HMVMXR0.cpp@ 80021

Last change on this file since 80021 was 80021, checked in by vboxsync, 5 years ago
VMM/HMVMXR0: Nested VMX: bugref:9180 nit.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 737.4 KB

Line
1	/* $Id: HMVMXR0.cpp 80021 2019-07-27 05:31:49Z vboxsync $ */
2	/** @file
3	* HM VMX (Intel VT-x) - Host Context Ring-0.
4	*/
5
6	/*
7	* Copyright (C) 2012-2019 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18
19	/*********************************************************************************************************************************
20	* Header Files *
21	*********************************************************************************************************************************/
22	#define LOG_GROUP LOG_GROUP_HM
23	#define VMCPU_INCL_CPUM_GST_CTX
24	#include <iprt/x86.h>
25	#include <iprt/asm-amd64-x86.h>
26	#include <iprt/thread.h>
27	#include <iprt/mem.h>
28	#include <iprt/mp.h>
29
30	#include <VBox/vmm/pdmapi.h>
31	#include <VBox/vmm/dbgf.h>
32	#include <VBox/vmm/iem.h>
33	#include <VBox/vmm/iom.h>
34	#include <VBox/vmm/selm.h>
35	#include <VBox/vmm/tm.h>
36	#include <VBox/vmm/em.h>
37	#include <VBox/vmm/gim.h>
38	#include <VBox/vmm/apic.h>
39	#ifdef VBOX_WITH_REM
40	# include <VBox/vmm/rem.h>
41	#endif
42	#include "HMInternal.h"
43	#include <VBox/vmm/vm.h>
44	#include <VBox/vmm/hmvmxinline.h>
45	#include "HMVMXR0.h"
46	#include "dtrace/VBoxVMM.h"
47
48	#ifdef DEBUG_ramshankar
49	# define HMVMX_ALWAYS_SAVE_GUEST_RFLAGS
50	# define HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
51	# define HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE
52	# define HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
53	# define HMVMX_ALWAYS_CLEAN_TRANSIENT
54	# define HMVMX_ALWAYS_CHECK_GUEST_STATE
55	# define HMVMX_ALWAYS_TRAP_ALL_XCPTS
56	# define HMVMX_ALWAYS_TRAP_PF
57	# define HMVMX_ALWAYS_FLUSH_TLB
58	# define HMVMX_ALWAYS_SWAP_EFER
59	#endif
60
61
62	/*********************************************************************************************************************************
63	* Defined Constants And Macros *
64	*********************************************************************************************************************************/
65	/** Use the function table. */
66	#define HMVMX_USE_FUNCTION_TABLE
67
68	/** Determine which tagged-TLB flush handler to use. */
69	#define HMVMX_FLUSH_TAGGED_TLB_EPT_VPID 0
70	#define HMVMX_FLUSH_TAGGED_TLB_EPT 1
71	#define HMVMX_FLUSH_TAGGED_TLB_VPID 2
72	#define HMVMX_FLUSH_TAGGED_TLB_NONE 3
73
74	/**
75	* Flags to skip redundant reads of some common VMCS fields that are not part of
76	* the guest-CPU or VCPU state but are needed while handling VM-exits.
77	*/
78	#define HMVMX_READ_IDT_VECTORING_INFO RT_BIT_32(0)
79	#define HMVMX_READ_IDT_VECTORING_ERROR_CODE RT_BIT_32(1)
80	#define HMVMX_READ_EXIT_QUALIFICATION RT_BIT_32(2)
81	#define HMVMX_READ_EXIT_INSTR_LEN RT_BIT_32(3)
82	#define HMVMX_READ_EXIT_INTERRUPTION_INFO RT_BIT_32(4)
83	#define HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE RT_BIT_32(5)
84	#define HMVMX_READ_EXIT_INSTR_INFO RT_BIT_32(6)
85	#define HMVMX_READ_GUEST_LINEAR_ADDR RT_BIT_32(7)
86	#define HMVMX_READ_GUEST_PHYSICAL_ADDR RT_BIT_32(8)
87
88	/** All the VMCS fields required for processing of exception/NMI VM-exits. */
89	#define HMVMX_READ_XCPT_INFO ( HMVMX_READ_EXIT_INTERRUPTION_INFO \
90	\| HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE \
91	\| HMVMX_READ_EXIT_INSTR_LEN \
92	\| HMVMX_READ_IDT_VECTORING_INFO \
93	\| HMVMX_READ_IDT_VECTORING_ERROR_CODE)
94
95	/** Assert that all the given fields have been read from the VMCS. */
96	#ifdef VBOX_STRICT
97	# define HMVMX_ASSERT_READ(a_pVmxTransient, a_fReadFields) \
98	do { \
99	uint32_t const fVmcsFieldRead = ASMAtomicUoReadU32(&pVmxTransient->fVmcsFieldsRead); \
100	RT_UNTRUSTED_NONVOLATILE_COPY_FENCE(); \
101	Assert((fVmcsFieldRead & (a_fReadFields)) == (a_fReadFields)); \
102	} while (0)
103	#else
104	# define HMVMX_ASSERT_READ(a_pVmxTransient, a_fReadFields) do { } while (0)
105	#endif
106
107	/**
108	* Subset of the guest-CPU state that is kept by VMX R0 code while executing the
109	* guest using hardware-assisted VMX.
110	*
111	* This excludes state like GPRs (other than RSP) which are always are
112	* swapped and restored across the world-switch and also registers like EFER,
113	* MSR which cannot be modified by the guest without causing a VM-exit.
114	*/
115	#define HMVMX_CPUMCTX_EXTRN_ALL ( CPUMCTX_EXTRN_RIP \
116	\| CPUMCTX_EXTRN_RFLAGS \
117	\| CPUMCTX_EXTRN_RSP \
118	\| CPUMCTX_EXTRN_SREG_MASK \
119	\| CPUMCTX_EXTRN_TABLE_MASK \
120	\| CPUMCTX_EXTRN_KERNEL_GS_BASE \
121	\| CPUMCTX_EXTRN_SYSCALL_MSRS \
122	\| CPUMCTX_EXTRN_SYSENTER_MSRS \
123	\| CPUMCTX_EXTRN_TSC_AUX \
124	\| CPUMCTX_EXTRN_OTHER_MSRS \
125	\| CPUMCTX_EXTRN_CR0 \
126	\| CPUMCTX_EXTRN_CR3 \
127	\| CPUMCTX_EXTRN_CR4 \
128	\| CPUMCTX_EXTRN_DR7 \
129	\| CPUMCTX_EXTRN_HWVIRT \
130	\| CPUMCTX_EXTRN_HM_VMX_MASK)
131
132	/**
133	* Exception bitmap mask for real-mode guests (real-on-v86).
134	*
135	* We need to intercept all exceptions manually except:
136	* - \#AC and \#DB are always intercepted to prevent the CPU from deadlocking
137	* due to bugs in Intel CPUs.
138	* - \#PF need not be intercepted even in real-mode if we have nested paging
139	* support.
140	*/
141	#define HMVMX_REAL_MODE_XCPT_MASK ( RT_BIT(X86_XCPT_DE) /* always: \| RT_BIT(X86_XCPT_DB) */ \| RT_BIT(X86_XCPT_NMI) \
142	\| RT_BIT(X86_XCPT_BP) \| RT_BIT(X86_XCPT_OF) \| RT_BIT(X86_XCPT_BR) \
143	\| RT_BIT(X86_XCPT_UD) \| RT_BIT(X86_XCPT_NM) \| RT_BIT(X86_XCPT_DF) \
144	\| RT_BIT(X86_XCPT_CO_SEG_OVERRUN) \| RT_BIT(X86_XCPT_TS) \| RT_BIT(X86_XCPT_NP) \
145	\| RT_BIT(X86_XCPT_SS) \| RT_BIT(X86_XCPT_GP) /* RT_BIT(X86_XCPT_PF) */ \
146	\| RT_BIT(X86_XCPT_MF) /* always: \| RT_BIT(X86_XCPT_AC) */ \| RT_BIT(X86_XCPT_MC) \
147	\| RT_BIT(X86_XCPT_XF))
148
149	/** Maximum VM-instruction error number. */
150	#define HMVMX_INSTR_ERROR_MAX 28
151
152	/** Profiling macro. */
153	#ifdef HM_PROFILE_EXIT_DISPATCH
154	# define HMVMX_START_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitDispatch, ed)
155	# define HMVMX_STOP_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitDispatch, ed)
156	#else
157	# define HMVMX_START_EXIT_DISPATCH_PROF() do { } while (0)
158	# define HMVMX_STOP_EXIT_DISPATCH_PROF() do { } while (0)
159	#endif
160
161	/** Assert that preemption is disabled or covered by thread-context hooks. */
162	#define HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu) Assert( VMMR0ThreadCtxHookIsEnabled((a_pVCpu)) \
163	\|\| !RTThreadPreemptIsEnabled(NIL_RTTHREAD))
164
165	/** Assert that we haven't migrated CPUs when thread-context hooks are not
166	* used. */
167	#define HMVMX_ASSERT_CPU_SAFE(a_pVCpu) AssertMsg( VMMR0ThreadCtxHookIsEnabled((a_pVCpu)) \
168	\|\| (a_pVCpu)->hm.s.idEnteredCpu == RTMpCpuId(), \
169	("Illegal migration! Entered on CPU %u Current %u\n", \
170	(a_pVCpu)->hm.s.idEnteredCpu, RTMpCpuId()))
171
172	/** Asserts that the given CPUMCTX_EXTRN_XXX bits are present in the guest-CPU
173	* context. */
174	#define HMVMX_CPUMCTX_ASSERT(a_pVCpu, a_fExtrnMbz) AssertMsg(!((a_pVCpu)->cpum.GstCtx.fExtrn & (a_fExtrnMbz)), \
175	("fExtrn=%#RX64 fExtrnMbz=%#RX64\n", \
176	(a_pVCpu)->cpum.GstCtx.fExtrn, (a_fExtrnMbz)))
177
178	/** Log the VM-exit reason with an easily visible marker to identify it in a
179	* potential sea of logging data. */
180	#define HMVMX_LOG_EXIT(a_pVCpu, a_uExitReason) \
181	do { \
182	Log4(("VM-exit: vcpu[%RU32] %85s -v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-\n", (a_pVCpu)->idCpu, \
183	HMGetVmxExitName(a_uExitReason))); \
184	} while (0) \
185
186
187	/*********************************************************************************************************************************
188	* Structures and Typedefs *
189	*********************************************************************************************************************************/
190	/**
191	* VMX per-VCPU transient state.
192	*
193	* A state structure for holding miscellaneous information across
194	* VMX non-root operation and restored after the transition.
195	*/
196	typedef struct VMXTRANSIENT
197	{
198	/** The host's rflags/eflags. */
199	RTCCUINTREG fEFlags;
200	#if HC_ARCH_BITS == 32
201	uint32_t u32Alignment0;
202	#endif
203	/** The guest's TPR value used for TPR shadowing. */
204	uint8_t u8GuestTpr;
205	/** Alignment. */
206	uint8_t abAlignment0[7];
207
208	/** The basic VM-exit reason. */
209	uint16_t uExitReason;
210	/** Alignment. */
211	uint16_t u16Alignment0;
212	/** The VM-exit interruption error code. */
213	uint32_t uExitIntErrorCode;
214	/** The VM-exit exit code qualification. */
215	uint64_t uExitQual;
216	/** The Guest-linear address. */
217	uint64_t uGuestLinearAddr;
218	/** The Guest-physical address. */
219	uint64_t uGuestPhysicalAddr;
220
221	/** The VM-exit interruption-information field. */
222	uint32_t uExitIntInfo;
223	/** The VM-exit instruction-length field. */
224	uint32_t cbInstr;
225	/** The VM-exit instruction-information field. */
226	VMXEXITINSTRINFO ExitInstrInfo;
227	/** Whether the VM-entry failed or not. */
228	bool fVMEntryFailed;
229	/** Whether we are currently executing a nested-guest. */
230	bool fIsNestedGuest;
231	/** Alignment. */
232	uint8_t abAlignment1[2];
233
234	/** The VM-entry interruption-information field. */
235	uint32_t uEntryIntInfo;
236	/** The VM-entry exception error code field. */
237	uint32_t uEntryXcptErrorCode;
238	/** The VM-entry instruction length field. */
239	uint32_t cbEntryInstr;
240
241	/** IDT-vectoring information field. */
242	uint32_t uIdtVectoringInfo;
243	/** IDT-vectoring error code. */
244	uint32_t uIdtVectoringErrorCode;
245
246	/** Mask of currently read VMCS fields; HMVMX_READ_XXX. */
247	uint32_t fVmcsFieldsRead;
248
249	/** Whether the guest debug state was active at the time of VM-exit. */
250	bool fWasGuestDebugStateActive;
251	/** Whether the hyper debug state was active at the time of VM-exit. */
252	bool fWasHyperDebugStateActive;
253	/** Whether TSC-offsetting and VMX-preemption timer was updated before VM-entry. */
254	bool fUpdatedTscOffsettingAndPreemptTimer;
255	/** Whether the VM-exit was caused by a page-fault during delivery of a
256	* contributory exception or a page-fault. */
257	bool fVectoringDoublePF;
258	/** Whether the VM-exit was caused by a page-fault during delivery of an
259	* external interrupt or NMI. */
260	bool fVectoringPF;
261	/** Whether the TSC_AUX MSR needs to be removed from the auto-load/store MSR
262	* area after VM-exit. */
263	bool fRemoveTscAuxMsr;
264	bool afAlignment0[2];
265
266	/** The VMCS info. object. */
267	PVMXVMCSINFO pVmcsInfo;
268	} VMXTRANSIENT;
269	AssertCompileMemberAlignment(VMXTRANSIENT, uExitReason, sizeof(uint64_t));
270	AssertCompileMemberAlignment(VMXTRANSIENT, uExitIntInfo, sizeof(uint64_t));
271	AssertCompileMemberAlignment(VMXTRANSIENT, uEntryIntInfo, sizeof(uint64_t));
272	AssertCompileMemberAlignment(VMXTRANSIENT, fWasGuestDebugStateActive, sizeof(uint64_t));
273	AssertCompileMemberAlignment(VMXTRANSIENT, pVmcsInfo, sizeof(uint64_t));
274	AssertCompileMemberSize(VMXTRANSIENT, ExitInstrInfo, sizeof(uint32_t));
275	/** Pointer to VMX transient state. */
276	typedef VMXTRANSIENT *PVMXTRANSIENT;
277	/** Pointer to a const VMX transient state. */
278	typedef const VMXTRANSIENT *PCVMXTRANSIENT;
279
280	/**
281	* Memory operand read or write access.
282	*/
283	typedef enum VMXMEMACCESS
284	{
285	VMXMEMACCESS_READ = 0,
286	VMXMEMACCESS_WRITE = 1
287	} VMXMEMACCESS;
288
289	/**
290	* VMX VM-exit handler.
291	*
292	* @returns Strict VBox status code (i.e. informational status codes too).
293	* @param pVCpu The cross context virtual CPU structure.
294	* @param pVmxTransient The VMX-transient structure.
295	*/
296	#ifndef HMVMX_USE_FUNCTION_TABLE
297	typedef VBOXSTRICTRC FNVMXEXITHANDLER(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient);
298	#else
299	typedef DECLCALLBACK(VBOXSTRICTRC) FNVMXEXITHANDLER(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient);
300	/** Pointer to VM-exit handler. */
301	typedef FNVMXEXITHANDLER *PFNVMXEXITHANDLER;
302	#endif
303
304	/**
305	* VMX VM-exit handler, non-strict status code.
306	*
307	* This is generally the same as FNVMXEXITHANDLER, the NSRC bit is just FYI.
308	*
309	* @returns VBox status code, no informational status code returned.
310	* @param pVCpu The cross context virtual CPU structure.
311	* @param pVmxTransient The VMX-transient structure.
312	*
313	* @remarks This is not used on anything returning VERR_EM_INTERPRETER as the
314	* use of that status code will be replaced with VINF_EM_SOMETHING
315	* later when switching over to IEM.
316	*/
317	#ifndef HMVMX_USE_FUNCTION_TABLE
318	typedef int FNVMXEXITHANDLERNSRC(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient);
319	#else
320	typedef FNVMXEXITHANDLER FNVMXEXITHANDLERNSRC;
321	#endif
322
323
324	/*********************************************************************************************************************************
325	* Internal Functions *
326	*********************************************************************************************************************************/
327	#ifndef HMVMX_USE_FUNCTION_TABLE
328	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient);
329	# define HMVMX_EXIT_DECL DECLINLINE(VBOXSTRICTRC)
330	# define HMVMX_EXIT_NSRC_DECL DECLINLINE(int)
331	#else
332	# define HMVMX_EXIT_DECL static DECLCALLBACK(VBOXSTRICTRC)
333	# define HMVMX_EXIT_NSRC_DECL HMVMX_EXIT_DECL
334	#endif
335	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
336	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExitNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient);
337	#endif
338
339	static int hmR0VmxImportGuestState(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo, uint64_t fWhat);
340	#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
341	static void hmR0VmxInitVmcsReadCache(PVMCPU pVCpu);
342	#endif
343
344	/** @name VM-exit handler prototypes.
345	* @{
346	*/
347	static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmi;
348	static FNVMXEXITHANDLER hmR0VmxExitExtInt;
349	static FNVMXEXITHANDLER hmR0VmxExitTripleFault;
350	static FNVMXEXITHANDLERNSRC hmR0VmxExitIntWindow;
351	static FNVMXEXITHANDLERNSRC hmR0VmxExitNmiWindow;
352	static FNVMXEXITHANDLER hmR0VmxExitTaskSwitch;
353	static FNVMXEXITHANDLER hmR0VmxExitCpuid;
354	static FNVMXEXITHANDLER hmR0VmxExitGetsec;
355	static FNVMXEXITHANDLER hmR0VmxExitHlt;
356	static FNVMXEXITHANDLERNSRC hmR0VmxExitInvd;
357	static FNVMXEXITHANDLER hmR0VmxExitInvlpg;
358	static FNVMXEXITHANDLER hmR0VmxExitRdpmc;
359	static FNVMXEXITHANDLER hmR0VmxExitVmcall;
360	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
361	static FNVMXEXITHANDLER hmR0VmxExitVmclear;
362	static FNVMXEXITHANDLER hmR0VmxExitVmlaunch;
363	static FNVMXEXITHANDLER hmR0VmxExitVmptrld;
364	static FNVMXEXITHANDLER hmR0VmxExitVmptrst;
365	static FNVMXEXITHANDLER hmR0VmxExitVmread;
366	static FNVMXEXITHANDLER hmR0VmxExitVmresume;
367	static FNVMXEXITHANDLER hmR0VmxExitVmwrite;
368	static FNVMXEXITHANDLER hmR0VmxExitVmxoff;
369	static FNVMXEXITHANDLER hmR0VmxExitVmxon;
370	static FNVMXEXITHANDLER hmR0VmxExitInvvpid;
371	#endif
372	static FNVMXEXITHANDLER hmR0VmxExitRdtsc;
373	static FNVMXEXITHANDLER hmR0VmxExitMovCRx;
374	static FNVMXEXITHANDLER hmR0VmxExitMovDRx;
375	static FNVMXEXITHANDLER hmR0VmxExitIoInstr;
376	static FNVMXEXITHANDLER hmR0VmxExitRdmsr;
377	static FNVMXEXITHANDLER hmR0VmxExitWrmsr;
378	static FNVMXEXITHANDLER hmR0VmxExitMwait;
379	static FNVMXEXITHANDLER hmR0VmxExitMtf;
380	static FNVMXEXITHANDLER hmR0VmxExitMonitor;
381	static FNVMXEXITHANDLER hmR0VmxExitPause;
382	static FNVMXEXITHANDLERNSRC hmR0VmxExitTprBelowThreshold;
383	static FNVMXEXITHANDLER hmR0VmxExitApicAccess;
384	static FNVMXEXITHANDLER hmR0VmxExitEptViolation;
385	static FNVMXEXITHANDLER hmR0VmxExitEptMisconfig;
386	static FNVMXEXITHANDLER hmR0VmxExitRdtscp;
387	static FNVMXEXITHANDLER hmR0VmxExitPreemptTimer;
388	static FNVMXEXITHANDLERNSRC hmR0VmxExitWbinvd;
389	static FNVMXEXITHANDLER hmR0VmxExitXsetbv;
390	static FNVMXEXITHANDLER hmR0VmxExitInvpcid;
391	static FNVMXEXITHANDLERNSRC hmR0VmxExitSetPendingXcptUD;
392	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrInvalidGuestState;
393	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrUnexpected;
394	/** @} */
395
396	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
397	/** @name Nested-guest VM-exit handler prototypes.
398	* @{
399	*/
400	static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmiNested;
401	static FNVMXEXITHANDLER hmR0VmxExitTripleFaultNested;
402	static FNVMXEXITHANDLERNSRC hmR0VmxExitIntWindowNested;
403	static FNVMXEXITHANDLERNSRC hmR0VmxExitNmiWindowNested;
404	static FNVMXEXITHANDLER hmR0VmxExitTaskSwitchNested;
405	static FNVMXEXITHANDLER hmR0VmxExitHltNested;
406	static FNVMXEXITHANDLER hmR0VmxExitInvlpgNested;
407	static FNVMXEXITHANDLER hmR0VmxExitRdpmcNested;
408	static FNVMXEXITHANDLER hmR0VmxExitVmreadVmwriteNested;
409	static FNVMXEXITHANDLER hmR0VmxExitRdtscNested;
410	static FNVMXEXITHANDLER hmR0VmxExitMovCRxNested;
411	static FNVMXEXITHANDLER hmR0VmxExitMovDRxNested;
412	static FNVMXEXITHANDLER hmR0VmxExitIoInstrNested;
413	static FNVMXEXITHANDLER hmR0VmxExitRdmsrNested;
414	static FNVMXEXITHANDLER hmR0VmxExitWrmsrNested;
415	static FNVMXEXITHANDLER hmR0VmxExitMwaitNested;
416	static FNVMXEXITHANDLER hmR0VmxExitMtfNested;
417	static FNVMXEXITHANDLER hmR0VmxExitMonitorNested;
418	static FNVMXEXITHANDLER hmR0VmxExitPauseNested;
419	static FNVMXEXITHANDLERNSRC hmR0VmxExitTprBelowThresholdNested;
420	static FNVMXEXITHANDLER hmR0VmxExitApicAccessNested;
421	static FNVMXEXITHANDLER hmR0VmxExitApicWriteNested;
422	static FNVMXEXITHANDLER hmR0VmxExitVirtEoiNested;
423	static FNVMXEXITHANDLER hmR0VmxExitRdtscpNested;
424	static FNVMXEXITHANDLERNSRC hmR0VmxExitWbinvdNested;
425	static FNVMXEXITHANDLER hmR0VmxExitInvpcidNested;
426	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrInvalidGuestStateNested;
427	static FNVMXEXITHANDLER hmR0VmxExitInstrNested;
428	static FNVMXEXITHANDLER hmR0VmxExitInstrWithInfoNested;
429	/** @} */
430	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
431
432
433	/*********************************************************************************************************************************
434	* Global Variables *
435	*********************************************************************************************************************************/
436	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
437	/**
438	* Array of all VMCS fields.
439	* Any fields added to the VT-x spec. should be added here.
440	*
441	* Currently only used to derive shadow VMCS fields for hardware-assisted execution
442	* of nested-guests.
443	*/
444	static const uint32_t g_aVmcsFields[] =
445	{
446	/* 16-bit control fields. */
447	VMX_VMCS16_VPID,
448	VMX_VMCS16_POSTED_INT_NOTIFY_VECTOR,
449	VMX_VMCS16_EPTP_INDEX,
450
451	/* 16-bit guest-state fields. */
452	VMX_VMCS16_GUEST_ES_SEL,
453	VMX_VMCS16_GUEST_CS_SEL,
454	VMX_VMCS16_GUEST_SS_SEL,
455	VMX_VMCS16_GUEST_DS_SEL,
456	VMX_VMCS16_GUEST_FS_SEL,
457	VMX_VMCS16_GUEST_GS_SEL,
458	VMX_VMCS16_GUEST_LDTR_SEL,
459	VMX_VMCS16_GUEST_TR_SEL,
460	VMX_VMCS16_GUEST_INTR_STATUS,
461	VMX_VMCS16_GUEST_PML_INDEX,
462
463	/* 16-bits host-state fields. */
464	VMX_VMCS16_HOST_ES_SEL,
465	VMX_VMCS16_HOST_CS_SEL,
466	VMX_VMCS16_HOST_SS_SEL,
467	VMX_VMCS16_HOST_DS_SEL,
468	VMX_VMCS16_HOST_FS_SEL,
469	VMX_VMCS16_HOST_GS_SEL,
470	VMX_VMCS16_HOST_TR_SEL,
471
472	/* 64-bit control fields. */
473	VMX_VMCS64_CTRL_IO_BITMAP_A_FULL,
474	VMX_VMCS64_CTRL_IO_BITMAP_A_HIGH,
475	VMX_VMCS64_CTRL_IO_BITMAP_B_FULL,
476	VMX_VMCS64_CTRL_IO_BITMAP_B_HIGH,
477	VMX_VMCS64_CTRL_MSR_BITMAP_FULL,
478	VMX_VMCS64_CTRL_MSR_BITMAP_HIGH,
479	VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL,
480	VMX_VMCS64_CTRL_EXIT_MSR_STORE_HIGH,
481	VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL,
482	VMX_VMCS64_CTRL_EXIT_MSR_LOAD_HIGH,
483	VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL,
484	VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_HIGH,
485	VMX_VMCS64_CTRL_EXEC_VMCS_PTR_FULL,
486	VMX_VMCS64_CTRL_EXEC_VMCS_PTR_HIGH,
487	VMX_VMCS64_CTRL_EXEC_PML_ADDR_FULL,
488	VMX_VMCS64_CTRL_EXEC_PML_ADDR_HIGH,
489	VMX_VMCS64_CTRL_TSC_OFFSET_FULL,
490	VMX_VMCS64_CTRL_TSC_OFFSET_HIGH,
491	VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL,
492	VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_HIGH,
493	VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL,
494	VMX_VMCS64_CTRL_APIC_ACCESSADDR_HIGH,
495	VMX_VMCS64_CTRL_POSTED_INTR_DESC_FULL,
496	VMX_VMCS64_CTRL_POSTED_INTR_DESC_HIGH,
497	VMX_VMCS64_CTRL_VMFUNC_CTRLS_FULL,
498	VMX_VMCS64_CTRL_VMFUNC_CTRLS_HIGH,
499	VMX_VMCS64_CTRL_EPTP_FULL,
500	VMX_VMCS64_CTRL_EPTP_HIGH,
501	VMX_VMCS64_CTRL_EOI_BITMAP_0_FULL,
502	VMX_VMCS64_CTRL_EOI_BITMAP_0_HIGH,
503	VMX_VMCS64_CTRL_EOI_BITMAP_1_FULL,
504	VMX_VMCS64_CTRL_EOI_BITMAP_1_HIGH,
505	VMX_VMCS64_CTRL_EOI_BITMAP_2_FULL,
506	VMX_VMCS64_CTRL_EOI_BITMAP_2_HIGH,
507	VMX_VMCS64_CTRL_EOI_BITMAP_3_FULL,
508	VMX_VMCS64_CTRL_EOI_BITMAP_3_HIGH,
509	VMX_VMCS64_CTRL_EPTP_LIST_FULL,
510	VMX_VMCS64_CTRL_EPTP_LIST_HIGH,
511	VMX_VMCS64_CTRL_VMREAD_BITMAP_FULL,
512	VMX_VMCS64_CTRL_VMREAD_BITMAP_HIGH,
513	VMX_VMCS64_CTRL_VMWRITE_BITMAP_FULL,
514	VMX_VMCS64_CTRL_VMWRITE_BITMAP_HIGH,
515	VMX_VMCS64_CTRL_VIRTXCPT_INFO_ADDR_FULL,
516	VMX_VMCS64_CTRL_VIRTXCPT_INFO_ADDR_HIGH,
517	VMX_VMCS64_CTRL_XSS_EXITING_BITMAP_FULL,
518	VMX_VMCS64_CTRL_XSS_EXITING_BITMAP_HIGH,
519	VMX_VMCS64_CTRL_ENCLS_EXITING_BITMAP_FULL,
520	VMX_VMCS64_CTRL_ENCLS_EXITING_BITMAP_HIGH,
521	VMX_VMCS64_CTRL_TSC_MULTIPLIER_FULL,
522	VMX_VMCS64_CTRL_TSC_MULTIPLIER_HIGH,
523
524	/* 64-bit read-only data fields. */
525	VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL,
526	VMX_VMCS64_RO_GUEST_PHYS_ADDR_HIGH,
527
528	/* 64-bit guest-state fields. */
529	VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL,
530	VMX_VMCS64_GUEST_VMCS_LINK_PTR_HIGH,
531	VMX_VMCS64_GUEST_DEBUGCTL_FULL,
532	VMX_VMCS64_GUEST_DEBUGCTL_HIGH,
533	VMX_VMCS64_GUEST_PAT_FULL,
534	VMX_VMCS64_GUEST_PAT_HIGH,
535	VMX_VMCS64_GUEST_EFER_FULL,
536	VMX_VMCS64_GUEST_EFER_HIGH,
537	VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL,
538	VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_HIGH,
539	VMX_VMCS64_GUEST_PDPTE0_FULL,
540	VMX_VMCS64_GUEST_PDPTE0_HIGH,
541	VMX_VMCS64_GUEST_PDPTE1_FULL,
542	VMX_VMCS64_GUEST_PDPTE1_HIGH,
543	VMX_VMCS64_GUEST_PDPTE2_FULL,
544	VMX_VMCS64_GUEST_PDPTE2_HIGH,
545	VMX_VMCS64_GUEST_PDPTE3_FULL,
546	VMX_VMCS64_GUEST_PDPTE3_HIGH,
547	VMX_VMCS64_GUEST_BNDCFGS_FULL,
548	VMX_VMCS64_GUEST_BNDCFGS_HIGH,
549
550	/* 64-bit host-state fields. */
551	VMX_VMCS64_HOST_PAT_FULL,
552	VMX_VMCS64_HOST_PAT_HIGH,
553	VMX_VMCS64_HOST_EFER_FULL,
554	VMX_VMCS64_HOST_EFER_HIGH,
555	VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_FULL,
556	VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_HIGH,
557
558	/* 32-bit control fields. */
559	VMX_VMCS32_CTRL_PIN_EXEC,
560	VMX_VMCS32_CTRL_PROC_EXEC,
561	VMX_VMCS32_CTRL_EXCEPTION_BITMAP,
562	VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK,
563	VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH,
564	VMX_VMCS32_CTRL_CR3_TARGET_COUNT,
565	VMX_VMCS32_CTRL_EXIT,
566	VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT,
567	VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT,
568	VMX_VMCS32_CTRL_ENTRY,
569	VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT,
570	VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO,
571	VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE,
572	VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH,
573	VMX_VMCS32_CTRL_TPR_THRESHOLD,
574	VMX_VMCS32_CTRL_PROC_EXEC2,
575	VMX_VMCS32_CTRL_PLE_GAP,
576	VMX_VMCS32_CTRL_PLE_WINDOW,
577
578	/* 32-bits read-only fields. */
579	VMX_VMCS32_RO_VM_INSTR_ERROR,
580	VMX_VMCS32_RO_EXIT_REASON,
581	VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO,
582	VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE,
583	VMX_VMCS32_RO_IDT_VECTORING_INFO,
584	VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE,
585	VMX_VMCS32_RO_EXIT_INSTR_LENGTH,
586	VMX_VMCS32_RO_EXIT_INSTR_INFO,
587
588	/* 32-bit guest-state fields. */
589	VMX_VMCS32_GUEST_ES_LIMIT,
590	VMX_VMCS32_GUEST_CS_LIMIT,
591	VMX_VMCS32_GUEST_SS_LIMIT,
592	VMX_VMCS32_GUEST_DS_LIMIT,
593	VMX_VMCS32_GUEST_FS_LIMIT,
594	VMX_VMCS32_GUEST_GS_LIMIT,
595	VMX_VMCS32_GUEST_LDTR_LIMIT,
596	VMX_VMCS32_GUEST_TR_LIMIT,
597	VMX_VMCS32_GUEST_GDTR_LIMIT,
598	VMX_VMCS32_GUEST_IDTR_LIMIT,
599	VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS,
600	VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS,
601	VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS,
602	VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS,
603	VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS,
604	VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS,
605	VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS,
606	VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS,
607	VMX_VMCS32_GUEST_INT_STATE,
608	VMX_VMCS32_GUEST_ACTIVITY_STATE,
609	VMX_VMCS32_GUEST_SMBASE,
610	VMX_VMCS32_GUEST_SYSENTER_CS,
611	VMX_VMCS32_PREEMPT_TIMER_VALUE,
612
613	/* 32-bit host-state fields. */
614	VMX_VMCS32_HOST_SYSENTER_CS,
615
616	/* Natural-width control fields. */
617	VMX_VMCS_CTRL_CR0_MASK,
618	VMX_VMCS_CTRL_CR4_MASK,
619	VMX_VMCS_CTRL_CR0_READ_SHADOW,
620	VMX_VMCS_CTRL_CR4_READ_SHADOW,
621	VMX_VMCS_CTRL_CR3_TARGET_VAL0,
622	VMX_VMCS_CTRL_CR3_TARGET_VAL1,
623	VMX_VMCS_CTRL_CR3_TARGET_VAL2,
624	VMX_VMCS_CTRL_CR3_TARGET_VAL3,
625
626	/* Natural-width read-only data fields. */
627	VMX_VMCS_RO_EXIT_QUALIFICATION,
628	VMX_VMCS_RO_IO_RCX,
629	VMX_VMCS_RO_IO_RSI,
630	VMX_VMCS_RO_IO_RDI,
631	VMX_VMCS_RO_IO_RIP,
632	VMX_VMCS_RO_GUEST_LINEAR_ADDR,
633
634	/* Natural-width guest-state field */
635	VMX_VMCS_GUEST_CR0,
636	VMX_VMCS_GUEST_CR3,
637	VMX_VMCS_GUEST_CR4,
638	VMX_VMCS_GUEST_ES_BASE,
639	VMX_VMCS_GUEST_CS_BASE,
640	VMX_VMCS_GUEST_SS_BASE,
641	VMX_VMCS_GUEST_DS_BASE,
642	VMX_VMCS_GUEST_FS_BASE,
643	VMX_VMCS_GUEST_GS_BASE,
644	VMX_VMCS_GUEST_LDTR_BASE,
645	VMX_VMCS_GUEST_TR_BASE,
646	VMX_VMCS_GUEST_GDTR_BASE,
647	VMX_VMCS_GUEST_IDTR_BASE,
648	VMX_VMCS_GUEST_DR7,
649	VMX_VMCS_GUEST_RSP,
650	VMX_VMCS_GUEST_RIP,
651	VMX_VMCS_GUEST_RFLAGS,
652	VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS,
653	VMX_VMCS_GUEST_SYSENTER_ESP,
654	VMX_VMCS_GUEST_SYSENTER_EIP,
655
656	/* Natural-width host-state fields */
657	VMX_VMCS_HOST_CR0,
658	VMX_VMCS_HOST_CR3,
659	VMX_VMCS_HOST_CR4,
660	VMX_VMCS_HOST_FS_BASE,
661	VMX_VMCS_HOST_GS_BASE,
662	VMX_VMCS_HOST_TR_BASE,
663	VMX_VMCS_HOST_GDTR_BASE,
664	VMX_VMCS_HOST_IDTR_BASE,
665	VMX_VMCS_HOST_SYSENTER_ESP,
666	VMX_VMCS_HOST_SYSENTER_EIP,
667	VMX_VMCS_HOST_RSP,
668	VMX_VMCS_HOST_RIP
669	};
670	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
671
672	#ifdef VMX_USE_CACHED_VMCS_ACCESSES
673	static const uint32_t g_aVmcsCacheSegBase[] =
674	{
675	VMX_VMCS_GUEST_ES_BASE_CACHE_IDX,
676	VMX_VMCS_GUEST_CS_BASE_CACHE_IDX,
677	VMX_VMCS_GUEST_SS_BASE_CACHE_IDX,
678	VMX_VMCS_GUEST_DS_BASE_CACHE_IDX,
679	VMX_VMCS_GUEST_FS_BASE_CACHE_IDX,
680	VMX_VMCS_GUEST_GS_BASE_CACHE_IDX
681	};
682	AssertCompile(RT_ELEMENTS(g_aVmcsCacheSegBase) == X86_SREG_COUNT);
683	#endif
684	static const uint32_t g_aVmcsSegBase[] =
685	{
686	VMX_VMCS_GUEST_ES_BASE,
687	VMX_VMCS_GUEST_CS_BASE,
688	VMX_VMCS_GUEST_SS_BASE,
689	VMX_VMCS_GUEST_DS_BASE,
690	VMX_VMCS_GUEST_FS_BASE,
691	VMX_VMCS_GUEST_GS_BASE
692	};
693	static const uint32_t g_aVmcsSegSel[] =
694	{
695	VMX_VMCS16_GUEST_ES_SEL,
696	VMX_VMCS16_GUEST_CS_SEL,
697	VMX_VMCS16_GUEST_SS_SEL,
698	VMX_VMCS16_GUEST_DS_SEL,
699	VMX_VMCS16_GUEST_FS_SEL,
700	VMX_VMCS16_GUEST_GS_SEL
701	};
702	static const uint32_t g_aVmcsSegLimit[] =
703	{
704	VMX_VMCS32_GUEST_ES_LIMIT,
705	VMX_VMCS32_GUEST_CS_LIMIT,
706	VMX_VMCS32_GUEST_SS_LIMIT,
707	VMX_VMCS32_GUEST_DS_LIMIT,
708	VMX_VMCS32_GUEST_FS_LIMIT,
709	VMX_VMCS32_GUEST_GS_LIMIT
710	};
711	static const uint32_t g_aVmcsSegAttr[] =
712	{
713	VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS,
714	VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS,
715	VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS,
716	VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS,
717	VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS,
718	VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS
719	};
720	AssertCompile(RT_ELEMENTS(g_aVmcsSegSel) == X86_SREG_COUNT);
721	AssertCompile(RT_ELEMENTS(g_aVmcsSegLimit) == X86_SREG_COUNT);
722	AssertCompile(RT_ELEMENTS(g_aVmcsSegBase) == X86_SREG_COUNT);
723	AssertCompile(RT_ELEMENTS(g_aVmcsSegAttr) == X86_SREG_COUNT);
724
725	#ifdef HMVMX_USE_FUNCTION_TABLE
726	/**
727	* VMX_EXIT dispatch table.
728	*/
729	static const PFNVMXEXITHANDLER g_apfnVMExitHandlers[VMX_EXIT_MAX + 1] =
730	{
731	/* 0 VMX_EXIT_XCPT_OR_NMI */ hmR0VmxExitXcptOrNmi,
732	/* 1 VMX_EXIT_EXT_INT */ hmR0VmxExitExtInt,
733	/* 2 VMX_EXIT_TRIPLE_FAULT */ hmR0VmxExitTripleFault,
734	/* 3 VMX_EXIT_INIT_SIGNAL */ hmR0VmxExitErrUnexpected,
735	/* 4 VMX_EXIT_SIPI */ hmR0VmxExitErrUnexpected,
736	/* 5 VMX_EXIT_IO_SMI */ hmR0VmxExitErrUnexpected,
737	/* 6 VMX_EXIT_SMI */ hmR0VmxExitErrUnexpected,
738	/* 7 VMX_EXIT_INT_WINDOW */ hmR0VmxExitIntWindow,
739	/* 8 VMX_EXIT_NMI_WINDOW */ hmR0VmxExitNmiWindow,
740	/* 9 VMX_EXIT_TASK_SWITCH */ hmR0VmxExitTaskSwitch,
741	/* 10 VMX_EXIT_CPUID */ hmR0VmxExitCpuid,
742	/* 11 VMX_EXIT_GETSEC */ hmR0VmxExitGetsec,
743	/* 12 VMX_EXIT_HLT */ hmR0VmxExitHlt,
744	/* 13 VMX_EXIT_INVD */ hmR0VmxExitInvd,
745	/* 14 VMX_EXIT_INVLPG */ hmR0VmxExitInvlpg,
746	/* 15 VMX_EXIT_RDPMC */ hmR0VmxExitRdpmc,
747	/* 16 VMX_EXIT_RDTSC */ hmR0VmxExitRdtsc,
748	/* 17 VMX_EXIT_RSM */ hmR0VmxExitErrUnexpected,
749	/* 18 VMX_EXIT_VMCALL */ hmR0VmxExitVmcall,
750	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
751	/* 19 VMX_EXIT_VMCLEAR */ hmR0VmxExitVmclear,
752	/* 20 VMX_EXIT_VMLAUNCH */ hmR0VmxExitVmlaunch,
753	/* 21 VMX_EXIT_VMPTRLD */ hmR0VmxExitVmptrld,
754	/* 22 VMX_EXIT_VMPTRST */ hmR0VmxExitVmptrst,
755	/* 23 VMX_EXIT_VMREAD */ hmR0VmxExitVmread,
756	/* 24 VMX_EXIT_VMRESUME */ hmR0VmxExitVmresume,
757	/* 25 VMX_EXIT_VMWRITE */ hmR0VmxExitVmwrite,
758	/* 26 VMX_EXIT_VMXOFF */ hmR0VmxExitVmxoff,
759	/* 27 VMX_EXIT_VMXON */ hmR0VmxExitVmxon,
760	#else
761	/* 19 VMX_EXIT_VMCLEAR */ hmR0VmxExitSetPendingXcptUD,
762	/* 20 VMX_EXIT_VMLAUNCH */ hmR0VmxExitSetPendingXcptUD,
763	/* 21 VMX_EXIT_VMPTRLD */ hmR0VmxExitSetPendingXcptUD,
764	/* 22 VMX_EXIT_VMPTRST */ hmR0VmxExitSetPendingXcptUD,
765	/* 23 VMX_EXIT_VMREAD */ hmR0VmxExitSetPendingXcptUD,
766	/* 24 VMX_EXIT_VMRESUME */ hmR0VmxExitSetPendingXcptUD,
767	/* 25 VMX_EXIT_VMWRITE */ hmR0VmxExitSetPendingXcptUD,
768	/* 26 VMX_EXIT_VMXOFF */ hmR0VmxExitSetPendingXcptUD,
769	/* 27 VMX_EXIT_VMXON */ hmR0VmxExitSetPendingXcptUD,
770	#endif
771	/* 28 VMX_EXIT_MOV_CRX */ hmR0VmxExitMovCRx,
772	/* 29 VMX_EXIT_MOV_DRX */ hmR0VmxExitMovDRx,
773	/* 30 VMX_EXIT_IO_INSTR */ hmR0VmxExitIoInstr,
774	/* 31 VMX_EXIT_RDMSR */ hmR0VmxExitRdmsr,
775	/* 32 VMX_EXIT_WRMSR */ hmR0VmxExitWrmsr,
776	/* 33 VMX_EXIT_ERR_INVALID_GUEST_STATE */ hmR0VmxExitErrInvalidGuestState,
777	/* 34 VMX_EXIT_ERR_MSR_LOAD */ hmR0VmxExitErrUnexpected,
778	/* 35 UNDEFINED */ hmR0VmxExitErrUnexpected,
779	/* 36 VMX_EXIT_MWAIT */ hmR0VmxExitMwait,
780	/* 37 VMX_EXIT_MTF */ hmR0VmxExitMtf,
781	/* 38 UNDEFINED */ hmR0VmxExitErrUnexpected,
782	/* 39 VMX_EXIT_MONITOR */ hmR0VmxExitMonitor,
783	/* 40 VMX_EXIT_PAUSE */ hmR0VmxExitPause,
784	/* 41 VMX_EXIT_ERR_MACHINE_CHECK */ hmR0VmxExitErrUnexpected,
785	/* 42 UNDEFINED */ hmR0VmxExitErrUnexpected,
786	/* 43 VMX_EXIT_TPR_BELOW_THRESHOLD */ hmR0VmxExitTprBelowThreshold,
787	/* 44 VMX_EXIT_APIC_ACCESS */ hmR0VmxExitApicAccess,
788	/* 45 VMX_EXIT_VIRTUALIZED_EOI */ hmR0VmxExitErrUnexpected,
789	/* 46 VMX_EXIT_GDTR_IDTR_ACCESS */ hmR0VmxExitErrUnexpected,
790	/* 47 VMX_EXIT_LDTR_TR_ACCESS */ hmR0VmxExitErrUnexpected,
791	/* 48 VMX_EXIT_EPT_VIOLATION */ hmR0VmxExitEptViolation,
792	/* 49 VMX_EXIT_EPT_MISCONFIG */ hmR0VmxExitEptMisconfig,
793	/* 50 VMX_EXIT_INVEPT */ hmR0VmxExitSetPendingXcptUD,
794	/* 51 VMX_EXIT_RDTSCP */ hmR0VmxExitRdtscp,
795	/* 52 VMX_EXIT_PREEMPT_TIMER */ hmR0VmxExitPreemptTimer,
796	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
797	/* 53 VMX_EXIT_INVVPID */ hmR0VmxExitInvvpid,
798	#else
799	/* 53 VMX_EXIT_INVVPID */ hmR0VmxExitSetPendingXcptUD,
800	#endif
801	/* 54 VMX_EXIT_WBINVD */ hmR0VmxExitWbinvd,
802	/* 55 VMX_EXIT_XSETBV */ hmR0VmxExitXsetbv,
803	/* 56 VMX_EXIT_APIC_WRITE */ hmR0VmxExitErrUnexpected,
804	/* 57 VMX_EXIT_RDRAND */ hmR0VmxExitErrUnexpected,
805	/* 58 VMX_EXIT_INVPCID */ hmR0VmxExitInvpcid,
806	/* 59 VMX_EXIT_VMFUNC */ hmR0VmxExitErrUnexpected,
807	/* 60 VMX_EXIT_ENCLS */ hmR0VmxExitErrUnexpected,
808	/* 61 VMX_EXIT_RDSEED */ hmR0VmxExitErrUnexpected,
809	/* 62 VMX_EXIT_PML_FULL */ hmR0VmxExitErrUnexpected,
810	/* 63 VMX_EXIT_XSAVES */ hmR0VmxExitErrUnexpected,
811	/* 64 VMX_EXIT_XRSTORS */ hmR0VmxExitErrUnexpected,
812	/* 65 UNDEFINED */ hmR0VmxExitErrUnexpected,
813	/* 66 VMX_EXIT_SPP_EVENT */ hmR0VmxExitErrUnexpected,
814	/* 67 VMX_EXIT_UMWAIT */ hmR0VmxExitErrUnexpected,
815	/* 68 VMX_EXIT_TPAUSE */ hmR0VmxExitErrUnexpected,
816	};
817	#endif /* HMVMX_USE_FUNCTION_TABLE */
818
819	#if defined(VBOX_STRICT) && defined(LOG_ENABLED)
820	static const char * const g_apszVmxInstrErrors[HMVMX_INSTR_ERROR_MAX + 1] =
821	{
822	/* 0 */ "(Not Used)",
823	/* 1 */ "VMCALL executed in VMX root operation.",
824	/* 2 */ "VMCLEAR with invalid physical address.",
825	/* 3 */ "VMCLEAR with VMXON pointer.",
826	/* 4 */ "VMLAUNCH with non-clear VMCS.",
827	/* 5 */ "VMRESUME with non-launched VMCS.",
828	/* 6 */ "VMRESUME after VMXOFF",
829	/* 7 */ "VM-entry with invalid control fields.",
830	/* 8 */ "VM-entry with invalid host state fields.",
831	/* 9 */ "VMPTRLD with invalid physical address.",
832	/* 10 */ "VMPTRLD with VMXON pointer.",
833	/* 11 */ "VMPTRLD with incorrect revision identifier.",
834	/* 12 */ "VMREAD/VMWRITE from/to unsupported VMCS component.",
835	/* 13 */ "VMWRITE to read-only VMCS component.",
836	/* 14 */ "(Not Used)",
837	/* 15 */ "VMXON executed in VMX root operation.",
838	/* 16 */ "VM-entry with invalid executive-VMCS pointer.",
839	/* 17 */ "VM-entry with non-launched executing VMCS.",
840	/* 18 */ "VM-entry with executive-VMCS pointer not VMXON pointer.",
841	/* 19 */ "VMCALL with non-clear VMCS.",
842	/* 20 */ "VMCALL with invalid VM-exit control fields.",
843	/* 21 */ "(Not Used)",
844	/* 22 */ "VMCALL with incorrect MSEG revision identifier.",
845	/* 23 */ "VMXOFF under dual monitor treatment of SMIs and SMM.",
846	/* 24 */ "VMCALL with invalid SMM-monitor features.",
847	/* 25 */ "VM-entry with invalid VM-execution control fields in executive VMCS.",
848	/* 26 */ "VM-entry with events blocked by MOV SS.",
849	/* 27 */ "(Not Used)",
850	/* 28 */ "Invalid operand to INVEPT/INVVPID."
851	};
852	#endif /* VBOX_STRICT && LOG_ENABLED */
853
854
855	/**
856	* Get the CR0 guest/host mask that does not change through the lifetime of a VM.
857	*
858	* Any bit set in this mask is owned by the host/hypervisor and would cause a
859	* VM-exit when modified by the guest.
860	*
861	* @returns The static CR0 guest/host mask.
862	* @param pVCpu The cross context virtual CPU structure.
863	*/
864	DECL_FORCE_INLINE(uint64_t) hmR0VmxGetFixedCr0Mask(PCVMCPU pVCpu)
865	{
866	/*
867	* Modifications to CR0 bits that VT-x ignores saving/restoring (CD, ET, NW) and
868	* to CR0 bits that we require for shadow paging (PG) by the guest must cause VM-exits.
869	*/
870	/** @todo Avoid intercepting CR0.PE with unrestricted guest execution. Fix PGM
871	* enmGuestMode to be in-sync with the current mode. See @bugref{6398}
872	* and @bugref{6944}. */
873	PVM pVM = pVCpu->CTX_SUFF(pVM);
874	return ( X86_CR0_PE
875	\| X86_CR0_NE
876	\| (pVM->hm.s.fNestedPaging ? 0 : X86_CR0_WP)
877	\| X86_CR0_PG
878	\| X86_CR0_ET /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.ET */
879	\| X86_CR0_CD /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.CD */
880	\| X86_CR0_NW); /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.NW */
881	}
882
883
884	/**
885	* Gets the CR4 guest/host mask that does not change through the lifetime of a VM.
886	*
887	* Any bit set in this mask is owned by the host/hypervisor and would cause a
888	* VM-exit when modified by the guest.
889	*
890	* @returns The static CR4 guest/host mask.
891	* @param pVCpu The cross context virtual CPU structure.
892	*/
893	DECL_FORCE_INLINE(uint64_t) hmR0VmxGetFixedCr4Mask(PCVMCPU pVCpu)
894	{
895	/*
896	* We need to look at the host features here (for e.g. OSXSAVE, PCID) because
897	* these bits are reserved on hardware that does not support them. Since the
898	* CPU cannot refer to our virtual CPUID, we need to intercept CR4 changes to
899	* these bits and handle it depending on whether we expose them to the guest.
900	*/
901	PVM pVM = pVCpu->CTX_SUFF(pVM);
902	bool const fXSaveRstor = pVM->cpum.ro.HostFeatures.fXSaveRstor;
903	bool const fPcid = pVM->cpum.ro.HostFeatures.fPcid;
904	return ( X86_CR4_VMXE
905	\| X86_CR4_VME
906	\| X86_CR4_PAE
907	\| X86_CR4_PGE
908	\| X86_CR4_PSE
909	\| (fXSaveRstor ? X86_CR4_OSXSAVE : 0)
910	\| (fPcid ? X86_CR4_PCIDE : 0));
911	}
912
913
914	/**
915	* Returns whether the the VM-exit MSR-store area differs from the VM-exit MSR-load
916	* area.
917	*
918	* @returns @c true if it's different, @c false otherwise.
919	* @param pVmcsInfo The VMCS info. object.
920	*/
921	DECL_FORCE_INLINE(bool) hmR0VmxIsSeparateExitMsrStoreAreaVmcs(PCVMXVMCSINFO pVmcsInfo)
922	{
923	return RT_BOOL( pVmcsInfo->pvGuestMsrStore != pVmcsInfo->pvGuestMsrLoad
924	&& pVmcsInfo->pvGuestMsrStore);
925	}
926
927
928	/**
929	* Sets the given Processor-based VM-execution controls.
930	*
931	* @param pVmxTransient The VMX-transient structure.
932	* @param uProcCtls The Processor-based VM-execution controls to set.
933	*/
934	static void hmR0VmxSetProcCtlsVmcs(PVMXTRANSIENT pVmxTransient, uint32_t uProcCtls)
935	{
936	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
937	if ((pVmcsInfo->u32ProcCtls & uProcCtls) != uProcCtls)
938	{
939	pVmcsInfo->u32ProcCtls \|= uProcCtls;
940	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
941	AssertRC(rc);
942	}
943	}
944
945
946	/**
947	* Removes the given Processor-based VM-execution controls.
948	*
949	* @param pVCpu The cross context virtual CPU structure.
950	* @param pVmxTransient The VMX-transient structure.
951	* @param uProcCtls The Processor-based VM-execution controls to remove.
952	*
953	* @remarks When executing a nested-guest, this will not remove any of the specified
954	* controls if the guest hypervisor has set any one of them.
955	*/
956	static void hmR0VmxRemoveProcCtlsVmcs(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uProcCtls)
957	{
958	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
959	if (pVmcsInfo->u32ProcCtls & uProcCtls)
960	{
961	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
962	bool const fRemoveCtls = !pVmxTransient->fIsNestedGuest
963	? true
964	: !CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, uProcCtls);
965	#else
966	NOREF(pVCpu);
967	bool const fRemoveCtls = true;
968	#endif
969	if (fRemoveCtls)
970	{
971	pVmcsInfo->u32ProcCtls &= ~uProcCtls;
972	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
973	AssertRC(rc);
974	}
975	}
976	}
977
978
979	/**
980	* Sets the TSC offset for the current VMCS.
981	*
982	* @param pVCpu The cross context virtual CPU structure.
983	* @param uTscOffset The TSC offset to set.
984	* @param pVmcsInfo The VMCS info. object.
985	*/
986	static void hmR0VmxSetTscOffsetVmcs(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo, uint64_t uTscOffset)
987	{
988	NOREF(pVCpu); /* Used implicitly by VMXWriteVmcs64 on 32-bit hosts. */
989	if (pVmcsInfo->u64TscOffset != uTscOffset)
990	{
991	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, uTscOffset);
992	AssertRC(rc);
993	pVmcsInfo->u64TscOffset = uTscOffset;
994	}
995	}
996
997
998	/**
999	* Adds one or more exceptions to the exception bitmap and commits it to the current
1000	* VMCS.
1001	*
1002	* @returns VBox status code.
1003	* @param pVmxTransient The VMX-transient structure.
1004	* @param uXcptMask The exception(s) to add.
1005	*/
1006	static int hmR0VmxAddXcptInterceptMask(PVMXTRANSIENT pVmxTransient, uint32_t uXcptMask)
1007	{
1008	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1009	uint32_t uXcptBitmap = pVmcsInfo->u32XcptBitmap;
1010	if ((uXcptBitmap & uXcptMask) != uXcptMask)
1011	{
1012	uXcptBitmap \|= uXcptMask;
1013	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
1014	AssertRCReturn(rc, rc);
1015	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
1016	}
1017	return VINF_SUCCESS;
1018	}
1019
1020
1021	/**
1022	* Adds an exception to the exception bitmap and commits it to the current VMCS.
1023	*
1024	* @returns VBox status code.
1025	* @param pVmxTransient The VMX-transient structure.
1026	* @param uXcpt The exception to add.
1027	*/
1028	static int hmR0VmxAddXcptIntercept(PVMXTRANSIENT pVmxTransient, uint8_t uXcpt)
1029	{
1030	Assert(uXcpt <= X86_XCPT_LAST);
1031	return hmR0VmxAddXcptInterceptMask(pVmxTransient, RT_BIT_32(uXcpt));
1032	}
1033
1034
1035	/**
1036	* Remove one or more exceptions from the exception bitmap and commits it to the
1037	* current VMCS.
1038	*
1039	* This takes care of not removing the exception intercept if a nested-guest
1040	* requires the exception to be intercepted.
1041	*
1042	* @returns VBox status code.
1043	* @param pVCpu The cross context virtual CPU structure.
1044	* @param pVmxTransient The VMX-transient structure.
1045	* @param uXcptMask The exception(s) to remove.
1046	*/
1047	static int hmR0VmxRemoveXcptInterceptMask(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uXcptMask)
1048	{
1049	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1050	uint32_t u32XcptBitmap = pVmcsInfo->u32XcptBitmap;
1051	if (u32XcptBitmap & uXcptMask)
1052	{
1053	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1054	if (!pVmxTransient->fIsNestedGuest)
1055	{ /* likely */ }
1056	else
1057	{
1058	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1059	uXcptMask &= ~pVmcsNstGst->u32XcptBitmap;
1060	}
1061	#endif
1062	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
1063	uXcptMask &= ~( RT_BIT(X86_XCPT_BP)
1064	\| RT_BIT(X86_XCPT_DE)
1065	\| RT_BIT(X86_XCPT_NM)
1066	\| RT_BIT(X86_XCPT_TS)
1067	\| RT_BIT(X86_XCPT_UD)
1068	\| RT_BIT(X86_XCPT_NP)
1069	\| RT_BIT(X86_XCPT_SS)
1070	\| RT_BIT(X86_XCPT_GP)
1071	\| RT_BIT(X86_XCPT_PF)
1072	\| RT_BIT(X86_XCPT_MF));
1073	#elif defined(HMVMX_ALWAYS_TRAP_PF)
1074	uXcptMask &= ~RT_BIT(X86_XCPT_PF);
1075	#endif
1076	if (uXcptMask)
1077	{
1078	/* Validate we are not removing any essential exception intercepts. */
1079	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging \|\| !(uXcptMask & RT_BIT(X86_XCPT_PF))); RT_NOREF(pVCpu);
1080	Assert(!(uXcptMask & RT_BIT(X86_XCPT_DB)));
1081	Assert(!(uXcptMask & RT_BIT(X86_XCPT_AC)));
1082
1083	/* Remove it from the exception bitmap. */
1084	u32XcptBitmap &= ~uXcptMask;
1085
1086	/* Commit and update the cache if necessary. */
1087	if (pVmcsInfo->u32XcptBitmap != u32XcptBitmap)
1088	{
1089	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
1090	AssertRCReturn(rc, rc);
1091	pVmcsInfo->u32XcptBitmap = u32XcptBitmap;
1092	}
1093	}
1094	}
1095	return VINF_SUCCESS;
1096	}
1097
1098
1099	/**
1100	* Remove an exceptions from the exception bitmap and commits it to the current
1101	* VMCS.
1102	*
1103	* @returns VBox status code.
1104	* @param pVCpu The cross context virtual CPU structure.
1105	* @param pVmxTransient The VMX-transient structure.
1106	* @param uXcpt The exception to remove.
1107	*/
1108	static int hmR0VmxRemoveXcptIntercept(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, uint8_t uXcpt)
1109	{
1110	return hmR0VmxRemoveXcptInterceptMask(pVCpu, pVmxTransient, RT_BIT(uXcpt));
1111	}
1112
1113
1114	/**
1115	* Loads the VMCS specified by the VMCS info. object.
1116	*
1117	* @returns VBox status code.
1118	* @param pVmcsInfo The VMCS info. object.
1119	*
1120	* @remarks Can be called with interrupts disabled.
1121	*/
1122	static int hmR0VmxLoadVmcs(PVMXVMCSINFO pVmcsInfo)
1123	{
1124	Assert(pVmcsInfo->HCPhysVmcs != 0 && pVmcsInfo->HCPhysVmcs != NIL_RTHCPHYS);
1125	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1126
1127	int rc = VMXLoadVmcs(pVmcsInfo->HCPhysVmcs);
1128	if (RT_SUCCESS(rc))
1129	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_CURRENT;
1130	return rc;
1131	}
1132
1133
1134	/**
1135	* Clears the VMCS specified by the VMCS info. object.
1136	*
1137	* @returns VBox status code.
1138	* @param pVmcsInfo The VMCS info. object.
1139	*
1140	* @remarks Can be called with interrupts disabled.
1141	*/
1142	static int hmR0VmxClearVmcs(PVMXVMCSINFO pVmcsInfo)
1143	{
1144	Assert(pVmcsInfo->HCPhysVmcs != 0 && pVmcsInfo->HCPhysVmcs != NIL_RTHCPHYS);
1145	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1146
1147	int rc = VMXClearVmcs(pVmcsInfo->HCPhysVmcs);
1148	if (RT_SUCCESS(rc))
1149	pVmcsInfo->fVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
1150	return rc;
1151	}
1152
1153
1154	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1155	/**
1156	* Loads the shadow VMCS specified by the VMCS info. object.
1157	*
1158	* @returns VBox status code.
1159	* @param pVmcsInfo The VMCS info. object.
1160	*
1161	* @remarks Can be called with interrupts disabled.
1162	*/
1163	static int hmR0VmxLoadShadowVmcs(PVMXVMCSINFO pVmcsInfo)
1164	{
1165	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1166	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
1167
1168	int rc = VMXLoadVmcs(pVmcsInfo->HCPhysShadowVmcs);
1169	if (RT_SUCCESS(rc))
1170	pVmcsInfo->fShadowVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_CURRENT;
1171	return rc;
1172	}
1173
1174
1175	/**
1176	* Clears the shadow VMCS specified by the VMCS info. object.
1177	*
1178	* @returns VBox status code.
1179	* @param pVmcsInfo The VMCS info. object.
1180	*
1181	* @remarks Can be called with interrupts disabled.
1182	*/
1183	static int hmR0VmxClearShadowVmcs(PVMXVMCSINFO pVmcsInfo)
1184	{
1185	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1186	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
1187
1188	int rc = VMXClearVmcs(pVmcsInfo->HCPhysShadowVmcs);
1189	if (RT_SUCCESS(rc))
1190	pVmcsInfo->fShadowVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
1191	return rc;
1192	}
1193
1194
1195	/**
1196	* Switches from and to the specified VMCSes.
1197	*
1198	* @returns VBox status code.
1199	* @param pVmcsInfoFrom The VMCS info. object we are switching from.
1200	* @param pVmcsInfoTo The VMCS info. object we are switching to.
1201	*
1202	* @remarks Called with interrupts disabled.
1203	*/
1204	static int hmR0VmxSwitchVmcs(PVMXVMCSINFO pVmcsInfoFrom, PVMXVMCSINFO pVmcsInfoTo)
1205	{
1206	/*
1207	* Clear the VMCS we are switching out if it has not already been cleared.
1208	* This will sync any CPU internal data back to the VMCS.
1209	*/
1210	if (pVmcsInfoFrom->fVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
1211	{
1212	int rc = hmR0VmxClearVmcs(pVmcsInfoFrom);
1213	if (RT_SUCCESS(rc))
1214	{
1215	/*
1216	* The shadow VMCS, if any, would not be active at this point since we
1217	* would have cleared it while importing the virtual hardware-virtualization
1218	* state as part the VMLAUNCH/VMRESUME VM-exit. Hence, there's no need to
1219	* clear the shadow VMCS here, just assert for safety.
1220	*/
1221	Assert(!pVmcsInfoFrom->pvShadowVmcs \|\| pVmcsInfoFrom->fShadowVmcsState == VMX_V_VMCS_LAUNCH_STATE_CLEAR);
1222	}
1223	else
1224	return rc;
1225	}
1226
1227	/*
1228	* Clear the VMCS we are switching to if it has not already been cleared.
1229	* This will initialize the VMCS launch state to "clear" required for loading it.
1230	*
1231	* See Intel spec. 31.6 "Preparation And Launching A Virtual Machine".
1232	*/
1233	if (pVmcsInfoTo->fVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
1234	{
1235	int rc = hmR0VmxClearVmcs(pVmcsInfoTo);
1236	if (RT_SUCCESS(rc))
1237	{ /* likely */ }
1238	else
1239	return rc;
1240	}
1241
1242	/*
1243	* Finally, load the VMCS we are switching to.
1244	*/
1245	return hmR0VmxLoadVmcs(pVmcsInfoTo);
1246	}
1247
1248
1249	/**
1250	* Switches between the guest VMCS and the nested-guest VMCS as specified by the
1251	* caller.
1252	*
1253	* @returns VBox status code.
1254	* @param pVCpu The cross context virtual CPU structure.
1255	* @param fSwitchToNstGstVmcs Whether to switch to the nested-guest VMCS (pass
1256	* true) or guest VMCS (pass false).
1257	*/
1258	static int hmR0VmxSwitchToGstOrNstGstVmcs(PVMCPU pVCpu, bool fSwitchToNstGstVmcs)
1259	{
1260	/* Ensure we have synced everything from the guest-CPU context to the VMCS before switching. */
1261	HMVMX_CPUMCTX_ASSERT(pVCpu, HMVMX_CPUMCTX_EXTRN_ALL);
1262
1263	PVMXVMCSINFO pVmcsInfoFrom;
1264	PVMXVMCSINFO pVmcsInfoTo;
1265	if (fSwitchToNstGstVmcs)
1266	{
1267	pVmcsInfoFrom = &pVCpu->hm.s.vmx.VmcsInfo;
1268	pVmcsInfoTo = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
1269	}
1270	else
1271	{
1272	pVmcsInfoFrom = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
1273	pVmcsInfoTo = &pVCpu->hm.s.vmx.VmcsInfo;
1274	}
1275
1276	/*
1277	* Disable interrupts to prevent being preempted while we switch the current VMCS as the
1278	* preemption hook code path acquires the current VMCS.
1279	*/
1280	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1281
1282	int rc = hmR0VmxSwitchVmcs(pVmcsInfoFrom, pVmcsInfoTo);
1283	if (RT_SUCCESS(rc))
1284	{
1285	pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs = fSwitchToNstGstVmcs;
1286
1287	/*
1288	* If we are switching to a VMCS that was executed on a different host CPU or was
1289	* never executed before, flag that we need to export the host state before executing
1290	* guest/nested-guest code using hardware-assisted VMX.
1291	*
1292	* This could probably be done in a preemptible context since the preemption hook
1293	* will flag the necessary change in host context. However, since preemption is
1294	* already disabled and to avoid making assumptions about host specific code in
1295	* RTMpCpuId when called with preemption enabled, we'll do this while preemption is
1296	* disabled.
1297	*/
1298	if (pVmcsInfoTo->idHostCpu == RTMpCpuId())
1299	{ /* likely */ }
1300	else
1301	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE);
1302
1303	ASMSetFlags(fEFlags);
1304
1305	/*
1306	* We use a different VM-exit MSR-store areas for the guest and nested-guest. Hence,
1307	* flag that we need to update the host MSR values there. Even if we decide in the
1308	* future to share the VM-exit MSR-store area page between the guest and nested-guest,
1309	* if its content differs, we would have to update the host MSRs anyway.
1310	*/
1311	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = false;
1312	}
1313	else
1314	ASMSetFlags(fEFlags);
1315	return rc;
1316	}
1317	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
1318
1319
1320	/**
1321	* Updates the VM's last error record.
1322	*
1323	* If there was a VMX instruction error, reads the error data from the VMCS and
1324	* updates VCPU's last error record as well.
1325	*
1326	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
1327	* Can be NULL if @a rc is not VERR_VMX_UNABLE_TO_START_VM or
1328	* VERR_VMX_INVALID_VMCS_FIELD.
1329	* @param rc The error code.
1330	*/
1331	static void hmR0VmxUpdateErrorRecord(PVMCPU pVCpu, int rc)
1332	{
1333	if ( rc == VERR_VMX_INVALID_VMCS_FIELD
1334	\|\| rc == VERR_VMX_UNABLE_TO_START_VM)
1335	{
1336	AssertPtrReturnVoid(pVCpu);
1337	VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
1338	}
1339	pVCpu->CTX_SUFF(pVM)->hm.s.rcInit = rc;
1340	}
1341
1342
1343	#ifdef VBOX_STRICT
1344	/**
1345	* Reads the VM-entry interruption-information field from the VMCS into the VMX
1346	* transient structure.
1347	*
1348	* @returns VBox status code.
1349	* @param pVmxTransient The VMX-transient structure.
1350	*
1351	* @remarks No-long-jump zone!!!
1352	*/
1353	DECLINLINE(int) hmR0VmxReadEntryIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
1354	{
1355	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &pVmxTransient->uEntryIntInfo);
1356	AssertRCReturn(rc, rc);
1357	return VINF_SUCCESS;
1358	}
1359
1360
1361	/**
1362	* Reads the VM-entry exception error code field from the VMCS into
1363	* the VMX transient structure.
1364	*
1365	* @returns VBox status code.
1366	* @param pVmxTransient The VMX-transient structure.
1367	*
1368	* @remarks No-long-jump zone!!!
1369	*/
1370	DECLINLINE(int) hmR0VmxReadEntryXcptErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1371	{
1372	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &pVmxTransient->uEntryXcptErrorCode);
1373	AssertRCReturn(rc, rc);
1374	return VINF_SUCCESS;
1375	}
1376
1377
1378	/**
1379	* Reads the VM-entry exception error code field from the VMCS into
1380	* the VMX transient structure.
1381	*
1382	* @returns VBox status code.
1383	* @param pVmxTransient The VMX-transient structure.
1384	*
1385	* @remarks No-long-jump zone!!!
1386	*/
1387	DECLINLINE(int) hmR0VmxReadEntryInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
1388	{
1389	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &pVmxTransient->cbEntryInstr);
1390	AssertRCReturn(rc, rc);
1391	return VINF_SUCCESS;
1392	}
1393	#endif /* VBOX_STRICT */
1394
1395
1396	/**
1397	* Reads the VM-exit interruption-information field from the VMCS into the VMX
1398	* transient structure.
1399	*
1400	* @returns VBox status code.
1401	* @param pVmxTransient The VMX-transient structure.
1402	*/
1403	DECLINLINE(int) hmR0VmxReadExitIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
1404	{
1405	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INTERRUPTION_INFO))
1406	{
1407	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
1408	AssertRCReturn(rc,rc);
1409	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INTERRUPTION_INFO;
1410	}
1411	return VINF_SUCCESS;
1412	}
1413
1414
1415	/**
1416	* Reads the VM-exit interruption error code from the VMCS into the VMX
1417	* transient structure.
1418	*
1419	* @returns VBox status code.
1420	* @param pVmxTransient The VMX-transient structure.
1421	*/
1422	DECLINLINE(int) hmR0VmxReadExitIntErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1423	{
1424	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE))
1425	{
1426	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
1427	AssertRCReturn(rc, rc);
1428	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE;
1429	}
1430	return VINF_SUCCESS;
1431	}
1432
1433
1434	/**
1435	* Reads the VM-exit instruction length field from the VMCS into the VMX
1436	* transient structure.
1437	*
1438	* @returns VBox status code.
1439	* @param pVmxTransient The VMX-transient structure.
1440	*/
1441	DECLINLINE(int) hmR0VmxReadExitInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
1442	{
1443	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INSTR_LEN))
1444	{
1445	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbInstr);
1446	AssertRCReturn(rc, rc);
1447	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INSTR_LEN;
1448	}
1449	return VINF_SUCCESS;
1450	}
1451
1452
1453	/**
1454	* Reads the VM-exit instruction-information field from the VMCS into
1455	* the VMX transient structure.
1456	*
1457	* @returns VBox status code.
1458	* @param pVmxTransient The VMX-transient structure.
1459	*/
1460	DECLINLINE(int) hmR0VmxReadExitInstrInfoVmcs(PVMXTRANSIENT pVmxTransient)
1461	{
1462	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INSTR_INFO))
1463	{
1464	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
1465	AssertRCReturn(rc, rc);
1466	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INSTR_INFO;
1467	}
1468	return VINF_SUCCESS;
1469	}
1470
1471
1472	/**
1473	* Reads the Exit Qualification from the VMCS into the VMX transient structure.
1474	*
1475	* @returns VBox status code.
1476	* @param pVCpu The cross context virtual CPU structure of the
1477	* calling EMT. (Required for the VMCS cache case.)
1478	* @param pVmxTransient The VMX-transient structure.
1479	*/
1480	DECLINLINE(int) hmR0VmxReadExitQualVmcs(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
1481	{
1482	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_QUALIFICATION))
1483	{
1484	int rc = VMXReadVmcsGstN(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQual); NOREF(pVCpu);
1485	AssertRCReturn(rc, rc);
1486	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_QUALIFICATION;
1487	}
1488	return VINF_SUCCESS;
1489	}
1490
1491
1492	/**
1493	* Reads the Guest-linear address from the VMCS into the VMX transient structure.
1494	*
1495	* @returns VBox status code.
1496	* @param pVCpu The cross context virtual CPU structure of the
1497	* calling EMT. (Required for the VMCS cache case.)
1498	* @param pVmxTransient The VMX-transient structure.
1499	*/
1500	DECLINLINE(int) hmR0VmxReadGuestLinearAddrVmcs(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
1501	{
1502	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_LINEAR_ADDR))
1503	{
1504	int rc = VMXReadVmcsGstN(VMX_VMCS_RO_GUEST_LINEAR_ADDR, &pVmxTransient->uGuestLinearAddr); NOREF(pVCpu);
1505	AssertRCReturn(rc, rc);
1506	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_LINEAR_ADDR;
1507	}
1508	return VINF_SUCCESS;
1509	}
1510
1511
1512	/**
1513	* Reads the Guest-physical address from the VMCS into the VMX transient structure.
1514	*
1515	* @returns VBox status code.
1516	* @param pVCpu The cross context virtual CPU structure of the
1517	* calling EMT. (Required for the VMCS cache case.)
1518	* @param pVmxTransient The VMX-transient structure.
1519	*/
1520	DECLINLINE(int) hmR0VmxReadGuestPhysicalAddrVmcs(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
1521	{
1522	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_PHYSICAL_ADDR))
1523	{
1524	int rc = VMXReadVmcs64(VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL, &pVmxTransient->uGuestPhysicalAddr); NOREF(pVCpu);
1525	AssertRCReturn(rc, rc);
1526	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_PHYSICAL_ADDR;
1527	}
1528	return VINF_SUCCESS;
1529	}
1530
1531
1532	/**
1533	* Reads the IDT-vectoring information field from the VMCS into the VMX
1534	* transient structure.
1535	*
1536	* @returns VBox status code.
1537	* @param pVmxTransient The VMX-transient structure.
1538	*
1539	* @remarks No-long-jump zone!!!
1540	*/
1541	DECLINLINE(int) hmR0VmxReadIdtVectoringInfoVmcs(PVMXTRANSIENT pVmxTransient)
1542	{
1543	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_IDT_VECTORING_INFO))
1544	{
1545	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_INFO, &pVmxTransient->uIdtVectoringInfo);
1546	AssertRCReturn(rc, rc);
1547	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_IDT_VECTORING_INFO;
1548	}
1549	return VINF_SUCCESS;
1550	}
1551
1552
1553	/**
1554	* Reads the IDT-vectoring error code from the VMCS into the VMX
1555	* transient structure.
1556	*
1557	* @returns VBox status code.
1558	* @param pVmxTransient The VMX-transient structure.
1559	*/
1560	DECLINLINE(int) hmR0VmxReadIdtVectoringErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1561	{
1562	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_IDT_VECTORING_ERROR_CODE))
1563	{
1564	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
1565	AssertRCReturn(rc, rc);
1566	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_IDT_VECTORING_ERROR_CODE;
1567	}
1568	return VINF_SUCCESS;
1569	}
1570
1571	#ifdef HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
1572	/**
1573	* Reads all relevant read-only VMCS fields into the VMX transient structure.
1574	*
1575	* @returns VBox status code.
1576	* @param pVCpu The cross context virtual CPU structure of the
1577	* calling EMT. (Required for the VMCS cache case.)
1578	* @param pVmxTransient The VMX-transient structure.
1579	*/
1580	static int hmR0VmxReadAllRoFieldsVmcs(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
1581	{
1582	NOREF(pVCpu); /* Used implicitly by VMXReadVmcsGstN on 32-bit hosts. */
1583	int rc = VMXReadVmcsGstN(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQual);
1584	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbInstr);
1585	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
1586	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_INFO, &pVmxTransient->uIdtVectoringInfo);
1587	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
1588	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
1589	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
1590	rc \|= VMXReadVmcsGstN(VMX_VMCS_RO_GUEST_LINEAR_ADDR, &pVmxTransient->uGuestLinearAddr);
1591	rc \|= VMXReadVmcsGstN(VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL, &pVmxTransient->uGuestPhysicalAddr);
1592	AssertRCReturn(rc, rc);
1593	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_QUALIFICATION
1594	\| HMVMX_READ_EXIT_INSTR_LEN
1595	\| HMVMX_READ_EXIT_INSTR_INFO
1596	\| HMVMX_READ_IDT_VECTORING_INFO
1597	\| HMVMX_READ_IDT_VECTORING_ERROR_CODE
1598	\| HMVMX_READ_EXIT_INTERRUPTION_INFO
1599	\| HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE
1600	\| HMVMX_READ_GUEST_LINEAR_ADDR
1601	\| HMVMX_READ_GUEST_PHYSICAL_ADDR;
1602	return VINF_SUCCESS;
1603	}
1604	#endif
1605
1606	/**
1607	* Enters VMX root mode operation on the current CPU.
1608	*
1609	* @returns VBox status code.
1610	* @param pVM The cross context VM structure. Can be
1611	* NULL, after a resume.
1612	* @param HCPhysCpuPage Physical address of the VMXON region.
1613	* @param pvCpuPage Pointer to the VMXON region.
1614	*/
1615	static int hmR0VmxEnterRootMode(PVM pVM, RTHCPHYS HCPhysCpuPage, void *pvCpuPage)
1616	{
1617	Assert(HCPhysCpuPage && HCPhysCpuPage != NIL_RTHCPHYS);
1618	Assert(RT_ALIGN_T(HCPhysCpuPage, _4K, RTHCPHYS) == HCPhysCpuPage);
1619	Assert(pvCpuPage);
1620	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1621
1622	if (pVM)
1623	{
1624	/* Write the VMCS revision identifier to the VMXON region. */
1625	(uint32_t )pvCpuPage = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_ID);
1626	}
1627
1628	/* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with CR4. */
1629	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1630
1631	/* Enable the VMX bit in CR4 if necessary. */
1632	RTCCUINTREG const uOldCr4 = SUPR0ChangeCR4(X86_CR4_VMXE, RTCCUINTREG_MAX);
1633
1634	/* Enter VMX root mode. */
1635	int rc = VMXEnable(HCPhysCpuPage);
1636	if (RT_FAILURE(rc))
1637	{
1638	if (!(uOldCr4 & X86_CR4_VMXE))
1639	SUPR0ChangeCR4(0 /* fOrMask */, ~X86_CR4_VMXE);
1640
1641	if (pVM)
1642	pVM->hm.s.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
1643	}
1644
1645	/* Restore interrupts. */
1646	ASMSetFlags(fEFlags);
1647	return rc;
1648	}
1649
1650
1651	/**
1652	* Exits VMX root mode operation on the current CPU.
1653	*
1654	* @returns VBox status code.
1655	*/
1656	static int hmR0VmxLeaveRootMode(void)
1657	{
1658	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1659
1660	/* Paranoid: Disable interrupts as, in theory, interrupts handlers might mess with CR4. */
1661	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1662
1663	/* If we're for some reason not in VMX root mode, then don't leave it. */
1664	RTCCUINTREG const uHostCr4 = ASMGetCR4();
1665
1666	int rc;
1667	if (uHostCr4 & X86_CR4_VMXE)
1668	{
1669	/* Exit VMX root mode and clear the VMX bit in CR4. */
1670	VMXDisable();
1671	SUPR0ChangeCR4(0 /* fOrMask */, ~X86_CR4_VMXE);
1672	rc = VINF_SUCCESS;
1673	}
1674	else
1675	rc = VERR_VMX_NOT_IN_VMX_ROOT_MODE;
1676
1677	/* Restore interrupts. */
1678	ASMSetFlags(fEFlags);
1679	return rc;
1680	}
1681
1682
1683	/**
1684	* Allocates and maps a physically contiguous page. The allocated page is
1685	* zero'd out (used by various VT-x structures).
1686	*
1687	* @returns IPRT status code.
1688	* @param pMemObj Pointer to the ring-0 memory object.
1689	* @param ppVirt Where to store the virtual address of the allocation.
1690	* @param pHCPhys Where to store the physical address of the allocation.
1691	*/
1692	static int hmR0VmxPageAllocZ(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys)
1693	{
1694	AssertPtr(pMemObj);
1695	AssertPtr(ppVirt);
1696	AssertPtr(pHCPhys);
1697	int rc = RTR0MemObjAllocCont(pMemObj, X86_PAGE_4K_SIZE, false /* fExecutable */);
1698	if (RT_FAILURE(rc))
1699	return rc;
1700	ppVirt = RTR0MemObjAddress(pMemObj);
1701	pHCPhys = RTR0MemObjGetPagePhysAddr(pMemObj, 0 /* iPage */);
1702	ASMMemZero32(*ppVirt, X86_PAGE_4K_SIZE);
1703	return VINF_SUCCESS;
1704	}
1705
1706
1707	/**
1708	* Frees and unmaps an allocated, physical page.
1709	*
1710	* @param pMemObj Pointer to the ring-0 memory object.
1711	* @param ppVirt Where to re-initialize the virtual address of allocation as
1712	* 0.
1713	* @param pHCPhys Where to re-initialize the physical address of the
1714	* allocation as 0.
1715	*/
1716	static void hmR0VmxPageFree(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys)
1717	{
1718	AssertPtr(pMemObj);
1719	AssertPtr(ppVirt);
1720	AssertPtr(pHCPhys);
1721	/* NULL is valid, accepted and ignored by the free function below. */
1722	RTR0MemObjFree(pMemObj, true / fFreeMappings */);
1723	*pMemObj = NIL_RTR0MEMOBJ;
1724	*ppVirt = NULL;
1725	*pHCPhys = NIL_RTHCPHYS;
1726	}
1727
1728
1729	/**
1730	* Initializes a VMCS info. object.
1731	*
1732	* @param pVmcsInfo The VMCS info. object.
1733	*/
1734	static void hmR0VmxInitVmcsInfo(PVMXVMCSINFO pVmcsInfo)
1735	{
1736	memset(pVmcsInfo, 0, sizeof(*pVmcsInfo));
1737
1738	Assert(pVmcsInfo->hMemObjVmcs == NIL_RTR0MEMOBJ);
1739	Assert(pVmcsInfo->hMemObjShadowVmcs == NIL_RTR0MEMOBJ);
1740	Assert(pVmcsInfo->hMemObjMsrBitmap == NIL_RTR0MEMOBJ);
1741	Assert(pVmcsInfo->hMemObjGuestMsrLoad == NIL_RTR0MEMOBJ);
1742	Assert(pVmcsInfo->hMemObjGuestMsrStore == NIL_RTR0MEMOBJ);
1743	Assert(pVmcsInfo->hMemObjHostMsrLoad == NIL_RTR0MEMOBJ);
1744	pVmcsInfo->HCPhysVmcs = NIL_RTHCPHYS;
1745	pVmcsInfo->HCPhysShadowVmcs = NIL_RTHCPHYS;
1746	pVmcsInfo->HCPhysMsrBitmap = NIL_RTHCPHYS;
1747	pVmcsInfo->HCPhysGuestMsrLoad = NIL_RTHCPHYS;
1748	pVmcsInfo->HCPhysGuestMsrStore = NIL_RTHCPHYS;
1749	pVmcsInfo->HCPhysHostMsrLoad = NIL_RTHCPHYS;
1750	pVmcsInfo->HCPhysVirtApic = NIL_RTHCPHYS;
1751	pVmcsInfo->HCPhysEPTP = NIL_RTHCPHYS;
1752	pVmcsInfo->u64VmcsLinkPtr = NIL_RTHCPHYS;
1753	pVmcsInfo->idHostCpu = NIL_RTCPUID;
1754	}
1755
1756
1757	/**
1758	* Frees the VT-x structures for a VMCS info. object.
1759	*
1760	* @param pVM The cross context VM structure.
1761	* @param pVmcsInfo The VMCS info. object.
1762	*/
1763	static void hmR0VmxFreeVmcsInfo(PVM pVM, PVMXVMCSINFO pVmcsInfo)
1764	{
1765	hmR0VmxPageFree(&pVmcsInfo->hMemObjVmcs, &pVmcsInfo->pvVmcs, &pVmcsInfo->HCPhysVmcs);
1766
1767	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1768	if (pVM->hm.s.vmx.fUseVmcsShadowing)
1769	hmR0VmxPageFree(&pVmcsInfo->hMemObjShadowVmcs, &pVmcsInfo->pvShadowVmcs, &pVmcsInfo->HCPhysShadowVmcs);
1770	#endif
1771
1772	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
1773	hmR0VmxPageFree(&pVmcsInfo->hMemObjMsrBitmap, &pVmcsInfo->pvMsrBitmap, &pVmcsInfo->HCPhysMsrBitmap);
1774
1775	hmR0VmxPageFree(&pVmcsInfo->hMemObjHostMsrLoad, &pVmcsInfo->pvHostMsrLoad, &pVmcsInfo->HCPhysHostMsrLoad);
1776	hmR0VmxPageFree(&pVmcsInfo->hMemObjGuestMsrLoad, &pVmcsInfo->pvGuestMsrLoad, &pVmcsInfo->HCPhysGuestMsrLoad);
1777	hmR0VmxPageFree(&pVmcsInfo->hMemObjGuestMsrStore, &pVmcsInfo->pvGuestMsrStore, &pVmcsInfo->HCPhysGuestMsrStore);
1778
1779	hmR0VmxInitVmcsInfo(pVmcsInfo);
1780	}
1781
1782
1783	/**
1784	* Allocates the VT-x structures for a VMCS info. object.
1785	*
1786	* @returns VBox status code.
1787	* @param pVCpu The cross context virtual CPU structure.
1788	* @param pVmcsInfo The VMCS info. object.
1789	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
1790	*/
1791	static int hmR0VmxAllocVmcsInfo(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
1792	{
1793	PVM pVM = pVCpu->CTX_SUFF(pVM);
1794
1795	/* Allocate the guest VM control structure (VMCS). */
1796	int rc = hmR0VmxPageAllocZ(&pVmcsInfo->hMemObjVmcs, &pVmcsInfo->pvVmcs, &pVmcsInfo->HCPhysVmcs);
1797	if (RT_SUCCESS(rc))
1798	{
1799	if (!fIsNstGstVmcs)
1800	{
1801	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1802	if (pVM->hm.s.vmx.fUseVmcsShadowing)
1803	rc = hmR0VmxPageAllocZ(&pVmcsInfo->hMemObjShadowVmcs, &pVmcsInfo->pvShadowVmcs, &pVmcsInfo->HCPhysShadowVmcs);
1804	#endif
1805	if (RT_SUCCESS(rc))
1806	{
1807	/* Get the allocated virtual-APIC page from the virtual APIC device. */
1808	if ( PDMHasApic(pVCpu->CTX_SUFF(pVM))
1809	&& (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW))
1810	{
1811	rc = APICGetApicPageForCpu(pVCpu, &pVmcsInfo->HCPhysVirtApic, (PRTR0PTR)&pVmcsInfo->pbVirtApic,
1812	NULL /* pR3Ptr /, NULL / pRCPtr */);
1813	}
1814	}
1815	}
1816	else
1817	{
1818	/* We don't yet support exposing VMCS shadowing to the guest. */
1819	Assert(pVmcsInfo->HCPhysShadowVmcs == NIL_RTHCPHYS);
1820	Assert(!pVmcsInfo->pvShadowVmcs);
1821
1822	/* Get the allocated virtual-APIC page from CPUM. */
1823	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW)
1824	{
1825	/** @todo NSTVMX: Get rid of this. There is no need to allocate a separate HC
1826	* page for this. Use the one provided by the nested-guest directly. */
1827	pVmcsInfo->pbVirtApic = (uint8_t *)CPUMGetGuestVmxVirtApicPage(pVCpu, &pVCpu->cpum.GstCtx,
1828	&pVmcsInfo->HCPhysVirtApic);
1829	Assert(pVmcsInfo->pbVirtApic);
1830	Assert(pVmcsInfo->HCPhysVirtApic && pVmcsInfo->HCPhysVirtApic != NIL_RTHCPHYS);
1831	}
1832	}
1833
1834	if (RT_SUCCESS(rc))
1835	{
1836	/*
1837	* Allocate the MSR-bitmap if supported by the CPU. The MSR-bitmap is for
1838	* transparent accesses of specific MSRs.
1839	*
1840	* If the condition for enabling MSR bitmaps changes here, don't forget to
1841	* update HMIsMsrBitmapActive().
1842	*
1843	* We don't share MSR bitmaps between the guest and nested-guest as we then
1844	* don't need to care about carefully restoring the guest MSR bitmap.
1845	* The guest visible nested-guest MSR bitmap needs to remain unchanged.
1846	* Hence, allocate a separate MSR bitmap for the guest and nested-guest.
1847	* We also don't need to re-initialize the nested-guest MSR bitmap here as
1848	* we do that later while merging VMCS.
1849	*/
1850	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
1851	{
1852	rc = hmR0VmxPageAllocZ(&pVmcsInfo->hMemObjMsrBitmap, &pVmcsInfo->pvMsrBitmap, &pVmcsInfo->HCPhysMsrBitmap);
1853	if ( RT_SUCCESS(rc)
1854	&& !fIsNstGstVmcs)
1855	ASMMemFill32(pVmcsInfo->pvMsrBitmap, X86_PAGE_4K_SIZE, UINT32_C(0xffffffff));
1856	}
1857
1858	if (RT_SUCCESS(rc))
1859	{
1860	/*
1861	* Allocate the VM-entry MSR-load area for the guest MSRs.
1862	*
1863	* Similar to MSR-bitmaps, we do not share the auto MSR-load/store are between
1864	* the guest and nested-guest.
1865	*/
1866	rc = hmR0VmxPageAllocZ(&pVmcsInfo->hMemObjGuestMsrLoad, &pVmcsInfo->pvGuestMsrLoad,
1867	&pVmcsInfo->HCPhysGuestMsrLoad);
1868	if (RT_SUCCESS(rc))
1869	{
1870	/*
1871	* We use the same page for VM-entry MSR-load and VM-exit MSR store areas.
1872	* These contain the guest MSRs to load on VM-entry and store on VM-exit.
1873	*/
1874	Assert(pVmcsInfo->hMemObjGuestMsrStore == NIL_RTR0MEMOBJ);
1875	pVmcsInfo->pvGuestMsrStore = pVmcsInfo->pvGuestMsrLoad;
1876	pVmcsInfo->HCPhysGuestMsrStore = pVmcsInfo->HCPhysGuestMsrLoad;
1877
1878	/* Allocate the VM-exit MSR-load page for the host MSRs. */
1879	rc = hmR0VmxPageAllocZ(&pVmcsInfo->hMemObjHostMsrLoad, &pVmcsInfo->pvHostMsrLoad,
1880	&pVmcsInfo->HCPhysHostMsrLoad);
1881	}
1882	}
1883	}
1884	}
1885
1886	return rc;
1887	}
1888
1889
1890	/**
1891	* Free all VT-x structures for the VM.
1892	*
1893	* @returns IPRT status code.
1894	* @param pVM The cross context VM structure.
1895	*/
1896	static void hmR0VmxStructsFree(PVM pVM)
1897	{
1898	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
1899	hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch);
1900	#endif
1901	hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess, &pVM->hm.s.vmx.HCPhysApicAccess);
1902
1903	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1904	if (pVM->hm.s.vmx.fUseVmcsShadowing)
1905	{
1906	RTMemFree(pVM->hm.s.vmx.paShadowVmcsFields);
1907	RTMemFree(pVM->hm.s.vmx.paShadowVmcsRoFields);
1908	hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjVmreadBitmap, &pVM->hm.s.vmx.pvVmreadBitmap, &pVM->hm.s.vmx.HCPhysVmreadBitmap);
1909	hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjVmwriteBitmap, &pVM->hm.s.vmx.pvVmwriteBitmap, &pVM->hm.s.vmx.HCPhysVmwriteBitmap);
1910	}
1911	#endif
1912
1913	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
1914	{
1915	PVMCPU pVCpu = &pVM->aCpus[idCpu];
1916	PVMXVMCSINFO pVmcsInfo = &pVCpu->hm.s.vmx.VmcsInfo;
1917	hmR0VmxFreeVmcsInfo(pVM, pVmcsInfo);
1918	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1919	if (pVM->cpum.ro.GuestFeatures.fVmx)
1920	{
1921	pVmcsInfo = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
1922	hmR0VmxFreeVmcsInfo(pVM, pVmcsInfo);
1923	}
1924	#endif
1925	}
1926	}
1927
1928
1929	/**
1930	* Allocate all VT-x structures for the VM.
1931	*
1932	* @returns IPRT status code.
1933	* @param pVM The cross context VM structure.
1934	*/
1935	static int hmR0VmxStructsAlloc(PVM pVM)
1936	{
1937	/*
1938	* Sanity check the VMCS size reported by the CPU as we assume 4KB allocations.
1939	* The VMCS size cannot be more than 4096 bytes.
1940	*
1941	* See Intel spec. Appendix A.1 "Basic VMX Information".
1942	*/
1943	uint32_t const cbVmcs = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_SIZE);
1944	if (cbVmcs <= X86_PAGE_4K_SIZE)
1945	{ /* likely */ }
1946	else
1947	{
1948	pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_INVALID_VMCS_SIZE;
1949	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
1950	}
1951
1952	/*
1953	* Initialize/check members up-front so we can cleanup en masse on allocation failures.
1954	*/
1955	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
1956	Assert(pVM->hm.s.vmx.hMemObjScratch == NIL_RTR0MEMOBJ);
1957	Assert(pVM->hm.s.vmx.pbScratch == NULL);
1958	pVM->hm.s.vmx.HCPhysScratch = NIL_RTHCPHYS;
1959	#endif
1960
1961	Assert(pVM->hm.s.vmx.hMemObjApicAccess == NIL_RTR0MEMOBJ);
1962	Assert(pVM->hm.s.vmx.pbApicAccess == NULL);
1963	pVM->hm.s.vmx.HCPhysApicAccess = NIL_RTHCPHYS;
1964
1965	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
1966	{
1967	PVMCPU pVCpu = &pVM->aCpus[idCpu];
1968	hmR0VmxInitVmcsInfo(&pVCpu->hm.s.vmx.VmcsInfo);
1969	hmR0VmxInitVmcsInfo(&pVCpu->hm.s.vmx.VmcsInfoNstGst);
1970	}
1971
1972	/*
1973	* Allocate per-VM VT-x structures.
1974	*/
1975	int rc = VINF_SUCCESS;
1976	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
1977	/* Allocate crash-dump magic scratch page. */
1978	rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch);
1979	if (RT_FAILURE(rc))
1980	{
1981	hmR0VmxStructsFree(pVM);
1982	return rc;
1983	}
1984	strcpy((char *)pVM->hm.s.vmx.pbScratch, "SCRATCH Magic");
1985	(uint64_t )(pVM->hm.s.vmx.pbScratch + 16) = UINT64_C(0xdeadbeefdeadbeef);
1986	#endif
1987
1988	/* Allocate the APIC-access page for trapping APIC accesses from the guest. */
1989	if (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
1990	{
1991	rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess,
1992	&pVM->hm.s.vmx.HCPhysApicAccess);
1993	if (RT_FAILURE(rc))
1994	{
1995	hmR0VmxStructsFree(pVM);
1996	return rc;
1997	}
1998	}
1999
2000	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2001	/* Allocate the shadow VMCS fields array, VMREAD, VMWRITE bitmaps.. */
2002	if (pVM->hm.s.vmx.fUseVmcsShadowing)
2003	{
2004	Assert(!pVM->hm.s.vmx.cShadowVmcsFields);
2005	Assert(!pVM->hm.s.vmx.cShadowVmcsRoFields);
2006	pVM->hm.s.vmx.paShadowVmcsFields = (uint32_t *)RTMemAllocZ(sizeof(g_aVmcsFields));
2007	pVM->hm.s.vmx.paShadowVmcsRoFields = (uint32_t *)RTMemAllocZ(sizeof(g_aVmcsFields));
2008	if (RT_LIKELY( pVM->hm.s.vmx.paShadowVmcsFields
2009	&& pVM->hm.s.vmx.paShadowVmcsRoFields))
2010	{
2011	rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjVmreadBitmap, &pVM->hm.s.vmx.pvVmreadBitmap,
2012	&pVM->hm.s.vmx.HCPhysVmreadBitmap);
2013	if (RT_SUCCESS(rc))
2014	{
2015	rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjVmwriteBitmap, &pVM->hm.s.vmx.pvVmwriteBitmap,
2016	&pVM->hm.s.vmx.HCPhysVmwriteBitmap);
2017	}
2018	}
2019	else
2020	rc = VERR_NO_MEMORY;
2021
2022	if (RT_FAILURE(rc))
2023	{
2024	hmR0VmxStructsFree(pVM);
2025	return rc;
2026	}
2027	}
2028	#endif
2029
2030	/*
2031	* Initialize per-VCPU VT-x structures.
2032	*/
2033	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
2034	{
2035	/* Allocate the guest VMCS structures. */
2036	PVMCPU pVCpu = &pVM->aCpus[idCpu];
2037	rc = hmR0VmxAllocVmcsInfo(pVCpu, &pVCpu->hm.s.vmx.VmcsInfo, false /* fIsNstGstVmcs */);
2038	if (RT_SUCCESS(rc))
2039	{
2040	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2041	/* Allocate the nested-guest VMCS structures, when the VMX feature is exposed to the guest. */
2042	if (pVM->cpum.ro.GuestFeatures.fVmx)
2043	{
2044	rc = hmR0VmxAllocVmcsInfo(pVCpu, &pVCpu->hm.s.vmx.VmcsInfoNstGst, true /* fIsNstGstVmcs */);
2045	if (RT_SUCCESS(rc))
2046	{ /* likely */ }
2047	else
2048	break;
2049	}
2050	#endif
2051	}
2052	else
2053	break;
2054	}
2055
2056	if (RT_FAILURE(rc))
2057	{
2058	hmR0VmxStructsFree(pVM);
2059	return rc;
2060	}
2061
2062	return VINF_SUCCESS;
2063	}
2064
2065	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2066	/**
2067	* Returns whether an MSR at the given MSR-bitmap offset is intercepted or not.
2068	*
2069	* @returns @c true if the MSR is intercepted, @c false otherwise.
2070	* @param pvMsrBitmap The MSR bitmap.
2071	* @param offMsr The MSR byte offset.
2072	* @param iBit The bit offset from the byte offset.
2073	*/
2074	DECLINLINE(bool) hmR0VmxIsMsrBitSet(const void *pvMsrBitmap, uint16_t offMsr, int32_t iBit)
2075	{
2076	uint8_t const * const pbMsrBitmap = (uint8_t const * const)pvMsrBitmap;
2077	Assert(pbMsrBitmap);
2078	Assert(offMsr + (iBit >> 3) <= X86_PAGE_4K_SIZE);
2079	return ASMBitTest(pbMsrBitmap + offMsr, iBit);
2080	}
2081	#endif
2082
2083	/**
2084	* Sets the permission bits for the specified MSR in the given MSR bitmap.
2085	*
2086	* If the passed VMCS is a nested-guest VMCS, this function ensures that the
2087	* read/write intercept is cleared from the MSR bitmap used for hardware-assisted
2088	* VMX execution of the nested-guest, only if nested-guest is also not intercepting
2089	* the read/write access of this MSR.
2090	*
2091	* @param pVCpu The cross context virtual CPU structure.
2092	* @param pVmcsInfo The VMCS info. object.
2093	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2094	* @param idMsr The MSR value.
2095	* @param fMsrpm The MSR permissions (see VMXMSRPM_XXX). This must
2096	* include both a read -and- a write permission!
2097	*
2098	* @sa CPUMGetVmxMsrPermission.
2099	* @remarks Can be called with interrupts disabled.
2100	*/
2101	static void hmR0VmxSetMsrPermission(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs, uint32_t idMsr, uint32_t fMsrpm)
2102	{
2103	uint8_t pbMsrBitmap = (uint8_t )pVmcsInfo->pvMsrBitmap;
2104	Assert(pbMsrBitmap);
2105	Assert(VMXMSRPM_IS_FLAG_VALID(fMsrpm));
2106
2107	/*
2108	* MSR-bitmap Layout:
2109	* Byte index MSR range Interpreted as
2110	* 0x000 - 0x3ff 0x00000000 - 0x00001fff Low MSR read bits.
2111	* 0x400 - 0x7ff 0xc0000000 - 0xc0001fff High MSR read bits.
2112	* 0x800 - 0xbff 0x00000000 - 0x00001fff Low MSR write bits.
2113	* 0xc00 - 0xfff 0xc0000000 - 0xc0001fff High MSR write bits.
2114	*
2115	* A bit corresponding to an MSR within the above range causes a VM-exit
2116	* if the bit is 1 on executions of RDMSR/WRMSR. If an MSR falls out of
2117	* the MSR range, it always cause a VM-exit.
2118	*
2119	* See Intel spec. 24.6.9 "MSR-Bitmap Address".
2120	*/
2121	uint16_t const offBitmapRead = 0;
2122	uint16_t const offBitmapWrite = 0x800;
2123	uint16_t offMsr;
2124	int32_t iBit;
2125	if (idMsr <= UINT32_C(0x00001fff))
2126	{
2127	offMsr = 0;
2128	iBit = idMsr;
2129	}
2130	else if (idMsr - UINT32_C(0xc0000000) <= UINT32_C(0x00001fff))
2131	{
2132	offMsr = 0x400;
2133	iBit = idMsr - UINT32_C(0xc0000000);
2134	}
2135	else
2136	AssertMsgFailedReturnVoid(("Invalid MSR %#RX32\n", idMsr));
2137
2138	/*
2139	* Set the MSR read permission.
2140	*/
2141	uint16_t const offMsrRead = offBitmapRead + offMsr;
2142	Assert(offMsrRead + (iBit >> 3) < offBitmapWrite);
2143	if (fMsrpm & VMXMSRPM_ALLOW_RD)
2144	{
2145	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2146	bool const fClear = !fIsNstGstVmcs ? true
2147	: !hmR0VmxIsMsrBitSet(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), offMsrRead, iBit);
2148	#else
2149	RT_NOREF2(pVCpu, fIsNstGstVmcs);
2150	bool const fClear = true;
2151	#endif
2152	if (fClear)
2153	ASMBitClear(pbMsrBitmap + offMsrRead, iBit);
2154	}
2155	else
2156	ASMBitSet(pbMsrBitmap + offMsrRead, iBit);
2157
2158	/*
2159	* Set the MSR write permission.
2160	*/
2161	uint16_t const offMsrWrite = offBitmapWrite + offMsr;
2162	Assert(offMsrWrite + (iBit >> 3) < X86_PAGE_4K_SIZE);
2163	if (fMsrpm & VMXMSRPM_ALLOW_WR)
2164	{
2165	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2166	bool const fClear = !fIsNstGstVmcs ? true
2167	: !hmR0VmxIsMsrBitSet(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), offMsrWrite, iBit);
2168	#else
2169	RT_NOREF2(pVCpu, fIsNstGstVmcs);
2170	bool const fClear = true;
2171	#endif
2172	if (fClear)
2173	ASMBitClear(pbMsrBitmap + offMsrWrite, iBit);
2174	}
2175	else
2176	ASMBitSet(pbMsrBitmap + offMsrWrite, iBit);
2177	}
2178
2179
2180	/**
2181	* Updates the VMCS with the number of effective MSRs in the auto-load/store MSR
2182	* area.
2183	*
2184	* @returns VBox status code.
2185	* @param pVCpu The cross context virtual CPU structure.
2186	* @param pVmcsInfo The VMCS info. object.
2187	* @param cMsrs The number of MSRs.
2188	*/
2189	static int hmR0VmxSetAutoLoadStoreMsrCount(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo, uint32_t cMsrs)
2190	{
2191	/* Shouldn't ever happen but there -is- a number. We're well within the recommended 512. */
2192	uint32_t const cMaxSupportedMsrs = VMX_MISC_MAX_MSRS(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.u64Misc);
2193	if (RT_LIKELY(cMsrs < cMaxSupportedMsrs))
2194	{
2195	/* Commit the MSR counts to the VMCS and update the cache. */
2196	if (pVmcsInfo->cEntryMsrLoad != cMsrs)
2197	{
2198	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, cMsrs);
2199	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, cMsrs);
2200	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, cMsrs);
2201	AssertRCReturn(rc, rc);
2202
2203	pVmcsInfo->cEntryMsrLoad = cMsrs;
2204	pVmcsInfo->cExitMsrStore = cMsrs;
2205	pVmcsInfo->cExitMsrLoad = cMsrs;
2206	}
2207	return VINF_SUCCESS;
2208	}
2209
2210	LogRel(("Auto-load/store MSR count exceeded! cMsrs=%u MaxSupported=%u\n", cMsrs, cMaxSupportedMsrs));
2211	pVCpu->hm.s.u32HMError = VMX_UFC_INSUFFICIENT_GUEST_MSR_STORAGE;
2212	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2213	}
2214
2215
2216	/**
2217	* Adds a new (or updates the value of an existing) guest/host MSR
2218	* pair to be swapped during the world-switch as part of the
2219	* auto-load/store MSR area in the VMCS.
2220	*
2221	* @returns VBox status code.
2222	* @param pVCpu The cross context virtual CPU structure.
2223	* @param pVmxTransient The VMX-transient structure.
2224	* @param idMsr The MSR.
2225	* @param uGuestMsrValue Value of the guest MSR.
2226	* @param fSetReadWrite Whether to set the guest read/write access of this
2227	* MSR (thus not causing a VM-exit).
2228	* @param fUpdateHostMsr Whether to update the value of the host MSR if
2229	* necessary.
2230	*/
2231	static int hmR0VmxAddAutoLoadStoreMsr(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t idMsr, uint64_t uGuestMsrValue,
2232	bool fSetReadWrite, bool fUpdateHostMsr)
2233	{
2234	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
2235	bool const fIsNstGstVmcs = pVmxTransient->fIsNestedGuest;
2236	PVMXAUTOMSR pGuestMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2237	uint32_t cMsrs = pVmcsInfo->cEntryMsrLoad;
2238	uint32_t i;
2239
2240	/* Paranoia. */
2241	Assert(pGuestMsrLoad);
2242
2243	LogFlowFunc(("pVCpu=%p idMsr=%#RX32 uGestMsrValue=%#RX64\n", pVCpu, idMsr, uGuestMsrValue));
2244
2245	/* Check if the MSR already exists in the VM-entry MSR-load area. */
2246	for (i = 0; i < cMsrs; i++)
2247	{
2248	if (pGuestMsrLoad[i].u32Msr == idMsr)
2249	break;
2250	}
2251
2252	bool fAdded = false;
2253	if (i == cMsrs)
2254	{
2255	/* The MSR does not exist, bump the MSR count to make room for the new MSR. */
2256	++cMsrs;
2257	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, pVmcsInfo, cMsrs);
2258	AssertMsgRCReturn(rc, ("Insufficient space to add MSR to VM-entry MSR-load/store area %u\n", idMsr), rc);
2259
2260	/* Set the guest to read/write this MSR without causing VM-exits. */
2261	if ( fSetReadWrite
2262	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS))
2263	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, fIsNstGstVmcs, idMsr, VMXMSRPM_ALLOW_RD_WR);
2264
2265	Log4Func(("Added MSR %#RX32, cMsrs=%u\n", idMsr, cMsrs));
2266	fAdded = true;
2267	}
2268
2269	/* Update the MSR value for the newly added or already existing MSR. */
2270	pGuestMsrLoad[i].u32Msr = idMsr;
2271	pGuestMsrLoad[i].u64Value = uGuestMsrValue;
2272
2273	/* Create the corresponding slot in the VM-exit MSR-store area if we use a different page. */
2274	if (hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo))
2275	{
2276	PVMXAUTOMSR pGuestMsrStore = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2277	pGuestMsrStore[i].u32Msr = idMsr;
2278	pGuestMsrStore[i].u64Value = uGuestMsrValue;
2279	}
2280
2281	/* Update the corresponding slot in the host MSR area. */
2282	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2283	Assert(pHostMsr != pVmcsInfo->pvGuestMsrLoad);
2284	Assert(pHostMsr != pVmcsInfo->pvGuestMsrStore);
2285	pHostMsr[i].u32Msr = idMsr;
2286
2287	/*
2288	* Only if the caller requests to update the host MSR value AND we've newly added the
2289	* MSR to the host MSR area do we actually update the value. Otherwise, it will be
2290	* updated by hmR0VmxUpdateAutoLoadHostMsrs().
2291	*
2292	* We do this for performance reasons since reading MSRs may be quite expensive.
2293	*/
2294	if (fAdded)
2295	{
2296	if (fUpdateHostMsr)
2297	{
2298	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2299	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2300	pHostMsr[i].u64Value = ASMRdMsr(idMsr);
2301	}
2302	else
2303	{
2304	/* Someone else can do the work. */
2305	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = false;
2306	}
2307	}
2308	return VINF_SUCCESS;
2309	}
2310
2311
2312	/**
2313	* Removes a guest/host MSR pair to be swapped during the world-switch from the
2314	* auto-load/store MSR area in the VMCS.
2315	*
2316	* @returns VBox status code.
2317	* @param pVCpu The cross context virtual CPU structure.
2318	* @param pVmxTransient The VMX-transient structure.
2319	* @param idMsr The MSR.
2320	*/
2321	static int hmR0VmxRemoveAutoLoadStoreMsr(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t idMsr)
2322	{
2323	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
2324	bool const fIsNstGstVmcs = pVmxTransient->fIsNestedGuest;
2325	PVMXAUTOMSR pGuestMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2326	uint32_t cMsrs = pVmcsInfo->cEntryMsrLoad;
2327
2328	LogFlowFunc(("pVCpu=%p idMsr=%#RX32\n", pVCpu, idMsr));
2329
2330	for (uint32_t i = 0; i < cMsrs; i++)
2331	{
2332	/* Find the MSR. */
2333	if (pGuestMsrLoad[i].u32Msr == idMsr)
2334	{
2335	/*
2336	* If it's the last MSR, we only need to reduce the MSR count.
2337	* If it's -not- the last MSR, copy the last MSR in place of it and reduce the MSR count.
2338	*/
2339	if (i < cMsrs - 1)
2340	{
2341	/* Remove it from the VM-entry MSR-load area. */
2342	pGuestMsrLoad[i].u32Msr = pGuestMsrLoad[cMsrs - 1].u32Msr;
2343	pGuestMsrLoad[i].u64Value = pGuestMsrLoad[cMsrs - 1].u64Value;
2344
2345	/* Remove it from the VM-exit MSR-store area if it's in a different page. */
2346	if (hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo))
2347	{
2348	PVMXAUTOMSR pGuestMsrStore = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2349	Assert(pGuestMsrStore[i].u32Msr == idMsr);
2350	pGuestMsrStore[i].u32Msr = pGuestMsrStore[cMsrs - 1].u32Msr;
2351	pGuestMsrStore[i].u64Value = pGuestMsrStore[cMsrs - 1].u64Value;
2352	}
2353
2354	/* Remove it from the VM-exit MSR-load area. */
2355	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2356	Assert(pHostMsr[i].u32Msr == idMsr);
2357	pHostMsr[i].u32Msr = pHostMsr[cMsrs - 1].u32Msr;
2358	pHostMsr[i].u64Value = pHostMsr[cMsrs - 1].u64Value;
2359	}
2360
2361	/* Reduce the count to reflect the removed MSR and bail. */
2362	--cMsrs;
2363	break;
2364	}
2365	}
2366
2367	/* Update the VMCS if the count changed (meaning the MSR was found and removed). */
2368	if (cMsrs != pVmcsInfo->cEntryMsrLoad)
2369	{
2370	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, pVmcsInfo, cMsrs);
2371	AssertRCReturn(rc, rc);
2372
2373	/* We're no longer swapping MSRs during the world-switch, intercept guest read/writes to them. */
2374	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
2375	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, fIsNstGstVmcs, idMsr, VMXMSRPM_EXIT_RD \| VMXMSRPM_EXIT_WR);
2376
2377	Log4Func(("Removed MSR %#RX32, cMsrs=%u\n", idMsr, cMsrs));
2378	return VINF_SUCCESS;
2379	}
2380
2381	return VERR_NOT_FOUND;
2382	}
2383
2384
2385	/**
2386	* Checks if the specified guest MSR is part of the VM-entry MSR-load area.
2387	*
2388	* @returns @c true if found, @c false otherwise.
2389	* @param pVmcsInfo The VMCS info. object.
2390	* @param idMsr The MSR to find.
2391	*/
2392	static bool hmR0VmxIsAutoLoadGuestMsr(PCVMXVMCSINFO pVmcsInfo, uint32_t idMsr)
2393	{
2394	PCVMXAUTOMSR pMsrs = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2395	uint32_t const cMsrs = pVmcsInfo->cEntryMsrLoad;
2396	Assert(pMsrs);
2397	Assert(sizeof(pMsrs) cMsrs <= X86_PAGE_4K_SIZE);
2398	for (uint32_t i = 0; i < cMsrs; i++)
2399	{
2400	if (pMsrs[i].u32Msr == idMsr)
2401	return true;
2402	}
2403	return false;
2404	}
2405
2406
2407	/**
2408	* Updates the value of all host MSRs in the VM-exit MSR-load area.
2409	*
2410	* @param pVCpu The cross context virtual CPU structure.
2411	* @param pVmcsInfo The VMCS info. object.
2412	*
2413	* @remarks No-long-jump zone!!!
2414	*/
2415	static void hmR0VmxUpdateAutoLoadHostMsrs(PCVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo)
2416	{
2417	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2418
2419	PVMXAUTOMSR pHostMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2420	uint32_t const cMsrs = pVmcsInfo->cExitMsrLoad;
2421	Assert(pHostMsrLoad);
2422	Assert(sizeof(pHostMsrLoad) cMsrs <= X86_PAGE_4K_SIZE);
2423	LogFlowFunc(("pVCpu=%p cMsrs=%u\n", pVCpu, cMsrs));
2424	for (uint32_t i = 0; i < cMsrs; i++)
2425	{
2426	/*
2427	* Performance hack for the host EFER MSR. We use the cached value rather than re-read it.
2428	* Strict builds will catch mismatches in hmR0VmxCheckAutoLoadStoreMsrs(). See @bugref{7368}.
2429	*/
2430	if (pHostMsrLoad[i].u32Msr == MSR_K6_EFER)
2431	pHostMsrLoad[i].u64Value = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostMsrEfer;
2432	else
2433	pHostMsrLoad[i].u64Value = ASMRdMsr(pHostMsrLoad[i].u32Msr);
2434	}
2435	}
2436
2437
2438	/**
2439	* Saves a set of host MSRs to allow read/write passthru access to the guest and
2440	* perform lazy restoration of the host MSRs while leaving VT-x.
2441	*
2442	* @param pVCpu The cross context virtual CPU structure.
2443	*
2444	* @remarks No-long-jump zone!!!
2445	*/
2446	static void hmR0VmxLazySaveHostMsrs(PVMCPU pVCpu)
2447	{
2448	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2449
2450	/*
2451	* Note: If you're adding MSRs here, make sure to update the MSR-bitmap accesses in hmR0VmxSetupVmcsProcCtls().
2452	*/
2453	if (!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST))
2454	{
2455	Assert(!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)); /* Guest MSRs better not be loaded now. */
2456	#if HC_ARCH_BITS == 64
2457	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
2458	{
2459	pVCpu->hm.s.vmx.u64HostMsrLStar = ASMRdMsr(MSR_K8_LSTAR);
2460	pVCpu->hm.s.vmx.u64HostMsrStar = ASMRdMsr(MSR_K6_STAR);
2461	pVCpu->hm.s.vmx.u64HostMsrSfMask = ASMRdMsr(MSR_K8_SF_MASK);
2462	pVCpu->hm.s.vmx.u64HostMsrKernelGsBase = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
2463	}
2464	#endif
2465	pVCpu->hm.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_SAVED_HOST;
2466	}
2467	}
2468
2469
2470	/**
2471	* Checks whether the MSR belongs to the set of guest MSRs that we restore
2472	* lazily while leaving VT-x.
2473	*
2474	* @returns true if it does, false otherwise.
2475	* @param pVCpu The cross context virtual CPU structure.
2476	* @param idMsr The MSR to check.
2477	*/
2478	static bool hmR0VmxIsLazyGuestMsr(PCVMCPU pVCpu, uint32_t idMsr)
2479	{
2480	NOREF(pVCpu);
2481	#if HC_ARCH_BITS == 64
2482	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
2483	{
2484	switch (idMsr)
2485	{
2486	case MSR_K8_LSTAR:
2487	case MSR_K6_STAR:
2488	case MSR_K8_SF_MASK:
2489	case MSR_K8_KERNEL_GS_BASE:
2490	return true;
2491	}
2492	}
2493	#else
2494	RT_NOREF(pVCpu, idMsr);
2495	#endif
2496	return false;
2497	}
2498
2499
2500	/**
2501	* Loads a set of guests MSRs to allow read/passthru to the guest.
2502	*
2503	* The name of this function is slightly confusing. This function does NOT
2504	* postpone loading, but loads the MSR right now. "hmR0VmxLazy" is simply a
2505	* common prefix for functions dealing with "lazy restoration" of the shared
2506	* MSRs.
2507	*
2508	* @param pVCpu The cross context virtual CPU structure.
2509	*
2510	* @remarks No-long-jump zone!!!
2511	*/
2512	static void hmR0VmxLazyLoadGuestMsrs(PVMCPU pVCpu)
2513	{
2514	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2515	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2516
2517	Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
2518	#if HC_ARCH_BITS == 64
2519	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
2520	{
2521	/*
2522	* If the guest MSRs are not loaded -and- if all the guest MSRs are identical
2523	* to the MSRs on the CPU (which are the saved host MSRs, see assertion above) then
2524	* we can skip a few MSR writes.
2525	*
2526	* Otherwise, it implies either 1. they're not loaded, or 2. they're loaded but the
2527	* guest MSR values in the guest-CPU context might be different to what's currently
2528	* loaded in the CPU. In either case, we need to write the new guest MSR values to the
2529	* CPU, see @bugref{8728}.
2530	*/
2531	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
2532	if ( !(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
2533	&& pCtx->msrKERNELGSBASE == pVCpu->hm.s.vmx.u64HostMsrKernelGsBase
2534	&& pCtx->msrLSTAR == pVCpu->hm.s.vmx.u64HostMsrLStar
2535	&& pCtx->msrSTAR == pVCpu->hm.s.vmx.u64HostMsrStar
2536	&& pCtx->msrSFMASK == pVCpu->hm.s.vmx.u64HostMsrSfMask)
2537	{
2538	#ifdef VBOX_STRICT
2539	Assert(ASMRdMsr(MSR_K8_KERNEL_GS_BASE) == pCtx->msrKERNELGSBASE);
2540	Assert(ASMRdMsr(MSR_K8_LSTAR) == pCtx->msrLSTAR);
2541	Assert(ASMRdMsr(MSR_K6_STAR) == pCtx->msrSTAR);
2542	Assert(ASMRdMsr(MSR_K8_SF_MASK) == pCtx->msrSFMASK);
2543	#endif
2544	}
2545	else
2546	{
2547	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pCtx->msrKERNELGSBASE);
2548	ASMWrMsr(MSR_K8_LSTAR, pCtx->msrLSTAR);
2549	ASMWrMsr(MSR_K6_STAR, pCtx->msrSTAR);
2550	ASMWrMsr(MSR_K8_SF_MASK, pCtx->msrSFMASK);
2551	}
2552	}
2553	#endif
2554	pVCpu->hm.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_LOADED_GUEST;
2555	}
2556
2557
2558	/**
2559	* Performs lazy restoration of the set of host MSRs if they were previously
2560	* loaded with guest MSR values.
2561	*
2562	* @param pVCpu The cross context virtual CPU structure.
2563	*
2564	* @remarks No-long-jump zone!!!
2565	* @remarks The guest MSRs should have been saved back into the guest-CPU
2566	* context by hmR0VmxImportGuestState()!!!
2567	*/
2568	static void hmR0VmxLazyRestoreHostMsrs(PVMCPU pVCpu)
2569	{
2570	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2571	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2572
2573	if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
2574	{
2575	Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
2576	#if HC_ARCH_BITS == 64
2577	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
2578	{
2579	ASMWrMsr(MSR_K8_LSTAR, pVCpu->hm.s.vmx.u64HostMsrLStar);
2580	ASMWrMsr(MSR_K6_STAR, pVCpu->hm.s.vmx.u64HostMsrStar);
2581	ASMWrMsr(MSR_K8_SF_MASK, pVCpu->hm.s.vmx.u64HostMsrSfMask);
2582	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pVCpu->hm.s.vmx.u64HostMsrKernelGsBase);
2583	}
2584	#endif
2585	}
2586	pVCpu->hm.s.vmx.fLazyMsrs &= ~(VMX_LAZY_MSRS_LOADED_GUEST \| VMX_LAZY_MSRS_SAVED_HOST);
2587	}
2588
2589
2590	/**
2591	* Verifies that our cached values of the VMCS fields are all consistent with
2592	* what's actually present in the VMCS.
2593	*
2594	* @returns VBox status code.
2595	* @retval VINF_SUCCESS if all our caches match their respective VMCS fields.
2596	* @retval VERR_VMX_VMCS_FIELD_CACHE_INVALID if a cache field doesn't match the
2597	* VMCS content. HMCPU error-field is
2598	* updated, see VMX_VCI_XXX.
2599	* @param pVCpu The cross context virtual CPU structure.
2600	* @param pVmcsInfo The VMCS info. object.
2601	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2602	*/
2603	static int hmR0VmxCheckVmcsCtls(PVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
2604	{
2605	const char * const pcszVmcs = fIsNstGstVmcs ? "Nested-guest VMCS" : "VMCS";
2606
2607	uint32_t u32Val;
2608	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
2609	AssertRCReturn(rc, rc);
2610	AssertMsgReturnStmt(pVmcsInfo->u32EntryCtls == u32Val,
2611	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32EntryCtls, u32Val),
2612	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_ENTRY,
2613	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2614
2615	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val);
2616	AssertRCReturn(rc, rc);
2617	AssertMsgReturnStmt(pVmcsInfo->u32ExitCtls == u32Val,
2618	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ExitCtls, u32Val),
2619	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_EXIT,
2620	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2621
2622	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val);
2623	AssertRCReturn(rc, rc);
2624	AssertMsgReturnStmt(pVmcsInfo->u32PinCtls == u32Val,
2625	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32PinCtls, u32Val),
2626	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PIN_EXEC,
2627	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2628
2629	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val);
2630	AssertRCReturn(rc, rc);
2631	AssertMsgReturnStmt(pVmcsInfo->u32ProcCtls == u32Val,
2632	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ProcCtls, u32Val),
2633	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC,
2634	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2635
2636	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
2637	{
2638	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val);
2639	AssertRCReturn(rc, rc);
2640	AssertMsgReturnStmt(pVmcsInfo->u32ProcCtls2 == u32Val,
2641	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ProcCtls2, u32Val),
2642	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC2,
2643	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2644	}
2645
2646	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, &u32Val);
2647	AssertRCReturn(rc, rc);
2648	AssertMsgReturnStmt(pVmcsInfo->u32XcptBitmap == u32Val,
2649	("%s exception bitmap mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32XcptBitmap, u32Val),
2650	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_XCPT_BITMAP,
2651	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2652
2653	uint64_t u64Val;
2654	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, &u64Val);
2655	AssertRCReturn(rc, rc);
2656	AssertMsgReturnStmt(pVmcsInfo->u64TscOffset == u64Val,
2657	("%s TSC offset mismatch: Cache=%#RX64 VMCS=%#RX64\n", pcszVmcs, pVmcsInfo->u64TscOffset, u64Val),
2658	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_TSC_OFFSET,
2659	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2660
2661	NOREF(pcszVmcs);
2662	return VINF_SUCCESS;
2663	}
2664
2665
2666	#ifdef VBOX_STRICT
2667	/**
2668	* Verifies that our cached host EFER MSR value has not changed since we cached it.
2669	*
2670	* @param pVCpu The cross context virtual CPU structure.
2671	* @param pVmcsInfo The VMCS info. object.
2672	*/
2673	static void hmR0VmxCheckHostEferMsr(PCVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo)
2674	{
2675	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2676
2677	if (pVmcsInfo->u32ExitCtls & VMX_EXIT_CTLS_LOAD_EFER_MSR)
2678	{
2679	uint64_t const uHostEferMsr = ASMRdMsr(MSR_K6_EFER);
2680	uint64_t const uHostEferMsrCache = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostMsrEfer;
2681	uint64_t uVmcsEferMsrVmcs;
2682	int rc = VMXReadVmcs64(VMX_VMCS64_HOST_EFER_FULL, &uVmcsEferMsrVmcs);
2683	AssertRC(rc);
2684
2685	AssertMsgReturnVoid(uHostEferMsr == uVmcsEferMsrVmcs,
2686	("EFER Host/VMCS mismatch! host=%#RX64 vmcs=%#RX64\n", uHostEferMsr, uVmcsEferMsrVmcs));
2687	AssertMsgReturnVoid(uHostEferMsr == uHostEferMsrCache,
2688	("EFER Host/Cache mismatch! host=%#RX64 cache=%#RX64\n", uHostEferMsr, uHostEferMsrCache));
2689	}
2690	}
2691
2692
2693	/**
2694	* Verifies whether the guest/host MSR pairs in the auto-load/store area in the
2695	* VMCS are correct.
2696	*
2697	* @param pVCpu The cross context virtual CPU structure.
2698	* @param pVmcsInfo The VMCS info. object.
2699	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2700	*/
2701	static void hmR0VmxCheckAutoLoadStoreMsrs(PVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
2702	{
2703	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2704
2705	/* Read the various MSR-area counts from the VMCS. */
2706	uint32_t cEntryLoadMsrs;
2707	uint32_t cExitStoreMsrs;
2708	uint32_t cExitLoadMsrs;
2709	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &cEntryLoadMsrs); AssertRC(rc);
2710	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &cExitStoreMsrs); AssertRC(rc);
2711	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &cExitLoadMsrs); AssertRC(rc);
2712
2713	/* Verify all the MSR counts are the same. */
2714	Assert(cEntryLoadMsrs == cExitStoreMsrs);
2715	Assert(cExitStoreMsrs == cExitLoadMsrs);
2716	uint32_t const cMsrs = cExitLoadMsrs;
2717
2718	/* Verify the MSR counts do not exceed the maximum count supported by the hardware. */
2719	Assert(cMsrs < VMX_MISC_MAX_MSRS(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.u64Misc));
2720
2721	/* Verify the MSR counts are within the allocated page size. */
2722	Assert(sizeof(VMXAUTOMSR) * cMsrs <= X86_PAGE_4K_SIZE);
2723
2724	/* Verify the relevant contents of the MSR areas match. */
2725	PCVMXAUTOMSR pGuestMsrLoad = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2726	PCVMXAUTOMSR pGuestMsrStore = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2727	PCVMXAUTOMSR pHostMsrLoad = (PCVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2728	bool const fSeparateExitMsrStorePage = hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo);
2729	for (uint32_t i = 0; i < cMsrs; i++)
2730	{
2731	/* Verify that the MSRs are paired properly and that the host MSR has the correct value. */
2732	if (fSeparateExitMsrStorePage)
2733	{
2734	AssertMsgReturnVoid(pGuestMsrLoad->u32Msr == pGuestMsrStore->u32Msr,
2735	("GuestMsrLoad=%#RX32 GuestMsrStore=%#RX32 cMsrs=%u\n",
2736	pGuestMsrLoad->u32Msr, pGuestMsrStore->u32Msr, cMsrs));
2737	}
2738
2739	AssertMsgReturnVoid(pHostMsrLoad->u32Msr == pGuestMsrLoad->u32Msr,
2740	("HostMsrLoad=%#RX32 GuestMsrLoad=%#RX32 cMsrs=%u\n",
2741	pHostMsrLoad->u32Msr, pGuestMsrLoad->u32Msr, cMsrs));
2742
2743	uint64_t const u64Msr = ASMRdMsr(pHostMsrLoad->u32Msr);
2744	AssertMsgReturnVoid(pHostMsrLoad->u64Value == u64Msr,
2745	("u32Msr=%#RX32 VMCS Value=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
2746	pHostMsrLoad->u32Msr, pHostMsrLoad->u64Value, u64Msr, cMsrs));
2747
2748	/* Verify that cached host EFER MSR matches what's loaded the CPU. */
2749	bool const fIsEferMsr = RT_BOOL(pHostMsrLoad->u32Msr == MSR_K6_EFER);
2750	if (fIsEferMsr)
2751	{
2752	AssertMsgReturnVoid(u64Msr == pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostMsrEfer,
2753	("Cached=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
2754	pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostMsrEfer, u64Msr, cMsrs));
2755	}
2756
2757	/* Verify that the accesses are as expected in the MSR bitmap for auto-load/store MSRs. */
2758	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
2759	{
2760	uint32_t const fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, pGuestMsrLoad->u32Msr);
2761	if (fIsEferMsr)
2762	{
2763	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_EXIT_RD), ("Passthru read for EFER MSR!?\n"));
2764	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_EXIT_WR), ("Passthru write for EFER MSR!?\n"));
2765	}
2766	else
2767	{
2768	if (!fIsNstGstVmcs)
2769	{
2770	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_ALLOW_RD_WR) == VMXMSRPM_ALLOW_RD_WR,
2771	("u32Msr=%#RX32 cMsrs=%u No passthru read/write!\n", pGuestMsrLoad->u32Msr, cMsrs));
2772	}
2773	else
2774	{
2775	/*
2776	* A nested-guest VMCS must -also- allow read/write passthrough for the MSR for us to
2777	* execute a nested-guest with MSR passthrough.
2778	*
2779	* Check if the nested-guest MSR bitmap allows passthrough, and if so, assert that we
2780	* allow passthrough too.
2781	*/
2782	void const *pvMsrBitmapNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap);
2783	Assert(pvMsrBitmapNstGst);
2784	uint32_t const fMsrpmNstGst = CPUMGetVmxMsrPermission(pvMsrBitmapNstGst, pGuestMsrLoad->u32Msr);
2785	AssertMsgReturnVoid(fMsrpm == fMsrpmNstGst,
2786	("u32Msr=%#RX32 cMsrs=%u Permission mismatch fMsrpm=%#x fMsrpmNstGst=%#x!\n",
2787	pGuestMsrLoad->u32Msr, cMsrs, fMsrpm, fMsrpmNstGst));
2788	}
2789	}
2790	}
2791
2792	/* Move to the next MSR. */
2793	pHostMsrLoad++;
2794	pGuestMsrLoad++;
2795	pGuestMsrStore++;
2796	}
2797	}
2798	#endif /* VBOX_STRICT */
2799
2800
2801	/**
2802	* Flushes the TLB using EPT.
2803	*
2804	* @returns VBox status code.
2805	* @param pVCpu The cross context virtual CPU structure of the calling
2806	* EMT. Can be NULL depending on @a enmTlbFlush.
2807	* @param pVmcsInfo The VMCS info. object. Can be NULL depending on @a
2808	* enmTlbFlush.
2809	* @param enmTlbFlush Type of flush.
2810	*
2811	* @remarks Caller is responsible for making sure this function is called only
2812	* when NestedPaging is supported and providing @a enmTlbFlush that is
2813	* supported by the CPU.
2814	* @remarks Can be called with interrupts disabled.
2815	*/
2816	static void hmR0VmxFlushEpt(PVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo, VMXTLBFLUSHEPT enmTlbFlush)
2817	{
2818	uint64_t au64Descriptor[2];
2819	if (enmTlbFlush == VMXTLBFLUSHEPT_ALL_CONTEXTS)
2820	au64Descriptor[0] = 0;
2821	else
2822	{
2823	Assert(pVCpu);
2824	Assert(pVmcsInfo);
2825	au64Descriptor[0] = pVmcsInfo->HCPhysEPTP;
2826	}
2827	au64Descriptor[1] = 0; /* MBZ. Intel spec. 33.3 "VMX Instructions" */
2828
2829	int rc = VMXR0InvEPT(enmTlbFlush, &au64Descriptor[0]);
2830	AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvEPT %#x %#RHp failed. rc=%Rrc\n", enmTlbFlush, au64Descriptor[0], rc));
2831
2832	if ( RT_SUCCESS(rc)
2833	&& pVCpu)
2834	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushNestedPaging);
2835	}
2836
2837
2838	/**
2839	* Flushes the TLB using VPID.
2840	*
2841	* @returns VBox status code.
2842	* @param pVCpu The cross context virtual CPU structure of the calling
2843	* EMT. Can be NULL depending on @a enmTlbFlush.
2844	* @param enmTlbFlush Type of flush.
2845	* @param GCPtr Virtual address of the page to flush (can be 0 depending
2846	* on @a enmTlbFlush).
2847	*
2848	* @remarks Can be called with interrupts disabled.
2849	*/
2850	static void hmR0VmxFlushVpid(PVMCPU pVCpu, VMXTLBFLUSHVPID enmTlbFlush, RTGCPTR GCPtr)
2851	{
2852	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fVpid);
2853
2854	uint64_t au64Descriptor[2];
2855	if (enmTlbFlush == VMXTLBFLUSHVPID_ALL_CONTEXTS)
2856	{
2857	au64Descriptor[0] = 0;
2858	au64Descriptor[1] = 0;
2859	}
2860	else
2861	{
2862	AssertPtr(pVCpu);
2863	AssertMsg(pVCpu->hm.s.uCurrentAsid != 0, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
2864	AssertMsg(pVCpu->hm.s.uCurrentAsid <= UINT16_MAX, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
2865	au64Descriptor[0] = pVCpu->hm.s.uCurrentAsid;
2866	au64Descriptor[1] = GCPtr;
2867	}
2868
2869	int rc = VMXR0InvVPID(enmTlbFlush, &au64Descriptor[0]);
2870	AssertMsg(rc == VINF_SUCCESS,
2871	("VMXR0InvVPID %#x %u %RGv failed with %Rrc\n", enmTlbFlush, pVCpu ? pVCpu->hm.s.uCurrentAsid : 0, GCPtr, rc));
2872
2873	if ( RT_SUCCESS(rc)
2874	&& pVCpu)
2875	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid);
2876	NOREF(rc);
2877	}
2878
2879
2880	/**
2881	* Invalidates a guest page by guest virtual address. Only relevant for EPT/VPID,
2882	* otherwise there is nothing really to invalidate.
2883	*
2884	* @returns VBox status code.
2885	* @param pVCpu The cross context virtual CPU structure.
2886	* @param GCVirt Guest virtual address of the page to invalidate.
2887	*/
2888	VMMR0DECL(int) VMXR0InvalidatePage(PVMCPU pVCpu, RTGCPTR GCVirt)
2889	{
2890	AssertPtr(pVCpu);
2891	LogFlowFunc(("pVCpu=%p GCVirt=%RGv\n", pVCpu, GCVirt));
2892
2893	if (!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TLB_FLUSH))
2894	{
2895	/*
2896	* We must invalidate the guest TLB entry in either case, we cannot ignore it even for
2897	* the EPT case. See @bugref{6043} and @bugref{6177}.
2898	*
2899	* Set the VMCPU_FF_TLB_FLUSH force flag and flush before VM-entry in hmR0VmxFlushTLB*()
2900	* as this function maybe called in a loop with individual addresses.
2901	*/
2902	PVM pVM = pVCpu->CTX_SUFF(pVM);
2903	if (pVM->hm.s.vmx.fVpid)
2904	{
2905	bool fVpidFlush = RT_BOOL(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR);
2906
2907	#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
2908	/*
2909	* Workaround Erratum BV75, AAJ159 and others that affect several Intel CPUs
2910	* where executing INVVPID outside 64-bit mode does not flush translations of
2911	* 64-bit linear addresses, see @bugref{6208#c72}.
2912	*/
2913	if (RT_HI_U32(GCVirt))
2914	fVpidFlush = false;
2915	#endif
2916
2917	if (fVpidFlush)
2918	{
2919	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_INDIV_ADDR, GCVirt);
2920	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt);
2921	}
2922	else
2923	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2924	}
2925	else if (pVM->hm.s.fNestedPaging)
2926	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2927	}
2928
2929	return VINF_SUCCESS;
2930	}
2931
2932
2933	/**
2934	* Dummy placeholder for tagged-TLB flush handling before VM-entry. Used in the
2935	* case where neither EPT nor VPID is supported by the CPU.
2936	*
2937	* @param pHostCpu The HM physical-CPU structure.
2938	* @param pVCpu The cross context virtual CPU structure.
2939	*
2940	* @remarks Called with interrupts disabled.
2941	*/
2942	static void hmR0VmxFlushTaggedTlbNone(PHMPHYSCPU pHostCpu, PVMCPU pVCpu)
2943	{
2944	AssertPtr(pVCpu);
2945	AssertPtr(pHostCpu);
2946
2947	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
2948
2949	Assert(pHostCpu->idCpu != NIL_RTCPUID);
2950	pVCpu->hm.s.idLastCpu = pHostCpu->idCpu;
2951	pVCpu->hm.s.cTlbFlushes = pHostCpu->cTlbFlushes;
2952	pVCpu->hm.s.fForceTLBFlush = false;
2953	return;
2954	}
2955
2956
2957	/**
2958	* Flushes the tagged-TLB entries for EPT+VPID CPUs as necessary.
2959	*
2960	* @param pHostCpu The HM physical-CPU structure.
2961	* @param pVCpu The cross context virtual CPU structure.
2962	* @param pVmcsInfo The VMCS info. object.
2963	*
2964	* @remarks All references to "ASID" in this function pertains to "VPID" in Intel's
2965	* nomenclature. The reason is, to avoid confusion in compare statements
2966	* since the host-CPU copies are named "ASID".
2967	*
2968	* @remarks Called with interrupts disabled.
2969	*/
2970	static void hmR0VmxFlushTaggedTlbBoth(PHMPHYSCPU pHostCpu, PVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo)
2971	{
2972	#ifdef VBOX_WITH_STATISTICS
2973	bool fTlbFlushed = false;
2974	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { fTlbFlushed = true; } while (0)
2975	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { \
2976	if (!fTlbFlushed) \
2977	STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); \
2978	} while (0)
2979	#else
2980	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { } while (0)
2981	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { } while (0)
2982	#endif
2983
2984	AssertPtr(pVCpu);
2985	AssertPtr(pHostCpu);
2986	Assert(pHostCpu->idCpu != NIL_RTCPUID);
2987
2988	PVM pVM = pVCpu->CTX_SUFF(pVM);
2989	AssertMsg(pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid,
2990	("hmR0VmxFlushTaggedTlbBoth cannot be invoked unless NestedPaging & VPID are enabled."
2991	"fNestedPaging=%RTbool fVpid=%RTbool", pVM->hm.s.fNestedPaging, pVM->hm.s.vmx.fVpid));
2992
2993	/*
2994	* Force a TLB flush for the first world-switch if the current CPU differs from the one we
2995	* ran on last. If the TLB flush count changed, another VM (VCPU rather) has hit the ASID
2996	* limit while flushing the TLB or the host CPU is online after a suspend/resume, so we
2997	* cannot reuse the current ASID anymore.
2998	*/
2999	if ( pVCpu->hm.s.idLastCpu != pHostCpu->idCpu
3000	\|\| pVCpu->hm.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3001	{
3002	++pHostCpu->uCurrentAsid;
3003	if (pHostCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
3004	{
3005	pHostCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0. */
3006	pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
3007	pHostCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
3008	}
3009
3010	pVCpu->hm.s.uCurrentAsid = pHostCpu->uCurrentAsid;
3011	pVCpu->hm.s.idLastCpu = pHostCpu->idCpu;
3012	pVCpu->hm.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3013
3014	/*
3015	* Flush by EPT when we get rescheduled to a new host CPU to ensure EPT-only tagged mappings are also
3016	* invalidated. We don't need to flush-by-VPID here as flushing by EPT covers it. See @bugref{6568}.
3017	*/
3018	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hm.s.vmx.enmTlbFlushEpt);
3019	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3020	HMVMX_SET_TAGGED_TLB_FLUSHED();
3021	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
3022	}
3023	else if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH)) /* Check for explicit TLB flushes. */
3024	{
3025	/*
3026	* Changes to the EPT paging structure by VMM requires flushing-by-EPT as the CPU
3027	* creates guest-physical (ie. only EPT-tagged) mappings while traversing the EPT
3028	* tables when EPT is in use. Flushing-by-VPID will only flush linear (only
3029	* VPID-tagged) and combined (EPT+VPID tagged) mappings but not guest-physical
3030	* mappings, see @bugref{6568}.
3031	*
3032	* See Intel spec. 28.3.2 "Creating and Using Cached Translation Information".
3033	*/
3034	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hm.s.vmx.enmTlbFlushEpt);
3035	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3036	HMVMX_SET_TAGGED_TLB_FLUSHED();
3037	}
3038	else if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3039	{
3040	/*
3041	* The nested-guest specifies its own guest-physical address to use as the APIC-access
3042	* address which requires flushing the TLB of EPT cached structures.
3043	*
3044	* See Intel spec. 28.3.3.4 "Guidelines for Use of the INVEPT Instruction".
3045	*/
3046	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hm.s.vmx.enmTlbFlushEpt);
3047	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3048	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3049	HMVMX_SET_TAGGED_TLB_FLUSHED();
3050	}
3051
3052
3053	pVCpu->hm.s.fForceTLBFlush = false;
3054	HMVMX_UPDATE_FLUSH_SKIPPED_STAT();
3055
3056	Assert(pVCpu->hm.s.idLastCpu == pHostCpu->idCpu);
3057	Assert(pVCpu->hm.s.cTlbFlushes == pHostCpu->cTlbFlushes);
3058	AssertMsg(pVCpu->hm.s.cTlbFlushes == pHostCpu->cTlbFlushes,
3059	("Flush count mismatch for cpu %d (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pHostCpu->cTlbFlushes));
3060	AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
3061	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pHostCpu->idCpu,
3062	pHostCpu->uCurrentAsid, pHostCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
3063	AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
3064	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pHostCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
3065
3066	/* Update VMCS with the VPID. */
3067	int rc = VMXWriteVmcs32(VMX_VMCS16_VPID, pVCpu->hm.s.uCurrentAsid);
3068	AssertRC(rc);
3069
3070	#undef HMVMX_SET_TAGGED_TLB_FLUSHED
3071	}
3072
3073
3074	/**
3075	* Flushes the tagged-TLB entries for EPT CPUs as necessary.
3076	*
3077	* @param pHostCpu The HM physical-CPU structure.
3078	* @param pVCpu The cross context virtual CPU structure.
3079	* @param pVmcsInfo The VMCS info. object.
3080	*
3081	* @remarks Called with interrupts disabled.
3082	*/
3083	static void hmR0VmxFlushTaggedTlbEpt(PHMPHYSCPU pHostCpu, PVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo)
3084	{
3085	AssertPtr(pVCpu);
3086	AssertPtr(pHostCpu);
3087	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3088	AssertMsg(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked without NestedPaging."));
3089	AssertMsg(!pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fVpid, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with VPID."));
3090
3091	/*
3092	* Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
3093	* A change in the TLB flush count implies the host CPU is online after a suspend/resume.
3094	*/
3095	if ( pVCpu->hm.s.idLastCpu != pHostCpu->idCpu
3096	\|\| pVCpu->hm.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3097	{
3098	pVCpu->hm.s.fForceTLBFlush = true;
3099	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3100	}
3101
3102	/* Check for explicit TLB flushes. */
3103	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
3104	{
3105	pVCpu->hm.s.fForceTLBFlush = true;
3106	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3107	}
3108
3109	/* Check for TLB flushes while switching to/from a nested-guest. */
3110	if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3111	{
3112	pVCpu->hm.s.fForceTLBFlush = true;
3113	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3114	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3115	}
3116
3117	pVCpu->hm.s.idLastCpu = pHostCpu->idCpu;
3118	pVCpu->hm.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3119
3120	if (pVCpu->hm.s.fForceTLBFlush)
3121	{
3122	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVCpu->CTX_SUFF(pVM)->hm.s.vmx.enmTlbFlushEpt);
3123	pVCpu->hm.s.fForceTLBFlush = false;
3124	}
3125	}
3126
3127
3128	/**
3129	* Flushes the tagged-TLB entries for VPID CPUs as necessary.
3130	*
3131	* @param pHostCpu The HM physical-CPU structure.
3132	* @param pVCpu The cross context virtual CPU structure.
3133	*
3134	* @remarks Called with interrupts disabled.
3135	*/
3136	static void hmR0VmxFlushTaggedTlbVpid(PHMPHYSCPU pHostCpu, PVMCPU pVCpu)
3137	{
3138	AssertPtr(pVCpu);
3139	AssertPtr(pHostCpu);
3140	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3141	AssertMsg(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fVpid, ("hmR0VmxFlushTlbVpid cannot be invoked without VPID."));
3142	AssertMsg(!pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging, ("hmR0VmxFlushTlbVpid cannot be invoked with NestedPaging"));
3143
3144	/*
3145	* Force a TLB flush for the first world switch if the current CPU differs from the one we
3146	* ran on last. If the TLB flush count changed, another VM (VCPU rather) has hit the ASID
3147	* limit while flushing the TLB or the host CPU is online after a suspend/resume, so we
3148	* cannot reuse the current ASID anymore.
3149	*/
3150	if ( pVCpu->hm.s.idLastCpu != pHostCpu->idCpu
3151	\|\| pVCpu->hm.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3152	{
3153	pVCpu->hm.s.fForceTLBFlush = true;
3154	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3155	}
3156
3157	/* Check for explicit TLB flushes. */
3158	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
3159	{
3160	/*
3161	* If we ever support VPID flush combinations other than ALL or SINGLE-context (see
3162	* hmR0VmxSetupTaggedTlb()) we would need to explicitly flush in this case (add an
3163	* fExplicitFlush = true here and change the pHostCpu->fFlushAsidBeforeUse check below to
3164	* include fExplicitFlush's too) - an obscure corner case.
3165	*/
3166	pVCpu->hm.s.fForceTLBFlush = true;
3167	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3168	}
3169
3170	/* Check for TLB flushes while switching to/from a nested-guest. */
3171	if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3172	{
3173	pVCpu->hm.s.fForceTLBFlush = true;
3174	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3175	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3176	}
3177
3178	PVM pVM = pVCpu->CTX_SUFF(pVM);
3179	pVCpu->hm.s.idLastCpu = pHostCpu->idCpu;
3180	if (pVCpu->hm.s.fForceTLBFlush)
3181	{
3182	++pHostCpu->uCurrentAsid;
3183	if (pHostCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
3184	{
3185	pHostCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0 */
3186	pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
3187	pHostCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
3188	}
3189
3190	pVCpu->hm.s.fForceTLBFlush = false;
3191	pVCpu->hm.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3192	pVCpu->hm.s.uCurrentAsid = pHostCpu->uCurrentAsid;
3193	if (pHostCpu->fFlushAsidBeforeUse)
3194	{
3195	if (pVM->hm.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_SINGLE_CONTEXT)
3196	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_SINGLE_CONTEXT, 0 /* GCPtr */);
3197	else if (pVM->hm.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_ALL_CONTEXTS)
3198	{
3199	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_ALL_CONTEXTS, 0 /* GCPtr */);
3200	pHostCpu->fFlushAsidBeforeUse = false;
3201	}
3202	else
3203	{
3204	/* hmR0VmxSetupTaggedTlb() ensures we never get here. Paranoia. */
3205	AssertMsgFailed(("Unsupported VPID-flush context type.\n"));
3206	}
3207	}
3208	}
3209
3210	AssertMsg(pVCpu->hm.s.cTlbFlushes == pHostCpu->cTlbFlushes,
3211	("Flush count mismatch for cpu %d (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pHostCpu->cTlbFlushes));
3212	AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
3213	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pHostCpu->idCpu,
3214	pHostCpu->uCurrentAsid, pHostCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
3215	AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
3216	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pHostCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
3217
3218	int rc = VMXWriteVmcs32(VMX_VMCS16_VPID, pVCpu->hm.s.uCurrentAsid);
3219	AssertRC(rc);
3220	}
3221
3222
3223	/**
3224	* Flushes the guest TLB entry based on CPU capabilities.
3225	*
3226	* @param pHostCpu The HM physical-CPU structure.
3227	* @param pVCpu The cross context virtual CPU structure.
3228	* @param pVmcsInfo The VMCS info. object.
3229	*
3230	* @remarks Called with interrupts disabled.
3231	*/
3232	static void hmR0VmxFlushTaggedTlb(PHMPHYSCPU pHostCpu, PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
3233	{
3234	#ifdef HMVMX_ALWAYS_FLUSH_TLB
3235	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
3236	#endif
3237	PVM pVM = pVCpu->CTX_SUFF(pVM);
3238	switch (pVM->hm.s.vmx.enmTlbFlushType)
3239	{
3240	case VMXTLBFLUSHTYPE_EPT_VPID: hmR0VmxFlushTaggedTlbBoth(pHostCpu, pVCpu, pVmcsInfo); break;
3241	case VMXTLBFLUSHTYPE_EPT: hmR0VmxFlushTaggedTlbEpt(pHostCpu, pVCpu, pVmcsInfo); break;
3242	case VMXTLBFLUSHTYPE_VPID: hmR0VmxFlushTaggedTlbVpid(pHostCpu, pVCpu); break;
3243	case VMXTLBFLUSHTYPE_NONE: hmR0VmxFlushTaggedTlbNone(pHostCpu, pVCpu); break;
3244	default:
3245	AssertMsgFailed(("Invalid flush-tag function identifier\n"));
3246	break;
3247	}
3248	/* Don't assert that VMCPU_FF_TLB_FLUSH should no longer be pending. It can be set by other EMTs. */
3249	}
3250
3251
3252	/**
3253	* Sets up the appropriate tagged TLB-flush level and handler for flushing guest
3254	* TLB entries from the host TLB before VM-entry.
3255	*
3256	* @returns VBox status code.
3257	* @param pVM The cross context VM structure.
3258	*/
3259	static int hmR0VmxSetupTaggedTlb(PVM pVM)
3260	{
3261	/*
3262	* Determine optimal flush type for nested paging.
3263	* We cannot ignore EPT if no suitable flush-types is supported by the CPU as we've already setup
3264	* unrestricted guest execution (see hmR3InitFinalizeR0()).
3265	*/
3266	if (pVM->hm.s.fNestedPaging)
3267	{
3268	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT)
3269	{
3270	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT)
3271	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_SINGLE_CONTEXT;
3272	else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
3273	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_ALL_CONTEXTS;
3274	else
3275	{
3276	/* Shouldn't happen. EPT is supported but no suitable flush-types supported. */
3277	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3278	pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_EPT_FLUSH_TYPE_UNSUPPORTED;
3279	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3280	}
3281
3282	/* Make sure the write-back cacheable memory type for EPT is supported. */
3283	if (RT_UNLIKELY(!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EMT_WB)))
3284	{
3285	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3286	pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_EPT_MEM_TYPE_NOT_WB;
3287	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3288	}
3289
3290	/* EPT requires a page-walk length of 4. */
3291	if (RT_UNLIKELY(!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_PAGE_WALK_LENGTH_4)))
3292	{
3293	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3294	pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_EPT_PAGE_WALK_LENGTH_UNSUPPORTED;
3295	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3296	}
3297	}
3298	else
3299	{
3300	/* Shouldn't happen. EPT is supported but INVEPT instruction is not supported. */
3301	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3302	pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_EPT_INVEPT_UNAVAILABLE;
3303	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3304	}
3305	}
3306
3307	/*
3308	* Determine optimal flush type for VPID.
3309	*/
3310	if (pVM->hm.s.vmx.fVpid)
3311	{
3312	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID)
3313	{
3314	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT)
3315	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_SINGLE_CONTEXT;
3316	else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS)
3317	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_ALL_CONTEXTS;
3318	else
3319	{
3320	/* Neither SINGLE nor ALL-context flush types for VPID is supported by the CPU. Ignore VPID capability. */
3321	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
3322	LogRelFunc(("Only INDIV_ADDR supported. Ignoring VPID.\n"));
3323	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
3324	LogRelFunc(("Only SINGLE_CONTEXT_RETAIN_GLOBALS supported. Ignoring VPID.\n"));
3325	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NOT_SUPPORTED;
3326	pVM->hm.s.vmx.fVpid = false;
3327	}
3328	}
3329	else
3330	{
3331	/* Shouldn't happen. VPID is supported but INVVPID is not supported by the CPU. Ignore VPID capability. */
3332	Log4Func(("VPID supported without INVEPT support. Ignoring VPID.\n"));
3333	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NOT_SUPPORTED;
3334	pVM->hm.s.vmx.fVpid = false;
3335	}
3336	}
3337
3338	/*
3339	* Setup the handler for flushing tagged-TLBs.
3340	*/
3341	if (pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid)
3342	pVM->hm.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_EPT_VPID;
3343	else if (pVM->hm.s.fNestedPaging)
3344	pVM->hm.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_EPT;
3345	else if (pVM->hm.s.vmx.fVpid)
3346	pVM->hm.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_VPID;
3347	else
3348	pVM->hm.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_NONE;
3349	return VINF_SUCCESS;
3350	}
3351
3352
3353	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3354	/**
3355	* Sets up the shadow VMCS fields arrays.
3356	*
3357	* This function builds arrays of VMCS fields to sync the shadow VMCS later while
3358	* executing the guest.
3359	*
3360	* @returns VBox status code.
3361	* @param pVM The cross context VM structure.
3362	*/
3363	static int hmR0VmxSetupShadowVmcsFieldsArrays(PVM pVM)
3364	{
3365	/*
3366	* Paranoia. Ensure we haven't exposed the VMWRITE-All VMX feature to the guest
3367	* when the host does not support it.
3368	*/
3369	bool const fGstVmwriteAll = pVM->cpum.ro.GuestFeatures.fVmxVmwriteAll;
3370	if ( !fGstVmwriteAll
3371	\|\| (pVM->hm.s.vmx.Msrs.u64Misc & VMX_MISC_VMWRITE_ALL))
3372	{ /* likely. */ }
3373	else
3374	{
3375	LogRelFunc(("VMX VMWRITE-All feature exposed to the guest but host CPU does not support it!\n"));
3376	pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_GST_HOST_VMWRITE_ALL;
3377	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3378	}
3379
3380	uint32_t const cVmcsFields = RT_ELEMENTS(g_aVmcsFields);
3381	uint32_t cRwFields = 0;
3382	uint32_t cRoFields = 0;
3383	for (uint32_t i = 0; i < cVmcsFields; i++)
3384	{
3385	VMXVMCSFIELD VmcsField;
3386	VmcsField.u = g_aVmcsFields[i];
3387
3388	/*
3389	* We will be writing "FULL" (64-bit) fields while syncing the shadow VMCS.
3390	* Therefore, "HIGH" (32-bit portion of 64-bit) fields must not be included
3391	* in the shadow VMCS fields array as they would be redundant.
3392	*
3393	* If the VMCS field depends on a CPU feature that is not exposed to the guest,
3394	* we must not include it in the shadow VMCS fields array. Guests attempting to
3395	* VMREAD/VMWRITE such VMCS fields would cause a VM-exit and we shall emulate
3396	* the required behavior.
3397	*/
3398	if ( VmcsField.n.fAccessType == VMX_VMCSFIELD_ACCESS_FULL
3399	&& CPUMIsGuestVmxVmcsFieldValid(pVM, VmcsField.u))
3400	{
3401	/*
3402	* Read-only fields are placed in a separate array so that while syncing shadow
3403	* VMCS fields later (which is more performance critical) we can avoid branches.
3404	*
3405	* However, if the guest can write to all fields (including read-only fields),
3406	* we treat it a as read/write field. Otherwise, writing to these fields would
3407	* cause a VMWRITE instruction error while syncing the shadow VMCS .
3408	*/
3409	if ( fGstVmwriteAll
3410	\|\| !HMVmxIsVmcsFieldReadOnly(VmcsField.u))
3411	pVM->hm.s.vmx.paShadowVmcsFields[cRwFields++] = VmcsField.u;
3412	else
3413	pVM->hm.s.vmx.paShadowVmcsRoFields[cRoFields++] = VmcsField.u;
3414	}
3415	}
3416
3417	/* Update the counts. */
3418	pVM->hm.s.vmx.cShadowVmcsFields = cRwFields;
3419	pVM->hm.s.vmx.cShadowVmcsRoFields = cRoFields;
3420	return VINF_SUCCESS;
3421	}
3422
3423
3424	/**
3425	* Sets up the VMREAD and VMWRITE bitmaps.
3426	*
3427	* @param pVM The cross context VM structure.
3428	*/
3429	static void hmR0VmxSetupVmreadVmwriteBitmaps(PVM pVM)
3430	{
3431	/*
3432	* By default, ensure guest attempts to acceses to any VMCS fields cause VM-exits.
3433	*/
3434	uint32_t const cbBitmap = X86_PAGE_4K_SIZE;
3435	uint8_t pbVmreadBitmap = (uint8_t )pVM->hm.s.vmx.pvVmreadBitmap;
3436	uint8_t pbVmwriteBitmap = (uint8_t )pVM->hm.s.vmx.pvVmwriteBitmap;
3437	ASMMemFill32(pbVmreadBitmap, cbBitmap, UINT32_C(0xffffffff));
3438	ASMMemFill32(pbVmwriteBitmap, cbBitmap, UINT32_C(0xffffffff));
3439
3440	/*
3441	* Skip intercepting VMREAD/VMWRITE to guest read/write fields in the
3442	* VMREAD and VMWRITE bitmaps.
3443	*/
3444	{
3445	uint32_t const *paShadowVmcsFields = pVM->hm.s.vmx.paShadowVmcsFields;
3446	uint32_t const cShadowVmcsFields = pVM->hm.s.vmx.cShadowVmcsFields;
3447	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
3448	{
3449	uint32_t const uVmcsField = paShadowVmcsFields[i];
3450	Assert(!(uVmcsField & VMX_VMCSFIELD_RSVD_MASK));
3451	Assert(uVmcsField >> 3 < cbBitmap);
3452	ASMBitClear(pbVmreadBitmap + (uVmcsField >> 3), uVmcsField & 7);
3453	ASMBitClear(pbVmwriteBitmap + (uVmcsField >> 3), uVmcsField & 7);
3454	}
3455	}
3456
3457	/*
3458	* Skip intercepting VMREAD for guest read-only fields in the VMREAD bitmap
3459	* if the host supports VMWRITE to all supported VMCS fields.
3460	*/
3461	if (pVM->hm.s.vmx.Msrs.u64Misc & VMX_MISC_VMWRITE_ALL)
3462	{
3463	uint32_t const *paShadowVmcsRoFields = pVM->hm.s.vmx.paShadowVmcsRoFields;
3464	uint32_t const cShadowVmcsRoFields = pVM->hm.s.vmx.cShadowVmcsRoFields;
3465	for (uint32_t i = 0; i < cShadowVmcsRoFields; i++)
3466	{
3467	uint32_t const uVmcsField = paShadowVmcsRoFields[i];
3468	Assert(!(uVmcsField & VMX_VMCSFIELD_RSVD_MASK));
3469	Assert(uVmcsField >> 3 < cbBitmap);
3470	ASMBitClear(pbVmreadBitmap + (uVmcsField >> 3), uVmcsField & 7);
3471	}
3472	}
3473	}
3474	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
3475
3476
3477	/**
3478	* Sets up the virtual-APIC page address for the VMCS.
3479	*
3480	* @returns VBox status code.
3481	* @param pVCpu The cross context virtual CPU structure.
3482	* @param pVmcsInfo The VMCS info. object.
3483	*/
3484	DECLINLINE(int) hmR0VmxSetupVmcsVirtApicAddr(PVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo)
3485	{
3486	NOREF(pVCpu); /* Used implicitly by VMXWriteVmcs64 on 32-bit hosts. */
3487	RTHCPHYS const HCPhysVirtApic = pVmcsInfo->HCPhysVirtApic;
3488	Assert(HCPhysVirtApic != NIL_RTHCPHYS);
3489	Assert(!(HCPhysVirtApic & 0xfff)); /* Bits 11:0 MBZ. */
3490	return VMXWriteVmcs64(VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL, HCPhysVirtApic);
3491	}
3492
3493
3494	/**
3495	* Sets up the MSR-bitmap address for the VMCS.
3496	*
3497	* @returns VBox status code.
3498	* @param pVCpu The cross context virtual CPU structure.
3499	* @param pVmcsInfo The VMCS info. object.
3500	*/
3501	DECLINLINE(int) hmR0VmxSetupVmcsMsrBitmapAddr(PVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo)
3502	{
3503	NOREF(pVCpu); /* Used implicitly by VMXWriteVmcs64 on 32-bit hosts. */
3504	RTHCPHYS const HCPhysMsrBitmap = pVmcsInfo->HCPhysMsrBitmap;
3505	Assert(HCPhysMsrBitmap != NIL_RTHCPHYS);
3506	Assert(!(HCPhysMsrBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3507	return VMXWriteVmcs64(VMX_VMCS64_CTRL_MSR_BITMAP_FULL, HCPhysMsrBitmap);
3508	}
3509
3510
3511	/**
3512	* Sets up the APIC-access page address for the VMCS.
3513	*
3514	* @returns VBox status code.
3515	* @param pVCpu The cross context virtual CPU structure.
3516	*/
3517	DECLINLINE(int) hmR0VmxSetupVmcsApicAccessAddr(PVMCPU pVCpu)
3518	{
3519	RTHCPHYS const HCPhysApicAccess = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.HCPhysApicAccess;
3520	Assert(HCPhysApicAccess != NIL_RTHCPHYS);
3521	Assert(!(HCPhysApicAccess & 0xfff)); /* Bits 11:0 MBZ. */
3522	return VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, HCPhysApicAccess);
3523	}
3524
3525
3526	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3527	/**
3528	* Sets up the VMREAD bitmap address for the VMCS.
3529	*
3530	* @returns VBox status code.
3531	* @param pVCpu The cross context virtual CPU structure.
3532	*/
3533	DECLINLINE(int) hmR0VmxSetupVmcsVmreadBitmapAddr(PVMCPU pVCpu)
3534	{
3535	RTHCPHYS const HCPhysVmreadBitmap = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.HCPhysVmreadBitmap;
3536	Assert(HCPhysVmreadBitmap != NIL_RTHCPHYS);
3537	Assert(!(HCPhysVmreadBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3538	return VMXWriteVmcs64(VMX_VMCS64_CTRL_VMREAD_BITMAP_FULL, HCPhysVmreadBitmap);
3539	}
3540
3541
3542	/**
3543	* Sets up the VMWRITE bitmap address for the VMCS.
3544	*
3545	* @returns VBox status code.
3546	* @param pVCpu The cross context virtual CPU structure.
3547	*/
3548	DECLINLINE(int) hmR0VmxSetupVmcsVmwriteBitmapAddr(PVMCPU pVCpu)
3549	{
3550	RTHCPHYS const HCPhysVmwriteBitmap = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.HCPhysVmwriteBitmap;
3551	Assert(HCPhysVmwriteBitmap != NIL_RTHCPHYS);
3552	Assert(!(HCPhysVmwriteBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3553	return VMXWriteVmcs64(VMX_VMCS64_CTRL_VMWRITE_BITMAP_FULL, HCPhysVmwriteBitmap);
3554	}
3555	#endif
3556
3557
3558	/**
3559	* Sets up the VM-entry MSR load, VM-exit MSR-store and VM-exit MSR-load addresses
3560	* in the VMCS.
3561	*
3562	* @returns VBox status code.
3563	* @param pVCpu The cross context virtual CPU structure.
3564	* @param pVmcsInfo The VMCS info. object.
3565	*/
3566	DECLINLINE(int) hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
3567	{
3568	NOREF(pVCpu); /* Used implicitly by VMXWriteVmcs64 on 32-bit hosts. */
3569
3570	RTHCPHYS const HCPhysGuestMsrLoad = pVmcsInfo->HCPhysGuestMsrLoad;
3571	Assert(HCPhysGuestMsrLoad != NIL_RTHCPHYS);
3572	Assert(!(HCPhysGuestMsrLoad & 0xf)); /* Bits 3:0 MBZ. */
3573
3574	RTHCPHYS const HCPhysGuestMsrStore = pVmcsInfo->HCPhysGuestMsrStore;
3575	Assert(HCPhysGuestMsrStore != NIL_RTHCPHYS);
3576	Assert(!(HCPhysGuestMsrStore & 0xf)); /* Bits 3:0 MBZ. */
3577
3578	RTHCPHYS const HCPhysHostMsrLoad = pVmcsInfo->HCPhysHostMsrLoad;
3579	Assert(HCPhysHostMsrLoad != NIL_RTHCPHYS);
3580	Assert(!(HCPhysHostMsrLoad & 0xf)); /* Bits 3:0 MBZ. */
3581
3582	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL, HCPhysGuestMsrLoad);
3583	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL, HCPhysGuestMsrStore);
3584	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL, HCPhysHostMsrLoad);
3585	AssertRCReturn(rc, rc);
3586	return VINF_SUCCESS;
3587	}
3588
3589
3590	/**
3591	* Sets up MSR permissions in the MSR bitmap of a VMCS info. object.
3592	*
3593	* @param pVCpu The cross context virtual CPU structure.
3594	* @param pVmcsInfo The VMCS info. object.
3595	*/
3596	static void hmR0VmxSetupVmcsMsrPermissions(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
3597	{
3598	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS);
3599
3600	/*
3601	* The guest can access the following MSRs (read, write) without causing
3602	* VM-exits; they are loaded/stored automatically using fields in the VMCS.
3603	*/
3604	PVM pVM = pVCpu->CTX_SUFF(pVM);
3605	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_CS, VMXMSRPM_ALLOW_RD_WR);
3606	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_ESP, VMXMSRPM_ALLOW_RD_WR);
3607	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_EIP, VMXMSRPM_ALLOW_RD_WR);
3608	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_GS_BASE, VMXMSRPM_ALLOW_RD_WR);
3609	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_FS_BASE, VMXMSRPM_ALLOW_RD_WR);
3610
3611	/*
3612	* The IA32_PRED_CMD and IA32_FLUSH_CMD MSRs are write-only and has no state
3613	* associated with then. We never need to intercept access (writes need to be
3614	* executed without causing a VM-exit, reads will #GP fault anyway).
3615	*
3616	* The IA32_SPEC_CTRL MSR is read/write and has state. We allow the guest to
3617	* read/write them. We swap the the guest/host MSR value using the
3618	* auto-load/store MSR area.
3619	*/
3620	if (pVM->cpum.ro.GuestFeatures.fIbpb)
3621	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_PRED_CMD, VMXMSRPM_ALLOW_RD_WR);
3622	if (pVM->cpum.ro.GuestFeatures.fFlushCmd)
3623	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_FLUSH_CMD, VMXMSRPM_ALLOW_RD_WR);
3624	if (pVM->cpum.ro.GuestFeatures.fIbrs)
3625	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SPEC_CTRL, VMXMSRPM_ALLOW_RD_WR);
3626
3627	#if HC_ARCH_BITS == 64
3628	/*
3629	* Allow full read/write access for the following MSRs (mandatory for VT-x)
3630	* required for 64-bit guests.
3631	*/
3632	if (pVM->hm.s.fAllow64BitGuests)
3633	{
3634	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_LSTAR, VMXMSRPM_ALLOW_RD_WR);
3635	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K6_STAR, VMXMSRPM_ALLOW_RD_WR);
3636	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_SF_MASK, VMXMSRPM_ALLOW_RD_WR);
3637	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_KERNEL_GS_BASE, VMXMSRPM_ALLOW_RD_WR);
3638	}
3639	#endif
3640
3641	/*
3642	* IA32_EFER MSR is always intercepted, see @bugref{9180#c37}.
3643	*/
3644	#ifdef VBOX_STRICT
3645	Assert(pVmcsInfo->pvMsrBitmap);
3646	uint32_t const fMsrpmEfer = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, MSR_K6_EFER);
3647	Assert(fMsrpmEfer == VMXMSRPM_EXIT_RD_WR);
3648	#endif
3649	}
3650
3651
3652	/**
3653	* Sets up pin-based VM-execution controls in the VMCS.
3654	*
3655	* @returns VBox status code.
3656	* @param pVCpu The cross context virtual CPU structure.
3657	* @param pVmcsInfo The VMCS info. object.
3658	*/
3659	static int hmR0VmxSetupVmcsPinCtls(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
3660	{
3661	PVM pVM = pVCpu->CTX_SUFF(pVM);
3662	uint32_t fVal = pVM->hm.s.vmx.Msrs.PinCtls.n.allowed0; /* Bits set here must always be set. */
3663	uint32_t const fZap = pVM->hm.s.vmx.Msrs.PinCtls.n.allowed1; /* Bits cleared here must always be cleared. */
3664
3665	fVal \|= VMX_PIN_CTLS_EXT_INT_EXIT /* External interrupts cause a VM-exit. */
3666	\| VMX_PIN_CTLS_NMI_EXIT; /* Non-maskable interrupts (NMIs) cause a VM-exit. */
3667
3668	if (pVM->hm.s.vmx.Msrs.PinCtls.n.allowed1 & VMX_PIN_CTLS_VIRT_NMI)
3669	fVal \|= VMX_PIN_CTLS_VIRT_NMI; /* Use virtual NMIs and virtual-NMI blocking features. */
3670
3671	/* Enable the VMX-preemption timer. */
3672	if (pVM->hm.s.vmx.fUsePreemptTimer)
3673	{
3674	Assert(pVM->hm.s.vmx.Msrs.PinCtls.n.allowed1 & VMX_PIN_CTLS_PREEMPT_TIMER);
3675	fVal \|= VMX_PIN_CTLS_PREEMPT_TIMER;
3676	}
3677
3678	#if 0
3679	/* Enable posted-interrupt processing. */
3680	if (pVM->hm.s.fPostedIntrs)
3681	{
3682	Assert(pVM->hm.s.vmx.Msrs.PinCtls.n.allowed1 & VMX_PIN_CTLS_POSTED_INT);
3683	Assert(pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_ACK_EXT_INT);
3684	fVal \|= VMX_PIN_CTLS_POSTED_INT;
3685	}
3686	#endif
3687
3688	if ((fVal & fZap) != fVal)
3689	{
3690	LogRelFunc(("Invalid pin-based VM-execution controls combo! Cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3691	pVM->hm.s.vmx.Msrs.PinCtls.n.allowed0, fVal, fZap));
3692	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PIN_EXEC;
3693	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3694	}
3695
3696	/* Commit it to the VMCS and update our cache. */
3697	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, fVal);
3698	AssertRCReturn(rc, rc);
3699	pVmcsInfo->u32PinCtls = fVal;
3700
3701	return VINF_SUCCESS;
3702	}
3703
3704
3705	/**
3706	* Sets up secondary processor-based VM-execution controls in the VMCS.
3707	*
3708	* @returns VBox status code.
3709	* @param pVCpu The cross context virtual CPU structure.
3710	* @param pVmcsInfo The VMCS info. object.
3711	*/
3712	static int hmR0VmxSetupVmcsProcCtls2(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
3713	{
3714	PVM pVM = pVCpu->CTX_SUFF(pVM);
3715	uint32_t fVal = pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed0; /* Bits set here must be set in the VMCS. */
3716	uint32_t const fZap = pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3717
3718	/* WBINVD causes a VM-exit. */
3719	if (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_WBINVD_EXIT)
3720	fVal \|= VMX_PROC_CTLS2_WBINVD_EXIT;
3721
3722	/* Enable EPT (aka nested-paging). */
3723	if (pVM->hm.s.fNestedPaging)
3724	fVal \|= VMX_PROC_CTLS2_EPT;
3725
3726	/* Enable the INVPCID instruction if we expose it to the guest and is supported
3727	by the hardware. Without this, guest executing INVPCID would cause a #UD. */
3728	if ( pVM->cpum.ro.GuestFeatures.fInvpcid
3729	&& (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_INVPCID))
3730	fVal \|= VMX_PROC_CTLS2_INVPCID;
3731
3732	/* Enable VPID. */
3733	if (pVM->hm.s.vmx.fVpid)
3734	fVal \|= VMX_PROC_CTLS2_VPID;
3735
3736	/* Enable unrestricted guest execution. */
3737	if (pVM->hm.s.vmx.fUnrestrictedGuest)
3738	fVal \|= VMX_PROC_CTLS2_UNRESTRICTED_GUEST;
3739
3740	#if 0
3741	if (pVM->hm.s.fVirtApicRegs)
3742	{
3743	/* Enable APIC-register virtualization. */
3744	Assert(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_APIC_REG_VIRT);
3745	fVal \|= VMX_PROC_CTLS2_APIC_REG_VIRT;
3746
3747	/* Enable virtual-interrupt delivery. */
3748	Assert(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_INTR_DELIVERY);
3749	fVal \|= VMX_PROC_CTLS2_VIRT_INTR_DELIVERY;
3750	}
3751	#endif
3752
3753	/* Virtualize-APIC accesses if supported by the CPU. The virtual-APIC page is
3754	where the TPR shadow resides. */
3755	/** @todo VIRT_X2APIC support, it's mutually exclusive with this. So must be
3756	* done dynamically. */
3757	if (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
3758	{
3759	fVal \|= VMX_PROC_CTLS2_VIRT_APIC_ACCESS;
3760	int rc = hmR0VmxSetupVmcsApicAccessAddr(pVCpu);
3761	AssertRCReturn(rc, rc);
3762	}
3763
3764	/* Enable the RDTSCP instruction if we expose it to the guest and is supported
3765	by the hardware. Without this, guest executing RDTSCP would cause a #UD. */
3766	if ( pVM->cpum.ro.GuestFeatures.fRdTscP
3767	&& (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_RDTSCP))
3768	fVal \|= VMX_PROC_CTLS2_RDTSCP;
3769
3770	/* Enable Pause-Loop exiting. */
3771	if ( (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT)
3772	&& pVM->hm.s.vmx.cPleGapTicks
3773	&& pVM->hm.s.vmx.cPleWindowTicks)
3774	{
3775	fVal \|= VMX_PROC_CTLS2_PAUSE_LOOP_EXIT;
3776
3777	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, pVM->hm.s.vmx.cPleGapTicks);
3778	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, pVM->hm.s.vmx.cPleWindowTicks);
3779	AssertRCReturn(rc, rc);
3780	}
3781
3782	if ((fVal & fZap) != fVal)
3783	{
3784	LogRelFunc(("Invalid secondary processor-based VM-execution controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3785	pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed0, fVal, fZap));
3786	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC2;
3787	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3788	}
3789
3790	/* Commit it to the VMCS and update our cache. */
3791	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, fVal);
3792	AssertRCReturn(rc, rc);
3793	pVmcsInfo->u32ProcCtls2 = fVal;
3794
3795	return VINF_SUCCESS;
3796	}
3797
3798
3799	/**
3800	* Sets up processor-based VM-execution controls in the VMCS.
3801	*
3802	* @returns VBox status code.
3803	* @param pVCpu The cross context virtual CPU structure.
3804	* @param pVmcsInfo The VMCS info. object.
3805	*/
3806	static int hmR0VmxSetupVmcsProcCtls(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
3807	{
3808	PVM pVM = pVCpu->CTX_SUFF(pVM);
3809
3810	uint32_t fVal = pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
3811	uint32_t const fZap = pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3812
3813	fVal \|= VMX_PROC_CTLS_HLT_EXIT /* HLT causes a VM-exit. */
3814	\| VMX_PROC_CTLS_USE_TSC_OFFSETTING /* Use TSC-offsetting. */
3815	\| VMX_PROC_CTLS_MOV_DR_EXIT /* MOV DRx causes a VM-exit. */
3816	\| VMX_PROC_CTLS_UNCOND_IO_EXIT /* All IO instructions cause a VM-exit. */
3817	\| VMX_PROC_CTLS_RDPMC_EXIT /* RDPMC causes a VM-exit. */
3818	\| VMX_PROC_CTLS_MONITOR_EXIT /* MONITOR causes a VM-exit. */
3819	\| VMX_PROC_CTLS_MWAIT_EXIT; /* MWAIT causes a VM-exit. */
3820
3821	/* We toggle VMX_PROC_CTLS_MOV_DR_EXIT later, check if it's not -always- needed to be set or clear. */
3822	if ( !(pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_MOV_DR_EXIT)
3823	\|\| (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed0 & VMX_PROC_CTLS_MOV_DR_EXIT))
3824	{
3825	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_MOV_DRX_EXIT;
3826	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3827	}
3828
3829	/* Without nested paging, INVLPG (also affects INVPCID) and MOV CR3 instructions should cause VM-exits. */
3830	if (!pVM->hm.s.fNestedPaging)
3831	{
3832	Assert(!pVM->hm.s.vmx.fUnrestrictedGuest);
3833	fVal \|= VMX_PROC_CTLS_INVLPG_EXIT
3834	\| VMX_PROC_CTLS_CR3_LOAD_EXIT
3835	\| VMX_PROC_CTLS_CR3_STORE_EXIT;
3836	}
3837
3838	/* Use TPR shadowing if supported by the CPU. */
3839	if ( PDMHasApic(pVM)
3840	&& pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW)
3841	{
3842	fVal \|= VMX_PROC_CTLS_USE_TPR_SHADOW; /* CR8 reads from the Virtual-APIC page. */
3843	/* CR8 writes cause a VM-exit based on TPR threshold. */
3844	Assert(!(fVal & VMX_PROC_CTLS_CR8_STORE_EXIT));
3845	Assert(!(fVal & VMX_PROC_CTLS_CR8_LOAD_EXIT));
3846	int rc = hmR0VmxSetupVmcsVirtApicAddr(pVCpu, pVmcsInfo);
3847	AssertRCReturn(rc, rc);
3848	}
3849	else
3850	{
3851	/* Some 32-bit CPUs do not support CR8 load/store exiting as MOV CR8 is
3852	invalid on 32-bit Intel CPUs. Set this control only for 64-bit guests. */
3853	if (pVM->hm.s.fAllow64BitGuests)
3854	{
3855	fVal \|= VMX_PROC_CTLS_CR8_STORE_EXIT /* CR8 reads cause a VM-exit. */
3856	\| VMX_PROC_CTLS_CR8_LOAD_EXIT; /* CR8 writes cause a VM-exit. */
3857	}
3858	}
3859
3860	/* Use MSR-bitmaps if supported by the CPU. */
3861	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
3862	{
3863	fVal \|= VMX_PROC_CTLS_USE_MSR_BITMAPS;
3864	int rc = hmR0VmxSetupVmcsMsrBitmapAddr(pVCpu, pVmcsInfo);
3865	AssertRCReturn(rc, rc);
3866	}
3867
3868	/* Use the secondary processor-based VM-execution controls if supported by the CPU. */
3869	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
3870	fVal \|= VMX_PROC_CTLS_USE_SECONDARY_CTLS;
3871
3872	if ((fVal & fZap) != fVal)
3873	{
3874	LogRelFunc(("Invalid processor-based VM-execution controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3875	pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed0, fVal, fZap));
3876	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC;
3877	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3878	}
3879
3880	/* Commit it to the VMCS and update our cache. */
3881	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, fVal);
3882	AssertRCReturn(rc, rc);
3883	pVmcsInfo->u32ProcCtls = fVal;
3884
3885	/* Set up MSR permissions that don't change through the lifetime of the VM. */
3886	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
3887	hmR0VmxSetupVmcsMsrPermissions(pVCpu, pVmcsInfo);
3888
3889	/* Set up secondary processor-based VM-execution controls if the CPU supports it. */
3890	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
3891	return hmR0VmxSetupVmcsProcCtls2(pVCpu, pVmcsInfo);
3892
3893	/* Sanity check, should not really happen. */
3894	if (RT_LIKELY(!pVM->hm.s.vmx.fUnrestrictedGuest))
3895	{ /* likely */ }
3896	else
3897	{
3898	pVCpu->hm.s.u32HMError = VMX_UFC_INVALID_UX_COMBO;
3899	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3900	}
3901
3902	/* Old CPUs without secondary processor-based VM-execution controls would end up here. */
3903	return VINF_SUCCESS;
3904	}
3905
3906
3907	/**
3908	* Sets up miscellaneous (everything other than Pin, Processor and secondary
3909	* Processor-based VM-execution) control fields in the VMCS.
3910	*
3911	* @returns VBox status code.
3912	* @param pVCpu The cross context virtual CPU structure.
3913	* @param pVmcsInfo The VMCS info. object.
3914	*/
3915	static int hmR0VmxSetupVmcsMiscCtls(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
3916	{
3917	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3918	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUseVmcsShadowing)
3919	{
3920	int rc = hmR0VmxSetupVmcsVmreadBitmapAddr(pVCpu);
3921	rc \|= hmR0VmxSetupVmcsVmwriteBitmapAddr(pVCpu);
3922	if (RT_SUCCESS(rc))
3923	{ /* likely */ }
3924	else
3925	{
3926	LogRelFunc(("Failed to setup VMREAD/VMWRITE bitmap addresses. rc=%Rrc\n", rc));
3927	return rc;
3928	}
3929	}
3930	#endif
3931
3932	Assert(pVmcsInfo->u64VmcsLinkPtr == NIL_RTHCPHYS);
3933	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
3934	if (RT_SUCCESS(rc))
3935	{
3936	rc = hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(pVCpu, pVmcsInfo);
3937	if (RT_SUCCESS(rc))
3938	{
3939	uint64_t const u64Cr0Mask = hmR0VmxGetFixedCr0Mask(pVCpu);
3940	uint64_t const u64Cr4Mask = hmR0VmxGetFixedCr4Mask(pVCpu);
3941	rc = VMXWriteVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, u64Cr0Mask);
3942	rc \|= VMXWriteVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, u64Cr4Mask);
3943	if (RT_SUCCESS(rc))
3944	{
3945	pVmcsInfo->u64Cr0Mask = u64Cr0Mask;
3946	pVmcsInfo->u64Cr4Mask = u64Cr4Mask;
3947	return VINF_SUCCESS;
3948	}
3949	LogRelFunc(("Failed to initialize VMCS CR0/CR4 guest/host mask. rc=%Rrc\n", rc));
3950	}
3951	else
3952	LogRelFunc(("Failed to initialize VMCS auto-load/store MSR addresses. rc=%Rrc\n", rc));
3953	}
3954	else
3955	LogRelFunc(("Failed to initialize VMCS link pointer. rc=%Rrc\n", rc));
3956	return rc;
3957	}
3958
3959
3960	/**
3961	* Sets up the initial exception bitmap in the VMCS based on static conditions.
3962	*
3963	* We shall setup those exception intercepts that don't change during the
3964	* lifetime of the VM here. The rest are done dynamically while loading the
3965	* guest state.
3966	*
3967	* @returns VBox status code.
3968	* @param pVCpu The cross context virtual CPU structure.
3969	* @param pVmcsInfo The VMCS info. object.
3970	*/
3971	static int hmR0VmxSetupVmcsXcptBitmap(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
3972	{
3973	/*
3974	* The following exceptions are always intercepted:
3975	*
3976	* #AC - To prevent the guest from hanging the CPU.
3977	* #DB - To maintain the DR6 state even when intercepting DRx reads/writes and
3978	* recursive #DBs can cause a CPU hang.
3979	* #PF - To sync our shadow page tables when nested-paging is not used.
3980	*/
3981	bool const fNestedPaging = pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging;
3982	uint32_t const uXcptBitmap = RT_BIT(X86_XCPT_AC)
3983	\| RT_BIT(X86_XCPT_DB)
3984	\| (fNestedPaging ? 0 : RT_BIT(X86_XCPT_PF));
3985
3986	/* Commit it to the VMCS. */
3987	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
3988	AssertRCReturn(rc, rc);
3989
3990	/* Update our cache of the exception bitmap. */
3991	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
3992	return VINF_SUCCESS;
3993	}
3994
3995
3996	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3997	/**
3998	* Sets up the VMCS for executing a nested-guest using hardware-assisted VMX.
3999	*
4000	* @returns VBox status code.
4001	* @param pVCpu The cross context virtual CPU structure.
4002	* @param pVmcsInfo The VMCS info. object.
4003	*/
4004	static int hmR0VmxSetupVmcsCtlsNested(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
4005	{
4006	PVM pVM = pVCpu->CTX_SUFF(pVM);
4007	Assert(pVmcsInfo->u64VmcsLinkPtr == NIL_RTHCPHYS);
4008	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
4009	if (RT_SUCCESS(rc))
4010	{
4011	rc = hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(pVCpu, pVmcsInfo);
4012	if (RT_SUCCESS(rc))
4013	{
4014	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
4015	rc = hmR0VmxSetupVmcsMsrBitmapAddr(pVCpu, pVmcsInfo);
4016	if (RT_SUCCESS(rc))
4017	return VINF_SUCCESS;
4018	LogRelFunc(("Failed to set up the MSR-bitmap address in the nested-guest VMCS. rc=%Rrc\n", rc));
4019	}
4020	else
4021	LogRelFunc(("Failed to set up the VMCS link pointer in the nested-guest VMCS. rc=%Rrc\n", rc));
4022	}
4023	else
4024	LogRelFunc(("Failed to set up the auto-load/store MSR addresses in the nested-guest VMCS. rc=%Rrc\n", rc));
4025
4026	return rc;
4027	}
4028	#endif
4029
4030
4031	/**
4032	* Sets up the VMCS for executing a guest (or nested-guest) using hardware-assisted
4033	* VMX.
4034	*
4035	* @returns VBox status code.
4036	* @param pVCpu The cross context virtual CPU structure.
4037	* @param pVmcsInfo The VMCS info. object.
4038	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
4039	*/
4040	static int hmR0VmxSetupVmcs(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
4041	{
4042	Assert(pVmcsInfo->pvVmcs);
4043	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4044
4045	/* Set the CPU specified revision identifier at the beginning of the VMCS structure. */
4046	PVM pVM = pVCpu->CTX_SUFF(pVM);
4047	(uint32_t )pVmcsInfo->pvVmcs = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_ID);
4048	const char * const pszVmcs = fIsNstGstVmcs ? "nested-guest VMCS" : "guest VMCS";
4049
4050	LogFlowFunc(("\n"));
4051
4052	/*
4053	* Initialize the VMCS using VMCLEAR before loading the VMCS.
4054	* See Intel spec. 31.6 "Preparation And Launching A Virtual Machine".
4055	*/
4056	int rc = hmR0VmxClearVmcs(pVmcsInfo);
4057	if (RT_SUCCESS(rc))
4058	{
4059	rc = hmR0VmxLoadVmcs(pVmcsInfo);
4060	if (RT_SUCCESS(rc))
4061	{
4062	if (!fIsNstGstVmcs)
4063	{
4064	rc = hmR0VmxSetupVmcsPinCtls(pVCpu, pVmcsInfo);
4065	if (RT_SUCCESS(rc))
4066	{
4067	rc = hmR0VmxSetupVmcsProcCtls(pVCpu, pVmcsInfo);
4068	if (RT_SUCCESS(rc))
4069	{
4070	rc = hmR0VmxSetupVmcsMiscCtls(pVCpu, pVmcsInfo);
4071	if (RT_SUCCESS(rc))
4072	{
4073	rc = hmR0VmxSetupVmcsXcptBitmap(pVCpu, pVmcsInfo);
4074	if (RT_SUCCESS(rc))
4075	{
4076	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4077	/*
4078	* If a shadow VMCS is allocated for the VMCS info. object, initialize the
4079	* VMCS revision ID and shadow VMCS indicator bit. Also, clear the VMCS
4080	* making it fit for use when VMCS shadowing is later enabled.
4081	*/
4082	if (pVmcsInfo->pvShadowVmcs)
4083	{
4084	VMXVMCSREVID VmcsRevId;
4085	VmcsRevId.u = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_ID);
4086	VmcsRevId.n.fIsShadowVmcs = 1;
4087	(uint32_t )pVmcsInfo->pvShadowVmcs = VmcsRevId.u;
4088	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
4089	if (RT_SUCCESS(rc))
4090	{ /* likely */ }
4091	else
4092	LogRelFunc(("Failed to initialize shadow VMCS. rc=%Rrc\n", rc));
4093	}
4094	#endif
4095	}
4096	else
4097	LogRelFunc(("Failed to initialize exception bitmap. rc=%Rrc\n", rc));
4098	}
4099	else
4100	LogRelFunc(("Failed to setup miscellaneous controls. rc=%Rrc\n", rc));
4101	}
4102	else
4103	LogRelFunc(("Failed to setup processor-based VM-execution controls. rc=%Rrc\n", rc));
4104	}
4105	else
4106	LogRelFunc(("Failed to setup pin-based controls. rc=%Rrc\n", rc));
4107	}
4108	else
4109	{
4110	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4111	rc = hmR0VmxSetupVmcsCtlsNested(pVCpu, pVmcsInfo);
4112	if (RT_SUCCESS(rc))
4113	{ /* likely */ }
4114	else
4115	LogRelFunc(("Failed to initialize nested-guest VMCS. rc=%Rrc\n", rc));
4116	#else
4117	AssertFailed();
4118	#endif
4119	}
4120	}
4121	else
4122	LogRelFunc(("Failed to load the %s. rc=%Rrc\n", rc, pszVmcs));
4123	}
4124	else
4125	LogRelFunc(("Failed to clear the %s. rc=%Rrc\n", rc, pszVmcs));
4126
4127	/* Sync any CPU internal VMCS data back into our VMCS in memory. */
4128	if (RT_SUCCESS(rc))
4129	{
4130	rc = hmR0VmxClearVmcs(pVmcsInfo);
4131	if (RT_SUCCESS(rc))
4132	{ /* likely */ }
4133	else
4134	LogRelFunc(("Failed to clear the %s post setup. rc=%Rrc\n", rc, pszVmcs));
4135	}
4136
4137	/*
4138	* Update the last-error record both for failures and success, so we
4139	* can propagate the status code back to ring-3 for diagnostics.
4140	*/
4141	hmR0VmxUpdateErrorRecord(pVCpu, rc);
4142	NOREF(pszVmcs);
4143	return rc;
4144	}
4145
4146
4147	/**
4148	* Does global VT-x initialization (called during module initialization).
4149	*
4150	* @returns VBox status code.
4151	*/
4152	VMMR0DECL(int) VMXR0GlobalInit(void)
4153	{
4154	#ifdef HMVMX_USE_FUNCTION_TABLE
4155	AssertCompile(VMX_EXIT_MAX + 1 == RT_ELEMENTS(g_apfnVMExitHandlers));
4156	# ifdef VBOX_STRICT
4157	for (unsigned i = 0; i < RT_ELEMENTS(g_apfnVMExitHandlers); i++)
4158	Assert(g_apfnVMExitHandlers[i]);
4159	# endif
4160	#endif
4161	return VINF_SUCCESS;
4162	}
4163
4164
4165	/**
4166	* Does global VT-x termination (called during module termination).
4167	*/
4168	VMMR0DECL(void) VMXR0GlobalTerm()
4169	{
4170	/* Nothing to do currently. */
4171	}
4172
4173
4174	/**
4175	* Sets up and activates VT-x on the current CPU.
4176	*
4177	* @returns VBox status code.
4178	* @param pHostCpu The HM physical-CPU structure.
4179	* @param pVM The cross context VM structure. Can be
4180	* NULL after a host resume operation.
4181	* @param pvCpuPage Pointer to the VMXON region (can be NULL if @a
4182	* fEnabledByHost is @c true).
4183	* @param HCPhysCpuPage Physical address of the VMXON region (can be 0 if
4184	* @a fEnabledByHost is @c true).
4185	* @param fEnabledByHost Set if SUPR0EnableVTx() or similar was used to
4186	* enable VT-x on the host.
4187	* @param pHwvirtMsrs Pointer to the hardware-virtualization MSRs.
4188	*/
4189	VMMR0DECL(int) VMXR0EnableCpu(PHMPHYSCPU pHostCpu, PVM pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost,
4190	PCSUPHWVIRTMSRS pHwvirtMsrs)
4191	{
4192	AssertPtr(pHostCpu);
4193	AssertPtr(pHwvirtMsrs);
4194	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4195
4196	/* Enable VT-x if it's not already enabled by the host. */
4197	if (!fEnabledByHost)
4198	{
4199	int rc = hmR0VmxEnterRootMode(pVM, HCPhysCpuPage, pvCpuPage);
4200	if (RT_FAILURE(rc))
4201	return rc;
4202	}
4203
4204	/*
4205	* Flush all EPT tagged-TLB entries (in case VirtualBox or any other hypervisor have been
4206	* using EPTPs) so we don't retain any stale guest-physical mappings which won't get
4207	* invalidated when flushing by VPID.
4208	*/
4209	if (pHwvirtMsrs->u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
4210	{
4211	hmR0VmxFlushEpt(NULL /* pVCpu /, NULL / pVmcsInfo */, VMXTLBFLUSHEPT_ALL_CONTEXTS);
4212	pHostCpu->fFlushAsidBeforeUse = false;
4213	}
4214	else
4215	pHostCpu->fFlushAsidBeforeUse = true;
4216
4217	/* Ensure each VCPU scheduled on this CPU gets a new VPID on resume. See @bugref{6255}. */
4218	++pHostCpu->cTlbFlushes;
4219
4220	return VINF_SUCCESS;
4221	}
4222
4223
4224	/**
4225	* Deactivates VT-x on the current CPU.
4226	*
4227	* @returns VBox status code.
4228	* @param pvCpuPage Pointer to the VMXON region.
4229	* @param HCPhysCpuPage Physical address of the VMXON region.
4230	*
4231	* @remarks This function should never be called when SUPR0EnableVTx() or
4232	* similar was used to enable VT-x on the host.
4233	*/
4234	VMMR0DECL(int) VMXR0DisableCpu(void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
4235	{
4236	RT_NOREF2(pvCpuPage, HCPhysCpuPage);
4237
4238	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4239	return hmR0VmxLeaveRootMode();
4240	}
4241
4242
4243	/**
4244	* Does per-VM VT-x initialization.
4245	*
4246	* @returns VBox status code.
4247	* @param pVM The cross context VM structure.
4248	*/
4249	VMMR0DECL(int) VMXR0InitVM(PVM pVM)
4250	{
4251	AssertPtr(pVM);
4252	LogFlowFunc(("pVM=%p\n", pVM));
4253
4254	int rc = hmR0VmxStructsAlloc(pVM);
4255	if (RT_FAILURE(rc))
4256	{
4257	LogRelFunc(("Failed to allocated VMX structures. rc=%Rrc\n", rc));
4258	return rc;
4259	}
4260
4261	return VINF_SUCCESS;
4262	}
4263
4264
4265	/**
4266	* Does per-VM VT-x termination.
4267	*
4268	* @returns VBox status code.
4269	* @param pVM The cross context VM structure.
4270	*/
4271	VMMR0DECL(int) VMXR0TermVM(PVM pVM)
4272	{
4273	AssertPtr(pVM);
4274	LogFlowFunc(("pVM=%p\n", pVM));
4275
4276	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4277	if (pVM->hm.s.vmx.hMemObjScratch != NIL_RTR0MEMOBJ)
4278	{
4279	Assert(pVM->hm.s.vmx.pvScratch);
4280	ASMMemZero32(pVM->hm.s.vmx.pvScratch, X86_PAGE_4K_SIZE);
4281	}
4282	#endif
4283	hmR0VmxStructsFree(pVM);
4284	return VINF_SUCCESS;
4285	}
4286
4287
4288	/**
4289	* Sets up the VM for execution using hardware-assisted VMX.
4290	* This function is only called once per-VM during initialization.
4291	*
4292	* @returns VBox status code.
4293	* @param pVM The cross context VM structure.
4294	*/
4295	VMMR0DECL(int) VMXR0SetupVM(PVM pVM)
4296	{
4297	AssertPtr(pVM);
4298	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4299
4300	LogFlowFunc(("pVM=%p\n", pVM));
4301
4302	/*
4303	* At least verify if VMX is enabled, since we can't check if we're in
4304	* VMX root mode or not without causing a #GP.
4305	*/
4306	RTCCUINTREG const uHostCr4 = ASMGetCR4();
4307	if (RT_LIKELY(uHostCr4 & X86_CR4_VMXE))
4308	{ /* likely */ }
4309	else
4310	return VERR_VMX_NOT_IN_VMX_ROOT_MODE;
4311
4312	/*
4313	* Without unrestricted guest execution, pRealModeTSS and pNonPagingModeEPTPageTable must
4314	* always be allocated. We no longer support the highly unlikely case of unrestricted guest
4315	* without pRealModeTSS, see hmR3InitFinalizeR0Intel().
4316	*/
4317	if ( !pVM->hm.s.vmx.fUnrestrictedGuest
4318	&& ( !pVM->hm.s.vmx.pNonPagingModeEPTPageTable
4319	\|\| !pVM->hm.s.vmx.pRealModeTSS))
4320	{
4321	LogRelFunc(("Invalid real-on-v86 state.\n"));
4322	return VERR_INTERNAL_ERROR;
4323	}
4324
4325	/* Initialize these always, see hmR3InitFinalizeR0().*/
4326	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NONE;
4327	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NONE;
4328
4329	/* Setup the tagged-TLB flush handlers. */
4330	int rc = hmR0VmxSetupTaggedTlb(pVM);
4331	if (RT_FAILURE(rc))
4332	{
4333	LogRelFunc(("Failed to setup tagged TLB. rc=%Rrc\n", rc));
4334	return rc;
4335	}
4336
4337	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4338	/* Setup the shadow VMCS fields array and VMREAD/VMWRITE bitmaps. */
4339	if (pVM->hm.s.vmx.fUseVmcsShadowing)
4340	{
4341	rc = hmR0VmxSetupShadowVmcsFieldsArrays(pVM);
4342	if (RT_SUCCESS(rc))
4343	hmR0VmxSetupVmreadVmwriteBitmaps(pVM);
4344	else
4345	{
4346	LogRelFunc(("Failed to setup shadow VMCS fields arrays. rc=%Rrc\n", rc));
4347	return rc;
4348	}
4349	}
4350	#endif
4351
4352	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
4353	{
4354	PVMCPU pVCpu = &pVM->aCpus[idCpu];
4355	Log4Func(("pVCpu=%p idCpu=%RU32\n", pVCpu, pVCpu->idCpu));
4356
4357	rc = hmR0VmxSetupVmcs(pVCpu, &pVCpu->hm.s.vmx.VmcsInfo, false /* fIsNstGstVmcs */);
4358	if (RT_SUCCESS(rc))
4359	{
4360	#if HC_ARCH_BITS == 32
4361	hmR0VmxInitVmcsReadCache(pVCpu);
4362	#endif
4363	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4364	if (pVM->cpum.ro.GuestFeatures.fVmx)
4365	{
4366	rc = hmR0VmxSetupVmcs(pVCpu, &pVCpu->hm.s.vmx.VmcsInfoNstGst, true /* fIsNstGstVmcs */);
4367	if (RT_SUCCESS(rc))
4368	{ /* likely */ }
4369	else
4370	{
4371	LogRelFunc(("Nested-guest VMCS setup failed. rc=%Rrc\n", rc));
4372	return rc;
4373	}
4374	}
4375	#endif
4376	}
4377	else
4378	{
4379	LogRelFunc(("VMCS setup failed. rc=%Rrc\n", rc));
4380	return rc;
4381	}
4382	}
4383
4384	return VINF_SUCCESS;
4385	}
4386
4387
4388	#if HC_ARCH_BITS == 32
4389	# ifdef VBOX_ENABLE_64_BITS_GUESTS
4390	/**
4391	* Check if guest state allows safe use of 32-bit switcher again.
4392	*
4393	* Segment bases and protected mode structures must be 32-bit addressable
4394	* because the 32-bit switcher will ignore high dword when writing these VMCS
4395	* fields. See @bugref{8432} for details.
4396	*
4397	* @returns true if safe, false if must continue to use the 64-bit switcher.
4398	* @param pCtx Pointer to the guest-CPU context.
4399	*
4400	* @remarks No-long-jump zone!!!
4401	*/
4402	static bool hmR0VmxIs32BitSwitcherSafe(PCCPUMCTX pCtx)
4403	{
4404	if (pCtx->gdtr.pGdt & UINT64_C(0xffffffff00000000)) return false;
4405	if (pCtx->idtr.pIdt & UINT64_C(0xffffffff00000000)) return false;
4406	if (pCtx->ldtr.u64Base & UINT64_C(0xffffffff00000000)) return false;
4407	if (pCtx->tr.u64Base & UINT64_C(0xffffffff00000000)) return false;
4408	if (pCtx->es.u64Base & UINT64_C(0xffffffff00000000)) return false;
4409	if (pCtx->cs.u64Base & UINT64_C(0xffffffff00000000)) return false;
4410	if (pCtx->ss.u64Base & UINT64_C(0xffffffff00000000)) return false;
4411	if (pCtx->ds.u64Base & UINT64_C(0xffffffff00000000)) return false;
4412	if (pCtx->fs.u64Base & UINT64_C(0xffffffff00000000)) return false;
4413	if (pCtx->gs.u64Base & UINT64_C(0xffffffff00000000)) return false;
4414
4415	/* All good, bases are 32-bit. */
4416	return true;
4417	}
4418	# endif /* VBOX_ENABLE_64_BITS_GUESTS */
4419
4420	# ifdef VBOX_STRICT
4421	static bool hmR0VmxIsValidWriteFieldInCache(uint32_t idxField)
4422	{
4423	switch (idxField)
4424	{
4425	case VMX_VMCS_GUEST_RIP:
4426	case VMX_VMCS_GUEST_RSP:
4427	case VMX_VMCS_GUEST_SYSENTER_EIP:
4428	case VMX_VMCS_GUEST_SYSENTER_ESP:
4429	case VMX_VMCS_GUEST_GDTR_BASE:
4430	case VMX_VMCS_GUEST_IDTR_BASE:
4431	case VMX_VMCS_GUEST_CS_BASE:
4432	case VMX_VMCS_GUEST_DS_BASE:
4433	case VMX_VMCS_GUEST_ES_BASE:
4434	case VMX_VMCS_GUEST_FS_BASE:
4435	case VMX_VMCS_GUEST_GS_BASE:
4436	case VMX_VMCS_GUEST_SS_BASE:
4437	case VMX_VMCS_GUEST_LDTR_BASE:
4438	case VMX_VMCS_GUEST_TR_BASE:
4439	case VMX_VMCS_GUEST_CR3:
4440	return true;
4441	}
4442	return false;
4443	}
4444
4445	static bool hmR0VmxIsValidReadFieldInCache(uint32_t idxField)
4446	{
4447	switch (idxField)
4448	{
4449	/* Read-only fields. */
4450	case VMX_VMCS_RO_EXIT_QUALIFICATION:
4451	return true;
4452	}
4453	/* Remaining readable fields should also be writable. */
4454	return hmR0VmxIsValidWriteFieldInCache(idxField);
4455	}
4456	# endif /* VBOX_STRICT */
4457
4458
4459	/**
4460	* Executes the specified handler in 64-bit mode.
4461	*
4462	* @returns VBox status code (no informational status codes).
4463	* @param pVCpu The cross context virtual CPU structure.
4464	* @param enmOp The operation to perform.
4465	* @param cParams Number of parameters.
4466	* @param paParam Array of 32-bit parameters.
4467	*/
4468	VMMR0DECL(int) VMXR0Execute64BitsHandler(PVMCPU pVCpu, HM64ON32OP enmOp, uint32_t cParams, uint32_t *paParam)
4469	{
4470	AssertPtr(pVCpu);
4471	PVM pVM = pVCpu->CTX_SUFF(pVM);
4472	AssertReturn(pVM->hm.s.pfnHost32ToGuest64R0, VERR_HM_NO_32_TO_64_SWITCHER);
4473	Assert(enmOp > HM64ON32OP_INVALID && enmOp < HM64ON32OP_END);
4474	Assert(pVCpu->hm.s.vmx.VmcsCache.Write.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VmcsCache.Write.aField));
4475	Assert(pVCpu->hm.s.vmx.VmcsCache.Read.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VmcsCache.Read.aField));
4476
4477	#ifdef VBOX_STRICT
4478	for (uint32_t i = 0; i < pVCpu->hm.s.vmx.VmcsCache.Write.cValidEntries; i++)
4479	Assert(hmR0VmxIsValidWriteFieldInCache(pVCpu->hm.s.vmx.VmcsCache.Write.aField[i]));
4480
4481	for (uint32_t i = 0; i <pVCpu->hm.s.vmx.VmcsCache.Read.cValidEntries; i++)
4482	Assert(hmR0VmxIsValidReadFieldInCache(pVCpu->hm.s.vmx.VmcsCache.Read.aField[i]));
4483	#endif
4484
4485	/* Disable interrupts. */
4486	RTCCUINTREG fOldEFlags = ASMIntDisableFlags();
4487
4488	#ifdef VBOX_WITH_VMMR0_DISABLE_LAPIC_NMI
4489	RTCPUID idHostCpu = RTMpCpuId();
4490	CPUMR0SetLApic(pVCpu, idHostCpu);
4491	#endif
4492
4493	/** @todo replace with hmR0VmxEnterRootMode() and hmR0VmxLeaveRootMode(). */
4494
4495	PCHMPHYSCPU pHostCpu = hmR0GetCurrentCpu();
4496	RTHCPHYS const HCPhysCpuPage = pHostCpu->HCPhysMemObj;
4497
4498	/* Clear VMCS. Marking it inactive, clearing implementation-specific data and writing VMCS data back to memory. */
4499	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
4500	hmR0VmxClearVmcs(pVmcsInfo);
4501
4502	/* Leave VMX root mode and disable VMX. */
4503	VMXDisable();
4504	SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
4505
4506	CPUMSetHyperESP(pVCpu, VMMGetStackRC(pVCpu));
4507	CPUMSetHyperEIP(pVCpu, enmOp);
4508	for (int i = (int)cParams - 1; i >= 0; i--)
4509	CPUMPushHyper(pVCpu, paParam[i]);
4510
4511	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatWorldSwitch3264, z);
4512
4513	/* Call the switcher. */
4514	int rc = pVM->hm.s.pfnHost32ToGuest64R0(pVM, RT_UOFFSETOF_DYN(VM, aCpus[pVCpu->idCpu].cpum) - RT_UOFFSETOF(VM, cpum));
4515	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatWorldSwitch3264, z);
4516
4517	/* Re-enable VMX to make sure the VMX instructions don't cause #UD faults. */
4518	SUPR0ChangeCR4(X86_CR4_VMXE, RTCCUINTREG_MAX);
4519
4520	/* Re-enter VMX root mode. */
4521	int rc2 = VMXEnable(HCPhysCpuPage);
4522	if (RT_FAILURE(rc2))
4523	{
4524	SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
4525	ASMSetFlags(fOldEFlags);
4526	pVM->hm.s.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
4527	return rc2;
4528	}
4529
4530	/* Restore the VMCS as the current VMCS. */
4531	rc2 = hmR0VmxLoadVmcs(pVmcsInfo);
4532	AssertRC(rc2);
4533	Assert(!(ASMGetFlags() & X86_EFL_IF));
4534	ASMSetFlags(fOldEFlags);
4535	return rc;
4536	}
4537
4538
4539	/**
4540	* Prepares for and executes VMLAUNCH (64-bit guests) for 32-bit hosts
4541	* supporting 64-bit guests.
4542	*
4543	* @returns VBox status code.
4544	* @param fResume Whether to VMLAUNCH or VMRESUME.
4545	* @param pCtx Pointer to the guest-CPU context.
4546	* @param pCache Pointer to the VMCS batch cache.
4547	* @param pVM The cross context VM structure.
4548	* @param pVCpu The cross context virtual CPU structure.
4549	*/
4550	DECLASM(int) VMXR0SwitcherStartVM64(RTHCUINT fResume, PCPUMCTX pCtx, PVMXVMCSCACHE pCache, PVM pVM, PVMCPU pVCpu)
4551	{
4552	NOREF(fResume);
4553
4554	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
4555	PCHMPHYSCPU pHostCpu = hmR0GetCurrentCpu();
4556	RTHCPHYS const HCPhysCpuPage = pHostCpu->HCPhysMemObj;
4557
4558	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4559	pCache->uPos = 1;
4560	pCache->interPD = PGMGetInterPaeCR3(pVM);
4561	pCache->pSwitcher = (uint64_t)pVM->hm.s.pfnHost32ToGuest64R0;
4562	#endif
4563
4564	#if defined(DEBUG) && defined(VMX_USE_CACHED_VMCS_ACCESSES)
4565	pCache->TestIn.HCPhysCpuPage = 0;
4566	pCache->TestIn.HCPhysVmcs = 0;
4567	pCache->TestIn.pCache = 0;
4568	pCache->TestOut.HCPhysVmcs = 0;
4569	pCache->TestOut.pCache = 0;
4570	pCache->TestOut.pCtx = 0;
4571	pCache->TestOut.eflags = 0;
4572	#else
4573	NOREF(pCache);
4574	#endif
4575
4576	uint32_t aParam[10];
4577	aParam[0] = RT_LO_U32(HCPhysCpuPage); /* Param 1: VMXON physical address - Lo. */
4578	aParam[1] = RT_HI_U32(HCPhysCpuPage); /* Param 1: VMXON physical address - Hi. */
4579	aParam[2] = RT_LO_U32(pVmcsInfo->HCPhysVmcs); /* Param 2: VMCS physical address - Lo. */
4580	aParam[3] = RT_HI_U32(pVmcsInfo->HCPhysVmcs); /* Param 2: VMCS physical address - Hi. */
4581	aParam[4] = VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VmcsCache);
4582	aParam[5] = 0;
4583	aParam[6] = VM_RC_ADDR(pVM, pVM);
4584	aParam[7] = 0;
4585	aParam[8] = VM_RC_ADDR(pVM, pVCpu);
4586	aParam[9] = 0;
4587
4588	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4589	pCtx->dr[4] = pVM->hm.s.vmx.pScratchPhys + 16 + 8;
4590	(uint32_t )(pVM->hm.s.vmx.pScratch + 16 + 8) = 1;
4591	#endif
4592	int rc = VMXR0Execute64BitsHandler(pVCpu, HM64ON32OP_VMXRCStartVM64, RT_ELEMENTS(aParam), &aParam[0]);
4593
4594	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4595	Assert((uint32_t )(pVM->hm.s.vmx.pScratch + 16 + 8) == 5);
4596	Assert(pCtx->dr[4] == 10);
4597	(uint32_t )(pVM->hm.s.vmx.pScratch + 16 + 8) = 0xff;
4598	#endif
4599
4600	#if defined(DEBUG) && defined(VMX_USE_CACHED_VMCS_ACCESSES)
4601	AssertMsg(pCache->TestIn.HCPhysCpuPage == HCPhysCpuPage, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysCpuPage, HCPhysCpuPage));
4602	AssertMsg(pCache->TestIn.HCPhysVmcs == pVmcsInfo->HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs,
4603	pVmcsInfo->HCPhysVmcs));
4604	AssertMsg(pCache->TestIn.HCPhysVmcs == pCache->TestOut.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs,
4605	pCache->TestOut.HCPhysVmcs));
4606	AssertMsg(pCache->TestIn.pCache == pCache->TestOut.pCache, ("%RGv vs %RGv\n", pCache->TestIn.pCache,
4607	pCache->TestOut.pCache));
4608	AssertMsg(pCache->TestIn.pCache == VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VmcsCache),
4609	("%RGv vs %RGv\n", pCache->TestIn.pCache, VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VmcsCache)));
4610	AssertMsg(pCache->TestIn.pCtx == pCache->TestOut.pCtx, ("%RGv vs %RGv\n", pCache->TestIn.pCtx,
4611	pCache->TestOut.pCtx));
4612	Assert(!(pCache->TestOut.eflags & X86_EFL_IF));
4613	#endif
4614	NOREF(pCtx);
4615	return rc;
4616	}
4617	#endif
4618
4619
4620	/**
4621	* Saves the host control registers (CR0, CR3, CR4) into the host-state area in
4622	* the VMCS.
4623	*
4624	* @returns VBox status code.
4625	*/
4626	static int hmR0VmxExportHostControlRegs(void)
4627	{
4628	RTCCUINTREG uReg = ASMGetCR0();
4629	int rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR0, uReg);
4630	AssertRCReturn(rc, rc);
4631
4632	uReg = ASMGetCR3();
4633	rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR3, uReg);
4634	AssertRCReturn(rc, rc);
4635
4636	uReg = ASMGetCR4();
4637	rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR4, uReg);
4638	AssertRCReturn(rc, rc);
4639	return rc;
4640	}
4641
4642
4643	/**
4644	* Saves the host segment registers and GDTR, IDTR, (TR, GS and FS bases) into
4645	* the host-state area in the VMCS.
4646	*
4647	* @returns VBox status code.
4648	* @param pVCpu The cross context virtual CPU structure.
4649	*/
4650	static int hmR0VmxExportHostSegmentRegs(PVMCPU pVCpu)
4651	{
4652	#if HC_ARCH_BITS == 64
4653	/**
4654	* Macro for adjusting host segment selectors to satisfy VT-x's VM-entry
4655	* requirements. See hmR0VmxExportHostSegmentRegs().
4656	*/
4657	# define VMXLOCAL_ADJUST_HOST_SEG(seg, selValue) \
4658	if ((selValue) & (X86_SEL_RPL \| X86_SEL_LDT)) \
4659	{ \
4660	bool fValidSelector = true; \
4661	if ((selValue) & X86_SEL_LDT) \
4662	{ \
4663	uint32_t uAttr = ASMGetSegAttr((selValue)); \
4664	fValidSelector = RT_BOOL(uAttr != UINT32_MAX && (uAttr & X86_DESC_P)); \
4665	} \
4666	if (fValidSelector) \
4667	{ \
4668	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##seg; \
4669	pVCpu->hm.s.vmx.RestoreHost.uHostSel##seg = (selValue); \
4670	} \
4671	(selValue) = 0; \
4672	}
4673
4674	/*
4675	* If we've executed guest code using hardware-assisted VMX, the host-state bits
4676	* will be messed up. We should -not- save the messed up state without restoring
4677	* the original host-state, see @bugref{7240}.
4678	*
4679	* This apparently can happen (most likely the FPU changes), deal with it rather than
4680	* asserting. Was observed booting Solaris 10u10 32-bit guest.
4681	*/
4682	if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
4683	&& (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
4684	{
4685	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags,
4686	pVCpu->idCpu));
4687	VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
4688	}
4689	pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
4690	#else
4691	RT_NOREF(pVCpu);
4692	#endif
4693
4694	/*
4695	* Host DS, ES, FS and GS segment registers.
4696	*/
4697	#if HC_ARCH_BITS == 64
4698	RTSEL uSelDS = ASMGetDS();
4699	RTSEL uSelES = ASMGetES();
4700	RTSEL uSelFS = ASMGetFS();
4701	RTSEL uSelGS = ASMGetGS();
4702	#else
4703	RTSEL uSelDS = 0;
4704	RTSEL uSelES = 0;
4705	RTSEL uSelFS = 0;
4706	RTSEL uSelGS = 0;
4707	#endif
4708
4709	/*
4710	* Host CS and SS segment registers.
4711	*/
4712	RTSEL uSelCS = ASMGetCS();
4713	RTSEL uSelSS = ASMGetSS();
4714
4715	/*
4716	* Host TR segment register.
4717	*/
4718	RTSEL uSelTR = ASMGetTR();
4719
4720	#if HC_ARCH_BITS == 64
4721	/*
4722	* Determine if the host segment registers are suitable for VT-x. Otherwise use zero to
4723	* gain VM-entry and restore them before we get preempted.
4724	*
4725	* See Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers".
4726	*/
4727	VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS);
4728	VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES);
4729	VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS);
4730	VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS);
4731	# undef VMXLOCAL_ADJUST_HOST_SEG
4732	#endif
4733
4734	/* Verification based on Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers" */
4735	Assert(!(uSelCS & X86_SEL_RPL)); Assert(!(uSelCS & X86_SEL_LDT));
4736	Assert(!(uSelSS & X86_SEL_RPL)); Assert(!(uSelSS & X86_SEL_LDT));
4737	Assert(!(uSelDS & X86_SEL_RPL)); Assert(!(uSelDS & X86_SEL_LDT));
4738	Assert(!(uSelES & X86_SEL_RPL)); Assert(!(uSelES & X86_SEL_LDT));
4739	Assert(!(uSelFS & X86_SEL_RPL)); Assert(!(uSelFS & X86_SEL_LDT));
4740	Assert(!(uSelGS & X86_SEL_RPL)); Assert(!(uSelGS & X86_SEL_LDT));
4741	Assert(!(uSelTR & X86_SEL_RPL)); Assert(!(uSelTR & X86_SEL_LDT));
4742	Assert(uSelCS);
4743	Assert(uSelTR);
4744
4745	/* Write these host selector fields into the host-state area in the VMCS. */
4746	int rc = VMXWriteVmcs32(VMX_VMCS16_HOST_CS_SEL, uSelCS);
4747	rc \|= VMXWriteVmcs32(VMX_VMCS16_HOST_SS_SEL, uSelSS);
4748	#if HC_ARCH_BITS == 64
4749	rc \|= VMXWriteVmcs32(VMX_VMCS16_HOST_DS_SEL, uSelDS);
4750	rc \|= VMXWriteVmcs32(VMX_VMCS16_HOST_ES_SEL, uSelES);
4751	rc \|= VMXWriteVmcs32(VMX_VMCS16_HOST_FS_SEL, uSelFS);
4752	rc \|= VMXWriteVmcs32(VMX_VMCS16_HOST_GS_SEL, uSelGS);
4753	#else
4754	NOREF(uSelDS);
4755	NOREF(uSelES);
4756	NOREF(uSelFS);
4757	NOREF(uSelGS);
4758	#endif
4759	rc \|= VMXWriteVmcs32(VMX_VMCS16_HOST_TR_SEL, uSelTR);
4760	AssertRCReturn(rc, rc);
4761
4762	/*
4763	* Host GDTR and IDTR.
4764	*/
4765	RTGDTR Gdtr;
4766	RTIDTR Idtr;
4767	RT_ZERO(Gdtr);
4768	RT_ZERO(Idtr);
4769	ASMGetGDTR(&Gdtr);
4770	ASMGetIDTR(&Idtr);
4771	rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_GDTR_BASE, Gdtr.pGdt);
4772	rc \|= VMXWriteVmcsHstN(VMX_VMCS_HOST_IDTR_BASE, Idtr.pIdt);
4773	AssertRCReturn(rc, rc);
4774
4775	#if HC_ARCH_BITS == 64
4776	/*
4777	* Determine if we need to manually need to restore the GDTR and IDTR limits as VT-x zaps
4778	* them to the maximum limit (0xffff) on every VM-exit.
4779	*/
4780	if (Gdtr.cbGdt != 0xffff)
4781	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_GDTR;
4782
4783	/*
4784	* IDT limit is effectively capped at 0xfff. (See Intel spec. 6.14.1 "64-Bit Mode IDT" and
4785	* Intel spec. 6.2 "Exception and Interrupt Vectors".) Therefore if the host has the limit
4786	* as 0xfff, VT-x bloating the limit to 0xffff shouldn't cause any different CPU behavior.
4787	* However, several hosts either insists on 0xfff being the limit (Windows Patch Guard) or
4788	* uses the limit for other purposes (darwin puts the CPU ID in there but botches sidt
4789	* alignment in at least one consumer). So, we're only allowing the IDTR.LIMIT to be left
4790	* at 0xffff on hosts where we are sure it won't cause trouble.
4791	*/
4792	# if defined(RT_OS_LINUX) \|\| defined(RT_OS_SOLARIS)
4793	if (Idtr.cbIdt < 0x0fff)
4794	# else
4795	if (Idtr.cbIdt != 0xffff)
4796	# endif
4797	{
4798	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_IDTR;
4799	AssertCompile(sizeof(Idtr) == sizeof(X86XDTR64));
4800	memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostIdtr, &Idtr, sizeof(X86XDTR64));
4801	}
4802	#endif
4803
4804	/*
4805	* Host TR base. Verify that TR selector doesn't point past the GDT. Masking off the TI
4806	* and RPL bits is effectively what the CPU does for "scaling by 8". TI is always 0 and
4807	* RPL should be too in most cases.
4808	*/
4809	AssertMsgReturn((uSelTR \| X86_SEL_RPL_LDT) <= Gdtr.cbGdt,
4810	("TR selector exceeds limit. TR=%RTsel cbGdt=%#x\n", uSelTR, Gdtr.cbGdt), VERR_VMX_INVALID_HOST_STATE);
4811
4812	PCX86DESCHC pDesc = (PCX86DESCHC)(Gdtr.pGdt + (uSelTR & X86_SEL_MASK));
4813	#if HC_ARCH_BITS == 64
4814	uintptr_t const uTRBase = X86DESC64_BASE(pDesc);
4815
4816	/*
4817	* VT-x unconditionally restores the TR limit to 0x67 and type to 11 (32-bit busy TSS) on
4818	* all VM-exits. The type is the same for 64-bit busy TSS[1]. The limit needs manual
4819	* restoration if the host has something else. Task switching is not supported in 64-bit
4820	* mode[2], but the limit still matters as IOPM is supported in 64-bit mode. Restoring the
4821	* limit lazily while returning to ring-3 is safe because IOPM is not applicable in ring-0.
4822	*
4823	* [1] See Intel spec. 3.5 "System Descriptor Types".
4824	* [2] See Intel spec. 7.2.3 "TSS Descriptor in 64-bit mode".
4825	*/
4826	PVM pVM = pVCpu->CTX_SUFF(pVM);
4827	Assert(pDesc->System.u4Type == 11);
4828	if ( pDesc->System.u16LimitLow != 0x67
4829	\|\| pDesc->System.u4LimitHigh)
4830	{
4831	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_TR;
4832	/* If the host has made GDT read-only, we would need to temporarily toggle CR0.WP before writing the GDT. */
4833	if (pVM->hm.s.fHostKernelFeatures & SUPKERNELFEATURES_GDT_READ_ONLY)
4834	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_READ_ONLY;
4835	pVCpu->hm.s.vmx.RestoreHost.uHostSelTR = uSelTR;
4836	}
4837
4838	/*
4839	* Store the GDTR as we need it when restoring the GDT and while restoring the TR.
4840	*/
4841	if (pVCpu->hm.s.vmx.fRestoreHostFlags & (VMX_RESTORE_HOST_GDTR \| VMX_RESTORE_HOST_SEL_TR))
4842	{
4843	AssertCompile(sizeof(Gdtr) == sizeof(X86XDTR64));
4844	memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64));
4845	if (pVM->hm.s.fHostKernelFeatures & SUPKERNELFEATURES_GDT_NEED_WRITABLE)
4846	{
4847	/* The GDT is read-only but the writable GDT is available. */
4848	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_NEED_WRITABLE;
4849	pVCpu->hm.s.vmx.RestoreHost.HostGdtrRw.cb = Gdtr.cbGdt;
4850	rc = SUPR0GetCurrentGdtRw(&pVCpu->hm.s.vmx.RestoreHost.HostGdtrRw.uAddr);
4851	AssertRCReturn(rc, rc);
4852	}
4853	}
4854	#else
4855	uintptr_t const uTRBase = X86DESC_BASE(pDesc);
4856	#endif
4857	rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_TR_BASE, uTRBase);
4858	AssertRCReturn(rc, rc);
4859
4860	/*
4861	* Host FS base and GS base.
4862	*/
4863	#if HC_ARCH_BITS == 64
4864	uint64_t const u64FSBase = ASMRdMsr(MSR_K8_FS_BASE);
4865	uint64_t const u64GSBase = ASMRdMsr(MSR_K8_GS_BASE);
4866	rc = VMXWriteVmcs64(VMX_VMCS_HOST_FS_BASE, u64FSBase);
4867	rc \|= VMXWriteVmcs64(VMX_VMCS_HOST_GS_BASE, u64GSBase);
4868	AssertRCReturn(rc, rc);
4869
4870	/* Store the base if we have to restore FS or GS manually as we need to restore the base as well. */
4871	if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_FS)
4872	pVCpu->hm.s.vmx.RestoreHost.uHostFSBase = u64FSBase;
4873	if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_GS)
4874	pVCpu->hm.s.vmx.RestoreHost.uHostGSBase = u64GSBase;
4875	#endif
4876	return VINF_SUCCESS;
4877	}
4878
4879
4880	/**
4881	* Exports certain host MSRs in the VM-exit MSR-load area and some in the
4882	* host-state area of the VMCS.
4883	*
4884	* These MSRs will be automatically restored on the host after every successful
4885	* VM-exit.
4886	*
4887	* @returns VBox status code.
4888	* @param pVCpu The cross context virtual CPU structure.
4889	*
4890	* @remarks No-long-jump zone!!!
4891	*/
4892	static int hmR0VmxExportHostMsrs(PVMCPU pVCpu)
4893	{
4894	AssertPtr(pVCpu);
4895
4896	/*
4897	* Save MSRs that we restore lazily (due to preemption or transition to ring-3)
4898	* rather than swapping them on every VM-entry.
4899	*/
4900	hmR0VmxLazySaveHostMsrs(pVCpu);
4901
4902	/*
4903	* Host Sysenter MSRs.
4904	*/
4905	int rc = VMXWriteVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, ASMRdMsr_Low(MSR_IA32_SYSENTER_CS));
4906	#if HC_ARCH_BITS == 32
4907	rc \|= VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr_Low(MSR_IA32_SYSENTER_ESP));
4908	rc \|= VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr_Low(MSR_IA32_SYSENTER_EIP));
4909	#else
4910	rc \|= VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP));
4911	rc \|= VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP));
4912	#endif
4913	AssertRCReturn(rc, rc);
4914
4915	/*
4916	* Host EFER MSR.
4917	*
4918	* If the CPU supports the newer VMCS controls for managing EFER, use it. Otherwise it's
4919	* done as part of auto-load/store MSR area in the VMCS, see hmR0VmxExportGuestMsrs().
4920	*/
4921	PVM pVM = pVCpu->CTX_SUFF(pVM);
4922	if (pVM->hm.s.vmx.fSupportsVmcsEfer)
4923	{
4924	rc = VMXWriteVmcs64(VMX_VMCS64_HOST_EFER_FULL, pVM->hm.s.vmx.u64HostMsrEfer);
4925	AssertRCReturn(rc, rc);
4926	}
4927
4928	/** @todo IA32_PERF_GLOBALCTRL, IA32_PAT also see
4929	* hmR0VmxExportGuestEntryExitCtls(). */
4930
4931	return VINF_SUCCESS;
4932	}
4933
4934
4935	/**
4936	* Figures out if we need to swap the EFER MSR which is particularly expensive.
4937	*
4938	* We check all relevant bits. For now, that's everything besides LMA/LME, as
4939	* these two bits are handled by VM-entry, see hmR0VMxExportGuestEntryExitCtls().
4940	*
4941	* @returns true if we need to load guest EFER, false otherwise.
4942	* @param pVCpu The cross context virtual CPU structure.
4943	*
4944	* @remarks Requires EFER, CR4.
4945	* @remarks No-long-jump zone!!!
4946	*/
4947	static bool hmR0VmxShouldSwapEferMsr(PCVMCPU pVCpu)
4948	{
4949	#ifdef HMVMX_ALWAYS_SWAP_EFER
4950	RT_NOREF(pVCpu);
4951	return true;
4952	#else
4953	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
4954	#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
4955	/* For 32-bit hosts running 64-bit guests, we always swap EFER MSR in the world-switcher. Nothing to do here. */
4956	if (CPUMIsGuestInLongModeEx(pCtx))
4957	return false;
4958	#endif
4959
4960	PVM pVM = pVCpu->CTX_SUFF(pVM);
4961	uint64_t const u64HostEfer = pVM->hm.s.vmx.u64HostMsrEfer;
4962	uint64_t const u64GuestEfer = pCtx->msrEFER;
4963
4964	/*
4965	* For 64-bit guests, if EFER.SCE bit differs, we need to swap the EFER MSR
4966	* to ensure that the guest's SYSCALL behaviour isn't broken, see @bugref{7386}.
4967	*/
4968	if ( CPUMIsGuestInLongModeEx(pCtx)
4969	&& (u64GuestEfer & MSR_K6_EFER_SCE) != (u64HostEfer & MSR_K6_EFER_SCE))
4970	return true;
4971
4972	/*
4973	* If the guest uses PAE and EFER.NXE bit differs, we need to swap the EFER MSR
4974	* as it affects guest paging. 64-bit paging implies CR4.PAE as well.
4975	*
4976	* See Intel spec. 4.5 "IA-32e Paging".
4977	* See Intel spec. 4.1.1 "Three Paging Modes".
4978	*
4979	* Verify that we always intercept CR4.PAE and CR0.PG bits, so we don't need to
4980	* import CR4 and CR0 from the VMCS here as those bits are always up to date.
4981	*/
4982	Assert(hmR0VmxGetFixedCr4Mask(pVCpu) & X86_CR4_PAE);
4983	Assert(hmR0VmxGetFixedCr0Mask(pVCpu) & X86_CR0_PG);
4984	if ( (pCtx->cr4 & X86_CR4_PAE)
4985	&& (pCtx->cr0 & X86_CR0_PG)
4986	&& (u64GuestEfer & MSR_K6_EFER_NXE) != (u64HostEfer & MSR_K6_EFER_NXE))
4987	{
4988	/* Assert that host is NX capable. */
4989	Assert(pVCpu->CTX_SUFF(pVM)->cpum.ro.HostFeatures.fNoExecute);
4990	return true;
4991	}
4992
4993	return false;
4994	#endif
4995	}
4996
4997	/**
4998	* Exports the guest state with appropriate VM-entry and VM-exit controls in the
4999	* VMCS.
5000	*
5001	* This is typically required when the guest changes paging mode.
5002	*
5003	* @returns VBox status code.
5004	* @param pVCpu The cross context virtual CPU structure.
5005	* @param pVmxTransient The VMX-transient structure.
5006	*
5007	* @remarks Requires EFER.
5008	* @remarks No-long-jump zone!!!
5009	*/
5010	static int hmR0VmxExportGuestEntryExitCtls(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
5011	{
5012	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_ENTRY_EXIT_CTLS)
5013	{
5014	PVM pVM = pVCpu->CTX_SUFF(pVM);
5015	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5016
5017	/*
5018	* VM-entry controls.
5019	*/
5020	{
5021	uint32_t fVal = pVM->hm.s.vmx.Msrs.EntryCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
5022	uint32_t const fZap = pVM->hm.s.vmx.Msrs.EntryCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
5023
5024	/*
5025	* Load the guest debug controls (DR7 and IA32_DEBUGCTL MSR) on VM-entry.
5026	* The first VT-x capable CPUs only supported the 1-setting of this bit.
5027	*
5028	* For nested-guests, this is a mandatory VM-entry control. It's also
5029	* required because we do not want to leak host bits to the nested-guest.
5030	*/
5031	fVal \|= VMX_ENTRY_CTLS_LOAD_DEBUG;
5032
5033	/*
5034	* Set if the guest is in long mode. This will set/clear the EFER.LMA bit on VM-entry.
5035	*
5036	* For nested-guests, the "IA-32e mode guest" control we initialize with what is
5037	* required to get the nested-guest working with hardware-assisted VMX execution.
5038	* It depends on the nested-guest's IA32_EFER.LMA bit. Remember, a guest hypervisor
5039	* can skip intercepting changes to the EFER MSR. This is why it it needs to be done
5040	* here rather than while merging the guest VMCS controls.
5041	*/
5042	if (CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
5043	fVal \|= VMX_ENTRY_CTLS_IA32E_MODE_GUEST;
5044	else
5045	Assert(!(fVal & VMX_ENTRY_CTLS_IA32E_MODE_GUEST));
5046
5047	/*
5048	* If the CPU supports the newer VMCS controls for managing guest/host EFER, use it.
5049	*
5050	* For nested-guests, we use the "load IA32_EFER" if the hardware supports it,
5051	* regardless of whether the nested-guest VMCS specifies it because we are free to
5052	* load whatever MSRs we require and we do not need to modify the guest visible copy
5053	* of the VM-entry MSR load area.
5054	*/
5055	if ( pVM->hm.s.vmx.fSupportsVmcsEfer
5056	&& hmR0VmxShouldSwapEferMsr(pVCpu))
5057	fVal \|= VMX_ENTRY_CTLS_LOAD_EFER_MSR;
5058	else
5059	Assert(!(fVal & VMX_ENTRY_CTLS_LOAD_EFER_MSR));
5060
5061	/*
5062	* The following should -not- be set (since we're not in SMM mode):
5063	* - VMX_ENTRY_CTLS_ENTRY_TO_SMM
5064	* - VMX_ENTRY_CTLS_DEACTIVATE_DUAL_MON
5065	*/
5066
5067	/** @todo VMX_ENTRY_CTLS_LOAD_PERF_MSR,
5068	* VMX_ENTRY_CTLS_LOAD_PAT_MSR. */
5069
5070	if ((fVal & fZap) == fVal)
5071	{ /* likely */ }
5072	else
5073	{
5074	Log4Func(("Invalid VM-entry controls combo! Cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
5075	pVM->hm.s.vmx.Msrs.EntryCtls.n.allowed0, fVal, fZap));
5076	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_ENTRY;
5077	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
5078	}
5079
5080	/* Commit it to the VMCS. */
5081	if (pVmcsInfo->u32EntryCtls != fVal)
5082	{
5083	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY, fVal);
5084	AssertRCReturn(rc, rc);
5085	pVmcsInfo->u32EntryCtls = fVal;
5086	}
5087	}
5088
5089	/*
5090	* VM-exit controls.
5091	*/
5092	{
5093	uint32_t fVal = pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
5094	uint32_t const fZap = pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
5095
5096	/*
5097	* Save debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x CPUs only
5098	* supported the 1-setting of this bit.
5099	*
5100	* For nested-guests, we set the "save debug controls" as the converse
5101	* "load debug controls" is mandatory for nested-guests anyway.
5102	*/
5103	fVal \|= VMX_EXIT_CTLS_SAVE_DEBUG;
5104
5105	/*
5106	* Set the host long mode active (EFER.LMA) bit (which Intel calls
5107	* "Host address-space size") if necessary. On VM-exit, VT-x sets both the
5108	* host EFER.LMA and EFER.LME bit to this value. See assertion in
5109	* hmR0VmxExportHostMsrs().
5110	*
5111	* For nested-guests, we always set this bit as we do not support 32-bit
5112	* hosts.
5113	*/
5114	#if HC_ARCH_BITS == 64
5115	fVal \|= VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE;
5116	#else
5117	Assert(!pVmxTransient->fIsNestedGuest);
5118	Assert( pVmcsInfo->pfnStartVM == VMXR0SwitcherStartVM64
5119	\|\| pVmcsInfo->pfnStartVM == VMXR0StartVM32);
5120	/* Set the host address-space size based on the switcher, not guest state. See @bugref{8432}. */
5121	if (pVmcsInfo->pfnStartVM == VMXR0SwitcherStartVM64)
5122	{
5123	/* The switcher returns to long mode, the EFER MSR is managed by the switcher. */
5124	fVal \|= VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE;
5125	}
5126	else
5127	Assert(!(fVal & VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE));
5128	#endif
5129
5130	/*
5131	* If the VMCS EFER MSR fields are supported by the hardware, we use it.
5132	*
5133	* For nested-guests, we should use the "save IA32_EFER" control if we also
5134	* used the "load IA32_EFER" control while exporting VM-entry controls.
5135	*/
5136	if ( pVM->hm.s.vmx.fSupportsVmcsEfer
5137	&& hmR0VmxShouldSwapEferMsr(pVCpu))
5138	{
5139	fVal \|= VMX_EXIT_CTLS_SAVE_EFER_MSR
5140	\| VMX_EXIT_CTLS_LOAD_EFER_MSR;
5141	}
5142
5143	/*
5144	* Enable saving of the VMX-preemption timer value on VM-exit.
5145	* For nested-guests, currently not exposed/used.
5146	*/
5147	if ( pVM->hm.s.vmx.fUsePreemptTimer
5148	&& (pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER))
5149	fVal \|= VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER;
5150
5151	/* Don't acknowledge external interrupts on VM-exit. We want to let the host do that. */
5152	Assert(!(fVal & VMX_EXIT_CTLS_ACK_EXT_INT));
5153
5154	/** @todo VMX_EXIT_CTLS_LOAD_PERF_MSR,
5155	* VMX_EXIT_CTLS_SAVE_PAT_MSR,
5156	* VMX_EXIT_CTLS_LOAD_PAT_MSR. */
5157
5158	if ((fVal & fZap) == fVal)
5159	{ /* likely */ }
5160	else
5161	{
5162	Log4Func(("Invalid VM-exit controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%R#X32\n",
5163	pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed0, fVal, fZap));
5164	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_EXIT;
5165	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
5166	}
5167
5168	/* Commit it to the VMCS. */
5169	if (pVmcsInfo->u32ExitCtls != fVal)
5170	{
5171	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT, fVal);
5172	AssertRCReturn(rc, rc);
5173	pVmcsInfo->u32ExitCtls = fVal;
5174	}
5175	}
5176
5177	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
5178	}
5179	return VINF_SUCCESS;
5180	}
5181
5182
5183	/**
5184	* Sets the TPR threshold in the VMCS.
5185	*
5186	* @returns VBox status code.
5187	* @param pVCpu The cross context virtual CPU structure.
5188	* @param pVmcsInfo The VMCS info. object.
5189	* @param u32TprThreshold The TPR threshold (task-priority class only).
5190	*/
5191	DECLINLINE(int) hmR0VmxApicSetTprThreshold(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo, uint32_t u32TprThreshold)
5192	{
5193	Assert(!(u32TprThreshold & ~VMX_TPR_THRESHOLD_MASK)); /* Bits 31:4 MBZ. */
5194	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
5195	RT_NOREF2(pVCpu, pVmcsInfo);
5196	return VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
5197	}
5198
5199
5200	/**
5201	* Exports the guest APIC TPR state into the VMCS.
5202	*
5203	* @returns VBox status code.
5204	* @param pVCpu The cross context virtual CPU structure.
5205	* @param pVmxTransient The VMX-transient structure.
5206	*
5207	* @remarks No-long-jump zone!!!
5208	*/
5209	static int hmR0VmxExportGuestApicTpr(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
5210	{
5211	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_APIC_TPR)
5212	{
5213	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_APIC_TPR);
5214
5215	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5216	if (!pVmxTransient->fIsNestedGuest)
5217	{
5218	if ( PDMHasApic(pVCpu->CTX_SUFF(pVM))
5219	&& APICIsEnabled(pVCpu))
5220	{
5221	/*
5222	* Setup TPR shadowing.
5223	*/
5224	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
5225	{
5226	bool fPendingIntr = false;
5227	uint8_t u8Tpr = 0;
5228	uint8_t u8PendingIntr = 0;
5229	int rc = APICGetTpr(pVCpu, &u8Tpr, &fPendingIntr, &u8PendingIntr);
5230	AssertRCReturn(rc, rc);
5231
5232	/*
5233	* If there are interrupts pending but masked by the TPR, instruct VT-x to
5234	* cause a TPR-below-threshold VM-exit when the guest lowers its TPR below the
5235	* priority of the pending interrupt so we can deliver the interrupt. If there
5236	* are no interrupts pending, set threshold to 0 to not cause any
5237	* TPR-below-threshold VM-exits.
5238	*/
5239	uint32_t u32TprThreshold = 0;
5240	if (fPendingIntr)
5241	{
5242	/* Bits 3:0 of the TPR threshold field correspond to bits 7:4 of the TPR
5243	(which is the Task-Priority Class). */
5244	const uint8_t u8PendingPriority = u8PendingIntr >> 4;
5245	const uint8_t u8TprPriority = u8Tpr >> 4;
5246	if (u8PendingPriority <= u8TprPriority)
5247	u32TprThreshold = u8PendingPriority;
5248	}
5249
5250	rc = hmR0VmxApicSetTprThreshold(pVCpu, pVmcsInfo, u32TprThreshold);
5251	AssertRCReturn(rc, rc);
5252	}
5253	}
5254	}
5255	/* else: the TPR threshold has already been updated while merging the nested-guest VMCS. */
5256	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_APIC_TPR);
5257	}
5258	return VINF_SUCCESS;
5259	}
5260
5261
5262	/**
5263	* Gets the guest interruptibility-state.
5264	*
5265	* @returns Guest's interruptibility-state.
5266	* @param pVCpu The cross context virtual CPU structure.
5267	* @param pVmxTransient The VMX-transient structure.
5268	*
5269	* @remarks No-long-jump zone!!!
5270	*/
5271	static uint32_t hmR0VmxGetGuestIntrState(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
5272	{
5273	/*
5274	* Check if we should inhibit interrupt delivery due to instructions like STI and MOV SS.
5275	*/
5276	uint32_t fIntrState = 0;
5277	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
5278	{
5279	/* If inhibition is active, RIP and RFLAGS should've been imported from the VMCS already. */
5280	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS);
5281
5282	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5283	if (pCtx->rip == EMGetInhibitInterruptsPC(pVCpu))
5284	{
5285	if (pCtx->eflags.Bits.u1IF)
5286	fIntrState = VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
5287	else
5288	fIntrState = VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS;
5289	}
5290	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
5291	{
5292	/*
5293	* We can clear the inhibit force flag as even if we go back to the recompiler
5294	* without executing guest code in VT-x, the flag's condition to be cleared is
5295	* met and thus the cleared state is correct.
5296	*/
5297	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
5298	}
5299	}
5300
5301	/*
5302	* NMIs to the guest are blocked after an NMI is injected until the guest executes an IRET. We only
5303	* bother with virtual-NMI blocking when we have support for virtual NMIs in the CPU, otherwise
5304	* setting this would block host-NMIs and IRET will not clear the blocking.
5305	*
5306	* We always set NMI-exiting so when the host receives an NMI we get a VM-exit.
5307	*
5308	* See Intel spec. 26.6.1 "Interruptibility state". See @bugref{7445}.
5309	*/
5310	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5311	if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
5312	&& CPUMIsGuestNmiBlocking(pVCpu))
5313	fIntrState \|= VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI;
5314
5315	return fIntrState;
5316	}
5317
5318
5319	/**
5320	* Exports the exception intercepts required for guest execution in the VMCS.
5321	*
5322	* @returns VBox status code.
5323	* @param pVCpu The cross context virtual CPU structure.
5324	* @param pVmxTransient The VMX-transient structure.
5325	*
5326	* @remarks No-long-jump zone!!!
5327	*/
5328	static int hmR0VmxExportGuestXcptIntercepts(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
5329	{
5330	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_XCPT_INTERCEPTS)
5331	{
5332	/* When executing a nested-guest, we do not need to trap GIM hypercalls by intercepting #UD. */
5333	if ( !pVmxTransient->fIsNestedGuest
5334	&& pVCpu->hm.s.fGIMTrapXcptUD)
5335	hmR0VmxAddXcptIntercept(pVmxTransient, X86_XCPT_UD);
5336	else
5337	hmR0VmxRemoveXcptIntercept(pVCpu, pVmxTransient, X86_XCPT_UD);
5338
5339	/* Other exception intercepts are handled elsewhere, e.g. while exporting guest CR0. */
5340	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_XCPT_INTERCEPTS);
5341	}
5342	return VINF_SUCCESS;
5343	}
5344
5345
5346	/**
5347	* Exports the guest's RIP into the guest-state area in the VMCS.
5348	*
5349	* @returns VBox status code.
5350	* @param pVCpu The cross context virtual CPU structure.
5351	*
5352	* @remarks No-long-jump zone!!!
5353	*/
5354	static int hmR0VmxExportGuestRip(PVMCPU pVCpu)
5355	{
5356	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RIP)
5357	{
5358	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP);
5359
5360	int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_RIP, pVCpu->cpum.GstCtx.rip);
5361	AssertRCReturn(rc, rc);
5362
5363	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RIP);
5364	Log4Func(("rip=%#RX64\n", pVCpu->cpum.GstCtx.rip));
5365	}
5366	return VINF_SUCCESS;
5367	}
5368
5369
5370	/**
5371	* Exports the guest's RSP into the guest-state area in the VMCS.
5372	*
5373	* @returns VBox status code.
5374	* @param pVCpu The cross context virtual CPU structure.
5375	*
5376	* @remarks No-long-jump zone!!!
5377	*/
5378	static int hmR0VmxExportGuestRsp(PVMCPU pVCpu)
5379	{
5380	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RSP)
5381	{
5382	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RSP);
5383
5384	int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_RSP, pVCpu->cpum.GstCtx.rsp);
5385	AssertRCReturn(rc, rc);
5386
5387	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RSP);
5388	}
5389	return VINF_SUCCESS;
5390	}
5391
5392
5393	/**
5394	* Exports the guest's RFLAGS into the guest-state area in the VMCS.
5395	*
5396	* @returns VBox status code.
5397	* @param pVCpu The cross context virtual CPU structure.
5398	* @param pVmxTransient The VMX-transient structure.
5399	*
5400	* @remarks No-long-jump zone!!!
5401	*/
5402	static int hmR0VmxExportGuestRflags(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
5403	{
5404	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RFLAGS)
5405	{
5406	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
5407
5408	/* Intel spec. 2.3.1 "System Flags and Fields in IA-32e Mode" claims the upper 32-bits of RFLAGS are reserved (MBZ).
5409	Let us assert it as such and use 32-bit VMWRITE. */
5410	Assert(!RT_HI_U32(pVCpu->cpum.GstCtx.rflags.u64));
5411	X86EFLAGS fEFlags = pVCpu->cpum.GstCtx.eflags;
5412	Assert(fEFlags.u32 & X86_EFL_RA1_MASK);
5413	Assert(!(fEFlags.u32 & ~(X86_EFL_1 \| X86_EFL_LIVE_MASK)));
5414
5415	/*
5416	* If we're emulating real-mode using Virtual 8086 mode, save the real-mode eflags so
5417	* we can restore them on VM-exit. Modify the real-mode guest's eflags so that VT-x
5418	* can run the real-mode guest code under Virtual 8086 mode.
5419	*/
5420	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5421	if (pVmcsInfo->RealMode.fRealOnV86Active)
5422	{
5423	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
5424	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
5425	Assert(!pVmxTransient->fIsNestedGuest);
5426	pVmcsInfo->RealMode.Eflags.u32 = fEFlags.u32; /* Save the original eflags of the real-mode guest. */
5427	fEFlags.Bits.u1VM = 1; /* Set the Virtual 8086 mode bit. */
5428	fEFlags.Bits.u2IOPL = 0; /* Change IOPL to 0, otherwise certain instructions won't fault. */
5429	}
5430
5431	int rc = VMXWriteVmcs32(VMX_VMCS_GUEST_RFLAGS, fEFlags.u32);
5432	AssertRCReturn(rc, rc);
5433
5434	/*
5435	* Setup pending debug exceptions if the guest is single-stepping using EFLAGS.TF.
5436	*
5437	* We must avoid setting any automatic debug exceptions delivery when single-stepping
5438	* through the hypervisor debugger using EFLAGS.TF.
5439	*/
5440	if ( !pVmxTransient->fIsNestedGuest
5441	&& !pVCpu->hm.s.fSingleInstruction
5442	&& fEFlags.Bits.u1TF)
5443	{
5444	/** @todo r=ramshankar: Warning!! We ASSUME EFLAGS.TF will not cleared on
5445	* premature trips to ring-3 esp since IEM does not yet handle it. */
5446	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, VMX_VMCS_GUEST_PENDING_DEBUG_XCPT_BS);
5447	AssertRCReturn(rc, rc);
5448	}
5449	/** @todo NSTVMX: Handling copying of VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS from
5450	* nested-guest VMCS. */
5451
5452	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RFLAGS);
5453	Log4Func(("eflags=%#RX32\n", fEFlags.u32));
5454	}
5455	return VINF_SUCCESS;
5456	}
5457
5458
5459	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5460	/**
5461	* Copies the nested-guest VMCS to the shadow VMCS.
5462	*
5463	* @returns VBox status code.
5464	* @param pVCpu The cross context virtual CPU structure.
5465	* @param pVmcsInfo The VMCS info. object.
5466	*
5467	* @remarks No-long-jump zone!!!
5468	*/
5469	static int hmR0VmxCopyNstGstToShadowVmcs(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
5470	{
5471	PVM pVM = pVCpu->CTX_SUFF(pVM);
5472	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5473
5474	/*
5475	* Disable interrupts so we don't get preempted while the shadow VMCS is the
5476	* current VMCS, as we may try saving guest lazy MSRs.
5477	*
5478	* Strictly speaking the lazy MSRs are not in the VMCS, but I'd rather not risk
5479	* calling the import VMCS code which is currently performing the guest MSR reads
5480	* (on 64-bit hosts) and accessing the auto-load/store MSR area on 32-bit hosts
5481	* and the rest of the VMX leave session machinery.
5482	*/
5483	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
5484
5485	int rc = hmR0VmxLoadShadowVmcs(pVmcsInfo);
5486	if (RT_SUCCESS(rc))
5487	{
5488	/*
5489	* Copy all guest read/write VMCS fields.
5490	*
5491	* We don't check for VMWRITE failures here for performance reasons and
5492	* because they are not expected to fail, barring irrecoverable conditions
5493	* like hardware errors.
5494	*/
5495	uint32_t const cShadowVmcsFields = pVM->hm.s.vmx.cShadowVmcsFields;
5496	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
5497	{
5498	uint64_t u64Val;
5499	uint32_t const uVmcsField = pVM->hm.s.vmx.paShadowVmcsFields[i];
5500	IEMReadVmxVmcsField(pVmcsNstGst, uVmcsField, &u64Val);
5501	VMXWriteVmcs64(uVmcsField, u64Val);
5502	}
5503
5504	/*
5505	* If the host CPU supports writing all VMCS fields, copy the guest read-only
5506	* VMCS fields, so the guest can VMREAD them without causing a VM-exit.
5507	*/
5508	if (pVM->hm.s.vmx.Msrs.u64Misc & VMX_MISC_VMWRITE_ALL)
5509	{
5510	uint32_t const cShadowVmcsRoFields = pVM->hm.s.vmx.cShadowVmcsRoFields;
5511	for (uint32_t i = 0; i < cShadowVmcsRoFields; i++)
5512	{
5513	uint64_t u64Val;
5514	uint32_t const uVmcsField = pVM->hm.s.vmx.paShadowVmcsRoFields[i];
5515	IEMReadVmxVmcsField(pVmcsNstGst, uVmcsField, &u64Val);
5516	VMXWriteVmcs64(uVmcsField, u64Val);
5517	}
5518	}
5519
5520	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
5521	rc \|= hmR0VmxLoadVmcs(pVmcsInfo);
5522	}
5523
5524	ASMSetFlags(fEFlags);
5525	return rc;
5526	}
5527
5528
5529	/**
5530	* Copies the shadow VMCS to the nested-guest VMCS.
5531	*
5532	* @returns VBox status code.
5533	* @param pVCpu The cross context virtual CPU structure.
5534	* @param pVmcsInfo The VMCS info. object.
5535	*
5536	* @remarks Called with interrupts disabled.
5537	*/
5538	static int hmR0VmxCopyShadowToNstGstVmcs(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
5539	{
5540	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
5541	PVM pVM = pVCpu->CTX_SUFF(pVM);
5542	PVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5543
5544	int rc = hmR0VmxLoadShadowVmcs(pVmcsInfo);
5545	if (RT_SUCCESS(rc))
5546	{
5547	/*
5548	* Copy guest read/write fields from the shadow VMCS.
5549	* Guest read-only fields cannot be modified, so no need to copy them.
5550	*
5551	* We don't check for VMREAD failures here for performance reasons and
5552	* because they are not expected to fail, barring irrecoverable conditions
5553	* like hardware errors.
5554	*/
5555	uint32_t const cShadowVmcsFields = pVM->hm.s.vmx.cShadowVmcsFields;
5556	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
5557	{
5558	uint64_t u64Val;
5559	uint32_t const uVmcsField = pVM->hm.s.vmx.paShadowVmcsFields[i];
5560	VMXReadVmcs64(uVmcsField, &u64Val);
5561	IEMWriteVmxVmcsField(pVmcsNstGst, uVmcsField, u64Val);
5562	}
5563
5564	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
5565	rc \|= hmR0VmxLoadVmcs(pVmcsInfo);
5566	}
5567	return rc;
5568	}
5569
5570
5571	/**
5572	* Enables VMCS shadowing for the given VMCS info. object.
5573	*
5574	* @param pVCpu The cross context virtual CPU structure.
5575	* @param pVmcsInfo The VMCS info. object.
5576	*
5577	* @remarks No-long-jump zone!!!
5578	*/
5579	static void hmR0VmxEnableVmcsShadowing(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
5580	{
5581	NOREF(pVCpu); /* Used implicitly by VMXWriteVmcs64 on 32-bit hosts. */
5582
5583	uint32_t uProcCtls2 = pVmcsInfo->u32ProcCtls2;
5584	if (!(uProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING))
5585	{
5586	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
5587	uProcCtls2 \|= VMX_PROC_CTLS2_VMCS_SHADOWING;
5588	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, uProcCtls2);
5589	rc \|= VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, pVmcsInfo->HCPhysShadowVmcs);
5590	AssertRC(rc);
5591
5592	pVmcsInfo->u32ProcCtls2 = uProcCtls2;
5593	pVmcsInfo->u64VmcsLinkPtr = pVmcsInfo->HCPhysShadowVmcs;
5594	Log4Func(("Enabled\n"));
5595	}
5596	}
5597
5598
5599	/**
5600	* Disables VMCS shadowing for the given VMCS info. object.
5601	*
5602	* @param pVCpu The cross context virtual CPU structure.
5603	* @param pVmcsInfo The VMCS info. object.
5604	*
5605	* @remarks No-long-jump zone!!!
5606	*/
5607	static void hmR0VmxDisableVmcsShadowing(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
5608	{
5609	NOREF(pVCpu); /* Used implicitly by VMXWriteVmcs64 on 32-bit hosts. */
5610
5611	/*
5612	* We want all VMREAD and VMWRITE instructions to cause VM-exits, so we clear the
5613	* VMCS shadowing control. However, VM-entry requires the shadow VMCS indicator bit
5614	* to match the VMCS shadowing control if the VMCS link pointer is not NIL_RTHCPHYS.
5615	* Hence, we must also reset the VMCS link pointer to ensure VM-entry does not fail.
5616	*
5617	* See Intel spec. 26.2.1.1 "VM-Execution Control Fields".
5618	* See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
5619	*/
5620	uint32_t uProcCtls2 = pVmcsInfo->u32ProcCtls2;
5621	if (uProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
5622	{
5623	uProcCtls2 &= ~VMX_PROC_CTLS2_VMCS_SHADOWING;
5624	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, uProcCtls2);
5625	rc \|= VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
5626	AssertRC(rc);
5627
5628	pVmcsInfo->u32ProcCtls2 = uProcCtls2;
5629	pVmcsInfo->u64VmcsLinkPtr = NIL_RTHCPHYS;
5630	Log4Func(("Disabled\n"));
5631	}
5632	}
5633	#endif
5634
5635
5636	/**
5637	* Exports the guest hardware-virtualization state.
5638	*
5639	* @returns VBox status code.
5640	* @param pVCpu The cross context virtual CPU structure.
5641	* @param pVmxTransient The VMX-transient structure.
5642	*
5643	* @remarks No-long-jump zone!!!
5644	*/
5645	static int hmR0VmxExportGuestHwvirtState(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
5646	{
5647	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_HWVIRT)
5648	{
5649	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5650	/*
5651	* Check if the VMX feature is exposed to the guest and if the host CPU supports
5652	* VMCS shadowing.
5653	*/
5654	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUseVmcsShadowing)
5655	{
5656	/*
5657	* If the guest hypervisor has loaded a current VMCS and is in VMX root mode,
5658	* copy the guest hypervisor's current VMCS into the shadow VMCS and enable
5659	* VMCS shadowing to skip intercepting some or all VMREAD/VMWRITE VM-exits.
5660	*
5661	* We check for VMX root mode here in case the guest executes VMXOFF without
5662	* clearing the current VMCS pointer and our VMXOFF instruction emulation does
5663	* not clear the current VMCS pointer.
5664	*/
5665	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5666	if ( CPUMIsGuestInVmxRootMode(&pVCpu->cpum.GstCtx)
5667	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx)
5668	&& CPUMIsGuestVmxCurrentVmcsValid(pVCpu, &pVCpu->cpum.GstCtx))
5669	{
5670	/* Paranoia. */
5671	Assert(!pVmxTransient->fIsNestedGuest);
5672
5673	/*
5674	* For performance reasons, also check if the guest hypervisor's current VMCS
5675	* was newly loaded or modified before copying it to the shadow VMCS.
5676	*/
5677	if (!pVCpu->hm.s.vmx.fCopiedNstGstToShadowVmcs)
5678	{
5679	int rc = hmR0VmxCopyNstGstToShadowVmcs(pVCpu, pVmcsInfo);
5680	AssertRCReturn(rc, rc);
5681	pVCpu->hm.s.vmx.fCopiedNstGstToShadowVmcs = true;
5682	}
5683	hmR0VmxEnableVmcsShadowing(pVCpu, pVmcsInfo);
5684	}
5685	else
5686	hmR0VmxDisableVmcsShadowing(pVCpu, pVmcsInfo);
5687	}
5688	#else
5689	NOREF(pVmxTransient);
5690	#endif
5691	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_HWVIRT);
5692	}
5693	return VINF_SUCCESS;
5694	}
5695
5696
5697	/**
5698	* Exports the guest CR0 control register into the guest-state area in the VMCS.
5699	*
5700	* The guest FPU state is always pre-loaded hence we don't need to bother about
5701	* sharing FPU related CR0 bits between the guest and host.
5702	*
5703	* @returns VBox status code.
5704	* @param pVCpu The cross context virtual CPU structure.
5705	* @param pVmxTransient The VMX-transient structure.
5706	*
5707	* @remarks No-long-jump zone!!!
5708	*/
5709	static int hmR0VmxExportGuestCR0(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
5710	{
5711	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR0)
5712	{
5713	PVM pVM = pVCpu->CTX_SUFF(pVM);
5714	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5715
5716	/*
5717	* Figure out fixed CR0 bits in VMX operation.
5718	*/
5719	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
5720	uint64_t fSetCr0 = pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1;
5721	uint64_t const fZapCr0 = pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 \| pVM->hm.s.vmx.Msrs.u64Cr0Fixed1;
5722	if (pVM->hm.s.vmx.fUnrestrictedGuest)
5723	fSetCr0 &= ~(uint64_t)(X86_CR0_PE \| X86_CR0_PG);
5724	else
5725	Assert((fSetCr0 & (X86_CR0_PE \| X86_CR0_PG)) == (X86_CR0_PE \| X86_CR0_PG));
5726
5727	if (!pVmxTransient->fIsNestedGuest)
5728	{
5729	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
5730	uint64_t u64GuestCr0 = pVCpu->cpum.GstCtx.cr0;
5731	uint64_t const u64ShadowCr0 = u64GuestCr0;
5732	Assert(!RT_HI_U32(u64GuestCr0));
5733
5734	/*
5735	* Setup VT-x's view of the guest CR0.
5736	*/
5737	uint32_t uProcCtls = pVmcsInfo->u32ProcCtls;
5738	if (pVM->hm.s.fNestedPaging)
5739	{
5740	if (CPUMIsGuestPagingEnabled(pVCpu))
5741	{
5742	/* The guest has paging enabled, let it access CR3 without causing a VM-exit if supported. */
5743	uProcCtls &= ~( VMX_PROC_CTLS_CR3_LOAD_EXIT
5744	\| VMX_PROC_CTLS_CR3_STORE_EXIT);
5745	}
5746	else
5747	{
5748	/* The guest doesn't have paging enabled, make CR3 access cause a VM-exit to update our shadow. */
5749	uProcCtls \|= VMX_PROC_CTLS_CR3_LOAD_EXIT
5750	\| VMX_PROC_CTLS_CR3_STORE_EXIT;
5751	}
5752
5753	/* If we have unrestricted guest execution, we never have to intercept CR3 reads. */
5754	if (pVM->hm.s.vmx.fUnrestrictedGuest)
5755	uProcCtls &= ~VMX_PROC_CTLS_CR3_STORE_EXIT;
5756	}
5757	else
5758	{
5759	/* Guest CPL 0 writes to its read-only pages should cause a #PF VM-exit. */
5760	u64GuestCr0 \|= X86_CR0_WP;
5761	}
5762
5763	/*
5764	* Guest FPU bits.
5765	*
5766	* Since we pre-load the guest FPU always before VM-entry there is no need to track lazy state
5767	* using CR0.TS.
5768	*
5769	* Intel spec. 23.8 "Restrictions on VMX operation" mentions that CR0.NE bit must always be
5770	* set on the first CPUs to support VT-x and no mention of with regards to UX in VM-entry checks.
5771	*/
5772	u64GuestCr0 \|= X86_CR0_NE;
5773
5774	/* If CR0.NE isn't set, we need to intercept #MF exceptions and report them to the guest differently. */
5775	bool const fInterceptMF = !(u64ShadowCr0 & X86_CR0_NE);
5776
5777	/*
5778	* Update exception intercepts.
5779	*/
5780	uint32_t uXcptBitmap = pVmcsInfo->u32XcptBitmap;
5781	if (pVmcsInfo->RealMode.fRealOnV86Active)
5782	{
5783	Assert(PDMVmmDevHeapIsEnabled(pVM));
5784	Assert(pVM->hm.s.vmx.pRealModeTSS);
5785	uXcptBitmap \|= HMVMX_REAL_MODE_XCPT_MASK;
5786	}
5787	else
5788	{
5789	/* For now, cleared here as mode-switches can happen outside HM/VT-x. See @bugref{7626#c11}. */
5790	uXcptBitmap &= ~HMVMX_REAL_MODE_XCPT_MASK;
5791	if (fInterceptMF)
5792	uXcptBitmap \|= RT_BIT(X86_XCPT_MF);
5793	}
5794
5795	/* Additional intercepts for debugging, define these yourself explicitly. */
5796	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
5797	uXcptBitmap \|= 0
5798	\| RT_BIT(X86_XCPT_BP)
5799	\| RT_BIT(X86_XCPT_DE)
5800	\| RT_BIT(X86_XCPT_NM)
5801	\| RT_BIT(X86_XCPT_TS)
5802	\| RT_BIT(X86_XCPT_UD)
5803	\| RT_BIT(X86_XCPT_NP)
5804	\| RT_BIT(X86_XCPT_SS)
5805	\| RT_BIT(X86_XCPT_GP)
5806	\| RT_BIT(X86_XCPT_PF)
5807	\| RT_BIT(X86_XCPT_MF)
5808	;
5809	#elif defined(HMVMX_ALWAYS_TRAP_PF)
5810	uXcptBitmap \|= RT_BIT(X86_XCPT_PF);
5811	#endif
5812	if (pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv)
5813	uXcptBitmap \|= RT_BIT(X86_XCPT_GP);
5814	Assert(pVM->hm.s.fNestedPaging \|\| (uXcptBitmap & RT_BIT(X86_XCPT_PF)));
5815
5816	/* Apply the hardware specified fixed CR0 bits and enable caching. */
5817	u64GuestCr0 \|= fSetCr0;
5818	u64GuestCr0 &= fZapCr0;
5819	u64GuestCr0 &= ~(uint64_t)(X86_CR0_CD \| X86_CR0_NW);
5820
5821	/* Commit the CR0 and related fields to the guest VMCS. */
5822	int rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR0, u64GuestCr0); /** @todo Fix to 64-bit when we drop 32-bit. */
5823	rc \|= VMXWriteVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, u64ShadowCr0);
5824	if (uProcCtls != pVmcsInfo->u32ProcCtls)
5825	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, uProcCtls);
5826	if (uXcptBitmap != pVmcsInfo->u32XcptBitmap)
5827	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
5828	AssertRCReturn(rc, rc);
5829
5830	/* Update our caches. */
5831	pVmcsInfo->u32ProcCtls = uProcCtls;
5832	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
5833
5834	Log4Func(("cr0=%#RX64 shadow=%#RX64 set=%#RX64 zap=%#RX64\n", u64GuestCr0, u64ShadowCr0, fSetCr0, fZapCr0));
5835	}
5836	else
5837	{
5838	/*
5839	* With nested-guests, we may have extended the guest/host mask here since we
5840	* merged in the outer guest's mask. Thus, the merged mask can include more bits
5841	* (to read from the nested-guest CR0 read-shadow) than the guest hypervisor
5842	* originally supplied. We must copy those bits from the nested-guest CR0 into
5843	* the nested-guest CR0 read-shadow.
5844	*/
5845	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
5846	uint64_t u64GuestCr0 = pVCpu->cpum.GstCtx.cr0;
5847	uint64_t const u64ShadowCr0 = CPUMGetGuestVmxMaskedCr0(pVCpu, &pVCpu->cpum.GstCtx, pVmcsInfo->u64Cr0Mask);
5848	Assert(!RT_HI_U32(u64GuestCr0));
5849	Assert(u64GuestCr0 & X86_CR0_NE);
5850
5851	/*
5852	* Apply the hardware specified fixed CR0 bits and enable caching.
5853	* Note! We could be altering our VMX emulation's fixed bits. We thus
5854	* need to re-apply them while importing CR0.
5855	*/
5856	u64GuestCr0 \|= fSetCr0;
5857	u64GuestCr0 &= fZapCr0;
5858	u64GuestCr0 &= ~(uint64_t)(X86_CR0_CD \| X86_CR0_NW);
5859
5860	/* Commit the CR0 and CR0 read-shadow to the nested-guest VMCS. */
5861	/** @todo NSTVMX: Fix to 64-bit when we drop 32-bit. */
5862	int rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR0, u64GuestCr0);
5863	rc \|= VMXWriteVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, u64ShadowCr0);
5864	AssertRCReturn(rc, rc);
5865
5866	Log4Func(("cr0=%#RX64 shadow=%#RX64 set=%#RX64 zap=%#RX64\n", u64GuestCr0, u64ShadowCr0, fSetCr0, fZapCr0));
5867	}
5868
5869	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR0);
5870	}
5871
5872	return VINF_SUCCESS;
5873	}
5874
5875
5876	/**
5877	* Exports the guest control registers (CR3, CR4) into the guest-state area
5878	* in the VMCS.
5879	*
5880	* @returns VBox strict status code.
5881	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
5882	* without unrestricted guest access and the VMMDev is not presently
5883	* mapped (e.g. EFI32).
5884	*
5885	* @param pVCpu The cross context virtual CPU structure.
5886	* @param pVmxTransient The VMX-transient structure.
5887	*
5888	* @remarks No-long-jump zone!!!
5889	*/
5890	static VBOXSTRICTRC hmR0VmxExportGuestCR3AndCR4(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
5891	{
5892	int rc = VINF_SUCCESS;
5893	PVM pVM = pVCpu->CTX_SUFF(pVM);
5894
5895	/*
5896	* Guest CR2.
5897	* It's always loaded in the assembler code. Nothing to do here.
5898	*/
5899
5900	/*
5901	* Guest CR3.
5902	*/
5903	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR3)
5904	{
5905	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR3);
5906
5907	RTGCPHYS GCPhysGuestCr3 = NIL_RTGCPHYS;
5908	if (pVM->hm.s.fNestedPaging)
5909	{
5910	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5911	pVmcsInfo->HCPhysEPTP = PGMGetHyperCR3(pVCpu);
5912
5913	/* Validate. See Intel spec. 28.2.2 "EPT Translation Mechanism" and 24.6.11 "Extended-Page-Table Pointer (EPTP)" */
5914	Assert(pVmcsInfo->HCPhysEPTP != NIL_RTHCPHYS);
5915	Assert(!(pVmcsInfo->HCPhysEPTP & UINT64_C(0xfff0000000000000)));
5916	Assert(!(pVmcsInfo->HCPhysEPTP & 0xfff));
5917
5918	/* VMX_EPT_MEMTYPE_WB support is already checked in hmR0VmxSetupTaggedTlb(). */
5919	pVmcsInfo->HCPhysEPTP \|= VMX_EPT_MEMTYPE_WB
5920	\| (VMX_EPT_PAGE_WALK_LENGTH_DEFAULT << VMX_EPT_PAGE_WALK_LENGTH_SHIFT);
5921
5922	/* Validate. See Intel spec. 26.2.1 "Checks on VMX Controls" */
5923	AssertMsg( ((pVmcsInfo->HCPhysEPTP >> 3) & 0x07) == 3 /* Bits 3:5 (EPT page walk length - 1) must be 3. */
5924	&& ((pVmcsInfo->HCPhysEPTP >> 7) & 0x1f) == 0, /* Bits 7:11 MBZ. */
5925	("EPTP %#RX64\n", pVmcsInfo->HCPhysEPTP));
5926	AssertMsg( !((pVmcsInfo->HCPhysEPTP >> 6) & 0x01) /* Bit 6 (EPT accessed & dirty bit). */
5927	\|\| (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EPT_ACCESS_DIRTY),
5928	("EPTP accessed/dirty bit not supported by CPU but set %#RX64\n", pVmcsInfo->HCPhysEPTP));
5929
5930	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, pVmcsInfo->HCPhysEPTP);
5931	AssertRCReturn(rc, rc);
5932
5933	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5934	if ( pVM->hm.s.vmx.fUnrestrictedGuest
5935	\|\| CPUMIsGuestPagingEnabledEx(pCtx))
5936	{
5937	/* If the guest is in PAE mode, pass the PDPEs to VT-x using the VMCS fields. */
5938	if (CPUMIsGuestInPAEModeEx(pCtx))
5939	{
5940	rc = PGMGstGetPaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
5941	AssertRCReturn(rc, rc);
5942	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, pVCpu->hm.s.aPdpes[0].u);
5943	rc \|= VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, pVCpu->hm.s.aPdpes[1].u);
5944	rc \|= VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, pVCpu->hm.s.aPdpes[2].u);
5945	rc \|= VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, pVCpu->hm.s.aPdpes[3].u);
5946	AssertRCReturn(rc, rc);
5947	}
5948
5949	/*
5950	* The guest's view of its CR3 is unblemished with nested paging when the
5951	* guest is using paging or we have unrestricted guest execution to handle
5952	* the guest when it's not using paging.
5953	*/
5954	GCPhysGuestCr3 = pCtx->cr3;
5955	}
5956	else
5957	{
5958	/*
5959	* The guest is not using paging, but the CPU (VT-x) has to. While the guest
5960	* thinks it accesses physical memory directly, we use our identity-mapped
5961	* page table to map guest-linear to guest-physical addresses. EPT takes care
5962	* of translating it to host-physical addresses.
5963	*/
5964	RTGCPHYS GCPhys;
5965	Assert(pVM->hm.s.vmx.pNonPagingModeEPTPageTable);
5966
5967	/* We obtain it here every time as the guest could have relocated this PCI region. */
5968	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys);
5969	if (RT_SUCCESS(rc))
5970	{ /* likely */ }
5971	else if (rc == VERR_PDM_DEV_HEAP_R3_TO_GCPHYS)
5972	{
5973	Log4Func(("VERR_PDM_DEV_HEAP_R3_TO_GCPHYS -> VINF_EM_RESCHEDULE_REM\n"));
5974	return VINF_EM_RESCHEDULE_REM; /* We cannot execute now, switch to REM/IEM till the guest maps in VMMDev. */
5975	}
5976	else
5977	AssertMsgFailedReturn(("%Rrc\n", rc), rc);
5978
5979	GCPhysGuestCr3 = GCPhys;
5980	}
5981
5982	Log4Func(("u32GuestCr3=%#RGp (GstN)\n", GCPhysGuestCr3));
5983	rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_CR3, GCPhysGuestCr3);
5984	AssertRCReturn(rc, rc);
5985	}
5986	else
5987	{
5988	/* Non-nested paging case, just use the hypervisor's CR3. */
5989	RTHCPHYS const HCPhysGuestCr3 = PGMGetHyperCR3(pVCpu);
5990
5991	Log4Func(("u32GuestCr3=%#RHv (HstN)\n", HCPhysGuestCr3));
5992	rc = VMXWriteVmcsHstN(VMX_VMCS_GUEST_CR3, HCPhysGuestCr3);
5993	AssertRCReturn(rc, rc);
5994	}
5995
5996	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR3);
5997	}
5998
5999	/*
6000	* Guest CR4.
6001	* ASSUMES this is done everytime we get in from ring-3! (XCR0)
6002	*/
6003	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR4)
6004	{
6005	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6006	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6007
6008	/*
6009	* Figure out fixed CR4 bits in VMX operation.
6010	*/
6011	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
6012	uint64_t const fSetCr4 = pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1;
6013	uint64_t const fZapCr4 = pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 \| pVM->hm.s.vmx.Msrs.u64Cr4Fixed1;
6014
6015	/*
6016	* With nested-guests, we may have extended the guest/host mask here (since we
6017	* merged in the outer guest's mask, see hmR0VmxMergeVmcsNested). This means, the
6018	* mask can include more bits (to read from the nested-guest CR4 read-shadow) than
6019	* the guest hypervisor originally supplied. Thus, we should, in essence, copy
6020	* those bits from the nested-guest CR4 into the nested-guest CR4 read-shadow.
6021	*/
6022	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
6023	uint64_t u64GuestCr4 = pCtx->cr4;
6024	uint64_t const u64ShadowCr4 = !pVmxTransient->fIsNestedGuest
6025	? pCtx->cr4
6026	: CPUMGetGuestVmxMaskedCr4(pVCpu, pCtx, pVmcsInfo->u64Cr4Mask);
6027	Assert(!RT_HI_U32(u64GuestCr4));
6028
6029	/*
6030	* Setup VT-x's view of the guest CR4.
6031	*
6032	* If we're emulating real-mode using virtual-8086 mode, we want to redirect software
6033	* interrupts to the 8086 program interrupt handler. Clear the VME bit (the interrupt
6034	* redirection bitmap is already all 0, see hmR3InitFinalizeR0())
6035	*
6036	* See Intel spec. 20.2 "Software Interrupt Handling Methods While in Virtual-8086 Mode".
6037	*/
6038	if (pVmcsInfo->RealMode.fRealOnV86Active)
6039	{
6040	Assert(pVM->hm.s.vmx.pRealModeTSS);
6041	Assert(PDMVmmDevHeapIsEnabled(pVM));
6042	u64GuestCr4 &= ~(uint64_t)X86_CR4_VME;
6043	}
6044
6045	if (pVM->hm.s.fNestedPaging)
6046	{
6047	if ( !CPUMIsGuestPagingEnabledEx(pCtx)
6048	&& !pVM->hm.s.vmx.fUnrestrictedGuest)
6049	{
6050	/* We use 4 MB pages in our identity mapping page table when the guest doesn't have paging. */
6051	u64GuestCr4 \|= X86_CR4_PSE;
6052	/* Our identity mapping is a 32-bit page directory. */
6053	u64GuestCr4 &= ~(uint64_t)X86_CR4_PAE;
6054	}
6055	/* else use guest CR4.*/
6056	}
6057	else
6058	{
6059	Assert(!pVmxTransient->fIsNestedGuest);
6060
6061	/*
6062	* The shadow paging modes and guest paging modes are different, the shadow is in accordance with the host
6063	* paging mode and thus we need to adjust VT-x's view of CR4 depending on our shadow page tables.
6064	*/
6065	switch (pVCpu->hm.s.enmShadowMode)
6066	{
6067	case PGMMODE_REAL: /* Real-mode. */
6068	case PGMMODE_PROTECTED: /* Protected mode without paging. */
6069	case PGMMODE_32_BIT: /* 32-bit paging. */
6070	{
6071	u64GuestCr4 &= ~(uint64_t)X86_CR4_PAE;
6072	break;
6073	}
6074
6075	case PGMMODE_PAE: /* PAE paging. */
6076	case PGMMODE_PAE_NX: /* PAE paging with NX. */
6077	{
6078	u64GuestCr4 \|= X86_CR4_PAE;
6079	break;
6080	}
6081
6082	case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
6083	case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
6084	#ifdef VBOX_ENABLE_64_BITS_GUESTS
6085	break;
6086	#endif
6087	default:
6088	AssertFailed();
6089	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
6090	}
6091	}
6092
6093	/*
6094	* Apply the hardware specified fixed CR4 bits (mainly CR4.VMXE).
6095	* Note! For nested-guests, we could be altering our VMX emulation's
6096	* fixed bits. We thus need to re-apply them while importing CR4.
6097	*/
6098	u64GuestCr4 \|= fSetCr4;
6099	u64GuestCr4 &= fZapCr4;
6100
6101	/* Commit the CR4 and CR4 read-shadow to the guest VMCS. */
6102	/** @todo Fix to 64-bit when we drop 32-bit. */
6103	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR4, u64GuestCr4);
6104	rc \|= VMXWriteVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, u64ShadowCr4);
6105	AssertRCReturn(rc, rc);
6106
6107	/* Whether to save/load/restore XCR0 during world switch depends on CR4.OSXSAVE and host+guest XCR0. */
6108	pVCpu->hm.s.fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
6109
6110	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR4);
6111
6112	Log4Func(("cr4=%#RX64 shadow=%#RX64 set=%#RX64 zap=%#RX64)\n", u64GuestCr4, u64ShadowCr4, fSetCr4, fZapCr4));
6113	}
6114	return rc;
6115	}
6116
6117
6118	/**
6119	* Exports the guest debug registers into the guest-state area in the VMCS.
6120	* The guest debug bits are partially shared with the host (e.g. DR6, DR0-3).
6121	*
6122	* This also sets up whether \#DB and MOV DRx accesses cause VM-exits.
6123	*
6124	* @returns VBox status code.
6125	* @param pVCpu The cross context virtual CPU structure.
6126	* @param pVmxTransient The VMX-transient structure.
6127	*
6128	* @remarks No-long-jump zone!!!
6129	*/
6130	static int hmR0VmxExportSharedDebugState(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
6131	{
6132	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
6133
6134	/** @todo NSTVMX: Figure out what we want to do with nested-guest instruction
6135	* stepping. */
6136	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6137	if (pVmxTransient->fIsNestedGuest)
6138	{
6139	int rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, CPUMGetGuestDR7(pVCpu));
6140	AssertRCReturn(rc, rc);
6141
6142	/* Always intercept Mov DRx accesses for the nested-guest for now. */
6143	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_MOV_DR_EXIT;
6144	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
6145	AssertRCReturn(rc, rc);
6146	return VINF_SUCCESS;
6147	}
6148
6149	#ifdef VBOX_STRICT
6150	/* Validate. Intel spec. 26.3.1.1 "Checks on Guest Controls Registers, Debug Registers, MSRs" */
6151	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
6152	{
6153	/* Validate. Intel spec. 17.2 "Debug Registers", recompiler paranoia checks. */
6154	Assert((pVCpu->cpum.GstCtx.dr[7] & (X86_DR7_MBZ_MASK \| X86_DR7_RAZ_MASK)) == 0);
6155	Assert((pVCpu->cpum.GstCtx.dr[7] & X86_DR7_RA1_MASK) == X86_DR7_RA1_MASK);
6156	}
6157	#endif
6158
6159	bool fSteppingDB = false;
6160	bool fInterceptMovDRx = false;
6161	uint32_t uProcCtls = pVmcsInfo->u32ProcCtls;
6162	if (pVCpu->hm.s.fSingleInstruction)
6163	{
6164	/* If the CPU supports the monitor trap flag, use it for single stepping in DBGF and avoid intercepting #DB. */
6165	PVM pVM = pVCpu->CTX_SUFF(pVM);
6166	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_MONITOR_TRAP_FLAG)
6167	{
6168	uProcCtls \|= VMX_PROC_CTLS_MONITOR_TRAP_FLAG;
6169	Assert(fSteppingDB == false);
6170	}
6171	else
6172	{
6173	pVCpu->cpum.GstCtx.eflags.u32 \|= X86_EFL_TF;
6174	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_GUEST_RFLAGS;
6175	pVCpu->hm.s.fClearTrapFlag = true;
6176	fSteppingDB = true;
6177	}
6178	}
6179
6180	uint32_t u32GuestDr7;
6181	if ( fSteppingDB
6182	\|\| (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK))
6183	{
6184	/*
6185	* Use the combined guest and host DRx values found in the hypervisor register set
6186	* because the hypervisor debugger has breakpoints active or someone is single stepping
6187	* on the host side without a monitor trap flag.
6188	*
6189	* Note! DBGF expects a clean DR6 state before executing guest code.
6190	*/
6191	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
6192	if ( CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx)
6193	&& !CPUMIsHyperDebugStateActivePending(pVCpu))
6194	{
6195	CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
6196	Assert(CPUMIsHyperDebugStateActivePending(pVCpu));
6197	Assert(!CPUMIsGuestDebugStateActivePending(pVCpu));
6198	}
6199	else
6200	#endif
6201	if (!CPUMIsHyperDebugStateActive(pVCpu))
6202	{
6203	CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
6204	Assert(CPUMIsHyperDebugStateActive(pVCpu));
6205	Assert(!CPUMIsGuestDebugStateActive(pVCpu));
6206	}
6207
6208	/* Update DR7 with the hypervisor value (other DRx registers are handled by CPUM one way or another). */
6209	u32GuestDr7 = (uint32_t)CPUMGetHyperDR7(pVCpu);
6210	pVCpu->hm.s.fUsingHyperDR7 = true;
6211	fInterceptMovDRx = true;
6212	}
6213	else
6214	{
6215	/*
6216	* If the guest has enabled debug registers, we need to load them prior to
6217	* executing guest code so they'll trigger at the right time.
6218	*/
6219	if (pVCpu->cpum.GstCtx.dr[7] & (X86_DR7_ENABLED_MASK \| X86_DR7_GD))
6220	{
6221	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
6222	if ( CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx)
6223	&& !CPUMIsGuestDebugStateActivePending(pVCpu))
6224	{
6225	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
6226	Assert(CPUMIsGuestDebugStateActivePending(pVCpu));
6227	Assert(!CPUMIsHyperDebugStateActivePending(pVCpu));
6228	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
6229	}
6230	else
6231	#endif
6232	if (!CPUMIsGuestDebugStateActive(pVCpu))
6233	{
6234	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
6235	Assert(CPUMIsGuestDebugStateActive(pVCpu));
6236	Assert(!CPUMIsHyperDebugStateActive(pVCpu));
6237	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
6238	}
6239	Assert(!fInterceptMovDRx);
6240	}
6241	/*
6242	* If no debugging enabled, we'll lazy load DR0-3. Unlike on AMD-V, we
6243	* must intercept #DB in order to maintain a correct DR6 guest value, and
6244	* because we need to intercept it to prevent nested #DBs from hanging the
6245	* CPU, we end up always having to intercept it. See hmR0VmxSetupVmcsXcptBitmap().
6246	*/
6247	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
6248	else if ( !CPUMIsGuestDebugStateActivePending(pVCpu)
6249	&& !CPUMIsGuestDebugStateActive(pVCpu))
6250	#else
6251	else if (!CPUMIsGuestDebugStateActive(pVCpu))
6252	#endif
6253	{
6254	fInterceptMovDRx = true;
6255	}
6256
6257	/* Update DR7 with the actual guest value. */
6258	u32GuestDr7 = pVCpu->cpum.GstCtx.dr[7];
6259	pVCpu->hm.s.fUsingHyperDR7 = false;
6260	}
6261
6262	if (fInterceptMovDRx)
6263	uProcCtls \|= VMX_PROC_CTLS_MOV_DR_EXIT;
6264	else
6265	uProcCtls &= ~VMX_PROC_CTLS_MOV_DR_EXIT;
6266
6267	/*
6268	* Update the processor-based VM-execution controls with the MOV-DRx intercepts and the
6269	* monitor-trap flag and update our cache.
6270	*/
6271	if (uProcCtls != pVmcsInfo->u32ProcCtls)
6272	{
6273	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, uProcCtls);
6274	AssertRCReturn(rc2, rc2);
6275	pVmcsInfo->u32ProcCtls = uProcCtls;
6276	}
6277
6278	/*
6279	* Update guest DR7.
6280	*/
6281	int rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, u32GuestDr7);
6282	AssertRCReturn(rc, rc);
6283
6284	/*
6285	* If we have forced EFLAGS.TF to be set because we're single-stepping in the hypervisor debugger,
6286	* we need to clear interrupt inhibition if any as otherwise it causes a VM-entry failure.
6287	*
6288	* See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
6289	*/
6290	if (fSteppingDB)
6291	{
6292	Assert(pVCpu->hm.s.fSingleInstruction);
6293	Assert(pVCpu->cpum.GstCtx.eflags.Bits.u1TF);
6294
6295	uint32_t fIntrState = 0;
6296	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
6297	AssertRCReturn(rc, rc);
6298
6299	if (fIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
6300	{
6301	fIntrState &= ~(VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
6302	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
6303	AssertRCReturn(rc, rc);
6304	}
6305	}
6306
6307	return VINF_SUCCESS;
6308	}
6309
6310
6311	#ifdef VBOX_STRICT
6312	/**
6313	* Strict function to validate segment registers.
6314	*
6315	* @param pVCpu The cross context virtual CPU structure.
6316	* @param pVmcsInfo The VMCS info. object.
6317	*
6318	* @remarks Will import guest CR0 on strict builds during validation of
6319	* segments.
6320	*/
6321	static void hmR0VmxValidateSegmentRegs(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
6322	{
6323	/*
6324	* Validate segment registers. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
6325	*
6326	* The reason we check for attribute value 0 in this function and not just the unusable bit is
6327	* because hmR0VmxExportGuestSegReg() only updates the VMCS' copy of the value with the
6328	* unusable bit and doesn't change the guest-context value.
6329	*/
6330	PVM pVM = pVCpu->CTX_SUFF(pVM);
6331	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6332	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR0);
6333	if ( !pVM->hm.s.vmx.fUnrestrictedGuest
6334	&& ( !CPUMIsGuestInRealModeEx(pCtx)
6335	&& !CPUMIsGuestInV86ModeEx(pCtx)))
6336	{
6337	/* Protected mode checks */
6338	/* CS */
6339	Assert(pCtx->cs.Attr.n.u1Present);
6340	Assert(!(pCtx->cs.Attr.u & 0xf00));
6341	Assert(!(pCtx->cs.Attr.u & 0xfffe0000));
6342	Assert( (pCtx->cs.u32Limit & 0xfff) == 0xfff
6343	\|\| !(pCtx->cs.Attr.n.u1Granularity));
6344	Assert( !(pCtx->cs.u32Limit & 0xfff00000)
6345	\|\| (pCtx->cs.Attr.n.u1Granularity));
6346	/* CS cannot be loaded with NULL in protected mode. */
6347	Assert(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE)); /** @todo is this really true even for 64-bit CS? */
6348	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
6349	Assert(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl);
6350	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
6351	Assert(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl);
6352	else
6353	AssertMsgFailed(("Invalid CS Type %#x\n", pCtx->cs.Attr.n.u2Dpl));
6354	/* SS */
6355	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
6356	Assert(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL));
6357	if ( !(pCtx->cr0 & X86_CR0_PE)
6358	\|\| pCtx->cs.Attr.n.u4Type == 3)
6359	{
6360	Assert(!pCtx->ss.Attr.n.u2Dpl);
6361	}
6362	if (pCtx->ss.Attr.u && !(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
6363	{
6364	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
6365	Assert(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7);
6366	Assert(pCtx->ss.Attr.n.u1Present);
6367	Assert(!(pCtx->ss.Attr.u & 0xf00));
6368	Assert(!(pCtx->ss.Attr.u & 0xfffe0000));
6369	Assert( (pCtx->ss.u32Limit & 0xfff) == 0xfff
6370	\|\| !(pCtx->ss.Attr.n.u1Granularity));
6371	Assert( !(pCtx->ss.u32Limit & 0xfff00000)
6372	\|\| (pCtx->ss.Attr.n.u1Granularity));
6373	}
6374	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxExportGuestSegReg(). */
6375	if (pCtx->ds.Attr.u && !(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
6376	{
6377	Assert(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6378	Assert(pCtx->ds.Attr.n.u1Present);
6379	Assert(pCtx->ds.Attr.n.u4Type > 11 \|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL));
6380	Assert(!(pCtx->ds.Attr.u & 0xf00));
6381	Assert(!(pCtx->ds.Attr.u & 0xfffe0000));
6382	Assert( (pCtx->ds.u32Limit & 0xfff) == 0xfff
6383	\|\| !(pCtx->ds.Attr.n.u1Granularity));
6384	Assert( !(pCtx->ds.u32Limit & 0xfff00000)
6385	\|\| (pCtx->ds.Attr.n.u1Granularity));
6386	Assert( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6387	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ));
6388	}
6389	if (pCtx->es.Attr.u && !(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
6390	{
6391	Assert(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6392	Assert(pCtx->es.Attr.n.u1Present);
6393	Assert(pCtx->es.Attr.n.u4Type > 11 \|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL));
6394	Assert(!(pCtx->es.Attr.u & 0xf00));
6395	Assert(!(pCtx->es.Attr.u & 0xfffe0000));
6396	Assert( (pCtx->es.u32Limit & 0xfff) == 0xfff
6397	\|\| !(pCtx->es.Attr.n.u1Granularity));
6398	Assert( !(pCtx->es.u32Limit & 0xfff00000)
6399	\|\| (pCtx->es.Attr.n.u1Granularity));
6400	Assert( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6401	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ));
6402	}
6403	if (pCtx->fs.Attr.u && !(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
6404	{
6405	Assert(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6406	Assert(pCtx->fs.Attr.n.u1Present);
6407	Assert(pCtx->fs.Attr.n.u4Type > 11 \|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL));
6408	Assert(!(pCtx->fs.Attr.u & 0xf00));
6409	Assert(!(pCtx->fs.Attr.u & 0xfffe0000));
6410	Assert( (pCtx->fs.u32Limit & 0xfff) == 0xfff
6411	\|\| !(pCtx->fs.Attr.n.u1Granularity));
6412	Assert( !(pCtx->fs.u32Limit & 0xfff00000)
6413	\|\| (pCtx->fs.Attr.n.u1Granularity));
6414	Assert( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6415	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ));
6416	}
6417	if (pCtx->gs.Attr.u && !(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
6418	{
6419	Assert(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6420	Assert(pCtx->gs.Attr.n.u1Present);
6421	Assert(pCtx->gs.Attr.n.u4Type > 11 \|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL));
6422	Assert(!(pCtx->gs.Attr.u & 0xf00));
6423	Assert(!(pCtx->gs.Attr.u & 0xfffe0000));
6424	Assert( (pCtx->gs.u32Limit & 0xfff) == 0xfff
6425	\|\| !(pCtx->gs.Attr.n.u1Granularity));
6426	Assert( !(pCtx->gs.u32Limit & 0xfff00000)
6427	\|\| (pCtx->gs.Attr.n.u1Granularity));
6428	Assert( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6429	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ));
6430	}
6431	/* 64-bit capable CPUs. */
6432	# if HC_ARCH_BITS == 64
6433	Assert(!RT_HI_U32(pCtx->cs.u64Base));
6434	Assert(!pCtx->ss.Attr.u \|\| !RT_HI_U32(pCtx->ss.u64Base));
6435	Assert(!pCtx->ds.Attr.u \|\| !RT_HI_U32(pCtx->ds.u64Base));
6436	Assert(!pCtx->es.Attr.u \|\| !RT_HI_U32(pCtx->es.u64Base));
6437	# endif
6438	}
6439	else if ( CPUMIsGuestInV86ModeEx(pCtx)
6440	\|\| ( CPUMIsGuestInRealModeEx(pCtx)
6441	&& !pVM->hm.s.vmx.fUnrestrictedGuest))
6442	{
6443	/* Real and v86 mode checks. */
6444	/* hmR0VmxExportGuestSegReg() writes the modified in VMCS. We want what we're feeding to VT-x. */
6445	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
6446	if (pVmcsInfo->RealMode.fRealOnV86Active)
6447	{
6448	u32CSAttr = 0xf3; u32SSAttr = 0xf3; u32DSAttr = 0xf3;
6449	u32ESAttr = 0xf3; u32FSAttr = 0xf3; u32GSAttr = 0xf3;
6450	}
6451	else
6452	{
6453	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u; u32DSAttr = pCtx->ds.Attr.u;
6454	u32ESAttr = pCtx->es.Attr.u; u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
6455	}
6456
6457	/* CS */
6458	AssertMsg((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), ("CS base %#x %#x\n", pCtx->cs.u64Base, pCtx->cs.Sel));
6459	Assert(pCtx->cs.u32Limit == 0xffff);
6460	Assert(u32CSAttr == 0xf3);
6461	/* SS */
6462	Assert(pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4);
6463	Assert(pCtx->ss.u32Limit == 0xffff);
6464	Assert(u32SSAttr == 0xf3);
6465	/* DS */
6466	Assert(pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4);
6467	Assert(pCtx->ds.u32Limit == 0xffff);
6468	Assert(u32DSAttr == 0xf3);
6469	/* ES */
6470	Assert(pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4);
6471	Assert(pCtx->es.u32Limit == 0xffff);
6472	Assert(u32ESAttr == 0xf3);
6473	/* FS */
6474	Assert(pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4);
6475	Assert(pCtx->fs.u32Limit == 0xffff);
6476	Assert(u32FSAttr == 0xf3);
6477	/* GS */
6478	Assert(pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4);
6479	Assert(pCtx->gs.u32Limit == 0xffff);
6480	Assert(u32GSAttr == 0xf3);
6481	/* 64-bit capable CPUs. */
6482	# if HC_ARCH_BITS == 64
6483	Assert(!RT_HI_U32(pCtx->cs.u64Base));
6484	Assert(!u32SSAttr \|\| !RT_HI_U32(pCtx->ss.u64Base));
6485	Assert(!u32DSAttr \|\| !RT_HI_U32(pCtx->ds.u64Base));
6486	Assert(!u32ESAttr \|\| !RT_HI_U32(pCtx->es.u64Base));
6487	# endif
6488	}
6489	}
6490	#endif /* VBOX_STRICT */
6491
6492
6493	/**
6494	* Exports a guest segment register into the guest-state area in the VMCS.
6495	*
6496	* @returns VBox status code.
6497	* @param pVCpu The cross context virtual CPU structure.
6498	* @param pVmcsInfo The VMCS info. object.
6499	* @param iSegReg The segment register number (X86_SREG_XXX).
6500	* @param pSelReg Pointer to the segment selector.
6501	*
6502	* @remarks No-long-jump zone!!!
6503	*/
6504	static int hmR0VmxExportGuestSegReg(PVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo, uint8_t iSegReg, PCCPUMSELREG pSelReg)
6505	{
6506	Assert(iSegReg < X86_SREG_COUNT);
6507	uint32_t const idxSel = g_aVmcsSegSel[iSegReg];
6508	uint32_t const idxLimit = g_aVmcsSegLimit[iSegReg];
6509	uint32_t const idxBase = g_aVmcsSegBase[iSegReg];
6510	uint32_t const idxAttr = g_aVmcsSegAttr[iSegReg];
6511
6512	uint32_t u32Access = pSelReg->Attr.u;
6513	if (pVmcsInfo->RealMode.fRealOnV86Active)
6514	{
6515	/* VT-x requires our real-using-v86 mode hack to override the segment access-right bits. */
6516	u32Access = 0xf3;
6517	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
6518	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
6519	RT_NOREF_PV(pVCpu);
6520	}
6521	else
6522	{
6523	/*
6524	* The way to differentiate between whether this is really a null selector or was just
6525	* a selector loaded with 0 in real-mode is using the segment attributes. A selector
6526	* loaded in real-mode with the value 0 is valid and usable in protected-mode and we
6527	* should -not- mark it as an unusable segment. Both the recompiler & VT-x ensures
6528	* NULL selectors loaded in protected-mode have their attribute as 0.
6529	*/
6530	if (!u32Access)
6531	u32Access = X86DESCATTR_UNUSABLE;
6532	}
6533
6534	/* Validate segment access rights. Refer to Intel spec. "26.3.1.2 Checks on Guest Segment Registers". */
6535	AssertMsg((u32Access & X86DESCATTR_UNUSABLE) \|\| (u32Access & X86_SEL_TYPE_ACCESSED),
6536	("Access bit not set for usable segment. idx=%#x sel=%#x attr %#x\n", idxBase, pSelReg, pSelReg->Attr.u));
6537
6538	/*
6539	* Commit it to the VMCS.
6540	*/
6541	int rc = VMXWriteVmcs32(idxSel, pSelReg->Sel);
6542	rc \|= VMXWriteVmcs32(idxLimit, pSelReg->u32Limit);
6543	rc \|= VMXWriteVmcsGstN(idxBase, pSelReg->u64Base);
6544	rc \|= VMXWriteVmcs32(idxAttr, u32Access);
6545	AssertRCReturn(rc, rc);
6546	return rc;
6547	}
6548
6549
6550	/**
6551	* Exports the guest segment registers, GDTR, IDTR, LDTR, TR into the guest-state
6552	* area in the VMCS.
6553	*
6554	* @returns VBox status code.
6555	* @param pVCpu The cross context virtual CPU structure.
6556	* @param pVmxTransient The VMX-transient structure.
6557	*
6558	* @remarks Will import guest CR0 on strict builds during validation of
6559	* segments.
6560	* @remarks No-long-jump zone!!!
6561	*/
6562	static int hmR0VmxExportGuestSegRegsXdtr(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
6563	{
6564	int rc = VERR_INTERNAL_ERROR_5;
6565	PVM pVM = pVCpu->CTX_SUFF(pVM);
6566	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6567	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6568
6569	/*
6570	* Guest Segment registers: CS, SS, DS, ES, FS, GS.
6571	*/
6572	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SREG_MASK)
6573	{
6574	#ifdef VBOX_WITH_REM
6575	if (!pVM->hm.s.vmx.fUnrestrictedGuest)
6576	{
6577	Assert(!pVmxTransient->fIsNestedGuest);
6578	Assert(pVM->hm.s.vmx.pRealModeTSS);
6579	AssertCompile(PGMMODE_REAL < PGMMODE_PROTECTED);
6580	if ( pVmcsInfo->fWasInRealMode
6581	&& PGMGetGuestMode(pVCpu) >= PGMMODE_PROTECTED)
6582	{
6583	/*
6584	* Notify the recompiler must flush its code-cache as the guest -may-
6585	* rewrite code it in real-mode (e.g. OpenBSD 4.0).
6586	*/
6587	REMFlushTBs(pVM);
6588	Log4Func(("Switch to protected mode detected!\n"));
6589	pVmcsInfo->fWasInRealMode = false;
6590	}
6591	}
6592	#endif
6593	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CS)
6594	{
6595	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CS);
6596	if (pVmcsInfo->RealMode.fRealOnV86Active)
6597	pVmcsInfo->RealMode.AttrCS.u = pCtx->cs.Attr.u;
6598	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_CS, &pCtx->cs);
6599	AssertRCReturn(rc, rc);
6600	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CS);
6601	}
6602
6603	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SS)
6604	{
6605	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SS);
6606	if (pVmcsInfo->RealMode.fRealOnV86Active)
6607	pVmcsInfo->RealMode.AttrSS.u = pCtx->ss.Attr.u;
6608	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_SS, &pCtx->ss);
6609	AssertRCReturn(rc, rc);
6610	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SS);
6611	}
6612
6613	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_DS)
6614	{
6615	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_DS);
6616	if (pVmcsInfo->RealMode.fRealOnV86Active)
6617	pVmcsInfo->RealMode.AttrDS.u = pCtx->ds.Attr.u;
6618	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_DS, &pCtx->ds);
6619	AssertRCReturn(rc, rc);
6620	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_DS);
6621	}
6622
6623	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_ES)
6624	{
6625	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_ES);
6626	if (pVmcsInfo->RealMode.fRealOnV86Active)
6627	pVmcsInfo->RealMode.AttrES.u = pCtx->es.Attr.u;
6628	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_ES, &pCtx->es);
6629	AssertRCReturn(rc, rc);
6630	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_ES);
6631	}
6632
6633	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_FS)
6634	{
6635	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_FS);
6636	if (pVmcsInfo->RealMode.fRealOnV86Active)
6637	pVmcsInfo->RealMode.AttrFS.u = pCtx->fs.Attr.u;
6638	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_FS, &pCtx->fs);
6639	AssertRCReturn(rc, rc);
6640	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_FS);
6641	}
6642
6643	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_GS)
6644	{
6645	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GS);
6646	if (pVmcsInfo->RealMode.fRealOnV86Active)
6647	pVmcsInfo->RealMode.AttrGS.u = pCtx->gs.Attr.u;
6648	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_GS, &pCtx->gs);
6649	AssertRCReturn(rc, rc);
6650	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_GS);
6651	}
6652
6653	#ifdef VBOX_STRICT
6654	hmR0VmxValidateSegmentRegs(pVCpu, pVmcsInfo);
6655	#endif
6656	Log4Func(("cs={%#04x base=%#RX64 limit=%#RX32 attr=%#RX32}\n", pCtx->cs.Sel, pCtx->cs.u64Base, pCtx->cs.u32Limit,
6657	pCtx->cs.Attr.u));
6658	}
6659
6660	/*
6661	* Guest TR.
6662	*/
6663	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_TR)
6664	{
6665	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_TR);
6666
6667	/*
6668	* Real-mode emulation using virtual-8086 mode with CR4.VME. Interrupt redirection is
6669	* achieved using the interrupt redirection bitmap (all bits cleared to let the guest
6670	* handle INT-n's) in the TSS. See hmR3InitFinalizeR0() to see how pRealModeTSS is setup.
6671	*/
6672	uint16_t u16Sel;
6673	uint32_t u32Limit;
6674	uint64_t u64Base;
6675	uint32_t u32AccessRights;
6676	if (!pVmcsInfo->RealMode.fRealOnV86Active)
6677	{
6678	u16Sel = pCtx->tr.Sel;
6679	u32Limit = pCtx->tr.u32Limit;
6680	u64Base = pCtx->tr.u64Base;
6681	u32AccessRights = pCtx->tr.Attr.u;
6682	}
6683	else
6684	{
6685	Assert(!pVmxTransient->fIsNestedGuest);
6686	Assert(pVM->hm.s.vmx.pRealModeTSS);
6687	Assert(PDMVmmDevHeapIsEnabled(pVM)); /* Guaranteed by HMCanExecuteGuest() -XXX- what about inner loop changes? */
6688
6689	/* We obtain it here every time as PCI regions could be reconfigured in the guest, changing the VMMDev base. */
6690	RTGCPHYS GCPhys;
6691	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys);
6692	AssertRCReturn(rc, rc);
6693
6694	X86DESCATTR DescAttr;
6695	DescAttr.u = 0;
6696	DescAttr.n.u1Present = 1;
6697	DescAttr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
6698
6699	u16Sel = 0;
6700	u32Limit = HM_VTX_TSS_SIZE;
6701	u64Base = GCPhys;
6702	u32AccessRights = DescAttr.u;
6703	}
6704
6705	/* Validate. */
6706	Assert(!(u16Sel & RT_BIT(2)));
6707	AssertMsg( (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_386_TSS_BUSY
6708	\|\| (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_286_TSS_BUSY, ("TSS is not busy!? %#x\n", u32AccessRights));
6709	AssertMsg(!(u32AccessRights & X86DESCATTR_UNUSABLE), ("TR unusable bit is not clear!? %#x\n", u32AccessRights));
6710	Assert(!(u32AccessRights & RT_BIT(4))); /* System MBZ.*/
6711	Assert(u32AccessRights & RT_BIT(7)); /* Present MB1.*/
6712	Assert(!(u32AccessRights & 0xf00)); /* 11:8 MBZ. */
6713	Assert(!(u32AccessRights & 0xfffe0000)); /* 31:17 MBZ. */
6714	Assert( (u32Limit & 0xfff) == 0xfff
6715	\|\| !(u32AccessRights & RT_BIT(15))); /* Granularity MBZ. */
6716	Assert( !(pCtx->tr.u32Limit & 0xfff00000)
6717	\|\| (u32AccessRights & RT_BIT(15))); /* Granularity MB1. */
6718
6719	rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_TR_SEL, u16Sel);
6720	rc \|= VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, u32Limit);
6721	rc \|= VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, u32AccessRights);
6722	rc \|= VMXWriteVmcsGstN(VMX_VMCS_GUEST_TR_BASE, u64Base);
6723	AssertRCReturn(rc, rc);
6724
6725	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_TR);
6726	Log4Func(("tr base=%#RX64 limit=%#RX32\n", pCtx->tr.u64Base, pCtx->tr.u32Limit));
6727	}
6728
6729	/*
6730	* Guest GDTR.
6731	*/
6732	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_GDTR)
6733	{
6734	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GDTR);
6735
6736	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, pCtx->gdtr.cbGdt);
6737	rc \|= VMXWriteVmcsGstN(VMX_VMCS_GUEST_GDTR_BASE, pCtx->gdtr.pGdt);
6738	AssertRCReturn(rc, rc);
6739
6740	/* Validate. */
6741	Assert(!(pCtx->gdtr.cbGdt & 0xffff0000)); /* Bits 31:16 MBZ. */
6742
6743	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_GDTR);
6744	Log4Func(("gdtr base=%#RX64 limit=%#RX32\n", pCtx->gdtr.pGdt, pCtx->gdtr.cbGdt));
6745	}
6746
6747	/*
6748	* Guest LDTR.
6749	*/
6750	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_LDTR)
6751	{
6752	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_LDTR);
6753
6754	/* The unusable bit is specific to VT-x, if it's a null selector mark it as an unusable segment. */
6755	uint32_t u32Access;
6756	if ( !pVmxTransient->fIsNestedGuest
6757	&& !pCtx->ldtr.Attr.u)
6758	u32Access = X86DESCATTR_UNUSABLE;
6759	else
6760	u32Access = pCtx->ldtr.Attr.u;
6761
6762	rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_LDTR_SEL, pCtx->ldtr.Sel);
6763	rc \|= VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, pCtx->ldtr.u32Limit);
6764	rc \|= VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, u32Access);
6765	rc \|= VMXWriteVmcsGstN(VMX_VMCS_GUEST_LDTR_BASE, pCtx->ldtr.u64Base);
6766	AssertRCReturn(rc, rc);
6767
6768	/* Validate. */
6769	if (!(u32Access & X86DESCATTR_UNUSABLE))
6770	{
6771	Assert(!(pCtx->ldtr.Sel & RT_BIT(2))); /* TI MBZ. */
6772	Assert(pCtx->ldtr.Attr.n.u4Type == 2); /* Type MB2 (LDT). */
6773	Assert(!pCtx->ldtr.Attr.n.u1DescType); /* System MBZ. */
6774	Assert(pCtx->ldtr.Attr.n.u1Present == 1); /* Present MB1. */
6775	Assert(!pCtx->ldtr.Attr.n.u4LimitHigh); /* 11:8 MBZ. */
6776	Assert(!(pCtx->ldtr.Attr.u & 0xfffe0000)); /* 31:17 MBZ. */
6777	Assert( (pCtx->ldtr.u32Limit & 0xfff) == 0xfff
6778	\|\| !pCtx->ldtr.Attr.n.u1Granularity); /* Granularity MBZ. */
6779	Assert( !(pCtx->ldtr.u32Limit & 0xfff00000)
6780	\|\| pCtx->ldtr.Attr.n.u1Granularity); /* Granularity MB1. */
6781	}
6782
6783	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_LDTR);
6784	Log4Func(("ldtr base=%#RX64 limit=%#RX32\n", pCtx->ldtr.u64Base, pCtx->ldtr.u32Limit));
6785	}
6786
6787	/*
6788	* Guest IDTR.
6789	*/
6790	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_IDTR)
6791	{
6792	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_IDTR);
6793
6794	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, pCtx->idtr.cbIdt);
6795	rc \|= VMXWriteVmcsGstN(VMX_VMCS_GUEST_IDTR_BASE, pCtx->idtr.pIdt);
6796	AssertRCReturn(rc, rc);
6797
6798	/* Validate. */
6799	Assert(!(pCtx->idtr.cbIdt & 0xffff0000)); /* Bits 31:16 MBZ. */
6800
6801	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_IDTR);
6802	Log4Func(("idtr base=%#RX64 limit=%#RX32\n", pCtx->idtr.pIdt, pCtx->idtr.cbIdt));
6803	}
6804
6805	return VINF_SUCCESS;
6806	}
6807
6808
6809	/**
6810	* Exports certain guest MSRs into the VM-entry MSR-load and VM-exit MSR-store
6811	* areas.
6812	*
6813	* These MSRs will automatically be loaded to the host CPU on every successful
6814	* VM-entry and stored from the host CPU on every successful VM-exit.
6815	*
6816	* We creates/updates MSR slots for the host MSRs in the VM-exit MSR-load area. The
6817	* actual host MSR values are not- updated here for performance reasons. See
6818	* hmR0VmxExportHostMsrs().
6819	*
6820	* We also exports the guest sysenter MSRs into the guest-state area in the VMCS.
6821	*
6822	* @returns VBox status code.
6823	* @param pVCpu The cross context virtual CPU structure.
6824	* @param pVmxTransient The VMX-transient structure.
6825	*
6826	* @remarks No-long-jump zone!!!
6827	*/
6828	static int hmR0VmxExportGuestMsrs(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
6829	{
6830	AssertPtr(pVCpu);
6831	AssertPtr(pVmxTransient);
6832
6833	PVM pVM = pVCpu->CTX_SUFF(pVM);
6834	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6835
6836	/*
6837	* MSRs that we use the auto-load/store MSR area in the VMCS.
6838	* For 64-bit hosts, we load/restore them lazily, see hmR0VmxLazyLoadGuestMsrs().
6839	* The host MSR values are updated when it's safe in hmR0VmxLazySaveHostMsrs().
6840	*
6841	* For nested-guests, the guests MSRs from the VM-entry MSR-load area are already
6842	* loaded (into the guest-CPU context) by the VMLAUNCH/VMRESUME instruction
6843	* emulation, nothing to do here.
6844	*/
6845	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_GUEST_AUTO_MSRS)
6846	{
6847	if ( !pVmxTransient->fIsNestedGuest
6848	&& pVM->hm.s.fAllow64BitGuests)
6849	{
6850	#if HC_ARCH_BITS == 32
6851	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SYSCALL_MSRS \| CPUMCTX_EXTRN_KERNEL_GS_BASE);
6852	Assert(!pVmxTransient->fIsNestedGuest);
6853
6854	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_LSTAR, pCtx->msrLSTAR, true, false);
6855	rc \|= hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_STAR, pCtx->msrSTAR, true, false);
6856	rc \|= hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_SF_MASK, pCtx->msrSFMASK, true, false);
6857	rc \|= hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_KERNEL_GS_BASE, pCtx->msrKERNELGSBASE, true, false);
6858	AssertRCReturn(rc, rc);
6859	#endif
6860	}
6861	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_GUEST_AUTO_MSRS);
6862	}
6863
6864	/*
6865	* Guest Sysenter MSRs.
6866	*/
6867	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_MSR_MASK)
6868	{
6869	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SYSENTER_MSRS);
6870
6871	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_CS_MSR)
6872	{
6873	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, pCtx->SysEnter.cs);
6874	AssertRCReturn(rc, rc);
6875	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_CS_MSR);
6876	}
6877
6878	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_EIP_MSR)
6879	{
6880	int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_SYSENTER_EIP, pCtx->SysEnter.eip);
6881	AssertRCReturn(rc, rc);
6882	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_EIP_MSR);
6883	}
6884
6885	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_ESP_MSR)
6886	{
6887	int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_SYSENTER_ESP, pCtx->SysEnter.esp);
6888	AssertRCReturn(rc, rc);
6889	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_ESP_MSR);
6890	}
6891	}
6892
6893	/*
6894	* Guest/host EFER MSR.
6895	*/
6896	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_EFER_MSR)
6897	{
6898	/* Whether we are using the VMCS to swap the EFER MSR must have been
6899	determined earlier while exporting VM-entry/VM-exit controls. */
6900	Assert(!(ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_ENTRY_EXIT_CTLS));
6901	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
6902
6903	if (hmR0VmxShouldSwapEferMsr(pVCpu))
6904	{
6905	/*
6906	* If the CPU supports VMCS controls for swapping EFER, use it. Otherwise, we have no option
6907	* but to use the auto-load store MSR area in the VMCS for swapping EFER. See @bugref{7368}.
6908	*/
6909	if (pVM->hm.s.vmx.fSupportsVmcsEfer)
6910	{
6911	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_EFER_FULL, pCtx->msrEFER);
6912	AssertRCReturn(rc, rc);
6913	}
6914	else
6915	{
6916	/*
6917	* We shall use the auto-load/store MSR area only for loading the EFER MSR but we must
6918	* continue to intercept guest read and write accesses to it, see @bugref{7386#c16}.
6919	*/
6920	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_EFER, pCtx->msrEFER,
6921	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6922	AssertRCReturn(rc, rc);
6923	}
6924	}
6925	else if (!pVM->hm.s.vmx.fSupportsVmcsEfer)
6926	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_EFER);
6927
6928	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_EFER_MSR);
6929	}
6930
6931	/*
6932	* Other MSRs.
6933	* Speculation Control (R/W).
6934	*/
6935	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_OTHER_MSRS)
6936	{
6937	HMVMX_CPUMCTX_ASSERT(pVCpu, HM_CHANGED_GUEST_OTHER_MSRS);
6938	if (pVM->cpum.ro.GuestFeatures.fIbrs)
6939	{
6940	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_IA32_SPEC_CTRL, CPUMGetGuestSpecCtrl(pVCpu),
6941	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6942	AssertRCReturn(rc, rc);
6943	}
6944	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_OTHER_MSRS);
6945	}
6946
6947	return VINF_SUCCESS;
6948	}
6949
6950
6951	/**
6952	* Selects up the appropriate function to run guest code.
6953	*
6954	* @returns VBox status code.
6955	* @param pVCpu The cross context virtual CPU structure.
6956	* @param pVmxTransient The VMX-transient structure.
6957	*
6958	* @remarks No-long-jump zone!!!
6959	*/
6960	static int hmR0VmxSelectVMRunHandler(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
6961	{
6962	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6963	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6964
6965	if (CPUMIsGuestInLongModeEx(pCtx))
6966	{
6967	#ifndef VBOX_ENABLE_64_BITS_GUESTS
6968	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
6969	#endif
6970	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests); /* Guaranteed by hmR3InitFinalizeR0(). */
6971	#if HC_ARCH_BITS == 32
6972	/* 32-bit host. We need to switch to 64-bit before running the 64-bit guest. */
6973	if (pVmcsInfo->pfnStartVM != VMXR0SwitcherStartVM64)
6974	{
6975	#ifdef VBOX_STRICT
6976	if (pVmcsInfo->pfnStartVM != NULL) /* Very first VM-entry would have saved host-state already, ignore it. */
6977	{
6978	/* Currently, all mode changes sends us back to ring-3, so these should be set. See @bugref{6944}. */
6979	uint64_t const fCtxChanged = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
6980	RT_UNTRUSTED_NONVOLATILE_COPY_FENCE();
6981	AssertMsg(fCtxChanged & (HM_CHANGED_VMX_ENTRY_EXIT_CTLS \| HM_CHANGED_GUEST_EFER_MSR),
6982	("fCtxChanged=%#RX64\n", fCtxChanged));
6983	}
6984	#endif
6985	pVmcsInfo->pfnStartVM = VMXR0SwitcherStartVM64;
6986
6987	/* Mark that we've switched to 64-bit handler, we can't safely switch back to 32-bit for
6988	the rest of the VM run (until VM reset). See @bugref{8432#c7}. */
6989	pVmcsInfo->fSwitchedTo64on32 = true;
6990	Log4Func(("Selected 64-bit switcher\n"));
6991	}
6992	#else
6993	/* 64-bit host. */
6994	pVmcsInfo->pfnStartVM = VMXR0StartVM64;
6995	#endif
6996	}
6997	else
6998	{
6999	/* Guest is not in long mode, use the 32-bit handler. */
7000	#if HC_ARCH_BITS == 32
7001	if ( pVmcsInfo->pfnStartVM != VMXR0StartVM32
7002	&& !pVmcsInfo->fSwitchedTo64on32 /* If set, guest mode change does not imply switcher change. */
7003	&& pVmcsInfo->pfnStartVM != NULL) /* Very first VM-entry would have saved host-state already, ignore it. */
7004	{
7005	# ifdef VBOX_STRICT
7006	/* Currently, all mode changes sends us back to ring-3, so these should be set. See @bugref{6944}. */
7007	uint64_t const fCtxChanged = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
7008	RT_UNTRUSTED_NONVOLATILE_COPY_FENCE();
7009	AssertMsg(fCtxChanged & (HM_CHANGED_VMX_ENTRY_EXIT_CTLS \| HM_CHANGED_GUEST_EFER_MSR),
7010	("fCtxChanged=%#RX64\n", fCtxChanged));
7011	# endif
7012	}
7013	# ifdef VBOX_ENABLE_64_BITS_GUESTS
7014	/*
7015	* Keep using the 64-bit switcher even though we're in 32-bit because of bad Intel
7016	* design, see @bugref{8432#c7}. If real-on-v86 mode is active, clear the 64-bit
7017	* switcher flag now because we know the guest is in a sane state where it's safe
7018	* to use the 32-bit switcher. Otherwise, check the guest state if it's safe to use
7019	* the much faster 32-bit switcher again.
7020	*/
7021	if (!pVmcsInfo->fSwitchedTo64on32)
7022	{
7023	if (pVmcsInfo->pfnStartVM != VMXR0StartVM32)
7024	Log4Func(("Selected 32-bit switcher\n"));
7025	pVmcsInfo->pfnStartVM = VMXR0StartVM32;
7026	}
7027	else
7028	{
7029	Assert(pVmcsInfo->pfnStartVM == VMXR0SwitcherStartVM64);
7030	if ( pVmcsInfo->RealMode.fRealOnV86Active
7031	\|\| hmR0VmxIs32BitSwitcherSafe(pCtx))
7032	{
7033	pVmcsInfo->fSwitchedTo64on32 = false;
7034	pVmcsInfo->pfnStartVM = VMXR0StartVM32;
7035	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_EFER_MSR
7036	\| HM_CHANGED_VMX_ENTRY_EXIT_CTLS
7037	\| HM_CHANGED_HOST_CONTEXT);
7038	Log4Func(("Selected 32-bit switcher (safe)\n"));
7039	}
7040	}
7041	# else
7042	pVmcsInfo->pfnStartVM = VMXR0StartVM32;
7043	# endif
7044	#else
7045	pVmcsInfo->pfnStartVM = VMXR0StartVM32;
7046	#endif
7047	}
7048	Assert(pVmcsInfo->pfnStartVM);
7049	return VINF_SUCCESS;
7050	}
7051
7052
7053	/**
7054	* Wrapper for running the guest code in VT-x.
7055	*
7056	* @returns VBox status code, no informational status codes.
7057	* @param pVCpu The cross context virtual CPU structure.
7058	* @param pVmxTransient The VMX-transient structure.
7059	*
7060	* @remarks No-long-jump zone!!!
7061	*/
7062	DECLINLINE(int) hmR0VmxRunGuest(PVMCPU pVCpu, PCVMXTRANSIENT pVmxTransient)
7063	{
7064	/* Mark that HM is the keeper of all guest-CPU registers now that we're going to execute guest code. */
7065	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7066	pCtx->fExtrn \|= HMVMX_CPUMCTX_EXTRN_ALL \| CPUMCTX_EXTRN_KEEPER_HM;
7067
7068	/** @todo Add stats for VMRESUME vs VMLAUNCH. */
7069
7070	/*
7071	* 64-bit Windows uses XMM registers in the kernel as the Microsoft compiler expresses
7072	* floating-point operations using SSE instructions. Some XMM registers (XMM6-XMM15) are
7073	* callee-saved and thus the need for this XMM wrapper.
7074	*
7075	* See MSDN "Configuring Programs for 64-bit/x64 Software Conventions / Register Usage".
7076	*/
7077	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
7078	bool const fResumeVM = RT_BOOL(pVmcsInfo->fVmcsState & VMX_V_VMCS_LAUNCH_STATE_LAUNCHED);
7079	PVM pVM = pVCpu->CTX_SUFF(pVM);
7080	#ifdef VBOX_WITH_KERNEL_USING_XMM
7081	int rc = hmR0VMXStartVMWrapXMM(fResumeVM, pCtx, &pVCpu->hm.s.vmx.VmcsCache, pVM, pVCpu, pVmcsInfo->pfnStartVM);
7082	#else
7083	int rc = pVmcsInfo->pfnStartVM(fResumeVM, pCtx, &pVCpu->hm.s.vmx.VmcsCache, pVM, pVCpu);
7084	#endif
7085	AssertMsg(rc <= VINF_SUCCESS, ("%Rrc\n", rc));
7086	return rc;
7087	}
7088
7089
7090	/**
7091	* Reports world-switch error and dumps some useful debug info.
7092	*
7093	* @param pVCpu The cross context virtual CPU structure.
7094	* @param rcVMRun The return code from VMLAUNCH/VMRESUME.
7095	* @param pVmxTransient The VMX-transient structure (only
7096	* exitReason updated).
7097	*/
7098	static void hmR0VmxReportWorldSwitchError(PVMCPU pVCpu, int rcVMRun, PVMXTRANSIENT pVmxTransient)
7099	{
7100	Assert(pVCpu);
7101	Assert(pVmxTransient);
7102	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
7103
7104	Log4Func(("VM-entry failure: %Rrc\n", rcVMRun));
7105	switch (rcVMRun)
7106	{
7107	case VERR_VMX_INVALID_VMXON_PTR:
7108	AssertFailed();
7109	break;
7110	case VINF_SUCCESS: /* VMLAUNCH/VMRESUME succeeded but VM-entry failed... yeah, true story. */
7111	case VERR_VMX_UNABLE_TO_START_VM: /* VMLAUNCH/VMRESUME itself failed. */
7112	{
7113	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &pVCpu->hm.s.vmx.LastError.u32ExitReason);
7114	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
7115	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
7116	AssertRC(rc);
7117
7118	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
7119	/* LastError.idCurrentCpu was already updated in hmR0VmxPreRunGuestCommitted().
7120	Cannot do it here as we may have been long preempted. */
7121
7122	#ifdef VBOX_STRICT
7123	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
7124	Log4(("uExitReason %#RX32 (VmxTransient %#RX16)\n", pVCpu->hm.s.vmx.LastError.u32ExitReason,
7125	pVmxTransient->uExitReason));
7126	Log4(("Exit Qualification %#RX64\n", pVmxTransient->uExitQual));
7127	Log4(("InstrError %#RX32\n", pVCpu->hm.s.vmx.LastError.u32InstrError));
7128	if (pVCpu->hm.s.vmx.LastError.u32InstrError <= HMVMX_INSTR_ERROR_MAX)
7129	Log4(("InstrError Desc. \"%s\"\n", g_apszVmxInstrErrors[pVCpu->hm.s.vmx.LastError.u32InstrError]));
7130	else
7131	Log4(("InstrError Desc. Range exceeded %u\n", HMVMX_INSTR_ERROR_MAX));
7132	Log4(("Entered host CPU %u\n", pVCpu->hm.s.vmx.LastError.idEnteredCpu));
7133	Log4(("Current host CPU %u\n", pVCpu->hm.s.vmx.LastError.idCurrentCpu));
7134
7135	/* VMX control bits. */
7136	uint32_t u32Val;
7137	uint64_t u64Val;
7138	RTHCUINTREG uHCReg;
7139	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val); AssertRC(rc);
7140	Log4(("VMX_VMCS32_CTRL_PIN_EXEC %#RX32\n", u32Val));
7141	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val); AssertRC(rc);
7142	Log4(("VMX_VMCS32_CTRL_PROC_EXEC %#RX32\n", u32Val));
7143	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
7144	{
7145	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val); AssertRC(rc);
7146	Log4(("VMX_VMCS32_CTRL_PROC_EXEC2 %#RX32\n", u32Val));
7147	}
7148	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val); AssertRC(rc);
7149	Log4(("VMX_VMCS32_CTRL_ENTRY %#RX32\n", u32Val));
7150	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val); AssertRC(rc);
7151	Log4(("VMX_VMCS32_CTRL_EXIT %#RX32\n", u32Val));
7152	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_CR3_TARGET_COUNT, &u32Val); AssertRC(rc);
7153	Log4(("VMX_VMCS32_CTRL_CR3_TARGET_COUNT %#RX32\n", u32Val));
7154	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32Val); AssertRC(rc);
7155	Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", u32Val));
7156	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &u32Val); AssertRC(rc);
7157	Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", u32Val));
7158	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &u32Val); AssertRC(rc);
7159	Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %u\n", u32Val));
7160	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, &u32Val); AssertRC(rc);
7161	Log4(("VMX_VMCS32_CTRL_TPR_THRESHOLD %u\n", u32Val));
7162	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &u32Val); AssertRC(rc);
7163	Log4(("VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT %u (guest MSRs)\n", u32Val));
7164	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &u32Val); AssertRC(rc);
7165	Log4(("VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT %u (host MSRs)\n", u32Val));
7166	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &u32Val); AssertRC(rc);
7167	Log4(("VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT %u (guest MSRs)\n", u32Val));
7168	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, &u32Val); AssertRC(rc);
7169	Log4(("VMX_VMCS32_CTRL_EXCEPTION_BITMAP %#RX32\n", u32Val));
7170	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, &u32Val); AssertRC(rc);
7171	Log4(("VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK %#RX32\n", u32Val));
7172	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, &u32Val); AssertRC(rc);
7173	Log4(("VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH %#RX32\n", u32Val));
7174	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, &uHCReg); AssertRC(rc);
7175	Log4(("VMX_VMCS_CTRL_CR0_MASK %#RHr\n", uHCReg));
7176	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uHCReg); AssertRC(rc);
7177	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
7178	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, &uHCReg); AssertRC(rc);
7179	Log4(("VMX_VMCS_CTRL_CR4_MASK %#RHr\n", uHCReg));
7180	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uHCReg); AssertRC(rc);
7181	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
7182	if (pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging)
7183	{
7184	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
7185	Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
7186	}
7187
7188	/* Guest bits. */
7189	rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RIP, &u64Val); AssertRC(rc);
7190	Log4(("Old Guest Rip %#RX64 New %#RX64\n", pVCpu->cpum.GstCtx.rip, u64Val));
7191	rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RSP, &u64Val); AssertRC(rc);
7192	Log4(("Old Guest Rsp %#RX64 New %#RX64\n", pVCpu->cpum.GstCtx.rsp, u64Val));
7193	rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Val); AssertRC(rc);
7194	Log4(("Old Guest Rflags %#RX32 New %#RX32\n", pVCpu->cpum.GstCtx.eflags.u32, u32Val));
7195	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fVpid)
7196	{
7197	rc = VMXReadVmcs32(VMX_VMCS16_VPID, &u32Val); AssertRC(rc);
7198	Log4(("VMX_VMCS16_VPID %u\n", u32Val));
7199	}
7200
7201	/* Host bits. */
7202	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR0, &uHCReg); AssertRC(rc);
7203	Log4(("Host CR0 %#RHr\n", uHCReg));
7204	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR3, &uHCReg); AssertRC(rc);
7205	Log4(("Host CR3 %#RHr\n", uHCReg));
7206	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR4, &uHCReg); AssertRC(rc);
7207	Log4(("Host CR4 %#RHr\n", uHCReg));
7208
7209	RTGDTR HostGdtr;
7210	PCX86DESCHC pDesc;
7211	ASMGetGDTR(&HostGdtr);
7212	rc = VMXReadVmcs32(VMX_VMCS16_HOST_CS_SEL, &u32Val); AssertRC(rc);
7213	Log4(("Host CS %#08x\n", u32Val));
7214	if (u32Val < HostGdtr.cbGdt)
7215	{
7216	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
7217	hmR0DumpDescriptor(pDesc, u32Val, "CS: ");
7218	}
7219
7220	rc = VMXReadVmcs32(VMX_VMCS16_HOST_DS_SEL, &u32Val); AssertRC(rc);
7221	Log4(("Host DS %#08x\n", u32Val));
7222	if (u32Val < HostGdtr.cbGdt)
7223	{
7224	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
7225	hmR0DumpDescriptor(pDesc, u32Val, "DS: ");
7226	}
7227
7228	rc = VMXReadVmcs32(VMX_VMCS16_HOST_ES_SEL, &u32Val); AssertRC(rc);
7229	Log4(("Host ES %#08x\n", u32Val));
7230	if (u32Val < HostGdtr.cbGdt)
7231	{
7232	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
7233	hmR0DumpDescriptor(pDesc, u32Val, "ES: ");
7234	}
7235
7236	rc = VMXReadVmcs32(VMX_VMCS16_HOST_FS_SEL, &u32Val); AssertRC(rc);
7237	Log4(("Host FS %#08x\n", u32Val));
7238	if (u32Val < HostGdtr.cbGdt)
7239	{
7240	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
7241	hmR0DumpDescriptor(pDesc, u32Val, "FS: ");
7242	}
7243
7244	rc = VMXReadVmcs32(VMX_VMCS16_HOST_GS_SEL, &u32Val); AssertRC(rc);
7245	Log4(("Host GS %#08x\n", u32Val));
7246	if (u32Val < HostGdtr.cbGdt)
7247	{
7248	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
7249	hmR0DumpDescriptor(pDesc, u32Val, "GS: ");
7250	}
7251
7252	rc = VMXReadVmcs32(VMX_VMCS16_HOST_SS_SEL, &u32Val); AssertRC(rc);
7253	Log4(("Host SS %#08x\n", u32Val));
7254	if (u32Val < HostGdtr.cbGdt)
7255	{
7256	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
7257	hmR0DumpDescriptor(pDesc, u32Val, "SS: ");
7258	}
7259
7260	rc = VMXReadVmcs32(VMX_VMCS16_HOST_TR_SEL, &u32Val); AssertRC(rc);
7261	Log4(("Host TR %#08x\n", u32Val));
7262	if (u32Val < HostGdtr.cbGdt)
7263	{
7264	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
7265	hmR0DumpDescriptor(pDesc, u32Val, "TR: ");
7266	}
7267
7268	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_TR_BASE, &uHCReg); AssertRC(rc);
7269	Log4(("Host TR Base %#RHv\n", uHCReg));
7270	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_GDTR_BASE, &uHCReg); AssertRC(rc);
7271	Log4(("Host GDTR Base %#RHv\n", uHCReg));
7272	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_IDTR_BASE, &uHCReg); AssertRC(rc);
7273	Log4(("Host IDTR Base %#RHv\n", uHCReg));
7274	rc = VMXReadVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, &u32Val); AssertRC(rc);
7275	Log4(("Host SYSENTER CS %#08x\n", u32Val));
7276	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_SYSENTER_EIP, &uHCReg); AssertRC(rc);
7277	Log4(("Host SYSENTER EIP %#RHv\n", uHCReg));
7278	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_SYSENTER_ESP, &uHCReg); AssertRC(rc);
7279	Log4(("Host SYSENTER ESP %#RHv\n", uHCReg));
7280	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_RSP, &uHCReg); AssertRC(rc);
7281	Log4(("Host RSP %#RHv\n", uHCReg));
7282	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_RIP, &uHCReg); AssertRC(rc);
7283	Log4(("Host RIP %#RHv\n", uHCReg));
7284	# if HC_ARCH_BITS == 64
7285	Log4(("MSR_K6_EFER = %#RX64\n", ASMRdMsr(MSR_K6_EFER)));
7286	Log4(("MSR_K8_CSTAR = %#RX64\n", ASMRdMsr(MSR_K8_CSTAR)));
7287	Log4(("MSR_K8_LSTAR = %#RX64\n", ASMRdMsr(MSR_K8_LSTAR)));
7288	Log4(("MSR_K6_STAR = %#RX64\n", ASMRdMsr(MSR_K6_STAR)));
7289	Log4(("MSR_K8_SF_MASK = %#RX64\n", ASMRdMsr(MSR_K8_SF_MASK)));
7290	Log4(("MSR_K8_KERNEL_GS_BASE = %#RX64\n", ASMRdMsr(MSR_K8_KERNEL_GS_BASE)));
7291	# endif
7292	#endif /* VBOX_STRICT */
7293	break;
7294	}
7295
7296	default:
7297	/* Impossible */
7298	AssertMsgFailed(("hmR0VmxReportWorldSwitchError %Rrc (%#x)\n", rcVMRun, rcVMRun));
7299	break;
7300	}
7301	}
7302
7303
7304	#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
7305	# ifndef VMX_USE_CACHED_VMCS_ACCESSES
7306	# error "VMX_USE_CACHED_VMCS_ACCESSES not defined when it should be!"
7307	# endif
7308
7309	/**
7310	* Initialize the VMCS-Read cache.
7311	*
7312	* The VMCS cache is used for 32-bit hosts running 64-bit guests (except 32-bit
7313	* Darwin which runs with 64-bit paging in 32-bit mode) for 64-bit fields that
7314	* cannot be accessed in 32-bit mode. Some 64-bit fields -can- be accessed
7315	* (those that have a 32-bit FULL & HIGH part).
7316	*
7317	* @param pVCpu The cross context virtual CPU structure.
7318	*/
7319	static void hmR0VmxInitVmcsReadCache(PVMCPU pVCpu)
7320	{
7321	#define VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, idxField) \
7322	do { \
7323	Assert(pCache->Read.aField[idxField##_CACHE_IDX] == 0); \
7324	pCache->Read.aField[idxField##_CACHE_IDX] = idxField; \
7325	pCache->Read.aFieldVal[idxField##_CACHE_IDX] = 0; \
7326	++cReadFields; \
7327	} while (0)
7328
7329	PVMXVMCSCACHE pCache = &pVCpu->hm.s.vmx.VmcsCache;
7330	uint32_t cReadFields = 0;
7331
7332	/*
7333	* Don't remove the #if 0'd fields in this code. They're listed here for consistency
7334	* and serve to indicate exceptions to the rules.
7335	*/
7336
7337	/* Guest-natural selector base fields. */
7338	#if 0
7339	/* These are 32-bit in practice. See Intel spec. 2.5 "Control Registers". */
7340	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR0);
7341	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR4);
7342	#endif
7343	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_ES_BASE);
7344	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CS_BASE);
7345	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SS_BASE);
7346	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_DS_BASE);
7347	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_FS_BASE);
7348	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_GS_BASE);
7349	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_LDTR_BASE);
7350	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_TR_BASE);
7351	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_GDTR_BASE);
7352	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_IDTR_BASE);
7353	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_RSP);
7354	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_RIP);
7355	#if 0
7356	/* Unused natural width guest-state fields. */
7357	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS);
7358	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR3); /* Handled in nested paging case */
7359	#endif
7360	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SYSENTER_ESP);
7361	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SYSENTER_EIP);
7362
7363	/* 64-bit guest-state fields; unused as we use two 32-bit VMREADs for
7364	these 64-bit fields (using "FULL" and "HIGH" fields). */
7365	#if 0
7366	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL);
7367	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_DEBUGCTL_FULL);
7368	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PAT_FULL);
7369	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_EFER_FULL);
7370	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL);
7371	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE0_FULL);
7372	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE1_FULL);
7373	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE2_FULL);
7374	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE3_FULL);
7375	#endif
7376
7377	/* Natural width guest-state fields. */
7378	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_RO_EXIT_QUALIFICATION);
7379	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_RO_GUEST_LINEAR_ADDR);
7380
7381	if (pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging)
7382	{
7383	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR3);
7384	AssertMsg(cReadFields == VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX, ("cReadFields=%u expected %u\n", cReadFields,
7385	VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX));
7386	pCache->Read.cValidEntries = VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX;
7387	}
7388	else
7389	{
7390	AssertMsg(cReadFields == VMX_VMCS_MAX_CACHE_IDX, ("cReadFields=%u expected %u\n", cReadFields, VMX_VMCS_MAX_CACHE_IDX));
7391	pCache->Read.cValidEntries = VMX_VMCS_MAX_CACHE_IDX;
7392	}
7393
7394	#undef VMXLOCAL_INIT_READ_CACHE_FIELD
7395	}
7396
7397
7398	/**
7399	* Writes a field into the VMCS. This can either directly invoke a VMWRITE or
7400	* queue up the VMWRITE by using the VMCS write cache (on 32-bit hosts, except
7401	* darwin, running 64-bit guests).
7402	*
7403	* @returns VBox status code.
7404	* @param pVCpu The cross context virtual CPU structure.
7405	* @param idxField The VMCS field encoding.
7406	* @param u64Val 16, 32 or 64-bit value.
7407	*/
7408	VMMR0DECL(int) VMXWriteVmcs64Ex(PVMCPU pVCpu, uint32_t idxField, uint64_t u64Val)
7409	{
7410	AssertPtr(pVCpu);
7411	int rc;
7412	switch (idxField)
7413	{
7414	/*
7415	* These fields consists of a "FULL" and a "HIGH" part which can be written to individually.
7416	*/
7417	/* 64-bit Control fields. */
7418	case VMX_VMCS64_CTRL_IO_BITMAP_A_FULL:
7419	case VMX_VMCS64_CTRL_IO_BITMAP_B_FULL:
7420	case VMX_VMCS64_CTRL_MSR_BITMAP_FULL:
7421	case VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL:
7422	case VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL:
7423	case VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL:
7424	case VMX_VMCS64_CTRL_EXEC_VMCS_PTR_FULL:
7425	case VMX_VMCS64_CTRL_TSC_OFFSET_FULL:
7426	case VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL:
7427	case VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL:
7428	case VMX_VMCS64_CTRL_VMFUNC_CTRLS_FULL:
7429	case VMX_VMCS64_CTRL_EPTP_FULL:
7430	case VMX_VMCS64_CTRL_EPTP_LIST_FULL:
7431	/* 64-bit Guest-state fields. */
7432	case VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL:
7433	case VMX_VMCS64_GUEST_DEBUGCTL_FULL:
7434	case VMX_VMCS64_GUEST_PAT_FULL:
7435	case VMX_VMCS64_GUEST_EFER_FULL:
7436	case VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL:
7437	case VMX_VMCS64_GUEST_PDPTE0_FULL:
7438	case VMX_VMCS64_GUEST_PDPTE1_FULL:
7439	case VMX_VMCS64_GUEST_PDPTE2_FULL:
7440	case VMX_VMCS64_GUEST_PDPTE3_FULL:
7441	/* 64-bit Host-state fields. */
7442	case VMX_VMCS64_HOST_PAT_FULL:
7443	case VMX_VMCS64_HOST_EFER_FULL:
7444	case VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_FULL:
7445	{
7446	rc = VMXWriteVmcs32(idxField, RT_LO_U32(u64Val));
7447	rc \|= VMXWriteVmcs32(idxField + 1, RT_HI_U32(u64Val));
7448	break;
7449	}
7450
7451	/*
7452	* These fields do not have high and low parts. Queue up the VMWRITE by using the VMCS write-cache (for 64-bit
7453	* values). When we switch the host to 64-bit mode for running 64-bit guests, these VMWRITEs get executed then.
7454	*/
7455	/* Natural-width Guest-state fields. */
7456	case VMX_VMCS_GUEST_CR3:
7457	case VMX_VMCS_GUEST_ES_BASE:
7458	case VMX_VMCS_GUEST_CS_BASE:
7459	case VMX_VMCS_GUEST_SS_BASE:
7460	case VMX_VMCS_GUEST_DS_BASE:
7461	case VMX_VMCS_GUEST_FS_BASE:
7462	case VMX_VMCS_GUEST_GS_BASE:
7463	case VMX_VMCS_GUEST_LDTR_BASE:
7464	case VMX_VMCS_GUEST_TR_BASE:
7465	case VMX_VMCS_GUEST_GDTR_BASE:
7466	case VMX_VMCS_GUEST_IDTR_BASE:
7467	case VMX_VMCS_GUEST_RSP:
7468	case VMX_VMCS_GUEST_RIP:
7469	case VMX_VMCS_GUEST_SYSENTER_ESP:
7470	case VMX_VMCS_GUEST_SYSENTER_EIP:
7471	{
7472	if (!(RT_HI_U32(u64Val)))
7473	{
7474	/* If this field is 64-bit, VT-x will zero out the top bits. */
7475	rc = VMXWriteVmcs32(idxField, RT_LO_U32(u64Val));
7476	}
7477	else
7478	{
7479	/* Assert that only the 32->64 switcher case should ever come here. */
7480	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests);
7481	rc = VMXWriteCachedVmcsEx(pVCpu, idxField, u64Val);
7482	}
7483	break;
7484	}
7485
7486	default:
7487	{
7488	AssertMsgFailed(("VMXWriteVmcs64Ex: Invalid field %#RX32 (pVCpu=%p u64Val=%#RX64)\n", idxField, pVCpu, u64Val));
7489	pVCpu->hm.s.u32HMError = idxField;
7490	rc = VERR_INVALID_PARAMETER;
7491	break;
7492	}
7493	}
7494	AssertRCReturn(rc, rc);
7495	return rc;
7496	}
7497
7498
7499	/**
7500	* Queue up a VMWRITE by using the VMCS write cache.
7501	* This is only used on 32-bit hosts (except darwin) for 64-bit guests.
7502	*
7503	* @param pVCpu The cross context virtual CPU structure.
7504	* @param idxField The VMCS field encoding.
7505	* @param u64Val 16, 32 or 64-bit value.
7506	*/
7507	VMMR0DECL(int) VMXWriteCachedVmcsEx(PVMCPU pVCpu, uint32_t idxField, uint64_t u64Val)
7508	{
7509	AssertPtr(pVCpu);
7510	PVMXVMCSCACHE pCache = &pVCpu->hm.s.vmx.VmcsCache;
7511
7512	AssertMsgReturn(pCache->Write.cValidEntries < VMX_VMCS_CACHE_MAX_ENTRY - 1,
7513	("entries=%u\n", pCache->Write.cValidEntries), VERR_ACCESS_DENIED);
7514
7515	/* Make sure there are no duplicates. */
7516	for (uint32_t i = 0; i < pCache->Write.cValidEntries; i++)
7517	{
7518	if (pCache->Write.aField[i] == idxField)
7519	{
7520	pCache->Write.aFieldVal[i] = u64Val;
7521	return VINF_SUCCESS;
7522	}
7523	}
7524
7525	pCache->Write.aField[pCache->Write.cValidEntries] = idxField;
7526	pCache->Write.aFieldVal[pCache->Write.cValidEntries] = u64Val;
7527	pCache->Write.cValidEntries++;
7528	return VINF_SUCCESS;
7529	}
7530	#endif /* HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) */
7531
7532
7533	/**
7534	* Sets up the usage of TSC-offsetting and updates the VMCS.
7535	*
7536	* If offsetting is not possible, cause VM-exits on RDTSC(P)s. Also sets up the
7537	* VMX-preemption timer.
7538	*
7539	* @returns VBox status code.
7540	* @param pVCpu The cross context virtual CPU structure.
7541	* @param pVmxTransient The VMX-transient structure.
7542	*
7543	* @remarks No-long-jump zone!!!
7544	*/
7545	static void hmR0VmxUpdateTscOffsettingAndPreemptTimer(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
7546	{
7547	bool fOffsettedTsc;
7548	bool fParavirtTsc;
7549	uint64_t uTscOffset;
7550	PVM pVM = pVCpu->CTX_SUFF(pVM);
7551	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
7552
7553	if (pVM->hm.s.vmx.fUsePreemptTimer)
7554	{
7555	uint64_t cTicksToDeadline = TMCpuTickGetDeadlineAndTscOffset(pVM, pVCpu, &uTscOffset, &fOffsettedTsc, &fParavirtTsc);
7556
7557	/* Make sure the returned values have sane upper and lower boundaries. */
7558	uint64_t u64CpuHz = SUPGetCpuHzFromGipBySetIndex(g_pSUPGlobalInfoPage, pVCpu->iHostCpuSet);
7559	cTicksToDeadline = RT_MIN(cTicksToDeadline, u64CpuHz / 64); /* 1/64th of a second */
7560	cTicksToDeadline = RT_MAX(cTicksToDeadline, u64CpuHz / 2048); /* 1/2048th of a second */
7561	cTicksToDeadline >>= pVM->hm.s.vmx.cPreemptTimerShift;
7562
7563	/** @todo r=ramshankar: We need to find a way to integrate nested-guest
7564	* preemption timers here. We probably need to clamp the preemption timer,
7565	* after converting the timer value to the host. */
7566	uint32_t cPreemptionTickCount = (uint32_t)RT_MIN(cTicksToDeadline, UINT32_MAX - 16);
7567	int rc = VMXWriteVmcs32(VMX_VMCS32_PREEMPT_TIMER_VALUE, cPreemptionTickCount);
7568	AssertRC(rc);
7569	}
7570	else
7571	fOffsettedTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &uTscOffset, &fParavirtTsc);
7572
7573	if (fParavirtTsc)
7574	{
7575	/* Currently neither Hyper-V nor KVM need to update their paravirt. TSC
7576	information before every VM-entry, hence disable it for performance sake. */
7577	#if 0
7578	int rc = GIMR0UpdateParavirtTsc(pVM, 0 /* u64Offset */);
7579	AssertRC(rc);
7580	#endif
7581	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
7582	}
7583
7584	if ( fOffsettedTsc
7585	&& RT_LIKELY(!pVCpu->hm.s.fDebugWantRdTscExit))
7586	{
7587	if (pVmxTransient->fIsNestedGuest)
7588	uTscOffset = CPUMApplyNestedGuestTscOffset(pVCpu, uTscOffset);
7589	hmR0VmxSetTscOffsetVmcs(pVCpu, pVmcsInfo, uTscOffset);
7590	hmR0VmxRemoveProcCtlsVmcs(pVCpu, pVmxTransient, VMX_PROC_CTLS_RDTSC_EXIT);
7591	}
7592	else
7593	{
7594	/* We can't use TSC-offsetting (non-fixed TSC, warp drive active etc.), VM-exit on RDTSC(P). */
7595	hmR0VmxSetProcCtlsVmcs(pVmxTransient, VMX_PROC_CTLS_RDTSC_EXIT);
7596	}
7597	}
7598
7599
7600	/**
7601	* Gets the IEM exception flags for the specified vector and IDT vectoring /
7602	* VM-exit interruption info type.
7603	*
7604	* @returns The IEM exception flags.
7605	* @param uVector The event vector.
7606	* @param uVmxEventType The VMX event type.
7607	*
7608	* @remarks This function currently only constructs flags required for
7609	* IEMEvaluateRecursiveXcpt and not the complete flags (e.g, error-code
7610	* and CR2 aspects of an exception are not included).
7611	*/
7612	static uint32_t hmR0VmxGetIemXcptFlags(uint8_t uVector, uint32_t uVmxEventType)
7613	{
7614	uint32_t fIemXcptFlags;
7615	switch (uVmxEventType)
7616	{
7617	case VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT:
7618	case VMX_IDT_VECTORING_INFO_TYPE_NMI:
7619	fIemXcptFlags = IEM_XCPT_FLAGS_T_CPU_XCPT;
7620	break;
7621
7622	case VMX_IDT_VECTORING_INFO_TYPE_EXT_INT:
7623	fIemXcptFlags = IEM_XCPT_FLAGS_T_EXT_INT;
7624	break;
7625
7626	case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
7627	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT \| IEM_XCPT_FLAGS_ICEBP_INSTR;
7628	break;
7629
7630	case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT:
7631	{
7632	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
7633	if (uVector == X86_XCPT_BP)
7634	fIemXcptFlags \|= IEM_XCPT_FLAGS_BP_INSTR;
7635	else if (uVector == X86_XCPT_OF)
7636	fIemXcptFlags \|= IEM_XCPT_FLAGS_OF_INSTR;
7637	else
7638	{
7639	fIemXcptFlags = 0;
7640	AssertMsgFailed(("Unexpected vector for software exception. uVector=%#x", uVector));
7641	}
7642	break;
7643	}
7644
7645	case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
7646	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
7647	break;
7648
7649	default:
7650	fIemXcptFlags = 0;
7651	AssertMsgFailed(("Unexpected vector type! uVmxEventType=%#x uVector=%#x", uVmxEventType, uVector));
7652	break;
7653	}
7654	return fIemXcptFlags;
7655	}
7656
7657
7658	/**
7659	* Sets an event as a pending event to be injected into the guest.
7660	*
7661	* @param pVCpu The cross context virtual CPU structure.
7662	* @param u32IntInfo The VM-entry interruption-information field.
7663	* @param cbInstr The VM-entry instruction length in bytes (for software
7664	* interrupts, exceptions and privileged software
7665	* exceptions).
7666	* @param u32ErrCode The VM-entry exception error code.
7667	* @param GCPtrFaultAddress The fault-address (CR2) in case it's a
7668	* page-fault.
7669	*/
7670	DECLINLINE(void) hmR0VmxSetPendingEvent(PVMCPU pVCpu, uint32_t u32IntInfo, uint32_t cbInstr, uint32_t u32ErrCode,
7671	RTGCUINTPTR GCPtrFaultAddress)
7672	{
7673	Assert(!pVCpu->hm.s.Event.fPending);
7674	pVCpu->hm.s.Event.fPending = true;
7675	pVCpu->hm.s.Event.u64IntInfo = u32IntInfo;
7676	pVCpu->hm.s.Event.u32ErrCode = u32ErrCode;
7677	pVCpu->hm.s.Event.cbInstr = cbInstr;
7678	pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress;
7679	}
7680
7681
7682	/**
7683	* Sets an external interrupt as pending-for-injection into the VM.
7684	*
7685	* @param pVCpu The cross context virtual CPU structure.
7686	* @param u8Interrupt The external interrupt vector.
7687	*/
7688	DECLINLINE(void) hmR0VmxSetPendingExtInt(PVMCPU pVCpu, uint8_t u8Interrupt)
7689	{
7690	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, u8Interrupt)
7691	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
7692	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7693	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7694	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7695	}
7696
7697
7698	/**
7699	* Sets an NMI (\#NMI) exception as pending-for-injection into the VM.
7700	*
7701	* @param pVCpu The cross context virtual CPU structure.
7702	*/
7703	DECLINLINE(void) hmR0VmxSetPendingXcptNmi(PVMCPU pVCpu)
7704	{
7705	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_NMI)
7706	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_NMI)
7707	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7708	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7709	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7710	}
7711
7712
7713	/**
7714	* Sets a double-fault (\#DF) exception as pending-for-injection into the VM.
7715	*
7716	* @param pVCpu The cross context virtual CPU structure.
7717	*/
7718	DECLINLINE(void) hmR0VmxSetPendingXcptDF(PVMCPU pVCpu)
7719	{
7720	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DF)
7721	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7722	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7723	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7724	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7725	}
7726
7727
7728	/**
7729	* Sets an invalid-opcode (\#UD) exception as pending-for-injection into the VM.
7730	*
7731	* @param pVCpu The cross context virtual CPU structure.
7732	*/
7733	DECLINLINE(void) hmR0VmxSetPendingXcptUD(PVMCPU pVCpu)
7734	{
7735	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_UD)
7736	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7737	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7738	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7739	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7740	}
7741
7742
7743	/**
7744	* Sets a debug (\#DB) exception as pending-for-injection into the VM.
7745	*
7746	* @param pVCpu The cross context virtual CPU structure.
7747	*/
7748	DECLINLINE(void) hmR0VmxSetPendingXcptDB(PVMCPU pVCpu)
7749	{
7750	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DB)
7751	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7752	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7753	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7754	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7755	}
7756
7757
7758	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7759	/**
7760	* Sets a general-protection (\#GP) exception as pending-for-injection into the VM.
7761	*
7762	* @param pVCpu The cross context virtual CPU structure.
7763	* @param u32ErrCode The error code for the general-protection exception.
7764	*/
7765	DECLINLINE(void) hmR0VmxSetPendingXcptGP(PVMCPU pVCpu, uint32_t u32ErrCode)
7766	{
7767	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_GP)
7768	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7769	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7770	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7771	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrCode, 0 / GCPtrFaultAddress */);
7772	}
7773
7774
7775	/**
7776	* Sets a stack (\#SS) exception as pending-for-injection into the VM.
7777	*
7778	* @param pVCpu The cross context virtual CPU structure.
7779	* @param u32ErrCode The error code for the stack exception.
7780	*/
7781	DECLINLINE(void) hmR0VmxSetPendingXcptSS(PVMCPU pVCpu, uint32_t u32ErrCode)
7782	{
7783	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_SS)
7784	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7785	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7786	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7787	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrCode, 0 / GCPtrFaultAddress */);
7788	}
7789	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
7790
7791
7792	/**
7793	* Fixes up attributes for the specified segment register.
7794	*
7795	* @param pVCpu The cross context virtual CPU structure.
7796	* @param pSelReg The segment register that needs fixing.
7797	* @param idxSel The VMCS field for the corresponding segment register.
7798	*/
7799	static void hmR0VmxFixUnusableSegRegAttr(PVMCPU pVCpu, PCPUMSELREG pSelReg, uint32_t idxSel)
7800	{
7801	Assert(pSelReg->Attr.u & X86DESCATTR_UNUSABLE);
7802
7803	/*
7804	* If VT-x marks the segment as unusable, most other bits remain undefined:
7805	* - For CS the L, D and G bits have meaning.
7806	* - For SS the DPL has meaning (it -is- the CPL for Intel and VBox).
7807	* - For the remaining data segments no bits are defined.
7808	*
7809	* The present bit and the unusable bit has been observed to be set at the
7810	* same time (the selector was supposed to be invalid as we started executing
7811	* a V8086 interrupt in ring-0).
7812	*
7813	* What should be important for the rest of the VBox code, is that the P bit is
7814	* cleared. Some of the other VBox code recognizes the unusable bit, but
7815	* AMD-V certainly don't, and REM doesn't really either. So, to be on the
7816	* safe side here, we'll strip off P and other bits we don't care about. If
7817	* any code breaks because Attr.u != 0 when Sel < 4, it should be fixed.
7818	*
7819	* See Intel spec. 27.3.2 "Saving Segment Registers and Descriptor-Table Registers".
7820	*/
7821	#ifdef VBOX_STRICT
7822	uint32_t const uAttr = pSelReg->Attr.u;
7823	#endif
7824
7825	/* Masking off: X86DESCATTR_P, X86DESCATTR_LIMIT_HIGH, and X86DESCATTR_AVL. The latter two are really irrelevant. */
7826	pSelReg->Attr.u &= X86DESCATTR_UNUSABLE \| X86DESCATTR_L \| X86DESCATTR_D \| X86DESCATTR_G
7827	\| X86DESCATTR_DPL \| X86DESCATTR_TYPE \| X86DESCATTR_DT;
7828
7829	#ifdef VBOX_STRICT
7830	VMMRZCallRing3Disable(pVCpu);
7831	Log4Func(("Unusable %#x: sel=%#x attr=%#x -> %#x\n", idxSel, pSelReg->Sel, uAttr, pSelReg->Attr.u));
7832	# ifdef DEBUG_bird
7833	AssertMsg((uAttr & ~X86DESCATTR_P) == pSelReg->Attr.u,
7834	("%#x: %#x != %#x (sel=%#x base=%#llx limit=%#x)\n",
7835	idxSel, uAttr, pSelReg->Attr.u, pSelReg->Sel, pSelReg->u64Base, pSelReg->u32Limit));
7836	# endif
7837	VMMRZCallRing3Enable(pVCpu);
7838	NOREF(uAttr);
7839	#endif
7840	RT_NOREF2(pVCpu, idxSel);
7841	}
7842
7843
7844	/**
7845	* Imports a guest segment register from the current VMCS into the guest-CPU
7846	* context.
7847	*
7848	* @returns VBox status code.
7849	* @param pVCpu The cross context virtual CPU structure.
7850	* @param iSegReg The segment register number (X86_SREG_XXX).
7851	*
7852	* @remarks Called with interrupts and/or preemption disabled, try not to assert and
7853	* do not log!
7854	*/
7855	static int hmR0VmxImportGuestSegReg(PVMCPU pVCpu, uint8_t iSegReg)
7856	{
7857	Assert(iSegReg < X86_SREG_COUNT);
7858
7859	uint32_t const idxSel = g_aVmcsSegSel[iSegReg];
7860	uint32_t const idxLimit = g_aVmcsSegLimit[iSegReg];
7861	uint32_t const idxAttr = g_aVmcsSegAttr[iSegReg];
7862	#ifdef VMX_USE_CACHED_VMCS_ACCESSES
7863	uint32_t const idxBase = g_aVmcsCacheSegBase[iSegReg];
7864	#else
7865	uint32_t const idxBase = g_aVmcsSegBase[iSegReg];
7866	#endif
7867	uint64_t u64Base;
7868	uint32_t u32Sel, u32Limit, u32Attr;
7869	int rc = VMXReadVmcs32(idxSel, &u32Sel);
7870	rc \|= VMXReadVmcs32(idxLimit, &u32Limit);
7871	rc \|= VMXReadVmcs32(idxAttr, &u32Attr);
7872	rc \|= VMXReadVmcsGstNByIdxVal(idxBase, &u64Base);
7873	if (RT_SUCCESS(rc))
7874	{
7875	PCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
7876	pSelReg->Sel = u32Sel;
7877	pSelReg->ValidSel = u32Sel;
7878	pSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
7879	pSelReg->u32Limit = u32Limit;
7880	pSelReg->u64Base = u64Base;
7881	pSelReg->Attr.u = u32Attr;
7882	if (u32Attr & X86DESCATTR_UNUSABLE)
7883	hmR0VmxFixUnusableSegRegAttr(pVCpu, pSelReg, idxSel);
7884	}
7885	return rc;
7886	}
7887
7888
7889	/**
7890	* Imports the guest LDTR from the current VMCS into the guest-CPU context.
7891	*
7892	* @returns VBox status code.
7893	* @param pVCpu The cross context virtual CPU structure.
7894	*
7895	* @remarks Called with interrupts and/or preemption disabled, try not to assert and
7896	* do not log!
7897	*/
7898	static int hmR0VmxImportGuestLdtr(PVMCPU pVCpu)
7899	{
7900	uint64_t u64Base;
7901	uint32_t u32Sel, u32Limit, u32Attr;
7902	int rc = VMXReadVmcs32(VMX_VMCS16_GUEST_LDTR_SEL, &u32Sel);
7903	rc \|= VMXReadVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, &u32Limit);
7904	rc \|= VMXReadVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, &u32Attr);
7905	rc \|= VMXReadVmcsGstN(VMX_VMCS_GUEST_LDTR_BASE, &u64Base);
7906	if (RT_SUCCESS(rc))
7907	{
7908	pVCpu->cpum.GstCtx.ldtr.Sel = u32Sel;
7909	pVCpu->cpum.GstCtx.ldtr.ValidSel = u32Sel;
7910	pVCpu->cpum.GstCtx.ldtr.fFlags = CPUMSELREG_FLAGS_VALID;
7911	pVCpu->cpum.GstCtx.ldtr.u32Limit = u32Limit;
7912	pVCpu->cpum.GstCtx.ldtr.u64Base = u64Base;
7913	pVCpu->cpum.GstCtx.ldtr.Attr.u = u32Attr;
7914	if (u32Attr & X86DESCATTR_UNUSABLE)
7915	hmR0VmxFixUnusableSegRegAttr(pVCpu, &pVCpu->cpum.GstCtx.ldtr, VMX_VMCS16_GUEST_LDTR_SEL);
7916	}
7917	return rc;
7918	}
7919
7920
7921	/**
7922	* Imports the guest TR from the current VMCS into the guest-CPU context.
7923	*
7924	* @returns VBox status code.
7925	* @param pVCpu The cross context virtual CPU structure.
7926	*
7927	* @remarks Called with interrupts and/or preemption disabled, try not to assert and
7928	* do not log!
7929	*/
7930	static int hmR0VmxImportGuestTr(PVMCPU pVCpu)
7931	{
7932	uint32_t u32Sel, u32Limit, u32Attr;
7933	uint64_t u64Base;
7934	int rc = VMXReadVmcs32(VMX_VMCS16_GUEST_TR_SEL, &u32Sel);
7935	rc \|= VMXReadVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, &u32Limit);
7936	rc \|= VMXReadVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, &u32Attr);
7937	rc \|= VMXReadVmcsGstN(VMX_VMCS_GUEST_TR_BASE, &u64Base);
7938	AssertRCReturn(rc, rc);
7939
7940	pVCpu->cpum.GstCtx.tr.Sel = u32Sel;
7941	pVCpu->cpum.GstCtx.tr.ValidSel = u32Sel;
7942	pVCpu->cpum.GstCtx.tr.fFlags = CPUMSELREG_FLAGS_VALID;
7943	pVCpu->cpum.GstCtx.tr.u32Limit = u32Limit;
7944	pVCpu->cpum.GstCtx.tr.u64Base = u64Base;
7945	pVCpu->cpum.GstCtx.tr.Attr.u = u32Attr;
7946	/* TR is the only selector that can never be unusable. */
7947	Assert(!(u32Attr & X86DESCATTR_UNUSABLE));
7948	return VINF_SUCCESS;
7949	}
7950
7951
7952	/**
7953	* Imports the guest RIP from the VMCS back into the guest-CPU context.
7954	*
7955	* @returns VBox status code.
7956	* @param pVCpu The cross context virtual CPU structure.
7957	*
7958	* @remarks Called with interrupts and/or preemption disabled, should not assert!
7959	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7960	* instead!!!
7961	*/
7962	static int hmR0VmxImportGuestRip(PVMCPU pVCpu)
7963	{
7964	uint64_t u64Val;
7965	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7966	if (pCtx->fExtrn & CPUMCTX_EXTRN_RIP)
7967	{
7968	int rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RIP, &u64Val);
7969	if (RT_SUCCESS(rc))
7970	{
7971	pCtx->rip = u64Val;
7972	EMR0HistoryUpdatePC(pVCpu, pCtx->rip, false);
7973	pCtx->fExtrn &= ~CPUMCTX_EXTRN_RIP;
7974	}
7975	return rc;
7976	}
7977	return VINF_SUCCESS;
7978	}
7979
7980
7981	/**
7982	* Imports the guest RFLAGS from the VMCS back into the guest-CPU context.
7983	*
7984	* @returns VBox status code.
7985	* @param pVCpu The cross context virtual CPU structure.
7986	* @param pVmcsInfo The VMCS info. object.
7987	*
7988	* @remarks Called with interrupts and/or preemption disabled, should not assert!
7989	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7990	* instead!!!
7991	*/
7992	static int hmR0VmxImportGuestRFlags(PVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo)
7993	{
7994	uint32_t u32Val;
7995	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7996	if (pCtx->fExtrn & CPUMCTX_EXTRN_RFLAGS)
7997	{
7998	int rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Val);
7999	if (RT_SUCCESS(rc))
8000	{
8001	pCtx->eflags.u32 = u32Val;
8002
8003	/* Restore eflags for real-on-v86-mode hack. */
8004	if (pVmcsInfo->RealMode.fRealOnV86Active)
8005	{
8006	pCtx->eflags.Bits.u1VM = 0;
8007	pCtx->eflags.Bits.u2IOPL = pVmcsInfo->RealMode.Eflags.Bits.u2IOPL;
8008	}
8009	}
8010	pCtx->fExtrn &= ~CPUMCTX_EXTRN_RFLAGS;
8011	return rc;
8012	}
8013	return VINF_SUCCESS;
8014	}
8015
8016
8017	/**
8018	* Imports the guest interruptibility-state from the VMCS back into the guest-CPU
8019	* context.
8020	*
8021	* @returns VBox status code.
8022	* @param pVCpu The cross context virtual CPU structure.
8023	* @param pVmcsInfo The VMCS info. object.
8024	*
8025	* @remarks Called with interrupts and/or preemption disabled, try not to assert and
8026	* do not log!
8027	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
8028	* instead!!!
8029	*/
8030	static int hmR0VmxImportGuestIntrState(PVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo)
8031	{
8032	uint32_t u32Val;
8033	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &u32Val);
8034	if (RT_SUCCESS(rc))
8035	{
8036	if (!u32Val)
8037	{
8038	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
8039	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
8040
8041	CPUMSetGuestNmiBlocking(pVCpu, false);
8042	}
8043	else
8044	{
8045	/*
8046	* We must import RIP here to set our EM interrupt-inhibited state.
8047	* We also import RFLAGS as our code that evaluates pending interrupts
8048	* before VM-entry requires it.
8049	*/
8050	rc = hmR0VmxImportGuestRip(pVCpu);
8051	rc \|= hmR0VmxImportGuestRFlags(pVCpu, pVmcsInfo);
8052	if (RT_SUCCESS(rc))
8053	{
8054	if (u32Val & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS \| VMX_VMCS_GUEST_INT_STATE_BLOCK_STI))
8055	EMSetInhibitInterruptsPC(pVCpu, pVCpu->cpum.GstCtx.rip);
8056	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
8057	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
8058
8059	bool const fNmiBlocking = RT_BOOL(u32Val & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI);
8060	CPUMSetGuestNmiBlocking(pVCpu, fNmiBlocking);
8061	}
8062	}
8063	}
8064	return rc;
8065	}
8066
8067
8068	/**
8069	* Worker for VMXR0ImportStateOnDemand.
8070	*
8071	* @returns VBox status code.
8072	* @param pVCpu The cross context virtual CPU structure.
8073	* @param pVmcsInfo The VMCS info. object.
8074	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
8075	*/
8076	static int hmR0VmxImportGuestState(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo, uint64_t fWhat)
8077	{
8078	#define VMXLOCAL_BREAK_RC(a_rc) \
8079	if (RT_SUCCESS(a_rc)) \
8080	{ } \
8081	else \
8082	break
8083
8084	int rc = VINF_SUCCESS;
8085	PVM pVM = pVCpu->CTX_SUFF(pVM);
8086	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8087	uint64_t u64Val;
8088	uint32_t u32Val;
8089
8090	/*
8091	* Note! This is hack to workaround a mysterious BSOD observed with release builds
8092	* on Windows 10 64-bit hosts. Profile and debug builds are not affected and
8093	* neither are other host platforms.
8094	*
8095	* Committing this temporarily as it prevents BSOD.
8096	*
8097	* Update: This is very likely a compiler optimization bug, see @bugref{9180}.
8098	*/
8099	#ifdef RT_OS_WINDOWS
8100	if (pVM == 0 \|\| pVM == (void *)(uintptr_t)-1)
8101	return VERR_HM_IPE_1;
8102	#endif
8103
8104	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatImportGuestState, x);
8105
8106	/*
8107	* We disable interrupts to make the updating of the state and in particular
8108	* the fExtrn modification atomic wrt to preemption hooks.
8109	*/
8110	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
8111
8112	fWhat &= pCtx->fExtrn;
8113	if (fWhat)
8114	{
8115	do
8116	{
8117	if (fWhat & CPUMCTX_EXTRN_RIP)
8118	{
8119	rc = hmR0VmxImportGuestRip(pVCpu);
8120	VMXLOCAL_BREAK_RC(rc);
8121	}
8122
8123	if (fWhat & CPUMCTX_EXTRN_RFLAGS)
8124	{
8125	rc = hmR0VmxImportGuestRFlags(pVCpu, pVmcsInfo);
8126	VMXLOCAL_BREAK_RC(rc);
8127	}
8128
8129	if (fWhat & CPUMCTX_EXTRN_HM_VMX_INT_STATE)
8130	{
8131	rc = hmR0VmxImportGuestIntrState(pVCpu, pVmcsInfo);
8132	VMXLOCAL_BREAK_RC(rc);
8133	}
8134
8135	if (fWhat & CPUMCTX_EXTRN_RSP)
8136	{
8137	rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RSP, &u64Val);
8138	VMXLOCAL_BREAK_RC(rc);
8139	pCtx->rsp = u64Val;
8140	}
8141
8142	if (fWhat & CPUMCTX_EXTRN_SREG_MASK)
8143	{
8144	bool const fRealOnV86Active = pVmcsInfo->RealMode.fRealOnV86Active;
8145	if (fWhat & CPUMCTX_EXTRN_CS)
8146	{
8147	rc \|= hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_CS);
8148	rc \|= hmR0VmxImportGuestRip(pVCpu);
8149	if (fRealOnV86Active)
8150	pCtx->cs.Attr.u = pVmcsInfo->RealMode.AttrCS.u;
8151	EMR0HistoryUpdatePC(pVCpu, pCtx->cs.u64Base + pCtx->rip, true /* fFlattened */);
8152	}
8153	if (fWhat & CPUMCTX_EXTRN_SS)
8154	{
8155	rc \|= hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_SS);
8156	if (fRealOnV86Active)
8157	pCtx->ss.Attr.u = pVmcsInfo->RealMode.AttrSS.u;
8158	}
8159	if (fWhat & CPUMCTX_EXTRN_DS)
8160	{
8161	rc \|= hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_DS);
8162	if (fRealOnV86Active)
8163	pCtx->ds.Attr.u = pVmcsInfo->RealMode.AttrDS.u;
8164	}
8165	if (fWhat & CPUMCTX_EXTRN_ES)
8166	{
8167	rc \|= hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_ES);
8168	if (fRealOnV86Active)
8169	pCtx->es.Attr.u = pVmcsInfo->RealMode.AttrES.u;
8170	}
8171	if (fWhat & CPUMCTX_EXTRN_FS)
8172	{
8173	rc \|= hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_FS);
8174	if (fRealOnV86Active)
8175	pCtx->fs.Attr.u = pVmcsInfo->RealMode.AttrFS.u;
8176	}
8177	if (fWhat & CPUMCTX_EXTRN_GS)
8178	{
8179	rc \|= hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_GS);
8180	if (fRealOnV86Active)
8181	pCtx->gs.Attr.u = pVmcsInfo->RealMode.AttrGS.u;
8182	}
8183	VMXLOCAL_BREAK_RC(rc);
8184	}
8185
8186	if (fWhat & CPUMCTX_EXTRN_TABLE_MASK)
8187	{
8188	if (fWhat & CPUMCTX_EXTRN_LDTR)
8189	rc \|= hmR0VmxImportGuestLdtr(pVCpu);
8190
8191	if (fWhat & CPUMCTX_EXTRN_GDTR)
8192	{
8193	rc \|= VMXReadVmcsGstN(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
8194	rc \|= VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val);
8195	pCtx->gdtr.pGdt = u64Val;
8196	pCtx->gdtr.cbGdt = u32Val;
8197	}
8198
8199	/* Guest IDTR. */
8200	if (fWhat & CPUMCTX_EXTRN_IDTR)
8201	{
8202	rc \|= VMXReadVmcsGstN(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
8203	rc \|= VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val);
8204	pCtx->idtr.pIdt = u64Val;
8205	pCtx->idtr.cbIdt = u32Val;
8206	}
8207
8208	/* Guest TR. */
8209	if (fWhat & CPUMCTX_EXTRN_TR)
8210	{
8211	/* Real-mode emulation using virtual-8086 mode has the fake TSS (pRealModeTSS) in TR,
8212	don't need to import that one. */
8213	if (!pVmcsInfo->RealMode.fRealOnV86Active)
8214	rc \|= hmR0VmxImportGuestTr(pVCpu);
8215	}
8216	VMXLOCAL_BREAK_RC(rc);
8217	}
8218
8219	if (fWhat & CPUMCTX_EXTRN_DR7)
8220	{
8221	if (!pVCpu->hm.s.fUsingHyperDR7)
8222	{
8223	/* Upper 32-bits are always zero. See Intel spec. 2.7.3 "Loading and Storing Debug Registers". */
8224	rc = VMXReadVmcs32(VMX_VMCS_GUEST_DR7, &u32Val);
8225	VMXLOCAL_BREAK_RC(rc);
8226	pCtx->dr[7] = u32Val;
8227	}
8228	}
8229
8230	if (fWhat & CPUMCTX_EXTRN_SYSENTER_MSRS)
8231	{
8232	rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_SYSENTER_EIP, &pCtx->SysEnter.eip);
8233	rc \|= VMXReadVmcsGstN(VMX_VMCS_GUEST_SYSENTER_ESP, &pCtx->SysEnter.esp);
8234	rc \|= VMXReadVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, &u32Val);
8235	pCtx->SysEnter.cs = u32Val;
8236	VMXLOCAL_BREAK_RC(rc);
8237	}
8238
8239	#if HC_ARCH_BITS == 64
8240	if (fWhat & CPUMCTX_EXTRN_KERNEL_GS_BASE)
8241	{
8242	if ( pVM->hm.s.fAllow64BitGuests
8243	&& (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
8244	pCtx->msrKERNELGSBASE = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
8245	}
8246
8247	if (fWhat & CPUMCTX_EXTRN_SYSCALL_MSRS)
8248	{
8249	if ( pVM->hm.s.fAllow64BitGuests
8250	&& (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
8251	{
8252	pCtx->msrLSTAR = ASMRdMsr(MSR_K8_LSTAR);
8253	pCtx->msrSTAR = ASMRdMsr(MSR_K6_STAR);
8254	pCtx->msrSFMASK = ASMRdMsr(MSR_K8_SF_MASK);
8255	}
8256	}
8257	#endif
8258
8259	if ( (fWhat & (CPUMCTX_EXTRN_TSC_AUX \| CPUMCTX_EXTRN_OTHER_MSRS))
8260	#if HC_ARCH_BITS == 32
8261	\|\| (fWhat & (CPUMCTX_EXTRN_KERNEL_GS_BASE \| CPUMCTX_EXTRN_SYSCALL_MSRS))
8262	#endif
8263	)
8264	{
8265	PCVMXAUTOMSR pMsrs = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
8266	uint32_t const cMsrs = pVmcsInfo->cExitMsrStore;
8267	Assert(pMsrs);
8268	Assert(cMsrs <= VMX_MISC_MAX_MSRS(pVM->hm.s.vmx.Msrs.u64Misc));
8269	Assert(sizeof(pMsrs) cMsrs <= X86_PAGE_4K_SIZE);
8270	for (uint32_t i = 0; i < cMsrs; i++)
8271	{
8272	uint32_t const idMsr = pMsrs[i].u32Msr;
8273	switch (idMsr)
8274	{
8275	case MSR_K8_TSC_AUX: CPUMSetGuestTscAux(pVCpu, pMsrs[i].u64Value); break;
8276	case MSR_IA32_SPEC_CTRL: CPUMSetGuestSpecCtrl(pVCpu, pMsrs[i].u64Value); break;
8277	case MSR_K6_EFER: /* Can't be changed without causing a VM-exit */ break;
8278	#if HC_ARCH_BITS == 32
8279	case MSR_K8_LSTAR: pCtx->msrLSTAR = pMsrs[i].u64Value; break;
8280	case MSR_K6_STAR: pCtx->msrSTAR = pMsrs[i].u64Value; break;
8281	case MSR_K8_SF_MASK: pCtx->msrSFMASK = pMsrs[i].u64Value; break;
8282	case MSR_K8_KERNEL_GS_BASE: pCtx->msrKERNELGSBASE = pMsrs[i].u64Value; break;
8283	#endif
8284	default:
8285	{
8286	pCtx->fExtrn = 0;
8287	pVCpu->hm.s.u32HMError = pMsrs->u32Msr;
8288	ASMSetFlags(fEFlags);
8289	AssertMsgFailed(("Unexpected MSR in auto-load/store area. idMsr=%#RX32 cMsrs=%u\n", idMsr, cMsrs));
8290	return VERR_HM_UNEXPECTED_LD_ST_MSR;
8291	}
8292	}
8293	}
8294	}
8295
8296	if (fWhat & CPUMCTX_EXTRN_CR_MASK)
8297	{
8298	uint64_t u64Shadow;
8299	if (fWhat & CPUMCTX_EXTRN_CR0)
8300	{
8301	/** @todo r=ramshankar: We only read 32-bits here for legacy/convenience reasons,
8302	* remove when we drop 32-bit host w/ 64-bit host support, see
8303	* @bugref{9180#c39}. */
8304	rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32Val);
8305	#if HC_ARCH_BITS == 32
8306	uint32_t u32Shadow;
8307	rc \|= VMXReadVmcs32(VMX_VMCS_CTRL_CR0_READ_SHADOW, &u32Shadow);
8308	u64Shadow = u32Shadow;
8309	#else
8310	rc \|= VMXReadVmcs64(VMX_VMCS_CTRL_CR0_READ_SHADOW, &u64Shadow);
8311	#endif
8312	VMXLOCAL_BREAK_RC(rc);
8313	u64Val = u32Val;
8314	u64Val = (u64Val & ~pVmcsInfo->u64Cr0Mask)
8315	\| (u64Shadow & pVmcsInfo->u64Cr0Mask);
8316	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8317	/*
8318	* Reapply the nested-guest's CR0 fixed bits that might have been altered while
8319	* exporting the nested-guest CR0 for executing using hardware-assisted VMX.
8320	*/
8321	if (CPUMIsGuestInVmxNonRootMode(pCtx))
8322	{
8323	u64Val \|= pCtx->hwvirt.vmx.Msrs.u64Cr0Fixed0;
8324	u64Val &= pCtx->hwvirt.vmx.Msrs.u64Cr0Fixed1;
8325	}
8326	#endif
8327	VMMRZCallRing3Disable(pVCpu); /* May call into PGM which has Log statements. */
8328	CPUMSetGuestCR0(pVCpu, u64Val);
8329	VMMRZCallRing3Enable(pVCpu);
8330	}
8331
8332	if (fWhat & CPUMCTX_EXTRN_CR4)
8333	{
8334	/** @todo r=ramshankar: We only read 32-bits here for legacy/convenience reasons,
8335	* remove when we drop 32-bit host w/ 64-bit host support, see
8336	* @bugref{9180#c39}. */
8337	rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR4, &u32Val);
8338	#if HC_ARCH_BITS == 32
8339	uint32_t u32Shadow;
8340	rc \|= VMXReadVmcs32(VMX_VMCS_CTRL_CR4_READ_SHADOW, &u32Shadow);
8341	u64Shadow = u32Shadow;
8342	#else
8343	rc \|= VMXReadVmcs64(VMX_VMCS_CTRL_CR4_READ_SHADOW, &u64Shadow);
8344	#endif
8345	VMXLOCAL_BREAK_RC(rc);
8346	u64Val = u32Val;
8347	u64Val = (u64Val & ~pVmcsInfo->u64Cr4Mask)
8348	\| (u64Shadow & pVmcsInfo->u64Cr4Mask);
8349	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8350	/*
8351	* Reapply the nested-guest's CR4 fixed bits that might have been altered while
8352	* exporting the nested-guest CR4 for executing using hardware-assisted VMX.
8353	*/
8354	if (CPUMIsGuestInVmxNonRootMode(pCtx))
8355	{
8356	u64Val \|= pCtx->hwvirt.vmx.Msrs.u64Cr4Fixed0;
8357	u64Val &= pCtx->hwvirt.vmx.Msrs.u64Cr4Fixed1;
8358	}
8359	#endif
8360	pCtx->cr4 = u64Val;
8361	}
8362
8363	if (fWhat & CPUMCTX_EXTRN_CR3)
8364	{
8365	/* CR0.PG bit changes are always intercepted, so it's up to date. */
8366	if ( pVM->hm.s.vmx.fUnrestrictedGuest
8367	\|\| ( pVM->hm.s.fNestedPaging
8368	&& CPUMIsGuestPagingEnabledEx(pCtx)))
8369	{
8370	rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_CR3, &u64Val);
8371	VMXLOCAL_BREAK_RC(rc);
8372	if (pCtx->cr3 != u64Val)
8373	{
8374	pCtx->cr3 = u64Val;
8375	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
8376	}
8377
8378	/* If the guest is in PAE mode, sync back the PDPE's into the guest state.
8379	Note: CR4.PAE, CR0.PG, EFER MSR changes are always intercepted, so they're up to date. */
8380	if (CPUMIsGuestInPAEModeEx(pCtx))
8381	{
8382	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &pVCpu->hm.s.aPdpes[0].u);
8383	rc \|= VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &pVCpu->hm.s.aPdpes[1].u);
8384	rc \|= VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &pVCpu->hm.s.aPdpes[2].u);
8385	rc \|= VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &pVCpu->hm.s.aPdpes[3].u);
8386	VMXLOCAL_BREAK_RC(rc);
8387	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES);
8388	}
8389	}
8390	}
8391	}
8392
8393	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8394	if (fWhat & CPUMCTX_EXTRN_HWVIRT)
8395	{
8396	if ( (pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
8397	&& !CPUMIsGuestInVmxNonRootMode(pCtx))
8398	{
8399	Assert(CPUMIsGuestInVmxRootMode(pCtx));
8400	rc = hmR0VmxCopyShadowToNstGstVmcs(pVCpu, pVmcsInfo);
8401	VMXLOCAL_BREAK_RC(rc);
8402	}
8403	}
8404	#endif
8405	} while (0);
8406
8407	if (RT_SUCCESS(rc))
8408	{
8409	/* Update fExtrn. */
8410	pCtx->fExtrn &= ~fWhat;
8411
8412	/* If everything has been imported, clear the HM keeper bit. */
8413	if (!(pCtx->fExtrn & HMVMX_CPUMCTX_EXTRN_ALL))
8414	{
8415	pCtx->fExtrn &= ~CPUMCTX_EXTRN_KEEPER_HM;
8416	Assert(!pCtx->fExtrn);
8417	}
8418	}
8419	}
8420	else
8421	AssertMsg(!pCtx->fExtrn \|\| (pCtx->fExtrn & HMVMX_CPUMCTX_EXTRN_ALL), ("%#RX64\n", pCtx->fExtrn));
8422
8423	ASMSetFlags(fEFlags);
8424
8425	STAM_PROFILE_ADV_STOP(& pVCpu->hm.s.StatImportGuestState, x);
8426
8427	if (RT_SUCCESS(rc))
8428	{ /* likely */ }
8429	else
8430	return rc;
8431
8432	/*
8433	* Honor any pending CR3 updates.
8434	*
8435	* Consider this scenario: VM-exit -> VMMRZCallRing3Enable() -> do stuff that causes a longjmp -> hmR0VmxCallRing3Callback()
8436	* -> VMMRZCallRing3Disable() -> hmR0VmxImportGuestState() -> Sets VMCPU_FF_HM_UPDATE_CR3 pending -> return from the longjmp
8437	* -> continue with VM-exit handling -> hmR0VmxImportGuestState() and here we are.
8438	*
8439	* The reason for such complicated handling is because VM-exits that call into PGM expect CR3 to be up-to-date and thus
8440	* if any CR3-saves -before- the VM-exit (longjmp) postponed the CR3 update via the force-flag, any VM-exit handler that
8441	* calls into PGM when it re-saves CR3 will end up here and we call PGMUpdateCR3(). This is why the code below should
8442	* -NOT- check if CPUMCTX_EXTRN_CR3 is set!
8443	*
8444	* The longjmp exit path can't check these CR3 force-flags and call code that takes a lock again. We cover for it here.
8445	*/
8446	if (VMMRZCallRing3IsEnabled(pVCpu))
8447	{
8448	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
8449	{
8450	Assert(!(ASMAtomicUoReadU64(&pCtx->fExtrn) & CPUMCTX_EXTRN_CR3));
8451	PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
8452	}
8453
8454	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
8455	PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
8456
8457	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
8458	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
8459	}
8460
8461	return VINF_SUCCESS;
8462	#undef VMXLOCAL_BREAK_RC
8463	}
8464
8465
8466	/**
8467	* Saves the guest state from the VMCS into the guest-CPU context.
8468	*
8469	* @returns VBox status code.
8470	* @param pVCpu The cross context virtual CPU structure.
8471	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
8472	*/
8473	VMMR0DECL(int) VMXR0ImportStateOnDemand(PVMCPU pVCpu, uint64_t fWhat)
8474	{
8475	AssertPtr(pVCpu);
8476	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8477	return hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fWhat);
8478	}
8479
8480
8481	/**
8482	* Check per-VM and per-VCPU force flag actions that require us to go back to
8483	* ring-3 for one reason or another.
8484	*
8485	* @returns Strict VBox status code (i.e. informational status codes too)
8486	* @retval VINF_SUCCESS if we don't have any actions that require going back to
8487	* ring-3.
8488	* @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync.
8489	* @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware
8490	* interrupts)
8491	* @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires
8492	* all EMTs to be in ring-3.
8493	* @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests.
8494	* @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return
8495	* to the EM loop.
8496	*
8497	* @param pVCpu The cross context virtual CPU structure.
8498	* @param fStepping Whether we are single-stepping the guest using the
8499	* hypervisor debugger.
8500	*/
8501	static VBOXSTRICTRC hmR0VmxCheckForceFlags(PVMCPU pVCpu, bool fStepping)
8502	{
8503	Assert(VMMRZCallRing3IsEnabled(pVCpu));
8504
8505	/*
8506	* Update pending interrupts into the APIC's IRR.
8507	*/
8508	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UPDATE_APIC))
8509	APICUpdatePendingInterrupts(pVCpu);
8510
8511	/*
8512	* Anything pending? Should be more likely than not if we're doing a good job.
8513	*/
8514	PVM pVM = pVCpu->CTX_SUFF(pVM);
8515	if ( !fStepping
8516	? !VM_FF_IS_ANY_SET(pVM, VM_FF_HP_R0_PRE_HM_MASK)
8517	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HP_R0_PRE_HM_MASK)
8518	: !VM_FF_IS_ANY_SET(pVM, VM_FF_HP_R0_PRE_HM_STEP_MASK)
8519	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) )
8520	return VINF_SUCCESS;
8521
8522	/* Pending PGM C3 sync. */
8523	if (VMCPU_FF_IS_ANY_SET(pVCpu,VMCPU_FF_PGM_SYNC_CR3 \| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL))
8524	{
8525	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8526	Assert(!(ASMAtomicUoReadU64(&pCtx->fExtrn) & (CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_CR3 \| CPUMCTX_EXTRN_CR4)));
8527	VBOXSTRICTRC rcStrict2 = PGMSyncCR3(pVCpu, pCtx->cr0, pCtx->cr3, pCtx->cr4,
8528	VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
8529	if (rcStrict2 != VINF_SUCCESS)
8530	{
8531	AssertRC(VBOXSTRICTRC_VAL(rcStrict2));
8532	Log4Func(("PGMSyncCR3 forcing us back to ring-3. rc2=%d\n", VBOXSTRICTRC_VAL(rcStrict2)));
8533	return rcStrict2;
8534	}
8535	}
8536
8537	/* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */
8538	if ( VM_FF_IS_ANY_SET(pVM, VM_FF_HM_TO_R3_MASK)
8539	\|\| VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
8540	{
8541	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
8542	int rc2 = RT_LIKELY(!VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_RAW_TO_R3 : VINF_EM_NO_MEMORY;
8543	Log4Func(("HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc2));
8544	return rc2;
8545	}
8546
8547	/* Pending VM request packets, such as hardware interrupts. */
8548	if ( VM_FF_IS_SET(pVM, VM_FF_REQUEST)
8549	\|\| VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_REQUEST))
8550	{
8551	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchVmReq);
8552	Log4Func(("Pending VM request forcing us back to ring-3\n"));
8553	return VINF_EM_PENDING_REQUEST;
8554	}
8555
8556	/* Pending PGM pool flushes. */
8557	if (VM_FF_IS_SET(pVM, VM_FF_PGM_POOL_FLUSH_PENDING))
8558	{
8559	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPgmPoolFlush);
8560	Log4Func(("PGM pool flush pending forcing us back to ring-3\n"));
8561	return VINF_PGM_POOL_FLUSH_PENDING;
8562	}
8563
8564	/* Pending DMA requests. */
8565	if (VM_FF_IS_SET(pVM, VM_FF_PDM_DMA))
8566	{
8567	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchDma);
8568	Log4Func(("Pending DMA request forcing us back to ring-3\n"));
8569	return VINF_EM_RAW_TO_R3;
8570	}
8571
8572	return VINF_SUCCESS;
8573	}
8574
8575
8576	/**
8577	* Converts any TRPM trap into a pending HM event. This is typically used when
8578	* entering from ring-3 (not longjmp returns).
8579	*
8580	* @param pVCpu The cross context virtual CPU structure.
8581	*/
8582	static void hmR0VmxTrpmTrapToPendingEvent(PVMCPU pVCpu)
8583	{
8584	Assert(TRPMHasTrap(pVCpu));
8585	Assert(!pVCpu->hm.s.Event.fPending);
8586
8587	uint8_t uVector;
8588	TRPMEVENT enmTrpmEvent;
8589	RTGCUINT uErrCode;
8590	RTGCUINTPTR GCPtrFaultAddress;
8591	uint8_t cbInstr;
8592
8593	int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr);
8594	AssertRC(rc);
8595
8596	uint32_t u32IntInfo;
8597	u32IntInfo = uVector \| VMX_IDT_VECTORING_INFO_VALID;
8598	u32IntInfo \|= HMTrpmEventTypeToVmxEventType(uVector, enmTrpmEvent);
8599
8600	rc = TRPMResetTrap(pVCpu);
8601	AssertRC(rc);
8602	Log4(("TRPM->HM event: u32IntInfo=%#RX32 enmTrpmEvent=%d cbInstr=%u uErrCode=%#RX32 GCPtrFaultAddress=%#RGv\n",
8603	u32IntInfo, enmTrpmEvent, cbInstr, uErrCode, GCPtrFaultAddress));
8604
8605	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, uErrCode, GCPtrFaultAddress);
8606	}
8607
8608
8609	/**
8610	* Converts the pending HM event into a TRPM trap.
8611	*
8612	* @param pVCpu The cross context virtual CPU structure.
8613	*/
8614	static void hmR0VmxPendingEventToTrpmTrap(PVMCPU pVCpu)
8615	{
8616	Assert(pVCpu->hm.s.Event.fPending);
8617
8618	/* If a trap was already pending, we did something wrong! */
8619	Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo /, NULL / pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP);
8620
8621	uint32_t const u32IntInfo = pVCpu->hm.s.Event.u64IntInfo;
8622	uint32_t const uVector = VMX_IDT_VECTORING_INFO_VECTOR(u32IntInfo);
8623	TRPMEVENT const enmTrapType = HMVmxEventTypeToTrpmEventType(u32IntInfo);
8624
8625	Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, enmTrapType));
8626
8627	int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType);
8628	AssertRC(rc);
8629
8630	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(u32IntInfo))
8631	TRPMSetErrorCode(pVCpu, pVCpu->hm.s.Event.u32ErrCode);
8632
8633	if (VMX_IDT_VECTORING_INFO_IS_XCPT_PF(u32IntInfo))
8634	TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress);
8635	else if (VMX_IDT_VECTORING_INFO_TYPE(u32IntInfo) == VMX_IDT_VECTORING_INFO_TYPE_SW_INT)
8636	TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr);
8637
8638	/* We're now done converting the pending event. */
8639	pVCpu->hm.s.Event.fPending = false;
8640	}
8641
8642
8643	/**
8644	* Sets the interrupt-window exiting control in the VMCS which instructs VT-x to
8645	* cause a VM-exit as soon as the guest is in a state to receive interrupts.
8646	*
8647	* @param pVCpu The cross context virtual CPU structure.
8648	* @param pVmcsInfo The VMCS info. object.
8649	*/
8650	static void hmR0VmxSetIntWindowExitVmcs(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
8651	{
8652	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_INT_WINDOW_EXIT)
8653	{
8654	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT))
8655	{
8656	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_INT_WINDOW_EXIT;
8657	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8658	AssertRC(rc);
8659	}
8660	} /* else we will deliver interrupts whenever the guest Vm-exits next and is in a state to receive the interrupt. */
8661	}
8662
8663
8664	/**
8665	* Clears the interrupt-window exiting control in the VMCS.
8666	*
8667	* @param pVmcsInfo The VMCS info. object.
8668	*/
8669	DECLINLINE(int) hmR0VmxClearIntWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8670	{
8671	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT)
8672	{
8673	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_INT_WINDOW_EXIT;
8674	return VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8675	}
8676	return VINF_SUCCESS;
8677	}
8678
8679
8680	/**
8681	* Sets the NMI-window exiting control in the VMCS which instructs VT-x to
8682	* cause a VM-exit as soon as the guest is in a state to receive NMIs.
8683	*
8684	* @param pVCpu The cross context virtual CPU structure.
8685	* @param pVmcsInfo The VMCS info. object.
8686	*/
8687	static void hmR0VmxSetNmiWindowExitVmcs(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo)
8688	{
8689	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
8690	{
8691	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))
8692	{
8693	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_NMI_WINDOW_EXIT;
8694	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8695	AssertRC(rc);
8696	Log4Func(("Setup NMI-window exiting\n"));
8697	}
8698	} /* else we will deliver NMIs whenever we VM-exit next, even possibly nesting NMIs. Can't be helped on ancient CPUs. */
8699	}
8700
8701
8702	/**
8703	* Clears the NMI-window exiting control in the VMCS.
8704	*
8705	* @param pVmcsInfo The VMCS info. object.
8706	*/
8707	DECLINLINE(int) hmR0VmxClearNmiWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8708	{
8709	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
8710	{
8711	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_NMI_WINDOW_EXIT;
8712	return VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8713	}
8714	return VINF_SUCCESS;
8715	}
8716
8717
8718	/**
8719	* Does the necessary state syncing before returning to ring-3 for any reason
8720	* (longjmp, preemption, voluntary exits to ring-3) from VT-x.
8721	*
8722	* @returns VBox status code.
8723	* @param pVCpu The cross context virtual CPU structure.
8724	* @param fImportState Whether to import the guest state from the VMCS back
8725	* to the guest-CPU context.
8726	*
8727	* @remarks No-long-jmp zone!!!
8728	*/
8729	static int hmR0VmxLeave(PVMCPU pVCpu, bool fImportState)
8730	{
8731	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8732	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8733
8734	RTCPUID const idCpu = RTMpCpuId();
8735	Log4Func(("HostCpuId=%u\n", idCpu));
8736
8737	/*
8738	* !!! IMPORTANT !!!
8739	* If you modify code here, check whether hmR0VmxCallRing3Callback() needs to be updated too.
8740	*/
8741
8742	/* Save the guest state if necessary. */
8743	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8744	if (fImportState)
8745	{
8746	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
8747	AssertRCReturn(rc, rc);
8748	}
8749
8750	/* Restore host FPU state if necessary. We will resync on next R0 reentry. */
8751	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
8752	Assert(!CPUMIsGuestFPUStateActive(pVCpu));
8753
8754	/* Restore host debug registers if necessary. We will resync on next R0 reentry. */
8755	#ifdef VBOX_STRICT
8756	if (CPUMIsHyperDebugStateActive(pVCpu))
8757	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MOV_DR_EXIT);
8758	#endif
8759	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
8760	Assert(!CPUMIsGuestDebugStateActive(pVCpu) && !CPUMIsGuestDebugStateActivePending(pVCpu));
8761	Assert(!CPUMIsHyperDebugStateActive(pVCpu) && !CPUMIsHyperDebugStateActivePending(pVCpu));
8762
8763	#if HC_ARCH_BITS == 64
8764	/* Restore host-state bits that VT-x only restores partially. */
8765	if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
8766	&& (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
8767	{
8768	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags, idCpu));
8769	VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
8770	}
8771	pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
8772	#endif
8773
8774	/* Restore the lazy host MSRs as we're leaving VT-x context. */
8775	if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
8776	{
8777	/* We shouldn't restore the host MSRs without saving the guest MSRs first. */
8778	if (!fImportState)
8779	{
8780	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_KERNEL_GS_BASE \| CPUMCTX_EXTRN_SYSCALL_MSRS);
8781	AssertRCReturn(rc, rc);
8782	}
8783	hmR0VmxLazyRestoreHostMsrs(pVCpu);
8784	Assert(!pVCpu->hm.s.vmx.fLazyMsrs);
8785	}
8786	else
8787	pVCpu->hm.s.vmx.fLazyMsrs = 0;
8788
8789	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
8790	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = false;
8791
8792	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry);
8793	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatImportGuestState);
8794	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExportGuestState);
8795	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatPreExit);
8796	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitHandling);
8797	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitIO);
8798	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitMovCRx);
8799	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitXcptNmi);
8800	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitVmentry);
8801	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
8802
8803	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
8804
8805	/** @todo This partially defeats the purpose of having preemption hooks.
8806	* The problem is, deregistering the hooks should be moved to a place that
8807	* lasts until the EMT is about to be destroyed not everytime while leaving HM
8808	* context.
8809	*/
8810	int rc = hmR0VmxClearVmcs(pVmcsInfo);
8811	AssertRCReturn(rc, rc);
8812
8813	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8814	/*
8815	* A valid shadow VMCS is made active as part of VM-entry. It is necessary to
8816	* clear a shadow VMCS before allowing that VMCS to become active on another
8817	* logical processor. We may or may not be importing guest state which clears
8818	* it, so cover for it here.
8819	*
8820	* See Intel spec. 24.11.1 "Software Use of Virtual-Machine Control Structures".
8821	*/
8822	if ( pVmcsInfo->pvShadowVmcs
8823	&& pVmcsInfo->fShadowVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
8824	{
8825	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
8826	AssertRCReturn(rc, rc);
8827	}
8828
8829	/*
8830	* Flag that we need to re-import the host state if we switch to this VMCS before
8831	* executing guest or nested-guest code.
8832	*/
8833	pVmcsInfo->idHostCpu = NIL_RTCPUID;
8834	#endif
8835
8836	Log4Func(("Cleared Vmcs. HostCpuId=%u\n", idCpu));
8837	NOREF(idCpu);
8838	return VINF_SUCCESS;
8839	}
8840
8841
8842	/**
8843	* Leaves the VT-x session.
8844	*
8845	* @returns VBox status code.
8846	* @param pVCpu The cross context virtual CPU structure.
8847	*
8848	* @remarks No-long-jmp zone!!!
8849	*/
8850	static int hmR0VmxLeaveSession(PVMCPU pVCpu)
8851	{
8852	HM_DISABLE_PREEMPT(pVCpu);
8853	HMVMX_ASSERT_CPU_SAFE(pVCpu);
8854	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8855	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8856
8857	/* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before
8858	and done this from the VMXR0ThreadCtxCallback(). */
8859	if (!pVCpu->hm.s.fLeaveDone)
8860	{
8861	int rc2 = hmR0VmxLeave(pVCpu, true /* fImportState */);
8862	AssertRCReturnStmt(rc2, HM_RESTORE_PREEMPT(), rc2);
8863	pVCpu->hm.s.fLeaveDone = true;
8864	}
8865	Assert(!pVCpu->cpum.GstCtx.fExtrn);
8866
8867	/*
8868	* !!! IMPORTANT !!!
8869	* If you modify code here, make sure to check whether hmR0VmxCallRing3Callback() needs to be updated too.
8870	*/
8871
8872	/* Deregister hook now that we've left HM context before re-enabling preemption. */
8873	/** @todo Deregistering here means we need to VMCLEAR always
8874	* (longjmp/exit-to-r3) in VT-x which is not efficient, eliminate need
8875	* for calling VMMR0ThreadCtxHookDisable here! */
8876	VMMR0ThreadCtxHookDisable(pVCpu);
8877
8878	/* Leave HM context. This takes care of local init (term). */
8879	int rc = HMR0LeaveCpu(pVCpu);
8880
8881	HM_RESTORE_PREEMPT();
8882	return rc;
8883	}
8884
8885
8886	/**
8887	* Does the necessary state syncing before doing a longjmp to ring-3.
8888	*
8889	* @returns VBox status code.
8890	* @param pVCpu The cross context virtual CPU structure.
8891	*
8892	* @remarks No-long-jmp zone!!!
8893	*/
8894	DECLINLINE(int) hmR0VmxLongJmpToRing3(PVMCPU pVCpu)
8895	{
8896	return hmR0VmxLeaveSession(pVCpu);
8897	}
8898
8899
8900	/**
8901	* Take necessary actions before going back to ring-3.
8902	*
8903	* An action requires us to go back to ring-3. This function does the necessary
8904	* steps before we can safely return to ring-3. This is not the same as longjmps
8905	* to ring-3, this is voluntary and prepares the guest so it may continue
8906	* executing outside HM (recompiler/IEM).
8907	*
8908	* @returns VBox status code.
8909	* @param pVCpu The cross context virtual CPU structure.
8910	* @param rcExit The reason for exiting to ring-3. Can be
8911	* VINF_VMM_UNKNOWN_RING3_CALL.
8912	*/
8913	static int hmR0VmxExitToRing3(PVMCPU pVCpu, VBOXSTRICTRC rcExit)
8914	{
8915	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
8916
8917	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8918	if (RT_UNLIKELY(rcExit == VERR_VMX_INVALID_VMCS_PTR))
8919	{
8920	VMXGetCurrentVmcs(&pVCpu->hm.s.vmx.LastError.HCPhysCurrentVmcs);
8921	pVCpu->hm.s.vmx.LastError.u32VmcsRev = (uint32_t )pVmcsInfo->pvVmcs;
8922	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
8923	/* LastError.idCurrentCpu was updated in hmR0VmxPreRunGuestCommitted(). */
8924	}
8925
8926	/* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */
8927	VMMRZCallRing3Disable(pVCpu);
8928	Log4Func(("rcExit=%d\n", VBOXSTRICTRC_VAL(rcExit)));
8929
8930	/*
8931	* Convert any pending HM events back to TRPM due to premature exits to ring-3.
8932	* We need to do this only on returns to ring-3 and not for longjmps to ring3.
8933	*
8934	* This is because execution may continue from ring-3 and we would need to inject
8935	* the event from there (hence place it back in TRPM).
8936	*/
8937	if (pVCpu->hm.s.Event.fPending)
8938	{
8939	hmR0VmxPendingEventToTrpmTrap(pVCpu);
8940	Assert(!pVCpu->hm.s.Event.fPending);
8941
8942	/* Clear the events from the VMCS. */
8943	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, 0);
8944	AssertRCReturn(rc, rc);
8945	}
8946	#ifdef VBOX_STRICT
8947	else
8948	{
8949	/*
8950	* Ensure we don't accidentally clear a pending HM event without clearing the VMCS.
8951	* This can be pretty hard to debug otherwise, interrupts might get injected twice
8952	* occasionally, see @bugref{9180#c42}.
8953	*
8954	* However, if the VM-entry failed, any VM entry-interruption info. field would
8955	* be left unmodified as the event would not have been injected to the guest. In
8956	* such cases, don't assert, we're not going to continue guest execution anyway.
8957	*/
8958	uint32_t uExitReason;
8959	uint32_t uEntryIntInfo;
8960	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
8961	rc \|= VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &uEntryIntInfo);
8962	AssertRC(rc);
8963	Assert(VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason) \|\| !VMX_ENTRY_INT_INFO_IS_VALID(uEntryIntInfo));
8964	}
8965	#endif
8966
8967	/*
8968	* Clear the interrupt-window and NMI-window VMCS controls as we could have got
8969	* a VM-exit with higher priority than interrupt-window or NMI-window VM-exits
8970	* (e.g. TPR below threshold).
8971	*/
8972	if (!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
8973	{
8974	int rc = hmR0VmxClearIntWindowExitVmcs(pVmcsInfo);
8975	rc \|= hmR0VmxClearNmiWindowExitVmcs(pVmcsInfo);
8976	AssertRCReturn(rc, rc);
8977	}
8978
8979	/* If we're emulating an instruction, we shouldn't have any TRPM traps pending
8980	and if we're injecting an event we should have a TRPM trap pending. */
8981	AssertMsg(rcExit != VINF_EM_RAW_INJECT_TRPM_EVENT \|\| TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
8982	#ifndef DEBUG_bird /* Triggered after firing an NMI against NT4SP1, possibly a triple fault in progress. */
8983	AssertMsg(rcExit != VINF_EM_RAW_EMULATE_INSTR \|\| !TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
8984	#endif
8985
8986	/* Save guest state and restore host state bits. */
8987	int rc = hmR0VmxLeaveSession(pVCpu);
8988	AssertRCReturn(rc, rc);
8989	STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
8990
8991	/* Thread-context hooks are unregistered at this point!!! */
8992
8993	/* Sync recompiler state. */
8994	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3);
8995	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR
8996	\| CPUM_CHANGED_LDTR
8997	\| CPUM_CHANGED_GDTR
8998	\| CPUM_CHANGED_IDTR
8999	\| CPUM_CHANGED_TR
9000	\| CPUM_CHANGED_HIDDEN_SEL_REGS);
9001	if ( pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging
9002	&& CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx))
9003	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH);
9004
9005	Assert(!pVCpu->hm.s.fClearTrapFlag);
9006
9007	/* Update the exit-to-ring 3 reason. */
9008	pVCpu->hm.s.rcLastExitToR3 = VBOXSTRICTRC_VAL(rcExit);
9009
9010	/* On our way back from ring-3 reload the guest state if there is a possibility of it being changed. */
9011	if ( rcExit != VINF_EM_RAW_INTERRUPT
9012	\|\| CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
9013	{
9014	Assert(!(pVCpu->cpum.GstCtx.fExtrn & HMVMX_CPUMCTX_EXTRN_ALL));
9015	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
9016	}
9017
9018	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3);
9019
9020	/* We do -not- want any longjmp notifications after this! We must return to ring-3 ASAP. */
9021	VMMRZCallRing3RemoveNotification(pVCpu);
9022	VMMRZCallRing3Enable(pVCpu);
9023
9024	return rc;
9025	}
9026
9027
9028	/**
9029	* VMMRZCallRing3() callback wrapper which saves the guest state before we
9030	* longjump to ring-3 and possibly get preempted.
9031	*
9032	* @returns VBox status code.
9033	* @param pVCpu The cross context virtual CPU structure.
9034	* @param enmOperation The operation causing the ring-3 longjump.
9035	* @param pvUser User argument, currently unused, NULL.
9036	*/
9037	static DECLCALLBACK(int) hmR0VmxCallRing3Callback(PVMCPU pVCpu, VMMCALLRING3 enmOperation, void *pvUser)
9038	{
9039	RT_NOREF(pvUser);
9040	if (enmOperation == VMMCALLRING3_VM_R0_ASSERTION)
9041	{
9042	/*
9043	* !!! IMPORTANT !!!
9044	* If you modify code here, check whether hmR0VmxLeave() and hmR0VmxLeaveSession() needs to be updated too.
9045	* This is a stripped down version which gets out ASAP, trying to not trigger any further assertions.
9046	*/
9047	VMMRZCallRing3RemoveNotification(pVCpu);
9048	VMMRZCallRing3Disable(pVCpu);
9049	RTTHREADPREEMPTSTATE PreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
9050	RTThreadPreemptDisable(&PreemptState);
9051
9052	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
9053	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
9054	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
9055	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
9056
9057	#if HC_ARCH_BITS == 64
9058	/* Restore host-state bits that VT-x only restores partially. */
9059	if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
9060	&& (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
9061	VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
9062	pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
9063	#endif
9064
9065	/* Restore the lazy host MSRs as we're leaving VT-x context. */
9066	if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
9067	hmR0VmxLazyRestoreHostMsrs(pVCpu);
9068
9069	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
9070	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = false;
9071	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
9072
9073	/* Clear the current VMCS data back to memory (shadow VMCS if any would have been
9074	cleared as part of importing the guest state above. */
9075	hmR0VmxClearVmcs(pVmcsInfo);
9076
9077	/** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
9078	VMMR0ThreadCtxHookDisable(pVCpu);
9079	HMR0LeaveCpu(pVCpu);
9080	RTThreadPreemptRestore(&PreemptState);
9081	return VINF_SUCCESS;
9082	}
9083
9084	Assert(pVCpu);
9085	Assert(pvUser);
9086	Assert(VMMRZCallRing3IsEnabled(pVCpu));
9087	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9088
9089	VMMRZCallRing3Disable(pVCpu);
9090	Assert(VMMR0IsLogFlushDisabled(pVCpu));
9091
9092	Log4Func((" -> hmR0VmxLongJmpToRing3 enmOperation=%d\n", enmOperation));
9093
9094	int rc = hmR0VmxLongJmpToRing3(pVCpu);
9095	AssertRCReturn(rc, rc);
9096
9097	VMMRZCallRing3Enable(pVCpu);
9098	return VINF_SUCCESS;
9099	}
9100
9101
9102	/**
9103	* Pushes a 2-byte value onto the real-mode (in virtual-8086 mode) guest's
9104	* stack.
9105	*
9106	* @returns Strict VBox status code (i.e. informational status codes too).
9107	* @retval VINF_EM_RESET if pushing a value to the stack caused a triple-fault.
9108	* @param pVCpu The cross context virtual CPU structure.
9109	* @param uValue The value to push to the guest stack.
9110	*/
9111	static VBOXSTRICTRC hmR0VmxRealModeGuestStackPush(PVMCPU pVCpu, uint16_t uValue)
9112	{
9113	/*
9114	* The stack limit is 0xffff in real-on-virtual 8086 mode. Real-mode with weird stack limits cannot be run in
9115	* virtual 8086 mode in VT-x. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
9116	* See Intel Instruction reference for PUSH and Intel spec. 22.33.1 "Segment Wraparound".
9117	*/
9118	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
9119	if (pCtx->sp == 1)
9120	return VINF_EM_RESET;
9121	pCtx->sp -= sizeof(uint16_t); /* May wrap around which is expected behaviour. */
9122	int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), pCtx->ss.u64Base + pCtx->sp, &uValue, sizeof(uint16_t));
9123	AssertRC(rc);
9124	return rc;
9125	}
9126
9127
9128	/**
9129	* Injects an event into the guest upon VM-entry by updating the relevant fields
9130	* in the VM-entry area in the VMCS.
9131	*
9132	* @returns Strict VBox status code (i.e. informational status codes too).
9133	* @retval VINF_SUCCESS if the event is successfully injected into the VMCS.
9134	* @retval VINF_EM_RESET if event injection resulted in a triple-fault.
9135	*
9136	* @param pVCpu The cross context virtual CPU structure.
9137	* @param pVmxTransient The VMX-transient structure.
9138	* @param pEvent The event being injected.
9139	* @param pfIntrState Pointer to the VT-x guest-interruptibility-state. This
9140	* will be updated if necessary. This cannot not be NULL.
9141	* @param fStepping Whether we're single-stepping guest execution and should
9142	* return VINF_EM_DBG_STEPPED if the event is injected
9143	* directly (registers modified by us, not by hardware on
9144	* VM-entry).
9145	*/
9146	static VBOXSTRICTRC hmR0VmxInjectEventVmcs(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, PCHMEVENT pEvent, bool fStepping,
9147	uint32_t *pfIntrState)
9148	{
9149	/* Intel spec. 24.8.3 "VM-Entry Controls for Event Injection" specifies the interruption-information field to be 32-bits. */
9150	AssertMsg(!RT_HI_U32(pEvent->u64IntInfo), ("%#RX64\n", pEvent->u64IntInfo));
9151	Assert(pfIntrState);
9152
9153	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
9154	uint32_t u32IntInfo = pEvent->u64IntInfo;
9155	uint32_t const u32ErrCode = pEvent->u32ErrCode;
9156	uint32_t const cbInstr = pEvent->cbInstr;
9157	RTGCUINTPTR const GCPtrFault = pEvent->GCPtrFaultAddress;
9158	uint8_t const uVector = VMX_ENTRY_INT_INFO_VECTOR(u32IntInfo);
9159	uint32_t const uIntType = VMX_ENTRY_INT_INFO_TYPE(u32IntInfo);
9160
9161	#ifdef VBOX_STRICT
9162	/*
9163	* Validate the error-code-valid bit for hardware exceptions.
9164	* No error codes for exceptions in real-mode.
9165	*
9166	* See Intel spec. 20.1.4 "Interrupt and Exception Handling"
9167	*/
9168	if ( uIntType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT
9169	&& !CPUMIsGuestInRealModeEx(pCtx))
9170	{
9171	switch (uVector)
9172	{
9173	case X86_XCPT_PF:
9174	case X86_XCPT_DF:
9175	case X86_XCPT_TS:
9176	case X86_XCPT_NP:
9177	case X86_XCPT_SS:
9178	case X86_XCPT_GP:
9179	case X86_XCPT_AC:
9180	AssertMsg(VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(u32IntInfo),
9181	("Error-code-valid bit not set for exception that has an error code uVector=%#x\n", uVector));
9182	RT_FALL_THRU();
9183	default:
9184	break;
9185	}
9186	}
9187
9188	/* Cannot inject an NMI when block-by-MOV SS is in effect. */
9189	Assert( uIntType != VMX_EXIT_INT_INFO_TYPE_NMI
9190	\|\| !(*pfIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
9191	#endif
9192
9193	STAM_COUNTER_INC(&pVCpu->hm.s.paStatInjectedIrqsR0[uVector & MASK_INJECT_IRQ_STAT]);
9194
9195	/*
9196	* Hardware interrupts & exceptions cannot be delivered through the software interrupt
9197	* redirection bitmap to the real mode task in virtual-8086 mode. We must jump to the
9198	* interrupt handler in the (real-mode) guest.
9199	*
9200	* See Intel spec. 20.3 "Interrupt and Exception handling in Virtual-8086 Mode".
9201	* See Intel spec. 20.1.4 "Interrupt and Exception Handling" for real-mode interrupt handling.
9202	*/
9203	if (CPUMIsGuestInRealModeEx(pCtx)) /* CR0.PE bit changes are always intercepted, so it's up to date. */
9204	{
9205	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest)
9206	{
9207	/*
9208	* For CPUs with unrestricted guest execution enabled and with the guest
9209	* in real-mode, we must not set the deliver-error-code bit.
9210	*
9211	* See Intel spec. 26.2.1.3 "VM-Entry Control Fields".
9212	*/
9213	u32IntInfo &= ~VMX_ENTRY_INT_INFO_ERROR_CODE_VALID;
9214	}
9215	else
9216	{
9217	PVM pVM = pVCpu->CTX_SUFF(pVM);
9218	Assert(PDMVmmDevHeapIsEnabled(pVM));
9219	Assert(pVM->hm.s.vmx.pRealModeTSS);
9220	Assert(!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
9221
9222	/* We require RIP, RSP, RFLAGS, CS, IDTR, import them. */
9223	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
9224	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_TABLE_MASK
9225	\| CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_RFLAGS);
9226	AssertRCReturn(rc2, rc2);
9227
9228	/* Check if the interrupt handler is present in the IVT (real-mode IDT). IDT limit is (4N - 1). */
9229	size_t const cbIdtEntry = sizeof(X86IDTR16);
9230	if (uVector * cbIdtEntry + (cbIdtEntry - 1) > pCtx->idtr.cbIdt)
9231	{
9232	/* If we are trying to inject a #DF with no valid IDT entry, return a triple-fault. */
9233	if (uVector == X86_XCPT_DF)
9234	return VINF_EM_RESET;
9235
9236	/* If we're injecting a #GP with no valid IDT entry, inject a double-fault.
9237	No error codes for exceptions in real-mode. */
9238	if (uVector == X86_XCPT_GP)
9239	{
9240	uint32_t const uXcptDfInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DF)
9241	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
9242	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
9243	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
9244	HMEVENT EventXcptDf;
9245	RT_ZERO(EventXcptDf);
9246	EventXcptDf.u64IntInfo = uXcptDfInfo;
9247	return hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &EventXcptDf, fStepping, pfIntrState);
9248	}
9249
9250	/*
9251	* If we're injecting an event with no valid IDT entry, inject a #GP.
9252	* No error codes for exceptions in real-mode.
9253	*
9254	* See Intel spec. 20.1.4 "Interrupt and Exception Handling"
9255	*/
9256	uint32_t const uXcptGpInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_GP)
9257	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
9258	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
9259	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
9260	HMEVENT EventXcptGp;
9261	RT_ZERO(EventXcptGp);
9262	EventXcptGp.u64IntInfo = uXcptGpInfo;
9263	return hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &EventXcptGp, fStepping, pfIntrState);
9264	}
9265
9266	/* Software exceptions (#BP and #OF exceptions thrown as a result of INT3 or INTO) */
9267	uint16_t uGuestIp = pCtx->ip;
9268	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_SW_XCPT)
9269	{
9270	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF);
9271	/* #BP and #OF are both benign traps, we need to resume the next instruction. */
9272	uGuestIp = pCtx->ip + (uint16_t)cbInstr;
9273	}
9274	else if (uIntType == VMX_ENTRY_INT_INFO_TYPE_SW_INT)
9275	uGuestIp = pCtx->ip + (uint16_t)cbInstr;
9276
9277	/* Get the code segment selector and offset from the IDT entry for the interrupt handler. */
9278	X86IDTR16 IdtEntry;
9279	RTGCPHYS const GCPhysIdtEntry = (RTGCPHYS)pCtx->idtr.pIdt + uVector * cbIdtEntry;
9280	rc2 = PGMPhysSimpleReadGCPhys(pVM, &IdtEntry, GCPhysIdtEntry, cbIdtEntry);
9281	AssertRCReturn(rc2, rc2);
9282
9283	/* Construct the stack frame for the interrupt/exception handler. */
9284	VBOXSTRICTRC rcStrict;
9285	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, pCtx->eflags.u32);
9286	if (rcStrict == VINF_SUCCESS)
9287	{
9288	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, pCtx->cs.Sel);
9289	if (rcStrict == VINF_SUCCESS)
9290	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, uGuestIp);
9291	}
9292
9293	/* Clear the required eflag bits and jump to the interrupt/exception handler. */
9294	if (rcStrict == VINF_SUCCESS)
9295	{
9296	pCtx->eflags.u32 &= ~(X86_EFL_IF \| X86_EFL_TF \| X86_EFL_RF \| X86_EFL_AC);
9297	pCtx->rip = IdtEntry.offSel;
9298	pCtx->cs.Sel = IdtEntry.uSel;
9299	pCtx->cs.ValidSel = IdtEntry.uSel;
9300	pCtx->cs.u64Base = IdtEntry.uSel << cbIdtEntry;
9301	if ( uIntType == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT
9302	&& uVector == X86_XCPT_PF)
9303	pCtx->cr2 = GCPtrFault;
9304
9305	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CS \| HM_CHANGED_GUEST_CR2
9306	\| HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
9307	\| HM_CHANGED_GUEST_RSP);
9308
9309	/*
9310	* If we delivered a hardware exception (other than an NMI) and if there was
9311	* block-by-STI in effect, we should clear it.
9312	*/
9313	if (*pfIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
9314	{
9315	Assert( uIntType != VMX_ENTRY_INT_INFO_TYPE_NMI
9316	&& uIntType != VMX_ENTRY_INT_INFO_TYPE_EXT_INT);
9317	Log4Func(("Clearing inhibition due to STI\n"));
9318	*pfIntrState &= ~VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
9319	}
9320
9321	Log4(("Injected real-mode: u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x Eflags=%#x CS:EIP=%04x:%04x\n",
9322	u32IntInfo, u32ErrCode, cbInstr, pCtx->eflags.u, pCtx->cs.Sel, pCtx->eip));
9323
9324	/*
9325	* The event has been truly dispatched to the guest. Mark it as no longer pending so
9326	* we don't attempt to undo it if we are returning to ring-3 before executing guest code.
9327	*/
9328	pVCpu->hm.s.Event.fPending = false;
9329
9330	/*
9331	* If we eventually support nested-guest execution without unrestricted guest execution,
9332	* we should set fInterceptEvents here.
9333	*/
9334	Assert(!pVmxTransient->fIsNestedGuest);
9335
9336	/* If we're stepping and we've changed cs:rip above, bail out of the VMX R0 execution loop. */
9337	if (fStepping)
9338	rcStrict = VINF_EM_DBG_STEPPED;
9339	}
9340	AssertMsg(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
9341	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
9342	return rcStrict;
9343	}
9344	}
9345
9346	/*
9347	* Validate.
9348	*/
9349	Assert(VMX_ENTRY_INT_INFO_IS_VALID(u32IntInfo)); /* Bit 31 (Valid bit) must be set by caller. */
9350	Assert(!(u32IntInfo & VMX_BF_ENTRY_INT_INFO_RSVD_12_30_MASK)); /* Bits 30:12 MBZ. */
9351
9352	/*
9353	* Inject the event into the VMCS.
9354	*/
9355	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, u32IntInfo);
9356	if (VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(u32IntInfo))
9357	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, u32ErrCode);
9358	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, cbInstr);
9359	AssertRCReturn(rc, rc);
9360
9361	/*
9362	* Update guest CR2 if this is a page-fault.
9363	*/
9364	if (VMX_ENTRY_INT_INFO_IS_XCPT_PF(u32IntInfo))
9365	pCtx->cr2 = GCPtrFault;
9366
9367	Log4(("Injecting u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x CR2=%#RX64\n", u32IntInfo, u32ErrCode, cbInstr, pCtx->cr2));
9368	return VINF_SUCCESS;
9369	}
9370
9371
9372	/**
9373	* Evaluates the event to be delivered to the guest and sets it as the pending
9374	* event.
9375	*
9376	* @returns Strict VBox status code (i.e. informational status codes too).
9377	* @param pVCpu The cross context virtual CPU structure.
9378	* @param pVmxTransient The VMX-transient structure.
9379	* @param pfIntrState Where to store the VT-x guest-interruptibility state.
9380	*/
9381	static VBOXSTRICTRC hmR0VmxEvaluatePendingEvent(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t *pfIntrState)
9382	{
9383	Assert(pfIntrState);
9384	Assert(!TRPMHasTrap(pVCpu));
9385
9386	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
9387	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
9388	bool const fIsNestedGuest = pVmxTransient->fIsNestedGuest;
9389
9390	/*
9391	* Get the current interruptibility-state of the guest or nested-guest and
9392	* then figure out what needs to be injected.
9393	*/
9394	uint32_t const fIntrState = hmR0VmxGetGuestIntrState(pVCpu, pVmxTransient);
9395	bool const fBlockMovSS = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
9396	bool const fBlockSti = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI);
9397	bool const fBlockNmi = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI);
9398
9399	/* We don't support block-by-SMI yet.*/
9400	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI));
9401
9402	/* Block-by-STI must not be set when interrupts are disabled. */
9403	if (fBlockSti)
9404	{
9405	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
9406	Assert(pCtx->eflags.Bits.u1IF);
9407	}
9408
9409	/* Update interruptibility state to the caller. */
9410	*pfIntrState = fIntrState;
9411
9412	/*
9413	* Toggling of interrupt force-flags here is safe since we update TRPM on
9414	* premature exits to ring-3 before executing guest code, see hmR0VmxExitToRing3().
9415	* We must NOT restore these force-flags.
9416	*/
9417
9418	/** @todo SMI. SMIs take priority over NMIs. */
9419
9420	/*
9421	* Check if an NMI is pending and if the guest or nested-guest can receive them.
9422	* NMIs take priority over external interrupts.
9423	*/
9424	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
9425	{
9426	/* On some CPUs block-by-STI also blocks NMIs. See Intel spec. 26.3.1.5 "Checks On Guest Non-Register State". */
9427	if ( !pVCpu->hm.s.Event.fPending
9428	&& !fBlockNmi
9429	&& !fBlockSti
9430	&& !fBlockMovSS)
9431	{
9432	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9433	if ( fIsNestedGuest
9434	&& CPUMIsGuestVmxPinCtlsSet(pVCpu, pCtx, VMX_PIN_CTLS_NMI_EXIT))
9435	return IEMExecVmxVmexitXcptNmi(pVCpu);
9436	#endif
9437	hmR0VmxSetPendingXcptNmi(pVCpu);
9438	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
9439	Log4Func(("Pending NMI\n"));
9440	}
9441	else if (!fIsNestedGuest)
9442	hmR0VmxSetNmiWindowExitVmcs(pVCpu, pVmcsInfo);
9443	/* else: for nested-guests, NMI-window exiting will be picked up when merging VMCS controls. */
9444	}
9445	/*
9446	* Check if an external interrupt (PIC/APIC) is pending and if the guest or nested-guest
9447	* can receive them. Once PDMGetInterrupt() returns a valid interrupt we -must- deliver
9448	* the interrupt. We can no longer re-request it from the APIC.
9449	*/
9450	else if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
9451	&& !pVCpu->hm.s.fSingleInstruction)
9452	{
9453	Assert(!DBGFIsStepping(pVCpu));
9454	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_RFLAGS);
9455	AssertRCReturn(rc, rc);
9456
9457	bool const fBlockInt = !(pCtx->eflags.u32 & X86_EFL_IF);
9458	if ( !pVCpu->hm.s.Event.fPending
9459	&& !fBlockInt
9460	&& !fBlockSti
9461	&& !fBlockMovSS)
9462	{
9463	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9464	if ( fIsNestedGuest
9465	&& CPUMIsGuestVmxPinCtlsSet(pVCpu, pCtx, VMX_PIN_CTLS_EXT_INT_EXIT)
9466	&& !CPUMIsGuestVmxExitCtlsSet(pVCpu, pCtx, VMX_EXIT_CTLS_ACK_EXT_INT))
9467	{
9468	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitExtInt(pVCpu, 0 /* uVector /, true / fIntPending */);
9469	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
9470	return rcStrict;
9471	}
9472	#endif
9473	uint8_t u8Interrupt;
9474	rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
9475	if (RT_SUCCESS(rc))
9476	{
9477	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9478	if ( fIsNestedGuest
9479	&& CPUMIsGuestVmxPinCtlsSet(pVCpu, pCtx, VMX_PIN_CTLS_EXT_INT_EXIT)
9480	&& CPUMIsGuestVmxExitCtlsSet(pVCpu, pCtx, VMX_EXIT_CTLS_ACK_EXT_INT))
9481	{
9482	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitExtInt(pVCpu, u8Interrupt, false /* fIntPending */);
9483	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
9484	return rcStrict;
9485	}
9486	#endif
9487	hmR0VmxSetPendingExtInt(pVCpu, u8Interrupt);
9488	Log4Func(("Pending external interrupt vector %#x\n", u8Interrupt));
9489	}
9490	else if (rc == VERR_APIC_INTR_MASKED_BY_TPR)
9491	{
9492	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchTprMaskedIrq);
9493
9494	if ( !fIsNestedGuest
9495	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW))
9496	hmR0VmxApicSetTprThreshold(pVCpu, pVmcsInfo, u8Interrupt >> 4);
9497	/* else: for nested-guests, TPR threshold is picked up while merging VMCS controls. */
9498
9499	/*
9500	* If the CPU doesn't have TPR shadowing, we will always get a VM-exit on TPR changes and
9501	* APICSetTpr() will end up setting the VMCPU_FF_INTERRUPT_APIC if required, so there is no
9502	* need to re-set this force-flag here.
9503	*/
9504	}
9505	else
9506	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
9507	}
9508	else if (!fIsNestedGuest)
9509	hmR0VmxSetIntWindowExitVmcs(pVCpu, pVmcsInfo);
9510	/* else: for nested-guests, interrupt-window exiting will be picked up when merging VMCS controls. */
9511	}
9512
9513	return VINF_SUCCESS;
9514	}
9515
9516
9517	/**
9518	* Injects any pending events into the guest if the guest is in a state to
9519	* receive them.
9520	*
9521	* @returns Strict VBox status code (i.e. informational status codes too).
9522	* @param pVCpu The cross context virtual CPU structure.
9523	* @param pVmxTransient The VMX-transient structure.
9524	* @param fIntrState The VT-x guest-interruptibility state.
9525	* @param fStepping Whether we are single-stepping the guest using the
9526	* hypervisor debugger and should return
9527	* VINF_EM_DBG_STEPPED if the event was dispatched
9528	* directly.
9529	*/
9530	static VBOXSTRICTRC hmR0VmxInjectPendingEvent(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t fIntrState, bool fStepping)
9531	{
9532	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9533	Assert(VMMRZCallRing3IsEnabled(pVCpu));
9534
9535	bool const fBlockMovSS = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
9536	bool const fBlockSti = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI);
9537
9538	Assert(!fBlockSti \|\| !(ASMAtomicUoReadU64(&pVCpu->cpum.GstCtx.fExtrn) & CPUMCTX_EXTRN_RFLAGS));
9539	Assert(!fBlockSti \|\| pVCpu->cpum.GstCtx.eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
9540	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
9541	Assert(!TRPMHasTrap(pVCpu));
9542
9543	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
9544	if (pVCpu->hm.s.Event.fPending)
9545	{
9546	/*
9547	* Do -not- clear any interrupt-window exiting control here. We might have an interrupt
9548	* pending even while injecting an event and in this case, we want a VM-exit as soon as
9549	* the guest is ready for the next interrupt, see @bugref{6208#c45}.
9550	*
9551	* See Intel spec. 26.6.5 "Interrupt-Window Exiting and Virtual-Interrupt Delivery".
9552	*/
9553	uint32_t const uIntType = VMX_ENTRY_INT_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
9554	#ifdef VBOX_STRICT
9555	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
9556	{
9557	bool const fBlockInt = !(pVCpu->cpum.GstCtx.eflags.u32 & X86_EFL_IF);
9558	Assert(!fBlockInt);
9559	Assert(!fBlockSti);
9560	Assert(!fBlockMovSS);
9561	}
9562	else if (uIntType == VMX_ENTRY_INT_INFO_TYPE_NMI)
9563	{
9564	bool const fBlockNmi = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI);
9565	Assert(!fBlockSti);
9566	Assert(!fBlockMovSS);
9567	Assert(!fBlockNmi);
9568	}
9569	#endif
9570	Log4(("Injecting pending event vcpu[%RU32] u64IntInfo=%#RX64 Type=%#RX32\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo,
9571	uIntType));
9572
9573	/*
9574	* Inject the event and get any changes to the guest-interruptibility state.
9575	*
9576	* The guest-interruptibility state may need to be updated if we inject the event
9577	* into the guest IDT ourselves (for real-on-v86 guest injecting software interrupts).
9578	*/
9579	rcStrict = hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &pVCpu->hm.s.Event, fStepping, &fIntrState);
9580	AssertRCReturn(VBOXSTRICTRC_VAL(rcStrict), rcStrict);
9581
9582	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
9583	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt);
9584	else
9585	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt);
9586	}
9587
9588	/*
9589	* Update the guest-interruptibility state.
9590	*
9591	* This is required for the real-on-v86 software interrupt injection case above, as well as
9592	* updates to the guest state from ring-3 or IEM/REM.
9593	*/
9594	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
9595	AssertRCReturn(rc, rc);
9596
9597	/*
9598	* There's no need to clear the VM-entry interruption-information field here if we're not
9599	* injecting anything. VT-x clears the valid bit on every VM-exit.
9600	*
9601	* See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
9602	*/
9603
9604	Assert(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping));
9605	NOREF(fBlockMovSS); NOREF(fBlockSti);
9606	return rcStrict;
9607	}
9608
9609
9610	/**
9611	* Enters the VT-x session.
9612	*
9613	* @returns VBox status code.
9614	* @param pVCpu The cross context virtual CPU structure.
9615	*/
9616	VMMR0DECL(int) VMXR0Enter(PVMCPU pVCpu)
9617	{
9618	AssertPtr(pVCpu);
9619	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fSupported);
9620	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9621
9622	LogFlowFunc(("pVCpu=%p\n", pVCpu));
9623	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
9624	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
9625
9626	#ifdef VBOX_STRICT
9627	/* At least verify VMX is enabled, since we can't check if we're in VMX root mode without #GP'ing. */
9628	RTCCUINTREG uHostCr4 = ASMGetCR4();
9629	if (!(uHostCr4 & X86_CR4_VMXE))
9630	{
9631	LogRelFunc(("X86_CR4_VMXE bit in CR4 is not set!\n"));
9632	return VERR_VMX_X86_CR4_VMXE_CLEARED;
9633	}
9634	#endif
9635
9636	/*
9637	* Load the appropriate VMCS as the current and active one.
9638	*/
9639	PVMXVMCSINFO pVmcsInfo;
9640	bool const fInNestedGuestMode = CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx);
9641	if (!fInNestedGuestMode)
9642	pVmcsInfo = &pVCpu->hm.s.vmx.VmcsInfo;
9643	else
9644	pVmcsInfo = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
9645	int rc = hmR0VmxLoadVmcs(pVmcsInfo);
9646	if (RT_SUCCESS(rc))
9647	{
9648	pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs = fInNestedGuestMode;
9649	pVCpu->hm.s.fLeaveDone = false;
9650	Log4Func(("Loaded Vmcs. HostCpuId=%u\n", RTMpCpuId()));
9651
9652	/*
9653	* Do the EMT scheduled L1D flush here if needed.
9654	*/
9655	if (pVCpu->CTX_SUFF(pVM)->hm.s.fL1dFlushOnSched)
9656	ASMWrMsr(MSR_IA32_FLUSH_CMD, MSR_IA32_FLUSH_CMD_F_L1D);
9657	else if (pVCpu->CTX_SUFF(pVM)->hm.s.fMdsClearOnSched)
9658	hmR0MdsClear();
9659	}
9660	return rc;
9661	}
9662
9663
9664	/**
9665	* The thread-context callback (only on platforms which support it).
9666	*
9667	* @param enmEvent The thread-context event.
9668	* @param pVCpu The cross context virtual CPU structure.
9669	* @param fGlobalInit Whether global VT-x/AMD-V init. was used.
9670	* @thread EMT(pVCpu)
9671	*/
9672	VMMR0DECL(void) VMXR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPU pVCpu, bool fGlobalInit)
9673	{
9674	AssertPtr(pVCpu);
9675	RT_NOREF1(fGlobalInit);
9676
9677	switch (enmEvent)
9678	{
9679	case RTTHREADCTXEVENT_OUT:
9680	{
9681	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9682	Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
9683	VMCPU_ASSERT_EMT(pVCpu);
9684
9685	/* No longjmps (logger flushes, locks) in this fragile context. */
9686	VMMRZCallRing3Disable(pVCpu);
9687	Log4Func(("Preempting: HostCpuId=%u\n", RTMpCpuId()));
9688
9689	/* Restore host-state (FPU, debug etc.) */
9690	if (!pVCpu->hm.s.fLeaveDone)
9691	{
9692	/*
9693	* Do -not- import the guest-state here as we might already be in the middle of importing
9694	* it, esp. bad if we're holding the PGM lock, see comment in hmR0VmxImportGuestState().
9695	*/
9696	hmR0VmxLeave(pVCpu, false /* fImportState */);
9697	pVCpu->hm.s.fLeaveDone = true;
9698	}
9699
9700	/* Leave HM context, takes care of local init (term). */
9701	int rc = HMR0LeaveCpu(pVCpu);
9702	AssertRC(rc);
9703
9704	/* Restore longjmp state. */
9705	VMMRZCallRing3Enable(pVCpu);
9706	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreempt);
9707	break;
9708	}
9709
9710	case RTTHREADCTXEVENT_IN:
9711	{
9712	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9713	Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
9714	VMCPU_ASSERT_EMT(pVCpu);
9715
9716	/* No longjmps here, as we don't want to trigger preemption (& its hook) while resuming. */
9717	VMMRZCallRing3Disable(pVCpu);
9718	Log4Func(("Resumed: HostCpuId=%u\n", RTMpCpuId()));
9719
9720	/* Initialize the bare minimum state required for HM. This takes care of
9721	initializing VT-x if necessary (onlined CPUs, local init etc.) */
9722	int rc = hmR0EnterCpu(pVCpu);
9723	AssertRC(rc);
9724	Assert( (pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
9725	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
9726
9727	/* Load the active VMCS as the current one. */
9728	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
9729	rc = hmR0VmxLoadVmcs(pVmcsInfo);
9730	AssertRC(rc);
9731	Log4Func(("Resumed: Loaded Vmcs. HostCpuId=%u\n", RTMpCpuId()));
9732	pVCpu->hm.s.fLeaveDone = false;
9733
9734	/* Do the EMT scheduled L1D flush if needed. */
9735	if (pVCpu->CTX_SUFF(pVM)->hm.s.fL1dFlushOnSched)
9736	ASMWrMsr(MSR_IA32_FLUSH_CMD, MSR_IA32_FLUSH_CMD_F_L1D);
9737
9738	/* Restore longjmp state. */
9739	VMMRZCallRing3Enable(pVCpu);
9740	break;
9741	}
9742
9743	default:
9744	break;
9745	}
9746	}
9747
9748
9749	/**
9750	* Exports the host state into the VMCS host-state area.
9751	* Sets up the VM-exit MSR-load area.
9752	*
9753	* The CPU state will be loaded from these fields on every successful VM-exit.
9754	*
9755	* @returns VBox status code.
9756	* @param pVCpu The cross context virtual CPU structure.
9757	*
9758	* @remarks No-long-jump zone!!!
9759	*/
9760	static int hmR0VmxExportHostState(PVMCPU pVCpu)
9761	{
9762	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9763
9764	int rc = VINF_SUCCESS;
9765	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT)
9766	{
9767	rc = hmR0VmxExportHostControlRegs();
9768	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9769
9770	rc = hmR0VmxExportHostSegmentRegs(pVCpu);
9771	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9772
9773	rc = hmR0VmxExportHostMsrs(pVCpu);
9774	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9775
9776	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_HOST_CONTEXT;
9777	}
9778	return rc;
9779	}
9780
9781
9782	/**
9783	* Saves the host state in the VMCS host-state.
9784	*
9785	* @returns VBox status code.
9786	* @param pVCpu The cross context virtual CPU structure.
9787	*
9788	* @remarks No-long-jump zone!!!
9789	*/
9790	VMMR0DECL(int) VMXR0ExportHostState(PVMCPU pVCpu)
9791	{
9792	AssertPtr(pVCpu);
9793	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9794
9795	/*
9796	* Export the host state here while entering HM context.
9797	* When thread-context hooks are used, we might get preempted and have to re-save the host
9798	* state but most of the time we won't be, so do it here before we disable interrupts.
9799	*/
9800	return hmR0VmxExportHostState(pVCpu);
9801	}
9802
9803
9804	/**
9805	* Exports the guest state into the VMCS guest-state area.
9806	*
9807	* The will typically be done before VM-entry when the guest-CPU state and the
9808	* VMCS state may potentially be out of sync.
9809	*
9810	* Sets up the VM-entry MSR-load and VM-exit MSR-store areas. Sets up the
9811	* VM-entry controls.
9812	* Sets up the appropriate VMX non-root function to execute guest code based on
9813	* the guest CPU mode.
9814	*
9815	* @returns VBox strict status code.
9816	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
9817	* without unrestricted guest execution and the VMMDev is not presently
9818	* mapped (e.g. EFI32).
9819	*
9820	* @param pVCpu The cross context virtual CPU structure.
9821	* @param pVmxTransient The VMX-transient structure.
9822	*
9823	* @remarks No-long-jump zone!!!
9824	*/
9825	static VBOXSTRICTRC hmR0VmxExportGuestState(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
9826	{
9827	AssertPtr(pVCpu);
9828	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9829	LogFlowFunc(("pVCpu=%p\n", pVCpu));
9830
9831	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExportGuestState, x);
9832
9833	/*
9834	* Determine real-on-v86 mode.
9835	* Used when the guest is in real-mode and unrestricted guest execution is not used.
9836	*/
9837	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
9838	if ( pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest
9839	\|\| !CPUMIsGuestInRealModeEx(&pVCpu->cpum.GstCtx))
9840	pVmcsInfo->RealMode. fRealOnV86Active = false;
9841	else
9842	{
9843	Assert(!pVmxTransient->fIsNestedGuest);
9844	pVmcsInfo->RealMode.fRealOnV86Active = true;
9845	}
9846
9847	/*
9848	* Any ordering dependency among the sub-functions below must be explicitly stated using comments.
9849	* Ideally, assert that the cross-dependent bits are up-to-date at the point of using it.
9850	*/
9851	/** @todo r=ramshankar: Move hmR0VmxSelectVMRunHandler inside
9852	* hmR0VmxExportGuestEntryExitCtls and do it conditionally. There shouldn't
9853	* be a need to evaluate this everytime since I'm pretty sure we intercept
9854	* all guest paging mode changes. */
9855	int rc = hmR0VmxSelectVMRunHandler(pVCpu, pVmxTransient);
9856	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9857
9858	rc = hmR0VmxExportGuestEntryExitCtls(pVCpu, pVmxTransient);
9859	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9860
9861	rc = hmR0VmxExportGuestCR0(pVCpu, pVmxTransient);
9862	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9863
9864	VBOXSTRICTRC rcStrict = hmR0VmxExportGuestCR3AndCR4(pVCpu, pVmxTransient);
9865	if (rcStrict == VINF_SUCCESS)
9866	{ /* likely */ }
9867	else
9868	{
9869	Assert(rcStrict == VINF_EM_RESCHEDULE_REM \|\| RT_FAILURE_NP(rcStrict));
9870	return rcStrict;
9871	}
9872
9873	rc = hmR0VmxExportGuestSegRegsXdtr(pVCpu, pVmxTransient);
9874	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9875
9876	rc = hmR0VmxExportGuestMsrs(pVCpu, pVmxTransient);
9877	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9878
9879	rc = hmR0VmxExportGuestApicTpr(pVCpu, pVmxTransient);
9880	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9881
9882	rc = hmR0VmxExportGuestXcptIntercepts(pVCpu, pVmxTransient);
9883	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9884
9885	rc = hmR0VmxExportGuestRip(pVCpu);
9886	rc \|= hmR0VmxExportGuestRsp(pVCpu);
9887	rc \|= hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9888	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9889
9890	rc = hmR0VmxExportGuestHwvirtState(pVCpu, pVmxTransient);
9891	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9892
9893	/* Clear any bits that may be set but exported unconditionally or unused/reserved bits. */
9894	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~( (HM_CHANGED_GUEST_GPRS_MASK & ~HM_CHANGED_GUEST_RSP)
9895	\| HM_CHANGED_GUEST_CR2
9896	\| (HM_CHANGED_GUEST_DR_MASK & ~HM_CHANGED_GUEST_DR7)
9897	\| HM_CHANGED_GUEST_X87
9898	\| HM_CHANGED_GUEST_SSE_AVX
9899	\| HM_CHANGED_GUEST_OTHER_XSAVE
9900	\| HM_CHANGED_GUEST_XCRx
9901	\| HM_CHANGED_GUEST_KERNEL_GS_BASE /* Part of lazy or auto load-store MSRs. */
9902	\| HM_CHANGED_GUEST_SYSCALL_MSRS /* Part of lazy or auto load-store MSRs. */
9903	\| HM_CHANGED_GUEST_TSC_AUX
9904	\| HM_CHANGED_GUEST_OTHER_MSRS
9905	\| (HM_CHANGED_KEEPER_STATE_MASK & ~HM_CHANGED_VMX_MASK)));
9906
9907	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExportGuestState, x);
9908	return rc;
9909	}
9910
9911
9912	/**
9913	* Exports the state shared between the host and guest into the VMCS.
9914	*
9915	* @param pVCpu The cross context virtual CPU structure.
9916	* @param pVmxTransient The VMX-transient structure.
9917	*
9918	* @remarks No-long-jump zone!!!
9919	*/
9920	static void hmR0VmxExportSharedState(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
9921	{
9922	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9923	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9924
9925	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_DR_MASK)
9926	{
9927	int rc = hmR0VmxExportSharedDebugState(pVCpu, pVmxTransient);
9928	AssertRC(rc);
9929	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_GUEST_DR_MASK;
9930
9931	/* Loading shared debug bits might have changed eflags.TF bit for debugging purposes. */
9932	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_RFLAGS)
9933	{
9934	rc = hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9935	AssertRC(rc);
9936	}
9937	}
9938
9939	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_GUEST_LAZY_MSRS)
9940	{
9941	hmR0VmxLazyLoadGuestMsrs(pVCpu);
9942	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_VMX_GUEST_LAZY_MSRS;
9943	}
9944
9945	AssertMsg(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE),
9946	("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
9947	}
9948
9949
9950	/**
9951	* Worker for loading the guest-state bits in the inner VT-x execution loop.
9952	*
9953	* @returns Strict VBox status code (i.e. informational status codes too).
9954	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
9955	* without unrestricted guest execution and the VMMDev is not presently
9956	* mapped (e.g. EFI32).
9957	*
9958	* @param pVCpu The cross context virtual CPU structure.
9959	* @param pVmxTransient The VMX-transient structure.
9960	*
9961	* @remarks No-long-jump zone!!!
9962	*/
9963	static VBOXSTRICTRC hmR0VmxExportGuestStateOptimal(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
9964	{
9965	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9966	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9967	Assert(VMMR0IsLogFlushDisabled(pVCpu));
9968
9969	#ifdef HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
9970	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
9971	#endif
9972
9973	/*
9974	* For many exits it's only RIP that changes and hence try to export it first
9975	* without going through a lot of change flag checks.
9976	*/
9977	VBOXSTRICTRC rcStrict;
9978	uint64_t fCtxChanged = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
9979	RT_UNTRUSTED_NONVOLATILE_COPY_FENCE();
9980	if ((fCtxChanged & (HM_CHANGED_ALL_GUEST & ~HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE)) == HM_CHANGED_GUEST_RIP)
9981	{
9982	rcStrict = hmR0VmxExportGuestRip(pVCpu);
9983	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
9984	{ /* likely */}
9985	else
9986	AssertMsgFailedReturn(("Failed to export guest RIP! rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)), rcStrict);
9987	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportMinimal);
9988	}
9989	else if (fCtxChanged & (HM_CHANGED_ALL_GUEST & ~HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
9990	{
9991	rcStrict = hmR0VmxExportGuestState(pVCpu, pVmxTransient);
9992	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
9993	{ /* likely */}
9994	else
9995	{
9996	AssertMsg(rcStrict == VINF_EM_RESCHEDULE_REM, ("Failed to export guest state! rc=%Rrc\n",
9997	VBOXSTRICTRC_VAL(rcStrict)));
9998	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9999	return rcStrict;
10000	}
10001	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportFull);
10002	}
10003	else
10004	rcStrict = VINF_SUCCESS;
10005
10006	#ifdef VBOX_STRICT
10007	/* All the guest state bits should be loaded except maybe the host context and/or the shared host/guest bits. */
10008	fCtxChanged = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
10009	RT_UNTRUSTED_NONVOLATILE_COPY_FENCE();
10010	AssertMsg(!(fCtxChanged & (HM_CHANGED_ALL_GUEST & ~HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE)),
10011	("fCtxChanged=%#RX64\n", fCtxChanged));
10012	#endif
10013	return rcStrict;
10014	}
10015
10016
10017	/**
10018	* Tries to determine what part of the guest-state VT-x has deemed as invalid
10019	* and update error record fields accordingly.
10020	*
10021	* @returns VMX_IGS_* error codes.
10022	* @retval VMX_IGS_REASON_NOT_FOUND if this function could not find anything
10023	* wrong with the guest state.
10024	*
10025	* @param pVCpu The cross context virtual CPU structure.
10026	* @param pVmcsInfo The VMCS info. object.
10027	*
10028	* @remarks This function assumes our cache of the VMCS controls
10029	* are valid, i.e. hmR0VmxCheckVmcsCtls() succeeded.
10030	*/
10031	static uint32_t hmR0VmxCheckGuestState(PVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo)
10032	{
10033	#define HMVMX_ERROR_BREAK(err) { uError = (err); break; }
10034	#define HMVMX_CHECK_BREAK(expr, err) do { \
10035	if (!(expr)) { uError = (err); break; } \
10036	} while (0)
10037
10038	int rc;
10039	PVM pVM = pVCpu->CTX_SUFF(pVM);
10040	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
10041	uint32_t uError = VMX_IGS_ERROR;
10042	uint32_t u32Val;
10043	bool const fUnrestrictedGuest = pVM->hm.s.vmx.fUnrestrictedGuest;
10044
10045	do
10046	{
10047	/*
10048	* CR0.
10049	*/
10050	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
10051	uint32_t fSetCr0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
10052	uint32_t const fZapCr0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 \| pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
10053	/* Exceptions for unrestricted guest execution for fixed CR0 bits (PE, PG).
10054	See Intel spec. 26.3.1 "Checks on Guest Control Registers, Debug Registers and MSRs." */
10055	if (fUnrestrictedGuest)
10056	fSetCr0 &= ~(X86_CR0_PE \| X86_CR0_PG);
10057
10058	uint32_t u32GuestCr0;
10059	rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32GuestCr0);
10060	AssertRCBreak(rc);
10061	HMVMX_CHECK_BREAK((u32GuestCr0 & fSetCr0) == fSetCr0, VMX_IGS_CR0_FIXED1);
10062	HMVMX_CHECK_BREAK(!(u32GuestCr0 & ~fZapCr0), VMX_IGS_CR0_FIXED0);
10063	if ( !fUnrestrictedGuest
10064	&& (u32GuestCr0 & X86_CR0_PG)
10065	&& !(u32GuestCr0 & X86_CR0_PE))
10066	{
10067	HMVMX_ERROR_BREAK(VMX_IGS_CR0_PG_PE_COMBO);
10068	}
10069
10070	/*
10071	* CR4.
10072	*/
10073	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
10074	uint64_t const fSetCr4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
10075	uint64_t const fZapCr4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 \| pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
10076
10077	uint32_t u32GuestCr4;
10078	rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR4, &u32GuestCr4);
10079	AssertRCBreak(rc);
10080	HMVMX_CHECK_BREAK((u32GuestCr4 & fSetCr4) == fSetCr4, VMX_IGS_CR4_FIXED1);
10081	HMVMX_CHECK_BREAK(!(u32GuestCr4 & ~fZapCr4), VMX_IGS_CR4_FIXED0);
10082
10083	/*
10084	* IA32_DEBUGCTL MSR.
10085	*/
10086	uint64_t u64Val;
10087	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, &u64Val);
10088	AssertRCBreak(rc);
10089	if ( (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
10090	&& (u64Val & 0xfffffe3c)) /* Bits 31:9, bits 5:2 MBZ. */
10091	{
10092	HMVMX_ERROR_BREAK(VMX_IGS_DEBUGCTL_MSR_RESERVED);
10093	}
10094	uint64_t u64DebugCtlMsr = u64Val;
10095
10096	#ifdef VBOX_STRICT
10097	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
10098	AssertRCBreak(rc);
10099	Assert(u32Val == pVmcsInfo->u32EntryCtls);
10100	#endif
10101	bool const fLongModeGuest = RT_BOOL(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
10102
10103	/*
10104	* RIP and RFLAGS.
10105	*/
10106	uint32_t u32Eflags;
10107	#if HC_ARCH_BITS == 64
10108	rc = VMXReadVmcs64(VMX_VMCS_GUEST_RIP, &u64Val);
10109	AssertRCBreak(rc);
10110	/* pCtx->rip can be different than the one in the VMCS (e.g. run guest code and VM-exits that don't update it). */
10111	if ( !fLongModeGuest
10112	\|\| !pCtx->cs.Attr.n.u1Long)
10113	{
10114	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffff00000000)), VMX_IGS_LONGMODE_RIP_INVALID);
10115	}
10116	/** @todo If the processor supports N < 64 linear-address bits, bits 63:N
10117	* must be identical if the "IA-32e mode guest" VM-entry
10118	* control is 1 and CS.L is 1. No check applies if the
10119	* CPU supports 64 linear-address bits. */
10120
10121	/* Flags in pCtx can be different (real-on-v86 for instance). We are only concerned about the VMCS contents here. */
10122	rc = VMXReadVmcs64(VMX_VMCS_GUEST_RFLAGS, &u64Val);
10123	AssertRCBreak(rc);
10124	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffc08028)), /* Bit 63:22, Bit 15, 5, 3 MBZ. */
10125	VMX_IGS_RFLAGS_RESERVED);
10126	HMVMX_CHECK_BREAK((u64Val & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
10127	u32Eflags = u64Val;
10128	#else
10129	rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Eflags);
10130	AssertRCBreak(rc);
10131	HMVMX_CHECK_BREAK(!(u32Eflags & 0xffc08028), VMX_IGS_RFLAGS_RESERVED); /* Bit 31:22, Bit 15, 5, 3 MBZ. */
10132	HMVMX_CHECK_BREAK((u32Eflags & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
10133	#endif
10134
10135	if ( fLongModeGuest
10136	\|\| ( fUnrestrictedGuest
10137	&& !(u32GuestCr0 & X86_CR0_PE)))
10138	{
10139	HMVMX_CHECK_BREAK(!(u32Eflags & X86_EFL_VM), VMX_IGS_RFLAGS_VM_INVALID);
10140	}
10141
10142	uint32_t u32EntryInfo;
10143	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo);
10144	AssertRCBreak(rc);
10145	if (VMX_ENTRY_INT_INFO_IS_EXT_INT(u32EntryInfo))
10146	HMVMX_CHECK_BREAK(u32Eflags & X86_EFL_IF, VMX_IGS_RFLAGS_IF_INVALID);
10147
10148	/*
10149	* 64-bit checks.
10150	*/
10151	#if HC_ARCH_BITS == 64
10152	if (fLongModeGuest)
10153	{
10154	HMVMX_CHECK_BREAK(u32GuestCr0 & X86_CR0_PG, VMX_IGS_CR0_PG_LONGMODE);
10155	HMVMX_CHECK_BREAK(u32GuestCr4 & X86_CR4_PAE, VMX_IGS_CR4_PAE_LONGMODE);
10156	}
10157
10158	if ( !fLongModeGuest
10159	&& (u32GuestCr4 & X86_CR4_PCIDE))
10160	{
10161	HMVMX_ERROR_BREAK(VMX_IGS_CR4_PCIDE);
10162	}
10163
10164	/** @todo CR3 field must be such that bits 63:52 and bits in the range
10165	* 51:32 beyond the processor's physical-address width are 0. */
10166
10167	if ( (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
10168	&& (pCtx->dr[7] & X86_DR7_MBZ_MASK))
10169	{
10170	HMVMX_ERROR_BREAK(VMX_IGS_DR7_RESERVED);
10171	}
10172
10173	rc = VMXReadVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, &u64Val);
10174	AssertRCBreak(rc);
10175	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_ESP_NOT_CANONICAL);
10176
10177	rc = VMXReadVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, &u64Val);
10178	AssertRCBreak(rc);
10179	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_EIP_NOT_CANONICAL);
10180	#endif
10181
10182	/*
10183	* PERF_GLOBAL MSR.
10184	*/
10185	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PERF_MSR)
10186	{
10187	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL, &u64Val);
10188	AssertRCBreak(rc);
10189	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffff8fffffffc)),
10190	VMX_IGS_PERF_GLOBAL_MSR_RESERVED); /* Bits 63:35, bits 31:2 MBZ. */
10191	}
10192
10193	/*
10194	* PAT MSR.
10195	*/
10196	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PAT_MSR)
10197	{
10198	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PAT_FULL, &u64Val);
10199	AssertRCBreak(rc);
10200	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0x707070707070707)), VMX_IGS_PAT_MSR_RESERVED);
10201	for (unsigned i = 0; i < 8; i++)
10202	{
10203	uint8_t u8Val = (u64Val & 0xff);
10204	if ( u8Val != 0 /* UC */
10205	&& u8Val != 1 /* WC */
10206	&& u8Val != 4 /* WT */
10207	&& u8Val != 5 /* WP */
10208	&& u8Val != 6 /* WB */
10209	&& u8Val != 7 /* UC- */)
10210	{
10211	HMVMX_ERROR_BREAK(VMX_IGS_PAT_MSR_INVALID);
10212	}
10213	u64Val >>= 8;
10214	}
10215	}
10216
10217	/*
10218	* EFER MSR.
10219	*/
10220	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_EFER_MSR)
10221	{
10222	Assert(pVM->hm.s.vmx.fSupportsVmcsEfer);
10223	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_EFER_FULL, &u64Val);
10224	AssertRCBreak(rc);
10225	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffffffffff2fe)),
10226	VMX_IGS_EFER_MSR_RESERVED); /* Bits 63:12, bit 9, bits 7:1 MBZ. */
10227	HMVMX_CHECK_BREAK(RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL( pVmcsInfo->u32EntryCtls
10228	& VMX_ENTRY_CTLS_IA32E_MODE_GUEST),
10229	VMX_IGS_EFER_LMA_GUEST_MODE_MISMATCH);
10230	/** @todo r=ramshankar: Unrestricted check here is probably wrong, see
10231	* iemVmxVmentryCheckGuestState(). */
10232	HMVMX_CHECK_BREAK( fUnrestrictedGuest
10233	\|\| !(u32GuestCr0 & X86_CR0_PG)
10234	\|\| RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL(u64Val & MSR_K6_EFER_LME),
10235	VMX_IGS_EFER_LMA_LME_MISMATCH);
10236	}
10237
10238	/*
10239	* Segment registers.
10240	*/
10241	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
10242	\|\| !(pCtx->ldtr.Sel & X86_SEL_LDT), VMX_IGS_LDTR_TI_INVALID);
10243	if (!(u32Eflags & X86_EFL_VM))
10244	{
10245	/* CS */
10246	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1Present, VMX_IGS_CS_ATTR_P_INVALID);
10247	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xf00), VMX_IGS_CS_ATTR_RESERVED);
10248	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xfffe0000), VMX_IGS_CS_ATTR_RESERVED);
10249	HMVMX_CHECK_BREAK( (pCtx->cs.u32Limit & 0xfff) == 0xfff
10250	\|\| !(pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
10251	HMVMX_CHECK_BREAK( !(pCtx->cs.u32Limit & 0xfff00000)
10252	\|\| (pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
10253	/* CS cannot be loaded with NULL in protected mode. */
10254	HMVMX_CHECK_BREAK(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_CS_ATTR_UNUSABLE);
10255	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1DescType, VMX_IGS_CS_ATTR_S_INVALID);
10256	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
10257	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_UNEQUAL);
10258	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
10259	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_MISMATCH);
10260	else if (pVM->hm.s.vmx.fUnrestrictedGuest && pCtx->cs.Attr.n.u4Type == 3)
10261	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == 0, VMX_IGS_CS_ATTR_DPL_INVALID);
10262	else
10263	HMVMX_ERROR_BREAK(VMX_IGS_CS_ATTR_TYPE_INVALID);
10264
10265	/* SS */
10266	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10267	\|\| (pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL), VMX_IGS_SS_CS_RPL_UNEQUAL);
10268	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL), VMX_IGS_SS_ATTR_DPL_RPL_UNEQUAL);
10269	if ( !(pCtx->cr0 & X86_CR0_PE)
10270	\|\| pCtx->cs.Attr.n.u4Type == 3)
10271	{
10272	HMVMX_CHECK_BREAK(!pCtx->ss.Attr.n.u2Dpl, VMX_IGS_SS_ATTR_DPL_INVALID);
10273	}
10274	if (!(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
10275	{
10276	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7, VMX_IGS_SS_ATTR_TYPE_INVALID);
10277	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u1Present, VMX_IGS_SS_ATTR_P_INVALID);
10278	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xf00), VMX_IGS_SS_ATTR_RESERVED);
10279	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xfffe0000), VMX_IGS_SS_ATTR_RESERVED);
10280	HMVMX_CHECK_BREAK( (pCtx->ss.u32Limit & 0xfff) == 0xfff
10281	\|\| !(pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
10282	HMVMX_CHECK_BREAK( !(pCtx->ss.u32Limit & 0xfff00000)
10283	\|\| (pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
10284	}
10285
10286	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxExportGuestSReg(). */
10287	if (!(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
10288	{
10289	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_DS_ATTR_A_INVALID);
10290	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u1Present, VMX_IGS_DS_ATTR_P_INVALID);
10291	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10292	\|\| pCtx->ds.Attr.n.u4Type > 11
10293	\|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
10294	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xf00), VMX_IGS_DS_ATTR_RESERVED);
10295	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xfffe0000), VMX_IGS_DS_ATTR_RESERVED);
10296	HMVMX_CHECK_BREAK( (pCtx->ds.u32Limit & 0xfff) == 0xfff
10297	\|\| !(pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
10298	HMVMX_CHECK_BREAK( !(pCtx->ds.u32Limit & 0xfff00000)
10299	\|\| (pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
10300	HMVMX_CHECK_BREAK( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10301	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_DS_ATTR_TYPE_INVALID);
10302	}
10303	if (!(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
10304	{
10305	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_ES_ATTR_A_INVALID);
10306	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u1Present, VMX_IGS_ES_ATTR_P_INVALID);
10307	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10308	\|\| pCtx->es.Attr.n.u4Type > 11
10309	\|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
10310	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xf00), VMX_IGS_ES_ATTR_RESERVED);
10311	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xfffe0000), VMX_IGS_ES_ATTR_RESERVED);
10312	HMVMX_CHECK_BREAK( (pCtx->es.u32Limit & 0xfff) == 0xfff
10313	\|\| !(pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
10314	HMVMX_CHECK_BREAK( !(pCtx->es.u32Limit & 0xfff00000)
10315	\|\| (pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
10316	HMVMX_CHECK_BREAK( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10317	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_ES_ATTR_TYPE_INVALID);
10318	}
10319	if (!(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
10320	{
10321	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_FS_ATTR_A_INVALID);
10322	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u1Present, VMX_IGS_FS_ATTR_P_INVALID);
10323	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10324	\|\| pCtx->fs.Attr.n.u4Type > 11
10325	\|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL), VMX_IGS_FS_ATTR_DPL_RPL_UNEQUAL);
10326	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xf00), VMX_IGS_FS_ATTR_RESERVED);
10327	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xfffe0000), VMX_IGS_FS_ATTR_RESERVED);
10328	HMVMX_CHECK_BREAK( (pCtx->fs.u32Limit & 0xfff) == 0xfff
10329	\|\| !(pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
10330	HMVMX_CHECK_BREAK( !(pCtx->fs.u32Limit & 0xfff00000)
10331	\|\| (pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
10332	HMVMX_CHECK_BREAK( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10333	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_FS_ATTR_TYPE_INVALID);
10334	}
10335	if (!(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
10336	{
10337	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_GS_ATTR_A_INVALID);
10338	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u1Present, VMX_IGS_GS_ATTR_P_INVALID);
10339	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10340	\|\| pCtx->gs.Attr.n.u4Type > 11
10341	\|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL), VMX_IGS_GS_ATTR_DPL_RPL_UNEQUAL);
10342	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xf00), VMX_IGS_GS_ATTR_RESERVED);
10343	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xfffe0000), VMX_IGS_GS_ATTR_RESERVED);
10344	HMVMX_CHECK_BREAK( (pCtx->gs.u32Limit & 0xfff) == 0xfff
10345	\|\| !(pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
10346	HMVMX_CHECK_BREAK( !(pCtx->gs.u32Limit & 0xfff00000)
10347	\|\| (pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
10348	HMVMX_CHECK_BREAK( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10349	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_GS_ATTR_TYPE_INVALID);
10350	}
10351	/* 64-bit capable CPUs. */
10352	#if HC_ARCH_BITS == 64
10353	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
10354	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
10355	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
10356	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
10357	HMVMX_CHECK_BREAK(!RT_HI_U32(pCtx->cs.u64Base), VMX_IGS_LONGMODE_CS_BASE_INVALID);
10358	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ss.u64Base),
10359	VMX_IGS_LONGMODE_SS_BASE_INVALID);
10360	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ds.u64Base),
10361	VMX_IGS_LONGMODE_DS_BASE_INVALID);
10362	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->es.u64Base),
10363	VMX_IGS_LONGMODE_ES_BASE_INVALID);
10364	#endif
10365	}
10366	else
10367	{
10368	/* V86 mode checks. */
10369	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
10370	if (pVmcsInfo->RealMode.fRealOnV86Active)
10371	{
10372	u32CSAttr = 0xf3; u32SSAttr = 0xf3;
10373	u32DSAttr = 0xf3; u32ESAttr = 0xf3;
10374	u32FSAttr = 0xf3; u32GSAttr = 0xf3;
10375	}
10376	else
10377	{
10378	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u;
10379	u32DSAttr = pCtx->ds.Attr.u; u32ESAttr = pCtx->es.Attr.u;
10380	u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
10381	}
10382
10383	/* CS */
10384	HMVMX_CHECK_BREAK((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), VMX_IGS_V86_CS_BASE_INVALID);
10385	HMVMX_CHECK_BREAK(pCtx->cs.u32Limit == 0xffff, VMX_IGS_V86_CS_LIMIT_INVALID);
10386	HMVMX_CHECK_BREAK(u32CSAttr == 0xf3, VMX_IGS_V86_CS_ATTR_INVALID);
10387	/* SS */
10388	HMVMX_CHECK_BREAK((pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4), VMX_IGS_V86_SS_BASE_INVALID);
10389	HMVMX_CHECK_BREAK(pCtx->ss.u32Limit == 0xffff, VMX_IGS_V86_SS_LIMIT_INVALID);
10390	HMVMX_CHECK_BREAK(u32SSAttr == 0xf3, VMX_IGS_V86_SS_ATTR_INVALID);
10391	/* DS */
10392	HMVMX_CHECK_BREAK((pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4), VMX_IGS_V86_DS_BASE_INVALID);
10393	HMVMX_CHECK_BREAK(pCtx->ds.u32Limit == 0xffff, VMX_IGS_V86_DS_LIMIT_INVALID);
10394	HMVMX_CHECK_BREAK(u32DSAttr == 0xf3, VMX_IGS_V86_DS_ATTR_INVALID);
10395	/* ES */
10396	HMVMX_CHECK_BREAK((pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4), VMX_IGS_V86_ES_BASE_INVALID);
10397	HMVMX_CHECK_BREAK(pCtx->es.u32Limit == 0xffff, VMX_IGS_V86_ES_LIMIT_INVALID);
10398	HMVMX_CHECK_BREAK(u32ESAttr == 0xf3, VMX_IGS_V86_ES_ATTR_INVALID);
10399	/* FS */
10400	HMVMX_CHECK_BREAK((pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4), VMX_IGS_V86_FS_BASE_INVALID);
10401	HMVMX_CHECK_BREAK(pCtx->fs.u32Limit == 0xffff, VMX_IGS_V86_FS_LIMIT_INVALID);
10402	HMVMX_CHECK_BREAK(u32FSAttr == 0xf3, VMX_IGS_V86_FS_ATTR_INVALID);
10403	/* GS */
10404	HMVMX_CHECK_BREAK((pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4), VMX_IGS_V86_GS_BASE_INVALID);
10405	HMVMX_CHECK_BREAK(pCtx->gs.u32Limit == 0xffff, VMX_IGS_V86_GS_LIMIT_INVALID);
10406	HMVMX_CHECK_BREAK(u32GSAttr == 0xf3, VMX_IGS_V86_GS_ATTR_INVALID);
10407	/* 64-bit capable CPUs. */
10408	#if HC_ARCH_BITS == 64
10409	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
10410	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
10411	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
10412	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
10413	HMVMX_CHECK_BREAK(!RT_HI_U32(pCtx->cs.u64Base), VMX_IGS_LONGMODE_CS_BASE_INVALID);
10414	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ss.u64Base),
10415	VMX_IGS_LONGMODE_SS_BASE_INVALID);
10416	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ds.u64Base),
10417	VMX_IGS_LONGMODE_DS_BASE_INVALID);
10418	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->es.u64Base),
10419	VMX_IGS_LONGMODE_ES_BASE_INVALID);
10420	#endif
10421	}
10422
10423	/*
10424	* TR.
10425	*/
10426	HMVMX_CHECK_BREAK(!(pCtx->tr.Sel & X86_SEL_LDT), VMX_IGS_TR_TI_INVALID);
10427	/* 64-bit capable CPUs. */
10428	#if HC_ARCH_BITS == 64
10429	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->tr.u64Base), VMX_IGS_TR_BASE_NOT_CANONICAL);
10430	#endif
10431	if (fLongModeGuest)
10432	{
10433	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u4Type == 11, /* 64-bit busy TSS. */
10434	VMX_IGS_LONGMODE_TR_ATTR_TYPE_INVALID);
10435	}
10436	else
10437	{
10438	HMVMX_CHECK_BREAK( pCtx->tr.Attr.n.u4Type == 3 /* 16-bit busy TSS. */
10439	\|\| pCtx->tr.Attr.n.u4Type == 11, /* 32-bit busy TSS.*/
10440	VMX_IGS_TR_ATTR_TYPE_INVALID);
10441	}
10442	HMVMX_CHECK_BREAK(!pCtx->tr.Attr.n.u1DescType, VMX_IGS_TR_ATTR_S_INVALID);
10443	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u1Present, VMX_IGS_TR_ATTR_P_INVALID);
10444	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & 0xf00), VMX_IGS_TR_ATTR_RESERVED); /* Bits 11:8 MBZ. */
10445	HMVMX_CHECK_BREAK( (pCtx->tr.u32Limit & 0xfff) == 0xfff
10446	\|\| !(pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
10447	HMVMX_CHECK_BREAK( !(pCtx->tr.u32Limit & 0xfff00000)
10448	\|\| (pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
10449	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_TR_ATTR_UNUSABLE);
10450
10451	/*
10452	* GDTR and IDTR.
10453	*/
10454	#if HC_ARCH_BITS == 64
10455	rc = VMXReadVmcs64(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
10456	AssertRCBreak(rc);
10457	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_GDTR_BASE_NOT_CANONICAL);
10458
10459	rc = VMXReadVmcs64(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
10460	AssertRCBreak(rc);
10461	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_IDTR_BASE_NOT_CANONICAL);
10462	#endif
10463
10464	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val);
10465	AssertRCBreak(rc);
10466	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_GDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
10467
10468	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val);
10469	AssertRCBreak(rc);
10470	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_IDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
10471
10472	/*
10473	* Guest Non-Register State.
10474	*/
10475	/* Activity State. */
10476	uint32_t u32ActivityState;
10477	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, &u32ActivityState);
10478	AssertRCBreak(rc);
10479	HMVMX_CHECK_BREAK( !u32ActivityState
10480	\|\| (u32ActivityState & RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Misc, VMX_BF_MISC_ACTIVITY_STATES)),
10481	VMX_IGS_ACTIVITY_STATE_INVALID);
10482	HMVMX_CHECK_BREAK( !(pCtx->ss.Attr.n.u2Dpl)
10483	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT, VMX_IGS_ACTIVITY_STATE_HLT_INVALID);
10484	uint32_t u32IntrState;
10485	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &u32IntrState);
10486	AssertRCBreak(rc);
10487	if ( u32IntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS
10488	\|\| u32IntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10489	{
10490	HMVMX_CHECK_BREAK(u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE, VMX_IGS_ACTIVITY_STATE_ACTIVE_INVALID);
10491	}
10492
10493	/** @todo Activity state and injecting interrupts. Left as a todo since we
10494	* currently don't use activity states but ACTIVE. */
10495
10496	HMVMX_CHECK_BREAK( !(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)
10497	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT, VMX_IGS_ACTIVITY_STATE_SIPI_WAIT_INVALID);
10498
10499	/* Guest interruptibility-state. */
10500	HMVMX_CHECK_BREAK(!(u32IntrState & 0xffffffe0), VMX_IGS_INTERRUPTIBILITY_STATE_RESERVED);
10501	HMVMX_CHECK_BREAK((u32IntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
10502	!= (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
10503	VMX_IGS_INTERRUPTIBILITY_STATE_STI_MOVSS_INVALID);
10504	HMVMX_CHECK_BREAK( (u32Eflags & X86_EFL_IF)
10505	\|\| !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI),
10506	VMX_IGS_INTERRUPTIBILITY_STATE_STI_EFL_INVALID);
10507	if (VMX_ENTRY_INT_INFO_IS_EXT_INT(u32EntryInfo))
10508	{
10509	HMVMX_CHECK_BREAK( !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10510	&& !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
10511	VMX_IGS_INTERRUPTIBILITY_STATE_EXT_INT_INVALID);
10512	}
10513	else if (VMX_ENTRY_INT_INFO_IS_XCPT_NMI(u32EntryInfo))
10514	{
10515	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
10516	VMX_IGS_INTERRUPTIBILITY_STATE_MOVSS_INVALID);
10517	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI),
10518	VMX_IGS_INTERRUPTIBILITY_STATE_STI_INVALID);
10519	}
10520	/** @todo Assumes the processor is not in SMM. */
10521	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI),
10522	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_INVALID);
10523	HMVMX_CHECK_BREAK( !(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)
10524	\|\| (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI),
10525	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_SMM_INVALID);
10526	if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
10527	&& VMX_ENTRY_INT_INFO_IS_XCPT_NMI(u32EntryInfo))
10528	{
10529	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI),
10530	VMX_IGS_INTERRUPTIBILITY_STATE_NMI_INVALID);
10531	}
10532
10533	/* Pending debug exceptions. */
10534	#if HC_ARCH_BITS == 64
10535	rc = VMXReadVmcs64(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, &u64Val);
10536	AssertRCBreak(rc);
10537	/* Bits 63:15, Bit 13, Bits 11:4 MBZ. */
10538	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffffaff0)), VMX_IGS_LONGMODE_PENDING_DEBUG_RESERVED);
10539	u32Val = u64Val; /* For pending debug exceptions checks below. */
10540	#else
10541	rc = VMXReadVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, &u32Val);
10542	AssertRCBreak(rc);
10543	/* Bits 31:15, Bit 13, Bits 11:4 MBZ. */
10544	HMVMX_CHECK_BREAK(!(u32Val & 0xffffaff0), VMX_IGS_PENDING_DEBUG_RESERVED);
10545	#endif
10546
10547	if ( (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10548	\|\| (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS)
10549	\|\| u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT)
10550	{
10551	if ( (u32Eflags & X86_EFL_TF)
10552	&& !(u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
10553	{
10554	/* Bit 14 is PendingDebug.BS. */
10555	HMVMX_CHECK_BREAK(u32Val & RT_BIT(14), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_SET);
10556	}
10557	if ( !(u32Eflags & X86_EFL_TF)
10558	\|\| (u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
10559	{
10560	/* Bit 14 is PendingDebug.BS. */
10561	HMVMX_CHECK_BREAK(!(u32Val & RT_BIT(14)), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_CLEAR);
10562	}
10563	}
10564
10565	/* VMCS link pointer. */
10566	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, &u64Val);
10567	AssertRCBreak(rc);
10568	if (u64Val != UINT64_C(0xffffffffffffffff))
10569	{
10570	HMVMX_CHECK_BREAK(!(u64Val & 0xfff), VMX_IGS_VMCS_LINK_PTR_RESERVED);
10571	/** @todo Bits beyond the processor's physical-address width MBZ. */
10572	/** @todo SMM checks. */
10573	Assert(pVmcsInfo->HCPhysShadowVmcs == u64Val);
10574	Assert(pVmcsInfo->pvShadowVmcs);
10575	VMXVMCSREVID VmcsRevId;
10576	VmcsRevId.u = (uint32_t )pVmcsInfo->pvShadowVmcs;
10577	HMVMX_CHECK_BREAK(VmcsRevId.n.u31RevisionId == RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_ID),
10578	VMX_IGS_VMCS_LINK_PTR_SHADOW_VMCS_ID_INVALID);
10579	HMVMX_CHECK_BREAK(VmcsRevId.n.fIsShadowVmcs == (uint32_t)!!(pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING),
10580	VMX_IGS_VMCS_LINK_PTR_NOT_SHADOW);
10581	}
10582
10583	/** @todo Checks on Guest Page-Directory-Pointer-Table Entries when guest is
10584	* not using nested paging? */
10585	if ( pVM->hm.s.fNestedPaging
10586	&& !fLongModeGuest
10587	&& CPUMIsGuestInPAEModeEx(pCtx))
10588	{
10589	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &u64Val);
10590	AssertRCBreak(rc);
10591	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10592
10593	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &u64Val);
10594	AssertRCBreak(rc);
10595	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10596
10597	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &u64Val);
10598	AssertRCBreak(rc);
10599	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10600
10601	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &u64Val);
10602	AssertRCBreak(rc);
10603	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10604	}
10605
10606	/* Shouldn't happen but distinguish it from AssertRCBreak() errors. */
10607	if (uError == VMX_IGS_ERROR)
10608	uError = VMX_IGS_REASON_NOT_FOUND;
10609	} while (0);
10610
10611	pVCpu->hm.s.u32HMError = uError;
10612	return uError;
10613
10614	#undef HMVMX_ERROR_BREAK
10615	#undef HMVMX_CHECK_BREAK
10616	}
10617
10618
10619	/**
10620	* Map the APIC-access page for virtualizing APIC accesses.
10621	*
10622	* This can cause a longjumps to R3 due to the acquisition of the PGM lock. Hence,
10623	* this not done as part of exporting guest state, see @bugref{8721}.
10624	*
10625	* @returns VBox status code.
10626	* @param pVCpu The cross context virtual CPU structure.
10627	*/
10628	static int hmR0VmxMapHCApicAccessPage(PVMCPU pVCpu)
10629	{
10630	PVM pVM = pVCpu->CTX_SUFF(pVM);
10631	uint64_t const u64MsrApicBase = APICGetBaseMsrNoCheck(pVCpu);
10632
10633	Assert(PDMHasApic(pVM));
10634	Assert(u64MsrApicBase);
10635
10636	RTGCPHYS const GCPhysApicBase = u64MsrApicBase & PAGE_BASE_GC_MASK;
10637	Log4Func(("Mappping HC APIC-access page at %#RGp\n", GCPhysApicBase));
10638
10639	/* Unalias the existing mapping. */
10640	int rc = PGMHandlerPhysicalReset(pVM, GCPhysApicBase);
10641	AssertRCReturn(rc, rc);
10642
10643	/* Map the HC APIC-access page in place of the MMIO page, also updates the shadow page tables if necessary. */
10644	Assert(pVM->hm.s.vmx.HCPhysApicAccess != NIL_RTHCPHYS);
10645	rc = IOMMMIOMapMMIOHCPage(pVM, pVCpu, GCPhysApicBase, pVM->hm.s.vmx.HCPhysApicAccess, X86_PTE_RW \| X86_PTE_P);
10646	AssertRCReturn(rc, rc);
10647
10648	/* Update the per-VCPU cache of the APIC base MSR. */
10649	pVCpu->hm.s.vmx.u64GstMsrApicBase = u64MsrApicBase;
10650	return VINF_SUCCESS;
10651	}
10652
10653
10654	/**
10655	* Worker function passed to RTMpOnSpecific() that is to be called on the target
10656	* CPU.
10657	*
10658	* @param idCpu The ID for the CPU the function is called on.
10659	* @param pvUser1 Null, not used.
10660	* @param pvUser2 Null, not used.
10661	*/
10662	static DECLCALLBACK(void) hmR0DispatchHostNmi(RTCPUID idCpu, void pvUser1, void pvUser2)
10663	{
10664	RT_NOREF3(idCpu, pvUser1, pvUser2);
10665	VMXDispatchHostNmi();
10666	}
10667
10668
10669	/**
10670	* Dispatching an NMI on the host CPU that received it.
10671	*
10672	* @returns VBox status code.
10673	* @param pVCpu The cross context virtual CPU structure.
10674	* @param pVmcsInfo The VMCS info. object corresponding to the VMCS that was
10675	* executing when receiving the host NMI in VMX non-root
10676	* operation.
10677	*/
10678	static int hmR0VmxExitHostNmi(PVMCPU pVCpu, PCVMXVMCSINFO pVmcsInfo)
10679	{
10680	RTCPUID const idCpu = pVmcsInfo->idHostCpu;
10681
10682	/*
10683	* We don't want to delay dispatching the NMI any more than we have to. However,
10684	* we have already chosen -not- to dispatch NMIs when interrupts were still disabled
10685	* after executing guest or nested-guest code for the following reasons:
10686	*
10687	* - We would need to perform VMREADs with interrupts disabled and is orders of
10688	* magnitude worse when we run as a guest hypervisor without VMCS shadowing
10689	* supported by the host hypervisor.
10690	*
10691	* - It affects the common VM-exit scenario and keeps interrupts disabled for a
10692	* longer period of time just for handling an edge case like host NMIs which do
10693	* not occur nearly as frequently as other VM-exits.
10694	*
10695	* Let's cover the most likely scenario first. Check if we are on the target CPU
10696	* and dispatch the NMI right away. This should be much faster than calling into
10697	* RTMpOnSpecific() machinery.
10698	*/
10699	bool fDispatched = false;
10700	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
10701	if (idCpu == RTMpCpuId())
10702	{
10703	VMXDispatchHostNmi();
10704	fDispatched = true;
10705	}
10706	ASMSetFlags(fEFlags);
10707	if (fDispatched)
10708	{
10709	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC);
10710	return VINF_SUCCESS;
10711	}
10712
10713	/*
10714	* RTMpOnSpecific() waits until the worker function has run on the target CPU. So
10715	* there should be no race or recursion even if we are unlucky enough to be preempted
10716	* (to the target CPU) without dispatching the host NMI above.
10717	*/
10718	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGCIpi);
10719	return RTMpOnSpecific(idCpu, &hmR0DispatchHostNmi, NULL /* pvUser1 /, NULL / pvUser2 */);
10720	}
10721
10722
10723	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10724	/**
10725	* Merges the guest with the nested-guest MSR bitmap in preparation of executing the
10726	* nested-guest using hardware-assisted VMX.
10727	*
10728	* @param pVCpu The cross context virtual CPU structure.
10729	* @param pVmcsInfoNstGst The nested-guest VMCS info. object.
10730	* @param pVmcsInfoGst The guest VMCS info. object.
10731	*/
10732	static void hmR0VmxMergeMsrBitmapNested(PCVMCPU pVCpu, PVMXVMCSINFO pVmcsInfoNstGst, PCVMXVMCSINFO pVmcsInfoGst)
10733	{
10734	uint32_t const cbMsrBitmap = X86_PAGE_4K_SIZE;
10735	uint64_t pu64MsrBitmap = (uint64_t )pVmcsInfoNstGst->pvMsrBitmap;
10736	Assert(pu64MsrBitmap);
10737
10738	/*
10739	* We merge the guest MSR bitmap with the nested-guest MSR bitmap such that any
10740	* MSR that is intercepted by the guest is also intercepted while executing the
10741	* nested-guest using hardware-assisted VMX.
10742	*
10743	* Note! If the nested-guest is not using an MSR bitmap, ever MSR must cause a
10744	* nested-guest VM-exit even if the outer guest is not intercepting some
10745	* MSRs. We cannot assume the caller has initialized the nested-guest
10746	* MSR bitmap in this case.
10747	*
10748	* The guest hypervisor may also switch whether it uses MSR bitmaps for
10749	* each VM-entry, hence initializing it once per-VM while setting up the
10750	* nested-guest VMCS is not sufficient.
10751	*/
10752	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
10753	if (pVmcsNstGst->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
10754	{
10755	uint64_t const pu64MsrBitmapNstGst = (uint64_t const )pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap);
10756	uint64_t const pu64MsrBitmapGst = (uint64_t const )pVmcsInfoGst->pvMsrBitmap;
10757	Assert(pu64MsrBitmapNstGst);
10758	Assert(pu64MsrBitmapGst);
10759
10760	uint32_t const cFrags = cbMsrBitmap / sizeof(uint64_t);
10761	for (uint32_t i = 0; i < cFrags; i++)
10762	pu64MsrBitmap[i] = pu64MsrBitmapNstGst[i] \| pu64MsrBitmapGst[i];
10763	}
10764	else
10765	ASMMemFill32(pu64MsrBitmap, cbMsrBitmap, UINT32_C(0xffffffff));
10766	}
10767
10768
10769	/**
10770	* Merges the guest VMCS in to the nested-guest VMCS controls in preparation of
10771	* hardware-assisted VMX execution of the nested-guest.
10772	*
10773	* For a guest, we don't modify these controls once we set up the VMCS and hence
10774	* this function is never called.
10775	*
10776	* For nested-guests since the guest hypervisor provides these controls on every
10777	* nested-guest VM-entry and could potentially change them everytime we need to
10778	* merge them before every nested-guest VM-entry.
10779	*
10780	* @returns VBox status code.
10781	* @param pVCpu The cross context virtual CPU structure.
10782	*/
10783	static int hmR0VmxMergeVmcsNested(PVMCPU pVCpu)
10784	{
10785	PVM pVM = pVCpu->CTX_SUFF(pVM);
10786	PCVMXVMCSINFO pVmcsInfoGst = &pVCpu->hm.s.vmx.VmcsInfo;
10787	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
10788	Assert(pVmcsNstGst);
10789
10790	/*
10791	* Merge the controls with the requirements of the guest VMCS.
10792	*
10793	* We do not need to validate the nested-guest VMX features specified in the nested-guest
10794	* VMCS with the features supported by the physical CPU as it's already done by the
10795	* VMLAUNCH/VMRESUME instruction emulation.
10796	*
10797	* This is because the VMX features exposed by CPUM (through CPUID/MSRs) to the guest are
10798	* derived from the VMX features supported by the physical CPU.
10799	*/
10800
10801	/* Pin-based VM-execution controls. */
10802	uint32_t const u32PinCtls = pVmcsNstGst->u32PinCtls \| pVmcsInfoGst->u32PinCtls;
10803
10804	/* Processor-based VM-execution controls. */
10805	uint32_t u32ProcCtls = (pVmcsNstGst->u32ProcCtls & ~VMX_PROC_CTLS_USE_IO_BITMAPS)
10806	\| (pVmcsInfoGst->u32ProcCtls & ~( VMX_PROC_CTLS_INT_WINDOW_EXIT
10807	\| VMX_PROC_CTLS_NMI_WINDOW_EXIT
10808	\| VMX_PROC_CTLS_USE_TPR_SHADOW
10809	\| VMX_PROC_CTLS_MONITOR_TRAP_FLAG));
10810
10811	/* Secondary processor-based VM-execution controls. */
10812	uint32_t const u32ProcCtls2 = (pVmcsNstGst->u32ProcCtls2 & ~VMX_PROC_CTLS2_VPID)
10813	\| (pVmcsInfoGst->u32ProcCtls2 & ~( VMX_PROC_CTLS2_VIRT_APIC_ACCESS
10814	\| VMX_PROC_CTLS2_INVPCID
10815	\| VMX_PROC_CTLS2_VMCS_SHADOWING
10816	\| VMX_PROC_CTLS2_RDTSCP
10817	\| VMX_PROC_CTLS2_XSAVES_XRSTORS
10818	\| VMX_PROC_CTLS2_APIC_REG_VIRT
10819	\| VMX_PROC_CTLS2_VIRT_INT_DELIVERY
10820	\| VMX_PROC_CTLS2_VMFUNC));
10821
10822	/*
10823	* VM-entry controls:
10824	* These controls contains state that depends on the nested-guest state (primarily
10825	* EFER MSR) and is thus not constant between VMLAUNCH/VMRESUME and the nested-guest
10826	* VM-exit. Although the guest hypervisor cannot change it, we need to in order to
10827	* properly continue executing the nested-guest if the EFER MSR changes but does not
10828	* cause a nested-guest VM-exits.
10829	*
10830	* VM-exit controls:
10831	* These controls specify the host state on return. We cannot use the controls from
10832	* the guest hypervisor state as is as it would contain the guest state rather than
10833	* the host state. Since the host state is subject to change (e.g. preemption, trips
10834	* to ring-3, longjmp and rescheduling to a different host CPU) they are not constant
10835	* through VMLAUNCH/VMRESUME and the nested-guest VM-exit.
10836	*
10837	* VM-entry MSR-load:
10838	* The guest MSRs from the VM-entry MSR-load area are already loaded into the guest-CPU
10839	* context by the VMLAUNCH/VMRESUME instruction emulation.
10840	*
10841	* VM-exit MSR-store:
10842	* The VM-exit emulation will take care of populating the MSRs from the guest-CPU context
10843	* back into the VM-exit MSR-store area.
10844	*
10845	* VM-exit MSR-load areas:
10846	* This must contain the real host MSRs with hardware-assisted VMX execution. Hence, we
10847	* can entirely ignore what the guest hypervisor wants to load here.
10848	*/
10849
10850	/*
10851	* Exception bitmap.
10852	*
10853	* We could remove #UD from the guest bitmap and merge it with the nested-guest bitmap
10854	* here (and avoid doing anything while exporting nested-guest state), but to keep the
10855	* code more flexible if intercepting exceptions become more dynamic in the future we do
10856	* it as part of exporting the nested-guest state.
10857	*/
10858	uint32_t const u32XcptBitmap = pVmcsNstGst->u32XcptBitmap \| pVmcsInfoGst->u32XcptBitmap;
10859
10860	/*
10861	* CR0/CR4 guest/host mask.
10862	*
10863	* Modifications by the nested-guest to CR0/CR4 bits owned by the host and the guest must
10864	* cause VM-exits, so we need to merge them here.
10865	*/
10866	uint64_t const u64Cr0Mask = pVmcsNstGst->u64Cr0Mask.u \| pVmcsInfoGst->u64Cr0Mask;
10867	uint64_t const u64Cr4Mask = pVmcsNstGst->u64Cr4Mask.u \| pVmcsInfoGst->u64Cr4Mask;
10868
10869	/*
10870	* Page-fault error-code mask and match.
10871	*
10872	* Although we require unrestricted guest execution (and thereby nested-paging) for
10873	* hardware-assisted VMX execution of nested-guests and thus the outer guest doesn't
10874	* normally intercept #PFs, it might intercept them for debugging purposes.
10875	*
10876	* If the outer guest is not intercepting #PFs, we can use the nested-guest #PF filters.
10877	* If the outer guest is intercepting #PFs, we must intercept all #PFs.
10878	*/
10879	uint32_t u32XcptPFMask;
10880	uint32_t u32XcptPFMatch;
10881	if (!(pVmcsInfoGst->u32XcptBitmap & RT_BIT(X86_XCPT_PF)))
10882	{
10883	u32XcptPFMask = pVmcsNstGst->u32XcptPFMask;
10884	u32XcptPFMatch = pVmcsNstGst->u32XcptPFMatch;
10885	}
10886	else
10887	{
10888	u32XcptPFMask = 0;
10889	u32XcptPFMatch = 0;
10890	}
10891
10892	/*
10893	* Pause-Loop exiting.
10894	*/
10895	uint32_t const cPleGapTicks = RT_MIN(pVM->hm.s.vmx.cPleGapTicks, pVmcsNstGst->u32PleGap);
10896	uint32_t const cPleWindowTicks = RT_MIN(pVM->hm.s.vmx.cPleWindowTicks, pVmcsNstGst->u32PleWindow);
10897
10898	/*
10899	* I/O Bitmap.
10900	*
10901	* We do not use the I/O bitmap that may be provided by the guest hypervisor as we always
10902	* intercept all I/O port accesses.
10903	*/
10904	Assert(u32ProcCtls & VMX_PROC_CTLS_UNCOND_IO_EXIT);
10905
10906	/*
10907	* VMCS shadowing.
10908	*
10909	* We do not yet expose VMCS shadowing to the guest and thus VMCS shadowing should not be
10910	* enabled while executing the nested-guest.
10911	*/
10912	Assert(!(u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING));
10913
10914	/*
10915	* APIC-access page.
10916	*/
10917	RTHCPHYS HCPhysApicAccess;
10918	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10919	{
10920	Assert(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS);
10921	RTGCPHYS const GCPhysApicAccess = pVmcsNstGst->u64AddrApicAccess.u;
10922
10923	/** @todo NSTVMX: This is not really correct but currently is required to make
10924	* things work. We need to re-register the page handler when we fallback to
10925	* IEM execution of the nested-guest! */
10926	PGMHandlerPhysicalDeregister(pVM, GCPhysApicAccess);
10927
10928	void *pvPage;
10929	PGMPAGEMAPLOCK PgMapLockApicAccess;
10930	int rc = PGMPhysGCPhys2CCPtr(pVM, GCPhysApicAccess, &pvPage, &PgMapLockApicAccess);
10931	if (RT_SUCCESS(rc))
10932	{
10933	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysApicAccess, &HCPhysApicAccess);
10934	AssertMsgRCReturn(rc, ("Failed to get host-physical address for APIC-access page at %#RGp\n", GCPhysApicAccess), rc);
10935
10936	/*
10937	* We can release the page lock here because the APIC-access page is never read or
10938	* written to but merely serves as a placeholder in the shadow/nested page tables
10939	* to cause VM-exits or re-direct the access to the virtual-APIC page.
10940	*/
10941	PGMPhysReleasePageMappingLock(pVCpu->CTX_SUFF(pVM), &PgMapLockApicAccess);
10942	}
10943	else
10944	return rc;
10945	}
10946	else
10947	HCPhysApicAccess = 0;
10948
10949	/*
10950	* Virtual-APIC page and TPR threshold.
10951	*/
10952	PVMXVMCSINFO pVmcsInfoNstGst = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
10953	RTHCPHYS HCPhysVirtApic;
10954	uint32_t u32TprThreshold;
10955	if (u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
10956	{
10957	Assert(pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW);
10958
10959	void *pvPage;
10960	RTGCPHYS const GCPhysVirtApic = pVmcsNstGst->u64AddrVirtApic.u;
10961	int rc = PGMPhysGCPhys2CCPtr(pVM, GCPhysVirtApic, &pvPage, &pVCpu->hm.s.vmx.PgMapLockVirtApic);
10962	AssertMsgRCReturn(rc, ("Failed to get current-context pointer for virtual-APIC page at %#RGp\n", GCPhysVirtApic), rc);
10963
10964	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysVirtApic, &HCPhysVirtApic);
10965	AssertMsgRCReturn(rc, ("Failed to get host-physical address for virtual-APIC page at %#RGp\n", GCPhysVirtApic), rc);
10966	pVCpu->hm.s.vmx.fVirtApicPageLocked = true;
10967
10968	u32TprThreshold = pVmcsNstGst->u32TprThreshold;
10969	}
10970	else
10971	{
10972	HCPhysVirtApic = 0;
10973	u32TprThreshold = 0;
10974
10975	/*
10976	* We must make sure CR8 reads/write must cause VM-exits when TPR shadowing is not
10977	* used by the guest hypervisor. Preventing MMIO accesses to the physical APIC will
10978	* be taken care of by EPT/shadow paging.
10979	*/
10980	if (pVM->hm.s.fAllow64BitGuests)
10981	{
10982	u32ProcCtls \|= VMX_PROC_CTLS_CR8_STORE_EXIT
10983	\| VMX_PROC_CTLS_CR8_LOAD_EXIT;
10984	}
10985	}
10986
10987	/*
10988	* Validate basic assumptions.
10989	*/
10990	Assert(pVM->hm.s.vmx.fAllowUnrestricted);
10991	Assert(pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS);
10992	Assert(hmGetVmxActiveVmcsInfo(pVCpu) == pVmcsInfoNstGst);
10993
10994	/*
10995	* Commit it to the nested-guest VMCS.
10996	*/
10997	int rc = VINF_SUCCESS;
10998	if (pVmcsInfoNstGst->u32PinCtls != u32PinCtls)
10999	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, u32PinCtls);
11000	if (pVmcsInfoNstGst->u32ProcCtls != u32ProcCtls)
11001	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, u32ProcCtls);
11002	if (pVmcsInfoNstGst->u32ProcCtls2 != u32ProcCtls2)
11003	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, u32ProcCtls2);
11004	if (pVmcsInfoNstGst->u32XcptBitmap != u32XcptBitmap)
11005	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
11006	if (pVmcsInfoNstGst->u64Cr0Mask != u64Cr0Mask)
11007	rc \|= VMXWriteVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, u64Cr0Mask);
11008	if (pVmcsInfoNstGst->u64Cr4Mask != u64Cr4Mask)
11009	rc \|= VMXWriteVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, u64Cr4Mask);
11010	if (pVmcsInfoNstGst->u32XcptPFMask != u32XcptPFMask)
11011	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, u32XcptPFMask);
11012	if (pVmcsInfoNstGst->u32XcptPFMatch != u32XcptPFMatch)
11013	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, u32XcptPFMatch);
11014	if ( !(u32ProcCtls & VMX_PROC_CTLS_PAUSE_EXIT)
11015	&& (u32ProcCtls2 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT))
11016	{
11017	Assert(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT);
11018	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, cPleGapTicks);
11019	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, cPleWindowTicks);
11020	}
11021	if (u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
11022	{
11023	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
11024	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL, HCPhysVirtApic);
11025	}
11026	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
11027	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, HCPhysApicAccess);
11028	AssertRCReturn(rc, rc);
11029
11030	/*
11031	* Update the nested-guest VMCS cache.
11032	*/
11033	pVmcsInfoNstGst->u32PinCtls = u32PinCtls;
11034	pVmcsInfoNstGst->u32ProcCtls = u32ProcCtls;
11035	pVmcsInfoNstGst->u32ProcCtls2 = u32ProcCtls2;
11036	pVmcsInfoNstGst->u32XcptBitmap = u32XcptBitmap;
11037	pVmcsInfoNstGst->u64Cr0Mask = u64Cr0Mask;
11038	pVmcsInfoNstGst->u64Cr4Mask = u64Cr4Mask;
11039	pVmcsInfoNstGst->u32XcptPFMask = u32XcptPFMask;
11040	pVmcsInfoNstGst->u32XcptPFMatch = u32XcptPFMatch;
11041	pVmcsInfoNstGst->HCPhysVirtApic = HCPhysVirtApic;
11042
11043	/*
11044	* We need to flush the TLB if we are switching the APIC-access page address.
11045	* See Intel spec. 28.3.3.4 "Guidelines for Use of the INVEPT Instruction".
11046	*/
11047	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
11048	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = true;
11049
11050	/*
11051	* MSR bitmap.
11052	*
11053	* The MSR bitmap address has already been initialized while setting up the nested-guest
11054	* VMCS, here we need to merge the MSR bitmaps.
11055	*/
11056	if (u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
11057	hmR0VmxMergeMsrBitmapNested(pVCpu, pVmcsInfoNstGst, pVmcsInfoGst);
11058
11059	return VINF_SUCCESS;
11060	}
11061	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
11062
11063
11064	/**
11065	* Does the preparations before executing guest code in VT-x.
11066	*
11067	* This may cause longjmps to ring-3 and may even result in rescheduling to the
11068	* recompiler/IEM. We must be cautious what we do here regarding committing
11069	* guest-state information into the VMCS assuming we assuredly execute the
11070	* guest in VT-x mode.
11071	*
11072	* If we fall back to the recompiler/IEM after updating the VMCS and clearing
11073	* the common-state (TRPM/forceflags), we must undo those changes so that the
11074	* recompiler/IEM can (and should) use them when it resumes guest execution.
11075	* Otherwise such operations must be done when we can no longer exit to ring-3.
11076	*
11077	* @returns Strict VBox status code (i.e. informational status codes too).
11078	* @retval VINF_SUCCESS if we can proceed with running the guest, interrupts
11079	* have been disabled.
11080	* @retval VINF_VMX_VMEXIT if a nested-guest VM-exit occurs (e.g., while evaluating
11081	* pending events).
11082	* @retval VINF_EM_RESET if a triple-fault occurs while injecting a
11083	* double-fault into the guest.
11084	* @retval VINF_EM_DBG_STEPPED if @a fStepping is true and an event was
11085	* dispatched directly.
11086	* @retval VINF_* scheduling changes, we have to go back to ring-3.
11087	*
11088	* @param pVCpu The cross context virtual CPU structure.
11089	* @param pVmxTransient The VMX-transient structure.
11090	* @param fStepping Whether we are single-stepping the guest in the
11091	* hypervisor debugger. Makes us ignore some of the reasons
11092	* for returning to ring-3, and return VINF_EM_DBG_STEPPED
11093	* if event dispatching took place.
11094	*/
11095	static VBOXSTRICTRC hmR0VmxPreRunGuest(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, bool fStepping)
11096	{
11097	Assert(VMMRZCallRing3IsEnabled(pVCpu));
11098
11099	#ifdef VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM
11100	if (pVmxTransient->fIsNestedGuest)
11101	{
11102	RT_NOREF2(pVCpu, fStepping);
11103	Log2Func(("Rescheduling to IEM due to nested-hwvirt or forced IEM exec -> VINF_EM_RESCHEDULE_REM\n"));
11104	return VINF_EM_RESCHEDULE_REM;
11105	}
11106	#endif
11107
11108	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
11109	PGMRZDynMapFlushAutoSet(pVCpu);
11110	#endif
11111
11112	/*
11113	* Check and process force flag actions, some of which might require us to go back to ring-3.
11114	*/
11115	VBOXSTRICTRC rcStrict = hmR0VmxCheckForceFlags(pVCpu, fStepping);
11116	if (rcStrict == VINF_SUCCESS)
11117	{ /* FFs don't get set all the time. */ }
11118	else
11119	return rcStrict;
11120
11121	/*
11122	* Virtualize memory-mapped accesses to the physical APIC (may take locks).
11123	*/
11124	/** @todo Doing this from ring-3 after VM setup phase causes a
11125	* VERR_IOM_MMIO_RANGE_NOT_FOUND guru while booting Visa 64 SMP VM. No
11126	* idea why atm. */
11127	PVM pVM = pVCpu->CTX_SUFF(pVM);
11128	if ( !pVCpu->hm.s.vmx.u64GstMsrApicBase
11129	&& (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
11130	&& PDMHasApic(pVM))
11131	{
11132	int rc = hmR0VmxMapHCApicAccessPage(pVCpu);
11133	AssertRCReturn(rc, rc);
11134	}
11135
11136	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11137	/*
11138	* Merge guest VMCS controls with the nested-guest VMCS controls.
11139	*
11140	* Even if we have not executed the guest prior to this (e.g. when resuming from a
11141	* saved state), we should be okay with merging controls as we initialize the
11142	* guest VMCS controls as part of VM setup phase.
11143	*/
11144	if ( pVmxTransient->fIsNestedGuest
11145	&& !pVCpu->hm.s.vmx.fMergedNstGstCtls)
11146	{
11147	int rc = hmR0VmxMergeVmcsNested(pVCpu);
11148	AssertRCReturn(rc, rc);
11149	pVCpu->hm.s.vmx.fMergedNstGstCtls = true;
11150	}
11151	#endif
11152
11153	/*
11154	* Evaluate events to be injected into the guest.
11155	*
11156	* Events in TRPM can be injected without inspecting the guest state.
11157	* If any new events (interrupts/NMI) are pending currently, we try to set up the
11158	* guest to cause a VM-exit the next time they are ready to receive the event.
11159	*
11160	* With nested-guests, evaluating pending events may cause VM-exits.
11161	*/
11162	if (TRPMHasTrap(pVCpu))
11163	hmR0VmxTrpmTrapToPendingEvent(pVCpu);
11164
11165	uint32_t fIntrState;
11166	rcStrict = hmR0VmxEvaluatePendingEvent(pVCpu, pVmxTransient, &fIntrState);
11167
11168	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11169	/*
11170	* While evaluating pending events if something failed (unlikely) or if we were
11171	* preparing to run a nested-guest but performed a nested-guest VM-exit, we should bail.
11172	*/
11173	if (rcStrict != VINF_SUCCESS)
11174	return rcStrict;
11175	if ( pVmxTransient->fIsNestedGuest
11176	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
11177	{
11178	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
11179	return VINF_VMX_VMEXIT;
11180	}
11181	#else
11182	Assert(rcStrict == VINF_SUCCESS);
11183	#endif
11184
11185	/*
11186	* Event injection may take locks (currently the PGM lock for real-on-v86 case) and thus
11187	* needs to be done with longjmps or interrupts + preemption enabled. Event injection might
11188	* also result in triple-faulting the VM.
11189	*
11190	* With nested-guests, the above does not apply since unrestricted guest execution is a
11191	* requirement. Regardless, we do this here to avoid duplicating code elsewhere.
11192	*/
11193	rcStrict = hmR0VmxInjectPendingEvent(pVCpu, pVmxTransient, fIntrState, fStepping);
11194	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
11195	{ /* likely */ }
11196	else
11197	{
11198	AssertMsg(rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
11199	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
11200	return rcStrict;
11201	}
11202
11203	/*
11204	* A longjump might result in importing CR3 even for VM-exits that don't necessarily
11205	* import CR3 themselves. We will need to update them here, as even as late as the above
11206	* hmR0VmxInjectPendingEvent() call may lazily import guest-CPU state on demand causing
11207	* the below force flags to be set.
11208	*/
11209	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
11210	{
11211	Assert(!(ASMAtomicUoReadU64(&pVCpu->cpum.GstCtx.fExtrn) & CPUMCTX_EXTRN_CR3));
11212	int rc2 = PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
11213	AssertMsgReturn(rc2 == VINF_SUCCESS \|\| rc2 == VINF_PGM_SYNC_CR3,
11214	("%Rrc\n", rc2), RT_FAILURE_NP(rc2) ? rc2 : VERR_IPE_UNEXPECTED_INFO_STATUS);
11215	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
11216	}
11217	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
11218	{
11219	PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
11220	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
11221	}
11222
11223	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11224	/* Paranoia. */
11225	Assert(!pVmxTransient->fIsNestedGuest \|\| CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
11226	#endif
11227
11228	/*
11229	* No longjmps to ring-3 from this point on!!!
11230	* Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
11231	* This also disables flushing of the R0-logger instance (if any).
11232	*/
11233	VMMRZCallRing3Disable(pVCpu);
11234
11235	/*
11236	* Export the guest state bits.
11237	*
11238	* We cannot perform longjmps while loading the guest state because we do not preserve the
11239	* host/guest state (although the VMCS will be preserved) across longjmps which can cause
11240	* CPU migration.
11241	*
11242	* If we are injecting events to a real-on-v86 mode guest, we would have updated RIP and some segment
11243	* registers. Hence, exporting of the guest state needs to be done -after- injection of events.
11244	*/
11245	rcStrict = hmR0VmxExportGuestStateOptimal(pVCpu, pVmxTransient);
11246	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
11247	{ /* likely */ }
11248	else
11249	{
11250	VMMRZCallRing3Enable(pVCpu);
11251	return rcStrict;
11252	}
11253
11254	/*
11255	* We disable interrupts so that we don't miss any interrupts that would flag preemption
11256	* (IPI/timers etc.) when thread-context hooks aren't used and we've been running with
11257	* preemption disabled for a while. Since this is purely to aid the
11258	* RTThreadPreemptIsPending() code, it doesn't matter that it may temporarily reenable and
11259	* disable interrupt on NT.
11260	*
11261	* We need to check for force-flags that could've possible been altered since we last
11262	* checked them (e.g. by PDMGetInterrupt() leaving the PDM critical section,
11263	* see @bugref{6398}).
11264	*
11265	* We also check a couple of other force-flags as a last opportunity to get the EMT back
11266	* to ring-3 before executing guest code.
11267	*/
11268	pVmxTransient->fEFlags = ASMIntDisableFlags();
11269
11270	if ( ( !VM_FF_IS_ANY_SET(pVM, VM_FF_EMT_RENDEZVOUS \| VM_FF_TM_VIRTUAL_SYNC)
11271	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
11272	\|\| ( fStepping /* Optimized for the non-stepping case, so a bit of unnecessary work when stepping. */
11273	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK & ~(VMCPU_FF_TIMER \| VMCPU_FF_PDM_CRITSECT))) )
11274	{
11275	if (!RTThreadPreemptIsPending(NIL_RTTHREAD))
11276	{
11277	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11278	/*
11279	* If we are executing a nested-guest make sure that we should intercept subsequent
11280	* events. The one we are injecting might be part of VM-entry. This is mainly to keep
11281	* the VM-exit instruction emulation happy.
11282	*/
11283	if (pVmxTransient->fIsNestedGuest)
11284	pVCpu->cpum.GstCtx.hwvirt.vmx.fInterceptEvents = true;
11285	#endif
11286
11287	/*
11288	* We've injected any pending events. This is really the point of no return (to ring-3).
11289	*
11290	* Note! The caller expects to continue with interrupts & longjmps disabled on successful
11291	* returns from this function, so do -not- enable them here.
11292	*/
11293	pVCpu->hm.s.Event.fPending = false;
11294	return VINF_SUCCESS;
11295	}
11296
11297	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPendingHostIrq);
11298	rcStrict = VINF_EM_RAW_INTERRUPT;
11299	}
11300	else
11301	{
11302	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
11303	rcStrict = VINF_EM_RAW_TO_R3;
11304	}
11305
11306	ASMSetFlags(pVmxTransient->fEFlags);
11307	VMMRZCallRing3Enable(pVCpu);
11308
11309	return rcStrict;
11310	}
11311
11312
11313	/**
11314	* Final preparations before executing guest code using hardware-assisted VMX.
11315	*
11316	* We can no longer get preempted to a different host CPU and there are no returns
11317	* to ring-3. We ignore any errors that may happen from this point (e.g. VMWRITE
11318	* failures), this function is not intended to fail sans unrecoverable hardware
11319	* errors.
11320	*
11321	* @param pVCpu The cross context virtual CPU structure.
11322	* @param pVmxTransient The VMX-transient structure.
11323	*
11324	* @remarks Called with preemption disabled.
11325	* @remarks No-long-jump zone!!!
11326	*/
11327	static void hmR0VmxPreRunGuestCommitted(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
11328	{
11329	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
11330	Assert(VMMR0IsLogFlushDisabled(pVCpu));
11331	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
11332	Assert(!pVCpu->hm.s.Event.fPending);
11333
11334	/*
11335	* Indicate start of guest execution and where poking EMT out of guest-context is recognized.
11336	*/
11337	VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
11338	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC);
11339
11340	PVM pVM = pVCpu->CTX_SUFF(pVM);
11341	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11342	PHMPHYSCPU pHostCpu = hmR0GetCurrentCpu();
11343	RTCPUID const idCurrentCpu = pHostCpu->idCpu;
11344
11345	if (!CPUMIsGuestFPUStateActive(pVCpu))
11346	{
11347	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestFpuState, x);
11348	if (CPUMR0LoadGuestFPU(pVM, pVCpu) == VINF_CPUM_HOST_CR0_MODIFIED)
11349	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_HOST_CONTEXT;
11350	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestFpuState, x);
11351	STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadGuestFpu);
11352	}
11353
11354	/*
11355	* Re-export the host state bits as we may've been preempted (only happens when
11356	* thread-context hooks are used or when the VM start function changes) or if
11357	* the host CR0 is modified while loading the guest FPU state above.
11358	*
11359	* The 64-on-32 switcher saves the (64-bit) host state into the VMCS and if we
11360	* changed the switcher back to 32-bit, we must save the 32-bit host state here,
11361	* see @bugref{8432}.
11362	*
11363	* This may also happen when switching to/from a nested-guest VMCS without leaving
11364	* ring-0.
11365	*/
11366	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT)
11367	{
11368	hmR0VmxExportHostState(pVCpu);
11369	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportHostState);
11370	}
11371	Assert(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT));
11372
11373	/*
11374	* Export the state shared between host and guest (FPU, debug, lazy MSRs).
11375	*/
11376	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE)
11377	hmR0VmxExportSharedState(pVCpu, pVmxTransient);
11378	AssertMsg(!pVCpu->hm.s.fCtxChanged, ("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
11379
11380	/*
11381	* Store status of the shared guest/host debug state at the time of VM-entry.
11382	*/
11383	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
11384	if (CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
11385	{
11386	pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActivePending(pVCpu);
11387	pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActivePending(pVCpu);
11388	}
11389	else
11390	#endif
11391	{
11392	pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
11393	pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
11394	}
11395
11396	/*
11397	* Always cache the TPR-shadow if the virtual-APIC page exists, thereby skipping
11398	* more than one conditional check. The post-run side of our code shall determine
11399	* if it needs to sync. the virtual APIC TPR with the TPR-shadow.
11400	*/
11401	if (pVmcsInfo->pbVirtApic)
11402	pVmxTransient->u8GuestTpr = pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR];
11403
11404	/*
11405	* Update the host MSRs values in the VM-exit MSR-load area.
11406	*/
11407	if (!pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs)
11408	{
11409	if (pVmcsInfo->cExitMsrLoad > 0)
11410	hmR0VmxUpdateAutoLoadHostMsrs(pVCpu, pVmcsInfo);
11411	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = true;
11412	}
11413
11414	/*
11415	* Evaluate if we need to intercept guest RDTSC/P accesses. Set up the
11416	* VMX-preemption timer based on the next virtual sync clock deadline.
11417	*/
11418	if ( !pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer
11419	\|\| idCurrentCpu != pVCpu->hm.s.idLastCpu)
11420	{
11421	hmR0VmxUpdateTscOffsettingAndPreemptTimer(pVCpu, pVmxTransient);
11422	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = true;
11423	}
11424
11425	/* Record statistics of how often we use TSC offsetting as opposed to intercepting RDTSC/P. */
11426	bool const fIsRdtscIntercepted = RT_BOOL(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_RDTSC_EXIT);
11427	if (!fIsRdtscIntercepted)
11428	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
11429	else
11430	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
11431
11432	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB flushing, set this across the world switch. */
11433	hmR0VmxFlushTaggedTlb(pHostCpu, pVCpu, pVmcsInfo); /* Invalidate the appropriate guest entries from the TLB. */
11434	Assert(idCurrentCpu == pVCpu->hm.s.idLastCpu);
11435	pVCpu->hm.s.vmx.LastError.idCurrentCpu = idCurrentCpu; /* Update the error reporting info. with the current host CPU. */
11436	pVmcsInfo->idHostCpu = idCurrentCpu; /* Update the CPU for which we updated host-state in this VMCS. */
11437
11438	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
11439
11440	TMNotifyStartOfExecution(pVCpu); /* Notify TM to resume its clocks when TSC is tied to execution,
11441	as we're about to start executing the guest . */
11442
11443	/*
11444	* Load the guest TSC_AUX MSR when we are not intercepting RDTSCP.
11445	*
11446	* This is done this late as updating the TSC offsetting/preemption timer above
11447	* figures out if we can skip intercepting RDTSCP by calculating the number of
11448	* host CPU ticks till the next virtual sync deadline (for the dynamic case).
11449	*/
11450	if ( (pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_RDTSCP)
11451	&& !fIsRdtscIntercepted)
11452	{
11453	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_TSC_AUX);
11454
11455	/* NB: Because we call hmR0VmxAddAutoLoadStoreMsr with fUpdateHostMsr=true,
11456	it's safe even after hmR0VmxUpdateAutoLoadHostMsrs has already been done. */
11457	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_TSC_AUX, CPUMGetGuestTscAux(pVCpu),
11458	true /* fSetReadWrite /, true / fUpdateHostMsr */);
11459	AssertRC(rc);
11460	Assert(!pVmxTransient->fRemoveTscAuxMsr);
11461	pVmxTransient->fRemoveTscAuxMsr = true;
11462	}
11463
11464	#ifdef VBOX_STRICT
11465	Assert(pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs);
11466	hmR0VmxCheckAutoLoadStoreMsrs(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest);
11467	hmR0VmxCheckHostEferMsr(pVCpu, pVmcsInfo);
11468	AssertRC(hmR0VmxCheckVmcsCtls(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest));
11469	#endif
11470
11471	#ifdef HMVMX_ALWAYS_CHECK_GUEST_STATE
11472	/** @todo r=ramshankar: We can now probably use iemVmxVmentryCheckGuestState here.
11473	* Add a PVMXMSRS parameter to it, so that IEM can look at the host MSRs,
11474	* see @bugref{9180#c54}. */
11475	uint32_t const uInvalidReason = hmR0VmxCheckGuestState(pVCpu, pVmcsInfo);
11476	if (uInvalidReason != VMX_IGS_REASON_NOT_FOUND)
11477	Log4(("hmR0VmxCheckGuestState returned %#x\n", uInvalidReason));
11478	#endif
11479	}
11480
11481
11482	/**
11483	* First C routine invoked after running guest code using hardware-assisted VMX.
11484	*
11485	* @param pVCpu The cross context virtual CPU structure.
11486	* @param pVmxTransient The VMX-transient structure.
11487	* @param rcVMRun Return code of VMLAUNCH/VMRESUME.
11488	*
11489	* @remarks Called with interrupts disabled, and returns with interrupts enabled!
11490	*
11491	* @remarks No-long-jump zone!!! This function will however re-enable longjmps
11492	* unconditionally when it is safe to do so.
11493	*/
11494	static void hmR0VmxPostRunGuest(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, int rcVMRun)
11495	{
11496	uint64_t const uHostTsc = ASMReadTSC(); /** @todo We can do a lot better here, see @bugref{9180#c38}. */
11497
11498	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB flushing. */
11499	ASMAtomicIncU32(&pVCpu->hm.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for EMT poking. */
11500	pVCpu->hm.s.fCtxChanged = 0; /* Exits/longjmps to ring-3 requires saving the guest state. */
11501	pVmxTransient->fVmcsFieldsRead = 0; /* Transient fields need to be read from the VMCS. */
11502	pVmxTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
11503	pVmxTransient->fVectoringDoublePF = false; /* Vectoring double page-fault needs to be determined later. */
11504
11505	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11506	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_RDTSC_EXIT))
11507	{
11508	uint64_t uGstTsc;
11509	if (!pVmxTransient->fIsNestedGuest)
11510	uGstTsc = uHostTsc + pVmcsInfo->u64TscOffset;
11511	else
11512	{
11513	uint64_t const uNstGstTsc = uHostTsc + pVmcsInfo->u64TscOffset;
11514	uGstTsc = CPUMRemoveNestedGuestTscOffset(pVCpu, uNstGstTsc);
11515	}
11516	TMCpuTickSetLastSeen(pVCpu, uGstTsc); /* Update TM with the guest TSC. */
11517	}
11518
11519	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatPreExit, x);
11520	TMNotifyEndOfExecution(pVCpu); /* Notify TM that the guest is no longer running. */
11521	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
11522
11523	#if HC_ARCH_BITS == 64
11524	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_REQUIRED; /* Some host state messed up by VMX needs restoring. */
11525	#endif
11526	#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
11527	/* The 64-on-32 switcher maintains VMCS-launch state on its own
11528	and we need to leave it alone here. */
11529	if (pVmcsInfo->pfnStartVM != VMXR0SwitcherStartVM64)
11530	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
11531	#else
11532	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
11533	#endif
11534	#ifdef VBOX_STRICT
11535	hmR0VmxCheckHostEferMsr(pVCpu, pVmcsInfo); /* Verify that the host EFER MSR wasn't modified. */
11536	#endif
11537	Assert(!ASMIntAreEnabled());
11538	ASMSetFlags(pVmxTransient->fEFlags); /* Enable interrupts. */
11539	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
11540
11541	#ifdef HMVMX_ALWAYS_CLEAN_TRANSIENT
11542	/*
11543	* Clean all the VMCS fields in the transient structure before reading
11544	* anything from the VMCS.
11545	*/
11546	pVmxTransient->uExitReason = 0;
11547	pVmxTransient->uExitIntErrorCode = 0;
11548	pVmxTransient->uExitQual = 0;
11549	pVmxTransient->uGuestLinearAddr = 0;
11550	pVmxTransient->uExitIntInfo = 0;
11551	pVmxTransient->cbInstr = 0;
11552	pVmxTransient->ExitInstrInfo.u = 0;
11553	pVmxTransient->uEntryIntInfo = 0;
11554	pVmxTransient->uEntryXcptErrorCode = 0;
11555	pVmxTransient->cbEntryInstr = 0;
11556	pVmxTransient->uIdtVectoringInfo = 0;
11557	pVmxTransient->uIdtVectoringErrorCode = 0;
11558	#endif
11559
11560	/*
11561	* Save the basic VM-exit reason and check if the VM-entry failed.
11562	* See Intel spec. 24.9.1 "Basic VM-exit Information".
11563	*/
11564	uint32_t uExitReason;
11565	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
11566	AssertRC(rc);
11567	pVmxTransient->uExitReason = VMX_EXIT_REASON_BASIC(uExitReason);
11568	pVmxTransient->fVMEntryFailed = VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason);
11569
11570	/*
11571	* Log the VM-exit before logging anything else as otherwise it might be a
11572	* tad confusing what happens before and after the world-switch.
11573	*/
11574	HMVMX_LOG_EXIT(pVCpu, uExitReason);
11575
11576	/*
11577	* Remove the TSC_AUX MSR from the auto-load/store MSR area and reset any MSR
11578	* bitmap permissions, if it was added before VM-entry.
11579	*/
11580	if (pVmxTransient->fRemoveTscAuxMsr)
11581	{
11582	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_TSC_AUX);
11583	pVmxTransient->fRemoveTscAuxMsr = false;
11584	}
11585
11586	/*
11587	* Check if VMLAUNCH/VMRESUME succeeded.
11588	* If this failed, we cause a guru meditation and cease further execution.
11589	*
11590	* However, if we are executing a nested-guest we might fail if we use the
11591	* fast path rather than fully emulating VMLAUNCH/VMRESUME instruction in IEM.
11592	*/
11593	if (RT_LIKELY(rcVMRun == VINF_SUCCESS))
11594	{
11595	/*
11596	* Update the VM-exit history array here even if the VM-entry failed due to:
11597	* - Invalid guest state.
11598	* - MSR loading.
11599	* - Machine-check event.
11600	*
11601	* In any of the above cases we will still have a "valid" VM-exit reason
11602	* despite @a fVMEntryFailed being false.
11603	*
11604	* See Intel spec. 26.7 "VM-Entry failures during or after loading guest state".
11605	*
11606	* Note! We don't have CS or RIP at this point. Will probably address that later
11607	* by amending the history entry added here.
11608	*/
11609	EMHistoryAddExit(pVCpu, EMEXIT_MAKE_FT(EMEXIT_F_KIND_VMX, pVmxTransient->uExitReason & EMEXIT_F_TYPE_MASK),
11610	UINT64_MAX, uHostTsc);
11611
11612	if (RT_LIKELY(!pVmxTransient->fVMEntryFailed))
11613	{
11614	VMMRZCallRing3Enable(pVCpu);
11615
11616	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
11617	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
11618
11619	#ifdef HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
11620	rc = hmR0VmxReadAllRoFieldsVmcs(pVCpu, pVmxTransient);
11621	AssertRC(rc);
11622	#endif
11623	#if defined(HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE) \|\| defined(HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE)
11624	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
11625	AssertRC(rc);
11626	#elif defined(HMVMX_ALWAYS_SAVE_GUEST_RFLAGS)
11627	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_RFLAGS);
11628	AssertRC(rc);
11629	#else
11630	/*
11631	* Import the guest-interruptibility state always as we need it while evaluating
11632	* injecting events on re-entry.
11633	*
11634	* We don't import CR0 (when unrestricted guest execution is unavailable) despite
11635	* checking for real-mode while exporting the state because all bits that cause
11636	* mode changes wrt CR0 are intercepted.
11637	*/
11638	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_HM_VMX_INT_STATE);
11639	AssertRC(rc);
11640	#endif
11641
11642	/*
11643	* Sync the TPR shadow with our APIC state.
11644	*/
11645	if ( !pVmxTransient->fIsNestedGuest
11646	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW))
11647	{
11648	Assert(pVmcsInfo->pbVirtApic);
11649	if (pVmxTransient->u8GuestTpr != pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR])
11650	{
11651	rc = APICSetTpr(pVCpu, pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR]);
11652	AssertRC(rc);
11653	ASMAtomicOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
11654	}
11655	}
11656
11657	Assert(VMMRZCallRing3IsEnabled(pVCpu));
11658	return;
11659	}
11660	}
11661	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11662	else if (pVmxTransient->fIsNestedGuest)
11663	AssertMsgFailed(("VMLAUNCH/VMRESUME failed but shouldn't happen when VMLAUNCH/VMRESUME was emulated in IEM!\n"));
11664	#endif
11665	else
11666	Log4Func(("VM-entry failure: rcVMRun=%Rrc fVMEntryFailed=%RTbool\n", rcVMRun, pVmxTransient->fVMEntryFailed));
11667
11668	VMMRZCallRing3Enable(pVCpu);
11669	}
11670
11671
11672	/**
11673	* Runs the guest code using hardware-assisted VMX the normal way.
11674	*
11675	* @returns VBox status code.
11676	* @param pVCpu The cross context virtual CPU structure.
11677	* @param pcLoops Pointer to the number of executed loops.
11678	*/
11679	static VBOXSTRICTRC hmR0VmxRunGuestCodeNormal(PVMCPU pVCpu, uint32_t *pcLoops)
11680	{
11681	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hm.s.cMaxResumeLoops;
11682	Assert(pcLoops);
11683	Assert(*pcLoops <= cMaxResumeLoops);
11684	Assert(!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
11685
11686	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11687	/*
11688	* Switch to the guest VMCS as we may have transitioned from executing the nested-guest
11689	* without leaving ring-0. Otherwise, if we came from ring-3 we would have loaded the
11690	* guest VMCS while entering the VMX ring-0 session.
11691	*/
11692	if (pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs)
11693	{
11694	int rc = hmR0VmxSwitchToGstOrNstGstVmcs(pVCpu, false /* fSwitchToNstGstVmcs */);
11695	if (RT_SUCCESS(rc))
11696	{ /* likely */ }
11697	else
11698	{
11699	LogRelFunc(("Failed to switch to the guest VMCS. rc=%Rrc\n", rc));
11700	return rc;
11701	}
11702	}
11703	#endif
11704
11705	VMXTRANSIENT VmxTransient;
11706	RT_ZERO(VmxTransient);
11707	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
11708
11709	/* Paranoia. */
11710	Assert(VmxTransient.pVmcsInfo == &pVCpu->hm.s.vmx.VmcsInfo);
11711
11712	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
11713	for (;;)
11714	{
11715	Assert(!HMR0SuspendPending());
11716	HMVMX_ASSERT_CPU_SAFE(pVCpu);
11717	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
11718
11719	/*
11720	* Preparatory work for running nested-guest code, this may force us to
11721	* return to ring-3.
11722	*
11723	* Warning! This bugger disables interrupts on VINF_SUCCESS!
11724	*/
11725	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, false /* fStepping */);
11726	if (rcStrict != VINF_SUCCESS)
11727	break;
11728
11729	/* Interrupts are disabled at this point! */
11730	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
11731	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
11732	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
11733	/* Interrupts are re-enabled at this point! */
11734
11735	/*
11736	* Check for errors with running the VM (VMLAUNCH/VMRESUME).
11737	*/
11738	if (RT_SUCCESS(rcRun))
11739	{ /* very likely */ }
11740	else
11741	{
11742	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
11743	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
11744	return rcRun;
11745	}
11746
11747	/*
11748	* Profile the VM-exit.
11749	*/
11750	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
11751	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
11752	STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
11753	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
11754	HMVMX_START_EXIT_DISPATCH_PROF();
11755
11756	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
11757
11758	/*
11759	* Handle the VM-exit.
11760	*/
11761	#ifdef HMVMX_USE_FUNCTION_TABLE
11762	rcStrict = g_apfnVMExitHandlers[VmxTransient.uExitReason](pVCpu, &VmxTransient);
11763	#else
11764	rcStrict = hmR0VmxHandleExit(pVCpu, &VmxTransient);
11765	#endif
11766	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
11767	if (rcStrict == VINF_SUCCESS)
11768	{
11769	if (++(*pcLoops) <= cMaxResumeLoops)
11770	continue;
11771	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
11772	rcStrict = VINF_EM_RAW_INTERRUPT;
11773	}
11774	break;
11775	}
11776
11777	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
11778	return rcStrict;
11779	}
11780
11781
11782	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11783	/**
11784	* Runs the nested-guest code using hardware-assisted VMX.
11785	*
11786	* @returns VBox status code.
11787	* @param pVCpu The cross context virtual CPU structure.
11788	* @param pcLoops Pointer to the number of executed loops.
11789	*
11790	* @sa hmR0VmxRunGuestCodeNormal.
11791	*/
11792	static VBOXSTRICTRC hmR0VmxRunGuestCodeNested(PVMCPU pVCpu, uint32_t *pcLoops)
11793	{
11794	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hm.s.cMaxResumeLoops;
11795	Assert(pcLoops);
11796	Assert(*pcLoops <= cMaxResumeLoops);
11797	Assert(CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
11798
11799	/*
11800	* Switch to the nested-guest VMCS as we may have transitioned from executing the
11801	* guest without leaving ring-0. Otherwise, if we came from ring-3 we would have
11802	* loaded the nested-guest VMCS while entering the VMX ring-0 session.
11803	*/
11804	if (!pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs)
11805	{
11806	int rc = hmR0VmxSwitchToGstOrNstGstVmcs(pVCpu, true /* fSwitchToNstGstVmcs */);
11807	if (RT_SUCCESS(rc))
11808	{ /* likely */ }
11809	else
11810	{
11811	LogRelFunc(("Failed to switch to the nested-guest VMCS. rc=%Rrc\n", rc));
11812	return rc;
11813	}
11814	}
11815
11816	VMXTRANSIENT VmxTransient;
11817	RT_ZERO(VmxTransient);
11818	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
11819	VmxTransient.fIsNestedGuest = true;
11820
11821	/* Paranoia. */
11822	Assert(VmxTransient.pVmcsInfo == &pVCpu->hm.s.vmx.VmcsInfoNstGst);
11823
11824	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
11825	for (;;)
11826	{
11827	Assert(!HMR0SuspendPending());
11828	HMVMX_ASSERT_CPU_SAFE(pVCpu);
11829	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
11830
11831	/*
11832	* Preparatory work for running guest code, this may force us to
11833	* return to ring-3.
11834	*
11835	* Warning! This bugger disables interrupts on VINF_SUCCESS!
11836	*/
11837	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, false /* fStepping */);
11838	if (rcStrict != VINF_SUCCESS)
11839	break;
11840
11841	/* Interrupts are disabled at this point! */
11842	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
11843	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
11844	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
11845	/* Interrupts are re-enabled at this point! */
11846
11847	/*
11848	* Check for errors with running the VM (VMLAUNCH/VMRESUME).
11849	*/
11850	if (RT_SUCCESS(rcRun))
11851	{ /* very likely */ }
11852	else
11853	{
11854	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
11855	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
11856	return rcRun;
11857	}
11858
11859	/*
11860	* Profile the VM-exit.
11861	*/
11862	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
11863	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
11864	STAM_COUNTER_INC(&pVCpu->hm.s.StatNestedExitAll);
11865	STAM_COUNTER_INC(&pVCpu->hm.s.paStatNestedExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
11866	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
11867	HMVMX_START_EXIT_DISPATCH_PROF();
11868
11869	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
11870
11871	/*
11872	* Handle the VM-exit.
11873	*/
11874	rcStrict = hmR0VmxHandleExitNested(pVCpu, &VmxTransient);
11875	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
11876	if (rcStrict == VINF_SUCCESS)
11877	{
11878	if (!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
11879	{
11880	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
11881	rcStrict = VINF_VMX_VMEXIT;
11882	}
11883	else
11884	{
11885	if (++(*pcLoops) <= cMaxResumeLoops)
11886	continue;
11887	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
11888	rcStrict = VINF_EM_RAW_INTERRUPT;
11889	}
11890	}
11891	else
11892	Assert(rcStrict != VINF_VMX_VMEXIT);
11893	break;
11894	}
11895
11896	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
11897	return rcStrict;
11898	}
11899	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
11900
11901
11902	/** @name Execution loop for single stepping, DBGF events and expensive Dtrace
11903	* probes.
11904	*
11905	* The following few functions and associated structure contains the bloat
11906	* necessary for providing detailed debug events and dtrace probes as well as
11907	* reliable host side single stepping. This works on the principle of
11908	* "subclassing" the normal execution loop and workers. We replace the loop
11909	* method completely and override selected helpers to add necessary adjustments
11910	* to their core operation.
11911	*
11912	* The goal is to keep the "parent" code lean and mean, so as not to sacrifice
11913	* any performance for debug and analysis features.
11914	*
11915	* @{
11916	*/
11917
11918	/**
11919	* Transient per-VCPU debug state of VMCS and related info. we save/restore in
11920	* the debug run loop.
11921	*/
11922	typedef struct VMXRUNDBGSTATE
11923	{
11924	/** The RIP we started executing at. This is for detecting that we stepped. */
11925	uint64_t uRipStart;
11926	/** The CS we started executing with. */
11927	uint16_t uCsStart;
11928
11929	/** Whether we've actually modified the 1st execution control field. */
11930	bool fModifiedProcCtls : 1;
11931	/** Whether we've actually modified the 2nd execution control field. */
11932	bool fModifiedProcCtls2 : 1;
11933	/** Whether we've actually modified the exception bitmap. */
11934	bool fModifiedXcptBitmap : 1;
11935
11936	/** We desire the modified the CR0 mask to be cleared. */
11937	bool fClearCr0Mask : 1;
11938	/** We desire the modified the CR4 mask to be cleared. */
11939	bool fClearCr4Mask : 1;
11940	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC. */
11941	uint32_t fCpe1Extra;
11942	/** Stuff we do not want in VMX_VMCS32_CTRL_PROC_EXEC. */
11943	uint32_t fCpe1Unwanted;
11944	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC2. */
11945	uint32_t fCpe2Extra;
11946	/** Extra stuff we need in VMX_VMCS32_CTRL_EXCEPTION_BITMAP. */
11947	uint32_t bmXcptExtra;
11948	/** The sequence number of the Dtrace provider settings the state was
11949	* configured against. */
11950	uint32_t uDtraceSettingsSeqNo;
11951	/** VM-exits to check (one bit per VM-exit). */
11952	uint32_t bmExitsToCheck[3];
11953
11954	/** The initial VMX_VMCS32_CTRL_PROC_EXEC value (helps with restore). */
11955	uint32_t fProcCtlsInitial;
11956	/** The initial VMX_VMCS32_CTRL_PROC_EXEC2 value (helps with restore). */
11957	uint32_t fProcCtls2Initial;
11958	/** The initial VMX_VMCS32_CTRL_EXCEPTION_BITMAP value (helps with restore). */
11959	uint32_t bmXcptInitial;
11960	} VMXRUNDBGSTATE;
11961	AssertCompileMemberSize(VMXRUNDBGSTATE, bmExitsToCheck, (VMX_EXIT_MAX + 1 + 31) / 32 * 4);
11962	typedef VMXRUNDBGSTATE *PVMXRUNDBGSTATE;
11963
11964
11965	/**
11966	* Initializes the VMXRUNDBGSTATE structure.
11967	*
11968	* @param pVCpu The cross context virtual CPU structure of the
11969	* calling EMT.
11970	* @param pVmxTransient The VMX-transient structure.
11971	* @param pDbgState The debug state to initialize.
11972	*/
11973	static void hmR0VmxRunDebugStateInit(PVMCPU pVCpu, PCVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11974	{
11975	pDbgState->uRipStart = pVCpu->cpum.GstCtx.rip;
11976	pDbgState->uCsStart = pVCpu->cpum.GstCtx.cs.Sel;
11977
11978	pDbgState->fModifiedProcCtls = false;
11979	pDbgState->fModifiedProcCtls2 = false;
11980	pDbgState->fModifiedXcptBitmap = false;
11981	pDbgState->fClearCr0Mask = false;
11982	pDbgState->fClearCr4Mask = false;
11983	pDbgState->fCpe1Extra = 0;
11984	pDbgState->fCpe1Unwanted = 0;
11985	pDbgState->fCpe2Extra = 0;
11986	pDbgState->bmXcptExtra = 0;
11987	pDbgState->fProcCtlsInitial = pVmxTransient->pVmcsInfo->u32ProcCtls;
11988	pDbgState->fProcCtls2Initial = pVmxTransient->pVmcsInfo->u32ProcCtls2;
11989	pDbgState->bmXcptInitial = pVmxTransient->pVmcsInfo->u32XcptBitmap;
11990	}
11991
11992
11993	/**
11994	* Updates the VMSC fields with changes requested by @a pDbgState.
11995	*
11996	* This is performed after hmR0VmxPreRunGuestDebugStateUpdate as well
11997	* immediately before executing guest code, i.e. when interrupts are disabled.
11998	* We don't check status codes here as we cannot easily assert or return in the
11999	* latter case.
12000	*
12001	* @param pVCpu The cross context virtual CPU structure.
12002	* @param pVmxTransient The VMX-transient structure.
12003	* @param pDbgState The debug state.
12004	*/
12005	static void hmR0VmxPreRunGuestDebugStateApply(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
12006	{
12007	/*
12008	* Ensure desired flags in VMCS control fields are set.
12009	* (Ignoring write failure here, as we're committed and it's just debug extras.)
12010	*
12011	* Note! We load the shadow CR0 & CR4 bits when we flag the clearing, so
12012	* there should be no stale data in pCtx at this point.
12013	*/
12014	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
12015	if ( (pVmcsInfo->u32ProcCtls & pDbgState->fCpe1Extra) != pDbgState->fCpe1Extra
12016	\|\| (pVmcsInfo->u32ProcCtls & pDbgState->fCpe1Unwanted))
12017	{
12018	pVmcsInfo->u32ProcCtls \|= pDbgState->fCpe1Extra;
12019	pVmcsInfo->u32ProcCtls &= ~pDbgState->fCpe1Unwanted;
12020	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
12021	Log6Func(("VMX_VMCS32_CTRL_PROC_EXEC: %#RX32\n", pVmcsInfo->u32ProcCtls));
12022	pDbgState->fModifiedProcCtls = true;
12023	}
12024
12025	if ((pVmcsInfo->u32ProcCtls2 & pDbgState->fCpe2Extra) != pDbgState->fCpe2Extra)
12026	{
12027	pVmcsInfo->u32ProcCtls2 \|= pDbgState->fCpe2Extra;
12028	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pVmcsInfo->u32ProcCtls2);
12029	Log6Func(("VMX_VMCS32_CTRL_PROC_EXEC2: %#RX32\n", pVmcsInfo->u32ProcCtls2));
12030	pDbgState->fModifiedProcCtls2 = true;
12031	}
12032
12033	if ((pVmcsInfo->u32XcptBitmap & pDbgState->bmXcptExtra) != pDbgState->bmXcptExtra)
12034	{
12035	pVmcsInfo->u32XcptBitmap \|= pDbgState->bmXcptExtra;
12036	VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVmcsInfo->u32XcptBitmap);
12037	Log6Func(("VMX_VMCS32_CTRL_EXCEPTION_BITMAP: %#RX32\n", pVmcsInfo->u32XcptBitmap));
12038	pDbgState->fModifiedXcptBitmap = true;
12039	}
12040
12041	if (pDbgState->fClearCr0Mask && pVmcsInfo->u64Cr0Mask != 0)
12042	{
12043	pVmcsInfo->u64Cr0Mask = 0;
12044	VMXWriteVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, 0);
12045	Log6Func(("VMX_VMCS_CTRL_CR0_MASK: 0\n"));
12046	}
12047
12048	if (pDbgState->fClearCr4Mask && pVmcsInfo->u64Cr4Mask != 0)
12049	{
12050	pVmcsInfo->u64Cr4Mask = 0;
12051	VMXWriteVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, 0);
12052	Log6Func(("VMX_VMCS_CTRL_CR4_MASK: 0\n"));
12053	}
12054
12055	NOREF(pVCpu);
12056	}
12057
12058
12059	/**
12060	* Restores VMCS fields that were changed by hmR0VmxPreRunGuestDebugStateApply for
12061	* re-entry next time around.
12062	*
12063	* @returns Strict VBox status code (i.e. informational status codes too).
12064	* @param pVCpu The cross context virtual CPU structure.
12065	* @param pVmxTransient The VMX-transient structure.
12066	* @param pDbgState The debug state.
12067	* @param rcStrict The return code from executing the guest using single
12068	* stepping.
12069	*/
12070	static VBOXSTRICTRC hmR0VmxRunDebugStateRevert(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState,
12071	VBOXSTRICTRC rcStrict)
12072	{
12073	/*
12074	* Restore VM-exit control settings as we may not reenter this function the
12075	* next time around.
12076	*/
12077	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
12078
12079	/* We reload the initial value, trigger what we can of recalculations the
12080	next time around. From the looks of things, that's all that's required atm. */
12081	if (pDbgState->fModifiedProcCtls)
12082	{
12083	if (!(pDbgState->fProcCtlsInitial & VMX_PROC_CTLS_MOV_DR_EXIT) && CPUMIsHyperDebugStateActive(pVCpu))
12084	pDbgState->fProcCtlsInitial \|= VMX_PROC_CTLS_MOV_DR_EXIT; /* Avoid assertion in hmR0VmxLeave */
12085	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pDbgState->fProcCtlsInitial);
12086	AssertRCReturn(rc2, rc2);
12087	pVmcsInfo->u32ProcCtls = pDbgState->fProcCtlsInitial;
12088	}
12089
12090	/* We're currently the only ones messing with this one, so just restore the
12091	cached value and reload the field. */
12092	if ( pDbgState->fModifiedProcCtls2
12093	&& pVmcsInfo->u32ProcCtls2 != pDbgState->fProcCtls2Initial)
12094	{
12095	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pDbgState->fProcCtls2Initial);
12096	AssertRCReturn(rc2, rc2);
12097	pVmcsInfo->u32ProcCtls2 = pDbgState->fProcCtls2Initial;
12098	}
12099
12100	/* If we've modified the exception bitmap, we restore it and trigger
12101	reloading and partial recalculation the next time around. */
12102	if (pDbgState->fModifiedXcptBitmap)
12103	pVmcsInfo->u32XcptBitmap = pDbgState->bmXcptInitial;
12104
12105	return rcStrict;
12106	}
12107
12108
12109	/**
12110	* Configures VM-exit controls for current DBGF and DTrace settings.
12111	*
12112	* This updates @a pDbgState and the VMCS execution control fields to reflect
12113	* the necessary VM-exits demanded by DBGF and DTrace.
12114	*
12115	* @param pVCpu The cross context virtual CPU structure.
12116	* @param pVmxTransient The VMX-transient structure. May update
12117	* fUpdatedTscOffsettingAndPreemptTimer.
12118	* @param pDbgState The debug state.
12119	*/
12120	static void hmR0VmxPreRunGuestDebugStateUpdate(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
12121	{
12122	/*
12123	* Take down the dtrace serial number so we can spot changes.
12124	*/
12125	pDbgState->uDtraceSettingsSeqNo = VBOXVMM_GET_SETTINGS_SEQ_NO();
12126	ASMCompilerBarrier();
12127
12128	/*
12129	* We'll rebuild most of the middle block of data members (holding the
12130	* current settings) as we go along here, so start by clearing it all.
12131	*/
12132	pDbgState->bmXcptExtra = 0;
12133	pDbgState->fCpe1Extra = 0;
12134	pDbgState->fCpe1Unwanted = 0;
12135	pDbgState->fCpe2Extra = 0;
12136	for (unsigned i = 0; i < RT_ELEMENTS(pDbgState->bmExitsToCheck); i++)
12137	pDbgState->bmExitsToCheck[i] = 0;
12138
12139	/*
12140	* Software interrupts (INT XXh) - no idea how to trigger these...
12141	*/
12142	PVM pVM = pVCpu->CTX_SUFF(pVM);
12143	if ( DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_INTERRUPT_SOFTWARE)
12144	\|\| VBOXVMM_INT_SOFTWARE_ENABLED())
12145	{
12146	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
12147	}
12148
12149	/*
12150	* INT3 breakpoints - triggered by #BP exceptions.
12151	*/
12152	if (pVM->dbgf.ro.cEnabledInt3Breakpoints > 0)
12153	pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
12154
12155	/*
12156	* Exception bitmap and XCPT events+probes.
12157	*/
12158	for (int iXcpt = 0; iXcpt < (DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST + 1); iXcpt++)
12159	if (DBGF_IS_EVENT_ENABLED(pVM, (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + iXcpt)))
12160	pDbgState->bmXcptExtra \|= RT_BIT_32(iXcpt);
12161
12162	if (VBOXVMM_XCPT_DE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DE);
12163	if (VBOXVMM_XCPT_DB_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DB);
12164	if (VBOXVMM_XCPT_BP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
12165	if (VBOXVMM_XCPT_OF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_OF);
12166	if (VBOXVMM_XCPT_BR_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BR);
12167	if (VBOXVMM_XCPT_UD_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_UD);
12168	if (VBOXVMM_XCPT_NM_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NM);
12169	if (VBOXVMM_XCPT_DF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DF);
12170	if (VBOXVMM_XCPT_TS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_TS);
12171	if (VBOXVMM_XCPT_NP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NP);
12172	if (VBOXVMM_XCPT_SS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SS);
12173	if (VBOXVMM_XCPT_GP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_GP);
12174	if (VBOXVMM_XCPT_PF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_PF);
12175	if (VBOXVMM_XCPT_MF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_MF);
12176	if (VBOXVMM_XCPT_AC_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_AC);
12177	if (VBOXVMM_XCPT_XF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_XF);
12178	if (VBOXVMM_XCPT_VE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_VE);
12179	if (VBOXVMM_XCPT_SX_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SX);
12180
12181	if (pDbgState->bmXcptExtra)
12182	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
12183
12184	/*
12185	* Process events and probes for VM-exits, making sure we get the wanted VM-exits.
12186	*
12187	* Note! This is the reverse of what hmR0VmxHandleExitDtraceEvents does.
12188	* So, when adding/changing/removing please don't forget to update it.
12189	*
12190	* Some of the macros are picking up local variables to save horizontal space,
12191	* (being able to see it in a table is the lesser evil here).
12192	*/
12193	#define IS_EITHER_ENABLED(a_pVM, a_EventSubName) \
12194	( DBGF_IS_EVENT_ENABLED(a_pVM, RT_CONCAT(DBGFEVENT_, a_EventSubName)) \
12195	\|\| RT_CONCAT3(VBOXVMM_, a_EventSubName, _ENABLED)() )
12196	#define SET_ONLY_XBM_IF_EITHER_EN(a_EventSubName, a_uExit) \
12197	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
12198	{ AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
12199	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
12200	} else do { } while (0)
12201	#define SET_CPE1_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec) \
12202	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
12203	{ \
12204	(pDbgState)->fCpe1Extra \|= (a_fCtrlProcExec); \
12205	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
12206	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
12207	} else do { } while (0)
12208	#define SET_CPEU_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fUnwantedCtrlProcExec) \
12209	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
12210	{ \
12211	(pDbgState)->fCpe1Unwanted \|= (a_fUnwantedCtrlProcExec); \
12212	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
12213	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
12214	} else do { } while (0)
12215	#define SET_CPE2_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec2) \
12216	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
12217	{ \
12218	(pDbgState)->fCpe2Extra \|= (a_fCtrlProcExec2); \
12219	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
12220	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
12221	} else do { } while (0)
12222
12223	SET_ONLY_XBM_IF_EITHER_EN(EXIT_TASK_SWITCH, VMX_EXIT_TASK_SWITCH); /* unconditional */
12224	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_VIOLATION, VMX_EXIT_EPT_VIOLATION); /* unconditional */
12225	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_MISCONFIG, VMX_EXIT_EPT_MISCONFIG); /* unconditional (unless #VE) */
12226	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_ACCESS, VMX_EXIT_APIC_ACCESS); /* feature dependent, nothing to enable here */
12227	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_WRITE, VMX_EXIT_APIC_WRITE); /* feature dependent, nothing to enable here */
12228
12229	SET_ONLY_XBM_IF_EITHER_EN(INSTR_CPUID, VMX_EXIT_CPUID); /* unconditional */
12230	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CPUID, VMX_EXIT_CPUID);
12231	SET_ONLY_XBM_IF_EITHER_EN(INSTR_GETSEC, VMX_EXIT_GETSEC); /* unconditional */
12232	SET_ONLY_XBM_IF_EITHER_EN( EXIT_GETSEC, VMX_EXIT_GETSEC);
12233	SET_CPE1_XBM_IF_EITHER_EN(INSTR_HALT, VMX_EXIT_HLT, VMX_PROC_CTLS_HLT_EXIT); /* paranoia */
12234	SET_ONLY_XBM_IF_EITHER_EN( EXIT_HALT, VMX_EXIT_HLT);
12235	SET_ONLY_XBM_IF_EITHER_EN(INSTR_INVD, VMX_EXIT_INVD); /* unconditional */
12236	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVD, VMX_EXIT_INVD);
12237	SET_CPE1_XBM_IF_EITHER_EN(INSTR_INVLPG, VMX_EXIT_INVLPG, VMX_PROC_CTLS_INVLPG_EXIT);
12238	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVLPG, VMX_EXIT_INVLPG);
12239	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDPMC, VMX_EXIT_RDPMC, VMX_PROC_CTLS_RDPMC_EXIT);
12240	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDPMC, VMX_EXIT_RDPMC);
12241	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSC, VMX_EXIT_RDTSC, VMX_PROC_CTLS_RDTSC_EXIT);
12242	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSC, VMX_EXIT_RDTSC);
12243	SET_ONLY_XBM_IF_EITHER_EN(INSTR_RSM, VMX_EXIT_RSM); /* unconditional */
12244	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RSM, VMX_EXIT_RSM);
12245	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMM_CALL, VMX_EXIT_VMCALL); /* unconditional */
12246	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMM_CALL, VMX_EXIT_VMCALL);
12247	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMCLEAR, VMX_EXIT_VMCLEAR); /* unconditional */
12248	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMCLEAR, VMX_EXIT_VMCLEAR);
12249	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH); /* unconditional */
12250	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH);
12251	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRLD, VMX_EXIT_VMPTRLD); /* unconditional */
12252	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRLD, VMX_EXIT_VMPTRLD);
12253	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRST, VMX_EXIT_VMPTRST); /* unconditional */
12254	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRST, VMX_EXIT_VMPTRST);
12255	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMREAD, VMX_EXIT_VMREAD); /* unconditional */
12256	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMREAD, VMX_EXIT_VMREAD);
12257	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMRESUME, VMX_EXIT_VMRESUME); /* unconditional */
12258	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMRESUME, VMX_EXIT_VMRESUME);
12259	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMWRITE, VMX_EXIT_VMWRITE); /* unconditional */
12260	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMWRITE, VMX_EXIT_VMWRITE);
12261	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXOFF, VMX_EXIT_VMXOFF); /* unconditional */
12262	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXOFF, VMX_EXIT_VMXOFF);
12263	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXON, VMX_EXIT_VMXON); /* unconditional */
12264	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXON, VMX_EXIT_VMXON);
12265
12266	if ( IS_EITHER_ENABLED(pVM, INSTR_CRX_READ)
12267	\|\| IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
12268	{
12269	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_CR4
12270	\| CPUMCTX_EXTRN_APIC_TPR);
12271	AssertRC(rc);
12272
12273	#if 0 /** @todo fix me */
12274	pDbgState->fClearCr0Mask = true;
12275	pDbgState->fClearCr4Mask = true;
12276	#endif
12277	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_READ))
12278	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_CR3_STORE_EXIT \| VMX_PROC_CTLS_CR8_STORE_EXIT;
12279	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
12280	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_CR3_LOAD_EXIT \| VMX_PROC_CTLS_CR8_LOAD_EXIT;
12281	pDbgState->fCpe1Unwanted \|= VMX_PROC_CTLS_USE_TPR_SHADOW; /* risky? */
12282	/* Note! We currently don't use VMX_VMCS32_CTRL_CR3_TARGET_COUNT. It would
12283	require clearing here and in the loop if we start using it. */
12284	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_CRX);
12285	}
12286	else
12287	{
12288	if (pDbgState->fClearCr0Mask)
12289	{
12290	pDbgState->fClearCr0Mask = false;
12291	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR0);
12292	}
12293	if (pDbgState->fClearCr4Mask)
12294	{
12295	pDbgState->fClearCr4Mask = false;
12296	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR4);
12297	}
12298	}
12299	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_READ, VMX_EXIT_MOV_CRX);
12300	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_WRITE, VMX_EXIT_MOV_CRX);
12301
12302	if ( IS_EITHER_ENABLED(pVM, INSTR_DRX_READ)
12303	\|\| IS_EITHER_ENABLED(pVM, INSTR_DRX_WRITE))
12304	{
12305	/** @todo later, need to fix handler as it assumes this won't usually happen. */
12306	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_DRX);
12307	}
12308	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_READ, VMX_EXIT_MOV_DRX);
12309	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_WRITE, VMX_EXIT_MOV_DRX);
12310
12311	SET_CPEU_XBM_IF_EITHER_EN(INSTR_RDMSR, VMX_EXIT_RDMSR, VMX_PROC_CTLS_USE_MSR_BITMAPS); /* risky clearing this? */
12312	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDMSR, VMX_EXIT_RDMSR);
12313	SET_CPEU_XBM_IF_EITHER_EN(INSTR_WRMSR, VMX_EXIT_WRMSR, VMX_PROC_CTLS_USE_MSR_BITMAPS);
12314	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WRMSR, VMX_EXIT_WRMSR);
12315	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MWAIT, VMX_EXIT_MWAIT, VMX_PROC_CTLS_MWAIT_EXIT); /* paranoia */
12316	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MWAIT, VMX_EXIT_MWAIT);
12317	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MONITOR, VMX_EXIT_MONITOR, VMX_PROC_CTLS_MONITOR_EXIT); /* paranoia */
12318	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MONITOR, VMX_EXIT_MONITOR);
12319	#if 0 /** @todo too slow, fix handler. */
12320	SET_CPE1_XBM_IF_EITHER_EN(INSTR_PAUSE, VMX_EXIT_PAUSE, VMX_PROC_CTLS_PAUSE_EXIT);
12321	#endif
12322	SET_ONLY_XBM_IF_EITHER_EN( EXIT_PAUSE, VMX_EXIT_PAUSE);
12323
12324	if ( IS_EITHER_ENABLED(pVM, INSTR_SGDT)
12325	\|\| IS_EITHER_ENABLED(pVM, INSTR_SIDT)
12326	\|\| IS_EITHER_ENABLED(pVM, INSTR_LGDT)
12327	\|\| IS_EITHER_ENABLED(pVM, INSTR_LIDT))
12328	{
12329	pDbgState->fCpe2Extra \|= VMX_PROC_CTLS2_DESC_TABLE_EXIT;
12330	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_GDTR_IDTR_ACCESS);
12331	}
12332	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SGDT, VMX_EXIT_GDTR_IDTR_ACCESS);
12333	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SIDT, VMX_EXIT_GDTR_IDTR_ACCESS);
12334	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LGDT, VMX_EXIT_GDTR_IDTR_ACCESS);
12335	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LIDT, VMX_EXIT_GDTR_IDTR_ACCESS);
12336
12337	if ( IS_EITHER_ENABLED(pVM, INSTR_SLDT)
12338	\|\| IS_EITHER_ENABLED(pVM, INSTR_STR)
12339	\|\| IS_EITHER_ENABLED(pVM, INSTR_LLDT)
12340	\|\| IS_EITHER_ENABLED(pVM, INSTR_LTR))
12341	{
12342	pDbgState->fCpe2Extra \|= VMX_PROC_CTLS2_DESC_TABLE_EXIT;
12343	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_LDTR_TR_ACCESS);
12344	}
12345	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SLDT, VMX_EXIT_LDTR_TR_ACCESS);
12346	SET_ONLY_XBM_IF_EITHER_EN( EXIT_STR, VMX_EXIT_LDTR_TR_ACCESS);
12347	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LLDT, VMX_EXIT_LDTR_TR_ACCESS);
12348	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LTR, VMX_EXIT_LDTR_TR_ACCESS);
12349
12350	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVEPT, VMX_EXIT_INVEPT); /* unconditional */
12351	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVEPT, VMX_EXIT_INVEPT);
12352	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSCP, VMX_EXIT_RDTSCP, VMX_PROC_CTLS_RDTSC_EXIT);
12353	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSCP, VMX_EXIT_RDTSCP);
12354	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVVPID, VMX_EXIT_INVVPID); /* unconditional */
12355	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVVPID, VMX_EXIT_INVVPID);
12356	SET_CPE2_XBM_IF_EITHER_EN(INSTR_WBINVD, VMX_EXIT_WBINVD, VMX_PROC_CTLS2_WBINVD_EXIT);
12357	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WBINVD, VMX_EXIT_WBINVD);
12358	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSETBV, VMX_EXIT_XSETBV); /* unconditional */
12359	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XSETBV, VMX_EXIT_XSETBV);
12360	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDRAND, VMX_EXIT_RDRAND, VMX_PROC_CTLS2_RDRAND_EXIT);
12361	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDRAND, VMX_EXIT_RDRAND);
12362	SET_CPE1_XBM_IF_EITHER_EN(INSTR_VMX_INVPCID, VMX_EXIT_INVPCID, VMX_PROC_CTLS_INVLPG_EXIT);
12363	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVPCID, VMX_EXIT_INVPCID);
12364	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMFUNC, VMX_EXIT_VMFUNC); /* unconditional for the current setup */
12365	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMFUNC, VMX_EXIT_VMFUNC);
12366	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDSEED, VMX_EXIT_RDSEED, VMX_PROC_CTLS2_RDSEED_EXIT);
12367	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDSEED, VMX_EXIT_RDSEED);
12368	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSAVES, VMX_EXIT_XSAVES); /* unconditional (enabled by host, guest cfg) */
12369	SET_ONLY_XBM_IF_EITHER_EN(EXIT_XSAVES, VMX_EXIT_XSAVES);
12370	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XRSTORS, VMX_EXIT_XRSTORS); /* unconditional (enabled by host, guest cfg) */
12371	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XRSTORS, VMX_EXIT_XRSTORS);
12372
12373	#undef IS_EITHER_ENABLED
12374	#undef SET_ONLY_XBM_IF_EITHER_EN
12375	#undef SET_CPE1_XBM_IF_EITHER_EN
12376	#undef SET_CPEU_XBM_IF_EITHER_EN
12377	#undef SET_CPE2_XBM_IF_EITHER_EN
12378
12379	/*
12380	* Sanitize the control stuff.
12381	*/
12382	pDbgState->fCpe2Extra &= pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1;
12383	if (pDbgState->fCpe2Extra)
12384	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_USE_SECONDARY_CTLS;
12385	pDbgState->fCpe1Extra &= pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1;
12386	pDbgState->fCpe1Unwanted &= ~pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed0;
12387	if (pVCpu->hm.s.fDebugWantRdTscExit != RT_BOOL(pDbgState->fCpe1Extra & VMX_PROC_CTLS_RDTSC_EXIT))
12388	{
12389	pVCpu->hm.s.fDebugWantRdTscExit ^= true;
12390	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
12391	}
12392
12393	Log6(("HM: debug state: cpe1=%#RX32 cpeu=%#RX32 cpe2=%#RX32%s%s\n",
12394	pDbgState->fCpe1Extra, pDbgState->fCpe1Unwanted, pDbgState->fCpe2Extra,
12395	pDbgState->fClearCr0Mask ? " clr-cr0" : "",
12396	pDbgState->fClearCr4Mask ? " clr-cr4" : ""));
12397	}
12398
12399
12400	/**
12401	* Fires off DBGF events and dtrace probes for a VM-exit, when it's
12402	* appropriate.
12403	*
12404	* The caller has checked the VM-exit against the
12405	* VMXRUNDBGSTATE::bmExitsToCheck bitmap. The caller has checked for NMIs
12406	* already, so we don't have to do that either.
12407	*
12408	* @returns Strict VBox status code (i.e. informational status codes too).
12409	* @param pVCpu The cross context virtual CPU structure.
12410	* @param pVmxTransient The VMX-transient structure.
12411	* @param uExitReason The VM-exit reason.
12412	*
12413	* @remarks The name of this function is displayed by dtrace, so keep it short
12414	* and to the point. No longer than 33 chars long, please.
12415	*/
12416	static VBOXSTRICTRC hmR0VmxHandleExitDtraceEvents(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uExitReason)
12417	{
12418	/*
12419	* Translate the event into a DBGF event (enmEvent + uEventArg) and at the
12420	* same time check whether any corresponding Dtrace event is enabled (fDtrace).
12421	*
12422	* Note! This is the reverse operation of what hmR0VmxPreRunGuestDebugStateUpdate
12423	* does. Must add/change/remove both places. Same ordering, please.
12424	*
12425	* Added/removed events must also be reflected in the next section
12426	* where we dispatch dtrace events.
12427	*/
12428	bool fDtrace1 = false;
12429	bool fDtrace2 = false;
12430	DBGFEVENTTYPE enmEvent1 = DBGFEVENT_END;
12431	DBGFEVENTTYPE enmEvent2 = DBGFEVENT_END;
12432	uint32_t uEventArg = 0;
12433	#define SET_EXIT(a_EventSubName) \
12434	do { \
12435	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
12436	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
12437	} while (0)
12438	#define SET_BOTH(a_EventSubName) \
12439	do { \
12440	enmEvent1 = RT_CONCAT(DBGFEVENT_INSTR_, a_EventSubName); \
12441	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
12442	fDtrace1 = RT_CONCAT3(VBOXVMM_INSTR_, a_EventSubName, _ENABLED)(); \
12443	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
12444	} while (0)
12445	switch (uExitReason)
12446	{
12447	case VMX_EXIT_MTF:
12448	return hmR0VmxExitMtf(pVCpu, pVmxTransient);
12449
12450	case VMX_EXIT_XCPT_OR_NMI:
12451	{
12452	uint8_t const idxVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
12453	switch (VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo))
12454	{
12455	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
12456	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
12457	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
12458	if (idxVector <= (unsigned)(DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST))
12459	{
12460	if (VMX_EXIT_INT_INFO_IS_ERROR_CODE_VALID(pVmxTransient->uExitIntInfo))
12461	{
12462	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
12463	uEventArg = pVmxTransient->uExitIntErrorCode;
12464	}
12465	enmEvent1 = (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + idxVector);
12466	switch (enmEvent1)
12467	{
12468	case DBGFEVENT_XCPT_DE: fDtrace1 = VBOXVMM_XCPT_DE_ENABLED(); break;
12469	case DBGFEVENT_XCPT_DB: fDtrace1 = VBOXVMM_XCPT_DB_ENABLED(); break;
12470	case DBGFEVENT_XCPT_BP: fDtrace1 = VBOXVMM_XCPT_BP_ENABLED(); break;
12471	case DBGFEVENT_XCPT_OF: fDtrace1 = VBOXVMM_XCPT_OF_ENABLED(); break;
12472	case DBGFEVENT_XCPT_BR: fDtrace1 = VBOXVMM_XCPT_BR_ENABLED(); break;
12473	case DBGFEVENT_XCPT_UD: fDtrace1 = VBOXVMM_XCPT_UD_ENABLED(); break;
12474	case DBGFEVENT_XCPT_NM: fDtrace1 = VBOXVMM_XCPT_NM_ENABLED(); break;
12475	case DBGFEVENT_XCPT_DF: fDtrace1 = VBOXVMM_XCPT_DF_ENABLED(); break;
12476	case DBGFEVENT_XCPT_TS: fDtrace1 = VBOXVMM_XCPT_TS_ENABLED(); break;
12477	case DBGFEVENT_XCPT_NP: fDtrace1 = VBOXVMM_XCPT_NP_ENABLED(); break;
12478	case DBGFEVENT_XCPT_SS: fDtrace1 = VBOXVMM_XCPT_SS_ENABLED(); break;
12479	case DBGFEVENT_XCPT_GP: fDtrace1 = VBOXVMM_XCPT_GP_ENABLED(); break;
12480	case DBGFEVENT_XCPT_PF: fDtrace1 = VBOXVMM_XCPT_PF_ENABLED(); break;
12481	case DBGFEVENT_XCPT_MF: fDtrace1 = VBOXVMM_XCPT_MF_ENABLED(); break;
12482	case DBGFEVENT_XCPT_AC: fDtrace1 = VBOXVMM_XCPT_AC_ENABLED(); break;
12483	case DBGFEVENT_XCPT_XF: fDtrace1 = VBOXVMM_XCPT_XF_ENABLED(); break;
12484	case DBGFEVENT_XCPT_VE: fDtrace1 = VBOXVMM_XCPT_VE_ENABLED(); break;
12485	case DBGFEVENT_XCPT_SX: fDtrace1 = VBOXVMM_XCPT_SX_ENABLED(); break;
12486	default: break;
12487	}
12488	}
12489	else
12490	AssertFailed();
12491	break;
12492
12493	case VMX_EXIT_INT_INFO_TYPE_SW_INT:
12494	uEventArg = idxVector;
12495	enmEvent1 = DBGFEVENT_INTERRUPT_SOFTWARE;
12496	fDtrace1 = VBOXVMM_INT_SOFTWARE_ENABLED();
12497	break;
12498	}
12499	break;
12500	}
12501
12502	case VMX_EXIT_TRIPLE_FAULT:
12503	enmEvent1 = DBGFEVENT_TRIPLE_FAULT;
12504	//fDtrace1 = VBOXVMM_EXIT_TRIPLE_FAULT_ENABLED();
12505	break;
12506	case VMX_EXIT_TASK_SWITCH: SET_EXIT(TASK_SWITCH); break;
12507	case VMX_EXIT_EPT_VIOLATION: SET_EXIT(VMX_EPT_VIOLATION); break;
12508	case VMX_EXIT_EPT_MISCONFIG: SET_EXIT(VMX_EPT_MISCONFIG); break;
12509	case VMX_EXIT_APIC_ACCESS: SET_EXIT(VMX_VAPIC_ACCESS); break;
12510	case VMX_EXIT_APIC_WRITE: SET_EXIT(VMX_VAPIC_WRITE); break;
12511
12512	/* Instruction specific VM-exits: */
12513	case VMX_EXIT_CPUID: SET_BOTH(CPUID); break;
12514	case VMX_EXIT_GETSEC: SET_BOTH(GETSEC); break;
12515	case VMX_EXIT_HLT: SET_BOTH(HALT); break;
12516	case VMX_EXIT_INVD: SET_BOTH(INVD); break;
12517	case VMX_EXIT_INVLPG: SET_BOTH(INVLPG); break;
12518	case VMX_EXIT_RDPMC: SET_BOTH(RDPMC); break;
12519	case VMX_EXIT_RDTSC: SET_BOTH(RDTSC); break;
12520	case VMX_EXIT_RSM: SET_BOTH(RSM); break;
12521	case VMX_EXIT_VMCALL: SET_BOTH(VMM_CALL); break;
12522	case VMX_EXIT_VMCLEAR: SET_BOTH(VMX_VMCLEAR); break;
12523	case VMX_EXIT_VMLAUNCH: SET_BOTH(VMX_VMLAUNCH); break;
12524	case VMX_EXIT_VMPTRLD: SET_BOTH(VMX_VMPTRLD); break;
12525	case VMX_EXIT_VMPTRST: SET_BOTH(VMX_VMPTRST); break;
12526	case VMX_EXIT_VMREAD: SET_BOTH(VMX_VMREAD); break;
12527	case VMX_EXIT_VMRESUME: SET_BOTH(VMX_VMRESUME); break;
12528	case VMX_EXIT_VMWRITE: SET_BOTH(VMX_VMWRITE); break;
12529	case VMX_EXIT_VMXOFF: SET_BOTH(VMX_VMXOFF); break;
12530	case VMX_EXIT_VMXON: SET_BOTH(VMX_VMXON); break;
12531	case VMX_EXIT_MOV_CRX:
12532	hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
12533	if (VMX_EXIT_QUAL_CRX_ACCESS(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_CRX_ACCESS_READ)
12534	SET_BOTH(CRX_READ);
12535	else
12536	SET_BOTH(CRX_WRITE);
12537	uEventArg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
12538	break;
12539	case VMX_EXIT_MOV_DRX:
12540	hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
12541	if ( VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual)
12542	== VMX_EXIT_QUAL_DRX_DIRECTION_READ)
12543	SET_BOTH(DRX_READ);
12544	else
12545	SET_BOTH(DRX_WRITE);
12546	uEventArg = VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual);
12547	break;
12548	case VMX_EXIT_RDMSR: SET_BOTH(RDMSR); break;
12549	case VMX_EXIT_WRMSR: SET_BOTH(WRMSR); break;
12550	case VMX_EXIT_MWAIT: SET_BOTH(MWAIT); break;
12551	case VMX_EXIT_MONITOR: SET_BOTH(MONITOR); break;
12552	case VMX_EXIT_PAUSE: SET_BOTH(PAUSE); break;
12553	case VMX_EXIT_GDTR_IDTR_ACCESS:
12554	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
12555	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_BF_XDTR_INSINFO_INSTR_ID))
12556	{
12557	case VMX_XDTR_INSINFO_II_SGDT: SET_BOTH(SGDT); break;
12558	case VMX_XDTR_INSINFO_II_SIDT: SET_BOTH(SIDT); break;
12559	case VMX_XDTR_INSINFO_II_LGDT: SET_BOTH(LGDT); break;
12560	case VMX_XDTR_INSINFO_II_LIDT: SET_BOTH(LIDT); break;
12561	}
12562	break;
12563
12564	case VMX_EXIT_LDTR_TR_ACCESS:
12565	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
12566	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_BF_YYTR_INSINFO_INSTR_ID))
12567	{
12568	case VMX_YYTR_INSINFO_II_SLDT: SET_BOTH(SLDT); break;
12569	case VMX_YYTR_INSINFO_II_STR: SET_BOTH(STR); break;
12570	case VMX_YYTR_INSINFO_II_LLDT: SET_BOTH(LLDT); break;
12571	case VMX_YYTR_INSINFO_II_LTR: SET_BOTH(LTR); break;
12572	}
12573	break;
12574
12575	case VMX_EXIT_INVEPT: SET_BOTH(VMX_INVEPT); break;
12576	case VMX_EXIT_RDTSCP: SET_BOTH(RDTSCP); break;
12577	case VMX_EXIT_INVVPID: SET_BOTH(VMX_INVVPID); break;
12578	case VMX_EXIT_WBINVD: SET_BOTH(WBINVD); break;
12579	case VMX_EXIT_XSETBV: SET_BOTH(XSETBV); break;
12580	case VMX_EXIT_RDRAND: SET_BOTH(RDRAND); break;
12581	case VMX_EXIT_INVPCID: SET_BOTH(VMX_INVPCID); break;
12582	case VMX_EXIT_VMFUNC: SET_BOTH(VMX_VMFUNC); break;
12583	case VMX_EXIT_RDSEED: SET_BOTH(RDSEED); break;
12584	case VMX_EXIT_XSAVES: SET_BOTH(XSAVES); break;
12585	case VMX_EXIT_XRSTORS: SET_BOTH(XRSTORS); break;
12586
12587	/* Events that aren't relevant at this point. */
12588	case VMX_EXIT_EXT_INT:
12589	case VMX_EXIT_INT_WINDOW:
12590	case VMX_EXIT_NMI_WINDOW:
12591	case VMX_EXIT_TPR_BELOW_THRESHOLD:
12592	case VMX_EXIT_PREEMPT_TIMER:
12593	case VMX_EXIT_IO_INSTR:
12594	break;
12595
12596	/* Errors and unexpected events. */
12597	case VMX_EXIT_INIT_SIGNAL:
12598	case VMX_EXIT_SIPI:
12599	case VMX_EXIT_IO_SMI:
12600	case VMX_EXIT_SMI:
12601	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
12602	case VMX_EXIT_ERR_MSR_LOAD:
12603	case VMX_EXIT_ERR_MACHINE_CHECK:
12604	case VMX_EXIT_PML_FULL:
12605	case VMX_EXIT_VIRTUALIZED_EOI:
12606	break;
12607
12608	default:
12609	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
12610	break;
12611	}
12612	#undef SET_BOTH
12613	#undef SET_EXIT
12614
12615	/*
12616	* Dtrace tracepoints go first. We do them here at once so we don't
12617	* have to copy the guest state saving and stuff a few dozen times.
12618	* Down side is that we've got to repeat the switch, though this time
12619	* we use enmEvent since the probes are a subset of what DBGF does.
12620	*/
12621	if (fDtrace1 \|\| fDtrace2)
12622	{
12623	hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
12624	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
12625	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
12626	switch (enmEvent1)
12627	{
12628	/** @todo consider which extra parameters would be helpful for each probe. */
12629	case DBGFEVENT_END: break;
12630	case DBGFEVENT_XCPT_DE: VBOXVMM_XCPT_DE(pVCpu, pCtx); break;
12631	case DBGFEVENT_XCPT_DB: VBOXVMM_XCPT_DB(pVCpu, pCtx, pCtx->dr[6]); break;
12632	case DBGFEVENT_XCPT_BP: VBOXVMM_XCPT_BP(pVCpu, pCtx); break;
12633	case DBGFEVENT_XCPT_OF: VBOXVMM_XCPT_OF(pVCpu, pCtx); break;
12634	case DBGFEVENT_XCPT_BR: VBOXVMM_XCPT_BR(pVCpu, pCtx); break;
12635	case DBGFEVENT_XCPT_UD: VBOXVMM_XCPT_UD(pVCpu, pCtx); break;
12636	case DBGFEVENT_XCPT_NM: VBOXVMM_XCPT_NM(pVCpu, pCtx); break;
12637	case DBGFEVENT_XCPT_DF: VBOXVMM_XCPT_DF(pVCpu, pCtx); break;
12638	case DBGFEVENT_XCPT_TS: VBOXVMM_XCPT_TS(pVCpu, pCtx, uEventArg); break;
12639	case DBGFEVENT_XCPT_NP: VBOXVMM_XCPT_NP(pVCpu, pCtx, uEventArg); break;
12640	case DBGFEVENT_XCPT_SS: VBOXVMM_XCPT_SS(pVCpu, pCtx, uEventArg); break;
12641	case DBGFEVENT_XCPT_GP: VBOXVMM_XCPT_GP(pVCpu, pCtx, uEventArg); break;
12642	case DBGFEVENT_XCPT_PF: VBOXVMM_XCPT_PF(pVCpu, pCtx, uEventArg, pCtx->cr2); break;
12643	case DBGFEVENT_XCPT_MF: VBOXVMM_XCPT_MF(pVCpu, pCtx); break;
12644	case DBGFEVENT_XCPT_AC: VBOXVMM_XCPT_AC(pVCpu, pCtx); break;
12645	case DBGFEVENT_XCPT_XF: VBOXVMM_XCPT_XF(pVCpu, pCtx); break;
12646	case DBGFEVENT_XCPT_VE: VBOXVMM_XCPT_VE(pVCpu, pCtx); break;
12647	case DBGFEVENT_XCPT_SX: VBOXVMM_XCPT_SX(pVCpu, pCtx, uEventArg); break;
12648	case DBGFEVENT_INTERRUPT_SOFTWARE: VBOXVMM_INT_SOFTWARE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12649	case DBGFEVENT_INSTR_CPUID: VBOXVMM_INSTR_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
12650	case DBGFEVENT_INSTR_GETSEC: VBOXVMM_INSTR_GETSEC(pVCpu, pCtx); break;
12651	case DBGFEVENT_INSTR_HALT: VBOXVMM_INSTR_HALT(pVCpu, pCtx); break;
12652	case DBGFEVENT_INSTR_INVD: VBOXVMM_INSTR_INVD(pVCpu, pCtx); break;
12653	case DBGFEVENT_INSTR_INVLPG: VBOXVMM_INSTR_INVLPG(pVCpu, pCtx); break;
12654	case DBGFEVENT_INSTR_RDPMC: VBOXVMM_INSTR_RDPMC(pVCpu, pCtx); break;
12655	case DBGFEVENT_INSTR_RDTSC: VBOXVMM_INSTR_RDTSC(pVCpu, pCtx); break;
12656	case DBGFEVENT_INSTR_RSM: VBOXVMM_INSTR_RSM(pVCpu, pCtx); break;
12657	case DBGFEVENT_INSTR_CRX_READ: VBOXVMM_INSTR_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12658	case DBGFEVENT_INSTR_CRX_WRITE: VBOXVMM_INSTR_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12659	case DBGFEVENT_INSTR_DRX_READ: VBOXVMM_INSTR_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12660	case DBGFEVENT_INSTR_DRX_WRITE: VBOXVMM_INSTR_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12661	case DBGFEVENT_INSTR_RDMSR: VBOXVMM_INSTR_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
12662	case DBGFEVENT_INSTR_WRMSR: VBOXVMM_INSTR_WRMSR(pVCpu, pCtx, pCtx->ecx,
12663	RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
12664	case DBGFEVENT_INSTR_MWAIT: VBOXVMM_INSTR_MWAIT(pVCpu, pCtx); break;
12665	case DBGFEVENT_INSTR_MONITOR: VBOXVMM_INSTR_MONITOR(pVCpu, pCtx); break;
12666	case DBGFEVENT_INSTR_PAUSE: VBOXVMM_INSTR_PAUSE(pVCpu, pCtx); break;
12667	case DBGFEVENT_INSTR_SGDT: VBOXVMM_INSTR_SGDT(pVCpu, pCtx); break;
12668	case DBGFEVENT_INSTR_SIDT: VBOXVMM_INSTR_SIDT(pVCpu, pCtx); break;
12669	case DBGFEVENT_INSTR_LGDT: VBOXVMM_INSTR_LGDT(pVCpu, pCtx); break;
12670	case DBGFEVENT_INSTR_LIDT: VBOXVMM_INSTR_LIDT(pVCpu, pCtx); break;
12671	case DBGFEVENT_INSTR_SLDT: VBOXVMM_INSTR_SLDT(pVCpu, pCtx); break;
12672	case DBGFEVENT_INSTR_STR: VBOXVMM_INSTR_STR(pVCpu, pCtx); break;
12673	case DBGFEVENT_INSTR_LLDT: VBOXVMM_INSTR_LLDT(pVCpu, pCtx); break;
12674	case DBGFEVENT_INSTR_LTR: VBOXVMM_INSTR_LTR(pVCpu, pCtx); break;
12675	case DBGFEVENT_INSTR_RDTSCP: VBOXVMM_INSTR_RDTSCP(pVCpu, pCtx); break;
12676	case DBGFEVENT_INSTR_WBINVD: VBOXVMM_INSTR_WBINVD(pVCpu, pCtx); break;
12677	case DBGFEVENT_INSTR_XSETBV: VBOXVMM_INSTR_XSETBV(pVCpu, pCtx); break;
12678	case DBGFEVENT_INSTR_RDRAND: VBOXVMM_INSTR_RDRAND(pVCpu, pCtx); break;
12679	case DBGFEVENT_INSTR_RDSEED: VBOXVMM_INSTR_RDSEED(pVCpu, pCtx); break;
12680	case DBGFEVENT_INSTR_XSAVES: VBOXVMM_INSTR_XSAVES(pVCpu, pCtx); break;
12681	case DBGFEVENT_INSTR_XRSTORS: VBOXVMM_INSTR_XRSTORS(pVCpu, pCtx); break;
12682	case DBGFEVENT_INSTR_VMM_CALL: VBOXVMM_INSTR_VMM_CALL(pVCpu, pCtx); break;
12683	case DBGFEVENT_INSTR_VMX_VMCLEAR: VBOXVMM_INSTR_VMX_VMCLEAR(pVCpu, pCtx); break;
12684	case DBGFEVENT_INSTR_VMX_VMLAUNCH: VBOXVMM_INSTR_VMX_VMLAUNCH(pVCpu, pCtx); break;
12685	case DBGFEVENT_INSTR_VMX_VMPTRLD: VBOXVMM_INSTR_VMX_VMPTRLD(pVCpu, pCtx); break;
12686	case DBGFEVENT_INSTR_VMX_VMPTRST: VBOXVMM_INSTR_VMX_VMPTRST(pVCpu, pCtx); break;
12687	case DBGFEVENT_INSTR_VMX_VMREAD: VBOXVMM_INSTR_VMX_VMREAD(pVCpu, pCtx); break;
12688	case DBGFEVENT_INSTR_VMX_VMRESUME: VBOXVMM_INSTR_VMX_VMRESUME(pVCpu, pCtx); break;
12689	case DBGFEVENT_INSTR_VMX_VMWRITE: VBOXVMM_INSTR_VMX_VMWRITE(pVCpu, pCtx); break;
12690	case DBGFEVENT_INSTR_VMX_VMXOFF: VBOXVMM_INSTR_VMX_VMXOFF(pVCpu, pCtx); break;
12691	case DBGFEVENT_INSTR_VMX_VMXON: VBOXVMM_INSTR_VMX_VMXON(pVCpu, pCtx); break;
12692	case DBGFEVENT_INSTR_VMX_INVEPT: VBOXVMM_INSTR_VMX_INVEPT(pVCpu, pCtx); break;
12693	case DBGFEVENT_INSTR_VMX_INVVPID: VBOXVMM_INSTR_VMX_INVVPID(pVCpu, pCtx); break;
12694	case DBGFEVENT_INSTR_VMX_INVPCID: VBOXVMM_INSTR_VMX_INVPCID(pVCpu, pCtx); break;
12695	case DBGFEVENT_INSTR_VMX_VMFUNC: VBOXVMM_INSTR_VMX_VMFUNC(pVCpu, pCtx); break;
12696	default: AssertMsgFailed(("enmEvent1=%d uExitReason=%d\n", enmEvent1, uExitReason)); break;
12697	}
12698	switch (enmEvent2)
12699	{
12700	/** @todo consider which extra parameters would be helpful for each probe. */
12701	case DBGFEVENT_END: break;
12702	case DBGFEVENT_EXIT_TASK_SWITCH: VBOXVMM_EXIT_TASK_SWITCH(pVCpu, pCtx); break;
12703	case DBGFEVENT_EXIT_CPUID: VBOXVMM_EXIT_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
12704	case DBGFEVENT_EXIT_GETSEC: VBOXVMM_EXIT_GETSEC(pVCpu, pCtx); break;
12705	case DBGFEVENT_EXIT_HALT: VBOXVMM_EXIT_HALT(pVCpu, pCtx); break;
12706	case DBGFEVENT_EXIT_INVD: VBOXVMM_EXIT_INVD(pVCpu, pCtx); break;
12707	case DBGFEVENT_EXIT_INVLPG: VBOXVMM_EXIT_INVLPG(pVCpu, pCtx); break;
12708	case DBGFEVENT_EXIT_RDPMC: VBOXVMM_EXIT_RDPMC(pVCpu, pCtx); break;
12709	case DBGFEVENT_EXIT_RDTSC: VBOXVMM_EXIT_RDTSC(pVCpu, pCtx); break;
12710	case DBGFEVENT_EXIT_RSM: VBOXVMM_EXIT_RSM(pVCpu, pCtx); break;
12711	case DBGFEVENT_EXIT_CRX_READ: VBOXVMM_EXIT_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12712	case DBGFEVENT_EXIT_CRX_WRITE: VBOXVMM_EXIT_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12713	case DBGFEVENT_EXIT_DRX_READ: VBOXVMM_EXIT_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12714	case DBGFEVENT_EXIT_DRX_WRITE: VBOXVMM_EXIT_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12715	case DBGFEVENT_EXIT_RDMSR: VBOXVMM_EXIT_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
12716	case DBGFEVENT_EXIT_WRMSR: VBOXVMM_EXIT_WRMSR(pVCpu, pCtx, pCtx->ecx,
12717	RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
12718	case DBGFEVENT_EXIT_MWAIT: VBOXVMM_EXIT_MWAIT(pVCpu, pCtx); break;
12719	case DBGFEVENT_EXIT_MONITOR: VBOXVMM_EXIT_MONITOR(pVCpu, pCtx); break;
12720	case DBGFEVENT_EXIT_PAUSE: VBOXVMM_EXIT_PAUSE(pVCpu, pCtx); break;
12721	case DBGFEVENT_EXIT_SGDT: VBOXVMM_EXIT_SGDT(pVCpu, pCtx); break;
12722	case DBGFEVENT_EXIT_SIDT: VBOXVMM_EXIT_SIDT(pVCpu, pCtx); break;
12723	case DBGFEVENT_EXIT_LGDT: VBOXVMM_EXIT_LGDT(pVCpu, pCtx); break;
12724	case DBGFEVENT_EXIT_LIDT: VBOXVMM_EXIT_LIDT(pVCpu, pCtx); break;
12725	case DBGFEVENT_EXIT_SLDT: VBOXVMM_EXIT_SLDT(pVCpu, pCtx); break;
12726	case DBGFEVENT_EXIT_STR: VBOXVMM_EXIT_STR(pVCpu, pCtx); break;
12727	case DBGFEVENT_EXIT_LLDT: VBOXVMM_EXIT_LLDT(pVCpu, pCtx); break;
12728	case DBGFEVENT_EXIT_LTR: VBOXVMM_EXIT_LTR(pVCpu, pCtx); break;
12729	case DBGFEVENT_EXIT_RDTSCP: VBOXVMM_EXIT_RDTSCP(pVCpu, pCtx); break;
12730	case DBGFEVENT_EXIT_WBINVD: VBOXVMM_EXIT_WBINVD(pVCpu, pCtx); break;
12731	case DBGFEVENT_EXIT_XSETBV: VBOXVMM_EXIT_XSETBV(pVCpu, pCtx); break;
12732	case DBGFEVENT_EXIT_RDRAND: VBOXVMM_EXIT_RDRAND(pVCpu, pCtx); break;
12733	case DBGFEVENT_EXIT_RDSEED: VBOXVMM_EXIT_RDSEED(pVCpu, pCtx); break;
12734	case DBGFEVENT_EXIT_XSAVES: VBOXVMM_EXIT_XSAVES(pVCpu, pCtx); break;
12735	case DBGFEVENT_EXIT_XRSTORS: VBOXVMM_EXIT_XRSTORS(pVCpu, pCtx); break;
12736	case DBGFEVENT_EXIT_VMM_CALL: VBOXVMM_EXIT_VMM_CALL(pVCpu, pCtx); break;
12737	case DBGFEVENT_EXIT_VMX_VMCLEAR: VBOXVMM_EXIT_VMX_VMCLEAR(pVCpu, pCtx); break;
12738	case DBGFEVENT_EXIT_VMX_VMLAUNCH: VBOXVMM_EXIT_VMX_VMLAUNCH(pVCpu, pCtx); break;
12739	case DBGFEVENT_EXIT_VMX_VMPTRLD: VBOXVMM_EXIT_VMX_VMPTRLD(pVCpu, pCtx); break;
12740	case DBGFEVENT_EXIT_VMX_VMPTRST: VBOXVMM_EXIT_VMX_VMPTRST(pVCpu, pCtx); break;
12741	case DBGFEVENT_EXIT_VMX_VMREAD: VBOXVMM_EXIT_VMX_VMREAD(pVCpu, pCtx); break;
12742	case DBGFEVENT_EXIT_VMX_VMRESUME: VBOXVMM_EXIT_VMX_VMRESUME(pVCpu, pCtx); break;
12743	case DBGFEVENT_EXIT_VMX_VMWRITE: VBOXVMM_EXIT_VMX_VMWRITE(pVCpu, pCtx); break;
12744	case DBGFEVENT_EXIT_VMX_VMXOFF: VBOXVMM_EXIT_VMX_VMXOFF(pVCpu, pCtx); break;
12745	case DBGFEVENT_EXIT_VMX_VMXON: VBOXVMM_EXIT_VMX_VMXON(pVCpu, pCtx); break;
12746	case DBGFEVENT_EXIT_VMX_INVEPT: VBOXVMM_EXIT_VMX_INVEPT(pVCpu, pCtx); break;
12747	case DBGFEVENT_EXIT_VMX_INVVPID: VBOXVMM_EXIT_VMX_INVVPID(pVCpu, pCtx); break;
12748	case DBGFEVENT_EXIT_VMX_INVPCID: VBOXVMM_EXIT_VMX_INVPCID(pVCpu, pCtx); break;
12749	case DBGFEVENT_EXIT_VMX_VMFUNC: VBOXVMM_EXIT_VMX_VMFUNC(pVCpu, pCtx); break;
12750	case DBGFEVENT_EXIT_VMX_EPT_MISCONFIG: VBOXVMM_EXIT_VMX_EPT_MISCONFIG(pVCpu, pCtx); break;
12751	case DBGFEVENT_EXIT_VMX_EPT_VIOLATION: VBOXVMM_EXIT_VMX_EPT_VIOLATION(pVCpu, pCtx); break;
12752	case DBGFEVENT_EXIT_VMX_VAPIC_ACCESS: VBOXVMM_EXIT_VMX_VAPIC_ACCESS(pVCpu, pCtx); break;
12753	case DBGFEVENT_EXIT_VMX_VAPIC_WRITE: VBOXVMM_EXIT_VMX_VAPIC_WRITE(pVCpu, pCtx); break;
12754	default: AssertMsgFailed(("enmEvent2=%d uExitReason=%d\n", enmEvent2, uExitReason)); break;
12755	}
12756	}
12757
12758	/*
12759	* Fire of the DBGF event, if enabled (our check here is just a quick one,
12760	* the DBGF call will do a full check).
12761	*
12762	* Note! DBGF sets DBGFEVENT_INTERRUPT_SOFTWARE in the bitmap.
12763	* Note! If we have to events, we prioritize the first, i.e. the instruction
12764	* one, in order to avoid event nesting.
12765	*/
12766	PVM pVM = pVCpu->CTX_SUFF(pVM);
12767	if ( enmEvent1 != DBGFEVENT_END
12768	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent1))
12769	{
12770	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12771	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent1, DBGFEVENTCTX_HM, 1, uEventArg);
12772	if (rcStrict != VINF_SUCCESS)
12773	return rcStrict;
12774	}
12775	else if ( enmEvent2 != DBGFEVENT_END
12776	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent2))
12777	{
12778	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12779	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent2, DBGFEVENTCTX_HM, 1, uEventArg);
12780	if (rcStrict != VINF_SUCCESS)
12781	return rcStrict;
12782	}
12783
12784	return VINF_SUCCESS;
12785	}
12786
12787
12788	/**
12789	* Single-stepping VM-exit filtering.
12790	*
12791	* This is preprocessing the VM-exits and deciding whether we've gotten far
12792	* enough to return VINF_EM_DBG_STEPPED already. If not, normal VM-exit
12793	* handling is performed.
12794	*
12795	* @returns Strict VBox status code (i.e. informational status codes too).
12796	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
12797	* @param pVmxTransient The VMX-transient structure.
12798	* @param pDbgState The debug state.
12799	*/
12800	DECLINLINE(VBOXSTRICTRC) hmR0VmxRunDebugHandleExit(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
12801	{
12802	/*
12803	* Expensive (saves context) generic dtrace VM-exit probe.
12804	*/
12805	uint32_t const uExitReason = pVmxTransient->uExitReason;
12806	if (!VBOXVMM_R0_HMVMX_VMEXIT_ENABLED())
12807	{ /* more likely */ }
12808	else
12809	{
12810	hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
12811	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
12812	AssertRC(rc);
12813	VBOXVMM_R0_HMVMX_VMEXIT(pVCpu, &pVCpu->cpum.GstCtx, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
12814	}
12815
12816	/*
12817	* Check for host NMI, just to get that out of the way.
12818	*/
12819	if (uExitReason != VMX_EXIT_XCPT_OR_NMI)
12820	{ /* normally likely */ }
12821	else
12822	{
12823	int rc2 = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
12824	AssertRCReturn(rc2, rc2);
12825	uint32_t const uIntType = VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo);
12826	if (uIntType == VMX_EXIT_INT_INFO_TYPE_NMI)
12827	return hmR0VmxExitHostNmi(pVCpu, pVmxTransient->pVmcsInfo);
12828	}
12829
12830	/*
12831	* Check for single stepping event if we're stepping.
12832	*/
12833	if (pVCpu->hm.s.fSingleInstruction)
12834	{
12835	switch (uExitReason)
12836	{
12837	case VMX_EXIT_MTF:
12838	return hmR0VmxExitMtf(pVCpu, pVmxTransient);
12839
12840	/* Various events: */
12841	case VMX_EXIT_XCPT_OR_NMI:
12842	case VMX_EXIT_EXT_INT:
12843	case VMX_EXIT_TRIPLE_FAULT:
12844	case VMX_EXIT_INT_WINDOW:
12845	case VMX_EXIT_NMI_WINDOW:
12846	case VMX_EXIT_TASK_SWITCH:
12847	case VMX_EXIT_TPR_BELOW_THRESHOLD:
12848	case VMX_EXIT_APIC_ACCESS:
12849	case VMX_EXIT_EPT_VIOLATION:
12850	case VMX_EXIT_EPT_MISCONFIG:
12851	case VMX_EXIT_PREEMPT_TIMER:
12852
12853	/* Instruction specific VM-exits: */
12854	case VMX_EXIT_CPUID:
12855	case VMX_EXIT_GETSEC:
12856	case VMX_EXIT_HLT:
12857	case VMX_EXIT_INVD:
12858	case VMX_EXIT_INVLPG:
12859	case VMX_EXIT_RDPMC:
12860	case VMX_EXIT_RDTSC:
12861	case VMX_EXIT_RSM:
12862	case VMX_EXIT_VMCALL:
12863	case VMX_EXIT_VMCLEAR:
12864	case VMX_EXIT_VMLAUNCH:
12865	case VMX_EXIT_VMPTRLD:
12866	case VMX_EXIT_VMPTRST:
12867	case VMX_EXIT_VMREAD:
12868	case VMX_EXIT_VMRESUME:
12869	case VMX_EXIT_VMWRITE:
12870	case VMX_EXIT_VMXOFF:
12871	case VMX_EXIT_VMXON:
12872	case VMX_EXIT_MOV_CRX:
12873	case VMX_EXIT_MOV_DRX:
12874	case VMX_EXIT_IO_INSTR:
12875	case VMX_EXIT_RDMSR:
12876	case VMX_EXIT_WRMSR:
12877	case VMX_EXIT_MWAIT:
12878	case VMX_EXIT_MONITOR:
12879	case VMX_EXIT_PAUSE:
12880	case VMX_EXIT_GDTR_IDTR_ACCESS:
12881	case VMX_EXIT_LDTR_TR_ACCESS:
12882	case VMX_EXIT_INVEPT:
12883	case VMX_EXIT_RDTSCP:
12884	case VMX_EXIT_INVVPID:
12885	case VMX_EXIT_WBINVD:
12886	case VMX_EXIT_XSETBV:
12887	case VMX_EXIT_RDRAND:
12888	case VMX_EXIT_INVPCID:
12889	case VMX_EXIT_VMFUNC:
12890	case VMX_EXIT_RDSEED:
12891	case VMX_EXIT_XSAVES:
12892	case VMX_EXIT_XRSTORS:
12893	{
12894	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12895	AssertRCReturn(rc, rc);
12896	if ( pVCpu->cpum.GstCtx.rip != pDbgState->uRipStart
12897	\|\| pVCpu->cpum.GstCtx.cs.Sel != pDbgState->uCsStart)
12898	return VINF_EM_DBG_STEPPED;
12899	break;
12900	}
12901
12902	/* Errors and unexpected events: */
12903	case VMX_EXIT_INIT_SIGNAL:
12904	case VMX_EXIT_SIPI:
12905	case VMX_EXIT_IO_SMI:
12906	case VMX_EXIT_SMI:
12907	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
12908	case VMX_EXIT_ERR_MSR_LOAD:
12909	case VMX_EXIT_ERR_MACHINE_CHECK:
12910	case VMX_EXIT_PML_FULL:
12911	case VMX_EXIT_VIRTUALIZED_EOI:
12912	case VMX_EXIT_APIC_WRITE: /* Some talk about this being fault like, so I guess we must process it? */
12913	break;
12914
12915	default:
12916	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
12917	break;
12918	}
12919	}
12920
12921	/*
12922	* Check for debugger event breakpoints and dtrace probes.
12923	*/
12924	if ( uExitReason < RT_ELEMENTS(pDbgState->bmExitsToCheck) * 32U
12925	&& ASMBitTest(pDbgState->bmExitsToCheck, uExitReason) )
12926	{
12927	VBOXSTRICTRC rcStrict = hmR0VmxHandleExitDtraceEvents(pVCpu, pVmxTransient, uExitReason);
12928	if (rcStrict != VINF_SUCCESS)
12929	return rcStrict;
12930	}
12931
12932	/*
12933	* Normal processing.
12934	*/
12935	#ifdef HMVMX_USE_FUNCTION_TABLE
12936	return g_apfnVMExitHandlers[uExitReason](pVCpu, pVmxTransient);
12937	#else
12938	return hmR0VmxHandleExit(pVCpu, pVmxTransient, uExitReason);
12939	#endif
12940	}
12941
12942
12943	/**
12944	* Single steps guest code using hardware-assisted VMX.
12945	*
12946	* This is -not- the same as the guest single-stepping itself (say using EFLAGS.TF)
12947	* but single-stepping through the hypervisor debugger.
12948	*
12949	* @returns Strict VBox status code (i.e. informational status codes too).
12950	* @param pVCpu The cross context virtual CPU structure.
12951	* @param pcLoops Pointer to the number of executed loops.
12952	*
12953	* @note Mostly the same as hmR0VmxRunGuestCodeNormal().
12954	*/
12955	static VBOXSTRICTRC hmR0VmxRunGuestCodeDebug(PVMCPU pVCpu, uint32_t *pcLoops)
12956	{
12957	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hm.s.cMaxResumeLoops;
12958	Assert(pcLoops);
12959	Assert(*pcLoops <= cMaxResumeLoops);
12960
12961	VMXTRANSIENT VmxTransient;
12962	RT_ZERO(VmxTransient);
12963	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
12964
12965	/* Set HMCPU indicators. */
12966	bool const fSavedSingleInstruction = pVCpu->hm.s.fSingleInstruction;
12967	pVCpu->hm.s.fSingleInstruction = pVCpu->hm.s.fSingleInstruction \|\| DBGFIsStepping(pVCpu);
12968	pVCpu->hm.s.fDebugWantRdTscExit = false;
12969	pVCpu->hm.s.fUsingDebugLoop = true;
12970
12971	/* State we keep to help modify and later restore the VMCS fields we alter, and for detecting steps. */
12972	VMXRUNDBGSTATE DbgState;
12973	hmR0VmxRunDebugStateInit(pVCpu, &VmxTransient, &DbgState);
12974	hmR0VmxPreRunGuestDebugStateUpdate(pVCpu, &VmxTransient, &DbgState);
12975
12976	/*
12977	* The loop.
12978	*/
12979	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
12980	for (;;)
12981	{
12982	Assert(!HMR0SuspendPending());
12983	HMVMX_ASSERT_CPU_SAFE(pVCpu);
12984	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
12985	bool fStepping = pVCpu->hm.s.fSingleInstruction;
12986
12987	/* Set up VM-execution controls the next two can respond to. */
12988	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &VmxTransient, &DbgState);
12989
12990	/*
12991	* Preparatory work for running guest code, this may force us to
12992	* return to ring-3.
12993	*
12994	* Warning! This bugger disables interrupts on VINF_SUCCESS!
12995	*/
12996	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, fStepping);
12997	if (rcStrict != VINF_SUCCESS)
12998	break;
12999
13000	/* Interrupts are disabled at this point! */
13001	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
13002
13003	/* Override any obnoxious code in the above two calls. */
13004	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &VmxTransient, &DbgState);
13005
13006	/*
13007	* Finally execute the guest.
13008	*/
13009	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
13010
13011	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
13012	/* Interrupts are re-enabled at this point! */
13013
13014	/* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
13015	if (RT_SUCCESS(rcRun))
13016	{ /* very likely */ }
13017	else
13018	{
13019	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
13020	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
13021	return rcRun;
13022	}
13023
13024	/* Profile the VM-exit. */
13025	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
13026	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
13027	STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
13028	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
13029	HMVMX_START_EXIT_DISPATCH_PROF();
13030
13031	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
13032
13033	/*
13034	* Handle the VM-exit - we quit earlier on certain VM-exits, see hmR0VmxHandleExitDebug().
13035	*/
13036	rcStrict = hmR0VmxRunDebugHandleExit(pVCpu, &VmxTransient, &DbgState);
13037	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
13038	if (rcStrict != VINF_SUCCESS)
13039	break;
13040	if (++(*pcLoops) > cMaxResumeLoops)
13041	{
13042	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
13043	rcStrict = VINF_EM_RAW_INTERRUPT;
13044	break;
13045	}
13046
13047	/*
13048	* Stepping: Did the RIP change, if so, consider it a single step.
13049	* Otherwise, make sure one of the TFs gets set.
13050	*/
13051	if (fStepping)
13052	{
13053	int rc = hmR0VmxImportGuestState(pVCpu, VmxTransient.pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
13054	AssertRC(rc);
13055	if ( pVCpu->cpum.GstCtx.rip != DbgState.uRipStart
13056	\|\| pVCpu->cpum.GstCtx.cs.Sel != DbgState.uCsStart)
13057	{
13058	rcStrict = VINF_EM_DBG_STEPPED;
13059	break;
13060	}
13061	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
13062	}
13063
13064	/*
13065	* Update when dtrace settings changes (DBGF kicks us, so no need to check).
13066	*/
13067	if (VBOXVMM_GET_SETTINGS_SEQ_NO() != DbgState.uDtraceSettingsSeqNo)
13068	hmR0VmxPreRunGuestDebugStateUpdate(pVCpu, &VmxTransient, &DbgState);
13069	}
13070
13071	/*
13072	* Clear the X86_EFL_TF if necessary.
13073	*/
13074	if (pVCpu->hm.s.fClearTrapFlag)
13075	{
13076	int rc = hmR0VmxImportGuestState(pVCpu, VmxTransient.pVmcsInfo, CPUMCTX_EXTRN_RFLAGS);
13077	AssertRC(rc);
13078	pVCpu->hm.s.fClearTrapFlag = false;
13079	pVCpu->cpum.GstCtx.eflags.Bits.u1TF = 0;
13080	}
13081	/** @todo there seems to be issues with the resume flag when the monitor trap
13082	* flag is pending without being used. Seen early in bios init when
13083	* accessing APIC page in protected mode. */
13084
13085	/*
13086	* Restore VM-exit control settings as we may not re-enter this function the
13087	* next time around.
13088	*/
13089	rcStrict = hmR0VmxRunDebugStateRevert(pVCpu, &VmxTransient, &DbgState, rcStrict);
13090
13091	/* Restore HMCPU indicators. */
13092	pVCpu->hm.s.fUsingDebugLoop = false;
13093	pVCpu->hm.s.fDebugWantRdTscExit = false;
13094	pVCpu->hm.s.fSingleInstruction = fSavedSingleInstruction;
13095
13096	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
13097	return rcStrict;
13098	}
13099
13100
13101	/** @} */
13102
13103
13104	/**
13105	* Checks if any expensive dtrace probes are enabled and we should go to the
13106	* debug loop.
13107	*
13108	* @returns true if we should use debug loop, false if not.
13109	*/
13110	static bool hmR0VmxAnyExpensiveProbesEnabled(void)
13111	{
13112	/* It's probably faster to OR the raw 32-bit counter variables together.
13113	Since the variables are in an array and the probes are next to one
13114	another (more or less), we have good locality. So, better read
13115	eight-nine cache lines ever time and only have one conditional, than
13116	128+ conditionals, right? */
13117	return ( VBOXVMM_R0_HMVMX_VMEXIT_ENABLED_RAW() /* expensive too due to context */
13118	\| VBOXVMM_XCPT_DE_ENABLED_RAW()
13119	\| VBOXVMM_XCPT_DB_ENABLED_RAW()
13120	\| VBOXVMM_XCPT_BP_ENABLED_RAW()
13121	\| VBOXVMM_XCPT_OF_ENABLED_RAW()
13122	\| VBOXVMM_XCPT_BR_ENABLED_RAW()
13123	\| VBOXVMM_XCPT_UD_ENABLED_RAW()
13124	\| VBOXVMM_XCPT_NM_ENABLED_RAW()
13125	\| VBOXVMM_XCPT_DF_ENABLED_RAW()
13126	\| VBOXVMM_XCPT_TS_ENABLED_RAW()
13127	\| VBOXVMM_XCPT_NP_ENABLED_RAW()
13128	\| VBOXVMM_XCPT_SS_ENABLED_RAW()
13129	\| VBOXVMM_XCPT_GP_ENABLED_RAW()
13130	\| VBOXVMM_XCPT_PF_ENABLED_RAW()
13131	\| VBOXVMM_XCPT_MF_ENABLED_RAW()
13132	\| VBOXVMM_XCPT_AC_ENABLED_RAW()
13133	\| VBOXVMM_XCPT_XF_ENABLED_RAW()
13134	\| VBOXVMM_XCPT_VE_ENABLED_RAW()
13135	\| VBOXVMM_XCPT_SX_ENABLED_RAW()
13136	\| VBOXVMM_INT_SOFTWARE_ENABLED_RAW()
13137	\| VBOXVMM_INT_HARDWARE_ENABLED_RAW()
13138	) != 0
13139	\|\| ( VBOXVMM_INSTR_HALT_ENABLED_RAW()
13140	\| VBOXVMM_INSTR_MWAIT_ENABLED_RAW()
13141	\| VBOXVMM_INSTR_MONITOR_ENABLED_RAW()
13142	\| VBOXVMM_INSTR_CPUID_ENABLED_RAW()
13143	\| VBOXVMM_INSTR_INVD_ENABLED_RAW()
13144	\| VBOXVMM_INSTR_WBINVD_ENABLED_RAW()
13145	\| VBOXVMM_INSTR_INVLPG_ENABLED_RAW()
13146	\| VBOXVMM_INSTR_RDTSC_ENABLED_RAW()
13147	\| VBOXVMM_INSTR_RDTSCP_ENABLED_RAW()
13148	\| VBOXVMM_INSTR_RDPMC_ENABLED_RAW()
13149	\| VBOXVMM_INSTR_RDMSR_ENABLED_RAW()
13150	\| VBOXVMM_INSTR_WRMSR_ENABLED_RAW()
13151	\| VBOXVMM_INSTR_CRX_READ_ENABLED_RAW()
13152	\| VBOXVMM_INSTR_CRX_WRITE_ENABLED_RAW()
13153	\| VBOXVMM_INSTR_DRX_READ_ENABLED_RAW()
13154	\| VBOXVMM_INSTR_DRX_WRITE_ENABLED_RAW()
13155	\| VBOXVMM_INSTR_PAUSE_ENABLED_RAW()
13156	\| VBOXVMM_INSTR_XSETBV_ENABLED_RAW()
13157	\| VBOXVMM_INSTR_SIDT_ENABLED_RAW()
13158	\| VBOXVMM_INSTR_LIDT_ENABLED_RAW()
13159	\| VBOXVMM_INSTR_SGDT_ENABLED_RAW()
13160	\| VBOXVMM_INSTR_LGDT_ENABLED_RAW()
13161	\| VBOXVMM_INSTR_SLDT_ENABLED_RAW()
13162	\| VBOXVMM_INSTR_LLDT_ENABLED_RAW()
13163	\| VBOXVMM_INSTR_STR_ENABLED_RAW()
13164	\| VBOXVMM_INSTR_LTR_ENABLED_RAW()
13165	\| VBOXVMM_INSTR_GETSEC_ENABLED_RAW()
13166	\| VBOXVMM_INSTR_RSM_ENABLED_RAW()
13167	\| VBOXVMM_INSTR_RDRAND_ENABLED_RAW()
13168	\| VBOXVMM_INSTR_RDSEED_ENABLED_RAW()
13169	\| VBOXVMM_INSTR_XSAVES_ENABLED_RAW()
13170	\| VBOXVMM_INSTR_XRSTORS_ENABLED_RAW()
13171	\| VBOXVMM_INSTR_VMM_CALL_ENABLED_RAW()
13172	\| VBOXVMM_INSTR_VMX_VMCLEAR_ENABLED_RAW()
13173	\| VBOXVMM_INSTR_VMX_VMLAUNCH_ENABLED_RAW()
13174	\| VBOXVMM_INSTR_VMX_VMPTRLD_ENABLED_RAW()
13175	\| VBOXVMM_INSTR_VMX_VMPTRST_ENABLED_RAW()
13176	\| VBOXVMM_INSTR_VMX_VMREAD_ENABLED_RAW()
13177	\| VBOXVMM_INSTR_VMX_VMRESUME_ENABLED_RAW()
13178	\| VBOXVMM_INSTR_VMX_VMWRITE_ENABLED_RAW()
13179	\| VBOXVMM_INSTR_VMX_VMXOFF_ENABLED_RAW()
13180	\| VBOXVMM_INSTR_VMX_VMXON_ENABLED_RAW()
13181	\| VBOXVMM_INSTR_VMX_VMFUNC_ENABLED_RAW()
13182	\| VBOXVMM_INSTR_VMX_INVEPT_ENABLED_RAW()
13183	\| VBOXVMM_INSTR_VMX_INVVPID_ENABLED_RAW()
13184	\| VBOXVMM_INSTR_VMX_INVPCID_ENABLED_RAW()
13185	) != 0
13186	\|\| ( VBOXVMM_EXIT_TASK_SWITCH_ENABLED_RAW()
13187	\| VBOXVMM_EXIT_HALT_ENABLED_RAW()
13188	\| VBOXVMM_EXIT_MWAIT_ENABLED_RAW()
13189	\| VBOXVMM_EXIT_MONITOR_ENABLED_RAW()
13190	\| VBOXVMM_EXIT_CPUID_ENABLED_RAW()
13191	\| VBOXVMM_EXIT_INVD_ENABLED_RAW()
13192	\| VBOXVMM_EXIT_WBINVD_ENABLED_RAW()
13193	\| VBOXVMM_EXIT_INVLPG_ENABLED_RAW()
13194	\| VBOXVMM_EXIT_RDTSC_ENABLED_RAW()
13195	\| VBOXVMM_EXIT_RDTSCP_ENABLED_RAW()
13196	\| VBOXVMM_EXIT_RDPMC_ENABLED_RAW()
13197	\| VBOXVMM_EXIT_RDMSR_ENABLED_RAW()
13198	\| VBOXVMM_EXIT_WRMSR_ENABLED_RAW()
13199	\| VBOXVMM_EXIT_CRX_READ_ENABLED_RAW()
13200	\| VBOXVMM_EXIT_CRX_WRITE_ENABLED_RAW()
13201	\| VBOXVMM_EXIT_DRX_READ_ENABLED_RAW()
13202	\| VBOXVMM_EXIT_DRX_WRITE_ENABLED_RAW()
13203	\| VBOXVMM_EXIT_PAUSE_ENABLED_RAW()
13204	\| VBOXVMM_EXIT_XSETBV_ENABLED_RAW()
13205	\| VBOXVMM_EXIT_SIDT_ENABLED_RAW()
13206	\| VBOXVMM_EXIT_LIDT_ENABLED_RAW()
13207	\| VBOXVMM_EXIT_SGDT_ENABLED_RAW()
13208	\| VBOXVMM_EXIT_LGDT_ENABLED_RAW()
13209	\| VBOXVMM_EXIT_SLDT_ENABLED_RAW()
13210	\| VBOXVMM_EXIT_LLDT_ENABLED_RAW()
13211	\| VBOXVMM_EXIT_STR_ENABLED_RAW()
13212	\| VBOXVMM_EXIT_LTR_ENABLED_RAW()
13213	\| VBOXVMM_EXIT_GETSEC_ENABLED_RAW()
13214	\| VBOXVMM_EXIT_RSM_ENABLED_RAW()
13215	\| VBOXVMM_EXIT_RDRAND_ENABLED_RAW()
13216	\| VBOXVMM_EXIT_RDSEED_ENABLED_RAW()
13217	\| VBOXVMM_EXIT_XSAVES_ENABLED_RAW()
13218	\| VBOXVMM_EXIT_XRSTORS_ENABLED_RAW()
13219	\| VBOXVMM_EXIT_VMM_CALL_ENABLED_RAW()
13220	\| VBOXVMM_EXIT_VMX_VMCLEAR_ENABLED_RAW()
13221	\| VBOXVMM_EXIT_VMX_VMLAUNCH_ENABLED_RAW()
13222	\| VBOXVMM_EXIT_VMX_VMPTRLD_ENABLED_RAW()
13223	\| VBOXVMM_EXIT_VMX_VMPTRST_ENABLED_RAW()
13224	\| VBOXVMM_EXIT_VMX_VMREAD_ENABLED_RAW()
13225	\| VBOXVMM_EXIT_VMX_VMRESUME_ENABLED_RAW()
13226	\| VBOXVMM_EXIT_VMX_VMWRITE_ENABLED_RAW()
13227	\| VBOXVMM_EXIT_VMX_VMXOFF_ENABLED_RAW()
13228	\| VBOXVMM_EXIT_VMX_VMXON_ENABLED_RAW()
13229	\| VBOXVMM_EXIT_VMX_VMFUNC_ENABLED_RAW()
13230	\| VBOXVMM_EXIT_VMX_INVEPT_ENABLED_RAW()
13231	\| VBOXVMM_EXIT_VMX_INVVPID_ENABLED_RAW()
13232	\| VBOXVMM_EXIT_VMX_INVPCID_ENABLED_RAW()
13233	\| VBOXVMM_EXIT_VMX_EPT_VIOLATION_ENABLED_RAW()
13234	\| VBOXVMM_EXIT_VMX_EPT_MISCONFIG_ENABLED_RAW()
13235	\| VBOXVMM_EXIT_VMX_VAPIC_ACCESS_ENABLED_RAW()
13236	\| VBOXVMM_EXIT_VMX_VAPIC_WRITE_ENABLED_RAW()
13237	) != 0;
13238	}
13239
13240
13241	/**
13242	* Runs the guest using hardware-assisted VMX.
13243	*
13244	* @returns Strict VBox status code (i.e. informational status codes too).
13245	* @param pVCpu The cross context virtual CPU structure.
13246	*/
13247	VMMR0DECL(VBOXSTRICTRC) VMXR0RunGuestCode(PVMCPU pVCpu)
13248	{
13249	AssertPtr(pVCpu);
13250	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
13251	Assert(VMMRZCallRing3IsEnabled(pVCpu));
13252	Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
13253	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
13254
13255	VMMRZCallRing3SetNotification(pVCpu, hmR0VmxCallRing3Callback, pCtx);
13256
13257	VBOXSTRICTRC rcStrict;
13258	uint32_t cLoops = 0;
13259	for (;;)
13260	{
13261	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13262	bool const fInNestedGuestMode = CPUMIsGuestInVmxNonRootMode(pCtx);
13263	#else
13264	bool const fInNestedGuestMode = false;
13265	#endif
13266	if (!fInNestedGuestMode)
13267	{
13268	if ( !pVCpu->hm.s.fUseDebugLoop
13269	&& (!VBOXVMM_ANY_PROBES_ENABLED() \|\| !hmR0VmxAnyExpensiveProbesEnabled())
13270	&& !DBGFIsStepping(pVCpu)
13271	&& !pVCpu->CTX_SUFF(pVM)->dbgf.ro.cEnabledInt3Breakpoints)
13272	rcStrict = hmR0VmxRunGuestCodeNormal(pVCpu, &cLoops);
13273	else
13274	rcStrict = hmR0VmxRunGuestCodeDebug(pVCpu, &cLoops);
13275	}
13276	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13277	else
13278	rcStrict = hmR0VmxRunGuestCodeNested(pVCpu, &cLoops);
13279
13280	if (rcStrict == VINF_VMX_VMLAUNCH_VMRESUME)
13281	{
13282	Assert(CPUMIsGuestInVmxNonRootMode(pCtx));
13283	continue;
13284	}
13285	if (rcStrict == VINF_VMX_VMEXIT)
13286	{
13287	Assert(!CPUMIsGuestInVmxNonRootMode(pCtx));
13288	continue;
13289	}
13290	#endif
13291	break;
13292	}
13293
13294	int const rcLoop = VBOXSTRICTRC_VAL(rcStrict);
13295	switch (rcLoop)
13296	{
13297	case VERR_EM_INTERPRETER: rcStrict = VINF_EM_RAW_EMULATE_INSTR; break;
13298	case VINF_EM_RESET: rcStrict = VINF_EM_TRIPLE_FAULT; break;
13299	}
13300
13301	int rc2 = hmR0VmxExitToRing3(pVCpu, rcStrict);
13302	if (RT_FAILURE(rc2))
13303	{
13304	pVCpu->hm.s.u32HMError = (uint32_t)VBOXSTRICTRC_VAL(rcStrict);
13305	rcStrict = rc2;
13306	}
13307	Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
13308	Assert(!VMMRZCallRing3IsNotificationSet(pVCpu));
13309	return rcStrict;
13310	}
13311
13312
13313	#ifndef HMVMX_USE_FUNCTION_TABLE
13314	/**
13315	* Handles a guest VM-exit from hardware-assisted VMX execution.
13316	*
13317	* @returns Strict VBox status code (i.e. informational status codes too).
13318	* @param pVCpu The cross context virtual CPU structure.
13319	* @param pVmxTransient The VMX-transient structure.
13320	*/
13321	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
13322	{
13323	#ifdef DEBUG_ramshankar
13324	#define VMEXIT_CALL_RET(a_fSave, a_CallExpr) \
13325	do { \
13326	if (a_fSave != 0) \
13327	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL); \
13328	VBOXSTRICTRC rcStrict = a_CallExpr; \
13329	if (a_fSave != 0) \
13330	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST); \
13331	return rcStrict; \
13332	} while (0)
13333	#else
13334	# define VMEXIT_CALL_RET(a_fSave, a_CallExpr) return a_CallExpr
13335	#endif
13336	uint32_t const uExitReason = pVmxTransient->uExitReason;
13337	switch (uExitReason)
13338	{
13339	case VMX_EXIT_EPT_MISCONFIG: VMEXIT_CALL_RET(0, hmR0VmxExitEptMisconfig(pVCpu, pVmxTransient));
13340	case VMX_EXIT_EPT_VIOLATION: VMEXIT_CALL_RET(0, hmR0VmxExitEptViolation(pVCpu, pVmxTransient));
13341	case VMX_EXIT_IO_INSTR: VMEXIT_CALL_RET(0, hmR0VmxExitIoInstr(pVCpu, pVmxTransient));
13342	case VMX_EXIT_CPUID: VMEXIT_CALL_RET(0, hmR0VmxExitCpuid(pVCpu, pVmxTransient));
13343	case VMX_EXIT_RDTSC: VMEXIT_CALL_RET(0, hmR0VmxExitRdtsc(pVCpu, pVmxTransient));
13344	case VMX_EXIT_RDTSCP: VMEXIT_CALL_RET(0, hmR0VmxExitRdtscp(pVCpu, pVmxTransient));
13345	case VMX_EXIT_APIC_ACCESS: VMEXIT_CALL_RET(0, hmR0VmxExitApicAccess(pVCpu, pVmxTransient));
13346	case VMX_EXIT_XCPT_OR_NMI: VMEXIT_CALL_RET(0, hmR0VmxExitXcptOrNmi(pVCpu, pVmxTransient));
13347	case VMX_EXIT_MOV_CRX: VMEXIT_CALL_RET(0, hmR0VmxExitMovCRx(pVCpu, pVmxTransient));
13348	case VMX_EXIT_EXT_INT: VMEXIT_CALL_RET(0, hmR0VmxExitExtInt(pVCpu, pVmxTransient));
13349	case VMX_EXIT_INT_WINDOW: VMEXIT_CALL_RET(0, hmR0VmxExitIntWindow(pVCpu, pVmxTransient));
13350	case VMX_EXIT_TPR_BELOW_THRESHOLD: VMEXIT_CALL_RET(0, hmR0VmxExitTprBelowThreshold(pVCpu, pVmxTransient));
13351	case VMX_EXIT_MWAIT: VMEXIT_CALL_RET(0, hmR0VmxExitMwait(pVCpu, pVmxTransient));
13352	case VMX_EXIT_MONITOR: VMEXIT_CALL_RET(0, hmR0VmxExitMonitor(pVCpu, pVmxTransient));
13353	case VMX_EXIT_TASK_SWITCH: VMEXIT_CALL_RET(0, hmR0VmxExitTaskSwitch(pVCpu, pVmxTransient));
13354	case VMX_EXIT_PREEMPT_TIMER: VMEXIT_CALL_RET(0, hmR0VmxExitPreemptTimer(pVCpu, pVmxTransient));
13355	case VMX_EXIT_RDMSR: VMEXIT_CALL_RET(0, hmR0VmxExitRdmsr(pVCpu, pVmxTransient));
13356	case VMX_EXIT_WRMSR: VMEXIT_CALL_RET(0, hmR0VmxExitWrmsr(pVCpu, pVmxTransient));
13357	case VMX_EXIT_VMCALL: VMEXIT_CALL_RET(0, hmR0VmxExitVmcall(pVCpu, pVmxTransient));
13358	case VMX_EXIT_MOV_DRX: VMEXIT_CALL_RET(0, hmR0VmxExitMovDRx(pVCpu, pVmxTransient));
13359	case VMX_EXIT_HLT: VMEXIT_CALL_RET(0, hmR0VmxExitHlt(pVCpu, pVmxTransient));
13360	case VMX_EXIT_INVD: VMEXIT_CALL_RET(0, hmR0VmxExitInvd(pVCpu, pVmxTransient));
13361	case VMX_EXIT_INVLPG: VMEXIT_CALL_RET(0, hmR0VmxExitInvlpg(pVCpu, pVmxTransient));
13362	case VMX_EXIT_MTF: VMEXIT_CALL_RET(0, hmR0VmxExitMtf(pVCpu, pVmxTransient));
13363	case VMX_EXIT_PAUSE: VMEXIT_CALL_RET(0, hmR0VmxExitPause(pVCpu, pVmxTransient));
13364	case VMX_EXIT_WBINVD: VMEXIT_CALL_RET(0, hmR0VmxExitWbinvd(pVCpu, pVmxTransient));
13365	case VMX_EXIT_XSETBV: VMEXIT_CALL_RET(0, hmR0VmxExitXsetbv(pVCpu, pVmxTransient));
13366	case VMX_EXIT_INVPCID: VMEXIT_CALL_RET(0, hmR0VmxExitInvpcid(pVCpu, pVmxTransient));
13367	case VMX_EXIT_GETSEC: VMEXIT_CALL_RET(0, hmR0VmxExitGetsec(pVCpu, pVmxTransient));
13368	case VMX_EXIT_RDPMC: VMEXIT_CALL_RET(0, hmR0VmxExitRdpmc(pVCpu, pVmxTransient));
13369	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13370	case VMX_EXIT_VMCLEAR: VMEXIT_CALL_RET(0, hmR0VmxExitVmclear(pVCpu, pVmxTransient));
13371	case VMX_EXIT_VMLAUNCH: VMEXIT_CALL_RET(0, hmR0VmxExitVmlaunch(pVCpu, pVmxTransient));
13372	case VMX_EXIT_VMPTRLD: VMEXIT_CALL_RET(0, hmR0VmxExitVmptrld(pVCpu, pVmxTransient));
13373	case VMX_EXIT_VMPTRST: VMEXIT_CALL_RET(0, hmR0VmxExitVmptrst(pVCpu, pVmxTransient));
13374	case VMX_EXIT_VMREAD: VMEXIT_CALL_RET(0, hmR0VmxExitVmread(pVCpu, pVmxTransient));
13375	case VMX_EXIT_VMRESUME: VMEXIT_CALL_RET(0, hmR0VmxExitVmwrite(pVCpu, pVmxTransient));
13376	case VMX_EXIT_VMWRITE: VMEXIT_CALL_RET(0, hmR0VmxExitVmresume(pVCpu, pVmxTransient));
13377	case VMX_EXIT_VMXOFF: VMEXIT_CALL_RET(0, hmR0VmxExitVmxoff(pVCpu, pVmxTransient));
13378	case VMX_EXIT_VMXON: VMEXIT_CALL_RET(0, hmR0VmxExitVmxon(pVCpu, pVmxTransient));
13379	case VMX_EXIT_INVVPID: VMEXIT_CALL_RET(0, hmR0VmxExitInvvpid(pVCpu, pVmxTransient));
13380	case VMX_EXIT_INVEPT: VMEXIT_CALL_RET(0, hmR0VmxExitSetPendingXcptUD(pVCpu, pVmxTransient));
13381	#else
13382	case VMX_EXIT_VMCLEAR:
13383	case VMX_EXIT_VMLAUNCH:
13384	case VMX_EXIT_VMPTRLD:
13385	case VMX_EXIT_VMPTRST:
13386	case VMX_EXIT_VMREAD:
13387	case VMX_EXIT_VMRESUME:
13388	case VMX_EXIT_VMWRITE:
13389	case VMX_EXIT_VMXOFF:
13390	case VMX_EXIT_VMXON:
13391	case VMX_EXIT_INVVPID:
13392	case VMX_EXIT_INVEPT:
13393	return hmR0VmxExitSetPendingXcptUD(pVCpu, pVmxTransient);
13394	#endif
13395
13396	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFault(pVCpu, pVmxTransient);
13397	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindow(pVCpu, pVmxTransient);
13398	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestState(pVCpu, pVmxTransient);
13399
13400	case VMX_EXIT_INIT_SIGNAL:
13401	case VMX_EXIT_SIPI:
13402	case VMX_EXIT_IO_SMI:
13403	case VMX_EXIT_SMI:
13404	case VMX_EXIT_ERR_MSR_LOAD:
13405	case VMX_EXIT_ERR_MACHINE_CHECK:
13406	case VMX_EXIT_PML_FULL:
13407	case VMX_EXIT_VIRTUALIZED_EOI:
13408	case VMX_EXIT_GDTR_IDTR_ACCESS:
13409	case VMX_EXIT_LDTR_TR_ACCESS:
13410	case VMX_EXIT_APIC_WRITE:
13411	case VMX_EXIT_RDRAND:
13412	case VMX_EXIT_RSM:
13413	case VMX_EXIT_VMFUNC:
13414	case VMX_EXIT_ENCLS:
13415	case VMX_EXIT_RDSEED:
13416	case VMX_EXIT_XSAVES:
13417	case VMX_EXIT_XRSTORS:
13418	case VMX_EXIT_UMWAIT:
13419	case VMX_EXIT_TPAUSE:
13420	default:
13421	return hmR0VmxExitErrUnexpected(pVCpu, pVmxTransient);
13422	}
13423	#undef VMEXIT_CALL_RET
13424	}
13425	#endif /* !HMVMX_USE_FUNCTION_TABLE */
13426
13427
13428	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13429	/**
13430	* Handles a nested-guest VM-exit from hardware-assisted VMX execution.
13431	*
13432	* @returns Strict VBox status code (i.e. informational status codes too).
13433	* @param pVCpu The cross context virtual CPU structure.
13434	* @param pVmxTransient The VMX-transient structure.
13435	*/
13436	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExitNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
13437	{
13438	uint32_t const uExitReason = pVmxTransient->uExitReason;
13439	switch (uExitReason)
13440	{
13441	case VMX_EXIT_EPT_MISCONFIG: return hmR0VmxExitEptMisconfig(pVCpu, pVmxTransient);
13442	case VMX_EXIT_EPT_VIOLATION: return hmR0VmxExitEptViolation(pVCpu, pVmxTransient);
13443	case VMX_EXIT_XCPT_OR_NMI: return hmR0VmxExitXcptOrNmiNested(pVCpu, pVmxTransient);
13444	case VMX_EXIT_IO_INSTR: return hmR0VmxExitIoInstrNested(pVCpu, pVmxTransient);
13445	case VMX_EXIT_HLT: return hmR0VmxExitHltNested(pVCpu, pVmxTransient);
13446
13447	/*
13448	* We shouldn't direct host physical interrupts to the nested-guest.
13449	*/
13450	case VMX_EXIT_EXT_INT:
13451	return hmR0VmxExitExtInt(pVCpu, pVmxTransient);
13452
13453	/*
13454	* Instructions that cause VM-exits unconditionally or the condition is
13455	* always is taken solely from the guest hypervisor (meaning if the VM-exit
13456	* happens, it's guaranteed to be a nested-guest VM-exit).
13457	*
13458	* - Provides VM-exit instruction length ONLY.
13459	*/
13460	case VMX_EXIT_CPUID: /* Unconditional. */
13461	case VMX_EXIT_VMCALL:
13462	case VMX_EXIT_GETSEC:
13463	case VMX_EXIT_INVD:
13464	case VMX_EXIT_XSETBV:
13465	case VMX_EXIT_VMLAUNCH:
13466	case VMX_EXIT_VMRESUME:
13467	case VMX_EXIT_VMXOFF:
13468	case VMX_EXIT_ENCLS: /* Condition specified solely by guest hypervisor. */
13469	case VMX_EXIT_VMFUNC:
13470	return hmR0VmxExitInstrNested(pVCpu, pVmxTransient);
13471
13472	/*
13473	* Instructions that cause VM-exits unconditionally or the condition is
13474	* always is taken solely from the guest hypervisor (meaning if the VM-exit
13475	* happens, it's guaranteed to be a nested-guest VM-exit).
13476	*
13477	* - Provides VM-exit instruction length.
13478	* - Provides VM-exit information.
13479	* - Optionally provides Exit qualification.
13480	*
13481	* Since Exit qualification is 0 for all VM-exits where it is not
13482	* applicable, reading and passing it to the guest should produce
13483	* defined behavior.
13484	*
13485	* See Intel spec. 27.2.1 "Basic VM-Exit Information".
13486	*/
13487	case VMX_EXIT_INVEPT: /* Unconditional. */
13488	case VMX_EXIT_INVVPID:
13489	case VMX_EXIT_VMCLEAR:
13490	case VMX_EXIT_VMPTRLD:
13491	case VMX_EXIT_VMPTRST:
13492	case VMX_EXIT_VMXON:
13493	case VMX_EXIT_GDTR_IDTR_ACCESS: /* Condition specified solely by guest hypervisor. */
13494	case VMX_EXIT_LDTR_TR_ACCESS:
13495	case VMX_EXIT_RDRAND:
13496	case VMX_EXIT_RDSEED:
13497	case VMX_EXIT_XSAVES:
13498	case VMX_EXIT_XRSTORS:
13499	case VMX_EXIT_UMWAIT:
13500	case VMX_EXIT_TPAUSE:
13501	return hmR0VmxExitInstrWithInfoNested(pVCpu, pVmxTransient);
13502
13503	case VMX_EXIT_RDTSC: return hmR0VmxExitRdtscNested(pVCpu, pVmxTransient);
13504	case VMX_EXIT_RDTSCP: return hmR0VmxExitRdtscpNested(pVCpu, pVmxTransient);
13505	case VMX_EXIT_RDMSR: return hmR0VmxExitRdmsrNested(pVCpu, pVmxTransient);
13506	case VMX_EXIT_WRMSR: return hmR0VmxExitWrmsrNested(pVCpu, pVmxTransient);
13507	case VMX_EXIT_INVLPG: return hmR0VmxExitInvlpgNested(pVCpu, pVmxTransient);
13508	case VMX_EXIT_INVPCID: return hmR0VmxExitInvpcidNested(pVCpu, pVmxTransient);
13509	case VMX_EXIT_TASK_SWITCH: return hmR0VmxExitTaskSwitchNested(pVCpu, pVmxTransient);
13510	case VMX_EXIT_WBINVD: return hmR0VmxExitWbinvdNested(pVCpu, pVmxTransient);
13511	case VMX_EXIT_MTF: return hmR0VmxExitMtfNested(pVCpu, pVmxTransient);
13512	case VMX_EXIT_APIC_ACCESS: return hmR0VmxExitApicAccessNested(pVCpu, pVmxTransient);
13513	case VMX_EXIT_APIC_WRITE: return hmR0VmxExitApicWriteNested(pVCpu, pVmxTransient);
13514	case VMX_EXIT_VIRTUALIZED_EOI: return hmR0VmxExitVirtEoiNested(pVCpu, pVmxTransient);
13515	case VMX_EXIT_MOV_CRX: return hmR0VmxExitMovCRxNested(pVCpu, pVmxTransient);
13516	case VMX_EXIT_INT_WINDOW: return hmR0VmxExitIntWindowNested(pVCpu, pVmxTransient);
13517	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindowNested(pVCpu, pVmxTransient);
13518	case VMX_EXIT_TPR_BELOW_THRESHOLD: return hmR0VmxExitTprBelowThresholdNested(pVCpu, pVmxTransient);
13519	case VMX_EXIT_MWAIT: return hmR0VmxExitMwaitNested(pVCpu, pVmxTransient);
13520	case VMX_EXIT_MONITOR: return hmR0VmxExitMonitorNested(pVCpu, pVmxTransient);
13521	case VMX_EXIT_PAUSE: return hmR0VmxExitPauseNested(pVCpu, pVmxTransient);
13522
13523	case VMX_EXIT_PREEMPT_TIMER:
13524	{
13525	/** @todo NSTVMX: Preempt timer. */
13526	return hmR0VmxExitPreemptTimer(pVCpu, pVmxTransient);
13527	}
13528
13529	case VMX_EXIT_MOV_DRX: return hmR0VmxExitMovDRxNested(pVCpu, pVmxTransient);
13530	case VMX_EXIT_RDPMC: return hmR0VmxExitRdpmcNested(pVCpu, pVmxTransient);
13531
13532	case VMX_EXIT_VMREAD:
13533	case VMX_EXIT_VMWRITE: return hmR0VmxExitVmreadVmwriteNested(pVCpu, pVmxTransient);
13534
13535	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFaultNested(pVCpu, pVmxTransient);
13536	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestStateNested(pVCpu, pVmxTransient);
13537
13538	case VMX_EXIT_INIT_SIGNAL:
13539	case VMX_EXIT_SIPI:
13540	case VMX_EXIT_IO_SMI:
13541	case VMX_EXIT_SMI:
13542	case VMX_EXIT_ERR_MSR_LOAD:
13543	case VMX_EXIT_ERR_MACHINE_CHECK:
13544	case VMX_EXIT_PML_FULL:
13545	case VMX_EXIT_RSM:
13546	default:
13547	return hmR0VmxExitErrUnexpected(pVCpu, pVmxTransient);
13548	}
13549	}
13550	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13551
13552
13553	/** @name VM-exit helpers.
13554	* @{
13555	*/
13556	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13557	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= VM-exit helpers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
13558	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13559
13560	/** Macro for VM-exits called unexpectedly. */
13561	#define HMVMX_UNEXPECTED_EXIT_RET(a_pVCpu, a_HmError) \
13562	do { \
13563	(a_pVCpu)->hm.s.u32HMError = (a_HmError); \
13564	return VERR_VMX_UNEXPECTED_EXIT; \
13565	} while (0)
13566
13567	#ifdef VBOX_STRICT
13568	/* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */
13569	# define HMVMX_ASSERT_PREEMPT_CPUID_VAR() \
13570	RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId()
13571
13572	# define HMVMX_ASSERT_PREEMPT_CPUID() \
13573	do { \
13574	RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \
13575	AssertMsg(idAssertCpu == idAssertCpuNow, ("VMX %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \
13576	} while (0)
13577
13578	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13579	do { \
13580	AssertPtr((a_pVCpu)); \
13581	AssertPtr((a_pVmxTransient)); \
13582	Assert((a_pVmxTransient)->fVMEntryFailed == false); \
13583	Assert((a_pVmxTransient)->pVmcsInfo); \
13584	Assert(ASMIntAreEnabled()); \
13585	HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu); \
13586	HMVMX_ASSERT_PREEMPT_CPUID_VAR(); \
13587	Log4Func(("vcpu[%RU32]\n", (a_pVCpu)->idCpu)); \
13588	HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu); \
13589	if (VMMR0IsLogFlushDisabled((a_pVCpu))) \
13590	HMVMX_ASSERT_PREEMPT_CPUID(); \
13591	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
13592	} while (0)
13593
13594	# define HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13595	do { \
13596	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient); \
13597	Assert((a_pVmxTransient)->fIsNestedGuest); \
13598	} while (0)
13599
13600	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13601	do { \
13602	Log4Func(("\n")); \
13603	} while (0)
13604	#else
13605	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13606	do { \
13607	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
13608	NOREF((a_pVCpu)); NOREF((a_pVmxTransient)); \
13609	} while (0)
13610
13611	# define HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13612	do { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient); } while (0)
13613
13614	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) do { } while (0)
13615	#endif
13616
13617	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13618	/** Macro that does the necessary privilege checks and intercepted VM-exits for
13619	* guests that attempted to execute a VMX instruction. */
13620	# define HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(a_pVCpu, a_uExitReason) \
13621	do \
13622	{ \
13623	VBOXSTRICTRC rcStrictTmp = hmR0VmxCheckExitDueToVmxInstr((a_pVCpu), (a_uExitReason)); \
13624	if (rcStrictTmp == VINF_SUCCESS) \
13625	{ /* likely */ } \
13626	else if (rcStrictTmp == VINF_HM_PENDING_XCPT) \
13627	{ \
13628	Assert((a_pVCpu)->hm.s.Event.fPending); \
13629	Log4Func(("Privilege checks failed -> %#x\n", VMX_ENTRY_INT_INFO_VECTOR((a_pVCpu)->hm.s.Event.u64IntInfo))); \
13630	return VINF_SUCCESS; \
13631	} \
13632	else \
13633	{ \
13634	int rcTmp = VBOXSTRICTRC_VAL(rcStrictTmp); \
13635	AssertMsgFailedReturn(("Unexpected failure. rc=%Rrc", rcTmp), rcTmp); \
13636	} \
13637	} while (0)
13638
13639	/** Macro that decodes a memory operand for an VM-exit caused by an instruction. */
13640	# define HMVMX_DECODE_MEM_OPERAND(a_pVCpu, a_uExitInstrInfo, a_uExitQual, a_enmMemAccess, a_pGCPtrEffAddr) \
13641	do \
13642	{ \
13643	VBOXSTRICTRC rcStrictTmp = hmR0VmxDecodeMemOperand((a_pVCpu), (a_uExitInstrInfo), (a_uExitQual), (a_enmMemAccess), \
13644	(a_pGCPtrEffAddr)); \
13645	if (rcStrictTmp == VINF_SUCCESS) \
13646	{ /* likely */ } \
13647	else if (rcStrictTmp == VINF_HM_PENDING_XCPT) \
13648	{ \
13649	uint8_t const uXcptTmp = VMX_ENTRY_INT_INFO_VECTOR((a_pVCpu)->hm.s.Event.u64IntInfo); \
13650	Log4Func(("Memory operand decoding failed, raising xcpt %#x\n", uXcptTmp)); \
13651	NOREF(uXcptTmp); \
13652	return VINF_SUCCESS; \
13653	} \
13654	else \
13655	{ \
13656	Log4Func(("hmR0VmxDecodeMemOperand failed. rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrictTmp))); \
13657	return rcStrictTmp; \
13658	} \
13659	} while (0)
13660	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13661
13662
13663	/**
13664	* Advances the guest RIP by the specified number of bytes.
13665	*
13666	* @param pVCpu The cross context virtual CPU structure.
13667	* @param cbInstr Number of bytes to advance the RIP by.
13668	*
13669	* @remarks No-long-jump zone!!!
13670	*/
13671	DECLINLINE(void) hmR0VmxAdvanceGuestRipBy(PVMCPU pVCpu, uint32_t cbInstr)
13672	{
13673	/* Advance the RIP. */
13674	pVCpu->cpum.GstCtx.rip += cbInstr;
13675	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
13676
13677	/* Update interrupt inhibition. */
13678	if ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)
13679	&& pVCpu->cpum.GstCtx.rip != EMGetInhibitInterruptsPC(pVCpu))
13680	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
13681	}
13682
13683
13684	/**
13685	* Advances the guest RIP after reading it from the VMCS.
13686	*
13687	* @returns VBox status code, no informational status codes.
13688	* @param pVCpu The cross context virtual CPU structure.
13689	* @param pVmxTransient The VMX-transient structure.
13690	*
13691	* @remarks No-long-jump zone!!!
13692	*/
13693	static int hmR0VmxAdvanceGuestRip(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
13694	{
13695	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13696	rc \|= hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS);
13697	AssertRCReturn(rc, rc);
13698
13699	hmR0VmxAdvanceGuestRipBy(pVCpu, pVmxTransient->cbInstr);
13700	return VINF_SUCCESS;
13701	}
13702
13703
13704	/**
13705	* Handle a condition that occurred while delivering an event through the guest or
13706	* nested-guest IDT.
13707	*
13708	* @returns Strict VBox status code (i.e. informational status codes too).
13709	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
13710	* @retval VINF_HM_DOUBLE_FAULT if a \#DF condition was detected and we ought
13711	* to continue execution of the guest which will delivery the \#DF.
13712	* @retval VINF_EM_RESET if we detected a triple-fault condition.
13713	* @retval VERR_EM_GUEST_CPU_HANG if we detected a guest CPU hang.
13714	*
13715	* @param pVCpu The cross context virtual CPU structure.
13716	* @param pVmxTransient The VMX-transient structure.
13717	*
13718	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13719	* Additionally, HMVMX_READ_EXIT_QUALIFICATION is required if the VM-exit
13720	* is due to an EPT violation, PML full or SPP-related event.
13721	*
13722	* @remarks No-long-jump zone!!!
13723	*/
13724	static VBOXSTRICTRC hmR0VmxCheckExitDueToEventDelivery(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
13725	{
13726	Assert(!pVCpu->hm.s.Event.fPending);
13727	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_XCPT_INFO);
13728	if ( pVmxTransient->uExitReason == VMX_EXIT_EPT_VIOLATION
13729	\|\| pVmxTransient->uExitReason == VMX_EXIT_PML_FULL
13730	\|\| pVmxTransient->uExitReason == VMX_EXIT_SPP_EVENT)
13731	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_EXIT_QUALIFICATION);
13732
13733	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
13734	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
13735	uint32_t const uIdtVectorInfo = pVmxTransient->uIdtVectoringInfo;
13736	uint32_t const uExitIntInfo = pVmxTransient->uExitIntInfo;
13737	if (VMX_IDT_VECTORING_INFO_IS_VALID(uIdtVectorInfo))
13738	{
13739	uint32_t const uIdtVector = VMX_IDT_VECTORING_INFO_VECTOR(uIdtVectorInfo);
13740	uint32_t const uIdtVectorType = VMX_IDT_VECTORING_INFO_TYPE(uIdtVectorInfo);
13741
13742	/*
13743	* If the event was a software interrupt (generated with INT n) or a software exception
13744	* (generated by INT3/INTO) or a privileged software exception (generated by INT1), we
13745	* can handle the VM-exit and continue guest execution which will re-execute the
13746	* instruction rather than re-injecting the exception, as that can cause premature
13747	* trips to ring-3 before injection and involve TRPM which currently has no way of
13748	* storing that these exceptions were caused by these instructions (ICEBP's #DB poses
13749	* the problem).
13750	*/
13751	IEMXCPTRAISE enmRaise;
13752	IEMXCPTRAISEINFO fRaiseInfo;
13753	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
13754	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
13755	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT)
13756	{
13757	enmRaise = IEMXCPTRAISE_REEXEC_INSTR;
13758	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
13759	}
13760	else if (VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo))
13761	{
13762	uint32_t const uExitVectorType = VMX_EXIT_INT_INFO_TYPE(uExitIntInfo);
13763	uint8_t const uExitVector = VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo);
13764	Assert(uExitVectorType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT);
13765
13766	uint32_t const fIdtVectorFlags = hmR0VmxGetIemXcptFlags(uIdtVector, uIdtVectorType);
13767	uint32_t const fExitVectorFlags = hmR0VmxGetIemXcptFlags(uExitVector, uExitVectorType);
13768
13769	enmRaise = IEMEvaluateRecursiveXcpt(pVCpu, fIdtVectorFlags, uIdtVector, fExitVectorFlags, uExitVector, &fRaiseInfo);
13770
13771	/* Determine a vectoring #PF condition, see comment in hmR0VmxExitXcptPF(). */
13772	if (fRaiseInfo & (IEMXCPTRAISEINFO_EXT_INT_PF \| IEMXCPTRAISEINFO_NMI_PF))
13773	{
13774	pVmxTransient->fVectoringPF = true;
13775	enmRaise = IEMXCPTRAISE_PREV_EVENT;
13776	}
13777	}
13778	else
13779	{
13780	/*
13781	* If an exception or hardware interrupt delivery caused an EPT violation/misconfig or APIC access
13782	* VM-exit, then the VM-exit interruption-information will not be valid and we end up here.
13783	* It is sufficient to reflect the original event to the guest after handling the VM-exit.
13784	*/
13785	Assert( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
13786	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
13787	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT);
13788	enmRaise = IEMXCPTRAISE_PREV_EVENT;
13789	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
13790	}
13791
13792	/*
13793	* On CPUs that support Virtual NMIs, if this VM-exit (be it an exception or EPT violation/misconfig
13794	* etc.) occurred while delivering the NMI, we need to clear the block-by-NMI field in the guest
13795	* interruptibility-state before re-delivering the NMI after handling the VM-exit. Otherwise the
13796	* subsequent VM-entry would fail, see @bugref{7445}.
13797	*
13798	* See Intel spec. 30.7.1.2 "Resuming Guest Software after Handling an Exception".
13799	*/
13800	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
13801	&& enmRaise == IEMXCPTRAISE_PREV_EVENT
13802	&& (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
13803	&& CPUMIsGuestNmiBlocking(pVCpu))
13804	{
13805	CPUMSetGuestNmiBlocking(pVCpu, false);
13806	}
13807
13808	switch (enmRaise)
13809	{
13810	case IEMXCPTRAISE_CURRENT_XCPT:
13811	{
13812	Log4Func(("IDT: Pending secondary Xcpt: idtinfo=%#RX64 exitinfo=%#RX64\n", uIdtVectorInfo, uExitIntInfo));
13813	Assert(rcStrict == VINF_SUCCESS);
13814	break;
13815	}
13816
13817	case IEMXCPTRAISE_PREV_EVENT:
13818	{
13819	uint32_t u32ErrCode;
13820	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(uIdtVectorInfo))
13821	u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
13822	else
13823	u32ErrCode = 0;
13824
13825	/* If uExitVector is #PF, CR2 value will be updated from the VMCS if it's a guest #PF, see hmR0VmxExitXcptPF(). */
13826	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflect);
13827	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(uIdtVectorInfo), 0 /* cbInstr */,
13828	u32ErrCode, pVCpu->cpum.GstCtx.cr2);
13829
13830	Log4Func(("IDT: Pending vectoring event %#RX64 Err=%#RX32\n", pVCpu->hm.s.Event.u64IntInfo,
13831	pVCpu->hm.s.Event.u32ErrCode));
13832	Assert(rcStrict == VINF_SUCCESS);
13833	break;
13834	}
13835
13836	case IEMXCPTRAISE_REEXEC_INSTR:
13837	Assert(rcStrict == VINF_SUCCESS);
13838	break;
13839
13840	case IEMXCPTRAISE_DOUBLE_FAULT:
13841	{
13842	/*
13843	* Determing a vectoring double #PF condition. Used later, when PGM evaluates the
13844	* second #PF as a guest #PF (and not a shadow #PF) and needs to be converted into a #DF.
13845	*/
13846	if (fRaiseInfo & IEMXCPTRAISEINFO_PF_PF)
13847	{
13848	pVmxTransient->fVectoringDoublePF = true;
13849	Log4Func(("IDT: Vectoring double #PF %#RX64 cr2=%#RX64\n", pVCpu->hm.s.Event.u64IntInfo,
13850	pVCpu->cpum.GstCtx.cr2));
13851	rcStrict = VINF_SUCCESS;
13852	}
13853	else
13854	{
13855	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectConvertDF);
13856	hmR0VmxSetPendingXcptDF(pVCpu);
13857	Log4Func(("IDT: Pending vectoring #DF %#RX64 uIdtVector=%#x uExitVector=%#x\n", pVCpu->hm.s.Event.u64IntInfo,
13858	uIdtVector, VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo)));
13859	rcStrict = VINF_HM_DOUBLE_FAULT;
13860	}
13861	break;
13862	}
13863
13864	case IEMXCPTRAISE_TRIPLE_FAULT:
13865	{
13866	Log4Func(("IDT: Pending vectoring triple-fault uIdt=%#x uExit=%#x\n", uIdtVector,
13867	VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo)));
13868	rcStrict = VINF_EM_RESET;
13869	break;
13870	}
13871
13872	case IEMXCPTRAISE_CPU_HANG:
13873	{
13874	Log4Func(("IDT: Bad guest! Entering CPU hang. fRaiseInfo=%#x\n", fRaiseInfo));
13875	rcStrict = VERR_EM_GUEST_CPU_HANG;
13876	break;
13877	}
13878
13879	default:
13880	{
13881	AssertMsgFailed(("IDT: vcpu[%RU32] Unexpected/invalid value! enmRaise=%#x\n", pVCpu->idCpu, enmRaise));
13882	rcStrict = VERR_VMX_IPE_2;
13883	break;
13884	}
13885	}
13886	}
13887	else if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
13888	&& !CPUMIsGuestNmiBlocking(pVCpu))
13889	{
13890	if ( VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo)
13891	&& VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo) != X86_XCPT_DF
13892	&& VMX_EXIT_INT_INFO_IS_NMI_UNBLOCK_IRET(uExitIntInfo))
13893	{
13894	/*
13895	* Execution of IRET caused a fault when NMI blocking was in effect (i.e we're in
13896	* the guest or nested-guest NMI handler). We need to set the block-by-NMI field so
13897	* that NMIs remain blocked until the IRET execution is completed.
13898	*
13899	* See Intel spec. 31.7.1.2 "Resuming Guest Software After Handling An Exception".
13900	*/
13901	CPUMSetGuestNmiBlocking(pVCpu, true);
13902	Log4Func(("Set NMI blocking. uExitReason=%u\n", pVmxTransient->uExitReason));
13903	}
13904	else if ( pVmxTransient->uExitReason == VMX_EXIT_EPT_VIOLATION
13905	\|\| pVmxTransient->uExitReason == VMX_EXIT_PML_FULL
13906	\|\| pVmxTransient->uExitReason == VMX_EXIT_SPP_EVENT)
13907	{
13908	/*
13909	* Execution of IRET caused an EPT violation, page-modification log-full event or
13910	* SPP-related event VM-exit when NMI blocking was in effect (i.e. we're in the
13911	* guest or nested-guest NMI handler). We need to set the block-by-NMI field so
13912	* that NMIs remain blocked until the IRET execution is completed.
13913	*
13914	* See Intel spec. 27.2.3 "Information about NMI unblocking due to IRET"
13915	*/
13916	if (VMX_EXIT_QUAL_EPT_IS_NMI_UNBLOCK_IRET(pVmxTransient->uExitQual))
13917	{
13918	CPUMSetGuestNmiBlocking(pVCpu, true);
13919	Log4Func(("Set NMI blocking. uExitReason=%u\n", pVmxTransient->uExitReason));
13920	}
13921	}
13922	}
13923
13924	Assert( rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_HM_DOUBLE_FAULT
13925	\|\| rcStrict == VINF_EM_RESET \|\| rcStrict == VERR_EM_GUEST_CPU_HANG);
13926	return rcStrict;
13927	}
13928
13929
13930	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13931	/**
13932	* Perform the relevant VMX instruction checks for VM-exits that occurred due to the
13933	* guest attempting to execute a VMX instruction.
13934	*
13935	* @returns Strict VBox status code (i.e. informational status codes too).
13936	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
13937	* @retval VINF_HM_PENDING_XCPT if an exception was raised.
13938	*
13939	* @param pVCpu The cross context virtual CPU structure.
13940	* @param uExitReason The VM-exit reason.
13941	*
13942	* @todo NSTVMX: Document other error codes when VM-exit is implemented.
13943	* @remarks No-long-jump zone!!!
13944	*/
13945	static VBOXSTRICTRC hmR0VmxCheckExitDueToVmxInstr(PVMCPU pVCpu, uint32_t uExitReason)
13946	{
13947	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS
13948	\| CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_EFER);
13949
13950	/*
13951	* The physical CPU would have already checked the CPU mode/code segment.
13952	* We shall just assert here for paranoia.
13953	* See Intel spec. 25.1.1 "Relative Priority of Faults and VM Exits".
13954	*/
13955	Assert(!CPUMIsGuestInRealOrV86ModeEx(&pVCpu->cpum.GstCtx));
13956	Assert( !CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx)
13957	\|\| CPUMIsGuestIn64BitCodeEx(&pVCpu->cpum.GstCtx));
13958
13959	if (uExitReason == VMX_EXIT_VMXON)
13960	{
13961	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
13962
13963	/*
13964	* We check CR4.VMXE because it is required to be always set while in VMX operation
13965	* by physical CPUs and our CR4 read-shadow is only consulted when executing specific
13966	* instructions (CLTS, LMSW, MOV CR, and SMSW) and thus doesn't affect CPU operation
13967	* otherwise (i.e. physical CPU won't automatically #UD if Cr4Shadow.VMXE is 0).
13968	*/
13969	if (!CPUMIsGuestVmxEnabled(&pVCpu->cpum.GstCtx))
13970	{
13971	Log4Func(("CR4.VMXE is not set -> #UD\n"));
13972	hmR0VmxSetPendingXcptUD(pVCpu);
13973	return VINF_HM_PENDING_XCPT;
13974	}
13975	}
13976	else if (!CPUMIsGuestInVmxRootMode(&pVCpu->cpum.GstCtx))
13977	{
13978	/*
13979	* The guest has not entered VMX operation but attempted to execute a VMX instruction
13980	* (other than VMXON), we need to raise a #UD.
13981	*/
13982	Log4Func(("Not in VMX root mode -> #UD\n"));
13983	hmR0VmxSetPendingXcptUD(pVCpu);
13984	return VINF_HM_PENDING_XCPT;
13985	}
13986
13987	/* All other checks (including VM-exit intercepts) are handled by IEM instruction emulation. */
13988	return VINF_SUCCESS;
13989	}
13990
13991	/**
13992	* Decodes the memory operand of an instruction that caused a VM-exit.
13993	*
13994	* The Exit qualification field provides the displacement field for memory
13995	* operand instructions, if any.
13996	*
13997	* @returns Strict VBox status code (i.e. informational status codes too).
13998	* @retval VINF_SUCCESS if the operand was successfully decoded.
13999	* @retval VINF_HM_PENDING_XCPT if an exception was raised while decoding the
14000	* operand.
14001	* @param pVCpu The cross context virtual CPU structure.
14002	* @param uExitInstrInfo The VM-exit instruction information field.
14003	* @param enmMemAccess The memory operand's access type (read or write).
14004	* @param GCPtrDisp The instruction displacement field, if any. For
14005	* RIP-relative addressing pass RIP + displacement here.
14006	* @param pGCPtrMem Where to store the effective destination memory address.
14007	*
14008	* @remarks Warning! This function ASSUMES the instruction cannot be used in real or
14009	* virtual-8086 mode hence skips those checks while verifying if the
14010	* segment is valid.
14011	*/
14012	static VBOXSTRICTRC hmR0VmxDecodeMemOperand(PVMCPU pVCpu, uint32_t uExitInstrInfo, RTGCPTR GCPtrDisp, VMXMEMACCESS enmMemAccess,
14013	PRTGCPTR pGCPtrMem)
14014	{
14015	Assert(pGCPtrMem);
14016	Assert(!CPUMIsGuestInRealOrV86Mode(pVCpu));
14017	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_EFER
14018	\| CPUMCTX_EXTRN_CR0);
14019
14020	static uint64_t const s_auAddrSizeMasks[] = { UINT64_C(0xffff), UINT64_C(0xffffffff), UINT64_C(0xffffffffffffffff) };
14021	static uint64_t const s_auAccessSizeMasks[] = { sizeof(uint16_t), sizeof(uint32_t), sizeof(uint64_t) };
14022	AssertCompile(RT_ELEMENTS(s_auAccessSizeMasks) == RT_ELEMENTS(s_auAddrSizeMasks));
14023
14024	VMXEXITINSTRINFO ExitInstrInfo;
14025	ExitInstrInfo.u = uExitInstrInfo;
14026	uint8_t const uAddrSize = ExitInstrInfo.All.u3AddrSize;
14027	uint8_t const iSegReg = ExitInstrInfo.All.iSegReg;
14028	bool const fIdxRegValid = !ExitInstrInfo.All.fIdxRegInvalid;
14029	uint8_t const iIdxReg = ExitInstrInfo.All.iIdxReg;
14030	uint8_t const uScale = ExitInstrInfo.All.u2Scaling;
14031	bool const fBaseRegValid = !ExitInstrInfo.All.fBaseRegInvalid;
14032	uint8_t const iBaseReg = ExitInstrInfo.All.iBaseReg;
14033	bool const fIsMemOperand = !ExitInstrInfo.All.fIsRegOperand;
14034	bool const fIsLongMode = CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx);
14035
14036	/*
14037	* Validate instruction information.
14038	* This shouldn't happen on real hardware but useful while testing our nested hardware-virtualization code.
14039	*/
14040	AssertLogRelMsgReturn(uAddrSize < RT_ELEMENTS(s_auAddrSizeMasks),
14041	("Invalid address size. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_1);
14042	AssertLogRelMsgReturn(iSegReg < X86_SREG_COUNT,
14043	("Invalid segment register. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_2);
14044	AssertLogRelMsgReturn(fIsMemOperand,
14045	("Expected memory operand. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_3);
14046
14047	/*
14048	* Compute the complete effective address.
14049	*
14050	* See AMD instruction spec. 1.4.2 "SIB Byte Format"
14051	* See AMD spec. 4.5.2 "Segment Registers".
14052	*/
14053	RTGCPTR GCPtrMem = GCPtrDisp;
14054	if (fBaseRegValid)
14055	GCPtrMem += pVCpu->cpum.GstCtx.aGRegs[iBaseReg].u64;
14056	if (fIdxRegValid)
14057	GCPtrMem += pVCpu->cpum.GstCtx.aGRegs[iIdxReg].u64 << uScale;
14058
14059	RTGCPTR const GCPtrOff = GCPtrMem;
14060	if ( !fIsLongMode
14061	\|\| iSegReg >= X86_SREG_FS)
14062	GCPtrMem += pVCpu->cpum.GstCtx.aSRegs[iSegReg].u64Base;
14063	GCPtrMem &= s_auAddrSizeMasks[uAddrSize];
14064
14065	/*
14066	* Validate effective address.
14067	* See AMD spec. 4.5.3 "Segment Registers in 64-Bit Mode".
14068	*/
14069	uint8_t const cbAccess = s_auAccessSizeMasks[uAddrSize];
14070	Assert(cbAccess > 0);
14071	if (fIsLongMode)
14072	{
14073	if (X86_IS_CANONICAL(GCPtrMem))
14074	{
14075	*pGCPtrMem = GCPtrMem;
14076	return VINF_SUCCESS;
14077	}
14078
14079	/** @todo r=ramshankar: We should probably raise \#SS or \#GP. See AMD spec. 4.12.2
14080	* "Data Limit Checks in 64-bit Mode". */
14081	Log4Func(("Long mode effective address is not canonical GCPtrMem=%#RX64\n", GCPtrMem));
14082	hmR0VmxSetPendingXcptGP(pVCpu, 0);
14083	return VINF_HM_PENDING_XCPT;
14084	}
14085
14086	/*
14087	* This is a watered down version of iemMemApplySegment().
14088	* Parts that are not applicable for VMX instructions like real-or-v8086 mode
14089	* and segment CPL/DPL checks are skipped.
14090	*/
14091	RTGCPTR32 const GCPtrFirst32 = (RTGCPTR32)GCPtrOff;
14092	RTGCPTR32 const GCPtrLast32 = GCPtrFirst32 + cbAccess - 1;
14093	PCCPUMSELREG pSel = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
14094
14095	/* Check if the segment is present and usable. */
14096	if ( pSel->Attr.n.u1Present
14097	&& !pSel->Attr.n.u1Unusable)
14098	{
14099	Assert(pSel->Attr.n.u1DescType);
14100	if (!(pSel->Attr.n.u4Type & X86_SEL_TYPE_CODE))
14101	{
14102	/* Check permissions for the data segment. */
14103	if ( enmMemAccess == VMXMEMACCESS_WRITE
14104	&& !(pSel->Attr.n.u4Type & X86_SEL_TYPE_WRITE))
14105	{
14106	Log4Func(("Data segment access invalid. iSegReg=%#x Attr=%#RX32\n", iSegReg, pSel->Attr.u));
14107	hmR0VmxSetPendingXcptGP(pVCpu, iSegReg);
14108	return VINF_HM_PENDING_XCPT;
14109	}
14110
14111	/* Check limits if it's a normal data segment. */
14112	if (!(pSel->Attr.n.u4Type & X86_SEL_TYPE_DOWN))
14113	{
14114	if ( GCPtrFirst32 > pSel->u32Limit
14115	\|\| GCPtrLast32 > pSel->u32Limit)
14116	{
14117	Log4Func(("Data segment limit exceeded. "
14118	"iSegReg=%#x GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n", iSegReg, GCPtrFirst32,
14119	GCPtrLast32, pSel->u32Limit));
14120	if (iSegReg == X86_SREG_SS)
14121	hmR0VmxSetPendingXcptSS(pVCpu, 0);
14122	else
14123	hmR0VmxSetPendingXcptGP(pVCpu, 0);
14124	return VINF_HM_PENDING_XCPT;
14125	}
14126	}
14127	else
14128	{
14129	/* Check limits if it's an expand-down data segment.
14130	Note! The upper boundary is defined by the B bit, not the G bit! */
14131	if ( GCPtrFirst32 < pSel->u32Limit + UINT32_C(1)
14132	\|\| GCPtrLast32 > (pSel->Attr.n.u1DefBig ? UINT32_MAX : UINT32_C(0xffff)))
14133	{
14134	Log4Func(("Expand-down data segment limit exceeded. "
14135	"iSegReg=%#x GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n", iSegReg, GCPtrFirst32,
14136	GCPtrLast32, pSel->u32Limit));
14137	if (iSegReg == X86_SREG_SS)
14138	hmR0VmxSetPendingXcptSS(pVCpu, 0);
14139	else
14140	hmR0VmxSetPendingXcptGP(pVCpu, 0);
14141	return VINF_HM_PENDING_XCPT;
14142	}
14143	}
14144	}
14145	else
14146	{
14147	/* Check permissions for the code segment. */
14148	if ( enmMemAccess == VMXMEMACCESS_WRITE
14149	\|\| ( enmMemAccess == VMXMEMACCESS_READ
14150	&& !(pSel->Attr.n.u4Type & X86_SEL_TYPE_READ)))
14151	{
14152	Log4Func(("Code segment access invalid. Attr=%#RX32\n", pSel->Attr.u));
14153	Assert(!CPUMIsGuestInRealOrV86ModeEx(&pVCpu->cpum.GstCtx));
14154	hmR0VmxSetPendingXcptGP(pVCpu, 0);
14155	return VINF_HM_PENDING_XCPT;
14156	}
14157
14158	/* Check limits for the code segment (normal/expand-down not applicable for code segments). */
14159	if ( GCPtrFirst32 > pSel->u32Limit
14160	\|\| GCPtrLast32 > pSel->u32Limit)
14161	{
14162	Log4Func(("Code segment limit exceeded. GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n",
14163	GCPtrFirst32, GCPtrLast32, pSel->u32Limit));
14164	if (iSegReg == X86_SREG_SS)
14165	hmR0VmxSetPendingXcptSS(pVCpu, 0);
14166	else
14167	hmR0VmxSetPendingXcptGP(pVCpu, 0);
14168	return VINF_HM_PENDING_XCPT;
14169	}
14170	}
14171	}
14172	else
14173	{
14174	Log4Func(("Not present or unusable segment. iSegReg=%#x Attr=%#RX32\n", iSegReg, pSel->Attr.u));
14175	hmR0VmxSetPendingXcptGP(pVCpu, 0);
14176	return VINF_HM_PENDING_XCPT;
14177	}
14178
14179	*pGCPtrMem = GCPtrMem;
14180	return VINF_SUCCESS;
14181	}
14182	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
14183
14184
14185	/**
14186	* VM-exit helper for LMSW.
14187	*/
14188	static VBOXSTRICTRC hmR0VmxExitLmsw(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint16_t uMsw, RTGCPTR GCPtrEffDst)
14189	{
14190	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14191	AssertRCReturn(rc, rc);
14192
14193	VBOXSTRICTRC rcStrict = IEMExecDecodedLmsw(pVCpu, cbInstr, uMsw, GCPtrEffDst);
14194	AssertMsg( rcStrict == VINF_SUCCESS
14195	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
14196
14197	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
14198	if (rcStrict == VINF_IEM_RAISED_XCPT)
14199	{
14200	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14201	rcStrict = VINF_SUCCESS;
14202	}
14203
14204	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitLmsw);
14205	Log4Func(("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
14206	return rcStrict;
14207	}
14208
14209
14210	/**
14211	* VM-exit helper for CLTS.
14212	*/
14213	static VBOXSTRICTRC hmR0VmxExitClts(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr)
14214	{
14215	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14216	AssertRCReturn(rc, rc);
14217
14218	VBOXSTRICTRC rcStrict = IEMExecDecodedClts(pVCpu, cbInstr);
14219	AssertMsg( rcStrict == VINF_SUCCESS
14220	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
14221
14222	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
14223	if (rcStrict == VINF_IEM_RAISED_XCPT)
14224	{
14225	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14226	rcStrict = VINF_SUCCESS;
14227	}
14228
14229	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitClts);
14230	Log4Func(("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
14231	return rcStrict;
14232	}
14233
14234
14235	/**
14236	* VM-exit helper for MOV from CRx (CRx read).
14237	*/
14238	static VBOXSTRICTRC hmR0VmxExitMovFromCrX(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint8_t iGReg, uint8_t iCrReg)
14239	{
14240	Assert(iCrReg < 16);
14241	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
14242
14243	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14244	AssertRCReturn(rc, rc);
14245
14246	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxRead(pVCpu, cbInstr, iGReg, iCrReg);
14247	AssertMsg( rcStrict == VINF_SUCCESS
14248	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
14249
14250	if (iGReg == X86_GREG_xSP)
14251	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_RSP);
14252	else
14253	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14254	#ifdef VBOX_WITH_STATISTICS
14255	switch (iCrReg)
14256	{
14257	case 0: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Read); break;
14258	case 2: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Read); break;
14259	case 3: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Read); break;
14260	case 4: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Read); break;
14261	case 8: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Read); break;
14262	}
14263	#endif
14264	Log4Func(("CR%d Read access rcStrict=%Rrc\n", iCrReg, VBOXSTRICTRC_VAL(rcStrict)));
14265	return rcStrict;
14266	}
14267
14268
14269	/**
14270	* VM-exit helper for MOV to CRx (CRx write).
14271	*/
14272	static VBOXSTRICTRC hmR0VmxExitMovToCrX(PVMCPU pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint8_t iGReg, uint8_t iCrReg)
14273	{
14274	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14275	AssertRCReturn(rc, rc);
14276
14277	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxWrite(pVCpu, cbInstr, iCrReg, iGReg);
14278	AssertMsg( rcStrict == VINF_SUCCESS
14279	\|\| rcStrict == VINF_IEM_RAISED_XCPT
14280	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
14281
14282	switch (iCrReg)
14283	{
14284	case 0:
14285	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
14286	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Write);
14287	Log4Func(("CR0 write. rcStrict=%Rrc CR0=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cr0));
14288	break;
14289
14290	case 2:
14291	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Write);
14292	/* Nothing to do here, CR2 it's not part of the VMCS. */
14293	break;
14294
14295	case 3:
14296	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR3);
14297	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Write);
14298	Log4Func(("CR3 write. rcStrict=%Rrc CR3=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cr3));
14299	break;
14300
14301	case 4:
14302	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR4);
14303	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Write);
14304	Log4Func(("CR4 write. rc=%Rrc CR4=%#RX64 fLoadSaveGuestXcr0=%u\n", VBOXSTRICTRC_VAL(rcStrict),
14305	pVCpu->cpum.GstCtx.cr4, pVCpu->hm.s.fLoadSaveGuestXcr0));
14306	break;
14307
14308	case 8:
14309	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged,
14310	HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_APIC_TPR);
14311	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Write);
14312	break;
14313
14314	default:
14315	AssertMsgFailed(("Invalid CRx register %#x\n", iCrReg));
14316	break;
14317	}
14318
14319	if (rcStrict == VINF_IEM_RAISED_XCPT)
14320	{
14321	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14322	rcStrict = VINF_SUCCESS;
14323	}
14324	return rcStrict;
14325	}
14326
14327
14328	/**
14329	* VM-exit exception handler for \#PF (Page-fault exception).
14330	*
14331	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14332	*/
14333	static VBOXSTRICTRC hmR0VmxExitXcptPF(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
14334	{
14335	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14336	PVM pVM = pVCpu->CTX_SUFF(pVM);
14337	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
14338	AssertRCReturn(rc, rc);
14339
14340	if (!pVM->hm.s.fNestedPaging)
14341	{ /* likely */ }
14342	else
14343	{
14344	#if !defined(HMVMX_ALWAYS_TRAP_ALL_XCPTS) && !defined(HMVMX_ALWAYS_TRAP_PF)
14345	Assert(pVmxTransient->fIsNestedGuest \|\| pVCpu->hm.s.fUsingDebugLoop);
14346	#endif
14347	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
14348	if (!pVmxTransient->fVectoringDoublePF)
14349	{
14350	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */,
14351	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual);
14352	}
14353	else
14354	{
14355	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
14356	Assert(!pVmxTransient->fIsNestedGuest);
14357	hmR0VmxSetPendingXcptDF(pVCpu);
14358	Log4Func(("Pending #DF due to vectoring #PF w/ NestedPaging\n"));
14359	}
14360	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
14361	return rc;
14362	}
14363
14364	Assert(!pVmxTransient->fIsNestedGuest);
14365
14366	/* If it's a vectoring #PF, emulate injecting the original event injection as PGMTrap0eHandler() is incapable
14367	of differentiating between instruction emulation and event injection that caused a #PF. See @bugref{6607}. */
14368	if (pVmxTransient->fVectoringPF)
14369	{
14370	Assert(pVCpu->hm.s.Event.fPending);
14371	return VINF_EM_RAW_INJECT_TRPM_EVENT;
14372	}
14373
14374	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14375	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14376	AssertRCReturn(rc, rc);
14377
14378	Log4Func(("#PF: cr2=%#RX64 cs:rip=%#04x:%#RX64 uErrCode %#RX32 cr3=%#RX64\n", pVmxTransient->uExitQual, pCtx->cs.Sel,
14379	pCtx->rip, pVmxTransient->uExitIntErrorCode, pCtx->cr3));
14380
14381	TRPMAssertXcptPF(pVCpu, pVmxTransient->uExitQual, (RTGCUINT)pVmxTransient->uExitIntErrorCode);
14382	rc = PGMTrap0eHandler(pVCpu, pVmxTransient->uExitIntErrorCode, CPUMCTX2CORE(pCtx), (RTGCPTR)pVmxTransient->uExitQual);
14383
14384	Log4Func(("#PF: rc=%Rrc\n", rc));
14385	if (rc == VINF_SUCCESS)
14386	{
14387	/*
14388	* This is typically a shadow page table sync or a MMIO instruction. But we may have
14389	* emulated something like LTR or a far jump. Any part of the CPU context may have changed.
14390	*/
14391	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14392	TRPMResetTrap(pVCpu);
14393	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
14394	return rc;
14395	}
14396
14397	if (rc == VINF_EM_RAW_GUEST_TRAP)
14398	{
14399	if (!pVmxTransient->fVectoringDoublePF)
14400	{
14401	/* It's a guest page fault and needs to be reflected to the guest. */
14402	uint32_t uGstErrorCode = TRPMGetErrorCode(pVCpu);
14403	TRPMResetTrap(pVCpu);
14404	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory #PF. */
14405	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */,
14406	uGstErrorCode, pVmxTransient->uExitQual);
14407	}
14408	else
14409	{
14410	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
14411	TRPMResetTrap(pVCpu);
14412	pVCpu->hm.s.Event.fPending = false; /* Clear pending #PF to replace it with #DF. */
14413	hmR0VmxSetPendingXcptDF(pVCpu);
14414	Log4Func(("#PF: Pending #DF due to vectoring #PF\n"));
14415	}
14416
14417	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
14418	return VINF_SUCCESS;
14419	}
14420
14421	TRPMResetTrap(pVCpu);
14422	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
14423	return rc;
14424	}
14425
14426
14427	/**
14428	* VM-exit exception handler for \#MF (Math Fault: floating point exception).
14429	*
14430	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14431	*/
14432	static VBOXSTRICTRC hmR0VmxExitXcptMF(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
14433	{
14434	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14435	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
14436
14437	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR0);
14438	AssertRCReturn(rc, rc);
14439
14440	if (!(pVCpu->cpum.GstCtx.cr0 & X86_CR0_NE))
14441	{
14442	/* Convert a #MF into a FERR -> IRQ 13. See @bugref{6117}. */
14443	rc = PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13, 1, 0 /* uTagSrc */);
14444
14445	/** @todo r=ramshankar: The Intel spec. does -not- specify that this VM-exit
14446	* provides VM-exit instruction length. If this causes problem later,
14447	* disassemble the instruction like it's done on AMD-V. */
14448	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14449	AssertRCReturn(rc2, rc2);
14450	return rc;
14451	}
14452
14453	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbInstr,
14454	pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14455	return VINF_SUCCESS;
14456	}
14457
14458
14459	/**
14460	* VM-exit exception handler for \#BP (Breakpoint exception).
14461	*
14462	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14463	*/
14464	static VBOXSTRICTRC hmR0VmxExitXcptBP(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
14465	{
14466	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14467	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP);
14468
14469	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14470	AssertRCReturn(rc, rc);
14471
14472	if (!pVmxTransient->fIsNestedGuest)
14473	rc = DBGFRZTrap03Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(&pVCpu->cpum.GstCtx));
14474	else
14475	rc = VINF_EM_RAW_GUEST_TRAP;
14476	if (rc == VINF_EM_RAW_GUEST_TRAP)
14477	{
14478	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbInstr,
14479	pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14480	}
14481
14482	Assert(rc == VINF_SUCCESS \|\| rc == VINF_EM_RAW_GUEST_TRAP \|\| rc == VINF_EM_DBG_BREAKPOINT);
14483	return rc;
14484	}
14485
14486
14487	/**
14488	* VM-exit exception handler for \#AC (Alignment-check exception).
14489	*
14490	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14491	*/
14492	static VBOXSTRICTRC hmR0VmxExitXcptAC(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
14493	{
14494	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14495	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC);
14496
14497	/* Re-inject it. We'll detect any nesting before getting here. */
14498	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbInstr,
14499	pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14500	return VINF_SUCCESS;
14501	}
14502
14503
14504	/**
14505	* VM-exit exception handler for \#DB (Debug exception).
14506	*
14507	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14508	*/
14509	static VBOXSTRICTRC hmR0VmxExitXcptDB(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
14510	{
14511	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14512	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
14513
14514	/*
14515	* Get the DR6-like values from the Exit qualification and pass it to DBGF for processing.
14516	*/
14517	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
14518	AssertRCReturn(rc, rc);
14519
14520	/* Refer Intel spec. Table 27-1. "Exit Qualifications for debug exceptions" for the format. */
14521	uint64_t const uDR6 = X86_DR6_INIT_VAL
14522	\| (pVmxTransient->uExitQual & ( X86_DR6_B0 \| X86_DR6_B1 \| X86_DR6_B2 \| X86_DR6_B3
14523	\| X86_DR6_BD \| X86_DR6_BS));
14524
14525	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14526	if (!pVmxTransient->fIsNestedGuest)
14527	rc = DBGFRZTrap01Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx), uDR6, pVCpu->hm.s.fSingleInstruction);
14528	else
14529	rc = VINF_EM_RAW_GUEST_TRAP;
14530	Log6Func(("rc=%Rrc\n", rc));
14531	if (rc == VINF_EM_RAW_GUEST_TRAP)
14532	{
14533	/*
14534	* The exception was for the guest. Update DR6, DR7.GD and
14535	* IA32_DEBUGCTL.LBR before forwarding it.
14536	* See Intel spec. 27.1 "Architectural State before a VM-Exit".
14537	*/
14538	VMMRZCallRing3Disable(pVCpu);
14539	HM_DISABLE_PREEMPT(pVCpu);
14540
14541	pCtx->dr[6] &= ~X86_DR6_B_MASK;
14542	pCtx->dr[6] \|= uDR6;
14543	if (CPUMIsGuestDebugStateActive(pVCpu))
14544	ASMSetDR6(pCtx->dr[6]);
14545
14546	HM_RESTORE_PREEMPT();
14547	VMMRZCallRing3Enable(pVCpu);
14548
14549	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_DR7);
14550	AssertRCReturn(rc, rc);
14551
14552	/* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */
14553	pCtx->dr[7] &= ~X86_DR7_GD;
14554
14555	/* Paranoia. */
14556	pCtx->dr[7] &= ~X86_DR7_RAZ_MASK;
14557	pCtx->dr[7] \|= X86_DR7_RA1_MASK;
14558
14559	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, (uint32_t)pCtx->dr[7]);
14560	AssertRCReturn(rc, rc);
14561
14562	/*
14563	* Raise #DB in the guest.
14564	*
14565	* It is important to reflect exactly what the VM-exit gave us (preserving the
14566	* interruption-type) rather than use hmR0VmxSetPendingXcptDB() as the #DB could've
14567	* been raised while executing ICEBP (INT1) and not the regular #DB. Thus it may
14568	* trigger different handling in the CPU (like skipping DPL checks), see @bugref{6398}.
14569	*
14570	* Intel re-documented ICEBP/INT1 on May 2018 previously documented as part of
14571	* Intel 386, see Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
14572	*/
14573	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbInstr,
14574	pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14575	return VINF_SUCCESS;
14576	}
14577
14578	/*
14579	* Not a guest trap, must be a hypervisor related debug event then.
14580	* Update DR6 in case someone is interested in it.
14581	*/
14582	AssertMsg(rc == VINF_EM_DBG_STEPPED \|\| rc == VINF_EM_DBG_BREAKPOINT, ("%Rrc\n", rc));
14583	AssertReturn(pVmxTransient->fWasHyperDebugStateActive, VERR_HM_IPE_5);
14584	CPUMSetHyperDR6(pVCpu, uDR6);
14585
14586	return rc;
14587	}
14588
14589
14590	/**
14591	* Hacks its way around the lovely mesa driver's backdoor accesses.
14592	*
14593	* @sa hmR0SvmHandleMesaDrvGp.
14594	*/
14595	static int hmR0VmxHandleMesaDrvGp(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, PCPUMCTX pCtx)
14596	{
14597	LogFunc(("cs:rip=%#04x:%#RX64 rcx=%#RX64 rbx=%#RX64\n", pCtx->cs.Sel, pCtx->rip, pCtx->rcx, pCtx->rbx));
14598	RT_NOREF(pCtx);
14599
14600	/* For now we'll just skip the instruction. */
14601	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14602	}
14603
14604
14605	/**
14606	* Checks if the \#GP'ing instruction is the mesa driver doing it's lovely
14607	* backdoor logging w/o checking what it is running inside.
14608	*
14609	* This recognizes an "IN EAX,DX" instruction executed in flat ring-3, with the
14610	* backdoor port and magic numbers loaded in registers.
14611	*
14612	* @returns true if it is, false if it isn't.
14613	* @sa hmR0SvmIsMesaDrvGp.
14614	*/
14615	DECLINLINE(bool) hmR0VmxIsMesaDrvGp(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient, PCPUMCTX pCtx)
14616	{
14617	/* 0xed: IN eAX,dx */
14618	uint8_t abInstr[1];
14619	if (pVmxTransient->cbInstr != sizeof(abInstr))
14620	return false;
14621
14622	/* Check that it is #GP(0). */
14623	if (pVmxTransient->uExitIntErrorCode != 0)
14624	return false;
14625
14626	/* Check magic and port. */
14627	Assert(!(pCtx->fExtrn & (CPUMCTX_EXTRN_RAX \| CPUMCTX_EXTRN_RDX \| CPUMCTX_EXTRN_RCX)));
14628	/Log(("hmR0VmxIsMesaDrvGp: rax=%RX64 rdx=%RX64\n", pCtx->rax, pCtx->rdx));/
14629	if (pCtx->rax != UINT32_C(0x564d5868))
14630	return false;
14631	if (pCtx->dx != UINT32_C(0x5658))
14632	return false;
14633
14634	/* Flat ring-3 CS. */
14635	AssertCompile(HMVMX_CPUMCTX_EXTRN_ALL & CPUMCTX_EXTRN_CS);
14636	Assert(!(pCtx->fExtrn & CPUMCTX_EXTRN_CS));
14637	/Log(("hmR0VmxIsMesaDrvGp: cs.Attr.n.u2Dpl=%d base=%Rx64\n", pCtx->cs.Attr.n.u2Dpl, pCtx->cs.u64Base));/
14638	if (pCtx->cs.Attr.n.u2Dpl != 3)
14639	return false;
14640	if (pCtx->cs.u64Base != 0)
14641	return false;
14642
14643	/* Check opcode. */
14644	AssertCompile(HMVMX_CPUMCTX_EXTRN_ALL & CPUMCTX_EXTRN_RIP);
14645	Assert(!(pCtx->fExtrn & CPUMCTX_EXTRN_RIP));
14646	int rc = PGMPhysSimpleReadGCPtr(pVCpu, abInstr, pCtx->rip, sizeof(abInstr));
14647	/Log(("hmR0VmxIsMesaDrvGp: PGMPhysSimpleReadGCPtr -> %Rrc %#x\n", rc, abInstr[0]));/
14648	if (RT_FAILURE(rc))
14649	return false;
14650	if (abInstr[0] != 0xed)
14651	return false;
14652
14653	return true;
14654	}
14655
14656
14657	/**
14658	* VM-exit exception handler for \#GP (General-protection exception).
14659	*
14660	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14661	*/
14662	static VBOXSTRICTRC hmR0VmxExitXcptGP(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
14663	{
14664	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14665	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP);
14666
14667	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14668	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14669	if (pVmcsInfo->RealMode.fRealOnV86Active)
14670	{ /* likely */ }
14671	else
14672	{
14673	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14674	Assert(pVCpu->hm.s.fUsingDebugLoop \|\| pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv \|\| pVmxTransient->fIsNestedGuest);
14675	#endif
14676	/*
14677	* If the guest is not in real-mode or we have unrestricted guest execution support, or if we are
14678	* executing a nested-guest, reflect #GP to the guest or nested-guest.
14679	*/
14680	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14681	AssertRCReturn(rc, rc);
14682	Log4Func(("Gst: cs:rip=%#04x:%#RX64 ErrorCode=%#x cr0=%#RX64 cpl=%u tr=%#04x\n", pCtx->cs.Sel, pCtx->rip,
14683	pVmxTransient->uExitIntErrorCode, pCtx->cr0, CPUMGetGuestCPL(pVCpu), pCtx->tr.Sel));
14684
14685	if ( pVmxTransient->fIsNestedGuest
14686	\|\| !pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv
14687	\|\| !hmR0VmxIsMesaDrvGp(pVCpu, pVmxTransient, pCtx))
14688	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14689	pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14690	else
14691	rc = hmR0VmxHandleMesaDrvGp(pVCpu, pVmxTransient, pCtx);
14692	return rc;
14693	}
14694
14695	Assert(CPUMIsGuestInRealModeEx(pCtx));
14696	Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest);
14697	Assert(!pVmxTransient->fIsNestedGuest);
14698
14699	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14700	AssertRCReturn(rc, rc);
14701
14702	VBOXSTRICTRC rcStrict = IEMExecOne(pVCpu);
14703	if (rcStrict == VINF_SUCCESS)
14704	{
14705	if (!CPUMIsGuestInRealModeEx(pCtx))
14706	{
14707	/*
14708	* The guest is no longer in real-mode, check if we can continue executing the
14709	* guest using hardware-assisted VMX. Otherwise, fall back to emulation.
14710	*/
14711	pVmcsInfo->RealMode.fRealOnV86Active = false;
14712	if (HMCanExecuteVmxGuest(pVCpu, pCtx))
14713	{
14714	Log4Func(("Mode changed but guest still suitable for executing using hardware-assisted VMX\n"));
14715	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14716	}
14717	else
14718	{
14719	Log4Func(("Mode changed -> VINF_EM_RESCHEDULE\n"));
14720	rcStrict = VINF_EM_RESCHEDULE;
14721	}
14722	}
14723	else
14724	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14725	}
14726	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14727	{
14728	rcStrict = VINF_SUCCESS;
14729	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14730	}
14731	return VBOXSTRICTRC_VAL(rcStrict);
14732	}
14733
14734
14735	/**
14736	* VM-exit exception handler wrapper for all other exceptions that are not handled
14737	* by a specific handler.
14738	*
14739	* This simply re-injects the exception back into the VM without any special
14740	* processing.
14741	*
14742	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14743	*/
14744	static VBOXSTRICTRC hmR0VmxExitXcptOthers(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
14745	{
14746	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14747
14748	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14749	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14750	AssertMsg(pVCpu->hm.s.fUsingDebugLoop \|\| pVmcsInfo->RealMode.fRealOnV86Active \|\| pVmxTransient->fIsNestedGuest,
14751	("uVector=%#x u32XcptBitmap=%#X32\n",
14752	VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo), pVmcsInfo->u32XcptBitmap));
14753	NOREF(pVmcsInfo);
14754	#endif
14755
14756	/*
14757	* Re-inject the exception into the guest. This cannot be a double-fault condition which
14758	* would have been handled while checking exits due to event delivery.
14759	*/
14760	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14761
14762	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14763	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
14764	AssertRCReturn(rc, rc);
14765	Log4Func(("Reinjecting Xcpt. uVector=%#x cs:rip=%#04x:%#RX64\n", uVector, pCtx->cs.Sel, pCtx->rip));
14766	#endif
14767
14768	#ifdef VBOX_WITH_STATISTICS
14769	switch (uVector)
14770	{
14771	case X86_XCPT_DE: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE); break;
14772	case X86_XCPT_DB: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB); break;
14773	case X86_XCPT_BP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP); break;
14774	case X86_XCPT_OF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
14775	case X86_XCPT_BR: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBR); break;
14776	case X86_XCPT_UD: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD); break;
14777	case X86_XCPT_NM: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
14778	case X86_XCPT_DF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDF); break;
14779	case X86_XCPT_TS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestTS); break;
14780	case X86_XCPT_NP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNP); break;
14781	case X86_XCPT_SS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestSS); break;
14782	case X86_XCPT_GP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP); break;
14783	case X86_XCPT_PF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF); break;
14784	case X86_XCPT_MF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF); break;
14785	case X86_XCPT_AC: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC); break;
14786	case X86_XCPT_XF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXF); break;
14787	default:
14788	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXcpUnk);
14789	break;
14790	}
14791	#endif
14792
14793	/* We should never call this function for a page-fault, we'd need to pass on the fault address below otherwise. */
14794	Assert(!VMX_EXIT_INT_INFO_IS_XCPT_PF(pVmxTransient->uExitIntInfo));
14795	NOREF(uVector);
14796
14797	/* Re-inject the original exception into the guest. */
14798	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbInstr,
14799	pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14800	return VINF_SUCCESS;
14801	}
14802
14803
14804	/**
14805	* VM-exit exception handler for all exceptions (except NMIs!).
14806	*
14807	* @remarks This may be called for both guests and nested-guests. Take care to not
14808	* make assumptions and avoid doing anything that is not relevant when
14809	* executing a nested-guest (e.g., Mesa driver hacks).
14810	*/
14811	static VBOXSTRICTRC hmR0VmxExitXcpt(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
14812	{
14813	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_XCPT_INFO);
14814
14815	/*
14816	* If this VM-exit occurred while delivering an event through the guest IDT, take
14817	* action based on the return code and additional hints (e.g. for page-faults)
14818	* that will be updated in the VMX transient structure.
14819	*/
14820	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
14821	if (rcStrict == VINF_SUCCESS)
14822	{
14823	/*
14824	* If an exception caused a VM-exit due to delivery of an event, the original
14825	* event may have to be re-injected into the guest. We shall reinject it and
14826	* continue guest execution. However, page-fault is a complicated case and
14827	* needs additional processing done in hmR0VmxExitXcptPF().
14828	*/
14829	Assert(VMX_EXIT_INT_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
14830	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14831	if ( !pVCpu->hm.s.Event.fPending
14832	\|\| uVector == X86_XCPT_PF)
14833	{
14834	switch (uVector)
14835	{
14836	case X86_XCPT_PF: return hmR0VmxExitXcptPF(pVCpu, pVmxTransient);
14837	case X86_XCPT_GP: return hmR0VmxExitXcptGP(pVCpu, pVmxTransient);
14838	case X86_XCPT_MF: return hmR0VmxExitXcptMF(pVCpu, pVmxTransient);
14839	case X86_XCPT_DB: return hmR0VmxExitXcptDB(pVCpu, pVmxTransient);
14840	case X86_XCPT_BP: return hmR0VmxExitXcptBP(pVCpu, pVmxTransient);
14841	case X86_XCPT_AC: return hmR0VmxExitXcptAC(pVCpu, pVmxTransient);
14842	default:
14843	return hmR0VmxExitXcptOthers(pVCpu, pVmxTransient);
14844	}
14845	}
14846	/* else: inject pending event before resuming guest execution. */
14847	}
14848	else if (rcStrict == VINF_HM_DOUBLE_FAULT)
14849	{
14850	Assert(pVCpu->hm.s.Event.fPending);
14851	rcStrict = VINF_SUCCESS;
14852	}
14853
14854	return rcStrict;
14855	}
14856	/** @} */
14857
14858
14859	/** @name VM-exit handlers.
14860	* @{
14861	*/
14862	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
14863	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
14864	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
14865
14866	/**
14867	* VM-exit handler for external interrupts (VMX_EXIT_EXT_INT).
14868	*/
14869	HMVMX_EXIT_DECL hmR0VmxExitExtInt(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
14870	{
14871	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14872	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt);
14873	/* Windows hosts (32-bit and 64-bit) have DPC latency issues. See @bugref{6853}. */
14874	if (VMMR0ThreadCtxHookIsEnabled(pVCpu))
14875	return VINF_SUCCESS;
14876	return VINF_EM_RAW_INTERRUPT;
14877	}
14878
14879
14880	/**
14881	* VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI). Conditional
14882	* VM-exit.
14883	*/
14884	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmi(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
14885	{
14886	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14887	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitXcptNmi, y3);
14888
14889	int rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
14890	AssertRCReturn(rc, rc);
14891
14892	uint32_t const uExitIntType = VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo);
14893	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14894	Assert(VMX_EXIT_INT_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
14895
14896	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14897	Assert( !(pVmcsInfo->u32ExitCtls & VMX_EXIT_CTLS_ACK_EXT_INT)
14898	&& uExitIntType != VMX_EXIT_INT_INFO_TYPE_EXT_INT);
14899	NOREF(pVmcsInfo);
14900
14901	VBOXSTRICTRC rcStrict;
14902	switch (uExitIntType)
14903	{
14904	/*
14905	* Host physical NMIs:
14906	* This cannot be a guest NMI as the only way for the guest to receive an NMI is if we
14907	* injected it ourselves and anything we inject is not going to cause a VM-exit directly
14908	* for the event being injected[1]. Go ahead and dispatch the NMI to the host[2].
14909	*
14910	* See Intel spec. 27.2.3 "Information for VM Exits During Event Delivery".
14911	* See Intel spec. 27.5.5 "Updating Non-Register State".
14912	*/
14913	case VMX_EXIT_INT_INFO_TYPE_NMI:
14914	{
14915	rcStrict = hmR0VmxExitHostNmi(pVCpu, pVmcsInfo);
14916	break;
14917	}
14918
14919	/*
14920	* Privileged software exceptions (#DB from ICEBP),
14921	* Software exceptions (#BP and #OF),
14922	* Hardware exceptions:
14923	* Process the required exceptions and resume guest execution if possible.
14924	*/
14925	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
14926	Assert(uVector == X86_XCPT_DB);
14927	RT_FALL_THRU();
14928	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
14929	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF \|\| uExitIntType == VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT);
14930	RT_FALL_THRU();
14931	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
14932	{
14933	NOREF(uVector);
14934	rc = hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
14935	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14936	rc \|= hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
14937	rc \|= hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
14938	AssertRCReturn(rc, rc);
14939
14940	rcStrict = hmR0VmxExitXcpt(pVCpu, pVmxTransient);
14941	break;
14942	}
14943
14944	default:
14945	{
14946	pVCpu->hm.s.u32HMError = pVmxTransient->uExitIntInfo;
14947	rcStrict = VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
14948	AssertMsgFailed(("Invalid/unexpected VM-exit interruption info %#x\n", pVmxTransient->uExitIntInfo));
14949	break;
14950	}
14951	}
14952
14953	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
14954	return rcStrict;
14955	}
14956
14957
14958	/**
14959	* VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
14960	*/
14961	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindow(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
14962	{
14963	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14964
14965	/* Indicate that we no longer need to VM-exit when the guest is ready to receive interrupts, it is now ready. */
14966	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14967	int rc = hmR0VmxClearIntWindowExitVmcs(pVmcsInfo);
14968	AssertRCReturn(rc, rc);
14969
14970	/* Evaluate and deliver pending events and resume guest execution. */
14971	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow);
14972	return VINF_SUCCESS;
14973	}
14974
14975
14976	/**
14977	* VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
14978	*/
14979	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindow(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
14980	{
14981	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14982
14983	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14984	if (RT_UNLIKELY(!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))) /** @todo NSTVMX: Turn this into an assertion. */
14985	{
14986	AssertMsgFailed(("Unexpected NMI-window exit.\n"));
14987	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
14988	}
14989
14990	Assert(!CPUMIsGuestNmiBlocking(pVCpu));
14991
14992	/*
14993	* If block-by-STI is set when we get this VM-exit, it means the CPU doesn't block NMIs following STI.
14994	* It is therefore safe to unblock STI and deliver the NMI ourselves. See @bugref{7445}.
14995	*/
14996	uint32_t fIntrState;
14997	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
14998	AssertRCReturn(rc, rc);
14999	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
15000	if (fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
15001	{
15002	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
15003	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
15004
15005	fIntrState &= ~VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
15006	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
15007	AssertRCReturn(rc, rc);
15008	}
15009
15010	/* Indicate that we no longer need to VM-exit when the guest is ready to receive NMIs, it is now ready */
15011	rc = hmR0VmxClearNmiWindowExitVmcs(pVmcsInfo);
15012	AssertRCReturn(rc, rc);
15013
15014	/* Evaluate and deliver pending events and resume guest execution. */
15015	return VINF_SUCCESS;
15016	}
15017
15018
15019	/**
15020	* VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
15021	*/
15022	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvd(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15023	{
15024	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15025	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15026	}
15027
15028
15029	/**
15030	* VM-exit handler for INVD (VMX_EXIT_INVD). Unconditional VM-exit.
15031	*/
15032	HMVMX_EXIT_NSRC_DECL hmR0VmxExitInvd(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15033	{
15034	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15035	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15036	}
15037
15038
15039	/**
15040	* VM-exit handler for CPUID (VMX_EXIT_CPUID). Unconditional VM-exit.
15041	*/
15042	HMVMX_EXIT_DECL hmR0VmxExitCpuid(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15043	{
15044	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15045
15046	/*
15047	* Get the state we need and update the exit history entry.
15048	*/
15049	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15050	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15051	rc \|= hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
15052	AssertRCReturn(rc, rc);
15053
15054	VBOXSTRICTRC rcStrict;
15055	PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
15056	EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_CPUID),
15057	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
15058	if (!pExitRec)
15059	{
15060	/*
15061	* Regular CPUID instruction execution.
15062	*/
15063	rcStrict = IEMExecDecodedCpuid(pVCpu, pVmxTransient->cbInstr);
15064	if (rcStrict == VINF_SUCCESS)
15065	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15066	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15067	{
15068	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15069	rcStrict = VINF_SUCCESS;
15070	}
15071	}
15072	else
15073	{
15074	/*
15075	* Frequent exit or something needing probing. Get state and call EMHistoryExec.
15076	*/
15077	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
15078	AssertRCReturn(rc2, rc2);
15079
15080	Log4(("CpuIdExit/%u: %04x:%08RX64: %#x/%#x -> EMHistoryExec\n",
15081	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.eax, pVCpu->cpum.GstCtx.ecx));
15082
15083	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
15084	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
15085
15086	Log4(("CpuIdExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
15087	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
15088	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
15089	}
15090	return rcStrict;
15091	}
15092
15093
15094	/**
15095	* VM-exit handler for GETSEC (VMX_EXIT_GETSEC). Unconditional VM-exit.
15096	*/
15097	HMVMX_EXIT_DECL hmR0VmxExitGetsec(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15098	{
15099	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15100
15101	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15102	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR4);
15103	AssertRCReturn(rc, rc);
15104
15105	if (pVCpu->cpum.GstCtx.cr4 & X86_CR4_SMXE)
15106	return VINF_EM_RAW_EMULATE_INSTR;
15107
15108	AssertMsgFailed(("hmR0VmxExitGetsec: Unexpected VM-exit when CR4.SMXE is 0.\n"));
15109	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
15110	}
15111
15112
15113	/**
15114	* VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
15115	*/
15116	HMVMX_EXIT_DECL hmR0VmxExitRdtsc(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15117	{
15118	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15119
15120	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15121	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
15122	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15123	AssertRCReturn(rc, rc);
15124
15125	VBOXSTRICTRC rcStrict = IEMExecDecodedRdtsc(pVCpu, pVmxTransient->cbInstr);
15126	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15127	{
15128	/* If we get a spurious VM-exit when TSC offsetting is enabled,
15129	we must reset offsetting on VM-entry. See @bugref{6634}. */
15130	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TSC_OFFSETTING)
15131	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
15132	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15133	}
15134	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15135	{
15136	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15137	rcStrict = VINF_SUCCESS;
15138	}
15139	return rcStrict;
15140	}
15141
15142
15143	/**
15144	* VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
15145	*/
15146	HMVMX_EXIT_DECL hmR0VmxExitRdtscp(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15147	{
15148	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15149
15150	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15151	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_TSC_AUX);
15152	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15153	AssertRCReturn(rc, rc);
15154
15155	VBOXSTRICTRC rcStrict = IEMExecDecodedRdtscp(pVCpu, pVmxTransient->cbInstr);
15156	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15157	{
15158	/* If we get a spurious VM-exit when TSC offsetting is enabled,
15159	we must reset offsetting on VM-reentry. See @bugref{6634}. */
15160	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TSC_OFFSETTING)
15161	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
15162	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15163	}
15164	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15165	{
15166	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15167	rcStrict = VINF_SUCCESS;
15168	}
15169	return rcStrict;
15170	}
15171
15172
15173	/**
15174	* VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
15175	*/
15176	HMVMX_EXIT_DECL hmR0VmxExitRdpmc(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15177	{
15178	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15179
15180	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15181	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR4 \| CPUMCTX_EXTRN_CR0
15182	\| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS);
15183	AssertRCReturn(rc, rc);
15184
15185	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15186	rc = EMInterpretRdpmc(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx));
15187	if (RT_LIKELY(rc == VINF_SUCCESS))
15188	{
15189	rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15190	Assert(pVmxTransient->cbInstr == 2);
15191	}
15192	else
15193	{
15194	AssertMsgFailed(("hmR0VmxExitRdpmc: EMInterpretRdpmc failed with %Rrc\n", rc));
15195	rc = VERR_EM_INTERPRETER;
15196	}
15197	return rc;
15198	}
15199
15200
15201	/**
15202	* VM-exit handler for VMCALL (VMX_EXIT_VMCALL). Unconditional VM-exit.
15203	*/
15204	HMVMX_EXIT_DECL hmR0VmxExitVmcall(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15205	{
15206	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15207
15208	VBOXSTRICTRC rcStrict = VERR_VMX_IPE_3;
15209	if (EMAreHypercallInstructionsEnabled(pVCpu))
15210	{
15211	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15212	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_CR0
15213	\| CPUMCTX_EXTRN_SS \| CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_EFER);
15214	AssertRCReturn(rc, rc);
15215
15216	/* Perform the hypercall. */
15217	rcStrict = GIMHypercall(pVCpu, &pVCpu->cpum.GstCtx);
15218	if (rcStrict == VINF_SUCCESS)
15219	{
15220	rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15221	AssertRCReturn(rc, rc);
15222	}
15223	else
15224	Assert( rcStrict == VINF_GIM_R3_HYPERCALL
15225	\|\| rcStrict == VINF_GIM_HYPERCALL_CONTINUING
15226	\|\| RT_FAILURE(rcStrict));
15227
15228	/* If the hypercall changes anything other than guest's general-purpose registers,
15229	we would need to reload the guest changed bits here before VM-entry. */
15230	}
15231	else
15232	Log4Func(("Hypercalls not enabled\n"));
15233
15234	/* If hypercalls are disabled or the hypercall failed for some reason, raise #UD and continue. */
15235	if (RT_FAILURE(rcStrict))
15236	{
15237	hmR0VmxSetPendingXcptUD(pVCpu);
15238	rcStrict = VINF_SUCCESS;
15239	}
15240
15241	return rcStrict;
15242	}
15243
15244
15245	/**
15246	* VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
15247	*/
15248	HMVMX_EXIT_DECL hmR0VmxExitInvlpg(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15249	{
15250	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15251	Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging \|\| pVCpu->hm.s.fUsingDebugLoop);
15252
15253	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15254	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
15255	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15256	rc \|= hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
15257	AssertRCReturn(rc, rc);
15258
15259	VBOXSTRICTRC rcStrict = IEMExecDecodedInvlpg(pVCpu, pVmxTransient->cbInstr, pVmxTransient->uExitQual);
15260
15261	if (rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_PGM_SYNC_CR3)
15262	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15263	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15264	{
15265	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15266	rcStrict = VINF_SUCCESS;
15267	}
15268	else
15269	AssertMsgFailed(("Unexpected IEMExecDecodedInvlpg(%#RX64) status: %Rrc\n", pVmxTransient->uExitQual,
15270	VBOXSTRICTRC_VAL(rcStrict)));
15271	return rcStrict;
15272	}
15273
15274
15275	/**
15276	* VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
15277	*/
15278	HMVMX_EXIT_DECL hmR0VmxExitMonitor(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15279	{
15280	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15281
15282	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15283	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK \| CPUMCTX_EXTRN_DS);
15284	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15285	AssertRCReturn(rc, rc);
15286
15287	VBOXSTRICTRC rcStrict = IEMExecDecodedMonitor(pVCpu, pVmxTransient->cbInstr);
15288	if (rcStrict == VINF_SUCCESS)
15289	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15290	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15291	{
15292	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15293	rcStrict = VINF_SUCCESS;
15294	}
15295
15296	return rcStrict;
15297	}
15298
15299
15300	/**
15301	* VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
15302	*/
15303	HMVMX_EXIT_DECL hmR0VmxExitMwait(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15304	{
15305	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15306
15307	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15308	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
15309	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15310	AssertRCReturn(rc, rc);
15311
15312	VBOXSTRICTRC rcStrict = IEMExecDecodedMwait(pVCpu, pVmxTransient->cbInstr);
15313	if (RT_SUCCESS(rcStrict))
15314	{
15315	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15316	if (EMMonitorWaitShouldContinue(pVCpu, &pVCpu->cpum.GstCtx))
15317	rcStrict = VINF_SUCCESS;
15318	}
15319
15320	return rcStrict;
15321	}
15322
15323
15324	/**
15325	* VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT). Unconditional
15326	* VM-exit.
15327	*/
15328	HMVMX_EXIT_DECL hmR0VmxExitTripleFault(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15329	{
15330	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15331	return VINF_EM_RESET;
15332	}
15333
15334
15335	/**
15336	* VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
15337	*/
15338	HMVMX_EXIT_DECL hmR0VmxExitHlt(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15339	{
15340	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15341
15342	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15343	AssertRCReturn(rc, rc);
15344
15345	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS); /* Advancing the RIP above should've imported eflags. */
15346	if (EMShouldContinueAfterHalt(pVCpu, &pVCpu->cpum.GstCtx)) /* Requires eflags. */
15347	rc = VINF_SUCCESS;
15348	else
15349	rc = VINF_EM_HALT;
15350
15351	if (rc != VINF_SUCCESS)
15352	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHltToR3);
15353	return rc;
15354	}
15355
15356
15357	/**
15358	* VM-exit handler for instructions that result in a \#UD exception delivered to
15359	* the guest.
15360	*/
15361	HMVMX_EXIT_NSRC_DECL hmR0VmxExitSetPendingXcptUD(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15362	{
15363	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15364	hmR0VmxSetPendingXcptUD(pVCpu);
15365	return VINF_SUCCESS;
15366	}
15367
15368
15369	/**
15370	* VM-exit handler for expiry of the VMX-preemption timer.
15371	*/
15372	HMVMX_EXIT_DECL hmR0VmxExitPreemptTimer(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15373	{
15374	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15375
15376	/* If the VMX-preemption timer has expired, reinitialize the preemption timer on next VM-entry. */
15377	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
15378
15379	/* If there are any timer events pending, fall back to ring-3, otherwise resume guest execution. */
15380	PVM pVM = pVCpu->CTX_SUFF(pVM);
15381	bool fTimersPending = TMTimerPollBool(pVM, pVCpu);
15382	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPreemptTimer);
15383	return fTimersPending ? VINF_EM_RAW_TIMER_PENDING : VINF_SUCCESS;
15384	}
15385
15386
15387	/**
15388	* VM-exit handler for XSETBV (VMX_EXIT_XSETBV). Unconditional VM-exit.
15389	*/
15390	HMVMX_EXIT_DECL hmR0VmxExitXsetbv(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15391	{
15392	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15393
15394	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15395	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15396	rc \|= hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_CR4);
15397	AssertRCReturn(rc, rc);
15398
15399	VBOXSTRICTRC rcStrict = IEMExecDecodedXsetbv(pVCpu, pVmxTransient->cbInstr);
15400	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, rcStrict != VINF_IEM_RAISED_XCPT ? HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
15401	: HM_CHANGED_RAISED_XCPT_MASK);
15402
15403	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15404	pVCpu->hm.s.fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
15405
15406	return rcStrict;
15407	}
15408
15409
15410	/**
15411	* VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
15412	*/
15413	HMVMX_EXIT_DECL hmR0VmxExitInvpcid(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15414	{
15415	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15416	/** @todo Use VM-exit instruction information. */
15417	return VERR_EM_INTERPRETER;
15418	}
15419
15420
15421	/**
15422	* VM-exit handler for invalid-guest-state (VMX_EXIT_ERR_INVALID_GUEST_STATE). Error
15423	* VM-exit.
15424	*/
15425	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrInvalidGuestState(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15426	{
15427	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15428	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
15429	AssertRCReturn(rc, rc);
15430
15431	rc = hmR0VmxCheckVmcsCtls(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest);
15432	if (RT_FAILURE(rc))
15433	return rc;
15434
15435	uint32_t const uInvalidReason = hmR0VmxCheckGuestState(pVCpu, pVmcsInfo);
15436	NOREF(uInvalidReason);
15437
15438	#ifdef VBOX_STRICT
15439	uint32_t fIntrState;
15440	RTHCUINTREG uHCReg;
15441	uint64_t u64Val;
15442	uint32_t u32Val;
15443	rc = hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
15444	rc \|= hmR0VmxReadEntryXcptErrorCodeVmcs(pVmxTransient);
15445	rc \|= hmR0VmxReadEntryInstrLenVmcs(pVmxTransient);
15446	rc \|= VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
15447	AssertRCReturn(rc, rc);
15448
15449	Log4(("uInvalidReason %u\n", uInvalidReason));
15450	Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", pVmxTransient->uEntryIntInfo));
15451	Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", pVmxTransient->uEntryXcptErrorCode));
15452	Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %#RX32\n", pVmxTransient->cbEntryInstr));
15453	Log4(("VMX_VMCS32_GUEST_INT_STATE %#RX32\n", fIntrState));
15454
15455	rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32Val); AssertRC(rc);
15456	Log4(("VMX_VMCS_GUEST_CR0 %#RX32\n", u32Val));
15457	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, &uHCReg); AssertRC(rc);
15458	Log4(("VMX_VMCS_CTRL_CR0_MASK %#RHr\n", uHCReg));
15459	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uHCReg); AssertRC(rc);
15460	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
15461	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, &uHCReg); AssertRC(rc);
15462	Log4(("VMX_VMCS_CTRL_CR4_MASK %#RHr\n", uHCReg));
15463	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uHCReg); AssertRC(rc);
15464	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
15465	if (pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging)
15466	{
15467	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
15468	Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
15469	}
15470
15471	hmR0DumpRegs(pVCpu);
15472	#endif
15473
15474	return VERR_VMX_INVALID_GUEST_STATE;
15475	}
15476
15477	/**
15478	* VM-exit handler for all undefined/unexpected reasons. Should never happen.
15479	*/
15480	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrUnexpected(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15481	{
15482	/*
15483	* Cummulative notes of all recognized but unexpected VM-exits.
15484	*
15485	* 1. This does -not- cover scenarios like like a page-fault VM-exit occurring when
15486	* nested-paging is used.
15487	*
15488	* 2. Any instruction that causes a VM-exit unconditionally (for e.g. VMXON) must be
15489	* emulated or a #UD must be raised in the guest. Therefore, we should -not- be using
15490	* this function (and thereby stop VM execution) for handling such instructions.
15491	*
15492	*
15493	* VMX_EXIT_INIT_SIGNAL:
15494	* INIT signals are blocked in VMX root operation by VMXON and by SMI in SMM.
15495	* It is -NOT- blocked in VMX non-root operation so we can, in theory, still get these
15496	* VM-exits. However, we should not receive INIT signals VM-exit while executing a VM.
15497	*
15498	* See Intel spec. 33.14.1 Default Treatment of SMI Delivery"
15499	* See Intel spec. 29.3 "VMX Instructions" for "VMXON".
15500	* See Intel spec. "23.8 Restrictions on VMX operation".
15501	*
15502	* VMX_EXIT_SIPI:
15503	* SIPI exits can only occur in VMX non-root operation when the "wait-for-SIPI" guest
15504	* activity state is used. We don't make use of it as our guests don't have direct
15505	* access to the host local APIC.
15506	*
15507	* See Intel spec. 25.3 "Other Causes of VM-exits".
15508	*
15509	* VMX_EXIT_IO_SMI:
15510	* VMX_EXIT_SMI:
15511	* This can only happen if we support dual-monitor treatment of SMI, which can be
15512	* activated by executing VMCALL in VMX root operation. Only an STM (SMM transfer
15513	* monitor) would get this VM-exit when we (the executive monitor) execute a VMCALL in
15514	* VMX root mode or receive an SMI. If we get here, something funny is going on.
15515	*
15516	* See Intel spec. 33.15.6 "Activating the Dual-Monitor Treatment"
15517	* See Intel spec. 25.3 "Other Causes of VM-Exits"
15518	*
15519	* VMX_EXIT_ERR_MSR_LOAD:
15520	* Failures while loading MSRs are part of the VM-entry MSR-load area are unexpected
15521	* and typically indicates a bug in the hypervisor code. We thus cannot not resume
15522	* execution.
15523	*
15524	* See Intel spec. 26.7 "VM-Entry Failures During Or After Loading Guest State".
15525	*
15526	* VMX_EXIT_ERR_MACHINE_CHECK:
15527	* Machine check exceptions indicates a fatal/unrecoverable hardware condition
15528	* including but not limited to system bus, ECC, parity, cache and TLB errors. A
15529	* #MC exception abort class exception is raised. We thus cannot assume a
15530	* reasonable chance of continuing any sort of execution and we bail.
15531	*
15532	* See Intel spec. 15.1 "Machine-check Architecture".
15533	* See Intel spec. 27.1 "Architectural State Before A VM Exit".
15534	*
15535	* VMX_EXIT_PML_FULL:
15536	* VMX_EXIT_VIRTUALIZED_EOI:
15537	* VMX_EXIT_APIC_WRITE:
15538	* We do not currently support any of these features and thus they are all unexpected
15539	* VM-exits.
15540	*
15541	* VMX_EXIT_GDTR_IDTR_ACCESS:
15542	* VMX_EXIT_LDTR_TR_ACCESS:
15543	* VMX_EXIT_RDRAND:
15544	* VMX_EXIT_RSM:
15545	* VMX_EXIT_VMFUNC:
15546	* VMX_EXIT_ENCLS:
15547	* VMX_EXIT_RDSEED:
15548	* VMX_EXIT_XSAVES:
15549	* VMX_EXIT_XRSTORS:
15550	* VMX_EXIT_UMWAIT:
15551	* VMX_EXIT_TPAUSE:
15552	* These VM-exits are -not- caused unconditionally by execution of the corresponding
15553	* instruction. Any VM-exit for these instructions indicate a hardware problem,
15554	* unsupported CPU modes (like SMM) or potentially corrupt VMCS controls.
15555	*
15556	* See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally".
15557	*/
15558	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15559	AssertMsgFailed(("Unexpected VM-exit %u\n", pVmxTransient->uExitReason));
15560	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
15561	}
15562
15563
15564	/**
15565	* VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
15566	*/
15567	HMVMX_EXIT_DECL hmR0VmxExitRdmsr(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15568	{
15569	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15570
15571	/** @todo Optimize this: We currently drag in in the whole MSR state
15572	* (CPUMCTX_EXTRN_ALL_MSRS) here. We should optimize this to only get
15573	* MSRs required. That would require changes to IEM and possibly CPUM too.
15574	* (Should probably do it lazy fashion from CPUMAllMsrs.cpp). */
15575	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15576	uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
15577	uint64_t fImport = IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK \| CPUMCTX_EXTRN_ALL_MSRS;
15578	switch (idMsr)
15579	{
15580	case MSR_K8_FS_BASE: fImport \|= CPUMCTX_EXTRN_FS; break;
15581	case MSR_K8_GS_BASE: fImport \|= CPUMCTX_EXTRN_GS; break;
15582	}
15583
15584	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15585	rc \|= hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImport);
15586	AssertRCReturn(rc, rc);
15587
15588	Log4Func(("ecx=%#RX32\n", idMsr));
15589
15590	#ifdef VBOX_STRICT
15591	Assert(!pVmxTransient->fIsNestedGuest);
15592	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
15593	{
15594	if ( hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr)
15595	&& idMsr != MSR_K6_EFER)
15596	{
15597	AssertMsgFailed(("Unexpected RDMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n", idMsr));
15598	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15599	}
15600	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15601	{
15602	Assert(pVmcsInfo->pvMsrBitmap);
15603	uint32_t fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, idMsr);
15604	if (fMsrpm & VMXMSRPM_ALLOW_RD)
15605	{
15606	AssertMsgFailed(("Unexpected RDMSR for a passthru lazy-restore MSR. ecx=%#RX32\n", idMsr));
15607	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15608	}
15609	}
15610	}
15611	#endif
15612
15613	VBOXSTRICTRC rcStrict = IEMExecDecodedRdmsr(pVCpu, pVmxTransient->cbInstr);
15614	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr);
15615	if (rcStrict == VINF_SUCCESS)
15616	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
15617	\| HM_CHANGED_GUEST_RAX \| HM_CHANGED_GUEST_RDX);
15618	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15619	{
15620	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15621	rcStrict = VINF_SUCCESS;
15622	}
15623	else
15624	AssertMsg(rcStrict == VINF_CPUM_R3_MSR_READ, ("Unexpected IEMExecDecodedRdmsr rc (%Rrc)\n", VBOXSTRICTRC_VAL(rcStrict)));
15625
15626	return rcStrict;
15627	}
15628
15629
15630	/**
15631	* VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
15632	*/
15633	HMVMX_EXIT_DECL hmR0VmxExitWrmsr(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15634	{
15635	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15636
15637	/** @todo Optimize this: We currently drag in in the whole MSR state
15638	* (CPUMCTX_EXTRN_ALL_MSRS) here. We should optimize this to only get
15639	* MSRs required. That would require changes to IEM and possibly CPUM too.
15640	* (Should probably do it lazy fashion from CPUMAllMsrs.cpp). */
15641	uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
15642	uint64_t fImport = IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK \| CPUMCTX_EXTRN_ALL_MSRS;
15643
15644	/*
15645	* The FS and GS base MSRs are not part of the above all-MSRs mask.
15646	* Although we don't need to fetch the base as it will be overwritten shortly, while
15647	* loading guest-state we would also load the entire segment register including limit
15648	* and attributes and thus we need to load them here.
15649	*/
15650	switch (idMsr)
15651	{
15652	case MSR_K8_FS_BASE: fImport \|= CPUMCTX_EXTRN_FS; break;
15653	case MSR_K8_GS_BASE: fImport \|= CPUMCTX_EXTRN_GS; break;
15654	}
15655
15656	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15657	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15658	rc \|= hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImport);
15659	AssertRCReturn(rc, rc);
15660
15661	Log4Func(("ecx=%#RX32 edx:eax=%#RX32:%#RX32\n", idMsr, pVCpu->cpum.GstCtx.edx, pVCpu->cpum.GstCtx.eax));
15662
15663	VBOXSTRICTRC rcStrict = IEMExecDecodedWrmsr(pVCpu, pVmxTransient->cbInstr);
15664	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr);
15665
15666	if (rcStrict == VINF_SUCCESS)
15667	{
15668	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15669
15670	/* If this is an X2APIC WRMSR access, update the APIC state as well. */
15671	if ( idMsr == MSR_IA32_APICBASE
15672	\|\| ( idMsr >= MSR_IA32_X2APIC_START
15673	&& idMsr <= MSR_IA32_X2APIC_END))
15674	{
15675	/*
15676	* We've already saved the APIC related guest-state (TPR) in post-run phase.
15677	* When full APIC register virtualization is implemented we'll have to make
15678	* sure APIC state is saved from the VMCS before IEM changes it.
15679	*/
15680	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
15681	}
15682	else if (idMsr == MSR_IA32_TSC) /* Windows 7 does this during bootup. See @bugref{6398}. */
15683	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
15684	else if (idMsr == MSR_K6_EFER)
15685	{
15686	/*
15687	* If the guest touches the EFER MSR we need to update the VM-Entry and VM-Exit controls
15688	* as well, even if it is -not- touching bits that cause paging mode changes (LMA/LME).
15689	* We care about the other bits as well, SCE and NXE. See @bugref{7368}.
15690	*/
15691	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_EFER_MSR \| HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
15692	}
15693
15694	/* Update MSRs that are part of the VMCS and auto-load/store area when MSR-bitmaps are not used. */
15695	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS))
15696	{
15697	switch (idMsr)
15698	{
15699	case MSR_IA32_SYSENTER_CS: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_CS_MSR); break;
15700	case MSR_IA32_SYSENTER_EIP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); break;
15701	case MSR_IA32_SYSENTER_ESP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); break;
15702	case MSR_K8_FS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_FS); break;
15703	case MSR_K8_GS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_GS); break;
15704	case MSR_K6_EFER: /* Nothing to do, already handled above. */ break;
15705	default:
15706	{
15707	if (hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr))
15708	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
15709	else if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15710	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_VMX_GUEST_LAZY_MSRS);
15711	break;
15712	}
15713	}
15714	}
15715	#ifdef VBOX_STRICT
15716	else
15717	{
15718	/* Paranoia. Validate that MSRs in the MSR-bitmaps with write-passthru are not intercepted. */
15719	switch (idMsr)
15720	{
15721	case MSR_IA32_SYSENTER_CS:
15722	case MSR_IA32_SYSENTER_EIP:
15723	case MSR_IA32_SYSENTER_ESP:
15724	case MSR_K8_FS_BASE:
15725	case MSR_K8_GS_BASE:
15726	{
15727	AssertMsgFailed(("Unexpected WRMSR for an MSR in the VMCS. ecx=%#RX32\n", idMsr));
15728	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15729	}
15730
15731	/* Writes to MSRs in auto-load/store area/swapped MSRs, shouldn't cause VM-exits with MSR-bitmaps. */
15732	default:
15733	{
15734	if (hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr))
15735	{
15736	/* EFER MSR writes are always intercepted. */
15737	if (idMsr != MSR_K6_EFER)
15738	{
15739	AssertMsgFailed(("Unexpected WRMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n",
15740	idMsr));
15741	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15742	}
15743	}
15744
15745	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15746	{
15747	Assert(pVmcsInfo->pvMsrBitmap);
15748	uint32_t fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, idMsr);
15749	if (fMsrpm & VMXMSRPM_ALLOW_WR)
15750	{
15751	AssertMsgFailed(("Unexpected WRMSR for passthru, lazy-restore MSR. ecx=%#RX32\n", idMsr));
15752	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15753	}
15754	}
15755	break;
15756	}
15757	}
15758	}
15759	#endif /* VBOX_STRICT */
15760	}
15761	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15762	{
15763	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15764	rcStrict = VINF_SUCCESS;
15765	}
15766	else
15767	AssertMsg(rcStrict == VINF_CPUM_R3_MSR_WRITE, ("Unexpected IEMExecDecodedWrmsr rc (%Rrc)\n", VBOXSTRICTRC_VAL(rcStrict)));
15768
15769	return rcStrict;
15770	}
15771
15772
15773	/**
15774	* VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
15775	*/
15776	HMVMX_EXIT_DECL hmR0VmxExitPause(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15777	{
15778	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15779
15780	/** @todo The guest has likely hit a contended spinlock. We might want to
15781	* poke a schedule different guest VCPU. */
15782	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15783	if (RT_SUCCESS(rc))
15784	return VINF_EM_RAW_INTERRUPT;
15785
15786	AssertMsgFailed(("hmR0VmxExitPause: Failed to increment RIP. rc=%Rrc\n", rc));
15787	return rc;
15788	}
15789
15790
15791	/**
15792	* VM-exit handler for when the TPR value is lowered below the specified
15793	* threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
15794	*/
15795	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThreshold(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15796	{
15797	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15798	Assert(pVmxTransient->pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
15799
15800	/*
15801	* The TPR shadow would've been synced with the APIC TPR in the post-run phase.
15802	* We'll re-evaluate pending interrupts and inject them before the next VM
15803	* entry so we can just continue execution here.
15804	*/
15805	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTprBelowThreshold);
15806	return VINF_SUCCESS;
15807	}
15808
15809
15810	/**
15811	* VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX). Conditional
15812	* VM-exit.
15813	*
15814	* @retval VINF_SUCCESS when guest execution can continue.
15815	* @retval VINF_PGM_SYNC_CR3 CR3 sync is required, back to ring-3.
15816	* @retval VERR_EM_RESCHEDULE_REM when we need to return to ring-3 due to
15817	* incompatible guest state for VMX execution (real-on-v86 case).
15818	*/
15819	HMVMX_EXIT_DECL hmR0VmxExitMovCRx(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15820	{
15821	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15822	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitMovCRx, y2);
15823
15824	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15825	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
15826	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15827	AssertRCReturn(rc, rc);
15828
15829	VBOXSTRICTRC rcStrict;
15830	PVM pVM = pVCpu->CTX_SUFF(pVM);
15831	uint64_t const uExitQual = pVmxTransient->uExitQual;
15832	uint32_t const uAccessType = VMX_EXIT_QUAL_CRX_ACCESS(uExitQual);
15833	switch (uAccessType)
15834	{
15835	/*
15836	* MOV to CRx.
15837	*/
15838	case VMX_EXIT_QUAL_CRX_ACCESS_WRITE:
15839	{
15840	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
15841	AssertRCReturn(rc, rc);
15842
15843	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
15844	uint32_t const uOldCr0 = pVCpu->cpum.GstCtx.cr0;
15845	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(uExitQual);
15846	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(uExitQual);
15847
15848	/*
15849	* MOV to CR3 only cause a VM-exit when one or more of the following are true:
15850	* - When nested paging isn't used.
15851	* - If the guest doesn't have paging enabled (intercept CR3 to update shadow page tables).
15852	* - We are executing in the VM debug loop.
15853	*/
15854	Assert( iCrReg != 3
15855	\|\| !pVM->hm.s.fNestedPaging
15856	\|\| !CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx)
15857	\|\| pVCpu->hm.s.fUsingDebugLoop);
15858
15859	/* MOV to CR8 writes only cause VM-exits when TPR shadow is not used. */
15860	Assert( iCrReg != 8
15861	\|\| !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW));
15862
15863	rcStrict = hmR0VmxExitMovToCrX(pVCpu, pVmcsInfo, pVmxTransient->cbInstr, iGReg, iCrReg);
15864	AssertMsg( rcStrict == VINF_SUCCESS
15865	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15866
15867	/*
15868	* This is a kludge for handling switches back to real mode when we try to use
15869	* V86 mode to run real mode code directly. Problem is that V86 mode cannot
15870	* deal with special selector values, so we have to return to ring-3 and run
15871	* there till the selector values are V86 mode compatible.
15872	*
15873	* Note! Using VINF_EM_RESCHEDULE_REM here rather than VINF_EM_RESCHEDULE since the
15874	* latter is an alias for VINF_IEM_RAISED_XCPT which is asserted at the end of
15875	* this function.
15876	*/
15877	if ( iCrReg == 0
15878	&& rcStrict == VINF_SUCCESS
15879	&& !pVM->hm.s.vmx.fUnrestrictedGuest
15880	&& CPUMIsGuestInRealModeEx(&pVCpu->cpum.GstCtx)
15881	&& (uOldCr0 & X86_CR0_PE)
15882	&& !(pVCpu->cpum.GstCtx.cr0 & X86_CR0_PE))
15883	{
15884	/** @todo Check selectors rather than returning all the time. */
15885	Assert(!pVmxTransient->fIsNestedGuest);
15886	Log4Func(("CR0 write, back to real mode -> VINF_EM_RESCHEDULE_REM\n"));
15887	rcStrict = VINF_EM_RESCHEDULE_REM;
15888	}
15889	break;
15890	}
15891
15892	/*
15893	* MOV from CRx.
15894	*/
15895	case VMX_EXIT_QUAL_CRX_ACCESS_READ:
15896	{
15897	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(uExitQual);
15898	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(uExitQual);
15899
15900	/*
15901	* MOV from CR3 only cause a VM-exit when one or more of the following are true:
15902	* - When nested paging isn't used.
15903	* - If the guest doesn't have paging enabled (pass guest's CR3 rather than our identity mapped CR3).
15904	* - We are executing in the VM debug loop.
15905	*/
15906	Assert( iCrReg != 3
15907	\|\| !pVM->hm.s.fNestedPaging
15908	\|\| !CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx)
15909	\|\| pVCpu->hm.s.fUsingDebugLoop);
15910
15911	/* MOV from CR8 reads only cause a VM-exit when the TPR shadow feature isn't enabled. */
15912	Assert( iCrReg != 8
15913	\|\| !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW));
15914
15915	rcStrict = hmR0VmxExitMovFromCrX(pVCpu, pVmcsInfo, pVmxTransient->cbInstr, iGReg, iCrReg);
15916	break;
15917	}
15918
15919	/*
15920	* CLTS (Clear Task-Switch Flag in CR0).
15921	*/
15922	case VMX_EXIT_QUAL_CRX_ACCESS_CLTS:
15923	{
15924	rcStrict = hmR0VmxExitClts(pVCpu, pVmcsInfo, pVmxTransient->cbInstr);
15925	break;
15926	}
15927
15928	/*
15929	* LMSW (Load Machine-Status Word into CR0).
15930	* LMSW cannot clear CR0.PE, so no fRealOnV86Active kludge needed here.
15931	*/
15932	case VMX_EXIT_QUAL_CRX_ACCESS_LMSW:
15933	{
15934	RTGCPTR GCPtrEffDst;
15935	uint8_t const cbInstr = pVmxTransient->cbInstr;
15936	uint16_t const uMsw = VMX_EXIT_QUAL_CRX_LMSW_DATA(uExitQual);
15937	bool const fMemOperand = VMX_EXIT_QUAL_CRX_LMSW_OP_MEM(uExitQual);
15938	if (fMemOperand)
15939	{
15940	rc = hmR0VmxReadGuestLinearAddrVmcs(pVCpu, pVmxTransient);
15941	AssertRCReturn(rc, rc);
15942	GCPtrEffDst = pVmxTransient->uGuestLinearAddr;
15943	}
15944	else
15945	GCPtrEffDst = NIL_RTGCPTR;
15946	rcStrict = hmR0VmxExitLmsw(pVCpu, pVmcsInfo, cbInstr, uMsw, GCPtrEffDst);
15947	break;
15948	}
15949
15950	default:
15951	{
15952	AssertMsgFailed(("Unrecognized Mov CRX access type %#x\n", uAccessType));
15953	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, uAccessType);
15954	}
15955	}
15956
15957	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS))
15958	== (HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS));
15959	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
15960
15961	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitMovCRx, y2);
15962	NOREF(pVM);
15963	return rcStrict;
15964	}
15965
15966
15967	/**
15968	* VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR). Conditional
15969	* VM-exit.
15970	*/
15971	HMVMX_EXIT_DECL hmR0VmxExitIoInstr(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
15972	{
15973	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15974	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitIO, y1);
15975
15976	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15977	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15978	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
15979	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15980	rc \|= hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_SREG_MASK
15981	\| CPUMCTX_EXTRN_EFER);
15982	/* EFER MSR also required for longmode checks in EMInterpretDisasCurrent(), but it's always up-to-date. */
15983	AssertRCReturn(rc, rc);
15984
15985	/* Refer Intel spec. 27-5. "Exit Qualifications for I/O Instructions" for the format. */
15986	uint32_t const uIOPort = VMX_EXIT_QUAL_IO_PORT(pVmxTransient->uExitQual);
15987	uint8_t const uIOSize = VMX_EXIT_QUAL_IO_SIZE(pVmxTransient->uExitQual);
15988	bool const fIOWrite = (VMX_EXIT_QUAL_IO_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_IO_DIRECTION_OUT);
15989	bool const fIOString = VMX_EXIT_QUAL_IO_IS_STRING(pVmxTransient->uExitQual);
15990	bool const fGstStepping = RT_BOOL(pCtx->eflags.Bits.u1TF);
15991	bool const fDbgStepping = pVCpu->hm.s.fSingleInstruction;
15992	AssertReturn(uIOSize <= 3 && uIOSize != 2, VERR_VMX_IPE_1);
15993
15994	/*
15995	* Update exit history to see if this exit can be optimized.
15996	*/
15997	VBOXSTRICTRC rcStrict;
15998	PCEMEXITREC pExitRec = NULL;
15999	if ( !fGstStepping
16000	&& !fDbgStepping)
16001	pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
16002	!fIOString
16003	? !fIOWrite
16004	? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_READ)
16005	: EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_WRITE)
16006	: !fIOWrite
16007	? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_READ)
16008	: EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_WRITE),
16009	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
16010	if (!pExitRec)
16011	{
16012	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses in bytes. */
16013	static uint32_t const s_aIOOpAnd[4] = { 0xff, 0xffff, 0, 0xffffffff }; /* AND masks for saving result in AL/AX/EAX. */
16014
16015	uint32_t const cbValue = s_aIOSizes[uIOSize];
16016	uint32_t const cbInstr = pVmxTransient->cbInstr;
16017	bool fUpdateRipAlready = false; /* ugly hack, should be temporary. */
16018	PVM pVM = pVCpu->CTX_SUFF(pVM);
16019	if (fIOString)
16020	{
16021	/*
16022	* INS/OUTS - I/O String instruction.
16023	*
16024	* Use instruction-information if available, otherwise fall back on
16025	* interpreting the instruction.
16026	*/
16027	Log4Func(("cs:rip=%#04x:%#RX64 %#06x/%u %c str\n", pCtx->cs.Sel, pCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
16028	AssertReturn(pCtx->dx == uIOPort, VERR_VMX_IPE_2);
16029	bool const fInsOutsInfo = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_INS_OUTS);
16030	if (fInsOutsInfo)
16031	{
16032	int rc2 = hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16033	AssertRCReturn(rc2, rc2);
16034	AssertReturn(pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize <= 2, VERR_VMX_IPE_3);
16035	AssertCompile(IEMMODE_16BIT == 0 && IEMMODE_32BIT == 1 && IEMMODE_64BIT == 2);
16036	IEMMODE const enmAddrMode = (IEMMODE)pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize;
16037	bool const fRep = VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual);
16038	if (fIOWrite)
16039	rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, fRep, cbInstr,
16040	pVmxTransient->ExitInstrInfo.StrIo.iSegReg, true /fIoChecked/);
16041	else
16042	{
16043	/*
16044	* The segment prefix for INS cannot be overridden and is always ES. We can safely assume X86_SREG_ES.
16045	* Hence "iSegReg" field is undefined in the instruction-information field in VT-x for INS.
16046	* See Intel Instruction spec. for "INS".
16047	* See Intel spec. Table 27-8 "Format of the VM-Exit Instruction-Information Field as Used for INS and OUTS".
16048	*/
16049	rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, fRep, cbInstr, true /fIoChecked/);
16050	}
16051	}
16052	else
16053	rcStrict = IEMExecOne(pVCpu);
16054
16055	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
16056	fUpdateRipAlready = true;
16057	}
16058	else
16059	{
16060	/*
16061	* IN/OUT - I/O instruction.
16062	*/
16063	Log4Func(("cs:rip=%04x:%08RX64 %#06x/%u %c\n", pCtx->cs.Sel, pCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
16064	uint32_t const uAndVal = s_aIOOpAnd[uIOSize];
16065	Assert(!VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual));
16066	if (fIOWrite)
16067	{
16068	rcStrict = IOMIOPortWrite(pVM, pVCpu, uIOPort, pCtx->eax & uAndVal, cbValue);
16069	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite);
16070	if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
16071	&& !pCtx->eflags.Bits.u1TF)
16072	rcStrict = EMRZSetPendingIoPortWrite(pVCpu, uIOPort, cbInstr, cbValue, pCtx->eax & uAndVal);
16073	}
16074	else
16075	{
16076	uint32_t u32Result = 0;
16077	rcStrict = IOMIOPortRead(pVM, pVCpu, uIOPort, &u32Result, cbValue);
16078	if (IOM_SUCCESS(rcStrict))
16079	{
16080	/* Save result of I/O IN instr. in AL/AX/EAX. */
16081	pCtx->eax = (pCtx->eax & ~uAndVal) \| (u32Result & uAndVal);
16082	}
16083	if ( rcStrict == VINF_IOM_R3_IOPORT_READ
16084	&& !pCtx->eflags.Bits.u1TF)
16085	rcStrict = EMRZSetPendingIoPortRead(pVCpu, uIOPort, cbInstr, cbValue);
16086	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead);
16087	}
16088	}
16089
16090	if (IOM_SUCCESS(rcStrict))
16091	{
16092	if (!fUpdateRipAlready)
16093	{
16094	hmR0VmxAdvanceGuestRipBy(pVCpu, cbInstr);
16095	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
16096	}
16097
16098	/*
16099	* INS/OUTS with REP prefix updates RFLAGS, can be observed with triple-fault guru
16100	* while booting Fedora 17 64-bit guest.
16101	*
16102	* See Intel Instruction reference for REP/REPE/REPZ/REPNE/REPNZ.
16103	*/
16104	if (fIOString)
16105	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RFLAGS);
16106
16107	/*
16108	* If any I/O breakpoints are armed, we need to check if one triggered
16109	* and take appropriate action.
16110	* Note that the I/O breakpoint type is undefined if CR4.DE is 0.
16111	*/
16112	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_DR7);
16113	AssertRCReturn(rc, rc);
16114
16115	/** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the
16116	* execution engines about whether hyper BPs and such are pending. */
16117	uint32_t const uDr7 = pCtx->dr[7];
16118	if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK)
16119	&& X86_DR7_ANY_RW_IO(uDr7)
16120	&& (pCtx->cr4 & X86_CR4_DE))
16121	\|\| DBGFBpIsHwIoArmed(pVM)))
16122	{
16123	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck);
16124
16125	/* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
16126	VMMRZCallRing3Disable(pVCpu);
16127	HM_DISABLE_PREEMPT(pVCpu);
16128
16129	bool fIsGuestDbgActive = CPUMR0DebugStateMaybeSaveGuest(pVCpu, true /* fDr6 */);
16130
16131	VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pCtx, uIOPort, cbValue);
16132	if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
16133	{
16134	/* Raise #DB. */
16135	if (fIsGuestDbgActive)
16136	ASMSetDR6(pCtx->dr[6]);
16137	if (pCtx->dr[7] != uDr7)
16138	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_GUEST_DR7;
16139
16140	hmR0VmxSetPendingXcptDB(pVCpu);
16141	}
16142	/* rcStrict is VINF_SUCCESS, VINF_IOM_R3_IOPORT_COMMIT_WRITE, or in [VINF_EM_FIRST..VINF_EM_LAST],
16143	however we can ditch VINF_IOM_R3_IOPORT_COMMIT_WRITE as it has VMCPU_FF_IOM as backup. */
16144	else if ( rcStrict2 != VINF_SUCCESS
16145	&& (rcStrict == VINF_SUCCESS \|\| rcStrict2 < rcStrict))
16146	rcStrict = rcStrict2;
16147	AssertCompile(VINF_EM_LAST < VINF_IOM_R3_IOPORT_COMMIT_WRITE);
16148
16149	HM_RESTORE_PREEMPT();
16150	VMMRZCallRing3Enable(pVCpu);
16151	}
16152	}
16153
16154	#ifdef VBOX_STRICT
16155	if ( rcStrict == VINF_IOM_R3_IOPORT_READ
16156	\|\| rcStrict == VINF_EM_PENDING_R3_IOPORT_READ)
16157	Assert(!fIOWrite);
16158	else if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
16159	\|\| rcStrict == VINF_IOM_R3_IOPORT_COMMIT_WRITE
16160	\|\| rcStrict == VINF_EM_PENDING_R3_IOPORT_WRITE)
16161	Assert(fIOWrite);
16162	else
16163	{
16164	# if 0 /** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST
16165	* statuses, that the VMM device and some others may return. See
16166	* IOM_SUCCESS() for guidance. */
16167	AssertMsg( RT_FAILURE(rcStrict)
16168	\|\| rcStrict == VINF_SUCCESS
16169	\|\| rcStrict == VINF_EM_RAW_EMULATE_INSTR
16170	\|\| rcStrict == VINF_EM_DBG_BREAKPOINT
16171	\|\| rcStrict == VINF_EM_RAW_GUEST_TRAP
16172	\|\| rcStrict == VINF_EM_RAW_TO_R3
16173	\|\| rcStrict == VINF_TRPM_XCPT_DISPATCHED, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
16174	# endif
16175	}
16176	#endif
16177	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitIO, y1);
16178	}
16179	else
16180	{
16181	/*
16182	* Frequent exit or something needing probing. Get state and call EMHistoryExec.
16183	*/
16184	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16185	AssertRCReturn(rc2, rc2);
16186	STAM_COUNTER_INC(!fIOString ? fIOWrite ? &pVCpu->hm.s.StatExitIOWrite : &pVCpu->hm.s.StatExitIORead
16187	: fIOWrite ? &pVCpu->hm.s.StatExitIOStringWrite : &pVCpu->hm.s.StatExitIOStringRead);
16188	Log4(("IOExit/%u: %04x:%08RX64: %s%s%s %#x LB %u -> EMHistoryExec\n",
16189	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
16190	VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual) ? "REP " : "",
16191	fIOWrite ? "OUT" : "IN", fIOString ? "S" : "", uIOPort, uIOSize));
16192
16193	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
16194	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16195
16196	Log4(("IOExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
16197	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
16198	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
16199	}
16200	return rcStrict;
16201	}
16202
16203
16204	/**
16205	* VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH). Unconditional
16206	* VM-exit.
16207	*/
16208	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitch(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16209	{
16210	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16211
16212	/* Check if this task-switch occurred while delivery an event through the guest IDT. */
16213	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
16214	AssertRCReturn(rc, rc);
16215	if (VMX_EXIT_QUAL_TASK_SWITCH_TYPE(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_TASK_SWITCH_TYPE_IDT)
16216	{
16217	rc = hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16218	AssertRCReturn(rc, rc);
16219	if (VMX_IDT_VECTORING_INFO_IS_VALID(pVmxTransient->uIdtVectoringInfo))
16220	{
16221	uint32_t uErrCode;
16222	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(pVmxTransient->uIdtVectoringInfo))
16223	{
16224	rc = hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16225	AssertRCReturn(rc, rc);
16226	uErrCode = pVmxTransient->uIdtVectoringErrorCode;
16227	}
16228	else
16229	uErrCode = 0;
16230
16231	RTGCUINTPTR GCPtrFaultAddress;
16232	if (VMX_IDT_VECTORING_INFO_IS_XCPT_PF(pVmxTransient->uIdtVectoringInfo))
16233	GCPtrFaultAddress = pVCpu->cpum.GstCtx.cr2;
16234	else
16235	GCPtrFaultAddress = 0;
16236
16237	rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16238	AssertRCReturn(rc, rc);
16239
16240	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(pVmxTransient->uIdtVectoringInfo),
16241	pVmxTransient->cbInstr, uErrCode, GCPtrFaultAddress);
16242
16243	Log4Func(("Pending event. uIntType=%#x uVector=%#x\n", VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo),
16244	VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo)));
16245	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
16246	return VINF_EM_RAW_INJECT_TRPM_EVENT;
16247	}
16248	}
16249
16250	/* Fall back to the interpreter to emulate the task-switch. */
16251	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
16252	return VERR_EM_INTERPRETER;
16253	}
16254
16255
16256	/**
16257	* VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional VM-exit.
16258	*/
16259	HMVMX_EXIT_DECL hmR0VmxExitMtf(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16260	{
16261	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16262
16263	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16264	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_MONITOR_TRAP_FLAG;
16265	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
16266	AssertRCReturn(rc, rc);
16267	return VINF_EM_DBG_STEPPED;
16268	}
16269
16270
16271	/**
16272	* VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional VM-exit.
16273	*/
16274	HMVMX_EXIT_DECL hmR0VmxExitApicAccess(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16275	{
16276	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16277	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitApicAccess);
16278
16279	int rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16280	rc \|= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16281	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16282	rc \|= hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16283	rc \|= hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16284	AssertRCReturn(rc, rc);
16285
16286	/*
16287	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16288	*/
16289	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16290	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16291	{
16292	/* For some crazy guest, if an event delivery causes an APIC-access VM-exit, go to instruction emulation. */
16293	if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
16294	{
16295	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
16296	return VINF_EM_RAW_INJECT_TRPM_EVENT;
16297	}
16298	}
16299	else
16300	{
16301	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16302	return rcStrict;
16303	}
16304
16305	/* IOMMIOPhysHandler() below may call into IEM, save the necessary state. */
16306	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16307	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16308	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
16309	AssertRCReturn(rc, rc);
16310
16311	/* See Intel spec. 27-6 "Exit Qualifications for APIC-access VM-exits from Linear Accesses & Guest-Phyiscal Addresses" */
16312	uint32_t const uAccessType = VMX_EXIT_QUAL_APIC_ACCESS_TYPE(pVmxTransient->uExitQual);
16313	switch (uAccessType)
16314	{
16315	case VMX_APIC_ACCESS_TYPE_LINEAR_WRITE:
16316	case VMX_APIC_ACCESS_TYPE_LINEAR_READ:
16317	{
16318	AssertMsg( !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
16319	\|\| VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual) != XAPIC_OFF_TPR,
16320	("hmR0VmxExitApicAccess: can't access TPR offset while using TPR shadowing.\n"));
16321
16322	RTGCPHYS GCPhys = pVCpu->hm.s.vmx.u64GstMsrApicBase; /* Always up-to-date, as it is not part of the VMCS. */
16323	GCPhys &= PAGE_BASE_GC_MASK;
16324	GCPhys += VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual);
16325	Log4Func(("Linear access uAccessType=%#x GCPhys=%#RGp Off=%#x\n", uAccessType, GCPhys,
16326	VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual)));
16327
16328	PVM pVM = pVCpu->CTX_SUFF(pVM);
16329	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16330	rcStrict = IOMMMIOPhysHandler(pVM, pVCpu,
16331	uAccessType == VMX_APIC_ACCESS_TYPE_LINEAR_READ ? 0 : X86_TRAP_PF_RW,
16332	CPUMCTX2CORE(pCtx), GCPhys);
16333	Log4Func(("IOMMMIOPhysHandler returned %Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
16334	if ( rcStrict == VINF_SUCCESS
16335	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16336	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16337	{
16338	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS
16339	\| HM_CHANGED_GUEST_APIC_TPR);
16340	rcStrict = VINF_SUCCESS;
16341	}
16342	break;
16343	}
16344
16345	default:
16346	{
16347	Log4Func(("uAccessType=%#x\n", uAccessType));
16348	rcStrict = VINF_EM_RAW_EMULATE_INSTR;
16349	break;
16350	}
16351	}
16352
16353	if (rcStrict != VINF_SUCCESS)
16354	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchApicAccessToR3);
16355	return rcStrict;
16356	}
16357
16358
16359	/**
16360	* VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX). Conditional
16361	* VM-exit.
16362	*/
16363	HMVMX_EXIT_DECL hmR0VmxExitMovDRx(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16364	{
16365	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16366	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16367
16368	/* We might get this VM-exit if the nested-guest is not intercepting MOV DRx accesses. */
16369	if (!pVmxTransient->fIsNestedGuest)
16370	{
16371	/* We should -not- get this VM-exit if the guest's debug registers were active. */
16372	if (pVmxTransient->fWasGuestDebugStateActive)
16373	{
16374	AssertMsgFailed(("Unexpected MOV DRx exit\n"));
16375	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
16376	}
16377
16378	if ( !pVCpu->hm.s.fSingleInstruction
16379	&& !pVmxTransient->fWasHyperDebugStateActive)
16380	{
16381	Assert(!DBGFIsStepping(pVCpu));
16382	Assert(pVmcsInfo->u32XcptBitmap & RT_BIT(X86_XCPT_DB));
16383
16384	/* Don't intercept MOV DRx any more. */
16385	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_MOV_DR_EXIT;
16386	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
16387	AssertRCReturn(rc, rc);
16388
16389	/* We're playing with the host CPU state here, make sure we can't preempt or longjmp. */
16390	VMMRZCallRing3Disable(pVCpu);
16391	HM_DISABLE_PREEMPT(pVCpu);
16392
16393	/* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */
16394	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
16395	Assert(CPUMIsGuestDebugStateActive(pVCpu) \|\| HC_ARCH_BITS == 32);
16396
16397	HM_RESTORE_PREEMPT();
16398	VMMRZCallRing3Enable(pVCpu);
16399
16400	#ifdef VBOX_WITH_STATISTICS
16401	rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
16402	AssertRCReturn(rc, rc);
16403	if (VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_DRX_DIRECTION_WRITE)
16404	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
16405	else
16406	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
16407	#endif
16408	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch);
16409	return VINF_SUCCESS;
16410	}
16411	}
16412
16413	/*
16414	* EMInterpretDRx[Write\|Read]() calls CPUMIsGuestIn64BitCode() which requires EFER MSR, CS.
16415	* The EFER MSR is always up-to-date.
16416	* Update the segment registers and DR7 from the CPU.
16417	*/
16418	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16419	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
16420	rc \|= hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_DR7);
16421	AssertRCReturn(rc, rc);
16422	Log4Func(("cs:rip=%#04x:%#RX64\n", pCtx->cs.Sel, pCtx->rip));
16423
16424	PVM pVM = pVCpu->CTX_SUFF(pVM);
16425	if (VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_DRX_DIRECTION_WRITE)
16426	{
16427	rc = EMInterpretDRxWrite(pVM, pVCpu, CPUMCTX2CORE(pCtx),
16428	VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual),
16429	VMX_EXIT_QUAL_DRX_GENREG(pVmxTransient->uExitQual));
16430	if (RT_SUCCESS(rc))
16431	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
16432	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
16433	}
16434	else
16435	{
16436	rc = EMInterpretDRxRead(pVM, pVCpu, CPUMCTX2CORE(pCtx),
16437	VMX_EXIT_QUAL_DRX_GENREG(pVmxTransient->uExitQual),
16438	VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual));
16439	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
16440	}
16441
16442	Assert(rc == VINF_SUCCESS \|\| rc == VERR_EM_INTERPRETER);
16443	if (RT_SUCCESS(rc))
16444	{
16445	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
16446	AssertRCReturn(rc2, rc2);
16447	return VINF_SUCCESS;
16448	}
16449	return rc;
16450	}
16451
16452
16453	/**
16454	* VM-exit handler for EPT misconfiguration (VMX_EXIT_EPT_MISCONFIG).
16455	* Conditional VM-exit.
16456	*/
16457	HMVMX_EXIT_DECL hmR0VmxExitEptMisconfig(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16458	{
16459	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16460	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
16461
16462	int rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16463	rc \|= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16464	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16465	rc \|= hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16466	rc \|= hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16467	AssertRCReturn(rc, rc);
16468
16469	/*
16470	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16471	*/
16472	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16473	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16474	{
16475	/*
16476	* In the unlikely case where delivering an event causes an EPT misconfig (MMIO), go back to
16477	* instruction emulation to inject the original event. Otherwise, injecting the original event
16478	* using hardware-assisted VMX would would trigger the same EPT misconfig VM-exit again.
16479	*/
16480	if (!pVCpu->hm.s.Event.fPending)
16481	{ /* likely */ }
16482	else
16483	{
16484	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
16485	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16486	/** @todo NSTVMX: Think about how this should be handled. */
16487	if (pVmxTransient->fIsNestedGuest)
16488	return VERR_VMX_IPE_3;
16489	#endif
16490	return VINF_EM_RAW_INJECT_TRPM_EVENT;
16491	}
16492	}
16493	else
16494	{
16495	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16496	return rcStrict;
16497	}
16498
16499	/*
16500	* Get sufficent state and update the exit history entry.
16501	*/
16502	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16503	rc = hmR0VmxReadGuestPhysicalAddrVmcs(pVCpu, pVmxTransient);
16504	rc \|= hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16505	AssertRCReturn(rc, rc);
16506
16507	RTGCPHYS const GCPhys = pVmxTransient->uGuestPhysicalAddr;
16508	PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
16509	EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_MMIO),
16510	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
16511	if (!pExitRec)
16512	{
16513	/*
16514	* If we succeed, resume guest execution.
16515	* If we fail in interpreting the instruction because we couldn't get the guest physical address
16516	* of the page containing the instruction via the guest's page tables (we would invalidate the guest page
16517	* in the host TLB), resume execution which would cause a guest page fault to let the guest handle this
16518	* weird case. See @bugref{6043}.
16519	*/
16520	PVM pVM = pVCpu->CTX_SUFF(pVM);
16521	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16522	rcStrict = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, PGMMODE_EPT, CPUMCTX2CORE(pCtx), GCPhys, UINT32_MAX);
16523	Log4Func(("At %#RGp RIP=%#RX64 rc=%Rrc\n", GCPhys, pCtx->rip, VBOXSTRICTRC_VAL(rcStrict)));
16524	if ( rcStrict == VINF_SUCCESS
16525	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16526	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16527	{
16528	/* Successfully handled MMIO operation. */
16529	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS
16530	\| HM_CHANGED_GUEST_APIC_TPR);
16531	rcStrict = VINF_SUCCESS;
16532	}
16533	}
16534	else
16535	{
16536	/*
16537	* Frequent exit or something needing probing. Call EMHistoryExec.
16538	*/
16539	Log4(("EptMisscfgExit/%u: %04x:%08RX64: %RGp -> EMHistoryExec\n",
16540	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, GCPhys));
16541
16542	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
16543	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16544
16545	Log4(("EptMisscfgExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
16546	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
16547	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
16548	}
16549	return rcStrict;
16550	}
16551
16552
16553	/**
16554	* VM-exit handler for EPT violation (VMX_EXIT_EPT_VIOLATION). Conditional
16555	* VM-exit.
16556	*/
16557	HMVMX_EXIT_DECL hmR0VmxExitEptViolation(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16558	{
16559	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16560	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
16561
16562	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
16563	rc \|= hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16564	rc \|= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16565	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16566	rc \|= hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16567	rc \|= hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16568	AssertRCReturn(rc, rc);
16569
16570	/*
16571	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16572	*/
16573	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16574	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16575	{
16576	/*
16577	* If delivery of an event causes an EPT violation (true nested #PF and not MMIO),
16578	* we shall resolve the nested #PF and re-inject the original event.
16579	*/
16580	if (pVCpu->hm.s.Event.fPending)
16581	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflectNPF);
16582	}
16583	else
16584	{
16585	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16586	return rcStrict;
16587	}
16588
16589	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16590	rc = hmR0VmxReadGuestPhysicalAddrVmcs(pVCpu, pVmxTransient);
16591	rc \|= hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16592	AssertRCReturn(rc, rc);
16593
16594	RTGCPHYS const GCPhys = pVmxTransient->uGuestPhysicalAddr;
16595	uint64_t const uExitQual = pVmxTransient->uExitQual;
16596	AssertMsg(((pVmxTransient->uExitQual >> 7) & 3) != 2, ("%#RX64", uExitQual));
16597
16598	RTGCUINT uErrorCode = 0;
16599	if (uExitQual & VMX_EXIT_QUAL_EPT_INSTR_FETCH)
16600	uErrorCode \|= X86_TRAP_PF_ID;
16601	if (uExitQual & VMX_EXIT_QUAL_EPT_DATA_WRITE)
16602	uErrorCode \|= X86_TRAP_PF_RW;
16603	if (uExitQual & VMX_EXIT_QUAL_EPT_ENTRY_PRESENT)
16604	uErrorCode \|= X86_TRAP_PF_P;
16605
16606	PVM pVM = pVCpu->CTX_SUFF(pVM);
16607	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16608	Log4Func(("at %#RX64 (%#RX64 errcode=%#x) cs:rip=%#04x:%#RX64\n", GCPhys, uExitQual, uErrorCode, pCtx->cs.Sel, pCtx->rip));
16609
16610	/*
16611	* Handle the pagefault trap for the nested shadow table.
16612	*/
16613	TRPMAssertXcptPF(pVCpu, GCPhys, uErrorCode);
16614	rcStrict = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, PGMMODE_EPT, uErrorCode, CPUMCTX2CORE(pCtx), GCPhys);
16615	TRPMResetTrap(pVCpu);
16616
16617	/* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}. */
16618	if ( rcStrict == VINF_SUCCESS
16619	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16620	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16621	{
16622	/* Successfully synced our nested page tables. */
16623	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf);
16624	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS);
16625	return VINF_SUCCESS;
16626	}
16627
16628	Log4Func(("EPT return to ring-3 rcStrict2=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
16629	return rcStrict;
16630	}
16631
16632
16633	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16634	/**
16635	* VM-exit handler for VMCLEAR (VMX_EXIT_VMCLEAR). Unconditional VM-exit.
16636	*/
16637	HMVMX_EXIT_DECL hmR0VmxExitVmclear(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16638	{
16639	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16640
16641	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16642	rc \|= hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16643	\| CPUMCTX_EXTRN_HWVIRT
16644	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16645	rc \|= hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16646	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
16647	AssertRCReturn(rc, rc);
16648
16649	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16650
16651	VMXVEXITINFO ExitInfo;
16652	RT_ZERO(ExitInfo);
16653	ExitInfo.uReason = pVmxTransient->uExitReason;
16654	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16655	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16656	ExitInfo.cbInstr = pVmxTransient->cbInstr;
16657	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16658
16659	VBOXSTRICTRC rcStrict = IEMExecDecodedVmclear(pVCpu, &ExitInfo);
16660	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16661	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16662	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16663	{
16664	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16665	rcStrict = VINF_SUCCESS;
16666	}
16667	return rcStrict;
16668	}
16669
16670
16671	/**
16672	* VM-exit handler for VMLAUNCH (VMX_EXIT_VMLAUNCH). Unconditional VM-exit.
16673	*/
16674	HMVMX_EXIT_DECL hmR0VmxExitVmlaunch(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16675	{
16676	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16677
16678	/* Import the entire VMCS state for now as we would be switching VMCS on successful VMLAUNCH,
16679	otherwise we could import just IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK. */
16680	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16681	rc \|= hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16682	AssertRCReturn(rc, rc);
16683
16684	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16685
16686	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
16687	VBOXSTRICTRC rcStrict = IEMExecDecodedVmlaunchVmresume(pVCpu, pVmxTransient->cbInstr, VMXINSTRID_VMLAUNCH);
16688	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
16689	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16690	{
16691	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16692	if (CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
16693	rcStrict = VINF_VMX_VMLAUNCH_VMRESUME;
16694	}
16695	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
16696	return rcStrict;
16697	}
16698
16699
16700	/**
16701	* VM-exit handler for VMPTRLD (VMX_EXIT_VMPTRLD). Unconditional VM-exit.
16702	*/
16703	HMVMX_EXIT_DECL hmR0VmxExitVmptrld(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16704	{
16705	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16706
16707	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16708	rc \|= hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16709	\| CPUMCTX_EXTRN_HWVIRT
16710	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16711	rc \|= hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16712	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
16713	AssertRCReturn(rc, rc);
16714
16715	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16716
16717	VMXVEXITINFO ExitInfo;
16718	RT_ZERO(ExitInfo);
16719	ExitInfo.uReason = pVmxTransient->uExitReason;
16720	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16721	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16722	ExitInfo.cbInstr = pVmxTransient->cbInstr;
16723	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16724
16725	VBOXSTRICTRC rcStrict = IEMExecDecodedVmptrld(pVCpu, &ExitInfo);
16726	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16727	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16728	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16729	{
16730	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16731	rcStrict = VINF_SUCCESS;
16732	}
16733	return rcStrict;
16734	}
16735
16736
16737	/**
16738	* VM-exit handler for VMPTRST (VMX_EXIT_VMPTRST). Unconditional VM-exit.
16739	*/
16740	HMVMX_EXIT_DECL hmR0VmxExitVmptrst(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16741	{
16742	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16743
16744	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16745	rc \|= hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16746	\| CPUMCTX_EXTRN_HWVIRT
16747	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16748	rc \|= hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16749	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
16750	AssertRCReturn(rc, rc);
16751
16752	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16753
16754	VMXVEXITINFO ExitInfo;
16755	RT_ZERO(ExitInfo);
16756	ExitInfo.uReason = pVmxTransient->uExitReason;
16757	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16758	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16759	ExitInfo.cbInstr = pVmxTransient->cbInstr;
16760	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_WRITE, &ExitInfo.GCPtrEffAddr);
16761
16762	VBOXSTRICTRC rcStrict = IEMExecDecodedVmptrst(pVCpu, &ExitInfo);
16763	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16764	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16765	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16766	{
16767	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16768	rcStrict = VINF_SUCCESS;
16769	}
16770	return rcStrict;
16771	}
16772
16773
16774	/**
16775	* VM-exit handler for VMREAD (VMX_EXIT_VMREAD). Conditional VM-exit.
16776	*/
16777	HMVMX_EXIT_DECL hmR0VmxExitVmread(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16778	{
16779	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16780
16781	/*
16782	* Strictly speaking we should not get VMREAD VM-exits for shadow VMCS fields and
16783	* thus might not need to import the shadow VMCS state, it's safer just in case
16784	* code elsewhere dares look at unsynced VMCS fields.
16785	*/
16786	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16787	rc \|= hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16788	\| CPUMCTX_EXTRN_HWVIRT
16789	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16790	rc \|= hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16791	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
16792	AssertRCReturn(rc, rc);
16793
16794	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16795
16796	VMXVEXITINFO ExitInfo;
16797	RT_ZERO(ExitInfo);
16798	ExitInfo.uReason = pVmxTransient->uExitReason;
16799	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16800	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16801	ExitInfo.cbInstr = pVmxTransient->cbInstr;
16802	if (!ExitInfo.InstrInfo.VmreadVmwrite.fIsRegOperand)
16803	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_WRITE, &ExitInfo.GCPtrEffAddr);
16804
16805	VBOXSTRICTRC rcStrict = IEMExecDecodedVmread(pVCpu, &ExitInfo);
16806	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16807	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16808	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16809	{
16810	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16811	rcStrict = VINF_SUCCESS;
16812	}
16813	return rcStrict;
16814	}
16815
16816
16817	/**
16818	* VM-exit handler for VMRESUME (VMX_EXIT_VMRESUME). Unconditional VM-exit.
16819	*/
16820	HMVMX_EXIT_DECL hmR0VmxExitVmresume(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16821	{
16822	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16823
16824	/* Import the entire VMCS state for now as we would be switching VMCS on successful VMRESUME,
16825	otherwise we could import just IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK. */
16826	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16827	rc \|= hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16828	AssertRCReturn(rc, rc);
16829
16830	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16831
16832	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
16833	VBOXSTRICTRC rcStrict = IEMExecDecodedVmlaunchVmresume(pVCpu, pVmxTransient->cbInstr, VMXINSTRID_VMRESUME);
16834	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
16835	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16836	{
16837	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16838	if (CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
16839	rcStrict = VINF_VMX_VMLAUNCH_VMRESUME;
16840	}
16841	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
16842	return rcStrict;
16843	}
16844
16845
16846	/**
16847	* VM-exit handler for VMWRITE (VMX_EXIT_VMWRITE). Conditional VM-exit.
16848	*/
16849	HMVMX_EXIT_DECL hmR0VmxExitVmwrite(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16850	{
16851	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16852
16853	/*
16854	* Although we should not get VMWRITE VM-exits for shadow VMCS fields, since our HM hook
16855	* gets invoked when IEM's VMWRITE instruction emulation modifies the current VMCS and it
16856	* flags re-loading the entire shadow VMCS, we should save the entire shadow VMCS here.
16857	*/
16858	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16859	rc \|= hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16860	\| CPUMCTX_EXTRN_HWVIRT
16861	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16862	rc \|= hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16863	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
16864	AssertRCReturn(rc, rc);
16865
16866	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16867
16868	VMXVEXITINFO ExitInfo;
16869	RT_ZERO(ExitInfo);
16870	ExitInfo.uReason = pVmxTransient->uExitReason;
16871	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16872	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16873	ExitInfo.cbInstr = pVmxTransient->cbInstr;
16874	if (!ExitInfo.InstrInfo.VmreadVmwrite.fIsRegOperand)
16875	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16876
16877	VBOXSTRICTRC rcStrict = IEMExecDecodedVmwrite(pVCpu, &ExitInfo);
16878	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16879	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16880	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16881	{
16882	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16883	rcStrict = VINF_SUCCESS;
16884	}
16885	return rcStrict;
16886	}
16887
16888
16889	/**
16890	* VM-exit handler for VMXOFF (VMX_EXIT_VMXOFF). Unconditional VM-exit.
16891	*/
16892	HMVMX_EXIT_DECL hmR0VmxExitVmxoff(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16893	{
16894	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16895
16896	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16897	rc \|= hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR4
16898	\| CPUMCTX_EXTRN_HWVIRT
16899	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
16900	AssertRCReturn(rc, rc);
16901
16902	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16903
16904	VBOXSTRICTRC rcStrict = IEMExecDecodedVmxoff(pVCpu, pVmxTransient->cbInstr);
16905	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16906	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_HWVIRT);
16907	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16908	{
16909	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16910	rcStrict = VINF_SUCCESS;
16911	}
16912	return rcStrict;
16913	}
16914
16915
16916	/**
16917	* VM-exit handler for VMXON (VMX_EXIT_VMXON). Unconditional VM-exit.
16918	*/
16919	HMVMX_EXIT_DECL hmR0VmxExitVmxon(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16920	{
16921	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16922
16923	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16924	rc \|= hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16925	\| CPUMCTX_EXTRN_HWVIRT
16926	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16927	rc \|= hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16928	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
16929	AssertRCReturn(rc, rc);
16930
16931	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16932
16933	VMXVEXITINFO ExitInfo;
16934	RT_ZERO(ExitInfo);
16935	ExitInfo.uReason = pVmxTransient->uExitReason;
16936	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16937	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16938	ExitInfo.cbInstr = pVmxTransient->cbInstr;
16939	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16940
16941	VBOXSTRICTRC rcStrict = IEMExecDecodedVmxon(pVCpu, &ExitInfo);
16942	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16943	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16944	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16945	{
16946	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16947	rcStrict = VINF_SUCCESS;
16948	}
16949	return rcStrict;
16950	}
16951
16952
16953	/**
16954	* VM-exit handler for INVVPID (VMX_EXIT_INVVPID). Unconditional VM-exit.
16955	*/
16956	HMVMX_EXIT_DECL hmR0VmxExitInvvpid(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
16957	{
16958	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16959
16960	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16961	rc \|= hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16962	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16963	rc \|= hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16964	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
16965	AssertRCReturn(rc, rc);
16966
16967	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16968
16969	VMXVEXITINFO ExitInfo;
16970	RT_ZERO(ExitInfo);
16971	ExitInfo.uReason = pVmxTransient->uExitReason;
16972	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16973	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16974	ExitInfo.cbInstr = pVmxTransient->cbInstr;
16975	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16976
16977	VBOXSTRICTRC rcStrict = IEMExecDecodedInvvpid(pVCpu, &ExitInfo);
16978	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16979	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16980	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16981	{
16982	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16983	rcStrict = VINF_SUCCESS;
16984	}
16985	return rcStrict;
16986	}
16987	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
16988	/** @} */
16989
16990
16991	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16992	/** @name Nested-guest VM-exit handlers.
16993	* @{
16994	*/
16995	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16996	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Nested-guest VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16997	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16998
16999	/**
17000	* Nested-guest VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI).
17001	* Conditional VM-exit.
17002	*/
17003	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmiNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17004	{
17005	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17006
17007	int rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
17008	AssertRCReturn(rc, rc);
17009
17010	uint64_t const uExitIntInfo = pVmxTransient->uExitIntInfo;
17011	uint32_t const uExitIntType = VMX_EXIT_INT_INFO_TYPE(uExitIntInfo);
17012	Assert(VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo));
17013
17014	switch (uExitIntType)
17015	{
17016	/*
17017	* Physical NMIs:
17018	* We shouldn't direct host physical NMIs to the nested-guest. Dispatch it to the host.
17019	*/
17020	case VMX_EXIT_INT_INFO_TYPE_NMI:
17021	return hmR0VmxExitHostNmi(pVCpu, pVmxTransient->pVmcsInfo);
17022
17023	/*
17024	* Hardware exceptions,
17025	* Software exceptions,
17026	* Privileged software exceptions:
17027	* Figure out if the exception must be delivered to the guest or the nested-guest.
17028	*/
17029	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
17030	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
17031	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
17032	{
17033	rc \|= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
17034	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17035	rc \|= hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
17036	rc \|= hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
17037	AssertRCReturn(rc, rc);
17038
17039	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
17040	bool const fIntercept = CPUMIsGuestVmxXcptInterceptSet(pVCpu, pCtx, VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo),
17041	pVmxTransient->uExitIntErrorCode);
17042	if (fIntercept)
17043	{
17044	/* Exit qualification is required for debug and page-fault exceptions. */
17045	rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
17046	AssertRCReturn(rc, rc);
17047
17048	/*
17049	* For VM-exits due to software exceptions (those generated by INT3 or INTO) and privileged
17050	* software exceptions (those generated by INT1/ICEBP) we need to supply the VM-exit instruction
17051	* length. However, if delivery of a software interrupt, software exception or privileged
17052	* software exception causes a VM-exit, that too provides the VM-exit instruction length.
17053	*/
17054	VMXVEXITINFO ExitInfo;
17055	RT_ZERO(ExitInfo);
17056	ExitInfo.uReason = pVmxTransient->uExitReason;
17057	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17058	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17059
17060	VMXVEXITEVENTINFO ExitEventInfo;
17061	RT_ZERO(ExitEventInfo);
17062	ExitEventInfo.uExitIntInfo = pVmxTransient->uExitIntInfo;
17063	ExitEventInfo.uExitIntErrCode = pVmxTransient->uExitIntErrorCode;
17064	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
17065	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
17066
17067	#ifdef DEBUG_ramshankar
17068	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
17069	Log4Func(("cs:rip=%#04x:%#RX64 %s err_code=%#x exit_qual=%#RX64\n", pCtx->cs.Sel, pCtx->rip,
17070	VMX_EXIT_INT_INFO_IS_XCPT_PF(pVmxTransient->uExitIntInfo) ? "#PF" : "Unk",
17071	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual));
17072	Log4Func(("idt_info=%#RX64 (%s) idt_errcode=%#RX32\n", pVmxTransient->uIdtVectoringInfo,
17073	VMX_IDT_VECTORING_INFO_IS_VALID(pVmxTransient->uIdtVectoringInfo) ? "Valid" : "Invalid",
17074	pVmxTransient->uIdtVectoringErrorCode));
17075	#endif
17076	return IEMExecVmxVmexitXcpt(pVCpu, &ExitInfo, &ExitEventInfo);
17077	}
17078
17079	/* Nested paging is currently a requirement, otherwise we would need to handle shadow #PFs in hmR0VmxExitXcptPF. */
17080	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
17081	return hmR0VmxExitXcpt(pVCpu, pVmxTransient);
17082	}
17083
17084	/*
17085	* Software interrupts:
17086	* VM-exits cannot be caused by software interrupts.
17087	*
17088	* External interrupts:
17089	* This should only happen when "acknowledge external interrupts on VM-exit"
17090	* control is set. However, we never set this when executing a guest or
17091	* nested-guest. For nested-guests it is emulated while injecting interrupts into
17092	* the guest.
17093	*/
17094	case VMX_EXIT_INT_INFO_TYPE_SW_INT:
17095	case VMX_EXIT_INT_INFO_TYPE_EXT_INT:
17096	default:
17097	{
17098	pVCpu->hm.s.u32HMError = pVmxTransient->uExitIntInfo;
17099	return VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
17100	}
17101	}
17102	}
17103
17104
17105	/**
17106	* Nested-guest VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT).
17107	* Unconditional VM-exit.
17108	*/
17109	HMVMX_EXIT_DECL hmR0VmxExitTripleFaultNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17110	{
17111	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17112	return IEMExecVmxVmexitTripleFault(pVCpu);
17113	}
17114
17115
17116	/**
17117	* Nested-guest VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
17118	*/
17119	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindowNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17120	{
17121	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17122
17123	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INT_WINDOW_EXIT))
17124	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
17125	return hmR0VmxExitIntWindow(pVCpu, pVmxTransient);
17126	}
17127
17128
17129	/**
17130	* Nested-guest VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
17131	*/
17132	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindowNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17133	{
17134	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17135
17136	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_NMI_WINDOW_EXIT))
17137	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
17138	return hmR0VmxExitIntWindow(pVCpu, pVmxTransient);
17139	}
17140
17141
17142	/**
17143	* Nested-guest VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH).
17144	* Unconditional VM-exit.
17145	*/
17146	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitchNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17147	{
17148	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17149
17150	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
17151	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17152	rc \|= hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
17153	rc \|= hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
17154	AssertRCReturn(rc, rc);
17155
17156	VMXVEXITINFO ExitInfo;
17157	RT_ZERO(ExitInfo);
17158	ExitInfo.uReason = pVmxTransient->uExitReason;
17159	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17160	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17161
17162	VMXVEXITEVENTINFO ExitEventInfo;
17163	RT_ZERO(ExitEventInfo);
17164	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
17165	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
17166	return IEMExecVmxVmexitTaskSwitch(pVCpu, &ExitInfo, &ExitEventInfo);
17167	}
17168
17169
17170	/**
17171	* Nested-guest VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
17172	*/
17173	HMVMX_EXIT_DECL hmR0VmxExitHltNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17174	{
17175	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17176
17177	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_HLT_EXIT))
17178	{
17179	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17180	AssertRCReturn(rc, rc);
17181	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbInstr);
17182	}
17183	return hmR0VmxExitHlt(pVCpu, pVmxTransient);
17184	}
17185
17186
17187	/**
17188	* Nested-guest VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
17189	*/
17190	HMVMX_EXIT_DECL hmR0VmxExitInvlpgNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17191	{
17192	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17193
17194	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INVLPG_EXIT))
17195	{
17196	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17197	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
17198	AssertRCReturn(rc, rc);
17199
17200	VMXVEXITINFO ExitInfo;
17201	RT_ZERO(ExitInfo);
17202	ExitInfo.uReason = pVmxTransient->uExitReason;
17203	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17204	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17205	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17206	}
17207	return hmR0VmxExitInvlpg(pVCpu, pVmxTransient);
17208	}
17209
17210
17211	/**
17212	* Nested-guest VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
17213	*/
17214	HMVMX_EXIT_DECL hmR0VmxExitRdpmcNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17215	{
17216	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17217
17218	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDPMC_EXIT))
17219	{
17220	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17221	AssertRCReturn(rc, rc);
17222	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbInstr);
17223	}
17224	return hmR0VmxExitRdpmc(pVCpu, pVmxTransient);
17225	}
17226
17227
17228	/**
17229	* Nested-guest VM-exit handler for VMREAD (VMX_EXIT_VMREAD) and VMWRITE
17230	* (VMX_EXIT_VMWRITE). Conditional VM-exit.
17231	*/
17232	HMVMX_EXIT_DECL hmR0VmxExitVmreadVmwriteNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17233	{
17234	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17235
17236	Assert( pVmxTransient->uExitReason == VMX_EXIT_VMREAD
17237	\|\| pVmxTransient->uExitReason == VMX_EXIT_VMWRITE);
17238
17239	int rc = hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17240	AssertRCReturn(rc, rc);
17241
17242	uint8_t const iGReg = pVmxTransient->ExitInstrInfo.VmreadVmwrite.iReg2;
17243	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
17244	uint64_t u64VmcsField = pVCpu->cpum.GstCtx.aGRegs[iGReg].u64;
17245
17246	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
17247	if (!CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
17248	u64VmcsField &= UINT64_C(0xffffffff);
17249
17250	if (CPUMIsGuestVmxVmreadVmwriteInterceptSet(pVCpu, pVmxTransient->uExitReason, u64VmcsField))
17251	{
17252	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17253	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
17254	AssertRCReturn(rc, rc);
17255
17256	VMXVEXITINFO ExitInfo;
17257	RT_ZERO(ExitInfo);
17258	ExitInfo.uReason = pVmxTransient->uExitReason;
17259	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17260	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17261	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17262	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17263	}
17264
17265	if (pVmxTransient->uExitReason == VMX_EXIT_VMREAD)
17266	return hmR0VmxExitVmread(pVCpu, pVmxTransient);
17267	return hmR0VmxExitVmwrite(pVCpu, pVmxTransient);
17268	}
17269
17270
17271	/**
17272	* Nested-guest VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
17273	*/
17274	HMVMX_EXIT_DECL hmR0VmxExitRdtscNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17275	{
17276	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17277
17278	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDTSC_EXIT))
17279	{
17280	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17281	AssertRCReturn(rc, rc);
17282	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbInstr);
17283	}
17284
17285	return hmR0VmxExitRdtsc(pVCpu, pVmxTransient);
17286	}
17287
17288
17289	/**
17290	* Nested-guest VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX).
17291	* Conditional VM-exit.
17292	*/
17293	HMVMX_EXIT_DECL hmR0VmxExitMovCRxNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17294	{
17295	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17296
17297	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
17298	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17299	AssertRCReturn(rc, rc);
17300
17301	VBOXSTRICTRC rcStrict;
17302	uint32_t const uAccessType = VMX_EXIT_QUAL_CRX_ACCESS(pVmxTransient->uExitQual);
17303	switch (uAccessType)
17304	{
17305	case VMX_EXIT_QUAL_CRX_ACCESS_WRITE:
17306	{
17307	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
17308	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(pVmxTransient->uExitQual);
17309	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
17310	uint64_t const uNewCrX = pVCpu->cpum.GstCtx.aGRegs[iGReg].u64;
17311
17312	bool fIntercept;
17313	switch (iCrReg)
17314	{
17315	case 0:
17316	case 4:
17317	fIntercept = CPUMIsGuestVmxMovToCr0Cr4InterceptSet(pVCpu, &pVCpu->cpum.GstCtx, iCrReg, uNewCrX);
17318	break;
17319
17320	case 3:
17321	fIntercept = CPUMIsGuestVmxMovToCr3InterceptSet(pVCpu, uNewCrX);
17322	break;
17323
17324	case 8:
17325	fIntercept = CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_CR8_LOAD_EXIT);
17326	break;
17327
17328	default:
17329	fIntercept = false;
17330	break;
17331	}
17332	if (fIntercept)
17333	{
17334	VMXVEXITINFO ExitInfo;
17335	RT_ZERO(ExitInfo);
17336	ExitInfo.uReason = pVmxTransient->uExitReason;
17337	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17338	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17339	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17340	}
17341	else
17342	rcStrict = hmR0VmxExitMovToCrX(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbInstr, iGReg, iCrReg);
17343	break;
17344	}
17345
17346	case VMX_EXIT_QUAL_CRX_ACCESS_READ:
17347	{
17348	/*
17349	* CR0/CR4 reads do not cause VM-exits, the read-shadow is used (subject to masking).
17350	* CR2 reads do not cause a VM-exit.
17351	* CR3 reads cause a VM-exit depending on the "CR3 store exiting" control.
17352	* CR8 reads cause a VM-exit depending on the "CR8 store exiting" control.
17353	*/
17354	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
17355	if ( iCrReg == 3
17356	\|\| iCrReg == 8)
17357	{
17358	static const uint32_t s_auCrXReadIntercepts[] = { 0, 0, 0, VMX_PROC_CTLS_CR3_STORE_EXIT, 0,
17359	0, 0, 0, VMX_PROC_CTLS_CR8_STORE_EXIT };
17360	uint32_t const uIntercept = s_auCrXReadIntercepts[iCrReg];
17361	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, uIntercept))
17362	{
17363	VMXVEXITINFO ExitInfo;
17364	RT_ZERO(ExitInfo);
17365	ExitInfo.uReason = pVmxTransient->uExitReason;
17366	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17367	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17368	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17369	}
17370	else
17371	{
17372	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(pVmxTransient->uExitQual);
17373	rcStrict = hmR0VmxExitMovFromCrX(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbInstr, iGReg, iCrReg);
17374	}
17375	}
17376	else
17377	{
17378	AssertMsgFailed(("MOV from CR%d VM-exit must not happen\n", iCrReg));
17379	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, iCrReg);
17380	}
17381	break;
17382	}
17383
17384	case VMX_EXIT_QUAL_CRX_ACCESS_CLTS:
17385	{
17386	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
17387	Assert(pVmcsNstGst);
17388	uint64_t const uGstHostMask = pVmcsNstGst->u64Cr0Mask.u;
17389	uint64_t const uReadShadow = pVmcsNstGst->u64Cr0ReadShadow.u;
17390	if ( (uGstHostMask & X86_CR0_TS)
17391	&& (uReadShadow & X86_CR0_TS))
17392	{
17393	VMXVEXITINFO ExitInfo;
17394	RT_ZERO(ExitInfo);
17395	ExitInfo.uReason = pVmxTransient->uExitReason;
17396	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17397	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17398	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17399	}
17400	else
17401	rcStrict = hmR0VmxExitClts(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbInstr);
17402	break;
17403	}
17404
17405	case VMX_EXIT_QUAL_CRX_ACCESS_LMSW: /* LMSW (Load Machine-Status Word into CR0) */
17406	{
17407	RTGCPTR GCPtrEffDst;
17408	uint16_t const uNewMsw = VMX_EXIT_QUAL_CRX_LMSW_DATA(pVmxTransient->uExitQual);
17409	bool const fMemOperand = VMX_EXIT_QUAL_CRX_LMSW_OP_MEM(pVmxTransient->uExitQual);
17410	if (fMemOperand)
17411	{
17412	rc = hmR0VmxReadGuestLinearAddrVmcs(pVCpu, pVmxTransient);
17413	AssertRCReturn(rc, rc);
17414	GCPtrEffDst = pVmxTransient->uGuestLinearAddr;
17415	}
17416	else
17417	GCPtrEffDst = NIL_RTGCPTR;
17418
17419	if (CPUMIsGuestVmxLmswInterceptSet(pVCpu, &pVCpu->cpum.GstCtx, uNewMsw))
17420	{
17421	VMXVEXITINFO ExitInfo;
17422	RT_ZERO(ExitInfo);
17423	ExitInfo.uReason = pVmxTransient->uExitReason;
17424	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17425	ExitInfo.u64GuestLinearAddr = GCPtrEffDst;
17426	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17427	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17428	}
17429	else
17430	rcStrict = hmR0VmxExitLmsw(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbInstr, uNewMsw, GCPtrEffDst);
17431	break;
17432	}
17433
17434	default:
17435	{
17436	AssertMsgFailed(("Unrecognized Mov CRX access type %#x\n", uAccessType));
17437	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, uAccessType);
17438	}
17439	}
17440
17441	if (rcStrict == VINF_IEM_RAISED_XCPT)
17442	{
17443	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
17444	rcStrict = VINF_SUCCESS;
17445	}
17446	return rcStrict;
17447	}
17448
17449
17450	/**
17451	* Nested-guest VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX).
17452	* Conditional VM-exit.
17453	*/
17454	HMVMX_EXIT_DECL hmR0VmxExitMovDRxNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17455	{
17456	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17457
17458	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MOV_DR_EXIT))
17459	{
17460	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
17461	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17462	AssertRCReturn(rc, rc);
17463
17464	VMXVEXITINFO ExitInfo;
17465	RT_ZERO(ExitInfo);
17466	ExitInfo.uReason = pVmxTransient->uExitReason;
17467	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17468	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17469	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17470	}
17471	return hmR0VmxExitMovDRx(pVCpu, pVmxTransient);
17472	}
17473
17474
17475	/**
17476	* Nested-guest VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR).
17477	* Conditional VM-exit.
17478	*/
17479	HMVMX_EXIT_DECL hmR0VmxExitIoInstrNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17480	{
17481	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17482
17483	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
17484	AssertRCReturn(rc, rc);
17485
17486	uint32_t const uIOPort = VMX_EXIT_QUAL_IO_PORT(pVmxTransient->uExitQual);
17487	uint8_t const uIOSize = VMX_EXIT_QUAL_IO_SIZE(pVmxTransient->uExitQual);
17488	AssertReturn(uIOSize <= 3 && uIOSize != 2, VERR_VMX_IPE_1);
17489
17490	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses in bytes. */
17491	uint8_t const cbAccess = s_aIOSizes[uIOSize];
17492	if (CPUMIsGuestVmxIoInterceptSet(pVCpu, uIOPort, cbAccess))
17493	{
17494	/*
17495	* IN/OUT instruction:
17496	* - Provides VM-exit instruction length.
17497	*
17498	* INS/OUTS instruction:
17499	* - Provides VM-exit instruction length.
17500	* - Provides Guest-linear address.
17501	* - Optionally provides VM-exit instruction info (depends on CPU feature).
17502	*/
17503	PVM pVM = pVCpu->CTX_SUFF(pVM);
17504	rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17505	AssertRCReturn(rc, rc);
17506
17507	/* Make sure we don't use stale/uninitialized VMX-transient info. below. */
17508	pVmxTransient->ExitInstrInfo.u = 0;
17509	pVmxTransient->uGuestLinearAddr = 0;
17510
17511	bool const fVmxInsOutsInfo = pVM->cpum.ro.GuestFeatures.fVmxInsOutInfo;
17512	bool const fIOString = VMX_EXIT_QUAL_IO_IS_STRING(pVmxTransient->uExitQual);
17513	if (fIOString)
17514	{
17515	rc \|= hmR0VmxReadGuestLinearAddrVmcs(pVCpu, pVmxTransient);
17516	if (fVmxInsOutsInfo)
17517	{
17518	Assert(RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_INS_OUTS)); /* Paranoia. */
17519	rc \|= hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17520	}
17521	}
17522	AssertRCReturn(rc, rc);
17523
17524	VMXVEXITINFO ExitInfo;
17525	RT_ZERO(ExitInfo);
17526	ExitInfo.uReason = pVmxTransient->uExitReason;
17527	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17528	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17529	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17530	ExitInfo.u64GuestLinearAddr = pVmxTransient->uGuestLinearAddr;
17531	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17532	}
17533	return hmR0VmxExitIoInstr(pVCpu, pVmxTransient);
17534	}
17535
17536
17537	/**
17538	* Nested-guest VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
17539	*/
17540	HMVMX_EXIT_DECL hmR0VmxExitRdmsrNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17541	{
17542	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17543
17544	uint32_t fMsrpm;
17545	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_MSR_BITMAPS))
17546	fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), pVCpu->cpum.GstCtx.ecx);
17547	else
17548	fMsrpm = VMXMSRPM_EXIT_RD;
17549
17550	if (fMsrpm & VMXMSRPM_EXIT_RD)
17551	{
17552	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17553	AssertRCReturn(rc, rc);
17554	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbInstr);
17555	}
17556	return hmR0VmxExitRdmsr(pVCpu, pVmxTransient);
17557	}
17558
17559
17560	/**
17561	* Nested-guest VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
17562	*/
17563	HMVMX_EXIT_DECL hmR0VmxExitWrmsrNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17564	{
17565	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17566
17567	uint32_t fMsrpm;
17568	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_MSR_BITMAPS))
17569	fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), pVCpu->cpum.GstCtx.ecx);
17570	else
17571	fMsrpm = VMXMSRPM_EXIT_WR;
17572
17573	if (fMsrpm & VMXMSRPM_EXIT_WR)
17574	{
17575	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17576	AssertRCReturn(rc, rc);
17577	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbInstr);
17578	}
17579	return hmR0VmxExitWrmsr(pVCpu, pVmxTransient);
17580	}
17581
17582
17583	/**
17584	* Nested-guest VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
17585	*/
17586	HMVMX_EXIT_DECL hmR0VmxExitMwaitNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17587	{
17588	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17589
17590	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MWAIT_EXIT))
17591	{
17592	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17593	AssertRCReturn(rc, rc);
17594	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbInstr);
17595	}
17596	return hmR0VmxExitMwait(pVCpu, pVmxTransient);
17597	}
17598
17599
17600	/**
17601	* Nested-guest VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional
17602	* VM-exit.
17603	*/
17604	HMVMX_EXIT_DECL hmR0VmxExitMtfNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17605	{
17606	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17607
17608	/** @todo NSTVMX: Should consider debugging nested-guests using VM debugger. */
17609	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
17610	}
17611
17612
17613	/**
17614	* Nested-guest VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
17615	*/
17616	HMVMX_EXIT_DECL hmR0VmxExitMonitorNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17617	{
17618	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17619
17620	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MONITOR_EXIT))
17621	{
17622	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17623	AssertRCReturn(rc, rc);
17624	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbInstr);
17625	}
17626	return hmR0VmxExitMonitor(pVCpu, pVmxTransient);
17627	}
17628
17629
17630	/**
17631	* Nested-guest VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
17632	*/
17633	HMVMX_EXIT_DECL hmR0VmxExitPauseNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17634	{
17635	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17636
17637	/** @todo NSTVMX: Think about this more. Does the outer guest need to intercept
17638	* PAUSE when executing a nested-guest? If it does not, we would not need
17639	* to check for the intercepts here. Just call VM-exit... */
17640
17641	/* The CPU would have already performed the necessary CPL checks for PAUSE-loop exiting. */
17642	if ( CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_PAUSE_EXIT)
17643	\|\| CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_PAUSE_LOOP_EXIT))
17644	{
17645	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17646	AssertRCReturn(rc, rc);
17647	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbInstr);
17648	}
17649	return hmR0VmxExitPause(pVCpu, pVmxTransient);
17650	}
17651
17652
17653	/**
17654	* Nested-guest VM-exit handler for when the TPR value is lowered below the
17655	* specified threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
17656	*/
17657	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThresholdNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17658	{
17659	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17660
17661	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_TPR_SHADOW))
17662	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
17663	return hmR0VmxExitTprBelowThreshold(pVCpu, pVmxTransient);
17664	}
17665
17666
17667	/**
17668	* Nested-guest VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional
17669	* VM-exit.
17670	*/
17671	HMVMX_EXIT_DECL hmR0VmxExitApicAccessNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17672	{
17673	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17674
17675	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17676	rc \|= hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
17677	rc \|= hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
17678	AssertRCReturn(rc, rc);
17679
17680	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_VIRT_APIC_ACCESS));
17681	rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
17682	AssertRCReturn(rc, rc);
17683
17684	VMXVEXITINFO ExitInfo;
17685	RT_ZERO(ExitInfo);
17686	ExitInfo.uReason = pVmxTransient->uExitReason;
17687	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17688	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17689
17690	VMXVEXITEVENTINFO ExitEventInfo;
17691	RT_ZERO(ExitEventInfo);
17692	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
17693	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
17694	return IEMExecVmxVmexitApicAccess(pVCpu, &ExitInfo, &ExitEventInfo);
17695	}
17696
17697
17698	/**
17699	* Nested-guest VM-exit handler for APIC write emulation (VMX_EXIT_APIC_WRITE).
17700	* Conditional VM-exit.
17701	*/
17702	HMVMX_EXIT_DECL hmR0VmxExitApicWriteNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17703	{
17704	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17705
17706	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_APIC_REG_VIRT));
17707	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
17708	AssertRCReturn(rc, rc);
17709
17710	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
17711	}
17712
17713
17714	/**
17715	* Nested-guest VM-exit handler for virtualized EOI (VMX_EXIT_VIRTUALIZED_EOI).
17716	* Conditional VM-exit.
17717	*/
17718	HMVMX_EXIT_DECL hmR0VmxExitVirtEoiNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17719	{
17720	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17721
17722	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_VIRT_INT_DELIVERY));
17723	int rc = hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
17724	AssertRCReturn(rc, rc);
17725
17726	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
17727	}
17728
17729
17730	/**
17731	* Nested-guest VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
17732	*/
17733	HMVMX_EXIT_DECL hmR0VmxExitRdtscpNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17734	{
17735	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17736
17737	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDTSC_EXIT))
17738	{
17739	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_RDTSCP));
17740	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17741	AssertRCReturn(rc, rc);
17742	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbInstr);
17743	}
17744	return hmR0VmxExitRdtscp(pVCpu, pVmxTransient);
17745	}
17746
17747
17748	/**
17749	* Nested-guest VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
17750	*/
17751	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvdNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17752	{
17753	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17754
17755	if (CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_WBINVD_EXIT))
17756	{
17757	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17758	AssertRCReturn(rc, rc);
17759	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbInstr);
17760	}
17761	return hmR0VmxExitWbinvd(pVCpu, pVmxTransient);
17762	}
17763
17764
17765	/**
17766	* Nested-guest VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
17767	*/
17768	HMVMX_EXIT_DECL hmR0VmxExitInvpcidNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17769	{
17770	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17771
17772	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INVLPG_EXIT))
17773	{
17774	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_INVPCID));
17775	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17776	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
17777	rc \|= hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17778	AssertRCReturn(rc, rc);
17779
17780	VMXVEXITINFO ExitInfo;
17781	RT_ZERO(ExitInfo);
17782	ExitInfo.uReason = pVmxTransient->uExitReason;
17783	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17784	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17785	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17786	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17787	}
17788	return hmR0VmxExitInvpcid(pVCpu, pVmxTransient);
17789	}
17790
17791
17792	/**
17793	* Nested-guest VM-exit handler for invalid-guest state
17794	* (VMX_EXIT_ERR_INVALID_GUEST_STATE). Error VM-exit.
17795	*/
17796	HMVMX_EXIT_DECL hmR0VmxExitErrInvalidGuestStateNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17797	{
17798	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17799
17800	/*
17801	* Currently this should never happen because we fully emulate VMLAUNCH/VMRESUME in IEM.
17802	* So if it does happen, it indicates a bug possibly in the hardware-assisted VMX code.
17803	* Handle it like it's in an invalid guest state of the outer guest.
17804	*
17805	* When the fast path is implemented, this should be changed to cause the corresponding
17806	* nested-guest VM-exit.
17807	*/
17808	return hmR0VmxExitErrInvalidGuestState(pVCpu, pVmxTransient);
17809	}
17810
17811
17812	/**
17813	* Nested-guest VM-exit handler for instructions that cause VM-exits uncondtionally
17814	* and only provide the instruction length.
17815	*
17816	* Unconditional VM-exit.
17817	*/
17818	HMVMX_EXIT_DECL hmR0VmxExitInstrNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17819	{
17820	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17821
17822	#ifdef VBOX_STRICT
17823	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
17824	switch (pVmxTransient->uExitReason)
17825	{
17826	case VMX_EXIT_ENCLS:
17827	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_ENCLS_EXIT));
17828	break;
17829
17830	case VMX_EXIT_VMFUNC:
17831	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_VMFUNC));
17832	break;
17833	}
17834	#endif
17835
17836	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17837	AssertRCReturn(rc, rc);
17838	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbInstr);
17839	}
17840
17841
17842	/**
17843	* Nested-guest VM-exit handler for instructions that provide instruction length as
17844	* well as more information.
17845	*
17846	* Unconditional VM-exit.
17847	*/
17848	HMVMX_EXIT_DECL hmR0VmxExitInstrWithInfoNested(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
17849	{
17850	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17851
17852	#ifdef VBOX_STRICT
17853	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
17854	switch (pVmxTransient->uExitReason)
17855	{
17856	case VMX_EXIT_GDTR_IDTR_ACCESS:
17857	case VMX_EXIT_LDTR_TR_ACCESS:
17858	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_DESC_TABLE_EXIT));
17859	break;
17860
17861	case VMX_EXIT_RDRAND:
17862	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_RDRAND_EXIT));
17863	break;
17864
17865	case VMX_EXIT_RDSEED:
17866	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_RDSEED_EXIT));
17867	break;
17868
17869	case VMX_EXIT_XSAVES:
17870	case VMX_EXIT_XRSTORS:
17871	/** @todo NSTVMX: Verify XSS-bitmap. */
17872	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_XSAVES_XRSTORS));
17873	break;
17874
17875	case VMX_EXIT_UMWAIT:
17876	case VMX_EXIT_TPAUSE:
17877	Assert(CPUMIsGuestVmxProcCtlsSet(pVCpu, pCtx, VMX_PROC_CTLS_RDTSC_EXIT));
17878	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_USER_WAIT_PAUSE));
17879	break;
17880	}
17881	#endif
17882
17883	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17884	rc \|= hmR0VmxReadExitQualVmcs(pVCpu, pVmxTransient);
17885	rc \|= hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17886	AssertRCReturn(rc, rc);
17887
17888	VMXVEXITINFO ExitInfo;
17889	RT_ZERO(ExitInfo);
17890	ExitInfo.uReason = pVmxTransient->uExitReason;
17891	ExitInfo.cbInstr = pVmxTransient->cbInstr;
17892	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17893	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17894	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17895	}
17896
17897	/** @} */
17898	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
17899

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/VMM/VMMR0/HMVMXR0.cpp@ 80021

Download in other formats: