HMVMXR0.cpp@ 90842

Last change on this file since 90842 was 90842, checked in by vboxsync, 3 years ago
VMM/HMVMXR0: bugref:10080 Cannot assume that a pending TRPM event here always implies it originated from VMLAUNCH/VMRESUME instruction emulation.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 739.4 KB

Line
1	/* $Id: HMVMXR0.cpp 90842 2021-08-24 11:40:31Z 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_HM_VMX_MASK)
127
128	/**
129	* Exception bitmap mask for real-mode guests (real-on-v86).
130	*
131	* We need to intercept all exceptions manually except:
132	* - \#AC and \#DB are always intercepted to prevent the CPU from deadlocking
133	* due to bugs in Intel CPUs.
134	* - \#PF need not be intercepted even in real-mode if we have nested paging
135	* support.
136	*/
137	#define HMVMX_REAL_MODE_XCPT_MASK ( RT_BIT(X86_XCPT_DE) /* always: \| RT_BIT(X86_XCPT_DB) */ \| RT_BIT(X86_XCPT_NMI) \
138	\| RT_BIT(X86_XCPT_BP) \| RT_BIT(X86_XCPT_OF) \| RT_BIT(X86_XCPT_BR) \
139	\| RT_BIT(X86_XCPT_UD) \| RT_BIT(X86_XCPT_NM) \| RT_BIT(X86_XCPT_DF) \
140	\| RT_BIT(X86_XCPT_CO_SEG_OVERRUN) \| RT_BIT(X86_XCPT_TS) \| RT_BIT(X86_XCPT_NP) \
141	\| RT_BIT(X86_XCPT_SS) \| RT_BIT(X86_XCPT_GP) /* RT_BIT(X86_XCPT_PF) */ \
142	\| RT_BIT(X86_XCPT_MF) /* always: \| RT_BIT(X86_XCPT_AC) */ \| RT_BIT(X86_XCPT_MC) \
143	\| RT_BIT(X86_XCPT_XF))
144
145	/** Maximum VM-instruction error number. */
146	#define HMVMX_INSTR_ERROR_MAX 28
147
148	/** Profiling macro. */
149	#ifdef HM_PROFILE_EXIT_DISPATCH
150	# define HMVMX_START_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitDispatch, ed)
151	# define HMVMX_STOP_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitDispatch, ed)
152	#else
153	# define HMVMX_START_EXIT_DISPATCH_PROF() do { } while (0)
154	# define HMVMX_STOP_EXIT_DISPATCH_PROF() do { } while (0)
155	#endif
156
157	/** Assert that preemption is disabled or covered by thread-context hooks. */
158	#define HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu) Assert( VMMR0ThreadCtxHookIsEnabled((a_pVCpu)) \
159	\|\| !RTThreadPreemptIsEnabled(NIL_RTTHREAD))
160
161	/** Assert that we haven't migrated CPUs when thread-context hooks are not
162	* used. */
163	#define HMVMX_ASSERT_CPU_SAFE(a_pVCpu) AssertMsg( VMMR0ThreadCtxHookIsEnabled((a_pVCpu)) \
164	\|\| (a_pVCpu)->hmr0.s.idEnteredCpu == RTMpCpuId(), \
165	("Illegal migration! Entered on CPU %u Current %u\n", \
166	(a_pVCpu)->hmr0.s.idEnteredCpu, RTMpCpuId()))
167
168	/** Asserts that the given CPUMCTX_EXTRN_XXX bits are present in the guest-CPU
169	* context. */
170	#define HMVMX_CPUMCTX_ASSERT(a_pVCpu, a_fExtrnMbz) AssertMsg(!((a_pVCpu)->cpum.GstCtx.fExtrn & (a_fExtrnMbz)), \
171	("fExtrn=%#RX64 fExtrnMbz=%#RX64\n", \
172	(a_pVCpu)->cpum.GstCtx.fExtrn, (a_fExtrnMbz)))
173
174	/** Log the VM-exit reason with an easily visible marker to identify it in a
175	* potential sea of logging data. */
176	#define HMVMX_LOG_EXIT(a_pVCpu, a_uExitReason) \
177	do { \
178	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, \
179	HMGetVmxExitName(a_uExitReason))); \
180	} while (0) \
181
182
183	/*********************************************************************************************************************************
184	* Structures and Typedefs *
185	*********************************************************************************************************************************/
186	/**
187	* VMX per-VCPU transient state.
188	*
189	* A state structure for holding miscellaneous information across
190	* VMX non-root operation and restored after the transition.
191	*
192	* Note: The members are ordered and aligned such that the most
193	* frequently used ones (in the guest execution loop) fall within
194	* the first cache line.
195	*/
196	typedef struct VMXTRANSIENT
197	{
198	/** Mask of currently read VMCS fields; HMVMX_READ_XXX. */
199	uint32_t fVmcsFieldsRead;
200	/** The guest's TPR value used for TPR shadowing. */
201	uint8_t u8GuestTpr;
202	uint8_t abAlignment0[3];
203
204	/** Whether the VM-exit was caused by a page-fault during delivery of an
205	* external interrupt or NMI. */
206	bool fVectoringPF;
207	/** Whether the VM-exit was caused by a page-fault during delivery of a
208	* contributory exception or a page-fault. */
209	bool fVectoringDoublePF;
210	/** Whether the VM-entry failed or not. */
211	bool fVMEntryFailed;
212	/** Whether the TSC_AUX MSR needs to be removed from the auto-load/store MSR
213	* area after VM-exit. */
214	bool fRemoveTscAuxMsr;
215	/** Whether TSC-offsetting and VMX-preemption timer was updated before VM-entry. */
216	bool fUpdatedTscOffsettingAndPreemptTimer;
217	/** Whether we are currently executing a nested-guest. */
218	bool fIsNestedGuest;
219	/** Whether the guest debug state was active at the time of VM-exit. */
220	bool fWasGuestDebugStateActive;
221	/** Whether the hyper debug state was active at the time of VM-exit. */
222	bool fWasHyperDebugStateActive;
223
224	/** The basic VM-exit reason. */
225	uint32_t uExitReason;
226	/** The VM-exit interruption error code. */
227	uint32_t uExitIntErrorCode;
228
229	/** The host's rflags/eflags. */
230	RTCCUINTREG fEFlags;
231
232	/** The VM-exit exit code qualification. */
233	uint64_t uExitQual;
234
235	/** The VMCS info. object. */
236	PVMXVMCSINFO pVmcsInfo;
237
238	/** The VM-exit interruption-information field. */
239	uint32_t uExitIntInfo;
240	/** The VM-exit instruction-length field. */
241	uint32_t cbExitInstr;
242
243	/** The VM-exit instruction-information field. */
244	VMXEXITINSTRINFO ExitInstrInfo;
245	/** IDT-vectoring information field. */
246	uint32_t uIdtVectoringInfo;
247
248	/** IDT-vectoring error code. */
249	uint32_t uIdtVectoringErrorCode;
250	uint32_t u32Alignment0;
251
252	/** The Guest-linear address. */
253	uint64_t uGuestLinearAddr;
254
255	/** The Guest-physical address. */
256	uint64_t uGuestPhysicalAddr;
257
258	/** The Guest pending-debug exceptions. */
259	uint64_t uGuestPendingDbgXcpts;
260
261	/** The VM-entry interruption-information field. */
262	uint32_t uEntryIntInfo;
263	/** The VM-entry exception error code field. */
264	uint32_t uEntryXcptErrorCode;
265
266	/** The VM-entry instruction length field. */
267	uint32_t cbEntryInstr;
268	} VMXTRANSIENT;
269	AssertCompileMemberSize(VMXTRANSIENT, ExitInstrInfo, sizeof(uint32_t));
270	AssertCompileMemberAlignment(VMXTRANSIENT, fVmcsFieldsRead, 8);
271	AssertCompileMemberAlignment(VMXTRANSIENT, fVectoringPF, 8);
272	AssertCompileMemberAlignment(VMXTRANSIENT, uExitReason, 8);
273	AssertCompileMemberAlignment(VMXTRANSIENT, fEFlags, 8);
274	AssertCompileMemberAlignment(VMXTRANSIENT, uExitQual, 8);
275	AssertCompileMemberAlignment(VMXTRANSIENT, pVmcsInfo, 8);
276	AssertCompileMemberAlignment(VMXTRANSIENT, uExitIntInfo, 8);
277	AssertCompileMemberAlignment(VMXTRANSIENT, ExitInstrInfo, 8);
278	AssertCompileMemberAlignment(VMXTRANSIENT, uIdtVectoringErrorCode, 8);
279	AssertCompileMemberAlignment(VMXTRANSIENT, uGuestLinearAddr, 8);
280	AssertCompileMemberAlignment(VMXTRANSIENT, uGuestPhysicalAddr, 8);
281	AssertCompileMemberAlignment(VMXTRANSIENT, uEntryIntInfo, 8);
282	AssertCompileMemberAlignment(VMXTRANSIENT, cbEntryInstr, 8);
283	/** Pointer to VMX transient state. */
284	typedef VMXTRANSIENT *PVMXTRANSIENT;
285	/** Pointer to a const VMX transient state. */
286	typedef const VMXTRANSIENT *PCVMXTRANSIENT;
287
288	/**
289	* VMX page allocation information.
290	*/
291	typedef struct
292	{
293	uint32_t fValid; /*< Whether to allocate this page (e.g, based on a CPU feature). /
294	uint32_t uPadding0; /*< Padding to ensure array of these structs are aligned to a multiple of 8. /
295	PRTHCPHYS pHCPhys; /*< Where to store the host-physical address of the allocation. /
296	PRTR0PTR ppVirt; /*< Where to store the host-virtual address of the allocation. /
297	} VMXPAGEALLOCINFO;
298	/** Pointer to VMX page-allocation info. */
299	typedef VMXPAGEALLOCINFO *PVMXPAGEALLOCINFO;
300	/** Pointer to a const VMX page-allocation info. */
301	typedef const VMXPAGEALLOCINFO *PCVMXPAGEALLOCINFO;
302	AssertCompileSizeAlignment(VMXPAGEALLOCINFO, 8);
303
304	/**
305	* Memory operand read or write access.
306	*/
307	typedef enum VMXMEMACCESS
308	{
309	VMXMEMACCESS_READ = 0,
310	VMXMEMACCESS_WRITE = 1
311	} VMXMEMACCESS;
312
313	/**
314	* VMX VM-exit handler.
315	*
316	* @returns Strict VBox status code (i.e. informational status codes too).
317	* @param pVCpu The cross context virtual CPU structure.
318	* @param pVmxTransient The VMX-transient structure.
319	*/
320	#ifndef HMVMX_USE_FUNCTION_TABLE
321	typedef VBOXSTRICTRC FNVMXEXITHANDLER(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
322	#else
323	typedef DECLCALLBACKTYPE(VBOXSTRICTRC, FNVMXEXITHANDLER,(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient));
324	/** Pointer to VM-exit handler. */
325	typedef FNVMXEXITHANDLER *PFNVMXEXITHANDLER;
326	#endif
327
328	/**
329	* VMX VM-exit handler, non-strict status code.
330	*
331	* This is generally the same as FNVMXEXITHANDLER, the NSRC bit is just FYI.
332	*
333	* @returns VBox status code, no informational status code returned.
334	* @param pVCpu The cross context virtual CPU structure.
335	* @param pVmxTransient The VMX-transient structure.
336	*
337	* @remarks This is not used on anything returning VERR_EM_INTERPRETER as the
338	* use of that status code will be replaced with VINF_EM_SOMETHING
339	* later when switching over to IEM.
340	*/
341	#ifndef HMVMX_USE_FUNCTION_TABLE
342	typedef int FNVMXEXITHANDLERNSRC(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
343	#else
344	typedef FNVMXEXITHANDLER FNVMXEXITHANDLERNSRC;
345	#endif
346
347
348	/*********************************************************************************************************************************
349	* Internal Functions *
350	*********************************************************************************************************************************/
351	#ifndef HMVMX_USE_FUNCTION_TABLE
352	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
353	# define HMVMX_EXIT_DECL DECLINLINE(VBOXSTRICTRC)
354	# define HMVMX_EXIT_NSRC_DECL DECLINLINE(int)
355	#else
356	# define HMVMX_EXIT_DECL static DECLCALLBACK(VBOXSTRICTRC)
357	# define HMVMX_EXIT_NSRC_DECL HMVMX_EXIT_DECL
358	#endif
359	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
360	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
361	#endif
362
363	static int hmR0VmxImportGuestState(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint64_t fWhat);
364
365	/** @name VM-exit handler prototypes.
366	* @{
367	*/
368	static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmi;
369	static FNVMXEXITHANDLER hmR0VmxExitExtInt;
370	static FNVMXEXITHANDLER hmR0VmxExitTripleFault;
371	static FNVMXEXITHANDLERNSRC hmR0VmxExitIntWindow;
372	static FNVMXEXITHANDLERNSRC hmR0VmxExitNmiWindow;
373	static FNVMXEXITHANDLER hmR0VmxExitTaskSwitch;
374	static FNVMXEXITHANDLER hmR0VmxExitCpuid;
375	static FNVMXEXITHANDLER hmR0VmxExitGetsec;
376	static FNVMXEXITHANDLER hmR0VmxExitHlt;
377	static FNVMXEXITHANDLERNSRC hmR0VmxExitInvd;
378	static FNVMXEXITHANDLER hmR0VmxExitInvlpg;
379	static FNVMXEXITHANDLER hmR0VmxExitRdpmc;
380	static FNVMXEXITHANDLER hmR0VmxExitVmcall;
381	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
382	static FNVMXEXITHANDLER hmR0VmxExitVmclear;
383	static FNVMXEXITHANDLER hmR0VmxExitVmlaunch;
384	static FNVMXEXITHANDLER hmR0VmxExitVmptrld;
385	static FNVMXEXITHANDLER hmR0VmxExitVmptrst;
386	static FNVMXEXITHANDLER hmR0VmxExitVmread;
387	static FNVMXEXITHANDLER hmR0VmxExitVmresume;
388	static FNVMXEXITHANDLER hmR0VmxExitVmwrite;
389	static FNVMXEXITHANDLER hmR0VmxExitVmxoff;
390	static FNVMXEXITHANDLER hmR0VmxExitVmxon;
391	static FNVMXEXITHANDLER hmR0VmxExitInvvpid;
392	#endif
393	static FNVMXEXITHANDLER hmR0VmxExitRdtsc;
394	static FNVMXEXITHANDLER hmR0VmxExitMovCRx;
395	static FNVMXEXITHANDLER hmR0VmxExitMovDRx;
396	static FNVMXEXITHANDLER hmR0VmxExitIoInstr;
397	static FNVMXEXITHANDLER hmR0VmxExitRdmsr;
398	static FNVMXEXITHANDLER hmR0VmxExitWrmsr;
399	static FNVMXEXITHANDLER hmR0VmxExitMwait;
400	static FNVMXEXITHANDLER hmR0VmxExitMtf;
401	static FNVMXEXITHANDLER hmR0VmxExitMonitor;
402	static FNVMXEXITHANDLER hmR0VmxExitPause;
403	static FNVMXEXITHANDLERNSRC hmR0VmxExitTprBelowThreshold;
404	static FNVMXEXITHANDLER hmR0VmxExitApicAccess;
405	static FNVMXEXITHANDLER hmR0VmxExitEptViolation;
406	static FNVMXEXITHANDLER hmR0VmxExitEptMisconfig;
407	static FNVMXEXITHANDLER hmR0VmxExitRdtscp;
408	static FNVMXEXITHANDLER hmR0VmxExitPreemptTimer;
409	static FNVMXEXITHANDLERNSRC hmR0VmxExitWbinvd;
410	static FNVMXEXITHANDLER hmR0VmxExitXsetbv;
411	static FNVMXEXITHANDLER hmR0VmxExitInvpcid;
412	static FNVMXEXITHANDLERNSRC hmR0VmxExitSetPendingXcptUD;
413	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrInvalidGuestState;
414	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrUnexpected;
415	/** @} */
416
417	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
418	/** @name Nested-guest VM-exit handler prototypes.
419	* @{
420	*/
421	static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmiNested;
422	static FNVMXEXITHANDLER hmR0VmxExitTripleFaultNested;
423	static FNVMXEXITHANDLERNSRC hmR0VmxExitIntWindowNested;
424	static FNVMXEXITHANDLERNSRC hmR0VmxExitNmiWindowNested;
425	static FNVMXEXITHANDLER hmR0VmxExitTaskSwitchNested;
426	static FNVMXEXITHANDLER hmR0VmxExitHltNested;
427	static FNVMXEXITHANDLER hmR0VmxExitInvlpgNested;
428	static FNVMXEXITHANDLER hmR0VmxExitRdpmcNested;
429	static FNVMXEXITHANDLER hmR0VmxExitVmreadVmwriteNested;
430	static FNVMXEXITHANDLER hmR0VmxExitRdtscNested;
431	static FNVMXEXITHANDLER hmR0VmxExitMovCRxNested;
432	static FNVMXEXITHANDLER hmR0VmxExitMovDRxNested;
433	static FNVMXEXITHANDLER hmR0VmxExitIoInstrNested;
434	static FNVMXEXITHANDLER hmR0VmxExitRdmsrNested;
435	static FNVMXEXITHANDLER hmR0VmxExitWrmsrNested;
436	static FNVMXEXITHANDLER hmR0VmxExitMwaitNested;
437	static FNVMXEXITHANDLER hmR0VmxExitMtfNested;
438	static FNVMXEXITHANDLER hmR0VmxExitMonitorNested;
439	static FNVMXEXITHANDLER hmR0VmxExitPauseNested;
440	static FNVMXEXITHANDLERNSRC hmR0VmxExitTprBelowThresholdNested;
441	static FNVMXEXITHANDLER hmR0VmxExitApicAccessNested;
442	static FNVMXEXITHANDLER hmR0VmxExitApicWriteNested;
443	static FNVMXEXITHANDLER hmR0VmxExitVirtEoiNested;
444	static FNVMXEXITHANDLER hmR0VmxExitRdtscpNested;
445	static FNVMXEXITHANDLERNSRC hmR0VmxExitWbinvdNested;
446	static FNVMXEXITHANDLER hmR0VmxExitInvpcidNested;
447	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrInvalidGuestStateNested;
448	static FNVMXEXITHANDLER hmR0VmxExitInstrNested;
449	static FNVMXEXITHANDLER hmR0VmxExitInstrWithInfoNested;
450	/** @} */
451	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
452
453
454	/*********************************************************************************************************************************
455	* Global Variables *
456	*********************************************************************************************************************************/
457	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
458	/**
459	* Array of all VMCS fields.
460	* Any fields added to the VT-x spec. should be added here.
461	*
462	* Currently only used to derive shadow VMCS fields for hardware-assisted execution
463	* of nested-guests.
464	*/
465	static const uint32_t g_aVmcsFields[] =
466	{
467	/* 16-bit control fields. */
468	VMX_VMCS16_VPID,
469	VMX_VMCS16_POSTED_INT_NOTIFY_VECTOR,
470	VMX_VMCS16_EPTP_INDEX,
471
472	/* 16-bit guest-state fields. */
473	VMX_VMCS16_GUEST_ES_SEL,
474	VMX_VMCS16_GUEST_CS_SEL,
475	VMX_VMCS16_GUEST_SS_SEL,
476	VMX_VMCS16_GUEST_DS_SEL,
477	VMX_VMCS16_GUEST_FS_SEL,
478	VMX_VMCS16_GUEST_GS_SEL,
479	VMX_VMCS16_GUEST_LDTR_SEL,
480	VMX_VMCS16_GUEST_TR_SEL,
481	VMX_VMCS16_GUEST_INTR_STATUS,
482	VMX_VMCS16_GUEST_PML_INDEX,
483
484	/* 16-bits host-state fields. */
485	VMX_VMCS16_HOST_ES_SEL,
486	VMX_VMCS16_HOST_CS_SEL,
487	VMX_VMCS16_HOST_SS_SEL,
488	VMX_VMCS16_HOST_DS_SEL,
489	VMX_VMCS16_HOST_FS_SEL,
490	VMX_VMCS16_HOST_GS_SEL,
491	VMX_VMCS16_HOST_TR_SEL,
492
493	/* 64-bit control fields. */
494	VMX_VMCS64_CTRL_IO_BITMAP_A_FULL,
495	VMX_VMCS64_CTRL_IO_BITMAP_A_HIGH,
496	VMX_VMCS64_CTRL_IO_BITMAP_B_FULL,
497	VMX_VMCS64_CTRL_IO_BITMAP_B_HIGH,
498	VMX_VMCS64_CTRL_MSR_BITMAP_FULL,
499	VMX_VMCS64_CTRL_MSR_BITMAP_HIGH,
500	VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL,
501	VMX_VMCS64_CTRL_EXIT_MSR_STORE_HIGH,
502	VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL,
503	VMX_VMCS64_CTRL_EXIT_MSR_LOAD_HIGH,
504	VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL,
505	VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_HIGH,
506	VMX_VMCS64_CTRL_EXEC_VMCS_PTR_FULL,
507	VMX_VMCS64_CTRL_EXEC_VMCS_PTR_HIGH,
508	VMX_VMCS64_CTRL_EXEC_PML_ADDR_FULL,
509	VMX_VMCS64_CTRL_EXEC_PML_ADDR_HIGH,
510	VMX_VMCS64_CTRL_TSC_OFFSET_FULL,
511	VMX_VMCS64_CTRL_TSC_OFFSET_HIGH,
512	VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL,
513	VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_HIGH,
514	VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL,
515	VMX_VMCS64_CTRL_APIC_ACCESSADDR_HIGH,
516	VMX_VMCS64_CTRL_POSTED_INTR_DESC_FULL,
517	VMX_VMCS64_CTRL_POSTED_INTR_DESC_HIGH,
518	VMX_VMCS64_CTRL_VMFUNC_CTRLS_FULL,
519	VMX_VMCS64_CTRL_VMFUNC_CTRLS_HIGH,
520	VMX_VMCS64_CTRL_EPTP_FULL,
521	VMX_VMCS64_CTRL_EPTP_HIGH,
522	VMX_VMCS64_CTRL_EOI_BITMAP_0_FULL,
523	VMX_VMCS64_CTRL_EOI_BITMAP_0_HIGH,
524	VMX_VMCS64_CTRL_EOI_BITMAP_1_FULL,
525	VMX_VMCS64_CTRL_EOI_BITMAP_1_HIGH,
526	VMX_VMCS64_CTRL_EOI_BITMAP_2_FULL,
527	VMX_VMCS64_CTRL_EOI_BITMAP_2_HIGH,
528	VMX_VMCS64_CTRL_EOI_BITMAP_3_FULL,
529	VMX_VMCS64_CTRL_EOI_BITMAP_3_HIGH,
530	VMX_VMCS64_CTRL_EPTP_LIST_FULL,
531	VMX_VMCS64_CTRL_EPTP_LIST_HIGH,
532	VMX_VMCS64_CTRL_VMREAD_BITMAP_FULL,
533	VMX_VMCS64_CTRL_VMREAD_BITMAP_HIGH,
534	VMX_VMCS64_CTRL_VMWRITE_BITMAP_FULL,
535	VMX_VMCS64_CTRL_VMWRITE_BITMAP_HIGH,
536	VMX_VMCS64_CTRL_VIRTXCPT_INFO_ADDR_FULL,
537	VMX_VMCS64_CTRL_VIRTXCPT_INFO_ADDR_HIGH,
538	VMX_VMCS64_CTRL_XSS_EXITING_BITMAP_FULL,
539	VMX_VMCS64_CTRL_XSS_EXITING_BITMAP_HIGH,
540	VMX_VMCS64_CTRL_ENCLS_EXITING_BITMAP_FULL,
541	VMX_VMCS64_CTRL_ENCLS_EXITING_BITMAP_HIGH,
542	VMX_VMCS64_CTRL_TSC_MULTIPLIER_FULL,
543	VMX_VMCS64_CTRL_TSC_MULTIPLIER_HIGH,
544
545	/* 64-bit read-only data fields. */
546	VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL,
547	VMX_VMCS64_RO_GUEST_PHYS_ADDR_HIGH,
548
549	/* 64-bit guest-state fields. */
550	VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL,
551	VMX_VMCS64_GUEST_VMCS_LINK_PTR_HIGH,
552	VMX_VMCS64_GUEST_DEBUGCTL_FULL,
553	VMX_VMCS64_GUEST_DEBUGCTL_HIGH,
554	VMX_VMCS64_GUEST_PAT_FULL,
555	VMX_VMCS64_GUEST_PAT_HIGH,
556	VMX_VMCS64_GUEST_EFER_FULL,
557	VMX_VMCS64_GUEST_EFER_HIGH,
558	VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL,
559	VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_HIGH,
560	VMX_VMCS64_GUEST_PDPTE0_FULL,
561	VMX_VMCS64_GUEST_PDPTE0_HIGH,
562	VMX_VMCS64_GUEST_PDPTE1_FULL,
563	VMX_VMCS64_GUEST_PDPTE1_HIGH,
564	VMX_VMCS64_GUEST_PDPTE2_FULL,
565	VMX_VMCS64_GUEST_PDPTE2_HIGH,
566	VMX_VMCS64_GUEST_PDPTE3_FULL,
567	VMX_VMCS64_GUEST_PDPTE3_HIGH,
568	VMX_VMCS64_GUEST_BNDCFGS_FULL,
569	VMX_VMCS64_GUEST_BNDCFGS_HIGH,
570
571	/* 64-bit host-state fields. */
572	VMX_VMCS64_HOST_PAT_FULL,
573	VMX_VMCS64_HOST_PAT_HIGH,
574	VMX_VMCS64_HOST_EFER_FULL,
575	VMX_VMCS64_HOST_EFER_HIGH,
576	VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_FULL,
577	VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_HIGH,
578
579	/* 32-bit control fields. */
580	VMX_VMCS32_CTRL_PIN_EXEC,
581	VMX_VMCS32_CTRL_PROC_EXEC,
582	VMX_VMCS32_CTRL_EXCEPTION_BITMAP,
583	VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK,
584	VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH,
585	VMX_VMCS32_CTRL_CR3_TARGET_COUNT,
586	VMX_VMCS32_CTRL_EXIT,
587	VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT,
588	VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT,
589	VMX_VMCS32_CTRL_ENTRY,
590	VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT,
591	VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO,
592	VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE,
593	VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH,
594	VMX_VMCS32_CTRL_TPR_THRESHOLD,
595	VMX_VMCS32_CTRL_PROC_EXEC2,
596	VMX_VMCS32_CTRL_PLE_GAP,
597	VMX_VMCS32_CTRL_PLE_WINDOW,
598
599	/* 32-bits read-only fields. */
600	VMX_VMCS32_RO_VM_INSTR_ERROR,
601	VMX_VMCS32_RO_EXIT_REASON,
602	VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO,
603	VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE,
604	VMX_VMCS32_RO_IDT_VECTORING_INFO,
605	VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE,
606	VMX_VMCS32_RO_EXIT_INSTR_LENGTH,
607	VMX_VMCS32_RO_EXIT_INSTR_INFO,
608
609	/* 32-bit guest-state fields. */
610	VMX_VMCS32_GUEST_ES_LIMIT,
611	VMX_VMCS32_GUEST_CS_LIMIT,
612	VMX_VMCS32_GUEST_SS_LIMIT,
613	VMX_VMCS32_GUEST_DS_LIMIT,
614	VMX_VMCS32_GUEST_FS_LIMIT,
615	VMX_VMCS32_GUEST_GS_LIMIT,
616	VMX_VMCS32_GUEST_LDTR_LIMIT,
617	VMX_VMCS32_GUEST_TR_LIMIT,
618	VMX_VMCS32_GUEST_GDTR_LIMIT,
619	VMX_VMCS32_GUEST_IDTR_LIMIT,
620	VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS,
621	VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS,
622	VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS,
623	VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS,
624	VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS,
625	VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS,
626	VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS,
627	VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS,
628	VMX_VMCS32_GUEST_INT_STATE,
629	VMX_VMCS32_GUEST_ACTIVITY_STATE,
630	VMX_VMCS32_GUEST_SMBASE,
631	VMX_VMCS32_GUEST_SYSENTER_CS,
632	VMX_VMCS32_PREEMPT_TIMER_VALUE,
633
634	/* 32-bit host-state fields. */
635	VMX_VMCS32_HOST_SYSENTER_CS,
636
637	/* Natural-width control fields. */
638	VMX_VMCS_CTRL_CR0_MASK,
639	VMX_VMCS_CTRL_CR4_MASK,
640	VMX_VMCS_CTRL_CR0_READ_SHADOW,
641	VMX_VMCS_CTRL_CR4_READ_SHADOW,
642	VMX_VMCS_CTRL_CR3_TARGET_VAL0,
643	VMX_VMCS_CTRL_CR3_TARGET_VAL1,
644	VMX_VMCS_CTRL_CR3_TARGET_VAL2,
645	VMX_VMCS_CTRL_CR3_TARGET_VAL3,
646
647	/* Natural-width read-only data fields. */
648	VMX_VMCS_RO_EXIT_QUALIFICATION,
649	VMX_VMCS_RO_IO_RCX,
650	VMX_VMCS_RO_IO_RSI,
651	VMX_VMCS_RO_IO_RDI,
652	VMX_VMCS_RO_IO_RIP,
653	VMX_VMCS_RO_GUEST_LINEAR_ADDR,
654
655	/* Natural-width guest-state field */
656	VMX_VMCS_GUEST_CR0,
657	VMX_VMCS_GUEST_CR3,
658	VMX_VMCS_GUEST_CR4,
659	VMX_VMCS_GUEST_ES_BASE,
660	VMX_VMCS_GUEST_CS_BASE,
661	VMX_VMCS_GUEST_SS_BASE,
662	VMX_VMCS_GUEST_DS_BASE,
663	VMX_VMCS_GUEST_FS_BASE,
664	VMX_VMCS_GUEST_GS_BASE,
665	VMX_VMCS_GUEST_LDTR_BASE,
666	VMX_VMCS_GUEST_TR_BASE,
667	VMX_VMCS_GUEST_GDTR_BASE,
668	VMX_VMCS_GUEST_IDTR_BASE,
669	VMX_VMCS_GUEST_DR7,
670	VMX_VMCS_GUEST_RSP,
671	VMX_VMCS_GUEST_RIP,
672	VMX_VMCS_GUEST_RFLAGS,
673	VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS,
674	VMX_VMCS_GUEST_SYSENTER_ESP,
675	VMX_VMCS_GUEST_SYSENTER_EIP,
676
677	/* Natural-width host-state fields */
678	VMX_VMCS_HOST_CR0,
679	VMX_VMCS_HOST_CR3,
680	VMX_VMCS_HOST_CR4,
681	VMX_VMCS_HOST_FS_BASE,
682	VMX_VMCS_HOST_GS_BASE,
683	VMX_VMCS_HOST_TR_BASE,
684	VMX_VMCS_HOST_GDTR_BASE,
685	VMX_VMCS_HOST_IDTR_BASE,
686	VMX_VMCS_HOST_SYSENTER_ESP,
687	VMX_VMCS_HOST_SYSENTER_EIP,
688	VMX_VMCS_HOST_RSP,
689	VMX_VMCS_HOST_RIP
690	};
691	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
692
693	#ifdef VBOX_STRICT
694	static const uint32_t g_aVmcsSegBase[] =
695	{
696	VMX_VMCS_GUEST_ES_BASE,
697	VMX_VMCS_GUEST_CS_BASE,
698	VMX_VMCS_GUEST_SS_BASE,
699	VMX_VMCS_GUEST_DS_BASE,
700	VMX_VMCS_GUEST_FS_BASE,
701	VMX_VMCS_GUEST_GS_BASE
702	};
703	static const uint32_t g_aVmcsSegSel[] =
704	{
705	VMX_VMCS16_GUEST_ES_SEL,
706	VMX_VMCS16_GUEST_CS_SEL,
707	VMX_VMCS16_GUEST_SS_SEL,
708	VMX_VMCS16_GUEST_DS_SEL,
709	VMX_VMCS16_GUEST_FS_SEL,
710	VMX_VMCS16_GUEST_GS_SEL
711	};
712	static const uint32_t g_aVmcsSegLimit[] =
713	{
714	VMX_VMCS32_GUEST_ES_LIMIT,
715	VMX_VMCS32_GUEST_CS_LIMIT,
716	VMX_VMCS32_GUEST_SS_LIMIT,
717	VMX_VMCS32_GUEST_DS_LIMIT,
718	VMX_VMCS32_GUEST_FS_LIMIT,
719	VMX_VMCS32_GUEST_GS_LIMIT
720	};
721	static const uint32_t g_aVmcsSegAttr[] =
722	{
723	VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS,
724	VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS,
725	VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS,
726	VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS,
727	VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS,
728	VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS
729	};
730	AssertCompile(RT_ELEMENTS(g_aVmcsSegSel) == X86_SREG_COUNT);
731	AssertCompile(RT_ELEMENTS(g_aVmcsSegLimit) == X86_SREG_COUNT);
732	AssertCompile(RT_ELEMENTS(g_aVmcsSegBase) == X86_SREG_COUNT);
733	AssertCompile(RT_ELEMENTS(g_aVmcsSegAttr) == X86_SREG_COUNT);
734	#endif /* VBOX_STRICT */
735
736	#ifdef HMVMX_USE_FUNCTION_TABLE
737	/**
738	* VMX_EXIT dispatch table.
739	*/
740	static const struct CLANG11NOTHROWWEIRDNESS { PFNVMXEXITHANDLER pfn; } g_aVMExitHandlers[VMX_EXIT_MAX + 1] =
741	{
742	/* 0 VMX_EXIT_XCPT_OR_NMI */ { hmR0VmxExitXcptOrNmi },
743	/* 1 VMX_EXIT_EXT_INT */ { hmR0VmxExitExtInt },
744	/* 2 VMX_EXIT_TRIPLE_FAULT */ { hmR0VmxExitTripleFault },
745	/* 3 VMX_EXIT_INIT_SIGNAL */ { hmR0VmxExitErrUnexpected },
746	/* 4 VMX_EXIT_SIPI */ { hmR0VmxExitErrUnexpected },
747	/* 5 VMX_EXIT_IO_SMI */ { hmR0VmxExitErrUnexpected },
748	/* 6 VMX_EXIT_SMI */ { hmR0VmxExitErrUnexpected },
749	/* 7 VMX_EXIT_INT_WINDOW */ { hmR0VmxExitIntWindow },
750	/* 8 VMX_EXIT_NMI_WINDOW */ { hmR0VmxExitNmiWindow },
751	/* 9 VMX_EXIT_TASK_SWITCH */ { hmR0VmxExitTaskSwitch },
752	/* 10 VMX_EXIT_CPUID */ { hmR0VmxExitCpuid },
753	/* 11 VMX_EXIT_GETSEC */ { hmR0VmxExitGetsec },
754	/* 12 VMX_EXIT_HLT */ { hmR0VmxExitHlt },
755	/* 13 VMX_EXIT_INVD */ { hmR0VmxExitInvd },
756	/* 14 VMX_EXIT_INVLPG */ { hmR0VmxExitInvlpg },
757	/* 15 VMX_EXIT_RDPMC */ { hmR0VmxExitRdpmc },
758	/* 16 VMX_EXIT_RDTSC */ { hmR0VmxExitRdtsc },
759	/* 17 VMX_EXIT_RSM */ { hmR0VmxExitErrUnexpected },
760	/* 18 VMX_EXIT_VMCALL */ { hmR0VmxExitVmcall },
761	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
762	/* 19 VMX_EXIT_VMCLEAR */ { hmR0VmxExitVmclear },
763	/* 20 VMX_EXIT_VMLAUNCH */ { hmR0VmxExitVmlaunch },
764	/* 21 VMX_EXIT_VMPTRLD */ { hmR0VmxExitVmptrld },
765	/* 22 VMX_EXIT_VMPTRST */ { hmR0VmxExitVmptrst },
766	/* 23 VMX_EXIT_VMREAD */ { hmR0VmxExitVmread },
767	/* 24 VMX_EXIT_VMRESUME */ { hmR0VmxExitVmresume },
768	/* 25 VMX_EXIT_VMWRITE */ { hmR0VmxExitVmwrite },
769	/* 26 VMX_EXIT_VMXOFF */ { hmR0VmxExitVmxoff },
770	/* 27 VMX_EXIT_VMXON */ { hmR0VmxExitVmxon },
771	#else
772	/* 19 VMX_EXIT_VMCLEAR */ { hmR0VmxExitSetPendingXcptUD },
773	/* 20 VMX_EXIT_VMLAUNCH */ { hmR0VmxExitSetPendingXcptUD },
774	/* 21 VMX_EXIT_VMPTRLD */ { hmR0VmxExitSetPendingXcptUD },
775	/* 22 VMX_EXIT_VMPTRST */ { hmR0VmxExitSetPendingXcptUD },
776	/* 23 VMX_EXIT_VMREAD */ { hmR0VmxExitSetPendingXcptUD },
777	/* 24 VMX_EXIT_VMRESUME */ { hmR0VmxExitSetPendingXcptUD },
778	/* 25 VMX_EXIT_VMWRITE */ { hmR0VmxExitSetPendingXcptUD },
779	/* 26 VMX_EXIT_VMXOFF */ { hmR0VmxExitSetPendingXcptUD },
780	/* 27 VMX_EXIT_VMXON */ { hmR0VmxExitSetPendingXcptUD },
781	#endif
782	/* 28 VMX_EXIT_MOV_CRX */ { hmR0VmxExitMovCRx },
783	/* 29 VMX_EXIT_MOV_DRX */ { hmR0VmxExitMovDRx },
784	/* 30 VMX_EXIT_IO_INSTR */ { hmR0VmxExitIoInstr },
785	/* 31 VMX_EXIT_RDMSR */ { hmR0VmxExitRdmsr },
786	/* 32 VMX_EXIT_WRMSR */ { hmR0VmxExitWrmsr },
787	/* 33 VMX_EXIT_ERR_INVALID_GUEST_STATE */ { hmR0VmxExitErrInvalidGuestState },
788	/* 34 VMX_EXIT_ERR_MSR_LOAD */ { hmR0VmxExitErrUnexpected },
789	/* 35 UNDEFINED */ { hmR0VmxExitErrUnexpected },
790	/* 36 VMX_EXIT_MWAIT */ { hmR0VmxExitMwait },
791	/* 37 VMX_EXIT_MTF */ { hmR0VmxExitMtf },
792	/* 38 UNDEFINED */ { hmR0VmxExitErrUnexpected },
793	/* 39 VMX_EXIT_MONITOR */ { hmR0VmxExitMonitor },
794	/* 40 VMX_EXIT_PAUSE */ { hmR0VmxExitPause },
795	/* 41 VMX_EXIT_ERR_MACHINE_CHECK */ { hmR0VmxExitErrUnexpected },
796	/* 42 UNDEFINED */ { hmR0VmxExitErrUnexpected },
797	/* 43 VMX_EXIT_TPR_BELOW_THRESHOLD */ { hmR0VmxExitTprBelowThreshold },
798	/* 44 VMX_EXIT_APIC_ACCESS */ { hmR0VmxExitApicAccess },
799	/* 45 VMX_EXIT_VIRTUALIZED_EOI */ { hmR0VmxExitErrUnexpected },
800	/* 46 VMX_EXIT_GDTR_IDTR_ACCESS */ { hmR0VmxExitErrUnexpected },
801	/* 47 VMX_EXIT_LDTR_TR_ACCESS */ { hmR0VmxExitErrUnexpected },
802	/* 48 VMX_EXIT_EPT_VIOLATION */ { hmR0VmxExitEptViolation },
803	/* 49 VMX_EXIT_EPT_MISCONFIG */ { hmR0VmxExitEptMisconfig },
804	/* 50 VMX_EXIT_INVEPT */ { hmR0VmxExitSetPendingXcptUD },
805	/* 51 VMX_EXIT_RDTSCP */ { hmR0VmxExitRdtscp },
806	/* 52 VMX_EXIT_PREEMPT_TIMER */ { hmR0VmxExitPreemptTimer },
807	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
808	/* 53 VMX_EXIT_INVVPID */ { hmR0VmxExitInvvpid },
809	#else
810	/* 53 VMX_EXIT_INVVPID */ { hmR0VmxExitSetPendingXcptUD },
811	#endif
812	/* 54 VMX_EXIT_WBINVD */ { hmR0VmxExitWbinvd },
813	/* 55 VMX_EXIT_XSETBV */ { hmR0VmxExitXsetbv },
814	/* 56 VMX_EXIT_APIC_WRITE */ { hmR0VmxExitErrUnexpected },
815	/* 57 VMX_EXIT_RDRAND */ { hmR0VmxExitErrUnexpected },
816	/* 58 VMX_EXIT_INVPCID */ { hmR0VmxExitInvpcid },
817	/* 59 VMX_EXIT_VMFUNC */ { hmR0VmxExitErrUnexpected },
818	/* 60 VMX_EXIT_ENCLS */ { hmR0VmxExitErrUnexpected },
819	/* 61 VMX_EXIT_RDSEED */ { hmR0VmxExitErrUnexpected },
820	/* 62 VMX_EXIT_PML_FULL */ { hmR0VmxExitErrUnexpected },
821	/* 63 VMX_EXIT_XSAVES */ { hmR0VmxExitErrUnexpected },
822	/* 64 VMX_EXIT_XRSTORS */ { hmR0VmxExitErrUnexpected },
823	/* 65 UNDEFINED */ { hmR0VmxExitErrUnexpected },
824	/* 66 VMX_EXIT_SPP_EVENT */ { hmR0VmxExitErrUnexpected },
825	/* 67 VMX_EXIT_UMWAIT */ { hmR0VmxExitErrUnexpected },
826	/* 68 VMX_EXIT_TPAUSE */ { hmR0VmxExitErrUnexpected },
827	};
828	#endif /* HMVMX_USE_FUNCTION_TABLE */
829
830	#if defined(VBOX_STRICT) && defined(LOG_ENABLED)
831	static const char * const g_apszVmxInstrErrors[HMVMX_INSTR_ERROR_MAX + 1] =
832	{
833	/* 0 */ "(Not Used)",
834	/* 1 */ "VMCALL executed in VMX root operation.",
835	/* 2 */ "VMCLEAR with invalid physical address.",
836	/* 3 */ "VMCLEAR with VMXON pointer.",
837	/* 4 */ "VMLAUNCH with non-clear VMCS.",
838	/* 5 */ "VMRESUME with non-launched VMCS.",
839	/* 6 */ "VMRESUME after VMXOFF",
840	/* 7 */ "VM-entry with invalid control fields.",
841	/* 8 */ "VM-entry with invalid host state fields.",
842	/* 9 */ "VMPTRLD with invalid physical address.",
843	/* 10 */ "VMPTRLD with VMXON pointer.",
844	/* 11 */ "VMPTRLD with incorrect revision identifier.",
845	/* 12 */ "VMREAD/VMWRITE from/to unsupported VMCS component.",
846	/* 13 */ "VMWRITE to read-only VMCS component.",
847	/* 14 */ "(Not Used)",
848	/* 15 */ "VMXON executed in VMX root operation.",
849	/* 16 */ "VM-entry with invalid executive-VMCS pointer.",
850	/* 17 */ "VM-entry with non-launched executing VMCS.",
851	/* 18 */ "VM-entry with executive-VMCS pointer not VMXON pointer.",
852	/* 19 */ "VMCALL with non-clear VMCS.",
853	/* 20 */ "VMCALL with invalid VM-exit control fields.",
854	/* 21 */ "(Not Used)",
855	/* 22 */ "VMCALL with incorrect MSEG revision identifier.",
856	/* 23 */ "VMXOFF under dual monitor treatment of SMIs and SMM.",
857	/* 24 */ "VMCALL with invalid SMM-monitor features.",
858	/* 25 */ "VM-entry with invalid VM-execution control fields in executive VMCS.",
859	/* 26 */ "VM-entry with events blocked by MOV SS.",
860	/* 27 */ "(Not Used)",
861	/* 28 */ "Invalid operand to INVEPT/INVVPID."
862	};
863	#endif /* VBOX_STRICT && LOG_ENABLED */
864
865
866	/**
867	* Checks if the given MSR is part of the lastbranch-from-IP MSR stack.
868	* @returns @c true if it's part of LBR stack, @c false otherwise.
869	*
870	* @param pVM The cross context VM structure.
871	* @param idMsr The MSR.
872	* @param pidxMsr Where to store the index of the MSR in the LBR MSR array.
873	* Optional, can be NULL.
874	*
875	* @remarks Must only be called when LBR is enabled.
876	*/
877	DECL_FORCE_INLINE(bool) hmR0VmxIsLbrBranchFromMsr(PCVMCC pVM, uint32_t idMsr, uint32_t *pidxMsr)
878	{
879	Assert(pVM->hmr0.s.vmx.fLbr);
880	Assert(pVM->hmr0.s.vmx.idLbrFromIpMsrFirst);
881	uint32_t const cLbrStack = pVM->hmr0.s.vmx.idLbrFromIpMsrLast - pVM->hmr0.s.vmx.idLbrFromIpMsrFirst + 1;
882	uint32_t const idxMsr = idMsr - pVM->hmr0.s.vmx.idLbrFromIpMsrFirst;
883	if (idxMsr < cLbrStack)
884	{
885	if (pidxMsr)
886	*pidxMsr = idxMsr;
887	return true;
888	}
889	return false;
890	}
891
892
893	/**
894	* Checks if the given MSR is part of the lastbranch-to-IP MSR stack.
895	* @returns @c true if it's part of LBR stack, @c false otherwise.
896	*
897	* @param pVM The cross context VM structure.
898	* @param idMsr The MSR.
899	* @param pidxMsr Where to store the index of the MSR in the LBR MSR array.
900	* Optional, can be NULL.
901	*
902	* @remarks Must only be called when LBR is enabled and when lastbranch-to-IP MSRs
903	* are supported by the CPU (see hmR0VmxSetupLbrMsrRange).
904	*/
905	DECL_FORCE_INLINE(bool) hmR0VmxIsLbrBranchToMsr(PCVMCC pVM, uint32_t idMsr, uint32_t *pidxMsr)
906	{
907	Assert(pVM->hmr0.s.vmx.fLbr);
908	if (pVM->hmr0.s.vmx.idLbrToIpMsrFirst)
909	{
910	uint32_t const cLbrStack = pVM->hmr0.s.vmx.idLbrToIpMsrLast - pVM->hmr0.s.vmx.idLbrToIpMsrFirst + 1;
911	uint32_t const idxMsr = idMsr - pVM->hmr0.s.vmx.idLbrToIpMsrFirst;
912	if (idxMsr < cLbrStack)
913	{
914	if (pidxMsr)
915	*pidxMsr = idxMsr;
916	return true;
917	}
918	}
919	return false;
920	}
921
922
923	/**
924	* Gets the CR0 guest/host mask.
925	*
926	* These bits typically does not change through the lifetime of a VM. Any bit set in
927	* this mask is owned by the host/hypervisor and would cause a VM-exit when modified
928	* by the guest.
929	*
930	* @returns The CR0 guest/host mask.
931	* @param pVCpu The cross context virtual CPU structure.
932	*/
933	static uint64_t hmR0VmxGetFixedCr0Mask(PCVMCPUCC pVCpu)
934	{
935	/*
936	* Modifications to CR0 bits that VT-x ignores saving/restoring (CD, ET, NW) and
937	* to CR0 bits that we require for shadow paging (PG) by the guest must cause VM-exits.
938	*
939	* Furthermore, modifications to any bits that are reserved/unspecified currently
940	* by the Intel spec. must also cause a VM-exit. This prevents unpredictable behavior
941	* when future CPUs specify and use currently reserved/unspecified bits.
942	*/
943	/** @todo Avoid intercepting CR0.PE with unrestricted guest execution. Fix PGM
944	* enmGuestMode to be in-sync with the current mode. See @bugref{6398}
945	* and @bugref{6944}. */
946	PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
947	return ( X86_CR0_PE
948	\| X86_CR0_NE
949	\| (pVM->hmr0.s.fNestedPaging ? 0 : X86_CR0_WP)
950	\| X86_CR0_PG
951	\| VMX_EXIT_HOST_CR0_IGNORE_MASK);
952	}
953
954
955	/**
956	* Gets the CR4 guest/host mask.
957	*
958	* These bits typically does not change through the lifetime of a VM. Any bit set in
959	* this mask is owned by the host/hypervisor and would cause a VM-exit when modified
960	* by the guest.
961	*
962	* @returns The CR4 guest/host mask.
963	* @param pVCpu The cross context virtual CPU structure.
964	*/
965	static uint64_t hmR0VmxGetFixedCr4Mask(PCVMCPUCC pVCpu)
966	{
967	/*
968	* We construct a mask of all CR4 bits that the guest can modify without causing
969	* a VM-exit. Then invert this mask to obtain all CR4 bits that should cause
970	* a VM-exit when the guest attempts to modify them when executing using
971	* hardware-assisted VMX.
972	*
973	* When a feature is not exposed to the guest (and may be present on the host),
974	* we want to intercept guest modifications to the bit so we can emulate proper
975	* behavior (e.g., #GP).
976	*
977	* Furthermore, only modifications to those bits that don't require immediate
978	* emulation is allowed. For e.g., PCIDE is excluded because the behavior
979	* depends on CR3 which might not always be the guest value while executing
980	* using hardware-assisted VMX.
981	*/
982	PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
983	bool const fFsGsBase = pVM->cpum.ro.GuestFeatures.fFsGsBase;
984	bool const fXSaveRstor = pVM->cpum.ro.GuestFeatures.fXSaveRstor;
985	bool const fFxSaveRstor = pVM->cpum.ro.GuestFeatures.fFxSaveRstor;
986
987	/*
988	* Paranoia.
989	* Ensure features exposed to the guest are present on the host.
990	*/
991	Assert(!fFsGsBase \|\| pVM->cpum.ro.HostFeatures.fFsGsBase);
992	Assert(!fXSaveRstor \|\| pVM->cpum.ro.HostFeatures.fXSaveRstor);
993	Assert(!fFxSaveRstor \|\| pVM->cpum.ro.HostFeatures.fFxSaveRstor);
994
995	uint64_t const fGstMask = ( X86_CR4_PVI
996	\| X86_CR4_TSD
997	\| X86_CR4_DE
998	\| X86_CR4_MCE
999	\| X86_CR4_PCE
1000	\| X86_CR4_OSXMMEEXCPT
1001	\| (fFsGsBase ? X86_CR4_FSGSBASE : 0)
1002	\| (fXSaveRstor ? X86_CR4_OSXSAVE : 0)
1003	\| (fFxSaveRstor ? X86_CR4_OSFXSR : 0));
1004	return ~fGstMask;
1005	}
1006
1007
1008	/**
1009	* Gets the active (in use) VMCS info. object for the specified VCPU.
1010	*
1011	* This is either the guest or nested-guest VMCS info. and need not necessarily
1012	* pertain to the "current" VMCS (in the VMX definition of the term). For instance,
1013	* if the VM-entry failed due to an invalid-guest state, we may have "cleared" the
1014	* current VMCS while returning to ring-3. However, the VMCS info. object for that
1015	* VMCS would still be active and returned here so that we could dump the VMCS
1016	* fields to ring-3 for diagnostics. This function is thus only used to
1017	* distinguish between the nested-guest or guest VMCS.
1018	*
1019	* @returns The active VMCS information.
1020	* @param pVCpu The cross context virtual CPU structure.
1021	*
1022	* @thread EMT.
1023	* @remarks This function may be called with preemption or interrupts disabled!
1024	*/
1025	DECLINLINE(PVMXVMCSINFO) hmGetVmxActiveVmcsInfo(PVMCPUCC pVCpu)
1026	{
1027	if (!pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs)
1028	return &pVCpu->hmr0.s.vmx.VmcsInfo;
1029	return &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
1030	}
1031
1032
1033	/**
1034	* Returns whether the VM-exit MSR-store area differs from the VM-exit MSR-load
1035	* area.
1036	*
1037	* @returns @c true if it's different, @c false otherwise.
1038	* @param pVmcsInfo The VMCS info. object.
1039	*/
1040	DECL_FORCE_INLINE(bool) hmR0VmxIsSeparateExitMsrStoreAreaVmcs(PCVMXVMCSINFO pVmcsInfo)
1041	{
1042	return RT_BOOL( pVmcsInfo->pvGuestMsrStore != pVmcsInfo->pvGuestMsrLoad
1043	&& pVmcsInfo->pvGuestMsrStore);
1044	}
1045
1046
1047	/**
1048	* Sets the given Processor-based VM-execution controls.
1049	*
1050	* @param pVmxTransient The VMX-transient structure.
1051	* @param uProcCtls The Processor-based VM-execution controls to set.
1052	*/
1053	static void hmR0VmxSetProcCtlsVmcs(PVMXTRANSIENT pVmxTransient, uint32_t uProcCtls)
1054	{
1055	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1056	if ((pVmcsInfo->u32ProcCtls & uProcCtls) != uProcCtls)
1057	{
1058	pVmcsInfo->u32ProcCtls \|= uProcCtls;
1059	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
1060	AssertRC(rc);
1061	}
1062	}
1063
1064
1065	/**
1066	* Removes the given Processor-based VM-execution controls.
1067	*
1068	* @param pVCpu The cross context virtual CPU structure.
1069	* @param pVmxTransient The VMX-transient structure.
1070	* @param uProcCtls The Processor-based VM-execution controls to remove.
1071	*
1072	* @remarks When executing a nested-guest, this will not remove any of the specified
1073	* controls if the nested hypervisor has set any one of them.
1074	*/
1075	static void hmR0VmxRemoveProcCtlsVmcs(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uProcCtls)
1076	{
1077	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1078	if (pVmcsInfo->u32ProcCtls & uProcCtls)
1079	{
1080	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1081	if ( !pVmxTransient->fIsNestedGuest
1082	\|\| !CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, uProcCtls))
1083	#else
1084	NOREF(pVCpu);
1085	if (!pVmxTransient->fIsNestedGuest)
1086	#endif
1087	{
1088	pVmcsInfo->u32ProcCtls &= ~uProcCtls;
1089	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
1090	AssertRC(rc);
1091	}
1092	}
1093	}
1094
1095
1096	/**
1097	* Sets the TSC offset for the current VMCS.
1098	*
1099	* @param uTscOffset The TSC offset to set.
1100	* @param pVmcsInfo The VMCS info. object.
1101	*/
1102	static void hmR0VmxSetTscOffsetVmcs(PVMXVMCSINFO pVmcsInfo, uint64_t uTscOffset)
1103	{
1104	if (pVmcsInfo->u64TscOffset != uTscOffset)
1105	{
1106	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, uTscOffset);
1107	AssertRC(rc);
1108	pVmcsInfo->u64TscOffset = uTscOffset;
1109	}
1110	}
1111
1112
1113	/**
1114	* Adds one or more exceptions to the exception bitmap and commits it to the current
1115	* VMCS.
1116	*
1117	* @param pVmxTransient The VMX-transient structure.
1118	* @param uXcptMask The exception(s) to add.
1119	*/
1120	static void hmR0VmxAddXcptInterceptMask(PCVMXTRANSIENT pVmxTransient, uint32_t uXcptMask)
1121	{
1122	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1123	uint32_t uXcptBitmap = pVmcsInfo->u32XcptBitmap;
1124	if ((uXcptBitmap & uXcptMask) != uXcptMask)
1125	{
1126	uXcptBitmap \|= uXcptMask;
1127	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
1128	AssertRC(rc);
1129	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
1130	}
1131	}
1132
1133
1134	/**
1135	* Adds an exception to the exception bitmap and commits it to the current VMCS.
1136	*
1137	* @param pVmxTransient The VMX-transient structure.
1138	* @param uXcpt The exception to add.
1139	*/
1140	static void hmR0VmxAddXcptIntercept(PCVMXTRANSIENT pVmxTransient, uint8_t uXcpt)
1141	{
1142	Assert(uXcpt <= X86_XCPT_LAST);
1143	hmR0VmxAddXcptInterceptMask(pVmxTransient, RT_BIT_32(uXcpt));
1144	}
1145
1146
1147	/**
1148	* Remove one or more exceptions from the exception bitmap and commits it to the
1149	* current VMCS.
1150	*
1151	* This takes care of not removing the exception intercept if a nested-guest
1152	* requires the exception to be intercepted.
1153	*
1154	* @returns VBox status code.
1155	* @param pVCpu The cross context virtual CPU structure.
1156	* @param pVmxTransient The VMX-transient structure.
1157	* @param uXcptMask The exception(s) to remove.
1158	*/
1159	static int hmR0VmxRemoveXcptInterceptMask(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t uXcptMask)
1160	{
1161	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1162	uint32_t u32XcptBitmap = pVmcsInfo->u32XcptBitmap;
1163	if (u32XcptBitmap & uXcptMask)
1164	{
1165	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1166	if (!pVmxTransient->fIsNestedGuest)
1167	{ /* likely */ }
1168	else
1169	{
1170	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1171	uXcptMask &= ~pVmcsNstGst->u32XcptBitmap;
1172	}
1173	#endif
1174	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
1175	uXcptMask &= ~( RT_BIT(X86_XCPT_BP)
1176	\| RT_BIT(X86_XCPT_DE)
1177	\| RT_BIT(X86_XCPT_NM)
1178	\| RT_BIT(X86_XCPT_TS)
1179	\| RT_BIT(X86_XCPT_UD)
1180	\| RT_BIT(X86_XCPT_NP)
1181	\| RT_BIT(X86_XCPT_SS)
1182	\| RT_BIT(X86_XCPT_GP)
1183	\| RT_BIT(X86_XCPT_PF)
1184	\| RT_BIT(X86_XCPT_MF));
1185	#elif defined(HMVMX_ALWAYS_TRAP_PF)
1186	uXcptMask &= ~RT_BIT(X86_XCPT_PF);
1187	#endif
1188	if (uXcptMask)
1189	{
1190	/* Validate we are not removing any essential exception intercepts. */
1191	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging \|\| !(uXcptMask & RT_BIT(X86_XCPT_PF)));
1192	NOREF(pVCpu);
1193	Assert(!(uXcptMask & RT_BIT(X86_XCPT_DB)));
1194	Assert(!(uXcptMask & RT_BIT(X86_XCPT_AC)));
1195
1196	/* Remove it from the exception bitmap. */
1197	u32XcptBitmap &= ~uXcptMask;
1198
1199	/* Commit and update the cache if necessary. */
1200	if (pVmcsInfo->u32XcptBitmap != u32XcptBitmap)
1201	{
1202	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
1203	AssertRC(rc);
1204	pVmcsInfo->u32XcptBitmap = u32XcptBitmap;
1205	}
1206	}
1207	}
1208	return VINF_SUCCESS;
1209	}
1210
1211
1212	/**
1213	* Remove an exceptions from the exception bitmap and commits it to the current
1214	* VMCS.
1215	*
1216	* @returns VBox status code.
1217	* @param pVCpu The cross context virtual CPU structure.
1218	* @param pVmxTransient The VMX-transient structure.
1219	* @param uXcpt The exception to remove.
1220	*/
1221	static int hmR0VmxRemoveXcptIntercept(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint8_t uXcpt)
1222	{
1223	return hmR0VmxRemoveXcptInterceptMask(pVCpu, pVmxTransient, RT_BIT(uXcpt));
1224	}
1225
1226
1227	/**
1228	* Loads the VMCS specified by the VMCS info. object.
1229	*
1230	* @returns VBox status code.
1231	* @param pVmcsInfo The VMCS info. object.
1232	*
1233	* @remarks Can be called with interrupts disabled.
1234	*/
1235	static int hmR0VmxLoadVmcs(PVMXVMCSINFO pVmcsInfo)
1236	{
1237	Assert(pVmcsInfo->HCPhysVmcs != 0 && pVmcsInfo->HCPhysVmcs != NIL_RTHCPHYS);
1238	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1239
1240	int rc = VMXLoadVmcs(pVmcsInfo->HCPhysVmcs);
1241	if (RT_SUCCESS(rc))
1242	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_CURRENT;
1243	return rc;
1244	}
1245
1246
1247	/**
1248	* Clears the VMCS specified by the VMCS info. object.
1249	*
1250	* @returns VBox status code.
1251	* @param pVmcsInfo The VMCS info. object.
1252	*
1253	* @remarks Can be called with interrupts disabled.
1254	*/
1255	static int hmR0VmxClearVmcs(PVMXVMCSINFO pVmcsInfo)
1256	{
1257	Assert(pVmcsInfo->HCPhysVmcs != 0 && pVmcsInfo->HCPhysVmcs != NIL_RTHCPHYS);
1258	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1259
1260	int rc = VMXClearVmcs(pVmcsInfo->HCPhysVmcs);
1261	if (RT_SUCCESS(rc))
1262	pVmcsInfo->fVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
1263	return rc;
1264	}
1265
1266
1267	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1268	/**
1269	* Loads the shadow VMCS specified by the VMCS info. object.
1270	*
1271	* @returns VBox status code.
1272	* @param pVmcsInfo The VMCS info. object.
1273	*
1274	* @remarks Can be called with interrupts disabled.
1275	*/
1276	static int hmR0VmxLoadShadowVmcs(PVMXVMCSINFO pVmcsInfo)
1277	{
1278	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1279	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
1280
1281	int rc = VMXLoadVmcs(pVmcsInfo->HCPhysShadowVmcs);
1282	if (RT_SUCCESS(rc))
1283	pVmcsInfo->fShadowVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_CURRENT;
1284	return rc;
1285	}
1286
1287
1288	/**
1289	* Clears the shadow VMCS specified by the VMCS info. object.
1290	*
1291	* @returns VBox status code.
1292	* @param pVmcsInfo The VMCS info. object.
1293	*
1294	* @remarks Can be called with interrupts disabled.
1295	*/
1296	static int hmR0VmxClearShadowVmcs(PVMXVMCSINFO pVmcsInfo)
1297	{
1298	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1299	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
1300
1301	int rc = VMXClearVmcs(pVmcsInfo->HCPhysShadowVmcs);
1302	if (RT_SUCCESS(rc))
1303	pVmcsInfo->fShadowVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
1304	return rc;
1305	}
1306
1307
1308	/**
1309	* Switches from and to the specified VMCSes.
1310	*
1311	* @returns VBox status code.
1312	* @param pVmcsInfoFrom The VMCS info. object we are switching from.
1313	* @param pVmcsInfoTo The VMCS info. object we are switching to.
1314	*
1315	* @remarks Called with interrupts disabled.
1316	*/
1317	static int hmR0VmxSwitchVmcs(PVMXVMCSINFO pVmcsInfoFrom, PVMXVMCSINFO pVmcsInfoTo)
1318	{
1319	/*
1320	* Clear the VMCS we are switching out if it has not already been cleared.
1321	* This will sync any CPU internal data back to the VMCS.
1322	*/
1323	if (pVmcsInfoFrom->fVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
1324	{
1325	int rc = hmR0VmxClearVmcs(pVmcsInfoFrom);
1326	if (RT_SUCCESS(rc))
1327	{
1328	/*
1329	* The shadow VMCS, if any, would not be active at this point since we
1330	* would have cleared it while importing the virtual hardware-virtualization
1331	* state as part the VMLAUNCH/VMRESUME VM-exit. Hence, there's no need to
1332	* clear the shadow VMCS here, just assert for safety.
1333	*/
1334	Assert(!pVmcsInfoFrom->pvShadowVmcs \|\| pVmcsInfoFrom->fShadowVmcsState == VMX_V_VMCS_LAUNCH_STATE_CLEAR);
1335	}
1336	else
1337	return rc;
1338	}
1339
1340	/*
1341	* Clear the VMCS we are switching to if it has not already been cleared.
1342	* This will initialize the VMCS launch state to "clear" required for loading it.
1343	*
1344	* See Intel spec. 31.6 "Preparation And Launching A Virtual Machine".
1345	*/
1346	if (pVmcsInfoTo->fVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
1347	{
1348	int rc = hmR0VmxClearVmcs(pVmcsInfoTo);
1349	if (RT_SUCCESS(rc))
1350	{ /* likely */ }
1351	else
1352	return rc;
1353	}
1354
1355	/*
1356	* Finally, load the VMCS we are switching to.
1357	*/
1358	return hmR0VmxLoadVmcs(pVmcsInfoTo);
1359	}
1360
1361
1362	/**
1363	* Switches between the guest VMCS and the nested-guest VMCS as specified by the
1364	* caller.
1365	*
1366	* @returns VBox status code.
1367	* @param pVCpu The cross context virtual CPU structure.
1368	* @param fSwitchToNstGstVmcs Whether to switch to the nested-guest VMCS (pass
1369	* true) or guest VMCS (pass false).
1370	*/
1371	static int hmR0VmxSwitchToGstOrNstGstVmcs(PVMCPUCC pVCpu, bool fSwitchToNstGstVmcs)
1372	{
1373	/* Ensure we have synced everything from the guest-CPU context to the VMCS before switching. */
1374	HMVMX_CPUMCTX_ASSERT(pVCpu, HMVMX_CPUMCTX_EXTRN_ALL);
1375
1376	PVMXVMCSINFO pVmcsInfoFrom;
1377	PVMXVMCSINFO pVmcsInfoTo;
1378	if (fSwitchToNstGstVmcs)
1379	{
1380	pVmcsInfoFrom = &pVCpu->hmr0.s.vmx.VmcsInfo;
1381	pVmcsInfoTo = &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
1382	}
1383	else
1384	{
1385	pVmcsInfoFrom = &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
1386	pVmcsInfoTo = &pVCpu->hmr0.s.vmx.VmcsInfo;
1387	}
1388
1389	/*
1390	* Disable interrupts to prevent being preempted while we switch the current VMCS as the
1391	* preemption hook code path acquires the current VMCS.
1392	*/
1393	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1394
1395	int rc = hmR0VmxSwitchVmcs(pVmcsInfoFrom, pVmcsInfoTo);
1396	if (RT_SUCCESS(rc))
1397	{
1398	pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs = fSwitchToNstGstVmcs;
1399	pVCpu->hm.s.vmx.fSwitchedToNstGstVmcsCopyForRing3 = fSwitchToNstGstVmcs;
1400
1401	/*
1402	* If we are switching to a VMCS that was executed on a different host CPU or was
1403	* never executed before, flag that we need to export the host state before executing
1404	* guest/nested-guest code using hardware-assisted VMX.
1405	*
1406	* This could probably be done in a preemptible context since the preemption hook
1407	* will flag the necessary change in host context. However, since preemption is
1408	* already disabled and to avoid making assumptions about host specific code in
1409	* RTMpCpuId when called with preemption enabled, we'll do this while preemption is
1410	* disabled.
1411	*/
1412	if (pVmcsInfoTo->idHostCpuState == RTMpCpuId())
1413	{ /* likely */ }
1414	else
1415	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE);
1416
1417	ASMSetFlags(fEFlags);
1418
1419	/*
1420	* We use a different VM-exit MSR-store areas for the guest and nested-guest. Hence,
1421	* flag that we need to update the host MSR values there. Even if we decide in the
1422	* future to share the VM-exit MSR-store area page between the guest and nested-guest,
1423	* if its content differs, we would have to update the host MSRs anyway.
1424	*/
1425	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = false;
1426	}
1427	else
1428	ASMSetFlags(fEFlags);
1429	return rc;
1430	}
1431	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
1432
1433
1434	/**
1435	* Updates the VM's last error record.
1436	*
1437	* If there was a VMX instruction error, reads the error data from the VMCS and
1438	* updates VCPU's last error record as well.
1439	*
1440	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
1441	* Can be NULL if @a rc is not VERR_VMX_UNABLE_TO_START_VM or
1442	* VERR_VMX_INVALID_VMCS_FIELD.
1443	* @param rc The error code.
1444	*/
1445	static void hmR0VmxUpdateErrorRecord(PVMCPUCC pVCpu, int rc)
1446	{
1447	if ( rc == VERR_VMX_INVALID_VMCS_FIELD
1448	\|\| rc == VERR_VMX_UNABLE_TO_START_VM)
1449	{
1450	AssertPtrReturnVoid(pVCpu);
1451	VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
1452	}
1453	pVCpu->CTX_SUFF(pVM)->hm.s.ForR3.rcInit = rc;
1454	}
1455
1456
1457	#ifdef VBOX_STRICT
1458	/**
1459	* Reads the VM-entry interruption-information field from the VMCS into the VMX
1460	* transient structure.
1461	*
1462	* @param pVmxTransient The VMX-transient structure.
1463	*/
1464	DECLINLINE(void) hmR0VmxReadEntryIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
1465	{
1466	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &pVmxTransient->uEntryIntInfo);
1467	AssertRC(rc);
1468	}
1469
1470
1471	/**
1472	* Reads the VM-entry exception error code field from the VMCS into
1473	* the VMX transient structure.
1474	*
1475	* @param pVmxTransient The VMX-transient structure.
1476	*/
1477	DECLINLINE(void) hmR0VmxReadEntryXcptErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1478	{
1479	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &pVmxTransient->uEntryXcptErrorCode);
1480	AssertRC(rc);
1481	}
1482
1483
1484	/**
1485	* Reads the VM-entry exception error code field from the VMCS into
1486	* the VMX transient structure.
1487	*
1488	* @param pVmxTransient The VMX-transient structure.
1489	*/
1490	DECLINLINE(void) hmR0VmxReadEntryInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
1491	{
1492	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &pVmxTransient->cbEntryInstr);
1493	AssertRC(rc);
1494	}
1495	#endif /* VBOX_STRICT */
1496
1497
1498	/**
1499	* Reads the VM-exit interruption-information field from the VMCS into the VMX
1500	* transient structure.
1501	*
1502	* @param pVmxTransient The VMX-transient structure.
1503	*/
1504	DECLINLINE(void) hmR0VmxReadExitIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
1505	{
1506	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INTERRUPTION_INFO))
1507	{
1508	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
1509	AssertRC(rc);
1510	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INTERRUPTION_INFO;
1511	}
1512	}
1513
1514
1515	/**
1516	* Reads the VM-exit interruption error code from the VMCS into the VMX
1517	* transient structure.
1518	*
1519	* @param pVmxTransient The VMX-transient structure.
1520	*/
1521	DECLINLINE(void) hmR0VmxReadExitIntErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1522	{
1523	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE))
1524	{
1525	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
1526	AssertRC(rc);
1527	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE;
1528	}
1529	}
1530
1531
1532	/**
1533	* Reads the VM-exit instruction length field from the VMCS into the VMX
1534	* transient structure.
1535	*
1536	* @param pVmxTransient The VMX-transient structure.
1537	*/
1538	DECLINLINE(void) hmR0VmxReadExitInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
1539	{
1540	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INSTR_LEN))
1541	{
1542	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbExitInstr);
1543	AssertRC(rc);
1544	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INSTR_LEN;
1545	}
1546	}
1547
1548
1549	/**
1550	* Reads the VM-exit instruction-information field from the VMCS into
1551	* the VMX transient structure.
1552	*
1553	* @param pVmxTransient The VMX-transient structure.
1554	*/
1555	DECLINLINE(void) hmR0VmxReadExitInstrInfoVmcs(PVMXTRANSIENT pVmxTransient)
1556	{
1557	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INSTR_INFO))
1558	{
1559	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
1560	AssertRC(rc);
1561	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INSTR_INFO;
1562	}
1563	}
1564
1565
1566	/**
1567	* Reads the Exit Qualification from the VMCS into the VMX transient structure.
1568	*
1569	* @param pVmxTransient The VMX-transient structure.
1570	*/
1571	DECLINLINE(void) hmR0VmxReadExitQualVmcs(PVMXTRANSIENT pVmxTransient)
1572	{
1573	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_QUALIFICATION))
1574	{
1575	int rc = VMXReadVmcsNw(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQual);
1576	AssertRC(rc);
1577	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_QUALIFICATION;
1578	}
1579	}
1580
1581
1582	/**
1583	* Reads the Guest-linear address from the VMCS into the VMX transient structure.
1584	*
1585	* @param pVmxTransient The VMX-transient structure.
1586	*/
1587	DECLINLINE(void) hmR0VmxReadGuestLinearAddrVmcs(PVMXTRANSIENT pVmxTransient)
1588	{
1589	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_LINEAR_ADDR))
1590	{
1591	int rc = VMXReadVmcsNw(VMX_VMCS_RO_GUEST_LINEAR_ADDR, &pVmxTransient->uGuestLinearAddr);
1592	AssertRC(rc);
1593	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_LINEAR_ADDR;
1594	}
1595	}
1596
1597
1598	/**
1599	* Reads the Guest-physical address from the VMCS into the VMX transient structure.
1600	*
1601	* @param pVmxTransient The VMX-transient structure.
1602	*/
1603	DECLINLINE(void) hmR0VmxReadGuestPhysicalAddrVmcs(PVMXTRANSIENT pVmxTransient)
1604	{
1605	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_PHYSICAL_ADDR))
1606	{
1607	int rc = VMXReadVmcs64(VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL, &pVmxTransient->uGuestPhysicalAddr);
1608	AssertRC(rc);
1609	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_PHYSICAL_ADDR;
1610	}
1611	}
1612
1613	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1614	/**
1615	* Reads the Guest pending-debug exceptions from the VMCS into the VMX transient
1616	* structure.
1617	*
1618	* @param pVmxTransient The VMX-transient structure.
1619	*/
1620	DECLINLINE(void) hmR0VmxReadGuestPendingDbgXctps(PVMXTRANSIENT pVmxTransient)
1621	{
1622	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_PENDING_DBG_XCPTS))
1623	{
1624	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, &pVmxTransient->uGuestPendingDbgXcpts);
1625	AssertRC(rc);
1626	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_PENDING_DBG_XCPTS;
1627	}
1628	}
1629	#endif
1630
1631	/**
1632	* Reads the IDT-vectoring information field from the VMCS into the VMX
1633	* transient structure.
1634	*
1635	* @param pVmxTransient The VMX-transient structure.
1636	*
1637	* @remarks No-long-jump zone!!!
1638	*/
1639	DECLINLINE(void) hmR0VmxReadIdtVectoringInfoVmcs(PVMXTRANSIENT pVmxTransient)
1640	{
1641	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_IDT_VECTORING_INFO))
1642	{
1643	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_INFO, &pVmxTransient->uIdtVectoringInfo);
1644	AssertRC(rc);
1645	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_IDT_VECTORING_INFO;
1646	}
1647	}
1648
1649
1650	/**
1651	* Reads the IDT-vectoring error code from the VMCS into the VMX
1652	* transient structure.
1653	*
1654	* @param pVmxTransient The VMX-transient structure.
1655	*/
1656	DECLINLINE(void) hmR0VmxReadIdtVectoringErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1657	{
1658	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_IDT_VECTORING_ERROR_CODE))
1659	{
1660	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
1661	AssertRC(rc);
1662	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_IDT_VECTORING_ERROR_CODE;
1663	}
1664	}
1665
1666	#ifdef HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
1667	/**
1668	* Reads all relevant read-only VMCS fields into the VMX transient structure.
1669	*
1670	* @param pVmxTransient The VMX-transient structure.
1671	*/
1672	static void hmR0VmxReadAllRoFieldsVmcs(PVMXTRANSIENT pVmxTransient)
1673	{
1674	int rc = VMXReadVmcsNw(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQual);
1675	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbExitInstr);
1676	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
1677	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_INFO, &pVmxTransient->uIdtVectoringInfo);
1678	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
1679	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
1680	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
1681	rc \|= VMXReadVmcsNw(VMX_VMCS_RO_GUEST_LINEAR_ADDR, &pVmxTransient->uGuestLinearAddr);
1682	rc \|= VMXReadVmcs64(VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL, &pVmxTransient->uGuestPhysicalAddr);
1683	AssertRC(rc);
1684	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_QUALIFICATION
1685	\| HMVMX_READ_EXIT_INSTR_LEN
1686	\| HMVMX_READ_EXIT_INSTR_INFO
1687	\| HMVMX_READ_IDT_VECTORING_INFO
1688	\| HMVMX_READ_IDT_VECTORING_ERROR_CODE
1689	\| HMVMX_READ_EXIT_INTERRUPTION_INFO
1690	\| HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE
1691	\| HMVMX_READ_GUEST_LINEAR_ADDR
1692	\| HMVMX_READ_GUEST_PHYSICAL_ADDR;
1693	}
1694	#endif
1695
1696	/**
1697	* Enters VMX root mode operation on the current CPU.
1698	*
1699	* @returns VBox status code.
1700	* @param pHostCpu The HM physical-CPU structure.
1701	* @param pVM The cross context VM structure. Can be
1702	* NULL, after a resume.
1703	* @param HCPhysCpuPage Physical address of the VMXON region.
1704	* @param pvCpuPage Pointer to the VMXON region.
1705	*/
1706	static int hmR0VmxEnterRootMode(PHMPHYSCPU pHostCpu, PVMCC pVM, RTHCPHYS HCPhysCpuPage, void *pvCpuPage)
1707	{
1708	Assert(pHostCpu);
1709	Assert(HCPhysCpuPage && HCPhysCpuPage != NIL_RTHCPHYS);
1710	Assert(RT_ALIGN_T(HCPhysCpuPage, _4K, RTHCPHYS) == HCPhysCpuPage);
1711	Assert(pvCpuPage);
1712	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1713
1714	if (pVM)
1715	{
1716	/* Write the VMCS revision identifier to the VMXON region. */
1717	(uint32_t )pvCpuPage = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_ID);
1718	}
1719
1720	/* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with CR4. */
1721	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1722
1723	/* Enable the VMX bit in CR4 if necessary. */
1724	RTCCUINTREG const uOldCr4 = SUPR0ChangeCR4(X86_CR4_VMXE, RTCCUINTREG_MAX);
1725
1726	/* Record whether VMXE was already prior to us enabling it above. */
1727	pHostCpu->fVmxeAlreadyEnabled = RT_BOOL(uOldCr4 & X86_CR4_VMXE);
1728
1729	/* Enter VMX root mode. */
1730	int rc = VMXEnable(HCPhysCpuPage);
1731	if (RT_FAILURE(rc))
1732	{
1733	/* Restore CR4.VMXE if it was not set prior to our attempt to set it above. */
1734	if (!pHostCpu->fVmxeAlreadyEnabled)
1735	SUPR0ChangeCR4(0 /* fOrMask */, ~(uint64_t)X86_CR4_VMXE);
1736
1737	if (pVM)
1738	pVM->hm.s.ForR3.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
1739	}
1740
1741	/* Restore interrupts. */
1742	ASMSetFlags(fEFlags);
1743	return rc;
1744	}
1745
1746
1747	/**
1748	* Exits VMX root mode operation on the current CPU.
1749	*
1750	* @returns VBox status code.
1751	* @param pHostCpu The HM physical-CPU structure.
1752	*/
1753	static int hmR0VmxLeaveRootMode(PHMPHYSCPU pHostCpu)
1754	{
1755	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1756
1757	/* Paranoid: Disable interrupts as, in theory, interrupts handlers might mess with CR4. */
1758	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1759
1760	/* If we're for some reason not in VMX root mode, then don't leave it. */
1761	RTCCUINTREG const uHostCr4 = ASMGetCR4();
1762
1763	int rc;
1764	if (uHostCr4 & X86_CR4_VMXE)
1765	{
1766	/* Exit VMX root mode and clear the VMX bit in CR4. */
1767	VMXDisable();
1768
1769	/* Clear CR4.VMXE only if it was clear prior to use setting it. */
1770	if (!pHostCpu->fVmxeAlreadyEnabled)
1771	SUPR0ChangeCR4(0 /* fOrMask */, ~(uint64_t)X86_CR4_VMXE);
1772
1773	rc = VINF_SUCCESS;
1774	}
1775	else
1776	rc = VERR_VMX_NOT_IN_VMX_ROOT_MODE;
1777
1778	/* Restore interrupts. */
1779	ASMSetFlags(fEFlags);
1780	return rc;
1781	}
1782
1783
1784	/**
1785	* Allocates pages specified as specified by an array of VMX page allocation info
1786	* objects.
1787	*
1788	* The pages contents are zero'd after allocation.
1789	*
1790	* @returns VBox status code.
1791	* @param phMemObj Where to return the handle to the allocation.
1792	* @param paAllocInfo The pointer to the first element of the VMX
1793	* page-allocation info object array.
1794	* @param cEntries The number of elements in the @a paAllocInfo array.
1795	*/
1796	static int hmR0VmxPagesAllocZ(PRTR0MEMOBJ phMemObj, PVMXPAGEALLOCINFO paAllocInfo, uint32_t cEntries)
1797	{
1798	*phMemObj = NIL_RTR0MEMOBJ;
1799
1800	/* Figure out how many pages to allocate. */
1801	uint32_t cPages = 0;
1802	for (uint32_t iPage = 0; iPage < cEntries; iPage++)
1803	cPages += !!paAllocInfo[iPage].fValid;
1804
1805	/* Allocate the pages. */
1806	if (cPages)
1807	{
1808	size_t const cbPages = cPages << PAGE_SHIFT;
1809	int rc = RTR0MemObjAllocPage(phMemObj, cbPages, false /* fExecutable */);
1810	if (RT_FAILURE(rc))
1811	return rc;
1812
1813	/* Zero the contents and assign each page to the corresponding VMX page-allocation entry. */
1814	void pvFirstPage = RTR0MemObjAddress(phMemObj);
1815	RT_BZERO(pvFirstPage, cbPages);
1816
1817	uint32_t iPage = 0;
1818	for (uint32_t i = 0; i < cEntries; i++)
1819	if (paAllocInfo[i].fValid)
1820	{
1821	RTHCPHYS const HCPhysPage = RTR0MemObjGetPagePhysAddr(*phMemObj, iPage);
1822	void pvPage = (void )((uintptr_t)pvFirstPage + (iPage << X86_PAGE_4K_SHIFT));
1823	Assert(HCPhysPage && HCPhysPage != NIL_RTHCPHYS);
1824	AssertPtr(pvPage);
1825
1826	Assert(paAllocInfo[iPage].pHCPhys);
1827	Assert(paAllocInfo[iPage].ppVirt);
1828	*paAllocInfo[iPage].pHCPhys = HCPhysPage;
1829	*paAllocInfo[iPage].ppVirt = pvPage;
1830
1831	/* Move to next page. */
1832	++iPage;
1833	}
1834
1835	/* Make sure all valid (requested) pages have been assigned. */
1836	Assert(iPage == cPages);
1837	}
1838	return VINF_SUCCESS;
1839	}
1840
1841
1842	/**
1843	* Frees pages allocated using hmR0VmxPagesAllocZ.
1844	*
1845	* @param phMemObj Pointer to the memory object handle. Will be set to
1846	* NIL.
1847	*/
1848	DECL_FORCE_INLINE(void) hmR0VmxPagesFree(PRTR0MEMOBJ phMemObj)
1849	{
1850	/* We can cleanup wholesale since it's all one allocation. */
1851	if (*phMemObj != NIL_RTR0MEMOBJ)
1852	{
1853	RTR0MemObjFree(phMemObj, true / fFreeMappings */);
1854	*phMemObj = NIL_RTR0MEMOBJ;
1855	}
1856	}
1857
1858
1859	/**
1860	* Initializes a VMCS info. object.
1861	*
1862	* @param pVmcsInfo The VMCS info. object.
1863	* @param pVmcsInfoShared The VMCS info. object shared with ring-3.
1864	*/
1865	static void hmR0VmxVmcsInfoInit(PVMXVMCSINFO pVmcsInfo, PVMXVMCSINFOSHARED pVmcsInfoShared)
1866	{
1867	RT_ZERO(*pVmcsInfo);
1868	RT_ZERO(*pVmcsInfoShared);
1869
1870	pVmcsInfo->pShared = pVmcsInfoShared;
1871	Assert(pVmcsInfo->hMemObj == NIL_RTR0MEMOBJ);
1872	pVmcsInfo->HCPhysVmcs = NIL_RTHCPHYS;
1873	pVmcsInfo->HCPhysShadowVmcs = NIL_RTHCPHYS;
1874	pVmcsInfo->HCPhysMsrBitmap = NIL_RTHCPHYS;
1875	pVmcsInfo->HCPhysGuestMsrLoad = NIL_RTHCPHYS;
1876	pVmcsInfo->HCPhysGuestMsrStore = NIL_RTHCPHYS;
1877	pVmcsInfo->HCPhysHostMsrLoad = NIL_RTHCPHYS;
1878	pVmcsInfo->HCPhysVirtApic = NIL_RTHCPHYS;
1879	pVmcsInfo->HCPhysEPTP = NIL_RTHCPHYS;
1880	pVmcsInfo->u64VmcsLinkPtr = NIL_RTHCPHYS;
1881	pVmcsInfo->idHostCpuState = NIL_RTCPUID;
1882	pVmcsInfo->idHostCpuExec = NIL_RTCPUID;
1883	}
1884
1885
1886	/**
1887	* Frees the VT-x structures for a VMCS info. object.
1888	*
1889	* @param pVmcsInfo The VMCS info. object.
1890	* @param pVmcsInfoShared The VMCS info. object shared with ring-3.
1891	*/
1892	static void hmR0VmxVmcsInfoFree(PVMXVMCSINFO pVmcsInfo, PVMXVMCSINFOSHARED pVmcsInfoShared)
1893	{
1894	hmR0VmxPagesFree(&pVmcsInfo->hMemObj);
1895	hmR0VmxVmcsInfoInit(pVmcsInfo, pVmcsInfoShared);
1896	}
1897
1898
1899	/**
1900	* Allocates the VT-x structures for a VMCS info. object.
1901	*
1902	* @returns VBox status code.
1903	* @param pVCpu The cross context virtual CPU structure.
1904	* @param pVmcsInfo The VMCS info. object.
1905	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
1906	*
1907	* @remarks The caller is expected to take care of any and all allocation failures.
1908	* This function will not perform any cleanup for failures half-way
1909	* through.
1910	*/
1911	static int hmR0VmxAllocVmcsInfo(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
1912	{
1913	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1914
1915	bool const fMsrBitmaps = RT_BOOL(g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS);
1916	bool const fShadowVmcs = !fIsNstGstVmcs ? pVM->hmr0.s.vmx.fUseVmcsShadowing : pVM->cpum.ro.GuestFeatures.fVmxVmcsShadowing;
1917	Assert(!pVM->cpum.ro.GuestFeatures.fVmxVmcsShadowing); /* VMCS shadowing is not yet exposed to the guest. */
1918	VMXPAGEALLOCINFO aAllocInfo[] =
1919	{
1920	{ true, 0 /* Unused */, &pVmcsInfo->HCPhysVmcs, &pVmcsInfo->pvVmcs },
1921	{ true, 0 /* Unused */, &pVmcsInfo->HCPhysGuestMsrLoad, &pVmcsInfo->pvGuestMsrLoad },
1922	{ true, 0 /* Unused */, &pVmcsInfo->HCPhysHostMsrLoad, &pVmcsInfo->pvHostMsrLoad },
1923	{ fMsrBitmaps, 0 /* Unused */, &pVmcsInfo->HCPhysMsrBitmap, &pVmcsInfo->pvMsrBitmap },
1924	{ fShadowVmcs, 0 /* Unused */, &pVmcsInfo->HCPhysShadowVmcs, &pVmcsInfo->pvShadowVmcs },
1925	};
1926
1927	int rc = hmR0VmxPagesAllocZ(&pVmcsInfo->hMemObj, &aAllocInfo[0], RT_ELEMENTS(aAllocInfo));
1928	if (RT_FAILURE(rc))
1929	return rc;
1930
1931	/*
1932	* We use the same page for VM-entry MSR-load and VM-exit MSR store areas.
1933	* Because they contain a symmetric list of guest MSRs to load on VM-entry and store on VM-exit.
1934	*/
1935	AssertCompile(RT_ELEMENTS(aAllocInfo) > 0);
1936	Assert(pVmcsInfo->HCPhysGuestMsrLoad != NIL_RTHCPHYS);
1937	pVmcsInfo->pvGuestMsrStore = pVmcsInfo->pvGuestMsrLoad;
1938	pVmcsInfo->HCPhysGuestMsrStore = pVmcsInfo->HCPhysGuestMsrLoad;
1939
1940	/*
1941	* Get the virtual-APIC page rather than allocating them again.
1942	*/
1943	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW)
1944	{
1945	if (!fIsNstGstVmcs)
1946	{
1947	if (PDMHasApic(pVM))
1948	{
1949	rc = APICGetApicPageForCpu(pVCpu, &pVmcsInfo->HCPhysVirtApic, (PRTR0PTR)&pVmcsInfo->pbVirtApic, NULL /pR3Ptr/);
1950	if (RT_FAILURE(rc))
1951	return rc;
1952	Assert(pVmcsInfo->pbVirtApic);
1953	Assert(pVmcsInfo->HCPhysVirtApic && pVmcsInfo->HCPhysVirtApic != NIL_RTHCPHYS);
1954	}
1955	}
1956	else
1957	{
1958	pVmcsInfo->pbVirtApic = (uint8_t *)CPUMGetGuestVmxVirtApicPage(&pVCpu->cpum.GstCtx, &pVmcsInfo->HCPhysVirtApic);
1959	Assert(pVmcsInfo->pbVirtApic);
1960	Assert(pVmcsInfo->HCPhysVirtApic && pVmcsInfo->HCPhysVirtApic != NIL_RTHCPHYS);
1961	}
1962	}
1963
1964	return VINF_SUCCESS;
1965	}
1966
1967
1968	/**
1969	* Free all VT-x structures for the VM.
1970	*
1971	* @returns IPRT status code.
1972	* @param pVM The cross context VM structure.
1973	*/
1974	static void hmR0VmxStructsFree(PVMCC pVM)
1975	{
1976	hmR0VmxPagesFree(&pVM->hmr0.s.vmx.hMemObj);
1977	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1978	if (pVM->hmr0.s.vmx.fUseVmcsShadowing)
1979	{
1980	RTMemFree(pVM->hmr0.s.vmx.paShadowVmcsFields);
1981	pVM->hmr0.s.vmx.paShadowVmcsFields = NULL;
1982	RTMemFree(pVM->hmr0.s.vmx.paShadowVmcsRoFields);
1983	pVM->hmr0.s.vmx.paShadowVmcsRoFields = NULL;
1984	}
1985	#endif
1986
1987	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
1988	{
1989	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
1990	hmR0VmxVmcsInfoFree(&pVCpu->hmr0.s.vmx.VmcsInfo, &pVCpu->hm.s.vmx.VmcsInfo);
1991	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1992	if (pVM->cpum.ro.GuestFeatures.fVmx)
1993	hmR0VmxVmcsInfoFree(&pVCpu->hmr0.s.vmx.VmcsInfoNstGst, &pVCpu->hm.s.vmx.VmcsInfoNstGst);
1994	#endif
1995	}
1996	}
1997
1998
1999	/**
2000	* Allocate all VT-x structures for the VM.
2001	*
2002	* @returns IPRT status code.
2003	* @param pVM The cross context VM structure.
2004	*
2005	* @remarks This functions will cleanup on memory allocation failures.
2006	*/
2007	static int hmR0VmxStructsAlloc(PVMCC pVM)
2008	{
2009	/*
2010	* Sanity check the VMCS size reported by the CPU as we assume 4KB allocations.
2011	* The VMCS size cannot be more than 4096 bytes.
2012	*
2013	* See Intel spec. Appendix A.1 "Basic VMX Information".
2014	*/
2015	uint32_t const cbVmcs = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_SIZE);
2016	if (cbVmcs <= X86_PAGE_4K_SIZE)
2017	{ /* likely */ }
2018	else
2019	{
2020	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_INVALID_VMCS_SIZE;
2021	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2022	}
2023
2024	/*
2025	* Allocate per-VM VT-x structures.
2026	*/
2027	bool const fVirtApicAccess = RT_BOOL(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS);
2028	bool const fUseVmcsShadowing = pVM->hmr0.s.vmx.fUseVmcsShadowing;
2029	VMXPAGEALLOCINFO aAllocInfo[] =
2030	{
2031	{ fVirtApicAccess, 0 /* Unused */, &pVM->hmr0.s.vmx.HCPhysApicAccess, (PRTR0PTR)&pVM->hmr0.s.vmx.pbApicAccess },
2032	{ fUseVmcsShadowing, 0 /* Unused */, &pVM->hmr0.s.vmx.HCPhysVmreadBitmap, &pVM->hmr0.s.vmx.pvVmreadBitmap },
2033	{ fUseVmcsShadowing, 0 /* Unused */, &pVM->hmr0.s.vmx.HCPhysVmwriteBitmap, &pVM->hmr0.s.vmx.pvVmwriteBitmap },
2034	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2035	{ true, 0 /* Unused */, &pVM->hmr0.s.vmx.HCPhysScratch, (PRTR0PTR)&pVM->hmr0.s.vmx.pbScratch },
2036	#endif
2037	};
2038
2039	int rc = hmR0VmxPagesAllocZ(&pVM->hmr0.s.vmx.hMemObj, &aAllocInfo[0], RT_ELEMENTS(aAllocInfo));
2040	if (RT_SUCCESS(rc))
2041	{
2042	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2043	/* Allocate the shadow VMCS-fields array. */
2044	if (fUseVmcsShadowing)
2045	{
2046	Assert(!pVM->hmr0.s.vmx.cShadowVmcsFields);
2047	Assert(!pVM->hmr0.s.vmx.cShadowVmcsRoFields);
2048	pVM->hmr0.s.vmx.paShadowVmcsFields = (uint32_t *)RTMemAllocZ(sizeof(g_aVmcsFields));
2049	pVM->hmr0.s.vmx.paShadowVmcsRoFields = (uint32_t *)RTMemAllocZ(sizeof(g_aVmcsFields));
2050	if (!pVM->hmr0.s.vmx.paShadowVmcsFields \|\| !pVM->hmr0.s.vmx.paShadowVmcsRoFields)
2051	rc = VERR_NO_MEMORY;
2052	}
2053	#endif
2054
2055	/*
2056	* Allocate per-VCPU VT-x structures.
2057	*/
2058	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus && RT_SUCCESS(rc); idCpu++)
2059	{
2060	/* Allocate the guest VMCS structures. */
2061	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
2062	rc = hmR0VmxAllocVmcsInfo(pVCpu, &pVCpu->hmr0.s.vmx.VmcsInfo, false /* fIsNstGstVmcs */);
2063
2064	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2065	/* Allocate the nested-guest VMCS structures, when the VMX feature is exposed to the guest. */
2066	if (pVM->cpum.ro.GuestFeatures.fVmx && RT_SUCCESS(rc))
2067	rc = hmR0VmxAllocVmcsInfo(pVCpu, &pVCpu->hmr0.s.vmx.VmcsInfoNstGst, true /* fIsNstGstVmcs */);
2068	#endif
2069	}
2070	if (RT_SUCCESS(rc))
2071	return VINF_SUCCESS;
2072	}
2073	hmR0VmxStructsFree(pVM);
2074	return rc;
2075	}
2076
2077
2078	/**
2079	* Pre-initializes non-zero fields in VMX structures that will be allocated.
2080	*
2081	* @param pVM The cross context VM structure.
2082	*/
2083	static void hmR0VmxStructsInit(PVMCC pVM)
2084	{
2085	/* Paranoia. */
2086	Assert(pVM->hmr0.s.vmx.pbApicAccess == NULL);
2087	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2088	Assert(pVM->hmr0.s.vmx.pbScratch == NULL);
2089	#endif
2090
2091	/*
2092	* Initialize members up-front so we can cleanup en masse on allocation failures.
2093	*/
2094	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2095	pVM->hmr0.s.vmx.HCPhysScratch = NIL_RTHCPHYS;
2096	#endif
2097	pVM->hmr0.s.vmx.HCPhysApicAccess = NIL_RTHCPHYS;
2098	pVM->hmr0.s.vmx.HCPhysVmreadBitmap = NIL_RTHCPHYS;
2099	pVM->hmr0.s.vmx.HCPhysVmwriteBitmap = NIL_RTHCPHYS;
2100	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
2101	{
2102	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
2103	hmR0VmxVmcsInfoInit(&pVCpu->hmr0.s.vmx.VmcsInfo, &pVCpu->hm.s.vmx.VmcsInfo);
2104	hmR0VmxVmcsInfoInit(&pVCpu->hmr0.s.vmx.VmcsInfoNstGst, &pVCpu->hm.s.vmx.VmcsInfoNstGst);
2105	}
2106	}
2107
2108	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2109	/**
2110	* Returns whether an MSR at the given MSR-bitmap offset is intercepted or not.
2111	*
2112	* @returns @c true if the MSR is intercepted, @c false otherwise.
2113	* @param pvMsrBitmap The MSR bitmap.
2114	* @param offMsr The MSR byte offset.
2115	* @param iBit The bit offset from the byte offset.
2116	*/
2117	DECLINLINE(bool) hmR0VmxIsMsrBitSet(const void *pvMsrBitmap, uint16_t offMsr, int32_t iBit)
2118	{
2119	uint8_t const * const pbMsrBitmap = (uint8_t const * const)pvMsrBitmap;
2120	Assert(pbMsrBitmap);
2121	Assert(offMsr + (iBit >> 3) <= X86_PAGE_4K_SIZE);
2122	return ASMBitTest(pbMsrBitmap + offMsr, iBit);
2123	}
2124	#endif
2125
2126	/**
2127	* Sets the permission bits for the specified MSR in the given MSR bitmap.
2128	*
2129	* If the passed VMCS is a nested-guest VMCS, this function ensures that the
2130	* read/write intercept is cleared from the MSR bitmap used for hardware-assisted
2131	* VMX execution of the nested-guest, only if nested-guest is also not intercepting
2132	* the read/write access of this MSR.
2133	*
2134	* @param pVCpu The cross context virtual CPU structure.
2135	* @param pVmcsInfo The VMCS info. object.
2136	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2137	* @param idMsr The MSR value.
2138	* @param fMsrpm The MSR permissions (see VMXMSRPM_XXX). This must
2139	* include both a read -and- a write permission!
2140	*
2141	* @sa CPUMGetVmxMsrPermission.
2142	* @remarks Can be called with interrupts disabled.
2143	*/
2144	static void hmR0VmxSetMsrPermission(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs, uint32_t idMsr, uint32_t fMsrpm)
2145	{
2146	uint8_t pbMsrBitmap = (uint8_t )pVmcsInfo->pvMsrBitmap;
2147	Assert(pbMsrBitmap);
2148	Assert(VMXMSRPM_IS_FLAG_VALID(fMsrpm));
2149
2150	/*
2151	* MSR-bitmap Layout:
2152	* Byte index MSR range Interpreted as
2153	* 0x000 - 0x3ff 0x00000000 - 0x00001fff Low MSR read bits.
2154	* 0x400 - 0x7ff 0xc0000000 - 0xc0001fff High MSR read bits.
2155	* 0x800 - 0xbff 0x00000000 - 0x00001fff Low MSR write bits.
2156	* 0xc00 - 0xfff 0xc0000000 - 0xc0001fff High MSR write bits.
2157	*
2158	* A bit corresponding to an MSR within the above range causes a VM-exit
2159	* if the bit is 1 on executions of RDMSR/WRMSR. If an MSR falls out of
2160	* the MSR range, it always cause a VM-exit.
2161	*
2162	* See Intel spec. 24.6.9 "MSR-Bitmap Address".
2163	*/
2164	uint16_t const offBitmapRead = 0;
2165	uint16_t const offBitmapWrite = 0x800;
2166	uint16_t offMsr;
2167	int32_t iBit;
2168	if (idMsr <= UINT32_C(0x00001fff))
2169	{
2170	offMsr = 0;
2171	iBit = idMsr;
2172	}
2173	else if (idMsr - UINT32_C(0xc0000000) <= UINT32_C(0x00001fff))
2174	{
2175	offMsr = 0x400;
2176	iBit = idMsr - UINT32_C(0xc0000000);
2177	}
2178	else
2179	AssertMsgFailedReturnVoid(("Invalid MSR %#RX32\n", idMsr));
2180
2181	/*
2182	* Set the MSR read permission.
2183	*/
2184	uint16_t const offMsrRead = offBitmapRead + offMsr;
2185	Assert(offMsrRead + (iBit >> 3) < offBitmapWrite);
2186	if (fMsrpm & VMXMSRPM_ALLOW_RD)
2187	{
2188	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2189	bool const fClear = !fIsNstGstVmcs ? true
2190	: !hmR0VmxIsMsrBitSet(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), offMsrRead, iBit);
2191	#else
2192	RT_NOREF2(pVCpu, fIsNstGstVmcs);
2193	bool const fClear = true;
2194	#endif
2195	if (fClear)
2196	ASMBitClear(pbMsrBitmap + offMsrRead, iBit);
2197	}
2198	else
2199	ASMBitSet(pbMsrBitmap + offMsrRead, iBit);
2200
2201	/*
2202	* Set the MSR write permission.
2203	*/
2204	uint16_t const offMsrWrite = offBitmapWrite + offMsr;
2205	Assert(offMsrWrite + (iBit >> 3) < X86_PAGE_4K_SIZE);
2206	if (fMsrpm & VMXMSRPM_ALLOW_WR)
2207	{
2208	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2209	bool const fClear = !fIsNstGstVmcs ? true
2210	: !hmR0VmxIsMsrBitSet(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), offMsrWrite, iBit);
2211	#else
2212	RT_NOREF2(pVCpu, fIsNstGstVmcs);
2213	bool const fClear = true;
2214	#endif
2215	if (fClear)
2216	ASMBitClear(pbMsrBitmap + offMsrWrite, iBit);
2217	}
2218	else
2219	ASMBitSet(pbMsrBitmap + offMsrWrite, iBit);
2220	}
2221
2222
2223	/**
2224	* Updates the VMCS with the number of effective MSRs in the auto-load/store MSR
2225	* area.
2226	*
2227	* @returns VBox status code.
2228	* @param pVCpu The cross context virtual CPU structure.
2229	* @param pVmcsInfo The VMCS info. object.
2230	* @param cMsrs The number of MSRs.
2231	*/
2232	static int hmR0VmxSetAutoLoadStoreMsrCount(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint32_t cMsrs)
2233	{
2234	/* Shouldn't ever happen but there -is- a number. We're well within the recommended 512. */
2235	uint32_t const cMaxSupportedMsrs = VMX_MISC_MAX_MSRS(g_HmMsrs.u.vmx.u64Misc);
2236	if (RT_LIKELY(cMsrs < cMaxSupportedMsrs))
2237	{
2238	/* Commit the MSR counts to the VMCS and update the cache. */
2239	if (pVmcsInfo->cEntryMsrLoad != cMsrs)
2240	{
2241	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, cMsrs); AssertRC(rc);
2242	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, cMsrs); AssertRC(rc);
2243	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, cMsrs); AssertRC(rc);
2244	pVmcsInfo->cEntryMsrLoad = cMsrs;
2245	pVmcsInfo->cExitMsrStore = cMsrs;
2246	pVmcsInfo->cExitMsrLoad = cMsrs;
2247	}
2248	return VINF_SUCCESS;
2249	}
2250
2251	LogRel(("Auto-load/store MSR count exceeded! cMsrs=%u MaxSupported=%u\n", cMsrs, cMaxSupportedMsrs));
2252	pVCpu->hm.s.u32HMError = VMX_UFC_INSUFFICIENT_GUEST_MSR_STORAGE;
2253	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2254	}
2255
2256
2257	/**
2258	* Adds a new (or updates the value of an existing) guest/host MSR
2259	* pair to be swapped during the world-switch as part of the
2260	* auto-load/store MSR area in the VMCS.
2261	*
2262	* @returns VBox status code.
2263	* @param pVCpu The cross context virtual CPU structure.
2264	* @param pVmxTransient The VMX-transient structure.
2265	* @param idMsr The MSR.
2266	* @param uGuestMsrValue Value of the guest MSR.
2267	* @param fSetReadWrite Whether to set the guest read/write access of this
2268	* MSR (thus not causing a VM-exit).
2269	* @param fUpdateHostMsr Whether to update the value of the host MSR if
2270	* necessary.
2271	*/
2272	static int hmR0VmxAddAutoLoadStoreMsr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t idMsr, uint64_t uGuestMsrValue,
2273	bool fSetReadWrite, bool fUpdateHostMsr)
2274	{
2275	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
2276	bool const fIsNstGstVmcs = pVmxTransient->fIsNestedGuest;
2277	PVMXAUTOMSR pGuestMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2278	uint32_t cMsrs = pVmcsInfo->cEntryMsrLoad;
2279	uint32_t i;
2280
2281	/* Paranoia. */
2282	Assert(pGuestMsrLoad);
2283
2284	#ifndef DEBUG_bird
2285	LogFlowFunc(("pVCpu=%p idMsr=%#RX32 uGuestMsrValue=%#RX64\n", pVCpu, idMsr, uGuestMsrValue));
2286	#endif
2287
2288	/* Check if the MSR already exists in the VM-entry MSR-load area. */
2289	for (i = 0; i < cMsrs; i++)
2290	{
2291	if (pGuestMsrLoad[i].u32Msr == idMsr)
2292	break;
2293	}
2294
2295	bool fAdded = false;
2296	if (i == cMsrs)
2297	{
2298	/* The MSR does not exist, bump the MSR count to make room for the new MSR. */
2299	++cMsrs;
2300	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, pVmcsInfo, cMsrs);
2301	AssertMsgRCReturn(rc, ("Insufficient space to add MSR to VM-entry MSR-load/store area %u\n", idMsr), rc);
2302
2303	/* Set the guest to read/write this MSR without causing VM-exits. */
2304	if ( fSetReadWrite
2305	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS))
2306	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, fIsNstGstVmcs, idMsr, VMXMSRPM_ALLOW_RD_WR);
2307
2308	Log4Func(("Added MSR %#RX32, cMsrs=%u\n", idMsr, cMsrs));
2309	fAdded = true;
2310	}
2311
2312	/* Update the MSR value for the newly added or already existing MSR. */
2313	pGuestMsrLoad[i].u32Msr = idMsr;
2314	pGuestMsrLoad[i].u64Value = uGuestMsrValue;
2315
2316	/* Create the corresponding slot in the VM-exit MSR-store area if we use a different page. */
2317	if (hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo))
2318	{
2319	PVMXAUTOMSR pGuestMsrStore = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2320	pGuestMsrStore[i].u32Msr = idMsr;
2321	pGuestMsrStore[i].u64Value = uGuestMsrValue;
2322	}
2323
2324	/* Update the corresponding slot in the host MSR area. */
2325	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2326	Assert(pHostMsr != pVmcsInfo->pvGuestMsrLoad);
2327	Assert(pHostMsr != pVmcsInfo->pvGuestMsrStore);
2328	pHostMsr[i].u32Msr = idMsr;
2329
2330	/*
2331	* Only if the caller requests to update the host MSR value AND we've newly added the
2332	* MSR to the host MSR area do we actually update the value. Otherwise, it will be
2333	* updated by hmR0VmxUpdateAutoLoadHostMsrs().
2334	*
2335	* We do this for performance reasons since reading MSRs may be quite expensive.
2336	*/
2337	if (fAdded)
2338	{
2339	if (fUpdateHostMsr)
2340	{
2341	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2342	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2343	pHostMsr[i].u64Value = ASMRdMsr(idMsr);
2344	}
2345	else
2346	{
2347	/* Someone else can do the work. */
2348	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = false;
2349	}
2350	}
2351	return VINF_SUCCESS;
2352	}
2353
2354
2355	/**
2356	* Removes a guest/host MSR pair to be swapped during the world-switch from the
2357	* auto-load/store MSR area in the VMCS.
2358	*
2359	* @returns VBox status code.
2360	* @param pVCpu The cross context virtual CPU structure.
2361	* @param pVmxTransient The VMX-transient structure.
2362	* @param idMsr The MSR.
2363	*/
2364	static int hmR0VmxRemoveAutoLoadStoreMsr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t idMsr)
2365	{
2366	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
2367	bool const fIsNstGstVmcs = pVmxTransient->fIsNestedGuest;
2368	PVMXAUTOMSR pGuestMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2369	uint32_t cMsrs = pVmcsInfo->cEntryMsrLoad;
2370
2371	#ifndef DEBUG_bird
2372	LogFlowFunc(("pVCpu=%p idMsr=%#RX32\n", pVCpu, idMsr));
2373	#endif
2374
2375	for (uint32_t i = 0; i < cMsrs; i++)
2376	{
2377	/* Find the MSR. */
2378	if (pGuestMsrLoad[i].u32Msr == idMsr)
2379	{
2380	/*
2381	* If it's the last MSR, we only need to reduce the MSR count.
2382	* If it's -not- the last MSR, copy the last MSR in place of it and reduce the MSR count.
2383	*/
2384	if (i < cMsrs - 1)
2385	{
2386	/* Remove it from the VM-entry MSR-load area. */
2387	pGuestMsrLoad[i].u32Msr = pGuestMsrLoad[cMsrs - 1].u32Msr;
2388	pGuestMsrLoad[i].u64Value = pGuestMsrLoad[cMsrs - 1].u64Value;
2389
2390	/* Remove it from the VM-exit MSR-store area if it's in a different page. */
2391	if (hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo))
2392	{
2393	PVMXAUTOMSR pGuestMsrStore = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2394	Assert(pGuestMsrStore[i].u32Msr == idMsr);
2395	pGuestMsrStore[i].u32Msr = pGuestMsrStore[cMsrs - 1].u32Msr;
2396	pGuestMsrStore[i].u64Value = pGuestMsrStore[cMsrs - 1].u64Value;
2397	}
2398
2399	/* Remove it from the VM-exit MSR-load area. */
2400	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2401	Assert(pHostMsr[i].u32Msr == idMsr);
2402	pHostMsr[i].u32Msr = pHostMsr[cMsrs - 1].u32Msr;
2403	pHostMsr[i].u64Value = pHostMsr[cMsrs - 1].u64Value;
2404	}
2405
2406	/* Reduce the count to reflect the removed MSR and bail. */
2407	--cMsrs;
2408	break;
2409	}
2410	}
2411
2412	/* Update the VMCS if the count changed (meaning the MSR was found and removed). */
2413	if (cMsrs != pVmcsInfo->cEntryMsrLoad)
2414	{
2415	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, pVmcsInfo, cMsrs);
2416	AssertRCReturn(rc, rc);
2417
2418	/* We're no longer swapping MSRs during the world-switch, intercept guest read/writes to them. */
2419	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
2420	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, fIsNstGstVmcs, idMsr, VMXMSRPM_EXIT_RD \| VMXMSRPM_EXIT_WR);
2421
2422	Log4Func(("Removed MSR %#RX32, cMsrs=%u\n", idMsr, cMsrs));
2423	return VINF_SUCCESS;
2424	}
2425
2426	return VERR_NOT_FOUND;
2427	}
2428
2429
2430	/**
2431	* Checks if the specified guest MSR is part of the VM-entry MSR-load area.
2432	*
2433	* @returns @c true if found, @c false otherwise.
2434	* @param pVmcsInfo The VMCS info. object.
2435	* @param idMsr The MSR to find.
2436	*/
2437	static bool hmR0VmxIsAutoLoadGuestMsr(PCVMXVMCSINFO pVmcsInfo, uint32_t idMsr)
2438	{
2439	PCVMXAUTOMSR pMsrs = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2440	uint32_t const cMsrs = pVmcsInfo->cEntryMsrLoad;
2441	Assert(pMsrs);
2442	Assert(sizeof(pMsrs) cMsrs <= X86_PAGE_4K_SIZE);
2443	for (uint32_t i = 0; i < cMsrs; i++)
2444	{
2445	if (pMsrs[i].u32Msr == idMsr)
2446	return true;
2447	}
2448	return false;
2449	}
2450
2451
2452	/**
2453	* Updates the value of all host MSRs in the VM-exit MSR-load area.
2454	*
2455	* @param pVCpu The cross context virtual CPU structure.
2456	* @param pVmcsInfo The VMCS info. object.
2457	*
2458	* @remarks No-long-jump zone!!!
2459	*/
2460	static void hmR0VmxUpdateAutoLoadHostMsrs(PCVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
2461	{
2462	RT_NOREF(pVCpu);
2463	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2464
2465	PVMXAUTOMSR pHostMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2466	uint32_t const cMsrs = pVmcsInfo->cExitMsrLoad;
2467	Assert(pHostMsrLoad);
2468	Assert(sizeof(pHostMsrLoad) cMsrs <= X86_PAGE_4K_SIZE);
2469	LogFlowFunc(("pVCpu=%p cMsrs=%u\n", pVCpu, cMsrs));
2470	for (uint32_t i = 0; i < cMsrs; i++)
2471	{
2472	/*
2473	* Performance hack for the host EFER MSR. We use the cached value rather than re-read it.
2474	* Strict builds will catch mismatches in hmR0VmxCheckAutoLoadStoreMsrs(). See @bugref{7368}.
2475	*/
2476	if (pHostMsrLoad[i].u32Msr == MSR_K6_EFER)
2477	pHostMsrLoad[i].u64Value = g_uHmVmxHostMsrEfer;
2478	else
2479	pHostMsrLoad[i].u64Value = ASMRdMsr(pHostMsrLoad[i].u32Msr);
2480	}
2481	}
2482
2483
2484	/**
2485	* Saves a set of host MSRs to allow read/write passthru access to the guest and
2486	* perform lazy restoration of the host MSRs while leaving VT-x.
2487	*
2488	* @param pVCpu The cross context virtual CPU structure.
2489	*
2490	* @remarks No-long-jump zone!!!
2491	*/
2492	static void hmR0VmxLazySaveHostMsrs(PVMCPUCC pVCpu)
2493	{
2494	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2495
2496	/*
2497	* Note: If you're adding MSRs here, make sure to update the MSR-bitmap accesses in hmR0VmxSetupVmcsProcCtls().
2498	*/
2499	if (!(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST))
2500	{
2501	Assert(!(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)); /* Guest MSRs better not be loaded now. */
2502	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fAllow64BitGuests)
2503	{
2504	pVCpu->hmr0.s.vmx.u64HostMsrLStar = ASMRdMsr(MSR_K8_LSTAR);
2505	pVCpu->hmr0.s.vmx.u64HostMsrStar = ASMRdMsr(MSR_K6_STAR);
2506	pVCpu->hmr0.s.vmx.u64HostMsrSfMask = ASMRdMsr(MSR_K8_SF_MASK);
2507	pVCpu->hmr0.s.vmx.u64HostMsrKernelGsBase = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
2508	}
2509	pVCpu->hmr0.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_SAVED_HOST;
2510	}
2511	}
2512
2513
2514	/**
2515	* Checks whether the MSR belongs to the set of guest MSRs that we restore
2516	* lazily while leaving VT-x.
2517	*
2518	* @returns true if it does, false otherwise.
2519	* @param pVCpu The cross context virtual CPU structure.
2520	* @param idMsr The MSR to check.
2521	*/
2522	static bool hmR0VmxIsLazyGuestMsr(PCVMCPUCC pVCpu, uint32_t idMsr)
2523	{
2524	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fAllow64BitGuests)
2525	{
2526	switch (idMsr)
2527	{
2528	case MSR_K8_LSTAR:
2529	case MSR_K6_STAR:
2530	case MSR_K8_SF_MASK:
2531	case MSR_K8_KERNEL_GS_BASE:
2532	return true;
2533	}
2534	}
2535	return false;
2536	}
2537
2538
2539	/**
2540	* Loads a set of guests MSRs to allow read/passthru to the guest.
2541	*
2542	* The name of this function is slightly confusing. This function does NOT
2543	* postpone loading, but loads the MSR right now. "hmR0VmxLazy" is simply a
2544	* common prefix for functions dealing with "lazy restoration" of the shared
2545	* MSRs.
2546	*
2547	* @param pVCpu The cross context virtual CPU structure.
2548	*
2549	* @remarks No-long-jump zone!!!
2550	*/
2551	static void hmR0VmxLazyLoadGuestMsrs(PVMCPUCC pVCpu)
2552	{
2553	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2554	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2555
2556	Assert(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
2557	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fAllow64BitGuests)
2558	{
2559	/*
2560	* If the guest MSRs are not loaded -and- if all the guest MSRs are identical
2561	* to the MSRs on the CPU (which are the saved host MSRs, see assertion above) then
2562	* we can skip a few MSR writes.
2563	*
2564	* Otherwise, it implies either 1. they're not loaded, or 2. they're loaded but the
2565	* guest MSR values in the guest-CPU context might be different to what's currently
2566	* loaded in the CPU. In either case, we need to write the new guest MSR values to the
2567	* CPU, see @bugref{8728}.
2568	*/
2569	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
2570	if ( !(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
2571	&& pCtx->msrKERNELGSBASE == pVCpu->hmr0.s.vmx.u64HostMsrKernelGsBase
2572	&& pCtx->msrLSTAR == pVCpu->hmr0.s.vmx.u64HostMsrLStar
2573	&& pCtx->msrSTAR == pVCpu->hmr0.s.vmx.u64HostMsrStar
2574	&& pCtx->msrSFMASK == pVCpu->hmr0.s.vmx.u64HostMsrSfMask)
2575	{
2576	#ifdef VBOX_STRICT
2577	Assert(ASMRdMsr(MSR_K8_KERNEL_GS_BASE) == pCtx->msrKERNELGSBASE);
2578	Assert(ASMRdMsr(MSR_K8_LSTAR) == pCtx->msrLSTAR);
2579	Assert(ASMRdMsr(MSR_K6_STAR) == pCtx->msrSTAR);
2580	Assert(ASMRdMsr(MSR_K8_SF_MASK) == pCtx->msrSFMASK);
2581	#endif
2582	}
2583	else
2584	{
2585	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pCtx->msrKERNELGSBASE);
2586	ASMWrMsr(MSR_K8_LSTAR, pCtx->msrLSTAR);
2587	ASMWrMsr(MSR_K6_STAR, pCtx->msrSTAR);
2588	/* The system call flag mask register isn't as benign and accepting of all
2589	values as the above, so mask it to avoid #GP'ing on corrupted input. */
2590	Assert(!(pCtx->msrSFMASK & ~(uint64_t)UINT32_MAX));
2591	ASMWrMsr(MSR_K8_SF_MASK, pCtx->msrSFMASK & UINT32_MAX);
2592	}
2593	}
2594	pVCpu->hmr0.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_LOADED_GUEST;
2595	}
2596
2597
2598	/**
2599	* Performs lazy restoration of the set of host MSRs if they were previously
2600	* loaded with guest MSR values.
2601	*
2602	* @param pVCpu The cross context virtual CPU structure.
2603	*
2604	* @remarks No-long-jump zone!!!
2605	* @remarks The guest MSRs should have been saved back into the guest-CPU
2606	* context by hmR0VmxImportGuestState()!!!
2607	*/
2608	static void hmR0VmxLazyRestoreHostMsrs(PVMCPUCC pVCpu)
2609	{
2610	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2611	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2612
2613	if (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
2614	{
2615	Assert(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
2616	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fAllow64BitGuests)
2617	{
2618	ASMWrMsr(MSR_K8_LSTAR, pVCpu->hmr0.s.vmx.u64HostMsrLStar);
2619	ASMWrMsr(MSR_K6_STAR, pVCpu->hmr0.s.vmx.u64HostMsrStar);
2620	ASMWrMsr(MSR_K8_SF_MASK, pVCpu->hmr0.s.vmx.u64HostMsrSfMask);
2621	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pVCpu->hmr0.s.vmx.u64HostMsrKernelGsBase);
2622	}
2623	}
2624	pVCpu->hmr0.s.vmx.fLazyMsrs &= ~(VMX_LAZY_MSRS_LOADED_GUEST \| VMX_LAZY_MSRS_SAVED_HOST);
2625	}
2626
2627
2628	/**
2629	* Verifies that our cached values of the VMCS fields are all consistent with
2630	* what's actually present in the VMCS.
2631	*
2632	* @returns VBox status code.
2633	* @retval VINF_SUCCESS if all our caches match their respective VMCS fields.
2634	* @retval VERR_VMX_VMCS_FIELD_CACHE_INVALID if a cache field doesn't match the
2635	* VMCS content. HMCPU error-field is
2636	* updated, see VMX_VCI_XXX.
2637	* @param pVCpu The cross context virtual CPU structure.
2638	* @param pVmcsInfo The VMCS info. object.
2639	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2640	*/
2641	static int hmR0VmxCheckCachedVmcsCtls(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
2642	{
2643	const char * const pcszVmcs = fIsNstGstVmcs ? "Nested-guest VMCS" : "VMCS";
2644
2645	uint32_t u32Val;
2646	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
2647	AssertRC(rc);
2648	AssertMsgReturnStmt(pVmcsInfo->u32EntryCtls == u32Val,
2649	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32EntryCtls, u32Val),
2650	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_ENTRY,
2651	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2652
2653	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val);
2654	AssertRC(rc);
2655	AssertMsgReturnStmt(pVmcsInfo->u32ExitCtls == u32Val,
2656	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ExitCtls, u32Val),
2657	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_EXIT,
2658	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2659
2660	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val);
2661	AssertRC(rc);
2662	AssertMsgReturnStmt(pVmcsInfo->u32PinCtls == u32Val,
2663	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32PinCtls, u32Val),
2664	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PIN_EXEC,
2665	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2666
2667	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val);
2668	AssertRC(rc);
2669	AssertMsgReturnStmt(pVmcsInfo->u32ProcCtls == u32Val,
2670	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ProcCtls, u32Val),
2671	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC,
2672	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2673
2674	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
2675	{
2676	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val);
2677	AssertRC(rc);
2678	AssertMsgReturnStmt(pVmcsInfo->u32ProcCtls2 == u32Val,
2679	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ProcCtls2, u32Val),
2680	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC2,
2681	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2682	}
2683
2684	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, &u32Val);
2685	AssertRC(rc);
2686	AssertMsgReturnStmt(pVmcsInfo->u32XcptBitmap == u32Val,
2687	("%s exception bitmap mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32XcptBitmap, u32Val),
2688	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_XCPT_BITMAP,
2689	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2690
2691	uint64_t u64Val;
2692	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, &u64Val);
2693	AssertRC(rc);
2694	AssertMsgReturnStmt(pVmcsInfo->u64TscOffset == u64Val,
2695	("%s TSC offset mismatch: Cache=%#RX64 VMCS=%#RX64\n", pcszVmcs, pVmcsInfo->u64TscOffset, u64Val),
2696	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_TSC_OFFSET,
2697	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2698
2699	NOREF(pcszVmcs);
2700	return VINF_SUCCESS;
2701	}
2702
2703	#ifdef VBOX_STRICT
2704
2705	/**
2706	* Verifies that our cached host EFER MSR value has not changed since we cached it.
2707	*
2708	* @param pVmcsInfo The VMCS info. object.
2709	*/
2710	static void hmR0VmxCheckHostEferMsr(PCVMXVMCSINFO pVmcsInfo)
2711	{
2712	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2713
2714	if (pVmcsInfo->u32ExitCtls & VMX_EXIT_CTLS_LOAD_EFER_MSR)
2715	{
2716	uint64_t const uHostEferMsr = ASMRdMsr(MSR_K6_EFER);
2717	uint64_t const uHostEferMsrCache = g_uHmVmxHostMsrEfer;
2718	uint64_t uVmcsEferMsrVmcs;
2719	int rc = VMXReadVmcs64(VMX_VMCS64_HOST_EFER_FULL, &uVmcsEferMsrVmcs);
2720	AssertRC(rc);
2721
2722	AssertMsgReturnVoid(uHostEferMsr == uVmcsEferMsrVmcs,
2723	("EFER Host/VMCS mismatch! host=%#RX64 vmcs=%#RX64\n", uHostEferMsr, uVmcsEferMsrVmcs));
2724	AssertMsgReturnVoid(uHostEferMsr == uHostEferMsrCache,
2725	("EFER Host/Cache mismatch! host=%#RX64 cache=%#RX64\n", uHostEferMsr, uHostEferMsrCache));
2726	}
2727	}
2728
2729
2730	/**
2731	* Verifies whether the guest/host MSR pairs in the auto-load/store area in the
2732	* VMCS are correct.
2733	*
2734	* @param pVCpu The cross context virtual CPU structure.
2735	* @param pVmcsInfo The VMCS info. object.
2736	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2737	*/
2738	static void hmR0VmxCheckAutoLoadStoreMsrs(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
2739	{
2740	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2741
2742	/* Read the various MSR-area counts from the VMCS. */
2743	uint32_t cEntryLoadMsrs;
2744	uint32_t cExitStoreMsrs;
2745	uint32_t cExitLoadMsrs;
2746	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &cEntryLoadMsrs); AssertRC(rc);
2747	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &cExitStoreMsrs); AssertRC(rc);
2748	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &cExitLoadMsrs); AssertRC(rc);
2749
2750	/* Verify all the MSR counts are the same. */
2751	Assert(cEntryLoadMsrs == cExitStoreMsrs);
2752	Assert(cExitStoreMsrs == cExitLoadMsrs);
2753	uint32_t const cMsrs = cExitLoadMsrs;
2754
2755	/* Verify the MSR counts do not exceed the maximum count supported by the hardware. */
2756	Assert(cMsrs < VMX_MISC_MAX_MSRS(g_HmMsrs.u.vmx.u64Misc));
2757
2758	/* Verify the MSR counts are within the allocated page size. */
2759	Assert(sizeof(VMXAUTOMSR) * cMsrs <= X86_PAGE_4K_SIZE);
2760
2761	/* Verify the relevant contents of the MSR areas match. */
2762	PCVMXAUTOMSR pGuestMsrLoad = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2763	PCVMXAUTOMSR pGuestMsrStore = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2764	PCVMXAUTOMSR pHostMsrLoad = (PCVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2765	bool const fSeparateExitMsrStorePage = hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo);
2766	for (uint32_t i = 0; i < cMsrs; i++)
2767	{
2768	/* Verify that the MSRs are paired properly and that the host MSR has the correct value. */
2769	if (fSeparateExitMsrStorePage)
2770	{
2771	AssertMsgReturnVoid(pGuestMsrLoad->u32Msr == pGuestMsrStore->u32Msr,
2772	("GuestMsrLoad=%#RX32 GuestMsrStore=%#RX32 cMsrs=%u\n",
2773	pGuestMsrLoad->u32Msr, pGuestMsrStore->u32Msr, cMsrs));
2774	}
2775
2776	AssertMsgReturnVoid(pHostMsrLoad->u32Msr == pGuestMsrLoad->u32Msr,
2777	("HostMsrLoad=%#RX32 GuestMsrLoad=%#RX32 cMsrs=%u\n",
2778	pHostMsrLoad->u32Msr, pGuestMsrLoad->u32Msr, cMsrs));
2779
2780	uint64_t const u64HostMsr = ASMRdMsr(pHostMsrLoad->u32Msr);
2781	AssertMsgReturnVoid(pHostMsrLoad->u64Value == u64HostMsr,
2782	("u32Msr=%#RX32 VMCS Value=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
2783	pHostMsrLoad->u32Msr, pHostMsrLoad->u64Value, u64HostMsr, cMsrs));
2784
2785	/* Verify that cached host EFER MSR matches what's loaded on the CPU. */
2786	bool const fIsEferMsr = RT_BOOL(pHostMsrLoad->u32Msr == MSR_K6_EFER);
2787	AssertMsgReturnVoid(!fIsEferMsr \|\| u64HostMsr == g_uHmVmxHostMsrEfer,
2788	("Cached=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n", g_uHmVmxHostMsrEfer, u64HostMsr, cMsrs));
2789
2790	/* Verify that the accesses are as expected in the MSR bitmap for auto-load/store MSRs. */
2791	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
2792	{
2793	uint32_t const fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, pGuestMsrLoad->u32Msr);
2794	if (fIsEferMsr)
2795	{
2796	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_EXIT_RD), ("Passthru read for EFER MSR!?\n"));
2797	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_EXIT_WR), ("Passthru write for EFER MSR!?\n"));
2798	}
2799	else
2800	{
2801	/* Verify LBR MSRs (used only for debugging) are intercepted. We don't passthru these MSRs to the guest yet. */
2802	PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
2803	if ( pVM->hmr0.s.vmx.fLbr
2804	&& ( hmR0VmxIsLbrBranchFromMsr(pVM, pGuestMsrLoad->u32Msr, NULL /* pidxMsr */)
2805	\|\| hmR0VmxIsLbrBranchToMsr(pVM, pGuestMsrLoad->u32Msr, NULL /* pidxMsr */)
2806	\|\| pGuestMsrLoad->u32Msr == pVM->hmr0.s.vmx.idLbrTosMsr))
2807	{
2808	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_MASK) == VMXMSRPM_EXIT_RD_WR,
2809	("u32Msr=%#RX32 cMsrs=%u Passthru read/write for LBR MSRs!\n",
2810	pGuestMsrLoad->u32Msr, cMsrs));
2811	}
2812	else if (!fIsNstGstVmcs)
2813	{
2814	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_MASK) == VMXMSRPM_ALLOW_RD_WR,
2815	("u32Msr=%#RX32 cMsrs=%u No passthru read/write!\n", pGuestMsrLoad->u32Msr, cMsrs));
2816	}
2817	else
2818	{
2819	/*
2820	* A nested-guest VMCS must -also- allow read/write passthrough for the MSR for us to
2821	* execute a nested-guest with MSR passthrough.
2822	*
2823	* Check if the nested-guest MSR bitmap allows passthrough, and if so, assert that we
2824	* allow passthrough too.
2825	*/
2826	void const *pvMsrBitmapNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap);
2827	Assert(pvMsrBitmapNstGst);
2828	uint32_t const fMsrpmNstGst = CPUMGetVmxMsrPermission(pvMsrBitmapNstGst, pGuestMsrLoad->u32Msr);
2829	AssertMsgReturnVoid(fMsrpm == fMsrpmNstGst,
2830	("u32Msr=%#RX32 cMsrs=%u Permission mismatch fMsrpm=%#x fMsrpmNstGst=%#x!\n",
2831	pGuestMsrLoad->u32Msr, cMsrs, fMsrpm, fMsrpmNstGst));
2832	}
2833	}
2834	}
2835
2836	/* Move to the next MSR. */
2837	pHostMsrLoad++;
2838	pGuestMsrLoad++;
2839	pGuestMsrStore++;
2840	}
2841	}
2842
2843	#endif /* VBOX_STRICT */
2844
2845	/**
2846	* Flushes the TLB using EPT.
2847	*
2848	* @returns VBox status code.
2849	* @param pVCpu The cross context virtual CPU structure of the calling
2850	* EMT. Can be NULL depending on @a enmTlbFlush.
2851	* @param pVmcsInfo The VMCS info. object. Can be NULL depending on @a
2852	* enmTlbFlush.
2853	* @param enmTlbFlush Type of flush.
2854	*
2855	* @remarks Caller is responsible for making sure this function is called only
2856	* when NestedPaging is supported and providing @a enmTlbFlush that is
2857	* supported by the CPU.
2858	* @remarks Can be called with interrupts disabled.
2859	*/
2860	static void hmR0VmxFlushEpt(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, VMXTLBFLUSHEPT enmTlbFlush)
2861	{
2862	uint64_t au64Descriptor[2];
2863	if (enmTlbFlush == VMXTLBFLUSHEPT_ALL_CONTEXTS)
2864	au64Descriptor[0] = 0;
2865	else
2866	{
2867	Assert(pVCpu);
2868	Assert(pVmcsInfo);
2869	au64Descriptor[0] = pVmcsInfo->HCPhysEPTP;
2870	}
2871	au64Descriptor[1] = 0; /* MBZ. Intel spec. 33.3 "VMX Instructions" */
2872
2873	int rc = VMXR0InvEPT(enmTlbFlush, &au64Descriptor[0]);
2874	AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvEPT %#x %#RHp failed. rc=%Rrc\n", enmTlbFlush, au64Descriptor[0], rc));
2875
2876	if ( RT_SUCCESS(rc)
2877	&& pVCpu)
2878	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushNestedPaging);
2879	}
2880
2881
2882	/**
2883	* Flushes the TLB using VPID.
2884	*
2885	* @returns VBox status code.
2886	* @param pVCpu The cross context virtual CPU structure of the calling
2887	* EMT. Can be NULL depending on @a enmTlbFlush.
2888	* @param enmTlbFlush Type of flush.
2889	* @param GCPtr Virtual address of the page to flush (can be 0 depending
2890	* on @a enmTlbFlush).
2891	*
2892	* @remarks Can be called with interrupts disabled.
2893	*/
2894	static void hmR0VmxFlushVpid(PVMCPUCC pVCpu, VMXTLBFLUSHVPID enmTlbFlush, RTGCPTR GCPtr)
2895	{
2896	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fVpid);
2897
2898	uint64_t au64Descriptor[2];
2899	if (enmTlbFlush == VMXTLBFLUSHVPID_ALL_CONTEXTS)
2900	{
2901	au64Descriptor[0] = 0;
2902	au64Descriptor[1] = 0;
2903	}
2904	else
2905	{
2906	AssertPtr(pVCpu);
2907	AssertMsg(pVCpu->hmr0.s.uCurrentAsid != 0, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hmr0.s.uCurrentAsid));
2908	AssertMsg(pVCpu->hmr0.s.uCurrentAsid <= UINT16_MAX, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hmr0.s.uCurrentAsid));
2909	au64Descriptor[0] = pVCpu->hmr0.s.uCurrentAsid;
2910	au64Descriptor[1] = GCPtr;
2911	}
2912
2913	int rc = VMXR0InvVPID(enmTlbFlush, &au64Descriptor[0]);
2914	AssertMsg(rc == VINF_SUCCESS,
2915	("VMXR0InvVPID %#x %u %RGv failed with %Rrc\n", enmTlbFlush, pVCpu ? pVCpu->hmr0.s.uCurrentAsid : 0, GCPtr, rc));
2916
2917	if ( RT_SUCCESS(rc)
2918	&& pVCpu)
2919	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid);
2920	NOREF(rc);
2921	}
2922
2923
2924	/**
2925	* Invalidates a guest page by guest virtual address. Only relevant for EPT/VPID,
2926	* otherwise there is nothing really to invalidate.
2927	*
2928	* @returns VBox status code.
2929	* @param pVCpu The cross context virtual CPU structure.
2930	* @param GCVirt Guest virtual address of the page to invalidate.
2931	*/
2932	VMMR0DECL(int) VMXR0InvalidatePage(PVMCPUCC pVCpu, RTGCPTR GCVirt)
2933	{
2934	AssertPtr(pVCpu);
2935	LogFlowFunc(("pVCpu=%p GCVirt=%RGv\n", pVCpu, GCVirt));
2936
2937	if (!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TLB_FLUSH))
2938	{
2939	/*
2940	* We must invalidate the guest TLB entry in either case, we cannot ignore it even for
2941	* the EPT case. See @bugref{6043} and @bugref{6177}.
2942	*
2943	* Set the VMCPU_FF_TLB_FLUSH force flag and flush before VM-entry in hmR0VmxFlushTLB*()
2944	* as this function maybe called in a loop with individual addresses.
2945	*/
2946	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2947	if (pVM->hmr0.s.vmx.fVpid)
2948	{
2949	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
2950	{
2951	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_INDIV_ADDR, GCVirt);
2952	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt);
2953	}
2954	else
2955	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2956	}
2957	else if (pVM->hmr0.s.fNestedPaging)
2958	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2959	}
2960
2961	return VINF_SUCCESS;
2962	}
2963
2964
2965	/**
2966	* Dummy placeholder for tagged-TLB flush handling before VM-entry. Used in the
2967	* case where neither EPT nor VPID is supported by the CPU.
2968	*
2969	* @param pHostCpu The HM physical-CPU structure.
2970	* @param pVCpu The cross context virtual CPU structure.
2971	*
2972	* @remarks Called with interrupts disabled.
2973	*/
2974	static void hmR0VmxFlushTaggedTlbNone(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu)
2975	{
2976	AssertPtr(pVCpu);
2977	AssertPtr(pHostCpu);
2978
2979	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
2980
2981	Assert(pHostCpu->idCpu != NIL_RTCPUID);
2982	pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
2983	pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
2984	pVCpu->hmr0.s.fForceTLBFlush = false;
2985	return;
2986	}
2987
2988
2989	/**
2990	* Flushes the tagged-TLB entries for EPT+VPID CPUs as necessary.
2991	*
2992	* @param pHostCpu The HM physical-CPU structure.
2993	* @param pVCpu The cross context virtual CPU structure.
2994	* @param pVmcsInfo The VMCS info. object.
2995	*
2996	* @remarks All references to "ASID" in this function pertains to "VPID" in Intel's
2997	* nomenclature. The reason is, to avoid confusion in compare statements
2998	* since the host-CPU copies are named "ASID".
2999	*
3000	* @remarks Called with interrupts disabled.
3001	*/
3002	static void hmR0VmxFlushTaggedTlbBoth(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
3003	{
3004	#ifdef VBOX_WITH_STATISTICS
3005	bool fTlbFlushed = false;
3006	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { fTlbFlushed = true; } while (0)
3007	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { \
3008	if (!fTlbFlushed) \
3009	STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); \
3010	} while (0)
3011	#else
3012	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { } while (0)
3013	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { } while (0)
3014	#endif
3015
3016	AssertPtr(pVCpu);
3017	AssertPtr(pHostCpu);
3018	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3019
3020	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3021	AssertMsg(pVM->hmr0.s.fNestedPaging && pVM->hmr0.s.vmx.fVpid,
3022	("hmR0VmxFlushTaggedTlbBoth cannot be invoked unless NestedPaging & VPID are enabled."
3023	"fNestedPaging=%RTbool fVpid=%RTbool", pVM->hmr0.s.fNestedPaging, pVM->hmr0.s.vmx.fVpid));
3024
3025	/*
3026	* Force a TLB flush for the first world-switch if the current CPU differs from the one we
3027	* ran on last. If the TLB flush count changed, another VM (VCPU rather) has hit the ASID
3028	* limit while flushing the TLB or the host CPU is online after a suspend/resume, so we
3029	* cannot reuse the current ASID anymore.
3030	*/
3031	if ( pVCpu->hmr0.s.idLastCpu != pHostCpu->idCpu
3032	\|\| pVCpu->hmr0.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3033	{
3034	++pHostCpu->uCurrentAsid;
3035	if (pHostCpu->uCurrentAsid >= g_uHmMaxAsid)
3036	{
3037	pHostCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0. */
3038	pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
3039	pHostCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
3040	}
3041
3042	pVCpu->hmr0.s.uCurrentAsid = pHostCpu->uCurrentAsid;
3043	pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
3044	pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3045
3046	/*
3047	* Flush by EPT when we get rescheduled to a new host CPU to ensure EPT-only tagged mappings are also
3048	* invalidated. We don't need to flush-by-VPID here as flushing by EPT covers it. See @bugref{6568}.
3049	*/
3050	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hmr0.s.vmx.enmTlbFlushEpt);
3051	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3052	HMVMX_SET_TAGGED_TLB_FLUSHED();
3053	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
3054	}
3055	else if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH)) /* Check for explicit TLB flushes. */
3056	{
3057	/*
3058	* Changes to the EPT paging structure by VMM requires flushing-by-EPT as the CPU
3059	* creates guest-physical (ie. only EPT-tagged) mappings while traversing the EPT
3060	* tables when EPT is in use. Flushing-by-VPID will only flush linear (only
3061	* VPID-tagged) and combined (EPT+VPID tagged) mappings but not guest-physical
3062	* mappings, see @bugref{6568}.
3063	*
3064	* See Intel spec. 28.3.2 "Creating and Using Cached Translation Information".
3065	*/
3066	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hmr0.s.vmx.enmTlbFlushEpt);
3067	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3068	HMVMX_SET_TAGGED_TLB_FLUSHED();
3069	}
3070	else if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3071	{
3072	/*
3073	* The nested-guest specifies its own guest-physical address to use as the APIC-access
3074	* address which requires flushing the TLB of EPT cached structures.
3075	*
3076	* See Intel spec. 28.3.3.4 "Guidelines for Use of the INVEPT Instruction".
3077	*/
3078	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hmr0.s.vmx.enmTlbFlushEpt);
3079	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3080	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3081	HMVMX_SET_TAGGED_TLB_FLUSHED();
3082	}
3083
3084
3085	pVCpu->hmr0.s.fForceTLBFlush = false;
3086	HMVMX_UPDATE_FLUSH_SKIPPED_STAT();
3087
3088	Assert(pVCpu->hmr0.s.idLastCpu == pHostCpu->idCpu);
3089	Assert(pVCpu->hmr0.s.cTlbFlushes == pHostCpu->cTlbFlushes);
3090	AssertMsg(pVCpu->hmr0.s.cTlbFlushes == pHostCpu->cTlbFlushes,
3091	("Flush count mismatch for cpu %d (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hmr0.s.cTlbFlushes, pHostCpu->cTlbFlushes));
3092	AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < g_uHmMaxAsid,
3093	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pHostCpu->idCpu,
3094	pHostCpu->uCurrentAsid, pHostCpu->cTlbFlushes, pVCpu->hmr0.s.idLastCpu, pVCpu->hmr0.s.cTlbFlushes));
3095	AssertMsg(pVCpu->hmr0.s.uCurrentAsid >= 1 && pVCpu->hmr0.s.uCurrentAsid < g_uHmMaxAsid,
3096	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pHostCpu->idCpu, pVCpu->hmr0.s.uCurrentAsid));
3097
3098	/* Update VMCS with the VPID. */
3099	int rc = VMXWriteVmcs16(VMX_VMCS16_VPID, pVCpu->hmr0.s.uCurrentAsid);
3100	AssertRC(rc);
3101
3102	#undef HMVMX_SET_TAGGED_TLB_FLUSHED
3103	}
3104
3105
3106	/**
3107	* Flushes the tagged-TLB entries for EPT CPUs as necessary.
3108	*
3109	* @param pHostCpu The HM physical-CPU structure.
3110	* @param pVCpu The cross context virtual CPU structure.
3111	* @param pVmcsInfo The VMCS info. object.
3112	*
3113	* @remarks Called with interrupts disabled.
3114	*/
3115	static void hmR0VmxFlushTaggedTlbEpt(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
3116	{
3117	AssertPtr(pVCpu);
3118	AssertPtr(pHostCpu);
3119	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3120	AssertMsg(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked without NestedPaging."));
3121	AssertMsg(!pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fVpid, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with VPID."));
3122
3123	/*
3124	* Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
3125	* A change in the TLB flush count implies the host CPU is online after a suspend/resume.
3126	*/
3127	if ( pVCpu->hmr0.s.idLastCpu != pHostCpu->idCpu
3128	\|\| pVCpu->hmr0.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3129	{
3130	pVCpu->hmr0.s.fForceTLBFlush = true;
3131	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3132	}
3133
3134	/* Check for explicit TLB flushes. */
3135	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
3136	{
3137	pVCpu->hmr0.s.fForceTLBFlush = true;
3138	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3139	}
3140
3141	/* Check for TLB flushes while switching to/from a nested-guest. */
3142	if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3143	{
3144	pVCpu->hmr0.s.fForceTLBFlush = true;
3145	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3146	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3147	}
3148
3149	pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
3150	pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3151
3152	if (pVCpu->hmr0.s.fForceTLBFlush)
3153	{
3154	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.enmTlbFlushEpt);
3155	pVCpu->hmr0.s.fForceTLBFlush = false;
3156	}
3157	}
3158
3159
3160	/**
3161	* Flushes the tagged-TLB entries for VPID CPUs as necessary.
3162	*
3163	* @param pHostCpu The HM physical-CPU structure.
3164	* @param pVCpu The cross context virtual CPU structure.
3165	*
3166	* @remarks Called with interrupts disabled.
3167	*/
3168	static void hmR0VmxFlushTaggedTlbVpid(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu)
3169	{
3170	AssertPtr(pVCpu);
3171	AssertPtr(pHostCpu);
3172	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3173	AssertMsg(pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fVpid, ("hmR0VmxFlushTlbVpid cannot be invoked without VPID."));
3174	AssertMsg(!pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging, ("hmR0VmxFlushTlbVpid cannot be invoked with NestedPaging"));
3175
3176	/*
3177	* Force a TLB flush for the first world switch if the current CPU differs from the one we
3178	* ran on last. If the TLB flush count changed, another VM (VCPU rather) has hit the ASID
3179	* limit while flushing the TLB or the host CPU is online after a suspend/resume, so we
3180	* cannot reuse the current ASID anymore.
3181	*/
3182	if ( pVCpu->hmr0.s.idLastCpu != pHostCpu->idCpu
3183	\|\| pVCpu->hmr0.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3184	{
3185	pVCpu->hmr0.s.fForceTLBFlush = true;
3186	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3187	}
3188
3189	/* Check for explicit TLB flushes. */
3190	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
3191	{
3192	/*
3193	* If we ever support VPID flush combinations other than ALL or SINGLE-context (see
3194	* hmR0VmxSetupTaggedTlb()) we would need to explicitly flush in this case (add an
3195	* fExplicitFlush = true here and change the pHostCpu->fFlushAsidBeforeUse check below to
3196	* include fExplicitFlush's too) - an obscure corner case.
3197	*/
3198	pVCpu->hmr0.s.fForceTLBFlush = true;
3199	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3200	}
3201
3202	/* Check for TLB flushes while switching to/from a nested-guest. */
3203	if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3204	{
3205	pVCpu->hmr0.s.fForceTLBFlush = true;
3206	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3207	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3208	}
3209
3210	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3211	pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
3212	if (pVCpu->hmr0.s.fForceTLBFlush)
3213	{
3214	++pHostCpu->uCurrentAsid;
3215	if (pHostCpu->uCurrentAsid >= g_uHmMaxAsid)
3216	{
3217	pHostCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0 */
3218	pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
3219	pHostCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
3220	}
3221
3222	pVCpu->hmr0.s.fForceTLBFlush = false;
3223	pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3224	pVCpu->hmr0.s.uCurrentAsid = pHostCpu->uCurrentAsid;
3225	if (pHostCpu->fFlushAsidBeforeUse)
3226	{
3227	if (pVM->hmr0.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_SINGLE_CONTEXT)
3228	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_SINGLE_CONTEXT, 0 /* GCPtr */);
3229	else if (pVM->hmr0.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_ALL_CONTEXTS)
3230	{
3231	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_ALL_CONTEXTS, 0 /* GCPtr */);
3232	pHostCpu->fFlushAsidBeforeUse = false;
3233	}
3234	else
3235	{
3236	/* hmR0VmxSetupTaggedTlb() ensures we never get here. Paranoia. */
3237	AssertMsgFailed(("Unsupported VPID-flush context type.\n"));
3238	}
3239	}
3240	}
3241
3242	AssertMsg(pVCpu->hmr0.s.cTlbFlushes == pHostCpu->cTlbFlushes,
3243	("Flush count mismatch for cpu %d (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hmr0.s.cTlbFlushes, pHostCpu->cTlbFlushes));
3244	AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < g_uHmMaxAsid,
3245	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pHostCpu->idCpu,
3246	pHostCpu->uCurrentAsid, pHostCpu->cTlbFlushes, pVCpu->hmr0.s.idLastCpu, pVCpu->hmr0.s.cTlbFlushes));
3247	AssertMsg(pVCpu->hmr0.s.uCurrentAsid >= 1 && pVCpu->hmr0.s.uCurrentAsid < g_uHmMaxAsid,
3248	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pHostCpu->idCpu, pVCpu->hmr0.s.uCurrentAsid));
3249
3250	int rc = VMXWriteVmcs16(VMX_VMCS16_VPID, pVCpu->hmr0.s.uCurrentAsid);
3251	AssertRC(rc);
3252	}
3253
3254
3255	/**
3256	* Flushes the guest TLB entry based on CPU capabilities.
3257	*
3258	* @param pHostCpu The HM physical-CPU structure.
3259	* @param pVCpu The cross context virtual CPU structure.
3260	* @param pVmcsInfo The VMCS info. object.
3261	*
3262	* @remarks Called with interrupts disabled.
3263	*/
3264	static void hmR0VmxFlushTaggedTlb(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3265	{
3266	#ifdef HMVMX_ALWAYS_FLUSH_TLB
3267	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
3268	#endif
3269	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3270	switch (pVM->hmr0.s.vmx.enmTlbFlushType)
3271	{
3272	case VMXTLBFLUSHTYPE_EPT_VPID: hmR0VmxFlushTaggedTlbBoth(pHostCpu, pVCpu, pVmcsInfo); break;
3273	case VMXTLBFLUSHTYPE_EPT: hmR0VmxFlushTaggedTlbEpt(pHostCpu, pVCpu, pVmcsInfo); break;
3274	case VMXTLBFLUSHTYPE_VPID: hmR0VmxFlushTaggedTlbVpid(pHostCpu, pVCpu); break;
3275	case VMXTLBFLUSHTYPE_NONE: hmR0VmxFlushTaggedTlbNone(pHostCpu, pVCpu); break;
3276	default:
3277	AssertMsgFailed(("Invalid flush-tag function identifier\n"));
3278	break;
3279	}
3280	/* Don't assert that VMCPU_FF_TLB_FLUSH should no longer be pending. It can be set by other EMTs. */
3281	}
3282
3283
3284	/**
3285	* Sets up the appropriate tagged TLB-flush level and handler for flushing guest
3286	* TLB entries from the host TLB before VM-entry.
3287	*
3288	* @returns VBox status code.
3289	* @param pVM The cross context VM structure.
3290	*/
3291	static int hmR0VmxSetupTaggedTlb(PVMCC pVM)
3292	{
3293	/*
3294	* Determine optimal flush type for nested paging.
3295	* We cannot ignore EPT if no suitable flush-types is supported by the CPU as we've already setup
3296	* unrestricted guest execution (see hmR3InitFinalizeR0()).
3297	*/
3298	if (pVM->hmr0.s.fNestedPaging)
3299	{
3300	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT)
3301	{
3302	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT)
3303	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_SINGLE_CONTEXT;
3304	else if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
3305	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_ALL_CONTEXTS;
3306	else
3307	{
3308	/* Shouldn't happen. EPT is supported but no suitable flush-types supported. */
3309	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3310	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_FLUSH_TYPE_UNSUPPORTED;
3311	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3312	}
3313
3314	/* Make sure the write-back cacheable memory type for EPT is supported. */
3315	if (RT_UNLIKELY(!(g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EMT_WB)))
3316	{
3317	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3318	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_MEM_TYPE_NOT_WB;
3319	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3320	}
3321
3322	/* EPT requires a page-walk length of 4. */
3323	if (RT_UNLIKELY(!(g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_PAGE_WALK_LENGTH_4)))
3324	{
3325	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3326	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_PAGE_WALK_LENGTH_UNSUPPORTED;
3327	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3328	}
3329	}
3330	else
3331	{
3332	/* Shouldn't happen. EPT is supported but INVEPT instruction is not supported. */
3333	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3334	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_INVEPT_UNAVAILABLE;
3335	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3336	}
3337	}
3338
3339	/*
3340	* Determine optimal flush type for VPID.
3341	*/
3342	if (pVM->hmr0.s.vmx.fVpid)
3343	{
3344	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID)
3345	{
3346	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT)
3347	pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_SINGLE_CONTEXT;
3348	else if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS)
3349	pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_ALL_CONTEXTS;
3350	else
3351	{
3352	/* Neither SINGLE nor ALL-context flush types for VPID is supported by the CPU. Ignore VPID capability. */
3353	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
3354	LogRelFunc(("Only INDIV_ADDR supported. Ignoring VPID.\n"));
3355	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
3356	LogRelFunc(("Only SINGLE_CONTEXT_RETAIN_GLOBALS supported. Ignoring VPID.\n"));
3357	pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NOT_SUPPORTED;
3358	pVM->hmr0.s.vmx.fVpid = false;
3359	}
3360	}
3361	else
3362	{
3363	/* Shouldn't happen. VPID is supported but INVVPID is not supported by the CPU. Ignore VPID capability. */
3364	Log4Func(("VPID supported without INVEPT support. Ignoring VPID.\n"));
3365	pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NOT_SUPPORTED;
3366	pVM->hmr0.s.vmx.fVpid = false;
3367	}
3368	}
3369
3370	/*
3371	* Setup the handler for flushing tagged-TLBs.
3372	*/
3373	if (pVM->hmr0.s.fNestedPaging && pVM->hmr0.s.vmx.fVpid)
3374	pVM->hmr0.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_EPT_VPID;
3375	else if (pVM->hmr0.s.fNestedPaging)
3376	pVM->hmr0.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_EPT;
3377	else if (pVM->hmr0.s.vmx.fVpid)
3378	pVM->hmr0.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_VPID;
3379	else
3380	pVM->hmr0.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_NONE;
3381
3382
3383	/*
3384	* Copy out the result to ring-3.
3385	*/
3386	pVM->hm.s.ForR3.vmx.fVpid = pVM->hmr0.s.vmx.fVpid;
3387	pVM->hm.s.ForR3.vmx.enmTlbFlushType = pVM->hmr0.s.vmx.enmTlbFlushType;
3388	pVM->hm.s.ForR3.vmx.enmTlbFlushEpt = pVM->hmr0.s.vmx.enmTlbFlushEpt;
3389	pVM->hm.s.ForR3.vmx.enmTlbFlushVpid = pVM->hmr0.s.vmx.enmTlbFlushVpid;
3390	return VINF_SUCCESS;
3391	}
3392
3393
3394	/**
3395	* Sets up the LBR MSR ranges based on the host CPU.
3396	*
3397	* @returns VBox status code.
3398	* @param pVM The cross context VM structure.
3399	*/
3400	static int hmR0VmxSetupLbrMsrRange(PVMCC pVM)
3401	{
3402	Assert(pVM->hmr0.s.vmx.fLbr);
3403	uint32_t idLbrFromIpMsrFirst;
3404	uint32_t idLbrFromIpMsrLast;
3405	uint32_t idLbrToIpMsrFirst;
3406	uint32_t idLbrToIpMsrLast;
3407	uint32_t idLbrTosMsr;
3408
3409	/*
3410	* Determine the LBR MSRs supported for this host CPU family and model.
3411	*
3412	* See Intel spec. 17.4.8 "LBR Stack".
3413	* See Intel "Model-Specific Registers" spec.
3414	*/
3415	uint32_t const uFamilyModel = (pVM->cpum.ro.HostFeatures.uFamily << 8)
3416	\| pVM->cpum.ro.HostFeatures.uModel;
3417	switch (uFamilyModel)
3418	{
3419	case 0x0f01: case 0x0f02:
3420	idLbrFromIpMsrFirst = MSR_P4_LASTBRANCH_0;
3421	idLbrFromIpMsrLast = MSR_P4_LASTBRANCH_3;
3422	idLbrToIpMsrFirst = 0x0;
3423	idLbrToIpMsrLast = 0x0;
3424	idLbrTosMsr = MSR_P4_LASTBRANCH_TOS;
3425	break;
3426
3427	case 0x065c: case 0x065f: case 0x064e: case 0x065e: case 0x068e:
3428	case 0x069e: case 0x0655: case 0x0666: case 0x067a: case 0x0667:
3429	case 0x066a: case 0x066c: case 0x067d: case 0x067e:
3430	idLbrFromIpMsrFirst = MSR_LASTBRANCH_0_FROM_IP;
3431	idLbrFromIpMsrLast = MSR_LASTBRANCH_31_FROM_IP;
3432	idLbrToIpMsrFirst = MSR_LASTBRANCH_0_TO_IP;
3433	idLbrToIpMsrLast = MSR_LASTBRANCH_31_TO_IP;
3434	idLbrTosMsr = MSR_LASTBRANCH_TOS;
3435	break;
3436
3437	case 0x063d: case 0x0647: case 0x064f: case 0x0656: case 0x063c:
3438	case 0x0645: case 0x0646: case 0x063f: case 0x062a: case 0x062d:
3439	case 0x063a: case 0x063e: case 0x061a: case 0x061e: case 0x061f:
3440	case 0x062e: case 0x0625: case 0x062c: case 0x062f:
3441	idLbrFromIpMsrFirst = MSR_LASTBRANCH_0_FROM_IP;
3442	idLbrFromIpMsrLast = MSR_LASTBRANCH_15_FROM_IP;
3443	idLbrToIpMsrFirst = MSR_LASTBRANCH_0_TO_IP;
3444	idLbrToIpMsrLast = MSR_LASTBRANCH_15_TO_IP;
3445	idLbrTosMsr = MSR_LASTBRANCH_TOS;
3446	break;
3447
3448	case 0x0617: case 0x061d: case 0x060f:
3449	idLbrFromIpMsrFirst = MSR_CORE2_LASTBRANCH_0_FROM_IP;
3450	idLbrFromIpMsrLast = MSR_CORE2_LASTBRANCH_3_FROM_IP;
3451	idLbrToIpMsrFirst = MSR_CORE2_LASTBRANCH_0_TO_IP;
3452	idLbrToIpMsrLast = MSR_CORE2_LASTBRANCH_3_TO_IP;
3453	idLbrTosMsr = MSR_CORE2_LASTBRANCH_TOS;
3454	break;
3455
3456	/* Atom and related microarchitectures we don't care about:
3457	case 0x0637: case 0x064a: case 0x064c: case 0x064d: case 0x065a:
3458	case 0x065d: case 0x061c: case 0x0626: case 0x0627: case 0x0635:
3459	case 0x0636: */
3460	/* All other CPUs: */
3461	default:
3462	{
3463	LogRelFunc(("Could not determine LBR stack size for the CPU model %#x\n", uFamilyModel));
3464	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_LBR_STACK_SIZE_UNKNOWN;
3465	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3466	}
3467	}
3468
3469	/*
3470	* Validate.
3471	*/
3472	uint32_t const cLbrStack = idLbrFromIpMsrLast - idLbrFromIpMsrFirst + 1;
3473	PCVMCPU pVCpu0 = VMCC_GET_CPU_0(pVM);
3474	AssertCompile( RT_ELEMENTS(pVCpu0->hm.s.vmx.VmcsInfo.au64LbrFromIpMsr)
3475	== RT_ELEMENTS(pVCpu0->hm.s.vmx.VmcsInfo.au64LbrToIpMsr));
3476	if (cLbrStack > RT_ELEMENTS(pVCpu0->hm.s.vmx.VmcsInfo.au64LbrFromIpMsr))
3477	{
3478	LogRelFunc(("LBR stack size of the CPU (%u) exceeds our buffer size\n", cLbrStack));
3479	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_LBR_STACK_SIZE_OVERFLOW;
3480	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3481	}
3482	NOREF(pVCpu0);
3483
3484	/*
3485	* Update the LBR info. to the VM struct. for use later.
3486	*/
3487	pVM->hmr0.s.vmx.idLbrTosMsr = idLbrTosMsr;
3488
3489	pVM->hm.s.ForR3.vmx.idLbrFromIpMsrFirst = pVM->hmr0.s.vmx.idLbrFromIpMsrFirst = idLbrFromIpMsrFirst;
3490	pVM->hm.s.ForR3.vmx.idLbrFromIpMsrLast = pVM->hmr0.s.vmx.idLbrFromIpMsrLast = idLbrFromIpMsrLast;
3491
3492	pVM->hm.s.ForR3.vmx.idLbrToIpMsrFirst = pVM->hmr0.s.vmx.idLbrToIpMsrFirst = idLbrToIpMsrFirst;
3493	pVM->hm.s.ForR3.vmx.idLbrToIpMsrLast = pVM->hmr0.s.vmx.idLbrToIpMsrLast = idLbrToIpMsrLast;
3494	return VINF_SUCCESS;
3495	}
3496
3497
3498	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3499	/**
3500	* Sets up the shadow VMCS fields arrays.
3501	*
3502	* This function builds arrays of VMCS fields to sync the shadow VMCS later while
3503	* executing the guest.
3504	*
3505	* @returns VBox status code.
3506	* @param pVM The cross context VM structure.
3507	*/
3508	static int hmR0VmxSetupShadowVmcsFieldsArrays(PVMCC pVM)
3509	{
3510	/*
3511	* Paranoia. Ensure we haven't exposed the VMWRITE-All VMX feature to the guest
3512	* when the host does not support it.
3513	*/
3514	bool const fGstVmwriteAll = pVM->cpum.ro.GuestFeatures.fVmxVmwriteAll;
3515	if ( !fGstVmwriteAll
3516	\|\| (g_HmMsrs.u.vmx.u64Misc & VMX_MISC_VMWRITE_ALL))
3517	{ /* likely. */ }
3518	else
3519	{
3520	LogRelFunc(("VMX VMWRITE-All feature exposed to the guest but host CPU does not support it!\n"));
3521	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_GST_HOST_VMWRITE_ALL;
3522	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3523	}
3524
3525	uint32_t const cVmcsFields = RT_ELEMENTS(g_aVmcsFields);
3526	uint32_t cRwFields = 0;
3527	uint32_t cRoFields = 0;
3528	for (uint32_t i = 0; i < cVmcsFields; i++)
3529	{
3530	VMXVMCSFIELD VmcsField;
3531	VmcsField.u = g_aVmcsFields[i];
3532
3533	/*
3534	* We will be writing "FULL" (64-bit) fields while syncing the shadow VMCS.
3535	* Therefore, "HIGH" (32-bit portion of 64-bit) fields must not be included
3536	* in the shadow VMCS fields array as they would be redundant.
3537	*
3538	* If the VMCS field depends on a CPU feature that is not exposed to the guest,
3539	* we must not include it in the shadow VMCS fields array. Guests attempting to
3540	* VMREAD/VMWRITE such VMCS fields would cause a VM-exit and we shall emulate
3541	* the required behavior.
3542	*/
3543	if ( VmcsField.n.fAccessType == VMX_VMCSFIELD_ACCESS_FULL
3544	&& CPUMIsGuestVmxVmcsFieldValid(pVM, VmcsField.u))
3545	{
3546	/*
3547	* Read-only fields are placed in a separate array so that while syncing shadow
3548	* VMCS fields later (which is more performance critical) we can avoid branches.
3549	*
3550	* However, if the guest can write to all fields (including read-only fields),
3551	* we treat it a as read/write field. Otherwise, writing to these fields would
3552	* cause a VMWRITE instruction error while syncing the shadow VMCS.
3553	*/
3554	if ( fGstVmwriteAll
3555	\|\| !VMXIsVmcsFieldReadOnly(VmcsField.u))
3556	pVM->hmr0.s.vmx.paShadowVmcsFields[cRwFields++] = VmcsField.u;
3557	else
3558	pVM->hmr0.s.vmx.paShadowVmcsRoFields[cRoFields++] = VmcsField.u;
3559	}
3560	}
3561
3562	/* Update the counts. */
3563	pVM->hmr0.s.vmx.cShadowVmcsFields = cRwFields;
3564	pVM->hmr0.s.vmx.cShadowVmcsRoFields = cRoFields;
3565	return VINF_SUCCESS;
3566	}
3567
3568
3569	/**
3570	* Sets up the VMREAD and VMWRITE bitmaps.
3571	*
3572	* @param pVM The cross context VM structure.
3573	*/
3574	static void hmR0VmxSetupVmreadVmwriteBitmaps(PVMCC pVM)
3575	{
3576	/*
3577	* By default, ensure guest attempts to access any VMCS fields cause VM-exits.
3578	*/
3579	uint32_t const cbBitmap = X86_PAGE_4K_SIZE;
3580	uint8_t pbVmreadBitmap = (uint8_t )pVM->hmr0.s.vmx.pvVmreadBitmap;
3581	uint8_t pbVmwriteBitmap = (uint8_t )pVM->hmr0.s.vmx.pvVmwriteBitmap;
3582	ASMMemFill32(pbVmreadBitmap, cbBitmap, UINT32_C(0xffffffff));
3583	ASMMemFill32(pbVmwriteBitmap, cbBitmap, UINT32_C(0xffffffff));
3584
3585	/*
3586	* Skip intercepting VMREAD/VMWRITE to guest read/write fields in the
3587	* VMREAD and VMWRITE bitmaps.
3588	*/
3589	{
3590	uint32_t const *paShadowVmcsFields = pVM->hmr0.s.vmx.paShadowVmcsFields;
3591	uint32_t const cShadowVmcsFields = pVM->hmr0.s.vmx.cShadowVmcsFields;
3592	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
3593	{
3594	uint32_t const uVmcsField = paShadowVmcsFields[i];
3595	Assert(!(uVmcsField & VMX_VMCSFIELD_RSVD_MASK));
3596	Assert(uVmcsField >> 3 < cbBitmap);
3597	ASMBitClear(pbVmreadBitmap + (uVmcsField >> 3), uVmcsField & 7);
3598	ASMBitClear(pbVmwriteBitmap + (uVmcsField >> 3), uVmcsField & 7);
3599	}
3600	}
3601
3602	/*
3603	* Skip intercepting VMREAD for guest read-only fields in the VMREAD bitmap
3604	* if the host supports VMWRITE to all supported VMCS fields.
3605	*/
3606	if (g_HmMsrs.u.vmx.u64Misc & VMX_MISC_VMWRITE_ALL)
3607	{
3608	uint32_t const *paShadowVmcsRoFields = pVM->hmr0.s.vmx.paShadowVmcsRoFields;
3609	uint32_t const cShadowVmcsRoFields = pVM->hmr0.s.vmx.cShadowVmcsRoFields;
3610	for (uint32_t i = 0; i < cShadowVmcsRoFields; i++)
3611	{
3612	uint32_t const uVmcsField = paShadowVmcsRoFields[i];
3613	Assert(!(uVmcsField & VMX_VMCSFIELD_RSVD_MASK));
3614	Assert(uVmcsField >> 3 < cbBitmap);
3615	ASMBitClear(pbVmreadBitmap + (uVmcsField >> 3), uVmcsField & 7);
3616	}
3617	}
3618	}
3619	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
3620
3621
3622	/**
3623	* Sets up the virtual-APIC page address for the VMCS.
3624	*
3625	* @param pVmcsInfo The VMCS info. object.
3626	*/
3627	DECLINLINE(void) hmR0VmxSetupVmcsVirtApicAddr(PCVMXVMCSINFO pVmcsInfo)
3628	{
3629	RTHCPHYS const HCPhysVirtApic = pVmcsInfo->HCPhysVirtApic;
3630	Assert(HCPhysVirtApic != NIL_RTHCPHYS);
3631	Assert(!(HCPhysVirtApic & 0xfff)); /* Bits 11:0 MBZ. */
3632	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL, HCPhysVirtApic);
3633	AssertRC(rc);
3634	}
3635
3636
3637	/**
3638	* Sets up the MSR-bitmap address for the VMCS.
3639	*
3640	* @param pVmcsInfo The VMCS info. object.
3641	*/
3642	DECLINLINE(void) hmR0VmxSetupVmcsMsrBitmapAddr(PCVMXVMCSINFO pVmcsInfo)
3643	{
3644	RTHCPHYS const HCPhysMsrBitmap = pVmcsInfo->HCPhysMsrBitmap;
3645	Assert(HCPhysMsrBitmap != NIL_RTHCPHYS);
3646	Assert(!(HCPhysMsrBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3647	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_MSR_BITMAP_FULL, HCPhysMsrBitmap);
3648	AssertRC(rc);
3649	}
3650
3651
3652	/**
3653	* Sets up the APIC-access page address for the VMCS.
3654	*
3655	* @param pVCpu The cross context virtual CPU structure.
3656	*/
3657	DECLINLINE(void) hmR0VmxSetupVmcsApicAccessAddr(PVMCPUCC pVCpu)
3658	{
3659	RTHCPHYS const HCPhysApicAccess = pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.HCPhysApicAccess;
3660	Assert(HCPhysApicAccess != NIL_RTHCPHYS);
3661	Assert(!(HCPhysApicAccess & 0xfff)); /* Bits 11:0 MBZ. */
3662	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, HCPhysApicAccess);
3663	AssertRC(rc);
3664	}
3665
3666	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3667
3668	/**
3669	* Sets up the VMREAD bitmap address for the VMCS.
3670	*
3671	* @param pVCpu The cross context virtual CPU structure.
3672	*/
3673	DECLINLINE(void) hmR0VmxSetupVmcsVmreadBitmapAddr(PVMCPUCC pVCpu)
3674	{
3675	RTHCPHYS const HCPhysVmreadBitmap = pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.HCPhysVmreadBitmap;
3676	Assert(HCPhysVmreadBitmap != NIL_RTHCPHYS);
3677	Assert(!(HCPhysVmreadBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3678	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VMREAD_BITMAP_FULL, HCPhysVmreadBitmap);
3679	AssertRC(rc);
3680	}
3681
3682
3683	/**
3684	* Sets up the VMWRITE bitmap address for the VMCS.
3685	*
3686	* @param pVCpu The cross context virtual CPU structure.
3687	*/
3688	DECLINLINE(void) hmR0VmxSetupVmcsVmwriteBitmapAddr(PVMCPUCC pVCpu)
3689	{
3690	RTHCPHYS const HCPhysVmwriteBitmap = pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.HCPhysVmwriteBitmap;
3691	Assert(HCPhysVmwriteBitmap != NIL_RTHCPHYS);
3692	Assert(!(HCPhysVmwriteBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3693	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VMWRITE_BITMAP_FULL, HCPhysVmwriteBitmap);
3694	AssertRC(rc);
3695	}
3696
3697	#endif
3698
3699	/**
3700	* Sets up the VM-entry MSR load, VM-exit MSR-store and VM-exit MSR-load addresses
3701	* in the VMCS.
3702	*
3703	* @returns VBox status code.
3704	* @param pVmcsInfo The VMCS info. object.
3705	*/
3706	DECLINLINE(int) hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(PVMXVMCSINFO pVmcsInfo)
3707	{
3708	RTHCPHYS const HCPhysGuestMsrLoad = pVmcsInfo->HCPhysGuestMsrLoad;
3709	Assert(HCPhysGuestMsrLoad != NIL_RTHCPHYS);
3710	Assert(!(HCPhysGuestMsrLoad & 0xf)); /* Bits 3:0 MBZ. */
3711
3712	RTHCPHYS const HCPhysGuestMsrStore = pVmcsInfo->HCPhysGuestMsrStore;
3713	Assert(HCPhysGuestMsrStore != NIL_RTHCPHYS);
3714	Assert(!(HCPhysGuestMsrStore & 0xf)); /* Bits 3:0 MBZ. */
3715
3716	RTHCPHYS const HCPhysHostMsrLoad = pVmcsInfo->HCPhysHostMsrLoad;
3717	Assert(HCPhysHostMsrLoad != NIL_RTHCPHYS);
3718	Assert(!(HCPhysHostMsrLoad & 0xf)); /* Bits 3:0 MBZ. */
3719
3720	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL, HCPhysGuestMsrLoad); AssertRC(rc);
3721	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL, HCPhysGuestMsrStore); AssertRC(rc);
3722	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL, HCPhysHostMsrLoad); AssertRC(rc);
3723	return VINF_SUCCESS;
3724	}
3725
3726
3727	/**
3728	* Sets up MSR permissions in the MSR bitmap of a VMCS info. object.
3729	*
3730	* @param pVCpu The cross context virtual CPU structure.
3731	* @param pVmcsInfo The VMCS info. object.
3732	*/
3733	static void hmR0VmxSetupVmcsMsrPermissions(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3734	{
3735	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS);
3736
3737	/*
3738	* By default, ensure guest attempts to access any MSR cause VM-exits.
3739	* This shall later be relaxed for specific MSRs as necessary.
3740	*
3741	* Note: For nested-guests, the entire bitmap will be merged prior to
3742	* executing the nested-guest using hardware-assisted VMX and hence there
3743	* is no need to perform this operation. See hmR0VmxMergeMsrBitmapNested.
3744	*/
3745	Assert(pVmcsInfo->pvMsrBitmap);
3746	ASMMemFill32(pVmcsInfo->pvMsrBitmap, X86_PAGE_4K_SIZE, UINT32_C(0xffffffff));
3747
3748	/*
3749	* The guest can access the following MSRs (read, write) without causing
3750	* VM-exits; they are loaded/stored automatically using fields in the VMCS.
3751	*/
3752	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3753	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_CS, VMXMSRPM_ALLOW_RD_WR);
3754	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_ESP, VMXMSRPM_ALLOW_RD_WR);
3755	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_EIP, VMXMSRPM_ALLOW_RD_WR);
3756	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_GS_BASE, VMXMSRPM_ALLOW_RD_WR);
3757	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_FS_BASE, VMXMSRPM_ALLOW_RD_WR);
3758
3759	/*
3760	* The IA32_PRED_CMD and IA32_FLUSH_CMD MSRs are write-only and has no state
3761	* associated with then. We never need to intercept access (writes need to be
3762	* executed without causing a VM-exit, reads will #GP fault anyway).
3763	*
3764	* The IA32_SPEC_CTRL MSR is read/write and has state. We allow the guest to
3765	* read/write them. We swap the guest/host MSR value using the
3766	* auto-load/store MSR area.
3767	*/
3768	if (pVM->cpum.ro.GuestFeatures.fIbpb)
3769	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_PRED_CMD, VMXMSRPM_ALLOW_RD_WR);
3770	if (pVM->cpum.ro.GuestFeatures.fFlushCmd)
3771	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_FLUSH_CMD, VMXMSRPM_ALLOW_RD_WR);
3772	if (pVM->cpum.ro.GuestFeatures.fIbrs)
3773	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SPEC_CTRL, VMXMSRPM_ALLOW_RD_WR);
3774
3775	/*
3776	* Allow full read/write access for the following MSRs (mandatory for VT-x)
3777	* required for 64-bit guests.
3778	*/
3779	if (pVM->hmr0.s.fAllow64BitGuests)
3780	{
3781	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_LSTAR, VMXMSRPM_ALLOW_RD_WR);
3782	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K6_STAR, VMXMSRPM_ALLOW_RD_WR);
3783	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_SF_MASK, VMXMSRPM_ALLOW_RD_WR);
3784	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_KERNEL_GS_BASE, VMXMSRPM_ALLOW_RD_WR);
3785	}
3786
3787	/*
3788	* IA32_EFER MSR is always intercepted, see @bugref{9180#c37}.
3789	*/
3790	#ifdef VBOX_STRICT
3791	Assert(pVmcsInfo->pvMsrBitmap);
3792	uint32_t const fMsrpmEfer = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, MSR_K6_EFER);
3793	Assert(fMsrpmEfer == VMXMSRPM_EXIT_RD_WR);
3794	#endif
3795	}
3796
3797
3798	/**
3799	* Sets up pin-based VM-execution controls in the VMCS.
3800	*
3801	* @returns VBox status code.
3802	* @param pVCpu The cross context virtual CPU structure.
3803	* @param pVmcsInfo The VMCS info. object.
3804	*/
3805	static int hmR0VmxSetupVmcsPinCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3806	{
3807	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3808	uint32_t fVal = g_HmMsrs.u.vmx.PinCtls.n.allowed0; /* Bits set here must always be set. */
3809	uint32_t const fZap = g_HmMsrs.u.vmx.PinCtls.n.allowed1; /* Bits cleared here must always be cleared. */
3810
3811	fVal \|= VMX_PIN_CTLS_EXT_INT_EXIT /* External interrupts cause a VM-exit. */
3812	\| VMX_PIN_CTLS_NMI_EXIT; /* Non-maskable interrupts (NMIs) cause a VM-exit. */
3813
3814	if (g_HmMsrs.u.vmx.PinCtls.n.allowed1 & VMX_PIN_CTLS_VIRT_NMI)
3815	fVal \|= VMX_PIN_CTLS_VIRT_NMI; /* Use virtual NMIs and virtual-NMI blocking features. */
3816
3817	/* Enable the VMX-preemption timer. */
3818	if (pVM->hmr0.s.vmx.fUsePreemptTimer)
3819	{
3820	Assert(g_HmMsrs.u.vmx.PinCtls.n.allowed1 & VMX_PIN_CTLS_PREEMPT_TIMER);
3821	fVal \|= VMX_PIN_CTLS_PREEMPT_TIMER;
3822	}
3823
3824	#if 0
3825	/* Enable posted-interrupt processing. */
3826	if (pVM->hm.s.fPostedIntrs)
3827	{
3828	Assert(g_HmMsrs.u.vmx.PinCtls.n.allowed1 & VMX_PIN_CTLS_POSTED_INT);
3829	Assert(g_HmMsrs.u.vmx.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_ACK_EXT_INT);
3830	fVal \|= VMX_PIN_CTLS_POSTED_INT;
3831	}
3832	#endif
3833
3834	if ((fVal & fZap) != fVal)
3835	{
3836	LogRelFunc(("Invalid pin-based VM-execution controls combo! Cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3837	g_HmMsrs.u.vmx.PinCtls.n.allowed0, fVal, fZap));
3838	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PIN_EXEC;
3839	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3840	}
3841
3842	/* Commit it to the VMCS and update our cache. */
3843	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, fVal);
3844	AssertRC(rc);
3845	pVmcsInfo->u32PinCtls = fVal;
3846
3847	return VINF_SUCCESS;
3848	}
3849
3850
3851	/**
3852	* Sets up secondary processor-based VM-execution controls in the VMCS.
3853	*
3854	* @returns VBox status code.
3855	* @param pVCpu The cross context virtual CPU structure.
3856	* @param pVmcsInfo The VMCS info. object.
3857	*/
3858	static int hmR0VmxSetupVmcsProcCtls2(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3859	{
3860	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3861	uint32_t fVal = g_HmMsrs.u.vmx.ProcCtls2.n.allowed0; /* Bits set here must be set in the VMCS. */
3862	uint32_t const fZap = g_HmMsrs.u.vmx.ProcCtls2.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3863
3864	/* WBINVD causes a VM-exit. */
3865	if (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_WBINVD_EXIT)
3866	fVal \|= VMX_PROC_CTLS2_WBINVD_EXIT;
3867
3868	/* Enable EPT (aka nested-paging). */
3869	if (pVM->hmr0.s.fNestedPaging)
3870	fVal \|= VMX_PROC_CTLS2_EPT;
3871
3872	/* Enable the INVPCID instruction if we expose it to the guest and is supported
3873	by the hardware. Without this, guest executing INVPCID would cause a #UD. */
3874	if ( pVM->cpum.ro.GuestFeatures.fInvpcid
3875	&& (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_INVPCID))
3876	fVal \|= VMX_PROC_CTLS2_INVPCID;
3877
3878	/* Enable VPID. */
3879	if (pVM->hmr0.s.vmx.fVpid)
3880	fVal \|= VMX_PROC_CTLS2_VPID;
3881
3882	/* Enable unrestricted guest execution. */
3883	if (pVM->hmr0.s.vmx.fUnrestrictedGuest)
3884	fVal \|= VMX_PROC_CTLS2_UNRESTRICTED_GUEST;
3885
3886	#if 0
3887	if (pVM->hm.s.fVirtApicRegs)
3888	{
3889	/* Enable APIC-register virtualization. */
3890	Assert(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_APIC_REG_VIRT);
3891	fVal \|= VMX_PROC_CTLS2_APIC_REG_VIRT;
3892
3893	/* Enable virtual-interrupt delivery. */
3894	Assert(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_INTR_DELIVERY);
3895	fVal \|= VMX_PROC_CTLS2_VIRT_INTR_DELIVERY;
3896	}
3897	#endif
3898
3899	/* Virtualize-APIC accesses if supported by the CPU. The virtual-APIC page is
3900	where the TPR shadow resides. */
3901	/** @todo VIRT_X2APIC support, it's mutually exclusive with this. So must be
3902	* done dynamically. */
3903	if (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
3904	{
3905	fVal \|= VMX_PROC_CTLS2_VIRT_APIC_ACCESS;
3906	hmR0VmxSetupVmcsApicAccessAddr(pVCpu);
3907	}
3908
3909	/* Enable the RDTSCP instruction if we expose it to the guest and is supported
3910	by the hardware. Without this, guest executing RDTSCP would cause a #UD. */
3911	if ( pVM->cpum.ro.GuestFeatures.fRdTscP
3912	&& (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_RDTSCP))
3913	fVal \|= VMX_PROC_CTLS2_RDTSCP;
3914
3915	/* Enable Pause-Loop exiting. */
3916	if ( (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT)
3917	&& pVM->hm.s.vmx.cPleGapTicks
3918	&& pVM->hm.s.vmx.cPleWindowTicks)
3919	{
3920	fVal \|= VMX_PROC_CTLS2_PAUSE_LOOP_EXIT;
3921
3922	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, pVM->hm.s.vmx.cPleGapTicks); AssertRC(rc);
3923	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, pVM->hm.s.vmx.cPleWindowTicks); AssertRC(rc);
3924	}
3925
3926	if ((fVal & fZap) != fVal)
3927	{
3928	LogRelFunc(("Invalid secondary processor-based VM-execution controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3929	g_HmMsrs.u.vmx.ProcCtls2.n.allowed0, fVal, fZap));
3930	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC2;
3931	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3932	}
3933
3934	/* Commit it to the VMCS and update our cache. */
3935	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, fVal);
3936	AssertRC(rc);
3937	pVmcsInfo->u32ProcCtls2 = fVal;
3938
3939	return VINF_SUCCESS;
3940	}
3941
3942
3943	/**
3944	* Sets up processor-based VM-execution controls in the VMCS.
3945	*
3946	* @returns VBox status code.
3947	* @param pVCpu The cross context virtual CPU structure.
3948	* @param pVmcsInfo The VMCS info. object.
3949	*/
3950	static int hmR0VmxSetupVmcsProcCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3951	{
3952	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3953	uint32_t fVal = g_HmMsrs.u.vmx.ProcCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
3954	uint32_t const fZap = g_HmMsrs.u.vmx.ProcCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3955
3956	fVal \|= VMX_PROC_CTLS_HLT_EXIT /* HLT causes a VM-exit. */
3957	\| VMX_PROC_CTLS_USE_TSC_OFFSETTING /* Use TSC-offsetting. */
3958	\| VMX_PROC_CTLS_MOV_DR_EXIT /* MOV DRx causes a VM-exit. */
3959	\| VMX_PROC_CTLS_UNCOND_IO_EXIT /* All IO instructions cause a VM-exit. */
3960	\| VMX_PROC_CTLS_RDPMC_EXIT /* RDPMC causes a VM-exit. */
3961	\| VMX_PROC_CTLS_MONITOR_EXIT /* MONITOR causes a VM-exit. */
3962	\| VMX_PROC_CTLS_MWAIT_EXIT; /* MWAIT causes a VM-exit. */
3963
3964	/* We toggle VMX_PROC_CTLS_MOV_DR_EXIT later, check if it's not -always- needed to be set or clear. */
3965	if ( !(g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_MOV_DR_EXIT)
3966	\|\| (g_HmMsrs.u.vmx.ProcCtls.n.allowed0 & VMX_PROC_CTLS_MOV_DR_EXIT))
3967	{
3968	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_MOV_DRX_EXIT;
3969	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3970	}
3971
3972	/* Without nested paging, INVLPG (also affects INVPCID) and MOV CR3 instructions should cause VM-exits. */
3973	if (!pVM->hmr0.s.fNestedPaging)
3974	{
3975	Assert(!pVM->hmr0.s.vmx.fUnrestrictedGuest);
3976	fVal \|= VMX_PROC_CTLS_INVLPG_EXIT
3977	\| VMX_PROC_CTLS_CR3_LOAD_EXIT
3978	\| VMX_PROC_CTLS_CR3_STORE_EXIT;
3979	}
3980
3981	/* Use TPR shadowing if supported by the CPU. */
3982	if ( PDMHasApic(pVM)
3983	&& (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW))
3984	{
3985	fVal \|= VMX_PROC_CTLS_USE_TPR_SHADOW; /* CR8 reads from the Virtual-APIC page. */
3986	/* CR8 writes cause a VM-exit based on TPR threshold. */
3987	Assert(!(fVal & VMX_PROC_CTLS_CR8_STORE_EXIT));
3988	Assert(!(fVal & VMX_PROC_CTLS_CR8_LOAD_EXIT));
3989	hmR0VmxSetupVmcsVirtApicAddr(pVmcsInfo);
3990	}
3991	else
3992	{
3993	/* Some 32-bit CPUs do not support CR8 load/store exiting as MOV CR8 is
3994	invalid on 32-bit Intel CPUs. Set this control only for 64-bit guests. */
3995	if (pVM->hmr0.s.fAllow64BitGuests)
3996	fVal \|= VMX_PROC_CTLS_CR8_STORE_EXIT /* CR8 reads cause a VM-exit. */
3997	\| VMX_PROC_CTLS_CR8_LOAD_EXIT; /* CR8 writes cause a VM-exit. */
3998	}
3999
4000	/* Use MSR-bitmaps if supported by the CPU. */
4001	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
4002	{
4003	fVal \|= VMX_PROC_CTLS_USE_MSR_BITMAPS;
4004	hmR0VmxSetupVmcsMsrBitmapAddr(pVmcsInfo);
4005	}
4006
4007	/* Use the secondary processor-based VM-execution controls if supported by the CPU. */
4008	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
4009	fVal \|= VMX_PROC_CTLS_USE_SECONDARY_CTLS;
4010
4011	if ((fVal & fZap) != fVal)
4012	{
4013	LogRelFunc(("Invalid processor-based VM-execution controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
4014	g_HmMsrs.u.vmx.ProcCtls.n.allowed0, fVal, fZap));
4015	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC;
4016	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
4017	}
4018
4019	/* Commit it to the VMCS and update our cache. */
4020	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, fVal);
4021	AssertRC(rc);
4022	pVmcsInfo->u32ProcCtls = fVal;
4023
4024	/* Set up MSR permissions that don't change through the lifetime of the VM. */
4025	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
4026	hmR0VmxSetupVmcsMsrPermissions(pVCpu, pVmcsInfo);
4027
4028	/* Set up secondary processor-based VM-execution controls if the CPU supports it. */
4029	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
4030	return hmR0VmxSetupVmcsProcCtls2(pVCpu, pVmcsInfo);
4031
4032	/* Sanity check, should not really happen. */
4033	if (RT_LIKELY(!pVM->hmr0.s.vmx.fUnrestrictedGuest))
4034	{ /* likely */ }
4035	else
4036	{
4037	pVCpu->hm.s.u32HMError = VMX_UFC_INVALID_UX_COMBO;
4038	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
4039	}
4040
4041	/* Old CPUs without secondary processor-based VM-execution controls would end up here. */
4042	return VINF_SUCCESS;
4043	}
4044
4045
4046	/**
4047	* Sets up miscellaneous (everything other than Pin, Processor and secondary
4048	* Processor-based VM-execution) control fields in the VMCS.
4049	*
4050	* @returns VBox status code.
4051	* @param pVCpu The cross context virtual CPU structure.
4052	* @param pVmcsInfo The VMCS info. object.
4053	*/
4054	static int hmR0VmxSetupVmcsMiscCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
4055	{
4056	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4057	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUseVmcsShadowing)
4058	{
4059	hmR0VmxSetupVmcsVmreadBitmapAddr(pVCpu);
4060	hmR0VmxSetupVmcsVmwriteBitmapAddr(pVCpu);
4061	}
4062	#endif
4063
4064	Assert(pVmcsInfo->u64VmcsLinkPtr == NIL_RTHCPHYS);
4065	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
4066	AssertRC(rc);
4067
4068	rc = hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(pVmcsInfo);
4069	if (RT_SUCCESS(rc))
4070	{
4071	uint64_t const u64Cr0Mask = hmR0VmxGetFixedCr0Mask(pVCpu);
4072	uint64_t const u64Cr4Mask = hmR0VmxGetFixedCr4Mask(pVCpu);
4073
4074	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, u64Cr0Mask); AssertRC(rc);
4075	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, u64Cr4Mask); AssertRC(rc);
4076
4077	pVmcsInfo->u64Cr0Mask = u64Cr0Mask;
4078	pVmcsInfo->u64Cr4Mask = u64Cr4Mask;
4079
4080	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fLbr)
4081	{
4082	rc = VMXWriteVmcsNw(VMX_VMCS64_GUEST_DEBUGCTL_FULL, MSR_IA32_DEBUGCTL_LBR);
4083	AssertRC(rc);
4084	}
4085	return VINF_SUCCESS;
4086	}
4087	else
4088	LogRelFunc(("Failed to initialize VMCS auto-load/store MSR addresses. rc=%Rrc\n", rc));
4089	return rc;
4090	}
4091
4092
4093	/**
4094	* Sets up the initial exception bitmap in the VMCS based on static conditions.
4095	*
4096	* We shall setup those exception intercepts that don't change during the
4097	* lifetime of the VM here. The rest are done dynamically while loading the
4098	* guest state.
4099	*
4100	* @param pVCpu The cross context virtual CPU structure.
4101	* @param pVmcsInfo The VMCS info. object.
4102	*/
4103	static void hmR0VmxSetupVmcsXcptBitmap(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
4104	{
4105	/*
4106	* The following exceptions are always intercepted:
4107	*
4108	* #AC - To prevent the guest from hanging the CPU and for dealing with
4109	* split-lock detecting host configs.
4110	* #DB - To maintain the DR6 state even when intercepting DRx reads/writes and
4111	* recursive #DBs can cause a CPU hang.
4112	* #PF - To sync our shadow page tables when nested-paging is not used.
4113	*/
4114	bool const fNestedPaging = pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging;
4115	uint32_t const uXcptBitmap = RT_BIT(X86_XCPT_AC)
4116	\| RT_BIT(X86_XCPT_DB)
4117	\| (fNestedPaging ? 0 : RT_BIT(X86_XCPT_PF));
4118
4119	/* Commit it to the VMCS. */
4120	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
4121	AssertRC(rc);
4122
4123	/* Update our cache of the exception bitmap. */
4124	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
4125	}
4126
4127
4128	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4129	/**
4130	* Sets up the VMCS for executing a nested-guest using hardware-assisted VMX.
4131	*
4132	* @returns VBox status code.
4133	* @param pVmcsInfo The VMCS info. object.
4134	*/
4135	static int hmR0VmxSetupVmcsCtlsNested(PVMXVMCSINFO pVmcsInfo)
4136	{
4137	Assert(pVmcsInfo->u64VmcsLinkPtr == NIL_RTHCPHYS);
4138	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
4139	AssertRC(rc);
4140
4141	rc = hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(pVmcsInfo);
4142	if (RT_SUCCESS(rc))
4143	{
4144	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
4145	hmR0VmxSetupVmcsMsrBitmapAddr(pVmcsInfo);
4146
4147	/* Paranoia - We've not yet initialized these, they shall be done while merging the VMCS. */
4148	Assert(!pVmcsInfo->u64Cr0Mask);
4149	Assert(!pVmcsInfo->u64Cr4Mask);
4150	return VINF_SUCCESS;
4151	}
4152	LogRelFunc(("Failed to set up the VMCS link pointer in the nested-guest VMCS. rc=%Rrc\n", rc));
4153	return rc;
4154	}
4155	#endif
4156
4157
4158	/**
4159	* Sets pfnStartVm to the best suited variant.
4160	*
4161	* This must be called whenever anything changes relative to the hmR0VmXStartVm
4162	* variant selection:
4163	* - pVCpu->hm.s.fLoadSaveGuestXcr0
4164	* - HM_WSF_IBPB_ENTRY in pVCpu->hmr0.s.fWorldSwitcher
4165	* - HM_WSF_IBPB_EXIT in pVCpu->hmr0.s.fWorldSwitcher
4166	* - Perhaps: CPUMIsGuestFPUStateActive() (windows only)
4167	* - Perhaps: CPUMCTX.fXStateMask (windows only)
4168	*
4169	* We currently ASSUME that neither HM_WSF_IBPB_ENTRY nor HM_WSF_IBPB_EXIT
4170	* cannot be changed at runtime.
4171	*/
4172	static void hmR0VmxUpdateStartVmFunction(PVMCPUCC pVCpu)
4173	{
4174	static const struct CLANGWORKAROUND { PFNHMVMXSTARTVM pfn; } s_aHmR0VmxStartVmFunctions[] =
4175	{
4176	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_SansL1dEntry_SansMdsEntry_SansIbpbExit },
4177	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_SansL1dEntry_SansMdsEntry_SansIbpbExit },
4178	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_SansL1dEntry_SansMdsEntry_SansIbpbExit },
4179	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_SansL1dEntry_SansMdsEntry_SansIbpbExit },
4180	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_WithL1dEntry_SansMdsEntry_SansIbpbExit },
4181	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_WithL1dEntry_SansMdsEntry_SansIbpbExit },
4182	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_WithL1dEntry_SansMdsEntry_SansIbpbExit },
4183	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_WithL1dEntry_SansMdsEntry_SansIbpbExit },
4184	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_SansL1dEntry_WithMdsEntry_SansIbpbExit },
4185	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_SansL1dEntry_WithMdsEntry_SansIbpbExit },
4186	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_SansL1dEntry_WithMdsEntry_SansIbpbExit },
4187	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_SansL1dEntry_WithMdsEntry_SansIbpbExit },
4188	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_WithL1dEntry_WithMdsEntry_SansIbpbExit },
4189	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_WithL1dEntry_WithMdsEntry_SansIbpbExit },
4190	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_WithL1dEntry_WithMdsEntry_SansIbpbExit },
4191	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_WithL1dEntry_WithMdsEntry_SansIbpbExit },
4192	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_SansL1dEntry_SansMdsEntry_WithIbpbExit },
4193	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_SansL1dEntry_SansMdsEntry_WithIbpbExit },
4194	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_SansL1dEntry_SansMdsEntry_WithIbpbExit },
4195	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_SansL1dEntry_SansMdsEntry_WithIbpbExit },
4196	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_WithL1dEntry_SansMdsEntry_WithIbpbExit },
4197	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_WithL1dEntry_SansMdsEntry_WithIbpbExit },
4198	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_WithL1dEntry_SansMdsEntry_WithIbpbExit },
4199	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_WithL1dEntry_SansMdsEntry_WithIbpbExit },
4200	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_SansL1dEntry_WithMdsEntry_WithIbpbExit },
4201	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_SansL1dEntry_WithMdsEntry_WithIbpbExit },
4202	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_SansL1dEntry_WithMdsEntry_WithIbpbExit },
4203	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_SansL1dEntry_WithMdsEntry_WithIbpbExit },
4204	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_WithL1dEntry_WithMdsEntry_WithIbpbExit },
4205	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_WithL1dEntry_WithMdsEntry_WithIbpbExit },
4206	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_WithL1dEntry_WithMdsEntry_WithIbpbExit },
4207	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_WithL1dEntry_WithMdsEntry_WithIbpbExit },
4208	};
4209	uintptr_t const idx = (pVCpu->hmr0.s.fLoadSaveGuestXcr0 ? 1 : 0)
4210	\| (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_IBPB_ENTRY ? 2 : 0)
4211	\| (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_L1D_ENTRY ? 4 : 0)
4212	\| (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_MDS_ENTRY ? 8 : 0)
4213	\| (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_IBPB_EXIT ? 16 : 0);
4214	PFNHMVMXSTARTVM const pfnStartVm = s_aHmR0VmxStartVmFunctions[idx].pfn;
4215	if (pVCpu->hmr0.s.vmx.pfnStartVm != pfnStartVm)
4216	pVCpu->hmr0.s.vmx.pfnStartVm = pfnStartVm;
4217	}
4218
4219
4220	/**
4221	* Selector FNHMSVMVMRUN implementation.
4222	*/
4223	static DECLCALLBACK(int) hmR0VmxStartVmSelector(PVMXVMCSINFO pVmcsInfo, PVMCPUCC pVCpu, bool fResume)
4224	{
4225	hmR0VmxUpdateStartVmFunction(pVCpu);
4226	return pVCpu->hmr0.s.vmx.pfnStartVm(pVmcsInfo, pVCpu, fResume);
4227	}
4228
4229
4230	/**
4231	* Sets up the VMCS for executing a guest (or nested-guest) using hardware-assisted
4232	* VMX.
4233	*
4234	* @returns VBox status code.
4235	* @param pVCpu The cross context virtual CPU structure.
4236	* @param pVmcsInfo The VMCS info. object.
4237	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
4238	*/
4239	static int hmR0VmxSetupVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
4240	{
4241	Assert(pVmcsInfo->pvVmcs);
4242	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4243
4244	/* Set the CPU specified revision identifier at the beginning of the VMCS structure. */
4245	(uint32_t )pVmcsInfo->pvVmcs = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_ID);
4246	const char * const pszVmcs = fIsNstGstVmcs ? "nested-guest VMCS" : "guest VMCS";
4247
4248	LogFlowFunc(("\n"));
4249
4250	/*
4251	* Initialize the VMCS using VMCLEAR before loading the VMCS.
4252	* See Intel spec. 31.6 "Preparation And Launching A Virtual Machine".
4253	*/
4254	int rc = hmR0VmxClearVmcs(pVmcsInfo);
4255	if (RT_SUCCESS(rc))
4256	{
4257	rc = hmR0VmxLoadVmcs(pVmcsInfo);
4258	if (RT_SUCCESS(rc))
4259	{
4260	/*
4261	* Initialize the hardware-assisted VMX execution handler for guest and nested-guest VMCS.
4262	* The host is always 64-bit since we no longer support 32-bit hosts.
4263	* Currently we have just a single handler for all guest modes as well, see @bugref{6208#c73}.
4264	*/
4265	if (!fIsNstGstVmcs)
4266	{
4267	rc = hmR0VmxSetupVmcsPinCtls(pVCpu, pVmcsInfo);
4268	if (RT_SUCCESS(rc))
4269	{
4270	rc = hmR0VmxSetupVmcsProcCtls(pVCpu, pVmcsInfo);
4271	if (RT_SUCCESS(rc))
4272	{
4273	rc = hmR0VmxSetupVmcsMiscCtls(pVCpu, pVmcsInfo);
4274	if (RT_SUCCESS(rc))
4275	{
4276	hmR0VmxSetupVmcsXcptBitmap(pVCpu, pVmcsInfo);
4277	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4278	/*
4279	* If a shadow VMCS is allocated for the VMCS info. object, initialize the
4280	* VMCS revision ID and shadow VMCS indicator bit. Also, clear the VMCS
4281	* making it fit for use when VMCS shadowing is later enabled.
4282	*/
4283	if (pVmcsInfo->pvShadowVmcs)
4284	{
4285	VMXVMCSREVID VmcsRevId;
4286	VmcsRevId.u = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_ID);
4287	VmcsRevId.n.fIsShadowVmcs = 1;
4288	(uint32_t )pVmcsInfo->pvShadowVmcs = VmcsRevId.u;
4289	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
4290	if (RT_SUCCESS(rc))
4291	{ /* likely */ }
4292	else
4293	LogRelFunc(("Failed to initialize shadow VMCS. rc=%Rrc\n", rc));
4294	}
4295	#endif
4296	}
4297	else
4298	LogRelFunc(("Failed to setup miscellaneous controls. rc=%Rrc\n", rc));
4299	}
4300	else
4301	LogRelFunc(("Failed to setup processor-based VM-execution controls. rc=%Rrc\n", rc));
4302	}
4303	else
4304	LogRelFunc(("Failed to setup pin-based controls. rc=%Rrc\n", rc));
4305	}
4306	else
4307	{
4308	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4309	rc = hmR0VmxSetupVmcsCtlsNested(pVmcsInfo);
4310	if (RT_SUCCESS(rc))
4311	{ /* likely */ }
4312	else
4313	LogRelFunc(("Failed to initialize nested-guest VMCS. rc=%Rrc\n", rc));
4314	#else
4315	AssertFailed();
4316	#endif
4317	}
4318	}
4319	else
4320	LogRelFunc(("Failed to load the %s. rc=%Rrc\n", rc, pszVmcs));
4321	}
4322	else
4323	LogRelFunc(("Failed to clear the %s. rc=%Rrc\n", rc, pszVmcs));
4324
4325	/* Sync any CPU internal VMCS data back into our VMCS in memory. */
4326	if (RT_SUCCESS(rc))
4327	{
4328	rc = hmR0VmxClearVmcs(pVmcsInfo);
4329	if (RT_SUCCESS(rc))
4330	{ /* likely */ }
4331	else
4332	LogRelFunc(("Failed to clear the %s post setup. rc=%Rrc\n", rc, pszVmcs));
4333	}
4334
4335	/*
4336	* Update the last-error record both for failures and success, so we
4337	* can propagate the status code back to ring-3 for diagnostics.
4338	*/
4339	hmR0VmxUpdateErrorRecord(pVCpu, rc);
4340	NOREF(pszVmcs);
4341	return rc;
4342	}
4343
4344
4345	/**
4346	* Does global VT-x initialization (called during module initialization).
4347	*
4348	* @returns VBox status code.
4349	*/
4350	VMMR0DECL(int) VMXR0GlobalInit(void)
4351	{
4352	#ifdef HMVMX_USE_FUNCTION_TABLE
4353	AssertCompile(VMX_EXIT_MAX + 1 == RT_ELEMENTS(g_aVMExitHandlers));
4354	# ifdef VBOX_STRICT
4355	for (unsigned i = 0; i < RT_ELEMENTS(g_aVMExitHandlers); i++)
4356	Assert(g_aVMExitHandlers[i].pfn);
4357	# endif
4358	#endif
4359	return VINF_SUCCESS;
4360	}
4361
4362
4363	/**
4364	* Does global VT-x termination (called during module termination).
4365	*/
4366	VMMR0DECL(void) VMXR0GlobalTerm()
4367	{
4368	/* Nothing to do currently. */
4369	}
4370
4371
4372	/**
4373	* Sets up and activates VT-x on the current CPU.
4374	*
4375	* @returns VBox status code.
4376	* @param pHostCpu The HM physical-CPU structure.
4377	* @param pVM The cross context VM structure. Can be
4378	* NULL after a host resume operation.
4379	* @param pvCpuPage Pointer to the VMXON region (can be NULL if @a
4380	* fEnabledByHost is @c true).
4381	* @param HCPhysCpuPage Physical address of the VMXON region (can be 0 if
4382	* @a fEnabledByHost is @c true).
4383	* @param fEnabledByHost Set if SUPR0EnableVTx() or similar was used to
4384	* enable VT-x on the host.
4385	* @param pHwvirtMsrs Pointer to the hardware-virtualization MSRs.
4386	*/
4387	VMMR0DECL(int) VMXR0EnableCpu(PHMPHYSCPU pHostCpu, PVMCC pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost,
4388	PCSUPHWVIRTMSRS pHwvirtMsrs)
4389	{
4390	AssertPtr(pHostCpu);
4391	AssertPtr(pHwvirtMsrs);
4392	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4393
4394	/* Enable VT-x if it's not already enabled by the host. */
4395	if (!fEnabledByHost)
4396	{
4397	int rc = hmR0VmxEnterRootMode(pHostCpu, pVM, HCPhysCpuPage, pvCpuPage);
4398	if (RT_FAILURE(rc))
4399	return rc;
4400	}
4401
4402	/*
4403	* Flush all EPT tagged-TLB entries (in case VirtualBox or any other hypervisor have been
4404	* using EPTPs) so we don't retain any stale guest-physical mappings which won't get
4405	* invalidated when flushing by VPID.
4406	*/
4407	if (pHwvirtMsrs->u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
4408	{
4409	hmR0VmxFlushEpt(NULL /* pVCpu /, NULL / pVmcsInfo */, VMXTLBFLUSHEPT_ALL_CONTEXTS);
4410	pHostCpu->fFlushAsidBeforeUse = false;
4411	}
4412	else
4413	pHostCpu->fFlushAsidBeforeUse = true;
4414
4415	/* Ensure each VCPU scheduled on this CPU gets a new VPID on resume. See @bugref{6255}. */
4416	++pHostCpu->cTlbFlushes;
4417
4418	return VINF_SUCCESS;
4419	}
4420
4421
4422	/**
4423	* Deactivates VT-x on the current CPU.
4424	*
4425	* @returns VBox status code.
4426	* @param pHostCpu The HM physical-CPU structure.
4427	* @param pvCpuPage Pointer to the VMXON region.
4428	* @param HCPhysCpuPage Physical address of the VMXON region.
4429	*
4430	* @remarks This function should never be called when SUPR0EnableVTx() or
4431	* similar was used to enable VT-x on the host.
4432	*/
4433	VMMR0DECL(int) VMXR0DisableCpu(PHMPHYSCPU pHostCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
4434	{
4435	RT_NOREF2(pvCpuPage, HCPhysCpuPage);
4436
4437	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4438	return hmR0VmxLeaveRootMode(pHostCpu);
4439	}
4440
4441
4442	/**
4443	* Does per-VM VT-x initialization.
4444	*
4445	* @returns VBox status code.
4446	* @param pVM The cross context VM structure.
4447	*/
4448	VMMR0DECL(int) VMXR0InitVM(PVMCC pVM)
4449	{
4450	AssertPtr(pVM);
4451	LogFlowFunc(("pVM=%p\n", pVM));
4452
4453	hmR0VmxStructsInit(pVM);
4454	int rc = hmR0VmxStructsAlloc(pVM);
4455	if (RT_FAILURE(rc))
4456	{
4457	LogRelFunc(("Failed to allocated VMX structures. rc=%Rrc\n", rc));
4458	return rc;
4459	}
4460
4461	/* Setup the crash dump page. */
4462	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4463	strcpy((char *)pVM->hmr0.s.vmx.pbScratch, "SCRATCH Magic");
4464	(uint64_t )(pVM->hmr0.s.vmx.pbScratch + 16) = UINT64_C(0xdeadbeefdeadbeef);
4465	#endif
4466	return VINF_SUCCESS;
4467	}
4468
4469
4470	/**
4471	* Does per-VM VT-x termination.
4472	*
4473	* @returns VBox status code.
4474	* @param pVM The cross context VM structure.
4475	*/
4476	VMMR0DECL(int) VMXR0TermVM(PVMCC pVM)
4477	{
4478	AssertPtr(pVM);
4479	LogFlowFunc(("pVM=%p\n", pVM));
4480
4481	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4482	if (pVM->hmr0.s.vmx.pbScratch)
4483	RT_BZERO(pVM->hmr0.s.vmx.pbScratch, X86_PAGE_4K_SIZE);
4484	#endif
4485	hmR0VmxStructsFree(pVM);
4486	return VINF_SUCCESS;
4487	}
4488
4489
4490	/**
4491	* Sets up the VM for execution using hardware-assisted VMX.
4492	* This function is only called once per-VM during initialization.
4493	*
4494	* @returns VBox status code.
4495	* @param pVM The cross context VM structure.
4496	*/
4497	VMMR0DECL(int) VMXR0SetupVM(PVMCC pVM)
4498	{
4499	AssertPtr(pVM);
4500	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4501
4502	LogFlowFunc(("pVM=%p\n", pVM));
4503
4504	/*
4505	* At least verify if VMX is enabled, since we can't check if we're in VMX root mode or not
4506	* without causing a #GP.
4507	*/
4508	RTCCUINTREG const uHostCr4 = ASMGetCR4();
4509	if (RT_LIKELY(uHostCr4 & X86_CR4_VMXE))
4510	{ /* likely */ }
4511	else
4512	return VERR_VMX_NOT_IN_VMX_ROOT_MODE;
4513
4514	/*
4515	* Check that nested paging is supported if enabled and copy over the flag to the
4516	* ring-0 only structure.
4517	*/
4518	bool const fNestedPaging = pVM->hm.s.fNestedPagingCfg;
4519	AssertReturn( !fNestedPaging
4520	\|\| (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_EPT), /** @todo use a ring-0 copy of ProcCtls2.n.allowed1 */
4521	VERR_INCOMPATIBLE_CONFIG);
4522	pVM->hmr0.s.fNestedPaging = fNestedPaging;
4523	pVM->hmr0.s.fAllow64BitGuests = pVM->hm.s.fAllow64BitGuestsCfg;
4524
4525	/*
4526	* Without unrestricted guest execution, pRealModeTSS and pNonPagingModeEPTPageTable must
4527	* always be allocated. We no longer support the highly unlikely case of unrestricted guest
4528	* without pRealModeTSS, see hmR3InitFinalizeR0Intel().
4529	*/
4530	bool const fUnrestrictedGuest = pVM->hm.s.vmx.fUnrestrictedGuestCfg;
4531	AssertReturn( !fUnrestrictedGuest
4532	\|\| ( (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST)
4533	&& fNestedPaging),
4534	VERR_INCOMPATIBLE_CONFIG);
4535	if ( !fUnrestrictedGuest
4536	&& ( !pVM->hm.s.vmx.pNonPagingModeEPTPageTable
4537	\|\| !pVM->hm.s.vmx.pRealModeTSS))
4538	{
4539	LogRelFunc(("Invalid real-on-v86 state.\n"));
4540	return VERR_INTERNAL_ERROR;
4541	}
4542	pVM->hmr0.s.vmx.fUnrestrictedGuest = fUnrestrictedGuest;
4543
4544	/* Initialize these always, see hmR3InitFinalizeR0().*/
4545	pVM->hm.s.ForR3.vmx.enmTlbFlushEpt = pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NONE;
4546	pVM->hm.s.ForR3.vmx.enmTlbFlushVpid = pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NONE;
4547
4548	/* Setup the tagged-TLB flush handlers. */
4549	int rc = hmR0VmxSetupTaggedTlb(pVM);
4550	if (RT_FAILURE(rc))
4551	{
4552	LogRelFunc(("Failed to setup tagged TLB. rc=%Rrc\n", rc));
4553	return rc;
4554	}
4555
4556	/* Determine LBR capabilities. */
4557	pVM->hmr0.s.vmx.fLbr = pVM->hm.s.vmx.fLbrCfg;
4558	if (pVM->hmr0.s.vmx.fLbr)
4559	{
4560	rc = hmR0VmxSetupLbrMsrRange(pVM);
4561	if (RT_FAILURE(rc))
4562	{
4563	LogRelFunc(("Failed to setup LBR MSR range. rc=%Rrc\n", rc));
4564	return rc;
4565	}
4566	}
4567
4568	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4569	/* Setup the shadow VMCS fields array and VMREAD/VMWRITE bitmaps. */
4570	if (pVM->hmr0.s.vmx.fUseVmcsShadowing)
4571	{
4572	rc = hmR0VmxSetupShadowVmcsFieldsArrays(pVM);
4573	if (RT_SUCCESS(rc))
4574	hmR0VmxSetupVmreadVmwriteBitmaps(pVM);
4575	else
4576	{
4577	LogRelFunc(("Failed to setup shadow VMCS fields arrays. rc=%Rrc\n", rc));
4578	return rc;
4579	}
4580	}
4581	#endif
4582
4583	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
4584	{
4585	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
4586	Log4Func(("pVCpu=%p idCpu=%RU32\n", pVCpu, pVCpu->idCpu));
4587
4588	pVCpu->hmr0.s.vmx.pfnStartVm = hmR0VmxStartVmSelector;
4589
4590	rc = hmR0VmxSetupVmcs(pVCpu, &pVCpu->hmr0.s.vmx.VmcsInfo, false /* fIsNstGstVmcs */);
4591	if (RT_SUCCESS(rc))
4592	{
4593	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4594	if (pVM->cpum.ro.GuestFeatures.fVmx)
4595	{
4596	rc = hmR0VmxSetupVmcs(pVCpu, &pVCpu->hmr0.s.vmx.VmcsInfoNstGst, true /* fIsNstGstVmcs */);
4597	if (RT_SUCCESS(rc))
4598	{ /* likely */ }
4599	else
4600	{
4601	LogRelFunc(("Nested-guest VMCS setup failed. rc=%Rrc\n", rc));
4602	return rc;
4603	}
4604	}
4605	#endif
4606	}
4607	else
4608	{
4609	LogRelFunc(("VMCS setup failed. rc=%Rrc\n", rc));
4610	return rc;
4611	}
4612	}
4613
4614	return VINF_SUCCESS;
4615	}
4616
4617
4618	/**
4619	* Saves the host control registers (CR0, CR3, CR4) into the host-state area in
4620	* the VMCS.
4621	* @returns CR4 for passing along to hmR0VmxExportHostSegmentRegs.
4622	*/
4623	static uint64_t hmR0VmxExportHostControlRegs(void)
4624	{
4625	int rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR0, ASMGetCR0()); AssertRC(rc);
4626	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR3, ASMGetCR3()); AssertRC(rc);
4627	uint64_t uHostCr4 = ASMGetCR4();
4628	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR4, uHostCr4); AssertRC(rc);
4629	return uHostCr4;
4630	}
4631
4632
4633	/**
4634	* Saves the host segment registers and GDTR, IDTR, (TR, GS and FS bases) into
4635	* the host-state area in the VMCS.
4636	*
4637	* @returns VBox status code.
4638	* @param pVCpu The cross context virtual CPU structure.
4639	* @param uHostCr4 The host CR4 value.
4640	*/
4641	static int hmR0VmxExportHostSegmentRegs(PVMCPUCC pVCpu, uint64_t uHostCr4)
4642	{
4643	/*
4644	* If we've executed guest code using hardware-assisted VMX, the host-state bits
4645	* will be messed up. We should -not- save the messed up state without restoring
4646	* the original host-state, see @bugref{7240}.
4647	*
4648	* This apparently can happen (most likely the FPU changes), deal with it rather than
4649	* asserting. Was observed booting Solaris 10u10 32-bit guest.
4650	*/
4651	if (pVCpu->hmr0.s.vmx.fRestoreHostFlags > VMX_RESTORE_HOST_REQUIRED)
4652	{
4653	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hmr0.s.vmx.fRestoreHostFlags,
4654	pVCpu->idCpu));
4655	VMXRestoreHostState(pVCpu->hmr0.s.vmx.fRestoreHostFlags, &pVCpu->hmr0.s.vmx.RestoreHost);
4656	pVCpu->hmr0.s.vmx.fRestoreHostFlags = 0;
4657	}
4658
4659	/*
4660	* Get all the host info.
4661	* ASSUME it is safe to use rdfsbase and friends if the CR4.FSGSBASE bit is set
4662	* without also checking the cpuid bit.
4663	*/
4664	uint32_t fRestoreHostFlags;
4665	#if RT_INLINE_ASM_EXTERNAL
4666	if (uHostCr4 & X86_CR4_FSGSBASE)
4667	{
4668	hmR0VmxExportHostSegmentRegsAsmHlp(&pVCpu->hmr0.s.vmx.RestoreHost, true /fHaveFsGsBase/);
4669	fRestoreHostFlags = VMX_RESTORE_HOST_CAN_USE_WRFSBASE_AND_WRGSBASE;
4670	}
4671	else
4672	{
4673	hmR0VmxExportHostSegmentRegsAsmHlp(&pVCpu->hmr0.s.vmx.RestoreHost, false /fHaveFsGsBase/);
4674	fRestoreHostFlags = 0;
4675	}
4676	RTSEL uSelES = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelES;
4677	RTSEL uSelDS = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelDS;
4678	RTSEL uSelFS = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelFS;
4679	RTSEL uSelGS = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelGS;
4680	#else
4681	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR = ASMGetTR();
4682	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelSS = ASMGetSS();
4683	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelCS = ASMGetCS();
4684	ASMGetGDTR((PRTGDTR)&pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr);
4685	ASMGetIDTR((PRTIDTR)&pVCpu->hmr0.s.vmx.RestoreHost.HostIdtr);
4686	if (uHostCr4 & X86_CR4_FSGSBASE)
4687	{
4688	pVCpu->hmr0.s.vmx.RestoreHost.uHostFSBase = ASMGetFSBase();
4689	pVCpu->hmr0.s.vmx.RestoreHost.uHostGSBase = ASMGetGSBase();
4690	fRestoreHostFlags = VMX_RESTORE_HOST_CAN_USE_WRFSBASE_AND_WRGSBASE;
4691	}
4692	else
4693	{
4694	pVCpu->hmr0.s.vmx.RestoreHost.uHostFSBase = ASMRdMsr(MSR_K8_FS_BASE);
4695	pVCpu->hmr0.s.vmx.RestoreHost.uHostGSBase = ASMRdMsr(MSR_K8_GS_BASE);
4696	fRestoreHostFlags = 0;
4697	}
4698	RTSEL uSelES, uSelDS, uSelFS, uSelGS;
4699	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelDS = uSelDS = ASMGetDS();
4700	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelES = uSelES = ASMGetES();
4701	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelFS = uSelFS = ASMGetFS();
4702	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelGS = uSelGS = ASMGetGS();
4703	#endif
4704
4705	/*
4706	* Determine if the host segment registers are suitable for VT-x. Otherwise use zero to
4707	* gain VM-entry and restore them before we get preempted.
4708	*
4709	* See Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers".
4710	*/
4711	RTSEL const uSelAll = uSelFS \| uSelGS \| uSelES \| uSelDS;
4712	if (uSelAll & (X86_SEL_RPL \| X86_SEL_LDT))
4713	{
4714	if (!(uSelAll & X86_SEL_LDT))
4715	{
4716	#define VMXLOCAL_ADJUST_HOST_SEG(a_Seg, a_uVmcsVar) \
4717	do { \
4718	(a_uVmcsVar) = pVCpu->hmr0.s.vmx.RestoreHost.uHostSel##a_Seg; \
4719	if ((a_uVmcsVar) & X86_SEL_RPL) \
4720	{ \
4721	fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##a_Seg; \
4722	(a_uVmcsVar) = 0; \
4723	} \
4724	} while (0)
4725	VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS);
4726	VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES);
4727	VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS);
4728	VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS);
4729	#undef VMXLOCAL_ADJUST_HOST_SEG
4730	}
4731	else
4732	{
4733	#define VMXLOCAL_ADJUST_HOST_SEG(a_Seg, a_uVmcsVar) \
4734	do { \
4735	(a_uVmcsVar) = pVCpu->hmr0.s.vmx.RestoreHost.uHostSel##a_Seg; \
4736	if ((a_uVmcsVar) & (X86_SEL_RPL \| X86_SEL_LDT)) \
4737	{ \
4738	if (!((a_uVmcsVar) & X86_SEL_LDT)) \
4739	fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##a_Seg; \
4740	else \
4741	{ \
4742	uint32_t const fAttr = ASMGetSegAttr(a_uVmcsVar); \
4743	if ((fAttr & X86_DESC_P) && fAttr != UINT32_MAX) \
4744	fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##a_Seg; \
4745	} \
4746	(a_uVmcsVar) = 0; \
4747	} \
4748	} while (0)
4749	VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS);
4750	VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES);
4751	VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS);
4752	VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS);
4753	#undef VMXLOCAL_ADJUST_HOST_SEG
4754	}
4755	}
4756
4757	/* Verification based on Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers" */
4758	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);
4759	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);
4760	Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelSS & X86_SEL_RPL)); Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelSS & X86_SEL_LDT));
4761	Assert(!(uSelDS & X86_SEL_RPL)); Assert(!(uSelDS & X86_SEL_LDT));
4762	Assert(!(uSelES & X86_SEL_RPL)); Assert(!(uSelES & X86_SEL_LDT));
4763	Assert(!(uSelFS & X86_SEL_RPL)); Assert(!(uSelFS & X86_SEL_LDT));
4764	Assert(!(uSelGS & X86_SEL_RPL)); Assert(!(uSelGS & X86_SEL_LDT));
4765
4766	/*
4767	* Determine if we need to manually need to restore the GDTR and IDTR limits as VT-x zaps
4768	* them to the maximum limit (0xffff) on every VM-exit.
4769	*/
4770	if (pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.cb != 0xffff)
4771	fRestoreHostFlags \|= VMX_RESTORE_HOST_GDTR;
4772
4773	/*
4774	* IDT limit is effectively capped at 0xfff. (See Intel spec. 6.14.1 "64-Bit Mode IDT" and
4775	* Intel spec. 6.2 "Exception and Interrupt Vectors".) Therefore if the host has the limit
4776	* as 0xfff, VT-x bloating the limit to 0xffff shouldn't cause any different CPU behavior.
4777	* However, several hosts either insists on 0xfff being the limit (Windows Patch Guard) or
4778	* uses the limit for other purposes (darwin puts the CPU ID in there but botches sidt
4779	* alignment in at least one consumer). So, we're only allowing the IDTR.LIMIT to be left
4780	* at 0xffff on hosts where we are sure it won't cause trouble.
4781	*/
4782	#if defined(RT_OS_LINUX) \|\| defined(RT_OS_SOLARIS)
4783	if (pVCpu->hmr0.s.vmx.RestoreHost.HostIdtr.cb < 0x0fff)
4784	#else
4785	if (pVCpu->hmr0.s.vmx.RestoreHost.HostIdtr.cb != 0xffff)
4786	#endif
4787	fRestoreHostFlags \|= VMX_RESTORE_HOST_IDTR;
4788
4789	/*
4790	* Host TR base. Verify that TR selector doesn't point past the GDT. Masking off the TI
4791	* and RPL bits is effectively what the CPU does for "scaling by 8". TI is always 0 and
4792	* RPL should be too in most cases.
4793	*/
4794	RTSEL const uSelTR = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR;
4795	AssertMsgReturn((uSelTR \| X86_SEL_RPL_LDT) <= pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.cb,
4796	("TR selector exceeds limit. TR=%RTsel cbGdt=%#x\n", uSelTR, pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.cb),
4797	VERR_VMX_INVALID_HOST_STATE);
4798
4799	PCX86DESCHC pDesc = (PCX86DESCHC)(pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.uAddr + (uSelTR & X86_SEL_MASK));
4800	uintptr_t const uTRBase = X86DESC64_BASE(pDesc);
4801
4802	/*
4803	* VT-x unconditionally restores the TR limit to 0x67 and type to 11 (32-bit busy TSS) on
4804	* all VM-exits. The type is the same for 64-bit busy TSS[1]. The limit needs manual
4805	* restoration if the host has something else. Task switching is not supported in 64-bit
4806	* mode[2], but the limit still matters as IOPM is supported in 64-bit mode. Restoring the
4807	* limit lazily while returning to ring-3 is safe because IOPM is not applicable in ring-0.
4808	*
4809	* [1] See Intel spec. 3.5 "System Descriptor Types".
4810	* [2] See Intel spec. 7.2.3 "TSS Descriptor in 64-bit mode".
4811	*/
4812	Assert(pDesc->System.u4Type == 11);
4813	if ( pDesc->System.u16LimitLow != 0x67
4814	\|\| pDesc->System.u4LimitHigh)
4815	{
4816	fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_TR;
4817
4818	/* If the host has made GDT read-only, we would need to temporarily toggle CR0.WP before writing the GDT. */
4819	if (g_fHmHostKernelFeatures & SUPKERNELFEATURES_GDT_READ_ONLY)
4820	fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_READ_ONLY;
4821	if (g_fHmHostKernelFeatures & SUPKERNELFEATURES_GDT_NEED_WRITABLE)
4822	{
4823	/* The GDT is read-only but the writable GDT is available. */
4824	fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_NEED_WRITABLE;
4825	pVCpu->hmr0.s.vmx.RestoreHost.HostGdtrRw.cb = pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.cb;
4826	int rc = SUPR0GetCurrentGdtRw(&pVCpu->hmr0.s.vmx.RestoreHost.HostGdtrRw.uAddr);
4827	AssertRCReturn(rc, rc);
4828	}
4829	}
4830
4831	pVCpu->hmr0.s.vmx.fRestoreHostFlags = fRestoreHostFlags;
4832
4833	/*
4834	* Do all the VMCS updates in one block to assist nested virtualization.
4835	*/
4836	int rc;
4837	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_CS_SEL, pVCpu->hmr0.s.vmx.RestoreHost.uHostSelCS); AssertRC(rc);
4838	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_SS_SEL, pVCpu->hmr0.s.vmx.RestoreHost.uHostSelSS); AssertRC(rc);
4839	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_DS_SEL, uSelDS); AssertRC(rc);
4840	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_ES_SEL, uSelES); AssertRC(rc);
4841	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_FS_SEL, uSelFS); AssertRC(rc);
4842	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_GS_SEL, uSelGS); AssertRC(rc);
4843	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_TR_SEL, pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR); AssertRC(rc);
4844	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_GDTR_BASE, pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.uAddr); AssertRC(rc);
4845	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_IDTR_BASE, pVCpu->hmr0.s.vmx.RestoreHost.HostIdtr.uAddr); AssertRC(rc);
4846	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_TR_BASE, uTRBase); AssertRC(rc);
4847	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_FS_BASE, pVCpu->hmr0.s.vmx.RestoreHost.uHostFSBase); AssertRC(rc);
4848	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_GS_BASE, pVCpu->hmr0.s.vmx.RestoreHost.uHostGSBase); AssertRC(rc);
4849
4850	return VINF_SUCCESS;
4851	}
4852
4853
4854	/**
4855	* Exports certain host MSRs in the VM-exit MSR-load area and some in the
4856	* host-state area of the VMCS.
4857	*
4858	* These MSRs will be automatically restored on the host after every successful
4859	* VM-exit.
4860	*
4861	* @param pVCpu The cross context virtual CPU structure.
4862	*
4863	* @remarks No-long-jump zone!!!
4864	*/
4865	static void hmR0VmxExportHostMsrs(PVMCPUCC pVCpu)
4866	{
4867	AssertPtr(pVCpu);
4868
4869	/*
4870	* Save MSRs that we restore lazily (due to preemption or transition to ring-3)
4871	* rather than swapping them on every VM-entry.
4872	*/
4873	hmR0VmxLazySaveHostMsrs(pVCpu);
4874
4875	/*
4876	* Host Sysenter MSRs.
4877	*/
4878	int rc = VMXWriteVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, ASMRdMsr_Low(MSR_IA32_SYSENTER_CS)); AssertRC(rc);
4879	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP)); AssertRC(rc);
4880	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP)); AssertRC(rc);
4881
4882	/*
4883	* Host EFER MSR.
4884	*
4885	* If the CPU supports the newer VMCS controls for managing EFER, use it. Otherwise it's
4886	* done as part of auto-load/store MSR area in the VMCS, see hmR0VmxExportGuestMsrs().
4887	*/
4888	if (g_fHmVmxSupportsVmcsEfer)
4889	{
4890	rc = VMXWriteVmcs64(VMX_VMCS64_HOST_EFER_FULL, g_uHmVmxHostMsrEfer);
4891	AssertRC(rc);
4892	}
4893
4894	/** @todo IA32_PERF_GLOBALCTRL, IA32_PAT also see
4895	* hmR0VmxExportGuestEntryExitCtls(). */
4896	}
4897
4898
4899	/**
4900	* Figures out if we need to swap the EFER MSR which is particularly expensive.
4901	*
4902	* We check all relevant bits. For now, that's everything besides LMA/LME, as
4903	* these two bits are handled by VM-entry, see hmR0VMxExportGuestEntryExitCtls().
4904	*
4905	* @returns true if we need to load guest EFER, false otherwise.
4906	* @param pVCpu The cross context virtual CPU structure.
4907	* @param pVmxTransient The VMX-transient structure.
4908	*
4909	* @remarks Requires EFER, CR4.
4910	* @remarks No-long-jump zone!!!
4911	*/
4912	static bool hmR0VmxShouldSwapEferMsr(PCVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
4913	{
4914	#ifdef HMVMX_ALWAYS_SWAP_EFER
4915	RT_NOREF2(pVCpu, pVmxTransient);
4916	return true;
4917	#else
4918	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
4919	uint64_t const u64HostEfer = g_uHmVmxHostMsrEfer;
4920	uint64_t const u64GuestEfer = pCtx->msrEFER;
4921
4922	# ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4923	/*
4924	* For nested-guests, we shall honor swapping the EFER MSR when requested by
4925	* the nested-guest.
4926	*/
4927	if ( pVmxTransient->fIsNestedGuest
4928	&& ( CPUMIsGuestVmxEntryCtlsSet(pCtx, VMX_ENTRY_CTLS_LOAD_EFER_MSR)
4929	\|\| CPUMIsGuestVmxExitCtlsSet(pCtx, VMX_EXIT_CTLS_SAVE_EFER_MSR)
4930	\|\| CPUMIsGuestVmxExitCtlsSet(pCtx, VMX_EXIT_CTLS_LOAD_EFER_MSR)))
4931	return true;
4932	# else
4933	RT_NOREF(pVmxTransient);
4934	#endif
4935
4936	/*
4937	* For 64-bit guests, if EFER.SCE bit differs, we need to swap the EFER MSR
4938	* to ensure that the guest's SYSCALL behaviour isn't broken, see @bugref{7386}.
4939	*/
4940	if ( CPUMIsGuestInLongModeEx(pCtx)
4941	&& (u64GuestEfer & MSR_K6_EFER_SCE) != (u64HostEfer & MSR_K6_EFER_SCE))
4942	return true;
4943
4944	/*
4945	* If the guest uses PAE and EFER.NXE bit differs, we need to swap the EFER MSR
4946	* as it affects guest paging. 64-bit paging implies CR4.PAE as well.
4947	*
4948	* See Intel spec. 4.5 "IA-32e Paging".
4949	* See Intel spec. 4.1.1 "Three Paging Modes".
4950	*
4951	* Verify that we always intercept CR4.PAE and CR0.PG bits, so we don't need to
4952	* import CR4 and CR0 from the VMCS here as those bits are always up to date.
4953	*/
4954	Assert(hmR0VmxGetFixedCr4Mask(pVCpu) & X86_CR4_PAE);
4955	Assert(hmR0VmxGetFixedCr0Mask(pVCpu) & X86_CR0_PG);
4956	if ( (pCtx->cr4 & X86_CR4_PAE)
4957	&& (pCtx->cr0 & X86_CR0_PG))
4958	{
4959	/*
4960	* If nested paging is not used, verify that the guest paging mode matches the
4961	* shadow paging mode which is/will be placed in the VMCS (which is what will
4962	* actually be used while executing the guest and not the CR4 shadow value).
4963	*/
4964	AssertMsg( pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging
4965	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_PAE
4966	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_PAE_NX
4967	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_AMD64
4968	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_AMD64_NX,
4969	("enmShadowMode=%u\n", pVCpu->hm.s.enmShadowMode));
4970	if ((u64GuestEfer & MSR_K6_EFER_NXE) != (u64HostEfer & MSR_K6_EFER_NXE))
4971	{
4972	/* Verify that the host is NX capable. */
4973	Assert(pVCpu->CTX_SUFF(pVM)->cpum.ro.HostFeatures.fNoExecute);
4974	return true;
4975	}
4976	}
4977
4978	return false;
4979	#endif
4980	}
4981
4982
4983	/**
4984	* Exports the guest state with appropriate VM-entry and VM-exit controls in the
4985	* VMCS.
4986	*
4987	* This is typically required when the guest changes paging mode.
4988	*
4989	* @returns VBox status code.
4990	* @param pVCpu The cross context virtual CPU structure.
4991	* @param pVmxTransient The VMX-transient structure.
4992	*
4993	* @remarks Requires EFER.
4994	* @remarks No-long-jump zone!!!
4995	*/
4996	static int hmR0VmxExportGuestEntryExitCtls(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
4997	{
4998	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_ENTRY_EXIT_CTLS)
4999	{
5000	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5001	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5002
5003	/*
5004	* VM-entry controls.
5005	*/
5006	{
5007	uint32_t fVal = g_HmMsrs.u.vmx.EntryCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
5008	uint32_t const fZap = g_HmMsrs.u.vmx.EntryCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
5009
5010	/*
5011	* Load the guest debug controls (DR7 and IA32_DEBUGCTL MSR) on VM-entry.
5012	* The first VT-x capable CPUs only supported the 1-setting of this bit.
5013	*
5014	* For nested-guests, this is a mandatory VM-entry control. It's also
5015	* required because we do not want to leak host bits to the nested-guest.
5016	*/
5017	fVal \|= VMX_ENTRY_CTLS_LOAD_DEBUG;
5018
5019	/*
5020	* Set if the guest is in long mode. This will set/clear the EFER.LMA bit on VM-entry.
5021	*
5022	* For nested-guests, the "IA-32e mode guest" control we initialize with what is
5023	* required to get the nested-guest working with hardware-assisted VMX execution.
5024	* It depends on the nested-guest's IA32_EFER.LMA bit. Remember, a nested hypervisor
5025	* can skip intercepting changes to the EFER MSR. This is why it needs to be done
5026	* here rather than while merging the guest VMCS controls.
5027	*/
5028	if (CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
5029	{
5030	Assert(pVCpu->cpum.GstCtx.msrEFER & MSR_K6_EFER_LME);
5031	fVal \|= VMX_ENTRY_CTLS_IA32E_MODE_GUEST;
5032	}
5033	else
5034	Assert(!(fVal & VMX_ENTRY_CTLS_IA32E_MODE_GUEST));
5035
5036	/*
5037	* If the CPU supports the newer VMCS controls for managing guest/host EFER, use it.
5038	*
5039	* For nested-guests, we use the "load IA32_EFER" if the hardware supports it,
5040	* regardless of whether the nested-guest VMCS specifies it because we are free to
5041	* load whatever MSRs we require and we do not need to modify the guest visible copy
5042	* of the VM-entry MSR load area.
5043	*/
5044	if ( g_fHmVmxSupportsVmcsEfer
5045	&& hmR0VmxShouldSwapEferMsr(pVCpu, pVmxTransient))
5046	fVal \|= VMX_ENTRY_CTLS_LOAD_EFER_MSR;
5047	else
5048	Assert(!(fVal & VMX_ENTRY_CTLS_LOAD_EFER_MSR));
5049
5050	/*
5051	* The following should -not- be set (since we're not in SMM mode):
5052	* - VMX_ENTRY_CTLS_ENTRY_TO_SMM
5053	* - VMX_ENTRY_CTLS_DEACTIVATE_DUAL_MON
5054	*/
5055
5056	/** @todo VMX_ENTRY_CTLS_LOAD_PERF_MSR,
5057	* VMX_ENTRY_CTLS_LOAD_PAT_MSR. */
5058
5059	if ((fVal & fZap) == fVal)
5060	{ /* likely */ }
5061	else
5062	{
5063	Log4Func(("Invalid VM-entry controls combo! Cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
5064	g_HmMsrs.u.vmx.EntryCtls.n.allowed0, fVal, fZap));
5065	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_ENTRY;
5066	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
5067	}
5068
5069	/* Commit it to the VMCS. */
5070	if (pVmcsInfo->u32EntryCtls != fVal)
5071	{
5072	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY, fVal);
5073	AssertRC(rc);
5074	pVmcsInfo->u32EntryCtls = fVal;
5075	}
5076	}
5077
5078	/*
5079	* VM-exit controls.
5080	*/
5081	{
5082	uint32_t fVal = g_HmMsrs.u.vmx.ExitCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
5083	uint32_t const fZap = g_HmMsrs.u.vmx.ExitCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
5084
5085	/*
5086	* Save debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x CPUs only
5087	* supported the 1-setting of this bit.
5088	*
5089	* For nested-guests, we set the "save debug controls" as the converse
5090	* "load debug controls" is mandatory for nested-guests anyway.
5091	*/
5092	fVal \|= VMX_EXIT_CTLS_SAVE_DEBUG;
5093
5094	/*
5095	* Set the host long mode active (EFER.LMA) bit (which Intel calls
5096	* "Host address-space size") if necessary. On VM-exit, VT-x sets both the
5097	* host EFER.LMA and EFER.LME bit to this value. See assertion in
5098	* hmR0VmxExportHostMsrs().
5099	*
5100	* For nested-guests, we always set this bit as we do not support 32-bit
5101	* hosts.
5102	*/
5103	fVal \|= VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE;
5104
5105	/*
5106	* If the VMCS EFER MSR fields are supported by the hardware, we use it.
5107	*
5108	* For nested-guests, we should use the "save IA32_EFER" control if we also
5109	* used the "load IA32_EFER" control while exporting VM-entry controls.
5110	*/
5111	if ( g_fHmVmxSupportsVmcsEfer
5112	&& hmR0VmxShouldSwapEferMsr(pVCpu, pVmxTransient))
5113	{
5114	fVal \|= VMX_EXIT_CTLS_SAVE_EFER_MSR
5115	\| VMX_EXIT_CTLS_LOAD_EFER_MSR;
5116	}
5117
5118	/*
5119	* Enable saving of the VMX-preemption timer value on VM-exit.
5120	* For nested-guests, currently not exposed/used.
5121	*/
5122	/** @todo r=bird: Measure performance hit because of this vs. always rewriting
5123	* the timer value. */
5124	if (pVM->hmr0.s.vmx.fUsePreemptTimer)
5125	{
5126	Assert(g_HmMsrs.u.vmx.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER);
5127	fVal \|= VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER;
5128	}
5129
5130	/* Don't acknowledge external interrupts on VM-exit. We want to let the host do that. */
5131	Assert(!(fVal & VMX_EXIT_CTLS_ACK_EXT_INT));
5132
5133	/** @todo VMX_EXIT_CTLS_LOAD_PERF_MSR,
5134	* VMX_EXIT_CTLS_SAVE_PAT_MSR,
5135	* VMX_EXIT_CTLS_LOAD_PAT_MSR. */
5136
5137	if ((fVal & fZap) == fVal)
5138	{ /* likely */ }
5139	else
5140	{
5141	Log4Func(("Invalid VM-exit controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%R#X32\n",
5142	g_HmMsrs.u.vmx.ExitCtls.n.allowed0, fVal, fZap));
5143	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_EXIT;
5144	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
5145	}
5146
5147	/* Commit it to the VMCS. */
5148	if (pVmcsInfo->u32ExitCtls != fVal)
5149	{
5150	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT, fVal);
5151	AssertRC(rc);
5152	pVmcsInfo->u32ExitCtls = fVal;
5153	}
5154	}
5155
5156	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
5157	}
5158	return VINF_SUCCESS;
5159	}
5160
5161
5162	/**
5163	* Sets the TPR threshold in the VMCS.
5164	*
5165	* @param pVmcsInfo The VMCS info. object.
5166	* @param u32TprThreshold The TPR threshold (task-priority class only).
5167	*/
5168	DECLINLINE(void) hmR0VmxApicSetTprThreshold(PVMXVMCSINFO pVmcsInfo, uint32_t u32TprThreshold)
5169	{
5170	Assert(!(u32TprThreshold & ~VMX_TPR_THRESHOLD_MASK)); /* Bits 31:4 MBZ. */
5171	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
5172	RT_NOREF(pVmcsInfo);
5173	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
5174	AssertRC(rc);
5175	}
5176
5177
5178	/**
5179	* Exports the guest APIC TPR state into the VMCS.
5180	*
5181	* @param pVCpu The cross context virtual CPU structure.
5182	* @param pVmxTransient The VMX-transient structure.
5183	*
5184	* @remarks No-long-jump zone!!!
5185	*/
5186	static void hmR0VmxExportGuestApicTpr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5187	{
5188	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_APIC_TPR)
5189	{
5190	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_APIC_TPR);
5191
5192	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5193	if (!pVmxTransient->fIsNestedGuest)
5194	{
5195	if ( PDMHasApic(pVCpu->CTX_SUFF(pVM))
5196	&& APICIsEnabled(pVCpu))
5197	{
5198	/*
5199	* Setup TPR shadowing.
5200	*/
5201	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
5202	{
5203	bool fPendingIntr = false;
5204	uint8_t u8Tpr = 0;
5205	uint8_t u8PendingIntr = 0;
5206	int rc = APICGetTpr(pVCpu, &u8Tpr, &fPendingIntr, &u8PendingIntr);
5207	AssertRC(rc);
5208
5209	/*
5210	* If there are interrupts pending but masked by the TPR, instruct VT-x to
5211	* cause a TPR-below-threshold VM-exit when the guest lowers its TPR below the
5212	* priority of the pending interrupt so we can deliver the interrupt. If there
5213	* are no interrupts pending, set threshold to 0 to not cause any
5214	* TPR-below-threshold VM-exits.
5215	*/
5216	uint32_t u32TprThreshold = 0;
5217	if (fPendingIntr)
5218	{
5219	/* Bits 3:0 of the TPR threshold field correspond to bits 7:4 of the TPR
5220	(which is the Task-Priority Class). */
5221	const uint8_t u8PendingPriority = u8PendingIntr >> 4;
5222	const uint8_t u8TprPriority = u8Tpr >> 4;
5223	if (u8PendingPriority <= u8TprPriority)
5224	u32TprThreshold = u8PendingPriority;
5225	}
5226
5227	hmR0VmxApicSetTprThreshold(pVmcsInfo, u32TprThreshold);
5228	}
5229	}
5230	}
5231	/* else: the TPR threshold has already been updated while merging the nested-guest VMCS. */
5232	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_APIC_TPR);
5233	}
5234	}
5235
5236
5237	/**
5238	* Gets the guest interruptibility-state and updates related force-flags.
5239	*
5240	* @returns Guest's interruptibility-state.
5241	* @param pVCpu The cross context virtual CPU structure.
5242	*
5243	* @remarks No-long-jump zone!!!
5244	*/
5245	static uint32_t hmR0VmxGetGuestIntrStateAndUpdateFFs(PVMCPUCC pVCpu)
5246	{
5247	/*
5248	* Check if we should inhibit interrupt delivery due to instructions like STI and MOV SS.
5249	*/
5250	uint32_t fIntrState = 0;
5251	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
5252	{
5253	/* If inhibition is active, RIP and RFLAGS should've been imported from the VMCS already. */
5254	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS);
5255
5256	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5257	if (pCtx->rip == EMGetInhibitInterruptsPC(pVCpu))
5258	{
5259	if (pCtx->eflags.Bits.u1IF)
5260	fIntrState = VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
5261	else
5262	fIntrState = VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS;
5263	}
5264	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
5265	{
5266	/*
5267	* We can clear the inhibit force flag as even if we go back to the recompiler
5268	* without executing guest code in VT-x, the flag's condition to be cleared is
5269	* met and thus the cleared state is correct.
5270	*/
5271	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
5272	}
5273	}
5274
5275	/*
5276	* Check if we should inhibit NMI delivery.
5277	*/
5278	if (CPUMIsGuestNmiBlocking(pVCpu))
5279	fIntrState \|= VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI;
5280
5281	/*
5282	* Validate.
5283	*/
5284	#ifdef VBOX_STRICT
5285	/* We don't support block-by-SMI yet.*/
5286	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI));
5287
5288	/* Block-by-STI must not be set when interrupts are disabled. */
5289	if (fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
5290	{
5291	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
5292	Assert(pVCpu->cpum.GstCtx.eflags.u & X86_EFL_IF);
5293	}
5294	#endif
5295
5296	return fIntrState;
5297	}
5298
5299
5300	/**
5301	* Exports the exception intercepts required for guest execution in the VMCS.
5302	*
5303	* @param pVCpu The cross context virtual CPU structure.
5304	* @param pVmxTransient The VMX-transient structure.
5305	*
5306	* @remarks No-long-jump zone!!!
5307	*/
5308	static void hmR0VmxExportGuestXcptIntercepts(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5309	{
5310	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_XCPT_INTERCEPTS)
5311	{
5312	/* When executing a nested-guest, we do not need to trap GIM hypercalls by intercepting #UD. */
5313	if ( !pVmxTransient->fIsNestedGuest
5314	&& pVCpu->hm.s.fGIMTrapXcptUD)
5315	hmR0VmxAddXcptIntercept(pVmxTransient, X86_XCPT_UD);
5316	else
5317	hmR0VmxRemoveXcptIntercept(pVCpu, pVmxTransient, X86_XCPT_UD);
5318
5319	/* Other exception intercepts are handled elsewhere, e.g. while exporting guest CR0. */
5320	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_XCPT_INTERCEPTS);
5321	}
5322	}
5323
5324
5325	/**
5326	* Exports the guest's RIP into the guest-state area in the VMCS.
5327	*
5328	* @param pVCpu The cross context virtual CPU structure.
5329	*
5330	* @remarks No-long-jump zone!!!
5331	*/
5332	static void hmR0VmxExportGuestRip(PVMCPUCC pVCpu)
5333	{
5334	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RIP)
5335	{
5336	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP);
5337
5338	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RIP, pVCpu->cpum.GstCtx.rip);
5339	AssertRC(rc);
5340
5341	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RIP);
5342	Log4Func(("rip=%#RX64\n", pVCpu->cpum.GstCtx.rip));
5343	}
5344	}
5345
5346
5347	/**
5348	* Exports the guest's RSP into the guest-state area in the VMCS.
5349	*
5350	* @param pVCpu The cross context virtual CPU structure.
5351	*
5352	* @remarks No-long-jump zone!!!
5353	*/
5354	static void hmR0VmxExportGuestRsp(PVMCPUCC pVCpu)
5355	{
5356	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RSP)
5357	{
5358	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RSP);
5359
5360	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RSP, pVCpu->cpum.GstCtx.rsp);
5361	AssertRC(rc);
5362
5363	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RSP);
5364	Log4Func(("rsp=%#RX64\n", pVCpu->cpum.GstCtx.rsp));
5365	}
5366	}
5367
5368
5369	/**
5370	* Exports the guest's RFLAGS into the guest-state area in the VMCS.
5371	*
5372	* @param pVCpu The cross context virtual CPU structure.
5373	* @param pVmxTransient The VMX-transient structure.
5374	*
5375	* @remarks No-long-jump zone!!!
5376	*/
5377	static void hmR0VmxExportGuestRflags(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5378	{
5379	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RFLAGS)
5380	{
5381	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
5382
5383	/* Intel spec. 2.3.1 "System Flags and Fields in IA-32e Mode" claims the upper 32-bits of RFLAGS are reserved (MBZ).
5384	Let us assert it as such and use 32-bit VMWRITE. */
5385	Assert(!RT_HI_U32(pVCpu->cpum.GstCtx.rflags.u64));
5386	X86EFLAGS fEFlags = pVCpu->cpum.GstCtx.eflags;
5387	Assert(fEFlags.u32 & X86_EFL_RA1_MASK);
5388	Assert(!(fEFlags.u32 & ~(X86_EFL_1 \| X86_EFL_LIVE_MASK)));
5389
5390	/*
5391	* If we're emulating real-mode using Virtual 8086 mode, save the real-mode eflags so
5392	* we can restore them on VM-exit. Modify the real-mode guest's eflags so that VT-x
5393	* can run the real-mode guest code under Virtual 8086 mode.
5394	*/
5395	PVMXVMCSINFOSHARED pVmcsInfo = pVmxTransient->pVmcsInfo->pShared;
5396	if (pVmcsInfo->RealMode.fRealOnV86Active)
5397	{
5398	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
5399	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
5400	Assert(!pVmxTransient->fIsNestedGuest);
5401	pVmcsInfo->RealMode.Eflags.u32 = fEFlags.u32; /* Save the original eflags of the real-mode guest. */
5402	fEFlags.Bits.u1VM = 1; /* Set the Virtual 8086 mode bit. */
5403	fEFlags.Bits.u2IOPL = 0; /* Change IOPL to 0, otherwise certain instructions won't fault. */
5404	}
5405
5406	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RFLAGS, fEFlags.u32);
5407	AssertRC(rc);
5408
5409	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RFLAGS);
5410	Log4Func(("eflags=%#RX32\n", fEFlags.u32));
5411	}
5412	}
5413
5414
5415	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5416	/**
5417	* Copies the nested-guest VMCS to the shadow VMCS.
5418	*
5419	* @returns VBox status code.
5420	* @param pVCpu The cross context virtual CPU structure.
5421	* @param pVmcsInfo The VMCS info. object.
5422	*
5423	* @remarks No-long-jump zone!!!
5424	*/
5425	static int hmR0VmxCopyNstGstToShadowVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
5426	{
5427	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5428	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5429
5430	/*
5431	* Disable interrupts so we don't get preempted while the shadow VMCS is the
5432	* current VMCS, as we may try saving guest lazy MSRs.
5433	*
5434	* Strictly speaking the lazy MSRs are not in the VMCS, but I'd rather not risk
5435	* calling the import VMCS code which is currently performing the guest MSR reads
5436	* (on 64-bit hosts) and accessing the auto-load/store MSR area on 32-bit hosts
5437	* and the rest of the VMX leave session machinery.
5438	*/
5439	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
5440
5441	int rc = hmR0VmxLoadShadowVmcs(pVmcsInfo);
5442	if (RT_SUCCESS(rc))
5443	{
5444	/*
5445	* Copy all guest read/write VMCS fields.
5446	*
5447	* We don't check for VMWRITE failures here for performance reasons and
5448	* because they are not expected to fail, barring irrecoverable conditions
5449	* like hardware errors.
5450	*/
5451	uint32_t const cShadowVmcsFields = pVM->hmr0.s.vmx.cShadowVmcsFields;
5452	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
5453	{
5454	uint64_t u64Val;
5455	uint32_t const uVmcsField = pVM->hmr0.s.vmx.paShadowVmcsFields[i];
5456	IEMReadVmxVmcsField(pVmcsNstGst, uVmcsField, &u64Val);
5457	VMXWriteVmcs64(uVmcsField, u64Val);
5458	}
5459
5460	/*
5461	* If the host CPU supports writing all VMCS fields, copy the guest read-only
5462	* VMCS fields, so the guest can VMREAD them without causing a VM-exit.
5463	*/
5464	if (g_HmMsrs.u.vmx.u64Misc & VMX_MISC_VMWRITE_ALL)
5465	{
5466	uint32_t const cShadowVmcsRoFields = pVM->hmr0.s.vmx.cShadowVmcsRoFields;
5467	for (uint32_t i = 0; i < cShadowVmcsRoFields; i++)
5468	{
5469	uint64_t u64Val;
5470	uint32_t const uVmcsField = pVM->hmr0.s.vmx.paShadowVmcsRoFields[i];
5471	IEMReadVmxVmcsField(pVmcsNstGst, uVmcsField, &u64Val);
5472	VMXWriteVmcs64(uVmcsField, u64Val);
5473	}
5474	}
5475
5476	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
5477	rc \|= hmR0VmxLoadVmcs(pVmcsInfo);
5478	}
5479
5480	ASMSetFlags(fEFlags);
5481	return rc;
5482	}
5483
5484
5485	/**
5486	* Copies the shadow VMCS to the nested-guest VMCS.
5487	*
5488	* @returns VBox status code.
5489	* @param pVCpu The cross context virtual CPU structure.
5490	* @param pVmcsInfo The VMCS info. object.
5491	*
5492	* @remarks Called with interrupts disabled.
5493	*/
5494	static int hmR0VmxCopyShadowToNstGstVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
5495	{
5496	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
5497	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5498	PVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5499
5500	int rc = hmR0VmxLoadShadowVmcs(pVmcsInfo);
5501	if (RT_SUCCESS(rc))
5502	{
5503	/*
5504	* Copy guest read/write fields from the shadow VMCS.
5505	* Guest read-only fields cannot be modified, so no need to copy them.
5506	*
5507	* We don't check for VMREAD failures here for performance reasons and
5508	* because they are not expected to fail, barring irrecoverable conditions
5509	* like hardware errors.
5510	*/
5511	uint32_t const cShadowVmcsFields = pVM->hmr0.s.vmx.cShadowVmcsFields;
5512	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
5513	{
5514	uint64_t u64Val;
5515	uint32_t const uVmcsField = pVM->hmr0.s.vmx.paShadowVmcsFields[i];
5516	VMXReadVmcs64(uVmcsField, &u64Val);
5517	IEMWriteVmxVmcsField(pVmcsNstGst, uVmcsField, u64Val);
5518	}
5519
5520	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
5521	rc \|= hmR0VmxLoadVmcs(pVmcsInfo);
5522	}
5523	return rc;
5524	}
5525
5526
5527	/**
5528	* Enables VMCS shadowing for the given VMCS info. object.
5529	*
5530	* @param pVmcsInfo The VMCS info. object.
5531	*
5532	* @remarks No-long-jump zone!!!
5533	*/
5534	static void hmR0VmxEnableVmcsShadowing(PVMXVMCSINFO pVmcsInfo)
5535	{
5536	uint32_t uProcCtls2 = pVmcsInfo->u32ProcCtls2;
5537	if (!(uProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING))
5538	{
5539	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
5540	uProcCtls2 \|= VMX_PROC_CTLS2_VMCS_SHADOWING;
5541	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, uProcCtls2); AssertRC(rc);
5542	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, pVmcsInfo->HCPhysShadowVmcs); AssertRC(rc);
5543	pVmcsInfo->u32ProcCtls2 = uProcCtls2;
5544	pVmcsInfo->u64VmcsLinkPtr = pVmcsInfo->HCPhysShadowVmcs;
5545	Log4Func(("Enabled\n"));
5546	}
5547	}
5548
5549
5550	/**
5551	* Disables VMCS shadowing for the given VMCS info. object.
5552	*
5553	* @param pVmcsInfo The VMCS info. object.
5554	*
5555	* @remarks No-long-jump zone!!!
5556	*/
5557	static void hmR0VmxDisableVmcsShadowing(PVMXVMCSINFO pVmcsInfo)
5558	{
5559	/*
5560	* We want all VMREAD and VMWRITE instructions to cause VM-exits, so we clear the
5561	* VMCS shadowing control. However, VM-entry requires the shadow VMCS indicator bit
5562	* to match the VMCS shadowing control if the VMCS link pointer is not NIL_RTHCPHYS.
5563	* Hence, we must also reset the VMCS link pointer to ensure VM-entry does not fail.
5564	*
5565	* See Intel spec. 26.2.1.1 "VM-Execution Control Fields".
5566	* See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
5567	*/
5568	uint32_t uProcCtls2 = pVmcsInfo->u32ProcCtls2;
5569	if (uProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
5570	{
5571	uProcCtls2 &= ~VMX_PROC_CTLS2_VMCS_SHADOWING;
5572	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, uProcCtls2); AssertRC(rc);
5573	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS); AssertRC(rc);
5574	pVmcsInfo->u32ProcCtls2 = uProcCtls2;
5575	pVmcsInfo->u64VmcsLinkPtr = NIL_RTHCPHYS;
5576	Log4Func(("Disabled\n"));
5577	}
5578	}
5579	#endif
5580
5581
5582	/**
5583	* Exports the guest hardware-virtualization state.
5584	*
5585	* @returns VBox status code.
5586	* @param pVCpu The cross context virtual CPU structure.
5587	* @param pVmxTransient The VMX-transient structure.
5588	*
5589	* @remarks No-long-jump zone!!!
5590	*/
5591	static int hmR0VmxExportGuestHwvirtState(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5592	{
5593	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_HWVIRT)
5594	{
5595	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5596	/*
5597	* Check if the VMX feature is exposed to the guest and if the host CPU supports
5598	* VMCS shadowing.
5599	*/
5600	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUseVmcsShadowing)
5601	{
5602	/*
5603	* If the nested hypervisor has loaded a current VMCS and is in VMX root mode,
5604	* copy the nested hypervisor's current VMCS into the shadow VMCS and enable
5605	* VMCS shadowing to skip intercepting some or all VMREAD/VMWRITE VM-exits.
5606	*
5607	* We check for VMX root mode here in case the guest executes VMXOFF without
5608	* clearing the current VMCS pointer and our VMXOFF instruction emulation does
5609	* not clear the current VMCS pointer.
5610	*/
5611	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5612	if ( CPUMIsGuestInVmxRootMode(&pVCpu->cpum.GstCtx)
5613	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx)
5614	&& CPUMIsGuestVmxCurrentVmcsValid(&pVCpu->cpum.GstCtx))
5615	{
5616	/* Paranoia. */
5617	Assert(!pVmxTransient->fIsNestedGuest);
5618
5619	/*
5620	* For performance reasons, also check if the nested hypervisor's current VMCS
5621	* was newly loaded or modified before copying it to the shadow VMCS.
5622	*/
5623	if (!pVCpu->hm.s.vmx.fCopiedNstGstToShadowVmcs)
5624	{
5625	int rc = hmR0VmxCopyNstGstToShadowVmcs(pVCpu, pVmcsInfo);
5626	AssertRCReturn(rc, rc);
5627	pVCpu->hm.s.vmx.fCopiedNstGstToShadowVmcs = true;
5628	}
5629	hmR0VmxEnableVmcsShadowing(pVmcsInfo);
5630	}
5631	else
5632	hmR0VmxDisableVmcsShadowing(pVmcsInfo);
5633	}
5634	#else
5635	NOREF(pVmxTransient);
5636	#endif
5637	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_HWVIRT);
5638	}
5639	return VINF_SUCCESS;
5640	}
5641
5642
5643	/**
5644	* Exports the guest CR0 control register into the guest-state area in the VMCS.
5645	*
5646	* The guest FPU state is always pre-loaded hence we don't need to bother about
5647	* sharing FPU related CR0 bits between the guest and host.
5648	*
5649	* @returns VBox status code.
5650	* @param pVCpu The cross context virtual CPU structure.
5651	* @param pVmxTransient The VMX-transient structure.
5652	*
5653	* @remarks No-long-jump zone!!!
5654	*/
5655	static int hmR0VmxExportGuestCR0(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5656	{
5657	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR0)
5658	{
5659	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5660	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5661
5662	uint64_t fSetCr0 = g_HmMsrs.u.vmx.u64Cr0Fixed0;
5663	uint64_t const fZapCr0 = g_HmMsrs.u.vmx.u64Cr0Fixed1;
5664	if (pVM->hmr0.s.vmx.fUnrestrictedGuest)
5665	fSetCr0 &= ~(uint64_t)(X86_CR0_PE \| X86_CR0_PG);
5666	else
5667	Assert((fSetCr0 & (X86_CR0_PE \| X86_CR0_PG)) == (X86_CR0_PE \| X86_CR0_PG));
5668
5669	if (!pVmxTransient->fIsNestedGuest)
5670	{
5671	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
5672	uint64_t u64GuestCr0 = pVCpu->cpum.GstCtx.cr0;
5673	uint64_t const u64ShadowCr0 = u64GuestCr0;
5674	Assert(!RT_HI_U32(u64GuestCr0));
5675
5676	/*
5677	* Setup VT-x's view of the guest CR0.
5678	*/
5679	uint32_t uProcCtls = pVmcsInfo->u32ProcCtls;
5680	if (pVM->hmr0.s.fNestedPaging)
5681	{
5682	if (CPUMIsGuestPagingEnabled(pVCpu))
5683	{
5684	/* The guest has paging enabled, let it access CR3 without causing a VM-exit if supported. */
5685	uProcCtls &= ~( VMX_PROC_CTLS_CR3_LOAD_EXIT
5686	\| VMX_PROC_CTLS_CR3_STORE_EXIT);
5687	}
5688	else
5689	{
5690	/* The guest doesn't have paging enabled, make CR3 access cause a VM-exit to update our shadow. */
5691	uProcCtls \|= VMX_PROC_CTLS_CR3_LOAD_EXIT
5692	\| VMX_PROC_CTLS_CR3_STORE_EXIT;
5693	}
5694
5695	/* If we have unrestricted guest execution, we never have to intercept CR3 reads. */
5696	if (pVM->hmr0.s.vmx.fUnrestrictedGuest)
5697	uProcCtls &= ~VMX_PROC_CTLS_CR3_STORE_EXIT;
5698	}
5699	else
5700	{
5701	/* Guest CPL 0 writes to its read-only pages should cause a #PF VM-exit. */
5702	u64GuestCr0 \|= X86_CR0_WP;
5703	}
5704
5705	/*
5706	* Guest FPU bits.
5707	*
5708	* Since we pre-load the guest FPU always before VM-entry there is no need to track lazy state
5709	* using CR0.TS.
5710	*
5711	* Intel spec. 23.8 "Restrictions on VMX operation" mentions that CR0.NE bit must always be
5712	* set on the first CPUs to support VT-x and no mention of with regards to UX in VM-entry checks.
5713	*/
5714	u64GuestCr0 \|= X86_CR0_NE;
5715
5716	/* If CR0.NE isn't set, we need to intercept #MF exceptions and report them to the guest differently. */
5717	bool const fInterceptMF = !(u64ShadowCr0 & X86_CR0_NE);
5718
5719	/*
5720	* Update exception intercepts.
5721	*/
5722	uint32_t uXcptBitmap = pVmcsInfo->u32XcptBitmap;
5723	if (pVmcsInfo->pShared->RealMode.fRealOnV86Active)
5724	{
5725	Assert(PDMVmmDevHeapIsEnabled(pVM));
5726	Assert(pVM->hm.s.vmx.pRealModeTSS);
5727	uXcptBitmap \|= HMVMX_REAL_MODE_XCPT_MASK;
5728	}
5729	else
5730	{
5731	/* For now, cleared here as mode-switches can happen outside HM/VT-x. See @bugref{7626#c11}. */
5732	uXcptBitmap &= ~HMVMX_REAL_MODE_XCPT_MASK;
5733	if (fInterceptMF)
5734	uXcptBitmap \|= RT_BIT(X86_XCPT_MF);
5735	}
5736
5737	/* Additional intercepts for debugging, define these yourself explicitly. */
5738	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
5739	uXcptBitmap \|= 0
5740	\| RT_BIT(X86_XCPT_BP)
5741	\| RT_BIT(X86_XCPT_DE)
5742	\| RT_BIT(X86_XCPT_NM)
5743	\| RT_BIT(X86_XCPT_TS)
5744	\| RT_BIT(X86_XCPT_UD)
5745	\| RT_BIT(X86_XCPT_NP)
5746	\| RT_BIT(X86_XCPT_SS)
5747	\| RT_BIT(X86_XCPT_GP)
5748	\| RT_BIT(X86_XCPT_PF)
5749	\| RT_BIT(X86_XCPT_MF)
5750	;
5751	#elif defined(HMVMX_ALWAYS_TRAP_PF)
5752	uXcptBitmap \|= RT_BIT(X86_XCPT_PF);
5753	#endif
5754	if (pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv)
5755	uXcptBitmap \|= RT_BIT(X86_XCPT_GP);
5756	Assert(pVM->hmr0.s.fNestedPaging \|\| (uXcptBitmap & RT_BIT(X86_XCPT_PF)));
5757
5758	/* Apply the hardware specified CR0 fixed bits and enable caching. */
5759	u64GuestCr0 \|= fSetCr0;
5760	u64GuestCr0 &= fZapCr0;
5761	u64GuestCr0 &= ~(uint64_t)(X86_CR0_CD \| X86_CR0_NW);
5762
5763	/* Commit the CR0 and related fields to the guest VMCS. */
5764	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR0, u64GuestCr0); AssertRC(rc);
5765	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, u64ShadowCr0); AssertRC(rc);
5766	if (uProcCtls != pVmcsInfo->u32ProcCtls)
5767	{
5768	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, uProcCtls);
5769	AssertRC(rc);
5770	}
5771	if (uXcptBitmap != pVmcsInfo->u32XcptBitmap)
5772	{
5773	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
5774	AssertRC(rc);
5775	}
5776
5777	/* Update our caches. */
5778	pVmcsInfo->u32ProcCtls = uProcCtls;
5779	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
5780
5781	Log4Func(("cr0=%#RX64 shadow=%#RX64 set=%#RX64 zap=%#RX64\n", u64GuestCr0, u64ShadowCr0, fSetCr0, fZapCr0));
5782	}
5783	else
5784	{
5785	/*
5786	* With nested-guests, we may have extended the guest/host mask here since we
5787	* merged in the outer guest's mask. Thus, the merged mask can include more bits
5788	* (to read from the nested-guest CR0 read-shadow) than the nested hypervisor
5789	* originally supplied. We must copy those bits from the nested-guest CR0 into
5790	* the nested-guest CR0 read-shadow.
5791	*/
5792	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
5793	uint64_t u64GuestCr0 = pVCpu->cpum.GstCtx.cr0;
5794	uint64_t const u64ShadowCr0 = CPUMGetGuestVmxMaskedCr0(&pVCpu->cpum.GstCtx, pVmcsInfo->u64Cr0Mask);
5795	Assert(!RT_HI_U32(u64GuestCr0));
5796	Assert(u64GuestCr0 & X86_CR0_NE);
5797
5798	/* Apply the hardware specified CR0 fixed bits and enable caching. */
5799	u64GuestCr0 \|= fSetCr0;
5800	u64GuestCr0 &= fZapCr0;
5801	u64GuestCr0 &= ~(uint64_t)(X86_CR0_CD \| X86_CR0_NW);
5802
5803	/* Commit the CR0 and CR0 read-shadow to the nested-guest VMCS. */
5804	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR0, u64GuestCr0); AssertRC(rc);
5805	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, u64ShadowCr0); AssertRC(rc);
5806
5807	Log4Func(("cr0=%#RX64 shadow=%#RX64 (set=%#RX64 zap=%#RX64)\n", u64GuestCr0, u64ShadowCr0, fSetCr0, fZapCr0));
5808	}
5809
5810	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR0);
5811	}
5812
5813	return VINF_SUCCESS;
5814	}
5815
5816
5817	/**
5818	* Exports the guest control registers (CR3, CR4) into the guest-state area
5819	* in the VMCS.
5820	*
5821	* @returns VBox strict status code.
5822	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
5823	* without unrestricted guest access and the VMMDev is not presently
5824	* mapped (e.g. EFI32).
5825	*
5826	* @param pVCpu The cross context virtual CPU structure.
5827	* @param pVmxTransient The VMX-transient structure.
5828	*
5829	* @remarks No-long-jump zone!!!
5830	*/
5831	static VBOXSTRICTRC hmR0VmxExportGuestCR3AndCR4(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5832	{
5833	int rc = VINF_SUCCESS;
5834	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5835
5836	/*
5837	* Guest CR2.
5838	* It's always loaded in the assembler code. Nothing to do here.
5839	*/
5840
5841	/*
5842	* Guest CR3.
5843	*/
5844	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR3)
5845	{
5846	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR3);
5847
5848	if (pVM->hmr0.s.fNestedPaging)
5849	{
5850	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5851	pVmcsInfo->HCPhysEPTP = PGMGetHyperCR3(pVCpu);
5852
5853	/* Validate. See Intel spec. 28.2.2 "EPT Translation Mechanism" and 24.6.11 "Extended-Page-Table Pointer (EPTP)" */
5854	Assert(pVmcsInfo->HCPhysEPTP != NIL_RTHCPHYS);
5855	Assert(!(pVmcsInfo->HCPhysEPTP & UINT64_C(0xfff0000000000000)));
5856	Assert(!(pVmcsInfo->HCPhysEPTP & 0xfff));
5857
5858	/* VMX_EPT_MEMTYPE_WB support is already checked in hmR0VmxSetupTaggedTlb(). */
5859	pVmcsInfo->HCPhysEPTP \|= VMX_EPT_MEMTYPE_WB
5860	\| (VMX_EPT_PAGE_WALK_LENGTH_DEFAULT << VMX_EPT_PAGE_WALK_LENGTH_SHIFT);
5861
5862	/* Validate. See Intel spec. 26.2.1 "Checks on VMX Controls" */
5863	AssertMsg( ((pVmcsInfo->HCPhysEPTP >> 3) & 0x07) == 3 /* Bits 3:5 (EPT page walk length - 1) must be 3. */
5864	&& ((pVmcsInfo->HCPhysEPTP >> 7) & 0x1f) == 0, /* Bits 7:11 MBZ. */
5865	("EPTP %#RX64\n", pVmcsInfo->HCPhysEPTP));
5866	AssertMsg( !((pVmcsInfo->HCPhysEPTP >> 6) & 0x01) /* Bit 6 (EPT accessed & dirty bit). */
5867	\|\| (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EPT_ACCESS_DIRTY),
5868	("EPTP accessed/dirty bit not supported by CPU but set %#RX64\n", pVmcsInfo->HCPhysEPTP));
5869
5870	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, pVmcsInfo->HCPhysEPTP);
5871	AssertRC(rc);
5872
5873	uint64_t u64GuestCr3;
5874	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5875	if ( pVM->hmr0.s.vmx.fUnrestrictedGuest
5876	\|\| CPUMIsGuestPagingEnabledEx(pCtx))
5877	{
5878	/* If the guest is in PAE mode, pass the PDPEs to VT-x using the VMCS fields. */
5879	if (CPUMIsGuestInPAEModeEx(pCtx))
5880	{
5881	rc = PGMGstGetPaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
5882	AssertRC(rc);
5883	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, pVCpu->hm.s.aPdpes[0].u); AssertRC(rc);
5884	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, pVCpu->hm.s.aPdpes[1].u); AssertRC(rc);
5885	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, pVCpu->hm.s.aPdpes[2].u); AssertRC(rc);
5886	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, pVCpu->hm.s.aPdpes[3].u); AssertRC(rc);
5887	}
5888
5889	/*
5890	* The guest's view of its CR3 is unblemished with nested paging when the
5891	* guest is using paging or we have unrestricted guest execution to handle
5892	* the guest when it's not using paging.
5893	*/
5894	u64GuestCr3 = pCtx->cr3;
5895	}
5896	else
5897	{
5898	/*
5899	* The guest is not using paging, but the CPU (VT-x) has to. While the guest
5900	* thinks it accesses physical memory directly, we use our identity-mapped
5901	* page table to map guest-linear to guest-physical addresses. EPT takes care
5902	* of translating it to host-physical addresses.
5903	*/
5904	RTGCPHYS GCPhys;
5905	Assert(pVM->hm.s.vmx.pNonPagingModeEPTPageTable);
5906
5907	/* We obtain it here every time as the guest could have relocated this PCI region. */
5908	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys);
5909	if (RT_SUCCESS(rc))
5910	{ /* likely */ }
5911	else if (rc == VERR_PDM_DEV_HEAP_R3_TO_GCPHYS)
5912	{
5913	Log4Func(("VERR_PDM_DEV_HEAP_R3_TO_GCPHYS -> VINF_EM_RESCHEDULE_REM\n"));
5914	return VINF_EM_RESCHEDULE_REM; /* We cannot execute now, switch to REM/IEM till the guest maps in VMMDev. */
5915	}
5916	else
5917	AssertMsgFailedReturn(("%Rrc\n", rc), rc);
5918
5919	u64GuestCr3 = GCPhys;
5920	}
5921
5922	Log4Func(("guest_cr3=%#RX64 (GstN)\n", u64GuestCr3));
5923	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR3, u64GuestCr3);
5924	AssertRC(rc);
5925	}
5926	else
5927	{
5928	Assert(!pVmxTransient->fIsNestedGuest);
5929	/* Non-nested paging case, just use the hypervisor's CR3. */
5930	RTHCPHYS const HCPhysGuestCr3 = PGMGetHyperCR3(pVCpu);
5931
5932	Log4Func(("guest_cr3=%#RX64 (HstN)\n", HCPhysGuestCr3));
5933	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR3, HCPhysGuestCr3);
5934	AssertRC(rc);
5935	}
5936
5937	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR3);
5938	}
5939
5940	/*
5941	* Guest CR4.
5942	* ASSUMES this is done everytime we get in from ring-3! (XCR0)
5943	*/
5944	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR4)
5945	{
5946	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5947	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5948
5949	uint64_t const fSetCr4 = g_HmMsrs.u.vmx.u64Cr4Fixed0;
5950	uint64_t const fZapCr4 = g_HmMsrs.u.vmx.u64Cr4Fixed1;
5951
5952	/*
5953	* With nested-guests, we may have extended the guest/host mask here (since we
5954	* merged in the outer guest's mask, see hmR0VmxMergeVmcsNested). This means, the
5955	* mask can include more bits (to read from the nested-guest CR4 read-shadow) than
5956	* the nested hypervisor originally supplied. Thus, we should, in essence, copy
5957	* those bits from the nested-guest CR4 into the nested-guest CR4 read-shadow.
5958	*/
5959	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
5960	uint64_t u64GuestCr4 = pCtx->cr4;
5961	uint64_t const u64ShadowCr4 = !pVmxTransient->fIsNestedGuest
5962	? pCtx->cr4
5963	: CPUMGetGuestVmxMaskedCr4(pCtx, pVmcsInfo->u64Cr4Mask);
5964	Assert(!RT_HI_U32(u64GuestCr4));
5965
5966	/*
5967	* Setup VT-x's view of the guest CR4.
5968	*
5969	* If we're emulating real-mode using virtual-8086 mode, we want to redirect software
5970	* interrupts to the 8086 program interrupt handler. Clear the VME bit (the interrupt
5971	* redirection bitmap is already all 0, see hmR3InitFinalizeR0())
5972	*
5973	* See Intel spec. 20.2 "Software Interrupt Handling Methods While in Virtual-8086 Mode".
5974	*/
5975	if (pVmcsInfo->pShared->RealMode.fRealOnV86Active)
5976	{
5977	Assert(pVM->hm.s.vmx.pRealModeTSS);
5978	Assert(PDMVmmDevHeapIsEnabled(pVM));
5979	u64GuestCr4 &= ~(uint64_t)X86_CR4_VME;
5980	}
5981
5982	if (pVM->hmr0.s.fNestedPaging)
5983	{
5984	if ( !CPUMIsGuestPagingEnabledEx(pCtx)
5985	&& !pVM->hmr0.s.vmx.fUnrestrictedGuest)
5986	{
5987	/* We use 4 MB pages in our identity mapping page table when the guest doesn't have paging. */
5988	u64GuestCr4 \|= X86_CR4_PSE;
5989	/* Our identity mapping is a 32-bit page directory. */
5990	u64GuestCr4 &= ~(uint64_t)X86_CR4_PAE;
5991	}
5992	/* else use guest CR4.*/
5993	}
5994	else
5995	{
5996	Assert(!pVmxTransient->fIsNestedGuest);
5997
5998	/*
5999	* The shadow paging modes and guest paging modes are different, the shadow is in accordance with the host
6000	* paging mode and thus we need to adjust VT-x's view of CR4 depending on our shadow page tables.
6001	*/
6002	switch (pVCpu->hm.s.enmShadowMode)
6003	{
6004	case PGMMODE_REAL: /* Real-mode. */
6005	case PGMMODE_PROTECTED: /* Protected mode without paging. */
6006	case PGMMODE_32_BIT: /* 32-bit paging. */
6007	{
6008	u64GuestCr4 &= ~(uint64_t)X86_CR4_PAE;
6009	break;
6010	}
6011
6012	case PGMMODE_PAE: /* PAE paging. */
6013	case PGMMODE_PAE_NX: /* PAE paging with NX. */
6014	{
6015	u64GuestCr4 \|= X86_CR4_PAE;
6016	break;
6017	}
6018
6019	case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
6020	case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
6021	{
6022	#ifdef VBOX_WITH_64_BITS_GUESTS
6023	/* For our assumption in hmR0VmxShouldSwapEferMsr. */
6024	Assert(u64GuestCr4 & X86_CR4_PAE);
6025	break;
6026	#endif
6027	}
6028	default:
6029	AssertFailed();
6030	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
6031	}
6032	}
6033
6034	/* Apply the hardware specified CR4 fixed bits (mainly CR4.VMXE). */
6035	u64GuestCr4 \|= fSetCr4;
6036	u64GuestCr4 &= fZapCr4;
6037
6038	/* Commit the CR4 and CR4 read-shadow to the guest VMCS. */
6039	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR4, u64GuestCr4); AssertRC(rc);
6040	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, u64ShadowCr4); AssertRC(rc);
6041
6042	/* Whether to save/load/restore XCR0 during world switch depends on CR4.OSXSAVE and host+guest XCR0. */
6043	bool const fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
6044	if (fLoadSaveGuestXcr0 != pVCpu->hmr0.s.fLoadSaveGuestXcr0)
6045	{
6046	pVCpu->hmr0.s.fLoadSaveGuestXcr0 = fLoadSaveGuestXcr0;
6047	hmR0VmxUpdateStartVmFunction(pVCpu);
6048	}
6049
6050	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR4);
6051
6052	Log4Func(("cr4=%#RX64 shadow=%#RX64 (set=%#RX64 zap=%#RX64)\n", u64GuestCr4, u64ShadowCr4, fSetCr4, fZapCr4));
6053	}
6054	return rc;
6055	}
6056
6057
6058	/**
6059	* Exports the guest debug registers into the guest-state area in the VMCS.
6060	* The guest debug bits are partially shared with the host (e.g. DR6, DR0-3).
6061	*
6062	* This also sets up whether \#DB and MOV DRx accesses cause VM-exits.
6063	*
6064	* @returns VBox status code.
6065	* @param pVCpu The cross context virtual CPU structure.
6066	* @param pVmxTransient The VMX-transient structure.
6067	*
6068	* @remarks No-long-jump zone!!!
6069	*/
6070	static int hmR0VmxExportSharedDebugState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
6071	{
6072	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
6073
6074	/** @todo NSTVMX: Figure out what we want to do with nested-guest instruction
6075	* stepping. */
6076	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6077	if (pVmxTransient->fIsNestedGuest)
6078	{
6079	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, CPUMGetGuestDR7(pVCpu));
6080	AssertRC(rc);
6081
6082	/*
6083	* We don't want to always intercept MOV DRx for nested-guests as it causes
6084	* problems when the nested hypervisor isn't intercepting them, see @bugref{10080}.
6085	* Instead, they are strictly only requested when the nested hypervisor intercepts
6086	* them -- handled while merging VMCS controls.
6087	*
6088	* If neither the outer nor the nested-hypervisor is intercepting MOV DRx,
6089	* then the nested-guest debug state should be actively loaded on the host so that
6090	* nested-guest reads its own debug registers without causing VM-exits.
6091	*/
6092	if ( !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MOV_DR_EXIT)
6093	&& !CPUMIsGuestDebugStateActive(pVCpu))
6094	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
6095	return VINF_SUCCESS;
6096	}
6097
6098	#ifdef VBOX_STRICT
6099	/* Validate. Intel spec. 26.3.1.1 "Checks on Guest Controls Registers, Debug Registers, MSRs" */
6100	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
6101	{
6102	/* Validate. Intel spec. 17.2 "Debug Registers", recompiler paranoia checks. */
6103	Assert((pVCpu->cpum.GstCtx.dr[7] & (X86_DR7_MBZ_MASK \| X86_DR7_RAZ_MASK)) == 0);
6104	Assert((pVCpu->cpum.GstCtx.dr[7] & X86_DR7_RA1_MASK) == X86_DR7_RA1_MASK);
6105	}
6106	#endif
6107
6108	bool fSteppingDB = false;
6109	bool fInterceptMovDRx = false;
6110	uint32_t uProcCtls = pVmcsInfo->u32ProcCtls;
6111	if (pVCpu->hm.s.fSingleInstruction)
6112	{
6113	/* If the CPU supports the monitor trap flag, use it for single stepping in DBGF and avoid intercepting #DB. */
6114	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_MONITOR_TRAP_FLAG)
6115	{
6116	uProcCtls \|= VMX_PROC_CTLS_MONITOR_TRAP_FLAG;
6117	Assert(fSteppingDB == false);
6118	}
6119	else
6120	{
6121	pVCpu->cpum.GstCtx.eflags.u32 \|= X86_EFL_TF;
6122	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_GUEST_RFLAGS;
6123	pVCpu->hmr0.s.fClearTrapFlag = true;
6124	fSteppingDB = true;
6125	}
6126	}
6127
6128	uint64_t u64GuestDr7;
6129	if ( fSteppingDB
6130	\|\| (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK))
6131	{
6132	/*
6133	* Use the combined guest and host DRx values found in the hypervisor register set
6134	* because the hypervisor debugger has breakpoints active or someone is single stepping
6135	* on the host side without a monitor trap flag.
6136	*
6137	* Note! DBGF expects a clean DR6 state before executing guest code.
6138	*/
6139	if (!CPUMIsHyperDebugStateActive(pVCpu))
6140	{
6141	CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
6142	Assert(CPUMIsHyperDebugStateActive(pVCpu));
6143	Assert(!CPUMIsGuestDebugStateActive(pVCpu));
6144	}
6145
6146	/* Update DR7 with the hypervisor value (other DRx registers are handled by CPUM one way or another). */
6147	u64GuestDr7 = CPUMGetHyperDR7(pVCpu);
6148	pVCpu->hmr0.s.fUsingHyperDR7 = true;
6149	fInterceptMovDRx = true;
6150	}
6151	else
6152	{
6153	/*
6154	* If the guest has enabled debug registers, we need to load them prior to
6155	* executing guest code so they'll trigger at the right time.
6156	*/
6157	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_DR7);
6158	if (pVCpu->cpum.GstCtx.dr[7] & (X86_DR7_ENABLED_MASK \| X86_DR7_GD))
6159	{
6160	if (!CPUMIsGuestDebugStateActive(pVCpu))
6161	{
6162	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
6163	Assert(CPUMIsGuestDebugStateActive(pVCpu));
6164	Assert(!CPUMIsHyperDebugStateActive(pVCpu));
6165	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
6166	}
6167	Assert(!fInterceptMovDRx);
6168	}
6169	else if (!CPUMIsGuestDebugStateActive(pVCpu))
6170	{
6171	/*
6172	* If no debugging enabled, we'll lazy load DR0-3. Unlike on AMD-V, we
6173	* must intercept #DB in order to maintain a correct DR6 guest value, and
6174	* because we need to intercept it to prevent nested #DBs from hanging the
6175	* CPU, we end up always having to intercept it. See hmR0VmxSetupVmcsXcptBitmap().
6176	*/
6177	fInterceptMovDRx = true;
6178	}
6179
6180	/* Update DR7 with the actual guest value. */
6181	u64GuestDr7 = pVCpu->cpum.GstCtx.dr[7];
6182	pVCpu->hmr0.s.fUsingHyperDR7 = false;
6183	}
6184
6185	if (fInterceptMovDRx)
6186	uProcCtls \|= VMX_PROC_CTLS_MOV_DR_EXIT;
6187	else
6188	uProcCtls &= ~VMX_PROC_CTLS_MOV_DR_EXIT;
6189
6190	/*
6191	* Update the processor-based VM-execution controls with the MOV-DRx intercepts and the
6192	* monitor-trap flag and update our cache.
6193	*/
6194	if (uProcCtls != pVmcsInfo->u32ProcCtls)
6195	{
6196	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, uProcCtls);
6197	AssertRC(rc);
6198	pVmcsInfo->u32ProcCtls = uProcCtls;
6199	}
6200
6201	/*
6202	* Update guest DR7.
6203	*/
6204	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, u64GuestDr7);
6205	AssertRC(rc);
6206
6207	/*
6208	* If we have forced EFLAGS.TF to be set because we're single-stepping in the hypervisor debugger,
6209	* we need to clear interrupt inhibition if any as otherwise it causes a VM-entry failure.
6210	*
6211	* See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
6212	*/
6213	if (fSteppingDB)
6214	{
6215	Assert(pVCpu->hm.s.fSingleInstruction);
6216	Assert(pVCpu->cpum.GstCtx.eflags.Bits.u1TF);
6217
6218	uint32_t fIntrState = 0;
6219	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
6220	AssertRC(rc);
6221
6222	if (fIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
6223	{
6224	fIntrState &= ~(VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
6225	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
6226	AssertRC(rc);
6227	}
6228	}
6229
6230	return VINF_SUCCESS;
6231	}
6232
6233
6234	#ifdef VBOX_STRICT
6235	/**
6236	* Strict function to validate segment registers.
6237	*
6238	* @param pVCpu The cross context virtual CPU structure.
6239	* @param pVmcsInfo The VMCS info. object.
6240	*
6241	* @remarks Will import guest CR0 on strict builds during validation of
6242	* segments.
6243	*/
6244	static void hmR0VmxValidateSegmentRegs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
6245	{
6246	/*
6247	* Validate segment registers. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
6248	*
6249	* The reason we check for attribute value 0 in this function and not just the unusable bit is
6250	* because hmR0VmxExportGuestSegReg() only updates the VMCS' copy of the value with the
6251	* unusable bit and doesn't change the guest-context value.
6252	*/
6253	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6254	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6255	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR0);
6256	if ( !pVM->hmr0.s.vmx.fUnrestrictedGuest
6257	&& ( !CPUMIsGuestInRealModeEx(pCtx)
6258	&& !CPUMIsGuestInV86ModeEx(pCtx)))
6259	{
6260	/* Protected mode checks */
6261	/* CS */
6262	Assert(pCtx->cs.Attr.n.u1Present);
6263	Assert(!(pCtx->cs.Attr.u & 0xf00));
6264	Assert(!(pCtx->cs.Attr.u & 0xfffe0000));
6265	Assert( (pCtx->cs.u32Limit & 0xfff) == 0xfff
6266	\|\| !(pCtx->cs.Attr.n.u1Granularity));
6267	Assert( !(pCtx->cs.u32Limit & 0xfff00000)
6268	\|\| (pCtx->cs.Attr.n.u1Granularity));
6269	/* CS cannot be loaded with NULL in protected mode. */
6270	Assert(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE)); /** @todo is this really true even for 64-bit CS? */
6271	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
6272	Assert(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl);
6273	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
6274	Assert(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl);
6275	else
6276	AssertMsgFailed(("Invalid CS Type %#x\n", pCtx->cs.Attr.n.u2Dpl));
6277	/* SS */
6278	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
6279	Assert(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL));
6280	if ( !(pCtx->cr0 & X86_CR0_PE)
6281	\|\| pCtx->cs.Attr.n.u4Type == 3)
6282	{
6283	Assert(!pCtx->ss.Attr.n.u2Dpl);
6284	}
6285	if (pCtx->ss.Attr.u && !(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
6286	{
6287	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
6288	Assert(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7);
6289	Assert(pCtx->ss.Attr.n.u1Present);
6290	Assert(!(pCtx->ss.Attr.u & 0xf00));
6291	Assert(!(pCtx->ss.Attr.u & 0xfffe0000));
6292	Assert( (pCtx->ss.u32Limit & 0xfff) == 0xfff
6293	\|\| !(pCtx->ss.Attr.n.u1Granularity));
6294	Assert( !(pCtx->ss.u32Limit & 0xfff00000)
6295	\|\| (pCtx->ss.Attr.n.u1Granularity));
6296	}
6297	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxExportGuestSegReg(). */
6298	if (pCtx->ds.Attr.u && !(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
6299	{
6300	Assert(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6301	Assert(pCtx->ds.Attr.n.u1Present);
6302	Assert(pCtx->ds.Attr.n.u4Type > 11 \|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL));
6303	Assert(!(pCtx->ds.Attr.u & 0xf00));
6304	Assert(!(pCtx->ds.Attr.u & 0xfffe0000));
6305	Assert( (pCtx->ds.u32Limit & 0xfff) == 0xfff
6306	\|\| !(pCtx->ds.Attr.n.u1Granularity));
6307	Assert( !(pCtx->ds.u32Limit & 0xfff00000)
6308	\|\| (pCtx->ds.Attr.n.u1Granularity));
6309	Assert( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6310	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ));
6311	}
6312	if (pCtx->es.Attr.u && !(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
6313	{
6314	Assert(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6315	Assert(pCtx->es.Attr.n.u1Present);
6316	Assert(pCtx->es.Attr.n.u4Type > 11 \|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL));
6317	Assert(!(pCtx->es.Attr.u & 0xf00));
6318	Assert(!(pCtx->es.Attr.u & 0xfffe0000));
6319	Assert( (pCtx->es.u32Limit & 0xfff) == 0xfff
6320	\|\| !(pCtx->es.Attr.n.u1Granularity));
6321	Assert( !(pCtx->es.u32Limit & 0xfff00000)
6322	\|\| (pCtx->es.Attr.n.u1Granularity));
6323	Assert( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6324	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ));
6325	}
6326	if (pCtx->fs.Attr.u && !(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
6327	{
6328	Assert(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6329	Assert(pCtx->fs.Attr.n.u1Present);
6330	Assert(pCtx->fs.Attr.n.u4Type > 11 \|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL));
6331	Assert(!(pCtx->fs.Attr.u & 0xf00));
6332	Assert(!(pCtx->fs.Attr.u & 0xfffe0000));
6333	Assert( (pCtx->fs.u32Limit & 0xfff) == 0xfff
6334	\|\| !(pCtx->fs.Attr.n.u1Granularity));
6335	Assert( !(pCtx->fs.u32Limit & 0xfff00000)
6336	\|\| (pCtx->fs.Attr.n.u1Granularity));
6337	Assert( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6338	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ));
6339	}
6340	if (pCtx->gs.Attr.u && !(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
6341	{
6342	Assert(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6343	Assert(pCtx->gs.Attr.n.u1Present);
6344	Assert(pCtx->gs.Attr.n.u4Type > 11 \|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL));
6345	Assert(!(pCtx->gs.Attr.u & 0xf00));
6346	Assert(!(pCtx->gs.Attr.u & 0xfffe0000));
6347	Assert( (pCtx->gs.u32Limit & 0xfff) == 0xfff
6348	\|\| !(pCtx->gs.Attr.n.u1Granularity));
6349	Assert( !(pCtx->gs.u32Limit & 0xfff00000)
6350	\|\| (pCtx->gs.Attr.n.u1Granularity));
6351	Assert( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6352	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ));
6353	}
6354	/* 64-bit capable CPUs. */
6355	Assert(!RT_HI_U32(pCtx->cs.u64Base));
6356	Assert(!pCtx->ss.Attr.u \|\| !RT_HI_U32(pCtx->ss.u64Base));
6357	Assert(!pCtx->ds.Attr.u \|\| !RT_HI_U32(pCtx->ds.u64Base));
6358	Assert(!pCtx->es.Attr.u \|\| !RT_HI_U32(pCtx->es.u64Base));
6359	}
6360	else if ( CPUMIsGuestInV86ModeEx(pCtx)
6361	\|\| ( CPUMIsGuestInRealModeEx(pCtx)
6362	&& !pVM->hmr0.s.vmx.fUnrestrictedGuest))
6363	{
6364	/* Real and v86 mode checks. */
6365	/* hmR0VmxExportGuestSegReg() writes the modified in VMCS. We want what we're feeding to VT-x. */
6366	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
6367	if (pVmcsInfo->pShared->RealMode.fRealOnV86Active)
6368	{
6369	u32CSAttr = 0xf3; u32SSAttr = 0xf3; u32DSAttr = 0xf3;
6370	u32ESAttr = 0xf3; u32FSAttr = 0xf3; u32GSAttr = 0xf3;
6371	}
6372	else
6373	{
6374	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u; u32DSAttr = pCtx->ds.Attr.u;
6375	u32ESAttr = pCtx->es.Attr.u; u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
6376	}
6377
6378	/* CS */
6379	AssertMsg((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), ("CS base %#x %#x\n", pCtx->cs.u64Base, pCtx->cs.Sel));
6380	Assert(pCtx->cs.u32Limit == 0xffff);
6381	Assert(u32CSAttr == 0xf3);
6382	/* SS */
6383	Assert(pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4);
6384	Assert(pCtx->ss.u32Limit == 0xffff);
6385	Assert(u32SSAttr == 0xf3);
6386	/* DS */
6387	Assert(pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4);
6388	Assert(pCtx->ds.u32Limit == 0xffff);
6389	Assert(u32DSAttr == 0xf3);
6390	/* ES */
6391	Assert(pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4);
6392	Assert(pCtx->es.u32Limit == 0xffff);
6393	Assert(u32ESAttr == 0xf3);
6394	/* FS */
6395	Assert(pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4);
6396	Assert(pCtx->fs.u32Limit == 0xffff);
6397	Assert(u32FSAttr == 0xf3);
6398	/* GS */
6399	Assert(pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4);
6400	Assert(pCtx->gs.u32Limit == 0xffff);
6401	Assert(u32GSAttr == 0xf3);
6402	/* 64-bit capable CPUs. */
6403	Assert(!RT_HI_U32(pCtx->cs.u64Base));
6404	Assert(!u32SSAttr \|\| !RT_HI_U32(pCtx->ss.u64Base));
6405	Assert(!u32DSAttr \|\| !RT_HI_U32(pCtx->ds.u64Base));
6406	Assert(!u32ESAttr \|\| !RT_HI_U32(pCtx->es.u64Base));
6407	}
6408	}
6409	#endif /* VBOX_STRICT */
6410
6411
6412	/**
6413	* Exports a guest segment register into the guest-state area in the VMCS.
6414	*
6415	* @returns VBox status code.
6416	* @param pVCpu The cross context virtual CPU structure.
6417	* @param pVmcsInfo The VMCS info. object.
6418	* @param iSegReg The segment register number (X86_SREG_XXX).
6419	* @param pSelReg Pointer to the segment selector.
6420	*
6421	* @remarks No-long-jump zone!!!
6422	*/
6423	static int hmR0VmxExportGuestSegReg(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, uint32_t iSegReg, PCCPUMSELREG pSelReg)
6424	{
6425	Assert(iSegReg < X86_SREG_COUNT);
6426
6427	uint32_t u32Access = pSelReg->Attr.u;
6428	if (!pVmcsInfo->pShared->RealMode.fRealOnV86Active)
6429	{
6430	/*
6431	* The way to differentiate between whether this is really a null selector or was just
6432	* a selector loaded with 0 in real-mode is using the segment attributes. A selector
6433	* loaded in real-mode with the value 0 is valid and usable in protected-mode and we
6434	* should -not- mark it as an unusable segment. Both the recompiler & VT-x ensures
6435	* NULL selectors loaded in protected-mode have their attribute as 0.
6436	*/
6437	if (u32Access)
6438	{ }
6439	else
6440	u32Access = X86DESCATTR_UNUSABLE;
6441	}
6442	else
6443	{
6444	/* VT-x requires our real-using-v86 mode hack to override the segment access-right bits. */
6445	u32Access = 0xf3;
6446	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
6447	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
6448	RT_NOREF_PV(pVCpu);
6449	}
6450
6451	/* Validate segment access rights. Refer to Intel spec. "26.3.1.2 Checks on Guest Segment Registers". */
6452	AssertMsg((u32Access & X86DESCATTR_UNUSABLE) \|\| (u32Access & X86_SEL_TYPE_ACCESSED),
6453	("Access bit not set for usable segment. %.2s sel=%#x attr %#x\n", "ESCSSSDSFSGS" + iSegReg * 2, pSelReg, pSelReg->Attr.u));
6454
6455	/*
6456	* Commit it to the VMCS.
6457	*/
6458	Assert((uint32_t)VMX_VMCS16_GUEST_SEG_SEL(iSegReg) == g_aVmcsSegSel[iSegReg]);
6459	Assert((uint32_t)VMX_VMCS32_GUEST_SEG_LIMIT(iSegReg) == g_aVmcsSegLimit[iSegReg]);
6460	Assert((uint32_t)VMX_VMCS32_GUEST_SEG_ACCESS_RIGHTS(iSegReg) == g_aVmcsSegAttr[iSegReg]);
6461	Assert((uint32_t)VMX_VMCS_GUEST_SEG_BASE(iSegReg) == g_aVmcsSegBase[iSegReg]);
6462	int rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_SEG_SEL(iSegReg), pSelReg->Sel); AssertRC(rc);
6463	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SEG_LIMIT(iSegReg), pSelReg->u32Limit); AssertRC(rc);
6464	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SEG_BASE(iSegReg), pSelReg->u64Base); AssertRC(rc);
6465	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SEG_ACCESS_RIGHTS(iSegReg), u32Access); AssertRC(rc);
6466	return VINF_SUCCESS;
6467	}
6468
6469
6470	/**
6471	* Exports the guest segment registers, GDTR, IDTR, LDTR, TR into the guest-state
6472	* area in the VMCS.
6473	*
6474	* @returns VBox status code.
6475	* @param pVCpu The cross context virtual CPU structure.
6476	* @param pVmxTransient The VMX-transient structure.
6477	*
6478	* @remarks Will import guest CR0 on strict builds during validation of
6479	* segments.
6480	* @remarks No-long-jump zone!!!
6481	*/
6482	static int hmR0VmxExportGuestSegRegsXdtr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6483	{
6484	int rc = VERR_INTERNAL_ERROR_5;
6485	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6486	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6487	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6488	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
6489
6490	/*
6491	* Guest Segment registers: CS, SS, DS, ES, FS, GS.
6492	*/
6493	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SREG_MASK)
6494	{
6495	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CS)
6496	{
6497	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CS);
6498	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6499	pVmcsInfoShared->RealMode.AttrCS.u = pCtx->cs.Attr.u;
6500	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_CS, &pCtx->cs);
6501	AssertRC(rc);
6502	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CS);
6503	}
6504
6505	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SS)
6506	{
6507	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SS);
6508	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6509	pVmcsInfoShared->RealMode.AttrSS.u = pCtx->ss.Attr.u;
6510	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_SS, &pCtx->ss);
6511	AssertRC(rc);
6512	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SS);
6513	}
6514
6515	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_DS)
6516	{
6517	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_DS);
6518	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6519	pVmcsInfoShared->RealMode.AttrDS.u = pCtx->ds.Attr.u;
6520	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_DS, &pCtx->ds);
6521	AssertRC(rc);
6522	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_DS);
6523	}
6524
6525	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_ES)
6526	{
6527	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_ES);
6528	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6529	pVmcsInfoShared->RealMode.AttrES.u = pCtx->es.Attr.u;
6530	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_ES, &pCtx->es);
6531	AssertRC(rc);
6532	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_ES);
6533	}
6534
6535	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_FS)
6536	{
6537	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_FS);
6538	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6539	pVmcsInfoShared->RealMode.AttrFS.u = pCtx->fs.Attr.u;
6540	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_FS, &pCtx->fs);
6541	AssertRC(rc);
6542	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_FS);
6543	}
6544
6545	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_GS)
6546	{
6547	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GS);
6548	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6549	pVmcsInfoShared->RealMode.AttrGS.u = pCtx->gs.Attr.u;
6550	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_GS, &pCtx->gs);
6551	AssertRC(rc);
6552	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_GS);
6553	}
6554
6555	#ifdef VBOX_STRICT
6556	hmR0VmxValidateSegmentRegs(pVCpu, pVmcsInfo);
6557	#endif
6558	Log4Func(("cs={%#04x base=%#RX64 limit=%#RX32 attr=%#RX32}\n", pCtx->cs.Sel, pCtx->cs.u64Base, pCtx->cs.u32Limit,
6559	pCtx->cs.Attr.u));
6560	}
6561
6562	/*
6563	* Guest TR.
6564	*/
6565	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_TR)
6566	{
6567	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_TR);
6568
6569	/*
6570	* Real-mode emulation using virtual-8086 mode with CR4.VME. Interrupt redirection is
6571	* achieved using the interrupt redirection bitmap (all bits cleared to let the guest
6572	* handle INT-n's) in the TSS. See hmR3InitFinalizeR0() to see how pRealModeTSS is setup.
6573	*/
6574	uint16_t u16Sel;
6575	uint32_t u32Limit;
6576	uint64_t u64Base;
6577	uint32_t u32AccessRights;
6578	if (!pVmcsInfoShared->RealMode.fRealOnV86Active)
6579	{
6580	u16Sel = pCtx->tr.Sel;
6581	u32Limit = pCtx->tr.u32Limit;
6582	u64Base = pCtx->tr.u64Base;
6583	u32AccessRights = pCtx->tr.Attr.u;
6584	}
6585	else
6586	{
6587	Assert(!pVmxTransient->fIsNestedGuest);
6588	Assert(pVM->hm.s.vmx.pRealModeTSS);
6589	Assert(PDMVmmDevHeapIsEnabled(pVM)); /* Guaranteed by HMCanExecuteGuest() -XXX- what about inner loop changes? */
6590
6591	/* We obtain it here every time as PCI regions could be reconfigured in the guest, changing the VMMDev base. */
6592	RTGCPHYS GCPhys;
6593	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys);
6594	AssertRCReturn(rc, rc);
6595
6596	X86DESCATTR DescAttr;
6597	DescAttr.u = 0;
6598	DescAttr.n.u1Present = 1;
6599	DescAttr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
6600
6601	u16Sel = 0;
6602	u32Limit = HM_VTX_TSS_SIZE;
6603	u64Base = GCPhys;
6604	u32AccessRights = DescAttr.u;
6605	}
6606
6607	/* Validate. */
6608	Assert(!(u16Sel & RT_BIT(2)));
6609	AssertMsg( (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_386_TSS_BUSY
6610	\|\| (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_286_TSS_BUSY, ("TSS is not busy!? %#x\n", u32AccessRights));
6611	AssertMsg(!(u32AccessRights & X86DESCATTR_UNUSABLE), ("TR unusable bit is not clear!? %#x\n", u32AccessRights));
6612	Assert(!(u32AccessRights & RT_BIT(4))); /* System MBZ.*/
6613	Assert(u32AccessRights & RT_BIT(7)); /* Present MB1.*/
6614	Assert(!(u32AccessRights & 0xf00)); /* 11:8 MBZ. */
6615	Assert(!(u32AccessRights & 0xfffe0000)); /* 31:17 MBZ. */
6616	Assert( (u32Limit & 0xfff) == 0xfff
6617	\|\| !(u32AccessRights & RT_BIT(15))); /* Granularity MBZ. */
6618	Assert( !(pCtx->tr.u32Limit & 0xfff00000)
6619	\|\| (u32AccessRights & RT_BIT(15))); /* Granularity MB1. */
6620
6621	rc = VMXWriteVmcs16(VMX_VMCS16_GUEST_TR_SEL, u16Sel); AssertRC(rc);
6622	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, u32Limit); AssertRC(rc);
6623	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, u32AccessRights); AssertRC(rc);
6624	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_TR_BASE, u64Base); AssertRC(rc);
6625
6626	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_TR);
6627	Log4Func(("tr base=%#RX64 limit=%#RX32\n", pCtx->tr.u64Base, pCtx->tr.u32Limit));
6628	}
6629
6630	/*
6631	* Guest GDTR.
6632	*/
6633	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_GDTR)
6634	{
6635	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GDTR);
6636
6637	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, pCtx->gdtr.cbGdt); AssertRC(rc);
6638	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, pCtx->gdtr.pGdt); AssertRC(rc);
6639
6640	/* Validate. */
6641	Assert(!(pCtx->gdtr.cbGdt & 0xffff0000)); /* Bits 31:16 MBZ. */
6642
6643	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_GDTR);
6644	Log4Func(("gdtr base=%#RX64 limit=%#RX32\n", pCtx->gdtr.pGdt, pCtx->gdtr.cbGdt));
6645	}
6646
6647	/*
6648	* Guest LDTR.
6649	*/
6650	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_LDTR)
6651	{
6652	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_LDTR);
6653
6654	/* The unusable bit is specific to VT-x, if it's a null selector mark it as an unusable segment. */
6655	uint32_t u32Access;
6656	if ( !pVmxTransient->fIsNestedGuest
6657	&& !pCtx->ldtr.Attr.u)
6658	u32Access = X86DESCATTR_UNUSABLE;
6659	else
6660	u32Access = pCtx->ldtr.Attr.u;
6661
6662	rc = VMXWriteVmcs16(VMX_VMCS16_GUEST_LDTR_SEL, pCtx->ldtr.Sel); AssertRC(rc);
6663	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, pCtx->ldtr.u32Limit); AssertRC(rc);
6664	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, u32Access); AssertRC(rc);
6665	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_LDTR_BASE, pCtx->ldtr.u64Base); AssertRC(rc);
6666
6667	/* Validate. */
6668	if (!(u32Access & X86DESCATTR_UNUSABLE))
6669	{
6670	Assert(!(pCtx->ldtr.Sel & RT_BIT(2))); /* TI MBZ. */
6671	Assert(pCtx->ldtr.Attr.n.u4Type == 2); /* Type MB2 (LDT). */
6672	Assert(!pCtx->ldtr.Attr.n.u1DescType); /* System MBZ. */
6673	Assert(pCtx->ldtr.Attr.n.u1Present == 1); /* Present MB1. */
6674	Assert(!pCtx->ldtr.Attr.n.u4LimitHigh); /* 11:8 MBZ. */
6675	Assert(!(pCtx->ldtr.Attr.u & 0xfffe0000)); /* 31:17 MBZ. */
6676	Assert( (pCtx->ldtr.u32Limit & 0xfff) == 0xfff
6677	\|\| !pCtx->ldtr.Attr.n.u1Granularity); /* Granularity MBZ. */
6678	Assert( !(pCtx->ldtr.u32Limit & 0xfff00000)
6679	\|\| pCtx->ldtr.Attr.n.u1Granularity); /* Granularity MB1. */
6680	}
6681
6682	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_LDTR);
6683	Log4Func(("ldtr base=%#RX64 limit=%#RX32\n", pCtx->ldtr.u64Base, pCtx->ldtr.u32Limit));
6684	}
6685
6686	/*
6687	* Guest IDTR.
6688	*/
6689	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_IDTR)
6690	{
6691	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_IDTR);
6692
6693	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, pCtx->idtr.cbIdt); AssertRC(rc);
6694	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, pCtx->idtr.pIdt); AssertRC(rc);
6695
6696	/* Validate. */
6697	Assert(!(pCtx->idtr.cbIdt & 0xffff0000)); /* Bits 31:16 MBZ. */
6698
6699	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_IDTR);
6700	Log4Func(("idtr base=%#RX64 limit=%#RX32\n", pCtx->idtr.pIdt, pCtx->idtr.cbIdt));
6701	}
6702
6703	return VINF_SUCCESS;
6704	}
6705
6706
6707	/**
6708	* Exports certain guest MSRs into the VM-entry MSR-load and VM-exit MSR-store
6709	* areas.
6710	*
6711	* These MSRs will automatically be loaded to the host CPU on every successful
6712	* VM-entry and stored from the host CPU on every successful VM-exit.
6713	*
6714	* We creates/updates MSR slots for the host MSRs in the VM-exit MSR-load area. The
6715	* actual host MSR values are not- updated here for performance reasons. See
6716	* hmR0VmxExportHostMsrs().
6717	*
6718	* We also exports the guest sysenter MSRs into the guest-state area in the VMCS.
6719	*
6720	* @returns VBox status code.
6721	* @param pVCpu The cross context virtual CPU structure.
6722	* @param pVmxTransient The VMX-transient structure.
6723	*
6724	* @remarks No-long-jump zone!!!
6725	*/
6726	static int hmR0VmxExportGuestMsrs(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6727	{
6728	AssertPtr(pVCpu);
6729	AssertPtr(pVmxTransient);
6730
6731	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6732	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6733
6734	/*
6735	* MSRs that we use the auto-load/store MSR area in the VMCS.
6736	* For 64-bit hosts, we load/restore them lazily, see hmR0VmxLazyLoadGuestMsrs(),
6737	* nothing to do here. The host MSR values are updated when it's safe in
6738	* hmR0VmxLazySaveHostMsrs().
6739	*
6740	* For nested-guests, the guests MSRs from the VM-entry MSR-load area are already
6741	* loaded (into the guest-CPU context) by the VMLAUNCH/VMRESUME instruction
6742	* emulation. The merged MSR permission bitmap will ensure that we get VM-exits
6743	* for any MSR that are not part of the lazy MSRs so we do not need to place
6744	* those MSRs into the auto-load/store MSR area. Nothing to do here.
6745	*/
6746	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_GUEST_AUTO_MSRS)
6747	{
6748	/* No auto-load/store MSRs currently. */
6749	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_GUEST_AUTO_MSRS);
6750	}
6751
6752	/*
6753	* Guest Sysenter MSRs.
6754	*/
6755	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_MSR_MASK)
6756	{
6757	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SYSENTER_MSRS);
6758
6759	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_CS_MSR)
6760	{
6761	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, pCtx->SysEnter.cs);
6762	AssertRC(rc);
6763	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_CS_MSR);
6764	}
6765
6766	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_EIP_MSR)
6767	{
6768	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SYSENTER_EIP, pCtx->SysEnter.eip);
6769	AssertRC(rc);
6770	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_EIP_MSR);
6771	}
6772
6773	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_ESP_MSR)
6774	{
6775	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SYSENTER_ESP, pCtx->SysEnter.esp);
6776	AssertRC(rc);
6777	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_ESP_MSR);
6778	}
6779	}
6780
6781	/*
6782	* Guest/host EFER MSR.
6783	*/
6784	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_EFER_MSR)
6785	{
6786	/* Whether we are using the VMCS to swap the EFER MSR must have been
6787	determined earlier while exporting VM-entry/VM-exit controls. */
6788	Assert(!(ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_ENTRY_EXIT_CTLS));
6789	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
6790
6791	if (hmR0VmxShouldSwapEferMsr(pVCpu, pVmxTransient))
6792	{
6793	/*
6794	* EFER.LME is written by software, while EFER.LMA is set by the CPU to (CR0.PG & EFER.LME).
6795	* This means a guest can set EFER.LME=1 while CR0.PG=0 and EFER.LMA can remain 0.
6796	* VT-x requires that "IA-32e mode guest" VM-entry control must be identical to EFER.LMA
6797	* and to CR0.PG. Without unrestricted execution, CR0.PG (used for VT-x, not the shadow)
6798	* must always be 1. This forces us to effectively clear both EFER.LMA and EFER.LME until
6799	* the guest has also set CR0.PG=1. Otherwise, we would run into an invalid-guest state
6800	* during VM-entry.
6801	*/
6802	uint64_t uGuestEferMsr = pCtx->msrEFER;
6803	if (!pVM->hmr0.s.vmx.fUnrestrictedGuest)
6804	{
6805	if (!(pCtx->msrEFER & MSR_K6_EFER_LMA))
6806	uGuestEferMsr &= ~MSR_K6_EFER_LME;
6807	else
6808	Assert((pCtx->msrEFER & (MSR_K6_EFER_LMA \| MSR_K6_EFER_LME)) == (MSR_K6_EFER_LMA \| MSR_K6_EFER_LME));
6809	}
6810
6811	/*
6812	* If the CPU supports VMCS controls for swapping EFER, use it. Otherwise, we have no option
6813	* but to use the auto-load store MSR area in the VMCS for swapping EFER. See @bugref{7368}.
6814	*/
6815	if (g_fHmVmxSupportsVmcsEfer)
6816	{
6817	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_EFER_FULL, uGuestEferMsr);
6818	AssertRC(rc);
6819	}
6820	else
6821	{
6822	/*
6823	* We shall use the auto-load/store MSR area only for loading the EFER MSR but we must
6824	* continue to intercept guest read and write accesses to it, see @bugref{7386#c16}.
6825	*/
6826	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_EFER, uGuestEferMsr,
6827	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6828	AssertRCReturn(rc, rc);
6829	}
6830
6831	Log4Func(("efer=%#RX64 shadow=%#RX64\n", uGuestEferMsr, pCtx->msrEFER));
6832	}
6833	else if (!g_fHmVmxSupportsVmcsEfer)
6834	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_EFER);
6835
6836	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_EFER_MSR);
6837	}
6838
6839	/*
6840	* Other MSRs.
6841	*/
6842	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_OTHER_MSRS)
6843	{
6844	/* Speculation Control (R/W). */
6845	HMVMX_CPUMCTX_ASSERT(pVCpu, HM_CHANGED_GUEST_OTHER_MSRS);
6846	if (pVM->cpum.ro.GuestFeatures.fIbrs)
6847	{
6848	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_IA32_SPEC_CTRL, CPUMGetGuestSpecCtrl(pVCpu),
6849	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6850	AssertRCReturn(rc, rc);
6851	}
6852
6853	/* Last Branch Record. */
6854	if (pVM->hmr0.s.vmx.fLbr)
6855	{
6856	PVMXVMCSINFOSHARED const pVmcsInfoShared = pVmxTransient->pVmcsInfo->pShared;
6857	uint32_t const idFromIpMsrStart = pVM->hmr0.s.vmx.idLbrFromIpMsrFirst;
6858	uint32_t const idToIpMsrStart = pVM->hmr0.s.vmx.idLbrToIpMsrFirst;
6859	uint32_t const cLbrStack = pVM->hmr0.s.vmx.idLbrFromIpMsrLast - pVM->hmr0.s.vmx.idLbrFromIpMsrFirst + 1;
6860	Assert(cLbrStack <= 32);
6861	for (uint32_t i = 0; i < cLbrStack; i++)
6862	{
6863	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, idFromIpMsrStart + i,
6864	pVmcsInfoShared->au64LbrFromIpMsr[i],
6865	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6866	AssertRCReturn(rc, rc);
6867
6868	/* Some CPUs don't have a Branch-To-IP MSR (P4 and related Xeons). */
6869	if (idToIpMsrStart != 0)
6870	{
6871	rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, idToIpMsrStart + i,
6872	pVmcsInfoShared->au64LbrToIpMsr[i],
6873	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6874	AssertRCReturn(rc, rc);
6875	}
6876	}
6877
6878	/* Add LBR top-of-stack MSR (which contains the index to the most recent record). */
6879	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, pVM->hmr0.s.vmx.idLbrTosMsr,
6880	pVmcsInfoShared->u64LbrTosMsr, false /* fSetReadWrite */,
6881	false /* fUpdateHostMsr */);
6882	AssertRCReturn(rc, rc);
6883	}
6884
6885	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_OTHER_MSRS);
6886	}
6887
6888	return VINF_SUCCESS;
6889	}
6890
6891
6892	/**
6893	* Wrapper for running the guest code in VT-x.
6894	*
6895	* @returns VBox status code, no informational status codes.
6896	* @param pVCpu The cross context virtual CPU structure.
6897	* @param pVmxTransient The VMX-transient structure.
6898	*
6899	* @remarks No-long-jump zone!!!
6900	*/
6901	DECLINLINE(int) hmR0VmxRunGuest(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6902	{
6903	/* Mark that HM is the keeper of all guest-CPU registers now that we're going to execute guest code. */
6904	pVCpu->cpum.GstCtx.fExtrn \|= HMVMX_CPUMCTX_EXTRN_ALL \| CPUMCTX_EXTRN_KEEPER_HM;
6905
6906	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6907	bool const fResumeVM = RT_BOOL(pVmcsInfo->fVmcsState & VMX_V_VMCS_LAUNCH_STATE_LAUNCHED);
6908	#ifdef VBOX_WITH_STATISTICS
6909	if (fResumeVM)
6910	STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxVmResume);
6911	else
6912	STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxVmLaunch);
6913	#endif
6914	int rc = pVCpu->hmr0.s.vmx.pfnStartVm(pVmcsInfo, pVCpu, fResumeVM);
6915	AssertMsg(rc <= VINF_SUCCESS, ("%Rrc\n", rc));
6916	return rc;
6917	}
6918
6919
6920	/**
6921	* Reports world-switch error and dumps some useful debug info.
6922	*
6923	* @param pVCpu The cross context virtual CPU structure.
6924	* @param rcVMRun The return code from VMLAUNCH/VMRESUME.
6925	* @param pVmxTransient The VMX-transient structure (only
6926	* exitReason updated).
6927	*/
6928	static void hmR0VmxReportWorldSwitchError(PVMCPUCC pVCpu, int rcVMRun, PVMXTRANSIENT pVmxTransient)
6929	{
6930	Assert(pVCpu);
6931	Assert(pVmxTransient);
6932	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
6933
6934	Log4Func(("VM-entry failure: %Rrc\n", rcVMRun));
6935	switch (rcVMRun)
6936	{
6937	case VERR_VMX_INVALID_VMXON_PTR:
6938	AssertFailed();
6939	break;
6940	case VINF_SUCCESS: /* VMLAUNCH/VMRESUME succeeded but VM-entry failed... yeah, true story. */
6941	case VERR_VMX_UNABLE_TO_START_VM: /* VMLAUNCH/VMRESUME itself failed. */
6942	{
6943	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &pVCpu->hm.s.vmx.LastError.u32ExitReason);
6944	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
6945	AssertRC(rc);
6946	hmR0VmxReadExitQualVmcs(pVmxTransient);
6947
6948	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hmr0.s.idEnteredCpu;
6949	/* LastError.idCurrentCpu was already updated in hmR0VmxPreRunGuestCommitted().
6950	Cannot do it here as we may have been long preempted. */
6951
6952	#ifdef VBOX_STRICT
6953	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
6954	Log4(("uExitReason %#RX32 (VmxTransient %#RX16)\n", pVCpu->hm.s.vmx.LastError.u32ExitReason,
6955	pVmxTransient->uExitReason));
6956	Log4(("Exit Qualification %#RX64\n", pVmxTransient->uExitQual));
6957	Log4(("InstrError %#RX32\n", pVCpu->hm.s.vmx.LastError.u32InstrError));
6958	if (pVCpu->hm.s.vmx.LastError.u32InstrError <= HMVMX_INSTR_ERROR_MAX)
6959	Log4(("InstrError Desc. \"%s\"\n", g_apszVmxInstrErrors[pVCpu->hm.s.vmx.LastError.u32InstrError]));
6960	else
6961	Log4(("InstrError Desc. Range exceeded %u\n", HMVMX_INSTR_ERROR_MAX));
6962	Log4(("Entered host CPU %u\n", pVCpu->hm.s.vmx.LastError.idEnteredCpu));
6963	Log4(("Current host CPU %u\n", pVCpu->hm.s.vmx.LastError.idCurrentCpu));
6964
6965	static struct
6966	{
6967	/** Name of the field to log. */
6968	const char *pszName;
6969	/** The VMCS field. */
6970	uint32_t uVmcsField;
6971	/** Whether host support of this field needs to be checked. */
6972	bool fCheckSupport;
6973	} const s_aVmcsFields[] =
6974	{
6975	{ "VMX_VMCS32_CTRL_PIN_EXEC", VMX_VMCS32_CTRL_PIN_EXEC, false },
6976	{ "VMX_VMCS32_CTRL_PROC_EXEC", VMX_VMCS32_CTRL_PROC_EXEC, false },
6977	{ "VMX_VMCS32_CTRL_PROC_EXEC2", VMX_VMCS32_CTRL_PROC_EXEC2, true },
6978	{ "VMX_VMCS32_CTRL_ENTRY", VMX_VMCS32_CTRL_ENTRY, false },
6979	{ "VMX_VMCS32_CTRL_EXIT", VMX_VMCS32_CTRL_EXIT, false },
6980	{ "VMX_VMCS32_CTRL_CR3_TARGET_COUNT", VMX_VMCS32_CTRL_CR3_TARGET_COUNT, false },
6981	{ "VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO", VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, false },
6982	{ "VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE", VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, false },
6983	{ "VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH", VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, false },
6984	{ "VMX_VMCS32_CTRL_TPR_THRESHOLD", VMX_VMCS32_CTRL_TPR_THRESHOLD, false },
6985	{ "VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT", VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, false },
6986	{ "VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT", VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, false },
6987	{ "VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT", VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, false },
6988	{ "VMX_VMCS32_CTRL_EXCEPTION_BITMAP", VMX_VMCS32_CTRL_EXCEPTION_BITMAP, false },
6989	{ "VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK", VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, false },
6990	{ "VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH", VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, false },
6991	{ "VMX_VMCS_CTRL_CR0_MASK", VMX_VMCS_CTRL_CR0_MASK, false },
6992	{ "VMX_VMCS_CTRL_CR0_READ_SHADOW", VMX_VMCS_CTRL_CR0_READ_SHADOW, false },
6993	{ "VMX_VMCS_CTRL_CR4_MASK", VMX_VMCS_CTRL_CR4_MASK, false },
6994	{ "VMX_VMCS_CTRL_CR4_READ_SHADOW", VMX_VMCS_CTRL_CR4_READ_SHADOW, false },
6995	{ "VMX_VMCS64_CTRL_EPTP_FULL", VMX_VMCS64_CTRL_EPTP_FULL, true },
6996	{ "VMX_VMCS_GUEST_RIP", VMX_VMCS_GUEST_RIP, false },
6997	{ "VMX_VMCS_GUEST_RSP", VMX_VMCS_GUEST_RSP, false },
6998	{ "VMX_VMCS_GUEST_RFLAGS", VMX_VMCS_GUEST_RFLAGS, false },
6999	{ "VMX_VMCS16_VPID", VMX_VMCS16_VPID, true, },
7000	{ "VMX_VMCS_HOST_CR0", VMX_VMCS_HOST_CR0, false },
7001	{ "VMX_VMCS_HOST_CR3", VMX_VMCS_HOST_CR3, false },
7002	{ "VMX_VMCS_HOST_CR4", VMX_VMCS_HOST_CR4, false },
7003	/* The order of selector fields below are fixed! */
7004	{ "VMX_VMCS16_HOST_ES_SEL", VMX_VMCS16_HOST_ES_SEL, false },
7005	{ "VMX_VMCS16_HOST_CS_SEL", VMX_VMCS16_HOST_CS_SEL, false },
7006	{ "VMX_VMCS16_HOST_SS_SEL", VMX_VMCS16_HOST_SS_SEL, false },
7007	{ "VMX_VMCS16_HOST_DS_SEL", VMX_VMCS16_HOST_DS_SEL, false },
7008	{ "VMX_VMCS16_HOST_FS_SEL", VMX_VMCS16_HOST_FS_SEL, false },
7009	{ "VMX_VMCS16_HOST_GS_SEL", VMX_VMCS16_HOST_GS_SEL, false },
7010	{ "VMX_VMCS16_HOST_TR_SEL", VMX_VMCS16_HOST_TR_SEL, false },
7011	/* End of ordered selector fields. */
7012	{ "VMX_VMCS_HOST_TR_BASE", VMX_VMCS_HOST_TR_BASE, false },
7013	{ "VMX_VMCS_HOST_GDTR_BASE", VMX_VMCS_HOST_GDTR_BASE, false },
7014	{ "VMX_VMCS_HOST_IDTR_BASE", VMX_VMCS_HOST_IDTR_BASE, false },
7015	{ "VMX_VMCS32_HOST_SYSENTER_CS", VMX_VMCS32_HOST_SYSENTER_CS, false },
7016	{ "VMX_VMCS_HOST_SYSENTER_EIP", VMX_VMCS_HOST_SYSENTER_EIP, false },
7017	{ "VMX_VMCS_HOST_SYSENTER_ESP", VMX_VMCS_HOST_SYSENTER_ESP, false },
7018	{ "VMX_VMCS_HOST_RSP", VMX_VMCS_HOST_RSP, false },
7019	{ "VMX_VMCS_HOST_RIP", VMX_VMCS_HOST_RIP, false }
7020	};
7021
7022	RTGDTR HostGdtr;
7023	ASMGetGDTR(&HostGdtr);
7024
7025	uint32_t const cVmcsFields = RT_ELEMENTS(s_aVmcsFields);
7026	for (uint32_t i = 0; i < cVmcsFields; i++)
7027	{
7028	uint32_t const uVmcsField = s_aVmcsFields[i].uVmcsField;
7029
7030	bool fSupported;
7031	if (!s_aVmcsFields[i].fCheckSupport)
7032	fSupported = true;
7033	else
7034	{
7035	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7036	switch (uVmcsField)
7037	{
7038	case VMX_VMCS64_CTRL_EPTP_FULL: fSupported = pVM->hmr0.s.fNestedPaging; break;
7039	case VMX_VMCS16_VPID: fSupported = pVM->hmr0.s.vmx.fVpid; break;
7040	case VMX_VMCS32_CTRL_PROC_EXEC2:
7041	fSupported = RT_BOOL(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS);
7042	break;
7043	default:
7044	AssertMsgFailedReturnVoid(("Failed to provide VMCS field support for %#RX32\n", uVmcsField));
7045	}
7046	}
7047
7048	if (fSupported)
7049	{
7050	uint8_t const uWidth = RT_BF_GET(uVmcsField, VMX_BF_VMCSFIELD_WIDTH);
7051	switch (uWidth)
7052	{
7053	case VMX_VMCSFIELD_WIDTH_16BIT:
7054	{
7055	uint16_t u16Val;
7056	rc = VMXReadVmcs16(uVmcsField, &u16Val);
7057	AssertRC(rc);
7058	Log4(("%-40s = %#RX16\n", s_aVmcsFields[i].pszName, u16Val));
7059
7060	if ( uVmcsField >= VMX_VMCS16_HOST_ES_SEL
7061	&& uVmcsField <= VMX_VMCS16_HOST_TR_SEL)
7062	{
7063	if (u16Val < HostGdtr.cbGdt)
7064	{
7065	/* Order of selectors in s_apszSel is fixed and matches the order in s_aVmcsFields. */
7066	static const char * const s_apszSel[] = { "Host ES", "Host CS", "Host SS", "Host DS",
7067	"Host FS", "Host GS", "Host TR" };
7068	uint8_t const idxSel = RT_BF_GET(uVmcsField, VMX_BF_VMCSFIELD_INDEX);
7069	Assert(idxSel < RT_ELEMENTS(s_apszSel));
7070	PCX86DESCHC pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u16Val & X86_SEL_MASK));
7071	hmR0DumpDescriptor(pDesc, u16Val, s_apszSel[idxSel]);
7072	}
7073	else
7074	Log4((" Selector value exceeds GDT limit!\n"));
7075	}
7076	break;
7077	}
7078
7079	case VMX_VMCSFIELD_WIDTH_32BIT:
7080	{
7081	uint32_t u32Val;
7082	rc = VMXReadVmcs32(uVmcsField, &u32Val);
7083	AssertRC(rc);
7084	Log4(("%-40s = %#RX32\n", s_aVmcsFields[i].pszName, u32Val));
7085	break;
7086	}
7087
7088	case VMX_VMCSFIELD_WIDTH_64BIT:
7089	case VMX_VMCSFIELD_WIDTH_NATURAL:
7090	{
7091	uint64_t u64Val;
7092	rc = VMXReadVmcs64(uVmcsField, &u64Val);
7093	AssertRC(rc);
7094	Log4(("%-40s = %#RX64\n", s_aVmcsFields[i].pszName, u64Val));
7095	break;
7096	}
7097	}
7098	}
7099	}
7100
7101	Log4(("MSR_K6_EFER = %#RX64\n", ASMRdMsr(MSR_K6_EFER)));
7102	Log4(("MSR_K8_CSTAR = %#RX64\n", ASMRdMsr(MSR_K8_CSTAR)));
7103	Log4(("MSR_K8_LSTAR = %#RX64\n", ASMRdMsr(MSR_K8_LSTAR)));
7104	Log4(("MSR_K6_STAR = %#RX64\n", ASMRdMsr(MSR_K6_STAR)));
7105	Log4(("MSR_K8_SF_MASK = %#RX64\n", ASMRdMsr(MSR_K8_SF_MASK)));
7106	Log4(("MSR_K8_KERNEL_GS_BASE = %#RX64\n", ASMRdMsr(MSR_K8_KERNEL_GS_BASE)));
7107	#endif /* VBOX_STRICT */
7108	break;
7109	}
7110
7111	default:
7112	/* Impossible */
7113	AssertMsgFailed(("hmR0VmxReportWorldSwitchError %Rrc (%#x)\n", rcVMRun, rcVMRun));
7114	break;
7115	}
7116	}
7117
7118
7119	/**
7120	* Sets up the usage of TSC-offsetting and updates the VMCS.
7121	*
7122	* If offsetting is not possible, cause VM-exits on RDTSC(P)s. Also sets up the
7123	* VMX-preemption timer.
7124	*
7125	* @returns VBox status code.
7126	* @param pVCpu The cross context virtual CPU structure.
7127	* @param pVmxTransient The VMX-transient structure.
7128	* @param idCurrentCpu The current CPU number.
7129	*
7130	* @remarks No-long-jump zone!!!
7131	*/
7132	static void hmR0VmxUpdateTscOffsettingAndPreemptTimer(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, RTCPUID idCurrentCpu)
7133	{
7134	bool fOffsettedTsc;
7135	bool fParavirtTsc;
7136	uint64_t uTscOffset;
7137	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7138	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
7139
7140	if (pVM->hmr0.s.vmx.fUsePreemptTimer)
7141	{
7142	/* The TMCpuTickGetDeadlineAndTscOffset function is expensive (calling it on
7143	every entry slowed down the bs2-test1 CPUID testcase by ~33% (on an 10980xe). */
7144	uint64_t cTicksToDeadline;
7145	if ( idCurrentCpu == pVCpu->hmr0.s.idLastCpu
7146	&& TMVirtualSyncIsCurrentDeadlineVersion(pVM, pVCpu->hmr0.s.vmx.uTscDeadlineVersion))
7147	{
7148	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatVmxPreemptionReusingDeadline);
7149	fOffsettedTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &uTscOffset, &fParavirtTsc);
7150	cTicksToDeadline = pVCpu->hmr0.s.vmx.uTscDeadline - SUPReadTsc();
7151	if ((int64_t)cTicksToDeadline > 0)
7152	{ /* hopefully */ }
7153	else
7154	{
7155	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatVmxPreemptionReusingDeadlineExpired);
7156	cTicksToDeadline = 0;
7157	}
7158	}
7159	else
7160	{
7161	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatVmxPreemptionRecalcingDeadline);
7162	cTicksToDeadline = TMCpuTickGetDeadlineAndTscOffset(pVM, pVCpu, &uTscOffset, &fOffsettedTsc, &fParavirtTsc,
7163	&pVCpu->hmr0.s.vmx.uTscDeadline,
7164	&pVCpu->hmr0.s.vmx.uTscDeadlineVersion);
7165	pVCpu->hmr0.s.vmx.uTscDeadline += cTicksToDeadline;
7166	if (cTicksToDeadline >= 128)
7167	{ /* hopefully */ }
7168	else
7169	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatVmxPreemptionRecalcingDeadlineExpired);
7170	}
7171
7172	/* Make sure the returned values have sane upper and lower boundaries. */
7173	uint64_t const u64CpuHz = SUPGetCpuHzFromGipBySetIndex(g_pSUPGlobalInfoPage, pVCpu->iHostCpuSet);
7174	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. */
7175	cTicksToDeadline = RT_MAX(cTicksToDeadline, u64CpuHz / 32678); /* 1/32768th of a second, ~30us. */
7176	cTicksToDeadline >>= pVM->hm.s.vmx.cPreemptTimerShift;
7177
7178	/** @todo r=ramshankar: We need to find a way to integrate nested-guest
7179	* preemption timers here. We probably need to clamp the preemption timer,
7180	* after converting the timer value to the host. */
7181	uint32_t const cPreemptionTickCount = (uint32_t)RT_MIN(cTicksToDeadline, UINT32_MAX - 16);
7182	int rc = VMXWriteVmcs32(VMX_VMCS32_PREEMPT_TIMER_VALUE, cPreemptionTickCount);
7183	AssertRC(rc);
7184	}
7185	else
7186	fOffsettedTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &uTscOffset, &fParavirtTsc);
7187
7188	if (fParavirtTsc)
7189	{
7190	/* Currently neither Hyper-V nor KVM need to update their paravirt. TSC
7191	information before every VM-entry, hence disable it for performance sake. */
7192	#if 0
7193	int rc = GIMR0UpdateParavirtTsc(pVM, 0 /* u64Offset */);
7194	AssertRC(rc);
7195	#endif
7196	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
7197	}
7198
7199	if ( fOffsettedTsc
7200	&& RT_LIKELY(!pVCpu->hmr0.s.fDebugWantRdTscExit))
7201	{
7202	if (pVmxTransient->fIsNestedGuest)
7203	uTscOffset = CPUMApplyNestedGuestTscOffset(pVCpu, uTscOffset);
7204	hmR0VmxSetTscOffsetVmcs(pVmcsInfo, uTscOffset);
7205	hmR0VmxRemoveProcCtlsVmcs(pVCpu, pVmxTransient, VMX_PROC_CTLS_RDTSC_EXIT);
7206	}
7207	else
7208	{
7209	/* We can't use TSC-offsetting (non-fixed TSC, warp drive active etc.), VM-exit on RDTSC(P). */
7210	hmR0VmxSetProcCtlsVmcs(pVmxTransient, VMX_PROC_CTLS_RDTSC_EXIT);
7211	}
7212	}
7213
7214
7215	/**
7216	* Gets the IEM exception flags for the specified vector and IDT vectoring /
7217	* VM-exit interruption info type.
7218	*
7219	* @returns The IEM exception flags.
7220	* @param uVector The event vector.
7221	* @param uVmxEventType The VMX event type.
7222	*
7223	* @remarks This function currently only constructs flags required for
7224	* IEMEvaluateRecursiveXcpt and not the complete flags (e.g, error-code
7225	* and CR2 aspects of an exception are not included).
7226	*/
7227	static uint32_t hmR0VmxGetIemXcptFlags(uint8_t uVector, uint32_t uVmxEventType)
7228	{
7229	uint32_t fIemXcptFlags;
7230	switch (uVmxEventType)
7231	{
7232	case VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT:
7233	case VMX_IDT_VECTORING_INFO_TYPE_NMI:
7234	fIemXcptFlags = IEM_XCPT_FLAGS_T_CPU_XCPT;
7235	break;
7236
7237	case VMX_IDT_VECTORING_INFO_TYPE_EXT_INT:
7238	fIemXcptFlags = IEM_XCPT_FLAGS_T_EXT_INT;
7239	break;
7240
7241	case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
7242	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT \| IEM_XCPT_FLAGS_ICEBP_INSTR;
7243	break;
7244
7245	case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT:
7246	{
7247	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
7248	if (uVector == X86_XCPT_BP)
7249	fIemXcptFlags \|= IEM_XCPT_FLAGS_BP_INSTR;
7250	else if (uVector == X86_XCPT_OF)
7251	fIemXcptFlags \|= IEM_XCPT_FLAGS_OF_INSTR;
7252	else
7253	{
7254	fIemXcptFlags = 0;
7255	AssertMsgFailed(("Unexpected vector for software exception. uVector=%#x", uVector));
7256	}
7257	break;
7258	}
7259
7260	case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
7261	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
7262	break;
7263
7264	default:
7265	fIemXcptFlags = 0;
7266	AssertMsgFailed(("Unexpected vector type! uVmxEventType=%#x uVector=%#x", uVmxEventType, uVector));
7267	break;
7268	}
7269	return fIemXcptFlags;
7270	}
7271
7272
7273	/**
7274	* Sets an event as a pending event to be injected into the guest.
7275	*
7276	* @param pVCpu The cross context virtual CPU structure.
7277	* @param u32IntInfo The VM-entry interruption-information field.
7278	* @param cbInstr The VM-entry instruction length in bytes (for
7279	* software interrupts, exceptions and privileged
7280	* software exceptions).
7281	* @param u32ErrCode The VM-entry exception error code.
7282	* @param GCPtrFaultAddress The fault-address (CR2) in case it's a
7283	* page-fault.
7284	*/
7285	DECLINLINE(void) hmR0VmxSetPendingEvent(PVMCPUCC pVCpu, uint32_t u32IntInfo, uint32_t cbInstr, uint32_t u32ErrCode,
7286	RTGCUINTPTR GCPtrFaultAddress)
7287	{
7288	Assert(!pVCpu->hm.s.Event.fPending);
7289	pVCpu->hm.s.Event.fPending = true;
7290	pVCpu->hm.s.Event.u64IntInfo = u32IntInfo;
7291	pVCpu->hm.s.Event.u32ErrCode = u32ErrCode;
7292	pVCpu->hm.s.Event.cbInstr = cbInstr;
7293	pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress;
7294	}
7295
7296
7297	/**
7298	* Sets an external interrupt as pending-for-injection into the VM.
7299	*
7300	* @param pVCpu The cross context virtual CPU structure.
7301	* @param u8Interrupt The external interrupt vector.
7302	*/
7303	DECLINLINE(void) hmR0VmxSetPendingExtInt(PVMCPUCC pVCpu, uint8_t u8Interrupt)
7304	{
7305	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, u8Interrupt)
7306	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
7307	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7308	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7309	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7310	}
7311
7312
7313	/**
7314	* Sets an NMI (\#NMI) exception as pending-for-injection into the VM.
7315	*
7316	* @param pVCpu The cross context virtual CPU structure.
7317	*/
7318	DECLINLINE(void) hmR0VmxSetPendingXcptNmi(PVMCPUCC pVCpu)
7319	{
7320	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_NMI)
7321	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_NMI)
7322	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7323	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7324	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7325	}
7326
7327
7328	/**
7329	* Sets a double-fault (\#DF) exception as pending-for-injection into the VM.
7330	*
7331	* @param pVCpu The cross context virtual CPU structure.
7332	*/
7333	DECLINLINE(void) hmR0VmxSetPendingXcptDF(PVMCPUCC pVCpu)
7334	{
7335	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DF)
7336	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7337	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7338	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7339	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7340	}
7341
7342
7343	/**
7344	* Sets an invalid-opcode (\#UD) exception as pending-for-injection into the VM.
7345	*
7346	* @param pVCpu The cross context virtual CPU structure.
7347	*/
7348	DECLINLINE(void) hmR0VmxSetPendingXcptUD(PVMCPUCC pVCpu)
7349	{
7350	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_UD)
7351	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7352	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7353	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7354	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7355	}
7356
7357
7358	/**
7359	* Sets a debug (\#DB) exception as pending-for-injection into the VM.
7360	*
7361	* @param pVCpu The cross context virtual CPU structure.
7362	*/
7363	DECLINLINE(void) hmR0VmxSetPendingXcptDB(PVMCPUCC pVCpu)
7364	{
7365	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DB)
7366	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7367	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7368	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7369	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7370	}
7371
7372
7373	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7374	/**
7375	* Sets a general-protection (\#GP) exception as pending-for-injection into the VM.
7376	*
7377	* @param pVCpu The cross context virtual CPU structure.
7378	* @param u32ErrCode The error code for the general-protection exception.
7379	*/
7380	DECLINLINE(void) hmR0VmxSetPendingXcptGP(PVMCPUCC pVCpu, uint32_t u32ErrCode)
7381	{
7382	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_GP)
7383	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7384	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7385	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7386	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrCode, 0 / GCPtrFaultAddress */);
7387	}
7388
7389
7390	/**
7391	* Sets a stack (\#SS) exception as pending-for-injection into the VM.
7392	*
7393	* @param pVCpu The cross context virtual CPU structure.
7394	* @param u32ErrCode The error code for the stack exception.
7395	*/
7396	DECLINLINE(void) hmR0VmxSetPendingXcptSS(PVMCPUCC pVCpu, uint32_t u32ErrCode)
7397	{
7398	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_SS)
7399	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7400	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7401	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7402	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrCode, 0 / GCPtrFaultAddress */);
7403	}
7404	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
7405
7406
7407	/**
7408	* Fixes up attributes for the specified segment register.
7409	*
7410	* @param pVCpu The cross context virtual CPU structure.
7411	* @param pSelReg The segment register that needs fixing.
7412	* @param pszRegName The register name (for logging and assertions).
7413	*/
7414	static void hmR0VmxFixUnusableSegRegAttr(PVMCPUCC pVCpu, PCPUMSELREG pSelReg, const char *pszRegName)
7415	{
7416	Assert(pSelReg->Attr.u & X86DESCATTR_UNUSABLE);
7417
7418	/*
7419	* If VT-x marks the segment as unusable, most other bits remain undefined:
7420	* - For CS the L, D and G bits have meaning.
7421	* - For SS the DPL has meaning (it -is- the CPL for Intel and VBox).
7422	* - For the remaining data segments no bits are defined.
7423	*
7424	* The present bit and the unusable bit has been observed to be set at the
7425	* same time (the selector was supposed to be invalid as we started executing
7426	* a V8086 interrupt in ring-0).
7427	*
7428	* What should be important for the rest of the VBox code, is that the P bit is
7429	* cleared. Some of the other VBox code recognizes the unusable bit, but
7430	* AMD-V certainly don't, and REM doesn't really either. So, to be on the
7431	* safe side here, we'll strip off P and other bits we don't care about. If
7432	* any code breaks because Attr.u != 0 when Sel < 4, it should be fixed.
7433	*
7434	* See Intel spec. 27.3.2 "Saving Segment Registers and Descriptor-Table Registers".
7435	*/
7436	#ifdef VBOX_STRICT
7437	uint32_t const uAttr = pSelReg->Attr.u;
7438	#endif
7439
7440	/* Masking off: X86DESCATTR_P, X86DESCATTR_LIMIT_HIGH, and X86DESCATTR_AVL. The latter two are really irrelevant. */
7441	pSelReg->Attr.u &= X86DESCATTR_UNUSABLE \| X86DESCATTR_L \| X86DESCATTR_D \| X86DESCATTR_G
7442	\| X86DESCATTR_DPL \| X86DESCATTR_TYPE \| X86DESCATTR_DT;
7443
7444	#ifdef VBOX_STRICT
7445	VMMRZCallRing3Disable(pVCpu);
7446	Log4Func(("Unusable %s: sel=%#x attr=%#x -> %#x\n", pszRegName, pSelReg->Sel, uAttr, pSelReg->Attr.u));
7447	# ifdef DEBUG_bird
7448	AssertMsg((uAttr & ~X86DESCATTR_P) == pSelReg->Attr.u,
7449	("%s: %#x != %#x (sel=%#x base=%#llx limit=%#x)\n",
7450	pszRegName, uAttr, pSelReg->Attr.u, pSelReg->Sel, pSelReg->u64Base, pSelReg->u32Limit));
7451	# endif
7452	VMMRZCallRing3Enable(pVCpu);
7453	NOREF(uAttr);
7454	#endif
7455	RT_NOREF2(pVCpu, pszRegName);
7456	}
7457
7458
7459	/**
7460	* Imports a guest segment register from the current VMCS into the guest-CPU
7461	* context.
7462	*
7463	* @param pVCpu The cross context virtual CPU structure.
7464	* @param iSegReg The segment register number (X86_SREG_XXX).
7465	*
7466	* @remarks Called with interrupts and/or preemption disabled.
7467	*/
7468	static void hmR0VmxImportGuestSegReg(PVMCPUCC pVCpu, uint32_t iSegReg)
7469	{
7470	Assert(iSegReg < X86_SREG_COUNT);
7471	Assert((uint32_t)VMX_VMCS16_GUEST_SEG_SEL(iSegReg) == g_aVmcsSegSel[iSegReg]);
7472	Assert((uint32_t)VMX_VMCS32_GUEST_SEG_LIMIT(iSegReg) == g_aVmcsSegLimit[iSegReg]);
7473	Assert((uint32_t)VMX_VMCS32_GUEST_SEG_ACCESS_RIGHTS(iSegReg) == g_aVmcsSegAttr[iSegReg]);
7474	Assert((uint32_t)VMX_VMCS_GUEST_SEG_BASE(iSegReg) == g_aVmcsSegBase[iSegReg]);
7475
7476	PCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
7477
7478	uint16_t u16Sel;
7479	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_SEG_SEL(iSegReg), &u16Sel); AssertRC(rc);
7480	pSelReg->Sel = u16Sel;
7481	pSelReg->ValidSel = u16Sel;
7482
7483	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SEG_LIMIT(iSegReg), &pSelReg->u32Limit); AssertRC(rc);
7484	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SEG_BASE(iSegReg), &pSelReg->u64Base); AssertRC(rc);
7485
7486	uint32_t u32Attr;
7487	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SEG_ACCESS_RIGHTS(iSegReg), &u32Attr); AssertRC(rc);
7488	pSelReg->Attr.u = u32Attr;
7489	if (u32Attr & X86DESCATTR_UNUSABLE)
7490	hmR0VmxFixUnusableSegRegAttr(pVCpu, pSelReg, "ES\0CS\0SS\0DS\0FS\0GS" + iSegReg * 3);
7491
7492	pSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
7493	}
7494
7495
7496	/**
7497	* Imports the guest LDTR from the current VMCS into the guest-CPU context.
7498	*
7499	* @param pVCpu The cross context virtual CPU structure.
7500	*
7501	* @remarks Called with interrupts and/or preemption disabled.
7502	*/
7503	static void hmR0VmxImportGuestLdtr(PVMCPUCC pVCpu)
7504	{
7505	uint16_t u16Sel;
7506	uint64_t u64Base;
7507	uint32_t u32Limit, u32Attr;
7508	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_LDTR_SEL, &u16Sel); AssertRC(rc);
7509	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, &u32Limit); AssertRC(rc);
7510	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, &u32Attr); AssertRC(rc);
7511	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_LDTR_BASE, &u64Base); AssertRC(rc);
7512
7513	pVCpu->cpum.GstCtx.ldtr.Sel = u16Sel;
7514	pVCpu->cpum.GstCtx.ldtr.ValidSel = u16Sel;
7515	pVCpu->cpum.GstCtx.ldtr.fFlags = CPUMSELREG_FLAGS_VALID;
7516	pVCpu->cpum.GstCtx.ldtr.u32Limit = u32Limit;
7517	pVCpu->cpum.GstCtx.ldtr.u64Base = u64Base;
7518	pVCpu->cpum.GstCtx.ldtr.Attr.u = u32Attr;
7519	if (u32Attr & X86DESCATTR_UNUSABLE)
7520	hmR0VmxFixUnusableSegRegAttr(pVCpu, &pVCpu->cpum.GstCtx.ldtr, "LDTR");
7521	}
7522
7523
7524	/**
7525	* Imports the guest TR from the current VMCS into the guest-CPU context.
7526	*
7527	* @param pVCpu The cross context virtual CPU structure.
7528	*
7529	* @remarks Called with interrupts and/or preemption disabled.
7530	*/
7531	static void hmR0VmxImportGuestTr(PVMCPUCC pVCpu)
7532	{
7533	uint16_t u16Sel;
7534	uint64_t u64Base;
7535	uint32_t u32Limit, u32Attr;
7536	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_TR_SEL, &u16Sel); AssertRC(rc);
7537	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, &u32Limit); AssertRC(rc);
7538	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, &u32Attr); AssertRC(rc);
7539	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_TR_BASE, &u64Base); AssertRC(rc);
7540
7541	pVCpu->cpum.GstCtx.tr.Sel = u16Sel;
7542	pVCpu->cpum.GstCtx.tr.ValidSel = u16Sel;
7543	pVCpu->cpum.GstCtx.tr.fFlags = CPUMSELREG_FLAGS_VALID;
7544	pVCpu->cpum.GstCtx.tr.u32Limit = u32Limit;
7545	pVCpu->cpum.GstCtx.tr.u64Base = u64Base;
7546	pVCpu->cpum.GstCtx.tr.Attr.u = u32Attr;
7547	/* TR is the only selector that can never be unusable. */
7548	Assert(!(u32Attr & X86DESCATTR_UNUSABLE));
7549	}
7550
7551
7552	/**
7553	* Imports the guest RIP from the VMCS back into the guest-CPU context.
7554	*
7555	* @param pVCpu The cross context virtual CPU structure.
7556	*
7557	* @remarks Called with interrupts and/or preemption disabled, should not assert!
7558	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7559	* instead!!!
7560	*/
7561	static void hmR0VmxImportGuestRip(PVMCPUCC pVCpu)
7562	{
7563	uint64_t u64Val;
7564	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7565	if (pCtx->fExtrn & CPUMCTX_EXTRN_RIP)
7566	{
7567	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RIP, &u64Val);
7568	AssertRC(rc);
7569
7570	pCtx->rip = u64Val;
7571	EMR0HistoryUpdatePC(pVCpu, pCtx->rip, false);
7572	pCtx->fExtrn &= ~CPUMCTX_EXTRN_RIP;
7573	}
7574	}
7575
7576
7577	/**
7578	* Imports the guest RFLAGS from the VMCS back into the guest-CPU context.
7579	*
7580	* @param pVCpu The cross context virtual CPU structure.
7581	* @param pVmcsInfo The VMCS info. object.
7582	*
7583	* @remarks Called with interrupts and/or preemption disabled, should not assert!
7584	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7585	* instead!!!
7586	*/
7587	static void hmR0VmxImportGuestRFlags(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
7588	{
7589	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7590	if (pCtx->fExtrn & CPUMCTX_EXTRN_RFLAGS)
7591	{
7592	uint64_t u64Val;
7593	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RFLAGS, &u64Val);
7594	AssertRC(rc);
7595
7596	pCtx->rflags.u64 = u64Val;
7597	PCVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
7598	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
7599	{
7600	pCtx->eflags.Bits.u1VM = 0;
7601	pCtx->eflags.Bits.u2IOPL = pVmcsInfoShared->RealMode.Eflags.Bits.u2IOPL;
7602	}
7603	pCtx->fExtrn &= ~CPUMCTX_EXTRN_RFLAGS;
7604	}
7605	}
7606
7607
7608	/**
7609	* Imports the guest interruptibility-state from the VMCS back into the guest-CPU
7610	* context.
7611	*
7612	* @param pVCpu The cross context virtual CPU structure.
7613	* @param pVmcsInfo The VMCS info. object.
7614	*
7615	* @remarks Called with interrupts and/or preemption disabled, try not to assert and
7616	* do not log!
7617	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7618	* instead!!!
7619	*/
7620	static void hmR0VmxImportGuestIntrState(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
7621	{
7622	uint32_t u32Val;
7623	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &u32Val); AssertRC(rc);
7624	if (!u32Val)
7625	{
7626	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
7627	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
7628	CPUMSetGuestNmiBlocking(pVCpu, false);
7629	}
7630	else
7631	{
7632	/*
7633	* We must import RIP here to set our EM interrupt-inhibited state.
7634	* We also import RFLAGS as our code that evaluates pending interrupts
7635	* before VM-entry requires it.
7636	*/
7637	hmR0VmxImportGuestRip(pVCpu);
7638	hmR0VmxImportGuestRFlags(pVCpu, pVmcsInfo);
7639
7640	if (u32Val & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS \| VMX_VMCS_GUEST_INT_STATE_BLOCK_STI))
7641	EMSetInhibitInterruptsPC(pVCpu, pVCpu->cpum.GstCtx.rip);
7642	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
7643	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
7644
7645	bool const fNmiBlocking = RT_BOOL(u32Val & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI);
7646	CPUMSetGuestNmiBlocking(pVCpu, fNmiBlocking);
7647	}
7648	}
7649
7650
7651	/**
7652	* Worker for VMXR0ImportStateOnDemand.
7653	*
7654	* @returns VBox status code.
7655	* @param pVCpu The cross context virtual CPU structure.
7656	* @param pVmcsInfo The VMCS info. object.
7657	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
7658	*/
7659	static int hmR0VmxImportGuestState(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint64_t fWhat)
7660	{
7661	int rc = VINF_SUCCESS;
7662	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7663	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7664	uint32_t u32Val;
7665
7666	/*
7667	* Note! This is hack to workaround a mysterious BSOD observed with release builds
7668	* on Windows 10 64-bit hosts. Profile and debug builds are not affected and
7669	* neither are other host platforms.
7670	*
7671	* Committing this temporarily as it prevents BSOD.
7672	*
7673	* Update: This is very likely a compiler optimization bug, see @bugref{9180}.
7674	*/
7675	#ifdef RT_OS_WINDOWS
7676	if (pVM == 0 \|\| pVM == (void *)(uintptr_t)-1)
7677	return VERR_HM_IPE_1;
7678	#endif
7679
7680	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatImportGuestState, x);
7681
7682	/*
7683	* We disable interrupts to make the updating of the state and in particular
7684	* the fExtrn modification atomic wrt to preemption hooks.
7685	*/
7686	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
7687
7688	fWhat &= pCtx->fExtrn;
7689	if (fWhat)
7690	{
7691	do
7692	{
7693	if (fWhat & CPUMCTX_EXTRN_RIP)
7694	hmR0VmxImportGuestRip(pVCpu);
7695
7696	if (fWhat & CPUMCTX_EXTRN_RFLAGS)
7697	hmR0VmxImportGuestRFlags(pVCpu, pVmcsInfo);
7698
7699	if (fWhat & CPUMCTX_EXTRN_HM_VMX_INT_STATE)
7700	hmR0VmxImportGuestIntrState(pVCpu, pVmcsInfo);
7701
7702	if (fWhat & CPUMCTX_EXTRN_RSP)
7703	{
7704	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RSP, &pCtx->rsp);
7705	AssertRC(rc);
7706	}
7707
7708	if (fWhat & CPUMCTX_EXTRN_SREG_MASK)
7709	{
7710	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
7711	bool const fRealOnV86Active = pVmcsInfoShared->RealMode.fRealOnV86Active;
7712	if (fWhat & CPUMCTX_EXTRN_CS)
7713	{
7714	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_CS);
7715	hmR0VmxImportGuestRip(pVCpu);
7716	if (fRealOnV86Active)
7717	pCtx->cs.Attr.u = pVmcsInfoShared->RealMode.AttrCS.u;
7718	EMR0HistoryUpdatePC(pVCpu, pCtx->cs.u64Base + pCtx->rip, true /* fFlattened */);
7719	}
7720	if (fWhat & CPUMCTX_EXTRN_SS)
7721	{
7722	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_SS);
7723	if (fRealOnV86Active)
7724	pCtx->ss.Attr.u = pVmcsInfoShared->RealMode.AttrSS.u;
7725	}
7726	if (fWhat & CPUMCTX_EXTRN_DS)
7727	{
7728	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_DS);
7729	if (fRealOnV86Active)
7730	pCtx->ds.Attr.u = pVmcsInfoShared->RealMode.AttrDS.u;
7731	}
7732	if (fWhat & CPUMCTX_EXTRN_ES)
7733	{
7734	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_ES);
7735	if (fRealOnV86Active)
7736	pCtx->es.Attr.u = pVmcsInfoShared->RealMode.AttrES.u;
7737	}
7738	if (fWhat & CPUMCTX_EXTRN_FS)
7739	{
7740	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_FS);
7741	if (fRealOnV86Active)
7742	pCtx->fs.Attr.u = pVmcsInfoShared->RealMode.AttrFS.u;
7743	}
7744	if (fWhat & CPUMCTX_EXTRN_GS)
7745	{
7746	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_GS);
7747	if (fRealOnV86Active)
7748	pCtx->gs.Attr.u = pVmcsInfoShared->RealMode.AttrGS.u;
7749	}
7750	}
7751
7752	if (fWhat & CPUMCTX_EXTRN_TABLE_MASK)
7753	{
7754	if (fWhat & CPUMCTX_EXTRN_LDTR)
7755	hmR0VmxImportGuestLdtr(pVCpu);
7756
7757	if (fWhat & CPUMCTX_EXTRN_GDTR)
7758	{
7759	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, &pCtx->gdtr.pGdt); AssertRC(rc);
7760	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val); AssertRC(rc);
7761	pCtx->gdtr.cbGdt = u32Val;
7762	}
7763
7764	/* Guest IDTR. */
7765	if (fWhat & CPUMCTX_EXTRN_IDTR)
7766	{
7767	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, &pCtx->idtr.pIdt); AssertRC(rc);
7768	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val); AssertRC(rc);
7769	pCtx->idtr.cbIdt = u32Val;
7770	}
7771
7772	/* Guest TR. */
7773	if (fWhat & CPUMCTX_EXTRN_TR)
7774	{
7775	/* Real-mode emulation using virtual-8086 mode has the fake TSS (pRealModeTSS) in TR,
7776	don't need to import that one. */
7777	if (!pVmcsInfo->pShared->RealMode.fRealOnV86Active)
7778	hmR0VmxImportGuestTr(pVCpu);
7779	}
7780	}
7781
7782	if (fWhat & CPUMCTX_EXTRN_DR7)
7783	{
7784	if (!pVCpu->hmr0.s.fUsingHyperDR7)
7785	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_DR7, &pCtx->dr[7]); AssertRC(rc);
7786	}
7787
7788	if (fWhat & CPUMCTX_EXTRN_SYSENTER_MSRS)
7789	{
7790	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SYSENTER_EIP, &pCtx->SysEnter.eip); AssertRC(rc);
7791	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SYSENTER_ESP, &pCtx->SysEnter.esp); AssertRC(rc);
7792	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, &u32Val); AssertRC(rc);
7793	pCtx->SysEnter.cs = u32Val;
7794	}
7795
7796	if (fWhat & CPUMCTX_EXTRN_KERNEL_GS_BASE)
7797	{
7798	if ( pVM->hmr0.s.fAllow64BitGuests
7799	&& (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
7800	pCtx->msrKERNELGSBASE = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
7801	}
7802
7803	if (fWhat & CPUMCTX_EXTRN_SYSCALL_MSRS)
7804	{
7805	if ( pVM->hmr0.s.fAllow64BitGuests
7806	&& (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
7807	{
7808	pCtx->msrLSTAR = ASMRdMsr(MSR_K8_LSTAR);
7809	pCtx->msrSTAR = ASMRdMsr(MSR_K6_STAR);
7810	pCtx->msrSFMASK = ASMRdMsr(MSR_K8_SF_MASK);
7811	}
7812	}
7813
7814	if (fWhat & (CPUMCTX_EXTRN_TSC_AUX \| CPUMCTX_EXTRN_OTHER_MSRS))
7815	{
7816	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
7817	PCVMXAUTOMSR pMsrs = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
7818	uint32_t const cMsrs = pVmcsInfo->cExitMsrStore;
7819	Assert(pMsrs);
7820	Assert(cMsrs <= VMX_MISC_MAX_MSRS(g_HmMsrs.u.vmx.u64Misc));
7821	Assert(sizeof(pMsrs) cMsrs <= X86_PAGE_4K_SIZE);
7822	for (uint32_t i = 0; i < cMsrs; i++)
7823	{
7824	uint32_t const idMsr = pMsrs[i].u32Msr;
7825	switch (idMsr)
7826	{
7827	case MSR_K8_TSC_AUX: CPUMSetGuestTscAux(pVCpu, pMsrs[i].u64Value); break;
7828	case MSR_IA32_SPEC_CTRL: CPUMSetGuestSpecCtrl(pVCpu, pMsrs[i].u64Value); break;
7829	case MSR_K6_EFER: /* Can't be changed without causing a VM-exit */ break;
7830	default:
7831	{
7832	uint32_t idxLbrMsr;
7833	if (pVM->hmr0.s.vmx.fLbr)
7834	{
7835	if (hmR0VmxIsLbrBranchFromMsr(pVM, idMsr, &idxLbrMsr))
7836	{
7837	Assert(idxLbrMsr < RT_ELEMENTS(pVmcsInfoShared->au64LbrFromIpMsr));
7838	pVmcsInfoShared->au64LbrFromIpMsr[idxLbrMsr] = pMsrs[i].u64Value;
7839	break;
7840	}
7841	if (hmR0VmxIsLbrBranchToMsr(pVM, idMsr, &idxLbrMsr))
7842	{
7843	Assert(idxLbrMsr < RT_ELEMENTS(pVmcsInfoShared->au64LbrFromIpMsr));
7844	pVmcsInfoShared->au64LbrToIpMsr[idxLbrMsr] = pMsrs[i].u64Value;
7845	break;
7846	}
7847	if (idMsr == pVM->hmr0.s.vmx.idLbrTosMsr)
7848	{
7849	pVmcsInfoShared->u64LbrTosMsr = pMsrs[i].u64Value;
7850	break;
7851	}
7852	/* Fallthru (no break) */
7853	}
7854	pCtx->fExtrn = 0;
7855	pVCpu->hm.s.u32HMError = pMsrs->u32Msr;
7856	ASMSetFlags(fEFlags);
7857	AssertMsgFailed(("Unexpected MSR in auto-load/store area. idMsr=%#RX32 cMsrs=%u\n", idMsr, cMsrs));
7858	return VERR_HM_UNEXPECTED_LD_ST_MSR;
7859	}
7860	}
7861	}
7862	}
7863
7864	if (fWhat & CPUMCTX_EXTRN_CR_MASK)
7865	{
7866	if (fWhat & CPUMCTX_EXTRN_CR0)
7867	{
7868	uint64_t u64Cr0;
7869	uint64_t u64Shadow;
7870	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64Cr0); AssertRC(rc);
7871	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, &u64Shadow); AssertRC(rc);
7872	#ifndef VBOX_WITH_NESTED_HWVIRT_VMX
7873	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7874	\| (u64Shadow & pVmcsInfo->u64Cr0Mask);
7875	#else
7876	if (!CPUMIsGuestInVmxNonRootMode(pCtx))
7877	{
7878	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7879	\| (u64Shadow & pVmcsInfo->u64Cr0Mask);
7880	}
7881	else
7882	{
7883	/*
7884	* We've merged the guest and nested-guest's CR0 guest/host mask while executing
7885	* the nested-guest using hardware-assisted VMX. Accordingly we need to
7886	* re-construct CR0. See @bugref{9180#c95} for details.
7887	*/
7888	PCVMXVMCSINFO pVmcsInfoGst = &pVCpu->hmr0.s.vmx.VmcsInfo;
7889	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
7890	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7891	\| (pVmcsNstGst->u64GuestCr0.u & pVmcsNstGst->u64Cr0Mask.u)
7892	\| (u64Shadow & (pVmcsInfoGst->u64Cr0Mask & ~pVmcsNstGst->u64Cr0Mask.u));
7893	}
7894	#endif
7895	VMMRZCallRing3Disable(pVCpu); /* May call into PGM which has Log statements. */
7896	CPUMSetGuestCR0(pVCpu, u64Cr0);
7897	VMMRZCallRing3Enable(pVCpu);
7898	}
7899
7900	if (fWhat & CPUMCTX_EXTRN_CR4)
7901	{
7902	uint64_t u64Cr4;
7903	uint64_t u64Shadow;
7904	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR4, &u64Cr4); AssertRC(rc);
7905	rc \|= VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, &u64Shadow); AssertRC(rc);
7906	#ifndef VBOX_WITH_NESTED_HWVIRT_VMX
7907	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7908	\| (u64Shadow & pVmcsInfo->u64Cr4Mask);
7909	#else
7910	if (!CPUMIsGuestInVmxNonRootMode(pCtx))
7911	{
7912	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7913	\| (u64Shadow & pVmcsInfo->u64Cr4Mask);
7914	}
7915	else
7916	{
7917	/*
7918	* We've merged the guest and nested-guest's CR4 guest/host mask while executing
7919	* the nested-guest using hardware-assisted VMX. Accordingly we need to
7920	* re-construct CR4. See @bugref{9180#c95} for details.
7921	*/
7922	PCVMXVMCSINFO pVmcsInfoGst = &pVCpu->hmr0.s.vmx.VmcsInfo;
7923	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
7924	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7925	\| (pVmcsNstGst->u64GuestCr4.u & pVmcsNstGst->u64Cr4Mask.u)
7926	\| (u64Shadow & (pVmcsInfoGst->u64Cr4Mask & ~pVmcsNstGst->u64Cr4Mask.u));
7927	}
7928	#endif
7929	pCtx->cr4 = u64Cr4;
7930	}
7931
7932	if (fWhat & CPUMCTX_EXTRN_CR3)
7933	{
7934	/* CR0.PG bit changes are always intercepted, so it's up to date. */
7935	if ( pVM->hmr0.s.vmx.fUnrestrictedGuest
7936	\|\| ( pVM->hmr0.s.fNestedPaging
7937	&& CPUMIsGuestPagingEnabledEx(pCtx)))
7938	{
7939	uint64_t u64Cr3;
7940	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR3, &u64Cr3); AssertRC(rc);
7941	if (pCtx->cr3 != u64Cr3)
7942	{
7943	pCtx->cr3 = u64Cr3;
7944	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
7945	}
7946
7947	/* If the guest is in PAE mode, sync back the PDPE's into the guest state.
7948	Note: CR4.PAE, CR0.PG, EFER MSR changes are always intercepted, so they're up to date. */
7949	if (CPUMIsGuestInPAEModeEx(pCtx))
7950	{
7951	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &pVCpu->hm.s.aPdpes[0].u); AssertRC(rc);
7952	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &pVCpu->hm.s.aPdpes[1].u); AssertRC(rc);
7953	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &pVCpu->hm.s.aPdpes[2].u); AssertRC(rc);
7954	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &pVCpu->hm.s.aPdpes[3].u); AssertRC(rc);
7955	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES);
7956	}
7957	}
7958	}
7959	}
7960
7961	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7962	if (fWhat & CPUMCTX_EXTRN_HWVIRT)
7963	{
7964	if ( (pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
7965	&& !CPUMIsGuestInVmxNonRootMode(pCtx))
7966	{
7967	Assert(CPUMIsGuestInVmxRootMode(pCtx));
7968	rc = hmR0VmxCopyShadowToNstGstVmcs(pVCpu, pVmcsInfo);
7969	if (RT_SUCCESS(rc))
7970	{ /* likely */ }
7971	else
7972	break;
7973	}
7974	}
7975	#endif
7976	} while (0);
7977
7978	if (RT_SUCCESS(rc))
7979	{
7980	/* Update fExtrn. */
7981	pCtx->fExtrn &= ~fWhat;
7982
7983	/* If everything has been imported, clear the HM keeper bit. */
7984	if (!(pCtx->fExtrn & HMVMX_CPUMCTX_EXTRN_ALL))
7985	{
7986	pCtx->fExtrn &= ~CPUMCTX_EXTRN_KEEPER_HM;
7987	Assert(!pCtx->fExtrn);
7988	}
7989	}
7990	}
7991	else
7992	AssertMsg(!pCtx->fExtrn \|\| (pCtx->fExtrn & HMVMX_CPUMCTX_EXTRN_ALL), ("%#RX64\n", pCtx->fExtrn));
7993
7994	/*
7995	* Restore interrupts.
7996	*/
7997	ASMSetFlags(fEFlags);
7998
7999	STAM_PROFILE_ADV_STOP(& pVCpu->hm.s.StatImportGuestState, x);
8000
8001	if (RT_SUCCESS(rc))
8002	{ /* likely */ }
8003	else
8004	return rc;
8005
8006	/*
8007	* Honor any pending CR3 updates.
8008	*
8009	* Consider this scenario: VM-exit -> VMMRZCallRing3Enable() -> do stuff that causes a longjmp -> VMXR0CallRing3Callback()
8010	* -> VMMRZCallRing3Disable() -> hmR0VmxImportGuestState() -> Sets VMCPU_FF_HM_UPDATE_CR3 pending -> return from the longjmp
8011	* -> continue with VM-exit handling -> hmR0VmxImportGuestState() and here we are.
8012	*
8013	* The reason for such complicated handling is because VM-exits that call into PGM expect CR3 to be up-to-date and thus
8014	* if any CR3-saves -before- the VM-exit (longjmp) postponed the CR3 update via the force-flag, any VM-exit handler that
8015	* calls into PGM when it re-saves CR3 will end up here and we call PGMUpdateCR3(). This is why the code below should
8016	* -NOT- check if CPUMCTX_EXTRN_CR3 is set!
8017	*
8018	* The longjmp exit path can't check these CR3 force-flags and call code that takes a lock again. We cover for it here.
8019	*/
8020	if (VMMRZCallRing3IsEnabled(pVCpu))
8021	{
8022	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
8023	{
8024	Assert(!(ASMAtomicUoReadU64(&pCtx->fExtrn) & CPUMCTX_EXTRN_CR3));
8025	PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
8026	}
8027
8028	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
8029	PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
8030
8031	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
8032	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
8033	}
8034
8035	return VINF_SUCCESS;
8036	}
8037
8038
8039	/**
8040	* Saves the guest state from the VMCS into the guest-CPU context.
8041	*
8042	* @returns VBox status code.
8043	* @param pVCpu The cross context virtual CPU structure.
8044	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
8045	*/
8046	VMMR0DECL(int) VMXR0ImportStateOnDemand(PVMCPUCC pVCpu, uint64_t fWhat)
8047	{
8048	AssertPtr(pVCpu);
8049	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8050	return hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fWhat);
8051	}
8052
8053
8054	/**
8055	* Check per-VM and per-VCPU force flag actions that require us to go back to
8056	* ring-3 for one reason or another.
8057	*
8058	* @returns Strict VBox status code (i.e. informational status codes too)
8059	* @retval VINF_SUCCESS if we don't have any actions that require going back to
8060	* ring-3.
8061	* @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync.
8062	* @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware
8063	* interrupts)
8064	* @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires
8065	* all EMTs to be in ring-3.
8066	* @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests.
8067	* @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return
8068	* to the EM loop.
8069	*
8070	* @param pVCpu The cross context virtual CPU structure.
8071	* @param pVmxTransient The VMX-transient structure.
8072	* @param fStepping Whether we are single-stepping the guest using the
8073	* hypervisor debugger.
8074	*
8075	* @remarks This might cause nested-guest VM-exits, caller must check if the guest
8076	* is no longer in VMX non-root mode.
8077	*/
8078	static VBOXSTRICTRC hmR0VmxCheckForceFlags(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, bool fStepping)
8079	{
8080	Assert(VMMRZCallRing3IsEnabled(pVCpu));
8081
8082	/*
8083	* Update pending interrupts into the APIC's IRR.
8084	*/
8085	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UPDATE_APIC))
8086	APICUpdatePendingInterrupts(pVCpu);
8087
8088	/*
8089	* Anything pending? Should be more likely than not if we're doing a good job.
8090	*/
8091	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
8092	if ( !fStepping
8093	? !VM_FF_IS_ANY_SET(pVM, VM_FF_HP_R0_PRE_HM_MASK)
8094	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HP_R0_PRE_HM_MASK)
8095	: !VM_FF_IS_ANY_SET(pVM, VM_FF_HP_R0_PRE_HM_STEP_MASK)
8096	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) )
8097	return VINF_SUCCESS;
8098
8099	/* Pending PGM C3 sync. */
8100	if (VMCPU_FF_IS_ANY_SET(pVCpu,VMCPU_FF_PGM_SYNC_CR3 \| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL))
8101	{
8102	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8103	Assert(!(ASMAtomicUoReadU64(&pCtx->fExtrn) & (CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_CR3 \| CPUMCTX_EXTRN_CR4)));
8104	VBOXSTRICTRC rcStrict = PGMSyncCR3(pVCpu, pCtx->cr0, pCtx->cr3, pCtx->cr4,
8105	VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
8106	if (rcStrict != VINF_SUCCESS)
8107	{
8108	AssertRC(VBOXSTRICTRC_VAL(rcStrict));
8109	Log4Func(("PGMSyncCR3 forcing us back to ring-3. rc2=%d\n", VBOXSTRICTRC_VAL(rcStrict)));
8110	return rcStrict;
8111	}
8112	}
8113
8114	/* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */
8115	if ( VM_FF_IS_ANY_SET(pVM, VM_FF_HM_TO_R3_MASK)
8116	\|\| VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
8117	{
8118	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
8119	int rc = RT_LIKELY(!VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_RAW_TO_R3 : VINF_EM_NO_MEMORY;
8120	Log4Func(("HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc));
8121	return rc;
8122	}
8123
8124	/* Pending VM request packets, such as hardware interrupts. */
8125	if ( VM_FF_IS_SET(pVM, VM_FF_REQUEST)
8126	\|\| VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_REQUEST))
8127	{
8128	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchVmReq);
8129	Log4Func(("Pending VM request forcing us back to ring-3\n"));
8130	return VINF_EM_PENDING_REQUEST;
8131	}
8132
8133	/* Pending PGM pool flushes. */
8134	if (VM_FF_IS_SET(pVM, VM_FF_PGM_POOL_FLUSH_PENDING))
8135	{
8136	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPgmPoolFlush);
8137	Log4Func(("PGM pool flush pending forcing us back to ring-3\n"));
8138	return VINF_PGM_POOL_FLUSH_PENDING;
8139	}
8140
8141	/* Pending DMA requests. */
8142	if (VM_FF_IS_SET(pVM, VM_FF_PDM_DMA))
8143	{
8144	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchDma);
8145	Log4Func(("Pending DMA request forcing us back to ring-3\n"));
8146	return VINF_EM_RAW_TO_R3;
8147	}
8148
8149	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8150	/*
8151	* Pending nested-guest events.
8152	*
8153	* Please note the priority of these events are specified and important.
8154	* See Intel spec. 29.4.3.2 "APIC-Write Emulation".
8155	* See Intel spec. 6.9 "Priority Among Simultaneous Exceptions And Interrupts".
8156	*/
8157	if (pVmxTransient->fIsNestedGuest)
8158	{
8159	/* Pending nested-guest APIC-write. */
8160	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_APIC_WRITE))
8161	{
8162	Log4Func(("Pending nested-guest APIC-write\n"));
8163	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitApicWrite(pVCpu);
8164	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8165	return rcStrict;
8166	}
8167
8168	/* Pending nested-guest monitor-trap flag (MTF). */
8169	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_MTF))
8170	{
8171	Log4Func(("Pending nested-guest MTF\n"));
8172	VBOXSTRICTRC rcStrict = IEMExecVmxVmexit(pVCpu, VMX_EXIT_MTF, 0 /* uExitQual */);
8173	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8174	return rcStrict;
8175	}
8176
8177	/* Pending nested-guest VMX-preemption timer expired. */
8178	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_PREEMPT_TIMER))
8179	{
8180	Log4Func(("Pending nested-guest preempt timer\n"));
8181	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitPreemptTimer(pVCpu);
8182	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8183	return rcStrict;
8184	}
8185	}
8186	#else
8187	NOREF(pVmxTransient);
8188	#endif
8189
8190	return VINF_SUCCESS;
8191	}
8192
8193
8194	/**
8195	* Converts any TRPM trap into a pending HM event. This is typically used when
8196	* entering from ring-3 (not longjmp returns).
8197	*
8198	* @param pVCpu The cross context virtual CPU structure.
8199	*/
8200	static void hmR0VmxTrpmTrapToPendingEvent(PVMCPUCC pVCpu)
8201	{
8202	Assert(TRPMHasTrap(pVCpu));
8203	Assert(!pVCpu->hm.s.Event.fPending);
8204
8205	uint8_t uVector;
8206	TRPMEVENT enmTrpmEvent;
8207	uint32_t uErrCode;
8208	RTGCUINTPTR GCPtrFaultAddress;
8209	uint8_t cbInstr;
8210	bool fIcebp;
8211
8212	int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr, &fIcebp);
8213	AssertRC(rc);
8214
8215	uint32_t u32IntInfo;
8216	u32IntInfo = uVector \| VMX_IDT_VECTORING_INFO_VALID;
8217	u32IntInfo \|= HMTrpmEventTypeToVmxEventType(uVector, enmTrpmEvent, fIcebp);
8218
8219	rc = TRPMResetTrap(pVCpu);
8220	AssertRC(rc);
8221	Log4(("TRPM->HM event: u32IntInfo=%#RX32 enmTrpmEvent=%d cbInstr=%u uErrCode=%#RX32 GCPtrFaultAddress=%#RGv\n",
8222	u32IntInfo, enmTrpmEvent, cbInstr, uErrCode, GCPtrFaultAddress));
8223
8224	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, uErrCode, GCPtrFaultAddress);
8225	}
8226
8227
8228	/**
8229	* Converts the pending HM event into a TRPM trap.
8230	*
8231	* @param pVCpu The cross context virtual CPU structure.
8232	*/
8233	static void hmR0VmxPendingEventToTrpmTrap(PVMCPUCC pVCpu)
8234	{
8235	Assert(pVCpu->hm.s.Event.fPending);
8236
8237	/* If a trap was already pending, we did something wrong! */
8238	Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo /, NULL / pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP);
8239
8240	uint32_t const u32IntInfo = pVCpu->hm.s.Event.u64IntInfo;
8241	uint32_t const uVector = VMX_IDT_VECTORING_INFO_VECTOR(u32IntInfo);
8242	TRPMEVENT const enmTrapType = HMVmxEventTypeToTrpmEventType(u32IntInfo);
8243
8244	Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, enmTrapType));
8245
8246	int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType);
8247	AssertRC(rc);
8248
8249	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(u32IntInfo))
8250	TRPMSetErrorCode(pVCpu, pVCpu->hm.s.Event.u32ErrCode);
8251
8252	if (VMX_IDT_VECTORING_INFO_IS_XCPT_PF(u32IntInfo))
8253	TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress);
8254	else
8255	{
8256	uint8_t const uVectorType = VMX_IDT_VECTORING_INFO_TYPE(u32IntInfo);
8257	switch (uVectorType)
8258	{
8259	case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
8260	TRPMSetTrapDueToIcebp(pVCpu);
8261	RT_FALL_THRU();
8262	case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
8263	case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT:
8264	{
8265	AssertMsg( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
8266	\|\| ( uVector == X86_XCPT_BP /* INT3 */
8267	\|\| uVector == X86_XCPT_OF /* INTO */
8268	\|\| uVector == X86_XCPT_DB /* INT1 (ICEBP) */),
8269	("Invalid vector: uVector=%#x uVectorType=%#x\n", uVector, uVectorType));
8270	TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr);
8271	break;
8272	}
8273	}
8274	}
8275
8276	/* We're now done converting the pending event. */
8277	pVCpu->hm.s.Event.fPending = false;
8278	}
8279
8280
8281	/**
8282	* Sets the interrupt-window exiting control in the VMCS which instructs VT-x to
8283	* cause a VM-exit as soon as the guest is in a state to receive interrupts.
8284	*
8285	* @param pVmcsInfo The VMCS info. object.
8286	*/
8287	static void hmR0VmxSetIntWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8288	{
8289	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_INT_WINDOW_EXIT)
8290	{
8291	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT))
8292	{
8293	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_INT_WINDOW_EXIT;
8294	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8295	AssertRC(rc);
8296	}
8297	} /* else we will deliver interrupts whenever the guest Vm-exits next and is in a state to receive the interrupt. */
8298	}
8299
8300
8301	/**
8302	* Clears the interrupt-window exiting control in the VMCS.
8303	*
8304	* @param pVmcsInfo The VMCS info. object.
8305	*/
8306	DECLINLINE(void) hmR0VmxClearIntWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8307	{
8308	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT)
8309	{
8310	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_INT_WINDOW_EXIT;
8311	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8312	AssertRC(rc);
8313	}
8314	}
8315
8316
8317	/**
8318	* Sets the NMI-window exiting control in the VMCS which instructs VT-x to
8319	* cause a VM-exit as soon as the guest is in a state to receive NMIs.
8320	*
8321	* @param pVmcsInfo The VMCS info. object.
8322	*/
8323	static void hmR0VmxSetNmiWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8324	{
8325	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
8326	{
8327	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))
8328	{
8329	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_NMI_WINDOW_EXIT;
8330	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8331	AssertRC(rc);
8332	Log4Func(("Setup NMI-window exiting\n"));
8333	}
8334	} /* else we will deliver NMIs whenever we VM-exit next, even possibly nesting NMIs. Can't be helped on ancient CPUs. */
8335	}
8336
8337
8338	/**
8339	* Clears the NMI-window exiting control in the VMCS.
8340	*
8341	* @param pVmcsInfo The VMCS info. object.
8342	*/
8343	DECLINLINE(void) hmR0VmxClearNmiWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8344	{
8345	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
8346	{
8347	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_NMI_WINDOW_EXIT;
8348	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8349	AssertRC(rc);
8350	}
8351	}
8352
8353
8354	/**
8355	* Does the necessary state syncing before returning to ring-3 for any reason
8356	* (longjmp, preemption, voluntary exits to ring-3) from VT-x.
8357	*
8358	* @returns VBox status code.
8359	* @param pVCpu The cross context virtual CPU structure.
8360	* @param fImportState Whether to import the guest state from the VMCS back
8361	* to the guest-CPU context.
8362	*
8363	* @remarks No-long-jmp zone!!!
8364	*/
8365	static int hmR0VmxLeave(PVMCPUCC pVCpu, bool fImportState)
8366	{
8367	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8368	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8369
8370	RTCPUID const idCpu = RTMpCpuId();
8371	Log4Func(("HostCpuId=%u\n", idCpu));
8372
8373	/*
8374	* !!! IMPORTANT !!!
8375	* If you modify code here, check whether VMXR0CallRing3Callback() needs to be updated too.
8376	*/
8377
8378	/* Save the guest state if necessary. */
8379	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8380	if (fImportState)
8381	{
8382	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
8383	AssertRCReturn(rc, rc);
8384	}
8385
8386	/* Restore host FPU state if necessary. We will resync on next R0 reentry. */
8387	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
8388	Assert(!CPUMIsGuestFPUStateActive(pVCpu));
8389
8390	/* Restore host debug registers if necessary. We will resync on next R0 reentry. */
8391	#ifdef VBOX_STRICT
8392	if (CPUMIsHyperDebugStateActive(pVCpu))
8393	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MOV_DR_EXIT);
8394	#endif
8395	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
8396	Assert(!CPUMIsGuestDebugStateActive(pVCpu));
8397	Assert(!CPUMIsHyperDebugStateActive(pVCpu));
8398
8399	/* Restore host-state bits that VT-x only restores partially. */
8400	if (pVCpu->hmr0.s.vmx.fRestoreHostFlags > VMX_RESTORE_HOST_REQUIRED)
8401	{
8402	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hmr0.s.vmx.fRestoreHostFlags, idCpu));
8403	VMXRestoreHostState(pVCpu->hmr0.s.vmx.fRestoreHostFlags, &pVCpu->hmr0.s.vmx.RestoreHost);
8404	}
8405	pVCpu->hmr0.s.vmx.fRestoreHostFlags = 0;
8406
8407	/* Restore the lazy host MSRs as we're leaving VT-x context. */
8408	if (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
8409	{
8410	/* We shouldn't restore the host MSRs without saving the guest MSRs first. */
8411	if (!fImportState)
8412	{
8413	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_KERNEL_GS_BASE \| CPUMCTX_EXTRN_SYSCALL_MSRS);
8414	AssertRCReturn(rc, rc);
8415	}
8416	hmR0VmxLazyRestoreHostMsrs(pVCpu);
8417	Assert(!pVCpu->hmr0.s.vmx.fLazyMsrs);
8418	}
8419	else
8420	pVCpu->hmr0.s.vmx.fLazyMsrs = 0;
8421
8422	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
8423	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = false;
8424
8425	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry);
8426	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatImportGuestState);
8427	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExportGuestState);
8428	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatPreExit);
8429	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitHandling);
8430	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitIO);
8431	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitMovCRx);
8432	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitXcptNmi);
8433	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitVmentry);
8434	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
8435
8436	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
8437
8438	/** @todo This partially defeats the purpose of having preemption hooks.
8439	* The problem is, deregistering the hooks should be moved to a place that
8440	* lasts until the EMT is about to be destroyed not everytime while leaving HM
8441	* context.
8442	*/
8443	int rc = hmR0VmxClearVmcs(pVmcsInfo);
8444	AssertRCReturn(rc, rc);
8445
8446	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8447	/*
8448	* A valid shadow VMCS is made active as part of VM-entry. It is necessary to
8449	* clear a shadow VMCS before allowing that VMCS to become active on another
8450	* logical processor. We may or may not be importing guest state which clears
8451	* it, so cover for it here.
8452	*
8453	* See Intel spec. 24.11.1 "Software Use of Virtual-Machine Control Structures".
8454	*/
8455	if ( pVmcsInfo->pvShadowVmcs
8456	&& pVmcsInfo->fShadowVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
8457	{
8458	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
8459	AssertRCReturn(rc, rc);
8460	}
8461
8462	/*
8463	* Flag that we need to re-export the host state if we switch to this VMCS before
8464	* executing guest or nested-guest code.
8465	*/
8466	pVmcsInfo->idHostCpuState = NIL_RTCPUID;
8467	#endif
8468
8469	Log4Func(("Cleared Vmcs. HostCpuId=%u\n", idCpu));
8470	NOREF(idCpu);
8471	return VINF_SUCCESS;
8472	}
8473
8474
8475	/**
8476	* Leaves the VT-x session.
8477	*
8478	* @returns VBox status code.
8479	* @param pVCpu The cross context virtual CPU structure.
8480	*
8481	* @remarks No-long-jmp zone!!!
8482	*/
8483	static int hmR0VmxLeaveSession(PVMCPUCC pVCpu)
8484	{
8485	HM_DISABLE_PREEMPT(pVCpu);
8486	HMVMX_ASSERT_CPU_SAFE(pVCpu);
8487	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8488	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8489
8490	/* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before
8491	and done this from the VMXR0ThreadCtxCallback(). */
8492	if (!pVCpu->hmr0.s.fLeaveDone)
8493	{
8494	int rc2 = hmR0VmxLeave(pVCpu, true /* fImportState */);
8495	AssertRCReturnStmt(rc2, HM_RESTORE_PREEMPT(), rc2);
8496	pVCpu->hmr0.s.fLeaveDone = true;
8497	}
8498	Assert(!pVCpu->cpum.GstCtx.fExtrn);
8499
8500	/*
8501	* !!! IMPORTANT !!!
8502	* If you modify code here, make sure to check whether VMXR0CallRing3Callback() needs to be updated too.
8503	*/
8504
8505	/* Deregister hook now that we've left HM context before re-enabling preemption. */
8506	/** @todo Deregistering here means we need to VMCLEAR always
8507	* (longjmp/exit-to-r3) in VT-x which is not efficient, eliminate need
8508	* for calling VMMR0ThreadCtxHookDisable here! */
8509	VMMR0ThreadCtxHookDisable(pVCpu);
8510
8511	/* Leave HM context. This takes care of local init (term) and deregistering the longjmp-to-ring-3 callback. */
8512	int rc = HMR0LeaveCpu(pVCpu);
8513	HM_RESTORE_PREEMPT();
8514	return rc;
8515	}
8516
8517
8518	/**
8519	* Does the necessary state syncing before doing a longjmp to ring-3.
8520	*
8521	* @returns VBox status code.
8522	* @param pVCpu The cross context virtual CPU structure.
8523	*
8524	* @remarks No-long-jmp zone!!!
8525	*/
8526	DECLINLINE(int) hmR0VmxLongJmpToRing3(PVMCPUCC pVCpu)
8527	{
8528	return hmR0VmxLeaveSession(pVCpu);
8529	}
8530
8531
8532	/**
8533	* Take necessary actions before going back to ring-3.
8534	*
8535	* An action requires us to go back to ring-3. This function does the necessary
8536	* steps before we can safely return to ring-3. This is not the same as longjmps
8537	* to ring-3, this is voluntary and prepares the guest so it may continue
8538	* executing outside HM (recompiler/IEM).
8539	*
8540	* @returns VBox status code.
8541	* @param pVCpu The cross context virtual CPU structure.
8542	* @param rcExit The reason for exiting to ring-3. Can be
8543	* VINF_VMM_UNKNOWN_RING3_CALL.
8544	*/
8545	static int hmR0VmxExitToRing3(PVMCPUCC pVCpu, VBOXSTRICTRC rcExit)
8546	{
8547	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
8548
8549	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8550	if (RT_UNLIKELY(rcExit == VERR_VMX_INVALID_VMCS_PTR))
8551	{
8552	VMXGetCurrentVmcs(&pVCpu->hm.s.vmx.LastError.HCPhysCurrentVmcs);
8553	pVCpu->hm.s.vmx.LastError.u32VmcsRev = (uint32_t )pVmcsInfo->pvVmcs;
8554	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hmr0.s.idEnteredCpu;
8555	/* LastError.idCurrentCpu was updated in hmR0VmxPreRunGuestCommitted(). */
8556	}
8557
8558	/* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */
8559	VMMRZCallRing3Disable(pVCpu);
8560	Log4Func(("rcExit=%d\n", VBOXSTRICTRC_VAL(rcExit)));
8561
8562	/*
8563	* Convert any pending HM events back to TRPM due to premature exits to ring-3.
8564	* We need to do this only on returns to ring-3 and not for longjmps to ring3.
8565	*
8566	* This is because execution may continue from ring-3 and we would need to inject
8567	* the event from there (hence place it back in TRPM).
8568	*/
8569	if (pVCpu->hm.s.Event.fPending)
8570	{
8571	hmR0VmxPendingEventToTrpmTrap(pVCpu);
8572	Assert(!pVCpu->hm.s.Event.fPending);
8573
8574	/* Clear the events from the VMCS. */
8575	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, 0); AssertRC(rc);
8576	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, 0); AssertRC(rc);
8577	}
8578	#ifdef VBOX_STRICT
8579	/*
8580	* We check for rcExit here since for errors like VERR_VMX_UNABLE_TO_START_VM (which are
8581	* fatal), we don't care about verifying duplicate injection of events. Errors like
8582	* VERR_EM_INTERPRET are converted to their VINF_* counterparts -prior- to calling this
8583	* function so those should and will be checked below.
8584	*/
8585	else if (RT_SUCCESS(rcExit))
8586	{
8587	/*
8588	* Ensure we don't accidentally clear a pending HM event without clearing the VMCS.
8589	* This can be pretty hard to debug otherwise, interrupts might get injected twice
8590	* occasionally, see @bugref{9180#c42}.
8591	*
8592	* However, if the VM-entry failed, any VM entry-interruption info. field would
8593	* be left unmodified as the event would not have been injected to the guest. In
8594	* such cases, don't assert, we're not going to continue guest execution anyway.
8595	*/
8596	uint32_t uExitReason;
8597	uint32_t uEntryIntInfo;
8598	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
8599	rc \|= VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &uEntryIntInfo);
8600	AssertRC(rc);
8601	AssertMsg(VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason) \|\| !VMX_ENTRY_INT_INFO_IS_VALID(uEntryIntInfo),
8602	("uExitReason=%#RX32 uEntryIntInfo=%#RX32 rcExit=%d\n", uExitReason, uEntryIntInfo, VBOXSTRICTRC_VAL(rcExit)));
8603	}
8604	#endif
8605
8606	/*
8607	* Clear the interrupt-window and NMI-window VMCS controls as we could have got
8608	* a VM-exit with higher priority than interrupt-window or NMI-window VM-exits
8609	* (e.g. TPR below threshold).
8610	*/
8611	if (!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
8612	{
8613	hmR0VmxClearIntWindowExitVmcs(pVmcsInfo);
8614	hmR0VmxClearNmiWindowExitVmcs(pVmcsInfo);
8615	}
8616
8617	/* If we're emulating an instruction, we shouldn't have any TRPM traps pending
8618	and if we're injecting an event we should have a TRPM trap pending. */
8619	AssertMsg(rcExit != VINF_EM_RAW_INJECT_TRPM_EVENT \|\| TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
8620	#ifndef DEBUG_bird /* Triggered after firing an NMI against NT4SP1, possibly a triple fault in progress. */
8621	AssertMsg(rcExit != VINF_EM_RAW_EMULATE_INSTR \|\| !TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
8622	#endif
8623
8624	/* Save guest state and restore host state bits. */
8625	int rc = hmR0VmxLeaveSession(pVCpu);
8626	AssertRCReturn(rc, rc);
8627	STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
8628
8629	/* Thread-context hooks are unregistered at this point!!! */
8630	/* Ring-3 callback notifications are unregistered at this point!!! */
8631
8632	/* Sync recompiler state. */
8633	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3);
8634	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR
8635	\| CPUM_CHANGED_LDTR
8636	\| CPUM_CHANGED_GDTR
8637	\| CPUM_CHANGED_IDTR
8638	\| CPUM_CHANGED_TR
8639	\| CPUM_CHANGED_HIDDEN_SEL_REGS);
8640	if ( pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging
8641	&& CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx))
8642	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH);
8643
8644	Assert(!pVCpu->hmr0.s.fClearTrapFlag);
8645
8646	/* Update the exit-to-ring 3 reason. */
8647	pVCpu->hm.s.rcLastExitToR3 = VBOXSTRICTRC_VAL(rcExit);
8648
8649	/* On our way back from ring-3 reload the guest state if there is a possibility of it being changed. */
8650	if ( rcExit != VINF_EM_RAW_INTERRUPT
8651	\|\| CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
8652	{
8653	Assert(!(pVCpu->cpum.GstCtx.fExtrn & HMVMX_CPUMCTX_EXTRN_ALL));
8654	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
8655	}
8656
8657	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3);
8658	VMMRZCallRing3Enable(pVCpu);
8659	return rc;
8660	}
8661
8662
8663	/**
8664	* VMMRZCallRing3() callback wrapper which saves the guest state before we
8665	* longjump to ring-3 and possibly get preempted.
8666	*
8667	* @returns VBox status code.
8668	* @param pVCpu The cross context virtual CPU structure.
8669	* @param enmOperation The operation causing the ring-3 longjump.
8670	*/
8671	VMMR0DECL(int) VMXR0CallRing3Callback(PVMCPUCC pVCpu, VMMCALLRING3 enmOperation)
8672	{
8673	if (enmOperation == VMMCALLRING3_VM_R0_ASSERTION)
8674	{
8675	/*
8676	* !!! IMPORTANT !!!
8677	* If you modify code here, check whether hmR0VmxLeave() and hmR0VmxLeaveSession() needs to be updated too.
8678	* This is a stripped down version which gets out ASAP, trying to not trigger any further assertions.
8679	*/
8680	VMMRZCallRing3RemoveNotification(pVCpu);
8681	VMMRZCallRing3Disable(pVCpu);
8682	HM_DISABLE_PREEMPT(pVCpu);
8683
8684	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8685	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
8686	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
8687	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
8688
8689	/* Restore host-state bits that VT-x only restores partially. */
8690	if (pVCpu->hmr0.s.vmx.fRestoreHostFlags > VMX_RESTORE_HOST_REQUIRED)
8691	VMXRestoreHostState(pVCpu->hmr0.s.vmx.fRestoreHostFlags, &pVCpu->hmr0.s.vmx.RestoreHost);
8692	pVCpu->hmr0.s.vmx.fRestoreHostFlags = 0;
8693
8694	/* Restore the lazy host MSRs as we're leaving VT-x context. */
8695	if (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
8696	hmR0VmxLazyRestoreHostMsrs(pVCpu);
8697
8698	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
8699	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = false;
8700	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
8701
8702	/* Clear the current VMCS data back to memory (shadow VMCS if any would have been
8703	cleared as part of importing the guest state above. */
8704	hmR0VmxClearVmcs(pVmcsInfo);
8705
8706	/** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
8707	VMMR0ThreadCtxHookDisable(pVCpu);
8708
8709	/* Leave HM context. This takes care of local init (term). */
8710	HMR0LeaveCpu(pVCpu);
8711	HM_RESTORE_PREEMPT();
8712	return VINF_SUCCESS;
8713	}
8714
8715	Assert(pVCpu);
8716	Assert(VMMRZCallRing3IsEnabled(pVCpu));
8717	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
8718
8719	VMMRZCallRing3Disable(pVCpu);
8720	Assert(VMMR0IsLogFlushDisabled(pVCpu));
8721
8722	Log4Func(("-> hmR0VmxLongJmpToRing3 enmOperation=%d\n", enmOperation));
8723
8724	int rc = hmR0VmxLongJmpToRing3(pVCpu);
8725	AssertRCReturn(rc, rc);
8726
8727	VMMRZCallRing3Enable(pVCpu);
8728	return VINF_SUCCESS;
8729	}
8730
8731
8732	/**
8733	* Pushes a 2-byte value onto the real-mode (in virtual-8086 mode) guest's
8734	* stack.
8735	*
8736	* @returns Strict VBox status code (i.e. informational status codes too).
8737	* @retval VINF_EM_RESET if pushing a value to the stack caused a triple-fault.
8738	* @param pVCpu The cross context virtual CPU structure.
8739	* @param uValue The value to push to the guest stack.
8740	*/
8741	static VBOXSTRICTRC hmR0VmxRealModeGuestStackPush(PVMCPUCC pVCpu, uint16_t uValue)
8742	{
8743	/*
8744	* The stack limit is 0xffff in real-on-virtual 8086 mode. Real-mode with weird stack limits cannot be run in
8745	* virtual 8086 mode in VT-x. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
8746	* See Intel Instruction reference for PUSH and Intel spec. 22.33.1 "Segment Wraparound".
8747	*/
8748	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8749	if (pCtx->sp == 1)
8750	return VINF_EM_RESET;
8751	pCtx->sp -= sizeof(uint16_t); /* May wrap around which is expected behaviour. */
8752	int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), pCtx->ss.u64Base + pCtx->sp, &uValue, sizeof(uint16_t));
8753	AssertRC(rc);
8754	return rc;
8755	}
8756
8757
8758	/**
8759	* Injects an event into the guest upon VM-entry by updating the relevant fields
8760	* in the VM-entry area in the VMCS.
8761	*
8762	* @returns Strict VBox status code (i.e. informational status codes too).
8763	* @retval VINF_SUCCESS if the event is successfully injected into the VMCS.
8764	* @retval VINF_EM_RESET if event injection resulted in a triple-fault.
8765	*
8766	* @param pVCpu The cross context virtual CPU structure.
8767	* @param pVmxTransient The VMX-transient structure.
8768	* @param pEvent The event being injected.
8769	* @param pfIntrState Pointer to the VT-x guest-interruptibility-state. This
8770	* will be updated if necessary. This cannot not be NULL.
8771	* @param fStepping Whether we're single-stepping guest execution and should
8772	* return VINF_EM_DBG_STEPPED if the event is injected
8773	* directly (registers modified by us, not by hardware on
8774	* VM-entry).
8775	*/
8776	static VBOXSTRICTRC hmR0VmxInjectEventVmcs(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, PCHMEVENT pEvent, bool fStepping,
8777	uint32_t *pfIntrState)
8778	{
8779	/* Intel spec. 24.8.3 "VM-Entry Controls for Event Injection" specifies the interruption-information field to be 32-bits. */
8780	AssertMsg(!RT_HI_U32(pEvent->u64IntInfo), ("%#RX64\n", pEvent->u64IntInfo));
8781	Assert(pfIntrState);
8782
8783	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8784	uint32_t u32IntInfo = pEvent->u64IntInfo;
8785	uint32_t const u32ErrCode = pEvent->u32ErrCode;
8786	uint32_t const cbInstr = pEvent->cbInstr;
8787	RTGCUINTPTR const GCPtrFault = pEvent->GCPtrFaultAddress;
8788	uint8_t const uVector = VMX_ENTRY_INT_INFO_VECTOR(u32IntInfo);
8789	uint32_t const uIntType = VMX_ENTRY_INT_INFO_TYPE(u32IntInfo);
8790
8791	#ifdef VBOX_STRICT
8792	/*
8793	* Validate the error-code-valid bit for hardware exceptions.
8794	* No error codes for exceptions in real-mode.
8795	*
8796	* See Intel spec. 20.1.4 "Interrupt and Exception Handling"
8797	*/
8798	if ( uIntType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT
8799	&& !CPUMIsGuestInRealModeEx(pCtx))
8800	{
8801	switch (uVector)
8802	{
8803	case X86_XCPT_PF:
8804	case X86_XCPT_DF:
8805	case X86_XCPT_TS:
8806	case X86_XCPT_NP:
8807	case X86_XCPT_SS:
8808	case X86_XCPT_GP:
8809	case X86_XCPT_AC:
8810	AssertMsg(VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(u32IntInfo),
8811	("Error-code-valid bit not set for exception that has an error code uVector=%#x\n", uVector));
8812	RT_FALL_THRU();
8813	default:
8814	break;
8815	}
8816	}
8817
8818	/* Cannot inject an NMI when block-by-MOV SS is in effect. */
8819	Assert( uIntType != VMX_EXIT_INT_INFO_TYPE_NMI
8820	\|\| !(*pfIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
8821	#endif
8822
8823	if ( uIntType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT
8824	\|\| uIntType == VMX_EXIT_INT_INFO_TYPE_NMI
8825	\|\| uIntType == VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT
8826	\|\| uIntType == VMX_EXIT_INT_INFO_TYPE_SW_XCPT)
8827	{
8828	Assert(uVector <= X86_XCPT_LAST);
8829	Assert(uIntType != VMX_EXIT_INT_INFO_TYPE_NMI \|\| uVector == X86_XCPT_NMI);
8830	Assert(uIntType != VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT \|\| uVector == X86_XCPT_DB);
8831	STAM_COUNTER_INC(&pVCpu->hm.s.paStatInjectedXcptsR0[uVector]);
8832	}
8833	else
8834	STAM_COUNTER_INC(&pVCpu->hm.s.paStatInjectedIrqsR0[uVector & MASK_INJECT_IRQ_STAT]);
8835
8836	/*
8837	* Hardware interrupts & exceptions cannot be delivered through the software interrupt
8838	* redirection bitmap to the real mode task in virtual-8086 mode. We must jump to the
8839	* interrupt handler in the (real-mode) guest.
8840	*
8841	* See Intel spec. 20.3 "Interrupt and Exception handling in Virtual-8086 Mode".
8842	* See Intel spec. 20.1.4 "Interrupt and Exception Handling" for real-mode interrupt handling.
8843	*/
8844	if (CPUMIsGuestInRealModeEx(pCtx)) /* CR0.PE bit changes are always intercepted, so it's up to date. */
8845	{
8846	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUnrestrictedGuest)
8847	{
8848	/*
8849	* For CPUs with unrestricted guest execution enabled and with the guest
8850	* in real-mode, we must not set the deliver-error-code bit.
8851	*
8852	* See Intel spec. 26.2.1.3 "VM-Entry Control Fields".
8853	*/
8854	u32IntInfo &= ~VMX_ENTRY_INT_INFO_ERROR_CODE_VALID;
8855	}
8856	else
8857	{
8858	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
8859	Assert(PDMVmmDevHeapIsEnabled(pVM));
8860	Assert(pVM->hm.s.vmx.pRealModeTSS);
8861	Assert(!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
8862
8863	/* We require RIP, RSP, RFLAGS, CS, IDTR, import them. */
8864	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
8865	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_TABLE_MASK
8866	\| CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_RFLAGS);
8867	AssertRCReturn(rc2, rc2);
8868
8869	/* Check if the interrupt handler is present in the IVT (real-mode IDT). IDT limit is (4N - 1). */
8870	size_t const cbIdtEntry = sizeof(X86IDTR16);
8871	if (uVector * cbIdtEntry + (cbIdtEntry - 1) > pCtx->idtr.cbIdt)
8872	{
8873	/* If we are trying to inject a #DF with no valid IDT entry, return a triple-fault. */
8874	if (uVector == X86_XCPT_DF)
8875	return VINF_EM_RESET;
8876
8877	/* If we're injecting a #GP with no valid IDT entry, inject a double-fault.
8878	No error codes for exceptions in real-mode. */
8879	if (uVector == X86_XCPT_GP)
8880	{
8881	uint32_t const uXcptDfInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DF)
8882	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
8883	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
8884	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
8885	HMEVENT EventXcptDf;
8886	RT_ZERO(EventXcptDf);
8887	EventXcptDf.u64IntInfo = uXcptDfInfo;
8888	return hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &EventXcptDf, fStepping, pfIntrState);
8889	}
8890
8891	/*
8892	* If we're injecting an event with no valid IDT entry, inject a #GP.
8893	* No error codes for exceptions in real-mode.
8894	*
8895	* See Intel spec. 20.1.4 "Interrupt and Exception Handling"
8896	*/
8897	uint32_t const uXcptGpInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_GP)
8898	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
8899	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
8900	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
8901	HMEVENT EventXcptGp;
8902	RT_ZERO(EventXcptGp);
8903	EventXcptGp.u64IntInfo = uXcptGpInfo;
8904	return hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &EventXcptGp, fStepping, pfIntrState);
8905	}
8906
8907	/* Software exceptions (#BP and #OF exceptions thrown as a result of INT3 or INTO) */
8908	uint16_t uGuestIp = pCtx->ip;
8909	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_SW_XCPT)
8910	{
8911	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF);
8912	/* #BP and #OF are both benign traps, we need to resume the next instruction. */
8913	uGuestIp = pCtx->ip + (uint16_t)cbInstr;
8914	}
8915	else if (uIntType == VMX_ENTRY_INT_INFO_TYPE_SW_INT)
8916	uGuestIp = pCtx->ip + (uint16_t)cbInstr;
8917
8918	/* Get the code segment selector and offset from the IDT entry for the interrupt handler. */
8919	X86IDTR16 IdtEntry;
8920	RTGCPHYS const GCPhysIdtEntry = (RTGCPHYS)pCtx->idtr.pIdt + uVector * cbIdtEntry;
8921	rc2 = PGMPhysSimpleReadGCPhys(pVM, &IdtEntry, GCPhysIdtEntry, cbIdtEntry);
8922	AssertRCReturn(rc2, rc2);
8923
8924	/* Construct the stack frame for the interrupt/exception handler. */
8925	VBOXSTRICTRC rcStrict;
8926	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, pCtx->eflags.u32);
8927	if (rcStrict == VINF_SUCCESS)
8928	{
8929	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, pCtx->cs.Sel);
8930	if (rcStrict == VINF_SUCCESS)
8931	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, uGuestIp);
8932	}
8933
8934	/* Clear the required eflag bits and jump to the interrupt/exception handler. */
8935	if (rcStrict == VINF_SUCCESS)
8936	{
8937	pCtx->eflags.u32 &= ~(X86_EFL_IF \| X86_EFL_TF \| X86_EFL_RF \| X86_EFL_AC);
8938	pCtx->rip = IdtEntry.offSel;
8939	pCtx->cs.Sel = IdtEntry.uSel;
8940	pCtx->cs.ValidSel = IdtEntry.uSel;
8941	pCtx->cs.u64Base = IdtEntry.uSel << cbIdtEntry;
8942	if ( uIntType == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT
8943	&& uVector == X86_XCPT_PF)
8944	pCtx->cr2 = GCPtrFault;
8945
8946	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CS \| HM_CHANGED_GUEST_CR2
8947	\| HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
8948	\| HM_CHANGED_GUEST_RSP);
8949
8950	/*
8951	* If we delivered a hardware exception (other than an NMI) and if there was
8952	* block-by-STI in effect, we should clear it.
8953	*/
8954	if (*pfIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
8955	{
8956	Assert( uIntType != VMX_ENTRY_INT_INFO_TYPE_NMI
8957	&& uIntType != VMX_ENTRY_INT_INFO_TYPE_EXT_INT);
8958	Log4Func(("Clearing inhibition due to STI\n"));
8959	*pfIntrState &= ~VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
8960	}
8961
8962	Log4(("Injected real-mode: u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x Eflags=%#x CS:EIP=%04x:%04x\n",
8963	u32IntInfo, u32ErrCode, cbInstr, pCtx->eflags.u, pCtx->cs.Sel, pCtx->eip));
8964
8965	/*
8966	* The event has been truly dispatched to the guest. Mark it as no longer pending so
8967	* we don't attempt to undo it if we are returning to ring-3 before executing guest code.
8968	*/
8969	pVCpu->hm.s.Event.fPending = false;
8970
8971	/*
8972	* If we eventually support nested-guest execution without unrestricted guest execution,
8973	* we should set fInterceptEvents here.
8974	*/
8975	Assert(!pVmxTransient->fIsNestedGuest);
8976
8977	/* If we're stepping and we've changed cs:rip above, bail out of the VMX R0 execution loop. */
8978	if (fStepping)
8979	rcStrict = VINF_EM_DBG_STEPPED;
8980	}
8981	AssertMsg(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
8982	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
8983	return rcStrict;
8984	}
8985	}
8986
8987	/*
8988	* Validate.
8989	*/
8990	Assert(VMX_ENTRY_INT_INFO_IS_VALID(u32IntInfo)); /* Bit 31 (Valid bit) must be set by caller. */
8991	Assert(!(u32IntInfo & VMX_BF_ENTRY_INT_INFO_RSVD_12_30_MASK)); /* Bits 30:12 MBZ. */
8992
8993	/*
8994	* Inject the event into the VMCS.
8995	*/
8996	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, u32IntInfo);
8997	if (VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(u32IntInfo))
8998	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, u32ErrCode);
8999	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, cbInstr);
9000	AssertRC(rc);
9001
9002	/*
9003	* Update guest CR2 if this is a page-fault.
9004	*/
9005	if (VMX_ENTRY_INT_INFO_IS_XCPT_PF(u32IntInfo))
9006	pCtx->cr2 = GCPtrFault;
9007
9008	Log4(("Injecting u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x CR2=%#RX64\n", u32IntInfo, u32ErrCode, cbInstr, pCtx->cr2));
9009	return VINF_SUCCESS;
9010	}
9011
9012
9013	/**
9014	* Evaluates the event to be delivered to the guest and sets it as the pending
9015	* event.
9016	*
9017	* Toggling of interrupt force-flags here is safe since we update TRPM on premature
9018	* exits to ring-3 before executing guest code, see hmR0VmxExitToRing3(). We must
9019	* NOT restore these force-flags.
9020	*
9021	* @returns Strict VBox status code (i.e. informational status codes too).
9022	* @param pVCpu The cross context virtual CPU structure.
9023	* @param pVmxTransient The VMX-transient structure.
9024	* @param pfIntrState Where to store the VT-x guest-interruptibility state.
9025	*/
9026	static VBOXSTRICTRC hmR0VmxEvaluatePendingEvent(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t *pfIntrState)
9027	{
9028	Assert(pfIntrState);
9029	Assert(!TRPMHasTrap(pVCpu));
9030
9031	/*
9032	* Compute/update guest-interruptibility state related FFs.
9033	* The FFs will be used below while evaluating events to be injected.
9034	*/
9035	*pfIntrState = hmR0VmxGetGuestIntrStateAndUpdateFFs(pVCpu);
9036
9037	/*
9038	* Evaluate if a new event needs to be injected.
9039	* An event that's already pending has already performed all necessary checks.
9040	*/
9041	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
9042	bool const fIsNestedGuest = pVmxTransient->fIsNestedGuest;
9043	if ( !pVCpu->hm.s.Event.fPending
9044	&& !VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
9045	{
9046	/** @todo SMI. SMIs take priority over NMIs. */
9047
9048	/*
9049	* NMIs.
9050	* NMIs take priority over external interrupts.
9051	*/
9052	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
9053	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
9054	{
9055	/*
9056	* For a guest, the FF always indicates the guest's ability to receive an NMI.
9057	*
9058	* For a nested-guest, the FF always indicates the outer guest's ability to
9059	* receive an NMI while the guest-interruptibility state bit depends on whether
9060	* the nested-hypervisor is using virtual-NMIs.
9061	*/
9062	if (!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_BLOCK_NMIS))
9063	{
9064	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9065	if ( fIsNestedGuest
9066	&& CPUMIsGuestVmxPinCtlsSet(pCtx, VMX_PIN_CTLS_NMI_EXIT))
9067	return IEMExecVmxVmexitXcptNmi(pVCpu);
9068	#endif
9069	hmR0VmxSetPendingXcptNmi(pVCpu);
9070	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
9071	Log4Func(("NMI pending injection\n"));
9072
9073	/* We've injected the NMI, bail. */
9074	return VINF_SUCCESS;
9075	}
9076	else if (!fIsNestedGuest)
9077	hmR0VmxSetNmiWindowExitVmcs(pVmcsInfo);
9078	}
9079
9080	/*
9081	* External interrupts (PIC/APIC).
9082	* Once PDMGetInterrupt() returns a valid interrupt we -must- deliver it.
9083	* We cannot re-request the interrupt from the controller again.
9084	*/
9085	if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
9086	&& !pVCpu->hm.s.fSingleInstruction)
9087	{
9088	Assert(!DBGFIsStepping(pVCpu));
9089	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_RFLAGS);
9090	AssertRC(rc);
9091
9092	/*
9093	* We must not check EFLAGS directly when executing a nested-guest, use
9094	* CPUMIsGuestPhysIntrEnabled() instead as EFLAGS.IF does not control the blocking of
9095	* external interrupts when "External interrupt exiting" is set. This fixes a nasty
9096	* SMP hang while executing nested-guest VCPUs on spinlocks which aren't rescued by
9097	* other VM-exits (like a preemption timer), see @bugref{9562#c18}.
9098	*
9099	* See Intel spec. 25.4.1 "Event Blocking".
9100	*/
9101	if (CPUMIsGuestPhysIntrEnabled(pVCpu))
9102	{
9103	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9104	if ( fIsNestedGuest
9105	&& CPUMIsGuestVmxPinCtlsSet(pCtx, VMX_PIN_CTLS_EXT_INT_EXIT))
9106	{
9107	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitExtInt(pVCpu, 0 /* uVector /, true / fIntPending */);
9108	if (rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE)
9109	return rcStrict;
9110	}
9111	#endif
9112	uint8_t u8Interrupt;
9113	rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
9114	if (RT_SUCCESS(rc))
9115	{
9116	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9117	if ( fIsNestedGuest
9118	&& CPUMIsGuestVmxPinCtlsSet(pCtx, VMX_PIN_CTLS_EXT_INT_EXIT))
9119	{
9120	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitExtInt(pVCpu, u8Interrupt, false /* fIntPending */);
9121	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
9122	return rcStrict;
9123	}
9124	#endif
9125	hmR0VmxSetPendingExtInt(pVCpu, u8Interrupt);
9126	Log4Func(("External interrupt (%#x) pending injection\n", u8Interrupt));
9127	}
9128	else if (rc == VERR_APIC_INTR_MASKED_BY_TPR)
9129	{
9130	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchTprMaskedIrq);
9131
9132	if ( !fIsNestedGuest
9133	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW))
9134	hmR0VmxApicSetTprThreshold(pVmcsInfo, u8Interrupt >> 4);
9135	/* else: for nested-guests, TPR threshold is picked up while merging VMCS controls. */
9136
9137	/*
9138	* If the CPU doesn't have TPR shadowing, we will always get a VM-exit on TPR changes and
9139	* APICSetTpr() will end up setting the VMCPU_FF_INTERRUPT_APIC if required, so there is no
9140	* need to re-set this force-flag here.
9141	*/
9142	}
9143	else
9144	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
9145
9146	/* We've injected the interrupt or taken necessary action, bail. */
9147	return VINF_SUCCESS;
9148	}
9149	if (!fIsNestedGuest)
9150	hmR0VmxSetIntWindowExitVmcs(pVmcsInfo);
9151	}
9152	}
9153	else if (!fIsNestedGuest)
9154	{
9155	/*
9156	* An event is being injected or we are in an interrupt shadow. Check if another event is
9157	* pending. If so, instruct VT-x to cause a VM-exit as soon as the guest is ready to accept
9158	* the pending event.
9159	*/
9160	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
9161	hmR0VmxSetNmiWindowExitVmcs(pVmcsInfo);
9162	else if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
9163	&& !pVCpu->hm.s.fSingleInstruction)
9164	hmR0VmxSetIntWindowExitVmcs(pVmcsInfo);
9165	}
9166	/* else: for nested-guests, NMI/interrupt-window exiting will be picked up when merging VMCS controls. */
9167
9168	return VINF_SUCCESS;
9169	}
9170
9171
9172	/**
9173	* Injects any pending events into the guest if the guest is in a state to
9174	* receive them.
9175	*
9176	* @returns Strict VBox status code (i.e. informational status codes too).
9177	* @param pVCpu The cross context virtual CPU structure.
9178	* @param pVmxTransient The VMX-transient structure.
9179	* @param fIntrState The VT-x guest-interruptibility state.
9180	* @param fStepping Whether we are single-stepping the guest using the
9181	* hypervisor debugger and should return
9182	* VINF_EM_DBG_STEPPED if the event was dispatched
9183	* directly.
9184	*/
9185	static VBOXSTRICTRC hmR0VmxInjectPendingEvent(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t fIntrState, bool fStepping)
9186	{
9187	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9188	Assert(VMMRZCallRing3IsEnabled(pVCpu));
9189
9190	#ifdef VBOX_STRICT
9191	/*
9192	* Verify guest-interruptibility state.
9193	*
9194	* We put this in a scoped block so we do not accidentally use fBlockSti or fBlockMovSS,
9195	* since injecting an event may modify the interruptibility state and we must thus always
9196	* use fIntrState.
9197	*/
9198	{
9199	bool const fBlockMovSS = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
9200	bool const fBlockSti = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI);
9201	Assert(!fBlockSti \|\| !(ASMAtomicUoReadU64(&pVCpu->cpum.GstCtx.fExtrn) & CPUMCTX_EXTRN_RFLAGS));
9202	Assert(!fBlockSti \|\| pVCpu->cpum.GstCtx.eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
9203	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
9204	Assert(!TRPMHasTrap(pVCpu));
9205	NOREF(fBlockMovSS); NOREF(fBlockSti);
9206	}
9207	#endif
9208
9209	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
9210	if (pVCpu->hm.s.Event.fPending)
9211	{
9212	/*
9213	* Do -not- clear any interrupt-window exiting control here. We might have an interrupt
9214	* pending even while injecting an event and in this case, we want a VM-exit as soon as
9215	* the guest is ready for the next interrupt, see @bugref{6208#c45}.
9216	*
9217	* See Intel spec. 26.6.5 "Interrupt-Window Exiting and Virtual-Interrupt Delivery".
9218	*/
9219	uint32_t const uIntType = VMX_ENTRY_INT_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
9220	#ifdef VBOX_STRICT
9221	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
9222	{
9223	Assert(pVCpu->cpum.GstCtx.eflags.u32 & X86_EFL_IF);
9224	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI));
9225	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
9226	}
9227	else if (uIntType == VMX_ENTRY_INT_INFO_TYPE_NMI)
9228	{
9229	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI));
9230	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI));
9231	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
9232	}
9233	#endif
9234	Log4(("Injecting pending event vcpu[%RU32] u64IntInfo=%#RX64 Type=%#RX32\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo,
9235	uIntType));
9236
9237	/*
9238	* Inject the event and get any changes to the guest-interruptibility state.
9239	*
9240	* The guest-interruptibility state may need to be updated if we inject the event
9241	* into the guest IDT ourselves (for real-on-v86 guest injecting software interrupts).
9242	*/
9243	rcStrict = hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &pVCpu->hm.s.Event, fStepping, &fIntrState);
9244	AssertRCReturn(VBOXSTRICTRC_VAL(rcStrict), rcStrict);
9245
9246	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
9247	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt);
9248	else
9249	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt);
9250	}
9251
9252	/*
9253	* Deliver any pending debug exceptions if the guest is single-stepping using EFLAGS.TF and
9254	* is an interrupt shadow (block-by-STI or block-by-MOV SS).
9255	*/
9256	if ( (fIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
9257	&& !pVmxTransient->fIsNestedGuest)
9258	{
9259	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
9260
9261	if (!pVCpu->hm.s.fSingleInstruction)
9262	{
9263	/*
9264	* Set or clear the BS bit depending on whether the trap flag is active or not. We need
9265	* to do both since we clear the BS bit from the VMCS while exiting to ring-3.
9266	*/
9267	Assert(!DBGFIsStepping(pVCpu));
9268	uint8_t const fTrapFlag = !!(pVCpu->cpum.GstCtx.eflags.u32 & X86_EFL_TF);
9269	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, fTrapFlag << VMX_BF_VMCS_PENDING_DBG_XCPT_BS_SHIFT);
9270	AssertRC(rc);
9271	}
9272	else
9273	{
9274	/*
9275	* We must not deliver a debug exception when single-stepping over STI/Mov-SS in the
9276	* hypervisor debugger using EFLAGS.TF but rather clear interrupt inhibition. However,
9277	* we take care of this case in hmR0VmxExportSharedDebugState and also the case if
9278	* we use MTF, so just make sure it's called before executing guest-code.
9279	*/
9280	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR_MASK);
9281	}
9282	}
9283	/* else: for nested-guest currently handling while merging controls. */
9284
9285	/*
9286	* Finally, update the guest-interruptibility state.
9287	*
9288	* This is required for the real-on-v86 software interrupt injection, for
9289	* pending debug exceptions as well as updates to the guest state from ring-3 (IEM).
9290	*/
9291	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
9292	AssertRC(rc);
9293
9294	/*
9295	* There's no need to clear the VM-entry interruption-information field here if we're not
9296	* injecting anything. VT-x clears the valid bit on every VM-exit.
9297	*
9298	* See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
9299	*/
9300
9301	Assert(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping));
9302	return rcStrict;
9303	}
9304
9305
9306	/**
9307	* Enters the VT-x session.
9308	*
9309	* @returns VBox status code.
9310	* @param pVCpu The cross context virtual CPU structure.
9311	*/
9312	VMMR0DECL(int) VMXR0Enter(PVMCPUCC pVCpu)
9313	{
9314	AssertPtr(pVCpu);
9315	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fSupported);
9316	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9317
9318	LogFlowFunc(("pVCpu=%p\n", pVCpu));
9319	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
9320	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
9321
9322	#ifdef VBOX_STRICT
9323	/* At least verify VMX is enabled, since we can't check if we're in VMX root mode without #GP'ing. */
9324	RTCCUINTREG uHostCr4 = ASMGetCR4();
9325	if (!(uHostCr4 & X86_CR4_VMXE))
9326	{
9327	LogRelFunc(("X86_CR4_VMXE bit in CR4 is not set!\n"));
9328	return VERR_VMX_X86_CR4_VMXE_CLEARED;
9329	}
9330	#endif
9331
9332	/*
9333	* Do the EMT scheduled L1D and MDS flush here if needed.
9334	*/
9335	if (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_L1D_SCHED)
9336	ASMWrMsr(MSR_IA32_FLUSH_CMD, MSR_IA32_FLUSH_CMD_F_L1D);
9337	else if (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_MDS_SCHED)
9338	hmR0MdsClear();
9339
9340	/*
9341	* Load the appropriate VMCS as the current and active one.
9342	*/
9343	PVMXVMCSINFO pVmcsInfo;
9344	bool const fInNestedGuestMode = CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx);
9345	if (!fInNestedGuestMode)
9346	pVmcsInfo = &pVCpu->hmr0.s.vmx.VmcsInfo;
9347	else
9348	pVmcsInfo = &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
9349	int rc = hmR0VmxLoadVmcs(pVmcsInfo);
9350	if (RT_SUCCESS(rc))
9351	{
9352	pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs = fInNestedGuestMode;
9353	pVCpu->hm.s.vmx.fSwitchedToNstGstVmcsCopyForRing3 = fInNestedGuestMode;
9354	pVCpu->hmr0.s.fLeaveDone = false;
9355	Log4Func(("Loaded Vmcs. HostCpuId=%u\n", RTMpCpuId()));
9356	}
9357	return rc;
9358	}
9359
9360
9361	/**
9362	* The thread-context callback.
9363	*
9364	* This is used together with RTThreadCtxHookCreate() on platforms which
9365	* supports it, and directly from VMMR0EmtPrepareForBlocking() and
9366	* VMMR0EmtResumeAfterBlocking() on platforms which don't.
9367	*
9368	* @param enmEvent The thread-context event.
9369	* @param pVCpu The cross context virtual CPU structure.
9370	* @param fGlobalInit Whether global VT-x/AMD-V init. was used.
9371	* @thread EMT(pVCpu)
9372	*/
9373	VMMR0DECL(void) VMXR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPUCC pVCpu, bool fGlobalInit)
9374	{
9375	AssertPtr(pVCpu);
9376	RT_NOREF1(fGlobalInit);
9377
9378	switch (enmEvent)
9379	{
9380	case RTTHREADCTXEVENT_OUT:
9381	{
9382	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9383	VMCPU_ASSERT_EMT(pVCpu);
9384
9385	/* No longjmps (logger flushes, locks) in this fragile context. */
9386	VMMRZCallRing3Disable(pVCpu);
9387	Log4Func(("Preempting: HostCpuId=%u\n", RTMpCpuId()));
9388
9389	/* Restore host-state (FPU, debug etc.) */
9390	if (!pVCpu->hmr0.s.fLeaveDone)
9391	{
9392	/*
9393	* Do -not- import the guest-state here as we might already be in the middle of importing
9394	* it, esp. bad if we're holding the PGM lock, see comment in hmR0VmxImportGuestState().
9395	*/
9396	hmR0VmxLeave(pVCpu, false /* fImportState */);
9397	pVCpu->hmr0.s.fLeaveDone = true;
9398	}
9399
9400	/* Leave HM context, takes care of local init (term). */
9401	int rc = HMR0LeaveCpu(pVCpu);
9402	AssertRC(rc);
9403
9404	/* Restore longjmp state. */
9405	VMMRZCallRing3Enable(pVCpu);
9406	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreempt);
9407	break;
9408	}
9409
9410	case RTTHREADCTXEVENT_IN:
9411	{
9412	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9413	VMCPU_ASSERT_EMT(pVCpu);
9414
9415	/* Do the EMT scheduled L1D and MDS flush here if needed. */
9416	if (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_L1D_SCHED)
9417	ASMWrMsr(MSR_IA32_FLUSH_CMD, MSR_IA32_FLUSH_CMD_F_L1D);
9418	else if (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_MDS_SCHED)
9419	hmR0MdsClear();
9420
9421	/* No longjmps here, as we don't want to trigger preemption (& its hook) while resuming. */
9422	VMMRZCallRing3Disable(pVCpu);
9423	Log4Func(("Resumed: HostCpuId=%u\n", RTMpCpuId()));
9424
9425	/* Initialize the bare minimum state required for HM. This takes care of
9426	initializing VT-x if necessary (onlined CPUs, local init etc.) */
9427	int rc = hmR0EnterCpu(pVCpu);
9428	AssertRC(rc);
9429	Assert( (pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
9430	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
9431
9432	/* Load the active VMCS as the current one. */
9433	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
9434	rc = hmR0VmxLoadVmcs(pVmcsInfo);
9435	AssertRC(rc);
9436	Log4Func(("Resumed: Loaded Vmcs. HostCpuId=%u\n", RTMpCpuId()));
9437	pVCpu->hmr0.s.fLeaveDone = false;
9438
9439	/* Restore longjmp state. */
9440	VMMRZCallRing3Enable(pVCpu);
9441	break;
9442	}
9443
9444	default:
9445	break;
9446	}
9447	}
9448
9449
9450	/**
9451	* Exports the host state into the VMCS host-state area.
9452	* Sets up the VM-exit MSR-load area.
9453	*
9454	* The CPU state will be loaded from these fields on every successful VM-exit.
9455	*
9456	* @returns VBox status code.
9457	* @param pVCpu The cross context virtual CPU structure.
9458	*
9459	* @remarks No-long-jump zone!!!
9460	*/
9461	static int hmR0VmxExportHostState(PVMCPUCC pVCpu)
9462	{
9463	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9464
9465	int rc = VINF_SUCCESS;
9466	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT)
9467	{
9468	uint64_t uHostCr4 = hmR0VmxExportHostControlRegs();
9469
9470	rc = hmR0VmxExportHostSegmentRegs(pVCpu, uHostCr4);
9471	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9472
9473	hmR0VmxExportHostMsrs(pVCpu);
9474
9475	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_HOST_CONTEXT;
9476	}
9477	return rc;
9478	}
9479
9480
9481	/**
9482	* Saves the host state in the VMCS host-state.
9483	*
9484	* @returns VBox status code.
9485	* @param pVCpu The cross context virtual CPU structure.
9486	*
9487	* @remarks No-long-jump zone!!!
9488	*/
9489	VMMR0DECL(int) VMXR0ExportHostState(PVMCPUCC pVCpu)
9490	{
9491	AssertPtr(pVCpu);
9492	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9493
9494	/*
9495	* Export the host state here while entering HM context.
9496	* When thread-context hooks are used, we might get preempted and have to re-save the host
9497	* state but most of the time we won't be, so do it here before we disable interrupts.
9498	*/
9499	return hmR0VmxExportHostState(pVCpu);
9500	}
9501
9502
9503	/**
9504	* Exports the guest state into the VMCS guest-state area.
9505	*
9506	* The will typically be done before VM-entry when the guest-CPU state and the
9507	* VMCS state may potentially be out of sync.
9508	*
9509	* Sets up the VM-entry MSR-load and VM-exit MSR-store areas. Sets up the
9510	* VM-entry controls.
9511	* Sets up the appropriate VMX non-root function to execute guest code based on
9512	* the guest CPU mode.
9513	*
9514	* @returns VBox strict status code.
9515	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
9516	* without unrestricted guest execution and the VMMDev is not presently
9517	* mapped (e.g. EFI32).
9518	*
9519	* @param pVCpu The cross context virtual CPU structure.
9520	* @param pVmxTransient The VMX-transient structure.
9521	*
9522	* @remarks No-long-jump zone!!!
9523	*/
9524	static VBOXSTRICTRC hmR0VmxExportGuestState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9525	{
9526	AssertPtr(pVCpu);
9527	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9528	LogFlowFunc(("pVCpu=%p\n", pVCpu));
9529
9530	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExportGuestState, x);
9531
9532	/*
9533	* Determine real-on-v86 mode.
9534	* Used when the guest is in real-mode and unrestricted guest execution is not used.
9535	*/
9536	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmxTransient->pVmcsInfo->pShared;
9537	if ( pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUnrestrictedGuest
9538	\|\| !CPUMIsGuestInRealModeEx(&pVCpu->cpum.GstCtx))
9539	pVmcsInfoShared->RealMode.fRealOnV86Active = false;
9540	else
9541	{
9542	Assert(!pVmxTransient->fIsNestedGuest);
9543	pVmcsInfoShared->RealMode.fRealOnV86Active = true;
9544	}
9545
9546	/*
9547	* Any ordering dependency among the sub-functions below must be explicitly stated using comments.
9548	* Ideally, assert that the cross-dependent bits are up-to-date at the point of using it.
9549	*/
9550	int rc = hmR0VmxExportGuestEntryExitCtls(pVCpu, pVmxTransient);
9551	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9552
9553	rc = hmR0VmxExportGuestCR0(pVCpu, pVmxTransient);
9554	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9555
9556	VBOXSTRICTRC rcStrict = hmR0VmxExportGuestCR3AndCR4(pVCpu, pVmxTransient);
9557	if (rcStrict == VINF_SUCCESS)
9558	{ /* likely */ }
9559	else
9560	{
9561	Assert(rcStrict == VINF_EM_RESCHEDULE_REM \|\| RT_FAILURE_NP(rcStrict));
9562	return rcStrict;
9563	}
9564
9565	rc = hmR0VmxExportGuestSegRegsXdtr(pVCpu, pVmxTransient);
9566	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9567
9568	rc = hmR0VmxExportGuestMsrs(pVCpu, pVmxTransient);
9569	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9570
9571	hmR0VmxExportGuestApicTpr(pVCpu, pVmxTransient);
9572	hmR0VmxExportGuestXcptIntercepts(pVCpu, pVmxTransient);
9573	hmR0VmxExportGuestRip(pVCpu);
9574	hmR0VmxExportGuestRsp(pVCpu);
9575	hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9576
9577	rc = hmR0VmxExportGuestHwvirtState(pVCpu, pVmxTransient);
9578	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9579
9580	/* Clear any bits that may be set but exported unconditionally or unused/reserved bits. */
9581	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~( (HM_CHANGED_GUEST_GPRS_MASK & ~HM_CHANGED_GUEST_RSP)
9582	\| HM_CHANGED_GUEST_CR2
9583	\| (HM_CHANGED_GUEST_DR_MASK & ~HM_CHANGED_GUEST_DR7)
9584	\| HM_CHANGED_GUEST_X87
9585	\| HM_CHANGED_GUEST_SSE_AVX
9586	\| HM_CHANGED_GUEST_OTHER_XSAVE
9587	\| HM_CHANGED_GUEST_XCRx
9588	\| HM_CHANGED_GUEST_KERNEL_GS_BASE /* Part of lazy or auto load-store MSRs. */
9589	\| HM_CHANGED_GUEST_SYSCALL_MSRS /* Part of lazy or auto load-store MSRs. */
9590	\| HM_CHANGED_GUEST_TSC_AUX
9591	\| HM_CHANGED_GUEST_OTHER_MSRS
9592	\| (HM_CHANGED_KEEPER_STATE_MASK & ~HM_CHANGED_VMX_MASK)));
9593
9594	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExportGuestState, x);
9595	return rc;
9596	}
9597
9598
9599	/**
9600	* Exports the state shared between the host and guest into the VMCS.
9601	*
9602	* @param pVCpu The cross context virtual CPU structure.
9603	* @param pVmxTransient The VMX-transient structure.
9604	*
9605	* @remarks No-long-jump zone!!!
9606	*/
9607	static void hmR0VmxExportSharedState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9608	{
9609	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9610	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9611
9612	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_DR_MASK)
9613	{
9614	int rc = hmR0VmxExportSharedDebugState(pVCpu, pVmxTransient);
9615	AssertRC(rc);
9616	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_GUEST_DR_MASK;
9617
9618	/* Loading shared debug bits might have changed eflags.TF bit for debugging purposes. */
9619	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_RFLAGS)
9620	hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9621	}
9622
9623	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_GUEST_LAZY_MSRS)
9624	{
9625	hmR0VmxLazyLoadGuestMsrs(pVCpu);
9626	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_VMX_GUEST_LAZY_MSRS;
9627	}
9628
9629	AssertMsg(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE),
9630	("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
9631	}
9632
9633
9634	/**
9635	* Worker for loading the guest-state bits in the inner VT-x execution loop.
9636	*
9637	* @returns Strict VBox status code (i.e. informational status codes too).
9638	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
9639	* without unrestricted guest execution and the VMMDev is not presently
9640	* mapped (e.g. EFI32).
9641	*
9642	* @param pVCpu The cross context virtual CPU structure.
9643	* @param pVmxTransient The VMX-transient structure.
9644	*
9645	* @remarks No-long-jump zone!!!
9646	*/
9647	static VBOXSTRICTRC hmR0VmxExportGuestStateOptimal(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9648	{
9649	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9650	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9651	Assert(VMMR0IsLogFlushDisabled(pVCpu));
9652
9653	#ifdef HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
9654	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
9655	#endif
9656
9657	/*
9658	* For many VM-exits only RIP/RSP/RFLAGS (and HWVIRT state when executing a nested-guest)
9659	* changes. First try to export only these without going through all other changed-flag checks.
9660	*/
9661	VBOXSTRICTRC rcStrict;
9662	uint64_t const fCtxMask = HM_CHANGED_ALL_GUEST & ~HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE;
9663	uint64_t const fMinimalMask = HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT;
9664	uint64_t const fCtxChanged = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
9665
9666	/* If only RIP/RSP/RFLAGS/HWVIRT changed, export only those (quicker, happens more often).*/
9667	if ( (fCtxChanged & fMinimalMask)
9668	&& !(fCtxChanged & (fCtxMask & ~fMinimalMask)))
9669	{
9670	hmR0VmxExportGuestRip(pVCpu);
9671	hmR0VmxExportGuestRsp(pVCpu);
9672	hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9673	rcStrict = hmR0VmxExportGuestHwvirtState(pVCpu, pVmxTransient);
9674	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportMinimal);
9675	}
9676	/* If anything else also changed, go through the full export routine and export as required. */
9677	else if (fCtxChanged & fCtxMask)
9678	{
9679	rcStrict = hmR0VmxExportGuestState(pVCpu, pVmxTransient);
9680	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
9681	{ /* likely */}
9682	else
9683	{
9684	AssertMsg(rcStrict == VINF_EM_RESCHEDULE_REM, ("Failed to export guest state! rc=%Rrc\n",
9685	VBOXSTRICTRC_VAL(rcStrict)));
9686	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9687	return rcStrict;
9688	}
9689	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportFull);
9690	}
9691	/* Nothing changed, nothing to load here. */
9692	else
9693	rcStrict = VINF_SUCCESS;
9694
9695	#ifdef VBOX_STRICT
9696	/* All the guest state bits should be loaded except maybe the host context and/or the shared host/guest bits. */
9697	uint64_t const fCtxChangedCur = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
9698	AssertMsg(!(fCtxChangedCur & fCtxMask), ("fCtxChangedCur=%#RX64\n", fCtxChangedCur));
9699	#endif
9700	return rcStrict;
9701	}
9702
9703
9704	/**
9705	* Tries to determine what part of the guest-state VT-x has deemed as invalid
9706	* and update error record fields accordingly.
9707	*
9708	* @returns VMX_IGS_* error codes.
9709	* @retval VMX_IGS_REASON_NOT_FOUND if this function could not find anything
9710	* wrong with the guest state.
9711	*
9712	* @param pVCpu The cross context virtual CPU structure.
9713	* @param pVmcsInfo The VMCS info. object.
9714	*
9715	* @remarks This function assumes our cache of the VMCS controls
9716	* are valid, i.e. hmR0VmxCheckCachedVmcsCtls() succeeded.
9717	*/
9718	static uint32_t hmR0VmxCheckGuestState(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
9719	{
9720	#define HMVMX_ERROR_BREAK(err) { uError = (err); break; }
9721	#define HMVMX_CHECK_BREAK(expr, err) do { \
9722	if (!(expr)) { uError = (err); break; } \
9723	} while (0)
9724
9725	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
9726	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
9727	uint32_t uError = VMX_IGS_ERROR;
9728	uint32_t u32IntrState = 0;
9729	bool const fUnrestrictedGuest = pVM->hmr0.s.vmx.fUnrestrictedGuest;
9730	do
9731	{
9732	int rc;
9733
9734	/*
9735	* Guest-interruptibility state.
9736	*
9737	* Read this first so that any check that fails prior to those that actually
9738	* require the guest-interruptibility state would still reflect the correct
9739	* VMCS value and avoids causing further confusion.
9740	*/
9741	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &u32IntrState);
9742	AssertRC(rc);
9743
9744	uint32_t u32Val;
9745	uint64_t u64Val;
9746
9747	/*
9748	* CR0.
9749	*/
9750	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
9751	uint64_t fSetCr0 = (g_HmMsrs.u.vmx.u64Cr0Fixed0 & g_HmMsrs.u.vmx.u64Cr0Fixed1);
9752	uint64_t const fZapCr0 = (g_HmMsrs.u.vmx.u64Cr0Fixed0 \| g_HmMsrs.u.vmx.u64Cr0Fixed1);
9753	/* Exceptions for unrestricted guest execution for CR0 fixed bits (PE, PG).
9754	See Intel spec. 26.3.1 "Checks on Guest Control Registers, Debug Registers and MSRs." */
9755	if (fUnrestrictedGuest)
9756	fSetCr0 &= ~(uint64_t)(X86_CR0_PE \| X86_CR0_PG);
9757
9758	uint64_t u64GuestCr0;
9759	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64GuestCr0);
9760	AssertRC(rc);
9761	HMVMX_CHECK_BREAK((u64GuestCr0 & fSetCr0) == fSetCr0, VMX_IGS_CR0_FIXED1);
9762	HMVMX_CHECK_BREAK(!(u64GuestCr0 & ~fZapCr0), VMX_IGS_CR0_FIXED0);
9763	if ( !fUnrestrictedGuest
9764	&& (u64GuestCr0 & X86_CR0_PG)
9765	&& !(u64GuestCr0 & X86_CR0_PE))
9766	HMVMX_ERROR_BREAK(VMX_IGS_CR0_PG_PE_COMBO);
9767
9768	/*
9769	* CR4.
9770	*/
9771	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
9772	uint64_t const fSetCr4 = (g_HmMsrs.u.vmx.u64Cr4Fixed0 & g_HmMsrs.u.vmx.u64Cr4Fixed1);
9773	uint64_t const fZapCr4 = (g_HmMsrs.u.vmx.u64Cr4Fixed0 \| g_HmMsrs.u.vmx.u64Cr4Fixed1);
9774
9775	uint64_t u64GuestCr4;
9776	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR4, &u64GuestCr4);
9777	AssertRC(rc);
9778	HMVMX_CHECK_BREAK((u64GuestCr4 & fSetCr4) == fSetCr4, VMX_IGS_CR4_FIXED1);
9779	HMVMX_CHECK_BREAK(!(u64GuestCr4 & ~fZapCr4), VMX_IGS_CR4_FIXED0);
9780
9781	/*
9782	* IA32_DEBUGCTL MSR.
9783	*/
9784	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, &u64Val);
9785	AssertRC(rc);
9786	if ( (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
9787	&& (u64Val & 0xfffffe3c)) /* Bits 31:9, bits 5:2 MBZ. */
9788	{
9789	HMVMX_ERROR_BREAK(VMX_IGS_DEBUGCTL_MSR_RESERVED);
9790	}
9791	uint64_t u64DebugCtlMsr = u64Val;
9792
9793	#ifdef VBOX_STRICT
9794	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
9795	AssertRC(rc);
9796	Assert(u32Val == pVmcsInfo->u32EntryCtls);
9797	#endif
9798	bool const fLongModeGuest = RT_BOOL(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
9799
9800	/*
9801	* RIP and RFLAGS.
9802	*/
9803	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RIP, &u64Val);
9804	AssertRC(rc);
9805	/* pCtx->rip can be different than the one in the VMCS (e.g. run guest code and VM-exits that don't update it). */
9806	if ( !fLongModeGuest
9807	\|\| !pCtx->cs.Attr.n.u1Long)
9808	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffff00000000)), VMX_IGS_LONGMODE_RIP_INVALID);
9809	/** @todo If the processor supports N < 64 linear-address bits, bits 63:N
9810	* must be identical if the "IA-32e mode guest" VM-entry
9811	* control is 1 and CS.L is 1. No check applies if the
9812	* CPU supports 64 linear-address bits. */
9813
9814	/* Flags in pCtx can be different (real-on-v86 for instance). We are only concerned about the VMCS contents here. */
9815	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RFLAGS, &u64Val);
9816	AssertRC(rc);
9817	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffc08028)), /* Bit 63:22, Bit 15, 5, 3 MBZ. */
9818	VMX_IGS_RFLAGS_RESERVED);
9819	HMVMX_CHECK_BREAK((u64Val & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
9820	uint32_t const u32Eflags = u64Val;
9821
9822	if ( fLongModeGuest
9823	\|\| ( fUnrestrictedGuest
9824	&& !(u64GuestCr0 & X86_CR0_PE)))
9825	{
9826	HMVMX_CHECK_BREAK(!(u32Eflags & X86_EFL_VM), VMX_IGS_RFLAGS_VM_INVALID);
9827	}
9828
9829	uint32_t u32EntryInfo;
9830	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo);
9831	AssertRC(rc);
9832	if (VMX_ENTRY_INT_INFO_IS_EXT_INT(u32EntryInfo))
9833	HMVMX_CHECK_BREAK(u32Eflags & X86_EFL_IF, VMX_IGS_RFLAGS_IF_INVALID);
9834
9835	/*
9836	* 64-bit checks.
9837	*/
9838	if (fLongModeGuest)
9839	{
9840	HMVMX_CHECK_BREAK(u64GuestCr0 & X86_CR0_PG, VMX_IGS_CR0_PG_LONGMODE);
9841	HMVMX_CHECK_BREAK(u64GuestCr4 & X86_CR4_PAE, VMX_IGS_CR4_PAE_LONGMODE);
9842	}
9843
9844	if ( !fLongModeGuest
9845	&& (u64GuestCr4 & X86_CR4_PCIDE))
9846	HMVMX_ERROR_BREAK(VMX_IGS_CR4_PCIDE);
9847
9848	/** @todo CR3 field must be such that bits 63:52 and bits in the range
9849	* 51:32 beyond the processor's physical-address width are 0. */
9850
9851	if ( (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
9852	&& (pCtx->dr[7] & X86_DR7_MBZ_MASK))
9853	HMVMX_ERROR_BREAK(VMX_IGS_DR7_RESERVED);
9854
9855	rc = VMXReadVmcsNw(VMX_VMCS_HOST_SYSENTER_ESP, &u64Val);
9856	AssertRC(rc);
9857	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_ESP_NOT_CANONICAL);
9858
9859	rc = VMXReadVmcsNw(VMX_VMCS_HOST_SYSENTER_EIP, &u64Val);
9860	AssertRC(rc);
9861	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_EIP_NOT_CANONICAL);
9862
9863	/*
9864	* PERF_GLOBAL MSR.
9865	*/
9866	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PERF_MSR)
9867	{
9868	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL, &u64Val);
9869	AssertRC(rc);
9870	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffff8fffffffc)),
9871	VMX_IGS_PERF_GLOBAL_MSR_RESERVED); /* Bits 63:35, bits 31:2 MBZ. */
9872	}
9873
9874	/*
9875	* PAT MSR.
9876	*/
9877	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PAT_MSR)
9878	{
9879	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PAT_FULL, &u64Val);
9880	AssertRC(rc);
9881	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0x707070707070707)), VMX_IGS_PAT_MSR_RESERVED);
9882	for (unsigned i = 0; i < 8; i++)
9883	{
9884	uint8_t u8Val = (u64Val & 0xff);
9885	if ( u8Val != 0 /* UC */
9886	&& u8Val != 1 /* WC */
9887	&& u8Val != 4 /* WT */
9888	&& u8Val != 5 /* WP */
9889	&& u8Val != 6 /* WB */
9890	&& u8Val != 7 /* UC- */)
9891	HMVMX_ERROR_BREAK(VMX_IGS_PAT_MSR_INVALID);
9892	u64Val >>= 8;
9893	}
9894	}
9895
9896	/*
9897	* EFER MSR.
9898	*/
9899	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_EFER_MSR)
9900	{
9901	Assert(g_fHmVmxSupportsVmcsEfer);
9902	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_EFER_FULL, &u64Val);
9903	AssertRC(rc);
9904	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffffffffff2fe)),
9905	VMX_IGS_EFER_MSR_RESERVED); /* Bits 63:12, bit 9, bits 7:1 MBZ. */
9906	HMVMX_CHECK_BREAK(RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL( pVmcsInfo->u32EntryCtls
9907	& VMX_ENTRY_CTLS_IA32E_MODE_GUEST),
9908	VMX_IGS_EFER_LMA_GUEST_MODE_MISMATCH);
9909	/** @todo r=ramshankar: Unrestricted check here is probably wrong, see
9910	* iemVmxVmentryCheckGuestState(). */
9911	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9912	\|\| !(u64GuestCr0 & X86_CR0_PG)
9913	\|\| RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL(u64Val & MSR_K6_EFER_LME),
9914	VMX_IGS_EFER_LMA_LME_MISMATCH);
9915	}
9916
9917	/*
9918	* Segment registers.
9919	*/
9920	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9921	\|\| !(pCtx->ldtr.Sel & X86_SEL_LDT), VMX_IGS_LDTR_TI_INVALID);
9922	if (!(u32Eflags & X86_EFL_VM))
9923	{
9924	/* CS */
9925	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1Present, VMX_IGS_CS_ATTR_P_INVALID);
9926	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xf00), VMX_IGS_CS_ATTR_RESERVED);
9927	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xfffe0000), VMX_IGS_CS_ATTR_RESERVED);
9928	HMVMX_CHECK_BREAK( (pCtx->cs.u32Limit & 0xfff) == 0xfff
9929	\|\| !(pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9930	HMVMX_CHECK_BREAK( !(pCtx->cs.u32Limit & 0xfff00000)
9931	\|\| (pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9932	/* CS cannot be loaded with NULL in protected mode. */
9933	HMVMX_CHECK_BREAK(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_CS_ATTR_UNUSABLE);
9934	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1DescType, VMX_IGS_CS_ATTR_S_INVALID);
9935	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
9936	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_UNEQUAL);
9937	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
9938	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_MISMATCH);
9939	else if (fUnrestrictedGuest && pCtx->cs.Attr.n.u4Type == 3)
9940	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == 0, VMX_IGS_CS_ATTR_DPL_INVALID);
9941	else
9942	HMVMX_ERROR_BREAK(VMX_IGS_CS_ATTR_TYPE_INVALID);
9943
9944	/* SS */
9945	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9946	\|\| (pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL), VMX_IGS_SS_CS_RPL_UNEQUAL);
9947	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL), VMX_IGS_SS_ATTR_DPL_RPL_UNEQUAL);
9948	if ( !(pCtx->cr0 & X86_CR0_PE)
9949	\|\| pCtx->cs.Attr.n.u4Type == 3)
9950	HMVMX_CHECK_BREAK(!pCtx->ss.Attr.n.u2Dpl, VMX_IGS_SS_ATTR_DPL_INVALID);
9951
9952	if (!(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
9953	{
9954	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7, VMX_IGS_SS_ATTR_TYPE_INVALID);
9955	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u1Present, VMX_IGS_SS_ATTR_P_INVALID);
9956	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xf00), VMX_IGS_SS_ATTR_RESERVED);
9957	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xfffe0000), VMX_IGS_SS_ATTR_RESERVED);
9958	HMVMX_CHECK_BREAK( (pCtx->ss.u32Limit & 0xfff) == 0xfff
9959	\|\| !(pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9960	HMVMX_CHECK_BREAK( !(pCtx->ss.u32Limit & 0xfff00000)
9961	\|\| (pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9962	}
9963
9964	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxExportGuestSReg(). */
9965	if (!(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
9966	{
9967	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_DS_ATTR_A_INVALID);
9968	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u1Present, VMX_IGS_DS_ATTR_P_INVALID);
9969	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9970	\|\| pCtx->ds.Attr.n.u4Type > 11
9971	\|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9972	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xf00), VMX_IGS_DS_ATTR_RESERVED);
9973	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xfffe0000), VMX_IGS_DS_ATTR_RESERVED);
9974	HMVMX_CHECK_BREAK( (pCtx->ds.u32Limit & 0xfff) == 0xfff
9975	\|\| !(pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9976	HMVMX_CHECK_BREAK( !(pCtx->ds.u32Limit & 0xfff00000)
9977	\|\| (pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9978	HMVMX_CHECK_BREAK( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9979	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_DS_ATTR_TYPE_INVALID);
9980	}
9981	if (!(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
9982	{
9983	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_ES_ATTR_A_INVALID);
9984	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u1Present, VMX_IGS_ES_ATTR_P_INVALID);
9985	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9986	\|\| pCtx->es.Attr.n.u4Type > 11
9987	\|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9988	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xf00), VMX_IGS_ES_ATTR_RESERVED);
9989	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xfffe0000), VMX_IGS_ES_ATTR_RESERVED);
9990	HMVMX_CHECK_BREAK( (pCtx->es.u32Limit & 0xfff) == 0xfff
9991	\|\| !(pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
9992	HMVMX_CHECK_BREAK( !(pCtx->es.u32Limit & 0xfff00000)
9993	\|\| (pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
9994	HMVMX_CHECK_BREAK( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9995	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_ES_ATTR_TYPE_INVALID);
9996	}
9997	if (!(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
9998	{
9999	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_FS_ATTR_A_INVALID);
10000	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u1Present, VMX_IGS_FS_ATTR_P_INVALID);
10001	HMVMX_CHECK_BREAK( fUnrestrictedGuest
10002	\|\| pCtx->fs.Attr.n.u4Type > 11
10003	\|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL), VMX_IGS_FS_ATTR_DPL_RPL_UNEQUAL);
10004	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xf00), VMX_IGS_FS_ATTR_RESERVED);
10005	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xfffe0000), VMX_IGS_FS_ATTR_RESERVED);
10006	HMVMX_CHECK_BREAK( (pCtx->fs.u32Limit & 0xfff) == 0xfff
10007	\|\| !(pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
10008	HMVMX_CHECK_BREAK( !(pCtx->fs.u32Limit & 0xfff00000)
10009	\|\| (pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
10010	HMVMX_CHECK_BREAK( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10011	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_FS_ATTR_TYPE_INVALID);
10012	}
10013	if (!(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
10014	{
10015	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_GS_ATTR_A_INVALID);
10016	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u1Present, VMX_IGS_GS_ATTR_P_INVALID);
10017	HMVMX_CHECK_BREAK( fUnrestrictedGuest
10018	\|\| pCtx->gs.Attr.n.u4Type > 11
10019	\|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL), VMX_IGS_GS_ATTR_DPL_RPL_UNEQUAL);
10020	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xf00), VMX_IGS_GS_ATTR_RESERVED);
10021	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xfffe0000), VMX_IGS_GS_ATTR_RESERVED);
10022	HMVMX_CHECK_BREAK( (pCtx->gs.u32Limit & 0xfff) == 0xfff
10023	\|\| !(pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
10024	HMVMX_CHECK_BREAK( !(pCtx->gs.u32Limit & 0xfff00000)
10025	\|\| (pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
10026	HMVMX_CHECK_BREAK( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10027	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_GS_ATTR_TYPE_INVALID);
10028	}
10029	/* 64-bit capable CPUs. */
10030	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
10031	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
10032	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
10033	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
10034	HMVMX_CHECK_BREAK(!RT_HI_U32(pCtx->cs.u64Base), VMX_IGS_LONGMODE_CS_BASE_INVALID);
10035	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ss.u64Base),
10036	VMX_IGS_LONGMODE_SS_BASE_INVALID);
10037	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ds.u64Base),
10038	VMX_IGS_LONGMODE_DS_BASE_INVALID);
10039	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->es.u64Base),
10040	VMX_IGS_LONGMODE_ES_BASE_INVALID);
10041	}
10042	else
10043	{
10044	/* V86 mode checks. */
10045	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
10046	if (pVmcsInfo->pShared->RealMode.fRealOnV86Active)
10047	{
10048	u32CSAttr = 0xf3; u32SSAttr = 0xf3;
10049	u32DSAttr = 0xf3; u32ESAttr = 0xf3;
10050	u32FSAttr = 0xf3; u32GSAttr = 0xf3;
10051	}
10052	else
10053	{
10054	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u;
10055	u32DSAttr = pCtx->ds.Attr.u; u32ESAttr = pCtx->es.Attr.u;
10056	u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
10057	}
10058
10059	/* CS */
10060	HMVMX_CHECK_BREAK((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), VMX_IGS_V86_CS_BASE_INVALID);
10061	HMVMX_CHECK_BREAK(pCtx->cs.u32Limit == 0xffff, VMX_IGS_V86_CS_LIMIT_INVALID);
10062	HMVMX_CHECK_BREAK(u32CSAttr == 0xf3, VMX_IGS_V86_CS_ATTR_INVALID);
10063	/* SS */
10064	HMVMX_CHECK_BREAK((pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4), VMX_IGS_V86_SS_BASE_INVALID);
10065	HMVMX_CHECK_BREAK(pCtx->ss.u32Limit == 0xffff, VMX_IGS_V86_SS_LIMIT_INVALID);
10066	HMVMX_CHECK_BREAK(u32SSAttr == 0xf3, VMX_IGS_V86_SS_ATTR_INVALID);
10067	/* DS */
10068	HMVMX_CHECK_BREAK((pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4), VMX_IGS_V86_DS_BASE_INVALID);
10069	HMVMX_CHECK_BREAK(pCtx->ds.u32Limit == 0xffff, VMX_IGS_V86_DS_LIMIT_INVALID);
10070	HMVMX_CHECK_BREAK(u32DSAttr == 0xf3, VMX_IGS_V86_DS_ATTR_INVALID);
10071	/* ES */
10072	HMVMX_CHECK_BREAK((pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4), VMX_IGS_V86_ES_BASE_INVALID);
10073	HMVMX_CHECK_BREAK(pCtx->es.u32Limit == 0xffff, VMX_IGS_V86_ES_LIMIT_INVALID);
10074	HMVMX_CHECK_BREAK(u32ESAttr == 0xf3, VMX_IGS_V86_ES_ATTR_INVALID);
10075	/* FS */
10076	HMVMX_CHECK_BREAK((pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4), VMX_IGS_V86_FS_BASE_INVALID);
10077	HMVMX_CHECK_BREAK(pCtx->fs.u32Limit == 0xffff, VMX_IGS_V86_FS_LIMIT_INVALID);
10078	HMVMX_CHECK_BREAK(u32FSAttr == 0xf3, VMX_IGS_V86_FS_ATTR_INVALID);
10079	/* GS */
10080	HMVMX_CHECK_BREAK((pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4), VMX_IGS_V86_GS_BASE_INVALID);
10081	HMVMX_CHECK_BREAK(pCtx->gs.u32Limit == 0xffff, VMX_IGS_V86_GS_LIMIT_INVALID);
10082	HMVMX_CHECK_BREAK(u32GSAttr == 0xf3, VMX_IGS_V86_GS_ATTR_INVALID);
10083	/* 64-bit capable CPUs. */
10084	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
10085	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
10086	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
10087	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
10088	HMVMX_CHECK_BREAK(!RT_HI_U32(pCtx->cs.u64Base), VMX_IGS_LONGMODE_CS_BASE_INVALID);
10089	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ss.u64Base),
10090	VMX_IGS_LONGMODE_SS_BASE_INVALID);
10091	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ds.u64Base),
10092	VMX_IGS_LONGMODE_DS_BASE_INVALID);
10093	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->es.u64Base),
10094	VMX_IGS_LONGMODE_ES_BASE_INVALID);
10095	}
10096
10097	/*
10098	* TR.
10099	*/
10100	HMVMX_CHECK_BREAK(!(pCtx->tr.Sel & X86_SEL_LDT), VMX_IGS_TR_TI_INVALID);
10101	/* 64-bit capable CPUs. */
10102	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->tr.u64Base), VMX_IGS_TR_BASE_NOT_CANONICAL);
10103	if (fLongModeGuest)
10104	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u4Type == 11, /* 64-bit busy TSS. */
10105	VMX_IGS_LONGMODE_TR_ATTR_TYPE_INVALID);
10106	else
10107	HMVMX_CHECK_BREAK( pCtx->tr.Attr.n.u4Type == 3 /* 16-bit busy TSS. */
10108	\|\| pCtx->tr.Attr.n.u4Type == 11, /* 32-bit busy TSS.*/
10109	VMX_IGS_TR_ATTR_TYPE_INVALID);
10110	HMVMX_CHECK_BREAK(!pCtx->tr.Attr.n.u1DescType, VMX_IGS_TR_ATTR_S_INVALID);
10111	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u1Present, VMX_IGS_TR_ATTR_P_INVALID);
10112	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & 0xf00), VMX_IGS_TR_ATTR_RESERVED); /* Bits 11:8 MBZ. */
10113	HMVMX_CHECK_BREAK( (pCtx->tr.u32Limit & 0xfff) == 0xfff
10114	\|\| !(pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
10115	HMVMX_CHECK_BREAK( !(pCtx->tr.u32Limit & 0xfff00000)
10116	\|\| (pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
10117	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_TR_ATTR_UNUSABLE);
10118
10119	/*
10120	* GDTR and IDTR (64-bit capable checks).
10121	*/
10122	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
10123	AssertRC(rc);
10124	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_GDTR_BASE_NOT_CANONICAL);
10125
10126	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
10127	AssertRC(rc);
10128	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_IDTR_BASE_NOT_CANONICAL);
10129
10130	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val);
10131	AssertRC(rc);
10132	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_GDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
10133
10134	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val);
10135	AssertRC(rc);
10136	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_IDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
10137
10138	/*
10139	* Guest Non-Register State.
10140	*/
10141	/* Activity State. */
10142	uint32_t u32ActivityState;
10143	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, &u32ActivityState);
10144	AssertRC(rc);
10145	HMVMX_CHECK_BREAK( !u32ActivityState
10146	\|\| (u32ActivityState & RT_BF_GET(g_HmMsrs.u.vmx.u64Misc, VMX_BF_MISC_ACTIVITY_STATES)),
10147	VMX_IGS_ACTIVITY_STATE_INVALID);
10148	HMVMX_CHECK_BREAK( !(pCtx->ss.Attr.n.u2Dpl)
10149	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT, VMX_IGS_ACTIVITY_STATE_HLT_INVALID);
10150
10151	if ( u32IntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS
10152	\|\| u32IntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10153	HMVMX_CHECK_BREAK(u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE, VMX_IGS_ACTIVITY_STATE_ACTIVE_INVALID);
10154
10155	/** @todo Activity state and injecting interrupts. Left as a todo since we
10156	* currently don't use activity states but ACTIVE. */
10157
10158	HMVMX_CHECK_BREAK( !(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)
10159	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT, VMX_IGS_ACTIVITY_STATE_SIPI_WAIT_INVALID);
10160
10161	/* Guest interruptibility-state. */
10162	HMVMX_CHECK_BREAK(!(u32IntrState & 0xffffffe0), VMX_IGS_INTERRUPTIBILITY_STATE_RESERVED);
10163	HMVMX_CHECK_BREAK((u32IntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
10164	!= (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
10165	VMX_IGS_INTERRUPTIBILITY_STATE_STI_MOVSS_INVALID);
10166	HMVMX_CHECK_BREAK( (u32Eflags & X86_EFL_IF)
10167	\|\| !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI),
10168	VMX_IGS_INTERRUPTIBILITY_STATE_STI_EFL_INVALID);
10169	if (VMX_ENTRY_INT_INFO_IS_EXT_INT(u32EntryInfo))
10170	{
10171	HMVMX_CHECK_BREAK( !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10172	&& !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
10173	VMX_IGS_INTERRUPTIBILITY_STATE_EXT_INT_INVALID);
10174	}
10175	else if (VMX_ENTRY_INT_INFO_IS_XCPT_NMI(u32EntryInfo))
10176	{
10177	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
10178	VMX_IGS_INTERRUPTIBILITY_STATE_MOVSS_INVALID);
10179	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI),
10180	VMX_IGS_INTERRUPTIBILITY_STATE_STI_INVALID);
10181	}
10182	/** @todo Assumes the processor is not in SMM. */
10183	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI),
10184	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_INVALID);
10185	HMVMX_CHECK_BREAK( !(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)
10186	\|\| (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI),
10187	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_SMM_INVALID);
10188	if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
10189	&& VMX_ENTRY_INT_INFO_IS_XCPT_NMI(u32EntryInfo))
10190	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI), VMX_IGS_INTERRUPTIBILITY_STATE_NMI_INVALID);
10191
10192	/* Pending debug exceptions. */
10193	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, &u64Val);
10194	AssertRC(rc);
10195	/* Bits 63:15, Bit 13, Bits 11:4 MBZ. */
10196	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffffaff0)), VMX_IGS_LONGMODE_PENDING_DEBUG_RESERVED);
10197	u32Val = u64Val; /* For pending debug exceptions checks below. */
10198
10199	if ( (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10200	\|\| (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS)
10201	\|\| u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT)
10202	{
10203	if ( (u32Eflags & X86_EFL_TF)
10204	&& !(u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
10205	{
10206	/* Bit 14 is PendingDebug.BS. */
10207	HMVMX_CHECK_BREAK(u32Val & RT_BIT(14), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_SET);
10208	}
10209	if ( !(u32Eflags & X86_EFL_TF)
10210	\|\| (u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
10211	{
10212	/* Bit 14 is PendingDebug.BS. */
10213	HMVMX_CHECK_BREAK(!(u32Val & RT_BIT(14)), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_CLEAR);
10214	}
10215	}
10216
10217	/* VMCS link pointer. */
10218	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, &u64Val);
10219	AssertRC(rc);
10220	if (u64Val != UINT64_C(0xffffffffffffffff))
10221	{
10222	HMVMX_CHECK_BREAK(!(u64Val & 0xfff), VMX_IGS_VMCS_LINK_PTR_RESERVED);
10223	/** @todo Bits beyond the processor's physical-address width MBZ. */
10224	/** @todo SMM checks. */
10225	Assert(pVmcsInfo->HCPhysShadowVmcs == u64Val);
10226	Assert(pVmcsInfo->pvShadowVmcs);
10227	VMXVMCSREVID VmcsRevId;
10228	VmcsRevId.u = (uint32_t )pVmcsInfo->pvShadowVmcs;
10229	HMVMX_CHECK_BREAK(VmcsRevId.n.u31RevisionId == RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_ID),
10230	VMX_IGS_VMCS_LINK_PTR_SHADOW_VMCS_ID_INVALID);
10231	HMVMX_CHECK_BREAK(VmcsRevId.n.fIsShadowVmcs == (uint32_t)!!(pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING),
10232	VMX_IGS_VMCS_LINK_PTR_NOT_SHADOW);
10233	}
10234
10235	/** @todo Checks on Guest Page-Directory-Pointer-Table Entries when guest is
10236	* not using nested paging? */
10237	if ( pVM->hmr0.s.fNestedPaging
10238	&& !fLongModeGuest
10239	&& CPUMIsGuestInPAEModeEx(pCtx))
10240	{
10241	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &u64Val);
10242	AssertRC(rc);
10243	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10244
10245	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &u64Val);
10246	AssertRC(rc);
10247	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10248
10249	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &u64Val);
10250	AssertRC(rc);
10251	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10252
10253	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &u64Val);
10254	AssertRC(rc);
10255	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10256	}
10257
10258	/* Shouldn't happen but distinguish it from AssertRCBreak() errors. */
10259	if (uError == VMX_IGS_ERROR)
10260	uError = VMX_IGS_REASON_NOT_FOUND;
10261	} while (0);
10262
10263	pVCpu->hm.s.u32HMError = uError;
10264	pVCpu->hm.s.vmx.LastError.u32GuestIntrState = u32IntrState;
10265	return uError;
10266
10267	#undef HMVMX_ERROR_BREAK
10268	#undef HMVMX_CHECK_BREAK
10269	}
10270
10271
10272	/**
10273	* Map the APIC-access page for virtualizing APIC accesses.
10274	*
10275	* This can cause a longjumps to R3 due to the acquisition of the PGM lock. Hence,
10276	* this not done as part of exporting guest state, see @bugref{8721}.
10277	*
10278	* @returns VBox status code.
10279	* @param pVCpu The cross context virtual CPU structure.
10280	*/
10281	static int hmR0VmxMapHCApicAccessPage(PVMCPUCC pVCpu)
10282	{
10283	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10284	uint64_t const u64MsrApicBase = APICGetBaseMsrNoCheck(pVCpu);
10285
10286	Assert(PDMHasApic(pVM));
10287	Assert(u64MsrApicBase);
10288
10289	RTGCPHYS const GCPhysApicBase = u64MsrApicBase & PAGE_BASE_GC_MASK;
10290	Log4Func(("Mappping HC APIC-access page at %#RGp\n", GCPhysApicBase));
10291
10292	/* Unalias the existing mapping. */
10293	int rc = PGMHandlerPhysicalReset(pVM, GCPhysApicBase);
10294	AssertRCReturn(rc, rc);
10295
10296	/* Map the HC APIC-access page in place of the MMIO page, also updates the shadow page tables if necessary. */
10297	Assert(pVM->hmr0.s.vmx.HCPhysApicAccess != NIL_RTHCPHYS);
10298	rc = IOMR0MmioMapMmioHCPage(pVM, pVCpu, GCPhysApicBase, pVM->hmr0.s.vmx.HCPhysApicAccess, X86_PTE_RW \| X86_PTE_P);
10299	AssertRCReturn(rc, rc);
10300
10301	/* Update the per-VCPU cache of the APIC base MSR. */
10302	pVCpu->hm.s.vmx.u64GstMsrApicBase = u64MsrApicBase;
10303	return VINF_SUCCESS;
10304	}
10305
10306
10307	/**
10308	* Worker function passed to RTMpOnSpecific() that is to be called on the target
10309	* CPU.
10310	*
10311	* @param idCpu The ID for the CPU the function is called on.
10312	* @param pvUser1 Null, not used.
10313	* @param pvUser2 Null, not used.
10314	*/
10315	static DECLCALLBACK(void) hmR0DispatchHostNmi(RTCPUID idCpu, void pvUser1, void pvUser2)
10316	{
10317	RT_NOREF3(idCpu, pvUser1, pvUser2);
10318	VMXDispatchHostNmi();
10319	}
10320
10321
10322	/**
10323	* Dispatching an NMI on the host CPU that received it.
10324	*
10325	* @returns VBox status code.
10326	* @param pVCpu The cross context virtual CPU structure.
10327	* @param pVmcsInfo The VMCS info. object corresponding to the VMCS that was
10328	* executing when receiving the host NMI in VMX non-root
10329	* operation.
10330	*/
10331	static int hmR0VmxExitHostNmi(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
10332	{
10333	RTCPUID const idCpu = pVmcsInfo->idHostCpuExec;
10334	Assert(idCpu != NIL_RTCPUID);
10335
10336	/*
10337	* We don't want to delay dispatching the NMI any more than we have to. However,
10338	* we have already chosen -not- to dispatch NMIs when interrupts were still disabled
10339	* after executing guest or nested-guest code for the following reasons:
10340	*
10341	* - We would need to perform VMREADs with interrupts disabled and is orders of
10342	* magnitude worse when we run as a nested hypervisor without VMCS shadowing
10343	* supported by the host hypervisor.
10344	*
10345	* - It affects the common VM-exit scenario and keeps interrupts disabled for a
10346	* longer period of time just for handling an edge case like host NMIs which do
10347	* not occur nearly as frequently as other VM-exits.
10348	*
10349	* Let's cover the most likely scenario first. Check if we are on the target CPU
10350	* and dispatch the NMI right away. This should be much faster than calling into
10351	* RTMpOnSpecific() machinery.
10352	*/
10353	bool fDispatched = false;
10354	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
10355	if (idCpu == RTMpCpuId())
10356	{
10357	VMXDispatchHostNmi();
10358	fDispatched = true;
10359	}
10360	ASMSetFlags(fEFlags);
10361	if (fDispatched)
10362	{
10363	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC);
10364	return VINF_SUCCESS;
10365	}
10366
10367	/*
10368	* RTMpOnSpecific() waits until the worker function has run on the target CPU. So
10369	* there should be no race or recursion even if we are unlucky enough to be preempted
10370	* (to the target CPU) without dispatching the host NMI above.
10371	*/
10372	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGCIpi);
10373	return RTMpOnSpecific(idCpu, &hmR0DispatchHostNmi, NULL /* pvUser1 /, NULL / pvUser2 */);
10374	}
10375
10376
10377	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10378	/**
10379	* Merges the guest with the nested-guest MSR bitmap in preparation of executing the
10380	* nested-guest using hardware-assisted VMX.
10381	*
10382	* @param pVCpu The cross context virtual CPU structure.
10383	* @param pVmcsInfoNstGst The nested-guest VMCS info. object.
10384	* @param pVmcsInfoGst The guest VMCS info. object.
10385	*/
10386	static void hmR0VmxMergeMsrBitmapNested(PCVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfoNstGst, PCVMXVMCSINFO pVmcsInfoGst)
10387	{
10388	uint32_t const cbMsrBitmap = X86_PAGE_4K_SIZE;
10389	uint64_t pu64MsrBitmap = (uint64_t )pVmcsInfoNstGst->pvMsrBitmap;
10390	Assert(pu64MsrBitmap);
10391
10392	/*
10393	* We merge the guest MSR bitmap with the nested-guest MSR bitmap such that any
10394	* MSR that is intercepted by the guest is also intercepted while executing the
10395	* nested-guest using hardware-assisted VMX.
10396	*
10397	* Note! If the nested-guest is not using an MSR bitmap, every MSR must cause a
10398	* nested-guest VM-exit even if the outer guest is not intercepting some
10399	* MSRs. We cannot assume the caller has initialized the nested-guest
10400	* MSR bitmap in this case.
10401	*
10402	* The nested hypervisor may also switch whether it uses MSR bitmaps for
10403	* each of its VM-entry, hence initializing it once per-VM while setting
10404	* up the nested-guest VMCS is not sufficient.
10405	*/
10406	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
10407	if (pVmcsNstGst->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
10408	{
10409	uint64_t const pu64MsrBitmapNstGst = (uint64_t const )pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap);
10410	uint64_t const pu64MsrBitmapGst = (uint64_t const )pVmcsInfoGst->pvMsrBitmap;
10411	Assert(pu64MsrBitmapNstGst);
10412	Assert(pu64MsrBitmapGst);
10413
10414	/** @todo Detect and use EVEX.POR? */
10415	uint32_t const cFrags = cbMsrBitmap / sizeof(uint64_t);
10416	for (uint32_t i = 0; i < cFrags; i++)
10417	pu64MsrBitmap[i] = pu64MsrBitmapNstGst[i] \| pu64MsrBitmapGst[i];
10418	}
10419	else
10420	ASMMemFill32(pu64MsrBitmap, cbMsrBitmap, UINT32_C(0xffffffff));
10421	}
10422
10423
10424	/**
10425	* Merges the guest VMCS in to the nested-guest VMCS controls in preparation of
10426	* hardware-assisted VMX execution of the nested-guest.
10427	*
10428	* For a guest, we don't modify these controls once we set up the VMCS and hence
10429	* this function is never called.
10430	*
10431	* For nested-guests since the nested hypervisor provides these controls on every
10432	* nested-guest VM-entry and could potentially change them everytime we need to
10433	* merge them before every nested-guest VM-entry.
10434	*
10435	* @returns VBox status code.
10436	* @param pVCpu The cross context virtual CPU structure.
10437	*/
10438	static int hmR0VmxMergeVmcsNested(PVMCPUCC pVCpu)
10439	{
10440	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10441	PCVMXVMCSINFO pVmcsInfoGst = &pVCpu->hmr0.s.vmx.VmcsInfo;
10442	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
10443	Assert(pVmcsNstGst);
10444
10445	/*
10446	* Merge the controls with the requirements of the guest VMCS.
10447	*
10448	* We do not need to validate the nested-guest VMX features specified in the nested-guest
10449	* VMCS with the features supported by the physical CPU as it's already done by the
10450	* VMLAUNCH/VMRESUME instruction emulation.
10451	*
10452	* This is because the VMX features exposed by CPUM (through CPUID/MSRs) to the guest are
10453	* derived from the VMX features supported by the physical CPU.
10454	*/
10455
10456	/* Pin-based VM-execution controls. */
10457	uint32_t const u32PinCtls = pVmcsNstGst->u32PinCtls \| pVmcsInfoGst->u32PinCtls;
10458
10459	/* Processor-based VM-execution controls. */
10460	uint32_t u32ProcCtls = (pVmcsNstGst->u32ProcCtls & ~VMX_PROC_CTLS_USE_IO_BITMAPS)
10461	\| (pVmcsInfoGst->u32ProcCtls & ~( VMX_PROC_CTLS_INT_WINDOW_EXIT
10462	\| VMX_PROC_CTLS_NMI_WINDOW_EXIT
10463	\| VMX_PROC_CTLS_MOV_DR_EXIT
10464	\| VMX_PROC_CTLS_USE_TPR_SHADOW
10465	\| VMX_PROC_CTLS_MONITOR_TRAP_FLAG));
10466
10467	/* Secondary processor-based VM-execution controls. */
10468	uint32_t const u32ProcCtls2 = (pVmcsNstGst->u32ProcCtls2 & ~VMX_PROC_CTLS2_VPID)
10469	\| (pVmcsInfoGst->u32ProcCtls2 & ~( VMX_PROC_CTLS2_VIRT_APIC_ACCESS
10470	\| VMX_PROC_CTLS2_INVPCID
10471	\| VMX_PROC_CTLS2_VMCS_SHADOWING
10472	\| VMX_PROC_CTLS2_RDTSCP
10473	\| VMX_PROC_CTLS2_XSAVES_XRSTORS
10474	\| VMX_PROC_CTLS2_APIC_REG_VIRT
10475	\| VMX_PROC_CTLS2_VIRT_INT_DELIVERY
10476	\| VMX_PROC_CTLS2_VMFUNC));
10477
10478	/*
10479	* VM-entry controls:
10480	* These controls contains state that depends on the nested-guest state (primarily
10481	* EFER MSR) and is thus not constant between VMLAUNCH/VMRESUME and the nested-guest
10482	* VM-exit. Although the nested hypervisor cannot change it, we need to in order to
10483	* properly continue executing the nested-guest if the EFER MSR changes but does not
10484	* cause a nested-guest VM-exits.
10485	*
10486	* VM-exit controls:
10487	* These controls specify the host state on return. We cannot use the controls from
10488	* the nested hypervisor state as is as it would contain the guest state rather than
10489	* the host state. Since the host state is subject to change (e.g. preemption, trips
10490	* to ring-3, longjmp and rescheduling to a different host CPU) they are not constant
10491	* through VMLAUNCH/VMRESUME and the nested-guest VM-exit.
10492	*
10493	* VM-entry MSR-load:
10494	* The guest MSRs from the VM-entry MSR-load area are already loaded into the guest-CPU
10495	* context by the VMLAUNCH/VMRESUME instruction emulation.
10496	*
10497	* VM-exit MSR-store:
10498	* The VM-exit emulation will take care of populating the MSRs from the guest-CPU context
10499	* back into the VM-exit MSR-store area.
10500	*
10501	* VM-exit MSR-load areas:
10502	* This must contain the real host MSRs with hardware-assisted VMX execution. Hence, we
10503	* can entirely ignore what the nested hypervisor wants to load here.
10504	*/
10505
10506	/*
10507	* Exception bitmap.
10508	*
10509	* We could remove #UD from the guest bitmap and merge it with the nested-guest bitmap
10510	* here (and avoid doing anything while exporting nested-guest state), but to keep the
10511	* code more flexible if intercepting exceptions become more dynamic in the future we do
10512	* it as part of exporting the nested-guest state.
10513	*/
10514	uint32_t const u32XcptBitmap = pVmcsNstGst->u32XcptBitmap \| pVmcsInfoGst->u32XcptBitmap;
10515
10516	/*
10517	* CR0/CR4 guest/host mask.
10518	*
10519	* Modifications by the nested-guest to CR0/CR4 bits owned by the host and the guest must
10520	* cause VM-exits, so we need to merge them here.
10521	*/
10522	uint64_t const u64Cr0Mask = pVmcsNstGst->u64Cr0Mask.u \| pVmcsInfoGst->u64Cr0Mask;
10523	uint64_t const u64Cr4Mask = pVmcsNstGst->u64Cr4Mask.u \| pVmcsInfoGst->u64Cr4Mask;
10524
10525	/*
10526	* Page-fault error-code mask and match.
10527	*
10528	* Although we require unrestricted guest execution (and thereby nested-paging) for
10529	* hardware-assisted VMX execution of nested-guests and thus the outer guest doesn't
10530	* normally intercept #PFs, it might intercept them for debugging purposes.
10531	*
10532	* If the outer guest is not intercepting #PFs, we can use the nested-guest #PF filters.
10533	* If the outer guest is intercepting #PFs, we must intercept all #PFs.
10534	*/
10535	uint32_t u32XcptPFMask;
10536	uint32_t u32XcptPFMatch;
10537	if (!(pVmcsInfoGst->u32XcptBitmap & RT_BIT(X86_XCPT_PF)))
10538	{
10539	u32XcptPFMask = pVmcsNstGst->u32XcptPFMask;
10540	u32XcptPFMatch = pVmcsNstGst->u32XcptPFMatch;
10541	}
10542	else
10543	{
10544	u32XcptPFMask = 0;
10545	u32XcptPFMatch = 0;
10546	}
10547
10548	/*
10549	* Pause-Loop exiting.
10550	*/
10551	/** @todo r=bird: given that both pVM->hm.s.vmx.cPleGapTicks and
10552	* pVM->hm.s.vmx.cPleWindowTicks defaults to zero, I cannot see how
10553	* this will work... */
10554	uint32_t const cPleGapTicks = RT_MIN(pVM->hm.s.vmx.cPleGapTicks, pVmcsNstGst->u32PleGap);
10555	uint32_t const cPleWindowTicks = RT_MIN(pVM->hm.s.vmx.cPleWindowTicks, pVmcsNstGst->u32PleWindow);
10556
10557	/*
10558	* Pending debug exceptions.
10559	* Currently just copy whatever the nested-guest provides us.
10560	*/
10561	uint64_t const uPendingDbgXcpts = pVmcsNstGst->u64GuestPendingDbgXcpts.u;
10562
10563	/*
10564	* I/O Bitmap.
10565	*
10566	* We do not use the I/O bitmap that may be provided by the nested hypervisor as we always
10567	* intercept all I/O port accesses.
10568	*/
10569	Assert(u32ProcCtls & VMX_PROC_CTLS_UNCOND_IO_EXIT);
10570	Assert(!(u32ProcCtls & VMX_PROC_CTLS_USE_IO_BITMAPS));
10571
10572	/*
10573	* VMCS shadowing.
10574	*
10575	* We do not yet expose VMCS shadowing to the guest and thus VMCS shadowing should not be
10576	* enabled while executing the nested-guest.
10577	*/
10578	Assert(!(u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING));
10579
10580	/*
10581	* APIC-access page.
10582	*/
10583	RTHCPHYS HCPhysApicAccess;
10584	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10585	{
10586	Assert(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS);
10587	RTGCPHYS const GCPhysApicAccess = pVmcsNstGst->u64AddrApicAccess.u;
10588
10589	/** @todo NSTVMX: This is not really correct but currently is required to make
10590	* things work. We need to re-enable the page handler when we fallback to
10591	* IEM execution of the nested-guest! */
10592	PGMHandlerPhysicalPageTempOff(pVM, GCPhysApicAccess, GCPhysApicAccess);
10593
10594	void *pvPage;
10595	PGMPAGEMAPLOCK PgLockApicAccess;
10596	int rc = PGMPhysGCPhys2CCPtr(pVM, GCPhysApicAccess, &pvPage, &PgLockApicAccess);
10597	if (RT_SUCCESS(rc))
10598	{
10599	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysApicAccess, &HCPhysApicAccess);
10600	AssertMsgRCReturn(rc, ("Failed to get host-physical address for APIC-access page at %#RGp\n", GCPhysApicAccess), rc);
10601
10602	/** @todo Handle proper releasing of page-mapping lock later. */
10603	PGMPhysReleasePageMappingLock(pVCpu->CTX_SUFF(pVM), &PgLockApicAccess);
10604	}
10605	else
10606	return rc;
10607	}
10608	else
10609	HCPhysApicAccess = 0;
10610
10611	/*
10612	* Virtual-APIC page and TPR threshold.
10613	*/
10614	RTHCPHYS HCPhysVirtApic;
10615	uint32_t u32TprThreshold;
10616	if (u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
10617	{
10618	Assert(g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW);
10619	RTGCPHYS const GCPhysVirtApic = pVmcsNstGst->u64AddrVirtApic.u;
10620
10621	void *pvPage;
10622	PGMPAGEMAPLOCK PgLockVirtApic;
10623	int rc = PGMPhysGCPhys2CCPtr(pVM, GCPhysVirtApic, &pvPage, &PgLockVirtApic);
10624	if (RT_SUCCESS(rc))
10625	{
10626	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysVirtApic, &HCPhysVirtApic);
10627	AssertMsgRCReturn(rc, ("Failed to get host-physical address for virtual-APIC page at %#RGp\n", GCPhysVirtApic), rc);
10628
10629	/** @todo Handle proper releasing of page-mapping lock later. */
10630	PGMPhysReleasePageMappingLock(pVCpu->CTX_SUFF(pVM), &PgLockVirtApic);
10631	}
10632	else
10633	return rc;
10634
10635	u32TprThreshold = pVmcsNstGst->u32TprThreshold;
10636	}
10637	else
10638	{
10639	HCPhysVirtApic = 0;
10640	u32TprThreshold = 0;
10641
10642	/*
10643	* We must make sure CR8 reads/write must cause VM-exits when TPR shadowing is not
10644	* used by the nested hypervisor. Preventing MMIO accesses to the physical APIC will
10645	* be taken care of by EPT/shadow paging.
10646	*/
10647	if (pVM->hmr0.s.fAllow64BitGuests)
10648	u32ProcCtls \|= VMX_PROC_CTLS_CR8_STORE_EXIT
10649	\| VMX_PROC_CTLS_CR8_LOAD_EXIT;
10650	}
10651
10652	/*
10653	* Validate basic assumptions.
10654	*/
10655	PVMXVMCSINFO pVmcsInfoNstGst = &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
10656	Assert(pVM->hmr0.s.vmx.fUnrestrictedGuest);
10657	Assert(g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS);
10658	Assert(hmGetVmxActiveVmcsInfo(pVCpu) == pVmcsInfoNstGst);
10659
10660	/*
10661	* Commit it to the nested-guest VMCS.
10662	*/
10663	int rc = VINF_SUCCESS;
10664	if (pVmcsInfoNstGst->u32PinCtls != u32PinCtls)
10665	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, u32PinCtls);
10666	if (pVmcsInfoNstGst->u32ProcCtls != u32ProcCtls)
10667	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, u32ProcCtls);
10668	if (pVmcsInfoNstGst->u32ProcCtls2 != u32ProcCtls2)
10669	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, u32ProcCtls2);
10670	if (pVmcsInfoNstGst->u32XcptBitmap != u32XcptBitmap)
10671	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
10672	if (pVmcsInfoNstGst->u64Cr0Mask != u64Cr0Mask)
10673	rc \|= VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, u64Cr0Mask);
10674	if (pVmcsInfoNstGst->u64Cr4Mask != u64Cr4Mask)
10675	rc \|= VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, u64Cr4Mask);
10676	if (pVmcsInfoNstGst->u32XcptPFMask != u32XcptPFMask)
10677	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, u32XcptPFMask);
10678	if (pVmcsInfoNstGst->u32XcptPFMatch != u32XcptPFMatch)
10679	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, u32XcptPFMatch);
10680	if ( !(u32ProcCtls & VMX_PROC_CTLS_PAUSE_EXIT)
10681	&& (u32ProcCtls2 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT))
10682	{
10683	Assert(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT);
10684	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, cPleGapTicks);
10685	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, cPleWindowTicks);
10686	}
10687	if (u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
10688	{
10689	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
10690	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL, HCPhysVirtApic);
10691	}
10692	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10693	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, HCPhysApicAccess);
10694	rc \|= VMXWriteVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, uPendingDbgXcpts);
10695	AssertRC(rc);
10696
10697	/*
10698	* Update the nested-guest VMCS cache.
10699	*/
10700	pVmcsInfoNstGst->u32PinCtls = u32PinCtls;
10701	pVmcsInfoNstGst->u32ProcCtls = u32ProcCtls;
10702	pVmcsInfoNstGst->u32ProcCtls2 = u32ProcCtls2;
10703	pVmcsInfoNstGst->u32XcptBitmap = u32XcptBitmap;
10704	pVmcsInfoNstGst->u64Cr0Mask = u64Cr0Mask;
10705	pVmcsInfoNstGst->u64Cr4Mask = u64Cr4Mask;
10706	pVmcsInfoNstGst->u32XcptPFMask = u32XcptPFMask;
10707	pVmcsInfoNstGst->u32XcptPFMatch = u32XcptPFMatch;
10708	pVmcsInfoNstGst->HCPhysVirtApic = HCPhysVirtApic;
10709
10710	/*
10711	* We need to flush the TLB if we are switching the APIC-access page address.
10712	* See Intel spec. 28.3.3.4 "Guidelines for Use of the INVEPT Instruction".
10713	*/
10714	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10715	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = true;
10716
10717	/*
10718	* MSR bitmap.
10719	*
10720	* The MSR bitmap address has already been initialized while setting up the nested-guest
10721	* VMCS, here we need to merge the MSR bitmaps.
10722	*/
10723	if (u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
10724	hmR0VmxMergeMsrBitmapNested(pVCpu, pVmcsInfoNstGst, pVmcsInfoGst);
10725
10726	return VINF_SUCCESS;
10727	}
10728	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
10729
10730
10731	/**
10732	* Does the preparations before executing guest code in VT-x.
10733	*
10734	* This may cause longjmps to ring-3 and may even result in rescheduling to the
10735	* recompiler/IEM. We must be cautious what we do here regarding committing
10736	* guest-state information into the VMCS assuming we assuredly execute the
10737	* guest in VT-x mode.
10738	*
10739	* If we fall back to the recompiler/IEM after updating the VMCS and clearing
10740	* the common-state (TRPM/forceflags), we must undo those changes so that the
10741	* recompiler/IEM can (and should) use them when it resumes guest execution.
10742	* Otherwise such operations must be done when we can no longer exit to ring-3.
10743	*
10744	* @returns Strict VBox status code (i.e. informational status codes too).
10745	* @retval VINF_SUCCESS if we can proceed with running the guest, interrupts
10746	* have been disabled.
10747	* @retval VINF_VMX_VMEXIT if a nested-guest VM-exit occurs (e.g., while evaluating
10748	* pending events).
10749	* @retval VINF_EM_RESET if a triple-fault occurs while injecting a
10750	* double-fault into the guest.
10751	* @retval VINF_EM_DBG_STEPPED if @a fStepping is true and an event was
10752	* dispatched directly.
10753	* @retval VINF_* scheduling changes, we have to go back to ring-3.
10754	*
10755	* @param pVCpu The cross context virtual CPU structure.
10756	* @param pVmxTransient The VMX-transient structure.
10757	* @param fStepping Whether we are single-stepping the guest in the
10758	* hypervisor debugger. Makes us ignore some of the reasons
10759	* for returning to ring-3, and return VINF_EM_DBG_STEPPED
10760	* if event dispatching took place.
10761	*/
10762	static VBOXSTRICTRC hmR0VmxPreRunGuest(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, bool fStepping)
10763	{
10764	Assert(VMMRZCallRing3IsEnabled(pVCpu));
10765
10766	Log4Func(("fIsNested=%RTbool fStepping=%RTbool\n", pVmxTransient->fIsNestedGuest, fStepping));
10767
10768	#ifdef VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM
10769	if (pVmxTransient->fIsNestedGuest)
10770	{
10771	RT_NOREF2(pVCpu, fStepping);
10772	Log2Func(("Rescheduling to IEM due to nested-hwvirt or forced IEM exec -> VINF_EM_RESCHEDULE_REM\n"));
10773	return VINF_EM_RESCHEDULE_REM;
10774	}
10775	#endif
10776
10777	/*
10778	* Check and process force flag actions, some of which might require us to go back to ring-3.
10779	*/
10780	VBOXSTRICTRC rcStrict = hmR0VmxCheckForceFlags(pVCpu, pVmxTransient, fStepping);
10781	if (rcStrict == VINF_SUCCESS)
10782	{
10783	/* FFs don't get set all the time. */
10784	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10785	if ( pVmxTransient->fIsNestedGuest
10786	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
10787	{
10788	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
10789	return VINF_VMX_VMEXIT;
10790	}
10791	#endif
10792	}
10793	else
10794	return rcStrict;
10795
10796	/*
10797	* Virtualize memory-mapped accesses to the physical APIC (may take locks).
10798	*/
10799	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10800	if ( !pVCpu->hm.s.vmx.u64GstMsrApicBase
10801	&& (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10802	&& PDMHasApic(pVM))
10803	{
10804	int rc = hmR0VmxMapHCApicAccessPage(pVCpu);
10805	AssertRCReturn(rc, rc);
10806	}
10807
10808	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10809	/*
10810	* Merge guest VMCS controls with the nested-guest VMCS controls.
10811	*
10812	* Even if we have not executed the guest prior to this (e.g. when resuming from a
10813	* saved state), we should be okay with merging controls as we initialize the
10814	* guest VMCS controls as part of VM setup phase.
10815	*/
10816	if ( pVmxTransient->fIsNestedGuest
10817	&& !pVCpu->hm.s.vmx.fMergedNstGstCtls)
10818	{
10819	int rc = hmR0VmxMergeVmcsNested(pVCpu);
10820	AssertRCReturn(rc, rc);
10821	pVCpu->hm.s.vmx.fMergedNstGstCtls = true;
10822	}
10823	#endif
10824
10825	/*
10826	* Evaluate events to be injected into the guest.
10827	*
10828	* Events in TRPM can be injected without inspecting the guest state.
10829	* If any new events (interrupts/NMI) are pending currently, we try to set up the
10830	* guest to cause a VM-exit the next time they are ready to receive the event.
10831	*/
10832	if (TRPMHasTrap(pVCpu))
10833	hmR0VmxTrpmTrapToPendingEvent(pVCpu);
10834
10835	uint32_t fIntrState;
10836	rcStrict = hmR0VmxEvaluatePendingEvent(pVCpu, pVmxTransient, &fIntrState);
10837
10838	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10839	/*
10840	* While evaluating pending events if something failed (unlikely) or if we were
10841	* preparing to run a nested-guest but performed a nested-guest VM-exit, we should bail.
10842	*/
10843	if (rcStrict != VINF_SUCCESS)
10844	return rcStrict;
10845	if ( pVmxTransient->fIsNestedGuest
10846	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
10847	{
10848	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
10849	return VINF_VMX_VMEXIT;
10850	}
10851	#else
10852	Assert(rcStrict == VINF_SUCCESS);
10853	#endif
10854
10855	/*
10856	* Event injection may take locks (currently the PGM lock for real-on-v86 case) and thus
10857	* needs to be done with longjmps or interrupts + preemption enabled. Event injection might
10858	* also result in triple-faulting the VM.
10859	*
10860	* With nested-guests, the above does not apply since unrestricted guest execution is a
10861	* requirement. Regardless, we do this here to avoid duplicating code elsewhere.
10862	*/
10863	rcStrict = hmR0VmxInjectPendingEvent(pVCpu, pVmxTransient, fIntrState, fStepping);
10864	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
10865	{ /* likely */ }
10866	else
10867	{
10868	AssertMsg(rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
10869	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
10870	return rcStrict;
10871	}
10872
10873	/*
10874	* A longjump might result in importing CR3 even for VM-exits that don't necessarily
10875	* import CR3 themselves. We will need to update them here, as even as late as the above
10876	* hmR0VmxInjectPendingEvent() call may lazily import guest-CPU state on demand causing
10877	* the below force flags to be set.
10878	*/
10879	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
10880	{
10881	Assert(!(ASMAtomicUoReadU64(&pVCpu->cpum.GstCtx.fExtrn) & CPUMCTX_EXTRN_CR3));
10882	int rc2 = PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
10883	AssertMsgReturn(rc2 == VINF_SUCCESS \|\| rc2 == VINF_PGM_SYNC_CR3,
10884	("%Rrc\n", rc2), RT_FAILURE_NP(rc2) ? rc2 : VERR_IPE_UNEXPECTED_INFO_STATUS);
10885	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
10886	}
10887	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
10888	{
10889	PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
10890	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
10891	}
10892
10893	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10894	/* Paranoia. */
10895	Assert(!pVmxTransient->fIsNestedGuest \|\| CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
10896	#endif
10897
10898	/*
10899	* No longjmps to ring-3 from this point on!!!
10900	* Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
10901	* This also disables flushing of the R0-logger instance (if any).
10902	*/
10903	VMMRZCallRing3Disable(pVCpu);
10904
10905	/*
10906	* Export the guest state bits.
10907	*
10908	* We cannot perform longjmps while loading the guest state because we do not preserve the
10909	* host/guest state (although the VMCS will be preserved) across longjmps which can cause
10910	* CPU migration.
10911	*
10912	* If we are injecting events to a real-on-v86 mode guest, we would have updated RIP and some segment
10913	* registers. Hence, exporting of the guest state needs to be done -after- injection of events.
10914	*/
10915	rcStrict = hmR0VmxExportGuestStateOptimal(pVCpu, pVmxTransient);
10916	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
10917	{ /* likely */ }
10918	else
10919	{
10920	VMMRZCallRing3Enable(pVCpu);
10921	return rcStrict;
10922	}
10923
10924	/*
10925	* We disable interrupts so that we don't miss any interrupts that would flag preemption
10926	* (IPI/timers etc.) when thread-context hooks aren't used and we've been running with
10927	* preemption disabled for a while. Since this is purely to aid the
10928	* RTThreadPreemptIsPending() code, it doesn't matter that it may temporarily reenable and
10929	* disable interrupt on NT.
10930	*
10931	* We need to check for force-flags that could've possible been altered since we last
10932	* checked them (e.g. by PDMGetInterrupt() leaving the PDM critical section,
10933	* see @bugref{6398}).
10934	*
10935	* We also check a couple of other force-flags as a last opportunity to get the EMT back
10936	* to ring-3 before executing guest code.
10937	*/
10938	pVmxTransient->fEFlags = ASMIntDisableFlags();
10939
10940	if ( ( !VM_FF_IS_ANY_SET(pVM, VM_FF_EMT_RENDEZVOUS \| VM_FF_TM_VIRTUAL_SYNC)
10941	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
10942	\|\| ( fStepping /* Optimized for the non-stepping case, so a bit of unnecessary work when stepping. */
10943	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK & ~(VMCPU_FF_TIMER \| VMCPU_FF_PDM_CRITSECT))) )
10944	{
10945	if (!RTThreadPreemptIsPending(NIL_RTTHREAD))
10946	{
10947	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10948	/*
10949	* If we are executing a nested-guest make sure that we should intercept subsequent
10950	* events. The one we are injecting might be part of VM-entry. This is mainly to keep
10951	* the VM-exit instruction emulation happy.
10952	*/
10953	if (pVmxTransient->fIsNestedGuest)
10954	CPUMSetGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx, true);
10955	#endif
10956
10957	/*
10958	* We've injected any pending events. This is really the point of no return (to ring-3).
10959	*
10960	* Note! The caller expects to continue with interrupts & longjmps disabled on successful
10961	* returns from this function, so do -not- enable them here.
10962	*/
10963	pVCpu->hm.s.Event.fPending = false;
10964	return VINF_SUCCESS;
10965	}
10966
10967	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPendingHostIrq);
10968	rcStrict = VINF_EM_RAW_INTERRUPT;
10969	}
10970	else
10971	{
10972	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
10973	rcStrict = VINF_EM_RAW_TO_R3;
10974	}
10975
10976	ASMSetFlags(pVmxTransient->fEFlags);
10977	VMMRZCallRing3Enable(pVCpu);
10978
10979	return rcStrict;
10980	}
10981
10982
10983	/**
10984	* Final preparations before executing guest code using hardware-assisted VMX.
10985	*
10986	* We can no longer get preempted to a different host CPU and there are no returns
10987	* to ring-3. We ignore any errors that may happen from this point (e.g. VMWRITE
10988	* failures), this function is not intended to fail sans unrecoverable hardware
10989	* errors.
10990	*
10991	* @param pVCpu The cross context virtual CPU structure.
10992	* @param pVmxTransient The VMX-transient structure.
10993	*
10994	* @remarks Called with preemption disabled.
10995	* @remarks No-long-jump zone!!!
10996	*/
10997	static void hmR0VmxPreRunGuestCommitted(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
10998	{
10999	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
11000	Assert(VMMR0IsLogFlushDisabled(pVCpu));
11001	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
11002	Assert(!pVCpu->hm.s.Event.fPending);
11003
11004	/*
11005	* Indicate start of guest execution and where poking EMT out of guest-context is recognized.
11006	*/
11007	VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
11008	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC);
11009
11010	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
11011	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11012	PHMPHYSCPU pHostCpu = hmR0GetCurrentCpu();
11013	RTCPUID const idCurrentCpu = pHostCpu->idCpu;
11014
11015	if (!CPUMIsGuestFPUStateActive(pVCpu))
11016	{
11017	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestFpuState, x);
11018	if (CPUMR0LoadGuestFPU(pVM, pVCpu) == VINF_CPUM_HOST_CR0_MODIFIED)
11019	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_HOST_CONTEXT;
11020	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestFpuState, x);
11021	STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadGuestFpu);
11022	}
11023
11024	/*
11025	* Re-export the host state bits as we may've been preempted (only happens when
11026	* thread-context hooks are used or when the VM start function changes) or if
11027	* the host CR0 is modified while loading the guest FPU state above.
11028	*
11029	* The 64-on-32 switcher saves the (64-bit) host state into the VMCS and if we
11030	* changed the switcher back to 32-bit, we must save the 32-bit host state here,
11031	* see @bugref{8432}.
11032	*
11033	* This may also happen when switching to/from a nested-guest VMCS without leaving
11034	* ring-0.
11035	*/
11036	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT)
11037	{
11038	hmR0VmxExportHostState(pVCpu);
11039	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportHostState);
11040	}
11041	Assert(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT));
11042
11043	/*
11044	* Export the state shared between host and guest (FPU, debug, lazy MSRs).
11045	*/
11046	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE)
11047	hmR0VmxExportSharedState(pVCpu, pVmxTransient);
11048	AssertMsg(!pVCpu->hm.s.fCtxChanged, ("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
11049
11050	/*
11051	* Store status of the shared guest/host debug state at the time of VM-entry.
11052	*/
11053	pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
11054	pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
11055
11056	/*
11057	* Always cache the TPR-shadow if the virtual-APIC page exists, thereby skipping
11058	* more than one conditional check. The post-run side of our code shall determine
11059	* if it needs to sync. the virtual APIC TPR with the TPR-shadow.
11060	*/
11061	if (pVmcsInfo->pbVirtApic)
11062	pVmxTransient->u8GuestTpr = pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR];
11063
11064	/*
11065	* Update the host MSRs values in the VM-exit MSR-load area.
11066	*/
11067	if (!pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs)
11068	{
11069	if (pVmcsInfo->cExitMsrLoad > 0)
11070	hmR0VmxUpdateAutoLoadHostMsrs(pVCpu, pVmcsInfo);
11071	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = true;
11072	}
11073
11074	/*
11075	* Evaluate if we need to intercept guest RDTSC/P accesses. Set up the
11076	* VMX-preemption timer based on the next virtual sync clock deadline.
11077	*/
11078	if ( !pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer
11079	\|\| idCurrentCpu != pVCpu->hmr0.s.idLastCpu)
11080	{
11081	hmR0VmxUpdateTscOffsettingAndPreemptTimer(pVCpu, pVmxTransient, idCurrentCpu);
11082	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = true;
11083	}
11084
11085	/* Record statistics of how often we use TSC offsetting as opposed to intercepting RDTSC/P. */
11086	bool const fIsRdtscIntercepted = RT_BOOL(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_RDTSC_EXIT);
11087	if (!fIsRdtscIntercepted)
11088	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
11089	else
11090	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
11091
11092	ASMAtomicUoWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB flushing, set this across the world switch. */
11093	hmR0VmxFlushTaggedTlb(pHostCpu, pVCpu, pVmcsInfo); /* Invalidate the appropriate guest entries from the TLB. */
11094	Assert(idCurrentCpu == pVCpu->hmr0.s.idLastCpu);
11095	pVCpu->hm.s.vmx.LastError.idCurrentCpu = idCurrentCpu; /* Record the error reporting info. with the current host CPU. */
11096	pVmcsInfo->idHostCpuState = idCurrentCpu; /* Record the CPU for which the host-state has been exported. */
11097	pVmcsInfo->idHostCpuExec = idCurrentCpu; /* Record the CPU on which we shall execute. */
11098
11099	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
11100
11101	TMNotifyStartOfExecution(pVM, pVCpu); /* Notify TM to resume its clocks when TSC is tied to execution,
11102	as we're about to start executing the guest. */
11103
11104	/*
11105	* Load the guest TSC_AUX MSR when we are not intercepting RDTSCP.
11106	*
11107	* This is done this late as updating the TSC offsetting/preemption timer above
11108	* figures out if we can skip intercepting RDTSCP by calculating the number of
11109	* host CPU ticks till the next virtual sync deadline (for the dynamic case).
11110	*/
11111	if ( (pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_RDTSCP)
11112	&& !fIsRdtscIntercepted)
11113	{
11114	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_TSC_AUX);
11115
11116	/* NB: Because we call hmR0VmxAddAutoLoadStoreMsr with fUpdateHostMsr=true,
11117	it's safe even after hmR0VmxUpdateAutoLoadHostMsrs has already been done. */
11118	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_TSC_AUX, CPUMGetGuestTscAux(pVCpu),
11119	true /* fSetReadWrite /, true / fUpdateHostMsr */);
11120	AssertRC(rc);
11121	Assert(!pVmxTransient->fRemoveTscAuxMsr);
11122	pVmxTransient->fRemoveTscAuxMsr = true;
11123	}
11124
11125	#ifdef VBOX_STRICT
11126	Assert(pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs);
11127	hmR0VmxCheckAutoLoadStoreMsrs(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest);
11128	hmR0VmxCheckHostEferMsr(pVmcsInfo);
11129	AssertRC(hmR0VmxCheckCachedVmcsCtls(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest));
11130	#endif
11131
11132	#ifdef HMVMX_ALWAYS_CHECK_GUEST_STATE
11133	/** @todo r=ramshankar: We can now probably use iemVmxVmentryCheckGuestState here.
11134	* Add a PVMXMSRS parameter to it, so that IEM can look at the host MSRs,
11135	* see @bugref{9180#c54}. */
11136	uint32_t const uInvalidReason = hmR0VmxCheckGuestState(pVCpu, pVmcsInfo);
11137	if (uInvalidReason != VMX_IGS_REASON_NOT_FOUND)
11138	Log4(("hmR0VmxCheckGuestState returned %#x\n", uInvalidReason));
11139	#endif
11140	}
11141
11142
11143	/**
11144	* First C routine invoked after running guest code using hardware-assisted VMX.
11145	*
11146	* @param pVCpu The cross context virtual CPU structure.
11147	* @param pVmxTransient The VMX-transient structure.
11148	* @param rcVMRun Return code of VMLAUNCH/VMRESUME.
11149	*
11150	* @remarks Called with interrupts disabled, and returns with interrupts enabled!
11151	*
11152	* @remarks No-long-jump zone!!! This function will however re-enable longjmps
11153	* unconditionally when it is safe to do so.
11154	*/
11155	static void hmR0VmxPostRunGuest(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, int rcVMRun)
11156	{
11157	ASMAtomicUoWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB flushing. */
11158	ASMAtomicIncU32(&pVCpu->hmr0.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for EMT poking. */
11159	pVCpu->hm.s.fCtxChanged = 0; /* Exits/longjmps to ring-3 requires saving the guest state. */
11160	pVmxTransient->fVmcsFieldsRead = 0; /* Transient fields need to be read from the VMCS. */
11161	pVmxTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
11162	pVmxTransient->fVectoringDoublePF = false; /* Vectoring double page-fault needs to be determined later. */
11163
11164	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11165	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_RDTSC_EXIT))
11166	{
11167	uint64_t uGstTsc;
11168	if (!pVmxTransient->fIsNestedGuest)
11169	uGstTsc = pVCpu->hmr0.s.uTscExit + pVmcsInfo->u64TscOffset;
11170	else
11171	{
11172	uint64_t const uNstGstTsc = pVCpu->hmr0.s.uTscExit + pVmcsInfo->u64TscOffset;
11173	uGstTsc = CPUMRemoveNestedGuestTscOffset(pVCpu, uNstGstTsc);
11174	}
11175	TMCpuTickSetLastSeen(pVCpu, uGstTsc); /* Update TM with the guest TSC. */
11176	}
11177
11178	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatPreExit, x);
11179	TMNotifyEndOfExecution(pVCpu->CTX_SUFF(pVM), pVCpu, pVCpu->hmr0.s.uTscExit); /* Notify TM that the guest is no longer running. */
11180	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
11181
11182	pVCpu->hmr0.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_REQUIRED; /* Some host state messed up by VMX needs restoring. */
11183	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
11184	#ifdef VBOX_STRICT
11185	hmR0VmxCheckHostEferMsr(pVmcsInfo); /* Verify that the host EFER MSR wasn't modified. */
11186	#endif
11187	Assert(!ASMIntAreEnabled());
11188	ASMSetFlags(pVmxTransient->fEFlags); /* Enable interrupts. */
11189	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
11190
11191	#ifdef HMVMX_ALWAYS_CLEAN_TRANSIENT
11192	/*
11193	* Clean all the VMCS fields in the transient structure before reading
11194	* anything from the VMCS.
11195	*/
11196	pVmxTransient->uExitReason = 0;
11197	pVmxTransient->uExitIntErrorCode = 0;
11198	pVmxTransient->uExitQual = 0;
11199	pVmxTransient->uGuestLinearAddr = 0;
11200	pVmxTransient->uExitIntInfo = 0;
11201	pVmxTransient->cbExitInstr = 0;
11202	pVmxTransient->ExitInstrInfo.u = 0;
11203	pVmxTransient->uEntryIntInfo = 0;
11204	pVmxTransient->uEntryXcptErrorCode = 0;
11205	pVmxTransient->cbEntryInstr = 0;
11206	pVmxTransient->uIdtVectoringInfo = 0;
11207	pVmxTransient->uIdtVectoringErrorCode = 0;
11208	#endif
11209
11210	/*
11211	* Save the basic VM-exit reason and check if the VM-entry failed.
11212	* See Intel spec. 24.9.1 "Basic VM-exit Information".
11213	*/
11214	uint32_t uExitReason;
11215	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
11216	AssertRC(rc);
11217	pVmxTransient->uExitReason = VMX_EXIT_REASON_BASIC(uExitReason);
11218	pVmxTransient->fVMEntryFailed = VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason);
11219
11220	/*
11221	* Log the VM-exit before logging anything else as otherwise it might be a
11222	* tad confusing what happens before and after the world-switch.
11223	*/
11224	HMVMX_LOG_EXIT(pVCpu, uExitReason);
11225
11226	/*
11227	* Remove the TSC_AUX MSR from the auto-load/store MSR area and reset any MSR
11228	* bitmap permissions, if it was added before VM-entry.
11229	*/
11230	if (pVmxTransient->fRemoveTscAuxMsr)
11231	{
11232	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_TSC_AUX);
11233	pVmxTransient->fRemoveTscAuxMsr = false;
11234	}
11235
11236	/*
11237	* Check if VMLAUNCH/VMRESUME succeeded.
11238	* If this failed, we cause a guru meditation and cease further execution.
11239	*
11240	* However, if we are executing a nested-guest we might fail if we use the
11241	* fast path rather than fully emulating VMLAUNCH/VMRESUME instruction in IEM.
11242	*/
11243	if (RT_LIKELY(rcVMRun == VINF_SUCCESS))
11244	{
11245	/*
11246	* Update the VM-exit history array here even if the VM-entry failed due to:
11247	* - Invalid guest state.
11248	* - MSR loading.
11249	* - Machine-check event.
11250	*
11251	* In any of the above cases we will still have a "valid" VM-exit reason
11252	* despite @a fVMEntryFailed being false.
11253	*
11254	* See Intel spec. 26.7 "VM-Entry failures during or after loading guest state".
11255	*
11256	* Note! We don't have CS or RIP at this point. Will probably address that later
11257	* by amending the history entry added here.
11258	*/
11259	EMHistoryAddExit(pVCpu, EMEXIT_MAKE_FT(EMEXIT_F_KIND_VMX, pVmxTransient->uExitReason & EMEXIT_F_TYPE_MASK),
11260	UINT64_MAX, pVCpu->hmr0.s.uTscExit);
11261
11262	if (RT_LIKELY(!pVmxTransient->fVMEntryFailed))
11263	{
11264	VMMRZCallRing3Enable(pVCpu);
11265
11266	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
11267	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
11268
11269	#ifdef HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
11270	hmR0VmxReadAllRoFieldsVmcs(pVmxTransient);
11271	#endif
11272
11273	/*
11274	* Import the guest-interruptibility state always as we need it while evaluating
11275	* injecting events on re-entry.
11276	*
11277	* We don't import CR0 (when unrestricted guest execution is unavailable) despite
11278	* checking for real-mode while exporting the state because all bits that cause
11279	* mode changes wrt CR0 are intercepted.
11280	*/
11281	uint64_t const fImportMask = CPUMCTX_EXTRN_HM_VMX_INT_STATE
11282	#if defined(HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE) \|\| defined(HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE)
11283	\| HMVMX_CPUMCTX_EXTRN_ALL
11284	#elif defined(HMVMX_ALWAYS_SAVE_GUEST_RFLAGS)
11285	\| CPUMCTX_EXTRN_RFLAGS
11286	#endif
11287	;
11288	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImportMask);
11289	AssertRC(rc);
11290
11291	/*
11292	* Sync the TPR shadow with our APIC state.
11293	*/
11294	if ( !pVmxTransient->fIsNestedGuest
11295	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW))
11296	{
11297	Assert(pVmcsInfo->pbVirtApic);
11298	if (pVmxTransient->u8GuestTpr != pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR])
11299	{
11300	rc = APICSetTpr(pVCpu, pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR]);
11301	AssertRC(rc);
11302	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
11303	}
11304	}
11305
11306	Assert(VMMRZCallRing3IsEnabled(pVCpu));
11307	Assert( pVmxTransient->fWasGuestDebugStateActive == false
11308	\|\| pVmxTransient->fWasHyperDebugStateActive == false);
11309	return;
11310	}
11311	}
11312	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11313	else if (pVmxTransient->fIsNestedGuest)
11314	AssertMsgFailed(("VMLAUNCH/VMRESUME failed but shouldn't happen when VMLAUNCH/VMRESUME was emulated in IEM!\n"));
11315	#endif
11316	else
11317	Log4Func(("VM-entry failure: rcVMRun=%Rrc fVMEntryFailed=%RTbool\n", rcVMRun, pVmxTransient->fVMEntryFailed));
11318
11319	VMMRZCallRing3Enable(pVCpu);
11320	}
11321
11322
11323	/**
11324	* Runs the guest code using hardware-assisted VMX the normal way.
11325	*
11326	* @returns VBox status code.
11327	* @param pVCpu The cross context virtual CPU structure.
11328	* @param pcLoops Pointer to the number of executed loops.
11329	*/
11330	static VBOXSTRICTRC hmR0VmxRunGuestCodeNormal(PVMCPUCC pVCpu, uint32_t *pcLoops)
11331	{
11332	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops;
11333	Assert(pcLoops);
11334	Assert(*pcLoops <= cMaxResumeLoops);
11335	Assert(!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
11336
11337	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11338	/*
11339	* Switch to the guest VMCS as we may have transitioned from executing the nested-guest
11340	* without leaving ring-0. Otherwise, if we came from ring-3 we would have loaded the
11341	* guest VMCS while entering the VMX ring-0 session.
11342	*/
11343	if (pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs)
11344	{
11345	int rc = hmR0VmxSwitchToGstOrNstGstVmcs(pVCpu, false /* fSwitchToNstGstVmcs */);
11346	if (RT_SUCCESS(rc))
11347	{ /* likely */ }
11348	else
11349	{
11350	LogRelFunc(("Failed to switch to the guest VMCS. rc=%Rrc\n", rc));
11351	return rc;
11352	}
11353	}
11354	#endif
11355
11356	VMXTRANSIENT VmxTransient;
11357	RT_ZERO(VmxTransient);
11358	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
11359
11360	/* Paranoia. */
11361	Assert(VmxTransient.pVmcsInfo == &pVCpu->hmr0.s.vmx.VmcsInfo);
11362
11363	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
11364	for (;;)
11365	{
11366	Assert(!HMR0SuspendPending());
11367	HMVMX_ASSERT_CPU_SAFE(pVCpu);
11368	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
11369
11370	/*
11371	* Preparatory work for running nested-guest code, this may force us to
11372	* return to ring-3.
11373	*
11374	* Warning! This bugger disables interrupts on VINF_SUCCESS!
11375	*/
11376	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, false /* fStepping */);
11377	if (rcStrict != VINF_SUCCESS)
11378	break;
11379
11380	/* Interrupts are disabled at this point! */
11381	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
11382	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
11383	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
11384	/* Interrupts are re-enabled at this point! */
11385
11386	/*
11387	* Check for errors with running the VM (VMLAUNCH/VMRESUME).
11388	*/
11389	if (RT_SUCCESS(rcRun))
11390	{ /* very likely */ }
11391	else
11392	{
11393	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
11394	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
11395	return rcRun;
11396	}
11397
11398	/*
11399	* Profile the VM-exit.
11400	*/
11401	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
11402	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
11403	STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
11404	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
11405	HMVMX_START_EXIT_DISPATCH_PROF();
11406
11407	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
11408
11409	/*
11410	* Handle the VM-exit.
11411	*/
11412	#ifdef HMVMX_USE_FUNCTION_TABLE
11413	rcStrict = g_aVMExitHandlers[VmxTransient.uExitReason].pfn(pVCpu, &VmxTransient);
11414	#else
11415	rcStrict = hmR0VmxHandleExit(pVCpu, &VmxTransient);
11416	#endif
11417	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
11418	if (rcStrict == VINF_SUCCESS)
11419	{
11420	if (++(*pcLoops) <= cMaxResumeLoops)
11421	continue;
11422	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
11423	rcStrict = VINF_EM_RAW_INTERRUPT;
11424	}
11425	break;
11426	}
11427
11428	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
11429	return rcStrict;
11430	}
11431
11432
11433	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11434	/**
11435	* Runs the nested-guest code using hardware-assisted VMX.
11436	*
11437	* @returns VBox status code.
11438	* @param pVCpu The cross context virtual CPU structure.
11439	* @param pcLoops Pointer to the number of executed loops.
11440	*
11441	* @sa hmR0VmxRunGuestCodeNormal.
11442	*/
11443	static VBOXSTRICTRC hmR0VmxRunGuestCodeNested(PVMCPUCC pVCpu, uint32_t *pcLoops)
11444	{
11445	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops;
11446	Assert(pcLoops);
11447	Assert(*pcLoops <= cMaxResumeLoops);
11448	Assert(CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
11449
11450	/*
11451	* Switch to the nested-guest VMCS as we may have transitioned from executing the
11452	* guest without leaving ring-0. Otherwise, if we came from ring-3 we would have
11453	* loaded the nested-guest VMCS while entering the VMX ring-0 session.
11454	*/
11455	if (!pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs)
11456	{
11457	int rc = hmR0VmxSwitchToGstOrNstGstVmcs(pVCpu, true /* fSwitchToNstGstVmcs */);
11458	if (RT_SUCCESS(rc))
11459	{ /* likely */ }
11460	else
11461	{
11462	LogRelFunc(("Failed to switch to the nested-guest VMCS. rc=%Rrc\n", rc));
11463	return rc;
11464	}
11465	}
11466
11467	VMXTRANSIENT VmxTransient;
11468	RT_ZERO(VmxTransient);
11469	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
11470	VmxTransient.fIsNestedGuest = true;
11471
11472	/* Paranoia. */
11473	Assert(VmxTransient.pVmcsInfo == &pVCpu->hmr0.s.vmx.VmcsInfoNstGst);
11474
11475	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
11476	for (;;)
11477	{
11478	Assert(!HMR0SuspendPending());
11479	HMVMX_ASSERT_CPU_SAFE(pVCpu);
11480	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
11481
11482	/*
11483	* Preparatory work for running guest code, this may force us to
11484	* return to ring-3.
11485	*
11486	* Warning! This bugger disables interrupts on VINF_SUCCESS!
11487	*/
11488	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, false /* fStepping */);
11489	if (rcStrict != VINF_SUCCESS)
11490	break;
11491
11492	/* Interrupts are disabled at this point! */
11493	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
11494	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
11495	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
11496	/* Interrupts are re-enabled at this point! */
11497
11498	/*
11499	* Check for errors with running the VM (VMLAUNCH/VMRESUME).
11500	*/
11501	if (RT_SUCCESS(rcRun))
11502	{ /* very likely */ }
11503	else
11504	{
11505	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
11506	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
11507	return rcRun;
11508	}
11509
11510	/*
11511	* Profile the VM-exit.
11512	*/
11513	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
11514	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
11515	STAM_COUNTER_INC(&pVCpu->hm.s.StatNestedExitAll);
11516	STAM_COUNTER_INC(&pVCpu->hm.s.paStatNestedExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
11517	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
11518	HMVMX_START_EXIT_DISPATCH_PROF();
11519
11520	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
11521
11522	/*
11523	* Handle the VM-exit.
11524	*/
11525	rcStrict = hmR0VmxHandleExitNested(pVCpu, &VmxTransient);
11526	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
11527	if (rcStrict == VINF_SUCCESS)
11528	{
11529	if (!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
11530	{
11531	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
11532	rcStrict = VINF_VMX_VMEXIT;
11533	}
11534	else
11535	{
11536	if (++(*pcLoops) <= cMaxResumeLoops)
11537	continue;
11538	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
11539	rcStrict = VINF_EM_RAW_INTERRUPT;
11540	}
11541	}
11542	else
11543	Assert(rcStrict != VINF_VMX_VMEXIT);
11544	break;
11545	}
11546
11547	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
11548	return rcStrict;
11549	}
11550	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
11551
11552
11553	/** @name Execution loop for single stepping, DBGF events and expensive Dtrace
11554	* probes.
11555	*
11556	* The following few functions and associated structure contains the bloat
11557	* necessary for providing detailed debug events and dtrace probes as well as
11558	* reliable host side single stepping. This works on the principle of
11559	* "subclassing" the normal execution loop and workers. We replace the loop
11560	* method completely and override selected helpers to add necessary adjustments
11561	* to their core operation.
11562	*
11563	* The goal is to keep the "parent" code lean and mean, so as not to sacrifice
11564	* any performance for debug and analysis features.
11565	*
11566	* @{
11567	*/
11568
11569	/**
11570	* Transient per-VCPU debug state of VMCS and related info. we save/restore in
11571	* the debug run loop.
11572	*/
11573	typedef struct VMXRUNDBGSTATE
11574	{
11575	/** The RIP we started executing at. This is for detecting that we stepped. */
11576	uint64_t uRipStart;
11577	/** The CS we started executing with. */
11578	uint16_t uCsStart;
11579
11580	/** Whether we've actually modified the 1st execution control field. */
11581	bool fModifiedProcCtls : 1;
11582	/** Whether we've actually modified the 2nd execution control field. */
11583	bool fModifiedProcCtls2 : 1;
11584	/** Whether we've actually modified the exception bitmap. */
11585	bool fModifiedXcptBitmap : 1;
11586
11587	/** We desire the modified the CR0 mask to be cleared. */
11588	bool fClearCr0Mask : 1;
11589	/** We desire the modified the CR4 mask to be cleared. */
11590	bool fClearCr4Mask : 1;
11591	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC. */
11592	uint32_t fCpe1Extra;
11593	/** Stuff we do not want in VMX_VMCS32_CTRL_PROC_EXEC. */
11594	uint32_t fCpe1Unwanted;
11595	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC2. */
11596	uint32_t fCpe2Extra;
11597	/** Extra stuff we need in VMX_VMCS32_CTRL_EXCEPTION_BITMAP. */
11598	uint32_t bmXcptExtra;
11599	/** The sequence number of the Dtrace provider settings the state was
11600	* configured against. */
11601	uint32_t uDtraceSettingsSeqNo;
11602	/** VM-exits to check (one bit per VM-exit). */
11603	uint32_t bmExitsToCheck[3];
11604
11605	/** The initial VMX_VMCS32_CTRL_PROC_EXEC value (helps with restore). */
11606	uint32_t fProcCtlsInitial;
11607	/** The initial VMX_VMCS32_CTRL_PROC_EXEC2 value (helps with restore). */
11608	uint32_t fProcCtls2Initial;
11609	/** The initial VMX_VMCS32_CTRL_EXCEPTION_BITMAP value (helps with restore). */
11610	uint32_t bmXcptInitial;
11611	} VMXRUNDBGSTATE;
11612	AssertCompileMemberSize(VMXRUNDBGSTATE, bmExitsToCheck, (VMX_EXIT_MAX + 1 + 31) / 32 * 4);
11613	typedef VMXRUNDBGSTATE *PVMXRUNDBGSTATE;
11614
11615
11616	/**
11617	* Initializes the VMXRUNDBGSTATE structure.
11618	*
11619	* @param pVCpu The cross context virtual CPU structure of the
11620	* calling EMT.
11621	* @param pVmxTransient The VMX-transient structure.
11622	* @param pDbgState The debug state to initialize.
11623	*/
11624	static void hmR0VmxRunDebugStateInit(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11625	{
11626	pDbgState->uRipStart = pVCpu->cpum.GstCtx.rip;
11627	pDbgState->uCsStart = pVCpu->cpum.GstCtx.cs.Sel;
11628
11629	pDbgState->fModifiedProcCtls = false;
11630	pDbgState->fModifiedProcCtls2 = false;
11631	pDbgState->fModifiedXcptBitmap = false;
11632	pDbgState->fClearCr0Mask = false;
11633	pDbgState->fClearCr4Mask = false;
11634	pDbgState->fCpe1Extra = 0;
11635	pDbgState->fCpe1Unwanted = 0;
11636	pDbgState->fCpe2Extra = 0;
11637	pDbgState->bmXcptExtra = 0;
11638	pDbgState->fProcCtlsInitial = pVmxTransient->pVmcsInfo->u32ProcCtls;
11639	pDbgState->fProcCtls2Initial = pVmxTransient->pVmcsInfo->u32ProcCtls2;
11640	pDbgState->bmXcptInitial = pVmxTransient->pVmcsInfo->u32XcptBitmap;
11641	}
11642
11643
11644	/**
11645	* Updates the VMSC fields with changes requested by @a pDbgState.
11646	*
11647	* This is performed after hmR0VmxPreRunGuestDebugStateUpdate as well
11648	* immediately before executing guest code, i.e. when interrupts are disabled.
11649	* We don't check status codes here as we cannot easily assert or return in the
11650	* latter case.
11651	*
11652	* @param pVCpu The cross context virtual CPU structure.
11653	* @param pVmxTransient The VMX-transient structure.
11654	* @param pDbgState The debug state.
11655	*/
11656	static void hmR0VmxPreRunGuestDebugStateApply(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11657	{
11658	/*
11659	* Ensure desired flags in VMCS control fields are set.
11660	* (Ignoring write failure here, as we're committed and it's just debug extras.)
11661	*
11662	* Note! We load the shadow CR0 & CR4 bits when we flag the clearing, so
11663	* there should be no stale data in pCtx at this point.
11664	*/
11665	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11666	if ( (pVmcsInfo->u32ProcCtls & pDbgState->fCpe1Extra) != pDbgState->fCpe1Extra
11667	\|\| (pVmcsInfo->u32ProcCtls & pDbgState->fCpe1Unwanted))
11668	{
11669	pVmcsInfo->u32ProcCtls \|= pDbgState->fCpe1Extra;
11670	pVmcsInfo->u32ProcCtls &= ~pDbgState->fCpe1Unwanted;
11671	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
11672	Log6Func(("VMX_VMCS32_CTRL_PROC_EXEC: %#RX32\n", pVmcsInfo->u32ProcCtls));
11673	pDbgState->fModifiedProcCtls = true;
11674	}
11675
11676	if ((pVmcsInfo->u32ProcCtls2 & pDbgState->fCpe2Extra) != pDbgState->fCpe2Extra)
11677	{
11678	pVmcsInfo->u32ProcCtls2 \|= pDbgState->fCpe2Extra;
11679	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pVmcsInfo->u32ProcCtls2);
11680	Log6Func(("VMX_VMCS32_CTRL_PROC_EXEC2: %#RX32\n", pVmcsInfo->u32ProcCtls2));
11681	pDbgState->fModifiedProcCtls2 = true;
11682	}
11683
11684	if ((pVmcsInfo->u32XcptBitmap & pDbgState->bmXcptExtra) != pDbgState->bmXcptExtra)
11685	{
11686	pVmcsInfo->u32XcptBitmap \|= pDbgState->bmXcptExtra;
11687	VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVmcsInfo->u32XcptBitmap);
11688	Log6Func(("VMX_VMCS32_CTRL_EXCEPTION_BITMAP: %#RX32\n", pVmcsInfo->u32XcptBitmap));
11689	pDbgState->fModifiedXcptBitmap = true;
11690	}
11691
11692	if (pDbgState->fClearCr0Mask && pVmcsInfo->u64Cr0Mask != 0)
11693	{
11694	pVmcsInfo->u64Cr0Mask = 0;
11695	VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, 0);
11696	Log6Func(("VMX_VMCS_CTRL_CR0_MASK: 0\n"));
11697	}
11698
11699	if (pDbgState->fClearCr4Mask && pVmcsInfo->u64Cr4Mask != 0)
11700	{
11701	pVmcsInfo->u64Cr4Mask = 0;
11702	VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, 0);
11703	Log6Func(("VMX_VMCS_CTRL_CR4_MASK: 0\n"));
11704	}
11705
11706	NOREF(pVCpu);
11707	}
11708
11709
11710	/**
11711	* Restores VMCS fields that were changed by hmR0VmxPreRunGuestDebugStateApply for
11712	* re-entry next time around.
11713	*
11714	* @returns Strict VBox status code (i.e. informational status codes too).
11715	* @param pVCpu The cross context virtual CPU structure.
11716	* @param pVmxTransient The VMX-transient structure.
11717	* @param pDbgState The debug state.
11718	* @param rcStrict The return code from executing the guest using single
11719	* stepping.
11720	*/
11721	static VBOXSTRICTRC hmR0VmxRunDebugStateRevert(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState,
11722	VBOXSTRICTRC rcStrict)
11723	{
11724	/*
11725	* Restore VM-exit control settings as we may not reenter this function the
11726	* next time around.
11727	*/
11728	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11729
11730	/* We reload the initial value, trigger what we can of recalculations the
11731	next time around. From the looks of things, that's all that's required atm. */
11732	if (pDbgState->fModifiedProcCtls)
11733	{
11734	if (!(pDbgState->fProcCtlsInitial & VMX_PROC_CTLS_MOV_DR_EXIT) && CPUMIsHyperDebugStateActive(pVCpu))
11735	pDbgState->fProcCtlsInitial \|= VMX_PROC_CTLS_MOV_DR_EXIT; /* Avoid assertion in hmR0VmxLeave */
11736	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pDbgState->fProcCtlsInitial);
11737	AssertRC(rc2);
11738	pVmcsInfo->u32ProcCtls = pDbgState->fProcCtlsInitial;
11739	}
11740
11741	/* We're currently the only ones messing with this one, so just restore the
11742	cached value and reload the field. */
11743	if ( pDbgState->fModifiedProcCtls2
11744	&& pVmcsInfo->u32ProcCtls2 != pDbgState->fProcCtls2Initial)
11745	{
11746	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pDbgState->fProcCtls2Initial);
11747	AssertRC(rc2);
11748	pVmcsInfo->u32ProcCtls2 = pDbgState->fProcCtls2Initial;
11749	}
11750
11751	/* If we've modified the exception bitmap, we restore it and trigger
11752	reloading and partial recalculation the next time around. */
11753	if (pDbgState->fModifiedXcptBitmap)
11754	pVmcsInfo->u32XcptBitmap = pDbgState->bmXcptInitial;
11755
11756	return rcStrict;
11757	}
11758
11759
11760	/**
11761	* Configures VM-exit controls for current DBGF and DTrace settings.
11762	*
11763	* This updates @a pDbgState and the VMCS execution control fields to reflect
11764	* the necessary VM-exits demanded by DBGF and DTrace.
11765	*
11766	* @param pVCpu The cross context virtual CPU structure.
11767	* @param pVmxTransient The VMX-transient structure. May update
11768	* fUpdatedTscOffsettingAndPreemptTimer.
11769	* @param pDbgState The debug state.
11770	*/
11771	static void hmR0VmxPreRunGuestDebugStateUpdate(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11772	{
11773	/*
11774	* Take down the dtrace serial number so we can spot changes.
11775	*/
11776	pDbgState->uDtraceSettingsSeqNo = VBOXVMM_GET_SETTINGS_SEQ_NO();
11777	ASMCompilerBarrier();
11778
11779	/*
11780	* We'll rebuild most of the middle block of data members (holding the
11781	* current settings) as we go along here, so start by clearing it all.
11782	*/
11783	pDbgState->bmXcptExtra = 0;
11784	pDbgState->fCpe1Extra = 0;
11785	pDbgState->fCpe1Unwanted = 0;
11786	pDbgState->fCpe2Extra = 0;
11787	for (unsigned i = 0; i < RT_ELEMENTS(pDbgState->bmExitsToCheck); i++)
11788	pDbgState->bmExitsToCheck[i] = 0;
11789
11790	/*
11791	* Software interrupts (INT XXh) - no idea how to trigger these...
11792	*/
11793	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
11794	if ( DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_INTERRUPT_SOFTWARE)
11795	\|\| VBOXVMM_INT_SOFTWARE_ENABLED())
11796	{
11797	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
11798	}
11799
11800	/*
11801	* INT3 breakpoints - triggered by #BP exceptions.
11802	*/
11803	if (pVM->dbgf.ro.cEnabledInt3Breakpoints > 0)
11804	pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
11805
11806	/*
11807	* Exception bitmap and XCPT events+probes.
11808	*/
11809	for (int iXcpt = 0; iXcpt < (DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST + 1); iXcpt++)
11810	if (DBGF_IS_EVENT_ENABLED(pVM, (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + iXcpt)))
11811	pDbgState->bmXcptExtra \|= RT_BIT_32(iXcpt);
11812
11813	if (VBOXVMM_XCPT_DE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DE);
11814	if (VBOXVMM_XCPT_DB_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DB);
11815	if (VBOXVMM_XCPT_BP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
11816	if (VBOXVMM_XCPT_OF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_OF);
11817	if (VBOXVMM_XCPT_BR_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BR);
11818	if (VBOXVMM_XCPT_UD_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_UD);
11819	if (VBOXVMM_XCPT_NM_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NM);
11820	if (VBOXVMM_XCPT_DF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DF);
11821	if (VBOXVMM_XCPT_TS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_TS);
11822	if (VBOXVMM_XCPT_NP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NP);
11823	if (VBOXVMM_XCPT_SS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SS);
11824	if (VBOXVMM_XCPT_GP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_GP);
11825	if (VBOXVMM_XCPT_PF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_PF);
11826	if (VBOXVMM_XCPT_MF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_MF);
11827	if (VBOXVMM_XCPT_AC_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_AC);
11828	if (VBOXVMM_XCPT_XF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_XF);
11829	if (VBOXVMM_XCPT_VE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_VE);
11830	if (VBOXVMM_XCPT_SX_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SX);
11831
11832	if (pDbgState->bmXcptExtra)
11833	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
11834
11835	/*
11836	* Process events and probes for VM-exits, making sure we get the wanted VM-exits.
11837	*
11838	* Note! This is the reverse of what hmR0VmxHandleExitDtraceEvents does.
11839	* So, when adding/changing/removing please don't forget to update it.
11840	*
11841	* Some of the macros are picking up local variables to save horizontal space,
11842	* (being able to see it in a table is the lesser evil here).
11843	*/
11844	#define IS_EITHER_ENABLED(a_pVM, a_EventSubName) \
11845	( DBGF_IS_EVENT_ENABLED(a_pVM, RT_CONCAT(DBGFEVENT_, a_EventSubName)) \
11846	\|\| RT_CONCAT3(VBOXVMM_, a_EventSubName, _ENABLED)() )
11847	#define SET_ONLY_XBM_IF_EITHER_EN(a_EventSubName, a_uExit) \
11848	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11849	{ AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11850	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11851	} else do { } while (0)
11852	#define SET_CPE1_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec) \
11853	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11854	{ \
11855	(pDbgState)->fCpe1Extra \|= (a_fCtrlProcExec); \
11856	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11857	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11858	} else do { } while (0)
11859	#define SET_CPEU_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fUnwantedCtrlProcExec) \
11860	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11861	{ \
11862	(pDbgState)->fCpe1Unwanted \|= (a_fUnwantedCtrlProcExec); \
11863	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11864	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11865	} else do { } while (0)
11866	#define SET_CPE2_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec2) \
11867	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11868	{ \
11869	(pDbgState)->fCpe2Extra \|= (a_fCtrlProcExec2); \
11870	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11871	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11872	} else do { } while (0)
11873
11874	SET_ONLY_XBM_IF_EITHER_EN(EXIT_TASK_SWITCH, VMX_EXIT_TASK_SWITCH); /* unconditional */
11875	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_VIOLATION, VMX_EXIT_EPT_VIOLATION); /* unconditional */
11876	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_MISCONFIG, VMX_EXIT_EPT_MISCONFIG); /* unconditional (unless #VE) */
11877	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_ACCESS, VMX_EXIT_APIC_ACCESS); /* feature dependent, nothing to enable here */
11878	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_WRITE, VMX_EXIT_APIC_WRITE); /* feature dependent, nothing to enable here */
11879
11880	SET_ONLY_XBM_IF_EITHER_EN(INSTR_CPUID, VMX_EXIT_CPUID); /* unconditional */
11881	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CPUID, VMX_EXIT_CPUID);
11882	SET_ONLY_XBM_IF_EITHER_EN(INSTR_GETSEC, VMX_EXIT_GETSEC); /* unconditional */
11883	SET_ONLY_XBM_IF_EITHER_EN( EXIT_GETSEC, VMX_EXIT_GETSEC);
11884	SET_CPE1_XBM_IF_EITHER_EN(INSTR_HALT, VMX_EXIT_HLT, VMX_PROC_CTLS_HLT_EXIT); /* paranoia */
11885	SET_ONLY_XBM_IF_EITHER_EN( EXIT_HALT, VMX_EXIT_HLT);
11886	SET_ONLY_XBM_IF_EITHER_EN(INSTR_INVD, VMX_EXIT_INVD); /* unconditional */
11887	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVD, VMX_EXIT_INVD);
11888	SET_CPE1_XBM_IF_EITHER_EN(INSTR_INVLPG, VMX_EXIT_INVLPG, VMX_PROC_CTLS_INVLPG_EXIT);
11889	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVLPG, VMX_EXIT_INVLPG);
11890	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDPMC, VMX_EXIT_RDPMC, VMX_PROC_CTLS_RDPMC_EXIT);
11891	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDPMC, VMX_EXIT_RDPMC);
11892	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSC, VMX_EXIT_RDTSC, VMX_PROC_CTLS_RDTSC_EXIT);
11893	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSC, VMX_EXIT_RDTSC);
11894	SET_ONLY_XBM_IF_EITHER_EN(INSTR_RSM, VMX_EXIT_RSM); /* unconditional */
11895	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RSM, VMX_EXIT_RSM);
11896	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMM_CALL, VMX_EXIT_VMCALL); /* unconditional */
11897	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMM_CALL, VMX_EXIT_VMCALL);
11898	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMCLEAR, VMX_EXIT_VMCLEAR); /* unconditional */
11899	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMCLEAR, VMX_EXIT_VMCLEAR);
11900	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH); /* unconditional */
11901	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH);
11902	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRLD, VMX_EXIT_VMPTRLD); /* unconditional */
11903	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRLD, VMX_EXIT_VMPTRLD);
11904	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRST, VMX_EXIT_VMPTRST); /* unconditional */
11905	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRST, VMX_EXIT_VMPTRST);
11906	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMREAD, VMX_EXIT_VMREAD); /* unconditional */
11907	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMREAD, VMX_EXIT_VMREAD);
11908	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMRESUME, VMX_EXIT_VMRESUME); /* unconditional */
11909	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMRESUME, VMX_EXIT_VMRESUME);
11910	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMWRITE, VMX_EXIT_VMWRITE); /* unconditional */
11911	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMWRITE, VMX_EXIT_VMWRITE);
11912	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXOFF, VMX_EXIT_VMXOFF); /* unconditional */
11913	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXOFF, VMX_EXIT_VMXOFF);
11914	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXON, VMX_EXIT_VMXON); /* unconditional */
11915	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXON, VMX_EXIT_VMXON);
11916
11917	if ( IS_EITHER_ENABLED(pVM, INSTR_CRX_READ)
11918	\|\| IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
11919	{
11920	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_CR4
11921	\| CPUMCTX_EXTRN_APIC_TPR);
11922	AssertRC(rc);
11923
11924	#if 0 /** @todo fix me */
11925	pDbgState->fClearCr0Mask = true;
11926	pDbgState->fClearCr4Mask = true;
11927	#endif
11928	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_READ))
11929	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_CR3_STORE_EXIT \| VMX_PROC_CTLS_CR8_STORE_EXIT;
11930	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
11931	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_CR3_LOAD_EXIT \| VMX_PROC_CTLS_CR8_LOAD_EXIT;
11932	pDbgState->fCpe1Unwanted \|= VMX_PROC_CTLS_USE_TPR_SHADOW; /* risky? */
11933	/* Note! We currently don't use VMX_VMCS32_CTRL_CR3_TARGET_COUNT. It would
11934	require clearing here and in the loop if we start using it. */
11935	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_CRX);
11936	}
11937	else
11938	{
11939	if (pDbgState->fClearCr0Mask)
11940	{
11941	pDbgState->fClearCr0Mask = false;
11942	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR0);
11943	}
11944	if (pDbgState->fClearCr4Mask)
11945	{
11946	pDbgState->fClearCr4Mask = false;
11947	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR4);
11948	}
11949	}
11950	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_READ, VMX_EXIT_MOV_CRX);
11951	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_WRITE, VMX_EXIT_MOV_CRX);
11952
11953	if ( IS_EITHER_ENABLED(pVM, INSTR_DRX_READ)
11954	\|\| IS_EITHER_ENABLED(pVM, INSTR_DRX_WRITE))
11955	{
11956	/** @todo later, need to fix handler as it assumes this won't usually happen. */
11957	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_DRX);
11958	}
11959	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_READ, VMX_EXIT_MOV_DRX);
11960	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_WRITE, VMX_EXIT_MOV_DRX);
11961
11962	SET_CPEU_XBM_IF_EITHER_EN(INSTR_RDMSR, VMX_EXIT_RDMSR, VMX_PROC_CTLS_USE_MSR_BITMAPS); /* risky clearing this? */
11963	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDMSR, VMX_EXIT_RDMSR);
11964	SET_CPEU_XBM_IF_EITHER_EN(INSTR_WRMSR, VMX_EXIT_WRMSR, VMX_PROC_CTLS_USE_MSR_BITMAPS);
11965	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WRMSR, VMX_EXIT_WRMSR);
11966	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MWAIT, VMX_EXIT_MWAIT, VMX_PROC_CTLS_MWAIT_EXIT); /* paranoia */
11967	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MWAIT, VMX_EXIT_MWAIT);
11968	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MONITOR, VMX_EXIT_MONITOR, VMX_PROC_CTLS_MONITOR_EXIT); /* paranoia */
11969	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MONITOR, VMX_EXIT_MONITOR);
11970	#if 0 /** @todo too slow, fix handler. */
11971	SET_CPE1_XBM_IF_EITHER_EN(INSTR_PAUSE, VMX_EXIT_PAUSE, VMX_PROC_CTLS_PAUSE_EXIT);
11972	#endif
11973	SET_ONLY_XBM_IF_EITHER_EN( EXIT_PAUSE, VMX_EXIT_PAUSE);
11974
11975	if ( IS_EITHER_ENABLED(pVM, INSTR_SGDT)
11976	\|\| IS_EITHER_ENABLED(pVM, INSTR_SIDT)
11977	\|\| IS_EITHER_ENABLED(pVM, INSTR_LGDT)
11978	\|\| IS_EITHER_ENABLED(pVM, INSTR_LIDT))
11979	{
11980	pDbgState->fCpe2Extra \|= VMX_PROC_CTLS2_DESC_TABLE_EXIT;
11981	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_GDTR_IDTR_ACCESS);
11982	}
11983	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SGDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11984	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SIDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11985	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LGDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11986	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LIDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11987
11988	if ( IS_EITHER_ENABLED(pVM, INSTR_SLDT)
11989	\|\| IS_EITHER_ENABLED(pVM, INSTR_STR)
11990	\|\| IS_EITHER_ENABLED(pVM, INSTR_LLDT)
11991	\|\| IS_EITHER_ENABLED(pVM, INSTR_LTR))
11992	{
11993	pDbgState->fCpe2Extra \|= VMX_PROC_CTLS2_DESC_TABLE_EXIT;
11994	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_LDTR_TR_ACCESS);
11995	}
11996	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SLDT, VMX_EXIT_LDTR_TR_ACCESS);
11997	SET_ONLY_XBM_IF_EITHER_EN( EXIT_STR, VMX_EXIT_LDTR_TR_ACCESS);
11998	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LLDT, VMX_EXIT_LDTR_TR_ACCESS);
11999	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LTR, VMX_EXIT_LDTR_TR_ACCESS);
12000
12001	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVEPT, VMX_EXIT_INVEPT); /* unconditional */
12002	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVEPT, VMX_EXIT_INVEPT);
12003	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSCP, VMX_EXIT_RDTSCP, VMX_PROC_CTLS_RDTSC_EXIT);
12004	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSCP, VMX_EXIT_RDTSCP);
12005	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVVPID, VMX_EXIT_INVVPID); /* unconditional */
12006	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVVPID, VMX_EXIT_INVVPID);
12007	SET_CPE2_XBM_IF_EITHER_EN(INSTR_WBINVD, VMX_EXIT_WBINVD, VMX_PROC_CTLS2_WBINVD_EXIT);
12008	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WBINVD, VMX_EXIT_WBINVD);
12009	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSETBV, VMX_EXIT_XSETBV); /* unconditional */
12010	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XSETBV, VMX_EXIT_XSETBV);
12011	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDRAND, VMX_EXIT_RDRAND, VMX_PROC_CTLS2_RDRAND_EXIT);
12012	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDRAND, VMX_EXIT_RDRAND);
12013	SET_CPE1_XBM_IF_EITHER_EN(INSTR_VMX_INVPCID, VMX_EXIT_INVPCID, VMX_PROC_CTLS_INVLPG_EXIT);
12014	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVPCID, VMX_EXIT_INVPCID);
12015	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMFUNC, VMX_EXIT_VMFUNC); /* unconditional for the current setup */
12016	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMFUNC, VMX_EXIT_VMFUNC);
12017	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDSEED, VMX_EXIT_RDSEED, VMX_PROC_CTLS2_RDSEED_EXIT);
12018	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDSEED, VMX_EXIT_RDSEED);
12019	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSAVES, VMX_EXIT_XSAVES); /* unconditional (enabled by host, guest cfg) */
12020	SET_ONLY_XBM_IF_EITHER_EN(EXIT_XSAVES, VMX_EXIT_XSAVES);
12021	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XRSTORS, VMX_EXIT_XRSTORS); /* unconditional (enabled by host, guest cfg) */
12022	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XRSTORS, VMX_EXIT_XRSTORS);
12023
12024	#undef IS_EITHER_ENABLED
12025	#undef SET_ONLY_XBM_IF_EITHER_EN
12026	#undef SET_CPE1_XBM_IF_EITHER_EN
12027	#undef SET_CPEU_XBM_IF_EITHER_EN
12028	#undef SET_CPE2_XBM_IF_EITHER_EN
12029
12030	/*
12031	* Sanitize the control stuff.
12032	*/
12033	pDbgState->fCpe2Extra &= g_HmMsrs.u.vmx.ProcCtls2.n.allowed1;
12034	if (pDbgState->fCpe2Extra)
12035	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_USE_SECONDARY_CTLS;
12036	pDbgState->fCpe1Extra &= g_HmMsrs.u.vmx.ProcCtls.n.allowed1;
12037	pDbgState->fCpe1Unwanted &= ~g_HmMsrs.u.vmx.ProcCtls.n.allowed0;
12038	if (pVCpu->hmr0.s.fDebugWantRdTscExit != RT_BOOL(pDbgState->fCpe1Extra & VMX_PROC_CTLS_RDTSC_EXIT))
12039	{
12040	pVCpu->hmr0.s.fDebugWantRdTscExit ^= true;
12041	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
12042	}
12043
12044	Log6(("HM: debug state: cpe1=%#RX32 cpeu=%#RX32 cpe2=%#RX32%s%s\n",
12045	pDbgState->fCpe1Extra, pDbgState->fCpe1Unwanted, pDbgState->fCpe2Extra,
12046	pDbgState->fClearCr0Mask ? " clr-cr0" : "",
12047	pDbgState->fClearCr4Mask ? " clr-cr4" : ""));
12048	}
12049
12050
12051	/**
12052	* Fires off DBGF events and dtrace probes for a VM-exit, when it's
12053	* appropriate.
12054	*
12055	* The caller has checked the VM-exit against the
12056	* VMXRUNDBGSTATE::bmExitsToCheck bitmap. The caller has checked for NMIs
12057	* already, so we don't have to do that either.
12058	*
12059	* @returns Strict VBox status code (i.e. informational status codes too).
12060	* @param pVCpu The cross context virtual CPU structure.
12061	* @param pVmxTransient The VMX-transient structure.
12062	* @param uExitReason The VM-exit reason.
12063	*
12064	* @remarks The name of this function is displayed by dtrace, so keep it short
12065	* and to the point. No longer than 33 chars long, please.
12066	*/
12067	static VBOXSTRICTRC hmR0VmxHandleExitDtraceEvents(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uExitReason)
12068	{
12069	/*
12070	* Translate the event into a DBGF event (enmEvent + uEventArg) and at the
12071	* same time check whether any corresponding Dtrace event is enabled (fDtrace).
12072	*
12073	* Note! This is the reverse operation of what hmR0VmxPreRunGuestDebugStateUpdate
12074	* does. Must add/change/remove both places. Same ordering, please.
12075	*
12076	* Added/removed events must also be reflected in the next section
12077	* where we dispatch dtrace events.
12078	*/
12079	bool fDtrace1 = false;
12080	bool fDtrace2 = false;
12081	DBGFEVENTTYPE enmEvent1 = DBGFEVENT_END;
12082	DBGFEVENTTYPE enmEvent2 = DBGFEVENT_END;
12083	uint32_t uEventArg = 0;
12084	#define SET_EXIT(a_EventSubName) \
12085	do { \
12086	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
12087	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
12088	} while (0)
12089	#define SET_BOTH(a_EventSubName) \
12090	do { \
12091	enmEvent1 = RT_CONCAT(DBGFEVENT_INSTR_, a_EventSubName); \
12092	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
12093	fDtrace1 = RT_CONCAT3(VBOXVMM_INSTR_, a_EventSubName, _ENABLED)(); \
12094	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
12095	} while (0)
12096	switch (uExitReason)
12097	{
12098	case VMX_EXIT_MTF:
12099	return hmR0VmxExitMtf(pVCpu, pVmxTransient);
12100
12101	case VMX_EXIT_XCPT_OR_NMI:
12102	{
12103	uint8_t const idxVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
12104	switch (VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo))
12105	{
12106	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
12107	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
12108	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
12109	if (idxVector <= (unsigned)(DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST))
12110	{
12111	if (VMX_EXIT_INT_INFO_IS_ERROR_CODE_VALID(pVmxTransient->uExitIntInfo))
12112	{
12113	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
12114	uEventArg = pVmxTransient->uExitIntErrorCode;
12115	}
12116	enmEvent1 = (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + idxVector);
12117	switch (enmEvent1)
12118	{
12119	case DBGFEVENT_XCPT_DE: fDtrace1 = VBOXVMM_XCPT_DE_ENABLED(); break;
12120	case DBGFEVENT_XCPT_DB: fDtrace1 = VBOXVMM_XCPT_DB_ENABLED(); break;
12121	case DBGFEVENT_XCPT_BP: fDtrace1 = VBOXVMM_XCPT_BP_ENABLED(); break;
12122	case DBGFEVENT_XCPT_OF: fDtrace1 = VBOXVMM_XCPT_OF_ENABLED(); break;
12123	case DBGFEVENT_XCPT_BR: fDtrace1 = VBOXVMM_XCPT_BR_ENABLED(); break;
12124	case DBGFEVENT_XCPT_UD: fDtrace1 = VBOXVMM_XCPT_UD_ENABLED(); break;
12125	case DBGFEVENT_XCPT_NM: fDtrace1 = VBOXVMM_XCPT_NM_ENABLED(); break;
12126	case DBGFEVENT_XCPT_DF: fDtrace1 = VBOXVMM_XCPT_DF_ENABLED(); break;
12127	case DBGFEVENT_XCPT_TS: fDtrace1 = VBOXVMM_XCPT_TS_ENABLED(); break;
12128	case DBGFEVENT_XCPT_NP: fDtrace1 = VBOXVMM_XCPT_NP_ENABLED(); break;
12129	case DBGFEVENT_XCPT_SS: fDtrace1 = VBOXVMM_XCPT_SS_ENABLED(); break;
12130	case DBGFEVENT_XCPT_GP: fDtrace1 = VBOXVMM_XCPT_GP_ENABLED(); break;
12131	case DBGFEVENT_XCPT_PF: fDtrace1 = VBOXVMM_XCPT_PF_ENABLED(); break;
12132	case DBGFEVENT_XCPT_MF: fDtrace1 = VBOXVMM_XCPT_MF_ENABLED(); break;
12133	case DBGFEVENT_XCPT_AC: fDtrace1 = VBOXVMM_XCPT_AC_ENABLED(); break;
12134	case DBGFEVENT_XCPT_XF: fDtrace1 = VBOXVMM_XCPT_XF_ENABLED(); break;
12135	case DBGFEVENT_XCPT_VE: fDtrace1 = VBOXVMM_XCPT_VE_ENABLED(); break;
12136	case DBGFEVENT_XCPT_SX: fDtrace1 = VBOXVMM_XCPT_SX_ENABLED(); break;
12137	default: break;
12138	}
12139	}
12140	else
12141	AssertFailed();
12142	break;
12143
12144	case VMX_EXIT_INT_INFO_TYPE_SW_INT:
12145	uEventArg = idxVector;
12146	enmEvent1 = DBGFEVENT_INTERRUPT_SOFTWARE;
12147	fDtrace1 = VBOXVMM_INT_SOFTWARE_ENABLED();
12148	break;
12149	}
12150	break;
12151	}
12152
12153	case VMX_EXIT_TRIPLE_FAULT:
12154	enmEvent1 = DBGFEVENT_TRIPLE_FAULT;
12155	//fDtrace1 = VBOXVMM_EXIT_TRIPLE_FAULT_ENABLED();
12156	break;
12157	case VMX_EXIT_TASK_SWITCH: SET_EXIT(TASK_SWITCH); break;
12158	case VMX_EXIT_EPT_VIOLATION: SET_EXIT(VMX_EPT_VIOLATION); break;
12159	case VMX_EXIT_EPT_MISCONFIG: SET_EXIT(VMX_EPT_MISCONFIG); break;
12160	case VMX_EXIT_APIC_ACCESS: SET_EXIT(VMX_VAPIC_ACCESS); break;
12161	case VMX_EXIT_APIC_WRITE: SET_EXIT(VMX_VAPIC_WRITE); break;
12162
12163	/* Instruction specific VM-exits: */
12164	case VMX_EXIT_CPUID: SET_BOTH(CPUID); break;
12165	case VMX_EXIT_GETSEC: SET_BOTH(GETSEC); break;
12166	case VMX_EXIT_HLT: SET_BOTH(HALT); break;
12167	case VMX_EXIT_INVD: SET_BOTH(INVD); break;
12168	case VMX_EXIT_INVLPG: SET_BOTH(INVLPG); break;
12169	case VMX_EXIT_RDPMC: SET_BOTH(RDPMC); break;
12170	case VMX_EXIT_RDTSC: SET_BOTH(RDTSC); break;
12171	case VMX_EXIT_RSM: SET_BOTH(RSM); break;
12172	case VMX_EXIT_VMCALL: SET_BOTH(VMM_CALL); break;
12173	case VMX_EXIT_VMCLEAR: SET_BOTH(VMX_VMCLEAR); break;
12174	case VMX_EXIT_VMLAUNCH: SET_BOTH(VMX_VMLAUNCH); break;
12175	case VMX_EXIT_VMPTRLD: SET_BOTH(VMX_VMPTRLD); break;
12176	case VMX_EXIT_VMPTRST: SET_BOTH(VMX_VMPTRST); break;
12177	case VMX_EXIT_VMREAD: SET_BOTH(VMX_VMREAD); break;
12178	case VMX_EXIT_VMRESUME: SET_BOTH(VMX_VMRESUME); break;
12179	case VMX_EXIT_VMWRITE: SET_BOTH(VMX_VMWRITE); break;
12180	case VMX_EXIT_VMXOFF: SET_BOTH(VMX_VMXOFF); break;
12181	case VMX_EXIT_VMXON: SET_BOTH(VMX_VMXON); break;
12182	case VMX_EXIT_MOV_CRX:
12183	hmR0VmxReadExitQualVmcs(pVmxTransient);
12184	if (VMX_EXIT_QUAL_CRX_ACCESS(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_CRX_ACCESS_READ)
12185	SET_BOTH(CRX_READ);
12186	else
12187	SET_BOTH(CRX_WRITE);
12188	uEventArg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
12189	break;
12190	case VMX_EXIT_MOV_DRX:
12191	hmR0VmxReadExitQualVmcs(pVmxTransient);
12192	if ( VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual)
12193	== VMX_EXIT_QUAL_DRX_DIRECTION_READ)
12194	SET_BOTH(DRX_READ);
12195	else
12196	SET_BOTH(DRX_WRITE);
12197	uEventArg = VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual);
12198	break;
12199	case VMX_EXIT_RDMSR: SET_BOTH(RDMSR); break;
12200	case VMX_EXIT_WRMSR: SET_BOTH(WRMSR); break;
12201	case VMX_EXIT_MWAIT: SET_BOTH(MWAIT); break;
12202	case VMX_EXIT_MONITOR: SET_BOTH(MONITOR); break;
12203	case VMX_EXIT_PAUSE: SET_BOTH(PAUSE); break;
12204	case VMX_EXIT_GDTR_IDTR_ACCESS:
12205	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
12206	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_BF_XDTR_INSINFO_INSTR_ID))
12207	{
12208	case VMX_XDTR_INSINFO_II_SGDT: SET_BOTH(SGDT); break;
12209	case VMX_XDTR_INSINFO_II_SIDT: SET_BOTH(SIDT); break;
12210	case VMX_XDTR_INSINFO_II_LGDT: SET_BOTH(LGDT); break;
12211	case VMX_XDTR_INSINFO_II_LIDT: SET_BOTH(LIDT); break;
12212	}
12213	break;
12214
12215	case VMX_EXIT_LDTR_TR_ACCESS:
12216	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
12217	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_BF_YYTR_INSINFO_INSTR_ID))
12218	{
12219	case VMX_YYTR_INSINFO_II_SLDT: SET_BOTH(SLDT); break;
12220	case VMX_YYTR_INSINFO_II_STR: SET_BOTH(STR); break;
12221	case VMX_YYTR_INSINFO_II_LLDT: SET_BOTH(LLDT); break;
12222	case VMX_YYTR_INSINFO_II_LTR: SET_BOTH(LTR); break;
12223	}
12224	break;
12225
12226	case VMX_EXIT_INVEPT: SET_BOTH(VMX_INVEPT); break;
12227	case VMX_EXIT_RDTSCP: SET_BOTH(RDTSCP); break;
12228	case VMX_EXIT_INVVPID: SET_BOTH(VMX_INVVPID); break;
12229	case VMX_EXIT_WBINVD: SET_BOTH(WBINVD); break;
12230	case VMX_EXIT_XSETBV: SET_BOTH(XSETBV); break;
12231	case VMX_EXIT_RDRAND: SET_BOTH(RDRAND); break;
12232	case VMX_EXIT_INVPCID: SET_BOTH(VMX_INVPCID); break;
12233	case VMX_EXIT_VMFUNC: SET_BOTH(VMX_VMFUNC); break;
12234	case VMX_EXIT_RDSEED: SET_BOTH(RDSEED); break;
12235	case VMX_EXIT_XSAVES: SET_BOTH(XSAVES); break;
12236	case VMX_EXIT_XRSTORS: SET_BOTH(XRSTORS); break;
12237
12238	/* Events that aren't relevant at this point. */
12239	case VMX_EXIT_EXT_INT:
12240	case VMX_EXIT_INT_WINDOW:
12241	case VMX_EXIT_NMI_WINDOW:
12242	case VMX_EXIT_TPR_BELOW_THRESHOLD:
12243	case VMX_EXIT_PREEMPT_TIMER:
12244	case VMX_EXIT_IO_INSTR:
12245	break;
12246
12247	/* Errors and unexpected events. */
12248	case VMX_EXIT_INIT_SIGNAL:
12249	case VMX_EXIT_SIPI:
12250	case VMX_EXIT_IO_SMI:
12251	case VMX_EXIT_SMI:
12252	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
12253	case VMX_EXIT_ERR_MSR_LOAD:
12254	case VMX_EXIT_ERR_MACHINE_CHECK:
12255	case VMX_EXIT_PML_FULL:
12256	case VMX_EXIT_VIRTUALIZED_EOI:
12257	break;
12258
12259	default:
12260	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
12261	break;
12262	}
12263	#undef SET_BOTH
12264	#undef SET_EXIT
12265
12266	/*
12267	* Dtrace tracepoints go first. We do them here at once so we don't
12268	* have to copy the guest state saving and stuff a few dozen times.
12269	* Down side is that we've got to repeat the switch, though this time
12270	* we use enmEvent since the probes are a subset of what DBGF does.
12271	*/
12272	if (fDtrace1 \|\| fDtrace2)
12273	{
12274	hmR0VmxReadExitQualVmcs(pVmxTransient);
12275	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
12276	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
12277	switch (enmEvent1)
12278	{
12279	/** @todo consider which extra parameters would be helpful for each probe. */
12280	case DBGFEVENT_END: break;
12281	case DBGFEVENT_XCPT_DE: VBOXVMM_XCPT_DE(pVCpu, pCtx); break;
12282	case DBGFEVENT_XCPT_DB: VBOXVMM_XCPT_DB(pVCpu, pCtx, pCtx->dr[6]); break;
12283	case DBGFEVENT_XCPT_BP: VBOXVMM_XCPT_BP(pVCpu, pCtx); break;
12284	case DBGFEVENT_XCPT_OF: VBOXVMM_XCPT_OF(pVCpu, pCtx); break;
12285	case DBGFEVENT_XCPT_BR: VBOXVMM_XCPT_BR(pVCpu, pCtx); break;
12286	case DBGFEVENT_XCPT_UD: VBOXVMM_XCPT_UD(pVCpu, pCtx); break;
12287	case DBGFEVENT_XCPT_NM: VBOXVMM_XCPT_NM(pVCpu, pCtx); break;
12288	case DBGFEVENT_XCPT_DF: VBOXVMM_XCPT_DF(pVCpu, pCtx); break;
12289	case DBGFEVENT_XCPT_TS: VBOXVMM_XCPT_TS(pVCpu, pCtx, uEventArg); break;
12290	case DBGFEVENT_XCPT_NP: VBOXVMM_XCPT_NP(pVCpu, pCtx, uEventArg); break;
12291	case DBGFEVENT_XCPT_SS: VBOXVMM_XCPT_SS(pVCpu, pCtx, uEventArg); break;
12292	case DBGFEVENT_XCPT_GP: VBOXVMM_XCPT_GP(pVCpu, pCtx, uEventArg); break;
12293	case DBGFEVENT_XCPT_PF: VBOXVMM_XCPT_PF(pVCpu, pCtx, uEventArg, pCtx->cr2); break;
12294	case DBGFEVENT_XCPT_MF: VBOXVMM_XCPT_MF(pVCpu, pCtx); break;
12295	case DBGFEVENT_XCPT_AC: VBOXVMM_XCPT_AC(pVCpu, pCtx); break;
12296	case DBGFEVENT_XCPT_XF: VBOXVMM_XCPT_XF(pVCpu, pCtx); break;
12297	case DBGFEVENT_XCPT_VE: VBOXVMM_XCPT_VE(pVCpu, pCtx); break;
12298	case DBGFEVENT_XCPT_SX: VBOXVMM_XCPT_SX(pVCpu, pCtx, uEventArg); break;
12299	case DBGFEVENT_INTERRUPT_SOFTWARE: VBOXVMM_INT_SOFTWARE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12300	case DBGFEVENT_INSTR_CPUID: VBOXVMM_INSTR_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
12301	case DBGFEVENT_INSTR_GETSEC: VBOXVMM_INSTR_GETSEC(pVCpu, pCtx); break;
12302	case DBGFEVENT_INSTR_HALT: VBOXVMM_INSTR_HALT(pVCpu, pCtx); break;
12303	case DBGFEVENT_INSTR_INVD: VBOXVMM_INSTR_INVD(pVCpu, pCtx); break;
12304	case DBGFEVENT_INSTR_INVLPG: VBOXVMM_INSTR_INVLPG(pVCpu, pCtx); break;
12305	case DBGFEVENT_INSTR_RDPMC: VBOXVMM_INSTR_RDPMC(pVCpu, pCtx); break;
12306	case DBGFEVENT_INSTR_RDTSC: VBOXVMM_INSTR_RDTSC(pVCpu, pCtx); break;
12307	case DBGFEVENT_INSTR_RSM: VBOXVMM_INSTR_RSM(pVCpu, pCtx); break;
12308	case DBGFEVENT_INSTR_CRX_READ: VBOXVMM_INSTR_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12309	case DBGFEVENT_INSTR_CRX_WRITE: VBOXVMM_INSTR_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12310	case DBGFEVENT_INSTR_DRX_READ: VBOXVMM_INSTR_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12311	case DBGFEVENT_INSTR_DRX_WRITE: VBOXVMM_INSTR_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12312	case DBGFEVENT_INSTR_RDMSR: VBOXVMM_INSTR_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
12313	case DBGFEVENT_INSTR_WRMSR: VBOXVMM_INSTR_WRMSR(pVCpu, pCtx, pCtx->ecx,
12314	RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
12315	case DBGFEVENT_INSTR_MWAIT: VBOXVMM_INSTR_MWAIT(pVCpu, pCtx); break;
12316	case DBGFEVENT_INSTR_MONITOR: VBOXVMM_INSTR_MONITOR(pVCpu, pCtx); break;
12317	case DBGFEVENT_INSTR_PAUSE: VBOXVMM_INSTR_PAUSE(pVCpu, pCtx); break;
12318	case DBGFEVENT_INSTR_SGDT: VBOXVMM_INSTR_SGDT(pVCpu, pCtx); break;
12319	case DBGFEVENT_INSTR_SIDT: VBOXVMM_INSTR_SIDT(pVCpu, pCtx); break;
12320	case DBGFEVENT_INSTR_LGDT: VBOXVMM_INSTR_LGDT(pVCpu, pCtx); break;
12321	case DBGFEVENT_INSTR_LIDT: VBOXVMM_INSTR_LIDT(pVCpu, pCtx); break;
12322	case DBGFEVENT_INSTR_SLDT: VBOXVMM_INSTR_SLDT(pVCpu, pCtx); break;
12323	case DBGFEVENT_INSTR_STR: VBOXVMM_INSTR_STR(pVCpu, pCtx); break;
12324	case DBGFEVENT_INSTR_LLDT: VBOXVMM_INSTR_LLDT(pVCpu, pCtx); break;
12325	case DBGFEVENT_INSTR_LTR: VBOXVMM_INSTR_LTR(pVCpu, pCtx); break;
12326	case DBGFEVENT_INSTR_RDTSCP: VBOXVMM_INSTR_RDTSCP(pVCpu, pCtx); break;
12327	case DBGFEVENT_INSTR_WBINVD: VBOXVMM_INSTR_WBINVD(pVCpu, pCtx); break;
12328	case DBGFEVENT_INSTR_XSETBV: VBOXVMM_INSTR_XSETBV(pVCpu, pCtx); break;
12329	case DBGFEVENT_INSTR_RDRAND: VBOXVMM_INSTR_RDRAND(pVCpu, pCtx); break;
12330	case DBGFEVENT_INSTR_RDSEED: VBOXVMM_INSTR_RDSEED(pVCpu, pCtx); break;
12331	case DBGFEVENT_INSTR_XSAVES: VBOXVMM_INSTR_XSAVES(pVCpu, pCtx); break;
12332	case DBGFEVENT_INSTR_XRSTORS: VBOXVMM_INSTR_XRSTORS(pVCpu, pCtx); break;
12333	case DBGFEVENT_INSTR_VMM_CALL: VBOXVMM_INSTR_VMM_CALL(pVCpu, pCtx); break;
12334	case DBGFEVENT_INSTR_VMX_VMCLEAR: VBOXVMM_INSTR_VMX_VMCLEAR(pVCpu, pCtx); break;
12335	case DBGFEVENT_INSTR_VMX_VMLAUNCH: VBOXVMM_INSTR_VMX_VMLAUNCH(pVCpu, pCtx); break;
12336	case DBGFEVENT_INSTR_VMX_VMPTRLD: VBOXVMM_INSTR_VMX_VMPTRLD(pVCpu, pCtx); break;
12337	case DBGFEVENT_INSTR_VMX_VMPTRST: VBOXVMM_INSTR_VMX_VMPTRST(pVCpu, pCtx); break;
12338	case DBGFEVENT_INSTR_VMX_VMREAD: VBOXVMM_INSTR_VMX_VMREAD(pVCpu, pCtx); break;
12339	case DBGFEVENT_INSTR_VMX_VMRESUME: VBOXVMM_INSTR_VMX_VMRESUME(pVCpu, pCtx); break;
12340	case DBGFEVENT_INSTR_VMX_VMWRITE: VBOXVMM_INSTR_VMX_VMWRITE(pVCpu, pCtx); break;
12341	case DBGFEVENT_INSTR_VMX_VMXOFF: VBOXVMM_INSTR_VMX_VMXOFF(pVCpu, pCtx); break;
12342	case DBGFEVENT_INSTR_VMX_VMXON: VBOXVMM_INSTR_VMX_VMXON(pVCpu, pCtx); break;
12343	case DBGFEVENT_INSTR_VMX_INVEPT: VBOXVMM_INSTR_VMX_INVEPT(pVCpu, pCtx); break;
12344	case DBGFEVENT_INSTR_VMX_INVVPID: VBOXVMM_INSTR_VMX_INVVPID(pVCpu, pCtx); break;
12345	case DBGFEVENT_INSTR_VMX_INVPCID: VBOXVMM_INSTR_VMX_INVPCID(pVCpu, pCtx); break;
12346	case DBGFEVENT_INSTR_VMX_VMFUNC: VBOXVMM_INSTR_VMX_VMFUNC(pVCpu, pCtx); break;
12347	default: AssertMsgFailed(("enmEvent1=%d uExitReason=%d\n", enmEvent1, uExitReason)); break;
12348	}
12349	switch (enmEvent2)
12350	{
12351	/** @todo consider which extra parameters would be helpful for each probe. */
12352	case DBGFEVENT_END: break;
12353	case DBGFEVENT_EXIT_TASK_SWITCH: VBOXVMM_EXIT_TASK_SWITCH(pVCpu, pCtx); break;
12354	case DBGFEVENT_EXIT_CPUID: VBOXVMM_EXIT_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
12355	case DBGFEVENT_EXIT_GETSEC: VBOXVMM_EXIT_GETSEC(pVCpu, pCtx); break;
12356	case DBGFEVENT_EXIT_HALT: VBOXVMM_EXIT_HALT(pVCpu, pCtx); break;
12357	case DBGFEVENT_EXIT_INVD: VBOXVMM_EXIT_INVD(pVCpu, pCtx); break;
12358	case DBGFEVENT_EXIT_INVLPG: VBOXVMM_EXIT_INVLPG(pVCpu, pCtx); break;
12359	case DBGFEVENT_EXIT_RDPMC: VBOXVMM_EXIT_RDPMC(pVCpu, pCtx); break;
12360	case DBGFEVENT_EXIT_RDTSC: VBOXVMM_EXIT_RDTSC(pVCpu, pCtx); break;
12361	case DBGFEVENT_EXIT_RSM: VBOXVMM_EXIT_RSM(pVCpu, pCtx); break;
12362	case DBGFEVENT_EXIT_CRX_READ: VBOXVMM_EXIT_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12363	case DBGFEVENT_EXIT_CRX_WRITE: VBOXVMM_EXIT_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12364	case DBGFEVENT_EXIT_DRX_READ: VBOXVMM_EXIT_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12365	case DBGFEVENT_EXIT_DRX_WRITE: VBOXVMM_EXIT_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12366	case DBGFEVENT_EXIT_RDMSR: VBOXVMM_EXIT_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
12367	case DBGFEVENT_EXIT_WRMSR: VBOXVMM_EXIT_WRMSR(pVCpu, pCtx, pCtx->ecx,
12368	RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
12369	case DBGFEVENT_EXIT_MWAIT: VBOXVMM_EXIT_MWAIT(pVCpu, pCtx); break;
12370	case DBGFEVENT_EXIT_MONITOR: VBOXVMM_EXIT_MONITOR(pVCpu, pCtx); break;
12371	case DBGFEVENT_EXIT_PAUSE: VBOXVMM_EXIT_PAUSE(pVCpu, pCtx); break;
12372	case DBGFEVENT_EXIT_SGDT: VBOXVMM_EXIT_SGDT(pVCpu, pCtx); break;
12373	case DBGFEVENT_EXIT_SIDT: VBOXVMM_EXIT_SIDT(pVCpu, pCtx); break;
12374	case DBGFEVENT_EXIT_LGDT: VBOXVMM_EXIT_LGDT(pVCpu, pCtx); break;
12375	case DBGFEVENT_EXIT_LIDT: VBOXVMM_EXIT_LIDT(pVCpu, pCtx); break;
12376	case DBGFEVENT_EXIT_SLDT: VBOXVMM_EXIT_SLDT(pVCpu, pCtx); break;
12377	case DBGFEVENT_EXIT_STR: VBOXVMM_EXIT_STR(pVCpu, pCtx); break;
12378	case DBGFEVENT_EXIT_LLDT: VBOXVMM_EXIT_LLDT(pVCpu, pCtx); break;
12379	case DBGFEVENT_EXIT_LTR: VBOXVMM_EXIT_LTR(pVCpu, pCtx); break;
12380	case DBGFEVENT_EXIT_RDTSCP: VBOXVMM_EXIT_RDTSCP(pVCpu, pCtx); break;
12381	case DBGFEVENT_EXIT_WBINVD: VBOXVMM_EXIT_WBINVD(pVCpu, pCtx); break;
12382	case DBGFEVENT_EXIT_XSETBV: VBOXVMM_EXIT_XSETBV(pVCpu, pCtx); break;
12383	case DBGFEVENT_EXIT_RDRAND: VBOXVMM_EXIT_RDRAND(pVCpu, pCtx); break;
12384	case DBGFEVENT_EXIT_RDSEED: VBOXVMM_EXIT_RDSEED(pVCpu, pCtx); break;
12385	case DBGFEVENT_EXIT_XSAVES: VBOXVMM_EXIT_XSAVES(pVCpu, pCtx); break;
12386	case DBGFEVENT_EXIT_XRSTORS: VBOXVMM_EXIT_XRSTORS(pVCpu, pCtx); break;
12387	case DBGFEVENT_EXIT_VMM_CALL: VBOXVMM_EXIT_VMM_CALL(pVCpu, pCtx); break;
12388	case DBGFEVENT_EXIT_VMX_VMCLEAR: VBOXVMM_EXIT_VMX_VMCLEAR(pVCpu, pCtx); break;
12389	case DBGFEVENT_EXIT_VMX_VMLAUNCH: VBOXVMM_EXIT_VMX_VMLAUNCH(pVCpu, pCtx); break;
12390	case DBGFEVENT_EXIT_VMX_VMPTRLD: VBOXVMM_EXIT_VMX_VMPTRLD(pVCpu, pCtx); break;
12391	case DBGFEVENT_EXIT_VMX_VMPTRST: VBOXVMM_EXIT_VMX_VMPTRST(pVCpu, pCtx); break;
12392	case DBGFEVENT_EXIT_VMX_VMREAD: VBOXVMM_EXIT_VMX_VMREAD(pVCpu, pCtx); break;
12393	case DBGFEVENT_EXIT_VMX_VMRESUME: VBOXVMM_EXIT_VMX_VMRESUME(pVCpu, pCtx); break;
12394	case DBGFEVENT_EXIT_VMX_VMWRITE: VBOXVMM_EXIT_VMX_VMWRITE(pVCpu, pCtx); break;
12395	case DBGFEVENT_EXIT_VMX_VMXOFF: VBOXVMM_EXIT_VMX_VMXOFF(pVCpu, pCtx); break;
12396	case DBGFEVENT_EXIT_VMX_VMXON: VBOXVMM_EXIT_VMX_VMXON(pVCpu, pCtx); break;
12397	case DBGFEVENT_EXIT_VMX_INVEPT: VBOXVMM_EXIT_VMX_INVEPT(pVCpu, pCtx); break;
12398	case DBGFEVENT_EXIT_VMX_INVVPID: VBOXVMM_EXIT_VMX_INVVPID(pVCpu, pCtx); break;
12399	case DBGFEVENT_EXIT_VMX_INVPCID: VBOXVMM_EXIT_VMX_INVPCID(pVCpu, pCtx); break;
12400	case DBGFEVENT_EXIT_VMX_VMFUNC: VBOXVMM_EXIT_VMX_VMFUNC(pVCpu, pCtx); break;
12401	case DBGFEVENT_EXIT_VMX_EPT_MISCONFIG: VBOXVMM_EXIT_VMX_EPT_MISCONFIG(pVCpu, pCtx); break;
12402	case DBGFEVENT_EXIT_VMX_EPT_VIOLATION: VBOXVMM_EXIT_VMX_EPT_VIOLATION(pVCpu, pCtx); break;
12403	case DBGFEVENT_EXIT_VMX_VAPIC_ACCESS: VBOXVMM_EXIT_VMX_VAPIC_ACCESS(pVCpu, pCtx); break;
12404	case DBGFEVENT_EXIT_VMX_VAPIC_WRITE: VBOXVMM_EXIT_VMX_VAPIC_WRITE(pVCpu, pCtx); break;
12405	default: AssertMsgFailed(("enmEvent2=%d uExitReason=%d\n", enmEvent2, uExitReason)); break;
12406	}
12407	}
12408
12409	/*
12410	* Fire of the DBGF event, if enabled (our check here is just a quick one,
12411	* the DBGF call will do a full check).
12412	*
12413	* Note! DBGF sets DBGFEVENT_INTERRUPT_SOFTWARE in the bitmap.
12414	* Note! If we have to events, we prioritize the first, i.e. the instruction
12415	* one, in order to avoid event nesting.
12416	*/
12417	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
12418	if ( enmEvent1 != DBGFEVENT_END
12419	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent1))
12420	{
12421	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12422	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent1, DBGFEVENTCTX_HM, 1, uEventArg);
12423	if (rcStrict != VINF_SUCCESS)
12424	return rcStrict;
12425	}
12426	else if ( enmEvent2 != DBGFEVENT_END
12427	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent2))
12428	{
12429	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12430	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent2, DBGFEVENTCTX_HM, 1, uEventArg);
12431	if (rcStrict != VINF_SUCCESS)
12432	return rcStrict;
12433	}
12434
12435	return VINF_SUCCESS;
12436	}
12437
12438
12439	/**
12440	* Single-stepping VM-exit filtering.
12441	*
12442	* This is preprocessing the VM-exits and deciding whether we've gotten far
12443	* enough to return VINF_EM_DBG_STEPPED already. If not, normal VM-exit
12444	* handling is performed.
12445	*
12446	* @returns Strict VBox status code (i.e. informational status codes too).
12447	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
12448	* @param pVmxTransient The VMX-transient structure.
12449	* @param pDbgState The debug state.
12450	*/
12451	DECLINLINE(VBOXSTRICTRC) hmR0VmxRunDebugHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
12452	{
12453	/*
12454	* Expensive (saves context) generic dtrace VM-exit probe.
12455	*/
12456	uint32_t const uExitReason = pVmxTransient->uExitReason;
12457	if (!VBOXVMM_R0_HMVMX_VMEXIT_ENABLED())
12458	{ /* more likely */ }
12459	else
12460	{
12461	hmR0VmxReadExitQualVmcs(pVmxTransient);
12462	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
12463	AssertRC(rc);
12464	VBOXVMM_R0_HMVMX_VMEXIT(pVCpu, &pVCpu->cpum.GstCtx, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
12465	}
12466
12467	/*
12468	* Check for host NMI, just to get that out of the way.
12469	*/
12470	if (uExitReason != VMX_EXIT_XCPT_OR_NMI)
12471	{ /* normally likely */ }
12472	else
12473	{
12474	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
12475	uint32_t const uIntType = VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo);
12476	if (uIntType == VMX_EXIT_INT_INFO_TYPE_NMI)
12477	return hmR0VmxExitHostNmi(pVCpu, pVmxTransient->pVmcsInfo);
12478	}
12479
12480	/*
12481	* Check for single stepping event if we're stepping.
12482	*/
12483	if (pVCpu->hm.s.fSingleInstruction)
12484	{
12485	switch (uExitReason)
12486	{
12487	case VMX_EXIT_MTF:
12488	return hmR0VmxExitMtf(pVCpu, pVmxTransient);
12489
12490	/* Various events: */
12491	case VMX_EXIT_XCPT_OR_NMI:
12492	case VMX_EXIT_EXT_INT:
12493	case VMX_EXIT_TRIPLE_FAULT:
12494	case VMX_EXIT_INT_WINDOW:
12495	case VMX_EXIT_NMI_WINDOW:
12496	case VMX_EXIT_TASK_SWITCH:
12497	case VMX_EXIT_TPR_BELOW_THRESHOLD:
12498	case VMX_EXIT_APIC_ACCESS:
12499	case VMX_EXIT_EPT_VIOLATION:
12500	case VMX_EXIT_EPT_MISCONFIG:
12501	case VMX_EXIT_PREEMPT_TIMER:
12502
12503	/* Instruction specific VM-exits: */
12504	case VMX_EXIT_CPUID:
12505	case VMX_EXIT_GETSEC:
12506	case VMX_EXIT_HLT:
12507	case VMX_EXIT_INVD:
12508	case VMX_EXIT_INVLPG:
12509	case VMX_EXIT_RDPMC:
12510	case VMX_EXIT_RDTSC:
12511	case VMX_EXIT_RSM:
12512	case VMX_EXIT_VMCALL:
12513	case VMX_EXIT_VMCLEAR:
12514	case VMX_EXIT_VMLAUNCH:
12515	case VMX_EXIT_VMPTRLD:
12516	case VMX_EXIT_VMPTRST:
12517	case VMX_EXIT_VMREAD:
12518	case VMX_EXIT_VMRESUME:
12519	case VMX_EXIT_VMWRITE:
12520	case VMX_EXIT_VMXOFF:
12521	case VMX_EXIT_VMXON:
12522	case VMX_EXIT_MOV_CRX:
12523	case VMX_EXIT_MOV_DRX:
12524	case VMX_EXIT_IO_INSTR:
12525	case VMX_EXIT_RDMSR:
12526	case VMX_EXIT_WRMSR:
12527	case VMX_EXIT_MWAIT:
12528	case VMX_EXIT_MONITOR:
12529	case VMX_EXIT_PAUSE:
12530	case VMX_EXIT_GDTR_IDTR_ACCESS:
12531	case VMX_EXIT_LDTR_TR_ACCESS:
12532	case VMX_EXIT_INVEPT:
12533	case VMX_EXIT_RDTSCP:
12534	case VMX_EXIT_INVVPID:
12535	case VMX_EXIT_WBINVD:
12536	case VMX_EXIT_XSETBV:
12537	case VMX_EXIT_RDRAND:
12538	case VMX_EXIT_INVPCID:
12539	case VMX_EXIT_VMFUNC:
12540	case VMX_EXIT_RDSEED:
12541	case VMX_EXIT_XSAVES:
12542	case VMX_EXIT_XRSTORS:
12543	{
12544	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12545	AssertRCReturn(rc, rc);
12546	if ( pVCpu->cpum.GstCtx.rip != pDbgState->uRipStart
12547	\|\| pVCpu->cpum.GstCtx.cs.Sel != pDbgState->uCsStart)
12548	return VINF_EM_DBG_STEPPED;
12549	break;
12550	}
12551
12552	/* Errors and unexpected events: */
12553	case VMX_EXIT_INIT_SIGNAL:
12554	case VMX_EXIT_SIPI:
12555	case VMX_EXIT_IO_SMI:
12556	case VMX_EXIT_SMI:
12557	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
12558	case VMX_EXIT_ERR_MSR_LOAD:
12559	case VMX_EXIT_ERR_MACHINE_CHECK:
12560	case VMX_EXIT_PML_FULL:
12561	case VMX_EXIT_VIRTUALIZED_EOI:
12562	case VMX_EXIT_APIC_WRITE: /* Some talk about this being fault like, so I guess we must process it? */
12563	break;
12564
12565	default:
12566	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
12567	break;
12568	}
12569	}
12570
12571	/*
12572	* Check for debugger event breakpoints and dtrace probes.
12573	*/
12574	if ( uExitReason < RT_ELEMENTS(pDbgState->bmExitsToCheck) * 32U
12575	&& ASMBitTest(pDbgState->bmExitsToCheck, uExitReason) )
12576	{
12577	VBOXSTRICTRC rcStrict = hmR0VmxHandleExitDtraceEvents(pVCpu, pVmxTransient, uExitReason);
12578	if (rcStrict != VINF_SUCCESS)
12579	return rcStrict;
12580	}
12581
12582	/*
12583	* Normal processing.
12584	*/
12585	#ifdef HMVMX_USE_FUNCTION_TABLE
12586	return g_aVMExitHandlers[uExitReason].pfn(pVCpu, pVmxTransient);
12587	#else
12588	return hmR0VmxHandleExit(pVCpu, pVmxTransient, uExitReason);
12589	#endif
12590	}
12591
12592
12593	/**
12594	* Single steps guest code using hardware-assisted VMX.
12595	*
12596	* This is -not- the same as the guest single-stepping itself (say using EFLAGS.TF)
12597	* but single-stepping through the hypervisor debugger.
12598	*
12599	* @returns Strict VBox status code (i.e. informational status codes too).
12600	* @param pVCpu The cross context virtual CPU structure.
12601	* @param pcLoops Pointer to the number of executed loops.
12602	*
12603	* @note Mostly the same as hmR0VmxRunGuestCodeNormal().
12604	*/
12605	static VBOXSTRICTRC hmR0VmxRunGuestCodeDebug(PVMCPUCC pVCpu, uint32_t *pcLoops)
12606	{
12607	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops;
12608	Assert(pcLoops);
12609	Assert(*pcLoops <= cMaxResumeLoops);
12610
12611	VMXTRANSIENT VmxTransient;
12612	RT_ZERO(VmxTransient);
12613	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
12614
12615	/* Set HMCPU indicators. */
12616	bool const fSavedSingleInstruction = pVCpu->hm.s.fSingleInstruction;
12617	pVCpu->hm.s.fSingleInstruction = pVCpu->hm.s.fSingleInstruction \|\| DBGFIsStepping(pVCpu);
12618	pVCpu->hmr0.s.fDebugWantRdTscExit = false;
12619	pVCpu->hmr0.s.fUsingDebugLoop = true;
12620
12621	/* State we keep to help modify and later restore the VMCS fields we alter, and for detecting steps. */
12622	VMXRUNDBGSTATE DbgState;
12623	hmR0VmxRunDebugStateInit(pVCpu, &VmxTransient, &DbgState);
12624	hmR0VmxPreRunGuestDebugStateUpdate(pVCpu, &VmxTransient, &DbgState);
12625
12626	/*
12627	* The loop.
12628	*/
12629	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
12630	for (;;)
12631	{
12632	Assert(!HMR0SuspendPending());
12633	HMVMX_ASSERT_CPU_SAFE(pVCpu);
12634	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
12635	bool fStepping = pVCpu->hm.s.fSingleInstruction;
12636
12637	/* Set up VM-execution controls the next two can respond to. */
12638	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &VmxTransient, &DbgState);
12639
12640	/*
12641	* Preparatory work for running guest code, this may force us to
12642	* return to ring-3.
12643	*
12644	* Warning! This bugger disables interrupts on VINF_SUCCESS!
12645	*/
12646	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, fStepping);
12647	if (rcStrict != VINF_SUCCESS)
12648	break;
12649
12650	/* Interrupts are disabled at this point! */
12651	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
12652
12653	/* Override any obnoxious code in the above two calls. */
12654	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &VmxTransient, &DbgState);
12655
12656	/*
12657	* Finally execute the guest.
12658	*/
12659	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
12660
12661	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
12662	/* Interrupts are re-enabled at this point! */
12663
12664	/* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
12665	if (RT_SUCCESS(rcRun))
12666	{ /* very likely */ }
12667	else
12668	{
12669	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
12670	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
12671	return rcRun;
12672	}
12673
12674	/* Profile the VM-exit. */
12675	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
12676	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
12677	STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
12678	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
12679	HMVMX_START_EXIT_DISPATCH_PROF();
12680
12681	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
12682
12683	/*
12684	* Handle the VM-exit - we quit earlier on certain VM-exits, see hmR0VmxHandleExitDebug().
12685	*/
12686	rcStrict = hmR0VmxRunDebugHandleExit(pVCpu, &VmxTransient, &DbgState);
12687	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
12688	if (rcStrict != VINF_SUCCESS)
12689	break;
12690	if (++(*pcLoops) > cMaxResumeLoops)
12691	{
12692	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
12693	rcStrict = VINF_EM_RAW_INTERRUPT;
12694	break;
12695	}
12696
12697	/*
12698	* Stepping: Did the RIP change, if so, consider it a single step.
12699	* Otherwise, make sure one of the TFs gets set.
12700	*/
12701	if (fStepping)
12702	{
12703	int rc = hmR0VmxImportGuestState(pVCpu, VmxTransient.pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12704	AssertRC(rc);
12705	if ( pVCpu->cpum.GstCtx.rip != DbgState.uRipStart
12706	\|\| pVCpu->cpum.GstCtx.cs.Sel != DbgState.uCsStart)
12707	{
12708	rcStrict = VINF_EM_DBG_STEPPED;
12709	break;
12710	}
12711	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
12712	}
12713
12714	/*
12715	* Update when dtrace settings changes (DBGF kicks us, so no need to check).
12716	*/
12717	if (VBOXVMM_GET_SETTINGS_SEQ_NO() != DbgState.uDtraceSettingsSeqNo)
12718	hmR0VmxPreRunGuestDebugStateUpdate(pVCpu, &VmxTransient, &DbgState);
12719
12720	/* Restore all controls applied by hmR0VmxPreRunGuestDebugStateApply above. */
12721	rcStrict = hmR0VmxRunDebugStateRevert(pVCpu, &VmxTransient, &DbgState, rcStrict);
12722	Assert(rcStrict == VINF_SUCCESS);
12723	}
12724
12725	/*
12726	* Clear the X86_EFL_TF if necessary.
12727	*/
12728	if (pVCpu->hmr0.s.fClearTrapFlag)
12729	{
12730	int rc = hmR0VmxImportGuestState(pVCpu, VmxTransient.pVmcsInfo, CPUMCTX_EXTRN_RFLAGS);
12731	AssertRC(rc);
12732	pVCpu->hmr0.s.fClearTrapFlag = false;
12733	pVCpu->cpum.GstCtx.eflags.Bits.u1TF = 0;
12734	}
12735	/** @todo there seems to be issues with the resume flag when the monitor trap
12736	* flag is pending without being used. Seen early in bios init when
12737	* accessing APIC page in protected mode. */
12738
12739	/* Restore HMCPU indicators. */
12740	pVCpu->hmr0.s.fUsingDebugLoop = false;
12741	pVCpu->hmr0.s.fDebugWantRdTscExit = false;
12742	pVCpu->hm.s.fSingleInstruction = fSavedSingleInstruction;
12743
12744	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
12745	return rcStrict;
12746	}
12747
12748
12749	/** @} */
12750
12751
12752	/**
12753	* Checks if any expensive dtrace probes are enabled and we should go to the
12754	* debug loop.
12755	*
12756	* @returns true if we should use debug loop, false if not.
12757	*/
12758	static bool hmR0VmxAnyExpensiveProbesEnabled(void)
12759	{
12760	/* It's probably faster to OR the raw 32-bit counter variables together.
12761	Since the variables are in an array and the probes are next to one
12762	another (more or less), we have good locality. So, better read
12763	eight-nine cache lines ever time and only have one conditional, than
12764	128+ conditionals, right? */
12765	return ( VBOXVMM_R0_HMVMX_VMEXIT_ENABLED_RAW() /* expensive too due to context */
12766	\| VBOXVMM_XCPT_DE_ENABLED_RAW()
12767	\| VBOXVMM_XCPT_DB_ENABLED_RAW()
12768	\| VBOXVMM_XCPT_BP_ENABLED_RAW()
12769	\| VBOXVMM_XCPT_OF_ENABLED_RAW()
12770	\| VBOXVMM_XCPT_BR_ENABLED_RAW()
12771	\| VBOXVMM_XCPT_UD_ENABLED_RAW()
12772	\| VBOXVMM_XCPT_NM_ENABLED_RAW()
12773	\| VBOXVMM_XCPT_DF_ENABLED_RAW()
12774	\| VBOXVMM_XCPT_TS_ENABLED_RAW()
12775	\| VBOXVMM_XCPT_NP_ENABLED_RAW()
12776	\| VBOXVMM_XCPT_SS_ENABLED_RAW()
12777	\| VBOXVMM_XCPT_GP_ENABLED_RAW()
12778	\| VBOXVMM_XCPT_PF_ENABLED_RAW()
12779	\| VBOXVMM_XCPT_MF_ENABLED_RAW()
12780	\| VBOXVMM_XCPT_AC_ENABLED_RAW()
12781	\| VBOXVMM_XCPT_XF_ENABLED_RAW()
12782	\| VBOXVMM_XCPT_VE_ENABLED_RAW()
12783	\| VBOXVMM_XCPT_SX_ENABLED_RAW()
12784	\| VBOXVMM_INT_SOFTWARE_ENABLED_RAW()
12785	\| VBOXVMM_INT_HARDWARE_ENABLED_RAW()
12786	) != 0
12787	\|\| ( VBOXVMM_INSTR_HALT_ENABLED_RAW()
12788	\| VBOXVMM_INSTR_MWAIT_ENABLED_RAW()
12789	\| VBOXVMM_INSTR_MONITOR_ENABLED_RAW()
12790	\| VBOXVMM_INSTR_CPUID_ENABLED_RAW()
12791	\| VBOXVMM_INSTR_INVD_ENABLED_RAW()
12792	\| VBOXVMM_INSTR_WBINVD_ENABLED_RAW()
12793	\| VBOXVMM_INSTR_INVLPG_ENABLED_RAW()
12794	\| VBOXVMM_INSTR_RDTSC_ENABLED_RAW()
12795	\| VBOXVMM_INSTR_RDTSCP_ENABLED_RAW()
12796	\| VBOXVMM_INSTR_RDPMC_ENABLED_RAW()
12797	\| VBOXVMM_INSTR_RDMSR_ENABLED_RAW()
12798	\| VBOXVMM_INSTR_WRMSR_ENABLED_RAW()
12799	\| VBOXVMM_INSTR_CRX_READ_ENABLED_RAW()
12800	\| VBOXVMM_INSTR_CRX_WRITE_ENABLED_RAW()
12801	\| VBOXVMM_INSTR_DRX_READ_ENABLED_RAW()
12802	\| VBOXVMM_INSTR_DRX_WRITE_ENABLED_RAW()
12803	\| VBOXVMM_INSTR_PAUSE_ENABLED_RAW()
12804	\| VBOXVMM_INSTR_XSETBV_ENABLED_RAW()
12805	\| VBOXVMM_INSTR_SIDT_ENABLED_RAW()
12806	\| VBOXVMM_INSTR_LIDT_ENABLED_RAW()
12807	\| VBOXVMM_INSTR_SGDT_ENABLED_RAW()
12808	\| VBOXVMM_INSTR_LGDT_ENABLED_RAW()
12809	\| VBOXVMM_INSTR_SLDT_ENABLED_RAW()
12810	\| VBOXVMM_INSTR_LLDT_ENABLED_RAW()
12811	\| VBOXVMM_INSTR_STR_ENABLED_RAW()
12812	\| VBOXVMM_INSTR_LTR_ENABLED_RAW()
12813	\| VBOXVMM_INSTR_GETSEC_ENABLED_RAW()
12814	\| VBOXVMM_INSTR_RSM_ENABLED_RAW()
12815	\| VBOXVMM_INSTR_RDRAND_ENABLED_RAW()
12816	\| VBOXVMM_INSTR_RDSEED_ENABLED_RAW()
12817	\| VBOXVMM_INSTR_XSAVES_ENABLED_RAW()
12818	\| VBOXVMM_INSTR_XRSTORS_ENABLED_RAW()
12819	\| VBOXVMM_INSTR_VMM_CALL_ENABLED_RAW()
12820	\| VBOXVMM_INSTR_VMX_VMCLEAR_ENABLED_RAW()
12821	\| VBOXVMM_INSTR_VMX_VMLAUNCH_ENABLED_RAW()
12822	\| VBOXVMM_INSTR_VMX_VMPTRLD_ENABLED_RAW()
12823	\| VBOXVMM_INSTR_VMX_VMPTRST_ENABLED_RAW()
12824	\| VBOXVMM_INSTR_VMX_VMREAD_ENABLED_RAW()
12825	\| VBOXVMM_INSTR_VMX_VMRESUME_ENABLED_RAW()
12826	\| VBOXVMM_INSTR_VMX_VMWRITE_ENABLED_RAW()
12827	\| VBOXVMM_INSTR_VMX_VMXOFF_ENABLED_RAW()
12828	\| VBOXVMM_INSTR_VMX_VMXON_ENABLED_RAW()
12829	\| VBOXVMM_INSTR_VMX_VMFUNC_ENABLED_RAW()
12830	\| VBOXVMM_INSTR_VMX_INVEPT_ENABLED_RAW()
12831	\| VBOXVMM_INSTR_VMX_INVVPID_ENABLED_RAW()
12832	\| VBOXVMM_INSTR_VMX_INVPCID_ENABLED_RAW()
12833	) != 0
12834	\|\| ( VBOXVMM_EXIT_TASK_SWITCH_ENABLED_RAW()
12835	\| VBOXVMM_EXIT_HALT_ENABLED_RAW()
12836	\| VBOXVMM_EXIT_MWAIT_ENABLED_RAW()
12837	\| VBOXVMM_EXIT_MONITOR_ENABLED_RAW()
12838	\| VBOXVMM_EXIT_CPUID_ENABLED_RAW()
12839	\| VBOXVMM_EXIT_INVD_ENABLED_RAW()
12840	\| VBOXVMM_EXIT_WBINVD_ENABLED_RAW()
12841	\| VBOXVMM_EXIT_INVLPG_ENABLED_RAW()
12842	\| VBOXVMM_EXIT_RDTSC_ENABLED_RAW()
12843	\| VBOXVMM_EXIT_RDTSCP_ENABLED_RAW()
12844	\| VBOXVMM_EXIT_RDPMC_ENABLED_RAW()
12845	\| VBOXVMM_EXIT_RDMSR_ENABLED_RAW()
12846	\| VBOXVMM_EXIT_WRMSR_ENABLED_RAW()
12847	\| VBOXVMM_EXIT_CRX_READ_ENABLED_RAW()
12848	\| VBOXVMM_EXIT_CRX_WRITE_ENABLED_RAW()
12849	\| VBOXVMM_EXIT_DRX_READ_ENABLED_RAW()
12850	\| VBOXVMM_EXIT_DRX_WRITE_ENABLED_RAW()
12851	\| VBOXVMM_EXIT_PAUSE_ENABLED_RAW()
12852	\| VBOXVMM_EXIT_XSETBV_ENABLED_RAW()
12853	\| VBOXVMM_EXIT_SIDT_ENABLED_RAW()
12854	\| VBOXVMM_EXIT_LIDT_ENABLED_RAW()
12855	\| VBOXVMM_EXIT_SGDT_ENABLED_RAW()
12856	\| VBOXVMM_EXIT_LGDT_ENABLED_RAW()
12857	\| VBOXVMM_EXIT_SLDT_ENABLED_RAW()
12858	\| VBOXVMM_EXIT_LLDT_ENABLED_RAW()
12859	\| VBOXVMM_EXIT_STR_ENABLED_RAW()
12860	\| VBOXVMM_EXIT_LTR_ENABLED_RAW()
12861	\| VBOXVMM_EXIT_GETSEC_ENABLED_RAW()
12862	\| VBOXVMM_EXIT_RSM_ENABLED_RAW()
12863	\| VBOXVMM_EXIT_RDRAND_ENABLED_RAW()
12864	\| VBOXVMM_EXIT_RDSEED_ENABLED_RAW()
12865	\| VBOXVMM_EXIT_XSAVES_ENABLED_RAW()
12866	\| VBOXVMM_EXIT_XRSTORS_ENABLED_RAW()
12867	\| VBOXVMM_EXIT_VMM_CALL_ENABLED_RAW()
12868	\| VBOXVMM_EXIT_VMX_VMCLEAR_ENABLED_RAW()
12869	\| VBOXVMM_EXIT_VMX_VMLAUNCH_ENABLED_RAW()
12870	\| VBOXVMM_EXIT_VMX_VMPTRLD_ENABLED_RAW()
12871	\| VBOXVMM_EXIT_VMX_VMPTRST_ENABLED_RAW()
12872	\| VBOXVMM_EXIT_VMX_VMREAD_ENABLED_RAW()
12873	\| VBOXVMM_EXIT_VMX_VMRESUME_ENABLED_RAW()
12874	\| VBOXVMM_EXIT_VMX_VMWRITE_ENABLED_RAW()
12875	\| VBOXVMM_EXIT_VMX_VMXOFF_ENABLED_RAW()
12876	\| VBOXVMM_EXIT_VMX_VMXON_ENABLED_RAW()
12877	\| VBOXVMM_EXIT_VMX_VMFUNC_ENABLED_RAW()
12878	\| VBOXVMM_EXIT_VMX_INVEPT_ENABLED_RAW()
12879	\| VBOXVMM_EXIT_VMX_INVVPID_ENABLED_RAW()
12880	\| VBOXVMM_EXIT_VMX_INVPCID_ENABLED_RAW()
12881	\| VBOXVMM_EXIT_VMX_EPT_VIOLATION_ENABLED_RAW()
12882	\| VBOXVMM_EXIT_VMX_EPT_MISCONFIG_ENABLED_RAW()
12883	\| VBOXVMM_EXIT_VMX_VAPIC_ACCESS_ENABLED_RAW()
12884	\| VBOXVMM_EXIT_VMX_VAPIC_WRITE_ENABLED_RAW()
12885	) != 0;
12886	}
12887
12888
12889	/**
12890	* Runs the guest using hardware-assisted VMX.
12891	*
12892	* @returns Strict VBox status code (i.e. informational status codes too).
12893	* @param pVCpu The cross context virtual CPU structure.
12894	*/
12895	VMMR0DECL(VBOXSTRICTRC) VMXR0RunGuestCode(PVMCPUCC pVCpu)
12896	{
12897	AssertPtr(pVCpu);
12898	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
12899	Assert(VMMRZCallRing3IsEnabled(pVCpu));
12900	Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
12901	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
12902
12903	VBOXSTRICTRC rcStrict;
12904	uint32_t cLoops = 0;
12905	for (;;)
12906	{
12907	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12908	bool const fInNestedGuestMode = CPUMIsGuestInVmxNonRootMode(pCtx);
12909	#else
12910	NOREF(pCtx);
12911	bool const fInNestedGuestMode = false;
12912	#endif
12913	if (!fInNestedGuestMode)
12914	{
12915	if ( !pVCpu->hm.s.fUseDebugLoop
12916	&& (!VBOXVMM_ANY_PROBES_ENABLED() \|\| !hmR0VmxAnyExpensiveProbesEnabled())
12917	&& !DBGFIsStepping(pVCpu)
12918	&& !pVCpu->CTX_SUFF(pVM)->dbgf.ro.cEnabledInt3Breakpoints)
12919	rcStrict = hmR0VmxRunGuestCodeNormal(pVCpu, &cLoops);
12920	else
12921	rcStrict = hmR0VmxRunGuestCodeDebug(pVCpu, &cLoops);
12922	}
12923	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12924	else
12925	rcStrict = hmR0VmxRunGuestCodeNested(pVCpu, &cLoops);
12926
12927	if (rcStrict == VINF_VMX_VMLAUNCH_VMRESUME)
12928	{
12929	Assert(CPUMIsGuestInVmxNonRootMode(pCtx));
12930	continue;
12931	}
12932	if (rcStrict == VINF_VMX_VMEXIT)
12933	{
12934	Assert(!CPUMIsGuestInVmxNonRootMode(pCtx));
12935	continue;
12936	}
12937	#endif
12938	break;
12939	}
12940
12941	int const rcLoop = VBOXSTRICTRC_VAL(rcStrict);
12942	switch (rcLoop)
12943	{
12944	case VERR_EM_INTERPRETER: rcStrict = VINF_EM_RAW_EMULATE_INSTR; break;
12945	case VINF_EM_RESET: rcStrict = VINF_EM_TRIPLE_FAULT; break;
12946	}
12947
12948	int rc2 = hmR0VmxExitToRing3(pVCpu, rcStrict);
12949	if (RT_FAILURE(rc2))
12950	{
12951	pVCpu->hm.s.u32HMError = (uint32_t)VBOXSTRICTRC_VAL(rcStrict);
12952	rcStrict = rc2;
12953	}
12954	Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
12955	Assert(!VMMRZCallRing3IsNotificationSet(pVCpu));
12956	return rcStrict;
12957	}
12958
12959
12960	#ifndef HMVMX_USE_FUNCTION_TABLE
12961	/**
12962	* Handles a guest VM-exit from hardware-assisted VMX execution.
12963	*
12964	* @returns Strict VBox status code (i.e. informational status codes too).
12965	* @param pVCpu The cross context virtual CPU structure.
12966	* @param pVmxTransient The VMX-transient structure.
12967	*/
12968	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
12969	{
12970	#ifdef DEBUG_ramshankar
12971	# define VMEXIT_CALL_RET(a_fSave, a_CallExpr) \
12972	do { \
12973	if (a_fSave != 0) \
12974	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL); \
12975	VBOXSTRICTRC rcStrict = a_CallExpr; \
12976	if (a_fSave != 0) \
12977	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST); \
12978	return rcStrict; \
12979	} while (0)
12980	#else
12981	# define VMEXIT_CALL_RET(a_fSave, a_CallExpr) return a_CallExpr
12982	#endif
12983	uint32_t const uExitReason = pVmxTransient->uExitReason;
12984	switch (uExitReason)
12985	{
12986	case VMX_EXIT_EPT_MISCONFIG: VMEXIT_CALL_RET(0, hmR0VmxExitEptMisconfig(pVCpu, pVmxTransient));
12987	case VMX_EXIT_EPT_VIOLATION: VMEXIT_CALL_RET(0, hmR0VmxExitEptViolation(pVCpu, pVmxTransient));
12988	case VMX_EXIT_IO_INSTR: VMEXIT_CALL_RET(0, hmR0VmxExitIoInstr(pVCpu, pVmxTransient));
12989	case VMX_EXIT_CPUID: VMEXIT_CALL_RET(0, hmR0VmxExitCpuid(pVCpu, pVmxTransient));
12990	case VMX_EXIT_RDTSC: VMEXIT_CALL_RET(0, hmR0VmxExitRdtsc(pVCpu, pVmxTransient));
12991	case VMX_EXIT_RDTSCP: VMEXIT_CALL_RET(0, hmR0VmxExitRdtscp(pVCpu, pVmxTransient));
12992	case VMX_EXIT_APIC_ACCESS: VMEXIT_CALL_RET(0, hmR0VmxExitApicAccess(pVCpu, pVmxTransient));
12993	case VMX_EXIT_XCPT_OR_NMI: VMEXIT_CALL_RET(0, hmR0VmxExitXcptOrNmi(pVCpu, pVmxTransient));
12994	case VMX_EXIT_MOV_CRX: VMEXIT_CALL_RET(0, hmR0VmxExitMovCRx(pVCpu, pVmxTransient));
12995	case VMX_EXIT_EXT_INT: VMEXIT_CALL_RET(0, hmR0VmxExitExtInt(pVCpu, pVmxTransient));
12996	case VMX_EXIT_INT_WINDOW: VMEXIT_CALL_RET(0, hmR0VmxExitIntWindow(pVCpu, pVmxTransient));
12997	case VMX_EXIT_TPR_BELOW_THRESHOLD: VMEXIT_CALL_RET(0, hmR0VmxExitTprBelowThreshold(pVCpu, pVmxTransient));
12998	case VMX_EXIT_MWAIT: VMEXIT_CALL_RET(0, hmR0VmxExitMwait(pVCpu, pVmxTransient));
12999	case VMX_EXIT_MONITOR: VMEXIT_CALL_RET(0, hmR0VmxExitMonitor(pVCpu, pVmxTransient));
13000	case VMX_EXIT_TASK_SWITCH: VMEXIT_CALL_RET(0, hmR0VmxExitTaskSwitch(pVCpu, pVmxTransient));
13001	case VMX_EXIT_PREEMPT_TIMER: VMEXIT_CALL_RET(0, hmR0VmxExitPreemptTimer(pVCpu, pVmxTransient));
13002	case VMX_EXIT_RDMSR: VMEXIT_CALL_RET(0, hmR0VmxExitRdmsr(pVCpu, pVmxTransient));
13003	case VMX_EXIT_WRMSR: VMEXIT_CALL_RET(0, hmR0VmxExitWrmsr(pVCpu, pVmxTransient));
13004	case VMX_EXIT_VMCALL: VMEXIT_CALL_RET(0, hmR0VmxExitVmcall(pVCpu, pVmxTransient));
13005	case VMX_EXIT_MOV_DRX: VMEXIT_CALL_RET(0, hmR0VmxExitMovDRx(pVCpu, pVmxTransient));
13006	case VMX_EXIT_HLT: VMEXIT_CALL_RET(0, hmR0VmxExitHlt(pVCpu, pVmxTransient));
13007	case VMX_EXIT_INVD: VMEXIT_CALL_RET(0, hmR0VmxExitInvd(pVCpu, pVmxTransient));
13008	case VMX_EXIT_INVLPG: VMEXIT_CALL_RET(0, hmR0VmxExitInvlpg(pVCpu, pVmxTransient));
13009	case VMX_EXIT_MTF: VMEXIT_CALL_RET(0, hmR0VmxExitMtf(pVCpu, pVmxTransient));
13010	case VMX_EXIT_PAUSE: VMEXIT_CALL_RET(0, hmR0VmxExitPause(pVCpu, pVmxTransient));
13011	case VMX_EXIT_WBINVD: VMEXIT_CALL_RET(0, hmR0VmxExitWbinvd(pVCpu, pVmxTransient));
13012	case VMX_EXIT_XSETBV: VMEXIT_CALL_RET(0, hmR0VmxExitXsetbv(pVCpu, pVmxTransient));
13013	case VMX_EXIT_INVPCID: VMEXIT_CALL_RET(0, hmR0VmxExitInvpcid(pVCpu, pVmxTransient));
13014	case VMX_EXIT_GETSEC: VMEXIT_CALL_RET(0, hmR0VmxExitGetsec(pVCpu, pVmxTransient));
13015	case VMX_EXIT_RDPMC: VMEXIT_CALL_RET(0, hmR0VmxExitRdpmc(pVCpu, pVmxTransient));
13016	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13017	case VMX_EXIT_VMCLEAR: VMEXIT_CALL_RET(0, hmR0VmxExitVmclear(pVCpu, pVmxTransient));
13018	case VMX_EXIT_VMLAUNCH: VMEXIT_CALL_RET(0, hmR0VmxExitVmlaunch(pVCpu, pVmxTransient));
13019	case VMX_EXIT_VMPTRLD: VMEXIT_CALL_RET(0, hmR0VmxExitVmptrld(pVCpu, pVmxTransient));
13020	case VMX_EXIT_VMPTRST: VMEXIT_CALL_RET(0, hmR0VmxExitVmptrst(pVCpu, pVmxTransient));
13021	case VMX_EXIT_VMREAD: VMEXIT_CALL_RET(0, hmR0VmxExitVmread(pVCpu, pVmxTransient));
13022	case VMX_EXIT_VMRESUME: VMEXIT_CALL_RET(0, hmR0VmxExitVmwrite(pVCpu, pVmxTransient));
13023	case VMX_EXIT_VMWRITE: VMEXIT_CALL_RET(0, hmR0VmxExitVmresume(pVCpu, pVmxTransient));
13024	case VMX_EXIT_VMXOFF: VMEXIT_CALL_RET(0, hmR0VmxExitVmxoff(pVCpu, pVmxTransient));
13025	case VMX_EXIT_VMXON: VMEXIT_CALL_RET(0, hmR0VmxExitVmxon(pVCpu, pVmxTransient));
13026	case VMX_EXIT_INVVPID: VMEXIT_CALL_RET(0, hmR0VmxExitInvvpid(pVCpu, pVmxTransient));
13027	case VMX_EXIT_INVEPT: VMEXIT_CALL_RET(0, hmR0VmxExitSetPendingXcptUD(pVCpu, pVmxTransient));
13028	#else
13029	case VMX_EXIT_VMCLEAR:
13030	case VMX_EXIT_VMLAUNCH:
13031	case VMX_EXIT_VMPTRLD:
13032	case VMX_EXIT_VMPTRST:
13033	case VMX_EXIT_VMREAD:
13034	case VMX_EXIT_VMRESUME:
13035	case VMX_EXIT_VMWRITE:
13036	case VMX_EXIT_VMXOFF:
13037	case VMX_EXIT_VMXON:
13038	case VMX_EXIT_INVVPID:
13039	case VMX_EXIT_INVEPT:
13040	return hmR0VmxExitSetPendingXcptUD(pVCpu, pVmxTransient);
13041	#endif
13042
13043	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFault(pVCpu, pVmxTransient);
13044	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindow(pVCpu, pVmxTransient);
13045	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestState(pVCpu, pVmxTransient);
13046
13047	case VMX_EXIT_INIT_SIGNAL:
13048	case VMX_EXIT_SIPI:
13049	case VMX_EXIT_IO_SMI:
13050	case VMX_EXIT_SMI:
13051	case VMX_EXIT_ERR_MSR_LOAD:
13052	case VMX_EXIT_ERR_MACHINE_CHECK:
13053	case VMX_EXIT_PML_FULL:
13054	case VMX_EXIT_VIRTUALIZED_EOI:
13055	case VMX_EXIT_GDTR_IDTR_ACCESS:
13056	case VMX_EXIT_LDTR_TR_ACCESS:
13057	case VMX_EXIT_APIC_WRITE:
13058	case VMX_EXIT_RDRAND:
13059	case VMX_EXIT_RSM:
13060	case VMX_EXIT_VMFUNC:
13061	case VMX_EXIT_ENCLS:
13062	case VMX_EXIT_RDSEED:
13063	case VMX_EXIT_XSAVES:
13064	case VMX_EXIT_XRSTORS:
13065	case VMX_EXIT_UMWAIT:
13066	case VMX_EXIT_TPAUSE:
13067	default:
13068	return hmR0VmxExitErrUnexpected(pVCpu, pVmxTransient);
13069	}
13070	#undef VMEXIT_CALL_RET
13071	}
13072	#endif /* !HMVMX_USE_FUNCTION_TABLE */
13073
13074
13075	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13076	/**
13077	* Handles a nested-guest VM-exit from hardware-assisted VMX execution.
13078	*
13079	* @returns Strict VBox status code (i.e. informational status codes too).
13080	* @param pVCpu The cross context virtual CPU structure.
13081	* @param pVmxTransient The VMX-transient structure.
13082	*/
13083	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13084	{
13085	uint32_t const uExitReason = pVmxTransient->uExitReason;
13086	switch (uExitReason)
13087	{
13088	case VMX_EXIT_EPT_MISCONFIG: return hmR0VmxExitEptMisconfig(pVCpu, pVmxTransient);
13089	case VMX_EXIT_EPT_VIOLATION: return hmR0VmxExitEptViolation(pVCpu, pVmxTransient);
13090	case VMX_EXIT_XCPT_OR_NMI: return hmR0VmxExitXcptOrNmiNested(pVCpu, pVmxTransient);
13091	case VMX_EXIT_IO_INSTR: return hmR0VmxExitIoInstrNested(pVCpu, pVmxTransient);
13092	case VMX_EXIT_HLT: return hmR0VmxExitHltNested(pVCpu, pVmxTransient);
13093
13094	/*
13095	* We shouldn't direct host physical interrupts to the nested-guest.
13096	*/
13097	case VMX_EXIT_EXT_INT:
13098	return hmR0VmxExitExtInt(pVCpu, pVmxTransient);
13099
13100	/*
13101	* Instructions that cause VM-exits unconditionally or the condition is
13102	* always is taken solely from the nested hypervisor (meaning if the VM-exit
13103	* happens, it's guaranteed to be a nested-guest VM-exit).
13104	*
13105	* - Provides VM-exit instruction length ONLY.
13106	*/
13107	case VMX_EXIT_CPUID: /* Unconditional. */
13108	case VMX_EXIT_VMCALL:
13109	case VMX_EXIT_GETSEC:
13110	case VMX_EXIT_INVD:
13111	case VMX_EXIT_XSETBV:
13112	case VMX_EXIT_VMLAUNCH:
13113	case VMX_EXIT_VMRESUME:
13114	case VMX_EXIT_VMXOFF:
13115	case VMX_EXIT_ENCLS: /* Condition specified solely by nested hypervisor. */
13116	case VMX_EXIT_VMFUNC:
13117	return hmR0VmxExitInstrNested(pVCpu, pVmxTransient);
13118
13119	/*
13120	* Instructions that cause VM-exits unconditionally or the condition is
13121	* always is taken solely from the nested hypervisor (meaning if the VM-exit
13122	* happens, it's guaranteed to be a nested-guest VM-exit).
13123	*
13124	* - Provides VM-exit instruction length.
13125	* - Provides VM-exit information.
13126	* - Optionally provides Exit qualification.
13127	*
13128	* Since Exit qualification is 0 for all VM-exits where it is not
13129	* applicable, reading and passing it to the guest should produce
13130	* defined behavior.
13131	*
13132	* See Intel spec. 27.2.1 "Basic VM-Exit Information".
13133	*/
13134	case VMX_EXIT_INVEPT: /* Unconditional. */
13135	case VMX_EXIT_INVVPID:
13136	case VMX_EXIT_VMCLEAR:
13137	case VMX_EXIT_VMPTRLD:
13138	case VMX_EXIT_VMPTRST:
13139	case VMX_EXIT_VMXON:
13140	case VMX_EXIT_GDTR_IDTR_ACCESS: /* Condition specified solely by nested hypervisor. */
13141	case VMX_EXIT_LDTR_TR_ACCESS:
13142	case VMX_EXIT_RDRAND:
13143	case VMX_EXIT_RDSEED:
13144	case VMX_EXIT_XSAVES:
13145	case VMX_EXIT_XRSTORS:
13146	case VMX_EXIT_UMWAIT:
13147	case VMX_EXIT_TPAUSE:
13148	return hmR0VmxExitInstrWithInfoNested(pVCpu, pVmxTransient);
13149
13150	case VMX_EXIT_RDTSC: return hmR0VmxExitRdtscNested(pVCpu, pVmxTransient);
13151	case VMX_EXIT_RDTSCP: return hmR0VmxExitRdtscpNested(pVCpu, pVmxTransient);
13152	case VMX_EXIT_RDMSR: return hmR0VmxExitRdmsrNested(pVCpu, pVmxTransient);
13153	case VMX_EXIT_WRMSR: return hmR0VmxExitWrmsrNested(pVCpu, pVmxTransient);
13154	case VMX_EXIT_INVLPG: return hmR0VmxExitInvlpgNested(pVCpu, pVmxTransient);
13155	case VMX_EXIT_INVPCID: return hmR0VmxExitInvpcidNested(pVCpu, pVmxTransient);
13156	case VMX_EXIT_TASK_SWITCH: return hmR0VmxExitTaskSwitchNested(pVCpu, pVmxTransient);
13157	case VMX_EXIT_WBINVD: return hmR0VmxExitWbinvdNested(pVCpu, pVmxTransient);
13158	case VMX_EXIT_MTF: return hmR0VmxExitMtfNested(pVCpu, pVmxTransient);
13159	case VMX_EXIT_APIC_ACCESS: return hmR0VmxExitApicAccessNested(pVCpu, pVmxTransient);
13160	case VMX_EXIT_APIC_WRITE: return hmR0VmxExitApicWriteNested(pVCpu, pVmxTransient);
13161	case VMX_EXIT_VIRTUALIZED_EOI: return hmR0VmxExitVirtEoiNested(pVCpu, pVmxTransient);
13162	case VMX_EXIT_MOV_CRX: return hmR0VmxExitMovCRxNested(pVCpu, pVmxTransient);
13163	case VMX_EXIT_INT_WINDOW: return hmR0VmxExitIntWindowNested(pVCpu, pVmxTransient);
13164	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindowNested(pVCpu, pVmxTransient);
13165	case VMX_EXIT_TPR_BELOW_THRESHOLD: return hmR0VmxExitTprBelowThresholdNested(pVCpu, pVmxTransient);
13166	case VMX_EXIT_MWAIT: return hmR0VmxExitMwaitNested(pVCpu, pVmxTransient);
13167	case VMX_EXIT_MONITOR: return hmR0VmxExitMonitorNested(pVCpu, pVmxTransient);
13168	case VMX_EXIT_PAUSE: return hmR0VmxExitPauseNested(pVCpu, pVmxTransient);
13169
13170	case VMX_EXIT_PREEMPT_TIMER:
13171	{
13172	/** @todo NSTVMX: Preempt timer. */
13173	return hmR0VmxExitPreemptTimer(pVCpu, pVmxTransient);
13174	}
13175
13176	case VMX_EXIT_MOV_DRX: return hmR0VmxExitMovDRxNested(pVCpu, pVmxTransient);
13177	case VMX_EXIT_RDPMC: return hmR0VmxExitRdpmcNested(pVCpu, pVmxTransient);
13178
13179	case VMX_EXIT_VMREAD:
13180	case VMX_EXIT_VMWRITE: return hmR0VmxExitVmreadVmwriteNested(pVCpu, pVmxTransient);
13181
13182	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFaultNested(pVCpu, pVmxTransient);
13183	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestStateNested(pVCpu, pVmxTransient);
13184
13185	case VMX_EXIT_INIT_SIGNAL:
13186	case VMX_EXIT_SIPI:
13187	case VMX_EXIT_IO_SMI:
13188	case VMX_EXIT_SMI:
13189	case VMX_EXIT_ERR_MSR_LOAD:
13190	case VMX_EXIT_ERR_MACHINE_CHECK:
13191	case VMX_EXIT_PML_FULL:
13192	case VMX_EXIT_RSM:
13193	default:
13194	return hmR0VmxExitErrUnexpected(pVCpu, pVmxTransient);
13195	}
13196	}
13197	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13198
13199
13200	/** @name VM-exit helpers.
13201	* @{
13202	*/
13203	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13204	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= VM-exit helpers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
13205	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13206
13207	/** Macro for VM-exits called unexpectedly. */
13208	#define HMVMX_UNEXPECTED_EXIT_RET(a_pVCpu, a_HmError) \
13209	do { \
13210	(a_pVCpu)->hm.s.u32HMError = (a_HmError); \
13211	return VERR_VMX_UNEXPECTED_EXIT; \
13212	} while (0)
13213
13214	#ifdef VBOX_STRICT
13215	/* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */
13216	# define HMVMX_ASSERT_PREEMPT_CPUID_VAR() \
13217	RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId()
13218
13219	# define HMVMX_ASSERT_PREEMPT_CPUID() \
13220	do { \
13221	RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \
13222	AssertMsg(idAssertCpu == idAssertCpuNow, ("VMX %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \
13223	} while (0)
13224
13225	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13226	do { \
13227	AssertPtr((a_pVCpu)); \
13228	AssertPtr((a_pVmxTransient)); \
13229	Assert((a_pVmxTransient)->fVMEntryFailed == false); \
13230	Assert((a_pVmxTransient)->pVmcsInfo); \
13231	Assert(ASMIntAreEnabled()); \
13232	HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu); \
13233	HMVMX_ASSERT_PREEMPT_CPUID_VAR(); \
13234	Log4Func(("vcpu[%RU32]\n", (a_pVCpu)->idCpu)); \
13235	HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu); \
13236	if (VMMR0IsLogFlushDisabled((a_pVCpu))) \
13237	HMVMX_ASSERT_PREEMPT_CPUID(); \
13238	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
13239	} while (0)
13240
13241	# define HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13242	do { \
13243	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient); \
13244	Assert((a_pVmxTransient)->fIsNestedGuest); \
13245	} while (0)
13246
13247	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13248	do { \
13249	Log4Func(("\n")); \
13250	} while (0)
13251	#else
13252	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13253	do { \
13254	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
13255	NOREF((a_pVCpu)); NOREF((a_pVmxTransient)); \
13256	} while (0)
13257
13258	# define HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13259	do { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient); } while (0)
13260
13261	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) do { } while (0)
13262	#endif
13263
13264	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13265	/** Macro that does the necessary privilege checks and intercepted VM-exits for
13266	* guests that attempted to execute a VMX instruction. */
13267	# define HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(a_pVCpu, a_uExitReason) \
13268	do \
13269	{ \
13270	VBOXSTRICTRC rcStrictTmp = hmR0VmxCheckExitDueToVmxInstr((a_pVCpu), (a_uExitReason)); \
13271	if (rcStrictTmp == VINF_SUCCESS) \
13272	{ /* likely */ } \
13273	else if (rcStrictTmp == VINF_HM_PENDING_XCPT) \
13274	{ \
13275	Assert((a_pVCpu)->hm.s.Event.fPending); \
13276	Log4Func(("Privilege checks failed -> %#x\n", VMX_ENTRY_INT_INFO_VECTOR((a_pVCpu)->hm.s.Event.u64IntInfo))); \
13277	return VINF_SUCCESS; \
13278	} \
13279	else \
13280	{ \
13281	int rcTmp = VBOXSTRICTRC_VAL(rcStrictTmp); \
13282	AssertMsgFailedReturn(("Unexpected failure. rc=%Rrc", rcTmp), rcTmp); \
13283	} \
13284	} while (0)
13285
13286	/** Macro that decodes a memory operand for an VM-exit caused by an instruction. */
13287	# define HMVMX_DECODE_MEM_OPERAND(a_pVCpu, a_uExitInstrInfo, a_uExitQual, a_enmMemAccess, a_pGCPtrEffAddr) \
13288	do \
13289	{ \
13290	VBOXSTRICTRC rcStrictTmp = hmR0VmxDecodeMemOperand((a_pVCpu), (a_uExitInstrInfo), (a_uExitQual), (a_enmMemAccess), \
13291	(a_pGCPtrEffAddr)); \
13292	if (rcStrictTmp == VINF_SUCCESS) \
13293	{ /* likely */ } \
13294	else if (rcStrictTmp == VINF_HM_PENDING_XCPT) \
13295	{ \
13296	uint8_t const uXcptTmp = VMX_ENTRY_INT_INFO_VECTOR((a_pVCpu)->hm.s.Event.u64IntInfo); \
13297	Log4Func(("Memory operand decoding failed, raising xcpt %#x\n", uXcptTmp)); \
13298	NOREF(uXcptTmp); \
13299	return VINF_SUCCESS; \
13300	} \
13301	else \
13302	{ \
13303	Log4Func(("hmR0VmxDecodeMemOperand failed. rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrictTmp))); \
13304	return rcStrictTmp; \
13305	} \
13306	} while (0)
13307	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13308
13309
13310	/**
13311	* Advances the guest RIP by the specified number of bytes.
13312	*
13313	* @param pVCpu The cross context virtual CPU structure.
13314	* @param cbInstr Number of bytes to advance the RIP by.
13315	*
13316	* @remarks No-long-jump zone!!!
13317	*/
13318	DECLINLINE(void) hmR0VmxAdvanceGuestRipBy(PVMCPUCC pVCpu, uint32_t cbInstr)
13319	{
13320	/* Advance the RIP. */
13321	pVCpu->cpum.GstCtx.rip += cbInstr;
13322	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
13323
13324	/* Update interrupt inhibition. */
13325	if ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)
13326	&& pVCpu->cpum.GstCtx.rip != EMGetInhibitInterruptsPC(pVCpu))
13327	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
13328	}
13329
13330
13331	/**
13332	* Advances the guest RIP after reading it from the VMCS.
13333	*
13334	* @returns VBox status code, no informational status codes.
13335	* @param pVCpu The cross context virtual CPU structure.
13336	* @param pVmxTransient The VMX-transient structure.
13337	*
13338	* @remarks No-long-jump zone!!!
13339	*/
13340	static int hmR0VmxAdvanceGuestRip(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13341	{
13342	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13343	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS);
13344	AssertRCReturn(rc, rc);
13345
13346	hmR0VmxAdvanceGuestRipBy(pVCpu, pVmxTransient->cbExitInstr);
13347	return VINF_SUCCESS;
13348	}
13349
13350
13351	/**
13352	* Handle a condition that occurred while delivering an event through the guest or
13353	* nested-guest IDT.
13354	*
13355	* @returns Strict VBox status code (i.e. informational status codes too).
13356	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
13357	* @retval VINF_HM_DOUBLE_FAULT if a \#DF condition was detected and we ought
13358	* to continue execution of the guest which will delivery the \#DF.
13359	* @retval VINF_EM_RESET if we detected a triple-fault condition.
13360	* @retval VERR_EM_GUEST_CPU_HANG if we detected a guest CPU hang.
13361	*
13362	* @param pVCpu The cross context virtual CPU structure.
13363	* @param pVmxTransient The VMX-transient structure.
13364	*
13365	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13366	* Additionally, HMVMX_READ_EXIT_QUALIFICATION is required if the VM-exit
13367	* is due to an EPT violation, PML full or SPP-related event.
13368	*
13369	* @remarks No-long-jump zone!!!
13370	*/
13371	static VBOXSTRICTRC hmR0VmxCheckExitDueToEventDelivery(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13372	{
13373	Assert(!pVCpu->hm.s.Event.fPending);
13374	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_XCPT_INFO);
13375	if ( pVmxTransient->uExitReason == VMX_EXIT_EPT_VIOLATION
13376	\|\| pVmxTransient->uExitReason == VMX_EXIT_PML_FULL
13377	\|\| pVmxTransient->uExitReason == VMX_EXIT_SPP_EVENT)
13378	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_EXIT_QUALIFICATION);
13379
13380	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
13381	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
13382	uint32_t const uIdtVectorInfo = pVmxTransient->uIdtVectoringInfo;
13383	uint32_t const uExitIntInfo = pVmxTransient->uExitIntInfo;
13384	if (VMX_IDT_VECTORING_INFO_IS_VALID(uIdtVectorInfo))
13385	{
13386	uint32_t const uIdtVector = VMX_IDT_VECTORING_INFO_VECTOR(uIdtVectorInfo);
13387	uint32_t const uIdtVectorType = VMX_IDT_VECTORING_INFO_TYPE(uIdtVectorInfo);
13388
13389	/*
13390	* If the event was a software interrupt (generated with INT n) or a software exception
13391	* (generated by INT3/INTO) or a privileged software exception (generated by INT1), we
13392	* can handle the VM-exit and continue guest execution which will re-execute the
13393	* instruction rather than re-injecting the exception, as that can cause premature
13394	* trips to ring-3 before injection and involve TRPM which currently has no way of
13395	* storing that these exceptions were caused by these instructions (ICEBP's #DB poses
13396	* the problem).
13397	*/
13398	IEMXCPTRAISE enmRaise;
13399	IEMXCPTRAISEINFO fRaiseInfo;
13400	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
13401	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
13402	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT)
13403	{
13404	enmRaise = IEMXCPTRAISE_REEXEC_INSTR;
13405	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
13406	}
13407	else if (VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo))
13408	{
13409	uint32_t const uExitVectorType = VMX_EXIT_INT_INFO_TYPE(uExitIntInfo);
13410	uint8_t const uExitVector = VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo);
13411	Assert(uExitVectorType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT);
13412
13413	uint32_t const fIdtVectorFlags = hmR0VmxGetIemXcptFlags(uIdtVector, uIdtVectorType);
13414	uint32_t const fExitVectorFlags = hmR0VmxGetIemXcptFlags(uExitVector, uExitVectorType);
13415
13416	enmRaise = IEMEvaluateRecursiveXcpt(pVCpu, fIdtVectorFlags, uIdtVector, fExitVectorFlags, uExitVector, &fRaiseInfo);
13417
13418	/* Determine a vectoring #PF condition, see comment in hmR0VmxExitXcptPF(). */
13419	if (fRaiseInfo & (IEMXCPTRAISEINFO_EXT_INT_PF \| IEMXCPTRAISEINFO_NMI_PF))
13420	{
13421	pVmxTransient->fVectoringPF = true;
13422	enmRaise = IEMXCPTRAISE_PREV_EVENT;
13423	}
13424	}
13425	else
13426	{
13427	/*
13428	* If an exception or hardware interrupt delivery caused an EPT violation/misconfig or APIC access
13429	* VM-exit, then the VM-exit interruption-information will not be valid and we end up here.
13430	* It is sufficient to reflect the original event to the guest after handling the VM-exit.
13431	*/
13432	Assert( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
13433	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
13434	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT);
13435	enmRaise = IEMXCPTRAISE_PREV_EVENT;
13436	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
13437	}
13438
13439	/*
13440	* On CPUs that support Virtual NMIs, if this VM-exit (be it an exception or EPT violation/misconfig
13441	* etc.) occurred while delivering the NMI, we need to clear the block-by-NMI field in the guest
13442	* interruptibility-state before re-delivering the NMI after handling the VM-exit. Otherwise the
13443	* subsequent VM-entry would fail, see @bugref{7445}.
13444	*
13445	* See Intel spec. 30.7.1.2 "Resuming Guest Software after Handling an Exception".
13446	*/
13447	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
13448	&& enmRaise == IEMXCPTRAISE_PREV_EVENT
13449	&& (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
13450	&& CPUMIsGuestNmiBlocking(pVCpu))
13451	{
13452	CPUMSetGuestNmiBlocking(pVCpu, false);
13453	}
13454
13455	switch (enmRaise)
13456	{
13457	case IEMXCPTRAISE_CURRENT_XCPT:
13458	{
13459	Log4Func(("IDT: Pending secondary Xcpt: idtinfo=%#RX64 exitinfo=%#RX64\n", uIdtVectorInfo, uExitIntInfo));
13460	Assert(rcStrict == VINF_SUCCESS);
13461	break;
13462	}
13463
13464	case IEMXCPTRAISE_PREV_EVENT:
13465	{
13466	uint32_t u32ErrCode;
13467	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(uIdtVectorInfo))
13468	u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
13469	else
13470	u32ErrCode = 0;
13471
13472	/* If uExitVector is #PF, CR2 value will be updated from the VMCS if it's a guest #PF, see hmR0VmxExitXcptPF(). */
13473	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflect);
13474	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(uIdtVectorInfo), 0 /* cbInstr */,
13475	u32ErrCode, pVCpu->cpum.GstCtx.cr2);
13476
13477	Log4Func(("IDT: Pending vectoring event %#RX64 Err=%#RX32\n", pVCpu->hm.s.Event.u64IntInfo,
13478	pVCpu->hm.s.Event.u32ErrCode));
13479	Assert(rcStrict == VINF_SUCCESS);
13480	break;
13481	}
13482
13483	case IEMXCPTRAISE_REEXEC_INSTR:
13484	Assert(rcStrict == VINF_SUCCESS);
13485	break;
13486
13487	case IEMXCPTRAISE_DOUBLE_FAULT:
13488	{
13489	/*
13490	* Determing a vectoring double #PF condition. Used later, when PGM evaluates the
13491	* second #PF as a guest #PF (and not a shadow #PF) and needs to be converted into a #DF.
13492	*/
13493	if (fRaiseInfo & IEMXCPTRAISEINFO_PF_PF)
13494	{
13495	pVmxTransient->fVectoringDoublePF = true;
13496	Log4Func(("IDT: Vectoring double #PF %#RX64 cr2=%#RX64\n", pVCpu->hm.s.Event.u64IntInfo,
13497	pVCpu->cpum.GstCtx.cr2));
13498	rcStrict = VINF_SUCCESS;
13499	}
13500	else
13501	{
13502	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectConvertDF);
13503	hmR0VmxSetPendingXcptDF(pVCpu);
13504	Log4Func(("IDT: Pending vectoring #DF %#RX64 uIdtVector=%#x uExitVector=%#x\n", pVCpu->hm.s.Event.u64IntInfo,
13505	uIdtVector, VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo)));
13506	rcStrict = VINF_HM_DOUBLE_FAULT;
13507	}
13508	break;
13509	}
13510
13511	case IEMXCPTRAISE_TRIPLE_FAULT:
13512	{
13513	Log4Func(("IDT: Pending vectoring triple-fault uIdt=%#x uExit=%#x\n", uIdtVector,
13514	VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo)));
13515	rcStrict = VINF_EM_RESET;
13516	break;
13517	}
13518
13519	case IEMXCPTRAISE_CPU_HANG:
13520	{
13521	Log4Func(("IDT: Bad guest! Entering CPU hang. fRaiseInfo=%#x\n", fRaiseInfo));
13522	rcStrict = VERR_EM_GUEST_CPU_HANG;
13523	break;
13524	}
13525
13526	default:
13527	{
13528	AssertMsgFailed(("IDT: vcpu[%RU32] Unexpected/invalid value! enmRaise=%#x\n", pVCpu->idCpu, enmRaise));
13529	rcStrict = VERR_VMX_IPE_2;
13530	break;
13531	}
13532	}
13533	}
13534	else if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
13535	&& !CPUMIsGuestNmiBlocking(pVCpu))
13536	{
13537	if ( VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo)
13538	&& VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo) != X86_XCPT_DF
13539	&& VMX_EXIT_INT_INFO_IS_NMI_UNBLOCK_IRET(uExitIntInfo))
13540	{
13541	/*
13542	* Execution of IRET caused a fault when NMI blocking was in effect (i.e we're in
13543	* the guest or nested-guest NMI handler). We need to set the block-by-NMI field so
13544	* that virtual NMIs remain blocked until the IRET execution is completed.
13545	*
13546	* See Intel spec. 31.7.1.2 "Resuming Guest Software After Handling An Exception".
13547	*/
13548	CPUMSetGuestNmiBlocking(pVCpu, true);
13549	Log4Func(("Set NMI blocking. uExitReason=%u\n", pVmxTransient->uExitReason));
13550	}
13551	else if ( pVmxTransient->uExitReason == VMX_EXIT_EPT_VIOLATION
13552	\|\| pVmxTransient->uExitReason == VMX_EXIT_PML_FULL
13553	\|\| pVmxTransient->uExitReason == VMX_EXIT_SPP_EVENT)
13554	{
13555	/*
13556	* Execution of IRET caused an EPT violation, page-modification log-full event or
13557	* SPP-related event VM-exit when NMI blocking was in effect (i.e. we're in the
13558	* guest or nested-guest NMI handler). We need to set the block-by-NMI field so
13559	* that virtual NMIs remain blocked until the IRET execution is completed.
13560	*
13561	* See Intel spec. 27.2.3 "Information about NMI unblocking due to IRET"
13562	*/
13563	if (VMX_EXIT_QUAL_EPT_IS_NMI_UNBLOCK_IRET(pVmxTransient->uExitQual))
13564	{
13565	CPUMSetGuestNmiBlocking(pVCpu, true);
13566	Log4Func(("Set NMI blocking. uExitReason=%u\n", pVmxTransient->uExitReason));
13567	}
13568	}
13569	}
13570
13571	Assert( rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_HM_DOUBLE_FAULT
13572	\|\| rcStrict == VINF_EM_RESET \|\| rcStrict == VERR_EM_GUEST_CPU_HANG);
13573	return rcStrict;
13574	}
13575
13576
13577	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13578	/**
13579	* Perform the relevant VMX instruction checks for VM-exits that occurred due to the
13580	* guest attempting to execute a VMX instruction.
13581	*
13582	* @returns Strict VBox status code (i.e. informational status codes too).
13583	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
13584	* @retval VINF_HM_PENDING_XCPT if an exception was raised.
13585	*
13586	* @param pVCpu The cross context virtual CPU structure.
13587	* @param uExitReason The VM-exit reason.
13588	*
13589	* @todo NSTVMX: Document other error codes when VM-exit is implemented.
13590	* @remarks No-long-jump zone!!!
13591	*/
13592	static VBOXSTRICTRC hmR0VmxCheckExitDueToVmxInstr(PVMCPUCC pVCpu, uint32_t uExitReason)
13593	{
13594	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS
13595	\| CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_EFER);
13596
13597	/*
13598	* The physical CPU would have already checked the CPU mode/code segment.
13599	* We shall just assert here for paranoia.
13600	* See Intel spec. 25.1.1 "Relative Priority of Faults and VM Exits".
13601	*/
13602	Assert(!CPUMIsGuestInRealOrV86ModeEx(&pVCpu->cpum.GstCtx));
13603	Assert( !CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx)
13604	\|\| CPUMIsGuestIn64BitCodeEx(&pVCpu->cpum.GstCtx));
13605
13606	if (uExitReason == VMX_EXIT_VMXON)
13607	{
13608	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
13609
13610	/*
13611	* We check CR4.VMXE because it is required to be always set while in VMX operation
13612	* by physical CPUs and our CR4 read-shadow is only consulted when executing specific
13613	* instructions (CLTS, LMSW, MOV CR, and SMSW) and thus doesn't affect CPU operation
13614	* otherwise (i.e. physical CPU won't automatically #UD if Cr4Shadow.VMXE is 0).
13615	*/
13616	if (!CPUMIsGuestVmxEnabled(&pVCpu->cpum.GstCtx))
13617	{
13618	Log4Func(("CR4.VMXE is not set -> #UD\n"));
13619	hmR0VmxSetPendingXcptUD(pVCpu);
13620	return VINF_HM_PENDING_XCPT;
13621	}
13622	}
13623	else if (!CPUMIsGuestInVmxRootMode(&pVCpu->cpum.GstCtx))
13624	{
13625	/*
13626	* The guest has not entered VMX operation but attempted to execute a VMX instruction
13627	* (other than VMXON), we need to raise a #UD.
13628	*/
13629	Log4Func(("Not in VMX root mode -> #UD\n"));
13630	hmR0VmxSetPendingXcptUD(pVCpu);
13631	return VINF_HM_PENDING_XCPT;
13632	}
13633
13634	/* All other checks (including VM-exit intercepts) are handled by IEM instruction emulation. */
13635	return VINF_SUCCESS;
13636	}
13637
13638
13639	/**
13640	* Decodes the memory operand of an instruction that caused a VM-exit.
13641	*
13642	* The Exit qualification field provides the displacement field for memory
13643	* operand instructions, if any.
13644	*
13645	* @returns Strict VBox status code (i.e. informational status codes too).
13646	* @retval VINF_SUCCESS if the operand was successfully decoded.
13647	* @retval VINF_HM_PENDING_XCPT if an exception was raised while decoding the
13648	* operand.
13649	* @param pVCpu The cross context virtual CPU structure.
13650	* @param uExitInstrInfo The VM-exit instruction information field.
13651	* @param enmMemAccess The memory operand's access type (read or write).
13652	* @param GCPtrDisp The instruction displacement field, if any. For
13653	* RIP-relative addressing pass RIP + displacement here.
13654	* @param pGCPtrMem Where to store the effective destination memory address.
13655	*
13656	* @remarks Warning! This function ASSUMES the instruction cannot be used in real or
13657	* virtual-8086 mode hence skips those checks while verifying if the
13658	* segment is valid.
13659	*/
13660	static VBOXSTRICTRC hmR0VmxDecodeMemOperand(PVMCPUCC pVCpu, uint32_t uExitInstrInfo, RTGCPTR GCPtrDisp, VMXMEMACCESS enmMemAccess,
13661	PRTGCPTR pGCPtrMem)
13662	{
13663	Assert(pGCPtrMem);
13664	Assert(!CPUMIsGuestInRealOrV86Mode(pVCpu));
13665	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_EFER
13666	\| CPUMCTX_EXTRN_CR0);
13667
13668	static uint64_t const s_auAddrSizeMasks[] = { UINT64_C(0xffff), UINT64_C(0xffffffff), UINT64_C(0xffffffffffffffff) };
13669	static uint64_t const s_auAccessSizeMasks[] = { sizeof(uint16_t), sizeof(uint32_t), sizeof(uint64_t) };
13670	AssertCompile(RT_ELEMENTS(s_auAccessSizeMasks) == RT_ELEMENTS(s_auAddrSizeMasks));
13671
13672	VMXEXITINSTRINFO ExitInstrInfo;
13673	ExitInstrInfo.u = uExitInstrInfo;
13674	uint8_t const uAddrSize = ExitInstrInfo.All.u3AddrSize;
13675	uint8_t const iSegReg = ExitInstrInfo.All.iSegReg;
13676	bool const fIdxRegValid = !ExitInstrInfo.All.fIdxRegInvalid;
13677	uint8_t const iIdxReg = ExitInstrInfo.All.iIdxReg;
13678	uint8_t const uScale = ExitInstrInfo.All.u2Scaling;
13679	bool const fBaseRegValid = !ExitInstrInfo.All.fBaseRegInvalid;
13680	uint8_t const iBaseReg = ExitInstrInfo.All.iBaseReg;
13681	bool const fIsMemOperand = !ExitInstrInfo.All.fIsRegOperand;
13682	bool const fIsLongMode = CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx);
13683
13684	/*
13685	* Validate instruction information.
13686	* This shouldn't happen on real hardware but useful while testing our nested hardware-virtualization code.
13687	*/
13688	AssertLogRelMsgReturn(uAddrSize < RT_ELEMENTS(s_auAddrSizeMasks),
13689	("Invalid address size. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_1);
13690	AssertLogRelMsgReturn(iSegReg < X86_SREG_COUNT,
13691	("Invalid segment register. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_2);
13692	AssertLogRelMsgReturn(fIsMemOperand,
13693	("Expected memory operand. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_3);
13694
13695	/*
13696	* Compute the complete effective address.
13697	*
13698	* See AMD instruction spec. 1.4.2 "SIB Byte Format"
13699	* See AMD spec. 4.5.2 "Segment Registers".
13700	*/
13701	RTGCPTR GCPtrMem = GCPtrDisp;
13702	if (fBaseRegValid)
13703	GCPtrMem += pVCpu->cpum.GstCtx.aGRegs[iBaseReg].u64;
13704	if (fIdxRegValid)
13705	GCPtrMem += pVCpu->cpum.GstCtx.aGRegs[iIdxReg].u64 << uScale;
13706
13707	RTGCPTR const GCPtrOff = GCPtrMem;
13708	if ( !fIsLongMode
13709	\|\| iSegReg >= X86_SREG_FS)
13710	GCPtrMem += pVCpu->cpum.GstCtx.aSRegs[iSegReg].u64Base;
13711	GCPtrMem &= s_auAddrSizeMasks[uAddrSize];
13712
13713	/*
13714	* Validate effective address.
13715	* See AMD spec. 4.5.3 "Segment Registers in 64-Bit Mode".
13716	*/
13717	uint8_t const cbAccess = s_auAccessSizeMasks[uAddrSize];
13718	Assert(cbAccess > 0);
13719	if (fIsLongMode)
13720	{
13721	if (X86_IS_CANONICAL(GCPtrMem))
13722	{
13723	*pGCPtrMem = GCPtrMem;
13724	return VINF_SUCCESS;
13725	}
13726
13727	/** @todo r=ramshankar: We should probably raise \#SS or \#GP. See AMD spec. 4.12.2
13728	* "Data Limit Checks in 64-bit Mode". */
13729	Log4Func(("Long mode effective address is not canonical GCPtrMem=%#RX64\n", GCPtrMem));
13730	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13731	return VINF_HM_PENDING_XCPT;
13732	}
13733
13734	/*
13735	* This is a watered down version of iemMemApplySegment().
13736	* Parts that are not applicable for VMX instructions like real-or-v8086 mode
13737	* and segment CPL/DPL checks are skipped.
13738	*/
13739	RTGCPTR32 const GCPtrFirst32 = (RTGCPTR32)GCPtrOff;
13740	RTGCPTR32 const GCPtrLast32 = GCPtrFirst32 + cbAccess - 1;
13741	PCCPUMSELREG pSel = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
13742
13743	/* Check if the segment is present and usable. */
13744	if ( pSel->Attr.n.u1Present
13745	&& !pSel->Attr.n.u1Unusable)
13746	{
13747	Assert(pSel->Attr.n.u1DescType);
13748	if (!(pSel->Attr.n.u4Type & X86_SEL_TYPE_CODE))
13749	{
13750	/* Check permissions for the data segment. */
13751	if ( enmMemAccess == VMXMEMACCESS_WRITE
13752	&& !(pSel->Attr.n.u4Type & X86_SEL_TYPE_WRITE))
13753	{
13754	Log4Func(("Data segment access invalid. iSegReg=%#x Attr=%#RX32\n", iSegReg, pSel->Attr.u));
13755	hmR0VmxSetPendingXcptGP(pVCpu, iSegReg);
13756	return VINF_HM_PENDING_XCPT;
13757	}
13758
13759	/* Check limits if it's a normal data segment. */
13760	if (!(pSel->Attr.n.u4Type & X86_SEL_TYPE_DOWN))
13761	{
13762	if ( GCPtrFirst32 > pSel->u32Limit
13763	\|\| GCPtrLast32 > pSel->u32Limit)
13764	{
13765	Log4Func(("Data segment limit exceeded. "
13766	"iSegReg=%#x GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n", iSegReg, GCPtrFirst32,
13767	GCPtrLast32, pSel->u32Limit));
13768	if (iSegReg == X86_SREG_SS)
13769	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13770	else
13771	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13772	return VINF_HM_PENDING_XCPT;
13773	}
13774	}
13775	else
13776	{
13777	/* Check limits if it's an expand-down data segment.
13778	Note! The upper boundary is defined by the B bit, not the G bit! */
13779	if ( GCPtrFirst32 < pSel->u32Limit + UINT32_C(1)
13780	\|\| GCPtrLast32 > (pSel->Attr.n.u1DefBig ? UINT32_MAX : UINT32_C(0xffff)))
13781	{
13782	Log4Func(("Expand-down data segment limit exceeded. "
13783	"iSegReg=%#x GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n", iSegReg, GCPtrFirst32,
13784	GCPtrLast32, pSel->u32Limit));
13785	if (iSegReg == X86_SREG_SS)
13786	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13787	else
13788	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13789	return VINF_HM_PENDING_XCPT;
13790	}
13791	}
13792	}
13793	else
13794	{
13795	/* Check permissions for the code segment. */
13796	if ( enmMemAccess == VMXMEMACCESS_WRITE
13797	\|\| ( enmMemAccess == VMXMEMACCESS_READ
13798	&& !(pSel->Attr.n.u4Type & X86_SEL_TYPE_READ)))
13799	{
13800	Log4Func(("Code segment access invalid. Attr=%#RX32\n", pSel->Attr.u));
13801	Assert(!CPUMIsGuestInRealOrV86ModeEx(&pVCpu->cpum.GstCtx));
13802	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13803	return VINF_HM_PENDING_XCPT;
13804	}
13805
13806	/* Check limits for the code segment (normal/expand-down not applicable for code segments). */
13807	if ( GCPtrFirst32 > pSel->u32Limit
13808	\|\| GCPtrLast32 > pSel->u32Limit)
13809	{
13810	Log4Func(("Code segment limit exceeded. GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n",
13811	GCPtrFirst32, GCPtrLast32, pSel->u32Limit));
13812	if (iSegReg == X86_SREG_SS)
13813	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13814	else
13815	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13816	return VINF_HM_PENDING_XCPT;
13817	}
13818	}
13819	}
13820	else
13821	{
13822	Log4Func(("Not present or unusable segment. iSegReg=%#x Attr=%#RX32\n", iSegReg, pSel->Attr.u));
13823	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13824	return VINF_HM_PENDING_XCPT;
13825	}
13826
13827	*pGCPtrMem = GCPtrMem;
13828	return VINF_SUCCESS;
13829	}
13830	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13831
13832
13833	/**
13834	* VM-exit helper for LMSW.
13835	*/
13836	static VBOXSTRICTRC hmR0VmxExitLmsw(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint16_t uMsw, RTGCPTR GCPtrEffDst)
13837	{
13838	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13839	AssertRCReturn(rc, rc);
13840
13841	VBOXSTRICTRC rcStrict = IEMExecDecodedLmsw(pVCpu, cbInstr, uMsw, GCPtrEffDst);
13842	AssertMsg( rcStrict == VINF_SUCCESS
13843	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13844
13845	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
13846	if (rcStrict == VINF_IEM_RAISED_XCPT)
13847	{
13848	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13849	rcStrict = VINF_SUCCESS;
13850	}
13851
13852	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitLmsw);
13853	Log4Func(("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13854	return rcStrict;
13855	}
13856
13857
13858	/**
13859	* VM-exit helper for CLTS.
13860	*/
13861	static VBOXSTRICTRC hmR0VmxExitClts(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr)
13862	{
13863	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13864	AssertRCReturn(rc, rc);
13865
13866	VBOXSTRICTRC rcStrict = IEMExecDecodedClts(pVCpu, cbInstr);
13867	AssertMsg( rcStrict == VINF_SUCCESS
13868	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13869
13870	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
13871	if (rcStrict == VINF_IEM_RAISED_XCPT)
13872	{
13873	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13874	rcStrict = VINF_SUCCESS;
13875	}
13876
13877	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitClts);
13878	Log4Func(("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13879	return rcStrict;
13880	}
13881
13882
13883	/**
13884	* VM-exit helper for MOV from CRx (CRx read).
13885	*/
13886	static VBOXSTRICTRC hmR0VmxExitMovFromCrX(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint8_t iGReg, uint8_t iCrReg)
13887	{
13888	Assert(iCrReg < 16);
13889	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
13890
13891	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13892	AssertRCReturn(rc, rc);
13893
13894	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxRead(pVCpu, cbInstr, iGReg, iCrReg);
13895	AssertMsg( rcStrict == VINF_SUCCESS
13896	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13897
13898	if (iGReg == X86_GREG_xSP)
13899	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_RSP);
13900	else
13901	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
13902	#ifdef VBOX_WITH_STATISTICS
13903	switch (iCrReg)
13904	{
13905	case 0: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Read); break;
13906	case 2: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Read); break;
13907	case 3: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Read); break;
13908	case 4: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Read); break;
13909	case 8: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Read); break;
13910	}
13911	#endif
13912	Log4Func(("CR%d Read access rcStrict=%Rrc\n", iCrReg, VBOXSTRICTRC_VAL(rcStrict)));
13913	return rcStrict;
13914	}
13915
13916
13917	/**
13918	* VM-exit helper for MOV to CRx (CRx write).
13919	*/
13920	static VBOXSTRICTRC hmR0VmxExitMovToCrX(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint8_t iGReg, uint8_t iCrReg)
13921	{
13922	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13923	AssertRCReturn(rc, rc);
13924
13925	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxWrite(pVCpu, cbInstr, iCrReg, iGReg);
13926	AssertMsg( rcStrict == VINF_SUCCESS
13927	\|\| rcStrict == VINF_IEM_RAISED_XCPT
13928	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13929
13930	switch (iCrReg)
13931	{
13932	case 0:
13933	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0
13934	\| HM_CHANGED_GUEST_EFER_MSR \| HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
13935	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Write);
13936	Log4Func(("CR0 write. rcStrict=%Rrc CR0=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cr0));
13937	break;
13938
13939	case 2:
13940	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Write);
13941	/* Nothing to do here, CR2 it's not part of the VMCS. */
13942	break;
13943
13944	case 3:
13945	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR3);
13946	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Write);
13947	Log4Func(("CR3 write. rcStrict=%Rrc CR3=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cr3));
13948	break;
13949
13950	case 4:
13951	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR4);
13952	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Write);
13953	Log4Func(("CR4 write. rc=%Rrc CR4=%#RX64 fLoadSaveGuestXcr0=%u\n", VBOXSTRICTRC_VAL(rcStrict),
13954	pVCpu->cpum.GstCtx.cr4, pVCpu->hmr0.s.fLoadSaveGuestXcr0));
13955	break;
13956
13957	case 8:
13958	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged,
13959	HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_APIC_TPR);
13960	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Write);
13961	break;
13962
13963	default:
13964	AssertMsgFailed(("Invalid CRx register %#x\n", iCrReg));
13965	break;
13966	}
13967
13968	if (rcStrict == VINF_IEM_RAISED_XCPT)
13969	{
13970	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13971	rcStrict = VINF_SUCCESS;
13972	}
13973	return rcStrict;
13974	}
13975
13976
13977	/**
13978	* VM-exit exception handler for \#PF (Page-fault exception).
13979	*
13980	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13981	*/
13982	static VBOXSTRICTRC hmR0VmxExitXcptPF(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13983	{
13984	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13985	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
13986	hmR0VmxReadExitQualVmcs(pVmxTransient);
13987
13988	if (!pVM->hmr0.s.fNestedPaging)
13989	{ /* likely */ }
13990	else
13991	{
13992	#if !defined(HMVMX_ALWAYS_TRAP_ALL_XCPTS) && !defined(HMVMX_ALWAYS_TRAP_PF)
13993	Assert(pVmxTransient->fIsNestedGuest \|\| pVCpu->hmr0.s.fUsingDebugLoop);
13994	#endif
13995	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
13996	if (!pVmxTransient->fVectoringDoublePF)
13997	{
13998	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */,
13999	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual);
14000	}
14001	else
14002	{
14003	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
14004	Assert(!pVmxTransient->fIsNestedGuest);
14005	hmR0VmxSetPendingXcptDF(pVCpu);
14006	Log4Func(("Pending #DF due to vectoring #PF w/ NestedPaging\n"));
14007	}
14008	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
14009	return VINF_SUCCESS;
14010	}
14011
14012	Assert(!pVmxTransient->fIsNestedGuest);
14013
14014	/* If it's a vectoring #PF, emulate injecting the original event injection as PGMTrap0eHandler() is incapable
14015	of differentiating between instruction emulation and event injection that caused a #PF. See @bugref{6607}. */
14016	if (pVmxTransient->fVectoringPF)
14017	{
14018	Assert(pVCpu->hm.s.Event.fPending);
14019	return VINF_EM_RAW_INJECT_TRPM_EVENT;
14020	}
14021
14022	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14023	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14024	AssertRCReturn(rc, rc);
14025
14026	Log4Func(("#PF: cs:rip=%#04x:%#RX64 err_code=%#RX32 exit_qual=%#RX64 cr3=%#RX64\n", pCtx->cs.Sel, pCtx->rip,
14027	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual, pCtx->cr3));
14028
14029	TRPMAssertXcptPF(pVCpu, pVmxTransient->uExitQual, (RTGCUINT)pVmxTransient->uExitIntErrorCode);
14030	rc = PGMTrap0eHandler(pVCpu, pVmxTransient->uExitIntErrorCode, CPUMCTX2CORE(pCtx), (RTGCPTR)pVmxTransient->uExitQual);
14031
14032	Log4Func(("#PF: rc=%Rrc\n", rc));
14033	if (rc == VINF_SUCCESS)
14034	{
14035	/*
14036	* This is typically a shadow page table sync or a MMIO instruction. But we may have
14037	* emulated something like LTR or a far jump. Any part of the CPU context may have changed.
14038	*/
14039	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14040	TRPMResetTrap(pVCpu);
14041	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
14042	return rc;
14043	}
14044
14045	if (rc == VINF_EM_RAW_GUEST_TRAP)
14046	{
14047	if (!pVmxTransient->fVectoringDoublePF)
14048	{
14049	/* It's a guest page fault and needs to be reflected to the guest. */
14050	uint32_t const uGstErrorCode = TRPMGetErrorCode(pVCpu);
14051	TRPMResetTrap(pVCpu);
14052	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory #PF. */
14053	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */,
14054	uGstErrorCode, pVmxTransient->uExitQual);
14055	}
14056	else
14057	{
14058	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
14059	TRPMResetTrap(pVCpu);
14060	pVCpu->hm.s.Event.fPending = false; /* Clear pending #PF to replace it with #DF. */
14061	hmR0VmxSetPendingXcptDF(pVCpu);
14062	Log4Func(("#PF: Pending #DF due to vectoring #PF\n"));
14063	}
14064
14065	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
14066	return VINF_SUCCESS;
14067	}
14068
14069	TRPMResetTrap(pVCpu);
14070	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
14071	return rc;
14072	}
14073
14074
14075	/**
14076	* VM-exit exception handler for \#MF (Math Fault: floating point exception).
14077	*
14078	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14079	*/
14080	static VBOXSTRICTRC hmR0VmxExitXcptMF(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14081	{
14082	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14083	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
14084
14085	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR0);
14086	AssertRCReturn(rc, rc);
14087
14088	if (!(pVCpu->cpum.GstCtx.cr0 & X86_CR0_NE))
14089	{
14090	/* Convert a #MF into a FERR -> IRQ 13. See @bugref{6117}. */
14091	rc = PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13, 1, 0 /* uTagSrc */);
14092
14093	/** @todo r=ramshankar: The Intel spec. does -not- specify that this VM-exit
14094	* provides VM-exit instruction length. If this causes problem later,
14095	* disassemble the instruction like it's done on AMD-V. */
14096	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14097	AssertRCReturn(rc2, rc2);
14098	return rc;
14099	}
14100
14101	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbExitInstr,
14102	pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14103	return VINF_SUCCESS;
14104	}
14105
14106
14107	/**
14108	* VM-exit exception handler for \#BP (Breakpoint exception).
14109	*
14110	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14111	*/
14112	static VBOXSTRICTRC hmR0VmxExitXcptBP(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14113	{
14114	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14115	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP);
14116
14117	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14118	AssertRCReturn(rc, rc);
14119
14120	if (!pVmxTransient->fIsNestedGuest)
14121	rc = DBGFTrap03Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(&pVCpu->cpum.GstCtx));
14122	else
14123	rc = VINF_EM_RAW_GUEST_TRAP;
14124
14125	if (rc == VINF_EM_RAW_GUEST_TRAP)
14126	{
14127	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14128	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14129	rc = VINF_SUCCESS;
14130	}
14131
14132	Assert(rc == VINF_SUCCESS \|\| rc == VINF_EM_DBG_BREAKPOINT);
14133	return rc;
14134	}
14135
14136
14137	/**
14138	* VM-exit exception handler for \#AC (Alignment-check exception).
14139	*
14140	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14141	*/
14142	static VBOXSTRICTRC hmR0VmxExitXcptAC(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14143	{
14144	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14145
14146	/*
14147	* Detect #ACs caused by host having enabled split-lock detection.
14148	* Emulate such instructions.
14149	*/
14150	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo,
14151	CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS \| CPUMCTX_EXTRN_CS);
14152	AssertRCReturn(rc, rc);
14153	/** @todo detect split lock in cpu feature? */
14154	if ( /* 1. If 486-style alignment checks aren't enabled, then this must be a split-lock exception */
14155	!(pVCpu->cpum.GstCtx.cr0 & X86_CR0_AM)
14156	/* 2. #AC cannot happen in rings 0-2 except for split-lock detection. */
14157	\|\| CPUMGetGuestCPL(pVCpu) != 3
14158	/* 3. When the EFLAGS.AC != 0 this can only be a split-lock case. */
14159	\|\| !(pVCpu->cpum.GstCtx.eflags.u & X86_EFL_AC) )
14160	{
14161	/*
14162	* Check for debug/trace events and import state accordingly.
14163	*/
14164	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitGuestACSplitLock);
14165	PVMCC pVM = pVCpu->pVMR0;
14166	if ( !DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_VMX_SPLIT_LOCK)
14167	&& !VBOXVMM_VMX_SPLIT_LOCK_ENABLED())
14168	{
14169	if (pVM->cCpus == 1)
14170	{
14171	#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... */
14172	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14173	#else
14174	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14175	#endif
14176	AssertRCReturn(rc, rc);
14177	}
14178	}
14179	else
14180	{
14181	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14182	AssertRCReturn(rc, rc);
14183
14184	VBOXVMM_XCPT_DF(pVCpu, &pVCpu->cpum.GstCtx);
14185
14186	if (DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_VMX_SPLIT_LOCK))
14187	{
14188	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, DBGFEVENT_VMX_SPLIT_LOCK, DBGFEVENTCTX_HM, 0);
14189	if (rcStrict != VINF_SUCCESS)
14190	return rcStrict;
14191	}
14192	}
14193
14194	/*
14195	* Emulate the instruction.
14196	*
14197	* We have to ignore the LOCK prefix here as we must not retrigger the
14198	* detection on the host. This isn't all that satisfactory, though...
14199	*/
14200	if (pVM->cCpus == 1)
14201	{
14202	Log8Func(("cs:rip=%#04x:%#RX64 rflags=%#RX64 cr0=%#RX64 split-lock #AC\n", pVCpu->cpum.GstCtx.cs.Sel,
14203	pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.rflags, pVCpu->cpum.GstCtx.cr0));
14204
14205	/** @todo For SMP configs we should do a rendezvous here. */
14206	VBOXSTRICTRC rcStrict = IEMExecOneIgnoreLock(pVCpu);
14207	if (rcStrict == VINF_SUCCESS)
14208	#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... */
14209	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged,
14210	HM_CHANGED_GUEST_RIP
14211	\| HM_CHANGED_GUEST_RFLAGS
14212	\| HM_CHANGED_GUEST_GPRS_MASK
14213	\| HM_CHANGED_GUEST_CS
14214	\| HM_CHANGED_GUEST_SS);
14215	#else
14216	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14217	#endif
14218	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14219	{
14220	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14221	rcStrict = VINF_SUCCESS;
14222	}
14223	return rcStrict;
14224	}
14225	Log8Func(("cs:rip=%#04x:%#RX64 rflags=%#RX64 cr0=%#RX64 split-lock #AC -> VINF_EM_EMULATE_SPLIT_LOCK\n",
14226	pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.rflags, pVCpu->cpum.GstCtx.cr0));
14227	return VINF_EM_EMULATE_SPLIT_LOCK;
14228	}
14229
14230	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC);
14231	Log8Func(("cs:rip=%#04x:%#RX64 rflags=%#RX64 cr0=%#RX64 cpl=%d -> #AC\n", pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
14232	pVCpu->cpum.GstCtx.rflags, pVCpu->cpum.GstCtx.cr0, CPUMGetGuestCPL(pVCpu) ));
14233
14234	/* Re-inject it. We'll detect any nesting before getting here. */
14235	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14236	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14237	return VINF_SUCCESS;
14238	}
14239
14240
14241	/**
14242	* VM-exit exception handler for \#DB (Debug exception).
14243	*
14244	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14245	*/
14246	static VBOXSTRICTRC hmR0VmxExitXcptDB(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14247	{
14248	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14249	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
14250
14251	/*
14252	* Get the DR6-like values from the Exit qualification and pass it to DBGF for processing.
14253	*/
14254	hmR0VmxReadExitQualVmcs(pVmxTransient);
14255
14256	/* Refer Intel spec. Table 27-1. "Exit Qualifications for debug exceptions" for the format. */
14257	uint64_t const uDR6 = X86_DR6_INIT_VAL
14258	\| (pVmxTransient->uExitQual & ( X86_DR6_B0 \| X86_DR6_B1 \| X86_DR6_B2 \| X86_DR6_B3
14259	\| X86_DR6_BD \| X86_DR6_BS));
14260
14261	int rc;
14262	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14263	if (!pVmxTransient->fIsNestedGuest)
14264	{
14265	rc = DBGFTrap01Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx), uDR6, pVCpu->hm.s.fSingleInstruction);
14266
14267	/*
14268	* Prevents stepping twice over the same instruction when the guest is stepping using
14269	* EFLAGS.TF and the hypervisor debugger is stepping using MTF.
14270	* Testcase: DOSQEMM, break (using "ba x 1") at cs:rip 0x70:0x774 and step (using "t").
14271	*/
14272	if ( rc == VINF_EM_DBG_STEPPED
14273	&& (pVmxTransient->pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MONITOR_TRAP_FLAG))
14274	{
14275	Assert(pVCpu->hm.s.fSingleInstruction);
14276	rc = VINF_EM_RAW_GUEST_TRAP;
14277	}
14278	}
14279	else
14280	rc = VINF_EM_RAW_GUEST_TRAP;
14281	Log6Func(("rc=%Rrc\n", rc));
14282	if (rc == VINF_EM_RAW_GUEST_TRAP)
14283	{
14284	/*
14285	* The exception was for the guest. Update DR6, DR7.GD and
14286	* IA32_DEBUGCTL.LBR before forwarding it.
14287	* See Intel spec. 27.1 "Architectural State before a VM-Exit".
14288	*/
14289	VMMRZCallRing3Disable(pVCpu);
14290	HM_DISABLE_PREEMPT(pVCpu);
14291
14292	pCtx->dr[6] &= ~X86_DR6_B_MASK;
14293	pCtx->dr[6] \|= uDR6;
14294	if (CPUMIsGuestDebugStateActive(pVCpu))
14295	ASMSetDR6(pCtx->dr[6]);
14296
14297	HM_RESTORE_PREEMPT();
14298	VMMRZCallRing3Enable(pVCpu);
14299
14300	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_DR7);
14301	AssertRCReturn(rc, rc);
14302
14303	/* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */
14304	pCtx->dr[7] &= ~(uint64_t)X86_DR7_GD;
14305
14306	/* Paranoia. */
14307	pCtx->dr[7] &= ~(uint64_t)X86_DR7_RAZ_MASK;
14308	pCtx->dr[7] \|= X86_DR7_RA1_MASK;
14309
14310	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, pCtx->dr[7]);
14311	AssertRC(rc);
14312
14313	/*
14314	* Raise #DB in the guest.
14315	*
14316	* It is important to reflect exactly what the VM-exit gave us (preserving the
14317	* interruption-type) rather than use hmR0VmxSetPendingXcptDB() as the #DB could've
14318	* been raised while executing ICEBP (INT1) and not the regular #DB. Thus it may
14319	* trigger different handling in the CPU (like skipping DPL checks), see @bugref{6398}.
14320	*
14321	* Intel re-documented ICEBP/INT1 on May 2018 previously documented as part of
14322	* Intel 386, see Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
14323	*/
14324	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14325	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14326	return VINF_SUCCESS;
14327	}
14328
14329	/*
14330	* Not a guest trap, must be a hypervisor related debug event then.
14331	* Update DR6 in case someone is interested in it.
14332	*/
14333	AssertMsg(rc == VINF_EM_DBG_STEPPED \|\| rc == VINF_EM_DBG_BREAKPOINT, ("%Rrc\n", rc));
14334	AssertReturn(pVmxTransient->fWasHyperDebugStateActive, VERR_HM_IPE_5);
14335	CPUMSetHyperDR6(pVCpu, uDR6);
14336
14337	return rc;
14338	}
14339
14340
14341	/**
14342	* Hacks its way around the lovely mesa driver's backdoor accesses.
14343	*
14344	* @sa hmR0SvmHandleMesaDrvGp.
14345	*/
14346	static int hmR0VmxHandleMesaDrvGp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PCPUMCTX pCtx)
14347	{
14348	LogFunc(("cs:rip=%#04x:%#RX64 rcx=%#RX64 rbx=%#RX64\n", pCtx->cs.Sel, pCtx->rip, pCtx->rcx, pCtx->rbx));
14349	RT_NOREF(pCtx);
14350
14351	/* For now we'll just skip the instruction. */
14352	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14353	}
14354
14355
14356	/**
14357	* Checks if the \#GP'ing instruction is the mesa driver doing it's lovely
14358	* backdoor logging w/o checking what it is running inside.
14359	*
14360	* This recognizes an "IN EAX,DX" instruction executed in flat ring-3, with the
14361	* backdoor port and magic numbers loaded in registers.
14362	*
14363	* @returns true if it is, false if it isn't.
14364	* @sa hmR0SvmIsMesaDrvGp.
14365	*/
14366	DECLINLINE(bool) hmR0VmxIsMesaDrvGp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PCPUMCTX pCtx)
14367	{
14368	/* 0xed: IN eAX,dx */
14369	uint8_t abInstr[1];
14370	if (pVmxTransient->cbExitInstr != sizeof(abInstr))
14371	return false;
14372
14373	/* Check that it is #GP(0). */
14374	if (pVmxTransient->uExitIntErrorCode != 0)
14375	return false;
14376
14377	/* Check magic and port. */
14378	Assert(!(pCtx->fExtrn & (CPUMCTX_EXTRN_RAX \| CPUMCTX_EXTRN_RDX \| CPUMCTX_EXTRN_RCX)));
14379	/Log(("hmR0VmxIsMesaDrvGp: rax=%RX64 rdx=%RX64\n", pCtx->rax, pCtx->rdx));/
14380	if (pCtx->rax != UINT32_C(0x564d5868))
14381	return false;
14382	if (pCtx->dx != UINT32_C(0x5658))
14383	return false;
14384
14385	/* Flat ring-3 CS. */
14386	AssertCompile(HMVMX_CPUMCTX_EXTRN_ALL & CPUMCTX_EXTRN_CS);
14387	Assert(!(pCtx->fExtrn & CPUMCTX_EXTRN_CS));
14388	/Log(("hmR0VmxIsMesaDrvGp: cs.Attr.n.u2Dpl=%d base=%Rx64\n", pCtx->cs.Attr.n.u2Dpl, pCtx->cs.u64Base));/
14389	if (pCtx->cs.Attr.n.u2Dpl != 3)
14390	return false;
14391	if (pCtx->cs.u64Base != 0)
14392	return false;
14393
14394	/* Check opcode. */
14395	AssertCompile(HMVMX_CPUMCTX_EXTRN_ALL & CPUMCTX_EXTRN_RIP);
14396	Assert(!(pCtx->fExtrn & CPUMCTX_EXTRN_RIP));
14397	int rc = PGMPhysSimpleReadGCPtr(pVCpu, abInstr, pCtx->rip, sizeof(abInstr));
14398	/Log(("hmR0VmxIsMesaDrvGp: PGMPhysSimpleReadGCPtr -> %Rrc %#x\n", rc, abInstr[0]));/
14399	if (RT_FAILURE(rc))
14400	return false;
14401	if (abInstr[0] != 0xed)
14402	return false;
14403
14404	return true;
14405	}
14406
14407
14408	/**
14409	* VM-exit exception handler for \#GP (General-protection exception).
14410	*
14411	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14412	*/
14413	static VBOXSTRICTRC hmR0VmxExitXcptGP(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14414	{
14415	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14416	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP);
14417
14418	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14419	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14420	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
14421	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
14422	{ /* likely */ }
14423	else
14424	{
14425	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14426	Assert(pVCpu->hmr0.s.fUsingDebugLoop \|\| pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv \|\| pVmxTransient->fIsNestedGuest);
14427	#endif
14428	/*
14429	* If the guest is not in real-mode or we have unrestricted guest execution support, or if we are
14430	* executing a nested-guest, reflect #GP to the guest or nested-guest.
14431	*/
14432	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14433	AssertRCReturn(rc, rc);
14434	Log4Func(("Gst: cs:rip=%#04x:%#RX64 ErrorCode=%#x cr0=%#RX64 cpl=%u tr=%#04x\n", pCtx->cs.Sel, pCtx->rip,
14435	pVmxTransient->uExitIntErrorCode, pCtx->cr0, CPUMGetGuestCPL(pVCpu), pCtx->tr.Sel));
14436
14437	if ( pVmxTransient->fIsNestedGuest
14438	\|\| !pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv
14439	\|\| !hmR0VmxIsMesaDrvGp(pVCpu, pVmxTransient, pCtx))
14440	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14441	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14442	else
14443	rc = hmR0VmxHandleMesaDrvGp(pVCpu, pVmxTransient, pCtx);
14444	return rc;
14445	}
14446
14447	Assert(CPUMIsGuestInRealModeEx(pCtx));
14448	Assert(!pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUnrestrictedGuest);
14449	Assert(!pVmxTransient->fIsNestedGuest);
14450
14451	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14452	AssertRCReturn(rc, rc);
14453
14454	VBOXSTRICTRC rcStrict = IEMExecOne(pVCpu);
14455	if (rcStrict == VINF_SUCCESS)
14456	{
14457	if (!CPUMIsGuestInRealModeEx(pCtx))
14458	{
14459	/*
14460	* The guest is no longer in real-mode, check if we can continue executing the
14461	* guest using hardware-assisted VMX. Otherwise, fall back to emulation.
14462	*/
14463	pVmcsInfoShared->RealMode.fRealOnV86Active = false;
14464	if (HMCanExecuteVmxGuest(pVCpu->pVMR0, pVCpu, pCtx))
14465	{
14466	Log4Func(("Mode changed but guest still suitable for executing using hardware-assisted VMX\n"));
14467	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14468	}
14469	else
14470	{
14471	Log4Func(("Mode changed -> VINF_EM_RESCHEDULE\n"));
14472	rcStrict = VINF_EM_RESCHEDULE;
14473	}
14474	}
14475	else
14476	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14477	}
14478	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14479	{
14480	rcStrict = VINF_SUCCESS;
14481	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14482	}
14483	return VBOXSTRICTRC_VAL(rcStrict);
14484	}
14485
14486
14487	/**
14488	* VM-exit exception handler wrapper for all other exceptions that are not handled
14489	* by a specific handler.
14490	*
14491	* This simply re-injects the exception back into the VM without any special
14492	* processing.
14493	*
14494	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14495	*/
14496	static VBOXSTRICTRC hmR0VmxExitXcptOthers(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14497	{
14498	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14499
14500	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14501	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14502	AssertMsg(pVCpu->hmr0.s.fUsingDebugLoop \|\| pVmcsInfo->pShared->RealMode.fRealOnV86Active \|\| pVmxTransient->fIsNestedGuest,
14503	("uVector=%#x u32XcptBitmap=%#X32\n",
14504	VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo), pVmcsInfo->u32XcptBitmap));
14505	NOREF(pVmcsInfo);
14506	#endif
14507
14508	/*
14509	* Re-inject the exception into the guest. This cannot be a double-fault condition which
14510	* would have been handled while checking exits due to event delivery.
14511	*/
14512	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14513
14514	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14515	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
14516	AssertRCReturn(rc, rc);
14517	Log4Func(("Reinjecting Xcpt. uVector=%#x cs:rip=%#04x:%#RX64\n", uVector, pCtx->cs.Sel, pCtx->rip));
14518	#endif
14519
14520	#ifdef VBOX_WITH_STATISTICS
14521	switch (uVector)
14522	{
14523	case X86_XCPT_DE: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE); break;
14524	case X86_XCPT_DB: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB); break;
14525	case X86_XCPT_BP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP); break;
14526	case X86_XCPT_OF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
14527	case X86_XCPT_BR: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBR); break;
14528	case X86_XCPT_UD: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD); break;
14529	case X86_XCPT_NM: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
14530	case X86_XCPT_DF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDF); break;
14531	case X86_XCPT_TS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestTS); break;
14532	case X86_XCPT_NP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNP); break;
14533	case X86_XCPT_SS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestSS); break;
14534	case X86_XCPT_GP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP); break;
14535	case X86_XCPT_PF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF); break;
14536	case X86_XCPT_MF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF); break;
14537	case X86_XCPT_AC: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC); break;
14538	case X86_XCPT_XF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXF); break;
14539	default:
14540	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXcpUnk);
14541	break;
14542	}
14543	#endif
14544
14545	/* We should never call this function for a page-fault, we'd need to pass on the fault address below otherwise. */
14546	Assert(!VMX_EXIT_INT_INFO_IS_XCPT_PF(pVmxTransient->uExitIntInfo));
14547	NOREF(uVector);
14548
14549	/* Re-inject the original exception into the guest. */
14550	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14551	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14552	return VINF_SUCCESS;
14553	}
14554
14555
14556	/**
14557	* VM-exit exception handler for all exceptions (except NMIs!).
14558	*
14559	* @remarks This may be called for both guests and nested-guests. Take care to not
14560	* make assumptions and avoid doing anything that is not relevant when
14561	* executing a nested-guest (e.g., Mesa driver hacks).
14562	*/
14563	static VBOXSTRICTRC hmR0VmxExitXcpt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14564	{
14565	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_XCPT_INFO);
14566
14567	/*
14568	* If this VM-exit occurred while delivering an event through the guest IDT, take
14569	* action based on the return code and additional hints (e.g. for page-faults)
14570	* that will be updated in the VMX transient structure.
14571	*/
14572	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
14573	if (rcStrict == VINF_SUCCESS)
14574	{
14575	/*
14576	* If an exception caused a VM-exit due to delivery of an event, the original
14577	* event may have to be re-injected into the guest. We shall reinject it and
14578	* continue guest execution. However, page-fault is a complicated case and
14579	* needs additional processing done in hmR0VmxExitXcptPF().
14580	*/
14581	Assert(VMX_EXIT_INT_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
14582	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14583	if ( !pVCpu->hm.s.Event.fPending
14584	\|\| uVector == X86_XCPT_PF)
14585	{
14586	switch (uVector)
14587	{
14588	case X86_XCPT_PF: return hmR0VmxExitXcptPF(pVCpu, pVmxTransient);
14589	case X86_XCPT_GP: return hmR0VmxExitXcptGP(pVCpu, pVmxTransient);
14590	case X86_XCPT_MF: return hmR0VmxExitXcptMF(pVCpu, pVmxTransient);
14591	case X86_XCPT_DB: return hmR0VmxExitXcptDB(pVCpu, pVmxTransient);
14592	case X86_XCPT_BP: return hmR0VmxExitXcptBP(pVCpu, pVmxTransient);
14593	case X86_XCPT_AC: return hmR0VmxExitXcptAC(pVCpu, pVmxTransient);
14594	default:
14595	return hmR0VmxExitXcptOthers(pVCpu, pVmxTransient);
14596	}
14597	}
14598	/* else: inject pending event before resuming guest execution. */
14599	}
14600	else if (rcStrict == VINF_HM_DOUBLE_FAULT)
14601	{
14602	Assert(pVCpu->hm.s.Event.fPending);
14603	rcStrict = VINF_SUCCESS;
14604	}
14605
14606	return rcStrict;
14607	}
14608	/** @} */
14609
14610
14611	/** @name VM-exit handlers.
14612	* @{
14613	*/
14614	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
14615	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
14616	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
14617
14618	/**
14619	* VM-exit handler for external interrupts (VMX_EXIT_EXT_INT).
14620	*/
14621	HMVMX_EXIT_DECL hmR0VmxExitExtInt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14622	{
14623	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14624	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt);
14625	/* Windows hosts (32-bit and 64-bit) have DPC latency issues. See @bugref{6853}. */
14626	if (VMMR0ThreadCtxHookIsEnabled(pVCpu))
14627	return VINF_SUCCESS;
14628	return VINF_EM_RAW_INTERRUPT;
14629	}
14630
14631
14632	/**
14633	* VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI). Conditional
14634	* VM-exit.
14635	*/
14636	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmi(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14637	{
14638	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14639	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitXcptNmi, y3);
14640
14641	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
14642
14643	uint32_t const uExitIntType = VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo);
14644	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14645	Assert(VMX_EXIT_INT_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
14646
14647	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14648	Assert( !(pVmcsInfo->u32ExitCtls & VMX_EXIT_CTLS_ACK_EXT_INT)
14649	&& uExitIntType != VMX_EXIT_INT_INFO_TYPE_EXT_INT);
14650	NOREF(pVmcsInfo);
14651
14652	VBOXSTRICTRC rcStrict;
14653	switch (uExitIntType)
14654	{
14655	/*
14656	* Host physical NMIs:
14657	* This cannot be a guest NMI as the only way for the guest to receive an NMI is if we
14658	* injected it ourselves and anything we inject is not going to cause a VM-exit directly
14659	* for the event being injected[1]. Go ahead and dispatch the NMI to the host[2].
14660	*
14661	* See Intel spec. 27.2.3 "Information for VM Exits During Event Delivery".
14662	* See Intel spec. 27.5.5 "Updating Non-Register State".
14663	*/
14664	case VMX_EXIT_INT_INFO_TYPE_NMI:
14665	{
14666	rcStrict = hmR0VmxExitHostNmi(pVCpu, pVmcsInfo);
14667	break;
14668	}
14669
14670	/*
14671	* Privileged software exceptions (#DB from ICEBP),
14672	* Software exceptions (#BP and #OF),
14673	* Hardware exceptions:
14674	* Process the required exceptions and resume guest execution if possible.
14675	*/
14676	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
14677	Assert(uVector == X86_XCPT_DB);
14678	RT_FALL_THRU();
14679	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
14680	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF \|\| uExitIntType == VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT);
14681	RT_FALL_THRU();
14682	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
14683	{
14684	NOREF(uVector);
14685	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
14686	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14687	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
14688	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
14689
14690	rcStrict = hmR0VmxExitXcpt(pVCpu, pVmxTransient);
14691	break;
14692	}
14693
14694	default:
14695	{
14696	pVCpu->hm.s.u32HMError = pVmxTransient->uExitIntInfo;
14697	rcStrict = VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
14698	AssertMsgFailed(("Invalid/unexpected VM-exit interruption info %#x\n", pVmxTransient->uExitIntInfo));
14699	break;
14700	}
14701	}
14702
14703	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
14704	return rcStrict;
14705	}
14706
14707
14708	/**
14709	* VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
14710	*/
14711	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindow(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14712	{
14713	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14714
14715	/* Indicate that we no longer need to VM-exit when the guest is ready to receive interrupts, it is now ready. */
14716	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14717	hmR0VmxClearIntWindowExitVmcs(pVmcsInfo);
14718
14719	/* Evaluate and deliver pending events and resume guest execution. */
14720	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow);
14721	return VINF_SUCCESS;
14722	}
14723
14724
14725	/**
14726	* VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
14727	*/
14728	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindow(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14729	{
14730	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14731
14732	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14733	if (RT_UNLIKELY(!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))) /** @todo NSTVMX: Turn this into an assertion. */
14734	{
14735	AssertMsgFailed(("Unexpected NMI-window exit.\n"));
14736	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
14737	}
14738
14739	Assert(!CPUMIsGuestNmiBlocking(pVCpu));
14740
14741	/*
14742	* If block-by-STI is set when we get this VM-exit, it means the CPU doesn't block NMIs following STI.
14743	* It is therefore safe to unblock STI and deliver the NMI ourselves. See @bugref{7445}.
14744	*/
14745	uint32_t fIntrState;
14746	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
14747	AssertRC(rc);
14748	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
14749	if (fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
14750	{
14751	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
14752	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
14753
14754	fIntrState &= ~VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
14755	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
14756	AssertRC(rc);
14757	}
14758
14759	/* Indicate that we no longer need to VM-exit when the guest is ready to receive NMIs, it is now ready */
14760	hmR0VmxClearNmiWindowExitVmcs(pVmcsInfo);
14761
14762	/* Evaluate and deliver pending events and resume guest execution. */
14763	return VINF_SUCCESS;
14764	}
14765
14766
14767	/**
14768	* VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
14769	*/
14770	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvd(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14771	{
14772	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14773	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14774	}
14775
14776
14777	/**
14778	* VM-exit handler for INVD (VMX_EXIT_INVD). Unconditional VM-exit.
14779	*/
14780	HMVMX_EXIT_NSRC_DECL hmR0VmxExitInvd(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14781	{
14782	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14783	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14784	}
14785
14786
14787	/**
14788	* VM-exit handler for CPUID (VMX_EXIT_CPUID). Unconditional VM-exit.
14789	*/
14790	HMVMX_EXIT_DECL hmR0VmxExitCpuid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14791	{
14792	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14793
14794	/*
14795	* Get the state we need and update the exit history entry.
14796	*/
14797	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14798	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14799
14800	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
14801	AssertRCReturn(rc, rc);
14802
14803	VBOXSTRICTRC rcStrict;
14804	PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
14805	EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_CPUID),
14806	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
14807	if (!pExitRec)
14808	{
14809	/*
14810	* Regular CPUID instruction execution.
14811	*/
14812	rcStrict = IEMExecDecodedCpuid(pVCpu, pVmxTransient->cbExitInstr);
14813	if (rcStrict == VINF_SUCCESS)
14814	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14815	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14816	{
14817	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14818	rcStrict = VINF_SUCCESS;
14819	}
14820	}
14821	else
14822	{
14823	/*
14824	* Frequent exit or something needing probing. Get state and call EMHistoryExec.
14825	*/
14826	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14827	AssertRCReturn(rc2, rc2);
14828
14829	Log4(("CpuIdExit/%u: %04x:%08RX64: %#x/%#x -> EMHistoryExec\n",
14830	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.eax, pVCpu->cpum.GstCtx.ecx));
14831
14832	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
14833	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14834
14835	Log4(("CpuIdExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
14836	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
14837	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
14838	}
14839	return rcStrict;
14840	}
14841
14842
14843	/**
14844	* VM-exit handler for GETSEC (VMX_EXIT_GETSEC). Unconditional VM-exit.
14845	*/
14846	HMVMX_EXIT_DECL hmR0VmxExitGetsec(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14847	{
14848	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14849
14850	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14851	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR4);
14852	AssertRCReturn(rc, rc);
14853
14854	if (pVCpu->cpum.GstCtx.cr4 & X86_CR4_SMXE)
14855	return VINF_EM_RAW_EMULATE_INSTR;
14856
14857	AssertMsgFailed(("hmR0VmxExitGetsec: Unexpected VM-exit when CR4.SMXE is 0.\n"));
14858	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
14859	}
14860
14861
14862	/**
14863	* VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
14864	*/
14865	HMVMX_EXIT_DECL hmR0VmxExitRdtsc(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14866	{
14867	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14868
14869	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14870	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14871	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14872	AssertRCReturn(rc, rc);
14873
14874	VBOXSTRICTRC rcStrict = IEMExecDecodedRdtsc(pVCpu, pVmxTransient->cbExitInstr);
14875	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
14876	{
14877	/* If we get a spurious VM-exit when TSC offsetting is enabled,
14878	we must reset offsetting on VM-entry. See @bugref{6634}. */
14879	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TSC_OFFSETTING)
14880	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14881	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14882	}
14883	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14884	{
14885	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14886	rcStrict = VINF_SUCCESS;
14887	}
14888	return rcStrict;
14889	}
14890
14891
14892	/**
14893	* VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
14894	*/
14895	HMVMX_EXIT_DECL hmR0VmxExitRdtscp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14896	{
14897	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14898
14899	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14900	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14901	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_TSC_AUX);
14902	AssertRCReturn(rc, rc);
14903
14904	VBOXSTRICTRC rcStrict = IEMExecDecodedRdtscp(pVCpu, pVmxTransient->cbExitInstr);
14905	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
14906	{
14907	/* If we get a spurious VM-exit when TSC offsetting is enabled,
14908	we must reset offsetting on VM-reentry. See @bugref{6634}. */
14909	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TSC_OFFSETTING)
14910	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14911	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14912	}
14913	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14914	{
14915	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14916	rcStrict = VINF_SUCCESS;
14917	}
14918	return rcStrict;
14919	}
14920
14921
14922	/**
14923	* VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
14924	*/
14925	HMVMX_EXIT_DECL hmR0VmxExitRdpmc(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14926	{
14927	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14928
14929	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14930	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR4 \| CPUMCTX_EXTRN_CR0
14931	\| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS);
14932	AssertRCReturn(rc, rc);
14933
14934	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14935	rc = EMInterpretRdpmc(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx));
14936	if (RT_LIKELY(rc == VINF_SUCCESS))
14937	{
14938	rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14939	Assert(pVmxTransient->cbExitInstr == 2);
14940	}
14941	else
14942	{
14943	AssertMsgFailed(("hmR0VmxExitRdpmc: EMInterpretRdpmc failed with %Rrc\n", rc));
14944	rc = VERR_EM_INTERPRETER;
14945	}
14946	return rc;
14947	}
14948
14949
14950	/**
14951	* VM-exit handler for VMCALL (VMX_EXIT_VMCALL). Unconditional VM-exit.
14952	*/
14953	HMVMX_EXIT_DECL hmR0VmxExitVmcall(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14954	{
14955	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14956
14957	VBOXSTRICTRC rcStrict = VERR_VMX_IPE_3;
14958	if (EMAreHypercallInstructionsEnabled(pVCpu))
14959	{
14960	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14961	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_CR0
14962	\| CPUMCTX_EXTRN_SS \| CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_EFER);
14963	AssertRCReturn(rc, rc);
14964
14965	/* Perform the hypercall. */
14966	rcStrict = GIMHypercall(pVCpu, &pVCpu->cpum.GstCtx);
14967	if (rcStrict == VINF_SUCCESS)
14968	{
14969	rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14970	AssertRCReturn(rc, rc);
14971	}
14972	else
14973	Assert( rcStrict == VINF_GIM_R3_HYPERCALL
14974	\|\| rcStrict == VINF_GIM_HYPERCALL_CONTINUING
14975	\|\| RT_FAILURE(rcStrict));
14976
14977	/* If the hypercall changes anything other than guest's general-purpose registers,
14978	we would need to reload the guest changed bits here before VM-entry. */
14979	}
14980	else
14981	Log4Func(("Hypercalls not enabled\n"));
14982
14983	/* If hypercalls are disabled or the hypercall failed for some reason, raise #UD and continue. */
14984	if (RT_FAILURE(rcStrict))
14985	{
14986	hmR0VmxSetPendingXcptUD(pVCpu);
14987	rcStrict = VINF_SUCCESS;
14988	}
14989
14990	return rcStrict;
14991	}
14992
14993
14994	/**
14995	* VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
14996	*/
14997	HMVMX_EXIT_DECL hmR0VmxExitInvlpg(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14998	{
14999	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15000	Assert(!pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging \|\| pVCpu->hmr0.s.fUsingDebugLoop);
15001
15002	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15003	hmR0VmxReadExitQualVmcs(pVmxTransient);
15004	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15005	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
15006	AssertRCReturn(rc, rc);
15007
15008	VBOXSTRICTRC rcStrict = IEMExecDecodedInvlpg(pVCpu, pVmxTransient->cbExitInstr, pVmxTransient->uExitQual);
15009
15010	if (rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_PGM_SYNC_CR3)
15011	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15012	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15013	{
15014	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15015	rcStrict = VINF_SUCCESS;
15016	}
15017	else
15018	AssertMsgFailed(("Unexpected IEMExecDecodedInvlpg(%#RX64) status: %Rrc\n", pVmxTransient->uExitQual,
15019	VBOXSTRICTRC_VAL(rcStrict)));
15020	return rcStrict;
15021	}
15022
15023
15024	/**
15025	* VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
15026	*/
15027	HMVMX_EXIT_DECL hmR0VmxExitMonitor(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15028	{
15029	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15030
15031	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15032	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15033	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK \| CPUMCTX_EXTRN_DS);
15034	AssertRCReturn(rc, rc);
15035
15036	VBOXSTRICTRC rcStrict = IEMExecDecodedMonitor(pVCpu, pVmxTransient->cbExitInstr);
15037	if (rcStrict == VINF_SUCCESS)
15038	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15039	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15040	{
15041	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15042	rcStrict = VINF_SUCCESS;
15043	}
15044
15045	return rcStrict;
15046	}
15047
15048
15049	/**
15050	* VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
15051	*/
15052	HMVMX_EXIT_DECL hmR0VmxExitMwait(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15053	{
15054	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15055
15056	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15057	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15058	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
15059	AssertRCReturn(rc, rc);
15060
15061	VBOXSTRICTRC rcStrict = IEMExecDecodedMwait(pVCpu, pVmxTransient->cbExitInstr);
15062	if (RT_SUCCESS(rcStrict))
15063	{
15064	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15065	if (EMMonitorWaitShouldContinue(pVCpu, &pVCpu->cpum.GstCtx))
15066	rcStrict = VINF_SUCCESS;
15067	}
15068
15069	return rcStrict;
15070	}
15071
15072
15073	/**
15074	* VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT). Unconditional
15075	* VM-exit.
15076	*/
15077	HMVMX_EXIT_DECL hmR0VmxExitTripleFault(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15078	{
15079	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15080	return VINF_EM_RESET;
15081	}
15082
15083
15084	/**
15085	* VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
15086	*/
15087	HMVMX_EXIT_DECL hmR0VmxExitHlt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15088	{
15089	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15090
15091	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15092	AssertRCReturn(rc, rc);
15093
15094	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS); /* Advancing the RIP above should've imported eflags. */
15095	if (EMShouldContinueAfterHalt(pVCpu, &pVCpu->cpum.GstCtx)) /* Requires eflags. */
15096	rc = VINF_SUCCESS;
15097	else
15098	rc = VINF_EM_HALT;
15099
15100	if (rc != VINF_SUCCESS)
15101	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHltToR3);
15102	return rc;
15103	}
15104
15105
15106	/**
15107	* VM-exit handler for instructions that result in a \#UD exception delivered to
15108	* the guest.
15109	*/
15110	HMVMX_EXIT_NSRC_DECL hmR0VmxExitSetPendingXcptUD(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15111	{
15112	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15113	hmR0VmxSetPendingXcptUD(pVCpu);
15114	return VINF_SUCCESS;
15115	}
15116
15117
15118	/**
15119	* VM-exit handler for expiry of the VMX-preemption timer.
15120	*/
15121	HMVMX_EXIT_DECL hmR0VmxExitPreemptTimer(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15122	{
15123	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15124
15125	/* If the VMX-preemption timer has expired, reinitialize the preemption timer on next VM-entry. */
15126	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
15127	Log12(("hmR0VmxExitPreemptTimer:\n"));
15128
15129	/* If there are any timer events pending, fall back to ring-3, otherwise resume guest execution. */
15130	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15131	bool fTimersPending = TMTimerPollBool(pVM, pVCpu);
15132	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitPreemptTimer);
15133	return fTimersPending ? VINF_EM_RAW_TIMER_PENDING : VINF_SUCCESS;
15134	}
15135
15136
15137	/**
15138	* VM-exit handler for XSETBV (VMX_EXIT_XSETBV). Unconditional VM-exit.
15139	*/
15140	HMVMX_EXIT_DECL hmR0VmxExitXsetbv(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15141	{
15142	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15143
15144	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15145	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15146	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_CR4);
15147	AssertRCReturn(rc, rc);
15148
15149	VBOXSTRICTRC rcStrict = IEMExecDecodedXsetbv(pVCpu, pVmxTransient->cbExitInstr);
15150	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, rcStrict != VINF_IEM_RAISED_XCPT ? HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
15151	: HM_CHANGED_RAISED_XCPT_MASK);
15152
15153	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15154	bool const fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
15155	if (fLoadSaveGuestXcr0 != pVCpu->hmr0.s.fLoadSaveGuestXcr0)
15156	{
15157	pVCpu->hmr0.s.fLoadSaveGuestXcr0 = fLoadSaveGuestXcr0;
15158	hmR0VmxUpdateStartVmFunction(pVCpu);
15159	}
15160
15161	return rcStrict;
15162	}
15163
15164
15165	/**
15166	* VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
15167	*/
15168	HMVMX_EXIT_DECL hmR0VmxExitInvpcid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15169	{
15170	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15171
15172	/** @todo Enable the new code after finding a reliably guest test-case. */
15173	#if 1
15174	return VERR_EM_INTERPRETER;
15175	#else
15176	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15177	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15178	hmR0VmxReadExitQualVmcs(pVmxTransient);
15179	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
15180	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
15181	AssertRCReturn(rc, rc);
15182
15183	/* Paranoia. Ensure this has a memory operand. */
15184	Assert(!pVmxTransient->ExitInstrInfo.Inv.u1Cleared0);
15185
15186	uint8_t const iGReg = pVmxTransient->ExitInstrInfo.VmreadVmwrite.iReg2;
15187	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
15188	uint64_t const uType = CPUMIsGuestIn64BitCode(pVCpu) ? pVCpu->cpum.GstCtx.aGRegs[iGReg].u64
15189	: pVCpu->cpum.GstCtx.aGRegs[iGReg].u32;
15190
15191	RTGCPTR GCPtrDesc;
15192	HMVMX_DECODE_MEM_OPERAND(pVCpu, pVmxTransient->ExitInstrInfo.u, pVmxTransient->uExitQual, VMXMEMACCESS_READ, &GCPtrDesc);
15193
15194	VBOXSTRICTRC rcStrict = IEMExecDecodedInvpcid(pVCpu, pVmxTransient->cbExitInstr, pVmxTransient->ExitInstrInfo.Inv.iSegReg,
15195	GCPtrDesc, uType);
15196	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15197	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15198	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15199	{
15200	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15201	rcStrict = VINF_SUCCESS;
15202	}
15203	return rcStrict;
15204	#endif
15205	}
15206
15207
15208	/**
15209	* VM-exit handler for invalid-guest-state (VMX_EXIT_ERR_INVALID_GUEST_STATE). Error
15210	* VM-exit.
15211	*/
15212	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrInvalidGuestState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15213	{
15214	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15215	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
15216	AssertRCReturn(rc, rc);
15217
15218	rc = hmR0VmxCheckCachedVmcsCtls(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest);
15219	if (RT_FAILURE(rc))
15220	return rc;
15221
15222	uint32_t const uInvalidReason = hmR0VmxCheckGuestState(pVCpu, pVmcsInfo);
15223	NOREF(uInvalidReason);
15224
15225	#ifdef VBOX_STRICT
15226	uint32_t fIntrState;
15227	uint64_t u64Val;
15228	hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
15229	hmR0VmxReadEntryXcptErrorCodeVmcs(pVmxTransient);
15230	hmR0VmxReadEntryInstrLenVmcs(pVmxTransient);
15231
15232	Log4(("uInvalidReason %u\n", uInvalidReason));
15233	Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", pVmxTransient->uEntryIntInfo));
15234	Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", pVmxTransient->uEntryXcptErrorCode));
15235	Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %#RX32\n", pVmxTransient->cbEntryInstr));
15236
15237	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState); AssertRC(rc);
15238	Log4(("VMX_VMCS32_GUEST_INT_STATE %#RX32\n", fIntrState));
15239	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64Val); AssertRC(rc);
15240	Log4(("VMX_VMCS_GUEST_CR0 %#RX64\n", u64Val));
15241	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_MASK, &u64Val); AssertRC(rc);
15242	Log4(("VMX_VMCS_CTRL_CR0_MASK %#RX64\n", u64Val));
15243	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, &u64Val); AssertRC(rc);
15244	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RX64\n", u64Val));
15245	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_MASK, &u64Val); AssertRC(rc);
15246	Log4(("VMX_VMCS_CTRL_CR4_MASK %#RX64\n", u64Val));
15247	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, &u64Val); AssertRC(rc);
15248	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RX64\n", u64Val));
15249	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging)
15250	{
15251	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
15252	Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
15253	}
15254	hmR0DumpRegs(pVCpu, HM_DUMP_REG_FLAGS_ALL);
15255	#endif
15256
15257	return VERR_VMX_INVALID_GUEST_STATE;
15258	}
15259
15260	/**
15261	* VM-exit handler for all undefined/unexpected reasons. Should never happen.
15262	*/
15263	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrUnexpected(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15264	{
15265	/*
15266	* Cumulative notes of all recognized but unexpected VM-exits.
15267	*
15268	* 1. This does -not- cover scenarios like a page-fault VM-exit occurring when
15269	* nested-paging is used.
15270	*
15271	* 2. Any instruction that causes a VM-exit unconditionally (for e.g. VMXON) must be
15272	* emulated or a #UD must be raised in the guest. Therefore, we should -not- be using
15273	* this function (and thereby stop VM execution) for handling such instructions.
15274	*
15275	*
15276	* VMX_EXIT_INIT_SIGNAL:
15277	* INIT signals are blocked in VMX root operation by VMXON and by SMI in SMM.
15278	* It is -NOT- blocked in VMX non-root operation so we can, in theory, still get these
15279	* VM-exits. However, we should not receive INIT signals VM-exit while executing a VM.
15280	*
15281	* See Intel spec. 33.14.1 Default Treatment of SMI Delivery"
15282	* See Intel spec. 29.3 "VMX Instructions" for "VMXON".
15283	* See Intel spec. "23.8 Restrictions on VMX operation".
15284	*
15285	* VMX_EXIT_SIPI:
15286	* SIPI exits can only occur in VMX non-root operation when the "wait-for-SIPI" guest
15287	* activity state is used. We don't make use of it as our guests don't have direct
15288	* access to the host local APIC.
15289	*
15290	* See Intel spec. 25.3 "Other Causes of VM-exits".
15291	*
15292	* VMX_EXIT_IO_SMI:
15293	* VMX_EXIT_SMI:
15294	* This can only happen if we support dual-monitor treatment of SMI, which can be
15295	* activated by executing VMCALL in VMX root operation. Only an STM (SMM transfer
15296	* monitor) would get this VM-exit when we (the executive monitor) execute a VMCALL in
15297	* VMX root mode or receive an SMI. If we get here, something funny is going on.
15298	*
15299	* See Intel spec. 33.15.6 "Activating the Dual-Monitor Treatment"
15300	* See Intel spec. 25.3 "Other Causes of VM-Exits"
15301	*
15302	* VMX_EXIT_ERR_MSR_LOAD:
15303	* Failures while loading MSRs are part of the VM-entry MSR-load area are unexpected
15304	* and typically indicates a bug in the hypervisor code. We thus cannot not resume
15305	* execution.
15306	*
15307	* See Intel spec. 26.7 "VM-Entry Failures During Or After Loading Guest State".
15308	*
15309	* VMX_EXIT_ERR_MACHINE_CHECK:
15310	* Machine check exceptions indicates a fatal/unrecoverable hardware condition
15311	* including but not limited to system bus, ECC, parity, cache and TLB errors. A
15312	* #MC exception abort class exception is raised. We thus cannot assume a
15313	* reasonable chance of continuing any sort of execution and we bail.
15314	*
15315	* See Intel spec. 15.1 "Machine-check Architecture".
15316	* See Intel spec. 27.1 "Architectural State Before A VM Exit".
15317	*
15318	* VMX_EXIT_PML_FULL:
15319	* VMX_EXIT_VIRTUALIZED_EOI:
15320	* VMX_EXIT_APIC_WRITE:
15321	* We do not currently support any of these features and thus they are all unexpected
15322	* VM-exits.
15323	*
15324	* VMX_EXIT_GDTR_IDTR_ACCESS:
15325	* VMX_EXIT_LDTR_TR_ACCESS:
15326	* VMX_EXIT_RDRAND:
15327	* VMX_EXIT_RSM:
15328	* VMX_EXIT_VMFUNC:
15329	* VMX_EXIT_ENCLS:
15330	* VMX_EXIT_RDSEED:
15331	* VMX_EXIT_XSAVES:
15332	* VMX_EXIT_XRSTORS:
15333	* VMX_EXIT_UMWAIT:
15334	* VMX_EXIT_TPAUSE:
15335	* These VM-exits are -not- caused unconditionally by execution of the corresponding
15336	* instruction. Any VM-exit for these instructions indicate a hardware problem,
15337	* unsupported CPU modes (like SMM) or potentially corrupt VMCS controls.
15338	*
15339	* See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally".
15340	*/
15341	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15342	AssertMsgFailed(("Unexpected VM-exit %u\n", pVmxTransient->uExitReason));
15343	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
15344	}
15345
15346
15347	/**
15348	* VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
15349	*/
15350	HMVMX_EXIT_DECL hmR0VmxExitRdmsr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15351	{
15352	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15353
15354	/** @todo Optimize this: We currently drag in the whole MSR state
15355	* (CPUMCTX_EXTRN_ALL_MSRS) here. We should optimize this to only get
15356	* MSRs required. That would require changes to IEM and possibly CPUM too.
15357	* (Should probably do it lazy fashion from CPUMAllMsrs.cpp). */
15358	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15359	uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
15360	uint64_t fImport = IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK \| CPUMCTX_EXTRN_ALL_MSRS;
15361	switch (idMsr)
15362	{
15363	case MSR_K8_FS_BASE: fImport \|= CPUMCTX_EXTRN_FS; break;
15364	case MSR_K8_GS_BASE: fImport \|= CPUMCTX_EXTRN_GS; break;
15365	}
15366
15367	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15368	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImport);
15369	AssertRCReturn(rc, rc);
15370
15371	Log4Func(("ecx=%#RX32\n", idMsr));
15372
15373	#ifdef VBOX_STRICT
15374	Assert(!pVmxTransient->fIsNestedGuest);
15375	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
15376	{
15377	if ( hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr)
15378	&& idMsr != MSR_K6_EFER)
15379	{
15380	AssertMsgFailed(("Unexpected RDMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n", idMsr));
15381	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15382	}
15383	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15384	{
15385	Assert(pVmcsInfo->pvMsrBitmap);
15386	uint32_t fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, idMsr);
15387	if (fMsrpm & VMXMSRPM_ALLOW_RD)
15388	{
15389	AssertMsgFailed(("Unexpected RDMSR for a passthru lazy-restore MSR. ecx=%#RX32\n", idMsr));
15390	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15391	}
15392	}
15393	}
15394	#endif
15395
15396	VBOXSTRICTRC rcStrict = IEMExecDecodedRdmsr(pVCpu, pVmxTransient->cbExitInstr);
15397	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr);
15398	if (rcStrict == VINF_SUCCESS)
15399	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15400	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15401	{
15402	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15403	rcStrict = VINF_SUCCESS;
15404	}
15405	else
15406	AssertMsg(rcStrict == VINF_CPUM_R3_MSR_READ \|\| rcStrict == VINF_EM_TRIPLE_FAULT,
15407	("Unexpected IEMExecDecodedRdmsr rc (%Rrc)\n", VBOXSTRICTRC_VAL(rcStrict)));
15408
15409	return rcStrict;
15410	}
15411
15412
15413	/**
15414	* VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
15415	*/
15416	HMVMX_EXIT_DECL hmR0VmxExitWrmsr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15417	{
15418	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15419
15420	/** @todo Optimize this: We currently drag in the whole MSR state
15421	* (CPUMCTX_EXTRN_ALL_MSRS) here. We should optimize this to only get
15422	* MSRs required. That would require changes to IEM and possibly CPUM too.
15423	* (Should probably do it lazy fashion from CPUMAllMsrs.cpp). */
15424	uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
15425	uint64_t fImport = IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK \| CPUMCTX_EXTRN_ALL_MSRS;
15426
15427	/*
15428	* The FS and GS base MSRs are not part of the above all-MSRs mask.
15429	* Although we don't need to fetch the base as it will be overwritten shortly, while
15430	* loading guest-state we would also load the entire segment register including limit
15431	* and attributes and thus we need to load them here.
15432	*/
15433	switch (idMsr)
15434	{
15435	case MSR_K8_FS_BASE: fImport \|= CPUMCTX_EXTRN_FS; break;
15436	case MSR_K8_GS_BASE: fImport \|= CPUMCTX_EXTRN_GS; break;
15437	}
15438
15439	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15440	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15441	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImport);
15442	AssertRCReturn(rc, rc);
15443
15444	Log4Func(("ecx=%#RX32 edx:eax=%#RX32:%#RX32\n", idMsr, pVCpu->cpum.GstCtx.edx, pVCpu->cpum.GstCtx.eax));
15445
15446	VBOXSTRICTRC rcStrict = IEMExecDecodedWrmsr(pVCpu, pVmxTransient->cbExitInstr);
15447	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr);
15448
15449	if (rcStrict == VINF_SUCCESS)
15450	{
15451	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15452
15453	/* If this is an X2APIC WRMSR access, update the APIC state as well. */
15454	if ( idMsr == MSR_IA32_APICBASE
15455	\|\| ( idMsr >= MSR_IA32_X2APIC_START
15456	&& idMsr <= MSR_IA32_X2APIC_END))
15457	{
15458	/*
15459	* We've already saved the APIC related guest-state (TPR) in post-run phase.
15460	* When full APIC register virtualization is implemented we'll have to make
15461	* sure APIC state is saved from the VMCS before IEM changes it.
15462	*/
15463	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
15464	}
15465	else if (idMsr == MSR_IA32_TSC) /* Windows 7 does this during bootup. See @bugref{6398}. */
15466	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
15467	else if (idMsr == MSR_K6_EFER)
15468	{
15469	/*
15470	* If the guest touches the EFER MSR we need to update the VM-Entry and VM-Exit controls
15471	* as well, even if it is -not- touching bits that cause paging mode changes (LMA/LME).
15472	* We care about the other bits as well, SCE and NXE. See @bugref{7368}.
15473	*/
15474	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_EFER_MSR \| HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
15475	}
15476
15477	/* Update MSRs that are part of the VMCS and auto-load/store area when MSR-bitmaps are not used. */
15478	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS))
15479	{
15480	switch (idMsr)
15481	{
15482	case MSR_IA32_SYSENTER_CS: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_CS_MSR); break;
15483	case MSR_IA32_SYSENTER_EIP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); break;
15484	case MSR_IA32_SYSENTER_ESP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); break;
15485	case MSR_K8_FS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_FS); break;
15486	case MSR_K8_GS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_GS); break;
15487	case MSR_K6_EFER: /* Nothing to do, already handled above. */ break;
15488	default:
15489	{
15490	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15491	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_VMX_GUEST_LAZY_MSRS);
15492	else if (hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr))
15493	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
15494	break;
15495	}
15496	}
15497	}
15498	#ifdef VBOX_STRICT
15499	else
15500	{
15501	/* Paranoia. Validate that MSRs in the MSR-bitmaps with write-passthru are not intercepted. */
15502	switch (idMsr)
15503	{
15504	case MSR_IA32_SYSENTER_CS:
15505	case MSR_IA32_SYSENTER_EIP:
15506	case MSR_IA32_SYSENTER_ESP:
15507	case MSR_K8_FS_BASE:
15508	case MSR_K8_GS_BASE:
15509	{
15510	AssertMsgFailed(("Unexpected WRMSR for an MSR in the VMCS. ecx=%#RX32\n", idMsr));
15511	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15512	}
15513
15514	/* Writes to MSRs in auto-load/store area/swapped MSRs, shouldn't cause VM-exits with MSR-bitmaps. */
15515	default:
15516	{
15517	if (hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr))
15518	{
15519	/* EFER MSR writes are always intercepted. */
15520	if (idMsr != MSR_K6_EFER)
15521	{
15522	AssertMsgFailed(("Unexpected WRMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n",
15523	idMsr));
15524	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15525	}
15526	}
15527
15528	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15529	{
15530	Assert(pVmcsInfo->pvMsrBitmap);
15531	uint32_t fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, idMsr);
15532	if (fMsrpm & VMXMSRPM_ALLOW_WR)
15533	{
15534	AssertMsgFailed(("Unexpected WRMSR for passthru, lazy-restore MSR. ecx=%#RX32\n", idMsr));
15535	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15536	}
15537	}
15538	break;
15539	}
15540	}
15541	}
15542	#endif /* VBOX_STRICT */
15543	}
15544	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15545	{
15546	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15547	rcStrict = VINF_SUCCESS;
15548	}
15549	else
15550	AssertMsg(rcStrict == VINF_CPUM_R3_MSR_WRITE \|\| rcStrict == VINF_EM_TRIPLE_FAULT,
15551	("Unexpected IEMExecDecodedWrmsr rc (%Rrc)\n", VBOXSTRICTRC_VAL(rcStrict)));
15552
15553	return rcStrict;
15554	}
15555
15556
15557	/**
15558	* VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
15559	*/
15560	HMVMX_EXIT_DECL hmR0VmxExitPause(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15561	{
15562	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15563
15564	/** @todo The guest has likely hit a contended spinlock. We might want to
15565	* poke a schedule different guest VCPU. */
15566	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15567	if (RT_SUCCESS(rc))
15568	return VINF_EM_RAW_INTERRUPT;
15569
15570	AssertMsgFailed(("hmR0VmxExitPause: Failed to increment RIP. rc=%Rrc\n", rc));
15571	return rc;
15572	}
15573
15574
15575	/**
15576	* VM-exit handler for when the TPR value is lowered below the specified
15577	* threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
15578	*/
15579	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThreshold(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15580	{
15581	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15582	Assert(pVmxTransient->pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
15583
15584	/*
15585	* The TPR shadow would've been synced with the APIC TPR in the post-run phase.
15586	* We'll re-evaluate pending interrupts and inject them before the next VM
15587	* entry so we can just continue execution here.
15588	*/
15589	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTprBelowThreshold);
15590	return VINF_SUCCESS;
15591	}
15592
15593
15594	/**
15595	* VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX). Conditional
15596	* VM-exit.
15597	*
15598	* @retval VINF_SUCCESS when guest execution can continue.
15599	* @retval VINF_PGM_SYNC_CR3 CR3 sync is required, back to ring-3.
15600	* @retval VERR_EM_RESCHEDULE_REM when we need to return to ring-3 due to
15601	* incompatible guest state for VMX execution (real-on-v86 case).
15602	*/
15603	HMVMX_EXIT_DECL hmR0VmxExitMovCRx(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15604	{
15605	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15606	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitMovCRx, y2);
15607
15608	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15609	hmR0VmxReadExitQualVmcs(pVmxTransient);
15610	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15611
15612	VBOXSTRICTRC rcStrict;
15613	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15614	uint64_t const uExitQual = pVmxTransient->uExitQual;
15615	uint32_t const uAccessType = VMX_EXIT_QUAL_CRX_ACCESS(uExitQual);
15616	switch (uAccessType)
15617	{
15618	/*
15619	* MOV to CRx.
15620	*/
15621	case VMX_EXIT_QUAL_CRX_ACCESS_WRITE:
15622	{
15623	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
15624	AssertRCReturn(rc, rc);
15625
15626	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
15627	uint32_t const uOldCr0 = pVCpu->cpum.GstCtx.cr0;
15628	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(uExitQual);
15629	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(uExitQual);
15630
15631	/*
15632	* MOV to CR3 only cause a VM-exit when one or more of the following are true:
15633	* - When nested paging isn't used.
15634	* - If the guest doesn't have paging enabled (intercept CR3 to update shadow page tables).
15635	* - We are executing in the VM debug loop.
15636	*/
15637	Assert( iCrReg != 3
15638	\|\| !pVM->hmr0.s.fNestedPaging
15639	\|\| !CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx)
15640	\|\| pVCpu->hmr0.s.fUsingDebugLoop);
15641
15642	/* MOV to CR8 writes only cause VM-exits when TPR shadow is not used. */
15643	Assert( iCrReg != 8
15644	\|\| !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW));
15645
15646	rcStrict = hmR0VmxExitMovToCrX(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
15647	AssertMsg( rcStrict == VINF_SUCCESS
15648	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15649
15650	/*
15651	* This is a kludge for handling switches back to real mode when we try to use
15652	* V86 mode to run real mode code directly. Problem is that V86 mode cannot
15653	* deal with special selector values, so we have to return to ring-3 and run
15654	* there till the selector values are V86 mode compatible.
15655	*
15656	* Note! Using VINF_EM_RESCHEDULE_REM here rather than VINF_EM_RESCHEDULE since the
15657	* latter is an alias for VINF_IEM_RAISED_XCPT which is asserted at the end of
15658	* this function.
15659	*/
15660	if ( iCrReg == 0
15661	&& rcStrict == VINF_SUCCESS
15662	&& !pVM->hmr0.s.vmx.fUnrestrictedGuest
15663	&& CPUMIsGuestInRealModeEx(&pVCpu->cpum.GstCtx)
15664	&& (uOldCr0 & X86_CR0_PE)
15665	&& !(pVCpu->cpum.GstCtx.cr0 & X86_CR0_PE))
15666	{
15667	/** @todo Check selectors rather than returning all the time. */
15668	Assert(!pVmxTransient->fIsNestedGuest);
15669	Log4Func(("CR0 write, back to real mode -> VINF_EM_RESCHEDULE_REM\n"));
15670	rcStrict = VINF_EM_RESCHEDULE_REM;
15671	}
15672	break;
15673	}
15674
15675	/*
15676	* MOV from CRx.
15677	*/
15678	case VMX_EXIT_QUAL_CRX_ACCESS_READ:
15679	{
15680	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(uExitQual);
15681	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(uExitQual);
15682
15683	/*
15684	* MOV from CR3 only cause a VM-exit when one or more of the following are true:
15685	* - When nested paging isn't used.
15686	* - If the guest doesn't have paging enabled (pass guest's CR3 rather than our identity mapped CR3).
15687	* - We are executing in the VM debug loop.
15688	*/
15689	Assert( iCrReg != 3
15690	\|\| !pVM->hmr0.s.fNestedPaging
15691	\|\| !CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx)
15692	\|\| pVCpu->hmr0.s.fLeaveDone);
15693
15694	/* MOV from CR8 reads only cause a VM-exit when the TPR shadow feature isn't enabled. */
15695	Assert( iCrReg != 8
15696	\|\| !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW));
15697
15698	rcStrict = hmR0VmxExitMovFromCrX(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
15699	break;
15700	}
15701
15702	/*
15703	* CLTS (Clear Task-Switch Flag in CR0).
15704	*/
15705	case VMX_EXIT_QUAL_CRX_ACCESS_CLTS:
15706	{
15707	rcStrict = hmR0VmxExitClts(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr);
15708	break;
15709	}
15710
15711	/*
15712	* LMSW (Load Machine-Status Word into CR0).
15713	* LMSW cannot clear CR0.PE, so no fRealOnV86Active kludge needed here.
15714	*/
15715	case VMX_EXIT_QUAL_CRX_ACCESS_LMSW:
15716	{
15717	RTGCPTR GCPtrEffDst;
15718	uint8_t const cbInstr = pVmxTransient->cbExitInstr;
15719	uint16_t const uMsw = VMX_EXIT_QUAL_CRX_LMSW_DATA(uExitQual);
15720	bool const fMemOperand = VMX_EXIT_QUAL_CRX_LMSW_OP_MEM(uExitQual);
15721	if (fMemOperand)
15722	{
15723	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
15724	GCPtrEffDst = pVmxTransient->uGuestLinearAddr;
15725	}
15726	else
15727	GCPtrEffDst = NIL_RTGCPTR;
15728	rcStrict = hmR0VmxExitLmsw(pVCpu, pVmcsInfo, cbInstr, uMsw, GCPtrEffDst);
15729	break;
15730	}
15731
15732	default:
15733	{
15734	AssertMsgFailed(("Unrecognized Mov CRX access type %#x\n", uAccessType));
15735	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, uAccessType);
15736	}
15737	}
15738
15739	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS))
15740	== (HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS));
15741	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
15742
15743	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitMovCRx, y2);
15744	NOREF(pVM);
15745	return rcStrict;
15746	}
15747
15748
15749	/**
15750	* VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR). Conditional
15751	* VM-exit.
15752	*/
15753	HMVMX_EXIT_DECL hmR0VmxExitIoInstr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15754	{
15755	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15756	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitIO, y1);
15757
15758	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15759	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15760	hmR0VmxReadExitQualVmcs(pVmxTransient);
15761	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15762	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_SREG_MASK
15763	\| CPUMCTX_EXTRN_EFER);
15764	/* EFER MSR also required for longmode checks in EMInterpretDisasCurrent(), but it's always up-to-date. */
15765	AssertRCReturn(rc, rc);
15766
15767	/* Refer Intel spec. 27-5. "Exit Qualifications for I/O Instructions" for the format. */
15768	uint32_t const uIOPort = VMX_EXIT_QUAL_IO_PORT(pVmxTransient->uExitQual);
15769	uint8_t const uIOSize = VMX_EXIT_QUAL_IO_SIZE(pVmxTransient->uExitQual);
15770	bool const fIOWrite = (VMX_EXIT_QUAL_IO_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_IO_DIRECTION_OUT);
15771	bool const fIOString = VMX_EXIT_QUAL_IO_IS_STRING(pVmxTransient->uExitQual);
15772	bool const fGstStepping = RT_BOOL(pCtx->eflags.Bits.u1TF);
15773	bool const fDbgStepping = pVCpu->hm.s.fSingleInstruction;
15774	AssertReturn(uIOSize <= 3 && uIOSize != 2, VERR_VMX_IPE_1);
15775
15776	/*
15777	* Update exit history to see if this exit can be optimized.
15778	*/
15779	VBOXSTRICTRC rcStrict;
15780	PCEMEXITREC pExitRec = NULL;
15781	if ( !fGstStepping
15782	&& !fDbgStepping)
15783	pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
15784	!fIOString
15785	? !fIOWrite
15786	? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_READ)
15787	: EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_WRITE)
15788	: !fIOWrite
15789	? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_READ)
15790	: EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_WRITE),
15791	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
15792	if (!pExitRec)
15793	{
15794	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses in bytes. */
15795	static uint32_t const s_aIOOpAnd[4] = { 0xff, 0xffff, 0, 0xffffffff }; /* AND masks for saving result in AL/AX/EAX. */
15796
15797	uint32_t const cbValue = s_aIOSizes[uIOSize];
15798	uint32_t const cbInstr = pVmxTransient->cbExitInstr;
15799	bool fUpdateRipAlready = false; /* ugly hack, should be temporary. */
15800	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15801	if (fIOString)
15802	{
15803	/*
15804	* INS/OUTS - I/O String instruction.
15805	*
15806	* Use instruction-information if available, otherwise fall back on
15807	* interpreting the instruction.
15808	*/
15809	Log4Func(("cs:rip=%#04x:%#RX64 %#06x/%u %c str\n", pCtx->cs.Sel, pCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
15810	AssertReturn(pCtx->dx == uIOPort, VERR_VMX_IPE_2);
15811	bool const fInsOutsInfo = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_INS_OUTS);
15812	if (fInsOutsInfo)
15813	{
15814	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15815	AssertReturn(pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize <= 2, VERR_VMX_IPE_3);
15816	AssertCompile(IEMMODE_16BIT == 0 && IEMMODE_32BIT == 1 && IEMMODE_64BIT == 2);
15817	IEMMODE const enmAddrMode = (IEMMODE)pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize;
15818	bool const fRep = VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual);
15819	if (fIOWrite)
15820	rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, fRep, cbInstr,
15821	pVmxTransient->ExitInstrInfo.StrIo.iSegReg, true /fIoChecked/);
15822	else
15823	{
15824	/*
15825	* The segment prefix for INS cannot be overridden and is always ES. We can safely assume X86_SREG_ES.
15826	* Hence "iSegReg" field is undefined in the instruction-information field in VT-x for INS.
15827	* See Intel Instruction spec. for "INS".
15828	* See Intel spec. Table 27-8 "Format of the VM-Exit Instruction-Information Field as Used for INS and OUTS".
15829	*/
15830	rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, fRep, cbInstr, true /fIoChecked/);
15831	}
15832	}
15833	else
15834	rcStrict = IEMExecOne(pVCpu);
15835
15836	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
15837	fUpdateRipAlready = true;
15838	}
15839	else
15840	{
15841	/*
15842	* IN/OUT - I/O instruction.
15843	*/
15844	Log4Func(("cs:rip=%04x:%08RX64 %#06x/%u %c\n", pCtx->cs.Sel, pCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
15845	uint32_t const uAndVal = s_aIOOpAnd[uIOSize];
15846	Assert(!VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual));
15847	if (fIOWrite)
15848	{
15849	rcStrict = IOMIOPortWrite(pVM, pVCpu, uIOPort, pCtx->eax & uAndVal, cbValue);
15850	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite);
15851	if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
15852	&& !pCtx->eflags.Bits.u1TF)
15853	rcStrict = EMRZSetPendingIoPortWrite(pVCpu, uIOPort, cbInstr, cbValue, pCtx->eax & uAndVal);
15854	}
15855	else
15856	{
15857	uint32_t u32Result = 0;
15858	rcStrict = IOMIOPortRead(pVM, pVCpu, uIOPort, &u32Result, cbValue);
15859	if (IOM_SUCCESS(rcStrict))
15860	{
15861	/* Save result of I/O IN instr. in AL/AX/EAX. */
15862	pCtx->eax = (pCtx->eax & ~uAndVal) \| (u32Result & uAndVal);
15863	}
15864	if ( rcStrict == VINF_IOM_R3_IOPORT_READ
15865	&& !pCtx->eflags.Bits.u1TF)
15866	rcStrict = EMRZSetPendingIoPortRead(pVCpu, uIOPort, cbInstr, cbValue);
15867	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead);
15868	}
15869	}
15870
15871	if (IOM_SUCCESS(rcStrict))
15872	{
15873	if (!fUpdateRipAlready)
15874	{
15875	hmR0VmxAdvanceGuestRipBy(pVCpu, cbInstr);
15876	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
15877	}
15878
15879	/*
15880	* INS/OUTS with REP prefix updates RFLAGS, can be observed with triple-fault guru
15881	* while booting Fedora 17 64-bit guest.
15882	*
15883	* See Intel Instruction reference for REP/REPE/REPZ/REPNE/REPNZ.
15884	*/
15885	if (fIOString)
15886	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RFLAGS);
15887
15888	/*
15889	* If any I/O breakpoints are armed, we need to check if one triggered
15890	* and take appropriate action.
15891	* Note that the I/O breakpoint type is undefined if CR4.DE is 0.
15892	*/
15893	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_DR7);
15894	AssertRCReturn(rc, rc);
15895
15896	/** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the
15897	* execution engines about whether hyper BPs and such are pending. */
15898	uint32_t const uDr7 = pCtx->dr[7];
15899	if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK)
15900	&& X86_DR7_ANY_RW_IO(uDr7)
15901	&& (pCtx->cr4 & X86_CR4_DE))
15902	\|\| DBGFBpIsHwIoArmed(pVM)))
15903	{
15904	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck);
15905
15906	/* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
15907	VMMRZCallRing3Disable(pVCpu);
15908	HM_DISABLE_PREEMPT(pVCpu);
15909
15910	bool fIsGuestDbgActive = CPUMR0DebugStateMaybeSaveGuest(pVCpu, true /* fDr6 */);
15911
15912	VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pCtx, uIOPort, cbValue);
15913	if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
15914	{
15915	/* Raise #DB. */
15916	if (fIsGuestDbgActive)
15917	ASMSetDR6(pCtx->dr[6]);
15918	if (pCtx->dr[7] != uDr7)
15919	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_GUEST_DR7;
15920
15921	hmR0VmxSetPendingXcptDB(pVCpu);
15922	}
15923	/* rcStrict is VINF_SUCCESS, VINF_IOM_R3_IOPORT_COMMIT_WRITE, or in [VINF_EM_FIRST..VINF_EM_LAST],
15924	however we can ditch VINF_IOM_R3_IOPORT_COMMIT_WRITE as it has VMCPU_FF_IOM as backup. */
15925	else if ( rcStrict2 != VINF_SUCCESS
15926	&& (rcStrict == VINF_SUCCESS \|\| rcStrict2 < rcStrict))
15927	rcStrict = rcStrict2;
15928	AssertCompile(VINF_EM_LAST < VINF_IOM_R3_IOPORT_COMMIT_WRITE);
15929
15930	HM_RESTORE_PREEMPT();
15931	VMMRZCallRing3Enable(pVCpu);
15932	}
15933	}
15934
15935	#ifdef VBOX_STRICT
15936	if ( rcStrict == VINF_IOM_R3_IOPORT_READ
15937	\|\| rcStrict == VINF_EM_PENDING_R3_IOPORT_READ)
15938	Assert(!fIOWrite);
15939	else if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
15940	\|\| rcStrict == VINF_IOM_R3_IOPORT_COMMIT_WRITE
15941	\|\| rcStrict == VINF_EM_PENDING_R3_IOPORT_WRITE)
15942	Assert(fIOWrite);
15943	else
15944	{
15945	# if 0 /** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST
15946	* statuses, that the VMM device and some others may return. See
15947	* IOM_SUCCESS() for guidance. */
15948	AssertMsg( RT_FAILURE(rcStrict)
15949	\|\| rcStrict == VINF_SUCCESS
15950	\|\| rcStrict == VINF_EM_RAW_EMULATE_INSTR
15951	\|\| rcStrict == VINF_EM_DBG_BREAKPOINT
15952	\|\| rcStrict == VINF_EM_RAW_GUEST_TRAP
15953	\|\| rcStrict == VINF_EM_RAW_TO_R3
15954	\|\| rcStrict == VINF_TRPM_XCPT_DISPATCHED, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15955	# endif
15956	}
15957	#endif
15958	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitIO, y1);
15959	}
15960	else
15961	{
15962	/*
15963	* Frequent exit or something needing probing. Get state and call EMHistoryExec.
15964	*/
15965	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
15966	AssertRCReturn(rc2, rc2);
15967	STAM_COUNTER_INC(!fIOString ? fIOWrite ? &pVCpu->hm.s.StatExitIOWrite : &pVCpu->hm.s.StatExitIORead
15968	: fIOWrite ? &pVCpu->hm.s.StatExitIOStringWrite : &pVCpu->hm.s.StatExitIOStringRead);
15969	Log4(("IOExit/%u: %04x:%08RX64: %s%s%s %#x LB %u -> EMHistoryExec\n",
15970	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
15971	VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual) ? "REP " : "",
15972	fIOWrite ? "OUT" : "IN", fIOString ? "S" : "", uIOPort, uIOSize));
15973
15974	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
15975	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
15976
15977	Log4(("IOExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
15978	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
15979	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
15980	}
15981	return rcStrict;
15982	}
15983
15984
15985	/**
15986	* VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH). Unconditional
15987	* VM-exit.
15988	*/
15989	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitch(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15990	{
15991	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15992
15993	/* Check if this task-switch occurred while delivery an event through the guest IDT. */
15994	hmR0VmxReadExitQualVmcs(pVmxTransient);
15995	if (VMX_EXIT_QUAL_TASK_SWITCH_TYPE(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_TASK_SWITCH_TYPE_IDT)
15996	{
15997	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
15998	if (VMX_IDT_VECTORING_INFO_IS_VALID(pVmxTransient->uIdtVectoringInfo))
15999	{
16000	uint32_t uErrCode;
16001	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(pVmxTransient->uIdtVectoringInfo))
16002	{
16003	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16004	uErrCode = pVmxTransient->uIdtVectoringErrorCode;
16005	}
16006	else
16007	uErrCode = 0;
16008
16009	RTGCUINTPTR GCPtrFaultAddress;
16010	if (VMX_IDT_VECTORING_INFO_IS_XCPT_PF(pVmxTransient->uIdtVectoringInfo))
16011	GCPtrFaultAddress = pVCpu->cpum.GstCtx.cr2;
16012	else
16013	GCPtrFaultAddress = 0;
16014
16015	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16016
16017	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(pVmxTransient->uIdtVectoringInfo),
16018	pVmxTransient->cbExitInstr, uErrCode, GCPtrFaultAddress);
16019
16020	Log4Func(("Pending event. uIntType=%#x uVector=%#x\n", VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo),
16021	VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo)));
16022	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
16023	return VINF_EM_RAW_INJECT_TRPM_EVENT;
16024	}
16025	}
16026
16027	/* Fall back to the interpreter to emulate the task-switch. */
16028	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
16029	return VERR_EM_INTERPRETER;
16030	}
16031
16032
16033	/**
16034	* VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional VM-exit.
16035	*/
16036	HMVMX_EXIT_DECL hmR0VmxExitMtf(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16037	{
16038	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16039
16040	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16041	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_MONITOR_TRAP_FLAG;
16042	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
16043	AssertRC(rc);
16044	return VINF_EM_DBG_STEPPED;
16045	}
16046
16047
16048	/**
16049	* VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional VM-exit.
16050	*/
16051	HMVMX_EXIT_DECL hmR0VmxExitApicAccess(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16052	{
16053	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16054	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitApicAccess);
16055
16056	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16057	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16058	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16059	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16060	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16061
16062	/*
16063	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16064	*/
16065	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16066	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16067	{
16068	/* For some crazy guest, if an event delivery causes an APIC-access VM-exit, go to instruction emulation. */
16069	if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
16070	{
16071	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
16072	return VINF_EM_RAW_INJECT_TRPM_EVENT;
16073	}
16074	}
16075	else
16076	{
16077	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16078	return rcStrict;
16079	}
16080
16081	/* IOMMIOPhysHandler() below may call into IEM, save the necessary state. */
16082	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16083	hmR0VmxReadExitQualVmcs(pVmxTransient);
16084	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16085	AssertRCReturn(rc, rc);
16086
16087	/* See Intel spec. 27-6 "Exit Qualifications for APIC-access VM-exits from Linear Accesses & Guest-Phyiscal Addresses" */
16088	uint32_t const uAccessType = VMX_EXIT_QUAL_APIC_ACCESS_TYPE(pVmxTransient->uExitQual);
16089	switch (uAccessType)
16090	{
16091	case VMX_APIC_ACCESS_TYPE_LINEAR_WRITE:
16092	case VMX_APIC_ACCESS_TYPE_LINEAR_READ:
16093	{
16094	AssertMsg( !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
16095	\|\| VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual) != XAPIC_OFF_TPR,
16096	("hmR0VmxExitApicAccess: can't access TPR offset while using TPR shadowing.\n"));
16097
16098	RTGCPHYS GCPhys = pVCpu->hm.s.vmx.u64GstMsrApicBase; /* Always up-to-date, as it is not part of the VMCS. */
16099	GCPhys &= PAGE_BASE_GC_MASK;
16100	GCPhys += VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual);
16101	Log4Func(("Linear access uAccessType=%#x GCPhys=%#RGp Off=%#x\n", uAccessType, GCPhys,
16102	VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual)));
16103
16104	rcStrict = IOMR0MmioPhysHandler(pVCpu->CTX_SUFF(pVM), pVCpu,
16105	uAccessType == VMX_APIC_ACCESS_TYPE_LINEAR_READ ? 0 : X86_TRAP_PF_RW, GCPhys);
16106	Log4Func(("IOMMMIOPhysHandler returned %Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
16107	if ( rcStrict == VINF_SUCCESS
16108	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16109	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16110	{
16111	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS
16112	\| HM_CHANGED_GUEST_APIC_TPR);
16113	rcStrict = VINF_SUCCESS;
16114	}
16115	break;
16116	}
16117
16118	default:
16119	{
16120	Log4Func(("uAccessType=%#x\n", uAccessType));
16121	rcStrict = VINF_EM_RAW_EMULATE_INSTR;
16122	break;
16123	}
16124	}
16125
16126	if (rcStrict != VINF_SUCCESS)
16127	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchApicAccessToR3);
16128	return rcStrict;
16129	}
16130
16131
16132	/**
16133	* VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX). Conditional
16134	* VM-exit.
16135	*/
16136	HMVMX_EXIT_DECL hmR0VmxExitMovDRx(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16137	{
16138	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16139	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16140
16141	/*
16142	* We might also get this VM-exit if the nested-guest isn't intercepting MOV DRx accesses.
16143	* In such a case, rather than disabling MOV DRx intercepts and resuming execution, we
16144	* must emulate the MOV DRx access.
16145	*/
16146	if (!pVmxTransient->fIsNestedGuest)
16147	{
16148	/* We should -not- get this VM-exit if the guest's debug registers were active. */
16149	if (pVmxTransient->fWasGuestDebugStateActive)
16150	{
16151	AssertMsgFailed(("Unexpected MOV DRx exit\n"));
16152	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
16153	}
16154
16155	if ( !pVCpu->hm.s.fSingleInstruction
16156	&& !pVmxTransient->fWasHyperDebugStateActive)
16157	{
16158	Assert(!DBGFIsStepping(pVCpu));
16159	Assert(pVmcsInfo->u32XcptBitmap & RT_BIT(X86_XCPT_DB));
16160
16161	/* Don't intercept MOV DRx any more. */
16162	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_MOV_DR_EXIT;
16163	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
16164	AssertRC(rc);
16165
16166	/* We're playing with the host CPU state here, make sure we can't preempt or longjmp. */
16167	VMMRZCallRing3Disable(pVCpu);
16168	HM_DISABLE_PREEMPT(pVCpu);
16169
16170	/* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */
16171	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
16172	Assert(CPUMIsGuestDebugStateActive(pVCpu));
16173
16174	HM_RESTORE_PREEMPT();
16175	VMMRZCallRing3Enable(pVCpu);
16176
16177	#ifdef VBOX_WITH_STATISTICS
16178	hmR0VmxReadExitQualVmcs(pVmxTransient);
16179	if (VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_DRX_DIRECTION_WRITE)
16180	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
16181	else
16182	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
16183	#endif
16184	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch);
16185	return VINF_SUCCESS;
16186	}
16187	}
16188
16189	/*
16190	* EMInterpretDRx[Write\|Read]() calls CPUMIsGuestIn64BitCode() which requires EFER MSR, CS.
16191	* The EFER MSR is always up-to-date.
16192	* Update the segment registers and DR7 from the CPU.
16193	*/
16194	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16195	hmR0VmxReadExitQualVmcs(pVmxTransient);
16196	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_DR7);
16197	AssertRCReturn(rc, rc);
16198	Log4Func(("cs:rip=%#04x:%#RX64\n", pCtx->cs.Sel, pCtx->rip));
16199
16200	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16201	if (VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_DRX_DIRECTION_WRITE)
16202	{
16203	rc = EMInterpretDRxWrite(pVM, pVCpu, CPUMCTX2CORE(pCtx),
16204	VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual),
16205	VMX_EXIT_QUAL_DRX_GENREG(pVmxTransient->uExitQual));
16206	if (RT_SUCCESS(rc))
16207	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
16208	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
16209	}
16210	else
16211	{
16212	rc = EMInterpretDRxRead(pVM, pVCpu, CPUMCTX2CORE(pCtx),
16213	VMX_EXIT_QUAL_DRX_GENREG(pVmxTransient->uExitQual),
16214	VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual));
16215	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
16216	}
16217
16218	Assert(rc == VINF_SUCCESS \|\| rc == VERR_EM_INTERPRETER);
16219	if (RT_SUCCESS(rc))
16220	{
16221	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
16222	AssertRCReturn(rc2, rc2);
16223	return VINF_SUCCESS;
16224	}
16225	return rc;
16226	}
16227
16228
16229	/**
16230	* VM-exit handler for EPT misconfiguration (VMX_EXIT_EPT_MISCONFIG).
16231	* Conditional VM-exit.
16232	*/
16233	HMVMX_EXIT_DECL hmR0VmxExitEptMisconfig(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16234	{
16235	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16236	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
16237
16238	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16239	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16240	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16241	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16242	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16243
16244	/*
16245	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16246	*/
16247	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16248	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16249	{
16250	/*
16251	* In the unlikely case where delivering an event causes an EPT misconfig (MMIO), go back to
16252	* instruction emulation to inject the original event. Otherwise, injecting the original event
16253	* using hardware-assisted VMX would trigger the same EPT misconfig VM-exit again.
16254	*/
16255	if (!pVCpu->hm.s.Event.fPending)
16256	{ /* likely */ }
16257	else
16258	{
16259	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
16260	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16261	/** @todo NSTVMX: Think about how this should be handled. */
16262	if (pVmxTransient->fIsNestedGuest)
16263	return VERR_VMX_IPE_3;
16264	#endif
16265	return VINF_EM_RAW_INJECT_TRPM_EVENT;
16266	}
16267	}
16268	else
16269	{
16270	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16271	return rcStrict;
16272	}
16273
16274	/*
16275	* Get sufficient state and update the exit history entry.
16276	*/
16277	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16278	hmR0VmxReadGuestPhysicalAddrVmcs(pVmxTransient);
16279	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16280	AssertRCReturn(rc, rc);
16281
16282	RTGCPHYS const GCPhys = pVmxTransient->uGuestPhysicalAddr;
16283	PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
16284	EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_MMIO),
16285	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
16286	if (!pExitRec)
16287	{
16288	/*
16289	* If we succeed, resume guest execution.
16290	* If we fail in interpreting the instruction because we couldn't get the guest physical address
16291	* of the page containing the instruction via the guest's page tables (we would invalidate the guest page
16292	* in the host TLB), resume execution which would cause a guest page fault to let the guest handle this
16293	* weird case. See @bugref{6043}.
16294	*/
16295	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16296	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16297	/** @todo bird: We can probably just go straight to IOM here and assume that
16298	* it's MMIO, then fall back on PGM if that hunch didn't work out so
16299	* well. However, we need to address that aliasing workarounds that
16300	* PGMR0Trap0eHandlerNPMisconfig implements. So, some care is needed.
16301	*
16302	* Might also be interesting to see if we can get this done more or
16303	* less locklessly inside IOM. Need to consider the lookup table
16304	* updating and use a bit more carefully first (or do all updates via
16305	* rendezvous) */
16306	rcStrict = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, PGMMODE_EPT, CPUMCTX2CORE(pCtx), GCPhys, UINT32_MAX);
16307	Log4Func(("At %#RGp RIP=%#RX64 rc=%Rrc\n", GCPhys, pCtx->rip, VBOXSTRICTRC_VAL(rcStrict)));
16308	if ( rcStrict == VINF_SUCCESS
16309	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16310	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16311	{
16312	/* Successfully handled MMIO operation. */
16313	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS
16314	\| HM_CHANGED_GUEST_APIC_TPR);
16315	rcStrict = VINF_SUCCESS;
16316	}
16317	}
16318	else
16319	{
16320	/*
16321	* Frequent exit or something needing probing. Call EMHistoryExec.
16322	*/
16323	Log4(("EptMisscfgExit/%u: %04x:%08RX64: %RGp -> EMHistoryExec\n",
16324	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, GCPhys));
16325
16326	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
16327	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16328
16329	Log4(("EptMisscfgExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
16330	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
16331	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
16332	}
16333	return rcStrict;
16334	}
16335
16336
16337	/**
16338	* VM-exit handler for EPT violation (VMX_EXIT_EPT_VIOLATION). Conditional
16339	* VM-exit.
16340	*/
16341	HMVMX_EXIT_DECL hmR0VmxExitEptViolation(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16342	{
16343	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16344	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
16345
16346	hmR0VmxReadExitQualVmcs(pVmxTransient);
16347	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16348	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16349	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16350	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16351	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16352
16353	/*
16354	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16355	*/
16356	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16357	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16358	{
16359	/*
16360	* If delivery of an event causes an EPT violation (true nested #PF and not MMIO),
16361	* we shall resolve the nested #PF and re-inject the original event.
16362	*/
16363	if (pVCpu->hm.s.Event.fPending)
16364	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflectNPF);
16365	}
16366	else
16367	{
16368	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16369	return rcStrict;
16370	}
16371
16372	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16373	hmR0VmxReadGuestPhysicalAddrVmcs(pVmxTransient);
16374	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16375	AssertRCReturn(rc, rc);
16376
16377	RTGCPHYS const GCPhys = pVmxTransient->uGuestPhysicalAddr;
16378	uint64_t const uExitQual = pVmxTransient->uExitQual;
16379	AssertMsg(((pVmxTransient->uExitQual >> 7) & 3) != 2, ("%#RX64", uExitQual));
16380
16381	RTGCUINT uErrorCode = 0;
16382	if (uExitQual & VMX_EXIT_QUAL_EPT_INSTR_FETCH)
16383	uErrorCode \|= X86_TRAP_PF_ID;
16384	if (uExitQual & VMX_EXIT_QUAL_EPT_DATA_WRITE)
16385	uErrorCode \|= X86_TRAP_PF_RW;
16386	if (uExitQual & VMX_EXIT_QUAL_EPT_ENTRY_PRESENT)
16387	uErrorCode \|= X86_TRAP_PF_P;
16388
16389	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16390	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16391	Log4Func(("at %#RX64 (%#RX64 errcode=%#x) cs:rip=%#04x:%#RX64\n", GCPhys, uExitQual, uErrorCode, pCtx->cs.Sel, pCtx->rip));
16392
16393	/*
16394	* Handle the pagefault trap for the nested shadow table.
16395	*/
16396	TRPMAssertXcptPF(pVCpu, GCPhys, uErrorCode);
16397	rcStrict = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, PGMMODE_EPT, uErrorCode, CPUMCTX2CORE(pCtx), GCPhys);
16398	TRPMResetTrap(pVCpu);
16399
16400	/* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}. */
16401	if ( rcStrict == VINF_SUCCESS
16402	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16403	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16404	{
16405	/* Successfully synced our nested page tables. */
16406	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf);
16407	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS);
16408	return VINF_SUCCESS;
16409	}
16410
16411	Log4Func(("EPT return to ring-3 rcStrict2=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
16412	return rcStrict;
16413	}
16414
16415
16416	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16417	/**
16418	* VM-exit handler for VMCLEAR (VMX_EXIT_VMCLEAR). Unconditional VM-exit.
16419	*/
16420	HMVMX_EXIT_DECL hmR0VmxExitVmclear(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16421	{
16422	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16423
16424	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16425	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16426	hmR0VmxReadExitQualVmcs(pVmxTransient);
16427	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16428	\| CPUMCTX_EXTRN_HWVIRT
16429	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16430	AssertRCReturn(rc, rc);
16431
16432	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16433
16434	VMXVEXITINFO ExitInfo;
16435	RT_ZERO(ExitInfo);
16436	ExitInfo.uReason = pVmxTransient->uExitReason;
16437	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16438	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16439	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16440	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16441
16442	VBOXSTRICTRC rcStrict = IEMExecDecodedVmclear(pVCpu, &ExitInfo);
16443	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16444	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16445	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16446	{
16447	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16448	rcStrict = VINF_SUCCESS;
16449	}
16450	return rcStrict;
16451	}
16452
16453
16454	/**
16455	* VM-exit handler for VMLAUNCH (VMX_EXIT_VMLAUNCH). Unconditional VM-exit.
16456	*/
16457	HMVMX_EXIT_DECL hmR0VmxExitVmlaunch(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16458	{
16459	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16460
16461	/* Import the entire VMCS state for now as we would be switching VMCS on successful VMLAUNCH,
16462	otherwise we could import just IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK. */
16463	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16464	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16465	AssertRCReturn(rc, rc);
16466
16467	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16468
16469	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
16470	VBOXSTRICTRC rcStrict = IEMExecDecodedVmlaunchVmresume(pVCpu, pVmxTransient->cbExitInstr, VMXINSTRID_VMLAUNCH);
16471	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
16472	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16473	{
16474	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16475	if (CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
16476	rcStrict = VINF_VMX_VMLAUNCH_VMRESUME;
16477	}
16478	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
16479	return rcStrict;
16480	}
16481
16482
16483	/**
16484	* VM-exit handler for VMPTRLD (VMX_EXIT_VMPTRLD). Unconditional VM-exit.
16485	*/
16486	HMVMX_EXIT_DECL hmR0VmxExitVmptrld(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16487	{
16488	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16489
16490	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16491	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16492	hmR0VmxReadExitQualVmcs(pVmxTransient);
16493	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16494	\| CPUMCTX_EXTRN_HWVIRT
16495	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16496	AssertRCReturn(rc, rc);
16497
16498	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16499
16500	VMXVEXITINFO ExitInfo;
16501	RT_ZERO(ExitInfo);
16502	ExitInfo.uReason = pVmxTransient->uExitReason;
16503	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16504	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16505	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16506	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16507
16508	VBOXSTRICTRC rcStrict = IEMExecDecodedVmptrld(pVCpu, &ExitInfo);
16509	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16510	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16511	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16512	{
16513	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16514	rcStrict = VINF_SUCCESS;
16515	}
16516	return rcStrict;
16517	}
16518
16519
16520	/**
16521	* VM-exit handler for VMPTRST (VMX_EXIT_VMPTRST). Unconditional VM-exit.
16522	*/
16523	HMVMX_EXIT_DECL hmR0VmxExitVmptrst(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16524	{
16525	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16526
16527	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16528	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16529	hmR0VmxReadExitQualVmcs(pVmxTransient);
16530	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16531	\| CPUMCTX_EXTRN_HWVIRT
16532	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16533	AssertRCReturn(rc, rc);
16534
16535	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16536
16537	VMXVEXITINFO ExitInfo;
16538	RT_ZERO(ExitInfo);
16539	ExitInfo.uReason = pVmxTransient->uExitReason;
16540	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16541	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16542	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16543	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_WRITE, &ExitInfo.GCPtrEffAddr);
16544
16545	VBOXSTRICTRC rcStrict = IEMExecDecodedVmptrst(pVCpu, &ExitInfo);
16546	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16547	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16548	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16549	{
16550	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16551	rcStrict = VINF_SUCCESS;
16552	}
16553	return rcStrict;
16554	}
16555
16556
16557	/**
16558	* VM-exit handler for VMREAD (VMX_EXIT_VMREAD). Conditional VM-exit.
16559	*/
16560	HMVMX_EXIT_DECL hmR0VmxExitVmread(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16561	{
16562	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16563
16564	/*
16565	* Strictly speaking we should not get VMREAD VM-exits for shadow VMCS fields and
16566	* thus might not need to import the shadow VMCS state, it's safer just in case
16567	* code elsewhere dares look at unsynced VMCS fields.
16568	*/
16569	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16570	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16571	hmR0VmxReadExitQualVmcs(pVmxTransient);
16572	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16573	\| CPUMCTX_EXTRN_HWVIRT
16574	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16575	AssertRCReturn(rc, rc);
16576
16577	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16578
16579	VMXVEXITINFO ExitInfo;
16580	RT_ZERO(ExitInfo);
16581	ExitInfo.uReason = pVmxTransient->uExitReason;
16582	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16583	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16584	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16585	if (!ExitInfo.InstrInfo.VmreadVmwrite.fIsRegOperand)
16586	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_WRITE, &ExitInfo.GCPtrEffAddr);
16587
16588	VBOXSTRICTRC rcStrict = IEMExecDecodedVmread(pVCpu, &ExitInfo);
16589	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16590	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16591	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16592	{
16593	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16594	rcStrict = VINF_SUCCESS;
16595	}
16596	return rcStrict;
16597	}
16598
16599
16600	/**
16601	* VM-exit handler for VMRESUME (VMX_EXIT_VMRESUME). Unconditional VM-exit.
16602	*/
16603	HMVMX_EXIT_DECL hmR0VmxExitVmresume(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16604	{
16605	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16606
16607	/* Import the entire VMCS state for now as we would be switching VMCS on successful VMRESUME,
16608	otherwise we could import just IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK. */
16609	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16610	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16611	AssertRCReturn(rc, rc);
16612
16613	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16614
16615	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
16616	VBOXSTRICTRC rcStrict = IEMExecDecodedVmlaunchVmresume(pVCpu, pVmxTransient->cbExitInstr, VMXINSTRID_VMRESUME);
16617	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
16618	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16619	{
16620	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16621	if (CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
16622	rcStrict = VINF_VMX_VMLAUNCH_VMRESUME;
16623	}
16624	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
16625	return rcStrict;
16626	}
16627
16628
16629	/**
16630	* VM-exit handler for VMWRITE (VMX_EXIT_VMWRITE). Conditional VM-exit.
16631	*/
16632	HMVMX_EXIT_DECL hmR0VmxExitVmwrite(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16633	{
16634	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16635
16636	/*
16637	* Although we should not get VMWRITE VM-exits for shadow VMCS fields, since our HM hook
16638	* gets invoked when IEM's VMWRITE instruction emulation modifies the current VMCS and it
16639	* flags re-loading the entire shadow VMCS, we should save the entire shadow VMCS here.
16640	*/
16641	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16642	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16643	hmR0VmxReadExitQualVmcs(pVmxTransient);
16644	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16645	\| CPUMCTX_EXTRN_HWVIRT
16646	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16647	AssertRCReturn(rc, rc);
16648
16649	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16650
16651	VMXVEXITINFO ExitInfo;
16652	RT_ZERO(ExitInfo);
16653	ExitInfo.uReason = pVmxTransient->uExitReason;
16654	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16655	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16656	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16657	if (!ExitInfo.InstrInfo.VmreadVmwrite.fIsRegOperand)
16658	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16659
16660	VBOXSTRICTRC rcStrict = IEMExecDecodedVmwrite(pVCpu, &ExitInfo);
16661	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16662	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16663	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16664	{
16665	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16666	rcStrict = VINF_SUCCESS;
16667	}
16668	return rcStrict;
16669	}
16670
16671
16672	/**
16673	* VM-exit handler for VMXOFF (VMX_EXIT_VMXOFF). Unconditional VM-exit.
16674	*/
16675	HMVMX_EXIT_DECL hmR0VmxExitVmxoff(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16676	{
16677	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16678
16679	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16680	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR4
16681	\| CPUMCTX_EXTRN_HWVIRT
16682	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
16683	AssertRCReturn(rc, rc);
16684
16685	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16686
16687	VBOXSTRICTRC rcStrict = IEMExecDecodedVmxoff(pVCpu, pVmxTransient->cbExitInstr);
16688	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16689	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_HWVIRT);
16690	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16691	{
16692	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16693	rcStrict = VINF_SUCCESS;
16694	}
16695	return rcStrict;
16696	}
16697
16698
16699	/**
16700	* VM-exit handler for VMXON (VMX_EXIT_VMXON). Unconditional VM-exit.
16701	*/
16702	HMVMX_EXIT_DECL hmR0VmxExitVmxon(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16703	{
16704	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16705
16706	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16707	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16708	hmR0VmxReadExitQualVmcs(pVmxTransient);
16709	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16710	\| CPUMCTX_EXTRN_HWVIRT
16711	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16712	AssertRCReturn(rc, rc);
16713
16714	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16715
16716	VMXVEXITINFO ExitInfo;
16717	RT_ZERO(ExitInfo);
16718	ExitInfo.uReason = pVmxTransient->uExitReason;
16719	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16720	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16721	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16722	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16723
16724	VBOXSTRICTRC rcStrict = IEMExecDecodedVmxon(pVCpu, &ExitInfo);
16725	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16726	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16727	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16728	{
16729	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16730	rcStrict = VINF_SUCCESS;
16731	}
16732	return rcStrict;
16733	}
16734
16735
16736	/**
16737	* VM-exit handler for INVVPID (VMX_EXIT_INVVPID). Unconditional VM-exit.
16738	*/
16739	HMVMX_EXIT_DECL hmR0VmxExitInvvpid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16740	{
16741	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16742
16743	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16744	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16745	hmR0VmxReadExitQualVmcs(pVmxTransient);
16746	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16747	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16748	AssertRCReturn(rc, rc);
16749
16750	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16751
16752	VMXVEXITINFO ExitInfo;
16753	RT_ZERO(ExitInfo);
16754	ExitInfo.uReason = pVmxTransient->uExitReason;
16755	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16756	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16757	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16758	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16759
16760	VBOXSTRICTRC rcStrict = IEMExecDecodedInvvpid(pVCpu, &ExitInfo);
16761	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16762	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16763	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16764	{
16765	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16766	rcStrict = VINF_SUCCESS;
16767	}
16768	return rcStrict;
16769	}
16770	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
16771	/** @} */
16772
16773
16774	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16775	/** @name Nested-guest VM-exit handlers.
16776	* @{
16777	*/
16778	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16779	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Nested-guest VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16780	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16781
16782	/**
16783	* Nested-guest VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI).
16784	* Conditional VM-exit.
16785	*/
16786	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmiNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16787	{
16788	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16789
16790	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16791
16792	uint64_t const uExitIntInfo = pVmxTransient->uExitIntInfo;
16793	uint32_t const uExitIntType = VMX_EXIT_INT_INFO_TYPE(uExitIntInfo);
16794	Assert(VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo));
16795
16796	switch (uExitIntType)
16797	{
16798	/*
16799	* Physical NMIs:
16800	* We shouldn't direct host physical NMIs to the nested-guest. Dispatch it to the host.
16801	*/
16802	case VMX_EXIT_INT_INFO_TYPE_NMI:
16803	return hmR0VmxExitHostNmi(pVCpu, pVmxTransient->pVmcsInfo);
16804
16805	/*
16806	* Hardware exceptions,
16807	* Software exceptions,
16808	* Privileged software exceptions:
16809	* Figure out if the exception must be delivered to the guest or the nested-guest.
16810	*/
16811	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
16812	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
16813	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
16814	{
16815	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16816	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16817	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16818	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16819
16820	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16821	bool const fIntercept = CPUMIsGuestVmxXcptInterceptSet(pCtx, VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo),
16822	pVmxTransient->uExitIntErrorCode);
16823	if (fIntercept)
16824	{
16825	/* Exit qualification is required for debug and page-fault exceptions. */
16826	hmR0VmxReadExitQualVmcs(pVmxTransient);
16827
16828	/*
16829	* For VM-exits due to software exceptions (those generated by INT3 or INTO) and privileged
16830	* software exceptions (those generated by INT1/ICEBP) we need to supply the VM-exit instruction
16831	* length. However, if delivery of a software interrupt, software exception or privileged
16832	* software exception causes a VM-exit, that too provides the VM-exit instruction length.
16833	*/
16834	VMXVEXITINFO ExitInfo;
16835	RT_ZERO(ExitInfo);
16836	ExitInfo.uReason = pVmxTransient->uExitReason;
16837	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16838	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16839
16840	VMXVEXITEVENTINFO ExitEventInfo;
16841	RT_ZERO(ExitEventInfo);
16842	ExitEventInfo.uExitIntInfo = pVmxTransient->uExitIntInfo;
16843	ExitEventInfo.uExitIntErrCode = pVmxTransient->uExitIntErrorCode;
16844	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
16845	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
16846
16847	#ifdef DEBUG_ramshankar
16848	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16849	Log4Func(("exit_int_info=%#RX32 err_code=%#RX32 exit_qual=%#RX64\n", pVmxTransient->uExitIntInfo,
16850	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual));
16851	if (VMX_IDT_VECTORING_INFO_IS_VALID(pVmxTransient->uIdtVectoringInfo))
16852	{
16853	Log4Func(("idt_info=%#RX32 idt_errcode=%#RX32 cr2=%#RX64\n", pVmxTransient->uIdtVectoringInfo,
16854	pVmxTransient->uIdtVectoringErrorCode, pCtx->cr2));
16855	}
16856	#endif
16857	return IEMExecVmxVmexitXcpt(pVCpu, &ExitInfo, &ExitEventInfo);
16858	}
16859
16860	/* Nested paging is currently a requirement, otherwise we would need to handle shadow #PFs in hmR0VmxExitXcptPF. */
16861	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
16862	return hmR0VmxExitXcpt(pVCpu, pVmxTransient);
16863	}
16864
16865	/*
16866	* Software interrupts:
16867	* VM-exits cannot be caused by software interrupts.
16868	*
16869	* External interrupts:
16870	* This should only happen when "acknowledge external interrupts on VM-exit"
16871	* control is set. However, we never set this when executing a guest or
16872	* nested-guest. For nested-guests it is emulated while injecting interrupts into
16873	* the guest.
16874	*/
16875	case VMX_EXIT_INT_INFO_TYPE_SW_INT:
16876	case VMX_EXIT_INT_INFO_TYPE_EXT_INT:
16877	default:
16878	{
16879	pVCpu->hm.s.u32HMError = pVmxTransient->uExitIntInfo;
16880	return VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
16881	}
16882	}
16883	}
16884
16885
16886	/**
16887	* Nested-guest VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT).
16888	* Unconditional VM-exit.
16889	*/
16890	HMVMX_EXIT_DECL hmR0VmxExitTripleFaultNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16891	{
16892	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16893	return IEMExecVmxVmexitTripleFault(pVCpu);
16894	}
16895
16896
16897	/**
16898	* Nested-guest VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
16899	*/
16900	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindowNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16901	{
16902	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16903
16904	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INT_WINDOW_EXIT))
16905	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
16906	return hmR0VmxExitIntWindow(pVCpu, pVmxTransient);
16907	}
16908
16909
16910	/**
16911	* Nested-guest VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
16912	*/
16913	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindowNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16914	{
16915	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16916
16917	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_NMI_WINDOW_EXIT))
16918	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
16919	return hmR0VmxExitIntWindow(pVCpu, pVmxTransient);
16920	}
16921
16922
16923	/**
16924	* Nested-guest VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH).
16925	* Unconditional VM-exit.
16926	*/
16927	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitchNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16928	{
16929	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16930
16931	hmR0VmxReadExitQualVmcs(pVmxTransient);
16932	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16933	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16934	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16935
16936	VMXVEXITINFO ExitInfo;
16937	RT_ZERO(ExitInfo);
16938	ExitInfo.uReason = pVmxTransient->uExitReason;
16939	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16940	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16941
16942	VMXVEXITEVENTINFO ExitEventInfo;
16943	RT_ZERO(ExitEventInfo);
16944	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
16945	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
16946	return IEMExecVmxVmexitTaskSwitch(pVCpu, &ExitInfo, &ExitEventInfo);
16947	}
16948
16949
16950	/**
16951	* Nested-guest VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
16952	*/
16953	HMVMX_EXIT_DECL hmR0VmxExitHltNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16954	{
16955	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16956
16957	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_HLT_EXIT))
16958	{
16959	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16960	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16961	}
16962	return hmR0VmxExitHlt(pVCpu, pVmxTransient);
16963	}
16964
16965
16966	/**
16967	* Nested-guest VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
16968	*/
16969	HMVMX_EXIT_DECL hmR0VmxExitInvlpgNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16970	{
16971	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16972
16973	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INVLPG_EXIT))
16974	{
16975	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16976	hmR0VmxReadExitQualVmcs(pVmxTransient);
16977
16978	VMXVEXITINFO ExitInfo;
16979	RT_ZERO(ExitInfo);
16980	ExitInfo.uReason = pVmxTransient->uExitReason;
16981	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16982	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16983	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16984	}
16985	return hmR0VmxExitInvlpg(pVCpu, pVmxTransient);
16986	}
16987
16988
16989	/**
16990	* Nested-guest VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
16991	*/
16992	HMVMX_EXIT_DECL hmR0VmxExitRdpmcNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16993	{
16994	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16995
16996	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDPMC_EXIT))
16997	{
16998	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16999	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17000	}
17001	return hmR0VmxExitRdpmc(pVCpu, pVmxTransient);
17002	}
17003
17004
17005	/**
17006	* Nested-guest VM-exit handler for VMREAD (VMX_EXIT_VMREAD) and VMWRITE
17007	* (VMX_EXIT_VMWRITE). Conditional VM-exit.
17008	*/
17009	HMVMX_EXIT_DECL hmR0VmxExitVmreadVmwriteNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17010	{
17011	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17012
17013	Assert( pVmxTransient->uExitReason == VMX_EXIT_VMREAD
17014	\|\| pVmxTransient->uExitReason == VMX_EXIT_VMWRITE);
17015
17016	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17017
17018	uint8_t const iGReg = pVmxTransient->ExitInstrInfo.VmreadVmwrite.iReg2;
17019	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
17020	uint64_t u64VmcsField = pVCpu->cpum.GstCtx.aGRegs[iGReg].u64;
17021
17022	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
17023	if (!CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
17024	u64VmcsField &= UINT64_C(0xffffffff);
17025
17026	if (CPUMIsGuestVmxVmreadVmwriteInterceptSet(pVCpu, pVmxTransient->uExitReason, u64VmcsField))
17027	{
17028	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17029	hmR0VmxReadExitQualVmcs(pVmxTransient);
17030
17031	VMXVEXITINFO ExitInfo;
17032	RT_ZERO(ExitInfo);
17033	ExitInfo.uReason = pVmxTransient->uExitReason;
17034	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17035	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17036	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17037	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17038	}
17039
17040	if (pVmxTransient->uExitReason == VMX_EXIT_VMREAD)
17041	return hmR0VmxExitVmread(pVCpu, pVmxTransient);
17042	return hmR0VmxExitVmwrite(pVCpu, pVmxTransient);
17043	}
17044
17045
17046	/**
17047	* Nested-guest VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
17048	*/
17049	HMVMX_EXIT_DECL hmR0VmxExitRdtscNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17050	{
17051	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17052
17053	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDTSC_EXIT))
17054	{
17055	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17056	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17057	}
17058
17059	return hmR0VmxExitRdtsc(pVCpu, pVmxTransient);
17060	}
17061
17062
17063	/**
17064	* Nested-guest VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX).
17065	* Conditional VM-exit.
17066	*/
17067	HMVMX_EXIT_DECL hmR0VmxExitMovCRxNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17068	{
17069	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17070
17071	hmR0VmxReadExitQualVmcs(pVmxTransient);
17072	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17073
17074	VBOXSTRICTRC rcStrict;
17075	uint32_t const uAccessType = VMX_EXIT_QUAL_CRX_ACCESS(pVmxTransient->uExitQual);
17076	switch (uAccessType)
17077	{
17078	case VMX_EXIT_QUAL_CRX_ACCESS_WRITE:
17079	{
17080	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
17081	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(pVmxTransient->uExitQual);
17082	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
17083	uint64_t const uNewCrX = pVCpu->cpum.GstCtx.aGRegs[iGReg].u64;
17084
17085	bool fIntercept;
17086	switch (iCrReg)
17087	{
17088	case 0:
17089	case 4:
17090	fIntercept = CPUMIsGuestVmxMovToCr0Cr4InterceptSet(&pVCpu->cpum.GstCtx, iCrReg, uNewCrX);
17091	break;
17092
17093	case 3:
17094	fIntercept = CPUMIsGuestVmxMovToCr3InterceptSet(pVCpu, uNewCrX);
17095	break;
17096
17097	case 8:
17098	fIntercept = CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_CR8_LOAD_EXIT);
17099	break;
17100
17101	default:
17102	fIntercept = false;
17103	break;
17104	}
17105	if (fIntercept)
17106	{
17107	VMXVEXITINFO ExitInfo;
17108	RT_ZERO(ExitInfo);
17109	ExitInfo.uReason = pVmxTransient->uExitReason;
17110	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17111	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17112	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17113	}
17114	else
17115	rcStrict = hmR0VmxExitMovToCrX(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
17116	break;
17117	}
17118
17119	case VMX_EXIT_QUAL_CRX_ACCESS_READ:
17120	{
17121	/*
17122	* CR0/CR4 reads do not cause VM-exits, the read-shadow is used (subject to masking).
17123	* CR2 reads do not cause a VM-exit.
17124	* CR3 reads cause a VM-exit depending on the "CR3 store exiting" control.
17125	* CR8 reads cause a VM-exit depending on the "CR8 store exiting" control.
17126	*/
17127	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
17128	if ( iCrReg == 3
17129	\|\| iCrReg == 8)
17130	{
17131	static const uint32_t s_auCrXReadIntercepts[] = { 0, 0, 0, VMX_PROC_CTLS_CR3_STORE_EXIT, 0,
17132	0, 0, 0, VMX_PROC_CTLS_CR8_STORE_EXIT };
17133	uint32_t const uIntercept = s_auCrXReadIntercepts[iCrReg];
17134	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, uIntercept))
17135	{
17136	VMXVEXITINFO ExitInfo;
17137	RT_ZERO(ExitInfo);
17138	ExitInfo.uReason = pVmxTransient->uExitReason;
17139	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17140	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17141	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17142	}
17143	else
17144	{
17145	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(pVmxTransient->uExitQual);
17146	rcStrict = hmR0VmxExitMovFromCrX(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
17147	}
17148	}
17149	else
17150	{
17151	AssertMsgFailed(("MOV from CR%d VM-exit must not happen\n", iCrReg));
17152	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, iCrReg);
17153	}
17154	break;
17155	}
17156
17157	case VMX_EXIT_QUAL_CRX_ACCESS_CLTS:
17158	{
17159	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
17160	Assert(pVmcsNstGst);
17161	uint64_t const uGstHostMask = pVmcsNstGst->u64Cr0Mask.u;
17162	uint64_t const uReadShadow = pVmcsNstGst->u64Cr0ReadShadow.u;
17163	if ( (uGstHostMask & X86_CR0_TS)
17164	&& (uReadShadow & X86_CR0_TS))
17165	{
17166	VMXVEXITINFO ExitInfo;
17167	RT_ZERO(ExitInfo);
17168	ExitInfo.uReason = pVmxTransient->uExitReason;
17169	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17170	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17171	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17172	}
17173	else
17174	rcStrict = hmR0VmxExitClts(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr);
17175	break;
17176	}
17177
17178	case VMX_EXIT_QUAL_CRX_ACCESS_LMSW: /* LMSW (Load Machine-Status Word into CR0) */
17179	{
17180	RTGCPTR GCPtrEffDst;
17181	uint16_t const uNewMsw = VMX_EXIT_QUAL_CRX_LMSW_DATA(pVmxTransient->uExitQual);
17182	bool const fMemOperand = VMX_EXIT_QUAL_CRX_LMSW_OP_MEM(pVmxTransient->uExitQual);
17183	if (fMemOperand)
17184	{
17185	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
17186	GCPtrEffDst = pVmxTransient->uGuestLinearAddr;
17187	}
17188	else
17189	GCPtrEffDst = NIL_RTGCPTR;
17190
17191	if (CPUMIsGuestVmxLmswInterceptSet(&pVCpu->cpum.GstCtx, uNewMsw))
17192	{
17193	VMXVEXITINFO ExitInfo;
17194	RT_ZERO(ExitInfo);
17195	ExitInfo.uReason = pVmxTransient->uExitReason;
17196	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17197	ExitInfo.u64GuestLinearAddr = GCPtrEffDst;
17198	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17199	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17200	}
17201	else
17202	rcStrict = hmR0VmxExitLmsw(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, uNewMsw, GCPtrEffDst);
17203	break;
17204	}
17205
17206	default:
17207	{
17208	AssertMsgFailed(("Unrecognized Mov CRX access type %#x\n", uAccessType));
17209	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, uAccessType);
17210	}
17211	}
17212
17213	if (rcStrict == VINF_IEM_RAISED_XCPT)
17214	{
17215	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
17216	rcStrict = VINF_SUCCESS;
17217	}
17218	return rcStrict;
17219	}
17220
17221
17222	/**
17223	* Nested-guest VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX).
17224	* Conditional VM-exit.
17225	*/
17226	HMVMX_EXIT_DECL hmR0VmxExitMovDRxNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17227	{
17228	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17229
17230	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MOV_DR_EXIT))
17231	{
17232	hmR0VmxReadExitQualVmcs(pVmxTransient);
17233	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17234
17235	VMXVEXITINFO ExitInfo;
17236	RT_ZERO(ExitInfo);
17237	ExitInfo.uReason = pVmxTransient->uExitReason;
17238	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17239	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17240	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17241	}
17242	return hmR0VmxExitMovDRx(pVCpu, pVmxTransient);
17243	}
17244
17245
17246	/**
17247	* Nested-guest VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR).
17248	* Conditional VM-exit.
17249	*/
17250	HMVMX_EXIT_DECL hmR0VmxExitIoInstrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17251	{
17252	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17253
17254	hmR0VmxReadExitQualVmcs(pVmxTransient);
17255
17256	uint32_t const uIOPort = VMX_EXIT_QUAL_IO_PORT(pVmxTransient->uExitQual);
17257	uint8_t const uIOSize = VMX_EXIT_QUAL_IO_SIZE(pVmxTransient->uExitQual);
17258	AssertReturn(uIOSize <= 3 && uIOSize != 2, VERR_VMX_IPE_1);
17259
17260	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses in bytes. */
17261	uint8_t const cbAccess = s_aIOSizes[uIOSize];
17262	if (CPUMIsGuestVmxIoInterceptSet(pVCpu, uIOPort, cbAccess))
17263	{
17264	/*
17265	* IN/OUT instruction:
17266	* - Provides VM-exit instruction length.
17267	*
17268	* INS/OUTS instruction:
17269	* - Provides VM-exit instruction length.
17270	* - Provides Guest-linear address.
17271	* - Optionally provides VM-exit instruction info (depends on CPU feature).
17272	*/
17273	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
17274	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17275
17276	/* Make sure we don't use stale/uninitialized VMX-transient info. below. */
17277	pVmxTransient->ExitInstrInfo.u = 0;
17278	pVmxTransient->uGuestLinearAddr = 0;
17279
17280	bool const fVmxInsOutsInfo = pVM->cpum.ro.GuestFeatures.fVmxInsOutInfo;
17281	bool const fIOString = VMX_EXIT_QUAL_IO_IS_STRING(pVmxTransient->uExitQual);
17282	if (fIOString)
17283	{
17284	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
17285	if (fVmxInsOutsInfo)
17286	{
17287	Assert(RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_INS_OUTS)); /* Paranoia. */
17288	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17289	}
17290	}
17291
17292	VMXVEXITINFO ExitInfo;
17293	RT_ZERO(ExitInfo);
17294	ExitInfo.uReason = pVmxTransient->uExitReason;
17295	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17296	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17297	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17298	ExitInfo.u64GuestLinearAddr = pVmxTransient->uGuestLinearAddr;
17299	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17300	}
17301	return hmR0VmxExitIoInstr(pVCpu, pVmxTransient);
17302	}
17303
17304
17305	/**
17306	* Nested-guest VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
17307	*/
17308	HMVMX_EXIT_DECL hmR0VmxExitRdmsrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17309	{
17310	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17311
17312	uint32_t fMsrpm;
17313	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_MSR_BITMAPS))
17314	fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), pVCpu->cpum.GstCtx.ecx);
17315	else
17316	fMsrpm = VMXMSRPM_EXIT_RD;
17317
17318	if (fMsrpm & VMXMSRPM_EXIT_RD)
17319	{
17320	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17321	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17322	}
17323	return hmR0VmxExitRdmsr(pVCpu, pVmxTransient);
17324	}
17325
17326
17327	/**
17328	* Nested-guest VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
17329	*/
17330	HMVMX_EXIT_DECL hmR0VmxExitWrmsrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17331	{
17332	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17333
17334	uint32_t fMsrpm;
17335	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_MSR_BITMAPS))
17336	fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), pVCpu->cpum.GstCtx.ecx);
17337	else
17338	fMsrpm = VMXMSRPM_EXIT_WR;
17339
17340	if (fMsrpm & VMXMSRPM_EXIT_WR)
17341	{
17342	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17343	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17344	}
17345	return hmR0VmxExitWrmsr(pVCpu, pVmxTransient);
17346	}
17347
17348
17349	/**
17350	* Nested-guest VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
17351	*/
17352	HMVMX_EXIT_DECL hmR0VmxExitMwaitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17353	{
17354	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17355
17356	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MWAIT_EXIT))
17357	{
17358	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17359	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17360	}
17361	return hmR0VmxExitMwait(pVCpu, pVmxTransient);
17362	}
17363
17364
17365	/**
17366	* Nested-guest VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional
17367	* VM-exit.
17368	*/
17369	HMVMX_EXIT_DECL hmR0VmxExitMtfNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17370	{
17371	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17372
17373	/** @todo NSTVMX: Should consider debugging nested-guests using VM debugger. */
17374	hmR0VmxReadGuestPendingDbgXctps(pVmxTransient);
17375	VMXVEXITINFO ExitInfo;
17376	RT_ZERO(ExitInfo);
17377	ExitInfo.uReason = pVmxTransient->uExitReason;
17378	ExitInfo.u64GuestPendingDbgXcpts = pVmxTransient->uGuestPendingDbgXcpts;
17379	return IEMExecVmxVmexitTrapLike(pVCpu, &ExitInfo);
17380	}
17381
17382
17383	/**
17384	* Nested-guest VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
17385	*/
17386	HMVMX_EXIT_DECL hmR0VmxExitMonitorNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17387	{
17388	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17389
17390	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MONITOR_EXIT))
17391	{
17392	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17393	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17394	}
17395	return hmR0VmxExitMonitor(pVCpu, pVmxTransient);
17396	}
17397
17398
17399	/**
17400	* Nested-guest VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
17401	*/
17402	HMVMX_EXIT_DECL hmR0VmxExitPauseNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17403	{
17404	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17405
17406	/** @todo NSTVMX: Think about this more. Does the outer guest need to intercept
17407	* PAUSE when executing a nested-guest? If it does not, we would not need
17408	* to check for the intercepts here. Just call VM-exit... */
17409
17410	/* The CPU would have already performed the necessary CPL checks for PAUSE-loop exiting. */
17411	if ( CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_PAUSE_EXIT)
17412	\|\| CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_PAUSE_LOOP_EXIT))
17413	{
17414	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17415	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17416	}
17417	return hmR0VmxExitPause(pVCpu, pVmxTransient);
17418	}
17419
17420
17421	/**
17422	* Nested-guest VM-exit handler for when the TPR value is lowered below the
17423	* specified threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
17424	*/
17425	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThresholdNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17426	{
17427	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17428
17429	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_TPR_SHADOW))
17430	{
17431	hmR0VmxReadGuestPendingDbgXctps(pVmxTransient);
17432	VMXVEXITINFO ExitInfo;
17433	RT_ZERO(ExitInfo);
17434	ExitInfo.uReason = pVmxTransient->uExitReason;
17435	ExitInfo.u64GuestPendingDbgXcpts = pVmxTransient->uGuestPendingDbgXcpts;
17436	return IEMExecVmxVmexitTrapLike(pVCpu, &ExitInfo);
17437	}
17438	return hmR0VmxExitTprBelowThreshold(pVCpu, pVmxTransient);
17439	}
17440
17441
17442	/**
17443	* Nested-guest VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional
17444	* VM-exit.
17445	*/
17446	HMVMX_EXIT_DECL hmR0VmxExitApicAccessNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17447	{
17448	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17449
17450	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17451	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
17452	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
17453	hmR0VmxReadExitQualVmcs(pVmxTransient);
17454
17455	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_VIRT_APIC_ACCESS));
17456
17457	Log4Func(("at offset %#x type=%u\n", VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual),
17458	VMX_EXIT_QUAL_APIC_ACCESS_TYPE(pVmxTransient->uExitQual)));
17459
17460	VMXVEXITINFO ExitInfo;
17461	RT_ZERO(ExitInfo);
17462	ExitInfo.uReason = pVmxTransient->uExitReason;
17463	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17464	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17465
17466	VMXVEXITEVENTINFO ExitEventInfo;
17467	RT_ZERO(ExitEventInfo);
17468	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
17469	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
17470	return IEMExecVmxVmexitApicAccess(pVCpu, &ExitInfo, &ExitEventInfo);
17471	}
17472
17473
17474	/**
17475	* Nested-guest VM-exit handler for APIC write emulation (VMX_EXIT_APIC_WRITE).
17476	* Conditional VM-exit.
17477	*/
17478	HMVMX_EXIT_DECL hmR0VmxExitApicWriteNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17479	{
17480	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17481
17482	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_APIC_REG_VIRT));
17483	hmR0VmxReadExitQualVmcs(pVmxTransient);
17484	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
17485	}
17486
17487
17488	/**
17489	* Nested-guest VM-exit handler for virtualized EOI (VMX_EXIT_VIRTUALIZED_EOI).
17490	* Conditional VM-exit.
17491	*/
17492	HMVMX_EXIT_DECL hmR0VmxExitVirtEoiNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17493	{
17494	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17495
17496	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_VIRT_INT_DELIVERY));
17497	hmR0VmxReadExitQualVmcs(pVmxTransient);
17498	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
17499	}
17500
17501
17502	/**
17503	* Nested-guest VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
17504	*/
17505	HMVMX_EXIT_DECL hmR0VmxExitRdtscpNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17506	{
17507	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17508
17509	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDTSC_EXIT))
17510	{
17511	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_RDTSCP));
17512	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17513	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17514	}
17515	return hmR0VmxExitRdtscp(pVCpu, pVmxTransient);
17516	}
17517
17518
17519	/**
17520	* Nested-guest VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
17521	*/
17522	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvdNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17523	{
17524	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17525
17526	if (CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_WBINVD_EXIT))
17527	{
17528	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17529	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17530	}
17531	return hmR0VmxExitWbinvd(pVCpu, pVmxTransient);
17532	}
17533
17534
17535	/**
17536	* Nested-guest VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
17537	*/
17538	HMVMX_EXIT_DECL hmR0VmxExitInvpcidNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17539	{
17540	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17541
17542	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INVLPG_EXIT))
17543	{
17544	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_INVPCID));
17545	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17546	hmR0VmxReadExitQualVmcs(pVmxTransient);
17547	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17548
17549	VMXVEXITINFO ExitInfo;
17550	RT_ZERO(ExitInfo);
17551	ExitInfo.uReason = pVmxTransient->uExitReason;
17552	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17553	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17554	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17555	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17556	}
17557	return hmR0VmxExitInvpcid(pVCpu, pVmxTransient);
17558	}
17559
17560
17561	/**
17562	* Nested-guest VM-exit handler for invalid-guest state
17563	* (VMX_EXIT_ERR_INVALID_GUEST_STATE). Error VM-exit.
17564	*/
17565	HMVMX_EXIT_DECL hmR0VmxExitErrInvalidGuestStateNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17566	{
17567	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17568
17569	/*
17570	* Currently this should never happen because we fully emulate VMLAUNCH/VMRESUME in IEM.
17571	* So if it does happen, it indicates a bug possibly in the hardware-assisted VMX code.
17572	* Handle it like it's in an invalid guest state of the outer guest.
17573	*
17574	* When the fast path is implemented, this should be changed to cause the corresponding
17575	* nested-guest VM-exit.
17576	*/
17577	return hmR0VmxExitErrInvalidGuestState(pVCpu, pVmxTransient);
17578	}
17579
17580
17581	/**
17582	* Nested-guest VM-exit handler for instructions that cause VM-exits uncondtionally
17583	* and only provide the instruction length.
17584	*
17585	* Unconditional VM-exit.
17586	*/
17587	HMVMX_EXIT_DECL hmR0VmxExitInstrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17588	{
17589	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17590
17591	#ifdef VBOX_STRICT
17592	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
17593	switch (pVmxTransient->uExitReason)
17594	{
17595	case VMX_EXIT_ENCLS:
17596	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_ENCLS_EXIT));
17597	break;
17598
17599	case VMX_EXIT_VMFUNC:
17600	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_VMFUNC));
17601	break;
17602	}
17603	#endif
17604
17605	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17606	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17607	}
17608
17609
17610	/**
17611	* Nested-guest VM-exit handler for instructions that provide instruction length as
17612	* well as more information.
17613	*
17614	* Unconditional VM-exit.
17615	*/
17616	HMVMX_EXIT_DECL hmR0VmxExitInstrWithInfoNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17617	{
17618	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17619
17620	#ifdef VBOX_STRICT
17621	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
17622	switch (pVmxTransient->uExitReason)
17623	{
17624	case VMX_EXIT_GDTR_IDTR_ACCESS:
17625	case VMX_EXIT_LDTR_TR_ACCESS:
17626	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_DESC_TABLE_EXIT));
17627	break;
17628
17629	case VMX_EXIT_RDRAND:
17630	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_RDRAND_EXIT));
17631	break;
17632
17633	case VMX_EXIT_RDSEED:
17634	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_RDSEED_EXIT));
17635	break;
17636
17637	case VMX_EXIT_XSAVES:
17638	case VMX_EXIT_XRSTORS:
17639	/** @todo NSTVMX: Verify XSS-bitmap. */
17640	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_XSAVES_XRSTORS));
17641	break;
17642
17643	case VMX_EXIT_UMWAIT:
17644	case VMX_EXIT_TPAUSE:
17645	Assert(CPUMIsGuestVmxProcCtlsSet(pCtx, VMX_PROC_CTLS_RDTSC_EXIT));
17646	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_USER_WAIT_PAUSE));
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@ 90842

Download in other formats: