HMVMXR0.cpp@ 92679

Last change on this file since 92679 was 92626, checked in by vboxsync, 3 years ago
VMM: Nested VMX: bugref:10092 Adjust PGM APIs and translate nested-guest CR3 prior to mapping them when switching mode and other places.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 740.0 KB

Line
1	/* $Id: HMVMXR0.cpp 92626 2021-11-29 12:32:58Z vboxsync $ */
2	/** @file
3	* HM VMX (Intel VT-x) - Host Context Ring-0.
4	*/
5
6	/*
7	* Copyright (C) 2012-2020 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	#include "HMInternal.h"
39	#include <VBox/vmm/vmcc.h>
40	#include <VBox/vmm/hmvmxinline.h>
41	#include "HMVMXR0.h"
42	#include "dtrace/VBoxVMM.h"
43
44	#ifdef DEBUG_ramshankar
45	# define HMVMX_ALWAYS_SAVE_GUEST_RFLAGS
46	# define HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
47	# define HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE
48	# define HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
49	# define HMVMX_ALWAYS_CLEAN_TRANSIENT
50	# define HMVMX_ALWAYS_CHECK_GUEST_STATE
51	# define HMVMX_ALWAYS_TRAP_ALL_XCPTS
52	# define HMVMX_ALWAYS_TRAP_PF
53	# define HMVMX_ALWAYS_FLUSH_TLB
54	# define HMVMX_ALWAYS_SWAP_EFER
55	#endif
56
57
58	/*********************************************************************************************************************************
59	* Defined Constants And Macros *
60	*********************************************************************************************************************************/
61	/** Use the function table. */
62	#define HMVMX_USE_FUNCTION_TABLE
63
64	/** Determine which tagged-TLB flush handler to use. */
65	#define HMVMX_FLUSH_TAGGED_TLB_EPT_VPID 0
66	#define HMVMX_FLUSH_TAGGED_TLB_EPT 1
67	#define HMVMX_FLUSH_TAGGED_TLB_VPID 2
68	#define HMVMX_FLUSH_TAGGED_TLB_NONE 3
69
70	/**
71	* Flags to skip redundant reads of some common VMCS fields that are not part of
72	* the guest-CPU or VCPU state but are needed while handling VM-exits.
73	*/
74	#define HMVMX_READ_IDT_VECTORING_INFO RT_BIT_32(0)
75	#define HMVMX_READ_IDT_VECTORING_ERROR_CODE RT_BIT_32(1)
76	#define HMVMX_READ_EXIT_QUALIFICATION RT_BIT_32(2)
77	#define HMVMX_READ_EXIT_INSTR_LEN RT_BIT_32(3)
78	#define HMVMX_READ_EXIT_INTERRUPTION_INFO RT_BIT_32(4)
79	#define HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE RT_BIT_32(5)
80	#define HMVMX_READ_EXIT_INSTR_INFO RT_BIT_32(6)
81	#define HMVMX_READ_GUEST_LINEAR_ADDR RT_BIT_32(7)
82	#define HMVMX_READ_GUEST_PHYSICAL_ADDR RT_BIT_32(8)
83	#define HMVMX_READ_GUEST_PENDING_DBG_XCPTS RT_BIT_32(9)
84
85	/** All the VMCS fields required for processing of exception/NMI VM-exits. */
86	#define HMVMX_READ_XCPT_INFO ( HMVMX_READ_EXIT_INTERRUPTION_INFO \
87	\| HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE \
88	\| HMVMX_READ_EXIT_INSTR_LEN \
89	\| HMVMX_READ_IDT_VECTORING_INFO \
90	\| HMVMX_READ_IDT_VECTORING_ERROR_CODE)
91
92	/** Assert that all the given fields have been read from the VMCS. */
93	#ifdef VBOX_STRICT
94	# define HMVMX_ASSERT_READ(a_pVmxTransient, a_fReadFields) \
95	do { \
96	uint32_t const fVmcsFieldRead = ASMAtomicUoReadU32(&pVmxTransient->fVmcsFieldsRead); \
97	Assert((fVmcsFieldRead & (a_fReadFields)) == (a_fReadFields)); \
98	} while (0)
99	#else
100	# define HMVMX_ASSERT_READ(a_pVmxTransient, a_fReadFields) do { } while (0)
101	#endif
102
103	/**
104	* Subset of the guest-CPU state that is kept by VMX R0 code while executing the
105	* guest using hardware-assisted VMX.
106	*
107	* This excludes state like GPRs (other than RSP) which are always are
108	* swapped and restored across the world-switch and also registers like EFER,
109	* MSR which cannot be modified by the guest without causing a VM-exit.
110	*/
111	#define HMVMX_CPUMCTX_EXTRN_ALL ( CPUMCTX_EXTRN_RIP \
112	\| CPUMCTX_EXTRN_RFLAGS \
113	\| CPUMCTX_EXTRN_RSP \
114	\| CPUMCTX_EXTRN_SREG_MASK \
115	\| CPUMCTX_EXTRN_TABLE_MASK \
116	\| CPUMCTX_EXTRN_KERNEL_GS_BASE \
117	\| CPUMCTX_EXTRN_SYSCALL_MSRS \
118	\| CPUMCTX_EXTRN_SYSENTER_MSRS \
119	\| CPUMCTX_EXTRN_TSC_AUX \
120	\| CPUMCTX_EXTRN_OTHER_MSRS \
121	\| CPUMCTX_EXTRN_CR0 \
122	\| CPUMCTX_EXTRN_CR3 \
123	\| CPUMCTX_EXTRN_CR4 \
124	\| CPUMCTX_EXTRN_DR7 \
125	\| CPUMCTX_EXTRN_HWVIRT \
126	\| CPUMCTX_EXTRN_INHIBIT_INT \
127	\| CPUMCTX_EXTRN_INHIBIT_NMI)
128
129	/**
130	* Exception bitmap mask for real-mode guests (real-on-v86).
131	*
132	* We need to intercept all exceptions manually except:
133	* - \#AC and \#DB are always intercepted to prevent the CPU from deadlocking
134	* due to bugs in Intel CPUs.
135	* - \#PF need not be intercepted even in real-mode if we have nested paging
136	* support.
137	*/
138	#define HMVMX_REAL_MODE_XCPT_MASK ( RT_BIT(X86_XCPT_DE) /* always: \| RT_BIT(X86_XCPT_DB) */ \| RT_BIT(X86_XCPT_NMI) \
139	\| RT_BIT(X86_XCPT_BP) \| RT_BIT(X86_XCPT_OF) \| RT_BIT(X86_XCPT_BR) \
140	\| RT_BIT(X86_XCPT_UD) \| RT_BIT(X86_XCPT_NM) \| RT_BIT(X86_XCPT_DF) \
141	\| RT_BIT(X86_XCPT_CO_SEG_OVERRUN) \| RT_BIT(X86_XCPT_TS) \| RT_BIT(X86_XCPT_NP) \
142	\| RT_BIT(X86_XCPT_SS) \| RT_BIT(X86_XCPT_GP) /* RT_BIT(X86_XCPT_PF) */ \
143	\| RT_BIT(X86_XCPT_MF) /* always: \| RT_BIT(X86_XCPT_AC) */ \| RT_BIT(X86_XCPT_MC) \
144	\| RT_BIT(X86_XCPT_XF))
145
146	/** Maximum VM-instruction error number. */
147	#define HMVMX_INSTR_ERROR_MAX 28
148
149	/** Profiling macro. */
150	#ifdef HM_PROFILE_EXIT_DISPATCH
151	# define HMVMX_START_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitDispatch, ed)
152	# define HMVMX_STOP_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitDispatch, ed)
153	#else
154	# define HMVMX_START_EXIT_DISPATCH_PROF() do { } while (0)
155	# define HMVMX_STOP_EXIT_DISPATCH_PROF() do { } while (0)
156	#endif
157
158	/** Assert that preemption is disabled or covered by thread-context hooks. */
159	#define HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu) Assert( VMMR0ThreadCtxHookIsEnabled((a_pVCpu)) \
160	\|\| !RTThreadPreemptIsEnabled(NIL_RTTHREAD))
161
162	/** Assert that we haven't migrated CPUs when thread-context hooks are not
163	* used. */
164	#define HMVMX_ASSERT_CPU_SAFE(a_pVCpu) AssertMsg( VMMR0ThreadCtxHookIsEnabled((a_pVCpu)) \
165	\|\| (a_pVCpu)->hmr0.s.idEnteredCpu == RTMpCpuId(), \
166	("Illegal migration! Entered on CPU %u Current %u\n", \
167	(a_pVCpu)->hmr0.s.idEnteredCpu, RTMpCpuId()))
168
169	/** Asserts that the given CPUMCTX_EXTRN_XXX bits are present in the guest-CPU
170	* context. */
171	#define HMVMX_CPUMCTX_ASSERT(a_pVCpu, a_fExtrnMbz) AssertMsg(!((a_pVCpu)->cpum.GstCtx.fExtrn & (a_fExtrnMbz)), \
172	("fExtrn=%#RX64 fExtrnMbz=%#RX64\n", \
173	(a_pVCpu)->cpum.GstCtx.fExtrn, (a_fExtrnMbz)))
174
175	/** Log the VM-exit reason with an easily visible marker to identify it in a
176	* potential sea of logging data. */
177	#define HMVMX_LOG_EXIT(a_pVCpu, a_uExitReason) \
178	do { \
179	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, \
180	HMGetVmxExitName(a_uExitReason))); \
181	} while (0) \
182
183
184	/*********************************************************************************************************************************
185	* Structures and Typedefs *
186	*********************************************************************************************************************************/
187	/**
188	* VMX per-VCPU transient state.
189	*
190	* A state structure for holding miscellaneous information across
191	* VMX non-root operation and restored after the transition.
192	*
193	* Note: The members are ordered and aligned such that the most
194	* frequently used ones (in the guest execution loop) fall within
195	* the first cache line.
196	*/
197	typedef struct VMXTRANSIENT
198	{
199	/** Mask of currently read VMCS fields; HMVMX_READ_XXX. */
200	uint32_t fVmcsFieldsRead;
201	/** The guest's TPR value used for TPR shadowing. */
202	uint8_t u8GuestTpr;
203	uint8_t abAlignment0[3];
204
205	/** Whether the VM-exit was caused by a page-fault during delivery of an
206	* external interrupt or NMI. */
207	bool fVectoringPF;
208	/** Whether the VM-exit was caused by a page-fault during delivery of a
209	* contributory exception or a page-fault. */
210	bool fVectoringDoublePF;
211	/** Whether the VM-entry failed or not. */
212	bool fVMEntryFailed;
213	/** Whether the TSC_AUX MSR needs to be removed from the auto-load/store MSR
214	* area after VM-exit. */
215	bool fRemoveTscAuxMsr;
216	/** Whether TSC-offsetting and VMX-preemption timer was updated before VM-entry. */
217	bool fUpdatedTscOffsettingAndPreemptTimer;
218	/** Whether we are currently executing a nested-guest. */
219	bool fIsNestedGuest;
220	/** Whether the guest debug state was active at the time of VM-exit. */
221	bool fWasGuestDebugStateActive;
222	/** Whether the hyper debug state was active at the time of VM-exit. */
223	bool fWasHyperDebugStateActive;
224
225	/** The basic VM-exit reason. */
226	uint32_t uExitReason;
227	/** The VM-exit interruption error code. */
228	uint32_t uExitIntErrorCode;
229
230	/** The host's rflags/eflags. */
231	RTCCUINTREG fEFlags;
232
233	/** The VM-exit exit code qualification. */
234	uint64_t uExitQual;
235
236	/** The VMCS info. object. */
237	PVMXVMCSINFO pVmcsInfo;
238
239	/** The VM-exit interruption-information field. */
240	uint32_t uExitIntInfo;
241	/** The VM-exit instruction-length field. */
242	uint32_t cbExitInstr;
243
244	/** The VM-exit instruction-information field. */
245	VMXEXITINSTRINFO ExitInstrInfo;
246	/** IDT-vectoring information field. */
247	uint32_t uIdtVectoringInfo;
248
249	/** IDT-vectoring error code. */
250	uint32_t uIdtVectoringErrorCode;
251	uint32_t u32Alignment0;
252
253	/** The Guest-linear address. */
254	uint64_t uGuestLinearAddr;
255
256	/** The Guest-physical address. */
257	uint64_t uGuestPhysicalAddr;
258
259	/** The Guest pending-debug exceptions. */
260	uint64_t uGuestPendingDbgXcpts;
261
262	/** The VM-entry interruption-information field. */
263	uint32_t uEntryIntInfo;
264	/** The VM-entry exception error code field. */
265	uint32_t uEntryXcptErrorCode;
266
267	/** The VM-entry instruction length field. */
268	uint32_t cbEntryInstr;
269	} VMXTRANSIENT;
270	AssertCompileMemberSize(VMXTRANSIENT, ExitInstrInfo, sizeof(uint32_t));
271	AssertCompileMemberAlignment(VMXTRANSIENT, fVmcsFieldsRead, 8);
272	AssertCompileMemberAlignment(VMXTRANSIENT, fVectoringPF, 8);
273	AssertCompileMemberAlignment(VMXTRANSIENT, uExitReason, 8);
274	AssertCompileMemberAlignment(VMXTRANSIENT, fEFlags, 8);
275	AssertCompileMemberAlignment(VMXTRANSIENT, uExitQual, 8);
276	AssertCompileMemberAlignment(VMXTRANSIENT, pVmcsInfo, 8);
277	AssertCompileMemberAlignment(VMXTRANSIENT, uExitIntInfo, 8);
278	AssertCompileMemberAlignment(VMXTRANSIENT, ExitInstrInfo, 8);
279	AssertCompileMemberAlignment(VMXTRANSIENT, uIdtVectoringErrorCode, 8);
280	AssertCompileMemberAlignment(VMXTRANSIENT, uGuestLinearAddr, 8);
281	AssertCompileMemberAlignment(VMXTRANSIENT, uGuestPhysicalAddr, 8);
282	AssertCompileMemberAlignment(VMXTRANSIENT, uEntryIntInfo, 8);
283	AssertCompileMemberAlignment(VMXTRANSIENT, cbEntryInstr, 8);
284	/** Pointer to VMX transient state. */
285	typedef VMXTRANSIENT *PVMXTRANSIENT;
286	/** Pointer to a const VMX transient state. */
287	typedef const VMXTRANSIENT *PCVMXTRANSIENT;
288
289	/**
290	* VMX page allocation information.
291	*/
292	typedef struct
293	{
294	uint32_t fValid; /*< Whether to allocate this page (e.g, based on a CPU feature). /
295	uint32_t uPadding0; /*< Padding to ensure array of these structs are aligned to a multiple of 8. /
296	PRTHCPHYS pHCPhys; /*< Where to store the host-physical address of the allocation. /
297	PRTR0PTR ppVirt; /*< Where to store the host-virtual address of the allocation. /
298	} VMXPAGEALLOCINFO;
299	/** Pointer to VMX page-allocation info. */
300	typedef VMXPAGEALLOCINFO *PVMXPAGEALLOCINFO;
301	/** Pointer to a const VMX page-allocation info. */
302	typedef const VMXPAGEALLOCINFO *PCVMXPAGEALLOCINFO;
303	AssertCompileSizeAlignment(VMXPAGEALLOCINFO, 8);
304
305	/**
306	* Memory operand read or write access.
307	*/
308	typedef enum VMXMEMACCESS
309	{
310	VMXMEMACCESS_READ = 0,
311	VMXMEMACCESS_WRITE = 1
312	} VMXMEMACCESS;
313
314	/**
315	* VMX VM-exit handler.
316	*
317	* @returns Strict VBox status code (i.e. informational status codes too).
318	* @param pVCpu The cross context virtual CPU structure.
319	* @param pVmxTransient The VMX-transient structure.
320	*/
321	#ifndef HMVMX_USE_FUNCTION_TABLE
322	typedef VBOXSTRICTRC FNVMXEXITHANDLER(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
323	#else
324	typedef DECLCALLBACKTYPE(VBOXSTRICTRC, FNVMXEXITHANDLER,(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient));
325	/** Pointer to VM-exit handler. */
326	typedef FNVMXEXITHANDLER *PFNVMXEXITHANDLER;
327	#endif
328
329	/**
330	* VMX VM-exit handler, non-strict status code.
331	*
332	* This is generally the same as FNVMXEXITHANDLER, the NSRC bit is just FYI.
333	*
334	* @returns VBox status code, no informational status code returned.
335	* @param pVCpu The cross context virtual CPU structure.
336	* @param pVmxTransient The VMX-transient structure.
337	*
338	* @remarks This is not used on anything returning VERR_EM_INTERPRETER as the
339	* use of that status code will be replaced with VINF_EM_SOMETHING
340	* later when switching over to IEM.
341	*/
342	#ifndef HMVMX_USE_FUNCTION_TABLE
343	typedef int FNVMXEXITHANDLERNSRC(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
344	#else
345	typedef FNVMXEXITHANDLER FNVMXEXITHANDLERNSRC;
346	#endif
347
348
349	/*********************************************************************************************************************************
350	* Internal Functions *
351	*********************************************************************************************************************************/
352	#ifndef HMVMX_USE_FUNCTION_TABLE
353	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
354	# define HMVMX_EXIT_DECL DECLINLINE(VBOXSTRICTRC)
355	# define HMVMX_EXIT_NSRC_DECL DECLINLINE(int)
356	#else
357	# define HMVMX_EXIT_DECL static DECLCALLBACK(VBOXSTRICTRC)
358	# define HMVMX_EXIT_NSRC_DECL HMVMX_EXIT_DECL
359	#endif
360	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
361	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
362	#endif
363
364	static int hmR0VmxImportGuestState(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint64_t fWhat);
365
366	/** @name VM-exit handler prototypes.
367	* @{
368	*/
369	static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmi;
370	static FNVMXEXITHANDLER hmR0VmxExitExtInt;
371	static FNVMXEXITHANDLER hmR0VmxExitTripleFault;
372	static FNVMXEXITHANDLERNSRC hmR0VmxExitIntWindow;
373	static FNVMXEXITHANDLERNSRC hmR0VmxExitNmiWindow;
374	static FNVMXEXITHANDLER hmR0VmxExitTaskSwitch;
375	static FNVMXEXITHANDLER hmR0VmxExitCpuid;
376	static FNVMXEXITHANDLER hmR0VmxExitGetsec;
377	static FNVMXEXITHANDLER hmR0VmxExitHlt;
378	static FNVMXEXITHANDLERNSRC hmR0VmxExitInvd;
379	static FNVMXEXITHANDLER hmR0VmxExitInvlpg;
380	static FNVMXEXITHANDLER hmR0VmxExitRdpmc;
381	static FNVMXEXITHANDLER hmR0VmxExitVmcall;
382	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
383	static FNVMXEXITHANDLER hmR0VmxExitVmclear;
384	static FNVMXEXITHANDLER hmR0VmxExitVmlaunch;
385	static FNVMXEXITHANDLER hmR0VmxExitVmptrld;
386	static FNVMXEXITHANDLER hmR0VmxExitVmptrst;
387	static FNVMXEXITHANDLER hmR0VmxExitVmread;
388	static FNVMXEXITHANDLER hmR0VmxExitVmresume;
389	static FNVMXEXITHANDLER hmR0VmxExitVmwrite;
390	static FNVMXEXITHANDLER hmR0VmxExitVmxoff;
391	static FNVMXEXITHANDLER hmR0VmxExitVmxon;
392	static FNVMXEXITHANDLER hmR0VmxExitInvvpid;
393	#endif
394	static FNVMXEXITHANDLER hmR0VmxExitRdtsc;
395	static FNVMXEXITHANDLER hmR0VmxExitMovCRx;
396	static FNVMXEXITHANDLER hmR0VmxExitMovDRx;
397	static FNVMXEXITHANDLER hmR0VmxExitIoInstr;
398	static FNVMXEXITHANDLER hmR0VmxExitRdmsr;
399	static FNVMXEXITHANDLER hmR0VmxExitWrmsr;
400	static FNVMXEXITHANDLER hmR0VmxExitMwait;
401	static FNVMXEXITHANDLER hmR0VmxExitMtf;
402	static FNVMXEXITHANDLER hmR0VmxExitMonitor;
403	static FNVMXEXITHANDLER hmR0VmxExitPause;
404	static FNVMXEXITHANDLERNSRC hmR0VmxExitTprBelowThreshold;
405	static FNVMXEXITHANDLER hmR0VmxExitApicAccess;
406	static FNVMXEXITHANDLER hmR0VmxExitEptViolation;
407	static FNVMXEXITHANDLER hmR0VmxExitEptMisconfig;
408	static FNVMXEXITHANDLER hmR0VmxExitRdtscp;
409	static FNVMXEXITHANDLER hmR0VmxExitPreemptTimer;
410	static FNVMXEXITHANDLERNSRC hmR0VmxExitWbinvd;
411	static FNVMXEXITHANDLER hmR0VmxExitXsetbv;
412	static FNVMXEXITHANDLER hmR0VmxExitInvpcid;
413	static FNVMXEXITHANDLERNSRC hmR0VmxExitSetPendingXcptUD;
414	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrInvalidGuestState;
415	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrUnexpected;
416	/** @} */
417
418	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
419	/** @name Nested-guest VM-exit handler prototypes.
420	* @{
421	*/
422	static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmiNested;
423	static FNVMXEXITHANDLER hmR0VmxExitTripleFaultNested;
424	static FNVMXEXITHANDLERNSRC hmR0VmxExitIntWindowNested;
425	static FNVMXEXITHANDLERNSRC hmR0VmxExitNmiWindowNested;
426	static FNVMXEXITHANDLER hmR0VmxExitTaskSwitchNested;
427	static FNVMXEXITHANDLER hmR0VmxExitHltNested;
428	static FNVMXEXITHANDLER hmR0VmxExitInvlpgNested;
429	static FNVMXEXITHANDLER hmR0VmxExitRdpmcNested;
430	static FNVMXEXITHANDLER hmR0VmxExitVmreadVmwriteNested;
431	static FNVMXEXITHANDLER hmR0VmxExitRdtscNested;
432	static FNVMXEXITHANDLER hmR0VmxExitMovCRxNested;
433	static FNVMXEXITHANDLER hmR0VmxExitMovDRxNested;
434	static FNVMXEXITHANDLER hmR0VmxExitIoInstrNested;
435	static FNVMXEXITHANDLER hmR0VmxExitRdmsrNested;
436	static FNVMXEXITHANDLER hmR0VmxExitWrmsrNested;
437	static FNVMXEXITHANDLER hmR0VmxExitMwaitNested;
438	static FNVMXEXITHANDLER hmR0VmxExitMtfNested;
439	static FNVMXEXITHANDLER hmR0VmxExitMonitorNested;
440	static FNVMXEXITHANDLER hmR0VmxExitPauseNested;
441	static FNVMXEXITHANDLERNSRC hmR0VmxExitTprBelowThresholdNested;
442	static FNVMXEXITHANDLER hmR0VmxExitApicAccessNested;
443	static FNVMXEXITHANDLER hmR0VmxExitApicWriteNested;
444	static FNVMXEXITHANDLER hmR0VmxExitVirtEoiNested;
445	static FNVMXEXITHANDLER hmR0VmxExitRdtscpNested;
446	static FNVMXEXITHANDLERNSRC hmR0VmxExitWbinvdNested;
447	static FNVMXEXITHANDLER hmR0VmxExitInvpcidNested;
448	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrInvalidGuestStateNested;
449	static FNVMXEXITHANDLER hmR0VmxExitInstrNested;
450	static FNVMXEXITHANDLER hmR0VmxExitInstrWithInfoNested;
451	/** @} */
452	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
453
454
455	/*********************************************************************************************************************************
456	* Global Variables *
457	*********************************************************************************************************************************/
458	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
459	/**
460	* Array of all VMCS fields.
461	* Any fields added to the VT-x spec. should be added here.
462	*
463	* Currently only used to derive shadow VMCS fields for hardware-assisted execution
464	* of nested-guests.
465	*/
466	static const uint32_t g_aVmcsFields[] =
467	{
468	/* 16-bit control fields. */
469	VMX_VMCS16_VPID,
470	VMX_VMCS16_POSTED_INT_NOTIFY_VECTOR,
471	VMX_VMCS16_EPTP_INDEX,
472
473	/* 16-bit guest-state fields. */
474	VMX_VMCS16_GUEST_ES_SEL,
475	VMX_VMCS16_GUEST_CS_SEL,
476	VMX_VMCS16_GUEST_SS_SEL,
477	VMX_VMCS16_GUEST_DS_SEL,
478	VMX_VMCS16_GUEST_FS_SEL,
479	VMX_VMCS16_GUEST_GS_SEL,
480	VMX_VMCS16_GUEST_LDTR_SEL,
481	VMX_VMCS16_GUEST_TR_SEL,
482	VMX_VMCS16_GUEST_INTR_STATUS,
483	VMX_VMCS16_GUEST_PML_INDEX,
484
485	/* 16-bits host-state fields. */
486	VMX_VMCS16_HOST_ES_SEL,
487	VMX_VMCS16_HOST_CS_SEL,
488	VMX_VMCS16_HOST_SS_SEL,
489	VMX_VMCS16_HOST_DS_SEL,
490	VMX_VMCS16_HOST_FS_SEL,
491	VMX_VMCS16_HOST_GS_SEL,
492	VMX_VMCS16_HOST_TR_SEL,
493
494	/* 64-bit control fields. */
495	VMX_VMCS64_CTRL_IO_BITMAP_A_FULL,
496	VMX_VMCS64_CTRL_IO_BITMAP_A_HIGH,
497	VMX_VMCS64_CTRL_IO_BITMAP_B_FULL,
498	VMX_VMCS64_CTRL_IO_BITMAP_B_HIGH,
499	VMX_VMCS64_CTRL_MSR_BITMAP_FULL,
500	VMX_VMCS64_CTRL_MSR_BITMAP_HIGH,
501	VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL,
502	VMX_VMCS64_CTRL_EXIT_MSR_STORE_HIGH,
503	VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL,
504	VMX_VMCS64_CTRL_EXIT_MSR_LOAD_HIGH,
505	VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL,
506	VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_HIGH,
507	VMX_VMCS64_CTRL_EXEC_VMCS_PTR_FULL,
508	VMX_VMCS64_CTRL_EXEC_VMCS_PTR_HIGH,
509	VMX_VMCS64_CTRL_EXEC_PML_ADDR_FULL,
510	VMX_VMCS64_CTRL_EXEC_PML_ADDR_HIGH,
511	VMX_VMCS64_CTRL_TSC_OFFSET_FULL,
512	VMX_VMCS64_CTRL_TSC_OFFSET_HIGH,
513	VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL,
514	VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_HIGH,
515	VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL,
516	VMX_VMCS64_CTRL_APIC_ACCESSADDR_HIGH,
517	VMX_VMCS64_CTRL_POSTED_INTR_DESC_FULL,
518	VMX_VMCS64_CTRL_POSTED_INTR_DESC_HIGH,
519	VMX_VMCS64_CTRL_VMFUNC_CTRLS_FULL,
520	VMX_VMCS64_CTRL_VMFUNC_CTRLS_HIGH,
521	VMX_VMCS64_CTRL_EPTP_FULL,
522	VMX_VMCS64_CTRL_EPTP_HIGH,
523	VMX_VMCS64_CTRL_EOI_BITMAP_0_FULL,
524	VMX_VMCS64_CTRL_EOI_BITMAP_0_HIGH,
525	VMX_VMCS64_CTRL_EOI_BITMAP_1_FULL,
526	VMX_VMCS64_CTRL_EOI_BITMAP_1_HIGH,
527	VMX_VMCS64_CTRL_EOI_BITMAP_2_FULL,
528	VMX_VMCS64_CTRL_EOI_BITMAP_2_HIGH,
529	VMX_VMCS64_CTRL_EOI_BITMAP_3_FULL,
530	VMX_VMCS64_CTRL_EOI_BITMAP_3_HIGH,
531	VMX_VMCS64_CTRL_EPTP_LIST_FULL,
532	VMX_VMCS64_CTRL_EPTP_LIST_HIGH,
533	VMX_VMCS64_CTRL_VMREAD_BITMAP_FULL,
534	VMX_VMCS64_CTRL_VMREAD_BITMAP_HIGH,
535	VMX_VMCS64_CTRL_VMWRITE_BITMAP_FULL,
536	VMX_VMCS64_CTRL_VMWRITE_BITMAP_HIGH,
537	VMX_VMCS64_CTRL_VE_XCPT_INFO_ADDR_FULL,
538	VMX_VMCS64_CTRL_VE_XCPT_INFO_ADDR_HIGH,
539	VMX_VMCS64_CTRL_XSS_EXITING_BITMAP_FULL,
540	VMX_VMCS64_CTRL_XSS_EXITING_BITMAP_HIGH,
541	VMX_VMCS64_CTRL_ENCLS_EXITING_BITMAP_FULL,
542	VMX_VMCS64_CTRL_ENCLS_EXITING_BITMAP_HIGH,
543	VMX_VMCS64_CTRL_SPPTP_FULL,
544	VMX_VMCS64_CTRL_SPPTP_HIGH,
545	VMX_VMCS64_CTRL_TSC_MULTIPLIER_FULL,
546	VMX_VMCS64_CTRL_TSC_MULTIPLIER_HIGH,
547	VMX_VMCS64_CTRL_PROC_EXEC3_FULL,
548	VMX_VMCS64_CTRL_PROC_EXEC3_HIGH,
549	VMX_VMCS64_CTRL_ENCLV_EXITING_BITMAP_FULL,
550	VMX_VMCS64_CTRL_ENCLV_EXITING_BITMAP_HIGH,
551
552	/* 64-bit read-only data fields. */
553	VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL,
554	VMX_VMCS64_RO_GUEST_PHYS_ADDR_HIGH,
555
556	/* 64-bit guest-state fields. */
557	VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL,
558	VMX_VMCS64_GUEST_VMCS_LINK_PTR_HIGH,
559	VMX_VMCS64_GUEST_DEBUGCTL_FULL,
560	VMX_VMCS64_GUEST_DEBUGCTL_HIGH,
561	VMX_VMCS64_GUEST_PAT_FULL,
562	VMX_VMCS64_GUEST_PAT_HIGH,
563	VMX_VMCS64_GUEST_EFER_FULL,
564	VMX_VMCS64_GUEST_EFER_HIGH,
565	VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL,
566	VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_HIGH,
567	VMX_VMCS64_GUEST_PDPTE0_FULL,
568	VMX_VMCS64_GUEST_PDPTE0_HIGH,
569	VMX_VMCS64_GUEST_PDPTE1_FULL,
570	VMX_VMCS64_GUEST_PDPTE1_HIGH,
571	VMX_VMCS64_GUEST_PDPTE2_FULL,
572	VMX_VMCS64_GUEST_PDPTE2_HIGH,
573	VMX_VMCS64_GUEST_PDPTE3_FULL,
574	VMX_VMCS64_GUEST_PDPTE3_HIGH,
575	VMX_VMCS64_GUEST_BNDCFGS_FULL,
576	VMX_VMCS64_GUEST_BNDCFGS_HIGH,
577	VMX_VMCS64_GUEST_RTIT_CTL_FULL,
578	VMX_VMCS64_GUEST_RTIT_CTL_HIGH,
579	VMX_VMCS64_GUEST_PKRS_FULL,
580	VMX_VMCS64_GUEST_PKRS_HIGH,
581
582	/* 64-bit host-state fields. */
583	VMX_VMCS64_HOST_PAT_FULL,
584	VMX_VMCS64_HOST_PAT_HIGH,
585	VMX_VMCS64_HOST_EFER_FULL,
586	VMX_VMCS64_HOST_EFER_HIGH,
587	VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_FULL,
588	VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_HIGH,
589	VMX_VMCS64_HOST_PKRS_FULL,
590	VMX_VMCS64_HOST_PKRS_HIGH,
591
592	/* 32-bit control fields. */
593	VMX_VMCS32_CTRL_PIN_EXEC,
594	VMX_VMCS32_CTRL_PROC_EXEC,
595	VMX_VMCS32_CTRL_EXCEPTION_BITMAP,
596	VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK,
597	VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH,
598	VMX_VMCS32_CTRL_CR3_TARGET_COUNT,
599	VMX_VMCS32_CTRL_EXIT,
600	VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT,
601	VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT,
602	VMX_VMCS32_CTRL_ENTRY,
603	VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT,
604	VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO,
605	VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE,
606	VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH,
607	VMX_VMCS32_CTRL_TPR_THRESHOLD,
608	VMX_VMCS32_CTRL_PROC_EXEC2,
609	VMX_VMCS32_CTRL_PLE_GAP,
610	VMX_VMCS32_CTRL_PLE_WINDOW,
611
612	/* 32-bits read-only fields. */
613	VMX_VMCS32_RO_VM_INSTR_ERROR,
614	VMX_VMCS32_RO_EXIT_REASON,
615	VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO,
616	VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE,
617	VMX_VMCS32_RO_IDT_VECTORING_INFO,
618	VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE,
619	VMX_VMCS32_RO_EXIT_INSTR_LENGTH,
620	VMX_VMCS32_RO_EXIT_INSTR_INFO,
621
622	/* 32-bit guest-state fields. */
623	VMX_VMCS32_GUEST_ES_LIMIT,
624	VMX_VMCS32_GUEST_CS_LIMIT,
625	VMX_VMCS32_GUEST_SS_LIMIT,
626	VMX_VMCS32_GUEST_DS_LIMIT,
627	VMX_VMCS32_GUEST_FS_LIMIT,
628	VMX_VMCS32_GUEST_GS_LIMIT,
629	VMX_VMCS32_GUEST_LDTR_LIMIT,
630	VMX_VMCS32_GUEST_TR_LIMIT,
631	VMX_VMCS32_GUEST_GDTR_LIMIT,
632	VMX_VMCS32_GUEST_IDTR_LIMIT,
633	VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS,
634	VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS,
635	VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS,
636	VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS,
637	VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS,
638	VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS,
639	VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS,
640	VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS,
641	VMX_VMCS32_GUEST_INT_STATE,
642	VMX_VMCS32_GUEST_ACTIVITY_STATE,
643	VMX_VMCS32_GUEST_SMBASE,
644	VMX_VMCS32_GUEST_SYSENTER_CS,
645	VMX_VMCS32_PREEMPT_TIMER_VALUE,
646
647	/* 32-bit host-state fields. */
648	VMX_VMCS32_HOST_SYSENTER_CS,
649
650	/* Natural-width control fields. */
651	VMX_VMCS_CTRL_CR0_MASK,
652	VMX_VMCS_CTRL_CR4_MASK,
653	VMX_VMCS_CTRL_CR0_READ_SHADOW,
654	VMX_VMCS_CTRL_CR4_READ_SHADOW,
655	VMX_VMCS_CTRL_CR3_TARGET_VAL0,
656	VMX_VMCS_CTRL_CR3_TARGET_VAL1,
657	VMX_VMCS_CTRL_CR3_TARGET_VAL2,
658	VMX_VMCS_CTRL_CR3_TARGET_VAL3,
659
660	/* Natural-width read-only data fields. */
661	VMX_VMCS_RO_EXIT_QUALIFICATION,
662	VMX_VMCS_RO_IO_RCX,
663	VMX_VMCS_RO_IO_RSI,
664	VMX_VMCS_RO_IO_RDI,
665	VMX_VMCS_RO_IO_RIP,
666	VMX_VMCS_RO_GUEST_LINEAR_ADDR,
667
668	/* Natural-width guest-state field */
669	VMX_VMCS_GUEST_CR0,
670	VMX_VMCS_GUEST_CR3,
671	VMX_VMCS_GUEST_CR4,
672	VMX_VMCS_GUEST_ES_BASE,
673	VMX_VMCS_GUEST_CS_BASE,
674	VMX_VMCS_GUEST_SS_BASE,
675	VMX_VMCS_GUEST_DS_BASE,
676	VMX_VMCS_GUEST_FS_BASE,
677	VMX_VMCS_GUEST_GS_BASE,
678	VMX_VMCS_GUEST_LDTR_BASE,
679	VMX_VMCS_GUEST_TR_BASE,
680	VMX_VMCS_GUEST_GDTR_BASE,
681	VMX_VMCS_GUEST_IDTR_BASE,
682	VMX_VMCS_GUEST_DR7,
683	VMX_VMCS_GUEST_RSP,
684	VMX_VMCS_GUEST_RIP,
685	VMX_VMCS_GUEST_RFLAGS,
686	VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS,
687	VMX_VMCS_GUEST_SYSENTER_ESP,
688	VMX_VMCS_GUEST_SYSENTER_EIP,
689	VMX_VMCS_GUEST_S_CET,
690	VMX_VMCS_GUEST_SSP,
691	VMX_VMCS_GUEST_INTR_SSP_TABLE_ADDR,
692
693	/* Natural-width host-state fields */
694	VMX_VMCS_HOST_CR0,
695	VMX_VMCS_HOST_CR3,
696	VMX_VMCS_HOST_CR4,
697	VMX_VMCS_HOST_FS_BASE,
698	VMX_VMCS_HOST_GS_BASE,
699	VMX_VMCS_HOST_TR_BASE,
700	VMX_VMCS_HOST_GDTR_BASE,
701	VMX_VMCS_HOST_IDTR_BASE,
702	VMX_VMCS_HOST_SYSENTER_ESP,
703	VMX_VMCS_HOST_SYSENTER_EIP,
704	VMX_VMCS_HOST_RSP,
705	VMX_VMCS_HOST_RIP,
706	VMX_VMCS_HOST_S_CET,
707	VMX_VMCS_HOST_SSP,
708	VMX_VMCS_HOST_INTR_SSP_TABLE_ADDR
709	};
710	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
711
712	#ifdef VBOX_STRICT
713	static const uint32_t g_aVmcsSegBase[] =
714	{
715	VMX_VMCS_GUEST_ES_BASE,
716	VMX_VMCS_GUEST_CS_BASE,
717	VMX_VMCS_GUEST_SS_BASE,
718	VMX_VMCS_GUEST_DS_BASE,
719	VMX_VMCS_GUEST_FS_BASE,
720	VMX_VMCS_GUEST_GS_BASE
721	};
722	static const uint32_t g_aVmcsSegSel[] =
723	{
724	VMX_VMCS16_GUEST_ES_SEL,
725	VMX_VMCS16_GUEST_CS_SEL,
726	VMX_VMCS16_GUEST_SS_SEL,
727	VMX_VMCS16_GUEST_DS_SEL,
728	VMX_VMCS16_GUEST_FS_SEL,
729	VMX_VMCS16_GUEST_GS_SEL
730	};
731	static const uint32_t g_aVmcsSegLimit[] =
732	{
733	VMX_VMCS32_GUEST_ES_LIMIT,
734	VMX_VMCS32_GUEST_CS_LIMIT,
735	VMX_VMCS32_GUEST_SS_LIMIT,
736	VMX_VMCS32_GUEST_DS_LIMIT,
737	VMX_VMCS32_GUEST_FS_LIMIT,
738	VMX_VMCS32_GUEST_GS_LIMIT
739	};
740	static const uint32_t g_aVmcsSegAttr[] =
741	{
742	VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS,
743	VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS,
744	VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS,
745	VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS,
746	VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS,
747	VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS
748	};
749	AssertCompile(RT_ELEMENTS(g_aVmcsSegSel) == X86_SREG_COUNT);
750	AssertCompile(RT_ELEMENTS(g_aVmcsSegLimit) == X86_SREG_COUNT);
751	AssertCompile(RT_ELEMENTS(g_aVmcsSegBase) == X86_SREG_COUNT);
752	AssertCompile(RT_ELEMENTS(g_aVmcsSegAttr) == X86_SREG_COUNT);
753	#endif /* VBOX_STRICT */
754
755	#ifdef HMVMX_USE_FUNCTION_TABLE
756	/**
757	* VMX_EXIT dispatch table.
758	*/
759	static const struct CLANG11NOTHROWWEIRDNESS { PFNVMXEXITHANDLER pfn; } g_aVMExitHandlers[VMX_EXIT_MAX + 1] =
760	{
761	/* 0 VMX_EXIT_XCPT_OR_NMI */ { hmR0VmxExitXcptOrNmi },
762	/* 1 VMX_EXIT_EXT_INT */ { hmR0VmxExitExtInt },
763	/* 2 VMX_EXIT_TRIPLE_FAULT */ { hmR0VmxExitTripleFault },
764	/* 3 VMX_EXIT_INIT_SIGNAL */ { hmR0VmxExitErrUnexpected },
765	/* 4 VMX_EXIT_SIPI */ { hmR0VmxExitErrUnexpected },
766	/* 5 VMX_EXIT_IO_SMI */ { hmR0VmxExitErrUnexpected },
767	/* 6 VMX_EXIT_SMI */ { hmR0VmxExitErrUnexpected },
768	/* 7 VMX_EXIT_INT_WINDOW */ { hmR0VmxExitIntWindow },
769	/* 8 VMX_EXIT_NMI_WINDOW */ { hmR0VmxExitNmiWindow },
770	/* 9 VMX_EXIT_TASK_SWITCH */ { hmR0VmxExitTaskSwitch },
771	/* 10 VMX_EXIT_CPUID */ { hmR0VmxExitCpuid },
772	/* 11 VMX_EXIT_GETSEC */ { hmR0VmxExitGetsec },
773	/* 12 VMX_EXIT_HLT */ { hmR0VmxExitHlt },
774	/* 13 VMX_EXIT_INVD */ { hmR0VmxExitInvd },
775	/* 14 VMX_EXIT_INVLPG */ { hmR0VmxExitInvlpg },
776	/* 15 VMX_EXIT_RDPMC */ { hmR0VmxExitRdpmc },
777	/* 16 VMX_EXIT_RDTSC */ { hmR0VmxExitRdtsc },
778	/* 17 VMX_EXIT_RSM */ { hmR0VmxExitErrUnexpected },
779	/* 18 VMX_EXIT_VMCALL */ { hmR0VmxExitVmcall },
780	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
781	/* 19 VMX_EXIT_VMCLEAR */ { hmR0VmxExitVmclear },
782	/* 20 VMX_EXIT_VMLAUNCH */ { hmR0VmxExitVmlaunch },
783	/* 21 VMX_EXIT_VMPTRLD */ { hmR0VmxExitVmptrld },
784	/* 22 VMX_EXIT_VMPTRST */ { hmR0VmxExitVmptrst },
785	/* 23 VMX_EXIT_VMREAD */ { hmR0VmxExitVmread },
786	/* 24 VMX_EXIT_VMRESUME */ { hmR0VmxExitVmresume },
787	/* 25 VMX_EXIT_VMWRITE */ { hmR0VmxExitVmwrite },
788	/* 26 VMX_EXIT_VMXOFF */ { hmR0VmxExitVmxoff },
789	/* 27 VMX_EXIT_VMXON */ { hmR0VmxExitVmxon },
790	#else
791	/* 19 VMX_EXIT_VMCLEAR */ { hmR0VmxExitSetPendingXcptUD },
792	/* 20 VMX_EXIT_VMLAUNCH */ { hmR0VmxExitSetPendingXcptUD },
793	/* 21 VMX_EXIT_VMPTRLD */ { hmR0VmxExitSetPendingXcptUD },
794	/* 22 VMX_EXIT_VMPTRST */ { hmR0VmxExitSetPendingXcptUD },
795	/* 23 VMX_EXIT_VMREAD */ { hmR0VmxExitSetPendingXcptUD },
796	/* 24 VMX_EXIT_VMRESUME */ { hmR0VmxExitSetPendingXcptUD },
797	/* 25 VMX_EXIT_VMWRITE */ { hmR0VmxExitSetPendingXcptUD },
798	/* 26 VMX_EXIT_VMXOFF */ { hmR0VmxExitSetPendingXcptUD },
799	/* 27 VMX_EXIT_VMXON */ { hmR0VmxExitSetPendingXcptUD },
800	#endif
801	/* 28 VMX_EXIT_MOV_CRX */ { hmR0VmxExitMovCRx },
802	/* 29 VMX_EXIT_MOV_DRX */ { hmR0VmxExitMovDRx },
803	/* 30 VMX_EXIT_IO_INSTR */ { hmR0VmxExitIoInstr },
804	/* 31 VMX_EXIT_RDMSR */ { hmR0VmxExitRdmsr },
805	/* 32 VMX_EXIT_WRMSR */ { hmR0VmxExitWrmsr },
806	/* 33 VMX_EXIT_ERR_INVALID_GUEST_STATE */ { hmR0VmxExitErrInvalidGuestState },
807	/* 34 VMX_EXIT_ERR_MSR_LOAD */ { hmR0VmxExitErrUnexpected },
808	/* 35 UNDEFINED */ { hmR0VmxExitErrUnexpected },
809	/* 36 VMX_EXIT_MWAIT */ { hmR0VmxExitMwait },
810	/* 37 VMX_EXIT_MTF */ { hmR0VmxExitMtf },
811	/* 38 UNDEFINED */ { hmR0VmxExitErrUnexpected },
812	/* 39 VMX_EXIT_MONITOR */ { hmR0VmxExitMonitor },
813	/* 40 VMX_EXIT_PAUSE */ { hmR0VmxExitPause },
814	/* 41 VMX_EXIT_ERR_MACHINE_CHECK */ { hmR0VmxExitErrUnexpected },
815	/* 42 UNDEFINED */ { hmR0VmxExitErrUnexpected },
816	/* 43 VMX_EXIT_TPR_BELOW_THRESHOLD */ { hmR0VmxExitTprBelowThreshold },
817	/* 44 VMX_EXIT_APIC_ACCESS */ { hmR0VmxExitApicAccess },
818	/* 45 VMX_EXIT_VIRTUALIZED_EOI */ { hmR0VmxExitErrUnexpected },
819	/* 46 VMX_EXIT_GDTR_IDTR_ACCESS */ { hmR0VmxExitErrUnexpected },
820	/* 47 VMX_EXIT_LDTR_TR_ACCESS */ { hmR0VmxExitErrUnexpected },
821	/* 48 VMX_EXIT_EPT_VIOLATION */ { hmR0VmxExitEptViolation },
822	/* 49 VMX_EXIT_EPT_MISCONFIG */ { hmR0VmxExitEptMisconfig },
823	/* 50 VMX_EXIT_INVEPT */ { hmR0VmxExitSetPendingXcptUD },
824	/* 51 VMX_EXIT_RDTSCP */ { hmR0VmxExitRdtscp },
825	/* 52 VMX_EXIT_PREEMPT_TIMER */ { hmR0VmxExitPreemptTimer },
826	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
827	/* 53 VMX_EXIT_INVVPID */ { hmR0VmxExitInvvpid },
828	#else
829	/* 53 VMX_EXIT_INVVPID */ { hmR0VmxExitSetPendingXcptUD },
830	#endif
831	/* 54 VMX_EXIT_WBINVD */ { hmR0VmxExitWbinvd },
832	/* 55 VMX_EXIT_XSETBV */ { hmR0VmxExitXsetbv },
833	/* 56 VMX_EXIT_APIC_WRITE */ { hmR0VmxExitErrUnexpected },
834	/* 57 VMX_EXIT_RDRAND */ { hmR0VmxExitErrUnexpected },
835	/* 58 VMX_EXIT_INVPCID */ { hmR0VmxExitInvpcid },
836	/* 59 VMX_EXIT_VMFUNC */ { hmR0VmxExitErrUnexpected },
837	/* 60 VMX_EXIT_ENCLS */ { hmR0VmxExitErrUnexpected },
838	/* 61 VMX_EXIT_RDSEED */ { hmR0VmxExitErrUnexpected },
839	/* 62 VMX_EXIT_PML_FULL */ { hmR0VmxExitErrUnexpected },
840	/* 63 VMX_EXIT_XSAVES */ { hmR0VmxExitErrUnexpected },
841	/* 64 VMX_EXIT_XRSTORS */ { hmR0VmxExitErrUnexpected },
842	/* 65 UNDEFINED */ { hmR0VmxExitErrUnexpected },
843	/* 66 VMX_EXIT_SPP_EVENT */ { hmR0VmxExitErrUnexpected },
844	/* 67 VMX_EXIT_UMWAIT */ { hmR0VmxExitErrUnexpected },
845	/* 68 VMX_EXIT_TPAUSE */ { hmR0VmxExitErrUnexpected },
846	/* 69 VMX_EXIT_LOADIWKEY */ { hmR0VmxExitErrUnexpected },
847	};
848	#endif /* HMVMX_USE_FUNCTION_TABLE */
849
850	#if defined(VBOX_STRICT) && defined(LOG_ENABLED)
851	static const char * const g_apszVmxInstrErrors[HMVMX_INSTR_ERROR_MAX + 1] =
852	{
853	/* 0 */ "(Not Used)",
854	/* 1 */ "VMCALL executed in VMX root operation.",
855	/* 2 */ "VMCLEAR with invalid physical address.",
856	/* 3 */ "VMCLEAR with VMXON pointer.",
857	/* 4 */ "VMLAUNCH with non-clear VMCS.",
858	/* 5 */ "VMRESUME with non-launched VMCS.",
859	/* 6 */ "VMRESUME after VMXOFF",
860	/* 7 */ "VM-entry with invalid control fields.",
861	/* 8 */ "VM-entry with invalid host state fields.",
862	/* 9 */ "VMPTRLD with invalid physical address.",
863	/* 10 */ "VMPTRLD with VMXON pointer.",
864	/* 11 */ "VMPTRLD with incorrect revision identifier.",
865	/* 12 */ "VMREAD/VMWRITE from/to unsupported VMCS component.",
866	/* 13 */ "VMWRITE to read-only VMCS component.",
867	/* 14 */ "(Not Used)",
868	/* 15 */ "VMXON executed in VMX root operation.",
869	/* 16 */ "VM-entry with invalid executive-VMCS pointer.",
870	/* 17 */ "VM-entry with non-launched executing VMCS.",
871	/* 18 */ "VM-entry with executive-VMCS pointer not VMXON pointer.",
872	/* 19 */ "VMCALL with non-clear VMCS.",
873	/* 20 */ "VMCALL with invalid VM-exit control fields.",
874	/* 21 */ "(Not Used)",
875	/* 22 */ "VMCALL with incorrect MSEG revision identifier.",
876	/* 23 */ "VMXOFF under dual monitor treatment of SMIs and SMM.",
877	/* 24 */ "VMCALL with invalid SMM-monitor features.",
878	/* 25 */ "VM-entry with invalid VM-execution control fields in executive VMCS.",
879	/* 26 */ "VM-entry with events blocked by MOV SS.",
880	/* 27 */ "(Not Used)",
881	/* 28 */ "Invalid operand to INVEPT/INVVPID."
882	};
883	#endif /* VBOX_STRICT && LOG_ENABLED */
884
885
886	/**
887	* Checks if the given MSR is part of the lastbranch-from-IP MSR stack.
888	* @returns @c true if it's part of LBR stack, @c false otherwise.
889	*
890	* @param pVM The cross context VM structure.
891	* @param idMsr The MSR.
892	* @param pidxMsr Where to store the index of the MSR in the LBR MSR array.
893	* Optional, can be NULL.
894	*
895	* @remarks Must only be called when LBR is enabled.
896	*/
897	DECL_FORCE_INLINE(bool) hmR0VmxIsLbrBranchFromMsr(PCVMCC pVM, uint32_t idMsr, uint32_t *pidxMsr)
898	{
899	Assert(pVM->hmr0.s.vmx.fLbr);
900	Assert(pVM->hmr0.s.vmx.idLbrFromIpMsrFirst);
901	uint32_t const cLbrStack = pVM->hmr0.s.vmx.idLbrFromIpMsrLast - pVM->hmr0.s.vmx.idLbrFromIpMsrFirst + 1;
902	uint32_t const idxMsr = idMsr - pVM->hmr0.s.vmx.idLbrFromIpMsrFirst;
903	if (idxMsr < cLbrStack)
904	{
905	if (pidxMsr)
906	*pidxMsr = idxMsr;
907	return true;
908	}
909	return false;
910	}
911
912
913	/**
914	* Checks if the given MSR is part of the lastbranch-to-IP MSR stack.
915	* @returns @c true if it's part of LBR stack, @c false otherwise.
916	*
917	* @param pVM The cross context VM structure.
918	* @param idMsr The MSR.
919	* @param pidxMsr Where to store the index of the MSR in the LBR MSR array.
920	* Optional, can be NULL.
921	*
922	* @remarks Must only be called when LBR is enabled and when lastbranch-to-IP MSRs
923	* are supported by the CPU (see hmR0VmxSetupLbrMsrRange).
924	*/
925	DECL_FORCE_INLINE(bool) hmR0VmxIsLbrBranchToMsr(PCVMCC pVM, uint32_t idMsr, uint32_t *pidxMsr)
926	{
927	Assert(pVM->hmr0.s.vmx.fLbr);
928	if (pVM->hmr0.s.vmx.idLbrToIpMsrFirst)
929	{
930	uint32_t const cLbrStack = pVM->hmr0.s.vmx.idLbrToIpMsrLast - pVM->hmr0.s.vmx.idLbrToIpMsrFirst + 1;
931	uint32_t const idxMsr = idMsr - pVM->hmr0.s.vmx.idLbrToIpMsrFirst;
932	if (idxMsr < cLbrStack)
933	{
934	if (pidxMsr)
935	*pidxMsr = idxMsr;
936	return true;
937	}
938	}
939	return false;
940	}
941
942
943	/**
944	* Gets the CR0 guest/host mask.
945	*
946	* These bits typically does not change through the lifetime of a VM. Any bit set in
947	* this mask is owned by the host/hypervisor and would cause a VM-exit when modified
948	* by the guest.
949	*
950	* @returns The CR0 guest/host mask.
951	* @param pVCpu The cross context virtual CPU structure.
952	*/
953	static uint64_t hmR0VmxGetFixedCr0Mask(PCVMCPUCC pVCpu)
954	{
955	/*
956	* Modifications to CR0 bits that VT-x ignores saving/restoring (CD, ET, NW) and
957	* to CR0 bits that we require for shadow paging (PG) by the guest must cause VM-exits.
958	*
959	* Furthermore, modifications to any bits that are reserved/unspecified currently
960	* by the Intel spec. must also cause a VM-exit. This prevents unpredictable behavior
961	* when future CPUs specify and use currently reserved/unspecified bits.
962	*/
963	/** @todo Avoid intercepting CR0.PE with unrestricted guest execution. Fix PGM
964	* enmGuestMode to be in-sync with the current mode. See @bugref{6398}
965	* and @bugref{6944}. */
966	PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
967	return ( X86_CR0_PE
968	\| X86_CR0_NE
969	\| (pVM->hmr0.s.fNestedPaging ? 0 : X86_CR0_WP)
970	\| X86_CR0_PG
971	\| VMX_EXIT_HOST_CR0_IGNORE_MASK);
972	}
973
974
975	/**
976	* Gets the CR4 guest/host mask.
977	*
978	* These bits typically does not change through the lifetime of a VM. Any bit set in
979	* this mask is owned by the host/hypervisor and would cause a VM-exit when modified
980	* by the guest.
981	*
982	* @returns The CR4 guest/host mask.
983	* @param pVCpu The cross context virtual CPU structure.
984	*/
985	static uint64_t hmR0VmxGetFixedCr4Mask(PCVMCPUCC pVCpu)
986	{
987	/*
988	* We construct a mask of all CR4 bits that the guest can modify without causing
989	* a VM-exit. Then invert this mask to obtain all CR4 bits that should cause
990	* a VM-exit when the guest attempts to modify them when executing using
991	* hardware-assisted VMX.
992	*
993	* When a feature is not exposed to the guest (and may be present on the host),
994	* we want to intercept guest modifications to the bit so we can emulate proper
995	* behavior (e.g., #GP).
996	*
997	* Furthermore, only modifications to those bits that don't require immediate
998	* emulation is allowed. For e.g., PCIDE is excluded because the behavior
999	* depends on CR3 which might not always be the guest value while executing
1000	* using hardware-assisted VMX.
1001	*/
1002	PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
1003	bool const fFsGsBase = pVM->cpum.ro.GuestFeatures.fFsGsBase;
1004	bool const fXSaveRstor = pVM->cpum.ro.GuestFeatures.fXSaveRstor;
1005	bool const fFxSaveRstor = pVM->cpum.ro.GuestFeatures.fFxSaveRstor;
1006
1007	/*
1008	* Paranoia.
1009	* Ensure features exposed to the guest are present on the host.
1010	*/
1011	Assert(!fFsGsBase \|\| pVM->cpum.ro.HostFeatures.fFsGsBase);
1012	Assert(!fXSaveRstor \|\| pVM->cpum.ro.HostFeatures.fXSaveRstor);
1013	Assert(!fFxSaveRstor \|\| pVM->cpum.ro.HostFeatures.fFxSaveRstor);
1014
1015	uint64_t const fGstMask = ( X86_CR4_PVI
1016	\| X86_CR4_TSD
1017	\| X86_CR4_DE
1018	\| X86_CR4_MCE
1019	\| X86_CR4_PCE
1020	\| X86_CR4_OSXMMEEXCPT
1021	\| (fFsGsBase ? X86_CR4_FSGSBASE : 0)
1022	\| (fXSaveRstor ? X86_CR4_OSXSAVE : 0)
1023	\| (fFxSaveRstor ? X86_CR4_OSFXSR : 0));
1024	return ~fGstMask;
1025	}
1026
1027
1028	/**
1029	* Gets the active (in use) VMCS info. object for the specified VCPU.
1030	*
1031	* This is either the guest or nested-guest VMCS info. and need not necessarily
1032	* pertain to the "current" VMCS (in the VMX definition of the term). For instance,
1033	* if the VM-entry failed due to an invalid-guest state, we may have "cleared" the
1034	* current VMCS while returning to ring-3. However, the VMCS info. object for that
1035	* VMCS would still be active and returned here so that we could dump the VMCS
1036	* fields to ring-3 for diagnostics. This function is thus only used to
1037	* distinguish between the nested-guest or guest VMCS.
1038	*
1039	* @returns The active VMCS information.
1040	* @param pVCpu The cross context virtual CPU structure.
1041	*
1042	* @thread EMT.
1043	* @remarks This function may be called with preemption or interrupts disabled!
1044	*/
1045	DECLINLINE(PVMXVMCSINFO) hmGetVmxActiveVmcsInfo(PVMCPUCC pVCpu)
1046	{
1047	if (!pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs)
1048	return &pVCpu->hmr0.s.vmx.VmcsInfo;
1049	return &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
1050	}
1051
1052
1053	/**
1054	* Returns whether the VM-exit MSR-store area differs from the VM-exit MSR-load
1055	* area.
1056	*
1057	* @returns @c true if it's different, @c false otherwise.
1058	* @param pVmcsInfo The VMCS info. object.
1059	*/
1060	DECL_FORCE_INLINE(bool) hmR0VmxIsSeparateExitMsrStoreAreaVmcs(PCVMXVMCSINFO pVmcsInfo)
1061	{
1062	return RT_BOOL( pVmcsInfo->pvGuestMsrStore != pVmcsInfo->pvGuestMsrLoad
1063	&& pVmcsInfo->pvGuestMsrStore);
1064	}
1065
1066
1067	/**
1068	* Sets the given Processor-based VM-execution controls.
1069	*
1070	* @param pVmxTransient The VMX-transient structure.
1071	* @param uProcCtls The Processor-based VM-execution controls to set.
1072	*/
1073	static void hmR0VmxSetProcCtlsVmcs(PVMXTRANSIENT pVmxTransient, uint32_t uProcCtls)
1074	{
1075	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1076	if ((pVmcsInfo->u32ProcCtls & uProcCtls) != uProcCtls)
1077	{
1078	pVmcsInfo->u32ProcCtls \|= uProcCtls;
1079	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
1080	AssertRC(rc);
1081	}
1082	}
1083
1084
1085	/**
1086	* Removes the given Processor-based VM-execution controls.
1087	*
1088	* @param pVCpu The cross context virtual CPU structure.
1089	* @param pVmxTransient The VMX-transient structure.
1090	* @param uProcCtls The Processor-based VM-execution controls to remove.
1091	*
1092	* @remarks When executing a nested-guest, this will not remove any of the specified
1093	* controls if the nested hypervisor has set any one of them.
1094	*/
1095	static void hmR0VmxRemoveProcCtlsVmcs(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uProcCtls)
1096	{
1097	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1098	if (pVmcsInfo->u32ProcCtls & uProcCtls)
1099	{
1100	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1101	if ( !pVmxTransient->fIsNestedGuest
1102	\|\| !CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, uProcCtls))
1103	#else
1104	NOREF(pVCpu);
1105	if (!pVmxTransient->fIsNestedGuest)
1106	#endif
1107	{
1108	pVmcsInfo->u32ProcCtls &= ~uProcCtls;
1109	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
1110	AssertRC(rc);
1111	}
1112	}
1113	}
1114
1115
1116	/**
1117	* Sets the TSC offset for the current VMCS.
1118	*
1119	* @param uTscOffset The TSC offset to set.
1120	* @param pVmcsInfo The VMCS info. object.
1121	*/
1122	static void hmR0VmxSetTscOffsetVmcs(PVMXVMCSINFO pVmcsInfo, uint64_t uTscOffset)
1123	{
1124	if (pVmcsInfo->u64TscOffset != uTscOffset)
1125	{
1126	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, uTscOffset);
1127	AssertRC(rc);
1128	pVmcsInfo->u64TscOffset = uTscOffset;
1129	}
1130	}
1131
1132
1133	/**
1134	* Adds one or more exceptions to the exception bitmap and commits it to the current
1135	* VMCS.
1136	*
1137	* @param pVmxTransient The VMX-transient structure.
1138	* @param uXcptMask The exception(s) to add.
1139	*/
1140	static void hmR0VmxAddXcptInterceptMask(PCVMXTRANSIENT pVmxTransient, uint32_t uXcptMask)
1141	{
1142	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1143	uint32_t uXcptBitmap = pVmcsInfo->u32XcptBitmap;
1144	if ((uXcptBitmap & uXcptMask) != uXcptMask)
1145	{
1146	uXcptBitmap \|= uXcptMask;
1147	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
1148	AssertRC(rc);
1149	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
1150	}
1151	}
1152
1153
1154	/**
1155	* Adds an exception to the exception bitmap and commits it to the current VMCS.
1156	*
1157	* @param pVmxTransient The VMX-transient structure.
1158	* @param uXcpt The exception to add.
1159	*/
1160	static void hmR0VmxAddXcptIntercept(PCVMXTRANSIENT pVmxTransient, uint8_t uXcpt)
1161	{
1162	Assert(uXcpt <= X86_XCPT_LAST);
1163	hmR0VmxAddXcptInterceptMask(pVmxTransient, RT_BIT_32(uXcpt));
1164	}
1165
1166
1167	/**
1168	* Remove one or more exceptions from the exception bitmap and commits it to the
1169	* current VMCS.
1170	*
1171	* This takes care of not removing the exception intercept if a nested-guest
1172	* requires the exception to be intercepted.
1173	*
1174	* @returns VBox status code.
1175	* @param pVCpu The cross context virtual CPU structure.
1176	* @param pVmxTransient The VMX-transient structure.
1177	* @param uXcptMask The exception(s) to remove.
1178	*/
1179	static int hmR0VmxRemoveXcptInterceptMask(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t uXcptMask)
1180	{
1181	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1182	uint32_t u32XcptBitmap = pVmcsInfo->u32XcptBitmap;
1183	if (u32XcptBitmap & uXcptMask)
1184	{
1185	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1186	if (!pVmxTransient->fIsNestedGuest)
1187	{ /* likely */ }
1188	else
1189	uXcptMask &= ~pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32XcptBitmap;
1190	#endif
1191	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
1192	uXcptMask &= ~( RT_BIT(X86_XCPT_BP)
1193	\| RT_BIT(X86_XCPT_DE)
1194	\| RT_BIT(X86_XCPT_NM)
1195	\| RT_BIT(X86_XCPT_TS)
1196	\| RT_BIT(X86_XCPT_UD)
1197	\| RT_BIT(X86_XCPT_NP)
1198	\| RT_BIT(X86_XCPT_SS)
1199	\| RT_BIT(X86_XCPT_GP)
1200	\| RT_BIT(X86_XCPT_PF)
1201	\| RT_BIT(X86_XCPT_MF));
1202	#elif defined(HMVMX_ALWAYS_TRAP_PF)
1203	uXcptMask &= ~RT_BIT(X86_XCPT_PF);
1204	#endif
1205	if (uXcptMask)
1206	{
1207	/* Validate we are not removing any essential exception intercepts. */
1208	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging \|\| !(uXcptMask & RT_BIT(X86_XCPT_PF)));
1209	NOREF(pVCpu);
1210	Assert(!(uXcptMask & RT_BIT(X86_XCPT_DB)));
1211	Assert(!(uXcptMask & RT_BIT(X86_XCPT_AC)));
1212
1213	/* Remove it from the exception bitmap. */
1214	u32XcptBitmap &= ~uXcptMask;
1215
1216	/* Commit and update the cache if necessary. */
1217	if (pVmcsInfo->u32XcptBitmap != u32XcptBitmap)
1218	{
1219	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
1220	AssertRC(rc);
1221	pVmcsInfo->u32XcptBitmap = u32XcptBitmap;
1222	}
1223	}
1224	}
1225	return VINF_SUCCESS;
1226	}
1227
1228
1229	/**
1230	* Remove an exceptions from the exception bitmap and commits it to the current
1231	* VMCS.
1232	*
1233	* @returns VBox status code.
1234	* @param pVCpu The cross context virtual CPU structure.
1235	* @param pVmxTransient The VMX-transient structure.
1236	* @param uXcpt The exception to remove.
1237	*/
1238	static int hmR0VmxRemoveXcptIntercept(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint8_t uXcpt)
1239	{
1240	return hmR0VmxRemoveXcptInterceptMask(pVCpu, pVmxTransient, RT_BIT(uXcpt));
1241	}
1242
1243
1244	/**
1245	* Loads the VMCS specified by the VMCS info. object.
1246	*
1247	* @returns VBox status code.
1248	* @param pVmcsInfo The VMCS info. object.
1249	*
1250	* @remarks Can be called with interrupts disabled.
1251	*/
1252	static int hmR0VmxLoadVmcs(PVMXVMCSINFO pVmcsInfo)
1253	{
1254	Assert(pVmcsInfo->HCPhysVmcs != 0 && pVmcsInfo->HCPhysVmcs != NIL_RTHCPHYS);
1255	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1256
1257	int rc = VMXLoadVmcs(pVmcsInfo->HCPhysVmcs);
1258	if (RT_SUCCESS(rc))
1259	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_CURRENT;
1260	return rc;
1261	}
1262
1263
1264	/**
1265	* Clears the VMCS specified by the VMCS info. object.
1266	*
1267	* @returns VBox status code.
1268	* @param pVmcsInfo The VMCS info. object.
1269	*
1270	* @remarks Can be called with interrupts disabled.
1271	*/
1272	static int hmR0VmxClearVmcs(PVMXVMCSINFO pVmcsInfo)
1273	{
1274	Assert(pVmcsInfo->HCPhysVmcs != 0 && pVmcsInfo->HCPhysVmcs != NIL_RTHCPHYS);
1275	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1276
1277	int rc = VMXClearVmcs(pVmcsInfo->HCPhysVmcs);
1278	if (RT_SUCCESS(rc))
1279	pVmcsInfo->fVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
1280	return rc;
1281	}
1282
1283
1284	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1285	/**
1286	* Loads the shadow VMCS specified by the VMCS info. object.
1287	*
1288	* @returns VBox status code.
1289	* @param pVmcsInfo The VMCS info. object.
1290	*
1291	* @remarks Can be called with interrupts disabled.
1292	*/
1293	static int hmR0VmxLoadShadowVmcs(PVMXVMCSINFO pVmcsInfo)
1294	{
1295	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1296	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
1297
1298	int rc = VMXLoadVmcs(pVmcsInfo->HCPhysShadowVmcs);
1299	if (RT_SUCCESS(rc))
1300	pVmcsInfo->fShadowVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_CURRENT;
1301	return rc;
1302	}
1303
1304
1305	/**
1306	* Clears the shadow VMCS specified by the VMCS info. object.
1307	*
1308	* @returns VBox status code.
1309	* @param pVmcsInfo The VMCS info. object.
1310	*
1311	* @remarks Can be called with interrupts disabled.
1312	*/
1313	static int hmR0VmxClearShadowVmcs(PVMXVMCSINFO pVmcsInfo)
1314	{
1315	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1316	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
1317
1318	int rc = VMXClearVmcs(pVmcsInfo->HCPhysShadowVmcs);
1319	if (RT_SUCCESS(rc))
1320	pVmcsInfo->fShadowVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
1321	return rc;
1322	}
1323
1324
1325	/**
1326	* Switches from and to the specified VMCSes.
1327	*
1328	* @returns VBox status code.
1329	* @param pVmcsInfoFrom The VMCS info. object we are switching from.
1330	* @param pVmcsInfoTo The VMCS info. object we are switching to.
1331	*
1332	* @remarks Called with interrupts disabled.
1333	*/
1334	static int hmR0VmxSwitchVmcs(PVMXVMCSINFO pVmcsInfoFrom, PVMXVMCSINFO pVmcsInfoTo)
1335	{
1336	/*
1337	* Clear the VMCS we are switching out if it has not already been cleared.
1338	* This will sync any CPU internal data back to the VMCS.
1339	*/
1340	if (pVmcsInfoFrom->fVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
1341	{
1342	int rc = hmR0VmxClearVmcs(pVmcsInfoFrom);
1343	if (RT_SUCCESS(rc))
1344	{
1345	/*
1346	* The shadow VMCS, if any, would not be active at this point since we
1347	* would have cleared it while importing the virtual hardware-virtualization
1348	* state as part the VMLAUNCH/VMRESUME VM-exit. Hence, there's no need to
1349	* clear the shadow VMCS here, just assert for safety.
1350	*/
1351	Assert(!pVmcsInfoFrom->pvShadowVmcs \|\| pVmcsInfoFrom->fShadowVmcsState == VMX_V_VMCS_LAUNCH_STATE_CLEAR);
1352	}
1353	else
1354	return rc;
1355	}
1356
1357	/*
1358	* Clear the VMCS we are switching to if it has not already been cleared.
1359	* This will initialize the VMCS launch state to "clear" required for loading it.
1360	*
1361	* See Intel spec. 31.6 "Preparation And Launching A Virtual Machine".
1362	*/
1363	if (pVmcsInfoTo->fVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
1364	{
1365	int rc = hmR0VmxClearVmcs(pVmcsInfoTo);
1366	if (RT_SUCCESS(rc))
1367	{ /* likely */ }
1368	else
1369	return rc;
1370	}
1371
1372	/*
1373	* Finally, load the VMCS we are switching to.
1374	*/
1375	return hmR0VmxLoadVmcs(pVmcsInfoTo);
1376	}
1377
1378
1379	/**
1380	* Switches between the guest VMCS and the nested-guest VMCS as specified by the
1381	* caller.
1382	*
1383	* @returns VBox status code.
1384	* @param pVCpu The cross context virtual CPU structure.
1385	* @param fSwitchToNstGstVmcs Whether to switch to the nested-guest VMCS (pass
1386	* true) or guest VMCS (pass false).
1387	*/
1388	static int hmR0VmxSwitchToGstOrNstGstVmcs(PVMCPUCC pVCpu, bool fSwitchToNstGstVmcs)
1389	{
1390	/* Ensure we have synced everything from the guest-CPU context to the VMCS before switching. */
1391	HMVMX_CPUMCTX_ASSERT(pVCpu, HMVMX_CPUMCTX_EXTRN_ALL);
1392
1393	PVMXVMCSINFO pVmcsInfoFrom;
1394	PVMXVMCSINFO pVmcsInfoTo;
1395	if (fSwitchToNstGstVmcs)
1396	{
1397	pVmcsInfoFrom = &pVCpu->hmr0.s.vmx.VmcsInfo;
1398	pVmcsInfoTo = &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
1399	}
1400	else
1401	{
1402	pVmcsInfoFrom = &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
1403	pVmcsInfoTo = &pVCpu->hmr0.s.vmx.VmcsInfo;
1404	}
1405
1406	/*
1407	* Disable interrupts to prevent being preempted while we switch the current VMCS as the
1408	* preemption hook code path acquires the current VMCS.
1409	*/
1410	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1411
1412	int rc = hmR0VmxSwitchVmcs(pVmcsInfoFrom, pVmcsInfoTo);
1413	if (RT_SUCCESS(rc))
1414	{
1415	pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs = fSwitchToNstGstVmcs;
1416	pVCpu->hm.s.vmx.fSwitchedToNstGstVmcsCopyForRing3 = fSwitchToNstGstVmcs;
1417
1418	/*
1419	* If we are switching to a VMCS that was executed on a different host CPU or was
1420	* never executed before, flag that we need to export the host state before executing
1421	* guest/nested-guest code using hardware-assisted VMX.
1422	*
1423	* This could probably be done in a preemptible context since the preemption hook
1424	* will flag the necessary change in host context. However, since preemption is
1425	* already disabled and to avoid making assumptions about host specific code in
1426	* RTMpCpuId when called with preemption enabled, we'll do this while preemption is
1427	* disabled.
1428	*/
1429	if (pVmcsInfoTo->idHostCpuState == RTMpCpuId())
1430	{ /* likely */ }
1431	else
1432	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE);
1433
1434	ASMSetFlags(fEFlags);
1435
1436	/*
1437	* We use a different VM-exit MSR-store areas for the guest and nested-guest. Hence,
1438	* flag that we need to update the host MSR values there. Even if we decide in the
1439	* future to share the VM-exit MSR-store area page between the guest and nested-guest,
1440	* if its content differs, we would have to update the host MSRs anyway.
1441	*/
1442	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = false;
1443	}
1444	else
1445	ASMSetFlags(fEFlags);
1446	return rc;
1447	}
1448	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
1449
1450
1451	/**
1452	* Updates the VM's last error record.
1453	*
1454	* If there was a VMX instruction error, reads the error data from the VMCS and
1455	* updates VCPU's last error record as well.
1456	*
1457	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
1458	* Can be NULL if @a rc is not VERR_VMX_UNABLE_TO_START_VM or
1459	* VERR_VMX_INVALID_VMCS_FIELD.
1460	* @param rc The error code.
1461	*/
1462	static void hmR0VmxUpdateErrorRecord(PVMCPUCC pVCpu, int rc)
1463	{
1464	if ( rc == VERR_VMX_INVALID_VMCS_FIELD
1465	\|\| rc == VERR_VMX_UNABLE_TO_START_VM)
1466	{
1467	AssertPtrReturnVoid(pVCpu);
1468	VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
1469	}
1470	pVCpu->CTX_SUFF(pVM)->hm.s.ForR3.rcInit = rc;
1471	}
1472
1473
1474	#ifdef VBOX_STRICT
1475	/**
1476	* Reads the VM-entry interruption-information field from the VMCS into the VMX
1477	* transient structure.
1478	*
1479	* @param pVmxTransient The VMX-transient structure.
1480	*/
1481	DECLINLINE(void) hmR0VmxReadEntryIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
1482	{
1483	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &pVmxTransient->uEntryIntInfo);
1484	AssertRC(rc);
1485	}
1486
1487
1488	/**
1489	* Reads the VM-entry exception error code field from the VMCS into
1490	* the VMX transient structure.
1491	*
1492	* @param pVmxTransient The VMX-transient structure.
1493	*/
1494	DECLINLINE(void) hmR0VmxReadEntryXcptErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1495	{
1496	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &pVmxTransient->uEntryXcptErrorCode);
1497	AssertRC(rc);
1498	}
1499
1500
1501	/**
1502	* Reads the VM-entry exception error code field from the VMCS into
1503	* the VMX transient structure.
1504	*
1505	* @param pVmxTransient The VMX-transient structure.
1506	*/
1507	DECLINLINE(void) hmR0VmxReadEntryInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
1508	{
1509	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &pVmxTransient->cbEntryInstr);
1510	AssertRC(rc);
1511	}
1512	#endif /* VBOX_STRICT */
1513
1514
1515	/**
1516	* Reads the VM-exit interruption-information field from the VMCS into the VMX
1517	* transient structure.
1518	*
1519	* @param pVmxTransient The VMX-transient structure.
1520	*/
1521	DECLINLINE(void) hmR0VmxReadExitIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
1522	{
1523	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INTERRUPTION_INFO))
1524	{
1525	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
1526	AssertRC(rc);
1527	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INTERRUPTION_INFO;
1528	}
1529	}
1530
1531
1532	/**
1533	* Reads the VM-exit interruption error code from the VMCS into the VMX
1534	* transient structure.
1535	*
1536	* @param pVmxTransient The VMX-transient structure.
1537	*/
1538	DECLINLINE(void) hmR0VmxReadExitIntErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1539	{
1540	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE))
1541	{
1542	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
1543	AssertRC(rc);
1544	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE;
1545	}
1546	}
1547
1548
1549	/**
1550	* Reads the VM-exit instruction length field from the VMCS into the VMX
1551	* transient structure.
1552	*
1553	* @param pVmxTransient The VMX-transient structure.
1554	*/
1555	DECLINLINE(void) hmR0VmxReadExitInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
1556	{
1557	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INSTR_LEN))
1558	{
1559	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbExitInstr);
1560	AssertRC(rc);
1561	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INSTR_LEN;
1562	}
1563	}
1564
1565
1566	/**
1567	* Reads the VM-exit instruction-information field from the VMCS into
1568	* the VMX transient structure.
1569	*
1570	* @param pVmxTransient The VMX-transient structure.
1571	*/
1572	DECLINLINE(void) hmR0VmxReadExitInstrInfoVmcs(PVMXTRANSIENT pVmxTransient)
1573	{
1574	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INSTR_INFO))
1575	{
1576	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
1577	AssertRC(rc);
1578	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INSTR_INFO;
1579	}
1580	}
1581
1582
1583	/**
1584	* Reads the Exit Qualification from the VMCS into the VMX transient structure.
1585	*
1586	* @param pVmxTransient The VMX-transient structure.
1587	*/
1588	DECLINLINE(void) hmR0VmxReadExitQualVmcs(PVMXTRANSIENT pVmxTransient)
1589	{
1590	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_QUALIFICATION))
1591	{
1592	int rc = VMXReadVmcsNw(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQual);
1593	AssertRC(rc);
1594	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_QUALIFICATION;
1595	}
1596	}
1597
1598
1599	/**
1600	* Reads the Guest-linear address from the VMCS into the VMX transient structure.
1601	*
1602	* @param pVmxTransient The VMX-transient structure.
1603	*/
1604	DECLINLINE(void) hmR0VmxReadGuestLinearAddrVmcs(PVMXTRANSIENT pVmxTransient)
1605	{
1606	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_LINEAR_ADDR))
1607	{
1608	int rc = VMXReadVmcsNw(VMX_VMCS_RO_GUEST_LINEAR_ADDR, &pVmxTransient->uGuestLinearAddr);
1609	AssertRC(rc);
1610	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_LINEAR_ADDR;
1611	}
1612	}
1613
1614
1615	/**
1616	* Reads the Guest-physical address from the VMCS into the VMX transient structure.
1617	*
1618	* @param pVmxTransient The VMX-transient structure.
1619	*/
1620	DECLINLINE(void) hmR0VmxReadGuestPhysicalAddrVmcs(PVMXTRANSIENT pVmxTransient)
1621	{
1622	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_PHYSICAL_ADDR))
1623	{
1624	int rc = VMXReadVmcs64(VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL, &pVmxTransient->uGuestPhysicalAddr);
1625	AssertRC(rc);
1626	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_PHYSICAL_ADDR;
1627	}
1628	}
1629
1630	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1631	/**
1632	* Reads the Guest pending-debug exceptions from the VMCS into the VMX transient
1633	* structure.
1634	*
1635	* @param pVmxTransient The VMX-transient structure.
1636	*/
1637	DECLINLINE(void) hmR0VmxReadGuestPendingDbgXctps(PVMXTRANSIENT pVmxTransient)
1638	{
1639	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_PENDING_DBG_XCPTS))
1640	{
1641	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, &pVmxTransient->uGuestPendingDbgXcpts);
1642	AssertRC(rc);
1643	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_PENDING_DBG_XCPTS;
1644	}
1645	}
1646	#endif
1647
1648	/**
1649	* Reads the IDT-vectoring information field from the VMCS into the VMX
1650	* transient structure.
1651	*
1652	* @param pVmxTransient The VMX-transient structure.
1653	*
1654	* @remarks No-long-jump zone!!!
1655	*/
1656	DECLINLINE(void) hmR0VmxReadIdtVectoringInfoVmcs(PVMXTRANSIENT pVmxTransient)
1657	{
1658	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_IDT_VECTORING_INFO))
1659	{
1660	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_INFO, &pVmxTransient->uIdtVectoringInfo);
1661	AssertRC(rc);
1662	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_IDT_VECTORING_INFO;
1663	}
1664	}
1665
1666
1667	/**
1668	* Reads the IDT-vectoring error code from the VMCS into the VMX
1669	* transient structure.
1670	*
1671	* @param pVmxTransient The VMX-transient structure.
1672	*/
1673	DECLINLINE(void) hmR0VmxReadIdtVectoringErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1674	{
1675	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_IDT_VECTORING_ERROR_CODE))
1676	{
1677	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
1678	AssertRC(rc);
1679	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_IDT_VECTORING_ERROR_CODE;
1680	}
1681	}
1682
1683	#ifdef HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
1684	/**
1685	* Reads all relevant read-only VMCS fields into the VMX transient structure.
1686	*
1687	* @param pVmxTransient The VMX-transient structure.
1688	*/
1689	static void hmR0VmxReadAllRoFieldsVmcs(PVMXTRANSIENT pVmxTransient)
1690	{
1691	int rc = VMXReadVmcsNw(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQual);
1692	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbExitInstr);
1693	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
1694	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_INFO, &pVmxTransient->uIdtVectoringInfo);
1695	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
1696	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
1697	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
1698	rc \|= VMXReadVmcsNw(VMX_VMCS_RO_GUEST_LINEAR_ADDR, &pVmxTransient->uGuestLinearAddr);
1699	rc \|= VMXReadVmcs64(VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL, &pVmxTransient->uGuestPhysicalAddr);
1700	AssertRC(rc);
1701	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_QUALIFICATION
1702	\| HMVMX_READ_EXIT_INSTR_LEN
1703	\| HMVMX_READ_EXIT_INSTR_INFO
1704	\| HMVMX_READ_IDT_VECTORING_INFO
1705	\| HMVMX_READ_IDT_VECTORING_ERROR_CODE
1706	\| HMVMX_READ_EXIT_INTERRUPTION_INFO
1707	\| HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE
1708	\| HMVMX_READ_GUEST_LINEAR_ADDR
1709	\| HMVMX_READ_GUEST_PHYSICAL_ADDR;
1710	}
1711	#endif
1712
1713	/**
1714	* Enters VMX root mode operation on the current CPU.
1715	*
1716	* @returns VBox status code.
1717	* @param pHostCpu The HM physical-CPU structure.
1718	* @param pVM The cross context VM structure. Can be
1719	* NULL, after a resume.
1720	* @param HCPhysCpuPage Physical address of the VMXON region.
1721	* @param pvCpuPage Pointer to the VMXON region.
1722	*/
1723	static int hmR0VmxEnterRootMode(PHMPHYSCPU pHostCpu, PVMCC pVM, RTHCPHYS HCPhysCpuPage, void *pvCpuPage)
1724	{
1725	Assert(pHostCpu);
1726	Assert(HCPhysCpuPage && HCPhysCpuPage != NIL_RTHCPHYS);
1727	Assert(RT_ALIGN_T(HCPhysCpuPage, _4K, RTHCPHYS) == HCPhysCpuPage);
1728	Assert(pvCpuPage);
1729	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1730
1731	if (pVM)
1732	{
1733	/* Write the VMCS revision identifier to the VMXON region. */
1734	(uint32_t )pvCpuPage = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_ID);
1735	}
1736
1737	/* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with CR4. */
1738	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1739
1740	/* Enable the VMX bit in CR4 if necessary. */
1741	RTCCUINTREG const uOldCr4 = SUPR0ChangeCR4(X86_CR4_VMXE, RTCCUINTREG_MAX);
1742
1743	/* Record whether VMXE was already prior to us enabling it above. */
1744	pHostCpu->fVmxeAlreadyEnabled = RT_BOOL(uOldCr4 & X86_CR4_VMXE);
1745
1746	/* Enter VMX root mode. */
1747	int rc = VMXEnable(HCPhysCpuPage);
1748	if (RT_FAILURE(rc))
1749	{
1750	/* Restore CR4.VMXE if it was not set prior to our attempt to set it above. */
1751	if (!pHostCpu->fVmxeAlreadyEnabled)
1752	SUPR0ChangeCR4(0 /* fOrMask */, ~(uint64_t)X86_CR4_VMXE);
1753
1754	if (pVM)
1755	pVM->hm.s.ForR3.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
1756	}
1757
1758	/* Restore interrupts. */
1759	ASMSetFlags(fEFlags);
1760	return rc;
1761	}
1762
1763
1764	/**
1765	* Exits VMX root mode operation on the current CPU.
1766	*
1767	* @returns VBox status code.
1768	* @param pHostCpu The HM physical-CPU structure.
1769	*/
1770	static int hmR0VmxLeaveRootMode(PHMPHYSCPU pHostCpu)
1771	{
1772	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1773
1774	/* Paranoid: Disable interrupts as, in theory, interrupts handlers might mess with CR4. */
1775	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1776
1777	/* If we're for some reason not in VMX root mode, then don't leave it. */
1778	RTCCUINTREG const uHostCr4 = ASMGetCR4();
1779
1780	int rc;
1781	if (uHostCr4 & X86_CR4_VMXE)
1782	{
1783	/* Exit VMX root mode and clear the VMX bit in CR4. */
1784	VMXDisable();
1785
1786	/* Clear CR4.VMXE only if it was clear prior to use setting it. */
1787	if (!pHostCpu->fVmxeAlreadyEnabled)
1788	SUPR0ChangeCR4(0 /* fOrMask */, ~(uint64_t)X86_CR4_VMXE);
1789
1790	rc = VINF_SUCCESS;
1791	}
1792	else
1793	rc = VERR_VMX_NOT_IN_VMX_ROOT_MODE;
1794
1795	/* Restore interrupts. */
1796	ASMSetFlags(fEFlags);
1797	return rc;
1798	}
1799
1800
1801	/**
1802	* Allocates pages specified as specified by an array of VMX page allocation info
1803	* objects.
1804	*
1805	* The pages contents are zero'd after allocation.
1806	*
1807	* @returns VBox status code.
1808	* @param phMemObj Where to return the handle to the allocation.
1809	* @param paAllocInfo The pointer to the first element of the VMX
1810	* page-allocation info object array.
1811	* @param cEntries The number of elements in the @a paAllocInfo array.
1812	*/
1813	static int hmR0VmxPagesAllocZ(PRTR0MEMOBJ phMemObj, PVMXPAGEALLOCINFO paAllocInfo, uint32_t cEntries)
1814	{
1815	*phMemObj = NIL_RTR0MEMOBJ;
1816
1817	/* Figure out how many pages to allocate. */
1818	uint32_t cPages = 0;
1819	for (uint32_t iPage = 0; iPage < cEntries; iPage++)
1820	cPages += !!paAllocInfo[iPage].fValid;
1821
1822	/* Allocate the pages. */
1823	if (cPages)
1824	{
1825	size_t const cbPages = cPages << PAGE_SHIFT;
1826	int rc = RTR0MemObjAllocPage(phMemObj, cbPages, false /* fExecutable */);
1827	if (RT_FAILURE(rc))
1828	return rc;
1829
1830	/* Zero the contents and assign each page to the corresponding VMX page-allocation entry. */
1831	void pvFirstPage = RTR0MemObjAddress(phMemObj);
1832	RT_BZERO(pvFirstPage, cbPages);
1833
1834	uint32_t iPage = 0;
1835	for (uint32_t i = 0; i < cEntries; i++)
1836	if (paAllocInfo[i].fValid)
1837	{
1838	RTHCPHYS const HCPhysPage = RTR0MemObjGetPagePhysAddr(*phMemObj, iPage);
1839	void pvPage = (void )((uintptr_t)pvFirstPage + (iPage << X86_PAGE_4K_SHIFT));
1840	Assert(HCPhysPage && HCPhysPage != NIL_RTHCPHYS);
1841	AssertPtr(pvPage);
1842
1843	Assert(paAllocInfo[iPage].pHCPhys);
1844	Assert(paAllocInfo[iPage].ppVirt);
1845	*paAllocInfo[iPage].pHCPhys = HCPhysPage;
1846	*paAllocInfo[iPage].ppVirt = pvPage;
1847
1848	/* Move to next page. */
1849	++iPage;
1850	}
1851
1852	/* Make sure all valid (requested) pages have been assigned. */
1853	Assert(iPage == cPages);
1854	}
1855	return VINF_SUCCESS;
1856	}
1857
1858
1859	/**
1860	* Frees pages allocated using hmR0VmxPagesAllocZ.
1861	*
1862	* @param phMemObj Pointer to the memory object handle. Will be set to
1863	* NIL.
1864	*/
1865	DECL_FORCE_INLINE(void) hmR0VmxPagesFree(PRTR0MEMOBJ phMemObj)
1866	{
1867	/* We can cleanup wholesale since it's all one allocation. */
1868	if (*phMemObj != NIL_RTR0MEMOBJ)
1869	{
1870	RTR0MemObjFree(phMemObj, true / fFreeMappings */);
1871	*phMemObj = NIL_RTR0MEMOBJ;
1872	}
1873	}
1874
1875
1876	/**
1877	* Initializes a VMCS info. object.
1878	*
1879	* @param pVmcsInfo The VMCS info. object.
1880	* @param pVmcsInfoShared The VMCS info. object shared with ring-3.
1881	*/
1882	static void hmR0VmxVmcsInfoInit(PVMXVMCSINFO pVmcsInfo, PVMXVMCSINFOSHARED pVmcsInfoShared)
1883	{
1884	RT_ZERO(*pVmcsInfo);
1885	RT_ZERO(*pVmcsInfoShared);
1886
1887	pVmcsInfo->pShared = pVmcsInfoShared;
1888	Assert(pVmcsInfo->hMemObj == NIL_RTR0MEMOBJ);
1889	pVmcsInfo->HCPhysVmcs = NIL_RTHCPHYS;
1890	pVmcsInfo->HCPhysShadowVmcs = NIL_RTHCPHYS;
1891	pVmcsInfo->HCPhysMsrBitmap = NIL_RTHCPHYS;
1892	pVmcsInfo->HCPhysGuestMsrLoad = NIL_RTHCPHYS;
1893	pVmcsInfo->HCPhysGuestMsrStore = NIL_RTHCPHYS;
1894	pVmcsInfo->HCPhysHostMsrLoad = NIL_RTHCPHYS;
1895	pVmcsInfo->HCPhysVirtApic = NIL_RTHCPHYS;
1896	pVmcsInfo->HCPhysEPTP = NIL_RTHCPHYS;
1897	pVmcsInfo->u64VmcsLinkPtr = NIL_RTHCPHYS;
1898	pVmcsInfo->idHostCpuState = NIL_RTCPUID;
1899	pVmcsInfo->idHostCpuExec = NIL_RTCPUID;
1900	}
1901
1902
1903	/**
1904	* Frees the VT-x structures for a VMCS info. object.
1905	*
1906	* @param pVmcsInfo The VMCS info. object.
1907	* @param pVmcsInfoShared The VMCS info. object shared with ring-3.
1908	*/
1909	static void hmR0VmxVmcsInfoFree(PVMXVMCSINFO pVmcsInfo, PVMXVMCSINFOSHARED pVmcsInfoShared)
1910	{
1911	hmR0VmxPagesFree(&pVmcsInfo->hMemObj);
1912	hmR0VmxVmcsInfoInit(pVmcsInfo, pVmcsInfoShared);
1913	}
1914
1915
1916	/**
1917	* Allocates the VT-x structures for a VMCS info. object.
1918	*
1919	* @returns VBox status code.
1920	* @param pVCpu The cross context virtual CPU structure.
1921	* @param pVmcsInfo The VMCS info. object.
1922	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
1923	*
1924	* @remarks The caller is expected to take care of any and all allocation failures.
1925	* This function will not perform any cleanup for failures half-way
1926	* through.
1927	*/
1928	static int hmR0VmxAllocVmcsInfo(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
1929	{
1930	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1931
1932	bool const fMsrBitmaps = RT_BOOL(g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS);
1933	bool const fShadowVmcs = !fIsNstGstVmcs ? pVM->hmr0.s.vmx.fUseVmcsShadowing : pVM->cpum.ro.GuestFeatures.fVmxVmcsShadowing;
1934	Assert(!pVM->cpum.ro.GuestFeatures.fVmxVmcsShadowing); /* VMCS shadowing is not yet exposed to the guest. */
1935	VMXPAGEALLOCINFO aAllocInfo[] =
1936	{
1937	{ true, 0 /* Unused */, &pVmcsInfo->HCPhysVmcs, &pVmcsInfo->pvVmcs },
1938	{ true, 0 /* Unused */, &pVmcsInfo->HCPhysGuestMsrLoad, &pVmcsInfo->pvGuestMsrLoad },
1939	{ true, 0 /* Unused */, &pVmcsInfo->HCPhysHostMsrLoad, &pVmcsInfo->pvHostMsrLoad },
1940	{ fMsrBitmaps, 0 /* Unused */, &pVmcsInfo->HCPhysMsrBitmap, &pVmcsInfo->pvMsrBitmap },
1941	{ fShadowVmcs, 0 /* Unused */, &pVmcsInfo->HCPhysShadowVmcs, &pVmcsInfo->pvShadowVmcs },
1942	};
1943
1944	int rc = hmR0VmxPagesAllocZ(&pVmcsInfo->hMemObj, &aAllocInfo[0], RT_ELEMENTS(aAllocInfo));
1945	if (RT_FAILURE(rc))
1946	return rc;
1947
1948	/*
1949	* We use the same page for VM-entry MSR-load and VM-exit MSR store areas.
1950	* Because they contain a symmetric list of guest MSRs to load on VM-entry and store on VM-exit.
1951	*/
1952	AssertCompile(RT_ELEMENTS(aAllocInfo) > 0);
1953	Assert(pVmcsInfo->HCPhysGuestMsrLoad != NIL_RTHCPHYS);
1954	pVmcsInfo->pvGuestMsrStore = pVmcsInfo->pvGuestMsrLoad;
1955	pVmcsInfo->HCPhysGuestMsrStore = pVmcsInfo->HCPhysGuestMsrLoad;
1956
1957	/*
1958	* Get the virtual-APIC page rather than allocating them again.
1959	*/
1960	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW)
1961	{
1962	if (!fIsNstGstVmcs)
1963	{
1964	if (PDMHasApic(pVM))
1965	{
1966	rc = APICGetApicPageForCpu(pVCpu, &pVmcsInfo->HCPhysVirtApic, (PRTR0PTR)&pVmcsInfo->pbVirtApic, NULL /pR3Ptr/);
1967	if (RT_FAILURE(rc))
1968	return rc;
1969	Assert(pVmcsInfo->pbVirtApic);
1970	Assert(pVmcsInfo->HCPhysVirtApic && pVmcsInfo->HCPhysVirtApic != NIL_RTHCPHYS);
1971	}
1972	}
1973	else
1974	{
1975	pVmcsInfo->pbVirtApic = &pVCpu->cpum.GstCtx.hwvirt.vmx.abVirtApicPage[0];
1976	pVmcsInfo->HCPhysVirtApic = GVMMR0ConvertGVMPtr2HCPhys(pVM, pVmcsInfo->pbVirtApic);
1977	Assert(pVmcsInfo->HCPhysVirtApic && pVmcsInfo->HCPhysVirtApic != NIL_RTHCPHYS);
1978	}
1979	}
1980
1981	return VINF_SUCCESS;
1982	}
1983
1984
1985	/**
1986	* Free all VT-x structures for the VM.
1987	*
1988	* @returns IPRT status code.
1989	* @param pVM The cross context VM structure.
1990	*/
1991	static void hmR0VmxStructsFree(PVMCC pVM)
1992	{
1993	hmR0VmxPagesFree(&pVM->hmr0.s.vmx.hMemObj);
1994	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1995	if (pVM->hmr0.s.vmx.fUseVmcsShadowing)
1996	{
1997	RTMemFree(pVM->hmr0.s.vmx.paShadowVmcsFields);
1998	pVM->hmr0.s.vmx.paShadowVmcsFields = NULL;
1999	RTMemFree(pVM->hmr0.s.vmx.paShadowVmcsRoFields);
2000	pVM->hmr0.s.vmx.paShadowVmcsRoFields = NULL;
2001	}
2002	#endif
2003
2004	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
2005	{
2006	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
2007	hmR0VmxVmcsInfoFree(&pVCpu->hmr0.s.vmx.VmcsInfo, &pVCpu->hm.s.vmx.VmcsInfo);
2008	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2009	if (pVM->cpum.ro.GuestFeatures.fVmx)
2010	hmR0VmxVmcsInfoFree(&pVCpu->hmr0.s.vmx.VmcsInfoNstGst, &pVCpu->hm.s.vmx.VmcsInfoNstGst);
2011	#endif
2012	}
2013	}
2014
2015
2016	/**
2017	* Allocate all VT-x structures for the VM.
2018	*
2019	* @returns IPRT status code.
2020	* @param pVM The cross context VM structure.
2021	*
2022	* @remarks This functions will cleanup on memory allocation failures.
2023	*/
2024	static int hmR0VmxStructsAlloc(PVMCC pVM)
2025	{
2026	/*
2027	* Sanity check the VMCS size reported by the CPU as we assume 4KB allocations.
2028	* The VMCS size cannot be more than 4096 bytes.
2029	*
2030	* See Intel spec. Appendix A.1 "Basic VMX Information".
2031	*/
2032	uint32_t const cbVmcs = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_SIZE);
2033	if (cbVmcs <= X86_PAGE_4K_SIZE)
2034	{ /* likely */ }
2035	else
2036	{
2037	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_INVALID_VMCS_SIZE;
2038	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2039	}
2040
2041	/*
2042	* Allocate per-VM VT-x structures.
2043	*/
2044	bool const fVirtApicAccess = RT_BOOL(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS);
2045	bool const fUseVmcsShadowing = pVM->hmr0.s.vmx.fUseVmcsShadowing;
2046	VMXPAGEALLOCINFO aAllocInfo[] =
2047	{
2048	{ fVirtApicAccess, 0 /* Unused */, &pVM->hmr0.s.vmx.HCPhysApicAccess, (PRTR0PTR)&pVM->hmr0.s.vmx.pbApicAccess },
2049	{ fUseVmcsShadowing, 0 /* Unused */, &pVM->hmr0.s.vmx.HCPhysVmreadBitmap, &pVM->hmr0.s.vmx.pvVmreadBitmap },
2050	{ fUseVmcsShadowing, 0 /* Unused */, &pVM->hmr0.s.vmx.HCPhysVmwriteBitmap, &pVM->hmr0.s.vmx.pvVmwriteBitmap },
2051	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2052	{ true, 0 /* Unused */, &pVM->hmr0.s.vmx.HCPhysScratch, (PRTR0PTR)&pVM->hmr0.s.vmx.pbScratch },
2053	#endif
2054	};
2055
2056	int rc = hmR0VmxPagesAllocZ(&pVM->hmr0.s.vmx.hMemObj, &aAllocInfo[0], RT_ELEMENTS(aAllocInfo));
2057	if (RT_SUCCESS(rc))
2058	{
2059	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2060	/* Allocate the shadow VMCS-fields array. */
2061	if (fUseVmcsShadowing)
2062	{
2063	Assert(!pVM->hmr0.s.vmx.cShadowVmcsFields);
2064	Assert(!pVM->hmr0.s.vmx.cShadowVmcsRoFields);
2065	pVM->hmr0.s.vmx.paShadowVmcsFields = (uint32_t *)RTMemAllocZ(sizeof(g_aVmcsFields));
2066	pVM->hmr0.s.vmx.paShadowVmcsRoFields = (uint32_t *)RTMemAllocZ(sizeof(g_aVmcsFields));
2067	if (!pVM->hmr0.s.vmx.paShadowVmcsFields \|\| !pVM->hmr0.s.vmx.paShadowVmcsRoFields)
2068	rc = VERR_NO_MEMORY;
2069	}
2070	#endif
2071
2072	/*
2073	* Allocate per-VCPU VT-x structures.
2074	*/
2075	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus && RT_SUCCESS(rc); idCpu++)
2076	{
2077	/* Allocate the guest VMCS structures. */
2078	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
2079	rc = hmR0VmxAllocVmcsInfo(pVCpu, &pVCpu->hmr0.s.vmx.VmcsInfo, false /* fIsNstGstVmcs */);
2080
2081	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2082	/* Allocate the nested-guest VMCS structures, when the VMX feature is exposed to the guest. */
2083	if (pVM->cpum.ro.GuestFeatures.fVmx && RT_SUCCESS(rc))
2084	rc = hmR0VmxAllocVmcsInfo(pVCpu, &pVCpu->hmr0.s.vmx.VmcsInfoNstGst, true /* fIsNstGstVmcs */);
2085	#endif
2086	}
2087	if (RT_SUCCESS(rc))
2088	return VINF_SUCCESS;
2089	}
2090	hmR0VmxStructsFree(pVM);
2091	return rc;
2092	}
2093
2094
2095	/**
2096	* Pre-initializes non-zero fields in VMX structures that will be allocated.
2097	*
2098	* @param pVM The cross context VM structure.
2099	*/
2100	static void hmR0VmxStructsInit(PVMCC pVM)
2101	{
2102	/* Paranoia. */
2103	Assert(pVM->hmr0.s.vmx.pbApicAccess == NULL);
2104	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2105	Assert(pVM->hmr0.s.vmx.pbScratch == NULL);
2106	#endif
2107
2108	/*
2109	* Initialize members up-front so we can cleanup en masse on allocation failures.
2110	*/
2111	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2112	pVM->hmr0.s.vmx.HCPhysScratch = NIL_RTHCPHYS;
2113	#endif
2114	pVM->hmr0.s.vmx.HCPhysApicAccess = NIL_RTHCPHYS;
2115	pVM->hmr0.s.vmx.HCPhysVmreadBitmap = NIL_RTHCPHYS;
2116	pVM->hmr0.s.vmx.HCPhysVmwriteBitmap = NIL_RTHCPHYS;
2117	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
2118	{
2119	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
2120	hmR0VmxVmcsInfoInit(&pVCpu->hmr0.s.vmx.VmcsInfo, &pVCpu->hm.s.vmx.VmcsInfo);
2121	hmR0VmxVmcsInfoInit(&pVCpu->hmr0.s.vmx.VmcsInfoNstGst, &pVCpu->hm.s.vmx.VmcsInfoNstGst);
2122	}
2123	}
2124
2125	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2126	/**
2127	* Returns whether an MSR at the given MSR-bitmap offset is intercepted or not.
2128	*
2129	* @returns @c true if the MSR is intercepted, @c false otherwise.
2130	* @param pbMsrBitmap The MSR bitmap.
2131	* @param offMsr The MSR byte offset.
2132	* @param iBit The bit offset from the byte offset.
2133	*/
2134	DECLINLINE(bool) hmR0VmxIsMsrBitSet(uint8_t const *pbMsrBitmap, uint16_t offMsr, int32_t iBit)
2135	{
2136	Assert(offMsr + (iBit >> 3) <= X86_PAGE_4K_SIZE);
2137	return ASMBitTest(pbMsrBitmap + offMsr, iBit);
2138	}
2139	#endif
2140
2141	/**
2142	* Sets the permission bits for the specified MSR in the given MSR bitmap.
2143	*
2144	* If the passed VMCS is a nested-guest VMCS, this function ensures that the
2145	* read/write intercept is cleared from the MSR bitmap used for hardware-assisted
2146	* VMX execution of the nested-guest, only if nested-guest is also not intercepting
2147	* the read/write access of this MSR.
2148	*
2149	* @param pVCpu The cross context virtual CPU structure.
2150	* @param pVmcsInfo The VMCS info. object.
2151	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2152	* @param idMsr The MSR value.
2153	* @param fMsrpm The MSR permissions (see VMXMSRPM_XXX). This must
2154	* include both a read -and- a write permission!
2155	*
2156	* @sa CPUMGetVmxMsrPermission.
2157	* @remarks Can be called with interrupts disabled.
2158	*/
2159	static void hmR0VmxSetMsrPermission(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs, uint32_t idMsr, uint32_t fMsrpm)
2160	{
2161	uint8_t pbMsrBitmap = (uint8_t )pVmcsInfo->pvMsrBitmap;
2162	Assert(pbMsrBitmap);
2163	Assert(VMXMSRPM_IS_FLAG_VALID(fMsrpm));
2164
2165	/*
2166	* MSR-bitmap Layout:
2167	* Byte index MSR range Interpreted as
2168	* 0x000 - 0x3ff 0x00000000 - 0x00001fff Low MSR read bits.
2169	* 0x400 - 0x7ff 0xc0000000 - 0xc0001fff High MSR read bits.
2170	* 0x800 - 0xbff 0x00000000 - 0x00001fff Low MSR write bits.
2171	* 0xc00 - 0xfff 0xc0000000 - 0xc0001fff High MSR write bits.
2172	*
2173	* A bit corresponding to an MSR within the above range causes a VM-exit
2174	* if the bit is 1 on executions of RDMSR/WRMSR. If an MSR falls out of
2175	* the MSR range, it always cause a VM-exit.
2176	*
2177	* See Intel spec. 24.6.9 "MSR-Bitmap Address".
2178	*/
2179	uint16_t const offBitmapRead = 0;
2180	uint16_t const offBitmapWrite = 0x800;
2181	uint16_t offMsr;
2182	int32_t iBit;
2183	if (idMsr <= UINT32_C(0x00001fff))
2184	{
2185	offMsr = 0;
2186	iBit = idMsr;
2187	}
2188	else if (idMsr - UINT32_C(0xc0000000) <= UINT32_C(0x00001fff))
2189	{
2190	offMsr = 0x400;
2191	iBit = idMsr - UINT32_C(0xc0000000);
2192	}
2193	else
2194	AssertMsgFailedReturnVoid(("Invalid MSR %#RX32\n", idMsr));
2195
2196	/*
2197	* Set the MSR read permission.
2198	*/
2199	uint16_t const offMsrRead = offBitmapRead + offMsr;
2200	Assert(offMsrRead + (iBit >> 3) < offBitmapWrite);
2201	if (fMsrpm & VMXMSRPM_ALLOW_RD)
2202	{
2203	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2204	bool const fClear = !fIsNstGstVmcs ? true
2205	: !hmR0VmxIsMsrBitSet(pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap, offMsrRead, iBit);
2206	#else
2207	RT_NOREF2(pVCpu, fIsNstGstVmcs);
2208	bool const fClear = true;
2209	#endif
2210	if (fClear)
2211	ASMBitClear(pbMsrBitmap + offMsrRead, iBit);
2212	}
2213	else
2214	ASMBitSet(pbMsrBitmap + offMsrRead, iBit);
2215
2216	/*
2217	* Set the MSR write permission.
2218	*/
2219	uint16_t const offMsrWrite = offBitmapWrite + offMsr;
2220	Assert(offMsrWrite + (iBit >> 3) < X86_PAGE_4K_SIZE);
2221	if (fMsrpm & VMXMSRPM_ALLOW_WR)
2222	{
2223	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2224	bool const fClear = !fIsNstGstVmcs ? true
2225	: !hmR0VmxIsMsrBitSet(pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap, offMsrWrite, iBit);
2226	#else
2227	RT_NOREF2(pVCpu, fIsNstGstVmcs);
2228	bool const fClear = true;
2229	#endif
2230	if (fClear)
2231	ASMBitClear(pbMsrBitmap + offMsrWrite, iBit);
2232	}
2233	else
2234	ASMBitSet(pbMsrBitmap + offMsrWrite, iBit);
2235	}
2236
2237
2238	/**
2239	* Updates the VMCS with the number of effective MSRs in the auto-load/store MSR
2240	* area.
2241	*
2242	* @returns VBox status code.
2243	* @param pVCpu The cross context virtual CPU structure.
2244	* @param pVmcsInfo The VMCS info. object.
2245	* @param cMsrs The number of MSRs.
2246	*/
2247	static int hmR0VmxSetAutoLoadStoreMsrCount(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint32_t cMsrs)
2248	{
2249	/* Shouldn't ever happen but there -is- a number. We're well within the recommended 512. */
2250	uint32_t const cMaxSupportedMsrs = VMX_MISC_MAX_MSRS(g_HmMsrs.u.vmx.u64Misc);
2251	if (RT_LIKELY(cMsrs < cMaxSupportedMsrs))
2252	{
2253	/* Commit the MSR counts to the VMCS and update the cache. */
2254	if (pVmcsInfo->cEntryMsrLoad != cMsrs)
2255	{
2256	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, cMsrs); AssertRC(rc);
2257	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, cMsrs); AssertRC(rc);
2258	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, cMsrs); AssertRC(rc);
2259	pVmcsInfo->cEntryMsrLoad = cMsrs;
2260	pVmcsInfo->cExitMsrStore = cMsrs;
2261	pVmcsInfo->cExitMsrLoad = cMsrs;
2262	}
2263	return VINF_SUCCESS;
2264	}
2265
2266	LogRel(("Auto-load/store MSR count exceeded! cMsrs=%u MaxSupported=%u\n", cMsrs, cMaxSupportedMsrs));
2267	pVCpu->hm.s.u32HMError = VMX_UFC_INSUFFICIENT_GUEST_MSR_STORAGE;
2268	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2269	}
2270
2271
2272	/**
2273	* Adds a new (or updates the value of an existing) guest/host MSR
2274	* pair to be swapped during the world-switch as part of the
2275	* auto-load/store MSR area in the VMCS.
2276	*
2277	* @returns VBox status code.
2278	* @param pVCpu The cross context virtual CPU structure.
2279	* @param pVmxTransient The VMX-transient structure.
2280	* @param idMsr The MSR.
2281	* @param uGuestMsrValue Value of the guest MSR.
2282	* @param fSetReadWrite Whether to set the guest read/write access of this
2283	* MSR (thus not causing a VM-exit).
2284	* @param fUpdateHostMsr Whether to update the value of the host MSR if
2285	* necessary.
2286	*/
2287	static int hmR0VmxAddAutoLoadStoreMsr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t idMsr, uint64_t uGuestMsrValue,
2288	bool fSetReadWrite, bool fUpdateHostMsr)
2289	{
2290	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
2291	bool const fIsNstGstVmcs = pVmxTransient->fIsNestedGuest;
2292	PVMXAUTOMSR pGuestMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2293	uint32_t cMsrs = pVmcsInfo->cEntryMsrLoad;
2294	uint32_t i;
2295
2296	/* Paranoia. */
2297	Assert(pGuestMsrLoad);
2298
2299	#ifndef DEBUG_bird
2300	LogFlowFunc(("pVCpu=%p idMsr=%#RX32 uGuestMsrValue=%#RX64\n", pVCpu, idMsr, uGuestMsrValue));
2301	#endif
2302
2303	/* Check if the MSR already exists in the VM-entry MSR-load area. */
2304	for (i = 0; i < cMsrs; i++)
2305	{
2306	if (pGuestMsrLoad[i].u32Msr == idMsr)
2307	break;
2308	}
2309
2310	bool fAdded = false;
2311	if (i == cMsrs)
2312	{
2313	/* The MSR does not exist, bump the MSR count to make room for the new MSR. */
2314	++cMsrs;
2315	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, pVmcsInfo, cMsrs);
2316	AssertMsgRCReturn(rc, ("Insufficient space to add MSR to VM-entry MSR-load/store area %u\n", idMsr), rc);
2317
2318	/* Set the guest to read/write this MSR without causing VM-exits. */
2319	if ( fSetReadWrite
2320	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS))
2321	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, fIsNstGstVmcs, idMsr, VMXMSRPM_ALLOW_RD_WR);
2322
2323	Log4Func(("Added MSR %#RX32, cMsrs=%u\n", idMsr, cMsrs));
2324	fAdded = true;
2325	}
2326
2327	/* Update the MSR value for the newly added or already existing MSR. */
2328	pGuestMsrLoad[i].u32Msr = idMsr;
2329	pGuestMsrLoad[i].u64Value = uGuestMsrValue;
2330
2331	/* Create the corresponding slot in the VM-exit MSR-store area if we use a different page. */
2332	if (hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo))
2333	{
2334	PVMXAUTOMSR pGuestMsrStore = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2335	pGuestMsrStore[i].u32Msr = idMsr;
2336	pGuestMsrStore[i].u64Value = uGuestMsrValue;
2337	}
2338
2339	/* Update the corresponding slot in the host MSR area. */
2340	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2341	Assert(pHostMsr != pVmcsInfo->pvGuestMsrLoad);
2342	Assert(pHostMsr != pVmcsInfo->pvGuestMsrStore);
2343	pHostMsr[i].u32Msr = idMsr;
2344
2345	/*
2346	* Only if the caller requests to update the host MSR value AND we've newly added the
2347	* MSR to the host MSR area do we actually update the value. Otherwise, it will be
2348	* updated by hmR0VmxUpdateAutoLoadHostMsrs().
2349	*
2350	* We do this for performance reasons since reading MSRs may be quite expensive.
2351	*/
2352	if (fAdded)
2353	{
2354	if (fUpdateHostMsr)
2355	{
2356	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2357	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2358	pHostMsr[i].u64Value = ASMRdMsr(idMsr);
2359	}
2360	else
2361	{
2362	/* Someone else can do the work. */
2363	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = false;
2364	}
2365	}
2366	return VINF_SUCCESS;
2367	}
2368
2369
2370	/**
2371	* Removes a guest/host MSR pair to be swapped during the world-switch from the
2372	* auto-load/store MSR area in the VMCS.
2373	*
2374	* @returns VBox status code.
2375	* @param pVCpu The cross context virtual CPU structure.
2376	* @param pVmxTransient The VMX-transient structure.
2377	* @param idMsr The MSR.
2378	*/
2379	static int hmR0VmxRemoveAutoLoadStoreMsr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t idMsr)
2380	{
2381	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
2382	bool const fIsNstGstVmcs = pVmxTransient->fIsNestedGuest;
2383	PVMXAUTOMSR pGuestMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2384	uint32_t cMsrs = pVmcsInfo->cEntryMsrLoad;
2385
2386	#ifndef DEBUG_bird
2387	LogFlowFunc(("pVCpu=%p idMsr=%#RX32\n", pVCpu, idMsr));
2388	#endif
2389
2390	for (uint32_t i = 0; i < cMsrs; i++)
2391	{
2392	/* Find the MSR. */
2393	if (pGuestMsrLoad[i].u32Msr == idMsr)
2394	{
2395	/*
2396	* If it's the last MSR, we only need to reduce the MSR count.
2397	* If it's -not- the last MSR, copy the last MSR in place of it and reduce the MSR count.
2398	*/
2399	if (i < cMsrs - 1)
2400	{
2401	/* Remove it from the VM-entry MSR-load area. */
2402	pGuestMsrLoad[i].u32Msr = pGuestMsrLoad[cMsrs - 1].u32Msr;
2403	pGuestMsrLoad[i].u64Value = pGuestMsrLoad[cMsrs - 1].u64Value;
2404
2405	/* Remove it from the VM-exit MSR-store area if it's in a different page. */
2406	if (hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo))
2407	{
2408	PVMXAUTOMSR pGuestMsrStore = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2409	Assert(pGuestMsrStore[i].u32Msr == idMsr);
2410	pGuestMsrStore[i].u32Msr = pGuestMsrStore[cMsrs - 1].u32Msr;
2411	pGuestMsrStore[i].u64Value = pGuestMsrStore[cMsrs - 1].u64Value;
2412	}
2413
2414	/* Remove it from the VM-exit MSR-load area. */
2415	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2416	Assert(pHostMsr[i].u32Msr == idMsr);
2417	pHostMsr[i].u32Msr = pHostMsr[cMsrs - 1].u32Msr;
2418	pHostMsr[i].u64Value = pHostMsr[cMsrs - 1].u64Value;
2419	}
2420
2421	/* Reduce the count to reflect the removed MSR and bail. */
2422	--cMsrs;
2423	break;
2424	}
2425	}
2426
2427	/* Update the VMCS if the count changed (meaning the MSR was found and removed). */
2428	if (cMsrs != pVmcsInfo->cEntryMsrLoad)
2429	{
2430	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, pVmcsInfo, cMsrs);
2431	AssertRCReturn(rc, rc);
2432
2433	/* We're no longer swapping MSRs during the world-switch, intercept guest read/writes to them. */
2434	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
2435	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, fIsNstGstVmcs, idMsr, VMXMSRPM_EXIT_RD \| VMXMSRPM_EXIT_WR);
2436
2437	Log4Func(("Removed MSR %#RX32, cMsrs=%u\n", idMsr, cMsrs));
2438	return VINF_SUCCESS;
2439	}
2440
2441	return VERR_NOT_FOUND;
2442	}
2443
2444
2445	/**
2446	* Checks if the specified guest MSR is part of the VM-entry MSR-load area.
2447	*
2448	* @returns @c true if found, @c false otherwise.
2449	* @param pVmcsInfo The VMCS info. object.
2450	* @param idMsr The MSR to find.
2451	*/
2452	static bool hmR0VmxIsAutoLoadGuestMsr(PCVMXVMCSINFO pVmcsInfo, uint32_t idMsr)
2453	{
2454	PCVMXAUTOMSR pMsrs = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2455	uint32_t const cMsrs = pVmcsInfo->cEntryMsrLoad;
2456	Assert(pMsrs);
2457	Assert(sizeof(pMsrs) cMsrs <= X86_PAGE_4K_SIZE);
2458	for (uint32_t i = 0; i < cMsrs; i++)
2459	{
2460	if (pMsrs[i].u32Msr == idMsr)
2461	return true;
2462	}
2463	return false;
2464	}
2465
2466
2467	/**
2468	* Updates the value of all host MSRs in the VM-exit MSR-load area.
2469	*
2470	* @param pVCpu The cross context virtual CPU structure.
2471	* @param pVmcsInfo The VMCS info. object.
2472	*
2473	* @remarks No-long-jump zone!!!
2474	*/
2475	static void hmR0VmxUpdateAutoLoadHostMsrs(PCVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
2476	{
2477	RT_NOREF(pVCpu);
2478	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2479
2480	PVMXAUTOMSR pHostMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2481	uint32_t const cMsrs = pVmcsInfo->cExitMsrLoad;
2482	Assert(pHostMsrLoad);
2483	Assert(sizeof(pHostMsrLoad) cMsrs <= X86_PAGE_4K_SIZE);
2484	LogFlowFunc(("pVCpu=%p cMsrs=%u\n", pVCpu, cMsrs));
2485	for (uint32_t i = 0; i < cMsrs; i++)
2486	{
2487	/*
2488	* Performance hack for the host EFER MSR. We use the cached value rather than re-read it.
2489	* Strict builds will catch mismatches in hmR0VmxCheckAutoLoadStoreMsrs(). See @bugref{7368}.
2490	*/
2491	if (pHostMsrLoad[i].u32Msr == MSR_K6_EFER)
2492	pHostMsrLoad[i].u64Value = g_uHmVmxHostMsrEfer;
2493	else
2494	pHostMsrLoad[i].u64Value = ASMRdMsr(pHostMsrLoad[i].u32Msr);
2495	}
2496	}
2497
2498
2499	/**
2500	* Saves a set of host MSRs to allow read/write passthru access to the guest and
2501	* perform lazy restoration of the host MSRs while leaving VT-x.
2502	*
2503	* @param pVCpu The cross context virtual CPU structure.
2504	*
2505	* @remarks No-long-jump zone!!!
2506	*/
2507	static void hmR0VmxLazySaveHostMsrs(PVMCPUCC pVCpu)
2508	{
2509	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2510
2511	/*
2512	* Note: If you're adding MSRs here, make sure to update the MSR-bitmap accesses in hmR0VmxSetupVmcsProcCtls().
2513	*/
2514	if (!(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST))
2515	{
2516	Assert(!(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)); /* Guest MSRs better not be loaded now. */
2517	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fAllow64BitGuests)
2518	{
2519	pVCpu->hmr0.s.vmx.u64HostMsrLStar = ASMRdMsr(MSR_K8_LSTAR);
2520	pVCpu->hmr0.s.vmx.u64HostMsrStar = ASMRdMsr(MSR_K6_STAR);
2521	pVCpu->hmr0.s.vmx.u64HostMsrSfMask = ASMRdMsr(MSR_K8_SF_MASK);
2522	pVCpu->hmr0.s.vmx.u64HostMsrKernelGsBase = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
2523	}
2524	pVCpu->hmr0.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_SAVED_HOST;
2525	}
2526	}
2527
2528
2529	/**
2530	* Checks whether the MSR belongs to the set of guest MSRs that we restore
2531	* lazily while leaving VT-x.
2532	*
2533	* @returns true if it does, false otherwise.
2534	* @param pVCpu The cross context virtual CPU structure.
2535	* @param idMsr The MSR to check.
2536	*/
2537	static bool hmR0VmxIsLazyGuestMsr(PCVMCPUCC pVCpu, uint32_t idMsr)
2538	{
2539	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fAllow64BitGuests)
2540	{
2541	switch (idMsr)
2542	{
2543	case MSR_K8_LSTAR:
2544	case MSR_K6_STAR:
2545	case MSR_K8_SF_MASK:
2546	case MSR_K8_KERNEL_GS_BASE:
2547	return true;
2548	}
2549	}
2550	return false;
2551	}
2552
2553
2554	/**
2555	* Loads a set of guests MSRs to allow read/passthru to the guest.
2556	*
2557	* The name of this function is slightly confusing. This function does NOT
2558	* postpone loading, but loads the MSR right now. "hmR0VmxLazy" is simply a
2559	* common prefix for functions dealing with "lazy restoration" of the shared
2560	* MSRs.
2561	*
2562	* @param pVCpu The cross context virtual CPU structure.
2563	*
2564	* @remarks No-long-jump zone!!!
2565	*/
2566	static void hmR0VmxLazyLoadGuestMsrs(PVMCPUCC pVCpu)
2567	{
2568	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2569	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2570
2571	Assert(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
2572	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fAllow64BitGuests)
2573	{
2574	/*
2575	* If the guest MSRs are not loaded -and- if all the guest MSRs are identical
2576	* to the MSRs on the CPU (which are the saved host MSRs, see assertion above) then
2577	* we can skip a few MSR writes.
2578	*
2579	* Otherwise, it implies either 1. they're not loaded, or 2. they're loaded but the
2580	* guest MSR values in the guest-CPU context might be different to what's currently
2581	* loaded in the CPU. In either case, we need to write the new guest MSR values to the
2582	* CPU, see @bugref{8728}.
2583	*/
2584	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
2585	if ( !(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
2586	&& pCtx->msrKERNELGSBASE == pVCpu->hmr0.s.vmx.u64HostMsrKernelGsBase
2587	&& pCtx->msrLSTAR == pVCpu->hmr0.s.vmx.u64HostMsrLStar
2588	&& pCtx->msrSTAR == pVCpu->hmr0.s.vmx.u64HostMsrStar
2589	&& pCtx->msrSFMASK == pVCpu->hmr0.s.vmx.u64HostMsrSfMask)
2590	{
2591	#ifdef VBOX_STRICT
2592	Assert(ASMRdMsr(MSR_K8_KERNEL_GS_BASE) == pCtx->msrKERNELGSBASE);
2593	Assert(ASMRdMsr(MSR_K8_LSTAR) == pCtx->msrLSTAR);
2594	Assert(ASMRdMsr(MSR_K6_STAR) == pCtx->msrSTAR);
2595	Assert(ASMRdMsr(MSR_K8_SF_MASK) == pCtx->msrSFMASK);
2596	#endif
2597	}
2598	else
2599	{
2600	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pCtx->msrKERNELGSBASE);
2601	ASMWrMsr(MSR_K8_LSTAR, pCtx->msrLSTAR);
2602	ASMWrMsr(MSR_K6_STAR, pCtx->msrSTAR);
2603	/* The system call flag mask register isn't as benign and accepting of all
2604	values as the above, so mask it to avoid #GP'ing on corrupted input. */
2605	Assert(!(pCtx->msrSFMASK & ~(uint64_t)UINT32_MAX));
2606	ASMWrMsr(MSR_K8_SF_MASK, pCtx->msrSFMASK & UINT32_MAX);
2607	}
2608	}
2609	pVCpu->hmr0.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_LOADED_GUEST;
2610	}
2611
2612
2613	/**
2614	* Performs lazy restoration of the set of host MSRs if they were previously
2615	* loaded with guest MSR values.
2616	*
2617	* @param pVCpu The cross context virtual CPU structure.
2618	*
2619	* @remarks No-long-jump zone!!!
2620	* @remarks The guest MSRs should have been saved back into the guest-CPU
2621	* context by hmR0VmxImportGuestState()!!!
2622	*/
2623	static void hmR0VmxLazyRestoreHostMsrs(PVMCPUCC pVCpu)
2624	{
2625	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2626	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2627
2628	if (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
2629	{
2630	Assert(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
2631	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fAllow64BitGuests)
2632	{
2633	ASMWrMsr(MSR_K8_LSTAR, pVCpu->hmr0.s.vmx.u64HostMsrLStar);
2634	ASMWrMsr(MSR_K6_STAR, pVCpu->hmr0.s.vmx.u64HostMsrStar);
2635	ASMWrMsr(MSR_K8_SF_MASK, pVCpu->hmr0.s.vmx.u64HostMsrSfMask);
2636	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pVCpu->hmr0.s.vmx.u64HostMsrKernelGsBase);
2637	}
2638	}
2639	pVCpu->hmr0.s.vmx.fLazyMsrs &= ~(VMX_LAZY_MSRS_LOADED_GUEST \| VMX_LAZY_MSRS_SAVED_HOST);
2640	}
2641
2642
2643	/**
2644	* Verifies that our cached values of the VMCS fields are all consistent with
2645	* what's actually present in the VMCS.
2646	*
2647	* @returns VBox status code.
2648	* @retval VINF_SUCCESS if all our caches match their respective VMCS fields.
2649	* @retval VERR_VMX_VMCS_FIELD_CACHE_INVALID if a cache field doesn't match the
2650	* VMCS content. HMCPU error-field is
2651	* updated, see VMX_VCI_XXX.
2652	* @param pVCpu The cross context virtual CPU structure.
2653	* @param pVmcsInfo The VMCS info. object.
2654	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2655	*/
2656	static int hmR0VmxCheckCachedVmcsCtls(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
2657	{
2658	const char * const pcszVmcs = fIsNstGstVmcs ? "Nested-guest VMCS" : "VMCS";
2659
2660	uint32_t u32Val;
2661	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
2662	AssertRC(rc);
2663	AssertMsgReturnStmt(pVmcsInfo->u32EntryCtls == u32Val,
2664	("%s entry controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32EntryCtls, u32Val),
2665	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_ENTRY,
2666	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2667
2668	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val);
2669	AssertRC(rc);
2670	AssertMsgReturnStmt(pVmcsInfo->u32ExitCtls == u32Val,
2671	("%s exit controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ExitCtls, u32Val),
2672	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_EXIT,
2673	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2674
2675	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val);
2676	AssertRC(rc);
2677	AssertMsgReturnStmt(pVmcsInfo->u32PinCtls == u32Val,
2678	("%s pin controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32PinCtls, u32Val),
2679	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PIN_EXEC,
2680	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2681
2682	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val);
2683	AssertRC(rc);
2684	AssertMsgReturnStmt(pVmcsInfo->u32ProcCtls == u32Val,
2685	("%s proc controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ProcCtls, u32Val),
2686	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC,
2687	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2688
2689	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
2690	{
2691	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val);
2692	AssertRC(rc);
2693	AssertMsgReturnStmt(pVmcsInfo->u32ProcCtls2 == u32Val,
2694	("%s proc2 controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ProcCtls2, u32Val),
2695	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC2,
2696	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2697	}
2698
2699	uint64_t u64Val;
2700	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TERTIARY_CTLS)
2701	{
2702	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_PROC_EXEC3_FULL, &u64Val);
2703	AssertRC(rc);
2704	AssertMsgReturnStmt(pVmcsInfo->u64ProcCtls3 == u64Val,
2705	("%s proc3 controls mismatch: Cache=%#RX32 VMCS=%#RX64\n", pcszVmcs, pVmcsInfo->u64ProcCtls3, u64Val),
2706	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC3,
2707	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2708	}
2709
2710	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, &u32Val);
2711	AssertRC(rc);
2712	AssertMsgReturnStmt(pVmcsInfo->u32XcptBitmap == u32Val,
2713	("%s exception bitmap mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32XcptBitmap, u32Val),
2714	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_XCPT_BITMAP,
2715	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2716
2717	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, &u64Val);
2718	AssertRC(rc);
2719	AssertMsgReturnStmt(pVmcsInfo->u64TscOffset == u64Val,
2720	("%s TSC offset mismatch: Cache=%#RX64 VMCS=%#RX64\n", pcszVmcs, pVmcsInfo->u64TscOffset, u64Val),
2721	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_TSC_OFFSET,
2722	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2723
2724	NOREF(pcszVmcs);
2725	return VINF_SUCCESS;
2726	}
2727
2728	#ifdef VBOX_STRICT
2729
2730	/**
2731	* Verifies that our cached host EFER MSR value has not changed since we cached it.
2732	*
2733	* @param pVmcsInfo The VMCS info. object.
2734	*/
2735	static void hmR0VmxCheckHostEferMsr(PCVMXVMCSINFO pVmcsInfo)
2736	{
2737	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2738
2739	if (pVmcsInfo->u32ExitCtls & VMX_EXIT_CTLS_LOAD_EFER_MSR)
2740	{
2741	uint64_t const uHostEferMsr = ASMRdMsr(MSR_K6_EFER);
2742	uint64_t const uHostEferMsrCache = g_uHmVmxHostMsrEfer;
2743	uint64_t uVmcsEferMsrVmcs;
2744	int rc = VMXReadVmcs64(VMX_VMCS64_HOST_EFER_FULL, &uVmcsEferMsrVmcs);
2745	AssertRC(rc);
2746
2747	AssertMsgReturnVoid(uHostEferMsr == uVmcsEferMsrVmcs,
2748	("EFER Host/VMCS mismatch! host=%#RX64 vmcs=%#RX64\n", uHostEferMsr, uVmcsEferMsrVmcs));
2749	AssertMsgReturnVoid(uHostEferMsr == uHostEferMsrCache,
2750	("EFER Host/Cache mismatch! host=%#RX64 cache=%#RX64\n", uHostEferMsr, uHostEferMsrCache));
2751	}
2752	}
2753
2754
2755	/**
2756	* Verifies whether the guest/host MSR pairs in the auto-load/store area in the
2757	* VMCS are correct.
2758	*
2759	* @param pVCpu The cross context virtual CPU structure.
2760	* @param pVmcsInfo The VMCS info. object.
2761	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2762	*/
2763	static void hmR0VmxCheckAutoLoadStoreMsrs(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
2764	{
2765	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2766
2767	/* Read the various MSR-area counts from the VMCS. */
2768	uint32_t cEntryLoadMsrs;
2769	uint32_t cExitStoreMsrs;
2770	uint32_t cExitLoadMsrs;
2771	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &cEntryLoadMsrs); AssertRC(rc);
2772	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &cExitStoreMsrs); AssertRC(rc);
2773	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &cExitLoadMsrs); AssertRC(rc);
2774
2775	/* Verify all the MSR counts are the same. */
2776	Assert(cEntryLoadMsrs == cExitStoreMsrs);
2777	Assert(cExitStoreMsrs == cExitLoadMsrs);
2778	uint32_t const cMsrs = cExitLoadMsrs;
2779
2780	/* Verify the MSR counts do not exceed the maximum count supported by the hardware. */
2781	Assert(cMsrs < VMX_MISC_MAX_MSRS(g_HmMsrs.u.vmx.u64Misc));
2782
2783	/* Verify the MSR counts are within the allocated page size. */
2784	Assert(sizeof(VMXAUTOMSR) * cMsrs <= X86_PAGE_4K_SIZE);
2785
2786	/* Verify the relevant contents of the MSR areas match. */
2787	PCVMXAUTOMSR pGuestMsrLoad = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2788	PCVMXAUTOMSR pGuestMsrStore = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2789	PCVMXAUTOMSR pHostMsrLoad = (PCVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2790	bool const fSeparateExitMsrStorePage = hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo);
2791	for (uint32_t i = 0; i < cMsrs; i++)
2792	{
2793	/* Verify that the MSRs are paired properly and that the host MSR has the correct value. */
2794	if (fSeparateExitMsrStorePage)
2795	{
2796	AssertMsgReturnVoid(pGuestMsrLoad->u32Msr == pGuestMsrStore->u32Msr,
2797	("GuestMsrLoad=%#RX32 GuestMsrStore=%#RX32 cMsrs=%u\n",
2798	pGuestMsrLoad->u32Msr, pGuestMsrStore->u32Msr, cMsrs));
2799	}
2800
2801	AssertMsgReturnVoid(pHostMsrLoad->u32Msr == pGuestMsrLoad->u32Msr,
2802	("HostMsrLoad=%#RX32 GuestMsrLoad=%#RX32 cMsrs=%u\n",
2803	pHostMsrLoad->u32Msr, pGuestMsrLoad->u32Msr, cMsrs));
2804
2805	uint64_t const u64HostMsr = ASMRdMsr(pHostMsrLoad->u32Msr);
2806	AssertMsgReturnVoid(pHostMsrLoad->u64Value == u64HostMsr,
2807	("u32Msr=%#RX32 VMCS Value=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
2808	pHostMsrLoad->u32Msr, pHostMsrLoad->u64Value, u64HostMsr, cMsrs));
2809
2810	/* Verify that cached host EFER MSR matches what's loaded on the CPU. */
2811	bool const fIsEferMsr = RT_BOOL(pHostMsrLoad->u32Msr == MSR_K6_EFER);
2812	AssertMsgReturnVoid(!fIsEferMsr \|\| u64HostMsr == g_uHmVmxHostMsrEfer,
2813	("Cached=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n", g_uHmVmxHostMsrEfer, u64HostMsr, cMsrs));
2814
2815	/* Verify that the accesses are as expected in the MSR bitmap for auto-load/store MSRs. */
2816	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
2817	{
2818	uint32_t const fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, pGuestMsrLoad->u32Msr);
2819	if (fIsEferMsr)
2820	{
2821	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_EXIT_RD), ("Passthru read for EFER MSR!?\n"));
2822	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_EXIT_WR), ("Passthru write for EFER MSR!?\n"));
2823	}
2824	else
2825	{
2826	/* Verify LBR MSRs (used only for debugging) are intercepted. We don't passthru these MSRs to the guest yet. */
2827	PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
2828	if ( pVM->hmr0.s.vmx.fLbr
2829	&& ( hmR0VmxIsLbrBranchFromMsr(pVM, pGuestMsrLoad->u32Msr, NULL /* pidxMsr */)
2830	\|\| hmR0VmxIsLbrBranchToMsr(pVM, pGuestMsrLoad->u32Msr, NULL /* pidxMsr */)
2831	\|\| pGuestMsrLoad->u32Msr == pVM->hmr0.s.vmx.idLbrTosMsr))
2832	{
2833	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_MASK) == VMXMSRPM_EXIT_RD_WR,
2834	("u32Msr=%#RX32 cMsrs=%u Passthru read/write for LBR MSRs!\n",
2835	pGuestMsrLoad->u32Msr, cMsrs));
2836	}
2837	else if (!fIsNstGstVmcs)
2838	{
2839	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_MASK) == VMXMSRPM_ALLOW_RD_WR,
2840	("u32Msr=%#RX32 cMsrs=%u No passthru read/write!\n", pGuestMsrLoad->u32Msr, cMsrs));
2841	}
2842	else
2843	{
2844	/*
2845	* A nested-guest VMCS must -also- allow read/write passthrough for the MSR for us to
2846	* execute a nested-guest with MSR passthrough.
2847	*
2848	* Check if the nested-guest MSR bitmap allows passthrough, and if so, assert that we
2849	* allow passthrough too.
2850	*/
2851	void const *pvMsrBitmapNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap;
2852	Assert(pvMsrBitmapNstGst);
2853	uint32_t const fMsrpmNstGst = CPUMGetVmxMsrPermission(pvMsrBitmapNstGst, pGuestMsrLoad->u32Msr);
2854	AssertMsgReturnVoid(fMsrpm == fMsrpmNstGst,
2855	("u32Msr=%#RX32 cMsrs=%u Permission mismatch fMsrpm=%#x fMsrpmNstGst=%#x!\n",
2856	pGuestMsrLoad->u32Msr, cMsrs, fMsrpm, fMsrpmNstGst));
2857	}
2858	}
2859	}
2860
2861	/* Move to the next MSR. */
2862	pHostMsrLoad++;
2863	pGuestMsrLoad++;
2864	pGuestMsrStore++;
2865	}
2866	}
2867
2868	#endif /* VBOX_STRICT */
2869
2870	/**
2871	* Flushes the TLB using EPT.
2872	*
2873	* @returns VBox status code.
2874	* @param pVCpu The cross context virtual CPU structure of the calling
2875	* EMT. Can be NULL depending on @a enmTlbFlush.
2876	* @param pVmcsInfo The VMCS info. object. Can be NULL depending on @a
2877	* enmTlbFlush.
2878	* @param enmTlbFlush Type of flush.
2879	*
2880	* @remarks Caller is responsible for making sure this function is called only
2881	* when NestedPaging is supported and providing @a enmTlbFlush that is
2882	* supported by the CPU.
2883	* @remarks Can be called with interrupts disabled.
2884	*/
2885	static void hmR0VmxFlushEpt(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, VMXTLBFLUSHEPT enmTlbFlush)
2886	{
2887	uint64_t au64Descriptor[2];
2888	if (enmTlbFlush == VMXTLBFLUSHEPT_ALL_CONTEXTS)
2889	au64Descriptor[0] = 0;
2890	else
2891	{
2892	Assert(pVCpu);
2893	Assert(pVmcsInfo);
2894	au64Descriptor[0] = pVmcsInfo->HCPhysEPTP;
2895	}
2896	au64Descriptor[1] = 0; /* MBZ. Intel spec. 33.3 "VMX Instructions" */
2897
2898	int rc = VMXR0InvEPT(enmTlbFlush, &au64Descriptor[0]);
2899	AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvEPT %#x %#RHp failed. rc=%Rrc\n", enmTlbFlush, au64Descriptor[0], rc));
2900
2901	if ( RT_SUCCESS(rc)
2902	&& pVCpu)
2903	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushNestedPaging);
2904	}
2905
2906
2907	/**
2908	* Flushes the TLB using VPID.
2909	*
2910	* @returns VBox status code.
2911	* @param pVCpu The cross context virtual CPU structure of the calling
2912	* EMT. Can be NULL depending on @a enmTlbFlush.
2913	* @param enmTlbFlush Type of flush.
2914	* @param GCPtr Virtual address of the page to flush (can be 0 depending
2915	* on @a enmTlbFlush).
2916	*
2917	* @remarks Can be called with interrupts disabled.
2918	*/
2919	static void hmR0VmxFlushVpid(PVMCPUCC pVCpu, VMXTLBFLUSHVPID enmTlbFlush, RTGCPTR GCPtr)
2920	{
2921	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fVpid);
2922
2923	uint64_t au64Descriptor[2];
2924	if (enmTlbFlush == VMXTLBFLUSHVPID_ALL_CONTEXTS)
2925	{
2926	au64Descriptor[0] = 0;
2927	au64Descriptor[1] = 0;
2928	}
2929	else
2930	{
2931	AssertPtr(pVCpu);
2932	AssertMsg(pVCpu->hmr0.s.uCurrentAsid != 0, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hmr0.s.uCurrentAsid));
2933	AssertMsg(pVCpu->hmr0.s.uCurrentAsid <= UINT16_MAX, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hmr0.s.uCurrentAsid));
2934	au64Descriptor[0] = pVCpu->hmr0.s.uCurrentAsid;
2935	au64Descriptor[1] = GCPtr;
2936	}
2937
2938	int rc = VMXR0InvVPID(enmTlbFlush, &au64Descriptor[0]);
2939	AssertMsg(rc == VINF_SUCCESS,
2940	("VMXR0InvVPID %#x %u %RGv failed with %Rrc\n", enmTlbFlush, pVCpu ? pVCpu->hmr0.s.uCurrentAsid : 0, GCPtr, rc));
2941
2942	if ( RT_SUCCESS(rc)
2943	&& pVCpu)
2944	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid);
2945	NOREF(rc);
2946	}
2947
2948
2949	/**
2950	* Invalidates a guest page by guest virtual address. Only relevant for EPT/VPID,
2951	* otherwise there is nothing really to invalidate.
2952	*
2953	* @returns VBox status code.
2954	* @param pVCpu The cross context virtual CPU structure.
2955	* @param GCVirt Guest virtual address of the page to invalidate.
2956	*/
2957	VMMR0DECL(int) VMXR0InvalidatePage(PVMCPUCC pVCpu, RTGCPTR GCVirt)
2958	{
2959	AssertPtr(pVCpu);
2960	LogFlowFunc(("pVCpu=%p GCVirt=%RGv\n", pVCpu, GCVirt));
2961
2962	if (!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TLB_FLUSH))
2963	{
2964	/*
2965	* We must invalidate the guest TLB entry in either case, we cannot ignore it even for
2966	* the EPT case. See @bugref{6043} and @bugref{6177}.
2967	*
2968	* Set the VMCPU_FF_TLB_FLUSH force flag and flush before VM-entry in hmR0VmxFlushTLB*()
2969	* as this function maybe called in a loop with individual addresses.
2970	*/
2971	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2972	if (pVM->hmr0.s.vmx.fVpid)
2973	{
2974	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
2975	{
2976	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_INDIV_ADDR, GCVirt);
2977	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt);
2978	}
2979	else
2980	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2981	}
2982	else if (pVM->hmr0.s.fNestedPaging)
2983	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2984	}
2985
2986	return VINF_SUCCESS;
2987	}
2988
2989
2990	/**
2991	* Dummy placeholder for tagged-TLB flush handling before VM-entry. Used in the
2992	* case where neither EPT nor VPID is supported by the CPU.
2993	*
2994	* @param pHostCpu The HM physical-CPU structure.
2995	* @param pVCpu The cross context virtual CPU structure.
2996	*
2997	* @remarks Called with interrupts disabled.
2998	*/
2999	static void hmR0VmxFlushTaggedTlbNone(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu)
3000	{
3001	AssertPtr(pVCpu);
3002	AssertPtr(pHostCpu);
3003
3004	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
3005
3006	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3007	pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
3008	pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3009	pVCpu->hmr0.s.fForceTLBFlush = false;
3010	return;
3011	}
3012
3013
3014	/**
3015	* Flushes the tagged-TLB entries for EPT+VPID CPUs as necessary.
3016	*
3017	* @param pHostCpu The HM physical-CPU structure.
3018	* @param pVCpu The cross context virtual CPU structure.
3019	* @param pVmcsInfo The VMCS info. object.
3020	*
3021	* @remarks All references to "ASID" in this function pertains to "VPID" in Intel's
3022	* nomenclature. The reason is, to avoid confusion in compare statements
3023	* since the host-CPU copies are named "ASID".
3024	*
3025	* @remarks Called with interrupts disabled.
3026	*/
3027	static void hmR0VmxFlushTaggedTlbBoth(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
3028	{
3029	#ifdef VBOX_WITH_STATISTICS
3030	bool fTlbFlushed = false;
3031	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { fTlbFlushed = true; } while (0)
3032	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { \
3033	if (!fTlbFlushed) \
3034	STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); \
3035	} while (0)
3036	#else
3037	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { } while (0)
3038	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { } while (0)
3039	#endif
3040
3041	AssertPtr(pVCpu);
3042	AssertPtr(pHostCpu);
3043	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3044
3045	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3046	AssertMsg(pVM->hmr0.s.fNestedPaging && pVM->hmr0.s.vmx.fVpid,
3047	("hmR0VmxFlushTaggedTlbBoth cannot be invoked unless NestedPaging & VPID are enabled."
3048	"fNestedPaging=%RTbool fVpid=%RTbool", pVM->hmr0.s.fNestedPaging, pVM->hmr0.s.vmx.fVpid));
3049
3050	/*
3051	* Force a TLB flush for the first world-switch if the current CPU differs from the one we
3052	* ran on last. If the TLB flush count changed, another VM (VCPU rather) has hit the ASID
3053	* limit while flushing the TLB or the host CPU is online after a suspend/resume, so we
3054	* cannot reuse the current ASID anymore.
3055	*/
3056	if ( pVCpu->hmr0.s.idLastCpu != pHostCpu->idCpu
3057	\|\| pVCpu->hmr0.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3058	{
3059	++pHostCpu->uCurrentAsid;
3060	if (pHostCpu->uCurrentAsid >= g_uHmMaxAsid)
3061	{
3062	pHostCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0. */
3063	pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
3064	pHostCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
3065	}
3066
3067	pVCpu->hmr0.s.uCurrentAsid = pHostCpu->uCurrentAsid;
3068	pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
3069	pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3070
3071	/*
3072	* Flush by EPT when we get rescheduled to a new host CPU to ensure EPT-only tagged mappings are also
3073	* invalidated. We don't need to flush-by-VPID here as flushing by EPT covers it. See @bugref{6568}.
3074	*/
3075	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hmr0.s.vmx.enmTlbFlushEpt);
3076	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3077	HMVMX_SET_TAGGED_TLB_FLUSHED();
3078	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
3079	}
3080	else if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH)) /* Check for explicit TLB flushes. */
3081	{
3082	/*
3083	* Changes to the EPT paging structure by VMM requires flushing-by-EPT as the CPU
3084	* creates guest-physical (ie. only EPT-tagged) mappings while traversing the EPT
3085	* tables when EPT is in use. Flushing-by-VPID will only flush linear (only
3086	* VPID-tagged) and combined (EPT+VPID tagged) mappings but not guest-physical
3087	* mappings, see @bugref{6568}.
3088	*
3089	* See Intel spec. 28.3.2 "Creating and Using Cached Translation Information".
3090	*/
3091	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hmr0.s.vmx.enmTlbFlushEpt);
3092	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3093	HMVMX_SET_TAGGED_TLB_FLUSHED();
3094	}
3095	else if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3096	{
3097	/*
3098	* The nested-guest specifies its own guest-physical address to use as the APIC-access
3099	* address which requires flushing the TLB of EPT cached structures.
3100	*
3101	* See Intel spec. 28.3.3.4 "Guidelines for Use of the INVEPT Instruction".
3102	*/
3103	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hmr0.s.vmx.enmTlbFlushEpt);
3104	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3105	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3106	HMVMX_SET_TAGGED_TLB_FLUSHED();
3107	}
3108
3109
3110	pVCpu->hmr0.s.fForceTLBFlush = false;
3111	HMVMX_UPDATE_FLUSH_SKIPPED_STAT();
3112
3113	Assert(pVCpu->hmr0.s.idLastCpu == pHostCpu->idCpu);
3114	Assert(pVCpu->hmr0.s.cTlbFlushes == pHostCpu->cTlbFlushes);
3115	AssertMsg(pVCpu->hmr0.s.cTlbFlushes == pHostCpu->cTlbFlushes,
3116	("Flush count mismatch for cpu %d (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hmr0.s.cTlbFlushes, pHostCpu->cTlbFlushes));
3117	AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < g_uHmMaxAsid,
3118	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pHostCpu->idCpu,
3119	pHostCpu->uCurrentAsid, pHostCpu->cTlbFlushes, pVCpu->hmr0.s.idLastCpu, pVCpu->hmr0.s.cTlbFlushes));
3120	AssertMsg(pVCpu->hmr0.s.uCurrentAsid >= 1 && pVCpu->hmr0.s.uCurrentAsid < g_uHmMaxAsid,
3121	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pHostCpu->idCpu, pVCpu->hmr0.s.uCurrentAsid));
3122
3123	/* Update VMCS with the VPID. */
3124	int rc = VMXWriteVmcs16(VMX_VMCS16_VPID, pVCpu->hmr0.s.uCurrentAsid);
3125	AssertRC(rc);
3126
3127	#undef HMVMX_SET_TAGGED_TLB_FLUSHED
3128	}
3129
3130
3131	/**
3132	* Flushes the tagged-TLB entries for EPT CPUs as necessary.
3133	*
3134	* @param pHostCpu The HM physical-CPU structure.
3135	* @param pVCpu The cross context virtual CPU structure.
3136	* @param pVmcsInfo The VMCS info. object.
3137	*
3138	* @remarks Called with interrupts disabled.
3139	*/
3140	static void hmR0VmxFlushTaggedTlbEpt(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
3141	{
3142	AssertPtr(pVCpu);
3143	AssertPtr(pHostCpu);
3144	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3145	AssertMsg(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked without NestedPaging."));
3146	AssertMsg(!pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fVpid, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with VPID."));
3147
3148	/*
3149	* Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
3150	* A change in the TLB flush count implies the host CPU is online after a suspend/resume.
3151	*/
3152	if ( pVCpu->hmr0.s.idLastCpu != pHostCpu->idCpu
3153	\|\| pVCpu->hmr0.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3154	{
3155	pVCpu->hmr0.s.fForceTLBFlush = true;
3156	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3157	}
3158
3159	/* Check for explicit TLB flushes. */
3160	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
3161	{
3162	pVCpu->hmr0.s.fForceTLBFlush = true;
3163	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3164	}
3165
3166	/* Check for TLB flushes while switching to/from a nested-guest. */
3167	if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3168	{
3169	pVCpu->hmr0.s.fForceTLBFlush = true;
3170	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3171	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3172	}
3173
3174	pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
3175	pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3176
3177	if (pVCpu->hmr0.s.fForceTLBFlush)
3178	{
3179	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.enmTlbFlushEpt);
3180	pVCpu->hmr0.s.fForceTLBFlush = false;
3181	}
3182	}
3183
3184
3185	/**
3186	* Flushes the tagged-TLB entries for VPID CPUs as necessary.
3187	*
3188	* @param pHostCpu The HM physical-CPU structure.
3189	* @param pVCpu The cross context virtual CPU structure.
3190	*
3191	* @remarks Called with interrupts disabled.
3192	*/
3193	static void hmR0VmxFlushTaggedTlbVpid(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu)
3194	{
3195	AssertPtr(pVCpu);
3196	AssertPtr(pHostCpu);
3197	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3198	AssertMsg(pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fVpid, ("hmR0VmxFlushTlbVpid cannot be invoked without VPID."));
3199	AssertMsg(!pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging, ("hmR0VmxFlushTlbVpid cannot be invoked with NestedPaging"));
3200
3201	/*
3202	* Force a TLB flush for the first world switch if the current CPU differs from the one we
3203	* ran on last. If the TLB flush count changed, another VM (VCPU rather) has hit the ASID
3204	* limit while flushing the TLB or the host CPU is online after a suspend/resume, so we
3205	* cannot reuse the current ASID anymore.
3206	*/
3207	if ( pVCpu->hmr0.s.idLastCpu != pHostCpu->idCpu
3208	\|\| pVCpu->hmr0.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3209	{
3210	pVCpu->hmr0.s.fForceTLBFlush = true;
3211	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3212	}
3213
3214	/* Check for explicit TLB flushes. */
3215	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
3216	{
3217	/*
3218	* If we ever support VPID flush combinations other than ALL or SINGLE-context (see
3219	* hmR0VmxSetupTaggedTlb()) we would need to explicitly flush in this case (add an
3220	* fExplicitFlush = true here and change the pHostCpu->fFlushAsidBeforeUse check below to
3221	* include fExplicitFlush's too) - an obscure corner case.
3222	*/
3223	pVCpu->hmr0.s.fForceTLBFlush = true;
3224	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3225	}
3226
3227	/* Check for TLB flushes while switching to/from a nested-guest. */
3228	if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3229	{
3230	pVCpu->hmr0.s.fForceTLBFlush = true;
3231	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3232	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3233	}
3234
3235	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3236	pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
3237	if (pVCpu->hmr0.s.fForceTLBFlush)
3238	{
3239	++pHostCpu->uCurrentAsid;
3240	if (pHostCpu->uCurrentAsid >= g_uHmMaxAsid)
3241	{
3242	pHostCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0 */
3243	pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
3244	pHostCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
3245	}
3246
3247	pVCpu->hmr0.s.fForceTLBFlush = false;
3248	pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3249	pVCpu->hmr0.s.uCurrentAsid = pHostCpu->uCurrentAsid;
3250	if (pHostCpu->fFlushAsidBeforeUse)
3251	{
3252	if (pVM->hmr0.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_SINGLE_CONTEXT)
3253	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_SINGLE_CONTEXT, 0 /* GCPtr */);
3254	else if (pVM->hmr0.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_ALL_CONTEXTS)
3255	{
3256	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_ALL_CONTEXTS, 0 /* GCPtr */);
3257	pHostCpu->fFlushAsidBeforeUse = false;
3258	}
3259	else
3260	{
3261	/* hmR0VmxSetupTaggedTlb() ensures we never get here. Paranoia. */
3262	AssertMsgFailed(("Unsupported VPID-flush context type.\n"));
3263	}
3264	}
3265	}
3266
3267	AssertMsg(pVCpu->hmr0.s.cTlbFlushes == pHostCpu->cTlbFlushes,
3268	("Flush count mismatch for cpu %d (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hmr0.s.cTlbFlushes, pHostCpu->cTlbFlushes));
3269	AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < g_uHmMaxAsid,
3270	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pHostCpu->idCpu,
3271	pHostCpu->uCurrentAsid, pHostCpu->cTlbFlushes, pVCpu->hmr0.s.idLastCpu, pVCpu->hmr0.s.cTlbFlushes));
3272	AssertMsg(pVCpu->hmr0.s.uCurrentAsid >= 1 && pVCpu->hmr0.s.uCurrentAsid < g_uHmMaxAsid,
3273	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pHostCpu->idCpu, pVCpu->hmr0.s.uCurrentAsid));
3274
3275	int rc = VMXWriteVmcs16(VMX_VMCS16_VPID, pVCpu->hmr0.s.uCurrentAsid);
3276	AssertRC(rc);
3277	}
3278
3279
3280	/**
3281	* Flushes the guest TLB entry based on CPU capabilities.
3282	*
3283	* @param pHostCpu The HM physical-CPU structure.
3284	* @param pVCpu The cross context virtual CPU structure.
3285	* @param pVmcsInfo The VMCS info. object.
3286	*
3287	* @remarks Called with interrupts disabled.
3288	*/
3289	static void hmR0VmxFlushTaggedTlb(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3290	{
3291	#ifdef HMVMX_ALWAYS_FLUSH_TLB
3292	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
3293	#endif
3294	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3295	switch (pVM->hmr0.s.vmx.enmTlbFlushType)
3296	{
3297	case VMXTLBFLUSHTYPE_EPT_VPID: hmR0VmxFlushTaggedTlbBoth(pHostCpu, pVCpu, pVmcsInfo); break;
3298	case VMXTLBFLUSHTYPE_EPT: hmR0VmxFlushTaggedTlbEpt(pHostCpu, pVCpu, pVmcsInfo); break;
3299	case VMXTLBFLUSHTYPE_VPID: hmR0VmxFlushTaggedTlbVpid(pHostCpu, pVCpu); break;
3300	case VMXTLBFLUSHTYPE_NONE: hmR0VmxFlushTaggedTlbNone(pHostCpu, pVCpu); break;
3301	default:
3302	AssertMsgFailed(("Invalid flush-tag function identifier\n"));
3303	break;
3304	}
3305	/* Don't assert that VMCPU_FF_TLB_FLUSH should no longer be pending. It can be set by other EMTs. */
3306	}
3307
3308
3309	/**
3310	* Sets up the appropriate tagged TLB-flush level and handler for flushing guest
3311	* TLB entries from the host TLB before VM-entry.
3312	*
3313	* @returns VBox status code.
3314	* @param pVM The cross context VM structure.
3315	*/
3316	static int hmR0VmxSetupTaggedTlb(PVMCC pVM)
3317	{
3318	/*
3319	* Determine optimal flush type for nested paging.
3320	* We cannot ignore EPT if no suitable flush-types is supported by the CPU as we've already setup
3321	* unrestricted guest execution (see hmR3InitFinalizeR0()).
3322	*/
3323	if (pVM->hmr0.s.fNestedPaging)
3324	{
3325	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT)
3326	{
3327	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT)
3328	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_SINGLE_CONTEXT;
3329	else if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
3330	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_ALL_CONTEXTS;
3331	else
3332	{
3333	/* Shouldn't happen. EPT is supported but no suitable flush-types supported. */
3334	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3335	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_FLUSH_TYPE_UNSUPPORTED;
3336	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3337	}
3338
3339	/* Make sure the write-back cacheable memory type for EPT is supported. */
3340	if (RT_UNLIKELY(!(g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_MEMTYPE_WB)))
3341	{
3342	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3343	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_MEM_TYPE_NOT_WB;
3344	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3345	}
3346
3347	/* EPT requires a page-walk length of 4. */
3348	if (RT_UNLIKELY(!(g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_PAGE_WALK_LENGTH_4)))
3349	{
3350	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3351	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_PAGE_WALK_LENGTH_UNSUPPORTED;
3352	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3353	}
3354	}
3355	else
3356	{
3357	/* Shouldn't happen. EPT is supported but INVEPT instruction is not supported. */
3358	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3359	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_INVEPT_UNAVAILABLE;
3360	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3361	}
3362	}
3363
3364	/*
3365	* Determine optimal flush type for VPID.
3366	*/
3367	if (pVM->hmr0.s.vmx.fVpid)
3368	{
3369	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID)
3370	{
3371	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT)
3372	pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_SINGLE_CONTEXT;
3373	else if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS)
3374	pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_ALL_CONTEXTS;
3375	else
3376	{
3377	/* Neither SINGLE nor ALL-context flush types for VPID is supported by the CPU. Ignore VPID capability. */
3378	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
3379	LogRelFunc(("Only INDIV_ADDR supported. Ignoring VPID.\n"));
3380	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
3381	LogRelFunc(("Only SINGLE_CONTEXT_RETAIN_GLOBALS supported. Ignoring VPID.\n"));
3382	pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NOT_SUPPORTED;
3383	pVM->hmr0.s.vmx.fVpid = false;
3384	}
3385	}
3386	else
3387	{
3388	/* Shouldn't happen. VPID is supported but INVVPID is not supported by the CPU. Ignore VPID capability. */
3389	Log4Func(("VPID supported without INVEPT support. Ignoring VPID.\n"));
3390	pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NOT_SUPPORTED;
3391	pVM->hmr0.s.vmx.fVpid = false;
3392	}
3393	}
3394
3395	/*
3396	* Setup the handler for flushing tagged-TLBs.
3397	*/
3398	if (pVM->hmr0.s.fNestedPaging && pVM->hmr0.s.vmx.fVpid)
3399	pVM->hmr0.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_EPT_VPID;
3400	else if (pVM->hmr0.s.fNestedPaging)
3401	pVM->hmr0.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_EPT;
3402	else if (pVM->hmr0.s.vmx.fVpid)
3403	pVM->hmr0.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_VPID;
3404	else
3405	pVM->hmr0.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_NONE;
3406
3407
3408	/*
3409	* Copy out the result to ring-3.
3410	*/
3411	pVM->hm.s.ForR3.vmx.fVpid = pVM->hmr0.s.vmx.fVpid;
3412	pVM->hm.s.ForR3.vmx.enmTlbFlushType = pVM->hmr0.s.vmx.enmTlbFlushType;
3413	pVM->hm.s.ForR3.vmx.enmTlbFlushEpt = pVM->hmr0.s.vmx.enmTlbFlushEpt;
3414	pVM->hm.s.ForR3.vmx.enmTlbFlushVpid = pVM->hmr0.s.vmx.enmTlbFlushVpid;
3415	return VINF_SUCCESS;
3416	}
3417
3418
3419	/**
3420	* Sets up the LBR MSR ranges based on the host CPU.
3421	*
3422	* @returns VBox status code.
3423	* @param pVM The cross context VM structure.
3424	*/
3425	static int hmR0VmxSetupLbrMsrRange(PVMCC pVM)
3426	{
3427	Assert(pVM->hmr0.s.vmx.fLbr);
3428	uint32_t idLbrFromIpMsrFirst;
3429	uint32_t idLbrFromIpMsrLast;
3430	uint32_t idLbrToIpMsrFirst;
3431	uint32_t idLbrToIpMsrLast;
3432	uint32_t idLbrTosMsr;
3433
3434	/*
3435	* Determine the LBR MSRs supported for this host CPU family and model.
3436	*
3437	* See Intel spec. 17.4.8 "LBR Stack".
3438	* See Intel "Model-Specific Registers" spec.
3439	*/
3440	uint32_t const uFamilyModel = (pVM->cpum.ro.HostFeatures.uFamily << 8)
3441	\| pVM->cpum.ro.HostFeatures.uModel;
3442	switch (uFamilyModel)
3443	{
3444	case 0x0f01: case 0x0f02:
3445	idLbrFromIpMsrFirst = MSR_P4_LASTBRANCH_0;
3446	idLbrFromIpMsrLast = MSR_P4_LASTBRANCH_3;
3447	idLbrToIpMsrFirst = 0x0;
3448	idLbrToIpMsrLast = 0x0;
3449	idLbrTosMsr = MSR_P4_LASTBRANCH_TOS;
3450	break;
3451
3452	case 0x065c: case 0x065f: case 0x064e: case 0x065e: case 0x068e:
3453	case 0x069e: case 0x0655: case 0x0666: case 0x067a: case 0x0667:
3454	case 0x066a: case 0x066c: case 0x067d: case 0x067e:
3455	idLbrFromIpMsrFirst = MSR_LASTBRANCH_0_FROM_IP;
3456	idLbrFromIpMsrLast = MSR_LASTBRANCH_31_FROM_IP;
3457	idLbrToIpMsrFirst = MSR_LASTBRANCH_0_TO_IP;
3458	idLbrToIpMsrLast = MSR_LASTBRANCH_31_TO_IP;
3459	idLbrTosMsr = MSR_LASTBRANCH_TOS;
3460	break;
3461
3462	case 0x063d: case 0x0647: case 0x064f: case 0x0656: case 0x063c:
3463	case 0x0645: case 0x0646: case 0x063f: case 0x062a: case 0x062d:
3464	case 0x063a: case 0x063e: case 0x061a: case 0x061e: case 0x061f:
3465	case 0x062e: case 0x0625: case 0x062c: case 0x062f:
3466	idLbrFromIpMsrFirst = MSR_LASTBRANCH_0_FROM_IP;
3467	idLbrFromIpMsrLast = MSR_LASTBRANCH_15_FROM_IP;
3468	idLbrToIpMsrFirst = MSR_LASTBRANCH_0_TO_IP;
3469	idLbrToIpMsrLast = MSR_LASTBRANCH_15_TO_IP;
3470	idLbrTosMsr = MSR_LASTBRANCH_TOS;
3471	break;
3472
3473	case 0x0617: case 0x061d: case 0x060f:
3474	idLbrFromIpMsrFirst = MSR_CORE2_LASTBRANCH_0_FROM_IP;
3475	idLbrFromIpMsrLast = MSR_CORE2_LASTBRANCH_3_FROM_IP;
3476	idLbrToIpMsrFirst = MSR_CORE2_LASTBRANCH_0_TO_IP;
3477	idLbrToIpMsrLast = MSR_CORE2_LASTBRANCH_3_TO_IP;
3478	idLbrTosMsr = MSR_CORE2_LASTBRANCH_TOS;
3479	break;
3480
3481	/* Atom and related microarchitectures we don't care about:
3482	case 0x0637: case 0x064a: case 0x064c: case 0x064d: case 0x065a:
3483	case 0x065d: case 0x061c: case 0x0626: case 0x0627: case 0x0635:
3484	case 0x0636: */
3485	/* All other CPUs: */
3486	default:
3487	{
3488	LogRelFunc(("Could not determine LBR stack size for the CPU model %#x\n", uFamilyModel));
3489	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_LBR_STACK_SIZE_UNKNOWN;
3490	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3491	}
3492	}
3493
3494	/*
3495	* Validate.
3496	*/
3497	uint32_t const cLbrStack = idLbrFromIpMsrLast - idLbrFromIpMsrFirst + 1;
3498	PCVMCPU pVCpu0 = VMCC_GET_CPU_0(pVM);
3499	AssertCompile( RT_ELEMENTS(pVCpu0->hm.s.vmx.VmcsInfo.au64LbrFromIpMsr)
3500	== RT_ELEMENTS(pVCpu0->hm.s.vmx.VmcsInfo.au64LbrToIpMsr));
3501	if (cLbrStack > RT_ELEMENTS(pVCpu0->hm.s.vmx.VmcsInfo.au64LbrFromIpMsr))
3502	{
3503	LogRelFunc(("LBR stack size of the CPU (%u) exceeds our buffer size\n", cLbrStack));
3504	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_LBR_STACK_SIZE_OVERFLOW;
3505	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3506	}
3507	NOREF(pVCpu0);
3508
3509	/*
3510	* Update the LBR info. to the VM struct. for use later.
3511	*/
3512	pVM->hmr0.s.vmx.idLbrTosMsr = idLbrTosMsr;
3513
3514	pVM->hm.s.ForR3.vmx.idLbrFromIpMsrFirst = pVM->hmr0.s.vmx.idLbrFromIpMsrFirst = idLbrFromIpMsrFirst;
3515	pVM->hm.s.ForR3.vmx.idLbrFromIpMsrLast = pVM->hmr0.s.vmx.idLbrFromIpMsrLast = idLbrFromIpMsrLast;
3516
3517	pVM->hm.s.ForR3.vmx.idLbrToIpMsrFirst = pVM->hmr0.s.vmx.idLbrToIpMsrFirst = idLbrToIpMsrFirst;
3518	pVM->hm.s.ForR3.vmx.idLbrToIpMsrLast = pVM->hmr0.s.vmx.idLbrToIpMsrLast = idLbrToIpMsrLast;
3519	return VINF_SUCCESS;
3520	}
3521
3522
3523	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3524	/**
3525	* Sets up the shadow VMCS fields arrays.
3526	*
3527	* This function builds arrays of VMCS fields to sync the shadow VMCS later while
3528	* executing the guest.
3529	*
3530	* @returns VBox status code.
3531	* @param pVM The cross context VM structure.
3532	*/
3533	static int hmR0VmxSetupShadowVmcsFieldsArrays(PVMCC pVM)
3534	{
3535	/*
3536	* Paranoia. Ensure we haven't exposed the VMWRITE-All VMX feature to the guest
3537	* when the host does not support it.
3538	*/
3539	bool const fGstVmwriteAll = pVM->cpum.ro.GuestFeatures.fVmxVmwriteAll;
3540	if ( !fGstVmwriteAll
3541	\|\| (g_HmMsrs.u.vmx.u64Misc & VMX_MISC_VMWRITE_ALL))
3542	{ /* likely. */ }
3543	else
3544	{
3545	LogRelFunc(("VMX VMWRITE-All feature exposed to the guest but host CPU does not support it!\n"));
3546	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_GST_HOST_VMWRITE_ALL;
3547	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3548	}
3549
3550	uint32_t const cVmcsFields = RT_ELEMENTS(g_aVmcsFields);
3551	uint32_t cRwFields = 0;
3552	uint32_t cRoFields = 0;
3553	for (uint32_t i = 0; i < cVmcsFields; i++)
3554	{
3555	VMXVMCSFIELD VmcsField;
3556	VmcsField.u = g_aVmcsFields[i];
3557
3558	/*
3559	* We will be writing "FULL" (64-bit) fields while syncing the shadow VMCS.
3560	* Therefore, "HIGH" (32-bit portion of 64-bit) fields must not be included
3561	* in the shadow VMCS fields array as they would be redundant.
3562	*
3563	* If the VMCS field depends on a CPU feature that is not exposed to the guest,
3564	* we must not include it in the shadow VMCS fields array. Guests attempting to
3565	* VMREAD/VMWRITE such VMCS fields would cause a VM-exit and we shall emulate
3566	* the required behavior.
3567	*/
3568	if ( VmcsField.n.fAccessType == VMX_VMCSFIELD_ACCESS_FULL
3569	&& CPUMIsGuestVmxVmcsFieldValid(pVM, VmcsField.u))
3570	{
3571	/*
3572	* Read-only fields are placed in a separate array so that while syncing shadow
3573	* VMCS fields later (which is more performance critical) we can avoid branches.
3574	*
3575	* However, if the guest can write to all fields (including read-only fields),
3576	* we treat it a as read/write field. Otherwise, writing to these fields would
3577	* cause a VMWRITE instruction error while syncing the shadow VMCS.
3578	*/
3579	if ( fGstVmwriteAll
3580	\|\| !VMXIsVmcsFieldReadOnly(VmcsField.u))
3581	pVM->hmr0.s.vmx.paShadowVmcsFields[cRwFields++] = VmcsField.u;
3582	else
3583	pVM->hmr0.s.vmx.paShadowVmcsRoFields[cRoFields++] = VmcsField.u;
3584	}
3585	}
3586
3587	/* Update the counts. */
3588	pVM->hmr0.s.vmx.cShadowVmcsFields = cRwFields;
3589	pVM->hmr0.s.vmx.cShadowVmcsRoFields = cRoFields;
3590	return VINF_SUCCESS;
3591	}
3592
3593
3594	/**
3595	* Sets up the VMREAD and VMWRITE bitmaps.
3596	*
3597	* @param pVM The cross context VM structure.
3598	*/
3599	static void hmR0VmxSetupVmreadVmwriteBitmaps(PVMCC pVM)
3600	{
3601	/*
3602	* By default, ensure guest attempts to access any VMCS fields cause VM-exits.
3603	*/
3604	uint32_t const cbBitmap = X86_PAGE_4K_SIZE;
3605	uint8_t pbVmreadBitmap = (uint8_t )pVM->hmr0.s.vmx.pvVmreadBitmap;
3606	uint8_t pbVmwriteBitmap = (uint8_t )pVM->hmr0.s.vmx.pvVmwriteBitmap;
3607	ASMMemFill32(pbVmreadBitmap, cbBitmap, UINT32_C(0xffffffff));
3608	ASMMemFill32(pbVmwriteBitmap, cbBitmap, UINT32_C(0xffffffff));
3609
3610	/*
3611	* Skip intercepting VMREAD/VMWRITE to guest read/write fields in the
3612	* VMREAD and VMWRITE bitmaps.
3613	*/
3614	{
3615	uint32_t const *paShadowVmcsFields = pVM->hmr0.s.vmx.paShadowVmcsFields;
3616	uint32_t const cShadowVmcsFields = pVM->hmr0.s.vmx.cShadowVmcsFields;
3617	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
3618	{
3619	uint32_t const uVmcsField = paShadowVmcsFields[i];
3620	Assert(!(uVmcsField & VMX_VMCSFIELD_RSVD_MASK));
3621	Assert(uVmcsField >> 3 < cbBitmap);
3622	ASMBitClear(pbVmreadBitmap + (uVmcsField >> 3), uVmcsField & 7);
3623	ASMBitClear(pbVmwriteBitmap + (uVmcsField >> 3), uVmcsField & 7);
3624	}
3625	}
3626
3627	/*
3628	* Skip intercepting VMREAD for guest read-only fields in the VMREAD bitmap
3629	* if the host supports VMWRITE to all supported VMCS fields.
3630	*/
3631	if (g_HmMsrs.u.vmx.u64Misc & VMX_MISC_VMWRITE_ALL)
3632	{
3633	uint32_t const *paShadowVmcsRoFields = pVM->hmr0.s.vmx.paShadowVmcsRoFields;
3634	uint32_t const cShadowVmcsRoFields = pVM->hmr0.s.vmx.cShadowVmcsRoFields;
3635	for (uint32_t i = 0; i < cShadowVmcsRoFields; i++)
3636	{
3637	uint32_t const uVmcsField = paShadowVmcsRoFields[i];
3638	Assert(!(uVmcsField & VMX_VMCSFIELD_RSVD_MASK));
3639	Assert(uVmcsField >> 3 < cbBitmap);
3640	ASMBitClear(pbVmreadBitmap + (uVmcsField >> 3), uVmcsField & 7);
3641	}
3642	}
3643	}
3644	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
3645
3646
3647	/**
3648	* Sets up the virtual-APIC page address for the VMCS.
3649	*
3650	* @param pVmcsInfo The VMCS info. object.
3651	*/
3652	DECLINLINE(void) hmR0VmxSetupVmcsVirtApicAddr(PCVMXVMCSINFO pVmcsInfo)
3653	{
3654	RTHCPHYS const HCPhysVirtApic = pVmcsInfo->HCPhysVirtApic;
3655	Assert(HCPhysVirtApic != NIL_RTHCPHYS);
3656	Assert(!(HCPhysVirtApic & 0xfff)); /* Bits 11:0 MBZ. */
3657	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL, HCPhysVirtApic);
3658	AssertRC(rc);
3659	}
3660
3661
3662	/**
3663	* Sets up the MSR-bitmap address for the VMCS.
3664	*
3665	* @param pVmcsInfo The VMCS info. object.
3666	*/
3667	DECLINLINE(void) hmR0VmxSetupVmcsMsrBitmapAddr(PCVMXVMCSINFO pVmcsInfo)
3668	{
3669	RTHCPHYS const HCPhysMsrBitmap = pVmcsInfo->HCPhysMsrBitmap;
3670	Assert(HCPhysMsrBitmap != NIL_RTHCPHYS);
3671	Assert(!(HCPhysMsrBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3672	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_MSR_BITMAP_FULL, HCPhysMsrBitmap);
3673	AssertRC(rc);
3674	}
3675
3676
3677	/**
3678	* Sets up the APIC-access page address for the VMCS.
3679	*
3680	* @param pVCpu The cross context virtual CPU structure.
3681	*/
3682	DECLINLINE(void) hmR0VmxSetupVmcsApicAccessAddr(PVMCPUCC pVCpu)
3683	{
3684	RTHCPHYS const HCPhysApicAccess = pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.HCPhysApicAccess;
3685	Assert(HCPhysApicAccess != NIL_RTHCPHYS);
3686	Assert(!(HCPhysApicAccess & 0xfff)); /* Bits 11:0 MBZ. */
3687	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, HCPhysApicAccess);
3688	AssertRC(rc);
3689	}
3690
3691	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3692
3693	/**
3694	* Sets up the VMREAD bitmap address for the VMCS.
3695	*
3696	* @param pVCpu The cross context virtual CPU structure.
3697	*/
3698	DECLINLINE(void) hmR0VmxSetupVmcsVmreadBitmapAddr(PVMCPUCC pVCpu)
3699	{
3700	RTHCPHYS const HCPhysVmreadBitmap = pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.HCPhysVmreadBitmap;
3701	Assert(HCPhysVmreadBitmap != NIL_RTHCPHYS);
3702	Assert(!(HCPhysVmreadBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3703	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VMREAD_BITMAP_FULL, HCPhysVmreadBitmap);
3704	AssertRC(rc);
3705	}
3706
3707
3708	/**
3709	* Sets up the VMWRITE bitmap address for the VMCS.
3710	*
3711	* @param pVCpu The cross context virtual CPU structure.
3712	*/
3713	DECLINLINE(void) hmR0VmxSetupVmcsVmwriteBitmapAddr(PVMCPUCC pVCpu)
3714	{
3715	RTHCPHYS const HCPhysVmwriteBitmap = pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.HCPhysVmwriteBitmap;
3716	Assert(HCPhysVmwriteBitmap != NIL_RTHCPHYS);
3717	Assert(!(HCPhysVmwriteBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3718	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VMWRITE_BITMAP_FULL, HCPhysVmwriteBitmap);
3719	AssertRC(rc);
3720	}
3721
3722	#endif
3723
3724	/**
3725	* Sets up the VM-entry MSR load, VM-exit MSR-store and VM-exit MSR-load addresses
3726	* in the VMCS.
3727	*
3728	* @returns VBox status code.
3729	* @param pVmcsInfo The VMCS info. object.
3730	*/
3731	DECLINLINE(int) hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(PVMXVMCSINFO pVmcsInfo)
3732	{
3733	RTHCPHYS const HCPhysGuestMsrLoad = pVmcsInfo->HCPhysGuestMsrLoad;
3734	Assert(HCPhysGuestMsrLoad != NIL_RTHCPHYS);
3735	Assert(!(HCPhysGuestMsrLoad & 0xf)); /* Bits 3:0 MBZ. */
3736
3737	RTHCPHYS const HCPhysGuestMsrStore = pVmcsInfo->HCPhysGuestMsrStore;
3738	Assert(HCPhysGuestMsrStore != NIL_RTHCPHYS);
3739	Assert(!(HCPhysGuestMsrStore & 0xf)); /* Bits 3:0 MBZ. */
3740
3741	RTHCPHYS const HCPhysHostMsrLoad = pVmcsInfo->HCPhysHostMsrLoad;
3742	Assert(HCPhysHostMsrLoad != NIL_RTHCPHYS);
3743	Assert(!(HCPhysHostMsrLoad & 0xf)); /* Bits 3:0 MBZ. */
3744
3745	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL, HCPhysGuestMsrLoad); AssertRC(rc);
3746	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL, HCPhysGuestMsrStore); AssertRC(rc);
3747	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL, HCPhysHostMsrLoad); AssertRC(rc);
3748	return VINF_SUCCESS;
3749	}
3750
3751
3752	/**
3753	* Sets up MSR permissions in the MSR bitmap of a VMCS info. object.
3754	*
3755	* @param pVCpu The cross context virtual CPU structure.
3756	* @param pVmcsInfo The VMCS info. object.
3757	*/
3758	static void hmR0VmxSetupVmcsMsrPermissions(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3759	{
3760	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS);
3761
3762	/*
3763	* By default, ensure guest attempts to access any MSR cause VM-exits.
3764	* This shall later be relaxed for specific MSRs as necessary.
3765	*
3766	* Note: For nested-guests, the entire bitmap will be merged prior to
3767	* executing the nested-guest using hardware-assisted VMX and hence there
3768	* is no need to perform this operation. See hmR0VmxMergeMsrBitmapNested.
3769	*/
3770	Assert(pVmcsInfo->pvMsrBitmap);
3771	ASMMemFill32(pVmcsInfo->pvMsrBitmap, X86_PAGE_4K_SIZE, UINT32_C(0xffffffff));
3772
3773	/*
3774	* The guest can access the following MSRs (read, write) without causing
3775	* VM-exits; they are loaded/stored automatically using fields in the VMCS.
3776	*/
3777	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3778	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_CS, VMXMSRPM_ALLOW_RD_WR);
3779	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_ESP, VMXMSRPM_ALLOW_RD_WR);
3780	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_EIP, VMXMSRPM_ALLOW_RD_WR);
3781	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_GS_BASE, VMXMSRPM_ALLOW_RD_WR);
3782	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_FS_BASE, VMXMSRPM_ALLOW_RD_WR);
3783
3784	/*
3785	* The IA32_PRED_CMD and IA32_FLUSH_CMD MSRs are write-only and has no state
3786	* associated with then. We never need to intercept access (writes need to be
3787	* executed without causing a VM-exit, reads will #GP fault anyway).
3788	*
3789	* The IA32_SPEC_CTRL MSR is read/write and has state. We allow the guest to
3790	* read/write them. We swap the guest/host MSR value using the
3791	* auto-load/store MSR area.
3792	*/
3793	if (pVM->cpum.ro.GuestFeatures.fIbpb)
3794	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_PRED_CMD, VMXMSRPM_ALLOW_RD_WR);
3795	if (pVM->cpum.ro.GuestFeatures.fFlushCmd)
3796	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_FLUSH_CMD, VMXMSRPM_ALLOW_RD_WR);
3797	if (pVM->cpum.ro.GuestFeatures.fIbrs)
3798	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SPEC_CTRL, VMXMSRPM_ALLOW_RD_WR);
3799
3800	/*
3801	* Allow full read/write access for the following MSRs (mandatory for VT-x)
3802	* required for 64-bit guests.
3803	*/
3804	if (pVM->hmr0.s.fAllow64BitGuests)
3805	{
3806	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_LSTAR, VMXMSRPM_ALLOW_RD_WR);
3807	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K6_STAR, VMXMSRPM_ALLOW_RD_WR);
3808	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_SF_MASK, VMXMSRPM_ALLOW_RD_WR);
3809	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_KERNEL_GS_BASE, VMXMSRPM_ALLOW_RD_WR);
3810	}
3811
3812	/*
3813	* IA32_EFER MSR is always intercepted, see @bugref{9180#c37}.
3814	*/
3815	#ifdef VBOX_STRICT
3816	Assert(pVmcsInfo->pvMsrBitmap);
3817	uint32_t const fMsrpmEfer = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, MSR_K6_EFER);
3818	Assert(fMsrpmEfer == VMXMSRPM_EXIT_RD_WR);
3819	#endif
3820	}
3821
3822
3823	/**
3824	* Sets up pin-based VM-execution controls in the VMCS.
3825	*
3826	* @returns VBox status code.
3827	* @param pVCpu The cross context virtual CPU structure.
3828	* @param pVmcsInfo The VMCS info. object.
3829	*/
3830	static int hmR0VmxSetupVmcsPinCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3831	{
3832	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3833	uint32_t fVal = g_HmMsrs.u.vmx.PinCtls.n.allowed0; /* Bits set here must always be set. */
3834	uint32_t const fZap = g_HmMsrs.u.vmx.PinCtls.n.allowed1; /* Bits cleared here must always be cleared. */
3835
3836	fVal \|= VMX_PIN_CTLS_EXT_INT_EXIT /* External interrupts cause a VM-exit. */
3837	\| VMX_PIN_CTLS_NMI_EXIT; /* Non-maskable interrupts (NMIs) cause a VM-exit. */
3838
3839	if (g_HmMsrs.u.vmx.PinCtls.n.allowed1 & VMX_PIN_CTLS_VIRT_NMI)
3840	fVal \|= VMX_PIN_CTLS_VIRT_NMI; /* Use virtual NMIs and virtual-NMI blocking features. */
3841
3842	/* Enable the VMX-preemption timer. */
3843	if (pVM->hmr0.s.vmx.fUsePreemptTimer)
3844	{
3845	Assert(g_HmMsrs.u.vmx.PinCtls.n.allowed1 & VMX_PIN_CTLS_PREEMPT_TIMER);
3846	fVal \|= VMX_PIN_CTLS_PREEMPT_TIMER;
3847	}
3848
3849	#if 0
3850	/* Enable posted-interrupt processing. */
3851	if (pVM->hm.s.fPostedIntrs)
3852	{
3853	Assert(g_HmMsrs.u.vmx.PinCtls.n.allowed1 & VMX_PIN_CTLS_POSTED_INT);
3854	Assert(g_HmMsrs.u.vmx.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_ACK_EXT_INT);
3855	fVal \|= VMX_PIN_CTLS_POSTED_INT;
3856	}
3857	#endif
3858
3859	if ((fVal & fZap) != fVal)
3860	{
3861	LogRelFunc(("Invalid pin-based VM-execution controls combo! Cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3862	g_HmMsrs.u.vmx.PinCtls.n.allowed0, fVal, fZap));
3863	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PIN_EXEC;
3864	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3865	}
3866
3867	/* Commit it to the VMCS and update our cache. */
3868	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, fVal);
3869	AssertRC(rc);
3870	pVmcsInfo->u32PinCtls = fVal;
3871
3872	return VINF_SUCCESS;
3873	}
3874
3875
3876	/**
3877	* Sets up secondary processor-based VM-execution controls in the VMCS.
3878	*
3879	* @returns VBox status code.
3880	* @param pVCpu The cross context virtual CPU structure.
3881	* @param pVmcsInfo The VMCS info. object.
3882	*/
3883	static int hmR0VmxSetupVmcsProcCtls2(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3884	{
3885	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3886	uint32_t fVal = g_HmMsrs.u.vmx.ProcCtls2.n.allowed0; /* Bits set here must be set in the VMCS. */
3887	uint32_t const fZap = g_HmMsrs.u.vmx.ProcCtls2.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3888
3889	/* WBINVD causes a VM-exit. */
3890	if (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_WBINVD_EXIT)
3891	fVal \|= VMX_PROC_CTLS2_WBINVD_EXIT;
3892
3893	/* Enable EPT (aka nested-paging). */
3894	if (pVM->hmr0.s.fNestedPaging)
3895	fVal \|= VMX_PROC_CTLS2_EPT;
3896
3897	/* Enable the INVPCID instruction if we expose it to the guest and is supported
3898	by the hardware. Without this, guest executing INVPCID would cause a #UD. */
3899	if ( pVM->cpum.ro.GuestFeatures.fInvpcid
3900	&& (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_INVPCID))
3901	fVal \|= VMX_PROC_CTLS2_INVPCID;
3902
3903	/* Enable VPID. */
3904	if (pVM->hmr0.s.vmx.fVpid)
3905	fVal \|= VMX_PROC_CTLS2_VPID;
3906
3907	/* Enable unrestricted guest execution. */
3908	if (pVM->hmr0.s.vmx.fUnrestrictedGuest)
3909	fVal \|= VMX_PROC_CTLS2_UNRESTRICTED_GUEST;
3910
3911	#if 0
3912	if (pVM->hm.s.fVirtApicRegs)
3913	{
3914	/* Enable APIC-register virtualization. */
3915	Assert(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_APIC_REG_VIRT);
3916	fVal \|= VMX_PROC_CTLS2_APIC_REG_VIRT;
3917
3918	/* Enable virtual-interrupt delivery. */
3919	Assert(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_INTR_DELIVERY);
3920	fVal \|= VMX_PROC_CTLS2_VIRT_INTR_DELIVERY;
3921	}
3922	#endif
3923
3924	/* Virtualize-APIC accesses if supported by the CPU. The virtual-APIC page is
3925	where the TPR shadow resides. */
3926	/** @todo VIRT_X2APIC support, it's mutually exclusive with this. So must be
3927	* done dynamically. */
3928	if (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
3929	{
3930	fVal \|= VMX_PROC_CTLS2_VIRT_APIC_ACCESS;
3931	hmR0VmxSetupVmcsApicAccessAddr(pVCpu);
3932	}
3933
3934	/* Enable the RDTSCP instruction if we expose it to the guest and is supported
3935	by the hardware. Without this, guest executing RDTSCP would cause a #UD. */
3936	if ( pVM->cpum.ro.GuestFeatures.fRdTscP
3937	&& (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_RDTSCP))
3938	fVal \|= VMX_PROC_CTLS2_RDTSCP;
3939
3940	/* Enable Pause-Loop exiting. */
3941	if ( (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT)
3942	&& pVM->hm.s.vmx.cPleGapTicks
3943	&& pVM->hm.s.vmx.cPleWindowTicks)
3944	{
3945	fVal \|= VMX_PROC_CTLS2_PAUSE_LOOP_EXIT;
3946
3947	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, pVM->hm.s.vmx.cPleGapTicks); AssertRC(rc);
3948	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, pVM->hm.s.vmx.cPleWindowTicks); AssertRC(rc);
3949	}
3950
3951	if ((fVal & fZap) != fVal)
3952	{
3953	LogRelFunc(("Invalid secondary processor-based VM-execution controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3954	g_HmMsrs.u.vmx.ProcCtls2.n.allowed0, fVal, fZap));
3955	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC2;
3956	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3957	}
3958
3959	/* Commit it to the VMCS and update our cache. */
3960	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, fVal);
3961	AssertRC(rc);
3962	pVmcsInfo->u32ProcCtls2 = fVal;
3963
3964	return VINF_SUCCESS;
3965	}
3966
3967
3968	/**
3969	* Sets up processor-based VM-execution controls in the VMCS.
3970	*
3971	* @returns VBox status code.
3972	* @param pVCpu The cross context virtual CPU structure.
3973	* @param pVmcsInfo The VMCS info. object.
3974	*/
3975	static int hmR0VmxSetupVmcsProcCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3976	{
3977	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3978	uint32_t fVal = g_HmMsrs.u.vmx.ProcCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
3979	uint32_t const fZap = g_HmMsrs.u.vmx.ProcCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3980
3981	fVal \|= VMX_PROC_CTLS_HLT_EXIT /* HLT causes a VM-exit. */
3982	\| VMX_PROC_CTLS_USE_TSC_OFFSETTING /* Use TSC-offsetting. */
3983	\| VMX_PROC_CTLS_MOV_DR_EXIT /* MOV DRx causes a VM-exit. */
3984	\| VMX_PROC_CTLS_UNCOND_IO_EXIT /* All IO instructions cause a VM-exit. */
3985	\| VMX_PROC_CTLS_RDPMC_EXIT /* RDPMC causes a VM-exit. */
3986	\| VMX_PROC_CTLS_MONITOR_EXIT /* MONITOR causes a VM-exit. */
3987	\| VMX_PROC_CTLS_MWAIT_EXIT; /* MWAIT causes a VM-exit. */
3988
3989	/* We toggle VMX_PROC_CTLS_MOV_DR_EXIT later, check if it's not -always- needed to be set or clear. */
3990	if ( !(g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_MOV_DR_EXIT)
3991	\|\| (g_HmMsrs.u.vmx.ProcCtls.n.allowed0 & VMX_PROC_CTLS_MOV_DR_EXIT))
3992	{
3993	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_MOV_DRX_EXIT;
3994	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3995	}
3996
3997	/* Without nested paging, INVLPG (also affects INVPCID) and MOV CR3 instructions should cause VM-exits. */
3998	if (!pVM->hmr0.s.fNestedPaging)
3999	{
4000	Assert(!pVM->hmr0.s.vmx.fUnrestrictedGuest);
4001	fVal \|= VMX_PROC_CTLS_INVLPG_EXIT
4002	\| VMX_PROC_CTLS_CR3_LOAD_EXIT
4003	\| VMX_PROC_CTLS_CR3_STORE_EXIT;
4004	}
4005
4006	/* Use TPR shadowing if supported by the CPU. */
4007	if ( PDMHasApic(pVM)
4008	&& (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW))
4009	{
4010	fVal \|= VMX_PROC_CTLS_USE_TPR_SHADOW; /* CR8 reads from the Virtual-APIC page. */
4011	/* CR8 writes cause a VM-exit based on TPR threshold. */
4012	Assert(!(fVal & VMX_PROC_CTLS_CR8_STORE_EXIT));
4013	Assert(!(fVal & VMX_PROC_CTLS_CR8_LOAD_EXIT));
4014	hmR0VmxSetupVmcsVirtApicAddr(pVmcsInfo);
4015	}
4016	else
4017	{
4018	/* Some 32-bit CPUs do not support CR8 load/store exiting as MOV CR8 is
4019	invalid on 32-bit Intel CPUs. Set this control only for 64-bit guests. */
4020	if (pVM->hmr0.s.fAllow64BitGuests)
4021	fVal \|= VMX_PROC_CTLS_CR8_STORE_EXIT /* CR8 reads cause a VM-exit. */
4022	\| VMX_PROC_CTLS_CR8_LOAD_EXIT; /* CR8 writes cause a VM-exit. */
4023	}
4024
4025	/* Use MSR-bitmaps if supported by the CPU. */
4026	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
4027	{
4028	fVal \|= VMX_PROC_CTLS_USE_MSR_BITMAPS;
4029	hmR0VmxSetupVmcsMsrBitmapAddr(pVmcsInfo);
4030	}
4031
4032	/* Use the secondary processor-based VM-execution controls if supported by the CPU. */
4033	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
4034	fVal \|= VMX_PROC_CTLS_USE_SECONDARY_CTLS;
4035
4036	if ((fVal & fZap) != fVal)
4037	{
4038	LogRelFunc(("Invalid processor-based VM-execution controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
4039	g_HmMsrs.u.vmx.ProcCtls.n.allowed0, fVal, fZap));
4040	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC;
4041	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
4042	}
4043
4044	/* Commit it to the VMCS and update our cache. */
4045	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, fVal);
4046	AssertRC(rc);
4047	pVmcsInfo->u32ProcCtls = fVal;
4048
4049	/* Set up MSR permissions that don't change through the lifetime of the VM. */
4050	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
4051	hmR0VmxSetupVmcsMsrPermissions(pVCpu, pVmcsInfo);
4052
4053	/* Set up secondary processor-based VM-execution controls if the CPU supports it. */
4054	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
4055	return hmR0VmxSetupVmcsProcCtls2(pVCpu, pVmcsInfo);
4056
4057	/* Sanity check, should not really happen. */
4058	if (RT_LIKELY(!pVM->hmr0.s.vmx.fUnrestrictedGuest))
4059	{ /* likely */ }
4060	else
4061	{
4062	pVCpu->hm.s.u32HMError = VMX_UFC_INVALID_UX_COMBO;
4063	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
4064	}
4065
4066	/* Old CPUs without secondary processor-based VM-execution controls would end up here. */
4067	return VINF_SUCCESS;
4068	}
4069
4070
4071	/**
4072	* Sets up miscellaneous (everything other than Pin, Processor and secondary
4073	* Processor-based VM-execution) control fields in the VMCS.
4074	*
4075	* @returns VBox status code.
4076	* @param pVCpu The cross context virtual CPU structure.
4077	* @param pVmcsInfo The VMCS info. object.
4078	*/
4079	static int hmR0VmxSetupVmcsMiscCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
4080	{
4081	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4082	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUseVmcsShadowing)
4083	{
4084	hmR0VmxSetupVmcsVmreadBitmapAddr(pVCpu);
4085	hmR0VmxSetupVmcsVmwriteBitmapAddr(pVCpu);
4086	}
4087	#endif
4088
4089	Assert(pVmcsInfo->u64VmcsLinkPtr == NIL_RTHCPHYS);
4090	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
4091	AssertRC(rc);
4092
4093	rc = hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(pVmcsInfo);
4094	if (RT_SUCCESS(rc))
4095	{
4096	uint64_t const u64Cr0Mask = hmR0VmxGetFixedCr0Mask(pVCpu);
4097	uint64_t const u64Cr4Mask = hmR0VmxGetFixedCr4Mask(pVCpu);
4098
4099	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, u64Cr0Mask); AssertRC(rc);
4100	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, u64Cr4Mask); AssertRC(rc);
4101
4102	pVmcsInfo->u64Cr0Mask = u64Cr0Mask;
4103	pVmcsInfo->u64Cr4Mask = u64Cr4Mask;
4104
4105	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fLbr)
4106	{
4107	rc = VMXWriteVmcsNw(VMX_VMCS64_GUEST_DEBUGCTL_FULL, MSR_IA32_DEBUGCTL_LBR);
4108	AssertRC(rc);
4109	}
4110	return VINF_SUCCESS;
4111	}
4112	else
4113	LogRelFunc(("Failed to initialize VMCS auto-load/store MSR addresses. rc=%Rrc\n", rc));
4114	return rc;
4115	}
4116
4117
4118	/**
4119	* Sets up the initial exception bitmap in the VMCS based on static conditions.
4120	*
4121	* We shall setup those exception intercepts that don't change during the
4122	* lifetime of the VM here. The rest are done dynamically while loading the
4123	* guest state.
4124	*
4125	* @param pVCpu The cross context virtual CPU structure.
4126	* @param pVmcsInfo The VMCS info. object.
4127	*/
4128	static void hmR0VmxSetupVmcsXcptBitmap(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
4129	{
4130	/*
4131	* The following exceptions are always intercepted:
4132	*
4133	* #AC - To prevent the guest from hanging the CPU and for dealing with
4134	* split-lock detecting host configs.
4135	* #DB - To maintain the DR6 state even when intercepting DRx reads/writes and
4136	* recursive #DBs can cause a CPU hang.
4137	* #PF - To sync our shadow page tables when nested-paging is not used.
4138	*/
4139	bool const fNestedPaging = pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging;
4140	uint32_t const uXcptBitmap = RT_BIT(X86_XCPT_AC)
4141	\| RT_BIT(X86_XCPT_DB)
4142	\| (fNestedPaging ? 0 : RT_BIT(X86_XCPT_PF));
4143
4144	/* Commit it to the VMCS. */
4145	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
4146	AssertRC(rc);
4147
4148	/* Update our cache of the exception bitmap. */
4149	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
4150	}
4151
4152
4153	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4154	/**
4155	* Sets up the VMCS for executing a nested-guest using hardware-assisted VMX.
4156	*
4157	* @returns VBox status code.
4158	* @param pVmcsInfo The VMCS info. object.
4159	*/
4160	static int hmR0VmxSetupVmcsCtlsNested(PVMXVMCSINFO pVmcsInfo)
4161	{
4162	Assert(pVmcsInfo->u64VmcsLinkPtr == NIL_RTHCPHYS);
4163	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
4164	AssertRC(rc);
4165
4166	rc = hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(pVmcsInfo);
4167	if (RT_SUCCESS(rc))
4168	{
4169	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
4170	hmR0VmxSetupVmcsMsrBitmapAddr(pVmcsInfo);
4171
4172	/* Paranoia - We've not yet initialized these, they shall be done while merging the VMCS. */
4173	Assert(!pVmcsInfo->u64Cr0Mask);
4174	Assert(!pVmcsInfo->u64Cr4Mask);
4175	return VINF_SUCCESS;
4176	}
4177	LogRelFunc(("Failed to set up the VMCS link pointer in the nested-guest VMCS. rc=%Rrc\n", rc));
4178	return rc;
4179	}
4180	#endif
4181
4182
4183	/**
4184	* Sets pfnStartVm to the best suited variant.
4185	*
4186	* This must be called whenever anything changes relative to the hmR0VmXStartVm
4187	* variant selection:
4188	* - pVCpu->hm.s.fLoadSaveGuestXcr0
4189	* - HM_WSF_IBPB_ENTRY in pVCpu->hmr0.s.fWorldSwitcher
4190	* - HM_WSF_IBPB_EXIT in pVCpu->hmr0.s.fWorldSwitcher
4191	* - Perhaps: CPUMIsGuestFPUStateActive() (windows only)
4192	* - Perhaps: CPUMCTX.fXStateMask (windows only)
4193	*
4194	* We currently ASSUME that neither HM_WSF_IBPB_ENTRY nor HM_WSF_IBPB_EXIT
4195	* cannot be changed at runtime.
4196	*/
4197	static void hmR0VmxUpdateStartVmFunction(PVMCPUCC pVCpu)
4198	{
4199	static const struct CLANGWORKAROUND { PFNHMVMXSTARTVM pfn; } s_aHmR0VmxStartVmFunctions[] =
4200	{
4201	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_SansL1dEntry_SansMdsEntry_SansIbpbExit },
4202	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_SansL1dEntry_SansMdsEntry_SansIbpbExit },
4203	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_SansL1dEntry_SansMdsEntry_SansIbpbExit },
4204	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_SansL1dEntry_SansMdsEntry_SansIbpbExit },
4205	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_WithL1dEntry_SansMdsEntry_SansIbpbExit },
4206	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_WithL1dEntry_SansMdsEntry_SansIbpbExit },
4207	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_WithL1dEntry_SansMdsEntry_SansIbpbExit },
4208	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_WithL1dEntry_SansMdsEntry_SansIbpbExit },
4209	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_SansL1dEntry_WithMdsEntry_SansIbpbExit },
4210	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_SansL1dEntry_WithMdsEntry_SansIbpbExit },
4211	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_SansL1dEntry_WithMdsEntry_SansIbpbExit },
4212	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_SansL1dEntry_WithMdsEntry_SansIbpbExit },
4213	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_WithL1dEntry_WithMdsEntry_SansIbpbExit },
4214	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_WithL1dEntry_WithMdsEntry_SansIbpbExit },
4215	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_WithL1dEntry_WithMdsEntry_SansIbpbExit },
4216	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_WithL1dEntry_WithMdsEntry_SansIbpbExit },
4217	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_SansL1dEntry_SansMdsEntry_WithIbpbExit },
4218	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_SansL1dEntry_SansMdsEntry_WithIbpbExit },
4219	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_SansL1dEntry_SansMdsEntry_WithIbpbExit },
4220	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_SansL1dEntry_SansMdsEntry_WithIbpbExit },
4221	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_WithL1dEntry_SansMdsEntry_WithIbpbExit },
4222	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_WithL1dEntry_SansMdsEntry_WithIbpbExit },
4223	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_WithL1dEntry_SansMdsEntry_WithIbpbExit },
4224	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_WithL1dEntry_SansMdsEntry_WithIbpbExit },
4225	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_SansL1dEntry_WithMdsEntry_WithIbpbExit },
4226	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_SansL1dEntry_WithMdsEntry_WithIbpbExit },
4227	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_SansL1dEntry_WithMdsEntry_WithIbpbExit },
4228	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_SansL1dEntry_WithMdsEntry_WithIbpbExit },
4229	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_WithL1dEntry_WithMdsEntry_WithIbpbExit },
4230	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_WithL1dEntry_WithMdsEntry_WithIbpbExit },
4231	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_WithL1dEntry_WithMdsEntry_WithIbpbExit },
4232	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_WithL1dEntry_WithMdsEntry_WithIbpbExit },
4233	};
4234	uintptr_t const idx = (pVCpu->hmr0.s.fLoadSaveGuestXcr0 ? 1 : 0)
4235	\| (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_IBPB_ENTRY ? 2 : 0)
4236	\| (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_L1D_ENTRY ? 4 : 0)
4237	\| (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_MDS_ENTRY ? 8 : 0)
4238	\| (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_IBPB_EXIT ? 16 : 0);
4239	PFNHMVMXSTARTVM const pfnStartVm = s_aHmR0VmxStartVmFunctions[idx].pfn;
4240	if (pVCpu->hmr0.s.vmx.pfnStartVm != pfnStartVm)
4241	pVCpu->hmr0.s.vmx.pfnStartVm = pfnStartVm;
4242	}
4243
4244
4245	/**
4246	* Selector FNHMSVMVMRUN implementation.
4247	*/
4248	static DECLCALLBACK(int) hmR0VmxStartVmSelector(PVMXVMCSINFO pVmcsInfo, PVMCPUCC pVCpu, bool fResume)
4249	{
4250	hmR0VmxUpdateStartVmFunction(pVCpu);
4251	return pVCpu->hmr0.s.vmx.pfnStartVm(pVmcsInfo, pVCpu, fResume);
4252	}
4253
4254
4255	/**
4256	* Sets up the VMCS for executing a guest (or nested-guest) using hardware-assisted
4257	* VMX.
4258	*
4259	* @returns VBox status code.
4260	* @param pVCpu The cross context virtual CPU structure.
4261	* @param pVmcsInfo The VMCS info. object.
4262	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
4263	*/
4264	static int hmR0VmxSetupVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
4265	{
4266	Assert(pVmcsInfo->pvVmcs);
4267	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4268
4269	/* Set the CPU specified revision identifier at the beginning of the VMCS structure. */
4270	(uint32_t )pVmcsInfo->pvVmcs = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_ID);
4271	const char * const pszVmcs = fIsNstGstVmcs ? "nested-guest VMCS" : "guest VMCS";
4272
4273	LogFlowFunc(("\n"));
4274
4275	/*
4276	* Initialize the VMCS using VMCLEAR before loading the VMCS.
4277	* See Intel spec. 31.6 "Preparation And Launching A Virtual Machine".
4278	*/
4279	int rc = hmR0VmxClearVmcs(pVmcsInfo);
4280	if (RT_SUCCESS(rc))
4281	{
4282	rc = hmR0VmxLoadVmcs(pVmcsInfo);
4283	if (RT_SUCCESS(rc))
4284	{
4285	/*
4286	* Initialize the hardware-assisted VMX execution handler for guest and nested-guest VMCS.
4287	* The host is always 64-bit since we no longer support 32-bit hosts.
4288	* Currently we have just a single handler for all guest modes as well, see @bugref{6208#c73}.
4289	*/
4290	if (!fIsNstGstVmcs)
4291	{
4292	rc = hmR0VmxSetupVmcsPinCtls(pVCpu, pVmcsInfo);
4293	if (RT_SUCCESS(rc))
4294	{
4295	rc = hmR0VmxSetupVmcsProcCtls(pVCpu, pVmcsInfo);
4296	if (RT_SUCCESS(rc))
4297	{
4298	rc = hmR0VmxSetupVmcsMiscCtls(pVCpu, pVmcsInfo);
4299	if (RT_SUCCESS(rc))
4300	{
4301	hmR0VmxSetupVmcsXcptBitmap(pVCpu, pVmcsInfo);
4302	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4303	/*
4304	* If a shadow VMCS is allocated for the VMCS info. object, initialize the
4305	* VMCS revision ID and shadow VMCS indicator bit. Also, clear the VMCS
4306	* making it fit for use when VMCS shadowing is later enabled.
4307	*/
4308	if (pVmcsInfo->pvShadowVmcs)
4309	{
4310	VMXVMCSREVID VmcsRevId;
4311	VmcsRevId.u = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_ID);
4312	VmcsRevId.n.fIsShadowVmcs = 1;
4313	(uint32_t )pVmcsInfo->pvShadowVmcs = VmcsRevId.u;
4314	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
4315	if (RT_SUCCESS(rc))
4316	{ /* likely */ }
4317	else
4318	LogRelFunc(("Failed to initialize shadow VMCS. rc=%Rrc\n", rc));
4319	}
4320	#endif
4321	}
4322	else
4323	LogRelFunc(("Failed to setup miscellaneous controls. rc=%Rrc\n", rc));
4324	}
4325	else
4326	LogRelFunc(("Failed to setup processor-based VM-execution controls. rc=%Rrc\n", rc));
4327	}
4328	else
4329	LogRelFunc(("Failed to setup pin-based controls. rc=%Rrc\n", rc));
4330	}
4331	else
4332	{
4333	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4334	rc = hmR0VmxSetupVmcsCtlsNested(pVmcsInfo);
4335	if (RT_SUCCESS(rc))
4336	{ /* likely */ }
4337	else
4338	LogRelFunc(("Failed to initialize nested-guest VMCS. rc=%Rrc\n", rc));
4339	#else
4340	AssertFailed();
4341	#endif
4342	}
4343	}
4344	else
4345	LogRelFunc(("Failed to load the %s. rc=%Rrc\n", rc, pszVmcs));
4346	}
4347	else
4348	LogRelFunc(("Failed to clear the %s. rc=%Rrc\n", rc, pszVmcs));
4349
4350	/* Sync any CPU internal VMCS data back into our VMCS in memory. */
4351	if (RT_SUCCESS(rc))
4352	{
4353	rc = hmR0VmxClearVmcs(pVmcsInfo);
4354	if (RT_SUCCESS(rc))
4355	{ /* likely */ }
4356	else
4357	LogRelFunc(("Failed to clear the %s post setup. rc=%Rrc\n", rc, pszVmcs));
4358	}
4359
4360	/*
4361	* Update the last-error record both for failures and success, so we
4362	* can propagate the status code back to ring-3 for diagnostics.
4363	*/
4364	hmR0VmxUpdateErrorRecord(pVCpu, rc);
4365	NOREF(pszVmcs);
4366	return rc;
4367	}
4368
4369
4370	/**
4371	* Does global VT-x initialization (called during module initialization).
4372	*
4373	* @returns VBox status code.
4374	*/
4375	VMMR0DECL(int) VMXR0GlobalInit(void)
4376	{
4377	#ifdef HMVMX_USE_FUNCTION_TABLE
4378	AssertCompile(VMX_EXIT_MAX + 1 == RT_ELEMENTS(g_aVMExitHandlers));
4379	# ifdef VBOX_STRICT
4380	for (unsigned i = 0; i < RT_ELEMENTS(g_aVMExitHandlers); i++)
4381	Assert(g_aVMExitHandlers[i].pfn);
4382	# endif
4383	#endif
4384	return VINF_SUCCESS;
4385	}
4386
4387
4388	/**
4389	* Does global VT-x termination (called during module termination).
4390	*/
4391	VMMR0DECL(void) VMXR0GlobalTerm()
4392	{
4393	/* Nothing to do currently. */
4394	}
4395
4396
4397	/**
4398	* Sets up and activates VT-x on the current CPU.
4399	*
4400	* @returns VBox status code.
4401	* @param pHostCpu The HM physical-CPU structure.
4402	* @param pVM The cross context VM structure. Can be
4403	* NULL after a host resume operation.
4404	* @param pvCpuPage Pointer to the VMXON region (can be NULL if @a
4405	* fEnabledByHost is @c true).
4406	* @param HCPhysCpuPage Physical address of the VMXON region (can be 0 if
4407	* @a fEnabledByHost is @c true).
4408	* @param fEnabledByHost Set if SUPR0EnableVTx() or similar was used to
4409	* enable VT-x on the host.
4410	* @param pHwvirtMsrs Pointer to the hardware-virtualization MSRs.
4411	*/
4412	VMMR0DECL(int) VMXR0EnableCpu(PHMPHYSCPU pHostCpu, PVMCC pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost,
4413	PCSUPHWVIRTMSRS pHwvirtMsrs)
4414	{
4415	AssertPtr(pHostCpu);
4416	AssertPtr(pHwvirtMsrs);
4417	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4418
4419	/* Enable VT-x if it's not already enabled by the host. */
4420	if (!fEnabledByHost)
4421	{
4422	int rc = hmR0VmxEnterRootMode(pHostCpu, pVM, HCPhysCpuPage, pvCpuPage);
4423	if (RT_FAILURE(rc))
4424	return rc;
4425	}
4426
4427	/*
4428	* Flush all EPT tagged-TLB entries (in case VirtualBox or any other hypervisor have been
4429	* using EPTPs) so we don't retain any stale guest-physical mappings which won't get
4430	* invalidated when flushing by VPID.
4431	*/
4432	if (pHwvirtMsrs->u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
4433	{
4434	hmR0VmxFlushEpt(NULL /* pVCpu /, NULL / pVmcsInfo */, VMXTLBFLUSHEPT_ALL_CONTEXTS);
4435	pHostCpu->fFlushAsidBeforeUse = false;
4436	}
4437	else
4438	pHostCpu->fFlushAsidBeforeUse = true;
4439
4440	/* Ensure each VCPU scheduled on this CPU gets a new VPID on resume. See @bugref{6255}. */
4441	++pHostCpu->cTlbFlushes;
4442
4443	return VINF_SUCCESS;
4444	}
4445
4446
4447	/**
4448	* Deactivates VT-x on the current CPU.
4449	*
4450	* @returns VBox status code.
4451	* @param pHostCpu The HM physical-CPU structure.
4452	* @param pvCpuPage Pointer to the VMXON region.
4453	* @param HCPhysCpuPage Physical address of the VMXON region.
4454	*
4455	* @remarks This function should never be called when SUPR0EnableVTx() or
4456	* similar was used to enable VT-x on the host.
4457	*/
4458	VMMR0DECL(int) VMXR0DisableCpu(PHMPHYSCPU pHostCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
4459	{
4460	RT_NOREF2(pvCpuPage, HCPhysCpuPage);
4461
4462	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4463	return hmR0VmxLeaveRootMode(pHostCpu);
4464	}
4465
4466
4467	/**
4468	* Does per-VM VT-x initialization.
4469	*
4470	* @returns VBox status code.
4471	* @param pVM The cross context VM structure.
4472	*/
4473	VMMR0DECL(int) VMXR0InitVM(PVMCC pVM)
4474	{
4475	AssertPtr(pVM);
4476	LogFlowFunc(("pVM=%p\n", pVM));
4477
4478	hmR0VmxStructsInit(pVM);
4479	int rc = hmR0VmxStructsAlloc(pVM);
4480	if (RT_FAILURE(rc))
4481	{
4482	LogRelFunc(("Failed to allocated VMX structures. rc=%Rrc\n", rc));
4483	return rc;
4484	}
4485
4486	/* Setup the crash dump page. */
4487	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4488	strcpy((char *)pVM->hmr0.s.vmx.pbScratch, "SCRATCH Magic");
4489	(uint64_t )(pVM->hmr0.s.vmx.pbScratch + 16) = UINT64_C(0xdeadbeefdeadbeef);
4490	#endif
4491	return VINF_SUCCESS;
4492	}
4493
4494
4495	/**
4496	* Does per-VM VT-x termination.
4497	*
4498	* @returns VBox status code.
4499	* @param pVM The cross context VM structure.
4500	*/
4501	VMMR0DECL(int) VMXR0TermVM(PVMCC pVM)
4502	{
4503	AssertPtr(pVM);
4504	LogFlowFunc(("pVM=%p\n", pVM));
4505
4506	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4507	if (pVM->hmr0.s.vmx.pbScratch)
4508	RT_BZERO(pVM->hmr0.s.vmx.pbScratch, X86_PAGE_4K_SIZE);
4509	#endif
4510	hmR0VmxStructsFree(pVM);
4511	return VINF_SUCCESS;
4512	}
4513
4514
4515	/**
4516	* Sets up the VM for execution using hardware-assisted VMX.
4517	* This function is only called once per-VM during initialization.
4518	*
4519	* @returns VBox status code.
4520	* @param pVM The cross context VM structure.
4521	*/
4522	VMMR0DECL(int) VMXR0SetupVM(PVMCC pVM)
4523	{
4524	AssertPtr(pVM);
4525	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4526
4527	LogFlowFunc(("pVM=%p\n", pVM));
4528
4529	/*
4530	* At least verify if VMX is enabled, since we can't check if we're in VMX root mode or not
4531	* without causing a #GP.
4532	*/
4533	RTCCUINTREG const uHostCr4 = ASMGetCR4();
4534	if (RT_LIKELY(uHostCr4 & X86_CR4_VMXE))
4535	{ /* likely */ }
4536	else
4537	return VERR_VMX_NOT_IN_VMX_ROOT_MODE;
4538
4539	/*
4540	* Check that nested paging is supported if enabled and copy over the flag to the
4541	* ring-0 only structure.
4542	*/
4543	bool const fNestedPaging = pVM->hm.s.fNestedPagingCfg;
4544	AssertReturn( !fNestedPaging
4545	\|\| (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_EPT), /** @todo use a ring-0 copy of ProcCtls2.n.allowed1 */
4546	VERR_INCOMPATIBLE_CONFIG);
4547	pVM->hmr0.s.fNestedPaging = fNestedPaging;
4548	pVM->hmr0.s.fAllow64BitGuests = pVM->hm.s.fAllow64BitGuestsCfg;
4549
4550	/*
4551	* Without unrestricted guest execution, pRealModeTSS and pNonPagingModeEPTPageTable must
4552	* always be allocated. We no longer support the highly unlikely case of unrestricted guest
4553	* without pRealModeTSS, see hmR3InitFinalizeR0Intel().
4554	*/
4555	bool const fUnrestrictedGuest = pVM->hm.s.vmx.fUnrestrictedGuestCfg;
4556	AssertReturn( !fUnrestrictedGuest
4557	\|\| ( (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST)
4558	&& fNestedPaging),
4559	VERR_INCOMPATIBLE_CONFIG);
4560	if ( !fUnrestrictedGuest
4561	&& ( !pVM->hm.s.vmx.pNonPagingModeEPTPageTable
4562	\|\| !pVM->hm.s.vmx.pRealModeTSS))
4563	{
4564	LogRelFunc(("Invalid real-on-v86 state.\n"));
4565	return VERR_INTERNAL_ERROR;
4566	}
4567	pVM->hmr0.s.vmx.fUnrestrictedGuest = fUnrestrictedGuest;
4568
4569	/* Initialize these always, see hmR3InitFinalizeR0().*/
4570	pVM->hm.s.ForR3.vmx.enmTlbFlushEpt = pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NONE;
4571	pVM->hm.s.ForR3.vmx.enmTlbFlushVpid = pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NONE;
4572
4573	/* Setup the tagged-TLB flush handlers. */
4574	int rc = hmR0VmxSetupTaggedTlb(pVM);
4575	if (RT_FAILURE(rc))
4576	{
4577	LogRelFunc(("Failed to setup tagged TLB. rc=%Rrc\n", rc));
4578	return rc;
4579	}
4580
4581	/* Determine LBR capabilities. */
4582	pVM->hmr0.s.vmx.fLbr = pVM->hm.s.vmx.fLbrCfg;
4583	if (pVM->hmr0.s.vmx.fLbr)
4584	{
4585	rc = hmR0VmxSetupLbrMsrRange(pVM);
4586	if (RT_FAILURE(rc))
4587	{
4588	LogRelFunc(("Failed to setup LBR MSR range. rc=%Rrc\n", rc));
4589	return rc;
4590	}
4591	}
4592
4593	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4594	/* Setup the shadow VMCS fields array and VMREAD/VMWRITE bitmaps. */
4595	if (pVM->hmr0.s.vmx.fUseVmcsShadowing)
4596	{
4597	rc = hmR0VmxSetupShadowVmcsFieldsArrays(pVM);
4598	if (RT_SUCCESS(rc))
4599	hmR0VmxSetupVmreadVmwriteBitmaps(pVM);
4600	else
4601	{
4602	LogRelFunc(("Failed to setup shadow VMCS fields arrays. rc=%Rrc\n", rc));
4603	return rc;
4604	}
4605	}
4606	#endif
4607
4608	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
4609	{
4610	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
4611	Log4Func(("pVCpu=%p idCpu=%RU32\n", pVCpu, pVCpu->idCpu));
4612
4613	pVCpu->hmr0.s.vmx.pfnStartVm = hmR0VmxStartVmSelector;
4614
4615	rc = hmR0VmxSetupVmcs(pVCpu, &pVCpu->hmr0.s.vmx.VmcsInfo, false /* fIsNstGstVmcs */);
4616	if (RT_SUCCESS(rc))
4617	{
4618	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4619	if (pVM->cpum.ro.GuestFeatures.fVmx)
4620	{
4621	rc = hmR0VmxSetupVmcs(pVCpu, &pVCpu->hmr0.s.vmx.VmcsInfoNstGst, true /* fIsNstGstVmcs */);
4622	if (RT_SUCCESS(rc))
4623	{ /* likely */ }
4624	else
4625	{
4626	LogRelFunc(("Nested-guest VMCS setup failed. rc=%Rrc\n", rc));
4627	return rc;
4628	}
4629	}
4630	#endif
4631	}
4632	else
4633	{
4634	LogRelFunc(("VMCS setup failed. rc=%Rrc\n", rc));
4635	return rc;
4636	}
4637	}
4638
4639	return VINF_SUCCESS;
4640	}
4641
4642
4643	/**
4644	* Saves the host control registers (CR0, CR3, CR4) into the host-state area in
4645	* the VMCS.
4646	* @returns CR4 for passing along to hmR0VmxExportHostSegmentRegs.
4647	*/
4648	static uint64_t hmR0VmxExportHostControlRegs(void)
4649	{
4650	int rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR0, ASMGetCR0()); AssertRC(rc);
4651	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR3, ASMGetCR3()); AssertRC(rc);
4652	uint64_t uHostCr4 = ASMGetCR4();
4653	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR4, uHostCr4); AssertRC(rc);
4654	return uHostCr4;
4655	}
4656
4657
4658	/**
4659	* Saves the host segment registers and GDTR, IDTR, (TR, GS and FS bases) into
4660	* the host-state area in the VMCS.
4661	*
4662	* @returns VBox status code.
4663	* @param pVCpu The cross context virtual CPU structure.
4664	* @param uHostCr4 The host CR4 value.
4665	*/
4666	static int hmR0VmxExportHostSegmentRegs(PVMCPUCC pVCpu, uint64_t uHostCr4)
4667	{
4668	/*
4669	* If we've executed guest code using hardware-assisted VMX, the host-state bits
4670	* will be messed up. We should -not- save the messed up state without restoring
4671	* the original host-state, see @bugref{7240}.
4672	*
4673	* This apparently can happen (most likely the FPU changes), deal with it rather than
4674	* asserting. Was observed booting Solaris 10u10 32-bit guest.
4675	*/
4676	if (pVCpu->hmr0.s.vmx.fRestoreHostFlags > VMX_RESTORE_HOST_REQUIRED)
4677	{
4678	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hmr0.s.vmx.fRestoreHostFlags,
4679	pVCpu->idCpu));
4680	VMXRestoreHostState(pVCpu->hmr0.s.vmx.fRestoreHostFlags, &pVCpu->hmr0.s.vmx.RestoreHost);
4681	pVCpu->hmr0.s.vmx.fRestoreHostFlags = 0;
4682	}
4683
4684	/*
4685	* Get all the host info.
4686	* ASSUME it is safe to use rdfsbase and friends if the CR4.FSGSBASE bit is set
4687	* without also checking the cpuid bit.
4688	*/
4689	uint32_t fRestoreHostFlags;
4690	#if RT_INLINE_ASM_EXTERNAL
4691	if (uHostCr4 & X86_CR4_FSGSBASE)
4692	{
4693	hmR0VmxExportHostSegmentRegsAsmHlp(&pVCpu->hmr0.s.vmx.RestoreHost, true /fHaveFsGsBase/);
4694	fRestoreHostFlags = VMX_RESTORE_HOST_CAN_USE_WRFSBASE_AND_WRGSBASE;
4695	}
4696	else
4697	{
4698	hmR0VmxExportHostSegmentRegsAsmHlp(&pVCpu->hmr0.s.vmx.RestoreHost, false /fHaveFsGsBase/);
4699	fRestoreHostFlags = 0;
4700	}
4701	RTSEL uSelES = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelES;
4702	RTSEL uSelDS = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelDS;
4703	RTSEL uSelFS = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelFS;
4704	RTSEL uSelGS = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelGS;
4705	#else
4706	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR = ASMGetTR();
4707	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelSS = ASMGetSS();
4708	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelCS = ASMGetCS();
4709	ASMGetGDTR((PRTGDTR)&pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr);
4710	ASMGetIDTR((PRTIDTR)&pVCpu->hmr0.s.vmx.RestoreHost.HostIdtr);
4711	if (uHostCr4 & X86_CR4_FSGSBASE)
4712	{
4713	pVCpu->hmr0.s.vmx.RestoreHost.uHostFSBase = ASMGetFSBase();
4714	pVCpu->hmr0.s.vmx.RestoreHost.uHostGSBase = ASMGetGSBase();
4715	fRestoreHostFlags = VMX_RESTORE_HOST_CAN_USE_WRFSBASE_AND_WRGSBASE;
4716	}
4717	else
4718	{
4719	pVCpu->hmr0.s.vmx.RestoreHost.uHostFSBase = ASMRdMsr(MSR_K8_FS_BASE);
4720	pVCpu->hmr0.s.vmx.RestoreHost.uHostGSBase = ASMRdMsr(MSR_K8_GS_BASE);
4721	fRestoreHostFlags = 0;
4722	}
4723	RTSEL uSelES, uSelDS, uSelFS, uSelGS;
4724	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelDS = uSelDS = ASMGetDS();
4725	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelES = uSelES = ASMGetES();
4726	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelFS = uSelFS = ASMGetFS();
4727	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelGS = uSelGS = ASMGetGS();
4728	#endif
4729
4730	/*
4731	* Determine if the host segment registers are suitable for VT-x. Otherwise use zero to
4732	* gain VM-entry and restore them before we get preempted.
4733	*
4734	* See Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers".
4735	*/
4736	RTSEL const uSelAll = uSelFS \| uSelGS \| uSelES \| uSelDS;
4737	if (uSelAll & (X86_SEL_RPL \| X86_SEL_LDT))
4738	{
4739	if (!(uSelAll & X86_SEL_LDT))
4740	{
4741	#define VMXLOCAL_ADJUST_HOST_SEG(a_Seg, a_uVmcsVar) \
4742	do { \
4743	(a_uVmcsVar) = pVCpu->hmr0.s.vmx.RestoreHost.uHostSel##a_Seg; \
4744	if ((a_uVmcsVar) & X86_SEL_RPL) \
4745	{ \
4746	fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##a_Seg; \
4747	(a_uVmcsVar) = 0; \
4748	} \
4749	} while (0)
4750	VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS);
4751	VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES);
4752	VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS);
4753	VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS);
4754	#undef VMXLOCAL_ADJUST_HOST_SEG
4755	}
4756	else
4757	{
4758	#define VMXLOCAL_ADJUST_HOST_SEG(a_Seg, a_uVmcsVar) \
4759	do { \
4760	(a_uVmcsVar) = pVCpu->hmr0.s.vmx.RestoreHost.uHostSel##a_Seg; \
4761	if ((a_uVmcsVar) & (X86_SEL_RPL \| X86_SEL_LDT)) \
4762	{ \
4763	if (!((a_uVmcsVar) & X86_SEL_LDT)) \
4764	fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##a_Seg; \
4765	else \
4766	{ \
4767	uint32_t const fAttr = ASMGetSegAttr(a_uVmcsVar); \
4768	if ((fAttr & X86_DESC_P) && fAttr != UINT32_MAX) \
4769	fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##a_Seg; \
4770	} \
4771	(a_uVmcsVar) = 0; \
4772	} \
4773	} while (0)
4774	VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS);
4775	VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES);
4776	VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS);
4777	VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS);
4778	#undef VMXLOCAL_ADJUST_HOST_SEG
4779	}
4780	}
4781
4782	/* Verification based on Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers" */
4783	Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR & X86_SEL_RPL)); Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR & X86_SEL_LDT)); Assert(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR);
4784	Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelCS & X86_SEL_RPL)); Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelCS & X86_SEL_LDT)); Assert(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelCS);
4785	Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelSS & X86_SEL_RPL)); Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelSS & X86_SEL_LDT));
4786	Assert(!(uSelDS & X86_SEL_RPL)); Assert(!(uSelDS & X86_SEL_LDT));
4787	Assert(!(uSelES & X86_SEL_RPL)); Assert(!(uSelES & X86_SEL_LDT));
4788	Assert(!(uSelFS & X86_SEL_RPL)); Assert(!(uSelFS & X86_SEL_LDT));
4789	Assert(!(uSelGS & X86_SEL_RPL)); Assert(!(uSelGS & X86_SEL_LDT));
4790
4791	/*
4792	* Determine if we need to manually need to restore the GDTR and IDTR limits as VT-x zaps
4793	* them to the maximum limit (0xffff) on every VM-exit.
4794	*/
4795	if (pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.cb != 0xffff)
4796	fRestoreHostFlags \|= VMX_RESTORE_HOST_GDTR;
4797
4798	/*
4799	* IDT limit is effectively capped at 0xfff. (See Intel spec. 6.14.1 "64-Bit Mode IDT" and
4800	* Intel spec. 6.2 "Exception and Interrupt Vectors".) Therefore if the host has the limit
4801	* as 0xfff, VT-x bloating the limit to 0xffff shouldn't cause any different CPU behavior.
4802	* However, several hosts either insists on 0xfff being the limit (Windows Patch Guard) or
4803	* uses the limit for other purposes (darwin puts the CPU ID in there but botches sidt
4804	* alignment in at least one consumer). So, we're only allowing the IDTR.LIMIT to be left
4805	* at 0xffff on hosts where we are sure it won't cause trouble.
4806	*/
4807	#if defined(RT_OS_LINUX) \|\| defined(RT_OS_SOLARIS)
4808	if (pVCpu->hmr0.s.vmx.RestoreHost.HostIdtr.cb < 0x0fff)
4809	#else
4810	if (pVCpu->hmr0.s.vmx.RestoreHost.HostIdtr.cb != 0xffff)
4811	#endif
4812	fRestoreHostFlags \|= VMX_RESTORE_HOST_IDTR;
4813
4814	/*
4815	* Host TR base. Verify that TR selector doesn't point past the GDT. Masking off the TI
4816	* and RPL bits is effectively what the CPU does for "scaling by 8". TI is always 0 and
4817	* RPL should be too in most cases.
4818	*/
4819	RTSEL const uSelTR = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR;
4820	AssertMsgReturn((uSelTR \| X86_SEL_RPL_LDT) <= pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.cb,
4821	("TR selector exceeds limit. TR=%RTsel cbGdt=%#x\n", uSelTR, pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.cb),
4822	VERR_VMX_INVALID_HOST_STATE);
4823
4824	PCX86DESCHC pDesc = (PCX86DESCHC)(pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.uAddr + (uSelTR & X86_SEL_MASK));
4825	uintptr_t const uTRBase = X86DESC64_BASE(pDesc);
4826
4827	/*
4828	* VT-x unconditionally restores the TR limit to 0x67 and type to 11 (32-bit busy TSS) on
4829	* all VM-exits. The type is the same for 64-bit busy TSS[1]. The limit needs manual
4830	* restoration if the host has something else. Task switching is not supported in 64-bit
4831	* mode[2], but the limit still matters as IOPM is supported in 64-bit mode. Restoring the
4832	* limit lazily while returning to ring-3 is safe because IOPM is not applicable in ring-0.
4833	*
4834	* [1] See Intel spec. 3.5 "System Descriptor Types".
4835	* [2] See Intel spec. 7.2.3 "TSS Descriptor in 64-bit mode".
4836	*/
4837	Assert(pDesc->System.u4Type == 11);
4838	if ( pDesc->System.u16LimitLow != 0x67
4839	\|\| pDesc->System.u4LimitHigh)
4840	{
4841	fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_TR;
4842
4843	/* If the host has made GDT read-only, we would need to temporarily toggle CR0.WP before writing the GDT. */
4844	if (g_fHmHostKernelFeatures & SUPKERNELFEATURES_GDT_READ_ONLY)
4845	fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_READ_ONLY;
4846	if (g_fHmHostKernelFeatures & SUPKERNELFEATURES_GDT_NEED_WRITABLE)
4847	{
4848	/* The GDT is read-only but the writable GDT is available. */
4849	fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_NEED_WRITABLE;
4850	pVCpu->hmr0.s.vmx.RestoreHost.HostGdtrRw.cb = pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.cb;
4851	int rc = SUPR0GetCurrentGdtRw(&pVCpu->hmr0.s.vmx.RestoreHost.HostGdtrRw.uAddr);
4852	AssertRCReturn(rc, rc);
4853	}
4854	}
4855
4856	pVCpu->hmr0.s.vmx.fRestoreHostFlags = fRestoreHostFlags;
4857
4858	/*
4859	* Do all the VMCS updates in one block to assist nested virtualization.
4860	*/
4861	int rc;
4862	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_CS_SEL, pVCpu->hmr0.s.vmx.RestoreHost.uHostSelCS); AssertRC(rc);
4863	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_SS_SEL, pVCpu->hmr0.s.vmx.RestoreHost.uHostSelSS); AssertRC(rc);
4864	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_DS_SEL, uSelDS); AssertRC(rc);
4865	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_ES_SEL, uSelES); AssertRC(rc);
4866	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_FS_SEL, uSelFS); AssertRC(rc);
4867	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_GS_SEL, uSelGS); AssertRC(rc);
4868	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_TR_SEL, pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR); AssertRC(rc);
4869	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_GDTR_BASE, pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.uAddr); AssertRC(rc);
4870	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_IDTR_BASE, pVCpu->hmr0.s.vmx.RestoreHost.HostIdtr.uAddr); AssertRC(rc);
4871	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_TR_BASE, uTRBase); AssertRC(rc);
4872	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_FS_BASE, pVCpu->hmr0.s.vmx.RestoreHost.uHostFSBase); AssertRC(rc);
4873	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_GS_BASE, pVCpu->hmr0.s.vmx.RestoreHost.uHostGSBase); AssertRC(rc);
4874
4875	return VINF_SUCCESS;
4876	}
4877
4878
4879	/**
4880	* Exports certain host MSRs in the VM-exit MSR-load area and some in the
4881	* host-state area of the VMCS.
4882	*
4883	* These MSRs will be automatically restored on the host after every successful
4884	* VM-exit.
4885	*
4886	* @param pVCpu The cross context virtual CPU structure.
4887	*
4888	* @remarks No-long-jump zone!!!
4889	*/
4890	static void hmR0VmxExportHostMsrs(PVMCPUCC pVCpu)
4891	{
4892	AssertPtr(pVCpu);
4893
4894	/*
4895	* Save MSRs that we restore lazily (due to preemption or transition to ring-3)
4896	* rather than swapping them on every VM-entry.
4897	*/
4898	hmR0VmxLazySaveHostMsrs(pVCpu);
4899
4900	/*
4901	* Host Sysenter MSRs.
4902	*/
4903	int rc = VMXWriteVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, ASMRdMsr_Low(MSR_IA32_SYSENTER_CS)); AssertRC(rc);
4904	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP)); AssertRC(rc);
4905	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP)); AssertRC(rc);
4906
4907	/*
4908	* Host EFER MSR.
4909	*
4910	* If the CPU supports the newer VMCS controls for managing EFER, use it. Otherwise it's
4911	* done as part of auto-load/store MSR area in the VMCS, see hmR0VmxExportGuestMsrs().
4912	*/
4913	if (g_fHmVmxSupportsVmcsEfer)
4914	{
4915	rc = VMXWriteVmcs64(VMX_VMCS64_HOST_EFER_FULL, g_uHmVmxHostMsrEfer);
4916	AssertRC(rc);
4917	}
4918
4919	/** @todo IA32_PERF_GLOBALCTRL, IA32_PAT also see
4920	* hmR0VmxExportGuestEntryExitCtls(). */
4921	}
4922
4923
4924	/**
4925	* Figures out if we need to swap the EFER MSR which is particularly expensive.
4926	*
4927	* We check all relevant bits. For now, that's everything besides LMA/LME, as
4928	* these two bits are handled by VM-entry, see hmR0VMxExportGuestEntryExitCtls().
4929	*
4930	* @returns true if we need to load guest EFER, false otherwise.
4931	* @param pVCpu The cross context virtual CPU structure.
4932	* @param pVmxTransient The VMX-transient structure.
4933	*
4934	* @remarks Requires EFER, CR4.
4935	* @remarks No-long-jump zone!!!
4936	*/
4937	static bool hmR0VmxShouldSwapEferMsr(PCVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
4938	{
4939	#ifdef HMVMX_ALWAYS_SWAP_EFER
4940	RT_NOREF2(pVCpu, pVmxTransient);
4941	return true;
4942	#else
4943	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
4944	uint64_t const u64HostEfer = g_uHmVmxHostMsrEfer;
4945	uint64_t const u64GuestEfer = pCtx->msrEFER;
4946
4947	# ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4948	/*
4949	* For nested-guests, we shall honor swapping the EFER MSR when requested by
4950	* the nested-guest.
4951	*/
4952	if ( pVmxTransient->fIsNestedGuest
4953	&& ( CPUMIsGuestVmxEntryCtlsSet(pCtx, VMX_ENTRY_CTLS_LOAD_EFER_MSR)
4954	\|\| CPUMIsGuestVmxExitCtlsSet(pCtx, VMX_EXIT_CTLS_SAVE_EFER_MSR)
4955	\|\| CPUMIsGuestVmxExitCtlsSet(pCtx, VMX_EXIT_CTLS_LOAD_EFER_MSR)))
4956	return true;
4957	# else
4958	RT_NOREF(pVmxTransient);
4959	#endif
4960
4961	/*
4962	* For 64-bit guests, if EFER.SCE bit differs, we need to swap the EFER MSR
4963	* to ensure that the guest's SYSCALL behaviour isn't broken, see @bugref{7386}.
4964	*/
4965	if ( CPUMIsGuestInLongModeEx(pCtx)
4966	&& (u64GuestEfer & MSR_K6_EFER_SCE) != (u64HostEfer & MSR_K6_EFER_SCE))
4967	return true;
4968
4969	/*
4970	* If the guest uses PAE and EFER.NXE bit differs, we need to swap the EFER MSR
4971	* as it affects guest paging. 64-bit paging implies CR4.PAE as well.
4972	*
4973	* See Intel spec. 4.5 "IA-32e Paging".
4974	* See Intel spec. 4.1.1 "Three Paging Modes".
4975	*
4976	* Verify that we always intercept CR4.PAE and CR0.PG bits, so we don't need to
4977	* import CR4 and CR0 from the VMCS here as those bits are always up to date.
4978	*/
4979	Assert(hmR0VmxGetFixedCr4Mask(pVCpu) & X86_CR4_PAE);
4980	Assert(hmR0VmxGetFixedCr0Mask(pVCpu) & X86_CR0_PG);
4981	if ( (pCtx->cr4 & X86_CR4_PAE)
4982	&& (pCtx->cr0 & X86_CR0_PG))
4983	{
4984	/*
4985	* If nested paging is not used, verify that the guest paging mode matches the
4986	* shadow paging mode which is/will be placed in the VMCS (which is what will
4987	* actually be used while executing the guest and not the CR4 shadow value).
4988	*/
4989	AssertMsg( pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging
4990	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_PAE
4991	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_PAE_NX
4992	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_AMD64
4993	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_AMD64_NX,
4994	("enmShadowMode=%u\n", pVCpu->hm.s.enmShadowMode));
4995	if ((u64GuestEfer & MSR_K6_EFER_NXE) != (u64HostEfer & MSR_K6_EFER_NXE))
4996	{
4997	/* Verify that the host is NX capable. */
4998	Assert(pVCpu->CTX_SUFF(pVM)->cpum.ro.HostFeatures.fNoExecute);
4999	return true;
5000	}
5001	}
5002
5003	return false;
5004	#endif
5005	}
5006
5007
5008	/**
5009	* Exports the guest state with appropriate VM-entry and VM-exit controls in the
5010	* VMCS.
5011	*
5012	* This is typically required when the guest changes paging mode.
5013	*
5014	* @returns VBox status code.
5015	* @param pVCpu The cross context virtual CPU structure.
5016	* @param pVmxTransient The VMX-transient structure.
5017	*
5018	* @remarks Requires EFER.
5019	* @remarks No-long-jump zone!!!
5020	*/
5021	static int hmR0VmxExportGuestEntryExitCtls(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5022	{
5023	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_ENTRY_EXIT_CTLS)
5024	{
5025	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5026	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5027
5028	/*
5029	* VM-entry controls.
5030	*/
5031	{
5032	uint32_t fVal = g_HmMsrs.u.vmx.EntryCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
5033	uint32_t const fZap = g_HmMsrs.u.vmx.EntryCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
5034
5035	/*
5036	* Load the guest debug controls (DR7 and IA32_DEBUGCTL MSR) on VM-entry.
5037	* The first VT-x capable CPUs only supported the 1-setting of this bit.
5038	*
5039	* For nested-guests, this is a mandatory VM-entry control. It's also
5040	* required because we do not want to leak host bits to the nested-guest.
5041	*/
5042	fVal \|= VMX_ENTRY_CTLS_LOAD_DEBUG;
5043
5044	/*
5045	* Set if the guest is in long mode. This will set/clear the EFER.LMA bit on VM-entry.
5046	*
5047	* For nested-guests, the "IA-32e mode guest" control we initialize with what is
5048	* required to get the nested-guest working with hardware-assisted VMX execution.
5049	* It depends on the nested-guest's IA32_EFER.LMA bit. Remember, a nested hypervisor
5050	* can skip intercepting changes to the EFER MSR. This is why it needs to be done
5051	* here rather than while merging the guest VMCS controls.
5052	*/
5053	if (CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
5054	{
5055	Assert(pVCpu->cpum.GstCtx.msrEFER & MSR_K6_EFER_LME);
5056	fVal \|= VMX_ENTRY_CTLS_IA32E_MODE_GUEST;
5057	}
5058	else
5059	Assert(!(fVal & VMX_ENTRY_CTLS_IA32E_MODE_GUEST));
5060
5061	/*
5062	* If the CPU supports the newer VMCS controls for managing guest/host EFER, use it.
5063	*
5064	* For nested-guests, we use the "load IA32_EFER" if the hardware supports it,
5065	* regardless of whether the nested-guest VMCS specifies it because we are free to
5066	* load whatever MSRs we require and we do not need to modify the guest visible copy
5067	* of the VM-entry MSR load area.
5068	*/
5069	if ( g_fHmVmxSupportsVmcsEfer
5070	&& hmR0VmxShouldSwapEferMsr(pVCpu, pVmxTransient))
5071	fVal \|= VMX_ENTRY_CTLS_LOAD_EFER_MSR;
5072	else
5073	Assert(!(fVal & VMX_ENTRY_CTLS_LOAD_EFER_MSR));
5074
5075	/*
5076	* The following should -not- be set (since we're not in SMM mode):
5077	* - VMX_ENTRY_CTLS_ENTRY_TO_SMM
5078	* - VMX_ENTRY_CTLS_DEACTIVATE_DUAL_MON
5079	*/
5080
5081	/** @todo VMX_ENTRY_CTLS_LOAD_PERF_MSR,
5082	* VMX_ENTRY_CTLS_LOAD_PAT_MSR. */
5083
5084	if ((fVal & fZap) == fVal)
5085	{ /* likely */ }
5086	else
5087	{
5088	Log4Func(("Invalid VM-entry controls combo! Cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
5089	g_HmMsrs.u.vmx.EntryCtls.n.allowed0, fVal, fZap));
5090	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_ENTRY;
5091	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
5092	}
5093
5094	/* Commit it to the VMCS. */
5095	if (pVmcsInfo->u32EntryCtls != fVal)
5096	{
5097	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY, fVal);
5098	AssertRC(rc);
5099	pVmcsInfo->u32EntryCtls = fVal;
5100	}
5101	}
5102
5103	/*
5104	* VM-exit controls.
5105	*/
5106	{
5107	uint32_t fVal = g_HmMsrs.u.vmx.ExitCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
5108	uint32_t const fZap = g_HmMsrs.u.vmx.ExitCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
5109
5110	/*
5111	* Save debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x CPUs only
5112	* supported the 1-setting of this bit.
5113	*
5114	* For nested-guests, we set the "save debug controls" as the converse
5115	* "load debug controls" is mandatory for nested-guests anyway.
5116	*/
5117	fVal \|= VMX_EXIT_CTLS_SAVE_DEBUG;
5118
5119	/*
5120	* Set the host long mode active (EFER.LMA) bit (which Intel calls
5121	* "Host address-space size") if necessary. On VM-exit, VT-x sets both the
5122	* host EFER.LMA and EFER.LME bit to this value. See assertion in
5123	* hmR0VmxExportHostMsrs().
5124	*
5125	* For nested-guests, we always set this bit as we do not support 32-bit
5126	* hosts.
5127	*/
5128	fVal \|= VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE;
5129
5130	/*
5131	* If the VMCS EFER MSR fields are supported by the hardware, we use it.
5132	*
5133	* For nested-guests, we should use the "save IA32_EFER" control if we also
5134	* used the "load IA32_EFER" control while exporting VM-entry controls.
5135	*/
5136	if ( g_fHmVmxSupportsVmcsEfer
5137	&& hmR0VmxShouldSwapEferMsr(pVCpu, pVmxTransient))
5138	{
5139	fVal \|= VMX_EXIT_CTLS_SAVE_EFER_MSR
5140	\| VMX_EXIT_CTLS_LOAD_EFER_MSR;
5141	}
5142
5143	/*
5144	* Enable saving of the VMX-preemption timer value on VM-exit.
5145	* For nested-guests, currently not exposed/used.
5146	*/
5147	/** @todo r=bird: Measure performance hit because of this vs. always rewriting
5148	* the timer value. */
5149	if (pVM->hmr0.s.vmx.fUsePreemptTimer)
5150	{
5151	Assert(g_HmMsrs.u.vmx.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER);
5152	fVal \|= VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER;
5153	}
5154
5155	/* Don't acknowledge external interrupts on VM-exit. We want to let the host do that. */
5156	Assert(!(fVal & VMX_EXIT_CTLS_ACK_EXT_INT));
5157
5158	/** @todo VMX_EXIT_CTLS_LOAD_PERF_MSR,
5159	* VMX_EXIT_CTLS_SAVE_PAT_MSR,
5160	* VMX_EXIT_CTLS_LOAD_PAT_MSR. */
5161
5162	if ((fVal & fZap) == fVal)
5163	{ /* likely */ }
5164	else
5165	{
5166	Log4Func(("Invalid VM-exit controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%R#X32\n",
5167	g_HmMsrs.u.vmx.ExitCtls.n.allowed0, fVal, fZap));
5168	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_EXIT;
5169	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
5170	}
5171
5172	/* Commit it to the VMCS. */
5173	if (pVmcsInfo->u32ExitCtls != fVal)
5174	{
5175	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT, fVal);
5176	AssertRC(rc);
5177	pVmcsInfo->u32ExitCtls = fVal;
5178	}
5179	}
5180
5181	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
5182	}
5183	return VINF_SUCCESS;
5184	}
5185
5186
5187	/**
5188	* Sets the TPR threshold in the VMCS.
5189	*
5190	* @param pVmcsInfo The VMCS info. object.
5191	* @param u32TprThreshold The TPR threshold (task-priority class only).
5192	*/
5193	DECLINLINE(void) hmR0VmxApicSetTprThreshold(PVMXVMCSINFO pVmcsInfo, uint32_t u32TprThreshold)
5194	{
5195	Assert(!(u32TprThreshold & ~VMX_TPR_THRESHOLD_MASK)); /* Bits 31:4 MBZ. */
5196	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
5197	RT_NOREF(pVmcsInfo);
5198	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
5199	AssertRC(rc);
5200	}
5201
5202
5203	/**
5204	* Exports the guest APIC TPR state into the VMCS.
5205	*
5206	* @param pVCpu The cross context virtual CPU structure.
5207	* @param pVmxTransient The VMX-transient structure.
5208	*
5209	* @remarks No-long-jump zone!!!
5210	*/
5211	static void hmR0VmxExportGuestApicTpr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5212	{
5213	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_APIC_TPR)
5214	{
5215	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_APIC_TPR);
5216
5217	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5218	if (!pVmxTransient->fIsNestedGuest)
5219	{
5220	if ( PDMHasApic(pVCpu->CTX_SUFF(pVM))
5221	&& APICIsEnabled(pVCpu))
5222	{
5223	/*
5224	* Setup TPR shadowing.
5225	*/
5226	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
5227	{
5228	bool fPendingIntr = false;
5229	uint8_t u8Tpr = 0;
5230	uint8_t u8PendingIntr = 0;
5231	int rc = APICGetTpr(pVCpu, &u8Tpr, &fPendingIntr, &u8PendingIntr);
5232	AssertRC(rc);
5233
5234	/*
5235	* If there are interrupts pending but masked by the TPR, instruct VT-x to
5236	* cause a TPR-below-threshold VM-exit when the guest lowers its TPR below the
5237	* priority of the pending interrupt so we can deliver the interrupt. If there
5238	* are no interrupts pending, set threshold to 0 to not cause any
5239	* TPR-below-threshold VM-exits.
5240	*/
5241	uint32_t u32TprThreshold = 0;
5242	if (fPendingIntr)
5243	{
5244	/* Bits 3:0 of the TPR threshold field correspond to bits 7:4 of the TPR
5245	(which is the Task-Priority Class). */
5246	const uint8_t u8PendingPriority = u8PendingIntr >> 4;
5247	const uint8_t u8TprPriority = u8Tpr >> 4;
5248	if (u8PendingPriority <= u8TprPriority)
5249	u32TprThreshold = u8PendingPriority;
5250	}
5251
5252	hmR0VmxApicSetTprThreshold(pVmcsInfo, u32TprThreshold);
5253	}
5254	}
5255	}
5256	/* else: the TPR threshold has already been updated while merging the nested-guest VMCS. */
5257	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_APIC_TPR);
5258	}
5259	}
5260
5261
5262	/**
5263	* Gets the guest interruptibility-state and updates related force-flags.
5264	*
5265	* @returns Guest's interruptibility-state.
5266	* @param pVCpu The cross context virtual CPU structure.
5267	*
5268	* @remarks No-long-jump zone!!!
5269	*/
5270	static uint32_t hmR0VmxGetGuestIntrStateAndUpdateFFs(PVMCPUCC pVCpu)
5271	{
5272	/*
5273	* Check if we should inhibit interrupt delivery due to instructions like STI and MOV SS.
5274	*/
5275	uint32_t fIntrState = 0;
5276	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
5277	{
5278	/* If inhibition is active, RIP and RFLAGS should've been imported from the VMCS already. */
5279	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS);
5280
5281	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5282	if (pCtx->rip == EMGetInhibitInterruptsPC(pVCpu))
5283	{
5284	if (pCtx->eflags.Bits.u1IF)
5285	fIntrState = VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
5286	else
5287	fIntrState = VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS;
5288	}
5289	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
5290	{
5291	/*
5292	* We can clear the inhibit force flag as even if we go back to the recompiler
5293	* without executing guest code in VT-x, the flag's condition to be cleared is
5294	* met and thus the cleared state is correct.
5295	*/
5296	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
5297	}
5298	}
5299
5300	/*
5301	* Check if we should inhibit NMI delivery.
5302	*/
5303	if (CPUMIsGuestNmiBlocking(pVCpu))
5304	fIntrState \|= VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI;
5305
5306	/*
5307	* Validate.
5308	*/
5309	#ifdef VBOX_STRICT
5310	/* We don't support block-by-SMI yet.*/
5311	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI));
5312
5313	/* Block-by-STI must not be set when interrupts are disabled. */
5314	if (fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
5315	{
5316	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
5317	Assert(pVCpu->cpum.GstCtx.eflags.u & X86_EFL_IF);
5318	}
5319	#endif
5320
5321	return fIntrState;
5322	}
5323
5324
5325	/**
5326	* Exports the exception intercepts required for guest execution in the VMCS.
5327	*
5328	* @param pVCpu The cross context virtual CPU structure.
5329	* @param pVmxTransient The VMX-transient structure.
5330	*
5331	* @remarks No-long-jump zone!!!
5332	*/
5333	static void hmR0VmxExportGuestXcptIntercepts(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5334	{
5335	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_XCPT_INTERCEPTS)
5336	{
5337	/* When executing a nested-guest, we do not need to trap GIM hypercalls by intercepting #UD. */
5338	if ( !pVmxTransient->fIsNestedGuest
5339	&& pVCpu->hm.s.fGIMTrapXcptUD)
5340	hmR0VmxAddXcptIntercept(pVmxTransient, X86_XCPT_UD);
5341	else
5342	hmR0VmxRemoveXcptIntercept(pVCpu, pVmxTransient, X86_XCPT_UD);
5343
5344	/* Other exception intercepts are handled elsewhere, e.g. while exporting guest CR0. */
5345	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_XCPT_INTERCEPTS);
5346	}
5347	}
5348
5349
5350	/**
5351	* Exports the guest's RIP into the guest-state area in the VMCS.
5352	*
5353	* @param pVCpu The cross context virtual CPU structure.
5354	*
5355	* @remarks No-long-jump zone!!!
5356	*/
5357	static void hmR0VmxExportGuestRip(PVMCPUCC pVCpu)
5358	{
5359	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RIP)
5360	{
5361	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP);
5362
5363	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RIP, pVCpu->cpum.GstCtx.rip);
5364	AssertRC(rc);
5365
5366	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RIP);
5367	Log4Func(("rip=%#RX64\n", pVCpu->cpum.GstCtx.rip));
5368	}
5369	}
5370
5371
5372	/**
5373	* Exports the guest's RSP into the guest-state area in the VMCS.
5374	*
5375	* @param pVCpu The cross context virtual CPU structure.
5376	*
5377	* @remarks No-long-jump zone!!!
5378	*/
5379	static void hmR0VmxExportGuestRsp(PVMCPUCC pVCpu)
5380	{
5381	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RSP)
5382	{
5383	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RSP);
5384
5385	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RSP, pVCpu->cpum.GstCtx.rsp);
5386	AssertRC(rc);
5387
5388	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RSP);
5389	Log4Func(("rsp=%#RX64\n", pVCpu->cpum.GstCtx.rsp));
5390	}
5391	}
5392
5393
5394	/**
5395	* Exports the guest's RFLAGS into the guest-state area in the VMCS.
5396	*
5397	* @param pVCpu The cross context virtual CPU structure.
5398	* @param pVmxTransient The VMX-transient structure.
5399	*
5400	* @remarks No-long-jump zone!!!
5401	*/
5402	static void hmR0VmxExportGuestRflags(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5403	{
5404	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RFLAGS)
5405	{
5406	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
5407
5408	/* Intel spec. 2.3.1 "System Flags and Fields in IA-32e Mode" claims the upper 32-bits of RFLAGS are reserved (MBZ).
5409	Let us assert it as such and use 32-bit VMWRITE. */
5410	Assert(!RT_HI_U32(pVCpu->cpum.GstCtx.rflags.u64));
5411	X86EFLAGS fEFlags = pVCpu->cpum.GstCtx.eflags;
5412	Assert(fEFlags.u32 & X86_EFL_RA1_MASK);
5413	Assert(!(fEFlags.u32 & ~(X86_EFL_1 \| X86_EFL_LIVE_MASK)));
5414
5415	/*
5416	* If we're emulating real-mode using Virtual 8086 mode, save the real-mode eflags so
5417	* we can restore them on VM-exit. Modify the real-mode guest's eflags so that VT-x
5418	* can run the real-mode guest code under Virtual 8086 mode.
5419	*/
5420	PVMXVMCSINFOSHARED pVmcsInfo = pVmxTransient->pVmcsInfo->pShared;
5421	if (pVmcsInfo->RealMode.fRealOnV86Active)
5422	{
5423	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
5424	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
5425	Assert(!pVmxTransient->fIsNestedGuest);
5426	pVmcsInfo->RealMode.Eflags.u32 = fEFlags.u32; /* Save the original eflags of the real-mode guest. */
5427	fEFlags.Bits.u1VM = 1; /* Set the Virtual 8086 mode bit. */
5428	fEFlags.Bits.u2IOPL = 0; /* Change IOPL to 0, otherwise certain instructions won't fault. */
5429	}
5430
5431	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RFLAGS, fEFlags.u32);
5432	AssertRC(rc);
5433
5434	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RFLAGS);
5435	Log4Func(("eflags=%#RX32\n", fEFlags.u32));
5436	}
5437	}
5438
5439
5440	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5441	/**
5442	* Copies the nested-guest VMCS to the shadow VMCS.
5443	*
5444	* @returns VBox status code.
5445	* @param pVCpu The cross context virtual CPU structure.
5446	* @param pVmcsInfo The VMCS info. object.
5447	*
5448	* @remarks No-long-jump zone!!!
5449	*/
5450	static int hmR0VmxCopyNstGstToShadowVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
5451	{
5452	PVMCC const pVM = pVCpu->CTX_SUFF(pVM);
5453	PCVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
5454
5455	/*
5456	* Disable interrupts so we don't get preempted while the shadow VMCS is the
5457	* current VMCS, as we may try saving guest lazy MSRs.
5458	*
5459	* Strictly speaking the lazy MSRs are not in the VMCS, but I'd rather not risk
5460	* calling the import VMCS code which is currently performing the guest MSR reads
5461	* (on 64-bit hosts) and accessing the auto-load/store MSR area on 32-bit hosts
5462	* and the rest of the VMX leave session machinery.
5463	*/
5464	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
5465
5466	int rc = hmR0VmxLoadShadowVmcs(pVmcsInfo);
5467	if (RT_SUCCESS(rc))
5468	{
5469	/*
5470	* Copy all guest read/write VMCS fields.
5471	*
5472	* We don't check for VMWRITE failures here for performance reasons and
5473	* because they are not expected to fail, barring irrecoverable conditions
5474	* like hardware errors.
5475	*/
5476	uint32_t const cShadowVmcsFields = pVM->hmr0.s.vmx.cShadowVmcsFields;
5477	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
5478	{
5479	uint64_t u64Val;
5480	uint32_t const uVmcsField = pVM->hmr0.s.vmx.paShadowVmcsFields[i];
5481	IEMReadVmxVmcsField(pVmcsNstGst, uVmcsField, &u64Val);
5482	VMXWriteVmcs64(uVmcsField, u64Val);
5483	}
5484
5485	/*
5486	* If the host CPU supports writing all VMCS fields, copy the guest read-only
5487	* VMCS fields, so the guest can VMREAD them without causing a VM-exit.
5488	*/
5489	if (g_HmMsrs.u.vmx.u64Misc & VMX_MISC_VMWRITE_ALL)
5490	{
5491	uint32_t const cShadowVmcsRoFields = pVM->hmr0.s.vmx.cShadowVmcsRoFields;
5492	for (uint32_t i = 0; i < cShadowVmcsRoFields; i++)
5493	{
5494	uint64_t u64Val;
5495	uint32_t const uVmcsField = pVM->hmr0.s.vmx.paShadowVmcsRoFields[i];
5496	IEMReadVmxVmcsField(pVmcsNstGst, uVmcsField, &u64Val);
5497	VMXWriteVmcs64(uVmcsField, u64Val);
5498	}
5499	}
5500
5501	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
5502	rc \|= hmR0VmxLoadVmcs(pVmcsInfo);
5503	}
5504
5505	ASMSetFlags(fEFlags);
5506	return rc;
5507	}
5508
5509
5510	/**
5511	* Copies the shadow VMCS to the nested-guest VMCS.
5512	*
5513	* @returns VBox status code.
5514	* @param pVCpu The cross context virtual CPU structure.
5515	* @param pVmcsInfo The VMCS info. object.
5516	*
5517	* @remarks Called with interrupts disabled.
5518	*/
5519	static int hmR0VmxCopyShadowToNstGstVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
5520	{
5521	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
5522	PVMCC const pVM = pVCpu->CTX_SUFF(pVM);
5523	PVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
5524
5525	int rc = hmR0VmxLoadShadowVmcs(pVmcsInfo);
5526	if (RT_SUCCESS(rc))
5527	{
5528	/*
5529	* Copy guest read/write fields from the shadow VMCS.
5530	* Guest read-only fields cannot be modified, so no need to copy them.
5531	*
5532	* We don't check for VMREAD failures here for performance reasons and
5533	* because they are not expected to fail, barring irrecoverable conditions
5534	* like hardware errors.
5535	*/
5536	uint32_t const cShadowVmcsFields = pVM->hmr0.s.vmx.cShadowVmcsFields;
5537	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
5538	{
5539	uint64_t u64Val;
5540	uint32_t const uVmcsField = pVM->hmr0.s.vmx.paShadowVmcsFields[i];
5541	VMXReadVmcs64(uVmcsField, &u64Val);
5542	IEMWriteVmxVmcsField(pVmcsNstGst, uVmcsField, u64Val);
5543	}
5544
5545	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
5546	rc \|= hmR0VmxLoadVmcs(pVmcsInfo);
5547	}
5548	return rc;
5549	}
5550
5551
5552	/**
5553	* Enables VMCS shadowing for the given VMCS info. object.
5554	*
5555	* @param pVmcsInfo The VMCS info. object.
5556	*
5557	* @remarks No-long-jump zone!!!
5558	*/
5559	static void hmR0VmxEnableVmcsShadowing(PVMXVMCSINFO pVmcsInfo)
5560	{
5561	uint32_t uProcCtls2 = pVmcsInfo->u32ProcCtls2;
5562	if (!(uProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING))
5563	{
5564	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
5565	uProcCtls2 \|= VMX_PROC_CTLS2_VMCS_SHADOWING;
5566	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, uProcCtls2); AssertRC(rc);
5567	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, pVmcsInfo->HCPhysShadowVmcs); AssertRC(rc);
5568	pVmcsInfo->u32ProcCtls2 = uProcCtls2;
5569	pVmcsInfo->u64VmcsLinkPtr = pVmcsInfo->HCPhysShadowVmcs;
5570	Log4Func(("Enabled\n"));
5571	}
5572	}
5573
5574
5575	/**
5576	* Disables VMCS shadowing for the given VMCS info. object.
5577	*
5578	* @param pVmcsInfo The VMCS info. object.
5579	*
5580	* @remarks No-long-jump zone!!!
5581	*/
5582	static void hmR0VmxDisableVmcsShadowing(PVMXVMCSINFO pVmcsInfo)
5583	{
5584	/*
5585	* We want all VMREAD and VMWRITE instructions to cause VM-exits, so we clear the
5586	* VMCS shadowing control. However, VM-entry requires the shadow VMCS indicator bit
5587	* to match the VMCS shadowing control if the VMCS link pointer is not NIL_RTHCPHYS.
5588	* Hence, we must also reset the VMCS link pointer to ensure VM-entry does not fail.
5589	*
5590	* See Intel spec. 26.2.1.1 "VM-Execution Control Fields".
5591	* See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
5592	*/
5593	uint32_t uProcCtls2 = pVmcsInfo->u32ProcCtls2;
5594	if (uProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
5595	{
5596	uProcCtls2 &= ~VMX_PROC_CTLS2_VMCS_SHADOWING;
5597	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, uProcCtls2); AssertRC(rc);
5598	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS); AssertRC(rc);
5599	pVmcsInfo->u32ProcCtls2 = uProcCtls2;
5600	pVmcsInfo->u64VmcsLinkPtr = NIL_RTHCPHYS;
5601	Log4Func(("Disabled\n"));
5602	}
5603	}
5604	#endif
5605
5606
5607	/**
5608	* Exports the guest hardware-virtualization state.
5609	*
5610	* @returns VBox status code.
5611	* @param pVCpu The cross context virtual CPU structure.
5612	* @param pVmxTransient The VMX-transient structure.
5613	*
5614	* @remarks No-long-jump zone!!!
5615	*/
5616	static int hmR0VmxExportGuestHwvirtState(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5617	{
5618	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_HWVIRT)
5619	{
5620	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5621	/*
5622	* Check if the VMX feature is exposed to the guest and if the host CPU supports
5623	* VMCS shadowing.
5624	*/
5625	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUseVmcsShadowing)
5626	{
5627	/*
5628	* If the nested hypervisor has loaded a current VMCS and is in VMX root mode,
5629	* copy the nested hypervisor's current VMCS into the shadow VMCS and enable
5630	* VMCS shadowing to skip intercepting some or all VMREAD/VMWRITE VM-exits.
5631	*
5632	* We check for VMX root mode here in case the guest executes VMXOFF without
5633	* clearing the current VMCS pointer and our VMXOFF instruction emulation does
5634	* not clear the current VMCS pointer.
5635	*/
5636	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5637	if ( CPUMIsGuestInVmxRootMode(&pVCpu->cpum.GstCtx)
5638	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx)
5639	&& CPUMIsGuestVmxCurrentVmcsValid(&pVCpu->cpum.GstCtx))
5640	{
5641	/* Paranoia. */
5642	Assert(!pVmxTransient->fIsNestedGuest);
5643
5644	/*
5645	* For performance reasons, also check if the nested hypervisor's current VMCS
5646	* was newly loaded or modified before copying it to the shadow VMCS.
5647	*/
5648	if (!pVCpu->hm.s.vmx.fCopiedNstGstToShadowVmcs)
5649	{
5650	int rc = hmR0VmxCopyNstGstToShadowVmcs(pVCpu, pVmcsInfo);
5651	AssertRCReturn(rc, rc);
5652	pVCpu->hm.s.vmx.fCopiedNstGstToShadowVmcs = true;
5653	}
5654	hmR0VmxEnableVmcsShadowing(pVmcsInfo);
5655	}
5656	else
5657	hmR0VmxDisableVmcsShadowing(pVmcsInfo);
5658	}
5659	#else
5660	NOREF(pVmxTransient);
5661	#endif
5662	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_HWVIRT);
5663	}
5664	return VINF_SUCCESS;
5665	}
5666
5667
5668	/**
5669	* Exports the guest CR0 control register into the guest-state area in the VMCS.
5670	*
5671	* The guest FPU state is always pre-loaded hence we don't need to bother about
5672	* sharing FPU related CR0 bits between the guest and host.
5673	*
5674	* @returns VBox status code.
5675	* @param pVCpu The cross context virtual CPU structure.
5676	* @param pVmxTransient The VMX-transient structure.
5677	*
5678	* @remarks No-long-jump zone!!!
5679	*/
5680	static int hmR0VmxExportGuestCR0(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5681	{
5682	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR0)
5683	{
5684	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5685	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5686
5687	uint64_t fSetCr0 = g_HmMsrs.u.vmx.u64Cr0Fixed0;
5688	uint64_t const fZapCr0 = g_HmMsrs.u.vmx.u64Cr0Fixed1;
5689	if (pVM->hmr0.s.vmx.fUnrestrictedGuest)
5690	fSetCr0 &= ~(uint64_t)(X86_CR0_PE \| X86_CR0_PG);
5691	else
5692	Assert((fSetCr0 & (X86_CR0_PE \| X86_CR0_PG)) == (X86_CR0_PE \| X86_CR0_PG));
5693
5694	if (!pVmxTransient->fIsNestedGuest)
5695	{
5696	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
5697	uint64_t u64GuestCr0 = pVCpu->cpum.GstCtx.cr0;
5698	uint64_t const u64ShadowCr0 = u64GuestCr0;
5699	Assert(!RT_HI_U32(u64GuestCr0));
5700
5701	/*
5702	* Setup VT-x's view of the guest CR0.
5703	*/
5704	uint32_t uProcCtls = pVmcsInfo->u32ProcCtls;
5705	if (pVM->hmr0.s.fNestedPaging)
5706	{
5707	if (CPUMIsGuestPagingEnabled(pVCpu))
5708	{
5709	/* The guest has paging enabled, let it access CR3 without causing a VM-exit if supported. */
5710	uProcCtls &= ~( VMX_PROC_CTLS_CR3_LOAD_EXIT
5711	\| VMX_PROC_CTLS_CR3_STORE_EXIT);
5712	}
5713	else
5714	{
5715	/* The guest doesn't have paging enabled, make CR3 access cause a VM-exit to update our shadow. */
5716	uProcCtls \|= VMX_PROC_CTLS_CR3_LOAD_EXIT
5717	\| VMX_PROC_CTLS_CR3_STORE_EXIT;
5718	}
5719
5720	/* If we have unrestricted guest execution, we never have to intercept CR3 reads. */
5721	if (pVM->hmr0.s.vmx.fUnrestrictedGuest)
5722	uProcCtls &= ~VMX_PROC_CTLS_CR3_STORE_EXIT;
5723	}
5724	else
5725	{
5726	/* Guest CPL 0 writes to its read-only pages should cause a #PF VM-exit. */
5727	u64GuestCr0 \|= X86_CR0_WP;
5728	}
5729
5730	/*
5731	* Guest FPU bits.
5732	*
5733	* Since we pre-load the guest FPU always before VM-entry there is no need to track lazy state
5734	* using CR0.TS.
5735	*
5736	* Intel spec. 23.8 "Restrictions on VMX operation" mentions that CR0.NE bit must always be
5737	* set on the first CPUs to support VT-x and no mention of with regards to UX in VM-entry checks.
5738	*/
5739	u64GuestCr0 \|= X86_CR0_NE;
5740
5741	/* If CR0.NE isn't set, we need to intercept #MF exceptions and report them to the guest differently. */
5742	bool const fInterceptMF = !(u64ShadowCr0 & X86_CR0_NE);
5743
5744	/*
5745	* Update exception intercepts.
5746	*/
5747	uint32_t uXcptBitmap = pVmcsInfo->u32XcptBitmap;
5748	if (pVmcsInfo->pShared->RealMode.fRealOnV86Active)
5749	{
5750	Assert(PDMVmmDevHeapIsEnabled(pVM));
5751	Assert(pVM->hm.s.vmx.pRealModeTSS);
5752	uXcptBitmap \|= HMVMX_REAL_MODE_XCPT_MASK;
5753	}
5754	else
5755	{
5756	/* For now, cleared here as mode-switches can happen outside HM/VT-x. See @bugref{7626#c11}. */
5757	uXcptBitmap &= ~HMVMX_REAL_MODE_XCPT_MASK;
5758	if (fInterceptMF)
5759	uXcptBitmap \|= RT_BIT(X86_XCPT_MF);
5760	}
5761
5762	/* Additional intercepts for debugging, define these yourself explicitly. */
5763	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
5764	uXcptBitmap \|= 0
5765	\| RT_BIT(X86_XCPT_BP)
5766	\| RT_BIT(X86_XCPT_DE)
5767	\| RT_BIT(X86_XCPT_NM)
5768	\| RT_BIT(X86_XCPT_TS)
5769	\| RT_BIT(X86_XCPT_UD)
5770	\| RT_BIT(X86_XCPT_NP)
5771	\| RT_BIT(X86_XCPT_SS)
5772	\| RT_BIT(X86_XCPT_GP)
5773	\| RT_BIT(X86_XCPT_PF)
5774	\| RT_BIT(X86_XCPT_MF)
5775	;
5776	#elif defined(HMVMX_ALWAYS_TRAP_PF)
5777	uXcptBitmap \|= RT_BIT(X86_XCPT_PF);
5778	#endif
5779	if (pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv)
5780	uXcptBitmap \|= RT_BIT(X86_XCPT_GP);
5781	Assert(pVM->hmr0.s.fNestedPaging \|\| (uXcptBitmap & RT_BIT(X86_XCPT_PF)));
5782
5783	/* Apply the hardware specified CR0 fixed bits and enable caching. */
5784	u64GuestCr0 \|= fSetCr0;
5785	u64GuestCr0 &= fZapCr0;
5786	u64GuestCr0 &= ~(uint64_t)(X86_CR0_CD \| X86_CR0_NW);
5787
5788	/* Commit the CR0 and related fields to the guest VMCS. */
5789	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR0, u64GuestCr0); AssertRC(rc);
5790	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, u64ShadowCr0); AssertRC(rc);
5791	if (uProcCtls != pVmcsInfo->u32ProcCtls)
5792	{
5793	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, uProcCtls);
5794	AssertRC(rc);
5795	}
5796	if (uXcptBitmap != pVmcsInfo->u32XcptBitmap)
5797	{
5798	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
5799	AssertRC(rc);
5800	}
5801
5802	/* Update our caches. */
5803	pVmcsInfo->u32ProcCtls = uProcCtls;
5804	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
5805
5806	Log4Func(("cr0=%#RX64 shadow=%#RX64 set=%#RX64 zap=%#RX64\n", u64GuestCr0, u64ShadowCr0, fSetCr0, fZapCr0));
5807	}
5808	else
5809	{
5810	/*
5811	* With nested-guests, we may have extended the guest/host mask here since we
5812	* merged in the outer guest's mask. Thus, the merged mask can include more bits
5813	* (to read from the nested-guest CR0 read-shadow) than the nested hypervisor
5814	* originally supplied. We must copy those bits from the nested-guest CR0 into
5815	* the nested-guest CR0 read-shadow.
5816	*/
5817	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
5818	uint64_t u64GuestCr0 = pVCpu->cpum.GstCtx.cr0;
5819	uint64_t const u64ShadowCr0 = CPUMGetGuestVmxMaskedCr0(&pVCpu->cpum.GstCtx, pVmcsInfo->u64Cr0Mask);
5820	Assert(!RT_HI_U32(u64GuestCr0));
5821	Assert(u64GuestCr0 & X86_CR0_NE);
5822
5823	/* Apply the hardware specified CR0 fixed bits and enable caching. */
5824	u64GuestCr0 \|= fSetCr0;
5825	u64GuestCr0 &= fZapCr0;
5826	u64GuestCr0 &= ~(uint64_t)(X86_CR0_CD \| X86_CR0_NW);
5827
5828	/* Commit the CR0 and CR0 read-shadow to the nested-guest VMCS. */
5829	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR0, u64GuestCr0); AssertRC(rc);
5830	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, u64ShadowCr0); AssertRC(rc);
5831
5832	Log4Func(("cr0=%#RX64 shadow=%#RX64 (set=%#RX64 zap=%#RX64)\n", u64GuestCr0, u64ShadowCr0, fSetCr0, fZapCr0));
5833	}
5834
5835	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR0);
5836	}
5837
5838	return VINF_SUCCESS;
5839	}
5840
5841
5842	/**
5843	* Exports the guest control registers (CR3, CR4) into the guest-state area
5844	* in the VMCS.
5845	*
5846	* @returns VBox strict status code.
5847	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
5848	* without unrestricted guest access and the VMMDev is not presently
5849	* mapped (e.g. EFI32).
5850	*
5851	* @param pVCpu The cross context virtual CPU structure.
5852	* @param pVmxTransient The VMX-transient structure.
5853	*
5854	* @remarks No-long-jump zone!!!
5855	*/
5856	static VBOXSTRICTRC hmR0VmxExportGuestCR3AndCR4(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5857	{
5858	int rc = VINF_SUCCESS;
5859	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5860
5861	/*
5862	* Guest CR2.
5863	* It's always loaded in the assembler code. Nothing to do here.
5864	*/
5865
5866	/*
5867	* Guest CR3.
5868	*/
5869	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR3)
5870	{
5871	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR3);
5872
5873	if (pVM->hmr0.s.fNestedPaging)
5874	{
5875	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5876	pVmcsInfo->HCPhysEPTP = PGMGetHyperCR3(pVCpu);
5877
5878	/* Validate. See Intel spec. 28.2.2 "EPT Translation Mechanism" and 24.6.11 "Extended-Page-Table Pointer (EPTP)" */
5879	Assert(pVmcsInfo->HCPhysEPTP != NIL_RTHCPHYS);
5880	Assert(!(pVmcsInfo->HCPhysEPTP & UINT64_C(0xfff0000000000000)));
5881	Assert(!(pVmcsInfo->HCPhysEPTP & 0xfff));
5882
5883	/* VMX_EPT_MEMTYPE_WB support is already checked in hmR0VmxSetupTaggedTlb(). */
5884	pVmcsInfo->HCPhysEPTP \|= RT_BF_MAKE(VMX_BF_EPTP_MEMTYPE, VMX_EPTP_MEMTYPE_WB)
5885	\| RT_BF_MAKE(VMX_BF_EPTP_PAGE_WALK_LENGTH, VMX_EPTP_PAGE_WALK_LENGTH_4);
5886
5887	/* Validate. See Intel spec. 26.2.1 "Checks on VMX Controls" */
5888	AssertMsg( ((pVmcsInfo->HCPhysEPTP >> 3) & 0x07) == 3 /* Bits 3:5 (EPT page walk length - 1) must be 3. */
5889	&& ((pVmcsInfo->HCPhysEPTP >> 7) & 0x1f) == 0, /* Bits 7:11 MBZ. */
5890	("EPTP %#RX64\n", pVmcsInfo->HCPhysEPTP));
5891	AssertMsg( !((pVmcsInfo->HCPhysEPTP >> 6) & 0x01) /* Bit 6 (EPT accessed & dirty bit). */
5892	\|\| (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_ACCESS_DIRTY),
5893	("EPTP accessed/dirty bit not supported by CPU but set %#RX64\n", pVmcsInfo->HCPhysEPTP));
5894
5895	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, pVmcsInfo->HCPhysEPTP);
5896	AssertRC(rc);
5897
5898	uint64_t u64GuestCr3;
5899	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5900	if ( pVM->hmr0.s.vmx.fUnrestrictedGuest
5901	\|\| CPUMIsGuestPagingEnabledEx(pCtx))
5902	{
5903	/* If the guest is in PAE mode, pass the PDPEs to VT-x using the VMCS fields. */
5904	if (CPUMIsGuestInPAEModeEx(pCtx))
5905	{
5906	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, pCtx->aPaePdpes[0].u); AssertRC(rc);
5907	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, pCtx->aPaePdpes[1].u); AssertRC(rc);
5908	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, pCtx->aPaePdpes[2].u); AssertRC(rc);
5909	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, pCtx->aPaePdpes[3].u); AssertRC(rc);
5910	}
5911
5912	/*
5913	* The guest's view of its CR3 is unblemished with nested paging when the
5914	* guest is using paging or we have unrestricted guest execution to handle
5915	* the guest when it's not using paging.
5916	*/
5917	u64GuestCr3 = pCtx->cr3;
5918	}
5919	else
5920	{
5921	/*
5922	* The guest is not using paging, but the CPU (VT-x) has to. While the guest
5923	* thinks it accesses physical memory directly, we use our identity-mapped
5924	* page table to map guest-linear to guest-physical addresses. EPT takes care
5925	* of translating it to host-physical addresses.
5926	*/
5927	RTGCPHYS GCPhys;
5928	Assert(pVM->hm.s.vmx.pNonPagingModeEPTPageTable);
5929
5930	/* We obtain it here every time as the guest could have relocated this PCI region. */
5931	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys);
5932	if (RT_SUCCESS(rc))
5933	{ /* likely */ }
5934	else if (rc == VERR_PDM_DEV_HEAP_R3_TO_GCPHYS)
5935	{
5936	Log4Func(("VERR_PDM_DEV_HEAP_R3_TO_GCPHYS -> VINF_EM_RESCHEDULE_REM\n"));
5937	return VINF_EM_RESCHEDULE_REM; /* We cannot execute now, switch to REM/IEM till the guest maps in VMMDev. */
5938	}
5939	else
5940	AssertMsgFailedReturn(("%Rrc\n", rc), rc);
5941
5942	u64GuestCr3 = GCPhys;
5943	}
5944
5945	Log4Func(("guest_cr3=%#RX64 (GstN)\n", u64GuestCr3));
5946	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR3, u64GuestCr3);
5947	AssertRC(rc);
5948	}
5949	else
5950	{
5951	Assert(!pVmxTransient->fIsNestedGuest);
5952	/* Non-nested paging case, just use the hypervisor's CR3. */
5953	RTHCPHYS const HCPhysGuestCr3 = PGMGetHyperCR3(pVCpu);
5954
5955	Log4Func(("guest_cr3=%#RX64 (HstN)\n", HCPhysGuestCr3));
5956	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR3, HCPhysGuestCr3);
5957	AssertRC(rc);
5958	}
5959
5960	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR3);
5961	}
5962
5963	/*
5964	* Guest CR4.
5965	* ASSUMES this is done everytime we get in from ring-3! (XCR0)
5966	*/
5967	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR4)
5968	{
5969	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5970	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5971
5972	uint64_t const fSetCr4 = g_HmMsrs.u.vmx.u64Cr4Fixed0;
5973	uint64_t const fZapCr4 = g_HmMsrs.u.vmx.u64Cr4Fixed1;
5974
5975	/*
5976	* With nested-guests, we may have extended the guest/host mask here (since we
5977	* merged in the outer guest's mask, see hmR0VmxMergeVmcsNested). This means, the
5978	* mask can include more bits (to read from the nested-guest CR4 read-shadow) than
5979	* the nested hypervisor originally supplied. Thus, we should, in essence, copy
5980	* those bits from the nested-guest CR4 into the nested-guest CR4 read-shadow.
5981	*/
5982	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
5983	uint64_t u64GuestCr4 = pCtx->cr4;
5984	uint64_t const u64ShadowCr4 = !pVmxTransient->fIsNestedGuest
5985	? pCtx->cr4
5986	: CPUMGetGuestVmxMaskedCr4(pCtx, pVmcsInfo->u64Cr4Mask);
5987	Assert(!RT_HI_U32(u64GuestCr4));
5988
5989	/*
5990	* Setup VT-x's view of the guest CR4.
5991	*
5992	* If we're emulating real-mode using virtual-8086 mode, we want to redirect software
5993	* interrupts to the 8086 program interrupt handler. Clear the VME bit (the interrupt
5994	* redirection bitmap is already all 0, see hmR3InitFinalizeR0())
5995	*
5996	* See Intel spec. 20.2 "Software Interrupt Handling Methods While in Virtual-8086 Mode".
5997	*/
5998	if (pVmcsInfo->pShared->RealMode.fRealOnV86Active)
5999	{
6000	Assert(pVM->hm.s.vmx.pRealModeTSS);
6001	Assert(PDMVmmDevHeapIsEnabled(pVM));
6002	u64GuestCr4 &= ~(uint64_t)X86_CR4_VME;
6003	}
6004
6005	if (pVM->hmr0.s.fNestedPaging)
6006	{
6007	if ( !CPUMIsGuestPagingEnabledEx(pCtx)
6008	&& !pVM->hmr0.s.vmx.fUnrestrictedGuest)
6009	{
6010	/* We use 4 MB pages in our identity mapping page table when the guest doesn't have paging. */
6011	u64GuestCr4 \|= X86_CR4_PSE;
6012	/* Our identity mapping is a 32-bit page directory. */
6013	u64GuestCr4 &= ~(uint64_t)X86_CR4_PAE;
6014	}
6015	/* else use guest CR4.*/
6016	}
6017	else
6018	{
6019	Assert(!pVmxTransient->fIsNestedGuest);
6020
6021	/*
6022	* The shadow paging modes and guest paging modes are different, the shadow is in accordance with the host
6023	* paging mode and thus we need to adjust VT-x's view of CR4 depending on our shadow page tables.
6024	*/
6025	switch (pVCpu->hm.s.enmShadowMode)
6026	{
6027	case PGMMODE_REAL: /* Real-mode. */
6028	case PGMMODE_PROTECTED: /* Protected mode without paging. */
6029	case PGMMODE_32_BIT: /* 32-bit paging. */
6030	{
6031	u64GuestCr4 &= ~(uint64_t)X86_CR4_PAE;
6032	break;
6033	}
6034
6035	case PGMMODE_PAE: /* PAE paging. */
6036	case PGMMODE_PAE_NX: /* PAE paging with NX. */
6037	{
6038	u64GuestCr4 \|= X86_CR4_PAE;
6039	break;
6040	}
6041
6042	case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
6043	case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
6044	{
6045	#ifdef VBOX_WITH_64_BITS_GUESTS
6046	/* For our assumption in hmR0VmxShouldSwapEferMsr. */
6047	Assert(u64GuestCr4 & X86_CR4_PAE);
6048	break;
6049	#endif
6050	}
6051	default:
6052	AssertFailed();
6053	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
6054	}
6055	}
6056
6057	/* Apply the hardware specified CR4 fixed bits (mainly CR4.VMXE). */
6058	u64GuestCr4 \|= fSetCr4;
6059	u64GuestCr4 &= fZapCr4;
6060
6061	/* Commit the CR4 and CR4 read-shadow to the guest VMCS. */
6062	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR4, u64GuestCr4); AssertRC(rc);
6063	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, u64ShadowCr4); AssertRC(rc);
6064
6065	/* Whether to save/load/restore XCR0 during world switch depends on CR4.OSXSAVE and host+guest XCR0. */
6066	bool const fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
6067	if (fLoadSaveGuestXcr0 != pVCpu->hmr0.s.fLoadSaveGuestXcr0)
6068	{
6069	pVCpu->hmr0.s.fLoadSaveGuestXcr0 = fLoadSaveGuestXcr0;
6070	hmR0VmxUpdateStartVmFunction(pVCpu);
6071	}
6072
6073	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR4);
6074
6075	Log4Func(("cr4=%#RX64 shadow=%#RX64 (set=%#RX64 zap=%#RX64)\n", u64GuestCr4, u64ShadowCr4, fSetCr4, fZapCr4));
6076	}
6077	return rc;
6078	}
6079
6080
6081	/**
6082	* Exports the guest debug registers into the guest-state area in the VMCS.
6083	* The guest debug bits are partially shared with the host (e.g. DR6, DR0-3).
6084	*
6085	* This also sets up whether \#DB and MOV DRx accesses cause VM-exits.
6086	*
6087	* @returns VBox status code.
6088	* @param pVCpu The cross context virtual CPU structure.
6089	* @param pVmxTransient The VMX-transient structure.
6090	*
6091	* @remarks No-long-jump zone!!!
6092	*/
6093	static int hmR0VmxExportSharedDebugState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
6094	{
6095	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
6096
6097	/** @todo NSTVMX: Figure out what we want to do with nested-guest instruction
6098	* stepping. */
6099	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6100	if (pVmxTransient->fIsNestedGuest)
6101	{
6102	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, CPUMGetGuestDR7(pVCpu));
6103	AssertRC(rc);
6104
6105	/*
6106	* We don't want to always intercept MOV DRx for nested-guests as it causes
6107	* problems when the nested hypervisor isn't intercepting them, see @bugref{10080}.
6108	* Instead, they are strictly only requested when the nested hypervisor intercepts
6109	* them -- handled while merging VMCS controls.
6110	*
6111	* If neither the outer nor the nested-hypervisor is intercepting MOV DRx,
6112	* then the nested-guest debug state should be actively loaded on the host so that
6113	* nested-guest reads its own debug registers without causing VM-exits.
6114	*/
6115	if ( !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MOV_DR_EXIT)
6116	&& !CPUMIsGuestDebugStateActive(pVCpu))
6117	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
6118	return VINF_SUCCESS;
6119	}
6120
6121	#ifdef VBOX_STRICT
6122	/* Validate. Intel spec. 26.3.1.1 "Checks on Guest Controls Registers, Debug Registers, MSRs" */
6123	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
6124	{
6125	/* Validate. Intel spec. 17.2 "Debug Registers", recompiler paranoia checks. */
6126	Assert((pVCpu->cpum.GstCtx.dr[7] & (X86_DR7_MBZ_MASK \| X86_DR7_RAZ_MASK)) == 0);
6127	Assert((pVCpu->cpum.GstCtx.dr[7] & X86_DR7_RA1_MASK) == X86_DR7_RA1_MASK);
6128	}
6129	#endif
6130
6131	bool fSteppingDB = false;
6132	bool fInterceptMovDRx = false;
6133	uint32_t uProcCtls = pVmcsInfo->u32ProcCtls;
6134	if (pVCpu->hm.s.fSingleInstruction)
6135	{
6136	/* If the CPU supports the monitor trap flag, use it for single stepping in DBGF and avoid intercepting #DB. */
6137	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_MONITOR_TRAP_FLAG)
6138	{
6139	uProcCtls \|= VMX_PROC_CTLS_MONITOR_TRAP_FLAG;
6140	Assert(fSteppingDB == false);
6141	}
6142	else
6143	{
6144	pVCpu->cpum.GstCtx.eflags.u32 \|= X86_EFL_TF;
6145	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_GUEST_RFLAGS;
6146	pVCpu->hmr0.s.fClearTrapFlag = true;
6147	fSteppingDB = true;
6148	}
6149	}
6150
6151	uint64_t u64GuestDr7;
6152	if ( fSteppingDB
6153	\|\| (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK))
6154	{
6155	/*
6156	* Use the combined guest and host DRx values found in the hypervisor register set
6157	* because the hypervisor debugger has breakpoints active or someone is single stepping
6158	* on the host side without a monitor trap flag.
6159	*
6160	* Note! DBGF expects a clean DR6 state before executing guest code.
6161	*/
6162	if (!CPUMIsHyperDebugStateActive(pVCpu))
6163	{
6164	CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
6165	Assert(CPUMIsHyperDebugStateActive(pVCpu));
6166	Assert(!CPUMIsGuestDebugStateActive(pVCpu));
6167	}
6168
6169	/* Update DR7 with the hypervisor value (other DRx registers are handled by CPUM one way or another). */
6170	u64GuestDr7 = CPUMGetHyperDR7(pVCpu);
6171	pVCpu->hmr0.s.fUsingHyperDR7 = true;
6172	fInterceptMovDRx = true;
6173	}
6174	else
6175	{
6176	/*
6177	* If the guest has enabled debug registers, we need to load them prior to
6178	* executing guest code so they'll trigger at the right time.
6179	*/
6180	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_DR7);
6181	if (pVCpu->cpum.GstCtx.dr[7] & (X86_DR7_ENABLED_MASK \| X86_DR7_GD))
6182	{
6183	if (!CPUMIsGuestDebugStateActive(pVCpu))
6184	{
6185	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
6186	Assert(CPUMIsGuestDebugStateActive(pVCpu));
6187	Assert(!CPUMIsHyperDebugStateActive(pVCpu));
6188	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
6189	}
6190	Assert(!fInterceptMovDRx);
6191	}
6192	else if (!CPUMIsGuestDebugStateActive(pVCpu))
6193	{
6194	/*
6195	* If no debugging enabled, we'll lazy load DR0-3. Unlike on AMD-V, we
6196	* must intercept #DB in order to maintain a correct DR6 guest value, and
6197	* because we need to intercept it to prevent nested #DBs from hanging the
6198	* CPU, we end up always having to intercept it. See hmR0VmxSetupVmcsXcptBitmap().
6199	*/
6200	fInterceptMovDRx = true;
6201	}
6202
6203	/* Update DR7 with the actual guest value. */
6204	u64GuestDr7 = pVCpu->cpum.GstCtx.dr[7];
6205	pVCpu->hmr0.s.fUsingHyperDR7 = false;
6206	}
6207
6208	if (fInterceptMovDRx)
6209	uProcCtls \|= VMX_PROC_CTLS_MOV_DR_EXIT;
6210	else
6211	uProcCtls &= ~VMX_PROC_CTLS_MOV_DR_EXIT;
6212
6213	/*
6214	* Update the processor-based VM-execution controls with the MOV-DRx intercepts and the
6215	* monitor-trap flag and update our cache.
6216	*/
6217	if (uProcCtls != pVmcsInfo->u32ProcCtls)
6218	{
6219	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, uProcCtls);
6220	AssertRC(rc);
6221	pVmcsInfo->u32ProcCtls = uProcCtls;
6222	}
6223
6224	/*
6225	* Update guest DR7.
6226	*/
6227	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, u64GuestDr7);
6228	AssertRC(rc);
6229
6230	/*
6231	* If we have forced EFLAGS.TF to be set because we're single-stepping in the hypervisor debugger,
6232	* we need to clear interrupt inhibition if any as otherwise it causes a VM-entry failure.
6233	*
6234	* See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
6235	*/
6236	if (fSteppingDB)
6237	{
6238	Assert(pVCpu->hm.s.fSingleInstruction);
6239	Assert(pVCpu->cpum.GstCtx.eflags.Bits.u1TF);
6240
6241	uint32_t fIntrState = 0;
6242	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
6243	AssertRC(rc);
6244
6245	if (fIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
6246	{
6247	fIntrState &= ~(VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
6248	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
6249	AssertRC(rc);
6250	}
6251	}
6252
6253	return VINF_SUCCESS;
6254	}
6255
6256
6257	#ifdef VBOX_STRICT
6258	/**
6259	* Strict function to validate segment registers.
6260	*
6261	* @param pVCpu The cross context virtual CPU structure.
6262	* @param pVmcsInfo The VMCS info. object.
6263	*
6264	* @remarks Will import guest CR0 on strict builds during validation of
6265	* segments.
6266	*/
6267	static void hmR0VmxValidateSegmentRegs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
6268	{
6269	/*
6270	* Validate segment registers. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
6271	*
6272	* The reason we check for attribute value 0 in this function and not just the unusable bit is
6273	* because hmR0VmxExportGuestSegReg() only updates the VMCS' copy of the value with the
6274	* unusable bit and doesn't change the guest-context value.
6275	*/
6276	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6277	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6278	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR0);
6279	if ( !pVM->hmr0.s.vmx.fUnrestrictedGuest
6280	&& ( !CPUMIsGuestInRealModeEx(pCtx)
6281	&& !CPUMIsGuestInV86ModeEx(pCtx)))
6282	{
6283	/* Protected mode checks */
6284	/* CS */
6285	Assert(pCtx->cs.Attr.n.u1Present);
6286	Assert(!(pCtx->cs.Attr.u & 0xf00));
6287	Assert(!(pCtx->cs.Attr.u & 0xfffe0000));
6288	Assert( (pCtx->cs.u32Limit & 0xfff) == 0xfff
6289	\|\| !(pCtx->cs.Attr.n.u1Granularity));
6290	Assert( !(pCtx->cs.u32Limit & 0xfff00000)
6291	\|\| (pCtx->cs.Attr.n.u1Granularity));
6292	/* CS cannot be loaded with NULL in protected mode. */
6293	Assert(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE)); /** @todo is this really true even for 64-bit CS? */
6294	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
6295	Assert(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl);
6296	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
6297	Assert(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl);
6298	else
6299	AssertMsgFailed(("Invalid CS Type %#x\n", pCtx->cs.Attr.n.u2Dpl));
6300	/* SS */
6301	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
6302	Assert(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL));
6303	if ( !(pCtx->cr0 & X86_CR0_PE)
6304	\|\| pCtx->cs.Attr.n.u4Type == 3)
6305	{
6306	Assert(!pCtx->ss.Attr.n.u2Dpl);
6307	}
6308	if (pCtx->ss.Attr.u && !(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
6309	{
6310	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
6311	Assert(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7);
6312	Assert(pCtx->ss.Attr.n.u1Present);
6313	Assert(!(pCtx->ss.Attr.u & 0xf00));
6314	Assert(!(pCtx->ss.Attr.u & 0xfffe0000));
6315	Assert( (pCtx->ss.u32Limit & 0xfff) == 0xfff
6316	\|\| !(pCtx->ss.Attr.n.u1Granularity));
6317	Assert( !(pCtx->ss.u32Limit & 0xfff00000)
6318	\|\| (pCtx->ss.Attr.n.u1Granularity));
6319	}
6320	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxExportGuestSegReg(). */
6321	if (pCtx->ds.Attr.u && !(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
6322	{
6323	Assert(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6324	Assert(pCtx->ds.Attr.n.u1Present);
6325	Assert(pCtx->ds.Attr.n.u4Type > 11 \|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL));
6326	Assert(!(pCtx->ds.Attr.u & 0xf00));
6327	Assert(!(pCtx->ds.Attr.u & 0xfffe0000));
6328	Assert( (pCtx->ds.u32Limit & 0xfff) == 0xfff
6329	\|\| !(pCtx->ds.Attr.n.u1Granularity));
6330	Assert( !(pCtx->ds.u32Limit & 0xfff00000)
6331	\|\| (pCtx->ds.Attr.n.u1Granularity));
6332	Assert( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6333	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ));
6334	}
6335	if (pCtx->es.Attr.u && !(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
6336	{
6337	Assert(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6338	Assert(pCtx->es.Attr.n.u1Present);
6339	Assert(pCtx->es.Attr.n.u4Type > 11 \|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL));
6340	Assert(!(pCtx->es.Attr.u & 0xf00));
6341	Assert(!(pCtx->es.Attr.u & 0xfffe0000));
6342	Assert( (pCtx->es.u32Limit & 0xfff) == 0xfff
6343	\|\| !(pCtx->es.Attr.n.u1Granularity));
6344	Assert( !(pCtx->es.u32Limit & 0xfff00000)
6345	\|\| (pCtx->es.Attr.n.u1Granularity));
6346	Assert( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6347	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ));
6348	}
6349	if (pCtx->fs.Attr.u && !(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
6350	{
6351	Assert(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6352	Assert(pCtx->fs.Attr.n.u1Present);
6353	Assert(pCtx->fs.Attr.n.u4Type > 11 \|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL));
6354	Assert(!(pCtx->fs.Attr.u & 0xf00));
6355	Assert(!(pCtx->fs.Attr.u & 0xfffe0000));
6356	Assert( (pCtx->fs.u32Limit & 0xfff) == 0xfff
6357	\|\| !(pCtx->fs.Attr.n.u1Granularity));
6358	Assert( !(pCtx->fs.u32Limit & 0xfff00000)
6359	\|\| (pCtx->fs.Attr.n.u1Granularity));
6360	Assert( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6361	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ));
6362	}
6363	if (pCtx->gs.Attr.u && !(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
6364	{
6365	Assert(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6366	Assert(pCtx->gs.Attr.n.u1Present);
6367	Assert(pCtx->gs.Attr.n.u4Type > 11 \|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL));
6368	Assert(!(pCtx->gs.Attr.u & 0xf00));
6369	Assert(!(pCtx->gs.Attr.u & 0xfffe0000));
6370	Assert( (pCtx->gs.u32Limit & 0xfff) == 0xfff
6371	\|\| !(pCtx->gs.Attr.n.u1Granularity));
6372	Assert( !(pCtx->gs.u32Limit & 0xfff00000)
6373	\|\| (pCtx->gs.Attr.n.u1Granularity));
6374	Assert( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6375	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ));
6376	}
6377	/* 64-bit capable CPUs. */
6378	Assert(!RT_HI_U32(pCtx->cs.u64Base));
6379	Assert(!pCtx->ss.Attr.u \|\| !RT_HI_U32(pCtx->ss.u64Base));
6380	Assert(!pCtx->ds.Attr.u \|\| !RT_HI_U32(pCtx->ds.u64Base));
6381	Assert(!pCtx->es.Attr.u \|\| !RT_HI_U32(pCtx->es.u64Base));
6382	}
6383	else if ( CPUMIsGuestInV86ModeEx(pCtx)
6384	\|\| ( CPUMIsGuestInRealModeEx(pCtx)
6385	&& !pVM->hmr0.s.vmx.fUnrestrictedGuest))
6386	{
6387	/* Real and v86 mode checks. */
6388	/* hmR0VmxExportGuestSegReg() writes the modified in VMCS. We want what we're feeding to VT-x. */
6389	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
6390	if (pVmcsInfo->pShared->RealMode.fRealOnV86Active)
6391	{
6392	u32CSAttr = 0xf3; u32SSAttr = 0xf3; u32DSAttr = 0xf3;
6393	u32ESAttr = 0xf3; u32FSAttr = 0xf3; u32GSAttr = 0xf3;
6394	}
6395	else
6396	{
6397	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u; u32DSAttr = pCtx->ds.Attr.u;
6398	u32ESAttr = pCtx->es.Attr.u; u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
6399	}
6400
6401	/* CS */
6402	AssertMsg((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), ("CS base %#x %#x\n", pCtx->cs.u64Base, pCtx->cs.Sel));
6403	Assert(pCtx->cs.u32Limit == 0xffff);
6404	Assert(u32CSAttr == 0xf3);
6405	/* SS */
6406	Assert(pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4);
6407	Assert(pCtx->ss.u32Limit == 0xffff);
6408	Assert(u32SSAttr == 0xf3);
6409	/* DS */
6410	Assert(pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4);
6411	Assert(pCtx->ds.u32Limit == 0xffff);
6412	Assert(u32DSAttr == 0xf3);
6413	/* ES */
6414	Assert(pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4);
6415	Assert(pCtx->es.u32Limit == 0xffff);
6416	Assert(u32ESAttr == 0xf3);
6417	/* FS */
6418	Assert(pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4);
6419	Assert(pCtx->fs.u32Limit == 0xffff);
6420	Assert(u32FSAttr == 0xf3);
6421	/* GS */
6422	Assert(pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4);
6423	Assert(pCtx->gs.u32Limit == 0xffff);
6424	Assert(u32GSAttr == 0xf3);
6425	/* 64-bit capable CPUs. */
6426	Assert(!RT_HI_U32(pCtx->cs.u64Base));
6427	Assert(!u32SSAttr \|\| !RT_HI_U32(pCtx->ss.u64Base));
6428	Assert(!u32DSAttr \|\| !RT_HI_U32(pCtx->ds.u64Base));
6429	Assert(!u32ESAttr \|\| !RT_HI_U32(pCtx->es.u64Base));
6430	}
6431	}
6432	#endif /* VBOX_STRICT */
6433
6434
6435	/**
6436	* Exports a guest segment register into the guest-state area in the VMCS.
6437	*
6438	* @returns VBox status code.
6439	* @param pVCpu The cross context virtual CPU structure.
6440	* @param pVmcsInfo The VMCS info. object.
6441	* @param iSegReg The segment register number (X86_SREG_XXX).
6442	* @param pSelReg Pointer to the segment selector.
6443	*
6444	* @remarks No-long-jump zone!!!
6445	*/
6446	static int hmR0VmxExportGuestSegReg(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, uint32_t iSegReg, PCCPUMSELREG pSelReg)
6447	{
6448	Assert(iSegReg < X86_SREG_COUNT);
6449
6450	uint32_t u32Access = pSelReg->Attr.u;
6451	if (!pVmcsInfo->pShared->RealMode.fRealOnV86Active)
6452	{
6453	/*
6454	* The way to differentiate between whether this is really a null selector or was just
6455	* a selector loaded with 0 in real-mode is using the segment attributes. A selector
6456	* loaded in real-mode with the value 0 is valid and usable in protected-mode and we
6457	* should -not- mark it as an unusable segment. Both the recompiler & VT-x ensures
6458	* NULL selectors loaded in protected-mode have their attribute as 0.
6459	*/
6460	if (u32Access)
6461	{ }
6462	else
6463	u32Access = X86DESCATTR_UNUSABLE;
6464	}
6465	else
6466	{
6467	/* VT-x requires our real-using-v86 mode hack to override the segment access-right bits. */
6468	u32Access = 0xf3;
6469	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
6470	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
6471	RT_NOREF_PV(pVCpu);
6472	}
6473
6474	/* Validate segment access rights. Refer to Intel spec. "26.3.1.2 Checks on Guest Segment Registers". */
6475	AssertMsg((u32Access & X86DESCATTR_UNUSABLE) \|\| (u32Access & X86_SEL_TYPE_ACCESSED),
6476	("Access bit not set for usable segment. %.2s sel=%#x attr %#x\n", "ESCSSSDSFSGS" + iSegReg * 2, pSelReg, pSelReg->Attr.u));
6477
6478	/*
6479	* Commit it to the VMCS.
6480	*/
6481	Assert((uint32_t)VMX_VMCS16_GUEST_SEG_SEL(iSegReg) == g_aVmcsSegSel[iSegReg]);
6482	Assert((uint32_t)VMX_VMCS32_GUEST_SEG_LIMIT(iSegReg) == g_aVmcsSegLimit[iSegReg]);
6483	Assert((uint32_t)VMX_VMCS32_GUEST_SEG_ACCESS_RIGHTS(iSegReg) == g_aVmcsSegAttr[iSegReg]);
6484	Assert((uint32_t)VMX_VMCS_GUEST_SEG_BASE(iSegReg) == g_aVmcsSegBase[iSegReg]);
6485	int rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_SEG_SEL(iSegReg), pSelReg->Sel); AssertRC(rc);
6486	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SEG_LIMIT(iSegReg), pSelReg->u32Limit); AssertRC(rc);
6487	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SEG_BASE(iSegReg), pSelReg->u64Base); AssertRC(rc);
6488	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SEG_ACCESS_RIGHTS(iSegReg), u32Access); AssertRC(rc);
6489	return VINF_SUCCESS;
6490	}
6491
6492
6493	/**
6494	* Exports the guest segment registers, GDTR, IDTR, LDTR, TR into the guest-state
6495	* area in the VMCS.
6496	*
6497	* @returns VBox status code.
6498	* @param pVCpu The cross context virtual CPU structure.
6499	* @param pVmxTransient The VMX-transient structure.
6500	*
6501	* @remarks Will import guest CR0 on strict builds during validation of
6502	* segments.
6503	* @remarks No-long-jump zone!!!
6504	*/
6505	static int hmR0VmxExportGuestSegRegsXdtr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6506	{
6507	int rc = VERR_INTERNAL_ERROR_5;
6508	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6509	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6510	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6511	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
6512
6513	/*
6514	* Guest Segment registers: CS, SS, DS, ES, FS, GS.
6515	*/
6516	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SREG_MASK)
6517	{
6518	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CS)
6519	{
6520	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CS);
6521	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6522	pVmcsInfoShared->RealMode.AttrCS.u = pCtx->cs.Attr.u;
6523	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_CS, &pCtx->cs);
6524	AssertRC(rc);
6525	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CS);
6526	}
6527
6528	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SS)
6529	{
6530	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SS);
6531	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6532	pVmcsInfoShared->RealMode.AttrSS.u = pCtx->ss.Attr.u;
6533	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_SS, &pCtx->ss);
6534	AssertRC(rc);
6535	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SS);
6536	}
6537
6538	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_DS)
6539	{
6540	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_DS);
6541	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6542	pVmcsInfoShared->RealMode.AttrDS.u = pCtx->ds.Attr.u;
6543	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_DS, &pCtx->ds);
6544	AssertRC(rc);
6545	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_DS);
6546	}
6547
6548	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_ES)
6549	{
6550	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_ES);
6551	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6552	pVmcsInfoShared->RealMode.AttrES.u = pCtx->es.Attr.u;
6553	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_ES, &pCtx->es);
6554	AssertRC(rc);
6555	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_ES);
6556	}
6557
6558	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_FS)
6559	{
6560	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_FS);
6561	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6562	pVmcsInfoShared->RealMode.AttrFS.u = pCtx->fs.Attr.u;
6563	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_FS, &pCtx->fs);
6564	AssertRC(rc);
6565	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_FS);
6566	}
6567
6568	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_GS)
6569	{
6570	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GS);
6571	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6572	pVmcsInfoShared->RealMode.AttrGS.u = pCtx->gs.Attr.u;
6573	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_GS, &pCtx->gs);
6574	AssertRC(rc);
6575	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_GS);
6576	}
6577
6578	#ifdef VBOX_STRICT
6579	hmR0VmxValidateSegmentRegs(pVCpu, pVmcsInfo);
6580	#endif
6581	Log4Func(("cs={%#04x base=%#RX64 limit=%#RX32 attr=%#RX32}\n", pCtx->cs.Sel, pCtx->cs.u64Base, pCtx->cs.u32Limit,
6582	pCtx->cs.Attr.u));
6583	}
6584
6585	/*
6586	* Guest TR.
6587	*/
6588	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_TR)
6589	{
6590	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_TR);
6591
6592	/*
6593	* Real-mode emulation using virtual-8086 mode with CR4.VME. Interrupt redirection is
6594	* achieved using the interrupt redirection bitmap (all bits cleared to let the guest
6595	* handle INT-n's) in the TSS. See hmR3InitFinalizeR0() to see how pRealModeTSS is setup.
6596	*/
6597	uint16_t u16Sel;
6598	uint32_t u32Limit;
6599	uint64_t u64Base;
6600	uint32_t u32AccessRights;
6601	if (!pVmcsInfoShared->RealMode.fRealOnV86Active)
6602	{
6603	u16Sel = pCtx->tr.Sel;
6604	u32Limit = pCtx->tr.u32Limit;
6605	u64Base = pCtx->tr.u64Base;
6606	u32AccessRights = pCtx->tr.Attr.u;
6607	}
6608	else
6609	{
6610	Assert(!pVmxTransient->fIsNestedGuest);
6611	Assert(pVM->hm.s.vmx.pRealModeTSS);
6612	Assert(PDMVmmDevHeapIsEnabled(pVM)); /* Guaranteed by HMCanExecuteGuest() -XXX- what about inner loop changes? */
6613
6614	/* We obtain it here every time as PCI regions could be reconfigured in the guest, changing the VMMDev base. */
6615	RTGCPHYS GCPhys;
6616	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys);
6617	AssertRCReturn(rc, rc);
6618
6619	X86DESCATTR DescAttr;
6620	DescAttr.u = 0;
6621	DescAttr.n.u1Present = 1;
6622	DescAttr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
6623
6624	u16Sel = 0;
6625	u32Limit = HM_VTX_TSS_SIZE;
6626	u64Base = GCPhys;
6627	u32AccessRights = DescAttr.u;
6628	}
6629
6630	/* Validate. */
6631	Assert(!(u16Sel & RT_BIT(2)));
6632	AssertMsg( (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_386_TSS_BUSY
6633	\|\| (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_286_TSS_BUSY, ("TSS is not busy!? %#x\n", u32AccessRights));
6634	AssertMsg(!(u32AccessRights & X86DESCATTR_UNUSABLE), ("TR unusable bit is not clear!? %#x\n", u32AccessRights));
6635	Assert(!(u32AccessRights & RT_BIT(4))); /* System MBZ.*/
6636	Assert(u32AccessRights & RT_BIT(7)); /* Present MB1.*/
6637	Assert(!(u32AccessRights & 0xf00)); /* 11:8 MBZ. */
6638	Assert(!(u32AccessRights & 0xfffe0000)); /* 31:17 MBZ. */
6639	Assert( (u32Limit & 0xfff) == 0xfff
6640	\|\| !(u32AccessRights & RT_BIT(15))); /* Granularity MBZ. */
6641	Assert( !(pCtx->tr.u32Limit & 0xfff00000)
6642	\|\| (u32AccessRights & RT_BIT(15))); /* Granularity MB1. */
6643
6644	rc = VMXWriteVmcs16(VMX_VMCS16_GUEST_TR_SEL, u16Sel); AssertRC(rc);
6645	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, u32Limit); AssertRC(rc);
6646	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, u32AccessRights); AssertRC(rc);
6647	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_TR_BASE, u64Base); AssertRC(rc);
6648
6649	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_TR);
6650	Log4Func(("tr base=%#RX64 limit=%#RX32\n", pCtx->tr.u64Base, pCtx->tr.u32Limit));
6651	}
6652
6653	/*
6654	* Guest GDTR.
6655	*/
6656	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_GDTR)
6657	{
6658	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GDTR);
6659
6660	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, pCtx->gdtr.cbGdt); AssertRC(rc);
6661	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, pCtx->gdtr.pGdt); AssertRC(rc);
6662
6663	/* Validate. */
6664	Assert(!(pCtx->gdtr.cbGdt & 0xffff0000)); /* Bits 31:16 MBZ. */
6665
6666	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_GDTR);
6667	Log4Func(("gdtr base=%#RX64 limit=%#RX32\n", pCtx->gdtr.pGdt, pCtx->gdtr.cbGdt));
6668	}
6669
6670	/*
6671	* Guest LDTR.
6672	*/
6673	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_LDTR)
6674	{
6675	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_LDTR);
6676
6677	/* The unusable bit is specific to VT-x, if it's a null selector mark it as an unusable segment. */
6678	uint32_t u32Access;
6679	if ( !pVmxTransient->fIsNestedGuest
6680	&& !pCtx->ldtr.Attr.u)
6681	u32Access = X86DESCATTR_UNUSABLE;
6682	else
6683	u32Access = pCtx->ldtr.Attr.u;
6684
6685	rc = VMXWriteVmcs16(VMX_VMCS16_GUEST_LDTR_SEL, pCtx->ldtr.Sel); AssertRC(rc);
6686	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, pCtx->ldtr.u32Limit); AssertRC(rc);
6687	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, u32Access); AssertRC(rc);
6688	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_LDTR_BASE, pCtx->ldtr.u64Base); AssertRC(rc);
6689
6690	/* Validate. */
6691	if (!(u32Access & X86DESCATTR_UNUSABLE))
6692	{
6693	Assert(!(pCtx->ldtr.Sel & RT_BIT(2))); /* TI MBZ. */
6694	Assert(pCtx->ldtr.Attr.n.u4Type == 2); /* Type MB2 (LDT). */
6695	Assert(!pCtx->ldtr.Attr.n.u1DescType); /* System MBZ. */
6696	Assert(pCtx->ldtr.Attr.n.u1Present == 1); /* Present MB1. */
6697	Assert(!pCtx->ldtr.Attr.n.u4LimitHigh); /* 11:8 MBZ. */
6698	Assert(!(pCtx->ldtr.Attr.u & 0xfffe0000)); /* 31:17 MBZ. */
6699	Assert( (pCtx->ldtr.u32Limit & 0xfff) == 0xfff
6700	\|\| !pCtx->ldtr.Attr.n.u1Granularity); /* Granularity MBZ. */
6701	Assert( !(pCtx->ldtr.u32Limit & 0xfff00000)
6702	\|\| pCtx->ldtr.Attr.n.u1Granularity); /* Granularity MB1. */
6703	}
6704
6705	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_LDTR);
6706	Log4Func(("ldtr base=%#RX64 limit=%#RX32\n", pCtx->ldtr.u64Base, pCtx->ldtr.u32Limit));
6707	}
6708
6709	/*
6710	* Guest IDTR.
6711	*/
6712	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_IDTR)
6713	{
6714	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_IDTR);
6715
6716	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, pCtx->idtr.cbIdt); AssertRC(rc);
6717	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, pCtx->idtr.pIdt); AssertRC(rc);
6718
6719	/* Validate. */
6720	Assert(!(pCtx->idtr.cbIdt & 0xffff0000)); /* Bits 31:16 MBZ. */
6721
6722	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_IDTR);
6723	Log4Func(("idtr base=%#RX64 limit=%#RX32\n", pCtx->idtr.pIdt, pCtx->idtr.cbIdt));
6724	}
6725
6726	return VINF_SUCCESS;
6727	}
6728
6729
6730	/**
6731	* Exports certain guest MSRs into the VM-entry MSR-load and VM-exit MSR-store
6732	* areas.
6733	*
6734	* These MSRs will automatically be loaded to the host CPU on every successful
6735	* VM-entry and stored from the host CPU on every successful VM-exit.
6736	*
6737	* We creates/updates MSR slots for the host MSRs in the VM-exit MSR-load area. The
6738	* actual host MSR values are not- updated here for performance reasons. See
6739	* hmR0VmxExportHostMsrs().
6740	*
6741	* We also exports the guest sysenter MSRs into the guest-state area in the VMCS.
6742	*
6743	* @returns VBox status code.
6744	* @param pVCpu The cross context virtual CPU structure.
6745	* @param pVmxTransient The VMX-transient structure.
6746	*
6747	* @remarks No-long-jump zone!!!
6748	*/
6749	static int hmR0VmxExportGuestMsrs(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6750	{
6751	AssertPtr(pVCpu);
6752	AssertPtr(pVmxTransient);
6753
6754	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6755	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6756
6757	/*
6758	* MSRs that we use the auto-load/store MSR area in the VMCS.
6759	* For 64-bit hosts, we load/restore them lazily, see hmR0VmxLazyLoadGuestMsrs(),
6760	* nothing to do here. The host MSR values are updated when it's safe in
6761	* hmR0VmxLazySaveHostMsrs().
6762	*
6763	* For nested-guests, the guests MSRs from the VM-entry MSR-load area are already
6764	* loaded (into the guest-CPU context) by the VMLAUNCH/VMRESUME instruction
6765	* emulation. The merged MSR permission bitmap will ensure that we get VM-exits
6766	* for any MSR that are not part of the lazy MSRs so we do not need to place
6767	* those MSRs into the auto-load/store MSR area. Nothing to do here.
6768	*/
6769	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_GUEST_AUTO_MSRS)
6770	{
6771	/* No auto-load/store MSRs currently. */
6772	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_GUEST_AUTO_MSRS);
6773	}
6774
6775	/*
6776	* Guest Sysenter MSRs.
6777	*/
6778	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_MSR_MASK)
6779	{
6780	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SYSENTER_MSRS);
6781
6782	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_CS_MSR)
6783	{
6784	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, pCtx->SysEnter.cs);
6785	AssertRC(rc);
6786	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_CS_MSR);
6787	}
6788
6789	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_EIP_MSR)
6790	{
6791	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SYSENTER_EIP, pCtx->SysEnter.eip);
6792	AssertRC(rc);
6793	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_EIP_MSR);
6794	}
6795
6796	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_ESP_MSR)
6797	{
6798	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SYSENTER_ESP, pCtx->SysEnter.esp);
6799	AssertRC(rc);
6800	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_ESP_MSR);
6801	}
6802	}
6803
6804	/*
6805	* Guest/host EFER MSR.
6806	*/
6807	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_EFER_MSR)
6808	{
6809	/* Whether we are using the VMCS to swap the EFER MSR must have been
6810	determined earlier while exporting VM-entry/VM-exit controls. */
6811	Assert(!(ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_ENTRY_EXIT_CTLS));
6812	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
6813
6814	if (hmR0VmxShouldSwapEferMsr(pVCpu, pVmxTransient))
6815	{
6816	/*
6817	* EFER.LME is written by software, while EFER.LMA is set by the CPU to (CR0.PG & EFER.LME).
6818	* This means a guest can set EFER.LME=1 while CR0.PG=0 and EFER.LMA can remain 0.
6819	* VT-x requires that "IA-32e mode guest" VM-entry control must be identical to EFER.LMA
6820	* and to CR0.PG. Without unrestricted execution, CR0.PG (used for VT-x, not the shadow)
6821	* must always be 1. This forces us to effectively clear both EFER.LMA and EFER.LME until
6822	* the guest has also set CR0.PG=1. Otherwise, we would run into an invalid-guest state
6823	* during VM-entry.
6824	*/
6825	uint64_t uGuestEferMsr = pCtx->msrEFER;
6826	if (!pVM->hmr0.s.vmx.fUnrestrictedGuest)
6827	{
6828	if (!(pCtx->msrEFER & MSR_K6_EFER_LMA))
6829	uGuestEferMsr &= ~MSR_K6_EFER_LME;
6830	else
6831	Assert((pCtx->msrEFER & (MSR_K6_EFER_LMA \| MSR_K6_EFER_LME)) == (MSR_K6_EFER_LMA \| MSR_K6_EFER_LME));
6832	}
6833
6834	/*
6835	* If the CPU supports VMCS controls for swapping EFER, use it. Otherwise, we have no option
6836	* but to use the auto-load store MSR area in the VMCS for swapping EFER. See @bugref{7368}.
6837	*/
6838	if (g_fHmVmxSupportsVmcsEfer)
6839	{
6840	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_EFER_FULL, uGuestEferMsr);
6841	AssertRC(rc);
6842	}
6843	else
6844	{
6845	/*
6846	* We shall use the auto-load/store MSR area only for loading the EFER MSR but we must
6847	* continue to intercept guest read and write accesses to it, see @bugref{7386#c16}.
6848	*/
6849	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_EFER, uGuestEferMsr,
6850	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6851	AssertRCReturn(rc, rc);
6852	}
6853
6854	Log4Func(("efer=%#RX64 shadow=%#RX64\n", uGuestEferMsr, pCtx->msrEFER));
6855	}
6856	else if (!g_fHmVmxSupportsVmcsEfer)
6857	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_EFER);
6858
6859	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_EFER_MSR);
6860	}
6861
6862	/*
6863	* Other MSRs.
6864	*/
6865	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_OTHER_MSRS)
6866	{
6867	/* Speculation Control (R/W). */
6868	HMVMX_CPUMCTX_ASSERT(pVCpu, HM_CHANGED_GUEST_OTHER_MSRS);
6869	if (pVM->cpum.ro.GuestFeatures.fIbrs)
6870	{
6871	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_IA32_SPEC_CTRL, CPUMGetGuestSpecCtrl(pVCpu),
6872	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6873	AssertRCReturn(rc, rc);
6874	}
6875
6876	/* Last Branch Record. */
6877	if (pVM->hmr0.s.vmx.fLbr)
6878	{
6879	PVMXVMCSINFOSHARED const pVmcsInfoShared = pVmxTransient->pVmcsInfo->pShared;
6880	uint32_t const idFromIpMsrStart = pVM->hmr0.s.vmx.idLbrFromIpMsrFirst;
6881	uint32_t const idToIpMsrStart = pVM->hmr0.s.vmx.idLbrToIpMsrFirst;
6882	uint32_t const cLbrStack = pVM->hmr0.s.vmx.idLbrFromIpMsrLast - pVM->hmr0.s.vmx.idLbrFromIpMsrFirst + 1;
6883	Assert(cLbrStack <= 32);
6884	for (uint32_t i = 0; i < cLbrStack; i++)
6885	{
6886	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, idFromIpMsrStart + i,
6887	pVmcsInfoShared->au64LbrFromIpMsr[i],
6888	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6889	AssertRCReturn(rc, rc);
6890
6891	/* Some CPUs don't have a Branch-To-IP MSR (P4 and related Xeons). */
6892	if (idToIpMsrStart != 0)
6893	{
6894	rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, idToIpMsrStart + i,
6895	pVmcsInfoShared->au64LbrToIpMsr[i],
6896	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6897	AssertRCReturn(rc, rc);
6898	}
6899	}
6900
6901	/* Add LBR top-of-stack MSR (which contains the index to the most recent record). */
6902	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, pVM->hmr0.s.vmx.idLbrTosMsr,
6903	pVmcsInfoShared->u64LbrTosMsr, false /* fSetReadWrite */,
6904	false /* fUpdateHostMsr */);
6905	AssertRCReturn(rc, rc);
6906	}
6907
6908	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_OTHER_MSRS);
6909	}
6910
6911	return VINF_SUCCESS;
6912	}
6913
6914
6915	/**
6916	* Wrapper for running the guest code in VT-x.
6917	*
6918	* @returns VBox status code, no informational status codes.
6919	* @param pVCpu The cross context virtual CPU structure.
6920	* @param pVmxTransient The VMX-transient structure.
6921	*
6922	* @remarks No-long-jump zone!!!
6923	*/
6924	DECLINLINE(int) hmR0VmxRunGuest(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6925	{
6926	/* Mark that HM is the keeper of all guest-CPU registers now that we're going to execute guest code. */
6927	pVCpu->cpum.GstCtx.fExtrn \|= HMVMX_CPUMCTX_EXTRN_ALL \| CPUMCTX_EXTRN_KEEPER_HM;
6928
6929	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6930	bool const fResumeVM = RT_BOOL(pVmcsInfo->fVmcsState & VMX_V_VMCS_LAUNCH_STATE_LAUNCHED);
6931	#ifdef VBOX_WITH_STATISTICS
6932	if (fResumeVM)
6933	STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxVmResume);
6934	else
6935	STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxVmLaunch);
6936	#endif
6937	int rc = pVCpu->hmr0.s.vmx.pfnStartVm(pVmcsInfo, pVCpu, fResumeVM);
6938	AssertMsg(rc <= VINF_SUCCESS, ("%Rrc\n", rc));
6939	return rc;
6940	}
6941
6942
6943	/**
6944	* Reports world-switch error and dumps some useful debug info.
6945	*
6946	* @param pVCpu The cross context virtual CPU structure.
6947	* @param rcVMRun The return code from VMLAUNCH/VMRESUME.
6948	* @param pVmxTransient The VMX-transient structure (only
6949	* exitReason updated).
6950	*/
6951	static void hmR0VmxReportWorldSwitchError(PVMCPUCC pVCpu, int rcVMRun, PVMXTRANSIENT pVmxTransient)
6952	{
6953	Assert(pVCpu);
6954	Assert(pVmxTransient);
6955	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
6956
6957	Log4Func(("VM-entry failure: %Rrc\n", rcVMRun));
6958	switch (rcVMRun)
6959	{
6960	case VERR_VMX_INVALID_VMXON_PTR:
6961	AssertFailed();
6962	break;
6963	case VINF_SUCCESS: /* VMLAUNCH/VMRESUME succeeded but VM-entry failed... yeah, true story. */
6964	case VERR_VMX_UNABLE_TO_START_VM: /* VMLAUNCH/VMRESUME itself failed. */
6965	{
6966	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &pVCpu->hm.s.vmx.LastError.u32ExitReason);
6967	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
6968	AssertRC(rc);
6969	hmR0VmxReadExitQualVmcs(pVmxTransient);
6970
6971	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hmr0.s.idEnteredCpu;
6972	/* LastError.idCurrentCpu was already updated in hmR0VmxPreRunGuestCommitted().
6973	Cannot do it here as we may have been long preempted. */
6974
6975	#ifdef VBOX_STRICT
6976	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
6977	Log4(("uExitReason %#RX32 (VmxTransient %#RX16)\n", pVCpu->hm.s.vmx.LastError.u32ExitReason,
6978	pVmxTransient->uExitReason));
6979	Log4(("Exit Qualification %#RX64\n", pVmxTransient->uExitQual));
6980	Log4(("InstrError %#RX32\n", pVCpu->hm.s.vmx.LastError.u32InstrError));
6981	if (pVCpu->hm.s.vmx.LastError.u32InstrError <= HMVMX_INSTR_ERROR_MAX)
6982	Log4(("InstrError Desc. \"%s\"\n", g_apszVmxInstrErrors[pVCpu->hm.s.vmx.LastError.u32InstrError]));
6983	else
6984	Log4(("InstrError Desc. Range exceeded %u\n", HMVMX_INSTR_ERROR_MAX));
6985	Log4(("Entered host CPU %u\n", pVCpu->hm.s.vmx.LastError.idEnteredCpu));
6986	Log4(("Current host CPU %u\n", pVCpu->hm.s.vmx.LastError.idCurrentCpu));
6987
6988	static struct
6989	{
6990	/** Name of the field to log. */
6991	const char *pszName;
6992	/** The VMCS field. */
6993	uint32_t uVmcsField;
6994	/** Whether host support of this field needs to be checked. */
6995	bool fCheckSupport;
6996	} const s_aVmcsFields[] =
6997	{
6998	{ "VMX_VMCS32_CTRL_PIN_EXEC", VMX_VMCS32_CTRL_PIN_EXEC, false },
6999	{ "VMX_VMCS32_CTRL_PROC_EXEC", VMX_VMCS32_CTRL_PROC_EXEC, false },
7000	{ "VMX_VMCS32_CTRL_PROC_EXEC2", VMX_VMCS32_CTRL_PROC_EXEC2, true },
7001	{ "VMX_VMCS32_CTRL_ENTRY", VMX_VMCS32_CTRL_ENTRY, false },
7002	{ "VMX_VMCS32_CTRL_EXIT", VMX_VMCS32_CTRL_EXIT, false },
7003	{ "VMX_VMCS32_CTRL_CR3_TARGET_COUNT", VMX_VMCS32_CTRL_CR3_TARGET_COUNT, false },
7004	{ "VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO", VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, false },
7005	{ "VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE", VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, false },
7006	{ "VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH", VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, false },
7007	{ "VMX_VMCS32_CTRL_TPR_THRESHOLD", VMX_VMCS32_CTRL_TPR_THRESHOLD, false },
7008	{ "VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT", VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, false },
7009	{ "VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT", VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, false },
7010	{ "VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT", VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, false },
7011	{ "VMX_VMCS32_CTRL_EXCEPTION_BITMAP", VMX_VMCS32_CTRL_EXCEPTION_BITMAP, false },
7012	{ "VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK", VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, false },
7013	{ "VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH", VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, false },
7014	{ "VMX_VMCS_CTRL_CR0_MASK", VMX_VMCS_CTRL_CR0_MASK, false },
7015	{ "VMX_VMCS_CTRL_CR0_READ_SHADOW", VMX_VMCS_CTRL_CR0_READ_SHADOW, false },
7016	{ "VMX_VMCS_CTRL_CR4_MASK", VMX_VMCS_CTRL_CR4_MASK, false },
7017	{ "VMX_VMCS_CTRL_CR4_READ_SHADOW", VMX_VMCS_CTRL_CR4_READ_SHADOW, false },
7018	{ "VMX_VMCS64_CTRL_EPTP_FULL", VMX_VMCS64_CTRL_EPTP_FULL, true },
7019	{ "VMX_VMCS_GUEST_RIP", VMX_VMCS_GUEST_RIP, false },
7020	{ "VMX_VMCS_GUEST_RSP", VMX_VMCS_GUEST_RSP, false },
7021	{ "VMX_VMCS_GUEST_RFLAGS", VMX_VMCS_GUEST_RFLAGS, false },
7022	{ "VMX_VMCS16_VPID", VMX_VMCS16_VPID, true, },
7023	{ "VMX_VMCS_HOST_CR0", VMX_VMCS_HOST_CR0, false },
7024	{ "VMX_VMCS_HOST_CR3", VMX_VMCS_HOST_CR3, false },
7025	{ "VMX_VMCS_HOST_CR4", VMX_VMCS_HOST_CR4, false },
7026	/* The order of selector fields below are fixed! */
7027	{ "VMX_VMCS16_HOST_ES_SEL", VMX_VMCS16_HOST_ES_SEL, false },
7028	{ "VMX_VMCS16_HOST_CS_SEL", VMX_VMCS16_HOST_CS_SEL, false },
7029	{ "VMX_VMCS16_HOST_SS_SEL", VMX_VMCS16_HOST_SS_SEL, false },
7030	{ "VMX_VMCS16_HOST_DS_SEL", VMX_VMCS16_HOST_DS_SEL, false },
7031	{ "VMX_VMCS16_HOST_FS_SEL", VMX_VMCS16_HOST_FS_SEL, false },
7032	{ "VMX_VMCS16_HOST_GS_SEL", VMX_VMCS16_HOST_GS_SEL, false },
7033	{ "VMX_VMCS16_HOST_TR_SEL", VMX_VMCS16_HOST_TR_SEL, false },
7034	/* End of ordered selector fields. */
7035	{ "VMX_VMCS_HOST_TR_BASE", VMX_VMCS_HOST_TR_BASE, false },
7036	{ "VMX_VMCS_HOST_GDTR_BASE", VMX_VMCS_HOST_GDTR_BASE, false },
7037	{ "VMX_VMCS_HOST_IDTR_BASE", VMX_VMCS_HOST_IDTR_BASE, false },
7038	{ "VMX_VMCS32_HOST_SYSENTER_CS", VMX_VMCS32_HOST_SYSENTER_CS, false },
7039	{ "VMX_VMCS_HOST_SYSENTER_EIP", VMX_VMCS_HOST_SYSENTER_EIP, false },
7040	{ "VMX_VMCS_HOST_SYSENTER_ESP", VMX_VMCS_HOST_SYSENTER_ESP, false },
7041	{ "VMX_VMCS_HOST_RSP", VMX_VMCS_HOST_RSP, false },
7042	{ "VMX_VMCS_HOST_RIP", VMX_VMCS_HOST_RIP, false }
7043	};
7044
7045	RTGDTR HostGdtr;
7046	ASMGetGDTR(&HostGdtr);
7047
7048	uint32_t const cVmcsFields = RT_ELEMENTS(s_aVmcsFields);
7049	for (uint32_t i = 0; i < cVmcsFields; i++)
7050	{
7051	uint32_t const uVmcsField = s_aVmcsFields[i].uVmcsField;
7052
7053	bool fSupported;
7054	if (!s_aVmcsFields[i].fCheckSupport)
7055	fSupported = true;
7056	else
7057	{
7058	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7059	switch (uVmcsField)
7060	{
7061	case VMX_VMCS64_CTRL_EPTP_FULL: fSupported = pVM->hmr0.s.fNestedPaging; break;
7062	case VMX_VMCS16_VPID: fSupported = pVM->hmr0.s.vmx.fVpid; break;
7063	case VMX_VMCS32_CTRL_PROC_EXEC2:
7064	fSupported = RT_BOOL(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS);
7065	break;
7066	default:
7067	AssertMsgFailedReturnVoid(("Failed to provide VMCS field support for %#RX32\n", uVmcsField));
7068	}
7069	}
7070
7071	if (fSupported)
7072	{
7073	uint8_t const uWidth = RT_BF_GET(uVmcsField, VMX_BF_VMCSFIELD_WIDTH);
7074	switch (uWidth)
7075	{
7076	case VMX_VMCSFIELD_WIDTH_16BIT:
7077	{
7078	uint16_t u16Val;
7079	rc = VMXReadVmcs16(uVmcsField, &u16Val);
7080	AssertRC(rc);
7081	Log4(("%-40s = %#RX16\n", s_aVmcsFields[i].pszName, u16Val));
7082
7083	if ( uVmcsField >= VMX_VMCS16_HOST_ES_SEL
7084	&& uVmcsField <= VMX_VMCS16_HOST_TR_SEL)
7085	{
7086	if (u16Val < HostGdtr.cbGdt)
7087	{
7088	/* Order of selectors in s_apszSel is fixed and matches the order in s_aVmcsFields. */
7089	static const char * const s_apszSel[] = { "Host ES", "Host CS", "Host SS", "Host DS",
7090	"Host FS", "Host GS", "Host TR" };
7091	uint8_t const idxSel = RT_BF_GET(uVmcsField, VMX_BF_VMCSFIELD_INDEX);
7092	Assert(idxSel < RT_ELEMENTS(s_apszSel));
7093	PCX86DESCHC pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u16Val & X86_SEL_MASK));
7094	hmR0DumpDescriptor(pDesc, u16Val, s_apszSel[idxSel]);
7095	}
7096	else
7097	Log4((" Selector value exceeds GDT limit!\n"));
7098	}
7099	break;
7100	}
7101
7102	case VMX_VMCSFIELD_WIDTH_32BIT:
7103	{
7104	uint32_t u32Val;
7105	rc = VMXReadVmcs32(uVmcsField, &u32Val);
7106	AssertRC(rc);
7107	Log4(("%-40s = %#RX32\n", s_aVmcsFields[i].pszName, u32Val));
7108	break;
7109	}
7110
7111	case VMX_VMCSFIELD_WIDTH_64BIT:
7112	case VMX_VMCSFIELD_WIDTH_NATURAL:
7113	{
7114	uint64_t u64Val;
7115	rc = VMXReadVmcs64(uVmcsField, &u64Val);
7116	AssertRC(rc);
7117	Log4(("%-40s = %#RX64\n", s_aVmcsFields[i].pszName, u64Val));
7118	break;
7119	}
7120	}
7121	}
7122	}
7123
7124	Log4(("MSR_K6_EFER = %#RX64\n", ASMRdMsr(MSR_K6_EFER)));
7125	Log4(("MSR_K8_CSTAR = %#RX64\n", ASMRdMsr(MSR_K8_CSTAR)));
7126	Log4(("MSR_K8_LSTAR = %#RX64\n", ASMRdMsr(MSR_K8_LSTAR)));
7127	Log4(("MSR_K6_STAR = %#RX64\n", ASMRdMsr(MSR_K6_STAR)));
7128	Log4(("MSR_K8_SF_MASK = %#RX64\n", ASMRdMsr(MSR_K8_SF_MASK)));
7129	Log4(("MSR_K8_KERNEL_GS_BASE = %#RX64\n", ASMRdMsr(MSR_K8_KERNEL_GS_BASE)));
7130	#endif /* VBOX_STRICT */
7131	break;
7132	}
7133
7134	default:
7135	/* Impossible */
7136	AssertMsgFailed(("hmR0VmxReportWorldSwitchError %Rrc (%#x)\n", rcVMRun, rcVMRun));
7137	break;
7138	}
7139	}
7140
7141
7142	/**
7143	* Sets up the usage of TSC-offsetting and updates the VMCS.
7144	*
7145	* If offsetting is not possible, cause VM-exits on RDTSC(P)s. Also sets up the
7146	* VMX-preemption timer.
7147	*
7148	* @returns VBox status code.
7149	* @param pVCpu The cross context virtual CPU structure.
7150	* @param pVmxTransient The VMX-transient structure.
7151	* @param idCurrentCpu The current CPU number.
7152	*
7153	* @remarks No-long-jump zone!!!
7154	*/
7155	static void hmR0VmxUpdateTscOffsettingAndPreemptTimer(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, RTCPUID idCurrentCpu)
7156	{
7157	bool fOffsettedTsc;
7158	bool fParavirtTsc;
7159	uint64_t uTscOffset;
7160	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7161	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
7162
7163	if (pVM->hmr0.s.vmx.fUsePreemptTimer)
7164	{
7165	/* The TMCpuTickGetDeadlineAndTscOffset function is expensive (calling it on
7166	every entry slowed down the bs2-test1 CPUID testcase by ~33% (on an 10980xe). */
7167	uint64_t cTicksToDeadline;
7168	if ( idCurrentCpu == pVCpu->hmr0.s.idLastCpu
7169	&& TMVirtualSyncIsCurrentDeadlineVersion(pVM, pVCpu->hmr0.s.vmx.uTscDeadlineVersion))
7170	{
7171	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatVmxPreemptionReusingDeadline);
7172	fOffsettedTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &uTscOffset, &fParavirtTsc);
7173	cTicksToDeadline = pVCpu->hmr0.s.vmx.uTscDeadline - SUPReadTsc();
7174	if ((int64_t)cTicksToDeadline > 0)
7175	{ /* hopefully */ }
7176	else
7177	{
7178	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatVmxPreemptionReusingDeadlineExpired);
7179	cTicksToDeadline = 0;
7180	}
7181	}
7182	else
7183	{
7184	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatVmxPreemptionRecalcingDeadline);
7185	cTicksToDeadline = TMCpuTickGetDeadlineAndTscOffset(pVM, pVCpu, &uTscOffset, &fOffsettedTsc, &fParavirtTsc,
7186	&pVCpu->hmr0.s.vmx.uTscDeadline,
7187	&pVCpu->hmr0.s.vmx.uTscDeadlineVersion);
7188	pVCpu->hmr0.s.vmx.uTscDeadline += cTicksToDeadline;
7189	if (cTicksToDeadline >= 128)
7190	{ /* hopefully */ }
7191	else
7192	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatVmxPreemptionRecalcingDeadlineExpired);
7193	}
7194
7195	/* Make sure the returned values have sane upper and lower boundaries. */
7196	uint64_t const u64CpuHz = SUPGetCpuHzFromGipBySetIndex(g_pSUPGlobalInfoPage, pVCpu->iHostCpuSet);
7197	cTicksToDeadline = RT_MIN(cTicksToDeadline, u64CpuHz / 64); /* 1/64th of a second, 15.625ms. / /* @todo r=bird: Once real+virtual timers move to separate thread, we can raise the upper limit (16ms isn't much). ASSUMES working poke cpu function. */
7198	cTicksToDeadline = RT_MAX(cTicksToDeadline, u64CpuHz / 32678); /* 1/32768th of a second, ~30us. */
7199	cTicksToDeadline >>= pVM->hm.s.vmx.cPreemptTimerShift;
7200
7201	/** @todo r=ramshankar: We need to find a way to integrate nested-guest
7202	* preemption timers here. We probably need to clamp the preemption timer,
7203	* after converting the timer value to the host. */
7204	uint32_t const cPreemptionTickCount = (uint32_t)RT_MIN(cTicksToDeadline, UINT32_MAX - 16);
7205	int rc = VMXWriteVmcs32(VMX_VMCS32_PREEMPT_TIMER_VALUE, cPreemptionTickCount);
7206	AssertRC(rc);
7207	}
7208	else
7209	fOffsettedTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &uTscOffset, &fParavirtTsc);
7210
7211	if (fParavirtTsc)
7212	{
7213	/* Currently neither Hyper-V nor KVM need to update their paravirt. TSC
7214	information before every VM-entry, hence disable it for performance sake. */
7215	#if 0
7216	int rc = GIMR0UpdateParavirtTsc(pVM, 0 /* u64Offset */);
7217	AssertRC(rc);
7218	#endif
7219	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
7220	}
7221
7222	if ( fOffsettedTsc
7223	&& RT_LIKELY(!pVCpu->hmr0.s.fDebugWantRdTscExit))
7224	{
7225	if (pVmxTransient->fIsNestedGuest)
7226	uTscOffset = CPUMApplyNestedGuestTscOffset(pVCpu, uTscOffset);
7227	hmR0VmxSetTscOffsetVmcs(pVmcsInfo, uTscOffset);
7228	hmR0VmxRemoveProcCtlsVmcs(pVCpu, pVmxTransient, VMX_PROC_CTLS_RDTSC_EXIT);
7229	}
7230	else
7231	{
7232	/* We can't use TSC-offsetting (non-fixed TSC, warp drive active etc.), VM-exit on RDTSC(P). */
7233	hmR0VmxSetProcCtlsVmcs(pVmxTransient, VMX_PROC_CTLS_RDTSC_EXIT);
7234	}
7235	}
7236
7237
7238	/**
7239	* Gets the IEM exception flags for the specified vector and IDT vectoring /
7240	* VM-exit interruption info type.
7241	*
7242	* @returns The IEM exception flags.
7243	* @param uVector The event vector.
7244	* @param uVmxEventType The VMX event type.
7245	*
7246	* @remarks This function currently only constructs flags required for
7247	* IEMEvaluateRecursiveXcpt and not the complete flags (e.g, error-code
7248	* and CR2 aspects of an exception are not included).
7249	*/
7250	static uint32_t hmR0VmxGetIemXcptFlags(uint8_t uVector, uint32_t uVmxEventType)
7251	{
7252	uint32_t fIemXcptFlags;
7253	switch (uVmxEventType)
7254	{
7255	case VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT:
7256	case VMX_IDT_VECTORING_INFO_TYPE_NMI:
7257	fIemXcptFlags = IEM_XCPT_FLAGS_T_CPU_XCPT;
7258	break;
7259
7260	case VMX_IDT_VECTORING_INFO_TYPE_EXT_INT:
7261	fIemXcptFlags = IEM_XCPT_FLAGS_T_EXT_INT;
7262	break;
7263
7264	case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
7265	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT \| IEM_XCPT_FLAGS_ICEBP_INSTR;
7266	break;
7267
7268	case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT:
7269	{
7270	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
7271	if (uVector == X86_XCPT_BP)
7272	fIemXcptFlags \|= IEM_XCPT_FLAGS_BP_INSTR;
7273	else if (uVector == X86_XCPT_OF)
7274	fIemXcptFlags \|= IEM_XCPT_FLAGS_OF_INSTR;
7275	else
7276	{
7277	fIemXcptFlags = 0;
7278	AssertMsgFailed(("Unexpected vector for software exception. uVector=%#x", uVector));
7279	}
7280	break;
7281	}
7282
7283	case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
7284	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
7285	break;
7286
7287	default:
7288	fIemXcptFlags = 0;
7289	AssertMsgFailed(("Unexpected vector type! uVmxEventType=%#x uVector=%#x", uVmxEventType, uVector));
7290	break;
7291	}
7292	return fIemXcptFlags;
7293	}
7294
7295
7296	/**
7297	* Sets an event as a pending event to be injected into the guest.
7298	*
7299	* @param pVCpu The cross context virtual CPU structure.
7300	* @param u32IntInfo The VM-entry interruption-information field.
7301	* @param cbInstr The VM-entry instruction length in bytes (for
7302	* software interrupts, exceptions and privileged
7303	* software exceptions).
7304	* @param u32ErrCode The VM-entry exception error code.
7305	* @param GCPtrFaultAddress The fault-address (CR2) in case it's a
7306	* page-fault.
7307	*/
7308	DECLINLINE(void) hmR0VmxSetPendingEvent(PVMCPUCC pVCpu, uint32_t u32IntInfo, uint32_t cbInstr, uint32_t u32ErrCode,
7309	RTGCUINTPTR GCPtrFaultAddress)
7310	{
7311	Assert(!pVCpu->hm.s.Event.fPending);
7312	pVCpu->hm.s.Event.fPending = true;
7313	pVCpu->hm.s.Event.u64IntInfo = u32IntInfo;
7314	pVCpu->hm.s.Event.u32ErrCode = u32ErrCode;
7315	pVCpu->hm.s.Event.cbInstr = cbInstr;
7316	pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress;
7317	}
7318
7319
7320	/**
7321	* Sets an external interrupt as pending-for-injection into the VM.
7322	*
7323	* @param pVCpu The cross context virtual CPU structure.
7324	* @param u8Interrupt The external interrupt vector.
7325	*/
7326	DECLINLINE(void) hmR0VmxSetPendingExtInt(PVMCPUCC pVCpu, uint8_t u8Interrupt)
7327	{
7328	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, u8Interrupt)
7329	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
7330	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7331	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7332	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7333	}
7334
7335
7336	/**
7337	* Sets an NMI (\#NMI) exception as pending-for-injection into the VM.
7338	*
7339	* @param pVCpu The cross context virtual CPU structure.
7340	*/
7341	DECLINLINE(void) hmR0VmxSetPendingXcptNmi(PVMCPUCC pVCpu)
7342	{
7343	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_NMI)
7344	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_NMI)
7345	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7346	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7347	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7348	}
7349
7350
7351	/**
7352	* Sets a double-fault (\#DF) exception as pending-for-injection into the VM.
7353	*
7354	* @param pVCpu The cross context virtual CPU structure.
7355	*/
7356	DECLINLINE(void) hmR0VmxSetPendingXcptDF(PVMCPUCC pVCpu)
7357	{
7358	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DF)
7359	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7360	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7361	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7362	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7363	}
7364
7365
7366	/**
7367	* Sets an invalid-opcode (\#UD) exception as pending-for-injection into the VM.
7368	*
7369	* @param pVCpu The cross context virtual CPU structure.
7370	*/
7371	DECLINLINE(void) hmR0VmxSetPendingXcptUD(PVMCPUCC pVCpu)
7372	{
7373	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_UD)
7374	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7375	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7376	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7377	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7378	}
7379
7380
7381	/**
7382	* Sets a debug (\#DB) exception as pending-for-injection into the VM.
7383	*
7384	* @param pVCpu The cross context virtual CPU structure.
7385	*/
7386	DECLINLINE(void) hmR0VmxSetPendingXcptDB(PVMCPUCC pVCpu)
7387	{
7388	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DB)
7389	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7390	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7391	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7392	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7393	}
7394
7395
7396	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7397	/**
7398	* Sets a general-protection (\#GP) exception as pending-for-injection into the VM.
7399	*
7400	* @param pVCpu The cross context virtual CPU structure.
7401	* @param u32ErrCode The error code for the general-protection exception.
7402	*/
7403	DECLINLINE(void) hmR0VmxSetPendingXcptGP(PVMCPUCC pVCpu, uint32_t u32ErrCode)
7404	{
7405	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_GP)
7406	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7407	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7408	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7409	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrCode, 0 / GCPtrFaultAddress */);
7410	}
7411
7412
7413	/**
7414	* Sets a stack (\#SS) exception as pending-for-injection into the VM.
7415	*
7416	* @param pVCpu The cross context virtual CPU structure.
7417	* @param u32ErrCode The error code for the stack exception.
7418	*/
7419	DECLINLINE(void) hmR0VmxSetPendingXcptSS(PVMCPUCC pVCpu, uint32_t u32ErrCode)
7420	{
7421	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_SS)
7422	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7423	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7424	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7425	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrCode, 0 / GCPtrFaultAddress */);
7426	}
7427	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
7428
7429
7430	/**
7431	* Fixes up attributes for the specified segment register.
7432	*
7433	* @param pVCpu The cross context virtual CPU structure.
7434	* @param pSelReg The segment register that needs fixing.
7435	* @param pszRegName The register name (for logging and assertions).
7436	*/
7437	static void hmR0VmxFixUnusableSegRegAttr(PVMCPUCC pVCpu, PCPUMSELREG pSelReg, const char *pszRegName)
7438	{
7439	Assert(pSelReg->Attr.u & X86DESCATTR_UNUSABLE);
7440
7441	/*
7442	* If VT-x marks the segment as unusable, most other bits remain undefined:
7443	* - For CS the L, D and G bits have meaning.
7444	* - For SS the DPL has meaning (it -is- the CPL for Intel and VBox).
7445	* - For the remaining data segments no bits are defined.
7446	*
7447	* The present bit and the unusable bit has been observed to be set at the
7448	* same time (the selector was supposed to be invalid as we started executing
7449	* a V8086 interrupt in ring-0).
7450	*
7451	* What should be important for the rest of the VBox code, is that the P bit is
7452	* cleared. Some of the other VBox code recognizes the unusable bit, but
7453	* AMD-V certainly don't, and REM doesn't really either. So, to be on the
7454	* safe side here, we'll strip off P and other bits we don't care about. If
7455	* any code breaks because Attr.u != 0 when Sel < 4, it should be fixed.
7456	*
7457	* See Intel spec. 27.3.2 "Saving Segment Registers and Descriptor-Table Registers".
7458	*/
7459	#ifdef VBOX_STRICT
7460	uint32_t const uAttr = pSelReg->Attr.u;
7461	#endif
7462
7463	/* Masking off: X86DESCATTR_P, X86DESCATTR_LIMIT_HIGH, and X86DESCATTR_AVL. The latter two are really irrelevant. */
7464	pSelReg->Attr.u &= X86DESCATTR_UNUSABLE \| X86DESCATTR_L \| X86DESCATTR_D \| X86DESCATTR_G
7465	\| X86DESCATTR_DPL \| X86DESCATTR_TYPE \| X86DESCATTR_DT;
7466
7467	#ifdef VBOX_STRICT
7468	VMMRZCallRing3Disable(pVCpu);
7469	Log4Func(("Unusable %s: sel=%#x attr=%#x -> %#x\n", pszRegName, pSelReg->Sel, uAttr, pSelReg->Attr.u));
7470	# ifdef DEBUG_bird
7471	AssertMsg((uAttr & ~X86DESCATTR_P) == pSelReg->Attr.u,
7472	("%s: %#x != %#x (sel=%#x base=%#llx limit=%#x)\n",
7473	pszRegName, uAttr, pSelReg->Attr.u, pSelReg->Sel, pSelReg->u64Base, pSelReg->u32Limit));
7474	# endif
7475	VMMRZCallRing3Enable(pVCpu);
7476	NOREF(uAttr);
7477	#endif
7478	RT_NOREF2(pVCpu, pszRegName);
7479	}
7480
7481
7482	/**
7483	* Imports a guest segment register from the current VMCS into the guest-CPU
7484	* context.
7485	*
7486	* @param pVCpu The cross context virtual CPU structure.
7487	* @param iSegReg The segment register number (X86_SREG_XXX).
7488	*
7489	* @remarks Called with interrupts and/or preemption disabled.
7490	*/
7491	static void hmR0VmxImportGuestSegReg(PVMCPUCC pVCpu, uint32_t iSegReg)
7492	{
7493	Assert(iSegReg < X86_SREG_COUNT);
7494	Assert((uint32_t)VMX_VMCS16_GUEST_SEG_SEL(iSegReg) == g_aVmcsSegSel[iSegReg]);
7495	Assert((uint32_t)VMX_VMCS32_GUEST_SEG_LIMIT(iSegReg) == g_aVmcsSegLimit[iSegReg]);
7496	Assert((uint32_t)VMX_VMCS32_GUEST_SEG_ACCESS_RIGHTS(iSegReg) == g_aVmcsSegAttr[iSegReg]);
7497	Assert((uint32_t)VMX_VMCS_GUEST_SEG_BASE(iSegReg) == g_aVmcsSegBase[iSegReg]);
7498
7499	PCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
7500
7501	uint16_t u16Sel;
7502	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_SEG_SEL(iSegReg), &u16Sel); AssertRC(rc);
7503	pSelReg->Sel = u16Sel;
7504	pSelReg->ValidSel = u16Sel;
7505
7506	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SEG_LIMIT(iSegReg), &pSelReg->u32Limit); AssertRC(rc);
7507	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SEG_BASE(iSegReg), &pSelReg->u64Base); AssertRC(rc);
7508
7509	uint32_t u32Attr;
7510	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SEG_ACCESS_RIGHTS(iSegReg), &u32Attr); AssertRC(rc);
7511	pSelReg->Attr.u = u32Attr;
7512	if (u32Attr & X86DESCATTR_UNUSABLE)
7513	hmR0VmxFixUnusableSegRegAttr(pVCpu, pSelReg, "ES\0CS\0SS\0DS\0FS\0GS" + iSegReg * 3);
7514
7515	pSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
7516	}
7517
7518
7519	/**
7520	* Imports the guest LDTR from the current VMCS into the guest-CPU context.
7521	*
7522	* @param pVCpu The cross context virtual CPU structure.
7523	*
7524	* @remarks Called with interrupts and/or preemption disabled.
7525	*/
7526	static void hmR0VmxImportGuestLdtr(PVMCPUCC pVCpu)
7527	{
7528	uint16_t u16Sel;
7529	uint64_t u64Base;
7530	uint32_t u32Limit, u32Attr;
7531	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_LDTR_SEL, &u16Sel); AssertRC(rc);
7532	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, &u32Limit); AssertRC(rc);
7533	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, &u32Attr); AssertRC(rc);
7534	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_LDTR_BASE, &u64Base); AssertRC(rc);
7535
7536	pVCpu->cpum.GstCtx.ldtr.Sel = u16Sel;
7537	pVCpu->cpum.GstCtx.ldtr.ValidSel = u16Sel;
7538	pVCpu->cpum.GstCtx.ldtr.fFlags = CPUMSELREG_FLAGS_VALID;
7539	pVCpu->cpum.GstCtx.ldtr.u32Limit = u32Limit;
7540	pVCpu->cpum.GstCtx.ldtr.u64Base = u64Base;
7541	pVCpu->cpum.GstCtx.ldtr.Attr.u = u32Attr;
7542	if (u32Attr & X86DESCATTR_UNUSABLE)
7543	hmR0VmxFixUnusableSegRegAttr(pVCpu, &pVCpu->cpum.GstCtx.ldtr, "LDTR");
7544	}
7545
7546
7547	/**
7548	* Imports the guest TR from the current VMCS into the guest-CPU context.
7549	*
7550	* @param pVCpu The cross context virtual CPU structure.
7551	*
7552	* @remarks Called with interrupts and/or preemption disabled.
7553	*/
7554	static void hmR0VmxImportGuestTr(PVMCPUCC pVCpu)
7555	{
7556	uint16_t u16Sel;
7557	uint64_t u64Base;
7558	uint32_t u32Limit, u32Attr;
7559	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_TR_SEL, &u16Sel); AssertRC(rc);
7560	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, &u32Limit); AssertRC(rc);
7561	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, &u32Attr); AssertRC(rc);
7562	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_TR_BASE, &u64Base); AssertRC(rc);
7563
7564	pVCpu->cpum.GstCtx.tr.Sel = u16Sel;
7565	pVCpu->cpum.GstCtx.tr.ValidSel = u16Sel;
7566	pVCpu->cpum.GstCtx.tr.fFlags = CPUMSELREG_FLAGS_VALID;
7567	pVCpu->cpum.GstCtx.tr.u32Limit = u32Limit;
7568	pVCpu->cpum.GstCtx.tr.u64Base = u64Base;
7569	pVCpu->cpum.GstCtx.tr.Attr.u = u32Attr;
7570	/* TR is the only selector that can never be unusable. */
7571	Assert(!(u32Attr & X86DESCATTR_UNUSABLE));
7572	}
7573
7574
7575	/**
7576	* Imports the guest RIP from the VMCS back into the guest-CPU context.
7577	*
7578	* @param pVCpu The cross context virtual CPU structure.
7579	*
7580	* @remarks Called with interrupts and/or preemption disabled, should not assert!
7581	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7582	* instead!!!
7583	*/
7584	static void hmR0VmxImportGuestRip(PVMCPUCC pVCpu)
7585	{
7586	uint64_t u64Val;
7587	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7588	if (pCtx->fExtrn & CPUMCTX_EXTRN_RIP)
7589	{
7590	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RIP, &u64Val);
7591	AssertRC(rc);
7592
7593	pCtx->rip = u64Val;
7594	EMHistoryUpdatePC(pVCpu, pCtx->rip, false);
7595	pCtx->fExtrn &= ~CPUMCTX_EXTRN_RIP;
7596	}
7597	}
7598
7599
7600	/**
7601	* Imports the guest RFLAGS from the VMCS back into the guest-CPU context.
7602	*
7603	* @param pVCpu The cross context virtual CPU structure.
7604	* @param pVmcsInfo The VMCS info. object.
7605	*
7606	* @remarks Called with interrupts and/or preemption disabled, should not assert!
7607	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7608	* instead!!!
7609	*/
7610	static void hmR0VmxImportGuestRFlags(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
7611	{
7612	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7613	if (pCtx->fExtrn & CPUMCTX_EXTRN_RFLAGS)
7614	{
7615	uint64_t u64Val;
7616	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RFLAGS, &u64Val);
7617	AssertRC(rc);
7618
7619	pCtx->rflags.u64 = u64Val;
7620	PCVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
7621	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
7622	{
7623	pCtx->eflags.Bits.u1VM = 0;
7624	pCtx->eflags.Bits.u2IOPL = pVmcsInfoShared->RealMode.Eflags.Bits.u2IOPL;
7625	}
7626	pCtx->fExtrn &= ~CPUMCTX_EXTRN_RFLAGS;
7627	}
7628	}
7629
7630
7631	/**
7632	* Imports the guest interruptibility-state from the VMCS back into the guest-CPU
7633	* context.
7634	*
7635	* @param pVCpu The cross context virtual CPU structure.
7636	* @param pVmcsInfo The VMCS info. object.
7637	*
7638	* @remarks Called with interrupts and/or preemption disabled, try not to assert and
7639	* do not log!
7640	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7641	* instead!!!
7642	*/
7643	static void hmR0VmxImportGuestIntrState(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
7644	{
7645	uint32_t u32Val;
7646	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &u32Val); AssertRC(rc);
7647	if (!u32Val)
7648	{
7649	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
7650	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
7651	CPUMSetGuestNmiBlocking(pVCpu, false);
7652	}
7653	else
7654	{
7655	/*
7656	* We must import RIP here to set our EM interrupt-inhibited state.
7657	* We also import RFLAGS as our code that evaluates pending interrupts
7658	* before VM-entry requires it.
7659	*/
7660	hmR0VmxImportGuestRip(pVCpu);
7661	hmR0VmxImportGuestRFlags(pVCpu, pVmcsInfo);
7662
7663	if (u32Val & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS \| VMX_VMCS_GUEST_INT_STATE_BLOCK_STI))
7664	EMSetInhibitInterruptsPC(pVCpu, pVCpu->cpum.GstCtx.rip);
7665	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
7666	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
7667
7668	bool const fNmiBlocking = RT_BOOL(u32Val & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI);
7669	CPUMSetGuestNmiBlocking(pVCpu, fNmiBlocking);
7670	}
7671	}
7672
7673
7674	/**
7675	* Worker for VMXR0ImportStateOnDemand.
7676	*
7677	* @returns VBox status code.
7678	* @param pVCpu The cross context virtual CPU structure.
7679	* @param pVmcsInfo The VMCS info. object.
7680	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
7681	*/
7682	static int hmR0VmxImportGuestState(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint64_t fWhat)
7683	{
7684	int rc = VINF_SUCCESS;
7685	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7686	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7687	uint32_t u32Val;
7688
7689	/*
7690	* Note! This is hack to workaround a mysterious BSOD observed with release builds
7691	* on Windows 10 64-bit hosts. Profile and debug builds are not affected and
7692	* neither are other host platforms.
7693	*
7694	* Committing this temporarily as it prevents BSOD.
7695	*
7696	* Update: This is very likely a compiler optimization bug, see @bugref{9180}.
7697	*/
7698	#ifdef RT_OS_WINDOWS
7699	if (pVM == 0 \|\| pVM == (void *)(uintptr_t)-1)
7700	return VERR_HM_IPE_1;
7701	#endif
7702
7703	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatImportGuestState, x);
7704
7705	/*
7706	* We disable interrupts to make the updating of the state and in particular
7707	* the fExtrn modification atomic wrt to preemption hooks.
7708	*/
7709	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
7710
7711	fWhat &= pCtx->fExtrn;
7712	if (fWhat)
7713	{
7714	do
7715	{
7716	if (fWhat & CPUMCTX_EXTRN_RIP)
7717	hmR0VmxImportGuestRip(pVCpu);
7718
7719	if (fWhat & CPUMCTX_EXTRN_RFLAGS)
7720	hmR0VmxImportGuestRFlags(pVCpu, pVmcsInfo);
7721
7722	if (fWhat & (CPUMCTX_EXTRN_INHIBIT_INT \| CPUMCTX_EXTRN_INHIBIT_NMI))
7723	hmR0VmxImportGuestIntrState(pVCpu, pVmcsInfo);
7724
7725	if (fWhat & CPUMCTX_EXTRN_RSP)
7726	{
7727	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RSP, &pCtx->rsp);
7728	AssertRC(rc);
7729	}
7730
7731	if (fWhat & CPUMCTX_EXTRN_SREG_MASK)
7732	{
7733	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
7734	bool const fRealOnV86Active = pVmcsInfoShared->RealMode.fRealOnV86Active;
7735	if (fWhat & CPUMCTX_EXTRN_CS)
7736	{
7737	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_CS);
7738	hmR0VmxImportGuestRip(pVCpu);
7739	if (fRealOnV86Active)
7740	pCtx->cs.Attr.u = pVmcsInfoShared->RealMode.AttrCS.u;
7741	EMHistoryUpdatePC(pVCpu, pCtx->cs.u64Base + pCtx->rip, true /* fFlattened */);
7742	}
7743	if (fWhat & CPUMCTX_EXTRN_SS)
7744	{
7745	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_SS);
7746	if (fRealOnV86Active)
7747	pCtx->ss.Attr.u = pVmcsInfoShared->RealMode.AttrSS.u;
7748	}
7749	if (fWhat & CPUMCTX_EXTRN_DS)
7750	{
7751	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_DS);
7752	if (fRealOnV86Active)
7753	pCtx->ds.Attr.u = pVmcsInfoShared->RealMode.AttrDS.u;
7754	}
7755	if (fWhat & CPUMCTX_EXTRN_ES)
7756	{
7757	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_ES);
7758	if (fRealOnV86Active)
7759	pCtx->es.Attr.u = pVmcsInfoShared->RealMode.AttrES.u;
7760	}
7761	if (fWhat & CPUMCTX_EXTRN_FS)
7762	{
7763	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_FS);
7764	if (fRealOnV86Active)
7765	pCtx->fs.Attr.u = pVmcsInfoShared->RealMode.AttrFS.u;
7766	}
7767	if (fWhat & CPUMCTX_EXTRN_GS)
7768	{
7769	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_GS);
7770	if (fRealOnV86Active)
7771	pCtx->gs.Attr.u = pVmcsInfoShared->RealMode.AttrGS.u;
7772	}
7773	}
7774
7775	if (fWhat & CPUMCTX_EXTRN_TABLE_MASK)
7776	{
7777	if (fWhat & CPUMCTX_EXTRN_LDTR)
7778	hmR0VmxImportGuestLdtr(pVCpu);
7779
7780	if (fWhat & CPUMCTX_EXTRN_GDTR)
7781	{
7782	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, &pCtx->gdtr.pGdt); AssertRC(rc);
7783	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val); AssertRC(rc);
7784	pCtx->gdtr.cbGdt = u32Val;
7785	}
7786
7787	/* Guest IDTR. */
7788	if (fWhat & CPUMCTX_EXTRN_IDTR)
7789	{
7790	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, &pCtx->idtr.pIdt); AssertRC(rc);
7791	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val); AssertRC(rc);
7792	pCtx->idtr.cbIdt = u32Val;
7793	}
7794
7795	/* Guest TR. */
7796	if (fWhat & CPUMCTX_EXTRN_TR)
7797	{
7798	/* Real-mode emulation using virtual-8086 mode has the fake TSS (pRealModeTSS) in TR,
7799	don't need to import that one. */
7800	if (!pVmcsInfo->pShared->RealMode.fRealOnV86Active)
7801	hmR0VmxImportGuestTr(pVCpu);
7802	}
7803	}
7804
7805	if (fWhat & CPUMCTX_EXTRN_DR7)
7806	{
7807	if (!pVCpu->hmr0.s.fUsingHyperDR7)
7808	{
7809	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_DR7, &pCtx->dr[7]);
7810	AssertRC(rc);
7811	}
7812	}
7813
7814	if (fWhat & CPUMCTX_EXTRN_SYSENTER_MSRS)
7815	{
7816	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SYSENTER_EIP, &pCtx->SysEnter.eip); AssertRC(rc);
7817	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SYSENTER_ESP, &pCtx->SysEnter.esp); AssertRC(rc);
7818	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, &u32Val); AssertRC(rc);
7819	pCtx->SysEnter.cs = u32Val;
7820	}
7821
7822	if (fWhat & CPUMCTX_EXTRN_KERNEL_GS_BASE)
7823	{
7824	if ( pVM->hmr0.s.fAllow64BitGuests
7825	&& (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
7826	pCtx->msrKERNELGSBASE = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
7827	}
7828
7829	if (fWhat & CPUMCTX_EXTRN_SYSCALL_MSRS)
7830	{
7831	if ( pVM->hmr0.s.fAllow64BitGuests
7832	&& (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
7833	{
7834	pCtx->msrLSTAR = ASMRdMsr(MSR_K8_LSTAR);
7835	pCtx->msrSTAR = ASMRdMsr(MSR_K6_STAR);
7836	pCtx->msrSFMASK = ASMRdMsr(MSR_K8_SF_MASK);
7837	}
7838	}
7839
7840	if (fWhat & (CPUMCTX_EXTRN_TSC_AUX \| CPUMCTX_EXTRN_OTHER_MSRS))
7841	{
7842	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
7843	PCVMXAUTOMSR pMsrs = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
7844	uint32_t const cMsrs = pVmcsInfo->cExitMsrStore;
7845	Assert(pMsrs);
7846	Assert(cMsrs <= VMX_MISC_MAX_MSRS(g_HmMsrs.u.vmx.u64Misc));
7847	Assert(sizeof(pMsrs) cMsrs <= X86_PAGE_4K_SIZE);
7848	for (uint32_t i = 0; i < cMsrs; i++)
7849	{
7850	uint32_t const idMsr = pMsrs[i].u32Msr;
7851	switch (idMsr)
7852	{
7853	case MSR_K8_TSC_AUX: CPUMSetGuestTscAux(pVCpu, pMsrs[i].u64Value); break;
7854	case MSR_IA32_SPEC_CTRL: CPUMSetGuestSpecCtrl(pVCpu, pMsrs[i].u64Value); break;
7855	case MSR_K6_EFER: /* Can't be changed without causing a VM-exit */ break;
7856	default:
7857	{
7858	uint32_t idxLbrMsr;
7859	if (pVM->hmr0.s.vmx.fLbr)
7860	{
7861	if (hmR0VmxIsLbrBranchFromMsr(pVM, idMsr, &idxLbrMsr))
7862	{
7863	Assert(idxLbrMsr < RT_ELEMENTS(pVmcsInfoShared->au64LbrFromIpMsr));
7864	pVmcsInfoShared->au64LbrFromIpMsr[idxLbrMsr] = pMsrs[i].u64Value;
7865	break;
7866	}
7867	if (hmR0VmxIsLbrBranchToMsr(pVM, idMsr, &idxLbrMsr))
7868	{
7869	Assert(idxLbrMsr < RT_ELEMENTS(pVmcsInfoShared->au64LbrFromIpMsr));
7870	pVmcsInfoShared->au64LbrToIpMsr[idxLbrMsr] = pMsrs[i].u64Value;
7871	break;
7872	}
7873	if (idMsr == pVM->hmr0.s.vmx.idLbrTosMsr)
7874	{
7875	pVmcsInfoShared->u64LbrTosMsr = pMsrs[i].u64Value;
7876	break;
7877	}
7878	/* Fallthru (no break) */
7879	}
7880	pCtx->fExtrn = 0;
7881	pVCpu->hm.s.u32HMError = pMsrs->u32Msr;
7882	ASMSetFlags(fEFlags);
7883	AssertMsgFailed(("Unexpected MSR in auto-load/store area. idMsr=%#RX32 cMsrs=%u\n", idMsr, cMsrs));
7884	return VERR_HM_UNEXPECTED_LD_ST_MSR;
7885	}
7886	}
7887	}
7888	}
7889
7890	if (fWhat & CPUMCTX_EXTRN_CR_MASK)
7891	{
7892	if (fWhat & CPUMCTX_EXTRN_CR0)
7893	{
7894	uint64_t u64Cr0;
7895	uint64_t u64Shadow;
7896	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64Cr0); AssertRC(rc);
7897	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, &u64Shadow); AssertRC(rc);
7898	#ifndef VBOX_WITH_NESTED_HWVIRT_VMX
7899	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7900	\| (u64Shadow & pVmcsInfo->u64Cr0Mask);
7901	#else
7902	if (!CPUMIsGuestInVmxNonRootMode(pCtx))
7903	{
7904	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7905	\| (u64Shadow & pVmcsInfo->u64Cr0Mask);
7906	}
7907	else
7908	{
7909	/*
7910	* We've merged the guest and nested-guest's CR0 guest/host mask while executing
7911	* the nested-guest using hardware-assisted VMX. Accordingly we need to
7912	* re-construct CR0. See @bugref{9180#c95} for details.
7913	*/
7914	PCVMXVMCSINFO const pVmcsInfoGst = &pVCpu->hmr0.s.vmx.VmcsInfo;
7915	PVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
7916	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7917	\| (pVmcsNstGst->u64GuestCr0.u & pVmcsNstGst->u64Cr0Mask.u)
7918	\| (u64Shadow & (pVmcsInfoGst->u64Cr0Mask & ~pVmcsNstGst->u64Cr0Mask.u));
7919	}
7920	#endif
7921	VMMRZCallRing3Disable(pVCpu); /* May call into PGM which has Log statements. */
7922	CPUMSetGuestCR0(pVCpu, u64Cr0);
7923	VMMRZCallRing3Enable(pVCpu);
7924	}
7925
7926	if (fWhat & CPUMCTX_EXTRN_CR4)
7927	{
7928	uint64_t u64Cr4;
7929	uint64_t u64Shadow;
7930	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR4, &u64Cr4); AssertRC(rc);
7931	rc \|= VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, &u64Shadow); AssertRC(rc);
7932	#ifndef VBOX_WITH_NESTED_HWVIRT_VMX
7933	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7934	\| (u64Shadow & pVmcsInfo->u64Cr4Mask);
7935	#else
7936	if (!CPUMIsGuestInVmxNonRootMode(pCtx))
7937	{
7938	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7939	\| (u64Shadow & pVmcsInfo->u64Cr4Mask);
7940	}
7941	else
7942	{
7943	/*
7944	* We've merged the guest and nested-guest's CR4 guest/host mask while executing
7945	* the nested-guest using hardware-assisted VMX. Accordingly we need to
7946	* re-construct CR4. See @bugref{9180#c95} for details.
7947	*/
7948	PCVMXVMCSINFO const pVmcsInfoGst = &pVCpu->hmr0.s.vmx.VmcsInfo;
7949	PVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
7950	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7951	\| (pVmcsNstGst->u64GuestCr4.u & pVmcsNstGst->u64Cr4Mask.u)
7952	\| (u64Shadow & (pVmcsInfoGst->u64Cr4Mask & ~pVmcsNstGst->u64Cr4Mask.u));
7953	}
7954	#endif
7955	pCtx->cr4 = u64Cr4;
7956	}
7957
7958	if (fWhat & CPUMCTX_EXTRN_CR3)
7959	{
7960	/* CR0.PG bit changes are always intercepted, so it's up to date. */
7961	if ( pVM->hmr0.s.vmx.fUnrestrictedGuest
7962	\|\| ( pVM->hmr0.s.fNestedPaging
7963	&& CPUMIsGuestPagingEnabledEx(pCtx)))
7964	{
7965	uint64_t u64Cr3;
7966	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR3, &u64Cr3); AssertRC(rc);
7967	if (pCtx->cr3 != u64Cr3)
7968	{
7969	pCtx->cr3 = u64Cr3;
7970	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
7971	}
7972
7973	/*
7974	* If the guest is in PAE mode, sync back the PDPE's into the guest state.
7975	* CR4.PAE, CR0.PG, EFER MSR changes are always intercepted, so they're up to date.
7976	*/
7977	if (CPUMIsGuestInPAEModeEx(pCtx))
7978	{
7979	X86PDPE aPaePdpes[4];
7980	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &aPaePdpes[0].u); AssertRC(rc);
7981	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &aPaePdpes[1].u); AssertRC(rc);
7982	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &aPaePdpes[2].u); AssertRC(rc);
7983	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &aPaePdpes[3].u); AssertRC(rc);
7984	if (memcmp(&aPaePdpes[0], &pCtx->aPaePdpes[0], sizeof(aPaePdpes)))
7985	{
7986	memcpy(&pCtx->aPaePdpes[0], &aPaePdpes[0], sizeof(aPaePdpes));
7987	/* PGM now updates PAE PDPTEs while updating CR3. */
7988	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
7989	}
7990	}
7991	}
7992	}
7993	}
7994
7995	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7996	if (fWhat & CPUMCTX_EXTRN_HWVIRT)
7997	{
7998	if ( (pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
7999	&& !CPUMIsGuestInVmxNonRootMode(pCtx))
8000	{
8001	Assert(CPUMIsGuestInVmxRootMode(pCtx));
8002	rc = hmR0VmxCopyShadowToNstGstVmcs(pVCpu, pVmcsInfo);
8003	if (RT_SUCCESS(rc))
8004	{ /* likely */ }
8005	else
8006	break;
8007	}
8008	}
8009	#endif
8010	} while (0);
8011
8012	if (RT_SUCCESS(rc))
8013	{
8014	/* Update fExtrn. */
8015	pCtx->fExtrn &= ~fWhat;
8016
8017	/* If everything has been imported, clear the HM keeper bit. */
8018	if (!(pCtx->fExtrn & HMVMX_CPUMCTX_EXTRN_ALL))
8019	{
8020	pCtx->fExtrn &= ~CPUMCTX_EXTRN_KEEPER_HM;
8021	Assert(!pCtx->fExtrn);
8022	}
8023	}
8024	}
8025	else
8026	AssertMsg(!pCtx->fExtrn \|\| (pCtx->fExtrn & HMVMX_CPUMCTX_EXTRN_ALL), ("%#RX64\n", pCtx->fExtrn));
8027
8028	/*
8029	* Restore interrupts.
8030	*/
8031	ASMSetFlags(fEFlags);
8032
8033	STAM_PROFILE_ADV_STOP(& pVCpu->hm.s.StatImportGuestState, x);
8034
8035	if (RT_SUCCESS(rc))
8036	{ /* likely */ }
8037	else
8038	return rc;
8039
8040	/*
8041	* Honor any pending CR3 updates.
8042	*
8043	* Consider this scenario: VM-exit -> VMMRZCallRing3Enable() -> do stuff that causes a longjmp -> VMXR0CallRing3Callback()
8044	* -> VMMRZCallRing3Disable() -> hmR0VmxImportGuestState() -> Sets VMCPU_FF_HM_UPDATE_CR3 pending -> return from the longjmp
8045	* -> continue with VM-exit handling -> hmR0VmxImportGuestState() and here we are.
8046	*
8047	* The reason for such complicated handling is because VM-exits that call into PGM expect CR3 to be up-to-date and thus
8048	* if any CR3-saves -before- the VM-exit (longjmp) postponed the CR3 update via the force-flag, any VM-exit handler that
8049	* calls into PGM when it re-saves CR3 will end up here and we call PGMUpdateCR3(). This is why the code below should
8050	* -NOT- check if CPUMCTX_EXTRN_CR3 is set!
8051	*
8052	* The longjmp exit path can't check these CR3 force-flags and call code that takes a lock again. We cover for it here.
8053	*
8054	* The force-flag is checked first as it's cheaper for potential superfluous calls to this function.
8055	*/
8056	if ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3)
8057	&& VMMRZCallRing3IsEnabled(pVCpu))
8058	{
8059	Assert(!(ASMAtomicUoReadU64(&pCtx->fExtrn) & CPUMCTX_EXTRN_CR3));
8060	PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
8061	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
8062	}
8063
8064	return VINF_SUCCESS;
8065	}
8066
8067
8068	/**
8069	* Saves the guest state from the VMCS into the guest-CPU context.
8070	*
8071	* @returns VBox status code.
8072	* @param pVCpu The cross context virtual CPU structure.
8073	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
8074	*/
8075	VMMR0DECL(int) VMXR0ImportStateOnDemand(PVMCPUCC pVCpu, uint64_t fWhat)
8076	{
8077	AssertPtr(pVCpu);
8078	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8079	return hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fWhat);
8080	}
8081
8082
8083	/**
8084	* Check per-VM and per-VCPU force flag actions that require us to go back to
8085	* ring-3 for one reason or another.
8086	*
8087	* @returns Strict VBox status code (i.e. informational status codes too)
8088	* @retval VINF_SUCCESS if we don't have any actions that require going back to
8089	* ring-3.
8090	* @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync.
8091	* @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware
8092	* interrupts)
8093	* @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires
8094	* all EMTs to be in ring-3.
8095	* @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests.
8096	* @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return
8097	* to the EM loop.
8098	*
8099	* @param pVCpu The cross context virtual CPU structure.
8100	* @param pVmxTransient The VMX-transient structure.
8101	* @param fStepping Whether we are single-stepping the guest using the
8102	* hypervisor debugger.
8103	*
8104	* @remarks This might cause nested-guest VM-exits, caller must check if the guest
8105	* is no longer in VMX non-root mode.
8106	*/
8107	static VBOXSTRICTRC hmR0VmxCheckForceFlags(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, bool fStepping)
8108	{
8109	Assert(VMMRZCallRing3IsEnabled(pVCpu));
8110
8111	/*
8112	* Update pending interrupts into the APIC's IRR.
8113	*/
8114	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UPDATE_APIC))
8115	APICUpdatePendingInterrupts(pVCpu);
8116
8117	/*
8118	* Anything pending? Should be more likely than not if we're doing a good job.
8119	*/
8120	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
8121	if ( !fStepping
8122	? !VM_FF_IS_ANY_SET(pVM, VM_FF_HP_R0_PRE_HM_MASK)
8123	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HP_R0_PRE_HM_MASK)
8124	: !VM_FF_IS_ANY_SET(pVM, VM_FF_HP_R0_PRE_HM_STEP_MASK)
8125	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) )
8126	return VINF_SUCCESS;
8127
8128	/* Pending PGM C3 sync. */
8129	if (VMCPU_FF_IS_ANY_SET(pVCpu,VMCPU_FF_PGM_SYNC_CR3 \| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL))
8130	{
8131	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8132	Assert(!(ASMAtomicUoReadU64(&pCtx->fExtrn) & (CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_CR3 \| CPUMCTX_EXTRN_CR4)));
8133	VBOXSTRICTRC rcStrict = PGMSyncCR3(pVCpu, pCtx->cr0, pCtx->cr3, pCtx->cr4,
8134	VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
8135	if (rcStrict != VINF_SUCCESS)
8136	{
8137	AssertRC(VBOXSTRICTRC_VAL(rcStrict));
8138	Log4Func(("PGMSyncCR3 forcing us back to ring-3. rc2=%d\n", VBOXSTRICTRC_VAL(rcStrict)));
8139	return rcStrict;
8140	}
8141	}
8142
8143	/* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */
8144	if ( VM_FF_IS_ANY_SET(pVM, VM_FF_HM_TO_R3_MASK)
8145	\|\| VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
8146	{
8147	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
8148	int rc = RT_LIKELY(!VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_RAW_TO_R3 : VINF_EM_NO_MEMORY;
8149	Log4Func(("HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc));
8150	return rc;
8151	}
8152
8153	/* Pending VM request packets, such as hardware interrupts. */
8154	if ( VM_FF_IS_SET(pVM, VM_FF_REQUEST)
8155	\|\| VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_REQUEST))
8156	{
8157	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchVmReq);
8158	Log4Func(("Pending VM request forcing us back to ring-3\n"));
8159	return VINF_EM_PENDING_REQUEST;
8160	}
8161
8162	/* Pending PGM pool flushes. */
8163	if (VM_FF_IS_SET(pVM, VM_FF_PGM_POOL_FLUSH_PENDING))
8164	{
8165	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPgmPoolFlush);
8166	Log4Func(("PGM pool flush pending forcing us back to ring-3\n"));
8167	return VINF_PGM_POOL_FLUSH_PENDING;
8168	}
8169
8170	/* Pending DMA requests. */
8171	if (VM_FF_IS_SET(pVM, VM_FF_PDM_DMA))
8172	{
8173	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchDma);
8174	Log4Func(("Pending DMA request forcing us back to ring-3\n"));
8175	return VINF_EM_RAW_TO_R3;
8176	}
8177
8178	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8179	/*
8180	* Pending nested-guest events.
8181	*
8182	* Please note the priority of these events are specified and important.
8183	* See Intel spec. 29.4.3.2 "APIC-Write Emulation".
8184	* See Intel spec. 6.9 "Priority Among Simultaneous Exceptions And Interrupts".
8185	*/
8186	if (pVmxTransient->fIsNestedGuest)
8187	{
8188	/* Pending nested-guest APIC-write. */
8189	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_APIC_WRITE))
8190	{
8191	Log4Func(("Pending nested-guest APIC-write\n"));
8192	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitApicWrite(pVCpu);
8193	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8194	return rcStrict;
8195	}
8196
8197	/* Pending nested-guest monitor-trap flag (MTF). */
8198	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_MTF))
8199	{
8200	Log4Func(("Pending nested-guest MTF\n"));
8201	VBOXSTRICTRC rcStrict = IEMExecVmxVmexit(pVCpu, VMX_EXIT_MTF, 0 /* uExitQual */);
8202	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8203	return rcStrict;
8204	}
8205
8206	/* Pending nested-guest VMX-preemption timer expired. */
8207	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_PREEMPT_TIMER))
8208	{
8209	Log4Func(("Pending nested-guest preempt timer\n"));
8210	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitPreemptTimer(pVCpu);
8211	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8212	return rcStrict;
8213	}
8214	}
8215	#else
8216	NOREF(pVmxTransient);
8217	#endif
8218
8219	return VINF_SUCCESS;
8220	}
8221
8222
8223	/**
8224	* Converts any TRPM trap into a pending HM event. This is typically used when
8225	* entering from ring-3 (not longjmp returns).
8226	*
8227	* @param pVCpu The cross context virtual CPU structure.
8228	*/
8229	static void hmR0VmxTrpmTrapToPendingEvent(PVMCPUCC pVCpu)
8230	{
8231	Assert(TRPMHasTrap(pVCpu));
8232	Assert(!pVCpu->hm.s.Event.fPending);
8233
8234	uint8_t uVector;
8235	TRPMEVENT enmTrpmEvent;
8236	uint32_t uErrCode;
8237	RTGCUINTPTR GCPtrFaultAddress;
8238	uint8_t cbInstr;
8239	bool fIcebp;
8240
8241	int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr, &fIcebp);
8242	AssertRC(rc);
8243
8244	uint32_t u32IntInfo;
8245	u32IntInfo = uVector \| VMX_IDT_VECTORING_INFO_VALID;
8246	u32IntInfo \|= HMTrpmEventTypeToVmxEventType(uVector, enmTrpmEvent, fIcebp);
8247
8248	rc = TRPMResetTrap(pVCpu);
8249	AssertRC(rc);
8250	Log4(("TRPM->HM event: u32IntInfo=%#RX32 enmTrpmEvent=%d cbInstr=%u uErrCode=%#RX32 GCPtrFaultAddress=%#RGv\n",
8251	u32IntInfo, enmTrpmEvent, cbInstr, uErrCode, GCPtrFaultAddress));
8252
8253	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, uErrCode, GCPtrFaultAddress);
8254	}
8255
8256
8257	/**
8258	* Converts the pending HM event into a TRPM trap.
8259	*
8260	* @param pVCpu The cross context virtual CPU structure.
8261	*/
8262	static void hmR0VmxPendingEventToTrpmTrap(PVMCPUCC pVCpu)
8263	{
8264	Assert(pVCpu->hm.s.Event.fPending);
8265
8266	/* If a trap was already pending, we did something wrong! */
8267	Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo /, NULL / pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP);
8268
8269	uint32_t const u32IntInfo = pVCpu->hm.s.Event.u64IntInfo;
8270	uint32_t const uVector = VMX_IDT_VECTORING_INFO_VECTOR(u32IntInfo);
8271	TRPMEVENT const enmTrapType = HMVmxEventTypeToTrpmEventType(u32IntInfo);
8272
8273	Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, enmTrapType));
8274
8275	int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType);
8276	AssertRC(rc);
8277
8278	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(u32IntInfo))
8279	TRPMSetErrorCode(pVCpu, pVCpu->hm.s.Event.u32ErrCode);
8280
8281	if (VMX_IDT_VECTORING_INFO_IS_XCPT_PF(u32IntInfo))
8282	TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress);
8283	else
8284	{
8285	uint8_t const uVectorType = VMX_IDT_VECTORING_INFO_TYPE(u32IntInfo);
8286	switch (uVectorType)
8287	{
8288	case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
8289	TRPMSetTrapDueToIcebp(pVCpu);
8290	RT_FALL_THRU();
8291	case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
8292	case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT:
8293	{
8294	AssertMsg( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
8295	\|\| ( uVector == X86_XCPT_BP /* INT3 */
8296	\|\| uVector == X86_XCPT_OF /* INTO */
8297	\|\| uVector == X86_XCPT_DB /* INT1 (ICEBP) */),
8298	("Invalid vector: uVector=%#x uVectorType=%#x\n", uVector, uVectorType));
8299	TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr);
8300	break;
8301	}
8302	}
8303	}
8304
8305	/* We're now done converting the pending event. */
8306	pVCpu->hm.s.Event.fPending = false;
8307	}
8308
8309
8310	/**
8311	* Sets the interrupt-window exiting control in the VMCS which instructs VT-x to
8312	* cause a VM-exit as soon as the guest is in a state to receive interrupts.
8313	*
8314	* @param pVmcsInfo The VMCS info. object.
8315	*/
8316	static void hmR0VmxSetIntWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8317	{
8318	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_INT_WINDOW_EXIT)
8319	{
8320	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT))
8321	{
8322	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_INT_WINDOW_EXIT;
8323	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8324	AssertRC(rc);
8325	}
8326	} /* else we will deliver interrupts whenever the guest Vm-exits next and is in a state to receive the interrupt. */
8327	}
8328
8329
8330	/**
8331	* Clears the interrupt-window exiting control in the VMCS.
8332	*
8333	* @param pVmcsInfo The VMCS info. object.
8334	*/
8335	DECLINLINE(void) hmR0VmxClearIntWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8336	{
8337	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT)
8338	{
8339	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_INT_WINDOW_EXIT;
8340	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8341	AssertRC(rc);
8342	}
8343	}
8344
8345
8346	/**
8347	* Sets the NMI-window exiting control in the VMCS which instructs VT-x to
8348	* cause a VM-exit as soon as the guest is in a state to receive NMIs.
8349	*
8350	* @param pVmcsInfo The VMCS info. object.
8351	*/
8352	static void hmR0VmxSetNmiWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8353	{
8354	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
8355	{
8356	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))
8357	{
8358	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_NMI_WINDOW_EXIT;
8359	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8360	AssertRC(rc);
8361	Log4Func(("Setup NMI-window exiting\n"));
8362	}
8363	} /* else we will deliver NMIs whenever we VM-exit next, even possibly nesting NMIs. Can't be helped on ancient CPUs. */
8364	}
8365
8366
8367	/**
8368	* Clears the NMI-window exiting control in the VMCS.
8369	*
8370	* @param pVmcsInfo The VMCS info. object.
8371	*/
8372	DECLINLINE(void) hmR0VmxClearNmiWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8373	{
8374	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
8375	{
8376	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_NMI_WINDOW_EXIT;
8377	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8378	AssertRC(rc);
8379	}
8380	}
8381
8382
8383	/**
8384	* Does the necessary state syncing before returning to ring-3 for any reason
8385	* (longjmp, preemption, voluntary exits to ring-3) from VT-x.
8386	*
8387	* @returns VBox status code.
8388	* @param pVCpu The cross context virtual CPU structure.
8389	* @param fImportState Whether to import the guest state from the VMCS back
8390	* to the guest-CPU context.
8391	*
8392	* @remarks No-long-jmp zone!!!
8393	*/
8394	static int hmR0VmxLeave(PVMCPUCC pVCpu, bool fImportState)
8395	{
8396	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8397	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8398
8399	RTCPUID const idCpu = RTMpCpuId();
8400	Log4Func(("HostCpuId=%u\n", idCpu));
8401
8402	/*
8403	* !!! IMPORTANT !!!
8404	* If you modify code here, check whether VMXR0CallRing3Callback() needs to be updated too.
8405	*/
8406
8407	/* Save the guest state if necessary. */
8408	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8409	if (fImportState)
8410	{
8411	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
8412	AssertRCReturn(rc, rc);
8413	}
8414
8415	/* Restore host FPU state if necessary. We will resync on next R0 reentry. */
8416	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
8417	Assert(!CPUMIsGuestFPUStateActive(pVCpu));
8418
8419	/* Restore host debug registers if necessary. We will resync on next R0 reentry. */
8420	#ifdef VBOX_STRICT
8421	if (CPUMIsHyperDebugStateActive(pVCpu))
8422	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MOV_DR_EXIT);
8423	#endif
8424	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
8425	Assert(!CPUMIsGuestDebugStateActive(pVCpu));
8426	Assert(!CPUMIsHyperDebugStateActive(pVCpu));
8427
8428	/* Restore host-state bits that VT-x only restores partially. */
8429	if (pVCpu->hmr0.s.vmx.fRestoreHostFlags > VMX_RESTORE_HOST_REQUIRED)
8430	{
8431	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hmr0.s.vmx.fRestoreHostFlags, idCpu));
8432	VMXRestoreHostState(pVCpu->hmr0.s.vmx.fRestoreHostFlags, &pVCpu->hmr0.s.vmx.RestoreHost);
8433	}
8434	pVCpu->hmr0.s.vmx.fRestoreHostFlags = 0;
8435
8436	/* Restore the lazy host MSRs as we're leaving VT-x context. */
8437	if (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
8438	{
8439	/* We shouldn't restore the host MSRs without saving the guest MSRs first. */
8440	if (!fImportState)
8441	{
8442	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_KERNEL_GS_BASE \| CPUMCTX_EXTRN_SYSCALL_MSRS);
8443	AssertRCReturn(rc, rc);
8444	}
8445	hmR0VmxLazyRestoreHostMsrs(pVCpu);
8446	Assert(!pVCpu->hmr0.s.vmx.fLazyMsrs);
8447	}
8448	else
8449	pVCpu->hmr0.s.vmx.fLazyMsrs = 0;
8450
8451	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
8452	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = false;
8453
8454	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry);
8455	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatImportGuestState);
8456	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExportGuestState);
8457	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatPreExit);
8458	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitHandling);
8459	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitIO);
8460	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitMovCRx);
8461	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitXcptNmi);
8462	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitVmentry);
8463	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
8464
8465	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
8466
8467	/** @todo This partially defeats the purpose of having preemption hooks.
8468	* The problem is, deregistering the hooks should be moved to a place that
8469	* lasts until the EMT is about to be destroyed not everytime while leaving HM
8470	* context.
8471	*/
8472	int rc = hmR0VmxClearVmcs(pVmcsInfo);
8473	AssertRCReturn(rc, rc);
8474
8475	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8476	/*
8477	* A valid shadow VMCS is made active as part of VM-entry. It is necessary to
8478	* clear a shadow VMCS before allowing that VMCS to become active on another
8479	* logical processor. We may or may not be importing guest state which clears
8480	* it, so cover for it here.
8481	*
8482	* See Intel spec. 24.11.1 "Software Use of Virtual-Machine Control Structures".
8483	*/
8484	if ( pVmcsInfo->pvShadowVmcs
8485	&& pVmcsInfo->fShadowVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
8486	{
8487	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
8488	AssertRCReturn(rc, rc);
8489	}
8490
8491	/*
8492	* Flag that we need to re-export the host state if we switch to this VMCS before
8493	* executing guest or nested-guest code.
8494	*/
8495	pVmcsInfo->idHostCpuState = NIL_RTCPUID;
8496	#endif
8497
8498	Log4Func(("Cleared Vmcs. HostCpuId=%u\n", idCpu));
8499	NOREF(idCpu);
8500	return VINF_SUCCESS;
8501	}
8502
8503
8504	/**
8505	* Leaves the VT-x session.
8506	*
8507	* @returns VBox status code.
8508	* @param pVCpu The cross context virtual CPU structure.
8509	*
8510	* @remarks No-long-jmp zone!!!
8511	*/
8512	static int hmR0VmxLeaveSession(PVMCPUCC pVCpu)
8513	{
8514	HM_DISABLE_PREEMPT(pVCpu);
8515	HMVMX_ASSERT_CPU_SAFE(pVCpu);
8516	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8517	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8518
8519	/* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before
8520	and done this from the VMXR0ThreadCtxCallback(). */
8521	if (!pVCpu->hmr0.s.fLeaveDone)
8522	{
8523	int rc2 = hmR0VmxLeave(pVCpu, true /* fImportState */);
8524	AssertRCReturnStmt(rc2, HM_RESTORE_PREEMPT(), rc2);
8525	pVCpu->hmr0.s.fLeaveDone = true;
8526	}
8527	Assert(!pVCpu->cpum.GstCtx.fExtrn);
8528
8529	/*
8530	* !!! IMPORTANT !!!
8531	* If you modify code here, make sure to check whether VMXR0CallRing3Callback() needs to be updated too.
8532	*/
8533
8534	/* Deregister hook now that we've left HM context before re-enabling preemption. */
8535	/** @todo Deregistering here means we need to VMCLEAR always
8536	* (longjmp/exit-to-r3) in VT-x which is not efficient, eliminate need
8537	* for calling VMMR0ThreadCtxHookDisable here! */
8538	VMMR0ThreadCtxHookDisable(pVCpu);
8539
8540	/* Leave HM context. This takes care of local init (term) and deregistering the longjmp-to-ring-3 callback. */
8541	int rc = HMR0LeaveCpu(pVCpu);
8542	HM_RESTORE_PREEMPT();
8543	return rc;
8544	}
8545
8546
8547	/**
8548	* Take necessary actions before going back to ring-3.
8549	*
8550	* An action requires us to go back to ring-3. This function does the necessary
8551	* steps before we can safely return to ring-3. This is not the same as longjmps
8552	* to ring-3, this is voluntary and prepares the guest so it may continue
8553	* executing outside HM (recompiler/IEM).
8554	*
8555	* @returns VBox status code.
8556	* @param pVCpu The cross context virtual CPU structure.
8557	* @param rcExit The reason for exiting to ring-3. Can be
8558	* VINF_VMM_UNKNOWN_RING3_CALL.
8559	*/
8560	static int hmR0VmxExitToRing3(PVMCPUCC pVCpu, VBOXSTRICTRC rcExit)
8561	{
8562	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
8563
8564	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8565	if (RT_UNLIKELY(rcExit == VERR_VMX_INVALID_VMCS_PTR))
8566	{
8567	VMXGetCurrentVmcs(&pVCpu->hm.s.vmx.LastError.HCPhysCurrentVmcs);
8568	pVCpu->hm.s.vmx.LastError.u32VmcsRev = (uint32_t )pVmcsInfo->pvVmcs;
8569	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hmr0.s.idEnteredCpu;
8570	/* LastError.idCurrentCpu was updated in hmR0VmxPreRunGuestCommitted(). */
8571	}
8572
8573	/* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */
8574	VMMRZCallRing3Disable(pVCpu);
8575	Log4Func(("rcExit=%d\n", VBOXSTRICTRC_VAL(rcExit)));
8576
8577	/*
8578	* Convert any pending HM events back to TRPM due to premature exits to ring-3.
8579	* We need to do this only on returns to ring-3 and not for longjmps to ring3.
8580	*
8581	* This is because execution may continue from ring-3 and we would need to inject
8582	* the event from there (hence place it back in TRPM).
8583	*/
8584	if (pVCpu->hm.s.Event.fPending)
8585	{
8586	hmR0VmxPendingEventToTrpmTrap(pVCpu);
8587	Assert(!pVCpu->hm.s.Event.fPending);
8588
8589	/* Clear the events from the VMCS. */
8590	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, 0); AssertRC(rc);
8591	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, 0); AssertRC(rc);
8592	}
8593	#ifdef VBOX_STRICT
8594	/*
8595	* We check for rcExit here since for errors like VERR_VMX_UNABLE_TO_START_VM (which are
8596	* fatal), we don't care about verifying duplicate injection of events. Errors like
8597	* VERR_EM_INTERPRET are converted to their VINF_* counterparts -prior- to calling this
8598	* function so those should and will be checked below.
8599	*/
8600	else if (RT_SUCCESS(rcExit))
8601	{
8602	/*
8603	* Ensure we don't accidentally clear a pending HM event without clearing the VMCS.
8604	* This can be pretty hard to debug otherwise, interrupts might get injected twice
8605	* occasionally, see @bugref{9180#c42}.
8606	*
8607	* However, if the VM-entry failed, any VM entry-interruption info. field would
8608	* be left unmodified as the event would not have been injected to the guest. In
8609	* such cases, don't assert, we're not going to continue guest execution anyway.
8610	*/
8611	uint32_t uExitReason;
8612	uint32_t uEntryIntInfo;
8613	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
8614	rc \|= VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &uEntryIntInfo);
8615	AssertRC(rc);
8616	AssertMsg(VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason) \|\| !VMX_ENTRY_INT_INFO_IS_VALID(uEntryIntInfo),
8617	("uExitReason=%#RX32 uEntryIntInfo=%#RX32 rcExit=%d\n", uExitReason, uEntryIntInfo, VBOXSTRICTRC_VAL(rcExit)));
8618	}
8619	#endif
8620
8621	/*
8622	* Clear the interrupt-window and NMI-window VMCS controls as we could have got
8623	* a VM-exit with higher priority than interrupt-window or NMI-window VM-exits
8624	* (e.g. TPR below threshold).
8625	*/
8626	if (!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
8627	{
8628	hmR0VmxClearIntWindowExitVmcs(pVmcsInfo);
8629	hmR0VmxClearNmiWindowExitVmcs(pVmcsInfo);
8630	}
8631
8632	/* If we're emulating an instruction, we shouldn't have any TRPM traps pending
8633	and if we're injecting an event we should have a TRPM trap pending. */
8634	AssertMsg(rcExit != VINF_EM_RAW_INJECT_TRPM_EVENT \|\| TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
8635	#ifndef DEBUG_bird /* Triggered after firing an NMI against NT4SP1, possibly a triple fault in progress. */
8636	AssertMsg(rcExit != VINF_EM_RAW_EMULATE_INSTR \|\| !TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
8637	#endif
8638
8639	/* Save guest state and restore host state bits. */
8640	int rc = hmR0VmxLeaveSession(pVCpu);
8641	AssertRCReturn(rc, rc);
8642	STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
8643
8644	/* Thread-context hooks are unregistered at this point!!! */
8645	/* Ring-3 callback notifications are unregistered at this point!!! */
8646
8647	/* Sync recompiler state. */
8648	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3);
8649	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR
8650	\| CPUM_CHANGED_LDTR
8651	\| CPUM_CHANGED_GDTR
8652	\| CPUM_CHANGED_IDTR
8653	\| CPUM_CHANGED_TR
8654	\| CPUM_CHANGED_HIDDEN_SEL_REGS);
8655	if ( pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging
8656	&& CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx))
8657	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH);
8658
8659	Assert(!pVCpu->hmr0.s.fClearTrapFlag);
8660
8661	/* Update the exit-to-ring 3 reason. */
8662	pVCpu->hm.s.rcLastExitToR3 = VBOXSTRICTRC_VAL(rcExit);
8663
8664	/* On our way back from ring-3 reload the guest state if there is a possibility of it being changed. */
8665	if ( rcExit != VINF_EM_RAW_INTERRUPT
8666	\|\| CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
8667	{
8668	Assert(!(pVCpu->cpum.GstCtx.fExtrn & HMVMX_CPUMCTX_EXTRN_ALL));
8669	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
8670	}
8671
8672	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3);
8673	VMMRZCallRing3Enable(pVCpu);
8674	return rc;
8675	}
8676
8677
8678	/**
8679	* VMMRZCallRing3() callback wrapper which saves the guest state before we
8680	* longjump due to a ring-0 assertion.
8681	*
8682	* @returns VBox status code.
8683	* @param pVCpu The cross context virtual CPU structure.
8684	*/
8685	VMMR0DECL(int) VMXR0AssertionCallback(PVMCPUCC pVCpu)
8686	{
8687	/*
8688	* !!! IMPORTANT !!!
8689	* If you modify code here, check whether hmR0VmxLeave() and hmR0VmxLeaveSession() needs to be updated too.
8690	* This is a stripped down version which gets out ASAP, trying to not trigger any further assertions.
8691	*/
8692	VMMR0AssertionRemoveNotification(pVCpu);
8693	VMMRZCallRing3Disable(pVCpu);
8694	HM_DISABLE_PREEMPT(pVCpu);
8695
8696	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8697	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
8698	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
8699	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
8700
8701	/* Restore host-state bits that VT-x only restores partially. */
8702	if (pVCpu->hmr0.s.vmx.fRestoreHostFlags > VMX_RESTORE_HOST_REQUIRED)
8703	VMXRestoreHostState(pVCpu->hmr0.s.vmx.fRestoreHostFlags, &pVCpu->hmr0.s.vmx.RestoreHost);
8704	pVCpu->hmr0.s.vmx.fRestoreHostFlags = 0;
8705
8706	/* Restore the lazy host MSRs as we're leaving VT-x context. */
8707	if (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
8708	hmR0VmxLazyRestoreHostMsrs(pVCpu);
8709
8710	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
8711	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = false;
8712	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
8713
8714	/* Clear the current VMCS data back to memory (shadow VMCS if any would have been
8715	cleared as part of importing the guest state above. */
8716	hmR0VmxClearVmcs(pVmcsInfo);
8717
8718	/** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
8719	VMMR0ThreadCtxHookDisable(pVCpu);
8720
8721	/* Leave HM context. This takes care of local init (term). */
8722	HMR0LeaveCpu(pVCpu);
8723	HM_RESTORE_PREEMPT();
8724	return VINF_SUCCESS;
8725	}
8726
8727
8728	/**
8729	* Pushes a 2-byte value onto the real-mode (in virtual-8086 mode) guest's
8730	* stack.
8731	*
8732	* @returns Strict VBox status code (i.e. informational status codes too).
8733	* @retval VINF_EM_RESET if pushing a value to the stack caused a triple-fault.
8734	* @param pVCpu The cross context virtual CPU structure.
8735	* @param uValue The value to push to the guest stack.
8736	*/
8737	static VBOXSTRICTRC hmR0VmxRealModeGuestStackPush(PVMCPUCC pVCpu, uint16_t uValue)
8738	{
8739	/*
8740	* The stack limit is 0xffff in real-on-virtual 8086 mode. Real-mode with weird stack limits cannot be run in
8741	* virtual 8086 mode in VT-x. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
8742	* See Intel Instruction reference for PUSH and Intel spec. 22.33.1 "Segment Wraparound".
8743	*/
8744	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8745	if (pCtx->sp == 1)
8746	return VINF_EM_RESET;
8747	pCtx->sp -= sizeof(uint16_t); /* May wrap around which is expected behaviour. */
8748	int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), pCtx->ss.u64Base + pCtx->sp, &uValue, sizeof(uint16_t));
8749	AssertRC(rc);
8750	return rc;
8751	}
8752
8753
8754	/**
8755	* Injects an event into the guest upon VM-entry by updating the relevant fields
8756	* in the VM-entry area in the VMCS.
8757	*
8758	* @returns Strict VBox status code (i.e. informational status codes too).
8759	* @retval VINF_SUCCESS if the event is successfully injected into the VMCS.
8760	* @retval VINF_EM_RESET if event injection resulted in a triple-fault.
8761	*
8762	* @param pVCpu The cross context virtual CPU structure.
8763	* @param pVmxTransient The VMX-transient structure.
8764	* @param pEvent The event being injected.
8765	* @param pfIntrState Pointer to the VT-x guest-interruptibility-state. This
8766	* will be updated if necessary. This cannot not be NULL.
8767	* @param fStepping Whether we're single-stepping guest execution and should
8768	* return VINF_EM_DBG_STEPPED if the event is injected
8769	* directly (registers modified by us, not by hardware on
8770	* VM-entry).
8771	*/
8772	static VBOXSTRICTRC hmR0VmxInjectEventVmcs(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, PCHMEVENT pEvent, bool fStepping,
8773	uint32_t *pfIntrState)
8774	{
8775	/* Intel spec. 24.8.3 "VM-Entry Controls for Event Injection" specifies the interruption-information field to be 32-bits. */
8776	AssertMsg(!RT_HI_U32(pEvent->u64IntInfo), ("%#RX64\n", pEvent->u64IntInfo));
8777	Assert(pfIntrState);
8778
8779	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8780	uint32_t u32IntInfo = pEvent->u64IntInfo;
8781	uint32_t const u32ErrCode = pEvent->u32ErrCode;
8782	uint32_t const cbInstr = pEvent->cbInstr;
8783	RTGCUINTPTR const GCPtrFault = pEvent->GCPtrFaultAddress;
8784	uint8_t const uVector = VMX_ENTRY_INT_INFO_VECTOR(u32IntInfo);
8785	uint32_t const uIntType = VMX_ENTRY_INT_INFO_TYPE(u32IntInfo);
8786
8787	#ifdef VBOX_STRICT
8788	/*
8789	* Validate the error-code-valid bit for hardware exceptions.
8790	* No error codes for exceptions in real-mode.
8791	*
8792	* See Intel spec. 20.1.4 "Interrupt and Exception Handling"
8793	*/
8794	if ( uIntType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT
8795	&& !CPUMIsGuestInRealModeEx(pCtx))
8796	{
8797	switch (uVector)
8798	{
8799	case X86_XCPT_PF:
8800	case X86_XCPT_DF:
8801	case X86_XCPT_TS:
8802	case X86_XCPT_NP:
8803	case X86_XCPT_SS:
8804	case X86_XCPT_GP:
8805	case X86_XCPT_AC:
8806	AssertMsg(VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(u32IntInfo),
8807	("Error-code-valid bit not set for exception that has an error code uVector=%#x\n", uVector));
8808	RT_FALL_THRU();
8809	default:
8810	break;
8811	}
8812	}
8813
8814	/* Cannot inject an NMI when block-by-MOV SS is in effect. */
8815	Assert( uIntType != VMX_EXIT_INT_INFO_TYPE_NMI
8816	\|\| !(*pfIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
8817	#endif
8818
8819	if ( uIntType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT
8820	\|\| uIntType == VMX_EXIT_INT_INFO_TYPE_NMI
8821	\|\| uIntType == VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT
8822	\|\| uIntType == VMX_EXIT_INT_INFO_TYPE_SW_XCPT)
8823	{
8824	Assert(uVector <= X86_XCPT_LAST);
8825	Assert(uIntType != VMX_EXIT_INT_INFO_TYPE_NMI \|\| uVector == X86_XCPT_NMI);
8826	Assert(uIntType != VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT \|\| uVector == X86_XCPT_DB);
8827	STAM_COUNTER_INC(&pVCpu->hm.s.aStatInjectedXcpts[uVector]);
8828	}
8829	else
8830	STAM_COUNTER_INC(&pVCpu->hm.s.aStatInjectedIrqs[uVector & MASK_INJECT_IRQ_STAT]);
8831
8832	/*
8833	* Hardware interrupts & exceptions cannot be delivered through the software interrupt
8834	* redirection bitmap to the real mode task in virtual-8086 mode. We must jump to the
8835	* interrupt handler in the (real-mode) guest.
8836	*
8837	* See Intel spec. 20.3 "Interrupt and Exception handling in Virtual-8086 Mode".
8838	* See Intel spec. 20.1.4 "Interrupt and Exception Handling" for real-mode interrupt handling.
8839	*/
8840	if (CPUMIsGuestInRealModeEx(pCtx)) /* CR0.PE bit changes are always intercepted, so it's up to date. */
8841	{
8842	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUnrestrictedGuest)
8843	{
8844	/*
8845	* For CPUs with unrestricted guest execution enabled and with the guest
8846	* in real-mode, we must not set the deliver-error-code bit.
8847	*
8848	* See Intel spec. 26.2.1.3 "VM-Entry Control Fields".
8849	*/
8850	u32IntInfo &= ~VMX_ENTRY_INT_INFO_ERROR_CODE_VALID;
8851	}
8852	else
8853	{
8854	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
8855	Assert(PDMVmmDevHeapIsEnabled(pVM));
8856	Assert(pVM->hm.s.vmx.pRealModeTSS);
8857	Assert(!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
8858
8859	/* We require RIP, RSP, RFLAGS, CS, IDTR, import them. */
8860	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
8861	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_TABLE_MASK
8862	\| CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_RFLAGS);
8863	AssertRCReturn(rc2, rc2);
8864
8865	/* Check if the interrupt handler is present in the IVT (real-mode IDT). IDT limit is (4N - 1). */
8866	size_t const cbIdtEntry = sizeof(X86IDTR16);
8867	if (uVector * cbIdtEntry + (cbIdtEntry - 1) > pCtx->idtr.cbIdt)
8868	{
8869	/* If we are trying to inject a #DF with no valid IDT entry, return a triple-fault. */
8870	if (uVector == X86_XCPT_DF)
8871	return VINF_EM_RESET;
8872
8873	/* If we're injecting a #GP with no valid IDT entry, inject a double-fault.
8874	No error codes for exceptions in real-mode. */
8875	if (uVector == X86_XCPT_GP)
8876	{
8877	uint32_t const uXcptDfInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DF)
8878	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
8879	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
8880	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
8881	HMEVENT EventXcptDf;
8882	RT_ZERO(EventXcptDf);
8883	EventXcptDf.u64IntInfo = uXcptDfInfo;
8884	return hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &EventXcptDf, fStepping, pfIntrState);
8885	}
8886
8887	/*
8888	* If we're injecting an event with no valid IDT entry, inject a #GP.
8889	* No error codes for exceptions in real-mode.
8890	*
8891	* See Intel spec. 20.1.4 "Interrupt and Exception Handling"
8892	*/
8893	uint32_t const uXcptGpInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_GP)
8894	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
8895	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
8896	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
8897	HMEVENT EventXcptGp;
8898	RT_ZERO(EventXcptGp);
8899	EventXcptGp.u64IntInfo = uXcptGpInfo;
8900	return hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &EventXcptGp, fStepping, pfIntrState);
8901	}
8902
8903	/* Software exceptions (#BP and #OF exceptions thrown as a result of INT3 or INTO) */
8904	uint16_t uGuestIp = pCtx->ip;
8905	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_SW_XCPT)
8906	{
8907	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF);
8908	/* #BP and #OF are both benign traps, we need to resume the next instruction. */
8909	uGuestIp = pCtx->ip + (uint16_t)cbInstr;
8910	}
8911	else if (uIntType == VMX_ENTRY_INT_INFO_TYPE_SW_INT)
8912	uGuestIp = pCtx->ip + (uint16_t)cbInstr;
8913
8914	/* Get the code segment selector and offset from the IDT entry for the interrupt handler. */
8915	X86IDTR16 IdtEntry;
8916	RTGCPHYS const GCPhysIdtEntry = (RTGCPHYS)pCtx->idtr.pIdt + uVector * cbIdtEntry;
8917	rc2 = PGMPhysSimpleReadGCPhys(pVM, &IdtEntry, GCPhysIdtEntry, cbIdtEntry);
8918	AssertRCReturn(rc2, rc2);
8919
8920	/* Construct the stack frame for the interrupt/exception handler. */
8921	VBOXSTRICTRC rcStrict;
8922	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, pCtx->eflags.u32);
8923	if (rcStrict == VINF_SUCCESS)
8924	{
8925	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, pCtx->cs.Sel);
8926	if (rcStrict == VINF_SUCCESS)
8927	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, uGuestIp);
8928	}
8929
8930	/* Clear the required eflag bits and jump to the interrupt/exception handler. */
8931	if (rcStrict == VINF_SUCCESS)
8932	{
8933	pCtx->eflags.u32 &= ~(X86_EFL_IF \| X86_EFL_TF \| X86_EFL_RF \| X86_EFL_AC);
8934	pCtx->rip = IdtEntry.offSel;
8935	pCtx->cs.Sel = IdtEntry.uSel;
8936	pCtx->cs.ValidSel = IdtEntry.uSel;
8937	pCtx->cs.u64Base = IdtEntry.uSel << cbIdtEntry;
8938	if ( uIntType == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT
8939	&& uVector == X86_XCPT_PF)
8940	pCtx->cr2 = GCPtrFault;
8941
8942	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CS \| HM_CHANGED_GUEST_CR2
8943	\| HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
8944	\| HM_CHANGED_GUEST_RSP);
8945
8946	/*
8947	* If we delivered a hardware exception (other than an NMI) and if there was
8948	* block-by-STI in effect, we should clear it.
8949	*/
8950	if (*pfIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
8951	{
8952	Assert( uIntType != VMX_ENTRY_INT_INFO_TYPE_NMI
8953	&& uIntType != VMX_ENTRY_INT_INFO_TYPE_EXT_INT);
8954	Log4Func(("Clearing inhibition due to STI\n"));
8955	*pfIntrState &= ~VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
8956	}
8957
8958	Log4(("Injected real-mode: u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x Eflags=%#x CS:EIP=%04x:%04x\n",
8959	u32IntInfo, u32ErrCode, cbInstr, pCtx->eflags.u, pCtx->cs.Sel, pCtx->eip));
8960
8961	/*
8962	* The event has been truly dispatched to the guest. Mark it as no longer pending so
8963	* we don't attempt to undo it if we are returning to ring-3 before executing guest code.
8964	*/
8965	pVCpu->hm.s.Event.fPending = false;
8966
8967	/*
8968	* If we eventually support nested-guest execution without unrestricted guest execution,
8969	* we should set fInterceptEvents here.
8970	*/
8971	Assert(!pVmxTransient->fIsNestedGuest);
8972
8973	/* If we're stepping and we've changed cs:rip above, bail out of the VMX R0 execution loop. */
8974	if (fStepping)
8975	rcStrict = VINF_EM_DBG_STEPPED;
8976	}
8977	AssertMsg(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
8978	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
8979	return rcStrict;
8980	}
8981	}
8982
8983	/*
8984	* Validate.
8985	*/
8986	Assert(VMX_ENTRY_INT_INFO_IS_VALID(u32IntInfo)); /* Bit 31 (Valid bit) must be set by caller. */
8987	Assert(!(u32IntInfo & VMX_BF_ENTRY_INT_INFO_RSVD_12_30_MASK)); /* Bits 30:12 MBZ. */
8988
8989	/*
8990	* Inject the event into the VMCS.
8991	*/
8992	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, u32IntInfo);
8993	if (VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(u32IntInfo))
8994	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, u32ErrCode);
8995	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, cbInstr);
8996	AssertRC(rc);
8997
8998	/*
8999	* Update guest CR2 if this is a page-fault.
9000	*/
9001	if (VMX_ENTRY_INT_INFO_IS_XCPT_PF(u32IntInfo))
9002	pCtx->cr2 = GCPtrFault;
9003
9004	Log4(("Injecting u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x CR2=%#RX64\n", u32IntInfo, u32ErrCode, cbInstr, pCtx->cr2));
9005	return VINF_SUCCESS;
9006	}
9007
9008
9009	/**
9010	* Evaluates the event to be delivered to the guest and sets it as the pending
9011	* event.
9012	*
9013	* Toggling of interrupt force-flags here is safe since we update TRPM on premature
9014	* exits to ring-3 before executing guest code, see hmR0VmxExitToRing3(). We must
9015	* NOT restore these force-flags.
9016	*
9017	* @returns Strict VBox status code (i.e. informational status codes too).
9018	* @param pVCpu The cross context virtual CPU structure.
9019	* @param pVmxTransient The VMX-transient structure.
9020	* @param pfIntrState Where to store the VT-x guest-interruptibility state.
9021	*/
9022	static VBOXSTRICTRC hmR0VmxEvaluatePendingEvent(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t *pfIntrState)
9023	{
9024	Assert(pfIntrState);
9025	Assert(!TRPMHasTrap(pVCpu));
9026
9027	/*
9028	* Compute/update guest-interruptibility state related FFs.
9029	* The FFs will be used below while evaluating events to be injected.
9030	*/
9031	*pfIntrState = hmR0VmxGetGuestIntrStateAndUpdateFFs(pVCpu);
9032
9033	/*
9034	* Evaluate if a new event needs to be injected.
9035	* An event that's already pending has already performed all necessary checks.
9036	*/
9037	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
9038	bool const fIsNestedGuest = pVmxTransient->fIsNestedGuest;
9039	if ( !pVCpu->hm.s.Event.fPending
9040	&& !VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
9041	{
9042	/** @todo SMI. SMIs take priority over NMIs. */
9043
9044	/*
9045	* NMIs.
9046	* NMIs take priority over external interrupts.
9047	*/
9048	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
9049	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
9050	{
9051	/*
9052	* For a guest, the FF always indicates the guest's ability to receive an NMI.
9053	*
9054	* For a nested-guest, the FF always indicates the outer guest's ability to
9055	* receive an NMI while the guest-interruptibility state bit depends on whether
9056	* the nested-hypervisor is using virtual-NMIs.
9057	*/
9058	if (!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_BLOCK_NMIS))
9059	{
9060	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9061	if ( fIsNestedGuest
9062	&& CPUMIsGuestVmxPinCtlsSet(pCtx, VMX_PIN_CTLS_NMI_EXIT))
9063	return IEMExecVmxVmexitXcptNmi(pVCpu);
9064	#endif
9065	hmR0VmxSetPendingXcptNmi(pVCpu);
9066	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
9067	Log4Func(("NMI pending injection\n"));
9068
9069	/* We've injected the NMI, bail. */
9070	return VINF_SUCCESS;
9071	}
9072	else if (!fIsNestedGuest)
9073	hmR0VmxSetNmiWindowExitVmcs(pVmcsInfo);
9074	}
9075
9076	/*
9077	* External interrupts (PIC/APIC).
9078	* Once PDMGetInterrupt() returns a valid interrupt we -must- deliver it.
9079	* We cannot re-request the interrupt from the controller again.
9080	*/
9081	if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
9082	&& !pVCpu->hm.s.fSingleInstruction)
9083	{
9084	Assert(!DBGFIsStepping(pVCpu));
9085	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_RFLAGS);
9086	AssertRC(rc);
9087
9088	/*
9089	* We must not check EFLAGS directly when executing a nested-guest, use
9090	* CPUMIsGuestPhysIntrEnabled() instead as EFLAGS.IF does not control the blocking of
9091	* external interrupts when "External interrupt exiting" is set. This fixes a nasty
9092	* SMP hang while executing nested-guest VCPUs on spinlocks which aren't rescued by
9093	* other VM-exits (like a preemption timer), see @bugref{9562#c18}.
9094	*
9095	* See Intel spec. 25.4.1 "Event Blocking".
9096	*/
9097	if (CPUMIsGuestPhysIntrEnabled(pVCpu))
9098	{
9099	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9100	if ( fIsNestedGuest
9101	&& CPUMIsGuestVmxPinCtlsSet(pCtx, VMX_PIN_CTLS_EXT_INT_EXIT))
9102	{
9103	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitExtInt(pVCpu, 0 /* uVector /, true / fIntPending */);
9104	if (rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE)
9105	return rcStrict;
9106	}
9107	#endif
9108	uint8_t u8Interrupt;
9109	rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
9110	if (RT_SUCCESS(rc))
9111	{
9112	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9113	if ( fIsNestedGuest
9114	&& CPUMIsGuestVmxPinCtlsSet(pCtx, VMX_PIN_CTLS_EXT_INT_EXIT))
9115	{
9116	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitExtInt(pVCpu, u8Interrupt, false /* fIntPending */);
9117	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
9118	return rcStrict;
9119	}
9120	#endif
9121	hmR0VmxSetPendingExtInt(pVCpu, u8Interrupt);
9122	Log4Func(("External interrupt (%#x) pending injection\n", u8Interrupt));
9123	}
9124	else if (rc == VERR_APIC_INTR_MASKED_BY_TPR)
9125	{
9126	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchTprMaskedIrq);
9127
9128	if ( !fIsNestedGuest
9129	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW))
9130	hmR0VmxApicSetTprThreshold(pVmcsInfo, u8Interrupt >> 4);
9131	/* else: for nested-guests, TPR threshold is picked up while merging VMCS controls. */
9132
9133	/*
9134	* If the CPU doesn't have TPR shadowing, we will always get a VM-exit on TPR changes and
9135	* APICSetTpr() will end up setting the VMCPU_FF_INTERRUPT_APIC if required, so there is no
9136	* need to re-set this force-flag here.
9137	*/
9138	}
9139	else
9140	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
9141
9142	/* We've injected the interrupt or taken necessary action, bail. */
9143	return VINF_SUCCESS;
9144	}
9145	if (!fIsNestedGuest)
9146	hmR0VmxSetIntWindowExitVmcs(pVmcsInfo);
9147	}
9148	}
9149	else if (!fIsNestedGuest)
9150	{
9151	/*
9152	* An event is being injected or we are in an interrupt shadow. Check if another event is
9153	* pending. If so, instruct VT-x to cause a VM-exit as soon as the guest is ready to accept
9154	* the pending event.
9155	*/
9156	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
9157	hmR0VmxSetNmiWindowExitVmcs(pVmcsInfo);
9158	else if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
9159	&& !pVCpu->hm.s.fSingleInstruction)
9160	hmR0VmxSetIntWindowExitVmcs(pVmcsInfo);
9161	}
9162	/* else: for nested-guests, NMI/interrupt-window exiting will be picked up when merging VMCS controls. */
9163
9164	return VINF_SUCCESS;
9165	}
9166
9167
9168	/**
9169	* Injects any pending events into the guest if the guest is in a state to
9170	* receive them.
9171	*
9172	* @returns Strict VBox status code (i.e. informational status codes too).
9173	* @param pVCpu The cross context virtual CPU structure.
9174	* @param pVmxTransient The VMX-transient structure.
9175	* @param fIntrState The VT-x guest-interruptibility state.
9176	* @param fStepping Whether we are single-stepping the guest using the
9177	* hypervisor debugger and should return
9178	* VINF_EM_DBG_STEPPED if the event was dispatched
9179	* directly.
9180	*/
9181	static VBOXSTRICTRC hmR0VmxInjectPendingEvent(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t fIntrState, bool fStepping)
9182	{
9183	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9184	Assert(VMMRZCallRing3IsEnabled(pVCpu));
9185
9186	#ifdef VBOX_STRICT
9187	/*
9188	* Verify guest-interruptibility state.
9189	*
9190	* We put this in a scoped block so we do not accidentally use fBlockSti or fBlockMovSS,
9191	* since injecting an event may modify the interruptibility state and we must thus always
9192	* use fIntrState.
9193	*/
9194	{
9195	bool const fBlockMovSS = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
9196	bool const fBlockSti = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI);
9197	Assert(!fBlockSti \|\| !(ASMAtomicUoReadU64(&pVCpu->cpum.GstCtx.fExtrn) & CPUMCTX_EXTRN_RFLAGS));
9198	Assert(!fBlockSti \|\| pVCpu->cpum.GstCtx.eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
9199	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
9200	Assert(!TRPMHasTrap(pVCpu));
9201	NOREF(fBlockMovSS); NOREF(fBlockSti);
9202	}
9203	#endif
9204
9205	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
9206	if (pVCpu->hm.s.Event.fPending)
9207	{
9208	/*
9209	* Do -not- clear any interrupt-window exiting control here. We might have an interrupt
9210	* pending even while injecting an event and in this case, we want a VM-exit as soon as
9211	* the guest is ready for the next interrupt, see @bugref{6208#c45}.
9212	*
9213	* See Intel spec. 26.6.5 "Interrupt-Window Exiting and Virtual-Interrupt Delivery".
9214	*/
9215	uint32_t const uIntType = VMX_ENTRY_INT_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
9216	#ifdef VBOX_STRICT
9217	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
9218	{
9219	Assert(pVCpu->cpum.GstCtx.eflags.u32 & X86_EFL_IF);
9220	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI));
9221	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
9222	}
9223	else if (uIntType == VMX_ENTRY_INT_INFO_TYPE_NMI)
9224	{
9225	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI));
9226	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI));
9227	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
9228	}
9229	#endif
9230	Log4(("Injecting pending event vcpu[%RU32] u64IntInfo=%#RX64 Type=%#RX32\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo,
9231	uIntType));
9232
9233	/*
9234	* Inject the event and get any changes to the guest-interruptibility state.
9235	*
9236	* The guest-interruptibility state may need to be updated if we inject the event
9237	* into the guest IDT ourselves (for real-on-v86 guest injecting software interrupts).
9238	*/
9239	rcStrict = hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &pVCpu->hm.s.Event, fStepping, &fIntrState);
9240	AssertRCReturn(VBOXSTRICTRC_VAL(rcStrict), rcStrict);
9241
9242	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
9243	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt);
9244	else
9245	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt);
9246	}
9247
9248	/*
9249	* Deliver any pending debug exceptions if the guest is single-stepping using EFLAGS.TF and
9250	* is an interrupt shadow (block-by-STI or block-by-MOV SS).
9251	*/
9252	if ( (fIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
9253	&& !pVmxTransient->fIsNestedGuest)
9254	{
9255	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
9256
9257	if (!pVCpu->hm.s.fSingleInstruction)
9258	{
9259	/*
9260	* Set or clear the BS bit depending on whether the trap flag is active or not. We need
9261	* to do both since we clear the BS bit from the VMCS while exiting to ring-3.
9262	*/
9263	Assert(!DBGFIsStepping(pVCpu));
9264	uint8_t const fTrapFlag = !!(pVCpu->cpum.GstCtx.eflags.u32 & X86_EFL_TF);
9265	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, fTrapFlag << VMX_BF_VMCS_PENDING_DBG_XCPT_BS_SHIFT);
9266	AssertRC(rc);
9267	}
9268	else
9269	{
9270	/*
9271	* We must not deliver a debug exception when single-stepping over STI/Mov-SS in the
9272	* hypervisor debugger using EFLAGS.TF but rather clear interrupt inhibition. However,
9273	* we take care of this case in hmR0VmxExportSharedDebugState and also the case if
9274	* we use MTF, so just make sure it's called before executing guest-code.
9275	*/
9276	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR_MASK);
9277	}
9278	}
9279	/* else: for nested-guest currently handling while merging controls. */
9280
9281	/*
9282	* Finally, update the guest-interruptibility state.
9283	*
9284	* This is required for the real-on-v86 software interrupt injection, for
9285	* pending debug exceptions as well as updates to the guest state from ring-3 (IEM).
9286	*/
9287	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
9288	AssertRC(rc);
9289
9290	/*
9291	* There's no need to clear the VM-entry interruption-information field here if we're not
9292	* injecting anything. VT-x clears the valid bit on every VM-exit.
9293	*
9294	* See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
9295	*/
9296
9297	Assert(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping));
9298	return rcStrict;
9299	}
9300
9301
9302	/**
9303	* Enters the VT-x session.
9304	*
9305	* @returns VBox status code.
9306	* @param pVCpu The cross context virtual CPU structure.
9307	*/
9308	VMMR0DECL(int) VMXR0Enter(PVMCPUCC pVCpu)
9309	{
9310	AssertPtr(pVCpu);
9311	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fSupported);
9312	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9313
9314	LogFlowFunc(("pVCpu=%p\n", pVCpu));
9315	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
9316	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
9317
9318	#ifdef VBOX_STRICT
9319	/* At least verify VMX is enabled, since we can't check if we're in VMX root mode without #GP'ing. */
9320	RTCCUINTREG uHostCr4 = ASMGetCR4();
9321	if (!(uHostCr4 & X86_CR4_VMXE))
9322	{
9323	LogRelFunc(("X86_CR4_VMXE bit in CR4 is not set!\n"));
9324	return VERR_VMX_X86_CR4_VMXE_CLEARED;
9325	}
9326	#endif
9327
9328	/*
9329	* Do the EMT scheduled L1D and MDS flush here if needed.
9330	*/
9331	if (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_L1D_SCHED)
9332	ASMWrMsr(MSR_IA32_FLUSH_CMD, MSR_IA32_FLUSH_CMD_F_L1D);
9333	else if (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_MDS_SCHED)
9334	hmR0MdsClear();
9335
9336	/*
9337	* Load the appropriate VMCS as the current and active one.
9338	*/
9339	PVMXVMCSINFO pVmcsInfo;
9340	bool const fInNestedGuestMode = CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx);
9341	if (!fInNestedGuestMode)
9342	pVmcsInfo = &pVCpu->hmr0.s.vmx.VmcsInfo;
9343	else
9344	pVmcsInfo = &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
9345	int rc = hmR0VmxLoadVmcs(pVmcsInfo);
9346	if (RT_SUCCESS(rc))
9347	{
9348	pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs = fInNestedGuestMode;
9349	pVCpu->hm.s.vmx.fSwitchedToNstGstVmcsCopyForRing3 = fInNestedGuestMode;
9350	pVCpu->hmr0.s.fLeaveDone = false;
9351	Log4Func(("Loaded Vmcs. HostCpuId=%u\n", RTMpCpuId()));
9352	}
9353	return rc;
9354	}
9355
9356
9357	/**
9358	* The thread-context callback.
9359	*
9360	* This is used together with RTThreadCtxHookCreate() on platforms which
9361	* supports it, and directly from VMMR0EmtPrepareForBlocking() and
9362	* VMMR0EmtResumeAfterBlocking() on platforms which don't.
9363	*
9364	* @param enmEvent The thread-context event.
9365	* @param pVCpu The cross context virtual CPU structure.
9366	* @param fGlobalInit Whether global VT-x/AMD-V init. was used.
9367	* @thread EMT(pVCpu)
9368	*/
9369	VMMR0DECL(void) VMXR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPUCC pVCpu, bool fGlobalInit)
9370	{
9371	AssertPtr(pVCpu);
9372	RT_NOREF1(fGlobalInit);
9373
9374	switch (enmEvent)
9375	{
9376	case RTTHREADCTXEVENT_OUT:
9377	{
9378	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9379	VMCPU_ASSERT_EMT(pVCpu);
9380
9381	/* No longjmps (logger flushes, locks) in this fragile context. */
9382	VMMRZCallRing3Disable(pVCpu);
9383	Log4Func(("Preempting: HostCpuId=%u\n", RTMpCpuId()));
9384
9385	/* Restore host-state (FPU, debug etc.) */
9386	if (!pVCpu->hmr0.s.fLeaveDone)
9387	{
9388	/*
9389	* Do -not- import the guest-state here as we might already be in the middle of importing
9390	* it, esp. bad if we're holding the PGM lock, see comment in hmR0VmxImportGuestState().
9391	*/
9392	hmR0VmxLeave(pVCpu, false /* fImportState */);
9393	pVCpu->hmr0.s.fLeaveDone = true;
9394	}
9395
9396	/* Leave HM context, takes care of local init (term). */
9397	int rc = HMR0LeaveCpu(pVCpu);
9398	AssertRC(rc);
9399
9400	/* Restore longjmp state. */
9401	VMMRZCallRing3Enable(pVCpu);
9402	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreempt);
9403	break;
9404	}
9405
9406	case RTTHREADCTXEVENT_IN:
9407	{
9408	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9409	VMCPU_ASSERT_EMT(pVCpu);
9410
9411	/* Do the EMT scheduled L1D and MDS flush here if needed. */
9412	if (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_L1D_SCHED)
9413	ASMWrMsr(MSR_IA32_FLUSH_CMD, MSR_IA32_FLUSH_CMD_F_L1D);
9414	else if (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_MDS_SCHED)
9415	hmR0MdsClear();
9416
9417	/* No longjmps here, as we don't want to trigger preemption (& its hook) while resuming. */
9418	VMMRZCallRing3Disable(pVCpu);
9419	Log4Func(("Resumed: HostCpuId=%u\n", RTMpCpuId()));
9420
9421	/* Initialize the bare minimum state required for HM. This takes care of
9422	initializing VT-x if necessary (onlined CPUs, local init etc.) */
9423	int rc = hmR0EnterCpu(pVCpu);
9424	AssertRC(rc);
9425	Assert( (pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
9426	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
9427
9428	/* Load the active VMCS as the current one. */
9429	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
9430	rc = hmR0VmxLoadVmcs(pVmcsInfo);
9431	AssertRC(rc);
9432	Log4Func(("Resumed: Loaded Vmcs. HostCpuId=%u\n", RTMpCpuId()));
9433	pVCpu->hmr0.s.fLeaveDone = false;
9434
9435	/* Restore longjmp state. */
9436	VMMRZCallRing3Enable(pVCpu);
9437	break;
9438	}
9439
9440	default:
9441	break;
9442	}
9443	}
9444
9445
9446	/**
9447	* Exports the host state into the VMCS host-state area.
9448	* Sets up the VM-exit MSR-load area.
9449	*
9450	* The CPU state will be loaded from these fields on every successful VM-exit.
9451	*
9452	* @returns VBox status code.
9453	* @param pVCpu The cross context virtual CPU structure.
9454	*
9455	* @remarks No-long-jump zone!!!
9456	*/
9457	static int hmR0VmxExportHostState(PVMCPUCC pVCpu)
9458	{
9459	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9460
9461	int rc = VINF_SUCCESS;
9462	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT)
9463	{
9464	uint64_t uHostCr4 = hmR0VmxExportHostControlRegs();
9465
9466	rc = hmR0VmxExportHostSegmentRegs(pVCpu, uHostCr4);
9467	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9468
9469	hmR0VmxExportHostMsrs(pVCpu);
9470
9471	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_HOST_CONTEXT;
9472	}
9473	return rc;
9474	}
9475
9476
9477	/**
9478	* Saves the host state in the VMCS host-state.
9479	*
9480	* @returns VBox status code.
9481	* @param pVCpu The cross context virtual CPU structure.
9482	*
9483	* @remarks No-long-jump zone!!!
9484	*/
9485	VMMR0DECL(int) VMXR0ExportHostState(PVMCPUCC pVCpu)
9486	{
9487	AssertPtr(pVCpu);
9488	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9489
9490	/*
9491	* Export the host state here while entering HM context.
9492	* When thread-context hooks are used, we might get preempted and have to re-save the host
9493	* state but most of the time we won't be, so do it here before we disable interrupts.
9494	*/
9495	return hmR0VmxExportHostState(pVCpu);
9496	}
9497
9498
9499	/**
9500	* Exports the guest state into the VMCS guest-state area.
9501	*
9502	* The will typically be done before VM-entry when the guest-CPU state and the
9503	* VMCS state may potentially be out of sync.
9504	*
9505	* Sets up the VM-entry MSR-load and VM-exit MSR-store areas. Sets up the
9506	* VM-entry controls.
9507	* Sets up the appropriate VMX non-root function to execute guest code based on
9508	* the guest CPU mode.
9509	*
9510	* @returns VBox strict status code.
9511	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
9512	* without unrestricted guest execution and the VMMDev is not presently
9513	* mapped (e.g. EFI32).
9514	*
9515	* @param pVCpu The cross context virtual CPU structure.
9516	* @param pVmxTransient The VMX-transient structure.
9517	*
9518	* @remarks No-long-jump zone!!!
9519	*/
9520	static VBOXSTRICTRC hmR0VmxExportGuestState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9521	{
9522	AssertPtr(pVCpu);
9523	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9524	LogFlowFunc(("pVCpu=%p\n", pVCpu));
9525
9526	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExportGuestState, x);
9527
9528	/*
9529	* Determine real-on-v86 mode.
9530	* Used when the guest is in real-mode and unrestricted guest execution is not used.
9531	*/
9532	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmxTransient->pVmcsInfo->pShared;
9533	if ( pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUnrestrictedGuest
9534	\|\| !CPUMIsGuestInRealModeEx(&pVCpu->cpum.GstCtx))
9535	pVmcsInfoShared->RealMode.fRealOnV86Active = false;
9536	else
9537	{
9538	Assert(!pVmxTransient->fIsNestedGuest);
9539	pVmcsInfoShared->RealMode.fRealOnV86Active = true;
9540	}
9541
9542	/*
9543	* Any ordering dependency among the sub-functions below must be explicitly stated using comments.
9544	* Ideally, assert that the cross-dependent bits are up-to-date at the point of using it.
9545	*/
9546	int rc = hmR0VmxExportGuestEntryExitCtls(pVCpu, pVmxTransient);
9547	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9548
9549	rc = hmR0VmxExportGuestCR0(pVCpu, pVmxTransient);
9550	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9551
9552	VBOXSTRICTRC rcStrict = hmR0VmxExportGuestCR3AndCR4(pVCpu, pVmxTransient);
9553	if (rcStrict == VINF_SUCCESS)
9554	{ /* likely */ }
9555	else
9556	{
9557	Assert(rcStrict == VINF_EM_RESCHEDULE_REM \|\| RT_FAILURE_NP(rcStrict));
9558	return rcStrict;
9559	}
9560
9561	rc = hmR0VmxExportGuestSegRegsXdtr(pVCpu, pVmxTransient);
9562	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9563
9564	rc = hmR0VmxExportGuestMsrs(pVCpu, pVmxTransient);
9565	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9566
9567	hmR0VmxExportGuestApicTpr(pVCpu, pVmxTransient);
9568	hmR0VmxExportGuestXcptIntercepts(pVCpu, pVmxTransient);
9569	hmR0VmxExportGuestRip(pVCpu);
9570	hmR0VmxExportGuestRsp(pVCpu);
9571	hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9572
9573	rc = hmR0VmxExportGuestHwvirtState(pVCpu, pVmxTransient);
9574	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9575
9576	/* Clear any bits that may be set but exported unconditionally or unused/reserved bits. */
9577	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~( (HM_CHANGED_GUEST_GPRS_MASK & ~HM_CHANGED_GUEST_RSP)
9578	\| HM_CHANGED_GUEST_CR2
9579	\| (HM_CHANGED_GUEST_DR_MASK & ~HM_CHANGED_GUEST_DR7)
9580	\| HM_CHANGED_GUEST_X87
9581	\| HM_CHANGED_GUEST_SSE_AVX
9582	\| HM_CHANGED_GUEST_OTHER_XSAVE
9583	\| HM_CHANGED_GUEST_XCRx
9584	\| HM_CHANGED_GUEST_KERNEL_GS_BASE /* Part of lazy or auto load-store MSRs. */
9585	\| HM_CHANGED_GUEST_SYSCALL_MSRS /* Part of lazy or auto load-store MSRs. */
9586	\| HM_CHANGED_GUEST_TSC_AUX
9587	\| HM_CHANGED_GUEST_OTHER_MSRS
9588	\| (HM_CHANGED_KEEPER_STATE_MASK & ~HM_CHANGED_VMX_MASK)));
9589
9590	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExportGuestState, x);
9591	return rc;
9592	}
9593
9594
9595	/**
9596	* Exports the state shared between the host and guest into the VMCS.
9597	*
9598	* @param pVCpu The cross context virtual CPU structure.
9599	* @param pVmxTransient The VMX-transient structure.
9600	*
9601	* @remarks No-long-jump zone!!!
9602	*/
9603	static void hmR0VmxExportSharedState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9604	{
9605	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9606	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9607
9608	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_DR_MASK)
9609	{
9610	int rc = hmR0VmxExportSharedDebugState(pVCpu, pVmxTransient);
9611	AssertRC(rc);
9612	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_GUEST_DR_MASK;
9613
9614	/* Loading shared debug bits might have changed eflags.TF bit for debugging purposes. */
9615	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_RFLAGS)
9616	hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9617	}
9618
9619	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_GUEST_LAZY_MSRS)
9620	{
9621	hmR0VmxLazyLoadGuestMsrs(pVCpu);
9622	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_VMX_GUEST_LAZY_MSRS;
9623	}
9624
9625	AssertMsg(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE),
9626	("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
9627	}
9628
9629
9630	/**
9631	* Worker for loading the guest-state bits in the inner VT-x execution loop.
9632	*
9633	* @returns Strict VBox status code (i.e. informational status codes too).
9634	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
9635	* without unrestricted guest execution and the VMMDev is not presently
9636	* mapped (e.g. EFI32).
9637	*
9638	* @param pVCpu The cross context virtual CPU structure.
9639	* @param pVmxTransient The VMX-transient structure.
9640	*
9641	* @remarks No-long-jump zone!!!
9642	*/
9643	static VBOXSTRICTRC hmR0VmxExportGuestStateOptimal(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9644	{
9645	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9646	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9647
9648	#ifdef HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
9649	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
9650	#endif
9651
9652	/*
9653	* For many VM-exits only RIP/RSP/RFLAGS (and HWVIRT state when executing a nested-guest)
9654	* changes. First try to export only these without going through all other changed-flag checks.
9655	*/
9656	VBOXSTRICTRC rcStrict;
9657	uint64_t const fCtxMask = HM_CHANGED_ALL_GUEST & ~HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE;
9658	uint64_t const fMinimalMask = HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT;
9659	uint64_t const fCtxChanged = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
9660
9661	/* If only RIP/RSP/RFLAGS/HWVIRT changed, export only those (quicker, happens more often).*/
9662	if ( (fCtxChanged & fMinimalMask)
9663	&& !(fCtxChanged & (fCtxMask & ~fMinimalMask)))
9664	{
9665	hmR0VmxExportGuestRip(pVCpu);
9666	hmR0VmxExportGuestRsp(pVCpu);
9667	hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9668	rcStrict = hmR0VmxExportGuestHwvirtState(pVCpu, pVmxTransient);
9669	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportMinimal);
9670	}
9671	/* If anything else also changed, go through the full export routine and export as required. */
9672	else if (fCtxChanged & fCtxMask)
9673	{
9674	rcStrict = hmR0VmxExportGuestState(pVCpu, pVmxTransient);
9675	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
9676	{ /* likely */}
9677	else
9678	{
9679	AssertMsg(rcStrict == VINF_EM_RESCHEDULE_REM, ("Failed to export guest state! rc=%Rrc\n",
9680	VBOXSTRICTRC_VAL(rcStrict)));
9681	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9682	return rcStrict;
9683	}
9684	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportFull);
9685	}
9686	/* Nothing changed, nothing to load here. */
9687	else
9688	rcStrict = VINF_SUCCESS;
9689
9690	#ifdef VBOX_STRICT
9691	/* All the guest state bits should be loaded except maybe the host context and/or the shared host/guest bits. */
9692	uint64_t const fCtxChangedCur = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
9693	AssertMsg(!(fCtxChangedCur & fCtxMask), ("fCtxChangedCur=%#RX64\n", fCtxChangedCur));
9694	#endif
9695	return rcStrict;
9696	}
9697
9698
9699	/**
9700	* Tries to determine what part of the guest-state VT-x has deemed as invalid
9701	* and update error record fields accordingly.
9702	*
9703	* @returns VMX_IGS_* error codes.
9704	* @retval VMX_IGS_REASON_NOT_FOUND if this function could not find anything
9705	* wrong with the guest state.
9706	*
9707	* @param pVCpu The cross context virtual CPU structure.
9708	* @param pVmcsInfo The VMCS info. object.
9709	*
9710	* @remarks This function assumes our cache of the VMCS controls
9711	* are valid, i.e. hmR0VmxCheckCachedVmcsCtls() succeeded.
9712	*/
9713	static uint32_t hmR0VmxCheckGuestState(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
9714	{
9715	#define HMVMX_ERROR_BREAK(err) { uError = (err); break; }
9716	#define HMVMX_CHECK_BREAK(expr, err) do { \
9717	if (!(expr)) { uError = (err); break; } \
9718	} while (0)
9719
9720	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
9721	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
9722	uint32_t uError = VMX_IGS_ERROR;
9723	uint32_t u32IntrState = 0;
9724	bool const fUnrestrictedGuest = pVM->hmr0.s.vmx.fUnrestrictedGuest;
9725	do
9726	{
9727	int rc;
9728
9729	/*
9730	* Guest-interruptibility state.
9731	*
9732	* Read this first so that any check that fails prior to those that actually
9733	* require the guest-interruptibility state would still reflect the correct
9734	* VMCS value and avoids causing further confusion.
9735	*/
9736	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &u32IntrState);
9737	AssertRC(rc);
9738
9739	uint32_t u32Val;
9740	uint64_t u64Val;
9741
9742	/*
9743	* CR0.
9744	*/
9745	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
9746	uint64_t fSetCr0 = (g_HmMsrs.u.vmx.u64Cr0Fixed0 & g_HmMsrs.u.vmx.u64Cr0Fixed1);
9747	uint64_t const fZapCr0 = (g_HmMsrs.u.vmx.u64Cr0Fixed0 \| g_HmMsrs.u.vmx.u64Cr0Fixed1);
9748	/* Exceptions for unrestricted guest execution for CR0 fixed bits (PE, PG).
9749	See Intel spec. 26.3.1 "Checks on Guest Control Registers, Debug Registers and MSRs." */
9750	if (fUnrestrictedGuest)
9751	fSetCr0 &= ~(uint64_t)(X86_CR0_PE \| X86_CR0_PG);
9752
9753	uint64_t u64GuestCr0;
9754	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64GuestCr0);
9755	AssertRC(rc);
9756	HMVMX_CHECK_BREAK((u64GuestCr0 & fSetCr0) == fSetCr0, VMX_IGS_CR0_FIXED1);
9757	HMVMX_CHECK_BREAK(!(u64GuestCr0 & ~fZapCr0), VMX_IGS_CR0_FIXED0);
9758	if ( !fUnrestrictedGuest
9759	&& (u64GuestCr0 & X86_CR0_PG)
9760	&& !(u64GuestCr0 & X86_CR0_PE))
9761	HMVMX_ERROR_BREAK(VMX_IGS_CR0_PG_PE_COMBO);
9762
9763	/*
9764	* CR4.
9765	*/
9766	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
9767	uint64_t const fSetCr4 = (g_HmMsrs.u.vmx.u64Cr4Fixed0 & g_HmMsrs.u.vmx.u64Cr4Fixed1);
9768	uint64_t const fZapCr4 = (g_HmMsrs.u.vmx.u64Cr4Fixed0 \| g_HmMsrs.u.vmx.u64Cr4Fixed1);
9769
9770	uint64_t u64GuestCr4;
9771	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR4, &u64GuestCr4);
9772	AssertRC(rc);
9773	HMVMX_CHECK_BREAK((u64GuestCr4 & fSetCr4) == fSetCr4, VMX_IGS_CR4_FIXED1);
9774	HMVMX_CHECK_BREAK(!(u64GuestCr4 & ~fZapCr4), VMX_IGS_CR4_FIXED0);
9775
9776	/*
9777	* IA32_DEBUGCTL MSR.
9778	*/
9779	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, &u64Val);
9780	AssertRC(rc);
9781	if ( (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
9782	&& (u64Val & 0xfffffe3c)) /* Bits 31:9, bits 5:2 MBZ. */
9783	{
9784	HMVMX_ERROR_BREAK(VMX_IGS_DEBUGCTL_MSR_RESERVED);
9785	}
9786	uint64_t u64DebugCtlMsr = u64Val;
9787
9788	#ifdef VBOX_STRICT
9789	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
9790	AssertRC(rc);
9791	Assert(u32Val == pVmcsInfo->u32EntryCtls);
9792	#endif
9793	bool const fLongModeGuest = RT_BOOL(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
9794
9795	/*
9796	* RIP and RFLAGS.
9797	*/
9798	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RIP, &u64Val);
9799	AssertRC(rc);
9800	/* pCtx->rip can be different than the one in the VMCS (e.g. run guest code and VM-exits that don't update it). */
9801	if ( !fLongModeGuest
9802	\|\| !pCtx->cs.Attr.n.u1Long)
9803	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffff00000000)), VMX_IGS_LONGMODE_RIP_INVALID);
9804	/** @todo If the processor supports N < 64 linear-address bits, bits 63:N
9805	* must be identical if the "IA-32e mode guest" VM-entry
9806	* control is 1 and CS.L is 1. No check applies if the
9807	* CPU supports 64 linear-address bits. */
9808
9809	/* Flags in pCtx can be different (real-on-v86 for instance). We are only concerned about the VMCS contents here. */
9810	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RFLAGS, &u64Val);
9811	AssertRC(rc);
9812	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffc08028)), /* Bit 63:22, Bit 15, 5, 3 MBZ. */
9813	VMX_IGS_RFLAGS_RESERVED);
9814	HMVMX_CHECK_BREAK((u64Val & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
9815	uint32_t const u32Eflags = u64Val;
9816
9817	if ( fLongModeGuest
9818	\|\| ( fUnrestrictedGuest
9819	&& !(u64GuestCr0 & X86_CR0_PE)))
9820	{
9821	HMVMX_CHECK_BREAK(!(u32Eflags & X86_EFL_VM), VMX_IGS_RFLAGS_VM_INVALID);
9822	}
9823
9824	uint32_t u32EntryInfo;
9825	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo);
9826	AssertRC(rc);
9827	if (VMX_ENTRY_INT_INFO_IS_EXT_INT(u32EntryInfo))
9828	HMVMX_CHECK_BREAK(u32Eflags & X86_EFL_IF, VMX_IGS_RFLAGS_IF_INVALID);
9829
9830	/*
9831	* 64-bit checks.
9832	*/
9833	if (fLongModeGuest)
9834	{
9835	HMVMX_CHECK_BREAK(u64GuestCr0 & X86_CR0_PG, VMX_IGS_CR0_PG_LONGMODE);
9836	HMVMX_CHECK_BREAK(u64GuestCr4 & X86_CR4_PAE, VMX_IGS_CR4_PAE_LONGMODE);
9837	}
9838
9839	if ( !fLongModeGuest
9840	&& (u64GuestCr4 & X86_CR4_PCIDE))
9841	HMVMX_ERROR_BREAK(VMX_IGS_CR4_PCIDE);
9842
9843	/** @todo CR3 field must be such that bits 63:52 and bits in the range
9844	* 51:32 beyond the processor's physical-address width are 0. */
9845
9846	if ( (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
9847	&& (pCtx->dr[7] & X86_DR7_MBZ_MASK))
9848	HMVMX_ERROR_BREAK(VMX_IGS_DR7_RESERVED);
9849
9850	rc = VMXReadVmcsNw(VMX_VMCS_HOST_SYSENTER_ESP, &u64Val);
9851	AssertRC(rc);
9852	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_ESP_NOT_CANONICAL);
9853
9854	rc = VMXReadVmcsNw(VMX_VMCS_HOST_SYSENTER_EIP, &u64Val);
9855	AssertRC(rc);
9856	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_EIP_NOT_CANONICAL);
9857
9858	/*
9859	* PERF_GLOBAL MSR.
9860	*/
9861	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PERF_MSR)
9862	{
9863	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL, &u64Val);
9864	AssertRC(rc);
9865	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffff8fffffffc)),
9866	VMX_IGS_PERF_GLOBAL_MSR_RESERVED); /* Bits 63:35, bits 31:2 MBZ. */
9867	}
9868
9869	/*
9870	* PAT MSR.
9871	*/
9872	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PAT_MSR)
9873	{
9874	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PAT_FULL, &u64Val);
9875	AssertRC(rc);
9876	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0x707070707070707)), VMX_IGS_PAT_MSR_RESERVED);
9877	for (unsigned i = 0; i < 8; i++)
9878	{
9879	uint8_t u8Val = (u64Val & 0xff);
9880	if ( u8Val != 0 /* UC */
9881	&& u8Val != 1 /* WC */
9882	&& u8Val != 4 /* WT */
9883	&& u8Val != 5 /* WP */
9884	&& u8Val != 6 /* WB */
9885	&& u8Val != 7 /* UC- */)
9886	HMVMX_ERROR_BREAK(VMX_IGS_PAT_MSR_INVALID);
9887	u64Val >>= 8;
9888	}
9889	}
9890
9891	/*
9892	* EFER MSR.
9893	*/
9894	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_EFER_MSR)
9895	{
9896	Assert(g_fHmVmxSupportsVmcsEfer);
9897	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_EFER_FULL, &u64Val);
9898	AssertRC(rc);
9899	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffffffffff2fe)),
9900	VMX_IGS_EFER_MSR_RESERVED); /* Bits 63:12, bit 9, bits 7:1 MBZ. */
9901	HMVMX_CHECK_BREAK(RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL( pVmcsInfo->u32EntryCtls
9902	& VMX_ENTRY_CTLS_IA32E_MODE_GUEST),
9903	VMX_IGS_EFER_LMA_GUEST_MODE_MISMATCH);
9904	/** @todo r=ramshankar: Unrestricted check here is probably wrong, see
9905	* iemVmxVmentryCheckGuestState(). */
9906	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9907	\|\| !(u64GuestCr0 & X86_CR0_PG)
9908	\|\| RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL(u64Val & MSR_K6_EFER_LME),
9909	VMX_IGS_EFER_LMA_LME_MISMATCH);
9910	}
9911
9912	/*
9913	* Segment registers.
9914	*/
9915	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9916	\|\| !(pCtx->ldtr.Sel & X86_SEL_LDT), VMX_IGS_LDTR_TI_INVALID);
9917	if (!(u32Eflags & X86_EFL_VM))
9918	{
9919	/* CS */
9920	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1Present, VMX_IGS_CS_ATTR_P_INVALID);
9921	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xf00), VMX_IGS_CS_ATTR_RESERVED);
9922	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xfffe0000), VMX_IGS_CS_ATTR_RESERVED);
9923	HMVMX_CHECK_BREAK( (pCtx->cs.u32Limit & 0xfff) == 0xfff
9924	\|\| !(pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9925	HMVMX_CHECK_BREAK( !(pCtx->cs.u32Limit & 0xfff00000)
9926	\|\| (pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9927	/* CS cannot be loaded with NULL in protected mode. */
9928	HMVMX_CHECK_BREAK(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_CS_ATTR_UNUSABLE);
9929	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1DescType, VMX_IGS_CS_ATTR_S_INVALID);
9930	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
9931	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_UNEQUAL);
9932	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
9933	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_MISMATCH);
9934	else if (fUnrestrictedGuest && pCtx->cs.Attr.n.u4Type == 3)
9935	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == 0, VMX_IGS_CS_ATTR_DPL_INVALID);
9936	else
9937	HMVMX_ERROR_BREAK(VMX_IGS_CS_ATTR_TYPE_INVALID);
9938
9939	/* SS */
9940	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9941	\|\| (pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL), VMX_IGS_SS_CS_RPL_UNEQUAL);
9942	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL), VMX_IGS_SS_ATTR_DPL_RPL_UNEQUAL);
9943	if ( !(pCtx->cr0 & X86_CR0_PE)
9944	\|\| pCtx->cs.Attr.n.u4Type == 3)
9945	HMVMX_CHECK_BREAK(!pCtx->ss.Attr.n.u2Dpl, VMX_IGS_SS_ATTR_DPL_INVALID);
9946
9947	if (!(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
9948	{
9949	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7, VMX_IGS_SS_ATTR_TYPE_INVALID);
9950	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u1Present, VMX_IGS_SS_ATTR_P_INVALID);
9951	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xf00), VMX_IGS_SS_ATTR_RESERVED);
9952	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xfffe0000), VMX_IGS_SS_ATTR_RESERVED);
9953	HMVMX_CHECK_BREAK( (pCtx->ss.u32Limit & 0xfff) == 0xfff
9954	\|\| !(pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9955	HMVMX_CHECK_BREAK( !(pCtx->ss.u32Limit & 0xfff00000)
9956	\|\| (pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9957	}
9958
9959	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxExportGuestSReg(). */
9960	if (!(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
9961	{
9962	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_DS_ATTR_A_INVALID);
9963	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u1Present, VMX_IGS_DS_ATTR_P_INVALID);
9964	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9965	\|\| pCtx->ds.Attr.n.u4Type > 11
9966	\|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9967	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xf00), VMX_IGS_DS_ATTR_RESERVED);
9968	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xfffe0000), VMX_IGS_DS_ATTR_RESERVED);
9969	HMVMX_CHECK_BREAK( (pCtx->ds.u32Limit & 0xfff) == 0xfff
9970	\|\| !(pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9971	HMVMX_CHECK_BREAK( !(pCtx->ds.u32Limit & 0xfff00000)
9972	\|\| (pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9973	HMVMX_CHECK_BREAK( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9974	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_DS_ATTR_TYPE_INVALID);
9975	}
9976	if (!(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
9977	{
9978	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_ES_ATTR_A_INVALID);
9979	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u1Present, VMX_IGS_ES_ATTR_P_INVALID);
9980	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9981	\|\| pCtx->es.Attr.n.u4Type > 11
9982	\|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9983	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xf00), VMX_IGS_ES_ATTR_RESERVED);
9984	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xfffe0000), VMX_IGS_ES_ATTR_RESERVED);
9985	HMVMX_CHECK_BREAK( (pCtx->es.u32Limit & 0xfff) == 0xfff
9986	\|\| !(pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
9987	HMVMX_CHECK_BREAK( !(pCtx->es.u32Limit & 0xfff00000)
9988	\|\| (pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
9989	HMVMX_CHECK_BREAK( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9990	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_ES_ATTR_TYPE_INVALID);
9991	}
9992	if (!(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
9993	{
9994	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_FS_ATTR_A_INVALID);
9995	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u1Present, VMX_IGS_FS_ATTR_P_INVALID);
9996	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9997	\|\| pCtx->fs.Attr.n.u4Type > 11
9998	\|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL), VMX_IGS_FS_ATTR_DPL_RPL_UNEQUAL);
9999	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xf00), VMX_IGS_FS_ATTR_RESERVED);
10000	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xfffe0000), VMX_IGS_FS_ATTR_RESERVED);
10001	HMVMX_CHECK_BREAK( (pCtx->fs.u32Limit & 0xfff) == 0xfff
10002	\|\| !(pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
10003	HMVMX_CHECK_BREAK( !(pCtx->fs.u32Limit & 0xfff00000)
10004	\|\| (pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
10005	HMVMX_CHECK_BREAK( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10006	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_FS_ATTR_TYPE_INVALID);
10007	}
10008	if (!(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
10009	{
10010	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_GS_ATTR_A_INVALID);
10011	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u1Present, VMX_IGS_GS_ATTR_P_INVALID);
10012	HMVMX_CHECK_BREAK( fUnrestrictedGuest
10013	\|\| pCtx->gs.Attr.n.u4Type > 11
10014	\|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL), VMX_IGS_GS_ATTR_DPL_RPL_UNEQUAL);
10015	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xf00), VMX_IGS_GS_ATTR_RESERVED);
10016	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xfffe0000), VMX_IGS_GS_ATTR_RESERVED);
10017	HMVMX_CHECK_BREAK( (pCtx->gs.u32Limit & 0xfff) == 0xfff
10018	\|\| !(pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
10019	HMVMX_CHECK_BREAK( !(pCtx->gs.u32Limit & 0xfff00000)
10020	\|\| (pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
10021	HMVMX_CHECK_BREAK( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10022	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_GS_ATTR_TYPE_INVALID);
10023	}
10024	/* 64-bit capable CPUs. */
10025	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
10026	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
10027	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
10028	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
10029	HMVMX_CHECK_BREAK(!RT_HI_U32(pCtx->cs.u64Base), VMX_IGS_LONGMODE_CS_BASE_INVALID);
10030	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ss.u64Base),
10031	VMX_IGS_LONGMODE_SS_BASE_INVALID);
10032	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ds.u64Base),
10033	VMX_IGS_LONGMODE_DS_BASE_INVALID);
10034	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->es.u64Base),
10035	VMX_IGS_LONGMODE_ES_BASE_INVALID);
10036	}
10037	else
10038	{
10039	/* V86 mode checks. */
10040	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
10041	if (pVmcsInfo->pShared->RealMode.fRealOnV86Active)
10042	{
10043	u32CSAttr = 0xf3; u32SSAttr = 0xf3;
10044	u32DSAttr = 0xf3; u32ESAttr = 0xf3;
10045	u32FSAttr = 0xf3; u32GSAttr = 0xf3;
10046	}
10047	else
10048	{
10049	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u;
10050	u32DSAttr = pCtx->ds.Attr.u; u32ESAttr = pCtx->es.Attr.u;
10051	u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
10052	}
10053
10054	/* CS */
10055	HMVMX_CHECK_BREAK((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), VMX_IGS_V86_CS_BASE_INVALID);
10056	HMVMX_CHECK_BREAK(pCtx->cs.u32Limit == 0xffff, VMX_IGS_V86_CS_LIMIT_INVALID);
10057	HMVMX_CHECK_BREAK(u32CSAttr == 0xf3, VMX_IGS_V86_CS_ATTR_INVALID);
10058	/* SS */
10059	HMVMX_CHECK_BREAK((pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4), VMX_IGS_V86_SS_BASE_INVALID);
10060	HMVMX_CHECK_BREAK(pCtx->ss.u32Limit == 0xffff, VMX_IGS_V86_SS_LIMIT_INVALID);
10061	HMVMX_CHECK_BREAK(u32SSAttr == 0xf3, VMX_IGS_V86_SS_ATTR_INVALID);
10062	/* DS */
10063	HMVMX_CHECK_BREAK((pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4), VMX_IGS_V86_DS_BASE_INVALID);
10064	HMVMX_CHECK_BREAK(pCtx->ds.u32Limit == 0xffff, VMX_IGS_V86_DS_LIMIT_INVALID);
10065	HMVMX_CHECK_BREAK(u32DSAttr == 0xf3, VMX_IGS_V86_DS_ATTR_INVALID);
10066	/* ES */
10067	HMVMX_CHECK_BREAK((pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4), VMX_IGS_V86_ES_BASE_INVALID);
10068	HMVMX_CHECK_BREAK(pCtx->es.u32Limit == 0xffff, VMX_IGS_V86_ES_LIMIT_INVALID);
10069	HMVMX_CHECK_BREAK(u32ESAttr == 0xf3, VMX_IGS_V86_ES_ATTR_INVALID);
10070	/* FS */
10071	HMVMX_CHECK_BREAK((pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4), VMX_IGS_V86_FS_BASE_INVALID);
10072	HMVMX_CHECK_BREAK(pCtx->fs.u32Limit == 0xffff, VMX_IGS_V86_FS_LIMIT_INVALID);
10073	HMVMX_CHECK_BREAK(u32FSAttr == 0xf3, VMX_IGS_V86_FS_ATTR_INVALID);
10074	/* GS */
10075	HMVMX_CHECK_BREAK((pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4), VMX_IGS_V86_GS_BASE_INVALID);
10076	HMVMX_CHECK_BREAK(pCtx->gs.u32Limit == 0xffff, VMX_IGS_V86_GS_LIMIT_INVALID);
10077	HMVMX_CHECK_BREAK(u32GSAttr == 0xf3, VMX_IGS_V86_GS_ATTR_INVALID);
10078	/* 64-bit capable CPUs. */
10079	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
10080	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
10081	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
10082	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
10083	HMVMX_CHECK_BREAK(!RT_HI_U32(pCtx->cs.u64Base), VMX_IGS_LONGMODE_CS_BASE_INVALID);
10084	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ss.u64Base),
10085	VMX_IGS_LONGMODE_SS_BASE_INVALID);
10086	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ds.u64Base),
10087	VMX_IGS_LONGMODE_DS_BASE_INVALID);
10088	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->es.u64Base),
10089	VMX_IGS_LONGMODE_ES_BASE_INVALID);
10090	}
10091
10092	/*
10093	* TR.
10094	*/
10095	HMVMX_CHECK_BREAK(!(pCtx->tr.Sel & X86_SEL_LDT), VMX_IGS_TR_TI_INVALID);
10096	/* 64-bit capable CPUs. */
10097	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->tr.u64Base), VMX_IGS_TR_BASE_NOT_CANONICAL);
10098	if (fLongModeGuest)
10099	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u4Type == 11, /* 64-bit busy TSS. */
10100	VMX_IGS_LONGMODE_TR_ATTR_TYPE_INVALID);
10101	else
10102	HMVMX_CHECK_BREAK( pCtx->tr.Attr.n.u4Type == 3 /* 16-bit busy TSS. */
10103	\|\| pCtx->tr.Attr.n.u4Type == 11, /* 32-bit busy TSS.*/
10104	VMX_IGS_TR_ATTR_TYPE_INVALID);
10105	HMVMX_CHECK_BREAK(!pCtx->tr.Attr.n.u1DescType, VMX_IGS_TR_ATTR_S_INVALID);
10106	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u1Present, VMX_IGS_TR_ATTR_P_INVALID);
10107	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & 0xf00), VMX_IGS_TR_ATTR_RESERVED); /* Bits 11:8 MBZ. */
10108	HMVMX_CHECK_BREAK( (pCtx->tr.u32Limit & 0xfff) == 0xfff
10109	\|\| !(pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
10110	HMVMX_CHECK_BREAK( !(pCtx->tr.u32Limit & 0xfff00000)
10111	\|\| (pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
10112	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_TR_ATTR_UNUSABLE);
10113
10114	/*
10115	* GDTR and IDTR (64-bit capable checks).
10116	*/
10117	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
10118	AssertRC(rc);
10119	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_GDTR_BASE_NOT_CANONICAL);
10120
10121	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
10122	AssertRC(rc);
10123	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_IDTR_BASE_NOT_CANONICAL);
10124
10125	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val);
10126	AssertRC(rc);
10127	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_GDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
10128
10129	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val);
10130	AssertRC(rc);
10131	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_IDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
10132
10133	/*
10134	* Guest Non-Register State.
10135	*/
10136	/* Activity State. */
10137	uint32_t u32ActivityState;
10138	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, &u32ActivityState);
10139	AssertRC(rc);
10140	HMVMX_CHECK_BREAK( !u32ActivityState
10141	\|\| (u32ActivityState & RT_BF_GET(g_HmMsrs.u.vmx.u64Misc, VMX_BF_MISC_ACTIVITY_STATES)),
10142	VMX_IGS_ACTIVITY_STATE_INVALID);
10143	HMVMX_CHECK_BREAK( !(pCtx->ss.Attr.n.u2Dpl)
10144	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT, VMX_IGS_ACTIVITY_STATE_HLT_INVALID);
10145
10146	if ( u32IntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS
10147	\|\| u32IntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10148	HMVMX_CHECK_BREAK(u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE, VMX_IGS_ACTIVITY_STATE_ACTIVE_INVALID);
10149
10150	/** @todo Activity state and injecting interrupts. Left as a todo since we
10151	* currently don't use activity states but ACTIVE. */
10152
10153	HMVMX_CHECK_BREAK( !(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)
10154	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT, VMX_IGS_ACTIVITY_STATE_SIPI_WAIT_INVALID);
10155
10156	/* Guest interruptibility-state. */
10157	HMVMX_CHECK_BREAK(!(u32IntrState & 0xffffffe0), VMX_IGS_INTERRUPTIBILITY_STATE_RESERVED);
10158	HMVMX_CHECK_BREAK((u32IntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
10159	!= (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
10160	VMX_IGS_INTERRUPTIBILITY_STATE_STI_MOVSS_INVALID);
10161	HMVMX_CHECK_BREAK( (u32Eflags & X86_EFL_IF)
10162	\|\| !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI),
10163	VMX_IGS_INTERRUPTIBILITY_STATE_STI_EFL_INVALID);
10164	if (VMX_ENTRY_INT_INFO_IS_EXT_INT(u32EntryInfo))
10165	{
10166	HMVMX_CHECK_BREAK( !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10167	&& !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
10168	VMX_IGS_INTERRUPTIBILITY_STATE_EXT_INT_INVALID);
10169	}
10170	else if (VMX_ENTRY_INT_INFO_IS_XCPT_NMI(u32EntryInfo))
10171	{
10172	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
10173	VMX_IGS_INTERRUPTIBILITY_STATE_MOVSS_INVALID);
10174	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI),
10175	VMX_IGS_INTERRUPTIBILITY_STATE_STI_INVALID);
10176	}
10177	/** @todo Assumes the processor is not in SMM. */
10178	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI),
10179	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_INVALID);
10180	HMVMX_CHECK_BREAK( !(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)
10181	\|\| (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI),
10182	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_SMM_INVALID);
10183	if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
10184	&& VMX_ENTRY_INT_INFO_IS_XCPT_NMI(u32EntryInfo))
10185	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI), VMX_IGS_INTERRUPTIBILITY_STATE_NMI_INVALID);
10186
10187	/* Pending debug exceptions. */
10188	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, &u64Val);
10189	AssertRC(rc);
10190	/* Bits 63:15, Bit 13, Bits 11:4 MBZ. */
10191	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffffaff0)), VMX_IGS_LONGMODE_PENDING_DEBUG_RESERVED);
10192	u32Val = u64Val; /* For pending debug exceptions checks below. */
10193
10194	if ( (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10195	\|\| (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS)
10196	\|\| u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT)
10197	{
10198	if ( (u32Eflags & X86_EFL_TF)
10199	&& !(u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
10200	{
10201	/* Bit 14 is PendingDebug.BS. */
10202	HMVMX_CHECK_BREAK(u32Val & RT_BIT(14), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_SET);
10203	}
10204	if ( !(u32Eflags & X86_EFL_TF)
10205	\|\| (u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
10206	{
10207	/* Bit 14 is PendingDebug.BS. */
10208	HMVMX_CHECK_BREAK(!(u32Val & RT_BIT(14)), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_CLEAR);
10209	}
10210	}
10211
10212	/* VMCS link pointer. */
10213	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, &u64Val);
10214	AssertRC(rc);
10215	if (u64Val != UINT64_C(0xffffffffffffffff))
10216	{
10217	HMVMX_CHECK_BREAK(!(u64Val & 0xfff), VMX_IGS_VMCS_LINK_PTR_RESERVED);
10218	/** @todo Bits beyond the processor's physical-address width MBZ. */
10219	/** @todo SMM checks. */
10220	Assert(pVmcsInfo->HCPhysShadowVmcs == u64Val);
10221	Assert(pVmcsInfo->pvShadowVmcs);
10222	VMXVMCSREVID VmcsRevId;
10223	VmcsRevId.u = (uint32_t )pVmcsInfo->pvShadowVmcs;
10224	HMVMX_CHECK_BREAK(VmcsRevId.n.u31RevisionId == RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_ID),
10225	VMX_IGS_VMCS_LINK_PTR_SHADOW_VMCS_ID_INVALID);
10226	HMVMX_CHECK_BREAK(VmcsRevId.n.fIsShadowVmcs == (uint32_t)!!(pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING),
10227	VMX_IGS_VMCS_LINK_PTR_NOT_SHADOW);
10228	}
10229
10230	/** @todo Checks on Guest Page-Directory-Pointer-Table Entries when guest is
10231	* not using nested paging? */
10232	if ( pVM->hmr0.s.fNestedPaging
10233	&& !fLongModeGuest
10234	&& CPUMIsGuestInPAEModeEx(pCtx))
10235	{
10236	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &u64Val);
10237	AssertRC(rc);
10238	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10239
10240	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &u64Val);
10241	AssertRC(rc);
10242	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10243
10244	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &u64Val);
10245	AssertRC(rc);
10246	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10247
10248	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &u64Val);
10249	AssertRC(rc);
10250	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10251	}
10252
10253	/* Shouldn't happen but distinguish it from AssertRCBreak() errors. */
10254	if (uError == VMX_IGS_ERROR)
10255	uError = VMX_IGS_REASON_NOT_FOUND;
10256	} while (0);
10257
10258	pVCpu->hm.s.u32HMError = uError;
10259	pVCpu->hm.s.vmx.LastError.u32GuestIntrState = u32IntrState;
10260	return uError;
10261
10262	#undef HMVMX_ERROR_BREAK
10263	#undef HMVMX_CHECK_BREAK
10264	}
10265
10266
10267	/**
10268	* Map the APIC-access page for virtualizing APIC accesses.
10269	*
10270	* This can cause a longjumps to R3 due to the acquisition of the PGM lock. Hence,
10271	* this not done as part of exporting guest state, see @bugref{8721}.
10272	*
10273	* @returns VBox status code.
10274	* @param pVCpu The cross context virtual CPU structure.
10275	*/
10276	static int hmR0VmxMapHCApicAccessPage(PVMCPUCC pVCpu)
10277	{
10278	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10279	uint64_t const u64MsrApicBase = APICGetBaseMsrNoCheck(pVCpu);
10280
10281	Assert(PDMHasApic(pVM));
10282	Assert(u64MsrApicBase);
10283
10284	RTGCPHYS const GCPhysApicBase = u64MsrApicBase & PAGE_BASE_GC_MASK;
10285	Log4Func(("Mappping HC APIC-access page at %#RGp\n", GCPhysApicBase));
10286
10287	/* Unalias the existing mapping. */
10288	int rc = PGMHandlerPhysicalReset(pVM, GCPhysApicBase);
10289	AssertRCReturn(rc, rc);
10290
10291	/* Map the HC APIC-access page in place of the MMIO page, also updates the shadow page tables if necessary. */
10292	Assert(pVM->hmr0.s.vmx.HCPhysApicAccess != NIL_RTHCPHYS);
10293	rc = IOMR0MmioMapMmioHCPage(pVM, pVCpu, GCPhysApicBase, pVM->hmr0.s.vmx.HCPhysApicAccess, X86_PTE_RW \| X86_PTE_P);
10294	AssertRCReturn(rc, rc);
10295
10296	/* Update the per-VCPU cache of the APIC base MSR. */
10297	pVCpu->hm.s.vmx.u64GstMsrApicBase = u64MsrApicBase;
10298	return VINF_SUCCESS;
10299	}
10300
10301
10302	/**
10303	* Worker function passed to RTMpOnSpecific() that is to be called on the target
10304	* CPU.
10305	*
10306	* @param idCpu The ID for the CPU the function is called on.
10307	* @param pvUser1 Null, not used.
10308	* @param pvUser2 Null, not used.
10309	*/
10310	static DECLCALLBACK(void) hmR0DispatchHostNmi(RTCPUID idCpu, void pvUser1, void pvUser2)
10311	{
10312	RT_NOREF3(idCpu, pvUser1, pvUser2);
10313	VMXDispatchHostNmi();
10314	}
10315
10316
10317	/**
10318	* Dispatching an NMI on the host CPU that received it.
10319	*
10320	* @returns VBox status code.
10321	* @param pVCpu The cross context virtual CPU structure.
10322	* @param pVmcsInfo The VMCS info. object corresponding to the VMCS that was
10323	* executing when receiving the host NMI in VMX non-root
10324	* operation.
10325	*/
10326	static int hmR0VmxExitHostNmi(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
10327	{
10328	RTCPUID const idCpu = pVmcsInfo->idHostCpuExec;
10329	Assert(idCpu != NIL_RTCPUID);
10330
10331	/*
10332	* We don't want to delay dispatching the NMI any more than we have to. However,
10333	* we have already chosen -not- to dispatch NMIs when interrupts were still disabled
10334	* after executing guest or nested-guest code for the following reasons:
10335	*
10336	* - We would need to perform VMREADs with interrupts disabled and is orders of
10337	* magnitude worse when we run as a nested hypervisor without VMCS shadowing
10338	* supported by the host hypervisor.
10339	*
10340	* - It affects the common VM-exit scenario and keeps interrupts disabled for a
10341	* longer period of time just for handling an edge case like host NMIs which do
10342	* not occur nearly as frequently as other VM-exits.
10343	*
10344	* Let's cover the most likely scenario first. Check if we are on the target CPU
10345	* and dispatch the NMI right away. This should be much faster than calling into
10346	* RTMpOnSpecific() machinery.
10347	*/
10348	bool fDispatched = false;
10349	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
10350	if (idCpu == RTMpCpuId())
10351	{
10352	VMXDispatchHostNmi();
10353	fDispatched = true;
10354	}
10355	ASMSetFlags(fEFlags);
10356	if (fDispatched)
10357	{
10358	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC);
10359	return VINF_SUCCESS;
10360	}
10361
10362	/*
10363	* RTMpOnSpecific() waits until the worker function has run on the target CPU. So
10364	* there should be no race or recursion even if we are unlucky enough to be preempted
10365	* (to the target CPU) without dispatching the host NMI above.
10366	*/
10367	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGCIpi);
10368	return RTMpOnSpecific(idCpu, &hmR0DispatchHostNmi, NULL /* pvUser1 /, NULL / pvUser2 */);
10369	}
10370
10371
10372	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10373	/**
10374	* Merges the guest with the nested-guest MSR bitmap in preparation of executing the
10375	* nested-guest using hardware-assisted VMX.
10376	*
10377	* @param pVCpu The cross context virtual CPU structure.
10378	* @param pVmcsInfoNstGst The nested-guest VMCS info. object.
10379	* @param pVmcsInfoGst The guest VMCS info. object.
10380	*/
10381	static void hmR0VmxMergeMsrBitmapNested(PCVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfoNstGst, PCVMXVMCSINFO pVmcsInfoGst)
10382	{
10383	uint32_t const cbMsrBitmap = X86_PAGE_4K_SIZE;
10384	uint64_t pu64MsrBitmap = (uint64_t )pVmcsInfoNstGst->pvMsrBitmap;
10385	Assert(pu64MsrBitmap);
10386
10387	/*
10388	* We merge the guest MSR bitmap with the nested-guest MSR bitmap such that any
10389	* MSR that is intercepted by the guest is also intercepted while executing the
10390	* nested-guest using hardware-assisted VMX.
10391	*
10392	* Note! If the nested-guest is not using an MSR bitmap, every MSR must cause a
10393	* nested-guest VM-exit even if the outer guest is not intercepting some
10394	* MSRs. We cannot assume the caller has initialized the nested-guest
10395	* MSR bitmap in this case.
10396	*
10397	* The nested hypervisor may also switch whether it uses MSR bitmaps for
10398	* each of its VM-entry, hence initializing it once per-VM while setting
10399	* up the nested-guest VMCS is not sufficient.
10400	*/
10401	PCVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
10402	if (pVmcsNstGst->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
10403	{
10404	uint64_t const pu64MsrBitmapNstGst = (uint64_t const )&pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap[0];
10405	uint64_t const pu64MsrBitmapGst = (uint64_t const )pVmcsInfoGst->pvMsrBitmap;
10406	Assert(pu64MsrBitmapNstGst);
10407	Assert(pu64MsrBitmapGst);
10408
10409	/** @todo Detect and use EVEX.POR? */
10410	uint32_t const cFrags = cbMsrBitmap / sizeof(uint64_t);
10411	for (uint32_t i = 0; i < cFrags; i++)
10412	pu64MsrBitmap[i] = pu64MsrBitmapNstGst[i] \| pu64MsrBitmapGst[i];
10413	}
10414	else
10415	ASMMemFill32(pu64MsrBitmap, cbMsrBitmap, UINT32_C(0xffffffff));
10416	}
10417
10418
10419	/**
10420	* Merges the guest VMCS in to the nested-guest VMCS controls in preparation of
10421	* hardware-assisted VMX execution of the nested-guest.
10422	*
10423	* For a guest, we don't modify these controls once we set up the VMCS and hence
10424	* this function is never called.
10425	*
10426	* For nested-guests since the nested hypervisor provides these controls on every
10427	* nested-guest VM-entry and could potentially change them everytime we need to
10428	* merge them before every nested-guest VM-entry.
10429	*
10430	* @returns VBox status code.
10431	* @param pVCpu The cross context virtual CPU structure.
10432	*/
10433	static int hmR0VmxMergeVmcsNested(PVMCPUCC pVCpu)
10434	{
10435	PVMCC const pVM = pVCpu->CTX_SUFF(pVM);
10436	PCVMXVMCSINFO const pVmcsInfoGst = &pVCpu->hmr0.s.vmx.VmcsInfo;
10437	PCVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
10438
10439	/*
10440	* Merge the controls with the requirements of the guest VMCS.
10441	*
10442	* We do not need to validate the nested-guest VMX features specified in the nested-guest
10443	* VMCS with the features supported by the physical CPU as it's already done by the
10444	* VMLAUNCH/VMRESUME instruction emulation.
10445	*
10446	* This is because the VMX features exposed by CPUM (through CPUID/MSRs) to the guest are
10447	* derived from the VMX features supported by the physical CPU.
10448	*/
10449
10450	/* Pin-based VM-execution controls. */
10451	uint32_t const u32PinCtls = pVmcsNstGst->u32PinCtls \| pVmcsInfoGst->u32PinCtls;
10452
10453	/* Processor-based VM-execution controls. */
10454	uint32_t u32ProcCtls = (pVmcsNstGst->u32ProcCtls & ~VMX_PROC_CTLS_USE_IO_BITMAPS)
10455	\| (pVmcsInfoGst->u32ProcCtls & ~( VMX_PROC_CTLS_INT_WINDOW_EXIT
10456	\| VMX_PROC_CTLS_NMI_WINDOW_EXIT
10457	\| VMX_PROC_CTLS_MOV_DR_EXIT
10458	\| VMX_PROC_CTLS_USE_TPR_SHADOW
10459	\| VMX_PROC_CTLS_MONITOR_TRAP_FLAG));
10460
10461	/* Secondary processor-based VM-execution controls. */
10462	uint32_t const u32ProcCtls2 = (pVmcsNstGst->u32ProcCtls2 & ~VMX_PROC_CTLS2_VPID)
10463	\| (pVmcsInfoGst->u32ProcCtls2 & ~( VMX_PROC_CTLS2_VIRT_APIC_ACCESS
10464	\| VMX_PROC_CTLS2_INVPCID
10465	\| VMX_PROC_CTLS2_VMCS_SHADOWING
10466	\| VMX_PROC_CTLS2_RDTSCP
10467	\| VMX_PROC_CTLS2_XSAVES_XRSTORS
10468	\| VMX_PROC_CTLS2_APIC_REG_VIRT
10469	\| VMX_PROC_CTLS2_VIRT_INT_DELIVERY
10470	\| VMX_PROC_CTLS2_VMFUNC));
10471
10472	/*
10473	* VM-entry controls:
10474	* These controls contains state that depends on the nested-guest state (primarily
10475	* EFER MSR) and is thus not constant between VMLAUNCH/VMRESUME and the nested-guest
10476	* VM-exit. Although the nested hypervisor cannot change it, we need to in order to
10477	* properly continue executing the nested-guest if the EFER MSR changes but does not
10478	* cause a nested-guest VM-exits.
10479	*
10480	* VM-exit controls:
10481	* These controls specify the host state on return. We cannot use the controls from
10482	* the nested hypervisor state as is as it would contain the guest state rather than
10483	* the host state. Since the host state is subject to change (e.g. preemption, trips
10484	* to ring-3, longjmp and rescheduling to a different host CPU) they are not constant
10485	* through VMLAUNCH/VMRESUME and the nested-guest VM-exit.
10486	*
10487	* VM-entry MSR-load:
10488	* The guest MSRs from the VM-entry MSR-load area are already loaded into the guest-CPU
10489	* context by the VMLAUNCH/VMRESUME instruction emulation.
10490	*
10491	* VM-exit MSR-store:
10492	* The VM-exit emulation will take care of populating the MSRs from the guest-CPU context
10493	* back into the VM-exit MSR-store area.
10494	*
10495	* VM-exit MSR-load areas:
10496	* This must contain the real host MSRs with hardware-assisted VMX execution. Hence, we
10497	* can entirely ignore what the nested hypervisor wants to load here.
10498	*/
10499
10500	/*
10501	* Exception bitmap.
10502	*
10503	* We could remove #UD from the guest bitmap and merge it with the nested-guest bitmap
10504	* here (and avoid doing anything while exporting nested-guest state), but to keep the
10505	* code more flexible if intercepting exceptions become more dynamic in the future we do
10506	* it as part of exporting the nested-guest state.
10507	*/
10508	uint32_t const u32XcptBitmap = pVmcsNstGst->u32XcptBitmap \| pVmcsInfoGst->u32XcptBitmap;
10509
10510	/*
10511	* CR0/CR4 guest/host mask.
10512	*
10513	* Modifications by the nested-guest to CR0/CR4 bits owned by the host and the guest must
10514	* cause VM-exits, so we need to merge them here.
10515	*/
10516	uint64_t const u64Cr0Mask = pVmcsNstGst->u64Cr0Mask.u \| pVmcsInfoGst->u64Cr0Mask;
10517	uint64_t const u64Cr4Mask = pVmcsNstGst->u64Cr4Mask.u \| pVmcsInfoGst->u64Cr4Mask;
10518
10519	/*
10520	* Page-fault error-code mask and match.
10521	*
10522	* Although we require unrestricted guest execution (and thereby nested-paging) for
10523	* hardware-assisted VMX execution of nested-guests and thus the outer guest doesn't
10524	* normally intercept #PFs, it might intercept them for debugging purposes.
10525	*
10526	* If the outer guest is not intercepting #PFs, we can use the nested-guest #PF filters.
10527	* If the outer guest is intercepting #PFs, we must intercept all #PFs.
10528	*/
10529	uint32_t u32XcptPFMask;
10530	uint32_t u32XcptPFMatch;
10531	if (!(pVmcsInfoGst->u32XcptBitmap & RT_BIT(X86_XCPT_PF)))
10532	{
10533	u32XcptPFMask = pVmcsNstGst->u32XcptPFMask;
10534	u32XcptPFMatch = pVmcsNstGst->u32XcptPFMatch;
10535	}
10536	else
10537	{
10538	u32XcptPFMask = 0;
10539	u32XcptPFMatch = 0;
10540	}
10541
10542	/*
10543	* Pause-Loop exiting.
10544	*/
10545	/** @todo r=bird: given that both pVM->hm.s.vmx.cPleGapTicks and
10546	* pVM->hm.s.vmx.cPleWindowTicks defaults to zero, I cannot see how
10547	* this will work... */
10548	uint32_t const cPleGapTicks = RT_MIN(pVM->hm.s.vmx.cPleGapTicks, pVmcsNstGst->u32PleGap);
10549	uint32_t const cPleWindowTicks = RT_MIN(pVM->hm.s.vmx.cPleWindowTicks, pVmcsNstGst->u32PleWindow);
10550
10551	/*
10552	* Pending debug exceptions.
10553	* Currently just copy whatever the nested-guest provides us.
10554	*/
10555	uint64_t const uPendingDbgXcpts = pVmcsNstGst->u64GuestPendingDbgXcpts.u;
10556
10557	/*
10558	* I/O Bitmap.
10559	*
10560	* We do not use the I/O bitmap that may be provided by the nested hypervisor as we always
10561	* intercept all I/O port accesses.
10562	*/
10563	Assert(u32ProcCtls & VMX_PROC_CTLS_UNCOND_IO_EXIT);
10564	Assert(!(u32ProcCtls & VMX_PROC_CTLS_USE_IO_BITMAPS));
10565
10566	/*
10567	* VMCS shadowing.
10568	*
10569	* We do not yet expose VMCS shadowing to the guest and thus VMCS shadowing should not be
10570	* enabled while executing the nested-guest.
10571	*/
10572	Assert(!(u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING));
10573
10574	/*
10575	* APIC-access page.
10576	*/
10577	RTHCPHYS HCPhysApicAccess;
10578	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10579	{
10580	Assert(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS);
10581	RTGCPHYS const GCPhysApicAccess = pVmcsNstGst->u64AddrApicAccess.u;
10582
10583	/** @todo NSTVMX: This is not really correct but currently is required to make
10584	* things work. We need to re-enable the page handler when we fallback to
10585	* IEM execution of the nested-guest! */
10586	PGMHandlerPhysicalPageTempOff(pVM, GCPhysApicAccess, GCPhysApicAccess);
10587
10588	void *pvPage;
10589	PGMPAGEMAPLOCK PgLockApicAccess;
10590	int rc = PGMPhysGCPhys2CCPtr(pVM, GCPhysApicAccess, &pvPage, &PgLockApicAccess);
10591	if (RT_SUCCESS(rc))
10592	{
10593	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysApicAccess, &HCPhysApicAccess);
10594	AssertMsgRCReturn(rc, ("Failed to get host-physical address for APIC-access page at %#RGp\n", GCPhysApicAccess), rc);
10595
10596	/** @todo Handle proper releasing of page-mapping lock later. */
10597	PGMPhysReleasePageMappingLock(pVCpu->CTX_SUFF(pVM), &PgLockApicAccess);
10598	}
10599	else
10600	return rc;
10601	}
10602	else
10603	HCPhysApicAccess = 0;
10604
10605	/*
10606	* Virtual-APIC page and TPR threshold.
10607	*/
10608	RTHCPHYS HCPhysVirtApic;
10609	uint32_t u32TprThreshold;
10610	if (u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
10611	{
10612	Assert(g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW);
10613	RTGCPHYS const GCPhysVirtApic = pVmcsNstGst->u64AddrVirtApic.u;
10614
10615	void *pvPage;
10616	PGMPAGEMAPLOCK PgLockVirtApic;
10617	int rc = PGMPhysGCPhys2CCPtr(pVM, GCPhysVirtApic, &pvPage, &PgLockVirtApic);
10618	if (RT_SUCCESS(rc))
10619	{
10620	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysVirtApic, &HCPhysVirtApic);
10621	AssertMsgRCReturn(rc, ("Failed to get host-physical address for virtual-APIC page at %#RGp\n", GCPhysVirtApic), rc);
10622
10623	/** @todo Handle proper releasing of page-mapping lock later. */
10624	PGMPhysReleasePageMappingLock(pVCpu->CTX_SUFF(pVM), &PgLockVirtApic);
10625	}
10626	else
10627	return rc;
10628
10629	u32TprThreshold = pVmcsNstGst->u32TprThreshold;
10630	}
10631	else
10632	{
10633	HCPhysVirtApic = 0;
10634	u32TprThreshold = 0;
10635
10636	/*
10637	* We must make sure CR8 reads/write must cause VM-exits when TPR shadowing is not
10638	* used by the nested hypervisor. Preventing MMIO accesses to the physical APIC will
10639	* be taken care of by EPT/shadow paging.
10640	*/
10641	if (pVM->hmr0.s.fAllow64BitGuests)
10642	u32ProcCtls \|= VMX_PROC_CTLS_CR8_STORE_EXIT
10643	\| VMX_PROC_CTLS_CR8_LOAD_EXIT;
10644	}
10645
10646	/*
10647	* Validate basic assumptions.
10648	*/
10649	PVMXVMCSINFO pVmcsInfoNstGst = &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
10650	Assert(pVM->hmr0.s.vmx.fUnrestrictedGuest);
10651	Assert(g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS);
10652	Assert(hmGetVmxActiveVmcsInfo(pVCpu) == pVmcsInfoNstGst);
10653
10654	/*
10655	* Commit it to the nested-guest VMCS.
10656	*/
10657	int rc = VINF_SUCCESS;
10658	if (pVmcsInfoNstGst->u32PinCtls != u32PinCtls)
10659	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, u32PinCtls);
10660	if (pVmcsInfoNstGst->u32ProcCtls != u32ProcCtls)
10661	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, u32ProcCtls);
10662	if (pVmcsInfoNstGst->u32ProcCtls2 != u32ProcCtls2)
10663	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, u32ProcCtls2);
10664	if (pVmcsInfoNstGst->u32XcptBitmap != u32XcptBitmap)
10665	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
10666	if (pVmcsInfoNstGst->u64Cr0Mask != u64Cr0Mask)
10667	rc \|= VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, u64Cr0Mask);
10668	if (pVmcsInfoNstGst->u64Cr4Mask != u64Cr4Mask)
10669	rc \|= VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, u64Cr4Mask);
10670	if (pVmcsInfoNstGst->u32XcptPFMask != u32XcptPFMask)
10671	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, u32XcptPFMask);
10672	if (pVmcsInfoNstGst->u32XcptPFMatch != u32XcptPFMatch)
10673	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, u32XcptPFMatch);
10674	if ( !(u32ProcCtls & VMX_PROC_CTLS_PAUSE_EXIT)
10675	&& (u32ProcCtls2 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT))
10676	{
10677	Assert(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT);
10678	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, cPleGapTicks);
10679	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, cPleWindowTicks);
10680	}
10681	if (u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
10682	{
10683	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
10684	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL, HCPhysVirtApic);
10685	}
10686	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10687	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, HCPhysApicAccess);
10688	rc \|= VMXWriteVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, uPendingDbgXcpts);
10689	AssertRC(rc);
10690
10691	/*
10692	* Update the nested-guest VMCS cache.
10693	*/
10694	pVmcsInfoNstGst->u32PinCtls = u32PinCtls;
10695	pVmcsInfoNstGst->u32ProcCtls = u32ProcCtls;
10696	pVmcsInfoNstGst->u32ProcCtls2 = u32ProcCtls2;
10697	pVmcsInfoNstGst->u32XcptBitmap = u32XcptBitmap;
10698	pVmcsInfoNstGst->u64Cr0Mask = u64Cr0Mask;
10699	pVmcsInfoNstGst->u64Cr4Mask = u64Cr4Mask;
10700	pVmcsInfoNstGst->u32XcptPFMask = u32XcptPFMask;
10701	pVmcsInfoNstGst->u32XcptPFMatch = u32XcptPFMatch;
10702	pVmcsInfoNstGst->HCPhysVirtApic = HCPhysVirtApic;
10703
10704	/*
10705	* We need to flush the TLB if we are switching the APIC-access page address.
10706	* See Intel spec. 28.3.3.4 "Guidelines for Use of the INVEPT Instruction".
10707	*/
10708	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10709	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = true;
10710
10711	/*
10712	* MSR bitmap.
10713	*
10714	* The MSR bitmap address has already been initialized while setting up the nested-guest
10715	* VMCS, here we need to merge the MSR bitmaps.
10716	*/
10717	if (u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
10718	hmR0VmxMergeMsrBitmapNested(pVCpu, pVmcsInfoNstGst, pVmcsInfoGst);
10719
10720	return VINF_SUCCESS;
10721	}
10722	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
10723
10724
10725	/**
10726	* Does the preparations before executing guest code in VT-x.
10727	*
10728	* This may cause longjmps to ring-3 and may even result in rescheduling to the
10729	* recompiler/IEM. We must be cautious what we do here regarding committing
10730	* guest-state information into the VMCS assuming we assuredly execute the
10731	* guest in VT-x mode.
10732	*
10733	* If we fall back to the recompiler/IEM after updating the VMCS and clearing
10734	* the common-state (TRPM/forceflags), we must undo those changes so that the
10735	* recompiler/IEM can (and should) use them when it resumes guest execution.
10736	* Otherwise such operations must be done when we can no longer exit to ring-3.
10737	*
10738	* @returns Strict VBox status code (i.e. informational status codes too).
10739	* @retval VINF_SUCCESS if we can proceed with running the guest, interrupts
10740	* have been disabled.
10741	* @retval VINF_VMX_VMEXIT if a nested-guest VM-exit occurs (e.g., while evaluating
10742	* pending events).
10743	* @retval VINF_EM_RESET if a triple-fault occurs while injecting a
10744	* double-fault into the guest.
10745	* @retval VINF_EM_DBG_STEPPED if @a fStepping is true and an event was
10746	* dispatched directly.
10747	* @retval VINF_* scheduling changes, we have to go back to ring-3.
10748	*
10749	* @param pVCpu The cross context virtual CPU structure.
10750	* @param pVmxTransient The VMX-transient structure.
10751	* @param fStepping Whether we are single-stepping the guest in the
10752	* hypervisor debugger. Makes us ignore some of the reasons
10753	* for returning to ring-3, and return VINF_EM_DBG_STEPPED
10754	* if event dispatching took place.
10755	*/
10756	static VBOXSTRICTRC hmR0VmxPreRunGuest(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, bool fStepping)
10757	{
10758	Assert(VMMRZCallRing3IsEnabled(pVCpu));
10759
10760	Log4Func(("fIsNested=%RTbool fStepping=%RTbool\n", pVmxTransient->fIsNestedGuest, fStepping));
10761
10762	#ifdef VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM
10763	if (pVmxTransient->fIsNestedGuest)
10764	{
10765	RT_NOREF2(pVCpu, fStepping);
10766	Log2Func(("Rescheduling to IEM due to nested-hwvirt or forced IEM exec -> VINF_EM_RESCHEDULE_REM\n"));
10767	return VINF_EM_RESCHEDULE_REM;
10768	}
10769	#endif
10770
10771	/*
10772	* Check and process force flag actions, some of which might require us to go back to ring-3.
10773	*/
10774	VBOXSTRICTRC rcStrict = hmR0VmxCheckForceFlags(pVCpu, pVmxTransient, fStepping);
10775	if (rcStrict == VINF_SUCCESS)
10776	{
10777	/* FFs don't get set all the time. */
10778	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10779	if ( pVmxTransient->fIsNestedGuest
10780	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
10781	{
10782	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
10783	return VINF_VMX_VMEXIT;
10784	}
10785	#endif
10786	}
10787	else
10788	return rcStrict;
10789
10790	/*
10791	* Virtualize memory-mapped accesses to the physical APIC (may take locks).
10792	*/
10793	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10794	if ( !pVCpu->hm.s.vmx.u64GstMsrApicBase
10795	&& (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10796	&& PDMHasApic(pVM))
10797	{
10798	int rc = hmR0VmxMapHCApicAccessPage(pVCpu);
10799	AssertRCReturn(rc, rc);
10800	}
10801
10802	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10803	/*
10804	* Merge guest VMCS controls with the nested-guest VMCS controls.
10805	*
10806	* Even if we have not executed the guest prior to this (e.g. when resuming from a
10807	* saved state), we should be okay with merging controls as we initialize the
10808	* guest VMCS controls as part of VM setup phase.
10809	*/
10810	if ( pVmxTransient->fIsNestedGuest
10811	&& !pVCpu->hm.s.vmx.fMergedNstGstCtls)
10812	{
10813	int rc = hmR0VmxMergeVmcsNested(pVCpu);
10814	AssertRCReturn(rc, rc);
10815	pVCpu->hm.s.vmx.fMergedNstGstCtls = true;
10816	}
10817	#endif
10818
10819	/*
10820	* Evaluate events to be injected into the guest.
10821	*
10822	* Events in TRPM can be injected without inspecting the guest state.
10823	* If any new events (interrupts/NMI) are pending currently, we try to set up the
10824	* guest to cause a VM-exit the next time they are ready to receive the event.
10825	*/
10826	if (TRPMHasTrap(pVCpu))
10827	hmR0VmxTrpmTrapToPendingEvent(pVCpu);
10828
10829	uint32_t fIntrState;
10830	rcStrict = hmR0VmxEvaluatePendingEvent(pVCpu, pVmxTransient, &fIntrState);
10831
10832	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10833	/*
10834	* While evaluating pending events if something failed (unlikely) or if we were
10835	* preparing to run a nested-guest but performed a nested-guest VM-exit, we should bail.
10836	*/
10837	if (rcStrict != VINF_SUCCESS)
10838	return rcStrict;
10839	if ( pVmxTransient->fIsNestedGuest
10840	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
10841	{
10842	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
10843	return VINF_VMX_VMEXIT;
10844	}
10845	#else
10846	Assert(rcStrict == VINF_SUCCESS);
10847	#endif
10848
10849	/*
10850	* Event injection may take locks (currently the PGM lock for real-on-v86 case) and thus
10851	* needs to be done with longjmps or interrupts + preemption enabled. Event injection might
10852	* also result in triple-faulting the VM.
10853	*
10854	* With nested-guests, the above does not apply since unrestricted guest execution is a
10855	* requirement. Regardless, we do this here to avoid duplicating code elsewhere.
10856	*/
10857	rcStrict = hmR0VmxInjectPendingEvent(pVCpu, pVmxTransient, fIntrState, fStepping);
10858	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
10859	{ /* likely */ }
10860	else
10861	{
10862	AssertMsg(rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
10863	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
10864	return rcStrict;
10865	}
10866
10867	/*
10868	* A longjump might result in importing CR3 even for VM-exits that don't necessarily
10869	* import CR3 themselves. We will need to update them here, as even as late as the above
10870	* hmR0VmxInjectPendingEvent() call may lazily import guest-CPU state on demand causing
10871	* the below force flags to be set.
10872	*/
10873	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
10874	{
10875	Assert(!(ASMAtomicUoReadU64(&pVCpu->cpum.GstCtx.fExtrn) & CPUMCTX_EXTRN_CR3));
10876	int rc2 = PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
10877	AssertMsgReturn(rc2 == VINF_SUCCESS \|\| rc2 == VINF_PGM_SYNC_CR3,
10878	("%Rrc\n", rc2), RT_FAILURE_NP(rc2) ? rc2 : VERR_IPE_UNEXPECTED_INFO_STATUS);
10879	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
10880	}
10881
10882	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10883	/* Paranoia. */
10884	Assert(!pVmxTransient->fIsNestedGuest \|\| CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
10885	#endif
10886
10887	/*
10888	* No longjmps to ring-3 from this point on!!!
10889	* Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
10890	* This also disables flushing of the R0-logger instance (if any).
10891	*/
10892	VMMRZCallRing3Disable(pVCpu);
10893
10894	/*
10895	* Export the guest state bits.
10896	*
10897	* We cannot perform longjmps while loading the guest state because we do not preserve the
10898	* host/guest state (although the VMCS will be preserved) across longjmps which can cause
10899	* CPU migration.
10900	*
10901	* If we are injecting events to a real-on-v86 mode guest, we would have updated RIP and some segment
10902	* registers. Hence, exporting of the guest state needs to be done -after- injection of events.
10903	*/
10904	rcStrict = hmR0VmxExportGuestStateOptimal(pVCpu, pVmxTransient);
10905	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
10906	{ /* likely */ }
10907	else
10908	{
10909	VMMRZCallRing3Enable(pVCpu);
10910	return rcStrict;
10911	}
10912
10913	/*
10914	* We disable interrupts so that we don't miss any interrupts that would flag preemption
10915	* (IPI/timers etc.) when thread-context hooks aren't used and we've been running with
10916	* preemption disabled for a while. Since this is purely to aid the
10917	* RTThreadPreemptIsPending() code, it doesn't matter that it may temporarily reenable and
10918	* disable interrupt on NT.
10919	*
10920	* We need to check for force-flags that could've possible been altered since we last
10921	* checked them (e.g. by PDMGetInterrupt() leaving the PDM critical section,
10922	* see @bugref{6398}).
10923	*
10924	* We also check a couple of other force-flags as a last opportunity to get the EMT back
10925	* to ring-3 before executing guest code.
10926	*/
10927	pVmxTransient->fEFlags = ASMIntDisableFlags();
10928
10929	if ( ( !VM_FF_IS_ANY_SET(pVM, VM_FF_EMT_RENDEZVOUS \| VM_FF_TM_VIRTUAL_SYNC)
10930	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
10931	\|\| ( fStepping /* Optimized for the non-stepping case, so a bit of unnecessary work when stepping. */
10932	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK & ~(VMCPU_FF_TIMER \| VMCPU_FF_PDM_CRITSECT))) )
10933	{
10934	if (!RTThreadPreemptIsPending(NIL_RTTHREAD))
10935	{
10936	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10937	/*
10938	* If we are executing a nested-guest make sure that we should intercept subsequent
10939	* events. The one we are injecting might be part of VM-entry. This is mainly to keep
10940	* the VM-exit instruction emulation happy.
10941	*/
10942	if (pVmxTransient->fIsNestedGuest)
10943	CPUMSetGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx, true);
10944	#endif
10945
10946	/*
10947	* We've injected any pending events. This is really the point of no return (to ring-3).
10948	*
10949	* Note! The caller expects to continue with interrupts & longjmps disabled on successful
10950	* returns from this function, so do -not- enable them here.
10951	*/
10952	pVCpu->hm.s.Event.fPending = false;
10953	return VINF_SUCCESS;
10954	}
10955
10956	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPendingHostIrq);
10957	rcStrict = VINF_EM_RAW_INTERRUPT;
10958	}
10959	else
10960	{
10961	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
10962	rcStrict = VINF_EM_RAW_TO_R3;
10963	}
10964
10965	ASMSetFlags(pVmxTransient->fEFlags);
10966	VMMRZCallRing3Enable(pVCpu);
10967
10968	return rcStrict;
10969	}
10970
10971
10972	/**
10973	* Final preparations before executing guest code using hardware-assisted VMX.
10974	*
10975	* We can no longer get preempted to a different host CPU and there are no returns
10976	* to ring-3. We ignore any errors that may happen from this point (e.g. VMWRITE
10977	* failures), this function is not intended to fail sans unrecoverable hardware
10978	* errors.
10979	*
10980	* @param pVCpu The cross context virtual CPU structure.
10981	* @param pVmxTransient The VMX-transient structure.
10982	*
10983	* @remarks Called with preemption disabled.
10984	* @remarks No-long-jump zone!!!
10985	*/
10986	static void hmR0VmxPreRunGuestCommitted(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
10987	{
10988	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
10989	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
10990	Assert(!pVCpu->hm.s.Event.fPending);
10991
10992	/*
10993	* Indicate start of guest execution and where poking EMT out of guest-context is recognized.
10994	*/
10995	VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
10996	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC);
10997
10998	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10999	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11000	PHMPHYSCPU pHostCpu = hmR0GetCurrentCpu();
11001	RTCPUID const idCurrentCpu = pHostCpu->idCpu;
11002
11003	if (!CPUMIsGuestFPUStateActive(pVCpu))
11004	{
11005	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestFpuState, x);
11006	if (CPUMR0LoadGuestFPU(pVM, pVCpu) == VINF_CPUM_HOST_CR0_MODIFIED)
11007	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_HOST_CONTEXT;
11008	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestFpuState, x);
11009	STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadGuestFpu);
11010	}
11011
11012	/*
11013	* Re-export the host state bits as we may've been preempted (only happens when
11014	* thread-context hooks are used or when the VM start function changes) or if
11015	* the host CR0 is modified while loading the guest FPU state above.
11016	*
11017	* The 64-on-32 switcher saves the (64-bit) host state into the VMCS and if we
11018	* changed the switcher back to 32-bit, we must save the 32-bit host state here,
11019	* see @bugref{8432}.
11020	*
11021	* This may also happen when switching to/from a nested-guest VMCS without leaving
11022	* ring-0.
11023	*/
11024	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT)
11025	{
11026	hmR0VmxExportHostState(pVCpu);
11027	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportHostState);
11028	}
11029	Assert(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT));
11030
11031	/*
11032	* Export the state shared between host and guest (FPU, debug, lazy MSRs).
11033	*/
11034	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE)
11035	hmR0VmxExportSharedState(pVCpu, pVmxTransient);
11036	AssertMsg(!pVCpu->hm.s.fCtxChanged, ("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
11037
11038	/*
11039	* Store status of the shared guest/host debug state at the time of VM-entry.
11040	*/
11041	pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
11042	pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
11043
11044	/*
11045	* Always cache the TPR-shadow if the virtual-APIC page exists, thereby skipping
11046	* more than one conditional check. The post-run side of our code shall determine
11047	* if it needs to sync. the virtual APIC TPR with the TPR-shadow.
11048	*/
11049	if (pVmcsInfo->pbVirtApic)
11050	pVmxTransient->u8GuestTpr = pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR];
11051
11052	/*
11053	* Update the host MSRs values in the VM-exit MSR-load area.
11054	*/
11055	if (!pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs)
11056	{
11057	if (pVmcsInfo->cExitMsrLoad > 0)
11058	hmR0VmxUpdateAutoLoadHostMsrs(pVCpu, pVmcsInfo);
11059	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = true;
11060	}
11061
11062	/*
11063	* Evaluate if we need to intercept guest RDTSC/P accesses. Set up the
11064	* VMX-preemption timer based on the next virtual sync clock deadline.
11065	*/
11066	if ( !pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer
11067	\|\| idCurrentCpu != pVCpu->hmr0.s.idLastCpu)
11068	{
11069	hmR0VmxUpdateTscOffsettingAndPreemptTimer(pVCpu, pVmxTransient, idCurrentCpu);
11070	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = true;
11071	}
11072
11073	/* Record statistics of how often we use TSC offsetting as opposed to intercepting RDTSC/P. */
11074	bool const fIsRdtscIntercepted = RT_BOOL(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_RDTSC_EXIT);
11075	if (!fIsRdtscIntercepted)
11076	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
11077	else
11078	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
11079
11080	ASMAtomicUoWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB flushing, set this across the world switch. */
11081	hmR0VmxFlushTaggedTlb(pHostCpu, pVCpu, pVmcsInfo); /* Invalidate the appropriate guest entries from the TLB. */
11082	Assert(idCurrentCpu == pVCpu->hmr0.s.idLastCpu);
11083	pVCpu->hm.s.vmx.LastError.idCurrentCpu = idCurrentCpu; /* Record the error reporting info. with the current host CPU. */
11084	pVmcsInfo->idHostCpuState = idCurrentCpu; /* Record the CPU for which the host-state has been exported. */
11085	pVmcsInfo->idHostCpuExec = idCurrentCpu; /* Record the CPU on which we shall execute. */
11086
11087	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
11088
11089	TMNotifyStartOfExecution(pVM, pVCpu); /* Notify TM to resume its clocks when TSC is tied to execution,
11090	as we're about to start executing the guest. */
11091
11092	/*
11093	* Load the guest TSC_AUX MSR when we are not intercepting RDTSCP.
11094	*
11095	* This is done this late as updating the TSC offsetting/preemption timer above
11096	* figures out if we can skip intercepting RDTSCP by calculating the number of
11097	* host CPU ticks till the next virtual sync deadline (for the dynamic case).
11098	*/
11099	if ( (pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_RDTSCP)
11100	&& !fIsRdtscIntercepted)
11101	{
11102	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_TSC_AUX);
11103
11104	/* NB: Because we call hmR0VmxAddAutoLoadStoreMsr with fUpdateHostMsr=true,
11105	it's safe even after hmR0VmxUpdateAutoLoadHostMsrs has already been done. */
11106	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_TSC_AUX, CPUMGetGuestTscAux(pVCpu),
11107	true /* fSetReadWrite /, true / fUpdateHostMsr */);
11108	AssertRC(rc);
11109	Assert(!pVmxTransient->fRemoveTscAuxMsr);
11110	pVmxTransient->fRemoveTscAuxMsr = true;
11111	}
11112
11113	#ifdef VBOX_STRICT
11114	Assert(pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs);
11115	hmR0VmxCheckAutoLoadStoreMsrs(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest);
11116	hmR0VmxCheckHostEferMsr(pVmcsInfo);
11117	AssertRC(hmR0VmxCheckCachedVmcsCtls(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest));
11118	#endif
11119
11120	#ifdef HMVMX_ALWAYS_CHECK_GUEST_STATE
11121	/** @todo r=ramshankar: We can now probably use iemVmxVmentryCheckGuestState here.
11122	* Add a PVMXMSRS parameter to it, so that IEM can look at the host MSRs,
11123	* see @bugref{9180#c54}. */
11124	uint32_t const uInvalidReason = hmR0VmxCheckGuestState(pVCpu, pVmcsInfo);
11125	if (uInvalidReason != VMX_IGS_REASON_NOT_FOUND)
11126	Log4(("hmR0VmxCheckGuestState returned %#x\n", uInvalidReason));
11127	#endif
11128	}
11129
11130
11131	/**
11132	* First C routine invoked after running guest code using hardware-assisted VMX.
11133	*
11134	* @param pVCpu The cross context virtual CPU structure.
11135	* @param pVmxTransient The VMX-transient structure.
11136	* @param rcVMRun Return code of VMLAUNCH/VMRESUME.
11137	*
11138	* @remarks Called with interrupts disabled, and returns with interrupts enabled!
11139	*
11140	* @remarks No-long-jump zone!!! This function will however re-enable longjmps
11141	* unconditionally when it is safe to do so.
11142	*/
11143	static void hmR0VmxPostRunGuest(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, int rcVMRun)
11144	{
11145	ASMAtomicUoWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB flushing. */
11146	ASMAtomicIncU32(&pVCpu->hmr0.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for EMT poking. */
11147	pVCpu->hm.s.fCtxChanged = 0; /* Exits/longjmps to ring-3 requires saving the guest state. */
11148	pVmxTransient->fVmcsFieldsRead = 0; /* Transient fields need to be read from the VMCS. */
11149	pVmxTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
11150	pVmxTransient->fVectoringDoublePF = false; /* Vectoring double page-fault needs to be determined later. */
11151
11152	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11153	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_RDTSC_EXIT))
11154	{
11155	uint64_t uGstTsc;
11156	if (!pVmxTransient->fIsNestedGuest)
11157	uGstTsc = pVCpu->hmr0.s.uTscExit + pVmcsInfo->u64TscOffset;
11158	else
11159	{
11160	uint64_t const uNstGstTsc = pVCpu->hmr0.s.uTscExit + pVmcsInfo->u64TscOffset;
11161	uGstTsc = CPUMRemoveNestedGuestTscOffset(pVCpu, uNstGstTsc);
11162	}
11163	TMCpuTickSetLastSeen(pVCpu, uGstTsc); /* Update TM with the guest TSC. */
11164	}
11165
11166	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatPreExit, x);
11167	TMNotifyEndOfExecution(pVCpu->CTX_SUFF(pVM), pVCpu, pVCpu->hmr0.s.uTscExit); /* Notify TM that the guest is no longer running. */
11168	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
11169
11170	pVCpu->hmr0.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_REQUIRED; /* Some host state messed up by VMX needs restoring. */
11171	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
11172	#ifdef VBOX_STRICT
11173	hmR0VmxCheckHostEferMsr(pVmcsInfo); /* Verify that the host EFER MSR wasn't modified. */
11174	#endif
11175	Assert(!ASMIntAreEnabled());
11176	ASMSetFlags(pVmxTransient->fEFlags); /* Enable interrupts. */
11177	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
11178
11179	#ifdef HMVMX_ALWAYS_CLEAN_TRANSIENT
11180	/*
11181	* Clean all the VMCS fields in the transient structure before reading
11182	* anything from the VMCS.
11183	*/
11184	pVmxTransient->uExitReason = 0;
11185	pVmxTransient->uExitIntErrorCode = 0;
11186	pVmxTransient->uExitQual = 0;
11187	pVmxTransient->uGuestLinearAddr = 0;
11188	pVmxTransient->uExitIntInfo = 0;
11189	pVmxTransient->cbExitInstr = 0;
11190	pVmxTransient->ExitInstrInfo.u = 0;
11191	pVmxTransient->uEntryIntInfo = 0;
11192	pVmxTransient->uEntryXcptErrorCode = 0;
11193	pVmxTransient->cbEntryInstr = 0;
11194	pVmxTransient->uIdtVectoringInfo = 0;
11195	pVmxTransient->uIdtVectoringErrorCode = 0;
11196	#endif
11197
11198	/*
11199	* Save the basic VM-exit reason and check if the VM-entry failed.
11200	* See Intel spec. 24.9.1 "Basic VM-exit Information".
11201	*/
11202	uint32_t uExitReason;
11203	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
11204	AssertRC(rc);
11205	pVmxTransient->uExitReason = VMX_EXIT_REASON_BASIC(uExitReason);
11206	pVmxTransient->fVMEntryFailed = VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason);
11207
11208	/*
11209	* Log the VM-exit before logging anything else as otherwise it might be a
11210	* tad confusing what happens before and after the world-switch.
11211	*/
11212	HMVMX_LOG_EXIT(pVCpu, uExitReason);
11213
11214	/*
11215	* Remove the TSC_AUX MSR from the auto-load/store MSR area and reset any MSR
11216	* bitmap permissions, if it was added before VM-entry.
11217	*/
11218	if (pVmxTransient->fRemoveTscAuxMsr)
11219	{
11220	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_TSC_AUX);
11221	pVmxTransient->fRemoveTscAuxMsr = false;
11222	}
11223
11224	/*
11225	* Check if VMLAUNCH/VMRESUME succeeded.
11226	* If this failed, we cause a guru meditation and cease further execution.
11227	*
11228	* However, if we are executing a nested-guest we might fail if we use the
11229	* fast path rather than fully emulating VMLAUNCH/VMRESUME instruction in IEM.
11230	*/
11231	if (RT_LIKELY(rcVMRun == VINF_SUCCESS))
11232	{
11233	/*
11234	* Update the VM-exit history array here even if the VM-entry failed due to:
11235	* - Invalid guest state.
11236	* - MSR loading.
11237	* - Machine-check event.
11238	*
11239	* In any of the above cases we will still have a "valid" VM-exit reason
11240	* despite @a fVMEntryFailed being false.
11241	*
11242	* See Intel spec. 26.7 "VM-Entry failures during or after loading guest state".
11243	*
11244	* Note! We don't have CS or RIP at this point. Will probably address that later
11245	* by amending the history entry added here.
11246	*/
11247	EMHistoryAddExit(pVCpu, EMEXIT_MAKE_FT(EMEXIT_F_KIND_VMX, pVmxTransient->uExitReason & EMEXIT_F_TYPE_MASK),
11248	UINT64_MAX, pVCpu->hmr0.s.uTscExit);
11249
11250	if (RT_LIKELY(!pVmxTransient->fVMEntryFailed))
11251	{
11252	VMMRZCallRing3Enable(pVCpu);
11253	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
11254
11255	#ifdef HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
11256	hmR0VmxReadAllRoFieldsVmcs(pVmxTransient);
11257	#endif
11258
11259	/*
11260	* Always import the guest-interruptibility state as we need it while evaluating
11261	* injecting events on re-entry.
11262	*
11263	* We don't import CR0 (when unrestricted guest execution is unavailable) despite
11264	* checking for real-mode while exporting the state because all bits that cause
11265	* mode changes wrt CR0 are intercepted.
11266	*/
11267	uint64_t const fImportMask = CPUMCTX_EXTRN_INHIBIT_INT
11268	\| CPUMCTX_EXTRN_INHIBIT_NMI
11269	#if defined(HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE) \|\| defined(HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE)
11270	\| HMVMX_CPUMCTX_EXTRN_ALL
11271	#elif defined(HMVMX_ALWAYS_SAVE_GUEST_RFLAGS)
11272	\| CPUMCTX_EXTRN_RFLAGS
11273	#endif
11274	;
11275	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImportMask);
11276	AssertRC(rc);
11277
11278	/*
11279	* Sync the TPR shadow with our APIC state.
11280	*/
11281	if ( !pVmxTransient->fIsNestedGuest
11282	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW))
11283	{
11284	Assert(pVmcsInfo->pbVirtApic);
11285	if (pVmxTransient->u8GuestTpr != pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR])
11286	{
11287	rc = APICSetTpr(pVCpu, pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR]);
11288	AssertRC(rc);
11289	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
11290	}
11291	}
11292
11293	Assert(VMMRZCallRing3IsEnabled(pVCpu));
11294	Assert( pVmxTransient->fWasGuestDebugStateActive == false
11295	\|\| pVmxTransient->fWasHyperDebugStateActive == false);
11296	return;
11297	}
11298	}
11299	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11300	else if (pVmxTransient->fIsNestedGuest)
11301	AssertMsgFailed(("VMLAUNCH/VMRESUME failed but shouldn't happen when VMLAUNCH/VMRESUME was emulated in IEM!\n"));
11302	#endif
11303	else
11304	Log4Func(("VM-entry failure: rcVMRun=%Rrc fVMEntryFailed=%RTbool\n", rcVMRun, pVmxTransient->fVMEntryFailed));
11305
11306	VMMRZCallRing3Enable(pVCpu);
11307	}
11308
11309
11310	/**
11311	* Runs the guest code using hardware-assisted VMX the normal way.
11312	*
11313	* @returns VBox status code.
11314	* @param pVCpu The cross context virtual CPU structure.
11315	* @param pcLoops Pointer to the number of executed loops.
11316	*/
11317	static VBOXSTRICTRC hmR0VmxRunGuestCodeNormal(PVMCPUCC pVCpu, uint32_t *pcLoops)
11318	{
11319	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops;
11320	Assert(pcLoops);
11321	Assert(*pcLoops <= cMaxResumeLoops);
11322	Assert(!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
11323
11324	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11325	/*
11326	* Switch to the guest VMCS as we may have transitioned from executing the nested-guest
11327	* without leaving ring-0. Otherwise, if we came from ring-3 we would have loaded the
11328	* guest VMCS while entering the VMX ring-0 session.
11329	*/
11330	if (pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs)
11331	{
11332	int rc = hmR0VmxSwitchToGstOrNstGstVmcs(pVCpu, false /* fSwitchToNstGstVmcs */);
11333	if (RT_SUCCESS(rc))
11334	{ /* likely */ }
11335	else
11336	{
11337	LogRelFunc(("Failed to switch to the guest VMCS. rc=%Rrc\n", rc));
11338	return rc;
11339	}
11340	}
11341	#endif
11342
11343	VMXTRANSIENT VmxTransient;
11344	RT_ZERO(VmxTransient);
11345	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
11346
11347	/* Paranoia. */
11348	Assert(VmxTransient.pVmcsInfo == &pVCpu->hmr0.s.vmx.VmcsInfo);
11349
11350	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
11351	for (;;)
11352	{
11353	Assert(!HMR0SuspendPending());
11354	HMVMX_ASSERT_CPU_SAFE(pVCpu);
11355	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
11356
11357	/*
11358	* Preparatory work for running nested-guest code, this may force us to
11359	* return to ring-3.
11360	*
11361	* Warning! This bugger disables interrupts on VINF_SUCCESS!
11362	*/
11363	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, false /* fStepping */);
11364	if (rcStrict != VINF_SUCCESS)
11365	break;
11366
11367	/* Interrupts are disabled at this point! */
11368	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
11369	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
11370	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
11371	/* Interrupts are re-enabled at this point! */
11372
11373	/*
11374	* Check for errors with running the VM (VMLAUNCH/VMRESUME).
11375	*/
11376	if (RT_SUCCESS(rcRun))
11377	{ /* very likely */ }
11378	else
11379	{
11380	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
11381	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
11382	return rcRun;
11383	}
11384
11385	/*
11386	* Profile the VM-exit.
11387	*/
11388	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
11389	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
11390	STAM_COUNTER_INC(&pVCpu->hm.s.aStatExitReason[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
11391	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
11392	HMVMX_START_EXIT_DISPATCH_PROF();
11393
11394	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
11395
11396	/*
11397	* Handle the VM-exit.
11398	*/
11399	#ifdef HMVMX_USE_FUNCTION_TABLE
11400	rcStrict = g_aVMExitHandlers[VmxTransient.uExitReason].pfn(pVCpu, &VmxTransient);
11401	#else
11402	rcStrict = hmR0VmxHandleExit(pVCpu, &VmxTransient);
11403	#endif
11404	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
11405	if (rcStrict == VINF_SUCCESS)
11406	{
11407	if (++(*pcLoops) <= cMaxResumeLoops)
11408	continue;
11409	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
11410	rcStrict = VINF_EM_RAW_INTERRUPT;
11411	}
11412	break;
11413	}
11414
11415	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
11416	return rcStrict;
11417	}
11418
11419
11420	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11421	/**
11422	* Runs the nested-guest code using hardware-assisted VMX.
11423	*
11424	* @returns VBox status code.
11425	* @param pVCpu The cross context virtual CPU structure.
11426	* @param pcLoops Pointer to the number of executed loops.
11427	*
11428	* @sa hmR0VmxRunGuestCodeNormal.
11429	*/
11430	static VBOXSTRICTRC hmR0VmxRunGuestCodeNested(PVMCPUCC pVCpu, uint32_t *pcLoops)
11431	{
11432	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops;
11433	Assert(pcLoops);
11434	Assert(*pcLoops <= cMaxResumeLoops);
11435	Assert(CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
11436
11437	/*
11438	* Switch to the nested-guest VMCS as we may have transitioned from executing the
11439	* guest without leaving ring-0. Otherwise, if we came from ring-3 we would have
11440	* loaded the nested-guest VMCS while entering the VMX ring-0 session.
11441	*/
11442	if (!pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs)
11443	{
11444	int rc = hmR0VmxSwitchToGstOrNstGstVmcs(pVCpu, true /* fSwitchToNstGstVmcs */);
11445	if (RT_SUCCESS(rc))
11446	{ /* likely */ }
11447	else
11448	{
11449	LogRelFunc(("Failed to switch to the nested-guest VMCS. rc=%Rrc\n", rc));
11450	return rc;
11451	}
11452	}
11453
11454	VMXTRANSIENT VmxTransient;
11455	RT_ZERO(VmxTransient);
11456	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
11457	VmxTransient.fIsNestedGuest = true;
11458
11459	/* Paranoia. */
11460	Assert(VmxTransient.pVmcsInfo == &pVCpu->hmr0.s.vmx.VmcsInfoNstGst);
11461
11462	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
11463	for (;;)
11464	{
11465	Assert(!HMR0SuspendPending());
11466	HMVMX_ASSERT_CPU_SAFE(pVCpu);
11467	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
11468
11469	/*
11470	* Preparatory work for running guest code, this may force us to
11471	* return to ring-3.
11472	*
11473	* Warning! This bugger disables interrupts on VINF_SUCCESS!
11474	*/
11475	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, false /* fStepping */);
11476	if (rcStrict != VINF_SUCCESS)
11477	break;
11478
11479	/* Interrupts are disabled at this point! */
11480	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
11481	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
11482	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
11483	/* Interrupts are re-enabled at this point! */
11484
11485	/*
11486	* Check for errors with running the VM (VMLAUNCH/VMRESUME).
11487	*/
11488	if (RT_SUCCESS(rcRun))
11489	{ /* very likely */ }
11490	else
11491	{
11492	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
11493	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
11494	return rcRun;
11495	}
11496
11497	/*
11498	* Profile the VM-exit.
11499	*/
11500	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
11501	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
11502	STAM_COUNTER_INC(&pVCpu->hm.s.StatNestedExitAll);
11503	STAM_COUNTER_INC(&pVCpu->hm.s.aStatNestedExitReason[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
11504	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
11505	HMVMX_START_EXIT_DISPATCH_PROF();
11506
11507	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
11508
11509	/*
11510	* Handle the VM-exit.
11511	*/
11512	rcStrict = hmR0VmxHandleExitNested(pVCpu, &VmxTransient);
11513	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
11514	if (rcStrict == VINF_SUCCESS)
11515	{
11516	if (!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
11517	{
11518	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
11519	rcStrict = VINF_VMX_VMEXIT;
11520	}
11521	else
11522	{
11523	if (++(*pcLoops) <= cMaxResumeLoops)
11524	continue;
11525	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
11526	rcStrict = VINF_EM_RAW_INTERRUPT;
11527	}
11528	}
11529	else
11530	Assert(rcStrict != VINF_VMX_VMEXIT);
11531	break;
11532	}
11533
11534	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
11535	return rcStrict;
11536	}
11537	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
11538
11539
11540	/** @name Execution loop for single stepping, DBGF events and expensive Dtrace
11541	* probes.
11542	*
11543	* The following few functions and associated structure contains the bloat
11544	* necessary for providing detailed debug events and dtrace probes as well as
11545	* reliable host side single stepping. This works on the principle of
11546	* "subclassing" the normal execution loop and workers. We replace the loop
11547	* method completely and override selected helpers to add necessary adjustments
11548	* to their core operation.
11549	*
11550	* The goal is to keep the "parent" code lean and mean, so as not to sacrifice
11551	* any performance for debug and analysis features.
11552	*
11553	* @{
11554	*/
11555
11556	/**
11557	* Transient per-VCPU debug state of VMCS and related info. we save/restore in
11558	* the debug run loop.
11559	*/
11560	typedef struct VMXRUNDBGSTATE
11561	{
11562	/** The RIP we started executing at. This is for detecting that we stepped. */
11563	uint64_t uRipStart;
11564	/** The CS we started executing with. */
11565	uint16_t uCsStart;
11566
11567	/** Whether we've actually modified the 1st execution control field. */
11568	bool fModifiedProcCtls : 1;
11569	/** Whether we've actually modified the 2nd execution control field. */
11570	bool fModifiedProcCtls2 : 1;
11571	/** Whether we've actually modified the exception bitmap. */
11572	bool fModifiedXcptBitmap : 1;
11573
11574	/** We desire the modified the CR0 mask to be cleared. */
11575	bool fClearCr0Mask : 1;
11576	/** We desire the modified the CR4 mask to be cleared. */
11577	bool fClearCr4Mask : 1;
11578	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC. */
11579	uint32_t fCpe1Extra;
11580	/** Stuff we do not want in VMX_VMCS32_CTRL_PROC_EXEC. */
11581	uint32_t fCpe1Unwanted;
11582	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC2. */
11583	uint32_t fCpe2Extra;
11584	/** Extra stuff we need in VMX_VMCS32_CTRL_EXCEPTION_BITMAP. */
11585	uint32_t bmXcptExtra;
11586	/** The sequence number of the Dtrace provider settings the state was
11587	* configured against. */
11588	uint32_t uDtraceSettingsSeqNo;
11589	/** VM-exits to check (one bit per VM-exit). */
11590	uint32_t bmExitsToCheck[3];
11591
11592	/** The initial VMX_VMCS32_CTRL_PROC_EXEC value (helps with restore). */
11593	uint32_t fProcCtlsInitial;
11594	/** The initial VMX_VMCS32_CTRL_PROC_EXEC2 value (helps with restore). */
11595	uint32_t fProcCtls2Initial;
11596	/** The initial VMX_VMCS32_CTRL_EXCEPTION_BITMAP value (helps with restore). */
11597	uint32_t bmXcptInitial;
11598	} VMXRUNDBGSTATE;
11599	AssertCompileMemberSize(VMXRUNDBGSTATE, bmExitsToCheck, (VMX_EXIT_MAX + 1 + 31) / 32 * 4);
11600	typedef VMXRUNDBGSTATE *PVMXRUNDBGSTATE;
11601
11602
11603	/**
11604	* Initializes the VMXRUNDBGSTATE structure.
11605	*
11606	* @param pVCpu The cross context virtual CPU structure of the
11607	* calling EMT.
11608	* @param pVmxTransient The VMX-transient structure.
11609	* @param pDbgState The debug state to initialize.
11610	*/
11611	static void hmR0VmxRunDebugStateInit(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11612	{
11613	pDbgState->uRipStart = pVCpu->cpum.GstCtx.rip;
11614	pDbgState->uCsStart = pVCpu->cpum.GstCtx.cs.Sel;
11615
11616	pDbgState->fModifiedProcCtls = false;
11617	pDbgState->fModifiedProcCtls2 = false;
11618	pDbgState->fModifiedXcptBitmap = false;
11619	pDbgState->fClearCr0Mask = false;
11620	pDbgState->fClearCr4Mask = false;
11621	pDbgState->fCpe1Extra = 0;
11622	pDbgState->fCpe1Unwanted = 0;
11623	pDbgState->fCpe2Extra = 0;
11624	pDbgState->bmXcptExtra = 0;
11625	pDbgState->fProcCtlsInitial = pVmxTransient->pVmcsInfo->u32ProcCtls;
11626	pDbgState->fProcCtls2Initial = pVmxTransient->pVmcsInfo->u32ProcCtls2;
11627	pDbgState->bmXcptInitial = pVmxTransient->pVmcsInfo->u32XcptBitmap;
11628	}
11629
11630
11631	/**
11632	* Updates the VMSC fields with changes requested by @a pDbgState.
11633	*
11634	* This is performed after hmR0VmxPreRunGuestDebugStateUpdate as well
11635	* immediately before executing guest code, i.e. when interrupts are disabled.
11636	* We don't check status codes here as we cannot easily assert or return in the
11637	* latter case.
11638	*
11639	* @param pVCpu The cross context virtual CPU structure.
11640	* @param pVmxTransient The VMX-transient structure.
11641	* @param pDbgState The debug state.
11642	*/
11643	static void hmR0VmxPreRunGuestDebugStateApply(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11644	{
11645	/*
11646	* Ensure desired flags in VMCS control fields are set.
11647	* (Ignoring write failure here, as we're committed and it's just debug extras.)
11648	*
11649	* Note! We load the shadow CR0 & CR4 bits when we flag the clearing, so
11650	* there should be no stale data in pCtx at this point.
11651	*/
11652	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11653	if ( (pVmcsInfo->u32ProcCtls & pDbgState->fCpe1Extra) != pDbgState->fCpe1Extra
11654	\|\| (pVmcsInfo->u32ProcCtls & pDbgState->fCpe1Unwanted))
11655	{
11656	pVmcsInfo->u32ProcCtls \|= pDbgState->fCpe1Extra;
11657	pVmcsInfo->u32ProcCtls &= ~pDbgState->fCpe1Unwanted;
11658	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
11659	Log6Func(("VMX_VMCS32_CTRL_PROC_EXEC: %#RX32\n", pVmcsInfo->u32ProcCtls));
11660	pDbgState->fModifiedProcCtls = true;
11661	}
11662
11663	if ((pVmcsInfo->u32ProcCtls2 & pDbgState->fCpe2Extra) != pDbgState->fCpe2Extra)
11664	{
11665	pVmcsInfo->u32ProcCtls2 \|= pDbgState->fCpe2Extra;
11666	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pVmcsInfo->u32ProcCtls2);
11667	Log6Func(("VMX_VMCS32_CTRL_PROC_EXEC2: %#RX32\n", pVmcsInfo->u32ProcCtls2));
11668	pDbgState->fModifiedProcCtls2 = true;
11669	}
11670
11671	if ((pVmcsInfo->u32XcptBitmap & pDbgState->bmXcptExtra) != pDbgState->bmXcptExtra)
11672	{
11673	pVmcsInfo->u32XcptBitmap \|= pDbgState->bmXcptExtra;
11674	VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVmcsInfo->u32XcptBitmap);
11675	Log6Func(("VMX_VMCS32_CTRL_EXCEPTION_BITMAP: %#RX32\n", pVmcsInfo->u32XcptBitmap));
11676	pDbgState->fModifiedXcptBitmap = true;
11677	}
11678
11679	if (pDbgState->fClearCr0Mask && pVmcsInfo->u64Cr0Mask != 0)
11680	{
11681	pVmcsInfo->u64Cr0Mask = 0;
11682	VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, 0);
11683	Log6Func(("VMX_VMCS_CTRL_CR0_MASK: 0\n"));
11684	}
11685
11686	if (pDbgState->fClearCr4Mask && pVmcsInfo->u64Cr4Mask != 0)
11687	{
11688	pVmcsInfo->u64Cr4Mask = 0;
11689	VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, 0);
11690	Log6Func(("VMX_VMCS_CTRL_CR4_MASK: 0\n"));
11691	}
11692
11693	NOREF(pVCpu);
11694	}
11695
11696
11697	/**
11698	* Restores VMCS fields that were changed by hmR0VmxPreRunGuestDebugStateApply for
11699	* re-entry next time around.
11700	*
11701	* @returns Strict VBox status code (i.e. informational status codes too).
11702	* @param pVCpu The cross context virtual CPU structure.
11703	* @param pVmxTransient The VMX-transient structure.
11704	* @param pDbgState The debug state.
11705	* @param rcStrict The return code from executing the guest using single
11706	* stepping.
11707	*/
11708	static VBOXSTRICTRC hmR0VmxRunDebugStateRevert(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState,
11709	VBOXSTRICTRC rcStrict)
11710	{
11711	/*
11712	* Restore VM-exit control settings as we may not reenter this function the
11713	* next time around.
11714	*/
11715	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11716
11717	/* We reload the initial value, trigger what we can of recalculations the
11718	next time around. From the looks of things, that's all that's required atm. */
11719	if (pDbgState->fModifiedProcCtls)
11720	{
11721	if (!(pDbgState->fProcCtlsInitial & VMX_PROC_CTLS_MOV_DR_EXIT) && CPUMIsHyperDebugStateActive(pVCpu))
11722	pDbgState->fProcCtlsInitial \|= VMX_PROC_CTLS_MOV_DR_EXIT; /* Avoid assertion in hmR0VmxLeave */
11723	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pDbgState->fProcCtlsInitial);
11724	AssertRC(rc2);
11725	pVmcsInfo->u32ProcCtls = pDbgState->fProcCtlsInitial;
11726	}
11727
11728	/* We're currently the only ones messing with this one, so just restore the
11729	cached value and reload the field. */
11730	if ( pDbgState->fModifiedProcCtls2
11731	&& pVmcsInfo->u32ProcCtls2 != pDbgState->fProcCtls2Initial)
11732	{
11733	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pDbgState->fProcCtls2Initial);
11734	AssertRC(rc2);
11735	pVmcsInfo->u32ProcCtls2 = pDbgState->fProcCtls2Initial;
11736	}
11737
11738	/* If we've modified the exception bitmap, we restore it and trigger
11739	reloading and partial recalculation the next time around. */
11740	if (pDbgState->fModifiedXcptBitmap)
11741	pVmcsInfo->u32XcptBitmap = pDbgState->bmXcptInitial;
11742
11743	return rcStrict;
11744	}
11745
11746
11747	/**
11748	* Configures VM-exit controls for current DBGF and DTrace settings.
11749	*
11750	* This updates @a pDbgState and the VMCS execution control fields to reflect
11751	* the necessary VM-exits demanded by DBGF and DTrace.
11752	*
11753	* @param pVCpu The cross context virtual CPU structure.
11754	* @param pVmxTransient The VMX-transient structure. May update
11755	* fUpdatedTscOffsettingAndPreemptTimer.
11756	* @param pDbgState The debug state.
11757	*/
11758	static void hmR0VmxPreRunGuestDebugStateUpdate(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11759	{
11760	/*
11761	* Take down the dtrace serial number so we can spot changes.
11762	*/
11763	pDbgState->uDtraceSettingsSeqNo = VBOXVMM_GET_SETTINGS_SEQ_NO();
11764	ASMCompilerBarrier();
11765
11766	/*
11767	* We'll rebuild most of the middle block of data members (holding the
11768	* current settings) as we go along here, so start by clearing it all.
11769	*/
11770	pDbgState->bmXcptExtra = 0;
11771	pDbgState->fCpe1Extra = 0;
11772	pDbgState->fCpe1Unwanted = 0;
11773	pDbgState->fCpe2Extra = 0;
11774	for (unsigned i = 0; i < RT_ELEMENTS(pDbgState->bmExitsToCheck); i++)
11775	pDbgState->bmExitsToCheck[i] = 0;
11776
11777	/*
11778	* Software interrupts (INT XXh) - no idea how to trigger these...
11779	*/
11780	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
11781	if ( DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_INTERRUPT_SOFTWARE)
11782	\|\| VBOXVMM_INT_SOFTWARE_ENABLED())
11783	{
11784	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
11785	}
11786
11787	/*
11788	* INT3 breakpoints - triggered by #BP exceptions.
11789	*/
11790	if (pVM->dbgf.ro.cEnabledInt3Breakpoints > 0)
11791	pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
11792
11793	/*
11794	* Exception bitmap and XCPT events+probes.
11795	*/
11796	for (int iXcpt = 0; iXcpt < (DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST + 1); iXcpt++)
11797	if (DBGF_IS_EVENT_ENABLED(pVM, (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + iXcpt)))
11798	pDbgState->bmXcptExtra \|= RT_BIT_32(iXcpt);
11799
11800	if (VBOXVMM_XCPT_DE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DE);
11801	if (VBOXVMM_XCPT_DB_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DB);
11802	if (VBOXVMM_XCPT_BP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
11803	if (VBOXVMM_XCPT_OF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_OF);
11804	if (VBOXVMM_XCPT_BR_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BR);
11805	if (VBOXVMM_XCPT_UD_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_UD);
11806	if (VBOXVMM_XCPT_NM_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NM);
11807	if (VBOXVMM_XCPT_DF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DF);
11808	if (VBOXVMM_XCPT_TS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_TS);
11809	if (VBOXVMM_XCPT_NP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NP);
11810	if (VBOXVMM_XCPT_SS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SS);
11811	if (VBOXVMM_XCPT_GP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_GP);
11812	if (VBOXVMM_XCPT_PF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_PF);
11813	if (VBOXVMM_XCPT_MF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_MF);
11814	if (VBOXVMM_XCPT_AC_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_AC);
11815	if (VBOXVMM_XCPT_XF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_XF);
11816	if (VBOXVMM_XCPT_VE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_VE);
11817	if (VBOXVMM_XCPT_SX_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SX);
11818
11819	if (pDbgState->bmXcptExtra)
11820	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
11821
11822	/*
11823	* Process events and probes for VM-exits, making sure we get the wanted VM-exits.
11824	*
11825	* Note! This is the reverse of what hmR0VmxHandleExitDtraceEvents does.
11826	* So, when adding/changing/removing please don't forget to update it.
11827	*
11828	* Some of the macros are picking up local variables to save horizontal space,
11829	* (being able to see it in a table is the lesser evil here).
11830	*/
11831	#define IS_EITHER_ENABLED(a_pVM, a_EventSubName) \
11832	( DBGF_IS_EVENT_ENABLED(a_pVM, RT_CONCAT(DBGFEVENT_, a_EventSubName)) \
11833	\|\| RT_CONCAT3(VBOXVMM_, a_EventSubName, _ENABLED)() )
11834	#define SET_ONLY_XBM_IF_EITHER_EN(a_EventSubName, a_uExit) \
11835	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11836	{ AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11837	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11838	} else do { } while (0)
11839	#define SET_CPE1_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec) \
11840	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11841	{ \
11842	(pDbgState)->fCpe1Extra \|= (a_fCtrlProcExec); \
11843	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11844	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11845	} else do { } while (0)
11846	#define SET_CPEU_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fUnwantedCtrlProcExec) \
11847	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11848	{ \
11849	(pDbgState)->fCpe1Unwanted \|= (a_fUnwantedCtrlProcExec); \
11850	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11851	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11852	} else do { } while (0)
11853	#define SET_CPE2_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec2) \
11854	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11855	{ \
11856	(pDbgState)->fCpe2Extra \|= (a_fCtrlProcExec2); \
11857	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11858	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11859	} else do { } while (0)
11860
11861	SET_ONLY_XBM_IF_EITHER_EN(EXIT_TASK_SWITCH, VMX_EXIT_TASK_SWITCH); /* unconditional */
11862	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_VIOLATION, VMX_EXIT_EPT_VIOLATION); /* unconditional */
11863	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_MISCONFIG, VMX_EXIT_EPT_MISCONFIG); /* unconditional (unless #VE) */
11864	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_ACCESS, VMX_EXIT_APIC_ACCESS); /* feature dependent, nothing to enable here */
11865	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_WRITE, VMX_EXIT_APIC_WRITE); /* feature dependent, nothing to enable here */
11866
11867	SET_ONLY_XBM_IF_EITHER_EN(INSTR_CPUID, VMX_EXIT_CPUID); /* unconditional */
11868	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CPUID, VMX_EXIT_CPUID);
11869	SET_ONLY_XBM_IF_EITHER_EN(INSTR_GETSEC, VMX_EXIT_GETSEC); /* unconditional */
11870	SET_ONLY_XBM_IF_EITHER_EN( EXIT_GETSEC, VMX_EXIT_GETSEC);
11871	SET_CPE1_XBM_IF_EITHER_EN(INSTR_HALT, VMX_EXIT_HLT, VMX_PROC_CTLS_HLT_EXIT); /* paranoia */
11872	SET_ONLY_XBM_IF_EITHER_EN( EXIT_HALT, VMX_EXIT_HLT);
11873	SET_ONLY_XBM_IF_EITHER_EN(INSTR_INVD, VMX_EXIT_INVD); /* unconditional */
11874	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVD, VMX_EXIT_INVD);
11875	SET_CPE1_XBM_IF_EITHER_EN(INSTR_INVLPG, VMX_EXIT_INVLPG, VMX_PROC_CTLS_INVLPG_EXIT);
11876	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVLPG, VMX_EXIT_INVLPG);
11877	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDPMC, VMX_EXIT_RDPMC, VMX_PROC_CTLS_RDPMC_EXIT);
11878	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDPMC, VMX_EXIT_RDPMC);
11879	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSC, VMX_EXIT_RDTSC, VMX_PROC_CTLS_RDTSC_EXIT);
11880	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSC, VMX_EXIT_RDTSC);
11881	SET_ONLY_XBM_IF_EITHER_EN(INSTR_RSM, VMX_EXIT_RSM); /* unconditional */
11882	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RSM, VMX_EXIT_RSM);
11883	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMM_CALL, VMX_EXIT_VMCALL); /* unconditional */
11884	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMM_CALL, VMX_EXIT_VMCALL);
11885	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMCLEAR, VMX_EXIT_VMCLEAR); /* unconditional */
11886	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMCLEAR, VMX_EXIT_VMCLEAR);
11887	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH); /* unconditional */
11888	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH);
11889	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRLD, VMX_EXIT_VMPTRLD); /* unconditional */
11890	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRLD, VMX_EXIT_VMPTRLD);
11891	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRST, VMX_EXIT_VMPTRST); /* unconditional */
11892	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRST, VMX_EXIT_VMPTRST);
11893	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMREAD, VMX_EXIT_VMREAD); /* unconditional */
11894	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMREAD, VMX_EXIT_VMREAD);
11895	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMRESUME, VMX_EXIT_VMRESUME); /* unconditional */
11896	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMRESUME, VMX_EXIT_VMRESUME);
11897	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMWRITE, VMX_EXIT_VMWRITE); /* unconditional */
11898	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMWRITE, VMX_EXIT_VMWRITE);
11899	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXOFF, VMX_EXIT_VMXOFF); /* unconditional */
11900	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXOFF, VMX_EXIT_VMXOFF);
11901	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXON, VMX_EXIT_VMXON); /* unconditional */
11902	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXON, VMX_EXIT_VMXON);
11903
11904	if ( IS_EITHER_ENABLED(pVM, INSTR_CRX_READ)
11905	\|\| IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
11906	{
11907	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_CR4
11908	\| CPUMCTX_EXTRN_APIC_TPR);
11909	AssertRC(rc);
11910
11911	#if 0 /** @todo fix me */
11912	pDbgState->fClearCr0Mask = true;
11913	pDbgState->fClearCr4Mask = true;
11914	#endif
11915	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_READ))
11916	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_CR3_STORE_EXIT \| VMX_PROC_CTLS_CR8_STORE_EXIT;
11917	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
11918	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_CR3_LOAD_EXIT \| VMX_PROC_CTLS_CR8_LOAD_EXIT;
11919	pDbgState->fCpe1Unwanted \|= VMX_PROC_CTLS_USE_TPR_SHADOW; /* risky? */
11920	/* Note! We currently don't use VMX_VMCS32_CTRL_CR3_TARGET_COUNT. It would
11921	require clearing here and in the loop if we start using it. */
11922	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_CRX);
11923	}
11924	else
11925	{
11926	if (pDbgState->fClearCr0Mask)
11927	{
11928	pDbgState->fClearCr0Mask = false;
11929	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR0);
11930	}
11931	if (pDbgState->fClearCr4Mask)
11932	{
11933	pDbgState->fClearCr4Mask = false;
11934	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR4);
11935	}
11936	}
11937	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_READ, VMX_EXIT_MOV_CRX);
11938	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_WRITE, VMX_EXIT_MOV_CRX);
11939
11940	if ( IS_EITHER_ENABLED(pVM, INSTR_DRX_READ)
11941	\|\| IS_EITHER_ENABLED(pVM, INSTR_DRX_WRITE))
11942	{
11943	/** @todo later, need to fix handler as it assumes this won't usually happen. */
11944	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_DRX);
11945	}
11946	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_READ, VMX_EXIT_MOV_DRX);
11947	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_WRITE, VMX_EXIT_MOV_DRX);
11948
11949	SET_CPEU_XBM_IF_EITHER_EN(INSTR_RDMSR, VMX_EXIT_RDMSR, VMX_PROC_CTLS_USE_MSR_BITMAPS); /* risky clearing this? */
11950	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDMSR, VMX_EXIT_RDMSR);
11951	SET_CPEU_XBM_IF_EITHER_EN(INSTR_WRMSR, VMX_EXIT_WRMSR, VMX_PROC_CTLS_USE_MSR_BITMAPS);
11952	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WRMSR, VMX_EXIT_WRMSR);
11953	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MWAIT, VMX_EXIT_MWAIT, VMX_PROC_CTLS_MWAIT_EXIT); /* paranoia */
11954	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MWAIT, VMX_EXIT_MWAIT);
11955	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MONITOR, VMX_EXIT_MONITOR, VMX_PROC_CTLS_MONITOR_EXIT); /* paranoia */
11956	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MONITOR, VMX_EXIT_MONITOR);
11957	#if 0 /** @todo too slow, fix handler. */
11958	SET_CPE1_XBM_IF_EITHER_EN(INSTR_PAUSE, VMX_EXIT_PAUSE, VMX_PROC_CTLS_PAUSE_EXIT);
11959	#endif
11960	SET_ONLY_XBM_IF_EITHER_EN( EXIT_PAUSE, VMX_EXIT_PAUSE);
11961
11962	if ( IS_EITHER_ENABLED(pVM, INSTR_SGDT)
11963	\|\| IS_EITHER_ENABLED(pVM, INSTR_SIDT)
11964	\|\| IS_EITHER_ENABLED(pVM, INSTR_LGDT)
11965	\|\| IS_EITHER_ENABLED(pVM, INSTR_LIDT))
11966	{
11967	pDbgState->fCpe2Extra \|= VMX_PROC_CTLS2_DESC_TABLE_EXIT;
11968	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_GDTR_IDTR_ACCESS);
11969	}
11970	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SGDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11971	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SIDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11972	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LGDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11973	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LIDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11974
11975	if ( IS_EITHER_ENABLED(pVM, INSTR_SLDT)
11976	\|\| IS_EITHER_ENABLED(pVM, INSTR_STR)
11977	\|\| IS_EITHER_ENABLED(pVM, INSTR_LLDT)
11978	\|\| IS_EITHER_ENABLED(pVM, INSTR_LTR))
11979	{
11980	pDbgState->fCpe2Extra \|= VMX_PROC_CTLS2_DESC_TABLE_EXIT;
11981	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_LDTR_TR_ACCESS);
11982	}
11983	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SLDT, VMX_EXIT_LDTR_TR_ACCESS);
11984	SET_ONLY_XBM_IF_EITHER_EN( EXIT_STR, VMX_EXIT_LDTR_TR_ACCESS);
11985	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LLDT, VMX_EXIT_LDTR_TR_ACCESS);
11986	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LTR, VMX_EXIT_LDTR_TR_ACCESS);
11987
11988	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVEPT, VMX_EXIT_INVEPT); /* unconditional */
11989	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVEPT, VMX_EXIT_INVEPT);
11990	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSCP, VMX_EXIT_RDTSCP, VMX_PROC_CTLS_RDTSC_EXIT);
11991	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSCP, VMX_EXIT_RDTSCP);
11992	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVVPID, VMX_EXIT_INVVPID); /* unconditional */
11993	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVVPID, VMX_EXIT_INVVPID);
11994	SET_CPE2_XBM_IF_EITHER_EN(INSTR_WBINVD, VMX_EXIT_WBINVD, VMX_PROC_CTLS2_WBINVD_EXIT);
11995	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WBINVD, VMX_EXIT_WBINVD);
11996	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSETBV, VMX_EXIT_XSETBV); /* unconditional */
11997	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XSETBV, VMX_EXIT_XSETBV);
11998	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDRAND, VMX_EXIT_RDRAND, VMX_PROC_CTLS2_RDRAND_EXIT);
11999	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDRAND, VMX_EXIT_RDRAND);
12000	SET_CPE1_XBM_IF_EITHER_EN(INSTR_VMX_INVPCID, VMX_EXIT_INVPCID, VMX_PROC_CTLS_INVLPG_EXIT);
12001	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVPCID, VMX_EXIT_INVPCID);
12002	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMFUNC, VMX_EXIT_VMFUNC); /* unconditional for the current setup */
12003	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMFUNC, VMX_EXIT_VMFUNC);
12004	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDSEED, VMX_EXIT_RDSEED, VMX_PROC_CTLS2_RDSEED_EXIT);
12005	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDSEED, VMX_EXIT_RDSEED);
12006	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSAVES, VMX_EXIT_XSAVES); /* unconditional (enabled by host, guest cfg) */
12007	SET_ONLY_XBM_IF_EITHER_EN(EXIT_XSAVES, VMX_EXIT_XSAVES);
12008	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XRSTORS, VMX_EXIT_XRSTORS); /* unconditional (enabled by host, guest cfg) */
12009	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XRSTORS, VMX_EXIT_XRSTORS);
12010
12011	#undef IS_EITHER_ENABLED
12012	#undef SET_ONLY_XBM_IF_EITHER_EN
12013	#undef SET_CPE1_XBM_IF_EITHER_EN
12014	#undef SET_CPEU_XBM_IF_EITHER_EN
12015	#undef SET_CPE2_XBM_IF_EITHER_EN
12016
12017	/*
12018	* Sanitize the control stuff.
12019	*/
12020	pDbgState->fCpe2Extra &= g_HmMsrs.u.vmx.ProcCtls2.n.allowed1;
12021	if (pDbgState->fCpe2Extra)
12022	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_USE_SECONDARY_CTLS;
12023	pDbgState->fCpe1Extra &= g_HmMsrs.u.vmx.ProcCtls.n.allowed1;
12024	pDbgState->fCpe1Unwanted &= ~g_HmMsrs.u.vmx.ProcCtls.n.allowed0;
12025	if (pVCpu->hmr0.s.fDebugWantRdTscExit != RT_BOOL(pDbgState->fCpe1Extra & VMX_PROC_CTLS_RDTSC_EXIT))
12026	{
12027	pVCpu->hmr0.s.fDebugWantRdTscExit ^= true;
12028	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
12029	}
12030
12031	Log6(("HM: debug state: cpe1=%#RX32 cpeu=%#RX32 cpe2=%#RX32%s%s\n",
12032	pDbgState->fCpe1Extra, pDbgState->fCpe1Unwanted, pDbgState->fCpe2Extra,
12033	pDbgState->fClearCr0Mask ? " clr-cr0" : "",
12034	pDbgState->fClearCr4Mask ? " clr-cr4" : ""));
12035	}
12036
12037
12038	/**
12039	* Fires off DBGF events and dtrace probes for a VM-exit, when it's
12040	* appropriate.
12041	*
12042	* The caller has checked the VM-exit against the
12043	* VMXRUNDBGSTATE::bmExitsToCheck bitmap. The caller has checked for NMIs
12044	* already, so we don't have to do that either.
12045	*
12046	* @returns Strict VBox status code (i.e. informational status codes too).
12047	* @param pVCpu The cross context virtual CPU structure.
12048	* @param pVmxTransient The VMX-transient structure.
12049	* @param uExitReason The VM-exit reason.
12050	*
12051	* @remarks The name of this function is displayed by dtrace, so keep it short
12052	* and to the point. No longer than 33 chars long, please.
12053	*/
12054	static VBOXSTRICTRC hmR0VmxHandleExitDtraceEvents(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uExitReason)
12055	{
12056	/*
12057	* Translate the event into a DBGF event (enmEvent + uEventArg) and at the
12058	* same time check whether any corresponding Dtrace event is enabled (fDtrace).
12059	*
12060	* Note! This is the reverse operation of what hmR0VmxPreRunGuestDebugStateUpdate
12061	* does. Must add/change/remove both places. Same ordering, please.
12062	*
12063	* Added/removed events must also be reflected in the next section
12064	* where we dispatch dtrace events.
12065	*/
12066	bool fDtrace1 = false;
12067	bool fDtrace2 = false;
12068	DBGFEVENTTYPE enmEvent1 = DBGFEVENT_END;
12069	DBGFEVENTTYPE enmEvent2 = DBGFEVENT_END;
12070	uint32_t uEventArg = 0;
12071	#define SET_EXIT(a_EventSubName) \
12072	do { \
12073	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
12074	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
12075	} while (0)
12076	#define SET_BOTH(a_EventSubName) \
12077	do { \
12078	enmEvent1 = RT_CONCAT(DBGFEVENT_INSTR_, a_EventSubName); \
12079	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
12080	fDtrace1 = RT_CONCAT3(VBOXVMM_INSTR_, a_EventSubName, _ENABLED)(); \
12081	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
12082	} while (0)
12083	switch (uExitReason)
12084	{
12085	case VMX_EXIT_MTF:
12086	return hmR0VmxExitMtf(pVCpu, pVmxTransient);
12087
12088	case VMX_EXIT_XCPT_OR_NMI:
12089	{
12090	uint8_t const idxVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
12091	switch (VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo))
12092	{
12093	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
12094	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
12095	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
12096	if (idxVector <= (unsigned)(DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST))
12097	{
12098	if (VMX_EXIT_INT_INFO_IS_ERROR_CODE_VALID(pVmxTransient->uExitIntInfo))
12099	{
12100	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
12101	uEventArg = pVmxTransient->uExitIntErrorCode;
12102	}
12103	enmEvent1 = (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + idxVector);
12104	switch (enmEvent1)
12105	{
12106	case DBGFEVENT_XCPT_DE: fDtrace1 = VBOXVMM_XCPT_DE_ENABLED(); break;
12107	case DBGFEVENT_XCPT_DB: fDtrace1 = VBOXVMM_XCPT_DB_ENABLED(); break;
12108	case DBGFEVENT_XCPT_BP: fDtrace1 = VBOXVMM_XCPT_BP_ENABLED(); break;
12109	case DBGFEVENT_XCPT_OF: fDtrace1 = VBOXVMM_XCPT_OF_ENABLED(); break;
12110	case DBGFEVENT_XCPT_BR: fDtrace1 = VBOXVMM_XCPT_BR_ENABLED(); break;
12111	case DBGFEVENT_XCPT_UD: fDtrace1 = VBOXVMM_XCPT_UD_ENABLED(); break;
12112	case DBGFEVENT_XCPT_NM: fDtrace1 = VBOXVMM_XCPT_NM_ENABLED(); break;
12113	case DBGFEVENT_XCPT_DF: fDtrace1 = VBOXVMM_XCPT_DF_ENABLED(); break;
12114	case DBGFEVENT_XCPT_TS: fDtrace1 = VBOXVMM_XCPT_TS_ENABLED(); break;
12115	case DBGFEVENT_XCPT_NP: fDtrace1 = VBOXVMM_XCPT_NP_ENABLED(); break;
12116	case DBGFEVENT_XCPT_SS: fDtrace1 = VBOXVMM_XCPT_SS_ENABLED(); break;
12117	case DBGFEVENT_XCPT_GP: fDtrace1 = VBOXVMM_XCPT_GP_ENABLED(); break;
12118	case DBGFEVENT_XCPT_PF: fDtrace1 = VBOXVMM_XCPT_PF_ENABLED(); break;
12119	case DBGFEVENT_XCPT_MF: fDtrace1 = VBOXVMM_XCPT_MF_ENABLED(); break;
12120	case DBGFEVENT_XCPT_AC: fDtrace1 = VBOXVMM_XCPT_AC_ENABLED(); break;
12121	case DBGFEVENT_XCPT_XF: fDtrace1 = VBOXVMM_XCPT_XF_ENABLED(); break;
12122	case DBGFEVENT_XCPT_VE: fDtrace1 = VBOXVMM_XCPT_VE_ENABLED(); break;
12123	case DBGFEVENT_XCPT_SX: fDtrace1 = VBOXVMM_XCPT_SX_ENABLED(); break;
12124	default: break;
12125	}
12126	}
12127	else
12128	AssertFailed();
12129	break;
12130
12131	case VMX_EXIT_INT_INFO_TYPE_SW_INT:
12132	uEventArg = idxVector;
12133	enmEvent1 = DBGFEVENT_INTERRUPT_SOFTWARE;
12134	fDtrace1 = VBOXVMM_INT_SOFTWARE_ENABLED();
12135	break;
12136	}
12137	break;
12138	}
12139
12140	case VMX_EXIT_TRIPLE_FAULT:
12141	enmEvent1 = DBGFEVENT_TRIPLE_FAULT;
12142	//fDtrace1 = VBOXVMM_EXIT_TRIPLE_FAULT_ENABLED();
12143	break;
12144	case VMX_EXIT_TASK_SWITCH: SET_EXIT(TASK_SWITCH); break;
12145	case VMX_EXIT_EPT_VIOLATION: SET_EXIT(VMX_EPT_VIOLATION); break;
12146	case VMX_EXIT_EPT_MISCONFIG: SET_EXIT(VMX_EPT_MISCONFIG); break;
12147	case VMX_EXIT_APIC_ACCESS: SET_EXIT(VMX_VAPIC_ACCESS); break;
12148	case VMX_EXIT_APIC_WRITE: SET_EXIT(VMX_VAPIC_WRITE); break;
12149
12150	/* Instruction specific VM-exits: */
12151	case VMX_EXIT_CPUID: SET_BOTH(CPUID); break;
12152	case VMX_EXIT_GETSEC: SET_BOTH(GETSEC); break;
12153	case VMX_EXIT_HLT: SET_BOTH(HALT); break;
12154	case VMX_EXIT_INVD: SET_BOTH(INVD); break;
12155	case VMX_EXIT_INVLPG: SET_BOTH(INVLPG); break;
12156	case VMX_EXIT_RDPMC: SET_BOTH(RDPMC); break;
12157	case VMX_EXIT_RDTSC: SET_BOTH(RDTSC); break;
12158	case VMX_EXIT_RSM: SET_BOTH(RSM); break;
12159	case VMX_EXIT_VMCALL: SET_BOTH(VMM_CALL); break;
12160	case VMX_EXIT_VMCLEAR: SET_BOTH(VMX_VMCLEAR); break;
12161	case VMX_EXIT_VMLAUNCH: SET_BOTH(VMX_VMLAUNCH); break;
12162	case VMX_EXIT_VMPTRLD: SET_BOTH(VMX_VMPTRLD); break;
12163	case VMX_EXIT_VMPTRST: SET_BOTH(VMX_VMPTRST); break;
12164	case VMX_EXIT_VMREAD: SET_BOTH(VMX_VMREAD); break;
12165	case VMX_EXIT_VMRESUME: SET_BOTH(VMX_VMRESUME); break;
12166	case VMX_EXIT_VMWRITE: SET_BOTH(VMX_VMWRITE); break;
12167	case VMX_EXIT_VMXOFF: SET_BOTH(VMX_VMXOFF); break;
12168	case VMX_EXIT_VMXON: SET_BOTH(VMX_VMXON); break;
12169	case VMX_EXIT_MOV_CRX:
12170	hmR0VmxReadExitQualVmcs(pVmxTransient);
12171	if (VMX_EXIT_QUAL_CRX_ACCESS(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_CRX_ACCESS_READ)
12172	SET_BOTH(CRX_READ);
12173	else
12174	SET_BOTH(CRX_WRITE);
12175	uEventArg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
12176	break;
12177	case VMX_EXIT_MOV_DRX:
12178	hmR0VmxReadExitQualVmcs(pVmxTransient);
12179	if ( VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual)
12180	== VMX_EXIT_QUAL_DRX_DIRECTION_READ)
12181	SET_BOTH(DRX_READ);
12182	else
12183	SET_BOTH(DRX_WRITE);
12184	uEventArg = VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual);
12185	break;
12186	case VMX_EXIT_RDMSR: SET_BOTH(RDMSR); break;
12187	case VMX_EXIT_WRMSR: SET_BOTH(WRMSR); break;
12188	case VMX_EXIT_MWAIT: SET_BOTH(MWAIT); break;
12189	case VMX_EXIT_MONITOR: SET_BOTH(MONITOR); break;
12190	case VMX_EXIT_PAUSE: SET_BOTH(PAUSE); break;
12191	case VMX_EXIT_GDTR_IDTR_ACCESS:
12192	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
12193	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_BF_XDTR_INSINFO_INSTR_ID))
12194	{
12195	case VMX_XDTR_INSINFO_II_SGDT: SET_BOTH(SGDT); break;
12196	case VMX_XDTR_INSINFO_II_SIDT: SET_BOTH(SIDT); break;
12197	case VMX_XDTR_INSINFO_II_LGDT: SET_BOTH(LGDT); break;
12198	case VMX_XDTR_INSINFO_II_LIDT: SET_BOTH(LIDT); break;
12199	}
12200	break;
12201
12202	case VMX_EXIT_LDTR_TR_ACCESS:
12203	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
12204	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_BF_YYTR_INSINFO_INSTR_ID))
12205	{
12206	case VMX_YYTR_INSINFO_II_SLDT: SET_BOTH(SLDT); break;
12207	case VMX_YYTR_INSINFO_II_STR: SET_BOTH(STR); break;
12208	case VMX_YYTR_INSINFO_II_LLDT: SET_BOTH(LLDT); break;
12209	case VMX_YYTR_INSINFO_II_LTR: SET_BOTH(LTR); break;
12210	}
12211	break;
12212
12213	case VMX_EXIT_INVEPT: SET_BOTH(VMX_INVEPT); break;
12214	case VMX_EXIT_RDTSCP: SET_BOTH(RDTSCP); break;
12215	case VMX_EXIT_INVVPID: SET_BOTH(VMX_INVVPID); break;
12216	case VMX_EXIT_WBINVD: SET_BOTH(WBINVD); break;
12217	case VMX_EXIT_XSETBV: SET_BOTH(XSETBV); break;
12218	case VMX_EXIT_RDRAND: SET_BOTH(RDRAND); break;
12219	case VMX_EXIT_INVPCID: SET_BOTH(VMX_INVPCID); break;
12220	case VMX_EXIT_VMFUNC: SET_BOTH(VMX_VMFUNC); break;
12221	case VMX_EXIT_RDSEED: SET_BOTH(RDSEED); break;
12222	case VMX_EXIT_XSAVES: SET_BOTH(XSAVES); break;
12223	case VMX_EXIT_XRSTORS: SET_BOTH(XRSTORS); break;
12224
12225	/* Events that aren't relevant at this point. */
12226	case VMX_EXIT_EXT_INT:
12227	case VMX_EXIT_INT_WINDOW:
12228	case VMX_EXIT_NMI_WINDOW:
12229	case VMX_EXIT_TPR_BELOW_THRESHOLD:
12230	case VMX_EXIT_PREEMPT_TIMER:
12231	case VMX_EXIT_IO_INSTR:
12232	break;
12233
12234	/* Errors and unexpected events. */
12235	case VMX_EXIT_INIT_SIGNAL:
12236	case VMX_EXIT_SIPI:
12237	case VMX_EXIT_IO_SMI:
12238	case VMX_EXIT_SMI:
12239	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
12240	case VMX_EXIT_ERR_MSR_LOAD:
12241	case VMX_EXIT_ERR_MACHINE_CHECK:
12242	case VMX_EXIT_PML_FULL:
12243	case VMX_EXIT_VIRTUALIZED_EOI:
12244	break;
12245
12246	default:
12247	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
12248	break;
12249	}
12250	#undef SET_BOTH
12251	#undef SET_EXIT
12252
12253	/*
12254	* Dtrace tracepoints go first. We do them here at once so we don't
12255	* have to copy the guest state saving and stuff a few dozen times.
12256	* Down side is that we've got to repeat the switch, though this time
12257	* we use enmEvent since the probes are a subset of what DBGF does.
12258	*/
12259	if (fDtrace1 \|\| fDtrace2)
12260	{
12261	hmR0VmxReadExitQualVmcs(pVmxTransient);
12262	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
12263	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
12264	switch (enmEvent1)
12265	{
12266	/** @todo consider which extra parameters would be helpful for each probe. */
12267	case DBGFEVENT_END: break;
12268	case DBGFEVENT_XCPT_DE: VBOXVMM_XCPT_DE(pVCpu, pCtx); break;
12269	case DBGFEVENT_XCPT_DB: VBOXVMM_XCPT_DB(pVCpu, pCtx, pCtx->dr[6]); break;
12270	case DBGFEVENT_XCPT_BP: VBOXVMM_XCPT_BP(pVCpu, pCtx); break;
12271	case DBGFEVENT_XCPT_OF: VBOXVMM_XCPT_OF(pVCpu, pCtx); break;
12272	case DBGFEVENT_XCPT_BR: VBOXVMM_XCPT_BR(pVCpu, pCtx); break;
12273	case DBGFEVENT_XCPT_UD: VBOXVMM_XCPT_UD(pVCpu, pCtx); break;
12274	case DBGFEVENT_XCPT_NM: VBOXVMM_XCPT_NM(pVCpu, pCtx); break;
12275	case DBGFEVENT_XCPT_DF: VBOXVMM_XCPT_DF(pVCpu, pCtx); break;
12276	case DBGFEVENT_XCPT_TS: VBOXVMM_XCPT_TS(pVCpu, pCtx, uEventArg); break;
12277	case DBGFEVENT_XCPT_NP: VBOXVMM_XCPT_NP(pVCpu, pCtx, uEventArg); break;
12278	case DBGFEVENT_XCPT_SS: VBOXVMM_XCPT_SS(pVCpu, pCtx, uEventArg); break;
12279	case DBGFEVENT_XCPT_GP: VBOXVMM_XCPT_GP(pVCpu, pCtx, uEventArg); break;
12280	case DBGFEVENT_XCPT_PF: VBOXVMM_XCPT_PF(pVCpu, pCtx, uEventArg, pCtx->cr2); break;
12281	case DBGFEVENT_XCPT_MF: VBOXVMM_XCPT_MF(pVCpu, pCtx); break;
12282	case DBGFEVENT_XCPT_AC: VBOXVMM_XCPT_AC(pVCpu, pCtx); break;
12283	case DBGFEVENT_XCPT_XF: VBOXVMM_XCPT_XF(pVCpu, pCtx); break;
12284	case DBGFEVENT_XCPT_VE: VBOXVMM_XCPT_VE(pVCpu, pCtx); break;
12285	case DBGFEVENT_XCPT_SX: VBOXVMM_XCPT_SX(pVCpu, pCtx, uEventArg); break;
12286	case DBGFEVENT_INTERRUPT_SOFTWARE: VBOXVMM_INT_SOFTWARE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12287	case DBGFEVENT_INSTR_CPUID: VBOXVMM_INSTR_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
12288	case DBGFEVENT_INSTR_GETSEC: VBOXVMM_INSTR_GETSEC(pVCpu, pCtx); break;
12289	case DBGFEVENT_INSTR_HALT: VBOXVMM_INSTR_HALT(pVCpu, pCtx); break;
12290	case DBGFEVENT_INSTR_INVD: VBOXVMM_INSTR_INVD(pVCpu, pCtx); break;
12291	case DBGFEVENT_INSTR_INVLPG: VBOXVMM_INSTR_INVLPG(pVCpu, pCtx); break;
12292	case DBGFEVENT_INSTR_RDPMC: VBOXVMM_INSTR_RDPMC(pVCpu, pCtx); break;
12293	case DBGFEVENT_INSTR_RDTSC: VBOXVMM_INSTR_RDTSC(pVCpu, pCtx); break;
12294	case DBGFEVENT_INSTR_RSM: VBOXVMM_INSTR_RSM(pVCpu, pCtx); break;
12295	case DBGFEVENT_INSTR_CRX_READ: VBOXVMM_INSTR_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12296	case DBGFEVENT_INSTR_CRX_WRITE: VBOXVMM_INSTR_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12297	case DBGFEVENT_INSTR_DRX_READ: VBOXVMM_INSTR_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12298	case DBGFEVENT_INSTR_DRX_WRITE: VBOXVMM_INSTR_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12299	case DBGFEVENT_INSTR_RDMSR: VBOXVMM_INSTR_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
12300	case DBGFEVENT_INSTR_WRMSR: VBOXVMM_INSTR_WRMSR(pVCpu, pCtx, pCtx->ecx,
12301	RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
12302	case DBGFEVENT_INSTR_MWAIT: VBOXVMM_INSTR_MWAIT(pVCpu, pCtx); break;
12303	case DBGFEVENT_INSTR_MONITOR: VBOXVMM_INSTR_MONITOR(pVCpu, pCtx); break;
12304	case DBGFEVENT_INSTR_PAUSE: VBOXVMM_INSTR_PAUSE(pVCpu, pCtx); break;
12305	case DBGFEVENT_INSTR_SGDT: VBOXVMM_INSTR_SGDT(pVCpu, pCtx); break;
12306	case DBGFEVENT_INSTR_SIDT: VBOXVMM_INSTR_SIDT(pVCpu, pCtx); break;
12307	case DBGFEVENT_INSTR_LGDT: VBOXVMM_INSTR_LGDT(pVCpu, pCtx); break;
12308	case DBGFEVENT_INSTR_LIDT: VBOXVMM_INSTR_LIDT(pVCpu, pCtx); break;
12309	case DBGFEVENT_INSTR_SLDT: VBOXVMM_INSTR_SLDT(pVCpu, pCtx); break;
12310	case DBGFEVENT_INSTR_STR: VBOXVMM_INSTR_STR(pVCpu, pCtx); break;
12311	case DBGFEVENT_INSTR_LLDT: VBOXVMM_INSTR_LLDT(pVCpu, pCtx); break;
12312	case DBGFEVENT_INSTR_LTR: VBOXVMM_INSTR_LTR(pVCpu, pCtx); break;
12313	case DBGFEVENT_INSTR_RDTSCP: VBOXVMM_INSTR_RDTSCP(pVCpu, pCtx); break;
12314	case DBGFEVENT_INSTR_WBINVD: VBOXVMM_INSTR_WBINVD(pVCpu, pCtx); break;
12315	case DBGFEVENT_INSTR_XSETBV: VBOXVMM_INSTR_XSETBV(pVCpu, pCtx); break;
12316	case DBGFEVENT_INSTR_RDRAND: VBOXVMM_INSTR_RDRAND(pVCpu, pCtx); break;
12317	case DBGFEVENT_INSTR_RDSEED: VBOXVMM_INSTR_RDSEED(pVCpu, pCtx); break;
12318	case DBGFEVENT_INSTR_XSAVES: VBOXVMM_INSTR_XSAVES(pVCpu, pCtx); break;
12319	case DBGFEVENT_INSTR_XRSTORS: VBOXVMM_INSTR_XRSTORS(pVCpu, pCtx); break;
12320	case DBGFEVENT_INSTR_VMM_CALL: VBOXVMM_INSTR_VMM_CALL(pVCpu, pCtx); break;
12321	case DBGFEVENT_INSTR_VMX_VMCLEAR: VBOXVMM_INSTR_VMX_VMCLEAR(pVCpu, pCtx); break;
12322	case DBGFEVENT_INSTR_VMX_VMLAUNCH: VBOXVMM_INSTR_VMX_VMLAUNCH(pVCpu, pCtx); break;
12323	case DBGFEVENT_INSTR_VMX_VMPTRLD: VBOXVMM_INSTR_VMX_VMPTRLD(pVCpu, pCtx); break;
12324	case DBGFEVENT_INSTR_VMX_VMPTRST: VBOXVMM_INSTR_VMX_VMPTRST(pVCpu, pCtx); break;
12325	case DBGFEVENT_INSTR_VMX_VMREAD: VBOXVMM_INSTR_VMX_VMREAD(pVCpu, pCtx); break;
12326	case DBGFEVENT_INSTR_VMX_VMRESUME: VBOXVMM_INSTR_VMX_VMRESUME(pVCpu, pCtx); break;
12327	case DBGFEVENT_INSTR_VMX_VMWRITE: VBOXVMM_INSTR_VMX_VMWRITE(pVCpu, pCtx); break;
12328	case DBGFEVENT_INSTR_VMX_VMXOFF: VBOXVMM_INSTR_VMX_VMXOFF(pVCpu, pCtx); break;
12329	case DBGFEVENT_INSTR_VMX_VMXON: VBOXVMM_INSTR_VMX_VMXON(pVCpu, pCtx); break;
12330	case DBGFEVENT_INSTR_VMX_INVEPT: VBOXVMM_INSTR_VMX_INVEPT(pVCpu, pCtx); break;
12331	case DBGFEVENT_INSTR_VMX_INVVPID: VBOXVMM_INSTR_VMX_INVVPID(pVCpu, pCtx); break;
12332	case DBGFEVENT_INSTR_VMX_INVPCID: VBOXVMM_INSTR_VMX_INVPCID(pVCpu, pCtx); break;
12333	case DBGFEVENT_INSTR_VMX_VMFUNC: VBOXVMM_INSTR_VMX_VMFUNC(pVCpu, pCtx); break;
12334	default: AssertMsgFailed(("enmEvent1=%d uExitReason=%d\n", enmEvent1, uExitReason)); break;
12335	}
12336	switch (enmEvent2)
12337	{
12338	/** @todo consider which extra parameters would be helpful for each probe. */
12339	case DBGFEVENT_END: break;
12340	case DBGFEVENT_EXIT_TASK_SWITCH: VBOXVMM_EXIT_TASK_SWITCH(pVCpu, pCtx); break;
12341	case DBGFEVENT_EXIT_CPUID: VBOXVMM_EXIT_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
12342	case DBGFEVENT_EXIT_GETSEC: VBOXVMM_EXIT_GETSEC(pVCpu, pCtx); break;
12343	case DBGFEVENT_EXIT_HALT: VBOXVMM_EXIT_HALT(pVCpu, pCtx); break;
12344	case DBGFEVENT_EXIT_INVD: VBOXVMM_EXIT_INVD(pVCpu, pCtx); break;
12345	case DBGFEVENT_EXIT_INVLPG: VBOXVMM_EXIT_INVLPG(pVCpu, pCtx); break;
12346	case DBGFEVENT_EXIT_RDPMC: VBOXVMM_EXIT_RDPMC(pVCpu, pCtx); break;
12347	case DBGFEVENT_EXIT_RDTSC: VBOXVMM_EXIT_RDTSC(pVCpu, pCtx); break;
12348	case DBGFEVENT_EXIT_RSM: VBOXVMM_EXIT_RSM(pVCpu, pCtx); break;
12349	case DBGFEVENT_EXIT_CRX_READ: VBOXVMM_EXIT_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12350	case DBGFEVENT_EXIT_CRX_WRITE: VBOXVMM_EXIT_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12351	case DBGFEVENT_EXIT_DRX_READ: VBOXVMM_EXIT_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12352	case DBGFEVENT_EXIT_DRX_WRITE: VBOXVMM_EXIT_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12353	case DBGFEVENT_EXIT_RDMSR: VBOXVMM_EXIT_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
12354	case DBGFEVENT_EXIT_WRMSR: VBOXVMM_EXIT_WRMSR(pVCpu, pCtx, pCtx->ecx,
12355	RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
12356	case DBGFEVENT_EXIT_MWAIT: VBOXVMM_EXIT_MWAIT(pVCpu, pCtx); break;
12357	case DBGFEVENT_EXIT_MONITOR: VBOXVMM_EXIT_MONITOR(pVCpu, pCtx); break;
12358	case DBGFEVENT_EXIT_PAUSE: VBOXVMM_EXIT_PAUSE(pVCpu, pCtx); break;
12359	case DBGFEVENT_EXIT_SGDT: VBOXVMM_EXIT_SGDT(pVCpu, pCtx); break;
12360	case DBGFEVENT_EXIT_SIDT: VBOXVMM_EXIT_SIDT(pVCpu, pCtx); break;
12361	case DBGFEVENT_EXIT_LGDT: VBOXVMM_EXIT_LGDT(pVCpu, pCtx); break;
12362	case DBGFEVENT_EXIT_LIDT: VBOXVMM_EXIT_LIDT(pVCpu, pCtx); break;
12363	case DBGFEVENT_EXIT_SLDT: VBOXVMM_EXIT_SLDT(pVCpu, pCtx); break;
12364	case DBGFEVENT_EXIT_STR: VBOXVMM_EXIT_STR(pVCpu, pCtx); break;
12365	case DBGFEVENT_EXIT_LLDT: VBOXVMM_EXIT_LLDT(pVCpu, pCtx); break;
12366	case DBGFEVENT_EXIT_LTR: VBOXVMM_EXIT_LTR(pVCpu, pCtx); break;
12367	case DBGFEVENT_EXIT_RDTSCP: VBOXVMM_EXIT_RDTSCP(pVCpu, pCtx); break;
12368	case DBGFEVENT_EXIT_WBINVD: VBOXVMM_EXIT_WBINVD(pVCpu, pCtx); break;
12369	case DBGFEVENT_EXIT_XSETBV: VBOXVMM_EXIT_XSETBV(pVCpu, pCtx); break;
12370	case DBGFEVENT_EXIT_RDRAND: VBOXVMM_EXIT_RDRAND(pVCpu, pCtx); break;
12371	case DBGFEVENT_EXIT_RDSEED: VBOXVMM_EXIT_RDSEED(pVCpu, pCtx); break;
12372	case DBGFEVENT_EXIT_XSAVES: VBOXVMM_EXIT_XSAVES(pVCpu, pCtx); break;
12373	case DBGFEVENT_EXIT_XRSTORS: VBOXVMM_EXIT_XRSTORS(pVCpu, pCtx); break;
12374	case DBGFEVENT_EXIT_VMM_CALL: VBOXVMM_EXIT_VMM_CALL(pVCpu, pCtx); break;
12375	case DBGFEVENT_EXIT_VMX_VMCLEAR: VBOXVMM_EXIT_VMX_VMCLEAR(pVCpu, pCtx); break;
12376	case DBGFEVENT_EXIT_VMX_VMLAUNCH: VBOXVMM_EXIT_VMX_VMLAUNCH(pVCpu, pCtx); break;
12377	case DBGFEVENT_EXIT_VMX_VMPTRLD: VBOXVMM_EXIT_VMX_VMPTRLD(pVCpu, pCtx); break;
12378	case DBGFEVENT_EXIT_VMX_VMPTRST: VBOXVMM_EXIT_VMX_VMPTRST(pVCpu, pCtx); break;
12379	case DBGFEVENT_EXIT_VMX_VMREAD: VBOXVMM_EXIT_VMX_VMREAD(pVCpu, pCtx); break;
12380	case DBGFEVENT_EXIT_VMX_VMRESUME: VBOXVMM_EXIT_VMX_VMRESUME(pVCpu, pCtx); break;
12381	case DBGFEVENT_EXIT_VMX_VMWRITE: VBOXVMM_EXIT_VMX_VMWRITE(pVCpu, pCtx); break;
12382	case DBGFEVENT_EXIT_VMX_VMXOFF: VBOXVMM_EXIT_VMX_VMXOFF(pVCpu, pCtx); break;
12383	case DBGFEVENT_EXIT_VMX_VMXON: VBOXVMM_EXIT_VMX_VMXON(pVCpu, pCtx); break;
12384	case DBGFEVENT_EXIT_VMX_INVEPT: VBOXVMM_EXIT_VMX_INVEPT(pVCpu, pCtx); break;
12385	case DBGFEVENT_EXIT_VMX_INVVPID: VBOXVMM_EXIT_VMX_INVVPID(pVCpu, pCtx); break;
12386	case DBGFEVENT_EXIT_VMX_INVPCID: VBOXVMM_EXIT_VMX_INVPCID(pVCpu, pCtx); break;
12387	case DBGFEVENT_EXIT_VMX_VMFUNC: VBOXVMM_EXIT_VMX_VMFUNC(pVCpu, pCtx); break;
12388	case DBGFEVENT_EXIT_VMX_EPT_MISCONFIG: VBOXVMM_EXIT_VMX_EPT_MISCONFIG(pVCpu, pCtx); break;
12389	case DBGFEVENT_EXIT_VMX_EPT_VIOLATION: VBOXVMM_EXIT_VMX_EPT_VIOLATION(pVCpu, pCtx); break;
12390	case DBGFEVENT_EXIT_VMX_VAPIC_ACCESS: VBOXVMM_EXIT_VMX_VAPIC_ACCESS(pVCpu, pCtx); break;
12391	case DBGFEVENT_EXIT_VMX_VAPIC_WRITE: VBOXVMM_EXIT_VMX_VAPIC_WRITE(pVCpu, pCtx); break;
12392	default: AssertMsgFailed(("enmEvent2=%d uExitReason=%d\n", enmEvent2, uExitReason)); break;
12393	}
12394	}
12395
12396	/*
12397	* Fire of the DBGF event, if enabled (our check here is just a quick one,
12398	* the DBGF call will do a full check).
12399	*
12400	* Note! DBGF sets DBGFEVENT_INTERRUPT_SOFTWARE in the bitmap.
12401	* Note! If we have to events, we prioritize the first, i.e. the instruction
12402	* one, in order to avoid event nesting.
12403	*/
12404	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
12405	if ( enmEvent1 != DBGFEVENT_END
12406	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent1))
12407	{
12408	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12409	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent1, DBGFEVENTCTX_HM, 1, uEventArg);
12410	if (rcStrict != VINF_SUCCESS)
12411	return rcStrict;
12412	}
12413	else if ( enmEvent2 != DBGFEVENT_END
12414	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent2))
12415	{
12416	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12417	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent2, DBGFEVENTCTX_HM, 1, uEventArg);
12418	if (rcStrict != VINF_SUCCESS)
12419	return rcStrict;
12420	}
12421
12422	return VINF_SUCCESS;
12423	}
12424
12425
12426	/**
12427	* Single-stepping VM-exit filtering.
12428	*
12429	* This is preprocessing the VM-exits and deciding whether we've gotten far
12430	* enough to return VINF_EM_DBG_STEPPED already. If not, normal VM-exit
12431	* handling is performed.
12432	*
12433	* @returns Strict VBox status code (i.e. informational status codes too).
12434	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
12435	* @param pVmxTransient The VMX-transient structure.
12436	* @param pDbgState The debug state.
12437	*/
12438	DECLINLINE(VBOXSTRICTRC) hmR0VmxRunDebugHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
12439	{
12440	/*
12441	* Expensive (saves context) generic dtrace VM-exit probe.
12442	*/
12443	uint32_t const uExitReason = pVmxTransient->uExitReason;
12444	if (!VBOXVMM_R0_HMVMX_VMEXIT_ENABLED())
12445	{ /* more likely */ }
12446	else
12447	{
12448	hmR0VmxReadExitQualVmcs(pVmxTransient);
12449	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
12450	AssertRC(rc);
12451	VBOXVMM_R0_HMVMX_VMEXIT(pVCpu, &pVCpu->cpum.GstCtx, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
12452	}
12453
12454	/*
12455	* Check for host NMI, just to get that out of the way.
12456	*/
12457	if (uExitReason != VMX_EXIT_XCPT_OR_NMI)
12458	{ /* normally likely */ }
12459	else
12460	{
12461	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
12462	uint32_t const uIntType = VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo);
12463	if (uIntType == VMX_EXIT_INT_INFO_TYPE_NMI)
12464	return hmR0VmxExitHostNmi(pVCpu, pVmxTransient->pVmcsInfo);
12465	}
12466
12467	/*
12468	* Check for single stepping event if we're stepping.
12469	*/
12470	if (pVCpu->hm.s.fSingleInstruction)
12471	{
12472	switch (uExitReason)
12473	{
12474	case VMX_EXIT_MTF:
12475	return hmR0VmxExitMtf(pVCpu, pVmxTransient);
12476
12477	/* Various events: */
12478	case VMX_EXIT_XCPT_OR_NMI:
12479	case VMX_EXIT_EXT_INT:
12480	case VMX_EXIT_TRIPLE_FAULT:
12481	case VMX_EXIT_INT_WINDOW:
12482	case VMX_EXIT_NMI_WINDOW:
12483	case VMX_EXIT_TASK_SWITCH:
12484	case VMX_EXIT_TPR_BELOW_THRESHOLD:
12485	case VMX_EXIT_APIC_ACCESS:
12486	case VMX_EXIT_EPT_VIOLATION:
12487	case VMX_EXIT_EPT_MISCONFIG:
12488	case VMX_EXIT_PREEMPT_TIMER:
12489
12490	/* Instruction specific VM-exits: */
12491	case VMX_EXIT_CPUID:
12492	case VMX_EXIT_GETSEC:
12493	case VMX_EXIT_HLT:
12494	case VMX_EXIT_INVD:
12495	case VMX_EXIT_INVLPG:
12496	case VMX_EXIT_RDPMC:
12497	case VMX_EXIT_RDTSC:
12498	case VMX_EXIT_RSM:
12499	case VMX_EXIT_VMCALL:
12500	case VMX_EXIT_VMCLEAR:
12501	case VMX_EXIT_VMLAUNCH:
12502	case VMX_EXIT_VMPTRLD:
12503	case VMX_EXIT_VMPTRST:
12504	case VMX_EXIT_VMREAD:
12505	case VMX_EXIT_VMRESUME:
12506	case VMX_EXIT_VMWRITE:
12507	case VMX_EXIT_VMXOFF:
12508	case VMX_EXIT_VMXON:
12509	case VMX_EXIT_MOV_CRX:
12510	case VMX_EXIT_MOV_DRX:
12511	case VMX_EXIT_IO_INSTR:
12512	case VMX_EXIT_RDMSR:
12513	case VMX_EXIT_WRMSR:
12514	case VMX_EXIT_MWAIT:
12515	case VMX_EXIT_MONITOR:
12516	case VMX_EXIT_PAUSE:
12517	case VMX_EXIT_GDTR_IDTR_ACCESS:
12518	case VMX_EXIT_LDTR_TR_ACCESS:
12519	case VMX_EXIT_INVEPT:
12520	case VMX_EXIT_RDTSCP:
12521	case VMX_EXIT_INVVPID:
12522	case VMX_EXIT_WBINVD:
12523	case VMX_EXIT_XSETBV:
12524	case VMX_EXIT_RDRAND:
12525	case VMX_EXIT_INVPCID:
12526	case VMX_EXIT_VMFUNC:
12527	case VMX_EXIT_RDSEED:
12528	case VMX_EXIT_XSAVES:
12529	case VMX_EXIT_XRSTORS:
12530	{
12531	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12532	AssertRCReturn(rc, rc);
12533	if ( pVCpu->cpum.GstCtx.rip != pDbgState->uRipStart
12534	\|\| pVCpu->cpum.GstCtx.cs.Sel != pDbgState->uCsStart)
12535	return VINF_EM_DBG_STEPPED;
12536	break;
12537	}
12538
12539	/* Errors and unexpected events: */
12540	case VMX_EXIT_INIT_SIGNAL:
12541	case VMX_EXIT_SIPI:
12542	case VMX_EXIT_IO_SMI:
12543	case VMX_EXIT_SMI:
12544	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
12545	case VMX_EXIT_ERR_MSR_LOAD:
12546	case VMX_EXIT_ERR_MACHINE_CHECK:
12547	case VMX_EXIT_PML_FULL:
12548	case VMX_EXIT_VIRTUALIZED_EOI:
12549	case VMX_EXIT_APIC_WRITE: /* Some talk about this being fault like, so I guess we must process it? */
12550	break;
12551
12552	default:
12553	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
12554	break;
12555	}
12556	}
12557
12558	/*
12559	* Check for debugger event breakpoints and dtrace probes.
12560	*/
12561	if ( uExitReason < RT_ELEMENTS(pDbgState->bmExitsToCheck) * 32U
12562	&& ASMBitTest(pDbgState->bmExitsToCheck, uExitReason) )
12563	{
12564	VBOXSTRICTRC rcStrict = hmR0VmxHandleExitDtraceEvents(pVCpu, pVmxTransient, uExitReason);
12565	if (rcStrict != VINF_SUCCESS)
12566	return rcStrict;
12567	}
12568
12569	/*
12570	* Normal processing.
12571	*/
12572	#ifdef HMVMX_USE_FUNCTION_TABLE
12573	return g_aVMExitHandlers[uExitReason].pfn(pVCpu, pVmxTransient);
12574	#else
12575	return hmR0VmxHandleExit(pVCpu, pVmxTransient, uExitReason);
12576	#endif
12577	}
12578
12579
12580	/**
12581	* Single steps guest code using hardware-assisted VMX.
12582	*
12583	* This is -not- the same as the guest single-stepping itself (say using EFLAGS.TF)
12584	* but single-stepping through the hypervisor debugger.
12585	*
12586	* @returns Strict VBox status code (i.e. informational status codes too).
12587	* @param pVCpu The cross context virtual CPU structure.
12588	* @param pcLoops Pointer to the number of executed loops.
12589	*
12590	* @note Mostly the same as hmR0VmxRunGuestCodeNormal().
12591	*/
12592	static VBOXSTRICTRC hmR0VmxRunGuestCodeDebug(PVMCPUCC pVCpu, uint32_t *pcLoops)
12593	{
12594	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops;
12595	Assert(pcLoops);
12596	Assert(*pcLoops <= cMaxResumeLoops);
12597
12598	VMXTRANSIENT VmxTransient;
12599	RT_ZERO(VmxTransient);
12600	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
12601
12602	/* Set HMCPU indicators. */
12603	bool const fSavedSingleInstruction = pVCpu->hm.s.fSingleInstruction;
12604	pVCpu->hm.s.fSingleInstruction = pVCpu->hm.s.fSingleInstruction \|\| DBGFIsStepping(pVCpu);
12605	pVCpu->hmr0.s.fDebugWantRdTscExit = false;
12606	pVCpu->hmr0.s.fUsingDebugLoop = true;
12607
12608	/* State we keep to help modify and later restore the VMCS fields we alter, and for detecting steps. */
12609	VMXRUNDBGSTATE DbgState;
12610	hmR0VmxRunDebugStateInit(pVCpu, &VmxTransient, &DbgState);
12611	hmR0VmxPreRunGuestDebugStateUpdate(pVCpu, &VmxTransient, &DbgState);
12612
12613	/*
12614	* The loop.
12615	*/
12616	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
12617	for (;;)
12618	{
12619	Assert(!HMR0SuspendPending());
12620	HMVMX_ASSERT_CPU_SAFE(pVCpu);
12621	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
12622	bool fStepping = pVCpu->hm.s.fSingleInstruction;
12623
12624	/* Set up VM-execution controls the next two can respond to. */
12625	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &VmxTransient, &DbgState);
12626
12627	/*
12628	* Preparatory work for running guest code, this may force us to
12629	* return to ring-3.
12630	*
12631	* Warning! This bugger disables interrupts on VINF_SUCCESS!
12632	*/
12633	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, fStepping);
12634	if (rcStrict != VINF_SUCCESS)
12635	break;
12636
12637	/* Interrupts are disabled at this point! */
12638	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
12639
12640	/* Override any obnoxious code in the above two calls. */
12641	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &VmxTransient, &DbgState);
12642
12643	/*
12644	* Finally execute the guest.
12645	*/
12646	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
12647
12648	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
12649	/* Interrupts are re-enabled at this point! */
12650
12651	/* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
12652	if (RT_SUCCESS(rcRun))
12653	{ /* very likely */ }
12654	else
12655	{
12656	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
12657	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
12658	return rcRun;
12659	}
12660
12661	/* Profile the VM-exit. */
12662	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
12663	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
12664	STAM_COUNTER_INC(&pVCpu->hm.s.aStatExitReason[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
12665	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
12666	HMVMX_START_EXIT_DISPATCH_PROF();
12667
12668	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
12669
12670	/*
12671	* Handle the VM-exit - we quit earlier on certain VM-exits, see hmR0VmxHandleExitDebug().
12672	*/
12673	rcStrict = hmR0VmxRunDebugHandleExit(pVCpu, &VmxTransient, &DbgState);
12674	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
12675	if (rcStrict != VINF_SUCCESS)
12676	break;
12677	if (++(*pcLoops) > cMaxResumeLoops)
12678	{
12679	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
12680	rcStrict = VINF_EM_RAW_INTERRUPT;
12681	break;
12682	}
12683
12684	/*
12685	* Stepping: Did the RIP change, if so, consider it a single step.
12686	* Otherwise, make sure one of the TFs gets set.
12687	*/
12688	if (fStepping)
12689	{
12690	int rc = hmR0VmxImportGuestState(pVCpu, VmxTransient.pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12691	AssertRC(rc);
12692	if ( pVCpu->cpum.GstCtx.rip != DbgState.uRipStart
12693	\|\| pVCpu->cpum.GstCtx.cs.Sel != DbgState.uCsStart)
12694	{
12695	rcStrict = VINF_EM_DBG_STEPPED;
12696	break;
12697	}
12698	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
12699	}
12700
12701	/*
12702	* Update when dtrace settings changes (DBGF kicks us, so no need to check).
12703	*/
12704	if (VBOXVMM_GET_SETTINGS_SEQ_NO() != DbgState.uDtraceSettingsSeqNo)
12705	hmR0VmxPreRunGuestDebugStateUpdate(pVCpu, &VmxTransient, &DbgState);
12706
12707	/* Restore all controls applied by hmR0VmxPreRunGuestDebugStateApply above. */
12708	rcStrict = hmR0VmxRunDebugStateRevert(pVCpu, &VmxTransient, &DbgState, rcStrict);
12709	Assert(rcStrict == VINF_SUCCESS);
12710	}
12711
12712	/*
12713	* Clear the X86_EFL_TF if necessary.
12714	*/
12715	if (pVCpu->hmr0.s.fClearTrapFlag)
12716	{
12717	int rc = hmR0VmxImportGuestState(pVCpu, VmxTransient.pVmcsInfo, CPUMCTX_EXTRN_RFLAGS);
12718	AssertRC(rc);
12719	pVCpu->hmr0.s.fClearTrapFlag = false;
12720	pVCpu->cpum.GstCtx.eflags.Bits.u1TF = 0;
12721	}
12722	/** @todo there seems to be issues with the resume flag when the monitor trap
12723	* flag is pending without being used. Seen early in bios init when
12724	* accessing APIC page in protected mode. */
12725
12726	/* Restore HMCPU indicators. */
12727	pVCpu->hmr0.s.fUsingDebugLoop = false;
12728	pVCpu->hmr0.s.fDebugWantRdTscExit = false;
12729	pVCpu->hm.s.fSingleInstruction = fSavedSingleInstruction;
12730
12731	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
12732	return rcStrict;
12733	}
12734
12735
12736	/** @} */
12737
12738
12739	/**
12740	* Checks if any expensive dtrace probes are enabled and we should go to the
12741	* debug loop.
12742	*
12743	* @returns true if we should use debug loop, false if not.
12744	*/
12745	static bool hmR0VmxAnyExpensiveProbesEnabled(void)
12746	{
12747	/* It's probably faster to OR the raw 32-bit counter variables together.
12748	Since the variables are in an array and the probes are next to one
12749	another (more or less), we have good locality. So, better read
12750	eight-nine cache lines ever time and only have one conditional, than
12751	128+ conditionals, right? */
12752	return ( VBOXVMM_R0_HMVMX_VMEXIT_ENABLED_RAW() /* expensive too due to context */
12753	\| VBOXVMM_XCPT_DE_ENABLED_RAW()
12754	\| VBOXVMM_XCPT_DB_ENABLED_RAW()
12755	\| VBOXVMM_XCPT_BP_ENABLED_RAW()
12756	\| VBOXVMM_XCPT_OF_ENABLED_RAW()
12757	\| VBOXVMM_XCPT_BR_ENABLED_RAW()
12758	\| VBOXVMM_XCPT_UD_ENABLED_RAW()
12759	\| VBOXVMM_XCPT_NM_ENABLED_RAW()
12760	\| VBOXVMM_XCPT_DF_ENABLED_RAW()
12761	\| VBOXVMM_XCPT_TS_ENABLED_RAW()
12762	\| VBOXVMM_XCPT_NP_ENABLED_RAW()
12763	\| VBOXVMM_XCPT_SS_ENABLED_RAW()
12764	\| VBOXVMM_XCPT_GP_ENABLED_RAW()
12765	\| VBOXVMM_XCPT_PF_ENABLED_RAW()
12766	\| VBOXVMM_XCPT_MF_ENABLED_RAW()
12767	\| VBOXVMM_XCPT_AC_ENABLED_RAW()
12768	\| VBOXVMM_XCPT_XF_ENABLED_RAW()
12769	\| VBOXVMM_XCPT_VE_ENABLED_RAW()
12770	\| VBOXVMM_XCPT_SX_ENABLED_RAW()
12771	\| VBOXVMM_INT_SOFTWARE_ENABLED_RAW()
12772	\| VBOXVMM_INT_HARDWARE_ENABLED_RAW()
12773	) != 0
12774	\|\| ( VBOXVMM_INSTR_HALT_ENABLED_RAW()
12775	\| VBOXVMM_INSTR_MWAIT_ENABLED_RAW()
12776	\| VBOXVMM_INSTR_MONITOR_ENABLED_RAW()
12777	\| VBOXVMM_INSTR_CPUID_ENABLED_RAW()
12778	\| VBOXVMM_INSTR_INVD_ENABLED_RAW()
12779	\| VBOXVMM_INSTR_WBINVD_ENABLED_RAW()
12780	\| VBOXVMM_INSTR_INVLPG_ENABLED_RAW()
12781	\| VBOXVMM_INSTR_RDTSC_ENABLED_RAW()
12782	\| VBOXVMM_INSTR_RDTSCP_ENABLED_RAW()
12783	\| VBOXVMM_INSTR_RDPMC_ENABLED_RAW()
12784	\| VBOXVMM_INSTR_RDMSR_ENABLED_RAW()
12785	\| VBOXVMM_INSTR_WRMSR_ENABLED_RAW()
12786	\| VBOXVMM_INSTR_CRX_READ_ENABLED_RAW()
12787	\| VBOXVMM_INSTR_CRX_WRITE_ENABLED_RAW()
12788	\| VBOXVMM_INSTR_DRX_READ_ENABLED_RAW()
12789	\| VBOXVMM_INSTR_DRX_WRITE_ENABLED_RAW()
12790	\| VBOXVMM_INSTR_PAUSE_ENABLED_RAW()
12791	\| VBOXVMM_INSTR_XSETBV_ENABLED_RAW()
12792	\| VBOXVMM_INSTR_SIDT_ENABLED_RAW()
12793	\| VBOXVMM_INSTR_LIDT_ENABLED_RAW()
12794	\| VBOXVMM_INSTR_SGDT_ENABLED_RAW()
12795	\| VBOXVMM_INSTR_LGDT_ENABLED_RAW()
12796	\| VBOXVMM_INSTR_SLDT_ENABLED_RAW()
12797	\| VBOXVMM_INSTR_LLDT_ENABLED_RAW()
12798	\| VBOXVMM_INSTR_STR_ENABLED_RAW()
12799	\| VBOXVMM_INSTR_LTR_ENABLED_RAW()
12800	\| VBOXVMM_INSTR_GETSEC_ENABLED_RAW()
12801	\| VBOXVMM_INSTR_RSM_ENABLED_RAW()
12802	\| VBOXVMM_INSTR_RDRAND_ENABLED_RAW()
12803	\| VBOXVMM_INSTR_RDSEED_ENABLED_RAW()
12804	\| VBOXVMM_INSTR_XSAVES_ENABLED_RAW()
12805	\| VBOXVMM_INSTR_XRSTORS_ENABLED_RAW()
12806	\| VBOXVMM_INSTR_VMM_CALL_ENABLED_RAW()
12807	\| VBOXVMM_INSTR_VMX_VMCLEAR_ENABLED_RAW()
12808	\| VBOXVMM_INSTR_VMX_VMLAUNCH_ENABLED_RAW()
12809	\| VBOXVMM_INSTR_VMX_VMPTRLD_ENABLED_RAW()
12810	\| VBOXVMM_INSTR_VMX_VMPTRST_ENABLED_RAW()
12811	\| VBOXVMM_INSTR_VMX_VMREAD_ENABLED_RAW()
12812	\| VBOXVMM_INSTR_VMX_VMRESUME_ENABLED_RAW()
12813	\| VBOXVMM_INSTR_VMX_VMWRITE_ENABLED_RAW()
12814	\| VBOXVMM_INSTR_VMX_VMXOFF_ENABLED_RAW()
12815	\| VBOXVMM_INSTR_VMX_VMXON_ENABLED_RAW()
12816	\| VBOXVMM_INSTR_VMX_VMFUNC_ENABLED_RAW()
12817	\| VBOXVMM_INSTR_VMX_INVEPT_ENABLED_RAW()
12818	\| VBOXVMM_INSTR_VMX_INVVPID_ENABLED_RAW()
12819	\| VBOXVMM_INSTR_VMX_INVPCID_ENABLED_RAW()
12820	) != 0
12821	\|\| ( VBOXVMM_EXIT_TASK_SWITCH_ENABLED_RAW()
12822	\| VBOXVMM_EXIT_HALT_ENABLED_RAW()
12823	\| VBOXVMM_EXIT_MWAIT_ENABLED_RAW()
12824	\| VBOXVMM_EXIT_MONITOR_ENABLED_RAW()
12825	\| VBOXVMM_EXIT_CPUID_ENABLED_RAW()
12826	\| VBOXVMM_EXIT_INVD_ENABLED_RAW()
12827	\| VBOXVMM_EXIT_WBINVD_ENABLED_RAW()
12828	\| VBOXVMM_EXIT_INVLPG_ENABLED_RAW()
12829	\| VBOXVMM_EXIT_RDTSC_ENABLED_RAW()
12830	\| VBOXVMM_EXIT_RDTSCP_ENABLED_RAW()
12831	\| VBOXVMM_EXIT_RDPMC_ENABLED_RAW()
12832	\| VBOXVMM_EXIT_RDMSR_ENABLED_RAW()
12833	\| VBOXVMM_EXIT_WRMSR_ENABLED_RAW()
12834	\| VBOXVMM_EXIT_CRX_READ_ENABLED_RAW()
12835	\| VBOXVMM_EXIT_CRX_WRITE_ENABLED_RAW()
12836	\| VBOXVMM_EXIT_DRX_READ_ENABLED_RAW()
12837	\| VBOXVMM_EXIT_DRX_WRITE_ENABLED_RAW()
12838	\| VBOXVMM_EXIT_PAUSE_ENABLED_RAW()
12839	\| VBOXVMM_EXIT_XSETBV_ENABLED_RAW()
12840	\| VBOXVMM_EXIT_SIDT_ENABLED_RAW()
12841	\| VBOXVMM_EXIT_LIDT_ENABLED_RAW()
12842	\| VBOXVMM_EXIT_SGDT_ENABLED_RAW()
12843	\| VBOXVMM_EXIT_LGDT_ENABLED_RAW()
12844	\| VBOXVMM_EXIT_SLDT_ENABLED_RAW()
12845	\| VBOXVMM_EXIT_LLDT_ENABLED_RAW()
12846	\| VBOXVMM_EXIT_STR_ENABLED_RAW()
12847	\| VBOXVMM_EXIT_LTR_ENABLED_RAW()
12848	\| VBOXVMM_EXIT_GETSEC_ENABLED_RAW()
12849	\| VBOXVMM_EXIT_RSM_ENABLED_RAW()
12850	\| VBOXVMM_EXIT_RDRAND_ENABLED_RAW()
12851	\| VBOXVMM_EXIT_RDSEED_ENABLED_RAW()
12852	\| VBOXVMM_EXIT_XSAVES_ENABLED_RAW()
12853	\| VBOXVMM_EXIT_XRSTORS_ENABLED_RAW()
12854	\| VBOXVMM_EXIT_VMM_CALL_ENABLED_RAW()
12855	\| VBOXVMM_EXIT_VMX_VMCLEAR_ENABLED_RAW()
12856	\| VBOXVMM_EXIT_VMX_VMLAUNCH_ENABLED_RAW()
12857	\| VBOXVMM_EXIT_VMX_VMPTRLD_ENABLED_RAW()
12858	\| VBOXVMM_EXIT_VMX_VMPTRST_ENABLED_RAW()
12859	\| VBOXVMM_EXIT_VMX_VMREAD_ENABLED_RAW()
12860	\| VBOXVMM_EXIT_VMX_VMRESUME_ENABLED_RAW()
12861	\| VBOXVMM_EXIT_VMX_VMWRITE_ENABLED_RAW()
12862	\| VBOXVMM_EXIT_VMX_VMXOFF_ENABLED_RAW()
12863	\| VBOXVMM_EXIT_VMX_VMXON_ENABLED_RAW()
12864	\| VBOXVMM_EXIT_VMX_VMFUNC_ENABLED_RAW()
12865	\| VBOXVMM_EXIT_VMX_INVEPT_ENABLED_RAW()
12866	\| VBOXVMM_EXIT_VMX_INVVPID_ENABLED_RAW()
12867	\| VBOXVMM_EXIT_VMX_INVPCID_ENABLED_RAW()
12868	\| VBOXVMM_EXIT_VMX_EPT_VIOLATION_ENABLED_RAW()
12869	\| VBOXVMM_EXIT_VMX_EPT_MISCONFIG_ENABLED_RAW()
12870	\| VBOXVMM_EXIT_VMX_VAPIC_ACCESS_ENABLED_RAW()
12871	\| VBOXVMM_EXIT_VMX_VAPIC_WRITE_ENABLED_RAW()
12872	) != 0;
12873	}
12874
12875
12876	/**
12877	* Runs the guest using hardware-assisted VMX.
12878	*
12879	* @returns Strict VBox status code (i.e. informational status codes too).
12880	* @param pVCpu The cross context virtual CPU structure.
12881	*/
12882	VMMR0DECL(VBOXSTRICTRC) VMXR0RunGuestCode(PVMCPUCC pVCpu)
12883	{
12884	AssertPtr(pVCpu);
12885	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
12886	Assert(VMMRZCallRing3IsEnabled(pVCpu));
12887	Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
12888	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
12889
12890	VBOXSTRICTRC rcStrict;
12891	uint32_t cLoops = 0;
12892	for (;;)
12893	{
12894	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12895	bool const fInNestedGuestMode = CPUMIsGuestInVmxNonRootMode(pCtx);
12896	#else
12897	NOREF(pCtx);
12898	bool const fInNestedGuestMode = false;
12899	#endif
12900	if (!fInNestedGuestMode)
12901	{
12902	if ( !pVCpu->hm.s.fUseDebugLoop
12903	&& (!VBOXVMM_ANY_PROBES_ENABLED() \|\| !hmR0VmxAnyExpensiveProbesEnabled())
12904	&& !DBGFIsStepping(pVCpu)
12905	&& !pVCpu->CTX_SUFF(pVM)->dbgf.ro.cEnabledInt3Breakpoints)
12906	rcStrict = hmR0VmxRunGuestCodeNormal(pVCpu, &cLoops);
12907	else
12908	rcStrict = hmR0VmxRunGuestCodeDebug(pVCpu, &cLoops);
12909	}
12910	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12911	else
12912	rcStrict = hmR0VmxRunGuestCodeNested(pVCpu, &cLoops);
12913
12914	if (rcStrict == VINF_VMX_VMLAUNCH_VMRESUME)
12915	{
12916	Assert(CPUMIsGuestInVmxNonRootMode(pCtx));
12917	continue;
12918	}
12919	if (rcStrict == VINF_VMX_VMEXIT)
12920	{
12921	Assert(!CPUMIsGuestInVmxNonRootMode(pCtx));
12922	continue;
12923	}
12924	#endif
12925	break;
12926	}
12927
12928	int const rcLoop = VBOXSTRICTRC_VAL(rcStrict);
12929	switch (rcLoop)
12930	{
12931	case VERR_EM_INTERPRETER: rcStrict = VINF_EM_RAW_EMULATE_INSTR; break;
12932	case VINF_EM_RESET: rcStrict = VINF_EM_TRIPLE_FAULT; break;
12933	}
12934
12935	int rc2 = hmR0VmxExitToRing3(pVCpu, rcStrict);
12936	if (RT_FAILURE(rc2))
12937	{
12938	pVCpu->hm.s.u32HMError = (uint32_t)VBOXSTRICTRC_VAL(rcStrict);
12939	rcStrict = rc2;
12940	}
12941	Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
12942	Assert(!VMMR0AssertionIsNotificationSet(pVCpu));
12943	return rcStrict;
12944	}
12945
12946
12947	#ifndef HMVMX_USE_FUNCTION_TABLE
12948	/**
12949	* Handles a guest VM-exit from hardware-assisted VMX execution.
12950	*
12951	* @returns Strict VBox status code (i.e. informational status codes too).
12952	* @param pVCpu The cross context virtual CPU structure.
12953	* @param pVmxTransient The VMX-transient structure.
12954	*/
12955	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
12956	{
12957	#ifdef DEBUG_ramshankar
12958	# define VMEXIT_CALL_RET(a_fSave, a_CallExpr) \
12959	do { \
12960	if (a_fSave != 0) \
12961	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL); \
12962	VBOXSTRICTRC rcStrict = a_CallExpr; \
12963	if (a_fSave != 0) \
12964	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST); \
12965	return rcStrict; \
12966	} while (0)
12967	#else
12968	# define VMEXIT_CALL_RET(a_fSave, a_CallExpr) return a_CallExpr
12969	#endif
12970	uint32_t const uExitReason = pVmxTransient->uExitReason;
12971	switch (uExitReason)
12972	{
12973	case VMX_EXIT_EPT_MISCONFIG: VMEXIT_CALL_RET(0, hmR0VmxExitEptMisconfig(pVCpu, pVmxTransient));
12974	case VMX_EXIT_EPT_VIOLATION: VMEXIT_CALL_RET(0, hmR0VmxExitEptViolation(pVCpu, pVmxTransient));
12975	case VMX_EXIT_IO_INSTR: VMEXIT_CALL_RET(0, hmR0VmxExitIoInstr(pVCpu, pVmxTransient));
12976	case VMX_EXIT_CPUID: VMEXIT_CALL_RET(0, hmR0VmxExitCpuid(pVCpu, pVmxTransient));
12977	case VMX_EXIT_RDTSC: VMEXIT_CALL_RET(0, hmR0VmxExitRdtsc(pVCpu, pVmxTransient));
12978	case VMX_EXIT_RDTSCP: VMEXIT_CALL_RET(0, hmR0VmxExitRdtscp(pVCpu, pVmxTransient));
12979	case VMX_EXIT_APIC_ACCESS: VMEXIT_CALL_RET(0, hmR0VmxExitApicAccess(pVCpu, pVmxTransient));
12980	case VMX_EXIT_XCPT_OR_NMI: VMEXIT_CALL_RET(0, hmR0VmxExitXcptOrNmi(pVCpu, pVmxTransient));
12981	case VMX_EXIT_MOV_CRX: VMEXIT_CALL_RET(0, hmR0VmxExitMovCRx(pVCpu, pVmxTransient));
12982	case VMX_EXIT_EXT_INT: VMEXIT_CALL_RET(0, hmR0VmxExitExtInt(pVCpu, pVmxTransient));
12983	case VMX_EXIT_INT_WINDOW: VMEXIT_CALL_RET(0, hmR0VmxExitIntWindow(pVCpu, pVmxTransient));
12984	case VMX_EXIT_TPR_BELOW_THRESHOLD: VMEXIT_CALL_RET(0, hmR0VmxExitTprBelowThreshold(pVCpu, pVmxTransient));
12985	case VMX_EXIT_MWAIT: VMEXIT_CALL_RET(0, hmR0VmxExitMwait(pVCpu, pVmxTransient));
12986	case VMX_EXIT_MONITOR: VMEXIT_CALL_RET(0, hmR0VmxExitMonitor(pVCpu, pVmxTransient));
12987	case VMX_EXIT_TASK_SWITCH: VMEXIT_CALL_RET(0, hmR0VmxExitTaskSwitch(pVCpu, pVmxTransient));
12988	case VMX_EXIT_PREEMPT_TIMER: VMEXIT_CALL_RET(0, hmR0VmxExitPreemptTimer(pVCpu, pVmxTransient));
12989	case VMX_EXIT_RDMSR: VMEXIT_CALL_RET(0, hmR0VmxExitRdmsr(pVCpu, pVmxTransient));
12990	case VMX_EXIT_WRMSR: VMEXIT_CALL_RET(0, hmR0VmxExitWrmsr(pVCpu, pVmxTransient));
12991	case VMX_EXIT_VMCALL: VMEXIT_CALL_RET(0, hmR0VmxExitVmcall(pVCpu, pVmxTransient));
12992	case VMX_EXIT_MOV_DRX: VMEXIT_CALL_RET(0, hmR0VmxExitMovDRx(pVCpu, pVmxTransient));
12993	case VMX_EXIT_HLT: VMEXIT_CALL_RET(0, hmR0VmxExitHlt(pVCpu, pVmxTransient));
12994	case VMX_EXIT_INVD: VMEXIT_CALL_RET(0, hmR0VmxExitInvd(pVCpu, pVmxTransient));
12995	case VMX_EXIT_INVLPG: VMEXIT_CALL_RET(0, hmR0VmxExitInvlpg(pVCpu, pVmxTransient));
12996	case VMX_EXIT_MTF: VMEXIT_CALL_RET(0, hmR0VmxExitMtf(pVCpu, pVmxTransient));
12997	case VMX_EXIT_PAUSE: VMEXIT_CALL_RET(0, hmR0VmxExitPause(pVCpu, pVmxTransient));
12998	case VMX_EXIT_WBINVD: VMEXIT_CALL_RET(0, hmR0VmxExitWbinvd(pVCpu, pVmxTransient));
12999	case VMX_EXIT_XSETBV: VMEXIT_CALL_RET(0, hmR0VmxExitXsetbv(pVCpu, pVmxTransient));
13000	case VMX_EXIT_INVPCID: VMEXIT_CALL_RET(0, hmR0VmxExitInvpcid(pVCpu, pVmxTransient));
13001	case VMX_EXIT_GETSEC: VMEXIT_CALL_RET(0, hmR0VmxExitGetsec(pVCpu, pVmxTransient));
13002	case VMX_EXIT_RDPMC: VMEXIT_CALL_RET(0, hmR0VmxExitRdpmc(pVCpu, pVmxTransient));
13003	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13004	case VMX_EXIT_VMCLEAR: VMEXIT_CALL_RET(0, hmR0VmxExitVmclear(pVCpu, pVmxTransient));
13005	case VMX_EXIT_VMLAUNCH: VMEXIT_CALL_RET(0, hmR0VmxExitVmlaunch(pVCpu, pVmxTransient));
13006	case VMX_EXIT_VMPTRLD: VMEXIT_CALL_RET(0, hmR0VmxExitVmptrld(pVCpu, pVmxTransient));
13007	case VMX_EXIT_VMPTRST: VMEXIT_CALL_RET(0, hmR0VmxExitVmptrst(pVCpu, pVmxTransient));
13008	case VMX_EXIT_VMREAD: VMEXIT_CALL_RET(0, hmR0VmxExitVmread(pVCpu, pVmxTransient));
13009	case VMX_EXIT_VMRESUME: VMEXIT_CALL_RET(0, hmR0VmxExitVmwrite(pVCpu, pVmxTransient));
13010	case VMX_EXIT_VMWRITE: VMEXIT_CALL_RET(0, hmR0VmxExitVmresume(pVCpu, pVmxTransient));
13011	case VMX_EXIT_VMXOFF: VMEXIT_CALL_RET(0, hmR0VmxExitVmxoff(pVCpu, pVmxTransient));
13012	case VMX_EXIT_VMXON: VMEXIT_CALL_RET(0, hmR0VmxExitVmxon(pVCpu, pVmxTransient));
13013	case VMX_EXIT_INVVPID: VMEXIT_CALL_RET(0, hmR0VmxExitInvvpid(pVCpu, pVmxTransient));
13014	case VMX_EXIT_INVEPT: VMEXIT_CALL_RET(0, hmR0VmxExitSetPendingXcptUD(pVCpu, pVmxTransient));
13015	#else
13016	case VMX_EXIT_VMCLEAR:
13017	case VMX_EXIT_VMLAUNCH:
13018	case VMX_EXIT_VMPTRLD:
13019	case VMX_EXIT_VMPTRST:
13020	case VMX_EXIT_VMREAD:
13021	case VMX_EXIT_VMRESUME:
13022	case VMX_EXIT_VMWRITE:
13023	case VMX_EXIT_VMXOFF:
13024	case VMX_EXIT_VMXON:
13025	case VMX_EXIT_INVVPID:
13026	case VMX_EXIT_INVEPT:
13027	return hmR0VmxExitSetPendingXcptUD(pVCpu, pVmxTransient);
13028	#endif
13029
13030	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFault(pVCpu, pVmxTransient);
13031	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindow(pVCpu, pVmxTransient);
13032	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestState(pVCpu, pVmxTransient);
13033
13034	case VMX_EXIT_INIT_SIGNAL:
13035	case VMX_EXIT_SIPI:
13036	case VMX_EXIT_IO_SMI:
13037	case VMX_EXIT_SMI:
13038	case VMX_EXIT_ERR_MSR_LOAD:
13039	case VMX_EXIT_ERR_MACHINE_CHECK:
13040	case VMX_EXIT_PML_FULL:
13041	case VMX_EXIT_VIRTUALIZED_EOI:
13042	case VMX_EXIT_GDTR_IDTR_ACCESS:
13043	case VMX_EXIT_LDTR_TR_ACCESS:
13044	case VMX_EXIT_APIC_WRITE:
13045	case VMX_EXIT_RDRAND:
13046	case VMX_EXIT_RSM:
13047	case VMX_EXIT_VMFUNC:
13048	case VMX_EXIT_ENCLS:
13049	case VMX_EXIT_RDSEED:
13050	case VMX_EXIT_XSAVES:
13051	case VMX_EXIT_XRSTORS:
13052	case VMX_EXIT_UMWAIT:
13053	case VMX_EXIT_TPAUSE:
13054	case VMX_EXIT_LOADIWKEY:
13055	default:
13056	return hmR0VmxExitErrUnexpected(pVCpu, pVmxTransient);
13057	}
13058	#undef VMEXIT_CALL_RET
13059	}
13060	#endif /* !HMVMX_USE_FUNCTION_TABLE */
13061
13062
13063	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13064	/**
13065	* Handles a nested-guest VM-exit from hardware-assisted VMX execution.
13066	*
13067	* @returns Strict VBox status code (i.e. informational status codes too).
13068	* @param pVCpu The cross context virtual CPU structure.
13069	* @param pVmxTransient The VMX-transient structure.
13070	*/
13071	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13072	{
13073	uint32_t const uExitReason = pVmxTransient->uExitReason;
13074	switch (uExitReason)
13075	{
13076	case VMX_EXIT_EPT_MISCONFIG: return hmR0VmxExitEptMisconfig(pVCpu, pVmxTransient);
13077	case VMX_EXIT_EPT_VIOLATION: return hmR0VmxExitEptViolation(pVCpu, pVmxTransient);
13078	case VMX_EXIT_XCPT_OR_NMI: return hmR0VmxExitXcptOrNmiNested(pVCpu, pVmxTransient);
13079	case VMX_EXIT_IO_INSTR: return hmR0VmxExitIoInstrNested(pVCpu, pVmxTransient);
13080	case VMX_EXIT_HLT: return hmR0VmxExitHltNested(pVCpu, pVmxTransient);
13081
13082	/*
13083	* We shouldn't direct host physical interrupts to the nested-guest.
13084	*/
13085	case VMX_EXIT_EXT_INT:
13086	return hmR0VmxExitExtInt(pVCpu, pVmxTransient);
13087
13088	/*
13089	* Instructions that cause VM-exits unconditionally or the condition is
13090	* always is taken solely from the nested hypervisor (meaning if the VM-exit
13091	* happens, it's guaranteed to be a nested-guest VM-exit).
13092	*
13093	* - Provides VM-exit instruction length ONLY.
13094	*/
13095	case VMX_EXIT_CPUID: /* Unconditional. */
13096	case VMX_EXIT_VMCALL:
13097	case VMX_EXIT_GETSEC:
13098	case VMX_EXIT_INVD:
13099	case VMX_EXIT_XSETBV:
13100	case VMX_EXIT_VMLAUNCH:
13101	case VMX_EXIT_VMRESUME:
13102	case VMX_EXIT_VMXOFF:
13103	case VMX_EXIT_ENCLS: /* Condition specified solely by nested hypervisor. */
13104	case VMX_EXIT_VMFUNC:
13105	return hmR0VmxExitInstrNested(pVCpu, pVmxTransient);
13106
13107	/*
13108	* Instructions that cause VM-exits unconditionally or the condition is
13109	* always is taken solely from the nested hypervisor (meaning if the VM-exit
13110	* happens, it's guaranteed to be a nested-guest VM-exit).
13111	*
13112	* - Provides VM-exit instruction length.
13113	* - Provides VM-exit information.
13114	* - Optionally provides Exit qualification.
13115	*
13116	* Since Exit qualification is 0 for all VM-exits where it is not
13117	* applicable, reading and passing it to the guest should produce
13118	* defined behavior.
13119	*
13120	* See Intel spec. 27.2.1 "Basic VM-Exit Information".
13121	*/
13122	case VMX_EXIT_INVEPT: /* Unconditional. */
13123	case VMX_EXIT_INVVPID:
13124	case VMX_EXIT_VMCLEAR:
13125	case VMX_EXIT_VMPTRLD:
13126	case VMX_EXIT_VMPTRST:
13127	case VMX_EXIT_VMXON:
13128	case VMX_EXIT_GDTR_IDTR_ACCESS: /* Condition specified solely by nested hypervisor. */
13129	case VMX_EXIT_LDTR_TR_ACCESS:
13130	case VMX_EXIT_RDRAND:
13131	case VMX_EXIT_RDSEED:
13132	case VMX_EXIT_XSAVES:
13133	case VMX_EXIT_XRSTORS:
13134	case VMX_EXIT_UMWAIT:
13135	case VMX_EXIT_TPAUSE:
13136	return hmR0VmxExitInstrWithInfoNested(pVCpu, pVmxTransient);
13137
13138	case VMX_EXIT_RDTSC: return hmR0VmxExitRdtscNested(pVCpu, pVmxTransient);
13139	case VMX_EXIT_RDTSCP: return hmR0VmxExitRdtscpNested(pVCpu, pVmxTransient);
13140	case VMX_EXIT_RDMSR: return hmR0VmxExitRdmsrNested(pVCpu, pVmxTransient);
13141	case VMX_EXIT_WRMSR: return hmR0VmxExitWrmsrNested(pVCpu, pVmxTransient);
13142	case VMX_EXIT_INVLPG: return hmR0VmxExitInvlpgNested(pVCpu, pVmxTransient);
13143	case VMX_EXIT_INVPCID: return hmR0VmxExitInvpcidNested(pVCpu, pVmxTransient);
13144	case VMX_EXIT_TASK_SWITCH: return hmR0VmxExitTaskSwitchNested(pVCpu, pVmxTransient);
13145	case VMX_EXIT_WBINVD: return hmR0VmxExitWbinvdNested(pVCpu, pVmxTransient);
13146	case VMX_EXIT_MTF: return hmR0VmxExitMtfNested(pVCpu, pVmxTransient);
13147	case VMX_EXIT_APIC_ACCESS: return hmR0VmxExitApicAccessNested(pVCpu, pVmxTransient);
13148	case VMX_EXIT_APIC_WRITE: return hmR0VmxExitApicWriteNested(pVCpu, pVmxTransient);
13149	case VMX_EXIT_VIRTUALIZED_EOI: return hmR0VmxExitVirtEoiNested(pVCpu, pVmxTransient);
13150	case VMX_EXIT_MOV_CRX: return hmR0VmxExitMovCRxNested(pVCpu, pVmxTransient);
13151	case VMX_EXIT_INT_WINDOW: return hmR0VmxExitIntWindowNested(pVCpu, pVmxTransient);
13152	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindowNested(pVCpu, pVmxTransient);
13153	case VMX_EXIT_TPR_BELOW_THRESHOLD: return hmR0VmxExitTprBelowThresholdNested(pVCpu, pVmxTransient);
13154	case VMX_EXIT_MWAIT: return hmR0VmxExitMwaitNested(pVCpu, pVmxTransient);
13155	case VMX_EXIT_MONITOR: return hmR0VmxExitMonitorNested(pVCpu, pVmxTransient);
13156	case VMX_EXIT_PAUSE: return hmR0VmxExitPauseNested(pVCpu, pVmxTransient);
13157
13158	case VMX_EXIT_PREEMPT_TIMER:
13159	{
13160	/** @todo NSTVMX: Preempt timer. */
13161	return hmR0VmxExitPreemptTimer(pVCpu, pVmxTransient);
13162	}
13163
13164	case VMX_EXIT_MOV_DRX: return hmR0VmxExitMovDRxNested(pVCpu, pVmxTransient);
13165	case VMX_EXIT_RDPMC: return hmR0VmxExitRdpmcNested(pVCpu, pVmxTransient);
13166
13167	case VMX_EXIT_VMREAD:
13168	case VMX_EXIT_VMWRITE: return hmR0VmxExitVmreadVmwriteNested(pVCpu, pVmxTransient);
13169
13170	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFaultNested(pVCpu, pVmxTransient);
13171	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestStateNested(pVCpu, pVmxTransient);
13172
13173	case VMX_EXIT_INIT_SIGNAL:
13174	case VMX_EXIT_SIPI:
13175	case VMX_EXIT_IO_SMI:
13176	case VMX_EXIT_SMI:
13177	case VMX_EXIT_ERR_MSR_LOAD:
13178	case VMX_EXIT_ERR_MACHINE_CHECK:
13179	case VMX_EXIT_PML_FULL:
13180	case VMX_EXIT_RSM:
13181	default:
13182	return hmR0VmxExitErrUnexpected(pVCpu, pVmxTransient);
13183	}
13184	}
13185	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13186
13187
13188	/** @name VM-exit helpers.
13189	* @{
13190	*/
13191	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13192	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= VM-exit helpers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
13193	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13194
13195	/** Macro for VM-exits called unexpectedly. */
13196	#define HMVMX_UNEXPECTED_EXIT_RET(a_pVCpu, a_HmError) \
13197	do { \
13198	(a_pVCpu)->hm.s.u32HMError = (a_HmError); \
13199	return VERR_VMX_UNEXPECTED_EXIT; \
13200	} while (0)
13201
13202	#ifdef VBOX_STRICT
13203	/* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */
13204	# define HMVMX_ASSERT_PREEMPT_CPUID_VAR() \
13205	RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId()
13206
13207	# define HMVMX_ASSERT_PREEMPT_CPUID() \
13208	do { \
13209	RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \
13210	AssertMsg(idAssertCpu == idAssertCpuNow, ("VMX %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \
13211	} while (0)
13212
13213	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13214	do { \
13215	AssertPtr((a_pVCpu)); \
13216	AssertPtr((a_pVmxTransient)); \
13217	Assert((a_pVmxTransient)->fVMEntryFailed == false); \
13218	Assert((a_pVmxTransient)->pVmcsInfo); \
13219	Assert(ASMIntAreEnabled()); \
13220	HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu); \
13221	HMVMX_ASSERT_PREEMPT_CPUID_VAR(); \
13222	Log4Func(("vcpu[%RU32]\n", (a_pVCpu)->idCpu)); \
13223	HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu); \
13224	if (!VMMRZCallRing3IsEnabled((a_pVCpu))) \
13225	HMVMX_ASSERT_PREEMPT_CPUID(); \
13226	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
13227	} while (0)
13228
13229	# define HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13230	do { \
13231	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient); \
13232	Assert((a_pVmxTransient)->fIsNestedGuest); \
13233	} while (0)
13234
13235	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13236	do { \
13237	Log4Func(("\n")); \
13238	} while (0)
13239	#else
13240	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13241	do { \
13242	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
13243	NOREF((a_pVCpu)); NOREF((a_pVmxTransient)); \
13244	} while (0)
13245
13246	# define HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13247	do { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient); } while (0)
13248
13249	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) do { } while (0)
13250	#endif
13251
13252	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13253	/** Macro that does the necessary privilege checks and intercepted VM-exits for
13254	* guests that attempted to execute a VMX instruction. */
13255	# define HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(a_pVCpu, a_uExitReason) \
13256	do \
13257	{ \
13258	VBOXSTRICTRC rcStrictTmp = hmR0VmxCheckExitDueToVmxInstr((a_pVCpu), (a_uExitReason)); \
13259	if (rcStrictTmp == VINF_SUCCESS) \
13260	{ /* likely */ } \
13261	else if (rcStrictTmp == VINF_HM_PENDING_XCPT) \
13262	{ \
13263	Assert((a_pVCpu)->hm.s.Event.fPending); \
13264	Log4Func(("Privilege checks failed -> %#x\n", VMX_ENTRY_INT_INFO_VECTOR((a_pVCpu)->hm.s.Event.u64IntInfo))); \
13265	return VINF_SUCCESS; \
13266	} \
13267	else \
13268	{ \
13269	int rcTmp = VBOXSTRICTRC_VAL(rcStrictTmp); \
13270	AssertMsgFailedReturn(("Unexpected failure. rc=%Rrc", rcTmp), rcTmp); \
13271	} \
13272	} while (0)
13273
13274	/** Macro that decodes a memory operand for an VM-exit caused by an instruction. */
13275	# define HMVMX_DECODE_MEM_OPERAND(a_pVCpu, a_uExitInstrInfo, a_uExitQual, a_enmMemAccess, a_pGCPtrEffAddr) \
13276	do \
13277	{ \
13278	VBOXSTRICTRC rcStrictTmp = hmR0VmxDecodeMemOperand((a_pVCpu), (a_uExitInstrInfo), (a_uExitQual), (a_enmMemAccess), \
13279	(a_pGCPtrEffAddr)); \
13280	if (rcStrictTmp == VINF_SUCCESS) \
13281	{ /* likely */ } \
13282	else if (rcStrictTmp == VINF_HM_PENDING_XCPT) \
13283	{ \
13284	uint8_t const uXcptTmp = VMX_ENTRY_INT_INFO_VECTOR((a_pVCpu)->hm.s.Event.u64IntInfo); \
13285	Log4Func(("Memory operand decoding failed, raising xcpt %#x\n", uXcptTmp)); \
13286	NOREF(uXcptTmp); \
13287	return VINF_SUCCESS; \
13288	} \
13289	else \
13290	{ \
13291	Log4Func(("hmR0VmxDecodeMemOperand failed. rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrictTmp))); \
13292	return rcStrictTmp; \
13293	} \
13294	} while (0)
13295	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13296
13297
13298	/**
13299	* Advances the guest RIP by the specified number of bytes.
13300	*
13301	* @param pVCpu The cross context virtual CPU structure.
13302	* @param cbInstr Number of bytes to advance the RIP by.
13303	*
13304	* @remarks No-long-jump zone!!!
13305	*/
13306	DECLINLINE(void) hmR0VmxAdvanceGuestRipBy(PVMCPUCC pVCpu, uint32_t cbInstr)
13307	{
13308	/* Advance the RIP. */
13309	pVCpu->cpum.GstCtx.rip += cbInstr;
13310	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
13311
13312	/* Update interrupt inhibition. */
13313	if ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)
13314	&& pVCpu->cpum.GstCtx.rip != EMGetInhibitInterruptsPC(pVCpu))
13315	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
13316	}
13317
13318
13319	/**
13320	* Advances the guest RIP after reading it from the VMCS.
13321	*
13322	* @returns VBox status code, no informational status codes.
13323	* @param pVCpu The cross context virtual CPU structure.
13324	* @param pVmxTransient The VMX-transient structure.
13325	*
13326	* @remarks No-long-jump zone!!!
13327	*/
13328	static int hmR0VmxAdvanceGuestRip(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13329	{
13330	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13331	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS);
13332	AssertRCReturn(rc, rc);
13333
13334	hmR0VmxAdvanceGuestRipBy(pVCpu, pVmxTransient->cbExitInstr);
13335	return VINF_SUCCESS;
13336	}
13337
13338
13339	/**
13340	* Handle a condition that occurred while delivering an event through the guest or
13341	* nested-guest IDT.
13342	*
13343	* @returns Strict VBox status code (i.e. informational status codes too).
13344	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
13345	* @retval VINF_HM_DOUBLE_FAULT if a \#DF condition was detected and we ought
13346	* to continue execution of the guest which will delivery the \#DF.
13347	* @retval VINF_EM_RESET if we detected a triple-fault condition.
13348	* @retval VERR_EM_GUEST_CPU_HANG if we detected a guest CPU hang.
13349	*
13350	* @param pVCpu The cross context virtual CPU structure.
13351	* @param pVmxTransient The VMX-transient structure.
13352	*
13353	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13354	* Additionally, HMVMX_READ_EXIT_QUALIFICATION is required if the VM-exit
13355	* is due to an EPT violation, PML full or SPP-related event.
13356	*
13357	* @remarks No-long-jump zone!!!
13358	*/
13359	static VBOXSTRICTRC hmR0VmxCheckExitDueToEventDelivery(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13360	{
13361	Assert(!pVCpu->hm.s.Event.fPending);
13362	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_XCPT_INFO);
13363	if ( pVmxTransient->uExitReason == VMX_EXIT_EPT_VIOLATION
13364	\|\| pVmxTransient->uExitReason == VMX_EXIT_PML_FULL
13365	\|\| pVmxTransient->uExitReason == VMX_EXIT_SPP_EVENT)
13366	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_EXIT_QUALIFICATION);
13367
13368	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
13369	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
13370	uint32_t const uIdtVectorInfo = pVmxTransient->uIdtVectoringInfo;
13371	uint32_t const uExitIntInfo = pVmxTransient->uExitIntInfo;
13372	if (VMX_IDT_VECTORING_INFO_IS_VALID(uIdtVectorInfo))
13373	{
13374	uint32_t const uIdtVector = VMX_IDT_VECTORING_INFO_VECTOR(uIdtVectorInfo);
13375	uint32_t const uIdtVectorType = VMX_IDT_VECTORING_INFO_TYPE(uIdtVectorInfo);
13376
13377	/*
13378	* If the event was a software interrupt (generated with INT n) or a software exception
13379	* (generated by INT3/INTO) or a privileged software exception (generated by INT1), we
13380	* can handle the VM-exit and continue guest execution which will re-execute the
13381	* instruction rather than re-injecting the exception, as that can cause premature
13382	* trips to ring-3 before injection and involve TRPM which currently has no way of
13383	* storing that these exceptions were caused by these instructions (ICEBP's #DB poses
13384	* the problem).
13385	*/
13386	IEMXCPTRAISE enmRaise;
13387	IEMXCPTRAISEINFO fRaiseInfo;
13388	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
13389	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
13390	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT)
13391	{
13392	enmRaise = IEMXCPTRAISE_REEXEC_INSTR;
13393	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
13394	}
13395	else if (VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo))
13396	{
13397	uint32_t const uExitVectorType = VMX_EXIT_INT_INFO_TYPE(uExitIntInfo);
13398	uint8_t const uExitVector = VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo);
13399	Assert(uExitVectorType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT);
13400
13401	uint32_t const fIdtVectorFlags = hmR0VmxGetIemXcptFlags(uIdtVector, uIdtVectorType);
13402	uint32_t const fExitVectorFlags = hmR0VmxGetIemXcptFlags(uExitVector, uExitVectorType);
13403
13404	enmRaise = IEMEvaluateRecursiveXcpt(pVCpu, fIdtVectorFlags, uIdtVector, fExitVectorFlags, uExitVector, &fRaiseInfo);
13405
13406	/* Determine a vectoring #PF condition, see comment in hmR0VmxExitXcptPF(). */
13407	if (fRaiseInfo & (IEMXCPTRAISEINFO_EXT_INT_PF \| IEMXCPTRAISEINFO_NMI_PF))
13408	{
13409	pVmxTransient->fVectoringPF = true;
13410	enmRaise = IEMXCPTRAISE_PREV_EVENT;
13411	}
13412	}
13413	else
13414	{
13415	/*
13416	* If an exception or hardware interrupt delivery caused an EPT violation/misconfig or APIC access
13417	* VM-exit, then the VM-exit interruption-information will not be valid and we end up here.
13418	* It is sufficient to reflect the original event to the guest after handling the VM-exit.
13419	*/
13420	Assert( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
13421	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
13422	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT);
13423	enmRaise = IEMXCPTRAISE_PREV_EVENT;
13424	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
13425	}
13426
13427	/*
13428	* On CPUs that support Virtual NMIs, if this VM-exit (be it an exception or EPT violation/misconfig
13429	* etc.) occurred while delivering the NMI, we need to clear the block-by-NMI field in the guest
13430	* interruptibility-state before re-delivering the NMI after handling the VM-exit. Otherwise the
13431	* subsequent VM-entry would fail, see @bugref{7445}.
13432	*
13433	* See Intel spec. 30.7.1.2 "Resuming Guest Software after Handling an Exception".
13434	*/
13435	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
13436	&& enmRaise == IEMXCPTRAISE_PREV_EVENT
13437	&& (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
13438	&& CPUMIsGuestNmiBlocking(pVCpu))
13439	{
13440	CPUMSetGuestNmiBlocking(pVCpu, false);
13441	}
13442
13443	switch (enmRaise)
13444	{
13445	case IEMXCPTRAISE_CURRENT_XCPT:
13446	{
13447	Log4Func(("IDT: Pending secondary Xcpt: idtinfo=%#RX64 exitinfo=%#RX64\n", uIdtVectorInfo, uExitIntInfo));
13448	Assert(rcStrict == VINF_SUCCESS);
13449	break;
13450	}
13451
13452	case IEMXCPTRAISE_PREV_EVENT:
13453	{
13454	uint32_t u32ErrCode;
13455	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(uIdtVectorInfo))
13456	u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
13457	else
13458	u32ErrCode = 0;
13459
13460	/* If uExitVector is #PF, CR2 value will be updated from the VMCS if it's a guest #PF, see hmR0VmxExitXcptPF(). */
13461	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflect);
13462	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(uIdtVectorInfo), 0 /* cbInstr */,
13463	u32ErrCode, pVCpu->cpum.GstCtx.cr2);
13464
13465	Log4Func(("IDT: Pending vectoring event %#RX64 Err=%#RX32\n", pVCpu->hm.s.Event.u64IntInfo,
13466	pVCpu->hm.s.Event.u32ErrCode));
13467	Assert(rcStrict == VINF_SUCCESS);
13468	break;
13469	}
13470
13471	case IEMXCPTRAISE_REEXEC_INSTR:
13472	Assert(rcStrict == VINF_SUCCESS);
13473	break;
13474
13475	case IEMXCPTRAISE_DOUBLE_FAULT:
13476	{
13477	/*
13478	* Determing a vectoring double #PF condition. Used later, when PGM evaluates the
13479	* second #PF as a guest #PF (and not a shadow #PF) and needs to be converted into a #DF.
13480	*/
13481	if (fRaiseInfo & IEMXCPTRAISEINFO_PF_PF)
13482	{
13483	pVmxTransient->fVectoringDoublePF = true;
13484	Log4Func(("IDT: Vectoring double #PF %#RX64 cr2=%#RX64\n", pVCpu->hm.s.Event.u64IntInfo,
13485	pVCpu->cpum.GstCtx.cr2));
13486	rcStrict = VINF_SUCCESS;
13487	}
13488	else
13489	{
13490	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectConvertDF);
13491	hmR0VmxSetPendingXcptDF(pVCpu);
13492	Log4Func(("IDT: Pending vectoring #DF %#RX64 uIdtVector=%#x uExitVector=%#x\n", pVCpu->hm.s.Event.u64IntInfo,
13493	uIdtVector, VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo)));
13494	rcStrict = VINF_HM_DOUBLE_FAULT;
13495	}
13496	break;
13497	}
13498
13499	case IEMXCPTRAISE_TRIPLE_FAULT:
13500	{
13501	Log4Func(("IDT: Pending vectoring triple-fault uIdt=%#x uExit=%#x\n", uIdtVector,
13502	VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo)));
13503	rcStrict = VINF_EM_RESET;
13504	break;
13505	}
13506
13507	case IEMXCPTRAISE_CPU_HANG:
13508	{
13509	Log4Func(("IDT: Bad guest! Entering CPU hang. fRaiseInfo=%#x\n", fRaiseInfo));
13510	rcStrict = VERR_EM_GUEST_CPU_HANG;
13511	break;
13512	}
13513
13514	default:
13515	{
13516	AssertMsgFailed(("IDT: vcpu[%RU32] Unexpected/invalid value! enmRaise=%#x\n", pVCpu->idCpu, enmRaise));
13517	rcStrict = VERR_VMX_IPE_2;
13518	break;
13519	}
13520	}
13521	}
13522	else if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
13523	&& !CPUMIsGuestNmiBlocking(pVCpu))
13524	{
13525	if ( VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo)
13526	&& VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo) != X86_XCPT_DF
13527	&& VMX_EXIT_INT_INFO_IS_NMI_UNBLOCK_IRET(uExitIntInfo))
13528	{
13529	/*
13530	* Execution of IRET caused a fault when NMI blocking was in effect (i.e we're in
13531	* the guest or nested-guest NMI handler). We need to set the block-by-NMI field so
13532	* that virtual NMIs remain blocked until the IRET execution is completed.
13533	*
13534	* See Intel spec. 31.7.1.2 "Resuming Guest Software After Handling An Exception".
13535	*/
13536	CPUMSetGuestNmiBlocking(pVCpu, true);
13537	Log4Func(("Set NMI blocking. uExitReason=%u\n", pVmxTransient->uExitReason));
13538	}
13539	else if ( pVmxTransient->uExitReason == VMX_EXIT_EPT_VIOLATION
13540	\|\| pVmxTransient->uExitReason == VMX_EXIT_PML_FULL
13541	\|\| pVmxTransient->uExitReason == VMX_EXIT_SPP_EVENT)
13542	{
13543	/*
13544	* Execution of IRET caused an EPT violation, page-modification log-full event or
13545	* SPP-related event VM-exit when NMI blocking was in effect (i.e. we're in the
13546	* guest or nested-guest NMI handler). We need to set the block-by-NMI field so
13547	* that virtual NMIs remain blocked until the IRET execution is completed.
13548	*
13549	* See Intel spec. 27.2.3 "Information about NMI unblocking due to IRET"
13550	*/
13551	if (VMX_EXIT_QUAL_EPT_IS_NMI_UNBLOCK_IRET(pVmxTransient->uExitQual))
13552	{
13553	CPUMSetGuestNmiBlocking(pVCpu, true);
13554	Log4Func(("Set NMI blocking. uExitReason=%u\n", pVmxTransient->uExitReason));
13555	}
13556	}
13557	}
13558
13559	Assert( rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_HM_DOUBLE_FAULT
13560	\|\| rcStrict == VINF_EM_RESET \|\| rcStrict == VERR_EM_GUEST_CPU_HANG);
13561	return rcStrict;
13562	}
13563
13564
13565	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13566	/**
13567	* Perform the relevant VMX instruction checks for VM-exits that occurred due to the
13568	* guest attempting to execute a VMX instruction.
13569	*
13570	* @returns Strict VBox status code (i.e. informational status codes too).
13571	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
13572	* @retval VINF_HM_PENDING_XCPT if an exception was raised.
13573	*
13574	* @param pVCpu The cross context virtual CPU structure.
13575	* @param uExitReason The VM-exit reason.
13576	*
13577	* @todo NSTVMX: Document other error codes when VM-exit is implemented.
13578	* @remarks No-long-jump zone!!!
13579	*/
13580	static VBOXSTRICTRC hmR0VmxCheckExitDueToVmxInstr(PVMCPUCC pVCpu, uint32_t uExitReason)
13581	{
13582	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS
13583	\| CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_EFER);
13584
13585	/*
13586	* The physical CPU would have already checked the CPU mode/code segment.
13587	* We shall just assert here for paranoia.
13588	* See Intel spec. 25.1.1 "Relative Priority of Faults and VM Exits".
13589	*/
13590	Assert(!CPUMIsGuestInRealOrV86ModeEx(&pVCpu->cpum.GstCtx));
13591	Assert( !CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx)
13592	\|\| CPUMIsGuestIn64BitCodeEx(&pVCpu->cpum.GstCtx));
13593
13594	if (uExitReason == VMX_EXIT_VMXON)
13595	{
13596	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
13597
13598	/*
13599	* We check CR4.VMXE because it is required to be always set while in VMX operation
13600	* by physical CPUs and our CR4 read-shadow is only consulted when executing specific
13601	* instructions (CLTS, LMSW, MOV CR, and SMSW) and thus doesn't affect CPU operation
13602	* otherwise (i.e. physical CPU won't automatically #UD if Cr4Shadow.VMXE is 0).
13603	*/
13604	if (!CPUMIsGuestVmxEnabled(&pVCpu->cpum.GstCtx))
13605	{
13606	Log4Func(("CR4.VMXE is not set -> #UD\n"));
13607	hmR0VmxSetPendingXcptUD(pVCpu);
13608	return VINF_HM_PENDING_XCPT;
13609	}
13610	}
13611	else if (!CPUMIsGuestInVmxRootMode(&pVCpu->cpum.GstCtx))
13612	{
13613	/*
13614	* The guest has not entered VMX operation but attempted to execute a VMX instruction
13615	* (other than VMXON), we need to raise a #UD.
13616	*/
13617	Log4Func(("Not in VMX root mode -> #UD\n"));
13618	hmR0VmxSetPendingXcptUD(pVCpu);
13619	return VINF_HM_PENDING_XCPT;
13620	}
13621
13622	/* All other checks (including VM-exit intercepts) are handled by IEM instruction emulation. */
13623	return VINF_SUCCESS;
13624	}
13625
13626
13627	/**
13628	* Decodes the memory operand of an instruction that caused a VM-exit.
13629	*
13630	* The Exit qualification field provides the displacement field for memory
13631	* operand instructions, if any.
13632	*
13633	* @returns Strict VBox status code (i.e. informational status codes too).
13634	* @retval VINF_SUCCESS if the operand was successfully decoded.
13635	* @retval VINF_HM_PENDING_XCPT if an exception was raised while decoding the
13636	* operand.
13637	* @param pVCpu The cross context virtual CPU structure.
13638	* @param uExitInstrInfo The VM-exit instruction information field.
13639	* @param enmMemAccess The memory operand's access type (read or write).
13640	* @param GCPtrDisp The instruction displacement field, if any. For
13641	* RIP-relative addressing pass RIP + displacement here.
13642	* @param pGCPtrMem Where to store the effective destination memory address.
13643	*
13644	* @remarks Warning! This function ASSUMES the instruction cannot be used in real or
13645	* virtual-8086 mode hence skips those checks while verifying if the
13646	* segment is valid.
13647	*/
13648	static VBOXSTRICTRC hmR0VmxDecodeMemOperand(PVMCPUCC pVCpu, uint32_t uExitInstrInfo, RTGCPTR GCPtrDisp, VMXMEMACCESS enmMemAccess,
13649	PRTGCPTR pGCPtrMem)
13650	{
13651	Assert(pGCPtrMem);
13652	Assert(!CPUMIsGuestInRealOrV86Mode(pVCpu));
13653	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_EFER
13654	\| CPUMCTX_EXTRN_CR0);
13655
13656	static uint64_t const s_auAddrSizeMasks[] = { UINT64_C(0xffff), UINT64_C(0xffffffff), UINT64_C(0xffffffffffffffff) };
13657	static uint64_t const s_auAccessSizeMasks[] = { sizeof(uint16_t), sizeof(uint32_t), sizeof(uint64_t) };
13658	AssertCompile(RT_ELEMENTS(s_auAccessSizeMasks) == RT_ELEMENTS(s_auAddrSizeMasks));
13659
13660	VMXEXITINSTRINFO ExitInstrInfo;
13661	ExitInstrInfo.u = uExitInstrInfo;
13662	uint8_t const uAddrSize = ExitInstrInfo.All.u3AddrSize;
13663	uint8_t const iSegReg = ExitInstrInfo.All.iSegReg;
13664	bool const fIdxRegValid = !ExitInstrInfo.All.fIdxRegInvalid;
13665	uint8_t const iIdxReg = ExitInstrInfo.All.iIdxReg;
13666	uint8_t const uScale = ExitInstrInfo.All.u2Scaling;
13667	bool const fBaseRegValid = !ExitInstrInfo.All.fBaseRegInvalid;
13668	uint8_t const iBaseReg = ExitInstrInfo.All.iBaseReg;
13669	bool const fIsMemOperand = !ExitInstrInfo.All.fIsRegOperand;
13670	bool const fIsLongMode = CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx);
13671
13672	/*
13673	* Validate instruction information.
13674	* This shouldn't happen on real hardware but useful while testing our nested hardware-virtualization code.
13675	*/
13676	AssertLogRelMsgReturn(uAddrSize < RT_ELEMENTS(s_auAddrSizeMasks),
13677	("Invalid address size. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_1);
13678	AssertLogRelMsgReturn(iSegReg < X86_SREG_COUNT,
13679	("Invalid segment register. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_2);
13680	AssertLogRelMsgReturn(fIsMemOperand,
13681	("Expected memory operand. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_3);
13682
13683	/*
13684	* Compute the complete effective address.
13685	*
13686	* See AMD instruction spec. 1.4.2 "SIB Byte Format"
13687	* See AMD spec. 4.5.2 "Segment Registers".
13688	*/
13689	RTGCPTR GCPtrMem = GCPtrDisp;
13690	if (fBaseRegValid)
13691	GCPtrMem += pVCpu->cpum.GstCtx.aGRegs[iBaseReg].u64;
13692	if (fIdxRegValid)
13693	GCPtrMem += pVCpu->cpum.GstCtx.aGRegs[iIdxReg].u64 << uScale;
13694
13695	RTGCPTR const GCPtrOff = GCPtrMem;
13696	if ( !fIsLongMode
13697	\|\| iSegReg >= X86_SREG_FS)
13698	GCPtrMem += pVCpu->cpum.GstCtx.aSRegs[iSegReg].u64Base;
13699	GCPtrMem &= s_auAddrSizeMasks[uAddrSize];
13700
13701	/*
13702	* Validate effective address.
13703	* See AMD spec. 4.5.3 "Segment Registers in 64-Bit Mode".
13704	*/
13705	uint8_t const cbAccess = s_auAccessSizeMasks[uAddrSize];
13706	Assert(cbAccess > 0);
13707	if (fIsLongMode)
13708	{
13709	if (X86_IS_CANONICAL(GCPtrMem))
13710	{
13711	*pGCPtrMem = GCPtrMem;
13712	return VINF_SUCCESS;
13713	}
13714
13715	/** @todo r=ramshankar: We should probably raise \#SS or \#GP. See AMD spec. 4.12.2
13716	* "Data Limit Checks in 64-bit Mode". */
13717	Log4Func(("Long mode effective address is not canonical GCPtrMem=%#RX64\n", GCPtrMem));
13718	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13719	return VINF_HM_PENDING_XCPT;
13720	}
13721
13722	/*
13723	* This is a watered down version of iemMemApplySegment().
13724	* Parts that are not applicable for VMX instructions like real-or-v8086 mode
13725	* and segment CPL/DPL checks are skipped.
13726	*/
13727	RTGCPTR32 const GCPtrFirst32 = (RTGCPTR32)GCPtrOff;
13728	RTGCPTR32 const GCPtrLast32 = GCPtrFirst32 + cbAccess - 1;
13729	PCCPUMSELREG pSel = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
13730
13731	/* Check if the segment is present and usable. */
13732	if ( pSel->Attr.n.u1Present
13733	&& !pSel->Attr.n.u1Unusable)
13734	{
13735	Assert(pSel->Attr.n.u1DescType);
13736	if (!(pSel->Attr.n.u4Type & X86_SEL_TYPE_CODE))
13737	{
13738	/* Check permissions for the data segment. */
13739	if ( enmMemAccess == VMXMEMACCESS_WRITE
13740	&& !(pSel->Attr.n.u4Type & X86_SEL_TYPE_WRITE))
13741	{
13742	Log4Func(("Data segment access invalid. iSegReg=%#x Attr=%#RX32\n", iSegReg, pSel->Attr.u));
13743	hmR0VmxSetPendingXcptGP(pVCpu, iSegReg);
13744	return VINF_HM_PENDING_XCPT;
13745	}
13746
13747	/* Check limits if it's a normal data segment. */
13748	if (!(pSel->Attr.n.u4Type & X86_SEL_TYPE_DOWN))
13749	{
13750	if ( GCPtrFirst32 > pSel->u32Limit
13751	\|\| GCPtrLast32 > pSel->u32Limit)
13752	{
13753	Log4Func(("Data segment limit exceeded. "
13754	"iSegReg=%#x GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n", iSegReg, GCPtrFirst32,
13755	GCPtrLast32, pSel->u32Limit));
13756	if (iSegReg == X86_SREG_SS)
13757	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13758	else
13759	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13760	return VINF_HM_PENDING_XCPT;
13761	}
13762	}
13763	else
13764	{
13765	/* Check limits if it's an expand-down data segment.
13766	Note! The upper boundary is defined by the B bit, not the G bit! */
13767	if ( GCPtrFirst32 < pSel->u32Limit + UINT32_C(1)
13768	\|\| GCPtrLast32 > (pSel->Attr.n.u1DefBig ? UINT32_MAX : UINT32_C(0xffff)))
13769	{
13770	Log4Func(("Expand-down data segment limit exceeded. "
13771	"iSegReg=%#x GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n", iSegReg, GCPtrFirst32,
13772	GCPtrLast32, pSel->u32Limit));
13773	if (iSegReg == X86_SREG_SS)
13774	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13775	else
13776	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13777	return VINF_HM_PENDING_XCPT;
13778	}
13779	}
13780	}
13781	else
13782	{
13783	/* Check permissions for the code segment. */
13784	if ( enmMemAccess == VMXMEMACCESS_WRITE
13785	\|\| ( enmMemAccess == VMXMEMACCESS_READ
13786	&& !(pSel->Attr.n.u4Type & X86_SEL_TYPE_READ)))
13787	{
13788	Log4Func(("Code segment access invalid. Attr=%#RX32\n", pSel->Attr.u));
13789	Assert(!CPUMIsGuestInRealOrV86ModeEx(&pVCpu->cpum.GstCtx));
13790	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13791	return VINF_HM_PENDING_XCPT;
13792	}
13793
13794	/* Check limits for the code segment (normal/expand-down not applicable for code segments). */
13795	if ( GCPtrFirst32 > pSel->u32Limit
13796	\|\| GCPtrLast32 > pSel->u32Limit)
13797	{
13798	Log4Func(("Code segment limit exceeded. GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n",
13799	GCPtrFirst32, GCPtrLast32, pSel->u32Limit));
13800	if (iSegReg == X86_SREG_SS)
13801	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13802	else
13803	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13804	return VINF_HM_PENDING_XCPT;
13805	}
13806	}
13807	}
13808	else
13809	{
13810	Log4Func(("Not present or unusable segment. iSegReg=%#x Attr=%#RX32\n", iSegReg, pSel->Attr.u));
13811	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13812	return VINF_HM_PENDING_XCPT;
13813	}
13814
13815	*pGCPtrMem = GCPtrMem;
13816	return VINF_SUCCESS;
13817	}
13818	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13819
13820
13821	/**
13822	* VM-exit helper for LMSW.
13823	*/
13824	static VBOXSTRICTRC hmR0VmxExitLmsw(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint16_t uMsw, RTGCPTR GCPtrEffDst)
13825	{
13826	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13827	AssertRCReturn(rc, rc);
13828
13829	VBOXSTRICTRC rcStrict = IEMExecDecodedLmsw(pVCpu, cbInstr, uMsw, GCPtrEffDst);
13830	AssertMsg( rcStrict == VINF_SUCCESS
13831	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13832
13833	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
13834	if (rcStrict == VINF_IEM_RAISED_XCPT)
13835	{
13836	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13837	rcStrict = VINF_SUCCESS;
13838	}
13839
13840	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitLmsw);
13841	Log4Func(("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13842	return rcStrict;
13843	}
13844
13845
13846	/**
13847	* VM-exit helper for CLTS.
13848	*/
13849	static VBOXSTRICTRC hmR0VmxExitClts(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr)
13850	{
13851	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13852	AssertRCReturn(rc, rc);
13853
13854	VBOXSTRICTRC rcStrict = IEMExecDecodedClts(pVCpu, cbInstr);
13855	AssertMsg( rcStrict == VINF_SUCCESS
13856	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13857
13858	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
13859	if (rcStrict == VINF_IEM_RAISED_XCPT)
13860	{
13861	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13862	rcStrict = VINF_SUCCESS;
13863	}
13864
13865	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitClts);
13866	Log4Func(("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13867	return rcStrict;
13868	}
13869
13870
13871	/**
13872	* VM-exit helper for MOV from CRx (CRx read).
13873	*/
13874	static VBOXSTRICTRC hmR0VmxExitMovFromCrX(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint8_t iGReg, uint8_t iCrReg)
13875	{
13876	Assert(iCrReg < 16);
13877	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
13878
13879	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13880	AssertRCReturn(rc, rc);
13881
13882	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxRead(pVCpu, cbInstr, iGReg, iCrReg);
13883	AssertMsg( rcStrict == VINF_SUCCESS
13884	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13885
13886	if (iGReg == X86_GREG_xSP)
13887	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_RSP);
13888	else
13889	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
13890	#ifdef VBOX_WITH_STATISTICS
13891	switch (iCrReg)
13892	{
13893	case 0: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Read); break;
13894	case 2: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Read); break;
13895	case 3: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Read); break;
13896	case 4: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Read); break;
13897	case 8: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Read); break;
13898	}
13899	#endif
13900	Log4Func(("CR%d Read access rcStrict=%Rrc\n", iCrReg, VBOXSTRICTRC_VAL(rcStrict)));
13901	return rcStrict;
13902	}
13903
13904
13905	/**
13906	* VM-exit helper for MOV to CRx (CRx write).
13907	*/
13908	static VBOXSTRICTRC hmR0VmxExitMovToCrX(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iGReg, uint8_t iCrReg)
13909	{
13910	HMVMX_CPUMCTX_ASSERT(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
13911
13912	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxWrite(pVCpu, cbInstr, iCrReg, iGReg);
13913	AssertMsg( rcStrict == VINF_SUCCESS
13914	\|\| rcStrict == VINF_IEM_RAISED_XCPT
13915	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13916
13917	switch (iCrReg)
13918	{
13919	case 0:
13920	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0
13921	\| HM_CHANGED_GUEST_EFER_MSR \| HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
13922	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Write);
13923	Log4Func(("CR0 write. rcStrict=%Rrc CR0=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cr0));
13924	break;
13925
13926	case 2:
13927	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Write);
13928	/* Nothing to do here, CR2 it's not part of the VMCS. */
13929	break;
13930
13931	case 3:
13932	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR3);
13933	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Write);
13934	Log4Func(("CR3 write. rcStrict=%Rrc CR3=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cr3));
13935	break;
13936
13937	case 4:
13938	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR4);
13939	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Write);
13940	Log4Func(("CR4 write. rc=%Rrc CR4=%#RX64 fLoadSaveGuestXcr0=%u\n", VBOXSTRICTRC_VAL(rcStrict),
13941	pVCpu->cpum.GstCtx.cr4, pVCpu->hmr0.s.fLoadSaveGuestXcr0));
13942	break;
13943
13944	case 8:
13945	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged,
13946	HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_APIC_TPR);
13947	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Write);
13948	break;
13949
13950	default:
13951	AssertMsgFailed(("Invalid CRx register %#x\n", iCrReg));
13952	break;
13953	}
13954
13955	if (rcStrict == VINF_IEM_RAISED_XCPT)
13956	{
13957	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13958	rcStrict = VINF_SUCCESS;
13959	}
13960	return rcStrict;
13961	}
13962
13963
13964	/**
13965	* VM-exit exception handler for \#PF (Page-fault exception).
13966	*
13967	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13968	*/
13969	static VBOXSTRICTRC hmR0VmxExitXcptPF(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13970	{
13971	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13972	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
13973	hmR0VmxReadExitQualVmcs(pVmxTransient);
13974
13975	if (!pVM->hmr0.s.fNestedPaging)
13976	{ /* likely */ }
13977	else
13978	{
13979	#if !defined(HMVMX_ALWAYS_TRAP_ALL_XCPTS) && !defined(HMVMX_ALWAYS_TRAP_PF)
13980	Assert(pVmxTransient->fIsNestedGuest \|\| pVCpu->hmr0.s.fUsingDebugLoop);
13981	#endif
13982	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
13983	if (!pVmxTransient->fVectoringDoublePF)
13984	{
13985	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */,
13986	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual);
13987	}
13988	else
13989	{
13990	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
13991	Assert(!pVmxTransient->fIsNestedGuest);
13992	hmR0VmxSetPendingXcptDF(pVCpu);
13993	Log4Func(("Pending #DF due to vectoring #PF w/ NestedPaging\n"));
13994	}
13995	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
13996	return VINF_SUCCESS;
13997	}
13998
13999	Assert(!pVmxTransient->fIsNestedGuest);
14000
14001	/* If it's a vectoring #PF, emulate injecting the original event injection as PGMTrap0eHandler() is incapable
14002	of differentiating between instruction emulation and event injection that caused a #PF. See @bugref{6607}. */
14003	if (pVmxTransient->fVectoringPF)
14004	{
14005	Assert(pVCpu->hm.s.Event.fPending);
14006	return VINF_EM_RAW_INJECT_TRPM_EVENT;
14007	}
14008
14009	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14010	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14011	AssertRCReturn(rc, rc);
14012
14013	Log4Func(("#PF: cs:rip=%#04x:%#RX64 err_code=%#RX32 exit_qual=%#RX64 cr3=%#RX64\n", pCtx->cs.Sel, pCtx->rip,
14014	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual, pCtx->cr3));
14015
14016	TRPMAssertXcptPF(pVCpu, pVmxTransient->uExitQual, (RTGCUINT)pVmxTransient->uExitIntErrorCode);
14017	rc = PGMTrap0eHandler(pVCpu, pVmxTransient->uExitIntErrorCode, CPUMCTX2CORE(pCtx), (RTGCPTR)pVmxTransient->uExitQual);
14018
14019	Log4Func(("#PF: rc=%Rrc\n", rc));
14020	if (rc == VINF_SUCCESS)
14021	{
14022	/*
14023	* This is typically a shadow page table sync or a MMIO instruction. But we may have
14024	* emulated something like LTR or a far jump. Any part of the CPU context may have changed.
14025	*/
14026	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14027	TRPMResetTrap(pVCpu);
14028	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
14029	return rc;
14030	}
14031
14032	if (rc == VINF_EM_RAW_GUEST_TRAP)
14033	{
14034	if (!pVmxTransient->fVectoringDoublePF)
14035	{
14036	/* It's a guest page fault and needs to be reflected to the guest. */
14037	uint32_t const uGstErrorCode = TRPMGetErrorCode(pVCpu);
14038	TRPMResetTrap(pVCpu);
14039	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory #PF. */
14040	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */,
14041	uGstErrorCode, pVmxTransient->uExitQual);
14042	}
14043	else
14044	{
14045	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
14046	TRPMResetTrap(pVCpu);
14047	pVCpu->hm.s.Event.fPending = false; /* Clear pending #PF to replace it with #DF. */
14048	hmR0VmxSetPendingXcptDF(pVCpu);
14049	Log4Func(("#PF: Pending #DF due to vectoring #PF\n"));
14050	}
14051
14052	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
14053	return VINF_SUCCESS;
14054	}
14055
14056	TRPMResetTrap(pVCpu);
14057	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
14058	return rc;
14059	}
14060
14061
14062	/**
14063	* VM-exit exception handler for \#MF (Math Fault: floating point exception).
14064	*
14065	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14066	*/
14067	static VBOXSTRICTRC hmR0VmxExitXcptMF(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14068	{
14069	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14070	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
14071
14072	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR0);
14073	AssertRCReturn(rc, rc);
14074
14075	if (!(pVCpu->cpum.GstCtx.cr0 & X86_CR0_NE))
14076	{
14077	/* Convert a #MF into a FERR -> IRQ 13. See @bugref{6117}. */
14078	rc = PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13, 1, 0 /* uTagSrc */);
14079
14080	/** @todo r=ramshankar: The Intel spec. does -not- specify that this VM-exit
14081	* provides VM-exit instruction length. If this causes problem later,
14082	* disassemble the instruction like it's done on AMD-V. */
14083	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14084	AssertRCReturn(rc2, rc2);
14085	return rc;
14086	}
14087
14088	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbExitInstr,
14089	pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14090	return VINF_SUCCESS;
14091	}
14092
14093
14094	/**
14095	* VM-exit exception handler for \#BP (Breakpoint exception).
14096	*
14097	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14098	*/
14099	static VBOXSTRICTRC hmR0VmxExitXcptBP(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14100	{
14101	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14102	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP);
14103
14104	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14105	AssertRCReturn(rc, rc);
14106
14107	if (!pVmxTransient->fIsNestedGuest)
14108	rc = DBGFTrap03Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(&pVCpu->cpum.GstCtx));
14109	else
14110	rc = VINF_EM_RAW_GUEST_TRAP;
14111
14112	if (rc == VINF_EM_RAW_GUEST_TRAP)
14113	{
14114	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14115	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14116	rc = VINF_SUCCESS;
14117	}
14118
14119	Assert(rc == VINF_SUCCESS \|\| rc == VINF_EM_DBG_BREAKPOINT);
14120	return rc;
14121	}
14122
14123
14124	/**
14125	* VM-exit exception handler for \#AC (Alignment-check exception).
14126	*
14127	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14128	*/
14129	static VBOXSTRICTRC hmR0VmxExitXcptAC(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14130	{
14131	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14132
14133	/*
14134	* Detect #ACs caused by host having enabled split-lock detection.
14135	* Emulate such instructions.
14136	*/
14137	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo,
14138	CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS \| CPUMCTX_EXTRN_CS);
14139	AssertRCReturn(rc, rc);
14140	/** @todo detect split lock in cpu feature? */
14141	if ( /* 1. If 486-style alignment checks aren't enabled, then this must be a split-lock exception */
14142	!(pVCpu->cpum.GstCtx.cr0 & X86_CR0_AM)
14143	/* 2. #AC cannot happen in rings 0-2 except for split-lock detection. */
14144	\|\| CPUMGetGuestCPL(pVCpu) != 3
14145	/* 3. When the EFLAGS.AC != 0 this can only be a split-lock case. */
14146	\|\| !(pVCpu->cpum.GstCtx.eflags.u & X86_EFL_AC) )
14147	{
14148	/*
14149	* Check for debug/trace events and import state accordingly.
14150	*/
14151	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitGuestACSplitLock);
14152	PVMCC pVM = pVCpu->pVMR0;
14153	if ( !DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_VMX_SPLIT_LOCK)
14154	&& !VBOXVMM_VMX_SPLIT_LOCK_ENABLED())
14155	{
14156	if (pVM->cCpus == 1)
14157	{
14158	#if 0 /** @todo r=bird: This is potentially wrong. Might have to just do a whole state sync above and mark everything changed to be safe... */
14159	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14160	#else
14161	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14162	#endif
14163	AssertRCReturn(rc, rc);
14164	}
14165	}
14166	else
14167	{
14168	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14169	AssertRCReturn(rc, rc);
14170
14171	VBOXVMM_XCPT_DF(pVCpu, &pVCpu->cpum.GstCtx);
14172
14173	if (DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_VMX_SPLIT_LOCK))
14174	{
14175	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, DBGFEVENT_VMX_SPLIT_LOCK, DBGFEVENTCTX_HM, 0);
14176	if (rcStrict != VINF_SUCCESS)
14177	return rcStrict;
14178	}
14179	}
14180
14181	/*
14182	* Emulate the instruction.
14183	*
14184	* We have to ignore the LOCK prefix here as we must not retrigger the
14185	* detection on the host. This isn't all that satisfactory, though...
14186	*/
14187	if (pVM->cCpus == 1)
14188	{
14189	Log8Func(("cs:rip=%#04x:%#RX64 rflags=%#RX64 cr0=%#RX64 split-lock #AC\n", pVCpu->cpum.GstCtx.cs.Sel,
14190	pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.rflags, pVCpu->cpum.GstCtx.cr0));
14191
14192	/** @todo For SMP configs we should do a rendezvous here. */
14193	VBOXSTRICTRC rcStrict = IEMExecOneIgnoreLock(pVCpu);
14194	if (rcStrict == VINF_SUCCESS)
14195	#if 0 /** @todo r=bird: This is potentially wrong. Might have to just do a whole state sync above and mark everything changed to be safe... */
14196	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged,
14197	HM_CHANGED_GUEST_RIP
14198	\| HM_CHANGED_GUEST_RFLAGS
14199	\| HM_CHANGED_GUEST_GPRS_MASK
14200	\| HM_CHANGED_GUEST_CS
14201	\| HM_CHANGED_GUEST_SS);
14202	#else
14203	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14204	#endif
14205	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14206	{
14207	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14208	rcStrict = VINF_SUCCESS;
14209	}
14210	return rcStrict;
14211	}
14212	Log8Func(("cs:rip=%#04x:%#RX64 rflags=%#RX64 cr0=%#RX64 split-lock #AC -> VINF_EM_EMULATE_SPLIT_LOCK\n",
14213	pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.rflags, pVCpu->cpum.GstCtx.cr0));
14214	return VINF_EM_EMULATE_SPLIT_LOCK;
14215	}
14216
14217	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC);
14218	Log8Func(("cs:rip=%#04x:%#RX64 rflags=%#RX64 cr0=%#RX64 cpl=%d -> #AC\n", pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
14219	pVCpu->cpum.GstCtx.rflags, pVCpu->cpum.GstCtx.cr0, CPUMGetGuestCPL(pVCpu) ));
14220
14221	/* Re-inject it. We'll detect any nesting before getting here. */
14222	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14223	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14224	return VINF_SUCCESS;
14225	}
14226
14227
14228	/**
14229	* VM-exit exception handler for \#DB (Debug exception).
14230	*
14231	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14232	*/
14233	static VBOXSTRICTRC hmR0VmxExitXcptDB(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14234	{
14235	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14236	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
14237
14238	/*
14239	* Get the DR6-like values from the Exit qualification and pass it to DBGF for processing.
14240	*/
14241	hmR0VmxReadExitQualVmcs(pVmxTransient);
14242
14243	/* Refer Intel spec. Table 27-1. "Exit Qualifications for debug exceptions" for the format. */
14244	uint64_t const uDR6 = X86_DR6_INIT_VAL
14245	\| (pVmxTransient->uExitQual & ( X86_DR6_B0 \| X86_DR6_B1 \| X86_DR6_B2 \| X86_DR6_B3
14246	\| X86_DR6_BD \| X86_DR6_BS));
14247
14248	int rc;
14249	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14250	if (!pVmxTransient->fIsNestedGuest)
14251	{
14252	rc = DBGFTrap01Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx), uDR6, pVCpu->hm.s.fSingleInstruction);
14253
14254	/*
14255	* Prevents stepping twice over the same instruction when the guest is stepping using
14256	* EFLAGS.TF and the hypervisor debugger is stepping using MTF.
14257	* Testcase: DOSQEMM, break (using "ba x 1") at cs:rip 0x70:0x774 and step (using "t").
14258	*/
14259	if ( rc == VINF_EM_DBG_STEPPED
14260	&& (pVmxTransient->pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MONITOR_TRAP_FLAG))
14261	{
14262	Assert(pVCpu->hm.s.fSingleInstruction);
14263	rc = VINF_EM_RAW_GUEST_TRAP;
14264	}
14265	}
14266	else
14267	rc = VINF_EM_RAW_GUEST_TRAP;
14268	Log6Func(("rc=%Rrc\n", rc));
14269	if (rc == VINF_EM_RAW_GUEST_TRAP)
14270	{
14271	/*
14272	* The exception was for the guest. Update DR6, DR7.GD and
14273	* IA32_DEBUGCTL.LBR before forwarding it.
14274	* See Intel spec. 27.1 "Architectural State before a VM-Exit".
14275	*/
14276	VMMRZCallRing3Disable(pVCpu);
14277	HM_DISABLE_PREEMPT(pVCpu);
14278
14279	pCtx->dr[6] &= ~X86_DR6_B_MASK;
14280	pCtx->dr[6] \|= uDR6;
14281	if (CPUMIsGuestDebugStateActive(pVCpu))
14282	ASMSetDR6(pCtx->dr[6]);
14283
14284	HM_RESTORE_PREEMPT();
14285	VMMRZCallRing3Enable(pVCpu);
14286
14287	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_DR7);
14288	AssertRCReturn(rc, rc);
14289
14290	/* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */
14291	pCtx->dr[7] &= ~(uint64_t)X86_DR7_GD;
14292
14293	/* Paranoia. */
14294	pCtx->dr[7] &= ~(uint64_t)X86_DR7_RAZ_MASK;
14295	pCtx->dr[7] \|= X86_DR7_RA1_MASK;
14296
14297	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, pCtx->dr[7]);
14298	AssertRC(rc);
14299
14300	/*
14301	* Raise #DB in the guest.
14302	*
14303	* It is important to reflect exactly what the VM-exit gave us (preserving the
14304	* interruption-type) rather than use hmR0VmxSetPendingXcptDB() as the #DB could've
14305	* been raised while executing ICEBP (INT1) and not the regular #DB. Thus it may
14306	* trigger different handling in the CPU (like skipping DPL checks), see @bugref{6398}.
14307	*
14308	* Intel re-documented ICEBP/INT1 on May 2018 previously documented as part of
14309	* Intel 386, see Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
14310	*/
14311	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14312	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14313	return VINF_SUCCESS;
14314	}
14315
14316	/*
14317	* Not a guest trap, must be a hypervisor related debug event then.
14318	* Update DR6 in case someone is interested in it.
14319	*/
14320	AssertMsg(rc == VINF_EM_DBG_STEPPED \|\| rc == VINF_EM_DBG_BREAKPOINT, ("%Rrc\n", rc));
14321	AssertReturn(pVmxTransient->fWasHyperDebugStateActive, VERR_HM_IPE_5);
14322	CPUMSetHyperDR6(pVCpu, uDR6);
14323
14324	return rc;
14325	}
14326
14327
14328	/**
14329	* Hacks its way around the lovely mesa driver's backdoor accesses.
14330	*
14331	* @sa hmR0SvmHandleMesaDrvGp.
14332	*/
14333	static int hmR0VmxHandleMesaDrvGp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PCPUMCTX pCtx)
14334	{
14335	LogFunc(("cs:rip=%#04x:%#RX64 rcx=%#RX64 rbx=%#RX64\n", pCtx->cs.Sel, pCtx->rip, pCtx->rcx, pCtx->rbx));
14336	RT_NOREF(pCtx);
14337
14338	/* For now we'll just skip the instruction. */
14339	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14340	}
14341
14342
14343	/**
14344	* Checks if the \#GP'ing instruction is the mesa driver doing it's lovely
14345	* backdoor logging w/o checking what it is running inside.
14346	*
14347	* This recognizes an "IN EAX,DX" instruction executed in flat ring-3, with the
14348	* backdoor port and magic numbers loaded in registers.
14349	*
14350	* @returns true if it is, false if it isn't.
14351	* @sa hmR0SvmIsMesaDrvGp.
14352	*/
14353	DECLINLINE(bool) hmR0VmxIsMesaDrvGp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PCPUMCTX pCtx)
14354	{
14355	/* 0xed: IN eAX,dx */
14356	uint8_t abInstr[1];
14357	if (pVmxTransient->cbExitInstr != sizeof(abInstr))
14358	return false;
14359
14360	/* Check that it is #GP(0). */
14361	if (pVmxTransient->uExitIntErrorCode != 0)
14362	return false;
14363
14364	/* Check magic and port. */
14365	Assert(!(pCtx->fExtrn & (CPUMCTX_EXTRN_RAX \| CPUMCTX_EXTRN_RDX \| CPUMCTX_EXTRN_RCX)));
14366	/Log(("hmR0VmxIsMesaDrvGp: rax=%RX64 rdx=%RX64\n", pCtx->rax, pCtx->rdx));/
14367	if (pCtx->rax != UINT32_C(0x564d5868))
14368	return false;
14369	if (pCtx->dx != UINT32_C(0x5658))
14370	return false;
14371
14372	/* Flat ring-3 CS. */
14373	AssertCompile(HMVMX_CPUMCTX_EXTRN_ALL & CPUMCTX_EXTRN_CS);
14374	Assert(!(pCtx->fExtrn & CPUMCTX_EXTRN_CS));
14375	/Log(("hmR0VmxIsMesaDrvGp: cs.Attr.n.u2Dpl=%d base=%Rx64\n", pCtx->cs.Attr.n.u2Dpl, pCtx->cs.u64Base));/
14376	if (pCtx->cs.Attr.n.u2Dpl != 3)
14377	return false;
14378	if (pCtx->cs.u64Base != 0)
14379	return false;
14380
14381	/* Check opcode. */
14382	AssertCompile(HMVMX_CPUMCTX_EXTRN_ALL & CPUMCTX_EXTRN_RIP);
14383	Assert(!(pCtx->fExtrn & CPUMCTX_EXTRN_RIP));
14384	int rc = PGMPhysSimpleReadGCPtr(pVCpu, abInstr, pCtx->rip, sizeof(abInstr));
14385	/Log(("hmR0VmxIsMesaDrvGp: PGMPhysSimpleReadGCPtr -> %Rrc %#x\n", rc, abInstr[0]));/
14386	if (RT_FAILURE(rc))
14387	return false;
14388	if (abInstr[0] != 0xed)
14389	return false;
14390
14391	return true;
14392	}
14393
14394
14395	/**
14396	* VM-exit exception handler for \#GP (General-protection exception).
14397	*
14398	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14399	*/
14400	static VBOXSTRICTRC hmR0VmxExitXcptGP(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14401	{
14402	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14403	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP);
14404
14405	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14406	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14407	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
14408	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
14409	{ /* likely */ }
14410	else
14411	{
14412	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14413	Assert(pVCpu->hmr0.s.fUsingDebugLoop \|\| pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv \|\| pVmxTransient->fIsNestedGuest);
14414	#endif
14415	/*
14416	* If the guest is not in real-mode or we have unrestricted guest execution support, or if we are
14417	* executing a nested-guest, reflect #GP to the guest or nested-guest.
14418	*/
14419	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14420	AssertRCReturn(rc, rc);
14421	Log4Func(("Gst: cs:rip=%#04x:%#RX64 ErrorCode=%#x cr0=%#RX64 cpl=%u tr=%#04x\n", pCtx->cs.Sel, pCtx->rip,
14422	pVmxTransient->uExitIntErrorCode, pCtx->cr0, CPUMGetGuestCPL(pVCpu), pCtx->tr.Sel));
14423
14424	if ( pVmxTransient->fIsNestedGuest
14425	\|\| !pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv
14426	\|\| !hmR0VmxIsMesaDrvGp(pVCpu, pVmxTransient, pCtx))
14427	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14428	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14429	else
14430	rc = hmR0VmxHandleMesaDrvGp(pVCpu, pVmxTransient, pCtx);
14431	return rc;
14432	}
14433
14434	Assert(CPUMIsGuestInRealModeEx(pCtx));
14435	Assert(!pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUnrestrictedGuest);
14436	Assert(!pVmxTransient->fIsNestedGuest);
14437
14438	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14439	AssertRCReturn(rc, rc);
14440
14441	VBOXSTRICTRC rcStrict = IEMExecOne(pVCpu);
14442	if (rcStrict == VINF_SUCCESS)
14443	{
14444	if (!CPUMIsGuestInRealModeEx(pCtx))
14445	{
14446	/*
14447	* The guest is no longer in real-mode, check if we can continue executing the
14448	* guest using hardware-assisted VMX. Otherwise, fall back to emulation.
14449	*/
14450	pVmcsInfoShared->RealMode.fRealOnV86Active = false;
14451	if (HMCanExecuteVmxGuest(pVCpu->pVMR0, pVCpu, pCtx))
14452	{
14453	Log4Func(("Mode changed but guest still suitable for executing using hardware-assisted VMX\n"));
14454	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14455	}
14456	else
14457	{
14458	Log4Func(("Mode changed -> VINF_EM_RESCHEDULE\n"));
14459	rcStrict = VINF_EM_RESCHEDULE;
14460	}
14461	}
14462	else
14463	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14464	}
14465	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14466	{
14467	rcStrict = VINF_SUCCESS;
14468	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14469	}
14470	return VBOXSTRICTRC_VAL(rcStrict);
14471	}
14472
14473
14474	/**
14475	* VM-exit exception handler wrapper for all other exceptions that are not handled
14476	* by a specific handler.
14477	*
14478	* This simply re-injects the exception back into the VM without any special
14479	* processing.
14480	*
14481	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14482	*/
14483	static VBOXSTRICTRC hmR0VmxExitXcptOthers(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14484	{
14485	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14486
14487	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14488	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14489	AssertMsg(pVCpu->hmr0.s.fUsingDebugLoop \|\| pVmcsInfo->pShared->RealMode.fRealOnV86Active \|\| pVmxTransient->fIsNestedGuest,
14490	("uVector=%#x u32XcptBitmap=%#X32\n",
14491	VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo), pVmcsInfo->u32XcptBitmap));
14492	NOREF(pVmcsInfo);
14493	#endif
14494
14495	/*
14496	* Re-inject the exception into the guest. This cannot be a double-fault condition which
14497	* would have been handled while checking exits due to event delivery.
14498	*/
14499	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14500
14501	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14502	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
14503	AssertRCReturn(rc, rc);
14504	Log4Func(("Reinjecting Xcpt. uVector=%#x cs:rip=%#04x:%#RX64\n", uVector, pCtx->cs.Sel, pCtx->rip));
14505	#endif
14506
14507	#ifdef VBOX_WITH_STATISTICS
14508	switch (uVector)
14509	{
14510	case X86_XCPT_DE: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE); break;
14511	case X86_XCPT_DB: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB); break;
14512	case X86_XCPT_BP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP); break;
14513	case X86_XCPT_OF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
14514	case X86_XCPT_BR: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBR); break;
14515	case X86_XCPT_UD: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD); break;
14516	case X86_XCPT_NM: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
14517	case X86_XCPT_DF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDF); break;
14518	case X86_XCPT_TS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestTS); break;
14519	case X86_XCPT_NP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNP); break;
14520	case X86_XCPT_SS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestSS); break;
14521	case X86_XCPT_GP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP); break;
14522	case X86_XCPT_PF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF); break;
14523	case X86_XCPT_MF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF); break;
14524	case X86_XCPT_AC: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC); break;
14525	case X86_XCPT_XF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXF); break;
14526	default:
14527	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXcpUnk);
14528	break;
14529	}
14530	#endif
14531
14532	/* We should never call this function for a page-fault, we'd need to pass on the fault address below otherwise. */
14533	Assert(!VMX_EXIT_INT_INFO_IS_XCPT_PF(pVmxTransient->uExitIntInfo));
14534	NOREF(uVector);
14535
14536	/* Re-inject the original exception into the guest. */
14537	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14538	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14539	return VINF_SUCCESS;
14540	}
14541
14542
14543	/**
14544	* VM-exit exception handler for all exceptions (except NMIs!).
14545	*
14546	* @remarks This may be called for both guests and nested-guests. Take care to not
14547	* make assumptions and avoid doing anything that is not relevant when
14548	* executing a nested-guest (e.g., Mesa driver hacks).
14549	*/
14550	static VBOXSTRICTRC hmR0VmxExitXcpt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14551	{
14552	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_XCPT_INFO);
14553
14554	/*
14555	* If this VM-exit occurred while delivering an event through the guest IDT, take
14556	* action based on the return code and additional hints (e.g. for page-faults)
14557	* that will be updated in the VMX transient structure.
14558	*/
14559	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
14560	if (rcStrict == VINF_SUCCESS)
14561	{
14562	/*
14563	* If an exception caused a VM-exit due to delivery of an event, the original
14564	* event may have to be re-injected into the guest. We shall reinject it and
14565	* continue guest execution. However, page-fault is a complicated case and
14566	* needs additional processing done in hmR0VmxExitXcptPF().
14567	*/
14568	Assert(VMX_EXIT_INT_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
14569	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14570	if ( !pVCpu->hm.s.Event.fPending
14571	\|\| uVector == X86_XCPT_PF)
14572	{
14573	switch (uVector)
14574	{
14575	case X86_XCPT_PF: return hmR0VmxExitXcptPF(pVCpu, pVmxTransient);
14576	case X86_XCPT_GP: return hmR0VmxExitXcptGP(pVCpu, pVmxTransient);
14577	case X86_XCPT_MF: return hmR0VmxExitXcptMF(pVCpu, pVmxTransient);
14578	case X86_XCPT_DB: return hmR0VmxExitXcptDB(pVCpu, pVmxTransient);
14579	case X86_XCPT_BP: return hmR0VmxExitXcptBP(pVCpu, pVmxTransient);
14580	case X86_XCPT_AC: return hmR0VmxExitXcptAC(pVCpu, pVmxTransient);
14581	default:
14582	return hmR0VmxExitXcptOthers(pVCpu, pVmxTransient);
14583	}
14584	}
14585	/* else: inject pending event before resuming guest execution. */
14586	}
14587	else if (rcStrict == VINF_HM_DOUBLE_FAULT)
14588	{
14589	Assert(pVCpu->hm.s.Event.fPending);
14590	rcStrict = VINF_SUCCESS;
14591	}
14592
14593	return rcStrict;
14594	}
14595	/** @} */
14596
14597
14598	/** @name VM-exit handlers.
14599	* @{
14600	*/
14601	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
14602	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
14603	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
14604
14605	/**
14606	* VM-exit handler for external interrupts (VMX_EXIT_EXT_INT).
14607	*/
14608	HMVMX_EXIT_DECL hmR0VmxExitExtInt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14609	{
14610	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14611	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt);
14612	/* Windows hosts (32-bit and 64-bit) have DPC latency issues. See @bugref{6853}. */
14613	if (VMMR0ThreadCtxHookIsEnabled(pVCpu))
14614	return VINF_SUCCESS;
14615	return VINF_EM_RAW_INTERRUPT;
14616	}
14617
14618
14619	/**
14620	* VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI). Conditional
14621	* VM-exit.
14622	*/
14623	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmi(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14624	{
14625	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14626	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitXcptNmi, y3);
14627
14628	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
14629
14630	uint32_t const uExitIntType = VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo);
14631	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14632	Assert(VMX_EXIT_INT_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
14633
14634	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14635	Assert( !(pVmcsInfo->u32ExitCtls & VMX_EXIT_CTLS_ACK_EXT_INT)
14636	&& uExitIntType != VMX_EXIT_INT_INFO_TYPE_EXT_INT);
14637	NOREF(pVmcsInfo);
14638
14639	VBOXSTRICTRC rcStrict;
14640	switch (uExitIntType)
14641	{
14642	/*
14643	* Host physical NMIs:
14644	* This cannot be a guest NMI as the only way for the guest to receive an NMI is if we
14645	* injected it ourselves and anything we inject is not going to cause a VM-exit directly
14646	* for the event being injected[1]. Go ahead and dispatch the NMI to the host[2].
14647	*
14648	* See Intel spec. 27.2.3 "Information for VM Exits During Event Delivery".
14649	* See Intel spec. 27.5.5 "Updating Non-Register State".
14650	*/
14651	case VMX_EXIT_INT_INFO_TYPE_NMI:
14652	{
14653	rcStrict = hmR0VmxExitHostNmi(pVCpu, pVmcsInfo);
14654	break;
14655	}
14656
14657	/*
14658	* Privileged software exceptions (#DB from ICEBP),
14659	* Software exceptions (#BP and #OF),
14660	* Hardware exceptions:
14661	* Process the required exceptions and resume guest execution if possible.
14662	*/
14663	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
14664	Assert(uVector == X86_XCPT_DB);
14665	RT_FALL_THRU();
14666	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
14667	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF \|\| uExitIntType == VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT);
14668	RT_FALL_THRU();
14669	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
14670	{
14671	NOREF(uVector);
14672	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
14673	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14674	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
14675	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
14676
14677	rcStrict = hmR0VmxExitXcpt(pVCpu, pVmxTransient);
14678	break;
14679	}
14680
14681	default:
14682	{
14683	pVCpu->hm.s.u32HMError = pVmxTransient->uExitIntInfo;
14684	rcStrict = VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
14685	AssertMsgFailed(("Invalid/unexpected VM-exit interruption info %#x\n", pVmxTransient->uExitIntInfo));
14686	break;
14687	}
14688	}
14689
14690	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
14691	return rcStrict;
14692	}
14693
14694
14695	/**
14696	* VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
14697	*/
14698	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindow(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14699	{
14700	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14701
14702	/* Indicate that we no longer need to VM-exit when the guest is ready to receive interrupts, it is now ready. */
14703	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14704	hmR0VmxClearIntWindowExitVmcs(pVmcsInfo);
14705
14706	/* Evaluate and deliver pending events and resume guest execution. */
14707	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow);
14708	return VINF_SUCCESS;
14709	}
14710
14711
14712	/**
14713	* VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
14714	*/
14715	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindow(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14716	{
14717	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14718
14719	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14720	if (RT_UNLIKELY(!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))) /** @todo NSTVMX: Turn this into an assertion. */
14721	{
14722	AssertMsgFailed(("Unexpected NMI-window exit.\n"));
14723	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
14724	}
14725
14726	Assert(!CPUMIsGuestNmiBlocking(pVCpu));
14727
14728	/*
14729	* If block-by-STI is set when we get this VM-exit, it means the CPU doesn't block NMIs following STI.
14730	* It is therefore safe to unblock STI and deliver the NMI ourselves. See @bugref{7445}.
14731	*/
14732	uint32_t fIntrState;
14733	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
14734	AssertRC(rc);
14735	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
14736	if (fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
14737	{
14738	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
14739	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
14740
14741	fIntrState &= ~VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
14742	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
14743	AssertRC(rc);
14744	}
14745
14746	/* Indicate that we no longer need to VM-exit when the guest is ready to receive NMIs, it is now ready */
14747	hmR0VmxClearNmiWindowExitVmcs(pVmcsInfo);
14748
14749	/* Evaluate and deliver pending events and resume guest execution. */
14750	return VINF_SUCCESS;
14751	}
14752
14753
14754	/**
14755	* VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
14756	*/
14757	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvd(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14758	{
14759	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14760	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14761	}
14762
14763
14764	/**
14765	* VM-exit handler for INVD (VMX_EXIT_INVD). Unconditional VM-exit.
14766	*/
14767	HMVMX_EXIT_NSRC_DECL hmR0VmxExitInvd(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14768	{
14769	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14770	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14771	}
14772
14773
14774	/**
14775	* VM-exit handler for CPUID (VMX_EXIT_CPUID). Unconditional VM-exit.
14776	*/
14777	HMVMX_EXIT_DECL hmR0VmxExitCpuid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14778	{
14779	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14780
14781	/*
14782	* Get the state we need and update the exit history entry.
14783	*/
14784	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14785	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14786
14787	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
14788	AssertRCReturn(rc, rc);
14789
14790	VBOXSTRICTRC rcStrict;
14791	PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
14792	EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_CPUID),
14793	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
14794	if (!pExitRec)
14795	{
14796	/*
14797	* Regular CPUID instruction execution.
14798	*/
14799	rcStrict = IEMExecDecodedCpuid(pVCpu, pVmxTransient->cbExitInstr);
14800	if (rcStrict == VINF_SUCCESS)
14801	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14802	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14803	{
14804	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14805	rcStrict = VINF_SUCCESS;
14806	}
14807	}
14808	else
14809	{
14810	/*
14811	* Frequent exit or something needing probing. Get state and call EMHistoryExec.
14812	*/
14813	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14814	AssertRCReturn(rc2, rc2);
14815
14816	Log4(("CpuIdExit/%u: %04x:%08RX64: %#x/%#x -> EMHistoryExec\n",
14817	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.eax, pVCpu->cpum.GstCtx.ecx));
14818
14819	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
14820	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14821
14822	Log4(("CpuIdExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
14823	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
14824	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
14825	}
14826	return rcStrict;
14827	}
14828
14829
14830	/**
14831	* VM-exit handler for GETSEC (VMX_EXIT_GETSEC). Unconditional VM-exit.
14832	*/
14833	HMVMX_EXIT_DECL hmR0VmxExitGetsec(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14834	{
14835	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14836
14837	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14838	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR4);
14839	AssertRCReturn(rc, rc);
14840
14841	if (pVCpu->cpum.GstCtx.cr4 & X86_CR4_SMXE)
14842	return VINF_EM_RAW_EMULATE_INSTR;
14843
14844	AssertMsgFailed(("hmR0VmxExitGetsec: Unexpected VM-exit when CR4.SMXE is 0.\n"));
14845	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
14846	}
14847
14848
14849	/**
14850	* VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
14851	*/
14852	HMVMX_EXIT_DECL hmR0VmxExitRdtsc(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14853	{
14854	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14855
14856	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14857	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14858	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14859	AssertRCReturn(rc, rc);
14860
14861	VBOXSTRICTRC rcStrict = IEMExecDecodedRdtsc(pVCpu, pVmxTransient->cbExitInstr);
14862	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
14863	{
14864	/* If we get a spurious VM-exit when TSC offsetting is enabled,
14865	we must reset offsetting on VM-entry. See @bugref{6634}. */
14866	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TSC_OFFSETTING)
14867	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14868	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14869	}
14870	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14871	{
14872	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14873	rcStrict = VINF_SUCCESS;
14874	}
14875	return rcStrict;
14876	}
14877
14878
14879	/**
14880	* VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
14881	*/
14882	HMVMX_EXIT_DECL hmR0VmxExitRdtscp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14883	{
14884	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14885
14886	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14887	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14888	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_TSC_AUX);
14889	AssertRCReturn(rc, rc);
14890
14891	VBOXSTRICTRC rcStrict = IEMExecDecodedRdtscp(pVCpu, pVmxTransient->cbExitInstr);
14892	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
14893	{
14894	/* If we get a spurious VM-exit when TSC offsetting is enabled,
14895	we must reset offsetting on VM-reentry. See @bugref{6634}. */
14896	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TSC_OFFSETTING)
14897	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14898	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14899	}
14900	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14901	{
14902	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14903	rcStrict = VINF_SUCCESS;
14904	}
14905	return rcStrict;
14906	}
14907
14908
14909	/**
14910	* VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
14911	*/
14912	HMVMX_EXIT_DECL hmR0VmxExitRdpmc(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14913	{
14914	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14915
14916	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14917	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR4 \| CPUMCTX_EXTRN_CR0
14918	\| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS);
14919	AssertRCReturn(rc, rc);
14920
14921	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14922	rc = EMInterpretRdpmc(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx));
14923	if (RT_LIKELY(rc == VINF_SUCCESS))
14924	{
14925	rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14926	Assert(pVmxTransient->cbExitInstr == 2);
14927	}
14928	else
14929	{
14930	AssertMsgFailed(("hmR0VmxExitRdpmc: EMInterpretRdpmc failed with %Rrc\n", rc));
14931	rc = VERR_EM_INTERPRETER;
14932	}
14933	return rc;
14934	}
14935
14936
14937	/**
14938	* VM-exit handler for VMCALL (VMX_EXIT_VMCALL). Unconditional VM-exit.
14939	*/
14940	HMVMX_EXIT_DECL hmR0VmxExitVmcall(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14941	{
14942	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14943
14944	VBOXSTRICTRC rcStrict = VERR_VMX_IPE_3;
14945	if (EMAreHypercallInstructionsEnabled(pVCpu))
14946	{
14947	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14948	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_CR0
14949	\| CPUMCTX_EXTRN_SS \| CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_EFER);
14950	AssertRCReturn(rc, rc);
14951
14952	/* Perform the hypercall. */
14953	rcStrict = GIMHypercall(pVCpu, &pVCpu->cpum.GstCtx);
14954	if (rcStrict == VINF_SUCCESS)
14955	{
14956	rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14957	AssertRCReturn(rc, rc);
14958	}
14959	else
14960	Assert( rcStrict == VINF_GIM_R3_HYPERCALL
14961	\|\| rcStrict == VINF_GIM_HYPERCALL_CONTINUING
14962	\|\| RT_FAILURE(rcStrict));
14963
14964	/* If the hypercall changes anything other than guest's general-purpose registers,
14965	we would need to reload the guest changed bits here before VM-entry. */
14966	}
14967	else
14968	Log4Func(("Hypercalls not enabled\n"));
14969
14970	/* If hypercalls are disabled or the hypercall failed for some reason, raise #UD and continue. */
14971	if (RT_FAILURE(rcStrict))
14972	{
14973	hmR0VmxSetPendingXcptUD(pVCpu);
14974	rcStrict = VINF_SUCCESS;
14975	}
14976
14977	return rcStrict;
14978	}
14979
14980
14981	/**
14982	* VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
14983	*/
14984	HMVMX_EXIT_DECL hmR0VmxExitInvlpg(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14985	{
14986	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14987	Assert(!pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging \|\| pVCpu->hmr0.s.fUsingDebugLoop);
14988
14989	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14990	hmR0VmxReadExitQualVmcs(pVmxTransient);
14991	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14992	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
14993	AssertRCReturn(rc, rc);
14994
14995	VBOXSTRICTRC rcStrict = IEMExecDecodedInvlpg(pVCpu, pVmxTransient->cbExitInstr, pVmxTransient->uExitQual);
14996
14997	if (rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_PGM_SYNC_CR3)
14998	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14999	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15000	{
15001	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15002	rcStrict = VINF_SUCCESS;
15003	}
15004	else
15005	AssertMsgFailed(("Unexpected IEMExecDecodedInvlpg(%#RX64) status: %Rrc\n", pVmxTransient->uExitQual,
15006	VBOXSTRICTRC_VAL(rcStrict)));
15007	return rcStrict;
15008	}
15009
15010
15011	/**
15012	* VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
15013	*/
15014	HMVMX_EXIT_DECL hmR0VmxExitMonitor(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15015	{
15016	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15017
15018	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15019	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15020	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK \| CPUMCTX_EXTRN_DS);
15021	AssertRCReturn(rc, rc);
15022
15023	VBOXSTRICTRC rcStrict = IEMExecDecodedMonitor(pVCpu, pVmxTransient->cbExitInstr);
15024	if (rcStrict == VINF_SUCCESS)
15025	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15026	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15027	{
15028	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15029	rcStrict = VINF_SUCCESS;
15030	}
15031
15032	return rcStrict;
15033	}
15034
15035
15036	/**
15037	* VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
15038	*/
15039	HMVMX_EXIT_DECL hmR0VmxExitMwait(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15040	{
15041	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15042
15043	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15044	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15045	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
15046	AssertRCReturn(rc, rc);
15047
15048	VBOXSTRICTRC rcStrict = IEMExecDecodedMwait(pVCpu, pVmxTransient->cbExitInstr);
15049	if (RT_SUCCESS(rcStrict))
15050	{
15051	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15052	if (EMMonitorWaitShouldContinue(pVCpu, &pVCpu->cpum.GstCtx))
15053	rcStrict = VINF_SUCCESS;
15054	}
15055
15056	return rcStrict;
15057	}
15058
15059
15060	/**
15061	* VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT). Unconditional
15062	* VM-exit.
15063	*/
15064	HMVMX_EXIT_DECL hmR0VmxExitTripleFault(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15065	{
15066	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15067	return VINF_EM_RESET;
15068	}
15069
15070
15071	/**
15072	* VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
15073	*/
15074	HMVMX_EXIT_DECL hmR0VmxExitHlt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15075	{
15076	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15077
15078	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15079	AssertRCReturn(rc, rc);
15080
15081	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS); /* Advancing the RIP above should've imported eflags. */
15082	if (EMShouldContinueAfterHalt(pVCpu, &pVCpu->cpum.GstCtx)) /* Requires eflags. */
15083	rc = VINF_SUCCESS;
15084	else
15085	rc = VINF_EM_HALT;
15086
15087	if (rc != VINF_SUCCESS)
15088	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHltToR3);
15089	return rc;
15090	}
15091
15092
15093	/**
15094	* VM-exit handler for instructions that result in a \#UD exception delivered to
15095	* the guest.
15096	*/
15097	HMVMX_EXIT_NSRC_DECL hmR0VmxExitSetPendingXcptUD(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15098	{
15099	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15100	hmR0VmxSetPendingXcptUD(pVCpu);
15101	return VINF_SUCCESS;
15102	}
15103
15104
15105	/**
15106	* VM-exit handler for expiry of the VMX-preemption timer.
15107	*/
15108	HMVMX_EXIT_DECL hmR0VmxExitPreemptTimer(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15109	{
15110	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15111
15112	/* If the VMX-preemption timer has expired, reinitialize the preemption timer on next VM-entry. */
15113	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
15114	Log12(("hmR0VmxExitPreemptTimer:\n"));
15115
15116	/* If there are any timer events pending, fall back to ring-3, otherwise resume guest execution. */
15117	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15118	bool fTimersPending = TMTimerPollBool(pVM, pVCpu);
15119	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitPreemptTimer);
15120	return fTimersPending ? VINF_EM_RAW_TIMER_PENDING : VINF_SUCCESS;
15121	}
15122
15123
15124	/**
15125	* VM-exit handler for XSETBV (VMX_EXIT_XSETBV). Unconditional VM-exit.
15126	*/
15127	HMVMX_EXIT_DECL hmR0VmxExitXsetbv(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15128	{
15129	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15130
15131	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15132	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15133	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_CR4);
15134	AssertRCReturn(rc, rc);
15135
15136	VBOXSTRICTRC rcStrict = IEMExecDecodedXsetbv(pVCpu, pVmxTransient->cbExitInstr);
15137	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, rcStrict != VINF_IEM_RAISED_XCPT ? HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
15138	: HM_CHANGED_RAISED_XCPT_MASK);
15139
15140	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15141	bool const fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
15142	if (fLoadSaveGuestXcr0 != pVCpu->hmr0.s.fLoadSaveGuestXcr0)
15143	{
15144	pVCpu->hmr0.s.fLoadSaveGuestXcr0 = fLoadSaveGuestXcr0;
15145	hmR0VmxUpdateStartVmFunction(pVCpu);
15146	}
15147
15148	return rcStrict;
15149	}
15150
15151
15152	/**
15153	* VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
15154	*/
15155	HMVMX_EXIT_DECL hmR0VmxExitInvpcid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15156	{
15157	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15158
15159	/** @todo Enable the new code after finding a reliably guest test-case. */
15160	#if 1
15161	return VERR_EM_INTERPRETER;
15162	#else
15163	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15164	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15165	hmR0VmxReadExitQualVmcs(pVmxTransient);
15166	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
15167	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
15168	AssertRCReturn(rc, rc);
15169
15170	/* Paranoia. Ensure this has a memory operand. */
15171	Assert(!pVmxTransient->ExitInstrInfo.Inv.u1Cleared0);
15172
15173	uint8_t const iGReg = pVmxTransient->ExitInstrInfo.VmreadVmwrite.iReg2;
15174	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
15175	uint64_t const uType = CPUMIsGuestIn64BitCode(pVCpu) ? pVCpu->cpum.GstCtx.aGRegs[iGReg].u64
15176	: pVCpu->cpum.GstCtx.aGRegs[iGReg].u32;
15177
15178	RTGCPTR GCPtrDesc;
15179	HMVMX_DECODE_MEM_OPERAND(pVCpu, pVmxTransient->ExitInstrInfo.u, pVmxTransient->uExitQual, VMXMEMACCESS_READ, &GCPtrDesc);
15180
15181	VBOXSTRICTRC rcStrict = IEMExecDecodedInvpcid(pVCpu, pVmxTransient->cbExitInstr, pVmxTransient->ExitInstrInfo.Inv.iSegReg,
15182	GCPtrDesc, uType);
15183	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15184	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15185	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15186	{
15187	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15188	rcStrict = VINF_SUCCESS;
15189	}
15190	return rcStrict;
15191	#endif
15192	}
15193
15194
15195	/**
15196	* VM-exit handler for invalid-guest-state (VMX_EXIT_ERR_INVALID_GUEST_STATE). Error
15197	* VM-exit.
15198	*/
15199	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrInvalidGuestState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15200	{
15201	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15202	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
15203	AssertRCReturn(rc, rc);
15204
15205	rc = hmR0VmxCheckCachedVmcsCtls(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest);
15206	if (RT_FAILURE(rc))
15207	return rc;
15208
15209	uint32_t const uInvalidReason = hmR0VmxCheckGuestState(pVCpu, pVmcsInfo);
15210	NOREF(uInvalidReason);
15211
15212	#ifdef VBOX_STRICT
15213	uint32_t fIntrState;
15214	uint64_t u64Val;
15215	hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
15216	hmR0VmxReadEntryXcptErrorCodeVmcs(pVmxTransient);
15217	hmR0VmxReadEntryInstrLenVmcs(pVmxTransient);
15218
15219	Log4(("uInvalidReason %u\n", uInvalidReason));
15220	Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", pVmxTransient->uEntryIntInfo));
15221	Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", pVmxTransient->uEntryXcptErrorCode));
15222	Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %#RX32\n", pVmxTransient->cbEntryInstr));
15223
15224	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState); AssertRC(rc);
15225	Log4(("VMX_VMCS32_GUEST_INT_STATE %#RX32\n", fIntrState));
15226	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64Val); AssertRC(rc);
15227	Log4(("VMX_VMCS_GUEST_CR0 %#RX64\n", u64Val));
15228	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_MASK, &u64Val); AssertRC(rc);
15229	Log4(("VMX_VMCS_CTRL_CR0_MASK %#RX64\n", u64Val));
15230	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, &u64Val); AssertRC(rc);
15231	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RX64\n", u64Val));
15232	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_MASK, &u64Val); AssertRC(rc);
15233	Log4(("VMX_VMCS_CTRL_CR4_MASK %#RX64\n", u64Val));
15234	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, &u64Val); AssertRC(rc);
15235	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RX64\n", u64Val));
15236	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging)
15237	{
15238	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
15239	Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
15240	}
15241	hmR0DumpRegs(pVCpu, HM_DUMP_REG_FLAGS_ALL);
15242	#endif
15243
15244	return VERR_VMX_INVALID_GUEST_STATE;
15245	}
15246
15247	/**
15248	* VM-exit handler for all undefined/unexpected reasons. Should never happen.
15249	*/
15250	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrUnexpected(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15251	{
15252	/*
15253	* Cumulative notes of all recognized but unexpected VM-exits.
15254	*
15255	* 1. This does -not- cover scenarios like a page-fault VM-exit occurring when
15256	* nested-paging is used.
15257	*
15258	* 2. Any instruction that causes a VM-exit unconditionally (for e.g. VMXON) must be
15259	* emulated or a #UD must be raised in the guest. Therefore, we should -not- be using
15260	* this function (and thereby stop VM execution) for handling such instructions.
15261	*
15262	*
15263	* VMX_EXIT_INIT_SIGNAL:
15264	* INIT signals are blocked in VMX root operation by VMXON and by SMI in SMM.
15265	* It is -NOT- blocked in VMX non-root operation so we can, in theory, still get these
15266	* VM-exits. However, we should not receive INIT signals VM-exit while executing a VM.
15267	*
15268	* See Intel spec. 33.14.1 Default Treatment of SMI Delivery"
15269	* See Intel spec. 29.3 "VMX Instructions" for "VMXON".
15270	* See Intel spec. "23.8 Restrictions on VMX operation".
15271	*
15272	* VMX_EXIT_SIPI:
15273	* SIPI exits can only occur in VMX non-root operation when the "wait-for-SIPI" guest
15274	* activity state is used. We don't make use of it as our guests don't have direct
15275	* access to the host local APIC.
15276	*
15277	* See Intel spec. 25.3 "Other Causes of VM-exits".
15278	*
15279	* VMX_EXIT_IO_SMI:
15280	* VMX_EXIT_SMI:
15281	* This can only happen if we support dual-monitor treatment of SMI, which can be
15282	* activated by executing VMCALL in VMX root operation. Only an STM (SMM transfer
15283	* monitor) would get this VM-exit when we (the executive monitor) execute a VMCALL in
15284	* VMX root mode or receive an SMI. If we get here, something funny is going on.
15285	*
15286	* See Intel spec. 33.15.6 "Activating the Dual-Monitor Treatment"
15287	* See Intel spec. 25.3 "Other Causes of VM-Exits"
15288	*
15289	* VMX_EXIT_ERR_MSR_LOAD:
15290	* Failures while loading MSRs are part of the VM-entry MSR-load area are unexpected
15291	* and typically indicates a bug in the hypervisor code. We thus cannot not resume
15292	* execution.
15293	*
15294	* See Intel spec. 26.7 "VM-Entry Failures During Or After Loading Guest State".
15295	*
15296	* VMX_EXIT_ERR_MACHINE_CHECK:
15297	* Machine check exceptions indicates a fatal/unrecoverable hardware condition
15298	* including but not limited to system bus, ECC, parity, cache and TLB errors. A
15299	* #MC exception abort class exception is raised. We thus cannot assume a
15300	* reasonable chance of continuing any sort of execution and we bail.
15301	*
15302	* See Intel spec. 15.1 "Machine-check Architecture".
15303	* See Intel spec. 27.1 "Architectural State Before A VM Exit".
15304	*
15305	* VMX_EXIT_PML_FULL:
15306	* VMX_EXIT_VIRTUALIZED_EOI:
15307	* VMX_EXIT_APIC_WRITE:
15308	* We do not currently support any of these features and thus they are all unexpected
15309	* VM-exits.
15310	*
15311	* VMX_EXIT_GDTR_IDTR_ACCESS:
15312	* VMX_EXIT_LDTR_TR_ACCESS:
15313	* VMX_EXIT_RDRAND:
15314	* VMX_EXIT_RSM:
15315	* VMX_EXIT_VMFUNC:
15316	* VMX_EXIT_ENCLS:
15317	* VMX_EXIT_RDSEED:
15318	* VMX_EXIT_XSAVES:
15319	* VMX_EXIT_XRSTORS:
15320	* VMX_EXIT_UMWAIT:
15321	* VMX_EXIT_TPAUSE:
15322	* VMX_EXIT_LOADIWKEY:
15323	* These VM-exits are -not- caused unconditionally by execution of the corresponding
15324	* instruction. Any VM-exit for these instructions indicate a hardware problem,
15325	* unsupported CPU modes (like SMM) or potentially corrupt VMCS controls.
15326	*
15327	* See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally".
15328	*/
15329	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15330	AssertMsgFailed(("Unexpected VM-exit %u\n", pVmxTransient->uExitReason));
15331	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
15332	}
15333
15334
15335	/**
15336	* VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
15337	*/
15338	HMVMX_EXIT_DECL hmR0VmxExitRdmsr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15339	{
15340	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15341
15342	/** @todo Optimize this: We currently drag in the whole MSR state
15343	* (CPUMCTX_EXTRN_ALL_MSRS) here. We should optimize this to only get
15344	* MSRs required. That would require changes to IEM and possibly CPUM too.
15345	* (Should probably do it lazy fashion from CPUMAllMsrs.cpp). */
15346	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15347	uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
15348	uint64_t fImport = IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK \| CPUMCTX_EXTRN_ALL_MSRS;
15349	switch (idMsr)
15350	{
15351	case MSR_K8_FS_BASE: fImport \|= CPUMCTX_EXTRN_FS; break;
15352	case MSR_K8_GS_BASE: fImport \|= CPUMCTX_EXTRN_GS; break;
15353	}
15354
15355	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15356	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImport);
15357	AssertRCReturn(rc, rc);
15358
15359	Log4Func(("ecx=%#RX32\n", idMsr));
15360
15361	#ifdef VBOX_STRICT
15362	Assert(!pVmxTransient->fIsNestedGuest);
15363	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
15364	{
15365	if ( hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr)
15366	&& idMsr != MSR_K6_EFER)
15367	{
15368	AssertMsgFailed(("Unexpected RDMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n", idMsr));
15369	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15370	}
15371	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15372	{
15373	Assert(pVmcsInfo->pvMsrBitmap);
15374	uint32_t fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, idMsr);
15375	if (fMsrpm & VMXMSRPM_ALLOW_RD)
15376	{
15377	AssertMsgFailed(("Unexpected RDMSR for a passthru lazy-restore MSR. ecx=%#RX32\n", idMsr));
15378	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15379	}
15380	}
15381	}
15382	#endif
15383
15384	VBOXSTRICTRC rcStrict = IEMExecDecodedRdmsr(pVCpu, pVmxTransient->cbExitInstr);
15385	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr);
15386	if (rcStrict == VINF_SUCCESS)
15387	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15388	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15389	{
15390	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15391	rcStrict = VINF_SUCCESS;
15392	}
15393	else
15394	AssertMsg(rcStrict == VINF_CPUM_R3_MSR_READ \|\| rcStrict == VINF_EM_TRIPLE_FAULT,
15395	("Unexpected IEMExecDecodedRdmsr rc (%Rrc)\n", VBOXSTRICTRC_VAL(rcStrict)));
15396
15397	return rcStrict;
15398	}
15399
15400
15401	/**
15402	* VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
15403	*/
15404	HMVMX_EXIT_DECL hmR0VmxExitWrmsr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15405	{
15406	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15407
15408	/** @todo Optimize this: We currently drag in the whole MSR state
15409	* (CPUMCTX_EXTRN_ALL_MSRS) here. We should optimize this to only get
15410	* MSRs required. That would require changes to IEM and possibly CPUM too.
15411	* (Should probably do it lazy fashion from CPUMAllMsrs.cpp). */
15412	uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
15413	uint64_t fImport = IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK \| CPUMCTX_EXTRN_ALL_MSRS;
15414
15415	/*
15416	* The FS and GS base MSRs are not part of the above all-MSRs mask.
15417	* Although we don't need to fetch the base as it will be overwritten shortly, while
15418	* loading guest-state we would also load the entire segment register including limit
15419	* and attributes and thus we need to load them here.
15420	*/
15421	switch (idMsr)
15422	{
15423	case MSR_K8_FS_BASE: fImport \|= CPUMCTX_EXTRN_FS; break;
15424	case MSR_K8_GS_BASE: fImport \|= CPUMCTX_EXTRN_GS; break;
15425	}
15426
15427	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15428	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15429	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImport);
15430	AssertRCReturn(rc, rc);
15431
15432	Log4Func(("ecx=%#RX32 edx:eax=%#RX32:%#RX32\n", idMsr, pVCpu->cpum.GstCtx.edx, pVCpu->cpum.GstCtx.eax));
15433
15434	VBOXSTRICTRC rcStrict = IEMExecDecodedWrmsr(pVCpu, pVmxTransient->cbExitInstr);
15435	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr);
15436
15437	if (rcStrict == VINF_SUCCESS)
15438	{
15439	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15440
15441	/* If this is an X2APIC WRMSR access, update the APIC state as well. */
15442	if ( idMsr == MSR_IA32_APICBASE
15443	\|\| ( idMsr >= MSR_IA32_X2APIC_START
15444	&& idMsr <= MSR_IA32_X2APIC_END))
15445	{
15446	/*
15447	* We've already saved the APIC related guest-state (TPR) in post-run phase.
15448	* When full APIC register virtualization is implemented we'll have to make
15449	* sure APIC state is saved from the VMCS before IEM changes it.
15450	*/
15451	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
15452	}
15453	else if (idMsr == MSR_IA32_TSC) /* Windows 7 does this during bootup. See @bugref{6398}. */
15454	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
15455	else if (idMsr == MSR_K6_EFER)
15456	{
15457	/*
15458	* If the guest touches the EFER MSR we need to update the VM-Entry and VM-Exit controls
15459	* as well, even if it is -not- touching bits that cause paging mode changes (LMA/LME).
15460	* We care about the other bits as well, SCE and NXE. See @bugref{7368}.
15461	*/
15462	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_EFER_MSR \| HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
15463	}
15464
15465	/* Update MSRs that are part of the VMCS and auto-load/store area when MSR-bitmaps are not used. */
15466	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS))
15467	{
15468	switch (idMsr)
15469	{
15470	case MSR_IA32_SYSENTER_CS: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_CS_MSR); break;
15471	case MSR_IA32_SYSENTER_EIP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); break;
15472	case MSR_IA32_SYSENTER_ESP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); break;
15473	case MSR_K8_FS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_FS); break;
15474	case MSR_K8_GS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_GS); break;
15475	case MSR_K6_EFER: /* Nothing to do, already handled above. */ break;
15476	default:
15477	{
15478	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15479	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_VMX_GUEST_LAZY_MSRS);
15480	else if (hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr))
15481	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
15482	break;
15483	}
15484	}
15485	}
15486	#ifdef VBOX_STRICT
15487	else
15488	{
15489	/* Paranoia. Validate that MSRs in the MSR-bitmaps with write-passthru are not intercepted. */
15490	switch (idMsr)
15491	{
15492	case MSR_IA32_SYSENTER_CS:
15493	case MSR_IA32_SYSENTER_EIP:
15494	case MSR_IA32_SYSENTER_ESP:
15495	case MSR_K8_FS_BASE:
15496	case MSR_K8_GS_BASE:
15497	{
15498	AssertMsgFailed(("Unexpected WRMSR for an MSR in the VMCS. ecx=%#RX32\n", idMsr));
15499	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15500	}
15501
15502	/* Writes to MSRs in auto-load/store area/swapped MSRs, shouldn't cause VM-exits with MSR-bitmaps. */
15503	default:
15504	{
15505	if (hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr))
15506	{
15507	/* EFER MSR writes are always intercepted. */
15508	if (idMsr != MSR_K6_EFER)
15509	{
15510	AssertMsgFailed(("Unexpected WRMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n",
15511	idMsr));
15512	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15513	}
15514	}
15515
15516	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15517	{
15518	Assert(pVmcsInfo->pvMsrBitmap);
15519	uint32_t fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, idMsr);
15520	if (fMsrpm & VMXMSRPM_ALLOW_WR)
15521	{
15522	AssertMsgFailed(("Unexpected WRMSR for passthru, lazy-restore MSR. ecx=%#RX32\n", idMsr));
15523	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15524	}
15525	}
15526	break;
15527	}
15528	}
15529	}
15530	#endif /* VBOX_STRICT */
15531	}
15532	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15533	{
15534	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15535	rcStrict = VINF_SUCCESS;
15536	}
15537	else
15538	AssertMsg(rcStrict == VINF_CPUM_R3_MSR_WRITE \|\| rcStrict == VINF_EM_TRIPLE_FAULT,
15539	("Unexpected IEMExecDecodedWrmsr rc (%Rrc)\n", VBOXSTRICTRC_VAL(rcStrict)));
15540
15541	return rcStrict;
15542	}
15543
15544
15545	/**
15546	* VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
15547	*/
15548	HMVMX_EXIT_DECL hmR0VmxExitPause(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15549	{
15550	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15551
15552	/** @todo The guest has likely hit a contended spinlock. We might want to
15553	* poke a schedule different guest VCPU. */
15554	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15555	if (RT_SUCCESS(rc))
15556	return VINF_EM_RAW_INTERRUPT;
15557
15558	AssertMsgFailed(("hmR0VmxExitPause: Failed to increment RIP. rc=%Rrc\n", rc));
15559	return rc;
15560	}
15561
15562
15563	/**
15564	* VM-exit handler for when the TPR value is lowered below the specified
15565	* threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
15566	*/
15567	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThreshold(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15568	{
15569	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15570	Assert(pVmxTransient->pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
15571
15572	/*
15573	* The TPR shadow would've been synced with the APIC TPR in the post-run phase.
15574	* We'll re-evaluate pending interrupts and inject them before the next VM
15575	* entry so we can just continue execution here.
15576	*/
15577	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTprBelowThreshold);
15578	return VINF_SUCCESS;
15579	}
15580
15581
15582	/**
15583	* VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX). Conditional
15584	* VM-exit.
15585	*
15586	* @retval VINF_SUCCESS when guest execution can continue.
15587	* @retval VINF_PGM_SYNC_CR3 CR3 sync is required, back to ring-3.
15588	* @retval VERR_EM_RESCHEDULE_REM when we need to return to ring-3 due to
15589	* incompatible guest state for VMX execution (real-on-v86 case).
15590	*/
15591	HMVMX_EXIT_DECL hmR0VmxExitMovCRx(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15592	{
15593	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15594	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitMovCRx, y2);
15595
15596	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15597	hmR0VmxReadExitQualVmcs(pVmxTransient);
15598	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15599
15600	VBOXSTRICTRC rcStrict;
15601	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15602	uint64_t const uExitQual = pVmxTransient->uExitQual;
15603	uint32_t const uAccessType = VMX_EXIT_QUAL_CRX_ACCESS(uExitQual);
15604	switch (uAccessType)
15605	{
15606	/*
15607	* MOV to CRx.
15608	*/
15609	case VMX_EXIT_QUAL_CRX_ACCESS_WRITE:
15610	{
15611	/*
15612	* When PAE paging is used, the CPU will reload PAE PDPTEs from CR3 when the guest
15613	* changes certain bits even in CR0, CR4 (and not just CR3). We are currently fine
15614	* since IEM_CPUMCTX_EXTRN_MUST_MASK (used below) includes CR3 which will import
15615	* PAE PDPTEs as well.
15616	*/
15617	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
15618	AssertRCReturn(rc, rc);
15619
15620	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
15621	uint32_t const uOldCr0 = pVCpu->cpum.GstCtx.cr0;
15622	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(uExitQual);
15623	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(uExitQual);
15624
15625	/*
15626	* MOV to CR3 only cause a VM-exit when one or more of the following are true:
15627	* - When nested paging isn't used.
15628	* - If the guest doesn't have paging enabled (intercept CR3 to update shadow page tables).
15629	* - We are executing in the VM debug loop.
15630	*/
15631	Assert( iCrReg != 3
15632	\|\| !pVM->hmr0.s.fNestedPaging
15633	\|\| !CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx)
15634	\|\| pVCpu->hmr0.s.fUsingDebugLoop);
15635
15636	/* MOV to CR8 writes only cause VM-exits when TPR shadow is not used. */
15637	Assert( iCrReg != 8
15638	\|\| !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW));
15639
15640	rcStrict = hmR0VmxExitMovToCrX(pVCpu, pVmxTransient->cbExitInstr, iGReg, iCrReg);
15641	AssertMsg( rcStrict == VINF_SUCCESS
15642	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15643
15644	/*
15645	* This is a kludge for handling switches back to real mode when we try to use
15646	* V86 mode to run real mode code directly. Problem is that V86 mode cannot
15647	* deal with special selector values, so we have to return to ring-3 and run
15648	* there till the selector values are V86 mode compatible.
15649	*
15650	* Note! Using VINF_EM_RESCHEDULE_REM here rather than VINF_EM_RESCHEDULE since the
15651	* latter is an alias for VINF_IEM_RAISED_XCPT which is asserted at the end of
15652	* this function.
15653	*/
15654	if ( iCrReg == 0
15655	&& rcStrict == VINF_SUCCESS
15656	&& !pVM->hmr0.s.vmx.fUnrestrictedGuest
15657	&& CPUMIsGuestInRealModeEx(&pVCpu->cpum.GstCtx)
15658	&& (uOldCr0 & X86_CR0_PE)
15659	&& !(pVCpu->cpum.GstCtx.cr0 & X86_CR0_PE))
15660	{
15661	/** @todo Check selectors rather than returning all the time. */
15662	Assert(!pVmxTransient->fIsNestedGuest);
15663	Log4Func(("CR0 write, back to real mode -> VINF_EM_RESCHEDULE_REM\n"));
15664	rcStrict = VINF_EM_RESCHEDULE_REM;
15665	}
15666	break;
15667	}
15668
15669	/*
15670	* MOV from CRx.
15671	*/
15672	case VMX_EXIT_QUAL_CRX_ACCESS_READ:
15673	{
15674	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(uExitQual);
15675	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(uExitQual);
15676
15677	/*
15678	* MOV from CR3 only cause a VM-exit when one or more of the following are true:
15679	* - When nested paging isn't used.
15680	* - If the guest doesn't have paging enabled (pass guest's CR3 rather than our identity mapped CR3).
15681	* - We are executing in the VM debug loop.
15682	*/
15683	Assert( iCrReg != 3
15684	\|\| !pVM->hmr0.s.fNestedPaging
15685	\|\| !CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx)
15686	\|\| pVCpu->hmr0.s.fLeaveDone);
15687
15688	/* MOV from CR8 reads only cause a VM-exit when the TPR shadow feature isn't enabled. */
15689	Assert( iCrReg != 8
15690	\|\| !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW));
15691
15692	rcStrict = hmR0VmxExitMovFromCrX(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
15693	break;
15694	}
15695
15696	/*
15697	* CLTS (Clear Task-Switch Flag in CR0).
15698	*/
15699	case VMX_EXIT_QUAL_CRX_ACCESS_CLTS:
15700	{
15701	rcStrict = hmR0VmxExitClts(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr);
15702	break;
15703	}
15704
15705	/*
15706	* LMSW (Load Machine-Status Word into CR0).
15707	* LMSW cannot clear CR0.PE, so no fRealOnV86Active kludge needed here.
15708	*/
15709	case VMX_EXIT_QUAL_CRX_ACCESS_LMSW:
15710	{
15711	RTGCPTR GCPtrEffDst;
15712	uint8_t const cbInstr = pVmxTransient->cbExitInstr;
15713	uint16_t const uMsw = VMX_EXIT_QUAL_CRX_LMSW_DATA(uExitQual);
15714	bool const fMemOperand = VMX_EXIT_QUAL_CRX_LMSW_OP_MEM(uExitQual);
15715	if (fMemOperand)
15716	{
15717	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
15718	GCPtrEffDst = pVmxTransient->uGuestLinearAddr;
15719	}
15720	else
15721	GCPtrEffDst = NIL_RTGCPTR;
15722	rcStrict = hmR0VmxExitLmsw(pVCpu, pVmcsInfo, cbInstr, uMsw, GCPtrEffDst);
15723	break;
15724	}
15725
15726	default:
15727	{
15728	AssertMsgFailed(("Unrecognized Mov CRX access type %#x\n", uAccessType));
15729	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, uAccessType);
15730	}
15731	}
15732
15733	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS))
15734	== (HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS));
15735	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
15736
15737	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitMovCRx, y2);
15738	NOREF(pVM);
15739	return rcStrict;
15740	}
15741
15742
15743	/**
15744	* VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR). Conditional
15745	* VM-exit.
15746	*/
15747	HMVMX_EXIT_DECL hmR0VmxExitIoInstr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15748	{
15749	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15750	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitIO, y1);
15751
15752	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15753	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15754	hmR0VmxReadExitQualVmcs(pVmxTransient);
15755	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15756	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_SREG_MASK
15757	\| CPUMCTX_EXTRN_EFER);
15758	/* EFER MSR also required for longmode checks in EMInterpretDisasCurrent(), but it's always up-to-date. */
15759	AssertRCReturn(rc, rc);
15760
15761	/* Refer Intel spec. 27-5. "Exit Qualifications for I/O Instructions" for the format. */
15762	uint32_t const uIOPort = VMX_EXIT_QUAL_IO_PORT(pVmxTransient->uExitQual);
15763	uint8_t const uIOSize = VMX_EXIT_QUAL_IO_SIZE(pVmxTransient->uExitQual);
15764	bool const fIOWrite = (VMX_EXIT_QUAL_IO_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_IO_DIRECTION_OUT);
15765	bool const fIOString = VMX_EXIT_QUAL_IO_IS_STRING(pVmxTransient->uExitQual);
15766	bool const fGstStepping = RT_BOOL(pCtx->eflags.Bits.u1TF);
15767	bool const fDbgStepping = pVCpu->hm.s.fSingleInstruction;
15768	AssertReturn(uIOSize <= 3 && uIOSize != 2, VERR_VMX_IPE_1);
15769
15770	/*
15771	* Update exit history to see if this exit can be optimized.
15772	*/
15773	VBOXSTRICTRC rcStrict;
15774	PCEMEXITREC pExitRec = NULL;
15775	if ( !fGstStepping
15776	&& !fDbgStepping)
15777	pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
15778	!fIOString
15779	? !fIOWrite
15780	? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_READ)
15781	: EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_WRITE)
15782	: !fIOWrite
15783	? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_READ)
15784	: EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_WRITE),
15785	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
15786	if (!pExitRec)
15787	{
15788	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses in bytes. */
15789	static uint32_t const s_aIOOpAnd[4] = { 0xff, 0xffff, 0, 0xffffffff }; /* AND masks for saving result in AL/AX/EAX. */
15790
15791	uint32_t const cbValue = s_aIOSizes[uIOSize];
15792	uint32_t const cbInstr = pVmxTransient->cbExitInstr;
15793	bool fUpdateRipAlready = false; /* ugly hack, should be temporary. */
15794	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15795	if (fIOString)
15796	{
15797	/*
15798	* INS/OUTS - I/O String instruction.
15799	*
15800	* Use instruction-information if available, otherwise fall back on
15801	* interpreting the instruction.
15802	*/
15803	Log4Func(("cs:rip=%#04x:%#RX64 %#06x/%u %c str\n", pCtx->cs.Sel, pCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
15804	AssertReturn(pCtx->dx == uIOPort, VERR_VMX_IPE_2);
15805	bool const fInsOutsInfo = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_INS_OUTS);
15806	if (fInsOutsInfo)
15807	{
15808	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15809	AssertReturn(pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize <= 2, VERR_VMX_IPE_3);
15810	AssertCompile(IEMMODE_16BIT == 0 && IEMMODE_32BIT == 1 && IEMMODE_64BIT == 2);
15811	IEMMODE const enmAddrMode = (IEMMODE)pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize;
15812	bool const fRep = VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual);
15813	if (fIOWrite)
15814	rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, fRep, cbInstr,
15815	pVmxTransient->ExitInstrInfo.StrIo.iSegReg, true /fIoChecked/);
15816	else
15817	{
15818	/*
15819	* The segment prefix for INS cannot be overridden and is always ES. We can safely assume X86_SREG_ES.
15820	* Hence "iSegReg" field is undefined in the instruction-information field in VT-x for INS.
15821	* See Intel Instruction spec. for "INS".
15822	* See Intel spec. Table 27-8 "Format of the VM-Exit Instruction-Information Field as Used for INS and OUTS".
15823	*/
15824	rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, fRep, cbInstr, true /fIoChecked/);
15825	}
15826	}
15827	else
15828	rcStrict = IEMExecOne(pVCpu);
15829
15830	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
15831	fUpdateRipAlready = true;
15832	}
15833	else
15834	{
15835	/*
15836	* IN/OUT - I/O instruction.
15837	*/
15838	Log4Func(("cs:rip=%04x:%08RX64 %#06x/%u %c\n", pCtx->cs.Sel, pCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
15839	uint32_t const uAndVal = s_aIOOpAnd[uIOSize];
15840	Assert(!VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual));
15841	if (fIOWrite)
15842	{
15843	rcStrict = IOMIOPortWrite(pVM, pVCpu, uIOPort, pCtx->eax & uAndVal, cbValue);
15844	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite);
15845	if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
15846	&& !pCtx->eflags.Bits.u1TF)
15847	rcStrict = EMRZSetPendingIoPortWrite(pVCpu, uIOPort, cbInstr, cbValue, pCtx->eax & uAndVal);
15848	}
15849	else
15850	{
15851	uint32_t u32Result = 0;
15852	rcStrict = IOMIOPortRead(pVM, pVCpu, uIOPort, &u32Result, cbValue);
15853	if (IOM_SUCCESS(rcStrict))
15854	{
15855	/* Save result of I/O IN instr. in AL/AX/EAX. */
15856	pCtx->eax = (pCtx->eax & ~uAndVal) \| (u32Result & uAndVal);
15857	}
15858	if ( rcStrict == VINF_IOM_R3_IOPORT_READ
15859	&& !pCtx->eflags.Bits.u1TF)
15860	rcStrict = EMRZSetPendingIoPortRead(pVCpu, uIOPort, cbInstr, cbValue);
15861	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead);
15862	}
15863	}
15864
15865	if (IOM_SUCCESS(rcStrict))
15866	{
15867	if (!fUpdateRipAlready)
15868	{
15869	hmR0VmxAdvanceGuestRipBy(pVCpu, cbInstr);
15870	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
15871	}
15872
15873	/*
15874	* INS/OUTS with REP prefix updates RFLAGS, can be observed with triple-fault guru
15875	* while booting Fedora 17 64-bit guest.
15876	*
15877	* See Intel Instruction reference for REP/REPE/REPZ/REPNE/REPNZ.
15878	*/
15879	if (fIOString)
15880	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RFLAGS);
15881
15882	/*
15883	* If any I/O breakpoints are armed, we need to check if one triggered
15884	* and take appropriate action.
15885	* Note that the I/O breakpoint type is undefined if CR4.DE is 0.
15886	*/
15887	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_DR7);
15888	AssertRCReturn(rc, rc);
15889
15890	/** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the
15891	* execution engines about whether hyper BPs and such are pending. */
15892	uint32_t const uDr7 = pCtx->dr[7];
15893	if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK)
15894	&& X86_DR7_ANY_RW_IO(uDr7)
15895	&& (pCtx->cr4 & X86_CR4_DE))
15896	\|\| DBGFBpIsHwIoArmed(pVM)))
15897	{
15898	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck);
15899
15900	/* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
15901	VMMRZCallRing3Disable(pVCpu);
15902	HM_DISABLE_PREEMPT(pVCpu);
15903
15904	bool fIsGuestDbgActive = CPUMR0DebugStateMaybeSaveGuest(pVCpu, true /* fDr6 */);
15905
15906	VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pCtx, uIOPort, cbValue);
15907	if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
15908	{
15909	/* Raise #DB. */
15910	if (fIsGuestDbgActive)
15911	ASMSetDR6(pCtx->dr[6]);
15912	if (pCtx->dr[7] != uDr7)
15913	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_GUEST_DR7;
15914
15915	hmR0VmxSetPendingXcptDB(pVCpu);
15916	}
15917	/* rcStrict is VINF_SUCCESS, VINF_IOM_R3_IOPORT_COMMIT_WRITE, or in [VINF_EM_FIRST..VINF_EM_LAST],
15918	however we can ditch VINF_IOM_R3_IOPORT_COMMIT_WRITE as it has VMCPU_FF_IOM as backup. */
15919	else if ( rcStrict2 != VINF_SUCCESS
15920	&& (rcStrict == VINF_SUCCESS \|\| rcStrict2 < rcStrict))
15921	rcStrict = rcStrict2;
15922	AssertCompile(VINF_EM_LAST < VINF_IOM_R3_IOPORT_COMMIT_WRITE);
15923
15924	HM_RESTORE_PREEMPT();
15925	VMMRZCallRing3Enable(pVCpu);
15926	}
15927	}
15928
15929	#ifdef VBOX_STRICT
15930	if ( rcStrict == VINF_IOM_R3_IOPORT_READ
15931	\|\| rcStrict == VINF_EM_PENDING_R3_IOPORT_READ)
15932	Assert(!fIOWrite);
15933	else if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
15934	\|\| rcStrict == VINF_IOM_R3_IOPORT_COMMIT_WRITE
15935	\|\| rcStrict == VINF_EM_PENDING_R3_IOPORT_WRITE)
15936	Assert(fIOWrite);
15937	else
15938	{
15939	# if 0 /** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST
15940	* statuses, that the VMM device and some others may return. See
15941	* IOM_SUCCESS() for guidance. */
15942	AssertMsg( RT_FAILURE(rcStrict)
15943	\|\| rcStrict == VINF_SUCCESS
15944	\|\| rcStrict == VINF_EM_RAW_EMULATE_INSTR
15945	\|\| rcStrict == VINF_EM_DBG_BREAKPOINT
15946	\|\| rcStrict == VINF_EM_RAW_GUEST_TRAP
15947	\|\| rcStrict == VINF_EM_RAW_TO_R3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15948	# endif
15949	}
15950	#endif
15951	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitIO, y1);
15952	}
15953	else
15954	{
15955	/*
15956	* Frequent exit or something needing probing. Get state and call EMHistoryExec.
15957	*/
15958	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
15959	AssertRCReturn(rc2, rc2);
15960	STAM_COUNTER_INC(!fIOString ? fIOWrite ? &pVCpu->hm.s.StatExitIOWrite : &pVCpu->hm.s.StatExitIORead
15961	: fIOWrite ? &pVCpu->hm.s.StatExitIOStringWrite : &pVCpu->hm.s.StatExitIOStringRead);
15962	Log4(("IOExit/%u: %04x:%08RX64: %s%s%s %#x LB %u -> EMHistoryExec\n",
15963	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
15964	VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual) ? "REP " : "",
15965	fIOWrite ? "OUT" : "IN", fIOString ? "S" : "", uIOPort, uIOSize));
15966
15967	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
15968	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
15969
15970	Log4(("IOExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
15971	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
15972	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
15973	}
15974	return rcStrict;
15975	}
15976
15977
15978	/**
15979	* VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH). Unconditional
15980	* VM-exit.
15981	*/
15982	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitch(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15983	{
15984	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15985
15986	/* Check if this task-switch occurred while delivery an event through the guest IDT. */
15987	hmR0VmxReadExitQualVmcs(pVmxTransient);
15988	if (VMX_EXIT_QUAL_TASK_SWITCH_TYPE(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_TASK_SWITCH_TYPE_IDT)
15989	{
15990	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
15991	if (VMX_IDT_VECTORING_INFO_IS_VALID(pVmxTransient->uIdtVectoringInfo))
15992	{
15993	uint32_t uErrCode;
15994	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(pVmxTransient->uIdtVectoringInfo))
15995	{
15996	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
15997	uErrCode = pVmxTransient->uIdtVectoringErrorCode;
15998	}
15999	else
16000	uErrCode = 0;
16001
16002	RTGCUINTPTR GCPtrFaultAddress;
16003	if (VMX_IDT_VECTORING_INFO_IS_XCPT_PF(pVmxTransient->uIdtVectoringInfo))
16004	GCPtrFaultAddress = pVCpu->cpum.GstCtx.cr2;
16005	else
16006	GCPtrFaultAddress = 0;
16007
16008	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16009
16010	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(pVmxTransient->uIdtVectoringInfo),
16011	pVmxTransient->cbExitInstr, uErrCode, GCPtrFaultAddress);
16012
16013	Log4Func(("Pending event. uIntType=%#x uVector=%#x\n", VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo),
16014	VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo)));
16015	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
16016	return VINF_EM_RAW_INJECT_TRPM_EVENT;
16017	}
16018	}
16019
16020	/* Fall back to the interpreter to emulate the task-switch. */
16021	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
16022	return VERR_EM_INTERPRETER;
16023	}
16024
16025
16026	/**
16027	* VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional VM-exit.
16028	*/
16029	HMVMX_EXIT_DECL hmR0VmxExitMtf(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16030	{
16031	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16032
16033	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16034	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_MONITOR_TRAP_FLAG;
16035	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
16036	AssertRC(rc);
16037	return VINF_EM_DBG_STEPPED;
16038	}
16039
16040
16041	/**
16042	* VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional VM-exit.
16043	*/
16044	HMVMX_EXIT_DECL hmR0VmxExitApicAccess(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16045	{
16046	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16047	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitApicAccess);
16048
16049	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16050	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16051	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16052	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16053	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16054
16055	/*
16056	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16057	*/
16058	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16059	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16060	{
16061	/* For some crazy guest, if an event delivery causes an APIC-access VM-exit, go to instruction emulation. */
16062	if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
16063	{
16064	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
16065	return VINF_EM_RAW_INJECT_TRPM_EVENT;
16066	}
16067	}
16068	else
16069	{
16070	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16071	return rcStrict;
16072	}
16073
16074	/* IOMMIOPhysHandler() below may call into IEM, save the necessary state. */
16075	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16076	hmR0VmxReadExitQualVmcs(pVmxTransient);
16077	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16078	AssertRCReturn(rc, rc);
16079
16080	/* See Intel spec. 27-6 "Exit Qualifications for APIC-access VM-exits from Linear Accesses & Guest-Phyiscal Addresses" */
16081	uint32_t const uAccessType = VMX_EXIT_QUAL_APIC_ACCESS_TYPE(pVmxTransient->uExitQual);
16082	switch (uAccessType)
16083	{
16084	case VMX_APIC_ACCESS_TYPE_LINEAR_WRITE:
16085	case VMX_APIC_ACCESS_TYPE_LINEAR_READ:
16086	{
16087	AssertMsg( !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
16088	\|\| VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual) != XAPIC_OFF_TPR,
16089	("hmR0VmxExitApicAccess: can't access TPR offset while using TPR shadowing.\n"));
16090
16091	RTGCPHYS GCPhys = pVCpu->hm.s.vmx.u64GstMsrApicBase; /* Always up-to-date, as it is not part of the VMCS. */
16092	GCPhys &= PAGE_BASE_GC_MASK;
16093	GCPhys += VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual);
16094	Log4Func(("Linear access uAccessType=%#x GCPhys=%#RGp Off=%#x\n", uAccessType, GCPhys,
16095	VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual)));
16096
16097	rcStrict = IOMR0MmioPhysHandler(pVCpu->CTX_SUFF(pVM), pVCpu,
16098	uAccessType == VMX_APIC_ACCESS_TYPE_LINEAR_READ ? 0 : X86_TRAP_PF_RW, GCPhys);
16099	Log4Func(("IOMMMIOPhysHandler returned %Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
16100	if ( rcStrict == VINF_SUCCESS
16101	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16102	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16103	{
16104	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS
16105	\| HM_CHANGED_GUEST_APIC_TPR);
16106	rcStrict = VINF_SUCCESS;
16107	}
16108	break;
16109	}
16110
16111	default:
16112	{
16113	Log4Func(("uAccessType=%#x\n", uAccessType));
16114	rcStrict = VINF_EM_RAW_EMULATE_INSTR;
16115	break;
16116	}
16117	}
16118
16119	if (rcStrict != VINF_SUCCESS)
16120	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchApicAccessToR3);
16121	return rcStrict;
16122	}
16123
16124
16125	/**
16126	* VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX). Conditional
16127	* VM-exit.
16128	*/
16129	HMVMX_EXIT_DECL hmR0VmxExitMovDRx(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16130	{
16131	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16132	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16133
16134	/*
16135	* We might also get this VM-exit if the nested-guest isn't intercepting MOV DRx accesses.
16136	* In such a case, rather than disabling MOV DRx intercepts and resuming execution, we
16137	* must emulate the MOV DRx access.
16138	*/
16139	if (!pVmxTransient->fIsNestedGuest)
16140	{
16141	/* We should -not- get this VM-exit if the guest's debug registers were active. */
16142	if (pVmxTransient->fWasGuestDebugStateActive)
16143	{
16144	AssertMsgFailed(("Unexpected MOV DRx exit\n"));
16145	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
16146	}
16147
16148	if ( !pVCpu->hm.s.fSingleInstruction
16149	&& !pVmxTransient->fWasHyperDebugStateActive)
16150	{
16151	Assert(!DBGFIsStepping(pVCpu));
16152	Assert(pVmcsInfo->u32XcptBitmap & RT_BIT(X86_XCPT_DB));
16153
16154	/* Don't intercept MOV DRx any more. */
16155	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_MOV_DR_EXIT;
16156	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
16157	AssertRC(rc);
16158
16159	/* We're playing with the host CPU state here, make sure we can't preempt or longjmp. */
16160	VMMRZCallRing3Disable(pVCpu);
16161	HM_DISABLE_PREEMPT(pVCpu);
16162
16163	/* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */
16164	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
16165	Assert(CPUMIsGuestDebugStateActive(pVCpu));
16166
16167	HM_RESTORE_PREEMPT();
16168	VMMRZCallRing3Enable(pVCpu);
16169
16170	#ifdef VBOX_WITH_STATISTICS
16171	hmR0VmxReadExitQualVmcs(pVmxTransient);
16172	if (VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_DRX_DIRECTION_WRITE)
16173	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
16174	else
16175	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
16176	#endif
16177	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch);
16178	return VINF_SUCCESS;
16179	}
16180	}
16181
16182	/*
16183	* EMInterpretDRx[Write\|Read]() calls CPUMIsGuestIn64BitCode() which requires EFER MSR, CS.
16184	* The EFER MSR is always up-to-date.
16185	* Update the segment registers and DR7 from the CPU.
16186	*/
16187	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16188	hmR0VmxReadExitQualVmcs(pVmxTransient);
16189	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_DR7);
16190	AssertRCReturn(rc, rc);
16191	Log4Func(("cs:rip=%#04x:%#RX64\n", pCtx->cs.Sel, pCtx->rip));
16192
16193	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16194	if (VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_DRX_DIRECTION_WRITE)
16195	{
16196	rc = EMInterpretDRxWrite(pVM, pVCpu, CPUMCTX2CORE(pCtx),
16197	VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual),
16198	VMX_EXIT_QUAL_DRX_GENREG(pVmxTransient->uExitQual));
16199	if (RT_SUCCESS(rc))
16200	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
16201	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
16202	}
16203	else
16204	{
16205	rc = EMInterpretDRxRead(pVM, pVCpu, CPUMCTX2CORE(pCtx),
16206	VMX_EXIT_QUAL_DRX_GENREG(pVmxTransient->uExitQual),
16207	VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual));
16208	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
16209	}
16210
16211	Assert(rc == VINF_SUCCESS \|\| rc == VERR_EM_INTERPRETER);
16212	if (RT_SUCCESS(rc))
16213	{
16214	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
16215	AssertRCReturn(rc2, rc2);
16216	return VINF_SUCCESS;
16217	}
16218	return rc;
16219	}
16220
16221
16222	/**
16223	* VM-exit handler for EPT misconfiguration (VMX_EXIT_EPT_MISCONFIG).
16224	* Conditional VM-exit.
16225	*/
16226	HMVMX_EXIT_DECL hmR0VmxExitEptMisconfig(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16227	{
16228	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16229	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
16230
16231	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16232	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16233	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16234	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16235	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16236
16237	/*
16238	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16239	*/
16240	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16241	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16242	{
16243	/*
16244	* In the unlikely case where delivering an event causes an EPT misconfig (MMIO), go back to
16245	* instruction emulation to inject the original event. Otherwise, injecting the original event
16246	* using hardware-assisted VMX would trigger the same EPT misconfig VM-exit again.
16247	*/
16248	if (!pVCpu->hm.s.Event.fPending)
16249	{ /* likely */ }
16250	else
16251	{
16252	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
16253	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16254	/** @todo NSTVMX: Think about how this should be handled. */
16255	if (pVmxTransient->fIsNestedGuest)
16256	return VERR_VMX_IPE_3;
16257	#endif
16258	return VINF_EM_RAW_INJECT_TRPM_EVENT;
16259	}
16260	}
16261	else
16262	{
16263	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16264	return rcStrict;
16265	}
16266
16267	/*
16268	* Get sufficient state and update the exit history entry.
16269	*/
16270	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16271	hmR0VmxReadGuestPhysicalAddrVmcs(pVmxTransient);
16272	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16273	AssertRCReturn(rc, rc);
16274
16275	RTGCPHYS const GCPhys = pVmxTransient->uGuestPhysicalAddr;
16276	PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
16277	EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_MMIO),
16278	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
16279	if (!pExitRec)
16280	{
16281	/*
16282	* If we succeed, resume guest execution.
16283	* If we fail in interpreting the instruction because we couldn't get the guest physical address
16284	* of the page containing the instruction via the guest's page tables (we would invalidate the guest page
16285	* in the host TLB), resume execution which would cause a guest page fault to let the guest handle this
16286	* weird case. See @bugref{6043}.
16287	*/
16288	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16289	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16290	/** @todo bird: We can probably just go straight to IOM here and assume that
16291	* it's MMIO, then fall back on PGM if that hunch didn't work out so
16292	* well. However, we need to address that aliasing workarounds that
16293	* PGMR0Trap0eHandlerNPMisconfig implements. So, some care is needed.
16294	*
16295	* Might also be interesting to see if we can get this done more or
16296	* less locklessly inside IOM. Need to consider the lookup table
16297	* updating and use a bit more carefully first (or do all updates via
16298	* rendezvous) */
16299	rcStrict = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, PGMMODE_EPT, CPUMCTX2CORE(pCtx), GCPhys, UINT32_MAX);
16300	Log4Func(("At %#RGp RIP=%#RX64 rc=%Rrc\n", GCPhys, pCtx->rip, VBOXSTRICTRC_VAL(rcStrict)));
16301	if ( rcStrict == VINF_SUCCESS
16302	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16303	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16304	{
16305	/* Successfully handled MMIO operation. */
16306	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS
16307	\| HM_CHANGED_GUEST_APIC_TPR);
16308	rcStrict = VINF_SUCCESS;
16309	}
16310	}
16311	else
16312	{
16313	/*
16314	* Frequent exit or something needing probing. Call EMHistoryExec.
16315	*/
16316	Log4(("EptMisscfgExit/%u: %04x:%08RX64: %RGp -> EMHistoryExec\n",
16317	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, GCPhys));
16318
16319	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
16320	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16321
16322	Log4(("EptMisscfgExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
16323	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
16324	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
16325	}
16326	return rcStrict;
16327	}
16328
16329
16330	/**
16331	* VM-exit handler for EPT violation (VMX_EXIT_EPT_VIOLATION). Conditional
16332	* VM-exit.
16333	*/
16334	HMVMX_EXIT_DECL hmR0VmxExitEptViolation(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16335	{
16336	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16337	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
16338
16339	hmR0VmxReadExitQualVmcs(pVmxTransient);
16340	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16341	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16342	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16343	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16344	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16345
16346	/*
16347	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16348	*/
16349	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16350	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16351	{
16352	/*
16353	* If delivery of an event causes an EPT violation (true nested #PF and not MMIO),
16354	* we shall resolve the nested #PF and re-inject the original event.
16355	*/
16356	if (pVCpu->hm.s.Event.fPending)
16357	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflectNPF);
16358	}
16359	else
16360	{
16361	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16362	return rcStrict;
16363	}
16364
16365	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16366	hmR0VmxReadGuestPhysicalAddrVmcs(pVmxTransient);
16367	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16368	AssertRCReturn(rc, rc);
16369
16370	RTGCPHYS const GCPhys = pVmxTransient->uGuestPhysicalAddr;
16371	uint64_t const uExitQual = pVmxTransient->uExitQual;
16372	AssertMsg(((pVmxTransient->uExitQual >> 7) & 3) != 2, ("%#RX64", uExitQual));
16373
16374	RTGCUINT uErrorCode = 0;
16375	if (uExitQual & VMX_EXIT_QUAL_EPT_ACCESS_INSTR_FETCH)
16376	uErrorCode \|= X86_TRAP_PF_ID;
16377	if (uExitQual & VMX_EXIT_QUAL_EPT_ACCESS_WRITE)
16378	uErrorCode \|= X86_TRAP_PF_RW;
16379	if (uExitQual & (VMX_EXIT_QUAL_EPT_ENTRY_READ \| VMX_EXIT_QUAL_EPT_ENTRY_WRITE \| VMX_EXIT_QUAL_EPT_ENTRY_EXECUTE))
16380	uErrorCode \|= X86_TRAP_PF_P;
16381
16382	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16383	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16384	Log4Func(("at %#RX64 (%#RX64 errcode=%#x) cs:rip=%#04x:%#RX64\n", GCPhys, uExitQual, uErrorCode, pCtx->cs.Sel, pCtx->rip));
16385
16386	/*
16387	* Handle the pagefault trap for the nested shadow table.
16388	*/
16389	TRPMAssertXcptPF(pVCpu, GCPhys, uErrorCode);
16390	rcStrict = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, PGMMODE_EPT, uErrorCode, CPUMCTX2CORE(pCtx), GCPhys);
16391	TRPMResetTrap(pVCpu);
16392
16393	/* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}. */
16394	if ( rcStrict == VINF_SUCCESS
16395	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16396	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16397	{
16398	/* Successfully synced our nested page tables. */
16399	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf);
16400	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS);
16401	return VINF_SUCCESS;
16402	}
16403
16404	Log4Func(("EPT return to ring-3 rcStrict2=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
16405	return rcStrict;
16406	}
16407
16408
16409	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16410	/**
16411	* VM-exit handler for VMCLEAR (VMX_EXIT_VMCLEAR). Unconditional VM-exit.
16412	*/
16413	HMVMX_EXIT_DECL hmR0VmxExitVmclear(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16414	{
16415	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16416
16417	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16418	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16419	hmR0VmxReadExitQualVmcs(pVmxTransient);
16420	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16421	\| CPUMCTX_EXTRN_HWVIRT
16422	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16423	AssertRCReturn(rc, rc);
16424
16425	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16426
16427	VMXVEXITINFO ExitInfo;
16428	RT_ZERO(ExitInfo);
16429	ExitInfo.uReason = pVmxTransient->uExitReason;
16430	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16431	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16432	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16433	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16434
16435	VBOXSTRICTRC rcStrict = IEMExecDecodedVmclear(pVCpu, &ExitInfo);
16436	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16437	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16438	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16439	{
16440	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16441	rcStrict = VINF_SUCCESS;
16442	}
16443	return rcStrict;
16444	}
16445
16446
16447	/**
16448	* VM-exit handler for VMLAUNCH (VMX_EXIT_VMLAUNCH). Unconditional VM-exit.
16449	*/
16450	HMVMX_EXIT_DECL hmR0VmxExitVmlaunch(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16451	{
16452	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16453
16454	/* Import the entire VMCS state for now as we would be switching VMCS on successful VMLAUNCH,
16455	otherwise we could import just IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK. */
16456	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16457	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16458	AssertRCReturn(rc, rc);
16459
16460	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16461
16462	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
16463	VBOXSTRICTRC rcStrict = IEMExecDecodedVmlaunchVmresume(pVCpu, pVmxTransient->cbExitInstr, VMXINSTRID_VMLAUNCH);
16464	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
16465	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16466	{
16467	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16468	if (CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
16469	rcStrict = VINF_VMX_VMLAUNCH_VMRESUME;
16470	}
16471	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
16472	return rcStrict;
16473	}
16474
16475
16476	/**
16477	* VM-exit handler for VMPTRLD (VMX_EXIT_VMPTRLD). Unconditional VM-exit.
16478	*/
16479	HMVMX_EXIT_DECL hmR0VmxExitVmptrld(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16480	{
16481	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16482
16483	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16484	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16485	hmR0VmxReadExitQualVmcs(pVmxTransient);
16486	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16487	\| CPUMCTX_EXTRN_HWVIRT
16488	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16489	AssertRCReturn(rc, rc);
16490
16491	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16492
16493	VMXVEXITINFO ExitInfo;
16494	RT_ZERO(ExitInfo);
16495	ExitInfo.uReason = pVmxTransient->uExitReason;
16496	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16497	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16498	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16499	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16500
16501	VBOXSTRICTRC rcStrict = IEMExecDecodedVmptrld(pVCpu, &ExitInfo);
16502	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16503	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16504	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16505	{
16506	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16507	rcStrict = VINF_SUCCESS;
16508	}
16509	return rcStrict;
16510	}
16511
16512
16513	/**
16514	* VM-exit handler for VMPTRST (VMX_EXIT_VMPTRST). Unconditional VM-exit.
16515	*/
16516	HMVMX_EXIT_DECL hmR0VmxExitVmptrst(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16517	{
16518	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16519
16520	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16521	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16522	hmR0VmxReadExitQualVmcs(pVmxTransient);
16523	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16524	\| CPUMCTX_EXTRN_HWVIRT
16525	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16526	AssertRCReturn(rc, rc);
16527
16528	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16529
16530	VMXVEXITINFO ExitInfo;
16531	RT_ZERO(ExitInfo);
16532	ExitInfo.uReason = pVmxTransient->uExitReason;
16533	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16534	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16535	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16536	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_WRITE, &ExitInfo.GCPtrEffAddr);
16537
16538	VBOXSTRICTRC rcStrict = IEMExecDecodedVmptrst(pVCpu, &ExitInfo);
16539	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16540	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16541	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16542	{
16543	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16544	rcStrict = VINF_SUCCESS;
16545	}
16546	return rcStrict;
16547	}
16548
16549
16550	/**
16551	* VM-exit handler for VMREAD (VMX_EXIT_VMREAD). Conditional VM-exit.
16552	*/
16553	HMVMX_EXIT_DECL hmR0VmxExitVmread(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16554	{
16555	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16556
16557	/*
16558	* Strictly speaking we should not get VMREAD VM-exits for shadow VMCS fields and
16559	* thus might not need to import the shadow VMCS state, it's safer just in case
16560	* code elsewhere dares look at unsynced VMCS fields.
16561	*/
16562	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16563	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16564	hmR0VmxReadExitQualVmcs(pVmxTransient);
16565	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16566	\| CPUMCTX_EXTRN_HWVIRT
16567	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16568	AssertRCReturn(rc, rc);
16569
16570	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16571
16572	VMXVEXITINFO ExitInfo;
16573	RT_ZERO(ExitInfo);
16574	ExitInfo.uReason = pVmxTransient->uExitReason;
16575	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16576	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16577	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16578	if (!ExitInfo.InstrInfo.VmreadVmwrite.fIsRegOperand)
16579	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_WRITE, &ExitInfo.GCPtrEffAddr);
16580
16581	VBOXSTRICTRC rcStrict = IEMExecDecodedVmread(pVCpu, &ExitInfo);
16582	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16583	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16584	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16585	{
16586	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16587	rcStrict = VINF_SUCCESS;
16588	}
16589	return rcStrict;
16590	}
16591
16592
16593	/**
16594	* VM-exit handler for VMRESUME (VMX_EXIT_VMRESUME). Unconditional VM-exit.
16595	*/
16596	HMVMX_EXIT_DECL hmR0VmxExitVmresume(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16597	{
16598	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16599
16600	/* Import the entire VMCS state for now as we would be switching VMCS on successful VMRESUME,
16601	otherwise we could import just IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK. */
16602	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16603	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16604	AssertRCReturn(rc, rc);
16605
16606	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16607
16608	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
16609	VBOXSTRICTRC rcStrict = IEMExecDecodedVmlaunchVmresume(pVCpu, pVmxTransient->cbExitInstr, VMXINSTRID_VMRESUME);
16610	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
16611	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16612	{
16613	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16614	if (CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
16615	rcStrict = VINF_VMX_VMLAUNCH_VMRESUME;
16616	}
16617	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
16618	return rcStrict;
16619	}
16620
16621
16622	/**
16623	* VM-exit handler for VMWRITE (VMX_EXIT_VMWRITE). Conditional VM-exit.
16624	*/
16625	HMVMX_EXIT_DECL hmR0VmxExitVmwrite(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16626	{
16627	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16628
16629	/*
16630	* Although we should not get VMWRITE VM-exits for shadow VMCS fields, since our HM hook
16631	* gets invoked when IEM's VMWRITE instruction emulation modifies the current VMCS and it
16632	* flags re-loading the entire shadow VMCS, we should save the entire shadow VMCS here.
16633	*/
16634	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16635	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16636	hmR0VmxReadExitQualVmcs(pVmxTransient);
16637	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16638	\| CPUMCTX_EXTRN_HWVIRT
16639	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16640	AssertRCReturn(rc, rc);
16641
16642	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16643
16644	VMXVEXITINFO ExitInfo;
16645	RT_ZERO(ExitInfo);
16646	ExitInfo.uReason = pVmxTransient->uExitReason;
16647	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16648	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16649	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16650	if (!ExitInfo.InstrInfo.VmreadVmwrite.fIsRegOperand)
16651	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16652
16653	VBOXSTRICTRC rcStrict = IEMExecDecodedVmwrite(pVCpu, &ExitInfo);
16654	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16655	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16656	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16657	{
16658	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16659	rcStrict = VINF_SUCCESS;
16660	}
16661	return rcStrict;
16662	}
16663
16664
16665	/**
16666	* VM-exit handler for VMXOFF (VMX_EXIT_VMXOFF). Unconditional VM-exit.
16667	*/
16668	HMVMX_EXIT_DECL hmR0VmxExitVmxoff(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16669	{
16670	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16671
16672	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16673	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR4
16674	\| CPUMCTX_EXTRN_HWVIRT
16675	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
16676	AssertRCReturn(rc, rc);
16677
16678	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16679
16680	VBOXSTRICTRC rcStrict = IEMExecDecodedVmxoff(pVCpu, pVmxTransient->cbExitInstr);
16681	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16682	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_HWVIRT);
16683	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16684	{
16685	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16686	rcStrict = VINF_SUCCESS;
16687	}
16688	return rcStrict;
16689	}
16690
16691
16692	/**
16693	* VM-exit handler for VMXON (VMX_EXIT_VMXON). Unconditional VM-exit.
16694	*/
16695	HMVMX_EXIT_DECL hmR0VmxExitVmxon(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16696	{
16697	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16698
16699	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16700	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16701	hmR0VmxReadExitQualVmcs(pVmxTransient);
16702	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16703	\| CPUMCTX_EXTRN_HWVIRT
16704	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16705	AssertRCReturn(rc, rc);
16706
16707	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16708
16709	VMXVEXITINFO ExitInfo;
16710	RT_ZERO(ExitInfo);
16711	ExitInfo.uReason = pVmxTransient->uExitReason;
16712	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16713	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16714	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16715	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16716
16717	VBOXSTRICTRC rcStrict = IEMExecDecodedVmxon(pVCpu, &ExitInfo);
16718	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16719	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16720	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16721	{
16722	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16723	rcStrict = VINF_SUCCESS;
16724	}
16725	return rcStrict;
16726	}
16727
16728
16729	/**
16730	* VM-exit handler for INVVPID (VMX_EXIT_INVVPID). Unconditional VM-exit.
16731	*/
16732	HMVMX_EXIT_DECL hmR0VmxExitInvvpid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16733	{
16734	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16735
16736	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16737	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16738	hmR0VmxReadExitQualVmcs(pVmxTransient);
16739	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16740	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16741	AssertRCReturn(rc, rc);
16742
16743	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16744
16745	VMXVEXITINFO ExitInfo;
16746	RT_ZERO(ExitInfo);
16747	ExitInfo.uReason = pVmxTransient->uExitReason;
16748	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16749	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16750	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16751	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16752
16753	VBOXSTRICTRC rcStrict = IEMExecDecodedInvvpid(pVCpu, &ExitInfo);
16754	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16755	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16756	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16757	{
16758	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16759	rcStrict = VINF_SUCCESS;
16760	}
16761	return rcStrict;
16762	}
16763	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
16764	/** @} */
16765
16766
16767	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16768	/** @name Nested-guest VM-exit handlers.
16769	* @{
16770	*/
16771	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16772	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Nested-guest VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16773	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16774
16775	/**
16776	* Nested-guest VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI).
16777	* Conditional VM-exit.
16778	*/
16779	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmiNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16780	{
16781	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16782
16783	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16784
16785	uint64_t const uExitIntInfo = pVmxTransient->uExitIntInfo;
16786	uint32_t const uExitIntType = VMX_EXIT_INT_INFO_TYPE(uExitIntInfo);
16787	Assert(VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo));
16788
16789	switch (uExitIntType)
16790	{
16791	/*
16792	* Physical NMIs:
16793	* We shouldn't direct host physical NMIs to the nested-guest. Dispatch it to the host.
16794	*/
16795	case VMX_EXIT_INT_INFO_TYPE_NMI:
16796	return hmR0VmxExitHostNmi(pVCpu, pVmxTransient->pVmcsInfo);
16797
16798	/*
16799	* Hardware exceptions,
16800	* Software exceptions,
16801	* Privileged software exceptions:
16802	* Figure out if the exception must be delivered to the guest or the nested-guest.
16803	*/
16804	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
16805	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
16806	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
16807	{
16808	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16809	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16810	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16811	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16812
16813	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16814	bool const fIntercept = CPUMIsGuestVmxXcptInterceptSet(pCtx, VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo),
16815	pVmxTransient->uExitIntErrorCode);
16816	if (fIntercept)
16817	{
16818	/* Exit qualification is required for debug and page-fault exceptions. */
16819	hmR0VmxReadExitQualVmcs(pVmxTransient);
16820
16821	/*
16822	* For VM-exits due to software exceptions (those generated by INT3 or INTO) and privileged
16823	* software exceptions (those generated by INT1/ICEBP) we need to supply the VM-exit instruction
16824	* length. However, if delivery of a software interrupt, software exception or privileged
16825	* software exception causes a VM-exit, that too provides the VM-exit instruction length.
16826	*/
16827	VMXVEXITINFO ExitInfo;
16828	RT_ZERO(ExitInfo);
16829	ExitInfo.uReason = pVmxTransient->uExitReason;
16830	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16831	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16832
16833	VMXVEXITEVENTINFO ExitEventInfo;
16834	RT_ZERO(ExitEventInfo);
16835	ExitEventInfo.uExitIntInfo = pVmxTransient->uExitIntInfo;
16836	ExitEventInfo.uExitIntErrCode = pVmxTransient->uExitIntErrorCode;
16837	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
16838	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
16839
16840	#ifdef DEBUG_ramshankar
16841	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16842	Log4Func(("exit_int_info=%#RX32 err_code=%#RX32 exit_qual=%#RX64\n", pVmxTransient->uExitIntInfo,
16843	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual));
16844	if (VMX_IDT_VECTORING_INFO_IS_VALID(pVmxTransient->uIdtVectoringInfo))
16845	{
16846	Log4Func(("idt_info=%#RX32 idt_errcode=%#RX32 cr2=%#RX64\n", pVmxTransient->uIdtVectoringInfo,
16847	pVmxTransient->uIdtVectoringErrorCode, pCtx->cr2));
16848	}
16849	#endif
16850	return IEMExecVmxVmexitXcpt(pVCpu, &ExitInfo, &ExitEventInfo);
16851	}
16852
16853	/* Nested paging is currently a requirement, otherwise we would need to handle shadow #PFs in hmR0VmxExitXcptPF. */
16854	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
16855	return hmR0VmxExitXcpt(pVCpu, pVmxTransient);
16856	}
16857
16858	/*
16859	* Software interrupts:
16860	* VM-exits cannot be caused by software interrupts.
16861	*
16862	* External interrupts:
16863	* This should only happen when "acknowledge external interrupts on VM-exit"
16864	* control is set. However, we never set this when executing a guest or
16865	* nested-guest. For nested-guests it is emulated while injecting interrupts into
16866	* the guest.
16867	*/
16868	case VMX_EXIT_INT_INFO_TYPE_SW_INT:
16869	case VMX_EXIT_INT_INFO_TYPE_EXT_INT:
16870	default:
16871	{
16872	pVCpu->hm.s.u32HMError = pVmxTransient->uExitIntInfo;
16873	return VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
16874	}
16875	}
16876	}
16877
16878
16879	/**
16880	* Nested-guest VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT).
16881	* Unconditional VM-exit.
16882	*/
16883	HMVMX_EXIT_DECL hmR0VmxExitTripleFaultNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16884	{
16885	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16886	return IEMExecVmxVmexitTripleFault(pVCpu);
16887	}
16888
16889
16890	/**
16891	* Nested-guest VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
16892	*/
16893	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindowNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16894	{
16895	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16896
16897	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INT_WINDOW_EXIT))
16898	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
16899	return hmR0VmxExitIntWindow(pVCpu, pVmxTransient);
16900	}
16901
16902
16903	/**
16904	* Nested-guest VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
16905	*/
16906	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindowNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16907	{
16908	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16909
16910	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_NMI_WINDOW_EXIT))
16911	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
16912	return hmR0VmxExitIntWindow(pVCpu, pVmxTransient);
16913	}
16914
16915
16916	/**
16917	* Nested-guest VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH).
16918	* Unconditional VM-exit.
16919	*/
16920	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitchNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16921	{
16922	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16923
16924	hmR0VmxReadExitQualVmcs(pVmxTransient);
16925	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16926	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16927	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16928
16929	VMXVEXITINFO ExitInfo;
16930	RT_ZERO(ExitInfo);
16931	ExitInfo.uReason = pVmxTransient->uExitReason;
16932	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16933	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16934
16935	VMXVEXITEVENTINFO ExitEventInfo;
16936	RT_ZERO(ExitEventInfo);
16937	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
16938	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
16939	return IEMExecVmxVmexitTaskSwitch(pVCpu, &ExitInfo, &ExitEventInfo);
16940	}
16941
16942
16943	/**
16944	* Nested-guest VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
16945	*/
16946	HMVMX_EXIT_DECL hmR0VmxExitHltNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16947	{
16948	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16949
16950	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_HLT_EXIT))
16951	{
16952	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16953	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16954	}
16955	return hmR0VmxExitHlt(pVCpu, pVmxTransient);
16956	}
16957
16958
16959	/**
16960	* Nested-guest VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
16961	*/
16962	HMVMX_EXIT_DECL hmR0VmxExitInvlpgNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16963	{
16964	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16965
16966	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INVLPG_EXIT))
16967	{
16968	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16969	hmR0VmxReadExitQualVmcs(pVmxTransient);
16970
16971	VMXVEXITINFO ExitInfo;
16972	RT_ZERO(ExitInfo);
16973	ExitInfo.uReason = pVmxTransient->uExitReason;
16974	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16975	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16976	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16977	}
16978	return hmR0VmxExitInvlpg(pVCpu, pVmxTransient);
16979	}
16980
16981
16982	/**
16983	* Nested-guest VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
16984	*/
16985	HMVMX_EXIT_DECL hmR0VmxExitRdpmcNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16986	{
16987	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16988
16989	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDPMC_EXIT))
16990	{
16991	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16992	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16993	}
16994	return hmR0VmxExitRdpmc(pVCpu, pVmxTransient);
16995	}
16996
16997
16998	/**
16999	* Nested-guest VM-exit handler for VMREAD (VMX_EXIT_VMREAD) and VMWRITE
17000	* (VMX_EXIT_VMWRITE). Conditional VM-exit.
17001	*/
17002	HMVMX_EXIT_DECL hmR0VmxExitVmreadVmwriteNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17003	{
17004	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17005
17006	Assert( pVmxTransient->uExitReason == VMX_EXIT_VMREAD
17007	\|\| pVmxTransient->uExitReason == VMX_EXIT_VMWRITE);
17008
17009	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17010
17011	uint8_t const iGReg = pVmxTransient->ExitInstrInfo.VmreadVmwrite.iReg2;
17012	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
17013	uint64_t u64VmcsField = pVCpu->cpum.GstCtx.aGRegs[iGReg].u64;
17014
17015	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
17016	if (!CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
17017	u64VmcsField &= UINT64_C(0xffffffff);
17018
17019	if (CPUMIsGuestVmxVmreadVmwriteInterceptSet(pVCpu, pVmxTransient->uExitReason, u64VmcsField))
17020	{
17021	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17022	hmR0VmxReadExitQualVmcs(pVmxTransient);
17023
17024	VMXVEXITINFO ExitInfo;
17025	RT_ZERO(ExitInfo);
17026	ExitInfo.uReason = pVmxTransient->uExitReason;
17027	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17028	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17029	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17030	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17031	}
17032
17033	if (pVmxTransient->uExitReason == VMX_EXIT_VMREAD)
17034	return hmR0VmxExitVmread(pVCpu, pVmxTransient);
17035	return hmR0VmxExitVmwrite(pVCpu, pVmxTransient);
17036	}
17037
17038
17039	/**
17040	* Nested-guest VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
17041	*/
17042	HMVMX_EXIT_DECL hmR0VmxExitRdtscNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17043	{
17044	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17045
17046	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDTSC_EXIT))
17047	{
17048	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17049	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17050	}
17051
17052	return hmR0VmxExitRdtsc(pVCpu, pVmxTransient);
17053	}
17054
17055
17056	/**
17057	* Nested-guest VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX).
17058	* Conditional VM-exit.
17059	*/
17060	HMVMX_EXIT_DECL hmR0VmxExitMovCRxNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17061	{
17062	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17063
17064	hmR0VmxReadExitQualVmcs(pVmxTransient);
17065	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17066
17067	VBOXSTRICTRC rcStrict;
17068	uint32_t const uAccessType = VMX_EXIT_QUAL_CRX_ACCESS(pVmxTransient->uExitQual);
17069	switch (uAccessType)
17070	{
17071	case VMX_EXIT_QUAL_CRX_ACCESS_WRITE:
17072	{
17073	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
17074	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(pVmxTransient->uExitQual);
17075	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
17076	uint64_t const uNewCrX = pVCpu->cpum.GstCtx.aGRegs[iGReg].u64;
17077
17078	bool fIntercept;
17079	switch (iCrReg)
17080	{
17081	case 0:
17082	case 4:
17083	fIntercept = CPUMIsGuestVmxMovToCr0Cr4InterceptSet(&pVCpu->cpum.GstCtx, iCrReg, uNewCrX);
17084	break;
17085
17086	case 3:
17087	fIntercept = CPUMIsGuestVmxMovToCr3InterceptSet(pVCpu, uNewCrX);
17088	break;
17089
17090	case 8:
17091	fIntercept = CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_CR8_LOAD_EXIT);
17092	break;
17093
17094	default:
17095	fIntercept = false;
17096	break;
17097	}
17098	if (fIntercept)
17099	{
17100	VMXVEXITINFO ExitInfo;
17101	RT_ZERO(ExitInfo);
17102	ExitInfo.uReason = pVmxTransient->uExitReason;
17103	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17104	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17105	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17106	}
17107	else
17108	{
17109	int const rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
17110	AssertRCReturn(rc, rc);
17111	rcStrict = hmR0VmxExitMovToCrX(pVCpu, pVmxTransient->cbExitInstr, iGReg, iCrReg);
17112	}
17113	break;
17114	}
17115
17116	case VMX_EXIT_QUAL_CRX_ACCESS_READ:
17117	{
17118	/*
17119	* CR0/CR4 reads do not cause VM-exits, the read-shadow is used (subject to masking).
17120	* CR2 reads do not cause a VM-exit.
17121	* CR3 reads cause a VM-exit depending on the "CR3 store exiting" control.
17122	* CR8 reads cause a VM-exit depending on the "CR8 store exiting" control.
17123	*/
17124	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
17125	if ( iCrReg == 3
17126	\|\| iCrReg == 8)
17127	{
17128	static const uint32_t s_auCrXReadIntercepts[] = { 0, 0, 0, VMX_PROC_CTLS_CR3_STORE_EXIT, 0,
17129	0, 0, 0, VMX_PROC_CTLS_CR8_STORE_EXIT };
17130	uint32_t const uIntercept = s_auCrXReadIntercepts[iCrReg];
17131	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, uIntercept))
17132	{
17133	VMXVEXITINFO ExitInfo;
17134	RT_ZERO(ExitInfo);
17135	ExitInfo.uReason = pVmxTransient->uExitReason;
17136	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17137	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17138	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17139	}
17140	else
17141	{
17142	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(pVmxTransient->uExitQual);
17143	rcStrict = hmR0VmxExitMovFromCrX(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
17144	}
17145	}
17146	else
17147	{
17148	AssertMsgFailed(("MOV from CR%d VM-exit must not happen\n", iCrReg));
17149	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, iCrReg);
17150	}
17151	break;
17152	}
17153
17154	case VMX_EXIT_QUAL_CRX_ACCESS_CLTS:
17155	{
17156	PCVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
17157	uint64_t const uGstHostMask = pVmcsNstGst->u64Cr0Mask.u;
17158	uint64_t const uReadShadow = pVmcsNstGst->u64Cr0ReadShadow.u;
17159	if ( (uGstHostMask & X86_CR0_TS)
17160	&& (uReadShadow & X86_CR0_TS))
17161	{
17162	VMXVEXITINFO ExitInfo;
17163	RT_ZERO(ExitInfo);
17164	ExitInfo.uReason = pVmxTransient->uExitReason;
17165	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17166	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17167	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17168	}
17169	else
17170	rcStrict = hmR0VmxExitClts(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr);
17171	break;
17172	}
17173
17174	case VMX_EXIT_QUAL_CRX_ACCESS_LMSW: /* LMSW (Load Machine-Status Word into CR0) */
17175	{
17176	RTGCPTR GCPtrEffDst;
17177	uint16_t const uNewMsw = VMX_EXIT_QUAL_CRX_LMSW_DATA(pVmxTransient->uExitQual);
17178	bool const fMemOperand = VMX_EXIT_QUAL_CRX_LMSW_OP_MEM(pVmxTransient->uExitQual);
17179	if (fMemOperand)
17180	{
17181	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
17182	GCPtrEffDst = pVmxTransient->uGuestLinearAddr;
17183	}
17184	else
17185	GCPtrEffDst = NIL_RTGCPTR;
17186
17187	if (CPUMIsGuestVmxLmswInterceptSet(&pVCpu->cpum.GstCtx, uNewMsw))
17188	{
17189	VMXVEXITINFO ExitInfo;
17190	RT_ZERO(ExitInfo);
17191	ExitInfo.uReason = pVmxTransient->uExitReason;
17192	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17193	ExitInfo.u64GuestLinearAddr = GCPtrEffDst;
17194	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17195	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17196	}
17197	else
17198	rcStrict = hmR0VmxExitLmsw(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, uNewMsw, GCPtrEffDst);
17199	break;
17200	}
17201
17202	default:
17203	{
17204	AssertMsgFailed(("Unrecognized Mov CRX access type %#x\n", uAccessType));
17205	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, uAccessType);
17206	}
17207	}
17208
17209	if (rcStrict == VINF_IEM_RAISED_XCPT)
17210	{
17211	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
17212	rcStrict = VINF_SUCCESS;
17213	}
17214	return rcStrict;
17215	}
17216
17217
17218	/**
17219	* Nested-guest VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX).
17220	* Conditional VM-exit.
17221	*/
17222	HMVMX_EXIT_DECL hmR0VmxExitMovDRxNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17223	{
17224	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17225
17226	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MOV_DR_EXIT))
17227	{
17228	hmR0VmxReadExitQualVmcs(pVmxTransient);
17229	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17230
17231	VMXVEXITINFO ExitInfo;
17232	RT_ZERO(ExitInfo);
17233	ExitInfo.uReason = pVmxTransient->uExitReason;
17234	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17235	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17236	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17237	}
17238	return hmR0VmxExitMovDRx(pVCpu, pVmxTransient);
17239	}
17240
17241
17242	/**
17243	* Nested-guest VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR).
17244	* Conditional VM-exit.
17245	*/
17246	HMVMX_EXIT_DECL hmR0VmxExitIoInstrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17247	{
17248	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17249
17250	hmR0VmxReadExitQualVmcs(pVmxTransient);
17251
17252	uint32_t const uIOPort = VMX_EXIT_QUAL_IO_PORT(pVmxTransient->uExitQual);
17253	uint8_t const uIOSize = VMX_EXIT_QUAL_IO_SIZE(pVmxTransient->uExitQual);
17254	AssertReturn(uIOSize <= 3 && uIOSize != 2, VERR_VMX_IPE_1);
17255
17256	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses in bytes. */
17257	uint8_t const cbAccess = s_aIOSizes[uIOSize];
17258	if (CPUMIsGuestVmxIoInterceptSet(pVCpu, uIOPort, cbAccess))
17259	{
17260	/*
17261	* IN/OUT instruction:
17262	* - Provides VM-exit instruction length.
17263	*
17264	* INS/OUTS instruction:
17265	* - Provides VM-exit instruction length.
17266	* - Provides Guest-linear address.
17267	* - Optionally provides VM-exit instruction info (depends on CPU feature).
17268	*/
17269	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
17270	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17271
17272	/* Make sure we don't use stale/uninitialized VMX-transient info. below. */
17273	pVmxTransient->ExitInstrInfo.u = 0;
17274	pVmxTransient->uGuestLinearAddr = 0;
17275
17276	bool const fVmxInsOutsInfo = pVM->cpum.ro.GuestFeatures.fVmxInsOutInfo;
17277	bool const fIOString = VMX_EXIT_QUAL_IO_IS_STRING(pVmxTransient->uExitQual);
17278	if (fIOString)
17279	{
17280	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
17281	if (fVmxInsOutsInfo)
17282	{
17283	Assert(RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_INS_OUTS)); /* Paranoia. */
17284	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17285	}
17286	}
17287
17288	VMXVEXITINFO ExitInfo;
17289	RT_ZERO(ExitInfo);
17290	ExitInfo.uReason = pVmxTransient->uExitReason;
17291	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17292	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17293	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17294	ExitInfo.u64GuestLinearAddr = pVmxTransient->uGuestLinearAddr;
17295	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17296	}
17297	return hmR0VmxExitIoInstr(pVCpu, pVmxTransient);
17298	}
17299
17300
17301	/**
17302	* Nested-guest VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
17303	*/
17304	HMVMX_EXIT_DECL hmR0VmxExitRdmsrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17305	{
17306	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17307
17308	uint32_t fMsrpm;
17309	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_MSR_BITMAPS))
17310	fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap, pVCpu->cpum.GstCtx.ecx);
17311	else
17312	fMsrpm = VMXMSRPM_EXIT_RD;
17313
17314	if (fMsrpm & VMXMSRPM_EXIT_RD)
17315	{
17316	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17317	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17318	}
17319	return hmR0VmxExitRdmsr(pVCpu, pVmxTransient);
17320	}
17321
17322
17323	/**
17324	* Nested-guest VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
17325	*/
17326	HMVMX_EXIT_DECL hmR0VmxExitWrmsrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17327	{
17328	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17329
17330	uint32_t fMsrpm;
17331	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_MSR_BITMAPS))
17332	fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap, pVCpu->cpum.GstCtx.ecx);
17333	else
17334	fMsrpm = VMXMSRPM_EXIT_WR;
17335
17336	if (fMsrpm & VMXMSRPM_EXIT_WR)
17337	{
17338	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17339	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17340	}
17341	return hmR0VmxExitWrmsr(pVCpu, pVmxTransient);
17342	}
17343
17344
17345	/**
17346	* Nested-guest VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
17347	*/
17348	HMVMX_EXIT_DECL hmR0VmxExitMwaitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17349	{
17350	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17351
17352	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MWAIT_EXIT))
17353	{
17354	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17355	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17356	}
17357	return hmR0VmxExitMwait(pVCpu, pVmxTransient);
17358	}
17359
17360
17361	/**
17362	* Nested-guest VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional
17363	* VM-exit.
17364	*/
17365	HMVMX_EXIT_DECL hmR0VmxExitMtfNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17366	{
17367	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17368
17369	/** @todo NSTVMX: Should consider debugging nested-guests using VM debugger. */
17370	hmR0VmxReadGuestPendingDbgXctps(pVmxTransient);
17371	VMXVEXITINFO ExitInfo;
17372	RT_ZERO(ExitInfo);
17373	ExitInfo.uReason = pVmxTransient->uExitReason;
17374	ExitInfo.u64GuestPendingDbgXcpts = pVmxTransient->uGuestPendingDbgXcpts;
17375	return IEMExecVmxVmexitTrapLike(pVCpu, &ExitInfo);
17376	}
17377
17378
17379	/**
17380	* Nested-guest VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
17381	*/
17382	HMVMX_EXIT_DECL hmR0VmxExitMonitorNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17383	{
17384	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17385
17386	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MONITOR_EXIT))
17387	{
17388	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17389	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17390	}
17391	return hmR0VmxExitMonitor(pVCpu, pVmxTransient);
17392	}
17393
17394
17395	/**
17396	* Nested-guest VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
17397	*/
17398	HMVMX_EXIT_DECL hmR0VmxExitPauseNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17399	{
17400	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17401
17402	/** @todo NSTVMX: Think about this more. Does the outer guest need to intercept
17403	* PAUSE when executing a nested-guest? If it does not, we would not need
17404	* to check for the intercepts here. Just call VM-exit... */
17405
17406	/* The CPU would have already performed the necessary CPL checks for PAUSE-loop exiting. */
17407	if ( CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_PAUSE_EXIT)
17408	\|\| CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_PAUSE_LOOP_EXIT))
17409	{
17410	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17411	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17412	}
17413	return hmR0VmxExitPause(pVCpu, pVmxTransient);
17414	}
17415
17416
17417	/**
17418	* Nested-guest VM-exit handler for when the TPR value is lowered below the
17419	* specified threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
17420	*/
17421	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThresholdNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17422	{
17423	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17424
17425	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_TPR_SHADOW))
17426	{
17427	hmR0VmxReadGuestPendingDbgXctps(pVmxTransient);
17428	VMXVEXITINFO ExitInfo;
17429	RT_ZERO(ExitInfo);
17430	ExitInfo.uReason = pVmxTransient->uExitReason;
17431	ExitInfo.u64GuestPendingDbgXcpts = pVmxTransient->uGuestPendingDbgXcpts;
17432	return IEMExecVmxVmexitTrapLike(pVCpu, &ExitInfo);
17433	}
17434	return hmR0VmxExitTprBelowThreshold(pVCpu, pVmxTransient);
17435	}
17436
17437
17438	/**
17439	* Nested-guest VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional
17440	* VM-exit.
17441	*/
17442	HMVMX_EXIT_DECL hmR0VmxExitApicAccessNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17443	{
17444	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17445
17446	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17447	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
17448	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
17449	hmR0VmxReadExitQualVmcs(pVmxTransient);
17450
17451	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_VIRT_APIC_ACCESS));
17452
17453	Log4Func(("at offset %#x type=%u\n", VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual),
17454	VMX_EXIT_QUAL_APIC_ACCESS_TYPE(pVmxTransient->uExitQual)));
17455
17456	VMXVEXITINFO ExitInfo;
17457	RT_ZERO(ExitInfo);
17458	ExitInfo.uReason = pVmxTransient->uExitReason;
17459	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17460	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17461
17462	VMXVEXITEVENTINFO ExitEventInfo;
17463	RT_ZERO(ExitEventInfo);
17464	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
17465	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
17466	return IEMExecVmxVmexitApicAccess(pVCpu, &ExitInfo, &ExitEventInfo);
17467	}
17468
17469
17470	/**
17471	* Nested-guest VM-exit handler for APIC write emulation (VMX_EXIT_APIC_WRITE).
17472	* Conditional VM-exit.
17473	*/
17474	HMVMX_EXIT_DECL hmR0VmxExitApicWriteNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17475	{
17476	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17477
17478	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_APIC_REG_VIRT));
17479	hmR0VmxReadExitQualVmcs(pVmxTransient);
17480	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
17481	}
17482
17483
17484	/**
17485	* Nested-guest VM-exit handler for virtualized EOI (VMX_EXIT_VIRTUALIZED_EOI).
17486	* Conditional VM-exit.
17487	*/
17488	HMVMX_EXIT_DECL hmR0VmxExitVirtEoiNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17489	{
17490	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17491
17492	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_VIRT_INT_DELIVERY));
17493	hmR0VmxReadExitQualVmcs(pVmxTransient);
17494	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
17495	}
17496
17497
17498	/**
17499	* Nested-guest VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
17500	*/
17501	HMVMX_EXIT_DECL hmR0VmxExitRdtscpNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17502	{
17503	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17504
17505	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDTSC_EXIT))
17506	{
17507	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_RDTSCP));
17508	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17509	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17510	}
17511	return hmR0VmxExitRdtscp(pVCpu, pVmxTransient);
17512	}
17513
17514
17515	/**
17516	* Nested-guest VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
17517	*/
17518	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvdNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17519	{
17520	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17521
17522	if (CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_WBINVD_EXIT))
17523	{
17524	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17525	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17526	}
17527	return hmR0VmxExitWbinvd(pVCpu, pVmxTransient);
17528	}
17529
17530
17531	/**
17532	* Nested-guest VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
17533	*/
17534	HMVMX_EXIT_DECL hmR0VmxExitInvpcidNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17535	{
17536	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17537
17538	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INVLPG_EXIT))
17539	{
17540	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_INVPCID));
17541	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17542	hmR0VmxReadExitQualVmcs(pVmxTransient);
17543	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17544
17545	VMXVEXITINFO ExitInfo;
17546	RT_ZERO(ExitInfo);
17547	ExitInfo.uReason = pVmxTransient->uExitReason;
17548	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17549	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17550	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17551	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17552	}
17553	return hmR0VmxExitInvpcid(pVCpu, pVmxTransient);
17554	}
17555
17556
17557	/**
17558	* Nested-guest VM-exit handler for invalid-guest state
17559	* (VMX_EXIT_ERR_INVALID_GUEST_STATE). Error VM-exit.
17560	*/
17561	HMVMX_EXIT_DECL hmR0VmxExitErrInvalidGuestStateNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17562	{
17563	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17564
17565	/*
17566	* Currently this should never happen because we fully emulate VMLAUNCH/VMRESUME in IEM.
17567	* So if it does happen, it indicates a bug possibly in the hardware-assisted VMX code.
17568	* Handle it like it's in an invalid guest state of the outer guest.
17569	*
17570	* When the fast path is implemented, this should be changed to cause the corresponding
17571	* nested-guest VM-exit.
17572	*/
17573	return hmR0VmxExitErrInvalidGuestState(pVCpu, pVmxTransient);
17574	}
17575
17576
17577	/**
17578	* Nested-guest VM-exit handler for instructions that cause VM-exits uncondtionally
17579	* and only provide the instruction length.
17580	*
17581	* Unconditional VM-exit.
17582	*/
17583	HMVMX_EXIT_DECL hmR0VmxExitInstrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17584	{
17585	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17586
17587	#ifdef VBOX_STRICT
17588	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
17589	switch (pVmxTransient->uExitReason)
17590	{
17591	case VMX_EXIT_ENCLS:
17592	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_ENCLS_EXIT));
17593	break;
17594
17595	case VMX_EXIT_VMFUNC:
17596	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_VMFUNC));
17597	break;
17598	}
17599	#endif
17600
17601	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17602	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17603	}
17604
17605
17606	/**
17607	* Nested-guest VM-exit handler for instructions that provide instruction length as
17608	* well as more information.
17609	*
17610	* Unconditional VM-exit.
17611	*/
17612	HMVMX_EXIT_DECL hmR0VmxExitInstrWithInfoNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17613	{
17614	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17615
17616	#ifdef VBOX_STRICT
17617	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
17618	switch (pVmxTransient->uExitReason)
17619	{
17620	case VMX_EXIT_GDTR_IDTR_ACCESS:
17621	case VMX_EXIT_LDTR_TR_ACCESS:
17622	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_DESC_TABLE_EXIT));
17623	break;
17624
17625	case VMX_EXIT_RDRAND:
17626	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_RDRAND_EXIT));
17627	break;
17628
17629	case VMX_EXIT_RDSEED:
17630	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_RDSEED_EXIT));
17631	break;
17632
17633	case VMX_EXIT_XSAVES:
17634	case VMX_EXIT_XRSTORS:
17635	/** @todo NSTVMX: Verify XSS-bitmap. */
17636	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_XSAVES_XRSTORS));
17637	break;
17638
17639	case VMX_EXIT_UMWAIT:
17640	case VMX_EXIT_TPAUSE:
17641	Assert(CPUMIsGuestVmxProcCtlsSet(pCtx, VMX_PROC_CTLS_RDTSC_EXIT));
17642	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_USER_WAIT_PAUSE));
17643	break;
17644
17645	case VMX_EXIT_LOADIWKEY:
17646	Assert(CPUMIsGuestVmxProcCtls3Set(pCtx, VMX_PROC_CTLS3_LOADIWKEY_EXIT));
17647	break;
17648	}
17649	#endif
17650
17651	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17652	hmR0VmxReadExitQualVmcs(pVmxTransient);
17653	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17654
17655	VMXVEXITINFO ExitInfo;
17656	RT_ZERO(ExitInfo);
17657	ExitInfo.uReason = pVmxTransient->uExitReason;
17658	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17659	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17660	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17661	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17662	}
17663
17664	/** @} */
17665	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
17666

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

source: vbox/trunk/src/VBox/VMM/VMMR0/HMVMXR0.cpp@ 92679

Download in other formats: