HMVMXR0.cpp@ 80580

Last change on this file since 80580 was 80580, checked in by vboxsync, 5 years ago
VMM/HMVMXR0: Nested VMX: bugref:9180 Build fix.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 702.7 KB

Line
1	/* $Id: HMVMXR0.cpp 80580 2019-09-04 08:49:59Z 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/tm.h>
35	#include <VBox/vmm/em.h>
36	#include <VBox/vmm/gim.h>
37	#include <VBox/vmm/apic.h>
38	#ifdef VBOX_WITH_REM
39	# include <VBox/vmm/rem.h>
40	#endif
41	#include "HMInternal.h"
42	#include <VBox/vmm/vmcc.h>
43	#include <VBox/vmm/hmvmxinline.h>
44	#include "HMVMXR0.h"
45	#include "dtrace/VBoxVMM.h"
46
47	#ifdef DEBUG_ramshankar
48	# define HMVMX_ALWAYS_SAVE_GUEST_RFLAGS
49	# define HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
50	# define HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE
51	# define HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
52	# define HMVMX_ALWAYS_CLEAN_TRANSIENT
53	# define HMVMX_ALWAYS_CHECK_GUEST_STATE
54	# define HMVMX_ALWAYS_TRAP_ALL_XCPTS
55	# define HMVMX_ALWAYS_TRAP_PF
56	# define HMVMX_ALWAYS_FLUSH_TLB
57	# define HMVMX_ALWAYS_SWAP_EFER
58	#endif
59
60
61	/*********************************************************************************************************************************
62	* Defined Constants And Macros *
63	*********************************************************************************************************************************/
64	/** Use the function table. */
65	#define HMVMX_USE_FUNCTION_TABLE
66
67	/** Determine which tagged-TLB flush handler to use. */
68	#define HMVMX_FLUSH_TAGGED_TLB_EPT_VPID 0
69	#define HMVMX_FLUSH_TAGGED_TLB_EPT 1
70	#define HMVMX_FLUSH_TAGGED_TLB_VPID 2
71	#define HMVMX_FLUSH_TAGGED_TLB_NONE 3
72
73	/**
74	* Flags to skip redundant reads of some common VMCS fields that are not part of
75	* the guest-CPU or VCPU state but are needed while handling VM-exits.
76	*/
77	#define HMVMX_READ_IDT_VECTORING_INFO RT_BIT_32(0)
78	#define HMVMX_READ_IDT_VECTORING_ERROR_CODE RT_BIT_32(1)
79	#define HMVMX_READ_EXIT_QUALIFICATION RT_BIT_32(2)
80	#define HMVMX_READ_EXIT_INSTR_LEN RT_BIT_32(3)
81	#define HMVMX_READ_EXIT_INTERRUPTION_INFO RT_BIT_32(4)
82	#define HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE RT_BIT_32(5)
83	#define HMVMX_READ_EXIT_INSTR_INFO RT_BIT_32(6)
84	#define HMVMX_READ_GUEST_LINEAR_ADDR RT_BIT_32(7)
85	#define HMVMX_READ_GUEST_PHYSICAL_ADDR RT_BIT_32(8)
86	#define HMVMX_READ_GUEST_PENDING_DBG_XCPTS RT_BIT_32(9)
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
201	/** The guest's TPR value used for TPR shadowing. */
202	uint8_t u8GuestTpr;
203	/** Alignment. */
204	uint8_t abAlignment0[7];
205
206	/** The basic VM-exit reason. */
207	uint16_t uExitReason;
208	/** Alignment. */
209	uint16_t u16Alignment0;
210	/** The VM-exit interruption error code. */
211	uint32_t uExitIntErrorCode;
212	/** The VM-exit exit code qualification. */
213	uint64_t uExitQual;
214	/** The Guest-linear address. */
215	uint64_t uGuestLinearAddr;
216	/** The Guest-physical address. */
217	uint64_t uGuestPhysicalAddr;
218	/** The Guest pending-debug exceptions. */
219	uint64_t uGuestPendingDbgXcpts;
220
221	/** The VM-exit interruption-information field. */
222	uint32_t uExitIntInfo;
223	/** The VM-exit instruction-length field. */
224	uint32_t cbExitInstr;
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(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
298	#else
299	typedef DECLCALLBACK(VBOXSTRICTRC) FNVMXEXITHANDLER(PVMCPUCC 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(PVMCPUCC 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(PVMCPUCC 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(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
337	#endif
338
339	static int hmR0VmxImportGuestState(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint64_t fWhat);
340
341	/** @name VM-exit handler prototypes.
342	* @{
343	*/
344	static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmi;
345	static FNVMXEXITHANDLER hmR0VmxExitExtInt;
346	static FNVMXEXITHANDLER hmR0VmxExitTripleFault;
347	static FNVMXEXITHANDLERNSRC hmR0VmxExitIntWindow;
348	static FNVMXEXITHANDLERNSRC hmR0VmxExitNmiWindow;
349	static FNVMXEXITHANDLER hmR0VmxExitTaskSwitch;
350	static FNVMXEXITHANDLER hmR0VmxExitCpuid;
351	static FNVMXEXITHANDLER hmR0VmxExitGetsec;
352	static FNVMXEXITHANDLER hmR0VmxExitHlt;
353	static FNVMXEXITHANDLERNSRC hmR0VmxExitInvd;
354	static FNVMXEXITHANDLER hmR0VmxExitInvlpg;
355	static FNVMXEXITHANDLER hmR0VmxExitRdpmc;
356	static FNVMXEXITHANDLER hmR0VmxExitVmcall;
357	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
358	static FNVMXEXITHANDLER hmR0VmxExitVmclear;
359	static FNVMXEXITHANDLER hmR0VmxExitVmlaunch;
360	static FNVMXEXITHANDLER hmR0VmxExitVmptrld;
361	static FNVMXEXITHANDLER hmR0VmxExitVmptrst;
362	static FNVMXEXITHANDLER hmR0VmxExitVmread;
363	static FNVMXEXITHANDLER hmR0VmxExitVmresume;
364	static FNVMXEXITHANDLER hmR0VmxExitVmwrite;
365	static FNVMXEXITHANDLER hmR0VmxExitVmxoff;
366	static FNVMXEXITHANDLER hmR0VmxExitVmxon;
367	static FNVMXEXITHANDLER hmR0VmxExitInvvpid;
368	#endif
369	static FNVMXEXITHANDLER hmR0VmxExitRdtsc;
370	static FNVMXEXITHANDLER hmR0VmxExitMovCRx;
371	static FNVMXEXITHANDLER hmR0VmxExitMovDRx;
372	static FNVMXEXITHANDLER hmR0VmxExitIoInstr;
373	static FNVMXEXITHANDLER hmR0VmxExitRdmsr;
374	static FNVMXEXITHANDLER hmR0VmxExitWrmsr;
375	static FNVMXEXITHANDLER hmR0VmxExitMwait;
376	static FNVMXEXITHANDLER hmR0VmxExitMtf;
377	static FNVMXEXITHANDLER hmR0VmxExitMonitor;
378	static FNVMXEXITHANDLER hmR0VmxExitPause;
379	static FNVMXEXITHANDLERNSRC hmR0VmxExitTprBelowThreshold;
380	static FNVMXEXITHANDLER hmR0VmxExitApicAccess;
381	static FNVMXEXITHANDLER hmR0VmxExitEptViolation;
382	static FNVMXEXITHANDLER hmR0VmxExitEptMisconfig;
383	static FNVMXEXITHANDLER hmR0VmxExitRdtscp;
384	static FNVMXEXITHANDLER hmR0VmxExitPreemptTimer;
385	static FNVMXEXITHANDLERNSRC hmR0VmxExitWbinvd;
386	static FNVMXEXITHANDLER hmR0VmxExitXsetbv;
387	static FNVMXEXITHANDLER hmR0VmxExitInvpcid;
388	static FNVMXEXITHANDLERNSRC hmR0VmxExitSetPendingXcptUD;
389	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrInvalidGuestState;
390	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrUnexpected;
391	/** @} */
392
393	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
394	/** @name Nested-guest VM-exit handler prototypes.
395	* @{
396	*/
397	static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmiNested;
398	static FNVMXEXITHANDLER hmR0VmxExitTripleFaultNested;
399	static FNVMXEXITHANDLERNSRC hmR0VmxExitIntWindowNested;
400	static FNVMXEXITHANDLERNSRC hmR0VmxExitNmiWindowNested;
401	static FNVMXEXITHANDLER hmR0VmxExitTaskSwitchNested;
402	static FNVMXEXITHANDLER hmR0VmxExitHltNested;
403	static FNVMXEXITHANDLER hmR0VmxExitInvlpgNested;
404	static FNVMXEXITHANDLER hmR0VmxExitRdpmcNested;
405	static FNVMXEXITHANDLER hmR0VmxExitVmreadVmwriteNested;
406	static FNVMXEXITHANDLER hmR0VmxExitRdtscNested;
407	static FNVMXEXITHANDLER hmR0VmxExitMovCRxNested;
408	static FNVMXEXITHANDLER hmR0VmxExitMovDRxNested;
409	static FNVMXEXITHANDLER hmR0VmxExitIoInstrNested;
410	static FNVMXEXITHANDLER hmR0VmxExitRdmsrNested;
411	static FNVMXEXITHANDLER hmR0VmxExitWrmsrNested;
412	static FNVMXEXITHANDLER hmR0VmxExitMwaitNested;
413	static FNVMXEXITHANDLER hmR0VmxExitMtfNested;
414	static FNVMXEXITHANDLER hmR0VmxExitMonitorNested;
415	static FNVMXEXITHANDLER hmR0VmxExitPauseNested;
416	static FNVMXEXITHANDLERNSRC hmR0VmxExitTprBelowThresholdNested;
417	static FNVMXEXITHANDLER hmR0VmxExitApicAccessNested;
418	static FNVMXEXITHANDLER hmR0VmxExitApicWriteNested;
419	static FNVMXEXITHANDLER hmR0VmxExitVirtEoiNested;
420	static FNVMXEXITHANDLER hmR0VmxExitRdtscpNested;
421	static FNVMXEXITHANDLERNSRC hmR0VmxExitWbinvdNested;
422	static FNVMXEXITHANDLER hmR0VmxExitInvpcidNested;
423	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrInvalidGuestStateNested;
424	static FNVMXEXITHANDLER hmR0VmxExitInstrNested;
425	static FNVMXEXITHANDLER hmR0VmxExitInstrWithInfoNested;
426	/** @} */
427	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
428
429
430	/*********************************************************************************************************************************
431	* Global Variables *
432	*********************************************************************************************************************************/
433	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
434	/**
435	* Array of all VMCS fields.
436	* Any fields added to the VT-x spec. should be added here.
437	*
438	* Currently only used to derive shadow VMCS fields for hardware-assisted execution
439	* of nested-guests.
440	*/
441	static const uint32_t g_aVmcsFields[] =
442	{
443	/* 16-bit control fields. */
444	VMX_VMCS16_VPID,
445	VMX_VMCS16_POSTED_INT_NOTIFY_VECTOR,
446	VMX_VMCS16_EPTP_INDEX,
447
448	/* 16-bit guest-state fields. */
449	VMX_VMCS16_GUEST_ES_SEL,
450	VMX_VMCS16_GUEST_CS_SEL,
451	VMX_VMCS16_GUEST_SS_SEL,
452	VMX_VMCS16_GUEST_DS_SEL,
453	VMX_VMCS16_GUEST_FS_SEL,
454	VMX_VMCS16_GUEST_GS_SEL,
455	VMX_VMCS16_GUEST_LDTR_SEL,
456	VMX_VMCS16_GUEST_TR_SEL,
457	VMX_VMCS16_GUEST_INTR_STATUS,
458	VMX_VMCS16_GUEST_PML_INDEX,
459
460	/* 16-bits host-state fields. */
461	VMX_VMCS16_HOST_ES_SEL,
462	VMX_VMCS16_HOST_CS_SEL,
463	VMX_VMCS16_HOST_SS_SEL,
464	VMX_VMCS16_HOST_DS_SEL,
465	VMX_VMCS16_HOST_FS_SEL,
466	VMX_VMCS16_HOST_GS_SEL,
467	VMX_VMCS16_HOST_TR_SEL,
468
469	/* 64-bit control fields. */
470	VMX_VMCS64_CTRL_IO_BITMAP_A_FULL,
471	VMX_VMCS64_CTRL_IO_BITMAP_A_HIGH,
472	VMX_VMCS64_CTRL_IO_BITMAP_B_FULL,
473	VMX_VMCS64_CTRL_IO_BITMAP_B_HIGH,
474	VMX_VMCS64_CTRL_MSR_BITMAP_FULL,
475	VMX_VMCS64_CTRL_MSR_BITMAP_HIGH,
476	VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL,
477	VMX_VMCS64_CTRL_EXIT_MSR_STORE_HIGH,
478	VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL,
479	VMX_VMCS64_CTRL_EXIT_MSR_LOAD_HIGH,
480	VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL,
481	VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_HIGH,
482	VMX_VMCS64_CTRL_EXEC_VMCS_PTR_FULL,
483	VMX_VMCS64_CTRL_EXEC_VMCS_PTR_HIGH,
484	VMX_VMCS64_CTRL_EXEC_PML_ADDR_FULL,
485	VMX_VMCS64_CTRL_EXEC_PML_ADDR_HIGH,
486	VMX_VMCS64_CTRL_TSC_OFFSET_FULL,
487	VMX_VMCS64_CTRL_TSC_OFFSET_HIGH,
488	VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL,
489	VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_HIGH,
490	VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL,
491	VMX_VMCS64_CTRL_APIC_ACCESSADDR_HIGH,
492	VMX_VMCS64_CTRL_POSTED_INTR_DESC_FULL,
493	VMX_VMCS64_CTRL_POSTED_INTR_DESC_HIGH,
494	VMX_VMCS64_CTRL_VMFUNC_CTRLS_FULL,
495	VMX_VMCS64_CTRL_VMFUNC_CTRLS_HIGH,
496	VMX_VMCS64_CTRL_EPTP_FULL,
497	VMX_VMCS64_CTRL_EPTP_HIGH,
498	VMX_VMCS64_CTRL_EOI_BITMAP_0_FULL,
499	VMX_VMCS64_CTRL_EOI_BITMAP_0_HIGH,
500	VMX_VMCS64_CTRL_EOI_BITMAP_1_FULL,
501	VMX_VMCS64_CTRL_EOI_BITMAP_1_HIGH,
502	VMX_VMCS64_CTRL_EOI_BITMAP_2_FULL,
503	VMX_VMCS64_CTRL_EOI_BITMAP_2_HIGH,
504	VMX_VMCS64_CTRL_EOI_BITMAP_3_FULL,
505	VMX_VMCS64_CTRL_EOI_BITMAP_3_HIGH,
506	VMX_VMCS64_CTRL_EPTP_LIST_FULL,
507	VMX_VMCS64_CTRL_EPTP_LIST_HIGH,
508	VMX_VMCS64_CTRL_VMREAD_BITMAP_FULL,
509	VMX_VMCS64_CTRL_VMREAD_BITMAP_HIGH,
510	VMX_VMCS64_CTRL_VMWRITE_BITMAP_FULL,
511	VMX_VMCS64_CTRL_VMWRITE_BITMAP_HIGH,
512	VMX_VMCS64_CTRL_VIRTXCPT_INFO_ADDR_FULL,
513	VMX_VMCS64_CTRL_VIRTXCPT_INFO_ADDR_HIGH,
514	VMX_VMCS64_CTRL_XSS_EXITING_BITMAP_FULL,
515	VMX_VMCS64_CTRL_XSS_EXITING_BITMAP_HIGH,
516	VMX_VMCS64_CTRL_ENCLS_EXITING_BITMAP_FULL,
517	VMX_VMCS64_CTRL_ENCLS_EXITING_BITMAP_HIGH,
518	VMX_VMCS64_CTRL_TSC_MULTIPLIER_FULL,
519	VMX_VMCS64_CTRL_TSC_MULTIPLIER_HIGH,
520
521	/* 64-bit read-only data fields. */
522	VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL,
523	VMX_VMCS64_RO_GUEST_PHYS_ADDR_HIGH,
524
525	/* 64-bit guest-state fields. */
526	VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL,
527	VMX_VMCS64_GUEST_VMCS_LINK_PTR_HIGH,
528	VMX_VMCS64_GUEST_DEBUGCTL_FULL,
529	VMX_VMCS64_GUEST_DEBUGCTL_HIGH,
530	VMX_VMCS64_GUEST_PAT_FULL,
531	VMX_VMCS64_GUEST_PAT_HIGH,
532	VMX_VMCS64_GUEST_EFER_FULL,
533	VMX_VMCS64_GUEST_EFER_HIGH,
534	VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL,
535	VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_HIGH,
536	VMX_VMCS64_GUEST_PDPTE0_FULL,
537	VMX_VMCS64_GUEST_PDPTE0_HIGH,
538	VMX_VMCS64_GUEST_PDPTE1_FULL,
539	VMX_VMCS64_GUEST_PDPTE1_HIGH,
540	VMX_VMCS64_GUEST_PDPTE2_FULL,
541	VMX_VMCS64_GUEST_PDPTE2_HIGH,
542	VMX_VMCS64_GUEST_PDPTE3_FULL,
543	VMX_VMCS64_GUEST_PDPTE3_HIGH,
544	VMX_VMCS64_GUEST_BNDCFGS_FULL,
545	VMX_VMCS64_GUEST_BNDCFGS_HIGH,
546
547	/* 64-bit host-state fields. */
548	VMX_VMCS64_HOST_PAT_FULL,
549	VMX_VMCS64_HOST_PAT_HIGH,
550	VMX_VMCS64_HOST_EFER_FULL,
551	VMX_VMCS64_HOST_EFER_HIGH,
552	VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_FULL,
553	VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_HIGH,
554
555	/* 32-bit control fields. */
556	VMX_VMCS32_CTRL_PIN_EXEC,
557	VMX_VMCS32_CTRL_PROC_EXEC,
558	VMX_VMCS32_CTRL_EXCEPTION_BITMAP,
559	VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK,
560	VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH,
561	VMX_VMCS32_CTRL_CR3_TARGET_COUNT,
562	VMX_VMCS32_CTRL_EXIT,
563	VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT,
564	VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT,
565	VMX_VMCS32_CTRL_ENTRY,
566	VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT,
567	VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO,
568	VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE,
569	VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH,
570	VMX_VMCS32_CTRL_TPR_THRESHOLD,
571	VMX_VMCS32_CTRL_PROC_EXEC2,
572	VMX_VMCS32_CTRL_PLE_GAP,
573	VMX_VMCS32_CTRL_PLE_WINDOW,
574
575	/* 32-bits read-only fields. */
576	VMX_VMCS32_RO_VM_INSTR_ERROR,
577	VMX_VMCS32_RO_EXIT_REASON,
578	VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO,
579	VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE,
580	VMX_VMCS32_RO_IDT_VECTORING_INFO,
581	VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE,
582	VMX_VMCS32_RO_EXIT_INSTR_LENGTH,
583	VMX_VMCS32_RO_EXIT_INSTR_INFO,
584
585	/* 32-bit guest-state fields. */
586	VMX_VMCS32_GUEST_ES_LIMIT,
587	VMX_VMCS32_GUEST_CS_LIMIT,
588	VMX_VMCS32_GUEST_SS_LIMIT,
589	VMX_VMCS32_GUEST_DS_LIMIT,
590	VMX_VMCS32_GUEST_FS_LIMIT,
591	VMX_VMCS32_GUEST_GS_LIMIT,
592	VMX_VMCS32_GUEST_LDTR_LIMIT,
593	VMX_VMCS32_GUEST_TR_LIMIT,
594	VMX_VMCS32_GUEST_GDTR_LIMIT,
595	VMX_VMCS32_GUEST_IDTR_LIMIT,
596	VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS,
597	VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS,
598	VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS,
599	VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS,
600	VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS,
601	VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS,
602	VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS,
603	VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS,
604	VMX_VMCS32_GUEST_INT_STATE,
605	VMX_VMCS32_GUEST_ACTIVITY_STATE,
606	VMX_VMCS32_GUEST_SMBASE,
607	VMX_VMCS32_GUEST_SYSENTER_CS,
608	VMX_VMCS32_PREEMPT_TIMER_VALUE,
609
610	/* 32-bit host-state fields. */
611	VMX_VMCS32_HOST_SYSENTER_CS,
612
613	/* Natural-width control fields. */
614	VMX_VMCS_CTRL_CR0_MASK,
615	VMX_VMCS_CTRL_CR4_MASK,
616	VMX_VMCS_CTRL_CR0_READ_SHADOW,
617	VMX_VMCS_CTRL_CR4_READ_SHADOW,
618	VMX_VMCS_CTRL_CR3_TARGET_VAL0,
619	VMX_VMCS_CTRL_CR3_TARGET_VAL1,
620	VMX_VMCS_CTRL_CR3_TARGET_VAL2,
621	VMX_VMCS_CTRL_CR3_TARGET_VAL3,
622
623	/* Natural-width read-only data fields. */
624	VMX_VMCS_RO_EXIT_QUALIFICATION,
625	VMX_VMCS_RO_IO_RCX,
626	VMX_VMCS_RO_IO_RSI,
627	VMX_VMCS_RO_IO_RDI,
628	VMX_VMCS_RO_IO_RIP,
629	VMX_VMCS_RO_GUEST_LINEAR_ADDR,
630
631	/* Natural-width guest-state field */
632	VMX_VMCS_GUEST_CR0,
633	VMX_VMCS_GUEST_CR3,
634	VMX_VMCS_GUEST_CR4,
635	VMX_VMCS_GUEST_ES_BASE,
636	VMX_VMCS_GUEST_CS_BASE,
637	VMX_VMCS_GUEST_SS_BASE,
638	VMX_VMCS_GUEST_DS_BASE,
639	VMX_VMCS_GUEST_FS_BASE,
640	VMX_VMCS_GUEST_GS_BASE,
641	VMX_VMCS_GUEST_LDTR_BASE,
642	VMX_VMCS_GUEST_TR_BASE,
643	VMX_VMCS_GUEST_GDTR_BASE,
644	VMX_VMCS_GUEST_IDTR_BASE,
645	VMX_VMCS_GUEST_DR7,
646	VMX_VMCS_GUEST_RSP,
647	VMX_VMCS_GUEST_RIP,
648	VMX_VMCS_GUEST_RFLAGS,
649	VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS,
650	VMX_VMCS_GUEST_SYSENTER_ESP,
651	VMX_VMCS_GUEST_SYSENTER_EIP,
652
653	/* Natural-width host-state fields */
654	VMX_VMCS_HOST_CR0,
655	VMX_VMCS_HOST_CR3,
656	VMX_VMCS_HOST_CR4,
657	VMX_VMCS_HOST_FS_BASE,
658	VMX_VMCS_HOST_GS_BASE,
659	VMX_VMCS_HOST_TR_BASE,
660	VMX_VMCS_HOST_GDTR_BASE,
661	VMX_VMCS_HOST_IDTR_BASE,
662	VMX_VMCS_HOST_SYSENTER_ESP,
663	VMX_VMCS_HOST_SYSENTER_EIP,
664	VMX_VMCS_HOST_RSP,
665	VMX_VMCS_HOST_RIP
666	};
667	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
668
669	static const uint32_t g_aVmcsSegBase[] =
670	{
671	VMX_VMCS_GUEST_ES_BASE,
672	VMX_VMCS_GUEST_CS_BASE,
673	VMX_VMCS_GUEST_SS_BASE,
674	VMX_VMCS_GUEST_DS_BASE,
675	VMX_VMCS_GUEST_FS_BASE,
676	VMX_VMCS_GUEST_GS_BASE
677	};
678	static const uint32_t g_aVmcsSegSel[] =
679	{
680	VMX_VMCS16_GUEST_ES_SEL,
681	VMX_VMCS16_GUEST_CS_SEL,
682	VMX_VMCS16_GUEST_SS_SEL,
683	VMX_VMCS16_GUEST_DS_SEL,
684	VMX_VMCS16_GUEST_FS_SEL,
685	VMX_VMCS16_GUEST_GS_SEL
686	};
687	static const uint32_t g_aVmcsSegLimit[] =
688	{
689	VMX_VMCS32_GUEST_ES_LIMIT,
690	VMX_VMCS32_GUEST_CS_LIMIT,
691	VMX_VMCS32_GUEST_SS_LIMIT,
692	VMX_VMCS32_GUEST_DS_LIMIT,
693	VMX_VMCS32_GUEST_FS_LIMIT,
694	VMX_VMCS32_GUEST_GS_LIMIT
695	};
696	static const uint32_t g_aVmcsSegAttr[] =
697	{
698	VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS,
699	VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS,
700	VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS,
701	VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS,
702	VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS,
703	VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS
704	};
705	AssertCompile(RT_ELEMENTS(g_aVmcsSegSel) == X86_SREG_COUNT);
706	AssertCompile(RT_ELEMENTS(g_aVmcsSegLimit) == X86_SREG_COUNT);
707	AssertCompile(RT_ELEMENTS(g_aVmcsSegBase) == X86_SREG_COUNT);
708	AssertCompile(RT_ELEMENTS(g_aVmcsSegAttr) == X86_SREG_COUNT);
709
710	#ifdef HMVMX_USE_FUNCTION_TABLE
711	/**
712	* VMX_EXIT dispatch table.
713	*/
714	static const PFNVMXEXITHANDLER g_apfnVMExitHandlers[VMX_EXIT_MAX + 1] =
715	{
716	/* 0 VMX_EXIT_XCPT_OR_NMI */ hmR0VmxExitXcptOrNmi,
717	/* 1 VMX_EXIT_EXT_INT */ hmR0VmxExitExtInt,
718	/* 2 VMX_EXIT_TRIPLE_FAULT */ hmR0VmxExitTripleFault,
719	/* 3 VMX_EXIT_INIT_SIGNAL */ hmR0VmxExitErrUnexpected,
720	/* 4 VMX_EXIT_SIPI */ hmR0VmxExitErrUnexpected,
721	/* 5 VMX_EXIT_IO_SMI */ hmR0VmxExitErrUnexpected,
722	/* 6 VMX_EXIT_SMI */ hmR0VmxExitErrUnexpected,
723	/* 7 VMX_EXIT_INT_WINDOW */ hmR0VmxExitIntWindow,
724	/* 8 VMX_EXIT_NMI_WINDOW */ hmR0VmxExitNmiWindow,
725	/* 9 VMX_EXIT_TASK_SWITCH */ hmR0VmxExitTaskSwitch,
726	/* 10 VMX_EXIT_CPUID */ hmR0VmxExitCpuid,
727	/* 11 VMX_EXIT_GETSEC */ hmR0VmxExitGetsec,
728	/* 12 VMX_EXIT_HLT */ hmR0VmxExitHlt,
729	/* 13 VMX_EXIT_INVD */ hmR0VmxExitInvd,
730	/* 14 VMX_EXIT_INVLPG */ hmR0VmxExitInvlpg,
731	/* 15 VMX_EXIT_RDPMC */ hmR0VmxExitRdpmc,
732	/* 16 VMX_EXIT_RDTSC */ hmR0VmxExitRdtsc,
733	/* 17 VMX_EXIT_RSM */ hmR0VmxExitErrUnexpected,
734	/* 18 VMX_EXIT_VMCALL */ hmR0VmxExitVmcall,
735	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
736	/* 19 VMX_EXIT_VMCLEAR */ hmR0VmxExitVmclear,
737	/* 20 VMX_EXIT_VMLAUNCH */ hmR0VmxExitVmlaunch,
738	/* 21 VMX_EXIT_VMPTRLD */ hmR0VmxExitVmptrld,
739	/* 22 VMX_EXIT_VMPTRST */ hmR0VmxExitVmptrst,
740	/* 23 VMX_EXIT_VMREAD */ hmR0VmxExitVmread,
741	/* 24 VMX_EXIT_VMRESUME */ hmR0VmxExitVmresume,
742	/* 25 VMX_EXIT_VMWRITE */ hmR0VmxExitVmwrite,
743	/* 26 VMX_EXIT_VMXOFF */ hmR0VmxExitVmxoff,
744	/* 27 VMX_EXIT_VMXON */ hmR0VmxExitVmxon,
745	#else
746	/* 19 VMX_EXIT_VMCLEAR */ hmR0VmxExitSetPendingXcptUD,
747	/* 20 VMX_EXIT_VMLAUNCH */ hmR0VmxExitSetPendingXcptUD,
748	/* 21 VMX_EXIT_VMPTRLD */ hmR0VmxExitSetPendingXcptUD,
749	/* 22 VMX_EXIT_VMPTRST */ hmR0VmxExitSetPendingXcptUD,
750	/* 23 VMX_EXIT_VMREAD */ hmR0VmxExitSetPendingXcptUD,
751	/* 24 VMX_EXIT_VMRESUME */ hmR0VmxExitSetPendingXcptUD,
752	/* 25 VMX_EXIT_VMWRITE */ hmR0VmxExitSetPendingXcptUD,
753	/* 26 VMX_EXIT_VMXOFF */ hmR0VmxExitSetPendingXcptUD,
754	/* 27 VMX_EXIT_VMXON */ hmR0VmxExitSetPendingXcptUD,
755	#endif
756	/* 28 VMX_EXIT_MOV_CRX */ hmR0VmxExitMovCRx,
757	/* 29 VMX_EXIT_MOV_DRX */ hmR0VmxExitMovDRx,
758	/* 30 VMX_EXIT_IO_INSTR */ hmR0VmxExitIoInstr,
759	/* 31 VMX_EXIT_RDMSR */ hmR0VmxExitRdmsr,
760	/* 32 VMX_EXIT_WRMSR */ hmR0VmxExitWrmsr,
761	/* 33 VMX_EXIT_ERR_INVALID_GUEST_STATE */ hmR0VmxExitErrInvalidGuestState,
762	/* 34 VMX_EXIT_ERR_MSR_LOAD */ hmR0VmxExitErrUnexpected,
763	/* 35 UNDEFINED */ hmR0VmxExitErrUnexpected,
764	/* 36 VMX_EXIT_MWAIT */ hmR0VmxExitMwait,
765	/* 37 VMX_EXIT_MTF */ hmR0VmxExitMtf,
766	/* 38 UNDEFINED */ hmR0VmxExitErrUnexpected,
767	/* 39 VMX_EXIT_MONITOR */ hmR0VmxExitMonitor,
768	/* 40 VMX_EXIT_PAUSE */ hmR0VmxExitPause,
769	/* 41 VMX_EXIT_ERR_MACHINE_CHECK */ hmR0VmxExitErrUnexpected,
770	/* 42 UNDEFINED */ hmR0VmxExitErrUnexpected,
771	/* 43 VMX_EXIT_TPR_BELOW_THRESHOLD */ hmR0VmxExitTprBelowThreshold,
772	/* 44 VMX_EXIT_APIC_ACCESS */ hmR0VmxExitApicAccess,
773	/* 45 VMX_EXIT_VIRTUALIZED_EOI */ hmR0VmxExitErrUnexpected,
774	/* 46 VMX_EXIT_GDTR_IDTR_ACCESS */ hmR0VmxExitErrUnexpected,
775	/* 47 VMX_EXIT_LDTR_TR_ACCESS */ hmR0VmxExitErrUnexpected,
776	/* 48 VMX_EXIT_EPT_VIOLATION */ hmR0VmxExitEptViolation,
777	/* 49 VMX_EXIT_EPT_MISCONFIG */ hmR0VmxExitEptMisconfig,
778	/* 50 VMX_EXIT_INVEPT */ hmR0VmxExitSetPendingXcptUD,
779	/* 51 VMX_EXIT_RDTSCP */ hmR0VmxExitRdtscp,
780	/* 52 VMX_EXIT_PREEMPT_TIMER */ hmR0VmxExitPreemptTimer,
781	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
782	/* 53 VMX_EXIT_INVVPID */ hmR0VmxExitInvvpid,
783	#else
784	/* 53 VMX_EXIT_INVVPID */ hmR0VmxExitSetPendingXcptUD,
785	#endif
786	/* 54 VMX_EXIT_WBINVD */ hmR0VmxExitWbinvd,
787	/* 55 VMX_EXIT_XSETBV */ hmR0VmxExitXsetbv,
788	/* 56 VMX_EXIT_APIC_WRITE */ hmR0VmxExitErrUnexpected,
789	/* 57 VMX_EXIT_RDRAND */ hmR0VmxExitErrUnexpected,
790	/* 58 VMX_EXIT_INVPCID */ hmR0VmxExitInvpcid,
791	/* 59 VMX_EXIT_VMFUNC */ hmR0VmxExitErrUnexpected,
792	/* 60 VMX_EXIT_ENCLS */ hmR0VmxExitErrUnexpected,
793	/* 61 VMX_EXIT_RDSEED */ hmR0VmxExitErrUnexpected,
794	/* 62 VMX_EXIT_PML_FULL */ hmR0VmxExitErrUnexpected,
795	/* 63 VMX_EXIT_XSAVES */ hmR0VmxExitErrUnexpected,
796	/* 64 VMX_EXIT_XRSTORS */ hmR0VmxExitErrUnexpected,
797	/* 65 UNDEFINED */ hmR0VmxExitErrUnexpected,
798	/* 66 VMX_EXIT_SPP_EVENT */ hmR0VmxExitErrUnexpected,
799	/* 67 VMX_EXIT_UMWAIT */ hmR0VmxExitErrUnexpected,
800	/* 68 VMX_EXIT_TPAUSE */ hmR0VmxExitErrUnexpected,
801	};
802	#endif /* HMVMX_USE_FUNCTION_TABLE */
803
804	#if defined(VBOX_STRICT) && defined(LOG_ENABLED)
805	static const char * const g_apszVmxInstrErrors[HMVMX_INSTR_ERROR_MAX + 1] =
806	{
807	/* 0 */ "(Not Used)",
808	/* 1 */ "VMCALL executed in VMX root operation.",
809	/* 2 */ "VMCLEAR with invalid physical address.",
810	/* 3 */ "VMCLEAR with VMXON pointer.",
811	/* 4 */ "VMLAUNCH with non-clear VMCS.",
812	/* 5 */ "VMRESUME with non-launched VMCS.",
813	/* 6 */ "VMRESUME after VMXOFF",
814	/* 7 */ "VM-entry with invalid control fields.",
815	/* 8 */ "VM-entry with invalid host state fields.",
816	/* 9 */ "VMPTRLD with invalid physical address.",
817	/* 10 */ "VMPTRLD with VMXON pointer.",
818	/* 11 */ "VMPTRLD with incorrect revision identifier.",
819	/* 12 */ "VMREAD/VMWRITE from/to unsupported VMCS component.",
820	/* 13 */ "VMWRITE to read-only VMCS component.",
821	/* 14 */ "(Not Used)",
822	/* 15 */ "VMXON executed in VMX root operation.",
823	/* 16 */ "VM-entry with invalid executive-VMCS pointer.",
824	/* 17 */ "VM-entry with non-launched executing VMCS.",
825	/* 18 */ "VM-entry with executive-VMCS pointer not VMXON pointer.",
826	/* 19 */ "VMCALL with non-clear VMCS.",
827	/* 20 */ "VMCALL with invalid VM-exit control fields.",
828	/* 21 */ "(Not Used)",
829	/* 22 */ "VMCALL with incorrect MSEG revision identifier.",
830	/* 23 */ "VMXOFF under dual monitor treatment of SMIs and SMM.",
831	/* 24 */ "VMCALL with invalid SMM-monitor features.",
832	/* 25 */ "VM-entry with invalid VM-execution control fields in executive VMCS.",
833	/* 26 */ "VM-entry with events blocked by MOV SS.",
834	/* 27 */ "(Not Used)",
835	/* 28 */ "Invalid operand to INVEPT/INVVPID."
836	};
837	#endif /* VBOX_STRICT && LOG_ENABLED */
838
839
840	/**
841	* Get the CR0 guest/host mask that does not change through the lifetime of a VM.
842	*
843	* Any bit set in this mask is owned by the host/hypervisor and would cause a
844	* VM-exit when modified by the guest.
845	*
846	* @returns The static CR0 guest/host mask.
847	* @param pVCpu The cross context virtual CPU structure.
848	*/
849	DECL_FORCE_INLINE(uint64_t) hmR0VmxGetFixedCr0Mask(PCVMCPUCC pVCpu)
850	{
851	/*
852	* Modifications to CR0 bits that VT-x ignores saving/restoring (CD, ET, NW) and
853	* to CR0 bits that we require for shadow paging (PG) by the guest must cause VM-exits.
854	*/
855	/** @todo Avoid intercepting CR0.PE with unrestricted guest execution. Fix PGM
856	* enmGuestMode to be in-sync with the current mode. See @bugref{6398}
857	* and @bugref{6944}. */
858	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
859	return ( X86_CR0_PE
860	\| X86_CR0_NE
861	\| (pVM->hm.s.fNestedPaging ? 0 : X86_CR0_WP)
862	\| X86_CR0_PG
863	\| X86_CR0_ET /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.ET */
864	\| X86_CR0_CD /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.CD */
865	\| X86_CR0_NW); /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.NW */
866	}
867
868
869	/**
870	* Gets the CR4 guest/host mask that does not change through the lifetime of a VM.
871	*
872	* Any bit set in this mask is owned by the host/hypervisor and would cause a
873	* VM-exit when modified by the guest.
874	*
875	* @returns The static CR4 guest/host mask.
876	* @param pVCpu The cross context virtual CPU structure.
877	*/
878	DECL_FORCE_INLINE(uint64_t) hmR0VmxGetFixedCr4Mask(PCVMCPUCC pVCpu)
879	{
880	/*
881	* We need to look at the host features here (for e.g. OSXSAVE, PCID) because
882	* these bits are reserved on hardware that does not support them. Since the
883	* CPU cannot refer to our virtual CPUID, we need to intercept CR4 changes to
884	* these bits and handle it depending on whether we expose them to the guest.
885	*/
886	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
887	bool const fXSaveRstor = pVM->cpum.ro.HostFeatures.fXSaveRstor;
888	bool const fPcid = pVM->cpum.ro.HostFeatures.fPcid;
889	return ( X86_CR4_VMXE
890	\| X86_CR4_VME
891	\| X86_CR4_PAE
892	\| X86_CR4_PGE
893	\| X86_CR4_PSE
894	\| (fXSaveRstor ? X86_CR4_OSXSAVE : 0)
895	\| (fPcid ? X86_CR4_PCIDE : 0));
896	}
897
898
899	/**
900	* Returns whether the the VM-exit MSR-store area differs from the VM-exit MSR-load
901	* area.
902	*
903	* @returns @c true if it's different, @c false otherwise.
904	* @param pVmcsInfo The VMCS info. object.
905	*/
906	DECL_FORCE_INLINE(bool) hmR0VmxIsSeparateExitMsrStoreAreaVmcs(PCVMXVMCSINFO pVmcsInfo)
907	{
908	return RT_BOOL( pVmcsInfo->pvGuestMsrStore != pVmcsInfo->pvGuestMsrLoad
909	&& pVmcsInfo->pvGuestMsrStore);
910	}
911
912
913	/**
914	* Sets the given Processor-based VM-execution controls.
915	*
916	* @param pVmxTransient The VMX-transient structure.
917	* @param uProcCtls The Processor-based VM-execution controls to set.
918	*/
919	static void hmR0VmxSetProcCtlsVmcs(PVMXTRANSIENT pVmxTransient, uint32_t uProcCtls)
920	{
921	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
922	if ((pVmcsInfo->u32ProcCtls & uProcCtls) != uProcCtls)
923	{
924	pVmcsInfo->u32ProcCtls \|= uProcCtls;
925	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
926	AssertRC(rc);
927	}
928	}
929
930
931	/**
932	* Removes the given Processor-based VM-execution controls.
933	*
934	* @param pVCpu The cross context virtual CPU structure.
935	* @param pVmxTransient The VMX-transient structure.
936	* @param uProcCtls The Processor-based VM-execution controls to remove.
937	*
938	* @remarks When executing a nested-guest, this will not remove any of the specified
939	* controls if the guest hypervisor has set any one of them.
940	*/
941	static void hmR0VmxRemoveProcCtlsVmcs(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uProcCtls)
942	{
943	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
944	if (pVmcsInfo->u32ProcCtls & uProcCtls)
945	{
946	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
947	bool const fRemoveCtls = !pVmxTransient->fIsNestedGuest
948	? true
949	: !CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, uProcCtls);
950	#else
951	NOREF(pVCpu);
952	bool const fRemoveCtls = true;
953	#endif
954	if (fRemoveCtls)
955	{
956	pVmcsInfo->u32ProcCtls &= ~uProcCtls;
957	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
958	AssertRC(rc);
959	}
960	}
961	}
962
963
964	/**
965	* Sets the TSC offset for the current VMCS.
966	*
967	* @param uTscOffset The TSC offset to set.
968	* @param pVmcsInfo The VMCS info. object.
969	*/
970	static void hmR0VmxSetTscOffsetVmcs(PVMXVMCSINFO pVmcsInfo, uint64_t uTscOffset)
971	{
972	if (pVmcsInfo->u64TscOffset != uTscOffset)
973	{
974	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, uTscOffset);
975	AssertRC(rc);
976	pVmcsInfo->u64TscOffset = uTscOffset;
977	}
978	}
979
980
981	/**
982	* Adds one or more exceptions to the exception bitmap and commits it to the current
983	* VMCS.
984	*
985	* @param pVmxTransient The VMX-transient structure.
986	* @param uXcptMask The exception(s) to add.
987	*/
988	static void hmR0VmxAddXcptInterceptMask(PVMXTRANSIENT pVmxTransient, uint32_t uXcptMask)
989	{
990	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
991	uint32_t uXcptBitmap = pVmcsInfo->u32XcptBitmap;
992	if ((uXcptBitmap & uXcptMask) != uXcptMask)
993	{
994	uXcptBitmap \|= uXcptMask;
995	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
996	AssertRC(rc);
997	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
998	}
999	}
1000
1001
1002	/**
1003	* Adds an exception to the exception bitmap and commits it to the current VMCS.
1004	*
1005	* @param pVmxTransient The VMX-transient structure.
1006	* @param uXcpt The exception to add.
1007	*/
1008	static void hmR0VmxAddXcptIntercept(PVMXTRANSIENT pVmxTransient, uint8_t uXcpt)
1009	{
1010	Assert(uXcpt <= X86_XCPT_LAST);
1011	hmR0VmxAddXcptInterceptMask(pVmxTransient, RT_BIT_32(uXcpt));
1012	}
1013
1014
1015	/**
1016	* Remove one or more exceptions from the exception bitmap and commits it to the
1017	* current VMCS.
1018	*
1019	* This takes care of not removing the exception intercept if a nested-guest
1020	* requires the exception to be intercepted.
1021	*
1022	* @returns VBox status code.
1023	* @param pVCpu The cross context virtual CPU structure.
1024	* @param pVmxTransient The VMX-transient structure.
1025	* @param uXcptMask The exception(s) to remove.
1026	*/
1027	static int hmR0VmxRemoveXcptInterceptMask(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uXcptMask)
1028	{
1029	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1030	uint32_t u32XcptBitmap = pVmcsInfo->u32XcptBitmap;
1031	if (u32XcptBitmap & uXcptMask)
1032	{
1033	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1034	if (!pVmxTransient->fIsNestedGuest)
1035	{ /* likely */ }
1036	else
1037	{
1038	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1039	uXcptMask &= ~pVmcsNstGst->u32XcptBitmap;
1040	}
1041	#endif
1042	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
1043	uXcptMask &= ~( RT_BIT(X86_XCPT_BP)
1044	\| RT_BIT(X86_XCPT_DE)
1045	\| RT_BIT(X86_XCPT_NM)
1046	\| RT_BIT(X86_XCPT_TS)
1047	\| RT_BIT(X86_XCPT_UD)
1048	\| RT_BIT(X86_XCPT_NP)
1049	\| RT_BIT(X86_XCPT_SS)
1050	\| RT_BIT(X86_XCPT_GP)
1051	\| RT_BIT(X86_XCPT_PF)
1052	\| RT_BIT(X86_XCPT_MF));
1053	#elif defined(HMVMX_ALWAYS_TRAP_PF)
1054	uXcptMask &= ~RT_BIT(X86_XCPT_PF);
1055	#endif
1056	if (uXcptMask)
1057	{
1058	/* Validate we are not removing any essential exception intercepts. */
1059	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging \|\| !(uXcptMask & RT_BIT(X86_XCPT_PF)));
1060	NOREF(pVCpu);
1061	Assert(!(uXcptMask & RT_BIT(X86_XCPT_DB)));
1062	Assert(!(uXcptMask & RT_BIT(X86_XCPT_AC)));
1063
1064	/* Remove it from the exception bitmap. */
1065	u32XcptBitmap &= ~uXcptMask;
1066
1067	/* Commit and update the cache if necessary. */
1068	if (pVmcsInfo->u32XcptBitmap != u32XcptBitmap)
1069	{
1070	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
1071	AssertRC(rc);
1072	pVmcsInfo->u32XcptBitmap = u32XcptBitmap;
1073	}
1074	}
1075	}
1076	return VINF_SUCCESS;
1077	}
1078
1079
1080	/**
1081	* Remove an exceptions from the exception bitmap and commits it to the current
1082	* VMCS.
1083	*
1084	* @returns VBox status code.
1085	* @param pVCpu The cross context virtual CPU structure.
1086	* @param pVmxTransient The VMX-transient structure.
1087	* @param uXcpt The exception to remove.
1088	*/
1089	static int hmR0VmxRemoveXcptIntercept(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint8_t uXcpt)
1090	{
1091	return hmR0VmxRemoveXcptInterceptMask(pVCpu, pVmxTransient, RT_BIT(uXcpt));
1092	}
1093
1094
1095	/**
1096	* Loads the VMCS specified by the VMCS info. object.
1097	*
1098	* @returns VBox status code.
1099	* @param pVmcsInfo The VMCS info. object.
1100	*
1101	* @remarks Can be called with interrupts disabled.
1102	*/
1103	static int hmR0VmxLoadVmcs(PVMXVMCSINFO pVmcsInfo)
1104	{
1105	Assert(pVmcsInfo->HCPhysVmcs != 0 && pVmcsInfo->HCPhysVmcs != NIL_RTHCPHYS);
1106	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1107
1108	int rc = VMXLoadVmcs(pVmcsInfo->HCPhysVmcs);
1109	if (RT_SUCCESS(rc))
1110	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_CURRENT;
1111	return rc;
1112	}
1113
1114
1115	/**
1116	* Clears the VMCS specified by the VMCS info. object.
1117	*
1118	* @returns VBox status code.
1119	* @param pVmcsInfo The VMCS info. object.
1120	*
1121	* @remarks Can be called with interrupts disabled.
1122	*/
1123	static int hmR0VmxClearVmcs(PVMXVMCSINFO pVmcsInfo)
1124	{
1125	Assert(pVmcsInfo->HCPhysVmcs != 0 && pVmcsInfo->HCPhysVmcs != NIL_RTHCPHYS);
1126	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1127
1128	int rc = VMXClearVmcs(pVmcsInfo->HCPhysVmcs);
1129	if (RT_SUCCESS(rc))
1130	pVmcsInfo->fVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
1131	return rc;
1132	}
1133
1134
1135	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1136	/**
1137	* Loads the shadow VMCS specified by the VMCS info. object.
1138	*
1139	* @returns VBox status code.
1140	* @param pVmcsInfo The VMCS info. object.
1141	*
1142	* @remarks Can be called with interrupts disabled.
1143	*/
1144	static int hmR0VmxLoadShadowVmcs(PVMXVMCSINFO pVmcsInfo)
1145	{
1146	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1147	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
1148
1149	int rc = VMXLoadVmcs(pVmcsInfo->HCPhysShadowVmcs);
1150	if (RT_SUCCESS(rc))
1151	pVmcsInfo->fShadowVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_CURRENT;
1152	return rc;
1153	}
1154
1155
1156	/**
1157	* Clears the shadow VMCS specified by the VMCS info. object.
1158	*
1159	* @returns VBox status code.
1160	* @param pVmcsInfo The VMCS info. object.
1161	*
1162	* @remarks Can be called with interrupts disabled.
1163	*/
1164	static int hmR0VmxClearShadowVmcs(PVMXVMCSINFO pVmcsInfo)
1165	{
1166	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1167	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
1168
1169	int rc = VMXClearVmcs(pVmcsInfo->HCPhysShadowVmcs);
1170	if (RT_SUCCESS(rc))
1171	pVmcsInfo->fShadowVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
1172	return rc;
1173	}
1174
1175
1176	/**
1177	* Switches from and to the specified VMCSes.
1178	*
1179	* @returns VBox status code.
1180	* @param pVmcsInfoFrom The VMCS info. object we are switching from.
1181	* @param pVmcsInfoTo The VMCS info. object we are switching to.
1182	*
1183	* @remarks Called with interrupts disabled.
1184	*/
1185	static int hmR0VmxSwitchVmcs(PVMXVMCSINFO pVmcsInfoFrom, PVMXVMCSINFO pVmcsInfoTo)
1186	{
1187	/*
1188	* Clear the VMCS we are switching out if it has not already been cleared.
1189	* This will sync any CPU internal data back to the VMCS.
1190	*/
1191	if (pVmcsInfoFrom->fVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
1192	{
1193	int rc = hmR0VmxClearVmcs(pVmcsInfoFrom);
1194	if (RT_SUCCESS(rc))
1195	{
1196	/*
1197	* The shadow VMCS, if any, would not be active at this point since we
1198	* would have cleared it while importing the virtual hardware-virtualization
1199	* state as part the VMLAUNCH/VMRESUME VM-exit. Hence, there's no need to
1200	* clear the shadow VMCS here, just assert for safety.
1201	*/
1202	Assert(!pVmcsInfoFrom->pvShadowVmcs \|\| pVmcsInfoFrom->fShadowVmcsState == VMX_V_VMCS_LAUNCH_STATE_CLEAR);
1203	}
1204	else
1205	return rc;
1206	}
1207
1208	/*
1209	* Clear the VMCS we are switching to if it has not already been cleared.
1210	* This will initialize the VMCS launch state to "clear" required for loading it.
1211	*
1212	* See Intel spec. 31.6 "Preparation And Launching A Virtual Machine".
1213	*/
1214	if (pVmcsInfoTo->fVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
1215	{
1216	int rc = hmR0VmxClearVmcs(pVmcsInfoTo);
1217	if (RT_SUCCESS(rc))
1218	{ /* likely */ }
1219	else
1220	return rc;
1221	}
1222
1223	/*
1224	* Finally, load the VMCS we are switching to.
1225	*/
1226	return hmR0VmxLoadVmcs(pVmcsInfoTo);
1227	}
1228
1229
1230	/**
1231	* Switches between the guest VMCS and the nested-guest VMCS as specified by the
1232	* caller.
1233	*
1234	* @returns VBox status code.
1235	* @param pVCpu The cross context virtual CPU structure.
1236	* @param fSwitchToNstGstVmcs Whether to switch to the nested-guest VMCS (pass
1237	* true) or guest VMCS (pass false).
1238	*/
1239	static int hmR0VmxSwitchToGstOrNstGstVmcs(PVMCPUCC pVCpu, bool fSwitchToNstGstVmcs)
1240	{
1241	/* Ensure we have synced everything from the guest-CPU context to the VMCS before switching. */
1242	HMVMX_CPUMCTX_ASSERT(pVCpu, HMVMX_CPUMCTX_EXTRN_ALL);
1243
1244	PVMXVMCSINFO pVmcsInfoFrom;
1245	PVMXVMCSINFO pVmcsInfoTo;
1246	if (fSwitchToNstGstVmcs)
1247	{
1248	pVmcsInfoFrom = &pVCpu->hm.s.vmx.VmcsInfo;
1249	pVmcsInfoTo = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
1250	}
1251	else
1252	{
1253	pVmcsInfoFrom = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
1254	pVmcsInfoTo = &pVCpu->hm.s.vmx.VmcsInfo;
1255	}
1256
1257	/*
1258	* Disable interrupts to prevent being preempted while we switch the current VMCS as the
1259	* preemption hook code path acquires the current VMCS.
1260	*/
1261	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1262
1263	int rc = hmR0VmxSwitchVmcs(pVmcsInfoFrom, pVmcsInfoTo);
1264	if (RT_SUCCESS(rc))
1265	{
1266	pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs = fSwitchToNstGstVmcs;
1267
1268	/*
1269	* If we are switching to a VMCS that was executed on a different host CPU or was
1270	* never executed before, flag that we need to export the host state before executing
1271	* guest/nested-guest code using hardware-assisted VMX.
1272	*
1273	* This could probably be done in a preemptible context since the preemption hook
1274	* will flag the necessary change in host context. However, since preemption is
1275	* already disabled and to avoid making assumptions about host specific code in
1276	* RTMpCpuId when called with preemption enabled, we'll do this while preemption is
1277	* disabled.
1278	*/
1279	if (pVmcsInfoTo->idHostCpuState == RTMpCpuId())
1280	{ /* likely */ }
1281	else
1282	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE);
1283
1284	ASMSetFlags(fEFlags);
1285
1286	/*
1287	* We use a different VM-exit MSR-store areas for the guest and nested-guest. Hence,
1288	* flag that we need to update the host MSR values there. Even if we decide in the
1289	* future to share the VM-exit MSR-store area page between the guest and nested-guest,
1290	* if its content differs, we would have to update the host MSRs anyway.
1291	*/
1292	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = false;
1293	}
1294	else
1295	ASMSetFlags(fEFlags);
1296	return rc;
1297	}
1298	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
1299
1300
1301	/**
1302	* Updates the VM's last error record.
1303	*
1304	* If there was a VMX instruction error, reads the error data from the VMCS and
1305	* updates VCPU's last error record as well.
1306	*
1307	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
1308	* Can be NULL if @a rc is not VERR_VMX_UNABLE_TO_START_VM or
1309	* VERR_VMX_INVALID_VMCS_FIELD.
1310	* @param rc The error code.
1311	*/
1312	static void hmR0VmxUpdateErrorRecord(PVMCPUCC pVCpu, int rc)
1313	{
1314	if ( rc == VERR_VMX_INVALID_VMCS_FIELD
1315	\|\| rc == VERR_VMX_UNABLE_TO_START_VM)
1316	{
1317	AssertPtrReturnVoid(pVCpu);
1318	VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
1319	}
1320	pVCpu->CTX_SUFF(pVM)->hm.s.rcInit = rc;
1321	}
1322
1323
1324	#ifdef VBOX_STRICT
1325	/**
1326	* Reads the VM-entry interruption-information field from the VMCS into the VMX
1327	* transient structure.
1328	*
1329	* @param pVmxTransient The VMX-transient structure.
1330	*/
1331	DECLINLINE(void) hmR0VmxReadEntryIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
1332	{
1333	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &pVmxTransient->uEntryIntInfo);
1334	AssertRC(rc);
1335	}
1336
1337
1338	/**
1339	* Reads the VM-entry exception error code field from the VMCS into
1340	* the VMX transient structure.
1341	*
1342	* @param pVmxTransient The VMX-transient structure.
1343	*/
1344	DECLINLINE(void) hmR0VmxReadEntryXcptErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1345	{
1346	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &pVmxTransient->uEntryXcptErrorCode);
1347	AssertRC(rc);
1348	}
1349
1350
1351	/**
1352	* Reads the VM-entry exception error code field from the VMCS into
1353	* the VMX transient structure.
1354	*
1355	* @param pVmxTransient The VMX-transient structure.
1356	*/
1357	DECLINLINE(void) hmR0VmxReadEntryInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
1358	{
1359	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &pVmxTransient->cbEntryInstr);
1360	AssertRC(rc);
1361	}
1362	#endif /* VBOX_STRICT */
1363
1364
1365	/**
1366	* Reads the VM-exit interruption-information field from the VMCS into the VMX
1367	* transient structure.
1368	*
1369	* @param pVmxTransient The VMX-transient structure.
1370	*/
1371	DECLINLINE(void) hmR0VmxReadExitIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
1372	{
1373	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INTERRUPTION_INFO))
1374	{
1375	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
1376	AssertRC(rc);
1377	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INTERRUPTION_INFO;
1378	}
1379	}
1380
1381
1382	/**
1383	* Reads the VM-exit interruption error code from the VMCS into the VMX
1384	* transient structure.
1385	*
1386	* @param pVmxTransient The VMX-transient structure.
1387	*/
1388	DECLINLINE(void) hmR0VmxReadExitIntErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1389	{
1390	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE))
1391	{
1392	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
1393	AssertRC(rc);
1394	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE;
1395	}
1396	}
1397
1398
1399	/**
1400	* Reads the VM-exit instruction length field from the VMCS into the VMX
1401	* transient structure.
1402	*
1403	* @param pVmxTransient The VMX-transient structure.
1404	*/
1405	DECLINLINE(void) hmR0VmxReadExitInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
1406	{
1407	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INSTR_LEN))
1408	{
1409	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbExitInstr);
1410	AssertRC(rc);
1411	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INSTR_LEN;
1412	}
1413	}
1414
1415
1416	/**
1417	* Reads the VM-exit instruction-information field from the VMCS into
1418	* the VMX transient structure.
1419	*
1420	* @param pVmxTransient The VMX-transient structure.
1421	*/
1422	DECLINLINE(void) hmR0VmxReadExitInstrInfoVmcs(PVMXTRANSIENT pVmxTransient)
1423	{
1424	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INSTR_INFO))
1425	{
1426	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
1427	AssertRC(rc);
1428	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INSTR_INFO;
1429	}
1430	}
1431
1432
1433	/**
1434	* Reads the Exit Qualification from the VMCS into the VMX transient structure.
1435	*
1436	* @param pVmxTransient The VMX-transient structure.
1437	*/
1438	DECLINLINE(void) hmR0VmxReadExitQualVmcs(PVMXTRANSIENT pVmxTransient)
1439	{
1440	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_QUALIFICATION))
1441	{
1442	int rc = VMXReadVmcsNw(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQual);
1443	AssertRC(rc);
1444	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_QUALIFICATION;
1445	}
1446	}
1447
1448
1449	/**
1450	* Reads the Guest-linear address from the VMCS into the VMX transient structure.
1451	*
1452	* @param pVmxTransient The VMX-transient structure.
1453	*/
1454	DECLINLINE(void) hmR0VmxReadGuestLinearAddrVmcs(PVMXTRANSIENT pVmxTransient)
1455	{
1456	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_LINEAR_ADDR))
1457	{
1458	int rc = VMXReadVmcsNw(VMX_VMCS_RO_GUEST_LINEAR_ADDR, &pVmxTransient->uGuestLinearAddr);
1459	AssertRC(rc);
1460	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_LINEAR_ADDR;
1461	}
1462	}
1463
1464
1465	/**
1466	* Reads the Guest-physical address from the VMCS into the VMX transient structure.
1467	*
1468	* @param pVmxTransient The VMX-transient structure.
1469	*/
1470	DECLINLINE(void) hmR0VmxReadGuestPhysicalAddrVmcs(PVMXTRANSIENT pVmxTransient)
1471	{
1472	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_PHYSICAL_ADDR))
1473	{
1474	int rc = VMXReadVmcs64(VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL, &pVmxTransient->uGuestPhysicalAddr);
1475	AssertRC(rc);
1476	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_PHYSICAL_ADDR;
1477	}
1478	}
1479
1480	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1481	/**
1482	* Reads the Guest pending-debug exceptions from the VMCS into the VMX transient
1483	* structure.
1484	*
1485	* @param pVmxTransient The VMX-transient structure.
1486	*/
1487	DECLINLINE(void) hmR0VmxReadGuestPendingDbgXctps(PVMXTRANSIENT pVmxTransient)
1488	{
1489	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_PENDING_DBG_XCPTS))
1490	{
1491	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, &pVmxTransient->uGuestPendingDbgXcpts);
1492	AssertRC(rc);
1493	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_PENDING_DBG_XCPTS;
1494	}
1495	}
1496	#endif
1497
1498	/**
1499	* Reads the IDT-vectoring information field from the VMCS into the VMX
1500	* transient structure.
1501	*
1502	* @param pVmxTransient The VMX-transient structure.
1503	*
1504	* @remarks No-long-jump zone!!!
1505	*/
1506	DECLINLINE(void) hmR0VmxReadIdtVectoringInfoVmcs(PVMXTRANSIENT pVmxTransient)
1507	{
1508	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_IDT_VECTORING_INFO))
1509	{
1510	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_INFO, &pVmxTransient->uIdtVectoringInfo);
1511	AssertRC(rc);
1512	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_IDT_VECTORING_INFO;
1513	}
1514	}
1515
1516
1517	/**
1518	* Reads the IDT-vectoring error code from the VMCS into the VMX
1519	* transient structure.
1520	*
1521	* @param pVmxTransient The VMX-transient structure.
1522	*/
1523	DECLINLINE(void) hmR0VmxReadIdtVectoringErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1524	{
1525	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_IDT_VECTORING_ERROR_CODE))
1526	{
1527	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
1528	AssertRC(rc);
1529	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_IDT_VECTORING_ERROR_CODE;
1530	}
1531	}
1532
1533	#ifdef HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
1534	/**
1535	* Reads all relevant read-only VMCS fields into the VMX transient structure.
1536	*
1537	* @param pVmxTransient The VMX-transient structure.
1538	*/
1539	static void hmR0VmxReadAllRoFieldsVmcs(PVMXTRANSIENT pVmxTransient)
1540	{
1541	int rc = VMXReadVmcsNw(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQual);
1542	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbExitInstr);
1543	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
1544	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_INFO, &pVmxTransient->uIdtVectoringInfo);
1545	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
1546	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
1547	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
1548	rc \|= VMXReadVmcsNw(VMX_VMCS_RO_GUEST_LINEAR_ADDR, &pVmxTransient->uGuestLinearAddr);
1549	rc \|= VMXReadVmcs64(VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL, &pVmxTransient->uGuestPhysicalAddr);
1550	AssertRC(rc);
1551	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_QUALIFICATION
1552	\| HMVMX_READ_EXIT_INSTR_LEN
1553	\| HMVMX_READ_EXIT_INSTR_INFO
1554	\| HMVMX_READ_IDT_VECTORING_INFO
1555	\| HMVMX_READ_IDT_VECTORING_ERROR_CODE
1556	\| HMVMX_READ_EXIT_INTERRUPTION_INFO
1557	\| HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE
1558	\| HMVMX_READ_GUEST_LINEAR_ADDR
1559	\| HMVMX_READ_GUEST_PHYSICAL_ADDR;
1560	}
1561	#endif
1562
1563	/**
1564	* Enters VMX root mode operation on the current CPU.
1565	*
1566	* @returns VBox status code.
1567	* @param pVM The cross context VM structure. Can be
1568	* NULL, after a resume.
1569	* @param HCPhysCpuPage Physical address of the VMXON region.
1570	* @param pvCpuPage Pointer to the VMXON region.
1571	*/
1572	static int hmR0VmxEnterRootMode(PVMCC pVM, RTHCPHYS HCPhysCpuPage, void *pvCpuPage)
1573	{
1574	Assert(HCPhysCpuPage && HCPhysCpuPage != NIL_RTHCPHYS);
1575	Assert(RT_ALIGN_T(HCPhysCpuPage, _4K, RTHCPHYS) == HCPhysCpuPage);
1576	Assert(pvCpuPage);
1577	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1578
1579	if (pVM)
1580	{
1581	/* Write the VMCS revision identifier to the VMXON region. */
1582	(uint32_t )pvCpuPage = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_ID);
1583	}
1584
1585	/* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with CR4. */
1586	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1587
1588	/* Enable the VMX bit in CR4 if necessary. */
1589	RTCCUINTREG const uOldCr4 = SUPR0ChangeCR4(X86_CR4_VMXE, RTCCUINTREG_MAX);
1590
1591	/* Enter VMX root mode. */
1592	int rc = VMXEnable(HCPhysCpuPage);
1593	if (RT_FAILURE(rc))
1594	{
1595	if (!(uOldCr4 & X86_CR4_VMXE))
1596	SUPR0ChangeCR4(0 /* fOrMask */, ~X86_CR4_VMXE);
1597
1598	if (pVM)
1599	pVM->hm.s.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
1600	}
1601
1602	/* Restore interrupts. */
1603	ASMSetFlags(fEFlags);
1604	return rc;
1605	}
1606
1607
1608	/**
1609	* Exits VMX root mode operation on the current CPU.
1610	*
1611	* @returns VBox status code.
1612	*/
1613	static int hmR0VmxLeaveRootMode(void)
1614	{
1615	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1616
1617	/* Paranoid: Disable interrupts as, in theory, interrupts handlers might mess with CR4. */
1618	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1619
1620	/* If we're for some reason not in VMX root mode, then don't leave it. */
1621	RTCCUINTREG const uHostCr4 = ASMGetCR4();
1622
1623	int rc;
1624	if (uHostCr4 & X86_CR4_VMXE)
1625	{
1626	/* Exit VMX root mode and clear the VMX bit in CR4. */
1627	VMXDisable();
1628	SUPR0ChangeCR4(0 /* fOrMask */, ~X86_CR4_VMXE);
1629	rc = VINF_SUCCESS;
1630	}
1631	else
1632	rc = VERR_VMX_NOT_IN_VMX_ROOT_MODE;
1633
1634	/* Restore interrupts. */
1635	ASMSetFlags(fEFlags);
1636	return rc;
1637	}
1638
1639
1640	/**
1641	* Allocates and maps a physically contiguous page. The allocated page is
1642	* zero'd out (used by various VT-x structures).
1643	*
1644	* @returns IPRT status code.
1645	* @param pMemObj Pointer to the ring-0 memory object.
1646	* @param ppVirt Where to store the virtual address of the allocation.
1647	* @param pHCPhys Where to store the physical address of the allocation.
1648	*/
1649	static int hmR0VmxPageAllocZ(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys)
1650	{
1651	AssertPtr(pMemObj);
1652	AssertPtr(ppVirt);
1653	AssertPtr(pHCPhys);
1654	int rc = RTR0MemObjAllocCont(pMemObj, X86_PAGE_4K_SIZE, false /* fExecutable */);
1655	if (RT_FAILURE(rc))
1656	return rc;
1657	ppVirt = RTR0MemObjAddress(pMemObj);
1658	pHCPhys = RTR0MemObjGetPagePhysAddr(pMemObj, 0 /* iPage */);
1659	ASMMemZero32(*ppVirt, X86_PAGE_4K_SIZE);
1660	return VINF_SUCCESS;
1661	}
1662
1663
1664	/**
1665	* Frees and unmaps an allocated, physical page.
1666	*
1667	* @param pMemObj Pointer to the ring-0 memory object.
1668	* @param ppVirt Where to re-initialize the virtual address of allocation as
1669	* 0.
1670	* @param pHCPhys Where to re-initialize the physical address of the
1671	* allocation as 0.
1672	*/
1673	static void hmR0VmxPageFree(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys)
1674	{
1675	AssertPtr(pMemObj);
1676	AssertPtr(ppVirt);
1677	AssertPtr(pHCPhys);
1678	/* NULL is valid, accepted and ignored by the free function below. */
1679	RTR0MemObjFree(pMemObj, true / fFreeMappings */);
1680	*pMemObj = NIL_RTR0MEMOBJ;
1681	*ppVirt = NULL;
1682	*pHCPhys = NIL_RTHCPHYS;
1683	}
1684
1685
1686	/**
1687	* Initializes a VMCS info. object.
1688	*
1689	* @param pVmcsInfo The VMCS info. object.
1690	*/
1691	static void hmR0VmxInitVmcsInfo(PVMXVMCSINFO pVmcsInfo)
1692	{
1693	memset(pVmcsInfo, 0, sizeof(*pVmcsInfo));
1694
1695	Assert(pVmcsInfo->hMemObjVmcs == NIL_RTR0MEMOBJ);
1696	Assert(pVmcsInfo->hMemObjShadowVmcs == NIL_RTR0MEMOBJ);
1697	Assert(pVmcsInfo->hMemObjMsrBitmap == NIL_RTR0MEMOBJ);
1698	Assert(pVmcsInfo->hMemObjGuestMsrLoad == NIL_RTR0MEMOBJ);
1699	Assert(pVmcsInfo->hMemObjGuestMsrStore == NIL_RTR0MEMOBJ);
1700	Assert(pVmcsInfo->hMemObjHostMsrLoad == NIL_RTR0MEMOBJ);
1701	pVmcsInfo->HCPhysVmcs = NIL_RTHCPHYS;
1702	pVmcsInfo->HCPhysShadowVmcs = NIL_RTHCPHYS;
1703	pVmcsInfo->HCPhysMsrBitmap = NIL_RTHCPHYS;
1704	pVmcsInfo->HCPhysGuestMsrLoad = NIL_RTHCPHYS;
1705	pVmcsInfo->HCPhysGuestMsrStore = NIL_RTHCPHYS;
1706	pVmcsInfo->HCPhysHostMsrLoad = NIL_RTHCPHYS;
1707	pVmcsInfo->HCPhysVirtApic = NIL_RTHCPHYS;
1708	pVmcsInfo->HCPhysEPTP = NIL_RTHCPHYS;
1709	pVmcsInfo->u64VmcsLinkPtr = NIL_RTHCPHYS;
1710	pVmcsInfo->idHostCpuState = NIL_RTCPUID;
1711	pVmcsInfo->idHostCpuExec = NIL_RTCPUID;
1712	}
1713
1714
1715	/**
1716	* Frees the VT-x structures for a VMCS info. object.
1717	*
1718	* @param pVM The cross context VM structure.
1719	* @param pVmcsInfo The VMCS info. object.
1720	*/
1721	static void hmR0VmxFreeVmcsInfo(PVMCC pVM, PVMXVMCSINFO pVmcsInfo)
1722	{
1723	hmR0VmxPageFree(&pVmcsInfo->hMemObjVmcs, &pVmcsInfo->pvVmcs, &pVmcsInfo->HCPhysVmcs);
1724
1725	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1726	if (pVM->hm.s.vmx.fUseVmcsShadowing)
1727	hmR0VmxPageFree(&pVmcsInfo->hMemObjShadowVmcs, &pVmcsInfo->pvShadowVmcs, &pVmcsInfo->HCPhysShadowVmcs);
1728	#endif
1729
1730	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
1731	hmR0VmxPageFree(&pVmcsInfo->hMemObjMsrBitmap, &pVmcsInfo->pvMsrBitmap, &pVmcsInfo->HCPhysMsrBitmap);
1732
1733	hmR0VmxPageFree(&pVmcsInfo->hMemObjHostMsrLoad, &pVmcsInfo->pvHostMsrLoad, &pVmcsInfo->HCPhysHostMsrLoad);
1734	hmR0VmxPageFree(&pVmcsInfo->hMemObjGuestMsrLoad, &pVmcsInfo->pvGuestMsrLoad, &pVmcsInfo->HCPhysGuestMsrLoad);
1735	hmR0VmxPageFree(&pVmcsInfo->hMemObjGuestMsrStore, &pVmcsInfo->pvGuestMsrStore, &pVmcsInfo->HCPhysGuestMsrStore);
1736
1737	hmR0VmxInitVmcsInfo(pVmcsInfo);
1738	}
1739
1740
1741	/**
1742	* Allocates the VT-x structures for a VMCS info. object.
1743	*
1744	* @returns VBox status code.
1745	* @param pVCpu The cross context virtual CPU structure.
1746	* @param pVmcsInfo The VMCS info. object.
1747	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
1748	*/
1749	static int hmR0VmxAllocVmcsInfo(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
1750	{
1751	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1752
1753	/* Allocate the guest VM control structure (VMCS). */
1754	int rc = hmR0VmxPageAllocZ(&pVmcsInfo->hMemObjVmcs, &pVmcsInfo->pvVmcs, &pVmcsInfo->HCPhysVmcs);
1755	if (RT_SUCCESS(rc))
1756	{
1757	if (!fIsNstGstVmcs)
1758	{
1759	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1760	if (pVM->hm.s.vmx.fUseVmcsShadowing)
1761	rc = hmR0VmxPageAllocZ(&pVmcsInfo->hMemObjShadowVmcs, &pVmcsInfo->pvShadowVmcs, &pVmcsInfo->HCPhysShadowVmcs);
1762	#endif
1763	if (RT_SUCCESS(rc))
1764	{
1765	/* Get the allocated virtual-APIC page from the virtual APIC device. */
1766	if ( PDMHasApic(pVCpu->CTX_SUFF(pVM))
1767	&& (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW))
1768	rc = APICGetApicPageForCpu(pVCpu, &pVmcsInfo->HCPhysVirtApic, (PRTR0PTR)&pVmcsInfo->pbVirtApic, NULL /pR3Ptr/);
1769	}
1770	}
1771	else
1772	{
1773	/* We don't yet support exposing VMCS shadowing to the guest. */
1774	Assert(pVmcsInfo->HCPhysShadowVmcs == NIL_RTHCPHYS);
1775	Assert(!pVmcsInfo->pvShadowVmcs);
1776
1777	/* Get the allocated virtual-APIC page from CPUM. */
1778	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW)
1779	{
1780	/** @todo NSTVMX: Get rid of this. There is no need to allocate a separate HC
1781	* page for this. Use the one provided by the nested-guest directly. */
1782	pVmcsInfo->pbVirtApic = (uint8_t *)CPUMGetGuestVmxVirtApicPage(pVCpu, &pVCpu->cpum.GstCtx,
1783	&pVmcsInfo->HCPhysVirtApic);
1784	Assert(pVmcsInfo->pbVirtApic);
1785	Assert(pVmcsInfo->HCPhysVirtApic && pVmcsInfo->HCPhysVirtApic != NIL_RTHCPHYS);
1786	}
1787	}
1788
1789	if (RT_SUCCESS(rc))
1790	{
1791	/*
1792	* Allocate the MSR-bitmap if supported by the CPU. The MSR-bitmap is for
1793	* transparent accesses of specific MSRs.
1794	*
1795	* If the condition for enabling MSR bitmaps changes here, don't forget to
1796	* update HMIsMsrBitmapActive().
1797	*
1798	* We don't share MSR bitmaps between the guest and nested-guest as we then
1799	* don't need to care about carefully restoring the guest MSR bitmap.
1800	* The guest visible nested-guest MSR bitmap needs to remain unchanged.
1801	* Hence, allocate a separate MSR bitmap for the guest and nested-guest.
1802	* We also don't need to re-initialize the nested-guest MSR bitmap here as
1803	* we do that later while merging VMCS.
1804	*/
1805	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
1806	{
1807	rc = hmR0VmxPageAllocZ(&pVmcsInfo->hMemObjMsrBitmap, &pVmcsInfo->pvMsrBitmap, &pVmcsInfo->HCPhysMsrBitmap);
1808	if ( RT_SUCCESS(rc)
1809	&& !fIsNstGstVmcs)
1810	ASMMemFill32(pVmcsInfo->pvMsrBitmap, X86_PAGE_4K_SIZE, UINT32_C(0xffffffff));
1811	}
1812
1813	if (RT_SUCCESS(rc))
1814	{
1815	/*
1816	* Allocate the VM-entry MSR-load area for the guest MSRs.
1817	*
1818	* Similar to MSR-bitmaps, we do not share the auto MSR-load/store are between
1819	* the guest and nested-guest.
1820	*/
1821	rc = hmR0VmxPageAllocZ(&pVmcsInfo->hMemObjGuestMsrLoad, &pVmcsInfo->pvGuestMsrLoad,
1822	&pVmcsInfo->HCPhysGuestMsrLoad);
1823	if (RT_SUCCESS(rc))
1824	{
1825	/*
1826	* We use the same page for VM-entry MSR-load and VM-exit MSR store areas.
1827	* These contain the guest MSRs to load on VM-entry and store on VM-exit.
1828	*/
1829	Assert(pVmcsInfo->hMemObjGuestMsrStore == NIL_RTR0MEMOBJ);
1830	pVmcsInfo->pvGuestMsrStore = pVmcsInfo->pvGuestMsrLoad;
1831	pVmcsInfo->HCPhysGuestMsrStore = pVmcsInfo->HCPhysGuestMsrLoad;
1832
1833	/* Allocate the VM-exit MSR-load page for the host MSRs. */
1834	rc = hmR0VmxPageAllocZ(&pVmcsInfo->hMemObjHostMsrLoad, &pVmcsInfo->pvHostMsrLoad,
1835	&pVmcsInfo->HCPhysHostMsrLoad);
1836	}
1837	}
1838	}
1839	}
1840
1841	return rc;
1842	}
1843
1844
1845	/**
1846	* Free all VT-x structures for the VM.
1847	*
1848	* @returns IPRT status code.
1849	* @param pVM The cross context VM structure.
1850	*/
1851	static void hmR0VmxStructsFree(PVMCC pVM)
1852	{
1853	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
1854	hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch);
1855	#endif
1856	hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess, &pVM->hm.s.vmx.HCPhysApicAccess);
1857
1858	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1859	if (pVM->hm.s.vmx.fUseVmcsShadowing)
1860	{
1861	RTMemFree(pVM->hm.s.vmx.paShadowVmcsFields);
1862	RTMemFree(pVM->hm.s.vmx.paShadowVmcsRoFields);
1863	hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjVmreadBitmap, &pVM->hm.s.vmx.pvVmreadBitmap, &pVM->hm.s.vmx.HCPhysVmreadBitmap);
1864	hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjVmwriteBitmap, &pVM->hm.s.vmx.pvVmwriteBitmap, &pVM->hm.s.vmx.HCPhysVmwriteBitmap);
1865	}
1866	#endif
1867
1868	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
1869	{
1870	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
1871	PVMXVMCSINFO pVmcsInfo = &pVCpu->hm.s.vmx.VmcsInfo;
1872	hmR0VmxFreeVmcsInfo(pVM, pVmcsInfo);
1873	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1874	if (pVM->cpum.ro.GuestFeatures.fVmx)
1875	{
1876	pVmcsInfo = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
1877	hmR0VmxFreeVmcsInfo(pVM, pVmcsInfo);
1878	}
1879	#endif
1880	}
1881	}
1882
1883
1884	/**
1885	* Allocate all VT-x structures for the VM.
1886	*
1887	* @returns IPRT status code.
1888	* @param pVM The cross context VM structure.
1889	*/
1890	static int hmR0VmxStructsAlloc(PVMCC pVM)
1891	{
1892	/*
1893	* Sanity check the VMCS size reported by the CPU as we assume 4KB allocations.
1894	* The VMCS size cannot be more than 4096 bytes.
1895	*
1896	* See Intel spec. Appendix A.1 "Basic VMX Information".
1897	*/
1898	uint32_t const cbVmcs = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_SIZE);
1899	if (cbVmcs <= X86_PAGE_4K_SIZE)
1900	{ /* likely */ }
1901	else
1902	{
1903	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_INVALID_VMCS_SIZE;
1904	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
1905	}
1906
1907	/*
1908	* Initialize/check members up-front so we can cleanup en masse on allocation failures.
1909	*/
1910	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
1911	Assert(pVM->hm.s.vmx.hMemObjScratch == NIL_RTR0MEMOBJ);
1912	Assert(pVM->hm.s.vmx.pbScratch == NULL);
1913	pVM->hm.s.vmx.HCPhysScratch = NIL_RTHCPHYS;
1914	#endif
1915
1916	Assert(pVM->hm.s.vmx.hMemObjApicAccess == NIL_RTR0MEMOBJ);
1917	Assert(pVM->hm.s.vmx.pbApicAccess == NULL);
1918	pVM->hm.s.vmx.HCPhysApicAccess = NIL_RTHCPHYS;
1919
1920	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
1921	{
1922	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
1923	hmR0VmxInitVmcsInfo(&pVCpu->hm.s.vmx.VmcsInfo);
1924	hmR0VmxInitVmcsInfo(&pVCpu->hm.s.vmx.VmcsInfoNstGst);
1925	}
1926
1927	/*
1928	* Allocate per-VM VT-x structures.
1929	*/
1930	int rc = VINF_SUCCESS;
1931	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
1932	/* Allocate crash-dump magic scratch page. */
1933	rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch);
1934	if (RT_FAILURE(rc))
1935	{
1936	hmR0VmxStructsFree(pVM);
1937	return rc;
1938	}
1939	strcpy((char *)pVM->hm.s.vmx.pbScratch, "SCRATCH Magic");
1940	(uint64_t )(pVM->hm.s.vmx.pbScratch + 16) = UINT64_C(0xdeadbeefdeadbeef);
1941	#endif
1942
1943	/* Allocate the APIC-access page for trapping APIC accesses from the guest. */
1944	if (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
1945	{
1946	rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess,
1947	&pVM->hm.s.vmx.HCPhysApicAccess);
1948	if (RT_FAILURE(rc))
1949	{
1950	hmR0VmxStructsFree(pVM);
1951	return rc;
1952	}
1953	}
1954
1955	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1956	/* Allocate the shadow VMCS fields array, VMREAD, VMWRITE bitmaps.. */
1957	if (pVM->hm.s.vmx.fUseVmcsShadowing)
1958	{
1959	Assert(!pVM->hm.s.vmx.cShadowVmcsFields);
1960	Assert(!pVM->hm.s.vmx.cShadowVmcsRoFields);
1961	pVM->hm.s.vmx.paShadowVmcsFields = (uint32_t *)RTMemAllocZ(sizeof(g_aVmcsFields));
1962	pVM->hm.s.vmx.paShadowVmcsRoFields = (uint32_t *)RTMemAllocZ(sizeof(g_aVmcsFields));
1963	if (RT_LIKELY( pVM->hm.s.vmx.paShadowVmcsFields
1964	&& pVM->hm.s.vmx.paShadowVmcsRoFields))
1965	{
1966	rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjVmreadBitmap, &pVM->hm.s.vmx.pvVmreadBitmap,
1967	&pVM->hm.s.vmx.HCPhysVmreadBitmap);
1968	if (RT_SUCCESS(rc))
1969	{
1970	rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjVmwriteBitmap, &pVM->hm.s.vmx.pvVmwriteBitmap,
1971	&pVM->hm.s.vmx.HCPhysVmwriteBitmap);
1972	}
1973	}
1974	else
1975	rc = VERR_NO_MEMORY;
1976
1977	if (RT_FAILURE(rc))
1978	{
1979	hmR0VmxStructsFree(pVM);
1980	return rc;
1981	}
1982	}
1983	#endif
1984
1985	/*
1986	* Initialize per-VCPU VT-x structures.
1987	*/
1988	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
1989	{
1990	/* Allocate the guest VMCS structures. */
1991	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
1992	rc = hmR0VmxAllocVmcsInfo(pVCpu, &pVCpu->hm.s.vmx.VmcsInfo, false /* fIsNstGstVmcs */);
1993	if (RT_SUCCESS(rc))
1994	{
1995	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1996	/* Allocate the nested-guest VMCS structures, when the VMX feature is exposed to the guest. */
1997	if (pVM->cpum.ro.GuestFeatures.fVmx)
1998	{
1999	rc = hmR0VmxAllocVmcsInfo(pVCpu, &pVCpu->hm.s.vmx.VmcsInfoNstGst, true /* fIsNstGstVmcs */);
2000	if (RT_SUCCESS(rc))
2001	{ /* likely */ }
2002	else
2003	break;
2004	}
2005	#endif
2006	}
2007	else
2008	break;
2009	}
2010
2011	if (RT_FAILURE(rc))
2012	{
2013	hmR0VmxStructsFree(pVM);
2014	return rc;
2015	}
2016
2017	return VINF_SUCCESS;
2018	}
2019
2020	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2021	/**
2022	* Returns whether an MSR at the given MSR-bitmap offset is intercepted or not.
2023	*
2024	* @returns @c true if the MSR is intercepted, @c false otherwise.
2025	* @param pvMsrBitmap The MSR bitmap.
2026	* @param offMsr The MSR byte offset.
2027	* @param iBit The bit offset from the byte offset.
2028	*/
2029	DECLINLINE(bool) hmR0VmxIsMsrBitSet(const void *pvMsrBitmap, uint16_t offMsr, int32_t iBit)
2030	{
2031	uint8_t const * const pbMsrBitmap = (uint8_t const * const)pvMsrBitmap;
2032	Assert(pbMsrBitmap);
2033	Assert(offMsr + (iBit >> 3) <= X86_PAGE_4K_SIZE);
2034	return ASMBitTest(pbMsrBitmap + offMsr, iBit);
2035	}
2036	#endif
2037
2038	/**
2039	* Sets the permission bits for the specified MSR in the given MSR bitmap.
2040	*
2041	* If the passed VMCS is a nested-guest VMCS, this function ensures that the
2042	* read/write intercept is cleared from the MSR bitmap used for hardware-assisted
2043	* VMX execution of the nested-guest, only if nested-guest is also not intercepting
2044	* the read/write access of this MSR.
2045	*
2046	* @param pVCpu The cross context virtual CPU structure.
2047	* @param pVmcsInfo The VMCS info. object.
2048	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2049	* @param idMsr The MSR value.
2050	* @param fMsrpm The MSR permissions (see VMXMSRPM_XXX). This must
2051	* include both a read -and- a write permission!
2052	*
2053	* @sa CPUMGetVmxMsrPermission.
2054	* @remarks Can be called with interrupts disabled.
2055	*/
2056	static void hmR0VmxSetMsrPermission(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs, uint32_t idMsr, uint32_t fMsrpm)
2057	{
2058	uint8_t pbMsrBitmap = (uint8_t )pVmcsInfo->pvMsrBitmap;
2059	Assert(pbMsrBitmap);
2060	Assert(VMXMSRPM_IS_FLAG_VALID(fMsrpm));
2061
2062	/*
2063	* MSR-bitmap Layout:
2064	* Byte index MSR range Interpreted as
2065	* 0x000 - 0x3ff 0x00000000 - 0x00001fff Low MSR read bits.
2066	* 0x400 - 0x7ff 0xc0000000 - 0xc0001fff High MSR read bits.
2067	* 0x800 - 0xbff 0x00000000 - 0x00001fff Low MSR write bits.
2068	* 0xc00 - 0xfff 0xc0000000 - 0xc0001fff High MSR write bits.
2069	*
2070	* A bit corresponding to an MSR within the above range causes a VM-exit
2071	* if the bit is 1 on executions of RDMSR/WRMSR. If an MSR falls out of
2072	* the MSR range, it always cause a VM-exit.
2073	*
2074	* See Intel spec. 24.6.9 "MSR-Bitmap Address".
2075	*/
2076	uint16_t const offBitmapRead = 0;
2077	uint16_t const offBitmapWrite = 0x800;
2078	uint16_t offMsr;
2079	int32_t iBit;
2080	if (idMsr <= UINT32_C(0x00001fff))
2081	{
2082	offMsr = 0;
2083	iBit = idMsr;
2084	}
2085	else if (idMsr - UINT32_C(0xc0000000) <= UINT32_C(0x00001fff))
2086	{
2087	offMsr = 0x400;
2088	iBit = idMsr - UINT32_C(0xc0000000);
2089	}
2090	else
2091	AssertMsgFailedReturnVoid(("Invalid MSR %#RX32\n", idMsr));
2092
2093	/*
2094	* Set the MSR read permission.
2095	*/
2096	uint16_t const offMsrRead = offBitmapRead + offMsr;
2097	Assert(offMsrRead + (iBit >> 3) < offBitmapWrite);
2098	if (fMsrpm & VMXMSRPM_ALLOW_RD)
2099	{
2100	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2101	bool const fClear = !fIsNstGstVmcs ? true
2102	: !hmR0VmxIsMsrBitSet(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), offMsrRead, iBit);
2103	#else
2104	RT_NOREF2(pVCpu, fIsNstGstVmcs);
2105	bool const fClear = true;
2106	#endif
2107	if (fClear)
2108	ASMBitClear(pbMsrBitmap + offMsrRead, iBit);
2109	}
2110	else
2111	ASMBitSet(pbMsrBitmap + offMsrRead, iBit);
2112
2113	/*
2114	* Set the MSR write permission.
2115	*/
2116	uint16_t const offMsrWrite = offBitmapWrite + offMsr;
2117	Assert(offMsrWrite + (iBit >> 3) < X86_PAGE_4K_SIZE);
2118	if (fMsrpm & VMXMSRPM_ALLOW_WR)
2119	{
2120	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2121	bool const fClear = !fIsNstGstVmcs ? true
2122	: !hmR0VmxIsMsrBitSet(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), offMsrWrite, iBit);
2123	#else
2124	RT_NOREF2(pVCpu, fIsNstGstVmcs);
2125	bool const fClear = true;
2126	#endif
2127	if (fClear)
2128	ASMBitClear(pbMsrBitmap + offMsrWrite, iBit);
2129	}
2130	else
2131	ASMBitSet(pbMsrBitmap + offMsrWrite, iBit);
2132	}
2133
2134
2135	/**
2136	* Updates the VMCS with the number of effective MSRs in the auto-load/store MSR
2137	* area.
2138	*
2139	* @returns VBox status code.
2140	* @param pVCpu The cross context virtual CPU structure.
2141	* @param pVmcsInfo The VMCS info. object.
2142	* @param cMsrs The number of MSRs.
2143	*/
2144	static int hmR0VmxSetAutoLoadStoreMsrCount(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint32_t cMsrs)
2145	{
2146	/* Shouldn't ever happen but there -is- a number. We're well within the recommended 512. */
2147	uint32_t const cMaxSupportedMsrs = VMX_MISC_MAX_MSRS(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.u64Misc);
2148	if (RT_LIKELY(cMsrs < cMaxSupportedMsrs))
2149	{
2150	/* Commit the MSR counts to the VMCS and update the cache. */
2151	if (pVmcsInfo->cEntryMsrLoad != cMsrs)
2152	{
2153	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, cMsrs); AssertRC(rc);
2154	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, cMsrs); AssertRC(rc);
2155	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, cMsrs); AssertRC(rc);
2156	pVmcsInfo->cEntryMsrLoad = cMsrs;
2157	pVmcsInfo->cExitMsrStore = cMsrs;
2158	pVmcsInfo->cExitMsrLoad = cMsrs;
2159	}
2160	return VINF_SUCCESS;
2161	}
2162
2163	LogRel(("Auto-load/store MSR count exceeded! cMsrs=%u MaxSupported=%u\n", cMsrs, cMaxSupportedMsrs));
2164	pVCpu->hm.s.u32HMError = VMX_UFC_INSUFFICIENT_GUEST_MSR_STORAGE;
2165	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2166	}
2167
2168
2169	/**
2170	* Adds a new (or updates the value of an existing) guest/host MSR
2171	* pair to be swapped during the world-switch as part of the
2172	* auto-load/store MSR area in the VMCS.
2173	*
2174	* @returns VBox status code.
2175	* @param pVCpu The cross context virtual CPU structure.
2176	* @param pVmxTransient The VMX-transient structure.
2177	* @param idMsr The MSR.
2178	* @param uGuestMsrValue Value of the guest MSR.
2179	* @param fSetReadWrite Whether to set the guest read/write access of this
2180	* MSR (thus not causing a VM-exit).
2181	* @param fUpdateHostMsr Whether to update the value of the host MSR if
2182	* necessary.
2183	*/
2184	static int hmR0VmxAddAutoLoadStoreMsr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t idMsr, uint64_t uGuestMsrValue,
2185	bool fSetReadWrite, bool fUpdateHostMsr)
2186	{
2187	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
2188	bool const fIsNstGstVmcs = pVmxTransient->fIsNestedGuest;
2189	PVMXAUTOMSR pGuestMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2190	uint32_t cMsrs = pVmcsInfo->cEntryMsrLoad;
2191	uint32_t i;
2192
2193	/* Paranoia. */
2194	Assert(pGuestMsrLoad);
2195
2196	LogFlowFunc(("pVCpu=%p idMsr=%#RX32 uGestMsrValue=%#RX64\n", pVCpu, idMsr, uGuestMsrValue));
2197
2198	/* Check if the MSR already exists in the VM-entry MSR-load area. */
2199	for (i = 0; i < cMsrs; i++)
2200	{
2201	if (pGuestMsrLoad[i].u32Msr == idMsr)
2202	break;
2203	}
2204
2205	bool fAdded = false;
2206	if (i == cMsrs)
2207	{
2208	/* The MSR does not exist, bump the MSR count to make room for the new MSR. */
2209	++cMsrs;
2210	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, pVmcsInfo, cMsrs);
2211	AssertMsgRCReturn(rc, ("Insufficient space to add MSR to VM-entry MSR-load/store area %u\n", idMsr), rc);
2212
2213	/* Set the guest to read/write this MSR without causing VM-exits. */
2214	if ( fSetReadWrite
2215	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS))
2216	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, fIsNstGstVmcs, idMsr, VMXMSRPM_ALLOW_RD_WR);
2217
2218	Log4Func(("Added MSR %#RX32, cMsrs=%u\n", idMsr, cMsrs));
2219	fAdded = true;
2220	}
2221
2222	/* Update the MSR value for the newly added or already existing MSR. */
2223	pGuestMsrLoad[i].u32Msr = idMsr;
2224	pGuestMsrLoad[i].u64Value = uGuestMsrValue;
2225
2226	/* Create the corresponding slot in the VM-exit MSR-store area if we use a different page. */
2227	if (hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo))
2228	{
2229	PVMXAUTOMSR pGuestMsrStore = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2230	pGuestMsrStore[i].u32Msr = idMsr;
2231	pGuestMsrStore[i].u64Value = uGuestMsrValue;
2232	}
2233
2234	/* Update the corresponding slot in the host MSR area. */
2235	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2236	Assert(pHostMsr != pVmcsInfo->pvGuestMsrLoad);
2237	Assert(pHostMsr != pVmcsInfo->pvGuestMsrStore);
2238	pHostMsr[i].u32Msr = idMsr;
2239
2240	/*
2241	* Only if the caller requests to update the host MSR value AND we've newly added the
2242	* MSR to the host MSR area do we actually update the value. Otherwise, it will be
2243	* updated by hmR0VmxUpdateAutoLoadHostMsrs().
2244	*
2245	* We do this for performance reasons since reading MSRs may be quite expensive.
2246	*/
2247	if (fAdded)
2248	{
2249	if (fUpdateHostMsr)
2250	{
2251	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2252	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2253	pHostMsr[i].u64Value = ASMRdMsr(idMsr);
2254	}
2255	else
2256	{
2257	/* Someone else can do the work. */
2258	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = false;
2259	}
2260	}
2261	return VINF_SUCCESS;
2262	}
2263
2264
2265	/**
2266	* Removes a guest/host MSR pair to be swapped during the world-switch from the
2267	* auto-load/store MSR area in the VMCS.
2268	*
2269	* @returns VBox status code.
2270	* @param pVCpu The cross context virtual CPU structure.
2271	* @param pVmxTransient The VMX-transient structure.
2272	* @param idMsr The MSR.
2273	*/
2274	static int hmR0VmxRemoveAutoLoadStoreMsr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t idMsr)
2275	{
2276	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
2277	bool const fIsNstGstVmcs = pVmxTransient->fIsNestedGuest;
2278	PVMXAUTOMSR pGuestMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2279	uint32_t cMsrs = pVmcsInfo->cEntryMsrLoad;
2280
2281	LogFlowFunc(("pVCpu=%p idMsr=%#RX32\n", pVCpu, idMsr));
2282
2283	for (uint32_t i = 0; i < cMsrs; i++)
2284	{
2285	/* Find the MSR. */
2286	if (pGuestMsrLoad[i].u32Msr == idMsr)
2287	{
2288	/*
2289	* If it's the last MSR, we only need to reduce the MSR count.
2290	* If it's -not- the last MSR, copy the last MSR in place of it and reduce the MSR count.
2291	*/
2292	if (i < cMsrs - 1)
2293	{
2294	/* Remove it from the VM-entry MSR-load area. */
2295	pGuestMsrLoad[i].u32Msr = pGuestMsrLoad[cMsrs - 1].u32Msr;
2296	pGuestMsrLoad[i].u64Value = pGuestMsrLoad[cMsrs - 1].u64Value;
2297
2298	/* Remove it from the VM-exit MSR-store area if it's in a different page. */
2299	if (hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo))
2300	{
2301	PVMXAUTOMSR pGuestMsrStore = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2302	Assert(pGuestMsrStore[i].u32Msr == idMsr);
2303	pGuestMsrStore[i].u32Msr = pGuestMsrStore[cMsrs - 1].u32Msr;
2304	pGuestMsrStore[i].u64Value = pGuestMsrStore[cMsrs - 1].u64Value;
2305	}
2306
2307	/* Remove it from the VM-exit MSR-load area. */
2308	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2309	Assert(pHostMsr[i].u32Msr == idMsr);
2310	pHostMsr[i].u32Msr = pHostMsr[cMsrs - 1].u32Msr;
2311	pHostMsr[i].u64Value = pHostMsr[cMsrs - 1].u64Value;
2312	}
2313
2314	/* Reduce the count to reflect the removed MSR and bail. */
2315	--cMsrs;
2316	break;
2317	}
2318	}
2319
2320	/* Update the VMCS if the count changed (meaning the MSR was found and removed). */
2321	if (cMsrs != pVmcsInfo->cEntryMsrLoad)
2322	{
2323	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, pVmcsInfo, cMsrs);
2324	AssertRCReturn(rc, rc);
2325
2326	/* We're no longer swapping MSRs during the world-switch, intercept guest read/writes to them. */
2327	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
2328	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, fIsNstGstVmcs, idMsr, VMXMSRPM_EXIT_RD \| VMXMSRPM_EXIT_WR);
2329
2330	Log4Func(("Removed MSR %#RX32, cMsrs=%u\n", idMsr, cMsrs));
2331	return VINF_SUCCESS;
2332	}
2333
2334	return VERR_NOT_FOUND;
2335	}
2336
2337
2338	/**
2339	* Checks if the specified guest MSR is part of the VM-entry MSR-load area.
2340	*
2341	* @returns @c true if found, @c false otherwise.
2342	* @param pVmcsInfo The VMCS info. object.
2343	* @param idMsr The MSR to find.
2344	*/
2345	static bool hmR0VmxIsAutoLoadGuestMsr(PCVMXVMCSINFO pVmcsInfo, uint32_t idMsr)
2346	{
2347	PCVMXAUTOMSR pMsrs = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2348	uint32_t const cMsrs = pVmcsInfo->cEntryMsrLoad;
2349	Assert(pMsrs);
2350	Assert(sizeof(pMsrs) cMsrs <= X86_PAGE_4K_SIZE);
2351	for (uint32_t i = 0; i < cMsrs; i++)
2352	{
2353	if (pMsrs[i].u32Msr == idMsr)
2354	return true;
2355	}
2356	return false;
2357	}
2358
2359
2360	/**
2361	* Updates the value of all host MSRs in the VM-exit MSR-load area.
2362	*
2363	* @param pVCpu The cross context virtual CPU structure.
2364	* @param pVmcsInfo The VMCS info. object.
2365	*
2366	* @remarks No-long-jump zone!!!
2367	*/
2368	static void hmR0VmxUpdateAutoLoadHostMsrs(PCVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
2369	{
2370	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2371
2372	PVMXAUTOMSR pHostMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2373	uint32_t const cMsrs = pVmcsInfo->cExitMsrLoad;
2374	Assert(pHostMsrLoad);
2375	Assert(sizeof(pHostMsrLoad) cMsrs <= X86_PAGE_4K_SIZE);
2376	LogFlowFunc(("pVCpu=%p cMsrs=%u\n", pVCpu, cMsrs));
2377	for (uint32_t i = 0; i < cMsrs; i++)
2378	{
2379	/*
2380	* Performance hack for the host EFER MSR. We use the cached value rather than re-read it.
2381	* Strict builds will catch mismatches in hmR0VmxCheckAutoLoadStoreMsrs(). See @bugref{7368}.
2382	*/
2383	if (pHostMsrLoad[i].u32Msr == MSR_K6_EFER)
2384	pHostMsrLoad[i].u64Value = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostMsrEfer;
2385	else
2386	pHostMsrLoad[i].u64Value = ASMRdMsr(pHostMsrLoad[i].u32Msr);
2387	}
2388	}
2389
2390
2391	/**
2392	* Saves a set of host MSRs to allow read/write passthru access to the guest and
2393	* perform lazy restoration of the host MSRs while leaving VT-x.
2394	*
2395	* @param pVCpu The cross context virtual CPU structure.
2396	*
2397	* @remarks No-long-jump zone!!!
2398	*/
2399	static void hmR0VmxLazySaveHostMsrs(PVMCPUCC pVCpu)
2400	{
2401	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2402
2403	/*
2404	* Note: If you're adding MSRs here, make sure to update the MSR-bitmap accesses in hmR0VmxSetupVmcsProcCtls().
2405	*/
2406	if (!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST))
2407	{
2408	Assert(!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)); /* Guest MSRs better not be loaded now. */
2409	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
2410	{
2411	pVCpu->hm.s.vmx.u64HostMsrLStar = ASMRdMsr(MSR_K8_LSTAR);
2412	pVCpu->hm.s.vmx.u64HostMsrStar = ASMRdMsr(MSR_K6_STAR);
2413	pVCpu->hm.s.vmx.u64HostMsrSfMask = ASMRdMsr(MSR_K8_SF_MASK);
2414	pVCpu->hm.s.vmx.u64HostMsrKernelGsBase = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
2415	}
2416	pVCpu->hm.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_SAVED_HOST;
2417	}
2418	}
2419
2420
2421	/**
2422	* Checks whether the MSR belongs to the set of guest MSRs that we restore
2423	* lazily while leaving VT-x.
2424	*
2425	* @returns true if it does, false otherwise.
2426	* @param pVCpu The cross context virtual CPU structure.
2427	* @param idMsr The MSR to check.
2428	*/
2429	static bool hmR0VmxIsLazyGuestMsr(PCVMCPUCC pVCpu, uint32_t idMsr)
2430	{
2431	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
2432	{
2433	switch (idMsr)
2434	{
2435	case MSR_K8_LSTAR:
2436	case MSR_K6_STAR:
2437	case MSR_K8_SF_MASK:
2438	case MSR_K8_KERNEL_GS_BASE:
2439	return true;
2440	}
2441	}
2442	return false;
2443	}
2444
2445
2446	/**
2447	* Loads a set of guests MSRs to allow read/passthru to the guest.
2448	*
2449	* The name of this function is slightly confusing. This function does NOT
2450	* postpone loading, but loads the MSR right now. "hmR0VmxLazy" is simply a
2451	* common prefix for functions dealing with "lazy restoration" of the shared
2452	* MSRs.
2453	*
2454	* @param pVCpu The cross context virtual CPU structure.
2455	*
2456	* @remarks No-long-jump zone!!!
2457	*/
2458	static void hmR0VmxLazyLoadGuestMsrs(PVMCPUCC pVCpu)
2459	{
2460	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2461	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2462
2463	Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
2464	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
2465	{
2466	/*
2467	* If the guest MSRs are not loaded -and- if all the guest MSRs are identical
2468	* to the MSRs on the CPU (which are the saved host MSRs, see assertion above) then
2469	* we can skip a few MSR writes.
2470	*
2471	* Otherwise, it implies either 1. they're not loaded, or 2. they're loaded but the
2472	* guest MSR values in the guest-CPU context might be different to what's currently
2473	* loaded in the CPU. In either case, we need to write the new guest MSR values to the
2474	* CPU, see @bugref{8728}.
2475	*/
2476	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
2477	if ( !(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
2478	&& pCtx->msrKERNELGSBASE == pVCpu->hm.s.vmx.u64HostMsrKernelGsBase
2479	&& pCtx->msrLSTAR == pVCpu->hm.s.vmx.u64HostMsrLStar
2480	&& pCtx->msrSTAR == pVCpu->hm.s.vmx.u64HostMsrStar
2481	&& pCtx->msrSFMASK == pVCpu->hm.s.vmx.u64HostMsrSfMask)
2482	{
2483	#ifdef VBOX_STRICT
2484	Assert(ASMRdMsr(MSR_K8_KERNEL_GS_BASE) == pCtx->msrKERNELGSBASE);
2485	Assert(ASMRdMsr(MSR_K8_LSTAR) == pCtx->msrLSTAR);
2486	Assert(ASMRdMsr(MSR_K6_STAR) == pCtx->msrSTAR);
2487	Assert(ASMRdMsr(MSR_K8_SF_MASK) == pCtx->msrSFMASK);
2488	#endif
2489	}
2490	else
2491	{
2492	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pCtx->msrKERNELGSBASE);
2493	ASMWrMsr(MSR_K8_LSTAR, pCtx->msrLSTAR);
2494	ASMWrMsr(MSR_K6_STAR, pCtx->msrSTAR);
2495	ASMWrMsr(MSR_K8_SF_MASK, pCtx->msrSFMASK);
2496	}
2497	}
2498	pVCpu->hm.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_LOADED_GUEST;
2499	}
2500
2501
2502	/**
2503	* Performs lazy restoration of the set of host MSRs if they were previously
2504	* loaded with guest MSR values.
2505	*
2506	* @param pVCpu The cross context virtual CPU structure.
2507	*
2508	* @remarks No-long-jump zone!!!
2509	* @remarks The guest MSRs should have been saved back into the guest-CPU
2510	* context by hmR0VmxImportGuestState()!!!
2511	*/
2512	static void hmR0VmxLazyRestoreHostMsrs(PVMCPUCC pVCpu)
2513	{
2514	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2515	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2516
2517	if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
2518	{
2519	Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
2520	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
2521	{
2522	ASMWrMsr(MSR_K8_LSTAR, pVCpu->hm.s.vmx.u64HostMsrLStar);
2523	ASMWrMsr(MSR_K6_STAR, pVCpu->hm.s.vmx.u64HostMsrStar);
2524	ASMWrMsr(MSR_K8_SF_MASK, pVCpu->hm.s.vmx.u64HostMsrSfMask);
2525	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pVCpu->hm.s.vmx.u64HostMsrKernelGsBase);
2526	}
2527	}
2528	pVCpu->hm.s.vmx.fLazyMsrs &= ~(VMX_LAZY_MSRS_LOADED_GUEST \| VMX_LAZY_MSRS_SAVED_HOST);
2529	}
2530
2531
2532	/**
2533	* Verifies that our cached values of the VMCS fields are all consistent with
2534	* what's actually present in the VMCS.
2535	*
2536	* @returns VBox status code.
2537	* @retval VINF_SUCCESS if all our caches match their respective VMCS fields.
2538	* @retval VERR_VMX_VMCS_FIELD_CACHE_INVALID if a cache field doesn't match the
2539	* VMCS content. HMCPU error-field is
2540	* updated, see VMX_VCI_XXX.
2541	* @param pVCpu The cross context virtual CPU structure.
2542	* @param pVmcsInfo The VMCS info. object.
2543	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2544	*/
2545	static int hmR0VmxCheckVmcsCtls(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
2546	{
2547	const char * const pcszVmcs = fIsNstGstVmcs ? "Nested-guest VMCS" : "VMCS";
2548
2549	uint32_t u32Val;
2550	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
2551	AssertRC(rc);
2552	AssertMsgReturnStmt(pVmcsInfo->u32EntryCtls == u32Val,
2553	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32EntryCtls, u32Val),
2554	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_ENTRY,
2555	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2556
2557	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val);
2558	AssertRC(rc);
2559	AssertMsgReturnStmt(pVmcsInfo->u32ExitCtls == u32Val,
2560	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ExitCtls, u32Val),
2561	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_EXIT,
2562	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2563
2564	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val);
2565	AssertRC(rc);
2566	AssertMsgReturnStmt(pVmcsInfo->u32PinCtls == u32Val,
2567	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32PinCtls, u32Val),
2568	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PIN_EXEC,
2569	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2570
2571	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val);
2572	AssertRC(rc);
2573	AssertMsgReturnStmt(pVmcsInfo->u32ProcCtls == u32Val,
2574	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ProcCtls, u32Val),
2575	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC,
2576	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2577
2578	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
2579	{
2580	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val);
2581	AssertRC(rc);
2582	AssertMsgReturnStmt(pVmcsInfo->u32ProcCtls2 == u32Val,
2583	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ProcCtls2, u32Val),
2584	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC2,
2585	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2586	}
2587
2588	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, &u32Val);
2589	AssertRC(rc);
2590	AssertMsgReturnStmt(pVmcsInfo->u32XcptBitmap == u32Val,
2591	("%s exception bitmap mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32XcptBitmap, u32Val),
2592	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_XCPT_BITMAP,
2593	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2594
2595	uint64_t u64Val;
2596	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, &u64Val);
2597	AssertRC(rc);
2598	AssertMsgReturnStmt(pVmcsInfo->u64TscOffset == u64Val,
2599	("%s TSC offset mismatch: Cache=%#RX64 VMCS=%#RX64\n", pcszVmcs, pVmcsInfo->u64TscOffset, u64Val),
2600	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_TSC_OFFSET,
2601	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2602
2603	NOREF(pcszVmcs);
2604	return VINF_SUCCESS;
2605	}
2606
2607
2608	#ifdef VBOX_STRICT
2609	/**
2610	* Verifies that our cached host EFER MSR value has not changed since we cached it.
2611	*
2612	* @param pVCpu The cross context virtual CPU structure.
2613	* @param pVmcsInfo The VMCS info. object.
2614	*/
2615	static void hmR0VmxCheckHostEferMsr(PCVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
2616	{
2617	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2618
2619	if (pVmcsInfo->u32ExitCtls & VMX_EXIT_CTLS_LOAD_EFER_MSR)
2620	{
2621	uint64_t const uHostEferMsr = ASMRdMsr(MSR_K6_EFER);
2622	uint64_t const uHostEferMsrCache = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostMsrEfer;
2623	uint64_t uVmcsEferMsrVmcs;
2624	int rc = VMXReadVmcs64(VMX_VMCS64_HOST_EFER_FULL, &uVmcsEferMsrVmcs);
2625	AssertRC(rc);
2626
2627	AssertMsgReturnVoid(uHostEferMsr == uVmcsEferMsrVmcs,
2628	("EFER Host/VMCS mismatch! host=%#RX64 vmcs=%#RX64\n", uHostEferMsr, uVmcsEferMsrVmcs));
2629	AssertMsgReturnVoid(uHostEferMsr == uHostEferMsrCache,
2630	("EFER Host/Cache mismatch! host=%#RX64 cache=%#RX64\n", uHostEferMsr, uHostEferMsrCache));
2631	}
2632	}
2633
2634
2635	/**
2636	* Verifies whether the guest/host MSR pairs in the auto-load/store area in the
2637	* VMCS are correct.
2638	*
2639	* @param pVCpu The cross context virtual CPU structure.
2640	* @param pVmcsInfo The VMCS info. object.
2641	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2642	*/
2643	static void hmR0VmxCheckAutoLoadStoreMsrs(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
2644	{
2645	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2646
2647	/* Read the various MSR-area counts from the VMCS. */
2648	uint32_t cEntryLoadMsrs;
2649	uint32_t cExitStoreMsrs;
2650	uint32_t cExitLoadMsrs;
2651	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &cEntryLoadMsrs); AssertRC(rc);
2652	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &cExitStoreMsrs); AssertRC(rc);
2653	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &cExitLoadMsrs); AssertRC(rc);
2654
2655	/* Verify all the MSR counts are the same. */
2656	Assert(cEntryLoadMsrs == cExitStoreMsrs);
2657	Assert(cExitStoreMsrs == cExitLoadMsrs);
2658	uint32_t const cMsrs = cExitLoadMsrs;
2659
2660	/* Verify the MSR counts do not exceed the maximum count supported by the hardware. */
2661	Assert(cMsrs < VMX_MISC_MAX_MSRS(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.u64Misc));
2662
2663	/* Verify the MSR counts are within the allocated page size. */
2664	Assert(sizeof(VMXAUTOMSR) * cMsrs <= X86_PAGE_4K_SIZE);
2665
2666	/* Verify the relevant contents of the MSR areas match. */
2667	PCVMXAUTOMSR pGuestMsrLoad = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2668	PCVMXAUTOMSR pGuestMsrStore = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2669	PCVMXAUTOMSR pHostMsrLoad = (PCVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2670	bool const fSeparateExitMsrStorePage = hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo);
2671	for (uint32_t i = 0; i < cMsrs; i++)
2672	{
2673	/* Verify that the MSRs are paired properly and that the host MSR has the correct value. */
2674	if (fSeparateExitMsrStorePage)
2675	{
2676	AssertMsgReturnVoid(pGuestMsrLoad->u32Msr == pGuestMsrStore->u32Msr,
2677	("GuestMsrLoad=%#RX32 GuestMsrStore=%#RX32 cMsrs=%u\n",
2678	pGuestMsrLoad->u32Msr, pGuestMsrStore->u32Msr, cMsrs));
2679	}
2680
2681	AssertMsgReturnVoid(pHostMsrLoad->u32Msr == pGuestMsrLoad->u32Msr,
2682	("HostMsrLoad=%#RX32 GuestMsrLoad=%#RX32 cMsrs=%u\n",
2683	pHostMsrLoad->u32Msr, pGuestMsrLoad->u32Msr, cMsrs));
2684
2685	uint64_t const u64Msr = ASMRdMsr(pHostMsrLoad->u32Msr);
2686	AssertMsgReturnVoid(pHostMsrLoad->u64Value == u64Msr,
2687	("u32Msr=%#RX32 VMCS Value=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
2688	pHostMsrLoad->u32Msr, pHostMsrLoad->u64Value, u64Msr, cMsrs));
2689
2690	/* Verify that cached host EFER MSR matches what's loaded the CPU. */
2691	bool const fIsEferMsr = RT_BOOL(pHostMsrLoad->u32Msr == MSR_K6_EFER);
2692	if (fIsEferMsr)
2693	{
2694	AssertMsgReturnVoid(u64Msr == pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostMsrEfer,
2695	("Cached=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
2696	pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostMsrEfer, u64Msr, cMsrs));
2697	}
2698
2699	/* Verify that the accesses are as expected in the MSR bitmap for auto-load/store MSRs. */
2700	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
2701	{
2702	uint32_t const fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, pGuestMsrLoad->u32Msr);
2703	if (fIsEferMsr)
2704	{
2705	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_EXIT_RD), ("Passthru read for EFER MSR!?\n"));
2706	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_EXIT_WR), ("Passthru write for EFER MSR!?\n"));
2707	}
2708	else
2709	{
2710	if (!fIsNstGstVmcs)
2711	{
2712	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_ALLOW_RD_WR) == VMXMSRPM_ALLOW_RD_WR,
2713	("u32Msr=%#RX32 cMsrs=%u No passthru read/write!\n", pGuestMsrLoad->u32Msr, cMsrs));
2714	}
2715	else
2716	{
2717	/*
2718	* A nested-guest VMCS must -also- allow read/write passthrough for the MSR for us to
2719	* execute a nested-guest with MSR passthrough.
2720	*
2721	* Check if the nested-guest MSR bitmap allows passthrough, and if so, assert that we
2722	* allow passthrough too.
2723	*/
2724	void const *pvMsrBitmapNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap);
2725	Assert(pvMsrBitmapNstGst);
2726	uint32_t const fMsrpmNstGst = CPUMGetVmxMsrPermission(pvMsrBitmapNstGst, pGuestMsrLoad->u32Msr);
2727	AssertMsgReturnVoid(fMsrpm == fMsrpmNstGst,
2728	("u32Msr=%#RX32 cMsrs=%u Permission mismatch fMsrpm=%#x fMsrpmNstGst=%#x!\n",
2729	pGuestMsrLoad->u32Msr, cMsrs, fMsrpm, fMsrpmNstGst));
2730	}
2731	}
2732	}
2733
2734	/* Move to the next MSR. */
2735	pHostMsrLoad++;
2736	pGuestMsrLoad++;
2737	pGuestMsrStore++;
2738	}
2739	}
2740	#endif /* VBOX_STRICT */
2741
2742
2743	/**
2744	* Flushes the TLB using EPT.
2745	*
2746	* @returns VBox status code.
2747	* @param pVCpu The cross context virtual CPU structure of the calling
2748	* EMT. Can be NULL depending on @a enmTlbFlush.
2749	* @param pVmcsInfo The VMCS info. object. Can be NULL depending on @a
2750	* enmTlbFlush.
2751	* @param enmTlbFlush Type of flush.
2752	*
2753	* @remarks Caller is responsible for making sure this function is called only
2754	* when NestedPaging is supported and providing @a enmTlbFlush that is
2755	* supported by the CPU.
2756	* @remarks Can be called with interrupts disabled.
2757	*/
2758	static void hmR0VmxFlushEpt(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, VMXTLBFLUSHEPT enmTlbFlush)
2759	{
2760	uint64_t au64Descriptor[2];
2761	if (enmTlbFlush == VMXTLBFLUSHEPT_ALL_CONTEXTS)
2762	au64Descriptor[0] = 0;
2763	else
2764	{
2765	Assert(pVCpu);
2766	Assert(pVmcsInfo);
2767	au64Descriptor[0] = pVmcsInfo->HCPhysEPTP;
2768	}
2769	au64Descriptor[1] = 0; /* MBZ. Intel spec. 33.3 "VMX Instructions" */
2770
2771	int rc = VMXR0InvEPT(enmTlbFlush, &au64Descriptor[0]);
2772	AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvEPT %#x %#RHp failed. rc=%Rrc\n", enmTlbFlush, au64Descriptor[0], rc));
2773
2774	if ( RT_SUCCESS(rc)
2775	&& pVCpu)
2776	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushNestedPaging);
2777	}
2778
2779
2780	/**
2781	* Flushes the TLB using VPID.
2782	*
2783	* @returns VBox status code.
2784	* @param pVCpu The cross context virtual CPU structure of the calling
2785	* EMT. Can be NULL depending on @a enmTlbFlush.
2786	* @param enmTlbFlush Type of flush.
2787	* @param GCPtr Virtual address of the page to flush (can be 0 depending
2788	* on @a enmTlbFlush).
2789	*
2790	* @remarks Can be called with interrupts disabled.
2791	*/
2792	static void hmR0VmxFlushVpid(PVMCPUCC pVCpu, VMXTLBFLUSHVPID enmTlbFlush, RTGCPTR GCPtr)
2793	{
2794	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fVpid);
2795
2796	uint64_t au64Descriptor[2];
2797	if (enmTlbFlush == VMXTLBFLUSHVPID_ALL_CONTEXTS)
2798	{
2799	au64Descriptor[0] = 0;
2800	au64Descriptor[1] = 0;
2801	}
2802	else
2803	{
2804	AssertPtr(pVCpu);
2805	AssertMsg(pVCpu->hm.s.uCurrentAsid != 0, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
2806	AssertMsg(pVCpu->hm.s.uCurrentAsid <= UINT16_MAX, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
2807	au64Descriptor[0] = pVCpu->hm.s.uCurrentAsid;
2808	au64Descriptor[1] = GCPtr;
2809	}
2810
2811	int rc = VMXR0InvVPID(enmTlbFlush, &au64Descriptor[0]);
2812	AssertMsg(rc == VINF_SUCCESS,
2813	("VMXR0InvVPID %#x %u %RGv failed with %Rrc\n", enmTlbFlush, pVCpu ? pVCpu->hm.s.uCurrentAsid : 0, GCPtr, rc));
2814
2815	if ( RT_SUCCESS(rc)
2816	&& pVCpu)
2817	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid);
2818	NOREF(rc);
2819	}
2820
2821
2822	/**
2823	* Invalidates a guest page by guest virtual address. Only relevant for EPT/VPID,
2824	* otherwise there is nothing really to invalidate.
2825	*
2826	* @returns VBox status code.
2827	* @param pVCpu The cross context virtual CPU structure.
2828	* @param GCVirt Guest virtual address of the page to invalidate.
2829	*/
2830	VMMR0DECL(int) VMXR0InvalidatePage(PVMCPUCC pVCpu, RTGCPTR GCVirt)
2831	{
2832	AssertPtr(pVCpu);
2833	LogFlowFunc(("pVCpu=%p GCVirt=%RGv\n", pVCpu, GCVirt));
2834
2835	if (!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TLB_FLUSH))
2836	{
2837	/*
2838	* We must invalidate the guest TLB entry in either case, we cannot ignore it even for
2839	* the EPT case. See @bugref{6043} and @bugref{6177}.
2840	*
2841	* Set the VMCPU_FF_TLB_FLUSH force flag and flush before VM-entry in hmR0VmxFlushTLB*()
2842	* as this function maybe called in a loop with individual addresses.
2843	*/
2844	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2845	if (pVM->hm.s.vmx.fVpid)
2846	{
2847	bool fVpidFlush = RT_BOOL(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR);
2848	if (fVpidFlush)
2849	{
2850	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_INDIV_ADDR, GCVirt);
2851	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt);
2852	}
2853	else
2854	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2855	}
2856	else if (pVM->hm.s.fNestedPaging)
2857	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2858	}
2859
2860	return VINF_SUCCESS;
2861	}
2862
2863
2864	/**
2865	* Dummy placeholder for tagged-TLB flush handling before VM-entry. Used in the
2866	* case where neither EPT nor VPID is supported by the CPU.
2867	*
2868	* @param pHostCpu The HM physical-CPU structure.
2869	* @param pVCpu The cross context virtual CPU structure.
2870	*
2871	* @remarks Called with interrupts disabled.
2872	*/
2873	static void hmR0VmxFlushTaggedTlbNone(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu)
2874	{
2875	AssertPtr(pVCpu);
2876	AssertPtr(pHostCpu);
2877
2878	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
2879
2880	Assert(pHostCpu->idCpu != NIL_RTCPUID);
2881	pVCpu->hm.s.idLastCpu = pHostCpu->idCpu;
2882	pVCpu->hm.s.cTlbFlushes = pHostCpu->cTlbFlushes;
2883	pVCpu->hm.s.fForceTLBFlush = false;
2884	return;
2885	}
2886
2887
2888	/**
2889	* Flushes the tagged-TLB entries for EPT+VPID CPUs as necessary.
2890	*
2891	* @param pHostCpu The HM physical-CPU structure.
2892	* @param pVCpu The cross context virtual CPU structure.
2893	* @param pVmcsInfo The VMCS info. object.
2894	*
2895	* @remarks All references to "ASID" in this function pertains to "VPID" in Intel's
2896	* nomenclature. The reason is, to avoid confusion in compare statements
2897	* since the host-CPU copies are named "ASID".
2898	*
2899	* @remarks Called with interrupts disabled.
2900	*/
2901	static void hmR0VmxFlushTaggedTlbBoth(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
2902	{
2903	#ifdef VBOX_WITH_STATISTICS
2904	bool fTlbFlushed = false;
2905	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { fTlbFlushed = true; } while (0)
2906	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { \
2907	if (!fTlbFlushed) \
2908	STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); \
2909	} while (0)
2910	#else
2911	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { } while (0)
2912	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { } while (0)
2913	#endif
2914
2915	AssertPtr(pVCpu);
2916	AssertPtr(pHostCpu);
2917	Assert(pHostCpu->idCpu != NIL_RTCPUID);
2918
2919	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2920	AssertMsg(pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid,
2921	("hmR0VmxFlushTaggedTlbBoth cannot be invoked unless NestedPaging & VPID are enabled."
2922	"fNestedPaging=%RTbool fVpid=%RTbool", pVM->hm.s.fNestedPaging, pVM->hm.s.vmx.fVpid));
2923
2924	/*
2925	* Force a TLB flush for the first world-switch if the current CPU differs from the one we
2926	* ran on last. If the TLB flush count changed, another VM (VCPU rather) has hit the ASID
2927	* limit while flushing the TLB or the host CPU is online after a suspend/resume, so we
2928	* cannot reuse the current ASID anymore.
2929	*/
2930	if ( pVCpu->hm.s.idLastCpu != pHostCpu->idCpu
2931	\|\| pVCpu->hm.s.cTlbFlushes != pHostCpu->cTlbFlushes)
2932	{
2933	++pHostCpu->uCurrentAsid;
2934	if (pHostCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
2935	{
2936	pHostCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0. */
2937	pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
2938	pHostCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
2939	}
2940
2941	pVCpu->hm.s.uCurrentAsid = pHostCpu->uCurrentAsid;
2942	pVCpu->hm.s.idLastCpu = pHostCpu->idCpu;
2943	pVCpu->hm.s.cTlbFlushes = pHostCpu->cTlbFlushes;
2944
2945	/*
2946	* Flush by EPT when we get rescheduled to a new host CPU to ensure EPT-only tagged mappings are also
2947	* invalidated. We don't need to flush-by-VPID here as flushing by EPT covers it. See @bugref{6568}.
2948	*/
2949	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hm.s.vmx.enmTlbFlushEpt);
2950	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
2951	HMVMX_SET_TAGGED_TLB_FLUSHED();
2952	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
2953	}
2954	else if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH)) /* Check for explicit TLB flushes. */
2955	{
2956	/*
2957	* Changes to the EPT paging structure by VMM requires flushing-by-EPT as the CPU
2958	* creates guest-physical (ie. only EPT-tagged) mappings while traversing the EPT
2959	* tables when EPT is in use. Flushing-by-VPID will only flush linear (only
2960	* VPID-tagged) and combined (EPT+VPID tagged) mappings but not guest-physical
2961	* mappings, see @bugref{6568}.
2962	*
2963	* See Intel spec. 28.3.2 "Creating and Using Cached Translation Information".
2964	*/
2965	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hm.s.vmx.enmTlbFlushEpt);
2966	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
2967	HMVMX_SET_TAGGED_TLB_FLUSHED();
2968	}
2969	else if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
2970	{
2971	/*
2972	* The nested-guest specifies its own guest-physical address to use as the APIC-access
2973	* address which requires flushing the TLB of EPT cached structures.
2974	*
2975	* See Intel spec. 28.3.3.4 "Guidelines for Use of the INVEPT Instruction".
2976	*/
2977	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hm.s.vmx.enmTlbFlushEpt);
2978	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
2979	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
2980	HMVMX_SET_TAGGED_TLB_FLUSHED();
2981	}
2982
2983
2984	pVCpu->hm.s.fForceTLBFlush = false;
2985	HMVMX_UPDATE_FLUSH_SKIPPED_STAT();
2986
2987	Assert(pVCpu->hm.s.idLastCpu == pHostCpu->idCpu);
2988	Assert(pVCpu->hm.s.cTlbFlushes == pHostCpu->cTlbFlushes);
2989	AssertMsg(pVCpu->hm.s.cTlbFlushes == pHostCpu->cTlbFlushes,
2990	("Flush count mismatch for cpu %d (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pHostCpu->cTlbFlushes));
2991	AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
2992	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pHostCpu->idCpu,
2993	pHostCpu->uCurrentAsid, pHostCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
2994	AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
2995	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pHostCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
2996
2997	/* Update VMCS with the VPID. */
2998	int rc = VMXWriteVmcs16(VMX_VMCS16_VPID, pVCpu->hm.s.uCurrentAsid);
2999	AssertRC(rc);
3000
3001	#undef HMVMX_SET_TAGGED_TLB_FLUSHED
3002	}
3003
3004
3005	/**
3006	* Flushes the tagged-TLB entries for EPT CPUs as necessary.
3007	*
3008	* @param pHostCpu The HM physical-CPU structure.
3009	* @param pVCpu The cross context virtual CPU structure.
3010	* @param pVmcsInfo The VMCS info. object.
3011	*
3012	* @remarks Called with interrupts disabled.
3013	*/
3014	static void hmR0VmxFlushTaggedTlbEpt(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
3015	{
3016	AssertPtr(pVCpu);
3017	AssertPtr(pHostCpu);
3018	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3019	AssertMsg(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked without NestedPaging."));
3020	AssertMsg(!pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fVpid, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with VPID."));
3021
3022	/*
3023	* Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
3024	* A change in the TLB flush count implies the host CPU is online after a suspend/resume.
3025	*/
3026	if ( pVCpu->hm.s.idLastCpu != pHostCpu->idCpu
3027	\|\| pVCpu->hm.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3028	{
3029	pVCpu->hm.s.fForceTLBFlush = true;
3030	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3031	}
3032
3033	/* Check for explicit TLB flushes. */
3034	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
3035	{
3036	pVCpu->hm.s.fForceTLBFlush = true;
3037	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3038	}
3039
3040	/* Check for TLB flushes while switching to/from a nested-guest. */
3041	if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3042	{
3043	pVCpu->hm.s.fForceTLBFlush = true;
3044	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3045	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3046	}
3047
3048	pVCpu->hm.s.idLastCpu = pHostCpu->idCpu;
3049	pVCpu->hm.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3050
3051	if (pVCpu->hm.s.fForceTLBFlush)
3052	{
3053	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVCpu->CTX_SUFF(pVM)->hm.s.vmx.enmTlbFlushEpt);
3054	pVCpu->hm.s.fForceTLBFlush = false;
3055	}
3056	}
3057
3058
3059	/**
3060	* Flushes the tagged-TLB entries for VPID CPUs as necessary.
3061	*
3062	* @param pHostCpu The HM physical-CPU structure.
3063	* @param pVCpu The cross context virtual CPU structure.
3064	*
3065	* @remarks Called with interrupts disabled.
3066	*/
3067	static void hmR0VmxFlushTaggedTlbVpid(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu)
3068	{
3069	AssertPtr(pVCpu);
3070	AssertPtr(pHostCpu);
3071	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3072	AssertMsg(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fVpid, ("hmR0VmxFlushTlbVpid cannot be invoked without VPID."));
3073	AssertMsg(!pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging, ("hmR0VmxFlushTlbVpid cannot be invoked with NestedPaging"));
3074
3075	/*
3076	* Force a TLB flush for the first world switch if the current CPU differs from the one we
3077	* ran on last. If the TLB flush count changed, another VM (VCPU rather) has hit the ASID
3078	* limit while flushing the TLB or the host CPU is online after a suspend/resume, so we
3079	* cannot reuse the current ASID anymore.
3080	*/
3081	if ( pVCpu->hm.s.idLastCpu != pHostCpu->idCpu
3082	\|\| pVCpu->hm.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3083	{
3084	pVCpu->hm.s.fForceTLBFlush = true;
3085	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3086	}
3087
3088	/* Check for explicit TLB flushes. */
3089	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
3090	{
3091	/*
3092	* If we ever support VPID flush combinations other than ALL or SINGLE-context (see
3093	* hmR0VmxSetupTaggedTlb()) we would need to explicitly flush in this case (add an
3094	* fExplicitFlush = true here and change the pHostCpu->fFlushAsidBeforeUse check below to
3095	* include fExplicitFlush's too) - an obscure corner case.
3096	*/
3097	pVCpu->hm.s.fForceTLBFlush = true;
3098	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3099	}
3100
3101	/* Check for TLB flushes while switching to/from a nested-guest. */
3102	if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3103	{
3104	pVCpu->hm.s.fForceTLBFlush = true;
3105	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3106	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3107	}
3108
3109	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3110	pVCpu->hm.s.idLastCpu = pHostCpu->idCpu;
3111	if (pVCpu->hm.s.fForceTLBFlush)
3112	{
3113	++pHostCpu->uCurrentAsid;
3114	if (pHostCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
3115	{
3116	pHostCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0 */
3117	pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
3118	pHostCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
3119	}
3120
3121	pVCpu->hm.s.fForceTLBFlush = false;
3122	pVCpu->hm.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3123	pVCpu->hm.s.uCurrentAsid = pHostCpu->uCurrentAsid;
3124	if (pHostCpu->fFlushAsidBeforeUse)
3125	{
3126	if (pVM->hm.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_SINGLE_CONTEXT)
3127	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_SINGLE_CONTEXT, 0 /* GCPtr */);
3128	else if (pVM->hm.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_ALL_CONTEXTS)
3129	{
3130	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_ALL_CONTEXTS, 0 /* GCPtr */);
3131	pHostCpu->fFlushAsidBeforeUse = false;
3132	}
3133	else
3134	{
3135	/* hmR0VmxSetupTaggedTlb() ensures we never get here. Paranoia. */
3136	AssertMsgFailed(("Unsupported VPID-flush context type.\n"));
3137	}
3138	}
3139	}
3140
3141	AssertMsg(pVCpu->hm.s.cTlbFlushes == pHostCpu->cTlbFlushes,
3142	("Flush count mismatch for cpu %d (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pHostCpu->cTlbFlushes));
3143	AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
3144	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pHostCpu->idCpu,
3145	pHostCpu->uCurrentAsid, pHostCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
3146	AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
3147	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pHostCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
3148
3149	int rc = VMXWriteVmcs16(VMX_VMCS16_VPID, pVCpu->hm.s.uCurrentAsid);
3150	AssertRC(rc);
3151	}
3152
3153
3154	/**
3155	* Flushes the guest TLB entry based on CPU capabilities.
3156	*
3157	* @param pHostCpu The HM physical-CPU structure.
3158	* @param pVCpu The cross context virtual CPU structure.
3159	* @param pVmcsInfo The VMCS info. object.
3160	*
3161	* @remarks Called with interrupts disabled.
3162	*/
3163	static void hmR0VmxFlushTaggedTlb(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3164	{
3165	#ifdef HMVMX_ALWAYS_FLUSH_TLB
3166	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
3167	#endif
3168	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3169	switch (pVM->hm.s.vmx.enmTlbFlushType)
3170	{
3171	case VMXTLBFLUSHTYPE_EPT_VPID: hmR0VmxFlushTaggedTlbBoth(pHostCpu, pVCpu, pVmcsInfo); break;
3172	case VMXTLBFLUSHTYPE_EPT: hmR0VmxFlushTaggedTlbEpt(pHostCpu, pVCpu, pVmcsInfo); break;
3173	case VMXTLBFLUSHTYPE_VPID: hmR0VmxFlushTaggedTlbVpid(pHostCpu, pVCpu); break;
3174	case VMXTLBFLUSHTYPE_NONE: hmR0VmxFlushTaggedTlbNone(pHostCpu, pVCpu); break;
3175	default:
3176	AssertMsgFailed(("Invalid flush-tag function identifier\n"));
3177	break;
3178	}
3179	/* Don't assert that VMCPU_FF_TLB_FLUSH should no longer be pending. It can be set by other EMTs. */
3180	}
3181
3182
3183	/**
3184	* Sets up the appropriate tagged TLB-flush level and handler for flushing guest
3185	* TLB entries from the host TLB before VM-entry.
3186	*
3187	* @returns VBox status code.
3188	* @param pVM The cross context VM structure.
3189	*/
3190	static int hmR0VmxSetupTaggedTlb(PVMCC pVM)
3191	{
3192	/*
3193	* Determine optimal flush type for nested paging.
3194	* We cannot ignore EPT if no suitable flush-types is supported by the CPU as we've already setup
3195	* unrestricted guest execution (see hmR3InitFinalizeR0()).
3196	*/
3197	if (pVM->hm.s.fNestedPaging)
3198	{
3199	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT)
3200	{
3201	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT)
3202	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_SINGLE_CONTEXT;
3203	else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
3204	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_ALL_CONTEXTS;
3205	else
3206	{
3207	/* Shouldn't happen. EPT is supported but no suitable flush-types supported. */
3208	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3209	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_FLUSH_TYPE_UNSUPPORTED;
3210	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3211	}
3212
3213	/* Make sure the write-back cacheable memory type for EPT is supported. */
3214	if (RT_UNLIKELY(!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EMT_WB)))
3215	{
3216	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3217	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_MEM_TYPE_NOT_WB;
3218	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3219	}
3220
3221	/* EPT requires a page-walk length of 4. */
3222	if (RT_UNLIKELY(!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_PAGE_WALK_LENGTH_4)))
3223	{
3224	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3225	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_PAGE_WALK_LENGTH_UNSUPPORTED;
3226	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3227	}
3228	}
3229	else
3230	{
3231	/* Shouldn't happen. EPT is supported but INVEPT instruction is not supported. */
3232	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3233	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_INVEPT_UNAVAILABLE;
3234	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3235	}
3236	}
3237
3238	/*
3239	* Determine optimal flush type for VPID.
3240	*/
3241	if (pVM->hm.s.vmx.fVpid)
3242	{
3243	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID)
3244	{
3245	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT)
3246	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_SINGLE_CONTEXT;
3247	else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS)
3248	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_ALL_CONTEXTS;
3249	else
3250	{
3251	/* Neither SINGLE nor ALL-context flush types for VPID is supported by the CPU. Ignore VPID capability. */
3252	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
3253	LogRelFunc(("Only INDIV_ADDR supported. Ignoring VPID.\n"));
3254	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
3255	LogRelFunc(("Only SINGLE_CONTEXT_RETAIN_GLOBALS supported. Ignoring VPID.\n"));
3256	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NOT_SUPPORTED;
3257	pVM->hm.s.vmx.fVpid = false;
3258	}
3259	}
3260	else
3261	{
3262	/* Shouldn't happen. VPID is supported but INVVPID is not supported by the CPU. Ignore VPID capability. */
3263	Log4Func(("VPID supported without INVEPT support. Ignoring VPID.\n"));
3264	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NOT_SUPPORTED;
3265	pVM->hm.s.vmx.fVpid = false;
3266	}
3267	}
3268
3269	/*
3270	* Setup the handler for flushing tagged-TLBs.
3271	*/
3272	if (pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid)
3273	pVM->hm.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_EPT_VPID;
3274	else if (pVM->hm.s.fNestedPaging)
3275	pVM->hm.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_EPT;
3276	else if (pVM->hm.s.vmx.fVpid)
3277	pVM->hm.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_VPID;
3278	else
3279	pVM->hm.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_NONE;
3280	return VINF_SUCCESS;
3281	}
3282
3283
3284	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3285	/**
3286	* Sets up the shadow VMCS fields arrays.
3287	*
3288	* This function builds arrays of VMCS fields to sync the shadow VMCS later while
3289	* executing the guest.
3290	*
3291	* @returns VBox status code.
3292	* @param pVM The cross context VM structure.
3293	*/
3294	static int hmR0VmxSetupShadowVmcsFieldsArrays(PVMCC pVM)
3295	{
3296	/*
3297	* Paranoia. Ensure we haven't exposed the VMWRITE-All VMX feature to the guest
3298	* when the host does not support it.
3299	*/
3300	bool const fGstVmwriteAll = pVM->cpum.ro.GuestFeatures.fVmxVmwriteAll;
3301	if ( !fGstVmwriteAll
3302	\|\| (pVM->hm.s.vmx.Msrs.u64Misc & VMX_MISC_VMWRITE_ALL))
3303	{ /* likely. */ }
3304	else
3305	{
3306	LogRelFunc(("VMX VMWRITE-All feature exposed to the guest but host CPU does not support it!\n"));
3307	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_GST_HOST_VMWRITE_ALL;
3308	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3309	}
3310
3311	uint32_t const cVmcsFields = RT_ELEMENTS(g_aVmcsFields);
3312	uint32_t cRwFields = 0;
3313	uint32_t cRoFields = 0;
3314	for (uint32_t i = 0; i < cVmcsFields; i++)
3315	{
3316	VMXVMCSFIELD VmcsField;
3317	VmcsField.u = g_aVmcsFields[i];
3318
3319	/*
3320	* We will be writing "FULL" (64-bit) fields while syncing the shadow VMCS.
3321	* Therefore, "HIGH" (32-bit portion of 64-bit) fields must not be included
3322	* in the shadow VMCS fields array as they would be redundant.
3323	*
3324	* If the VMCS field depends on a CPU feature that is not exposed to the guest,
3325	* we must not include it in the shadow VMCS fields array. Guests attempting to
3326	* VMREAD/VMWRITE such VMCS fields would cause a VM-exit and we shall emulate
3327	* the required behavior.
3328	*/
3329	if ( VmcsField.n.fAccessType == VMX_VMCSFIELD_ACCESS_FULL
3330	&& CPUMIsGuestVmxVmcsFieldValid(pVM, VmcsField.u))
3331	{
3332	/*
3333	* Read-only fields are placed in a separate array so that while syncing shadow
3334	* VMCS fields later (which is more performance critical) we can avoid branches.
3335	*
3336	* However, if the guest can write to all fields (including read-only fields),
3337	* we treat it a as read/write field. Otherwise, writing to these fields would
3338	* cause a VMWRITE instruction error while syncing the shadow VMCS .
3339	*/
3340	if ( fGstVmwriteAll
3341	\|\| !VMXIsVmcsFieldReadOnly(VmcsField.u))
3342	pVM->hm.s.vmx.paShadowVmcsFields[cRwFields++] = VmcsField.u;
3343	else
3344	pVM->hm.s.vmx.paShadowVmcsRoFields[cRoFields++] = VmcsField.u;
3345	}
3346	}
3347
3348	/* Update the counts. */
3349	pVM->hm.s.vmx.cShadowVmcsFields = cRwFields;
3350	pVM->hm.s.vmx.cShadowVmcsRoFields = cRoFields;
3351	return VINF_SUCCESS;
3352	}
3353
3354
3355	/**
3356	* Sets up the VMREAD and VMWRITE bitmaps.
3357	*
3358	* @param pVM The cross context VM structure.
3359	*/
3360	static void hmR0VmxSetupVmreadVmwriteBitmaps(PVMCC pVM)
3361	{
3362	/*
3363	* By default, ensure guest attempts to acceses to any VMCS fields cause VM-exits.
3364	*/
3365	uint32_t const cbBitmap = X86_PAGE_4K_SIZE;
3366	uint8_t pbVmreadBitmap = (uint8_t )pVM->hm.s.vmx.pvVmreadBitmap;
3367	uint8_t pbVmwriteBitmap = (uint8_t )pVM->hm.s.vmx.pvVmwriteBitmap;
3368	ASMMemFill32(pbVmreadBitmap, cbBitmap, UINT32_C(0xffffffff));
3369	ASMMemFill32(pbVmwriteBitmap, cbBitmap, UINT32_C(0xffffffff));
3370
3371	/*
3372	* Skip intercepting VMREAD/VMWRITE to guest read/write fields in the
3373	* VMREAD and VMWRITE bitmaps.
3374	*/
3375	{
3376	uint32_t const *paShadowVmcsFields = pVM->hm.s.vmx.paShadowVmcsFields;
3377	uint32_t const cShadowVmcsFields = pVM->hm.s.vmx.cShadowVmcsFields;
3378	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
3379	{
3380	uint32_t const uVmcsField = paShadowVmcsFields[i];
3381	Assert(!(uVmcsField & VMX_VMCSFIELD_RSVD_MASK));
3382	Assert(uVmcsField >> 3 < cbBitmap);
3383	ASMBitClear(pbVmreadBitmap + (uVmcsField >> 3), uVmcsField & 7);
3384	ASMBitClear(pbVmwriteBitmap + (uVmcsField >> 3), uVmcsField & 7);
3385	}
3386	}
3387
3388	/*
3389	* Skip intercepting VMREAD for guest read-only fields in the VMREAD bitmap
3390	* if the host supports VMWRITE to all supported VMCS fields.
3391	*/
3392	if (pVM->hm.s.vmx.Msrs.u64Misc & VMX_MISC_VMWRITE_ALL)
3393	{
3394	uint32_t const *paShadowVmcsRoFields = pVM->hm.s.vmx.paShadowVmcsRoFields;
3395	uint32_t const cShadowVmcsRoFields = pVM->hm.s.vmx.cShadowVmcsRoFields;
3396	for (uint32_t i = 0; i < cShadowVmcsRoFields; i++)
3397	{
3398	uint32_t const uVmcsField = paShadowVmcsRoFields[i];
3399	Assert(!(uVmcsField & VMX_VMCSFIELD_RSVD_MASK));
3400	Assert(uVmcsField >> 3 < cbBitmap);
3401	ASMBitClear(pbVmreadBitmap + (uVmcsField >> 3), uVmcsField & 7);
3402	}
3403	}
3404	}
3405	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
3406
3407
3408	/**
3409	* Sets up the virtual-APIC page address for the VMCS.
3410	*
3411	* @param pVmcsInfo The VMCS info. object.
3412	*/
3413	DECLINLINE(void) hmR0VmxSetupVmcsVirtApicAddr(PCVMXVMCSINFO pVmcsInfo)
3414	{
3415	RTHCPHYS const HCPhysVirtApic = pVmcsInfo->HCPhysVirtApic;
3416	Assert(HCPhysVirtApic != NIL_RTHCPHYS);
3417	Assert(!(HCPhysVirtApic & 0xfff)); /* Bits 11:0 MBZ. */
3418	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL, HCPhysVirtApic);
3419	AssertRC(rc);
3420	}
3421
3422
3423	/**
3424	* Sets up the MSR-bitmap address for the VMCS.
3425	*
3426	* @param pVmcsInfo The VMCS info. object.
3427	*/
3428	DECLINLINE(void) hmR0VmxSetupVmcsMsrBitmapAddr(PCVMXVMCSINFO pVmcsInfo)
3429	{
3430	RTHCPHYS const HCPhysMsrBitmap = pVmcsInfo->HCPhysMsrBitmap;
3431	Assert(HCPhysMsrBitmap != NIL_RTHCPHYS);
3432	Assert(!(HCPhysMsrBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3433	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_MSR_BITMAP_FULL, HCPhysMsrBitmap);
3434	AssertRC(rc);
3435	}
3436
3437
3438	/**
3439	* Sets up the APIC-access page address for the VMCS.
3440	*
3441	* @param pVCpu The cross context virtual CPU structure.
3442	*/
3443	DECLINLINE(void) hmR0VmxSetupVmcsApicAccessAddr(PVMCPUCC pVCpu)
3444	{
3445	RTHCPHYS const HCPhysApicAccess = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.HCPhysApicAccess;
3446	Assert(HCPhysApicAccess != NIL_RTHCPHYS);
3447	Assert(!(HCPhysApicAccess & 0xfff)); /* Bits 11:0 MBZ. */
3448	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, HCPhysApicAccess);
3449	AssertRC(rc);
3450	}
3451
3452
3453	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3454	/**
3455	* Sets up the VMREAD bitmap address for the VMCS.
3456	*
3457	* @param pVCpu The cross context virtual CPU structure.
3458	*/
3459	DECLINLINE(void) hmR0VmxSetupVmcsVmreadBitmapAddr(PVMCPUCC pVCpu)
3460	{
3461	RTHCPHYS const HCPhysVmreadBitmap = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.HCPhysVmreadBitmap;
3462	Assert(HCPhysVmreadBitmap != NIL_RTHCPHYS);
3463	Assert(!(HCPhysVmreadBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3464	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VMREAD_BITMAP_FULL, HCPhysVmreadBitmap);
3465	AssertRC(rc);
3466	}
3467
3468
3469	/**
3470	* Sets up the VMWRITE bitmap address for the VMCS.
3471	*
3472	* @param pVCpu The cross context virtual CPU structure.
3473	*/
3474	DECLINLINE(void) hmR0VmxSetupVmcsVmwriteBitmapAddr(PVMCPUCC pVCpu)
3475	{
3476	RTHCPHYS const HCPhysVmwriteBitmap = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.HCPhysVmwriteBitmap;
3477	Assert(HCPhysVmwriteBitmap != NIL_RTHCPHYS);
3478	Assert(!(HCPhysVmwriteBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3479	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VMWRITE_BITMAP_FULL, HCPhysVmwriteBitmap);
3480	AssertRC(rc);
3481	}
3482	#endif
3483
3484
3485	/**
3486	* Sets up the VM-entry MSR load, VM-exit MSR-store and VM-exit MSR-load addresses
3487	* in the VMCS.
3488	*
3489	* @returns VBox status code.
3490	* @param pVmcsInfo The VMCS info. object.
3491	*/
3492	DECLINLINE(int) hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(PVMXVMCSINFO pVmcsInfo)
3493	{
3494	RTHCPHYS const HCPhysGuestMsrLoad = pVmcsInfo->HCPhysGuestMsrLoad;
3495	Assert(HCPhysGuestMsrLoad != NIL_RTHCPHYS);
3496	Assert(!(HCPhysGuestMsrLoad & 0xf)); /* Bits 3:0 MBZ. */
3497
3498	RTHCPHYS const HCPhysGuestMsrStore = pVmcsInfo->HCPhysGuestMsrStore;
3499	Assert(HCPhysGuestMsrStore != NIL_RTHCPHYS);
3500	Assert(!(HCPhysGuestMsrStore & 0xf)); /* Bits 3:0 MBZ. */
3501
3502	RTHCPHYS const HCPhysHostMsrLoad = pVmcsInfo->HCPhysHostMsrLoad;
3503	Assert(HCPhysHostMsrLoad != NIL_RTHCPHYS);
3504	Assert(!(HCPhysHostMsrLoad & 0xf)); /* Bits 3:0 MBZ. */
3505
3506	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL, HCPhysGuestMsrLoad); AssertRC(rc);
3507	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL, HCPhysGuestMsrStore); AssertRC(rc);
3508	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL, HCPhysHostMsrLoad); AssertRC(rc);
3509	return VINF_SUCCESS;
3510	}
3511
3512
3513	/**
3514	* Sets up MSR permissions in the MSR bitmap of a VMCS info. object.
3515	*
3516	* @param pVCpu The cross context virtual CPU structure.
3517	* @param pVmcsInfo The VMCS info. object.
3518	*/
3519	static void hmR0VmxSetupVmcsMsrPermissions(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3520	{
3521	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS);
3522
3523	/*
3524	* The guest can access the following MSRs (read, write) without causing
3525	* VM-exits; they are loaded/stored automatically using fields in the VMCS.
3526	*/
3527	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3528	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_CS, VMXMSRPM_ALLOW_RD_WR);
3529	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_ESP, VMXMSRPM_ALLOW_RD_WR);
3530	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_EIP, VMXMSRPM_ALLOW_RD_WR);
3531	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_GS_BASE, VMXMSRPM_ALLOW_RD_WR);
3532	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_FS_BASE, VMXMSRPM_ALLOW_RD_WR);
3533
3534	/*
3535	* The IA32_PRED_CMD and IA32_FLUSH_CMD MSRs are write-only and has no state
3536	* associated with then. We never need to intercept access (writes need to be
3537	* executed without causing a VM-exit, reads will #GP fault anyway).
3538	*
3539	* The IA32_SPEC_CTRL MSR is read/write and has state. We allow the guest to
3540	* read/write them. We swap the the guest/host MSR value using the
3541	* auto-load/store MSR area.
3542	*/
3543	if (pVM->cpum.ro.GuestFeatures.fIbpb)
3544	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_PRED_CMD, VMXMSRPM_ALLOW_RD_WR);
3545	if (pVM->cpum.ro.GuestFeatures.fFlushCmd)
3546	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_FLUSH_CMD, VMXMSRPM_ALLOW_RD_WR);
3547	if (pVM->cpum.ro.GuestFeatures.fIbrs)
3548	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SPEC_CTRL, VMXMSRPM_ALLOW_RD_WR);
3549
3550	/*
3551	* Allow full read/write access for the following MSRs (mandatory for VT-x)
3552	* required for 64-bit guests.
3553	*/
3554	if (pVM->hm.s.fAllow64BitGuests)
3555	{
3556	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_LSTAR, VMXMSRPM_ALLOW_RD_WR);
3557	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K6_STAR, VMXMSRPM_ALLOW_RD_WR);
3558	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_SF_MASK, VMXMSRPM_ALLOW_RD_WR);
3559	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_KERNEL_GS_BASE, VMXMSRPM_ALLOW_RD_WR);
3560	}
3561
3562	/*
3563	* IA32_EFER MSR is always intercepted, see @bugref{9180#c37}.
3564	*/
3565	#ifdef VBOX_STRICT
3566	Assert(pVmcsInfo->pvMsrBitmap);
3567	uint32_t const fMsrpmEfer = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, MSR_K6_EFER);
3568	Assert(fMsrpmEfer == VMXMSRPM_EXIT_RD_WR);
3569	#endif
3570	}
3571
3572
3573	/**
3574	* Sets up pin-based VM-execution controls in the VMCS.
3575	*
3576	* @returns VBox status code.
3577	* @param pVCpu The cross context virtual CPU structure.
3578	* @param pVmcsInfo The VMCS info. object.
3579	*/
3580	static int hmR0VmxSetupVmcsPinCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3581	{
3582	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3583	uint32_t fVal = pVM->hm.s.vmx.Msrs.PinCtls.n.allowed0; /* Bits set here must always be set. */
3584	uint32_t const fZap = pVM->hm.s.vmx.Msrs.PinCtls.n.allowed1; /* Bits cleared here must always be cleared. */
3585
3586	fVal \|= VMX_PIN_CTLS_EXT_INT_EXIT /* External interrupts cause a VM-exit. */
3587	\| VMX_PIN_CTLS_NMI_EXIT; /* Non-maskable interrupts (NMIs) cause a VM-exit. */
3588
3589	if (pVM->hm.s.vmx.Msrs.PinCtls.n.allowed1 & VMX_PIN_CTLS_VIRT_NMI)
3590	fVal \|= VMX_PIN_CTLS_VIRT_NMI; /* Use virtual NMIs and virtual-NMI blocking features. */
3591
3592	/* Enable the VMX-preemption timer. */
3593	if (pVM->hm.s.vmx.fUsePreemptTimer)
3594	{
3595	Assert(pVM->hm.s.vmx.Msrs.PinCtls.n.allowed1 & VMX_PIN_CTLS_PREEMPT_TIMER);
3596	fVal \|= VMX_PIN_CTLS_PREEMPT_TIMER;
3597	}
3598
3599	#if 0
3600	/* Enable posted-interrupt processing. */
3601	if (pVM->hm.s.fPostedIntrs)
3602	{
3603	Assert(pVM->hm.s.vmx.Msrs.PinCtls.n.allowed1 & VMX_PIN_CTLS_POSTED_INT);
3604	Assert(pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_ACK_EXT_INT);
3605	fVal \|= VMX_PIN_CTLS_POSTED_INT;
3606	}
3607	#endif
3608
3609	if ((fVal & fZap) != fVal)
3610	{
3611	LogRelFunc(("Invalid pin-based VM-execution controls combo! Cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3612	pVM->hm.s.vmx.Msrs.PinCtls.n.allowed0, fVal, fZap));
3613	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PIN_EXEC;
3614	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3615	}
3616
3617	/* Commit it to the VMCS and update our cache. */
3618	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, fVal);
3619	AssertRC(rc);
3620	pVmcsInfo->u32PinCtls = fVal;
3621
3622	return VINF_SUCCESS;
3623	}
3624
3625
3626	/**
3627	* Sets up secondary processor-based VM-execution controls in the VMCS.
3628	*
3629	* @returns VBox status code.
3630	* @param pVCpu The cross context virtual CPU structure.
3631	* @param pVmcsInfo The VMCS info. object.
3632	*/
3633	static int hmR0VmxSetupVmcsProcCtls2(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3634	{
3635	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3636	uint32_t fVal = pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed0; /* Bits set here must be set in the VMCS. */
3637	uint32_t const fZap = pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3638
3639	/* WBINVD causes a VM-exit. */
3640	if (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_WBINVD_EXIT)
3641	fVal \|= VMX_PROC_CTLS2_WBINVD_EXIT;
3642
3643	/* Enable EPT (aka nested-paging). */
3644	if (pVM->hm.s.fNestedPaging)
3645	fVal \|= VMX_PROC_CTLS2_EPT;
3646
3647	/* Enable the INVPCID instruction if we expose it to the guest and is supported
3648	by the hardware. Without this, guest executing INVPCID would cause a #UD. */
3649	if ( pVM->cpum.ro.GuestFeatures.fInvpcid
3650	&& (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_INVPCID))
3651	fVal \|= VMX_PROC_CTLS2_INVPCID;
3652
3653	/* Enable VPID. */
3654	if (pVM->hm.s.vmx.fVpid)
3655	fVal \|= VMX_PROC_CTLS2_VPID;
3656
3657	/* Enable unrestricted guest execution. */
3658	if (pVM->hm.s.vmx.fUnrestrictedGuest)
3659	fVal \|= VMX_PROC_CTLS2_UNRESTRICTED_GUEST;
3660
3661	#if 0
3662	if (pVM->hm.s.fVirtApicRegs)
3663	{
3664	/* Enable APIC-register virtualization. */
3665	Assert(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_APIC_REG_VIRT);
3666	fVal \|= VMX_PROC_CTLS2_APIC_REG_VIRT;
3667
3668	/* Enable virtual-interrupt delivery. */
3669	Assert(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_INTR_DELIVERY);
3670	fVal \|= VMX_PROC_CTLS2_VIRT_INTR_DELIVERY;
3671	}
3672	#endif
3673
3674	/* Virtualize-APIC accesses if supported by the CPU. The virtual-APIC page is
3675	where the TPR shadow resides. */
3676	/** @todo VIRT_X2APIC support, it's mutually exclusive with this. So must be
3677	* done dynamically. */
3678	if (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
3679	{
3680	fVal \|= VMX_PROC_CTLS2_VIRT_APIC_ACCESS;
3681	hmR0VmxSetupVmcsApicAccessAddr(pVCpu);
3682	}
3683
3684	/* Enable the RDTSCP instruction if we expose it to the guest and is supported
3685	by the hardware. Without this, guest executing RDTSCP would cause a #UD. */
3686	if ( pVM->cpum.ro.GuestFeatures.fRdTscP
3687	&& (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_RDTSCP))
3688	fVal \|= VMX_PROC_CTLS2_RDTSCP;
3689
3690	/* Enable Pause-Loop exiting. */
3691	if ( (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT)
3692	&& pVM->hm.s.vmx.cPleGapTicks
3693	&& pVM->hm.s.vmx.cPleWindowTicks)
3694	{
3695	fVal \|= VMX_PROC_CTLS2_PAUSE_LOOP_EXIT;
3696
3697	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, pVM->hm.s.vmx.cPleGapTicks); AssertRC(rc);
3698	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, pVM->hm.s.vmx.cPleWindowTicks); AssertRC(rc);
3699	}
3700
3701	if ((fVal & fZap) != fVal)
3702	{
3703	LogRelFunc(("Invalid secondary processor-based VM-execution controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3704	pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed0, fVal, fZap));
3705	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC2;
3706	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3707	}
3708
3709	/* Commit it to the VMCS and update our cache. */
3710	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, fVal);
3711	AssertRC(rc);
3712	pVmcsInfo->u32ProcCtls2 = fVal;
3713
3714	return VINF_SUCCESS;
3715	}
3716
3717
3718	/**
3719	* Sets up processor-based VM-execution controls in the VMCS.
3720	*
3721	* @returns VBox status code.
3722	* @param pVCpu The cross context virtual CPU structure.
3723	* @param pVmcsInfo The VMCS info. object.
3724	*/
3725	static int hmR0VmxSetupVmcsProcCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3726	{
3727	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3728
3729	uint32_t fVal = pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
3730	uint32_t const fZap = pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3731
3732	fVal \|= VMX_PROC_CTLS_HLT_EXIT /* HLT causes a VM-exit. */
3733	\| VMX_PROC_CTLS_USE_TSC_OFFSETTING /* Use TSC-offsetting. */
3734	\| VMX_PROC_CTLS_MOV_DR_EXIT /* MOV DRx causes a VM-exit. */
3735	\| VMX_PROC_CTLS_UNCOND_IO_EXIT /* All IO instructions cause a VM-exit. */
3736	\| VMX_PROC_CTLS_RDPMC_EXIT /* RDPMC causes a VM-exit. */
3737	\| VMX_PROC_CTLS_MONITOR_EXIT /* MONITOR causes a VM-exit. */
3738	\| VMX_PROC_CTLS_MWAIT_EXIT; /* MWAIT causes a VM-exit. */
3739
3740	/* We toggle VMX_PROC_CTLS_MOV_DR_EXIT later, check if it's not -always- needed to be set or clear. */
3741	if ( !(pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_MOV_DR_EXIT)
3742	\|\| (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed0 & VMX_PROC_CTLS_MOV_DR_EXIT))
3743	{
3744	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_MOV_DRX_EXIT;
3745	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3746	}
3747
3748	/* Without nested paging, INVLPG (also affects INVPCID) and MOV CR3 instructions should cause VM-exits. */
3749	if (!pVM->hm.s.fNestedPaging)
3750	{
3751	Assert(!pVM->hm.s.vmx.fUnrestrictedGuest);
3752	fVal \|= VMX_PROC_CTLS_INVLPG_EXIT
3753	\| VMX_PROC_CTLS_CR3_LOAD_EXIT
3754	\| VMX_PROC_CTLS_CR3_STORE_EXIT;
3755	}
3756
3757	/* Use TPR shadowing if supported by the CPU. */
3758	if ( PDMHasApic(pVM)
3759	&& pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW)
3760	{
3761	fVal \|= VMX_PROC_CTLS_USE_TPR_SHADOW; /* CR8 reads from the Virtual-APIC page. */
3762	/* CR8 writes cause a VM-exit based on TPR threshold. */
3763	Assert(!(fVal & VMX_PROC_CTLS_CR8_STORE_EXIT));
3764	Assert(!(fVal & VMX_PROC_CTLS_CR8_LOAD_EXIT));
3765	hmR0VmxSetupVmcsVirtApicAddr(pVmcsInfo);
3766	}
3767	else
3768	{
3769	/* Some 32-bit CPUs do not support CR8 load/store exiting as MOV CR8 is
3770	invalid on 32-bit Intel CPUs. Set this control only for 64-bit guests. */
3771	if (pVM->hm.s.fAllow64BitGuests)
3772	{
3773	fVal \|= VMX_PROC_CTLS_CR8_STORE_EXIT /* CR8 reads cause a VM-exit. */
3774	\| VMX_PROC_CTLS_CR8_LOAD_EXIT; /* CR8 writes cause a VM-exit. */
3775	}
3776	}
3777
3778	/* Use MSR-bitmaps if supported by the CPU. */
3779	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
3780	{
3781	fVal \|= VMX_PROC_CTLS_USE_MSR_BITMAPS;
3782	hmR0VmxSetupVmcsMsrBitmapAddr(pVmcsInfo);
3783	}
3784
3785	/* Use the secondary processor-based VM-execution controls if supported by the CPU. */
3786	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
3787	fVal \|= VMX_PROC_CTLS_USE_SECONDARY_CTLS;
3788
3789	if ((fVal & fZap) != fVal)
3790	{
3791	LogRelFunc(("Invalid processor-based VM-execution controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3792	pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed0, fVal, fZap));
3793	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC;
3794	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3795	}
3796
3797	/* Commit it to the VMCS and update our cache. */
3798	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, fVal);
3799	AssertRC(rc);
3800	pVmcsInfo->u32ProcCtls = fVal;
3801
3802	/* Set up MSR permissions that don't change through the lifetime of the VM. */
3803	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
3804	hmR0VmxSetupVmcsMsrPermissions(pVCpu, pVmcsInfo);
3805
3806	/* Set up secondary processor-based VM-execution controls if the CPU supports it. */
3807	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
3808	return hmR0VmxSetupVmcsProcCtls2(pVCpu, pVmcsInfo);
3809
3810	/* Sanity check, should not really happen. */
3811	if (RT_LIKELY(!pVM->hm.s.vmx.fUnrestrictedGuest))
3812	{ /* likely */ }
3813	else
3814	{
3815	pVCpu->hm.s.u32HMError = VMX_UFC_INVALID_UX_COMBO;
3816	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3817	}
3818
3819	/* Old CPUs without secondary processor-based VM-execution controls would end up here. */
3820	return VINF_SUCCESS;
3821	}
3822
3823
3824	/**
3825	* Sets up miscellaneous (everything other than Pin, Processor and secondary
3826	* Processor-based VM-execution) control fields in the VMCS.
3827	*
3828	* @returns VBox status code.
3829	* @param pVCpu The cross context virtual CPU structure.
3830	* @param pVmcsInfo The VMCS info. object.
3831	*/
3832	static int hmR0VmxSetupVmcsMiscCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3833	{
3834	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3835	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUseVmcsShadowing)
3836	{
3837	hmR0VmxSetupVmcsVmreadBitmapAddr(pVCpu);
3838	hmR0VmxSetupVmcsVmwriteBitmapAddr(pVCpu);
3839	}
3840	#endif
3841
3842	Assert(pVmcsInfo->u64VmcsLinkPtr == NIL_RTHCPHYS);
3843	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
3844	AssertRC(rc);
3845
3846	rc = hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(pVmcsInfo);
3847	if (RT_SUCCESS(rc))
3848	{
3849	uint64_t const u64Cr0Mask = hmR0VmxGetFixedCr0Mask(pVCpu);
3850	uint64_t const u64Cr4Mask = hmR0VmxGetFixedCr4Mask(pVCpu);
3851
3852	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, u64Cr0Mask); AssertRC(rc);
3853	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, u64Cr4Mask); AssertRC(rc);
3854
3855	pVmcsInfo->u64Cr0Mask = u64Cr0Mask;
3856	pVmcsInfo->u64Cr4Mask = u64Cr4Mask;
3857	return VINF_SUCCESS;
3858	}
3859	else
3860	LogRelFunc(("Failed to initialize VMCS auto-load/store MSR addresses. rc=%Rrc\n", rc));
3861	return rc;
3862	}
3863
3864
3865	/**
3866	* Sets up the initial exception bitmap in the VMCS based on static conditions.
3867	*
3868	* We shall setup those exception intercepts that don't change during the
3869	* lifetime of the VM here. The rest are done dynamically while loading the
3870	* guest state.
3871	*
3872	* @param pVCpu The cross context virtual CPU structure.
3873	* @param pVmcsInfo The VMCS info. object.
3874	*/
3875	static void hmR0VmxSetupVmcsXcptBitmap(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3876	{
3877	/*
3878	* The following exceptions are always intercepted:
3879	*
3880	* #AC - To prevent the guest from hanging the CPU.
3881	* #DB - To maintain the DR6 state even when intercepting DRx reads/writes and
3882	* recursive #DBs can cause a CPU hang.
3883	* #PF - To sync our shadow page tables when nested-paging is not used.
3884	*/
3885	bool const fNestedPaging = pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging;
3886	uint32_t const uXcptBitmap = RT_BIT(X86_XCPT_AC)
3887	\| RT_BIT(X86_XCPT_DB)
3888	\| (fNestedPaging ? 0 : RT_BIT(X86_XCPT_PF));
3889
3890	/* Commit it to the VMCS. */
3891	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
3892	AssertRC(rc);
3893
3894	/* Update our cache of the exception bitmap. */
3895	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
3896	}
3897
3898
3899	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3900	/**
3901	* Sets up the VMCS for executing a nested-guest using hardware-assisted VMX.
3902	*
3903	* @returns VBox status code.
3904	* @param pVCpu The cross context virtual CPU structure.
3905	* @param pVmcsInfo The VMCS info. object.
3906	*/
3907	static int hmR0VmxSetupVmcsCtlsNested(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3908	{
3909	Assert(pVmcsInfo->u64VmcsLinkPtr == NIL_RTHCPHYS);
3910	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
3911	AssertRC(rc);
3912
3913	rc = hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(pVmcsInfo);
3914	if (RT_SUCCESS(rc))
3915	{
3916	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
3917	hmR0VmxSetupVmcsMsrBitmapAddr(pVmcsInfo);
3918	return VINF_SUCCESS;
3919	}
3920	else
3921	LogRelFunc(("Failed to set up the VMCS link pointer in the nested-guest VMCS. rc=%Rrc\n", rc));
3922	return rc;
3923	}
3924	#endif
3925
3926
3927	/**
3928	* Sets up the VMCS for executing a guest (or nested-guest) using hardware-assisted
3929	* VMX.
3930	*
3931	* @returns VBox status code.
3932	* @param pVCpu The cross context virtual CPU structure.
3933	* @param pVmcsInfo The VMCS info. object.
3934	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
3935	*/
3936	static int hmR0VmxSetupVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
3937	{
3938	Assert(pVmcsInfo->pvVmcs);
3939	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
3940
3941	/* Set the CPU specified revision identifier at the beginning of the VMCS structure. */
3942	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3943	(uint32_t )pVmcsInfo->pvVmcs = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_ID);
3944	const char * const pszVmcs = fIsNstGstVmcs ? "nested-guest VMCS" : "guest VMCS";
3945
3946	LogFlowFunc(("\n"));
3947
3948	/*
3949	* Initialize the VMCS using VMCLEAR before loading the VMCS.
3950	* See Intel spec. 31.6 "Preparation And Launching A Virtual Machine".
3951	*/
3952	int rc = hmR0VmxClearVmcs(pVmcsInfo);
3953	if (RT_SUCCESS(rc))
3954	{
3955	rc = hmR0VmxLoadVmcs(pVmcsInfo);
3956	if (RT_SUCCESS(rc))
3957	{
3958	if (!fIsNstGstVmcs)
3959	{
3960	rc = hmR0VmxSetupVmcsPinCtls(pVCpu, pVmcsInfo);
3961	if (RT_SUCCESS(rc))
3962	{
3963	rc = hmR0VmxSetupVmcsProcCtls(pVCpu, pVmcsInfo);
3964	if (RT_SUCCESS(rc))
3965	{
3966	rc = hmR0VmxSetupVmcsMiscCtls(pVCpu, pVmcsInfo);
3967	if (RT_SUCCESS(rc))
3968	{
3969	hmR0VmxSetupVmcsXcptBitmap(pVCpu, pVmcsInfo);
3970	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3971	/*
3972	* If a shadow VMCS is allocated for the VMCS info. object, initialize the
3973	* VMCS revision ID and shadow VMCS indicator bit. Also, clear the VMCS
3974	* making it fit for use when VMCS shadowing is later enabled.
3975	*/
3976	if (pVmcsInfo->pvShadowVmcs)
3977	{
3978	VMXVMCSREVID VmcsRevId;
3979	VmcsRevId.u = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_ID);
3980	VmcsRevId.n.fIsShadowVmcs = 1;
3981	(uint32_t )pVmcsInfo->pvShadowVmcs = VmcsRevId.u;
3982	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
3983	if (RT_SUCCESS(rc))
3984	{ /* likely */ }
3985	else
3986	LogRelFunc(("Failed to initialize shadow VMCS. rc=%Rrc\n", rc));
3987	}
3988	#endif
3989	}
3990	else
3991	LogRelFunc(("Failed to setup miscellaneous controls. rc=%Rrc\n", rc));
3992	}
3993	else
3994	LogRelFunc(("Failed to setup processor-based VM-execution controls. rc=%Rrc\n", rc));
3995	}
3996	else
3997	LogRelFunc(("Failed to setup pin-based controls. rc=%Rrc\n", rc));
3998	}
3999	else
4000	{
4001	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4002	rc = hmR0VmxSetupVmcsCtlsNested(pVCpu, pVmcsInfo);
4003	if (RT_SUCCESS(rc))
4004	{ /* likely */ }
4005	else
4006	LogRelFunc(("Failed to initialize nested-guest VMCS. rc=%Rrc\n", rc));
4007	#else
4008	AssertFailed();
4009	#endif
4010	}
4011	}
4012	else
4013	LogRelFunc(("Failed to load the %s. rc=%Rrc\n", rc, pszVmcs));
4014	}
4015	else
4016	LogRelFunc(("Failed to clear the %s. rc=%Rrc\n", rc, pszVmcs));
4017
4018	/* Sync any CPU internal VMCS data back into our VMCS in memory. */
4019	if (RT_SUCCESS(rc))
4020	{
4021	rc = hmR0VmxClearVmcs(pVmcsInfo);
4022	if (RT_SUCCESS(rc))
4023	{ /* likely */ }
4024	else
4025	LogRelFunc(("Failed to clear the %s post setup. rc=%Rrc\n", rc, pszVmcs));
4026	}
4027
4028	/*
4029	* Update the last-error record both for failures and success, so we
4030	* can propagate the status code back to ring-3 for diagnostics.
4031	*/
4032	hmR0VmxUpdateErrorRecord(pVCpu, rc);
4033	NOREF(pszVmcs);
4034	return rc;
4035	}
4036
4037
4038	/**
4039	* Does global VT-x initialization (called during module initialization).
4040	*
4041	* @returns VBox status code.
4042	*/
4043	VMMR0DECL(int) VMXR0GlobalInit(void)
4044	{
4045	#ifdef HMVMX_USE_FUNCTION_TABLE
4046	AssertCompile(VMX_EXIT_MAX + 1 == RT_ELEMENTS(g_apfnVMExitHandlers));
4047	# ifdef VBOX_STRICT
4048	for (unsigned i = 0; i < RT_ELEMENTS(g_apfnVMExitHandlers); i++)
4049	Assert(g_apfnVMExitHandlers[i]);
4050	# endif
4051	#endif
4052	return VINF_SUCCESS;
4053	}
4054
4055
4056	/**
4057	* Does global VT-x termination (called during module termination).
4058	*/
4059	VMMR0DECL(void) VMXR0GlobalTerm()
4060	{
4061	/* Nothing to do currently. */
4062	}
4063
4064
4065	/**
4066	* Sets up and activates VT-x on the current CPU.
4067	*
4068	* @returns VBox status code.
4069	* @param pHostCpu The HM physical-CPU structure.
4070	* @param pVM The cross context VM structure. Can be
4071	* NULL after a host resume operation.
4072	* @param pvCpuPage Pointer to the VMXON region (can be NULL if @a
4073	* fEnabledByHost is @c true).
4074	* @param HCPhysCpuPage Physical address of the VMXON region (can be 0 if
4075	* @a fEnabledByHost is @c true).
4076	* @param fEnabledByHost Set if SUPR0EnableVTx() or similar was used to
4077	* enable VT-x on the host.
4078	* @param pHwvirtMsrs Pointer to the hardware-virtualization MSRs.
4079	*/
4080	VMMR0DECL(int) VMXR0EnableCpu(PHMPHYSCPU pHostCpu, PVMCC pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost,
4081	PCSUPHWVIRTMSRS pHwvirtMsrs)
4082	{
4083	AssertPtr(pHostCpu);
4084	AssertPtr(pHwvirtMsrs);
4085	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4086
4087	/* Enable VT-x if it's not already enabled by the host. */
4088	if (!fEnabledByHost)
4089	{
4090	int rc = hmR0VmxEnterRootMode(pVM, HCPhysCpuPage, pvCpuPage);
4091	if (RT_FAILURE(rc))
4092	return rc;
4093	}
4094
4095	/*
4096	* Flush all EPT tagged-TLB entries (in case VirtualBox or any other hypervisor have been
4097	* using EPTPs) so we don't retain any stale guest-physical mappings which won't get
4098	* invalidated when flushing by VPID.
4099	*/
4100	if (pHwvirtMsrs->u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
4101	{
4102	hmR0VmxFlushEpt(NULL /* pVCpu /, NULL / pVmcsInfo */, VMXTLBFLUSHEPT_ALL_CONTEXTS);
4103	pHostCpu->fFlushAsidBeforeUse = false;
4104	}
4105	else
4106	pHostCpu->fFlushAsidBeforeUse = true;
4107
4108	/* Ensure each VCPU scheduled on this CPU gets a new VPID on resume. See @bugref{6255}. */
4109	++pHostCpu->cTlbFlushes;
4110
4111	return VINF_SUCCESS;
4112	}
4113
4114
4115	/**
4116	* Deactivates VT-x on the current CPU.
4117	*
4118	* @returns VBox status code.
4119	* @param pvCpuPage Pointer to the VMXON region.
4120	* @param HCPhysCpuPage Physical address of the VMXON region.
4121	*
4122	* @remarks This function should never be called when SUPR0EnableVTx() or
4123	* similar was used to enable VT-x on the host.
4124	*/
4125	VMMR0DECL(int) VMXR0DisableCpu(void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
4126	{
4127	RT_NOREF2(pvCpuPage, HCPhysCpuPage);
4128
4129	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4130	return hmR0VmxLeaveRootMode();
4131	}
4132
4133
4134	/**
4135	* Does per-VM VT-x initialization.
4136	*
4137	* @returns VBox status code.
4138	* @param pVM The cross context VM structure.
4139	*/
4140	VMMR0DECL(int) VMXR0InitVM(PVMCC pVM)
4141	{
4142	AssertPtr(pVM);
4143	LogFlowFunc(("pVM=%p\n", pVM));
4144
4145	int rc = hmR0VmxStructsAlloc(pVM);
4146	if (RT_FAILURE(rc))
4147	{
4148	LogRelFunc(("Failed to allocated VMX structures. rc=%Rrc\n", rc));
4149	return rc;
4150	}
4151
4152	return VINF_SUCCESS;
4153	}
4154
4155
4156	/**
4157	* Does per-VM VT-x termination.
4158	*
4159	* @returns VBox status code.
4160	* @param pVM The cross context VM structure.
4161	*/
4162	VMMR0DECL(int) VMXR0TermVM(PVMCC pVM)
4163	{
4164	AssertPtr(pVM);
4165	LogFlowFunc(("pVM=%p\n", pVM));
4166
4167	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4168	if (pVM->hm.s.vmx.hMemObjScratch != NIL_RTR0MEMOBJ)
4169	{
4170	Assert(pVM->hm.s.vmx.pvScratch);
4171	ASMMemZero32(pVM->hm.s.vmx.pvScratch, X86_PAGE_4K_SIZE);
4172	}
4173	#endif
4174	hmR0VmxStructsFree(pVM);
4175	return VINF_SUCCESS;
4176	}
4177
4178
4179	/**
4180	* Sets up the VM for execution using hardware-assisted VMX.
4181	* This function is only called once per-VM during initialization.
4182	*
4183	* @returns VBox status code.
4184	* @param pVM The cross context VM structure.
4185	*/
4186	VMMR0DECL(int) VMXR0SetupVM(PVMCC pVM)
4187	{
4188	AssertPtr(pVM);
4189	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4190
4191	LogFlowFunc(("pVM=%p\n", pVM));
4192
4193	/*
4194	* At least verify if VMX is enabled, since we can't check if we're in
4195	* VMX root mode or not without causing a #GP.
4196	*/
4197	RTCCUINTREG const uHostCr4 = ASMGetCR4();
4198	if (RT_LIKELY(uHostCr4 & X86_CR4_VMXE))
4199	{ /* likely */ }
4200	else
4201	return VERR_VMX_NOT_IN_VMX_ROOT_MODE;
4202
4203	/*
4204	* Without unrestricted guest execution, pRealModeTSS and pNonPagingModeEPTPageTable must
4205	* always be allocated. We no longer support the highly unlikely case of unrestricted guest
4206	* without pRealModeTSS, see hmR3InitFinalizeR0Intel().
4207	*/
4208	if ( !pVM->hm.s.vmx.fUnrestrictedGuest
4209	&& ( !pVM->hm.s.vmx.pNonPagingModeEPTPageTable
4210	\|\| !pVM->hm.s.vmx.pRealModeTSS))
4211	{
4212	LogRelFunc(("Invalid real-on-v86 state.\n"));
4213	return VERR_INTERNAL_ERROR;
4214	}
4215
4216	/* Initialize these always, see hmR3InitFinalizeR0().*/
4217	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NONE;
4218	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NONE;
4219
4220	/* Setup the tagged-TLB flush handlers. */
4221	int rc = hmR0VmxSetupTaggedTlb(pVM);
4222	if (RT_FAILURE(rc))
4223	{
4224	LogRelFunc(("Failed to setup tagged TLB. rc=%Rrc\n", rc));
4225	return rc;
4226	}
4227
4228	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4229	/* Setup the shadow VMCS fields array and VMREAD/VMWRITE bitmaps. */
4230	if (pVM->hm.s.vmx.fUseVmcsShadowing)
4231	{
4232	rc = hmR0VmxSetupShadowVmcsFieldsArrays(pVM);
4233	if (RT_SUCCESS(rc))
4234	hmR0VmxSetupVmreadVmwriteBitmaps(pVM);
4235	else
4236	{
4237	LogRelFunc(("Failed to setup shadow VMCS fields arrays. rc=%Rrc\n", rc));
4238	return rc;
4239	}
4240	}
4241	#endif
4242
4243	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
4244	{
4245	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
4246	Log4Func(("pVCpu=%p idCpu=%RU32\n", pVCpu, pVCpu->idCpu));
4247
4248	rc = hmR0VmxSetupVmcs(pVCpu, &pVCpu->hm.s.vmx.VmcsInfo, false /* fIsNstGstVmcs */);
4249	if (RT_SUCCESS(rc))
4250	{
4251	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4252	if (pVM->cpum.ro.GuestFeatures.fVmx)
4253	{
4254	rc = hmR0VmxSetupVmcs(pVCpu, &pVCpu->hm.s.vmx.VmcsInfoNstGst, true /* fIsNstGstVmcs */);
4255	if (RT_SUCCESS(rc))
4256	{ /* likely */ }
4257	else
4258	{
4259	LogRelFunc(("Nested-guest VMCS setup failed. rc=%Rrc\n", rc));
4260	return rc;
4261	}
4262	}
4263	#endif
4264	}
4265	else
4266	{
4267	LogRelFunc(("VMCS setup failed. rc=%Rrc\n", rc));
4268	return rc;
4269	}
4270	}
4271
4272	return VINF_SUCCESS;
4273	}
4274
4275
4276	/**
4277	* Saves the host control registers (CR0, CR3, CR4) into the host-state area in
4278	* the VMCS.
4279	*/
4280	static void hmR0VmxExportHostControlRegs(void)
4281	{
4282	int rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR0, ASMGetCR0()); AssertRC(rc);
4283	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR3, ASMGetCR3()); AssertRC(rc);
4284	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR4, ASMGetCR4()); AssertRC(rc);
4285	}
4286
4287
4288	/**
4289	* Saves the host segment registers and GDTR, IDTR, (TR, GS and FS bases) into
4290	* the host-state area in the VMCS.
4291	*
4292	* @returns VBox status code.
4293	* @param pVCpu The cross context virtual CPU structure.
4294	*/
4295	static int hmR0VmxExportHostSegmentRegs(PVMCPUCC pVCpu)
4296	{
4297	/**
4298	* Macro for adjusting host segment selectors to satisfy VT-x's VM-entry
4299	* requirements. See hmR0VmxExportHostSegmentRegs().
4300	*/
4301	#define VMXLOCAL_ADJUST_HOST_SEG(a_Seg, a_selValue) \
4302	if ((a_selValue) & (X86_SEL_RPL \| X86_SEL_LDT)) \
4303	{ \
4304	bool fValidSelector = true; \
4305	if ((a_selValue) & X86_SEL_LDT) \
4306	{ \
4307	uint32_t const uAttr = ASMGetSegAttr(a_selValue); \
4308	fValidSelector = RT_BOOL(uAttr != UINT32_MAX && (uAttr & X86_DESC_P)); \
4309	} \
4310	if (fValidSelector) \
4311	{ \
4312	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##a_Seg; \
4313	pVCpu->hm.s.vmx.RestoreHost.uHostSel##a_Seg = (a_selValue); \
4314	} \
4315	(a_selValue) = 0; \
4316	}
4317
4318	/*
4319	* If we've executed guest code using hardware-assisted VMX, the host-state bits
4320	* will be messed up. We should -not- save the messed up state without restoring
4321	* the original host-state, see @bugref{7240}.
4322	*
4323	* This apparently can happen (most likely the FPU changes), deal with it rather than
4324	* asserting. Was observed booting Solaris 10u10 32-bit guest.
4325	*/
4326	if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
4327	&& (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
4328	{
4329	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags,
4330	pVCpu->idCpu));
4331	VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
4332	}
4333	pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
4334
4335	/*
4336	* Host segment registers.
4337	*/
4338	RTSEL uSelES = ASMGetES();
4339	RTSEL uSelCS = ASMGetCS();
4340	RTSEL uSelSS = ASMGetSS();
4341	RTSEL uSelDS = ASMGetDS();
4342	RTSEL uSelFS = ASMGetFS();
4343	RTSEL uSelGS = ASMGetGS();
4344	RTSEL uSelTR = ASMGetTR();
4345
4346	/*
4347	* Determine if the host segment registers are suitable for VT-x. Otherwise use zero to
4348	* gain VM-entry and restore them before we get preempted.
4349	*
4350	* See Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers".
4351	*/
4352	VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS);
4353	VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES);
4354	VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS);
4355	VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS);
4356
4357	/* Verification based on Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers" */
4358	Assert(!(uSelCS & X86_SEL_RPL)); Assert(!(uSelCS & X86_SEL_LDT));
4359	Assert(!(uSelSS & X86_SEL_RPL)); Assert(!(uSelSS & X86_SEL_LDT));
4360	Assert(!(uSelDS & X86_SEL_RPL)); Assert(!(uSelDS & X86_SEL_LDT));
4361	Assert(!(uSelES & X86_SEL_RPL)); Assert(!(uSelES & X86_SEL_LDT));
4362	Assert(!(uSelFS & X86_SEL_RPL)); Assert(!(uSelFS & X86_SEL_LDT));
4363	Assert(!(uSelGS & X86_SEL_RPL)); Assert(!(uSelGS & X86_SEL_LDT));
4364	Assert(!(uSelTR & X86_SEL_RPL)); Assert(!(uSelTR & X86_SEL_LDT));
4365	Assert(uSelCS);
4366	Assert(uSelTR);
4367
4368	/* Write these host selector fields into the host-state area in the VMCS. */
4369	int rc = VMXWriteVmcs16(VMX_VMCS16_HOST_CS_SEL, uSelCS); AssertRC(rc);
4370	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_SS_SEL, uSelSS); AssertRC(rc);
4371	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_DS_SEL, uSelDS); AssertRC(rc);
4372	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_ES_SEL, uSelES); AssertRC(rc);
4373	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_FS_SEL, uSelFS); AssertRC(rc);
4374	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_GS_SEL, uSelGS); AssertRC(rc);
4375	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_TR_SEL, uSelTR); AssertRC(rc);
4376
4377	/*
4378	* Host GDTR and IDTR.
4379	*/
4380	RTGDTR Gdtr;
4381	RTIDTR Idtr;
4382	RT_ZERO(Gdtr);
4383	RT_ZERO(Idtr);
4384	ASMGetGDTR(&Gdtr);
4385	ASMGetIDTR(&Idtr);
4386	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_GDTR_BASE, Gdtr.pGdt); AssertRC(rc);
4387	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_IDTR_BASE, Idtr.pIdt); AssertRC(rc);
4388
4389	/*
4390	* Determine if we need to manually need to restore the GDTR and IDTR limits as VT-x zaps
4391	* them to the maximum limit (0xffff) on every VM-exit.
4392	*/
4393	if (Gdtr.cbGdt != 0xffff)
4394	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_GDTR;
4395
4396	/*
4397	* IDT limit is effectively capped at 0xfff. (See Intel spec. 6.14.1 "64-Bit Mode IDT" and
4398	* Intel spec. 6.2 "Exception and Interrupt Vectors".) Therefore if the host has the limit
4399	* as 0xfff, VT-x bloating the limit to 0xffff shouldn't cause any different CPU behavior.
4400	* However, several hosts either insists on 0xfff being the limit (Windows Patch Guard) or
4401	* uses the limit for other purposes (darwin puts the CPU ID in there but botches sidt
4402	* alignment in at least one consumer). So, we're only allowing the IDTR.LIMIT to be left
4403	* at 0xffff on hosts where we are sure it won't cause trouble.
4404	*/
4405	#if defined(RT_OS_LINUX) \|\| defined(RT_OS_SOLARIS)
4406	if (Idtr.cbIdt < 0x0fff)
4407	#else
4408	if (Idtr.cbIdt != 0xffff)
4409	#endif
4410	{
4411	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_IDTR;
4412	AssertCompile(sizeof(Idtr) == sizeof(X86XDTR64));
4413	memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostIdtr, &Idtr, sizeof(X86XDTR64));
4414	}
4415
4416	/*
4417	* Host TR base. Verify that TR selector doesn't point past the GDT. Masking off the TI
4418	* and RPL bits is effectively what the CPU does for "scaling by 8". TI is always 0 and
4419	* RPL should be too in most cases.
4420	*/
4421	AssertMsgReturn((uSelTR \| X86_SEL_RPL_LDT) <= Gdtr.cbGdt,
4422	("TR selector exceeds limit. TR=%RTsel cbGdt=%#x\n", uSelTR, Gdtr.cbGdt), VERR_VMX_INVALID_HOST_STATE);
4423
4424	PCX86DESCHC pDesc = (PCX86DESCHC)(Gdtr.pGdt + (uSelTR & X86_SEL_MASK));
4425	uintptr_t const uTRBase = X86DESC64_BASE(pDesc);
4426
4427	/*
4428	* VT-x unconditionally restores the TR limit to 0x67 and type to 11 (32-bit busy TSS) on
4429	* all VM-exits. The type is the same for 64-bit busy TSS[1]. The limit needs manual
4430	* restoration if the host has something else. Task switching is not supported in 64-bit
4431	* mode[2], but the limit still matters as IOPM is supported in 64-bit mode. Restoring the
4432	* limit lazily while returning to ring-3 is safe because IOPM is not applicable in ring-0.
4433	*
4434	* [1] See Intel spec. 3.5 "System Descriptor Types".
4435	* [2] See Intel spec. 7.2.3 "TSS Descriptor in 64-bit mode".
4436	*/
4437	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4438	Assert(pDesc->System.u4Type == 11);
4439	if ( pDesc->System.u16LimitLow != 0x67
4440	\|\| pDesc->System.u4LimitHigh)
4441	{
4442	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_TR;
4443	/* If the host has made GDT read-only, we would need to temporarily toggle CR0.WP before writing the GDT. */
4444	if (pVM->hm.s.fHostKernelFeatures & SUPKERNELFEATURES_GDT_READ_ONLY)
4445	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_READ_ONLY;
4446	pVCpu->hm.s.vmx.RestoreHost.uHostSelTR = uSelTR;
4447	}
4448
4449	/*
4450	* Store the GDTR as we need it when restoring the GDT and while restoring the TR.
4451	*/
4452	if (pVCpu->hm.s.vmx.fRestoreHostFlags & (VMX_RESTORE_HOST_GDTR \| VMX_RESTORE_HOST_SEL_TR))
4453	{
4454	AssertCompile(sizeof(Gdtr) == sizeof(X86XDTR64));
4455	memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64));
4456	if (pVM->hm.s.fHostKernelFeatures & SUPKERNELFEATURES_GDT_NEED_WRITABLE)
4457	{
4458	/* The GDT is read-only but the writable GDT is available. */
4459	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_NEED_WRITABLE;
4460	pVCpu->hm.s.vmx.RestoreHost.HostGdtrRw.cb = Gdtr.cbGdt;
4461	rc = SUPR0GetCurrentGdtRw(&pVCpu->hm.s.vmx.RestoreHost.HostGdtrRw.uAddr);
4462	AssertRCReturn(rc, rc);
4463	}
4464	}
4465
4466	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_TR_BASE, uTRBase);
4467	AssertRC(rc);
4468
4469	/*
4470	* Host FS base and GS base.
4471	*/
4472	uint64_t const u64FSBase = ASMRdMsr(MSR_K8_FS_BASE);
4473	uint64_t const u64GSBase = ASMRdMsr(MSR_K8_GS_BASE);
4474	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_FS_BASE, u64FSBase); AssertRC(rc);
4475	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_GS_BASE, u64GSBase); AssertRC(rc);
4476
4477	/* Store the base if we have to restore FS or GS manually as we need to restore the base as well. */
4478	if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_FS)
4479	pVCpu->hm.s.vmx.RestoreHost.uHostFSBase = u64FSBase;
4480	if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_GS)
4481	pVCpu->hm.s.vmx.RestoreHost.uHostGSBase = u64GSBase;
4482
4483	return VINF_SUCCESS;
4484	#undef VMXLOCAL_ADJUST_HOST_SEG
4485	}
4486
4487
4488	/**
4489	* Exports certain host MSRs in the VM-exit MSR-load area and some in the
4490	* host-state area of the VMCS.
4491	*
4492	* These MSRs will be automatically restored on the host after every successful
4493	* VM-exit.
4494	*
4495	* @param pVCpu The cross context virtual CPU structure.
4496	*
4497	* @remarks No-long-jump zone!!!
4498	*/
4499	static void hmR0VmxExportHostMsrs(PVMCPUCC pVCpu)
4500	{
4501	AssertPtr(pVCpu);
4502
4503	/*
4504	* Save MSRs that we restore lazily (due to preemption or transition to ring-3)
4505	* rather than swapping them on every VM-entry.
4506	*/
4507	hmR0VmxLazySaveHostMsrs(pVCpu);
4508
4509	/*
4510	* Host Sysenter MSRs.
4511	*/
4512	int rc = VMXWriteVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, ASMRdMsr_Low(MSR_IA32_SYSENTER_CS)); AssertRC(rc);
4513	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP)); AssertRC(rc);
4514	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP)); AssertRC(rc);
4515
4516	/*
4517	* Host EFER MSR.
4518	*
4519	* If the CPU supports the newer VMCS controls for managing EFER, use it. Otherwise it's
4520	* done as part of auto-load/store MSR area in the VMCS, see hmR0VmxExportGuestMsrs().
4521	*/
4522	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4523	if (pVM->hm.s.vmx.fSupportsVmcsEfer)
4524	{
4525	rc = VMXWriteVmcs64(VMX_VMCS64_HOST_EFER_FULL, pVM->hm.s.vmx.u64HostMsrEfer);
4526	AssertRC(rc);
4527	}
4528
4529	/** @todo IA32_PERF_GLOBALCTRL, IA32_PAT also see
4530	* hmR0VmxExportGuestEntryExitCtls(). */
4531	}
4532
4533
4534	/**
4535	* Figures out if we need to swap the EFER MSR which is particularly expensive.
4536	*
4537	* We check all relevant bits. For now, that's everything besides LMA/LME, as
4538	* these two bits are handled by VM-entry, see hmR0VMxExportGuestEntryExitCtls().
4539	*
4540	* @returns true if we need to load guest EFER, false otherwise.
4541	* @param pVCpu The cross context virtual CPU structure.
4542	*
4543	* @remarks Requires EFER, CR4.
4544	* @remarks No-long-jump zone!!!
4545	*/
4546	static bool hmR0VmxShouldSwapEferMsr(PCVMCPUCC pVCpu)
4547	{
4548	#ifdef HMVMX_ALWAYS_SWAP_EFER
4549	RT_NOREF(pVCpu);
4550	return true;
4551	#else
4552	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
4553	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4554	uint64_t const u64HostEfer = pVM->hm.s.vmx.u64HostMsrEfer;
4555	uint64_t const u64GuestEfer = pCtx->msrEFER;
4556
4557	/*
4558	* For 64-bit guests, if EFER.SCE bit differs, we need to swap the EFER MSR
4559	* to ensure that the guest's SYSCALL behaviour isn't broken, see @bugref{7386}.
4560	*/
4561	if ( CPUMIsGuestInLongModeEx(pCtx)
4562	&& (u64GuestEfer & MSR_K6_EFER_SCE) != (u64HostEfer & MSR_K6_EFER_SCE))
4563	return true;
4564
4565	/*
4566	* If the guest uses PAE and EFER.NXE bit differs, we need to swap the EFER MSR
4567	* as it affects guest paging. 64-bit paging implies CR4.PAE as well.
4568	*
4569	* See Intel spec. 4.5 "IA-32e Paging".
4570	* See Intel spec. 4.1.1 "Three Paging Modes".
4571	*
4572	* Verify that we always intercept CR4.PAE and CR0.PG bits, so we don't need to
4573	* import CR4 and CR0 from the VMCS here as those bits are always up to date.
4574	*/
4575	Assert(hmR0VmxGetFixedCr4Mask(pVCpu) & X86_CR4_PAE);
4576	Assert(hmR0VmxGetFixedCr0Mask(pVCpu) & X86_CR0_PG);
4577	if ( (pCtx->cr4 & X86_CR4_PAE)
4578	&& (pCtx->cr0 & X86_CR0_PG)
4579	&& (u64GuestEfer & MSR_K6_EFER_NXE) != (u64HostEfer & MSR_K6_EFER_NXE))
4580	{
4581	/* Assert that host is NX capable. */
4582	Assert(pVCpu->CTX_SUFF(pVM)->cpum.ro.HostFeatures.fNoExecute);
4583	return true;
4584	}
4585
4586	return false;
4587	#endif
4588	}
4589
4590	/**
4591	* Exports the guest state with appropriate VM-entry and VM-exit controls in the
4592	* VMCS.
4593	*
4594	* This is typically required when the guest changes paging mode.
4595	*
4596	* @returns VBox status code.
4597	* @param pVCpu The cross context virtual CPU structure.
4598	* @param pVmxTransient The VMX-transient structure.
4599	*
4600	* @remarks Requires EFER.
4601	* @remarks No-long-jump zone!!!
4602	*/
4603	static int hmR0VmxExportGuestEntryExitCtls(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
4604	{
4605	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_ENTRY_EXIT_CTLS)
4606	{
4607	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4608	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
4609
4610	/*
4611	* VM-entry controls.
4612	*/
4613	{
4614	uint32_t fVal = pVM->hm.s.vmx.Msrs.EntryCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
4615	uint32_t const fZap = pVM->hm.s.vmx.Msrs.EntryCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
4616
4617	/*
4618	* Load the guest debug controls (DR7 and IA32_DEBUGCTL MSR) on VM-entry.
4619	* The first VT-x capable CPUs only supported the 1-setting of this bit.
4620	*
4621	* For nested-guests, this is a mandatory VM-entry control. It's also
4622	* required because we do not want to leak host bits to the nested-guest.
4623	*/
4624	fVal \|= VMX_ENTRY_CTLS_LOAD_DEBUG;
4625
4626	/*
4627	* Set if the guest is in long mode. This will set/clear the EFER.LMA bit on VM-entry.
4628	*
4629	* For nested-guests, the "IA-32e mode guest" control we initialize with what is
4630	* required to get the nested-guest working with hardware-assisted VMX execution.
4631	* It depends on the nested-guest's IA32_EFER.LMA bit. Remember, a guest hypervisor
4632	* can skip intercepting changes to the EFER MSR. This is why it it needs to be done
4633	* here rather than while merging the guest VMCS controls.
4634	*/
4635	if (CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
4636	fVal \|= VMX_ENTRY_CTLS_IA32E_MODE_GUEST;
4637	else
4638	Assert(!(fVal & VMX_ENTRY_CTLS_IA32E_MODE_GUEST));
4639
4640	/*
4641	* If the CPU supports the newer VMCS controls for managing guest/host EFER, use it.
4642	*
4643	* For nested-guests, we use the "load IA32_EFER" if the hardware supports it,
4644	* regardless of whether the nested-guest VMCS specifies it because we are free to
4645	* load whatever MSRs we require and we do not need to modify the guest visible copy
4646	* of the VM-entry MSR load area.
4647	*/
4648	if ( pVM->hm.s.vmx.fSupportsVmcsEfer
4649	&& hmR0VmxShouldSwapEferMsr(pVCpu))
4650	fVal \|= VMX_ENTRY_CTLS_LOAD_EFER_MSR;
4651	else
4652	Assert(!(fVal & VMX_ENTRY_CTLS_LOAD_EFER_MSR));
4653
4654	/*
4655	* The following should -not- be set (since we're not in SMM mode):
4656	* - VMX_ENTRY_CTLS_ENTRY_TO_SMM
4657	* - VMX_ENTRY_CTLS_DEACTIVATE_DUAL_MON
4658	*/
4659
4660	/** @todo VMX_ENTRY_CTLS_LOAD_PERF_MSR,
4661	* VMX_ENTRY_CTLS_LOAD_PAT_MSR. */
4662
4663	if ((fVal & fZap) == fVal)
4664	{ /* likely */ }
4665	else
4666	{
4667	Log4Func(("Invalid VM-entry controls combo! Cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
4668	pVM->hm.s.vmx.Msrs.EntryCtls.n.allowed0, fVal, fZap));
4669	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_ENTRY;
4670	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
4671	}
4672
4673	/* Commit it to the VMCS. */
4674	if (pVmcsInfo->u32EntryCtls != fVal)
4675	{
4676	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY, fVal);
4677	AssertRC(rc);
4678	pVmcsInfo->u32EntryCtls = fVal;
4679	}
4680	}
4681
4682	/*
4683	* VM-exit controls.
4684	*/
4685	{
4686	uint32_t fVal = pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
4687	uint32_t const fZap = pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
4688
4689	/*
4690	* Save debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x CPUs only
4691	* supported the 1-setting of this bit.
4692	*
4693	* For nested-guests, we set the "save debug controls" as the converse
4694	* "load debug controls" is mandatory for nested-guests anyway.
4695	*/
4696	fVal \|= VMX_EXIT_CTLS_SAVE_DEBUG;
4697
4698	/*
4699	* Set the host long mode active (EFER.LMA) bit (which Intel calls
4700	* "Host address-space size") if necessary. On VM-exit, VT-x sets both the
4701	* host EFER.LMA and EFER.LME bit to this value. See assertion in
4702	* hmR0VmxExportHostMsrs().
4703	*
4704	* For nested-guests, we always set this bit as we do not support 32-bit
4705	* hosts.
4706	*/
4707	fVal \|= VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE;
4708
4709	/*
4710	* If the VMCS EFER MSR fields are supported by the hardware, we use it.
4711	*
4712	* For nested-guests, we should use the "save IA32_EFER" control if we also
4713	* used the "load IA32_EFER" control while exporting VM-entry controls.
4714	*/
4715	if ( pVM->hm.s.vmx.fSupportsVmcsEfer
4716	&& hmR0VmxShouldSwapEferMsr(pVCpu))
4717	{
4718	fVal \|= VMX_EXIT_CTLS_SAVE_EFER_MSR
4719	\| VMX_EXIT_CTLS_LOAD_EFER_MSR;
4720	}
4721
4722	/*
4723	* Enable saving of the VMX-preemption timer value on VM-exit.
4724	* For nested-guests, currently not exposed/used.
4725	*/
4726	if ( pVM->hm.s.vmx.fUsePreemptTimer
4727	&& (pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER))
4728	fVal \|= VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER;
4729
4730	/* Don't acknowledge external interrupts on VM-exit. We want to let the host do that. */
4731	Assert(!(fVal & VMX_EXIT_CTLS_ACK_EXT_INT));
4732
4733	/** @todo VMX_EXIT_CTLS_LOAD_PERF_MSR,
4734	* VMX_EXIT_CTLS_SAVE_PAT_MSR,
4735	* VMX_EXIT_CTLS_LOAD_PAT_MSR. */
4736
4737	if ((fVal & fZap) == fVal)
4738	{ /* likely */ }
4739	else
4740	{
4741	Log4Func(("Invalid VM-exit controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%R#X32\n",
4742	pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed0, fVal, fZap));
4743	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_EXIT;
4744	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
4745	}
4746
4747	/* Commit it to the VMCS. */
4748	if (pVmcsInfo->u32ExitCtls != fVal)
4749	{
4750	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT, fVal);
4751	AssertRC(rc);
4752	pVmcsInfo->u32ExitCtls = fVal;
4753	}
4754	}
4755
4756	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
4757	}
4758	return VINF_SUCCESS;
4759	}
4760
4761
4762	/**
4763	* Sets the TPR threshold in the VMCS.
4764	*
4765	* @param pVmcsInfo The VMCS info. object.
4766	* @param u32TprThreshold The TPR threshold (task-priority class only).
4767	*/
4768	DECLINLINE(void) hmR0VmxApicSetTprThreshold(PVMXVMCSINFO pVmcsInfo, uint32_t u32TprThreshold)
4769	{
4770	Assert(!(u32TprThreshold & ~VMX_TPR_THRESHOLD_MASK)); /* Bits 31:4 MBZ. */
4771	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
4772	RT_NOREF(pVmcsInfo);
4773	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
4774	AssertRC(rc);
4775	}
4776
4777
4778	/**
4779	* Exports the guest APIC TPR state into the VMCS.
4780	*
4781	* @returns VBox status code.
4782	* @param pVCpu The cross context virtual CPU structure.
4783	* @param pVmxTransient The VMX-transient structure.
4784	*
4785	* @remarks No-long-jump zone!!!
4786	*/
4787	static int hmR0VmxExportGuestApicTpr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
4788	{
4789	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_APIC_TPR)
4790	{
4791	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_APIC_TPR);
4792
4793	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
4794	if (!pVmxTransient->fIsNestedGuest)
4795	{
4796	if ( PDMHasApic(pVCpu->CTX_SUFF(pVM))
4797	&& APICIsEnabled(pVCpu))
4798	{
4799	/*
4800	* Setup TPR shadowing.
4801	*/
4802	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
4803	{
4804	bool fPendingIntr = false;
4805	uint8_t u8Tpr = 0;
4806	uint8_t u8PendingIntr = 0;
4807	int rc = APICGetTpr(pVCpu, &u8Tpr, &fPendingIntr, &u8PendingIntr);
4808	AssertRCReturn(rc, rc);
4809
4810	/*
4811	* If there are interrupts pending but masked by the TPR, instruct VT-x to
4812	* cause a TPR-below-threshold VM-exit when the guest lowers its TPR below the
4813	* priority of the pending interrupt so we can deliver the interrupt. If there
4814	* are no interrupts pending, set threshold to 0 to not cause any
4815	* TPR-below-threshold VM-exits.
4816	*/
4817	uint32_t u32TprThreshold = 0;
4818	if (fPendingIntr)
4819	{
4820	/* Bits 3:0 of the TPR threshold field correspond to bits 7:4 of the TPR
4821	(which is the Task-Priority Class). */
4822	const uint8_t u8PendingPriority = u8PendingIntr >> 4;
4823	const uint8_t u8TprPriority = u8Tpr >> 4;
4824	if (u8PendingPriority <= u8TprPriority)
4825	u32TprThreshold = u8PendingPriority;
4826	}
4827
4828	hmR0VmxApicSetTprThreshold(pVmcsInfo, u32TprThreshold);
4829	}
4830	}
4831	}
4832	/* else: the TPR threshold has already been updated while merging the nested-guest VMCS. */
4833	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_APIC_TPR);
4834	}
4835	return VINF_SUCCESS;
4836	}
4837
4838
4839	/**
4840	* Gets the guest interruptibility-state.
4841	*
4842	* @returns Guest's interruptibility-state.
4843	* @param pVCpu The cross context virtual CPU structure.
4844	* @param pVmxTransient The VMX-transient structure.
4845	*
4846	* @remarks No-long-jump zone!!!
4847	*/
4848	static uint32_t hmR0VmxGetGuestIntrState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
4849	{
4850	/*
4851	* Check if we should inhibit interrupt delivery due to instructions like STI and MOV SS.
4852	*/
4853	uint32_t fIntrState = 0;
4854	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
4855	{
4856	/* If inhibition is active, RIP and RFLAGS should've been imported from the VMCS already. */
4857	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS);
4858
4859	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
4860	if (pCtx->rip == EMGetInhibitInterruptsPC(pVCpu))
4861	{
4862	if (pCtx->eflags.Bits.u1IF)
4863	fIntrState = VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
4864	else
4865	fIntrState = VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS;
4866	}
4867	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
4868	{
4869	/*
4870	* We can clear the inhibit force flag as even if we go back to the recompiler
4871	* without executing guest code in VT-x, the flag's condition to be cleared is
4872	* met and thus the cleared state is correct.
4873	*/
4874	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
4875	}
4876	}
4877
4878	/*
4879	* NMIs to the guest are blocked after an NMI is injected until the guest executes an IRET. We only
4880	* bother with virtual-NMI blocking when we have support for virtual NMIs in the CPU, otherwise
4881	* setting this would block host-NMIs and IRET will not clear the blocking.
4882	*
4883	* We always set NMI-exiting so when the host receives an NMI we get a VM-exit.
4884	*
4885	* See Intel spec. 26.6.1 "Interruptibility state". See @bugref{7445}.
4886	*/
4887	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
4888	if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
4889	&& CPUMIsGuestNmiBlocking(pVCpu))
4890	fIntrState \|= VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI;
4891
4892	return fIntrState;
4893	}
4894
4895
4896	/**
4897	* Exports the exception intercepts required for guest execution in the VMCS.
4898	*
4899	* @returns VBox status code.
4900	* @param pVCpu The cross context virtual CPU structure.
4901	* @param pVmxTransient The VMX-transient structure.
4902	*
4903	* @remarks No-long-jump zone!!!
4904	*/
4905	static int hmR0VmxExportGuestXcptIntercepts(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
4906	{
4907	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_XCPT_INTERCEPTS)
4908	{
4909	/* When executing a nested-guest, we do not need to trap GIM hypercalls by intercepting #UD. */
4910	if ( !pVmxTransient->fIsNestedGuest
4911	&& pVCpu->hm.s.fGIMTrapXcptUD)
4912	hmR0VmxAddXcptIntercept(pVmxTransient, X86_XCPT_UD);
4913	else
4914	hmR0VmxRemoveXcptIntercept(pVCpu, pVmxTransient, X86_XCPT_UD);
4915
4916	/* Other exception intercepts are handled elsewhere, e.g. while exporting guest CR0. */
4917	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_XCPT_INTERCEPTS);
4918	}
4919	return VINF_SUCCESS;
4920	}
4921
4922
4923	/**
4924	* Exports the guest's RIP into the guest-state area in the VMCS.
4925	*
4926	* @returns VBox status code.
4927	* @param pVCpu The cross context virtual CPU structure.
4928	*
4929	* @remarks No-long-jump zone!!!
4930	*/
4931	static int hmR0VmxExportGuestRip(PVMCPUCC pVCpu)
4932	{
4933	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RIP)
4934	{
4935	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP);
4936
4937	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RIP, pVCpu->cpum.GstCtx.rip);
4938	AssertRC(rc);
4939
4940	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RIP);
4941	Log4Func(("rip=%#RX64\n", pVCpu->cpum.GstCtx.rip));
4942	}
4943	return VINF_SUCCESS;
4944	}
4945
4946
4947	/**
4948	* Exports the guest's RSP into the guest-state area in the VMCS.
4949	*
4950	* @returns VBox status code.
4951	* @param pVCpu The cross context virtual CPU structure.
4952	*
4953	* @remarks No-long-jump zone!!!
4954	*/
4955	static int hmR0VmxExportGuestRsp(PVMCPUCC pVCpu)
4956	{
4957	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RSP)
4958	{
4959	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RSP);
4960
4961	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RSP, pVCpu->cpum.GstCtx.rsp);
4962	AssertRC(rc);
4963
4964	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RSP);
4965	Log4Func(("rsp=%#RX64\n", pVCpu->cpum.GstCtx.rsp));
4966	}
4967	return VINF_SUCCESS;
4968	}
4969
4970
4971	/**
4972	* Exports the guest's RFLAGS into the guest-state area in the VMCS.
4973	*
4974	* @returns VBox status code.
4975	* @param pVCpu The cross context virtual CPU structure.
4976	* @param pVmxTransient The VMX-transient structure.
4977	*
4978	* @remarks No-long-jump zone!!!
4979	*/
4980	static int hmR0VmxExportGuestRflags(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
4981	{
4982	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RFLAGS)
4983	{
4984	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
4985
4986	/* Intel spec. 2.3.1 "System Flags and Fields in IA-32e Mode" claims the upper 32-bits of RFLAGS are reserved (MBZ).
4987	Let us assert it as such and use 32-bit VMWRITE. */
4988	Assert(!RT_HI_U32(pVCpu->cpum.GstCtx.rflags.u64));
4989	X86EFLAGS fEFlags = pVCpu->cpum.GstCtx.eflags;
4990	Assert(fEFlags.u32 & X86_EFL_RA1_MASK);
4991	Assert(!(fEFlags.u32 & ~(X86_EFL_1 \| X86_EFL_LIVE_MASK)));
4992
4993	/*
4994	* If we're emulating real-mode using Virtual 8086 mode, save the real-mode eflags so
4995	* we can restore them on VM-exit. Modify the real-mode guest's eflags so that VT-x
4996	* can run the real-mode guest code under Virtual 8086 mode.
4997	*/
4998	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
4999	if (pVmcsInfo->RealMode.fRealOnV86Active)
5000	{
5001	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
5002	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
5003	Assert(!pVmxTransient->fIsNestedGuest);
5004	pVmcsInfo->RealMode.Eflags.u32 = fEFlags.u32; /* Save the original eflags of the real-mode guest. */
5005	fEFlags.Bits.u1VM = 1; /* Set the Virtual 8086 mode bit. */
5006	fEFlags.Bits.u2IOPL = 0; /* Change IOPL to 0, otherwise certain instructions won't fault. */
5007	}
5008
5009	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RFLAGS, fEFlags.u32);
5010	AssertRC(rc);
5011
5012	/*
5013	* Setup pending debug exceptions if the guest is single-stepping using EFLAGS.TF.
5014	*
5015	* We must avoid setting any automatic debug exceptions delivery when single-stepping
5016	* through the hypervisor debugger using EFLAGS.TF.
5017	*/
5018	if ( !pVmxTransient->fIsNestedGuest
5019	&& !pVCpu->hm.s.fSingleInstruction
5020	&& fEFlags.Bits.u1TF)
5021	{
5022	/** @todo r=ramshankar: Warning!! We ASSUME EFLAGS.TF will not cleared on
5023	* premature trips to ring-3 esp since IEM does not yet handle it. */
5024	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, VMX_VMCS_GUEST_PENDING_DEBUG_XCPT_BS);
5025	AssertRC(rc);
5026	}
5027	/* else: for nested-guest currently handling while merging controls. */
5028
5029	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RFLAGS);
5030	Log4Func(("eflags=%#RX32\n", fEFlags.u32));
5031	}
5032	return VINF_SUCCESS;
5033	}
5034
5035
5036	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5037	/**
5038	* Copies the nested-guest VMCS to the shadow VMCS.
5039	*
5040	* @returns VBox status code.
5041	* @param pVCpu The cross context virtual CPU structure.
5042	* @param pVmcsInfo The VMCS info. object.
5043	*
5044	* @remarks No-long-jump zone!!!
5045	*/
5046	static int hmR0VmxCopyNstGstToShadowVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
5047	{
5048	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5049	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5050
5051	/*
5052	* Disable interrupts so we don't get preempted while the shadow VMCS is the
5053	* current VMCS, as we may try saving guest lazy MSRs.
5054	*
5055	* Strictly speaking the lazy MSRs are not in the VMCS, but I'd rather not risk
5056	* calling the import VMCS code which is currently performing the guest MSR reads
5057	* (on 64-bit hosts) and accessing the auto-load/store MSR area on 32-bit hosts
5058	* and the rest of the VMX leave session machinery.
5059	*/
5060	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
5061
5062	int rc = hmR0VmxLoadShadowVmcs(pVmcsInfo);
5063	if (RT_SUCCESS(rc))
5064	{
5065	/*
5066	* Copy all guest read/write VMCS fields.
5067	*
5068	* We don't check for VMWRITE failures here for performance reasons and
5069	* because they are not expected to fail, barring irrecoverable conditions
5070	* like hardware errors.
5071	*/
5072	uint32_t const cShadowVmcsFields = pVM->hm.s.vmx.cShadowVmcsFields;
5073	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
5074	{
5075	uint64_t u64Val;
5076	uint32_t const uVmcsField = pVM->hm.s.vmx.paShadowVmcsFields[i];
5077	IEMReadVmxVmcsField(pVmcsNstGst, uVmcsField, &u64Val);
5078	VMXWriteVmcs64(uVmcsField, u64Val);
5079	}
5080
5081	/*
5082	* If the host CPU supports writing all VMCS fields, copy the guest read-only
5083	* VMCS fields, so the guest can VMREAD them without causing a VM-exit.
5084	*/
5085	if (pVM->hm.s.vmx.Msrs.u64Misc & VMX_MISC_VMWRITE_ALL)
5086	{
5087	uint32_t const cShadowVmcsRoFields = pVM->hm.s.vmx.cShadowVmcsRoFields;
5088	for (uint32_t i = 0; i < cShadowVmcsRoFields; i++)
5089	{
5090	uint64_t u64Val;
5091	uint32_t const uVmcsField = pVM->hm.s.vmx.paShadowVmcsRoFields[i];
5092	IEMReadVmxVmcsField(pVmcsNstGst, uVmcsField, &u64Val);
5093	VMXWriteVmcs64(uVmcsField, u64Val);
5094	}
5095	}
5096
5097	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
5098	rc \|= hmR0VmxLoadVmcs(pVmcsInfo);
5099	}
5100
5101	ASMSetFlags(fEFlags);
5102	return rc;
5103	}
5104
5105
5106	/**
5107	* Copies the shadow VMCS to the nested-guest VMCS.
5108	*
5109	* @returns VBox status code.
5110	* @param pVCpu The cross context virtual CPU structure.
5111	* @param pVmcsInfo The VMCS info. object.
5112	*
5113	* @remarks Called with interrupts disabled.
5114	*/
5115	static int hmR0VmxCopyShadowToNstGstVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
5116	{
5117	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
5118	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5119	PVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5120
5121	int rc = hmR0VmxLoadShadowVmcs(pVmcsInfo);
5122	if (RT_SUCCESS(rc))
5123	{
5124	/*
5125	* Copy guest read/write fields from the shadow VMCS.
5126	* Guest read-only fields cannot be modified, so no need to copy them.
5127	*
5128	* We don't check for VMREAD failures here for performance reasons and
5129	* because they are not expected to fail, barring irrecoverable conditions
5130	* like hardware errors.
5131	*/
5132	uint32_t const cShadowVmcsFields = pVM->hm.s.vmx.cShadowVmcsFields;
5133	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
5134	{
5135	uint64_t u64Val;
5136	uint32_t const uVmcsField = pVM->hm.s.vmx.paShadowVmcsFields[i];
5137	VMXReadVmcs64(uVmcsField, &u64Val);
5138	IEMWriteVmxVmcsField(pVmcsNstGst, uVmcsField, u64Val);
5139	}
5140
5141	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
5142	rc \|= hmR0VmxLoadVmcs(pVmcsInfo);
5143	}
5144	return rc;
5145	}
5146
5147
5148	/**
5149	* Enables VMCS shadowing for the given VMCS info. object.
5150	*
5151	* @param pVmcsInfo The VMCS info. object.
5152	*
5153	* @remarks No-long-jump zone!!!
5154	*/
5155	static void hmR0VmxEnableVmcsShadowing(PVMXVMCSINFO pVmcsInfo)
5156	{
5157	uint32_t uProcCtls2 = pVmcsInfo->u32ProcCtls2;
5158	if (!(uProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING))
5159	{
5160	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
5161	uProcCtls2 \|= VMX_PROC_CTLS2_VMCS_SHADOWING;
5162	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, uProcCtls2); AssertRC(rc);
5163	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, pVmcsInfo->HCPhysShadowVmcs); AssertRC(rc);
5164	pVmcsInfo->u32ProcCtls2 = uProcCtls2;
5165	pVmcsInfo->u64VmcsLinkPtr = pVmcsInfo->HCPhysShadowVmcs;
5166	Log4Func(("Enabled\n"));
5167	}
5168	}
5169
5170
5171	/**
5172	* Disables VMCS shadowing for the given VMCS info. object.
5173	*
5174	* @param pVmcsInfo The VMCS info. object.
5175	*
5176	* @remarks No-long-jump zone!!!
5177	*/
5178	static void hmR0VmxDisableVmcsShadowing(PVMXVMCSINFO pVmcsInfo)
5179	{
5180	/*
5181	* We want all VMREAD and VMWRITE instructions to cause VM-exits, so we clear the
5182	* VMCS shadowing control. However, VM-entry requires the shadow VMCS indicator bit
5183	* to match the VMCS shadowing control if the VMCS link pointer is not NIL_RTHCPHYS.
5184	* Hence, we must also reset the VMCS link pointer to ensure VM-entry does not fail.
5185	*
5186	* See Intel spec. 26.2.1.1 "VM-Execution Control Fields".
5187	* See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
5188	*/
5189	uint32_t uProcCtls2 = pVmcsInfo->u32ProcCtls2;
5190	if (uProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
5191	{
5192	uProcCtls2 &= ~VMX_PROC_CTLS2_VMCS_SHADOWING;
5193	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, uProcCtls2); AssertRC(rc);
5194	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS); AssertRC(rc);
5195	pVmcsInfo->u32ProcCtls2 = uProcCtls2;
5196	pVmcsInfo->u64VmcsLinkPtr = NIL_RTHCPHYS;
5197	Log4Func(("Disabled\n"));
5198	}
5199	}
5200	#endif
5201
5202
5203	/**
5204	* Exports the guest hardware-virtualization state.
5205	*
5206	* @returns VBox status code.
5207	* @param pVCpu The cross context virtual CPU structure.
5208	* @param pVmxTransient The VMX-transient structure.
5209	*
5210	* @remarks No-long-jump zone!!!
5211	*/
5212	static int hmR0VmxExportGuestHwvirtState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
5213	{
5214	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_HWVIRT)
5215	{
5216	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5217	/*
5218	* Check if the VMX feature is exposed to the guest and if the host CPU supports
5219	* VMCS shadowing.
5220	*/
5221	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUseVmcsShadowing)
5222	{
5223	/*
5224	* If the guest hypervisor has loaded a current VMCS and is in VMX root mode,
5225	* copy the guest hypervisor's current VMCS into the shadow VMCS and enable
5226	* VMCS shadowing to skip intercepting some or all VMREAD/VMWRITE VM-exits.
5227	*
5228	* We check for VMX root mode here in case the guest executes VMXOFF without
5229	* clearing the current VMCS pointer and our VMXOFF instruction emulation does
5230	* not clear the current VMCS pointer.
5231	*/
5232	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5233	if ( CPUMIsGuestInVmxRootMode(&pVCpu->cpum.GstCtx)
5234	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx)
5235	&& CPUMIsGuestVmxCurrentVmcsValid(pVCpu, &pVCpu->cpum.GstCtx))
5236	{
5237	/* Paranoia. */
5238	Assert(!pVmxTransient->fIsNestedGuest);
5239
5240	/*
5241	* For performance reasons, also check if the guest hypervisor's current VMCS
5242	* was newly loaded or modified before copying it to the shadow VMCS.
5243	*/
5244	if (!pVCpu->hm.s.vmx.fCopiedNstGstToShadowVmcs)
5245	{
5246	int rc = hmR0VmxCopyNstGstToShadowVmcs(pVCpu, pVmcsInfo);
5247	AssertRCReturn(rc, rc);
5248	pVCpu->hm.s.vmx.fCopiedNstGstToShadowVmcs = true;
5249	}
5250	hmR0VmxEnableVmcsShadowing(pVmcsInfo);
5251	}
5252	else
5253	hmR0VmxDisableVmcsShadowing(pVmcsInfo);
5254	}
5255	#else
5256	NOREF(pVmxTransient);
5257	#endif
5258	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_HWVIRT);
5259	}
5260	return VINF_SUCCESS;
5261	}
5262
5263
5264	/**
5265	* Exports the guest CR0 control register into the guest-state area in the VMCS.
5266	*
5267	* The guest FPU state is always pre-loaded hence we don't need to bother about
5268	* sharing FPU related CR0 bits between the guest and host.
5269	*
5270	* @returns VBox status code.
5271	* @param pVCpu The cross context virtual CPU structure.
5272	* @param pVmxTransient The VMX-transient structure.
5273	*
5274	* @remarks No-long-jump zone!!!
5275	*/
5276	static int hmR0VmxExportGuestCR0(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
5277	{
5278	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR0)
5279	{
5280	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5281	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5282
5283	/*
5284	* Figure out fixed CR0 bits in VMX operation.
5285	*/
5286	uint64_t fSetCr0 = pVM->hm.s.vmx.Msrs.u64Cr0Fixed0;
5287	uint64_t const fZapCr0 = pVM->hm.s.vmx.Msrs.u64Cr0Fixed1;
5288	if (pVM->hm.s.vmx.fUnrestrictedGuest)
5289	fSetCr0 &= ~(uint64_t)(X86_CR0_PE \| X86_CR0_PG);
5290	else
5291	Assert((fSetCr0 & (X86_CR0_PE \| X86_CR0_PG)) == (X86_CR0_PE \| X86_CR0_PG));
5292
5293	if (!pVmxTransient->fIsNestedGuest)
5294	{
5295	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
5296	uint64_t u64GuestCr0 = pVCpu->cpum.GstCtx.cr0;
5297	uint64_t const u64ShadowCr0 = u64GuestCr0;
5298	Assert(!RT_HI_U32(u64GuestCr0));
5299
5300	/*
5301	* Setup VT-x's view of the guest CR0.
5302	*/
5303	uint32_t uProcCtls = pVmcsInfo->u32ProcCtls;
5304	if (pVM->hm.s.fNestedPaging)
5305	{
5306	if (CPUMIsGuestPagingEnabled(pVCpu))
5307	{
5308	/* The guest has paging enabled, let it access CR3 without causing a VM-exit if supported. */
5309	uProcCtls &= ~( VMX_PROC_CTLS_CR3_LOAD_EXIT
5310	\| VMX_PROC_CTLS_CR3_STORE_EXIT);
5311	}
5312	else
5313	{
5314	/* The guest doesn't have paging enabled, make CR3 access cause a VM-exit to update our shadow. */
5315	uProcCtls \|= VMX_PROC_CTLS_CR3_LOAD_EXIT
5316	\| VMX_PROC_CTLS_CR3_STORE_EXIT;
5317	}
5318
5319	/* If we have unrestricted guest execution, we never have to intercept CR3 reads. */
5320	if (pVM->hm.s.vmx.fUnrestrictedGuest)
5321	uProcCtls &= ~VMX_PROC_CTLS_CR3_STORE_EXIT;
5322	}
5323	else
5324	{
5325	/* Guest CPL 0 writes to its read-only pages should cause a #PF VM-exit. */
5326	u64GuestCr0 \|= X86_CR0_WP;
5327	}
5328
5329	/*
5330	* Guest FPU bits.
5331	*
5332	* Since we pre-load the guest FPU always before VM-entry there is no need to track lazy state
5333	* using CR0.TS.
5334	*
5335	* Intel spec. 23.8 "Restrictions on VMX operation" mentions that CR0.NE bit must always be
5336	* set on the first CPUs to support VT-x and no mention of with regards to UX in VM-entry checks.
5337	*/
5338	u64GuestCr0 \|= X86_CR0_NE;
5339
5340	/* If CR0.NE isn't set, we need to intercept #MF exceptions and report them to the guest differently. */
5341	bool const fInterceptMF = !(u64ShadowCr0 & X86_CR0_NE);
5342
5343	/*
5344	* Update exception intercepts.
5345	*/
5346	uint32_t uXcptBitmap = pVmcsInfo->u32XcptBitmap;
5347	if (pVmcsInfo->RealMode.fRealOnV86Active)
5348	{
5349	Assert(PDMVmmDevHeapIsEnabled(pVM));
5350	Assert(pVM->hm.s.vmx.pRealModeTSS);
5351	uXcptBitmap \|= HMVMX_REAL_MODE_XCPT_MASK;
5352	}
5353	else
5354	{
5355	/* For now, cleared here as mode-switches can happen outside HM/VT-x. See @bugref{7626#c11}. */
5356	uXcptBitmap &= ~HMVMX_REAL_MODE_XCPT_MASK;
5357	if (fInterceptMF)
5358	uXcptBitmap \|= RT_BIT(X86_XCPT_MF);
5359	}
5360
5361	/* Additional intercepts for debugging, define these yourself explicitly. */
5362	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
5363	uXcptBitmap \|= 0
5364	\| RT_BIT(X86_XCPT_BP)
5365	\| RT_BIT(X86_XCPT_DE)
5366	\| RT_BIT(X86_XCPT_NM)
5367	\| RT_BIT(X86_XCPT_TS)
5368	\| RT_BIT(X86_XCPT_UD)
5369	\| RT_BIT(X86_XCPT_NP)
5370	\| RT_BIT(X86_XCPT_SS)
5371	\| RT_BIT(X86_XCPT_GP)
5372	\| RT_BIT(X86_XCPT_PF)
5373	\| RT_BIT(X86_XCPT_MF)
5374	;
5375	#elif defined(HMVMX_ALWAYS_TRAP_PF)
5376	uXcptBitmap \|= RT_BIT(X86_XCPT_PF);
5377	#endif
5378	if (pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv)
5379	uXcptBitmap \|= RT_BIT(X86_XCPT_GP);
5380	Assert(pVM->hm.s.fNestedPaging \|\| (uXcptBitmap & RT_BIT(X86_XCPT_PF)));
5381
5382	/* Apply the hardware specified fixed CR0 bits and enable caching. */
5383	u64GuestCr0 \|= fSetCr0;
5384	u64GuestCr0 &= fZapCr0;
5385	u64GuestCr0 &= ~(uint64_t)(X86_CR0_CD \| X86_CR0_NW);
5386
5387	/* Commit the CR0 and related fields to the guest VMCS. */
5388	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR0, u64GuestCr0); AssertRC(rc);
5389	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, u64ShadowCr0); AssertRC(rc);
5390	if (uProcCtls != pVmcsInfo->u32ProcCtls)
5391	{
5392	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, uProcCtls);
5393	AssertRC(rc);
5394	}
5395	if (uXcptBitmap != pVmcsInfo->u32XcptBitmap)
5396	{
5397	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
5398	AssertRC(rc);
5399	}
5400
5401	/* Update our caches. */
5402	pVmcsInfo->u32ProcCtls = uProcCtls;
5403	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
5404
5405	Log4Func(("cr0=%#RX64 shadow=%#RX64 set=%#RX64 zap=%#RX64\n", u64GuestCr0, u64ShadowCr0, fSetCr0, fZapCr0));
5406	}
5407	else
5408	{
5409	/*
5410	* With nested-guests, we may have extended the guest/host mask here since we
5411	* merged in the outer guest's mask. Thus, the merged mask can include more bits
5412	* (to read from the nested-guest CR0 read-shadow) than the guest hypervisor
5413	* originally supplied. We must copy those bits from the nested-guest CR0 into
5414	* the nested-guest CR0 read-shadow.
5415	*/
5416	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
5417	uint64_t u64GuestCr0 = pVCpu->cpum.GstCtx.cr0;
5418	uint64_t const u64ShadowCr0 = CPUMGetGuestVmxMaskedCr0(pVCpu, &pVCpu->cpum.GstCtx, pVmcsInfo->u64Cr0Mask);
5419	Assert(!RT_HI_U32(u64GuestCr0));
5420	Assert(u64GuestCr0 & X86_CR0_NE);
5421
5422	/*
5423	* Apply the hardware specified fixed CR0 bits and enable caching.
5424	* Note! We could be altering our VMX emulation's fixed bits. We thus
5425	* need to re-apply them while importing CR0.
5426	*/
5427	u64GuestCr0 \|= fSetCr0;
5428	u64GuestCr0 &= fZapCr0;
5429	u64GuestCr0 &= ~(uint64_t)(X86_CR0_CD \| X86_CR0_NW);
5430
5431	/* Commit the CR0 and CR0 read-shadow to the nested-guest VMCS. */
5432	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR0, u64GuestCr0); AssertRC(rc);
5433	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, u64ShadowCr0); AssertRC(rc);
5434
5435	Log4Func(("cr0=%#RX64 shadow=%#RX64 (set=%#RX64 zap=%#RX64)\n", u64GuestCr0, u64ShadowCr0, fSetCr0, fZapCr0));
5436	}
5437
5438	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR0);
5439	}
5440
5441	return VINF_SUCCESS;
5442	}
5443
5444
5445	/**
5446	* Exports the guest control registers (CR3, CR4) into the guest-state area
5447	* in the VMCS.
5448	*
5449	* @returns VBox strict status code.
5450	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
5451	* without unrestricted guest access and the VMMDev is not presently
5452	* mapped (e.g. EFI32).
5453	*
5454	* @param pVCpu The cross context virtual CPU structure.
5455	* @param pVmxTransient The VMX-transient structure.
5456	*
5457	* @remarks No-long-jump zone!!!
5458	*/
5459	static VBOXSTRICTRC hmR0VmxExportGuestCR3AndCR4(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
5460	{
5461	int rc = VINF_SUCCESS;
5462	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5463
5464	/*
5465	* Guest CR2.
5466	* It's always loaded in the assembler code. Nothing to do here.
5467	*/
5468
5469	/*
5470	* Guest CR3.
5471	*/
5472	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR3)
5473	{
5474	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR3);
5475
5476	if (pVM->hm.s.fNestedPaging)
5477	{
5478	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5479	pVmcsInfo->HCPhysEPTP = PGMGetHyperCR3(pVCpu);
5480
5481	/* Validate. See Intel spec. 28.2.2 "EPT Translation Mechanism" and 24.6.11 "Extended-Page-Table Pointer (EPTP)" */
5482	Assert(pVmcsInfo->HCPhysEPTP != NIL_RTHCPHYS);
5483	Assert(!(pVmcsInfo->HCPhysEPTP & UINT64_C(0xfff0000000000000)));
5484	Assert(!(pVmcsInfo->HCPhysEPTP & 0xfff));
5485
5486	/* VMX_EPT_MEMTYPE_WB support is already checked in hmR0VmxSetupTaggedTlb(). */
5487	pVmcsInfo->HCPhysEPTP \|= VMX_EPT_MEMTYPE_WB
5488	\| (VMX_EPT_PAGE_WALK_LENGTH_DEFAULT << VMX_EPT_PAGE_WALK_LENGTH_SHIFT);
5489
5490	/* Validate. See Intel spec. 26.2.1 "Checks on VMX Controls" */
5491	AssertMsg( ((pVmcsInfo->HCPhysEPTP >> 3) & 0x07) == 3 /* Bits 3:5 (EPT page walk length - 1) must be 3. */
5492	&& ((pVmcsInfo->HCPhysEPTP >> 7) & 0x1f) == 0, /* Bits 7:11 MBZ. */
5493	("EPTP %#RX64\n", pVmcsInfo->HCPhysEPTP));
5494	AssertMsg( !((pVmcsInfo->HCPhysEPTP >> 6) & 0x01) /* Bit 6 (EPT accessed & dirty bit). */
5495	\|\| (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EPT_ACCESS_DIRTY),
5496	("EPTP accessed/dirty bit not supported by CPU but set %#RX64\n", pVmcsInfo->HCPhysEPTP));
5497
5498	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, pVmcsInfo->HCPhysEPTP);
5499	AssertRC(rc);
5500
5501	uint64_t u64GuestCr3;
5502	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5503	if ( pVM->hm.s.vmx.fUnrestrictedGuest
5504	\|\| CPUMIsGuestPagingEnabledEx(pCtx))
5505	{
5506	/* If the guest is in PAE mode, pass the PDPEs to VT-x using the VMCS fields. */
5507	if (CPUMIsGuestInPAEModeEx(pCtx))
5508	{
5509	rc = PGMGstGetPaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
5510	AssertRC(rc);
5511	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, pVCpu->hm.s.aPdpes[0].u); AssertRC(rc);
5512	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, pVCpu->hm.s.aPdpes[1].u); AssertRC(rc);
5513	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, pVCpu->hm.s.aPdpes[2].u); AssertRC(rc);
5514	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, pVCpu->hm.s.aPdpes[3].u); AssertRC(rc);
5515	}
5516
5517	/*
5518	* The guest's view of its CR3 is unblemished with nested paging when the
5519	* guest is using paging or we have unrestricted guest execution to handle
5520	* the guest when it's not using paging.
5521	*/
5522	u64GuestCr3 = pCtx->cr3;
5523	}
5524	else
5525	{
5526	/*
5527	* The guest is not using paging, but the CPU (VT-x) has to. While the guest
5528	* thinks it accesses physical memory directly, we use our identity-mapped
5529	* page table to map guest-linear to guest-physical addresses. EPT takes care
5530	* of translating it to host-physical addresses.
5531	*/
5532	RTGCPHYS GCPhys;
5533	Assert(pVM->hm.s.vmx.pNonPagingModeEPTPageTable);
5534
5535	/* We obtain it here every time as the guest could have relocated this PCI region. */
5536	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys);
5537	if (RT_SUCCESS(rc))
5538	{ /* likely */ }
5539	else if (rc == VERR_PDM_DEV_HEAP_R3_TO_GCPHYS)
5540	{
5541	Log4Func(("VERR_PDM_DEV_HEAP_R3_TO_GCPHYS -> VINF_EM_RESCHEDULE_REM\n"));
5542	return VINF_EM_RESCHEDULE_REM; /* We cannot execute now, switch to REM/IEM till the guest maps in VMMDev. */
5543	}
5544	else
5545	AssertMsgFailedReturn(("%Rrc\n", rc), rc);
5546
5547	u64GuestCr3 = GCPhys;
5548	}
5549
5550	Log4Func(("guest_cr3=%#RX64 (GstN)\n", u64GuestCr3));
5551	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR3, u64GuestCr3);
5552	AssertRC(rc);
5553	}
5554	else
5555	{
5556	Assert(!pVmxTransient->fIsNestedGuest);
5557	/* Non-nested paging case, just use the hypervisor's CR3. */
5558	RTHCPHYS const HCPhysGuestCr3 = PGMGetHyperCR3(pVCpu);
5559
5560	Log4Func(("guest_cr3=%#RX64 (HstN)\n", HCPhysGuestCr3));
5561	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR3, HCPhysGuestCr3);
5562	AssertRC(rc);
5563	}
5564
5565	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR3);
5566	}
5567
5568	/*
5569	* Guest CR4.
5570	* ASSUMES this is done everytime we get in from ring-3! (XCR0)
5571	*/
5572	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR4)
5573	{
5574	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5575	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5576
5577	/*
5578	* Figure out fixed CR4 bits in VMX operation.
5579	*/
5580	uint64_t const fSetCr4 = pVM->hm.s.vmx.Msrs.u64Cr4Fixed0;
5581	uint64_t const fZapCr4 = pVM->hm.s.vmx.Msrs.u64Cr4Fixed1;
5582
5583	/*
5584	* With nested-guests, we may have extended the guest/host mask here (since we
5585	* merged in the outer guest's mask, see hmR0VmxMergeVmcsNested). This means, the
5586	* mask can include more bits (to read from the nested-guest CR4 read-shadow) than
5587	* the guest hypervisor originally supplied. Thus, we should, in essence, copy
5588	* those bits from the nested-guest CR4 into the nested-guest CR4 read-shadow.
5589	*/
5590	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
5591	uint64_t u64GuestCr4 = pCtx->cr4;
5592	uint64_t const u64ShadowCr4 = !pVmxTransient->fIsNestedGuest
5593	? pCtx->cr4
5594	: CPUMGetGuestVmxMaskedCr4(pVCpu, pCtx, pVmcsInfo->u64Cr4Mask);
5595	Assert(!RT_HI_U32(u64GuestCr4));
5596
5597	/*
5598	* Setup VT-x's view of the guest CR4.
5599	*
5600	* If we're emulating real-mode using virtual-8086 mode, we want to redirect software
5601	* interrupts to the 8086 program interrupt handler. Clear the VME bit (the interrupt
5602	* redirection bitmap is already all 0, see hmR3InitFinalizeR0())
5603	*
5604	* See Intel spec. 20.2 "Software Interrupt Handling Methods While in Virtual-8086 Mode".
5605	*/
5606	if (pVmcsInfo->RealMode.fRealOnV86Active)
5607	{
5608	Assert(pVM->hm.s.vmx.pRealModeTSS);
5609	Assert(PDMVmmDevHeapIsEnabled(pVM));
5610	u64GuestCr4 &= ~(uint64_t)X86_CR4_VME;
5611	}
5612
5613	if (pVM->hm.s.fNestedPaging)
5614	{
5615	if ( !CPUMIsGuestPagingEnabledEx(pCtx)
5616	&& !pVM->hm.s.vmx.fUnrestrictedGuest)
5617	{
5618	/* We use 4 MB pages in our identity mapping page table when the guest doesn't have paging. */
5619	u64GuestCr4 \|= X86_CR4_PSE;
5620	/* Our identity mapping is a 32-bit page directory. */
5621	u64GuestCr4 &= ~(uint64_t)X86_CR4_PAE;
5622	}
5623	/* else use guest CR4.*/
5624	}
5625	else
5626	{
5627	Assert(!pVmxTransient->fIsNestedGuest);
5628
5629	/*
5630	* The shadow paging modes and guest paging modes are different, the shadow is in accordance with the host
5631	* paging mode and thus we need to adjust VT-x's view of CR4 depending on our shadow page tables.
5632	*/
5633	switch (pVCpu->hm.s.enmShadowMode)
5634	{
5635	case PGMMODE_REAL: /* Real-mode. */
5636	case PGMMODE_PROTECTED: /* Protected mode without paging. */
5637	case PGMMODE_32_BIT: /* 32-bit paging. */
5638	{
5639	u64GuestCr4 &= ~(uint64_t)X86_CR4_PAE;
5640	break;
5641	}
5642
5643	case PGMMODE_PAE: /* PAE paging. */
5644	case PGMMODE_PAE_NX: /* PAE paging with NX. */
5645	{
5646	u64GuestCr4 \|= X86_CR4_PAE;
5647	break;
5648	}
5649
5650	case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
5651	case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
5652	#ifdef VBOX_WITH_64_BITS_GUESTS
5653	break;
5654	#endif
5655	default:
5656	AssertFailed();
5657	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
5658	}
5659	}
5660
5661	/*
5662	* Apply the hardware specified fixed CR4 bits (mainly CR4.VMXE).
5663	* Note! For nested-guests, we could be altering our VMX emulation's
5664	* fixed bits. We thus need to re-apply them while importing CR4.
5665	*/
5666	u64GuestCr4 \|= fSetCr4;
5667	u64GuestCr4 &= fZapCr4;
5668
5669	/* Commit the CR4 and CR4 read-shadow to the guest VMCS. */
5670	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR4, u64GuestCr4); AssertRC(rc);
5671	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, u64ShadowCr4); AssertRC(rc);
5672
5673	/* Whether to save/load/restore XCR0 during world switch depends on CR4.OSXSAVE and host+guest XCR0. */
5674	pVCpu->hm.s.fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
5675
5676	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR4);
5677
5678	Log4Func(("cr4=%#RX64 shadow=%#RX64 (set=%#RX64 zap=%#RX64)\n", u64GuestCr4, u64ShadowCr4, fSetCr4, fZapCr4));
5679	}
5680	return rc;
5681	}
5682
5683
5684	/**
5685	* Exports the guest debug registers into the guest-state area in the VMCS.
5686	* The guest debug bits are partially shared with the host (e.g. DR6, DR0-3).
5687	*
5688	* This also sets up whether \#DB and MOV DRx accesses cause VM-exits.
5689	*
5690	* @returns VBox status code.
5691	* @param pVCpu The cross context virtual CPU structure.
5692	* @param pVmxTransient The VMX-transient structure.
5693	*
5694	* @remarks No-long-jump zone!!!
5695	*/
5696	static int hmR0VmxExportSharedDebugState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
5697	{
5698	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
5699
5700	/** @todo NSTVMX: Figure out what we want to do with nested-guest instruction
5701	* stepping. */
5702	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5703	if (pVmxTransient->fIsNestedGuest)
5704	{
5705	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, CPUMGetGuestDR7(pVCpu));
5706	AssertRC(rc);
5707
5708	/* Always intercept Mov DRx accesses for the nested-guest for now. */
5709	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_MOV_DR_EXIT;
5710	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
5711	AssertRC(rc);
5712	return VINF_SUCCESS;
5713	}
5714
5715	#ifdef VBOX_STRICT
5716	/* Validate. Intel spec. 26.3.1.1 "Checks on Guest Controls Registers, Debug Registers, MSRs" */
5717	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
5718	{
5719	/* Validate. Intel spec. 17.2 "Debug Registers", recompiler paranoia checks. */
5720	Assert((pVCpu->cpum.GstCtx.dr[7] & (X86_DR7_MBZ_MASK \| X86_DR7_RAZ_MASK)) == 0);
5721	Assert((pVCpu->cpum.GstCtx.dr[7] & X86_DR7_RA1_MASK) == X86_DR7_RA1_MASK);
5722	}
5723	#endif
5724
5725	bool fSteppingDB = false;
5726	bool fInterceptMovDRx = false;
5727	uint32_t uProcCtls = pVmcsInfo->u32ProcCtls;
5728	if (pVCpu->hm.s.fSingleInstruction)
5729	{
5730	/* If the CPU supports the monitor trap flag, use it for single stepping in DBGF and avoid intercepting #DB. */
5731	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5732	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_MONITOR_TRAP_FLAG)
5733	{
5734	uProcCtls \|= VMX_PROC_CTLS_MONITOR_TRAP_FLAG;
5735	Assert(fSteppingDB == false);
5736	}
5737	else
5738	{
5739	pVCpu->cpum.GstCtx.eflags.u32 \|= X86_EFL_TF;
5740	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_GUEST_RFLAGS;
5741	pVCpu->hm.s.fClearTrapFlag = true;
5742	fSteppingDB = true;
5743	}
5744	}
5745
5746	uint64_t u64GuestDr7;
5747	if ( fSteppingDB
5748	\|\| (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK))
5749	{
5750	/*
5751	* Use the combined guest and host DRx values found in the hypervisor register set
5752	* because the hypervisor debugger has breakpoints active or someone is single stepping
5753	* on the host side without a monitor trap flag.
5754	*
5755	* Note! DBGF expects a clean DR6 state before executing guest code.
5756	*/
5757	if (!CPUMIsHyperDebugStateActive(pVCpu))
5758	{
5759	CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
5760	Assert(CPUMIsHyperDebugStateActive(pVCpu));
5761	Assert(!CPUMIsGuestDebugStateActive(pVCpu));
5762	}
5763
5764	/* Update DR7 with the hypervisor value (other DRx registers are handled by CPUM one way or another). */
5765	u64GuestDr7 = CPUMGetHyperDR7(pVCpu);
5766	pVCpu->hm.s.fUsingHyperDR7 = true;
5767	fInterceptMovDRx = true;
5768	}
5769	else
5770	{
5771	/*
5772	* If the guest has enabled debug registers, we need to load them prior to
5773	* executing guest code so they'll trigger at the right time.
5774	*/
5775	if (pVCpu->cpum.GstCtx.dr[7] & (X86_DR7_ENABLED_MASK \| X86_DR7_GD))
5776	{
5777	if (!CPUMIsGuestDebugStateActive(pVCpu))
5778	{
5779	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
5780	Assert(CPUMIsGuestDebugStateActive(pVCpu));
5781	Assert(!CPUMIsHyperDebugStateActive(pVCpu));
5782	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
5783	}
5784	Assert(!fInterceptMovDRx);
5785	}
5786	else if (!CPUMIsGuestDebugStateActive(pVCpu))
5787	{
5788	/*
5789	* If no debugging enabled, we'll lazy load DR0-3. Unlike on AMD-V, we
5790	* must intercept #DB in order to maintain a correct DR6 guest value, and
5791	* because we need to intercept it to prevent nested #DBs from hanging the
5792	* CPU, we end up always having to intercept it. See hmR0VmxSetupVmcsXcptBitmap().
5793	*/
5794	fInterceptMovDRx = true;
5795	}
5796
5797	/* Update DR7 with the actual guest value. */
5798	u64GuestDr7 = pVCpu->cpum.GstCtx.dr[7];
5799	pVCpu->hm.s.fUsingHyperDR7 = false;
5800	}
5801
5802	if (fInterceptMovDRx)
5803	uProcCtls \|= VMX_PROC_CTLS_MOV_DR_EXIT;
5804	else
5805	uProcCtls &= ~VMX_PROC_CTLS_MOV_DR_EXIT;
5806
5807	/*
5808	* Update the processor-based VM-execution controls with the MOV-DRx intercepts and the
5809	* monitor-trap flag and update our cache.
5810	*/
5811	if (uProcCtls != pVmcsInfo->u32ProcCtls)
5812	{
5813	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, uProcCtls);
5814	AssertRC(rc);
5815	pVmcsInfo->u32ProcCtls = uProcCtls;
5816	}
5817
5818	/*
5819	* Update guest DR7.
5820	*/
5821	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, u64GuestDr7);
5822	AssertRC(rc);
5823
5824	/*
5825	* If we have forced EFLAGS.TF to be set because we're single-stepping in the hypervisor debugger,
5826	* we need to clear interrupt inhibition if any as otherwise it causes a VM-entry failure.
5827	*
5828	* See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
5829	*/
5830	if (fSteppingDB)
5831	{
5832	Assert(pVCpu->hm.s.fSingleInstruction);
5833	Assert(pVCpu->cpum.GstCtx.eflags.Bits.u1TF);
5834
5835	uint32_t fIntrState = 0;
5836	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
5837	AssertRC(rc);
5838
5839	if (fIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
5840	{
5841	fIntrState &= ~(VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
5842	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
5843	AssertRC(rc);
5844	}
5845	}
5846
5847	return VINF_SUCCESS;
5848	}
5849
5850
5851	#ifdef VBOX_STRICT
5852	/**
5853	* Strict function to validate segment registers.
5854	*
5855	* @param pVCpu The cross context virtual CPU structure.
5856	* @param pVmcsInfo The VMCS info. object.
5857	*
5858	* @remarks Will import guest CR0 on strict builds during validation of
5859	* segments.
5860	*/
5861	static void hmR0VmxValidateSegmentRegs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
5862	{
5863	/*
5864	* Validate segment registers. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
5865	*
5866	* The reason we check for attribute value 0 in this function and not just the unusable bit is
5867	* because hmR0VmxExportGuestSegReg() only updates the VMCS' copy of the value with the
5868	* unusable bit and doesn't change the guest-context value.
5869	*/
5870	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5871	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5872	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR0);
5873	if ( !pVM->hm.s.vmx.fUnrestrictedGuest
5874	&& ( !CPUMIsGuestInRealModeEx(pCtx)
5875	&& !CPUMIsGuestInV86ModeEx(pCtx)))
5876	{
5877	/* Protected mode checks */
5878	/* CS */
5879	Assert(pCtx->cs.Attr.n.u1Present);
5880	Assert(!(pCtx->cs.Attr.u & 0xf00));
5881	Assert(!(pCtx->cs.Attr.u & 0xfffe0000));
5882	Assert( (pCtx->cs.u32Limit & 0xfff) == 0xfff
5883	\|\| !(pCtx->cs.Attr.n.u1Granularity));
5884	Assert( !(pCtx->cs.u32Limit & 0xfff00000)
5885	\|\| (pCtx->cs.Attr.n.u1Granularity));
5886	/* CS cannot be loaded with NULL in protected mode. */
5887	Assert(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE)); /** @todo is this really true even for 64-bit CS? */
5888	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
5889	Assert(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl);
5890	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
5891	Assert(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl);
5892	else
5893	AssertMsgFailed(("Invalid CS Type %#x\n", pCtx->cs.Attr.n.u2Dpl));
5894	/* SS */
5895	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
5896	Assert(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL));
5897	if ( !(pCtx->cr0 & X86_CR0_PE)
5898	\|\| pCtx->cs.Attr.n.u4Type == 3)
5899	{
5900	Assert(!pCtx->ss.Attr.n.u2Dpl);
5901	}
5902	if (pCtx->ss.Attr.u && !(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
5903	{
5904	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
5905	Assert(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7);
5906	Assert(pCtx->ss.Attr.n.u1Present);
5907	Assert(!(pCtx->ss.Attr.u & 0xf00));
5908	Assert(!(pCtx->ss.Attr.u & 0xfffe0000));
5909	Assert( (pCtx->ss.u32Limit & 0xfff) == 0xfff
5910	\|\| !(pCtx->ss.Attr.n.u1Granularity));
5911	Assert( !(pCtx->ss.u32Limit & 0xfff00000)
5912	\|\| (pCtx->ss.Attr.n.u1Granularity));
5913	}
5914	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxExportGuestSegReg(). */
5915	if (pCtx->ds.Attr.u && !(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
5916	{
5917	Assert(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
5918	Assert(pCtx->ds.Attr.n.u1Present);
5919	Assert(pCtx->ds.Attr.n.u4Type > 11 \|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL));
5920	Assert(!(pCtx->ds.Attr.u & 0xf00));
5921	Assert(!(pCtx->ds.Attr.u & 0xfffe0000));
5922	Assert( (pCtx->ds.u32Limit & 0xfff) == 0xfff
5923	\|\| !(pCtx->ds.Attr.n.u1Granularity));
5924	Assert( !(pCtx->ds.u32Limit & 0xfff00000)
5925	\|\| (pCtx->ds.Attr.n.u1Granularity));
5926	Assert( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
5927	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ));
5928	}
5929	if (pCtx->es.Attr.u && !(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
5930	{
5931	Assert(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
5932	Assert(pCtx->es.Attr.n.u1Present);
5933	Assert(pCtx->es.Attr.n.u4Type > 11 \|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL));
5934	Assert(!(pCtx->es.Attr.u & 0xf00));
5935	Assert(!(pCtx->es.Attr.u & 0xfffe0000));
5936	Assert( (pCtx->es.u32Limit & 0xfff) == 0xfff
5937	\|\| !(pCtx->es.Attr.n.u1Granularity));
5938	Assert( !(pCtx->es.u32Limit & 0xfff00000)
5939	\|\| (pCtx->es.Attr.n.u1Granularity));
5940	Assert( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
5941	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ));
5942	}
5943	if (pCtx->fs.Attr.u && !(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
5944	{
5945	Assert(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
5946	Assert(pCtx->fs.Attr.n.u1Present);
5947	Assert(pCtx->fs.Attr.n.u4Type > 11 \|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL));
5948	Assert(!(pCtx->fs.Attr.u & 0xf00));
5949	Assert(!(pCtx->fs.Attr.u & 0xfffe0000));
5950	Assert( (pCtx->fs.u32Limit & 0xfff) == 0xfff
5951	\|\| !(pCtx->fs.Attr.n.u1Granularity));
5952	Assert( !(pCtx->fs.u32Limit & 0xfff00000)
5953	\|\| (pCtx->fs.Attr.n.u1Granularity));
5954	Assert( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
5955	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ));
5956	}
5957	if (pCtx->gs.Attr.u && !(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
5958	{
5959	Assert(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
5960	Assert(pCtx->gs.Attr.n.u1Present);
5961	Assert(pCtx->gs.Attr.n.u4Type > 11 \|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL));
5962	Assert(!(pCtx->gs.Attr.u & 0xf00));
5963	Assert(!(pCtx->gs.Attr.u & 0xfffe0000));
5964	Assert( (pCtx->gs.u32Limit & 0xfff) == 0xfff
5965	\|\| !(pCtx->gs.Attr.n.u1Granularity));
5966	Assert( !(pCtx->gs.u32Limit & 0xfff00000)
5967	\|\| (pCtx->gs.Attr.n.u1Granularity));
5968	Assert( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
5969	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ));
5970	}
5971	/* 64-bit capable CPUs. */
5972	Assert(!RT_HI_U32(pCtx->cs.u64Base));
5973	Assert(!pCtx->ss.Attr.u \|\| !RT_HI_U32(pCtx->ss.u64Base));
5974	Assert(!pCtx->ds.Attr.u \|\| !RT_HI_U32(pCtx->ds.u64Base));
5975	Assert(!pCtx->es.Attr.u \|\| !RT_HI_U32(pCtx->es.u64Base));
5976	}
5977	else if ( CPUMIsGuestInV86ModeEx(pCtx)
5978	\|\| ( CPUMIsGuestInRealModeEx(pCtx)
5979	&& !pVM->hm.s.vmx.fUnrestrictedGuest))
5980	{
5981	/* Real and v86 mode checks. */
5982	/* hmR0VmxExportGuestSegReg() writes the modified in VMCS. We want what we're feeding to VT-x. */
5983	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
5984	if (pVmcsInfo->RealMode.fRealOnV86Active)
5985	{
5986	u32CSAttr = 0xf3; u32SSAttr = 0xf3; u32DSAttr = 0xf3;
5987	u32ESAttr = 0xf3; u32FSAttr = 0xf3; u32GSAttr = 0xf3;
5988	}
5989	else
5990	{
5991	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u; u32DSAttr = pCtx->ds.Attr.u;
5992	u32ESAttr = pCtx->es.Attr.u; u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
5993	}
5994
5995	/* CS */
5996	AssertMsg((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), ("CS base %#x %#x\n", pCtx->cs.u64Base, pCtx->cs.Sel));
5997	Assert(pCtx->cs.u32Limit == 0xffff);
5998	Assert(u32CSAttr == 0xf3);
5999	/* SS */
6000	Assert(pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4);
6001	Assert(pCtx->ss.u32Limit == 0xffff);
6002	Assert(u32SSAttr == 0xf3);
6003	/* DS */
6004	Assert(pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4);
6005	Assert(pCtx->ds.u32Limit == 0xffff);
6006	Assert(u32DSAttr == 0xf3);
6007	/* ES */
6008	Assert(pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4);
6009	Assert(pCtx->es.u32Limit == 0xffff);
6010	Assert(u32ESAttr == 0xf3);
6011	/* FS */
6012	Assert(pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4);
6013	Assert(pCtx->fs.u32Limit == 0xffff);
6014	Assert(u32FSAttr == 0xf3);
6015	/* GS */
6016	Assert(pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4);
6017	Assert(pCtx->gs.u32Limit == 0xffff);
6018	Assert(u32GSAttr == 0xf3);
6019	/* 64-bit capable CPUs. */
6020	Assert(!RT_HI_U32(pCtx->cs.u64Base));
6021	Assert(!u32SSAttr \|\| !RT_HI_U32(pCtx->ss.u64Base));
6022	Assert(!u32DSAttr \|\| !RT_HI_U32(pCtx->ds.u64Base));
6023	Assert(!u32ESAttr \|\| !RT_HI_U32(pCtx->es.u64Base));
6024	}
6025	}
6026	#endif /* VBOX_STRICT */
6027
6028
6029	/**
6030	* Exports a guest segment register into the guest-state area in the VMCS.
6031	*
6032	* @returns VBox status code.
6033	* @param pVCpu The cross context virtual CPU structure.
6034	* @param pVmcsInfo The VMCS info. object.
6035	* @param iSegReg The segment register number (X86_SREG_XXX).
6036	* @param pSelReg Pointer to the segment selector.
6037	*
6038	* @remarks No-long-jump zone!!!
6039	*/
6040	static int hmR0VmxExportGuestSegReg(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, uint8_t iSegReg, PCCPUMSELREG pSelReg)
6041	{
6042	Assert(iSegReg < X86_SREG_COUNT);
6043	uint32_t const idxSel = g_aVmcsSegSel[iSegReg];
6044	uint32_t const idxLimit = g_aVmcsSegLimit[iSegReg];
6045	uint32_t const idxBase = g_aVmcsSegBase[iSegReg];
6046	uint32_t const idxAttr = g_aVmcsSegAttr[iSegReg];
6047
6048	uint32_t u32Access = pSelReg->Attr.u;
6049	if (pVmcsInfo->RealMode.fRealOnV86Active)
6050	{
6051	/* VT-x requires our real-using-v86 mode hack to override the segment access-right bits. */
6052	u32Access = 0xf3;
6053	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
6054	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
6055	RT_NOREF_PV(pVCpu);
6056	}
6057	else
6058	{
6059	/*
6060	* The way to differentiate between whether this is really a null selector or was just
6061	* a selector loaded with 0 in real-mode is using the segment attributes. A selector
6062	* loaded in real-mode with the value 0 is valid and usable in protected-mode and we
6063	* should -not- mark it as an unusable segment. Both the recompiler & VT-x ensures
6064	* NULL selectors loaded in protected-mode have their attribute as 0.
6065	*/
6066	if (!u32Access)
6067	u32Access = X86DESCATTR_UNUSABLE;
6068	}
6069
6070	/* Validate segment access rights. Refer to Intel spec. "26.3.1.2 Checks on Guest Segment Registers". */
6071	AssertMsg((u32Access & X86DESCATTR_UNUSABLE) \|\| (u32Access & X86_SEL_TYPE_ACCESSED),
6072	("Access bit not set for usable segment. idx=%#x sel=%#x attr %#x\n", idxBase, pSelReg, pSelReg->Attr.u));
6073
6074	/*
6075	* Commit it to the VMCS.
6076	*/
6077	int rc = VMXWriteVmcs32(idxSel, pSelReg->Sel); AssertRC(rc);
6078	rc = VMXWriteVmcs32(idxLimit, pSelReg->u32Limit); AssertRC(rc);
6079	rc = VMXWriteVmcsNw(idxBase, pSelReg->u64Base); AssertRC(rc);
6080	rc = VMXWriteVmcs32(idxAttr, u32Access); AssertRC(rc);
6081	return VINF_SUCCESS;
6082	}
6083
6084
6085	/**
6086	* Exports the guest segment registers, GDTR, IDTR, LDTR, TR into the guest-state
6087	* area in the VMCS.
6088	*
6089	* @returns VBox status code.
6090	* @param pVCpu The cross context virtual CPU structure.
6091	* @param pVmxTransient The VMX-transient structure.
6092	*
6093	* @remarks Will import guest CR0 on strict builds during validation of
6094	* segments.
6095	* @remarks No-long-jump zone!!!
6096	*/
6097	static int hmR0VmxExportGuestSegRegsXdtr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
6098	{
6099	int rc = VERR_INTERNAL_ERROR_5;
6100	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6101	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6102	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6103
6104	/*
6105	* Guest Segment registers: CS, SS, DS, ES, FS, GS.
6106	*/
6107	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SREG_MASK)
6108	{
6109	#ifdef VBOX_WITH_REM
6110	if (!pVM->hm.s.vmx.fUnrestrictedGuest)
6111	{
6112	Assert(!pVmxTransient->fIsNestedGuest);
6113	Assert(pVM->hm.s.vmx.pRealModeTSS);
6114	AssertCompile(PGMMODE_REAL < PGMMODE_PROTECTED);
6115	if ( pVmcsInfo->fWasInRealMode
6116	&& PGMGetGuestMode(pVCpu) >= PGMMODE_PROTECTED)
6117	{
6118	/*
6119	* Notify the recompiler must flush its code-cache as the guest -may-
6120	* rewrite code it in real-mode (e.g. OpenBSD 4.0).
6121	*/
6122	REMFlushTBs(pVM);
6123	Log4Func(("Switch to protected mode detected!\n"));
6124	pVmcsInfo->fWasInRealMode = false;
6125	}
6126	}
6127	#endif
6128	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CS)
6129	{
6130	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CS);
6131	if (pVmcsInfo->RealMode.fRealOnV86Active)
6132	pVmcsInfo->RealMode.AttrCS.u = pCtx->cs.Attr.u;
6133	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_CS, &pCtx->cs);
6134	AssertRC(rc);
6135	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CS);
6136	}
6137
6138	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SS)
6139	{
6140	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SS);
6141	if (pVmcsInfo->RealMode.fRealOnV86Active)
6142	pVmcsInfo->RealMode.AttrSS.u = pCtx->ss.Attr.u;
6143	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_SS, &pCtx->ss);
6144	AssertRC(rc);
6145	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SS);
6146	}
6147
6148	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_DS)
6149	{
6150	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_DS);
6151	if (pVmcsInfo->RealMode.fRealOnV86Active)
6152	pVmcsInfo->RealMode.AttrDS.u = pCtx->ds.Attr.u;
6153	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_DS, &pCtx->ds);
6154	AssertRC(rc);
6155	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_DS);
6156	}
6157
6158	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_ES)
6159	{
6160	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_ES);
6161	if (pVmcsInfo->RealMode.fRealOnV86Active)
6162	pVmcsInfo->RealMode.AttrES.u = pCtx->es.Attr.u;
6163	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_ES, &pCtx->es);
6164	AssertRC(rc);
6165	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_ES);
6166	}
6167
6168	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_FS)
6169	{
6170	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_FS);
6171	if (pVmcsInfo->RealMode.fRealOnV86Active)
6172	pVmcsInfo->RealMode.AttrFS.u = pCtx->fs.Attr.u;
6173	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_FS, &pCtx->fs);
6174	AssertRC(rc);
6175	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_FS);
6176	}
6177
6178	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_GS)
6179	{
6180	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GS);
6181	if (pVmcsInfo->RealMode.fRealOnV86Active)
6182	pVmcsInfo->RealMode.AttrGS.u = pCtx->gs.Attr.u;
6183	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_GS, &pCtx->gs);
6184	AssertRC(rc);
6185	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_GS);
6186	}
6187
6188	#ifdef VBOX_STRICT
6189	hmR0VmxValidateSegmentRegs(pVCpu, pVmcsInfo);
6190	#endif
6191	Log4Func(("cs={%#04x base=%#RX64 limit=%#RX32 attr=%#RX32}\n", pCtx->cs.Sel, pCtx->cs.u64Base, pCtx->cs.u32Limit,
6192	pCtx->cs.Attr.u));
6193	}
6194
6195	/*
6196	* Guest TR.
6197	*/
6198	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_TR)
6199	{
6200	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_TR);
6201
6202	/*
6203	* Real-mode emulation using virtual-8086 mode with CR4.VME. Interrupt redirection is
6204	* achieved using the interrupt redirection bitmap (all bits cleared to let the guest
6205	* handle INT-n's) in the TSS. See hmR3InitFinalizeR0() to see how pRealModeTSS is setup.
6206	*/
6207	uint16_t u16Sel;
6208	uint32_t u32Limit;
6209	uint64_t u64Base;
6210	uint32_t u32AccessRights;
6211	if (!pVmcsInfo->RealMode.fRealOnV86Active)
6212	{
6213	u16Sel = pCtx->tr.Sel;
6214	u32Limit = pCtx->tr.u32Limit;
6215	u64Base = pCtx->tr.u64Base;
6216	u32AccessRights = pCtx->tr.Attr.u;
6217	}
6218	else
6219	{
6220	Assert(!pVmxTransient->fIsNestedGuest);
6221	Assert(pVM->hm.s.vmx.pRealModeTSS);
6222	Assert(PDMVmmDevHeapIsEnabled(pVM)); /* Guaranteed by HMCanExecuteGuest() -XXX- what about inner loop changes? */
6223
6224	/* We obtain it here every time as PCI regions could be reconfigured in the guest, changing the VMMDev base. */
6225	RTGCPHYS GCPhys;
6226	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys);
6227	AssertRCReturn(rc, rc);
6228
6229	X86DESCATTR DescAttr;
6230	DescAttr.u = 0;
6231	DescAttr.n.u1Present = 1;
6232	DescAttr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
6233
6234	u16Sel = 0;
6235	u32Limit = HM_VTX_TSS_SIZE;
6236	u64Base = GCPhys;
6237	u32AccessRights = DescAttr.u;
6238	}
6239
6240	/* Validate. */
6241	Assert(!(u16Sel & RT_BIT(2)));
6242	AssertMsg( (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_386_TSS_BUSY
6243	\|\| (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_286_TSS_BUSY, ("TSS is not busy!? %#x\n", u32AccessRights));
6244	AssertMsg(!(u32AccessRights & X86DESCATTR_UNUSABLE), ("TR unusable bit is not clear!? %#x\n", u32AccessRights));
6245	Assert(!(u32AccessRights & RT_BIT(4))); /* System MBZ.*/
6246	Assert(u32AccessRights & RT_BIT(7)); /* Present MB1.*/
6247	Assert(!(u32AccessRights & 0xf00)); /* 11:8 MBZ. */
6248	Assert(!(u32AccessRights & 0xfffe0000)); /* 31:17 MBZ. */
6249	Assert( (u32Limit & 0xfff) == 0xfff
6250	\|\| !(u32AccessRights & RT_BIT(15))); /* Granularity MBZ. */
6251	Assert( !(pCtx->tr.u32Limit & 0xfff00000)
6252	\|\| (u32AccessRights & RT_BIT(15))); /* Granularity MB1. */
6253
6254	rc = VMXWriteVmcs16(VMX_VMCS16_GUEST_TR_SEL, u16Sel); AssertRC(rc);
6255	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, u32Limit); AssertRC(rc);
6256	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, u32AccessRights); AssertRC(rc);
6257	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_TR_BASE, u64Base); AssertRC(rc);
6258
6259	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_TR);
6260	Log4Func(("tr base=%#RX64 limit=%#RX32\n", pCtx->tr.u64Base, pCtx->tr.u32Limit));
6261	}
6262
6263	/*
6264	* Guest GDTR.
6265	*/
6266	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_GDTR)
6267	{
6268	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GDTR);
6269
6270	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, pCtx->gdtr.cbGdt); AssertRC(rc);
6271	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, pCtx->gdtr.pGdt); AssertRC(rc);
6272
6273	/* Validate. */
6274	Assert(!(pCtx->gdtr.cbGdt & 0xffff0000)); /* Bits 31:16 MBZ. */
6275
6276	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_GDTR);
6277	Log4Func(("gdtr base=%#RX64 limit=%#RX32\n", pCtx->gdtr.pGdt, pCtx->gdtr.cbGdt));
6278	}
6279
6280	/*
6281	* Guest LDTR.
6282	*/
6283	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_LDTR)
6284	{
6285	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_LDTR);
6286
6287	/* The unusable bit is specific to VT-x, if it's a null selector mark it as an unusable segment. */
6288	uint32_t u32Access;
6289	if ( !pVmxTransient->fIsNestedGuest
6290	&& !pCtx->ldtr.Attr.u)
6291	u32Access = X86DESCATTR_UNUSABLE;
6292	else
6293	u32Access = pCtx->ldtr.Attr.u;
6294
6295	rc = VMXWriteVmcs16(VMX_VMCS16_GUEST_LDTR_SEL, pCtx->ldtr.Sel); AssertRC(rc);
6296	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, pCtx->ldtr.u32Limit); AssertRC(rc);
6297	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, u32Access); AssertRC(rc);
6298	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_LDTR_BASE, pCtx->ldtr.u64Base); AssertRC(rc);
6299
6300	/* Validate. */
6301	if (!(u32Access & X86DESCATTR_UNUSABLE))
6302	{
6303	Assert(!(pCtx->ldtr.Sel & RT_BIT(2))); /* TI MBZ. */
6304	Assert(pCtx->ldtr.Attr.n.u4Type == 2); /* Type MB2 (LDT). */
6305	Assert(!pCtx->ldtr.Attr.n.u1DescType); /* System MBZ. */
6306	Assert(pCtx->ldtr.Attr.n.u1Present == 1); /* Present MB1. */
6307	Assert(!pCtx->ldtr.Attr.n.u4LimitHigh); /* 11:8 MBZ. */
6308	Assert(!(pCtx->ldtr.Attr.u & 0xfffe0000)); /* 31:17 MBZ. */
6309	Assert( (pCtx->ldtr.u32Limit & 0xfff) == 0xfff
6310	\|\| !pCtx->ldtr.Attr.n.u1Granularity); /* Granularity MBZ. */
6311	Assert( !(pCtx->ldtr.u32Limit & 0xfff00000)
6312	\|\| pCtx->ldtr.Attr.n.u1Granularity); /* Granularity MB1. */
6313	}
6314
6315	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_LDTR);
6316	Log4Func(("ldtr base=%#RX64 limit=%#RX32\n", pCtx->ldtr.u64Base, pCtx->ldtr.u32Limit));
6317	}
6318
6319	/*
6320	* Guest IDTR.
6321	*/
6322	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_IDTR)
6323	{
6324	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_IDTR);
6325
6326	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, pCtx->idtr.cbIdt); AssertRC(rc);
6327	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, pCtx->idtr.pIdt); AssertRC(rc);
6328
6329	/* Validate. */
6330	Assert(!(pCtx->idtr.cbIdt & 0xffff0000)); /* Bits 31:16 MBZ. */
6331
6332	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_IDTR);
6333	Log4Func(("idtr base=%#RX64 limit=%#RX32\n", pCtx->idtr.pIdt, pCtx->idtr.cbIdt));
6334	}
6335
6336	return VINF_SUCCESS;
6337	}
6338
6339
6340	/**
6341	* Exports certain guest MSRs into the VM-entry MSR-load and VM-exit MSR-store
6342	* areas.
6343	*
6344	* These MSRs will automatically be loaded to the host CPU on every successful
6345	* VM-entry and stored from the host CPU on every successful VM-exit.
6346	*
6347	* We creates/updates MSR slots for the host MSRs in the VM-exit MSR-load area. The
6348	* actual host MSR values are not- updated here for performance reasons. See
6349	* hmR0VmxExportHostMsrs().
6350	*
6351	* We also exports the guest sysenter MSRs into the guest-state area in the VMCS.
6352	*
6353	* @returns VBox status code.
6354	* @param pVCpu The cross context virtual CPU structure.
6355	* @param pVmxTransient The VMX-transient structure.
6356	*
6357	* @remarks No-long-jump zone!!!
6358	*/
6359	static int hmR0VmxExportGuestMsrs(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
6360	{
6361	AssertPtr(pVCpu);
6362	AssertPtr(pVmxTransient);
6363
6364	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6365	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6366
6367	/*
6368	* MSRs that we use the auto-load/store MSR area in the VMCS.
6369	* For 64-bit hosts, we load/restore them lazily, see hmR0VmxLazyLoadGuestMsrs(),
6370	* nothing to do here. The host MSR values are updated when it's safe in
6371	* hmR0VmxLazySaveHostMsrs().
6372	*
6373	* For nested-guests, the guests MSRs from the VM-entry MSR-load area are already
6374	* loaded (into the guest-CPU context) by the VMLAUNCH/VMRESUME instruction
6375	* emulation, nothing to do here.
6376	*/
6377	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_GUEST_AUTO_MSRS)
6378	{
6379	/* No auto-load/store MSRs currently. */
6380	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_GUEST_AUTO_MSRS);
6381	}
6382
6383	/*
6384	* Guest Sysenter MSRs.
6385	*/
6386	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_MSR_MASK)
6387	{
6388	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SYSENTER_MSRS);
6389
6390	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_CS_MSR)
6391	{
6392	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, pCtx->SysEnter.cs);
6393	AssertRC(rc);
6394	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_CS_MSR);
6395	}
6396
6397	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_EIP_MSR)
6398	{
6399	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SYSENTER_EIP, pCtx->SysEnter.eip);
6400	AssertRC(rc);
6401	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_EIP_MSR);
6402	}
6403
6404	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_ESP_MSR)
6405	{
6406	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SYSENTER_ESP, pCtx->SysEnter.esp);
6407	AssertRC(rc);
6408	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_ESP_MSR);
6409	}
6410	}
6411
6412	/*
6413	* Guest/host EFER MSR.
6414	*/
6415	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_EFER_MSR)
6416	{
6417	/* Whether we are using the VMCS to swap the EFER MSR must have been
6418	determined earlier while exporting VM-entry/VM-exit controls. */
6419	Assert(!(ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_ENTRY_EXIT_CTLS));
6420	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
6421
6422	if (hmR0VmxShouldSwapEferMsr(pVCpu))
6423	{
6424	/*
6425	* If the CPU supports VMCS controls for swapping EFER, use it. Otherwise, we have no option
6426	* but to use the auto-load store MSR area in the VMCS for swapping EFER. See @bugref{7368}.
6427	*/
6428	if (pVM->hm.s.vmx.fSupportsVmcsEfer)
6429	{
6430	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_EFER_FULL, pCtx->msrEFER);
6431	AssertRC(rc);
6432	}
6433	else
6434	{
6435	/*
6436	* We shall use the auto-load/store MSR area only for loading the EFER MSR but we must
6437	* continue to intercept guest read and write accesses to it, see @bugref{7386#c16}.
6438	*/
6439	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_EFER, pCtx->msrEFER,
6440	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6441	AssertRCReturn(rc, rc);
6442	}
6443	}
6444	else if (!pVM->hm.s.vmx.fSupportsVmcsEfer)
6445	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_EFER);
6446
6447	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_EFER_MSR);
6448	}
6449
6450	/*
6451	* Other MSRs.
6452	* Speculation Control (R/W).
6453	*/
6454	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_OTHER_MSRS)
6455	{
6456	HMVMX_CPUMCTX_ASSERT(pVCpu, HM_CHANGED_GUEST_OTHER_MSRS);
6457	if (pVM->cpum.ro.GuestFeatures.fIbrs)
6458	{
6459	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_IA32_SPEC_CTRL, CPUMGetGuestSpecCtrl(pVCpu),
6460	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6461	AssertRCReturn(rc, rc);
6462	}
6463	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_OTHER_MSRS);
6464	}
6465
6466	return VINF_SUCCESS;
6467	}
6468
6469
6470	/**
6471	* Selects up the appropriate function to run guest code.
6472	*
6473	* @returns VBox status code.
6474	* @param pVCpu The cross context virtual CPU structure.
6475	* @param pVmxTransient The VMX-transient structure.
6476	*
6477	* @remarks No-long-jump zone!!!
6478	*/
6479	static int hmR0VmxSelectVMRunHandler(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
6480	{
6481	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6482	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6483
6484	if (CPUMIsGuestInLongModeEx(pCtx))
6485	{
6486	#ifndef VBOX_WITH_64_BITS_GUESTS
6487	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
6488	#else
6489	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests); /* Guaranteed by hmR3InitFinalizeR0(). */
6490	/* Guest is in long mode, use the 64-bit handler (host is 64-bit). */
6491	pVmcsInfo->pfnStartVM = VMXR0StartVM64;
6492	#endif
6493	}
6494	else
6495	{
6496	/* Guest is not in long mode, use the 32-bit handler. */
6497	pVmcsInfo->pfnStartVM = VMXR0StartVM32;
6498	}
6499	Assert(pVmcsInfo->pfnStartVM);
6500	return VINF_SUCCESS;
6501	}
6502
6503
6504	/**
6505	* Wrapper for running the guest code in VT-x.
6506	*
6507	* @returns VBox status code, no informational status codes.
6508	* @param pVCpu The cross context virtual CPU structure.
6509	* @param pVmxTransient The VMX-transient structure.
6510	*
6511	* @remarks No-long-jump zone!!!
6512	*/
6513	DECLINLINE(int) hmR0VmxRunGuest(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6514	{
6515	/* Mark that HM is the keeper of all guest-CPU registers now that we're going to execute guest code. */
6516	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6517	pCtx->fExtrn \|= HMVMX_CPUMCTX_EXTRN_ALL \| CPUMCTX_EXTRN_KEEPER_HM;
6518
6519	/** @todo Add stats for VMRESUME vs VMLAUNCH. */
6520
6521	/*
6522	* 64-bit Windows uses XMM registers in the kernel as the Microsoft compiler expresses
6523	* floating-point operations using SSE instructions. Some XMM registers (XMM6-XMM15) are
6524	* callee-saved and thus the need for this XMM wrapper.
6525	*
6526	* See MSDN "Configuring Programs for 64-bit/x64 Software Conventions / Register Usage".
6527	*/
6528	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6529	bool const fResumeVM = RT_BOOL(pVmcsInfo->fVmcsState & VMX_V_VMCS_LAUNCH_STATE_LAUNCHED);
6530	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6531	#ifdef VBOX_WITH_KERNEL_USING_XMM
6532	int rc = hmR0VMXStartVMWrapXMM(fResumeVM, pCtx, NULL /pvUnused/, pVM, pVCpu, pVmcsInfo->pfnStartVM);
6533	#else
6534	int rc = pVmcsInfo->pfnStartVM(fResumeVM, pCtx, NULL /pvUnused/, pVM, pVCpu);
6535	#endif
6536	AssertMsg(rc <= VINF_SUCCESS, ("%Rrc\n", rc));
6537	return rc;
6538	}
6539
6540
6541	/**
6542	* Reports world-switch error and dumps some useful debug info.
6543	*
6544	* @param pVCpu The cross context virtual CPU structure.
6545	* @param rcVMRun The return code from VMLAUNCH/VMRESUME.
6546	* @param pVmxTransient The VMX-transient structure (only
6547	* exitReason updated).
6548	*/
6549	static void hmR0VmxReportWorldSwitchError(PVMCPUCC pVCpu, int rcVMRun, PVMXTRANSIENT pVmxTransient)
6550	{
6551	Assert(pVCpu);
6552	Assert(pVmxTransient);
6553	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
6554
6555	Log4Func(("VM-entry failure: %Rrc\n", rcVMRun));
6556	switch (rcVMRun)
6557	{
6558	case VERR_VMX_INVALID_VMXON_PTR:
6559	AssertFailed();
6560	break;
6561	case VINF_SUCCESS: /* VMLAUNCH/VMRESUME succeeded but VM-entry failed... yeah, true story. */
6562	case VERR_VMX_UNABLE_TO_START_VM: /* VMLAUNCH/VMRESUME itself failed. */
6563	{
6564	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &pVCpu->hm.s.vmx.LastError.u32ExitReason);
6565	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
6566	AssertRC(rc);
6567	hmR0VmxReadExitQualVmcs(pVmxTransient);
6568
6569	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
6570	/* LastError.idCurrentCpu was already updated in hmR0VmxPreRunGuestCommitted().
6571	Cannot do it here as we may have been long preempted. */
6572
6573	#ifdef VBOX_STRICT
6574	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
6575	Log4(("uExitReason %#RX32 (VmxTransient %#RX16)\n", pVCpu->hm.s.vmx.LastError.u32ExitReason,
6576	pVmxTransient->uExitReason));
6577	Log4(("Exit Qualification %#RX64\n", pVmxTransient->uExitQual));
6578	Log4(("InstrError %#RX32\n", pVCpu->hm.s.vmx.LastError.u32InstrError));
6579	if (pVCpu->hm.s.vmx.LastError.u32InstrError <= HMVMX_INSTR_ERROR_MAX)
6580	Log4(("InstrError Desc. \"%s\"\n", g_apszVmxInstrErrors[pVCpu->hm.s.vmx.LastError.u32InstrError]));
6581	else
6582	Log4(("InstrError Desc. Range exceeded %u\n", HMVMX_INSTR_ERROR_MAX));
6583	Log4(("Entered host CPU %u\n", pVCpu->hm.s.vmx.LastError.idEnteredCpu));
6584	Log4(("Current host CPU %u\n", pVCpu->hm.s.vmx.LastError.idCurrentCpu));
6585
6586	static struct
6587	{
6588	/** Name of the field to log. */
6589	const char *pszName;
6590	/** The VMCS field. */
6591	uint32_t uVmcsField;
6592	/** Whether host support of this field needs to be checked. */
6593	bool fCheckSupport;
6594	} const s_aVmcsFields[] =
6595	{
6596	{ "VMX_VMCS32_CTRL_PIN_EXEC", VMX_VMCS32_CTRL_PIN_EXEC, false },
6597	{ "VMX_VMCS32_CTRL_PROC_EXEC", VMX_VMCS32_CTRL_PROC_EXEC, false },
6598	{ "VMX_VMCS32_CTRL_PROC_EXEC2", VMX_VMCS32_CTRL_PROC_EXEC2, true },
6599	{ "VMX_VMCS32_CTRL_ENTRY", VMX_VMCS32_CTRL_ENTRY, false },
6600	{ "VMX_VMCS32_CTRL_EXIT", VMX_VMCS32_CTRL_EXIT, false },
6601	{ "VMX_VMCS32_CTRL_CR3_TARGET_COUNT", VMX_VMCS32_CTRL_CR3_TARGET_COUNT, false },
6602	{ "VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO", VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, false },
6603	{ "VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE", VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, false },
6604	{ "VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH", VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, false },
6605	{ "VMX_VMCS32_CTRL_TPR_THRESHOLD", VMX_VMCS32_CTRL_TPR_THRESHOLD, false },
6606	{ "VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT", VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, false },
6607	{ "VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT", VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, false },
6608	{ "VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT", VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, false },
6609	{ "VMX_VMCS32_CTRL_EXCEPTION_BITMAP", VMX_VMCS32_CTRL_EXCEPTION_BITMAP, false },
6610	{ "VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK", VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, false },
6611	{ "VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH", VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, false },
6612	{ "VMX_VMCS_CTRL_CR0_MASK", VMX_VMCS_CTRL_CR0_MASK, false },
6613	{ "VMX_VMCS_CTRL_CR0_READ_SHADOW", VMX_VMCS_CTRL_CR0_READ_SHADOW, false },
6614	{ "VMX_VMCS_CTRL_CR4_MASK", VMX_VMCS_CTRL_CR4_MASK, false },
6615	{ "VMX_VMCS_CTRL_CR4_READ_SHADOW", VMX_VMCS_CTRL_CR4_READ_SHADOW, false },
6616	{ "VMX_VMCS64_CTRL_EPTP_FULL", VMX_VMCS64_CTRL_EPTP_FULL, true },
6617	{ "VMX_VMCS_GUEST_RIP", VMX_VMCS_GUEST_RIP, false },
6618	{ "VMX_VMCS_GUEST_RSP", VMX_VMCS_GUEST_RSP, false },
6619	{ "VMX_VMCS_GUEST_RFLAGS", VMX_VMCS_GUEST_RFLAGS, false },
6620	{ "VMX_VMCS16_VPID", VMX_VMCS16_VPID, true, },
6621	{ "VMX_VMCS_HOST_CR0", VMX_VMCS_HOST_CR0, false },
6622	{ "VMX_VMCS_HOST_CR3", VMX_VMCS_HOST_CR3, false },
6623	{ "VMX_VMCS_HOST_CR4", VMX_VMCS_HOST_CR4, false },
6624	/* The order of selector fields below are fixed! */
6625	{ "VMX_VMCS16_HOST_ES_SEL", VMX_VMCS16_HOST_ES_SEL, false },
6626	{ "VMX_VMCS16_HOST_CS_SEL", VMX_VMCS16_HOST_CS_SEL, false },
6627	{ "VMX_VMCS16_HOST_SS_SEL", VMX_VMCS16_HOST_SS_SEL, false },
6628	{ "VMX_VMCS16_HOST_DS_SEL", VMX_VMCS16_HOST_DS_SEL, false },
6629	{ "VMX_VMCS16_HOST_FS_SEL", VMX_VMCS16_HOST_FS_SEL, false },
6630	{ "VMX_VMCS16_HOST_GS_SEL", VMX_VMCS16_HOST_GS_SEL, false },
6631	{ "VMX_VMCS16_HOST_TR_SEL", VMX_VMCS16_HOST_TR_SEL, false },
6632	/* End of ordered selector fields. */
6633	{ "VMX_VMCS_HOST_TR_BASE", VMX_VMCS_HOST_TR_BASE, false },
6634	{ "VMX_VMCS_HOST_GDTR_BASE", VMX_VMCS_HOST_GDTR_BASE, false },
6635	{ "VMX_VMCS_HOST_IDTR_BASE", VMX_VMCS_HOST_IDTR_BASE, false },
6636	{ "VMX_VMCS32_HOST_SYSENTER_CS", VMX_VMCS32_HOST_SYSENTER_CS, false },
6637	{ "VMX_VMCS_HOST_SYSENTER_EIP", VMX_VMCS_HOST_SYSENTER_EIP, false },
6638	{ "VMX_VMCS_HOST_SYSENTER_ESP", VMX_VMCS_HOST_SYSENTER_ESP, false },
6639	{ "VMX_VMCS_HOST_RSP", VMX_VMCS_HOST_RSP, false },
6640	{ "VMX_VMCS_HOST_RIP", VMX_VMCS_HOST_RIP, false }
6641	};
6642
6643	RTGDTR HostGdtr;
6644	ASMGetGDTR(&HostGdtr);
6645
6646	uint32_t const cVmcsFields = RT_ELEMENTS(s_aVmcsFields);
6647	for (uint32_t i = 0; i < cVmcsFields; i++)
6648	{
6649	uint32_t const uVmcsField = s_aVmcsFields[i].uVmcsField;
6650
6651	bool fSupported;
6652	if (!s_aVmcsFields[i].fCheckSupport)
6653	fSupported = true;
6654	else
6655	{
6656	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6657	switch (uVmcsField)
6658	{
6659	case VMX_VMCS64_CTRL_EPTP_FULL: fSupported = pVM->hm.s.fNestedPaging; break;
6660	case VMX_VMCS16_VPID: fSupported = pVM->hm.s.vmx.fVpid; break;
6661	case VMX_VMCS32_CTRL_PROC_EXEC2:
6662	fSupported = RT_BOOL(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS);
6663	break;
6664	default:
6665	AssertMsgFailedReturnVoid(("Failed to provide VMCS field support for %#RX32\n", uVmcsField));
6666	}
6667	}
6668
6669	if (fSupported)
6670	{
6671	uint8_t const uWidth = RT_BF_GET(uVmcsField, VMX_BF_VMCSFIELD_WIDTH);
6672	switch (uWidth)
6673	{
6674	case VMX_VMCSFIELD_WIDTH_16BIT:
6675	{
6676	uint16_t u16Val;
6677	rc = VMXReadVmcs16(uVmcsField, &u16Val);
6678	AssertRC(rc);
6679	Log4(("%-40s = %#RX16\n", s_aVmcsFields[i].pszName, u16Val));
6680
6681	if ( uVmcsField >= VMX_VMCS16_HOST_ES_SEL
6682	&& uVmcsField <= VMX_VMCS16_HOST_TR_SEL)
6683	{
6684	if (u16Val < HostGdtr.cbGdt)
6685	{
6686	/* Order of selectors in s_apszSel is fixed and matches the order in s_aVmcsFields. */
6687	static const char * const s_apszSel[] = { "Host ES", "Host CS", "Host SS", "Host DS",
6688	"Host FS", "Host GS", "Host TR" };
6689	uint8_t const idxSel = RT_BF_GET(uVmcsField, VMX_BF_VMCSFIELD_INDEX);
6690	Assert(idxSel < RT_ELEMENTS(s_apszSel));
6691	PCX86DESCHC pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u16Val & X86_SEL_MASK));
6692	hmR0DumpDescriptor(pDesc, u16Val, s_apszSel[idxSel]);
6693	}
6694	else
6695	Log4((" Selector value exceeds GDT limit!\n"));
6696	}
6697	break;
6698	}
6699
6700	case VMX_VMCSFIELD_WIDTH_32BIT:
6701	{
6702	uint32_t u32Val;
6703	rc = VMXReadVmcs32(uVmcsField, &u32Val);
6704	AssertRC(rc);
6705	Log4(("%-40s = %#RX32\n", s_aVmcsFields[i].pszName, u32Val));
6706	break;
6707	}
6708
6709	case VMX_VMCSFIELD_WIDTH_64BIT:
6710	case VMX_VMCSFIELD_WIDTH_NATURAL:
6711	{
6712	uint64_t u64Val;
6713	rc = VMXReadVmcs64(uVmcsField, &u64Val);
6714	AssertRC(rc);
6715	Log4(("%-40s = %#RX64\n", s_aVmcsFields[i].pszName, u64Val));
6716	break;
6717	}
6718	}
6719	}
6720	}
6721
6722	Log4(("MSR_K6_EFER = %#RX64\n", ASMRdMsr(MSR_K6_EFER)));
6723	Log4(("MSR_K8_CSTAR = %#RX64\n", ASMRdMsr(MSR_K8_CSTAR)));
6724	Log4(("MSR_K8_LSTAR = %#RX64\n", ASMRdMsr(MSR_K8_LSTAR)));
6725	Log4(("MSR_K6_STAR = %#RX64\n", ASMRdMsr(MSR_K6_STAR)));
6726	Log4(("MSR_K8_SF_MASK = %#RX64\n", ASMRdMsr(MSR_K8_SF_MASK)));
6727	Log4(("MSR_K8_KERNEL_GS_BASE = %#RX64\n", ASMRdMsr(MSR_K8_KERNEL_GS_BASE)));
6728	#endif /* VBOX_STRICT */
6729	break;
6730	}
6731
6732	default:
6733	/* Impossible */
6734	AssertMsgFailed(("hmR0VmxReportWorldSwitchError %Rrc (%#x)\n", rcVMRun, rcVMRun));
6735	break;
6736	}
6737	}
6738
6739
6740	/**
6741	* Sets up the usage of TSC-offsetting and updates the VMCS.
6742	*
6743	* If offsetting is not possible, cause VM-exits on RDTSC(P)s. Also sets up the
6744	* VMX-preemption timer.
6745	*
6746	* @returns VBox status code.
6747	* @param pVCpu The cross context virtual CPU structure.
6748	* @param pVmxTransient The VMX-transient structure.
6749	*
6750	* @remarks No-long-jump zone!!!
6751	*/
6752	static void hmR0VmxUpdateTscOffsettingAndPreemptTimer(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
6753	{
6754	bool fOffsettedTsc;
6755	bool fParavirtTsc;
6756	uint64_t uTscOffset;
6757	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6758	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
6759
6760	if (pVM->hm.s.vmx.fUsePreemptTimer)
6761	{
6762	uint64_t cTicksToDeadline = TMCpuTickGetDeadlineAndTscOffset(pVM, pVCpu, &uTscOffset, &fOffsettedTsc, &fParavirtTsc);
6763
6764	/* Make sure the returned values have sane upper and lower boundaries. */
6765	uint64_t u64CpuHz = SUPGetCpuHzFromGipBySetIndex(g_pSUPGlobalInfoPage, pVCpu->iHostCpuSet);
6766	cTicksToDeadline = RT_MIN(cTicksToDeadline, u64CpuHz / 64); /* 1/64th of a second */
6767	cTicksToDeadline = RT_MAX(cTicksToDeadline, u64CpuHz / 2048); /* 1/2048th of a second */
6768	cTicksToDeadline >>= pVM->hm.s.vmx.cPreemptTimerShift;
6769
6770	/** @todo r=ramshankar: We need to find a way to integrate nested-guest
6771	* preemption timers here. We probably need to clamp the preemption timer,
6772	* after converting the timer value to the host. */
6773	uint32_t const cPreemptionTickCount = (uint32_t)RT_MIN(cTicksToDeadline, UINT32_MAX - 16);
6774	int rc = VMXWriteVmcs32(VMX_VMCS32_PREEMPT_TIMER_VALUE, cPreemptionTickCount);
6775	AssertRC(rc);
6776	}
6777	else
6778	fOffsettedTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &uTscOffset, &fParavirtTsc);
6779
6780	if (fParavirtTsc)
6781	{
6782	/* Currently neither Hyper-V nor KVM need to update their paravirt. TSC
6783	information before every VM-entry, hence disable it for performance sake. */
6784	#if 0
6785	int rc = GIMR0UpdateParavirtTsc(pVM, 0 /* u64Offset */);
6786	AssertRC(rc);
6787	#endif
6788	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
6789	}
6790
6791	if ( fOffsettedTsc
6792	&& RT_LIKELY(!pVCpu->hm.s.fDebugWantRdTscExit))
6793	{
6794	if (pVmxTransient->fIsNestedGuest)
6795	uTscOffset = CPUMApplyNestedGuestTscOffset(pVCpu, uTscOffset);
6796	hmR0VmxSetTscOffsetVmcs(pVmcsInfo, uTscOffset);
6797	hmR0VmxRemoveProcCtlsVmcs(pVCpu, pVmxTransient, VMX_PROC_CTLS_RDTSC_EXIT);
6798	}
6799	else
6800	{
6801	/* We can't use TSC-offsetting (non-fixed TSC, warp drive active etc.), VM-exit on RDTSC(P). */
6802	hmR0VmxSetProcCtlsVmcs(pVmxTransient, VMX_PROC_CTLS_RDTSC_EXIT);
6803	}
6804	}
6805
6806
6807	/**
6808	* Gets the IEM exception flags for the specified vector and IDT vectoring /
6809	* VM-exit interruption info type.
6810	*
6811	* @returns The IEM exception flags.
6812	* @param uVector The event vector.
6813	* @param uVmxEventType The VMX event type.
6814	*
6815	* @remarks This function currently only constructs flags required for
6816	* IEMEvaluateRecursiveXcpt and not the complete flags (e.g, error-code
6817	* and CR2 aspects of an exception are not included).
6818	*/
6819	static uint32_t hmR0VmxGetIemXcptFlags(uint8_t uVector, uint32_t uVmxEventType)
6820	{
6821	uint32_t fIemXcptFlags;
6822	switch (uVmxEventType)
6823	{
6824	case VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT:
6825	case VMX_IDT_VECTORING_INFO_TYPE_NMI:
6826	fIemXcptFlags = IEM_XCPT_FLAGS_T_CPU_XCPT;
6827	break;
6828
6829	case VMX_IDT_VECTORING_INFO_TYPE_EXT_INT:
6830	fIemXcptFlags = IEM_XCPT_FLAGS_T_EXT_INT;
6831	break;
6832
6833	case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
6834	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT \| IEM_XCPT_FLAGS_ICEBP_INSTR;
6835	break;
6836
6837	case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT:
6838	{
6839	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
6840	if (uVector == X86_XCPT_BP)
6841	fIemXcptFlags \|= IEM_XCPT_FLAGS_BP_INSTR;
6842	else if (uVector == X86_XCPT_OF)
6843	fIemXcptFlags \|= IEM_XCPT_FLAGS_OF_INSTR;
6844	else
6845	{
6846	fIemXcptFlags = 0;
6847	AssertMsgFailed(("Unexpected vector for software exception. uVector=%#x", uVector));
6848	}
6849	break;
6850	}
6851
6852	case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
6853	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
6854	break;
6855
6856	default:
6857	fIemXcptFlags = 0;
6858	AssertMsgFailed(("Unexpected vector type! uVmxEventType=%#x uVector=%#x", uVmxEventType, uVector));
6859	break;
6860	}
6861	return fIemXcptFlags;
6862	}
6863
6864
6865	/**
6866	* Sets an event as a pending event to be injected into the guest.
6867	*
6868	* @param pVCpu The cross context virtual CPU structure.
6869	* @param u32IntInfo The VM-entry interruption-information field.
6870	* @param cbInstr The VM-entry instruction length in bytes (for
6871	* software interrupts, exceptions and privileged
6872	* software exceptions).
6873	* @param u32ErrCode The VM-entry exception error code.
6874	* @param GCPtrFaultAddress The fault-address (CR2) in case it's a
6875	* page-fault.
6876	*/
6877	DECLINLINE(void) hmR0VmxSetPendingEvent(PVMCPUCC pVCpu, uint32_t u32IntInfo, uint32_t cbInstr, uint32_t u32ErrCode,
6878	RTGCUINTPTR GCPtrFaultAddress)
6879	{
6880	Assert(!pVCpu->hm.s.Event.fPending);
6881	pVCpu->hm.s.Event.fPending = true;
6882	pVCpu->hm.s.Event.u64IntInfo = u32IntInfo;
6883	pVCpu->hm.s.Event.u32ErrCode = u32ErrCode;
6884	pVCpu->hm.s.Event.cbInstr = cbInstr;
6885	pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress;
6886	}
6887
6888
6889	/**
6890	* Sets an external interrupt as pending-for-injection into the VM.
6891	*
6892	* @param pVCpu The cross context virtual CPU structure.
6893	* @param u8Interrupt The external interrupt vector.
6894	*/
6895	DECLINLINE(void) hmR0VmxSetPendingExtInt(PVMCPUCC pVCpu, uint8_t u8Interrupt)
6896	{
6897	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, u8Interrupt)
6898	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
6899	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
6900	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
6901	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
6902	}
6903
6904
6905	/**
6906	* Sets an NMI (\#NMI) exception as pending-for-injection into the VM.
6907	*
6908	* @param pVCpu The cross context virtual CPU structure.
6909	*/
6910	DECLINLINE(void) hmR0VmxSetPendingXcptNmi(PVMCPUCC pVCpu)
6911	{
6912	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_NMI)
6913	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_NMI)
6914	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
6915	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
6916	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
6917	}
6918
6919
6920	/**
6921	* Sets a double-fault (\#DF) exception as pending-for-injection into the VM.
6922	*
6923	* @param pVCpu The cross context virtual CPU structure.
6924	*/
6925	DECLINLINE(void) hmR0VmxSetPendingXcptDF(PVMCPUCC pVCpu)
6926	{
6927	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DF)
6928	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
6929	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
6930	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
6931	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
6932	}
6933
6934
6935	/**
6936	* Sets an invalid-opcode (\#UD) exception as pending-for-injection into the VM.
6937	*
6938	* @param pVCpu The cross context virtual CPU structure.
6939	*/
6940	DECLINLINE(void) hmR0VmxSetPendingXcptUD(PVMCPUCC pVCpu)
6941	{
6942	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_UD)
6943	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
6944	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
6945	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
6946	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
6947	}
6948
6949
6950	/**
6951	* Sets a debug (\#DB) exception as pending-for-injection into the VM.
6952	*
6953	* @param pVCpu The cross context virtual CPU structure.
6954	*/
6955	DECLINLINE(void) hmR0VmxSetPendingXcptDB(PVMCPUCC pVCpu)
6956	{
6957	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DB)
6958	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
6959	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
6960	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
6961	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
6962	}
6963
6964
6965	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
6966	/**
6967	* Sets a general-protection (\#GP) exception as pending-for-injection into the VM.
6968	*
6969	* @param pVCpu The cross context virtual CPU structure.
6970	* @param u32ErrCode The error code for the general-protection exception.
6971	*/
6972	DECLINLINE(void) hmR0VmxSetPendingXcptGP(PVMCPUCC pVCpu, uint32_t u32ErrCode)
6973	{
6974	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_GP)
6975	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
6976	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
6977	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
6978	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrCode, 0 / GCPtrFaultAddress */);
6979	}
6980
6981
6982	/**
6983	* Sets a stack (\#SS) exception as pending-for-injection into the VM.
6984	*
6985	* @param pVCpu The cross context virtual CPU structure.
6986	* @param u32ErrCode The error code for the stack exception.
6987	*/
6988	DECLINLINE(void) hmR0VmxSetPendingXcptSS(PVMCPUCC pVCpu, uint32_t u32ErrCode)
6989	{
6990	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_SS)
6991	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
6992	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
6993	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
6994	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrCode, 0 / GCPtrFaultAddress */);
6995	}
6996	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
6997
6998
6999	/**
7000	* Fixes up attributes for the specified segment register.
7001	*
7002	* @param pVCpu The cross context virtual CPU structure.
7003	* @param pSelReg The segment register that needs fixing.
7004	* @param idxSel The VMCS field for the corresponding segment register.
7005	*/
7006	static void hmR0VmxFixUnusableSegRegAttr(PVMCPUCC pVCpu, PCPUMSELREG pSelReg, uint32_t idxSel)
7007	{
7008	Assert(pSelReg->Attr.u & X86DESCATTR_UNUSABLE);
7009
7010	/*
7011	* If VT-x marks the segment as unusable, most other bits remain undefined:
7012	* - For CS the L, D and G bits have meaning.
7013	* - For SS the DPL has meaning (it -is- the CPL for Intel and VBox).
7014	* - For the remaining data segments no bits are defined.
7015	*
7016	* The present bit and the unusable bit has been observed to be set at the
7017	* same time (the selector was supposed to be invalid as we started executing
7018	* a V8086 interrupt in ring-0).
7019	*
7020	* What should be important for the rest of the VBox code, is that the P bit is
7021	* cleared. Some of the other VBox code recognizes the unusable bit, but
7022	* AMD-V certainly don't, and REM doesn't really either. So, to be on the
7023	* safe side here, we'll strip off P and other bits we don't care about. If
7024	* any code breaks because Attr.u != 0 when Sel < 4, it should be fixed.
7025	*
7026	* See Intel spec. 27.3.2 "Saving Segment Registers and Descriptor-Table Registers".
7027	*/
7028	#ifdef VBOX_STRICT
7029	uint32_t const uAttr = pSelReg->Attr.u;
7030	#endif
7031
7032	/* Masking off: X86DESCATTR_P, X86DESCATTR_LIMIT_HIGH, and X86DESCATTR_AVL. The latter two are really irrelevant. */
7033	pSelReg->Attr.u &= X86DESCATTR_UNUSABLE \| X86DESCATTR_L \| X86DESCATTR_D \| X86DESCATTR_G
7034	\| X86DESCATTR_DPL \| X86DESCATTR_TYPE \| X86DESCATTR_DT;
7035
7036	#ifdef VBOX_STRICT
7037	VMMRZCallRing3Disable(pVCpu);
7038	Log4Func(("Unusable %#x: sel=%#x attr=%#x -> %#x\n", idxSel, pSelReg->Sel, uAttr, pSelReg->Attr.u));
7039	# ifdef DEBUG_bird
7040	AssertMsg((uAttr & ~X86DESCATTR_P) == pSelReg->Attr.u,
7041	("%#x: %#x != %#x (sel=%#x base=%#llx limit=%#x)\n",
7042	idxSel, uAttr, pSelReg->Attr.u, pSelReg->Sel, pSelReg->u64Base, pSelReg->u32Limit));
7043	# endif
7044	VMMRZCallRing3Enable(pVCpu);
7045	NOREF(uAttr);
7046	#endif
7047	RT_NOREF2(pVCpu, idxSel);
7048	}
7049
7050
7051	/**
7052	* Imports a guest segment register from the current VMCS into the guest-CPU
7053	* context.
7054	*
7055	* @param pVCpu The cross context virtual CPU structure.
7056	* @param iSegReg The segment register number (X86_SREG_XXX).
7057	*
7058	* @remarks Called with interrupts and/or preemption disabled.
7059	*/
7060	static void hmR0VmxImportGuestSegReg(PVMCPUCC pVCpu, uint8_t iSegReg)
7061	{
7062	Assert(iSegReg < X86_SREG_COUNT);
7063
7064	uint32_t const idxSel = g_aVmcsSegSel[iSegReg];
7065	uint32_t const idxLimit = g_aVmcsSegLimit[iSegReg];
7066	uint32_t const idxAttr = g_aVmcsSegAttr[iSegReg];
7067	uint32_t const idxBase = g_aVmcsSegBase[iSegReg];
7068
7069	uint16_t u16Sel;
7070	uint64_t u64Base;
7071	uint32_t u32Limit, u32Attr;
7072	int rc = VMXReadVmcs16(idxSel, &u16Sel); AssertRC(rc);
7073	rc = VMXReadVmcs32(idxLimit, &u32Limit); AssertRC(rc);
7074	rc = VMXReadVmcs32(idxAttr, &u32Attr); AssertRC(rc);
7075	rc = VMXReadVmcsNw(idxBase, &u64Base); AssertRC(rc);
7076
7077	PCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
7078	pSelReg->Sel = u16Sel;
7079	pSelReg->ValidSel = u16Sel;
7080	pSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
7081	pSelReg->u32Limit = u32Limit;
7082	pSelReg->u64Base = u64Base;
7083	pSelReg->Attr.u = u32Attr;
7084	if (u32Attr & X86DESCATTR_UNUSABLE)
7085	hmR0VmxFixUnusableSegRegAttr(pVCpu, pSelReg, idxSel);
7086	}
7087
7088
7089	/**
7090	* Imports the guest LDTR from the current VMCS into the guest-CPU context.
7091	*
7092	* @param pVCpu The cross context virtual CPU structure.
7093	*
7094	* @remarks Called with interrupts and/or preemption disabled.
7095	*/
7096	static void hmR0VmxImportGuestLdtr(PVMCPUCC pVCpu)
7097	{
7098	uint16_t u16Sel;
7099	uint64_t u64Base;
7100	uint32_t u32Limit, u32Attr;
7101	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_LDTR_SEL, &u16Sel); AssertRC(rc);
7102	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, &u32Limit); AssertRC(rc);
7103	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, &u32Attr); AssertRC(rc);
7104	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_LDTR_BASE, &u64Base); AssertRC(rc);
7105
7106	pVCpu->cpum.GstCtx.ldtr.Sel = u16Sel;
7107	pVCpu->cpum.GstCtx.ldtr.ValidSel = u16Sel;
7108	pVCpu->cpum.GstCtx.ldtr.fFlags = CPUMSELREG_FLAGS_VALID;
7109	pVCpu->cpum.GstCtx.ldtr.u32Limit = u32Limit;
7110	pVCpu->cpum.GstCtx.ldtr.u64Base = u64Base;
7111	pVCpu->cpum.GstCtx.ldtr.Attr.u = u32Attr;
7112	if (u32Attr & X86DESCATTR_UNUSABLE)
7113	hmR0VmxFixUnusableSegRegAttr(pVCpu, &pVCpu->cpum.GstCtx.ldtr, VMX_VMCS16_GUEST_LDTR_SEL);
7114	}
7115
7116
7117	/**
7118	* Imports the guest TR from the current VMCS into the guest-CPU context.
7119	*
7120	* @param pVCpu The cross context virtual CPU structure.
7121	*
7122	* @remarks Called with interrupts and/or preemption disabled.
7123	*/
7124	static void hmR0VmxImportGuestTr(PVMCPUCC pVCpu)
7125	{
7126	uint16_t u16Sel;
7127	uint64_t u64Base;
7128	uint32_t u32Limit, u32Attr;
7129	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_TR_SEL, &u16Sel); AssertRC(rc);
7130	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, &u32Limit); AssertRC(rc);
7131	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, &u32Attr); AssertRC(rc);
7132	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_TR_BASE, &u64Base); AssertRC(rc);
7133
7134	pVCpu->cpum.GstCtx.tr.Sel = u16Sel;
7135	pVCpu->cpum.GstCtx.tr.ValidSel = u16Sel;
7136	pVCpu->cpum.GstCtx.tr.fFlags = CPUMSELREG_FLAGS_VALID;
7137	pVCpu->cpum.GstCtx.tr.u32Limit = u32Limit;
7138	pVCpu->cpum.GstCtx.tr.u64Base = u64Base;
7139	pVCpu->cpum.GstCtx.tr.Attr.u = u32Attr;
7140	/* TR is the only selector that can never be unusable. */
7141	Assert(!(u32Attr & X86DESCATTR_UNUSABLE));
7142	}
7143
7144
7145	/**
7146	* Imports the guest RIP from the VMCS back into the guest-CPU context.
7147	*
7148	* @param pVCpu The cross context virtual CPU structure.
7149	*
7150	* @remarks Called with interrupts and/or preemption disabled, should not assert!
7151	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7152	* instead!!!
7153	*/
7154	static void hmR0VmxImportGuestRip(PVMCPUCC pVCpu)
7155	{
7156	uint64_t u64Val;
7157	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7158	if (pCtx->fExtrn & CPUMCTX_EXTRN_RIP)
7159	{
7160	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RIP, &u64Val);
7161	AssertRC(rc);
7162
7163	pCtx->rip = u64Val;
7164	EMR0HistoryUpdatePC(pVCpu, pCtx->rip, false);
7165	pCtx->fExtrn &= ~CPUMCTX_EXTRN_RIP;
7166	}
7167	}
7168
7169
7170	/**
7171	* Imports the guest RFLAGS from the VMCS back into the guest-CPU context.
7172	*
7173	* @param pVCpu The cross context virtual CPU structure.
7174	* @param pVmcsInfo The VMCS info. object.
7175	*
7176	* @remarks Called with interrupts and/or preemption disabled, should not assert!
7177	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7178	* instead!!!
7179	*/
7180	static void hmR0VmxImportGuestRFlags(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
7181	{
7182	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7183	if (pCtx->fExtrn & CPUMCTX_EXTRN_RFLAGS)
7184	{
7185	uint64_t u64Val;
7186	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RFLAGS, &u64Val);
7187	AssertRC(rc);
7188
7189	pCtx->rflags.u64 = u64Val;
7190	if (pVmcsInfo->RealMode.fRealOnV86Active)
7191	{
7192	pCtx->eflags.Bits.u1VM = 0;
7193	pCtx->eflags.Bits.u2IOPL = pVmcsInfo->RealMode.Eflags.Bits.u2IOPL;
7194	}
7195	pCtx->fExtrn &= ~CPUMCTX_EXTRN_RFLAGS;
7196	}
7197	}
7198
7199
7200	/**
7201	* Imports the guest interruptibility-state from the VMCS back into the guest-CPU
7202	* context.
7203	*
7204	* @param pVCpu The cross context virtual CPU structure.
7205	* @param pVmcsInfo The VMCS info. object.
7206	*
7207	* @remarks Called with interrupts and/or preemption disabled, try not to assert and
7208	* do not log!
7209	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7210	* instead!!!
7211	*/
7212	static void hmR0VmxImportGuestIntrState(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
7213	{
7214	uint32_t u32Val;
7215	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &u32Val); AssertRC(rc);
7216	if (!u32Val)
7217	{
7218	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
7219	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
7220	CPUMSetGuestNmiBlocking(pVCpu, false);
7221	}
7222	else
7223	{
7224	/*
7225	* We must import RIP here to set our EM interrupt-inhibited state.
7226	* We also import RFLAGS as our code that evaluates pending interrupts
7227	* before VM-entry requires it.
7228	*/
7229	hmR0VmxImportGuestRip(pVCpu);
7230	hmR0VmxImportGuestRFlags(pVCpu, pVmcsInfo);
7231
7232	if (u32Val & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS \| VMX_VMCS_GUEST_INT_STATE_BLOCK_STI))
7233	EMSetInhibitInterruptsPC(pVCpu, pVCpu->cpum.GstCtx.rip);
7234	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
7235	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
7236
7237	bool const fNmiBlocking = RT_BOOL(u32Val & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI);
7238	CPUMSetGuestNmiBlocking(pVCpu, fNmiBlocking);
7239	}
7240	}
7241
7242
7243	/**
7244	* Worker for VMXR0ImportStateOnDemand.
7245	*
7246	* @returns VBox status code.
7247	* @param pVCpu The cross context virtual CPU structure.
7248	* @param pVmcsInfo The VMCS info. object.
7249	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
7250	*/
7251	static int hmR0VmxImportGuestState(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint64_t fWhat)
7252	{
7253	int rc = VINF_SUCCESS;
7254	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7255	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7256	uint32_t u32Val;
7257
7258	/*
7259	* Note! This is hack to workaround a mysterious BSOD observed with release builds
7260	* on Windows 10 64-bit hosts. Profile and debug builds are not affected and
7261	* neither are other host platforms.
7262	*
7263	* Committing this temporarily as it prevents BSOD.
7264	*
7265	* Update: This is very likely a compiler optimization bug, see @bugref{9180}.
7266	*/
7267	#ifdef RT_OS_WINDOWS
7268	if (pVM == 0 \|\| pVM == (void *)(uintptr_t)-1)
7269	return VERR_HM_IPE_1;
7270	#endif
7271
7272	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatImportGuestState, x);
7273
7274	/*
7275	* We disable interrupts to make the updating of the state and in particular
7276	* the fExtrn modification atomic wrt to preemption hooks.
7277	*/
7278	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
7279
7280	fWhat &= pCtx->fExtrn;
7281	if (fWhat)
7282	{
7283	do
7284	{
7285	if (fWhat & CPUMCTX_EXTRN_RIP)
7286	hmR0VmxImportGuestRip(pVCpu);
7287
7288	if (fWhat & CPUMCTX_EXTRN_RFLAGS)
7289	hmR0VmxImportGuestRFlags(pVCpu, pVmcsInfo);
7290
7291	if (fWhat & CPUMCTX_EXTRN_HM_VMX_INT_STATE)
7292	hmR0VmxImportGuestIntrState(pVCpu, pVmcsInfo);
7293
7294	if (fWhat & CPUMCTX_EXTRN_RSP)
7295	{
7296	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RSP, &pCtx->rsp);
7297	AssertRC(rc);
7298	}
7299
7300	if (fWhat & CPUMCTX_EXTRN_SREG_MASK)
7301	{
7302	bool const fRealOnV86Active = pVmcsInfo->RealMode.fRealOnV86Active;
7303	if (fWhat & CPUMCTX_EXTRN_CS)
7304	{
7305	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_CS);
7306	hmR0VmxImportGuestRip(pVCpu);
7307	if (fRealOnV86Active)
7308	pCtx->cs.Attr.u = pVmcsInfo->RealMode.AttrCS.u;
7309	EMR0HistoryUpdatePC(pVCpu, pCtx->cs.u64Base + pCtx->rip, true /* fFlattened */);
7310	}
7311	if (fWhat & CPUMCTX_EXTRN_SS)
7312	{
7313	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_SS);
7314	if (fRealOnV86Active)
7315	pCtx->ss.Attr.u = pVmcsInfo->RealMode.AttrSS.u;
7316	}
7317	if (fWhat & CPUMCTX_EXTRN_DS)
7318	{
7319	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_DS);
7320	if (fRealOnV86Active)
7321	pCtx->ds.Attr.u = pVmcsInfo->RealMode.AttrDS.u;
7322	}
7323	if (fWhat & CPUMCTX_EXTRN_ES)
7324	{
7325	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_ES);
7326	if (fRealOnV86Active)
7327	pCtx->es.Attr.u = pVmcsInfo->RealMode.AttrES.u;
7328	}
7329	if (fWhat & CPUMCTX_EXTRN_FS)
7330	{
7331	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_FS);
7332	if (fRealOnV86Active)
7333	pCtx->fs.Attr.u = pVmcsInfo->RealMode.AttrFS.u;
7334	}
7335	if (fWhat & CPUMCTX_EXTRN_GS)
7336	{
7337	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_GS);
7338	if (fRealOnV86Active)
7339	pCtx->gs.Attr.u = pVmcsInfo->RealMode.AttrGS.u;
7340	}
7341	}
7342
7343	if (fWhat & CPUMCTX_EXTRN_TABLE_MASK)
7344	{
7345	if (fWhat & CPUMCTX_EXTRN_LDTR)
7346	hmR0VmxImportGuestLdtr(pVCpu);
7347
7348	if (fWhat & CPUMCTX_EXTRN_GDTR)
7349	{
7350	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, &pCtx->gdtr.pGdt); AssertRC(rc);
7351	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val); AssertRC(rc);
7352	pCtx->gdtr.cbGdt = u32Val;
7353	}
7354
7355	/* Guest IDTR. */
7356	if (fWhat & CPUMCTX_EXTRN_IDTR)
7357	{
7358	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, &pCtx->idtr.pIdt); AssertRC(rc);
7359	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val); AssertRC(rc);
7360	pCtx->idtr.cbIdt = u32Val;
7361	}
7362
7363	/* Guest TR. */
7364	if (fWhat & CPUMCTX_EXTRN_TR)
7365	{
7366	/* Real-mode emulation using virtual-8086 mode has the fake TSS (pRealModeTSS) in TR,
7367	don't need to import that one. */
7368	if (!pVmcsInfo->RealMode.fRealOnV86Active)
7369	hmR0VmxImportGuestTr(pVCpu);
7370	}
7371	}
7372
7373	if (fWhat & CPUMCTX_EXTRN_DR7)
7374	{
7375	if (!pVCpu->hm.s.fUsingHyperDR7)
7376	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_DR7, &pCtx->dr[7]); AssertRC(rc);
7377	}
7378
7379	if (fWhat & CPUMCTX_EXTRN_SYSENTER_MSRS)
7380	{
7381	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SYSENTER_EIP, &pCtx->SysEnter.eip); AssertRC(rc);
7382	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SYSENTER_ESP, &pCtx->SysEnter.esp); AssertRC(rc);
7383	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, &u32Val); AssertRC(rc);
7384	pCtx->SysEnter.cs = u32Val;
7385	}
7386
7387	if (fWhat & CPUMCTX_EXTRN_KERNEL_GS_BASE)
7388	{
7389	if ( pVM->hm.s.fAllow64BitGuests
7390	&& (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
7391	pCtx->msrKERNELGSBASE = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
7392	}
7393
7394	if (fWhat & CPUMCTX_EXTRN_SYSCALL_MSRS)
7395	{
7396	if ( pVM->hm.s.fAllow64BitGuests
7397	&& (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
7398	{
7399	pCtx->msrLSTAR = ASMRdMsr(MSR_K8_LSTAR);
7400	pCtx->msrSTAR = ASMRdMsr(MSR_K6_STAR);
7401	pCtx->msrSFMASK = ASMRdMsr(MSR_K8_SF_MASK);
7402	}
7403	}
7404
7405	if (fWhat & (CPUMCTX_EXTRN_TSC_AUX \| CPUMCTX_EXTRN_OTHER_MSRS))
7406	{
7407	PCVMXAUTOMSR pMsrs = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
7408	uint32_t const cMsrs = pVmcsInfo->cExitMsrStore;
7409	Assert(pMsrs);
7410	Assert(cMsrs <= VMX_MISC_MAX_MSRS(pVM->hm.s.vmx.Msrs.u64Misc));
7411	Assert(sizeof(pMsrs) cMsrs <= X86_PAGE_4K_SIZE);
7412	for (uint32_t i = 0; i < cMsrs; i++)
7413	{
7414	uint32_t const idMsr = pMsrs[i].u32Msr;
7415	switch (idMsr)
7416	{
7417	case MSR_K8_TSC_AUX: CPUMSetGuestTscAux(pVCpu, pMsrs[i].u64Value); break;
7418	case MSR_IA32_SPEC_CTRL: CPUMSetGuestSpecCtrl(pVCpu, pMsrs[i].u64Value); break;
7419	case MSR_K6_EFER: /* Can't be changed without causing a VM-exit */ break;
7420	default:
7421	{
7422	pCtx->fExtrn = 0;
7423	pVCpu->hm.s.u32HMError = pMsrs->u32Msr;
7424	ASMSetFlags(fEFlags);
7425	AssertMsgFailed(("Unexpected MSR in auto-load/store area. idMsr=%#RX32 cMsrs=%u\n", idMsr, cMsrs));
7426	return VERR_HM_UNEXPECTED_LD_ST_MSR;
7427	}
7428	}
7429	}
7430	}
7431
7432	if (fWhat & CPUMCTX_EXTRN_CR_MASK)
7433	{
7434	if (fWhat & CPUMCTX_EXTRN_CR0)
7435	{
7436	uint64_t u64Cr0;
7437	uint64_t u64Shadow;
7438	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64Cr0); AssertRC(rc);
7439	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, &u64Shadow); AssertRC(rc);
7440	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7441	\| (u64Shadow & pVmcsInfo->u64Cr0Mask);
7442	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7443	/*
7444	* Reapply the nested-guest's CR0 fixed bits that might have been altered while
7445	* exporting the nested-guest CR0 for executing using hardware-assisted VMX.
7446	*/
7447	if (CPUMIsGuestInVmxNonRootMode(pCtx))
7448	{
7449	u64Cr0 \|= pCtx->hwvirt.vmx.Msrs.u64Cr0Fixed0;
7450	u64Cr0 &= pCtx->hwvirt.vmx.Msrs.u64Cr0Fixed1;
7451	}
7452	#endif
7453	VMMRZCallRing3Disable(pVCpu); /* May call into PGM which has Log statements. */
7454	CPUMSetGuestCR0(pVCpu, u64Cr0);
7455	VMMRZCallRing3Enable(pVCpu);
7456	}
7457
7458	if (fWhat & CPUMCTX_EXTRN_CR4)
7459	{
7460	uint64_t u64Cr4;
7461	uint64_t u64Shadow;
7462	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR4, &u64Cr4); AssertRC(rc);
7463	rc \|= VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, &u64Shadow); AssertRC(rc);
7464	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7465	\| (u64Shadow & pVmcsInfo->u64Cr4Mask);
7466	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7467	/*
7468	* Reapply the nested-guest's CR4 fixed bits that might have been altered while
7469	* exporting the nested-guest CR4 for executing using hardware-assisted VMX.
7470	*/
7471	if (CPUMIsGuestInVmxNonRootMode(pCtx))
7472	{
7473	u64Cr4 \|= pCtx->hwvirt.vmx.Msrs.u64Cr4Fixed0;
7474	u64Cr4 &= pCtx->hwvirt.vmx.Msrs.u64Cr4Fixed1;
7475	}
7476	#endif
7477	pCtx->cr4 = u64Cr4;
7478	}
7479
7480	if (fWhat & CPUMCTX_EXTRN_CR3)
7481	{
7482	/* CR0.PG bit changes are always intercepted, so it's up to date. */
7483	if ( pVM->hm.s.vmx.fUnrestrictedGuest
7484	\|\| ( pVM->hm.s.fNestedPaging
7485	&& CPUMIsGuestPagingEnabledEx(pCtx)))
7486	{
7487	uint64_t u64Cr3;
7488	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR3, &u64Cr3); AssertRC(rc);
7489	if (pCtx->cr3 != u64Cr3)
7490	{
7491	pCtx->cr3 = u64Cr3;
7492	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
7493	}
7494
7495	/* If the guest is in PAE mode, sync back the PDPE's into the guest state.
7496	Note: CR4.PAE, CR0.PG, EFER MSR changes are always intercepted, so they're up to date. */
7497	if (CPUMIsGuestInPAEModeEx(pCtx))
7498	{
7499	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &pVCpu->hm.s.aPdpes[0].u); AssertRC(rc);
7500	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &pVCpu->hm.s.aPdpes[1].u); AssertRC(rc);
7501	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &pVCpu->hm.s.aPdpes[2].u); AssertRC(rc);
7502	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &pVCpu->hm.s.aPdpes[3].u); AssertRC(rc);
7503	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES);
7504	}
7505	}
7506	}
7507	}
7508
7509	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7510	if (fWhat & CPUMCTX_EXTRN_HWVIRT)
7511	{
7512	if ( (pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
7513	&& !CPUMIsGuestInVmxNonRootMode(pCtx))
7514	{
7515	Assert(CPUMIsGuestInVmxRootMode(pCtx));
7516	rc = hmR0VmxCopyShadowToNstGstVmcs(pVCpu, pVmcsInfo);
7517	if (RT_SUCCESS(rc))
7518	{ /* likely */ }
7519	else
7520	break;
7521	}
7522	}
7523	#endif
7524	} while (0);
7525
7526	if (RT_SUCCESS(rc))
7527	{
7528	/* Update fExtrn. */
7529	pCtx->fExtrn &= ~fWhat;
7530
7531	/* If everything has been imported, clear the HM keeper bit. */
7532	if (!(pCtx->fExtrn & HMVMX_CPUMCTX_EXTRN_ALL))
7533	{
7534	pCtx->fExtrn &= ~CPUMCTX_EXTRN_KEEPER_HM;
7535	Assert(!pCtx->fExtrn);
7536	}
7537	}
7538	}
7539	else
7540	AssertMsg(!pCtx->fExtrn \|\| (pCtx->fExtrn & HMVMX_CPUMCTX_EXTRN_ALL), ("%#RX64\n", pCtx->fExtrn));
7541
7542	/*
7543	* Restore interrupts.
7544	*/
7545	ASMSetFlags(fEFlags);
7546
7547	STAM_PROFILE_ADV_STOP(& pVCpu->hm.s.StatImportGuestState, x);
7548
7549	if (RT_SUCCESS(rc))
7550	{ /* likely */ }
7551	else
7552	return rc;
7553
7554	/*
7555	* Honor any pending CR3 updates.
7556	*
7557	* Consider this scenario: VM-exit -> VMMRZCallRing3Enable() -> do stuff that causes a longjmp -> hmR0VmxCallRing3Callback()
7558	* -> VMMRZCallRing3Disable() -> hmR0VmxImportGuestState() -> Sets VMCPU_FF_HM_UPDATE_CR3 pending -> return from the longjmp
7559	* -> continue with VM-exit handling -> hmR0VmxImportGuestState() and here we are.
7560	*
7561	* The reason for such complicated handling is because VM-exits that call into PGM expect CR3 to be up-to-date and thus
7562	* if any CR3-saves -before- the VM-exit (longjmp) postponed the CR3 update via the force-flag, any VM-exit handler that
7563	* calls into PGM when it re-saves CR3 will end up here and we call PGMUpdateCR3(). This is why the code below should
7564	* -NOT- check if CPUMCTX_EXTRN_CR3 is set!
7565	*
7566	* The longjmp exit path can't check these CR3 force-flags and call code that takes a lock again. We cover for it here.
7567	*/
7568	if (VMMRZCallRing3IsEnabled(pVCpu))
7569	{
7570	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
7571	{
7572	Assert(!(ASMAtomicUoReadU64(&pCtx->fExtrn) & CPUMCTX_EXTRN_CR3));
7573	PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
7574	}
7575
7576	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
7577	PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
7578
7579	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
7580	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
7581	}
7582
7583	return VINF_SUCCESS;
7584	}
7585
7586
7587	/**
7588	* Saves the guest state from the VMCS into the guest-CPU context.
7589	*
7590	* @returns VBox status code.
7591	* @param pVCpu The cross context virtual CPU structure.
7592	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
7593	*/
7594	VMMR0DECL(int) VMXR0ImportStateOnDemand(PVMCPUCC pVCpu, uint64_t fWhat)
7595	{
7596	AssertPtr(pVCpu);
7597	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
7598	return hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fWhat);
7599	}
7600
7601
7602	/**
7603	* Check per-VM and per-VCPU force flag actions that require us to go back to
7604	* ring-3 for one reason or another.
7605	*
7606	* @returns Strict VBox status code (i.e. informational status codes too)
7607	* @retval VINF_SUCCESS if we don't have any actions that require going back to
7608	* ring-3.
7609	* @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync.
7610	* @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware
7611	* interrupts)
7612	* @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires
7613	* all EMTs to be in ring-3.
7614	* @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests.
7615	* @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return
7616	* to the EM loop.
7617	*
7618	* @param pVCpu The cross context virtual CPU structure.
7619	* @param fStepping Whether we are single-stepping the guest using the
7620	* hypervisor debugger.
7621	*
7622	* @remarks This might cause nested-guest VM-exits, caller must check if the guest
7623	* is no longer in VMX non-root mode.
7624	*/
7625	static VBOXSTRICTRC hmR0VmxCheckForceFlags(PVMCPUCC pVCpu, bool fStepping)
7626	{
7627	Assert(VMMRZCallRing3IsEnabled(pVCpu));
7628
7629	/*
7630	* Update pending interrupts into the APIC's IRR.
7631	*/
7632	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UPDATE_APIC))
7633	APICUpdatePendingInterrupts(pVCpu);
7634
7635	/*
7636	* Anything pending? Should be more likely than not if we're doing a good job.
7637	*/
7638	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7639	if ( !fStepping
7640	? !VM_FF_IS_ANY_SET(pVM, VM_FF_HP_R0_PRE_HM_MASK)
7641	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HP_R0_PRE_HM_MASK)
7642	: !VM_FF_IS_ANY_SET(pVM, VM_FF_HP_R0_PRE_HM_STEP_MASK)
7643	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) )
7644	return VINF_SUCCESS;
7645
7646	/* Pending PGM C3 sync. */
7647	if (VMCPU_FF_IS_ANY_SET(pVCpu,VMCPU_FF_PGM_SYNC_CR3 \| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL))
7648	{
7649	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7650	Assert(!(ASMAtomicUoReadU64(&pCtx->fExtrn) & (CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_CR3 \| CPUMCTX_EXTRN_CR4)));
7651	VBOXSTRICTRC rcStrict = PGMSyncCR3(pVCpu, pCtx->cr0, pCtx->cr3, pCtx->cr4,
7652	VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
7653	if (rcStrict != VINF_SUCCESS)
7654	{
7655	AssertRC(VBOXSTRICTRC_VAL(rcStrict));
7656	Log4Func(("PGMSyncCR3 forcing us back to ring-3. rc2=%d\n", VBOXSTRICTRC_VAL(rcStrict)));
7657	return rcStrict;
7658	}
7659	}
7660
7661	/* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */
7662	if ( VM_FF_IS_ANY_SET(pVM, VM_FF_HM_TO_R3_MASK)
7663	\|\| VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
7664	{
7665	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
7666	int rc = RT_LIKELY(!VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_RAW_TO_R3 : VINF_EM_NO_MEMORY;
7667	Log4Func(("HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc));
7668	return rc;
7669	}
7670
7671	/* Pending VM request packets, such as hardware interrupts. */
7672	if ( VM_FF_IS_SET(pVM, VM_FF_REQUEST)
7673	\|\| VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_REQUEST))
7674	{
7675	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchVmReq);
7676	Log4Func(("Pending VM request forcing us back to ring-3\n"));
7677	return VINF_EM_PENDING_REQUEST;
7678	}
7679
7680	/* Pending PGM pool flushes. */
7681	if (VM_FF_IS_SET(pVM, VM_FF_PGM_POOL_FLUSH_PENDING))
7682	{
7683	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPgmPoolFlush);
7684	Log4Func(("PGM pool flush pending forcing us back to ring-3\n"));
7685	return VINF_PGM_POOL_FLUSH_PENDING;
7686	}
7687
7688	/* Pending DMA requests. */
7689	if (VM_FF_IS_SET(pVM, VM_FF_PDM_DMA))
7690	{
7691	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchDma);
7692	Log4Func(("Pending DMA request forcing us back to ring-3\n"));
7693	return VINF_EM_RAW_TO_R3;
7694	}
7695
7696	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7697	/* Pending nested-guest APIC-write (has highest priority among nested-guest FFs). */
7698	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_APIC_WRITE))
7699	{
7700	Log4Func(("Pending nested-guest APIC-write\n"));
7701	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitApicWrite(pVCpu);
7702	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
7703	return rcStrict;
7704	}
7705	/** @todo VMCPU_FF_VMX_MTF, VMCPU_FF_VMX_PREEMPT_TIMER */
7706	#endif
7707
7708	return VINF_SUCCESS;
7709	}
7710
7711
7712	/**
7713	* Converts any TRPM trap into a pending HM event. This is typically used when
7714	* entering from ring-3 (not longjmp returns).
7715	*
7716	* @param pVCpu The cross context virtual CPU structure.
7717	*/
7718	static void hmR0VmxTrpmTrapToPendingEvent(PVMCPUCC pVCpu)
7719	{
7720	Assert(TRPMHasTrap(pVCpu));
7721	Assert(!pVCpu->hm.s.Event.fPending);
7722
7723	uint8_t uVector;
7724	TRPMEVENT enmTrpmEvent;
7725	RTGCUINT uErrCode;
7726	RTGCUINTPTR GCPtrFaultAddress;
7727	uint8_t cbInstr;
7728
7729	int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr);
7730	AssertRC(rc);
7731
7732	uint32_t u32IntInfo;
7733	u32IntInfo = uVector \| VMX_IDT_VECTORING_INFO_VALID;
7734	u32IntInfo \|= HMTrpmEventTypeToVmxEventType(uVector, enmTrpmEvent);
7735
7736	rc = TRPMResetTrap(pVCpu);
7737	AssertRC(rc);
7738	Log4(("TRPM->HM event: u32IntInfo=%#RX32 enmTrpmEvent=%d cbInstr=%u uErrCode=%#RX32 GCPtrFaultAddress=%#RGv\n",
7739	u32IntInfo, enmTrpmEvent, cbInstr, uErrCode, GCPtrFaultAddress));
7740
7741	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, uErrCode, GCPtrFaultAddress);
7742	}
7743
7744
7745	/**
7746	* Converts the pending HM event into a TRPM trap.
7747	*
7748	* @param pVCpu The cross context virtual CPU structure.
7749	*/
7750	static void hmR0VmxPendingEventToTrpmTrap(PVMCPUCC pVCpu)
7751	{
7752	Assert(pVCpu->hm.s.Event.fPending);
7753
7754	/* If a trap was already pending, we did something wrong! */
7755	Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo /, NULL / pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP);
7756
7757	uint32_t const u32IntInfo = pVCpu->hm.s.Event.u64IntInfo;
7758	uint32_t const uVector = VMX_IDT_VECTORING_INFO_VECTOR(u32IntInfo);
7759	TRPMEVENT const enmTrapType = HMVmxEventTypeToTrpmEventType(u32IntInfo);
7760
7761	Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, enmTrapType));
7762
7763	int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType);
7764	AssertRC(rc);
7765
7766	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(u32IntInfo))
7767	TRPMSetErrorCode(pVCpu, pVCpu->hm.s.Event.u32ErrCode);
7768
7769	if (VMX_IDT_VECTORING_INFO_IS_XCPT_PF(u32IntInfo))
7770	TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress);
7771	else if (VMX_IDT_VECTORING_INFO_TYPE(u32IntInfo) == VMX_IDT_VECTORING_INFO_TYPE_SW_INT)
7772	TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr);
7773
7774	/* We're now done converting the pending event. */
7775	pVCpu->hm.s.Event.fPending = false;
7776	}
7777
7778
7779	/**
7780	* Sets the interrupt-window exiting control in the VMCS which instructs VT-x to
7781	* cause a VM-exit as soon as the guest is in a state to receive interrupts.
7782	*
7783	* @param pVCpu The cross context virtual CPU structure.
7784	* @param pVmcsInfo The VMCS info. object.
7785	*/
7786	static void hmR0VmxSetIntWindowExitVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
7787	{
7788	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_INT_WINDOW_EXIT)
7789	{
7790	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT))
7791	{
7792	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_INT_WINDOW_EXIT;
7793	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
7794	AssertRC(rc);
7795	}
7796	} /* else we will deliver interrupts whenever the guest Vm-exits next and is in a state to receive the interrupt. */
7797	}
7798
7799
7800	/**
7801	* Clears the interrupt-window exiting control in the VMCS.
7802	*
7803	* @param pVmcsInfo The VMCS info. object.
7804	*/
7805	DECLINLINE(void) hmR0VmxClearIntWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
7806	{
7807	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT)
7808	{
7809	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_INT_WINDOW_EXIT;
7810	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
7811	AssertRC(rc);
7812	}
7813	}
7814
7815
7816	/**
7817	* Sets the NMI-window exiting control in the VMCS which instructs VT-x to
7818	* cause a VM-exit as soon as the guest is in a state to receive NMIs.
7819	*
7820	* @param pVCpu The cross context virtual CPU structure.
7821	* @param pVmcsInfo The VMCS info. object.
7822	*/
7823	static void hmR0VmxSetNmiWindowExitVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
7824	{
7825	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
7826	{
7827	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))
7828	{
7829	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_NMI_WINDOW_EXIT;
7830	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
7831	AssertRC(rc);
7832	Log4Func(("Setup NMI-window exiting\n"));
7833	}
7834	} /* else we will deliver NMIs whenever we VM-exit next, even possibly nesting NMIs. Can't be helped on ancient CPUs. */
7835	}
7836
7837
7838	/**
7839	* Clears the NMI-window exiting control in the VMCS.
7840	*
7841	* @param pVmcsInfo The VMCS info. object.
7842	*/
7843	DECLINLINE(void) hmR0VmxClearNmiWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
7844	{
7845	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
7846	{
7847	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_NMI_WINDOW_EXIT;
7848	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
7849	AssertRC(rc);
7850	}
7851	}
7852
7853
7854	/**
7855	* Does the necessary state syncing before returning to ring-3 for any reason
7856	* (longjmp, preemption, voluntary exits to ring-3) from VT-x.
7857	*
7858	* @returns VBox status code.
7859	* @param pVCpu The cross context virtual CPU structure.
7860	* @param fImportState Whether to import the guest state from the VMCS back
7861	* to the guest-CPU context.
7862	*
7863	* @remarks No-long-jmp zone!!!
7864	*/
7865	static int hmR0VmxLeave(PVMCPUCC pVCpu, bool fImportState)
7866	{
7867	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
7868	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
7869
7870	RTCPUID const idCpu = RTMpCpuId();
7871	Log4Func(("HostCpuId=%u\n", idCpu));
7872
7873	/*
7874	* !!! IMPORTANT !!!
7875	* If you modify code here, check whether hmR0VmxCallRing3Callback() needs to be updated too.
7876	*/
7877
7878	/* Save the guest state if necessary. */
7879	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
7880	if (fImportState)
7881	{
7882	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
7883	AssertRCReturn(rc, rc);
7884	}
7885
7886	/* Restore host FPU state if necessary. We will resync on next R0 reentry. */
7887	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
7888	Assert(!CPUMIsGuestFPUStateActive(pVCpu));
7889
7890	/* Restore host debug registers if necessary. We will resync on next R0 reentry. */
7891	#ifdef VBOX_STRICT
7892	if (CPUMIsHyperDebugStateActive(pVCpu))
7893	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MOV_DR_EXIT);
7894	#endif
7895	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
7896	Assert(!CPUMIsGuestDebugStateActive(pVCpu) && !CPUMIsGuestDebugStateActivePending(pVCpu));
7897	Assert(!CPUMIsHyperDebugStateActive(pVCpu) && !CPUMIsHyperDebugStateActivePending(pVCpu));
7898
7899	/* Restore host-state bits that VT-x only restores partially. */
7900	if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
7901	&& (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
7902	{
7903	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags, idCpu));
7904	VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
7905	}
7906	pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
7907
7908	/* Restore the lazy host MSRs as we're leaving VT-x context. */
7909	if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
7910	{
7911	/* We shouldn't restore the host MSRs without saving the guest MSRs first. */
7912	if (!fImportState)
7913	{
7914	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_KERNEL_GS_BASE \| CPUMCTX_EXTRN_SYSCALL_MSRS);
7915	AssertRCReturn(rc, rc);
7916	}
7917	hmR0VmxLazyRestoreHostMsrs(pVCpu);
7918	Assert(!pVCpu->hm.s.vmx.fLazyMsrs);
7919	}
7920	else
7921	pVCpu->hm.s.vmx.fLazyMsrs = 0;
7922
7923	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
7924	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = false;
7925
7926	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry);
7927	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatImportGuestState);
7928	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExportGuestState);
7929	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatPreExit);
7930	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitHandling);
7931	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitIO);
7932	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitMovCRx);
7933	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitXcptNmi);
7934	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitVmentry);
7935	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
7936
7937	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
7938
7939	/** @todo This partially defeats the purpose of having preemption hooks.
7940	* The problem is, deregistering the hooks should be moved to a place that
7941	* lasts until the EMT is about to be destroyed not everytime while leaving HM
7942	* context.
7943	*/
7944	int rc = hmR0VmxClearVmcs(pVmcsInfo);
7945	AssertRCReturn(rc, rc);
7946
7947	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7948	/*
7949	* A valid shadow VMCS is made active as part of VM-entry. It is necessary to
7950	* clear a shadow VMCS before allowing that VMCS to become active on another
7951	* logical processor. We may or may not be importing guest state which clears
7952	* it, so cover for it here.
7953	*
7954	* See Intel spec. 24.11.1 "Software Use of Virtual-Machine Control Structures".
7955	*/
7956	if ( pVmcsInfo->pvShadowVmcs
7957	&& pVmcsInfo->fShadowVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
7958	{
7959	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
7960	AssertRCReturn(rc, rc);
7961	}
7962
7963	/*
7964	* Flag that we need to re-export the host state if we switch to this VMCS before
7965	* executing guest or nested-guest code.
7966	*/
7967	pVmcsInfo->idHostCpuState = NIL_RTCPUID;
7968	#endif
7969
7970	Log4Func(("Cleared Vmcs. HostCpuId=%u\n", idCpu));
7971	NOREF(idCpu);
7972	return VINF_SUCCESS;
7973	}
7974
7975
7976	/**
7977	* Leaves the VT-x session.
7978	*
7979	* @returns VBox status code.
7980	* @param pVCpu The cross context virtual CPU structure.
7981	*
7982	* @remarks No-long-jmp zone!!!
7983	*/
7984	static int hmR0VmxLeaveSession(PVMCPUCC pVCpu)
7985	{
7986	HM_DISABLE_PREEMPT(pVCpu);
7987	HMVMX_ASSERT_CPU_SAFE(pVCpu);
7988	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
7989	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
7990
7991	/* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before
7992	and done this from the VMXR0ThreadCtxCallback(). */
7993	if (!pVCpu->hm.s.fLeaveDone)
7994	{
7995	int rc2 = hmR0VmxLeave(pVCpu, true /* fImportState */);
7996	AssertRCReturnStmt(rc2, HM_RESTORE_PREEMPT(), rc2);
7997	pVCpu->hm.s.fLeaveDone = true;
7998	}
7999	Assert(!pVCpu->cpum.GstCtx.fExtrn);
8000
8001	/*
8002	* !!! IMPORTANT !!!
8003	* If you modify code here, make sure to check whether hmR0VmxCallRing3Callback() needs to be updated too.
8004	*/
8005
8006	/* Deregister hook now that we've left HM context before re-enabling preemption. */
8007	/** @todo Deregistering here means we need to VMCLEAR always
8008	* (longjmp/exit-to-r3) in VT-x which is not efficient, eliminate need
8009	* for calling VMMR0ThreadCtxHookDisable here! */
8010	VMMR0ThreadCtxHookDisable(pVCpu);
8011
8012	/* Leave HM context. This takes care of local init (term). */
8013	int rc = HMR0LeaveCpu(pVCpu);
8014
8015	HM_RESTORE_PREEMPT();
8016	return rc;
8017	}
8018
8019
8020	/**
8021	* Does the necessary state syncing before doing a longjmp to ring-3.
8022	*
8023	* @returns VBox status code.
8024	* @param pVCpu The cross context virtual CPU structure.
8025	*
8026	* @remarks No-long-jmp zone!!!
8027	*/
8028	DECLINLINE(int) hmR0VmxLongJmpToRing3(PVMCPUCC pVCpu)
8029	{
8030	return hmR0VmxLeaveSession(pVCpu);
8031	}
8032
8033
8034	/**
8035	* Take necessary actions before going back to ring-3.
8036	*
8037	* An action requires us to go back to ring-3. This function does the necessary
8038	* steps before we can safely return to ring-3. This is not the same as longjmps
8039	* to ring-3, this is voluntary and prepares the guest so it may continue
8040	* executing outside HM (recompiler/IEM).
8041	*
8042	* @returns VBox status code.
8043	* @param pVCpu The cross context virtual CPU structure.
8044	* @param rcExit The reason for exiting to ring-3. Can be
8045	* VINF_VMM_UNKNOWN_RING3_CALL.
8046	*/
8047	static int hmR0VmxExitToRing3(PVMCPUCC pVCpu, VBOXSTRICTRC rcExit)
8048	{
8049	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
8050
8051	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8052	if (RT_UNLIKELY(rcExit == VERR_VMX_INVALID_VMCS_PTR))
8053	{
8054	VMXGetCurrentVmcs(&pVCpu->hm.s.vmx.LastError.HCPhysCurrentVmcs);
8055	pVCpu->hm.s.vmx.LastError.u32VmcsRev = (uint32_t )pVmcsInfo->pvVmcs;
8056	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
8057	/* LastError.idCurrentCpu was updated in hmR0VmxPreRunGuestCommitted(). */
8058	}
8059
8060	/* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */
8061	VMMRZCallRing3Disable(pVCpu);
8062	Log4Func(("rcExit=%d\n", VBOXSTRICTRC_VAL(rcExit)));
8063
8064	/*
8065	* Convert any pending HM events back to TRPM due to premature exits to ring-3.
8066	* We need to do this only on returns to ring-3 and not for longjmps to ring3.
8067	*
8068	* This is because execution may continue from ring-3 and we would need to inject
8069	* the event from there (hence place it back in TRPM).
8070	*/
8071	if (pVCpu->hm.s.Event.fPending)
8072	{
8073	hmR0VmxPendingEventToTrpmTrap(pVCpu);
8074	Assert(!pVCpu->hm.s.Event.fPending);
8075
8076	/* Clear the events from the VMCS. */
8077	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, 0);
8078	AssertRC(rc);
8079	}
8080	#ifdef VBOX_STRICT
8081	else
8082	{
8083	/*
8084	* Ensure we don't accidentally clear a pending HM event without clearing the VMCS.
8085	* This can be pretty hard to debug otherwise, interrupts might get injected twice
8086	* occasionally, see @bugref{9180#c42}.
8087	*
8088	* However, if the VM-entry failed, any VM entry-interruption info. field would
8089	* be left unmodified as the event would not have been injected to the guest. In
8090	* such cases, don't assert, we're not going to continue guest execution anyway.
8091	*/
8092	uint32_t uExitReason;
8093	uint32_t uEntryIntInfo;
8094	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
8095	rc \|= VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &uEntryIntInfo);
8096	AssertRC(rc);
8097	Assert(VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason) \|\| !VMX_ENTRY_INT_INFO_IS_VALID(uEntryIntInfo));
8098	}
8099	#endif
8100
8101	/*
8102	* Clear the interrupt-window and NMI-window VMCS controls as we could have got
8103	* a VM-exit with higher priority than interrupt-window or NMI-window VM-exits
8104	* (e.g. TPR below threshold).
8105	*/
8106	if (!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
8107	{
8108	hmR0VmxClearIntWindowExitVmcs(pVmcsInfo);
8109	hmR0VmxClearNmiWindowExitVmcs(pVmcsInfo);
8110	}
8111
8112	/* If we're emulating an instruction, we shouldn't have any TRPM traps pending
8113	and if we're injecting an event we should have a TRPM trap pending. */
8114	AssertMsg(rcExit != VINF_EM_RAW_INJECT_TRPM_EVENT \|\| TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
8115	#ifndef DEBUG_bird /* Triggered after firing an NMI against NT4SP1, possibly a triple fault in progress. */
8116	AssertMsg(rcExit != VINF_EM_RAW_EMULATE_INSTR \|\| !TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
8117	#endif
8118
8119	/* Save guest state and restore host state bits. */
8120	int rc = hmR0VmxLeaveSession(pVCpu);
8121	AssertRCReturn(rc, rc);
8122	STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
8123
8124	/* Thread-context hooks are unregistered at this point!!! */
8125
8126	/* Sync recompiler state. */
8127	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3);
8128	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR
8129	\| CPUM_CHANGED_LDTR
8130	\| CPUM_CHANGED_GDTR
8131	\| CPUM_CHANGED_IDTR
8132	\| CPUM_CHANGED_TR
8133	\| CPUM_CHANGED_HIDDEN_SEL_REGS);
8134	if ( pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging
8135	&& CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx))
8136	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH);
8137
8138	Assert(!pVCpu->hm.s.fClearTrapFlag);
8139
8140	/* Update the exit-to-ring 3 reason. */
8141	pVCpu->hm.s.rcLastExitToR3 = VBOXSTRICTRC_VAL(rcExit);
8142
8143	/* On our way back from ring-3 reload the guest state if there is a possibility of it being changed. */
8144	if ( rcExit != VINF_EM_RAW_INTERRUPT
8145	\|\| CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
8146	{
8147	Assert(!(pVCpu->cpum.GstCtx.fExtrn & HMVMX_CPUMCTX_EXTRN_ALL));
8148	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
8149	}
8150
8151	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3);
8152
8153	/* We do -not- want any longjmp notifications after this! We must return to ring-3 ASAP. */
8154	VMMRZCallRing3RemoveNotification(pVCpu);
8155	VMMRZCallRing3Enable(pVCpu);
8156
8157	return rc;
8158	}
8159
8160
8161	/**
8162	* VMMRZCallRing3() callback wrapper which saves the guest state before we
8163	* longjump to ring-3 and possibly get preempted.
8164	*
8165	* @returns VBox status code.
8166	* @param pVCpu The cross context virtual CPU structure.
8167	* @param enmOperation The operation causing the ring-3 longjump.
8168	* @param pvUser User argument, currently unused, NULL.
8169	*/
8170	static DECLCALLBACK(int) hmR0VmxCallRing3Callback(PVMCPUCC pVCpu, VMMCALLRING3 enmOperation, void *pvUser)
8171	{
8172	RT_NOREF(pvUser);
8173	if (enmOperation == VMMCALLRING3_VM_R0_ASSERTION)
8174	{
8175	/*
8176	* !!! IMPORTANT !!!
8177	* If you modify code here, check whether hmR0VmxLeave() and hmR0VmxLeaveSession() needs to be updated too.
8178	* This is a stripped down version which gets out ASAP, trying to not trigger any further assertions.
8179	*/
8180	VMMRZCallRing3RemoveNotification(pVCpu);
8181	VMMRZCallRing3Disable(pVCpu);
8182	RTTHREADPREEMPTSTATE PreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
8183	RTThreadPreemptDisable(&PreemptState);
8184
8185	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8186	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
8187	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
8188	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
8189
8190	/* Restore host-state bits that VT-x only restores partially. */
8191	if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
8192	&& (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
8193	VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
8194	pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
8195
8196	/* Restore the lazy host MSRs as we're leaving VT-x context. */
8197	if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
8198	hmR0VmxLazyRestoreHostMsrs(pVCpu);
8199
8200	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
8201	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = false;
8202	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
8203
8204	/* Clear the current VMCS data back to memory (shadow VMCS if any would have been
8205	cleared as part of importing the guest state above. */
8206	hmR0VmxClearVmcs(pVmcsInfo);
8207
8208	/** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
8209	VMMR0ThreadCtxHookDisable(pVCpu);
8210	HMR0LeaveCpu(pVCpu);
8211	RTThreadPreemptRestore(&PreemptState);
8212	return VINF_SUCCESS;
8213	}
8214
8215	Assert(pVCpu);
8216	Assert(pvUser);
8217	Assert(VMMRZCallRing3IsEnabled(pVCpu));
8218	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
8219
8220	VMMRZCallRing3Disable(pVCpu);
8221	Assert(VMMR0IsLogFlushDisabled(pVCpu));
8222
8223	Log4Func(("-> hmR0VmxLongJmpToRing3 enmOperation=%d\n", enmOperation));
8224
8225	int rc = hmR0VmxLongJmpToRing3(pVCpu);
8226	AssertRCReturn(rc, rc);
8227
8228	VMMRZCallRing3Enable(pVCpu);
8229	return VINF_SUCCESS;
8230	}
8231
8232
8233	/**
8234	* Pushes a 2-byte value onto the real-mode (in virtual-8086 mode) guest's
8235	* stack.
8236	*
8237	* @returns Strict VBox status code (i.e. informational status codes too).
8238	* @retval VINF_EM_RESET if pushing a value to the stack caused a triple-fault.
8239	* @param pVCpu The cross context virtual CPU structure.
8240	* @param uValue The value to push to the guest stack.
8241	*/
8242	static VBOXSTRICTRC hmR0VmxRealModeGuestStackPush(PVMCPUCC pVCpu, uint16_t uValue)
8243	{
8244	/*
8245	* The stack limit is 0xffff in real-on-virtual 8086 mode. Real-mode with weird stack limits cannot be run in
8246	* virtual 8086 mode in VT-x. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
8247	* See Intel Instruction reference for PUSH and Intel spec. 22.33.1 "Segment Wraparound".
8248	*/
8249	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8250	if (pCtx->sp == 1)
8251	return VINF_EM_RESET;
8252	pCtx->sp -= sizeof(uint16_t); /* May wrap around which is expected behaviour. */
8253	int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), pCtx->ss.u64Base + pCtx->sp, &uValue, sizeof(uint16_t));
8254	AssertRC(rc);
8255	return rc;
8256	}
8257
8258
8259	/**
8260	* Injects an event into the guest upon VM-entry by updating the relevant fields
8261	* in the VM-entry area in the VMCS.
8262	*
8263	* @returns Strict VBox status code (i.e. informational status codes too).
8264	* @retval VINF_SUCCESS if the event is successfully injected into the VMCS.
8265	* @retval VINF_EM_RESET if event injection resulted in a triple-fault.
8266	*
8267	* @param pVCpu The cross context virtual CPU structure.
8268	* @param pVmxTransient The VMX-transient structure.
8269	* @param pEvent The event being injected.
8270	* @param pfIntrState Pointer to the VT-x guest-interruptibility-state. This
8271	* will be updated if necessary. This cannot not be NULL.
8272	* @param fStepping Whether we're single-stepping guest execution and should
8273	* return VINF_EM_DBG_STEPPED if the event is injected
8274	* directly (registers modified by us, not by hardware on
8275	* VM-entry).
8276	*/
8277	static VBOXSTRICTRC hmR0VmxInjectEventVmcs(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PCHMEVENT pEvent, bool fStepping,
8278	uint32_t *pfIntrState)
8279	{
8280	/* Intel spec. 24.8.3 "VM-Entry Controls for Event Injection" specifies the interruption-information field to be 32-bits. */
8281	AssertMsg(!RT_HI_U32(pEvent->u64IntInfo), ("%#RX64\n", pEvent->u64IntInfo));
8282	Assert(pfIntrState);
8283
8284	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8285	uint32_t u32IntInfo = pEvent->u64IntInfo;
8286	uint32_t const u32ErrCode = pEvent->u32ErrCode;
8287	uint32_t const cbInstr = pEvent->cbInstr;
8288	RTGCUINTPTR const GCPtrFault = pEvent->GCPtrFaultAddress;
8289	uint8_t const uVector = VMX_ENTRY_INT_INFO_VECTOR(u32IntInfo);
8290	uint32_t const uIntType = VMX_ENTRY_INT_INFO_TYPE(u32IntInfo);
8291
8292	#ifdef VBOX_STRICT
8293	/*
8294	* Validate the error-code-valid bit for hardware exceptions.
8295	* No error codes for exceptions in real-mode.
8296	*
8297	* See Intel spec. 20.1.4 "Interrupt and Exception Handling"
8298	*/
8299	if ( uIntType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT
8300	&& !CPUMIsGuestInRealModeEx(pCtx))
8301	{
8302	switch (uVector)
8303	{
8304	case X86_XCPT_PF:
8305	case X86_XCPT_DF:
8306	case X86_XCPT_TS:
8307	case X86_XCPT_NP:
8308	case X86_XCPT_SS:
8309	case X86_XCPT_GP:
8310	case X86_XCPT_AC:
8311	AssertMsg(VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(u32IntInfo),
8312	("Error-code-valid bit not set for exception that has an error code uVector=%#x\n", uVector));
8313	RT_FALL_THRU();
8314	default:
8315	break;
8316	}
8317	}
8318
8319	/* Cannot inject an NMI when block-by-MOV SS is in effect. */
8320	Assert( uIntType != VMX_EXIT_INT_INFO_TYPE_NMI
8321	\|\| !(*pfIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
8322	#endif
8323
8324	STAM_COUNTER_INC(&pVCpu->hm.s.paStatInjectedIrqsR0[uVector & MASK_INJECT_IRQ_STAT]);
8325
8326	/*
8327	* Hardware interrupts & exceptions cannot be delivered through the software interrupt
8328	* redirection bitmap to the real mode task in virtual-8086 mode. We must jump to the
8329	* interrupt handler in the (real-mode) guest.
8330	*
8331	* See Intel spec. 20.3 "Interrupt and Exception handling in Virtual-8086 Mode".
8332	* See Intel spec. 20.1.4 "Interrupt and Exception Handling" for real-mode interrupt handling.
8333	*/
8334	if (CPUMIsGuestInRealModeEx(pCtx)) /* CR0.PE bit changes are always intercepted, so it's up to date. */
8335	{
8336	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest)
8337	{
8338	/*
8339	* For CPUs with unrestricted guest execution enabled and with the guest
8340	* in real-mode, we must not set the deliver-error-code bit.
8341	*
8342	* See Intel spec. 26.2.1.3 "VM-Entry Control Fields".
8343	*/
8344	u32IntInfo &= ~VMX_ENTRY_INT_INFO_ERROR_CODE_VALID;
8345	}
8346	else
8347	{
8348	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
8349	Assert(PDMVmmDevHeapIsEnabled(pVM));
8350	Assert(pVM->hm.s.vmx.pRealModeTSS);
8351	Assert(!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
8352
8353	/* We require RIP, RSP, RFLAGS, CS, IDTR, import them. */
8354	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
8355	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_TABLE_MASK
8356	\| CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_RFLAGS);
8357	AssertRCReturn(rc2, rc2);
8358
8359	/* Check if the interrupt handler is present in the IVT (real-mode IDT). IDT limit is (4N - 1). */
8360	size_t const cbIdtEntry = sizeof(X86IDTR16);
8361	if (uVector * cbIdtEntry + (cbIdtEntry - 1) > pCtx->idtr.cbIdt)
8362	{
8363	/* If we are trying to inject a #DF with no valid IDT entry, return a triple-fault. */
8364	if (uVector == X86_XCPT_DF)
8365	return VINF_EM_RESET;
8366
8367	/* If we're injecting a #GP with no valid IDT entry, inject a double-fault.
8368	No error codes for exceptions in real-mode. */
8369	if (uVector == X86_XCPT_GP)
8370	{
8371	uint32_t const uXcptDfInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DF)
8372	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
8373	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
8374	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
8375	HMEVENT EventXcptDf;
8376	RT_ZERO(EventXcptDf);
8377	EventXcptDf.u64IntInfo = uXcptDfInfo;
8378	return hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &EventXcptDf, fStepping, pfIntrState);
8379	}
8380
8381	/*
8382	* If we're injecting an event with no valid IDT entry, inject a #GP.
8383	* No error codes for exceptions in real-mode.
8384	*
8385	* See Intel spec. 20.1.4 "Interrupt and Exception Handling"
8386	*/
8387	uint32_t const uXcptGpInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_GP)
8388	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
8389	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
8390	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
8391	HMEVENT EventXcptGp;
8392	RT_ZERO(EventXcptGp);
8393	EventXcptGp.u64IntInfo = uXcptGpInfo;
8394	return hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &EventXcptGp, fStepping, pfIntrState);
8395	}
8396
8397	/* Software exceptions (#BP and #OF exceptions thrown as a result of INT3 or INTO) */
8398	uint16_t uGuestIp = pCtx->ip;
8399	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_SW_XCPT)
8400	{
8401	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF);
8402	/* #BP and #OF are both benign traps, we need to resume the next instruction. */
8403	uGuestIp = pCtx->ip + (uint16_t)cbInstr;
8404	}
8405	else if (uIntType == VMX_ENTRY_INT_INFO_TYPE_SW_INT)
8406	uGuestIp = pCtx->ip + (uint16_t)cbInstr;
8407
8408	/* Get the code segment selector and offset from the IDT entry for the interrupt handler. */
8409	X86IDTR16 IdtEntry;
8410	RTGCPHYS const GCPhysIdtEntry = (RTGCPHYS)pCtx->idtr.pIdt + uVector * cbIdtEntry;
8411	rc2 = PGMPhysSimpleReadGCPhys(pVM, &IdtEntry, GCPhysIdtEntry, cbIdtEntry);
8412	AssertRCReturn(rc2, rc2);
8413
8414	/* Construct the stack frame for the interrupt/exception handler. */
8415	VBOXSTRICTRC rcStrict;
8416	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, pCtx->eflags.u32);
8417	if (rcStrict == VINF_SUCCESS)
8418	{
8419	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, pCtx->cs.Sel);
8420	if (rcStrict == VINF_SUCCESS)
8421	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, uGuestIp);
8422	}
8423
8424	/* Clear the required eflag bits and jump to the interrupt/exception handler. */
8425	if (rcStrict == VINF_SUCCESS)
8426	{
8427	pCtx->eflags.u32 &= ~(X86_EFL_IF \| X86_EFL_TF \| X86_EFL_RF \| X86_EFL_AC);
8428	pCtx->rip = IdtEntry.offSel;
8429	pCtx->cs.Sel = IdtEntry.uSel;
8430	pCtx->cs.ValidSel = IdtEntry.uSel;
8431	pCtx->cs.u64Base = IdtEntry.uSel << cbIdtEntry;
8432	if ( uIntType == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT
8433	&& uVector == X86_XCPT_PF)
8434	pCtx->cr2 = GCPtrFault;
8435
8436	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CS \| HM_CHANGED_GUEST_CR2
8437	\| HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
8438	\| HM_CHANGED_GUEST_RSP);
8439
8440	/*
8441	* If we delivered a hardware exception (other than an NMI) and if there was
8442	* block-by-STI in effect, we should clear it.
8443	*/
8444	if (*pfIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
8445	{
8446	Assert( uIntType != VMX_ENTRY_INT_INFO_TYPE_NMI
8447	&& uIntType != VMX_ENTRY_INT_INFO_TYPE_EXT_INT);
8448	Log4Func(("Clearing inhibition due to STI\n"));
8449	*pfIntrState &= ~VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
8450	}
8451
8452	Log4(("Injected real-mode: u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x Eflags=%#x CS:EIP=%04x:%04x\n",
8453	u32IntInfo, u32ErrCode, cbInstr, pCtx->eflags.u, pCtx->cs.Sel, pCtx->eip));
8454
8455	/*
8456	* The event has been truly dispatched to the guest. Mark it as no longer pending so
8457	* we don't attempt to undo it if we are returning to ring-3 before executing guest code.
8458	*/
8459	pVCpu->hm.s.Event.fPending = false;
8460
8461	/*
8462	* If we eventually support nested-guest execution without unrestricted guest execution,
8463	* we should set fInterceptEvents here.
8464	*/
8465	Assert(!pVmxTransient->fIsNestedGuest);
8466
8467	/* If we're stepping and we've changed cs:rip above, bail out of the VMX R0 execution loop. */
8468	if (fStepping)
8469	rcStrict = VINF_EM_DBG_STEPPED;
8470	}
8471	AssertMsg(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
8472	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
8473	return rcStrict;
8474	}
8475	}
8476
8477	/*
8478	* Validate.
8479	*/
8480	Assert(VMX_ENTRY_INT_INFO_IS_VALID(u32IntInfo)); /* Bit 31 (Valid bit) must be set by caller. */
8481	Assert(!(u32IntInfo & VMX_BF_ENTRY_INT_INFO_RSVD_12_30_MASK)); /* Bits 30:12 MBZ. */
8482
8483	/*
8484	* Inject the event into the VMCS.
8485	*/
8486	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, u32IntInfo);
8487	if (VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(u32IntInfo))
8488	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, u32ErrCode);
8489	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, cbInstr);
8490	AssertRC(rc);
8491
8492	/*
8493	* Update guest CR2 if this is a page-fault.
8494	*/
8495	if (VMX_ENTRY_INT_INFO_IS_XCPT_PF(u32IntInfo))
8496	pCtx->cr2 = GCPtrFault;
8497
8498	Log4(("Injecting u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x CR2=%#RX64\n", u32IntInfo, u32ErrCode, cbInstr, pCtx->cr2));
8499	return VINF_SUCCESS;
8500	}
8501
8502
8503	/**
8504	* Evaluates the event to be delivered to the guest and sets it as the pending
8505	* event.
8506	*
8507	* @returns Strict VBox status code (i.e. informational status codes too).
8508	* @param pVCpu The cross context virtual CPU structure.
8509	* @param pVmxTransient The VMX-transient structure.
8510	* @param pfIntrState Where to store the VT-x guest-interruptibility state.
8511	*/
8512	static VBOXSTRICTRC hmR0VmxEvaluatePendingEvent(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t *pfIntrState)
8513	{
8514	Assert(pfIntrState);
8515	Assert(!TRPMHasTrap(pVCpu));
8516
8517	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8518	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
8519	bool const fIsNestedGuest = pVmxTransient->fIsNestedGuest;
8520
8521	/*
8522	* Get the current interruptibility-state of the guest or nested-guest and
8523	* then figure out what needs to be injected.
8524	*/
8525	uint32_t const fIntrState = hmR0VmxGetGuestIntrState(pVCpu, pVmxTransient);
8526	bool const fBlockMovSS = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
8527	bool const fBlockSti = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI);
8528	bool const fBlockNmi = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI);
8529
8530	/* We don't support block-by-SMI yet.*/
8531	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI));
8532
8533	/* Block-by-STI must not be set when interrupts are disabled. */
8534	if (fBlockSti)
8535	{
8536	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
8537	Assert(pCtx->eflags.Bits.u1IF);
8538	}
8539
8540	/* Update interruptibility state to the caller. */
8541	*pfIntrState = fIntrState;
8542
8543	/*
8544	* Toggling of interrupt force-flags here is safe since we update TRPM on
8545	* premature exits to ring-3 before executing guest code, see hmR0VmxExitToRing3().
8546	* We must NOT restore these force-flags.
8547	*/
8548
8549	/** @todo SMI. SMIs take priority over NMIs. */
8550
8551	/*
8552	* Check if an NMI is pending and if the guest or nested-guest can receive them.
8553	* NMIs take priority over external interrupts.
8554	*/
8555	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
8556	{
8557	/* On some CPUs block-by-STI also blocks NMIs. See Intel spec. 26.3.1.5 "Checks On Guest Non-Register State". */
8558	if ( !pVCpu->hm.s.Event.fPending
8559	&& !fBlockNmi
8560	&& !fBlockSti
8561	&& !fBlockMovSS)
8562	{
8563	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8564	if ( fIsNestedGuest
8565	&& CPUMIsGuestVmxPinCtlsSet(pVCpu, pCtx, VMX_PIN_CTLS_NMI_EXIT))
8566	return IEMExecVmxVmexitXcptNmi(pVCpu);
8567	#endif
8568	hmR0VmxSetPendingXcptNmi(pVCpu);
8569	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
8570	Log4Func(("Pending NMI\n"));
8571	}
8572	else if (!fIsNestedGuest)
8573	hmR0VmxSetNmiWindowExitVmcs(pVCpu, pVmcsInfo);
8574	/* else: for nested-guests, NMI-window exiting will be picked up when merging VMCS controls. */
8575	}
8576	/*
8577	* Check if an external interrupt (PIC/APIC) is pending and if the guest or nested-guest
8578	* can receive them. Once PDMGetInterrupt() returns a valid interrupt we -must- deliver
8579	* the interrupt. We can no longer re-request it from the APIC.
8580	*/
8581	else if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
8582	&& !pVCpu->hm.s.fSingleInstruction)
8583	{
8584	Assert(!DBGFIsStepping(pVCpu));
8585	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_RFLAGS);
8586	AssertRCReturn(rc, rc);
8587
8588	bool const fBlockInt = !(pCtx->eflags.u32 & X86_EFL_IF);
8589	if ( !pVCpu->hm.s.Event.fPending
8590	&& !fBlockInt
8591	&& !fBlockSti
8592	&& !fBlockMovSS)
8593	{
8594	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8595	if ( fIsNestedGuest
8596	&& CPUMIsGuestVmxPinCtlsSet(pVCpu, pCtx, VMX_PIN_CTLS_EXT_INT_EXIT)
8597	&& !CPUMIsGuestVmxExitCtlsSet(pVCpu, pCtx, VMX_EXIT_CTLS_ACK_EXT_INT))
8598	{
8599	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitExtInt(pVCpu, 0 /* uVector /, true / fIntPending */);
8600	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8601	return rcStrict;
8602	}
8603	#endif
8604	uint8_t u8Interrupt;
8605	rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
8606	if (RT_SUCCESS(rc))
8607	{
8608	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8609	if ( fIsNestedGuest
8610	&& CPUMIsGuestVmxPinCtlsSet(pVCpu, pCtx, VMX_PIN_CTLS_EXT_INT_EXIT)
8611	&& CPUMIsGuestVmxExitCtlsSet(pVCpu, pCtx, VMX_EXIT_CTLS_ACK_EXT_INT))
8612	{
8613	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitExtInt(pVCpu, u8Interrupt, false /* fIntPending */);
8614	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8615	return rcStrict;
8616	}
8617	#endif
8618	hmR0VmxSetPendingExtInt(pVCpu, u8Interrupt);
8619	Log4Func(("Pending external interrupt vector %#x\n", u8Interrupt));
8620	}
8621	else if (rc == VERR_APIC_INTR_MASKED_BY_TPR)
8622	{
8623	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchTprMaskedIrq);
8624
8625	if ( !fIsNestedGuest
8626	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW))
8627	hmR0VmxApicSetTprThreshold(pVmcsInfo, u8Interrupt >> 4);
8628	/* else: for nested-guests, TPR threshold is picked up while merging VMCS controls. */
8629
8630	/*
8631	* If the CPU doesn't have TPR shadowing, we will always get a VM-exit on TPR changes and
8632	* APICSetTpr() will end up setting the VMCPU_FF_INTERRUPT_APIC if required, so there is no
8633	* need to re-set this force-flag here.
8634	*/
8635	}
8636	else
8637	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
8638	}
8639	else if (!fIsNestedGuest)
8640	hmR0VmxSetIntWindowExitVmcs(pVCpu, pVmcsInfo);
8641	/* else: for nested-guests, interrupt-window exiting will be picked up when merging VMCS controls. */
8642	}
8643
8644	return VINF_SUCCESS;
8645	}
8646
8647
8648	/**
8649	* Injects any pending events into the guest if the guest is in a state to
8650	* receive them.
8651	*
8652	* @returns Strict VBox status code (i.e. informational status codes too).
8653	* @param pVCpu The cross context virtual CPU structure.
8654	* @param pVmxTransient The VMX-transient structure.
8655	* @param fIntrState The VT-x guest-interruptibility state.
8656	* @param fStepping Whether we are single-stepping the guest using the
8657	* hypervisor debugger and should return
8658	* VINF_EM_DBG_STEPPED if the event was dispatched
8659	* directly.
8660	*/
8661	static VBOXSTRICTRC hmR0VmxInjectPendingEvent(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t fIntrState, bool fStepping)
8662	{
8663	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
8664	Assert(VMMRZCallRing3IsEnabled(pVCpu));
8665
8666	bool const fBlockMovSS = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
8667	bool const fBlockSti = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI);
8668
8669	Assert(!fBlockSti \|\| !(ASMAtomicUoReadU64(&pVCpu->cpum.GstCtx.fExtrn) & CPUMCTX_EXTRN_RFLAGS));
8670	Assert(!fBlockSti \|\| pVCpu->cpum.GstCtx.eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
8671	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
8672	Assert(!TRPMHasTrap(pVCpu));
8673
8674	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
8675	if (pVCpu->hm.s.Event.fPending)
8676	{
8677	/*
8678	* Do -not- clear any interrupt-window exiting control here. We might have an interrupt
8679	* pending even while injecting an event and in this case, we want a VM-exit as soon as
8680	* the guest is ready for the next interrupt, see @bugref{6208#c45}.
8681	*
8682	* See Intel spec. 26.6.5 "Interrupt-Window Exiting and Virtual-Interrupt Delivery".
8683	*/
8684	uint32_t const uIntType = VMX_ENTRY_INT_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
8685	#ifdef VBOX_STRICT
8686	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
8687	{
8688	bool const fBlockInt = !(pVCpu->cpum.GstCtx.eflags.u32 & X86_EFL_IF);
8689	Assert(!fBlockInt);
8690	Assert(!fBlockSti);
8691	Assert(!fBlockMovSS);
8692	}
8693	else if (uIntType == VMX_ENTRY_INT_INFO_TYPE_NMI)
8694	{
8695	bool const fBlockNmi = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI);
8696	Assert(!fBlockSti);
8697	Assert(!fBlockMovSS);
8698	Assert(!fBlockNmi);
8699	}
8700	#endif
8701	Log4(("Injecting pending event vcpu[%RU32] u64IntInfo=%#RX64 Type=%#RX32\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo,
8702	uIntType));
8703
8704	/*
8705	* Inject the event and get any changes to the guest-interruptibility state.
8706	*
8707	* The guest-interruptibility state may need to be updated if we inject the event
8708	* into the guest IDT ourselves (for real-on-v86 guest injecting software interrupts).
8709	*/
8710	rcStrict = hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &pVCpu->hm.s.Event, fStepping, &fIntrState);
8711	AssertRCReturn(VBOXSTRICTRC_VAL(rcStrict), rcStrict);
8712
8713	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
8714	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt);
8715	else
8716	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt);
8717	}
8718
8719	/*
8720	* Update the guest-interruptibility state.
8721	*
8722	* This is required for the real-on-v86 software interrupt injection case above, as well as
8723	* updates to the guest state from ring-3 or IEM/REM.
8724	*/
8725	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
8726	AssertRC(rc);
8727
8728	/*
8729	* There's no need to clear the VM-entry interruption-information field here if we're not
8730	* injecting anything. VT-x clears the valid bit on every VM-exit.
8731	*
8732	* See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
8733	*/
8734
8735	Assert(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping));
8736	NOREF(fBlockMovSS); NOREF(fBlockSti);
8737	return rcStrict;
8738	}
8739
8740
8741	/**
8742	* Enters the VT-x session.
8743	*
8744	* @returns VBox status code.
8745	* @param pVCpu The cross context virtual CPU structure.
8746	*/
8747	VMMR0DECL(int) VMXR0Enter(PVMCPUCC pVCpu)
8748	{
8749	AssertPtr(pVCpu);
8750	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fSupported);
8751	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8752
8753	LogFlowFunc(("pVCpu=%p\n", pVCpu));
8754	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
8755	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
8756
8757	#ifdef VBOX_STRICT
8758	/* At least verify VMX is enabled, since we can't check if we're in VMX root mode without #GP'ing. */
8759	RTCCUINTREG uHostCr4 = ASMGetCR4();
8760	if (!(uHostCr4 & X86_CR4_VMXE))
8761	{
8762	LogRelFunc(("X86_CR4_VMXE bit in CR4 is not set!\n"));
8763	return VERR_VMX_X86_CR4_VMXE_CLEARED;
8764	}
8765	#endif
8766
8767	/*
8768	* Load the appropriate VMCS as the current and active one.
8769	*/
8770	PVMXVMCSINFO pVmcsInfo;
8771	bool const fInNestedGuestMode = CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx);
8772	if (!fInNestedGuestMode)
8773	pVmcsInfo = &pVCpu->hm.s.vmx.VmcsInfo;
8774	else
8775	pVmcsInfo = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
8776	int rc = hmR0VmxLoadVmcs(pVmcsInfo);
8777	if (RT_SUCCESS(rc))
8778	{
8779	pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs = fInNestedGuestMode;
8780	pVCpu->hm.s.fLeaveDone = false;
8781	Log4Func(("Loaded Vmcs. HostCpuId=%u\n", RTMpCpuId()));
8782
8783	/*
8784	* Do the EMT scheduled L1D flush here if needed.
8785	*/
8786	if (pVCpu->CTX_SUFF(pVM)->hm.s.fL1dFlushOnSched)
8787	ASMWrMsr(MSR_IA32_FLUSH_CMD, MSR_IA32_FLUSH_CMD_F_L1D);
8788	else if (pVCpu->CTX_SUFF(pVM)->hm.s.fMdsClearOnSched)
8789	hmR0MdsClear();
8790	}
8791	return rc;
8792	}
8793
8794
8795	/**
8796	* The thread-context callback (only on platforms which support it).
8797	*
8798	* @param enmEvent The thread-context event.
8799	* @param pVCpu The cross context virtual CPU structure.
8800	* @param fGlobalInit Whether global VT-x/AMD-V init. was used.
8801	* @thread EMT(pVCpu)
8802	*/
8803	VMMR0DECL(void) VMXR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPUCC pVCpu, bool fGlobalInit)
8804	{
8805	AssertPtr(pVCpu);
8806	RT_NOREF1(fGlobalInit);
8807
8808	switch (enmEvent)
8809	{
8810	case RTTHREADCTXEVENT_OUT:
8811	{
8812	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8813	Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
8814	VMCPU_ASSERT_EMT(pVCpu);
8815
8816	/* No longjmps (logger flushes, locks) in this fragile context. */
8817	VMMRZCallRing3Disable(pVCpu);
8818	Log4Func(("Preempting: HostCpuId=%u\n", RTMpCpuId()));
8819
8820	/* Restore host-state (FPU, debug etc.) */
8821	if (!pVCpu->hm.s.fLeaveDone)
8822	{
8823	/*
8824	* Do -not- import the guest-state here as we might already be in the middle of importing
8825	* it, esp. bad if we're holding the PGM lock, see comment in hmR0VmxImportGuestState().
8826	*/
8827	hmR0VmxLeave(pVCpu, false /* fImportState */);
8828	pVCpu->hm.s.fLeaveDone = true;
8829	}
8830
8831	/* Leave HM context, takes care of local init (term). */
8832	int rc = HMR0LeaveCpu(pVCpu);
8833	AssertRC(rc);
8834
8835	/* Restore longjmp state. */
8836	VMMRZCallRing3Enable(pVCpu);
8837	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreempt);
8838	break;
8839	}
8840
8841	case RTTHREADCTXEVENT_IN:
8842	{
8843	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8844	Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
8845	VMCPU_ASSERT_EMT(pVCpu);
8846
8847	/* No longjmps here, as we don't want to trigger preemption (& its hook) while resuming. */
8848	VMMRZCallRing3Disable(pVCpu);
8849	Log4Func(("Resumed: HostCpuId=%u\n", RTMpCpuId()));
8850
8851	/* Initialize the bare minimum state required for HM. This takes care of
8852	initializing VT-x if necessary (onlined CPUs, local init etc.) */
8853	int rc = hmR0EnterCpu(pVCpu);
8854	AssertRC(rc);
8855	Assert( (pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
8856	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
8857
8858	/* Load the active VMCS as the current one. */
8859	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8860	rc = hmR0VmxLoadVmcs(pVmcsInfo);
8861	AssertRC(rc);
8862	Log4Func(("Resumed: Loaded Vmcs. HostCpuId=%u\n", RTMpCpuId()));
8863	pVCpu->hm.s.fLeaveDone = false;
8864
8865	/* Do the EMT scheduled L1D flush if needed. */
8866	if (pVCpu->CTX_SUFF(pVM)->hm.s.fL1dFlushOnSched)
8867	ASMWrMsr(MSR_IA32_FLUSH_CMD, MSR_IA32_FLUSH_CMD_F_L1D);
8868
8869	/* Restore longjmp state. */
8870	VMMRZCallRing3Enable(pVCpu);
8871	break;
8872	}
8873
8874	default:
8875	break;
8876	}
8877	}
8878
8879
8880	/**
8881	* Exports the host state into the VMCS host-state area.
8882	* Sets up the VM-exit MSR-load area.
8883	*
8884	* The CPU state will be loaded from these fields on every successful VM-exit.
8885	*
8886	* @returns VBox status code.
8887	* @param pVCpu The cross context virtual CPU structure.
8888	*
8889	* @remarks No-long-jump zone!!!
8890	*/
8891	static int hmR0VmxExportHostState(PVMCPUCC pVCpu)
8892	{
8893	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8894
8895	int rc = VINF_SUCCESS;
8896	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT)
8897	{
8898	hmR0VmxExportHostControlRegs();
8899
8900	rc = hmR0VmxExportHostSegmentRegs(pVCpu);
8901	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
8902
8903	hmR0VmxExportHostMsrs(pVCpu);
8904
8905	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_HOST_CONTEXT;
8906	}
8907	return rc;
8908	}
8909
8910
8911	/**
8912	* Saves the host state in the VMCS host-state.
8913	*
8914	* @returns VBox status code.
8915	* @param pVCpu The cross context virtual CPU structure.
8916	*
8917	* @remarks No-long-jump zone!!!
8918	*/
8919	VMMR0DECL(int) VMXR0ExportHostState(PVMCPUCC pVCpu)
8920	{
8921	AssertPtr(pVCpu);
8922	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8923
8924	/*
8925	* Export the host state here while entering HM context.
8926	* When thread-context hooks are used, we might get preempted and have to re-save the host
8927	* state but most of the time we won't be, so do it here before we disable interrupts.
8928	*/
8929	return hmR0VmxExportHostState(pVCpu);
8930	}
8931
8932
8933	/**
8934	* Exports the guest state into the VMCS guest-state area.
8935	*
8936	* The will typically be done before VM-entry when the guest-CPU state and the
8937	* VMCS state may potentially be out of sync.
8938	*
8939	* Sets up the VM-entry MSR-load and VM-exit MSR-store areas. Sets up the
8940	* VM-entry controls.
8941	* Sets up the appropriate VMX non-root function to execute guest code based on
8942	* the guest CPU mode.
8943	*
8944	* @returns VBox strict status code.
8945	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
8946	* without unrestricted guest execution and the VMMDev is not presently
8947	* mapped (e.g. EFI32).
8948	*
8949	* @param pVCpu The cross context virtual CPU structure.
8950	* @param pVmxTransient The VMX-transient structure.
8951	*
8952	* @remarks No-long-jump zone!!!
8953	*/
8954	static VBOXSTRICTRC hmR0VmxExportGuestState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
8955	{
8956	AssertPtr(pVCpu);
8957	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
8958	LogFlowFunc(("pVCpu=%p\n", pVCpu));
8959
8960	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExportGuestState, x);
8961
8962	/*
8963	* Determine real-on-v86 mode.
8964	* Used when the guest is in real-mode and unrestricted guest execution is not used.
8965	*/
8966	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
8967	if ( pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest
8968	\|\| !CPUMIsGuestInRealModeEx(&pVCpu->cpum.GstCtx))
8969	pVmcsInfo->RealMode. fRealOnV86Active = false;
8970	else
8971	{
8972	Assert(!pVmxTransient->fIsNestedGuest);
8973	pVmcsInfo->RealMode.fRealOnV86Active = true;
8974	}
8975
8976	/*
8977	* Any ordering dependency among the sub-functions below must be explicitly stated using comments.
8978	* Ideally, assert that the cross-dependent bits are up-to-date at the point of using it.
8979	*/
8980	/** @todo r=ramshankar: Move hmR0VmxSelectVMRunHandler inside
8981	* hmR0VmxExportGuestEntryExitCtls and do it conditionally. There shouldn't
8982	* be a need to evaluate this everytime since I'm pretty sure we intercept
8983	* all guest paging mode changes. */
8984	int rc = hmR0VmxSelectVMRunHandler(pVCpu, pVmxTransient);
8985	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
8986
8987	rc = hmR0VmxExportGuestEntryExitCtls(pVCpu, pVmxTransient);
8988	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
8989
8990	rc = hmR0VmxExportGuestCR0(pVCpu, pVmxTransient);
8991	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
8992
8993	VBOXSTRICTRC rcStrict = hmR0VmxExportGuestCR3AndCR4(pVCpu, pVmxTransient);
8994	if (rcStrict == VINF_SUCCESS)
8995	{ /* likely */ }
8996	else
8997	{
8998	Assert(rcStrict == VINF_EM_RESCHEDULE_REM \|\| RT_FAILURE_NP(rcStrict));
8999	return rcStrict;
9000	}
9001
9002	rc = hmR0VmxExportGuestSegRegsXdtr(pVCpu, pVmxTransient);
9003	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9004
9005	rc = hmR0VmxExportGuestMsrs(pVCpu, pVmxTransient);
9006	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9007
9008	rc = hmR0VmxExportGuestApicTpr(pVCpu, pVmxTransient);
9009	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9010
9011	rc = hmR0VmxExportGuestXcptIntercepts(pVCpu, pVmxTransient);
9012	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9013
9014	rc = hmR0VmxExportGuestRip(pVCpu);
9015	rc \|= hmR0VmxExportGuestRsp(pVCpu);
9016	rc \|= hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9017	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9018
9019	rc = hmR0VmxExportGuestHwvirtState(pVCpu, pVmxTransient);
9020	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9021
9022	/* Clear any bits that may be set but exported unconditionally or unused/reserved bits. */
9023	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~( (HM_CHANGED_GUEST_GPRS_MASK & ~HM_CHANGED_GUEST_RSP)
9024	\| HM_CHANGED_GUEST_CR2
9025	\| (HM_CHANGED_GUEST_DR_MASK & ~HM_CHANGED_GUEST_DR7)
9026	\| HM_CHANGED_GUEST_X87
9027	\| HM_CHANGED_GUEST_SSE_AVX
9028	\| HM_CHANGED_GUEST_OTHER_XSAVE
9029	\| HM_CHANGED_GUEST_XCRx
9030	\| HM_CHANGED_GUEST_KERNEL_GS_BASE /* Part of lazy or auto load-store MSRs. */
9031	\| HM_CHANGED_GUEST_SYSCALL_MSRS /* Part of lazy or auto load-store MSRs. */
9032	\| HM_CHANGED_GUEST_TSC_AUX
9033	\| HM_CHANGED_GUEST_OTHER_MSRS
9034	\| (HM_CHANGED_KEEPER_STATE_MASK & ~HM_CHANGED_VMX_MASK)));
9035
9036	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExportGuestState, x);
9037	return rc;
9038	}
9039
9040
9041	/**
9042	* Exports the state shared between the host and guest into the VMCS.
9043	*
9044	* @param pVCpu The cross context virtual CPU structure.
9045	* @param pVmxTransient The VMX-transient structure.
9046	*
9047	* @remarks No-long-jump zone!!!
9048	*/
9049	static void hmR0VmxExportSharedState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9050	{
9051	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9052	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9053
9054	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_DR_MASK)
9055	{
9056	int rc = hmR0VmxExportSharedDebugState(pVCpu, pVmxTransient);
9057	AssertRC(rc);
9058	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_GUEST_DR_MASK;
9059
9060	/* Loading shared debug bits might have changed eflags.TF bit for debugging purposes. */
9061	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_RFLAGS)
9062	{
9063	rc = hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9064	AssertRC(rc);
9065	}
9066	}
9067
9068	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_GUEST_LAZY_MSRS)
9069	{
9070	hmR0VmxLazyLoadGuestMsrs(pVCpu);
9071	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_VMX_GUEST_LAZY_MSRS;
9072	}
9073
9074	AssertMsg(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE),
9075	("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
9076	}
9077
9078
9079	/**
9080	* Worker for loading the guest-state bits in the inner VT-x execution loop.
9081	*
9082	* @returns Strict VBox status code (i.e. informational status codes too).
9083	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
9084	* without unrestricted guest execution and the VMMDev is not presently
9085	* mapped (e.g. EFI32).
9086	*
9087	* @param pVCpu The cross context virtual CPU structure.
9088	* @param pVmxTransient The VMX-transient structure.
9089	*
9090	* @remarks No-long-jump zone!!!
9091	*/
9092	static VBOXSTRICTRC hmR0VmxExportGuestStateOptimal(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9093	{
9094	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9095	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9096	Assert(VMMR0IsLogFlushDisabled(pVCpu));
9097
9098	#ifdef HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
9099	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
9100	#endif
9101
9102	/*
9103	* For many exits it's only RIP that changes and hence try to export it first
9104	* without going through a lot of change flag checks.
9105	*/
9106	VBOXSTRICTRC rcStrict;
9107	uint64_t fCtxChanged = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
9108	RT_UNTRUSTED_NONVOLATILE_COPY_FENCE();
9109	if ((fCtxChanged & (HM_CHANGED_ALL_GUEST & ~HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE)) == HM_CHANGED_GUEST_RIP)
9110	{
9111	rcStrict = hmR0VmxExportGuestRip(pVCpu);
9112	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
9113	{ /* likely */}
9114	else
9115	AssertMsgFailedReturn(("Failed to export guest RIP! rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)), rcStrict);
9116	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportMinimal);
9117	}
9118	else if (fCtxChanged & (HM_CHANGED_ALL_GUEST & ~HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
9119	{
9120	rcStrict = hmR0VmxExportGuestState(pVCpu, pVmxTransient);
9121	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
9122	{ /* likely */}
9123	else
9124	{
9125	AssertMsg(rcStrict == VINF_EM_RESCHEDULE_REM, ("Failed to export guest state! rc=%Rrc\n",
9126	VBOXSTRICTRC_VAL(rcStrict)));
9127	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9128	return rcStrict;
9129	}
9130	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportFull);
9131	}
9132	else
9133	rcStrict = VINF_SUCCESS;
9134
9135	#ifdef VBOX_STRICT
9136	/* All the guest state bits should be loaded except maybe the host context and/or the shared host/guest bits. */
9137	fCtxChanged = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
9138	RT_UNTRUSTED_NONVOLATILE_COPY_FENCE();
9139	AssertMsg(!(fCtxChanged & (HM_CHANGED_ALL_GUEST & ~HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE)),
9140	("fCtxChanged=%#RX64\n", fCtxChanged));
9141	#endif
9142	return rcStrict;
9143	}
9144
9145
9146	/**
9147	* Tries to determine what part of the guest-state VT-x has deemed as invalid
9148	* and update error record fields accordingly.
9149	*
9150	* @returns VMX_IGS_* error codes.
9151	* @retval VMX_IGS_REASON_NOT_FOUND if this function could not find anything
9152	* wrong with the guest state.
9153	*
9154	* @param pVCpu The cross context virtual CPU structure.
9155	* @param pVmcsInfo The VMCS info. object.
9156	*
9157	* @remarks This function assumes our cache of the VMCS controls
9158	* are valid, i.e. hmR0VmxCheckVmcsCtls() succeeded.
9159	*/
9160	static uint32_t hmR0VmxCheckGuestState(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
9161	{
9162	#define HMVMX_ERROR_BREAK(err) { uError = (err); break; }
9163	#define HMVMX_CHECK_BREAK(expr, err) do { \
9164	if (!(expr)) { uError = (err); break; } \
9165	} while (0)
9166
9167	int rc;
9168	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
9169	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
9170	uint32_t uError = VMX_IGS_ERROR;
9171	uint32_t u32Val;
9172	bool const fUnrestrictedGuest = pVM->hm.s.vmx.fUnrestrictedGuest;
9173
9174	do
9175	{
9176	/*
9177	* CR0.
9178	*/
9179	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
9180	uint64_t fSetCr0 = (pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
9181	uint64_t const fZapCr0 = (pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 \| pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
9182	/* Exceptions for unrestricted guest execution for fixed CR0 bits (PE, PG).
9183	See Intel spec. 26.3.1 "Checks on Guest Control Registers, Debug Registers and MSRs." */
9184	if (fUnrestrictedGuest)
9185	fSetCr0 &= ~(uint64_t)(X86_CR0_PE \| X86_CR0_PG);
9186
9187	uint64_t u64GuestCr0;
9188	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64GuestCr0);
9189	AssertRC(rc);
9190	HMVMX_CHECK_BREAK((u64GuestCr0 & fSetCr0) == fSetCr0, VMX_IGS_CR0_FIXED1);
9191	HMVMX_CHECK_BREAK(!(u64GuestCr0 & ~fZapCr0), VMX_IGS_CR0_FIXED0);
9192	if ( !fUnrestrictedGuest
9193	&& (u64GuestCr0 & X86_CR0_PG)
9194	&& !(u64GuestCr0 & X86_CR0_PE))
9195	{
9196	HMVMX_ERROR_BREAK(VMX_IGS_CR0_PG_PE_COMBO);
9197	}
9198
9199	/*
9200	* CR4.
9201	*/
9202	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
9203	uint64_t const fSetCr4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
9204	uint64_t const fZapCr4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 \| pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
9205
9206	uint64_t u64GuestCr4;
9207	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR4, &u64GuestCr4);
9208	AssertRC(rc);
9209	HMVMX_CHECK_BREAK((u64GuestCr4 & fSetCr4) == fSetCr4, VMX_IGS_CR4_FIXED1);
9210	HMVMX_CHECK_BREAK(!(u64GuestCr4 & ~fZapCr4), VMX_IGS_CR4_FIXED0);
9211
9212	/*
9213	* IA32_DEBUGCTL MSR.
9214	*/
9215	uint64_t u64Val;
9216	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, &u64Val);
9217	AssertRC(rc);
9218	if ( (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
9219	&& (u64Val & 0xfffffe3c)) /* Bits 31:9, bits 5:2 MBZ. */
9220	{
9221	HMVMX_ERROR_BREAK(VMX_IGS_DEBUGCTL_MSR_RESERVED);
9222	}
9223	uint64_t u64DebugCtlMsr = u64Val;
9224
9225	#ifdef VBOX_STRICT
9226	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
9227	AssertRC(rc);
9228	Assert(u32Val == pVmcsInfo->u32EntryCtls);
9229	#endif
9230	bool const fLongModeGuest = RT_BOOL(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
9231
9232	/*
9233	* RIP and RFLAGS.
9234	*/
9235	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RIP, &u64Val);
9236	AssertRC(rc);
9237	/* pCtx->rip can be different than the one in the VMCS (e.g. run guest code and VM-exits that don't update it). */
9238	if ( !fLongModeGuest
9239	\|\| !pCtx->cs.Attr.n.u1Long)
9240	{
9241	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffff00000000)), VMX_IGS_LONGMODE_RIP_INVALID);
9242	}
9243	/** @todo If the processor supports N < 64 linear-address bits, bits 63:N
9244	* must be identical if the "IA-32e mode guest" VM-entry
9245	* control is 1 and CS.L is 1. No check applies if the
9246	* CPU supports 64 linear-address bits. */
9247
9248	/* Flags in pCtx can be different (real-on-v86 for instance). We are only concerned about the VMCS contents here. */
9249	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RFLAGS, &u64Val);
9250	AssertRC(rc);
9251	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffc08028)), /* Bit 63:22, Bit 15, 5, 3 MBZ. */
9252	VMX_IGS_RFLAGS_RESERVED);
9253	HMVMX_CHECK_BREAK((u64Val & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
9254	uint32_t const u32Eflags = u64Val;
9255
9256	if ( fLongModeGuest
9257	\|\| ( fUnrestrictedGuest
9258	&& !(u64GuestCr0 & X86_CR0_PE)))
9259	{
9260	HMVMX_CHECK_BREAK(!(u32Eflags & X86_EFL_VM), VMX_IGS_RFLAGS_VM_INVALID);
9261	}
9262
9263	uint32_t u32EntryInfo;
9264	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo);
9265	AssertRC(rc);
9266	if (VMX_ENTRY_INT_INFO_IS_EXT_INT(u32EntryInfo))
9267	HMVMX_CHECK_BREAK(u32Eflags & X86_EFL_IF, VMX_IGS_RFLAGS_IF_INVALID);
9268
9269	/*
9270	* 64-bit checks.
9271	*/
9272	if (fLongModeGuest)
9273	{
9274	HMVMX_CHECK_BREAK(u64GuestCr0 & X86_CR0_PG, VMX_IGS_CR0_PG_LONGMODE);
9275	HMVMX_CHECK_BREAK(u64GuestCr4 & X86_CR4_PAE, VMX_IGS_CR4_PAE_LONGMODE);
9276	}
9277
9278	if ( !fLongModeGuest
9279	&& (u64GuestCr4 & X86_CR4_PCIDE))
9280	{
9281	HMVMX_ERROR_BREAK(VMX_IGS_CR4_PCIDE);
9282	}
9283
9284	/** @todo CR3 field must be such that bits 63:52 and bits in the range
9285	* 51:32 beyond the processor's physical-address width are 0. */
9286
9287	if ( (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
9288	&& (pCtx->dr[7] & X86_DR7_MBZ_MASK))
9289	{
9290	HMVMX_ERROR_BREAK(VMX_IGS_DR7_RESERVED);
9291	}
9292
9293	rc = VMXReadVmcsNw(VMX_VMCS_HOST_SYSENTER_ESP, &u64Val);
9294	AssertRC(rc);
9295	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_ESP_NOT_CANONICAL);
9296
9297	rc = VMXReadVmcsNw(VMX_VMCS_HOST_SYSENTER_EIP, &u64Val);
9298	AssertRC(rc);
9299	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_EIP_NOT_CANONICAL);
9300
9301	/*
9302	* PERF_GLOBAL MSR.
9303	*/
9304	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PERF_MSR)
9305	{
9306	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL, &u64Val);
9307	AssertRC(rc);
9308	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffff8fffffffc)),
9309	VMX_IGS_PERF_GLOBAL_MSR_RESERVED); /* Bits 63:35, bits 31:2 MBZ. */
9310	}
9311
9312	/*
9313	* PAT MSR.
9314	*/
9315	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PAT_MSR)
9316	{
9317	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PAT_FULL, &u64Val);
9318	AssertRC(rc);
9319	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0x707070707070707)), VMX_IGS_PAT_MSR_RESERVED);
9320	for (unsigned i = 0; i < 8; i++)
9321	{
9322	uint8_t u8Val = (u64Val & 0xff);
9323	if ( u8Val != 0 /* UC */
9324	&& u8Val != 1 /* WC */
9325	&& u8Val != 4 /* WT */
9326	&& u8Val != 5 /* WP */
9327	&& u8Val != 6 /* WB */
9328	&& u8Val != 7 /* UC- */)
9329	{
9330	HMVMX_ERROR_BREAK(VMX_IGS_PAT_MSR_INVALID);
9331	}
9332	u64Val >>= 8;
9333	}
9334	}
9335
9336	/*
9337	* EFER MSR.
9338	*/
9339	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_EFER_MSR)
9340	{
9341	Assert(pVM->hm.s.vmx.fSupportsVmcsEfer);
9342	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_EFER_FULL, &u64Val);
9343	AssertRC(rc);
9344	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffffffffff2fe)),
9345	VMX_IGS_EFER_MSR_RESERVED); /* Bits 63:12, bit 9, bits 7:1 MBZ. */
9346	HMVMX_CHECK_BREAK(RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL( pVmcsInfo->u32EntryCtls
9347	& VMX_ENTRY_CTLS_IA32E_MODE_GUEST),
9348	VMX_IGS_EFER_LMA_GUEST_MODE_MISMATCH);
9349	/** @todo r=ramshankar: Unrestricted check here is probably wrong, see
9350	* iemVmxVmentryCheckGuestState(). */
9351	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9352	\|\| !(u64GuestCr0 & X86_CR0_PG)
9353	\|\| RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL(u64Val & MSR_K6_EFER_LME),
9354	VMX_IGS_EFER_LMA_LME_MISMATCH);
9355	}
9356
9357	/*
9358	* Segment registers.
9359	*/
9360	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9361	\|\| !(pCtx->ldtr.Sel & X86_SEL_LDT), VMX_IGS_LDTR_TI_INVALID);
9362	if (!(u32Eflags & X86_EFL_VM))
9363	{
9364	/* CS */
9365	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1Present, VMX_IGS_CS_ATTR_P_INVALID);
9366	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xf00), VMX_IGS_CS_ATTR_RESERVED);
9367	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xfffe0000), VMX_IGS_CS_ATTR_RESERVED);
9368	HMVMX_CHECK_BREAK( (pCtx->cs.u32Limit & 0xfff) == 0xfff
9369	\|\| !(pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9370	HMVMX_CHECK_BREAK( !(pCtx->cs.u32Limit & 0xfff00000)
9371	\|\| (pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9372	/* CS cannot be loaded with NULL in protected mode. */
9373	HMVMX_CHECK_BREAK(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_CS_ATTR_UNUSABLE);
9374	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1DescType, VMX_IGS_CS_ATTR_S_INVALID);
9375	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
9376	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_UNEQUAL);
9377	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
9378	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_MISMATCH);
9379	else if (pVM->hm.s.vmx.fUnrestrictedGuest && pCtx->cs.Attr.n.u4Type == 3)
9380	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == 0, VMX_IGS_CS_ATTR_DPL_INVALID);
9381	else
9382	HMVMX_ERROR_BREAK(VMX_IGS_CS_ATTR_TYPE_INVALID);
9383
9384	/* SS */
9385	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9386	\|\| (pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL), VMX_IGS_SS_CS_RPL_UNEQUAL);
9387	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL), VMX_IGS_SS_ATTR_DPL_RPL_UNEQUAL);
9388	if ( !(pCtx->cr0 & X86_CR0_PE)
9389	\|\| pCtx->cs.Attr.n.u4Type == 3)
9390	{
9391	HMVMX_CHECK_BREAK(!pCtx->ss.Attr.n.u2Dpl, VMX_IGS_SS_ATTR_DPL_INVALID);
9392	}
9393	if (!(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
9394	{
9395	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7, VMX_IGS_SS_ATTR_TYPE_INVALID);
9396	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u1Present, VMX_IGS_SS_ATTR_P_INVALID);
9397	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xf00), VMX_IGS_SS_ATTR_RESERVED);
9398	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xfffe0000), VMX_IGS_SS_ATTR_RESERVED);
9399	HMVMX_CHECK_BREAK( (pCtx->ss.u32Limit & 0xfff) == 0xfff
9400	\|\| !(pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9401	HMVMX_CHECK_BREAK( !(pCtx->ss.u32Limit & 0xfff00000)
9402	\|\| (pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9403	}
9404
9405	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxExportGuestSReg(). */
9406	if (!(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
9407	{
9408	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_DS_ATTR_A_INVALID);
9409	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u1Present, VMX_IGS_DS_ATTR_P_INVALID);
9410	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9411	\|\| pCtx->ds.Attr.n.u4Type > 11
9412	\|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9413	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xf00), VMX_IGS_DS_ATTR_RESERVED);
9414	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xfffe0000), VMX_IGS_DS_ATTR_RESERVED);
9415	HMVMX_CHECK_BREAK( (pCtx->ds.u32Limit & 0xfff) == 0xfff
9416	\|\| !(pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9417	HMVMX_CHECK_BREAK( !(pCtx->ds.u32Limit & 0xfff00000)
9418	\|\| (pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9419	HMVMX_CHECK_BREAK( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9420	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_DS_ATTR_TYPE_INVALID);
9421	}
9422	if (!(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
9423	{
9424	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_ES_ATTR_A_INVALID);
9425	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u1Present, VMX_IGS_ES_ATTR_P_INVALID);
9426	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9427	\|\| pCtx->es.Attr.n.u4Type > 11
9428	\|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9429	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xf00), VMX_IGS_ES_ATTR_RESERVED);
9430	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xfffe0000), VMX_IGS_ES_ATTR_RESERVED);
9431	HMVMX_CHECK_BREAK( (pCtx->es.u32Limit & 0xfff) == 0xfff
9432	\|\| !(pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
9433	HMVMX_CHECK_BREAK( !(pCtx->es.u32Limit & 0xfff00000)
9434	\|\| (pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
9435	HMVMX_CHECK_BREAK( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9436	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_ES_ATTR_TYPE_INVALID);
9437	}
9438	if (!(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
9439	{
9440	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_FS_ATTR_A_INVALID);
9441	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u1Present, VMX_IGS_FS_ATTR_P_INVALID);
9442	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9443	\|\| pCtx->fs.Attr.n.u4Type > 11
9444	\|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL), VMX_IGS_FS_ATTR_DPL_RPL_UNEQUAL);
9445	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xf00), VMX_IGS_FS_ATTR_RESERVED);
9446	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xfffe0000), VMX_IGS_FS_ATTR_RESERVED);
9447	HMVMX_CHECK_BREAK( (pCtx->fs.u32Limit & 0xfff) == 0xfff
9448	\|\| !(pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
9449	HMVMX_CHECK_BREAK( !(pCtx->fs.u32Limit & 0xfff00000)
9450	\|\| (pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
9451	HMVMX_CHECK_BREAK( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9452	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_FS_ATTR_TYPE_INVALID);
9453	}
9454	if (!(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
9455	{
9456	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_GS_ATTR_A_INVALID);
9457	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u1Present, VMX_IGS_GS_ATTR_P_INVALID);
9458	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9459	\|\| pCtx->gs.Attr.n.u4Type > 11
9460	\|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL), VMX_IGS_GS_ATTR_DPL_RPL_UNEQUAL);
9461	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xf00), VMX_IGS_GS_ATTR_RESERVED);
9462	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xfffe0000), VMX_IGS_GS_ATTR_RESERVED);
9463	HMVMX_CHECK_BREAK( (pCtx->gs.u32Limit & 0xfff) == 0xfff
9464	\|\| !(pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
9465	HMVMX_CHECK_BREAK( !(pCtx->gs.u32Limit & 0xfff00000)
9466	\|\| (pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
9467	HMVMX_CHECK_BREAK( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9468	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_GS_ATTR_TYPE_INVALID);
9469	}
9470	/* 64-bit capable CPUs. */
9471	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
9472	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
9473	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9474	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
9475	HMVMX_CHECK_BREAK(!RT_HI_U32(pCtx->cs.u64Base), VMX_IGS_LONGMODE_CS_BASE_INVALID);
9476	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ss.u64Base),
9477	VMX_IGS_LONGMODE_SS_BASE_INVALID);
9478	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ds.u64Base),
9479	VMX_IGS_LONGMODE_DS_BASE_INVALID);
9480	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->es.u64Base),
9481	VMX_IGS_LONGMODE_ES_BASE_INVALID);
9482	}
9483	else
9484	{
9485	/* V86 mode checks. */
9486	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
9487	if (pVmcsInfo->RealMode.fRealOnV86Active)
9488	{
9489	u32CSAttr = 0xf3; u32SSAttr = 0xf3;
9490	u32DSAttr = 0xf3; u32ESAttr = 0xf3;
9491	u32FSAttr = 0xf3; u32GSAttr = 0xf3;
9492	}
9493	else
9494	{
9495	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u;
9496	u32DSAttr = pCtx->ds.Attr.u; u32ESAttr = pCtx->es.Attr.u;
9497	u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
9498	}
9499
9500	/* CS */
9501	HMVMX_CHECK_BREAK((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), VMX_IGS_V86_CS_BASE_INVALID);
9502	HMVMX_CHECK_BREAK(pCtx->cs.u32Limit == 0xffff, VMX_IGS_V86_CS_LIMIT_INVALID);
9503	HMVMX_CHECK_BREAK(u32CSAttr == 0xf3, VMX_IGS_V86_CS_ATTR_INVALID);
9504	/* SS */
9505	HMVMX_CHECK_BREAK((pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4), VMX_IGS_V86_SS_BASE_INVALID);
9506	HMVMX_CHECK_BREAK(pCtx->ss.u32Limit == 0xffff, VMX_IGS_V86_SS_LIMIT_INVALID);
9507	HMVMX_CHECK_BREAK(u32SSAttr == 0xf3, VMX_IGS_V86_SS_ATTR_INVALID);
9508	/* DS */
9509	HMVMX_CHECK_BREAK((pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4), VMX_IGS_V86_DS_BASE_INVALID);
9510	HMVMX_CHECK_BREAK(pCtx->ds.u32Limit == 0xffff, VMX_IGS_V86_DS_LIMIT_INVALID);
9511	HMVMX_CHECK_BREAK(u32DSAttr == 0xf3, VMX_IGS_V86_DS_ATTR_INVALID);
9512	/* ES */
9513	HMVMX_CHECK_BREAK((pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4), VMX_IGS_V86_ES_BASE_INVALID);
9514	HMVMX_CHECK_BREAK(pCtx->es.u32Limit == 0xffff, VMX_IGS_V86_ES_LIMIT_INVALID);
9515	HMVMX_CHECK_BREAK(u32ESAttr == 0xf3, VMX_IGS_V86_ES_ATTR_INVALID);
9516	/* FS */
9517	HMVMX_CHECK_BREAK((pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4), VMX_IGS_V86_FS_BASE_INVALID);
9518	HMVMX_CHECK_BREAK(pCtx->fs.u32Limit == 0xffff, VMX_IGS_V86_FS_LIMIT_INVALID);
9519	HMVMX_CHECK_BREAK(u32FSAttr == 0xf3, VMX_IGS_V86_FS_ATTR_INVALID);
9520	/* GS */
9521	HMVMX_CHECK_BREAK((pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4), VMX_IGS_V86_GS_BASE_INVALID);
9522	HMVMX_CHECK_BREAK(pCtx->gs.u32Limit == 0xffff, VMX_IGS_V86_GS_LIMIT_INVALID);
9523	HMVMX_CHECK_BREAK(u32GSAttr == 0xf3, VMX_IGS_V86_GS_ATTR_INVALID);
9524	/* 64-bit capable CPUs. */
9525	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
9526	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
9527	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9528	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
9529	HMVMX_CHECK_BREAK(!RT_HI_U32(pCtx->cs.u64Base), VMX_IGS_LONGMODE_CS_BASE_INVALID);
9530	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ss.u64Base),
9531	VMX_IGS_LONGMODE_SS_BASE_INVALID);
9532	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ds.u64Base),
9533	VMX_IGS_LONGMODE_DS_BASE_INVALID);
9534	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->es.u64Base),
9535	VMX_IGS_LONGMODE_ES_BASE_INVALID);
9536	}
9537
9538	/*
9539	* TR.
9540	*/
9541	HMVMX_CHECK_BREAK(!(pCtx->tr.Sel & X86_SEL_LDT), VMX_IGS_TR_TI_INVALID);
9542	/* 64-bit capable CPUs. */
9543	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->tr.u64Base), VMX_IGS_TR_BASE_NOT_CANONICAL);
9544	if (fLongModeGuest)
9545	{
9546	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u4Type == 11, /* 64-bit busy TSS. */
9547	VMX_IGS_LONGMODE_TR_ATTR_TYPE_INVALID);
9548	}
9549	else
9550	{
9551	HMVMX_CHECK_BREAK( pCtx->tr.Attr.n.u4Type == 3 /* 16-bit busy TSS. */
9552	\|\| pCtx->tr.Attr.n.u4Type == 11, /* 32-bit busy TSS.*/
9553	VMX_IGS_TR_ATTR_TYPE_INVALID);
9554	}
9555	HMVMX_CHECK_BREAK(!pCtx->tr.Attr.n.u1DescType, VMX_IGS_TR_ATTR_S_INVALID);
9556	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u1Present, VMX_IGS_TR_ATTR_P_INVALID);
9557	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & 0xf00), VMX_IGS_TR_ATTR_RESERVED); /* Bits 11:8 MBZ. */
9558	HMVMX_CHECK_BREAK( (pCtx->tr.u32Limit & 0xfff) == 0xfff
9559	\|\| !(pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
9560	HMVMX_CHECK_BREAK( !(pCtx->tr.u32Limit & 0xfff00000)
9561	\|\| (pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
9562	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_TR_ATTR_UNUSABLE);
9563
9564	/*
9565	* GDTR and IDTR (64-bit capable checks).
9566	*/
9567	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
9568	AssertRC(rc);
9569	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_GDTR_BASE_NOT_CANONICAL);
9570
9571	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
9572	AssertRC(rc);
9573	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_IDTR_BASE_NOT_CANONICAL);
9574
9575	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val);
9576	AssertRC(rc);
9577	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_GDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
9578
9579	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val);
9580	AssertRC(rc);
9581	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_IDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
9582
9583	/*
9584	* Guest Non-Register State.
9585	*/
9586	/* Activity State. */
9587	uint32_t u32ActivityState;
9588	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, &u32ActivityState);
9589	AssertRC(rc);
9590	HMVMX_CHECK_BREAK( !u32ActivityState
9591	\|\| (u32ActivityState & RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Misc, VMX_BF_MISC_ACTIVITY_STATES)),
9592	VMX_IGS_ACTIVITY_STATE_INVALID);
9593	HMVMX_CHECK_BREAK( !(pCtx->ss.Attr.n.u2Dpl)
9594	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT, VMX_IGS_ACTIVITY_STATE_HLT_INVALID);
9595	uint32_t u32IntrState;
9596	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &u32IntrState);
9597	AssertRC(rc);
9598	if ( u32IntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS
9599	\|\| u32IntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
9600	{
9601	HMVMX_CHECK_BREAK(u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE, VMX_IGS_ACTIVITY_STATE_ACTIVE_INVALID);
9602	}
9603
9604	/** @todo Activity state and injecting interrupts. Left as a todo since we
9605	* currently don't use activity states but ACTIVE. */
9606
9607	HMVMX_CHECK_BREAK( !(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)
9608	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT, VMX_IGS_ACTIVITY_STATE_SIPI_WAIT_INVALID);
9609
9610	/* Guest interruptibility-state. */
9611	HMVMX_CHECK_BREAK(!(u32IntrState & 0xffffffe0), VMX_IGS_INTERRUPTIBILITY_STATE_RESERVED);
9612	HMVMX_CHECK_BREAK((u32IntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
9613	!= (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
9614	VMX_IGS_INTERRUPTIBILITY_STATE_STI_MOVSS_INVALID);
9615	HMVMX_CHECK_BREAK( (u32Eflags & X86_EFL_IF)
9616	\|\| !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI),
9617	VMX_IGS_INTERRUPTIBILITY_STATE_STI_EFL_INVALID);
9618	if (VMX_ENTRY_INT_INFO_IS_EXT_INT(u32EntryInfo))
9619	{
9620	HMVMX_CHECK_BREAK( !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
9621	&& !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
9622	VMX_IGS_INTERRUPTIBILITY_STATE_EXT_INT_INVALID);
9623	}
9624	else if (VMX_ENTRY_INT_INFO_IS_XCPT_NMI(u32EntryInfo))
9625	{
9626	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
9627	VMX_IGS_INTERRUPTIBILITY_STATE_MOVSS_INVALID);
9628	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI),
9629	VMX_IGS_INTERRUPTIBILITY_STATE_STI_INVALID);
9630	}
9631	/** @todo Assumes the processor is not in SMM. */
9632	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI),
9633	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_INVALID);
9634	HMVMX_CHECK_BREAK( !(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)
9635	\|\| (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI),
9636	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_SMM_INVALID);
9637	if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
9638	&& VMX_ENTRY_INT_INFO_IS_XCPT_NMI(u32EntryInfo))
9639	{
9640	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI),
9641	VMX_IGS_INTERRUPTIBILITY_STATE_NMI_INVALID);
9642	}
9643
9644	/* Pending debug exceptions. */
9645	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, &u64Val);
9646	AssertRC(rc);
9647	/* Bits 63:15, Bit 13, Bits 11:4 MBZ. */
9648	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffffaff0)), VMX_IGS_LONGMODE_PENDING_DEBUG_RESERVED);
9649	u32Val = u64Val; /* For pending debug exceptions checks below. */
9650
9651	if ( (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
9652	\|\| (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS)
9653	\|\| u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT)
9654	{
9655	if ( (u32Eflags & X86_EFL_TF)
9656	&& !(u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
9657	{
9658	/* Bit 14 is PendingDebug.BS. */
9659	HMVMX_CHECK_BREAK(u32Val & RT_BIT(14), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_SET);
9660	}
9661	if ( !(u32Eflags & X86_EFL_TF)
9662	\|\| (u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
9663	{
9664	/* Bit 14 is PendingDebug.BS. */
9665	HMVMX_CHECK_BREAK(!(u32Val & RT_BIT(14)), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_CLEAR);
9666	}
9667	}
9668
9669	/* VMCS link pointer. */
9670	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, &u64Val);
9671	AssertRC(rc);
9672	if (u64Val != UINT64_C(0xffffffffffffffff))
9673	{
9674	HMVMX_CHECK_BREAK(!(u64Val & 0xfff), VMX_IGS_VMCS_LINK_PTR_RESERVED);
9675	/** @todo Bits beyond the processor's physical-address width MBZ. */
9676	/** @todo SMM checks. */
9677	Assert(pVmcsInfo->HCPhysShadowVmcs == u64Val);
9678	Assert(pVmcsInfo->pvShadowVmcs);
9679	VMXVMCSREVID VmcsRevId;
9680	VmcsRevId.u = (uint32_t )pVmcsInfo->pvShadowVmcs;
9681	HMVMX_CHECK_BREAK(VmcsRevId.n.u31RevisionId == RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_ID),
9682	VMX_IGS_VMCS_LINK_PTR_SHADOW_VMCS_ID_INVALID);
9683	HMVMX_CHECK_BREAK(VmcsRevId.n.fIsShadowVmcs == (uint32_t)!!(pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING),
9684	VMX_IGS_VMCS_LINK_PTR_NOT_SHADOW);
9685	}
9686
9687	/** @todo Checks on Guest Page-Directory-Pointer-Table Entries when guest is
9688	* not using nested paging? */
9689	if ( pVM->hm.s.fNestedPaging
9690	&& !fLongModeGuest
9691	&& CPUMIsGuestInPAEModeEx(pCtx))
9692	{
9693	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &u64Val);
9694	AssertRC(rc);
9695	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
9696
9697	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &u64Val);
9698	AssertRC(rc);
9699	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
9700
9701	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &u64Val);
9702	AssertRC(rc);
9703	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
9704
9705	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &u64Val);
9706	AssertRC(rc);
9707	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
9708	}
9709
9710	/* Shouldn't happen but distinguish it from AssertRCBreak() errors. */
9711	if (uError == VMX_IGS_ERROR)
9712	uError = VMX_IGS_REASON_NOT_FOUND;
9713	} while (0);
9714
9715	pVCpu->hm.s.u32HMError = uError;
9716	return uError;
9717
9718	#undef HMVMX_ERROR_BREAK
9719	#undef HMVMX_CHECK_BREAK
9720	}
9721
9722
9723	/**
9724	* Map the APIC-access page for virtualizing APIC accesses.
9725	*
9726	* This can cause a longjumps to R3 due to the acquisition of the PGM lock. Hence,
9727	* this not done as part of exporting guest state, see @bugref{8721}.
9728	*
9729	* @returns VBox status code.
9730	* @param pVCpu The cross context virtual CPU structure.
9731	*/
9732	static int hmR0VmxMapHCApicAccessPage(PVMCPUCC pVCpu)
9733	{
9734	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
9735	uint64_t const u64MsrApicBase = APICGetBaseMsrNoCheck(pVCpu);
9736
9737	Assert(PDMHasApic(pVM));
9738	Assert(u64MsrApicBase);
9739
9740	RTGCPHYS const GCPhysApicBase = u64MsrApicBase & PAGE_BASE_GC_MASK;
9741	Log4Func(("Mappping HC APIC-access page at %#RGp\n", GCPhysApicBase));
9742
9743	/* Unalias the existing mapping. */
9744	int rc = PGMHandlerPhysicalReset(pVM, GCPhysApicBase);
9745	AssertRCReturn(rc, rc);
9746
9747	/* Map the HC APIC-access page in place of the MMIO page, also updates the shadow page tables if necessary. */
9748	Assert(pVM->hm.s.vmx.HCPhysApicAccess != NIL_RTHCPHYS);
9749	rc = IOMMMIOMapMMIOHCPage(pVM, pVCpu, GCPhysApicBase, pVM->hm.s.vmx.HCPhysApicAccess, X86_PTE_RW \| X86_PTE_P);
9750	AssertRCReturn(rc, rc);
9751
9752	/* Update the per-VCPU cache of the APIC base MSR. */
9753	pVCpu->hm.s.vmx.u64GstMsrApicBase = u64MsrApicBase;
9754	return VINF_SUCCESS;
9755	}
9756
9757
9758	/**
9759	* Worker function passed to RTMpOnSpecific() that is to be called on the target
9760	* CPU.
9761	*
9762	* @param idCpu The ID for the CPU the function is called on.
9763	* @param pvUser1 Null, not used.
9764	* @param pvUser2 Null, not used.
9765	*/
9766	static DECLCALLBACK(void) hmR0DispatchHostNmi(RTCPUID idCpu, void pvUser1, void pvUser2)
9767	{
9768	RT_NOREF3(idCpu, pvUser1, pvUser2);
9769	VMXDispatchHostNmi();
9770	}
9771
9772
9773	/**
9774	* Dispatching an NMI on the host CPU that received it.
9775	*
9776	* @returns VBox status code.
9777	* @param pVCpu The cross context virtual CPU structure.
9778	* @param pVmcsInfo The VMCS info. object corresponding to the VMCS that was
9779	* executing when receiving the host NMI in VMX non-root
9780	* operation.
9781	*/
9782	static int hmR0VmxExitHostNmi(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
9783	{
9784	RTCPUID const idCpu = pVmcsInfo->idHostCpuExec;
9785	Assert(idCpu != NIL_RTCPUID);
9786
9787	/*
9788	* We don't want to delay dispatching the NMI any more than we have to. However,
9789	* we have already chosen -not- to dispatch NMIs when interrupts were still disabled
9790	* after executing guest or nested-guest code for the following reasons:
9791	*
9792	* - We would need to perform VMREADs with interrupts disabled and is orders of
9793	* magnitude worse when we run as a guest hypervisor without VMCS shadowing
9794	* supported by the host hypervisor.
9795	*
9796	* - It affects the common VM-exit scenario and keeps interrupts disabled for a
9797	* longer period of time just for handling an edge case like host NMIs which do
9798	* not occur nearly as frequently as other VM-exits.
9799	*
9800	* Let's cover the most likely scenario first. Check if we are on the target CPU
9801	* and dispatch the NMI right away. This should be much faster than calling into
9802	* RTMpOnSpecific() machinery.
9803	*/
9804	bool fDispatched = false;
9805	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
9806	if (idCpu == RTMpCpuId())
9807	{
9808	VMXDispatchHostNmi();
9809	fDispatched = true;
9810	}
9811	ASMSetFlags(fEFlags);
9812	if (fDispatched)
9813	{
9814	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC);
9815	return VINF_SUCCESS;
9816	}
9817
9818	/*
9819	* RTMpOnSpecific() waits until the worker function has run on the target CPU. So
9820	* there should be no race or recursion even if we are unlucky enough to be preempted
9821	* (to the target CPU) without dispatching the host NMI above.
9822	*/
9823	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGCIpi);
9824	return RTMpOnSpecific(idCpu, &hmR0DispatchHostNmi, NULL /* pvUser1 /, NULL / pvUser2 */);
9825	}
9826
9827
9828	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9829	/**
9830	* Merges the guest with the nested-guest MSR bitmap in preparation of executing the
9831	* nested-guest using hardware-assisted VMX.
9832	*
9833	* @param pVCpu The cross context virtual CPU structure.
9834	* @param pVmcsInfoNstGst The nested-guest VMCS info. object.
9835	* @param pVmcsInfoGst The guest VMCS info. object.
9836	*/
9837	static void hmR0VmxMergeMsrBitmapNested(PCVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfoNstGst, PCVMXVMCSINFO pVmcsInfoGst)
9838	{
9839	uint32_t const cbMsrBitmap = X86_PAGE_4K_SIZE;
9840	uint64_t pu64MsrBitmap = (uint64_t )pVmcsInfoNstGst->pvMsrBitmap;
9841	Assert(pu64MsrBitmap);
9842
9843	/*
9844	* We merge the guest MSR bitmap with the nested-guest MSR bitmap such that any
9845	* MSR that is intercepted by the guest is also intercepted while executing the
9846	* nested-guest using hardware-assisted VMX.
9847	*
9848	* Note! If the nested-guest is not using an MSR bitmap, ever MSR must cause a
9849	* nested-guest VM-exit even if the outer guest is not intercepting some
9850	* MSRs. We cannot assume the caller has initialized the nested-guest
9851	* MSR bitmap in this case.
9852	*
9853	* The guest hypervisor may also switch whether it uses MSR bitmaps for
9854	* each VM-entry, hence initializing it once per-VM while setting up the
9855	* nested-guest VMCS is not sufficient.
9856	*/
9857	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
9858	if (pVmcsNstGst->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
9859	{
9860	uint64_t const pu64MsrBitmapNstGst = (uint64_t const )pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap);
9861	uint64_t const pu64MsrBitmapGst = (uint64_t const )pVmcsInfoGst->pvMsrBitmap;
9862	Assert(pu64MsrBitmapNstGst);
9863	Assert(pu64MsrBitmapGst);
9864
9865	uint32_t const cFrags = cbMsrBitmap / sizeof(uint64_t);
9866	for (uint32_t i = 0; i < cFrags; i++)
9867	pu64MsrBitmap[i] = pu64MsrBitmapNstGst[i] \| pu64MsrBitmapGst[i];
9868	}
9869	else
9870	ASMMemFill32(pu64MsrBitmap, cbMsrBitmap, UINT32_C(0xffffffff));
9871	}
9872
9873
9874	/**
9875	* Merges the guest VMCS in to the nested-guest VMCS controls in preparation of
9876	* hardware-assisted VMX execution of the nested-guest.
9877	*
9878	* For a guest, we don't modify these controls once we set up the VMCS and hence
9879	* this function is never called.
9880	*
9881	* For nested-guests since the guest hypervisor provides these controls on every
9882	* nested-guest VM-entry and could potentially change them everytime we need to
9883	* merge them before every nested-guest VM-entry.
9884	*
9885	* @returns VBox status code.
9886	* @param pVCpu The cross context virtual CPU structure.
9887	*/
9888	static int hmR0VmxMergeVmcsNested(PVMCPUCC pVCpu)
9889	{
9890	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
9891	PCVMXVMCSINFO pVmcsInfoGst = &pVCpu->hm.s.vmx.VmcsInfo;
9892	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
9893	Assert(pVmcsNstGst);
9894
9895	/*
9896	* Merge the controls with the requirements of the guest VMCS.
9897	*
9898	* We do not need to validate the nested-guest VMX features specified in the nested-guest
9899	* VMCS with the features supported by the physical CPU as it's already done by the
9900	* VMLAUNCH/VMRESUME instruction emulation.
9901	*
9902	* This is because the VMX features exposed by CPUM (through CPUID/MSRs) to the guest are
9903	* derived from the VMX features supported by the physical CPU.
9904	*/
9905
9906	/* Pin-based VM-execution controls. */
9907	uint32_t const u32PinCtls = pVmcsNstGst->u32PinCtls \| pVmcsInfoGst->u32PinCtls;
9908
9909	/* Processor-based VM-execution controls. */
9910	uint32_t u32ProcCtls = (pVmcsNstGst->u32ProcCtls & ~VMX_PROC_CTLS_USE_IO_BITMAPS)
9911	\| (pVmcsInfoGst->u32ProcCtls & ~( VMX_PROC_CTLS_INT_WINDOW_EXIT
9912	\| VMX_PROC_CTLS_NMI_WINDOW_EXIT
9913	\| VMX_PROC_CTLS_USE_TPR_SHADOW
9914	\| VMX_PROC_CTLS_MONITOR_TRAP_FLAG));
9915
9916	/* Secondary processor-based VM-execution controls. */
9917	uint32_t const u32ProcCtls2 = (pVmcsNstGst->u32ProcCtls2 & ~VMX_PROC_CTLS2_VPID)
9918	\| (pVmcsInfoGst->u32ProcCtls2 & ~( VMX_PROC_CTLS2_VIRT_APIC_ACCESS
9919	\| VMX_PROC_CTLS2_INVPCID
9920	\| VMX_PROC_CTLS2_VMCS_SHADOWING
9921	\| VMX_PROC_CTLS2_RDTSCP
9922	\| VMX_PROC_CTLS2_XSAVES_XRSTORS
9923	\| VMX_PROC_CTLS2_APIC_REG_VIRT
9924	\| VMX_PROC_CTLS2_VIRT_INT_DELIVERY
9925	\| VMX_PROC_CTLS2_VMFUNC));
9926
9927	/*
9928	* VM-entry controls:
9929	* These controls contains state that depends on the nested-guest state (primarily
9930	* EFER MSR) and is thus not constant between VMLAUNCH/VMRESUME and the nested-guest
9931	* VM-exit. Although the guest hypervisor cannot change it, we need to in order to
9932	* properly continue executing the nested-guest if the EFER MSR changes but does not
9933	* cause a nested-guest VM-exits.
9934	*
9935	* VM-exit controls:
9936	* These controls specify the host state on return. We cannot use the controls from
9937	* the guest hypervisor state as is as it would contain the guest state rather than
9938	* the host state. Since the host state is subject to change (e.g. preemption, trips
9939	* to ring-3, longjmp and rescheduling to a different host CPU) they are not constant
9940	* through VMLAUNCH/VMRESUME and the nested-guest VM-exit.
9941	*
9942	* VM-entry MSR-load:
9943	* The guest MSRs from the VM-entry MSR-load area are already loaded into the guest-CPU
9944	* context by the VMLAUNCH/VMRESUME instruction emulation.
9945	*
9946	* VM-exit MSR-store:
9947	* The VM-exit emulation will take care of populating the MSRs from the guest-CPU context
9948	* back into the VM-exit MSR-store area.
9949	*
9950	* VM-exit MSR-load areas:
9951	* This must contain the real host MSRs with hardware-assisted VMX execution. Hence, we
9952	* can entirely ignore what the guest hypervisor wants to load here.
9953	*/
9954
9955	/*
9956	* Exception bitmap.
9957	*
9958	* We could remove #UD from the guest bitmap and merge it with the nested-guest bitmap
9959	* here (and avoid doing anything while exporting nested-guest state), but to keep the
9960	* code more flexible if intercepting exceptions become more dynamic in the future we do
9961	* it as part of exporting the nested-guest state.
9962	*/
9963	uint32_t const u32XcptBitmap = pVmcsNstGst->u32XcptBitmap \| pVmcsInfoGst->u32XcptBitmap;
9964
9965	/*
9966	* CR0/CR4 guest/host mask.
9967	*
9968	* Modifications by the nested-guest to CR0/CR4 bits owned by the host and the guest must
9969	* cause VM-exits, so we need to merge them here.
9970	*/
9971	uint64_t const u64Cr0Mask = pVmcsNstGst->u64Cr0Mask.u \| pVmcsInfoGst->u64Cr0Mask;
9972	uint64_t const u64Cr4Mask = pVmcsNstGst->u64Cr4Mask.u \| pVmcsInfoGst->u64Cr4Mask;
9973
9974	/*
9975	* Page-fault error-code mask and match.
9976	*
9977	* Although we require unrestricted guest execution (and thereby nested-paging) for
9978	* hardware-assisted VMX execution of nested-guests and thus the outer guest doesn't
9979	* normally intercept #PFs, it might intercept them for debugging purposes.
9980	*
9981	* If the outer guest is not intercepting #PFs, we can use the nested-guest #PF filters.
9982	* If the outer guest is intercepting #PFs, we must intercept all #PFs.
9983	*/
9984	uint32_t u32XcptPFMask;
9985	uint32_t u32XcptPFMatch;
9986	if (!(pVmcsInfoGst->u32XcptBitmap & RT_BIT(X86_XCPT_PF)))
9987	{
9988	u32XcptPFMask = pVmcsNstGst->u32XcptPFMask;
9989	u32XcptPFMatch = pVmcsNstGst->u32XcptPFMatch;
9990	}
9991	else
9992	{
9993	u32XcptPFMask = 0;
9994	u32XcptPFMatch = 0;
9995	}
9996
9997	/*
9998	* Pause-Loop exiting.
9999	*/
10000	uint32_t const cPleGapTicks = RT_MIN(pVM->hm.s.vmx.cPleGapTicks, pVmcsNstGst->u32PleGap);
10001	uint32_t const cPleWindowTicks = RT_MIN(pVM->hm.s.vmx.cPleWindowTicks, pVmcsNstGst->u32PleWindow);
10002
10003	/*
10004	* Pending debug exceptions.
10005	* Currently just copy whatever the nested-guest provides us.
10006	*/
10007	uint64_t const uPendingDbgXcpts = pVmcsNstGst->u64GuestPendingDbgXcpts.u;
10008
10009	/*
10010	* I/O Bitmap.
10011	*
10012	* We do not use the I/O bitmap that may be provided by the guest hypervisor as we always
10013	* intercept all I/O port accesses.
10014	*/
10015	Assert(u32ProcCtls & VMX_PROC_CTLS_UNCOND_IO_EXIT);
10016	Assert(!(u32ProcCtls & VMX_PROC_CTLS_USE_IO_BITMAPS));
10017
10018	/*
10019	* VMCS shadowing.
10020	*
10021	* We do not yet expose VMCS shadowing to the guest and thus VMCS shadowing should not be
10022	* enabled while executing the nested-guest.
10023	*/
10024	Assert(!(u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING));
10025
10026	/*
10027	* APIC-access page.
10028	*/
10029	RTHCPHYS HCPhysApicAccess;
10030	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10031	{
10032	Assert(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS);
10033	RTGCPHYS const GCPhysApicAccess = pVmcsNstGst->u64AddrApicAccess.u;
10034
10035	/** @todo NSTVMX: This is not really correct but currently is required to make
10036	* things work. We need to re-enable the page handler when we fallback to
10037	* IEM execution of the nested-guest! */
10038	PGMHandlerPhysicalPageTempOff(pVM, GCPhysApicAccess, GCPhysApicAccess);
10039
10040	void *pvPage;
10041	PGMPAGEMAPLOCK PgLockApicAccess;
10042	int rc = PGMPhysGCPhys2CCPtr(pVM, GCPhysApicAccess, &pvPage, &PgLockApicAccess);
10043	if (RT_SUCCESS(rc))
10044	{
10045	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysApicAccess, &HCPhysApicAccess);
10046	AssertMsgRCReturn(rc, ("Failed to get host-physical address for APIC-access page at %#RGp\n", GCPhysApicAccess), rc);
10047
10048	/** @todo Handle proper releasing of page-mapping lock later. */
10049	PGMPhysReleasePageMappingLock(pVCpu->CTX_SUFF(pVM), &PgLockApicAccess);
10050	}
10051	else
10052	return rc;
10053	}
10054	else
10055	HCPhysApicAccess = 0;
10056
10057	/*
10058	* Virtual-APIC page and TPR threshold.
10059	*/
10060	PVMXVMCSINFO pVmcsInfoNstGst = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
10061	RTHCPHYS HCPhysVirtApic;
10062	uint32_t u32TprThreshold;
10063	if (u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
10064	{
10065	Assert(pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW);
10066	RTGCPHYS const GCPhysVirtApic = pVmcsNstGst->u64AddrVirtApic.u;
10067
10068	void *pvPage;
10069	PGMPAGEMAPLOCK PgLockVirtApic;
10070	int rc = PGMPhysGCPhys2CCPtr(pVM, GCPhysVirtApic, &pvPage, &PgLockVirtApic);
10071	if (RT_SUCCESS(rc))
10072	{
10073	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysVirtApic, &HCPhysVirtApic);
10074	AssertMsgRCReturn(rc, ("Failed to get host-physical address for virtual-APIC page at %#RGp\n", GCPhysVirtApic), rc);
10075
10076	/** @todo Handle proper releasing of page-mapping lock later. */
10077	PGMPhysReleasePageMappingLock(pVCpu->CTX_SUFF(pVM), &PgLockVirtApic);
10078	}
10079	else
10080	return rc;
10081
10082	u32TprThreshold = pVmcsNstGst->u32TprThreshold;
10083	}
10084	else
10085	{
10086	HCPhysVirtApic = 0;
10087	u32TprThreshold = 0;
10088
10089	/*
10090	* We must make sure CR8 reads/write must cause VM-exits when TPR shadowing is not
10091	* used by the guest hypervisor. Preventing MMIO accesses to the physical APIC will
10092	* be taken care of by EPT/shadow paging.
10093	*/
10094	if (pVM->hm.s.fAllow64BitGuests)
10095	{
10096	u32ProcCtls \|= VMX_PROC_CTLS_CR8_STORE_EXIT
10097	\| VMX_PROC_CTLS_CR8_LOAD_EXIT;
10098	}
10099	}
10100
10101	/*
10102	* Validate basic assumptions.
10103	*/
10104	Assert(pVM->hm.s.vmx.fAllowUnrestricted);
10105	Assert(pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS);
10106	Assert(hmGetVmxActiveVmcsInfo(pVCpu) == pVmcsInfoNstGst);
10107
10108	/*
10109	* Commit it to the nested-guest VMCS.
10110	*/
10111	int rc = VINF_SUCCESS;
10112	if (pVmcsInfoNstGst->u32PinCtls != u32PinCtls)
10113	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, u32PinCtls);
10114	if (pVmcsInfoNstGst->u32ProcCtls != u32ProcCtls)
10115	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, u32ProcCtls);
10116	if (pVmcsInfoNstGst->u32ProcCtls2 != u32ProcCtls2)
10117	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, u32ProcCtls2);
10118	if (pVmcsInfoNstGst->u32XcptBitmap != u32XcptBitmap)
10119	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
10120	if (pVmcsInfoNstGst->u64Cr0Mask != u64Cr0Mask)
10121	rc \|= VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, u64Cr0Mask);
10122	if (pVmcsInfoNstGst->u64Cr4Mask != u64Cr4Mask)
10123	rc \|= VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, u64Cr4Mask);
10124	if (pVmcsInfoNstGst->u32XcptPFMask != u32XcptPFMask)
10125	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, u32XcptPFMask);
10126	if (pVmcsInfoNstGst->u32XcptPFMatch != u32XcptPFMatch)
10127	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, u32XcptPFMatch);
10128	if ( !(u32ProcCtls & VMX_PROC_CTLS_PAUSE_EXIT)
10129	&& (u32ProcCtls2 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT))
10130	{
10131	Assert(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT);
10132	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, cPleGapTicks);
10133	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, cPleWindowTicks);
10134	}
10135	if (u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
10136	{
10137	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
10138	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL, HCPhysVirtApic);
10139	}
10140	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10141	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, HCPhysApicAccess);
10142	rc \|= VMXWriteVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, uPendingDbgXcpts);
10143	AssertRC(rc);
10144
10145	/*
10146	* Update the nested-guest VMCS cache.
10147	*/
10148	pVmcsInfoNstGst->u32PinCtls = u32PinCtls;
10149	pVmcsInfoNstGst->u32ProcCtls = u32ProcCtls;
10150	pVmcsInfoNstGst->u32ProcCtls2 = u32ProcCtls2;
10151	pVmcsInfoNstGst->u32XcptBitmap = u32XcptBitmap;
10152	pVmcsInfoNstGst->u64Cr0Mask = u64Cr0Mask;
10153	pVmcsInfoNstGst->u64Cr4Mask = u64Cr4Mask;
10154	pVmcsInfoNstGst->u32XcptPFMask = u32XcptPFMask;
10155	pVmcsInfoNstGst->u32XcptPFMatch = u32XcptPFMatch;
10156	pVmcsInfoNstGst->HCPhysVirtApic = HCPhysVirtApic;
10157
10158	/*
10159	* We need to flush the TLB if we are switching the APIC-access page address.
10160	* See Intel spec. 28.3.3.4 "Guidelines for Use of the INVEPT Instruction".
10161	*/
10162	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10163	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = true;
10164
10165	/*
10166	* MSR bitmap.
10167	*
10168	* The MSR bitmap address has already been initialized while setting up the nested-guest
10169	* VMCS, here we need to merge the MSR bitmaps.
10170	*/
10171	if (u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
10172	hmR0VmxMergeMsrBitmapNested(pVCpu, pVmcsInfoNstGst, pVmcsInfoGst);
10173
10174	return VINF_SUCCESS;
10175	}
10176	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
10177
10178
10179	/**
10180	* Does the preparations before executing guest code in VT-x.
10181	*
10182	* This may cause longjmps to ring-3 and may even result in rescheduling to the
10183	* recompiler/IEM. We must be cautious what we do here regarding committing
10184	* guest-state information into the VMCS assuming we assuredly execute the
10185	* guest in VT-x mode.
10186	*
10187	* If we fall back to the recompiler/IEM after updating the VMCS and clearing
10188	* the common-state (TRPM/forceflags), we must undo those changes so that the
10189	* recompiler/IEM can (and should) use them when it resumes guest execution.
10190	* Otherwise such operations must be done when we can no longer exit to ring-3.
10191	*
10192	* @returns Strict VBox status code (i.e. informational status codes too).
10193	* @retval VINF_SUCCESS if we can proceed with running the guest, interrupts
10194	* have been disabled.
10195	* @retval VINF_VMX_VMEXIT if a nested-guest VM-exit occurs (e.g., while evaluating
10196	* pending events).
10197	* @retval VINF_EM_RESET if a triple-fault occurs while injecting a
10198	* double-fault into the guest.
10199	* @retval VINF_EM_DBG_STEPPED if @a fStepping is true and an event was
10200	* dispatched directly.
10201	* @retval VINF_* scheduling changes, we have to go back to ring-3.
10202	*
10203	* @param pVCpu The cross context virtual CPU structure.
10204	* @param pVmxTransient The VMX-transient structure.
10205	* @param fStepping Whether we are single-stepping the guest in the
10206	* hypervisor debugger. Makes us ignore some of the reasons
10207	* for returning to ring-3, and return VINF_EM_DBG_STEPPED
10208	* if event dispatching took place.
10209	*/
10210	static VBOXSTRICTRC hmR0VmxPreRunGuest(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, bool fStepping)
10211	{
10212	Assert(VMMRZCallRing3IsEnabled(pVCpu));
10213
10214	Log4Func(("fIsNested=%RTbool fStepping=%RTbool\n", pVmxTransient->fIsNestedGuest, fStepping));
10215
10216	#ifdef VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM
10217	if (pVmxTransient->fIsNestedGuest)
10218	{
10219	RT_NOREF2(pVCpu, fStepping);
10220	Log2Func(("Rescheduling to IEM due to nested-hwvirt or forced IEM exec -> VINF_EM_RESCHEDULE_REM\n"));
10221	return VINF_EM_RESCHEDULE_REM;
10222	}
10223	#endif
10224
10225	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
10226	PGMRZDynMapFlushAutoSet(pVCpu);
10227	#endif
10228
10229	/*
10230	* Check and process force flag actions, some of which might require us to go back to ring-3.
10231	*/
10232	VBOXSTRICTRC rcStrict = hmR0VmxCheckForceFlags(pVCpu, fStepping);
10233	if (rcStrict == VINF_SUCCESS)
10234	{
10235	/* FFs don't get set all the time. */
10236	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10237	if ( pVmxTransient->fIsNestedGuest
10238	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
10239	{
10240	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
10241	return VINF_VMX_VMEXIT;
10242	}
10243	#endif
10244	}
10245	else
10246	return rcStrict;
10247
10248	/*
10249	* Virtualize memory-mapped accesses to the physical APIC (may take locks).
10250	*/
10251	/** @todo Doing this from ring-3 after VM setup phase causes a
10252	* VERR_IOM_MMIO_RANGE_NOT_FOUND guru while booting Visa 64 SMP VM. No
10253	* idea why atm. */
10254	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10255	if ( !pVCpu->hm.s.vmx.u64GstMsrApicBase
10256	&& (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10257	&& PDMHasApic(pVM))
10258	{
10259	int rc = hmR0VmxMapHCApicAccessPage(pVCpu);
10260	AssertRCReturn(rc, rc);
10261	}
10262
10263	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10264	/*
10265	* Merge guest VMCS controls with the nested-guest VMCS controls.
10266	*
10267	* Even if we have not executed the guest prior to this (e.g. when resuming from a
10268	* saved state), we should be okay with merging controls as we initialize the
10269	* guest VMCS controls as part of VM setup phase.
10270	*/
10271	if ( pVmxTransient->fIsNestedGuest
10272	&& !pVCpu->hm.s.vmx.fMergedNstGstCtls)
10273	{
10274	int rc = hmR0VmxMergeVmcsNested(pVCpu);
10275	AssertRCReturn(rc, rc);
10276	pVCpu->hm.s.vmx.fMergedNstGstCtls = true;
10277	}
10278	#endif
10279
10280	/*
10281	* Evaluate events to be injected into the guest.
10282	*
10283	* Events in TRPM can be injected without inspecting the guest state.
10284	* If any new events (interrupts/NMI) are pending currently, we try to set up the
10285	* guest to cause a VM-exit the next time they are ready to receive the event.
10286	*
10287	* With nested-guests, evaluating pending events may cause VM-exits.
10288	*/
10289	if (TRPMHasTrap(pVCpu))
10290	{
10291	Assert(!pVmxTransient->fIsNestedGuest \|\| !pVCpu->cpum.GstCtx.hwvirt.vmx.fInterceptEvents);
10292	hmR0VmxTrpmTrapToPendingEvent(pVCpu);
10293	}
10294
10295	uint32_t fIntrState;
10296	rcStrict = hmR0VmxEvaluatePendingEvent(pVCpu, pVmxTransient, &fIntrState);
10297
10298	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10299	/*
10300	* While evaluating pending events if something failed (unlikely) or if we were
10301	* preparing to run a nested-guest but performed a nested-guest VM-exit, we should bail.
10302	*/
10303	if (rcStrict != VINF_SUCCESS)
10304	return rcStrict;
10305	if ( pVmxTransient->fIsNestedGuest
10306	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
10307	{
10308	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
10309	return VINF_VMX_VMEXIT;
10310	}
10311	#else
10312	Assert(rcStrict == VINF_SUCCESS);
10313	#endif
10314
10315	/*
10316	* Event injection may take locks (currently the PGM lock for real-on-v86 case) and thus
10317	* needs to be done with longjmps or interrupts + preemption enabled. Event injection might
10318	* also result in triple-faulting the VM.
10319	*
10320	* With nested-guests, the above does not apply since unrestricted guest execution is a
10321	* requirement. Regardless, we do this here to avoid duplicating code elsewhere.
10322	*/
10323	rcStrict = hmR0VmxInjectPendingEvent(pVCpu, pVmxTransient, fIntrState, fStepping);
10324	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
10325	{ /* likely */ }
10326	else
10327	{
10328	AssertMsg(rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
10329	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
10330	return rcStrict;
10331	}
10332
10333	/*
10334	* A longjump might result in importing CR3 even for VM-exits that don't necessarily
10335	* import CR3 themselves. We will need to update them here, as even as late as the above
10336	* hmR0VmxInjectPendingEvent() call may lazily import guest-CPU state on demand causing
10337	* the below force flags to be set.
10338	*/
10339	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
10340	{
10341	Assert(!(ASMAtomicUoReadU64(&pVCpu->cpum.GstCtx.fExtrn) & CPUMCTX_EXTRN_CR3));
10342	int rc2 = PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
10343	AssertMsgReturn(rc2 == VINF_SUCCESS \|\| rc2 == VINF_PGM_SYNC_CR3,
10344	("%Rrc\n", rc2), RT_FAILURE_NP(rc2) ? rc2 : VERR_IPE_UNEXPECTED_INFO_STATUS);
10345	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
10346	}
10347	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
10348	{
10349	PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
10350	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
10351	}
10352
10353	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10354	/* Paranoia. */
10355	Assert(!pVmxTransient->fIsNestedGuest \|\| CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
10356	#endif
10357
10358	/*
10359	* No longjmps to ring-3 from this point on!!!
10360	* Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
10361	* This also disables flushing of the R0-logger instance (if any).
10362	*/
10363	VMMRZCallRing3Disable(pVCpu);
10364
10365	/*
10366	* Export the guest state bits.
10367	*
10368	* We cannot perform longjmps while loading the guest state because we do not preserve the
10369	* host/guest state (although the VMCS will be preserved) across longjmps which can cause
10370	* CPU migration.
10371	*
10372	* If we are injecting events to a real-on-v86 mode guest, we would have updated RIP and some segment
10373	* registers. Hence, exporting of the guest state needs to be done -after- injection of events.
10374	*/
10375	rcStrict = hmR0VmxExportGuestStateOptimal(pVCpu, pVmxTransient);
10376	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
10377	{ /* likely */ }
10378	else
10379	{
10380	VMMRZCallRing3Enable(pVCpu);
10381	return rcStrict;
10382	}
10383
10384	/*
10385	* We disable interrupts so that we don't miss any interrupts that would flag preemption
10386	* (IPI/timers etc.) when thread-context hooks aren't used and we've been running with
10387	* preemption disabled for a while. Since this is purely to aid the
10388	* RTThreadPreemptIsPending() code, it doesn't matter that it may temporarily reenable and
10389	* disable interrupt on NT.
10390	*
10391	* We need to check for force-flags that could've possible been altered since we last
10392	* checked them (e.g. by PDMGetInterrupt() leaving the PDM critical section,
10393	* see @bugref{6398}).
10394	*
10395	* We also check a couple of other force-flags as a last opportunity to get the EMT back
10396	* to ring-3 before executing guest code.
10397	*/
10398	pVmxTransient->fEFlags = ASMIntDisableFlags();
10399
10400	if ( ( !VM_FF_IS_ANY_SET(pVM, VM_FF_EMT_RENDEZVOUS \| VM_FF_TM_VIRTUAL_SYNC)
10401	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
10402	\|\| ( fStepping /* Optimized for the non-stepping case, so a bit of unnecessary work when stepping. */
10403	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK & ~(VMCPU_FF_TIMER \| VMCPU_FF_PDM_CRITSECT))) )
10404	{
10405	if (!RTThreadPreemptIsPending(NIL_RTTHREAD))
10406	{
10407	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10408	/*
10409	* If we are executing a nested-guest make sure that we should intercept subsequent
10410	* events. The one we are injecting might be part of VM-entry. This is mainly to keep
10411	* the VM-exit instruction emulation happy.
10412	*/
10413	if (pVmxTransient->fIsNestedGuest)
10414	pVCpu->cpum.GstCtx.hwvirt.vmx.fInterceptEvents = true;
10415	#endif
10416
10417	/*
10418	* We've injected any pending events. This is really the point of no return (to ring-3).
10419	*
10420	* Note! The caller expects to continue with interrupts & longjmps disabled on successful
10421	* returns from this function, so do -not- enable them here.
10422	*/
10423	pVCpu->hm.s.Event.fPending = false;
10424	return VINF_SUCCESS;
10425	}
10426
10427	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPendingHostIrq);
10428	rcStrict = VINF_EM_RAW_INTERRUPT;
10429	}
10430	else
10431	{
10432	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
10433	rcStrict = VINF_EM_RAW_TO_R3;
10434	}
10435
10436	ASMSetFlags(pVmxTransient->fEFlags);
10437	VMMRZCallRing3Enable(pVCpu);
10438
10439	return rcStrict;
10440	}
10441
10442
10443	/**
10444	* Final preparations before executing guest code using hardware-assisted VMX.
10445	*
10446	* We can no longer get preempted to a different host CPU and there are no returns
10447	* to ring-3. We ignore any errors that may happen from this point (e.g. VMWRITE
10448	* failures), this function is not intended to fail sans unrecoverable hardware
10449	* errors.
10450	*
10451	* @param pVCpu The cross context virtual CPU structure.
10452	* @param pVmxTransient The VMX-transient structure.
10453	*
10454	* @remarks Called with preemption disabled.
10455	* @remarks No-long-jump zone!!!
10456	*/
10457	static void hmR0VmxPreRunGuestCommitted(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
10458	{
10459	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
10460	Assert(VMMR0IsLogFlushDisabled(pVCpu));
10461	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
10462	Assert(!pVCpu->hm.s.Event.fPending);
10463
10464	/*
10465	* Indicate start of guest execution and where poking EMT out of guest-context is recognized.
10466	*/
10467	VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
10468	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC);
10469
10470	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10471	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
10472	PHMPHYSCPU pHostCpu = hmR0GetCurrentCpu();
10473	RTCPUID const idCurrentCpu = pHostCpu->idCpu;
10474
10475	if (!CPUMIsGuestFPUStateActive(pVCpu))
10476	{
10477	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestFpuState, x);
10478	if (CPUMR0LoadGuestFPU(pVM, pVCpu) == VINF_CPUM_HOST_CR0_MODIFIED)
10479	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_HOST_CONTEXT;
10480	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestFpuState, x);
10481	STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadGuestFpu);
10482	}
10483
10484	/*
10485	* Re-export the host state bits as we may've been preempted (only happens when
10486	* thread-context hooks are used or when the VM start function changes) or if
10487	* the host CR0 is modified while loading the guest FPU state above.
10488	*
10489	* The 64-on-32 switcher saves the (64-bit) host state into the VMCS and if we
10490	* changed the switcher back to 32-bit, we must save the 32-bit host state here,
10491	* see @bugref{8432}.
10492	*
10493	* This may also happen when switching to/from a nested-guest VMCS without leaving
10494	* ring-0.
10495	*/
10496	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT)
10497	{
10498	hmR0VmxExportHostState(pVCpu);
10499	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportHostState);
10500	}
10501	Assert(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT));
10502
10503	/*
10504	* Export the state shared between host and guest (FPU, debug, lazy MSRs).
10505	*/
10506	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE)
10507	hmR0VmxExportSharedState(pVCpu, pVmxTransient);
10508	AssertMsg(!pVCpu->hm.s.fCtxChanged, ("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
10509
10510	/*
10511	* Store status of the shared guest/host debug state at the time of VM-entry.
10512	*/
10513	pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
10514	pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
10515
10516	/*
10517	* Always cache the TPR-shadow if the virtual-APIC page exists, thereby skipping
10518	* more than one conditional check. The post-run side of our code shall determine
10519	* if it needs to sync. the virtual APIC TPR with the TPR-shadow.
10520	*/
10521	if (pVmcsInfo->pbVirtApic)
10522	pVmxTransient->u8GuestTpr = pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR];
10523
10524	/*
10525	* Update the host MSRs values in the VM-exit MSR-load area.
10526	*/
10527	if (!pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs)
10528	{
10529	if (pVmcsInfo->cExitMsrLoad > 0)
10530	hmR0VmxUpdateAutoLoadHostMsrs(pVCpu, pVmcsInfo);
10531	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = true;
10532	}
10533
10534	/*
10535	* Evaluate if we need to intercept guest RDTSC/P accesses. Set up the
10536	* VMX-preemption timer based on the next virtual sync clock deadline.
10537	*/
10538	if ( !pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer
10539	\|\| idCurrentCpu != pVCpu->hm.s.idLastCpu)
10540	{
10541	hmR0VmxUpdateTscOffsettingAndPreemptTimer(pVCpu, pVmxTransient);
10542	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = true;
10543	}
10544
10545	/* Record statistics of how often we use TSC offsetting as opposed to intercepting RDTSC/P. */
10546	bool const fIsRdtscIntercepted = RT_BOOL(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_RDTSC_EXIT);
10547	if (!fIsRdtscIntercepted)
10548	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
10549	else
10550	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
10551
10552	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB flushing, set this across the world switch. */
10553	hmR0VmxFlushTaggedTlb(pHostCpu, pVCpu, pVmcsInfo); /* Invalidate the appropriate guest entries from the TLB. */
10554	Assert(idCurrentCpu == pVCpu->hm.s.idLastCpu);
10555	pVCpu->hm.s.vmx.LastError.idCurrentCpu = idCurrentCpu; /* Record the error reporting info. with the current host CPU. */
10556	pVmcsInfo->idHostCpuState = idCurrentCpu; /* Record the CPU for which the host-state has been exported. */
10557	pVmcsInfo->idHostCpuExec = idCurrentCpu; /* Record the CPU on which we shall execute. */
10558
10559	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
10560
10561	TMNotifyStartOfExecution(pVM, pVCpu); /* Notify TM to resume its clocks when TSC is tied to execution,
10562	as we're about to start executing the guest. */
10563
10564	/*
10565	* Load the guest TSC_AUX MSR when we are not intercepting RDTSCP.
10566	*
10567	* This is done this late as updating the TSC offsetting/preemption timer above
10568	* figures out if we can skip intercepting RDTSCP by calculating the number of
10569	* host CPU ticks till the next virtual sync deadline (for the dynamic case).
10570	*/
10571	if ( (pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_RDTSCP)
10572	&& !fIsRdtscIntercepted)
10573	{
10574	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_TSC_AUX);
10575
10576	/* NB: Because we call hmR0VmxAddAutoLoadStoreMsr with fUpdateHostMsr=true,
10577	it's safe even after hmR0VmxUpdateAutoLoadHostMsrs has already been done. */
10578	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_TSC_AUX, CPUMGetGuestTscAux(pVCpu),
10579	true /* fSetReadWrite /, true / fUpdateHostMsr */);
10580	AssertRC(rc);
10581	Assert(!pVmxTransient->fRemoveTscAuxMsr);
10582	pVmxTransient->fRemoveTscAuxMsr = true;
10583	}
10584
10585	#ifdef VBOX_STRICT
10586	Assert(pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs);
10587	hmR0VmxCheckAutoLoadStoreMsrs(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest);
10588	hmR0VmxCheckHostEferMsr(pVCpu, pVmcsInfo);
10589	AssertRC(hmR0VmxCheckVmcsCtls(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest));
10590	#endif
10591
10592	#ifdef HMVMX_ALWAYS_CHECK_GUEST_STATE
10593	/** @todo r=ramshankar: We can now probably use iemVmxVmentryCheckGuestState here.
10594	* Add a PVMXMSRS parameter to it, so that IEM can look at the host MSRs,
10595	* see @bugref{9180#c54}. */
10596	uint32_t const uInvalidReason = hmR0VmxCheckGuestState(pVCpu, pVmcsInfo);
10597	if (uInvalidReason != VMX_IGS_REASON_NOT_FOUND)
10598	Log4(("hmR0VmxCheckGuestState returned %#x\n", uInvalidReason));
10599	#endif
10600	}
10601
10602
10603	/**
10604	* First C routine invoked after running guest code using hardware-assisted VMX.
10605	*
10606	* @param pVCpu The cross context virtual CPU structure.
10607	* @param pVmxTransient The VMX-transient structure.
10608	* @param rcVMRun Return code of VMLAUNCH/VMRESUME.
10609	*
10610	* @remarks Called with interrupts disabled, and returns with interrupts enabled!
10611	*
10612	* @remarks No-long-jump zone!!! This function will however re-enable longjmps
10613	* unconditionally when it is safe to do so.
10614	*/
10615	static void hmR0VmxPostRunGuest(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, int rcVMRun)
10616	{
10617	uint64_t const uHostTsc = ASMReadTSC(); /** @todo We can do a lot better here, see @bugref{9180#c38}. */
10618
10619	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB flushing. */
10620	ASMAtomicIncU32(&pVCpu->hm.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for EMT poking. */
10621	pVCpu->hm.s.fCtxChanged = 0; /* Exits/longjmps to ring-3 requires saving the guest state. */
10622	pVmxTransient->fVmcsFieldsRead = 0; /* Transient fields need to be read from the VMCS. */
10623	pVmxTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
10624	pVmxTransient->fVectoringDoublePF = false; /* Vectoring double page-fault needs to be determined later. */
10625
10626	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
10627	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_RDTSC_EXIT))
10628	{
10629	uint64_t uGstTsc;
10630	if (!pVmxTransient->fIsNestedGuest)
10631	uGstTsc = uHostTsc + pVmcsInfo->u64TscOffset;
10632	else
10633	{
10634	uint64_t const uNstGstTsc = uHostTsc + pVmcsInfo->u64TscOffset;
10635	uGstTsc = CPUMRemoveNestedGuestTscOffset(pVCpu, uNstGstTsc);
10636	}
10637	TMCpuTickSetLastSeen(pVCpu, uGstTsc); /* Update TM with the guest TSC. */
10638	}
10639
10640	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatPreExit, x);
10641	TMNotifyEndOfExecution(pVCpu->CTX_SUFF(pVM), pVCpu); /* Notify TM that the guest is no longer running. */
10642	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
10643
10644	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_REQUIRED; /* Some host state messed up by VMX needs restoring. */
10645	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
10646	#ifdef VBOX_STRICT
10647	hmR0VmxCheckHostEferMsr(pVCpu, pVmcsInfo); /* Verify that the host EFER MSR wasn't modified. */
10648	#endif
10649	Assert(!ASMIntAreEnabled());
10650	ASMSetFlags(pVmxTransient->fEFlags); /* Enable interrupts. */
10651	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
10652
10653	#ifdef HMVMX_ALWAYS_CLEAN_TRANSIENT
10654	/*
10655	* Clean all the VMCS fields in the transient structure before reading
10656	* anything from the VMCS.
10657	*/
10658	pVmxTransient->uExitReason = 0;
10659	pVmxTransient->uExitIntErrorCode = 0;
10660	pVmxTransient->uExitQual = 0;
10661	pVmxTransient->uGuestLinearAddr = 0;
10662	pVmxTransient->uExitIntInfo = 0;
10663	pVmxTransient->cbExitInstr = 0;
10664	pVmxTransient->ExitInstrInfo.u = 0;
10665	pVmxTransient->uEntryIntInfo = 0;
10666	pVmxTransient->uEntryXcptErrorCode = 0;
10667	pVmxTransient->cbEntryInstr = 0;
10668	pVmxTransient->uIdtVectoringInfo = 0;
10669	pVmxTransient->uIdtVectoringErrorCode = 0;
10670	#endif
10671
10672	/*
10673	* Save the basic VM-exit reason and check if the VM-entry failed.
10674	* See Intel spec. 24.9.1 "Basic VM-exit Information".
10675	*/
10676	uint32_t uExitReason;
10677	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
10678	AssertRC(rc);
10679	pVmxTransient->uExitReason = VMX_EXIT_REASON_BASIC(uExitReason);
10680	pVmxTransient->fVMEntryFailed = VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason);
10681
10682	/*
10683	* Log the VM-exit before logging anything else as otherwise it might be a
10684	* tad confusing what happens before and after the world-switch.
10685	*/
10686	HMVMX_LOG_EXIT(pVCpu, uExitReason);
10687
10688	/*
10689	* Remove the TSC_AUX MSR from the auto-load/store MSR area and reset any MSR
10690	* bitmap permissions, if it was added before VM-entry.
10691	*/
10692	if (pVmxTransient->fRemoveTscAuxMsr)
10693	{
10694	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_TSC_AUX);
10695	pVmxTransient->fRemoveTscAuxMsr = false;
10696	}
10697
10698	/*
10699	* Check if VMLAUNCH/VMRESUME succeeded.
10700	* If this failed, we cause a guru meditation and cease further execution.
10701	*
10702	* However, if we are executing a nested-guest we might fail if we use the
10703	* fast path rather than fully emulating VMLAUNCH/VMRESUME instruction in IEM.
10704	*/
10705	if (RT_LIKELY(rcVMRun == VINF_SUCCESS))
10706	{
10707	/*
10708	* Update the VM-exit history array here even if the VM-entry failed due to:
10709	* - Invalid guest state.
10710	* - MSR loading.
10711	* - Machine-check event.
10712	*
10713	* In any of the above cases we will still have a "valid" VM-exit reason
10714	* despite @a fVMEntryFailed being false.
10715	*
10716	* See Intel spec. 26.7 "VM-Entry failures during or after loading guest state".
10717	*
10718	* Note! We don't have CS or RIP at this point. Will probably address that later
10719	* by amending the history entry added here.
10720	*/
10721	EMHistoryAddExit(pVCpu, EMEXIT_MAKE_FT(EMEXIT_F_KIND_VMX, pVmxTransient->uExitReason & EMEXIT_F_TYPE_MASK),
10722	UINT64_MAX, uHostTsc);
10723
10724	if (RT_LIKELY(!pVmxTransient->fVMEntryFailed))
10725	{
10726	VMMRZCallRing3Enable(pVCpu);
10727
10728	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
10729	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
10730
10731	#ifdef HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
10732	hmR0VmxReadAllRoFieldsVmcs(pVmxTransient);
10733	#endif
10734	#if defined(HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE) \|\| defined(HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE)
10735	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
10736	AssertRC(rc);
10737	#elif defined(HMVMX_ALWAYS_SAVE_GUEST_RFLAGS)
10738	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_RFLAGS);
10739	AssertRC(rc);
10740	#else
10741	/*
10742	* Import the guest-interruptibility state always as we need it while evaluating
10743	* injecting events on re-entry.
10744	*
10745	* We don't import CR0 (when unrestricted guest execution is unavailable) despite
10746	* checking for real-mode while exporting the state because all bits that cause
10747	* mode changes wrt CR0 are intercepted.
10748	*/
10749	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_HM_VMX_INT_STATE);
10750	AssertRC(rc);
10751	#endif
10752
10753	/*
10754	* Sync the TPR shadow with our APIC state.
10755	*/
10756	if ( !pVmxTransient->fIsNestedGuest
10757	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW))
10758	{
10759	Assert(pVmcsInfo->pbVirtApic);
10760	if (pVmxTransient->u8GuestTpr != pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR])
10761	{
10762	rc = APICSetTpr(pVCpu, pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR]);
10763	AssertRC(rc);
10764	ASMAtomicOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
10765	}
10766	}
10767
10768	Assert(VMMRZCallRing3IsEnabled(pVCpu));
10769	return;
10770	}
10771	}
10772	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10773	else if (pVmxTransient->fIsNestedGuest)
10774	AssertMsgFailed(("VMLAUNCH/VMRESUME failed but shouldn't happen when VMLAUNCH/VMRESUME was emulated in IEM!\n"));
10775	#endif
10776	else
10777	Log4Func(("VM-entry failure: rcVMRun=%Rrc fVMEntryFailed=%RTbool\n", rcVMRun, pVmxTransient->fVMEntryFailed));
10778
10779	VMMRZCallRing3Enable(pVCpu);
10780	}
10781
10782
10783	/**
10784	* Runs the guest code using hardware-assisted VMX the normal way.
10785	*
10786	* @returns VBox status code.
10787	* @param pVCpu The cross context virtual CPU structure.
10788	* @param pcLoops Pointer to the number of executed loops.
10789	*/
10790	static VBOXSTRICTRC hmR0VmxRunGuestCodeNormal(PVMCPUCC pVCpu, uint32_t *pcLoops)
10791	{
10792	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hm.s.cMaxResumeLoops;
10793	Assert(pcLoops);
10794	Assert(*pcLoops <= cMaxResumeLoops);
10795	Assert(!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
10796
10797	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10798	/*
10799	* Switch to the guest VMCS as we may have transitioned from executing the nested-guest
10800	* without leaving ring-0. Otherwise, if we came from ring-3 we would have loaded the
10801	* guest VMCS while entering the VMX ring-0 session.
10802	*/
10803	if (pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs)
10804	{
10805	int rc = hmR0VmxSwitchToGstOrNstGstVmcs(pVCpu, false /* fSwitchToNstGstVmcs */);
10806	if (RT_SUCCESS(rc))
10807	{ /* likely */ }
10808	else
10809	{
10810	LogRelFunc(("Failed to switch to the guest VMCS. rc=%Rrc\n", rc));
10811	return rc;
10812	}
10813	}
10814	#endif
10815
10816	VMXTRANSIENT VmxTransient;
10817	RT_ZERO(VmxTransient);
10818	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
10819
10820	/* Paranoia. */
10821	Assert(VmxTransient.pVmcsInfo == &pVCpu->hm.s.vmx.VmcsInfo);
10822
10823	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
10824	for (;;)
10825	{
10826	Assert(!HMR0SuspendPending());
10827	HMVMX_ASSERT_CPU_SAFE(pVCpu);
10828	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
10829
10830	/*
10831	* Preparatory work for running nested-guest code, this may force us to
10832	* return to ring-3.
10833	*
10834	* Warning! This bugger disables interrupts on VINF_SUCCESS!
10835	*/
10836	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, false /* fStepping */);
10837	if (rcStrict != VINF_SUCCESS)
10838	break;
10839
10840	/* Interrupts are disabled at this point! */
10841	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
10842	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
10843	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
10844	/* Interrupts are re-enabled at this point! */
10845
10846	/*
10847	* Check for errors with running the VM (VMLAUNCH/VMRESUME).
10848	*/
10849	if (RT_SUCCESS(rcRun))
10850	{ /* very likely */ }
10851	else
10852	{
10853	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
10854	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
10855	return rcRun;
10856	}
10857
10858	/*
10859	* Profile the VM-exit.
10860	*/
10861	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
10862	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
10863	STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
10864	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
10865	HMVMX_START_EXIT_DISPATCH_PROF();
10866
10867	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
10868
10869	/*
10870	* Handle the VM-exit.
10871	*/
10872	#ifdef HMVMX_USE_FUNCTION_TABLE
10873	rcStrict = g_apfnVMExitHandlers[VmxTransient.uExitReason](pVCpu, &VmxTransient);
10874	#else
10875	rcStrict = hmR0VmxHandleExit(pVCpu, &VmxTransient);
10876	#endif
10877	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
10878	if (rcStrict == VINF_SUCCESS)
10879	{
10880	if (++(*pcLoops) <= cMaxResumeLoops)
10881	continue;
10882	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
10883	rcStrict = VINF_EM_RAW_INTERRUPT;
10884	}
10885	break;
10886	}
10887
10888	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
10889	return rcStrict;
10890	}
10891
10892
10893	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10894	/**
10895	* Runs the nested-guest code using hardware-assisted VMX.
10896	*
10897	* @returns VBox status code.
10898	* @param pVCpu The cross context virtual CPU structure.
10899	* @param pcLoops Pointer to the number of executed loops.
10900	*
10901	* @sa hmR0VmxRunGuestCodeNormal.
10902	*/
10903	static VBOXSTRICTRC hmR0VmxRunGuestCodeNested(PVMCPUCC pVCpu, uint32_t *pcLoops)
10904	{
10905	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hm.s.cMaxResumeLoops;
10906	Assert(pcLoops);
10907	Assert(*pcLoops <= cMaxResumeLoops);
10908	Assert(CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
10909
10910	/*
10911	* Switch to the nested-guest VMCS as we may have transitioned from executing the
10912	* guest without leaving ring-0. Otherwise, if we came from ring-3 we would have
10913	* loaded the nested-guest VMCS while entering the VMX ring-0 session.
10914	*/
10915	if (!pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs)
10916	{
10917	int rc = hmR0VmxSwitchToGstOrNstGstVmcs(pVCpu, true /* fSwitchToNstGstVmcs */);
10918	if (RT_SUCCESS(rc))
10919	{ /* likely */ }
10920	else
10921	{
10922	LogRelFunc(("Failed to switch to the nested-guest VMCS. rc=%Rrc\n", rc));
10923	return rc;
10924	}
10925	}
10926
10927	VMXTRANSIENT VmxTransient;
10928	RT_ZERO(VmxTransient);
10929	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
10930	VmxTransient.fIsNestedGuest = true;
10931
10932	/* Paranoia. */
10933	Assert(VmxTransient.pVmcsInfo == &pVCpu->hm.s.vmx.VmcsInfoNstGst);
10934
10935	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
10936	for (;;)
10937	{
10938	Assert(!HMR0SuspendPending());
10939	HMVMX_ASSERT_CPU_SAFE(pVCpu);
10940	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
10941
10942	/*
10943	* Preparatory work for running guest code, this may force us to
10944	* return to ring-3.
10945	*
10946	* Warning! This bugger disables interrupts on VINF_SUCCESS!
10947	*/
10948	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, false /* fStepping */);
10949	if (rcStrict != VINF_SUCCESS)
10950	break;
10951
10952	/* Interrupts are disabled at this point! */
10953	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
10954	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
10955	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
10956	/* Interrupts are re-enabled at this point! */
10957
10958	/*
10959	* Check for errors with running the VM (VMLAUNCH/VMRESUME).
10960	*/
10961	if (RT_SUCCESS(rcRun))
10962	{ /* very likely */ }
10963	else
10964	{
10965	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
10966	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
10967	return rcRun;
10968	}
10969
10970	/*
10971	* Profile the VM-exit.
10972	*/
10973	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
10974	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
10975	STAM_COUNTER_INC(&pVCpu->hm.s.StatNestedExitAll);
10976	STAM_COUNTER_INC(&pVCpu->hm.s.paStatNestedExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
10977	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
10978	HMVMX_START_EXIT_DISPATCH_PROF();
10979
10980	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
10981
10982	/*
10983	* Handle the VM-exit.
10984	*/
10985	rcStrict = hmR0VmxHandleExitNested(pVCpu, &VmxTransient);
10986	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
10987	if (rcStrict == VINF_SUCCESS)
10988	{
10989	if (!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
10990	{
10991	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
10992	rcStrict = VINF_VMX_VMEXIT;
10993	}
10994	else
10995	{
10996	if (++(*pcLoops) <= cMaxResumeLoops)
10997	continue;
10998	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
10999	rcStrict = VINF_EM_RAW_INTERRUPT;
11000	}
11001	}
11002	else
11003	Assert(rcStrict != VINF_VMX_VMEXIT);
11004	break;
11005	}
11006
11007	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
11008	return rcStrict;
11009	}
11010	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
11011
11012
11013	/** @name Execution loop for single stepping, DBGF events and expensive Dtrace
11014	* probes.
11015	*
11016	* The following few functions and associated structure contains the bloat
11017	* necessary for providing detailed debug events and dtrace probes as well as
11018	* reliable host side single stepping. This works on the principle of
11019	* "subclassing" the normal execution loop and workers. We replace the loop
11020	* method completely and override selected helpers to add necessary adjustments
11021	* to their core operation.
11022	*
11023	* The goal is to keep the "parent" code lean and mean, so as not to sacrifice
11024	* any performance for debug and analysis features.
11025	*
11026	* @{
11027	*/
11028
11029	/**
11030	* Transient per-VCPU debug state of VMCS and related info. we save/restore in
11031	* the debug run loop.
11032	*/
11033	typedef struct VMXRUNDBGSTATE
11034	{
11035	/** The RIP we started executing at. This is for detecting that we stepped. */
11036	uint64_t uRipStart;
11037	/** The CS we started executing with. */
11038	uint16_t uCsStart;
11039
11040	/** Whether we've actually modified the 1st execution control field. */
11041	bool fModifiedProcCtls : 1;
11042	/** Whether we've actually modified the 2nd execution control field. */
11043	bool fModifiedProcCtls2 : 1;
11044	/** Whether we've actually modified the exception bitmap. */
11045	bool fModifiedXcptBitmap : 1;
11046
11047	/** We desire the modified the CR0 mask to be cleared. */
11048	bool fClearCr0Mask : 1;
11049	/** We desire the modified the CR4 mask to be cleared. */
11050	bool fClearCr4Mask : 1;
11051	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC. */
11052	uint32_t fCpe1Extra;
11053	/** Stuff we do not want in VMX_VMCS32_CTRL_PROC_EXEC. */
11054	uint32_t fCpe1Unwanted;
11055	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC2. */
11056	uint32_t fCpe2Extra;
11057	/** Extra stuff we need in VMX_VMCS32_CTRL_EXCEPTION_BITMAP. */
11058	uint32_t bmXcptExtra;
11059	/** The sequence number of the Dtrace provider settings the state was
11060	* configured against. */
11061	uint32_t uDtraceSettingsSeqNo;
11062	/** VM-exits to check (one bit per VM-exit). */
11063	uint32_t bmExitsToCheck[3];
11064
11065	/** The initial VMX_VMCS32_CTRL_PROC_EXEC value (helps with restore). */
11066	uint32_t fProcCtlsInitial;
11067	/** The initial VMX_VMCS32_CTRL_PROC_EXEC2 value (helps with restore). */
11068	uint32_t fProcCtls2Initial;
11069	/** The initial VMX_VMCS32_CTRL_EXCEPTION_BITMAP value (helps with restore). */
11070	uint32_t bmXcptInitial;
11071	} VMXRUNDBGSTATE;
11072	AssertCompileMemberSize(VMXRUNDBGSTATE, bmExitsToCheck, (VMX_EXIT_MAX + 1 + 31) / 32 * 4);
11073	typedef VMXRUNDBGSTATE *PVMXRUNDBGSTATE;
11074
11075
11076	/**
11077	* Initializes the VMXRUNDBGSTATE structure.
11078	*
11079	* @param pVCpu The cross context virtual CPU structure of the
11080	* calling EMT.
11081	* @param pVmxTransient The VMX-transient structure.
11082	* @param pDbgState The debug state to initialize.
11083	*/
11084	static void hmR0VmxRunDebugStateInit(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11085	{
11086	pDbgState->uRipStart = pVCpu->cpum.GstCtx.rip;
11087	pDbgState->uCsStart = pVCpu->cpum.GstCtx.cs.Sel;
11088
11089	pDbgState->fModifiedProcCtls = false;
11090	pDbgState->fModifiedProcCtls2 = false;
11091	pDbgState->fModifiedXcptBitmap = false;
11092	pDbgState->fClearCr0Mask = false;
11093	pDbgState->fClearCr4Mask = false;
11094	pDbgState->fCpe1Extra = 0;
11095	pDbgState->fCpe1Unwanted = 0;
11096	pDbgState->fCpe2Extra = 0;
11097	pDbgState->bmXcptExtra = 0;
11098	pDbgState->fProcCtlsInitial = pVmxTransient->pVmcsInfo->u32ProcCtls;
11099	pDbgState->fProcCtls2Initial = pVmxTransient->pVmcsInfo->u32ProcCtls2;
11100	pDbgState->bmXcptInitial = pVmxTransient->pVmcsInfo->u32XcptBitmap;
11101	}
11102
11103
11104	/**
11105	* Updates the VMSC fields with changes requested by @a pDbgState.
11106	*
11107	* This is performed after hmR0VmxPreRunGuestDebugStateUpdate as well
11108	* immediately before executing guest code, i.e. when interrupts are disabled.
11109	* We don't check status codes here as we cannot easily assert or return in the
11110	* latter case.
11111	*
11112	* @param pVCpu The cross context virtual CPU structure.
11113	* @param pVmxTransient The VMX-transient structure.
11114	* @param pDbgState The debug state.
11115	*/
11116	static void hmR0VmxPreRunGuestDebugStateApply(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11117	{
11118	/*
11119	* Ensure desired flags in VMCS control fields are set.
11120	* (Ignoring write failure here, as we're committed and it's just debug extras.)
11121	*
11122	* Note! We load the shadow CR0 & CR4 bits when we flag the clearing, so
11123	* there should be no stale data in pCtx at this point.
11124	*/
11125	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11126	if ( (pVmcsInfo->u32ProcCtls & pDbgState->fCpe1Extra) != pDbgState->fCpe1Extra
11127	\|\| (pVmcsInfo->u32ProcCtls & pDbgState->fCpe1Unwanted))
11128	{
11129	pVmcsInfo->u32ProcCtls \|= pDbgState->fCpe1Extra;
11130	pVmcsInfo->u32ProcCtls &= ~pDbgState->fCpe1Unwanted;
11131	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
11132	Log6Func(("VMX_VMCS32_CTRL_PROC_EXEC: %#RX32\n", pVmcsInfo->u32ProcCtls));
11133	pDbgState->fModifiedProcCtls = true;
11134	}
11135
11136	if ((pVmcsInfo->u32ProcCtls2 & pDbgState->fCpe2Extra) != pDbgState->fCpe2Extra)
11137	{
11138	pVmcsInfo->u32ProcCtls2 \|= pDbgState->fCpe2Extra;
11139	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pVmcsInfo->u32ProcCtls2);
11140	Log6Func(("VMX_VMCS32_CTRL_PROC_EXEC2: %#RX32\n", pVmcsInfo->u32ProcCtls2));
11141	pDbgState->fModifiedProcCtls2 = true;
11142	}
11143
11144	if ((pVmcsInfo->u32XcptBitmap & pDbgState->bmXcptExtra) != pDbgState->bmXcptExtra)
11145	{
11146	pVmcsInfo->u32XcptBitmap \|= pDbgState->bmXcptExtra;
11147	VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVmcsInfo->u32XcptBitmap);
11148	Log6Func(("VMX_VMCS32_CTRL_EXCEPTION_BITMAP: %#RX32\n", pVmcsInfo->u32XcptBitmap));
11149	pDbgState->fModifiedXcptBitmap = true;
11150	}
11151
11152	if (pDbgState->fClearCr0Mask && pVmcsInfo->u64Cr0Mask != 0)
11153	{
11154	pVmcsInfo->u64Cr0Mask = 0;
11155	VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, 0);
11156	Log6Func(("VMX_VMCS_CTRL_CR0_MASK: 0\n"));
11157	}
11158
11159	if (pDbgState->fClearCr4Mask && pVmcsInfo->u64Cr4Mask != 0)
11160	{
11161	pVmcsInfo->u64Cr4Mask = 0;
11162	VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, 0);
11163	Log6Func(("VMX_VMCS_CTRL_CR4_MASK: 0\n"));
11164	}
11165
11166	NOREF(pVCpu);
11167	}
11168
11169
11170	/**
11171	* Restores VMCS fields that were changed by hmR0VmxPreRunGuestDebugStateApply for
11172	* re-entry next time around.
11173	*
11174	* @returns Strict VBox status code (i.e. informational status codes too).
11175	* @param pVCpu The cross context virtual CPU structure.
11176	* @param pVmxTransient The VMX-transient structure.
11177	* @param pDbgState The debug state.
11178	* @param rcStrict The return code from executing the guest using single
11179	* stepping.
11180	*/
11181	static VBOXSTRICTRC hmR0VmxRunDebugStateRevert(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState,
11182	VBOXSTRICTRC rcStrict)
11183	{
11184	/*
11185	* Restore VM-exit control settings as we may not reenter this function the
11186	* next time around.
11187	*/
11188	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11189
11190	/* We reload the initial value, trigger what we can of recalculations the
11191	next time around. From the looks of things, that's all that's required atm. */
11192	if (pDbgState->fModifiedProcCtls)
11193	{
11194	if (!(pDbgState->fProcCtlsInitial & VMX_PROC_CTLS_MOV_DR_EXIT) && CPUMIsHyperDebugStateActive(pVCpu))
11195	pDbgState->fProcCtlsInitial \|= VMX_PROC_CTLS_MOV_DR_EXIT; /* Avoid assertion in hmR0VmxLeave */
11196	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pDbgState->fProcCtlsInitial);
11197	AssertRC(rc2);
11198	pVmcsInfo->u32ProcCtls = pDbgState->fProcCtlsInitial;
11199	}
11200
11201	/* We're currently the only ones messing with this one, so just restore the
11202	cached value and reload the field. */
11203	if ( pDbgState->fModifiedProcCtls2
11204	&& pVmcsInfo->u32ProcCtls2 != pDbgState->fProcCtls2Initial)
11205	{
11206	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pDbgState->fProcCtls2Initial);
11207	AssertRC(rc2);
11208	pVmcsInfo->u32ProcCtls2 = pDbgState->fProcCtls2Initial;
11209	}
11210
11211	/* If we've modified the exception bitmap, we restore it and trigger
11212	reloading and partial recalculation the next time around. */
11213	if (pDbgState->fModifiedXcptBitmap)
11214	pVmcsInfo->u32XcptBitmap = pDbgState->bmXcptInitial;
11215
11216	return rcStrict;
11217	}
11218
11219
11220	/**
11221	* Configures VM-exit controls for current DBGF and DTrace settings.
11222	*
11223	* This updates @a pDbgState and the VMCS execution control fields to reflect
11224	* the necessary VM-exits demanded by DBGF and DTrace.
11225	*
11226	* @param pVCpu The cross context virtual CPU structure.
11227	* @param pVmxTransient The VMX-transient structure. May update
11228	* fUpdatedTscOffsettingAndPreemptTimer.
11229	* @param pDbgState The debug state.
11230	*/
11231	static void hmR0VmxPreRunGuestDebugStateUpdate(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11232	{
11233	/*
11234	* Take down the dtrace serial number so we can spot changes.
11235	*/
11236	pDbgState->uDtraceSettingsSeqNo = VBOXVMM_GET_SETTINGS_SEQ_NO();
11237	ASMCompilerBarrier();
11238
11239	/*
11240	* We'll rebuild most of the middle block of data members (holding the
11241	* current settings) as we go along here, so start by clearing it all.
11242	*/
11243	pDbgState->bmXcptExtra = 0;
11244	pDbgState->fCpe1Extra = 0;
11245	pDbgState->fCpe1Unwanted = 0;
11246	pDbgState->fCpe2Extra = 0;
11247	for (unsigned i = 0; i < RT_ELEMENTS(pDbgState->bmExitsToCheck); i++)
11248	pDbgState->bmExitsToCheck[i] = 0;
11249
11250	/*
11251	* Software interrupts (INT XXh) - no idea how to trigger these...
11252	*/
11253	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
11254	if ( DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_INTERRUPT_SOFTWARE)
11255	\|\| VBOXVMM_INT_SOFTWARE_ENABLED())
11256	{
11257	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
11258	}
11259
11260	/*
11261	* INT3 breakpoints - triggered by #BP exceptions.
11262	*/
11263	if (pVM->dbgf.ro.cEnabledInt3Breakpoints > 0)
11264	pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
11265
11266	/*
11267	* Exception bitmap and XCPT events+probes.
11268	*/
11269	for (int iXcpt = 0; iXcpt < (DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST + 1); iXcpt++)
11270	if (DBGF_IS_EVENT_ENABLED(pVM, (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + iXcpt)))
11271	pDbgState->bmXcptExtra \|= RT_BIT_32(iXcpt);
11272
11273	if (VBOXVMM_XCPT_DE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DE);
11274	if (VBOXVMM_XCPT_DB_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DB);
11275	if (VBOXVMM_XCPT_BP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
11276	if (VBOXVMM_XCPT_OF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_OF);
11277	if (VBOXVMM_XCPT_BR_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BR);
11278	if (VBOXVMM_XCPT_UD_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_UD);
11279	if (VBOXVMM_XCPT_NM_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NM);
11280	if (VBOXVMM_XCPT_DF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DF);
11281	if (VBOXVMM_XCPT_TS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_TS);
11282	if (VBOXVMM_XCPT_NP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NP);
11283	if (VBOXVMM_XCPT_SS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SS);
11284	if (VBOXVMM_XCPT_GP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_GP);
11285	if (VBOXVMM_XCPT_PF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_PF);
11286	if (VBOXVMM_XCPT_MF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_MF);
11287	if (VBOXVMM_XCPT_AC_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_AC);
11288	if (VBOXVMM_XCPT_XF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_XF);
11289	if (VBOXVMM_XCPT_VE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_VE);
11290	if (VBOXVMM_XCPT_SX_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SX);
11291
11292	if (pDbgState->bmXcptExtra)
11293	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
11294
11295	/*
11296	* Process events and probes for VM-exits, making sure we get the wanted VM-exits.
11297	*
11298	* Note! This is the reverse of what hmR0VmxHandleExitDtraceEvents does.
11299	* So, when adding/changing/removing please don't forget to update it.
11300	*
11301	* Some of the macros are picking up local variables to save horizontal space,
11302	* (being able to see it in a table is the lesser evil here).
11303	*/
11304	#define IS_EITHER_ENABLED(a_pVM, a_EventSubName) \
11305	( DBGF_IS_EVENT_ENABLED(a_pVM, RT_CONCAT(DBGFEVENT_, a_EventSubName)) \
11306	\|\| RT_CONCAT3(VBOXVMM_, a_EventSubName, _ENABLED)() )
11307	#define SET_ONLY_XBM_IF_EITHER_EN(a_EventSubName, a_uExit) \
11308	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11309	{ AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11310	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11311	} else do { } while (0)
11312	#define SET_CPE1_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec) \
11313	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11314	{ \
11315	(pDbgState)->fCpe1Extra \|= (a_fCtrlProcExec); \
11316	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11317	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11318	} else do { } while (0)
11319	#define SET_CPEU_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fUnwantedCtrlProcExec) \
11320	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11321	{ \
11322	(pDbgState)->fCpe1Unwanted \|= (a_fUnwantedCtrlProcExec); \
11323	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11324	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11325	} else do { } while (0)
11326	#define SET_CPE2_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec2) \
11327	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11328	{ \
11329	(pDbgState)->fCpe2Extra \|= (a_fCtrlProcExec2); \
11330	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11331	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11332	} else do { } while (0)
11333
11334	SET_ONLY_XBM_IF_EITHER_EN(EXIT_TASK_SWITCH, VMX_EXIT_TASK_SWITCH); /* unconditional */
11335	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_VIOLATION, VMX_EXIT_EPT_VIOLATION); /* unconditional */
11336	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_MISCONFIG, VMX_EXIT_EPT_MISCONFIG); /* unconditional (unless #VE) */
11337	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_ACCESS, VMX_EXIT_APIC_ACCESS); /* feature dependent, nothing to enable here */
11338	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_WRITE, VMX_EXIT_APIC_WRITE); /* feature dependent, nothing to enable here */
11339
11340	SET_ONLY_XBM_IF_EITHER_EN(INSTR_CPUID, VMX_EXIT_CPUID); /* unconditional */
11341	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CPUID, VMX_EXIT_CPUID);
11342	SET_ONLY_XBM_IF_EITHER_EN(INSTR_GETSEC, VMX_EXIT_GETSEC); /* unconditional */
11343	SET_ONLY_XBM_IF_EITHER_EN( EXIT_GETSEC, VMX_EXIT_GETSEC);
11344	SET_CPE1_XBM_IF_EITHER_EN(INSTR_HALT, VMX_EXIT_HLT, VMX_PROC_CTLS_HLT_EXIT); /* paranoia */
11345	SET_ONLY_XBM_IF_EITHER_EN( EXIT_HALT, VMX_EXIT_HLT);
11346	SET_ONLY_XBM_IF_EITHER_EN(INSTR_INVD, VMX_EXIT_INVD); /* unconditional */
11347	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVD, VMX_EXIT_INVD);
11348	SET_CPE1_XBM_IF_EITHER_EN(INSTR_INVLPG, VMX_EXIT_INVLPG, VMX_PROC_CTLS_INVLPG_EXIT);
11349	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVLPG, VMX_EXIT_INVLPG);
11350	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDPMC, VMX_EXIT_RDPMC, VMX_PROC_CTLS_RDPMC_EXIT);
11351	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDPMC, VMX_EXIT_RDPMC);
11352	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSC, VMX_EXIT_RDTSC, VMX_PROC_CTLS_RDTSC_EXIT);
11353	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSC, VMX_EXIT_RDTSC);
11354	SET_ONLY_XBM_IF_EITHER_EN(INSTR_RSM, VMX_EXIT_RSM); /* unconditional */
11355	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RSM, VMX_EXIT_RSM);
11356	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMM_CALL, VMX_EXIT_VMCALL); /* unconditional */
11357	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMM_CALL, VMX_EXIT_VMCALL);
11358	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMCLEAR, VMX_EXIT_VMCLEAR); /* unconditional */
11359	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMCLEAR, VMX_EXIT_VMCLEAR);
11360	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH); /* unconditional */
11361	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH);
11362	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRLD, VMX_EXIT_VMPTRLD); /* unconditional */
11363	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRLD, VMX_EXIT_VMPTRLD);
11364	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRST, VMX_EXIT_VMPTRST); /* unconditional */
11365	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRST, VMX_EXIT_VMPTRST);
11366	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMREAD, VMX_EXIT_VMREAD); /* unconditional */
11367	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMREAD, VMX_EXIT_VMREAD);
11368	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMRESUME, VMX_EXIT_VMRESUME); /* unconditional */
11369	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMRESUME, VMX_EXIT_VMRESUME);
11370	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMWRITE, VMX_EXIT_VMWRITE); /* unconditional */
11371	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMWRITE, VMX_EXIT_VMWRITE);
11372	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXOFF, VMX_EXIT_VMXOFF); /* unconditional */
11373	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXOFF, VMX_EXIT_VMXOFF);
11374	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXON, VMX_EXIT_VMXON); /* unconditional */
11375	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXON, VMX_EXIT_VMXON);
11376
11377	if ( IS_EITHER_ENABLED(pVM, INSTR_CRX_READ)
11378	\|\| IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
11379	{
11380	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_CR4
11381	\| CPUMCTX_EXTRN_APIC_TPR);
11382	AssertRC(rc);
11383
11384	#if 0 /** @todo fix me */
11385	pDbgState->fClearCr0Mask = true;
11386	pDbgState->fClearCr4Mask = true;
11387	#endif
11388	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_READ))
11389	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_CR3_STORE_EXIT \| VMX_PROC_CTLS_CR8_STORE_EXIT;
11390	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
11391	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_CR3_LOAD_EXIT \| VMX_PROC_CTLS_CR8_LOAD_EXIT;
11392	pDbgState->fCpe1Unwanted \|= VMX_PROC_CTLS_USE_TPR_SHADOW; /* risky? */
11393	/* Note! We currently don't use VMX_VMCS32_CTRL_CR3_TARGET_COUNT. It would
11394	require clearing here and in the loop if we start using it. */
11395	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_CRX);
11396	}
11397	else
11398	{
11399	if (pDbgState->fClearCr0Mask)
11400	{
11401	pDbgState->fClearCr0Mask = false;
11402	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR0);
11403	}
11404	if (pDbgState->fClearCr4Mask)
11405	{
11406	pDbgState->fClearCr4Mask = false;
11407	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR4);
11408	}
11409	}
11410	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_READ, VMX_EXIT_MOV_CRX);
11411	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_WRITE, VMX_EXIT_MOV_CRX);
11412
11413	if ( IS_EITHER_ENABLED(pVM, INSTR_DRX_READ)
11414	\|\| IS_EITHER_ENABLED(pVM, INSTR_DRX_WRITE))
11415	{
11416	/** @todo later, need to fix handler as it assumes this won't usually happen. */
11417	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_DRX);
11418	}
11419	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_READ, VMX_EXIT_MOV_DRX);
11420	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_WRITE, VMX_EXIT_MOV_DRX);
11421
11422	SET_CPEU_XBM_IF_EITHER_EN(INSTR_RDMSR, VMX_EXIT_RDMSR, VMX_PROC_CTLS_USE_MSR_BITMAPS); /* risky clearing this? */
11423	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDMSR, VMX_EXIT_RDMSR);
11424	SET_CPEU_XBM_IF_EITHER_EN(INSTR_WRMSR, VMX_EXIT_WRMSR, VMX_PROC_CTLS_USE_MSR_BITMAPS);
11425	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WRMSR, VMX_EXIT_WRMSR);
11426	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MWAIT, VMX_EXIT_MWAIT, VMX_PROC_CTLS_MWAIT_EXIT); /* paranoia */
11427	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MWAIT, VMX_EXIT_MWAIT);
11428	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MONITOR, VMX_EXIT_MONITOR, VMX_PROC_CTLS_MONITOR_EXIT); /* paranoia */
11429	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MONITOR, VMX_EXIT_MONITOR);
11430	#if 0 /** @todo too slow, fix handler. */
11431	SET_CPE1_XBM_IF_EITHER_EN(INSTR_PAUSE, VMX_EXIT_PAUSE, VMX_PROC_CTLS_PAUSE_EXIT);
11432	#endif
11433	SET_ONLY_XBM_IF_EITHER_EN( EXIT_PAUSE, VMX_EXIT_PAUSE);
11434
11435	if ( IS_EITHER_ENABLED(pVM, INSTR_SGDT)
11436	\|\| IS_EITHER_ENABLED(pVM, INSTR_SIDT)
11437	\|\| IS_EITHER_ENABLED(pVM, INSTR_LGDT)
11438	\|\| IS_EITHER_ENABLED(pVM, INSTR_LIDT))
11439	{
11440	pDbgState->fCpe2Extra \|= VMX_PROC_CTLS2_DESC_TABLE_EXIT;
11441	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_GDTR_IDTR_ACCESS);
11442	}
11443	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SGDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11444	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SIDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11445	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LGDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11446	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LIDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11447
11448	if ( IS_EITHER_ENABLED(pVM, INSTR_SLDT)
11449	\|\| IS_EITHER_ENABLED(pVM, INSTR_STR)
11450	\|\| IS_EITHER_ENABLED(pVM, INSTR_LLDT)
11451	\|\| IS_EITHER_ENABLED(pVM, INSTR_LTR))
11452	{
11453	pDbgState->fCpe2Extra \|= VMX_PROC_CTLS2_DESC_TABLE_EXIT;
11454	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_LDTR_TR_ACCESS);
11455	}
11456	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SLDT, VMX_EXIT_LDTR_TR_ACCESS);
11457	SET_ONLY_XBM_IF_EITHER_EN( EXIT_STR, VMX_EXIT_LDTR_TR_ACCESS);
11458	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LLDT, VMX_EXIT_LDTR_TR_ACCESS);
11459	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LTR, VMX_EXIT_LDTR_TR_ACCESS);
11460
11461	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVEPT, VMX_EXIT_INVEPT); /* unconditional */
11462	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVEPT, VMX_EXIT_INVEPT);
11463	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSCP, VMX_EXIT_RDTSCP, VMX_PROC_CTLS_RDTSC_EXIT);
11464	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSCP, VMX_EXIT_RDTSCP);
11465	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVVPID, VMX_EXIT_INVVPID); /* unconditional */
11466	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVVPID, VMX_EXIT_INVVPID);
11467	SET_CPE2_XBM_IF_EITHER_EN(INSTR_WBINVD, VMX_EXIT_WBINVD, VMX_PROC_CTLS2_WBINVD_EXIT);
11468	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WBINVD, VMX_EXIT_WBINVD);
11469	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSETBV, VMX_EXIT_XSETBV); /* unconditional */
11470	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XSETBV, VMX_EXIT_XSETBV);
11471	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDRAND, VMX_EXIT_RDRAND, VMX_PROC_CTLS2_RDRAND_EXIT);
11472	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDRAND, VMX_EXIT_RDRAND);
11473	SET_CPE1_XBM_IF_EITHER_EN(INSTR_VMX_INVPCID, VMX_EXIT_INVPCID, VMX_PROC_CTLS_INVLPG_EXIT);
11474	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVPCID, VMX_EXIT_INVPCID);
11475	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMFUNC, VMX_EXIT_VMFUNC); /* unconditional for the current setup */
11476	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMFUNC, VMX_EXIT_VMFUNC);
11477	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDSEED, VMX_EXIT_RDSEED, VMX_PROC_CTLS2_RDSEED_EXIT);
11478	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDSEED, VMX_EXIT_RDSEED);
11479	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSAVES, VMX_EXIT_XSAVES); /* unconditional (enabled by host, guest cfg) */
11480	SET_ONLY_XBM_IF_EITHER_EN(EXIT_XSAVES, VMX_EXIT_XSAVES);
11481	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XRSTORS, VMX_EXIT_XRSTORS); /* unconditional (enabled by host, guest cfg) */
11482	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XRSTORS, VMX_EXIT_XRSTORS);
11483
11484	#undef IS_EITHER_ENABLED
11485	#undef SET_ONLY_XBM_IF_EITHER_EN
11486	#undef SET_CPE1_XBM_IF_EITHER_EN
11487	#undef SET_CPEU_XBM_IF_EITHER_EN
11488	#undef SET_CPE2_XBM_IF_EITHER_EN
11489
11490	/*
11491	* Sanitize the control stuff.
11492	*/
11493	pDbgState->fCpe2Extra &= pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1;
11494	if (pDbgState->fCpe2Extra)
11495	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_USE_SECONDARY_CTLS;
11496	pDbgState->fCpe1Extra &= pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1;
11497	pDbgState->fCpe1Unwanted &= ~pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed0;
11498	if (pVCpu->hm.s.fDebugWantRdTscExit != RT_BOOL(pDbgState->fCpe1Extra & VMX_PROC_CTLS_RDTSC_EXIT))
11499	{
11500	pVCpu->hm.s.fDebugWantRdTscExit ^= true;
11501	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
11502	}
11503
11504	Log6(("HM: debug state: cpe1=%#RX32 cpeu=%#RX32 cpe2=%#RX32%s%s\n",
11505	pDbgState->fCpe1Extra, pDbgState->fCpe1Unwanted, pDbgState->fCpe2Extra,
11506	pDbgState->fClearCr0Mask ? " clr-cr0" : "",
11507	pDbgState->fClearCr4Mask ? " clr-cr4" : ""));
11508	}
11509
11510
11511	/**
11512	* Fires off DBGF events and dtrace probes for a VM-exit, when it's
11513	* appropriate.
11514	*
11515	* The caller has checked the VM-exit against the
11516	* VMXRUNDBGSTATE::bmExitsToCheck bitmap. The caller has checked for NMIs
11517	* already, so we don't have to do that either.
11518	*
11519	* @returns Strict VBox status code (i.e. informational status codes too).
11520	* @param pVCpu The cross context virtual CPU structure.
11521	* @param pVmxTransient The VMX-transient structure.
11522	* @param uExitReason The VM-exit reason.
11523	*
11524	* @remarks The name of this function is displayed by dtrace, so keep it short
11525	* and to the point. No longer than 33 chars long, please.
11526	*/
11527	static VBOXSTRICTRC hmR0VmxHandleExitDtraceEvents(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uExitReason)
11528	{
11529	/*
11530	* Translate the event into a DBGF event (enmEvent + uEventArg) and at the
11531	* same time check whether any corresponding Dtrace event is enabled (fDtrace).
11532	*
11533	* Note! This is the reverse operation of what hmR0VmxPreRunGuestDebugStateUpdate
11534	* does. Must add/change/remove both places. Same ordering, please.
11535	*
11536	* Added/removed events must also be reflected in the next section
11537	* where we dispatch dtrace events.
11538	*/
11539	bool fDtrace1 = false;
11540	bool fDtrace2 = false;
11541	DBGFEVENTTYPE enmEvent1 = DBGFEVENT_END;
11542	DBGFEVENTTYPE enmEvent2 = DBGFEVENT_END;
11543	uint32_t uEventArg = 0;
11544	#define SET_EXIT(a_EventSubName) \
11545	do { \
11546	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
11547	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
11548	} while (0)
11549	#define SET_BOTH(a_EventSubName) \
11550	do { \
11551	enmEvent1 = RT_CONCAT(DBGFEVENT_INSTR_, a_EventSubName); \
11552	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
11553	fDtrace1 = RT_CONCAT3(VBOXVMM_INSTR_, a_EventSubName, _ENABLED)(); \
11554	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
11555	} while (0)
11556	switch (uExitReason)
11557	{
11558	case VMX_EXIT_MTF:
11559	return hmR0VmxExitMtf(pVCpu, pVmxTransient);
11560
11561	case VMX_EXIT_XCPT_OR_NMI:
11562	{
11563	uint8_t const idxVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
11564	switch (VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo))
11565	{
11566	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
11567	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
11568	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
11569	if (idxVector <= (unsigned)(DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST))
11570	{
11571	if (VMX_EXIT_INT_INFO_IS_ERROR_CODE_VALID(pVmxTransient->uExitIntInfo))
11572	{
11573	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
11574	uEventArg = pVmxTransient->uExitIntErrorCode;
11575	}
11576	enmEvent1 = (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + idxVector);
11577	switch (enmEvent1)
11578	{
11579	case DBGFEVENT_XCPT_DE: fDtrace1 = VBOXVMM_XCPT_DE_ENABLED(); break;
11580	case DBGFEVENT_XCPT_DB: fDtrace1 = VBOXVMM_XCPT_DB_ENABLED(); break;
11581	case DBGFEVENT_XCPT_BP: fDtrace1 = VBOXVMM_XCPT_BP_ENABLED(); break;
11582	case DBGFEVENT_XCPT_OF: fDtrace1 = VBOXVMM_XCPT_OF_ENABLED(); break;
11583	case DBGFEVENT_XCPT_BR: fDtrace1 = VBOXVMM_XCPT_BR_ENABLED(); break;
11584	case DBGFEVENT_XCPT_UD: fDtrace1 = VBOXVMM_XCPT_UD_ENABLED(); break;
11585	case DBGFEVENT_XCPT_NM: fDtrace1 = VBOXVMM_XCPT_NM_ENABLED(); break;
11586	case DBGFEVENT_XCPT_DF: fDtrace1 = VBOXVMM_XCPT_DF_ENABLED(); break;
11587	case DBGFEVENT_XCPT_TS: fDtrace1 = VBOXVMM_XCPT_TS_ENABLED(); break;
11588	case DBGFEVENT_XCPT_NP: fDtrace1 = VBOXVMM_XCPT_NP_ENABLED(); break;
11589	case DBGFEVENT_XCPT_SS: fDtrace1 = VBOXVMM_XCPT_SS_ENABLED(); break;
11590	case DBGFEVENT_XCPT_GP: fDtrace1 = VBOXVMM_XCPT_GP_ENABLED(); break;
11591	case DBGFEVENT_XCPT_PF: fDtrace1 = VBOXVMM_XCPT_PF_ENABLED(); break;
11592	case DBGFEVENT_XCPT_MF: fDtrace1 = VBOXVMM_XCPT_MF_ENABLED(); break;
11593	case DBGFEVENT_XCPT_AC: fDtrace1 = VBOXVMM_XCPT_AC_ENABLED(); break;
11594	case DBGFEVENT_XCPT_XF: fDtrace1 = VBOXVMM_XCPT_XF_ENABLED(); break;
11595	case DBGFEVENT_XCPT_VE: fDtrace1 = VBOXVMM_XCPT_VE_ENABLED(); break;
11596	case DBGFEVENT_XCPT_SX: fDtrace1 = VBOXVMM_XCPT_SX_ENABLED(); break;
11597	default: break;
11598	}
11599	}
11600	else
11601	AssertFailed();
11602	break;
11603
11604	case VMX_EXIT_INT_INFO_TYPE_SW_INT:
11605	uEventArg = idxVector;
11606	enmEvent1 = DBGFEVENT_INTERRUPT_SOFTWARE;
11607	fDtrace1 = VBOXVMM_INT_SOFTWARE_ENABLED();
11608	break;
11609	}
11610	break;
11611	}
11612
11613	case VMX_EXIT_TRIPLE_FAULT:
11614	enmEvent1 = DBGFEVENT_TRIPLE_FAULT;
11615	//fDtrace1 = VBOXVMM_EXIT_TRIPLE_FAULT_ENABLED();
11616	break;
11617	case VMX_EXIT_TASK_SWITCH: SET_EXIT(TASK_SWITCH); break;
11618	case VMX_EXIT_EPT_VIOLATION: SET_EXIT(VMX_EPT_VIOLATION); break;
11619	case VMX_EXIT_EPT_MISCONFIG: SET_EXIT(VMX_EPT_MISCONFIG); break;
11620	case VMX_EXIT_APIC_ACCESS: SET_EXIT(VMX_VAPIC_ACCESS); break;
11621	case VMX_EXIT_APIC_WRITE: SET_EXIT(VMX_VAPIC_WRITE); break;
11622
11623	/* Instruction specific VM-exits: */
11624	case VMX_EXIT_CPUID: SET_BOTH(CPUID); break;
11625	case VMX_EXIT_GETSEC: SET_BOTH(GETSEC); break;
11626	case VMX_EXIT_HLT: SET_BOTH(HALT); break;
11627	case VMX_EXIT_INVD: SET_BOTH(INVD); break;
11628	case VMX_EXIT_INVLPG: SET_BOTH(INVLPG); break;
11629	case VMX_EXIT_RDPMC: SET_BOTH(RDPMC); break;
11630	case VMX_EXIT_RDTSC: SET_BOTH(RDTSC); break;
11631	case VMX_EXIT_RSM: SET_BOTH(RSM); break;
11632	case VMX_EXIT_VMCALL: SET_BOTH(VMM_CALL); break;
11633	case VMX_EXIT_VMCLEAR: SET_BOTH(VMX_VMCLEAR); break;
11634	case VMX_EXIT_VMLAUNCH: SET_BOTH(VMX_VMLAUNCH); break;
11635	case VMX_EXIT_VMPTRLD: SET_BOTH(VMX_VMPTRLD); break;
11636	case VMX_EXIT_VMPTRST: SET_BOTH(VMX_VMPTRST); break;
11637	case VMX_EXIT_VMREAD: SET_BOTH(VMX_VMREAD); break;
11638	case VMX_EXIT_VMRESUME: SET_BOTH(VMX_VMRESUME); break;
11639	case VMX_EXIT_VMWRITE: SET_BOTH(VMX_VMWRITE); break;
11640	case VMX_EXIT_VMXOFF: SET_BOTH(VMX_VMXOFF); break;
11641	case VMX_EXIT_VMXON: SET_BOTH(VMX_VMXON); break;
11642	case VMX_EXIT_MOV_CRX:
11643	hmR0VmxReadExitQualVmcs(pVmxTransient);
11644	if (VMX_EXIT_QUAL_CRX_ACCESS(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_CRX_ACCESS_READ)
11645	SET_BOTH(CRX_READ);
11646	else
11647	SET_BOTH(CRX_WRITE);
11648	uEventArg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
11649	break;
11650	case VMX_EXIT_MOV_DRX:
11651	hmR0VmxReadExitQualVmcs(pVmxTransient);
11652	if ( VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual)
11653	== VMX_EXIT_QUAL_DRX_DIRECTION_READ)
11654	SET_BOTH(DRX_READ);
11655	else
11656	SET_BOTH(DRX_WRITE);
11657	uEventArg = VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual);
11658	break;
11659	case VMX_EXIT_RDMSR: SET_BOTH(RDMSR); break;
11660	case VMX_EXIT_WRMSR: SET_BOTH(WRMSR); break;
11661	case VMX_EXIT_MWAIT: SET_BOTH(MWAIT); break;
11662	case VMX_EXIT_MONITOR: SET_BOTH(MONITOR); break;
11663	case VMX_EXIT_PAUSE: SET_BOTH(PAUSE); break;
11664	case VMX_EXIT_GDTR_IDTR_ACCESS:
11665	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
11666	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_BF_XDTR_INSINFO_INSTR_ID))
11667	{
11668	case VMX_XDTR_INSINFO_II_SGDT: SET_BOTH(SGDT); break;
11669	case VMX_XDTR_INSINFO_II_SIDT: SET_BOTH(SIDT); break;
11670	case VMX_XDTR_INSINFO_II_LGDT: SET_BOTH(LGDT); break;
11671	case VMX_XDTR_INSINFO_II_LIDT: SET_BOTH(LIDT); break;
11672	}
11673	break;
11674
11675	case VMX_EXIT_LDTR_TR_ACCESS:
11676	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
11677	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_BF_YYTR_INSINFO_INSTR_ID))
11678	{
11679	case VMX_YYTR_INSINFO_II_SLDT: SET_BOTH(SLDT); break;
11680	case VMX_YYTR_INSINFO_II_STR: SET_BOTH(STR); break;
11681	case VMX_YYTR_INSINFO_II_LLDT: SET_BOTH(LLDT); break;
11682	case VMX_YYTR_INSINFO_II_LTR: SET_BOTH(LTR); break;
11683	}
11684	break;
11685
11686	case VMX_EXIT_INVEPT: SET_BOTH(VMX_INVEPT); break;
11687	case VMX_EXIT_RDTSCP: SET_BOTH(RDTSCP); break;
11688	case VMX_EXIT_INVVPID: SET_BOTH(VMX_INVVPID); break;
11689	case VMX_EXIT_WBINVD: SET_BOTH(WBINVD); break;
11690	case VMX_EXIT_XSETBV: SET_BOTH(XSETBV); break;
11691	case VMX_EXIT_RDRAND: SET_BOTH(RDRAND); break;
11692	case VMX_EXIT_INVPCID: SET_BOTH(VMX_INVPCID); break;
11693	case VMX_EXIT_VMFUNC: SET_BOTH(VMX_VMFUNC); break;
11694	case VMX_EXIT_RDSEED: SET_BOTH(RDSEED); break;
11695	case VMX_EXIT_XSAVES: SET_BOTH(XSAVES); break;
11696	case VMX_EXIT_XRSTORS: SET_BOTH(XRSTORS); break;
11697
11698	/* Events that aren't relevant at this point. */
11699	case VMX_EXIT_EXT_INT:
11700	case VMX_EXIT_INT_WINDOW:
11701	case VMX_EXIT_NMI_WINDOW:
11702	case VMX_EXIT_TPR_BELOW_THRESHOLD:
11703	case VMX_EXIT_PREEMPT_TIMER:
11704	case VMX_EXIT_IO_INSTR:
11705	break;
11706
11707	/* Errors and unexpected events. */
11708	case VMX_EXIT_INIT_SIGNAL:
11709	case VMX_EXIT_SIPI:
11710	case VMX_EXIT_IO_SMI:
11711	case VMX_EXIT_SMI:
11712	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
11713	case VMX_EXIT_ERR_MSR_LOAD:
11714	case VMX_EXIT_ERR_MACHINE_CHECK:
11715	case VMX_EXIT_PML_FULL:
11716	case VMX_EXIT_VIRTUALIZED_EOI:
11717	break;
11718
11719	default:
11720	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
11721	break;
11722	}
11723	#undef SET_BOTH
11724	#undef SET_EXIT
11725
11726	/*
11727	* Dtrace tracepoints go first. We do them here at once so we don't
11728	* have to copy the guest state saving and stuff a few dozen times.
11729	* Down side is that we've got to repeat the switch, though this time
11730	* we use enmEvent since the probes are a subset of what DBGF does.
11731	*/
11732	if (fDtrace1 \|\| fDtrace2)
11733	{
11734	hmR0VmxReadExitQualVmcs(pVmxTransient);
11735	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
11736	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
11737	switch (enmEvent1)
11738	{
11739	/** @todo consider which extra parameters would be helpful for each probe. */
11740	case DBGFEVENT_END: break;
11741	case DBGFEVENT_XCPT_DE: VBOXVMM_XCPT_DE(pVCpu, pCtx); break;
11742	case DBGFEVENT_XCPT_DB: VBOXVMM_XCPT_DB(pVCpu, pCtx, pCtx->dr[6]); break;
11743	case DBGFEVENT_XCPT_BP: VBOXVMM_XCPT_BP(pVCpu, pCtx); break;
11744	case DBGFEVENT_XCPT_OF: VBOXVMM_XCPT_OF(pVCpu, pCtx); break;
11745	case DBGFEVENT_XCPT_BR: VBOXVMM_XCPT_BR(pVCpu, pCtx); break;
11746	case DBGFEVENT_XCPT_UD: VBOXVMM_XCPT_UD(pVCpu, pCtx); break;
11747	case DBGFEVENT_XCPT_NM: VBOXVMM_XCPT_NM(pVCpu, pCtx); break;
11748	case DBGFEVENT_XCPT_DF: VBOXVMM_XCPT_DF(pVCpu, pCtx); break;
11749	case DBGFEVENT_XCPT_TS: VBOXVMM_XCPT_TS(pVCpu, pCtx, uEventArg); break;
11750	case DBGFEVENT_XCPT_NP: VBOXVMM_XCPT_NP(pVCpu, pCtx, uEventArg); break;
11751	case DBGFEVENT_XCPT_SS: VBOXVMM_XCPT_SS(pVCpu, pCtx, uEventArg); break;
11752	case DBGFEVENT_XCPT_GP: VBOXVMM_XCPT_GP(pVCpu, pCtx, uEventArg); break;
11753	case DBGFEVENT_XCPT_PF: VBOXVMM_XCPT_PF(pVCpu, pCtx, uEventArg, pCtx->cr2); break;
11754	case DBGFEVENT_XCPT_MF: VBOXVMM_XCPT_MF(pVCpu, pCtx); break;
11755	case DBGFEVENT_XCPT_AC: VBOXVMM_XCPT_AC(pVCpu, pCtx); break;
11756	case DBGFEVENT_XCPT_XF: VBOXVMM_XCPT_XF(pVCpu, pCtx); break;
11757	case DBGFEVENT_XCPT_VE: VBOXVMM_XCPT_VE(pVCpu, pCtx); break;
11758	case DBGFEVENT_XCPT_SX: VBOXVMM_XCPT_SX(pVCpu, pCtx, uEventArg); break;
11759	case DBGFEVENT_INTERRUPT_SOFTWARE: VBOXVMM_INT_SOFTWARE(pVCpu, pCtx, (uint8_t)uEventArg); break;
11760	case DBGFEVENT_INSTR_CPUID: VBOXVMM_INSTR_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
11761	case DBGFEVENT_INSTR_GETSEC: VBOXVMM_INSTR_GETSEC(pVCpu, pCtx); break;
11762	case DBGFEVENT_INSTR_HALT: VBOXVMM_INSTR_HALT(pVCpu, pCtx); break;
11763	case DBGFEVENT_INSTR_INVD: VBOXVMM_INSTR_INVD(pVCpu, pCtx); break;
11764	case DBGFEVENT_INSTR_INVLPG: VBOXVMM_INSTR_INVLPG(pVCpu, pCtx); break;
11765	case DBGFEVENT_INSTR_RDPMC: VBOXVMM_INSTR_RDPMC(pVCpu, pCtx); break;
11766	case DBGFEVENT_INSTR_RDTSC: VBOXVMM_INSTR_RDTSC(pVCpu, pCtx); break;
11767	case DBGFEVENT_INSTR_RSM: VBOXVMM_INSTR_RSM(pVCpu, pCtx); break;
11768	case DBGFEVENT_INSTR_CRX_READ: VBOXVMM_INSTR_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
11769	case DBGFEVENT_INSTR_CRX_WRITE: VBOXVMM_INSTR_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
11770	case DBGFEVENT_INSTR_DRX_READ: VBOXVMM_INSTR_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
11771	case DBGFEVENT_INSTR_DRX_WRITE: VBOXVMM_INSTR_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
11772	case DBGFEVENT_INSTR_RDMSR: VBOXVMM_INSTR_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
11773	case DBGFEVENT_INSTR_WRMSR: VBOXVMM_INSTR_WRMSR(pVCpu, pCtx, pCtx->ecx,
11774	RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
11775	case DBGFEVENT_INSTR_MWAIT: VBOXVMM_INSTR_MWAIT(pVCpu, pCtx); break;
11776	case DBGFEVENT_INSTR_MONITOR: VBOXVMM_INSTR_MONITOR(pVCpu, pCtx); break;
11777	case DBGFEVENT_INSTR_PAUSE: VBOXVMM_INSTR_PAUSE(pVCpu, pCtx); break;
11778	case DBGFEVENT_INSTR_SGDT: VBOXVMM_INSTR_SGDT(pVCpu, pCtx); break;
11779	case DBGFEVENT_INSTR_SIDT: VBOXVMM_INSTR_SIDT(pVCpu, pCtx); break;
11780	case DBGFEVENT_INSTR_LGDT: VBOXVMM_INSTR_LGDT(pVCpu, pCtx); break;
11781	case DBGFEVENT_INSTR_LIDT: VBOXVMM_INSTR_LIDT(pVCpu, pCtx); break;
11782	case DBGFEVENT_INSTR_SLDT: VBOXVMM_INSTR_SLDT(pVCpu, pCtx); break;
11783	case DBGFEVENT_INSTR_STR: VBOXVMM_INSTR_STR(pVCpu, pCtx); break;
11784	case DBGFEVENT_INSTR_LLDT: VBOXVMM_INSTR_LLDT(pVCpu, pCtx); break;
11785	case DBGFEVENT_INSTR_LTR: VBOXVMM_INSTR_LTR(pVCpu, pCtx); break;
11786	case DBGFEVENT_INSTR_RDTSCP: VBOXVMM_INSTR_RDTSCP(pVCpu, pCtx); break;
11787	case DBGFEVENT_INSTR_WBINVD: VBOXVMM_INSTR_WBINVD(pVCpu, pCtx); break;
11788	case DBGFEVENT_INSTR_XSETBV: VBOXVMM_INSTR_XSETBV(pVCpu, pCtx); break;
11789	case DBGFEVENT_INSTR_RDRAND: VBOXVMM_INSTR_RDRAND(pVCpu, pCtx); break;
11790	case DBGFEVENT_INSTR_RDSEED: VBOXVMM_INSTR_RDSEED(pVCpu, pCtx); break;
11791	case DBGFEVENT_INSTR_XSAVES: VBOXVMM_INSTR_XSAVES(pVCpu, pCtx); break;
11792	case DBGFEVENT_INSTR_XRSTORS: VBOXVMM_INSTR_XRSTORS(pVCpu, pCtx); break;
11793	case DBGFEVENT_INSTR_VMM_CALL: VBOXVMM_INSTR_VMM_CALL(pVCpu, pCtx); break;
11794	case DBGFEVENT_INSTR_VMX_VMCLEAR: VBOXVMM_INSTR_VMX_VMCLEAR(pVCpu, pCtx); break;
11795	case DBGFEVENT_INSTR_VMX_VMLAUNCH: VBOXVMM_INSTR_VMX_VMLAUNCH(pVCpu, pCtx); break;
11796	case DBGFEVENT_INSTR_VMX_VMPTRLD: VBOXVMM_INSTR_VMX_VMPTRLD(pVCpu, pCtx); break;
11797	case DBGFEVENT_INSTR_VMX_VMPTRST: VBOXVMM_INSTR_VMX_VMPTRST(pVCpu, pCtx); break;
11798	case DBGFEVENT_INSTR_VMX_VMREAD: VBOXVMM_INSTR_VMX_VMREAD(pVCpu, pCtx); break;
11799	case DBGFEVENT_INSTR_VMX_VMRESUME: VBOXVMM_INSTR_VMX_VMRESUME(pVCpu, pCtx); break;
11800	case DBGFEVENT_INSTR_VMX_VMWRITE: VBOXVMM_INSTR_VMX_VMWRITE(pVCpu, pCtx); break;
11801	case DBGFEVENT_INSTR_VMX_VMXOFF: VBOXVMM_INSTR_VMX_VMXOFF(pVCpu, pCtx); break;
11802	case DBGFEVENT_INSTR_VMX_VMXON: VBOXVMM_INSTR_VMX_VMXON(pVCpu, pCtx); break;
11803	case DBGFEVENT_INSTR_VMX_INVEPT: VBOXVMM_INSTR_VMX_INVEPT(pVCpu, pCtx); break;
11804	case DBGFEVENT_INSTR_VMX_INVVPID: VBOXVMM_INSTR_VMX_INVVPID(pVCpu, pCtx); break;
11805	case DBGFEVENT_INSTR_VMX_INVPCID: VBOXVMM_INSTR_VMX_INVPCID(pVCpu, pCtx); break;
11806	case DBGFEVENT_INSTR_VMX_VMFUNC: VBOXVMM_INSTR_VMX_VMFUNC(pVCpu, pCtx); break;
11807	default: AssertMsgFailed(("enmEvent1=%d uExitReason=%d\n", enmEvent1, uExitReason)); break;
11808	}
11809	switch (enmEvent2)
11810	{
11811	/** @todo consider which extra parameters would be helpful for each probe. */
11812	case DBGFEVENT_END: break;
11813	case DBGFEVENT_EXIT_TASK_SWITCH: VBOXVMM_EXIT_TASK_SWITCH(pVCpu, pCtx); break;
11814	case DBGFEVENT_EXIT_CPUID: VBOXVMM_EXIT_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
11815	case DBGFEVENT_EXIT_GETSEC: VBOXVMM_EXIT_GETSEC(pVCpu, pCtx); break;
11816	case DBGFEVENT_EXIT_HALT: VBOXVMM_EXIT_HALT(pVCpu, pCtx); break;
11817	case DBGFEVENT_EXIT_INVD: VBOXVMM_EXIT_INVD(pVCpu, pCtx); break;
11818	case DBGFEVENT_EXIT_INVLPG: VBOXVMM_EXIT_INVLPG(pVCpu, pCtx); break;
11819	case DBGFEVENT_EXIT_RDPMC: VBOXVMM_EXIT_RDPMC(pVCpu, pCtx); break;
11820	case DBGFEVENT_EXIT_RDTSC: VBOXVMM_EXIT_RDTSC(pVCpu, pCtx); break;
11821	case DBGFEVENT_EXIT_RSM: VBOXVMM_EXIT_RSM(pVCpu, pCtx); break;
11822	case DBGFEVENT_EXIT_CRX_READ: VBOXVMM_EXIT_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
11823	case DBGFEVENT_EXIT_CRX_WRITE: VBOXVMM_EXIT_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
11824	case DBGFEVENT_EXIT_DRX_READ: VBOXVMM_EXIT_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
11825	case DBGFEVENT_EXIT_DRX_WRITE: VBOXVMM_EXIT_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
11826	case DBGFEVENT_EXIT_RDMSR: VBOXVMM_EXIT_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
11827	case DBGFEVENT_EXIT_WRMSR: VBOXVMM_EXIT_WRMSR(pVCpu, pCtx, pCtx->ecx,
11828	RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
11829	case DBGFEVENT_EXIT_MWAIT: VBOXVMM_EXIT_MWAIT(pVCpu, pCtx); break;
11830	case DBGFEVENT_EXIT_MONITOR: VBOXVMM_EXIT_MONITOR(pVCpu, pCtx); break;
11831	case DBGFEVENT_EXIT_PAUSE: VBOXVMM_EXIT_PAUSE(pVCpu, pCtx); break;
11832	case DBGFEVENT_EXIT_SGDT: VBOXVMM_EXIT_SGDT(pVCpu, pCtx); break;
11833	case DBGFEVENT_EXIT_SIDT: VBOXVMM_EXIT_SIDT(pVCpu, pCtx); break;
11834	case DBGFEVENT_EXIT_LGDT: VBOXVMM_EXIT_LGDT(pVCpu, pCtx); break;
11835	case DBGFEVENT_EXIT_LIDT: VBOXVMM_EXIT_LIDT(pVCpu, pCtx); break;
11836	case DBGFEVENT_EXIT_SLDT: VBOXVMM_EXIT_SLDT(pVCpu, pCtx); break;
11837	case DBGFEVENT_EXIT_STR: VBOXVMM_EXIT_STR(pVCpu, pCtx); break;
11838	case DBGFEVENT_EXIT_LLDT: VBOXVMM_EXIT_LLDT(pVCpu, pCtx); break;
11839	case DBGFEVENT_EXIT_LTR: VBOXVMM_EXIT_LTR(pVCpu, pCtx); break;
11840	case DBGFEVENT_EXIT_RDTSCP: VBOXVMM_EXIT_RDTSCP(pVCpu, pCtx); break;
11841	case DBGFEVENT_EXIT_WBINVD: VBOXVMM_EXIT_WBINVD(pVCpu, pCtx); break;
11842	case DBGFEVENT_EXIT_XSETBV: VBOXVMM_EXIT_XSETBV(pVCpu, pCtx); break;
11843	case DBGFEVENT_EXIT_RDRAND: VBOXVMM_EXIT_RDRAND(pVCpu, pCtx); break;
11844	case DBGFEVENT_EXIT_RDSEED: VBOXVMM_EXIT_RDSEED(pVCpu, pCtx); break;
11845	case DBGFEVENT_EXIT_XSAVES: VBOXVMM_EXIT_XSAVES(pVCpu, pCtx); break;
11846	case DBGFEVENT_EXIT_XRSTORS: VBOXVMM_EXIT_XRSTORS(pVCpu, pCtx); break;
11847	case DBGFEVENT_EXIT_VMM_CALL: VBOXVMM_EXIT_VMM_CALL(pVCpu, pCtx); break;
11848	case DBGFEVENT_EXIT_VMX_VMCLEAR: VBOXVMM_EXIT_VMX_VMCLEAR(pVCpu, pCtx); break;
11849	case DBGFEVENT_EXIT_VMX_VMLAUNCH: VBOXVMM_EXIT_VMX_VMLAUNCH(pVCpu, pCtx); break;
11850	case DBGFEVENT_EXIT_VMX_VMPTRLD: VBOXVMM_EXIT_VMX_VMPTRLD(pVCpu, pCtx); break;
11851	case DBGFEVENT_EXIT_VMX_VMPTRST: VBOXVMM_EXIT_VMX_VMPTRST(pVCpu, pCtx); break;
11852	case DBGFEVENT_EXIT_VMX_VMREAD: VBOXVMM_EXIT_VMX_VMREAD(pVCpu, pCtx); break;
11853	case DBGFEVENT_EXIT_VMX_VMRESUME: VBOXVMM_EXIT_VMX_VMRESUME(pVCpu, pCtx); break;
11854	case DBGFEVENT_EXIT_VMX_VMWRITE: VBOXVMM_EXIT_VMX_VMWRITE(pVCpu, pCtx); break;
11855	case DBGFEVENT_EXIT_VMX_VMXOFF: VBOXVMM_EXIT_VMX_VMXOFF(pVCpu, pCtx); break;
11856	case DBGFEVENT_EXIT_VMX_VMXON: VBOXVMM_EXIT_VMX_VMXON(pVCpu, pCtx); break;
11857	case DBGFEVENT_EXIT_VMX_INVEPT: VBOXVMM_EXIT_VMX_INVEPT(pVCpu, pCtx); break;
11858	case DBGFEVENT_EXIT_VMX_INVVPID: VBOXVMM_EXIT_VMX_INVVPID(pVCpu, pCtx); break;
11859	case DBGFEVENT_EXIT_VMX_INVPCID: VBOXVMM_EXIT_VMX_INVPCID(pVCpu, pCtx); break;
11860	case DBGFEVENT_EXIT_VMX_VMFUNC: VBOXVMM_EXIT_VMX_VMFUNC(pVCpu, pCtx); break;
11861	case DBGFEVENT_EXIT_VMX_EPT_MISCONFIG: VBOXVMM_EXIT_VMX_EPT_MISCONFIG(pVCpu, pCtx); break;
11862	case DBGFEVENT_EXIT_VMX_EPT_VIOLATION: VBOXVMM_EXIT_VMX_EPT_VIOLATION(pVCpu, pCtx); break;
11863	case DBGFEVENT_EXIT_VMX_VAPIC_ACCESS: VBOXVMM_EXIT_VMX_VAPIC_ACCESS(pVCpu, pCtx); break;
11864	case DBGFEVENT_EXIT_VMX_VAPIC_WRITE: VBOXVMM_EXIT_VMX_VAPIC_WRITE(pVCpu, pCtx); break;
11865	default: AssertMsgFailed(("enmEvent2=%d uExitReason=%d\n", enmEvent2, uExitReason)); break;
11866	}
11867	}
11868
11869	/*
11870	* Fire of the DBGF event, if enabled (our check here is just a quick one,
11871	* the DBGF call will do a full check).
11872	*
11873	* Note! DBGF sets DBGFEVENT_INTERRUPT_SOFTWARE in the bitmap.
11874	* Note! If we have to events, we prioritize the first, i.e. the instruction
11875	* one, in order to avoid event nesting.
11876	*/
11877	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
11878	if ( enmEvent1 != DBGFEVENT_END
11879	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent1))
11880	{
11881	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
11882	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent1, DBGFEVENTCTX_HM, 1, uEventArg);
11883	if (rcStrict != VINF_SUCCESS)
11884	return rcStrict;
11885	}
11886	else if ( enmEvent2 != DBGFEVENT_END
11887	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent2))
11888	{
11889	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
11890	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent2, DBGFEVENTCTX_HM, 1, uEventArg);
11891	if (rcStrict != VINF_SUCCESS)
11892	return rcStrict;
11893	}
11894
11895	return VINF_SUCCESS;
11896	}
11897
11898
11899	/**
11900	* Single-stepping VM-exit filtering.
11901	*
11902	* This is preprocessing the VM-exits and deciding whether we've gotten far
11903	* enough to return VINF_EM_DBG_STEPPED already. If not, normal VM-exit
11904	* handling is performed.
11905	*
11906	* @returns Strict VBox status code (i.e. informational status codes too).
11907	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
11908	* @param pVmxTransient The VMX-transient structure.
11909	* @param pDbgState The debug state.
11910	*/
11911	DECLINLINE(VBOXSTRICTRC) hmR0VmxRunDebugHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11912	{
11913	/*
11914	* Expensive (saves context) generic dtrace VM-exit probe.
11915	*/
11916	uint32_t const uExitReason = pVmxTransient->uExitReason;
11917	if (!VBOXVMM_R0_HMVMX_VMEXIT_ENABLED())
11918	{ /* more likely */ }
11919	else
11920	{
11921	hmR0VmxReadExitQualVmcs(pVmxTransient);
11922	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
11923	AssertRC(rc);
11924	VBOXVMM_R0_HMVMX_VMEXIT(pVCpu, &pVCpu->cpum.GstCtx, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
11925	}
11926
11927	/*
11928	* Check for host NMI, just to get that out of the way.
11929	*/
11930	if (uExitReason != VMX_EXIT_XCPT_OR_NMI)
11931	{ /* normally likely */ }
11932	else
11933	{
11934	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
11935	uint32_t const uIntType = VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo);
11936	if (uIntType == VMX_EXIT_INT_INFO_TYPE_NMI)
11937	return hmR0VmxExitHostNmi(pVCpu, pVmxTransient->pVmcsInfo);
11938	}
11939
11940	/*
11941	* Check for single stepping event if we're stepping.
11942	*/
11943	if (pVCpu->hm.s.fSingleInstruction)
11944	{
11945	switch (uExitReason)
11946	{
11947	case VMX_EXIT_MTF:
11948	return hmR0VmxExitMtf(pVCpu, pVmxTransient);
11949
11950	/* Various events: */
11951	case VMX_EXIT_XCPT_OR_NMI:
11952	case VMX_EXIT_EXT_INT:
11953	case VMX_EXIT_TRIPLE_FAULT:
11954	case VMX_EXIT_INT_WINDOW:
11955	case VMX_EXIT_NMI_WINDOW:
11956	case VMX_EXIT_TASK_SWITCH:
11957	case VMX_EXIT_TPR_BELOW_THRESHOLD:
11958	case VMX_EXIT_APIC_ACCESS:
11959	case VMX_EXIT_EPT_VIOLATION:
11960	case VMX_EXIT_EPT_MISCONFIG:
11961	case VMX_EXIT_PREEMPT_TIMER:
11962
11963	/* Instruction specific VM-exits: */
11964	case VMX_EXIT_CPUID:
11965	case VMX_EXIT_GETSEC:
11966	case VMX_EXIT_HLT:
11967	case VMX_EXIT_INVD:
11968	case VMX_EXIT_INVLPG:
11969	case VMX_EXIT_RDPMC:
11970	case VMX_EXIT_RDTSC:
11971	case VMX_EXIT_RSM:
11972	case VMX_EXIT_VMCALL:
11973	case VMX_EXIT_VMCLEAR:
11974	case VMX_EXIT_VMLAUNCH:
11975	case VMX_EXIT_VMPTRLD:
11976	case VMX_EXIT_VMPTRST:
11977	case VMX_EXIT_VMREAD:
11978	case VMX_EXIT_VMRESUME:
11979	case VMX_EXIT_VMWRITE:
11980	case VMX_EXIT_VMXOFF:
11981	case VMX_EXIT_VMXON:
11982	case VMX_EXIT_MOV_CRX:
11983	case VMX_EXIT_MOV_DRX:
11984	case VMX_EXIT_IO_INSTR:
11985	case VMX_EXIT_RDMSR:
11986	case VMX_EXIT_WRMSR:
11987	case VMX_EXIT_MWAIT:
11988	case VMX_EXIT_MONITOR:
11989	case VMX_EXIT_PAUSE:
11990	case VMX_EXIT_GDTR_IDTR_ACCESS:
11991	case VMX_EXIT_LDTR_TR_ACCESS:
11992	case VMX_EXIT_INVEPT:
11993	case VMX_EXIT_RDTSCP:
11994	case VMX_EXIT_INVVPID:
11995	case VMX_EXIT_WBINVD:
11996	case VMX_EXIT_XSETBV:
11997	case VMX_EXIT_RDRAND:
11998	case VMX_EXIT_INVPCID:
11999	case VMX_EXIT_VMFUNC:
12000	case VMX_EXIT_RDSEED:
12001	case VMX_EXIT_XSAVES:
12002	case VMX_EXIT_XRSTORS:
12003	{
12004	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12005	AssertRCReturn(rc, rc);
12006	if ( pVCpu->cpum.GstCtx.rip != pDbgState->uRipStart
12007	\|\| pVCpu->cpum.GstCtx.cs.Sel != pDbgState->uCsStart)
12008	return VINF_EM_DBG_STEPPED;
12009	break;
12010	}
12011
12012	/* Errors and unexpected events: */
12013	case VMX_EXIT_INIT_SIGNAL:
12014	case VMX_EXIT_SIPI:
12015	case VMX_EXIT_IO_SMI:
12016	case VMX_EXIT_SMI:
12017	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
12018	case VMX_EXIT_ERR_MSR_LOAD:
12019	case VMX_EXIT_ERR_MACHINE_CHECK:
12020	case VMX_EXIT_PML_FULL:
12021	case VMX_EXIT_VIRTUALIZED_EOI:
12022	case VMX_EXIT_APIC_WRITE: /* Some talk about this being fault like, so I guess we must process it? */
12023	break;
12024
12025	default:
12026	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
12027	break;
12028	}
12029	}
12030
12031	/*
12032	* Check for debugger event breakpoints and dtrace probes.
12033	*/
12034	if ( uExitReason < RT_ELEMENTS(pDbgState->bmExitsToCheck) * 32U
12035	&& ASMBitTest(pDbgState->bmExitsToCheck, uExitReason) )
12036	{
12037	VBOXSTRICTRC rcStrict = hmR0VmxHandleExitDtraceEvents(pVCpu, pVmxTransient, uExitReason);
12038	if (rcStrict != VINF_SUCCESS)
12039	return rcStrict;
12040	}
12041
12042	/*
12043	* Normal processing.
12044	*/
12045	#ifdef HMVMX_USE_FUNCTION_TABLE
12046	return g_apfnVMExitHandlers[uExitReason](pVCpu, pVmxTransient);
12047	#else
12048	return hmR0VmxHandleExit(pVCpu, pVmxTransient, uExitReason);
12049	#endif
12050	}
12051
12052
12053	/**
12054	* Single steps guest code using hardware-assisted VMX.
12055	*
12056	* This is -not- the same as the guest single-stepping itself (say using EFLAGS.TF)
12057	* but single-stepping through the hypervisor debugger.
12058	*
12059	* @returns Strict VBox status code (i.e. informational status codes too).
12060	* @param pVCpu The cross context virtual CPU structure.
12061	* @param pcLoops Pointer to the number of executed loops.
12062	*
12063	* @note Mostly the same as hmR0VmxRunGuestCodeNormal().
12064	*/
12065	static VBOXSTRICTRC hmR0VmxRunGuestCodeDebug(PVMCPUCC pVCpu, uint32_t *pcLoops)
12066	{
12067	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hm.s.cMaxResumeLoops;
12068	Assert(pcLoops);
12069	Assert(*pcLoops <= cMaxResumeLoops);
12070
12071	VMXTRANSIENT VmxTransient;
12072	RT_ZERO(VmxTransient);
12073	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
12074
12075	/* Set HMCPU indicators. */
12076	bool const fSavedSingleInstruction = pVCpu->hm.s.fSingleInstruction;
12077	pVCpu->hm.s.fSingleInstruction = pVCpu->hm.s.fSingleInstruction \|\| DBGFIsStepping(pVCpu);
12078	pVCpu->hm.s.fDebugWantRdTscExit = false;
12079	pVCpu->hm.s.fUsingDebugLoop = true;
12080
12081	/* State we keep to help modify and later restore the VMCS fields we alter, and for detecting steps. */
12082	VMXRUNDBGSTATE DbgState;
12083	hmR0VmxRunDebugStateInit(pVCpu, &VmxTransient, &DbgState);
12084	hmR0VmxPreRunGuestDebugStateUpdate(pVCpu, &VmxTransient, &DbgState);
12085
12086	/*
12087	* The loop.
12088	*/
12089	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
12090	for (;;)
12091	{
12092	Assert(!HMR0SuspendPending());
12093	HMVMX_ASSERT_CPU_SAFE(pVCpu);
12094	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
12095	bool fStepping = pVCpu->hm.s.fSingleInstruction;
12096
12097	/* Set up VM-execution controls the next two can respond to. */
12098	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &VmxTransient, &DbgState);
12099
12100	/*
12101	* Preparatory work for running guest code, this may force us to
12102	* return to ring-3.
12103	*
12104	* Warning! This bugger disables interrupts on VINF_SUCCESS!
12105	*/
12106	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, fStepping);
12107	if (rcStrict != VINF_SUCCESS)
12108	break;
12109
12110	/* Interrupts are disabled at this point! */
12111	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
12112
12113	/* Override any obnoxious code in the above two calls. */
12114	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &VmxTransient, &DbgState);
12115
12116	/*
12117	* Finally execute the guest.
12118	*/
12119	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
12120
12121	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
12122	/* Interrupts are re-enabled at this point! */
12123
12124	/* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
12125	if (RT_SUCCESS(rcRun))
12126	{ /* very likely */ }
12127	else
12128	{
12129	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
12130	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
12131	return rcRun;
12132	}
12133
12134	/* Profile the VM-exit. */
12135	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
12136	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
12137	STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
12138	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
12139	HMVMX_START_EXIT_DISPATCH_PROF();
12140
12141	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
12142
12143	/*
12144	* Handle the VM-exit - we quit earlier on certain VM-exits, see hmR0VmxHandleExitDebug().
12145	*/
12146	rcStrict = hmR0VmxRunDebugHandleExit(pVCpu, &VmxTransient, &DbgState);
12147	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
12148	if (rcStrict != VINF_SUCCESS)
12149	break;
12150	if (++(*pcLoops) > cMaxResumeLoops)
12151	{
12152	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
12153	rcStrict = VINF_EM_RAW_INTERRUPT;
12154	break;
12155	}
12156
12157	/*
12158	* Stepping: Did the RIP change, if so, consider it a single step.
12159	* Otherwise, make sure one of the TFs gets set.
12160	*/
12161	if (fStepping)
12162	{
12163	int rc = hmR0VmxImportGuestState(pVCpu, VmxTransient.pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12164	AssertRC(rc);
12165	if ( pVCpu->cpum.GstCtx.rip != DbgState.uRipStart
12166	\|\| pVCpu->cpum.GstCtx.cs.Sel != DbgState.uCsStart)
12167	{
12168	rcStrict = VINF_EM_DBG_STEPPED;
12169	break;
12170	}
12171	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
12172	}
12173
12174	/*
12175	* Update when dtrace settings changes (DBGF kicks us, so no need to check).
12176	*/
12177	if (VBOXVMM_GET_SETTINGS_SEQ_NO() != DbgState.uDtraceSettingsSeqNo)
12178	hmR0VmxPreRunGuestDebugStateUpdate(pVCpu, &VmxTransient, &DbgState);
12179	}
12180
12181	/*
12182	* Clear the X86_EFL_TF if necessary.
12183	*/
12184	if (pVCpu->hm.s.fClearTrapFlag)
12185	{
12186	int rc = hmR0VmxImportGuestState(pVCpu, VmxTransient.pVmcsInfo, CPUMCTX_EXTRN_RFLAGS);
12187	AssertRC(rc);
12188	pVCpu->hm.s.fClearTrapFlag = false;
12189	pVCpu->cpum.GstCtx.eflags.Bits.u1TF = 0;
12190	}
12191	/** @todo there seems to be issues with the resume flag when the monitor trap
12192	* flag is pending without being used. Seen early in bios init when
12193	* accessing APIC page in protected mode. */
12194
12195	/*
12196	* Restore VM-exit control settings as we may not re-enter this function the
12197	* next time around.
12198	*/
12199	rcStrict = hmR0VmxRunDebugStateRevert(pVCpu, &VmxTransient, &DbgState, rcStrict);
12200
12201	/* Restore HMCPU indicators. */
12202	pVCpu->hm.s.fUsingDebugLoop = false;
12203	pVCpu->hm.s.fDebugWantRdTscExit = false;
12204	pVCpu->hm.s.fSingleInstruction = fSavedSingleInstruction;
12205
12206	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
12207	return rcStrict;
12208	}
12209
12210
12211	/** @} */
12212
12213
12214	/**
12215	* Checks if any expensive dtrace probes are enabled and we should go to the
12216	* debug loop.
12217	*
12218	* @returns true if we should use debug loop, false if not.
12219	*/
12220	static bool hmR0VmxAnyExpensiveProbesEnabled(void)
12221	{
12222	/* It's probably faster to OR the raw 32-bit counter variables together.
12223	Since the variables are in an array and the probes are next to one
12224	another (more or less), we have good locality. So, better read
12225	eight-nine cache lines ever time and only have one conditional, than
12226	128+ conditionals, right? */
12227	return ( VBOXVMM_R0_HMVMX_VMEXIT_ENABLED_RAW() /* expensive too due to context */
12228	\| VBOXVMM_XCPT_DE_ENABLED_RAW()
12229	\| VBOXVMM_XCPT_DB_ENABLED_RAW()
12230	\| VBOXVMM_XCPT_BP_ENABLED_RAW()
12231	\| VBOXVMM_XCPT_OF_ENABLED_RAW()
12232	\| VBOXVMM_XCPT_BR_ENABLED_RAW()
12233	\| VBOXVMM_XCPT_UD_ENABLED_RAW()
12234	\| VBOXVMM_XCPT_NM_ENABLED_RAW()
12235	\| VBOXVMM_XCPT_DF_ENABLED_RAW()
12236	\| VBOXVMM_XCPT_TS_ENABLED_RAW()
12237	\| VBOXVMM_XCPT_NP_ENABLED_RAW()
12238	\| VBOXVMM_XCPT_SS_ENABLED_RAW()
12239	\| VBOXVMM_XCPT_GP_ENABLED_RAW()
12240	\| VBOXVMM_XCPT_PF_ENABLED_RAW()
12241	\| VBOXVMM_XCPT_MF_ENABLED_RAW()
12242	\| VBOXVMM_XCPT_AC_ENABLED_RAW()
12243	\| VBOXVMM_XCPT_XF_ENABLED_RAW()
12244	\| VBOXVMM_XCPT_VE_ENABLED_RAW()
12245	\| VBOXVMM_XCPT_SX_ENABLED_RAW()
12246	\| VBOXVMM_INT_SOFTWARE_ENABLED_RAW()
12247	\| VBOXVMM_INT_HARDWARE_ENABLED_RAW()
12248	) != 0
12249	\|\| ( VBOXVMM_INSTR_HALT_ENABLED_RAW()
12250	\| VBOXVMM_INSTR_MWAIT_ENABLED_RAW()
12251	\| VBOXVMM_INSTR_MONITOR_ENABLED_RAW()
12252	\| VBOXVMM_INSTR_CPUID_ENABLED_RAW()
12253	\| VBOXVMM_INSTR_INVD_ENABLED_RAW()
12254	\| VBOXVMM_INSTR_WBINVD_ENABLED_RAW()
12255	\| VBOXVMM_INSTR_INVLPG_ENABLED_RAW()
12256	\| VBOXVMM_INSTR_RDTSC_ENABLED_RAW()
12257	\| VBOXVMM_INSTR_RDTSCP_ENABLED_RAW()
12258	\| VBOXVMM_INSTR_RDPMC_ENABLED_RAW()
12259	\| VBOXVMM_INSTR_RDMSR_ENABLED_RAW()
12260	\| VBOXVMM_INSTR_WRMSR_ENABLED_RAW()
12261	\| VBOXVMM_INSTR_CRX_READ_ENABLED_RAW()
12262	\| VBOXVMM_INSTR_CRX_WRITE_ENABLED_RAW()
12263	\| VBOXVMM_INSTR_DRX_READ_ENABLED_RAW()
12264	\| VBOXVMM_INSTR_DRX_WRITE_ENABLED_RAW()
12265	\| VBOXVMM_INSTR_PAUSE_ENABLED_RAW()
12266	\| VBOXVMM_INSTR_XSETBV_ENABLED_RAW()
12267	\| VBOXVMM_INSTR_SIDT_ENABLED_RAW()
12268	\| VBOXVMM_INSTR_LIDT_ENABLED_RAW()
12269	\| VBOXVMM_INSTR_SGDT_ENABLED_RAW()
12270	\| VBOXVMM_INSTR_LGDT_ENABLED_RAW()
12271	\| VBOXVMM_INSTR_SLDT_ENABLED_RAW()
12272	\| VBOXVMM_INSTR_LLDT_ENABLED_RAW()
12273	\| VBOXVMM_INSTR_STR_ENABLED_RAW()
12274	\| VBOXVMM_INSTR_LTR_ENABLED_RAW()
12275	\| VBOXVMM_INSTR_GETSEC_ENABLED_RAW()
12276	\| VBOXVMM_INSTR_RSM_ENABLED_RAW()
12277	\| VBOXVMM_INSTR_RDRAND_ENABLED_RAW()
12278	\| VBOXVMM_INSTR_RDSEED_ENABLED_RAW()
12279	\| VBOXVMM_INSTR_XSAVES_ENABLED_RAW()
12280	\| VBOXVMM_INSTR_XRSTORS_ENABLED_RAW()
12281	\| VBOXVMM_INSTR_VMM_CALL_ENABLED_RAW()
12282	\| VBOXVMM_INSTR_VMX_VMCLEAR_ENABLED_RAW()
12283	\| VBOXVMM_INSTR_VMX_VMLAUNCH_ENABLED_RAW()
12284	\| VBOXVMM_INSTR_VMX_VMPTRLD_ENABLED_RAW()
12285	\| VBOXVMM_INSTR_VMX_VMPTRST_ENABLED_RAW()
12286	\| VBOXVMM_INSTR_VMX_VMREAD_ENABLED_RAW()
12287	\| VBOXVMM_INSTR_VMX_VMRESUME_ENABLED_RAW()
12288	\| VBOXVMM_INSTR_VMX_VMWRITE_ENABLED_RAW()
12289	\| VBOXVMM_INSTR_VMX_VMXOFF_ENABLED_RAW()
12290	\| VBOXVMM_INSTR_VMX_VMXON_ENABLED_RAW()
12291	\| VBOXVMM_INSTR_VMX_VMFUNC_ENABLED_RAW()
12292	\| VBOXVMM_INSTR_VMX_INVEPT_ENABLED_RAW()
12293	\| VBOXVMM_INSTR_VMX_INVVPID_ENABLED_RAW()
12294	\| VBOXVMM_INSTR_VMX_INVPCID_ENABLED_RAW()
12295	) != 0
12296	\|\| ( VBOXVMM_EXIT_TASK_SWITCH_ENABLED_RAW()
12297	\| VBOXVMM_EXIT_HALT_ENABLED_RAW()
12298	\| VBOXVMM_EXIT_MWAIT_ENABLED_RAW()
12299	\| VBOXVMM_EXIT_MONITOR_ENABLED_RAW()
12300	\| VBOXVMM_EXIT_CPUID_ENABLED_RAW()
12301	\| VBOXVMM_EXIT_INVD_ENABLED_RAW()
12302	\| VBOXVMM_EXIT_WBINVD_ENABLED_RAW()
12303	\| VBOXVMM_EXIT_INVLPG_ENABLED_RAW()
12304	\| VBOXVMM_EXIT_RDTSC_ENABLED_RAW()
12305	\| VBOXVMM_EXIT_RDTSCP_ENABLED_RAW()
12306	\| VBOXVMM_EXIT_RDPMC_ENABLED_RAW()
12307	\| VBOXVMM_EXIT_RDMSR_ENABLED_RAW()
12308	\| VBOXVMM_EXIT_WRMSR_ENABLED_RAW()
12309	\| VBOXVMM_EXIT_CRX_READ_ENABLED_RAW()
12310	\| VBOXVMM_EXIT_CRX_WRITE_ENABLED_RAW()
12311	\| VBOXVMM_EXIT_DRX_READ_ENABLED_RAW()
12312	\| VBOXVMM_EXIT_DRX_WRITE_ENABLED_RAW()
12313	\| VBOXVMM_EXIT_PAUSE_ENABLED_RAW()
12314	\| VBOXVMM_EXIT_XSETBV_ENABLED_RAW()
12315	\| VBOXVMM_EXIT_SIDT_ENABLED_RAW()
12316	\| VBOXVMM_EXIT_LIDT_ENABLED_RAW()
12317	\| VBOXVMM_EXIT_SGDT_ENABLED_RAW()
12318	\| VBOXVMM_EXIT_LGDT_ENABLED_RAW()
12319	\| VBOXVMM_EXIT_SLDT_ENABLED_RAW()
12320	\| VBOXVMM_EXIT_LLDT_ENABLED_RAW()
12321	\| VBOXVMM_EXIT_STR_ENABLED_RAW()
12322	\| VBOXVMM_EXIT_LTR_ENABLED_RAW()
12323	\| VBOXVMM_EXIT_GETSEC_ENABLED_RAW()
12324	\| VBOXVMM_EXIT_RSM_ENABLED_RAW()
12325	\| VBOXVMM_EXIT_RDRAND_ENABLED_RAW()
12326	\| VBOXVMM_EXIT_RDSEED_ENABLED_RAW()
12327	\| VBOXVMM_EXIT_XSAVES_ENABLED_RAW()
12328	\| VBOXVMM_EXIT_XRSTORS_ENABLED_RAW()
12329	\| VBOXVMM_EXIT_VMM_CALL_ENABLED_RAW()
12330	\| VBOXVMM_EXIT_VMX_VMCLEAR_ENABLED_RAW()
12331	\| VBOXVMM_EXIT_VMX_VMLAUNCH_ENABLED_RAW()
12332	\| VBOXVMM_EXIT_VMX_VMPTRLD_ENABLED_RAW()
12333	\| VBOXVMM_EXIT_VMX_VMPTRST_ENABLED_RAW()
12334	\| VBOXVMM_EXIT_VMX_VMREAD_ENABLED_RAW()
12335	\| VBOXVMM_EXIT_VMX_VMRESUME_ENABLED_RAW()
12336	\| VBOXVMM_EXIT_VMX_VMWRITE_ENABLED_RAW()
12337	\| VBOXVMM_EXIT_VMX_VMXOFF_ENABLED_RAW()
12338	\| VBOXVMM_EXIT_VMX_VMXON_ENABLED_RAW()
12339	\| VBOXVMM_EXIT_VMX_VMFUNC_ENABLED_RAW()
12340	\| VBOXVMM_EXIT_VMX_INVEPT_ENABLED_RAW()
12341	\| VBOXVMM_EXIT_VMX_INVVPID_ENABLED_RAW()
12342	\| VBOXVMM_EXIT_VMX_INVPCID_ENABLED_RAW()
12343	\| VBOXVMM_EXIT_VMX_EPT_VIOLATION_ENABLED_RAW()
12344	\| VBOXVMM_EXIT_VMX_EPT_MISCONFIG_ENABLED_RAW()
12345	\| VBOXVMM_EXIT_VMX_VAPIC_ACCESS_ENABLED_RAW()
12346	\| VBOXVMM_EXIT_VMX_VAPIC_WRITE_ENABLED_RAW()
12347	) != 0;
12348	}
12349
12350
12351	/**
12352	* Runs the guest using hardware-assisted VMX.
12353	*
12354	* @returns Strict VBox status code (i.e. informational status codes too).
12355	* @param pVCpu The cross context virtual CPU structure.
12356	*/
12357	VMMR0DECL(VBOXSTRICTRC) VMXR0RunGuestCode(PVMCPUCC pVCpu)
12358	{
12359	AssertPtr(pVCpu);
12360	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
12361	Assert(VMMRZCallRing3IsEnabled(pVCpu));
12362	Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
12363	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
12364
12365	VMMRZCallRing3SetNotification(pVCpu, hmR0VmxCallRing3Callback, pCtx);
12366
12367	VBOXSTRICTRC rcStrict;
12368	uint32_t cLoops = 0;
12369	for (;;)
12370	{
12371	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12372	bool const fInNestedGuestMode = CPUMIsGuestInVmxNonRootMode(pCtx);
12373	#else
12374	bool const fInNestedGuestMode = false;
12375	#endif
12376	if (!fInNestedGuestMode)
12377	{
12378	if ( !pVCpu->hm.s.fUseDebugLoop
12379	&& (!VBOXVMM_ANY_PROBES_ENABLED() \|\| !hmR0VmxAnyExpensiveProbesEnabled())
12380	&& !DBGFIsStepping(pVCpu)
12381	&& !pVCpu->CTX_SUFF(pVM)->dbgf.ro.cEnabledInt3Breakpoints)
12382	rcStrict = hmR0VmxRunGuestCodeNormal(pVCpu, &cLoops);
12383	else
12384	rcStrict = hmR0VmxRunGuestCodeDebug(pVCpu, &cLoops);
12385	}
12386	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12387	else
12388	rcStrict = hmR0VmxRunGuestCodeNested(pVCpu, &cLoops);
12389
12390	if (rcStrict == VINF_VMX_VMLAUNCH_VMRESUME)
12391	{
12392	Assert(CPUMIsGuestInVmxNonRootMode(pCtx));
12393	continue;
12394	}
12395	if (rcStrict == VINF_VMX_VMEXIT)
12396	{
12397	Assert(!CPUMIsGuestInVmxNonRootMode(pCtx));
12398	continue;
12399	}
12400	#endif
12401	break;
12402	}
12403
12404	int const rcLoop = VBOXSTRICTRC_VAL(rcStrict);
12405	switch (rcLoop)
12406	{
12407	case VERR_EM_INTERPRETER: rcStrict = VINF_EM_RAW_EMULATE_INSTR; break;
12408	case VINF_EM_RESET: rcStrict = VINF_EM_TRIPLE_FAULT; break;
12409	}
12410
12411	int rc2 = hmR0VmxExitToRing3(pVCpu, rcStrict);
12412	if (RT_FAILURE(rc2))
12413	{
12414	pVCpu->hm.s.u32HMError = (uint32_t)VBOXSTRICTRC_VAL(rcStrict);
12415	rcStrict = rc2;
12416	}
12417	Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
12418	Assert(!VMMRZCallRing3IsNotificationSet(pVCpu));
12419	return rcStrict;
12420	}
12421
12422
12423	#ifndef HMVMX_USE_FUNCTION_TABLE
12424	/**
12425	* Handles a guest VM-exit from hardware-assisted VMX execution.
12426	*
12427	* @returns Strict VBox status code (i.e. informational status codes too).
12428	* @param pVCpu The cross context virtual CPU structure.
12429	* @param pVmxTransient The VMX-transient structure.
12430	*/
12431	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
12432	{
12433	#ifdef DEBUG_ramshankar
12434	# define VMEXIT_CALL_RET(a_fSave, a_CallExpr) \
12435	do { \
12436	if (a_fSave != 0) \
12437	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL); \
12438	VBOXSTRICTRC rcStrict = a_CallExpr; \
12439	if (a_fSave != 0) \
12440	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST); \
12441	return rcStrict; \
12442	} while (0)
12443	#else
12444	# define VMEXIT_CALL_RET(a_fSave, a_CallExpr) return a_CallExpr
12445	#endif
12446	uint32_t const uExitReason = pVmxTransient->uExitReason;
12447	switch (uExitReason)
12448	{
12449	case VMX_EXIT_EPT_MISCONFIG: VMEXIT_CALL_RET(0, hmR0VmxExitEptMisconfig(pVCpu, pVmxTransient));
12450	case VMX_EXIT_EPT_VIOLATION: VMEXIT_CALL_RET(0, hmR0VmxExitEptViolation(pVCpu, pVmxTransient));
12451	case VMX_EXIT_IO_INSTR: VMEXIT_CALL_RET(0, hmR0VmxExitIoInstr(pVCpu, pVmxTransient));
12452	case VMX_EXIT_CPUID: VMEXIT_CALL_RET(0, hmR0VmxExitCpuid(pVCpu, pVmxTransient));
12453	case VMX_EXIT_RDTSC: VMEXIT_CALL_RET(0, hmR0VmxExitRdtsc(pVCpu, pVmxTransient));
12454	case VMX_EXIT_RDTSCP: VMEXIT_CALL_RET(0, hmR0VmxExitRdtscp(pVCpu, pVmxTransient));
12455	case VMX_EXIT_APIC_ACCESS: VMEXIT_CALL_RET(0, hmR0VmxExitApicAccess(pVCpu, pVmxTransient));
12456	case VMX_EXIT_XCPT_OR_NMI: VMEXIT_CALL_RET(0, hmR0VmxExitXcptOrNmi(pVCpu, pVmxTransient));
12457	case VMX_EXIT_MOV_CRX: VMEXIT_CALL_RET(0, hmR0VmxExitMovCRx(pVCpu, pVmxTransient));
12458	case VMX_EXIT_EXT_INT: VMEXIT_CALL_RET(0, hmR0VmxExitExtInt(pVCpu, pVmxTransient));
12459	case VMX_EXIT_INT_WINDOW: VMEXIT_CALL_RET(0, hmR0VmxExitIntWindow(pVCpu, pVmxTransient));
12460	case VMX_EXIT_TPR_BELOW_THRESHOLD: VMEXIT_CALL_RET(0, hmR0VmxExitTprBelowThreshold(pVCpu, pVmxTransient));
12461	case VMX_EXIT_MWAIT: VMEXIT_CALL_RET(0, hmR0VmxExitMwait(pVCpu, pVmxTransient));
12462	case VMX_EXIT_MONITOR: VMEXIT_CALL_RET(0, hmR0VmxExitMonitor(pVCpu, pVmxTransient));
12463	case VMX_EXIT_TASK_SWITCH: VMEXIT_CALL_RET(0, hmR0VmxExitTaskSwitch(pVCpu, pVmxTransient));
12464	case VMX_EXIT_PREEMPT_TIMER: VMEXIT_CALL_RET(0, hmR0VmxExitPreemptTimer(pVCpu, pVmxTransient));
12465	case VMX_EXIT_RDMSR: VMEXIT_CALL_RET(0, hmR0VmxExitRdmsr(pVCpu, pVmxTransient));
12466	case VMX_EXIT_WRMSR: VMEXIT_CALL_RET(0, hmR0VmxExitWrmsr(pVCpu, pVmxTransient));
12467	case VMX_EXIT_VMCALL: VMEXIT_CALL_RET(0, hmR0VmxExitVmcall(pVCpu, pVmxTransient));
12468	case VMX_EXIT_MOV_DRX: VMEXIT_CALL_RET(0, hmR0VmxExitMovDRx(pVCpu, pVmxTransient));
12469	case VMX_EXIT_HLT: VMEXIT_CALL_RET(0, hmR0VmxExitHlt(pVCpu, pVmxTransient));
12470	case VMX_EXIT_INVD: VMEXIT_CALL_RET(0, hmR0VmxExitInvd(pVCpu, pVmxTransient));
12471	case VMX_EXIT_INVLPG: VMEXIT_CALL_RET(0, hmR0VmxExitInvlpg(pVCpu, pVmxTransient));
12472	case VMX_EXIT_MTF: VMEXIT_CALL_RET(0, hmR0VmxExitMtf(pVCpu, pVmxTransient));
12473	case VMX_EXIT_PAUSE: VMEXIT_CALL_RET(0, hmR0VmxExitPause(pVCpu, pVmxTransient));
12474	case VMX_EXIT_WBINVD: VMEXIT_CALL_RET(0, hmR0VmxExitWbinvd(pVCpu, pVmxTransient));
12475	case VMX_EXIT_XSETBV: VMEXIT_CALL_RET(0, hmR0VmxExitXsetbv(pVCpu, pVmxTransient));
12476	case VMX_EXIT_INVPCID: VMEXIT_CALL_RET(0, hmR0VmxExitInvpcid(pVCpu, pVmxTransient));
12477	case VMX_EXIT_GETSEC: VMEXIT_CALL_RET(0, hmR0VmxExitGetsec(pVCpu, pVmxTransient));
12478	case VMX_EXIT_RDPMC: VMEXIT_CALL_RET(0, hmR0VmxExitRdpmc(pVCpu, pVmxTransient));
12479	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12480	case VMX_EXIT_VMCLEAR: VMEXIT_CALL_RET(0, hmR0VmxExitVmclear(pVCpu, pVmxTransient));
12481	case VMX_EXIT_VMLAUNCH: VMEXIT_CALL_RET(0, hmR0VmxExitVmlaunch(pVCpu, pVmxTransient));
12482	case VMX_EXIT_VMPTRLD: VMEXIT_CALL_RET(0, hmR0VmxExitVmptrld(pVCpu, pVmxTransient));
12483	case VMX_EXIT_VMPTRST: VMEXIT_CALL_RET(0, hmR0VmxExitVmptrst(pVCpu, pVmxTransient));
12484	case VMX_EXIT_VMREAD: VMEXIT_CALL_RET(0, hmR0VmxExitVmread(pVCpu, pVmxTransient));
12485	case VMX_EXIT_VMRESUME: VMEXIT_CALL_RET(0, hmR0VmxExitVmwrite(pVCpu, pVmxTransient));
12486	case VMX_EXIT_VMWRITE: VMEXIT_CALL_RET(0, hmR0VmxExitVmresume(pVCpu, pVmxTransient));
12487	case VMX_EXIT_VMXOFF: VMEXIT_CALL_RET(0, hmR0VmxExitVmxoff(pVCpu, pVmxTransient));
12488	case VMX_EXIT_VMXON: VMEXIT_CALL_RET(0, hmR0VmxExitVmxon(pVCpu, pVmxTransient));
12489	case VMX_EXIT_INVVPID: VMEXIT_CALL_RET(0, hmR0VmxExitInvvpid(pVCpu, pVmxTransient));
12490	case VMX_EXIT_INVEPT: VMEXIT_CALL_RET(0, hmR0VmxExitSetPendingXcptUD(pVCpu, pVmxTransient));
12491	#else
12492	case VMX_EXIT_VMCLEAR:
12493	case VMX_EXIT_VMLAUNCH:
12494	case VMX_EXIT_VMPTRLD:
12495	case VMX_EXIT_VMPTRST:
12496	case VMX_EXIT_VMREAD:
12497	case VMX_EXIT_VMRESUME:
12498	case VMX_EXIT_VMWRITE:
12499	case VMX_EXIT_VMXOFF:
12500	case VMX_EXIT_VMXON:
12501	case VMX_EXIT_INVVPID:
12502	case VMX_EXIT_INVEPT:
12503	return hmR0VmxExitSetPendingXcptUD(pVCpu, pVmxTransient);
12504	#endif
12505
12506	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFault(pVCpu, pVmxTransient);
12507	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindow(pVCpu, pVmxTransient);
12508	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestState(pVCpu, pVmxTransient);
12509
12510	case VMX_EXIT_INIT_SIGNAL:
12511	case VMX_EXIT_SIPI:
12512	case VMX_EXIT_IO_SMI:
12513	case VMX_EXIT_SMI:
12514	case VMX_EXIT_ERR_MSR_LOAD:
12515	case VMX_EXIT_ERR_MACHINE_CHECK:
12516	case VMX_EXIT_PML_FULL:
12517	case VMX_EXIT_VIRTUALIZED_EOI:
12518	case VMX_EXIT_GDTR_IDTR_ACCESS:
12519	case VMX_EXIT_LDTR_TR_ACCESS:
12520	case VMX_EXIT_APIC_WRITE:
12521	case VMX_EXIT_RDRAND:
12522	case VMX_EXIT_RSM:
12523	case VMX_EXIT_VMFUNC:
12524	case VMX_EXIT_ENCLS:
12525	case VMX_EXIT_RDSEED:
12526	case VMX_EXIT_XSAVES:
12527	case VMX_EXIT_XRSTORS:
12528	case VMX_EXIT_UMWAIT:
12529	case VMX_EXIT_TPAUSE:
12530	default:
12531	return hmR0VmxExitErrUnexpected(pVCpu, pVmxTransient);
12532	}
12533	#undef VMEXIT_CALL_RET
12534	}
12535	#endif /* !HMVMX_USE_FUNCTION_TABLE */
12536
12537
12538	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12539	/**
12540	* Handles a nested-guest VM-exit from hardware-assisted VMX execution.
12541	*
12542	* @returns Strict VBox status code (i.e. informational status codes too).
12543	* @param pVCpu The cross context virtual CPU structure.
12544	* @param pVmxTransient The VMX-transient structure.
12545	*/
12546	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
12547	{
12548	/** @todo NSTVMX: Remove after debugging page-fault issue. */
12549	#ifdef DEBUG_ramshankar
12550	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
12551	Log4Func(("cs:rip=%#04x:%#RX64 rsp=%#RX64 eflags=%#RX32 cr3=%#RX64\n", pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
12552	pVCpu->cpum.GstCtx.rsp, pVCpu->cpum.GstCtx.eflags.u32, pVCpu->cpum.GstCtx.cr3));
12553	#endif
12554
12555	uint32_t const uExitReason = pVmxTransient->uExitReason;
12556	switch (uExitReason)
12557	{
12558	case VMX_EXIT_EPT_MISCONFIG: return hmR0VmxExitEptMisconfig(pVCpu, pVmxTransient);
12559	case VMX_EXIT_EPT_VIOLATION: return hmR0VmxExitEptViolation(pVCpu, pVmxTransient);
12560	case VMX_EXIT_XCPT_OR_NMI: return hmR0VmxExitXcptOrNmiNested(pVCpu, pVmxTransient);
12561	case VMX_EXIT_IO_INSTR: return hmR0VmxExitIoInstrNested(pVCpu, pVmxTransient);
12562	case VMX_EXIT_HLT: return hmR0VmxExitHltNested(pVCpu, pVmxTransient);
12563
12564	/*
12565	* We shouldn't direct host physical interrupts to the nested-guest.
12566	*/
12567	case VMX_EXIT_EXT_INT:
12568	return hmR0VmxExitExtInt(pVCpu, pVmxTransient);
12569
12570	/*
12571	* Instructions that cause VM-exits unconditionally or the condition is
12572	* always is taken solely from the guest hypervisor (meaning if the VM-exit
12573	* happens, it's guaranteed to be a nested-guest VM-exit).
12574	*
12575	* - Provides VM-exit instruction length ONLY.
12576	*/
12577	case VMX_EXIT_CPUID: /* Unconditional. */
12578	case VMX_EXIT_VMCALL:
12579	case VMX_EXIT_GETSEC:
12580	case VMX_EXIT_INVD:
12581	case VMX_EXIT_XSETBV:
12582	case VMX_EXIT_VMLAUNCH:
12583	case VMX_EXIT_VMRESUME:
12584	case VMX_EXIT_VMXOFF:
12585	case VMX_EXIT_ENCLS: /* Condition specified solely by guest hypervisor. */
12586	case VMX_EXIT_VMFUNC:
12587	return hmR0VmxExitInstrNested(pVCpu, pVmxTransient);
12588
12589	/*
12590	* Instructions that cause VM-exits unconditionally or the condition is
12591	* always is taken solely from the guest hypervisor (meaning if the VM-exit
12592	* happens, it's guaranteed to be a nested-guest VM-exit).
12593	*
12594	* - Provides VM-exit instruction length.
12595	* - Provides VM-exit information.
12596	* - Optionally provides Exit qualification.
12597	*
12598	* Since Exit qualification is 0 for all VM-exits where it is not
12599	* applicable, reading and passing it to the guest should produce
12600	* defined behavior.
12601	*
12602	* See Intel spec. 27.2.1 "Basic VM-Exit Information".
12603	*/
12604	case VMX_EXIT_INVEPT: /* Unconditional. */
12605	case VMX_EXIT_INVVPID:
12606	case VMX_EXIT_VMCLEAR:
12607	case VMX_EXIT_VMPTRLD:
12608	case VMX_EXIT_VMPTRST:
12609	case VMX_EXIT_VMXON:
12610	case VMX_EXIT_GDTR_IDTR_ACCESS: /* Condition specified solely by guest hypervisor. */
12611	case VMX_EXIT_LDTR_TR_ACCESS:
12612	case VMX_EXIT_RDRAND:
12613	case VMX_EXIT_RDSEED:
12614	case VMX_EXIT_XSAVES:
12615	case VMX_EXIT_XRSTORS:
12616	case VMX_EXIT_UMWAIT:
12617	case VMX_EXIT_TPAUSE:
12618	return hmR0VmxExitInstrWithInfoNested(pVCpu, pVmxTransient);
12619
12620	case VMX_EXIT_RDTSC: return hmR0VmxExitRdtscNested(pVCpu, pVmxTransient);
12621	case VMX_EXIT_RDTSCP: return hmR0VmxExitRdtscpNested(pVCpu, pVmxTransient);
12622	case VMX_EXIT_RDMSR: return hmR0VmxExitRdmsrNested(pVCpu, pVmxTransient);
12623	case VMX_EXIT_WRMSR: return hmR0VmxExitWrmsrNested(pVCpu, pVmxTransient);
12624	case VMX_EXIT_INVLPG: return hmR0VmxExitInvlpgNested(pVCpu, pVmxTransient);
12625	case VMX_EXIT_INVPCID: return hmR0VmxExitInvpcidNested(pVCpu, pVmxTransient);
12626	case VMX_EXIT_TASK_SWITCH: return hmR0VmxExitTaskSwitchNested(pVCpu, pVmxTransient);
12627	case VMX_EXIT_WBINVD: return hmR0VmxExitWbinvdNested(pVCpu, pVmxTransient);
12628	case VMX_EXIT_MTF: return hmR0VmxExitMtfNested(pVCpu, pVmxTransient);
12629	case VMX_EXIT_APIC_ACCESS: return hmR0VmxExitApicAccessNested(pVCpu, pVmxTransient);
12630	case VMX_EXIT_APIC_WRITE: return hmR0VmxExitApicWriteNested(pVCpu, pVmxTransient);
12631	case VMX_EXIT_VIRTUALIZED_EOI: return hmR0VmxExitVirtEoiNested(pVCpu, pVmxTransient);
12632	case VMX_EXIT_MOV_CRX: return hmR0VmxExitMovCRxNested(pVCpu, pVmxTransient);
12633	case VMX_EXIT_INT_WINDOW: return hmR0VmxExitIntWindowNested(pVCpu, pVmxTransient);
12634	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindowNested(pVCpu, pVmxTransient);
12635	case VMX_EXIT_TPR_BELOW_THRESHOLD: return hmR0VmxExitTprBelowThresholdNested(pVCpu, pVmxTransient);
12636	case VMX_EXIT_MWAIT: return hmR0VmxExitMwaitNested(pVCpu, pVmxTransient);
12637	case VMX_EXIT_MONITOR: return hmR0VmxExitMonitorNested(pVCpu, pVmxTransient);
12638	case VMX_EXIT_PAUSE: return hmR0VmxExitPauseNested(pVCpu, pVmxTransient);
12639
12640	case VMX_EXIT_PREEMPT_TIMER:
12641	{
12642	/** @todo NSTVMX: Preempt timer. */
12643	return hmR0VmxExitPreemptTimer(pVCpu, pVmxTransient);
12644	}
12645
12646	case VMX_EXIT_MOV_DRX: return hmR0VmxExitMovDRxNested(pVCpu, pVmxTransient);
12647	case VMX_EXIT_RDPMC: return hmR0VmxExitRdpmcNested(pVCpu, pVmxTransient);
12648
12649	case VMX_EXIT_VMREAD:
12650	case VMX_EXIT_VMWRITE: return hmR0VmxExitVmreadVmwriteNested(pVCpu, pVmxTransient);
12651
12652	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFaultNested(pVCpu, pVmxTransient);
12653	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestStateNested(pVCpu, pVmxTransient);
12654
12655	case VMX_EXIT_INIT_SIGNAL:
12656	case VMX_EXIT_SIPI:
12657	case VMX_EXIT_IO_SMI:
12658	case VMX_EXIT_SMI:
12659	case VMX_EXIT_ERR_MSR_LOAD:
12660	case VMX_EXIT_ERR_MACHINE_CHECK:
12661	case VMX_EXIT_PML_FULL:
12662	case VMX_EXIT_RSM:
12663	default:
12664	return hmR0VmxExitErrUnexpected(pVCpu, pVmxTransient);
12665	}
12666	}
12667	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
12668
12669
12670	/** @name VM-exit helpers.
12671	* @{
12672	*/
12673	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
12674	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= VM-exit helpers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
12675	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
12676
12677	/** Macro for VM-exits called unexpectedly. */
12678	#define HMVMX_UNEXPECTED_EXIT_RET(a_pVCpu, a_HmError) \
12679	do { \
12680	(a_pVCpu)->hm.s.u32HMError = (a_HmError); \
12681	return VERR_VMX_UNEXPECTED_EXIT; \
12682	} while (0)
12683
12684	#ifdef VBOX_STRICT
12685	/* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */
12686	# define HMVMX_ASSERT_PREEMPT_CPUID_VAR() \
12687	RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId()
12688
12689	# define HMVMX_ASSERT_PREEMPT_CPUID() \
12690	do { \
12691	RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \
12692	AssertMsg(idAssertCpu == idAssertCpuNow, ("VMX %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \
12693	} while (0)
12694
12695	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
12696	do { \
12697	AssertPtr((a_pVCpu)); \
12698	AssertPtr((a_pVmxTransient)); \
12699	Assert((a_pVmxTransient)->fVMEntryFailed == false); \
12700	Assert((a_pVmxTransient)->pVmcsInfo); \
12701	Assert(ASMIntAreEnabled()); \
12702	HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu); \
12703	HMVMX_ASSERT_PREEMPT_CPUID_VAR(); \
12704	Log4Func(("vcpu[%RU32]\n", (a_pVCpu)->idCpu)); \
12705	HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu); \
12706	if (VMMR0IsLogFlushDisabled((a_pVCpu))) \
12707	HMVMX_ASSERT_PREEMPT_CPUID(); \
12708	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
12709	} while (0)
12710
12711	# define HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
12712	do { \
12713	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient); \
12714	Assert((a_pVmxTransient)->fIsNestedGuest); \
12715	} while (0)
12716
12717	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
12718	do { \
12719	Log4Func(("\n")); \
12720	} while (0)
12721	#else
12722	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
12723	do { \
12724	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
12725	NOREF((a_pVCpu)); NOREF((a_pVmxTransient)); \
12726	} while (0)
12727
12728	# define HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
12729	do { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient); } while (0)
12730
12731	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) do { } while (0)
12732	#endif
12733
12734	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12735	/** Macro that does the necessary privilege checks and intercepted VM-exits for
12736	* guests that attempted to execute a VMX instruction. */
12737	# define HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(a_pVCpu, a_uExitReason) \
12738	do \
12739	{ \
12740	VBOXSTRICTRC rcStrictTmp = hmR0VmxCheckExitDueToVmxInstr((a_pVCpu), (a_uExitReason)); \
12741	if (rcStrictTmp == VINF_SUCCESS) \
12742	{ /* likely */ } \
12743	else if (rcStrictTmp == VINF_HM_PENDING_XCPT) \
12744	{ \
12745	Assert((a_pVCpu)->hm.s.Event.fPending); \
12746	Log4Func(("Privilege checks failed -> %#x\n", VMX_ENTRY_INT_INFO_VECTOR((a_pVCpu)->hm.s.Event.u64IntInfo))); \
12747	return VINF_SUCCESS; \
12748	} \
12749	else \
12750	{ \
12751	int rcTmp = VBOXSTRICTRC_VAL(rcStrictTmp); \
12752	AssertMsgFailedReturn(("Unexpected failure. rc=%Rrc", rcTmp), rcTmp); \
12753	} \
12754	} while (0)
12755
12756	/** Macro that decodes a memory operand for an VM-exit caused by an instruction. */
12757	# define HMVMX_DECODE_MEM_OPERAND(a_pVCpu, a_uExitInstrInfo, a_uExitQual, a_enmMemAccess, a_pGCPtrEffAddr) \
12758	do \
12759	{ \
12760	VBOXSTRICTRC rcStrictTmp = hmR0VmxDecodeMemOperand((a_pVCpu), (a_uExitInstrInfo), (a_uExitQual), (a_enmMemAccess), \
12761	(a_pGCPtrEffAddr)); \
12762	if (rcStrictTmp == VINF_SUCCESS) \
12763	{ /* likely */ } \
12764	else if (rcStrictTmp == VINF_HM_PENDING_XCPT) \
12765	{ \
12766	uint8_t const uXcptTmp = VMX_ENTRY_INT_INFO_VECTOR((a_pVCpu)->hm.s.Event.u64IntInfo); \
12767	Log4Func(("Memory operand decoding failed, raising xcpt %#x\n", uXcptTmp)); \
12768	NOREF(uXcptTmp); \
12769	return VINF_SUCCESS; \
12770	} \
12771	else \
12772	{ \
12773	Log4Func(("hmR0VmxDecodeMemOperand failed. rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrictTmp))); \
12774	return rcStrictTmp; \
12775	} \
12776	} while (0)
12777	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
12778
12779
12780	/**
12781	* Advances the guest RIP by the specified number of bytes.
12782	*
12783	* @param pVCpu The cross context virtual CPU structure.
12784	* @param cbInstr Number of bytes to advance the RIP by.
12785	*
12786	* @remarks No-long-jump zone!!!
12787	*/
12788	DECLINLINE(void) hmR0VmxAdvanceGuestRipBy(PVMCPUCC pVCpu, uint32_t cbInstr)
12789	{
12790	/* Advance the RIP. */
12791	pVCpu->cpum.GstCtx.rip += cbInstr;
12792	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
12793
12794	/* Update interrupt inhibition. */
12795	if ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)
12796	&& pVCpu->cpum.GstCtx.rip != EMGetInhibitInterruptsPC(pVCpu))
12797	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
12798	}
12799
12800
12801	/**
12802	* Advances the guest RIP after reading it from the VMCS.
12803	*
12804	* @returns VBox status code, no informational status codes.
12805	* @param pVCpu The cross context virtual CPU structure.
12806	* @param pVmxTransient The VMX-transient structure.
12807	*
12808	* @remarks No-long-jump zone!!!
12809	*/
12810	static int hmR0VmxAdvanceGuestRip(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
12811	{
12812	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
12813	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS);
12814	AssertRCReturn(rc, rc);
12815
12816	hmR0VmxAdvanceGuestRipBy(pVCpu, pVmxTransient->cbExitInstr);
12817	return VINF_SUCCESS;
12818	}
12819
12820
12821	/**
12822	* Handle a condition that occurred while delivering an event through the guest or
12823	* nested-guest IDT.
12824	*
12825	* @returns Strict VBox status code (i.e. informational status codes too).
12826	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
12827	* @retval VINF_HM_DOUBLE_FAULT if a \#DF condition was detected and we ought
12828	* to continue execution of the guest which will delivery the \#DF.
12829	* @retval VINF_EM_RESET if we detected a triple-fault condition.
12830	* @retval VERR_EM_GUEST_CPU_HANG if we detected a guest CPU hang.
12831	*
12832	* @param pVCpu The cross context virtual CPU structure.
12833	* @param pVmxTransient The VMX-transient structure.
12834	*
12835	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
12836	* Additionally, HMVMX_READ_EXIT_QUALIFICATION is required if the VM-exit
12837	* is due to an EPT violation, PML full or SPP-related event.
12838	*
12839	* @remarks No-long-jump zone!!!
12840	*/
12841	static VBOXSTRICTRC hmR0VmxCheckExitDueToEventDelivery(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
12842	{
12843	Assert(!pVCpu->hm.s.Event.fPending);
12844	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_XCPT_INFO);
12845	if ( pVmxTransient->uExitReason == VMX_EXIT_EPT_VIOLATION
12846	\|\| pVmxTransient->uExitReason == VMX_EXIT_PML_FULL
12847	\|\| pVmxTransient->uExitReason == VMX_EXIT_SPP_EVENT)
12848	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_EXIT_QUALIFICATION);
12849
12850	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
12851	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
12852	uint32_t const uIdtVectorInfo = pVmxTransient->uIdtVectoringInfo;
12853	uint32_t const uExitIntInfo = pVmxTransient->uExitIntInfo;
12854	if (VMX_IDT_VECTORING_INFO_IS_VALID(uIdtVectorInfo))
12855	{
12856	uint32_t const uIdtVector = VMX_IDT_VECTORING_INFO_VECTOR(uIdtVectorInfo);
12857	uint32_t const uIdtVectorType = VMX_IDT_VECTORING_INFO_TYPE(uIdtVectorInfo);
12858
12859	/*
12860	* If the event was a software interrupt (generated with INT n) or a software exception
12861	* (generated by INT3/INTO) or a privileged software exception (generated by INT1), we
12862	* can handle the VM-exit and continue guest execution which will re-execute the
12863	* instruction rather than re-injecting the exception, as that can cause premature
12864	* trips to ring-3 before injection and involve TRPM which currently has no way of
12865	* storing that these exceptions were caused by these instructions (ICEBP's #DB poses
12866	* the problem).
12867	*/
12868	IEMXCPTRAISE enmRaise;
12869	IEMXCPTRAISEINFO fRaiseInfo;
12870	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
12871	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
12872	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT)
12873	{
12874	enmRaise = IEMXCPTRAISE_REEXEC_INSTR;
12875	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
12876	}
12877	else if (VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo))
12878	{
12879	uint32_t const uExitVectorType = VMX_EXIT_INT_INFO_TYPE(uExitIntInfo);
12880	uint8_t const uExitVector = VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo);
12881	Assert(uExitVectorType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT);
12882
12883	uint32_t const fIdtVectorFlags = hmR0VmxGetIemXcptFlags(uIdtVector, uIdtVectorType);
12884	uint32_t const fExitVectorFlags = hmR0VmxGetIemXcptFlags(uExitVector, uExitVectorType);
12885
12886	enmRaise = IEMEvaluateRecursiveXcpt(pVCpu, fIdtVectorFlags, uIdtVector, fExitVectorFlags, uExitVector, &fRaiseInfo);
12887
12888	/* Determine a vectoring #PF condition, see comment in hmR0VmxExitXcptPF(). */
12889	if (fRaiseInfo & (IEMXCPTRAISEINFO_EXT_INT_PF \| IEMXCPTRAISEINFO_NMI_PF))
12890	{
12891	pVmxTransient->fVectoringPF = true;
12892	enmRaise = IEMXCPTRAISE_PREV_EVENT;
12893	}
12894	}
12895	else
12896	{
12897	/*
12898	* If an exception or hardware interrupt delivery caused an EPT violation/misconfig or APIC access
12899	* VM-exit, then the VM-exit interruption-information will not be valid and we end up here.
12900	* It is sufficient to reflect the original event to the guest after handling the VM-exit.
12901	*/
12902	Assert( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
12903	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
12904	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT);
12905	enmRaise = IEMXCPTRAISE_PREV_EVENT;
12906	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
12907	}
12908
12909	/*
12910	* On CPUs that support Virtual NMIs, if this VM-exit (be it an exception or EPT violation/misconfig
12911	* etc.) occurred while delivering the NMI, we need to clear the block-by-NMI field in the guest
12912	* interruptibility-state before re-delivering the NMI after handling the VM-exit. Otherwise the
12913	* subsequent VM-entry would fail, see @bugref{7445}.
12914	*
12915	* See Intel spec. 30.7.1.2 "Resuming Guest Software after Handling an Exception".
12916	*/
12917	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
12918	&& enmRaise == IEMXCPTRAISE_PREV_EVENT
12919	&& (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
12920	&& CPUMIsGuestNmiBlocking(pVCpu))
12921	{
12922	CPUMSetGuestNmiBlocking(pVCpu, false);
12923	}
12924
12925	switch (enmRaise)
12926	{
12927	case IEMXCPTRAISE_CURRENT_XCPT:
12928	{
12929	Log4Func(("IDT: Pending secondary Xcpt: idtinfo=%#RX64 exitinfo=%#RX64\n", uIdtVectorInfo, uExitIntInfo));
12930	Assert(rcStrict == VINF_SUCCESS);
12931	break;
12932	}
12933
12934	case IEMXCPTRAISE_PREV_EVENT:
12935	{
12936	uint32_t u32ErrCode;
12937	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(uIdtVectorInfo))
12938	u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
12939	else
12940	u32ErrCode = 0;
12941
12942	/* If uExitVector is #PF, CR2 value will be updated from the VMCS if it's a guest #PF, see hmR0VmxExitXcptPF(). */
12943	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflect);
12944	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(uIdtVectorInfo), 0 /* cbInstr */,
12945	u32ErrCode, pVCpu->cpum.GstCtx.cr2);
12946
12947	Log4Func(("IDT: Pending vectoring event %#RX64 Err=%#RX32\n", pVCpu->hm.s.Event.u64IntInfo,
12948	pVCpu->hm.s.Event.u32ErrCode));
12949	Assert(rcStrict == VINF_SUCCESS);
12950	break;
12951	}
12952
12953	case IEMXCPTRAISE_REEXEC_INSTR:
12954	Assert(rcStrict == VINF_SUCCESS);
12955	break;
12956
12957	case IEMXCPTRAISE_DOUBLE_FAULT:
12958	{
12959	/*
12960	* Determing a vectoring double #PF condition. Used later, when PGM evaluates the
12961	* second #PF as a guest #PF (and not a shadow #PF) and needs to be converted into a #DF.
12962	*/
12963	if (fRaiseInfo & IEMXCPTRAISEINFO_PF_PF)
12964	{
12965	pVmxTransient->fVectoringDoublePF = true;
12966	Log4Func(("IDT: Vectoring double #PF %#RX64 cr2=%#RX64\n", pVCpu->hm.s.Event.u64IntInfo,
12967	pVCpu->cpum.GstCtx.cr2));
12968	rcStrict = VINF_SUCCESS;
12969	}
12970	else
12971	{
12972	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectConvertDF);
12973	hmR0VmxSetPendingXcptDF(pVCpu);
12974	Log4Func(("IDT: Pending vectoring #DF %#RX64 uIdtVector=%#x uExitVector=%#x\n", pVCpu->hm.s.Event.u64IntInfo,
12975	uIdtVector, VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo)));
12976	rcStrict = VINF_HM_DOUBLE_FAULT;
12977	}
12978	break;
12979	}
12980
12981	case IEMXCPTRAISE_TRIPLE_FAULT:
12982	{
12983	Log4Func(("IDT: Pending vectoring triple-fault uIdt=%#x uExit=%#x\n", uIdtVector,
12984	VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo)));
12985	rcStrict = VINF_EM_RESET;
12986	break;
12987	}
12988
12989	case IEMXCPTRAISE_CPU_HANG:
12990	{
12991	Log4Func(("IDT: Bad guest! Entering CPU hang. fRaiseInfo=%#x\n", fRaiseInfo));
12992	rcStrict = VERR_EM_GUEST_CPU_HANG;
12993	break;
12994	}
12995
12996	default:
12997	{
12998	AssertMsgFailed(("IDT: vcpu[%RU32] Unexpected/invalid value! enmRaise=%#x\n", pVCpu->idCpu, enmRaise));
12999	rcStrict = VERR_VMX_IPE_2;
13000	break;
13001	}
13002	}
13003	}
13004	else if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
13005	&& !CPUMIsGuestNmiBlocking(pVCpu))
13006	{
13007	if ( VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo)
13008	&& VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo) != X86_XCPT_DF
13009	&& VMX_EXIT_INT_INFO_IS_NMI_UNBLOCK_IRET(uExitIntInfo))
13010	{
13011	/*
13012	* Execution of IRET caused a fault when NMI blocking was in effect (i.e we're in
13013	* the guest or nested-guest NMI handler). We need to set the block-by-NMI field so
13014	* that NMIs remain blocked until the IRET execution is completed.
13015	*
13016	* See Intel spec. 31.7.1.2 "Resuming Guest Software After Handling An Exception".
13017	*/
13018	CPUMSetGuestNmiBlocking(pVCpu, true);
13019	Log4Func(("Set NMI blocking. uExitReason=%u\n", pVmxTransient->uExitReason));
13020	}
13021	else if ( pVmxTransient->uExitReason == VMX_EXIT_EPT_VIOLATION
13022	\|\| pVmxTransient->uExitReason == VMX_EXIT_PML_FULL
13023	\|\| pVmxTransient->uExitReason == VMX_EXIT_SPP_EVENT)
13024	{
13025	/*
13026	* Execution of IRET caused an EPT violation, page-modification log-full event or
13027	* SPP-related event VM-exit when NMI blocking was in effect (i.e. we're in the
13028	* guest or nested-guest NMI handler). We need to set the block-by-NMI field so
13029	* that NMIs remain blocked until the IRET execution is completed.
13030	*
13031	* See Intel spec. 27.2.3 "Information about NMI unblocking due to IRET"
13032	*/
13033	if (VMX_EXIT_QUAL_EPT_IS_NMI_UNBLOCK_IRET(pVmxTransient->uExitQual))
13034	{
13035	CPUMSetGuestNmiBlocking(pVCpu, true);
13036	Log4Func(("Set NMI blocking. uExitReason=%u\n", pVmxTransient->uExitReason));
13037	}
13038	}
13039	}
13040
13041	Assert( rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_HM_DOUBLE_FAULT
13042	\|\| rcStrict == VINF_EM_RESET \|\| rcStrict == VERR_EM_GUEST_CPU_HANG);
13043	return rcStrict;
13044	}
13045
13046
13047	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13048	/**
13049	* Perform the relevant VMX instruction checks for VM-exits that occurred due to the
13050	* guest attempting to execute a VMX instruction.
13051	*
13052	* @returns Strict VBox status code (i.e. informational status codes too).
13053	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
13054	* @retval VINF_HM_PENDING_XCPT if an exception was raised.
13055	*
13056	* @param pVCpu The cross context virtual CPU structure.
13057	* @param uExitReason The VM-exit reason.
13058	*
13059	* @todo NSTVMX: Document other error codes when VM-exit is implemented.
13060	* @remarks No-long-jump zone!!!
13061	*/
13062	static VBOXSTRICTRC hmR0VmxCheckExitDueToVmxInstr(PVMCPUCC pVCpu, uint32_t uExitReason)
13063	{
13064	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS
13065	\| CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_EFER);
13066
13067	/*
13068	* The physical CPU would have already checked the CPU mode/code segment.
13069	* We shall just assert here for paranoia.
13070	* See Intel spec. 25.1.1 "Relative Priority of Faults and VM Exits".
13071	*/
13072	Assert(!CPUMIsGuestInRealOrV86ModeEx(&pVCpu->cpum.GstCtx));
13073	Assert( !CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx)
13074	\|\| CPUMIsGuestIn64BitCodeEx(&pVCpu->cpum.GstCtx));
13075
13076	if (uExitReason == VMX_EXIT_VMXON)
13077	{
13078	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
13079
13080	/*
13081	* We check CR4.VMXE because it is required to be always set while in VMX operation
13082	* by physical CPUs and our CR4 read-shadow is only consulted when executing specific
13083	* instructions (CLTS, LMSW, MOV CR, and SMSW) and thus doesn't affect CPU operation
13084	* otherwise (i.e. physical CPU won't automatically #UD if Cr4Shadow.VMXE is 0).
13085	*/
13086	if (!CPUMIsGuestVmxEnabled(&pVCpu->cpum.GstCtx))
13087	{
13088	Log4Func(("CR4.VMXE is not set -> #UD\n"));
13089	hmR0VmxSetPendingXcptUD(pVCpu);
13090	return VINF_HM_PENDING_XCPT;
13091	}
13092	}
13093	else if (!CPUMIsGuestInVmxRootMode(&pVCpu->cpum.GstCtx))
13094	{
13095	/*
13096	* The guest has not entered VMX operation but attempted to execute a VMX instruction
13097	* (other than VMXON), we need to raise a #UD.
13098	*/
13099	Log4Func(("Not in VMX root mode -> #UD\n"));
13100	hmR0VmxSetPendingXcptUD(pVCpu);
13101	return VINF_HM_PENDING_XCPT;
13102	}
13103
13104	/* All other checks (including VM-exit intercepts) are handled by IEM instruction emulation. */
13105	return VINF_SUCCESS;
13106	}
13107
13108
13109	/**
13110	* Decodes the memory operand of an instruction that caused a VM-exit.
13111	*
13112	* The Exit qualification field provides the displacement field for memory
13113	* operand instructions, if any.
13114	*
13115	* @returns Strict VBox status code (i.e. informational status codes too).
13116	* @retval VINF_SUCCESS if the operand was successfully decoded.
13117	* @retval VINF_HM_PENDING_XCPT if an exception was raised while decoding the
13118	* operand.
13119	* @param pVCpu The cross context virtual CPU structure.
13120	* @param uExitInstrInfo The VM-exit instruction information field.
13121	* @param enmMemAccess The memory operand's access type (read or write).
13122	* @param GCPtrDisp The instruction displacement field, if any. For
13123	* RIP-relative addressing pass RIP + displacement here.
13124	* @param pGCPtrMem Where to store the effective destination memory address.
13125	*
13126	* @remarks Warning! This function ASSUMES the instruction cannot be used in real or
13127	* virtual-8086 mode hence skips those checks while verifying if the
13128	* segment is valid.
13129	*/
13130	static VBOXSTRICTRC hmR0VmxDecodeMemOperand(PVMCPUCC pVCpu, uint32_t uExitInstrInfo, RTGCPTR GCPtrDisp, VMXMEMACCESS enmMemAccess,
13131	PRTGCPTR pGCPtrMem)
13132	{
13133	Assert(pGCPtrMem);
13134	Assert(!CPUMIsGuestInRealOrV86Mode(pVCpu));
13135	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_EFER
13136	\| CPUMCTX_EXTRN_CR0);
13137
13138	static uint64_t const s_auAddrSizeMasks[] = { UINT64_C(0xffff), UINT64_C(0xffffffff), UINT64_C(0xffffffffffffffff) };
13139	static uint64_t const s_auAccessSizeMasks[] = { sizeof(uint16_t), sizeof(uint32_t), sizeof(uint64_t) };
13140	AssertCompile(RT_ELEMENTS(s_auAccessSizeMasks) == RT_ELEMENTS(s_auAddrSizeMasks));
13141
13142	VMXEXITINSTRINFO ExitInstrInfo;
13143	ExitInstrInfo.u = uExitInstrInfo;
13144	uint8_t const uAddrSize = ExitInstrInfo.All.u3AddrSize;
13145	uint8_t const iSegReg = ExitInstrInfo.All.iSegReg;
13146	bool const fIdxRegValid = !ExitInstrInfo.All.fIdxRegInvalid;
13147	uint8_t const iIdxReg = ExitInstrInfo.All.iIdxReg;
13148	uint8_t const uScale = ExitInstrInfo.All.u2Scaling;
13149	bool const fBaseRegValid = !ExitInstrInfo.All.fBaseRegInvalid;
13150	uint8_t const iBaseReg = ExitInstrInfo.All.iBaseReg;
13151	bool const fIsMemOperand = !ExitInstrInfo.All.fIsRegOperand;
13152	bool const fIsLongMode = CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx);
13153
13154	/*
13155	* Validate instruction information.
13156	* This shouldn't happen on real hardware but useful while testing our nested hardware-virtualization code.
13157	*/
13158	AssertLogRelMsgReturn(uAddrSize < RT_ELEMENTS(s_auAddrSizeMasks),
13159	("Invalid address size. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_1);
13160	AssertLogRelMsgReturn(iSegReg < X86_SREG_COUNT,
13161	("Invalid segment register. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_2);
13162	AssertLogRelMsgReturn(fIsMemOperand,
13163	("Expected memory operand. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_3);
13164
13165	/*
13166	* Compute the complete effective address.
13167	*
13168	* See AMD instruction spec. 1.4.2 "SIB Byte Format"
13169	* See AMD spec. 4.5.2 "Segment Registers".
13170	*/
13171	RTGCPTR GCPtrMem = GCPtrDisp;
13172	if (fBaseRegValid)
13173	GCPtrMem += pVCpu->cpum.GstCtx.aGRegs[iBaseReg].u64;
13174	if (fIdxRegValid)
13175	GCPtrMem += pVCpu->cpum.GstCtx.aGRegs[iIdxReg].u64 << uScale;
13176
13177	RTGCPTR const GCPtrOff = GCPtrMem;
13178	if ( !fIsLongMode
13179	\|\| iSegReg >= X86_SREG_FS)
13180	GCPtrMem += pVCpu->cpum.GstCtx.aSRegs[iSegReg].u64Base;
13181	GCPtrMem &= s_auAddrSizeMasks[uAddrSize];
13182
13183	/*
13184	* Validate effective address.
13185	* See AMD spec. 4.5.3 "Segment Registers in 64-Bit Mode".
13186	*/
13187	uint8_t const cbAccess = s_auAccessSizeMasks[uAddrSize];
13188	Assert(cbAccess > 0);
13189	if (fIsLongMode)
13190	{
13191	if (X86_IS_CANONICAL(GCPtrMem))
13192	{
13193	*pGCPtrMem = GCPtrMem;
13194	return VINF_SUCCESS;
13195	}
13196
13197	/** @todo r=ramshankar: We should probably raise \#SS or \#GP. See AMD spec. 4.12.2
13198	* "Data Limit Checks in 64-bit Mode". */
13199	Log4Func(("Long mode effective address is not canonical GCPtrMem=%#RX64\n", GCPtrMem));
13200	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13201	return VINF_HM_PENDING_XCPT;
13202	}
13203
13204	/*
13205	* This is a watered down version of iemMemApplySegment().
13206	* Parts that are not applicable for VMX instructions like real-or-v8086 mode
13207	* and segment CPL/DPL checks are skipped.
13208	*/
13209	RTGCPTR32 const GCPtrFirst32 = (RTGCPTR32)GCPtrOff;
13210	RTGCPTR32 const GCPtrLast32 = GCPtrFirst32 + cbAccess - 1;
13211	PCCPUMSELREG pSel = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
13212
13213	/* Check if the segment is present and usable. */
13214	if ( pSel->Attr.n.u1Present
13215	&& !pSel->Attr.n.u1Unusable)
13216	{
13217	Assert(pSel->Attr.n.u1DescType);
13218	if (!(pSel->Attr.n.u4Type & X86_SEL_TYPE_CODE))
13219	{
13220	/* Check permissions for the data segment. */
13221	if ( enmMemAccess == VMXMEMACCESS_WRITE
13222	&& !(pSel->Attr.n.u4Type & X86_SEL_TYPE_WRITE))
13223	{
13224	Log4Func(("Data segment access invalid. iSegReg=%#x Attr=%#RX32\n", iSegReg, pSel->Attr.u));
13225	hmR0VmxSetPendingXcptGP(pVCpu, iSegReg);
13226	return VINF_HM_PENDING_XCPT;
13227	}
13228
13229	/* Check limits if it's a normal data segment. */
13230	if (!(pSel->Attr.n.u4Type & X86_SEL_TYPE_DOWN))
13231	{
13232	if ( GCPtrFirst32 > pSel->u32Limit
13233	\|\| GCPtrLast32 > pSel->u32Limit)
13234	{
13235	Log4Func(("Data segment limit exceeded. "
13236	"iSegReg=%#x GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n", iSegReg, GCPtrFirst32,
13237	GCPtrLast32, pSel->u32Limit));
13238	if (iSegReg == X86_SREG_SS)
13239	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13240	else
13241	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13242	return VINF_HM_PENDING_XCPT;
13243	}
13244	}
13245	else
13246	{
13247	/* Check limits if it's an expand-down data segment.
13248	Note! The upper boundary is defined by the B bit, not the G bit! */
13249	if ( GCPtrFirst32 < pSel->u32Limit + UINT32_C(1)
13250	\|\| GCPtrLast32 > (pSel->Attr.n.u1DefBig ? UINT32_MAX : UINT32_C(0xffff)))
13251	{
13252	Log4Func(("Expand-down data segment limit exceeded. "
13253	"iSegReg=%#x GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n", iSegReg, GCPtrFirst32,
13254	GCPtrLast32, pSel->u32Limit));
13255	if (iSegReg == X86_SREG_SS)
13256	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13257	else
13258	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13259	return VINF_HM_PENDING_XCPT;
13260	}
13261	}
13262	}
13263	else
13264	{
13265	/* Check permissions for the code segment. */
13266	if ( enmMemAccess == VMXMEMACCESS_WRITE
13267	\|\| ( enmMemAccess == VMXMEMACCESS_READ
13268	&& !(pSel->Attr.n.u4Type & X86_SEL_TYPE_READ)))
13269	{
13270	Log4Func(("Code segment access invalid. Attr=%#RX32\n", pSel->Attr.u));
13271	Assert(!CPUMIsGuestInRealOrV86ModeEx(&pVCpu->cpum.GstCtx));
13272	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13273	return VINF_HM_PENDING_XCPT;
13274	}
13275
13276	/* Check limits for the code segment (normal/expand-down not applicable for code segments). */
13277	if ( GCPtrFirst32 > pSel->u32Limit
13278	\|\| GCPtrLast32 > pSel->u32Limit)
13279	{
13280	Log4Func(("Code segment limit exceeded. GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n",
13281	GCPtrFirst32, GCPtrLast32, pSel->u32Limit));
13282	if (iSegReg == X86_SREG_SS)
13283	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13284	else
13285	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13286	return VINF_HM_PENDING_XCPT;
13287	}
13288	}
13289	}
13290	else
13291	{
13292	Log4Func(("Not present or unusable segment. iSegReg=%#x Attr=%#RX32\n", iSegReg, pSel->Attr.u));
13293	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13294	return VINF_HM_PENDING_XCPT;
13295	}
13296
13297	*pGCPtrMem = GCPtrMem;
13298	return VINF_SUCCESS;
13299	}
13300	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13301
13302
13303	/**
13304	* VM-exit helper for LMSW.
13305	*/
13306	static VBOXSTRICTRC hmR0VmxExitLmsw(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint16_t uMsw, RTGCPTR GCPtrEffDst)
13307	{
13308	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13309	AssertRCReturn(rc, rc);
13310
13311	VBOXSTRICTRC rcStrict = IEMExecDecodedLmsw(pVCpu, cbInstr, uMsw, GCPtrEffDst);
13312	AssertMsg( rcStrict == VINF_SUCCESS
13313	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13314
13315	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
13316	if (rcStrict == VINF_IEM_RAISED_XCPT)
13317	{
13318	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13319	rcStrict = VINF_SUCCESS;
13320	}
13321
13322	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitLmsw);
13323	Log4Func(("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13324	return rcStrict;
13325	}
13326
13327
13328	/**
13329	* VM-exit helper for CLTS.
13330	*/
13331	static VBOXSTRICTRC hmR0VmxExitClts(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr)
13332	{
13333	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13334	AssertRCReturn(rc, rc);
13335
13336	VBOXSTRICTRC rcStrict = IEMExecDecodedClts(pVCpu, cbInstr);
13337	AssertMsg( rcStrict == VINF_SUCCESS
13338	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13339
13340	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
13341	if (rcStrict == VINF_IEM_RAISED_XCPT)
13342	{
13343	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13344	rcStrict = VINF_SUCCESS;
13345	}
13346
13347	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitClts);
13348	Log4Func(("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13349	return rcStrict;
13350	}
13351
13352
13353	/**
13354	* VM-exit helper for MOV from CRx (CRx read).
13355	*/
13356	static VBOXSTRICTRC hmR0VmxExitMovFromCrX(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint8_t iGReg, uint8_t iCrReg)
13357	{
13358	Assert(iCrReg < 16);
13359	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
13360
13361	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13362	AssertRCReturn(rc, rc);
13363
13364	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxRead(pVCpu, cbInstr, iGReg, iCrReg);
13365	AssertMsg( rcStrict == VINF_SUCCESS
13366	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13367
13368	if (iGReg == X86_GREG_xSP)
13369	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_RSP);
13370	else
13371	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
13372	#ifdef VBOX_WITH_STATISTICS
13373	switch (iCrReg)
13374	{
13375	case 0: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Read); break;
13376	case 2: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Read); break;
13377	case 3: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Read); break;
13378	case 4: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Read); break;
13379	case 8: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Read); break;
13380	}
13381	#endif
13382	Log4Func(("CR%d Read access rcStrict=%Rrc\n", iCrReg, VBOXSTRICTRC_VAL(rcStrict)));
13383	return rcStrict;
13384	}
13385
13386
13387	/**
13388	* VM-exit helper for MOV to CRx (CRx write).
13389	*/
13390	static VBOXSTRICTRC hmR0VmxExitMovToCrX(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint8_t iGReg, uint8_t iCrReg)
13391	{
13392	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13393	AssertRCReturn(rc, rc);
13394
13395	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxWrite(pVCpu, cbInstr, iCrReg, iGReg);
13396	AssertMsg( rcStrict == VINF_SUCCESS
13397	\|\| rcStrict == VINF_IEM_RAISED_XCPT
13398	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13399
13400	switch (iCrReg)
13401	{
13402	case 0:
13403	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
13404	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Write);
13405	Log4Func(("CR0 write. rcStrict=%Rrc CR0=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cr0));
13406	break;
13407
13408	case 2:
13409	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Write);
13410	/* Nothing to do here, CR2 it's not part of the VMCS. */
13411	break;
13412
13413	case 3:
13414	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR3);
13415	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Write);
13416	Log4Func(("CR3 write. rcStrict=%Rrc CR3=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cr3));
13417	break;
13418
13419	case 4:
13420	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR4);
13421	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Write);
13422	Log4Func(("CR4 write. rc=%Rrc CR4=%#RX64 fLoadSaveGuestXcr0=%u\n", VBOXSTRICTRC_VAL(rcStrict),
13423	pVCpu->cpum.GstCtx.cr4, pVCpu->hm.s.fLoadSaveGuestXcr0));
13424	break;
13425
13426	case 8:
13427	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged,
13428	HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_APIC_TPR);
13429	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Write);
13430	break;
13431
13432	default:
13433	AssertMsgFailed(("Invalid CRx register %#x\n", iCrReg));
13434	break;
13435	}
13436
13437	if (rcStrict == VINF_IEM_RAISED_XCPT)
13438	{
13439	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13440	rcStrict = VINF_SUCCESS;
13441	}
13442	return rcStrict;
13443	}
13444
13445
13446	/**
13447	* VM-exit exception handler for \#PF (Page-fault exception).
13448	*
13449	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13450	*/
13451	static VBOXSTRICTRC hmR0VmxExitXcptPF(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13452	{
13453	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13454	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
13455	hmR0VmxReadExitQualVmcs(pVmxTransient);
13456
13457	if (!pVM->hm.s.fNestedPaging)
13458	{ /* likely */ }
13459	else
13460	{
13461	#if !defined(HMVMX_ALWAYS_TRAP_ALL_XCPTS) && !defined(HMVMX_ALWAYS_TRAP_PF)
13462	Assert(pVmxTransient->fIsNestedGuest \|\| pVCpu->hm.s.fUsingDebugLoop);
13463	#endif
13464	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
13465	if (!pVmxTransient->fVectoringDoublePF)
13466	{
13467	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */,
13468	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual);
13469	}
13470	else
13471	{
13472	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
13473	Assert(!pVmxTransient->fIsNestedGuest);
13474	hmR0VmxSetPendingXcptDF(pVCpu);
13475	Log4Func(("Pending #DF due to vectoring #PF w/ NestedPaging\n"));
13476	}
13477	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
13478	return VINF_SUCCESS;
13479	}
13480
13481	Assert(!pVmxTransient->fIsNestedGuest);
13482
13483	/* If it's a vectoring #PF, emulate injecting the original event injection as PGMTrap0eHandler() is incapable
13484	of differentiating between instruction emulation and event injection that caused a #PF. See @bugref{6607}. */
13485	if (pVmxTransient->fVectoringPF)
13486	{
13487	Assert(pVCpu->hm.s.Event.fPending);
13488	return VINF_EM_RAW_INJECT_TRPM_EVENT;
13489	}
13490
13491	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
13492	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
13493	AssertRCReturn(rc, rc);
13494
13495	Log4Func(("#PF: cs:rip=%#04x:%#RX64 err_code=%#RX32 exit_qual=%#RX64 cr3=%#RX64\n", pCtx->cs.Sel, pCtx->rip,
13496	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual, pCtx->cr3));
13497
13498	TRPMAssertXcptPF(pVCpu, pVmxTransient->uExitQual, (RTGCUINT)pVmxTransient->uExitIntErrorCode);
13499	rc = PGMTrap0eHandler(pVCpu, pVmxTransient->uExitIntErrorCode, CPUMCTX2CORE(pCtx), (RTGCPTR)pVmxTransient->uExitQual);
13500
13501	Log4Func(("#PF: rc=%Rrc\n", rc));
13502	if (rc == VINF_SUCCESS)
13503	{
13504	/*
13505	* This is typically a shadow page table sync or a MMIO instruction. But we may have
13506	* emulated something like LTR or a far jump. Any part of the CPU context may have changed.
13507	*/
13508	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
13509	TRPMResetTrap(pVCpu);
13510	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
13511	return rc;
13512	}
13513
13514	if (rc == VINF_EM_RAW_GUEST_TRAP)
13515	{
13516	if (!pVmxTransient->fVectoringDoublePF)
13517	{
13518	/* It's a guest page fault and needs to be reflected to the guest. */
13519	uint32_t uGstErrorCode = TRPMGetErrorCode(pVCpu);
13520	TRPMResetTrap(pVCpu);
13521	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory #PF. */
13522	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */,
13523	uGstErrorCode, pVmxTransient->uExitQual);
13524	}
13525	else
13526	{
13527	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
13528	TRPMResetTrap(pVCpu);
13529	pVCpu->hm.s.Event.fPending = false; /* Clear pending #PF to replace it with #DF. */
13530	hmR0VmxSetPendingXcptDF(pVCpu);
13531	Log4Func(("#PF: Pending #DF due to vectoring #PF\n"));
13532	}
13533
13534	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
13535	return VINF_SUCCESS;
13536	}
13537
13538	TRPMResetTrap(pVCpu);
13539	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
13540	return rc;
13541	}
13542
13543
13544	/**
13545	* VM-exit exception handler for \#MF (Math Fault: floating point exception).
13546	*
13547	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13548	*/
13549	static VBOXSTRICTRC hmR0VmxExitXcptMF(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13550	{
13551	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13552	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
13553
13554	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR0);
13555	AssertRCReturn(rc, rc);
13556
13557	if (!(pVCpu->cpum.GstCtx.cr0 & X86_CR0_NE))
13558	{
13559	/* Convert a #MF into a FERR -> IRQ 13. See @bugref{6117}. */
13560	rc = PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13, 1, 0 /* uTagSrc */);
13561
13562	/** @todo r=ramshankar: The Intel spec. does -not- specify that this VM-exit
13563	* provides VM-exit instruction length. If this causes problem later,
13564	* disassemble the instruction like it's done on AMD-V. */
13565	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
13566	AssertRCReturn(rc2, rc2);
13567	return rc;
13568	}
13569
13570	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbExitInstr,
13571	pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13572	return VINF_SUCCESS;
13573	}
13574
13575
13576	/**
13577	* VM-exit exception handler for \#BP (Breakpoint exception).
13578	*
13579	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13580	*/
13581	static VBOXSTRICTRC hmR0VmxExitXcptBP(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13582	{
13583	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13584	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP);
13585
13586	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
13587	AssertRCReturn(rc, rc);
13588
13589	if (!pVmxTransient->fIsNestedGuest)
13590	rc = DBGFRZTrap03Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(&pVCpu->cpum.GstCtx));
13591	else
13592	rc = VINF_EM_RAW_GUEST_TRAP;
13593
13594	if (rc == VINF_EM_RAW_GUEST_TRAP)
13595	{
13596	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13597	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13598	rc = VINF_SUCCESS;
13599	}
13600
13601	Assert(rc == VINF_SUCCESS \|\| rc == VINF_EM_DBG_BREAKPOINT);
13602	return rc;
13603	}
13604
13605
13606	/**
13607	* VM-exit exception handler for \#AC (Alignment-check exception).
13608	*
13609	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13610	*/
13611	static VBOXSTRICTRC hmR0VmxExitXcptAC(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13612	{
13613	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13614	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC);
13615
13616	/* Re-inject it. We'll detect any nesting before getting here. */
13617	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13618	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13619	return VINF_SUCCESS;
13620	}
13621
13622
13623	/**
13624	* VM-exit exception handler for \#DB (Debug exception).
13625	*
13626	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13627	*/
13628	static VBOXSTRICTRC hmR0VmxExitXcptDB(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13629	{
13630	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13631	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
13632
13633	/*
13634	* Get the DR6-like values from the Exit qualification and pass it to DBGF for processing.
13635	*/
13636	hmR0VmxReadExitQualVmcs(pVmxTransient);
13637
13638	/* Refer Intel spec. Table 27-1. "Exit Qualifications for debug exceptions" for the format. */
13639	uint64_t const uDR6 = X86_DR6_INIT_VAL
13640	\| (pVmxTransient->uExitQual & ( X86_DR6_B0 \| X86_DR6_B1 \| X86_DR6_B2 \| X86_DR6_B3
13641	\| X86_DR6_BD \| X86_DR6_BS));
13642
13643	int rc;
13644	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
13645	if (!pVmxTransient->fIsNestedGuest)
13646	rc = DBGFRZTrap01Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx), uDR6, pVCpu->hm.s.fSingleInstruction);
13647	else
13648	rc = VINF_EM_RAW_GUEST_TRAP;
13649	Log6Func(("rc=%Rrc\n", rc));
13650	if (rc == VINF_EM_RAW_GUEST_TRAP)
13651	{
13652	/*
13653	* The exception was for the guest. Update DR6, DR7.GD and
13654	* IA32_DEBUGCTL.LBR before forwarding it.
13655	* See Intel spec. 27.1 "Architectural State before a VM-Exit".
13656	*/
13657	VMMRZCallRing3Disable(pVCpu);
13658	HM_DISABLE_PREEMPT(pVCpu);
13659
13660	pCtx->dr[6] &= ~X86_DR6_B_MASK;
13661	pCtx->dr[6] \|= uDR6;
13662	if (CPUMIsGuestDebugStateActive(pVCpu))
13663	ASMSetDR6(pCtx->dr[6]);
13664
13665	HM_RESTORE_PREEMPT();
13666	VMMRZCallRing3Enable(pVCpu);
13667
13668	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_DR7);
13669	AssertRCReturn(rc, rc);
13670
13671	/* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */
13672	pCtx->dr[7] &= ~(uint64_t)X86_DR7_GD;
13673
13674	/* Paranoia. */
13675	pCtx->dr[7] &= ~(uint64_t)X86_DR7_RAZ_MASK;
13676	pCtx->dr[7] \|= X86_DR7_RA1_MASK;
13677
13678	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, pCtx->dr[7]);
13679	AssertRC(rc);
13680
13681	/*
13682	* Raise #DB in the guest.
13683	*
13684	* It is important to reflect exactly what the VM-exit gave us (preserving the
13685	* interruption-type) rather than use hmR0VmxSetPendingXcptDB() as the #DB could've
13686	* been raised while executing ICEBP (INT1) and not the regular #DB. Thus it may
13687	* trigger different handling in the CPU (like skipping DPL checks), see @bugref{6398}.
13688	*
13689	* Intel re-documented ICEBP/INT1 on May 2018 previously documented as part of
13690	* Intel 386, see Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
13691	*/
13692	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13693	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13694	return VINF_SUCCESS;
13695	}
13696
13697	/*
13698	* Not a guest trap, must be a hypervisor related debug event then.
13699	* Update DR6 in case someone is interested in it.
13700	*/
13701	AssertMsg(rc == VINF_EM_DBG_STEPPED \|\| rc == VINF_EM_DBG_BREAKPOINT, ("%Rrc\n", rc));
13702	AssertReturn(pVmxTransient->fWasHyperDebugStateActive, VERR_HM_IPE_5);
13703	CPUMSetHyperDR6(pVCpu, uDR6);
13704
13705	return rc;
13706	}
13707
13708
13709	/**
13710	* Hacks its way around the lovely mesa driver's backdoor accesses.
13711	*
13712	* @sa hmR0SvmHandleMesaDrvGp.
13713	*/
13714	static int hmR0VmxHandleMesaDrvGp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PCPUMCTX pCtx)
13715	{
13716	LogFunc(("cs:rip=%#04x:%#RX64 rcx=%#RX64 rbx=%#RX64\n", pCtx->cs.Sel, pCtx->rip, pCtx->rcx, pCtx->rbx));
13717	RT_NOREF(pCtx);
13718
13719	/* For now we'll just skip the instruction. */
13720	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
13721	}
13722
13723
13724	/**
13725	* Checks if the \#GP'ing instruction is the mesa driver doing it's lovely
13726	* backdoor logging w/o checking what it is running inside.
13727	*
13728	* This recognizes an "IN EAX,DX" instruction executed in flat ring-3, with the
13729	* backdoor port and magic numbers loaded in registers.
13730	*
13731	* @returns true if it is, false if it isn't.
13732	* @sa hmR0SvmIsMesaDrvGp.
13733	*/
13734	DECLINLINE(bool) hmR0VmxIsMesaDrvGp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PCPUMCTX pCtx)
13735	{
13736	/* 0xed: IN eAX,dx */
13737	uint8_t abInstr[1];
13738	if (pVmxTransient->cbExitInstr != sizeof(abInstr))
13739	return false;
13740
13741	/* Check that it is #GP(0). */
13742	if (pVmxTransient->uExitIntErrorCode != 0)
13743	return false;
13744
13745	/* Check magic and port. */
13746	Assert(!(pCtx->fExtrn & (CPUMCTX_EXTRN_RAX \| CPUMCTX_EXTRN_RDX \| CPUMCTX_EXTRN_RCX)));
13747	/Log(("hmR0VmxIsMesaDrvGp: rax=%RX64 rdx=%RX64\n", pCtx->rax, pCtx->rdx));/
13748	if (pCtx->rax != UINT32_C(0x564d5868))
13749	return false;
13750	if (pCtx->dx != UINT32_C(0x5658))
13751	return false;
13752
13753	/* Flat ring-3 CS. */
13754	AssertCompile(HMVMX_CPUMCTX_EXTRN_ALL & CPUMCTX_EXTRN_CS);
13755	Assert(!(pCtx->fExtrn & CPUMCTX_EXTRN_CS));
13756	/Log(("hmR0VmxIsMesaDrvGp: cs.Attr.n.u2Dpl=%d base=%Rx64\n", pCtx->cs.Attr.n.u2Dpl, pCtx->cs.u64Base));/
13757	if (pCtx->cs.Attr.n.u2Dpl != 3)
13758	return false;
13759	if (pCtx->cs.u64Base != 0)
13760	return false;
13761
13762	/* Check opcode. */
13763	AssertCompile(HMVMX_CPUMCTX_EXTRN_ALL & CPUMCTX_EXTRN_RIP);
13764	Assert(!(pCtx->fExtrn & CPUMCTX_EXTRN_RIP));
13765	int rc = PGMPhysSimpleReadGCPtr(pVCpu, abInstr, pCtx->rip, sizeof(abInstr));
13766	/Log(("hmR0VmxIsMesaDrvGp: PGMPhysSimpleReadGCPtr -> %Rrc %#x\n", rc, abInstr[0]));/
13767	if (RT_FAILURE(rc))
13768	return false;
13769	if (abInstr[0] != 0xed)
13770	return false;
13771
13772	return true;
13773	}
13774
13775
13776	/**
13777	* VM-exit exception handler for \#GP (General-protection exception).
13778	*
13779	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13780	*/
13781	static VBOXSTRICTRC hmR0VmxExitXcptGP(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13782	{
13783	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13784	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP);
13785
13786	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
13787	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
13788	if (pVmcsInfo->RealMode.fRealOnV86Active)
13789	{ /* likely */ }
13790	else
13791	{
13792	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
13793	Assert(pVCpu->hm.s.fUsingDebugLoop \|\| pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv \|\| pVmxTransient->fIsNestedGuest);
13794	#endif
13795	/*
13796	* If the guest is not in real-mode or we have unrestricted guest execution support, or if we are
13797	* executing a nested-guest, reflect #GP to the guest or nested-guest.
13798	*/
13799	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
13800	AssertRCReturn(rc, rc);
13801	Log4Func(("Gst: cs:rip=%#04x:%#RX64 ErrorCode=%#x cr0=%#RX64 cpl=%u tr=%#04x\n", pCtx->cs.Sel, pCtx->rip,
13802	pVmxTransient->uExitIntErrorCode, pCtx->cr0, CPUMGetGuestCPL(pVCpu), pCtx->tr.Sel));
13803
13804	if ( pVmxTransient->fIsNestedGuest
13805	\|\| !pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv
13806	\|\| !hmR0VmxIsMesaDrvGp(pVCpu, pVmxTransient, pCtx))
13807	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13808	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13809	else
13810	rc = hmR0VmxHandleMesaDrvGp(pVCpu, pVmxTransient, pCtx);
13811	return rc;
13812	}
13813
13814	Assert(CPUMIsGuestInRealModeEx(pCtx));
13815	Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest);
13816	Assert(!pVmxTransient->fIsNestedGuest);
13817
13818	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
13819	AssertRCReturn(rc, rc);
13820
13821	VBOXSTRICTRC rcStrict = IEMExecOne(pVCpu);
13822	if (rcStrict == VINF_SUCCESS)
13823	{
13824	if (!CPUMIsGuestInRealModeEx(pCtx))
13825	{
13826	/*
13827	* The guest is no longer in real-mode, check if we can continue executing the
13828	* guest using hardware-assisted VMX. Otherwise, fall back to emulation.
13829	*/
13830	pVmcsInfo->RealMode.fRealOnV86Active = false;
13831	if (HMCanExecuteVmxGuest(pVCpu->pVMR0, pVCpu, pCtx))
13832	{
13833	Log4Func(("Mode changed but guest still suitable for executing using hardware-assisted VMX\n"));
13834	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
13835	}
13836	else
13837	{
13838	Log4Func(("Mode changed -> VINF_EM_RESCHEDULE\n"));
13839	rcStrict = VINF_EM_RESCHEDULE;
13840	}
13841	}
13842	else
13843	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
13844	}
13845	else if (rcStrict == VINF_IEM_RAISED_XCPT)
13846	{
13847	rcStrict = VINF_SUCCESS;
13848	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13849	}
13850	return VBOXSTRICTRC_VAL(rcStrict);
13851	}
13852
13853
13854	/**
13855	* VM-exit exception handler wrapper for all other exceptions that are not handled
13856	* by a specific handler.
13857	*
13858	* This simply re-injects the exception back into the VM without any special
13859	* processing.
13860	*
13861	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13862	*/
13863	static VBOXSTRICTRC hmR0VmxExitXcptOthers(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13864	{
13865	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13866
13867	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
13868	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
13869	AssertMsg(pVCpu->hm.s.fUsingDebugLoop \|\| pVmcsInfo->RealMode.fRealOnV86Active \|\| pVmxTransient->fIsNestedGuest,
13870	("uVector=%#x u32XcptBitmap=%#X32\n",
13871	VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo), pVmcsInfo->u32XcptBitmap));
13872	NOREF(pVmcsInfo);
13873	#endif
13874
13875	/*
13876	* Re-inject the exception into the guest. This cannot be a double-fault condition which
13877	* would have been handled while checking exits due to event delivery.
13878	*/
13879	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
13880
13881	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
13882	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
13883	AssertRCReturn(rc, rc);
13884	Log4Func(("Reinjecting Xcpt. uVector=%#x cs:rip=%#04x:%#RX64\n", uVector, pCtx->cs.Sel, pCtx->rip));
13885	#endif
13886
13887	#ifdef VBOX_WITH_STATISTICS
13888	switch (uVector)
13889	{
13890	case X86_XCPT_DE: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE); break;
13891	case X86_XCPT_DB: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB); break;
13892	case X86_XCPT_BP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP); break;
13893	case X86_XCPT_OF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
13894	case X86_XCPT_BR: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBR); break;
13895	case X86_XCPT_UD: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD); break;
13896	case X86_XCPT_NM: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
13897	case X86_XCPT_DF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDF); break;
13898	case X86_XCPT_TS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestTS); break;
13899	case X86_XCPT_NP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNP); break;
13900	case X86_XCPT_SS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestSS); break;
13901	case X86_XCPT_GP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP); break;
13902	case X86_XCPT_PF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF); break;
13903	case X86_XCPT_MF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF); break;
13904	case X86_XCPT_AC: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC); break;
13905	case X86_XCPT_XF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXF); break;
13906	default:
13907	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXcpUnk);
13908	break;
13909	}
13910	#endif
13911
13912	/* We should never call this function for a page-fault, we'd need to pass on the fault address below otherwise. */
13913	Assert(!VMX_EXIT_INT_INFO_IS_XCPT_PF(pVmxTransient->uExitIntInfo));
13914	NOREF(uVector);
13915
13916	/* Re-inject the original exception into the guest. */
13917	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13918	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13919	return VINF_SUCCESS;
13920	}
13921
13922
13923	/**
13924	* VM-exit exception handler for all exceptions (except NMIs!).
13925	*
13926	* @remarks This may be called for both guests and nested-guests. Take care to not
13927	* make assumptions and avoid doing anything that is not relevant when
13928	* executing a nested-guest (e.g., Mesa driver hacks).
13929	*/
13930	static VBOXSTRICTRC hmR0VmxExitXcpt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13931	{
13932	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_XCPT_INFO);
13933
13934	/*
13935	* If this VM-exit occurred while delivering an event through the guest IDT, take
13936	* action based on the return code and additional hints (e.g. for page-faults)
13937	* that will be updated in the VMX transient structure.
13938	*/
13939	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
13940	if (rcStrict == VINF_SUCCESS)
13941	{
13942	/*
13943	* If an exception caused a VM-exit due to delivery of an event, the original
13944	* event may have to be re-injected into the guest. We shall reinject it and
13945	* continue guest execution. However, page-fault is a complicated case and
13946	* needs additional processing done in hmR0VmxExitXcptPF().
13947	*/
13948	Assert(VMX_EXIT_INT_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
13949	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
13950	if ( !pVCpu->hm.s.Event.fPending
13951	\|\| uVector == X86_XCPT_PF)
13952	{
13953	switch (uVector)
13954	{
13955	case X86_XCPT_PF: return hmR0VmxExitXcptPF(pVCpu, pVmxTransient);
13956	case X86_XCPT_GP: return hmR0VmxExitXcptGP(pVCpu, pVmxTransient);
13957	case X86_XCPT_MF: return hmR0VmxExitXcptMF(pVCpu, pVmxTransient);
13958	case X86_XCPT_DB: return hmR0VmxExitXcptDB(pVCpu, pVmxTransient);
13959	case X86_XCPT_BP: return hmR0VmxExitXcptBP(pVCpu, pVmxTransient);
13960	case X86_XCPT_AC: return hmR0VmxExitXcptAC(pVCpu, pVmxTransient);
13961	default:
13962	return hmR0VmxExitXcptOthers(pVCpu, pVmxTransient);
13963	}
13964	}
13965	/* else: inject pending event before resuming guest execution. */
13966	}
13967	else if (rcStrict == VINF_HM_DOUBLE_FAULT)
13968	{
13969	Assert(pVCpu->hm.s.Event.fPending);
13970	rcStrict = VINF_SUCCESS;
13971	}
13972
13973	return rcStrict;
13974	}
13975	/** @} */
13976
13977
13978	/** @name VM-exit handlers.
13979	* @{
13980	*/
13981	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13982	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
13983	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13984
13985	/**
13986	* VM-exit handler for external interrupts (VMX_EXIT_EXT_INT).
13987	*/
13988	HMVMX_EXIT_DECL hmR0VmxExitExtInt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13989	{
13990	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13991	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt);
13992	/* Windows hosts (32-bit and 64-bit) have DPC latency issues. See @bugref{6853}. */
13993	if (VMMR0ThreadCtxHookIsEnabled(pVCpu))
13994	return VINF_SUCCESS;
13995	return VINF_EM_RAW_INTERRUPT;
13996	}
13997
13998
13999	/**
14000	* VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI). Conditional
14001	* VM-exit.
14002	*/
14003	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmi(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14004	{
14005	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14006	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitXcptNmi, y3);
14007
14008	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
14009
14010	uint32_t const uExitIntType = VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo);
14011	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14012	Assert(VMX_EXIT_INT_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
14013
14014	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14015	Assert( !(pVmcsInfo->u32ExitCtls & VMX_EXIT_CTLS_ACK_EXT_INT)
14016	&& uExitIntType != VMX_EXIT_INT_INFO_TYPE_EXT_INT);
14017	NOREF(pVmcsInfo);
14018
14019	VBOXSTRICTRC rcStrict;
14020	switch (uExitIntType)
14021	{
14022	/*
14023	* Host physical NMIs:
14024	* This cannot be a guest NMI as the only way for the guest to receive an NMI is if we
14025	* injected it ourselves and anything we inject is not going to cause a VM-exit directly
14026	* for the event being injected[1]. Go ahead and dispatch the NMI to the host[2].
14027	*
14028	* See Intel spec. 27.2.3 "Information for VM Exits During Event Delivery".
14029	* See Intel spec. 27.5.5 "Updating Non-Register State".
14030	*/
14031	case VMX_EXIT_INT_INFO_TYPE_NMI:
14032	{
14033	rcStrict = hmR0VmxExitHostNmi(pVCpu, pVmcsInfo);
14034	break;
14035	}
14036
14037	/*
14038	* Privileged software exceptions (#DB from ICEBP),
14039	* Software exceptions (#BP and #OF),
14040	* Hardware exceptions:
14041	* Process the required exceptions and resume guest execution if possible.
14042	*/
14043	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
14044	Assert(uVector == X86_XCPT_DB);
14045	RT_FALL_THRU();
14046	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
14047	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF \|\| uExitIntType == VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT);
14048	RT_FALL_THRU();
14049	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
14050	{
14051	NOREF(uVector);
14052	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
14053	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14054	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
14055	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
14056
14057	rcStrict = hmR0VmxExitXcpt(pVCpu, pVmxTransient);
14058	break;
14059	}
14060
14061	default:
14062	{
14063	pVCpu->hm.s.u32HMError = pVmxTransient->uExitIntInfo;
14064	rcStrict = VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
14065	AssertMsgFailed(("Invalid/unexpected VM-exit interruption info %#x\n", pVmxTransient->uExitIntInfo));
14066	break;
14067	}
14068	}
14069
14070	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
14071	return rcStrict;
14072	}
14073
14074
14075	/**
14076	* VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
14077	*/
14078	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindow(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14079	{
14080	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14081
14082	/* Indicate that we no longer need to VM-exit when the guest is ready to receive interrupts, it is now ready. */
14083	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14084	hmR0VmxClearIntWindowExitVmcs(pVmcsInfo);
14085
14086	/* Evaluate and deliver pending events and resume guest execution. */
14087	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow);
14088	return VINF_SUCCESS;
14089	}
14090
14091
14092	/**
14093	* VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
14094	*/
14095	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindow(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14096	{
14097	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14098
14099	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14100	if (RT_UNLIKELY(!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))) /** @todo NSTVMX: Turn this into an assertion. */
14101	{
14102	AssertMsgFailed(("Unexpected NMI-window exit.\n"));
14103	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
14104	}
14105
14106	Assert(!CPUMIsGuestNmiBlocking(pVCpu));
14107
14108	/*
14109	* If block-by-STI is set when we get this VM-exit, it means the CPU doesn't block NMIs following STI.
14110	* It is therefore safe to unblock STI and deliver the NMI ourselves. See @bugref{7445}.
14111	*/
14112	uint32_t fIntrState;
14113	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
14114	AssertRC(rc);
14115	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
14116	if (fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
14117	{
14118	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
14119	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
14120
14121	fIntrState &= ~VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
14122	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
14123	AssertRC(rc);
14124	}
14125
14126	/* Indicate that we no longer need to VM-exit when the guest is ready to receive NMIs, it is now ready */
14127	hmR0VmxClearNmiWindowExitVmcs(pVmcsInfo);
14128
14129	/* Evaluate and deliver pending events and resume guest execution. */
14130	return VINF_SUCCESS;
14131	}
14132
14133
14134	/**
14135	* VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
14136	*/
14137	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvd(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14138	{
14139	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14140	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14141	}
14142
14143
14144	/**
14145	* VM-exit handler for INVD (VMX_EXIT_INVD). Unconditional VM-exit.
14146	*/
14147	HMVMX_EXIT_NSRC_DECL hmR0VmxExitInvd(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14148	{
14149	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14150	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14151	}
14152
14153
14154	/**
14155	* VM-exit handler for CPUID (VMX_EXIT_CPUID). Unconditional VM-exit.
14156	*/
14157	HMVMX_EXIT_DECL hmR0VmxExitCpuid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14158	{
14159	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14160
14161	/*
14162	* Get the state we need and update the exit history entry.
14163	*/
14164	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14165	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14166
14167	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
14168	AssertRCReturn(rc, rc);
14169
14170	VBOXSTRICTRC rcStrict;
14171	PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
14172	EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_CPUID),
14173	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
14174	if (!pExitRec)
14175	{
14176	/*
14177	* Regular CPUID instruction execution.
14178	*/
14179	rcStrict = IEMExecDecodedCpuid(pVCpu, pVmxTransient->cbExitInstr);
14180	if (rcStrict == VINF_SUCCESS)
14181	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14182	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14183	{
14184	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14185	rcStrict = VINF_SUCCESS;
14186	}
14187	}
14188	else
14189	{
14190	/*
14191	* Frequent exit or something needing probing. Get state and call EMHistoryExec.
14192	*/
14193	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14194	AssertRCReturn(rc2, rc2);
14195
14196	Log4(("CpuIdExit/%u: %04x:%08RX64: %#x/%#x -> EMHistoryExec\n",
14197	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.eax, pVCpu->cpum.GstCtx.ecx));
14198
14199	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
14200	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14201
14202	Log4(("CpuIdExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
14203	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
14204	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
14205	}
14206	return rcStrict;
14207	}
14208
14209
14210	/**
14211	* VM-exit handler for GETSEC (VMX_EXIT_GETSEC). Unconditional VM-exit.
14212	*/
14213	HMVMX_EXIT_DECL hmR0VmxExitGetsec(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14214	{
14215	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14216
14217	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14218	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR4);
14219	AssertRCReturn(rc, rc);
14220
14221	if (pVCpu->cpum.GstCtx.cr4 & X86_CR4_SMXE)
14222	return VINF_EM_RAW_EMULATE_INSTR;
14223
14224	AssertMsgFailed(("hmR0VmxExitGetsec: Unexpected VM-exit when CR4.SMXE is 0.\n"));
14225	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
14226	}
14227
14228
14229	/**
14230	* VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
14231	*/
14232	HMVMX_EXIT_DECL hmR0VmxExitRdtsc(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14233	{
14234	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14235
14236	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14237	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14238	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14239	AssertRCReturn(rc, rc);
14240
14241	VBOXSTRICTRC rcStrict = IEMExecDecodedRdtsc(pVCpu, pVmxTransient->cbExitInstr);
14242	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
14243	{
14244	/* If we get a spurious VM-exit when TSC offsetting is enabled,
14245	we must reset offsetting on VM-entry. See @bugref{6634}. */
14246	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TSC_OFFSETTING)
14247	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14248	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14249	}
14250	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14251	{
14252	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14253	rcStrict = VINF_SUCCESS;
14254	}
14255	return rcStrict;
14256	}
14257
14258
14259	/**
14260	* VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
14261	*/
14262	HMVMX_EXIT_DECL hmR0VmxExitRdtscp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14263	{
14264	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14265
14266	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14267	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14268	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_TSC_AUX);
14269	AssertRCReturn(rc, rc);
14270
14271	VBOXSTRICTRC rcStrict = IEMExecDecodedRdtscp(pVCpu, pVmxTransient->cbExitInstr);
14272	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
14273	{
14274	/* If we get a spurious VM-exit when TSC offsetting is enabled,
14275	we must reset offsetting on VM-reentry. See @bugref{6634}. */
14276	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TSC_OFFSETTING)
14277	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14278	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14279	}
14280	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14281	{
14282	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14283	rcStrict = VINF_SUCCESS;
14284	}
14285	return rcStrict;
14286	}
14287
14288
14289	/**
14290	* VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
14291	*/
14292	HMVMX_EXIT_DECL hmR0VmxExitRdpmc(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14293	{
14294	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14295
14296	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14297	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR4 \| CPUMCTX_EXTRN_CR0
14298	\| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS);
14299	AssertRCReturn(rc, rc);
14300
14301	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14302	rc = EMInterpretRdpmc(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx));
14303	if (RT_LIKELY(rc == VINF_SUCCESS))
14304	{
14305	rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14306	Assert(pVmxTransient->cbExitInstr == 2);
14307	}
14308	else
14309	{
14310	AssertMsgFailed(("hmR0VmxExitRdpmc: EMInterpretRdpmc failed with %Rrc\n", rc));
14311	rc = VERR_EM_INTERPRETER;
14312	}
14313	return rc;
14314	}
14315
14316
14317	/**
14318	* VM-exit handler for VMCALL (VMX_EXIT_VMCALL). Unconditional VM-exit.
14319	*/
14320	HMVMX_EXIT_DECL hmR0VmxExitVmcall(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14321	{
14322	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14323
14324	VBOXSTRICTRC rcStrict = VERR_VMX_IPE_3;
14325	if (EMAreHypercallInstructionsEnabled(pVCpu))
14326	{
14327	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14328	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_CR0
14329	\| CPUMCTX_EXTRN_SS \| CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_EFER);
14330	AssertRCReturn(rc, rc);
14331
14332	/* Perform the hypercall. */
14333	rcStrict = GIMHypercall(pVCpu, &pVCpu->cpum.GstCtx);
14334	if (rcStrict == VINF_SUCCESS)
14335	{
14336	rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14337	AssertRCReturn(rc, rc);
14338	}
14339	else
14340	Assert( rcStrict == VINF_GIM_R3_HYPERCALL
14341	\|\| rcStrict == VINF_GIM_HYPERCALL_CONTINUING
14342	\|\| RT_FAILURE(rcStrict));
14343
14344	/* If the hypercall changes anything other than guest's general-purpose registers,
14345	we would need to reload the guest changed bits here before VM-entry. */
14346	}
14347	else
14348	Log4Func(("Hypercalls not enabled\n"));
14349
14350	/* If hypercalls are disabled or the hypercall failed for some reason, raise #UD and continue. */
14351	if (RT_FAILURE(rcStrict))
14352	{
14353	hmR0VmxSetPendingXcptUD(pVCpu);
14354	rcStrict = VINF_SUCCESS;
14355	}
14356
14357	return rcStrict;
14358	}
14359
14360
14361	/**
14362	* VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
14363	*/
14364	HMVMX_EXIT_DECL hmR0VmxExitInvlpg(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14365	{
14366	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14367	Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging \|\| pVCpu->hm.s.fUsingDebugLoop);
14368
14369	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14370	hmR0VmxReadExitQualVmcs(pVmxTransient);
14371	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14372	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
14373	AssertRCReturn(rc, rc);
14374
14375	VBOXSTRICTRC rcStrict = IEMExecDecodedInvlpg(pVCpu, pVmxTransient->cbExitInstr, pVmxTransient->uExitQual);
14376
14377	if (rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_PGM_SYNC_CR3)
14378	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14379	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14380	{
14381	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14382	rcStrict = VINF_SUCCESS;
14383	}
14384	else
14385	AssertMsgFailed(("Unexpected IEMExecDecodedInvlpg(%#RX64) status: %Rrc\n", pVmxTransient->uExitQual,
14386	VBOXSTRICTRC_VAL(rcStrict)));
14387	return rcStrict;
14388	}
14389
14390
14391	/**
14392	* VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
14393	*/
14394	HMVMX_EXIT_DECL hmR0VmxExitMonitor(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14395	{
14396	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14397
14398	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14399	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14400	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK \| CPUMCTX_EXTRN_DS);
14401	AssertRCReturn(rc, rc);
14402
14403	VBOXSTRICTRC rcStrict = IEMExecDecodedMonitor(pVCpu, pVmxTransient->cbExitInstr);
14404	if (rcStrict == VINF_SUCCESS)
14405	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14406	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14407	{
14408	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14409	rcStrict = VINF_SUCCESS;
14410	}
14411
14412	return rcStrict;
14413	}
14414
14415
14416	/**
14417	* VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
14418	*/
14419	HMVMX_EXIT_DECL hmR0VmxExitMwait(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14420	{
14421	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14422
14423	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14424	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14425	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
14426	AssertRCReturn(rc, rc);
14427
14428	VBOXSTRICTRC rcStrict = IEMExecDecodedMwait(pVCpu, pVmxTransient->cbExitInstr);
14429	if (RT_SUCCESS(rcStrict))
14430	{
14431	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14432	if (EMMonitorWaitShouldContinue(pVCpu, &pVCpu->cpum.GstCtx))
14433	rcStrict = VINF_SUCCESS;
14434	}
14435
14436	return rcStrict;
14437	}
14438
14439
14440	/**
14441	* VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT). Unconditional
14442	* VM-exit.
14443	*/
14444	HMVMX_EXIT_DECL hmR0VmxExitTripleFault(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14445	{
14446	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14447	return VINF_EM_RESET;
14448	}
14449
14450
14451	/**
14452	* VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
14453	*/
14454	HMVMX_EXIT_DECL hmR0VmxExitHlt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14455	{
14456	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14457
14458	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14459	AssertRCReturn(rc, rc);
14460
14461	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS); /* Advancing the RIP above should've imported eflags. */
14462	if (EMShouldContinueAfterHalt(pVCpu, &pVCpu->cpum.GstCtx)) /* Requires eflags. */
14463	rc = VINF_SUCCESS;
14464	else
14465	rc = VINF_EM_HALT;
14466
14467	if (rc != VINF_SUCCESS)
14468	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHltToR3);
14469	return rc;
14470	}
14471
14472
14473	/**
14474	* VM-exit handler for instructions that result in a \#UD exception delivered to
14475	* the guest.
14476	*/
14477	HMVMX_EXIT_NSRC_DECL hmR0VmxExitSetPendingXcptUD(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14478	{
14479	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14480	hmR0VmxSetPendingXcptUD(pVCpu);
14481	return VINF_SUCCESS;
14482	}
14483
14484
14485	/**
14486	* VM-exit handler for expiry of the VMX-preemption timer.
14487	*/
14488	HMVMX_EXIT_DECL hmR0VmxExitPreemptTimer(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14489	{
14490	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14491
14492	/* If the VMX-preemption timer has expired, reinitialize the preemption timer on next VM-entry. */
14493	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14494
14495	/* If there are any timer events pending, fall back to ring-3, otherwise resume guest execution. */
14496	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
14497	bool fTimersPending = TMTimerPollBool(pVM, pVCpu);
14498	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPreemptTimer);
14499	return fTimersPending ? VINF_EM_RAW_TIMER_PENDING : VINF_SUCCESS;
14500	}
14501
14502
14503	/**
14504	* VM-exit handler for XSETBV (VMX_EXIT_XSETBV). Unconditional VM-exit.
14505	*/
14506	HMVMX_EXIT_DECL hmR0VmxExitXsetbv(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14507	{
14508	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14509
14510	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14511	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14512	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_CR4);
14513	AssertRCReturn(rc, rc);
14514
14515	VBOXSTRICTRC rcStrict = IEMExecDecodedXsetbv(pVCpu, pVmxTransient->cbExitInstr);
14516	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, rcStrict != VINF_IEM_RAISED_XCPT ? HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
14517	: HM_CHANGED_RAISED_XCPT_MASK);
14518
14519	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14520	pVCpu->hm.s.fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
14521
14522	return rcStrict;
14523	}
14524
14525
14526	/**
14527	* VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
14528	*/
14529	HMVMX_EXIT_DECL hmR0VmxExitInvpcid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14530	{
14531	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14532	/** @todo Use VM-exit instruction information. */
14533	return VERR_EM_INTERPRETER;
14534	}
14535
14536
14537	/**
14538	* VM-exit handler for invalid-guest-state (VMX_EXIT_ERR_INVALID_GUEST_STATE). Error
14539	* VM-exit.
14540	*/
14541	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrInvalidGuestState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14542	{
14543	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14544	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14545	AssertRCReturn(rc, rc);
14546
14547	rc = hmR0VmxCheckVmcsCtls(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest);
14548	if (RT_FAILURE(rc))
14549	return rc;
14550
14551	uint32_t const uInvalidReason = hmR0VmxCheckGuestState(pVCpu, pVmcsInfo);
14552	NOREF(uInvalidReason);
14553
14554	#ifdef VBOX_STRICT
14555	uint32_t fIntrState;
14556	uint64_t u64Val;
14557	hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
14558	hmR0VmxReadEntryXcptErrorCodeVmcs(pVmxTransient);
14559	hmR0VmxReadEntryInstrLenVmcs(pVmxTransient);
14560
14561	Log4(("uInvalidReason %u\n", uInvalidReason));
14562	Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", pVmxTransient->uEntryIntInfo));
14563	Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", pVmxTransient->uEntryXcptErrorCode));
14564	Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %#RX32\n", pVmxTransient->cbEntryInstr));
14565
14566	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState); AssertRC(rc);
14567	Log4(("VMX_VMCS32_GUEST_INT_STATE %#RX32\n", fIntrState));
14568	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64Val); AssertRC(rc);
14569	Log4(("VMX_VMCS_GUEST_CR0 %#RX64\n", u64Val));
14570	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_MASK, &u64Val); AssertRC(rc);
14571	Log4(("VMX_VMCS_CTRL_CR0_MASK %#RX64\n", u64Val));
14572	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, &u64Val); AssertRC(rc);
14573	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RX64\n", u64Val));
14574	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_MASK, &u64Val); AssertRC(rc);
14575	Log4(("VMX_VMCS_CTRL_CR4_MASK %#RX64\n", u64Val));
14576	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, &u64Val); AssertRC(rc);
14577	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RX64\n", u64Val));
14578	if (pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging)
14579	{
14580	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
14581	Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
14582	}
14583	hmR0DumpRegs(pVCpu, HM_DUMP_REG_FLAGS_ALL);
14584	#endif
14585
14586	return VERR_VMX_INVALID_GUEST_STATE;
14587	}
14588
14589	/**
14590	* VM-exit handler for all undefined/unexpected reasons. Should never happen.
14591	*/
14592	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrUnexpected(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14593	{
14594	/*
14595	* Cummulative notes of all recognized but unexpected VM-exits.
14596	*
14597	* 1. This does -not- cover scenarios like like a page-fault VM-exit occurring when
14598	* nested-paging is used.
14599	*
14600	* 2. Any instruction that causes a VM-exit unconditionally (for e.g. VMXON) must be
14601	* emulated or a #UD must be raised in the guest. Therefore, we should -not- be using
14602	* this function (and thereby stop VM execution) for handling such instructions.
14603	*
14604	*
14605	* VMX_EXIT_INIT_SIGNAL:
14606	* INIT signals are blocked in VMX root operation by VMXON and by SMI in SMM.
14607	* It is -NOT- blocked in VMX non-root operation so we can, in theory, still get these
14608	* VM-exits. However, we should not receive INIT signals VM-exit while executing a VM.
14609	*
14610	* See Intel spec. 33.14.1 Default Treatment of SMI Delivery"
14611	* See Intel spec. 29.3 "VMX Instructions" for "VMXON".
14612	* See Intel spec. "23.8 Restrictions on VMX operation".
14613	*
14614	* VMX_EXIT_SIPI:
14615	* SIPI exits can only occur in VMX non-root operation when the "wait-for-SIPI" guest
14616	* activity state is used. We don't make use of it as our guests don't have direct
14617	* access to the host local APIC.
14618	*
14619	* See Intel spec. 25.3 "Other Causes of VM-exits".
14620	*
14621	* VMX_EXIT_IO_SMI:
14622	* VMX_EXIT_SMI:
14623	* This can only happen if we support dual-monitor treatment of SMI, which can be
14624	* activated by executing VMCALL in VMX root operation. Only an STM (SMM transfer
14625	* monitor) would get this VM-exit when we (the executive monitor) execute a VMCALL in
14626	* VMX root mode or receive an SMI. If we get here, something funny is going on.
14627	*
14628	* See Intel spec. 33.15.6 "Activating the Dual-Monitor Treatment"
14629	* See Intel spec. 25.3 "Other Causes of VM-Exits"
14630	*
14631	* VMX_EXIT_ERR_MSR_LOAD:
14632	* Failures while loading MSRs are part of the VM-entry MSR-load area are unexpected
14633	* and typically indicates a bug in the hypervisor code. We thus cannot not resume
14634	* execution.
14635	*
14636	* See Intel spec. 26.7 "VM-Entry Failures During Or After Loading Guest State".
14637	*
14638	* VMX_EXIT_ERR_MACHINE_CHECK:
14639	* Machine check exceptions indicates a fatal/unrecoverable hardware condition
14640	* including but not limited to system bus, ECC, parity, cache and TLB errors. A
14641	* #MC exception abort class exception is raised. We thus cannot assume a
14642	* reasonable chance of continuing any sort of execution and we bail.
14643	*
14644	* See Intel spec. 15.1 "Machine-check Architecture".
14645	* See Intel spec. 27.1 "Architectural State Before A VM Exit".
14646	*
14647	* VMX_EXIT_PML_FULL:
14648	* VMX_EXIT_VIRTUALIZED_EOI:
14649	* VMX_EXIT_APIC_WRITE:
14650	* We do not currently support any of these features and thus they are all unexpected
14651	* VM-exits.
14652	*
14653	* VMX_EXIT_GDTR_IDTR_ACCESS:
14654	* VMX_EXIT_LDTR_TR_ACCESS:
14655	* VMX_EXIT_RDRAND:
14656	* VMX_EXIT_RSM:
14657	* VMX_EXIT_VMFUNC:
14658	* VMX_EXIT_ENCLS:
14659	* VMX_EXIT_RDSEED:
14660	* VMX_EXIT_XSAVES:
14661	* VMX_EXIT_XRSTORS:
14662	* VMX_EXIT_UMWAIT:
14663	* VMX_EXIT_TPAUSE:
14664	* These VM-exits are -not- caused unconditionally by execution of the corresponding
14665	* instruction. Any VM-exit for these instructions indicate a hardware problem,
14666	* unsupported CPU modes (like SMM) or potentially corrupt VMCS controls.
14667	*
14668	* See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally".
14669	*/
14670	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14671	AssertMsgFailed(("Unexpected VM-exit %u\n", pVmxTransient->uExitReason));
14672	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
14673	}
14674
14675
14676	/**
14677	* VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
14678	*/
14679	HMVMX_EXIT_DECL hmR0VmxExitRdmsr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14680	{
14681	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14682
14683	/** @todo Optimize this: We currently drag in in the whole MSR state
14684	* (CPUMCTX_EXTRN_ALL_MSRS) here. We should optimize this to only get
14685	* MSRs required. That would require changes to IEM and possibly CPUM too.
14686	* (Should probably do it lazy fashion from CPUMAllMsrs.cpp). */
14687	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14688	uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
14689	uint64_t fImport = IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK \| CPUMCTX_EXTRN_ALL_MSRS;
14690	switch (idMsr)
14691	{
14692	case MSR_K8_FS_BASE: fImport \|= CPUMCTX_EXTRN_FS; break;
14693	case MSR_K8_GS_BASE: fImport \|= CPUMCTX_EXTRN_GS; break;
14694	}
14695
14696	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14697	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImport);
14698	AssertRCReturn(rc, rc);
14699
14700	Log4Func(("ecx=%#RX32\n", idMsr));
14701
14702	#ifdef VBOX_STRICT
14703	Assert(!pVmxTransient->fIsNestedGuest);
14704	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
14705	{
14706	if ( hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr)
14707	&& idMsr != MSR_K6_EFER)
14708	{
14709	AssertMsgFailed(("Unexpected RDMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n", idMsr));
14710	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
14711	}
14712	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
14713	{
14714	Assert(pVmcsInfo->pvMsrBitmap);
14715	uint32_t fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, idMsr);
14716	if (fMsrpm & VMXMSRPM_ALLOW_RD)
14717	{
14718	AssertMsgFailed(("Unexpected RDMSR for a passthru lazy-restore MSR. ecx=%#RX32\n", idMsr));
14719	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
14720	}
14721	}
14722	}
14723	#endif
14724
14725	VBOXSTRICTRC rcStrict = IEMExecDecodedRdmsr(pVCpu, pVmxTransient->cbExitInstr);
14726	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr);
14727	if (rcStrict == VINF_SUCCESS)
14728	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
14729	\| HM_CHANGED_GUEST_RAX \| HM_CHANGED_GUEST_RDX);
14730	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14731	{
14732	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14733	rcStrict = VINF_SUCCESS;
14734	}
14735	else
14736	AssertMsg(rcStrict == VINF_CPUM_R3_MSR_READ, ("Unexpected IEMExecDecodedRdmsr rc (%Rrc)\n", VBOXSTRICTRC_VAL(rcStrict)));
14737
14738	return rcStrict;
14739	}
14740
14741
14742	/**
14743	* VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
14744	*/
14745	HMVMX_EXIT_DECL hmR0VmxExitWrmsr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14746	{
14747	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14748
14749	/** @todo Optimize this: We currently drag in in the whole MSR state
14750	* (CPUMCTX_EXTRN_ALL_MSRS) here. We should optimize this to only get
14751	* MSRs required. That would require changes to IEM and possibly CPUM too.
14752	* (Should probably do it lazy fashion from CPUMAllMsrs.cpp). */
14753	uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
14754	uint64_t fImport = IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK \| CPUMCTX_EXTRN_ALL_MSRS;
14755
14756	/*
14757	* The FS and GS base MSRs are not part of the above all-MSRs mask.
14758	* Although we don't need to fetch the base as it will be overwritten shortly, while
14759	* loading guest-state we would also load the entire segment register including limit
14760	* and attributes and thus we need to load them here.
14761	*/
14762	switch (idMsr)
14763	{
14764	case MSR_K8_FS_BASE: fImport \|= CPUMCTX_EXTRN_FS; break;
14765	case MSR_K8_GS_BASE: fImport \|= CPUMCTX_EXTRN_GS; break;
14766	}
14767
14768	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14769	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14770	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImport);
14771	AssertRCReturn(rc, rc);
14772
14773	Log4Func(("ecx=%#RX32 edx:eax=%#RX32:%#RX32\n", idMsr, pVCpu->cpum.GstCtx.edx, pVCpu->cpum.GstCtx.eax));
14774
14775	VBOXSTRICTRC rcStrict = IEMExecDecodedWrmsr(pVCpu, pVmxTransient->cbExitInstr);
14776	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr);
14777
14778	if (rcStrict == VINF_SUCCESS)
14779	{
14780	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14781
14782	/* If this is an X2APIC WRMSR access, update the APIC state as well. */
14783	if ( idMsr == MSR_IA32_APICBASE
14784	\|\| ( idMsr >= MSR_IA32_X2APIC_START
14785	&& idMsr <= MSR_IA32_X2APIC_END))
14786	{
14787	/*
14788	* We've already saved the APIC related guest-state (TPR) in post-run phase.
14789	* When full APIC register virtualization is implemented we'll have to make
14790	* sure APIC state is saved from the VMCS before IEM changes it.
14791	*/
14792	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
14793	}
14794	else if (idMsr == MSR_IA32_TSC) /* Windows 7 does this during bootup. See @bugref{6398}. */
14795	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14796	else if (idMsr == MSR_K6_EFER)
14797	{
14798	/*
14799	* If the guest touches the EFER MSR we need to update the VM-Entry and VM-Exit controls
14800	* as well, even if it is -not- touching bits that cause paging mode changes (LMA/LME).
14801	* We care about the other bits as well, SCE and NXE. See @bugref{7368}.
14802	*/
14803	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_EFER_MSR \| HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
14804	}
14805
14806	/* Update MSRs that are part of the VMCS and auto-load/store area when MSR-bitmaps are not used. */
14807	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS))
14808	{
14809	switch (idMsr)
14810	{
14811	case MSR_IA32_SYSENTER_CS: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_CS_MSR); break;
14812	case MSR_IA32_SYSENTER_EIP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); break;
14813	case MSR_IA32_SYSENTER_ESP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); break;
14814	case MSR_K8_FS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_FS); break;
14815	case MSR_K8_GS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_GS); break;
14816	case MSR_K6_EFER: /* Nothing to do, already handled above. */ break;
14817	default:
14818	{
14819	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
14820	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_VMX_GUEST_LAZY_MSRS);
14821	else if (hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr))
14822	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
14823	break;
14824	}
14825	}
14826	}
14827	#ifdef VBOX_STRICT
14828	else
14829	{
14830	/* Paranoia. Validate that MSRs in the MSR-bitmaps with write-passthru are not intercepted. */
14831	switch (idMsr)
14832	{
14833	case MSR_IA32_SYSENTER_CS:
14834	case MSR_IA32_SYSENTER_EIP:
14835	case MSR_IA32_SYSENTER_ESP:
14836	case MSR_K8_FS_BASE:
14837	case MSR_K8_GS_BASE:
14838	{
14839	AssertMsgFailed(("Unexpected WRMSR for an MSR in the VMCS. ecx=%#RX32\n", idMsr));
14840	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
14841	}
14842
14843	/* Writes to MSRs in auto-load/store area/swapped MSRs, shouldn't cause VM-exits with MSR-bitmaps. */
14844	default:
14845	{
14846	if (hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr))
14847	{
14848	/* EFER MSR writes are always intercepted. */
14849	if (idMsr != MSR_K6_EFER)
14850	{
14851	AssertMsgFailed(("Unexpected WRMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n",
14852	idMsr));
14853	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
14854	}
14855	}
14856
14857	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
14858	{
14859	Assert(pVmcsInfo->pvMsrBitmap);
14860	uint32_t fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, idMsr);
14861	if (fMsrpm & VMXMSRPM_ALLOW_WR)
14862	{
14863	AssertMsgFailed(("Unexpected WRMSR for passthru, lazy-restore MSR. ecx=%#RX32\n", idMsr));
14864	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
14865	}
14866	}
14867	break;
14868	}
14869	}
14870	}
14871	#endif /* VBOX_STRICT */
14872	}
14873	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14874	{
14875	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14876	rcStrict = VINF_SUCCESS;
14877	}
14878	else
14879	AssertMsg(rcStrict == VINF_CPUM_R3_MSR_WRITE, ("Unexpected IEMExecDecodedWrmsr rc (%Rrc)\n", VBOXSTRICTRC_VAL(rcStrict)));
14880
14881	return rcStrict;
14882	}
14883
14884
14885	/**
14886	* VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
14887	*/
14888	HMVMX_EXIT_DECL hmR0VmxExitPause(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14889	{
14890	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14891
14892	/** @todo The guest has likely hit a contended spinlock. We might want to
14893	* poke a schedule different guest VCPU. */
14894	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14895	if (RT_SUCCESS(rc))
14896	return VINF_EM_RAW_INTERRUPT;
14897
14898	AssertMsgFailed(("hmR0VmxExitPause: Failed to increment RIP. rc=%Rrc\n", rc));
14899	return rc;
14900	}
14901
14902
14903	/**
14904	* VM-exit handler for when the TPR value is lowered below the specified
14905	* threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
14906	*/
14907	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThreshold(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14908	{
14909	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14910	Assert(pVmxTransient->pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
14911
14912	/*
14913	* The TPR shadow would've been synced with the APIC TPR in the post-run phase.
14914	* We'll re-evaluate pending interrupts and inject them before the next VM
14915	* entry so we can just continue execution here.
14916	*/
14917	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTprBelowThreshold);
14918	return VINF_SUCCESS;
14919	}
14920
14921
14922	/**
14923	* VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX). Conditional
14924	* VM-exit.
14925	*
14926	* @retval VINF_SUCCESS when guest execution can continue.
14927	* @retval VINF_PGM_SYNC_CR3 CR3 sync is required, back to ring-3.
14928	* @retval VERR_EM_RESCHEDULE_REM when we need to return to ring-3 due to
14929	* incompatible guest state for VMX execution (real-on-v86 case).
14930	*/
14931	HMVMX_EXIT_DECL hmR0VmxExitMovCRx(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14932	{
14933	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14934	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitMovCRx, y2);
14935
14936	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14937	hmR0VmxReadExitQualVmcs(pVmxTransient);
14938	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14939
14940	VBOXSTRICTRC rcStrict;
14941	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
14942	uint64_t const uExitQual = pVmxTransient->uExitQual;
14943	uint32_t const uAccessType = VMX_EXIT_QUAL_CRX_ACCESS(uExitQual);
14944	switch (uAccessType)
14945	{
14946	/*
14947	* MOV to CRx.
14948	*/
14949	case VMX_EXIT_QUAL_CRX_ACCESS_WRITE:
14950	{
14951	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14952	AssertRCReturn(rc, rc);
14953
14954	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
14955	uint32_t const uOldCr0 = pVCpu->cpum.GstCtx.cr0;
14956	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(uExitQual);
14957	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(uExitQual);
14958
14959	/*
14960	* MOV to CR3 only cause a VM-exit when one or more of the following are true:
14961	* - When nested paging isn't used.
14962	* - If the guest doesn't have paging enabled (intercept CR3 to update shadow page tables).
14963	* - We are executing in the VM debug loop.
14964	*/
14965	Assert( iCrReg != 3
14966	\|\| !pVM->hm.s.fNestedPaging
14967	\|\| !CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx)
14968	\|\| pVCpu->hm.s.fUsingDebugLoop);
14969
14970	/* MOV to CR8 writes only cause VM-exits when TPR shadow is not used. */
14971	Assert( iCrReg != 8
14972	\|\| !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW));
14973
14974	rcStrict = hmR0VmxExitMovToCrX(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
14975	AssertMsg( rcStrict == VINF_SUCCESS
14976	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
14977
14978	/*
14979	* This is a kludge for handling switches back to real mode when we try to use
14980	* V86 mode to run real mode code directly. Problem is that V86 mode cannot
14981	* deal with special selector values, so we have to return to ring-3 and run
14982	* there till the selector values are V86 mode compatible.
14983	*
14984	* Note! Using VINF_EM_RESCHEDULE_REM here rather than VINF_EM_RESCHEDULE since the
14985	* latter is an alias for VINF_IEM_RAISED_XCPT which is asserted at the end of
14986	* this function.
14987	*/
14988	if ( iCrReg == 0
14989	&& rcStrict == VINF_SUCCESS
14990	&& !pVM->hm.s.vmx.fUnrestrictedGuest
14991	&& CPUMIsGuestInRealModeEx(&pVCpu->cpum.GstCtx)
14992	&& (uOldCr0 & X86_CR0_PE)
14993	&& !(pVCpu->cpum.GstCtx.cr0 & X86_CR0_PE))
14994	{
14995	/** @todo Check selectors rather than returning all the time. */
14996	Assert(!pVmxTransient->fIsNestedGuest);
14997	Log4Func(("CR0 write, back to real mode -> VINF_EM_RESCHEDULE_REM\n"));
14998	rcStrict = VINF_EM_RESCHEDULE_REM;
14999	}
15000	break;
15001	}
15002
15003	/*
15004	* MOV from CRx.
15005	*/
15006	case VMX_EXIT_QUAL_CRX_ACCESS_READ:
15007	{
15008	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(uExitQual);
15009	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(uExitQual);
15010
15011	/*
15012	* MOV from CR3 only cause a VM-exit when one or more of the following are true:
15013	* - When nested paging isn't used.
15014	* - If the guest doesn't have paging enabled (pass guest's CR3 rather than our identity mapped CR3).
15015	* - We are executing in the VM debug loop.
15016	*/
15017	Assert( iCrReg != 3
15018	\|\| !pVM->hm.s.fNestedPaging
15019	\|\| !CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx)
15020	\|\| pVCpu->hm.s.fUsingDebugLoop);
15021
15022	/* MOV from CR8 reads only cause a VM-exit when the TPR shadow feature isn't enabled. */
15023	Assert( iCrReg != 8
15024	\|\| !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW));
15025
15026	rcStrict = hmR0VmxExitMovFromCrX(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
15027	break;
15028	}
15029
15030	/*
15031	* CLTS (Clear Task-Switch Flag in CR0).
15032	*/
15033	case VMX_EXIT_QUAL_CRX_ACCESS_CLTS:
15034	{
15035	rcStrict = hmR0VmxExitClts(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr);
15036	break;
15037	}
15038
15039	/*
15040	* LMSW (Load Machine-Status Word into CR0).
15041	* LMSW cannot clear CR0.PE, so no fRealOnV86Active kludge needed here.
15042	*/
15043	case VMX_EXIT_QUAL_CRX_ACCESS_LMSW:
15044	{
15045	RTGCPTR GCPtrEffDst;
15046	uint8_t const cbInstr = pVmxTransient->cbExitInstr;
15047	uint16_t const uMsw = VMX_EXIT_QUAL_CRX_LMSW_DATA(uExitQual);
15048	bool const fMemOperand = VMX_EXIT_QUAL_CRX_LMSW_OP_MEM(uExitQual);
15049	if (fMemOperand)
15050	{
15051	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
15052	GCPtrEffDst = pVmxTransient->uGuestLinearAddr;
15053	}
15054	else
15055	GCPtrEffDst = NIL_RTGCPTR;
15056	rcStrict = hmR0VmxExitLmsw(pVCpu, pVmcsInfo, cbInstr, uMsw, GCPtrEffDst);
15057	break;
15058	}
15059
15060	default:
15061	{
15062	AssertMsgFailed(("Unrecognized Mov CRX access type %#x\n", uAccessType));
15063	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, uAccessType);
15064	}
15065	}
15066
15067	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS))
15068	== (HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS));
15069	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
15070
15071	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitMovCRx, y2);
15072	NOREF(pVM);
15073	return rcStrict;
15074	}
15075
15076
15077	/**
15078	* VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR). Conditional
15079	* VM-exit.
15080	*/
15081	HMVMX_EXIT_DECL hmR0VmxExitIoInstr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15082	{
15083	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15084	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitIO, y1);
15085
15086	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15087	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15088	hmR0VmxReadExitQualVmcs(pVmxTransient);
15089	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15090	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_SREG_MASK
15091	\| CPUMCTX_EXTRN_EFER);
15092	/* EFER MSR also required for longmode checks in EMInterpretDisasCurrent(), but it's always up-to-date. */
15093	AssertRCReturn(rc, rc);
15094
15095	/* Refer Intel spec. 27-5. "Exit Qualifications for I/O Instructions" for the format. */
15096	uint32_t const uIOPort = VMX_EXIT_QUAL_IO_PORT(pVmxTransient->uExitQual);
15097	uint8_t const uIOSize = VMX_EXIT_QUAL_IO_SIZE(pVmxTransient->uExitQual);
15098	bool const fIOWrite = (VMX_EXIT_QUAL_IO_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_IO_DIRECTION_OUT);
15099	bool const fIOString = VMX_EXIT_QUAL_IO_IS_STRING(pVmxTransient->uExitQual);
15100	bool const fGstStepping = RT_BOOL(pCtx->eflags.Bits.u1TF);
15101	bool const fDbgStepping = pVCpu->hm.s.fSingleInstruction;
15102	AssertReturn(uIOSize <= 3 && uIOSize != 2, VERR_VMX_IPE_1);
15103
15104	/*
15105	* Update exit history to see if this exit can be optimized.
15106	*/
15107	VBOXSTRICTRC rcStrict;
15108	PCEMEXITREC pExitRec = NULL;
15109	if ( !fGstStepping
15110	&& !fDbgStepping)
15111	pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
15112	!fIOString
15113	? !fIOWrite
15114	? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_READ)
15115	: EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_WRITE)
15116	: !fIOWrite
15117	? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_READ)
15118	: EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_WRITE),
15119	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
15120	if (!pExitRec)
15121	{
15122	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses in bytes. */
15123	static uint32_t const s_aIOOpAnd[4] = { 0xff, 0xffff, 0, 0xffffffff }; /* AND masks for saving result in AL/AX/EAX. */
15124
15125	uint32_t const cbValue = s_aIOSizes[uIOSize];
15126	uint32_t const cbInstr = pVmxTransient->cbExitInstr;
15127	bool fUpdateRipAlready = false; /* ugly hack, should be temporary. */
15128	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15129	if (fIOString)
15130	{
15131	/*
15132	* INS/OUTS - I/O String instruction.
15133	*
15134	* Use instruction-information if available, otherwise fall back on
15135	* interpreting the instruction.
15136	*/
15137	Log4Func(("cs:rip=%#04x:%#RX64 %#06x/%u %c str\n", pCtx->cs.Sel, pCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
15138	AssertReturn(pCtx->dx == uIOPort, VERR_VMX_IPE_2);
15139	bool const fInsOutsInfo = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_INS_OUTS);
15140	if (fInsOutsInfo)
15141	{
15142	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15143	AssertReturn(pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize <= 2, VERR_VMX_IPE_3);
15144	AssertCompile(IEMMODE_16BIT == 0 && IEMMODE_32BIT == 1 && IEMMODE_64BIT == 2);
15145	IEMMODE const enmAddrMode = (IEMMODE)pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize;
15146	bool const fRep = VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual);
15147	if (fIOWrite)
15148	rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, fRep, cbInstr,
15149	pVmxTransient->ExitInstrInfo.StrIo.iSegReg, true /fIoChecked/);
15150	else
15151	{
15152	/*
15153	* The segment prefix for INS cannot be overridden and is always ES. We can safely assume X86_SREG_ES.
15154	* Hence "iSegReg" field is undefined in the instruction-information field in VT-x for INS.
15155	* See Intel Instruction spec. for "INS".
15156	* See Intel spec. Table 27-8 "Format of the VM-Exit Instruction-Information Field as Used for INS and OUTS".
15157	*/
15158	rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, fRep, cbInstr, true /fIoChecked/);
15159	}
15160	}
15161	else
15162	rcStrict = IEMExecOne(pVCpu);
15163
15164	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
15165	fUpdateRipAlready = true;
15166	}
15167	else
15168	{
15169	/*
15170	* IN/OUT - I/O instruction.
15171	*/
15172	Log4Func(("cs:rip=%04x:%08RX64 %#06x/%u %c\n", pCtx->cs.Sel, pCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
15173	uint32_t const uAndVal = s_aIOOpAnd[uIOSize];
15174	Assert(!VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual));
15175	if (fIOWrite)
15176	{
15177	rcStrict = IOMIOPortWrite(pVM, pVCpu, uIOPort, pCtx->eax & uAndVal, cbValue);
15178	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite);
15179	if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
15180	&& !pCtx->eflags.Bits.u1TF)
15181	rcStrict = EMRZSetPendingIoPortWrite(pVCpu, uIOPort, cbInstr, cbValue, pCtx->eax & uAndVal);
15182	}
15183	else
15184	{
15185	uint32_t u32Result = 0;
15186	rcStrict = IOMIOPortRead(pVM, pVCpu, uIOPort, &u32Result, cbValue);
15187	if (IOM_SUCCESS(rcStrict))
15188	{
15189	/* Save result of I/O IN instr. in AL/AX/EAX. */
15190	pCtx->eax = (pCtx->eax & ~uAndVal) \| (u32Result & uAndVal);
15191	}
15192	if ( rcStrict == VINF_IOM_R3_IOPORT_READ
15193	&& !pCtx->eflags.Bits.u1TF)
15194	rcStrict = EMRZSetPendingIoPortRead(pVCpu, uIOPort, cbInstr, cbValue);
15195	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead);
15196	}
15197	}
15198
15199	if (IOM_SUCCESS(rcStrict))
15200	{
15201	if (!fUpdateRipAlready)
15202	{
15203	hmR0VmxAdvanceGuestRipBy(pVCpu, cbInstr);
15204	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
15205	}
15206
15207	/*
15208	* INS/OUTS with REP prefix updates RFLAGS, can be observed with triple-fault guru
15209	* while booting Fedora 17 64-bit guest.
15210	*
15211	* See Intel Instruction reference for REP/REPE/REPZ/REPNE/REPNZ.
15212	*/
15213	if (fIOString)
15214	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RFLAGS);
15215
15216	/*
15217	* If any I/O breakpoints are armed, we need to check if one triggered
15218	* and take appropriate action.
15219	* Note that the I/O breakpoint type is undefined if CR4.DE is 0.
15220	*/
15221	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_DR7);
15222	AssertRCReturn(rc, rc);
15223
15224	/** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the
15225	* execution engines about whether hyper BPs and such are pending. */
15226	uint32_t const uDr7 = pCtx->dr[7];
15227	if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK)
15228	&& X86_DR7_ANY_RW_IO(uDr7)
15229	&& (pCtx->cr4 & X86_CR4_DE))
15230	\|\| DBGFBpIsHwIoArmed(pVM)))
15231	{
15232	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck);
15233
15234	/* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
15235	VMMRZCallRing3Disable(pVCpu);
15236	HM_DISABLE_PREEMPT(pVCpu);
15237
15238	bool fIsGuestDbgActive = CPUMR0DebugStateMaybeSaveGuest(pVCpu, true /* fDr6 */);
15239
15240	VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pCtx, uIOPort, cbValue);
15241	if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
15242	{
15243	/* Raise #DB. */
15244	if (fIsGuestDbgActive)
15245	ASMSetDR6(pCtx->dr[6]);
15246	if (pCtx->dr[7] != uDr7)
15247	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_GUEST_DR7;
15248
15249	hmR0VmxSetPendingXcptDB(pVCpu);
15250	}
15251	/* rcStrict is VINF_SUCCESS, VINF_IOM_R3_IOPORT_COMMIT_WRITE, or in [VINF_EM_FIRST..VINF_EM_LAST],
15252	however we can ditch VINF_IOM_R3_IOPORT_COMMIT_WRITE as it has VMCPU_FF_IOM as backup. */
15253	else if ( rcStrict2 != VINF_SUCCESS
15254	&& (rcStrict == VINF_SUCCESS \|\| rcStrict2 < rcStrict))
15255	rcStrict = rcStrict2;
15256	AssertCompile(VINF_EM_LAST < VINF_IOM_R3_IOPORT_COMMIT_WRITE);
15257
15258	HM_RESTORE_PREEMPT();
15259	VMMRZCallRing3Enable(pVCpu);
15260	}
15261	}
15262
15263	#ifdef VBOX_STRICT
15264	if ( rcStrict == VINF_IOM_R3_IOPORT_READ
15265	\|\| rcStrict == VINF_EM_PENDING_R3_IOPORT_READ)
15266	Assert(!fIOWrite);
15267	else if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
15268	\|\| rcStrict == VINF_IOM_R3_IOPORT_COMMIT_WRITE
15269	\|\| rcStrict == VINF_EM_PENDING_R3_IOPORT_WRITE)
15270	Assert(fIOWrite);
15271	else
15272	{
15273	# if 0 /** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST
15274	* statuses, that the VMM device and some others may return. See
15275	* IOM_SUCCESS() for guidance. */
15276	AssertMsg( RT_FAILURE(rcStrict)
15277	\|\| rcStrict == VINF_SUCCESS
15278	\|\| rcStrict == VINF_EM_RAW_EMULATE_INSTR
15279	\|\| rcStrict == VINF_EM_DBG_BREAKPOINT
15280	\|\| rcStrict == VINF_EM_RAW_GUEST_TRAP
15281	\|\| rcStrict == VINF_EM_RAW_TO_R3
15282	\|\| rcStrict == VINF_TRPM_XCPT_DISPATCHED, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15283	# endif
15284	}
15285	#endif
15286	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitIO, y1);
15287	}
15288	else
15289	{
15290	/*
15291	* Frequent exit or something needing probing. Get state and call EMHistoryExec.
15292	*/
15293	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
15294	AssertRCReturn(rc2, rc2);
15295	STAM_COUNTER_INC(!fIOString ? fIOWrite ? &pVCpu->hm.s.StatExitIOWrite : &pVCpu->hm.s.StatExitIORead
15296	: fIOWrite ? &pVCpu->hm.s.StatExitIOStringWrite : &pVCpu->hm.s.StatExitIOStringRead);
15297	Log4(("IOExit/%u: %04x:%08RX64: %s%s%s %#x LB %u -> EMHistoryExec\n",
15298	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
15299	VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual) ? "REP " : "",
15300	fIOWrite ? "OUT" : "IN", fIOString ? "S" : "", uIOPort, uIOSize));
15301
15302	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
15303	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
15304
15305	Log4(("IOExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
15306	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
15307	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
15308	}
15309	return rcStrict;
15310	}
15311
15312
15313	/**
15314	* VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH). Unconditional
15315	* VM-exit.
15316	*/
15317	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitch(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15318	{
15319	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15320
15321	/* Check if this task-switch occurred while delivery an event through the guest IDT. */
15322	hmR0VmxReadExitQualVmcs(pVmxTransient);
15323	if (VMX_EXIT_QUAL_TASK_SWITCH_TYPE(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_TASK_SWITCH_TYPE_IDT)
15324	{
15325	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
15326	if (VMX_IDT_VECTORING_INFO_IS_VALID(pVmxTransient->uIdtVectoringInfo))
15327	{
15328	uint32_t uErrCode;
15329	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(pVmxTransient->uIdtVectoringInfo))
15330	{
15331	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
15332	uErrCode = pVmxTransient->uIdtVectoringErrorCode;
15333	}
15334	else
15335	uErrCode = 0;
15336
15337	RTGCUINTPTR GCPtrFaultAddress;
15338	if (VMX_IDT_VECTORING_INFO_IS_XCPT_PF(pVmxTransient->uIdtVectoringInfo))
15339	GCPtrFaultAddress = pVCpu->cpum.GstCtx.cr2;
15340	else
15341	GCPtrFaultAddress = 0;
15342
15343	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15344
15345	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(pVmxTransient->uIdtVectoringInfo),
15346	pVmxTransient->cbExitInstr, uErrCode, GCPtrFaultAddress);
15347
15348	Log4Func(("Pending event. uIntType=%#x uVector=%#x\n", VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo),
15349	VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo)));
15350	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
15351	return VINF_EM_RAW_INJECT_TRPM_EVENT;
15352	}
15353	}
15354
15355	/* Fall back to the interpreter to emulate the task-switch. */
15356	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
15357	return VERR_EM_INTERPRETER;
15358	}
15359
15360
15361	/**
15362	* VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional VM-exit.
15363	*/
15364	HMVMX_EXIT_DECL hmR0VmxExitMtf(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15365	{
15366	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15367
15368	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15369	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_MONITOR_TRAP_FLAG;
15370	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
15371	AssertRC(rc);
15372	return VINF_EM_DBG_STEPPED;
15373	}
15374
15375
15376	/**
15377	* VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional VM-exit.
15378	*/
15379	HMVMX_EXIT_DECL hmR0VmxExitApicAccess(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15380	{
15381	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15382	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitApicAccess);
15383
15384	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
15385	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
15386	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15387	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
15388	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
15389
15390	/*
15391	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
15392	*/
15393	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
15394	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15395	{
15396	/* For some crazy guest, if an event delivery causes an APIC-access VM-exit, go to instruction emulation. */
15397	if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
15398	{
15399	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
15400	return VINF_EM_RAW_INJECT_TRPM_EVENT;
15401	}
15402	}
15403	else
15404	{
15405	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
15406	return rcStrict;
15407	}
15408
15409	/* IOMMIOPhysHandler() below may call into IEM, save the necessary state. */
15410	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15411	hmR0VmxReadExitQualVmcs(pVmxTransient);
15412	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
15413	AssertRCReturn(rc, rc);
15414
15415	/* See Intel spec. 27-6 "Exit Qualifications for APIC-access VM-exits from Linear Accesses & Guest-Phyiscal Addresses" */
15416	uint32_t const uAccessType = VMX_EXIT_QUAL_APIC_ACCESS_TYPE(pVmxTransient->uExitQual);
15417	switch (uAccessType)
15418	{
15419	case VMX_APIC_ACCESS_TYPE_LINEAR_WRITE:
15420	case VMX_APIC_ACCESS_TYPE_LINEAR_READ:
15421	{
15422	AssertMsg( !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
15423	\|\| VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual) != XAPIC_OFF_TPR,
15424	("hmR0VmxExitApicAccess: can't access TPR offset while using TPR shadowing.\n"));
15425
15426	RTGCPHYS GCPhys = pVCpu->hm.s.vmx.u64GstMsrApicBase; /* Always up-to-date, as it is not part of the VMCS. */
15427	GCPhys &= PAGE_BASE_GC_MASK;
15428	GCPhys += VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual);
15429	Log4Func(("Linear access uAccessType=%#x GCPhys=%#RGp Off=%#x\n", uAccessType, GCPhys,
15430	VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual)));
15431
15432	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15433	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15434	rcStrict = IOMMMIOPhysHandler(pVM, pVCpu,
15435	uAccessType == VMX_APIC_ACCESS_TYPE_LINEAR_READ ? 0 : X86_TRAP_PF_RW,
15436	CPUMCTX2CORE(pCtx), GCPhys);
15437	Log4Func(("IOMMMIOPhysHandler returned %Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15438	if ( rcStrict == VINF_SUCCESS
15439	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
15440	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
15441	{
15442	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS
15443	\| HM_CHANGED_GUEST_APIC_TPR);
15444	rcStrict = VINF_SUCCESS;
15445	}
15446	break;
15447	}
15448
15449	default:
15450	{
15451	Log4Func(("uAccessType=%#x\n", uAccessType));
15452	rcStrict = VINF_EM_RAW_EMULATE_INSTR;
15453	break;
15454	}
15455	}
15456
15457	if (rcStrict != VINF_SUCCESS)
15458	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchApicAccessToR3);
15459	return rcStrict;
15460	}
15461
15462
15463	/**
15464	* VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX). Conditional
15465	* VM-exit.
15466	*/
15467	HMVMX_EXIT_DECL hmR0VmxExitMovDRx(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15468	{
15469	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15470	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15471
15472	/* We might get this VM-exit if the nested-guest is not intercepting MOV DRx accesses. */
15473	if (!pVmxTransient->fIsNestedGuest)
15474	{
15475	/* We should -not- get this VM-exit if the guest's debug registers were active. */
15476	if (pVmxTransient->fWasGuestDebugStateActive)
15477	{
15478	AssertMsgFailed(("Unexpected MOV DRx exit\n"));
15479	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
15480	}
15481
15482	if ( !pVCpu->hm.s.fSingleInstruction
15483	&& !pVmxTransient->fWasHyperDebugStateActive)
15484	{
15485	Assert(!DBGFIsStepping(pVCpu));
15486	Assert(pVmcsInfo->u32XcptBitmap & RT_BIT(X86_XCPT_DB));
15487
15488	/* Don't intercept MOV DRx any more. */
15489	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_MOV_DR_EXIT;
15490	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
15491	AssertRC(rc);
15492
15493	/* We're playing with the host CPU state here, make sure we can't preempt or longjmp. */
15494	VMMRZCallRing3Disable(pVCpu);
15495	HM_DISABLE_PREEMPT(pVCpu);
15496
15497	/* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */
15498	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
15499	Assert(CPUMIsGuestDebugStateActive(pVCpu));
15500
15501	HM_RESTORE_PREEMPT();
15502	VMMRZCallRing3Enable(pVCpu);
15503
15504	#ifdef VBOX_WITH_STATISTICS
15505	hmR0VmxReadExitQualVmcs(pVmxTransient);
15506	if (VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_DRX_DIRECTION_WRITE)
15507	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
15508	else
15509	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
15510	#endif
15511	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch);
15512	return VINF_SUCCESS;
15513	}
15514	}
15515
15516	/*
15517	* EMInterpretDRx[Write\|Read]() calls CPUMIsGuestIn64BitCode() which requires EFER MSR, CS.
15518	* The EFER MSR is always up-to-date.
15519	* Update the segment registers and DR7 from the CPU.
15520	*/
15521	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15522	hmR0VmxReadExitQualVmcs(pVmxTransient);
15523	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_DR7);
15524	AssertRCReturn(rc, rc);
15525	Log4Func(("cs:rip=%#04x:%#RX64\n", pCtx->cs.Sel, pCtx->rip));
15526
15527	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15528	if (VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_DRX_DIRECTION_WRITE)
15529	{
15530	rc = EMInterpretDRxWrite(pVM, pVCpu, CPUMCTX2CORE(pCtx),
15531	VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual),
15532	VMX_EXIT_QUAL_DRX_GENREG(pVmxTransient->uExitQual));
15533	if (RT_SUCCESS(rc))
15534	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
15535	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
15536	}
15537	else
15538	{
15539	rc = EMInterpretDRxRead(pVM, pVCpu, CPUMCTX2CORE(pCtx),
15540	VMX_EXIT_QUAL_DRX_GENREG(pVmxTransient->uExitQual),
15541	VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual));
15542	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
15543	}
15544
15545	Assert(rc == VINF_SUCCESS \|\| rc == VERR_EM_INTERPRETER);
15546	if (RT_SUCCESS(rc))
15547	{
15548	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15549	AssertRCReturn(rc2, rc2);
15550	return VINF_SUCCESS;
15551	}
15552	return rc;
15553	}
15554
15555
15556	/**
15557	* VM-exit handler for EPT misconfiguration (VMX_EXIT_EPT_MISCONFIG).
15558	* Conditional VM-exit.
15559	*/
15560	HMVMX_EXIT_DECL hmR0VmxExitEptMisconfig(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15561	{
15562	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15563	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
15564
15565	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
15566	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
15567	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15568	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
15569	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
15570
15571	/*
15572	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
15573	*/
15574	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
15575	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15576	{
15577	/*
15578	* In the unlikely case where delivering an event causes an EPT misconfig (MMIO), go back to
15579	* instruction emulation to inject the original event. Otherwise, injecting the original event
15580	* using hardware-assisted VMX would would trigger the same EPT misconfig VM-exit again.
15581	*/
15582	if (!pVCpu->hm.s.Event.fPending)
15583	{ /* likely */ }
15584	else
15585	{
15586	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
15587	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
15588	/** @todo NSTVMX: Think about how this should be handled. */
15589	if (pVmxTransient->fIsNestedGuest)
15590	return VERR_VMX_IPE_3;
15591	#endif
15592	return VINF_EM_RAW_INJECT_TRPM_EVENT;
15593	}
15594	}
15595	else
15596	{
15597	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
15598	return rcStrict;
15599	}
15600
15601	/*
15602	* Get sufficent state and update the exit history entry.
15603	*/
15604	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15605	hmR0VmxReadGuestPhysicalAddrVmcs(pVmxTransient);
15606	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
15607	AssertRCReturn(rc, rc);
15608
15609	RTGCPHYS const GCPhys = pVmxTransient->uGuestPhysicalAddr;
15610	PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
15611	EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_MMIO),
15612	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
15613	if (!pExitRec)
15614	{
15615	/*
15616	* If we succeed, resume guest execution.
15617	* If we fail in interpreting the instruction because we couldn't get the guest physical address
15618	* of the page containing the instruction via the guest's page tables (we would invalidate the guest page
15619	* in the host TLB), resume execution which would cause a guest page fault to let the guest handle this
15620	* weird case. See @bugref{6043}.
15621	*/
15622	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15623	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15624	rcStrict = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, PGMMODE_EPT, CPUMCTX2CORE(pCtx), GCPhys, UINT32_MAX);
15625	Log4Func(("At %#RGp RIP=%#RX64 rc=%Rrc\n", GCPhys, pCtx->rip, VBOXSTRICTRC_VAL(rcStrict)));
15626	if ( rcStrict == VINF_SUCCESS
15627	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
15628	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
15629	{
15630	/* Successfully handled MMIO operation. */
15631	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS
15632	\| HM_CHANGED_GUEST_APIC_TPR);
15633	rcStrict = VINF_SUCCESS;
15634	}
15635	}
15636	else
15637	{
15638	/*
15639	* Frequent exit or something needing probing. Call EMHistoryExec.
15640	*/
15641	Log4(("EptMisscfgExit/%u: %04x:%08RX64: %RGp -> EMHistoryExec\n",
15642	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, GCPhys));
15643
15644	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
15645	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
15646
15647	Log4(("EptMisscfgExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
15648	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
15649	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
15650	}
15651	return rcStrict;
15652	}
15653
15654
15655	/**
15656	* VM-exit handler for EPT violation (VMX_EXIT_EPT_VIOLATION). Conditional
15657	* VM-exit.
15658	*/
15659	HMVMX_EXIT_DECL hmR0VmxExitEptViolation(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15660	{
15661	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15662	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
15663
15664	hmR0VmxReadExitQualVmcs(pVmxTransient);
15665	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
15666	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
15667	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15668	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
15669	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
15670
15671	/*
15672	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
15673	*/
15674	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
15675	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15676	{
15677	/*
15678	* If delivery of an event causes an EPT violation (true nested #PF and not MMIO),
15679	* we shall resolve the nested #PF and re-inject the original event.
15680	*/
15681	if (pVCpu->hm.s.Event.fPending)
15682	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflectNPF);
15683	}
15684	else
15685	{
15686	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
15687	return rcStrict;
15688	}
15689
15690	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15691	hmR0VmxReadGuestPhysicalAddrVmcs(pVmxTransient);
15692	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
15693	AssertRCReturn(rc, rc);
15694
15695	RTGCPHYS const GCPhys = pVmxTransient->uGuestPhysicalAddr;
15696	uint64_t const uExitQual = pVmxTransient->uExitQual;
15697	AssertMsg(((pVmxTransient->uExitQual >> 7) & 3) != 2, ("%#RX64", uExitQual));
15698
15699	RTGCUINT uErrorCode = 0;
15700	if (uExitQual & VMX_EXIT_QUAL_EPT_INSTR_FETCH)
15701	uErrorCode \|= X86_TRAP_PF_ID;
15702	if (uExitQual & VMX_EXIT_QUAL_EPT_DATA_WRITE)
15703	uErrorCode \|= X86_TRAP_PF_RW;
15704	if (uExitQual & VMX_EXIT_QUAL_EPT_ENTRY_PRESENT)
15705	uErrorCode \|= X86_TRAP_PF_P;
15706
15707	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15708	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15709	Log4Func(("at %#RX64 (%#RX64 errcode=%#x) cs:rip=%#04x:%#RX64\n", GCPhys, uExitQual, uErrorCode, pCtx->cs.Sel, pCtx->rip));
15710
15711	/*
15712	* Handle the pagefault trap for the nested shadow table.
15713	*/
15714	TRPMAssertXcptPF(pVCpu, GCPhys, uErrorCode);
15715	rcStrict = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, PGMMODE_EPT, uErrorCode, CPUMCTX2CORE(pCtx), GCPhys);
15716	TRPMResetTrap(pVCpu);
15717
15718	/* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}. */
15719	if ( rcStrict == VINF_SUCCESS
15720	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
15721	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
15722	{
15723	/* Successfully synced our nested page tables. */
15724	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf);
15725	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS);
15726	return VINF_SUCCESS;
15727	}
15728
15729	Log4Func(("EPT return to ring-3 rcStrict2=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15730	return rcStrict;
15731	}
15732
15733
15734	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
15735	/**
15736	* VM-exit handler for VMCLEAR (VMX_EXIT_VMCLEAR). Unconditional VM-exit.
15737	*/
15738	HMVMX_EXIT_DECL hmR0VmxExitVmclear(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15739	{
15740	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15741
15742	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15743	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15744	hmR0VmxReadExitQualVmcs(pVmxTransient);
15745	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
15746	\| CPUMCTX_EXTRN_HWVIRT
15747	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
15748	AssertRCReturn(rc, rc);
15749
15750	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
15751
15752	VMXVEXITINFO ExitInfo;
15753	RT_ZERO(ExitInfo);
15754	ExitInfo.uReason = pVmxTransient->uExitReason;
15755	ExitInfo.u64Qual = pVmxTransient->uExitQual;
15756	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
15757	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
15758	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
15759
15760	VBOXSTRICTRC rcStrict = IEMExecDecodedVmclear(pVCpu, &ExitInfo);
15761	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15762	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
15763	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15764	{
15765	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15766	rcStrict = VINF_SUCCESS;
15767	}
15768	return rcStrict;
15769	}
15770
15771
15772	/**
15773	* VM-exit handler for VMLAUNCH (VMX_EXIT_VMLAUNCH). Unconditional VM-exit.
15774	*/
15775	HMVMX_EXIT_DECL hmR0VmxExitVmlaunch(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15776	{
15777	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15778
15779	/* Import the entire VMCS state for now as we would be switching VMCS on successful VMLAUNCH,
15780	otherwise we could import just IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK. */
15781	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15782	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
15783	AssertRCReturn(rc, rc);
15784
15785	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
15786
15787	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
15788	VBOXSTRICTRC rcStrict = IEMExecDecodedVmlaunchVmresume(pVCpu, pVmxTransient->cbExitInstr, VMXINSTRID_VMLAUNCH);
15789	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
15790	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15791	{
15792	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
15793	if (CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
15794	rcStrict = VINF_VMX_VMLAUNCH_VMRESUME;
15795	}
15796	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
15797	return rcStrict;
15798	}
15799
15800
15801	/**
15802	* VM-exit handler for VMPTRLD (VMX_EXIT_VMPTRLD). Unconditional VM-exit.
15803	*/
15804	HMVMX_EXIT_DECL hmR0VmxExitVmptrld(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15805	{
15806	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15807
15808	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15809	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15810	hmR0VmxReadExitQualVmcs(pVmxTransient);
15811	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
15812	\| CPUMCTX_EXTRN_HWVIRT
15813	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
15814	AssertRCReturn(rc, rc);
15815
15816	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
15817
15818	VMXVEXITINFO ExitInfo;
15819	RT_ZERO(ExitInfo);
15820	ExitInfo.uReason = pVmxTransient->uExitReason;
15821	ExitInfo.u64Qual = pVmxTransient->uExitQual;
15822	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
15823	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
15824	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
15825
15826	VBOXSTRICTRC rcStrict = IEMExecDecodedVmptrld(pVCpu, &ExitInfo);
15827	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15828	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
15829	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15830	{
15831	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15832	rcStrict = VINF_SUCCESS;
15833	}
15834	return rcStrict;
15835	}
15836
15837
15838	/**
15839	* VM-exit handler for VMPTRST (VMX_EXIT_VMPTRST). Unconditional VM-exit.
15840	*/
15841	HMVMX_EXIT_DECL hmR0VmxExitVmptrst(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15842	{
15843	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15844
15845	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15846	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15847	hmR0VmxReadExitQualVmcs(pVmxTransient);
15848	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
15849	\| CPUMCTX_EXTRN_HWVIRT
15850	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
15851	AssertRCReturn(rc, rc);
15852
15853	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
15854
15855	VMXVEXITINFO ExitInfo;
15856	RT_ZERO(ExitInfo);
15857	ExitInfo.uReason = pVmxTransient->uExitReason;
15858	ExitInfo.u64Qual = pVmxTransient->uExitQual;
15859	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
15860	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
15861	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_WRITE, &ExitInfo.GCPtrEffAddr);
15862
15863	VBOXSTRICTRC rcStrict = IEMExecDecodedVmptrst(pVCpu, &ExitInfo);
15864	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15865	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15866	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15867	{
15868	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15869	rcStrict = VINF_SUCCESS;
15870	}
15871	return rcStrict;
15872	}
15873
15874
15875	/**
15876	* VM-exit handler for VMREAD (VMX_EXIT_VMREAD). Conditional VM-exit.
15877	*/
15878	HMVMX_EXIT_DECL hmR0VmxExitVmread(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15879	{
15880	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15881
15882	/*
15883	* Strictly speaking we should not get VMREAD VM-exits for shadow VMCS fields and
15884	* thus might not need to import the shadow VMCS state, it's safer just in case
15885	* code elsewhere dares look at unsynced VMCS fields.
15886	*/
15887	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15888	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15889	hmR0VmxReadExitQualVmcs(pVmxTransient);
15890	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
15891	\| CPUMCTX_EXTRN_HWVIRT
15892	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
15893	AssertRCReturn(rc, rc);
15894
15895	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
15896
15897	VMXVEXITINFO ExitInfo;
15898	RT_ZERO(ExitInfo);
15899	ExitInfo.uReason = pVmxTransient->uExitReason;
15900	ExitInfo.u64Qual = pVmxTransient->uExitQual;
15901	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
15902	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
15903	if (!ExitInfo.InstrInfo.VmreadVmwrite.fIsRegOperand)
15904	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_WRITE, &ExitInfo.GCPtrEffAddr);
15905
15906	VBOXSTRICTRC rcStrict = IEMExecDecodedVmread(pVCpu, &ExitInfo);
15907	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15908	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15909	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15910	{
15911	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15912	rcStrict = VINF_SUCCESS;
15913	}
15914	return rcStrict;
15915	}
15916
15917
15918	/**
15919	* VM-exit handler for VMRESUME (VMX_EXIT_VMRESUME). Unconditional VM-exit.
15920	*/
15921	HMVMX_EXIT_DECL hmR0VmxExitVmresume(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15922	{
15923	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15924
15925	/* Import the entire VMCS state for now as we would be switching VMCS on successful VMRESUME,
15926	otherwise we could import just IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK. */
15927	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15928	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
15929	AssertRCReturn(rc, rc);
15930
15931	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
15932
15933	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
15934	VBOXSTRICTRC rcStrict = IEMExecDecodedVmlaunchVmresume(pVCpu, pVmxTransient->cbExitInstr, VMXINSTRID_VMRESUME);
15935	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
15936	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15937	{
15938	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
15939	if (CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
15940	rcStrict = VINF_VMX_VMLAUNCH_VMRESUME;
15941	}
15942	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
15943	return rcStrict;
15944	}
15945
15946
15947	/**
15948	* VM-exit handler for VMWRITE (VMX_EXIT_VMWRITE). Conditional VM-exit.
15949	*/
15950	HMVMX_EXIT_DECL hmR0VmxExitVmwrite(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15951	{
15952	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15953
15954	/*
15955	* Although we should not get VMWRITE VM-exits for shadow VMCS fields, since our HM hook
15956	* gets invoked when IEM's VMWRITE instruction emulation modifies the current VMCS and it
15957	* flags re-loading the entire shadow VMCS, we should save the entire shadow VMCS here.
15958	*/
15959	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15960	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15961	hmR0VmxReadExitQualVmcs(pVmxTransient);
15962	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
15963	\| CPUMCTX_EXTRN_HWVIRT
15964	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
15965	AssertRCReturn(rc, rc);
15966
15967	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
15968
15969	VMXVEXITINFO ExitInfo;
15970	RT_ZERO(ExitInfo);
15971	ExitInfo.uReason = pVmxTransient->uExitReason;
15972	ExitInfo.u64Qual = pVmxTransient->uExitQual;
15973	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
15974	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
15975	if (!ExitInfo.InstrInfo.VmreadVmwrite.fIsRegOperand)
15976	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
15977
15978	VBOXSTRICTRC rcStrict = IEMExecDecodedVmwrite(pVCpu, &ExitInfo);
15979	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15980	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
15981	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15982	{
15983	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15984	rcStrict = VINF_SUCCESS;
15985	}
15986	return rcStrict;
15987	}
15988
15989
15990	/**
15991	* VM-exit handler for VMXOFF (VMX_EXIT_VMXOFF). Unconditional VM-exit.
15992	*/
15993	HMVMX_EXIT_DECL hmR0VmxExitVmxoff(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15994	{
15995	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15996
15997	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15998	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR4
15999	\| CPUMCTX_EXTRN_HWVIRT
16000	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
16001	AssertRCReturn(rc, rc);
16002
16003	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16004
16005	VBOXSTRICTRC rcStrict = IEMExecDecodedVmxoff(pVCpu, pVmxTransient->cbExitInstr);
16006	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16007	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_HWVIRT);
16008	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16009	{
16010	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16011	rcStrict = VINF_SUCCESS;
16012	}
16013	return rcStrict;
16014	}
16015
16016
16017	/**
16018	* VM-exit handler for VMXON (VMX_EXIT_VMXON). Unconditional VM-exit.
16019	*/
16020	HMVMX_EXIT_DECL hmR0VmxExitVmxon(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16021	{
16022	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16023
16024	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16025	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16026	hmR0VmxReadExitQualVmcs(pVmxTransient);
16027	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16028	\| CPUMCTX_EXTRN_HWVIRT
16029	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16030	AssertRCReturn(rc, rc);
16031
16032	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16033
16034	VMXVEXITINFO ExitInfo;
16035	RT_ZERO(ExitInfo);
16036	ExitInfo.uReason = pVmxTransient->uExitReason;
16037	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16038	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16039	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16040	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16041
16042	VBOXSTRICTRC rcStrict = IEMExecDecodedVmxon(pVCpu, &ExitInfo);
16043	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16044	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16045	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16046	{
16047	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16048	rcStrict = VINF_SUCCESS;
16049	}
16050	return rcStrict;
16051	}
16052
16053
16054	/**
16055	* VM-exit handler for INVVPID (VMX_EXIT_INVVPID). Unconditional VM-exit.
16056	*/
16057	HMVMX_EXIT_DECL hmR0VmxExitInvvpid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16058	{
16059	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16060
16061	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16062	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16063	hmR0VmxReadExitQualVmcs(pVmxTransient);
16064	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16065	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16066	AssertRCReturn(rc, rc);
16067
16068	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16069
16070	VMXVEXITINFO ExitInfo;
16071	RT_ZERO(ExitInfo);
16072	ExitInfo.uReason = pVmxTransient->uExitReason;
16073	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16074	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16075	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16076	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16077
16078	VBOXSTRICTRC rcStrict = IEMExecDecodedInvvpid(pVCpu, &ExitInfo);
16079	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16080	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16081	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16082	{
16083	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16084	rcStrict = VINF_SUCCESS;
16085	}
16086	return rcStrict;
16087	}
16088	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
16089	/** @} */
16090
16091
16092	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16093	/** @name Nested-guest VM-exit handlers.
16094	* @{
16095	*/
16096	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16097	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Nested-guest VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16098	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16099
16100	/**
16101	* Nested-guest VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI).
16102	* Conditional VM-exit.
16103	*/
16104	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmiNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16105	{
16106	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16107
16108	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16109
16110	uint64_t const uExitIntInfo = pVmxTransient->uExitIntInfo;
16111	uint32_t const uExitIntType = VMX_EXIT_INT_INFO_TYPE(uExitIntInfo);
16112	Assert(VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo));
16113
16114	switch (uExitIntType)
16115	{
16116	/*
16117	* Physical NMIs:
16118	* We shouldn't direct host physical NMIs to the nested-guest. Dispatch it to the host.
16119	*/
16120	case VMX_EXIT_INT_INFO_TYPE_NMI:
16121	return hmR0VmxExitHostNmi(pVCpu, pVmxTransient->pVmcsInfo);
16122
16123	/*
16124	* Hardware exceptions,
16125	* Software exceptions,
16126	* Privileged software exceptions:
16127	* Figure out if the exception must be delivered to the guest or the nested-guest.
16128	*/
16129	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
16130	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
16131	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
16132	{
16133	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16134	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16135	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16136	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16137
16138	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16139	bool const fIntercept = CPUMIsGuestVmxXcptInterceptSet(pVCpu, pCtx, VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo),
16140	pVmxTransient->uExitIntErrorCode);
16141	if (fIntercept)
16142	{
16143	/* Exit qualification is required for debug and page-fault exceptions. */
16144	hmR0VmxReadExitQualVmcs(pVmxTransient);
16145
16146	/*
16147	* For VM-exits due to software exceptions (those generated by INT3 or INTO) and privileged
16148	* software exceptions (those generated by INT1/ICEBP) we need to supply the VM-exit instruction
16149	* length. However, if delivery of a software interrupt, software exception or privileged
16150	* software exception causes a VM-exit, that too provides the VM-exit instruction length.
16151	*/
16152	VMXVEXITINFO ExitInfo;
16153	RT_ZERO(ExitInfo);
16154	ExitInfo.uReason = pVmxTransient->uExitReason;
16155	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16156	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16157
16158	VMXVEXITEVENTINFO ExitEventInfo;
16159	RT_ZERO(ExitEventInfo);
16160	ExitEventInfo.uExitIntInfo = pVmxTransient->uExitIntInfo;
16161	ExitEventInfo.uExitIntErrCode = pVmxTransient->uExitIntErrorCode;
16162	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
16163	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
16164
16165	#ifdef DEBUG_ramshankar
16166	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16167	Log4Func(("exit_int_info=%#RX32 err_code=%#RX32 exit_qual=%#RX64\n", pVmxTransient->uExitIntInfo,
16168	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual));
16169	if (VMX_IDT_VECTORING_INFO_IS_VALID(pVmxTransient->uIdtVectoringInfo))
16170	{
16171	Log4Func(("idt_info=%#RX32 idt_errcode=%#RX32 cr2=%#RX64\n", pVmxTransient->uIdtVectoringInfo,
16172	pVmxTransient->uIdtVectoringErrorCode, pCtx->cr2));
16173	}
16174	#endif
16175	return IEMExecVmxVmexitXcpt(pVCpu, &ExitInfo, &ExitEventInfo);
16176	}
16177
16178	/* Nested paging is currently a requirement, otherwise we would need to handle shadow #PFs in hmR0VmxExitXcptPF. */
16179	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
16180	return hmR0VmxExitXcpt(pVCpu, pVmxTransient);
16181	}
16182
16183	/*
16184	* Software interrupts:
16185	* VM-exits cannot be caused by software interrupts.
16186	*
16187	* External interrupts:
16188	* This should only happen when "acknowledge external interrupts on VM-exit"
16189	* control is set. However, we never set this when executing a guest or
16190	* nested-guest. For nested-guests it is emulated while injecting interrupts into
16191	* the guest.
16192	*/
16193	case VMX_EXIT_INT_INFO_TYPE_SW_INT:
16194	case VMX_EXIT_INT_INFO_TYPE_EXT_INT:
16195	default:
16196	{
16197	pVCpu->hm.s.u32HMError = pVmxTransient->uExitIntInfo;
16198	return VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
16199	}
16200	}
16201	}
16202
16203
16204	/**
16205	* Nested-guest VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT).
16206	* Unconditional VM-exit.
16207	*/
16208	HMVMX_EXIT_DECL hmR0VmxExitTripleFaultNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16209	{
16210	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16211	return IEMExecVmxVmexitTripleFault(pVCpu);
16212	}
16213
16214
16215	/**
16216	* Nested-guest VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
16217	*/
16218	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindowNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16219	{
16220	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16221
16222	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INT_WINDOW_EXIT))
16223	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
16224	return hmR0VmxExitIntWindow(pVCpu, pVmxTransient);
16225	}
16226
16227
16228	/**
16229	* Nested-guest VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
16230	*/
16231	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindowNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16232	{
16233	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16234
16235	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_NMI_WINDOW_EXIT))
16236	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
16237	return hmR0VmxExitIntWindow(pVCpu, pVmxTransient);
16238	}
16239
16240
16241	/**
16242	* Nested-guest VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH).
16243	* Unconditional VM-exit.
16244	*/
16245	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitchNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16246	{
16247	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16248
16249	hmR0VmxReadExitQualVmcs(pVmxTransient);
16250	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16251	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16252	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16253
16254	VMXVEXITINFO ExitInfo;
16255	RT_ZERO(ExitInfo);
16256	ExitInfo.uReason = pVmxTransient->uExitReason;
16257	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16258	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16259
16260	VMXVEXITEVENTINFO ExitEventInfo;
16261	RT_ZERO(ExitEventInfo);
16262	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
16263	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
16264	return IEMExecVmxVmexitTaskSwitch(pVCpu, &ExitInfo, &ExitEventInfo);
16265	}
16266
16267
16268	/**
16269	* Nested-guest VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
16270	*/
16271	HMVMX_EXIT_DECL hmR0VmxExitHltNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16272	{
16273	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16274
16275	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_HLT_EXIT))
16276	{
16277	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16278	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16279	}
16280	return hmR0VmxExitHlt(pVCpu, pVmxTransient);
16281	}
16282
16283
16284	/**
16285	* Nested-guest VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
16286	*/
16287	HMVMX_EXIT_DECL hmR0VmxExitInvlpgNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16288	{
16289	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16290
16291	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INVLPG_EXIT))
16292	{
16293	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16294	hmR0VmxReadExitQualVmcs(pVmxTransient);
16295
16296	VMXVEXITINFO ExitInfo;
16297	RT_ZERO(ExitInfo);
16298	ExitInfo.uReason = pVmxTransient->uExitReason;
16299	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16300	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16301	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16302	}
16303	return hmR0VmxExitInvlpg(pVCpu, pVmxTransient);
16304	}
16305
16306
16307	/**
16308	* Nested-guest VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
16309	*/
16310	HMVMX_EXIT_DECL hmR0VmxExitRdpmcNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16311	{
16312	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16313
16314	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDPMC_EXIT))
16315	{
16316	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16317	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16318	}
16319	return hmR0VmxExitRdpmc(pVCpu, pVmxTransient);
16320	}
16321
16322
16323	/**
16324	* Nested-guest VM-exit handler for VMREAD (VMX_EXIT_VMREAD) and VMWRITE
16325	* (VMX_EXIT_VMWRITE). Conditional VM-exit.
16326	*/
16327	HMVMX_EXIT_DECL hmR0VmxExitVmreadVmwriteNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16328	{
16329	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16330
16331	Assert( pVmxTransient->uExitReason == VMX_EXIT_VMREAD
16332	\|\| pVmxTransient->uExitReason == VMX_EXIT_VMWRITE);
16333
16334	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16335
16336	uint8_t const iGReg = pVmxTransient->ExitInstrInfo.VmreadVmwrite.iReg2;
16337	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
16338	uint64_t u64VmcsField = pVCpu->cpum.GstCtx.aGRegs[iGReg].u64;
16339
16340	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
16341	if (!CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
16342	u64VmcsField &= UINT64_C(0xffffffff);
16343
16344	if (CPUMIsGuestVmxVmreadVmwriteInterceptSet(pVCpu, pVmxTransient->uExitReason, u64VmcsField))
16345	{
16346	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16347	hmR0VmxReadExitQualVmcs(pVmxTransient);
16348
16349	VMXVEXITINFO ExitInfo;
16350	RT_ZERO(ExitInfo);
16351	ExitInfo.uReason = pVmxTransient->uExitReason;
16352	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16353	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16354	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
16355	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16356	}
16357
16358	if (pVmxTransient->uExitReason == VMX_EXIT_VMREAD)
16359	return hmR0VmxExitVmread(pVCpu, pVmxTransient);
16360	return hmR0VmxExitVmwrite(pVCpu, pVmxTransient);
16361	}
16362
16363
16364	/**
16365	* Nested-guest VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
16366	*/
16367	HMVMX_EXIT_DECL hmR0VmxExitRdtscNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16368	{
16369	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16370
16371	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDTSC_EXIT))
16372	{
16373	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16374	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16375	}
16376
16377	return hmR0VmxExitRdtsc(pVCpu, pVmxTransient);
16378	}
16379
16380
16381	/**
16382	* Nested-guest VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX).
16383	* Conditional VM-exit.
16384	*/
16385	HMVMX_EXIT_DECL hmR0VmxExitMovCRxNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16386	{
16387	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16388
16389	hmR0VmxReadExitQualVmcs(pVmxTransient);
16390	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16391
16392	VBOXSTRICTRC rcStrict;
16393	uint32_t const uAccessType = VMX_EXIT_QUAL_CRX_ACCESS(pVmxTransient->uExitQual);
16394	switch (uAccessType)
16395	{
16396	case VMX_EXIT_QUAL_CRX_ACCESS_WRITE:
16397	{
16398	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
16399	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(pVmxTransient->uExitQual);
16400	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
16401	uint64_t const uNewCrX = pVCpu->cpum.GstCtx.aGRegs[iGReg].u64;
16402
16403	bool fIntercept;
16404	switch (iCrReg)
16405	{
16406	case 0:
16407	case 4:
16408	fIntercept = CPUMIsGuestVmxMovToCr0Cr4InterceptSet(pVCpu, &pVCpu->cpum.GstCtx, iCrReg, uNewCrX);
16409	break;
16410
16411	case 3:
16412	fIntercept = CPUMIsGuestVmxMovToCr3InterceptSet(pVCpu, uNewCrX);
16413	break;
16414
16415	case 8:
16416	fIntercept = CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_CR8_LOAD_EXIT);
16417	break;
16418
16419	default:
16420	fIntercept = false;
16421	break;
16422	}
16423	if (fIntercept)
16424	{
16425	VMXVEXITINFO ExitInfo;
16426	RT_ZERO(ExitInfo);
16427	ExitInfo.uReason = pVmxTransient->uExitReason;
16428	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16429	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16430	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16431	}
16432	else
16433	rcStrict = hmR0VmxExitMovToCrX(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
16434	break;
16435	}
16436
16437	case VMX_EXIT_QUAL_CRX_ACCESS_READ:
16438	{
16439	/*
16440	* CR0/CR4 reads do not cause VM-exits, the read-shadow is used (subject to masking).
16441	* CR2 reads do not cause a VM-exit.
16442	* CR3 reads cause a VM-exit depending on the "CR3 store exiting" control.
16443	* CR8 reads cause a VM-exit depending on the "CR8 store exiting" control.
16444	*/
16445	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
16446	if ( iCrReg == 3
16447	\|\| iCrReg == 8)
16448	{
16449	static const uint32_t s_auCrXReadIntercepts[] = { 0, 0, 0, VMX_PROC_CTLS_CR3_STORE_EXIT, 0,
16450	0, 0, 0, VMX_PROC_CTLS_CR8_STORE_EXIT };
16451	uint32_t const uIntercept = s_auCrXReadIntercepts[iCrReg];
16452	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, uIntercept))
16453	{
16454	VMXVEXITINFO ExitInfo;
16455	RT_ZERO(ExitInfo);
16456	ExitInfo.uReason = pVmxTransient->uExitReason;
16457	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16458	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16459	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16460	}
16461	else
16462	{
16463	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(pVmxTransient->uExitQual);
16464	rcStrict = hmR0VmxExitMovFromCrX(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
16465	}
16466	}
16467	else
16468	{
16469	AssertMsgFailed(("MOV from CR%d VM-exit must not happen\n", iCrReg));
16470	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, iCrReg);
16471	}
16472	break;
16473	}
16474
16475	case VMX_EXIT_QUAL_CRX_ACCESS_CLTS:
16476	{
16477	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
16478	Assert(pVmcsNstGst);
16479	uint64_t const uGstHostMask = pVmcsNstGst->u64Cr0Mask.u;
16480	uint64_t const uReadShadow = pVmcsNstGst->u64Cr0ReadShadow.u;
16481	if ( (uGstHostMask & X86_CR0_TS)
16482	&& (uReadShadow & X86_CR0_TS))
16483	{
16484	VMXVEXITINFO ExitInfo;
16485	RT_ZERO(ExitInfo);
16486	ExitInfo.uReason = pVmxTransient->uExitReason;
16487	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16488	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16489	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16490	}
16491	else
16492	rcStrict = hmR0VmxExitClts(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr);
16493	break;
16494	}
16495
16496	case VMX_EXIT_QUAL_CRX_ACCESS_LMSW: /* LMSW (Load Machine-Status Word into CR0) */
16497	{
16498	RTGCPTR GCPtrEffDst;
16499	uint16_t const uNewMsw = VMX_EXIT_QUAL_CRX_LMSW_DATA(pVmxTransient->uExitQual);
16500	bool const fMemOperand = VMX_EXIT_QUAL_CRX_LMSW_OP_MEM(pVmxTransient->uExitQual);
16501	if (fMemOperand)
16502	{
16503	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
16504	GCPtrEffDst = pVmxTransient->uGuestLinearAddr;
16505	}
16506	else
16507	GCPtrEffDst = NIL_RTGCPTR;
16508
16509	if (CPUMIsGuestVmxLmswInterceptSet(pVCpu, &pVCpu->cpum.GstCtx, uNewMsw))
16510	{
16511	VMXVEXITINFO ExitInfo;
16512	RT_ZERO(ExitInfo);
16513	ExitInfo.uReason = pVmxTransient->uExitReason;
16514	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16515	ExitInfo.u64GuestLinearAddr = GCPtrEffDst;
16516	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16517	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16518	}
16519	else
16520	rcStrict = hmR0VmxExitLmsw(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, uNewMsw, GCPtrEffDst);
16521	break;
16522	}
16523
16524	default:
16525	{
16526	AssertMsgFailed(("Unrecognized Mov CRX access type %#x\n", uAccessType));
16527	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, uAccessType);
16528	}
16529	}
16530
16531	if (rcStrict == VINF_IEM_RAISED_XCPT)
16532	{
16533	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16534	rcStrict = VINF_SUCCESS;
16535	}
16536	return rcStrict;
16537	}
16538
16539
16540	/**
16541	* Nested-guest VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX).
16542	* Conditional VM-exit.
16543	*/
16544	HMVMX_EXIT_DECL hmR0VmxExitMovDRxNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16545	{
16546	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16547
16548	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MOV_DR_EXIT))
16549	{
16550	hmR0VmxReadExitQualVmcs(pVmxTransient);
16551	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16552
16553	VMXVEXITINFO ExitInfo;
16554	RT_ZERO(ExitInfo);
16555	ExitInfo.uReason = pVmxTransient->uExitReason;
16556	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16557	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16558	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16559	}
16560	return hmR0VmxExitMovDRx(pVCpu, pVmxTransient);
16561	}
16562
16563
16564	/**
16565	* Nested-guest VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR).
16566	* Conditional VM-exit.
16567	*/
16568	HMVMX_EXIT_DECL hmR0VmxExitIoInstrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16569	{
16570	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16571
16572	hmR0VmxReadExitQualVmcs(pVmxTransient);
16573
16574	uint32_t const uIOPort = VMX_EXIT_QUAL_IO_PORT(pVmxTransient->uExitQual);
16575	uint8_t const uIOSize = VMX_EXIT_QUAL_IO_SIZE(pVmxTransient->uExitQual);
16576	AssertReturn(uIOSize <= 3 && uIOSize != 2, VERR_VMX_IPE_1);
16577
16578	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses in bytes. */
16579	uint8_t const cbAccess = s_aIOSizes[uIOSize];
16580	if (CPUMIsGuestVmxIoInterceptSet(pVCpu, uIOPort, cbAccess))
16581	{
16582	/*
16583	* IN/OUT instruction:
16584	* - Provides VM-exit instruction length.
16585	*
16586	* INS/OUTS instruction:
16587	* - Provides VM-exit instruction length.
16588	* - Provides Guest-linear address.
16589	* - Optionally provides VM-exit instruction info (depends on CPU feature).
16590	*/
16591	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16592	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16593
16594	/* Make sure we don't use stale/uninitialized VMX-transient info. below. */
16595	pVmxTransient->ExitInstrInfo.u = 0;
16596	pVmxTransient->uGuestLinearAddr = 0;
16597
16598	bool const fVmxInsOutsInfo = pVM->cpum.ro.GuestFeatures.fVmxInsOutInfo;
16599	bool const fIOString = VMX_EXIT_QUAL_IO_IS_STRING(pVmxTransient->uExitQual);
16600	if (fIOString)
16601	{
16602	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
16603	if (fVmxInsOutsInfo)
16604	{
16605	Assert(RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_INS_OUTS)); /* Paranoia. */
16606	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16607	}
16608	}
16609
16610	VMXVEXITINFO ExitInfo;
16611	RT_ZERO(ExitInfo);
16612	ExitInfo.uReason = pVmxTransient->uExitReason;
16613	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16614	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16615	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
16616	ExitInfo.u64GuestLinearAddr = pVmxTransient->uGuestLinearAddr;
16617	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16618	}
16619	return hmR0VmxExitIoInstr(pVCpu, pVmxTransient);
16620	}
16621
16622
16623	/**
16624	* Nested-guest VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
16625	*/
16626	HMVMX_EXIT_DECL hmR0VmxExitRdmsrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16627	{
16628	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16629
16630	uint32_t fMsrpm;
16631	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_MSR_BITMAPS))
16632	fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), pVCpu->cpum.GstCtx.ecx);
16633	else
16634	fMsrpm = VMXMSRPM_EXIT_RD;
16635
16636	if (fMsrpm & VMXMSRPM_EXIT_RD)
16637	{
16638	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16639	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16640	}
16641	return hmR0VmxExitRdmsr(pVCpu, pVmxTransient);
16642	}
16643
16644
16645	/**
16646	* Nested-guest VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
16647	*/
16648	HMVMX_EXIT_DECL hmR0VmxExitWrmsrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16649	{
16650	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16651
16652	uint32_t fMsrpm;
16653	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_MSR_BITMAPS))
16654	fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), pVCpu->cpum.GstCtx.ecx);
16655	else
16656	fMsrpm = VMXMSRPM_EXIT_WR;
16657
16658	if (fMsrpm & VMXMSRPM_EXIT_WR)
16659	{
16660	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16661	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16662	}
16663	return hmR0VmxExitWrmsr(pVCpu, pVmxTransient);
16664	}
16665
16666
16667	/**
16668	* Nested-guest VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
16669	*/
16670	HMVMX_EXIT_DECL hmR0VmxExitMwaitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16671	{
16672	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16673
16674	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MWAIT_EXIT))
16675	{
16676	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16677	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16678	}
16679	return hmR0VmxExitMwait(pVCpu, pVmxTransient);
16680	}
16681
16682
16683	/**
16684	* Nested-guest VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional
16685	* VM-exit.
16686	*/
16687	HMVMX_EXIT_DECL hmR0VmxExitMtfNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16688	{
16689	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16690
16691	/** @todo NSTVMX: Should consider debugging nested-guests using VM debugger. */
16692	hmR0VmxReadGuestPendingDbgXctps(pVmxTransient);
16693	VMXVEXITINFO ExitInfo;
16694	RT_ZERO(ExitInfo);
16695	ExitInfo.uReason = pVmxTransient->uExitReason;
16696	ExitInfo.u64GuestPendingDbgXcpts = pVmxTransient->uGuestPendingDbgXcpts;
16697	return IEMExecVmxVmexitTrapLike(pVCpu, &ExitInfo);
16698	}
16699
16700
16701	/**
16702	* Nested-guest VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
16703	*/
16704	HMVMX_EXIT_DECL hmR0VmxExitMonitorNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16705	{
16706	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16707
16708	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MONITOR_EXIT))
16709	{
16710	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16711	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16712	}
16713	return hmR0VmxExitMonitor(pVCpu, pVmxTransient);
16714	}
16715
16716
16717	/**
16718	* Nested-guest VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
16719	*/
16720	HMVMX_EXIT_DECL hmR0VmxExitPauseNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16721	{
16722	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16723
16724	/** @todo NSTVMX: Think about this more. Does the outer guest need to intercept
16725	* PAUSE when executing a nested-guest? If it does not, we would not need
16726	* to check for the intercepts here. Just call VM-exit... */
16727
16728	/* The CPU would have already performed the necessary CPL checks for PAUSE-loop exiting. */
16729	if ( CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_PAUSE_EXIT)
16730	\|\| CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_PAUSE_LOOP_EXIT))
16731	{
16732	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16733	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16734	}
16735	return hmR0VmxExitPause(pVCpu, pVmxTransient);
16736	}
16737
16738
16739	/**
16740	* Nested-guest VM-exit handler for when the TPR value is lowered below the
16741	* specified threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
16742	*/
16743	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThresholdNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16744	{
16745	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16746
16747	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_TPR_SHADOW))
16748	{
16749	hmR0VmxReadGuestPendingDbgXctps(pVmxTransient);
16750	VMXVEXITINFO ExitInfo;
16751	RT_ZERO(ExitInfo);
16752	ExitInfo.uReason = pVmxTransient->uExitReason;
16753	ExitInfo.u64GuestPendingDbgXcpts = pVmxTransient->uGuestPendingDbgXcpts;
16754	return IEMExecVmxVmexitTrapLike(pVCpu, &ExitInfo);
16755	}
16756	return hmR0VmxExitTprBelowThreshold(pVCpu, pVmxTransient);
16757	}
16758
16759
16760	/**
16761	* Nested-guest VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional
16762	* VM-exit.
16763	*/
16764	HMVMX_EXIT_DECL hmR0VmxExitApicAccessNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16765	{
16766	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16767
16768	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16769	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16770	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16771	hmR0VmxReadExitQualVmcs(pVmxTransient);
16772
16773	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_VIRT_APIC_ACCESS));
16774
16775	Log4Func(("at offset %#x type=%u\n", VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual),
16776	VMX_EXIT_QUAL_APIC_ACCESS_TYPE(pVmxTransient->uExitQual)));
16777
16778	VMXVEXITINFO ExitInfo;
16779	RT_ZERO(ExitInfo);
16780	ExitInfo.uReason = pVmxTransient->uExitReason;
16781	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16782	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16783
16784	VMXVEXITEVENTINFO ExitEventInfo;
16785	RT_ZERO(ExitEventInfo);
16786	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
16787	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
16788	return IEMExecVmxVmexitApicAccess(pVCpu, &ExitInfo, &ExitEventInfo);
16789	}
16790
16791
16792	/**
16793	* Nested-guest VM-exit handler for APIC write emulation (VMX_EXIT_APIC_WRITE).
16794	* Conditional VM-exit.
16795	*/
16796	HMVMX_EXIT_DECL hmR0VmxExitApicWriteNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16797	{
16798	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16799
16800	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_APIC_REG_VIRT));
16801	hmR0VmxReadExitQualVmcs(pVmxTransient);
16802	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
16803	}
16804
16805
16806	/**
16807	* Nested-guest VM-exit handler for virtualized EOI (VMX_EXIT_VIRTUALIZED_EOI).
16808	* Conditional VM-exit.
16809	*/
16810	HMVMX_EXIT_DECL hmR0VmxExitVirtEoiNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16811	{
16812	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16813
16814	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_VIRT_INT_DELIVERY));
16815	hmR0VmxReadExitQualVmcs(pVmxTransient);
16816	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
16817	}
16818
16819
16820	/**
16821	* Nested-guest VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
16822	*/
16823	HMVMX_EXIT_DECL hmR0VmxExitRdtscpNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16824	{
16825	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16826
16827	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDTSC_EXIT))
16828	{
16829	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_RDTSCP));
16830	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16831	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16832	}
16833	return hmR0VmxExitRdtscp(pVCpu, pVmxTransient);
16834	}
16835
16836
16837	/**
16838	* Nested-guest VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
16839	*/
16840	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvdNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16841	{
16842	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16843
16844	if (CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_WBINVD_EXIT))
16845	{
16846	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16847	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16848	}
16849	return hmR0VmxExitWbinvd(pVCpu, pVmxTransient);
16850	}
16851
16852
16853	/**
16854	* Nested-guest VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
16855	*/
16856	HMVMX_EXIT_DECL hmR0VmxExitInvpcidNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16857	{
16858	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16859
16860	if (CPUMIsGuestVmxProcCtlsSet(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INVLPG_EXIT))
16861	{
16862	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, &pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_INVPCID));
16863	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16864	hmR0VmxReadExitQualVmcs(pVmxTransient);
16865	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16866
16867	VMXVEXITINFO ExitInfo;
16868	RT_ZERO(ExitInfo);
16869	ExitInfo.uReason = pVmxTransient->uExitReason;
16870	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16871	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16872	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
16873	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16874	}
16875	return hmR0VmxExitInvpcid(pVCpu, pVmxTransient);
16876	}
16877
16878
16879	/**
16880	* Nested-guest VM-exit handler for invalid-guest state
16881	* (VMX_EXIT_ERR_INVALID_GUEST_STATE). Error VM-exit.
16882	*/
16883	HMVMX_EXIT_DECL hmR0VmxExitErrInvalidGuestStateNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16884	{
16885	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16886
16887	/*
16888	* Currently this should never happen because we fully emulate VMLAUNCH/VMRESUME in IEM.
16889	* So if it does happen, it indicates a bug possibly in the hardware-assisted VMX code.
16890	* Handle it like it's in an invalid guest state of the outer guest.
16891	*
16892	* When the fast path is implemented, this should be changed to cause the corresponding
16893	* nested-guest VM-exit.
16894	*/
16895	return hmR0VmxExitErrInvalidGuestState(pVCpu, pVmxTransient);
16896	}
16897
16898
16899	/**
16900	* Nested-guest VM-exit handler for instructions that cause VM-exits uncondtionally
16901	* and only provide the instruction length.
16902	*
16903	* Unconditional VM-exit.
16904	*/
16905	HMVMX_EXIT_DECL hmR0VmxExitInstrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16906	{
16907	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16908
16909	#ifdef VBOX_STRICT
16910	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16911	switch (pVmxTransient->uExitReason)
16912	{
16913	case VMX_EXIT_ENCLS:
16914	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_ENCLS_EXIT));
16915	break;
16916
16917	case VMX_EXIT_VMFUNC:
16918	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_VMFUNC));
16919	break;
16920	}
16921	#endif
16922
16923	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16924	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16925	}
16926
16927
16928	/**
16929	* Nested-guest VM-exit handler for instructions that provide instruction length as
16930	* well as more information.
16931	*
16932	* Unconditional VM-exit.
16933	*/
16934	HMVMX_EXIT_DECL hmR0VmxExitInstrWithInfoNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16935	{
16936	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16937
16938	#ifdef VBOX_STRICT
16939	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16940	switch (pVmxTransient->uExitReason)
16941	{
16942	case VMX_EXIT_GDTR_IDTR_ACCESS:
16943	case VMX_EXIT_LDTR_TR_ACCESS:
16944	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_DESC_TABLE_EXIT));
16945	break;
16946
16947	case VMX_EXIT_RDRAND:
16948	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_RDRAND_EXIT));
16949	break;
16950
16951	case VMX_EXIT_RDSEED:
16952	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_RDSEED_EXIT));
16953	break;
16954
16955	case VMX_EXIT_XSAVES:
16956	case VMX_EXIT_XRSTORS:
16957	/** @todo NSTVMX: Verify XSS-bitmap. */
16958	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_XSAVES_XRSTORS));
16959	break;
16960
16961	case VMX_EXIT_UMWAIT:
16962	case VMX_EXIT_TPAUSE:
16963	Assert(CPUMIsGuestVmxProcCtlsSet(pVCpu, pCtx, VMX_PROC_CTLS_RDTSC_EXIT));
16964	Assert(CPUMIsGuestVmxProcCtls2Set(pVCpu, pCtx, VMX_PROC_CTLS2_USER_WAIT_PAUSE));
16965	break;
16966	}
16967	#endif
16968
16969	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16970	hmR0VmxReadExitQualVmcs(pVmxTransient);
16971	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16972
16973	VMXVEXITINFO ExitInfo;
16974	RT_ZERO(ExitInfo);
16975	ExitInfo.uReason = pVmxTransient->uExitReason;
16976	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16977	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16978	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
16979	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16980	}
16981
16982	/** @} */
16983	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
16984

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

source: vbox/trunk/src/VBox/VMM/VMMR0/HMVMXR0.cpp@ 80580

Download in other formats: