HMVMXR0.cpp@ 85765

Last change on this file since 85765 was 85765, checked in by vboxsync, 4 years ago
VMM/HMVMXR0: Fix assertion with exception bitmaps by restoring debug state in the inner loop itself.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 721.7 KB

Line
1	/* $Id: HMVMXR0.cpp 85765 2020-08-14 11:35:19Z 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)->hm.s.idEnteredCpu == RTMpCpuId(), \
165	("Illegal migration! Entered on CPU %u Current %u\n", \
166	(a_pVCpu)->hm.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	static const uint32_t g_aVmcsSegBase[] =
694	{
695	VMX_VMCS_GUEST_ES_BASE,
696	VMX_VMCS_GUEST_CS_BASE,
697	VMX_VMCS_GUEST_SS_BASE,
698	VMX_VMCS_GUEST_DS_BASE,
699	VMX_VMCS_GUEST_FS_BASE,
700	VMX_VMCS_GUEST_GS_BASE
701	};
702	static const uint32_t g_aVmcsSegSel[] =
703	{
704	VMX_VMCS16_GUEST_ES_SEL,
705	VMX_VMCS16_GUEST_CS_SEL,
706	VMX_VMCS16_GUEST_SS_SEL,
707	VMX_VMCS16_GUEST_DS_SEL,
708	VMX_VMCS16_GUEST_FS_SEL,
709	VMX_VMCS16_GUEST_GS_SEL
710	};
711	static const uint32_t g_aVmcsSegLimit[] =
712	{
713	VMX_VMCS32_GUEST_ES_LIMIT,
714	VMX_VMCS32_GUEST_CS_LIMIT,
715	VMX_VMCS32_GUEST_SS_LIMIT,
716	VMX_VMCS32_GUEST_DS_LIMIT,
717	VMX_VMCS32_GUEST_FS_LIMIT,
718	VMX_VMCS32_GUEST_GS_LIMIT
719	};
720	static const uint32_t g_aVmcsSegAttr[] =
721	{
722	VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS,
723	VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS,
724	VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS,
725	VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS,
726	VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS,
727	VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS
728	};
729	AssertCompile(RT_ELEMENTS(g_aVmcsSegSel) == X86_SREG_COUNT);
730	AssertCompile(RT_ELEMENTS(g_aVmcsSegLimit) == X86_SREG_COUNT);
731	AssertCompile(RT_ELEMENTS(g_aVmcsSegBase) == X86_SREG_COUNT);
732	AssertCompile(RT_ELEMENTS(g_aVmcsSegAttr) == X86_SREG_COUNT);
733
734	#ifdef HMVMX_USE_FUNCTION_TABLE
735	/**
736	* VMX_EXIT dispatch table.
737	*/
738	static const struct CLANG11NOTHROWWEIRDNESS { PFNVMXEXITHANDLER pfn; } g_aVMExitHandlers[VMX_EXIT_MAX + 1] =
739	{
740	/* 0 VMX_EXIT_XCPT_OR_NMI */ { hmR0VmxExitXcptOrNmi },
741	/* 1 VMX_EXIT_EXT_INT */ { hmR0VmxExitExtInt },
742	/* 2 VMX_EXIT_TRIPLE_FAULT */ { hmR0VmxExitTripleFault },
743	/* 3 VMX_EXIT_INIT_SIGNAL */ { hmR0VmxExitErrUnexpected },
744	/* 4 VMX_EXIT_SIPI */ { hmR0VmxExitErrUnexpected },
745	/* 5 VMX_EXIT_IO_SMI */ { hmR0VmxExitErrUnexpected },
746	/* 6 VMX_EXIT_SMI */ { hmR0VmxExitErrUnexpected },
747	/* 7 VMX_EXIT_INT_WINDOW */ { hmR0VmxExitIntWindow },
748	/* 8 VMX_EXIT_NMI_WINDOW */ { hmR0VmxExitNmiWindow },
749	/* 9 VMX_EXIT_TASK_SWITCH */ { hmR0VmxExitTaskSwitch },
750	/* 10 VMX_EXIT_CPUID */ { hmR0VmxExitCpuid },
751	/* 11 VMX_EXIT_GETSEC */ { hmR0VmxExitGetsec },
752	/* 12 VMX_EXIT_HLT */ { hmR0VmxExitHlt },
753	/* 13 VMX_EXIT_INVD */ { hmR0VmxExitInvd },
754	/* 14 VMX_EXIT_INVLPG */ { hmR0VmxExitInvlpg },
755	/* 15 VMX_EXIT_RDPMC */ { hmR0VmxExitRdpmc },
756	/* 16 VMX_EXIT_RDTSC */ { hmR0VmxExitRdtsc },
757	/* 17 VMX_EXIT_RSM */ { hmR0VmxExitErrUnexpected },
758	/* 18 VMX_EXIT_VMCALL */ { hmR0VmxExitVmcall },
759	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
760	/* 19 VMX_EXIT_VMCLEAR */ { hmR0VmxExitVmclear },
761	/* 20 VMX_EXIT_VMLAUNCH */ { hmR0VmxExitVmlaunch },
762	/* 21 VMX_EXIT_VMPTRLD */ { hmR0VmxExitVmptrld },
763	/* 22 VMX_EXIT_VMPTRST */ { hmR0VmxExitVmptrst },
764	/* 23 VMX_EXIT_VMREAD */ { hmR0VmxExitVmread },
765	/* 24 VMX_EXIT_VMRESUME */ { hmR0VmxExitVmresume },
766	/* 25 VMX_EXIT_VMWRITE */ { hmR0VmxExitVmwrite },
767	/* 26 VMX_EXIT_VMXOFF */ { hmR0VmxExitVmxoff },
768	/* 27 VMX_EXIT_VMXON */ { hmR0VmxExitVmxon },
769	#else
770	/* 19 VMX_EXIT_VMCLEAR */ { hmR0VmxExitSetPendingXcptUD },
771	/* 20 VMX_EXIT_VMLAUNCH */ { hmR0VmxExitSetPendingXcptUD },
772	/* 21 VMX_EXIT_VMPTRLD */ { hmR0VmxExitSetPendingXcptUD },
773	/* 22 VMX_EXIT_VMPTRST */ { hmR0VmxExitSetPendingXcptUD },
774	/* 23 VMX_EXIT_VMREAD */ { hmR0VmxExitSetPendingXcptUD },
775	/* 24 VMX_EXIT_VMRESUME */ { hmR0VmxExitSetPendingXcptUD },
776	/* 25 VMX_EXIT_VMWRITE */ { hmR0VmxExitSetPendingXcptUD },
777	/* 26 VMX_EXIT_VMXOFF */ { hmR0VmxExitSetPendingXcptUD },
778	/* 27 VMX_EXIT_VMXON */ { hmR0VmxExitSetPendingXcptUD },
779	#endif
780	/* 28 VMX_EXIT_MOV_CRX */ { hmR0VmxExitMovCRx },
781	/* 29 VMX_EXIT_MOV_DRX */ { hmR0VmxExitMovDRx },
782	/* 30 VMX_EXIT_IO_INSTR */ { hmR0VmxExitIoInstr },
783	/* 31 VMX_EXIT_RDMSR */ { hmR0VmxExitRdmsr },
784	/* 32 VMX_EXIT_WRMSR */ { hmR0VmxExitWrmsr },
785	/* 33 VMX_EXIT_ERR_INVALID_GUEST_STATE */ { hmR0VmxExitErrInvalidGuestState },
786	/* 34 VMX_EXIT_ERR_MSR_LOAD */ { hmR0VmxExitErrUnexpected },
787	/* 35 UNDEFINED */ { hmR0VmxExitErrUnexpected },
788	/* 36 VMX_EXIT_MWAIT */ { hmR0VmxExitMwait },
789	/* 37 VMX_EXIT_MTF */ { hmR0VmxExitMtf },
790	/* 38 UNDEFINED */ { hmR0VmxExitErrUnexpected },
791	/* 39 VMX_EXIT_MONITOR */ { hmR0VmxExitMonitor },
792	/* 40 VMX_EXIT_PAUSE */ { hmR0VmxExitPause },
793	/* 41 VMX_EXIT_ERR_MACHINE_CHECK */ { hmR0VmxExitErrUnexpected },
794	/* 42 UNDEFINED */ { hmR0VmxExitErrUnexpected },
795	/* 43 VMX_EXIT_TPR_BELOW_THRESHOLD */ { hmR0VmxExitTprBelowThreshold },
796	/* 44 VMX_EXIT_APIC_ACCESS */ { hmR0VmxExitApicAccess },
797	/* 45 VMX_EXIT_VIRTUALIZED_EOI */ { hmR0VmxExitErrUnexpected },
798	/* 46 VMX_EXIT_GDTR_IDTR_ACCESS */ { hmR0VmxExitErrUnexpected },
799	/* 47 VMX_EXIT_LDTR_TR_ACCESS */ { hmR0VmxExitErrUnexpected },
800	/* 48 VMX_EXIT_EPT_VIOLATION */ { hmR0VmxExitEptViolation },
801	/* 49 VMX_EXIT_EPT_MISCONFIG */ { hmR0VmxExitEptMisconfig },
802	/* 50 VMX_EXIT_INVEPT */ { hmR0VmxExitSetPendingXcptUD },
803	/* 51 VMX_EXIT_RDTSCP */ { hmR0VmxExitRdtscp },
804	/* 52 VMX_EXIT_PREEMPT_TIMER */ { hmR0VmxExitPreemptTimer },
805	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
806	/* 53 VMX_EXIT_INVVPID */ { hmR0VmxExitInvvpid },
807	#else
808	/* 53 VMX_EXIT_INVVPID */ { hmR0VmxExitSetPendingXcptUD },
809	#endif
810	/* 54 VMX_EXIT_WBINVD */ { hmR0VmxExitWbinvd },
811	/* 55 VMX_EXIT_XSETBV */ { hmR0VmxExitXsetbv },
812	/* 56 VMX_EXIT_APIC_WRITE */ { hmR0VmxExitErrUnexpected },
813	/* 57 VMX_EXIT_RDRAND */ { hmR0VmxExitErrUnexpected },
814	/* 58 VMX_EXIT_INVPCID */ { hmR0VmxExitInvpcid },
815	/* 59 VMX_EXIT_VMFUNC */ { hmR0VmxExitErrUnexpected },
816	/* 60 VMX_EXIT_ENCLS */ { hmR0VmxExitErrUnexpected },
817	/* 61 VMX_EXIT_RDSEED */ { hmR0VmxExitErrUnexpected },
818	/* 62 VMX_EXIT_PML_FULL */ { hmR0VmxExitErrUnexpected },
819	/* 63 VMX_EXIT_XSAVES */ { hmR0VmxExitErrUnexpected },
820	/* 64 VMX_EXIT_XRSTORS */ { hmR0VmxExitErrUnexpected },
821	/* 65 UNDEFINED */ { hmR0VmxExitErrUnexpected },
822	/* 66 VMX_EXIT_SPP_EVENT */ { hmR0VmxExitErrUnexpected },
823	/* 67 VMX_EXIT_UMWAIT */ { hmR0VmxExitErrUnexpected },
824	/* 68 VMX_EXIT_TPAUSE */ { hmR0VmxExitErrUnexpected },
825	};
826	#endif /* HMVMX_USE_FUNCTION_TABLE */
827
828	#if defined(VBOX_STRICT) && defined(LOG_ENABLED)
829	static const char * const g_apszVmxInstrErrors[HMVMX_INSTR_ERROR_MAX + 1] =
830	{
831	/* 0 */ "(Not Used)",
832	/* 1 */ "VMCALL executed in VMX root operation.",
833	/* 2 */ "VMCLEAR with invalid physical address.",
834	/* 3 */ "VMCLEAR with VMXON pointer.",
835	/* 4 */ "VMLAUNCH with non-clear VMCS.",
836	/* 5 */ "VMRESUME with non-launched VMCS.",
837	/* 6 */ "VMRESUME after VMXOFF",
838	/* 7 */ "VM-entry with invalid control fields.",
839	/* 8 */ "VM-entry with invalid host state fields.",
840	/* 9 */ "VMPTRLD with invalid physical address.",
841	/* 10 */ "VMPTRLD with VMXON pointer.",
842	/* 11 */ "VMPTRLD with incorrect revision identifier.",
843	/* 12 */ "VMREAD/VMWRITE from/to unsupported VMCS component.",
844	/* 13 */ "VMWRITE to read-only VMCS component.",
845	/* 14 */ "(Not Used)",
846	/* 15 */ "VMXON executed in VMX root operation.",
847	/* 16 */ "VM-entry with invalid executive-VMCS pointer.",
848	/* 17 */ "VM-entry with non-launched executing VMCS.",
849	/* 18 */ "VM-entry with executive-VMCS pointer not VMXON pointer.",
850	/* 19 */ "VMCALL with non-clear VMCS.",
851	/* 20 */ "VMCALL with invalid VM-exit control fields.",
852	/* 21 */ "(Not Used)",
853	/* 22 */ "VMCALL with incorrect MSEG revision identifier.",
854	/* 23 */ "VMXOFF under dual monitor treatment of SMIs and SMM.",
855	/* 24 */ "VMCALL with invalid SMM-monitor features.",
856	/* 25 */ "VM-entry with invalid VM-execution control fields in executive VMCS.",
857	/* 26 */ "VM-entry with events blocked by MOV SS.",
858	/* 27 */ "(Not Used)",
859	/* 28 */ "Invalid operand to INVEPT/INVVPID."
860	};
861	#endif /* VBOX_STRICT && LOG_ENABLED */
862
863
864	/**
865	* Checks if the given MSR is part of the lastbranch-from-IP MSR stack.
866	* @returns @c true if it's part of LBR stack, @c false otherwise.
867	*
868	* @param pVM The cross context VM structure.
869	* @param idMsr The MSR.
870	* @param pidxMsr Where to store the index of the MSR in the LBR MSR array.
871	* Optional, can be NULL.
872	*
873	* @remarks Must only be called when LBR is enabled.
874	*/
875	DECL_FORCE_INLINE(bool) hmR0VmxIsLbrBranchFromMsr(PCVM pVM, uint32_t idMsr, uint32_t *pidxMsr)
876	{
877	Assert(pVM->hm.s.vmx.fLbr);
878	Assert(pVM->hm.s.vmx.idLbrFromIpMsrFirst);
879	uint32_t const cLbrStack = pVM->hm.s.vmx.idLbrFromIpMsrLast - pVM->hm.s.vmx.idLbrFromIpMsrFirst + 1;
880	uint32_t const idxMsr = idMsr - pVM->hm.s.vmx.idLbrFromIpMsrFirst;
881	if (idxMsr < cLbrStack)
882	{
883	if (pidxMsr)
884	*pidxMsr = idxMsr;
885	return true;
886	}
887	return false;
888	}
889
890
891	/**
892	* Checks if the given MSR is part of the lastbranch-to-IP MSR stack.
893	* @returns @c true if it's part of LBR stack, @c false otherwise.
894	*
895	* @param pVM The cross context VM structure.
896	* @param idMsr The MSR.
897	* @param pidxMsr Where to store the index of the MSR in the LBR MSR array.
898	* Optional, can be NULL.
899	*
900	* @remarks Must only be called when LBR is enabled and when lastbranch-to-IP MSRs
901	* are supported by the CPU (see hmR0VmxSetupLbrMsrRange).
902	*/
903	DECL_FORCE_INLINE(bool) hmR0VmxIsLbrBranchToMsr(PCVM pVM, uint32_t idMsr, uint32_t *pidxMsr)
904	{
905	Assert(pVM->hm.s.vmx.fLbr);
906	if (pVM->hm.s.vmx.idLbrToIpMsrFirst)
907	{
908	uint32_t const cLbrStack = pVM->hm.s.vmx.idLbrToIpMsrLast - pVM->hm.s.vmx.idLbrToIpMsrFirst + 1;
909	uint32_t const idxMsr = idMsr - pVM->hm.s.vmx.idLbrToIpMsrFirst;
910	if (idxMsr < cLbrStack)
911	{
912	if (pidxMsr)
913	*pidxMsr = idxMsr;
914	return true;
915	}
916	}
917	return false;
918	}
919
920
921	/**
922	* Gets the CR0 guest/host mask.
923	*
924	* These bits typically does not change through the lifetime of a VM. Any bit set in
925	* this mask is owned by the host/hypervisor and would cause a VM-exit when modified
926	* by the guest.
927	*
928	* @returns The CR0 guest/host mask.
929	* @param pVCpu The cross context virtual CPU structure.
930	*/
931	static uint64_t hmR0VmxGetFixedCr0Mask(PCVMCPUCC pVCpu)
932	{
933	/*
934	* Modifications to CR0 bits that VT-x ignores saving/restoring (CD, ET, NW) and
935	* to CR0 bits that we require for shadow paging (PG) by the guest must cause VM-exits.
936	*
937	* Furthermore, modifications to any bits that are reserved/unspecified currently
938	* by the Intel spec. must also cause a VM-exit. This prevents unpredictable behavior
939	* when future CPUs specify and use currently reserved/unspecified bits.
940	*/
941	/** @todo Avoid intercepting CR0.PE with unrestricted guest execution. Fix PGM
942	* enmGuestMode to be in-sync with the current mode. See @bugref{6398}
943	* and @bugref{6944}. */
944	PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
945	return ( X86_CR0_PE
946	\| X86_CR0_NE
947	\| (pVM->hm.s.fNestedPaging ? 0 : X86_CR0_WP)
948	\| X86_CR0_PG
949	\| VMX_EXIT_HOST_CR0_IGNORE_MASK);
950	}
951
952
953	/**
954	* Gets the CR4 guest/host mask.
955	*
956	* These bits typically does not change through the lifetime of a VM. Any bit set in
957	* this mask is owned by the host/hypervisor and would cause a VM-exit when modified
958	* by the guest.
959	*
960	* @returns The CR4 guest/host mask.
961	* @param pVCpu The cross context virtual CPU structure.
962	*/
963	static uint64_t hmR0VmxGetFixedCr4Mask(PCVMCPUCC pVCpu)
964	{
965	/*
966	* We construct a mask of all CR4 bits that the guest can modify without causing
967	* a VM-exit. Then invert this mask to obtain all CR4 bits that should cause
968	* a VM-exit when the guest attempts to modify them when executing using
969	* hardware-assisted VMX.
970	*
971	* When a feature is not exposed to the guest (and may be present on the host),
972	* we want to intercept guest modifications to the bit so we can emulate proper
973	* behavior (e.g., #GP).
974	*
975	* Furthermore, only modifications to those bits that don't require immediate
976	* emulation is allowed. For e.g., PCIDE is excluded because the behavior
977	* depends on CR3 which might not always be the guest value while executing
978	* using hardware-assisted VMX.
979	*/
980	PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
981	bool const fFsGsBase = pVM->cpum.ro.GuestFeatures.fFsGsBase;
982	bool const fXSaveRstor = pVM->cpum.ro.GuestFeatures.fXSaveRstor;
983	bool const fFxSaveRstor = pVM->cpum.ro.GuestFeatures.fFxSaveRstor;
984
985	/*
986	* Paranoia.
987	* Ensure features exposed to the guest are present on the host.
988	*/
989	Assert(!fFsGsBase \|\| pVM->cpum.ro.HostFeatures.fFsGsBase);
990	Assert(!fXSaveRstor \|\| pVM->cpum.ro.HostFeatures.fXSaveRstor);
991	Assert(!fFxSaveRstor \|\| pVM->cpum.ro.HostFeatures.fFxSaveRstor);
992
993	uint64_t const fGstMask = ( X86_CR4_PVI
994	\| X86_CR4_TSD
995	\| X86_CR4_DE
996	\| X86_CR4_MCE
997	\| X86_CR4_PCE
998	\| X86_CR4_OSXMMEEXCPT
999	\| (fFsGsBase ? X86_CR4_FSGSBASE : 0)
1000	\| (fXSaveRstor ? X86_CR4_OSXSAVE : 0)
1001	\| (fFxSaveRstor ? X86_CR4_OSFXSR : 0));
1002	return ~fGstMask;
1003	}
1004
1005
1006	/**
1007	* Returns whether the the VM-exit MSR-store area differs from the VM-exit MSR-load
1008	* area.
1009	*
1010	* @returns @c true if it's different, @c false otherwise.
1011	* @param pVmcsInfo The VMCS info. object.
1012	*/
1013	DECL_FORCE_INLINE(bool) hmR0VmxIsSeparateExitMsrStoreAreaVmcs(PCVMXVMCSINFO pVmcsInfo)
1014	{
1015	return RT_BOOL( pVmcsInfo->pvGuestMsrStore != pVmcsInfo->pvGuestMsrLoad
1016	&& pVmcsInfo->pvGuestMsrStore);
1017	}
1018
1019
1020	/**
1021	* Sets the given Processor-based VM-execution controls.
1022	*
1023	* @param pVmxTransient The VMX-transient structure.
1024	* @param uProcCtls The Processor-based VM-execution controls to set.
1025	*/
1026	static void hmR0VmxSetProcCtlsVmcs(PVMXTRANSIENT pVmxTransient, uint32_t uProcCtls)
1027	{
1028	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1029	if ((pVmcsInfo->u32ProcCtls & uProcCtls) != uProcCtls)
1030	{
1031	pVmcsInfo->u32ProcCtls \|= uProcCtls;
1032	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
1033	AssertRC(rc);
1034	}
1035	}
1036
1037
1038	/**
1039	* Removes the given Processor-based VM-execution controls.
1040	*
1041	* @param pVCpu The cross context virtual CPU structure.
1042	* @param pVmxTransient The VMX-transient structure.
1043	* @param uProcCtls The Processor-based VM-execution controls to remove.
1044	*
1045	* @remarks When executing a nested-guest, this will not remove any of the specified
1046	* controls if the nested hypervisor has set any one of them.
1047	*/
1048	static void hmR0VmxRemoveProcCtlsVmcs(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uProcCtls)
1049	{
1050	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1051	if (pVmcsInfo->u32ProcCtls & uProcCtls)
1052	{
1053	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1054	bool const fRemoveCtls = !pVmxTransient->fIsNestedGuest
1055	? true
1056	: !CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, uProcCtls);
1057	#else
1058	NOREF(pVCpu);
1059	bool const fRemoveCtls = true;
1060	#endif
1061	if (fRemoveCtls)
1062	{
1063	pVmcsInfo->u32ProcCtls &= ~uProcCtls;
1064	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
1065	AssertRC(rc);
1066	}
1067	}
1068	}
1069
1070
1071	/**
1072	* Sets the TSC offset for the current VMCS.
1073	*
1074	* @param uTscOffset The TSC offset to set.
1075	* @param pVmcsInfo The VMCS info. object.
1076	*/
1077	static void hmR0VmxSetTscOffsetVmcs(PVMXVMCSINFO pVmcsInfo, uint64_t uTscOffset)
1078	{
1079	if (pVmcsInfo->u64TscOffset != uTscOffset)
1080	{
1081	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, uTscOffset);
1082	AssertRC(rc);
1083	pVmcsInfo->u64TscOffset = uTscOffset;
1084	}
1085	}
1086
1087
1088	/**
1089	* Adds one or more exceptions to the exception bitmap and commits it to the current
1090	* VMCS.
1091	*
1092	* @param pVmxTransient The VMX-transient structure.
1093	* @param uXcptMask The exception(s) to add.
1094	*/
1095	static void hmR0VmxAddXcptInterceptMask(PCVMXTRANSIENT pVmxTransient, uint32_t uXcptMask)
1096	{
1097	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1098	uint32_t uXcptBitmap = pVmcsInfo->u32XcptBitmap;
1099	if ((uXcptBitmap & uXcptMask) != uXcptMask)
1100	{
1101	uXcptBitmap \|= uXcptMask;
1102	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
1103	AssertRC(rc);
1104	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
1105	}
1106	}
1107
1108
1109	/**
1110	* Adds an exception to the exception bitmap and commits it to the current VMCS.
1111	*
1112	* @param pVmxTransient The VMX-transient structure.
1113	* @param uXcpt The exception to add.
1114	*/
1115	static void hmR0VmxAddXcptIntercept(PCVMXTRANSIENT pVmxTransient, uint8_t uXcpt)
1116	{
1117	Assert(uXcpt <= X86_XCPT_LAST);
1118	hmR0VmxAddXcptInterceptMask(pVmxTransient, RT_BIT_32(uXcpt));
1119	}
1120
1121
1122	/**
1123	* Remove one or more exceptions from the exception bitmap and commits it to the
1124	* current VMCS.
1125	*
1126	* This takes care of not removing the exception intercept if a nested-guest
1127	* requires the exception to be intercepted.
1128	*
1129	* @returns VBox status code.
1130	* @param pVCpu The cross context virtual CPU structure.
1131	* @param pVmxTransient The VMX-transient structure.
1132	* @param uXcptMask The exception(s) to remove.
1133	*/
1134	static int hmR0VmxRemoveXcptInterceptMask(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t uXcptMask)
1135	{
1136	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1137	uint32_t u32XcptBitmap = pVmcsInfo->u32XcptBitmap;
1138	if (u32XcptBitmap & uXcptMask)
1139	{
1140	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1141	if (!pVmxTransient->fIsNestedGuest)
1142	{ /* likely */ }
1143	else
1144	{
1145	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1146	uXcptMask &= ~pVmcsNstGst->u32XcptBitmap;
1147	}
1148	#endif
1149	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
1150	uXcptMask &= ~( RT_BIT(X86_XCPT_BP)
1151	\| RT_BIT(X86_XCPT_DE)
1152	\| RT_BIT(X86_XCPT_NM)
1153	\| RT_BIT(X86_XCPT_TS)
1154	\| RT_BIT(X86_XCPT_UD)
1155	\| RT_BIT(X86_XCPT_NP)
1156	\| RT_BIT(X86_XCPT_SS)
1157	\| RT_BIT(X86_XCPT_GP)
1158	\| RT_BIT(X86_XCPT_PF)
1159	\| RT_BIT(X86_XCPT_MF));
1160	#elif defined(HMVMX_ALWAYS_TRAP_PF)
1161	uXcptMask &= ~RT_BIT(X86_XCPT_PF);
1162	#endif
1163	if (uXcptMask)
1164	{
1165	/* Validate we are not removing any essential exception intercepts. */
1166	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging \|\| !(uXcptMask & RT_BIT(X86_XCPT_PF)));
1167	NOREF(pVCpu);
1168	Assert(!(uXcptMask & RT_BIT(X86_XCPT_DB)));
1169	Assert(!(uXcptMask & RT_BIT(X86_XCPT_AC)));
1170
1171	/* Remove it from the exception bitmap. */
1172	u32XcptBitmap &= ~uXcptMask;
1173
1174	/* Commit and update the cache if necessary. */
1175	if (pVmcsInfo->u32XcptBitmap != u32XcptBitmap)
1176	{
1177	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
1178	AssertRC(rc);
1179	pVmcsInfo->u32XcptBitmap = u32XcptBitmap;
1180	}
1181	}
1182	}
1183	return VINF_SUCCESS;
1184	}
1185
1186
1187	/**
1188	* Remove an exceptions from the exception bitmap and commits it to the current
1189	* VMCS.
1190	*
1191	* @returns VBox status code.
1192	* @param pVCpu The cross context virtual CPU structure.
1193	* @param pVmxTransient The VMX-transient structure.
1194	* @param uXcpt The exception to remove.
1195	*/
1196	static int hmR0VmxRemoveXcptIntercept(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint8_t uXcpt)
1197	{
1198	return hmR0VmxRemoveXcptInterceptMask(pVCpu, pVmxTransient, RT_BIT(uXcpt));
1199	}
1200
1201
1202	/**
1203	* Loads the VMCS specified by the VMCS info. object.
1204	*
1205	* @returns VBox status code.
1206	* @param pVmcsInfo The VMCS info. object.
1207	*
1208	* @remarks Can be called with interrupts disabled.
1209	*/
1210	static int hmR0VmxLoadVmcs(PVMXVMCSINFO pVmcsInfo)
1211	{
1212	Assert(pVmcsInfo->HCPhysVmcs != 0 && pVmcsInfo->HCPhysVmcs != NIL_RTHCPHYS);
1213	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1214
1215	int rc = VMXLoadVmcs(pVmcsInfo->HCPhysVmcs);
1216	if (RT_SUCCESS(rc))
1217	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_CURRENT;
1218	return rc;
1219	}
1220
1221
1222	/**
1223	* Clears the VMCS specified by the VMCS info. object.
1224	*
1225	* @returns VBox status code.
1226	* @param pVmcsInfo The VMCS info. object.
1227	*
1228	* @remarks Can be called with interrupts disabled.
1229	*/
1230	static int hmR0VmxClearVmcs(PVMXVMCSINFO pVmcsInfo)
1231	{
1232	Assert(pVmcsInfo->HCPhysVmcs != 0 && pVmcsInfo->HCPhysVmcs != NIL_RTHCPHYS);
1233	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1234
1235	int rc = VMXClearVmcs(pVmcsInfo->HCPhysVmcs);
1236	if (RT_SUCCESS(rc))
1237	pVmcsInfo->fVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
1238	return rc;
1239	}
1240
1241
1242	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1243	/**
1244	* Loads the shadow VMCS specified by the VMCS info. object.
1245	*
1246	* @returns VBox status code.
1247	* @param pVmcsInfo The VMCS info. object.
1248	*
1249	* @remarks Can be called with interrupts disabled.
1250	*/
1251	static int hmR0VmxLoadShadowVmcs(PVMXVMCSINFO pVmcsInfo)
1252	{
1253	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1254	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
1255
1256	int rc = VMXLoadVmcs(pVmcsInfo->HCPhysShadowVmcs);
1257	if (RT_SUCCESS(rc))
1258	pVmcsInfo->fShadowVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_CURRENT;
1259	return rc;
1260	}
1261
1262
1263	/**
1264	* Clears the shadow VMCS specified by the VMCS info. object.
1265	*
1266	* @returns VBox status code.
1267	* @param pVmcsInfo The VMCS info. object.
1268	*
1269	* @remarks Can be called with interrupts disabled.
1270	*/
1271	static int hmR0VmxClearShadowVmcs(PVMXVMCSINFO pVmcsInfo)
1272	{
1273	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1274	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
1275
1276	int rc = VMXClearVmcs(pVmcsInfo->HCPhysShadowVmcs);
1277	if (RT_SUCCESS(rc))
1278	pVmcsInfo->fShadowVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
1279	return rc;
1280	}
1281
1282
1283	/**
1284	* Switches from and to the specified VMCSes.
1285	*
1286	* @returns VBox status code.
1287	* @param pVmcsInfoFrom The VMCS info. object we are switching from.
1288	* @param pVmcsInfoTo The VMCS info. object we are switching to.
1289	*
1290	* @remarks Called with interrupts disabled.
1291	*/
1292	static int hmR0VmxSwitchVmcs(PVMXVMCSINFO pVmcsInfoFrom, PVMXVMCSINFO pVmcsInfoTo)
1293	{
1294	/*
1295	* Clear the VMCS we are switching out if it has not already been cleared.
1296	* This will sync any CPU internal data back to the VMCS.
1297	*/
1298	if (pVmcsInfoFrom->fVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
1299	{
1300	int rc = hmR0VmxClearVmcs(pVmcsInfoFrom);
1301	if (RT_SUCCESS(rc))
1302	{
1303	/*
1304	* The shadow VMCS, if any, would not be active at this point since we
1305	* would have cleared it while importing the virtual hardware-virtualization
1306	* state as part the VMLAUNCH/VMRESUME VM-exit. Hence, there's no need to
1307	* clear the shadow VMCS here, just assert for safety.
1308	*/
1309	Assert(!pVmcsInfoFrom->pvShadowVmcs \|\| pVmcsInfoFrom->fShadowVmcsState == VMX_V_VMCS_LAUNCH_STATE_CLEAR);
1310	}
1311	else
1312	return rc;
1313	}
1314
1315	/*
1316	* Clear the VMCS we are switching to if it has not already been cleared.
1317	* This will initialize the VMCS launch state to "clear" required for loading it.
1318	*
1319	* See Intel spec. 31.6 "Preparation And Launching A Virtual Machine".
1320	*/
1321	if (pVmcsInfoTo->fVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
1322	{
1323	int rc = hmR0VmxClearVmcs(pVmcsInfoTo);
1324	if (RT_SUCCESS(rc))
1325	{ /* likely */ }
1326	else
1327	return rc;
1328	}
1329
1330	/*
1331	* Finally, load the VMCS we are switching to.
1332	*/
1333	return hmR0VmxLoadVmcs(pVmcsInfoTo);
1334	}
1335
1336
1337	/**
1338	* Switches between the guest VMCS and the nested-guest VMCS as specified by the
1339	* caller.
1340	*
1341	* @returns VBox status code.
1342	* @param pVCpu The cross context virtual CPU structure.
1343	* @param fSwitchToNstGstVmcs Whether to switch to the nested-guest VMCS (pass
1344	* true) or guest VMCS (pass false).
1345	*/
1346	static int hmR0VmxSwitchToGstOrNstGstVmcs(PVMCPUCC pVCpu, bool fSwitchToNstGstVmcs)
1347	{
1348	/* Ensure we have synced everything from the guest-CPU context to the VMCS before switching. */
1349	HMVMX_CPUMCTX_ASSERT(pVCpu, HMVMX_CPUMCTX_EXTRN_ALL);
1350
1351	PVMXVMCSINFO pVmcsInfoFrom;
1352	PVMXVMCSINFO pVmcsInfoTo;
1353	if (fSwitchToNstGstVmcs)
1354	{
1355	pVmcsInfoFrom = &pVCpu->hm.s.vmx.VmcsInfo;
1356	pVmcsInfoTo = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
1357	}
1358	else
1359	{
1360	pVmcsInfoFrom = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
1361	pVmcsInfoTo = &pVCpu->hm.s.vmx.VmcsInfo;
1362	}
1363
1364	/*
1365	* Disable interrupts to prevent being preempted while we switch the current VMCS as the
1366	* preemption hook code path acquires the current VMCS.
1367	*/
1368	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1369
1370	int rc = hmR0VmxSwitchVmcs(pVmcsInfoFrom, pVmcsInfoTo);
1371	if (RT_SUCCESS(rc))
1372	{
1373	pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs = fSwitchToNstGstVmcs;
1374
1375	/*
1376	* If we are switching to a VMCS that was executed on a different host CPU or was
1377	* never executed before, flag that we need to export the host state before executing
1378	* guest/nested-guest code using hardware-assisted VMX.
1379	*
1380	* This could probably be done in a preemptible context since the preemption hook
1381	* will flag the necessary change in host context. However, since preemption is
1382	* already disabled and to avoid making assumptions about host specific code in
1383	* RTMpCpuId when called with preemption enabled, we'll do this while preemption is
1384	* disabled.
1385	*/
1386	if (pVmcsInfoTo->idHostCpuState == RTMpCpuId())
1387	{ /* likely */ }
1388	else
1389	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE);
1390
1391	ASMSetFlags(fEFlags);
1392
1393	/*
1394	* We use a different VM-exit MSR-store areas for the guest and nested-guest. Hence,
1395	* flag that we need to update the host MSR values there. Even if we decide in the
1396	* future to share the VM-exit MSR-store area page between the guest and nested-guest,
1397	* if its content differs, we would have to update the host MSRs anyway.
1398	*/
1399	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = false;
1400	}
1401	else
1402	ASMSetFlags(fEFlags);
1403	return rc;
1404	}
1405	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
1406
1407
1408	/**
1409	* Updates the VM's last error record.
1410	*
1411	* If there was a VMX instruction error, reads the error data from the VMCS and
1412	* updates VCPU's last error record as well.
1413	*
1414	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
1415	* Can be NULL if @a rc is not VERR_VMX_UNABLE_TO_START_VM or
1416	* VERR_VMX_INVALID_VMCS_FIELD.
1417	* @param rc The error code.
1418	*/
1419	static void hmR0VmxUpdateErrorRecord(PVMCPUCC pVCpu, int rc)
1420	{
1421	if ( rc == VERR_VMX_INVALID_VMCS_FIELD
1422	\|\| rc == VERR_VMX_UNABLE_TO_START_VM)
1423	{
1424	AssertPtrReturnVoid(pVCpu);
1425	VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
1426	}
1427	pVCpu->CTX_SUFF(pVM)->hm.s.rcInit = rc;
1428	}
1429
1430
1431	#ifdef VBOX_STRICT
1432	/**
1433	* Reads the VM-entry interruption-information field from the VMCS into the VMX
1434	* transient structure.
1435	*
1436	* @param pVmxTransient The VMX-transient structure.
1437	*/
1438	DECLINLINE(void) hmR0VmxReadEntryIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
1439	{
1440	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &pVmxTransient->uEntryIntInfo);
1441	AssertRC(rc);
1442	}
1443
1444
1445	/**
1446	* Reads the VM-entry exception error code field from the VMCS into
1447	* the VMX transient structure.
1448	*
1449	* @param pVmxTransient The VMX-transient structure.
1450	*/
1451	DECLINLINE(void) hmR0VmxReadEntryXcptErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1452	{
1453	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &pVmxTransient->uEntryXcptErrorCode);
1454	AssertRC(rc);
1455	}
1456
1457
1458	/**
1459	* Reads the VM-entry exception error code field from the VMCS into
1460	* the VMX transient structure.
1461	*
1462	* @param pVmxTransient The VMX-transient structure.
1463	*/
1464	DECLINLINE(void) hmR0VmxReadEntryInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
1465	{
1466	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &pVmxTransient->cbEntryInstr);
1467	AssertRC(rc);
1468	}
1469	#endif /* VBOX_STRICT */
1470
1471
1472	/**
1473	* Reads the VM-exit interruption-information field from the VMCS into the VMX
1474	* transient structure.
1475	*
1476	* @param pVmxTransient The VMX-transient structure.
1477	*/
1478	DECLINLINE(void) hmR0VmxReadExitIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
1479	{
1480	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INTERRUPTION_INFO))
1481	{
1482	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
1483	AssertRC(rc);
1484	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INTERRUPTION_INFO;
1485	}
1486	}
1487
1488
1489	/**
1490	* Reads the VM-exit interruption error code from the VMCS into the VMX
1491	* transient structure.
1492	*
1493	* @param pVmxTransient The VMX-transient structure.
1494	*/
1495	DECLINLINE(void) hmR0VmxReadExitIntErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1496	{
1497	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE))
1498	{
1499	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
1500	AssertRC(rc);
1501	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE;
1502	}
1503	}
1504
1505
1506	/**
1507	* Reads the VM-exit instruction length field from the VMCS into the VMX
1508	* transient structure.
1509	*
1510	* @param pVmxTransient The VMX-transient structure.
1511	*/
1512	DECLINLINE(void) hmR0VmxReadExitInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
1513	{
1514	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INSTR_LEN))
1515	{
1516	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbExitInstr);
1517	AssertRC(rc);
1518	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INSTR_LEN;
1519	}
1520	}
1521
1522
1523	/**
1524	* Reads the VM-exit instruction-information field from the VMCS into
1525	* the VMX transient structure.
1526	*
1527	* @param pVmxTransient The VMX-transient structure.
1528	*/
1529	DECLINLINE(void) hmR0VmxReadExitInstrInfoVmcs(PVMXTRANSIENT pVmxTransient)
1530	{
1531	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INSTR_INFO))
1532	{
1533	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
1534	AssertRC(rc);
1535	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INSTR_INFO;
1536	}
1537	}
1538
1539
1540	/**
1541	* Reads the Exit Qualification from the VMCS into the VMX transient structure.
1542	*
1543	* @param pVmxTransient The VMX-transient structure.
1544	*/
1545	DECLINLINE(void) hmR0VmxReadExitQualVmcs(PVMXTRANSIENT pVmxTransient)
1546	{
1547	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_QUALIFICATION))
1548	{
1549	int rc = VMXReadVmcsNw(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQual);
1550	AssertRC(rc);
1551	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_QUALIFICATION;
1552	}
1553	}
1554
1555
1556	/**
1557	* Reads the Guest-linear address from the VMCS into the VMX transient structure.
1558	*
1559	* @param pVmxTransient The VMX-transient structure.
1560	*/
1561	DECLINLINE(void) hmR0VmxReadGuestLinearAddrVmcs(PVMXTRANSIENT pVmxTransient)
1562	{
1563	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_LINEAR_ADDR))
1564	{
1565	int rc = VMXReadVmcsNw(VMX_VMCS_RO_GUEST_LINEAR_ADDR, &pVmxTransient->uGuestLinearAddr);
1566	AssertRC(rc);
1567	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_LINEAR_ADDR;
1568	}
1569	}
1570
1571
1572	/**
1573	* Reads the Guest-physical address from the VMCS into the VMX transient structure.
1574	*
1575	* @param pVmxTransient The VMX-transient structure.
1576	*/
1577	DECLINLINE(void) hmR0VmxReadGuestPhysicalAddrVmcs(PVMXTRANSIENT pVmxTransient)
1578	{
1579	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_PHYSICAL_ADDR))
1580	{
1581	int rc = VMXReadVmcs64(VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL, &pVmxTransient->uGuestPhysicalAddr);
1582	AssertRC(rc);
1583	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_PHYSICAL_ADDR;
1584	}
1585	}
1586
1587	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1588	/**
1589	* Reads the Guest pending-debug exceptions from the VMCS into the VMX transient
1590	* structure.
1591	*
1592	* @param pVmxTransient The VMX-transient structure.
1593	*/
1594	DECLINLINE(void) hmR0VmxReadGuestPendingDbgXctps(PVMXTRANSIENT pVmxTransient)
1595	{
1596	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_PENDING_DBG_XCPTS))
1597	{
1598	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, &pVmxTransient->uGuestPendingDbgXcpts);
1599	AssertRC(rc);
1600	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_PENDING_DBG_XCPTS;
1601	}
1602	}
1603	#endif
1604
1605	/**
1606	* Reads the IDT-vectoring information field from the VMCS into the VMX
1607	* transient structure.
1608	*
1609	* @param pVmxTransient The VMX-transient structure.
1610	*
1611	* @remarks No-long-jump zone!!!
1612	*/
1613	DECLINLINE(void) hmR0VmxReadIdtVectoringInfoVmcs(PVMXTRANSIENT pVmxTransient)
1614	{
1615	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_IDT_VECTORING_INFO))
1616	{
1617	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_INFO, &pVmxTransient->uIdtVectoringInfo);
1618	AssertRC(rc);
1619	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_IDT_VECTORING_INFO;
1620	}
1621	}
1622
1623
1624	/**
1625	* Reads the IDT-vectoring error code from the VMCS into the VMX
1626	* transient structure.
1627	*
1628	* @param pVmxTransient The VMX-transient structure.
1629	*/
1630	DECLINLINE(void) hmR0VmxReadIdtVectoringErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1631	{
1632	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_IDT_VECTORING_ERROR_CODE))
1633	{
1634	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
1635	AssertRC(rc);
1636	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_IDT_VECTORING_ERROR_CODE;
1637	}
1638	}
1639
1640	#ifdef HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
1641	/**
1642	* Reads all relevant read-only VMCS fields into the VMX transient structure.
1643	*
1644	* @param pVmxTransient The VMX-transient structure.
1645	*/
1646	static void hmR0VmxReadAllRoFieldsVmcs(PVMXTRANSIENT pVmxTransient)
1647	{
1648	int rc = VMXReadVmcsNw(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQual);
1649	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbExitInstr);
1650	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
1651	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_INFO, &pVmxTransient->uIdtVectoringInfo);
1652	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
1653	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
1654	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
1655	rc \|= VMXReadVmcsNw(VMX_VMCS_RO_GUEST_LINEAR_ADDR, &pVmxTransient->uGuestLinearAddr);
1656	rc \|= VMXReadVmcs64(VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL, &pVmxTransient->uGuestPhysicalAddr);
1657	AssertRC(rc);
1658	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_QUALIFICATION
1659	\| HMVMX_READ_EXIT_INSTR_LEN
1660	\| HMVMX_READ_EXIT_INSTR_INFO
1661	\| HMVMX_READ_IDT_VECTORING_INFO
1662	\| HMVMX_READ_IDT_VECTORING_ERROR_CODE
1663	\| HMVMX_READ_EXIT_INTERRUPTION_INFO
1664	\| HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE
1665	\| HMVMX_READ_GUEST_LINEAR_ADDR
1666	\| HMVMX_READ_GUEST_PHYSICAL_ADDR;
1667	}
1668	#endif
1669
1670	/**
1671	* Enters VMX root mode operation on the current CPU.
1672	*
1673	* @returns VBox status code.
1674	* @param pHostCpu The HM physical-CPU structure.
1675	* @param pVM The cross context VM structure. Can be
1676	* NULL, after a resume.
1677	* @param HCPhysCpuPage Physical address of the VMXON region.
1678	* @param pvCpuPage Pointer to the VMXON region.
1679	*/
1680	static int hmR0VmxEnterRootMode(PHMPHYSCPU pHostCpu, PVMCC pVM, RTHCPHYS HCPhysCpuPage, void *pvCpuPage)
1681	{
1682	Assert(pHostCpu);
1683	Assert(HCPhysCpuPage && HCPhysCpuPage != NIL_RTHCPHYS);
1684	Assert(RT_ALIGN_T(HCPhysCpuPage, _4K, RTHCPHYS) == HCPhysCpuPage);
1685	Assert(pvCpuPage);
1686	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1687
1688	if (pVM)
1689	{
1690	/* Write the VMCS revision identifier to the VMXON region. */
1691	(uint32_t )pvCpuPage = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_ID);
1692	}
1693
1694	/* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with CR4. */
1695	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1696
1697	/* Enable the VMX bit in CR4 if necessary. */
1698	RTCCUINTREG const uOldCr4 = SUPR0ChangeCR4(X86_CR4_VMXE, RTCCUINTREG_MAX);
1699
1700	/* Record whether VMXE was already prior to us enabling it above. */
1701	pHostCpu->fVmxeAlreadyEnabled = RT_BOOL(uOldCr4 & X86_CR4_VMXE);
1702
1703	/* Enter VMX root mode. */
1704	int rc = VMXEnable(HCPhysCpuPage);
1705	if (RT_FAILURE(rc))
1706	{
1707	/* Restore CR4.VMXE if it was not set prior to our attempt to set it above. */
1708	if (!pHostCpu->fVmxeAlreadyEnabled)
1709	SUPR0ChangeCR4(0 /* fOrMask */, ~(uint64_t)X86_CR4_VMXE);
1710
1711	if (pVM)
1712	pVM->hm.s.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
1713	}
1714
1715	/* Restore interrupts. */
1716	ASMSetFlags(fEFlags);
1717	return rc;
1718	}
1719
1720
1721	/**
1722	* Exits VMX root mode operation on the current CPU.
1723	*
1724	* @returns VBox status code.
1725	* @param pHostCpu The HM physical-CPU structure.
1726	*/
1727	static int hmR0VmxLeaveRootMode(PHMPHYSCPU pHostCpu)
1728	{
1729	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1730
1731	/* Paranoid: Disable interrupts as, in theory, interrupts handlers might mess with CR4. */
1732	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1733
1734	/* If we're for some reason not in VMX root mode, then don't leave it. */
1735	RTCCUINTREG const uHostCr4 = ASMGetCR4();
1736
1737	int rc;
1738	if (uHostCr4 & X86_CR4_VMXE)
1739	{
1740	/* Exit VMX root mode and clear the VMX bit in CR4. */
1741	VMXDisable();
1742
1743	/* Clear CR4.VMXE only if it was clear prior to use setting it. */
1744	if (!pHostCpu->fVmxeAlreadyEnabled)
1745	SUPR0ChangeCR4(0 /* fOrMask */, ~(uint64_t)X86_CR4_VMXE);
1746
1747	rc = VINF_SUCCESS;
1748	}
1749	else
1750	rc = VERR_VMX_NOT_IN_VMX_ROOT_MODE;
1751
1752	/* Restore interrupts. */
1753	ASMSetFlags(fEFlags);
1754	return rc;
1755	}
1756
1757
1758	/**
1759	* Allocates pages specified as specified by an array of VMX page allocation info
1760	* objects.
1761	*
1762	* The pages contents are zero'd after allocation.
1763	*
1764	* @returns VBox status code.
1765	* @param hMemObj The ring-0 memory object associated with the allocation.
1766	* @param paAllocInfo The pointer to the first element of the VMX
1767	* page-allocation info object array.
1768	* @param cEntries The number of elements in the @a paAllocInfo array.
1769	*/
1770	static int hmR0VmxPagesAllocZ(RTR0MEMOBJ hMemObj, PVMXPAGEALLOCINFO paAllocInfo, uint32_t cEntries)
1771	{
1772	/* Figure out how many pages to allocate. */
1773	uint32_t cPages = 0;
1774	for (uint32_t iPage = 0; iPage < cEntries; iPage++)
1775	cPages += !!paAllocInfo[iPage].fValid;
1776
1777	/* Allocate the pages. */
1778	if (cPages)
1779	{
1780	size_t const cbPages = cPages << X86_PAGE_4K_SHIFT;
1781	int rc = RTR0MemObjAllocPage(&hMemObj, cbPages, false /* fExecutable */);
1782	if (RT_FAILURE(rc))
1783	return rc;
1784
1785	/* Zero the contents and assign each page to the corresponding VMX page-allocation entry. */
1786	void *pvFirstPage = RTR0MemObjAddress(hMemObj);
1787	ASMMemZero32(pvFirstPage, cbPages);
1788
1789	uint32_t iPage = 0;
1790	for (uint32_t i = 0; i < cEntries; i++)
1791	if (paAllocInfo[i].fValid)
1792	{
1793	RTHCPHYS const HCPhysPage = RTR0MemObjGetPagePhysAddr(hMemObj, iPage);
1794	void pvPage = (void )((uintptr_t)pvFirstPage + (iPage << X86_PAGE_4K_SHIFT));
1795	Assert(HCPhysPage && HCPhysPage != NIL_RTHCPHYS);
1796	AssertPtr(pvPage);
1797
1798	Assert(paAllocInfo[iPage].pHCPhys);
1799	Assert(paAllocInfo[iPage].ppVirt);
1800	*paAllocInfo[iPage].pHCPhys = HCPhysPage;
1801	*paAllocInfo[iPage].ppVirt = pvPage;
1802
1803	/* Move to next page. */
1804	++iPage;
1805	}
1806
1807	/* Make sure all valid (requested) pages have been assigned. */
1808	Assert(iPage == cPages);
1809	}
1810	return VINF_SUCCESS;
1811	}
1812
1813
1814	/**
1815	* Frees pages allocated using hmR0VmxPagesAllocZ.
1816	*
1817	* @param hMemObj The ring-0 memory object associated with the allocation.
1818	*/
1819	DECL_FORCE_INLINE(void) hmR0VmxPagesFree(RTR0MEMOBJ hMemObj)
1820	{
1821	/* We can cleanup wholesale since it's all one allocation. */
1822	RTR0MemObjFree(hMemObj, true /* fFreeMappings */);
1823	}
1824
1825
1826	/**
1827	* Initializes a VMCS info. object.
1828	*
1829	* @param pVmcsInfo The VMCS info. object.
1830	*/
1831	static void hmR0VmxVmcsInfoInit(PVMXVMCSINFO pVmcsInfo)
1832	{
1833	memset(pVmcsInfo, 0, sizeof(*pVmcsInfo));
1834
1835	Assert(pVmcsInfo->hMemObj == NIL_RTR0MEMOBJ);
1836	pVmcsInfo->HCPhysVmcs = NIL_RTHCPHYS;
1837	pVmcsInfo->HCPhysShadowVmcs = NIL_RTHCPHYS;
1838	pVmcsInfo->HCPhysMsrBitmap = NIL_RTHCPHYS;
1839	pVmcsInfo->HCPhysGuestMsrLoad = NIL_RTHCPHYS;
1840	pVmcsInfo->HCPhysGuestMsrStore = NIL_RTHCPHYS;
1841	pVmcsInfo->HCPhysHostMsrLoad = NIL_RTHCPHYS;
1842	pVmcsInfo->HCPhysVirtApic = NIL_RTHCPHYS;
1843	pVmcsInfo->HCPhysEPTP = NIL_RTHCPHYS;
1844	pVmcsInfo->u64VmcsLinkPtr = NIL_RTHCPHYS;
1845	pVmcsInfo->idHostCpuState = NIL_RTCPUID;
1846	pVmcsInfo->idHostCpuExec = NIL_RTCPUID;
1847	}
1848
1849
1850	/**
1851	* Frees the VT-x structures for a VMCS info. object.
1852	*
1853	* @param pVmcsInfo The VMCS info. object.
1854	*/
1855	static void hmR0VmxVmcsInfoFree(PVMXVMCSINFO pVmcsInfo)
1856	{
1857	if (pVmcsInfo->hMemObj != NIL_RTR0MEMOBJ)
1858	{
1859	hmR0VmxPagesFree(pVmcsInfo->hMemObj);
1860	hmR0VmxVmcsInfoInit(pVmcsInfo);
1861	}
1862	}
1863
1864
1865	/**
1866	* Allocates the VT-x structures for a VMCS info. object.
1867	*
1868	* @returns VBox status code.
1869	* @param pVCpu The cross context virtual CPU structure.
1870	* @param pVmcsInfo The VMCS info. object.
1871	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
1872	*
1873	* @remarks The caller is expected to take care of any and all allocation failures.
1874	* This function will not perform any cleanup for failures half-way
1875	* through.
1876	*/
1877	static int hmR0VmxAllocVmcsInfo(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
1878	{
1879	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1880
1881	bool const fMsrBitmaps = RT_BOOL(pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS);
1882	bool const fShadowVmcs = !fIsNstGstVmcs ? pVM->hm.s.vmx.fUseVmcsShadowing : pVM->cpum.ro.GuestFeatures.fVmxVmcsShadowing;
1883	Assert(!pVM->cpum.ro.GuestFeatures.fVmxVmcsShadowing); /* VMCS shadowing is not yet exposed to the guest. */
1884	VMXPAGEALLOCINFO aAllocInfo[] = {
1885	{ true, 0 /* Unused */, &pVmcsInfo->HCPhysVmcs, &pVmcsInfo->pvVmcs },
1886	{ true, 0 /* Unused */, &pVmcsInfo->HCPhysGuestMsrLoad, &pVmcsInfo->pvGuestMsrLoad },
1887	{ true, 0 /* Unused */, &pVmcsInfo->HCPhysHostMsrLoad, &pVmcsInfo->pvHostMsrLoad },
1888	{ fMsrBitmaps, 0 /* Unused */, &pVmcsInfo->HCPhysMsrBitmap, &pVmcsInfo->pvMsrBitmap },
1889	{ fShadowVmcs, 0 /* Unused */, &pVmcsInfo->HCPhysShadowVmcs, &pVmcsInfo->pvShadowVmcs },
1890	};
1891
1892	int rc = hmR0VmxPagesAllocZ(pVmcsInfo->hMemObj, &aAllocInfo[0], RT_ELEMENTS(aAllocInfo));
1893	if (RT_FAILURE(rc))
1894	return rc;
1895
1896	/*
1897	* We use the same page for VM-entry MSR-load and VM-exit MSR store areas.
1898	* Because they contain a symmetric list of guest MSRs to load on VM-entry and store on VM-exit.
1899	*/
1900	AssertCompile(RT_ELEMENTS(aAllocInfo) > 0);
1901	Assert(pVmcsInfo->HCPhysGuestMsrLoad != NIL_RTHCPHYS);
1902	pVmcsInfo->pvGuestMsrStore = pVmcsInfo->pvGuestMsrLoad;
1903	pVmcsInfo->HCPhysGuestMsrStore = pVmcsInfo->HCPhysGuestMsrLoad;
1904
1905	/*
1906	* Get the virtual-APIC page rather than allocating them again.
1907	*/
1908	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW)
1909	{
1910	if (!fIsNstGstVmcs)
1911	{
1912	if (PDMHasApic(pVM))
1913	{
1914	rc = APICGetApicPageForCpu(pVCpu, &pVmcsInfo->HCPhysVirtApic, (PRTR0PTR)&pVmcsInfo->pbVirtApic, NULL /pR3Ptr/);
1915	if (RT_FAILURE(rc))
1916	return rc;
1917	Assert(pVmcsInfo->pbVirtApic);
1918	Assert(pVmcsInfo->HCPhysVirtApic && pVmcsInfo->HCPhysVirtApic != NIL_RTHCPHYS);
1919	}
1920	}
1921	else
1922	{
1923	pVmcsInfo->pbVirtApic = (uint8_t *)CPUMGetGuestVmxVirtApicPage(&pVCpu->cpum.GstCtx, &pVmcsInfo->HCPhysVirtApic);
1924	Assert(pVmcsInfo->pbVirtApic);
1925	Assert(pVmcsInfo->HCPhysVirtApic && pVmcsInfo->HCPhysVirtApic != NIL_RTHCPHYS);
1926	}
1927	}
1928
1929	return VINF_SUCCESS;
1930	}
1931
1932
1933	/**
1934	* Free all VT-x structures for the VM.
1935	*
1936	* @returns IPRT status code.
1937	* @param pVM The cross context VM structure.
1938	*/
1939	static void hmR0VmxStructsFree(PVMCC pVM)
1940	{
1941	hmR0VmxPagesFree(pVM->hm.s.vmx.hMemObj);
1942	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1943	if (pVM->hm.s.vmx.fUseVmcsShadowing)
1944	{
1945	RTMemFree(pVM->hm.s.vmx.paShadowVmcsFields);
1946	RTMemFree(pVM->hm.s.vmx.paShadowVmcsRoFields);
1947	}
1948	#endif
1949
1950	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
1951	{
1952	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
1953	hmR0VmxVmcsInfoFree(&pVCpu->hm.s.vmx.VmcsInfo);
1954	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1955	if (pVM->cpum.ro.GuestFeatures.fVmx)
1956	hmR0VmxVmcsInfoFree(&pVCpu->hm.s.vmx.VmcsInfoNstGst);
1957	#endif
1958	}
1959	}
1960
1961
1962	/**
1963	* Allocate all VT-x structures for the VM.
1964	*
1965	* @returns IPRT status code.
1966	* @param pVM The cross context VM structure.
1967	*
1968	* @remarks This functions will cleanup on memory allocation failures.
1969	*/
1970	static int hmR0VmxStructsAlloc(PVMCC pVM)
1971	{
1972	/*
1973	* Sanity check the VMCS size reported by the CPU as we assume 4KB allocations.
1974	* The VMCS size cannot be more than 4096 bytes.
1975	*
1976	* See Intel spec. Appendix A.1 "Basic VMX Information".
1977	*/
1978	uint32_t const cbVmcs = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_SIZE);
1979	if (cbVmcs <= X86_PAGE_4K_SIZE)
1980	{ /* likely */ }
1981	else
1982	{
1983	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_INVALID_VMCS_SIZE;
1984	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
1985	}
1986
1987	/*
1988	* Allocate per-VM VT-x structures.
1989	*/
1990	bool const fVirtApicAccess = RT_BOOL(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS);
1991	bool const fUseVmcsShadowing = pVM->hm.s.vmx.fUseVmcsShadowing;
1992	VMXPAGEALLOCINFO aAllocInfo[] = {
1993	{ fVirtApicAccess, 0 /* Unused */, &pVM->hm.s.vmx.HCPhysApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess },
1994	{ fUseVmcsShadowing, 0 /* Unused */, &pVM->hm.s.vmx.HCPhysVmreadBitmap, &pVM->hm.s.vmx.pvVmreadBitmap },
1995	{ fUseVmcsShadowing, 0 /* Unused */, &pVM->hm.s.vmx.HCPhysVmwriteBitmap, &pVM->hm.s.vmx.pvVmwriteBitmap },
1996	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
1997	{ true, 0 /* Unused */, &pVM->hm.s.vmx.HCPhysScratch, &(PRTR0PTR)pVM->hm.s.vmx.pbScratch },
1998	#endif
1999	};
2000
2001	int rc = hmR0VmxPagesAllocZ(pVM->hm.s.vmx.hMemObj, &aAllocInfo[0], RT_ELEMENTS(aAllocInfo));
2002	if (RT_FAILURE(rc))
2003	goto cleanup;
2004
2005	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2006	/* Allocate the shadow VMCS-fields array. */
2007	if (fUseVmcsShadowing)
2008	{
2009	Assert(!pVM->hm.s.vmx.cShadowVmcsFields);
2010	Assert(!pVM->hm.s.vmx.cShadowVmcsRoFields);
2011	pVM->hm.s.vmx.paShadowVmcsFields = (uint32_t *)RTMemAllocZ(sizeof(g_aVmcsFields));
2012	pVM->hm.s.vmx.paShadowVmcsRoFields = (uint32_t *)RTMemAllocZ(sizeof(g_aVmcsFields));
2013	if (RT_LIKELY( pVM->hm.s.vmx.paShadowVmcsFields
2014	&& pVM->hm.s.vmx.paShadowVmcsRoFields))
2015	{ /* likely */ }
2016	else
2017	{
2018	rc = VERR_NO_MEMORY;
2019	goto cleanup;
2020	}
2021	}
2022	#endif
2023
2024	/*
2025	* Allocate per-VCPU VT-x structures.
2026	*/
2027	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
2028	{
2029	/* Allocate the guest VMCS structures. */
2030	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
2031	rc = hmR0VmxAllocVmcsInfo(pVCpu, &pVCpu->hm.s.vmx.VmcsInfo, false /* fIsNstGstVmcs */);
2032	if (RT_FAILURE(rc))
2033	goto cleanup;
2034
2035	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2036	/* Allocate the nested-guest VMCS structures, when the VMX feature is exposed to the guest. */
2037	if (pVM->cpum.ro.GuestFeatures.fVmx)
2038	{
2039	rc = hmR0VmxAllocVmcsInfo(pVCpu, &pVCpu->hm.s.vmx.VmcsInfoNstGst, true /* fIsNstGstVmcs */);
2040	if (RT_FAILURE(rc))
2041	goto cleanup;
2042	}
2043	#endif
2044	}
2045
2046	return VINF_SUCCESS;
2047
2048	cleanup:
2049	hmR0VmxStructsFree(pVM);
2050	Assert(rc != VINF_SUCCESS);
2051	return rc;
2052	}
2053
2054
2055	/**
2056	* Pre-initializes non-zero fields in VMX structures that will be allocated.
2057	*
2058	* @param pVM The cross context VM structure.
2059	*/
2060	static void hmR0VmxStructsInit(PVMCC pVM)
2061	{
2062	/* Paranoia. */
2063	Assert(pVM->hm.s.vmx.pbApicAccess == NULL);
2064	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2065	Assert(pVM->hm.s.vmx.pbScratch == NULL);
2066	#endif
2067
2068	/*
2069	* Initialize members up-front so we can cleanup en masse on allocation failures.
2070	*/
2071	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2072	pVM->hm.s.vmx.HCPhysScratch = NIL_RTHCPHYS;
2073	#endif
2074	pVM->hm.s.vmx.HCPhysApicAccess = NIL_RTHCPHYS;
2075	pVM->hm.s.vmx.HCPhysVmreadBitmap = NIL_RTHCPHYS;
2076	pVM->hm.s.vmx.HCPhysVmwriteBitmap = NIL_RTHCPHYS;
2077	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
2078	{
2079	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
2080	hmR0VmxVmcsInfoInit(&pVCpu->hm.s.vmx.VmcsInfo);
2081	hmR0VmxVmcsInfoInit(&pVCpu->hm.s.vmx.VmcsInfoNstGst);
2082	}
2083	}
2084
2085	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2086	/**
2087	* Returns whether an MSR at the given MSR-bitmap offset is intercepted or not.
2088	*
2089	* @returns @c true if the MSR is intercepted, @c false otherwise.
2090	* @param pvMsrBitmap The MSR bitmap.
2091	* @param offMsr The MSR byte offset.
2092	* @param iBit The bit offset from the byte offset.
2093	*/
2094	DECLINLINE(bool) hmR0VmxIsMsrBitSet(const void *pvMsrBitmap, uint16_t offMsr, int32_t iBit)
2095	{
2096	uint8_t const * const pbMsrBitmap = (uint8_t const * const)pvMsrBitmap;
2097	Assert(pbMsrBitmap);
2098	Assert(offMsr + (iBit >> 3) <= X86_PAGE_4K_SIZE);
2099	return ASMBitTest(pbMsrBitmap + offMsr, iBit);
2100	}
2101	#endif
2102
2103	/**
2104	* Sets the permission bits for the specified MSR in the given MSR bitmap.
2105	*
2106	* If the passed VMCS is a nested-guest VMCS, this function ensures that the
2107	* read/write intercept is cleared from the MSR bitmap used for hardware-assisted
2108	* VMX execution of the nested-guest, only if nested-guest is also not intercepting
2109	* the read/write access of this MSR.
2110	*
2111	* @param pVCpu The cross context virtual CPU structure.
2112	* @param pVmcsInfo The VMCS info. object.
2113	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2114	* @param idMsr The MSR value.
2115	* @param fMsrpm The MSR permissions (see VMXMSRPM_XXX). This must
2116	* include both a read -and- a write permission!
2117	*
2118	* @sa CPUMGetVmxMsrPermission.
2119	* @remarks Can be called with interrupts disabled.
2120	*/
2121	static void hmR0VmxSetMsrPermission(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs, uint32_t idMsr, uint32_t fMsrpm)
2122	{
2123	uint8_t pbMsrBitmap = (uint8_t )pVmcsInfo->pvMsrBitmap;
2124	Assert(pbMsrBitmap);
2125	Assert(VMXMSRPM_IS_FLAG_VALID(fMsrpm));
2126
2127	/*
2128	* MSR-bitmap Layout:
2129	* Byte index MSR range Interpreted as
2130	* 0x000 - 0x3ff 0x00000000 - 0x00001fff Low MSR read bits.
2131	* 0x400 - 0x7ff 0xc0000000 - 0xc0001fff High MSR read bits.
2132	* 0x800 - 0xbff 0x00000000 - 0x00001fff Low MSR write bits.
2133	* 0xc00 - 0xfff 0xc0000000 - 0xc0001fff High MSR write bits.
2134	*
2135	* A bit corresponding to an MSR within the above range causes a VM-exit
2136	* if the bit is 1 on executions of RDMSR/WRMSR. If an MSR falls out of
2137	* the MSR range, it always cause a VM-exit.
2138	*
2139	* See Intel spec. 24.6.9 "MSR-Bitmap Address".
2140	*/
2141	uint16_t const offBitmapRead = 0;
2142	uint16_t const offBitmapWrite = 0x800;
2143	uint16_t offMsr;
2144	int32_t iBit;
2145	if (idMsr <= UINT32_C(0x00001fff))
2146	{
2147	offMsr = 0;
2148	iBit = idMsr;
2149	}
2150	else if (idMsr - UINT32_C(0xc0000000) <= UINT32_C(0x00001fff))
2151	{
2152	offMsr = 0x400;
2153	iBit = idMsr - UINT32_C(0xc0000000);
2154	}
2155	else
2156	AssertMsgFailedReturnVoid(("Invalid MSR %#RX32\n", idMsr));
2157
2158	/*
2159	* Set the MSR read permission.
2160	*/
2161	uint16_t const offMsrRead = offBitmapRead + offMsr;
2162	Assert(offMsrRead + (iBit >> 3) < offBitmapWrite);
2163	if (fMsrpm & VMXMSRPM_ALLOW_RD)
2164	{
2165	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2166	bool const fClear = !fIsNstGstVmcs ? true
2167	: !hmR0VmxIsMsrBitSet(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), offMsrRead, iBit);
2168	#else
2169	RT_NOREF2(pVCpu, fIsNstGstVmcs);
2170	bool const fClear = true;
2171	#endif
2172	if (fClear)
2173	ASMBitClear(pbMsrBitmap + offMsrRead, iBit);
2174	}
2175	else
2176	ASMBitSet(pbMsrBitmap + offMsrRead, iBit);
2177
2178	/*
2179	* Set the MSR write permission.
2180	*/
2181	uint16_t const offMsrWrite = offBitmapWrite + offMsr;
2182	Assert(offMsrWrite + (iBit >> 3) < X86_PAGE_4K_SIZE);
2183	if (fMsrpm & VMXMSRPM_ALLOW_WR)
2184	{
2185	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2186	bool const fClear = !fIsNstGstVmcs ? true
2187	: !hmR0VmxIsMsrBitSet(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), offMsrWrite, iBit);
2188	#else
2189	RT_NOREF2(pVCpu, fIsNstGstVmcs);
2190	bool const fClear = true;
2191	#endif
2192	if (fClear)
2193	ASMBitClear(pbMsrBitmap + offMsrWrite, iBit);
2194	}
2195	else
2196	ASMBitSet(pbMsrBitmap + offMsrWrite, iBit);
2197	}
2198
2199
2200	/**
2201	* Updates the VMCS with the number of effective MSRs in the auto-load/store MSR
2202	* area.
2203	*
2204	* @returns VBox status code.
2205	* @param pVCpu The cross context virtual CPU structure.
2206	* @param pVmcsInfo The VMCS info. object.
2207	* @param cMsrs The number of MSRs.
2208	*/
2209	static int hmR0VmxSetAutoLoadStoreMsrCount(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint32_t cMsrs)
2210	{
2211	/* Shouldn't ever happen but there -is- a number. We're well within the recommended 512. */
2212	uint32_t const cMaxSupportedMsrs = VMX_MISC_MAX_MSRS(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.u64Misc);
2213	if (RT_LIKELY(cMsrs < cMaxSupportedMsrs))
2214	{
2215	/* Commit the MSR counts to the VMCS and update the cache. */
2216	if (pVmcsInfo->cEntryMsrLoad != cMsrs)
2217	{
2218	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, cMsrs); AssertRC(rc);
2219	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, cMsrs); AssertRC(rc);
2220	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, cMsrs); AssertRC(rc);
2221	pVmcsInfo->cEntryMsrLoad = cMsrs;
2222	pVmcsInfo->cExitMsrStore = cMsrs;
2223	pVmcsInfo->cExitMsrLoad = cMsrs;
2224	}
2225	return VINF_SUCCESS;
2226	}
2227
2228	LogRel(("Auto-load/store MSR count exceeded! cMsrs=%u MaxSupported=%u\n", cMsrs, cMaxSupportedMsrs));
2229	pVCpu->hm.s.u32HMError = VMX_UFC_INSUFFICIENT_GUEST_MSR_STORAGE;
2230	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2231	}
2232
2233
2234	/**
2235	* Adds a new (or updates the value of an existing) guest/host MSR
2236	* pair to be swapped during the world-switch as part of the
2237	* auto-load/store MSR area in the VMCS.
2238	*
2239	* @returns VBox status code.
2240	* @param pVCpu The cross context virtual CPU structure.
2241	* @param pVmxTransient The VMX-transient structure.
2242	* @param idMsr The MSR.
2243	* @param uGuestMsrValue Value of the guest MSR.
2244	* @param fSetReadWrite Whether to set the guest read/write access of this
2245	* MSR (thus not causing a VM-exit).
2246	* @param fUpdateHostMsr Whether to update the value of the host MSR if
2247	* necessary.
2248	*/
2249	static int hmR0VmxAddAutoLoadStoreMsr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t idMsr, uint64_t uGuestMsrValue,
2250	bool fSetReadWrite, bool fUpdateHostMsr)
2251	{
2252	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
2253	bool const fIsNstGstVmcs = pVmxTransient->fIsNestedGuest;
2254	PVMXAUTOMSR pGuestMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2255	uint32_t cMsrs = pVmcsInfo->cEntryMsrLoad;
2256	uint32_t i;
2257
2258	/* Paranoia. */
2259	Assert(pGuestMsrLoad);
2260
2261	LogFlowFunc(("pVCpu=%p idMsr=%#RX32 uGestMsrValue=%#RX64\n", pVCpu, idMsr, uGuestMsrValue));
2262
2263	/* Check if the MSR already exists in the VM-entry MSR-load area. */
2264	for (i = 0; i < cMsrs; i++)
2265	{
2266	if (pGuestMsrLoad[i].u32Msr == idMsr)
2267	break;
2268	}
2269
2270	bool fAdded = false;
2271	if (i == cMsrs)
2272	{
2273	/* The MSR does not exist, bump the MSR count to make room for the new MSR. */
2274	++cMsrs;
2275	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, pVmcsInfo, cMsrs);
2276	AssertMsgRCReturn(rc, ("Insufficient space to add MSR to VM-entry MSR-load/store area %u\n", idMsr), rc);
2277
2278	/* Set the guest to read/write this MSR without causing VM-exits. */
2279	if ( fSetReadWrite
2280	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS))
2281	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, fIsNstGstVmcs, idMsr, VMXMSRPM_ALLOW_RD_WR);
2282
2283	Log4Func(("Added MSR %#RX32, cMsrs=%u\n", idMsr, cMsrs));
2284	fAdded = true;
2285	}
2286
2287	/* Update the MSR value for the newly added or already existing MSR. */
2288	pGuestMsrLoad[i].u32Msr = idMsr;
2289	pGuestMsrLoad[i].u64Value = uGuestMsrValue;
2290
2291	/* Create the corresponding slot in the VM-exit MSR-store area if we use a different page. */
2292	if (hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo))
2293	{
2294	PVMXAUTOMSR pGuestMsrStore = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2295	pGuestMsrStore[i].u32Msr = idMsr;
2296	pGuestMsrStore[i].u64Value = uGuestMsrValue;
2297	}
2298
2299	/* Update the corresponding slot in the host MSR area. */
2300	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2301	Assert(pHostMsr != pVmcsInfo->pvGuestMsrLoad);
2302	Assert(pHostMsr != pVmcsInfo->pvGuestMsrStore);
2303	pHostMsr[i].u32Msr = idMsr;
2304
2305	/*
2306	* Only if the caller requests to update the host MSR value AND we've newly added the
2307	* MSR to the host MSR area do we actually update the value. Otherwise, it will be
2308	* updated by hmR0VmxUpdateAutoLoadHostMsrs().
2309	*
2310	* We do this for performance reasons since reading MSRs may be quite expensive.
2311	*/
2312	if (fAdded)
2313	{
2314	if (fUpdateHostMsr)
2315	{
2316	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2317	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2318	pHostMsr[i].u64Value = ASMRdMsr(idMsr);
2319	}
2320	else
2321	{
2322	/* Someone else can do the work. */
2323	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = false;
2324	}
2325	}
2326	return VINF_SUCCESS;
2327	}
2328
2329
2330	/**
2331	* Removes a guest/host MSR pair to be swapped during the world-switch from the
2332	* auto-load/store MSR area in the VMCS.
2333	*
2334	* @returns VBox status code.
2335	* @param pVCpu The cross context virtual CPU structure.
2336	* @param pVmxTransient The VMX-transient structure.
2337	* @param idMsr The MSR.
2338	*/
2339	static int hmR0VmxRemoveAutoLoadStoreMsr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t idMsr)
2340	{
2341	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
2342	bool const fIsNstGstVmcs = pVmxTransient->fIsNestedGuest;
2343	PVMXAUTOMSR pGuestMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2344	uint32_t cMsrs = pVmcsInfo->cEntryMsrLoad;
2345
2346	LogFlowFunc(("pVCpu=%p idMsr=%#RX32\n", pVCpu, idMsr));
2347
2348	for (uint32_t i = 0; i < cMsrs; i++)
2349	{
2350	/* Find the MSR. */
2351	if (pGuestMsrLoad[i].u32Msr == idMsr)
2352	{
2353	/*
2354	* If it's the last MSR, we only need to reduce the MSR count.
2355	* If it's -not- the last MSR, copy the last MSR in place of it and reduce the MSR count.
2356	*/
2357	if (i < cMsrs - 1)
2358	{
2359	/* Remove it from the VM-entry MSR-load area. */
2360	pGuestMsrLoad[i].u32Msr = pGuestMsrLoad[cMsrs - 1].u32Msr;
2361	pGuestMsrLoad[i].u64Value = pGuestMsrLoad[cMsrs - 1].u64Value;
2362
2363	/* Remove it from the VM-exit MSR-store area if it's in a different page. */
2364	if (hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo))
2365	{
2366	PVMXAUTOMSR pGuestMsrStore = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2367	Assert(pGuestMsrStore[i].u32Msr == idMsr);
2368	pGuestMsrStore[i].u32Msr = pGuestMsrStore[cMsrs - 1].u32Msr;
2369	pGuestMsrStore[i].u64Value = pGuestMsrStore[cMsrs - 1].u64Value;
2370	}
2371
2372	/* Remove it from the VM-exit MSR-load area. */
2373	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2374	Assert(pHostMsr[i].u32Msr == idMsr);
2375	pHostMsr[i].u32Msr = pHostMsr[cMsrs - 1].u32Msr;
2376	pHostMsr[i].u64Value = pHostMsr[cMsrs - 1].u64Value;
2377	}
2378
2379	/* Reduce the count to reflect the removed MSR and bail. */
2380	--cMsrs;
2381	break;
2382	}
2383	}
2384
2385	/* Update the VMCS if the count changed (meaning the MSR was found and removed). */
2386	if (cMsrs != pVmcsInfo->cEntryMsrLoad)
2387	{
2388	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, pVmcsInfo, cMsrs);
2389	AssertRCReturn(rc, rc);
2390
2391	/* We're no longer swapping MSRs during the world-switch, intercept guest read/writes to them. */
2392	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
2393	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, fIsNstGstVmcs, idMsr, VMXMSRPM_EXIT_RD \| VMXMSRPM_EXIT_WR);
2394
2395	Log4Func(("Removed MSR %#RX32, cMsrs=%u\n", idMsr, cMsrs));
2396	return VINF_SUCCESS;
2397	}
2398
2399	return VERR_NOT_FOUND;
2400	}
2401
2402
2403	/**
2404	* Checks if the specified guest MSR is part of the VM-entry MSR-load area.
2405	*
2406	* @returns @c true if found, @c false otherwise.
2407	* @param pVmcsInfo The VMCS info. object.
2408	* @param idMsr The MSR to find.
2409	*/
2410	static bool hmR0VmxIsAutoLoadGuestMsr(PCVMXVMCSINFO pVmcsInfo, uint32_t idMsr)
2411	{
2412	PCVMXAUTOMSR pMsrs = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2413	uint32_t const cMsrs = pVmcsInfo->cEntryMsrLoad;
2414	Assert(pMsrs);
2415	Assert(sizeof(pMsrs) cMsrs <= X86_PAGE_4K_SIZE);
2416	for (uint32_t i = 0; i < cMsrs; i++)
2417	{
2418	if (pMsrs[i].u32Msr == idMsr)
2419	return true;
2420	}
2421	return false;
2422	}
2423
2424
2425	/**
2426	* Updates the value of all host MSRs in the VM-exit MSR-load area.
2427	*
2428	* @param pVCpu The cross context virtual CPU structure.
2429	* @param pVmcsInfo The VMCS info. object.
2430	*
2431	* @remarks No-long-jump zone!!!
2432	*/
2433	static void hmR0VmxUpdateAutoLoadHostMsrs(PCVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
2434	{
2435	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2436
2437	PVMXAUTOMSR pHostMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2438	uint32_t const cMsrs = pVmcsInfo->cExitMsrLoad;
2439	Assert(pHostMsrLoad);
2440	Assert(sizeof(pHostMsrLoad) cMsrs <= X86_PAGE_4K_SIZE);
2441	LogFlowFunc(("pVCpu=%p cMsrs=%u\n", pVCpu, cMsrs));
2442	for (uint32_t i = 0; i < cMsrs; i++)
2443	{
2444	/*
2445	* Performance hack for the host EFER MSR. We use the cached value rather than re-read it.
2446	* Strict builds will catch mismatches in hmR0VmxCheckAutoLoadStoreMsrs(). See @bugref{7368}.
2447	*/
2448	if (pHostMsrLoad[i].u32Msr == MSR_K6_EFER)
2449	pHostMsrLoad[i].u64Value = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostMsrEfer;
2450	else
2451	pHostMsrLoad[i].u64Value = ASMRdMsr(pHostMsrLoad[i].u32Msr);
2452	}
2453	}
2454
2455
2456	/**
2457	* Saves a set of host MSRs to allow read/write passthru access to the guest and
2458	* perform lazy restoration of the host MSRs while leaving VT-x.
2459	*
2460	* @param pVCpu The cross context virtual CPU structure.
2461	*
2462	* @remarks No-long-jump zone!!!
2463	*/
2464	static void hmR0VmxLazySaveHostMsrs(PVMCPUCC pVCpu)
2465	{
2466	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2467
2468	/*
2469	* Note: If you're adding MSRs here, make sure to update the MSR-bitmap accesses in hmR0VmxSetupVmcsProcCtls().
2470	*/
2471	if (!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST))
2472	{
2473	Assert(!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)); /* Guest MSRs better not be loaded now. */
2474	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
2475	{
2476	pVCpu->hm.s.vmx.u64HostMsrLStar = ASMRdMsr(MSR_K8_LSTAR);
2477	pVCpu->hm.s.vmx.u64HostMsrStar = ASMRdMsr(MSR_K6_STAR);
2478	pVCpu->hm.s.vmx.u64HostMsrSfMask = ASMRdMsr(MSR_K8_SF_MASK);
2479	pVCpu->hm.s.vmx.u64HostMsrKernelGsBase = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
2480	}
2481	pVCpu->hm.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_SAVED_HOST;
2482	}
2483	}
2484
2485
2486	/**
2487	* Checks whether the MSR belongs to the set of guest MSRs that we restore
2488	* lazily while leaving VT-x.
2489	*
2490	* @returns true if it does, false otherwise.
2491	* @param pVCpu The cross context virtual CPU structure.
2492	* @param idMsr The MSR to check.
2493	*/
2494	static bool hmR0VmxIsLazyGuestMsr(PCVMCPUCC pVCpu, uint32_t idMsr)
2495	{
2496	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
2497	{
2498	switch (idMsr)
2499	{
2500	case MSR_K8_LSTAR:
2501	case MSR_K6_STAR:
2502	case MSR_K8_SF_MASK:
2503	case MSR_K8_KERNEL_GS_BASE:
2504	return true;
2505	}
2506	}
2507	return false;
2508	}
2509
2510
2511	/**
2512	* Loads a set of guests MSRs to allow read/passthru to the guest.
2513	*
2514	* The name of this function is slightly confusing. This function does NOT
2515	* postpone loading, but loads the MSR right now. "hmR0VmxLazy" is simply a
2516	* common prefix for functions dealing with "lazy restoration" of the shared
2517	* MSRs.
2518	*
2519	* @param pVCpu The cross context virtual CPU structure.
2520	*
2521	* @remarks No-long-jump zone!!!
2522	*/
2523	static void hmR0VmxLazyLoadGuestMsrs(PVMCPUCC pVCpu)
2524	{
2525	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2526	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2527
2528	Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
2529	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
2530	{
2531	/*
2532	* If the guest MSRs are not loaded -and- if all the guest MSRs are identical
2533	* to the MSRs on the CPU (which are the saved host MSRs, see assertion above) then
2534	* we can skip a few MSR writes.
2535	*
2536	* Otherwise, it implies either 1. they're not loaded, or 2. they're loaded but the
2537	* guest MSR values in the guest-CPU context might be different to what's currently
2538	* loaded in the CPU. In either case, we need to write the new guest MSR values to the
2539	* CPU, see @bugref{8728}.
2540	*/
2541	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
2542	if ( !(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
2543	&& pCtx->msrKERNELGSBASE == pVCpu->hm.s.vmx.u64HostMsrKernelGsBase
2544	&& pCtx->msrLSTAR == pVCpu->hm.s.vmx.u64HostMsrLStar
2545	&& pCtx->msrSTAR == pVCpu->hm.s.vmx.u64HostMsrStar
2546	&& pCtx->msrSFMASK == pVCpu->hm.s.vmx.u64HostMsrSfMask)
2547	{
2548	#ifdef VBOX_STRICT
2549	Assert(ASMRdMsr(MSR_K8_KERNEL_GS_BASE) == pCtx->msrKERNELGSBASE);
2550	Assert(ASMRdMsr(MSR_K8_LSTAR) == pCtx->msrLSTAR);
2551	Assert(ASMRdMsr(MSR_K6_STAR) == pCtx->msrSTAR);
2552	Assert(ASMRdMsr(MSR_K8_SF_MASK) == pCtx->msrSFMASK);
2553	#endif
2554	}
2555	else
2556	{
2557	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pCtx->msrKERNELGSBASE);
2558	ASMWrMsr(MSR_K8_LSTAR, pCtx->msrLSTAR);
2559	ASMWrMsr(MSR_K6_STAR, pCtx->msrSTAR);
2560	ASMWrMsr(MSR_K8_SF_MASK, pCtx->msrSFMASK);
2561	}
2562	}
2563	pVCpu->hm.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_LOADED_GUEST;
2564	}
2565
2566
2567	/**
2568	* Performs lazy restoration of the set of host MSRs if they were previously
2569	* loaded with guest MSR values.
2570	*
2571	* @param pVCpu The cross context virtual CPU structure.
2572	*
2573	* @remarks No-long-jump zone!!!
2574	* @remarks The guest MSRs should have been saved back into the guest-CPU
2575	* context by hmR0VmxImportGuestState()!!!
2576	*/
2577	static void hmR0VmxLazyRestoreHostMsrs(PVMCPUCC pVCpu)
2578	{
2579	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2580	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2581
2582	if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
2583	{
2584	Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
2585	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
2586	{
2587	ASMWrMsr(MSR_K8_LSTAR, pVCpu->hm.s.vmx.u64HostMsrLStar);
2588	ASMWrMsr(MSR_K6_STAR, pVCpu->hm.s.vmx.u64HostMsrStar);
2589	ASMWrMsr(MSR_K8_SF_MASK, pVCpu->hm.s.vmx.u64HostMsrSfMask);
2590	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pVCpu->hm.s.vmx.u64HostMsrKernelGsBase);
2591	}
2592	}
2593	pVCpu->hm.s.vmx.fLazyMsrs &= ~(VMX_LAZY_MSRS_LOADED_GUEST \| VMX_LAZY_MSRS_SAVED_HOST);
2594	}
2595
2596
2597	/**
2598	* Verifies that our cached values of the VMCS fields are all consistent with
2599	* what's actually present in the VMCS.
2600	*
2601	* @returns VBox status code.
2602	* @retval VINF_SUCCESS if all our caches match their respective VMCS fields.
2603	* @retval VERR_VMX_VMCS_FIELD_CACHE_INVALID if a cache field doesn't match the
2604	* VMCS content. HMCPU error-field is
2605	* updated, see VMX_VCI_XXX.
2606	* @param pVCpu The cross context virtual CPU structure.
2607	* @param pVmcsInfo The VMCS info. object.
2608	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2609	*/
2610	static int hmR0VmxCheckCachedVmcsCtls(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
2611	{
2612	const char * const pcszVmcs = fIsNstGstVmcs ? "Nested-guest VMCS" : "VMCS";
2613
2614	uint32_t u32Val;
2615	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
2616	AssertRC(rc);
2617	AssertMsgReturnStmt(pVmcsInfo->u32EntryCtls == u32Val,
2618	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32EntryCtls, u32Val),
2619	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_ENTRY,
2620	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2621
2622	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val);
2623	AssertRC(rc);
2624	AssertMsgReturnStmt(pVmcsInfo->u32ExitCtls == u32Val,
2625	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ExitCtls, u32Val),
2626	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_EXIT,
2627	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2628
2629	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val);
2630	AssertRC(rc);
2631	AssertMsgReturnStmt(pVmcsInfo->u32PinCtls == u32Val,
2632	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32PinCtls, u32Val),
2633	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PIN_EXEC,
2634	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2635
2636	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val);
2637	AssertRC(rc);
2638	AssertMsgReturnStmt(pVmcsInfo->u32ProcCtls == u32Val,
2639	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ProcCtls, u32Val),
2640	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC,
2641	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2642
2643	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
2644	{
2645	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val);
2646	AssertRC(rc);
2647	AssertMsgReturnStmt(pVmcsInfo->u32ProcCtls2 == u32Val,
2648	("%s controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ProcCtls2, u32Val),
2649	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC2,
2650	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2651	}
2652
2653	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, &u32Val);
2654	AssertRC(rc);
2655	AssertMsgReturnStmt(pVmcsInfo->u32XcptBitmap == u32Val,
2656	("%s exception bitmap mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32XcptBitmap, u32Val),
2657	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_XCPT_BITMAP,
2658	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2659
2660	uint64_t u64Val;
2661	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, &u64Val);
2662	AssertRC(rc);
2663	AssertMsgReturnStmt(pVmcsInfo->u64TscOffset == u64Val,
2664	("%s TSC offset mismatch: Cache=%#RX64 VMCS=%#RX64\n", pcszVmcs, pVmcsInfo->u64TscOffset, u64Val),
2665	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_TSC_OFFSET,
2666	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2667
2668	NOREF(pcszVmcs);
2669	return VINF_SUCCESS;
2670	}
2671
2672
2673	#ifdef VBOX_STRICT
2674	/**
2675	* Verifies that our cached host EFER MSR value has not changed since we cached it.
2676	*
2677	* @param pVCpu The cross context virtual CPU structure.
2678	* @param pVmcsInfo The VMCS info. object.
2679	*/
2680	static void hmR0VmxCheckHostEferMsr(PCVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
2681	{
2682	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2683
2684	if (pVmcsInfo->u32ExitCtls & VMX_EXIT_CTLS_LOAD_EFER_MSR)
2685	{
2686	uint64_t const uHostEferMsr = ASMRdMsr(MSR_K6_EFER);
2687	uint64_t const uHostEferMsrCache = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostMsrEfer;
2688	uint64_t uVmcsEferMsrVmcs;
2689	int rc = VMXReadVmcs64(VMX_VMCS64_HOST_EFER_FULL, &uVmcsEferMsrVmcs);
2690	AssertRC(rc);
2691
2692	AssertMsgReturnVoid(uHostEferMsr == uVmcsEferMsrVmcs,
2693	("EFER Host/VMCS mismatch! host=%#RX64 vmcs=%#RX64\n", uHostEferMsr, uVmcsEferMsrVmcs));
2694	AssertMsgReturnVoid(uHostEferMsr == uHostEferMsrCache,
2695	("EFER Host/Cache mismatch! host=%#RX64 cache=%#RX64\n", uHostEferMsr, uHostEferMsrCache));
2696	}
2697	}
2698
2699
2700	/**
2701	* Verifies whether the guest/host MSR pairs in the auto-load/store area in the
2702	* VMCS are correct.
2703	*
2704	* @param pVCpu The cross context virtual CPU structure.
2705	* @param pVmcsInfo The VMCS info. object.
2706	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2707	*/
2708	static void hmR0VmxCheckAutoLoadStoreMsrs(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
2709	{
2710	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2711
2712	/* Read the various MSR-area counts from the VMCS. */
2713	uint32_t cEntryLoadMsrs;
2714	uint32_t cExitStoreMsrs;
2715	uint32_t cExitLoadMsrs;
2716	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &cEntryLoadMsrs); AssertRC(rc);
2717	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &cExitStoreMsrs); AssertRC(rc);
2718	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &cExitLoadMsrs); AssertRC(rc);
2719
2720	/* Verify all the MSR counts are the same. */
2721	Assert(cEntryLoadMsrs == cExitStoreMsrs);
2722	Assert(cExitStoreMsrs == cExitLoadMsrs);
2723	uint32_t const cMsrs = cExitLoadMsrs;
2724
2725	/* Verify the MSR counts do not exceed the maximum count supported by the hardware. */
2726	Assert(cMsrs < VMX_MISC_MAX_MSRS(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.u64Misc));
2727
2728	/* Verify the MSR counts are within the allocated page size. */
2729	Assert(sizeof(VMXAUTOMSR) * cMsrs <= X86_PAGE_4K_SIZE);
2730
2731	/* Verify the relevant contents of the MSR areas match. */
2732	PCVMXAUTOMSR pGuestMsrLoad = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2733	PCVMXAUTOMSR pGuestMsrStore = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2734	PCVMXAUTOMSR pHostMsrLoad = (PCVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2735	bool const fSeparateExitMsrStorePage = hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo);
2736	for (uint32_t i = 0; i < cMsrs; i++)
2737	{
2738	/* Verify that the MSRs are paired properly and that the host MSR has the correct value. */
2739	if (fSeparateExitMsrStorePage)
2740	{
2741	AssertMsgReturnVoid(pGuestMsrLoad->u32Msr == pGuestMsrStore->u32Msr,
2742	("GuestMsrLoad=%#RX32 GuestMsrStore=%#RX32 cMsrs=%u\n",
2743	pGuestMsrLoad->u32Msr, pGuestMsrStore->u32Msr, cMsrs));
2744	}
2745
2746	AssertMsgReturnVoid(pHostMsrLoad->u32Msr == pGuestMsrLoad->u32Msr,
2747	("HostMsrLoad=%#RX32 GuestMsrLoad=%#RX32 cMsrs=%u\n",
2748	pHostMsrLoad->u32Msr, pGuestMsrLoad->u32Msr, cMsrs));
2749
2750	uint64_t const u64HostMsr = ASMRdMsr(pHostMsrLoad->u32Msr);
2751	AssertMsgReturnVoid(pHostMsrLoad->u64Value == u64HostMsr,
2752	("u32Msr=%#RX32 VMCS Value=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
2753	pHostMsrLoad->u32Msr, pHostMsrLoad->u64Value, u64HostMsr, cMsrs));
2754
2755	/* Verify that cached host EFER MSR matches what's loaded the CPU. */
2756	bool const fIsEferMsr = RT_BOOL(pHostMsrLoad->u32Msr == MSR_K6_EFER);
2757	if (fIsEferMsr)
2758	{
2759	AssertMsgReturnVoid(u64HostMsr == pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostMsrEfer,
2760	("Cached=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
2761	pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostMsrEfer, u64HostMsr, cMsrs));
2762	}
2763
2764	/* Verify that the accesses are as expected in the MSR bitmap for auto-load/store MSRs. */
2765	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
2766	{
2767	uint32_t const fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, pGuestMsrLoad->u32Msr);
2768	if (fIsEferMsr)
2769	{
2770	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_EXIT_RD), ("Passthru read for EFER MSR!?\n"));
2771	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_EXIT_WR), ("Passthru write for EFER MSR!?\n"));
2772	}
2773	else
2774	{
2775	/* Verify LBR MSRs (used only for debugging) are intercepted. We don't passthru these MSRs to the guest yet. */
2776	PCVM pVM = pVCpu->CTX_SUFF(pVM);
2777	if ( pVM->hm.s.vmx.fLbr
2778	&& ( hmR0VmxIsLbrBranchFromMsr(pVM, pGuestMsrLoad->u32Msr, NULL /* pidxMsr */)
2779	\|\| hmR0VmxIsLbrBranchToMsr(pVM, pGuestMsrLoad->u32Msr, NULL /* pidxMsr */)
2780	\|\| pGuestMsrLoad->u32Msr == pVM->hm.s.vmx.idLbrTosMsr))
2781	{
2782	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_MASK) == VMXMSRPM_EXIT_RD_WR,
2783	("u32Msr=%#RX32 cMsrs=%u Passthru read/write for LBR MSRs!\n",
2784	pGuestMsrLoad->u32Msr, cMsrs));
2785	}
2786	else if (!fIsNstGstVmcs)
2787	{
2788	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_MASK) == VMXMSRPM_ALLOW_RD_WR,
2789	("u32Msr=%#RX32 cMsrs=%u No passthru read/write!\n", pGuestMsrLoad->u32Msr, cMsrs));
2790	}
2791	else
2792	{
2793	/*
2794	* A nested-guest VMCS must -also- allow read/write passthrough for the MSR for us to
2795	* execute a nested-guest with MSR passthrough.
2796	*
2797	* Check if the nested-guest MSR bitmap allows passthrough, and if so, assert that we
2798	* allow passthrough too.
2799	*/
2800	void const *pvMsrBitmapNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap);
2801	Assert(pvMsrBitmapNstGst);
2802	uint32_t const fMsrpmNstGst = CPUMGetVmxMsrPermission(pvMsrBitmapNstGst, pGuestMsrLoad->u32Msr);
2803	AssertMsgReturnVoid(fMsrpm == fMsrpmNstGst,
2804	("u32Msr=%#RX32 cMsrs=%u Permission mismatch fMsrpm=%#x fMsrpmNstGst=%#x!\n",
2805	pGuestMsrLoad->u32Msr, cMsrs, fMsrpm, fMsrpmNstGst));
2806	}
2807	}
2808	}
2809
2810	/* Move to the next MSR. */
2811	pHostMsrLoad++;
2812	pGuestMsrLoad++;
2813	pGuestMsrStore++;
2814	}
2815	}
2816	#endif /* VBOX_STRICT */
2817
2818
2819	/**
2820	* Flushes the TLB using EPT.
2821	*
2822	* @returns VBox status code.
2823	* @param pVCpu The cross context virtual CPU structure of the calling
2824	* EMT. Can be NULL depending on @a enmTlbFlush.
2825	* @param pVmcsInfo The VMCS info. object. Can be NULL depending on @a
2826	* enmTlbFlush.
2827	* @param enmTlbFlush Type of flush.
2828	*
2829	* @remarks Caller is responsible for making sure this function is called only
2830	* when NestedPaging is supported and providing @a enmTlbFlush that is
2831	* supported by the CPU.
2832	* @remarks Can be called with interrupts disabled.
2833	*/
2834	static void hmR0VmxFlushEpt(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, VMXTLBFLUSHEPT enmTlbFlush)
2835	{
2836	uint64_t au64Descriptor[2];
2837	if (enmTlbFlush == VMXTLBFLUSHEPT_ALL_CONTEXTS)
2838	au64Descriptor[0] = 0;
2839	else
2840	{
2841	Assert(pVCpu);
2842	Assert(pVmcsInfo);
2843	au64Descriptor[0] = pVmcsInfo->HCPhysEPTP;
2844	}
2845	au64Descriptor[1] = 0; /* MBZ. Intel spec. 33.3 "VMX Instructions" */
2846
2847	int rc = VMXR0InvEPT(enmTlbFlush, &au64Descriptor[0]);
2848	AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvEPT %#x %#RHp failed. rc=%Rrc\n", enmTlbFlush, au64Descriptor[0], rc));
2849
2850	if ( RT_SUCCESS(rc)
2851	&& pVCpu)
2852	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushNestedPaging);
2853	}
2854
2855
2856	/**
2857	* Flushes the TLB using VPID.
2858	*
2859	* @returns VBox status code.
2860	* @param pVCpu The cross context virtual CPU structure of the calling
2861	* EMT. Can be NULL depending on @a enmTlbFlush.
2862	* @param enmTlbFlush Type of flush.
2863	* @param GCPtr Virtual address of the page to flush (can be 0 depending
2864	* on @a enmTlbFlush).
2865	*
2866	* @remarks Can be called with interrupts disabled.
2867	*/
2868	static void hmR0VmxFlushVpid(PVMCPUCC pVCpu, VMXTLBFLUSHVPID enmTlbFlush, RTGCPTR GCPtr)
2869	{
2870	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fVpid);
2871
2872	uint64_t au64Descriptor[2];
2873	if (enmTlbFlush == VMXTLBFLUSHVPID_ALL_CONTEXTS)
2874	{
2875	au64Descriptor[0] = 0;
2876	au64Descriptor[1] = 0;
2877	}
2878	else
2879	{
2880	AssertPtr(pVCpu);
2881	AssertMsg(pVCpu->hm.s.uCurrentAsid != 0, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
2882	AssertMsg(pVCpu->hm.s.uCurrentAsid <= UINT16_MAX, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
2883	au64Descriptor[0] = pVCpu->hm.s.uCurrentAsid;
2884	au64Descriptor[1] = GCPtr;
2885	}
2886
2887	int rc = VMXR0InvVPID(enmTlbFlush, &au64Descriptor[0]);
2888	AssertMsg(rc == VINF_SUCCESS,
2889	("VMXR0InvVPID %#x %u %RGv failed with %Rrc\n", enmTlbFlush, pVCpu ? pVCpu->hm.s.uCurrentAsid : 0, GCPtr, rc));
2890
2891	if ( RT_SUCCESS(rc)
2892	&& pVCpu)
2893	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid);
2894	NOREF(rc);
2895	}
2896
2897
2898	/**
2899	* Invalidates a guest page by guest virtual address. Only relevant for EPT/VPID,
2900	* otherwise there is nothing really to invalidate.
2901	*
2902	* @returns VBox status code.
2903	* @param pVCpu The cross context virtual CPU structure.
2904	* @param GCVirt Guest virtual address of the page to invalidate.
2905	*/
2906	VMMR0DECL(int) VMXR0InvalidatePage(PVMCPUCC pVCpu, RTGCPTR GCVirt)
2907	{
2908	AssertPtr(pVCpu);
2909	LogFlowFunc(("pVCpu=%p GCVirt=%RGv\n", pVCpu, GCVirt));
2910
2911	if (!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TLB_FLUSH))
2912	{
2913	/*
2914	* We must invalidate the guest TLB entry in either case, we cannot ignore it even for
2915	* the EPT case. See @bugref{6043} and @bugref{6177}.
2916	*
2917	* Set the VMCPU_FF_TLB_FLUSH force flag and flush before VM-entry in hmR0VmxFlushTLB*()
2918	* as this function maybe called in a loop with individual addresses.
2919	*/
2920	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2921	if (pVM->hm.s.vmx.fVpid)
2922	{
2923	bool fVpidFlush = RT_BOOL(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR);
2924	if (fVpidFlush)
2925	{
2926	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_INDIV_ADDR, GCVirt);
2927	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt);
2928	}
2929	else
2930	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2931	}
2932	else if (pVM->hm.s.fNestedPaging)
2933	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2934	}
2935
2936	return VINF_SUCCESS;
2937	}
2938
2939
2940	/**
2941	* Dummy placeholder for tagged-TLB flush handling before VM-entry. Used in the
2942	* case where neither EPT nor VPID is supported by the CPU.
2943	*
2944	* @param pHostCpu The HM physical-CPU structure.
2945	* @param pVCpu The cross context virtual CPU structure.
2946	*
2947	* @remarks Called with interrupts disabled.
2948	*/
2949	static void hmR0VmxFlushTaggedTlbNone(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu)
2950	{
2951	AssertPtr(pVCpu);
2952	AssertPtr(pHostCpu);
2953
2954	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
2955
2956	Assert(pHostCpu->idCpu != NIL_RTCPUID);
2957	pVCpu->hm.s.idLastCpu = pHostCpu->idCpu;
2958	pVCpu->hm.s.cTlbFlushes = pHostCpu->cTlbFlushes;
2959	pVCpu->hm.s.fForceTLBFlush = false;
2960	return;
2961	}
2962
2963
2964	/**
2965	* Flushes the tagged-TLB entries for EPT+VPID CPUs as necessary.
2966	*
2967	* @param pHostCpu The HM physical-CPU structure.
2968	* @param pVCpu The cross context virtual CPU structure.
2969	* @param pVmcsInfo The VMCS info. object.
2970	*
2971	* @remarks All references to "ASID" in this function pertains to "VPID" in Intel's
2972	* nomenclature. The reason is, to avoid confusion in compare statements
2973	* since the host-CPU copies are named "ASID".
2974	*
2975	* @remarks Called with interrupts disabled.
2976	*/
2977	static void hmR0VmxFlushTaggedTlbBoth(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
2978	{
2979	#ifdef VBOX_WITH_STATISTICS
2980	bool fTlbFlushed = false;
2981	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { fTlbFlushed = true; } while (0)
2982	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { \
2983	if (!fTlbFlushed) \
2984	STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); \
2985	} while (0)
2986	#else
2987	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { } while (0)
2988	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { } while (0)
2989	#endif
2990
2991	AssertPtr(pVCpu);
2992	AssertPtr(pHostCpu);
2993	Assert(pHostCpu->idCpu != NIL_RTCPUID);
2994
2995	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2996	AssertMsg(pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid,
2997	("hmR0VmxFlushTaggedTlbBoth cannot be invoked unless NestedPaging & VPID are enabled."
2998	"fNestedPaging=%RTbool fVpid=%RTbool", pVM->hm.s.fNestedPaging, pVM->hm.s.vmx.fVpid));
2999
3000	/*
3001	* Force a TLB flush for the first world-switch if the current CPU differs from the one we
3002	* ran on last. If the TLB flush count changed, another VM (VCPU rather) has hit the ASID
3003	* limit while flushing the TLB or the host CPU is online after a suspend/resume, so we
3004	* cannot reuse the current ASID anymore.
3005	*/
3006	if ( pVCpu->hm.s.idLastCpu != pHostCpu->idCpu
3007	\|\| pVCpu->hm.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3008	{
3009	++pHostCpu->uCurrentAsid;
3010	if (pHostCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
3011	{
3012	pHostCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0. */
3013	pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
3014	pHostCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
3015	}
3016
3017	pVCpu->hm.s.uCurrentAsid = pHostCpu->uCurrentAsid;
3018	pVCpu->hm.s.idLastCpu = pHostCpu->idCpu;
3019	pVCpu->hm.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3020
3021	/*
3022	* Flush by EPT when we get rescheduled to a new host CPU to ensure EPT-only tagged mappings are also
3023	* invalidated. We don't need to flush-by-VPID here as flushing by EPT covers it. See @bugref{6568}.
3024	*/
3025	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hm.s.vmx.enmTlbFlushEpt);
3026	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3027	HMVMX_SET_TAGGED_TLB_FLUSHED();
3028	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
3029	}
3030	else if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH)) /* Check for explicit TLB flushes. */
3031	{
3032	/*
3033	* Changes to the EPT paging structure by VMM requires flushing-by-EPT as the CPU
3034	* creates guest-physical (ie. only EPT-tagged) mappings while traversing the EPT
3035	* tables when EPT is in use. Flushing-by-VPID will only flush linear (only
3036	* VPID-tagged) and combined (EPT+VPID tagged) mappings but not guest-physical
3037	* mappings, see @bugref{6568}.
3038	*
3039	* See Intel spec. 28.3.2 "Creating and Using Cached Translation Information".
3040	*/
3041	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hm.s.vmx.enmTlbFlushEpt);
3042	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3043	HMVMX_SET_TAGGED_TLB_FLUSHED();
3044	}
3045	else if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3046	{
3047	/*
3048	* The nested-guest specifies its own guest-physical address to use as the APIC-access
3049	* address which requires flushing the TLB of EPT cached structures.
3050	*
3051	* See Intel spec. 28.3.3.4 "Guidelines for Use of the INVEPT Instruction".
3052	*/
3053	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hm.s.vmx.enmTlbFlushEpt);
3054	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3055	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3056	HMVMX_SET_TAGGED_TLB_FLUSHED();
3057	}
3058
3059
3060	pVCpu->hm.s.fForceTLBFlush = false;
3061	HMVMX_UPDATE_FLUSH_SKIPPED_STAT();
3062
3063	Assert(pVCpu->hm.s.idLastCpu == pHostCpu->idCpu);
3064	Assert(pVCpu->hm.s.cTlbFlushes == pHostCpu->cTlbFlushes);
3065	AssertMsg(pVCpu->hm.s.cTlbFlushes == pHostCpu->cTlbFlushes,
3066	("Flush count mismatch for cpu %d (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pHostCpu->cTlbFlushes));
3067	AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
3068	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pHostCpu->idCpu,
3069	pHostCpu->uCurrentAsid, pHostCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
3070	AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
3071	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pHostCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
3072
3073	/* Update VMCS with the VPID. */
3074	int rc = VMXWriteVmcs16(VMX_VMCS16_VPID, pVCpu->hm.s.uCurrentAsid);
3075	AssertRC(rc);
3076
3077	#undef HMVMX_SET_TAGGED_TLB_FLUSHED
3078	}
3079
3080
3081	/**
3082	* Flushes the tagged-TLB entries for EPT CPUs as necessary.
3083	*
3084	* @param pHostCpu The HM physical-CPU structure.
3085	* @param pVCpu The cross context virtual CPU structure.
3086	* @param pVmcsInfo The VMCS info. object.
3087	*
3088	* @remarks Called with interrupts disabled.
3089	*/
3090	static void hmR0VmxFlushTaggedTlbEpt(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
3091	{
3092	AssertPtr(pVCpu);
3093	AssertPtr(pHostCpu);
3094	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3095	AssertMsg(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked without NestedPaging."));
3096	AssertMsg(!pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fVpid, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with VPID."));
3097
3098	/*
3099	* Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
3100	* A change in the TLB flush count implies the host CPU is online after a suspend/resume.
3101	*/
3102	if ( pVCpu->hm.s.idLastCpu != pHostCpu->idCpu
3103	\|\| pVCpu->hm.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3104	{
3105	pVCpu->hm.s.fForceTLBFlush = true;
3106	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3107	}
3108
3109	/* Check for explicit TLB flushes. */
3110	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
3111	{
3112	pVCpu->hm.s.fForceTLBFlush = true;
3113	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3114	}
3115
3116	/* Check for TLB flushes while switching to/from a nested-guest. */
3117	if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3118	{
3119	pVCpu->hm.s.fForceTLBFlush = true;
3120	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3121	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3122	}
3123
3124	pVCpu->hm.s.idLastCpu = pHostCpu->idCpu;
3125	pVCpu->hm.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3126
3127	if (pVCpu->hm.s.fForceTLBFlush)
3128	{
3129	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVCpu->CTX_SUFF(pVM)->hm.s.vmx.enmTlbFlushEpt);
3130	pVCpu->hm.s.fForceTLBFlush = false;
3131	}
3132	}
3133
3134
3135	/**
3136	* Flushes the tagged-TLB entries for VPID CPUs as necessary.
3137	*
3138	* @param pHostCpu The HM physical-CPU structure.
3139	* @param pVCpu The cross context virtual CPU structure.
3140	*
3141	* @remarks Called with interrupts disabled.
3142	*/
3143	static void hmR0VmxFlushTaggedTlbVpid(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu)
3144	{
3145	AssertPtr(pVCpu);
3146	AssertPtr(pHostCpu);
3147	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3148	AssertMsg(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fVpid, ("hmR0VmxFlushTlbVpid cannot be invoked without VPID."));
3149	AssertMsg(!pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging, ("hmR0VmxFlushTlbVpid cannot be invoked with NestedPaging"));
3150
3151	/*
3152	* Force a TLB flush for the first world switch if the current CPU differs from the one we
3153	* ran on last. If the TLB flush count changed, another VM (VCPU rather) has hit the ASID
3154	* limit while flushing the TLB or the host CPU is online after a suspend/resume, so we
3155	* cannot reuse the current ASID anymore.
3156	*/
3157	if ( pVCpu->hm.s.idLastCpu != pHostCpu->idCpu
3158	\|\| pVCpu->hm.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3159	{
3160	pVCpu->hm.s.fForceTLBFlush = true;
3161	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3162	}
3163
3164	/* Check for explicit TLB flushes. */
3165	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
3166	{
3167	/*
3168	* If we ever support VPID flush combinations other than ALL or SINGLE-context (see
3169	* hmR0VmxSetupTaggedTlb()) we would need to explicitly flush in this case (add an
3170	* fExplicitFlush = true here and change the pHostCpu->fFlushAsidBeforeUse check below to
3171	* include fExplicitFlush's too) - an obscure corner case.
3172	*/
3173	pVCpu->hm.s.fForceTLBFlush = true;
3174	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3175	}
3176
3177	/* Check for TLB flushes while switching to/from a nested-guest. */
3178	if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3179	{
3180	pVCpu->hm.s.fForceTLBFlush = true;
3181	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3182	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3183	}
3184
3185	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3186	pVCpu->hm.s.idLastCpu = pHostCpu->idCpu;
3187	if (pVCpu->hm.s.fForceTLBFlush)
3188	{
3189	++pHostCpu->uCurrentAsid;
3190	if (pHostCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
3191	{
3192	pHostCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0 */
3193	pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
3194	pHostCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
3195	}
3196
3197	pVCpu->hm.s.fForceTLBFlush = false;
3198	pVCpu->hm.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3199	pVCpu->hm.s.uCurrentAsid = pHostCpu->uCurrentAsid;
3200	if (pHostCpu->fFlushAsidBeforeUse)
3201	{
3202	if (pVM->hm.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_SINGLE_CONTEXT)
3203	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_SINGLE_CONTEXT, 0 /* GCPtr */);
3204	else if (pVM->hm.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_ALL_CONTEXTS)
3205	{
3206	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_ALL_CONTEXTS, 0 /* GCPtr */);
3207	pHostCpu->fFlushAsidBeforeUse = false;
3208	}
3209	else
3210	{
3211	/* hmR0VmxSetupTaggedTlb() ensures we never get here. Paranoia. */
3212	AssertMsgFailed(("Unsupported VPID-flush context type.\n"));
3213	}
3214	}
3215	}
3216
3217	AssertMsg(pVCpu->hm.s.cTlbFlushes == pHostCpu->cTlbFlushes,
3218	("Flush count mismatch for cpu %d (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pHostCpu->cTlbFlushes));
3219	AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
3220	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pHostCpu->idCpu,
3221	pHostCpu->uCurrentAsid, pHostCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
3222	AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
3223	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pHostCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
3224
3225	int rc = VMXWriteVmcs16(VMX_VMCS16_VPID, pVCpu->hm.s.uCurrentAsid);
3226	AssertRC(rc);
3227	}
3228
3229
3230	/**
3231	* Flushes the guest TLB entry based on CPU capabilities.
3232	*
3233	* @param pHostCpu The HM physical-CPU structure.
3234	* @param pVCpu The cross context virtual CPU structure.
3235	* @param pVmcsInfo The VMCS info. object.
3236	*
3237	* @remarks Called with interrupts disabled.
3238	*/
3239	static void hmR0VmxFlushTaggedTlb(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3240	{
3241	#ifdef HMVMX_ALWAYS_FLUSH_TLB
3242	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
3243	#endif
3244	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3245	switch (pVM->hm.s.vmx.enmTlbFlushType)
3246	{
3247	case VMXTLBFLUSHTYPE_EPT_VPID: hmR0VmxFlushTaggedTlbBoth(pHostCpu, pVCpu, pVmcsInfo); break;
3248	case VMXTLBFLUSHTYPE_EPT: hmR0VmxFlushTaggedTlbEpt(pHostCpu, pVCpu, pVmcsInfo); break;
3249	case VMXTLBFLUSHTYPE_VPID: hmR0VmxFlushTaggedTlbVpid(pHostCpu, pVCpu); break;
3250	case VMXTLBFLUSHTYPE_NONE: hmR0VmxFlushTaggedTlbNone(pHostCpu, pVCpu); break;
3251	default:
3252	AssertMsgFailed(("Invalid flush-tag function identifier\n"));
3253	break;
3254	}
3255	/* Don't assert that VMCPU_FF_TLB_FLUSH should no longer be pending. It can be set by other EMTs. */
3256	}
3257
3258
3259	/**
3260	* Sets up the appropriate tagged TLB-flush level and handler for flushing guest
3261	* TLB entries from the host TLB before VM-entry.
3262	*
3263	* @returns VBox status code.
3264	* @param pVM The cross context VM structure.
3265	*/
3266	static int hmR0VmxSetupTaggedTlb(PVMCC pVM)
3267	{
3268	/*
3269	* Determine optimal flush type for nested paging.
3270	* We cannot ignore EPT if no suitable flush-types is supported by the CPU as we've already setup
3271	* unrestricted guest execution (see hmR3InitFinalizeR0()).
3272	*/
3273	if (pVM->hm.s.fNestedPaging)
3274	{
3275	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT)
3276	{
3277	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT)
3278	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_SINGLE_CONTEXT;
3279	else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
3280	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_ALL_CONTEXTS;
3281	else
3282	{
3283	/* Shouldn't happen. EPT is supported but no suitable flush-types supported. */
3284	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3285	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_FLUSH_TYPE_UNSUPPORTED;
3286	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3287	}
3288
3289	/* Make sure the write-back cacheable memory type for EPT is supported. */
3290	if (RT_UNLIKELY(!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EMT_WB)))
3291	{
3292	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3293	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_MEM_TYPE_NOT_WB;
3294	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3295	}
3296
3297	/* EPT requires a page-walk length of 4. */
3298	if (RT_UNLIKELY(!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_PAGE_WALK_LENGTH_4)))
3299	{
3300	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3301	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_PAGE_WALK_LENGTH_UNSUPPORTED;
3302	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3303	}
3304	}
3305	else
3306	{
3307	/* Shouldn't happen. EPT is supported but INVEPT instruction is not supported. */
3308	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3309	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_INVEPT_UNAVAILABLE;
3310	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3311	}
3312	}
3313
3314	/*
3315	* Determine optimal flush type for VPID.
3316	*/
3317	if (pVM->hm.s.vmx.fVpid)
3318	{
3319	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID)
3320	{
3321	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT)
3322	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_SINGLE_CONTEXT;
3323	else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS)
3324	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_ALL_CONTEXTS;
3325	else
3326	{
3327	/* Neither SINGLE nor ALL-context flush types for VPID is supported by the CPU. Ignore VPID capability. */
3328	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
3329	LogRelFunc(("Only INDIV_ADDR supported. Ignoring VPID.\n"));
3330	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
3331	LogRelFunc(("Only SINGLE_CONTEXT_RETAIN_GLOBALS supported. Ignoring VPID.\n"));
3332	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NOT_SUPPORTED;
3333	pVM->hm.s.vmx.fVpid = false;
3334	}
3335	}
3336	else
3337	{
3338	/* Shouldn't happen. VPID is supported but INVVPID is not supported by the CPU. Ignore VPID capability. */
3339	Log4Func(("VPID supported without INVEPT support. Ignoring VPID.\n"));
3340	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NOT_SUPPORTED;
3341	pVM->hm.s.vmx.fVpid = false;
3342	}
3343	}
3344
3345	/*
3346	* Setup the handler for flushing tagged-TLBs.
3347	*/
3348	if (pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid)
3349	pVM->hm.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_EPT_VPID;
3350	else if (pVM->hm.s.fNestedPaging)
3351	pVM->hm.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_EPT;
3352	else if (pVM->hm.s.vmx.fVpid)
3353	pVM->hm.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_VPID;
3354	else
3355	pVM->hm.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_NONE;
3356	return VINF_SUCCESS;
3357	}
3358
3359
3360	/**
3361	* Sets up the LBR MSR ranges based on the host CPU.
3362	*
3363	* @returns VBox status code.
3364	* @param pVM The cross context VM structure.
3365	*/
3366	static int hmR0VmxSetupLbrMsrRange(PVMCC pVM)
3367	{
3368	Assert(pVM->hm.s.vmx.fLbr);
3369	uint32_t idLbrFromIpMsrFirst;
3370	uint32_t idLbrFromIpMsrLast;
3371	uint32_t idLbrToIpMsrFirst;
3372	uint32_t idLbrToIpMsrLast;
3373	uint32_t idLbrTosMsr;
3374
3375	/*
3376	* Determine the LBR MSRs supported for this host CPU family and model.
3377	*
3378	* See Intel spec. 17.4.8 "LBR Stack".
3379	* See Intel "Model-Specific Registers" spec.
3380	*/
3381	uint32_t const uFamilyModel = (pVM->cpum.ro.HostFeatures.uFamily << 8)
3382	\| pVM->cpum.ro.HostFeatures.uModel;
3383	switch (uFamilyModel)
3384	{
3385	case 0x0f01: case 0x0f02:
3386	idLbrFromIpMsrFirst = MSR_P4_LASTBRANCH_0;
3387	idLbrFromIpMsrLast = MSR_P4_LASTBRANCH_3;
3388	idLbrToIpMsrFirst = 0x0;
3389	idLbrToIpMsrLast = 0x0;
3390	idLbrTosMsr = MSR_P4_LASTBRANCH_TOS;
3391	break;
3392
3393	case 0x065c: case 0x065f: case 0x064e: case 0x065e: case 0x068e:
3394	case 0x069e: case 0x0655: case 0x0666: case 0x067a: case 0x0667:
3395	case 0x066a: case 0x066c: case 0x067d: case 0x067e:
3396	idLbrFromIpMsrFirst = MSR_LASTBRANCH_0_FROM_IP;
3397	idLbrFromIpMsrLast = MSR_LASTBRANCH_31_FROM_IP;
3398	idLbrToIpMsrFirst = MSR_LASTBRANCH_0_TO_IP;
3399	idLbrToIpMsrLast = MSR_LASTBRANCH_31_TO_IP;
3400	idLbrTosMsr = MSR_LASTBRANCH_TOS;
3401	break;
3402
3403	case 0x063d: case 0x0647: case 0x064f: case 0x0656: case 0x063c:
3404	case 0x0645: case 0x0646: case 0x063f: case 0x062a: case 0x062d:
3405	case 0x063a: case 0x063e: case 0x061a: case 0x061e: case 0x061f:
3406	case 0x062e: case 0x0625: case 0x062c: case 0x062f:
3407	idLbrFromIpMsrFirst = MSR_LASTBRANCH_0_FROM_IP;
3408	idLbrFromIpMsrLast = MSR_LASTBRANCH_15_FROM_IP;
3409	idLbrToIpMsrFirst = MSR_LASTBRANCH_0_TO_IP;
3410	idLbrToIpMsrLast = MSR_LASTBRANCH_15_TO_IP;
3411	idLbrTosMsr = MSR_LASTBRANCH_TOS;
3412	break;
3413
3414	case 0x0617: case 0x061d: case 0x060f:
3415	idLbrFromIpMsrFirst = MSR_CORE2_LASTBRANCH_0_FROM_IP;
3416	idLbrFromIpMsrLast = MSR_CORE2_LASTBRANCH_3_FROM_IP;
3417	idLbrToIpMsrFirst = MSR_CORE2_LASTBRANCH_0_TO_IP;
3418	idLbrToIpMsrLast = MSR_CORE2_LASTBRANCH_3_TO_IP;
3419	idLbrTosMsr = MSR_CORE2_LASTBRANCH_TOS;
3420	break;
3421
3422	/* Atom and related microarchitectures we don't care about:
3423	case 0x0637: case 0x064a: case 0x064c: case 0x064d: case 0x065a:
3424	case 0x065d: case 0x061c: case 0x0626: case 0x0627: case 0x0635:
3425	case 0x0636: */
3426	/* All other CPUs: */
3427	default:
3428	{
3429	LogRelFunc(("Could not determine LBR stack size for the CPU model %#x\n", uFamilyModel));
3430	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_LBR_STACK_SIZE_UNKNOWN;
3431	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3432	}
3433	}
3434
3435	/*
3436	* Validate.
3437	*/
3438	uint32_t const cLbrStack = idLbrFromIpMsrLast - idLbrFromIpMsrFirst + 1;
3439	PCVMCPU pVCpu0 = VMCC_GET_CPU_0(pVM);
3440	AssertCompile( RT_ELEMENTS(pVCpu0->hm.s.vmx.VmcsInfo.au64LbrFromIpMsr)
3441	== RT_ELEMENTS(pVCpu0->hm.s.vmx.VmcsInfo.au64LbrToIpMsr));
3442	if (cLbrStack > RT_ELEMENTS(pVCpu0->hm.s.vmx.VmcsInfo.au64LbrFromIpMsr))
3443	{
3444	LogRelFunc(("LBR stack size of the CPU (%u) exceeds our buffer size\n", cLbrStack));
3445	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_LBR_STACK_SIZE_OVERFLOW;
3446	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3447	}
3448	NOREF(pVCpu0);
3449
3450	/*
3451	* Update the LBR info. to the VM struct. for use later.
3452	*/
3453	pVM->hm.s.vmx.idLbrTosMsr = idLbrTosMsr;
3454	pVM->hm.s.vmx.idLbrFromIpMsrFirst = idLbrFromIpMsrFirst;
3455	pVM->hm.s.vmx.idLbrFromIpMsrLast = idLbrFromIpMsrLast;
3456
3457	pVM->hm.s.vmx.idLbrToIpMsrFirst = idLbrToIpMsrFirst;
3458	pVM->hm.s.vmx.idLbrToIpMsrLast = idLbrToIpMsrLast;
3459	return VINF_SUCCESS;
3460	}
3461
3462
3463	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3464	/**
3465	* Sets up the shadow VMCS fields arrays.
3466	*
3467	* This function builds arrays of VMCS fields to sync the shadow VMCS later while
3468	* executing the guest.
3469	*
3470	* @returns VBox status code.
3471	* @param pVM The cross context VM structure.
3472	*/
3473	static int hmR0VmxSetupShadowVmcsFieldsArrays(PVMCC pVM)
3474	{
3475	/*
3476	* Paranoia. Ensure we haven't exposed the VMWRITE-All VMX feature to the guest
3477	* when the host does not support it.
3478	*/
3479	bool const fGstVmwriteAll = pVM->cpum.ro.GuestFeatures.fVmxVmwriteAll;
3480	if ( !fGstVmwriteAll
3481	\|\| (pVM->hm.s.vmx.Msrs.u64Misc & VMX_MISC_VMWRITE_ALL))
3482	{ /* likely. */ }
3483	else
3484	{
3485	LogRelFunc(("VMX VMWRITE-All feature exposed to the guest but host CPU does not support it!\n"));
3486	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_GST_HOST_VMWRITE_ALL;
3487	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3488	}
3489
3490	uint32_t const cVmcsFields = RT_ELEMENTS(g_aVmcsFields);
3491	uint32_t cRwFields = 0;
3492	uint32_t cRoFields = 0;
3493	for (uint32_t i = 0; i < cVmcsFields; i++)
3494	{
3495	VMXVMCSFIELD VmcsField;
3496	VmcsField.u = g_aVmcsFields[i];
3497
3498	/*
3499	* We will be writing "FULL" (64-bit) fields while syncing the shadow VMCS.
3500	* Therefore, "HIGH" (32-bit portion of 64-bit) fields must not be included
3501	* in the shadow VMCS fields array as they would be redundant.
3502	*
3503	* If the VMCS field depends on a CPU feature that is not exposed to the guest,
3504	* we must not include it in the shadow VMCS fields array. Guests attempting to
3505	* VMREAD/VMWRITE such VMCS fields would cause a VM-exit and we shall emulate
3506	* the required behavior.
3507	*/
3508	if ( VmcsField.n.fAccessType == VMX_VMCSFIELD_ACCESS_FULL
3509	&& CPUMIsGuestVmxVmcsFieldValid(pVM, VmcsField.u))
3510	{
3511	/*
3512	* Read-only fields are placed in a separate array so that while syncing shadow
3513	* VMCS fields later (which is more performance critical) we can avoid branches.
3514	*
3515	* However, if the guest can write to all fields (including read-only fields),
3516	* we treat it a as read/write field. Otherwise, writing to these fields would
3517	* cause a VMWRITE instruction error while syncing the shadow VMCS.
3518	*/
3519	if ( fGstVmwriteAll
3520	\|\| !VMXIsVmcsFieldReadOnly(VmcsField.u))
3521	pVM->hm.s.vmx.paShadowVmcsFields[cRwFields++] = VmcsField.u;
3522	else
3523	pVM->hm.s.vmx.paShadowVmcsRoFields[cRoFields++] = VmcsField.u;
3524	}
3525	}
3526
3527	/* Update the counts. */
3528	pVM->hm.s.vmx.cShadowVmcsFields = cRwFields;
3529	pVM->hm.s.vmx.cShadowVmcsRoFields = cRoFields;
3530	return VINF_SUCCESS;
3531	}
3532
3533
3534	/**
3535	* Sets up the VMREAD and VMWRITE bitmaps.
3536	*
3537	* @param pVM The cross context VM structure.
3538	*/
3539	static void hmR0VmxSetupVmreadVmwriteBitmaps(PVMCC pVM)
3540	{
3541	/*
3542	* By default, ensure guest attempts to access any VMCS fields cause VM-exits.
3543	*/
3544	uint32_t const cbBitmap = X86_PAGE_4K_SIZE;
3545	uint8_t pbVmreadBitmap = (uint8_t )pVM->hm.s.vmx.pvVmreadBitmap;
3546	uint8_t pbVmwriteBitmap = (uint8_t )pVM->hm.s.vmx.pvVmwriteBitmap;
3547	ASMMemFill32(pbVmreadBitmap, cbBitmap, UINT32_C(0xffffffff));
3548	ASMMemFill32(pbVmwriteBitmap, cbBitmap, UINT32_C(0xffffffff));
3549
3550	/*
3551	* Skip intercepting VMREAD/VMWRITE to guest read/write fields in the
3552	* VMREAD and VMWRITE bitmaps.
3553	*/
3554	{
3555	uint32_t const *paShadowVmcsFields = pVM->hm.s.vmx.paShadowVmcsFields;
3556	uint32_t const cShadowVmcsFields = pVM->hm.s.vmx.cShadowVmcsFields;
3557	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
3558	{
3559	uint32_t const uVmcsField = paShadowVmcsFields[i];
3560	Assert(!(uVmcsField & VMX_VMCSFIELD_RSVD_MASK));
3561	Assert(uVmcsField >> 3 < cbBitmap);
3562	ASMBitClear(pbVmreadBitmap + (uVmcsField >> 3), uVmcsField & 7);
3563	ASMBitClear(pbVmwriteBitmap + (uVmcsField >> 3), uVmcsField & 7);
3564	}
3565	}
3566
3567	/*
3568	* Skip intercepting VMREAD for guest read-only fields in the VMREAD bitmap
3569	* if the host supports VMWRITE to all supported VMCS fields.
3570	*/
3571	if (pVM->hm.s.vmx.Msrs.u64Misc & VMX_MISC_VMWRITE_ALL)
3572	{
3573	uint32_t const *paShadowVmcsRoFields = pVM->hm.s.vmx.paShadowVmcsRoFields;
3574	uint32_t const cShadowVmcsRoFields = pVM->hm.s.vmx.cShadowVmcsRoFields;
3575	for (uint32_t i = 0; i < cShadowVmcsRoFields; i++)
3576	{
3577	uint32_t const uVmcsField = paShadowVmcsRoFields[i];
3578	Assert(!(uVmcsField & VMX_VMCSFIELD_RSVD_MASK));
3579	Assert(uVmcsField >> 3 < cbBitmap);
3580	ASMBitClear(pbVmreadBitmap + (uVmcsField >> 3), uVmcsField & 7);
3581	}
3582	}
3583	}
3584	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
3585
3586
3587	/**
3588	* Sets up the virtual-APIC page address for the VMCS.
3589	*
3590	* @param pVmcsInfo The VMCS info. object.
3591	*/
3592	DECLINLINE(void) hmR0VmxSetupVmcsVirtApicAddr(PCVMXVMCSINFO pVmcsInfo)
3593	{
3594	RTHCPHYS const HCPhysVirtApic = pVmcsInfo->HCPhysVirtApic;
3595	Assert(HCPhysVirtApic != NIL_RTHCPHYS);
3596	Assert(!(HCPhysVirtApic & 0xfff)); /* Bits 11:0 MBZ. */
3597	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL, HCPhysVirtApic);
3598	AssertRC(rc);
3599	}
3600
3601
3602	/**
3603	* Sets up the MSR-bitmap address for the VMCS.
3604	*
3605	* @param pVmcsInfo The VMCS info. object.
3606	*/
3607	DECLINLINE(void) hmR0VmxSetupVmcsMsrBitmapAddr(PCVMXVMCSINFO pVmcsInfo)
3608	{
3609	RTHCPHYS const HCPhysMsrBitmap = pVmcsInfo->HCPhysMsrBitmap;
3610	Assert(HCPhysMsrBitmap != NIL_RTHCPHYS);
3611	Assert(!(HCPhysMsrBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3612	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_MSR_BITMAP_FULL, HCPhysMsrBitmap);
3613	AssertRC(rc);
3614	}
3615
3616
3617	/**
3618	* Sets up the APIC-access page address for the VMCS.
3619	*
3620	* @param pVCpu The cross context virtual CPU structure.
3621	*/
3622	DECLINLINE(void) hmR0VmxSetupVmcsApicAccessAddr(PVMCPUCC pVCpu)
3623	{
3624	RTHCPHYS const HCPhysApicAccess = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.HCPhysApicAccess;
3625	Assert(HCPhysApicAccess != NIL_RTHCPHYS);
3626	Assert(!(HCPhysApicAccess & 0xfff)); /* Bits 11:0 MBZ. */
3627	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, HCPhysApicAccess);
3628	AssertRC(rc);
3629	}
3630
3631
3632	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3633	/**
3634	* Sets up the VMREAD bitmap address for the VMCS.
3635	*
3636	* @param pVCpu The cross context virtual CPU structure.
3637	*/
3638	DECLINLINE(void) hmR0VmxSetupVmcsVmreadBitmapAddr(PVMCPUCC pVCpu)
3639	{
3640	RTHCPHYS const HCPhysVmreadBitmap = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.HCPhysVmreadBitmap;
3641	Assert(HCPhysVmreadBitmap != NIL_RTHCPHYS);
3642	Assert(!(HCPhysVmreadBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3643	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VMREAD_BITMAP_FULL, HCPhysVmreadBitmap);
3644	AssertRC(rc);
3645	}
3646
3647
3648	/**
3649	* Sets up the VMWRITE bitmap address for the VMCS.
3650	*
3651	* @param pVCpu The cross context virtual CPU structure.
3652	*/
3653	DECLINLINE(void) hmR0VmxSetupVmcsVmwriteBitmapAddr(PVMCPUCC pVCpu)
3654	{
3655	RTHCPHYS const HCPhysVmwriteBitmap = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.HCPhysVmwriteBitmap;
3656	Assert(HCPhysVmwriteBitmap != NIL_RTHCPHYS);
3657	Assert(!(HCPhysVmwriteBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3658	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VMWRITE_BITMAP_FULL, HCPhysVmwriteBitmap);
3659	AssertRC(rc);
3660	}
3661	#endif
3662
3663
3664	/**
3665	* Sets up the VM-entry MSR load, VM-exit MSR-store and VM-exit MSR-load addresses
3666	* in the VMCS.
3667	*
3668	* @returns VBox status code.
3669	* @param pVmcsInfo The VMCS info. object.
3670	*/
3671	DECLINLINE(int) hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(PVMXVMCSINFO pVmcsInfo)
3672	{
3673	RTHCPHYS const HCPhysGuestMsrLoad = pVmcsInfo->HCPhysGuestMsrLoad;
3674	Assert(HCPhysGuestMsrLoad != NIL_RTHCPHYS);
3675	Assert(!(HCPhysGuestMsrLoad & 0xf)); /* Bits 3:0 MBZ. */
3676
3677	RTHCPHYS const HCPhysGuestMsrStore = pVmcsInfo->HCPhysGuestMsrStore;
3678	Assert(HCPhysGuestMsrStore != NIL_RTHCPHYS);
3679	Assert(!(HCPhysGuestMsrStore & 0xf)); /* Bits 3:0 MBZ. */
3680
3681	RTHCPHYS const HCPhysHostMsrLoad = pVmcsInfo->HCPhysHostMsrLoad;
3682	Assert(HCPhysHostMsrLoad != NIL_RTHCPHYS);
3683	Assert(!(HCPhysHostMsrLoad & 0xf)); /* Bits 3:0 MBZ. */
3684
3685	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL, HCPhysGuestMsrLoad); AssertRC(rc);
3686	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL, HCPhysGuestMsrStore); AssertRC(rc);
3687	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL, HCPhysHostMsrLoad); AssertRC(rc);
3688	return VINF_SUCCESS;
3689	}
3690
3691
3692	/**
3693	* Sets up MSR permissions in the MSR bitmap of a VMCS info. object.
3694	*
3695	* @param pVCpu The cross context virtual CPU structure.
3696	* @param pVmcsInfo The VMCS info. object.
3697	*/
3698	static void hmR0VmxSetupVmcsMsrPermissions(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3699	{
3700	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS);
3701
3702	/*
3703	* By default, ensure guest attempts to access any MSR cause VM-exits.
3704	* This shall later be relaxed for specific MSRs as necessary.
3705	*
3706	* Note: For nested-guests, the entire bitmap will be merged prior to
3707	* executing the nested-guest using hardware-assisted VMX and hence there
3708	* is no need to perform this operation. See hmR0VmxMergeMsrBitmapNested.
3709	*/
3710	Assert(pVmcsInfo->pvMsrBitmap);
3711	ASMMemFill32(pVmcsInfo->pvMsrBitmap, X86_PAGE_4K_SIZE, UINT32_C(0xffffffff));
3712
3713	/*
3714	* The guest can access the following MSRs (read, write) without causing
3715	* VM-exits; they are loaded/stored automatically using fields in the VMCS.
3716	*/
3717	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3718	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_CS, VMXMSRPM_ALLOW_RD_WR);
3719	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_ESP, VMXMSRPM_ALLOW_RD_WR);
3720	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_EIP, VMXMSRPM_ALLOW_RD_WR);
3721	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_GS_BASE, VMXMSRPM_ALLOW_RD_WR);
3722	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_FS_BASE, VMXMSRPM_ALLOW_RD_WR);
3723
3724	/*
3725	* The IA32_PRED_CMD and IA32_FLUSH_CMD MSRs are write-only and has no state
3726	* associated with then. We never need to intercept access (writes need to be
3727	* executed without causing a VM-exit, reads will #GP fault anyway).
3728	*
3729	* The IA32_SPEC_CTRL MSR is read/write and has state. We allow the guest to
3730	* read/write them. We swap the the guest/host MSR value using the
3731	* auto-load/store MSR area.
3732	*/
3733	if (pVM->cpum.ro.GuestFeatures.fIbpb)
3734	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_PRED_CMD, VMXMSRPM_ALLOW_RD_WR);
3735	if (pVM->cpum.ro.GuestFeatures.fFlushCmd)
3736	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_FLUSH_CMD, VMXMSRPM_ALLOW_RD_WR);
3737	if (pVM->cpum.ro.GuestFeatures.fIbrs)
3738	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SPEC_CTRL, VMXMSRPM_ALLOW_RD_WR);
3739
3740	/*
3741	* Allow full read/write access for the following MSRs (mandatory for VT-x)
3742	* required for 64-bit guests.
3743	*/
3744	if (pVM->hm.s.fAllow64BitGuests)
3745	{
3746	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_LSTAR, VMXMSRPM_ALLOW_RD_WR);
3747	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K6_STAR, VMXMSRPM_ALLOW_RD_WR);
3748	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_SF_MASK, VMXMSRPM_ALLOW_RD_WR);
3749	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_KERNEL_GS_BASE, VMXMSRPM_ALLOW_RD_WR);
3750	}
3751
3752	/*
3753	* IA32_EFER MSR is always intercepted, see @bugref{9180#c37}.
3754	*/
3755	#ifdef VBOX_STRICT
3756	Assert(pVmcsInfo->pvMsrBitmap);
3757	uint32_t const fMsrpmEfer = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, MSR_K6_EFER);
3758	Assert(fMsrpmEfer == VMXMSRPM_EXIT_RD_WR);
3759	#endif
3760	}
3761
3762
3763	/**
3764	* Sets up pin-based VM-execution controls in the VMCS.
3765	*
3766	* @returns VBox status code.
3767	* @param pVCpu The cross context virtual CPU structure.
3768	* @param pVmcsInfo The VMCS info. object.
3769	*/
3770	static int hmR0VmxSetupVmcsPinCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3771	{
3772	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3773	uint32_t fVal = pVM->hm.s.vmx.Msrs.PinCtls.n.allowed0; /* Bits set here must always be set. */
3774	uint32_t const fZap = pVM->hm.s.vmx.Msrs.PinCtls.n.allowed1; /* Bits cleared here must always be cleared. */
3775
3776	fVal \|= VMX_PIN_CTLS_EXT_INT_EXIT /* External interrupts cause a VM-exit. */
3777	\| VMX_PIN_CTLS_NMI_EXIT; /* Non-maskable interrupts (NMIs) cause a VM-exit. */
3778
3779	if (pVM->hm.s.vmx.Msrs.PinCtls.n.allowed1 & VMX_PIN_CTLS_VIRT_NMI)
3780	fVal \|= VMX_PIN_CTLS_VIRT_NMI; /* Use virtual NMIs and virtual-NMI blocking features. */
3781
3782	/* Enable the VMX-preemption timer. */
3783	if (pVM->hm.s.vmx.fUsePreemptTimer)
3784	{
3785	Assert(pVM->hm.s.vmx.Msrs.PinCtls.n.allowed1 & VMX_PIN_CTLS_PREEMPT_TIMER);
3786	fVal \|= VMX_PIN_CTLS_PREEMPT_TIMER;
3787	}
3788
3789	#if 0
3790	/* Enable posted-interrupt processing. */
3791	if (pVM->hm.s.fPostedIntrs)
3792	{
3793	Assert(pVM->hm.s.vmx.Msrs.PinCtls.n.allowed1 & VMX_PIN_CTLS_POSTED_INT);
3794	Assert(pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_ACK_EXT_INT);
3795	fVal \|= VMX_PIN_CTLS_POSTED_INT;
3796	}
3797	#endif
3798
3799	if ((fVal & fZap) != fVal)
3800	{
3801	LogRelFunc(("Invalid pin-based VM-execution controls combo! Cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3802	pVM->hm.s.vmx.Msrs.PinCtls.n.allowed0, fVal, fZap));
3803	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PIN_EXEC;
3804	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3805	}
3806
3807	/* Commit it to the VMCS and update our cache. */
3808	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, fVal);
3809	AssertRC(rc);
3810	pVmcsInfo->u32PinCtls = fVal;
3811
3812	return VINF_SUCCESS;
3813	}
3814
3815
3816	/**
3817	* Sets up secondary processor-based VM-execution controls in the VMCS.
3818	*
3819	* @returns VBox status code.
3820	* @param pVCpu The cross context virtual CPU structure.
3821	* @param pVmcsInfo The VMCS info. object.
3822	*/
3823	static int hmR0VmxSetupVmcsProcCtls2(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3824	{
3825	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3826	uint32_t fVal = pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed0; /* Bits set here must be set in the VMCS. */
3827	uint32_t const fZap = pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3828
3829	/* WBINVD causes a VM-exit. */
3830	if (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_WBINVD_EXIT)
3831	fVal \|= VMX_PROC_CTLS2_WBINVD_EXIT;
3832
3833	/* Enable EPT (aka nested-paging). */
3834	if (pVM->hm.s.fNestedPaging)
3835	fVal \|= VMX_PROC_CTLS2_EPT;
3836
3837	/* Enable the INVPCID instruction if we expose it to the guest and is supported
3838	by the hardware. Without this, guest executing INVPCID would cause a #UD. */
3839	if ( pVM->cpum.ro.GuestFeatures.fInvpcid
3840	&& (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_INVPCID))
3841	fVal \|= VMX_PROC_CTLS2_INVPCID;
3842
3843	/* Enable VPID. */
3844	if (pVM->hm.s.vmx.fVpid)
3845	fVal \|= VMX_PROC_CTLS2_VPID;
3846
3847	/* Enable unrestricted guest execution. */
3848	if (pVM->hm.s.vmx.fUnrestrictedGuest)
3849	fVal \|= VMX_PROC_CTLS2_UNRESTRICTED_GUEST;
3850
3851	#if 0
3852	if (pVM->hm.s.fVirtApicRegs)
3853	{
3854	/* Enable APIC-register virtualization. */
3855	Assert(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_APIC_REG_VIRT);
3856	fVal \|= VMX_PROC_CTLS2_APIC_REG_VIRT;
3857
3858	/* Enable virtual-interrupt delivery. */
3859	Assert(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_INTR_DELIVERY);
3860	fVal \|= VMX_PROC_CTLS2_VIRT_INTR_DELIVERY;
3861	}
3862	#endif
3863
3864	/* Virtualize-APIC accesses if supported by the CPU. The virtual-APIC page is
3865	where the TPR shadow resides. */
3866	/** @todo VIRT_X2APIC support, it's mutually exclusive with this. So must be
3867	* done dynamically. */
3868	if (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
3869	{
3870	fVal \|= VMX_PROC_CTLS2_VIRT_APIC_ACCESS;
3871	hmR0VmxSetupVmcsApicAccessAddr(pVCpu);
3872	}
3873
3874	/* Enable the RDTSCP instruction if we expose it to the guest and is supported
3875	by the hardware. Without this, guest executing RDTSCP would cause a #UD. */
3876	if ( pVM->cpum.ro.GuestFeatures.fRdTscP
3877	&& (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_RDTSCP))
3878	fVal \|= VMX_PROC_CTLS2_RDTSCP;
3879
3880	/* Enable Pause-Loop exiting. */
3881	if ( (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT)
3882	&& pVM->hm.s.vmx.cPleGapTicks
3883	&& pVM->hm.s.vmx.cPleWindowTicks)
3884	{
3885	fVal \|= VMX_PROC_CTLS2_PAUSE_LOOP_EXIT;
3886
3887	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, pVM->hm.s.vmx.cPleGapTicks); AssertRC(rc);
3888	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, pVM->hm.s.vmx.cPleWindowTicks); AssertRC(rc);
3889	}
3890
3891	if ((fVal & fZap) != fVal)
3892	{
3893	LogRelFunc(("Invalid secondary processor-based VM-execution controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3894	pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed0, fVal, fZap));
3895	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC2;
3896	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3897	}
3898
3899	/* Commit it to the VMCS and update our cache. */
3900	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, fVal);
3901	AssertRC(rc);
3902	pVmcsInfo->u32ProcCtls2 = fVal;
3903
3904	return VINF_SUCCESS;
3905	}
3906
3907
3908	/**
3909	* Sets up processor-based VM-execution controls in the VMCS.
3910	*
3911	* @returns VBox status code.
3912	* @param pVCpu The cross context virtual CPU structure.
3913	* @param pVmcsInfo The VMCS info. object.
3914	*/
3915	static int hmR0VmxSetupVmcsProcCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3916	{
3917	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3918	uint32_t fVal = pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
3919	uint32_t const fZap = pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3920
3921	fVal \|= VMX_PROC_CTLS_HLT_EXIT /* HLT causes a VM-exit. */
3922	\| VMX_PROC_CTLS_USE_TSC_OFFSETTING /* Use TSC-offsetting. */
3923	\| VMX_PROC_CTLS_MOV_DR_EXIT /* MOV DRx causes a VM-exit. */
3924	\| VMX_PROC_CTLS_UNCOND_IO_EXIT /* All IO instructions cause a VM-exit. */
3925	\| VMX_PROC_CTLS_RDPMC_EXIT /* RDPMC causes a VM-exit. */
3926	\| VMX_PROC_CTLS_MONITOR_EXIT /* MONITOR causes a VM-exit. */
3927	\| VMX_PROC_CTLS_MWAIT_EXIT; /* MWAIT causes a VM-exit. */
3928
3929	/* We toggle VMX_PROC_CTLS_MOV_DR_EXIT later, check if it's not -always- needed to be set or clear. */
3930	if ( !(pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_MOV_DR_EXIT)
3931	\|\| (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed0 & VMX_PROC_CTLS_MOV_DR_EXIT))
3932	{
3933	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_MOV_DRX_EXIT;
3934	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3935	}
3936
3937	/* Without nested paging, INVLPG (also affects INVPCID) and MOV CR3 instructions should cause VM-exits. */
3938	if (!pVM->hm.s.fNestedPaging)
3939	{
3940	Assert(!pVM->hm.s.vmx.fUnrestrictedGuest);
3941	fVal \|= VMX_PROC_CTLS_INVLPG_EXIT
3942	\| VMX_PROC_CTLS_CR3_LOAD_EXIT
3943	\| VMX_PROC_CTLS_CR3_STORE_EXIT;
3944	}
3945
3946	/* Use TPR shadowing if supported by the CPU. */
3947	if ( PDMHasApic(pVM)
3948	&& (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW))
3949	{
3950	fVal \|= VMX_PROC_CTLS_USE_TPR_SHADOW; /* CR8 reads from the Virtual-APIC page. */
3951	/* CR8 writes cause a VM-exit based on TPR threshold. */
3952	Assert(!(fVal & VMX_PROC_CTLS_CR8_STORE_EXIT));
3953	Assert(!(fVal & VMX_PROC_CTLS_CR8_LOAD_EXIT));
3954	hmR0VmxSetupVmcsVirtApicAddr(pVmcsInfo);
3955	}
3956	else
3957	{
3958	/* Some 32-bit CPUs do not support CR8 load/store exiting as MOV CR8 is
3959	invalid on 32-bit Intel CPUs. Set this control only for 64-bit guests. */
3960	if (pVM->hm.s.fAllow64BitGuests)
3961	{
3962	fVal \|= VMX_PROC_CTLS_CR8_STORE_EXIT /* CR8 reads cause a VM-exit. */
3963	\| VMX_PROC_CTLS_CR8_LOAD_EXIT; /* CR8 writes cause a VM-exit. */
3964	}
3965	}
3966
3967	/* Use MSR-bitmaps if supported by the CPU. */
3968	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
3969	{
3970	fVal \|= VMX_PROC_CTLS_USE_MSR_BITMAPS;
3971	hmR0VmxSetupVmcsMsrBitmapAddr(pVmcsInfo);
3972	}
3973
3974	/* Use the secondary processor-based VM-execution controls if supported by the CPU. */
3975	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
3976	fVal \|= VMX_PROC_CTLS_USE_SECONDARY_CTLS;
3977
3978	if ((fVal & fZap) != fVal)
3979	{
3980	LogRelFunc(("Invalid processor-based VM-execution controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3981	pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed0, fVal, fZap));
3982	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC;
3983	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3984	}
3985
3986	/* Commit it to the VMCS and update our cache. */
3987	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, fVal);
3988	AssertRC(rc);
3989	pVmcsInfo->u32ProcCtls = fVal;
3990
3991	/* Set up MSR permissions that don't change through the lifetime of the VM. */
3992	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
3993	hmR0VmxSetupVmcsMsrPermissions(pVCpu, pVmcsInfo);
3994
3995	/* Set up secondary processor-based VM-execution controls if the CPU supports it. */
3996	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
3997	return hmR0VmxSetupVmcsProcCtls2(pVCpu, pVmcsInfo);
3998
3999	/* Sanity check, should not really happen. */
4000	if (RT_LIKELY(!pVM->hm.s.vmx.fUnrestrictedGuest))
4001	{ /* likely */ }
4002	else
4003	{
4004	pVCpu->hm.s.u32HMError = VMX_UFC_INVALID_UX_COMBO;
4005	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
4006	}
4007
4008	/* Old CPUs without secondary processor-based VM-execution controls would end up here. */
4009	return VINF_SUCCESS;
4010	}
4011
4012
4013	/**
4014	* Sets up miscellaneous (everything other than Pin, Processor and secondary
4015	* Processor-based VM-execution) control fields in the VMCS.
4016	*
4017	* @returns VBox status code.
4018	* @param pVCpu The cross context virtual CPU structure.
4019	* @param pVmcsInfo The VMCS info. object.
4020	*/
4021	static int hmR0VmxSetupVmcsMiscCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
4022	{
4023	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4024	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUseVmcsShadowing)
4025	{
4026	hmR0VmxSetupVmcsVmreadBitmapAddr(pVCpu);
4027	hmR0VmxSetupVmcsVmwriteBitmapAddr(pVCpu);
4028	}
4029	#endif
4030
4031	Assert(pVmcsInfo->u64VmcsLinkPtr == NIL_RTHCPHYS);
4032	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
4033	AssertRC(rc);
4034
4035	rc = hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(pVmcsInfo);
4036	if (RT_SUCCESS(rc))
4037	{
4038	uint64_t const u64Cr0Mask = hmR0VmxGetFixedCr0Mask(pVCpu);
4039	uint64_t const u64Cr4Mask = hmR0VmxGetFixedCr4Mask(pVCpu);
4040
4041	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, u64Cr0Mask); AssertRC(rc);
4042	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, u64Cr4Mask); AssertRC(rc);
4043
4044	pVmcsInfo->u64Cr0Mask = u64Cr0Mask;
4045	pVmcsInfo->u64Cr4Mask = u64Cr4Mask;
4046
4047	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fLbr)
4048	{
4049	rc = VMXWriteVmcsNw(VMX_VMCS64_GUEST_DEBUGCTL_FULL, MSR_IA32_DEBUGCTL_LBR);
4050	AssertRC(rc);
4051	}
4052	return VINF_SUCCESS;
4053	}
4054	else
4055	LogRelFunc(("Failed to initialize VMCS auto-load/store MSR addresses. rc=%Rrc\n", rc));
4056	return rc;
4057	}
4058
4059
4060	/**
4061	* Sets up the initial exception bitmap in the VMCS based on static conditions.
4062	*
4063	* We shall setup those exception intercepts that don't change during the
4064	* lifetime of the VM here. The rest are done dynamically while loading the
4065	* guest state.
4066	*
4067	* @param pVCpu The cross context virtual CPU structure.
4068	* @param pVmcsInfo The VMCS info. object.
4069	*/
4070	static void hmR0VmxSetupVmcsXcptBitmap(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
4071	{
4072	/*
4073	* The following exceptions are always intercepted:
4074	*
4075	* #AC - To prevent the guest from hanging the CPU.
4076	* #DB - To maintain the DR6 state even when intercepting DRx reads/writes and
4077	* recursive #DBs can cause a CPU hang.
4078	* #PF - To sync our shadow page tables when nested-paging is not used.
4079	*/
4080	bool const fNestedPaging = pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging;
4081	uint32_t const uXcptBitmap = RT_BIT(X86_XCPT_AC)
4082	\| RT_BIT(X86_XCPT_DB)
4083	\| (fNestedPaging ? 0 : RT_BIT(X86_XCPT_PF));
4084
4085	/* Commit it to the VMCS. */
4086	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
4087	AssertRC(rc);
4088
4089	/* Update our cache of the exception bitmap. */
4090	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
4091	}
4092
4093
4094	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4095	/**
4096	* Sets up the VMCS for executing a nested-guest using hardware-assisted VMX.
4097	*
4098	* @returns VBox status code.
4099	* @param pVCpu The cross context virtual CPU structure.
4100	* @param pVmcsInfo The VMCS info. object.
4101	*/
4102	static int hmR0VmxSetupVmcsCtlsNested(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
4103	{
4104	Assert(pVmcsInfo->u64VmcsLinkPtr == NIL_RTHCPHYS);
4105	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
4106	AssertRC(rc);
4107
4108	rc = hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(pVmcsInfo);
4109	if (RT_SUCCESS(rc))
4110	{
4111	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
4112	hmR0VmxSetupVmcsMsrBitmapAddr(pVmcsInfo);
4113
4114	/* Paranoia - We've not yet initialized these, they shall be done while merging the VMCS. */
4115	Assert(!pVmcsInfo->u64Cr0Mask);
4116	Assert(!pVmcsInfo->u64Cr4Mask);
4117	return VINF_SUCCESS;
4118	}
4119	else
4120	LogRelFunc(("Failed to set up the VMCS link pointer in the nested-guest VMCS. rc=%Rrc\n", rc));
4121	return rc;
4122	}
4123	#endif
4124
4125
4126	/**
4127	* Sets up the VMCS for executing a guest (or nested-guest) using hardware-assisted
4128	* VMX.
4129	*
4130	* @returns VBox status code.
4131	* @param pVCpu The cross context virtual CPU structure.
4132	* @param pVmcsInfo The VMCS info. object.
4133	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
4134	*/
4135	static int hmR0VmxSetupVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
4136	{
4137	Assert(pVmcsInfo->pvVmcs);
4138	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4139
4140	/* Set the CPU specified revision identifier at the beginning of the VMCS structure. */
4141	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4142	(uint32_t )pVmcsInfo->pvVmcs = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_ID);
4143	const char * const pszVmcs = fIsNstGstVmcs ? "nested-guest VMCS" : "guest VMCS";
4144
4145	LogFlowFunc(("\n"));
4146
4147	/*
4148	* Initialize the VMCS using VMCLEAR before loading the VMCS.
4149	* See Intel spec. 31.6 "Preparation And Launching A Virtual Machine".
4150	*/
4151	int rc = hmR0VmxClearVmcs(pVmcsInfo);
4152	if (RT_SUCCESS(rc))
4153	{
4154	rc = hmR0VmxLoadVmcs(pVmcsInfo);
4155	if (RT_SUCCESS(rc))
4156	{
4157	/*
4158	* Initialize the hardware-assisted VMX execution handler for guest and nested-guest VMCS.
4159	* The host is always 64-bit since we no longer support 32-bit hosts.
4160	* Currently we have just a single handler for all guest modes as well, see @bugref{6208#c73}.
4161	*/
4162	pVmcsInfo->pfnStartVM = VMXR0StartVM64;
4163	if (!fIsNstGstVmcs)
4164	{
4165	rc = hmR0VmxSetupVmcsPinCtls(pVCpu, pVmcsInfo);
4166	if (RT_SUCCESS(rc))
4167	{
4168	rc = hmR0VmxSetupVmcsProcCtls(pVCpu, pVmcsInfo);
4169	if (RT_SUCCESS(rc))
4170	{
4171	rc = hmR0VmxSetupVmcsMiscCtls(pVCpu, pVmcsInfo);
4172	if (RT_SUCCESS(rc))
4173	{
4174	hmR0VmxSetupVmcsXcptBitmap(pVCpu, pVmcsInfo);
4175	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4176	/*
4177	* If a shadow VMCS is allocated for the VMCS info. object, initialize the
4178	* VMCS revision ID and shadow VMCS indicator bit. Also, clear the VMCS
4179	* making it fit for use when VMCS shadowing is later enabled.
4180	*/
4181	if (pVmcsInfo->pvShadowVmcs)
4182	{
4183	VMXVMCSREVID VmcsRevId;
4184	VmcsRevId.u = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_ID);
4185	VmcsRevId.n.fIsShadowVmcs = 1;
4186	(uint32_t )pVmcsInfo->pvShadowVmcs = VmcsRevId.u;
4187	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
4188	if (RT_SUCCESS(rc))
4189	{ /* likely */ }
4190	else
4191	LogRelFunc(("Failed to initialize shadow VMCS. rc=%Rrc\n", rc));
4192	}
4193	#endif
4194	}
4195	else
4196	LogRelFunc(("Failed to setup miscellaneous controls. rc=%Rrc\n", rc));
4197	}
4198	else
4199	LogRelFunc(("Failed to setup processor-based VM-execution controls. rc=%Rrc\n", rc));
4200	}
4201	else
4202	LogRelFunc(("Failed to setup pin-based controls. rc=%Rrc\n", rc));
4203	}
4204	else
4205	{
4206	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4207	rc = hmR0VmxSetupVmcsCtlsNested(pVCpu, pVmcsInfo);
4208	if (RT_SUCCESS(rc))
4209	{ /* likely */ }
4210	else
4211	LogRelFunc(("Failed to initialize nested-guest VMCS. rc=%Rrc\n", rc));
4212	#else
4213	AssertFailed();
4214	#endif
4215	}
4216	}
4217	else
4218	LogRelFunc(("Failed to load the %s. rc=%Rrc\n", rc, pszVmcs));
4219	}
4220	else
4221	LogRelFunc(("Failed to clear the %s. rc=%Rrc\n", rc, pszVmcs));
4222
4223	/* Sync any CPU internal VMCS data back into our VMCS in memory. */
4224	if (RT_SUCCESS(rc))
4225	{
4226	rc = hmR0VmxClearVmcs(pVmcsInfo);
4227	if (RT_SUCCESS(rc))
4228	{ /* likely */ }
4229	else
4230	LogRelFunc(("Failed to clear the %s post setup. rc=%Rrc\n", rc, pszVmcs));
4231	}
4232
4233	/*
4234	* Update the last-error record both for failures and success, so we
4235	* can propagate the status code back to ring-3 for diagnostics.
4236	*/
4237	hmR0VmxUpdateErrorRecord(pVCpu, rc);
4238	NOREF(pszVmcs);
4239	return rc;
4240	}
4241
4242
4243	/**
4244	* Does global VT-x initialization (called during module initialization).
4245	*
4246	* @returns VBox status code.
4247	*/
4248	VMMR0DECL(int) VMXR0GlobalInit(void)
4249	{
4250	#ifdef HMVMX_USE_FUNCTION_TABLE
4251	AssertCompile(VMX_EXIT_MAX + 1 == RT_ELEMENTS(g_aVMExitHandlers));
4252	# ifdef VBOX_STRICT
4253	for (unsigned i = 0; i < RT_ELEMENTS(g_aVMExitHandlers); i++)
4254	Assert(g_aVMExitHandlers[i].pfn);
4255	# endif
4256	#endif
4257	return VINF_SUCCESS;
4258	}
4259
4260
4261	/**
4262	* Does global VT-x termination (called during module termination).
4263	*/
4264	VMMR0DECL(void) VMXR0GlobalTerm()
4265	{
4266	/* Nothing to do currently. */
4267	}
4268
4269
4270	/**
4271	* Sets up and activates VT-x on the current CPU.
4272	*
4273	* @returns VBox status code.
4274	* @param pHostCpu The HM physical-CPU structure.
4275	* @param pVM The cross context VM structure. Can be
4276	* NULL after a host resume operation.
4277	* @param pvCpuPage Pointer to the VMXON region (can be NULL if @a
4278	* fEnabledByHost is @c true).
4279	* @param HCPhysCpuPage Physical address of the VMXON region (can be 0 if
4280	* @a fEnabledByHost is @c true).
4281	* @param fEnabledByHost Set if SUPR0EnableVTx() or similar was used to
4282	* enable VT-x on the host.
4283	* @param pHwvirtMsrs Pointer to the hardware-virtualization MSRs.
4284	*/
4285	VMMR0DECL(int) VMXR0EnableCpu(PHMPHYSCPU pHostCpu, PVMCC pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost,
4286	PCSUPHWVIRTMSRS pHwvirtMsrs)
4287	{
4288	AssertPtr(pHostCpu);
4289	AssertPtr(pHwvirtMsrs);
4290	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4291
4292	/* Enable VT-x if it's not already enabled by the host. */
4293	if (!fEnabledByHost)
4294	{
4295	int rc = hmR0VmxEnterRootMode(pHostCpu, pVM, HCPhysCpuPage, pvCpuPage);
4296	if (RT_FAILURE(rc))
4297	return rc;
4298	}
4299
4300	/*
4301	* Flush all EPT tagged-TLB entries (in case VirtualBox or any other hypervisor have been
4302	* using EPTPs) so we don't retain any stale guest-physical mappings which won't get
4303	* invalidated when flushing by VPID.
4304	*/
4305	if (pHwvirtMsrs->u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
4306	{
4307	hmR0VmxFlushEpt(NULL /* pVCpu /, NULL / pVmcsInfo */, VMXTLBFLUSHEPT_ALL_CONTEXTS);
4308	pHostCpu->fFlushAsidBeforeUse = false;
4309	}
4310	else
4311	pHostCpu->fFlushAsidBeforeUse = true;
4312
4313	/* Ensure each VCPU scheduled on this CPU gets a new VPID on resume. See @bugref{6255}. */
4314	++pHostCpu->cTlbFlushes;
4315
4316	return VINF_SUCCESS;
4317	}
4318
4319
4320	/**
4321	* Deactivates VT-x on the current CPU.
4322	*
4323	* @returns VBox status code.
4324	* @param pHostCpu The HM physical-CPU structure.
4325	* @param pvCpuPage Pointer to the VMXON region.
4326	* @param HCPhysCpuPage Physical address of the VMXON region.
4327	*
4328	* @remarks This function should never be called when SUPR0EnableVTx() or
4329	* similar was used to enable VT-x on the host.
4330	*/
4331	VMMR0DECL(int) VMXR0DisableCpu(PHMPHYSCPU pHostCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
4332	{
4333	RT_NOREF2(pvCpuPage, HCPhysCpuPage);
4334
4335	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4336	return hmR0VmxLeaveRootMode(pHostCpu);
4337	}
4338
4339
4340	/**
4341	* Does per-VM VT-x initialization.
4342	*
4343	* @returns VBox status code.
4344	* @param pVM The cross context VM structure.
4345	*/
4346	VMMR0DECL(int) VMXR0InitVM(PVMCC pVM)
4347	{
4348	AssertPtr(pVM);
4349	LogFlowFunc(("pVM=%p\n", pVM));
4350
4351	hmR0VmxStructsInit(pVM);
4352	int rc = hmR0VmxStructsAlloc(pVM);
4353	if (RT_FAILURE(rc))
4354	{
4355	LogRelFunc(("Failed to allocated VMX structures. rc=%Rrc\n", rc));
4356	return rc;
4357	}
4358
4359	/* Setup the crash dump page. */
4360	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4361	strcpy((char *)pVM->hm.s.vmx.pbScratch, "SCRATCH Magic");
4362	(uint64_t )(pVM->hm.s.vmx.pbScratch + 16) = UINT64_C(0xdeadbeefdeadbeef);
4363	#endif
4364	return VINF_SUCCESS;
4365	}
4366
4367
4368	/**
4369	* Does per-VM VT-x termination.
4370	*
4371	* @returns VBox status code.
4372	* @param pVM The cross context VM structure.
4373	*/
4374	VMMR0DECL(int) VMXR0TermVM(PVMCC pVM)
4375	{
4376	AssertPtr(pVM);
4377	LogFlowFunc(("pVM=%p\n", pVM));
4378
4379	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4380	if (pVM->hm.s.vmx.hMemObjScratch != NIL_RTR0MEMOBJ)
4381	{
4382	Assert(pVM->hm.s.vmx.pvScratch);
4383	ASMMemZero32(pVM->hm.s.vmx.pvScratch, X86_PAGE_4K_SIZE);
4384	}
4385	#endif
4386	hmR0VmxStructsFree(pVM);
4387	return VINF_SUCCESS;
4388	}
4389
4390
4391	/**
4392	* Sets up the VM for execution using hardware-assisted VMX.
4393	* This function is only called once per-VM during initialization.
4394	*
4395	* @returns VBox status code.
4396	* @param pVM The cross context VM structure.
4397	*/
4398	VMMR0DECL(int) VMXR0SetupVM(PVMCC pVM)
4399	{
4400	AssertPtr(pVM);
4401	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4402
4403	LogFlowFunc(("pVM=%p\n", pVM));
4404
4405	/*
4406	* At least verify if VMX is enabled, since we can't check if we're in VMX root mode or not
4407	* without causing a #GP.
4408	*/
4409	RTCCUINTREG const uHostCr4 = ASMGetCR4();
4410	if (RT_LIKELY(uHostCr4 & X86_CR4_VMXE))
4411	{ /* likely */ }
4412	else
4413	return VERR_VMX_NOT_IN_VMX_ROOT_MODE;
4414
4415	/*
4416	* Without unrestricted guest execution, pRealModeTSS and pNonPagingModeEPTPageTable must
4417	* always be allocated. We no longer support the highly unlikely case of unrestricted guest
4418	* without pRealModeTSS, see hmR3InitFinalizeR0Intel().
4419	*/
4420	if ( !pVM->hm.s.vmx.fUnrestrictedGuest
4421	&& ( !pVM->hm.s.vmx.pNonPagingModeEPTPageTable
4422	\|\| !pVM->hm.s.vmx.pRealModeTSS))
4423	{
4424	LogRelFunc(("Invalid real-on-v86 state.\n"));
4425	return VERR_INTERNAL_ERROR;
4426	}
4427
4428	/* Initialize these always, see hmR3InitFinalizeR0().*/
4429	pVM->hm.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NONE;
4430	pVM->hm.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NONE;
4431
4432	/* Setup the tagged-TLB flush handlers. */
4433	int rc = hmR0VmxSetupTaggedTlb(pVM);
4434	if (RT_FAILURE(rc))
4435	{
4436	LogRelFunc(("Failed to setup tagged TLB. rc=%Rrc\n", rc));
4437	return rc;
4438	}
4439
4440	/* Determine LBR capabilities. */
4441	if (pVM->hm.s.vmx.fLbr)
4442	{
4443	rc = hmR0VmxSetupLbrMsrRange(pVM);
4444	if (RT_FAILURE(rc))
4445	{
4446	LogRelFunc(("Failed to setup LBR MSR range. rc=%Rrc\n", rc));
4447	return rc;
4448	}
4449	}
4450
4451	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4452	/* Setup the shadow VMCS fields array and VMREAD/VMWRITE bitmaps. */
4453	if (pVM->hm.s.vmx.fUseVmcsShadowing)
4454	{
4455	rc = hmR0VmxSetupShadowVmcsFieldsArrays(pVM);
4456	if (RT_SUCCESS(rc))
4457	hmR0VmxSetupVmreadVmwriteBitmaps(pVM);
4458	else
4459	{
4460	LogRelFunc(("Failed to setup shadow VMCS fields arrays. rc=%Rrc\n", rc));
4461	return rc;
4462	}
4463	}
4464	#endif
4465
4466	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
4467	{
4468	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
4469	Log4Func(("pVCpu=%p idCpu=%RU32\n", pVCpu, pVCpu->idCpu));
4470
4471	rc = hmR0VmxSetupVmcs(pVCpu, &pVCpu->hm.s.vmx.VmcsInfo, false /* fIsNstGstVmcs */);
4472	if (RT_SUCCESS(rc))
4473	{
4474	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4475	if (pVM->cpum.ro.GuestFeatures.fVmx)
4476	{
4477	rc = hmR0VmxSetupVmcs(pVCpu, &pVCpu->hm.s.vmx.VmcsInfoNstGst, true /* fIsNstGstVmcs */);
4478	if (RT_SUCCESS(rc))
4479	{ /* likely */ }
4480	else
4481	{
4482	LogRelFunc(("Nested-guest VMCS setup failed. rc=%Rrc\n", rc));
4483	return rc;
4484	}
4485	}
4486	#endif
4487	}
4488	else
4489	{
4490	LogRelFunc(("VMCS setup failed. rc=%Rrc\n", rc));
4491	return rc;
4492	}
4493	}
4494
4495	return VINF_SUCCESS;
4496	}
4497
4498
4499	/**
4500	* Saves the host control registers (CR0, CR3, CR4) into the host-state area in
4501	* the VMCS.
4502	*/
4503	static void hmR0VmxExportHostControlRegs(void)
4504	{
4505	int rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR0, ASMGetCR0()); AssertRC(rc);
4506	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR3, ASMGetCR3()); AssertRC(rc);
4507	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR4, ASMGetCR4()); AssertRC(rc);
4508	}
4509
4510
4511	/**
4512	* Saves the host segment registers and GDTR, IDTR, (TR, GS and FS bases) into
4513	* the host-state area in the VMCS.
4514	*
4515	* @returns VBox status code.
4516	* @param pVCpu The cross context virtual CPU structure.
4517	*/
4518	static int hmR0VmxExportHostSegmentRegs(PVMCPUCC pVCpu)
4519	{
4520	/**
4521	* Macro for adjusting host segment selectors to satisfy VT-x's VM-entry
4522	* requirements. See hmR0VmxExportHostSegmentRegs().
4523	*/
4524	#define VMXLOCAL_ADJUST_HOST_SEG(a_Seg, a_selValue) \
4525	if ((a_selValue) & (X86_SEL_RPL \| X86_SEL_LDT)) \
4526	{ \
4527	bool fValidSelector = true; \
4528	if ((a_selValue) & X86_SEL_LDT) \
4529	{ \
4530	uint32_t const uAttr = ASMGetSegAttr(a_selValue); \
4531	fValidSelector = RT_BOOL(uAttr != UINT32_MAX && (uAttr & X86_DESC_P)); \
4532	} \
4533	if (fValidSelector) \
4534	{ \
4535	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##a_Seg; \
4536	pVCpu->hm.s.vmx.RestoreHost.uHostSel##a_Seg = (a_selValue); \
4537	} \
4538	(a_selValue) = 0; \
4539	}
4540
4541	/*
4542	* If we've executed guest code using hardware-assisted VMX, the host-state bits
4543	* will be messed up. We should -not- save the messed up state without restoring
4544	* the original host-state, see @bugref{7240}.
4545	*
4546	* This apparently can happen (most likely the FPU changes), deal with it rather than
4547	* asserting. Was observed booting Solaris 10u10 32-bit guest.
4548	*/
4549	if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
4550	&& (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
4551	{
4552	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags,
4553	pVCpu->idCpu));
4554	VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
4555	}
4556	pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
4557
4558	/*
4559	* Host segment registers.
4560	*/
4561	RTSEL uSelES = ASMGetES();
4562	RTSEL uSelCS = ASMGetCS();
4563	RTSEL uSelSS = ASMGetSS();
4564	RTSEL uSelDS = ASMGetDS();
4565	RTSEL uSelFS = ASMGetFS();
4566	RTSEL uSelGS = ASMGetGS();
4567	RTSEL uSelTR = ASMGetTR();
4568
4569	/*
4570	* Determine if the host segment registers are suitable for VT-x. Otherwise use zero to
4571	* gain VM-entry and restore them before we get preempted.
4572	*
4573	* See Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers".
4574	*/
4575	VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS);
4576	VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES);
4577	VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS);
4578	VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS);
4579
4580	/* Verification based on Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers" */
4581	Assert(!(uSelCS & X86_SEL_RPL)); Assert(!(uSelCS & X86_SEL_LDT));
4582	Assert(!(uSelSS & X86_SEL_RPL)); Assert(!(uSelSS & X86_SEL_LDT));
4583	Assert(!(uSelDS & X86_SEL_RPL)); Assert(!(uSelDS & X86_SEL_LDT));
4584	Assert(!(uSelES & X86_SEL_RPL)); Assert(!(uSelES & X86_SEL_LDT));
4585	Assert(!(uSelFS & X86_SEL_RPL)); Assert(!(uSelFS & X86_SEL_LDT));
4586	Assert(!(uSelGS & X86_SEL_RPL)); Assert(!(uSelGS & X86_SEL_LDT));
4587	Assert(!(uSelTR & X86_SEL_RPL)); Assert(!(uSelTR & X86_SEL_LDT));
4588	Assert(uSelCS);
4589	Assert(uSelTR);
4590
4591	/* Write these host selector fields into the host-state area in the VMCS. */
4592	int rc = VMXWriteVmcs16(VMX_VMCS16_HOST_CS_SEL, uSelCS); AssertRC(rc);
4593	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_SS_SEL, uSelSS); AssertRC(rc);
4594	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_DS_SEL, uSelDS); AssertRC(rc);
4595	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_ES_SEL, uSelES); AssertRC(rc);
4596	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_FS_SEL, uSelFS); AssertRC(rc);
4597	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_GS_SEL, uSelGS); AssertRC(rc);
4598	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_TR_SEL, uSelTR); AssertRC(rc);
4599
4600	/*
4601	* Host GDTR and IDTR.
4602	*/
4603	RTGDTR Gdtr;
4604	RTIDTR Idtr;
4605	RT_ZERO(Gdtr);
4606	RT_ZERO(Idtr);
4607	ASMGetGDTR(&Gdtr);
4608	ASMGetIDTR(&Idtr);
4609	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_GDTR_BASE, Gdtr.pGdt); AssertRC(rc);
4610	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_IDTR_BASE, Idtr.pIdt); AssertRC(rc);
4611
4612	/*
4613	* Determine if we need to manually need to restore the GDTR and IDTR limits as VT-x zaps
4614	* them to the maximum limit (0xffff) on every VM-exit.
4615	*/
4616	if (Gdtr.cbGdt != 0xffff)
4617	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_GDTR;
4618
4619	/*
4620	* IDT limit is effectively capped at 0xfff. (See Intel spec. 6.14.1 "64-Bit Mode IDT" and
4621	* Intel spec. 6.2 "Exception and Interrupt Vectors".) Therefore if the host has the limit
4622	* as 0xfff, VT-x bloating the limit to 0xffff shouldn't cause any different CPU behavior.
4623	* However, several hosts either insists on 0xfff being the limit (Windows Patch Guard) or
4624	* uses the limit for other purposes (darwin puts the CPU ID in there but botches sidt
4625	* alignment in at least one consumer). So, we're only allowing the IDTR.LIMIT to be left
4626	* at 0xffff on hosts where we are sure it won't cause trouble.
4627	*/
4628	#if defined(RT_OS_LINUX) \|\| defined(RT_OS_SOLARIS)
4629	if (Idtr.cbIdt < 0x0fff)
4630	#else
4631	if (Idtr.cbIdt != 0xffff)
4632	#endif
4633	{
4634	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_IDTR;
4635	AssertCompile(sizeof(Idtr) == sizeof(X86XDTR64));
4636	memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostIdtr, &Idtr, sizeof(X86XDTR64));
4637	}
4638
4639	/*
4640	* Host TR base. Verify that TR selector doesn't point past the GDT. Masking off the TI
4641	* and RPL bits is effectively what the CPU does for "scaling by 8". TI is always 0 and
4642	* RPL should be too in most cases.
4643	*/
4644	AssertMsgReturn((uSelTR \| X86_SEL_RPL_LDT) <= Gdtr.cbGdt,
4645	("TR selector exceeds limit. TR=%RTsel cbGdt=%#x\n", uSelTR, Gdtr.cbGdt), VERR_VMX_INVALID_HOST_STATE);
4646
4647	PCX86DESCHC pDesc = (PCX86DESCHC)(Gdtr.pGdt + (uSelTR & X86_SEL_MASK));
4648	uintptr_t const uTRBase = X86DESC64_BASE(pDesc);
4649
4650	/*
4651	* VT-x unconditionally restores the TR limit to 0x67 and type to 11 (32-bit busy TSS) on
4652	* all VM-exits. The type is the same for 64-bit busy TSS[1]. The limit needs manual
4653	* restoration if the host has something else. Task switching is not supported in 64-bit
4654	* mode[2], but the limit still matters as IOPM is supported in 64-bit mode. Restoring the
4655	* limit lazily while returning to ring-3 is safe because IOPM is not applicable in ring-0.
4656	*
4657	* [1] See Intel spec. 3.5 "System Descriptor Types".
4658	* [2] See Intel spec. 7.2.3 "TSS Descriptor in 64-bit mode".
4659	*/
4660	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4661	Assert(pDesc->System.u4Type == 11);
4662	if ( pDesc->System.u16LimitLow != 0x67
4663	\|\| pDesc->System.u4LimitHigh)
4664	{
4665	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_TR;
4666	/* If the host has made GDT read-only, we would need to temporarily toggle CR0.WP before writing the GDT. */
4667	if (pVM->hm.s.fHostKernelFeatures & SUPKERNELFEATURES_GDT_READ_ONLY)
4668	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_READ_ONLY;
4669	pVCpu->hm.s.vmx.RestoreHost.uHostSelTR = uSelTR;
4670	}
4671
4672	/*
4673	* Store the GDTR as we need it when restoring the GDT and while restoring the TR.
4674	*/
4675	if (pVCpu->hm.s.vmx.fRestoreHostFlags & (VMX_RESTORE_HOST_GDTR \| VMX_RESTORE_HOST_SEL_TR))
4676	{
4677	AssertCompile(sizeof(Gdtr) == sizeof(X86XDTR64));
4678	memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64));
4679	if (pVM->hm.s.fHostKernelFeatures & SUPKERNELFEATURES_GDT_NEED_WRITABLE)
4680	{
4681	/* The GDT is read-only but the writable GDT is available. */
4682	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_NEED_WRITABLE;
4683	pVCpu->hm.s.vmx.RestoreHost.HostGdtrRw.cb = Gdtr.cbGdt;
4684	rc = SUPR0GetCurrentGdtRw(&pVCpu->hm.s.vmx.RestoreHost.HostGdtrRw.uAddr);
4685	AssertRCReturn(rc, rc);
4686	}
4687	}
4688
4689	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_TR_BASE, uTRBase);
4690	AssertRC(rc);
4691
4692	/*
4693	* Host FS base and GS base.
4694	*/
4695	uint64_t const u64FSBase = ASMRdMsr(MSR_K8_FS_BASE);
4696	uint64_t const u64GSBase = ASMRdMsr(MSR_K8_GS_BASE);
4697	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_FS_BASE, u64FSBase); AssertRC(rc);
4698	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_GS_BASE, u64GSBase); AssertRC(rc);
4699
4700	/* Store the base if we have to restore FS or GS manually as we need to restore the base as well. */
4701	if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_FS)
4702	pVCpu->hm.s.vmx.RestoreHost.uHostFSBase = u64FSBase;
4703	if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_GS)
4704	pVCpu->hm.s.vmx.RestoreHost.uHostGSBase = u64GSBase;
4705
4706	return VINF_SUCCESS;
4707	#undef VMXLOCAL_ADJUST_HOST_SEG
4708	}
4709
4710
4711	/**
4712	* Exports certain host MSRs in the VM-exit MSR-load area and some in the
4713	* host-state area of the VMCS.
4714	*
4715	* These MSRs will be automatically restored on the host after every successful
4716	* VM-exit.
4717	*
4718	* @param pVCpu The cross context virtual CPU structure.
4719	*
4720	* @remarks No-long-jump zone!!!
4721	*/
4722	static void hmR0VmxExportHostMsrs(PVMCPUCC pVCpu)
4723	{
4724	AssertPtr(pVCpu);
4725
4726	/*
4727	* Save MSRs that we restore lazily (due to preemption or transition to ring-3)
4728	* rather than swapping them on every VM-entry.
4729	*/
4730	hmR0VmxLazySaveHostMsrs(pVCpu);
4731
4732	/*
4733	* Host Sysenter MSRs.
4734	*/
4735	int rc = VMXWriteVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, ASMRdMsr_Low(MSR_IA32_SYSENTER_CS)); AssertRC(rc);
4736	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP)); AssertRC(rc);
4737	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP)); AssertRC(rc);
4738
4739	/*
4740	* Host EFER MSR.
4741	*
4742	* If the CPU supports the newer VMCS controls for managing EFER, use it. Otherwise it's
4743	* done as part of auto-load/store MSR area in the VMCS, see hmR0VmxExportGuestMsrs().
4744	*/
4745	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4746	if (pVM->hm.s.vmx.fSupportsVmcsEfer)
4747	{
4748	rc = VMXWriteVmcs64(VMX_VMCS64_HOST_EFER_FULL, pVM->hm.s.vmx.u64HostMsrEfer);
4749	AssertRC(rc);
4750	}
4751
4752	/** @todo IA32_PERF_GLOBALCTRL, IA32_PAT also see
4753	* hmR0VmxExportGuestEntryExitCtls(). */
4754	}
4755
4756
4757	/**
4758	* Figures out if we need to swap the EFER MSR which is particularly expensive.
4759	*
4760	* We check all relevant bits. For now, that's everything besides LMA/LME, as
4761	* these two bits are handled by VM-entry, see hmR0VMxExportGuestEntryExitCtls().
4762	*
4763	* @returns true if we need to load guest EFER, false otherwise.
4764	* @param pVCpu The cross context virtual CPU structure.
4765	* @param pVmxTransient The VMX-transient structure.
4766	*
4767	* @remarks Requires EFER, CR4.
4768	* @remarks No-long-jump zone!!!
4769	*/
4770	static bool hmR0VmxShouldSwapEferMsr(PCVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
4771	{
4772	#ifdef HMVMX_ALWAYS_SWAP_EFER
4773	RT_NOREF2(pVCpu, pVmxTransient);
4774	return true;
4775	#else
4776	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
4777	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4778	uint64_t const u64HostEfer = pVM->hm.s.vmx.u64HostMsrEfer;
4779	uint64_t const u64GuestEfer = pCtx->msrEFER;
4780
4781	# ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4782	/*
4783	* For nested-guests, we shall honor swapping the EFER MSR when requested by
4784	* the nested-guest.
4785	*/
4786	if ( pVmxTransient->fIsNestedGuest
4787	&& ( CPUMIsGuestVmxEntryCtlsSet(pCtx, VMX_ENTRY_CTLS_LOAD_EFER_MSR)
4788	\|\| CPUMIsGuestVmxExitCtlsSet(pCtx, VMX_EXIT_CTLS_SAVE_EFER_MSR)
4789	\|\| CPUMIsGuestVmxExitCtlsSet(pCtx, VMX_EXIT_CTLS_LOAD_EFER_MSR)))
4790	return true;
4791	# else
4792	RT_NOREF(pVmxTransient);
4793	#endif
4794
4795	/*
4796	* For 64-bit guests, if EFER.SCE bit differs, we need to swap the EFER MSR
4797	* to ensure that the guest's SYSCALL behaviour isn't broken, see @bugref{7386}.
4798	*/
4799	if ( CPUMIsGuestInLongModeEx(pCtx)
4800	&& (u64GuestEfer & MSR_K6_EFER_SCE) != (u64HostEfer & MSR_K6_EFER_SCE))
4801	return true;
4802
4803	/*
4804	* If the guest uses PAE and EFER.NXE bit differs, we need to swap the EFER MSR
4805	* as it affects guest paging. 64-bit paging implies CR4.PAE as well.
4806	*
4807	* See Intel spec. 4.5 "IA-32e Paging".
4808	* See Intel spec. 4.1.1 "Three Paging Modes".
4809	*
4810	* Verify that we always intercept CR4.PAE and CR0.PG bits, so we don't need to
4811	* import CR4 and CR0 from the VMCS here as those bits are always up to date.
4812	*/
4813	Assert(hmR0VmxGetFixedCr4Mask(pVCpu) & X86_CR4_PAE);
4814	Assert(hmR0VmxGetFixedCr0Mask(pVCpu) & X86_CR0_PG);
4815	if ( (pCtx->cr4 & X86_CR4_PAE)
4816	&& (pCtx->cr0 & X86_CR0_PG))
4817	{
4818	/*
4819	* If nested paging is not used, verify that the guest paging mode matches the
4820	* shadow paging mode which is/will be placed in the VMCS (which is what will
4821	* actually be used while executing the guest and not the CR4 shadow value).
4822	*/
4823	AssertMsg(pVM->hm.s.fNestedPaging \|\| ( pVCpu->hm.s.enmShadowMode == PGMMODE_PAE
4824	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_PAE_NX
4825	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_AMD64
4826	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_AMD64_NX),
4827	("enmShadowMode=%u\n", pVCpu->hm.s.enmShadowMode));
4828	if ((u64GuestEfer & MSR_K6_EFER_NXE) != (u64HostEfer & MSR_K6_EFER_NXE))
4829	{
4830	/* Verify that the host is NX capable. */
4831	Assert(pVCpu->CTX_SUFF(pVM)->cpum.ro.HostFeatures.fNoExecute);
4832	return true;
4833	}
4834	}
4835
4836	return false;
4837	#endif
4838	}
4839
4840
4841	/**
4842	* Exports the guest state with appropriate VM-entry and VM-exit controls in the
4843	* VMCS.
4844	*
4845	* This is typically required when the guest changes paging mode.
4846	*
4847	* @returns VBox status code.
4848	* @param pVCpu The cross context virtual CPU structure.
4849	* @param pVmxTransient The VMX-transient structure.
4850	*
4851	* @remarks Requires EFER.
4852	* @remarks No-long-jump zone!!!
4853	*/
4854	static int hmR0VmxExportGuestEntryExitCtls(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
4855	{
4856	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_ENTRY_EXIT_CTLS)
4857	{
4858	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4859	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
4860
4861	/*
4862	* VM-entry controls.
4863	*/
4864	{
4865	uint32_t fVal = pVM->hm.s.vmx.Msrs.EntryCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
4866	uint32_t const fZap = pVM->hm.s.vmx.Msrs.EntryCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
4867
4868	/*
4869	* Load the guest debug controls (DR7 and IA32_DEBUGCTL MSR) on VM-entry.
4870	* The first VT-x capable CPUs only supported the 1-setting of this bit.
4871	*
4872	* For nested-guests, this is a mandatory VM-entry control. It's also
4873	* required because we do not want to leak host bits to the nested-guest.
4874	*/
4875	fVal \|= VMX_ENTRY_CTLS_LOAD_DEBUG;
4876
4877	/*
4878	* Set if the guest is in long mode. This will set/clear the EFER.LMA bit on VM-entry.
4879	*
4880	* For nested-guests, the "IA-32e mode guest" control we initialize with what is
4881	* required to get the nested-guest working with hardware-assisted VMX execution.
4882	* It depends on the nested-guest's IA32_EFER.LMA bit. Remember, a nested hypervisor
4883	* can skip intercepting changes to the EFER MSR. This is why it it needs to be done
4884	* here rather than while merging the guest VMCS controls.
4885	*/
4886	if (CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
4887	{
4888	Assert(pVCpu->cpum.GstCtx.msrEFER & MSR_K6_EFER_LME);
4889	fVal \|= VMX_ENTRY_CTLS_IA32E_MODE_GUEST;
4890	}
4891	else
4892	Assert(!(fVal & VMX_ENTRY_CTLS_IA32E_MODE_GUEST));
4893
4894	/*
4895	* If the CPU supports the newer VMCS controls for managing guest/host EFER, use it.
4896	*
4897	* For nested-guests, we use the "load IA32_EFER" if the hardware supports it,
4898	* regardless of whether the nested-guest VMCS specifies it because we are free to
4899	* load whatever MSRs we require and we do not need to modify the guest visible copy
4900	* of the VM-entry MSR load area.
4901	*/
4902	if ( pVM->hm.s.vmx.fSupportsVmcsEfer
4903	&& hmR0VmxShouldSwapEferMsr(pVCpu, pVmxTransient))
4904	fVal \|= VMX_ENTRY_CTLS_LOAD_EFER_MSR;
4905	else
4906	Assert(!(fVal & VMX_ENTRY_CTLS_LOAD_EFER_MSR));
4907
4908	/*
4909	* The following should -not- be set (since we're not in SMM mode):
4910	* - VMX_ENTRY_CTLS_ENTRY_TO_SMM
4911	* - VMX_ENTRY_CTLS_DEACTIVATE_DUAL_MON
4912	*/
4913
4914	/** @todo VMX_ENTRY_CTLS_LOAD_PERF_MSR,
4915	* VMX_ENTRY_CTLS_LOAD_PAT_MSR. */
4916
4917	if ((fVal & fZap) == fVal)
4918	{ /* likely */ }
4919	else
4920	{
4921	Log4Func(("Invalid VM-entry controls combo! Cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
4922	pVM->hm.s.vmx.Msrs.EntryCtls.n.allowed0, fVal, fZap));
4923	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_ENTRY;
4924	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
4925	}
4926
4927	/* Commit it to the VMCS. */
4928	if (pVmcsInfo->u32EntryCtls != fVal)
4929	{
4930	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY, fVal);
4931	AssertRC(rc);
4932	pVmcsInfo->u32EntryCtls = fVal;
4933	}
4934	}
4935
4936	/*
4937	* VM-exit controls.
4938	*/
4939	{
4940	uint32_t fVal = pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
4941	uint32_t const fZap = pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
4942
4943	/*
4944	* Save debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x CPUs only
4945	* supported the 1-setting of this bit.
4946	*
4947	* For nested-guests, we set the "save debug controls" as the converse
4948	* "load debug controls" is mandatory for nested-guests anyway.
4949	*/
4950	fVal \|= VMX_EXIT_CTLS_SAVE_DEBUG;
4951
4952	/*
4953	* Set the host long mode active (EFER.LMA) bit (which Intel calls
4954	* "Host address-space size") if necessary. On VM-exit, VT-x sets both the
4955	* host EFER.LMA and EFER.LME bit to this value. See assertion in
4956	* hmR0VmxExportHostMsrs().
4957	*
4958	* For nested-guests, we always set this bit as we do not support 32-bit
4959	* hosts.
4960	*/
4961	fVal \|= VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE;
4962
4963	/*
4964	* If the VMCS EFER MSR fields are supported by the hardware, we use it.
4965	*
4966	* For nested-guests, we should use the "save IA32_EFER" control if we also
4967	* used the "load IA32_EFER" control while exporting VM-entry controls.
4968	*/
4969	if ( pVM->hm.s.vmx.fSupportsVmcsEfer
4970	&& hmR0VmxShouldSwapEferMsr(pVCpu, pVmxTransient))
4971	{
4972	fVal \|= VMX_EXIT_CTLS_SAVE_EFER_MSR
4973	\| VMX_EXIT_CTLS_LOAD_EFER_MSR;
4974	}
4975
4976	/*
4977	* Enable saving of the VMX-preemption timer value on VM-exit.
4978	* For nested-guests, currently not exposed/used.
4979	*/
4980	if ( pVM->hm.s.vmx.fUsePreemptTimer
4981	&& (pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER))
4982	fVal \|= VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER;
4983
4984	/* Don't acknowledge external interrupts on VM-exit. We want to let the host do that. */
4985	Assert(!(fVal & VMX_EXIT_CTLS_ACK_EXT_INT));
4986
4987	/** @todo VMX_EXIT_CTLS_LOAD_PERF_MSR,
4988	* VMX_EXIT_CTLS_SAVE_PAT_MSR,
4989	* VMX_EXIT_CTLS_LOAD_PAT_MSR. */
4990
4991	if ((fVal & fZap) == fVal)
4992	{ /* likely */ }
4993	else
4994	{
4995	Log4Func(("Invalid VM-exit controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%R#X32\n",
4996	pVM->hm.s.vmx.Msrs.ExitCtls.n.allowed0, fVal, fZap));
4997	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_EXIT;
4998	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
4999	}
5000
5001	/* Commit it to the VMCS. */
5002	if (pVmcsInfo->u32ExitCtls != fVal)
5003	{
5004	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT, fVal);
5005	AssertRC(rc);
5006	pVmcsInfo->u32ExitCtls = fVal;
5007	}
5008	}
5009
5010	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
5011	}
5012	return VINF_SUCCESS;
5013	}
5014
5015
5016	/**
5017	* Sets the TPR threshold in the VMCS.
5018	*
5019	* @param pVmcsInfo The VMCS info. object.
5020	* @param u32TprThreshold The TPR threshold (task-priority class only).
5021	*/
5022	DECLINLINE(void) hmR0VmxApicSetTprThreshold(PVMXVMCSINFO pVmcsInfo, uint32_t u32TprThreshold)
5023	{
5024	Assert(!(u32TprThreshold & ~VMX_TPR_THRESHOLD_MASK)); /* Bits 31:4 MBZ. */
5025	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
5026	RT_NOREF(pVmcsInfo);
5027	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
5028	AssertRC(rc);
5029	}
5030
5031
5032	/**
5033	* Exports the guest APIC TPR state into the VMCS.
5034	*
5035	* @param pVCpu The cross context virtual CPU structure.
5036	* @param pVmxTransient The VMX-transient structure.
5037	*
5038	* @remarks No-long-jump zone!!!
5039	*/
5040	static void hmR0VmxExportGuestApicTpr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5041	{
5042	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_APIC_TPR)
5043	{
5044	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_APIC_TPR);
5045
5046	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5047	if (!pVmxTransient->fIsNestedGuest)
5048	{
5049	if ( PDMHasApic(pVCpu->CTX_SUFF(pVM))
5050	&& APICIsEnabled(pVCpu))
5051	{
5052	/*
5053	* Setup TPR shadowing.
5054	*/
5055	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
5056	{
5057	bool fPendingIntr = false;
5058	uint8_t u8Tpr = 0;
5059	uint8_t u8PendingIntr = 0;
5060	int rc = APICGetTpr(pVCpu, &u8Tpr, &fPendingIntr, &u8PendingIntr);
5061	AssertRC(rc);
5062
5063	/*
5064	* If there are interrupts pending but masked by the TPR, instruct VT-x to
5065	* cause a TPR-below-threshold VM-exit when the guest lowers its TPR below the
5066	* priority of the pending interrupt so we can deliver the interrupt. If there
5067	* are no interrupts pending, set threshold to 0 to not cause any
5068	* TPR-below-threshold VM-exits.
5069	*/
5070	uint32_t u32TprThreshold = 0;
5071	if (fPendingIntr)
5072	{
5073	/* Bits 3:0 of the TPR threshold field correspond to bits 7:4 of the TPR
5074	(which is the Task-Priority Class). */
5075	const uint8_t u8PendingPriority = u8PendingIntr >> 4;
5076	const uint8_t u8TprPriority = u8Tpr >> 4;
5077	if (u8PendingPriority <= u8TprPriority)
5078	u32TprThreshold = u8PendingPriority;
5079	}
5080
5081	hmR0VmxApicSetTprThreshold(pVmcsInfo, u32TprThreshold);
5082	}
5083	}
5084	}
5085	/* else: the TPR threshold has already been updated while merging the nested-guest VMCS. */
5086	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_APIC_TPR);
5087	}
5088	}
5089
5090
5091	/**
5092	* Gets the guest interruptibility-state and updates related force-flags.
5093	*
5094	* @returns Guest's interruptibility-state.
5095	* @param pVCpu The cross context virtual CPU structure.
5096	*
5097	* @remarks No-long-jump zone!!!
5098	*/
5099	static uint32_t hmR0VmxGetGuestIntrStateAndUpdateFFs(PVMCPUCC pVCpu)
5100	{
5101	/*
5102	* Check if we should inhibit interrupt delivery due to instructions like STI and MOV SS.
5103	*/
5104	uint32_t fIntrState = 0;
5105	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
5106	{
5107	/* If inhibition is active, RIP and RFLAGS should've been imported from the VMCS already. */
5108	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS);
5109
5110	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5111	if (pCtx->rip == EMGetInhibitInterruptsPC(pVCpu))
5112	{
5113	if (pCtx->eflags.Bits.u1IF)
5114	fIntrState = VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
5115	else
5116	fIntrState = VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS;
5117	}
5118	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
5119	{
5120	/*
5121	* We can clear the inhibit force flag as even if we go back to the recompiler
5122	* without executing guest code in VT-x, the flag's condition to be cleared is
5123	* met and thus the cleared state is correct.
5124	*/
5125	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
5126	}
5127	}
5128
5129	/*
5130	* Check if we should inhibit NMI delivery.
5131	*/
5132	if (CPUMIsGuestNmiBlocking(pVCpu))
5133	fIntrState \|= VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI;
5134
5135	/*
5136	* Validate.
5137	*/
5138	#ifdef VBOX_STRICT
5139	/* We don't support block-by-SMI yet.*/
5140	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI));
5141
5142	/* Block-by-STI must not be set when interrupts are disabled. */
5143	if (fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
5144	{
5145	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
5146	Assert(pVCpu->cpum.GstCtx.eflags.u & X86_EFL_IF);
5147	}
5148	#endif
5149
5150	return fIntrState;
5151	}
5152
5153
5154	/**
5155	* Exports the exception intercepts required for guest execution in the VMCS.
5156	*
5157	* @param pVCpu The cross context virtual CPU structure.
5158	* @param pVmxTransient The VMX-transient structure.
5159	*
5160	* @remarks No-long-jump zone!!!
5161	*/
5162	static void hmR0VmxExportGuestXcptIntercepts(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5163	{
5164	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_XCPT_INTERCEPTS)
5165	{
5166	/* When executing a nested-guest, we do not need to trap GIM hypercalls by intercepting #UD. */
5167	if ( !pVmxTransient->fIsNestedGuest
5168	&& pVCpu->hm.s.fGIMTrapXcptUD)
5169	hmR0VmxAddXcptIntercept(pVmxTransient, X86_XCPT_UD);
5170	else
5171	hmR0VmxRemoveXcptIntercept(pVCpu, pVmxTransient, X86_XCPT_UD);
5172
5173	/* Other exception intercepts are handled elsewhere, e.g. while exporting guest CR0. */
5174	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_XCPT_INTERCEPTS);
5175	}
5176	}
5177
5178
5179	/**
5180	* Exports the guest's RIP into the guest-state area in the VMCS.
5181	*
5182	* @param pVCpu The cross context virtual CPU structure.
5183	*
5184	* @remarks No-long-jump zone!!!
5185	*/
5186	static void hmR0VmxExportGuestRip(PVMCPUCC pVCpu)
5187	{
5188	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RIP)
5189	{
5190	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP);
5191
5192	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RIP, pVCpu->cpum.GstCtx.rip);
5193	AssertRC(rc);
5194
5195	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RIP);
5196	Log4Func(("rip=%#RX64\n", pVCpu->cpum.GstCtx.rip));
5197	}
5198	}
5199
5200
5201	/**
5202	* Exports the guest's RSP into the guest-state area in the VMCS.
5203	*
5204	* @param pVCpu The cross context virtual CPU structure.
5205	*
5206	* @remarks No-long-jump zone!!!
5207	*/
5208	static void hmR0VmxExportGuestRsp(PVMCPUCC pVCpu)
5209	{
5210	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RSP)
5211	{
5212	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RSP);
5213
5214	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RSP, pVCpu->cpum.GstCtx.rsp);
5215	AssertRC(rc);
5216
5217	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RSP);
5218	Log4Func(("rsp=%#RX64\n", pVCpu->cpum.GstCtx.rsp));
5219	}
5220	}
5221
5222
5223	/**
5224	* Exports the guest's RFLAGS into the guest-state area in the VMCS.
5225	*
5226	* @param pVCpu The cross context virtual CPU structure.
5227	* @param pVmxTransient The VMX-transient structure.
5228	*
5229	* @remarks No-long-jump zone!!!
5230	*/
5231	static void hmR0VmxExportGuestRflags(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5232	{
5233	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RFLAGS)
5234	{
5235	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
5236
5237	/* Intel spec. 2.3.1 "System Flags and Fields in IA-32e Mode" claims the upper 32-bits of RFLAGS are reserved (MBZ).
5238	Let us assert it as such and use 32-bit VMWRITE. */
5239	Assert(!RT_HI_U32(pVCpu->cpum.GstCtx.rflags.u64));
5240	X86EFLAGS fEFlags = pVCpu->cpum.GstCtx.eflags;
5241	Assert(fEFlags.u32 & X86_EFL_RA1_MASK);
5242	Assert(!(fEFlags.u32 & ~(X86_EFL_1 \| X86_EFL_LIVE_MASK)));
5243
5244	/*
5245	* If we're emulating real-mode using Virtual 8086 mode, save the real-mode eflags so
5246	* we can restore them on VM-exit. Modify the real-mode guest's eflags so that VT-x
5247	* can run the real-mode guest code under Virtual 8086 mode.
5248	*/
5249	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5250	if (pVmcsInfo->RealMode.fRealOnV86Active)
5251	{
5252	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
5253	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
5254	Assert(!pVmxTransient->fIsNestedGuest);
5255	pVmcsInfo->RealMode.Eflags.u32 = fEFlags.u32; /* Save the original eflags of the real-mode guest. */
5256	fEFlags.Bits.u1VM = 1; /* Set the Virtual 8086 mode bit. */
5257	fEFlags.Bits.u2IOPL = 0; /* Change IOPL to 0, otherwise certain instructions won't fault. */
5258	}
5259
5260	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RFLAGS, fEFlags.u32);
5261	AssertRC(rc);
5262
5263	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RFLAGS);
5264	Log4Func(("eflags=%#RX32\n", fEFlags.u32));
5265	}
5266	}
5267
5268
5269	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5270	/**
5271	* Copies the nested-guest VMCS to the shadow VMCS.
5272	*
5273	* @returns VBox status code.
5274	* @param pVCpu The cross context virtual CPU structure.
5275	* @param pVmcsInfo The VMCS info. object.
5276	*
5277	* @remarks No-long-jump zone!!!
5278	*/
5279	static int hmR0VmxCopyNstGstToShadowVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
5280	{
5281	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5282	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5283
5284	/*
5285	* Disable interrupts so we don't get preempted while the shadow VMCS is the
5286	* current VMCS, as we may try saving guest lazy MSRs.
5287	*
5288	* Strictly speaking the lazy MSRs are not in the VMCS, but I'd rather not risk
5289	* calling the import VMCS code which is currently performing the guest MSR reads
5290	* (on 64-bit hosts) and accessing the auto-load/store MSR area on 32-bit hosts
5291	* and the rest of the VMX leave session machinery.
5292	*/
5293	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
5294
5295	int rc = hmR0VmxLoadShadowVmcs(pVmcsInfo);
5296	if (RT_SUCCESS(rc))
5297	{
5298	/*
5299	* Copy all guest read/write VMCS fields.
5300	*
5301	* We don't check for VMWRITE failures here for performance reasons and
5302	* because they are not expected to fail, barring irrecoverable conditions
5303	* like hardware errors.
5304	*/
5305	uint32_t const cShadowVmcsFields = pVM->hm.s.vmx.cShadowVmcsFields;
5306	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
5307	{
5308	uint64_t u64Val;
5309	uint32_t const uVmcsField = pVM->hm.s.vmx.paShadowVmcsFields[i];
5310	IEMReadVmxVmcsField(pVmcsNstGst, uVmcsField, &u64Val);
5311	VMXWriteVmcs64(uVmcsField, u64Val);
5312	}
5313
5314	/*
5315	* If the host CPU supports writing all VMCS fields, copy the guest read-only
5316	* VMCS fields, so the guest can VMREAD them without causing a VM-exit.
5317	*/
5318	if (pVM->hm.s.vmx.Msrs.u64Misc & VMX_MISC_VMWRITE_ALL)
5319	{
5320	uint32_t const cShadowVmcsRoFields = pVM->hm.s.vmx.cShadowVmcsRoFields;
5321	for (uint32_t i = 0; i < cShadowVmcsRoFields; i++)
5322	{
5323	uint64_t u64Val;
5324	uint32_t const uVmcsField = pVM->hm.s.vmx.paShadowVmcsRoFields[i];
5325	IEMReadVmxVmcsField(pVmcsNstGst, uVmcsField, &u64Val);
5326	VMXWriteVmcs64(uVmcsField, u64Val);
5327	}
5328	}
5329
5330	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
5331	rc \|= hmR0VmxLoadVmcs(pVmcsInfo);
5332	}
5333
5334	ASMSetFlags(fEFlags);
5335	return rc;
5336	}
5337
5338
5339	/**
5340	* Copies the shadow VMCS to the nested-guest VMCS.
5341	*
5342	* @returns VBox status code.
5343	* @param pVCpu The cross context virtual CPU structure.
5344	* @param pVmcsInfo The VMCS info. object.
5345	*
5346	* @remarks Called with interrupts disabled.
5347	*/
5348	static int hmR0VmxCopyShadowToNstGstVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
5349	{
5350	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
5351	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5352	PVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5353
5354	int rc = hmR0VmxLoadShadowVmcs(pVmcsInfo);
5355	if (RT_SUCCESS(rc))
5356	{
5357	/*
5358	* Copy guest read/write fields from the shadow VMCS.
5359	* Guest read-only fields cannot be modified, so no need to copy them.
5360	*
5361	* We don't check for VMREAD failures here for performance reasons and
5362	* because they are not expected to fail, barring irrecoverable conditions
5363	* like hardware errors.
5364	*/
5365	uint32_t const cShadowVmcsFields = pVM->hm.s.vmx.cShadowVmcsFields;
5366	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
5367	{
5368	uint64_t u64Val;
5369	uint32_t const uVmcsField = pVM->hm.s.vmx.paShadowVmcsFields[i];
5370	VMXReadVmcs64(uVmcsField, &u64Val);
5371	IEMWriteVmxVmcsField(pVmcsNstGst, uVmcsField, u64Val);
5372	}
5373
5374	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
5375	rc \|= hmR0VmxLoadVmcs(pVmcsInfo);
5376	}
5377	return rc;
5378	}
5379
5380
5381	/**
5382	* Enables VMCS shadowing for the given VMCS info. object.
5383	*
5384	* @param pVmcsInfo The VMCS info. object.
5385	*
5386	* @remarks No-long-jump zone!!!
5387	*/
5388	static void hmR0VmxEnableVmcsShadowing(PVMXVMCSINFO pVmcsInfo)
5389	{
5390	uint32_t uProcCtls2 = pVmcsInfo->u32ProcCtls2;
5391	if (!(uProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING))
5392	{
5393	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
5394	uProcCtls2 \|= VMX_PROC_CTLS2_VMCS_SHADOWING;
5395	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, uProcCtls2); AssertRC(rc);
5396	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, pVmcsInfo->HCPhysShadowVmcs); AssertRC(rc);
5397	pVmcsInfo->u32ProcCtls2 = uProcCtls2;
5398	pVmcsInfo->u64VmcsLinkPtr = pVmcsInfo->HCPhysShadowVmcs;
5399	Log4Func(("Enabled\n"));
5400	}
5401	}
5402
5403
5404	/**
5405	* Disables VMCS shadowing for the given VMCS info. object.
5406	*
5407	* @param pVmcsInfo The VMCS info. object.
5408	*
5409	* @remarks No-long-jump zone!!!
5410	*/
5411	static void hmR0VmxDisableVmcsShadowing(PVMXVMCSINFO pVmcsInfo)
5412	{
5413	/*
5414	* We want all VMREAD and VMWRITE instructions to cause VM-exits, so we clear the
5415	* VMCS shadowing control. However, VM-entry requires the shadow VMCS indicator bit
5416	* to match the VMCS shadowing control if the VMCS link pointer is not NIL_RTHCPHYS.
5417	* Hence, we must also reset the VMCS link pointer to ensure VM-entry does not fail.
5418	*
5419	* See Intel spec. 26.2.1.1 "VM-Execution Control Fields".
5420	* See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
5421	*/
5422	uint32_t uProcCtls2 = pVmcsInfo->u32ProcCtls2;
5423	if (uProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
5424	{
5425	uProcCtls2 &= ~VMX_PROC_CTLS2_VMCS_SHADOWING;
5426	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, uProcCtls2); AssertRC(rc);
5427	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS); AssertRC(rc);
5428	pVmcsInfo->u32ProcCtls2 = uProcCtls2;
5429	pVmcsInfo->u64VmcsLinkPtr = NIL_RTHCPHYS;
5430	Log4Func(("Disabled\n"));
5431	}
5432	}
5433	#endif
5434
5435
5436	/**
5437	* Exports the guest hardware-virtualization state.
5438	*
5439	* @returns VBox status code.
5440	* @param pVCpu The cross context virtual CPU structure.
5441	* @param pVmxTransient The VMX-transient structure.
5442	*
5443	* @remarks No-long-jump zone!!!
5444	*/
5445	static int hmR0VmxExportGuestHwvirtState(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5446	{
5447	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_HWVIRT)
5448	{
5449	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5450	/*
5451	* Check if the VMX feature is exposed to the guest and if the host CPU supports
5452	* VMCS shadowing.
5453	*/
5454	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUseVmcsShadowing)
5455	{
5456	/*
5457	* If the nested hypervisor has loaded a current VMCS and is in VMX root mode,
5458	* copy the nested hypervisor's current VMCS into the shadow VMCS and enable
5459	* VMCS shadowing to skip intercepting some or all VMREAD/VMWRITE VM-exits.
5460	*
5461	* We check for VMX root mode here in case the guest executes VMXOFF without
5462	* clearing the current VMCS pointer and our VMXOFF instruction emulation does
5463	* not clear the current VMCS pointer.
5464	*/
5465	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5466	if ( CPUMIsGuestInVmxRootMode(&pVCpu->cpum.GstCtx)
5467	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx)
5468	&& CPUMIsGuestVmxCurrentVmcsValid(&pVCpu->cpum.GstCtx))
5469	{
5470	/* Paranoia. */
5471	Assert(!pVmxTransient->fIsNestedGuest);
5472
5473	/*
5474	* For performance reasons, also check if the nested hypervisor's current VMCS
5475	* was newly loaded or modified before copying it to the shadow VMCS.
5476	*/
5477	if (!pVCpu->hm.s.vmx.fCopiedNstGstToShadowVmcs)
5478	{
5479	int rc = hmR0VmxCopyNstGstToShadowVmcs(pVCpu, pVmcsInfo);
5480	AssertRCReturn(rc, rc);
5481	pVCpu->hm.s.vmx.fCopiedNstGstToShadowVmcs = true;
5482	}
5483	hmR0VmxEnableVmcsShadowing(pVmcsInfo);
5484	}
5485	else
5486	hmR0VmxDisableVmcsShadowing(pVmcsInfo);
5487	}
5488	#else
5489	NOREF(pVmxTransient);
5490	#endif
5491	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_HWVIRT);
5492	}
5493	return VINF_SUCCESS;
5494	}
5495
5496
5497	/**
5498	* Exports the guest CR0 control register into the guest-state area in the VMCS.
5499	*
5500	* The guest FPU state is always pre-loaded hence we don't need to bother about
5501	* sharing FPU related CR0 bits between the guest and host.
5502	*
5503	* @returns VBox status code.
5504	* @param pVCpu The cross context virtual CPU structure.
5505	* @param pVmxTransient The VMX-transient structure.
5506	*
5507	* @remarks No-long-jump zone!!!
5508	*/
5509	static int hmR0VmxExportGuestCR0(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5510	{
5511	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR0)
5512	{
5513	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5514	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5515
5516	uint64_t fSetCr0 = pVM->hm.s.vmx.Msrs.u64Cr0Fixed0;
5517	uint64_t const fZapCr0 = pVM->hm.s.vmx.Msrs.u64Cr0Fixed1;
5518	if (pVM->hm.s.vmx.fUnrestrictedGuest)
5519	fSetCr0 &= ~(uint64_t)(X86_CR0_PE \| X86_CR0_PG);
5520	else
5521	Assert((fSetCr0 & (X86_CR0_PE \| X86_CR0_PG)) == (X86_CR0_PE \| X86_CR0_PG));
5522
5523	if (!pVmxTransient->fIsNestedGuest)
5524	{
5525	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
5526	uint64_t u64GuestCr0 = pVCpu->cpum.GstCtx.cr0;
5527	uint64_t const u64ShadowCr0 = u64GuestCr0;
5528	Assert(!RT_HI_U32(u64GuestCr0));
5529
5530	/*
5531	* Setup VT-x's view of the guest CR0.
5532	*/
5533	uint32_t uProcCtls = pVmcsInfo->u32ProcCtls;
5534	if (pVM->hm.s.fNestedPaging)
5535	{
5536	if (CPUMIsGuestPagingEnabled(pVCpu))
5537	{
5538	/* The guest has paging enabled, let it access CR3 without causing a VM-exit if supported. */
5539	uProcCtls &= ~( VMX_PROC_CTLS_CR3_LOAD_EXIT
5540	\| VMX_PROC_CTLS_CR3_STORE_EXIT);
5541	}
5542	else
5543	{
5544	/* The guest doesn't have paging enabled, make CR3 access cause a VM-exit to update our shadow. */
5545	uProcCtls \|= VMX_PROC_CTLS_CR3_LOAD_EXIT
5546	\| VMX_PROC_CTLS_CR3_STORE_EXIT;
5547	}
5548
5549	/* If we have unrestricted guest execution, we never have to intercept CR3 reads. */
5550	if (pVM->hm.s.vmx.fUnrestrictedGuest)
5551	uProcCtls &= ~VMX_PROC_CTLS_CR3_STORE_EXIT;
5552	}
5553	else
5554	{
5555	/* Guest CPL 0 writes to its read-only pages should cause a #PF VM-exit. */
5556	u64GuestCr0 \|= X86_CR0_WP;
5557	}
5558
5559	/*
5560	* Guest FPU bits.
5561	*
5562	* Since we pre-load the guest FPU always before VM-entry there is no need to track lazy state
5563	* using CR0.TS.
5564	*
5565	* Intel spec. 23.8 "Restrictions on VMX operation" mentions that CR0.NE bit must always be
5566	* set on the first CPUs to support VT-x and no mention of with regards to UX in VM-entry checks.
5567	*/
5568	u64GuestCr0 \|= X86_CR0_NE;
5569
5570	/* If CR0.NE isn't set, we need to intercept #MF exceptions and report them to the guest differently. */
5571	bool const fInterceptMF = !(u64ShadowCr0 & X86_CR0_NE);
5572
5573	/*
5574	* Update exception intercepts.
5575	*/
5576	uint32_t uXcptBitmap = pVmcsInfo->u32XcptBitmap;
5577	if (pVmcsInfo->RealMode.fRealOnV86Active)
5578	{
5579	Assert(PDMVmmDevHeapIsEnabled(pVM));
5580	Assert(pVM->hm.s.vmx.pRealModeTSS);
5581	uXcptBitmap \|= HMVMX_REAL_MODE_XCPT_MASK;
5582	}
5583	else
5584	{
5585	/* For now, cleared here as mode-switches can happen outside HM/VT-x. See @bugref{7626#c11}. */
5586	uXcptBitmap &= ~HMVMX_REAL_MODE_XCPT_MASK;
5587	if (fInterceptMF)
5588	uXcptBitmap \|= RT_BIT(X86_XCPT_MF);
5589	}
5590
5591	/* Additional intercepts for debugging, define these yourself explicitly. */
5592	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
5593	uXcptBitmap \|= 0
5594	\| RT_BIT(X86_XCPT_BP)
5595	\| RT_BIT(X86_XCPT_DE)
5596	\| RT_BIT(X86_XCPT_NM)
5597	\| RT_BIT(X86_XCPT_TS)
5598	\| RT_BIT(X86_XCPT_UD)
5599	\| RT_BIT(X86_XCPT_NP)
5600	\| RT_BIT(X86_XCPT_SS)
5601	\| RT_BIT(X86_XCPT_GP)
5602	\| RT_BIT(X86_XCPT_PF)
5603	\| RT_BIT(X86_XCPT_MF)
5604	;
5605	#elif defined(HMVMX_ALWAYS_TRAP_PF)
5606	uXcptBitmap \|= RT_BIT(X86_XCPT_PF);
5607	#endif
5608	if (pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv)
5609	uXcptBitmap \|= RT_BIT(X86_XCPT_GP);
5610	Assert(pVM->hm.s.fNestedPaging \|\| (uXcptBitmap & RT_BIT(X86_XCPT_PF)));
5611
5612	/* Apply the hardware specified CR0 fixed bits and enable caching. */
5613	u64GuestCr0 \|= fSetCr0;
5614	u64GuestCr0 &= fZapCr0;
5615	u64GuestCr0 &= ~(uint64_t)(X86_CR0_CD \| X86_CR0_NW);
5616
5617	/* Commit the CR0 and related fields to the guest VMCS. */
5618	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR0, u64GuestCr0); AssertRC(rc);
5619	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, u64ShadowCr0); AssertRC(rc);
5620	if (uProcCtls != pVmcsInfo->u32ProcCtls)
5621	{
5622	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, uProcCtls);
5623	AssertRC(rc);
5624	}
5625	if (uXcptBitmap != pVmcsInfo->u32XcptBitmap)
5626	{
5627	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
5628	AssertRC(rc);
5629	}
5630
5631	/* Update our caches. */
5632	pVmcsInfo->u32ProcCtls = uProcCtls;
5633	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
5634
5635	Log4Func(("cr0=%#RX64 shadow=%#RX64 set=%#RX64 zap=%#RX64\n", u64GuestCr0, u64ShadowCr0, fSetCr0, fZapCr0));
5636	}
5637	else
5638	{
5639	/*
5640	* With nested-guests, we may have extended the guest/host mask here since we
5641	* merged in the outer guest's mask. Thus, the merged mask can include more bits
5642	* (to read from the nested-guest CR0 read-shadow) than the nested hypervisor
5643	* originally supplied. We must copy those bits from the nested-guest CR0 into
5644	* the nested-guest CR0 read-shadow.
5645	*/
5646	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
5647	uint64_t u64GuestCr0 = pVCpu->cpum.GstCtx.cr0;
5648	uint64_t const u64ShadowCr0 = CPUMGetGuestVmxMaskedCr0(&pVCpu->cpum.GstCtx, pVmcsInfo->u64Cr0Mask);
5649	Assert(!RT_HI_U32(u64GuestCr0));
5650	Assert(u64GuestCr0 & X86_CR0_NE);
5651
5652	/* Apply the hardware specified CR0 fixed bits and enable caching. */
5653	u64GuestCr0 \|= fSetCr0;
5654	u64GuestCr0 &= fZapCr0;
5655	u64GuestCr0 &= ~(uint64_t)(X86_CR0_CD \| X86_CR0_NW);
5656
5657	/* Commit the CR0 and CR0 read-shadow to the nested-guest VMCS. */
5658	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR0, u64GuestCr0); AssertRC(rc);
5659	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, u64ShadowCr0); AssertRC(rc);
5660
5661	Log4Func(("cr0=%#RX64 shadow=%#RX64 (set=%#RX64 zap=%#RX64)\n", u64GuestCr0, u64ShadowCr0, fSetCr0, fZapCr0));
5662	}
5663
5664	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR0);
5665	}
5666
5667	return VINF_SUCCESS;
5668	}
5669
5670
5671	/**
5672	* Exports the guest control registers (CR3, CR4) into the guest-state area
5673	* in the VMCS.
5674	*
5675	* @returns VBox strict status code.
5676	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
5677	* without unrestricted guest access and the VMMDev is not presently
5678	* mapped (e.g. EFI32).
5679	*
5680	* @param pVCpu The cross context virtual CPU structure.
5681	* @param pVmxTransient The VMX-transient structure.
5682	*
5683	* @remarks No-long-jump zone!!!
5684	*/
5685	static VBOXSTRICTRC hmR0VmxExportGuestCR3AndCR4(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5686	{
5687	int rc = VINF_SUCCESS;
5688	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5689
5690	/*
5691	* Guest CR2.
5692	* It's always loaded in the assembler code. Nothing to do here.
5693	*/
5694
5695	/*
5696	* Guest CR3.
5697	*/
5698	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR3)
5699	{
5700	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR3);
5701
5702	if (pVM->hm.s.fNestedPaging)
5703	{
5704	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5705	pVmcsInfo->HCPhysEPTP = PGMGetHyperCR3(pVCpu);
5706
5707	/* Validate. See Intel spec. 28.2.2 "EPT Translation Mechanism" and 24.6.11 "Extended-Page-Table Pointer (EPTP)" */
5708	Assert(pVmcsInfo->HCPhysEPTP != NIL_RTHCPHYS);
5709	Assert(!(pVmcsInfo->HCPhysEPTP & UINT64_C(0xfff0000000000000)));
5710	Assert(!(pVmcsInfo->HCPhysEPTP & 0xfff));
5711
5712	/* VMX_EPT_MEMTYPE_WB support is already checked in hmR0VmxSetupTaggedTlb(). */
5713	pVmcsInfo->HCPhysEPTP \|= VMX_EPT_MEMTYPE_WB
5714	\| (VMX_EPT_PAGE_WALK_LENGTH_DEFAULT << VMX_EPT_PAGE_WALK_LENGTH_SHIFT);
5715
5716	/* Validate. See Intel spec. 26.2.1 "Checks on VMX Controls" */
5717	AssertMsg( ((pVmcsInfo->HCPhysEPTP >> 3) & 0x07) == 3 /* Bits 3:5 (EPT page walk length - 1) must be 3. */
5718	&& ((pVmcsInfo->HCPhysEPTP >> 7) & 0x1f) == 0, /* Bits 7:11 MBZ. */
5719	("EPTP %#RX64\n", pVmcsInfo->HCPhysEPTP));
5720	AssertMsg( !((pVmcsInfo->HCPhysEPTP >> 6) & 0x01) /* Bit 6 (EPT accessed & dirty bit). */
5721	\|\| (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EPT_ACCESS_DIRTY),
5722	("EPTP accessed/dirty bit not supported by CPU but set %#RX64\n", pVmcsInfo->HCPhysEPTP));
5723
5724	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, pVmcsInfo->HCPhysEPTP);
5725	AssertRC(rc);
5726
5727	uint64_t u64GuestCr3;
5728	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5729	if ( pVM->hm.s.vmx.fUnrestrictedGuest
5730	\|\| CPUMIsGuestPagingEnabledEx(pCtx))
5731	{
5732	/* If the guest is in PAE mode, pass the PDPEs to VT-x using the VMCS fields. */
5733	if (CPUMIsGuestInPAEModeEx(pCtx))
5734	{
5735	rc = PGMGstGetPaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
5736	AssertRC(rc);
5737	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, pVCpu->hm.s.aPdpes[0].u); AssertRC(rc);
5738	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, pVCpu->hm.s.aPdpes[1].u); AssertRC(rc);
5739	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, pVCpu->hm.s.aPdpes[2].u); AssertRC(rc);
5740	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, pVCpu->hm.s.aPdpes[3].u); AssertRC(rc);
5741	}
5742
5743	/*
5744	* The guest's view of its CR3 is unblemished with nested paging when the
5745	* guest is using paging or we have unrestricted guest execution to handle
5746	* the guest when it's not using paging.
5747	*/
5748	u64GuestCr3 = pCtx->cr3;
5749	}
5750	else
5751	{
5752	/*
5753	* The guest is not using paging, but the CPU (VT-x) has to. While the guest
5754	* thinks it accesses physical memory directly, we use our identity-mapped
5755	* page table to map guest-linear to guest-physical addresses. EPT takes care
5756	* of translating it to host-physical addresses.
5757	*/
5758	RTGCPHYS GCPhys;
5759	Assert(pVM->hm.s.vmx.pNonPagingModeEPTPageTable);
5760
5761	/* We obtain it here every time as the guest could have relocated this PCI region. */
5762	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys);
5763	if (RT_SUCCESS(rc))
5764	{ /* likely */ }
5765	else if (rc == VERR_PDM_DEV_HEAP_R3_TO_GCPHYS)
5766	{
5767	Log4Func(("VERR_PDM_DEV_HEAP_R3_TO_GCPHYS -> VINF_EM_RESCHEDULE_REM\n"));
5768	return VINF_EM_RESCHEDULE_REM; /* We cannot execute now, switch to REM/IEM till the guest maps in VMMDev. */
5769	}
5770	else
5771	AssertMsgFailedReturn(("%Rrc\n", rc), rc);
5772
5773	u64GuestCr3 = GCPhys;
5774	}
5775
5776	Log4Func(("guest_cr3=%#RX64 (GstN)\n", u64GuestCr3));
5777	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR3, u64GuestCr3);
5778	AssertRC(rc);
5779	}
5780	else
5781	{
5782	Assert(!pVmxTransient->fIsNestedGuest);
5783	/* Non-nested paging case, just use the hypervisor's CR3. */
5784	RTHCPHYS const HCPhysGuestCr3 = PGMGetHyperCR3(pVCpu);
5785
5786	Log4Func(("guest_cr3=%#RX64 (HstN)\n", HCPhysGuestCr3));
5787	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR3, HCPhysGuestCr3);
5788	AssertRC(rc);
5789	}
5790
5791	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR3);
5792	}
5793
5794	/*
5795	* Guest CR4.
5796	* ASSUMES this is done everytime we get in from ring-3! (XCR0)
5797	*/
5798	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR4)
5799	{
5800	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5801	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5802
5803	uint64_t const fSetCr4 = pVM->hm.s.vmx.Msrs.u64Cr4Fixed0;
5804	uint64_t const fZapCr4 = pVM->hm.s.vmx.Msrs.u64Cr4Fixed1;
5805
5806	/*
5807	* With nested-guests, we may have extended the guest/host mask here (since we
5808	* merged in the outer guest's mask, see hmR0VmxMergeVmcsNested). This means, the
5809	* mask can include more bits (to read from the nested-guest CR4 read-shadow) than
5810	* the nested hypervisor originally supplied. Thus, we should, in essence, copy
5811	* those bits from the nested-guest CR4 into the nested-guest CR4 read-shadow.
5812	*/
5813	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
5814	uint64_t u64GuestCr4 = pCtx->cr4;
5815	uint64_t const u64ShadowCr4 = !pVmxTransient->fIsNestedGuest
5816	? pCtx->cr4
5817	: CPUMGetGuestVmxMaskedCr4(pCtx, pVmcsInfo->u64Cr4Mask);
5818	Assert(!RT_HI_U32(u64GuestCr4));
5819
5820	/*
5821	* Setup VT-x's view of the guest CR4.
5822	*
5823	* If we're emulating real-mode using virtual-8086 mode, we want to redirect software
5824	* interrupts to the 8086 program interrupt handler. Clear the VME bit (the interrupt
5825	* redirection bitmap is already all 0, see hmR3InitFinalizeR0())
5826	*
5827	* See Intel spec. 20.2 "Software Interrupt Handling Methods While in Virtual-8086 Mode".
5828	*/
5829	if (pVmcsInfo->RealMode.fRealOnV86Active)
5830	{
5831	Assert(pVM->hm.s.vmx.pRealModeTSS);
5832	Assert(PDMVmmDevHeapIsEnabled(pVM));
5833	u64GuestCr4 &= ~(uint64_t)X86_CR4_VME;
5834	}
5835
5836	if (pVM->hm.s.fNestedPaging)
5837	{
5838	if ( !CPUMIsGuestPagingEnabledEx(pCtx)
5839	&& !pVM->hm.s.vmx.fUnrestrictedGuest)
5840	{
5841	/* We use 4 MB pages in our identity mapping page table when the guest doesn't have paging. */
5842	u64GuestCr4 \|= X86_CR4_PSE;
5843	/* Our identity mapping is a 32-bit page directory. */
5844	u64GuestCr4 &= ~(uint64_t)X86_CR4_PAE;
5845	}
5846	/* else use guest CR4.*/
5847	}
5848	else
5849	{
5850	Assert(!pVmxTransient->fIsNestedGuest);
5851
5852	/*
5853	* The shadow paging modes and guest paging modes are different, the shadow is in accordance with the host
5854	* paging mode and thus we need to adjust VT-x's view of CR4 depending on our shadow page tables.
5855	*/
5856	switch (pVCpu->hm.s.enmShadowMode)
5857	{
5858	case PGMMODE_REAL: /* Real-mode. */
5859	case PGMMODE_PROTECTED: /* Protected mode without paging. */
5860	case PGMMODE_32_BIT: /* 32-bit paging. */
5861	{
5862	u64GuestCr4 &= ~(uint64_t)X86_CR4_PAE;
5863	break;
5864	}
5865
5866	case PGMMODE_PAE: /* PAE paging. */
5867	case PGMMODE_PAE_NX: /* PAE paging with NX. */
5868	{
5869	u64GuestCr4 \|= X86_CR4_PAE;
5870	break;
5871	}
5872
5873	case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
5874	case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
5875	{
5876	#ifdef VBOX_WITH_64_BITS_GUESTS
5877	/* For our assumption in hmR0VmxShouldSwapEferMsr. */
5878	Assert(u64GuestCr4 & X86_CR4_PAE);
5879	break;
5880	#endif
5881	}
5882	default:
5883	AssertFailed();
5884	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
5885	}
5886	}
5887
5888	/* Apply the hardware specified CR4 fixed bits (mainly CR4.VMXE). */
5889	u64GuestCr4 \|= fSetCr4;
5890	u64GuestCr4 &= fZapCr4;
5891
5892	/* Commit the CR4 and CR4 read-shadow to the guest VMCS. */
5893	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR4, u64GuestCr4); AssertRC(rc);
5894	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, u64ShadowCr4); AssertRC(rc);
5895
5896	/* Whether to save/load/restore XCR0 during world switch depends on CR4.OSXSAVE and host+guest XCR0. */
5897	pVCpu->hm.s.fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
5898
5899	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR4);
5900
5901	Log4Func(("cr4=%#RX64 shadow=%#RX64 (set=%#RX64 zap=%#RX64)\n", u64GuestCr4, u64ShadowCr4, fSetCr4, fZapCr4));
5902	}
5903	return rc;
5904	}
5905
5906
5907	/**
5908	* Exports the guest debug registers into the guest-state area in the VMCS.
5909	* The guest debug bits are partially shared with the host (e.g. DR6, DR0-3).
5910	*
5911	* This also sets up whether \#DB and MOV DRx accesses cause VM-exits.
5912	*
5913	* @returns VBox status code.
5914	* @param pVCpu The cross context virtual CPU structure.
5915	* @param pVmxTransient The VMX-transient structure.
5916	*
5917	* @remarks No-long-jump zone!!!
5918	*/
5919	static int hmR0VmxExportSharedDebugState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
5920	{
5921	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
5922
5923	/** @todo NSTVMX: Figure out what we want to do with nested-guest instruction
5924	* stepping. */
5925	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5926	if (pVmxTransient->fIsNestedGuest)
5927	{
5928	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, CPUMGetGuestDR7(pVCpu));
5929	AssertRC(rc);
5930
5931	/* Always intercept Mov DRx accesses for the nested-guest for now. */
5932	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_MOV_DR_EXIT;
5933	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
5934	AssertRC(rc);
5935	return VINF_SUCCESS;
5936	}
5937
5938	#ifdef VBOX_STRICT
5939	/* Validate. Intel spec. 26.3.1.1 "Checks on Guest Controls Registers, Debug Registers, MSRs" */
5940	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
5941	{
5942	/* Validate. Intel spec. 17.2 "Debug Registers", recompiler paranoia checks. */
5943	Assert((pVCpu->cpum.GstCtx.dr[7] & (X86_DR7_MBZ_MASK \| X86_DR7_RAZ_MASK)) == 0);
5944	Assert((pVCpu->cpum.GstCtx.dr[7] & X86_DR7_RA1_MASK) == X86_DR7_RA1_MASK);
5945	}
5946	#endif
5947
5948	bool fSteppingDB = false;
5949	bool fInterceptMovDRx = false;
5950	uint32_t uProcCtls = pVmcsInfo->u32ProcCtls;
5951	if (pVCpu->hm.s.fSingleInstruction)
5952	{
5953	/* If the CPU supports the monitor trap flag, use it for single stepping in DBGF and avoid intercepting #DB. */
5954	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5955	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_MONITOR_TRAP_FLAG)
5956	{
5957	uProcCtls \|= VMX_PROC_CTLS_MONITOR_TRAP_FLAG;
5958	Assert(fSteppingDB == false);
5959	}
5960	else
5961	{
5962	pVCpu->cpum.GstCtx.eflags.u32 \|= X86_EFL_TF;
5963	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_GUEST_RFLAGS;
5964	pVCpu->hm.s.fClearTrapFlag = true;
5965	fSteppingDB = true;
5966	}
5967	}
5968
5969	uint64_t u64GuestDr7;
5970	if ( fSteppingDB
5971	\|\| (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK))
5972	{
5973	/*
5974	* Use the combined guest and host DRx values found in the hypervisor register set
5975	* because the hypervisor debugger has breakpoints active or someone is single stepping
5976	* on the host side without a monitor trap flag.
5977	*
5978	* Note! DBGF expects a clean DR6 state before executing guest code.
5979	*/
5980	if (!CPUMIsHyperDebugStateActive(pVCpu))
5981	{
5982	CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
5983	Assert(CPUMIsHyperDebugStateActive(pVCpu));
5984	Assert(!CPUMIsGuestDebugStateActive(pVCpu));
5985	}
5986
5987	/* Update DR7 with the hypervisor value (other DRx registers are handled by CPUM one way or another). */
5988	u64GuestDr7 = CPUMGetHyperDR7(pVCpu);
5989	pVCpu->hm.s.fUsingHyperDR7 = true;
5990	fInterceptMovDRx = true;
5991	}
5992	else
5993	{
5994	/*
5995	* If the guest has enabled debug registers, we need to load them prior to
5996	* executing guest code so they'll trigger at the right time.
5997	*/
5998	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_DR7);
5999	if (pVCpu->cpum.GstCtx.dr[7] & (X86_DR7_ENABLED_MASK \| X86_DR7_GD))
6000	{
6001	if (!CPUMIsGuestDebugStateActive(pVCpu))
6002	{
6003	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
6004	Assert(CPUMIsGuestDebugStateActive(pVCpu));
6005	Assert(!CPUMIsHyperDebugStateActive(pVCpu));
6006	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
6007	}
6008	Assert(!fInterceptMovDRx);
6009	}
6010	else if (!CPUMIsGuestDebugStateActive(pVCpu))
6011	{
6012	/*
6013	* If no debugging enabled, we'll lazy load DR0-3. Unlike on AMD-V, we
6014	* must intercept #DB in order to maintain a correct DR6 guest value, and
6015	* because we need to intercept it to prevent nested #DBs from hanging the
6016	* CPU, we end up always having to intercept it. See hmR0VmxSetupVmcsXcptBitmap().
6017	*/
6018	fInterceptMovDRx = true;
6019	}
6020
6021	/* Update DR7 with the actual guest value. */
6022	u64GuestDr7 = pVCpu->cpum.GstCtx.dr[7];
6023	pVCpu->hm.s.fUsingHyperDR7 = false;
6024	}
6025
6026	if (fInterceptMovDRx)
6027	uProcCtls \|= VMX_PROC_CTLS_MOV_DR_EXIT;
6028	else
6029	uProcCtls &= ~VMX_PROC_CTLS_MOV_DR_EXIT;
6030
6031	/*
6032	* Update the processor-based VM-execution controls with the MOV-DRx intercepts and the
6033	* monitor-trap flag and update our cache.
6034	*/
6035	if (uProcCtls != pVmcsInfo->u32ProcCtls)
6036	{
6037	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, uProcCtls);
6038	AssertRC(rc);
6039	pVmcsInfo->u32ProcCtls = uProcCtls;
6040	}
6041
6042	/*
6043	* Update guest DR7.
6044	*/
6045	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, u64GuestDr7);
6046	AssertRC(rc);
6047
6048	/*
6049	* If we have forced EFLAGS.TF to be set because we're single-stepping in the hypervisor debugger,
6050	* we need to clear interrupt inhibition if any as otherwise it causes a VM-entry failure.
6051	*
6052	* See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
6053	*/
6054	if (fSteppingDB)
6055	{
6056	Assert(pVCpu->hm.s.fSingleInstruction);
6057	Assert(pVCpu->cpum.GstCtx.eflags.Bits.u1TF);
6058
6059	uint32_t fIntrState = 0;
6060	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
6061	AssertRC(rc);
6062
6063	if (fIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
6064	{
6065	fIntrState &= ~(VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
6066	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
6067	AssertRC(rc);
6068	}
6069	}
6070
6071	return VINF_SUCCESS;
6072	}
6073
6074
6075	#ifdef VBOX_STRICT
6076	/**
6077	* Strict function to validate segment registers.
6078	*
6079	* @param pVCpu The cross context virtual CPU structure.
6080	* @param pVmcsInfo The VMCS info. object.
6081	*
6082	* @remarks Will import guest CR0 on strict builds during validation of
6083	* segments.
6084	*/
6085	static void hmR0VmxValidateSegmentRegs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
6086	{
6087	/*
6088	* Validate segment registers. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
6089	*
6090	* The reason we check for attribute value 0 in this function and not just the unusable bit is
6091	* because hmR0VmxExportGuestSegReg() only updates the VMCS' copy of the value with the
6092	* unusable bit and doesn't change the guest-context value.
6093	*/
6094	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6095	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6096	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR0);
6097	if ( !pVM->hm.s.vmx.fUnrestrictedGuest
6098	&& ( !CPUMIsGuestInRealModeEx(pCtx)
6099	&& !CPUMIsGuestInV86ModeEx(pCtx)))
6100	{
6101	/* Protected mode checks */
6102	/* CS */
6103	Assert(pCtx->cs.Attr.n.u1Present);
6104	Assert(!(pCtx->cs.Attr.u & 0xf00));
6105	Assert(!(pCtx->cs.Attr.u & 0xfffe0000));
6106	Assert( (pCtx->cs.u32Limit & 0xfff) == 0xfff
6107	\|\| !(pCtx->cs.Attr.n.u1Granularity));
6108	Assert( !(pCtx->cs.u32Limit & 0xfff00000)
6109	\|\| (pCtx->cs.Attr.n.u1Granularity));
6110	/* CS cannot be loaded with NULL in protected mode. */
6111	Assert(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE)); /** @todo is this really true even for 64-bit CS? */
6112	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
6113	Assert(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl);
6114	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
6115	Assert(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl);
6116	else
6117	AssertMsgFailed(("Invalid CS Type %#x\n", pCtx->cs.Attr.n.u2Dpl));
6118	/* SS */
6119	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
6120	Assert(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL));
6121	if ( !(pCtx->cr0 & X86_CR0_PE)
6122	\|\| pCtx->cs.Attr.n.u4Type == 3)
6123	{
6124	Assert(!pCtx->ss.Attr.n.u2Dpl);
6125	}
6126	if (pCtx->ss.Attr.u && !(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
6127	{
6128	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
6129	Assert(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7);
6130	Assert(pCtx->ss.Attr.n.u1Present);
6131	Assert(!(pCtx->ss.Attr.u & 0xf00));
6132	Assert(!(pCtx->ss.Attr.u & 0xfffe0000));
6133	Assert( (pCtx->ss.u32Limit & 0xfff) == 0xfff
6134	\|\| !(pCtx->ss.Attr.n.u1Granularity));
6135	Assert( !(pCtx->ss.u32Limit & 0xfff00000)
6136	\|\| (pCtx->ss.Attr.n.u1Granularity));
6137	}
6138	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxExportGuestSegReg(). */
6139	if (pCtx->ds.Attr.u && !(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
6140	{
6141	Assert(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6142	Assert(pCtx->ds.Attr.n.u1Present);
6143	Assert(pCtx->ds.Attr.n.u4Type > 11 \|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL));
6144	Assert(!(pCtx->ds.Attr.u & 0xf00));
6145	Assert(!(pCtx->ds.Attr.u & 0xfffe0000));
6146	Assert( (pCtx->ds.u32Limit & 0xfff) == 0xfff
6147	\|\| !(pCtx->ds.Attr.n.u1Granularity));
6148	Assert( !(pCtx->ds.u32Limit & 0xfff00000)
6149	\|\| (pCtx->ds.Attr.n.u1Granularity));
6150	Assert( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6151	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ));
6152	}
6153	if (pCtx->es.Attr.u && !(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
6154	{
6155	Assert(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6156	Assert(pCtx->es.Attr.n.u1Present);
6157	Assert(pCtx->es.Attr.n.u4Type > 11 \|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL));
6158	Assert(!(pCtx->es.Attr.u & 0xf00));
6159	Assert(!(pCtx->es.Attr.u & 0xfffe0000));
6160	Assert( (pCtx->es.u32Limit & 0xfff) == 0xfff
6161	\|\| !(pCtx->es.Attr.n.u1Granularity));
6162	Assert( !(pCtx->es.u32Limit & 0xfff00000)
6163	\|\| (pCtx->es.Attr.n.u1Granularity));
6164	Assert( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6165	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ));
6166	}
6167	if (pCtx->fs.Attr.u && !(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
6168	{
6169	Assert(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6170	Assert(pCtx->fs.Attr.n.u1Present);
6171	Assert(pCtx->fs.Attr.n.u4Type > 11 \|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL));
6172	Assert(!(pCtx->fs.Attr.u & 0xf00));
6173	Assert(!(pCtx->fs.Attr.u & 0xfffe0000));
6174	Assert( (pCtx->fs.u32Limit & 0xfff) == 0xfff
6175	\|\| !(pCtx->fs.Attr.n.u1Granularity));
6176	Assert( !(pCtx->fs.u32Limit & 0xfff00000)
6177	\|\| (pCtx->fs.Attr.n.u1Granularity));
6178	Assert( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6179	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ));
6180	}
6181	if (pCtx->gs.Attr.u && !(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
6182	{
6183	Assert(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6184	Assert(pCtx->gs.Attr.n.u1Present);
6185	Assert(pCtx->gs.Attr.n.u4Type > 11 \|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL));
6186	Assert(!(pCtx->gs.Attr.u & 0xf00));
6187	Assert(!(pCtx->gs.Attr.u & 0xfffe0000));
6188	Assert( (pCtx->gs.u32Limit & 0xfff) == 0xfff
6189	\|\| !(pCtx->gs.Attr.n.u1Granularity));
6190	Assert( !(pCtx->gs.u32Limit & 0xfff00000)
6191	\|\| (pCtx->gs.Attr.n.u1Granularity));
6192	Assert( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6193	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ));
6194	}
6195	/* 64-bit capable CPUs. */
6196	Assert(!RT_HI_U32(pCtx->cs.u64Base));
6197	Assert(!pCtx->ss.Attr.u \|\| !RT_HI_U32(pCtx->ss.u64Base));
6198	Assert(!pCtx->ds.Attr.u \|\| !RT_HI_U32(pCtx->ds.u64Base));
6199	Assert(!pCtx->es.Attr.u \|\| !RT_HI_U32(pCtx->es.u64Base));
6200	}
6201	else if ( CPUMIsGuestInV86ModeEx(pCtx)
6202	\|\| ( CPUMIsGuestInRealModeEx(pCtx)
6203	&& !pVM->hm.s.vmx.fUnrestrictedGuest))
6204	{
6205	/* Real and v86 mode checks. */
6206	/* hmR0VmxExportGuestSegReg() writes the modified in VMCS. We want what we're feeding to VT-x. */
6207	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
6208	if (pVmcsInfo->RealMode.fRealOnV86Active)
6209	{
6210	u32CSAttr = 0xf3; u32SSAttr = 0xf3; u32DSAttr = 0xf3;
6211	u32ESAttr = 0xf3; u32FSAttr = 0xf3; u32GSAttr = 0xf3;
6212	}
6213	else
6214	{
6215	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u; u32DSAttr = pCtx->ds.Attr.u;
6216	u32ESAttr = pCtx->es.Attr.u; u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
6217	}
6218
6219	/* CS */
6220	AssertMsg((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), ("CS base %#x %#x\n", pCtx->cs.u64Base, pCtx->cs.Sel));
6221	Assert(pCtx->cs.u32Limit == 0xffff);
6222	Assert(u32CSAttr == 0xf3);
6223	/* SS */
6224	Assert(pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4);
6225	Assert(pCtx->ss.u32Limit == 0xffff);
6226	Assert(u32SSAttr == 0xf3);
6227	/* DS */
6228	Assert(pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4);
6229	Assert(pCtx->ds.u32Limit == 0xffff);
6230	Assert(u32DSAttr == 0xf3);
6231	/* ES */
6232	Assert(pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4);
6233	Assert(pCtx->es.u32Limit == 0xffff);
6234	Assert(u32ESAttr == 0xf3);
6235	/* FS */
6236	Assert(pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4);
6237	Assert(pCtx->fs.u32Limit == 0xffff);
6238	Assert(u32FSAttr == 0xf3);
6239	/* GS */
6240	Assert(pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4);
6241	Assert(pCtx->gs.u32Limit == 0xffff);
6242	Assert(u32GSAttr == 0xf3);
6243	/* 64-bit capable CPUs. */
6244	Assert(!RT_HI_U32(pCtx->cs.u64Base));
6245	Assert(!u32SSAttr \|\| !RT_HI_U32(pCtx->ss.u64Base));
6246	Assert(!u32DSAttr \|\| !RT_HI_U32(pCtx->ds.u64Base));
6247	Assert(!u32ESAttr \|\| !RT_HI_U32(pCtx->es.u64Base));
6248	}
6249	}
6250	#endif /* VBOX_STRICT */
6251
6252
6253	/**
6254	* Exports a guest segment register into the guest-state area in the VMCS.
6255	*
6256	* @returns VBox status code.
6257	* @param pVCpu The cross context virtual CPU structure.
6258	* @param pVmcsInfo The VMCS info. object.
6259	* @param iSegReg The segment register number (X86_SREG_XXX).
6260	* @param pSelReg Pointer to the segment selector.
6261	*
6262	* @remarks No-long-jump zone!!!
6263	*/
6264	static int hmR0VmxExportGuestSegReg(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, uint8_t iSegReg, PCCPUMSELREG pSelReg)
6265	{
6266	Assert(iSegReg < X86_SREG_COUNT);
6267	uint32_t const idxSel = g_aVmcsSegSel[iSegReg];
6268	uint32_t const idxLimit = g_aVmcsSegLimit[iSegReg];
6269	uint32_t const idxBase = g_aVmcsSegBase[iSegReg];
6270	uint32_t const idxAttr = g_aVmcsSegAttr[iSegReg];
6271
6272	uint32_t u32Access = pSelReg->Attr.u;
6273	if (pVmcsInfo->RealMode.fRealOnV86Active)
6274	{
6275	/* VT-x requires our real-using-v86 mode hack to override the segment access-right bits. */
6276	u32Access = 0xf3;
6277	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
6278	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
6279	RT_NOREF_PV(pVCpu);
6280	}
6281	else
6282	{
6283	/*
6284	* The way to differentiate between whether this is really a null selector or was just
6285	* a selector loaded with 0 in real-mode is using the segment attributes. A selector
6286	* loaded in real-mode with the value 0 is valid and usable in protected-mode and we
6287	* should -not- mark it as an unusable segment. Both the recompiler & VT-x ensures
6288	* NULL selectors loaded in protected-mode have their attribute as 0.
6289	*/
6290	if (!u32Access)
6291	u32Access = X86DESCATTR_UNUSABLE;
6292	}
6293
6294	/* Validate segment access rights. Refer to Intel spec. "26.3.1.2 Checks on Guest Segment Registers". */
6295	AssertMsg((u32Access & X86DESCATTR_UNUSABLE) \|\| (u32Access & X86_SEL_TYPE_ACCESSED),
6296	("Access bit not set for usable segment. idx=%#x sel=%#x attr %#x\n", idxBase, pSelReg, pSelReg->Attr.u));
6297
6298	/*
6299	* Commit it to the VMCS.
6300	*/
6301	int rc = VMXWriteVmcs32(idxSel, pSelReg->Sel); AssertRC(rc);
6302	rc = VMXWriteVmcs32(idxLimit, pSelReg->u32Limit); AssertRC(rc);
6303	rc = VMXWriteVmcsNw(idxBase, pSelReg->u64Base); AssertRC(rc);
6304	rc = VMXWriteVmcs32(idxAttr, u32Access); AssertRC(rc);
6305	return VINF_SUCCESS;
6306	}
6307
6308
6309	/**
6310	* Exports the guest segment registers, GDTR, IDTR, LDTR, TR into the guest-state
6311	* area in the VMCS.
6312	*
6313	* @returns VBox status code.
6314	* @param pVCpu The cross context virtual CPU structure.
6315	* @param pVmxTransient The VMX-transient structure.
6316	*
6317	* @remarks Will import guest CR0 on strict builds during validation of
6318	* segments.
6319	* @remarks No-long-jump zone!!!
6320	*/
6321	static int hmR0VmxExportGuestSegRegsXdtr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6322	{
6323	int rc = VERR_INTERNAL_ERROR_5;
6324	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6325	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6326	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6327
6328	/*
6329	* Guest Segment registers: CS, SS, DS, ES, FS, GS.
6330	*/
6331	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SREG_MASK)
6332	{
6333	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CS)
6334	{
6335	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CS);
6336	if (pVmcsInfo->RealMode.fRealOnV86Active)
6337	pVmcsInfo->RealMode.AttrCS.u = pCtx->cs.Attr.u;
6338	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_CS, &pCtx->cs);
6339	AssertRC(rc);
6340	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CS);
6341	}
6342
6343	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SS)
6344	{
6345	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SS);
6346	if (pVmcsInfo->RealMode.fRealOnV86Active)
6347	pVmcsInfo->RealMode.AttrSS.u = pCtx->ss.Attr.u;
6348	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_SS, &pCtx->ss);
6349	AssertRC(rc);
6350	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SS);
6351	}
6352
6353	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_DS)
6354	{
6355	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_DS);
6356	if (pVmcsInfo->RealMode.fRealOnV86Active)
6357	pVmcsInfo->RealMode.AttrDS.u = pCtx->ds.Attr.u;
6358	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_DS, &pCtx->ds);
6359	AssertRC(rc);
6360	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_DS);
6361	}
6362
6363	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_ES)
6364	{
6365	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_ES);
6366	if (pVmcsInfo->RealMode.fRealOnV86Active)
6367	pVmcsInfo->RealMode.AttrES.u = pCtx->es.Attr.u;
6368	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_ES, &pCtx->es);
6369	AssertRC(rc);
6370	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_ES);
6371	}
6372
6373	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_FS)
6374	{
6375	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_FS);
6376	if (pVmcsInfo->RealMode.fRealOnV86Active)
6377	pVmcsInfo->RealMode.AttrFS.u = pCtx->fs.Attr.u;
6378	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_FS, &pCtx->fs);
6379	AssertRC(rc);
6380	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_FS);
6381	}
6382
6383	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_GS)
6384	{
6385	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GS);
6386	if (pVmcsInfo->RealMode.fRealOnV86Active)
6387	pVmcsInfo->RealMode.AttrGS.u = pCtx->gs.Attr.u;
6388	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_GS, &pCtx->gs);
6389	AssertRC(rc);
6390	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_GS);
6391	}
6392
6393	#ifdef VBOX_STRICT
6394	hmR0VmxValidateSegmentRegs(pVCpu, pVmcsInfo);
6395	#endif
6396	Log4Func(("cs={%#04x base=%#RX64 limit=%#RX32 attr=%#RX32}\n", pCtx->cs.Sel, pCtx->cs.u64Base, pCtx->cs.u32Limit,
6397	pCtx->cs.Attr.u));
6398	}
6399
6400	/*
6401	* Guest TR.
6402	*/
6403	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_TR)
6404	{
6405	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_TR);
6406
6407	/*
6408	* Real-mode emulation using virtual-8086 mode with CR4.VME. Interrupt redirection is
6409	* achieved using the interrupt redirection bitmap (all bits cleared to let the guest
6410	* handle INT-n's) in the TSS. See hmR3InitFinalizeR0() to see how pRealModeTSS is setup.
6411	*/
6412	uint16_t u16Sel;
6413	uint32_t u32Limit;
6414	uint64_t u64Base;
6415	uint32_t u32AccessRights;
6416	if (!pVmcsInfo->RealMode.fRealOnV86Active)
6417	{
6418	u16Sel = pCtx->tr.Sel;
6419	u32Limit = pCtx->tr.u32Limit;
6420	u64Base = pCtx->tr.u64Base;
6421	u32AccessRights = pCtx->tr.Attr.u;
6422	}
6423	else
6424	{
6425	Assert(!pVmxTransient->fIsNestedGuest);
6426	Assert(pVM->hm.s.vmx.pRealModeTSS);
6427	Assert(PDMVmmDevHeapIsEnabled(pVM)); /* Guaranteed by HMCanExecuteGuest() -XXX- what about inner loop changes? */
6428
6429	/* We obtain it here every time as PCI regions could be reconfigured in the guest, changing the VMMDev base. */
6430	RTGCPHYS GCPhys;
6431	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys);
6432	AssertRCReturn(rc, rc);
6433
6434	X86DESCATTR DescAttr;
6435	DescAttr.u = 0;
6436	DescAttr.n.u1Present = 1;
6437	DescAttr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
6438
6439	u16Sel = 0;
6440	u32Limit = HM_VTX_TSS_SIZE;
6441	u64Base = GCPhys;
6442	u32AccessRights = DescAttr.u;
6443	}
6444
6445	/* Validate. */
6446	Assert(!(u16Sel & RT_BIT(2)));
6447	AssertMsg( (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_386_TSS_BUSY
6448	\|\| (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_286_TSS_BUSY, ("TSS is not busy!? %#x\n", u32AccessRights));
6449	AssertMsg(!(u32AccessRights & X86DESCATTR_UNUSABLE), ("TR unusable bit is not clear!? %#x\n", u32AccessRights));
6450	Assert(!(u32AccessRights & RT_BIT(4))); /* System MBZ.*/
6451	Assert(u32AccessRights & RT_BIT(7)); /* Present MB1.*/
6452	Assert(!(u32AccessRights & 0xf00)); /* 11:8 MBZ. */
6453	Assert(!(u32AccessRights & 0xfffe0000)); /* 31:17 MBZ. */
6454	Assert( (u32Limit & 0xfff) == 0xfff
6455	\|\| !(u32AccessRights & RT_BIT(15))); /* Granularity MBZ. */
6456	Assert( !(pCtx->tr.u32Limit & 0xfff00000)
6457	\|\| (u32AccessRights & RT_BIT(15))); /* Granularity MB1. */
6458
6459	rc = VMXWriteVmcs16(VMX_VMCS16_GUEST_TR_SEL, u16Sel); AssertRC(rc);
6460	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, u32Limit); AssertRC(rc);
6461	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, u32AccessRights); AssertRC(rc);
6462	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_TR_BASE, u64Base); AssertRC(rc);
6463
6464	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_TR);
6465	Log4Func(("tr base=%#RX64 limit=%#RX32\n", pCtx->tr.u64Base, pCtx->tr.u32Limit));
6466	}
6467
6468	/*
6469	* Guest GDTR.
6470	*/
6471	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_GDTR)
6472	{
6473	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GDTR);
6474
6475	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, pCtx->gdtr.cbGdt); AssertRC(rc);
6476	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, pCtx->gdtr.pGdt); AssertRC(rc);
6477
6478	/* Validate. */
6479	Assert(!(pCtx->gdtr.cbGdt & 0xffff0000)); /* Bits 31:16 MBZ. */
6480
6481	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_GDTR);
6482	Log4Func(("gdtr base=%#RX64 limit=%#RX32\n", pCtx->gdtr.pGdt, pCtx->gdtr.cbGdt));
6483	}
6484
6485	/*
6486	* Guest LDTR.
6487	*/
6488	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_LDTR)
6489	{
6490	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_LDTR);
6491
6492	/* The unusable bit is specific to VT-x, if it's a null selector mark it as an unusable segment. */
6493	uint32_t u32Access;
6494	if ( !pVmxTransient->fIsNestedGuest
6495	&& !pCtx->ldtr.Attr.u)
6496	u32Access = X86DESCATTR_UNUSABLE;
6497	else
6498	u32Access = pCtx->ldtr.Attr.u;
6499
6500	rc = VMXWriteVmcs16(VMX_VMCS16_GUEST_LDTR_SEL, pCtx->ldtr.Sel); AssertRC(rc);
6501	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, pCtx->ldtr.u32Limit); AssertRC(rc);
6502	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, u32Access); AssertRC(rc);
6503	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_LDTR_BASE, pCtx->ldtr.u64Base); AssertRC(rc);
6504
6505	/* Validate. */
6506	if (!(u32Access & X86DESCATTR_UNUSABLE))
6507	{
6508	Assert(!(pCtx->ldtr.Sel & RT_BIT(2))); /* TI MBZ. */
6509	Assert(pCtx->ldtr.Attr.n.u4Type == 2); /* Type MB2 (LDT). */
6510	Assert(!pCtx->ldtr.Attr.n.u1DescType); /* System MBZ. */
6511	Assert(pCtx->ldtr.Attr.n.u1Present == 1); /* Present MB1. */
6512	Assert(!pCtx->ldtr.Attr.n.u4LimitHigh); /* 11:8 MBZ. */
6513	Assert(!(pCtx->ldtr.Attr.u & 0xfffe0000)); /* 31:17 MBZ. */
6514	Assert( (pCtx->ldtr.u32Limit & 0xfff) == 0xfff
6515	\|\| !pCtx->ldtr.Attr.n.u1Granularity); /* Granularity MBZ. */
6516	Assert( !(pCtx->ldtr.u32Limit & 0xfff00000)
6517	\|\| pCtx->ldtr.Attr.n.u1Granularity); /* Granularity MB1. */
6518	}
6519
6520	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_LDTR);
6521	Log4Func(("ldtr base=%#RX64 limit=%#RX32\n", pCtx->ldtr.u64Base, pCtx->ldtr.u32Limit));
6522	}
6523
6524	/*
6525	* Guest IDTR.
6526	*/
6527	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_IDTR)
6528	{
6529	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_IDTR);
6530
6531	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, pCtx->idtr.cbIdt); AssertRC(rc);
6532	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, pCtx->idtr.pIdt); AssertRC(rc);
6533
6534	/* Validate. */
6535	Assert(!(pCtx->idtr.cbIdt & 0xffff0000)); /* Bits 31:16 MBZ. */
6536
6537	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_IDTR);
6538	Log4Func(("idtr base=%#RX64 limit=%#RX32\n", pCtx->idtr.pIdt, pCtx->idtr.cbIdt));
6539	}
6540
6541	return VINF_SUCCESS;
6542	}
6543
6544
6545	/**
6546	* Exports certain guest MSRs into the VM-entry MSR-load and VM-exit MSR-store
6547	* areas.
6548	*
6549	* These MSRs will automatically be loaded to the host CPU on every successful
6550	* VM-entry and stored from the host CPU on every successful VM-exit.
6551	*
6552	* We creates/updates MSR slots for the host MSRs in the VM-exit MSR-load area. The
6553	* actual host MSR values are not- updated here for performance reasons. See
6554	* hmR0VmxExportHostMsrs().
6555	*
6556	* We also exports the guest sysenter MSRs into the guest-state area in the VMCS.
6557	*
6558	* @returns VBox status code.
6559	* @param pVCpu The cross context virtual CPU structure.
6560	* @param pVmxTransient The VMX-transient structure.
6561	*
6562	* @remarks No-long-jump zone!!!
6563	*/
6564	static int hmR0VmxExportGuestMsrs(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6565	{
6566	AssertPtr(pVCpu);
6567	AssertPtr(pVmxTransient);
6568
6569	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6570	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6571
6572	/*
6573	* MSRs that we use the auto-load/store MSR area in the VMCS.
6574	* For 64-bit hosts, we load/restore them lazily, see hmR0VmxLazyLoadGuestMsrs(),
6575	* nothing to do here. The host MSR values are updated when it's safe in
6576	* hmR0VmxLazySaveHostMsrs().
6577	*
6578	* For nested-guests, the guests MSRs from the VM-entry MSR-load area are already
6579	* loaded (into the guest-CPU context) by the VMLAUNCH/VMRESUME instruction
6580	* emulation. The merged MSR permission bitmap will ensure that we get VM-exits
6581	* for any MSR that are not part of the lazy MSRs so we do not need to place
6582	* those MSRs into the auto-load/store MSR area. Nothing to do here.
6583	*/
6584	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_GUEST_AUTO_MSRS)
6585	{
6586	/* No auto-load/store MSRs currently. */
6587	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_GUEST_AUTO_MSRS);
6588	}
6589
6590	/*
6591	* Guest Sysenter MSRs.
6592	*/
6593	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_MSR_MASK)
6594	{
6595	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SYSENTER_MSRS);
6596
6597	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_CS_MSR)
6598	{
6599	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, pCtx->SysEnter.cs);
6600	AssertRC(rc);
6601	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_CS_MSR);
6602	}
6603
6604	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_EIP_MSR)
6605	{
6606	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SYSENTER_EIP, pCtx->SysEnter.eip);
6607	AssertRC(rc);
6608	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_EIP_MSR);
6609	}
6610
6611	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_ESP_MSR)
6612	{
6613	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SYSENTER_ESP, pCtx->SysEnter.esp);
6614	AssertRC(rc);
6615	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_ESP_MSR);
6616	}
6617	}
6618
6619	/*
6620	* Guest/host EFER MSR.
6621	*/
6622	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_EFER_MSR)
6623	{
6624	/* Whether we are using the VMCS to swap the EFER MSR must have been
6625	determined earlier while exporting VM-entry/VM-exit controls. */
6626	Assert(!(ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_ENTRY_EXIT_CTLS));
6627	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
6628
6629	if (hmR0VmxShouldSwapEferMsr(pVCpu, pVmxTransient))
6630	{
6631	/*
6632	* EFER.LME is written by software, while EFER.LMA is set by the CPU to (CR0.PG & EFER.LME).
6633	* This means a guest can set EFER.LME=1 while CR0.PG=0 and EFER.LMA can remain 0.
6634	* VT-x requires that "IA-32e mode guest" VM-entry control must be identical to EFER.LMA
6635	* and to CR0.PG. Without unrestricted execution, CR0.PG (used for VT-x, not the shadow)
6636	* must always be 1. This forces us to effectively clear both EFER.LMA and EFER.LME until
6637	* the guest has also set CR0.PG=1. Otherwise, we would run into an invalid-guest state
6638	* during VM-entry.
6639	*/
6640	uint64_t uGuestEferMsr = pCtx->msrEFER;
6641	if (!pVM->hm.s.vmx.fUnrestrictedGuest)
6642	{
6643	if (!(pCtx->msrEFER & MSR_K6_EFER_LMA))
6644	uGuestEferMsr &= ~MSR_K6_EFER_LME;
6645	else
6646	Assert((pCtx->msrEFER & (MSR_K6_EFER_LMA \| MSR_K6_EFER_LME)) == (MSR_K6_EFER_LMA \| MSR_K6_EFER_LME));
6647	}
6648
6649	/*
6650	* If the CPU supports VMCS controls for swapping EFER, use it. Otherwise, we have no option
6651	* but to use the auto-load store MSR area in the VMCS for swapping EFER. See @bugref{7368}.
6652	*/
6653	if (pVM->hm.s.vmx.fSupportsVmcsEfer)
6654	{
6655	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_EFER_FULL, uGuestEferMsr);
6656	AssertRC(rc);
6657	}
6658	else
6659	{
6660	/*
6661	* We shall use the auto-load/store MSR area only for loading the EFER MSR but we must
6662	* continue to intercept guest read and write accesses to it, see @bugref{7386#c16}.
6663	*/
6664	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_EFER, uGuestEferMsr,
6665	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6666	AssertRCReturn(rc, rc);
6667	}
6668
6669	Log4Func(("efer=%#RX64 shadow=%#RX64\n", uGuestEferMsr, pCtx->msrEFER));
6670	}
6671	else if (!pVM->hm.s.vmx.fSupportsVmcsEfer)
6672	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_EFER);
6673
6674	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_EFER_MSR);
6675	}
6676
6677	/*
6678	* Other MSRs.
6679	*/
6680	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_OTHER_MSRS)
6681	{
6682	/* Speculation Control (R/W). */
6683	HMVMX_CPUMCTX_ASSERT(pVCpu, HM_CHANGED_GUEST_OTHER_MSRS);
6684	if (pVM->cpum.ro.GuestFeatures.fIbrs)
6685	{
6686	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_IA32_SPEC_CTRL, CPUMGetGuestSpecCtrl(pVCpu),
6687	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6688	AssertRCReturn(rc, rc);
6689	}
6690
6691	/* Last Branch Record. */
6692	if (pVM->hm.s.vmx.fLbr)
6693	{
6694	uint32_t const idFromIpMsrStart = pVM->hm.s.vmx.idLbrFromIpMsrFirst;
6695	uint32_t const idToIpMsrStart = pVM->hm.s.vmx.idLbrToIpMsrFirst;
6696	uint32_t const cLbrStack = pVM->hm.s.vmx.idLbrFromIpMsrLast - pVM->hm.s.vmx.idLbrFromIpMsrFirst + 1;
6697	Assert(cLbrStack <= 32);
6698	for (uint32_t i = 0; i < cLbrStack; i++)
6699	{
6700	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, idFromIpMsrStart + i,
6701	pVmxTransient->pVmcsInfo->au64LbrFromIpMsr[i],
6702	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6703	AssertRCReturn(rc, rc);
6704
6705	/* Some CPUs don't have a Branch-To-IP MSR (P4 and related Xeons). */
6706	if (idToIpMsrStart != 0)
6707	{
6708	rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, idToIpMsrStart + i,
6709	pVmxTransient->pVmcsInfo->au64LbrToIpMsr[i],
6710	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6711	AssertRCReturn(rc, rc);
6712	}
6713	}
6714
6715	/* Add LBR top-of-stack MSR (which contains the index to the most recent record). */
6716	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, pVM->hm.s.vmx.idLbrTosMsr,
6717	pVmxTransient->pVmcsInfo->u64LbrTosMsr, false /* fSetReadWrite */,
6718	false /* fUpdateHostMsr */);
6719	AssertRCReturn(rc, rc);
6720	}
6721
6722	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_OTHER_MSRS);
6723	}
6724
6725	return VINF_SUCCESS;
6726	}
6727
6728
6729	/**
6730	* Wrapper for running the guest code in VT-x.
6731	*
6732	* @returns VBox status code, no informational status codes.
6733	* @param pVCpu The cross context virtual CPU structure.
6734	* @param pVmxTransient The VMX-transient structure.
6735	*
6736	* @remarks No-long-jump zone!!!
6737	*/
6738	DECLINLINE(int) hmR0VmxRunGuest(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6739	{
6740	/* Mark that HM is the keeper of all guest-CPU registers now that we're going to execute guest code. */
6741	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6742	pCtx->fExtrn \|= HMVMX_CPUMCTX_EXTRN_ALL \| CPUMCTX_EXTRN_KEEPER_HM;
6743
6744	/** @todo Add stats for VMRESUME vs VMLAUNCH. */
6745
6746	/*
6747	* 64-bit Windows uses XMM registers in the kernel as the Microsoft compiler expresses
6748	* floating-point operations using SSE instructions. Some XMM registers (XMM6-XMM15) are
6749	* callee-saved and thus the need for this XMM wrapper.
6750	*
6751	* See MSDN "Configuring Programs for 64-bit/x64 Software Conventions / Register Usage".
6752	*/
6753	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6754	bool const fResumeVM = RT_BOOL(pVmcsInfo->fVmcsState & VMX_V_VMCS_LAUNCH_STATE_LAUNCHED);
6755	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6756	#ifdef VBOX_WITH_KERNEL_USING_XMM
6757	int rc = hmR0VMXStartVMWrapXMM(fResumeVM, pCtx, NULL /pvUnused/, pVM, pVCpu, pVmcsInfo->pfnStartVM);
6758	#else
6759	int rc = pVmcsInfo->pfnStartVM(fResumeVM, pCtx, NULL /pvUnused/, pVM, pVCpu);
6760	#endif
6761	AssertMsg(rc <= VINF_SUCCESS, ("%Rrc\n", rc));
6762	return rc;
6763	}
6764
6765
6766	/**
6767	* Reports world-switch error and dumps some useful debug info.
6768	*
6769	* @param pVCpu The cross context virtual CPU structure.
6770	* @param rcVMRun The return code from VMLAUNCH/VMRESUME.
6771	* @param pVmxTransient The VMX-transient structure (only
6772	* exitReason updated).
6773	*/
6774	static void hmR0VmxReportWorldSwitchError(PVMCPUCC pVCpu, int rcVMRun, PVMXTRANSIENT pVmxTransient)
6775	{
6776	Assert(pVCpu);
6777	Assert(pVmxTransient);
6778	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
6779
6780	Log4Func(("VM-entry failure: %Rrc\n", rcVMRun));
6781	switch (rcVMRun)
6782	{
6783	case VERR_VMX_INVALID_VMXON_PTR:
6784	AssertFailed();
6785	break;
6786	case VINF_SUCCESS: /* VMLAUNCH/VMRESUME succeeded but VM-entry failed... yeah, true story. */
6787	case VERR_VMX_UNABLE_TO_START_VM: /* VMLAUNCH/VMRESUME itself failed. */
6788	{
6789	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &pVCpu->hm.s.vmx.LastError.u32ExitReason);
6790	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
6791	AssertRC(rc);
6792	hmR0VmxReadExitQualVmcs(pVmxTransient);
6793
6794	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
6795	/* LastError.idCurrentCpu was already updated in hmR0VmxPreRunGuestCommitted().
6796	Cannot do it here as we may have been long preempted. */
6797
6798	#ifdef VBOX_STRICT
6799	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
6800	Log4(("uExitReason %#RX32 (VmxTransient %#RX16)\n", pVCpu->hm.s.vmx.LastError.u32ExitReason,
6801	pVmxTransient->uExitReason));
6802	Log4(("Exit Qualification %#RX64\n", pVmxTransient->uExitQual));
6803	Log4(("InstrError %#RX32\n", pVCpu->hm.s.vmx.LastError.u32InstrError));
6804	if (pVCpu->hm.s.vmx.LastError.u32InstrError <= HMVMX_INSTR_ERROR_MAX)
6805	Log4(("InstrError Desc. \"%s\"\n", g_apszVmxInstrErrors[pVCpu->hm.s.vmx.LastError.u32InstrError]));
6806	else
6807	Log4(("InstrError Desc. Range exceeded %u\n", HMVMX_INSTR_ERROR_MAX));
6808	Log4(("Entered host CPU %u\n", pVCpu->hm.s.vmx.LastError.idEnteredCpu));
6809	Log4(("Current host CPU %u\n", pVCpu->hm.s.vmx.LastError.idCurrentCpu));
6810
6811	static struct
6812	{
6813	/** Name of the field to log. */
6814	const char *pszName;
6815	/** The VMCS field. */
6816	uint32_t uVmcsField;
6817	/** Whether host support of this field needs to be checked. */
6818	bool fCheckSupport;
6819	} const s_aVmcsFields[] =
6820	{
6821	{ "VMX_VMCS32_CTRL_PIN_EXEC", VMX_VMCS32_CTRL_PIN_EXEC, false },
6822	{ "VMX_VMCS32_CTRL_PROC_EXEC", VMX_VMCS32_CTRL_PROC_EXEC, false },
6823	{ "VMX_VMCS32_CTRL_PROC_EXEC2", VMX_VMCS32_CTRL_PROC_EXEC2, true },
6824	{ "VMX_VMCS32_CTRL_ENTRY", VMX_VMCS32_CTRL_ENTRY, false },
6825	{ "VMX_VMCS32_CTRL_EXIT", VMX_VMCS32_CTRL_EXIT, false },
6826	{ "VMX_VMCS32_CTRL_CR3_TARGET_COUNT", VMX_VMCS32_CTRL_CR3_TARGET_COUNT, false },
6827	{ "VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO", VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, false },
6828	{ "VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE", VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, false },
6829	{ "VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH", VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, false },
6830	{ "VMX_VMCS32_CTRL_TPR_THRESHOLD", VMX_VMCS32_CTRL_TPR_THRESHOLD, false },
6831	{ "VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT", VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, false },
6832	{ "VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT", VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, false },
6833	{ "VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT", VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, false },
6834	{ "VMX_VMCS32_CTRL_EXCEPTION_BITMAP", VMX_VMCS32_CTRL_EXCEPTION_BITMAP, false },
6835	{ "VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK", VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, false },
6836	{ "VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH", VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, false },
6837	{ "VMX_VMCS_CTRL_CR0_MASK", VMX_VMCS_CTRL_CR0_MASK, false },
6838	{ "VMX_VMCS_CTRL_CR0_READ_SHADOW", VMX_VMCS_CTRL_CR0_READ_SHADOW, false },
6839	{ "VMX_VMCS_CTRL_CR4_MASK", VMX_VMCS_CTRL_CR4_MASK, false },
6840	{ "VMX_VMCS_CTRL_CR4_READ_SHADOW", VMX_VMCS_CTRL_CR4_READ_SHADOW, false },
6841	{ "VMX_VMCS64_CTRL_EPTP_FULL", VMX_VMCS64_CTRL_EPTP_FULL, true },
6842	{ "VMX_VMCS_GUEST_RIP", VMX_VMCS_GUEST_RIP, false },
6843	{ "VMX_VMCS_GUEST_RSP", VMX_VMCS_GUEST_RSP, false },
6844	{ "VMX_VMCS_GUEST_RFLAGS", VMX_VMCS_GUEST_RFLAGS, false },
6845	{ "VMX_VMCS16_VPID", VMX_VMCS16_VPID, true, },
6846	{ "VMX_VMCS_HOST_CR0", VMX_VMCS_HOST_CR0, false },
6847	{ "VMX_VMCS_HOST_CR3", VMX_VMCS_HOST_CR3, false },
6848	{ "VMX_VMCS_HOST_CR4", VMX_VMCS_HOST_CR4, false },
6849	/* The order of selector fields below are fixed! */
6850	{ "VMX_VMCS16_HOST_ES_SEL", VMX_VMCS16_HOST_ES_SEL, false },
6851	{ "VMX_VMCS16_HOST_CS_SEL", VMX_VMCS16_HOST_CS_SEL, false },
6852	{ "VMX_VMCS16_HOST_SS_SEL", VMX_VMCS16_HOST_SS_SEL, false },
6853	{ "VMX_VMCS16_HOST_DS_SEL", VMX_VMCS16_HOST_DS_SEL, false },
6854	{ "VMX_VMCS16_HOST_FS_SEL", VMX_VMCS16_HOST_FS_SEL, false },
6855	{ "VMX_VMCS16_HOST_GS_SEL", VMX_VMCS16_HOST_GS_SEL, false },
6856	{ "VMX_VMCS16_HOST_TR_SEL", VMX_VMCS16_HOST_TR_SEL, false },
6857	/* End of ordered selector fields. */
6858	{ "VMX_VMCS_HOST_TR_BASE", VMX_VMCS_HOST_TR_BASE, false },
6859	{ "VMX_VMCS_HOST_GDTR_BASE", VMX_VMCS_HOST_GDTR_BASE, false },
6860	{ "VMX_VMCS_HOST_IDTR_BASE", VMX_VMCS_HOST_IDTR_BASE, false },
6861	{ "VMX_VMCS32_HOST_SYSENTER_CS", VMX_VMCS32_HOST_SYSENTER_CS, false },
6862	{ "VMX_VMCS_HOST_SYSENTER_EIP", VMX_VMCS_HOST_SYSENTER_EIP, false },
6863	{ "VMX_VMCS_HOST_SYSENTER_ESP", VMX_VMCS_HOST_SYSENTER_ESP, false },
6864	{ "VMX_VMCS_HOST_RSP", VMX_VMCS_HOST_RSP, false },
6865	{ "VMX_VMCS_HOST_RIP", VMX_VMCS_HOST_RIP, false }
6866	};
6867
6868	RTGDTR HostGdtr;
6869	ASMGetGDTR(&HostGdtr);
6870
6871	uint32_t const cVmcsFields = RT_ELEMENTS(s_aVmcsFields);
6872	for (uint32_t i = 0; i < cVmcsFields; i++)
6873	{
6874	uint32_t const uVmcsField = s_aVmcsFields[i].uVmcsField;
6875
6876	bool fSupported;
6877	if (!s_aVmcsFields[i].fCheckSupport)
6878	fSupported = true;
6879	else
6880	{
6881	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6882	switch (uVmcsField)
6883	{
6884	case VMX_VMCS64_CTRL_EPTP_FULL: fSupported = pVM->hm.s.fNestedPaging; break;
6885	case VMX_VMCS16_VPID: fSupported = pVM->hm.s.vmx.fVpid; break;
6886	case VMX_VMCS32_CTRL_PROC_EXEC2:
6887	fSupported = RT_BOOL(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS);
6888	break;
6889	default:
6890	AssertMsgFailedReturnVoid(("Failed to provide VMCS field support for %#RX32\n", uVmcsField));
6891	}
6892	}
6893
6894	if (fSupported)
6895	{
6896	uint8_t const uWidth = RT_BF_GET(uVmcsField, VMX_BF_VMCSFIELD_WIDTH);
6897	switch (uWidth)
6898	{
6899	case VMX_VMCSFIELD_WIDTH_16BIT:
6900	{
6901	uint16_t u16Val;
6902	rc = VMXReadVmcs16(uVmcsField, &u16Val);
6903	AssertRC(rc);
6904	Log4(("%-40s = %#RX16\n", s_aVmcsFields[i].pszName, u16Val));
6905
6906	if ( uVmcsField >= VMX_VMCS16_HOST_ES_SEL
6907	&& uVmcsField <= VMX_VMCS16_HOST_TR_SEL)
6908	{
6909	if (u16Val < HostGdtr.cbGdt)
6910	{
6911	/* Order of selectors in s_apszSel is fixed and matches the order in s_aVmcsFields. */
6912	static const char * const s_apszSel[] = { "Host ES", "Host CS", "Host SS", "Host DS",
6913	"Host FS", "Host GS", "Host TR" };
6914	uint8_t const idxSel = RT_BF_GET(uVmcsField, VMX_BF_VMCSFIELD_INDEX);
6915	Assert(idxSel < RT_ELEMENTS(s_apszSel));
6916	PCX86DESCHC pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u16Val & X86_SEL_MASK));
6917	hmR0DumpDescriptor(pDesc, u16Val, s_apszSel[idxSel]);
6918	}
6919	else
6920	Log4((" Selector value exceeds GDT limit!\n"));
6921	}
6922	break;
6923	}
6924
6925	case VMX_VMCSFIELD_WIDTH_32BIT:
6926	{
6927	uint32_t u32Val;
6928	rc = VMXReadVmcs32(uVmcsField, &u32Val);
6929	AssertRC(rc);
6930	Log4(("%-40s = %#RX32\n", s_aVmcsFields[i].pszName, u32Val));
6931	break;
6932	}
6933
6934	case VMX_VMCSFIELD_WIDTH_64BIT:
6935	case VMX_VMCSFIELD_WIDTH_NATURAL:
6936	{
6937	uint64_t u64Val;
6938	rc = VMXReadVmcs64(uVmcsField, &u64Val);
6939	AssertRC(rc);
6940	Log4(("%-40s = %#RX64\n", s_aVmcsFields[i].pszName, u64Val));
6941	break;
6942	}
6943	}
6944	}
6945	}
6946
6947	Log4(("MSR_K6_EFER = %#RX64\n", ASMRdMsr(MSR_K6_EFER)));
6948	Log4(("MSR_K8_CSTAR = %#RX64\n", ASMRdMsr(MSR_K8_CSTAR)));
6949	Log4(("MSR_K8_LSTAR = %#RX64\n", ASMRdMsr(MSR_K8_LSTAR)));
6950	Log4(("MSR_K6_STAR = %#RX64\n", ASMRdMsr(MSR_K6_STAR)));
6951	Log4(("MSR_K8_SF_MASK = %#RX64\n", ASMRdMsr(MSR_K8_SF_MASK)));
6952	Log4(("MSR_K8_KERNEL_GS_BASE = %#RX64\n", ASMRdMsr(MSR_K8_KERNEL_GS_BASE)));
6953	#endif /* VBOX_STRICT */
6954	break;
6955	}
6956
6957	default:
6958	/* Impossible */
6959	AssertMsgFailed(("hmR0VmxReportWorldSwitchError %Rrc (%#x)\n", rcVMRun, rcVMRun));
6960	break;
6961	}
6962	}
6963
6964
6965	/**
6966	* Sets up the usage of TSC-offsetting and updates the VMCS.
6967	*
6968	* If offsetting is not possible, cause VM-exits on RDTSC(P)s. Also sets up the
6969	* VMX-preemption timer.
6970	*
6971	* @returns VBox status code.
6972	* @param pVCpu The cross context virtual CPU structure.
6973	* @param pVmxTransient The VMX-transient structure.
6974	*
6975	* @remarks No-long-jump zone!!!
6976	*/
6977	static void hmR0VmxUpdateTscOffsettingAndPreemptTimer(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
6978	{
6979	bool fOffsettedTsc;
6980	bool fParavirtTsc;
6981	uint64_t uTscOffset;
6982	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6983	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
6984
6985	if (pVM->hm.s.vmx.fUsePreemptTimer)
6986	{
6987	uint64_t cTicksToDeadline = TMCpuTickGetDeadlineAndTscOffset(pVM, pVCpu, &uTscOffset, &fOffsettedTsc, &fParavirtTsc);
6988
6989	/* Make sure the returned values have sane upper and lower boundaries. */
6990	uint64_t u64CpuHz = SUPGetCpuHzFromGipBySetIndex(g_pSUPGlobalInfoPage, pVCpu->iHostCpuSet);
6991	cTicksToDeadline = RT_MIN(cTicksToDeadline, u64CpuHz / 64); /* 1/64th of a second */
6992	cTicksToDeadline = RT_MAX(cTicksToDeadline, u64CpuHz / 2048); /* 1/2048th of a second */
6993	cTicksToDeadline >>= pVM->hm.s.vmx.cPreemptTimerShift;
6994
6995	/** @todo r=ramshankar: We need to find a way to integrate nested-guest
6996	* preemption timers here. We probably need to clamp the preemption timer,
6997	* after converting the timer value to the host. */
6998	uint32_t const cPreemptionTickCount = (uint32_t)RT_MIN(cTicksToDeadline, UINT32_MAX - 16);
6999	int rc = VMXWriteVmcs32(VMX_VMCS32_PREEMPT_TIMER_VALUE, cPreemptionTickCount);
7000	AssertRC(rc);
7001	}
7002	else
7003	fOffsettedTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &uTscOffset, &fParavirtTsc);
7004
7005	if (fParavirtTsc)
7006	{
7007	/* Currently neither Hyper-V nor KVM need to update their paravirt. TSC
7008	information before every VM-entry, hence disable it for performance sake. */
7009	#if 0
7010	int rc = GIMR0UpdateParavirtTsc(pVM, 0 /* u64Offset */);
7011	AssertRC(rc);
7012	#endif
7013	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
7014	}
7015
7016	if ( fOffsettedTsc
7017	&& RT_LIKELY(!pVCpu->hm.s.fDebugWantRdTscExit))
7018	{
7019	if (pVmxTransient->fIsNestedGuest)
7020	uTscOffset = CPUMApplyNestedGuestTscOffset(pVCpu, uTscOffset);
7021	hmR0VmxSetTscOffsetVmcs(pVmcsInfo, uTscOffset);
7022	hmR0VmxRemoveProcCtlsVmcs(pVCpu, pVmxTransient, VMX_PROC_CTLS_RDTSC_EXIT);
7023	}
7024	else
7025	{
7026	/* We can't use TSC-offsetting (non-fixed TSC, warp drive active etc.), VM-exit on RDTSC(P). */
7027	hmR0VmxSetProcCtlsVmcs(pVmxTransient, VMX_PROC_CTLS_RDTSC_EXIT);
7028	}
7029	}
7030
7031
7032	/**
7033	* Gets the IEM exception flags for the specified vector and IDT vectoring /
7034	* VM-exit interruption info type.
7035	*
7036	* @returns The IEM exception flags.
7037	* @param uVector The event vector.
7038	* @param uVmxEventType The VMX event type.
7039	*
7040	* @remarks This function currently only constructs flags required for
7041	* IEMEvaluateRecursiveXcpt and not the complete flags (e.g, error-code
7042	* and CR2 aspects of an exception are not included).
7043	*/
7044	static uint32_t hmR0VmxGetIemXcptFlags(uint8_t uVector, uint32_t uVmxEventType)
7045	{
7046	uint32_t fIemXcptFlags;
7047	switch (uVmxEventType)
7048	{
7049	case VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT:
7050	case VMX_IDT_VECTORING_INFO_TYPE_NMI:
7051	fIemXcptFlags = IEM_XCPT_FLAGS_T_CPU_XCPT;
7052	break;
7053
7054	case VMX_IDT_VECTORING_INFO_TYPE_EXT_INT:
7055	fIemXcptFlags = IEM_XCPT_FLAGS_T_EXT_INT;
7056	break;
7057
7058	case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
7059	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT \| IEM_XCPT_FLAGS_ICEBP_INSTR;
7060	break;
7061
7062	case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT:
7063	{
7064	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
7065	if (uVector == X86_XCPT_BP)
7066	fIemXcptFlags \|= IEM_XCPT_FLAGS_BP_INSTR;
7067	else if (uVector == X86_XCPT_OF)
7068	fIemXcptFlags \|= IEM_XCPT_FLAGS_OF_INSTR;
7069	else
7070	{
7071	fIemXcptFlags = 0;
7072	AssertMsgFailed(("Unexpected vector for software exception. uVector=%#x", uVector));
7073	}
7074	break;
7075	}
7076
7077	case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
7078	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
7079	break;
7080
7081	default:
7082	fIemXcptFlags = 0;
7083	AssertMsgFailed(("Unexpected vector type! uVmxEventType=%#x uVector=%#x", uVmxEventType, uVector));
7084	break;
7085	}
7086	return fIemXcptFlags;
7087	}
7088
7089
7090	/**
7091	* Sets an event as a pending event to be injected into the guest.
7092	*
7093	* @param pVCpu The cross context virtual CPU structure.
7094	* @param u32IntInfo The VM-entry interruption-information field.
7095	* @param cbInstr The VM-entry instruction length in bytes (for
7096	* software interrupts, exceptions and privileged
7097	* software exceptions).
7098	* @param u32ErrCode The VM-entry exception error code.
7099	* @param GCPtrFaultAddress The fault-address (CR2) in case it's a
7100	* page-fault.
7101	*/
7102	DECLINLINE(void) hmR0VmxSetPendingEvent(PVMCPUCC pVCpu, uint32_t u32IntInfo, uint32_t cbInstr, uint32_t u32ErrCode,
7103	RTGCUINTPTR GCPtrFaultAddress)
7104	{
7105	Assert(!pVCpu->hm.s.Event.fPending);
7106	pVCpu->hm.s.Event.fPending = true;
7107	pVCpu->hm.s.Event.u64IntInfo = u32IntInfo;
7108	pVCpu->hm.s.Event.u32ErrCode = u32ErrCode;
7109	pVCpu->hm.s.Event.cbInstr = cbInstr;
7110	pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress;
7111	}
7112
7113
7114	/**
7115	* Sets an external interrupt as pending-for-injection into the VM.
7116	*
7117	* @param pVCpu The cross context virtual CPU structure.
7118	* @param u8Interrupt The external interrupt vector.
7119	*/
7120	DECLINLINE(void) hmR0VmxSetPendingExtInt(PVMCPUCC pVCpu, uint8_t u8Interrupt)
7121	{
7122	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, u8Interrupt)
7123	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
7124	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7125	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7126	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7127	}
7128
7129
7130	/**
7131	* Sets an NMI (\#NMI) exception as pending-for-injection into the VM.
7132	*
7133	* @param pVCpu The cross context virtual CPU structure.
7134	*/
7135	DECLINLINE(void) hmR0VmxSetPendingXcptNmi(PVMCPUCC pVCpu)
7136	{
7137	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_NMI)
7138	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_NMI)
7139	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7140	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7141	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7142	}
7143
7144
7145	/**
7146	* Sets a double-fault (\#DF) exception as pending-for-injection into the VM.
7147	*
7148	* @param pVCpu The cross context virtual CPU structure.
7149	*/
7150	DECLINLINE(void) hmR0VmxSetPendingXcptDF(PVMCPUCC pVCpu)
7151	{
7152	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DF)
7153	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7154	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7155	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7156	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7157	}
7158
7159
7160	/**
7161	* Sets an invalid-opcode (\#UD) exception as pending-for-injection into the VM.
7162	*
7163	* @param pVCpu The cross context virtual CPU structure.
7164	*/
7165	DECLINLINE(void) hmR0VmxSetPendingXcptUD(PVMCPUCC pVCpu)
7166	{
7167	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_UD)
7168	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7169	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7170	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7171	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7172	}
7173
7174
7175	/**
7176	* Sets a debug (\#DB) exception as pending-for-injection into the VM.
7177	*
7178	* @param pVCpu The cross context virtual CPU structure.
7179	*/
7180	DECLINLINE(void) hmR0VmxSetPendingXcptDB(PVMCPUCC pVCpu)
7181	{
7182	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DB)
7183	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7184	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7185	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7186	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7187	}
7188
7189
7190	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7191	/**
7192	* Sets a general-protection (\#GP) exception as pending-for-injection into the VM.
7193	*
7194	* @param pVCpu The cross context virtual CPU structure.
7195	* @param u32ErrCode The error code for the general-protection exception.
7196	*/
7197	DECLINLINE(void) hmR0VmxSetPendingXcptGP(PVMCPUCC pVCpu, uint32_t u32ErrCode)
7198	{
7199	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_GP)
7200	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7201	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7202	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7203	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrCode, 0 / GCPtrFaultAddress */);
7204	}
7205
7206
7207	/**
7208	* Sets a stack (\#SS) exception as pending-for-injection into the VM.
7209	*
7210	* @param pVCpu The cross context virtual CPU structure.
7211	* @param u32ErrCode The error code for the stack exception.
7212	*/
7213	DECLINLINE(void) hmR0VmxSetPendingXcptSS(PVMCPUCC pVCpu, uint32_t u32ErrCode)
7214	{
7215	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_SS)
7216	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7217	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7218	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7219	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrCode, 0 / GCPtrFaultAddress */);
7220	}
7221	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
7222
7223
7224	/**
7225	* Fixes up attributes for the specified segment register.
7226	*
7227	* @param pVCpu The cross context virtual CPU structure.
7228	* @param pSelReg The segment register that needs fixing.
7229	* @param idxSel The VMCS field for the corresponding segment register.
7230	*/
7231	static void hmR0VmxFixUnusableSegRegAttr(PVMCPUCC pVCpu, PCPUMSELREG pSelReg, uint32_t idxSel)
7232	{
7233	Assert(pSelReg->Attr.u & X86DESCATTR_UNUSABLE);
7234
7235	/*
7236	* If VT-x marks the segment as unusable, most other bits remain undefined:
7237	* - For CS the L, D and G bits have meaning.
7238	* - For SS the DPL has meaning (it -is- the CPL for Intel and VBox).
7239	* - For the remaining data segments no bits are defined.
7240	*
7241	* The present bit and the unusable bit has been observed to be set at the
7242	* same time (the selector was supposed to be invalid as we started executing
7243	* a V8086 interrupt in ring-0).
7244	*
7245	* What should be important for the rest of the VBox code, is that the P bit is
7246	* cleared. Some of the other VBox code recognizes the unusable bit, but
7247	* AMD-V certainly don't, and REM doesn't really either. So, to be on the
7248	* safe side here, we'll strip off P and other bits we don't care about. If
7249	* any code breaks because Attr.u != 0 when Sel < 4, it should be fixed.
7250	*
7251	* See Intel spec. 27.3.2 "Saving Segment Registers and Descriptor-Table Registers".
7252	*/
7253	#ifdef VBOX_STRICT
7254	uint32_t const uAttr = pSelReg->Attr.u;
7255	#endif
7256
7257	/* Masking off: X86DESCATTR_P, X86DESCATTR_LIMIT_HIGH, and X86DESCATTR_AVL. The latter two are really irrelevant. */
7258	pSelReg->Attr.u &= X86DESCATTR_UNUSABLE \| X86DESCATTR_L \| X86DESCATTR_D \| X86DESCATTR_G
7259	\| X86DESCATTR_DPL \| X86DESCATTR_TYPE \| X86DESCATTR_DT;
7260
7261	#ifdef VBOX_STRICT
7262	VMMRZCallRing3Disable(pVCpu);
7263	Log4Func(("Unusable %#x: sel=%#x attr=%#x -> %#x\n", idxSel, pSelReg->Sel, uAttr, pSelReg->Attr.u));
7264	# ifdef DEBUG_bird
7265	AssertMsg((uAttr & ~X86DESCATTR_P) == pSelReg->Attr.u,
7266	("%#x: %#x != %#x (sel=%#x base=%#llx limit=%#x)\n",
7267	idxSel, uAttr, pSelReg->Attr.u, pSelReg->Sel, pSelReg->u64Base, pSelReg->u32Limit));
7268	# endif
7269	VMMRZCallRing3Enable(pVCpu);
7270	NOREF(uAttr);
7271	#endif
7272	RT_NOREF2(pVCpu, idxSel);
7273	}
7274
7275
7276	/**
7277	* Imports a guest segment register from the current VMCS into the guest-CPU
7278	* context.
7279	*
7280	* @param pVCpu The cross context virtual CPU structure.
7281	* @param iSegReg The segment register number (X86_SREG_XXX).
7282	*
7283	* @remarks Called with interrupts and/or preemption disabled.
7284	*/
7285	static void hmR0VmxImportGuestSegReg(PVMCPUCC pVCpu, uint8_t iSegReg)
7286	{
7287	Assert(iSegReg < X86_SREG_COUNT);
7288
7289	uint32_t const idxSel = g_aVmcsSegSel[iSegReg];
7290	uint32_t const idxLimit = g_aVmcsSegLimit[iSegReg];
7291	uint32_t const idxAttr = g_aVmcsSegAttr[iSegReg];
7292	uint32_t const idxBase = g_aVmcsSegBase[iSegReg];
7293
7294	uint16_t u16Sel;
7295	uint64_t u64Base;
7296	uint32_t u32Limit, u32Attr;
7297	int rc = VMXReadVmcs16(idxSel, &u16Sel); AssertRC(rc);
7298	rc = VMXReadVmcs32(idxLimit, &u32Limit); AssertRC(rc);
7299	rc = VMXReadVmcs32(idxAttr, &u32Attr); AssertRC(rc);
7300	rc = VMXReadVmcsNw(idxBase, &u64Base); AssertRC(rc);
7301
7302	PCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
7303	pSelReg->Sel = u16Sel;
7304	pSelReg->ValidSel = u16Sel;
7305	pSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
7306	pSelReg->u32Limit = u32Limit;
7307	pSelReg->u64Base = u64Base;
7308	pSelReg->Attr.u = u32Attr;
7309	if (u32Attr & X86DESCATTR_UNUSABLE)
7310	hmR0VmxFixUnusableSegRegAttr(pVCpu, pSelReg, idxSel);
7311	}
7312
7313
7314	/**
7315	* Imports the guest LDTR from the current VMCS into the guest-CPU context.
7316	*
7317	* @param pVCpu The cross context virtual CPU structure.
7318	*
7319	* @remarks Called with interrupts and/or preemption disabled.
7320	*/
7321	static void hmR0VmxImportGuestLdtr(PVMCPUCC pVCpu)
7322	{
7323	uint16_t u16Sel;
7324	uint64_t u64Base;
7325	uint32_t u32Limit, u32Attr;
7326	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_LDTR_SEL, &u16Sel); AssertRC(rc);
7327	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, &u32Limit); AssertRC(rc);
7328	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, &u32Attr); AssertRC(rc);
7329	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_LDTR_BASE, &u64Base); AssertRC(rc);
7330
7331	pVCpu->cpum.GstCtx.ldtr.Sel = u16Sel;
7332	pVCpu->cpum.GstCtx.ldtr.ValidSel = u16Sel;
7333	pVCpu->cpum.GstCtx.ldtr.fFlags = CPUMSELREG_FLAGS_VALID;
7334	pVCpu->cpum.GstCtx.ldtr.u32Limit = u32Limit;
7335	pVCpu->cpum.GstCtx.ldtr.u64Base = u64Base;
7336	pVCpu->cpum.GstCtx.ldtr.Attr.u = u32Attr;
7337	if (u32Attr & X86DESCATTR_UNUSABLE)
7338	hmR0VmxFixUnusableSegRegAttr(pVCpu, &pVCpu->cpum.GstCtx.ldtr, VMX_VMCS16_GUEST_LDTR_SEL);
7339	}
7340
7341
7342	/**
7343	* Imports the guest TR from the current VMCS into the guest-CPU context.
7344	*
7345	* @param pVCpu The cross context virtual CPU structure.
7346	*
7347	* @remarks Called with interrupts and/or preemption disabled.
7348	*/
7349	static void hmR0VmxImportGuestTr(PVMCPUCC pVCpu)
7350	{
7351	uint16_t u16Sel;
7352	uint64_t u64Base;
7353	uint32_t u32Limit, u32Attr;
7354	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_TR_SEL, &u16Sel); AssertRC(rc);
7355	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, &u32Limit); AssertRC(rc);
7356	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, &u32Attr); AssertRC(rc);
7357	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_TR_BASE, &u64Base); AssertRC(rc);
7358
7359	pVCpu->cpum.GstCtx.tr.Sel = u16Sel;
7360	pVCpu->cpum.GstCtx.tr.ValidSel = u16Sel;
7361	pVCpu->cpum.GstCtx.tr.fFlags = CPUMSELREG_FLAGS_VALID;
7362	pVCpu->cpum.GstCtx.tr.u32Limit = u32Limit;
7363	pVCpu->cpum.GstCtx.tr.u64Base = u64Base;
7364	pVCpu->cpum.GstCtx.tr.Attr.u = u32Attr;
7365	/* TR is the only selector that can never be unusable. */
7366	Assert(!(u32Attr & X86DESCATTR_UNUSABLE));
7367	}
7368
7369
7370	/**
7371	* Imports the guest RIP from the VMCS back into the guest-CPU context.
7372	*
7373	* @param pVCpu The cross context virtual CPU structure.
7374	*
7375	* @remarks Called with interrupts and/or preemption disabled, should not assert!
7376	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7377	* instead!!!
7378	*/
7379	static void hmR0VmxImportGuestRip(PVMCPUCC pVCpu)
7380	{
7381	uint64_t u64Val;
7382	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7383	if (pCtx->fExtrn & CPUMCTX_EXTRN_RIP)
7384	{
7385	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RIP, &u64Val);
7386	AssertRC(rc);
7387
7388	pCtx->rip = u64Val;
7389	EMR0HistoryUpdatePC(pVCpu, pCtx->rip, false);
7390	pCtx->fExtrn &= ~CPUMCTX_EXTRN_RIP;
7391	}
7392	}
7393
7394
7395	/**
7396	* Imports the guest RFLAGS from the VMCS back into the guest-CPU context.
7397	*
7398	* @param pVCpu The cross context virtual CPU structure.
7399	* @param pVmcsInfo The VMCS info. object.
7400	*
7401	* @remarks Called with interrupts and/or preemption disabled, should not assert!
7402	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7403	* instead!!!
7404	*/
7405	static void hmR0VmxImportGuestRFlags(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
7406	{
7407	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7408	if (pCtx->fExtrn & CPUMCTX_EXTRN_RFLAGS)
7409	{
7410	uint64_t u64Val;
7411	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RFLAGS, &u64Val);
7412	AssertRC(rc);
7413
7414	pCtx->rflags.u64 = u64Val;
7415	if (pVmcsInfo->RealMode.fRealOnV86Active)
7416	{
7417	pCtx->eflags.Bits.u1VM = 0;
7418	pCtx->eflags.Bits.u2IOPL = pVmcsInfo->RealMode.Eflags.Bits.u2IOPL;
7419	}
7420	pCtx->fExtrn &= ~CPUMCTX_EXTRN_RFLAGS;
7421	}
7422	}
7423
7424
7425	/**
7426	* Imports the guest interruptibility-state from the VMCS back into the guest-CPU
7427	* context.
7428	*
7429	* @param pVCpu The cross context virtual CPU structure.
7430	* @param pVmcsInfo The VMCS info. object.
7431	*
7432	* @remarks Called with interrupts and/or preemption disabled, try not to assert and
7433	* do not log!
7434	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7435	* instead!!!
7436	*/
7437	static void hmR0VmxImportGuestIntrState(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
7438	{
7439	uint32_t u32Val;
7440	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &u32Val); AssertRC(rc);
7441	if (!u32Val)
7442	{
7443	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
7444	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
7445	CPUMSetGuestNmiBlocking(pVCpu, false);
7446	}
7447	else
7448	{
7449	/*
7450	* We must import RIP here to set our EM interrupt-inhibited state.
7451	* We also import RFLAGS as our code that evaluates pending interrupts
7452	* before VM-entry requires it.
7453	*/
7454	hmR0VmxImportGuestRip(pVCpu);
7455	hmR0VmxImportGuestRFlags(pVCpu, pVmcsInfo);
7456
7457	if (u32Val & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS \| VMX_VMCS_GUEST_INT_STATE_BLOCK_STI))
7458	EMSetInhibitInterruptsPC(pVCpu, pVCpu->cpum.GstCtx.rip);
7459	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
7460	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
7461
7462	bool const fNmiBlocking = RT_BOOL(u32Val & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI);
7463	CPUMSetGuestNmiBlocking(pVCpu, fNmiBlocking);
7464	}
7465	}
7466
7467
7468	/**
7469	* Worker for VMXR0ImportStateOnDemand.
7470	*
7471	* @returns VBox status code.
7472	* @param pVCpu The cross context virtual CPU structure.
7473	* @param pVmcsInfo The VMCS info. object.
7474	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
7475	*/
7476	static int hmR0VmxImportGuestState(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint64_t fWhat)
7477	{
7478	int rc = VINF_SUCCESS;
7479	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7480	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7481	uint32_t u32Val;
7482
7483	/*
7484	* Note! This is hack to workaround a mysterious BSOD observed with release builds
7485	* on Windows 10 64-bit hosts. Profile and debug builds are not affected and
7486	* neither are other host platforms.
7487	*
7488	* Committing this temporarily as it prevents BSOD.
7489	*
7490	* Update: This is very likely a compiler optimization bug, see @bugref{9180}.
7491	*/
7492	#ifdef RT_OS_WINDOWS
7493	if (pVM == 0 \|\| pVM == (void *)(uintptr_t)-1)
7494	return VERR_HM_IPE_1;
7495	#endif
7496
7497	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatImportGuestState, x);
7498
7499	/*
7500	* We disable interrupts to make the updating of the state and in particular
7501	* the fExtrn modification atomic wrt to preemption hooks.
7502	*/
7503	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
7504
7505	fWhat &= pCtx->fExtrn;
7506	if (fWhat)
7507	{
7508	do
7509	{
7510	if (fWhat & CPUMCTX_EXTRN_RIP)
7511	hmR0VmxImportGuestRip(pVCpu);
7512
7513	if (fWhat & CPUMCTX_EXTRN_RFLAGS)
7514	hmR0VmxImportGuestRFlags(pVCpu, pVmcsInfo);
7515
7516	if (fWhat & CPUMCTX_EXTRN_HM_VMX_INT_STATE)
7517	hmR0VmxImportGuestIntrState(pVCpu, pVmcsInfo);
7518
7519	if (fWhat & CPUMCTX_EXTRN_RSP)
7520	{
7521	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RSP, &pCtx->rsp);
7522	AssertRC(rc);
7523	}
7524
7525	if (fWhat & CPUMCTX_EXTRN_SREG_MASK)
7526	{
7527	bool const fRealOnV86Active = pVmcsInfo->RealMode.fRealOnV86Active;
7528	if (fWhat & CPUMCTX_EXTRN_CS)
7529	{
7530	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_CS);
7531	hmR0VmxImportGuestRip(pVCpu);
7532	if (fRealOnV86Active)
7533	pCtx->cs.Attr.u = pVmcsInfo->RealMode.AttrCS.u;
7534	EMR0HistoryUpdatePC(pVCpu, pCtx->cs.u64Base + pCtx->rip, true /* fFlattened */);
7535	}
7536	if (fWhat & CPUMCTX_EXTRN_SS)
7537	{
7538	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_SS);
7539	if (fRealOnV86Active)
7540	pCtx->ss.Attr.u = pVmcsInfo->RealMode.AttrSS.u;
7541	}
7542	if (fWhat & CPUMCTX_EXTRN_DS)
7543	{
7544	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_DS);
7545	if (fRealOnV86Active)
7546	pCtx->ds.Attr.u = pVmcsInfo->RealMode.AttrDS.u;
7547	}
7548	if (fWhat & CPUMCTX_EXTRN_ES)
7549	{
7550	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_ES);
7551	if (fRealOnV86Active)
7552	pCtx->es.Attr.u = pVmcsInfo->RealMode.AttrES.u;
7553	}
7554	if (fWhat & CPUMCTX_EXTRN_FS)
7555	{
7556	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_FS);
7557	if (fRealOnV86Active)
7558	pCtx->fs.Attr.u = pVmcsInfo->RealMode.AttrFS.u;
7559	}
7560	if (fWhat & CPUMCTX_EXTRN_GS)
7561	{
7562	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_GS);
7563	if (fRealOnV86Active)
7564	pCtx->gs.Attr.u = pVmcsInfo->RealMode.AttrGS.u;
7565	}
7566	}
7567
7568	if (fWhat & CPUMCTX_EXTRN_TABLE_MASK)
7569	{
7570	if (fWhat & CPUMCTX_EXTRN_LDTR)
7571	hmR0VmxImportGuestLdtr(pVCpu);
7572
7573	if (fWhat & CPUMCTX_EXTRN_GDTR)
7574	{
7575	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, &pCtx->gdtr.pGdt); AssertRC(rc);
7576	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val); AssertRC(rc);
7577	pCtx->gdtr.cbGdt = u32Val;
7578	}
7579
7580	/* Guest IDTR. */
7581	if (fWhat & CPUMCTX_EXTRN_IDTR)
7582	{
7583	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, &pCtx->idtr.pIdt); AssertRC(rc);
7584	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val); AssertRC(rc);
7585	pCtx->idtr.cbIdt = u32Val;
7586	}
7587
7588	/* Guest TR. */
7589	if (fWhat & CPUMCTX_EXTRN_TR)
7590	{
7591	/* Real-mode emulation using virtual-8086 mode has the fake TSS (pRealModeTSS) in TR,
7592	don't need to import that one. */
7593	if (!pVmcsInfo->RealMode.fRealOnV86Active)
7594	hmR0VmxImportGuestTr(pVCpu);
7595	}
7596	}
7597
7598	if (fWhat & CPUMCTX_EXTRN_DR7)
7599	{
7600	if (!pVCpu->hm.s.fUsingHyperDR7)
7601	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_DR7, &pCtx->dr[7]); AssertRC(rc);
7602	}
7603
7604	if (fWhat & CPUMCTX_EXTRN_SYSENTER_MSRS)
7605	{
7606	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SYSENTER_EIP, &pCtx->SysEnter.eip); AssertRC(rc);
7607	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SYSENTER_ESP, &pCtx->SysEnter.esp); AssertRC(rc);
7608	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, &u32Val); AssertRC(rc);
7609	pCtx->SysEnter.cs = u32Val;
7610	}
7611
7612	if (fWhat & CPUMCTX_EXTRN_KERNEL_GS_BASE)
7613	{
7614	if ( pVM->hm.s.fAllow64BitGuests
7615	&& (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
7616	pCtx->msrKERNELGSBASE = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
7617	}
7618
7619	if (fWhat & CPUMCTX_EXTRN_SYSCALL_MSRS)
7620	{
7621	if ( pVM->hm.s.fAllow64BitGuests
7622	&& (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
7623	{
7624	pCtx->msrLSTAR = ASMRdMsr(MSR_K8_LSTAR);
7625	pCtx->msrSTAR = ASMRdMsr(MSR_K6_STAR);
7626	pCtx->msrSFMASK = ASMRdMsr(MSR_K8_SF_MASK);
7627	}
7628	}
7629
7630	if (fWhat & (CPUMCTX_EXTRN_TSC_AUX \| CPUMCTX_EXTRN_OTHER_MSRS))
7631	{
7632	PCVMXAUTOMSR pMsrs = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
7633	uint32_t const cMsrs = pVmcsInfo->cExitMsrStore;
7634	Assert(pMsrs);
7635	Assert(cMsrs <= VMX_MISC_MAX_MSRS(pVM->hm.s.vmx.Msrs.u64Misc));
7636	Assert(sizeof(pMsrs) cMsrs <= X86_PAGE_4K_SIZE);
7637	for (uint32_t i = 0; i < cMsrs; i++)
7638	{
7639	uint32_t const idMsr = pMsrs[i].u32Msr;
7640	switch (idMsr)
7641	{
7642	case MSR_K8_TSC_AUX: CPUMSetGuestTscAux(pVCpu, pMsrs[i].u64Value); break;
7643	case MSR_IA32_SPEC_CTRL: CPUMSetGuestSpecCtrl(pVCpu, pMsrs[i].u64Value); break;
7644	case MSR_K6_EFER: /* Can't be changed without causing a VM-exit */ break;
7645	default:
7646	{
7647	uint32_t idxLbrMsr;
7648	if (pVM->hm.s.vmx.fLbr)
7649	{
7650	if (hmR0VmxIsLbrBranchFromMsr(pVM, idMsr, &idxLbrMsr))
7651	{
7652	Assert(idxLbrMsr < RT_ELEMENTS(pVmcsInfo->au64LbrFromIpMsr));
7653	pVmcsInfo->au64LbrFromIpMsr[idxLbrMsr] = pMsrs[i].u64Value;
7654	break;
7655	}
7656	else if (hmR0VmxIsLbrBranchToMsr(pVM, idMsr, &idxLbrMsr))
7657	{
7658	Assert(idxLbrMsr < RT_ELEMENTS(pVmcsInfo->au64LbrFromIpMsr));
7659	pVmcsInfo->au64LbrToIpMsr[idxLbrMsr] = pMsrs[i].u64Value;
7660	break;
7661	}
7662	else if (idMsr == pVM->hm.s.vmx.idLbrTosMsr)
7663	{
7664	pVmcsInfo->u64LbrTosMsr = pMsrs[i].u64Value;
7665	break;
7666	}
7667	/* Fallthru (no break) */
7668	}
7669	pCtx->fExtrn = 0;
7670	pVCpu->hm.s.u32HMError = pMsrs->u32Msr;
7671	ASMSetFlags(fEFlags);
7672	AssertMsgFailed(("Unexpected MSR in auto-load/store area. idMsr=%#RX32 cMsrs=%u\n", idMsr, cMsrs));
7673	return VERR_HM_UNEXPECTED_LD_ST_MSR;
7674	}
7675	}
7676	}
7677	}
7678
7679	if (fWhat & CPUMCTX_EXTRN_CR_MASK)
7680	{
7681	if (fWhat & CPUMCTX_EXTRN_CR0)
7682	{
7683	uint64_t u64Cr0;
7684	uint64_t u64Shadow;
7685	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64Cr0); AssertRC(rc);
7686	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, &u64Shadow); AssertRC(rc);
7687	#ifndef VBOX_WITH_NESTED_HWVIRT_VMX
7688	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7689	\| (u64Shadow & pVmcsInfo->u64Cr0Mask);
7690	#else
7691	if (!CPUMIsGuestInVmxNonRootMode(pCtx))
7692	{
7693	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7694	\| (u64Shadow & pVmcsInfo->u64Cr0Mask);
7695	}
7696	else
7697	{
7698	/*
7699	* We've merged the guest and nested-guest's CR0 guest/host mask while executing
7700	* the nested-guest using hardware-assisted VMX. Accordingly we need to
7701	* re-construct CR0. See @bugref{9180#c95} for details.
7702	*/
7703	PCVMXVMCSINFO pVmcsInfoGst = &pVCpu->hm.s.vmx.VmcsInfo;
7704	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
7705	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7706	\| (pVmcsNstGst->u64GuestCr0.u & pVmcsNstGst->u64Cr0Mask.u)
7707	\| (u64Shadow & (pVmcsInfoGst->u64Cr0Mask & ~pVmcsNstGst->u64Cr0Mask.u));
7708	}
7709	#endif
7710	VMMRZCallRing3Disable(pVCpu); /* May call into PGM which has Log statements. */
7711	CPUMSetGuestCR0(pVCpu, u64Cr0);
7712	VMMRZCallRing3Enable(pVCpu);
7713	}
7714
7715	if (fWhat & CPUMCTX_EXTRN_CR4)
7716	{
7717	uint64_t u64Cr4;
7718	uint64_t u64Shadow;
7719	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR4, &u64Cr4); AssertRC(rc);
7720	rc \|= VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, &u64Shadow); AssertRC(rc);
7721	#ifndef VBOX_WITH_NESTED_HWVIRT_VMX
7722	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7723	\| (u64Shadow & pVmcsInfo->u64Cr4Mask);
7724	#else
7725	if (!CPUMIsGuestInVmxNonRootMode(pCtx))
7726	{
7727	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7728	\| (u64Shadow & pVmcsInfo->u64Cr4Mask);
7729	}
7730	else
7731	{
7732	/*
7733	* We've merged the guest and nested-guest's CR4 guest/host mask while executing
7734	* the nested-guest using hardware-assisted VMX. Accordingly we need to
7735	* re-construct CR4. See @bugref{9180#c95} for details.
7736	*/
7737	PCVMXVMCSINFO pVmcsInfoGst = &pVCpu->hm.s.vmx.VmcsInfo;
7738	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
7739	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7740	\| (pVmcsNstGst->u64GuestCr4.u & pVmcsNstGst->u64Cr4Mask.u)
7741	\| (u64Shadow & (pVmcsInfoGst->u64Cr4Mask & ~pVmcsNstGst->u64Cr4Mask.u));
7742	}
7743	#endif
7744	pCtx->cr4 = u64Cr4;
7745	}
7746
7747	if (fWhat & CPUMCTX_EXTRN_CR3)
7748	{
7749	/* CR0.PG bit changes are always intercepted, so it's up to date. */
7750	if ( pVM->hm.s.vmx.fUnrestrictedGuest
7751	\|\| ( pVM->hm.s.fNestedPaging
7752	&& CPUMIsGuestPagingEnabledEx(pCtx)))
7753	{
7754	uint64_t u64Cr3;
7755	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR3, &u64Cr3); AssertRC(rc);
7756	if (pCtx->cr3 != u64Cr3)
7757	{
7758	pCtx->cr3 = u64Cr3;
7759	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
7760	}
7761
7762	/* If the guest is in PAE mode, sync back the PDPE's into the guest state.
7763	Note: CR4.PAE, CR0.PG, EFER MSR changes are always intercepted, so they're up to date. */
7764	if (CPUMIsGuestInPAEModeEx(pCtx))
7765	{
7766	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &pVCpu->hm.s.aPdpes[0].u); AssertRC(rc);
7767	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &pVCpu->hm.s.aPdpes[1].u); AssertRC(rc);
7768	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &pVCpu->hm.s.aPdpes[2].u); AssertRC(rc);
7769	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &pVCpu->hm.s.aPdpes[3].u); AssertRC(rc);
7770	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES);
7771	}
7772	}
7773	}
7774	}
7775
7776	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7777	if (fWhat & CPUMCTX_EXTRN_HWVIRT)
7778	{
7779	if ( (pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
7780	&& !CPUMIsGuestInVmxNonRootMode(pCtx))
7781	{
7782	Assert(CPUMIsGuestInVmxRootMode(pCtx));
7783	rc = hmR0VmxCopyShadowToNstGstVmcs(pVCpu, pVmcsInfo);
7784	if (RT_SUCCESS(rc))
7785	{ /* likely */ }
7786	else
7787	break;
7788	}
7789	}
7790	#endif
7791	} while (0);
7792
7793	if (RT_SUCCESS(rc))
7794	{
7795	/* Update fExtrn. */
7796	pCtx->fExtrn &= ~fWhat;
7797
7798	/* If everything has been imported, clear the HM keeper bit. */
7799	if (!(pCtx->fExtrn & HMVMX_CPUMCTX_EXTRN_ALL))
7800	{
7801	pCtx->fExtrn &= ~CPUMCTX_EXTRN_KEEPER_HM;
7802	Assert(!pCtx->fExtrn);
7803	}
7804	}
7805	}
7806	else
7807	AssertMsg(!pCtx->fExtrn \|\| (pCtx->fExtrn & HMVMX_CPUMCTX_EXTRN_ALL), ("%#RX64\n", pCtx->fExtrn));
7808
7809	/*
7810	* Restore interrupts.
7811	*/
7812	ASMSetFlags(fEFlags);
7813
7814	STAM_PROFILE_ADV_STOP(& pVCpu->hm.s.StatImportGuestState, x);
7815
7816	if (RT_SUCCESS(rc))
7817	{ /* likely */ }
7818	else
7819	return rc;
7820
7821	/*
7822	* Honor any pending CR3 updates.
7823	*
7824	* Consider this scenario: VM-exit -> VMMRZCallRing3Enable() -> do stuff that causes a longjmp -> VMXR0CallRing3Callback()
7825	* -> VMMRZCallRing3Disable() -> hmR0VmxImportGuestState() -> Sets VMCPU_FF_HM_UPDATE_CR3 pending -> return from the longjmp
7826	* -> continue with VM-exit handling -> hmR0VmxImportGuestState() and here we are.
7827	*
7828	* The reason for such complicated handling is because VM-exits that call into PGM expect CR3 to be up-to-date and thus
7829	* if any CR3-saves -before- the VM-exit (longjmp) postponed the CR3 update via the force-flag, any VM-exit handler that
7830	* calls into PGM when it re-saves CR3 will end up here and we call PGMUpdateCR3(). This is why the code below should
7831	* -NOT- check if CPUMCTX_EXTRN_CR3 is set!
7832	*
7833	* The longjmp exit path can't check these CR3 force-flags and call code that takes a lock again. We cover for it here.
7834	*/
7835	if (VMMRZCallRing3IsEnabled(pVCpu))
7836	{
7837	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
7838	{
7839	Assert(!(ASMAtomicUoReadU64(&pCtx->fExtrn) & CPUMCTX_EXTRN_CR3));
7840	PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
7841	}
7842
7843	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
7844	PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
7845
7846	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
7847	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
7848	}
7849
7850	return VINF_SUCCESS;
7851	}
7852
7853
7854	/**
7855	* Saves the guest state from the VMCS into the guest-CPU context.
7856	*
7857	* @returns VBox status code.
7858	* @param pVCpu The cross context virtual CPU structure.
7859	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
7860	*/
7861	VMMR0DECL(int) VMXR0ImportStateOnDemand(PVMCPUCC pVCpu, uint64_t fWhat)
7862	{
7863	AssertPtr(pVCpu);
7864	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
7865	return hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fWhat);
7866	}
7867
7868
7869	/**
7870	* Check per-VM and per-VCPU force flag actions that require us to go back to
7871	* ring-3 for one reason or another.
7872	*
7873	* @returns Strict VBox status code (i.e. informational status codes too)
7874	* @retval VINF_SUCCESS if we don't have any actions that require going back to
7875	* ring-3.
7876	* @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync.
7877	* @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware
7878	* interrupts)
7879	* @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires
7880	* all EMTs to be in ring-3.
7881	* @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests.
7882	* @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return
7883	* to the EM loop.
7884	*
7885	* @param pVCpu The cross context virtual CPU structure.
7886	* @param pVmxTransient The VMX-transient structure.
7887	* @param fStepping Whether we are single-stepping the guest using the
7888	* hypervisor debugger.
7889	*
7890	* @remarks This might cause nested-guest VM-exits, caller must check if the guest
7891	* is no longer in VMX non-root mode.
7892	*/
7893	static VBOXSTRICTRC hmR0VmxCheckForceFlags(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, bool fStepping)
7894	{
7895	Assert(VMMRZCallRing3IsEnabled(pVCpu));
7896
7897	/*
7898	* Update pending interrupts into the APIC's IRR.
7899	*/
7900	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UPDATE_APIC))
7901	APICUpdatePendingInterrupts(pVCpu);
7902
7903	/*
7904	* Anything pending? Should be more likely than not if we're doing a good job.
7905	*/
7906	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7907	if ( !fStepping
7908	? !VM_FF_IS_ANY_SET(pVM, VM_FF_HP_R0_PRE_HM_MASK)
7909	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HP_R0_PRE_HM_MASK)
7910	: !VM_FF_IS_ANY_SET(pVM, VM_FF_HP_R0_PRE_HM_STEP_MASK)
7911	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) )
7912	return VINF_SUCCESS;
7913
7914	/* Pending PGM C3 sync. */
7915	if (VMCPU_FF_IS_ANY_SET(pVCpu,VMCPU_FF_PGM_SYNC_CR3 \| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL))
7916	{
7917	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7918	Assert(!(ASMAtomicUoReadU64(&pCtx->fExtrn) & (CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_CR3 \| CPUMCTX_EXTRN_CR4)));
7919	VBOXSTRICTRC rcStrict = PGMSyncCR3(pVCpu, pCtx->cr0, pCtx->cr3, pCtx->cr4,
7920	VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
7921	if (rcStrict != VINF_SUCCESS)
7922	{
7923	AssertRC(VBOXSTRICTRC_VAL(rcStrict));
7924	Log4Func(("PGMSyncCR3 forcing us back to ring-3. rc2=%d\n", VBOXSTRICTRC_VAL(rcStrict)));
7925	return rcStrict;
7926	}
7927	}
7928
7929	/* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */
7930	if ( VM_FF_IS_ANY_SET(pVM, VM_FF_HM_TO_R3_MASK)
7931	\|\| VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
7932	{
7933	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
7934	int rc = RT_LIKELY(!VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_RAW_TO_R3 : VINF_EM_NO_MEMORY;
7935	Log4Func(("HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc));
7936	return rc;
7937	}
7938
7939	/* Pending VM request packets, such as hardware interrupts. */
7940	if ( VM_FF_IS_SET(pVM, VM_FF_REQUEST)
7941	\|\| VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_REQUEST))
7942	{
7943	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchVmReq);
7944	Log4Func(("Pending VM request forcing us back to ring-3\n"));
7945	return VINF_EM_PENDING_REQUEST;
7946	}
7947
7948	/* Pending PGM pool flushes. */
7949	if (VM_FF_IS_SET(pVM, VM_FF_PGM_POOL_FLUSH_PENDING))
7950	{
7951	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPgmPoolFlush);
7952	Log4Func(("PGM pool flush pending forcing us back to ring-3\n"));
7953	return VINF_PGM_POOL_FLUSH_PENDING;
7954	}
7955
7956	/* Pending DMA requests. */
7957	if (VM_FF_IS_SET(pVM, VM_FF_PDM_DMA))
7958	{
7959	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchDma);
7960	Log4Func(("Pending DMA request forcing us back to ring-3\n"));
7961	return VINF_EM_RAW_TO_R3;
7962	}
7963
7964	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7965	/*
7966	* Pending nested-guest events.
7967	*
7968	* Please note the priority of these events are specified and important.
7969	* See Intel spec. 29.4.3.2 "APIC-Write Emulation".
7970	* See Intel spec. 6.9 "Priority Among Simultaneous Exceptions And Interrupts".
7971	*/
7972	if (pVmxTransient->fIsNestedGuest)
7973	{
7974	/* Pending nested-guest APIC-write. */
7975	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_APIC_WRITE))
7976	{
7977	Log4Func(("Pending nested-guest APIC-write\n"));
7978	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitApicWrite(pVCpu);
7979	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
7980	return rcStrict;
7981	}
7982
7983	/* Pending nested-guest monitor-trap flag (MTF). */
7984	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_MTF))
7985	{
7986	Log4Func(("Pending nested-guest MTF\n"));
7987	VBOXSTRICTRC rcStrict = IEMExecVmxVmexit(pVCpu, VMX_EXIT_MTF, 0 /* uExitQual */);
7988	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
7989	return rcStrict;
7990	}
7991
7992	/* Pending nested-guest VMX-preemption timer expired. */
7993	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_PREEMPT_TIMER))
7994	{
7995	Log4Func(("Pending nested-guest MTF\n"));
7996	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitPreemptTimer(pVCpu);
7997	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
7998	return rcStrict;
7999	}
8000	}
8001	#else
8002	NOREF(pVmxTransient);
8003	#endif
8004
8005	return VINF_SUCCESS;
8006	}
8007
8008
8009	/**
8010	* Converts any TRPM trap into a pending HM event. This is typically used when
8011	* entering from ring-3 (not longjmp returns).
8012	*
8013	* @param pVCpu The cross context virtual CPU structure.
8014	*/
8015	static void hmR0VmxTrpmTrapToPendingEvent(PVMCPUCC pVCpu)
8016	{
8017	Assert(TRPMHasTrap(pVCpu));
8018	Assert(!pVCpu->hm.s.Event.fPending);
8019
8020	uint8_t uVector;
8021	TRPMEVENT enmTrpmEvent;
8022	uint32_t uErrCode;
8023	RTGCUINTPTR GCPtrFaultAddress;
8024	uint8_t cbInstr;
8025	bool fIcebp;
8026
8027	int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr, &fIcebp);
8028	AssertRC(rc);
8029
8030	uint32_t u32IntInfo;
8031	u32IntInfo = uVector \| VMX_IDT_VECTORING_INFO_VALID;
8032	u32IntInfo \|= HMTrpmEventTypeToVmxEventType(uVector, enmTrpmEvent, fIcebp);
8033
8034	rc = TRPMResetTrap(pVCpu);
8035	AssertRC(rc);
8036	Log4(("TRPM->HM event: u32IntInfo=%#RX32 enmTrpmEvent=%d cbInstr=%u uErrCode=%#RX32 GCPtrFaultAddress=%#RGv\n",
8037	u32IntInfo, enmTrpmEvent, cbInstr, uErrCode, GCPtrFaultAddress));
8038
8039	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, uErrCode, GCPtrFaultAddress);
8040	}
8041
8042
8043	/**
8044	* Converts the pending HM event into a TRPM trap.
8045	*
8046	* @param pVCpu The cross context virtual CPU structure.
8047	*/
8048	static void hmR0VmxPendingEventToTrpmTrap(PVMCPUCC pVCpu)
8049	{
8050	Assert(pVCpu->hm.s.Event.fPending);
8051
8052	/* If a trap was already pending, we did something wrong! */
8053	Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo /, NULL / pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP);
8054
8055	uint32_t const u32IntInfo = pVCpu->hm.s.Event.u64IntInfo;
8056	uint32_t const uVector = VMX_IDT_VECTORING_INFO_VECTOR(u32IntInfo);
8057	TRPMEVENT const enmTrapType = HMVmxEventTypeToTrpmEventType(u32IntInfo);
8058
8059	Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, enmTrapType));
8060
8061	int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType);
8062	AssertRC(rc);
8063
8064	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(u32IntInfo))
8065	TRPMSetErrorCode(pVCpu, pVCpu->hm.s.Event.u32ErrCode);
8066
8067	if (VMX_IDT_VECTORING_INFO_IS_XCPT_PF(u32IntInfo))
8068	TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress);
8069	else
8070	{
8071	uint8_t const uVectorType = VMX_IDT_VECTORING_INFO_TYPE(u32IntInfo);
8072	switch (uVectorType)
8073	{
8074	case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
8075	TRPMSetTrapDueToIcebp(pVCpu);
8076	RT_FALL_THRU();
8077	case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
8078	case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT:
8079	{
8080	AssertMsg( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
8081	\|\| ( uVector == X86_XCPT_BP /* INT3 */
8082	\|\| uVector == X86_XCPT_OF /* INTO */
8083	\|\| uVector == X86_XCPT_DB /* INT1 (ICEBP) */),
8084	("Invalid vector: uVector=%#x uVectorType=%#x\n", uVector, uVectorType));
8085	TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr);
8086	break;
8087	}
8088	}
8089	}
8090
8091	/* We're now done converting the pending event. */
8092	pVCpu->hm.s.Event.fPending = false;
8093	}
8094
8095
8096	/**
8097	* Sets the interrupt-window exiting control in the VMCS which instructs VT-x to
8098	* cause a VM-exit as soon as the guest is in a state to receive interrupts.
8099	*
8100	* @param pVCpu The cross context virtual CPU structure.
8101	* @param pVmcsInfo The VMCS info. object.
8102	*/
8103	static void hmR0VmxSetIntWindowExitVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
8104	{
8105	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_INT_WINDOW_EXIT)
8106	{
8107	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT))
8108	{
8109	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_INT_WINDOW_EXIT;
8110	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8111	AssertRC(rc);
8112	}
8113	} /* else we will deliver interrupts whenever the guest Vm-exits next and is in a state to receive the interrupt. */
8114	}
8115
8116
8117	/**
8118	* Clears the interrupt-window exiting control in the VMCS.
8119	*
8120	* @param pVmcsInfo The VMCS info. object.
8121	*/
8122	DECLINLINE(void) hmR0VmxClearIntWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8123	{
8124	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT)
8125	{
8126	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_INT_WINDOW_EXIT;
8127	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8128	AssertRC(rc);
8129	}
8130	}
8131
8132
8133	/**
8134	* Sets the NMI-window exiting control in the VMCS which instructs VT-x to
8135	* cause a VM-exit as soon as the guest is in a state to receive NMIs.
8136	*
8137	* @param pVCpu The cross context virtual CPU structure.
8138	* @param pVmcsInfo The VMCS info. object.
8139	*/
8140	static void hmR0VmxSetNmiWindowExitVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
8141	{
8142	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
8143	{
8144	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))
8145	{
8146	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_NMI_WINDOW_EXIT;
8147	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8148	AssertRC(rc);
8149	Log4Func(("Setup NMI-window exiting\n"));
8150	}
8151	} /* else we will deliver NMIs whenever we VM-exit next, even possibly nesting NMIs. Can't be helped on ancient CPUs. */
8152	}
8153
8154
8155	/**
8156	* Clears the NMI-window exiting control in the VMCS.
8157	*
8158	* @param pVmcsInfo The VMCS info. object.
8159	*/
8160	DECLINLINE(void) hmR0VmxClearNmiWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8161	{
8162	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
8163	{
8164	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_NMI_WINDOW_EXIT;
8165	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8166	AssertRC(rc);
8167	}
8168	}
8169
8170
8171	/**
8172	* Does the necessary state syncing before returning to ring-3 for any reason
8173	* (longjmp, preemption, voluntary exits to ring-3) from VT-x.
8174	*
8175	* @returns VBox status code.
8176	* @param pVCpu The cross context virtual CPU structure.
8177	* @param fImportState Whether to import the guest state from the VMCS back
8178	* to the guest-CPU context.
8179	*
8180	* @remarks No-long-jmp zone!!!
8181	*/
8182	static int hmR0VmxLeave(PVMCPUCC pVCpu, bool fImportState)
8183	{
8184	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8185	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8186
8187	RTCPUID const idCpu = RTMpCpuId();
8188	Log4Func(("HostCpuId=%u\n", idCpu));
8189
8190	/*
8191	* !!! IMPORTANT !!!
8192	* If you modify code here, check whether VMXR0CallRing3Callback() needs to be updated too.
8193	*/
8194
8195	/* Save the guest state if necessary. */
8196	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8197	if (fImportState)
8198	{
8199	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
8200	AssertRCReturn(rc, rc);
8201	}
8202
8203	/* Restore host FPU state if necessary. We will resync on next R0 reentry. */
8204	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
8205	Assert(!CPUMIsGuestFPUStateActive(pVCpu));
8206
8207	/* Restore host debug registers if necessary. We will resync on next R0 reentry. */
8208	#ifdef VBOX_STRICT
8209	if (CPUMIsHyperDebugStateActive(pVCpu))
8210	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MOV_DR_EXIT);
8211	#endif
8212	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
8213	Assert(!CPUMIsGuestDebugStateActive(pVCpu) && !CPUMIsGuestDebugStateActivePending(pVCpu));
8214	Assert(!CPUMIsHyperDebugStateActive(pVCpu) && !CPUMIsHyperDebugStateActivePending(pVCpu));
8215
8216	/* Restore host-state bits that VT-x only restores partially. */
8217	if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
8218	&& (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
8219	{
8220	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags, idCpu));
8221	VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
8222	}
8223	pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
8224
8225	/* Restore the lazy host MSRs as we're leaving VT-x context. */
8226	if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
8227	{
8228	/* We shouldn't restore the host MSRs without saving the guest MSRs first. */
8229	if (!fImportState)
8230	{
8231	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_KERNEL_GS_BASE \| CPUMCTX_EXTRN_SYSCALL_MSRS);
8232	AssertRCReturn(rc, rc);
8233	}
8234	hmR0VmxLazyRestoreHostMsrs(pVCpu);
8235	Assert(!pVCpu->hm.s.vmx.fLazyMsrs);
8236	}
8237	else
8238	pVCpu->hm.s.vmx.fLazyMsrs = 0;
8239
8240	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
8241	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = false;
8242
8243	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry);
8244	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatImportGuestState);
8245	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExportGuestState);
8246	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatPreExit);
8247	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitHandling);
8248	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitIO);
8249	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitMovCRx);
8250	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitXcptNmi);
8251	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitVmentry);
8252	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
8253
8254	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
8255
8256	/** @todo This partially defeats the purpose of having preemption hooks.
8257	* The problem is, deregistering the hooks should be moved to a place that
8258	* lasts until the EMT is about to be destroyed not everytime while leaving HM
8259	* context.
8260	*/
8261	int rc = hmR0VmxClearVmcs(pVmcsInfo);
8262	AssertRCReturn(rc, rc);
8263
8264	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8265	/*
8266	* A valid shadow VMCS is made active as part of VM-entry. It is necessary to
8267	* clear a shadow VMCS before allowing that VMCS to become active on another
8268	* logical processor. We may or may not be importing guest state which clears
8269	* it, so cover for it here.
8270	*
8271	* See Intel spec. 24.11.1 "Software Use of Virtual-Machine Control Structures".
8272	*/
8273	if ( pVmcsInfo->pvShadowVmcs
8274	&& pVmcsInfo->fShadowVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
8275	{
8276	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
8277	AssertRCReturn(rc, rc);
8278	}
8279
8280	/*
8281	* Flag that we need to re-export the host state if we switch to this VMCS before
8282	* executing guest or nested-guest code.
8283	*/
8284	pVmcsInfo->idHostCpuState = NIL_RTCPUID;
8285	#endif
8286
8287	Log4Func(("Cleared Vmcs. HostCpuId=%u\n", idCpu));
8288	NOREF(idCpu);
8289	return VINF_SUCCESS;
8290	}
8291
8292
8293	/**
8294	* Leaves the VT-x session.
8295	*
8296	* @returns VBox status code.
8297	* @param pVCpu The cross context virtual CPU structure.
8298	*
8299	* @remarks No-long-jmp zone!!!
8300	*/
8301	static int hmR0VmxLeaveSession(PVMCPUCC pVCpu)
8302	{
8303	HM_DISABLE_PREEMPT(pVCpu);
8304	HMVMX_ASSERT_CPU_SAFE(pVCpu);
8305	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8306	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8307
8308	/* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before
8309	and done this from the VMXR0ThreadCtxCallback(). */
8310	if (!pVCpu->hm.s.fLeaveDone)
8311	{
8312	int rc2 = hmR0VmxLeave(pVCpu, true /* fImportState */);
8313	AssertRCReturnStmt(rc2, HM_RESTORE_PREEMPT(), rc2);
8314	pVCpu->hm.s.fLeaveDone = true;
8315	}
8316	Assert(!pVCpu->cpum.GstCtx.fExtrn);
8317
8318	/*
8319	* !!! IMPORTANT !!!
8320	* If you modify code here, make sure to check whether VMXR0CallRing3Callback() needs to be updated too.
8321	*/
8322
8323	/* Deregister hook now that we've left HM context before re-enabling preemption. */
8324	/** @todo Deregistering here means we need to VMCLEAR always
8325	* (longjmp/exit-to-r3) in VT-x which is not efficient, eliminate need
8326	* for calling VMMR0ThreadCtxHookDisable here! */
8327	VMMR0ThreadCtxHookDisable(pVCpu);
8328
8329	/* Leave HM context. This takes care of local init (term) and deregistering the longjmp-to-ring-3 callback. */
8330	int rc = HMR0LeaveCpu(pVCpu);
8331	HM_RESTORE_PREEMPT();
8332	return rc;
8333	}
8334
8335
8336	/**
8337	* Does the necessary state syncing before doing a longjmp to ring-3.
8338	*
8339	* @returns VBox status code.
8340	* @param pVCpu The cross context virtual CPU structure.
8341	*
8342	* @remarks No-long-jmp zone!!!
8343	*/
8344	DECLINLINE(int) hmR0VmxLongJmpToRing3(PVMCPUCC pVCpu)
8345	{
8346	return hmR0VmxLeaveSession(pVCpu);
8347	}
8348
8349
8350	/**
8351	* Take necessary actions before going back to ring-3.
8352	*
8353	* An action requires us to go back to ring-3. This function does the necessary
8354	* steps before we can safely return to ring-3. This is not the same as longjmps
8355	* to ring-3, this is voluntary and prepares the guest so it may continue
8356	* executing outside HM (recompiler/IEM).
8357	*
8358	* @returns VBox status code.
8359	* @param pVCpu The cross context virtual CPU structure.
8360	* @param rcExit The reason for exiting to ring-3. Can be
8361	* VINF_VMM_UNKNOWN_RING3_CALL.
8362	*/
8363	static int hmR0VmxExitToRing3(PVMCPUCC pVCpu, VBOXSTRICTRC rcExit)
8364	{
8365	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
8366
8367	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8368	if (RT_UNLIKELY(rcExit == VERR_VMX_INVALID_VMCS_PTR))
8369	{
8370	VMXGetCurrentVmcs(&pVCpu->hm.s.vmx.LastError.HCPhysCurrentVmcs);
8371	pVCpu->hm.s.vmx.LastError.u32VmcsRev = (uint32_t )pVmcsInfo->pvVmcs;
8372	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
8373	/* LastError.idCurrentCpu was updated in hmR0VmxPreRunGuestCommitted(). */
8374	}
8375
8376	/* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */
8377	VMMRZCallRing3Disable(pVCpu);
8378	Log4Func(("rcExit=%d\n", VBOXSTRICTRC_VAL(rcExit)));
8379
8380	/*
8381	* Convert any pending HM events back to TRPM due to premature exits to ring-3.
8382	* We need to do this only on returns to ring-3 and not for longjmps to ring3.
8383	*
8384	* This is because execution may continue from ring-3 and we would need to inject
8385	* the event from there (hence place it back in TRPM).
8386	*/
8387	if (pVCpu->hm.s.Event.fPending)
8388	{
8389	hmR0VmxPendingEventToTrpmTrap(pVCpu);
8390	Assert(!pVCpu->hm.s.Event.fPending);
8391
8392	/* Clear the events from the VMCS. */
8393	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, 0); AssertRC(rc);
8394	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, 0); AssertRC(rc);
8395	}
8396	#ifdef VBOX_STRICT
8397	/*
8398	* We check for rcExit here since for errors like VERR_VMX_UNABLE_TO_START_VM (which are
8399	* fatal), we don't care about verifying duplicate injection of events. Errors like
8400	* VERR_EM_INTERPRET are converted to their VINF_* counterparts -prior- to calling this
8401	* function so those should and will be checked below.
8402	*/
8403	else if (RT_SUCCESS(rcExit))
8404	{
8405	/*
8406	* Ensure we don't accidentally clear a pending HM event without clearing the VMCS.
8407	* This can be pretty hard to debug otherwise, interrupts might get injected twice
8408	* occasionally, see @bugref{9180#c42}.
8409	*
8410	* However, if the VM-entry failed, any VM entry-interruption info. field would
8411	* be left unmodified as the event would not have been injected to the guest. In
8412	* such cases, don't assert, we're not going to continue guest execution anyway.
8413	*/
8414	uint32_t uExitReason;
8415	uint32_t uEntryIntInfo;
8416	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
8417	rc \|= VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &uEntryIntInfo);
8418	AssertRC(rc);
8419	AssertMsg(VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason) \|\| !VMX_ENTRY_INT_INFO_IS_VALID(uEntryIntInfo),
8420	("uExitReason=%#RX32 uEntryIntInfo=%#RX32 rcExit=%d\n", uExitReason, uEntryIntInfo, VBOXSTRICTRC_VAL(rcExit)));
8421	}
8422	#endif
8423
8424	/*
8425	* Clear the interrupt-window and NMI-window VMCS controls as we could have got
8426	* a VM-exit with higher priority than interrupt-window or NMI-window VM-exits
8427	* (e.g. TPR below threshold).
8428	*/
8429	if (!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
8430	{
8431	hmR0VmxClearIntWindowExitVmcs(pVmcsInfo);
8432	hmR0VmxClearNmiWindowExitVmcs(pVmcsInfo);
8433	}
8434
8435	/* If we're emulating an instruction, we shouldn't have any TRPM traps pending
8436	and if we're injecting an event we should have a TRPM trap pending. */
8437	AssertMsg(rcExit != VINF_EM_RAW_INJECT_TRPM_EVENT \|\| TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
8438	#ifndef DEBUG_bird /* Triggered after firing an NMI against NT4SP1, possibly a triple fault in progress. */
8439	AssertMsg(rcExit != VINF_EM_RAW_EMULATE_INSTR \|\| !TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
8440	#endif
8441
8442	/* Save guest state and restore host state bits. */
8443	int rc = hmR0VmxLeaveSession(pVCpu);
8444	AssertRCReturn(rc, rc);
8445	STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
8446
8447	/* Thread-context hooks are unregistered at this point!!! */
8448	/* Ring-3 callback notifications are unregistered at this point!!! */
8449
8450	/* Sync recompiler state. */
8451	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3);
8452	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR
8453	\| CPUM_CHANGED_LDTR
8454	\| CPUM_CHANGED_GDTR
8455	\| CPUM_CHANGED_IDTR
8456	\| CPUM_CHANGED_TR
8457	\| CPUM_CHANGED_HIDDEN_SEL_REGS);
8458	if ( pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging
8459	&& CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx))
8460	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH);
8461
8462	Assert(!pVCpu->hm.s.fClearTrapFlag);
8463
8464	/* Update the exit-to-ring 3 reason. */
8465	pVCpu->hm.s.rcLastExitToR3 = VBOXSTRICTRC_VAL(rcExit);
8466
8467	/* On our way back from ring-3 reload the guest state if there is a possibility of it being changed. */
8468	if ( rcExit != VINF_EM_RAW_INTERRUPT
8469	\|\| CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
8470	{
8471	Assert(!(pVCpu->cpum.GstCtx.fExtrn & HMVMX_CPUMCTX_EXTRN_ALL));
8472	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
8473	}
8474
8475	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3);
8476	VMMRZCallRing3Enable(pVCpu);
8477	return rc;
8478	}
8479
8480
8481	/**
8482	* VMMRZCallRing3() callback wrapper which saves the guest state before we
8483	* longjump to ring-3 and possibly get preempted.
8484	*
8485	* @returns VBox status code.
8486	* @param pVCpu The cross context virtual CPU structure.
8487	* @param enmOperation The operation causing the ring-3 longjump.
8488	*/
8489	VMMR0DECL(int) VMXR0CallRing3Callback(PVMCPUCC pVCpu, VMMCALLRING3 enmOperation)
8490	{
8491	if (enmOperation == VMMCALLRING3_VM_R0_ASSERTION)
8492	{
8493	/*
8494	* !!! IMPORTANT !!!
8495	* If you modify code here, check whether hmR0VmxLeave() and hmR0VmxLeaveSession() needs to be updated too.
8496	* This is a stripped down version which gets out ASAP, trying to not trigger any further assertions.
8497	*/
8498	VMMRZCallRing3RemoveNotification(pVCpu);
8499	VMMRZCallRing3Disable(pVCpu);
8500	HM_DISABLE_PREEMPT(pVCpu);
8501
8502	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8503	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
8504	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
8505	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
8506
8507	/* Restore host-state bits that VT-x only restores partially. */
8508	if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
8509	&& (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
8510	VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
8511	pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
8512
8513	/* Restore the lazy host MSRs as we're leaving VT-x context. */
8514	if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
8515	hmR0VmxLazyRestoreHostMsrs(pVCpu);
8516
8517	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
8518	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = false;
8519	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
8520
8521	/* Clear the current VMCS data back to memory (shadow VMCS if any would have been
8522	cleared as part of importing the guest state above. */
8523	hmR0VmxClearVmcs(pVmcsInfo);
8524
8525	/** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
8526	VMMR0ThreadCtxHookDisable(pVCpu);
8527
8528	/* Leave HM context. This takes care of local init (term). */
8529	HMR0LeaveCpu(pVCpu);
8530	HM_RESTORE_PREEMPT();
8531	return VINF_SUCCESS;
8532	}
8533
8534	Assert(pVCpu);
8535	Assert(VMMRZCallRing3IsEnabled(pVCpu));
8536	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
8537
8538	VMMRZCallRing3Disable(pVCpu);
8539	Assert(VMMR0IsLogFlushDisabled(pVCpu));
8540
8541	Log4Func(("-> hmR0VmxLongJmpToRing3 enmOperation=%d\n", enmOperation));
8542
8543	int rc = hmR0VmxLongJmpToRing3(pVCpu);
8544	AssertRCReturn(rc, rc);
8545
8546	VMMRZCallRing3Enable(pVCpu);
8547	return VINF_SUCCESS;
8548	}
8549
8550
8551	/**
8552	* Pushes a 2-byte value onto the real-mode (in virtual-8086 mode) guest's
8553	* stack.
8554	*
8555	* @returns Strict VBox status code (i.e. informational status codes too).
8556	* @retval VINF_EM_RESET if pushing a value to the stack caused a triple-fault.
8557	* @param pVCpu The cross context virtual CPU structure.
8558	* @param uValue The value to push to the guest stack.
8559	*/
8560	static VBOXSTRICTRC hmR0VmxRealModeGuestStackPush(PVMCPUCC pVCpu, uint16_t uValue)
8561	{
8562	/*
8563	* The stack limit is 0xffff in real-on-virtual 8086 mode. Real-mode with weird stack limits cannot be run in
8564	* virtual 8086 mode in VT-x. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
8565	* See Intel Instruction reference for PUSH and Intel spec. 22.33.1 "Segment Wraparound".
8566	*/
8567	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8568	if (pCtx->sp == 1)
8569	return VINF_EM_RESET;
8570	pCtx->sp -= sizeof(uint16_t); /* May wrap around which is expected behaviour. */
8571	int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), pCtx->ss.u64Base + pCtx->sp, &uValue, sizeof(uint16_t));
8572	AssertRC(rc);
8573	return rc;
8574	}
8575
8576
8577	/**
8578	* Injects an event into the guest upon VM-entry by updating the relevant fields
8579	* in the VM-entry area in the VMCS.
8580	*
8581	* @returns Strict VBox status code (i.e. informational status codes too).
8582	* @retval VINF_SUCCESS if the event is successfully injected into the VMCS.
8583	* @retval VINF_EM_RESET if event injection resulted in a triple-fault.
8584	*
8585	* @param pVCpu The cross context virtual CPU structure.
8586	* @param pVmxTransient The VMX-transient structure.
8587	* @param pEvent The event being injected.
8588	* @param pfIntrState Pointer to the VT-x guest-interruptibility-state. This
8589	* will be updated if necessary. This cannot not be NULL.
8590	* @param fStepping Whether we're single-stepping guest execution and should
8591	* return VINF_EM_DBG_STEPPED if the event is injected
8592	* directly (registers modified by us, not by hardware on
8593	* VM-entry).
8594	*/
8595	static VBOXSTRICTRC hmR0VmxInjectEventVmcs(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, PCHMEVENT pEvent, bool fStepping,
8596	uint32_t *pfIntrState)
8597	{
8598	/* Intel spec. 24.8.3 "VM-Entry Controls for Event Injection" specifies the interruption-information field to be 32-bits. */
8599	AssertMsg(!RT_HI_U32(pEvent->u64IntInfo), ("%#RX64\n", pEvent->u64IntInfo));
8600	Assert(pfIntrState);
8601
8602	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8603	uint32_t u32IntInfo = pEvent->u64IntInfo;
8604	uint32_t const u32ErrCode = pEvent->u32ErrCode;
8605	uint32_t const cbInstr = pEvent->cbInstr;
8606	RTGCUINTPTR const GCPtrFault = pEvent->GCPtrFaultAddress;
8607	uint8_t const uVector = VMX_ENTRY_INT_INFO_VECTOR(u32IntInfo);
8608	uint32_t const uIntType = VMX_ENTRY_INT_INFO_TYPE(u32IntInfo);
8609
8610	#ifdef VBOX_STRICT
8611	/*
8612	* Validate the error-code-valid bit for hardware exceptions.
8613	* No error codes for exceptions in real-mode.
8614	*
8615	* See Intel spec. 20.1.4 "Interrupt and Exception Handling"
8616	*/
8617	if ( uIntType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT
8618	&& !CPUMIsGuestInRealModeEx(pCtx))
8619	{
8620	switch (uVector)
8621	{
8622	case X86_XCPT_PF:
8623	case X86_XCPT_DF:
8624	case X86_XCPT_TS:
8625	case X86_XCPT_NP:
8626	case X86_XCPT_SS:
8627	case X86_XCPT_GP:
8628	case X86_XCPT_AC:
8629	AssertMsg(VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(u32IntInfo),
8630	("Error-code-valid bit not set for exception that has an error code uVector=%#x\n", uVector));
8631	RT_FALL_THRU();
8632	default:
8633	break;
8634	}
8635	}
8636
8637	/* Cannot inject an NMI when block-by-MOV SS is in effect. */
8638	Assert( uIntType != VMX_EXIT_INT_INFO_TYPE_NMI
8639	\|\| !(*pfIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
8640	#endif
8641
8642	if ( uIntType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT
8643	\|\| uIntType == VMX_EXIT_INT_INFO_TYPE_NMI
8644	\|\| uIntType == VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT
8645	\|\| uIntType == VMX_EXIT_INT_INFO_TYPE_SW_XCPT)
8646	{
8647	Assert(uVector <= X86_XCPT_LAST);
8648	Assert(uIntType != VMX_EXIT_INT_INFO_TYPE_NMI \|\| uVector == X86_XCPT_NMI);
8649	Assert(uIntType != VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT \|\| uVector == X86_XCPT_DB);
8650	STAM_COUNTER_INC(&pVCpu->hm.s.paStatInjectedXcptsR0[uVector]);
8651	}
8652	else
8653	STAM_COUNTER_INC(&pVCpu->hm.s.paStatInjectedIrqsR0[uVector & MASK_INJECT_IRQ_STAT]);
8654
8655	/*
8656	* Hardware interrupts & exceptions cannot be delivered through the software interrupt
8657	* redirection bitmap to the real mode task in virtual-8086 mode. We must jump to the
8658	* interrupt handler in the (real-mode) guest.
8659	*
8660	* See Intel spec. 20.3 "Interrupt and Exception handling in Virtual-8086 Mode".
8661	* See Intel spec. 20.1.4 "Interrupt and Exception Handling" for real-mode interrupt handling.
8662	*/
8663	if (CPUMIsGuestInRealModeEx(pCtx)) /* CR0.PE bit changes are always intercepted, so it's up to date. */
8664	{
8665	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest)
8666	{
8667	/*
8668	* For CPUs with unrestricted guest execution enabled and with the guest
8669	* in real-mode, we must not set the deliver-error-code bit.
8670	*
8671	* See Intel spec. 26.2.1.3 "VM-Entry Control Fields".
8672	*/
8673	u32IntInfo &= ~VMX_ENTRY_INT_INFO_ERROR_CODE_VALID;
8674	}
8675	else
8676	{
8677	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
8678	Assert(PDMVmmDevHeapIsEnabled(pVM));
8679	Assert(pVM->hm.s.vmx.pRealModeTSS);
8680	Assert(!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
8681
8682	/* We require RIP, RSP, RFLAGS, CS, IDTR, import them. */
8683	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
8684	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_TABLE_MASK
8685	\| CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_RFLAGS);
8686	AssertRCReturn(rc2, rc2);
8687
8688	/* Check if the interrupt handler is present in the IVT (real-mode IDT). IDT limit is (4N - 1). */
8689	size_t const cbIdtEntry = sizeof(X86IDTR16);
8690	if (uVector * cbIdtEntry + (cbIdtEntry - 1) > pCtx->idtr.cbIdt)
8691	{
8692	/* If we are trying to inject a #DF with no valid IDT entry, return a triple-fault. */
8693	if (uVector == X86_XCPT_DF)
8694	return VINF_EM_RESET;
8695
8696	/* If we're injecting a #GP with no valid IDT entry, inject a double-fault.
8697	No error codes for exceptions in real-mode. */
8698	if (uVector == X86_XCPT_GP)
8699	{
8700	uint32_t const uXcptDfInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DF)
8701	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
8702	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
8703	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
8704	HMEVENT EventXcptDf;
8705	RT_ZERO(EventXcptDf);
8706	EventXcptDf.u64IntInfo = uXcptDfInfo;
8707	return hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &EventXcptDf, fStepping, pfIntrState);
8708	}
8709
8710	/*
8711	* If we're injecting an event with no valid IDT entry, inject a #GP.
8712	* No error codes for exceptions in real-mode.
8713	*
8714	* See Intel spec. 20.1.4 "Interrupt and Exception Handling"
8715	*/
8716	uint32_t const uXcptGpInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_GP)
8717	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
8718	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
8719	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
8720	HMEVENT EventXcptGp;
8721	RT_ZERO(EventXcptGp);
8722	EventXcptGp.u64IntInfo = uXcptGpInfo;
8723	return hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &EventXcptGp, fStepping, pfIntrState);
8724	}
8725
8726	/* Software exceptions (#BP and #OF exceptions thrown as a result of INT3 or INTO) */
8727	uint16_t uGuestIp = pCtx->ip;
8728	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_SW_XCPT)
8729	{
8730	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF);
8731	/* #BP and #OF are both benign traps, we need to resume the next instruction. */
8732	uGuestIp = pCtx->ip + (uint16_t)cbInstr;
8733	}
8734	else if (uIntType == VMX_ENTRY_INT_INFO_TYPE_SW_INT)
8735	uGuestIp = pCtx->ip + (uint16_t)cbInstr;
8736
8737	/* Get the code segment selector and offset from the IDT entry for the interrupt handler. */
8738	X86IDTR16 IdtEntry;
8739	RTGCPHYS const GCPhysIdtEntry = (RTGCPHYS)pCtx->idtr.pIdt + uVector * cbIdtEntry;
8740	rc2 = PGMPhysSimpleReadGCPhys(pVM, &IdtEntry, GCPhysIdtEntry, cbIdtEntry);
8741	AssertRCReturn(rc2, rc2);
8742
8743	/* Construct the stack frame for the interrupt/exception handler. */
8744	VBOXSTRICTRC rcStrict;
8745	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, pCtx->eflags.u32);
8746	if (rcStrict == VINF_SUCCESS)
8747	{
8748	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, pCtx->cs.Sel);
8749	if (rcStrict == VINF_SUCCESS)
8750	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, uGuestIp);
8751	}
8752
8753	/* Clear the required eflag bits and jump to the interrupt/exception handler. */
8754	if (rcStrict == VINF_SUCCESS)
8755	{
8756	pCtx->eflags.u32 &= ~(X86_EFL_IF \| X86_EFL_TF \| X86_EFL_RF \| X86_EFL_AC);
8757	pCtx->rip = IdtEntry.offSel;
8758	pCtx->cs.Sel = IdtEntry.uSel;
8759	pCtx->cs.ValidSel = IdtEntry.uSel;
8760	pCtx->cs.u64Base = IdtEntry.uSel << cbIdtEntry;
8761	if ( uIntType == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT
8762	&& uVector == X86_XCPT_PF)
8763	pCtx->cr2 = GCPtrFault;
8764
8765	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CS \| HM_CHANGED_GUEST_CR2
8766	\| HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
8767	\| HM_CHANGED_GUEST_RSP);
8768
8769	/*
8770	* If we delivered a hardware exception (other than an NMI) and if there was
8771	* block-by-STI in effect, we should clear it.
8772	*/
8773	if (*pfIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
8774	{
8775	Assert( uIntType != VMX_ENTRY_INT_INFO_TYPE_NMI
8776	&& uIntType != VMX_ENTRY_INT_INFO_TYPE_EXT_INT);
8777	Log4Func(("Clearing inhibition due to STI\n"));
8778	*pfIntrState &= ~VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
8779	}
8780
8781	Log4(("Injected real-mode: u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x Eflags=%#x CS:EIP=%04x:%04x\n",
8782	u32IntInfo, u32ErrCode, cbInstr, pCtx->eflags.u, pCtx->cs.Sel, pCtx->eip));
8783
8784	/*
8785	* The event has been truly dispatched to the guest. Mark it as no longer pending so
8786	* we don't attempt to undo it if we are returning to ring-3 before executing guest code.
8787	*/
8788	pVCpu->hm.s.Event.fPending = false;
8789
8790	/*
8791	* If we eventually support nested-guest execution without unrestricted guest execution,
8792	* we should set fInterceptEvents here.
8793	*/
8794	Assert(!pVmxTransient->fIsNestedGuest);
8795
8796	/* If we're stepping and we've changed cs:rip above, bail out of the VMX R0 execution loop. */
8797	if (fStepping)
8798	rcStrict = VINF_EM_DBG_STEPPED;
8799	}
8800	AssertMsg(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
8801	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
8802	return rcStrict;
8803	}
8804	}
8805
8806	/*
8807	* Validate.
8808	*/
8809	Assert(VMX_ENTRY_INT_INFO_IS_VALID(u32IntInfo)); /* Bit 31 (Valid bit) must be set by caller. */
8810	Assert(!(u32IntInfo & VMX_BF_ENTRY_INT_INFO_RSVD_12_30_MASK)); /* Bits 30:12 MBZ. */
8811
8812	/*
8813	* Inject the event into the VMCS.
8814	*/
8815	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, u32IntInfo);
8816	if (VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(u32IntInfo))
8817	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, u32ErrCode);
8818	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, cbInstr);
8819	AssertRC(rc);
8820
8821	/*
8822	* Update guest CR2 if this is a page-fault.
8823	*/
8824	if (VMX_ENTRY_INT_INFO_IS_XCPT_PF(u32IntInfo))
8825	pCtx->cr2 = GCPtrFault;
8826
8827	Log4(("Injecting u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x CR2=%#RX64\n", u32IntInfo, u32ErrCode, cbInstr, pCtx->cr2));
8828	return VINF_SUCCESS;
8829	}
8830
8831
8832	/**
8833	* Evaluates the event to be delivered to the guest and sets it as the pending
8834	* event.
8835	*
8836	* Toggling of interrupt force-flags here is safe since we update TRPM on premature
8837	* exits to ring-3 before executing guest code, see hmR0VmxExitToRing3(). We must
8838	* NOT restore these force-flags.
8839	*
8840	* @returns Strict VBox status code (i.e. informational status codes too).
8841	* @param pVCpu The cross context virtual CPU structure.
8842	* @param pVmxTransient The VMX-transient structure.
8843	* @param pfIntrState Where to store the VT-x guest-interruptibility state.
8844	*/
8845	static VBOXSTRICTRC hmR0VmxEvaluatePendingEvent(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t *pfIntrState)
8846	{
8847	Assert(pfIntrState);
8848	Assert(!TRPMHasTrap(pVCpu));
8849
8850	/*
8851	* Compute/update guest-interruptibility state related FFs.
8852	* The FFs will be used below while evaluating events to be injected.
8853	*/
8854	*pfIntrState = hmR0VmxGetGuestIntrStateAndUpdateFFs(pVCpu);
8855
8856	/*
8857	* Evaluate if a new event needs to be injected.
8858	* An event that's already pending has already performed all necessary checks.
8859	*/
8860	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
8861	bool const fIsNestedGuest = pVmxTransient->fIsNestedGuest;
8862	if ( !pVCpu->hm.s.Event.fPending
8863	&& !VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
8864	{
8865	/** @todo SMI. SMIs take priority over NMIs. */
8866
8867	/*
8868	* NMIs.
8869	* NMIs take priority over external interrupts.
8870	*/
8871	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8872	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
8873	{
8874	/*
8875	* For a guest, the FF always indicates the guest's ability to receive an NMI.
8876	*
8877	* For a nested-guest, the FF always indicates the outer guest's ability to
8878	* receive an NMI while the guest-interruptibility state bit depends on whether
8879	* the nested-hypervisor is using virtual-NMIs.
8880	*/
8881	if (!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_BLOCK_NMIS))
8882	{
8883	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8884	if ( fIsNestedGuest
8885	&& CPUMIsGuestVmxPinCtlsSet(pCtx, VMX_PIN_CTLS_NMI_EXIT))
8886	return IEMExecVmxVmexitXcptNmi(pVCpu);
8887	#endif
8888	hmR0VmxSetPendingXcptNmi(pVCpu);
8889	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
8890	Log4Func(("NMI pending injection\n"));
8891
8892	/* We've injected the NMI, bail. */
8893	return VINF_SUCCESS;
8894	}
8895	else if (!fIsNestedGuest)
8896	hmR0VmxSetNmiWindowExitVmcs(pVCpu, pVmcsInfo);
8897	}
8898
8899	/*
8900	* External interrupts (PIC/APIC).
8901	* Once PDMGetInterrupt() returns a valid interrupt we -must- deliver it.
8902	* We cannot re-request the interrupt from the controller again.
8903	*/
8904	if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
8905	&& !pVCpu->hm.s.fSingleInstruction)
8906	{
8907	Assert(!DBGFIsStepping(pVCpu));
8908	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_RFLAGS);
8909	AssertRC(rc);
8910
8911	if (pCtx->eflags.u32 & X86_EFL_IF)
8912	{
8913	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8914	if ( fIsNestedGuest
8915	&& CPUMIsGuestVmxPinCtlsSet(pCtx, VMX_PIN_CTLS_EXT_INT_EXIT)
8916	&& !CPUMIsGuestVmxExitCtlsSet(pCtx, VMX_EXIT_CTLS_ACK_EXT_INT))
8917	{
8918	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitExtInt(pVCpu, 0 /* uVector /, true / fIntPending */);
8919	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8920	return rcStrict;
8921	}
8922	#endif
8923	uint8_t u8Interrupt;
8924	rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
8925	if (RT_SUCCESS(rc))
8926	{
8927	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8928	if ( fIsNestedGuest
8929	&& CPUMIsGuestVmxPinCtlsSet(pCtx, VMX_PIN_CTLS_EXT_INT_EXIT)
8930	&& CPUMIsGuestVmxExitCtlsSet(pCtx, VMX_EXIT_CTLS_ACK_EXT_INT))
8931	{
8932	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitExtInt(pVCpu, u8Interrupt, false /* fIntPending */);
8933	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8934	return rcStrict;
8935	}
8936	#endif
8937	hmR0VmxSetPendingExtInt(pVCpu, u8Interrupt);
8938	Log4Func(("External interrupt (%#x) pending injection\n", u8Interrupt));
8939	}
8940	else if (rc == VERR_APIC_INTR_MASKED_BY_TPR)
8941	{
8942	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchTprMaskedIrq);
8943
8944	if ( !fIsNestedGuest
8945	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW))
8946	hmR0VmxApicSetTprThreshold(pVmcsInfo, u8Interrupt >> 4);
8947	/* else: for nested-guests, TPR threshold is picked up while merging VMCS controls. */
8948
8949	/*
8950	* If the CPU doesn't have TPR shadowing, we will always get a VM-exit on TPR changes and
8951	* APICSetTpr() will end up setting the VMCPU_FF_INTERRUPT_APIC if required, so there is no
8952	* need to re-set this force-flag here.
8953	*/
8954	}
8955	else
8956	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
8957
8958	/* We've injected the interrupt or taken necessary action, bail. */
8959	return VINF_SUCCESS;
8960	}
8961	else if (!fIsNestedGuest)
8962	hmR0VmxSetIntWindowExitVmcs(pVCpu, pVmcsInfo);
8963	}
8964	}
8965	else if (!fIsNestedGuest)
8966	{
8967	/*
8968	* An event is being injected or we are in an interrupt shadow. Check if another event is
8969	* pending. If so, instruct VT-x to cause a VM-exit as soon as the guest is ready to accept
8970	* the pending event.
8971	*/
8972	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
8973	hmR0VmxSetNmiWindowExitVmcs(pVCpu, pVmcsInfo);
8974	else if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
8975	&& !pVCpu->hm.s.fSingleInstruction)
8976	hmR0VmxSetIntWindowExitVmcs(pVCpu, pVmcsInfo);
8977	}
8978	/* else: for nested-guests, NMI/interrupt-window exiting will be picked up when merging VMCS controls. */
8979
8980	return VINF_SUCCESS;
8981	}
8982
8983
8984	/**
8985	* Injects any pending events into the guest if the guest is in a state to
8986	* receive them.
8987	*
8988	* @returns Strict VBox status code (i.e. informational status codes too).
8989	* @param pVCpu The cross context virtual CPU structure.
8990	* @param pVmxTransient The VMX-transient structure.
8991	* @param fIntrState The VT-x guest-interruptibility state.
8992	* @param fStepping Whether we are single-stepping the guest using the
8993	* hypervisor debugger and should return
8994	* VINF_EM_DBG_STEPPED if the event was dispatched
8995	* directly.
8996	*/
8997	static VBOXSTRICTRC hmR0VmxInjectPendingEvent(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t fIntrState, bool fStepping)
8998	{
8999	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9000	Assert(VMMRZCallRing3IsEnabled(pVCpu));
9001
9002	#ifdef VBOX_STRICT
9003	/*
9004	* Verify guest-interruptibility state.
9005	*
9006	* We put this in a scoped block so we do not accidentally use fBlockSti or fBlockMovSS,
9007	* since injecting an event may modify the interruptibility state and we must thus always
9008	* use fIntrState.
9009	*/
9010	{
9011	bool const fBlockMovSS = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
9012	bool const fBlockSti = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI);
9013	Assert(!fBlockSti \|\| !(ASMAtomicUoReadU64(&pVCpu->cpum.GstCtx.fExtrn) & CPUMCTX_EXTRN_RFLAGS));
9014	Assert(!fBlockSti \|\| pVCpu->cpum.GstCtx.eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
9015	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
9016	Assert(!TRPMHasTrap(pVCpu));
9017	NOREF(fBlockMovSS); NOREF(fBlockSti);
9018	}
9019	#endif
9020
9021	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
9022	if (pVCpu->hm.s.Event.fPending)
9023	{
9024	/*
9025	* Do -not- clear any interrupt-window exiting control here. We might have an interrupt
9026	* pending even while injecting an event and in this case, we want a VM-exit as soon as
9027	* the guest is ready for the next interrupt, see @bugref{6208#c45}.
9028	*
9029	* See Intel spec. 26.6.5 "Interrupt-Window Exiting and Virtual-Interrupt Delivery".
9030	*/
9031	uint32_t const uIntType = VMX_ENTRY_INT_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
9032	#ifdef VBOX_STRICT
9033	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
9034	{
9035	Assert(pVCpu->cpum.GstCtx.eflags.u32 & X86_EFL_IF);
9036	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI));
9037	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
9038	}
9039	else if (uIntType == VMX_ENTRY_INT_INFO_TYPE_NMI)
9040	{
9041	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI));
9042	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI));
9043	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
9044	}
9045	#endif
9046	Log4(("Injecting pending event vcpu[%RU32] u64IntInfo=%#RX64 Type=%#RX32\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo,
9047	uIntType));
9048
9049	/*
9050	* Inject the event and get any changes to the guest-interruptibility state.
9051	*
9052	* The guest-interruptibility state may need to be updated if we inject the event
9053	* into the guest IDT ourselves (for real-on-v86 guest injecting software interrupts).
9054	*/
9055	rcStrict = hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &pVCpu->hm.s.Event, fStepping, &fIntrState);
9056	AssertRCReturn(VBOXSTRICTRC_VAL(rcStrict), rcStrict);
9057
9058	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
9059	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt);
9060	else
9061	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt);
9062	}
9063
9064	/*
9065	* Deliver any pending debug exceptions if the guest is single-stepping using EFLAGS.TF and
9066	* is an interrupt shadow (block-by-STI or block-by-MOV SS).
9067	*/
9068	if ( (fIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
9069	&& !pVmxTransient->fIsNestedGuest)
9070	{
9071	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
9072
9073	if (!pVCpu->hm.s.fSingleInstruction)
9074	{
9075	/*
9076	* Set or clear the BS bit depending on whether the trap flag is active or not. We need
9077	* to do both since we clear the BS bit from the VMCS while exiting to ring-3.
9078	*/
9079	Assert(!DBGFIsStepping(pVCpu));
9080	uint8_t const fTrapFlag = !!(pVCpu->cpum.GstCtx.eflags.u32 & X86_EFL_TF);
9081	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, fTrapFlag << VMX_BF_VMCS_PENDING_DBG_XCPT_BS_SHIFT);
9082	AssertRC(rc);
9083	}
9084	else
9085	{
9086	/*
9087	* We must not deliver a debug exception when single-stepping over STI/Mov-SS in the
9088	* hypervisor debugger using EFLAGS.TF but rather clear interrupt inhibition. However,
9089	* we take care of this case in hmR0VmxExportSharedDebugState and also the case if
9090	* we use MTF, so just make sure it's called before executing guest-code.
9091	*/
9092	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR_MASK);
9093	}
9094	}
9095	/* else: for nested-guest currently handling while merging controls. */
9096
9097	/*
9098	* Finally, update the guest-interruptibility state.
9099	*
9100	* This is required for the real-on-v86 software interrupt injection, for
9101	* pending debug exceptions as well as updates to the guest state from ring-3 (IEM).
9102	*/
9103	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
9104	AssertRC(rc);
9105
9106	/*
9107	* There's no need to clear the VM-entry interruption-information field here if we're not
9108	* injecting anything. VT-x clears the valid bit on every VM-exit.
9109	*
9110	* See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
9111	*/
9112
9113	Assert(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping));
9114	return rcStrict;
9115	}
9116
9117
9118	/**
9119	* Enters the VT-x session.
9120	*
9121	* @returns VBox status code.
9122	* @param pVCpu The cross context virtual CPU structure.
9123	*/
9124	VMMR0DECL(int) VMXR0Enter(PVMCPUCC pVCpu)
9125	{
9126	AssertPtr(pVCpu);
9127	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fSupported);
9128	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9129
9130	LogFlowFunc(("pVCpu=%p\n", pVCpu));
9131	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
9132	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
9133
9134	#ifdef VBOX_STRICT
9135	/* At least verify VMX is enabled, since we can't check if we're in VMX root mode without #GP'ing. */
9136	RTCCUINTREG uHostCr4 = ASMGetCR4();
9137	if (!(uHostCr4 & X86_CR4_VMXE))
9138	{
9139	LogRelFunc(("X86_CR4_VMXE bit in CR4 is not set!\n"));
9140	return VERR_VMX_X86_CR4_VMXE_CLEARED;
9141	}
9142	#endif
9143
9144	/*
9145	* Load the appropriate VMCS as the current and active one.
9146	*/
9147	PVMXVMCSINFO pVmcsInfo;
9148	bool const fInNestedGuestMode = CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx);
9149	if (!fInNestedGuestMode)
9150	pVmcsInfo = &pVCpu->hm.s.vmx.VmcsInfo;
9151	else
9152	pVmcsInfo = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
9153	int rc = hmR0VmxLoadVmcs(pVmcsInfo);
9154	if (RT_SUCCESS(rc))
9155	{
9156	pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs = fInNestedGuestMode;
9157	pVCpu->hm.s.fLeaveDone = false;
9158	Log4Func(("Loaded Vmcs. HostCpuId=%u\n", RTMpCpuId()));
9159
9160	/*
9161	* Do the EMT scheduled L1D flush here if needed.
9162	*/
9163	if (pVCpu->CTX_SUFF(pVM)->hm.s.fL1dFlushOnSched)
9164	ASMWrMsr(MSR_IA32_FLUSH_CMD, MSR_IA32_FLUSH_CMD_F_L1D);
9165	else if (pVCpu->CTX_SUFF(pVM)->hm.s.fMdsClearOnSched)
9166	hmR0MdsClear();
9167	}
9168	return rc;
9169	}
9170
9171
9172	/**
9173	* The thread-context callback (only on platforms which support it).
9174	*
9175	* @param enmEvent The thread-context event.
9176	* @param pVCpu The cross context virtual CPU structure.
9177	* @param fGlobalInit Whether global VT-x/AMD-V init. was used.
9178	* @thread EMT(pVCpu)
9179	*/
9180	VMMR0DECL(void) VMXR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPUCC pVCpu, bool fGlobalInit)
9181	{
9182	AssertPtr(pVCpu);
9183	RT_NOREF1(fGlobalInit);
9184
9185	switch (enmEvent)
9186	{
9187	case RTTHREADCTXEVENT_OUT:
9188	{
9189	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9190	Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
9191	VMCPU_ASSERT_EMT(pVCpu);
9192
9193	/* No longjmps (logger flushes, locks) in this fragile context. */
9194	VMMRZCallRing3Disable(pVCpu);
9195	Log4Func(("Preempting: HostCpuId=%u\n", RTMpCpuId()));
9196
9197	/* Restore host-state (FPU, debug etc.) */
9198	if (!pVCpu->hm.s.fLeaveDone)
9199	{
9200	/*
9201	* Do -not- import the guest-state here as we might already be in the middle of importing
9202	* it, esp. bad if we're holding the PGM lock, see comment in hmR0VmxImportGuestState().
9203	*/
9204	hmR0VmxLeave(pVCpu, false /* fImportState */);
9205	pVCpu->hm.s.fLeaveDone = true;
9206	}
9207
9208	/* Leave HM context, takes care of local init (term). */
9209	int rc = HMR0LeaveCpu(pVCpu);
9210	AssertRC(rc);
9211
9212	/* Restore longjmp state. */
9213	VMMRZCallRing3Enable(pVCpu);
9214	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreempt);
9215	break;
9216	}
9217
9218	case RTTHREADCTXEVENT_IN:
9219	{
9220	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9221	Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
9222	VMCPU_ASSERT_EMT(pVCpu);
9223
9224	/* No longjmps here, as we don't want to trigger preemption (& its hook) while resuming. */
9225	VMMRZCallRing3Disable(pVCpu);
9226	Log4Func(("Resumed: HostCpuId=%u\n", RTMpCpuId()));
9227
9228	/* Initialize the bare minimum state required for HM. This takes care of
9229	initializing VT-x if necessary (onlined CPUs, local init etc.) */
9230	int rc = hmR0EnterCpu(pVCpu);
9231	AssertRC(rc);
9232	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
9233	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
9234
9235	/* Load the active VMCS as the current one. */
9236	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
9237	rc = hmR0VmxLoadVmcs(pVmcsInfo);
9238	AssertRC(rc);
9239	Log4Func(("Resumed: Loaded Vmcs. HostCpuId=%u\n", RTMpCpuId()));
9240	pVCpu->hm.s.fLeaveDone = false;
9241
9242	/* Do the EMT scheduled L1D flush if needed. */
9243	if (pVCpu->CTX_SUFF(pVM)->hm.s.fL1dFlushOnSched)
9244	ASMWrMsr(MSR_IA32_FLUSH_CMD, MSR_IA32_FLUSH_CMD_F_L1D);
9245
9246	/* Restore longjmp state. */
9247	VMMRZCallRing3Enable(pVCpu);
9248	break;
9249	}
9250
9251	default:
9252	break;
9253	}
9254	}
9255
9256
9257	/**
9258	* Exports the host state into the VMCS host-state area.
9259	* Sets up the VM-exit MSR-load area.
9260	*
9261	* The CPU state will be loaded from these fields on every successful VM-exit.
9262	*
9263	* @returns VBox status code.
9264	* @param pVCpu The cross context virtual CPU structure.
9265	*
9266	* @remarks No-long-jump zone!!!
9267	*/
9268	static int hmR0VmxExportHostState(PVMCPUCC pVCpu)
9269	{
9270	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9271
9272	int rc = VINF_SUCCESS;
9273	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT)
9274	{
9275	hmR0VmxExportHostControlRegs();
9276
9277	rc = hmR0VmxExportHostSegmentRegs(pVCpu);
9278	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9279
9280	hmR0VmxExportHostMsrs(pVCpu);
9281
9282	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_HOST_CONTEXT;
9283	}
9284	return rc;
9285	}
9286
9287
9288	/**
9289	* Saves the host state in the VMCS host-state.
9290	*
9291	* @returns VBox status code.
9292	* @param pVCpu The cross context virtual CPU structure.
9293	*
9294	* @remarks No-long-jump zone!!!
9295	*/
9296	VMMR0DECL(int) VMXR0ExportHostState(PVMCPUCC pVCpu)
9297	{
9298	AssertPtr(pVCpu);
9299	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9300
9301	/*
9302	* Export the host state here while entering HM context.
9303	* When thread-context hooks are used, we might get preempted and have to re-save the host
9304	* state but most of the time we won't be, so do it here before we disable interrupts.
9305	*/
9306	return hmR0VmxExportHostState(pVCpu);
9307	}
9308
9309
9310	/**
9311	* Exports the guest state into the VMCS guest-state area.
9312	*
9313	* The will typically be done before VM-entry when the guest-CPU state and the
9314	* VMCS state may potentially be out of sync.
9315	*
9316	* Sets up the VM-entry MSR-load and VM-exit MSR-store areas. Sets up the
9317	* VM-entry controls.
9318	* Sets up the appropriate VMX non-root function to execute guest code based on
9319	* the guest CPU mode.
9320	*
9321	* @returns VBox strict status code.
9322	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
9323	* without unrestricted guest execution and the VMMDev is not presently
9324	* mapped (e.g. EFI32).
9325	*
9326	* @param pVCpu The cross context virtual CPU structure.
9327	* @param pVmxTransient The VMX-transient structure.
9328	*
9329	* @remarks No-long-jump zone!!!
9330	*/
9331	static VBOXSTRICTRC hmR0VmxExportGuestState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9332	{
9333	AssertPtr(pVCpu);
9334	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9335	LogFlowFunc(("pVCpu=%p\n", pVCpu));
9336
9337	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExportGuestState, x);
9338
9339	/*
9340	* Determine real-on-v86 mode.
9341	* Used when the guest is in real-mode and unrestricted guest execution is not used.
9342	*/
9343	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
9344	if ( pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest
9345	\|\| !CPUMIsGuestInRealModeEx(&pVCpu->cpum.GstCtx))
9346	pVmcsInfo->RealMode.fRealOnV86Active = false;
9347	else
9348	{
9349	Assert(!pVmxTransient->fIsNestedGuest);
9350	pVmcsInfo->RealMode.fRealOnV86Active = true;
9351	}
9352
9353	/*
9354	* Any ordering dependency among the sub-functions below must be explicitly stated using comments.
9355	* Ideally, assert that the cross-dependent bits are up-to-date at the point of using it.
9356	*/
9357	int rc = hmR0VmxExportGuestEntryExitCtls(pVCpu, pVmxTransient);
9358	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9359
9360	rc = hmR0VmxExportGuestCR0(pVCpu, pVmxTransient);
9361	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9362
9363	VBOXSTRICTRC rcStrict = hmR0VmxExportGuestCR3AndCR4(pVCpu, pVmxTransient);
9364	if (rcStrict == VINF_SUCCESS)
9365	{ /* likely */ }
9366	else
9367	{
9368	Assert(rcStrict == VINF_EM_RESCHEDULE_REM \|\| RT_FAILURE_NP(rcStrict));
9369	return rcStrict;
9370	}
9371
9372	rc = hmR0VmxExportGuestSegRegsXdtr(pVCpu, pVmxTransient);
9373	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9374
9375	rc = hmR0VmxExportGuestMsrs(pVCpu, pVmxTransient);
9376	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9377
9378	hmR0VmxExportGuestApicTpr(pVCpu, pVmxTransient);
9379	hmR0VmxExportGuestXcptIntercepts(pVCpu, pVmxTransient);
9380	hmR0VmxExportGuestRip(pVCpu);
9381	hmR0VmxExportGuestRsp(pVCpu);
9382	hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9383
9384	rc = hmR0VmxExportGuestHwvirtState(pVCpu, pVmxTransient);
9385	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9386
9387	/* Clear any bits that may be set but exported unconditionally or unused/reserved bits. */
9388	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~( (HM_CHANGED_GUEST_GPRS_MASK & ~HM_CHANGED_GUEST_RSP)
9389	\| HM_CHANGED_GUEST_CR2
9390	\| (HM_CHANGED_GUEST_DR_MASK & ~HM_CHANGED_GUEST_DR7)
9391	\| HM_CHANGED_GUEST_X87
9392	\| HM_CHANGED_GUEST_SSE_AVX
9393	\| HM_CHANGED_GUEST_OTHER_XSAVE
9394	\| HM_CHANGED_GUEST_XCRx
9395	\| HM_CHANGED_GUEST_KERNEL_GS_BASE /* Part of lazy or auto load-store MSRs. */
9396	\| HM_CHANGED_GUEST_SYSCALL_MSRS /* Part of lazy or auto load-store MSRs. */
9397	\| HM_CHANGED_GUEST_TSC_AUX
9398	\| HM_CHANGED_GUEST_OTHER_MSRS
9399	\| (HM_CHANGED_KEEPER_STATE_MASK & ~HM_CHANGED_VMX_MASK)));
9400
9401	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExportGuestState, x);
9402	return rc;
9403	}
9404
9405
9406	/**
9407	* Exports the state shared between the host and guest into the VMCS.
9408	*
9409	* @param pVCpu The cross context virtual CPU structure.
9410	* @param pVmxTransient The VMX-transient structure.
9411	*
9412	* @remarks No-long-jump zone!!!
9413	*/
9414	static void hmR0VmxExportSharedState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9415	{
9416	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9417	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9418
9419	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_DR_MASK)
9420	{
9421	int rc = hmR0VmxExportSharedDebugState(pVCpu, pVmxTransient);
9422	AssertRC(rc);
9423	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_GUEST_DR_MASK;
9424
9425	/* Loading shared debug bits might have changed eflags.TF bit for debugging purposes. */
9426	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_RFLAGS)
9427	hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9428	}
9429
9430	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_GUEST_LAZY_MSRS)
9431	{
9432	hmR0VmxLazyLoadGuestMsrs(pVCpu);
9433	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_VMX_GUEST_LAZY_MSRS;
9434	}
9435
9436	AssertMsg(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE),
9437	("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
9438	}
9439
9440
9441	/**
9442	* Worker for loading the guest-state bits in the inner VT-x execution loop.
9443	*
9444	* @returns Strict VBox status code (i.e. informational status codes too).
9445	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
9446	* without unrestricted guest execution and the VMMDev is not presently
9447	* mapped (e.g. EFI32).
9448	*
9449	* @param pVCpu The cross context virtual CPU structure.
9450	* @param pVmxTransient The VMX-transient structure.
9451	*
9452	* @remarks No-long-jump zone!!!
9453	*/
9454	static VBOXSTRICTRC hmR0VmxExportGuestStateOptimal(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9455	{
9456	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9457	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9458	Assert(VMMR0IsLogFlushDisabled(pVCpu));
9459
9460	#ifdef HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
9461	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
9462	#endif
9463
9464	/*
9465	* For many VM-exits only RIP/RSP/RFLAGS (and HWVIRT state when executing a nested-guest)
9466	* changes. First try to export only these without going through all other changed-flag checks.
9467	*/
9468	VBOXSTRICTRC rcStrict;
9469	uint64_t const fCtxMask = HM_CHANGED_ALL_GUEST & ~HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE;
9470	uint64_t const fMinimalMask = HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT;
9471	uint64_t const fCtxChanged = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
9472
9473	/* If only RIP/RSP/RFLAGS/HWVIRT changed, export only those (quicker, happens more often).*/
9474	if ( (fCtxChanged & fMinimalMask)
9475	&& !(fCtxChanged & (fCtxMask & ~fMinimalMask)))
9476	{
9477	hmR0VmxExportGuestRip(pVCpu);
9478	hmR0VmxExportGuestRsp(pVCpu);
9479	hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9480	rcStrict = hmR0VmxExportGuestHwvirtState(pVCpu, pVmxTransient);
9481	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportMinimal);
9482	}
9483	/* If anything else also changed, go through the full export routine and export as required. */
9484	else if (fCtxChanged & fCtxMask)
9485	{
9486	rcStrict = hmR0VmxExportGuestState(pVCpu, pVmxTransient);
9487	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
9488	{ /* likely */}
9489	else
9490	{
9491	AssertMsg(rcStrict == VINF_EM_RESCHEDULE_REM, ("Failed to export guest state! rc=%Rrc\n",
9492	VBOXSTRICTRC_VAL(rcStrict)));
9493	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9494	return rcStrict;
9495	}
9496	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportFull);
9497	}
9498	/* Nothing changed, nothing to load here. */
9499	else
9500	rcStrict = VINF_SUCCESS;
9501
9502	#ifdef VBOX_STRICT
9503	/* All the guest state bits should be loaded except maybe the host context and/or the shared host/guest bits. */
9504	uint64_t const fCtxChangedCur = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
9505	AssertMsg(!(fCtxChangedCur & fCtxMask), ("fCtxChangedCur=%#RX64\n", fCtxChangedCur));
9506	#endif
9507	return rcStrict;
9508	}
9509
9510
9511	/**
9512	* Tries to determine what part of the guest-state VT-x has deemed as invalid
9513	* and update error record fields accordingly.
9514	*
9515	* @returns VMX_IGS_* error codes.
9516	* @retval VMX_IGS_REASON_NOT_FOUND if this function could not find anything
9517	* wrong with the guest state.
9518	*
9519	* @param pVCpu The cross context virtual CPU structure.
9520	* @param pVmcsInfo The VMCS info. object.
9521	*
9522	* @remarks This function assumes our cache of the VMCS controls
9523	* are valid, i.e. hmR0VmxCheckCachedVmcsCtls() succeeded.
9524	*/
9525	static uint32_t hmR0VmxCheckGuestState(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
9526	{
9527	#define HMVMX_ERROR_BREAK(err) { uError = (err); break; }
9528	#define HMVMX_CHECK_BREAK(expr, err) do { \
9529	if (!(expr)) { uError = (err); break; } \
9530	} while (0)
9531
9532	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
9533	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
9534	uint32_t uError = VMX_IGS_ERROR;
9535	uint32_t u32IntrState = 0;
9536	bool const fUnrestrictedGuest = pVM->hm.s.vmx.fUnrestrictedGuest;
9537	do
9538	{
9539	int rc;
9540
9541	/*
9542	* Guest-interruptibility state.
9543	*
9544	* Read this first so that any check that fails prior to those that actually
9545	* require the guest-interruptibility state would still reflect the correct
9546	* VMCS value and avoids causing further confusion.
9547	*/
9548	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &u32IntrState);
9549	AssertRC(rc);
9550
9551	uint32_t u32Val;
9552	uint64_t u64Val;
9553
9554	/*
9555	* CR0.
9556	*/
9557	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
9558	uint64_t fSetCr0 = (pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
9559	uint64_t const fZapCr0 = (pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 \| pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
9560	/* Exceptions for unrestricted guest execution for CR0 fixed bits (PE, PG).
9561	See Intel spec. 26.3.1 "Checks on Guest Control Registers, Debug Registers and MSRs." */
9562	if (fUnrestrictedGuest)
9563	fSetCr0 &= ~(uint64_t)(X86_CR0_PE \| X86_CR0_PG);
9564
9565	uint64_t u64GuestCr0;
9566	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64GuestCr0);
9567	AssertRC(rc);
9568	HMVMX_CHECK_BREAK((u64GuestCr0 & fSetCr0) == fSetCr0, VMX_IGS_CR0_FIXED1);
9569	HMVMX_CHECK_BREAK(!(u64GuestCr0 & ~fZapCr0), VMX_IGS_CR0_FIXED0);
9570	if ( !fUnrestrictedGuest
9571	&& (u64GuestCr0 & X86_CR0_PG)
9572	&& !(u64GuestCr0 & X86_CR0_PE))
9573	HMVMX_ERROR_BREAK(VMX_IGS_CR0_PG_PE_COMBO);
9574
9575	/*
9576	* CR4.
9577	*/
9578	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
9579	uint64_t const fSetCr4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
9580	uint64_t const fZapCr4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 \| pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
9581
9582	uint64_t u64GuestCr4;
9583	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR4, &u64GuestCr4);
9584	AssertRC(rc);
9585	HMVMX_CHECK_BREAK((u64GuestCr4 & fSetCr4) == fSetCr4, VMX_IGS_CR4_FIXED1);
9586	HMVMX_CHECK_BREAK(!(u64GuestCr4 & ~fZapCr4), VMX_IGS_CR4_FIXED0);
9587
9588	/*
9589	* IA32_DEBUGCTL MSR.
9590	*/
9591	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, &u64Val);
9592	AssertRC(rc);
9593	if ( (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
9594	&& (u64Val & 0xfffffe3c)) /* Bits 31:9, bits 5:2 MBZ. */
9595	{
9596	HMVMX_ERROR_BREAK(VMX_IGS_DEBUGCTL_MSR_RESERVED);
9597	}
9598	uint64_t u64DebugCtlMsr = u64Val;
9599
9600	#ifdef VBOX_STRICT
9601	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
9602	AssertRC(rc);
9603	Assert(u32Val == pVmcsInfo->u32EntryCtls);
9604	#endif
9605	bool const fLongModeGuest = RT_BOOL(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
9606
9607	/*
9608	* RIP and RFLAGS.
9609	*/
9610	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RIP, &u64Val);
9611	AssertRC(rc);
9612	/* pCtx->rip can be different than the one in the VMCS (e.g. run guest code and VM-exits that don't update it). */
9613	if ( !fLongModeGuest
9614	\|\| !pCtx->cs.Attr.n.u1Long)
9615	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffff00000000)), VMX_IGS_LONGMODE_RIP_INVALID);
9616	/** @todo If the processor supports N < 64 linear-address bits, bits 63:N
9617	* must be identical if the "IA-32e mode guest" VM-entry
9618	* control is 1 and CS.L is 1. No check applies if the
9619	* CPU supports 64 linear-address bits. */
9620
9621	/* Flags in pCtx can be different (real-on-v86 for instance). We are only concerned about the VMCS contents here. */
9622	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RFLAGS, &u64Val);
9623	AssertRC(rc);
9624	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffc08028)), /* Bit 63:22, Bit 15, 5, 3 MBZ. */
9625	VMX_IGS_RFLAGS_RESERVED);
9626	HMVMX_CHECK_BREAK((u64Val & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
9627	uint32_t const u32Eflags = u64Val;
9628
9629	if ( fLongModeGuest
9630	\|\| ( fUnrestrictedGuest
9631	&& !(u64GuestCr0 & X86_CR0_PE)))
9632	{
9633	HMVMX_CHECK_BREAK(!(u32Eflags & X86_EFL_VM), VMX_IGS_RFLAGS_VM_INVALID);
9634	}
9635
9636	uint32_t u32EntryInfo;
9637	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo);
9638	AssertRC(rc);
9639	if (VMX_ENTRY_INT_INFO_IS_EXT_INT(u32EntryInfo))
9640	HMVMX_CHECK_BREAK(u32Eflags & X86_EFL_IF, VMX_IGS_RFLAGS_IF_INVALID);
9641
9642	/*
9643	* 64-bit checks.
9644	*/
9645	if (fLongModeGuest)
9646	{
9647	HMVMX_CHECK_BREAK(u64GuestCr0 & X86_CR0_PG, VMX_IGS_CR0_PG_LONGMODE);
9648	HMVMX_CHECK_BREAK(u64GuestCr4 & X86_CR4_PAE, VMX_IGS_CR4_PAE_LONGMODE);
9649	}
9650
9651	if ( !fLongModeGuest
9652	&& (u64GuestCr4 & X86_CR4_PCIDE))
9653	HMVMX_ERROR_BREAK(VMX_IGS_CR4_PCIDE);
9654
9655	/** @todo CR3 field must be such that bits 63:52 and bits in the range
9656	* 51:32 beyond the processor's physical-address width are 0. */
9657
9658	if ( (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
9659	&& (pCtx->dr[7] & X86_DR7_MBZ_MASK))
9660	HMVMX_ERROR_BREAK(VMX_IGS_DR7_RESERVED);
9661
9662	rc = VMXReadVmcsNw(VMX_VMCS_HOST_SYSENTER_ESP, &u64Val);
9663	AssertRC(rc);
9664	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_ESP_NOT_CANONICAL);
9665
9666	rc = VMXReadVmcsNw(VMX_VMCS_HOST_SYSENTER_EIP, &u64Val);
9667	AssertRC(rc);
9668	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_EIP_NOT_CANONICAL);
9669
9670	/*
9671	* PERF_GLOBAL MSR.
9672	*/
9673	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PERF_MSR)
9674	{
9675	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL, &u64Val);
9676	AssertRC(rc);
9677	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffff8fffffffc)),
9678	VMX_IGS_PERF_GLOBAL_MSR_RESERVED); /* Bits 63:35, bits 31:2 MBZ. */
9679	}
9680
9681	/*
9682	* PAT MSR.
9683	*/
9684	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PAT_MSR)
9685	{
9686	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PAT_FULL, &u64Val);
9687	AssertRC(rc);
9688	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0x707070707070707)), VMX_IGS_PAT_MSR_RESERVED);
9689	for (unsigned i = 0; i < 8; i++)
9690	{
9691	uint8_t u8Val = (u64Val & 0xff);
9692	if ( u8Val != 0 /* UC */
9693	&& u8Val != 1 /* WC */
9694	&& u8Val != 4 /* WT */
9695	&& u8Val != 5 /* WP */
9696	&& u8Val != 6 /* WB */
9697	&& u8Val != 7 /* UC- */)
9698	HMVMX_ERROR_BREAK(VMX_IGS_PAT_MSR_INVALID);
9699	u64Val >>= 8;
9700	}
9701	}
9702
9703	/*
9704	* EFER MSR.
9705	*/
9706	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_EFER_MSR)
9707	{
9708	Assert(pVM->hm.s.vmx.fSupportsVmcsEfer);
9709	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_EFER_FULL, &u64Val);
9710	AssertRC(rc);
9711	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffffffffff2fe)),
9712	VMX_IGS_EFER_MSR_RESERVED); /* Bits 63:12, bit 9, bits 7:1 MBZ. */
9713	HMVMX_CHECK_BREAK(RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL( pVmcsInfo->u32EntryCtls
9714	& VMX_ENTRY_CTLS_IA32E_MODE_GUEST),
9715	VMX_IGS_EFER_LMA_GUEST_MODE_MISMATCH);
9716	/** @todo r=ramshankar: Unrestricted check here is probably wrong, see
9717	* iemVmxVmentryCheckGuestState(). */
9718	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9719	\|\| !(u64GuestCr0 & X86_CR0_PG)
9720	\|\| RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL(u64Val & MSR_K6_EFER_LME),
9721	VMX_IGS_EFER_LMA_LME_MISMATCH);
9722	}
9723
9724	/*
9725	* Segment registers.
9726	*/
9727	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9728	\|\| !(pCtx->ldtr.Sel & X86_SEL_LDT), VMX_IGS_LDTR_TI_INVALID);
9729	if (!(u32Eflags & X86_EFL_VM))
9730	{
9731	/* CS */
9732	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1Present, VMX_IGS_CS_ATTR_P_INVALID);
9733	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xf00), VMX_IGS_CS_ATTR_RESERVED);
9734	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xfffe0000), VMX_IGS_CS_ATTR_RESERVED);
9735	HMVMX_CHECK_BREAK( (pCtx->cs.u32Limit & 0xfff) == 0xfff
9736	\|\| !(pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9737	HMVMX_CHECK_BREAK( !(pCtx->cs.u32Limit & 0xfff00000)
9738	\|\| (pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9739	/* CS cannot be loaded with NULL in protected mode. */
9740	HMVMX_CHECK_BREAK(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_CS_ATTR_UNUSABLE);
9741	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1DescType, VMX_IGS_CS_ATTR_S_INVALID);
9742	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
9743	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_UNEQUAL);
9744	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
9745	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_MISMATCH);
9746	else if (pVM->hm.s.vmx.fUnrestrictedGuest && pCtx->cs.Attr.n.u4Type == 3)
9747	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == 0, VMX_IGS_CS_ATTR_DPL_INVALID);
9748	else
9749	HMVMX_ERROR_BREAK(VMX_IGS_CS_ATTR_TYPE_INVALID);
9750
9751	/* SS */
9752	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9753	\|\| (pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL), VMX_IGS_SS_CS_RPL_UNEQUAL);
9754	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL), VMX_IGS_SS_ATTR_DPL_RPL_UNEQUAL);
9755	if ( !(pCtx->cr0 & X86_CR0_PE)
9756	\|\| pCtx->cs.Attr.n.u4Type == 3)
9757	HMVMX_CHECK_BREAK(!pCtx->ss.Attr.n.u2Dpl, VMX_IGS_SS_ATTR_DPL_INVALID);
9758
9759	if (!(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
9760	{
9761	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7, VMX_IGS_SS_ATTR_TYPE_INVALID);
9762	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u1Present, VMX_IGS_SS_ATTR_P_INVALID);
9763	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xf00), VMX_IGS_SS_ATTR_RESERVED);
9764	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xfffe0000), VMX_IGS_SS_ATTR_RESERVED);
9765	HMVMX_CHECK_BREAK( (pCtx->ss.u32Limit & 0xfff) == 0xfff
9766	\|\| !(pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9767	HMVMX_CHECK_BREAK( !(pCtx->ss.u32Limit & 0xfff00000)
9768	\|\| (pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9769	}
9770
9771	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxExportGuestSReg(). */
9772	if (!(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
9773	{
9774	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_DS_ATTR_A_INVALID);
9775	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u1Present, VMX_IGS_DS_ATTR_P_INVALID);
9776	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9777	\|\| pCtx->ds.Attr.n.u4Type > 11
9778	\|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9779	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xf00), VMX_IGS_DS_ATTR_RESERVED);
9780	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xfffe0000), VMX_IGS_DS_ATTR_RESERVED);
9781	HMVMX_CHECK_BREAK( (pCtx->ds.u32Limit & 0xfff) == 0xfff
9782	\|\| !(pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9783	HMVMX_CHECK_BREAK( !(pCtx->ds.u32Limit & 0xfff00000)
9784	\|\| (pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9785	HMVMX_CHECK_BREAK( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9786	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_DS_ATTR_TYPE_INVALID);
9787	}
9788	if (!(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
9789	{
9790	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_ES_ATTR_A_INVALID);
9791	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u1Present, VMX_IGS_ES_ATTR_P_INVALID);
9792	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9793	\|\| pCtx->es.Attr.n.u4Type > 11
9794	\|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9795	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xf00), VMX_IGS_ES_ATTR_RESERVED);
9796	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xfffe0000), VMX_IGS_ES_ATTR_RESERVED);
9797	HMVMX_CHECK_BREAK( (pCtx->es.u32Limit & 0xfff) == 0xfff
9798	\|\| !(pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
9799	HMVMX_CHECK_BREAK( !(pCtx->es.u32Limit & 0xfff00000)
9800	\|\| (pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
9801	HMVMX_CHECK_BREAK( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9802	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_ES_ATTR_TYPE_INVALID);
9803	}
9804	if (!(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
9805	{
9806	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_FS_ATTR_A_INVALID);
9807	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u1Present, VMX_IGS_FS_ATTR_P_INVALID);
9808	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9809	\|\| pCtx->fs.Attr.n.u4Type > 11
9810	\|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL), VMX_IGS_FS_ATTR_DPL_RPL_UNEQUAL);
9811	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xf00), VMX_IGS_FS_ATTR_RESERVED);
9812	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xfffe0000), VMX_IGS_FS_ATTR_RESERVED);
9813	HMVMX_CHECK_BREAK( (pCtx->fs.u32Limit & 0xfff) == 0xfff
9814	\|\| !(pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
9815	HMVMX_CHECK_BREAK( !(pCtx->fs.u32Limit & 0xfff00000)
9816	\|\| (pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
9817	HMVMX_CHECK_BREAK( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9818	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_FS_ATTR_TYPE_INVALID);
9819	}
9820	if (!(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
9821	{
9822	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_GS_ATTR_A_INVALID);
9823	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u1Present, VMX_IGS_GS_ATTR_P_INVALID);
9824	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9825	\|\| pCtx->gs.Attr.n.u4Type > 11
9826	\|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL), VMX_IGS_GS_ATTR_DPL_RPL_UNEQUAL);
9827	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xf00), VMX_IGS_GS_ATTR_RESERVED);
9828	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xfffe0000), VMX_IGS_GS_ATTR_RESERVED);
9829	HMVMX_CHECK_BREAK( (pCtx->gs.u32Limit & 0xfff) == 0xfff
9830	\|\| !(pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
9831	HMVMX_CHECK_BREAK( !(pCtx->gs.u32Limit & 0xfff00000)
9832	\|\| (pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
9833	HMVMX_CHECK_BREAK( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9834	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_GS_ATTR_TYPE_INVALID);
9835	}
9836	/* 64-bit capable CPUs. */
9837	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
9838	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
9839	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9840	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
9841	HMVMX_CHECK_BREAK(!RT_HI_U32(pCtx->cs.u64Base), VMX_IGS_LONGMODE_CS_BASE_INVALID);
9842	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ss.u64Base),
9843	VMX_IGS_LONGMODE_SS_BASE_INVALID);
9844	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ds.u64Base),
9845	VMX_IGS_LONGMODE_DS_BASE_INVALID);
9846	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->es.u64Base),
9847	VMX_IGS_LONGMODE_ES_BASE_INVALID);
9848	}
9849	else
9850	{
9851	/* V86 mode checks. */
9852	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
9853	if (pVmcsInfo->RealMode.fRealOnV86Active)
9854	{
9855	u32CSAttr = 0xf3; u32SSAttr = 0xf3;
9856	u32DSAttr = 0xf3; u32ESAttr = 0xf3;
9857	u32FSAttr = 0xf3; u32GSAttr = 0xf3;
9858	}
9859	else
9860	{
9861	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u;
9862	u32DSAttr = pCtx->ds.Attr.u; u32ESAttr = pCtx->es.Attr.u;
9863	u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
9864	}
9865
9866	/* CS */
9867	HMVMX_CHECK_BREAK((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), VMX_IGS_V86_CS_BASE_INVALID);
9868	HMVMX_CHECK_BREAK(pCtx->cs.u32Limit == 0xffff, VMX_IGS_V86_CS_LIMIT_INVALID);
9869	HMVMX_CHECK_BREAK(u32CSAttr == 0xf3, VMX_IGS_V86_CS_ATTR_INVALID);
9870	/* SS */
9871	HMVMX_CHECK_BREAK((pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4), VMX_IGS_V86_SS_BASE_INVALID);
9872	HMVMX_CHECK_BREAK(pCtx->ss.u32Limit == 0xffff, VMX_IGS_V86_SS_LIMIT_INVALID);
9873	HMVMX_CHECK_BREAK(u32SSAttr == 0xf3, VMX_IGS_V86_SS_ATTR_INVALID);
9874	/* DS */
9875	HMVMX_CHECK_BREAK((pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4), VMX_IGS_V86_DS_BASE_INVALID);
9876	HMVMX_CHECK_BREAK(pCtx->ds.u32Limit == 0xffff, VMX_IGS_V86_DS_LIMIT_INVALID);
9877	HMVMX_CHECK_BREAK(u32DSAttr == 0xf3, VMX_IGS_V86_DS_ATTR_INVALID);
9878	/* ES */
9879	HMVMX_CHECK_BREAK((pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4), VMX_IGS_V86_ES_BASE_INVALID);
9880	HMVMX_CHECK_BREAK(pCtx->es.u32Limit == 0xffff, VMX_IGS_V86_ES_LIMIT_INVALID);
9881	HMVMX_CHECK_BREAK(u32ESAttr == 0xf3, VMX_IGS_V86_ES_ATTR_INVALID);
9882	/* FS */
9883	HMVMX_CHECK_BREAK((pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4), VMX_IGS_V86_FS_BASE_INVALID);
9884	HMVMX_CHECK_BREAK(pCtx->fs.u32Limit == 0xffff, VMX_IGS_V86_FS_LIMIT_INVALID);
9885	HMVMX_CHECK_BREAK(u32FSAttr == 0xf3, VMX_IGS_V86_FS_ATTR_INVALID);
9886	/* GS */
9887	HMVMX_CHECK_BREAK((pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4), VMX_IGS_V86_GS_BASE_INVALID);
9888	HMVMX_CHECK_BREAK(pCtx->gs.u32Limit == 0xffff, VMX_IGS_V86_GS_LIMIT_INVALID);
9889	HMVMX_CHECK_BREAK(u32GSAttr == 0xf3, VMX_IGS_V86_GS_ATTR_INVALID);
9890	/* 64-bit capable CPUs. */
9891	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
9892	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
9893	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9894	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
9895	HMVMX_CHECK_BREAK(!RT_HI_U32(pCtx->cs.u64Base), VMX_IGS_LONGMODE_CS_BASE_INVALID);
9896	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ss.u64Base),
9897	VMX_IGS_LONGMODE_SS_BASE_INVALID);
9898	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ds.u64Base),
9899	VMX_IGS_LONGMODE_DS_BASE_INVALID);
9900	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->es.u64Base),
9901	VMX_IGS_LONGMODE_ES_BASE_INVALID);
9902	}
9903
9904	/*
9905	* TR.
9906	*/
9907	HMVMX_CHECK_BREAK(!(pCtx->tr.Sel & X86_SEL_LDT), VMX_IGS_TR_TI_INVALID);
9908	/* 64-bit capable CPUs. */
9909	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->tr.u64Base), VMX_IGS_TR_BASE_NOT_CANONICAL);
9910	if (fLongModeGuest)
9911	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u4Type == 11, /* 64-bit busy TSS. */
9912	VMX_IGS_LONGMODE_TR_ATTR_TYPE_INVALID);
9913	else
9914	HMVMX_CHECK_BREAK( pCtx->tr.Attr.n.u4Type == 3 /* 16-bit busy TSS. */
9915	\|\| pCtx->tr.Attr.n.u4Type == 11, /* 32-bit busy TSS.*/
9916	VMX_IGS_TR_ATTR_TYPE_INVALID);
9917	HMVMX_CHECK_BREAK(!pCtx->tr.Attr.n.u1DescType, VMX_IGS_TR_ATTR_S_INVALID);
9918	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u1Present, VMX_IGS_TR_ATTR_P_INVALID);
9919	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & 0xf00), VMX_IGS_TR_ATTR_RESERVED); /* Bits 11:8 MBZ. */
9920	HMVMX_CHECK_BREAK( (pCtx->tr.u32Limit & 0xfff) == 0xfff
9921	\|\| !(pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
9922	HMVMX_CHECK_BREAK( !(pCtx->tr.u32Limit & 0xfff00000)
9923	\|\| (pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
9924	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_TR_ATTR_UNUSABLE);
9925
9926	/*
9927	* GDTR and IDTR (64-bit capable checks).
9928	*/
9929	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
9930	AssertRC(rc);
9931	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_GDTR_BASE_NOT_CANONICAL);
9932
9933	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
9934	AssertRC(rc);
9935	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_IDTR_BASE_NOT_CANONICAL);
9936
9937	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val);
9938	AssertRC(rc);
9939	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_GDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
9940
9941	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val);
9942	AssertRC(rc);
9943	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_IDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
9944
9945	/*
9946	* Guest Non-Register State.
9947	*/
9948	/* Activity State. */
9949	uint32_t u32ActivityState;
9950	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, &u32ActivityState);
9951	AssertRC(rc);
9952	HMVMX_CHECK_BREAK( !u32ActivityState
9953	\|\| (u32ActivityState & RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Misc, VMX_BF_MISC_ACTIVITY_STATES)),
9954	VMX_IGS_ACTIVITY_STATE_INVALID);
9955	HMVMX_CHECK_BREAK( !(pCtx->ss.Attr.n.u2Dpl)
9956	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT, VMX_IGS_ACTIVITY_STATE_HLT_INVALID);
9957
9958	if ( u32IntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS
9959	\|\| u32IntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
9960	HMVMX_CHECK_BREAK(u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE, VMX_IGS_ACTIVITY_STATE_ACTIVE_INVALID);
9961
9962	/** @todo Activity state and injecting interrupts. Left as a todo since we
9963	* currently don't use activity states but ACTIVE. */
9964
9965	HMVMX_CHECK_BREAK( !(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)
9966	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT, VMX_IGS_ACTIVITY_STATE_SIPI_WAIT_INVALID);
9967
9968	/* Guest interruptibility-state. */
9969	HMVMX_CHECK_BREAK(!(u32IntrState & 0xffffffe0), VMX_IGS_INTERRUPTIBILITY_STATE_RESERVED);
9970	HMVMX_CHECK_BREAK((u32IntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
9971	!= (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
9972	VMX_IGS_INTERRUPTIBILITY_STATE_STI_MOVSS_INVALID);
9973	HMVMX_CHECK_BREAK( (u32Eflags & X86_EFL_IF)
9974	\|\| !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI),
9975	VMX_IGS_INTERRUPTIBILITY_STATE_STI_EFL_INVALID);
9976	if (VMX_ENTRY_INT_INFO_IS_EXT_INT(u32EntryInfo))
9977	{
9978	HMVMX_CHECK_BREAK( !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
9979	&& !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
9980	VMX_IGS_INTERRUPTIBILITY_STATE_EXT_INT_INVALID);
9981	}
9982	else if (VMX_ENTRY_INT_INFO_IS_XCPT_NMI(u32EntryInfo))
9983	{
9984	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
9985	VMX_IGS_INTERRUPTIBILITY_STATE_MOVSS_INVALID);
9986	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI),
9987	VMX_IGS_INTERRUPTIBILITY_STATE_STI_INVALID);
9988	}
9989	/** @todo Assumes the processor is not in SMM. */
9990	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI),
9991	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_INVALID);
9992	HMVMX_CHECK_BREAK( !(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)
9993	\|\| (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI),
9994	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_SMM_INVALID);
9995	if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
9996	&& VMX_ENTRY_INT_INFO_IS_XCPT_NMI(u32EntryInfo))
9997	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI), VMX_IGS_INTERRUPTIBILITY_STATE_NMI_INVALID);
9998
9999	/* Pending debug exceptions. */
10000	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, &u64Val);
10001	AssertRC(rc);
10002	/* Bits 63:15, Bit 13, Bits 11:4 MBZ. */
10003	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffffaff0)), VMX_IGS_LONGMODE_PENDING_DEBUG_RESERVED);
10004	u32Val = u64Val; /* For pending debug exceptions checks below. */
10005
10006	if ( (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10007	\|\| (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS)
10008	\|\| u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT)
10009	{
10010	if ( (u32Eflags & X86_EFL_TF)
10011	&& !(u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
10012	{
10013	/* Bit 14 is PendingDebug.BS. */
10014	HMVMX_CHECK_BREAK(u32Val & RT_BIT(14), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_SET);
10015	}
10016	if ( !(u32Eflags & X86_EFL_TF)
10017	\|\| (u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
10018	{
10019	/* Bit 14 is PendingDebug.BS. */
10020	HMVMX_CHECK_BREAK(!(u32Val & RT_BIT(14)), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_CLEAR);
10021	}
10022	}
10023
10024	/* VMCS link pointer. */
10025	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, &u64Val);
10026	AssertRC(rc);
10027	if (u64Val != UINT64_C(0xffffffffffffffff))
10028	{
10029	HMVMX_CHECK_BREAK(!(u64Val & 0xfff), VMX_IGS_VMCS_LINK_PTR_RESERVED);
10030	/** @todo Bits beyond the processor's physical-address width MBZ. */
10031	/** @todo SMM checks. */
10032	Assert(pVmcsInfo->HCPhysShadowVmcs == u64Val);
10033	Assert(pVmcsInfo->pvShadowVmcs);
10034	VMXVMCSREVID VmcsRevId;
10035	VmcsRevId.u = (uint32_t )pVmcsInfo->pvShadowVmcs;
10036	HMVMX_CHECK_BREAK(VmcsRevId.n.u31RevisionId == RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_ID),
10037	VMX_IGS_VMCS_LINK_PTR_SHADOW_VMCS_ID_INVALID);
10038	HMVMX_CHECK_BREAK(VmcsRevId.n.fIsShadowVmcs == (uint32_t)!!(pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING),
10039	VMX_IGS_VMCS_LINK_PTR_NOT_SHADOW);
10040	}
10041
10042	/** @todo Checks on Guest Page-Directory-Pointer-Table Entries when guest is
10043	* not using nested paging? */
10044	if ( pVM->hm.s.fNestedPaging
10045	&& !fLongModeGuest
10046	&& CPUMIsGuestInPAEModeEx(pCtx))
10047	{
10048	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &u64Val);
10049	AssertRC(rc);
10050	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10051
10052	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &u64Val);
10053	AssertRC(rc);
10054	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10055
10056	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &u64Val);
10057	AssertRC(rc);
10058	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10059
10060	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &u64Val);
10061	AssertRC(rc);
10062	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10063	}
10064
10065	/* Shouldn't happen but distinguish it from AssertRCBreak() errors. */
10066	if (uError == VMX_IGS_ERROR)
10067	uError = VMX_IGS_REASON_NOT_FOUND;
10068	} while (0);
10069
10070	pVCpu->hm.s.u32HMError = uError;
10071	pVCpu->hm.s.vmx.LastError.u32GuestIntrState = u32IntrState;
10072	return uError;
10073
10074	#undef HMVMX_ERROR_BREAK
10075	#undef HMVMX_CHECK_BREAK
10076	}
10077
10078
10079	/**
10080	* Map the APIC-access page for virtualizing APIC accesses.
10081	*
10082	* This can cause a longjumps to R3 due to the acquisition of the PGM lock. Hence,
10083	* this not done as part of exporting guest state, see @bugref{8721}.
10084	*
10085	* @returns VBox status code.
10086	* @param pVCpu The cross context virtual CPU structure.
10087	*/
10088	static int hmR0VmxMapHCApicAccessPage(PVMCPUCC pVCpu)
10089	{
10090	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10091	uint64_t const u64MsrApicBase = APICGetBaseMsrNoCheck(pVCpu);
10092
10093	Assert(PDMHasApic(pVM));
10094	Assert(u64MsrApicBase);
10095
10096	RTGCPHYS const GCPhysApicBase = u64MsrApicBase & PAGE_BASE_GC_MASK;
10097	Log4Func(("Mappping HC APIC-access page at %#RGp\n", GCPhysApicBase));
10098
10099	/* Unalias the existing mapping. */
10100	int rc = PGMHandlerPhysicalReset(pVM, GCPhysApicBase);
10101	AssertRCReturn(rc, rc);
10102
10103	/* Map the HC APIC-access page in place of the MMIO page, also updates the shadow page tables if necessary. */
10104	Assert(pVM->hm.s.vmx.HCPhysApicAccess != NIL_RTHCPHYS);
10105	rc = IOMR0MmioMapMmioHCPage(pVM, pVCpu, GCPhysApicBase, pVM->hm.s.vmx.HCPhysApicAccess, X86_PTE_RW \| X86_PTE_P);
10106	AssertRCReturn(rc, rc);
10107
10108	/* Update the per-VCPU cache of the APIC base MSR. */
10109	pVCpu->hm.s.vmx.u64GstMsrApicBase = u64MsrApicBase;
10110	return VINF_SUCCESS;
10111	}
10112
10113
10114	/**
10115	* Worker function passed to RTMpOnSpecific() that is to be called on the target
10116	* CPU.
10117	*
10118	* @param idCpu The ID for the CPU the function is called on.
10119	* @param pvUser1 Null, not used.
10120	* @param pvUser2 Null, not used.
10121	*/
10122	static DECLCALLBACK(void) hmR0DispatchHostNmi(RTCPUID idCpu, void pvUser1, void pvUser2)
10123	{
10124	RT_NOREF3(idCpu, pvUser1, pvUser2);
10125	VMXDispatchHostNmi();
10126	}
10127
10128
10129	/**
10130	* Dispatching an NMI on the host CPU that received it.
10131	*
10132	* @returns VBox status code.
10133	* @param pVCpu The cross context virtual CPU structure.
10134	* @param pVmcsInfo The VMCS info. object corresponding to the VMCS that was
10135	* executing when receiving the host NMI in VMX non-root
10136	* operation.
10137	*/
10138	static int hmR0VmxExitHostNmi(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
10139	{
10140	RTCPUID const idCpu = pVmcsInfo->idHostCpuExec;
10141	Assert(idCpu != NIL_RTCPUID);
10142
10143	/*
10144	* We don't want to delay dispatching the NMI any more than we have to. However,
10145	* we have already chosen -not- to dispatch NMIs when interrupts were still disabled
10146	* after executing guest or nested-guest code for the following reasons:
10147	*
10148	* - We would need to perform VMREADs with interrupts disabled and is orders of
10149	* magnitude worse when we run as a nested hypervisor without VMCS shadowing
10150	* supported by the host hypervisor.
10151	*
10152	* - It affects the common VM-exit scenario and keeps interrupts disabled for a
10153	* longer period of time just for handling an edge case like host NMIs which do
10154	* not occur nearly as frequently as other VM-exits.
10155	*
10156	* Let's cover the most likely scenario first. Check if we are on the target CPU
10157	* and dispatch the NMI right away. This should be much faster than calling into
10158	* RTMpOnSpecific() machinery.
10159	*/
10160	bool fDispatched = false;
10161	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
10162	if (idCpu == RTMpCpuId())
10163	{
10164	VMXDispatchHostNmi();
10165	fDispatched = true;
10166	}
10167	ASMSetFlags(fEFlags);
10168	if (fDispatched)
10169	{
10170	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC);
10171	return VINF_SUCCESS;
10172	}
10173
10174	/*
10175	* RTMpOnSpecific() waits until the worker function has run on the target CPU. So
10176	* there should be no race or recursion even if we are unlucky enough to be preempted
10177	* (to the target CPU) without dispatching the host NMI above.
10178	*/
10179	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGCIpi);
10180	return RTMpOnSpecific(idCpu, &hmR0DispatchHostNmi, NULL /* pvUser1 /, NULL / pvUser2 */);
10181	}
10182
10183
10184	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10185	/**
10186	* Merges the guest with the nested-guest MSR bitmap in preparation of executing the
10187	* nested-guest using hardware-assisted VMX.
10188	*
10189	* @param pVCpu The cross context virtual CPU structure.
10190	* @param pVmcsInfoNstGst The nested-guest VMCS info. object.
10191	* @param pVmcsInfoGst The guest VMCS info. object.
10192	*/
10193	static void hmR0VmxMergeMsrBitmapNested(PCVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfoNstGst, PCVMXVMCSINFO pVmcsInfoGst)
10194	{
10195	uint32_t const cbMsrBitmap = X86_PAGE_4K_SIZE;
10196	uint64_t pu64MsrBitmap = (uint64_t )pVmcsInfoNstGst->pvMsrBitmap;
10197	Assert(pu64MsrBitmap);
10198
10199	/*
10200	* We merge the guest MSR bitmap with the nested-guest MSR bitmap such that any
10201	* MSR that is intercepted by the guest is also intercepted while executing the
10202	* nested-guest using hardware-assisted VMX.
10203	*
10204	* Note! If the nested-guest is not using an MSR bitmap, every MSR must cause a
10205	* nested-guest VM-exit even if the outer guest is not intercepting some
10206	* MSRs. We cannot assume the caller has initialized the nested-guest
10207	* MSR bitmap in this case.
10208	*
10209	* The nested hypervisor may also switch whether it uses MSR bitmaps for
10210	* each of its VM-entry, hence initializing it once per-VM while setting
10211	* up the nested-guest VMCS is not sufficient.
10212	*/
10213	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
10214	if (pVmcsNstGst->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
10215	{
10216	uint64_t const pu64MsrBitmapNstGst = (uint64_t const )pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap);
10217	uint64_t const pu64MsrBitmapGst = (uint64_t const )pVmcsInfoGst->pvMsrBitmap;
10218	Assert(pu64MsrBitmapNstGst);
10219	Assert(pu64MsrBitmapGst);
10220
10221	uint32_t const cFrags = cbMsrBitmap / sizeof(uint64_t);
10222	for (uint32_t i = 0; i < cFrags; i++)
10223	pu64MsrBitmap[i] = pu64MsrBitmapNstGst[i] \| pu64MsrBitmapGst[i];
10224	}
10225	else
10226	ASMMemFill32(pu64MsrBitmap, cbMsrBitmap, UINT32_C(0xffffffff));
10227	}
10228
10229
10230	/**
10231	* Merges the guest VMCS in to the nested-guest VMCS controls in preparation of
10232	* hardware-assisted VMX execution of the nested-guest.
10233	*
10234	* For a guest, we don't modify these controls once we set up the VMCS and hence
10235	* this function is never called.
10236	*
10237	* For nested-guests since the nested hypervisor provides these controls on every
10238	* nested-guest VM-entry and could potentially change them everytime we need to
10239	* merge them before every nested-guest VM-entry.
10240	*
10241	* @returns VBox status code.
10242	* @param pVCpu The cross context virtual CPU structure.
10243	*/
10244	static int hmR0VmxMergeVmcsNested(PVMCPUCC pVCpu)
10245	{
10246	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10247	PCVMXVMCSINFO pVmcsInfoGst = &pVCpu->hm.s.vmx.VmcsInfo;
10248	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
10249	Assert(pVmcsNstGst);
10250
10251	/*
10252	* Merge the controls with the requirements of the guest VMCS.
10253	*
10254	* We do not need to validate the nested-guest VMX features specified in the nested-guest
10255	* VMCS with the features supported by the physical CPU as it's already done by the
10256	* VMLAUNCH/VMRESUME instruction emulation.
10257	*
10258	* This is because the VMX features exposed by CPUM (through CPUID/MSRs) to the guest are
10259	* derived from the VMX features supported by the physical CPU.
10260	*/
10261
10262	/* Pin-based VM-execution controls. */
10263	uint32_t const u32PinCtls = pVmcsNstGst->u32PinCtls \| pVmcsInfoGst->u32PinCtls;
10264
10265	/* Processor-based VM-execution controls. */
10266	uint32_t u32ProcCtls = (pVmcsNstGst->u32ProcCtls & ~VMX_PROC_CTLS_USE_IO_BITMAPS)
10267	\| (pVmcsInfoGst->u32ProcCtls & ~( VMX_PROC_CTLS_INT_WINDOW_EXIT
10268	\| VMX_PROC_CTLS_NMI_WINDOW_EXIT
10269	\| VMX_PROC_CTLS_USE_TPR_SHADOW
10270	\| VMX_PROC_CTLS_MONITOR_TRAP_FLAG));
10271
10272	/* Secondary processor-based VM-execution controls. */
10273	uint32_t const u32ProcCtls2 = (pVmcsNstGst->u32ProcCtls2 & ~VMX_PROC_CTLS2_VPID)
10274	\| (pVmcsInfoGst->u32ProcCtls2 & ~( VMX_PROC_CTLS2_VIRT_APIC_ACCESS
10275	\| VMX_PROC_CTLS2_INVPCID
10276	\| VMX_PROC_CTLS2_VMCS_SHADOWING
10277	\| VMX_PROC_CTLS2_RDTSCP
10278	\| VMX_PROC_CTLS2_XSAVES_XRSTORS
10279	\| VMX_PROC_CTLS2_APIC_REG_VIRT
10280	\| VMX_PROC_CTLS2_VIRT_INT_DELIVERY
10281	\| VMX_PROC_CTLS2_VMFUNC));
10282
10283	/*
10284	* VM-entry controls:
10285	* These controls contains state that depends on the nested-guest state (primarily
10286	* EFER MSR) and is thus not constant between VMLAUNCH/VMRESUME and the nested-guest
10287	* VM-exit. Although the nested hypervisor cannot change it, we need to in order to
10288	* properly continue executing the nested-guest if the EFER MSR changes but does not
10289	* cause a nested-guest VM-exits.
10290	*
10291	* VM-exit controls:
10292	* These controls specify the host state on return. We cannot use the controls from
10293	* the nested hypervisor state as is as it would contain the guest state rather than
10294	* the host state. Since the host state is subject to change (e.g. preemption, trips
10295	* to ring-3, longjmp and rescheduling to a different host CPU) they are not constant
10296	* through VMLAUNCH/VMRESUME and the nested-guest VM-exit.
10297	*
10298	* VM-entry MSR-load:
10299	* The guest MSRs from the VM-entry MSR-load area are already loaded into the guest-CPU
10300	* context by the VMLAUNCH/VMRESUME instruction emulation.
10301	*
10302	* VM-exit MSR-store:
10303	* The VM-exit emulation will take care of populating the MSRs from the guest-CPU context
10304	* back into the VM-exit MSR-store area.
10305	*
10306	* VM-exit MSR-load areas:
10307	* This must contain the real host MSRs with hardware-assisted VMX execution. Hence, we
10308	* can entirely ignore what the nested hypervisor wants to load here.
10309	*/
10310
10311	/*
10312	* Exception bitmap.
10313	*
10314	* We could remove #UD from the guest bitmap and merge it with the nested-guest bitmap
10315	* here (and avoid doing anything while exporting nested-guest state), but to keep the
10316	* code more flexible if intercepting exceptions become more dynamic in the future we do
10317	* it as part of exporting the nested-guest state.
10318	*/
10319	uint32_t const u32XcptBitmap = pVmcsNstGst->u32XcptBitmap \| pVmcsInfoGst->u32XcptBitmap;
10320
10321	/*
10322	* CR0/CR4 guest/host mask.
10323	*
10324	* Modifications by the nested-guest to CR0/CR4 bits owned by the host and the guest must
10325	* cause VM-exits, so we need to merge them here.
10326	*/
10327	uint64_t const u64Cr0Mask = pVmcsNstGst->u64Cr0Mask.u \| pVmcsInfoGst->u64Cr0Mask;
10328	uint64_t const u64Cr4Mask = pVmcsNstGst->u64Cr4Mask.u \| pVmcsInfoGst->u64Cr4Mask;
10329
10330	/*
10331	* Page-fault error-code mask and match.
10332	*
10333	* Although we require unrestricted guest execution (and thereby nested-paging) for
10334	* hardware-assisted VMX execution of nested-guests and thus the outer guest doesn't
10335	* normally intercept #PFs, it might intercept them for debugging purposes.
10336	*
10337	* If the outer guest is not intercepting #PFs, we can use the nested-guest #PF filters.
10338	* If the outer guest is intercepting #PFs, we must intercept all #PFs.
10339	*/
10340	uint32_t u32XcptPFMask;
10341	uint32_t u32XcptPFMatch;
10342	if (!(pVmcsInfoGst->u32XcptBitmap & RT_BIT(X86_XCPT_PF)))
10343	{
10344	u32XcptPFMask = pVmcsNstGst->u32XcptPFMask;
10345	u32XcptPFMatch = pVmcsNstGst->u32XcptPFMatch;
10346	}
10347	else
10348	{
10349	u32XcptPFMask = 0;
10350	u32XcptPFMatch = 0;
10351	}
10352
10353	/*
10354	* Pause-Loop exiting.
10355	*/
10356	uint32_t const cPleGapTicks = RT_MIN(pVM->hm.s.vmx.cPleGapTicks, pVmcsNstGst->u32PleGap);
10357	uint32_t const cPleWindowTicks = RT_MIN(pVM->hm.s.vmx.cPleWindowTicks, pVmcsNstGst->u32PleWindow);
10358
10359	/*
10360	* Pending debug exceptions.
10361	* Currently just copy whatever the nested-guest provides us.
10362	*/
10363	uint64_t const uPendingDbgXcpts = pVmcsNstGst->u64GuestPendingDbgXcpts.u;
10364
10365	/*
10366	* I/O Bitmap.
10367	*
10368	* We do not use the I/O bitmap that may be provided by the nested hypervisor as we always
10369	* intercept all I/O port accesses.
10370	*/
10371	Assert(u32ProcCtls & VMX_PROC_CTLS_UNCOND_IO_EXIT);
10372	Assert(!(u32ProcCtls & VMX_PROC_CTLS_USE_IO_BITMAPS));
10373
10374	/*
10375	* VMCS shadowing.
10376	*
10377	* We do not yet expose VMCS shadowing to the guest and thus VMCS shadowing should not be
10378	* enabled while executing the nested-guest.
10379	*/
10380	Assert(!(u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING));
10381
10382	/*
10383	* APIC-access page.
10384	*/
10385	RTHCPHYS HCPhysApicAccess;
10386	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10387	{
10388	Assert(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS);
10389	RTGCPHYS const GCPhysApicAccess = pVmcsNstGst->u64AddrApicAccess.u;
10390
10391	/** @todo NSTVMX: This is not really correct but currently is required to make
10392	* things work. We need to re-enable the page handler when we fallback to
10393	* IEM execution of the nested-guest! */
10394	PGMHandlerPhysicalPageTempOff(pVM, GCPhysApicAccess, GCPhysApicAccess);
10395
10396	void *pvPage;
10397	PGMPAGEMAPLOCK PgLockApicAccess;
10398	int rc = PGMPhysGCPhys2CCPtr(pVM, GCPhysApicAccess, &pvPage, &PgLockApicAccess);
10399	if (RT_SUCCESS(rc))
10400	{
10401	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysApicAccess, &HCPhysApicAccess);
10402	AssertMsgRCReturn(rc, ("Failed to get host-physical address for APIC-access page at %#RGp\n", GCPhysApicAccess), rc);
10403
10404	/** @todo Handle proper releasing of page-mapping lock later. */
10405	PGMPhysReleasePageMappingLock(pVCpu->CTX_SUFF(pVM), &PgLockApicAccess);
10406	}
10407	else
10408	return rc;
10409	}
10410	else
10411	HCPhysApicAccess = 0;
10412
10413	/*
10414	* Virtual-APIC page and TPR threshold.
10415	*/
10416	RTHCPHYS HCPhysVirtApic;
10417	uint32_t u32TprThreshold;
10418	if (u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
10419	{
10420	Assert(pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW);
10421	RTGCPHYS const GCPhysVirtApic = pVmcsNstGst->u64AddrVirtApic.u;
10422
10423	void *pvPage;
10424	PGMPAGEMAPLOCK PgLockVirtApic;
10425	int rc = PGMPhysGCPhys2CCPtr(pVM, GCPhysVirtApic, &pvPage, &PgLockVirtApic);
10426	if (RT_SUCCESS(rc))
10427	{
10428	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysVirtApic, &HCPhysVirtApic);
10429	AssertMsgRCReturn(rc, ("Failed to get host-physical address for virtual-APIC page at %#RGp\n", GCPhysVirtApic), rc);
10430
10431	/** @todo Handle proper releasing of page-mapping lock later. */
10432	PGMPhysReleasePageMappingLock(pVCpu->CTX_SUFF(pVM), &PgLockVirtApic);
10433	}
10434	else
10435	return rc;
10436
10437	u32TprThreshold = pVmcsNstGst->u32TprThreshold;
10438	}
10439	else
10440	{
10441	HCPhysVirtApic = 0;
10442	u32TprThreshold = 0;
10443
10444	/*
10445	* We must make sure CR8 reads/write must cause VM-exits when TPR shadowing is not
10446	* used by the nested hypervisor. Preventing MMIO accesses to the physical APIC will
10447	* be taken care of by EPT/shadow paging.
10448	*/
10449	if (pVM->hm.s.fAllow64BitGuests)
10450	{
10451	u32ProcCtls \|= VMX_PROC_CTLS_CR8_STORE_EXIT
10452	\| VMX_PROC_CTLS_CR8_LOAD_EXIT;
10453	}
10454	}
10455
10456	/*
10457	* Validate basic assumptions.
10458	*/
10459	PVMXVMCSINFO pVmcsInfoNstGst = &pVCpu->hm.s.vmx.VmcsInfoNstGst;
10460	Assert(pVM->hm.s.vmx.fAllowUnrestricted);
10461	Assert(pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS);
10462	Assert(hmGetVmxActiveVmcsInfo(pVCpu) == pVmcsInfoNstGst);
10463
10464	/*
10465	* Commit it to the nested-guest VMCS.
10466	*/
10467	int rc = VINF_SUCCESS;
10468	if (pVmcsInfoNstGst->u32PinCtls != u32PinCtls)
10469	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, u32PinCtls);
10470	if (pVmcsInfoNstGst->u32ProcCtls != u32ProcCtls)
10471	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, u32ProcCtls);
10472	if (pVmcsInfoNstGst->u32ProcCtls2 != u32ProcCtls2)
10473	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, u32ProcCtls2);
10474	if (pVmcsInfoNstGst->u32XcptBitmap != u32XcptBitmap)
10475	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
10476	if (pVmcsInfoNstGst->u64Cr0Mask != u64Cr0Mask)
10477	rc \|= VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, u64Cr0Mask);
10478	if (pVmcsInfoNstGst->u64Cr4Mask != u64Cr4Mask)
10479	rc \|= VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, u64Cr4Mask);
10480	if (pVmcsInfoNstGst->u32XcptPFMask != u32XcptPFMask)
10481	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, u32XcptPFMask);
10482	if (pVmcsInfoNstGst->u32XcptPFMatch != u32XcptPFMatch)
10483	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, u32XcptPFMatch);
10484	if ( !(u32ProcCtls & VMX_PROC_CTLS_PAUSE_EXIT)
10485	&& (u32ProcCtls2 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT))
10486	{
10487	Assert(pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT);
10488	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, cPleGapTicks);
10489	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, cPleWindowTicks);
10490	}
10491	if (u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
10492	{
10493	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
10494	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL, HCPhysVirtApic);
10495	}
10496	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10497	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, HCPhysApicAccess);
10498	rc \|= VMXWriteVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, uPendingDbgXcpts);
10499	AssertRC(rc);
10500
10501	/*
10502	* Update the nested-guest VMCS cache.
10503	*/
10504	pVmcsInfoNstGst->u32PinCtls = u32PinCtls;
10505	pVmcsInfoNstGst->u32ProcCtls = u32ProcCtls;
10506	pVmcsInfoNstGst->u32ProcCtls2 = u32ProcCtls2;
10507	pVmcsInfoNstGst->u32XcptBitmap = u32XcptBitmap;
10508	pVmcsInfoNstGst->u64Cr0Mask = u64Cr0Mask;
10509	pVmcsInfoNstGst->u64Cr4Mask = u64Cr4Mask;
10510	pVmcsInfoNstGst->u32XcptPFMask = u32XcptPFMask;
10511	pVmcsInfoNstGst->u32XcptPFMatch = u32XcptPFMatch;
10512	pVmcsInfoNstGst->HCPhysVirtApic = HCPhysVirtApic;
10513
10514	/*
10515	* We need to flush the TLB if we are switching the APIC-access page address.
10516	* See Intel spec. 28.3.3.4 "Guidelines for Use of the INVEPT Instruction".
10517	*/
10518	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10519	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = true;
10520
10521	/*
10522	* MSR bitmap.
10523	*
10524	* The MSR bitmap address has already been initialized while setting up the nested-guest
10525	* VMCS, here we need to merge the MSR bitmaps.
10526	*/
10527	if (u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
10528	hmR0VmxMergeMsrBitmapNested(pVCpu, pVmcsInfoNstGst, pVmcsInfoGst);
10529
10530	return VINF_SUCCESS;
10531	}
10532	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
10533
10534
10535	/**
10536	* Does the preparations before executing guest code in VT-x.
10537	*
10538	* This may cause longjmps to ring-3 and may even result in rescheduling to the
10539	* recompiler/IEM. We must be cautious what we do here regarding committing
10540	* guest-state information into the VMCS assuming we assuredly execute the
10541	* guest in VT-x mode.
10542	*
10543	* If we fall back to the recompiler/IEM after updating the VMCS and clearing
10544	* the common-state (TRPM/forceflags), we must undo those changes so that the
10545	* recompiler/IEM can (and should) use them when it resumes guest execution.
10546	* Otherwise such operations must be done when we can no longer exit to ring-3.
10547	*
10548	* @returns Strict VBox status code (i.e. informational status codes too).
10549	* @retval VINF_SUCCESS if we can proceed with running the guest, interrupts
10550	* have been disabled.
10551	* @retval VINF_VMX_VMEXIT if a nested-guest VM-exit occurs (e.g., while evaluating
10552	* pending events).
10553	* @retval VINF_EM_RESET if a triple-fault occurs while injecting a
10554	* double-fault into the guest.
10555	* @retval VINF_EM_DBG_STEPPED if @a fStepping is true and an event was
10556	* dispatched directly.
10557	* @retval VINF_* scheduling changes, we have to go back to ring-3.
10558	*
10559	* @param pVCpu The cross context virtual CPU structure.
10560	* @param pVmxTransient The VMX-transient structure.
10561	* @param fStepping Whether we are single-stepping the guest in the
10562	* hypervisor debugger. Makes us ignore some of the reasons
10563	* for returning to ring-3, and return VINF_EM_DBG_STEPPED
10564	* if event dispatching took place.
10565	*/
10566	static VBOXSTRICTRC hmR0VmxPreRunGuest(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, bool fStepping)
10567	{
10568	Assert(VMMRZCallRing3IsEnabled(pVCpu));
10569
10570	Log4Func(("fIsNested=%RTbool fStepping=%RTbool\n", pVmxTransient->fIsNestedGuest, fStepping));
10571
10572	#ifdef VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM
10573	if (pVmxTransient->fIsNestedGuest)
10574	{
10575	RT_NOREF2(pVCpu, fStepping);
10576	Log2Func(("Rescheduling to IEM due to nested-hwvirt or forced IEM exec -> VINF_EM_RESCHEDULE_REM\n"));
10577	return VINF_EM_RESCHEDULE_REM;
10578	}
10579	#endif
10580
10581	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
10582	PGMRZDynMapFlushAutoSet(pVCpu);
10583	#endif
10584
10585	/*
10586	* Check and process force flag actions, some of which might require us to go back to ring-3.
10587	*/
10588	VBOXSTRICTRC rcStrict = hmR0VmxCheckForceFlags(pVCpu, pVmxTransient, fStepping);
10589	if (rcStrict == VINF_SUCCESS)
10590	{
10591	/* FFs don't get set all the time. */
10592	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10593	if ( pVmxTransient->fIsNestedGuest
10594	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
10595	{
10596	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
10597	return VINF_VMX_VMEXIT;
10598	}
10599	#endif
10600	}
10601	else
10602	return rcStrict;
10603
10604	/*
10605	* Virtualize memory-mapped accesses to the physical APIC (may take locks).
10606	*/
10607	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10608	if ( !pVCpu->hm.s.vmx.u64GstMsrApicBase
10609	&& (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10610	&& PDMHasApic(pVM))
10611	{
10612	int rc = hmR0VmxMapHCApicAccessPage(pVCpu);
10613	AssertRCReturn(rc, rc);
10614	}
10615
10616	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10617	/*
10618	* Merge guest VMCS controls with the nested-guest VMCS controls.
10619	*
10620	* Even if we have not executed the guest prior to this (e.g. when resuming from a
10621	* saved state), we should be okay with merging controls as we initialize the
10622	* guest VMCS controls as part of VM setup phase.
10623	*/
10624	if ( pVmxTransient->fIsNestedGuest
10625	&& !pVCpu->hm.s.vmx.fMergedNstGstCtls)
10626	{
10627	int rc = hmR0VmxMergeVmcsNested(pVCpu);
10628	AssertRCReturn(rc, rc);
10629	pVCpu->hm.s.vmx.fMergedNstGstCtls = true;
10630	}
10631	#endif
10632
10633	/*
10634	* Evaluate events to be injected into the guest.
10635	*
10636	* Events in TRPM can be injected without inspecting the guest state.
10637	* If any new events (interrupts/NMI) are pending currently, we try to set up the
10638	* guest to cause a VM-exit the next time they are ready to receive the event.
10639	*
10640	* With nested-guests, evaluating pending events may cause VM-exits. Also, verify
10641	* that the event in TRPM that we will inject using hardware-assisted VMX is -not-
10642	* subject to interecption. Otherwise, we should have checked and injected them
10643	* manually elsewhere (IEM).
10644	*/
10645	if (TRPMHasTrap(pVCpu))
10646	{
10647	Assert(!pVmxTransient->fIsNestedGuest \|\| !CPUMIsGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx));
10648	hmR0VmxTrpmTrapToPendingEvent(pVCpu);
10649	}
10650
10651	uint32_t fIntrState;
10652	rcStrict = hmR0VmxEvaluatePendingEvent(pVCpu, pVmxTransient, &fIntrState);
10653
10654	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10655	/*
10656	* While evaluating pending events if something failed (unlikely) or if we were
10657	* preparing to run a nested-guest but performed a nested-guest VM-exit, we should bail.
10658	*/
10659	if (rcStrict != VINF_SUCCESS)
10660	return rcStrict;
10661	if ( pVmxTransient->fIsNestedGuest
10662	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
10663	{
10664	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
10665	return VINF_VMX_VMEXIT;
10666	}
10667	#else
10668	Assert(rcStrict == VINF_SUCCESS);
10669	#endif
10670
10671	/*
10672	* Event injection may take locks (currently the PGM lock for real-on-v86 case) and thus
10673	* needs to be done with longjmps or interrupts + preemption enabled. Event injection might
10674	* also result in triple-faulting the VM.
10675	*
10676	* With nested-guests, the above does not apply since unrestricted guest execution is a
10677	* requirement. Regardless, we do this here to avoid duplicating code elsewhere.
10678	*/
10679	rcStrict = hmR0VmxInjectPendingEvent(pVCpu, pVmxTransient, fIntrState, fStepping);
10680	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
10681	{ /* likely */ }
10682	else
10683	{
10684	AssertMsg(rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
10685	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
10686	return rcStrict;
10687	}
10688
10689	/*
10690	* A longjump might result in importing CR3 even for VM-exits that don't necessarily
10691	* import CR3 themselves. We will need to update them here, as even as late as the above
10692	* hmR0VmxInjectPendingEvent() call may lazily import guest-CPU state on demand causing
10693	* the below force flags to be set.
10694	*/
10695	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
10696	{
10697	Assert(!(ASMAtomicUoReadU64(&pVCpu->cpum.GstCtx.fExtrn) & CPUMCTX_EXTRN_CR3));
10698	int rc2 = PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
10699	AssertMsgReturn(rc2 == VINF_SUCCESS \|\| rc2 == VINF_PGM_SYNC_CR3,
10700	("%Rrc\n", rc2), RT_FAILURE_NP(rc2) ? rc2 : VERR_IPE_UNEXPECTED_INFO_STATUS);
10701	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
10702	}
10703	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
10704	{
10705	PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
10706	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
10707	}
10708
10709	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10710	/* Paranoia. */
10711	Assert(!pVmxTransient->fIsNestedGuest \|\| CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
10712	#endif
10713
10714	/*
10715	* No longjmps to ring-3 from this point on!!!
10716	* Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
10717	* This also disables flushing of the R0-logger instance (if any).
10718	*/
10719	VMMRZCallRing3Disable(pVCpu);
10720
10721	/*
10722	* Export the guest state bits.
10723	*
10724	* We cannot perform longjmps while loading the guest state because we do not preserve the
10725	* host/guest state (although the VMCS will be preserved) across longjmps which can cause
10726	* CPU migration.
10727	*
10728	* If we are injecting events to a real-on-v86 mode guest, we would have updated RIP and some segment
10729	* registers. Hence, exporting of the guest state needs to be done -after- injection of events.
10730	*/
10731	rcStrict = hmR0VmxExportGuestStateOptimal(pVCpu, pVmxTransient);
10732	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
10733	{ /* likely */ }
10734	else
10735	{
10736	VMMRZCallRing3Enable(pVCpu);
10737	return rcStrict;
10738	}
10739
10740	/*
10741	* We disable interrupts so that we don't miss any interrupts that would flag preemption
10742	* (IPI/timers etc.) when thread-context hooks aren't used and we've been running with
10743	* preemption disabled for a while. Since this is purely to aid the
10744	* RTThreadPreemptIsPending() code, it doesn't matter that it may temporarily reenable and
10745	* disable interrupt on NT.
10746	*
10747	* We need to check for force-flags that could've possible been altered since we last
10748	* checked them (e.g. by PDMGetInterrupt() leaving the PDM critical section,
10749	* see @bugref{6398}).
10750	*
10751	* We also check a couple of other force-flags as a last opportunity to get the EMT back
10752	* to ring-3 before executing guest code.
10753	*/
10754	pVmxTransient->fEFlags = ASMIntDisableFlags();
10755
10756	if ( ( !VM_FF_IS_ANY_SET(pVM, VM_FF_EMT_RENDEZVOUS \| VM_FF_TM_VIRTUAL_SYNC)
10757	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
10758	\|\| ( fStepping /* Optimized for the non-stepping case, so a bit of unnecessary work when stepping. */
10759	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK & ~(VMCPU_FF_TIMER \| VMCPU_FF_PDM_CRITSECT))) )
10760	{
10761	if (!RTThreadPreemptIsPending(NIL_RTTHREAD))
10762	{
10763	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10764	/*
10765	* If we are executing a nested-guest make sure that we should intercept subsequent
10766	* events. The one we are injecting might be part of VM-entry. This is mainly to keep
10767	* the VM-exit instruction emulation happy.
10768	*/
10769	if (pVmxTransient->fIsNestedGuest)
10770	CPUMSetGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx, true);
10771	#endif
10772
10773	/*
10774	* We've injected any pending events. This is really the point of no return (to ring-3).
10775	*
10776	* Note! The caller expects to continue with interrupts & longjmps disabled on successful
10777	* returns from this function, so do -not- enable them here.
10778	*/
10779	pVCpu->hm.s.Event.fPending = false;
10780	return VINF_SUCCESS;
10781	}
10782
10783	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPendingHostIrq);
10784	rcStrict = VINF_EM_RAW_INTERRUPT;
10785	}
10786	else
10787	{
10788	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
10789	rcStrict = VINF_EM_RAW_TO_R3;
10790	}
10791
10792	ASMSetFlags(pVmxTransient->fEFlags);
10793	VMMRZCallRing3Enable(pVCpu);
10794
10795	return rcStrict;
10796	}
10797
10798
10799	/**
10800	* Final preparations before executing guest code using hardware-assisted VMX.
10801	*
10802	* We can no longer get preempted to a different host CPU and there are no returns
10803	* to ring-3. We ignore any errors that may happen from this point (e.g. VMWRITE
10804	* failures), this function is not intended to fail sans unrecoverable hardware
10805	* errors.
10806	*
10807	* @param pVCpu The cross context virtual CPU structure.
10808	* @param pVmxTransient The VMX-transient structure.
10809	*
10810	* @remarks Called with preemption disabled.
10811	* @remarks No-long-jump zone!!!
10812	*/
10813	static void hmR0VmxPreRunGuestCommitted(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
10814	{
10815	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
10816	Assert(VMMR0IsLogFlushDisabled(pVCpu));
10817	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
10818	Assert(!pVCpu->hm.s.Event.fPending);
10819
10820	/*
10821	* Indicate start of guest execution and where poking EMT out of guest-context is recognized.
10822	*/
10823	VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
10824	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC);
10825
10826	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10827	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
10828	PHMPHYSCPU pHostCpu = hmR0GetCurrentCpu();
10829	RTCPUID const idCurrentCpu = pHostCpu->idCpu;
10830
10831	if (!CPUMIsGuestFPUStateActive(pVCpu))
10832	{
10833	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestFpuState, x);
10834	if (CPUMR0LoadGuestFPU(pVM, pVCpu) == VINF_CPUM_HOST_CR0_MODIFIED)
10835	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_HOST_CONTEXT;
10836	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestFpuState, x);
10837	STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadGuestFpu);
10838	}
10839
10840	/*
10841	* Re-export the host state bits as we may've been preempted (only happens when
10842	* thread-context hooks are used or when the VM start function changes) or if
10843	* the host CR0 is modified while loading the guest FPU state above.
10844	*
10845	* The 64-on-32 switcher saves the (64-bit) host state into the VMCS and if we
10846	* changed the switcher back to 32-bit, we must save the 32-bit host state here,
10847	* see @bugref{8432}.
10848	*
10849	* This may also happen when switching to/from a nested-guest VMCS without leaving
10850	* ring-0.
10851	*/
10852	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT)
10853	{
10854	hmR0VmxExportHostState(pVCpu);
10855	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportHostState);
10856	}
10857	Assert(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT));
10858
10859	/*
10860	* Export the state shared between host and guest (FPU, debug, lazy MSRs).
10861	*/
10862	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE)
10863	hmR0VmxExportSharedState(pVCpu, pVmxTransient);
10864	AssertMsg(!pVCpu->hm.s.fCtxChanged, ("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
10865
10866	/*
10867	* Store status of the shared guest/host debug state at the time of VM-entry.
10868	*/
10869	pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
10870	pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
10871
10872	/*
10873	* Always cache the TPR-shadow if the virtual-APIC page exists, thereby skipping
10874	* more than one conditional check. The post-run side of our code shall determine
10875	* if it needs to sync. the virtual APIC TPR with the TPR-shadow.
10876	*/
10877	if (pVmcsInfo->pbVirtApic)
10878	pVmxTransient->u8GuestTpr = pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR];
10879
10880	/*
10881	* Update the host MSRs values in the VM-exit MSR-load area.
10882	*/
10883	if (!pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs)
10884	{
10885	if (pVmcsInfo->cExitMsrLoad > 0)
10886	hmR0VmxUpdateAutoLoadHostMsrs(pVCpu, pVmcsInfo);
10887	pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs = true;
10888	}
10889
10890	/*
10891	* Evaluate if we need to intercept guest RDTSC/P accesses. Set up the
10892	* VMX-preemption timer based on the next virtual sync clock deadline.
10893	*/
10894	if ( !pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer
10895	\|\| idCurrentCpu != pVCpu->hm.s.idLastCpu)
10896	{
10897	hmR0VmxUpdateTscOffsettingAndPreemptTimer(pVCpu, pVmxTransient);
10898	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = true;
10899	}
10900
10901	/* Record statistics of how often we use TSC offsetting as opposed to intercepting RDTSC/P. */
10902	bool const fIsRdtscIntercepted = RT_BOOL(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_RDTSC_EXIT);
10903	if (!fIsRdtscIntercepted)
10904	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
10905	else
10906	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
10907
10908	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB flushing, set this across the world switch. */
10909	hmR0VmxFlushTaggedTlb(pHostCpu, pVCpu, pVmcsInfo); /* Invalidate the appropriate guest entries from the TLB. */
10910	Assert(idCurrentCpu == pVCpu->hm.s.idLastCpu);
10911	pVCpu->hm.s.vmx.LastError.idCurrentCpu = idCurrentCpu; /* Record the error reporting info. with the current host CPU. */
10912	pVmcsInfo->idHostCpuState = idCurrentCpu; /* Record the CPU for which the host-state has been exported. */
10913	pVmcsInfo->idHostCpuExec = idCurrentCpu; /* Record the CPU on which we shall execute. */
10914
10915	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
10916
10917	TMNotifyStartOfExecution(pVM, pVCpu); /* Notify TM to resume its clocks when TSC is tied to execution,
10918	as we're about to start executing the guest. */
10919
10920	/*
10921	* Load the guest TSC_AUX MSR when we are not intercepting RDTSCP.
10922	*
10923	* This is done this late as updating the TSC offsetting/preemption timer above
10924	* figures out if we can skip intercepting RDTSCP by calculating the number of
10925	* host CPU ticks till the next virtual sync deadline (for the dynamic case).
10926	*/
10927	if ( (pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_RDTSCP)
10928	&& !fIsRdtscIntercepted)
10929	{
10930	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_TSC_AUX);
10931
10932	/* NB: Because we call hmR0VmxAddAutoLoadStoreMsr with fUpdateHostMsr=true,
10933	it's safe even after hmR0VmxUpdateAutoLoadHostMsrs has already been done. */
10934	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_TSC_AUX, CPUMGetGuestTscAux(pVCpu),
10935	true /* fSetReadWrite /, true / fUpdateHostMsr */);
10936	AssertRC(rc);
10937	Assert(!pVmxTransient->fRemoveTscAuxMsr);
10938	pVmxTransient->fRemoveTscAuxMsr = true;
10939	}
10940
10941	#ifdef VBOX_STRICT
10942	Assert(pVCpu->hm.s.vmx.fUpdatedHostAutoMsrs);
10943	hmR0VmxCheckAutoLoadStoreMsrs(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest);
10944	hmR0VmxCheckHostEferMsr(pVCpu, pVmcsInfo);
10945	AssertRC(hmR0VmxCheckCachedVmcsCtls(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest));
10946	#endif
10947
10948	#ifdef HMVMX_ALWAYS_CHECK_GUEST_STATE
10949	/** @todo r=ramshankar: We can now probably use iemVmxVmentryCheckGuestState here.
10950	* Add a PVMXMSRS parameter to it, so that IEM can look at the host MSRs,
10951	* see @bugref{9180#c54}. */
10952	uint32_t const uInvalidReason = hmR0VmxCheckGuestState(pVCpu, pVmcsInfo);
10953	if (uInvalidReason != VMX_IGS_REASON_NOT_FOUND)
10954	Log4(("hmR0VmxCheckGuestState returned %#x\n", uInvalidReason));
10955	#endif
10956	}
10957
10958
10959	/**
10960	* First C routine invoked after running guest code using hardware-assisted VMX.
10961	*
10962	* @param pVCpu The cross context virtual CPU structure.
10963	* @param pVmxTransient The VMX-transient structure.
10964	* @param rcVMRun Return code of VMLAUNCH/VMRESUME.
10965	*
10966	* @remarks Called with interrupts disabled, and returns with interrupts enabled!
10967	*
10968	* @remarks No-long-jump zone!!! This function will however re-enable longjmps
10969	* unconditionally when it is safe to do so.
10970	*/
10971	static void hmR0VmxPostRunGuest(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, int rcVMRun)
10972	{
10973	uint64_t const uHostTsc = ASMReadTSC(); /** @todo We can do a lot better here, see @bugref{9180#c38}. */
10974
10975	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB flushing. */
10976	ASMAtomicIncU32(&pVCpu->hm.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for EMT poking. */
10977	pVCpu->hm.s.fCtxChanged = 0; /* Exits/longjmps to ring-3 requires saving the guest state. */
10978	pVmxTransient->fVmcsFieldsRead = 0; /* Transient fields need to be read from the VMCS. */
10979	pVmxTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
10980	pVmxTransient->fVectoringDoublePF = false; /* Vectoring double page-fault needs to be determined later. */
10981
10982	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
10983	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_RDTSC_EXIT))
10984	{
10985	uint64_t uGstTsc;
10986	if (!pVmxTransient->fIsNestedGuest)
10987	uGstTsc = uHostTsc + pVmcsInfo->u64TscOffset;
10988	else
10989	{
10990	uint64_t const uNstGstTsc = uHostTsc + pVmcsInfo->u64TscOffset;
10991	uGstTsc = CPUMRemoveNestedGuestTscOffset(pVCpu, uNstGstTsc);
10992	}
10993	TMCpuTickSetLastSeen(pVCpu, uGstTsc); /* Update TM with the guest TSC. */
10994	}
10995
10996	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatPreExit, x);
10997	TMNotifyEndOfExecution(pVCpu->CTX_SUFF(pVM), pVCpu); /* Notify TM that the guest is no longer running. */
10998	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
10999
11000	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_REQUIRED; /* Some host state messed up by VMX needs restoring. */
11001	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
11002	#ifdef VBOX_STRICT
11003	hmR0VmxCheckHostEferMsr(pVCpu, pVmcsInfo); /* Verify that the host EFER MSR wasn't modified. */
11004	#endif
11005	Assert(!ASMIntAreEnabled());
11006	ASMSetFlags(pVmxTransient->fEFlags); /* Enable interrupts. */
11007	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
11008
11009	#ifdef HMVMX_ALWAYS_CLEAN_TRANSIENT
11010	/*
11011	* Clean all the VMCS fields in the transient structure before reading
11012	* anything from the VMCS.
11013	*/
11014	pVmxTransient->uExitReason = 0;
11015	pVmxTransient->uExitIntErrorCode = 0;
11016	pVmxTransient->uExitQual = 0;
11017	pVmxTransient->uGuestLinearAddr = 0;
11018	pVmxTransient->uExitIntInfo = 0;
11019	pVmxTransient->cbExitInstr = 0;
11020	pVmxTransient->ExitInstrInfo.u = 0;
11021	pVmxTransient->uEntryIntInfo = 0;
11022	pVmxTransient->uEntryXcptErrorCode = 0;
11023	pVmxTransient->cbEntryInstr = 0;
11024	pVmxTransient->uIdtVectoringInfo = 0;
11025	pVmxTransient->uIdtVectoringErrorCode = 0;
11026	#endif
11027
11028	/*
11029	* Save the basic VM-exit reason and check if the VM-entry failed.
11030	* See Intel spec. 24.9.1 "Basic VM-exit Information".
11031	*/
11032	uint32_t uExitReason;
11033	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
11034	AssertRC(rc);
11035	pVmxTransient->uExitReason = VMX_EXIT_REASON_BASIC(uExitReason);
11036	pVmxTransient->fVMEntryFailed = VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason);
11037
11038	/*
11039	* Log the VM-exit before logging anything else as otherwise it might be a
11040	* tad confusing what happens before and after the world-switch.
11041	*/
11042	HMVMX_LOG_EXIT(pVCpu, uExitReason);
11043
11044	/*
11045	* Remove the TSC_AUX MSR from the auto-load/store MSR area and reset any MSR
11046	* bitmap permissions, if it was added before VM-entry.
11047	*/
11048	if (pVmxTransient->fRemoveTscAuxMsr)
11049	{
11050	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_TSC_AUX);
11051	pVmxTransient->fRemoveTscAuxMsr = false;
11052	}
11053
11054	/*
11055	* Check if VMLAUNCH/VMRESUME succeeded.
11056	* If this failed, we cause a guru meditation and cease further execution.
11057	*
11058	* However, if we are executing a nested-guest we might fail if we use the
11059	* fast path rather than fully emulating VMLAUNCH/VMRESUME instruction in IEM.
11060	*/
11061	if (RT_LIKELY(rcVMRun == VINF_SUCCESS))
11062	{
11063	/*
11064	* Update the VM-exit history array here even if the VM-entry failed due to:
11065	* - Invalid guest state.
11066	* - MSR loading.
11067	* - Machine-check event.
11068	*
11069	* In any of the above cases we will still have a "valid" VM-exit reason
11070	* despite @a fVMEntryFailed being false.
11071	*
11072	* See Intel spec. 26.7 "VM-Entry failures during or after loading guest state".
11073	*
11074	* Note! We don't have CS or RIP at this point. Will probably address that later
11075	* by amending the history entry added here.
11076	*/
11077	EMHistoryAddExit(pVCpu, EMEXIT_MAKE_FT(EMEXIT_F_KIND_VMX, pVmxTransient->uExitReason & EMEXIT_F_TYPE_MASK),
11078	UINT64_MAX, uHostTsc);
11079
11080	if (RT_LIKELY(!pVmxTransient->fVMEntryFailed))
11081	{
11082	VMMRZCallRing3Enable(pVCpu);
11083
11084	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
11085	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
11086
11087	#ifdef HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
11088	hmR0VmxReadAllRoFieldsVmcs(pVmxTransient);
11089	#endif
11090
11091	/*
11092	* Import the guest-interruptibility state always as we need it while evaluating
11093	* injecting events on re-entry.
11094	*
11095	* We don't import CR0 (when unrestricted guest execution is unavailable) despite
11096	* checking for real-mode while exporting the state because all bits that cause
11097	* mode changes wrt CR0 are intercepted.
11098	*/
11099	uint64_t const fImportMask = CPUMCTX_EXTRN_HM_VMX_INT_STATE
11100	#if defined(HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE) \|\| defined(HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE)
11101	\| HMVMX_CPUMCTX_EXTRN_ALL
11102	#elif defined(HMVMX_ALWAYS_SAVE_GUEST_RFLAGS)
11103	\| CPUMCTX_EXTRN_RFLAGS
11104	#endif
11105	;
11106	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImportMask);
11107	AssertRC(rc);
11108
11109	/*
11110	* Sync the TPR shadow with our APIC state.
11111	*/
11112	if ( !pVmxTransient->fIsNestedGuest
11113	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW))
11114	{
11115	Assert(pVmcsInfo->pbVirtApic);
11116	if (pVmxTransient->u8GuestTpr != pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR])
11117	{
11118	rc = APICSetTpr(pVCpu, pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR]);
11119	AssertRC(rc);
11120	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
11121	}
11122	}
11123
11124	Assert(VMMRZCallRing3IsEnabled(pVCpu));
11125	return;
11126	}
11127	}
11128	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11129	else if (pVmxTransient->fIsNestedGuest)
11130	AssertMsgFailed(("VMLAUNCH/VMRESUME failed but shouldn't happen when VMLAUNCH/VMRESUME was emulated in IEM!\n"));
11131	#endif
11132	else
11133	Log4Func(("VM-entry failure: rcVMRun=%Rrc fVMEntryFailed=%RTbool\n", rcVMRun, pVmxTransient->fVMEntryFailed));
11134
11135	VMMRZCallRing3Enable(pVCpu);
11136	}
11137
11138
11139	/**
11140	* Runs the guest code using hardware-assisted VMX the normal way.
11141	*
11142	* @returns VBox status code.
11143	* @param pVCpu The cross context virtual CPU structure.
11144	* @param pcLoops Pointer to the number of executed loops.
11145	*/
11146	static VBOXSTRICTRC hmR0VmxRunGuestCodeNormal(PVMCPUCC pVCpu, uint32_t *pcLoops)
11147	{
11148	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hm.s.cMaxResumeLoops;
11149	Assert(pcLoops);
11150	Assert(*pcLoops <= cMaxResumeLoops);
11151	Assert(!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
11152
11153	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11154	/*
11155	* Switch to the guest VMCS as we may have transitioned from executing the nested-guest
11156	* without leaving ring-0. Otherwise, if we came from ring-3 we would have loaded the
11157	* guest VMCS while entering the VMX ring-0 session.
11158	*/
11159	if (pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs)
11160	{
11161	int rc = hmR0VmxSwitchToGstOrNstGstVmcs(pVCpu, false /* fSwitchToNstGstVmcs */);
11162	if (RT_SUCCESS(rc))
11163	{ /* likely */ }
11164	else
11165	{
11166	LogRelFunc(("Failed to switch to the guest VMCS. rc=%Rrc\n", rc));
11167	return rc;
11168	}
11169	}
11170	#endif
11171
11172	VMXTRANSIENT VmxTransient;
11173	RT_ZERO(VmxTransient);
11174	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
11175
11176	/* Paranoia. */
11177	Assert(VmxTransient.pVmcsInfo == &pVCpu->hm.s.vmx.VmcsInfo);
11178
11179	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
11180	for (;;)
11181	{
11182	Assert(!HMR0SuspendPending());
11183	HMVMX_ASSERT_CPU_SAFE(pVCpu);
11184	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
11185
11186	/*
11187	* Preparatory work for running nested-guest code, this may force us to
11188	* return to ring-3.
11189	*
11190	* Warning! This bugger disables interrupts on VINF_SUCCESS!
11191	*/
11192	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, false /* fStepping */);
11193	if (rcStrict != VINF_SUCCESS)
11194	break;
11195
11196	/* Interrupts are disabled at this point! */
11197	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
11198	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
11199	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
11200	/* Interrupts are re-enabled at this point! */
11201
11202	/*
11203	* Check for errors with running the VM (VMLAUNCH/VMRESUME).
11204	*/
11205	if (RT_SUCCESS(rcRun))
11206	{ /* very likely */ }
11207	else
11208	{
11209	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
11210	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
11211	return rcRun;
11212	}
11213
11214	/*
11215	* Profile the VM-exit.
11216	*/
11217	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
11218	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
11219	STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
11220	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
11221	HMVMX_START_EXIT_DISPATCH_PROF();
11222
11223	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
11224
11225	/*
11226	* Handle the VM-exit.
11227	*/
11228	#ifdef HMVMX_USE_FUNCTION_TABLE
11229	rcStrict = g_aVMExitHandlers[VmxTransient.uExitReason].pfn(pVCpu, &VmxTransient);
11230	#else
11231	rcStrict = hmR0VmxHandleExit(pVCpu, &VmxTransient);
11232	#endif
11233	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
11234	if (rcStrict == VINF_SUCCESS)
11235	{
11236	if (++(*pcLoops) <= cMaxResumeLoops)
11237	continue;
11238	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
11239	rcStrict = VINF_EM_RAW_INTERRUPT;
11240	}
11241	break;
11242	}
11243
11244	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
11245	return rcStrict;
11246	}
11247
11248
11249	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11250	/**
11251	* Runs the nested-guest code using hardware-assisted VMX.
11252	*
11253	* @returns VBox status code.
11254	* @param pVCpu The cross context virtual CPU structure.
11255	* @param pcLoops Pointer to the number of executed loops.
11256	*
11257	* @sa hmR0VmxRunGuestCodeNormal.
11258	*/
11259	static VBOXSTRICTRC hmR0VmxRunGuestCodeNested(PVMCPUCC pVCpu, uint32_t *pcLoops)
11260	{
11261	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hm.s.cMaxResumeLoops;
11262	Assert(pcLoops);
11263	Assert(*pcLoops <= cMaxResumeLoops);
11264	Assert(CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
11265
11266	/*
11267	* Switch to the nested-guest VMCS as we may have transitioned from executing the
11268	* guest without leaving ring-0. Otherwise, if we came from ring-3 we would have
11269	* loaded the nested-guest VMCS while entering the VMX ring-0 session.
11270	*/
11271	if (!pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs)
11272	{
11273	int rc = hmR0VmxSwitchToGstOrNstGstVmcs(pVCpu, true /* fSwitchToNstGstVmcs */);
11274	if (RT_SUCCESS(rc))
11275	{ /* likely */ }
11276	else
11277	{
11278	LogRelFunc(("Failed to switch to the nested-guest VMCS. rc=%Rrc\n", rc));
11279	return rc;
11280	}
11281	}
11282
11283	VMXTRANSIENT VmxTransient;
11284	RT_ZERO(VmxTransient);
11285	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
11286	VmxTransient.fIsNestedGuest = true;
11287
11288	/* Paranoia. */
11289	Assert(VmxTransient.pVmcsInfo == &pVCpu->hm.s.vmx.VmcsInfoNstGst);
11290
11291	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
11292	for (;;)
11293	{
11294	Assert(!HMR0SuspendPending());
11295	HMVMX_ASSERT_CPU_SAFE(pVCpu);
11296	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
11297
11298	/*
11299	* Preparatory work for running guest code, this may force us to
11300	* return to ring-3.
11301	*
11302	* Warning! This bugger disables interrupts on VINF_SUCCESS!
11303	*/
11304	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, false /* fStepping */);
11305	if (rcStrict != VINF_SUCCESS)
11306	break;
11307
11308	/* Interrupts are disabled at this point! */
11309	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
11310	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
11311	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
11312	/* Interrupts are re-enabled at this point! */
11313
11314	/*
11315	* Check for errors with running the VM (VMLAUNCH/VMRESUME).
11316	*/
11317	if (RT_SUCCESS(rcRun))
11318	{ /* very likely */ }
11319	else
11320	{
11321	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
11322	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
11323	return rcRun;
11324	}
11325
11326	/*
11327	* Profile the VM-exit.
11328	*/
11329	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
11330	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
11331	STAM_COUNTER_INC(&pVCpu->hm.s.StatNestedExitAll);
11332	STAM_COUNTER_INC(&pVCpu->hm.s.paStatNestedExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
11333	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
11334	HMVMX_START_EXIT_DISPATCH_PROF();
11335
11336	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
11337
11338	/*
11339	* Handle the VM-exit.
11340	*/
11341	rcStrict = hmR0VmxHandleExitNested(pVCpu, &VmxTransient);
11342	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
11343	if (rcStrict == VINF_SUCCESS)
11344	{
11345	if (!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
11346	{
11347	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
11348	rcStrict = VINF_VMX_VMEXIT;
11349	}
11350	else
11351	{
11352	if (++(*pcLoops) <= cMaxResumeLoops)
11353	continue;
11354	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
11355	rcStrict = VINF_EM_RAW_INTERRUPT;
11356	}
11357	}
11358	else
11359	Assert(rcStrict != VINF_VMX_VMEXIT);
11360	break;
11361	}
11362
11363	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
11364	return rcStrict;
11365	}
11366	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
11367
11368
11369	/** @name Execution loop for single stepping, DBGF events and expensive Dtrace
11370	* probes.
11371	*
11372	* The following few functions and associated structure contains the bloat
11373	* necessary for providing detailed debug events and dtrace probes as well as
11374	* reliable host side single stepping. This works on the principle of
11375	* "subclassing" the normal execution loop and workers. We replace the loop
11376	* method completely and override selected helpers to add necessary adjustments
11377	* to their core operation.
11378	*
11379	* The goal is to keep the "parent" code lean and mean, so as not to sacrifice
11380	* any performance for debug and analysis features.
11381	*
11382	* @{
11383	*/
11384
11385	/**
11386	* Transient per-VCPU debug state of VMCS and related info. we save/restore in
11387	* the debug run loop.
11388	*/
11389	typedef struct VMXRUNDBGSTATE
11390	{
11391	/** The RIP we started executing at. This is for detecting that we stepped. */
11392	uint64_t uRipStart;
11393	/** The CS we started executing with. */
11394	uint16_t uCsStart;
11395
11396	/** Whether we've actually modified the 1st execution control field. */
11397	bool fModifiedProcCtls : 1;
11398	/** Whether we've actually modified the 2nd execution control field. */
11399	bool fModifiedProcCtls2 : 1;
11400	/** Whether we've actually modified the exception bitmap. */
11401	bool fModifiedXcptBitmap : 1;
11402
11403	/** We desire the modified the CR0 mask to be cleared. */
11404	bool fClearCr0Mask : 1;
11405	/** We desire the modified the CR4 mask to be cleared. */
11406	bool fClearCr4Mask : 1;
11407	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC. */
11408	uint32_t fCpe1Extra;
11409	/** Stuff we do not want in VMX_VMCS32_CTRL_PROC_EXEC. */
11410	uint32_t fCpe1Unwanted;
11411	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC2. */
11412	uint32_t fCpe2Extra;
11413	/** Extra stuff we need in VMX_VMCS32_CTRL_EXCEPTION_BITMAP. */
11414	uint32_t bmXcptExtra;
11415	/** The sequence number of the Dtrace provider settings the state was
11416	* configured against. */
11417	uint32_t uDtraceSettingsSeqNo;
11418	/** VM-exits to check (one bit per VM-exit). */
11419	uint32_t bmExitsToCheck[3];
11420
11421	/** The initial VMX_VMCS32_CTRL_PROC_EXEC value (helps with restore). */
11422	uint32_t fProcCtlsInitial;
11423	/** The initial VMX_VMCS32_CTRL_PROC_EXEC2 value (helps with restore). */
11424	uint32_t fProcCtls2Initial;
11425	/** The initial VMX_VMCS32_CTRL_EXCEPTION_BITMAP value (helps with restore). */
11426	uint32_t bmXcptInitial;
11427	} VMXRUNDBGSTATE;
11428	AssertCompileMemberSize(VMXRUNDBGSTATE, bmExitsToCheck, (VMX_EXIT_MAX + 1 + 31) / 32 * 4);
11429	typedef VMXRUNDBGSTATE *PVMXRUNDBGSTATE;
11430
11431
11432	/**
11433	* Initializes the VMXRUNDBGSTATE structure.
11434	*
11435	* @param pVCpu The cross context virtual CPU structure of the
11436	* calling EMT.
11437	* @param pVmxTransient The VMX-transient structure.
11438	* @param pDbgState The debug state to initialize.
11439	*/
11440	static void hmR0VmxRunDebugStateInit(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11441	{
11442	pDbgState->uRipStart = pVCpu->cpum.GstCtx.rip;
11443	pDbgState->uCsStart = pVCpu->cpum.GstCtx.cs.Sel;
11444
11445	pDbgState->fModifiedProcCtls = false;
11446	pDbgState->fModifiedProcCtls2 = false;
11447	pDbgState->fModifiedXcptBitmap = false;
11448	pDbgState->fClearCr0Mask = false;
11449	pDbgState->fClearCr4Mask = false;
11450	pDbgState->fCpe1Extra = 0;
11451	pDbgState->fCpe1Unwanted = 0;
11452	pDbgState->fCpe2Extra = 0;
11453	pDbgState->bmXcptExtra = 0;
11454	pDbgState->fProcCtlsInitial = pVmxTransient->pVmcsInfo->u32ProcCtls;
11455	pDbgState->fProcCtls2Initial = pVmxTransient->pVmcsInfo->u32ProcCtls2;
11456	pDbgState->bmXcptInitial = pVmxTransient->pVmcsInfo->u32XcptBitmap;
11457	}
11458
11459
11460	/**
11461	* Updates the VMSC fields with changes requested by @a pDbgState.
11462	*
11463	* This is performed after hmR0VmxPreRunGuestDebugStateUpdate as well
11464	* immediately before executing guest code, i.e. when interrupts are disabled.
11465	* We don't check status codes here as we cannot easily assert or return in the
11466	* latter case.
11467	*
11468	* @param pVCpu The cross context virtual CPU structure.
11469	* @param pVmxTransient The VMX-transient structure.
11470	* @param pDbgState The debug state.
11471	*/
11472	static void hmR0VmxPreRunGuestDebugStateApply(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11473	{
11474	/*
11475	* Ensure desired flags in VMCS control fields are set.
11476	* (Ignoring write failure here, as we're committed and it's just debug extras.)
11477	*
11478	* Note! We load the shadow CR0 & CR4 bits when we flag the clearing, so
11479	* there should be no stale data in pCtx at this point.
11480	*/
11481	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11482	if ( (pVmcsInfo->u32ProcCtls & pDbgState->fCpe1Extra) != pDbgState->fCpe1Extra
11483	\|\| (pVmcsInfo->u32ProcCtls & pDbgState->fCpe1Unwanted))
11484	{
11485	pVmcsInfo->u32ProcCtls \|= pDbgState->fCpe1Extra;
11486	pVmcsInfo->u32ProcCtls &= ~pDbgState->fCpe1Unwanted;
11487	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
11488	Log6Func(("VMX_VMCS32_CTRL_PROC_EXEC: %#RX32\n", pVmcsInfo->u32ProcCtls));
11489	pDbgState->fModifiedProcCtls = true;
11490	}
11491
11492	if ((pVmcsInfo->u32ProcCtls2 & pDbgState->fCpe2Extra) != pDbgState->fCpe2Extra)
11493	{
11494	pVmcsInfo->u32ProcCtls2 \|= pDbgState->fCpe2Extra;
11495	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pVmcsInfo->u32ProcCtls2);
11496	Log6Func(("VMX_VMCS32_CTRL_PROC_EXEC2: %#RX32\n", pVmcsInfo->u32ProcCtls2));
11497	pDbgState->fModifiedProcCtls2 = true;
11498	}
11499
11500	if ((pVmcsInfo->u32XcptBitmap & pDbgState->bmXcptExtra) != pDbgState->bmXcptExtra)
11501	{
11502	pVmcsInfo->u32XcptBitmap \|= pDbgState->bmXcptExtra;
11503	VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVmcsInfo->u32XcptBitmap);
11504	Log6Func(("VMX_VMCS32_CTRL_EXCEPTION_BITMAP: %#RX32\n", pVmcsInfo->u32XcptBitmap));
11505	pDbgState->fModifiedXcptBitmap = true;
11506	}
11507
11508	if (pDbgState->fClearCr0Mask && pVmcsInfo->u64Cr0Mask != 0)
11509	{
11510	pVmcsInfo->u64Cr0Mask = 0;
11511	VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, 0);
11512	Log6Func(("VMX_VMCS_CTRL_CR0_MASK: 0\n"));
11513	}
11514
11515	if (pDbgState->fClearCr4Mask && pVmcsInfo->u64Cr4Mask != 0)
11516	{
11517	pVmcsInfo->u64Cr4Mask = 0;
11518	VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, 0);
11519	Log6Func(("VMX_VMCS_CTRL_CR4_MASK: 0\n"));
11520	}
11521
11522	NOREF(pVCpu);
11523	}
11524
11525
11526	/**
11527	* Restores VMCS fields that were changed by hmR0VmxPreRunGuestDebugStateApply for
11528	* re-entry next time around.
11529	*
11530	* @returns Strict VBox status code (i.e. informational status codes too).
11531	* @param pVCpu The cross context virtual CPU structure.
11532	* @param pVmxTransient The VMX-transient structure.
11533	* @param pDbgState The debug state.
11534	* @param rcStrict The return code from executing the guest using single
11535	* stepping.
11536	*/
11537	static VBOXSTRICTRC hmR0VmxRunDebugStateRevert(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState,
11538	VBOXSTRICTRC rcStrict)
11539	{
11540	/*
11541	* Restore VM-exit control settings as we may not reenter this function the
11542	* next time around.
11543	*/
11544	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11545
11546	/* We reload the initial value, trigger what we can of recalculations the
11547	next time around. From the looks of things, that's all that's required atm. */
11548	if (pDbgState->fModifiedProcCtls)
11549	{
11550	if (!(pDbgState->fProcCtlsInitial & VMX_PROC_CTLS_MOV_DR_EXIT) && CPUMIsHyperDebugStateActive(pVCpu))
11551	pDbgState->fProcCtlsInitial \|= VMX_PROC_CTLS_MOV_DR_EXIT; /* Avoid assertion in hmR0VmxLeave */
11552	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pDbgState->fProcCtlsInitial);
11553	AssertRC(rc2);
11554	pVmcsInfo->u32ProcCtls = pDbgState->fProcCtlsInitial;
11555	}
11556
11557	/* We're currently the only ones messing with this one, so just restore the
11558	cached value and reload the field. */
11559	if ( pDbgState->fModifiedProcCtls2
11560	&& pVmcsInfo->u32ProcCtls2 != pDbgState->fProcCtls2Initial)
11561	{
11562	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pDbgState->fProcCtls2Initial);
11563	AssertRC(rc2);
11564	pVmcsInfo->u32ProcCtls2 = pDbgState->fProcCtls2Initial;
11565	}
11566
11567	/* If we've modified the exception bitmap, we restore it and trigger
11568	reloading and partial recalculation the next time around. */
11569	if (pDbgState->fModifiedXcptBitmap)
11570	pVmcsInfo->u32XcptBitmap = pDbgState->bmXcptInitial;
11571
11572	return rcStrict;
11573	}
11574
11575
11576	/**
11577	* Configures VM-exit controls for current DBGF and DTrace settings.
11578	*
11579	* This updates @a pDbgState and the VMCS execution control fields to reflect
11580	* the necessary VM-exits demanded by DBGF and DTrace.
11581	*
11582	* @param pVCpu The cross context virtual CPU structure.
11583	* @param pVmxTransient The VMX-transient structure. May update
11584	* fUpdatedTscOffsettingAndPreemptTimer.
11585	* @param pDbgState The debug state.
11586	*/
11587	static void hmR0VmxPreRunGuestDebugStateUpdate(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11588	{
11589	/*
11590	* Take down the dtrace serial number so we can spot changes.
11591	*/
11592	pDbgState->uDtraceSettingsSeqNo = VBOXVMM_GET_SETTINGS_SEQ_NO();
11593	ASMCompilerBarrier();
11594
11595	/*
11596	* We'll rebuild most of the middle block of data members (holding the
11597	* current settings) as we go along here, so start by clearing it all.
11598	*/
11599	pDbgState->bmXcptExtra = 0;
11600	pDbgState->fCpe1Extra = 0;
11601	pDbgState->fCpe1Unwanted = 0;
11602	pDbgState->fCpe2Extra = 0;
11603	for (unsigned i = 0; i < RT_ELEMENTS(pDbgState->bmExitsToCheck); i++)
11604	pDbgState->bmExitsToCheck[i] = 0;
11605
11606	/*
11607	* Software interrupts (INT XXh) - no idea how to trigger these...
11608	*/
11609	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
11610	if ( DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_INTERRUPT_SOFTWARE)
11611	\|\| VBOXVMM_INT_SOFTWARE_ENABLED())
11612	{
11613	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
11614	}
11615
11616	/*
11617	* INT3 breakpoints - triggered by #BP exceptions.
11618	*/
11619	if (pVM->dbgf.ro.cEnabledInt3Breakpoints > 0)
11620	pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
11621
11622	/*
11623	* Exception bitmap and XCPT events+probes.
11624	*/
11625	for (int iXcpt = 0; iXcpt < (DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST + 1); iXcpt++)
11626	if (DBGF_IS_EVENT_ENABLED(pVM, (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + iXcpt)))
11627	pDbgState->bmXcptExtra \|= RT_BIT_32(iXcpt);
11628
11629	if (VBOXVMM_XCPT_DE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DE);
11630	if (VBOXVMM_XCPT_DB_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DB);
11631	if (VBOXVMM_XCPT_BP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
11632	if (VBOXVMM_XCPT_OF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_OF);
11633	if (VBOXVMM_XCPT_BR_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BR);
11634	if (VBOXVMM_XCPT_UD_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_UD);
11635	if (VBOXVMM_XCPT_NM_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NM);
11636	if (VBOXVMM_XCPT_DF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DF);
11637	if (VBOXVMM_XCPT_TS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_TS);
11638	if (VBOXVMM_XCPT_NP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NP);
11639	if (VBOXVMM_XCPT_SS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SS);
11640	if (VBOXVMM_XCPT_GP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_GP);
11641	if (VBOXVMM_XCPT_PF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_PF);
11642	if (VBOXVMM_XCPT_MF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_MF);
11643	if (VBOXVMM_XCPT_AC_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_AC);
11644	if (VBOXVMM_XCPT_XF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_XF);
11645	if (VBOXVMM_XCPT_VE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_VE);
11646	if (VBOXVMM_XCPT_SX_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SX);
11647
11648	if (pDbgState->bmXcptExtra)
11649	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
11650
11651	/*
11652	* Process events and probes for VM-exits, making sure we get the wanted VM-exits.
11653	*
11654	* Note! This is the reverse of what hmR0VmxHandleExitDtraceEvents does.
11655	* So, when adding/changing/removing please don't forget to update it.
11656	*
11657	* Some of the macros are picking up local variables to save horizontal space,
11658	* (being able to see it in a table is the lesser evil here).
11659	*/
11660	#define IS_EITHER_ENABLED(a_pVM, a_EventSubName) \
11661	( DBGF_IS_EVENT_ENABLED(a_pVM, RT_CONCAT(DBGFEVENT_, a_EventSubName)) \
11662	\|\| RT_CONCAT3(VBOXVMM_, a_EventSubName, _ENABLED)() )
11663	#define SET_ONLY_XBM_IF_EITHER_EN(a_EventSubName, a_uExit) \
11664	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11665	{ AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11666	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11667	} else do { } while (0)
11668	#define SET_CPE1_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec) \
11669	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11670	{ \
11671	(pDbgState)->fCpe1Extra \|= (a_fCtrlProcExec); \
11672	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11673	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11674	} else do { } while (0)
11675	#define SET_CPEU_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fUnwantedCtrlProcExec) \
11676	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11677	{ \
11678	(pDbgState)->fCpe1Unwanted \|= (a_fUnwantedCtrlProcExec); \
11679	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11680	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11681	} else do { } while (0)
11682	#define SET_CPE2_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec2) \
11683	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11684	{ \
11685	(pDbgState)->fCpe2Extra \|= (a_fCtrlProcExec2); \
11686	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11687	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11688	} else do { } while (0)
11689
11690	SET_ONLY_XBM_IF_EITHER_EN(EXIT_TASK_SWITCH, VMX_EXIT_TASK_SWITCH); /* unconditional */
11691	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_VIOLATION, VMX_EXIT_EPT_VIOLATION); /* unconditional */
11692	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_MISCONFIG, VMX_EXIT_EPT_MISCONFIG); /* unconditional (unless #VE) */
11693	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_ACCESS, VMX_EXIT_APIC_ACCESS); /* feature dependent, nothing to enable here */
11694	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_WRITE, VMX_EXIT_APIC_WRITE); /* feature dependent, nothing to enable here */
11695
11696	SET_ONLY_XBM_IF_EITHER_EN(INSTR_CPUID, VMX_EXIT_CPUID); /* unconditional */
11697	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CPUID, VMX_EXIT_CPUID);
11698	SET_ONLY_XBM_IF_EITHER_EN(INSTR_GETSEC, VMX_EXIT_GETSEC); /* unconditional */
11699	SET_ONLY_XBM_IF_EITHER_EN( EXIT_GETSEC, VMX_EXIT_GETSEC);
11700	SET_CPE1_XBM_IF_EITHER_EN(INSTR_HALT, VMX_EXIT_HLT, VMX_PROC_CTLS_HLT_EXIT); /* paranoia */
11701	SET_ONLY_XBM_IF_EITHER_EN( EXIT_HALT, VMX_EXIT_HLT);
11702	SET_ONLY_XBM_IF_EITHER_EN(INSTR_INVD, VMX_EXIT_INVD); /* unconditional */
11703	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVD, VMX_EXIT_INVD);
11704	SET_CPE1_XBM_IF_EITHER_EN(INSTR_INVLPG, VMX_EXIT_INVLPG, VMX_PROC_CTLS_INVLPG_EXIT);
11705	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVLPG, VMX_EXIT_INVLPG);
11706	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDPMC, VMX_EXIT_RDPMC, VMX_PROC_CTLS_RDPMC_EXIT);
11707	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDPMC, VMX_EXIT_RDPMC);
11708	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSC, VMX_EXIT_RDTSC, VMX_PROC_CTLS_RDTSC_EXIT);
11709	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSC, VMX_EXIT_RDTSC);
11710	SET_ONLY_XBM_IF_EITHER_EN(INSTR_RSM, VMX_EXIT_RSM); /* unconditional */
11711	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RSM, VMX_EXIT_RSM);
11712	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMM_CALL, VMX_EXIT_VMCALL); /* unconditional */
11713	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMM_CALL, VMX_EXIT_VMCALL);
11714	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMCLEAR, VMX_EXIT_VMCLEAR); /* unconditional */
11715	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMCLEAR, VMX_EXIT_VMCLEAR);
11716	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH); /* unconditional */
11717	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH);
11718	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRLD, VMX_EXIT_VMPTRLD); /* unconditional */
11719	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRLD, VMX_EXIT_VMPTRLD);
11720	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRST, VMX_EXIT_VMPTRST); /* unconditional */
11721	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRST, VMX_EXIT_VMPTRST);
11722	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMREAD, VMX_EXIT_VMREAD); /* unconditional */
11723	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMREAD, VMX_EXIT_VMREAD);
11724	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMRESUME, VMX_EXIT_VMRESUME); /* unconditional */
11725	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMRESUME, VMX_EXIT_VMRESUME);
11726	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMWRITE, VMX_EXIT_VMWRITE); /* unconditional */
11727	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMWRITE, VMX_EXIT_VMWRITE);
11728	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXOFF, VMX_EXIT_VMXOFF); /* unconditional */
11729	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXOFF, VMX_EXIT_VMXOFF);
11730	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXON, VMX_EXIT_VMXON); /* unconditional */
11731	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXON, VMX_EXIT_VMXON);
11732
11733	if ( IS_EITHER_ENABLED(pVM, INSTR_CRX_READ)
11734	\|\| IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
11735	{
11736	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_CR4
11737	\| CPUMCTX_EXTRN_APIC_TPR);
11738	AssertRC(rc);
11739
11740	#if 0 /** @todo fix me */
11741	pDbgState->fClearCr0Mask = true;
11742	pDbgState->fClearCr4Mask = true;
11743	#endif
11744	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_READ))
11745	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_CR3_STORE_EXIT \| VMX_PROC_CTLS_CR8_STORE_EXIT;
11746	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
11747	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_CR3_LOAD_EXIT \| VMX_PROC_CTLS_CR8_LOAD_EXIT;
11748	pDbgState->fCpe1Unwanted \|= VMX_PROC_CTLS_USE_TPR_SHADOW; /* risky? */
11749	/* Note! We currently don't use VMX_VMCS32_CTRL_CR3_TARGET_COUNT. It would
11750	require clearing here and in the loop if we start using it. */
11751	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_CRX);
11752	}
11753	else
11754	{
11755	if (pDbgState->fClearCr0Mask)
11756	{
11757	pDbgState->fClearCr0Mask = false;
11758	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR0);
11759	}
11760	if (pDbgState->fClearCr4Mask)
11761	{
11762	pDbgState->fClearCr4Mask = false;
11763	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR4);
11764	}
11765	}
11766	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_READ, VMX_EXIT_MOV_CRX);
11767	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_WRITE, VMX_EXIT_MOV_CRX);
11768
11769	if ( IS_EITHER_ENABLED(pVM, INSTR_DRX_READ)
11770	\|\| IS_EITHER_ENABLED(pVM, INSTR_DRX_WRITE))
11771	{
11772	/** @todo later, need to fix handler as it assumes this won't usually happen. */
11773	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_DRX);
11774	}
11775	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_READ, VMX_EXIT_MOV_DRX);
11776	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_WRITE, VMX_EXIT_MOV_DRX);
11777
11778	SET_CPEU_XBM_IF_EITHER_EN(INSTR_RDMSR, VMX_EXIT_RDMSR, VMX_PROC_CTLS_USE_MSR_BITMAPS); /* risky clearing this? */
11779	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDMSR, VMX_EXIT_RDMSR);
11780	SET_CPEU_XBM_IF_EITHER_EN(INSTR_WRMSR, VMX_EXIT_WRMSR, VMX_PROC_CTLS_USE_MSR_BITMAPS);
11781	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WRMSR, VMX_EXIT_WRMSR);
11782	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MWAIT, VMX_EXIT_MWAIT, VMX_PROC_CTLS_MWAIT_EXIT); /* paranoia */
11783	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MWAIT, VMX_EXIT_MWAIT);
11784	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MONITOR, VMX_EXIT_MONITOR, VMX_PROC_CTLS_MONITOR_EXIT); /* paranoia */
11785	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MONITOR, VMX_EXIT_MONITOR);
11786	#if 0 /** @todo too slow, fix handler. */
11787	SET_CPE1_XBM_IF_EITHER_EN(INSTR_PAUSE, VMX_EXIT_PAUSE, VMX_PROC_CTLS_PAUSE_EXIT);
11788	#endif
11789	SET_ONLY_XBM_IF_EITHER_EN( EXIT_PAUSE, VMX_EXIT_PAUSE);
11790
11791	if ( IS_EITHER_ENABLED(pVM, INSTR_SGDT)
11792	\|\| IS_EITHER_ENABLED(pVM, INSTR_SIDT)
11793	\|\| IS_EITHER_ENABLED(pVM, INSTR_LGDT)
11794	\|\| IS_EITHER_ENABLED(pVM, INSTR_LIDT))
11795	{
11796	pDbgState->fCpe2Extra \|= VMX_PROC_CTLS2_DESC_TABLE_EXIT;
11797	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_GDTR_IDTR_ACCESS);
11798	}
11799	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SGDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11800	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SIDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11801	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LGDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11802	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LIDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11803
11804	if ( IS_EITHER_ENABLED(pVM, INSTR_SLDT)
11805	\|\| IS_EITHER_ENABLED(pVM, INSTR_STR)
11806	\|\| IS_EITHER_ENABLED(pVM, INSTR_LLDT)
11807	\|\| IS_EITHER_ENABLED(pVM, INSTR_LTR))
11808	{
11809	pDbgState->fCpe2Extra \|= VMX_PROC_CTLS2_DESC_TABLE_EXIT;
11810	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_LDTR_TR_ACCESS);
11811	}
11812	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SLDT, VMX_EXIT_LDTR_TR_ACCESS);
11813	SET_ONLY_XBM_IF_EITHER_EN( EXIT_STR, VMX_EXIT_LDTR_TR_ACCESS);
11814	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LLDT, VMX_EXIT_LDTR_TR_ACCESS);
11815	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LTR, VMX_EXIT_LDTR_TR_ACCESS);
11816
11817	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVEPT, VMX_EXIT_INVEPT); /* unconditional */
11818	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVEPT, VMX_EXIT_INVEPT);
11819	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSCP, VMX_EXIT_RDTSCP, VMX_PROC_CTLS_RDTSC_EXIT);
11820	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSCP, VMX_EXIT_RDTSCP);
11821	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVVPID, VMX_EXIT_INVVPID); /* unconditional */
11822	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVVPID, VMX_EXIT_INVVPID);
11823	SET_CPE2_XBM_IF_EITHER_EN(INSTR_WBINVD, VMX_EXIT_WBINVD, VMX_PROC_CTLS2_WBINVD_EXIT);
11824	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WBINVD, VMX_EXIT_WBINVD);
11825	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSETBV, VMX_EXIT_XSETBV); /* unconditional */
11826	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XSETBV, VMX_EXIT_XSETBV);
11827	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDRAND, VMX_EXIT_RDRAND, VMX_PROC_CTLS2_RDRAND_EXIT);
11828	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDRAND, VMX_EXIT_RDRAND);
11829	SET_CPE1_XBM_IF_EITHER_EN(INSTR_VMX_INVPCID, VMX_EXIT_INVPCID, VMX_PROC_CTLS_INVLPG_EXIT);
11830	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVPCID, VMX_EXIT_INVPCID);
11831	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMFUNC, VMX_EXIT_VMFUNC); /* unconditional for the current setup */
11832	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMFUNC, VMX_EXIT_VMFUNC);
11833	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDSEED, VMX_EXIT_RDSEED, VMX_PROC_CTLS2_RDSEED_EXIT);
11834	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDSEED, VMX_EXIT_RDSEED);
11835	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSAVES, VMX_EXIT_XSAVES); /* unconditional (enabled by host, guest cfg) */
11836	SET_ONLY_XBM_IF_EITHER_EN(EXIT_XSAVES, VMX_EXIT_XSAVES);
11837	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XRSTORS, VMX_EXIT_XRSTORS); /* unconditional (enabled by host, guest cfg) */
11838	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XRSTORS, VMX_EXIT_XRSTORS);
11839
11840	#undef IS_EITHER_ENABLED
11841	#undef SET_ONLY_XBM_IF_EITHER_EN
11842	#undef SET_CPE1_XBM_IF_EITHER_EN
11843	#undef SET_CPEU_XBM_IF_EITHER_EN
11844	#undef SET_CPE2_XBM_IF_EITHER_EN
11845
11846	/*
11847	* Sanitize the control stuff.
11848	*/
11849	pDbgState->fCpe2Extra &= pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1;
11850	if (pDbgState->fCpe2Extra)
11851	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_USE_SECONDARY_CTLS;
11852	pDbgState->fCpe1Extra &= pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1;
11853	pDbgState->fCpe1Unwanted &= ~pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed0;
11854	if (pVCpu->hm.s.fDebugWantRdTscExit != RT_BOOL(pDbgState->fCpe1Extra & VMX_PROC_CTLS_RDTSC_EXIT))
11855	{
11856	pVCpu->hm.s.fDebugWantRdTscExit ^= true;
11857	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
11858	}
11859
11860	Log6(("HM: debug state: cpe1=%#RX32 cpeu=%#RX32 cpe2=%#RX32%s%s\n",
11861	pDbgState->fCpe1Extra, pDbgState->fCpe1Unwanted, pDbgState->fCpe2Extra,
11862	pDbgState->fClearCr0Mask ? " clr-cr0" : "",
11863	pDbgState->fClearCr4Mask ? " clr-cr4" : ""));
11864	}
11865
11866
11867	/**
11868	* Fires off DBGF events and dtrace probes for a VM-exit, when it's
11869	* appropriate.
11870	*
11871	* The caller has checked the VM-exit against the
11872	* VMXRUNDBGSTATE::bmExitsToCheck bitmap. The caller has checked for NMIs
11873	* already, so we don't have to do that either.
11874	*
11875	* @returns Strict VBox status code (i.e. informational status codes too).
11876	* @param pVCpu The cross context virtual CPU structure.
11877	* @param pVmxTransient The VMX-transient structure.
11878	* @param uExitReason The VM-exit reason.
11879	*
11880	* @remarks The name of this function is displayed by dtrace, so keep it short
11881	* and to the point. No longer than 33 chars long, please.
11882	*/
11883	static VBOXSTRICTRC hmR0VmxHandleExitDtraceEvents(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uExitReason)
11884	{
11885	/*
11886	* Translate the event into a DBGF event (enmEvent + uEventArg) and at the
11887	* same time check whether any corresponding Dtrace event is enabled (fDtrace).
11888	*
11889	* Note! This is the reverse operation of what hmR0VmxPreRunGuestDebugStateUpdate
11890	* does. Must add/change/remove both places. Same ordering, please.
11891	*
11892	* Added/removed events must also be reflected in the next section
11893	* where we dispatch dtrace events.
11894	*/
11895	bool fDtrace1 = false;
11896	bool fDtrace2 = false;
11897	DBGFEVENTTYPE enmEvent1 = DBGFEVENT_END;
11898	DBGFEVENTTYPE enmEvent2 = DBGFEVENT_END;
11899	uint32_t uEventArg = 0;
11900	#define SET_EXIT(a_EventSubName) \
11901	do { \
11902	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
11903	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
11904	} while (0)
11905	#define SET_BOTH(a_EventSubName) \
11906	do { \
11907	enmEvent1 = RT_CONCAT(DBGFEVENT_INSTR_, a_EventSubName); \
11908	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
11909	fDtrace1 = RT_CONCAT3(VBOXVMM_INSTR_, a_EventSubName, _ENABLED)(); \
11910	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
11911	} while (0)
11912	switch (uExitReason)
11913	{
11914	case VMX_EXIT_MTF:
11915	return hmR0VmxExitMtf(pVCpu, pVmxTransient);
11916
11917	case VMX_EXIT_XCPT_OR_NMI:
11918	{
11919	uint8_t const idxVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
11920	switch (VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo))
11921	{
11922	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
11923	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
11924	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
11925	if (idxVector <= (unsigned)(DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST))
11926	{
11927	if (VMX_EXIT_INT_INFO_IS_ERROR_CODE_VALID(pVmxTransient->uExitIntInfo))
11928	{
11929	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
11930	uEventArg = pVmxTransient->uExitIntErrorCode;
11931	}
11932	enmEvent1 = (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + idxVector);
11933	switch (enmEvent1)
11934	{
11935	case DBGFEVENT_XCPT_DE: fDtrace1 = VBOXVMM_XCPT_DE_ENABLED(); break;
11936	case DBGFEVENT_XCPT_DB: fDtrace1 = VBOXVMM_XCPT_DB_ENABLED(); break;
11937	case DBGFEVENT_XCPT_BP: fDtrace1 = VBOXVMM_XCPT_BP_ENABLED(); break;
11938	case DBGFEVENT_XCPT_OF: fDtrace1 = VBOXVMM_XCPT_OF_ENABLED(); break;
11939	case DBGFEVENT_XCPT_BR: fDtrace1 = VBOXVMM_XCPT_BR_ENABLED(); break;
11940	case DBGFEVENT_XCPT_UD: fDtrace1 = VBOXVMM_XCPT_UD_ENABLED(); break;
11941	case DBGFEVENT_XCPT_NM: fDtrace1 = VBOXVMM_XCPT_NM_ENABLED(); break;
11942	case DBGFEVENT_XCPT_DF: fDtrace1 = VBOXVMM_XCPT_DF_ENABLED(); break;
11943	case DBGFEVENT_XCPT_TS: fDtrace1 = VBOXVMM_XCPT_TS_ENABLED(); break;
11944	case DBGFEVENT_XCPT_NP: fDtrace1 = VBOXVMM_XCPT_NP_ENABLED(); break;
11945	case DBGFEVENT_XCPT_SS: fDtrace1 = VBOXVMM_XCPT_SS_ENABLED(); break;
11946	case DBGFEVENT_XCPT_GP: fDtrace1 = VBOXVMM_XCPT_GP_ENABLED(); break;
11947	case DBGFEVENT_XCPT_PF: fDtrace1 = VBOXVMM_XCPT_PF_ENABLED(); break;
11948	case DBGFEVENT_XCPT_MF: fDtrace1 = VBOXVMM_XCPT_MF_ENABLED(); break;
11949	case DBGFEVENT_XCPT_AC: fDtrace1 = VBOXVMM_XCPT_AC_ENABLED(); break;
11950	case DBGFEVENT_XCPT_XF: fDtrace1 = VBOXVMM_XCPT_XF_ENABLED(); break;
11951	case DBGFEVENT_XCPT_VE: fDtrace1 = VBOXVMM_XCPT_VE_ENABLED(); break;
11952	case DBGFEVENT_XCPT_SX: fDtrace1 = VBOXVMM_XCPT_SX_ENABLED(); break;
11953	default: break;
11954	}
11955	}
11956	else
11957	AssertFailed();
11958	break;
11959
11960	case VMX_EXIT_INT_INFO_TYPE_SW_INT:
11961	uEventArg = idxVector;
11962	enmEvent1 = DBGFEVENT_INTERRUPT_SOFTWARE;
11963	fDtrace1 = VBOXVMM_INT_SOFTWARE_ENABLED();
11964	break;
11965	}
11966	break;
11967	}
11968
11969	case VMX_EXIT_TRIPLE_FAULT:
11970	enmEvent1 = DBGFEVENT_TRIPLE_FAULT;
11971	//fDtrace1 = VBOXVMM_EXIT_TRIPLE_FAULT_ENABLED();
11972	break;
11973	case VMX_EXIT_TASK_SWITCH: SET_EXIT(TASK_SWITCH); break;
11974	case VMX_EXIT_EPT_VIOLATION: SET_EXIT(VMX_EPT_VIOLATION); break;
11975	case VMX_EXIT_EPT_MISCONFIG: SET_EXIT(VMX_EPT_MISCONFIG); break;
11976	case VMX_EXIT_APIC_ACCESS: SET_EXIT(VMX_VAPIC_ACCESS); break;
11977	case VMX_EXIT_APIC_WRITE: SET_EXIT(VMX_VAPIC_WRITE); break;
11978
11979	/* Instruction specific VM-exits: */
11980	case VMX_EXIT_CPUID: SET_BOTH(CPUID); break;
11981	case VMX_EXIT_GETSEC: SET_BOTH(GETSEC); break;
11982	case VMX_EXIT_HLT: SET_BOTH(HALT); break;
11983	case VMX_EXIT_INVD: SET_BOTH(INVD); break;
11984	case VMX_EXIT_INVLPG: SET_BOTH(INVLPG); break;
11985	case VMX_EXIT_RDPMC: SET_BOTH(RDPMC); break;
11986	case VMX_EXIT_RDTSC: SET_BOTH(RDTSC); break;
11987	case VMX_EXIT_RSM: SET_BOTH(RSM); break;
11988	case VMX_EXIT_VMCALL: SET_BOTH(VMM_CALL); break;
11989	case VMX_EXIT_VMCLEAR: SET_BOTH(VMX_VMCLEAR); break;
11990	case VMX_EXIT_VMLAUNCH: SET_BOTH(VMX_VMLAUNCH); break;
11991	case VMX_EXIT_VMPTRLD: SET_BOTH(VMX_VMPTRLD); break;
11992	case VMX_EXIT_VMPTRST: SET_BOTH(VMX_VMPTRST); break;
11993	case VMX_EXIT_VMREAD: SET_BOTH(VMX_VMREAD); break;
11994	case VMX_EXIT_VMRESUME: SET_BOTH(VMX_VMRESUME); break;
11995	case VMX_EXIT_VMWRITE: SET_BOTH(VMX_VMWRITE); break;
11996	case VMX_EXIT_VMXOFF: SET_BOTH(VMX_VMXOFF); break;
11997	case VMX_EXIT_VMXON: SET_BOTH(VMX_VMXON); break;
11998	case VMX_EXIT_MOV_CRX:
11999	hmR0VmxReadExitQualVmcs(pVmxTransient);
12000	if (VMX_EXIT_QUAL_CRX_ACCESS(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_CRX_ACCESS_READ)
12001	SET_BOTH(CRX_READ);
12002	else
12003	SET_BOTH(CRX_WRITE);
12004	uEventArg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
12005	break;
12006	case VMX_EXIT_MOV_DRX:
12007	hmR0VmxReadExitQualVmcs(pVmxTransient);
12008	if ( VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual)
12009	== VMX_EXIT_QUAL_DRX_DIRECTION_READ)
12010	SET_BOTH(DRX_READ);
12011	else
12012	SET_BOTH(DRX_WRITE);
12013	uEventArg = VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual);
12014	break;
12015	case VMX_EXIT_RDMSR: SET_BOTH(RDMSR); break;
12016	case VMX_EXIT_WRMSR: SET_BOTH(WRMSR); break;
12017	case VMX_EXIT_MWAIT: SET_BOTH(MWAIT); break;
12018	case VMX_EXIT_MONITOR: SET_BOTH(MONITOR); break;
12019	case VMX_EXIT_PAUSE: SET_BOTH(PAUSE); break;
12020	case VMX_EXIT_GDTR_IDTR_ACCESS:
12021	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
12022	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_BF_XDTR_INSINFO_INSTR_ID))
12023	{
12024	case VMX_XDTR_INSINFO_II_SGDT: SET_BOTH(SGDT); break;
12025	case VMX_XDTR_INSINFO_II_SIDT: SET_BOTH(SIDT); break;
12026	case VMX_XDTR_INSINFO_II_LGDT: SET_BOTH(LGDT); break;
12027	case VMX_XDTR_INSINFO_II_LIDT: SET_BOTH(LIDT); break;
12028	}
12029	break;
12030
12031	case VMX_EXIT_LDTR_TR_ACCESS:
12032	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
12033	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_BF_YYTR_INSINFO_INSTR_ID))
12034	{
12035	case VMX_YYTR_INSINFO_II_SLDT: SET_BOTH(SLDT); break;
12036	case VMX_YYTR_INSINFO_II_STR: SET_BOTH(STR); break;
12037	case VMX_YYTR_INSINFO_II_LLDT: SET_BOTH(LLDT); break;
12038	case VMX_YYTR_INSINFO_II_LTR: SET_BOTH(LTR); break;
12039	}
12040	break;
12041
12042	case VMX_EXIT_INVEPT: SET_BOTH(VMX_INVEPT); break;
12043	case VMX_EXIT_RDTSCP: SET_BOTH(RDTSCP); break;
12044	case VMX_EXIT_INVVPID: SET_BOTH(VMX_INVVPID); break;
12045	case VMX_EXIT_WBINVD: SET_BOTH(WBINVD); break;
12046	case VMX_EXIT_XSETBV: SET_BOTH(XSETBV); break;
12047	case VMX_EXIT_RDRAND: SET_BOTH(RDRAND); break;
12048	case VMX_EXIT_INVPCID: SET_BOTH(VMX_INVPCID); break;
12049	case VMX_EXIT_VMFUNC: SET_BOTH(VMX_VMFUNC); break;
12050	case VMX_EXIT_RDSEED: SET_BOTH(RDSEED); break;
12051	case VMX_EXIT_XSAVES: SET_BOTH(XSAVES); break;
12052	case VMX_EXIT_XRSTORS: SET_BOTH(XRSTORS); break;
12053
12054	/* Events that aren't relevant at this point. */
12055	case VMX_EXIT_EXT_INT:
12056	case VMX_EXIT_INT_WINDOW:
12057	case VMX_EXIT_NMI_WINDOW:
12058	case VMX_EXIT_TPR_BELOW_THRESHOLD:
12059	case VMX_EXIT_PREEMPT_TIMER:
12060	case VMX_EXIT_IO_INSTR:
12061	break;
12062
12063	/* Errors and unexpected events. */
12064	case VMX_EXIT_INIT_SIGNAL:
12065	case VMX_EXIT_SIPI:
12066	case VMX_EXIT_IO_SMI:
12067	case VMX_EXIT_SMI:
12068	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
12069	case VMX_EXIT_ERR_MSR_LOAD:
12070	case VMX_EXIT_ERR_MACHINE_CHECK:
12071	case VMX_EXIT_PML_FULL:
12072	case VMX_EXIT_VIRTUALIZED_EOI:
12073	break;
12074
12075	default:
12076	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
12077	break;
12078	}
12079	#undef SET_BOTH
12080	#undef SET_EXIT
12081
12082	/*
12083	* Dtrace tracepoints go first. We do them here at once so we don't
12084	* have to copy the guest state saving and stuff a few dozen times.
12085	* Down side is that we've got to repeat the switch, though this time
12086	* we use enmEvent since the probes are a subset of what DBGF does.
12087	*/
12088	if (fDtrace1 \|\| fDtrace2)
12089	{
12090	hmR0VmxReadExitQualVmcs(pVmxTransient);
12091	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
12092	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
12093	switch (enmEvent1)
12094	{
12095	/** @todo consider which extra parameters would be helpful for each probe. */
12096	case DBGFEVENT_END: break;
12097	case DBGFEVENT_XCPT_DE: VBOXVMM_XCPT_DE(pVCpu, pCtx); break;
12098	case DBGFEVENT_XCPT_DB: VBOXVMM_XCPT_DB(pVCpu, pCtx, pCtx->dr[6]); break;
12099	case DBGFEVENT_XCPT_BP: VBOXVMM_XCPT_BP(pVCpu, pCtx); break;
12100	case DBGFEVENT_XCPT_OF: VBOXVMM_XCPT_OF(pVCpu, pCtx); break;
12101	case DBGFEVENT_XCPT_BR: VBOXVMM_XCPT_BR(pVCpu, pCtx); break;
12102	case DBGFEVENT_XCPT_UD: VBOXVMM_XCPT_UD(pVCpu, pCtx); break;
12103	case DBGFEVENT_XCPT_NM: VBOXVMM_XCPT_NM(pVCpu, pCtx); break;
12104	case DBGFEVENT_XCPT_DF: VBOXVMM_XCPT_DF(pVCpu, pCtx); break;
12105	case DBGFEVENT_XCPT_TS: VBOXVMM_XCPT_TS(pVCpu, pCtx, uEventArg); break;
12106	case DBGFEVENT_XCPT_NP: VBOXVMM_XCPT_NP(pVCpu, pCtx, uEventArg); break;
12107	case DBGFEVENT_XCPT_SS: VBOXVMM_XCPT_SS(pVCpu, pCtx, uEventArg); break;
12108	case DBGFEVENT_XCPT_GP: VBOXVMM_XCPT_GP(pVCpu, pCtx, uEventArg); break;
12109	case DBGFEVENT_XCPT_PF: VBOXVMM_XCPT_PF(pVCpu, pCtx, uEventArg, pCtx->cr2); break;
12110	case DBGFEVENT_XCPT_MF: VBOXVMM_XCPT_MF(pVCpu, pCtx); break;
12111	case DBGFEVENT_XCPT_AC: VBOXVMM_XCPT_AC(pVCpu, pCtx); break;
12112	case DBGFEVENT_XCPT_XF: VBOXVMM_XCPT_XF(pVCpu, pCtx); break;
12113	case DBGFEVENT_XCPT_VE: VBOXVMM_XCPT_VE(pVCpu, pCtx); break;
12114	case DBGFEVENT_XCPT_SX: VBOXVMM_XCPT_SX(pVCpu, pCtx, uEventArg); break;
12115	case DBGFEVENT_INTERRUPT_SOFTWARE: VBOXVMM_INT_SOFTWARE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12116	case DBGFEVENT_INSTR_CPUID: VBOXVMM_INSTR_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
12117	case DBGFEVENT_INSTR_GETSEC: VBOXVMM_INSTR_GETSEC(pVCpu, pCtx); break;
12118	case DBGFEVENT_INSTR_HALT: VBOXVMM_INSTR_HALT(pVCpu, pCtx); break;
12119	case DBGFEVENT_INSTR_INVD: VBOXVMM_INSTR_INVD(pVCpu, pCtx); break;
12120	case DBGFEVENT_INSTR_INVLPG: VBOXVMM_INSTR_INVLPG(pVCpu, pCtx); break;
12121	case DBGFEVENT_INSTR_RDPMC: VBOXVMM_INSTR_RDPMC(pVCpu, pCtx); break;
12122	case DBGFEVENT_INSTR_RDTSC: VBOXVMM_INSTR_RDTSC(pVCpu, pCtx); break;
12123	case DBGFEVENT_INSTR_RSM: VBOXVMM_INSTR_RSM(pVCpu, pCtx); break;
12124	case DBGFEVENT_INSTR_CRX_READ: VBOXVMM_INSTR_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12125	case DBGFEVENT_INSTR_CRX_WRITE: VBOXVMM_INSTR_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12126	case DBGFEVENT_INSTR_DRX_READ: VBOXVMM_INSTR_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12127	case DBGFEVENT_INSTR_DRX_WRITE: VBOXVMM_INSTR_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12128	case DBGFEVENT_INSTR_RDMSR: VBOXVMM_INSTR_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
12129	case DBGFEVENT_INSTR_WRMSR: VBOXVMM_INSTR_WRMSR(pVCpu, pCtx, pCtx->ecx,
12130	RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
12131	case DBGFEVENT_INSTR_MWAIT: VBOXVMM_INSTR_MWAIT(pVCpu, pCtx); break;
12132	case DBGFEVENT_INSTR_MONITOR: VBOXVMM_INSTR_MONITOR(pVCpu, pCtx); break;
12133	case DBGFEVENT_INSTR_PAUSE: VBOXVMM_INSTR_PAUSE(pVCpu, pCtx); break;
12134	case DBGFEVENT_INSTR_SGDT: VBOXVMM_INSTR_SGDT(pVCpu, pCtx); break;
12135	case DBGFEVENT_INSTR_SIDT: VBOXVMM_INSTR_SIDT(pVCpu, pCtx); break;
12136	case DBGFEVENT_INSTR_LGDT: VBOXVMM_INSTR_LGDT(pVCpu, pCtx); break;
12137	case DBGFEVENT_INSTR_LIDT: VBOXVMM_INSTR_LIDT(pVCpu, pCtx); break;
12138	case DBGFEVENT_INSTR_SLDT: VBOXVMM_INSTR_SLDT(pVCpu, pCtx); break;
12139	case DBGFEVENT_INSTR_STR: VBOXVMM_INSTR_STR(pVCpu, pCtx); break;
12140	case DBGFEVENT_INSTR_LLDT: VBOXVMM_INSTR_LLDT(pVCpu, pCtx); break;
12141	case DBGFEVENT_INSTR_LTR: VBOXVMM_INSTR_LTR(pVCpu, pCtx); break;
12142	case DBGFEVENT_INSTR_RDTSCP: VBOXVMM_INSTR_RDTSCP(pVCpu, pCtx); break;
12143	case DBGFEVENT_INSTR_WBINVD: VBOXVMM_INSTR_WBINVD(pVCpu, pCtx); break;
12144	case DBGFEVENT_INSTR_XSETBV: VBOXVMM_INSTR_XSETBV(pVCpu, pCtx); break;
12145	case DBGFEVENT_INSTR_RDRAND: VBOXVMM_INSTR_RDRAND(pVCpu, pCtx); break;
12146	case DBGFEVENT_INSTR_RDSEED: VBOXVMM_INSTR_RDSEED(pVCpu, pCtx); break;
12147	case DBGFEVENT_INSTR_XSAVES: VBOXVMM_INSTR_XSAVES(pVCpu, pCtx); break;
12148	case DBGFEVENT_INSTR_XRSTORS: VBOXVMM_INSTR_XRSTORS(pVCpu, pCtx); break;
12149	case DBGFEVENT_INSTR_VMM_CALL: VBOXVMM_INSTR_VMM_CALL(pVCpu, pCtx); break;
12150	case DBGFEVENT_INSTR_VMX_VMCLEAR: VBOXVMM_INSTR_VMX_VMCLEAR(pVCpu, pCtx); break;
12151	case DBGFEVENT_INSTR_VMX_VMLAUNCH: VBOXVMM_INSTR_VMX_VMLAUNCH(pVCpu, pCtx); break;
12152	case DBGFEVENT_INSTR_VMX_VMPTRLD: VBOXVMM_INSTR_VMX_VMPTRLD(pVCpu, pCtx); break;
12153	case DBGFEVENT_INSTR_VMX_VMPTRST: VBOXVMM_INSTR_VMX_VMPTRST(pVCpu, pCtx); break;
12154	case DBGFEVENT_INSTR_VMX_VMREAD: VBOXVMM_INSTR_VMX_VMREAD(pVCpu, pCtx); break;
12155	case DBGFEVENT_INSTR_VMX_VMRESUME: VBOXVMM_INSTR_VMX_VMRESUME(pVCpu, pCtx); break;
12156	case DBGFEVENT_INSTR_VMX_VMWRITE: VBOXVMM_INSTR_VMX_VMWRITE(pVCpu, pCtx); break;
12157	case DBGFEVENT_INSTR_VMX_VMXOFF: VBOXVMM_INSTR_VMX_VMXOFF(pVCpu, pCtx); break;
12158	case DBGFEVENT_INSTR_VMX_VMXON: VBOXVMM_INSTR_VMX_VMXON(pVCpu, pCtx); break;
12159	case DBGFEVENT_INSTR_VMX_INVEPT: VBOXVMM_INSTR_VMX_INVEPT(pVCpu, pCtx); break;
12160	case DBGFEVENT_INSTR_VMX_INVVPID: VBOXVMM_INSTR_VMX_INVVPID(pVCpu, pCtx); break;
12161	case DBGFEVENT_INSTR_VMX_INVPCID: VBOXVMM_INSTR_VMX_INVPCID(pVCpu, pCtx); break;
12162	case DBGFEVENT_INSTR_VMX_VMFUNC: VBOXVMM_INSTR_VMX_VMFUNC(pVCpu, pCtx); break;
12163	default: AssertMsgFailed(("enmEvent1=%d uExitReason=%d\n", enmEvent1, uExitReason)); break;
12164	}
12165	switch (enmEvent2)
12166	{
12167	/** @todo consider which extra parameters would be helpful for each probe. */
12168	case DBGFEVENT_END: break;
12169	case DBGFEVENT_EXIT_TASK_SWITCH: VBOXVMM_EXIT_TASK_SWITCH(pVCpu, pCtx); break;
12170	case DBGFEVENT_EXIT_CPUID: VBOXVMM_EXIT_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
12171	case DBGFEVENT_EXIT_GETSEC: VBOXVMM_EXIT_GETSEC(pVCpu, pCtx); break;
12172	case DBGFEVENT_EXIT_HALT: VBOXVMM_EXIT_HALT(pVCpu, pCtx); break;
12173	case DBGFEVENT_EXIT_INVD: VBOXVMM_EXIT_INVD(pVCpu, pCtx); break;
12174	case DBGFEVENT_EXIT_INVLPG: VBOXVMM_EXIT_INVLPG(pVCpu, pCtx); break;
12175	case DBGFEVENT_EXIT_RDPMC: VBOXVMM_EXIT_RDPMC(pVCpu, pCtx); break;
12176	case DBGFEVENT_EXIT_RDTSC: VBOXVMM_EXIT_RDTSC(pVCpu, pCtx); break;
12177	case DBGFEVENT_EXIT_RSM: VBOXVMM_EXIT_RSM(pVCpu, pCtx); break;
12178	case DBGFEVENT_EXIT_CRX_READ: VBOXVMM_EXIT_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12179	case DBGFEVENT_EXIT_CRX_WRITE: VBOXVMM_EXIT_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12180	case DBGFEVENT_EXIT_DRX_READ: VBOXVMM_EXIT_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12181	case DBGFEVENT_EXIT_DRX_WRITE: VBOXVMM_EXIT_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12182	case DBGFEVENT_EXIT_RDMSR: VBOXVMM_EXIT_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
12183	case DBGFEVENT_EXIT_WRMSR: VBOXVMM_EXIT_WRMSR(pVCpu, pCtx, pCtx->ecx,
12184	RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
12185	case DBGFEVENT_EXIT_MWAIT: VBOXVMM_EXIT_MWAIT(pVCpu, pCtx); break;
12186	case DBGFEVENT_EXIT_MONITOR: VBOXVMM_EXIT_MONITOR(pVCpu, pCtx); break;
12187	case DBGFEVENT_EXIT_PAUSE: VBOXVMM_EXIT_PAUSE(pVCpu, pCtx); break;
12188	case DBGFEVENT_EXIT_SGDT: VBOXVMM_EXIT_SGDT(pVCpu, pCtx); break;
12189	case DBGFEVENT_EXIT_SIDT: VBOXVMM_EXIT_SIDT(pVCpu, pCtx); break;
12190	case DBGFEVENT_EXIT_LGDT: VBOXVMM_EXIT_LGDT(pVCpu, pCtx); break;
12191	case DBGFEVENT_EXIT_LIDT: VBOXVMM_EXIT_LIDT(pVCpu, pCtx); break;
12192	case DBGFEVENT_EXIT_SLDT: VBOXVMM_EXIT_SLDT(pVCpu, pCtx); break;
12193	case DBGFEVENT_EXIT_STR: VBOXVMM_EXIT_STR(pVCpu, pCtx); break;
12194	case DBGFEVENT_EXIT_LLDT: VBOXVMM_EXIT_LLDT(pVCpu, pCtx); break;
12195	case DBGFEVENT_EXIT_LTR: VBOXVMM_EXIT_LTR(pVCpu, pCtx); break;
12196	case DBGFEVENT_EXIT_RDTSCP: VBOXVMM_EXIT_RDTSCP(pVCpu, pCtx); break;
12197	case DBGFEVENT_EXIT_WBINVD: VBOXVMM_EXIT_WBINVD(pVCpu, pCtx); break;
12198	case DBGFEVENT_EXIT_XSETBV: VBOXVMM_EXIT_XSETBV(pVCpu, pCtx); break;
12199	case DBGFEVENT_EXIT_RDRAND: VBOXVMM_EXIT_RDRAND(pVCpu, pCtx); break;
12200	case DBGFEVENT_EXIT_RDSEED: VBOXVMM_EXIT_RDSEED(pVCpu, pCtx); break;
12201	case DBGFEVENT_EXIT_XSAVES: VBOXVMM_EXIT_XSAVES(pVCpu, pCtx); break;
12202	case DBGFEVENT_EXIT_XRSTORS: VBOXVMM_EXIT_XRSTORS(pVCpu, pCtx); break;
12203	case DBGFEVENT_EXIT_VMM_CALL: VBOXVMM_EXIT_VMM_CALL(pVCpu, pCtx); break;
12204	case DBGFEVENT_EXIT_VMX_VMCLEAR: VBOXVMM_EXIT_VMX_VMCLEAR(pVCpu, pCtx); break;
12205	case DBGFEVENT_EXIT_VMX_VMLAUNCH: VBOXVMM_EXIT_VMX_VMLAUNCH(pVCpu, pCtx); break;
12206	case DBGFEVENT_EXIT_VMX_VMPTRLD: VBOXVMM_EXIT_VMX_VMPTRLD(pVCpu, pCtx); break;
12207	case DBGFEVENT_EXIT_VMX_VMPTRST: VBOXVMM_EXIT_VMX_VMPTRST(pVCpu, pCtx); break;
12208	case DBGFEVENT_EXIT_VMX_VMREAD: VBOXVMM_EXIT_VMX_VMREAD(pVCpu, pCtx); break;
12209	case DBGFEVENT_EXIT_VMX_VMRESUME: VBOXVMM_EXIT_VMX_VMRESUME(pVCpu, pCtx); break;
12210	case DBGFEVENT_EXIT_VMX_VMWRITE: VBOXVMM_EXIT_VMX_VMWRITE(pVCpu, pCtx); break;
12211	case DBGFEVENT_EXIT_VMX_VMXOFF: VBOXVMM_EXIT_VMX_VMXOFF(pVCpu, pCtx); break;
12212	case DBGFEVENT_EXIT_VMX_VMXON: VBOXVMM_EXIT_VMX_VMXON(pVCpu, pCtx); break;
12213	case DBGFEVENT_EXIT_VMX_INVEPT: VBOXVMM_EXIT_VMX_INVEPT(pVCpu, pCtx); break;
12214	case DBGFEVENT_EXIT_VMX_INVVPID: VBOXVMM_EXIT_VMX_INVVPID(pVCpu, pCtx); break;
12215	case DBGFEVENT_EXIT_VMX_INVPCID: VBOXVMM_EXIT_VMX_INVPCID(pVCpu, pCtx); break;
12216	case DBGFEVENT_EXIT_VMX_VMFUNC: VBOXVMM_EXIT_VMX_VMFUNC(pVCpu, pCtx); break;
12217	case DBGFEVENT_EXIT_VMX_EPT_MISCONFIG: VBOXVMM_EXIT_VMX_EPT_MISCONFIG(pVCpu, pCtx); break;
12218	case DBGFEVENT_EXIT_VMX_EPT_VIOLATION: VBOXVMM_EXIT_VMX_EPT_VIOLATION(pVCpu, pCtx); break;
12219	case DBGFEVENT_EXIT_VMX_VAPIC_ACCESS: VBOXVMM_EXIT_VMX_VAPIC_ACCESS(pVCpu, pCtx); break;
12220	case DBGFEVENT_EXIT_VMX_VAPIC_WRITE: VBOXVMM_EXIT_VMX_VAPIC_WRITE(pVCpu, pCtx); break;
12221	default: AssertMsgFailed(("enmEvent2=%d uExitReason=%d\n", enmEvent2, uExitReason)); break;
12222	}
12223	}
12224
12225	/*
12226	* Fire of the DBGF event, if enabled (our check here is just a quick one,
12227	* the DBGF call will do a full check).
12228	*
12229	* Note! DBGF sets DBGFEVENT_INTERRUPT_SOFTWARE in the bitmap.
12230	* Note! If we have to events, we prioritize the first, i.e. the instruction
12231	* one, in order to avoid event nesting.
12232	*/
12233	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
12234	if ( enmEvent1 != DBGFEVENT_END
12235	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent1))
12236	{
12237	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12238	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent1, DBGFEVENTCTX_HM, 1, uEventArg);
12239	if (rcStrict != VINF_SUCCESS)
12240	return rcStrict;
12241	}
12242	else if ( enmEvent2 != DBGFEVENT_END
12243	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent2))
12244	{
12245	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12246	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent2, DBGFEVENTCTX_HM, 1, uEventArg);
12247	if (rcStrict != VINF_SUCCESS)
12248	return rcStrict;
12249	}
12250
12251	return VINF_SUCCESS;
12252	}
12253
12254
12255	/**
12256	* Single-stepping VM-exit filtering.
12257	*
12258	* This is preprocessing the VM-exits and deciding whether we've gotten far
12259	* enough to return VINF_EM_DBG_STEPPED already. If not, normal VM-exit
12260	* handling is performed.
12261	*
12262	* @returns Strict VBox status code (i.e. informational status codes too).
12263	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
12264	* @param pVmxTransient The VMX-transient structure.
12265	* @param pDbgState The debug state.
12266	*/
12267	DECLINLINE(VBOXSTRICTRC) hmR0VmxRunDebugHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
12268	{
12269	/*
12270	* Expensive (saves context) generic dtrace VM-exit probe.
12271	*/
12272	uint32_t const uExitReason = pVmxTransient->uExitReason;
12273	if (!VBOXVMM_R0_HMVMX_VMEXIT_ENABLED())
12274	{ /* more likely */ }
12275	else
12276	{
12277	hmR0VmxReadExitQualVmcs(pVmxTransient);
12278	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
12279	AssertRC(rc);
12280	VBOXVMM_R0_HMVMX_VMEXIT(pVCpu, &pVCpu->cpum.GstCtx, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
12281	}
12282
12283	/*
12284	* Check for host NMI, just to get that out of the way.
12285	*/
12286	if (uExitReason != VMX_EXIT_XCPT_OR_NMI)
12287	{ /* normally likely */ }
12288	else
12289	{
12290	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
12291	uint32_t const uIntType = VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo);
12292	if (uIntType == VMX_EXIT_INT_INFO_TYPE_NMI)
12293	return hmR0VmxExitHostNmi(pVCpu, pVmxTransient->pVmcsInfo);
12294	}
12295
12296	/*
12297	* Check for single stepping event if we're stepping.
12298	*/
12299	if (pVCpu->hm.s.fSingleInstruction)
12300	{
12301	switch (uExitReason)
12302	{
12303	case VMX_EXIT_MTF:
12304	return hmR0VmxExitMtf(pVCpu, pVmxTransient);
12305
12306	/* Various events: */
12307	case VMX_EXIT_XCPT_OR_NMI:
12308	case VMX_EXIT_EXT_INT:
12309	case VMX_EXIT_TRIPLE_FAULT:
12310	case VMX_EXIT_INT_WINDOW:
12311	case VMX_EXIT_NMI_WINDOW:
12312	case VMX_EXIT_TASK_SWITCH:
12313	case VMX_EXIT_TPR_BELOW_THRESHOLD:
12314	case VMX_EXIT_APIC_ACCESS:
12315	case VMX_EXIT_EPT_VIOLATION:
12316	case VMX_EXIT_EPT_MISCONFIG:
12317	case VMX_EXIT_PREEMPT_TIMER:
12318
12319	/* Instruction specific VM-exits: */
12320	case VMX_EXIT_CPUID:
12321	case VMX_EXIT_GETSEC:
12322	case VMX_EXIT_HLT:
12323	case VMX_EXIT_INVD:
12324	case VMX_EXIT_INVLPG:
12325	case VMX_EXIT_RDPMC:
12326	case VMX_EXIT_RDTSC:
12327	case VMX_EXIT_RSM:
12328	case VMX_EXIT_VMCALL:
12329	case VMX_EXIT_VMCLEAR:
12330	case VMX_EXIT_VMLAUNCH:
12331	case VMX_EXIT_VMPTRLD:
12332	case VMX_EXIT_VMPTRST:
12333	case VMX_EXIT_VMREAD:
12334	case VMX_EXIT_VMRESUME:
12335	case VMX_EXIT_VMWRITE:
12336	case VMX_EXIT_VMXOFF:
12337	case VMX_EXIT_VMXON:
12338	case VMX_EXIT_MOV_CRX:
12339	case VMX_EXIT_MOV_DRX:
12340	case VMX_EXIT_IO_INSTR:
12341	case VMX_EXIT_RDMSR:
12342	case VMX_EXIT_WRMSR:
12343	case VMX_EXIT_MWAIT:
12344	case VMX_EXIT_MONITOR:
12345	case VMX_EXIT_PAUSE:
12346	case VMX_EXIT_GDTR_IDTR_ACCESS:
12347	case VMX_EXIT_LDTR_TR_ACCESS:
12348	case VMX_EXIT_INVEPT:
12349	case VMX_EXIT_RDTSCP:
12350	case VMX_EXIT_INVVPID:
12351	case VMX_EXIT_WBINVD:
12352	case VMX_EXIT_XSETBV:
12353	case VMX_EXIT_RDRAND:
12354	case VMX_EXIT_INVPCID:
12355	case VMX_EXIT_VMFUNC:
12356	case VMX_EXIT_RDSEED:
12357	case VMX_EXIT_XSAVES:
12358	case VMX_EXIT_XRSTORS:
12359	{
12360	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12361	AssertRCReturn(rc, rc);
12362	if ( pVCpu->cpum.GstCtx.rip != pDbgState->uRipStart
12363	\|\| pVCpu->cpum.GstCtx.cs.Sel != pDbgState->uCsStart)
12364	return VINF_EM_DBG_STEPPED;
12365	break;
12366	}
12367
12368	/* Errors and unexpected events: */
12369	case VMX_EXIT_INIT_SIGNAL:
12370	case VMX_EXIT_SIPI:
12371	case VMX_EXIT_IO_SMI:
12372	case VMX_EXIT_SMI:
12373	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
12374	case VMX_EXIT_ERR_MSR_LOAD:
12375	case VMX_EXIT_ERR_MACHINE_CHECK:
12376	case VMX_EXIT_PML_FULL:
12377	case VMX_EXIT_VIRTUALIZED_EOI:
12378	case VMX_EXIT_APIC_WRITE: /* Some talk about this being fault like, so I guess we must process it? */
12379	break;
12380
12381	default:
12382	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
12383	break;
12384	}
12385	}
12386
12387	/*
12388	* Check for debugger event breakpoints and dtrace probes.
12389	*/
12390	if ( uExitReason < RT_ELEMENTS(pDbgState->bmExitsToCheck) * 32U
12391	&& ASMBitTest(pDbgState->bmExitsToCheck, uExitReason) )
12392	{
12393	VBOXSTRICTRC rcStrict = hmR0VmxHandleExitDtraceEvents(pVCpu, pVmxTransient, uExitReason);
12394	if (rcStrict != VINF_SUCCESS)
12395	return rcStrict;
12396	}
12397
12398	/*
12399	* Normal processing.
12400	*/
12401	#ifdef HMVMX_USE_FUNCTION_TABLE
12402	return g_aVMExitHandlers[uExitReason].pfn(pVCpu, pVmxTransient);
12403	#else
12404	return hmR0VmxHandleExit(pVCpu, pVmxTransient, uExitReason);
12405	#endif
12406	}
12407
12408
12409	/**
12410	* Single steps guest code using hardware-assisted VMX.
12411	*
12412	* This is -not- the same as the guest single-stepping itself (say using EFLAGS.TF)
12413	* but single-stepping through the hypervisor debugger.
12414	*
12415	* @returns Strict VBox status code (i.e. informational status codes too).
12416	* @param pVCpu The cross context virtual CPU structure.
12417	* @param pcLoops Pointer to the number of executed loops.
12418	*
12419	* @note Mostly the same as hmR0VmxRunGuestCodeNormal().
12420	*/
12421	static VBOXSTRICTRC hmR0VmxRunGuestCodeDebug(PVMCPUCC pVCpu, uint32_t *pcLoops)
12422	{
12423	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hm.s.cMaxResumeLoops;
12424	Assert(pcLoops);
12425	Assert(*pcLoops <= cMaxResumeLoops);
12426
12427	VMXTRANSIENT VmxTransient;
12428	RT_ZERO(VmxTransient);
12429	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
12430
12431	/* Set HMCPU indicators. */
12432	bool const fSavedSingleInstruction = pVCpu->hm.s.fSingleInstruction;
12433	pVCpu->hm.s.fSingleInstruction = pVCpu->hm.s.fSingleInstruction \|\| DBGFIsStepping(pVCpu);
12434	pVCpu->hm.s.fDebugWantRdTscExit = false;
12435	pVCpu->hm.s.fUsingDebugLoop = true;
12436
12437	/* State we keep to help modify and later restore the VMCS fields we alter, and for detecting steps. */
12438	VMXRUNDBGSTATE DbgState;
12439	hmR0VmxRunDebugStateInit(pVCpu, &VmxTransient, &DbgState);
12440	hmR0VmxPreRunGuestDebugStateUpdate(pVCpu, &VmxTransient, &DbgState);
12441
12442	/*
12443	* The loop.
12444	*/
12445	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
12446	for (;;)
12447	{
12448	Assert(!HMR0SuspendPending());
12449	HMVMX_ASSERT_CPU_SAFE(pVCpu);
12450	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
12451	bool fStepping = pVCpu->hm.s.fSingleInstruction;
12452
12453	/* Set up VM-execution controls the next two can respond to. */
12454	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &VmxTransient, &DbgState);
12455
12456	/*
12457	* Preparatory work for running guest code, this may force us to
12458	* return to ring-3.
12459	*
12460	* Warning! This bugger disables interrupts on VINF_SUCCESS!
12461	*/
12462	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, fStepping);
12463	if (rcStrict != VINF_SUCCESS)
12464	break;
12465
12466	/* Interrupts are disabled at this point! */
12467	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
12468
12469	/* Override any obnoxious code in the above two calls. */
12470	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &VmxTransient, &DbgState);
12471
12472	/*
12473	* Finally execute the guest.
12474	*/
12475	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
12476
12477	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
12478	/* Interrupts are re-enabled at this point! */
12479
12480	/* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
12481	if (RT_SUCCESS(rcRun))
12482	{ /* very likely */ }
12483	else
12484	{
12485	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
12486	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
12487	return rcRun;
12488	}
12489
12490	/* Profile the VM-exit. */
12491	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
12492	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
12493	STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
12494	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
12495	HMVMX_START_EXIT_DISPATCH_PROF();
12496
12497	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
12498
12499	/*
12500	* Handle the VM-exit - we quit earlier on certain VM-exits, see hmR0VmxHandleExitDebug().
12501	*/
12502	rcStrict = hmR0VmxRunDebugHandleExit(pVCpu, &VmxTransient, &DbgState);
12503	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
12504	if (rcStrict != VINF_SUCCESS)
12505	break;
12506	if (++(*pcLoops) > cMaxResumeLoops)
12507	{
12508	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
12509	rcStrict = VINF_EM_RAW_INTERRUPT;
12510	break;
12511	}
12512
12513	/*
12514	* Stepping: Did the RIP change, if so, consider it a single step.
12515	* Otherwise, make sure one of the TFs gets set.
12516	*/
12517	if (fStepping)
12518	{
12519	int rc = hmR0VmxImportGuestState(pVCpu, VmxTransient.pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12520	AssertRC(rc);
12521	if ( pVCpu->cpum.GstCtx.rip != DbgState.uRipStart
12522	\|\| pVCpu->cpum.GstCtx.cs.Sel != DbgState.uCsStart)
12523	{
12524	rcStrict = VINF_EM_DBG_STEPPED;
12525	break;
12526	}
12527	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
12528	}
12529
12530	/*
12531	* Update when dtrace settings changes (DBGF kicks us, so no need to check).
12532	*/
12533	if (VBOXVMM_GET_SETTINGS_SEQ_NO() != DbgState.uDtraceSettingsSeqNo)
12534	hmR0VmxPreRunGuestDebugStateUpdate(pVCpu, &VmxTransient, &DbgState);
12535
12536	/* Restore all controls applied by hmR0VmxPreRunGuestDebugStateApply above. */
12537	rcStrict = hmR0VmxRunDebugStateRevert(pVCpu, &VmxTransient, &DbgState, rcStrict);
12538	Assert(rcStrict == VINF_SUCCESS);
12539	}
12540
12541	/*
12542	* Clear the X86_EFL_TF if necessary.
12543	*/
12544	if (pVCpu->hm.s.fClearTrapFlag)
12545	{
12546	int rc = hmR0VmxImportGuestState(pVCpu, VmxTransient.pVmcsInfo, CPUMCTX_EXTRN_RFLAGS);
12547	AssertRC(rc);
12548	pVCpu->hm.s.fClearTrapFlag = false;
12549	pVCpu->cpum.GstCtx.eflags.Bits.u1TF = 0;
12550	}
12551	/** @todo there seems to be issues with the resume flag when the monitor trap
12552	* flag is pending without being used. Seen early in bios init when
12553	* accessing APIC page in protected mode. */
12554
12555	/* Restore HMCPU indicators. */
12556	pVCpu->hm.s.fUsingDebugLoop = false;
12557	pVCpu->hm.s.fDebugWantRdTscExit = false;
12558	pVCpu->hm.s.fSingleInstruction = fSavedSingleInstruction;
12559
12560	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
12561	return rcStrict;
12562	}
12563
12564
12565	/** @} */
12566
12567
12568	/**
12569	* Checks if any expensive dtrace probes are enabled and we should go to the
12570	* debug loop.
12571	*
12572	* @returns true if we should use debug loop, false if not.
12573	*/
12574	static bool hmR0VmxAnyExpensiveProbesEnabled(void)
12575	{
12576	/* It's probably faster to OR the raw 32-bit counter variables together.
12577	Since the variables are in an array and the probes are next to one
12578	another (more or less), we have good locality. So, better read
12579	eight-nine cache lines ever time and only have one conditional, than
12580	128+ conditionals, right? */
12581	return ( VBOXVMM_R0_HMVMX_VMEXIT_ENABLED_RAW() /* expensive too due to context */
12582	\| VBOXVMM_XCPT_DE_ENABLED_RAW()
12583	\| VBOXVMM_XCPT_DB_ENABLED_RAW()
12584	\| VBOXVMM_XCPT_BP_ENABLED_RAW()
12585	\| VBOXVMM_XCPT_OF_ENABLED_RAW()
12586	\| VBOXVMM_XCPT_BR_ENABLED_RAW()
12587	\| VBOXVMM_XCPT_UD_ENABLED_RAW()
12588	\| VBOXVMM_XCPT_NM_ENABLED_RAW()
12589	\| VBOXVMM_XCPT_DF_ENABLED_RAW()
12590	\| VBOXVMM_XCPT_TS_ENABLED_RAW()
12591	\| VBOXVMM_XCPT_NP_ENABLED_RAW()
12592	\| VBOXVMM_XCPT_SS_ENABLED_RAW()
12593	\| VBOXVMM_XCPT_GP_ENABLED_RAW()
12594	\| VBOXVMM_XCPT_PF_ENABLED_RAW()
12595	\| VBOXVMM_XCPT_MF_ENABLED_RAW()
12596	\| VBOXVMM_XCPT_AC_ENABLED_RAW()
12597	\| VBOXVMM_XCPT_XF_ENABLED_RAW()
12598	\| VBOXVMM_XCPT_VE_ENABLED_RAW()
12599	\| VBOXVMM_XCPT_SX_ENABLED_RAW()
12600	\| VBOXVMM_INT_SOFTWARE_ENABLED_RAW()
12601	\| VBOXVMM_INT_HARDWARE_ENABLED_RAW()
12602	) != 0
12603	\|\| ( VBOXVMM_INSTR_HALT_ENABLED_RAW()
12604	\| VBOXVMM_INSTR_MWAIT_ENABLED_RAW()
12605	\| VBOXVMM_INSTR_MONITOR_ENABLED_RAW()
12606	\| VBOXVMM_INSTR_CPUID_ENABLED_RAW()
12607	\| VBOXVMM_INSTR_INVD_ENABLED_RAW()
12608	\| VBOXVMM_INSTR_WBINVD_ENABLED_RAW()
12609	\| VBOXVMM_INSTR_INVLPG_ENABLED_RAW()
12610	\| VBOXVMM_INSTR_RDTSC_ENABLED_RAW()
12611	\| VBOXVMM_INSTR_RDTSCP_ENABLED_RAW()
12612	\| VBOXVMM_INSTR_RDPMC_ENABLED_RAW()
12613	\| VBOXVMM_INSTR_RDMSR_ENABLED_RAW()
12614	\| VBOXVMM_INSTR_WRMSR_ENABLED_RAW()
12615	\| VBOXVMM_INSTR_CRX_READ_ENABLED_RAW()
12616	\| VBOXVMM_INSTR_CRX_WRITE_ENABLED_RAW()
12617	\| VBOXVMM_INSTR_DRX_READ_ENABLED_RAW()
12618	\| VBOXVMM_INSTR_DRX_WRITE_ENABLED_RAW()
12619	\| VBOXVMM_INSTR_PAUSE_ENABLED_RAW()
12620	\| VBOXVMM_INSTR_XSETBV_ENABLED_RAW()
12621	\| VBOXVMM_INSTR_SIDT_ENABLED_RAW()
12622	\| VBOXVMM_INSTR_LIDT_ENABLED_RAW()
12623	\| VBOXVMM_INSTR_SGDT_ENABLED_RAW()
12624	\| VBOXVMM_INSTR_LGDT_ENABLED_RAW()
12625	\| VBOXVMM_INSTR_SLDT_ENABLED_RAW()
12626	\| VBOXVMM_INSTR_LLDT_ENABLED_RAW()
12627	\| VBOXVMM_INSTR_STR_ENABLED_RAW()
12628	\| VBOXVMM_INSTR_LTR_ENABLED_RAW()
12629	\| VBOXVMM_INSTR_GETSEC_ENABLED_RAW()
12630	\| VBOXVMM_INSTR_RSM_ENABLED_RAW()
12631	\| VBOXVMM_INSTR_RDRAND_ENABLED_RAW()
12632	\| VBOXVMM_INSTR_RDSEED_ENABLED_RAW()
12633	\| VBOXVMM_INSTR_XSAVES_ENABLED_RAW()
12634	\| VBOXVMM_INSTR_XRSTORS_ENABLED_RAW()
12635	\| VBOXVMM_INSTR_VMM_CALL_ENABLED_RAW()
12636	\| VBOXVMM_INSTR_VMX_VMCLEAR_ENABLED_RAW()
12637	\| VBOXVMM_INSTR_VMX_VMLAUNCH_ENABLED_RAW()
12638	\| VBOXVMM_INSTR_VMX_VMPTRLD_ENABLED_RAW()
12639	\| VBOXVMM_INSTR_VMX_VMPTRST_ENABLED_RAW()
12640	\| VBOXVMM_INSTR_VMX_VMREAD_ENABLED_RAW()
12641	\| VBOXVMM_INSTR_VMX_VMRESUME_ENABLED_RAW()
12642	\| VBOXVMM_INSTR_VMX_VMWRITE_ENABLED_RAW()
12643	\| VBOXVMM_INSTR_VMX_VMXOFF_ENABLED_RAW()
12644	\| VBOXVMM_INSTR_VMX_VMXON_ENABLED_RAW()
12645	\| VBOXVMM_INSTR_VMX_VMFUNC_ENABLED_RAW()
12646	\| VBOXVMM_INSTR_VMX_INVEPT_ENABLED_RAW()
12647	\| VBOXVMM_INSTR_VMX_INVVPID_ENABLED_RAW()
12648	\| VBOXVMM_INSTR_VMX_INVPCID_ENABLED_RAW()
12649	) != 0
12650	\|\| ( VBOXVMM_EXIT_TASK_SWITCH_ENABLED_RAW()
12651	\| VBOXVMM_EXIT_HALT_ENABLED_RAW()
12652	\| VBOXVMM_EXIT_MWAIT_ENABLED_RAW()
12653	\| VBOXVMM_EXIT_MONITOR_ENABLED_RAW()
12654	\| VBOXVMM_EXIT_CPUID_ENABLED_RAW()
12655	\| VBOXVMM_EXIT_INVD_ENABLED_RAW()
12656	\| VBOXVMM_EXIT_WBINVD_ENABLED_RAW()
12657	\| VBOXVMM_EXIT_INVLPG_ENABLED_RAW()
12658	\| VBOXVMM_EXIT_RDTSC_ENABLED_RAW()
12659	\| VBOXVMM_EXIT_RDTSCP_ENABLED_RAW()
12660	\| VBOXVMM_EXIT_RDPMC_ENABLED_RAW()
12661	\| VBOXVMM_EXIT_RDMSR_ENABLED_RAW()
12662	\| VBOXVMM_EXIT_WRMSR_ENABLED_RAW()
12663	\| VBOXVMM_EXIT_CRX_READ_ENABLED_RAW()
12664	\| VBOXVMM_EXIT_CRX_WRITE_ENABLED_RAW()
12665	\| VBOXVMM_EXIT_DRX_READ_ENABLED_RAW()
12666	\| VBOXVMM_EXIT_DRX_WRITE_ENABLED_RAW()
12667	\| VBOXVMM_EXIT_PAUSE_ENABLED_RAW()
12668	\| VBOXVMM_EXIT_XSETBV_ENABLED_RAW()
12669	\| VBOXVMM_EXIT_SIDT_ENABLED_RAW()
12670	\| VBOXVMM_EXIT_LIDT_ENABLED_RAW()
12671	\| VBOXVMM_EXIT_SGDT_ENABLED_RAW()
12672	\| VBOXVMM_EXIT_LGDT_ENABLED_RAW()
12673	\| VBOXVMM_EXIT_SLDT_ENABLED_RAW()
12674	\| VBOXVMM_EXIT_LLDT_ENABLED_RAW()
12675	\| VBOXVMM_EXIT_STR_ENABLED_RAW()
12676	\| VBOXVMM_EXIT_LTR_ENABLED_RAW()
12677	\| VBOXVMM_EXIT_GETSEC_ENABLED_RAW()
12678	\| VBOXVMM_EXIT_RSM_ENABLED_RAW()
12679	\| VBOXVMM_EXIT_RDRAND_ENABLED_RAW()
12680	\| VBOXVMM_EXIT_RDSEED_ENABLED_RAW()
12681	\| VBOXVMM_EXIT_XSAVES_ENABLED_RAW()
12682	\| VBOXVMM_EXIT_XRSTORS_ENABLED_RAW()
12683	\| VBOXVMM_EXIT_VMM_CALL_ENABLED_RAW()
12684	\| VBOXVMM_EXIT_VMX_VMCLEAR_ENABLED_RAW()
12685	\| VBOXVMM_EXIT_VMX_VMLAUNCH_ENABLED_RAW()
12686	\| VBOXVMM_EXIT_VMX_VMPTRLD_ENABLED_RAW()
12687	\| VBOXVMM_EXIT_VMX_VMPTRST_ENABLED_RAW()
12688	\| VBOXVMM_EXIT_VMX_VMREAD_ENABLED_RAW()
12689	\| VBOXVMM_EXIT_VMX_VMRESUME_ENABLED_RAW()
12690	\| VBOXVMM_EXIT_VMX_VMWRITE_ENABLED_RAW()
12691	\| VBOXVMM_EXIT_VMX_VMXOFF_ENABLED_RAW()
12692	\| VBOXVMM_EXIT_VMX_VMXON_ENABLED_RAW()
12693	\| VBOXVMM_EXIT_VMX_VMFUNC_ENABLED_RAW()
12694	\| VBOXVMM_EXIT_VMX_INVEPT_ENABLED_RAW()
12695	\| VBOXVMM_EXIT_VMX_INVVPID_ENABLED_RAW()
12696	\| VBOXVMM_EXIT_VMX_INVPCID_ENABLED_RAW()
12697	\| VBOXVMM_EXIT_VMX_EPT_VIOLATION_ENABLED_RAW()
12698	\| VBOXVMM_EXIT_VMX_EPT_MISCONFIG_ENABLED_RAW()
12699	\| VBOXVMM_EXIT_VMX_VAPIC_ACCESS_ENABLED_RAW()
12700	\| VBOXVMM_EXIT_VMX_VAPIC_WRITE_ENABLED_RAW()
12701	) != 0;
12702	}
12703
12704
12705	/**
12706	* Runs the guest using hardware-assisted VMX.
12707	*
12708	* @returns Strict VBox status code (i.e. informational status codes too).
12709	* @param pVCpu The cross context virtual CPU structure.
12710	*/
12711	VMMR0DECL(VBOXSTRICTRC) VMXR0RunGuestCode(PVMCPUCC pVCpu)
12712	{
12713	AssertPtr(pVCpu);
12714	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
12715	Assert(VMMRZCallRing3IsEnabled(pVCpu));
12716	Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
12717	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
12718
12719	VBOXSTRICTRC rcStrict;
12720	uint32_t cLoops = 0;
12721	for (;;)
12722	{
12723	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12724	bool const fInNestedGuestMode = CPUMIsGuestInVmxNonRootMode(pCtx);
12725	#else
12726	NOREF(pCtx);
12727	bool const fInNestedGuestMode = false;
12728	#endif
12729	if (!fInNestedGuestMode)
12730	{
12731	if ( !pVCpu->hm.s.fUseDebugLoop
12732	&& (!VBOXVMM_ANY_PROBES_ENABLED() \|\| !hmR0VmxAnyExpensiveProbesEnabled())
12733	&& !DBGFIsStepping(pVCpu)
12734	&& !pVCpu->CTX_SUFF(pVM)->dbgf.ro.cEnabledInt3Breakpoints)
12735	rcStrict = hmR0VmxRunGuestCodeNormal(pVCpu, &cLoops);
12736	else
12737	rcStrict = hmR0VmxRunGuestCodeDebug(pVCpu, &cLoops);
12738	}
12739	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12740	else
12741	rcStrict = hmR0VmxRunGuestCodeNested(pVCpu, &cLoops);
12742
12743	if (rcStrict == VINF_VMX_VMLAUNCH_VMRESUME)
12744	{
12745	Assert(CPUMIsGuestInVmxNonRootMode(pCtx));
12746	continue;
12747	}
12748	if (rcStrict == VINF_VMX_VMEXIT)
12749	{
12750	Assert(!CPUMIsGuestInVmxNonRootMode(pCtx));
12751	continue;
12752	}
12753	#endif
12754	break;
12755	}
12756
12757	int const rcLoop = VBOXSTRICTRC_VAL(rcStrict);
12758	switch (rcLoop)
12759	{
12760	case VERR_EM_INTERPRETER: rcStrict = VINF_EM_RAW_EMULATE_INSTR; break;
12761	case VINF_EM_RESET: rcStrict = VINF_EM_TRIPLE_FAULT; break;
12762	}
12763
12764	int rc2 = hmR0VmxExitToRing3(pVCpu, rcStrict);
12765	if (RT_FAILURE(rc2))
12766	{
12767	pVCpu->hm.s.u32HMError = (uint32_t)VBOXSTRICTRC_VAL(rcStrict);
12768	rcStrict = rc2;
12769	}
12770	Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
12771	Assert(!VMMRZCallRing3IsNotificationSet(pVCpu));
12772	return rcStrict;
12773	}
12774
12775
12776	#ifndef HMVMX_USE_FUNCTION_TABLE
12777	/**
12778	* Handles a guest VM-exit from hardware-assisted VMX execution.
12779	*
12780	* @returns Strict VBox status code (i.e. informational status codes too).
12781	* @param pVCpu The cross context virtual CPU structure.
12782	* @param pVmxTransient The VMX-transient structure.
12783	*/
12784	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
12785	{
12786	#ifdef DEBUG_ramshankar
12787	# define VMEXIT_CALL_RET(a_fSave, a_CallExpr) \
12788	do { \
12789	if (a_fSave != 0) \
12790	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL); \
12791	VBOXSTRICTRC rcStrict = a_CallExpr; \
12792	if (a_fSave != 0) \
12793	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST); \
12794	return rcStrict; \
12795	} while (0)
12796	#else
12797	# define VMEXIT_CALL_RET(a_fSave, a_CallExpr) return a_CallExpr
12798	#endif
12799	uint32_t const uExitReason = pVmxTransient->uExitReason;
12800	switch (uExitReason)
12801	{
12802	case VMX_EXIT_EPT_MISCONFIG: VMEXIT_CALL_RET(0, hmR0VmxExitEptMisconfig(pVCpu, pVmxTransient));
12803	case VMX_EXIT_EPT_VIOLATION: VMEXIT_CALL_RET(0, hmR0VmxExitEptViolation(pVCpu, pVmxTransient));
12804	case VMX_EXIT_IO_INSTR: VMEXIT_CALL_RET(0, hmR0VmxExitIoInstr(pVCpu, pVmxTransient));
12805	case VMX_EXIT_CPUID: VMEXIT_CALL_RET(0, hmR0VmxExitCpuid(pVCpu, pVmxTransient));
12806	case VMX_EXIT_RDTSC: VMEXIT_CALL_RET(0, hmR0VmxExitRdtsc(pVCpu, pVmxTransient));
12807	case VMX_EXIT_RDTSCP: VMEXIT_CALL_RET(0, hmR0VmxExitRdtscp(pVCpu, pVmxTransient));
12808	case VMX_EXIT_APIC_ACCESS: VMEXIT_CALL_RET(0, hmR0VmxExitApicAccess(pVCpu, pVmxTransient));
12809	case VMX_EXIT_XCPT_OR_NMI: VMEXIT_CALL_RET(0, hmR0VmxExitXcptOrNmi(pVCpu, pVmxTransient));
12810	case VMX_EXIT_MOV_CRX: VMEXIT_CALL_RET(0, hmR0VmxExitMovCRx(pVCpu, pVmxTransient));
12811	case VMX_EXIT_EXT_INT: VMEXIT_CALL_RET(0, hmR0VmxExitExtInt(pVCpu, pVmxTransient));
12812	case VMX_EXIT_INT_WINDOW: VMEXIT_CALL_RET(0, hmR0VmxExitIntWindow(pVCpu, pVmxTransient));
12813	case VMX_EXIT_TPR_BELOW_THRESHOLD: VMEXIT_CALL_RET(0, hmR0VmxExitTprBelowThreshold(pVCpu, pVmxTransient));
12814	case VMX_EXIT_MWAIT: VMEXIT_CALL_RET(0, hmR0VmxExitMwait(pVCpu, pVmxTransient));
12815	case VMX_EXIT_MONITOR: VMEXIT_CALL_RET(0, hmR0VmxExitMonitor(pVCpu, pVmxTransient));
12816	case VMX_EXIT_TASK_SWITCH: VMEXIT_CALL_RET(0, hmR0VmxExitTaskSwitch(pVCpu, pVmxTransient));
12817	case VMX_EXIT_PREEMPT_TIMER: VMEXIT_CALL_RET(0, hmR0VmxExitPreemptTimer(pVCpu, pVmxTransient));
12818	case VMX_EXIT_RDMSR: VMEXIT_CALL_RET(0, hmR0VmxExitRdmsr(pVCpu, pVmxTransient));
12819	case VMX_EXIT_WRMSR: VMEXIT_CALL_RET(0, hmR0VmxExitWrmsr(pVCpu, pVmxTransient));
12820	case VMX_EXIT_VMCALL: VMEXIT_CALL_RET(0, hmR0VmxExitVmcall(pVCpu, pVmxTransient));
12821	case VMX_EXIT_MOV_DRX: VMEXIT_CALL_RET(0, hmR0VmxExitMovDRx(pVCpu, pVmxTransient));
12822	case VMX_EXIT_HLT: VMEXIT_CALL_RET(0, hmR0VmxExitHlt(pVCpu, pVmxTransient));
12823	case VMX_EXIT_INVD: VMEXIT_CALL_RET(0, hmR0VmxExitInvd(pVCpu, pVmxTransient));
12824	case VMX_EXIT_INVLPG: VMEXIT_CALL_RET(0, hmR0VmxExitInvlpg(pVCpu, pVmxTransient));
12825	case VMX_EXIT_MTF: VMEXIT_CALL_RET(0, hmR0VmxExitMtf(pVCpu, pVmxTransient));
12826	case VMX_EXIT_PAUSE: VMEXIT_CALL_RET(0, hmR0VmxExitPause(pVCpu, pVmxTransient));
12827	case VMX_EXIT_WBINVD: VMEXIT_CALL_RET(0, hmR0VmxExitWbinvd(pVCpu, pVmxTransient));
12828	case VMX_EXIT_XSETBV: VMEXIT_CALL_RET(0, hmR0VmxExitXsetbv(pVCpu, pVmxTransient));
12829	case VMX_EXIT_INVPCID: VMEXIT_CALL_RET(0, hmR0VmxExitInvpcid(pVCpu, pVmxTransient));
12830	case VMX_EXIT_GETSEC: VMEXIT_CALL_RET(0, hmR0VmxExitGetsec(pVCpu, pVmxTransient));
12831	case VMX_EXIT_RDPMC: VMEXIT_CALL_RET(0, hmR0VmxExitRdpmc(pVCpu, pVmxTransient));
12832	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12833	case VMX_EXIT_VMCLEAR: VMEXIT_CALL_RET(0, hmR0VmxExitVmclear(pVCpu, pVmxTransient));
12834	case VMX_EXIT_VMLAUNCH: VMEXIT_CALL_RET(0, hmR0VmxExitVmlaunch(pVCpu, pVmxTransient));
12835	case VMX_EXIT_VMPTRLD: VMEXIT_CALL_RET(0, hmR0VmxExitVmptrld(pVCpu, pVmxTransient));
12836	case VMX_EXIT_VMPTRST: VMEXIT_CALL_RET(0, hmR0VmxExitVmptrst(pVCpu, pVmxTransient));
12837	case VMX_EXIT_VMREAD: VMEXIT_CALL_RET(0, hmR0VmxExitVmread(pVCpu, pVmxTransient));
12838	case VMX_EXIT_VMRESUME: VMEXIT_CALL_RET(0, hmR0VmxExitVmwrite(pVCpu, pVmxTransient));
12839	case VMX_EXIT_VMWRITE: VMEXIT_CALL_RET(0, hmR0VmxExitVmresume(pVCpu, pVmxTransient));
12840	case VMX_EXIT_VMXOFF: VMEXIT_CALL_RET(0, hmR0VmxExitVmxoff(pVCpu, pVmxTransient));
12841	case VMX_EXIT_VMXON: VMEXIT_CALL_RET(0, hmR0VmxExitVmxon(pVCpu, pVmxTransient));
12842	case VMX_EXIT_INVVPID: VMEXIT_CALL_RET(0, hmR0VmxExitInvvpid(pVCpu, pVmxTransient));
12843	case VMX_EXIT_INVEPT: VMEXIT_CALL_RET(0, hmR0VmxExitSetPendingXcptUD(pVCpu, pVmxTransient));
12844	#else
12845	case VMX_EXIT_VMCLEAR:
12846	case VMX_EXIT_VMLAUNCH:
12847	case VMX_EXIT_VMPTRLD:
12848	case VMX_EXIT_VMPTRST:
12849	case VMX_EXIT_VMREAD:
12850	case VMX_EXIT_VMRESUME:
12851	case VMX_EXIT_VMWRITE:
12852	case VMX_EXIT_VMXOFF:
12853	case VMX_EXIT_VMXON:
12854	case VMX_EXIT_INVVPID:
12855	case VMX_EXIT_INVEPT:
12856	return hmR0VmxExitSetPendingXcptUD(pVCpu, pVmxTransient);
12857	#endif
12858
12859	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFault(pVCpu, pVmxTransient);
12860	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindow(pVCpu, pVmxTransient);
12861	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestState(pVCpu, pVmxTransient);
12862
12863	case VMX_EXIT_INIT_SIGNAL:
12864	case VMX_EXIT_SIPI:
12865	case VMX_EXIT_IO_SMI:
12866	case VMX_EXIT_SMI:
12867	case VMX_EXIT_ERR_MSR_LOAD:
12868	case VMX_EXIT_ERR_MACHINE_CHECK:
12869	case VMX_EXIT_PML_FULL:
12870	case VMX_EXIT_VIRTUALIZED_EOI:
12871	case VMX_EXIT_GDTR_IDTR_ACCESS:
12872	case VMX_EXIT_LDTR_TR_ACCESS:
12873	case VMX_EXIT_APIC_WRITE:
12874	case VMX_EXIT_RDRAND:
12875	case VMX_EXIT_RSM:
12876	case VMX_EXIT_VMFUNC:
12877	case VMX_EXIT_ENCLS:
12878	case VMX_EXIT_RDSEED:
12879	case VMX_EXIT_XSAVES:
12880	case VMX_EXIT_XRSTORS:
12881	case VMX_EXIT_UMWAIT:
12882	case VMX_EXIT_TPAUSE:
12883	default:
12884	return hmR0VmxExitErrUnexpected(pVCpu, pVmxTransient);
12885	}
12886	#undef VMEXIT_CALL_RET
12887	}
12888	#endif /* !HMVMX_USE_FUNCTION_TABLE */
12889
12890
12891	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12892	/**
12893	* Handles a nested-guest VM-exit from hardware-assisted VMX execution.
12894	*
12895	* @returns Strict VBox status code (i.e. informational status codes too).
12896	* @param pVCpu The cross context virtual CPU structure.
12897	* @param pVmxTransient The VMX-transient structure.
12898	*/
12899	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
12900	{
12901	uint32_t const uExitReason = pVmxTransient->uExitReason;
12902	switch (uExitReason)
12903	{
12904	case VMX_EXIT_EPT_MISCONFIG: return hmR0VmxExitEptMisconfig(pVCpu, pVmxTransient);
12905	case VMX_EXIT_EPT_VIOLATION: return hmR0VmxExitEptViolation(pVCpu, pVmxTransient);
12906	case VMX_EXIT_XCPT_OR_NMI: return hmR0VmxExitXcptOrNmiNested(pVCpu, pVmxTransient);
12907	case VMX_EXIT_IO_INSTR: return hmR0VmxExitIoInstrNested(pVCpu, pVmxTransient);
12908	case VMX_EXIT_HLT: return hmR0VmxExitHltNested(pVCpu, pVmxTransient);
12909
12910	/*
12911	* We shouldn't direct host physical interrupts to the nested-guest.
12912	*/
12913	case VMX_EXIT_EXT_INT:
12914	return hmR0VmxExitExtInt(pVCpu, pVmxTransient);
12915
12916	/*
12917	* Instructions that cause VM-exits unconditionally or the condition is
12918	* always is taken solely from the nested hypervisor (meaning if the VM-exit
12919	* happens, it's guaranteed to be a nested-guest VM-exit).
12920	*
12921	* - Provides VM-exit instruction length ONLY.
12922	*/
12923	case VMX_EXIT_CPUID: /* Unconditional. */
12924	case VMX_EXIT_VMCALL:
12925	case VMX_EXIT_GETSEC:
12926	case VMX_EXIT_INVD:
12927	case VMX_EXIT_XSETBV:
12928	case VMX_EXIT_VMLAUNCH:
12929	case VMX_EXIT_VMRESUME:
12930	case VMX_EXIT_VMXOFF:
12931	case VMX_EXIT_ENCLS: /* Condition specified solely by nested hypervisor. */
12932	case VMX_EXIT_VMFUNC:
12933	return hmR0VmxExitInstrNested(pVCpu, pVmxTransient);
12934
12935	/*
12936	* Instructions that cause VM-exits unconditionally or the condition is
12937	* always is taken solely from the nested hypervisor (meaning if the VM-exit
12938	* happens, it's guaranteed to be a nested-guest VM-exit).
12939	*
12940	* - Provides VM-exit instruction length.
12941	* - Provides VM-exit information.
12942	* - Optionally provides Exit qualification.
12943	*
12944	* Since Exit qualification is 0 for all VM-exits where it is not
12945	* applicable, reading and passing it to the guest should produce
12946	* defined behavior.
12947	*
12948	* See Intel spec. 27.2.1 "Basic VM-Exit Information".
12949	*/
12950	case VMX_EXIT_INVEPT: /* Unconditional. */
12951	case VMX_EXIT_INVVPID:
12952	case VMX_EXIT_VMCLEAR:
12953	case VMX_EXIT_VMPTRLD:
12954	case VMX_EXIT_VMPTRST:
12955	case VMX_EXIT_VMXON:
12956	case VMX_EXIT_GDTR_IDTR_ACCESS: /* Condition specified solely by nested hypervisor. */
12957	case VMX_EXIT_LDTR_TR_ACCESS:
12958	case VMX_EXIT_RDRAND:
12959	case VMX_EXIT_RDSEED:
12960	case VMX_EXIT_XSAVES:
12961	case VMX_EXIT_XRSTORS:
12962	case VMX_EXIT_UMWAIT:
12963	case VMX_EXIT_TPAUSE:
12964	return hmR0VmxExitInstrWithInfoNested(pVCpu, pVmxTransient);
12965
12966	case VMX_EXIT_RDTSC: return hmR0VmxExitRdtscNested(pVCpu, pVmxTransient);
12967	case VMX_EXIT_RDTSCP: return hmR0VmxExitRdtscpNested(pVCpu, pVmxTransient);
12968	case VMX_EXIT_RDMSR: return hmR0VmxExitRdmsrNested(pVCpu, pVmxTransient);
12969	case VMX_EXIT_WRMSR: return hmR0VmxExitWrmsrNested(pVCpu, pVmxTransient);
12970	case VMX_EXIT_INVLPG: return hmR0VmxExitInvlpgNested(pVCpu, pVmxTransient);
12971	case VMX_EXIT_INVPCID: return hmR0VmxExitInvpcidNested(pVCpu, pVmxTransient);
12972	case VMX_EXIT_TASK_SWITCH: return hmR0VmxExitTaskSwitchNested(pVCpu, pVmxTransient);
12973	case VMX_EXIT_WBINVD: return hmR0VmxExitWbinvdNested(pVCpu, pVmxTransient);
12974	case VMX_EXIT_MTF: return hmR0VmxExitMtfNested(pVCpu, pVmxTransient);
12975	case VMX_EXIT_APIC_ACCESS: return hmR0VmxExitApicAccessNested(pVCpu, pVmxTransient);
12976	case VMX_EXIT_APIC_WRITE: return hmR0VmxExitApicWriteNested(pVCpu, pVmxTransient);
12977	case VMX_EXIT_VIRTUALIZED_EOI: return hmR0VmxExitVirtEoiNested(pVCpu, pVmxTransient);
12978	case VMX_EXIT_MOV_CRX: return hmR0VmxExitMovCRxNested(pVCpu, pVmxTransient);
12979	case VMX_EXIT_INT_WINDOW: return hmR0VmxExitIntWindowNested(pVCpu, pVmxTransient);
12980	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindowNested(pVCpu, pVmxTransient);
12981	case VMX_EXIT_TPR_BELOW_THRESHOLD: return hmR0VmxExitTprBelowThresholdNested(pVCpu, pVmxTransient);
12982	case VMX_EXIT_MWAIT: return hmR0VmxExitMwaitNested(pVCpu, pVmxTransient);
12983	case VMX_EXIT_MONITOR: return hmR0VmxExitMonitorNested(pVCpu, pVmxTransient);
12984	case VMX_EXIT_PAUSE: return hmR0VmxExitPauseNested(pVCpu, pVmxTransient);
12985
12986	case VMX_EXIT_PREEMPT_TIMER:
12987	{
12988	/** @todo NSTVMX: Preempt timer. */
12989	return hmR0VmxExitPreemptTimer(pVCpu, pVmxTransient);
12990	}
12991
12992	case VMX_EXIT_MOV_DRX: return hmR0VmxExitMovDRxNested(pVCpu, pVmxTransient);
12993	case VMX_EXIT_RDPMC: return hmR0VmxExitRdpmcNested(pVCpu, pVmxTransient);
12994
12995	case VMX_EXIT_VMREAD:
12996	case VMX_EXIT_VMWRITE: return hmR0VmxExitVmreadVmwriteNested(pVCpu, pVmxTransient);
12997
12998	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFaultNested(pVCpu, pVmxTransient);
12999	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestStateNested(pVCpu, pVmxTransient);
13000
13001	case VMX_EXIT_INIT_SIGNAL:
13002	case VMX_EXIT_SIPI:
13003	case VMX_EXIT_IO_SMI:
13004	case VMX_EXIT_SMI:
13005	case VMX_EXIT_ERR_MSR_LOAD:
13006	case VMX_EXIT_ERR_MACHINE_CHECK:
13007	case VMX_EXIT_PML_FULL:
13008	case VMX_EXIT_RSM:
13009	default:
13010	return hmR0VmxExitErrUnexpected(pVCpu, pVmxTransient);
13011	}
13012	}
13013	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13014
13015
13016	/** @name VM-exit helpers.
13017	* @{
13018	*/
13019	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13020	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= VM-exit helpers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
13021	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13022
13023	/** Macro for VM-exits called unexpectedly. */
13024	#define HMVMX_UNEXPECTED_EXIT_RET(a_pVCpu, a_HmError) \
13025	do { \
13026	(a_pVCpu)->hm.s.u32HMError = (a_HmError); \
13027	return VERR_VMX_UNEXPECTED_EXIT; \
13028	} while (0)
13029
13030	#ifdef VBOX_STRICT
13031	/* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */
13032	# define HMVMX_ASSERT_PREEMPT_CPUID_VAR() \
13033	RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId()
13034
13035	# define HMVMX_ASSERT_PREEMPT_CPUID() \
13036	do { \
13037	RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \
13038	AssertMsg(idAssertCpu == idAssertCpuNow, ("VMX %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \
13039	} while (0)
13040
13041	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13042	do { \
13043	AssertPtr((a_pVCpu)); \
13044	AssertPtr((a_pVmxTransient)); \
13045	Assert((a_pVmxTransient)->fVMEntryFailed == false); \
13046	Assert((a_pVmxTransient)->pVmcsInfo); \
13047	Assert(ASMIntAreEnabled()); \
13048	HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu); \
13049	HMVMX_ASSERT_PREEMPT_CPUID_VAR(); \
13050	Log4Func(("vcpu[%RU32]\n", (a_pVCpu)->idCpu)); \
13051	HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu); \
13052	if (VMMR0IsLogFlushDisabled((a_pVCpu))) \
13053	HMVMX_ASSERT_PREEMPT_CPUID(); \
13054	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
13055	} while (0)
13056
13057	# define HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13058	do { \
13059	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient); \
13060	Assert((a_pVmxTransient)->fIsNestedGuest); \
13061	} while (0)
13062
13063	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13064	do { \
13065	Log4Func(("\n")); \
13066	} while (0)
13067	#else
13068	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13069	do { \
13070	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
13071	NOREF((a_pVCpu)); NOREF((a_pVmxTransient)); \
13072	} while (0)
13073
13074	# define HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13075	do { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient); } while (0)
13076
13077	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) do { } while (0)
13078	#endif
13079
13080	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13081	/** Macro that does the necessary privilege checks and intercepted VM-exits for
13082	* guests that attempted to execute a VMX instruction. */
13083	# define HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(a_pVCpu, a_uExitReason) \
13084	do \
13085	{ \
13086	VBOXSTRICTRC rcStrictTmp = hmR0VmxCheckExitDueToVmxInstr((a_pVCpu), (a_uExitReason)); \
13087	if (rcStrictTmp == VINF_SUCCESS) \
13088	{ /* likely */ } \
13089	else if (rcStrictTmp == VINF_HM_PENDING_XCPT) \
13090	{ \
13091	Assert((a_pVCpu)->hm.s.Event.fPending); \
13092	Log4Func(("Privilege checks failed -> %#x\n", VMX_ENTRY_INT_INFO_VECTOR((a_pVCpu)->hm.s.Event.u64IntInfo))); \
13093	return VINF_SUCCESS; \
13094	} \
13095	else \
13096	{ \
13097	int rcTmp = VBOXSTRICTRC_VAL(rcStrictTmp); \
13098	AssertMsgFailedReturn(("Unexpected failure. rc=%Rrc", rcTmp), rcTmp); \
13099	} \
13100	} while (0)
13101
13102	/** Macro that decodes a memory operand for an VM-exit caused by an instruction. */
13103	# define HMVMX_DECODE_MEM_OPERAND(a_pVCpu, a_uExitInstrInfo, a_uExitQual, a_enmMemAccess, a_pGCPtrEffAddr) \
13104	do \
13105	{ \
13106	VBOXSTRICTRC rcStrictTmp = hmR0VmxDecodeMemOperand((a_pVCpu), (a_uExitInstrInfo), (a_uExitQual), (a_enmMemAccess), \
13107	(a_pGCPtrEffAddr)); \
13108	if (rcStrictTmp == VINF_SUCCESS) \
13109	{ /* likely */ } \
13110	else if (rcStrictTmp == VINF_HM_PENDING_XCPT) \
13111	{ \
13112	uint8_t const uXcptTmp = VMX_ENTRY_INT_INFO_VECTOR((a_pVCpu)->hm.s.Event.u64IntInfo); \
13113	Log4Func(("Memory operand decoding failed, raising xcpt %#x\n", uXcptTmp)); \
13114	NOREF(uXcptTmp); \
13115	return VINF_SUCCESS; \
13116	} \
13117	else \
13118	{ \
13119	Log4Func(("hmR0VmxDecodeMemOperand failed. rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrictTmp))); \
13120	return rcStrictTmp; \
13121	} \
13122	} while (0)
13123	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13124
13125
13126	/**
13127	* Advances the guest RIP by the specified number of bytes.
13128	*
13129	* @param pVCpu The cross context virtual CPU structure.
13130	* @param cbInstr Number of bytes to advance the RIP by.
13131	*
13132	* @remarks No-long-jump zone!!!
13133	*/
13134	DECLINLINE(void) hmR0VmxAdvanceGuestRipBy(PVMCPUCC pVCpu, uint32_t cbInstr)
13135	{
13136	/* Advance the RIP. */
13137	pVCpu->cpum.GstCtx.rip += cbInstr;
13138	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
13139
13140	/* Update interrupt inhibition. */
13141	if ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)
13142	&& pVCpu->cpum.GstCtx.rip != EMGetInhibitInterruptsPC(pVCpu))
13143	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
13144	}
13145
13146
13147	/**
13148	* Advances the guest RIP after reading it from the VMCS.
13149	*
13150	* @returns VBox status code, no informational status codes.
13151	* @param pVCpu The cross context virtual CPU structure.
13152	* @param pVmxTransient The VMX-transient structure.
13153	*
13154	* @remarks No-long-jump zone!!!
13155	*/
13156	static int hmR0VmxAdvanceGuestRip(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13157	{
13158	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13159	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS);
13160	AssertRCReturn(rc, rc);
13161
13162	hmR0VmxAdvanceGuestRipBy(pVCpu, pVmxTransient->cbExitInstr);
13163	return VINF_SUCCESS;
13164	}
13165
13166
13167	/**
13168	* Handle a condition that occurred while delivering an event through the guest or
13169	* nested-guest IDT.
13170	*
13171	* @returns Strict VBox status code (i.e. informational status codes too).
13172	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
13173	* @retval VINF_HM_DOUBLE_FAULT if a \#DF condition was detected and we ought
13174	* to continue execution of the guest which will delivery the \#DF.
13175	* @retval VINF_EM_RESET if we detected a triple-fault condition.
13176	* @retval VERR_EM_GUEST_CPU_HANG if we detected a guest CPU hang.
13177	*
13178	* @param pVCpu The cross context virtual CPU structure.
13179	* @param pVmxTransient The VMX-transient structure.
13180	*
13181	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13182	* Additionally, HMVMX_READ_EXIT_QUALIFICATION is required if the VM-exit
13183	* is due to an EPT violation, PML full or SPP-related event.
13184	*
13185	* @remarks No-long-jump zone!!!
13186	*/
13187	static VBOXSTRICTRC hmR0VmxCheckExitDueToEventDelivery(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13188	{
13189	Assert(!pVCpu->hm.s.Event.fPending);
13190	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_XCPT_INFO);
13191	if ( pVmxTransient->uExitReason == VMX_EXIT_EPT_VIOLATION
13192	\|\| pVmxTransient->uExitReason == VMX_EXIT_PML_FULL
13193	\|\| pVmxTransient->uExitReason == VMX_EXIT_SPP_EVENT)
13194	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_EXIT_QUALIFICATION);
13195
13196	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
13197	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
13198	uint32_t const uIdtVectorInfo = pVmxTransient->uIdtVectoringInfo;
13199	uint32_t const uExitIntInfo = pVmxTransient->uExitIntInfo;
13200	if (VMX_IDT_VECTORING_INFO_IS_VALID(uIdtVectorInfo))
13201	{
13202	uint32_t const uIdtVector = VMX_IDT_VECTORING_INFO_VECTOR(uIdtVectorInfo);
13203	uint32_t const uIdtVectorType = VMX_IDT_VECTORING_INFO_TYPE(uIdtVectorInfo);
13204
13205	/*
13206	* If the event was a software interrupt (generated with INT n) or a software exception
13207	* (generated by INT3/INTO) or a privileged software exception (generated by INT1), we
13208	* can handle the VM-exit and continue guest execution which will re-execute the
13209	* instruction rather than re-injecting the exception, as that can cause premature
13210	* trips to ring-3 before injection and involve TRPM which currently has no way of
13211	* storing that these exceptions were caused by these instructions (ICEBP's #DB poses
13212	* the problem).
13213	*/
13214	IEMXCPTRAISE enmRaise;
13215	IEMXCPTRAISEINFO fRaiseInfo;
13216	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
13217	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
13218	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT)
13219	{
13220	enmRaise = IEMXCPTRAISE_REEXEC_INSTR;
13221	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
13222	}
13223	else if (VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo))
13224	{
13225	uint32_t const uExitVectorType = VMX_EXIT_INT_INFO_TYPE(uExitIntInfo);
13226	uint8_t const uExitVector = VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo);
13227	Assert(uExitVectorType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT);
13228
13229	uint32_t const fIdtVectorFlags = hmR0VmxGetIemXcptFlags(uIdtVector, uIdtVectorType);
13230	uint32_t const fExitVectorFlags = hmR0VmxGetIemXcptFlags(uExitVector, uExitVectorType);
13231
13232	enmRaise = IEMEvaluateRecursiveXcpt(pVCpu, fIdtVectorFlags, uIdtVector, fExitVectorFlags, uExitVector, &fRaiseInfo);
13233
13234	/* Determine a vectoring #PF condition, see comment in hmR0VmxExitXcptPF(). */
13235	if (fRaiseInfo & (IEMXCPTRAISEINFO_EXT_INT_PF \| IEMXCPTRAISEINFO_NMI_PF))
13236	{
13237	pVmxTransient->fVectoringPF = true;
13238	enmRaise = IEMXCPTRAISE_PREV_EVENT;
13239	}
13240	}
13241	else
13242	{
13243	/*
13244	* If an exception or hardware interrupt delivery caused an EPT violation/misconfig or APIC access
13245	* VM-exit, then the VM-exit interruption-information will not be valid and we end up here.
13246	* It is sufficient to reflect the original event to the guest after handling the VM-exit.
13247	*/
13248	Assert( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
13249	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
13250	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT);
13251	enmRaise = IEMXCPTRAISE_PREV_EVENT;
13252	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
13253	}
13254
13255	/*
13256	* On CPUs that support Virtual NMIs, if this VM-exit (be it an exception or EPT violation/misconfig
13257	* etc.) occurred while delivering the NMI, we need to clear the block-by-NMI field in the guest
13258	* interruptibility-state before re-delivering the NMI after handling the VM-exit. Otherwise the
13259	* subsequent VM-entry would fail, see @bugref{7445}.
13260	*
13261	* See Intel spec. 30.7.1.2 "Resuming Guest Software after Handling an Exception".
13262	*/
13263	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
13264	&& enmRaise == IEMXCPTRAISE_PREV_EVENT
13265	&& (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
13266	&& CPUMIsGuestNmiBlocking(pVCpu))
13267	{
13268	CPUMSetGuestNmiBlocking(pVCpu, false);
13269	}
13270
13271	switch (enmRaise)
13272	{
13273	case IEMXCPTRAISE_CURRENT_XCPT:
13274	{
13275	Log4Func(("IDT: Pending secondary Xcpt: idtinfo=%#RX64 exitinfo=%#RX64\n", uIdtVectorInfo, uExitIntInfo));
13276	Assert(rcStrict == VINF_SUCCESS);
13277	break;
13278	}
13279
13280	case IEMXCPTRAISE_PREV_EVENT:
13281	{
13282	uint32_t u32ErrCode;
13283	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(uIdtVectorInfo))
13284	u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
13285	else
13286	u32ErrCode = 0;
13287
13288	/* If uExitVector is #PF, CR2 value will be updated from the VMCS if it's a guest #PF, see hmR0VmxExitXcptPF(). */
13289	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflect);
13290	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(uIdtVectorInfo), 0 /* cbInstr */,
13291	u32ErrCode, pVCpu->cpum.GstCtx.cr2);
13292
13293	Log4Func(("IDT: Pending vectoring event %#RX64 Err=%#RX32\n", pVCpu->hm.s.Event.u64IntInfo,
13294	pVCpu->hm.s.Event.u32ErrCode));
13295	Assert(rcStrict == VINF_SUCCESS);
13296	break;
13297	}
13298
13299	case IEMXCPTRAISE_REEXEC_INSTR:
13300	Assert(rcStrict == VINF_SUCCESS);
13301	break;
13302
13303	case IEMXCPTRAISE_DOUBLE_FAULT:
13304	{
13305	/*
13306	* Determing a vectoring double #PF condition. Used later, when PGM evaluates the
13307	* second #PF as a guest #PF (and not a shadow #PF) and needs to be converted into a #DF.
13308	*/
13309	if (fRaiseInfo & IEMXCPTRAISEINFO_PF_PF)
13310	{
13311	pVmxTransient->fVectoringDoublePF = true;
13312	Log4Func(("IDT: Vectoring double #PF %#RX64 cr2=%#RX64\n", pVCpu->hm.s.Event.u64IntInfo,
13313	pVCpu->cpum.GstCtx.cr2));
13314	rcStrict = VINF_SUCCESS;
13315	}
13316	else
13317	{
13318	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectConvertDF);
13319	hmR0VmxSetPendingXcptDF(pVCpu);
13320	Log4Func(("IDT: Pending vectoring #DF %#RX64 uIdtVector=%#x uExitVector=%#x\n", pVCpu->hm.s.Event.u64IntInfo,
13321	uIdtVector, VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo)));
13322	rcStrict = VINF_HM_DOUBLE_FAULT;
13323	}
13324	break;
13325	}
13326
13327	case IEMXCPTRAISE_TRIPLE_FAULT:
13328	{
13329	Log4Func(("IDT: Pending vectoring triple-fault uIdt=%#x uExit=%#x\n", uIdtVector,
13330	VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo)));
13331	rcStrict = VINF_EM_RESET;
13332	break;
13333	}
13334
13335	case IEMXCPTRAISE_CPU_HANG:
13336	{
13337	Log4Func(("IDT: Bad guest! Entering CPU hang. fRaiseInfo=%#x\n", fRaiseInfo));
13338	rcStrict = VERR_EM_GUEST_CPU_HANG;
13339	break;
13340	}
13341
13342	default:
13343	{
13344	AssertMsgFailed(("IDT: vcpu[%RU32] Unexpected/invalid value! enmRaise=%#x\n", pVCpu->idCpu, enmRaise));
13345	rcStrict = VERR_VMX_IPE_2;
13346	break;
13347	}
13348	}
13349	}
13350	else if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
13351	&& !CPUMIsGuestNmiBlocking(pVCpu))
13352	{
13353	if ( VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo)
13354	&& VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo) != X86_XCPT_DF
13355	&& VMX_EXIT_INT_INFO_IS_NMI_UNBLOCK_IRET(uExitIntInfo))
13356	{
13357	/*
13358	* Execution of IRET caused a fault when NMI blocking was in effect (i.e we're in
13359	* the guest or nested-guest NMI handler). We need to set the block-by-NMI field so
13360	* that virtual NMIs remain blocked until the IRET execution is completed.
13361	*
13362	* See Intel spec. 31.7.1.2 "Resuming Guest Software After Handling An Exception".
13363	*/
13364	CPUMSetGuestNmiBlocking(pVCpu, true);
13365	Log4Func(("Set NMI blocking. uExitReason=%u\n", pVmxTransient->uExitReason));
13366	}
13367	else if ( pVmxTransient->uExitReason == VMX_EXIT_EPT_VIOLATION
13368	\|\| pVmxTransient->uExitReason == VMX_EXIT_PML_FULL
13369	\|\| pVmxTransient->uExitReason == VMX_EXIT_SPP_EVENT)
13370	{
13371	/*
13372	* Execution of IRET caused an EPT violation, page-modification log-full event or
13373	* SPP-related event VM-exit when NMI blocking was in effect (i.e. we're in the
13374	* guest or nested-guest NMI handler). We need to set the block-by-NMI field so
13375	* that virtual NMIs remain blocked until the IRET execution is completed.
13376	*
13377	* See Intel spec. 27.2.3 "Information about NMI unblocking due to IRET"
13378	*/
13379	if (VMX_EXIT_QUAL_EPT_IS_NMI_UNBLOCK_IRET(pVmxTransient->uExitQual))
13380	{
13381	CPUMSetGuestNmiBlocking(pVCpu, true);
13382	Log4Func(("Set NMI blocking. uExitReason=%u\n", pVmxTransient->uExitReason));
13383	}
13384	}
13385	}
13386
13387	Assert( rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_HM_DOUBLE_FAULT
13388	\|\| rcStrict == VINF_EM_RESET \|\| rcStrict == VERR_EM_GUEST_CPU_HANG);
13389	return rcStrict;
13390	}
13391
13392
13393	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13394	/**
13395	* Perform the relevant VMX instruction checks for VM-exits that occurred due to the
13396	* guest attempting to execute a VMX instruction.
13397	*
13398	* @returns Strict VBox status code (i.e. informational status codes too).
13399	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
13400	* @retval VINF_HM_PENDING_XCPT if an exception was raised.
13401	*
13402	* @param pVCpu The cross context virtual CPU structure.
13403	* @param uExitReason The VM-exit reason.
13404	*
13405	* @todo NSTVMX: Document other error codes when VM-exit is implemented.
13406	* @remarks No-long-jump zone!!!
13407	*/
13408	static VBOXSTRICTRC hmR0VmxCheckExitDueToVmxInstr(PVMCPUCC pVCpu, uint32_t uExitReason)
13409	{
13410	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS
13411	\| CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_EFER);
13412
13413	/*
13414	* The physical CPU would have already checked the CPU mode/code segment.
13415	* We shall just assert here for paranoia.
13416	* See Intel spec. 25.1.1 "Relative Priority of Faults and VM Exits".
13417	*/
13418	Assert(!CPUMIsGuestInRealOrV86ModeEx(&pVCpu->cpum.GstCtx));
13419	Assert( !CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx)
13420	\|\| CPUMIsGuestIn64BitCodeEx(&pVCpu->cpum.GstCtx));
13421
13422	if (uExitReason == VMX_EXIT_VMXON)
13423	{
13424	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
13425
13426	/*
13427	* We check CR4.VMXE because it is required to be always set while in VMX operation
13428	* by physical CPUs and our CR4 read-shadow is only consulted when executing specific
13429	* instructions (CLTS, LMSW, MOV CR, and SMSW) and thus doesn't affect CPU operation
13430	* otherwise (i.e. physical CPU won't automatically #UD if Cr4Shadow.VMXE is 0).
13431	*/
13432	if (!CPUMIsGuestVmxEnabled(&pVCpu->cpum.GstCtx))
13433	{
13434	Log4Func(("CR4.VMXE is not set -> #UD\n"));
13435	hmR0VmxSetPendingXcptUD(pVCpu);
13436	return VINF_HM_PENDING_XCPT;
13437	}
13438	}
13439	else if (!CPUMIsGuestInVmxRootMode(&pVCpu->cpum.GstCtx))
13440	{
13441	/*
13442	* The guest has not entered VMX operation but attempted to execute a VMX instruction
13443	* (other than VMXON), we need to raise a #UD.
13444	*/
13445	Log4Func(("Not in VMX root mode -> #UD\n"));
13446	hmR0VmxSetPendingXcptUD(pVCpu);
13447	return VINF_HM_PENDING_XCPT;
13448	}
13449
13450	/* All other checks (including VM-exit intercepts) are handled by IEM instruction emulation. */
13451	return VINF_SUCCESS;
13452	}
13453
13454
13455	/**
13456	* Decodes the memory operand of an instruction that caused a VM-exit.
13457	*
13458	* The Exit qualification field provides the displacement field for memory
13459	* operand instructions, if any.
13460	*
13461	* @returns Strict VBox status code (i.e. informational status codes too).
13462	* @retval VINF_SUCCESS if the operand was successfully decoded.
13463	* @retval VINF_HM_PENDING_XCPT if an exception was raised while decoding the
13464	* operand.
13465	* @param pVCpu The cross context virtual CPU structure.
13466	* @param uExitInstrInfo The VM-exit instruction information field.
13467	* @param enmMemAccess The memory operand's access type (read or write).
13468	* @param GCPtrDisp The instruction displacement field, if any. For
13469	* RIP-relative addressing pass RIP + displacement here.
13470	* @param pGCPtrMem Where to store the effective destination memory address.
13471	*
13472	* @remarks Warning! This function ASSUMES the instruction cannot be used in real or
13473	* virtual-8086 mode hence skips those checks while verifying if the
13474	* segment is valid.
13475	*/
13476	static VBOXSTRICTRC hmR0VmxDecodeMemOperand(PVMCPUCC pVCpu, uint32_t uExitInstrInfo, RTGCPTR GCPtrDisp, VMXMEMACCESS enmMemAccess,
13477	PRTGCPTR pGCPtrMem)
13478	{
13479	Assert(pGCPtrMem);
13480	Assert(!CPUMIsGuestInRealOrV86Mode(pVCpu));
13481	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_EFER
13482	\| CPUMCTX_EXTRN_CR0);
13483
13484	static uint64_t const s_auAddrSizeMasks[] = { UINT64_C(0xffff), UINT64_C(0xffffffff), UINT64_C(0xffffffffffffffff) };
13485	static uint64_t const s_auAccessSizeMasks[] = { sizeof(uint16_t), sizeof(uint32_t), sizeof(uint64_t) };
13486	AssertCompile(RT_ELEMENTS(s_auAccessSizeMasks) == RT_ELEMENTS(s_auAddrSizeMasks));
13487
13488	VMXEXITINSTRINFO ExitInstrInfo;
13489	ExitInstrInfo.u = uExitInstrInfo;
13490	uint8_t const uAddrSize = ExitInstrInfo.All.u3AddrSize;
13491	uint8_t const iSegReg = ExitInstrInfo.All.iSegReg;
13492	bool const fIdxRegValid = !ExitInstrInfo.All.fIdxRegInvalid;
13493	uint8_t const iIdxReg = ExitInstrInfo.All.iIdxReg;
13494	uint8_t const uScale = ExitInstrInfo.All.u2Scaling;
13495	bool const fBaseRegValid = !ExitInstrInfo.All.fBaseRegInvalid;
13496	uint8_t const iBaseReg = ExitInstrInfo.All.iBaseReg;
13497	bool const fIsMemOperand = !ExitInstrInfo.All.fIsRegOperand;
13498	bool const fIsLongMode = CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx);
13499
13500	/*
13501	* Validate instruction information.
13502	* This shouldn't happen on real hardware but useful while testing our nested hardware-virtualization code.
13503	*/
13504	AssertLogRelMsgReturn(uAddrSize < RT_ELEMENTS(s_auAddrSizeMasks),
13505	("Invalid address size. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_1);
13506	AssertLogRelMsgReturn(iSegReg < X86_SREG_COUNT,
13507	("Invalid segment register. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_2);
13508	AssertLogRelMsgReturn(fIsMemOperand,
13509	("Expected memory operand. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_3);
13510
13511	/*
13512	* Compute the complete effective address.
13513	*
13514	* See AMD instruction spec. 1.4.2 "SIB Byte Format"
13515	* See AMD spec. 4.5.2 "Segment Registers".
13516	*/
13517	RTGCPTR GCPtrMem = GCPtrDisp;
13518	if (fBaseRegValid)
13519	GCPtrMem += pVCpu->cpum.GstCtx.aGRegs[iBaseReg].u64;
13520	if (fIdxRegValid)
13521	GCPtrMem += pVCpu->cpum.GstCtx.aGRegs[iIdxReg].u64 << uScale;
13522
13523	RTGCPTR const GCPtrOff = GCPtrMem;
13524	if ( !fIsLongMode
13525	\|\| iSegReg >= X86_SREG_FS)
13526	GCPtrMem += pVCpu->cpum.GstCtx.aSRegs[iSegReg].u64Base;
13527	GCPtrMem &= s_auAddrSizeMasks[uAddrSize];
13528
13529	/*
13530	* Validate effective address.
13531	* See AMD spec. 4.5.3 "Segment Registers in 64-Bit Mode".
13532	*/
13533	uint8_t const cbAccess = s_auAccessSizeMasks[uAddrSize];
13534	Assert(cbAccess > 0);
13535	if (fIsLongMode)
13536	{
13537	if (X86_IS_CANONICAL(GCPtrMem))
13538	{
13539	*pGCPtrMem = GCPtrMem;
13540	return VINF_SUCCESS;
13541	}
13542
13543	/** @todo r=ramshankar: We should probably raise \#SS or \#GP. See AMD spec. 4.12.2
13544	* "Data Limit Checks in 64-bit Mode". */
13545	Log4Func(("Long mode effective address is not canonical GCPtrMem=%#RX64\n", GCPtrMem));
13546	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13547	return VINF_HM_PENDING_XCPT;
13548	}
13549
13550	/*
13551	* This is a watered down version of iemMemApplySegment().
13552	* Parts that are not applicable for VMX instructions like real-or-v8086 mode
13553	* and segment CPL/DPL checks are skipped.
13554	*/
13555	RTGCPTR32 const GCPtrFirst32 = (RTGCPTR32)GCPtrOff;
13556	RTGCPTR32 const GCPtrLast32 = GCPtrFirst32 + cbAccess - 1;
13557	PCCPUMSELREG pSel = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
13558
13559	/* Check if the segment is present and usable. */
13560	if ( pSel->Attr.n.u1Present
13561	&& !pSel->Attr.n.u1Unusable)
13562	{
13563	Assert(pSel->Attr.n.u1DescType);
13564	if (!(pSel->Attr.n.u4Type & X86_SEL_TYPE_CODE))
13565	{
13566	/* Check permissions for the data segment. */
13567	if ( enmMemAccess == VMXMEMACCESS_WRITE
13568	&& !(pSel->Attr.n.u4Type & X86_SEL_TYPE_WRITE))
13569	{
13570	Log4Func(("Data segment access invalid. iSegReg=%#x Attr=%#RX32\n", iSegReg, pSel->Attr.u));
13571	hmR0VmxSetPendingXcptGP(pVCpu, iSegReg);
13572	return VINF_HM_PENDING_XCPT;
13573	}
13574
13575	/* Check limits if it's a normal data segment. */
13576	if (!(pSel->Attr.n.u4Type & X86_SEL_TYPE_DOWN))
13577	{
13578	if ( GCPtrFirst32 > pSel->u32Limit
13579	\|\| GCPtrLast32 > pSel->u32Limit)
13580	{
13581	Log4Func(("Data segment limit exceeded. "
13582	"iSegReg=%#x GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n", iSegReg, GCPtrFirst32,
13583	GCPtrLast32, pSel->u32Limit));
13584	if (iSegReg == X86_SREG_SS)
13585	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13586	else
13587	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13588	return VINF_HM_PENDING_XCPT;
13589	}
13590	}
13591	else
13592	{
13593	/* Check limits if it's an expand-down data segment.
13594	Note! The upper boundary is defined by the B bit, not the G bit! */
13595	if ( GCPtrFirst32 < pSel->u32Limit + UINT32_C(1)
13596	\|\| GCPtrLast32 > (pSel->Attr.n.u1DefBig ? UINT32_MAX : UINT32_C(0xffff)))
13597	{
13598	Log4Func(("Expand-down data segment limit exceeded. "
13599	"iSegReg=%#x GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n", iSegReg, GCPtrFirst32,
13600	GCPtrLast32, pSel->u32Limit));
13601	if (iSegReg == X86_SREG_SS)
13602	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13603	else
13604	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13605	return VINF_HM_PENDING_XCPT;
13606	}
13607	}
13608	}
13609	else
13610	{
13611	/* Check permissions for the code segment. */
13612	if ( enmMemAccess == VMXMEMACCESS_WRITE
13613	\|\| ( enmMemAccess == VMXMEMACCESS_READ
13614	&& !(pSel->Attr.n.u4Type & X86_SEL_TYPE_READ)))
13615	{
13616	Log4Func(("Code segment access invalid. Attr=%#RX32\n", pSel->Attr.u));
13617	Assert(!CPUMIsGuestInRealOrV86ModeEx(&pVCpu->cpum.GstCtx));
13618	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13619	return VINF_HM_PENDING_XCPT;
13620	}
13621
13622	/* Check limits for the code segment (normal/expand-down not applicable for code segments). */
13623	if ( GCPtrFirst32 > pSel->u32Limit
13624	\|\| GCPtrLast32 > pSel->u32Limit)
13625	{
13626	Log4Func(("Code segment limit exceeded. GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n",
13627	GCPtrFirst32, GCPtrLast32, pSel->u32Limit));
13628	if (iSegReg == X86_SREG_SS)
13629	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13630	else
13631	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13632	return VINF_HM_PENDING_XCPT;
13633	}
13634	}
13635	}
13636	else
13637	{
13638	Log4Func(("Not present or unusable segment. iSegReg=%#x Attr=%#RX32\n", iSegReg, pSel->Attr.u));
13639	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13640	return VINF_HM_PENDING_XCPT;
13641	}
13642
13643	*pGCPtrMem = GCPtrMem;
13644	return VINF_SUCCESS;
13645	}
13646	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13647
13648
13649	/**
13650	* VM-exit helper for LMSW.
13651	*/
13652	static VBOXSTRICTRC hmR0VmxExitLmsw(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint16_t uMsw, RTGCPTR GCPtrEffDst)
13653	{
13654	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13655	AssertRCReturn(rc, rc);
13656
13657	VBOXSTRICTRC rcStrict = IEMExecDecodedLmsw(pVCpu, cbInstr, uMsw, GCPtrEffDst);
13658	AssertMsg( rcStrict == VINF_SUCCESS
13659	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13660
13661	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
13662	if (rcStrict == VINF_IEM_RAISED_XCPT)
13663	{
13664	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13665	rcStrict = VINF_SUCCESS;
13666	}
13667
13668	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitLmsw);
13669	Log4Func(("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13670	return rcStrict;
13671	}
13672
13673
13674	/**
13675	* VM-exit helper for CLTS.
13676	*/
13677	static VBOXSTRICTRC hmR0VmxExitClts(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr)
13678	{
13679	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13680	AssertRCReturn(rc, rc);
13681
13682	VBOXSTRICTRC rcStrict = IEMExecDecodedClts(pVCpu, cbInstr);
13683	AssertMsg( rcStrict == VINF_SUCCESS
13684	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13685
13686	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
13687	if (rcStrict == VINF_IEM_RAISED_XCPT)
13688	{
13689	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13690	rcStrict = VINF_SUCCESS;
13691	}
13692
13693	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitClts);
13694	Log4Func(("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13695	return rcStrict;
13696	}
13697
13698
13699	/**
13700	* VM-exit helper for MOV from CRx (CRx read).
13701	*/
13702	static VBOXSTRICTRC hmR0VmxExitMovFromCrX(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint8_t iGReg, uint8_t iCrReg)
13703	{
13704	Assert(iCrReg < 16);
13705	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
13706
13707	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13708	AssertRCReturn(rc, rc);
13709
13710	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxRead(pVCpu, cbInstr, iGReg, iCrReg);
13711	AssertMsg( rcStrict == VINF_SUCCESS
13712	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13713
13714	if (iGReg == X86_GREG_xSP)
13715	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_RSP);
13716	else
13717	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
13718	#ifdef VBOX_WITH_STATISTICS
13719	switch (iCrReg)
13720	{
13721	case 0: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Read); break;
13722	case 2: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Read); break;
13723	case 3: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Read); break;
13724	case 4: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Read); break;
13725	case 8: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Read); break;
13726	}
13727	#endif
13728	Log4Func(("CR%d Read access rcStrict=%Rrc\n", iCrReg, VBOXSTRICTRC_VAL(rcStrict)));
13729	return rcStrict;
13730	}
13731
13732
13733	/**
13734	* VM-exit helper for MOV to CRx (CRx write).
13735	*/
13736	static VBOXSTRICTRC hmR0VmxExitMovToCrX(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint8_t iGReg, uint8_t iCrReg)
13737	{
13738	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13739	AssertRCReturn(rc, rc);
13740
13741	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxWrite(pVCpu, cbInstr, iCrReg, iGReg);
13742	AssertMsg( rcStrict == VINF_SUCCESS
13743	\|\| rcStrict == VINF_IEM_RAISED_XCPT
13744	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13745
13746	switch (iCrReg)
13747	{
13748	case 0:
13749	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0
13750	\| HM_CHANGED_GUEST_EFER_MSR \| HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
13751	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Write);
13752	Log4Func(("CR0 write. rcStrict=%Rrc CR0=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cr0));
13753	break;
13754
13755	case 2:
13756	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Write);
13757	/* Nothing to do here, CR2 it's not part of the VMCS. */
13758	break;
13759
13760	case 3:
13761	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR3);
13762	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Write);
13763	Log4Func(("CR3 write. rcStrict=%Rrc CR3=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cr3));
13764	break;
13765
13766	case 4:
13767	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR4);
13768	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Write);
13769	Log4Func(("CR4 write. rc=%Rrc CR4=%#RX64 fLoadSaveGuestXcr0=%u\n", VBOXSTRICTRC_VAL(rcStrict),
13770	pVCpu->cpum.GstCtx.cr4, pVCpu->hm.s.fLoadSaveGuestXcr0));
13771	break;
13772
13773	case 8:
13774	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged,
13775	HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_APIC_TPR);
13776	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Write);
13777	break;
13778
13779	default:
13780	AssertMsgFailed(("Invalid CRx register %#x\n", iCrReg));
13781	break;
13782	}
13783
13784	if (rcStrict == VINF_IEM_RAISED_XCPT)
13785	{
13786	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13787	rcStrict = VINF_SUCCESS;
13788	}
13789	return rcStrict;
13790	}
13791
13792
13793	/**
13794	* VM-exit exception handler for \#PF (Page-fault exception).
13795	*
13796	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13797	*/
13798	static VBOXSTRICTRC hmR0VmxExitXcptPF(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13799	{
13800	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13801	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
13802	hmR0VmxReadExitQualVmcs(pVmxTransient);
13803
13804	if (!pVM->hm.s.fNestedPaging)
13805	{ /* likely */ }
13806	else
13807	{
13808	#if !defined(HMVMX_ALWAYS_TRAP_ALL_XCPTS) && !defined(HMVMX_ALWAYS_TRAP_PF)
13809	Assert(pVmxTransient->fIsNestedGuest \|\| pVCpu->hm.s.fUsingDebugLoop);
13810	#endif
13811	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
13812	if (!pVmxTransient->fVectoringDoublePF)
13813	{
13814	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */,
13815	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual);
13816	}
13817	else
13818	{
13819	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
13820	Assert(!pVmxTransient->fIsNestedGuest);
13821	hmR0VmxSetPendingXcptDF(pVCpu);
13822	Log4Func(("Pending #DF due to vectoring #PF w/ NestedPaging\n"));
13823	}
13824	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
13825	return VINF_SUCCESS;
13826	}
13827
13828	Assert(!pVmxTransient->fIsNestedGuest);
13829
13830	/* If it's a vectoring #PF, emulate injecting the original event injection as PGMTrap0eHandler() is incapable
13831	of differentiating between instruction emulation and event injection that caused a #PF. See @bugref{6607}. */
13832	if (pVmxTransient->fVectoringPF)
13833	{
13834	Assert(pVCpu->hm.s.Event.fPending);
13835	return VINF_EM_RAW_INJECT_TRPM_EVENT;
13836	}
13837
13838	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
13839	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
13840	AssertRCReturn(rc, rc);
13841
13842	Log4Func(("#PF: cs:rip=%#04x:%#RX64 err_code=%#RX32 exit_qual=%#RX64 cr3=%#RX64\n", pCtx->cs.Sel, pCtx->rip,
13843	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual, pCtx->cr3));
13844
13845	TRPMAssertXcptPF(pVCpu, pVmxTransient->uExitQual, (RTGCUINT)pVmxTransient->uExitIntErrorCode);
13846	rc = PGMTrap0eHandler(pVCpu, pVmxTransient->uExitIntErrorCode, CPUMCTX2CORE(pCtx), (RTGCPTR)pVmxTransient->uExitQual);
13847
13848	Log4Func(("#PF: rc=%Rrc\n", rc));
13849	if (rc == VINF_SUCCESS)
13850	{
13851	/*
13852	* This is typically a shadow page table sync or a MMIO instruction. But we may have
13853	* emulated something like LTR or a far jump. Any part of the CPU context may have changed.
13854	*/
13855	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
13856	TRPMResetTrap(pVCpu);
13857	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
13858	return rc;
13859	}
13860
13861	if (rc == VINF_EM_RAW_GUEST_TRAP)
13862	{
13863	if (!pVmxTransient->fVectoringDoublePF)
13864	{
13865	/* It's a guest page fault and needs to be reflected to the guest. */
13866	uint32_t const uGstErrorCode = TRPMGetErrorCode(pVCpu);
13867	TRPMResetTrap(pVCpu);
13868	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory #PF. */
13869	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */,
13870	uGstErrorCode, pVmxTransient->uExitQual);
13871	}
13872	else
13873	{
13874	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
13875	TRPMResetTrap(pVCpu);
13876	pVCpu->hm.s.Event.fPending = false; /* Clear pending #PF to replace it with #DF. */
13877	hmR0VmxSetPendingXcptDF(pVCpu);
13878	Log4Func(("#PF: Pending #DF due to vectoring #PF\n"));
13879	}
13880
13881	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
13882	return VINF_SUCCESS;
13883	}
13884
13885	TRPMResetTrap(pVCpu);
13886	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
13887	return rc;
13888	}
13889
13890
13891	/**
13892	* VM-exit exception handler for \#MF (Math Fault: floating point exception).
13893	*
13894	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13895	*/
13896	static VBOXSTRICTRC hmR0VmxExitXcptMF(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13897	{
13898	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13899	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
13900
13901	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR0);
13902	AssertRCReturn(rc, rc);
13903
13904	if (!(pVCpu->cpum.GstCtx.cr0 & X86_CR0_NE))
13905	{
13906	/* Convert a #MF into a FERR -> IRQ 13. See @bugref{6117}. */
13907	rc = PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13, 1, 0 /* uTagSrc */);
13908
13909	/** @todo r=ramshankar: The Intel spec. does -not- specify that this VM-exit
13910	* provides VM-exit instruction length. If this causes problem later,
13911	* disassemble the instruction like it's done on AMD-V. */
13912	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
13913	AssertRCReturn(rc2, rc2);
13914	return rc;
13915	}
13916
13917	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbExitInstr,
13918	pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13919	return VINF_SUCCESS;
13920	}
13921
13922
13923	/**
13924	* VM-exit exception handler for \#BP (Breakpoint exception).
13925	*
13926	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13927	*/
13928	static VBOXSTRICTRC hmR0VmxExitXcptBP(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13929	{
13930	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13931	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP);
13932
13933	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
13934	AssertRCReturn(rc, rc);
13935
13936	if (!pVmxTransient->fIsNestedGuest)
13937	rc = DBGFRZTrap03Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(&pVCpu->cpum.GstCtx));
13938	else
13939	rc = VINF_EM_RAW_GUEST_TRAP;
13940
13941	if (rc == VINF_EM_RAW_GUEST_TRAP)
13942	{
13943	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13944	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13945	rc = VINF_SUCCESS;
13946	}
13947
13948	Assert(rc == VINF_SUCCESS \|\| rc == VINF_EM_DBG_BREAKPOINT);
13949	return rc;
13950	}
13951
13952
13953	/**
13954	* VM-exit exception handler for \#AC (Alignment-check exception).
13955	*
13956	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13957	*/
13958	static VBOXSTRICTRC hmR0VmxExitXcptAC(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13959	{
13960	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13961	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC);
13962
13963	/* Re-inject it. We'll detect any nesting before getting here. */
13964	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13965	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13966	return VINF_SUCCESS;
13967	}
13968
13969
13970	/**
13971	* VM-exit exception handler for \#DB (Debug exception).
13972	*
13973	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13974	*/
13975	static VBOXSTRICTRC hmR0VmxExitXcptDB(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13976	{
13977	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
13978	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
13979
13980	/*
13981	* Get the DR6-like values from the Exit qualification and pass it to DBGF for processing.
13982	*/
13983	hmR0VmxReadExitQualVmcs(pVmxTransient);
13984
13985	/* Refer Intel spec. Table 27-1. "Exit Qualifications for debug exceptions" for the format. */
13986	uint64_t const uDR6 = X86_DR6_INIT_VAL
13987	\| (pVmxTransient->uExitQual & ( X86_DR6_B0 \| X86_DR6_B1 \| X86_DR6_B2 \| X86_DR6_B3
13988	\| X86_DR6_BD \| X86_DR6_BS));
13989
13990	int rc;
13991	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
13992	if (!pVmxTransient->fIsNestedGuest)
13993	{
13994	rc = DBGFRZTrap01Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx), uDR6, pVCpu->hm.s.fSingleInstruction);
13995
13996	/*
13997	* Prevents stepping twice over the same instruction when the guest is stepping using
13998	* EFLAGS.TF and the hypervisor debugger is stepping using MTF.
13999	* Testcase: DOSQEMM, break (using "ba x 1") at cs:rip 0x70:0x774 and step (using "t").
14000	*/
14001	if ( rc == VINF_EM_DBG_STEPPED
14002	&& (pVmxTransient->pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MONITOR_TRAP_FLAG))
14003	{
14004	Assert(pVCpu->hm.s.fSingleInstruction);
14005	rc = VINF_EM_RAW_GUEST_TRAP;
14006	}
14007	}
14008	else
14009	rc = VINF_EM_RAW_GUEST_TRAP;
14010	Log6Func(("rc=%Rrc\n", rc));
14011	if (rc == VINF_EM_RAW_GUEST_TRAP)
14012	{
14013	/*
14014	* The exception was for the guest. Update DR6, DR7.GD and
14015	* IA32_DEBUGCTL.LBR before forwarding it.
14016	* See Intel spec. 27.1 "Architectural State before a VM-Exit".
14017	*/
14018	VMMRZCallRing3Disable(pVCpu);
14019	HM_DISABLE_PREEMPT(pVCpu);
14020
14021	pCtx->dr[6] &= ~X86_DR6_B_MASK;
14022	pCtx->dr[6] \|= uDR6;
14023	if (CPUMIsGuestDebugStateActive(pVCpu))
14024	ASMSetDR6(pCtx->dr[6]);
14025
14026	HM_RESTORE_PREEMPT();
14027	VMMRZCallRing3Enable(pVCpu);
14028
14029	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_DR7);
14030	AssertRCReturn(rc, rc);
14031
14032	/* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */
14033	pCtx->dr[7] &= ~(uint64_t)X86_DR7_GD;
14034
14035	/* Paranoia. */
14036	pCtx->dr[7] &= ~(uint64_t)X86_DR7_RAZ_MASK;
14037	pCtx->dr[7] \|= X86_DR7_RA1_MASK;
14038
14039	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, pCtx->dr[7]);
14040	AssertRC(rc);
14041
14042	/*
14043	* Raise #DB in the guest.
14044	*
14045	* It is important to reflect exactly what the VM-exit gave us (preserving the
14046	* interruption-type) rather than use hmR0VmxSetPendingXcptDB() as the #DB could've
14047	* been raised while executing ICEBP (INT1) and not the regular #DB. Thus it may
14048	* trigger different handling in the CPU (like skipping DPL checks), see @bugref{6398}.
14049	*
14050	* Intel re-documented ICEBP/INT1 on May 2018 previously documented as part of
14051	* Intel 386, see Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
14052	*/
14053	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14054	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14055	return VINF_SUCCESS;
14056	}
14057
14058	/*
14059	* Not a guest trap, must be a hypervisor related debug event then.
14060	* Update DR6 in case someone is interested in it.
14061	*/
14062	AssertMsg(rc == VINF_EM_DBG_STEPPED \|\| rc == VINF_EM_DBG_BREAKPOINT, ("%Rrc\n", rc));
14063	AssertReturn(pVmxTransient->fWasHyperDebugStateActive, VERR_HM_IPE_5);
14064	CPUMSetHyperDR6(pVCpu, uDR6);
14065
14066	return rc;
14067	}
14068
14069
14070	/**
14071	* Hacks its way around the lovely mesa driver's backdoor accesses.
14072	*
14073	* @sa hmR0SvmHandleMesaDrvGp.
14074	*/
14075	static int hmR0VmxHandleMesaDrvGp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PCPUMCTX pCtx)
14076	{
14077	LogFunc(("cs:rip=%#04x:%#RX64 rcx=%#RX64 rbx=%#RX64\n", pCtx->cs.Sel, pCtx->rip, pCtx->rcx, pCtx->rbx));
14078	RT_NOREF(pCtx);
14079
14080	/* For now we'll just skip the instruction. */
14081	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14082	}
14083
14084
14085	/**
14086	* Checks if the \#GP'ing instruction is the mesa driver doing it's lovely
14087	* backdoor logging w/o checking what it is running inside.
14088	*
14089	* This recognizes an "IN EAX,DX" instruction executed in flat ring-3, with the
14090	* backdoor port and magic numbers loaded in registers.
14091	*
14092	* @returns true if it is, false if it isn't.
14093	* @sa hmR0SvmIsMesaDrvGp.
14094	*/
14095	DECLINLINE(bool) hmR0VmxIsMesaDrvGp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PCPUMCTX pCtx)
14096	{
14097	/* 0xed: IN eAX,dx */
14098	uint8_t abInstr[1];
14099	if (pVmxTransient->cbExitInstr != sizeof(abInstr))
14100	return false;
14101
14102	/* Check that it is #GP(0). */
14103	if (pVmxTransient->uExitIntErrorCode != 0)
14104	return false;
14105
14106	/* Check magic and port. */
14107	Assert(!(pCtx->fExtrn & (CPUMCTX_EXTRN_RAX \| CPUMCTX_EXTRN_RDX \| CPUMCTX_EXTRN_RCX)));
14108	/Log(("hmR0VmxIsMesaDrvGp: rax=%RX64 rdx=%RX64\n", pCtx->rax, pCtx->rdx));/
14109	if (pCtx->rax != UINT32_C(0x564d5868))
14110	return false;
14111	if (pCtx->dx != UINT32_C(0x5658))
14112	return false;
14113
14114	/* Flat ring-3 CS. */
14115	AssertCompile(HMVMX_CPUMCTX_EXTRN_ALL & CPUMCTX_EXTRN_CS);
14116	Assert(!(pCtx->fExtrn & CPUMCTX_EXTRN_CS));
14117	/Log(("hmR0VmxIsMesaDrvGp: cs.Attr.n.u2Dpl=%d base=%Rx64\n", pCtx->cs.Attr.n.u2Dpl, pCtx->cs.u64Base));/
14118	if (pCtx->cs.Attr.n.u2Dpl != 3)
14119	return false;
14120	if (pCtx->cs.u64Base != 0)
14121	return false;
14122
14123	/* Check opcode. */
14124	AssertCompile(HMVMX_CPUMCTX_EXTRN_ALL & CPUMCTX_EXTRN_RIP);
14125	Assert(!(pCtx->fExtrn & CPUMCTX_EXTRN_RIP));
14126	int rc = PGMPhysSimpleReadGCPtr(pVCpu, abInstr, pCtx->rip, sizeof(abInstr));
14127	/Log(("hmR0VmxIsMesaDrvGp: PGMPhysSimpleReadGCPtr -> %Rrc %#x\n", rc, abInstr[0]));/
14128	if (RT_FAILURE(rc))
14129	return false;
14130	if (abInstr[0] != 0xed)
14131	return false;
14132
14133	return true;
14134	}
14135
14136
14137	/**
14138	* VM-exit exception handler for \#GP (General-protection exception).
14139	*
14140	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14141	*/
14142	static VBOXSTRICTRC hmR0VmxExitXcptGP(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14143	{
14144	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14145	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP);
14146
14147	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14148	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14149	if (pVmcsInfo->RealMode.fRealOnV86Active)
14150	{ /* likely */ }
14151	else
14152	{
14153	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14154	Assert(pVCpu->hm.s.fUsingDebugLoop \|\| pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv \|\| pVmxTransient->fIsNestedGuest);
14155	#endif
14156	/*
14157	* If the guest is not in real-mode or we have unrestricted guest execution support, or if we are
14158	* executing a nested-guest, reflect #GP to the guest or nested-guest.
14159	*/
14160	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14161	AssertRCReturn(rc, rc);
14162	Log4Func(("Gst: cs:rip=%#04x:%#RX64 ErrorCode=%#x cr0=%#RX64 cpl=%u tr=%#04x\n", pCtx->cs.Sel, pCtx->rip,
14163	pVmxTransient->uExitIntErrorCode, pCtx->cr0, CPUMGetGuestCPL(pVCpu), pCtx->tr.Sel));
14164
14165	if ( pVmxTransient->fIsNestedGuest
14166	\|\| !pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv
14167	\|\| !hmR0VmxIsMesaDrvGp(pVCpu, pVmxTransient, pCtx))
14168	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14169	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14170	else
14171	rc = hmR0VmxHandleMesaDrvGp(pVCpu, pVmxTransient, pCtx);
14172	return rc;
14173	}
14174
14175	Assert(CPUMIsGuestInRealModeEx(pCtx));
14176	Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest);
14177	Assert(!pVmxTransient->fIsNestedGuest);
14178
14179	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14180	AssertRCReturn(rc, rc);
14181
14182	VBOXSTRICTRC rcStrict = IEMExecOne(pVCpu);
14183	if (rcStrict == VINF_SUCCESS)
14184	{
14185	if (!CPUMIsGuestInRealModeEx(pCtx))
14186	{
14187	/*
14188	* The guest is no longer in real-mode, check if we can continue executing the
14189	* guest using hardware-assisted VMX. Otherwise, fall back to emulation.
14190	*/
14191	pVmcsInfo->RealMode.fRealOnV86Active = false;
14192	if (HMCanExecuteVmxGuest(pVCpu->pVMR0, pVCpu, pCtx))
14193	{
14194	Log4Func(("Mode changed but guest still suitable for executing using hardware-assisted VMX\n"));
14195	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14196	}
14197	else
14198	{
14199	Log4Func(("Mode changed -> VINF_EM_RESCHEDULE\n"));
14200	rcStrict = VINF_EM_RESCHEDULE;
14201	}
14202	}
14203	else
14204	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14205	}
14206	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14207	{
14208	rcStrict = VINF_SUCCESS;
14209	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14210	}
14211	return VBOXSTRICTRC_VAL(rcStrict);
14212	}
14213
14214
14215	/**
14216	* VM-exit exception handler wrapper for all other exceptions that are not handled
14217	* by a specific handler.
14218	*
14219	* This simply re-injects the exception back into the VM without any special
14220	* processing.
14221	*
14222	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14223	*/
14224	static VBOXSTRICTRC hmR0VmxExitXcptOthers(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14225	{
14226	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14227
14228	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14229	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14230	AssertMsg(pVCpu->hm.s.fUsingDebugLoop \|\| pVmcsInfo->RealMode.fRealOnV86Active \|\| pVmxTransient->fIsNestedGuest,
14231	("uVector=%#x u32XcptBitmap=%#X32\n",
14232	VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo), pVmcsInfo->u32XcptBitmap));
14233	NOREF(pVmcsInfo);
14234	#endif
14235
14236	/*
14237	* Re-inject the exception into the guest. This cannot be a double-fault condition which
14238	* would have been handled while checking exits due to event delivery.
14239	*/
14240	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14241
14242	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14243	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
14244	AssertRCReturn(rc, rc);
14245	Log4Func(("Reinjecting Xcpt. uVector=%#x cs:rip=%#04x:%#RX64\n", uVector, pCtx->cs.Sel, pCtx->rip));
14246	#endif
14247
14248	#ifdef VBOX_WITH_STATISTICS
14249	switch (uVector)
14250	{
14251	case X86_XCPT_DE: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE); break;
14252	case X86_XCPT_DB: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB); break;
14253	case X86_XCPT_BP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP); break;
14254	case X86_XCPT_OF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
14255	case X86_XCPT_BR: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBR); break;
14256	case X86_XCPT_UD: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD); break;
14257	case X86_XCPT_NM: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
14258	case X86_XCPT_DF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDF); break;
14259	case X86_XCPT_TS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestTS); break;
14260	case X86_XCPT_NP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNP); break;
14261	case X86_XCPT_SS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestSS); break;
14262	case X86_XCPT_GP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP); break;
14263	case X86_XCPT_PF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF); break;
14264	case X86_XCPT_MF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF); break;
14265	case X86_XCPT_AC: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC); break;
14266	case X86_XCPT_XF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXF); break;
14267	default:
14268	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXcpUnk);
14269	break;
14270	}
14271	#endif
14272
14273	/* We should never call this function for a page-fault, we'd need to pass on the fault address below otherwise. */
14274	Assert(!VMX_EXIT_INT_INFO_IS_XCPT_PF(pVmxTransient->uExitIntInfo));
14275	NOREF(uVector);
14276
14277	/* Re-inject the original exception into the guest. */
14278	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14279	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14280	return VINF_SUCCESS;
14281	}
14282
14283
14284	/**
14285	* VM-exit exception handler for all exceptions (except NMIs!).
14286	*
14287	* @remarks This may be called for both guests and nested-guests. Take care to not
14288	* make assumptions and avoid doing anything that is not relevant when
14289	* executing a nested-guest (e.g., Mesa driver hacks).
14290	*/
14291	static VBOXSTRICTRC hmR0VmxExitXcpt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14292	{
14293	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_XCPT_INFO);
14294
14295	/*
14296	* If this VM-exit occurred while delivering an event through the guest IDT, take
14297	* action based on the return code and additional hints (e.g. for page-faults)
14298	* that will be updated in the VMX transient structure.
14299	*/
14300	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
14301	if (rcStrict == VINF_SUCCESS)
14302	{
14303	/*
14304	* If an exception caused a VM-exit due to delivery of an event, the original
14305	* event may have to be re-injected into the guest. We shall reinject it and
14306	* continue guest execution. However, page-fault is a complicated case and
14307	* needs additional processing done in hmR0VmxExitXcptPF().
14308	*/
14309	Assert(VMX_EXIT_INT_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
14310	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14311	if ( !pVCpu->hm.s.Event.fPending
14312	\|\| uVector == X86_XCPT_PF)
14313	{
14314	switch (uVector)
14315	{
14316	case X86_XCPT_PF: return hmR0VmxExitXcptPF(pVCpu, pVmxTransient);
14317	case X86_XCPT_GP: return hmR0VmxExitXcptGP(pVCpu, pVmxTransient);
14318	case X86_XCPT_MF: return hmR0VmxExitXcptMF(pVCpu, pVmxTransient);
14319	case X86_XCPT_DB: return hmR0VmxExitXcptDB(pVCpu, pVmxTransient);
14320	case X86_XCPT_BP: return hmR0VmxExitXcptBP(pVCpu, pVmxTransient);
14321	case X86_XCPT_AC: return hmR0VmxExitXcptAC(pVCpu, pVmxTransient);
14322	default:
14323	return hmR0VmxExitXcptOthers(pVCpu, pVmxTransient);
14324	}
14325	}
14326	/* else: inject pending event before resuming guest execution. */
14327	}
14328	else if (rcStrict == VINF_HM_DOUBLE_FAULT)
14329	{
14330	Assert(pVCpu->hm.s.Event.fPending);
14331	rcStrict = VINF_SUCCESS;
14332	}
14333
14334	return rcStrict;
14335	}
14336	/** @} */
14337
14338
14339	/** @name VM-exit handlers.
14340	* @{
14341	*/
14342	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
14343	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
14344	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
14345
14346	/**
14347	* VM-exit handler for external interrupts (VMX_EXIT_EXT_INT).
14348	*/
14349	HMVMX_EXIT_DECL hmR0VmxExitExtInt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14350	{
14351	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14352	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt);
14353	/* Windows hosts (32-bit and 64-bit) have DPC latency issues. See @bugref{6853}. */
14354	if (VMMR0ThreadCtxHookIsEnabled(pVCpu))
14355	return VINF_SUCCESS;
14356	return VINF_EM_RAW_INTERRUPT;
14357	}
14358
14359
14360	/**
14361	* VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI). Conditional
14362	* VM-exit.
14363	*/
14364	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmi(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14365	{
14366	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14367	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitXcptNmi, y3);
14368
14369	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
14370
14371	uint32_t const uExitIntType = VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo);
14372	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14373	Assert(VMX_EXIT_INT_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
14374
14375	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14376	Assert( !(pVmcsInfo->u32ExitCtls & VMX_EXIT_CTLS_ACK_EXT_INT)
14377	&& uExitIntType != VMX_EXIT_INT_INFO_TYPE_EXT_INT);
14378	NOREF(pVmcsInfo);
14379
14380	VBOXSTRICTRC rcStrict;
14381	switch (uExitIntType)
14382	{
14383	/*
14384	* Host physical NMIs:
14385	* This cannot be a guest NMI as the only way for the guest to receive an NMI is if we
14386	* injected it ourselves and anything we inject is not going to cause a VM-exit directly
14387	* for the event being injected[1]. Go ahead and dispatch the NMI to the host[2].
14388	*
14389	* See Intel spec. 27.2.3 "Information for VM Exits During Event Delivery".
14390	* See Intel spec. 27.5.5 "Updating Non-Register State".
14391	*/
14392	case VMX_EXIT_INT_INFO_TYPE_NMI:
14393	{
14394	rcStrict = hmR0VmxExitHostNmi(pVCpu, pVmcsInfo);
14395	break;
14396	}
14397
14398	/*
14399	* Privileged software exceptions (#DB from ICEBP),
14400	* Software exceptions (#BP and #OF),
14401	* Hardware exceptions:
14402	* Process the required exceptions and resume guest execution if possible.
14403	*/
14404	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
14405	Assert(uVector == X86_XCPT_DB);
14406	RT_FALL_THRU();
14407	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
14408	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF \|\| uExitIntType == VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT);
14409	RT_FALL_THRU();
14410	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
14411	{
14412	NOREF(uVector);
14413	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
14414	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14415	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
14416	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
14417
14418	rcStrict = hmR0VmxExitXcpt(pVCpu, pVmxTransient);
14419	break;
14420	}
14421
14422	default:
14423	{
14424	pVCpu->hm.s.u32HMError = pVmxTransient->uExitIntInfo;
14425	rcStrict = VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
14426	AssertMsgFailed(("Invalid/unexpected VM-exit interruption info %#x\n", pVmxTransient->uExitIntInfo));
14427	break;
14428	}
14429	}
14430
14431	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
14432	return rcStrict;
14433	}
14434
14435
14436	/**
14437	* VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
14438	*/
14439	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindow(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14440	{
14441	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14442
14443	/* Indicate that we no longer need to VM-exit when the guest is ready to receive interrupts, it is now ready. */
14444	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14445	hmR0VmxClearIntWindowExitVmcs(pVmcsInfo);
14446
14447	/* Evaluate and deliver pending events and resume guest execution. */
14448	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow);
14449	return VINF_SUCCESS;
14450	}
14451
14452
14453	/**
14454	* VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
14455	*/
14456	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindow(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14457	{
14458	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14459
14460	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14461	if (RT_UNLIKELY(!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))) /** @todo NSTVMX: Turn this into an assertion. */
14462	{
14463	AssertMsgFailed(("Unexpected NMI-window exit.\n"));
14464	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
14465	}
14466
14467	Assert(!CPUMIsGuestNmiBlocking(pVCpu));
14468
14469	/*
14470	* If block-by-STI is set when we get this VM-exit, it means the CPU doesn't block NMIs following STI.
14471	* It is therefore safe to unblock STI and deliver the NMI ourselves. See @bugref{7445}.
14472	*/
14473	uint32_t fIntrState;
14474	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
14475	AssertRC(rc);
14476	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
14477	if (fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
14478	{
14479	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
14480	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
14481
14482	fIntrState &= ~VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
14483	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
14484	AssertRC(rc);
14485	}
14486
14487	/* Indicate that we no longer need to VM-exit when the guest is ready to receive NMIs, it is now ready */
14488	hmR0VmxClearNmiWindowExitVmcs(pVmcsInfo);
14489
14490	/* Evaluate and deliver pending events and resume guest execution. */
14491	return VINF_SUCCESS;
14492	}
14493
14494
14495	/**
14496	* VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
14497	*/
14498	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvd(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14499	{
14500	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14501	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14502	}
14503
14504
14505	/**
14506	* VM-exit handler for INVD (VMX_EXIT_INVD). Unconditional VM-exit.
14507	*/
14508	HMVMX_EXIT_NSRC_DECL hmR0VmxExitInvd(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14509	{
14510	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14511	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14512	}
14513
14514
14515	/**
14516	* VM-exit handler for CPUID (VMX_EXIT_CPUID). Unconditional VM-exit.
14517	*/
14518	HMVMX_EXIT_DECL hmR0VmxExitCpuid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14519	{
14520	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14521
14522	/*
14523	* Get the state we need and update the exit history entry.
14524	*/
14525	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14526	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14527
14528	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
14529	AssertRCReturn(rc, rc);
14530
14531	VBOXSTRICTRC rcStrict;
14532	PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
14533	EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_CPUID),
14534	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
14535	if (!pExitRec)
14536	{
14537	/*
14538	* Regular CPUID instruction execution.
14539	*/
14540	rcStrict = IEMExecDecodedCpuid(pVCpu, pVmxTransient->cbExitInstr);
14541	if (rcStrict == VINF_SUCCESS)
14542	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14543	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14544	{
14545	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14546	rcStrict = VINF_SUCCESS;
14547	}
14548	}
14549	else
14550	{
14551	/*
14552	* Frequent exit or something needing probing. Get state and call EMHistoryExec.
14553	*/
14554	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14555	AssertRCReturn(rc2, rc2);
14556
14557	Log4(("CpuIdExit/%u: %04x:%08RX64: %#x/%#x -> EMHistoryExec\n",
14558	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.eax, pVCpu->cpum.GstCtx.ecx));
14559
14560	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
14561	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14562
14563	Log4(("CpuIdExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
14564	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
14565	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
14566	}
14567	return rcStrict;
14568	}
14569
14570
14571	/**
14572	* VM-exit handler for GETSEC (VMX_EXIT_GETSEC). Unconditional VM-exit.
14573	*/
14574	HMVMX_EXIT_DECL hmR0VmxExitGetsec(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14575	{
14576	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14577
14578	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14579	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR4);
14580	AssertRCReturn(rc, rc);
14581
14582	if (pVCpu->cpum.GstCtx.cr4 & X86_CR4_SMXE)
14583	return VINF_EM_RAW_EMULATE_INSTR;
14584
14585	AssertMsgFailed(("hmR0VmxExitGetsec: Unexpected VM-exit when CR4.SMXE is 0.\n"));
14586	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
14587	}
14588
14589
14590	/**
14591	* VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
14592	*/
14593	HMVMX_EXIT_DECL hmR0VmxExitRdtsc(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14594	{
14595	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14596
14597	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14598	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14599	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14600	AssertRCReturn(rc, rc);
14601
14602	VBOXSTRICTRC rcStrict = IEMExecDecodedRdtsc(pVCpu, pVmxTransient->cbExitInstr);
14603	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
14604	{
14605	/* If we get a spurious VM-exit when TSC offsetting is enabled,
14606	we must reset offsetting on VM-entry. See @bugref{6634}. */
14607	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TSC_OFFSETTING)
14608	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14609	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14610	}
14611	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14612	{
14613	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14614	rcStrict = VINF_SUCCESS;
14615	}
14616	return rcStrict;
14617	}
14618
14619
14620	/**
14621	* VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
14622	*/
14623	HMVMX_EXIT_DECL hmR0VmxExitRdtscp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14624	{
14625	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14626
14627	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14628	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14629	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_TSC_AUX);
14630	AssertRCReturn(rc, rc);
14631
14632	VBOXSTRICTRC rcStrict = IEMExecDecodedRdtscp(pVCpu, pVmxTransient->cbExitInstr);
14633	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
14634	{
14635	/* If we get a spurious VM-exit when TSC offsetting is enabled,
14636	we must reset offsetting on VM-reentry. See @bugref{6634}. */
14637	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TSC_OFFSETTING)
14638	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14639	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14640	}
14641	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14642	{
14643	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14644	rcStrict = VINF_SUCCESS;
14645	}
14646	return rcStrict;
14647	}
14648
14649
14650	/**
14651	* VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
14652	*/
14653	HMVMX_EXIT_DECL hmR0VmxExitRdpmc(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14654	{
14655	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14656
14657	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14658	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR4 \| CPUMCTX_EXTRN_CR0
14659	\| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS);
14660	AssertRCReturn(rc, rc);
14661
14662	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14663	rc = EMInterpretRdpmc(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx));
14664	if (RT_LIKELY(rc == VINF_SUCCESS))
14665	{
14666	rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14667	Assert(pVmxTransient->cbExitInstr == 2);
14668	}
14669	else
14670	{
14671	AssertMsgFailed(("hmR0VmxExitRdpmc: EMInterpretRdpmc failed with %Rrc\n", rc));
14672	rc = VERR_EM_INTERPRETER;
14673	}
14674	return rc;
14675	}
14676
14677
14678	/**
14679	* VM-exit handler for VMCALL (VMX_EXIT_VMCALL). Unconditional VM-exit.
14680	*/
14681	HMVMX_EXIT_DECL hmR0VmxExitVmcall(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14682	{
14683	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14684
14685	VBOXSTRICTRC rcStrict = VERR_VMX_IPE_3;
14686	if (EMAreHypercallInstructionsEnabled(pVCpu))
14687	{
14688	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14689	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_CR0
14690	\| CPUMCTX_EXTRN_SS \| CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_EFER);
14691	AssertRCReturn(rc, rc);
14692
14693	/* Perform the hypercall. */
14694	rcStrict = GIMHypercall(pVCpu, &pVCpu->cpum.GstCtx);
14695	if (rcStrict == VINF_SUCCESS)
14696	{
14697	rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14698	AssertRCReturn(rc, rc);
14699	}
14700	else
14701	Assert( rcStrict == VINF_GIM_R3_HYPERCALL
14702	\|\| rcStrict == VINF_GIM_HYPERCALL_CONTINUING
14703	\|\| RT_FAILURE(rcStrict));
14704
14705	/* If the hypercall changes anything other than guest's general-purpose registers,
14706	we would need to reload the guest changed bits here before VM-entry. */
14707	}
14708	else
14709	Log4Func(("Hypercalls not enabled\n"));
14710
14711	/* If hypercalls are disabled or the hypercall failed for some reason, raise #UD and continue. */
14712	if (RT_FAILURE(rcStrict))
14713	{
14714	hmR0VmxSetPendingXcptUD(pVCpu);
14715	rcStrict = VINF_SUCCESS;
14716	}
14717
14718	return rcStrict;
14719	}
14720
14721
14722	/**
14723	* VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
14724	*/
14725	HMVMX_EXIT_DECL hmR0VmxExitInvlpg(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14726	{
14727	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14728	Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging \|\| pVCpu->hm.s.fUsingDebugLoop);
14729
14730	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14731	hmR0VmxReadExitQualVmcs(pVmxTransient);
14732	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14733	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
14734	AssertRCReturn(rc, rc);
14735
14736	VBOXSTRICTRC rcStrict = IEMExecDecodedInvlpg(pVCpu, pVmxTransient->cbExitInstr, pVmxTransient->uExitQual);
14737
14738	if (rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_PGM_SYNC_CR3)
14739	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14740	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14741	{
14742	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14743	rcStrict = VINF_SUCCESS;
14744	}
14745	else
14746	AssertMsgFailed(("Unexpected IEMExecDecodedInvlpg(%#RX64) status: %Rrc\n", pVmxTransient->uExitQual,
14747	VBOXSTRICTRC_VAL(rcStrict)));
14748	return rcStrict;
14749	}
14750
14751
14752	/**
14753	* VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
14754	*/
14755	HMVMX_EXIT_DECL hmR0VmxExitMonitor(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14756	{
14757	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14758
14759	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14760	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14761	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK \| CPUMCTX_EXTRN_DS);
14762	AssertRCReturn(rc, rc);
14763
14764	VBOXSTRICTRC rcStrict = IEMExecDecodedMonitor(pVCpu, pVmxTransient->cbExitInstr);
14765	if (rcStrict == VINF_SUCCESS)
14766	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14767	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14768	{
14769	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14770	rcStrict = VINF_SUCCESS;
14771	}
14772
14773	return rcStrict;
14774	}
14775
14776
14777	/**
14778	* VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
14779	*/
14780	HMVMX_EXIT_DECL hmR0VmxExitMwait(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14781	{
14782	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14783
14784	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14785	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14786	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
14787	AssertRCReturn(rc, rc);
14788
14789	VBOXSTRICTRC rcStrict = IEMExecDecodedMwait(pVCpu, pVmxTransient->cbExitInstr);
14790	if (RT_SUCCESS(rcStrict))
14791	{
14792	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14793	if (EMMonitorWaitShouldContinue(pVCpu, &pVCpu->cpum.GstCtx))
14794	rcStrict = VINF_SUCCESS;
14795	}
14796
14797	return rcStrict;
14798	}
14799
14800
14801	/**
14802	* VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT). Unconditional
14803	* VM-exit.
14804	*/
14805	HMVMX_EXIT_DECL hmR0VmxExitTripleFault(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14806	{
14807	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14808	return VINF_EM_RESET;
14809	}
14810
14811
14812	/**
14813	* VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
14814	*/
14815	HMVMX_EXIT_DECL hmR0VmxExitHlt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14816	{
14817	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14818
14819	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14820	AssertRCReturn(rc, rc);
14821
14822	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS); /* Advancing the RIP above should've imported eflags. */
14823	if (EMShouldContinueAfterHalt(pVCpu, &pVCpu->cpum.GstCtx)) /* Requires eflags. */
14824	rc = VINF_SUCCESS;
14825	else
14826	rc = VINF_EM_HALT;
14827
14828	if (rc != VINF_SUCCESS)
14829	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHltToR3);
14830	return rc;
14831	}
14832
14833
14834	/**
14835	* VM-exit handler for instructions that result in a \#UD exception delivered to
14836	* the guest.
14837	*/
14838	HMVMX_EXIT_NSRC_DECL hmR0VmxExitSetPendingXcptUD(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14839	{
14840	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14841	hmR0VmxSetPendingXcptUD(pVCpu);
14842	return VINF_SUCCESS;
14843	}
14844
14845
14846	/**
14847	* VM-exit handler for expiry of the VMX-preemption timer.
14848	*/
14849	HMVMX_EXIT_DECL hmR0VmxExitPreemptTimer(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14850	{
14851	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14852
14853	/* If the VMX-preemption timer has expired, reinitialize the preemption timer on next VM-entry. */
14854	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14855
14856	/* If there are any timer events pending, fall back to ring-3, otherwise resume guest execution. */
14857	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
14858	bool fTimersPending = TMTimerPollBool(pVM, pVCpu);
14859	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPreemptTimer);
14860	return fTimersPending ? VINF_EM_RAW_TIMER_PENDING : VINF_SUCCESS;
14861	}
14862
14863
14864	/**
14865	* VM-exit handler for XSETBV (VMX_EXIT_XSETBV). Unconditional VM-exit.
14866	*/
14867	HMVMX_EXIT_DECL hmR0VmxExitXsetbv(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14868	{
14869	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14870
14871	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14872	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14873	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_CR4);
14874	AssertRCReturn(rc, rc);
14875
14876	VBOXSTRICTRC rcStrict = IEMExecDecodedXsetbv(pVCpu, pVmxTransient->cbExitInstr);
14877	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, rcStrict != VINF_IEM_RAISED_XCPT ? HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
14878	: HM_CHANGED_RAISED_XCPT_MASK);
14879
14880	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14881	pVCpu->hm.s.fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
14882
14883	return rcStrict;
14884	}
14885
14886
14887	/**
14888	* VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
14889	*/
14890	HMVMX_EXIT_DECL hmR0VmxExitInvpcid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14891	{
14892	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14893
14894	/** @todo Enable the new code after finding a reliably guest test-case. */
14895	#if 1
14896	return VERR_EM_INTERPRETER;
14897	#else
14898	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14899	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
14900	hmR0VmxReadExitQualVmcs(pVmxTransient);
14901	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
14902	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
14903	AssertRCReturn(rc, rc);
14904
14905	/* Paranoia. Ensure this has a memory operand. */
14906	Assert(!pVmxTransient->ExitInstrInfo.Inv.u1Cleared0);
14907
14908	uint8_t const iGReg = pVmxTransient->ExitInstrInfo.VmreadVmwrite.iReg2;
14909	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
14910	uint64_t const uType = CPUMIsGuestIn64BitCode(pVCpu) ? pVCpu->cpum.GstCtx.aGRegs[iGReg].u64
14911	: pVCpu->cpum.GstCtx.aGRegs[iGReg].u32;
14912
14913	RTGCPTR GCPtrDesc;
14914	HMVMX_DECODE_MEM_OPERAND(pVCpu, pVmxTransient->ExitInstrInfo.u, pVmxTransient->uExitQual, VMXMEMACCESS_READ, &GCPtrDesc);
14915
14916	VBOXSTRICTRC rcStrict = IEMExecDecodedInvpcid(pVCpu, pVmxTransient->cbExitInstr, pVmxTransient->ExitInstrInfo.Inv.iSegReg,
14917	GCPtrDesc, uType);
14918	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
14919	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14920	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14921	{
14922	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14923	rcStrict = VINF_SUCCESS;
14924	}
14925	return rcStrict;
14926	#endif
14927	}
14928
14929
14930	/**
14931	* VM-exit handler for invalid-guest-state (VMX_EXIT_ERR_INVALID_GUEST_STATE). Error
14932	* VM-exit.
14933	*/
14934	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrInvalidGuestState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14935	{
14936	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14937	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14938	AssertRCReturn(rc, rc);
14939
14940	rc = hmR0VmxCheckCachedVmcsCtls(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest);
14941	if (RT_FAILURE(rc))
14942	return rc;
14943
14944	uint32_t const uInvalidReason = hmR0VmxCheckGuestState(pVCpu, pVmcsInfo);
14945	NOREF(uInvalidReason);
14946
14947	#ifdef VBOX_STRICT
14948	uint32_t fIntrState;
14949	uint64_t u64Val;
14950	hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
14951	hmR0VmxReadEntryXcptErrorCodeVmcs(pVmxTransient);
14952	hmR0VmxReadEntryInstrLenVmcs(pVmxTransient);
14953
14954	Log4(("uInvalidReason %u\n", uInvalidReason));
14955	Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", pVmxTransient->uEntryIntInfo));
14956	Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", pVmxTransient->uEntryXcptErrorCode));
14957	Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %#RX32\n", pVmxTransient->cbEntryInstr));
14958
14959	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState); AssertRC(rc);
14960	Log4(("VMX_VMCS32_GUEST_INT_STATE %#RX32\n", fIntrState));
14961	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64Val); AssertRC(rc);
14962	Log4(("VMX_VMCS_GUEST_CR0 %#RX64\n", u64Val));
14963	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_MASK, &u64Val); AssertRC(rc);
14964	Log4(("VMX_VMCS_CTRL_CR0_MASK %#RX64\n", u64Val));
14965	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, &u64Val); AssertRC(rc);
14966	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RX64\n", u64Val));
14967	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_MASK, &u64Val); AssertRC(rc);
14968	Log4(("VMX_VMCS_CTRL_CR4_MASK %#RX64\n", u64Val));
14969	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, &u64Val); AssertRC(rc);
14970	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RX64\n", u64Val));
14971	if (pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging)
14972	{
14973	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
14974	Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
14975	}
14976	hmR0DumpRegs(pVCpu, HM_DUMP_REG_FLAGS_ALL);
14977	#endif
14978
14979	return VERR_VMX_INVALID_GUEST_STATE;
14980	}
14981
14982	/**
14983	* VM-exit handler for all undefined/unexpected reasons. Should never happen.
14984	*/
14985	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrUnexpected(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14986	{
14987	/*
14988	* Cumulative notes of all recognized but unexpected VM-exits.
14989	*
14990	* 1. This does -not- cover scenarios like a page-fault VM-exit occurring when
14991	* nested-paging is used.
14992	*
14993	* 2. Any instruction that causes a VM-exit unconditionally (for e.g. VMXON) must be
14994	* emulated or a #UD must be raised in the guest. Therefore, we should -not- be using
14995	* this function (and thereby stop VM execution) for handling such instructions.
14996	*
14997	*
14998	* VMX_EXIT_INIT_SIGNAL:
14999	* INIT signals are blocked in VMX root operation by VMXON and by SMI in SMM.
15000	* It is -NOT- blocked in VMX non-root operation so we can, in theory, still get these
15001	* VM-exits. However, we should not receive INIT signals VM-exit while executing a VM.
15002	*
15003	* See Intel spec. 33.14.1 Default Treatment of SMI Delivery"
15004	* See Intel spec. 29.3 "VMX Instructions" for "VMXON".
15005	* See Intel spec. "23.8 Restrictions on VMX operation".
15006	*
15007	* VMX_EXIT_SIPI:
15008	* SIPI exits can only occur in VMX non-root operation when the "wait-for-SIPI" guest
15009	* activity state is used. We don't make use of it as our guests don't have direct
15010	* access to the host local APIC.
15011	*
15012	* See Intel spec. 25.3 "Other Causes of VM-exits".
15013	*
15014	* VMX_EXIT_IO_SMI:
15015	* VMX_EXIT_SMI:
15016	* This can only happen if we support dual-monitor treatment of SMI, which can be
15017	* activated by executing VMCALL in VMX root operation. Only an STM (SMM transfer
15018	* monitor) would get this VM-exit when we (the executive monitor) execute a VMCALL in
15019	* VMX root mode or receive an SMI. If we get here, something funny is going on.
15020	*
15021	* See Intel spec. 33.15.6 "Activating the Dual-Monitor Treatment"
15022	* See Intel spec. 25.3 "Other Causes of VM-Exits"
15023	*
15024	* VMX_EXIT_ERR_MSR_LOAD:
15025	* Failures while loading MSRs are part of the VM-entry MSR-load area are unexpected
15026	* and typically indicates a bug in the hypervisor code. We thus cannot not resume
15027	* execution.
15028	*
15029	* See Intel spec. 26.7 "VM-Entry Failures During Or After Loading Guest State".
15030	*
15031	* VMX_EXIT_ERR_MACHINE_CHECK:
15032	* Machine check exceptions indicates a fatal/unrecoverable hardware condition
15033	* including but not limited to system bus, ECC, parity, cache and TLB errors. A
15034	* #MC exception abort class exception is raised. We thus cannot assume a
15035	* reasonable chance of continuing any sort of execution and we bail.
15036	*
15037	* See Intel spec. 15.1 "Machine-check Architecture".
15038	* See Intel spec. 27.1 "Architectural State Before A VM Exit".
15039	*
15040	* VMX_EXIT_PML_FULL:
15041	* VMX_EXIT_VIRTUALIZED_EOI:
15042	* VMX_EXIT_APIC_WRITE:
15043	* We do not currently support any of these features and thus they are all unexpected
15044	* VM-exits.
15045	*
15046	* VMX_EXIT_GDTR_IDTR_ACCESS:
15047	* VMX_EXIT_LDTR_TR_ACCESS:
15048	* VMX_EXIT_RDRAND:
15049	* VMX_EXIT_RSM:
15050	* VMX_EXIT_VMFUNC:
15051	* VMX_EXIT_ENCLS:
15052	* VMX_EXIT_RDSEED:
15053	* VMX_EXIT_XSAVES:
15054	* VMX_EXIT_XRSTORS:
15055	* VMX_EXIT_UMWAIT:
15056	* VMX_EXIT_TPAUSE:
15057	* These VM-exits are -not- caused unconditionally by execution of the corresponding
15058	* instruction. Any VM-exit for these instructions indicate a hardware problem,
15059	* unsupported CPU modes (like SMM) or potentially corrupt VMCS controls.
15060	*
15061	* See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally".
15062	*/
15063	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15064	AssertMsgFailed(("Unexpected VM-exit %u\n", pVmxTransient->uExitReason));
15065	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
15066	}
15067
15068
15069	/**
15070	* VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
15071	*/
15072	HMVMX_EXIT_DECL hmR0VmxExitRdmsr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15073	{
15074	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15075
15076	/** @todo Optimize this: We currently drag in the whole MSR state
15077	* (CPUMCTX_EXTRN_ALL_MSRS) here. We should optimize this to only get
15078	* MSRs required. That would require changes to IEM and possibly CPUM too.
15079	* (Should probably do it lazy fashion from CPUMAllMsrs.cpp). */
15080	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15081	uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
15082	uint64_t fImport = IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK \| CPUMCTX_EXTRN_ALL_MSRS;
15083	switch (idMsr)
15084	{
15085	case MSR_K8_FS_BASE: fImport \|= CPUMCTX_EXTRN_FS; break;
15086	case MSR_K8_GS_BASE: fImport \|= CPUMCTX_EXTRN_GS; break;
15087	}
15088
15089	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15090	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImport);
15091	AssertRCReturn(rc, rc);
15092
15093	Log4Func(("ecx=%#RX32\n", idMsr));
15094
15095	#ifdef VBOX_STRICT
15096	Assert(!pVmxTransient->fIsNestedGuest);
15097	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
15098	{
15099	if ( hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr)
15100	&& idMsr != MSR_K6_EFER)
15101	{
15102	AssertMsgFailed(("Unexpected RDMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n", idMsr));
15103	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15104	}
15105	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15106	{
15107	Assert(pVmcsInfo->pvMsrBitmap);
15108	uint32_t fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, idMsr);
15109	if (fMsrpm & VMXMSRPM_ALLOW_RD)
15110	{
15111	AssertMsgFailed(("Unexpected RDMSR for a passthru lazy-restore MSR. ecx=%#RX32\n", idMsr));
15112	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15113	}
15114	}
15115	}
15116	#endif
15117
15118	VBOXSTRICTRC rcStrict = IEMExecDecodedRdmsr(pVCpu, pVmxTransient->cbExitInstr);
15119	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr);
15120	if (rcStrict == VINF_SUCCESS)
15121	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15122	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15123	{
15124	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15125	rcStrict = VINF_SUCCESS;
15126	}
15127	else
15128	AssertMsg(rcStrict == VINF_CPUM_R3_MSR_READ, ("Unexpected IEMExecDecodedRdmsr rc (%Rrc)\n", VBOXSTRICTRC_VAL(rcStrict)));
15129
15130	return rcStrict;
15131	}
15132
15133
15134	/**
15135	* VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
15136	*/
15137	HMVMX_EXIT_DECL hmR0VmxExitWrmsr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15138	{
15139	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15140
15141	/** @todo Optimize this: We currently drag in the whole MSR state
15142	* (CPUMCTX_EXTRN_ALL_MSRS) here. We should optimize this to only get
15143	* MSRs required. That would require changes to IEM and possibly CPUM too.
15144	* (Should probably do it lazy fashion from CPUMAllMsrs.cpp). */
15145	uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
15146	uint64_t fImport = IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK \| CPUMCTX_EXTRN_ALL_MSRS;
15147
15148	/*
15149	* The FS and GS base MSRs are not part of the above all-MSRs mask.
15150	* Although we don't need to fetch the base as it will be overwritten shortly, while
15151	* loading guest-state we would also load the entire segment register including limit
15152	* and attributes and thus we need to load them here.
15153	*/
15154	switch (idMsr)
15155	{
15156	case MSR_K8_FS_BASE: fImport \|= CPUMCTX_EXTRN_FS; break;
15157	case MSR_K8_GS_BASE: fImport \|= CPUMCTX_EXTRN_GS; break;
15158	}
15159
15160	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15161	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15162	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImport);
15163	AssertRCReturn(rc, rc);
15164
15165	Log4Func(("ecx=%#RX32 edx:eax=%#RX32:%#RX32\n", idMsr, pVCpu->cpum.GstCtx.edx, pVCpu->cpum.GstCtx.eax));
15166
15167	VBOXSTRICTRC rcStrict = IEMExecDecodedWrmsr(pVCpu, pVmxTransient->cbExitInstr);
15168	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr);
15169
15170	if (rcStrict == VINF_SUCCESS)
15171	{
15172	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15173
15174	/* If this is an X2APIC WRMSR access, update the APIC state as well. */
15175	if ( idMsr == MSR_IA32_APICBASE
15176	\|\| ( idMsr >= MSR_IA32_X2APIC_START
15177	&& idMsr <= MSR_IA32_X2APIC_END))
15178	{
15179	/*
15180	* We've already saved the APIC related guest-state (TPR) in post-run phase.
15181	* When full APIC register virtualization is implemented we'll have to make
15182	* sure APIC state is saved from the VMCS before IEM changes it.
15183	*/
15184	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
15185	}
15186	else if (idMsr == MSR_IA32_TSC) /* Windows 7 does this during bootup. See @bugref{6398}. */
15187	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
15188	else if (idMsr == MSR_K6_EFER)
15189	{
15190	/*
15191	* If the guest touches the EFER MSR we need to update the VM-Entry and VM-Exit controls
15192	* as well, even if it is -not- touching bits that cause paging mode changes (LMA/LME).
15193	* We care about the other bits as well, SCE and NXE. See @bugref{7368}.
15194	*/
15195	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_EFER_MSR \| HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
15196	}
15197
15198	/* Update MSRs that are part of the VMCS and auto-load/store area when MSR-bitmaps are not used. */
15199	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS))
15200	{
15201	switch (idMsr)
15202	{
15203	case MSR_IA32_SYSENTER_CS: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_CS_MSR); break;
15204	case MSR_IA32_SYSENTER_EIP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); break;
15205	case MSR_IA32_SYSENTER_ESP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); break;
15206	case MSR_K8_FS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_FS); break;
15207	case MSR_K8_GS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_GS); break;
15208	case MSR_K6_EFER: /* Nothing to do, already handled above. */ break;
15209	default:
15210	{
15211	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15212	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_VMX_GUEST_LAZY_MSRS);
15213	else if (hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr))
15214	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
15215	break;
15216	}
15217	}
15218	}
15219	#ifdef VBOX_STRICT
15220	else
15221	{
15222	/* Paranoia. Validate that MSRs in the MSR-bitmaps with write-passthru are not intercepted. */
15223	switch (idMsr)
15224	{
15225	case MSR_IA32_SYSENTER_CS:
15226	case MSR_IA32_SYSENTER_EIP:
15227	case MSR_IA32_SYSENTER_ESP:
15228	case MSR_K8_FS_BASE:
15229	case MSR_K8_GS_BASE:
15230	{
15231	AssertMsgFailed(("Unexpected WRMSR for an MSR in the VMCS. ecx=%#RX32\n", idMsr));
15232	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15233	}
15234
15235	/* Writes to MSRs in auto-load/store area/swapped MSRs, shouldn't cause VM-exits with MSR-bitmaps. */
15236	default:
15237	{
15238	if (hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr))
15239	{
15240	/* EFER MSR writes are always intercepted. */
15241	if (idMsr != MSR_K6_EFER)
15242	{
15243	AssertMsgFailed(("Unexpected WRMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n",
15244	idMsr));
15245	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15246	}
15247	}
15248
15249	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15250	{
15251	Assert(pVmcsInfo->pvMsrBitmap);
15252	uint32_t fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, idMsr);
15253	if (fMsrpm & VMXMSRPM_ALLOW_WR)
15254	{
15255	AssertMsgFailed(("Unexpected WRMSR for passthru, lazy-restore MSR. ecx=%#RX32\n", idMsr));
15256	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15257	}
15258	}
15259	break;
15260	}
15261	}
15262	}
15263	#endif /* VBOX_STRICT */
15264	}
15265	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15266	{
15267	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15268	rcStrict = VINF_SUCCESS;
15269	}
15270	else
15271	AssertMsg(rcStrict == VINF_CPUM_R3_MSR_WRITE, ("Unexpected IEMExecDecodedWrmsr rc (%Rrc)\n", VBOXSTRICTRC_VAL(rcStrict)));
15272
15273	return rcStrict;
15274	}
15275
15276
15277	/**
15278	* VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
15279	*/
15280	HMVMX_EXIT_DECL hmR0VmxExitPause(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15281	{
15282	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15283
15284	/** @todo The guest has likely hit a contended spinlock. We might want to
15285	* poke a schedule different guest VCPU. */
15286	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15287	if (RT_SUCCESS(rc))
15288	return VINF_EM_RAW_INTERRUPT;
15289
15290	AssertMsgFailed(("hmR0VmxExitPause: Failed to increment RIP. rc=%Rrc\n", rc));
15291	return rc;
15292	}
15293
15294
15295	/**
15296	* VM-exit handler for when the TPR value is lowered below the specified
15297	* threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
15298	*/
15299	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThreshold(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15300	{
15301	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15302	Assert(pVmxTransient->pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
15303
15304	/*
15305	* The TPR shadow would've been synced with the APIC TPR in the post-run phase.
15306	* We'll re-evaluate pending interrupts and inject them before the next VM
15307	* entry so we can just continue execution here.
15308	*/
15309	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTprBelowThreshold);
15310	return VINF_SUCCESS;
15311	}
15312
15313
15314	/**
15315	* VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX). Conditional
15316	* VM-exit.
15317	*
15318	* @retval VINF_SUCCESS when guest execution can continue.
15319	* @retval VINF_PGM_SYNC_CR3 CR3 sync is required, back to ring-3.
15320	* @retval VERR_EM_RESCHEDULE_REM when we need to return to ring-3 due to
15321	* incompatible guest state for VMX execution (real-on-v86 case).
15322	*/
15323	HMVMX_EXIT_DECL hmR0VmxExitMovCRx(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15324	{
15325	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15326	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitMovCRx, y2);
15327
15328	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15329	hmR0VmxReadExitQualVmcs(pVmxTransient);
15330	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15331
15332	VBOXSTRICTRC rcStrict;
15333	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15334	uint64_t const uExitQual = pVmxTransient->uExitQual;
15335	uint32_t const uAccessType = VMX_EXIT_QUAL_CRX_ACCESS(uExitQual);
15336	switch (uAccessType)
15337	{
15338	/*
15339	* MOV to CRx.
15340	*/
15341	case VMX_EXIT_QUAL_CRX_ACCESS_WRITE:
15342	{
15343	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
15344	AssertRCReturn(rc, rc);
15345
15346	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
15347	uint32_t const uOldCr0 = pVCpu->cpum.GstCtx.cr0;
15348	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(uExitQual);
15349	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(uExitQual);
15350
15351	/*
15352	* MOV to CR3 only cause a VM-exit when one or more of the following are true:
15353	* - When nested paging isn't used.
15354	* - If the guest doesn't have paging enabled (intercept CR3 to update shadow page tables).
15355	* - We are executing in the VM debug loop.
15356	*/
15357	Assert( iCrReg != 3
15358	\|\| !pVM->hm.s.fNestedPaging
15359	\|\| !CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx)
15360	\|\| pVCpu->hm.s.fUsingDebugLoop);
15361
15362	/* MOV to CR8 writes only cause VM-exits when TPR shadow is not used. */
15363	Assert( iCrReg != 8
15364	\|\| !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW));
15365
15366	rcStrict = hmR0VmxExitMovToCrX(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
15367	AssertMsg( rcStrict == VINF_SUCCESS
15368	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15369
15370	/*
15371	* This is a kludge for handling switches back to real mode when we try to use
15372	* V86 mode to run real mode code directly. Problem is that V86 mode cannot
15373	* deal with special selector values, so we have to return to ring-3 and run
15374	* there till the selector values are V86 mode compatible.
15375	*
15376	* Note! Using VINF_EM_RESCHEDULE_REM here rather than VINF_EM_RESCHEDULE since the
15377	* latter is an alias for VINF_IEM_RAISED_XCPT which is asserted at the end of
15378	* this function.
15379	*/
15380	if ( iCrReg == 0
15381	&& rcStrict == VINF_SUCCESS
15382	&& !pVM->hm.s.vmx.fUnrestrictedGuest
15383	&& CPUMIsGuestInRealModeEx(&pVCpu->cpum.GstCtx)
15384	&& (uOldCr0 & X86_CR0_PE)
15385	&& !(pVCpu->cpum.GstCtx.cr0 & X86_CR0_PE))
15386	{
15387	/** @todo Check selectors rather than returning all the time. */
15388	Assert(!pVmxTransient->fIsNestedGuest);
15389	Log4Func(("CR0 write, back to real mode -> VINF_EM_RESCHEDULE_REM\n"));
15390	rcStrict = VINF_EM_RESCHEDULE_REM;
15391	}
15392	break;
15393	}
15394
15395	/*
15396	* MOV from CRx.
15397	*/
15398	case VMX_EXIT_QUAL_CRX_ACCESS_READ:
15399	{
15400	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(uExitQual);
15401	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(uExitQual);
15402
15403	/*
15404	* MOV from CR3 only cause a VM-exit when one or more of the following are true:
15405	* - When nested paging isn't used.
15406	* - If the guest doesn't have paging enabled (pass guest's CR3 rather than our identity mapped CR3).
15407	* - We are executing in the VM debug loop.
15408	*/
15409	Assert( iCrReg != 3
15410	\|\| !pVM->hm.s.fNestedPaging
15411	\|\| !CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx)
15412	\|\| pVCpu->hm.s.fUsingDebugLoop);
15413
15414	/* MOV from CR8 reads only cause a VM-exit when the TPR shadow feature isn't enabled. */
15415	Assert( iCrReg != 8
15416	\|\| !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW));
15417
15418	rcStrict = hmR0VmxExitMovFromCrX(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
15419	break;
15420	}
15421
15422	/*
15423	* CLTS (Clear Task-Switch Flag in CR0).
15424	*/
15425	case VMX_EXIT_QUAL_CRX_ACCESS_CLTS:
15426	{
15427	rcStrict = hmR0VmxExitClts(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr);
15428	break;
15429	}
15430
15431	/*
15432	* LMSW (Load Machine-Status Word into CR0).
15433	* LMSW cannot clear CR0.PE, so no fRealOnV86Active kludge needed here.
15434	*/
15435	case VMX_EXIT_QUAL_CRX_ACCESS_LMSW:
15436	{
15437	RTGCPTR GCPtrEffDst;
15438	uint8_t const cbInstr = pVmxTransient->cbExitInstr;
15439	uint16_t const uMsw = VMX_EXIT_QUAL_CRX_LMSW_DATA(uExitQual);
15440	bool const fMemOperand = VMX_EXIT_QUAL_CRX_LMSW_OP_MEM(uExitQual);
15441	if (fMemOperand)
15442	{
15443	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
15444	GCPtrEffDst = pVmxTransient->uGuestLinearAddr;
15445	}
15446	else
15447	GCPtrEffDst = NIL_RTGCPTR;
15448	rcStrict = hmR0VmxExitLmsw(pVCpu, pVmcsInfo, cbInstr, uMsw, GCPtrEffDst);
15449	break;
15450	}
15451
15452	default:
15453	{
15454	AssertMsgFailed(("Unrecognized Mov CRX access type %#x\n", uAccessType));
15455	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, uAccessType);
15456	}
15457	}
15458
15459	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS))
15460	== (HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS));
15461	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
15462
15463	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitMovCRx, y2);
15464	NOREF(pVM);
15465	return rcStrict;
15466	}
15467
15468
15469	/**
15470	* VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR). Conditional
15471	* VM-exit.
15472	*/
15473	HMVMX_EXIT_DECL hmR0VmxExitIoInstr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15474	{
15475	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15476	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitIO, y1);
15477
15478	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15479	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15480	hmR0VmxReadExitQualVmcs(pVmxTransient);
15481	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15482	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_SREG_MASK
15483	\| CPUMCTX_EXTRN_EFER);
15484	/* EFER MSR also required for longmode checks in EMInterpretDisasCurrent(), but it's always up-to-date. */
15485	AssertRCReturn(rc, rc);
15486
15487	/* Refer Intel spec. 27-5. "Exit Qualifications for I/O Instructions" for the format. */
15488	uint32_t const uIOPort = VMX_EXIT_QUAL_IO_PORT(pVmxTransient->uExitQual);
15489	uint8_t const uIOSize = VMX_EXIT_QUAL_IO_SIZE(pVmxTransient->uExitQual);
15490	bool const fIOWrite = (VMX_EXIT_QUAL_IO_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_IO_DIRECTION_OUT);
15491	bool const fIOString = VMX_EXIT_QUAL_IO_IS_STRING(pVmxTransient->uExitQual);
15492	bool const fGstStepping = RT_BOOL(pCtx->eflags.Bits.u1TF);
15493	bool const fDbgStepping = pVCpu->hm.s.fSingleInstruction;
15494	AssertReturn(uIOSize <= 3 && uIOSize != 2, VERR_VMX_IPE_1);
15495
15496	/*
15497	* Update exit history to see if this exit can be optimized.
15498	*/
15499	VBOXSTRICTRC rcStrict;
15500	PCEMEXITREC pExitRec = NULL;
15501	if ( !fGstStepping
15502	&& !fDbgStepping)
15503	pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
15504	!fIOString
15505	? !fIOWrite
15506	? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_READ)
15507	: EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_WRITE)
15508	: !fIOWrite
15509	? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_READ)
15510	: EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_WRITE),
15511	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
15512	if (!pExitRec)
15513	{
15514	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses in bytes. */
15515	static uint32_t const s_aIOOpAnd[4] = { 0xff, 0xffff, 0, 0xffffffff }; /* AND masks for saving result in AL/AX/EAX. */
15516
15517	uint32_t const cbValue = s_aIOSizes[uIOSize];
15518	uint32_t const cbInstr = pVmxTransient->cbExitInstr;
15519	bool fUpdateRipAlready = false; /* ugly hack, should be temporary. */
15520	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15521	if (fIOString)
15522	{
15523	/*
15524	* INS/OUTS - I/O String instruction.
15525	*
15526	* Use instruction-information if available, otherwise fall back on
15527	* interpreting the instruction.
15528	*/
15529	Log4Func(("cs:rip=%#04x:%#RX64 %#06x/%u %c str\n", pCtx->cs.Sel, pCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
15530	AssertReturn(pCtx->dx == uIOPort, VERR_VMX_IPE_2);
15531	bool const fInsOutsInfo = RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_INS_OUTS);
15532	if (fInsOutsInfo)
15533	{
15534	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15535	AssertReturn(pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize <= 2, VERR_VMX_IPE_3);
15536	AssertCompile(IEMMODE_16BIT == 0 && IEMMODE_32BIT == 1 && IEMMODE_64BIT == 2);
15537	IEMMODE const enmAddrMode = (IEMMODE)pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize;
15538	bool const fRep = VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual);
15539	if (fIOWrite)
15540	rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, fRep, cbInstr,
15541	pVmxTransient->ExitInstrInfo.StrIo.iSegReg, true /fIoChecked/);
15542	else
15543	{
15544	/*
15545	* The segment prefix for INS cannot be overridden and is always ES. We can safely assume X86_SREG_ES.
15546	* Hence "iSegReg" field is undefined in the instruction-information field in VT-x for INS.
15547	* See Intel Instruction spec. for "INS".
15548	* See Intel spec. Table 27-8 "Format of the VM-Exit Instruction-Information Field as Used for INS and OUTS".
15549	*/
15550	rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, fRep, cbInstr, true /fIoChecked/);
15551	}
15552	}
15553	else
15554	rcStrict = IEMExecOne(pVCpu);
15555
15556	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
15557	fUpdateRipAlready = true;
15558	}
15559	else
15560	{
15561	/*
15562	* IN/OUT - I/O instruction.
15563	*/
15564	Log4Func(("cs:rip=%04x:%08RX64 %#06x/%u %c\n", pCtx->cs.Sel, pCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
15565	uint32_t const uAndVal = s_aIOOpAnd[uIOSize];
15566	Assert(!VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual));
15567	if (fIOWrite)
15568	{
15569	rcStrict = IOMIOPortWrite(pVM, pVCpu, uIOPort, pCtx->eax & uAndVal, cbValue);
15570	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite);
15571	if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
15572	&& !pCtx->eflags.Bits.u1TF)
15573	rcStrict = EMRZSetPendingIoPortWrite(pVCpu, uIOPort, cbInstr, cbValue, pCtx->eax & uAndVal);
15574	}
15575	else
15576	{
15577	uint32_t u32Result = 0;
15578	rcStrict = IOMIOPortRead(pVM, pVCpu, uIOPort, &u32Result, cbValue);
15579	if (IOM_SUCCESS(rcStrict))
15580	{
15581	/* Save result of I/O IN instr. in AL/AX/EAX. */
15582	pCtx->eax = (pCtx->eax & ~uAndVal) \| (u32Result & uAndVal);
15583	}
15584	if ( rcStrict == VINF_IOM_R3_IOPORT_READ
15585	&& !pCtx->eflags.Bits.u1TF)
15586	rcStrict = EMRZSetPendingIoPortRead(pVCpu, uIOPort, cbInstr, cbValue);
15587	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead);
15588	}
15589	}
15590
15591	if (IOM_SUCCESS(rcStrict))
15592	{
15593	if (!fUpdateRipAlready)
15594	{
15595	hmR0VmxAdvanceGuestRipBy(pVCpu, cbInstr);
15596	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
15597	}
15598
15599	/*
15600	* INS/OUTS with REP prefix updates RFLAGS, can be observed with triple-fault guru
15601	* while booting Fedora 17 64-bit guest.
15602	*
15603	* See Intel Instruction reference for REP/REPE/REPZ/REPNE/REPNZ.
15604	*/
15605	if (fIOString)
15606	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RFLAGS);
15607
15608	/*
15609	* If any I/O breakpoints are armed, we need to check if one triggered
15610	* and take appropriate action.
15611	* Note that the I/O breakpoint type is undefined if CR4.DE is 0.
15612	*/
15613	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_DR7);
15614	AssertRCReturn(rc, rc);
15615
15616	/** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the
15617	* execution engines about whether hyper BPs and such are pending. */
15618	uint32_t const uDr7 = pCtx->dr[7];
15619	if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK)
15620	&& X86_DR7_ANY_RW_IO(uDr7)
15621	&& (pCtx->cr4 & X86_CR4_DE))
15622	\|\| DBGFBpIsHwIoArmed(pVM)))
15623	{
15624	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck);
15625
15626	/* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
15627	VMMRZCallRing3Disable(pVCpu);
15628	HM_DISABLE_PREEMPT(pVCpu);
15629
15630	bool fIsGuestDbgActive = CPUMR0DebugStateMaybeSaveGuest(pVCpu, true /* fDr6 */);
15631
15632	VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pCtx, uIOPort, cbValue);
15633	if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
15634	{
15635	/* Raise #DB. */
15636	if (fIsGuestDbgActive)
15637	ASMSetDR6(pCtx->dr[6]);
15638	if (pCtx->dr[7] != uDr7)
15639	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_GUEST_DR7;
15640
15641	hmR0VmxSetPendingXcptDB(pVCpu);
15642	}
15643	/* rcStrict is VINF_SUCCESS, VINF_IOM_R3_IOPORT_COMMIT_WRITE, or in [VINF_EM_FIRST..VINF_EM_LAST],
15644	however we can ditch VINF_IOM_R3_IOPORT_COMMIT_WRITE as it has VMCPU_FF_IOM as backup. */
15645	else if ( rcStrict2 != VINF_SUCCESS
15646	&& (rcStrict == VINF_SUCCESS \|\| rcStrict2 < rcStrict))
15647	rcStrict = rcStrict2;
15648	AssertCompile(VINF_EM_LAST < VINF_IOM_R3_IOPORT_COMMIT_WRITE);
15649
15650	HM_RESTORE_PREEMPT();
15651	VMMRZCallRing3Enable(pVCpu);
15652	}
15653	}
15654
15655	#ifdef VBOX_STRICT
15656	if ( rcStrict == VINF_IOM_R3_IOPORT_READ
15657	\|\| rcStrict == VINF_EM_PENDING_R3_IOPORT_READ)
15658	Assert(!fIOWrite);
15659	else if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
15660	\|\| rcStrict == VINF_IOM_R3_IOPORT_COMMIT_WRITE
15661	\|\| rcStrict == VINF_EM_PENDING_R3_IOPORT_WRITE)
15662	Assert(fIOWrite);
15663	else
15664	{
15665	# if 0 /** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST
15666	* statuses, that the VMM device and some others may return. See
15667	* IOM_SUCCESS() for guidance. */
15668	AssertMsg( RT_FAILURE(rcStrict)
15669	\|\| rcStrict == VINF_SUCCESS
15670	\|\| rcStrict == VINF_EM_RAW_EMULATE_INSTR
15671	\|\| rcStrict == VINF_EM_DBG_BREAKPOINT
15672	\|\| rcStrict == VINF_EM_RAW_GUEST_TRAP
15673	\|\| rcStrict == VINF_EM_RAW_TO_R3
15674	\|\| rcStrict == VINF_TRPM_XCPT_DISPATCHED, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15675	# endif
15676	}
15677	#endif
15678	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitIO, y1);
15679	}
15680	else
15681	{
15682	/*
15683	* Frequent exit or something needing probing. Get state and call EMHistoryExec.
15684	*/
15685	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
15686	AssertRCReturn(rc2, rc2);
15687	STAM_COUNTER_INC(!fIOString ? fIOWrite ? &pVCpu->hm.s.StatExitIOWrite : &pVCpu->hm.s.StatExitIORead
15688	: fIOWrite ? &pVCpu->hm.s.StatExitIOStringWrite : &pVCpu->hm.s.StatExitIOStringRead);
15689	Log4(("IOExit/%u: %04x:%08RX64: %s%s%s %#x LB %u -> EMHistoryExec\n",
15690	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
15691	VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual) ? "REP " : "",
15692	fIOWrite ? "OUT" : "IN", fIOString ? "S" : "", uIOPort, uIOSize));
15693
15694	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
15695	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
15696
15697	Log4(("IOExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
15698	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
15699	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
15700	}
15701	return rcStrict;
15702	}
15703
15704
15705	/**
15706	* VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH). Unconditional
15707	* VM-exit.
15708	*/
15709	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitch(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15710	{
15711	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15712
15713	/* Check if this task-switch occurred while delivery an event through the guest IDT. */
15714	hmR0VmxReadExitQualVmcs(pVmxTransient);
15715	if (VMX_EXIT_QUAL_TASK_SWITCH_TYPE(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_TASK_SWITCH_TYPE_IDT)
15716	{
15717	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
15718	if (VMX_IDT_VECTORING_INFO_IS_VALID(pVmxTransient->uIdtVectoringInfo))
15719	{
15720	uint32_t uErrCode;
15721	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(pVmxTransient->uIdtVectoringInfo))
15722	{
15723	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
15724	uErrCode = pVmxTransient->uIdtVectoringErrorCode;
15725	}
15726	else
15727	uErrCode = 0;
15728
15729	RTGCUINTPTR GCPtrFaultAddress;
15730	if (VMX_IDT_VECTORING_INFO_IS_XCPT_PF(pVmxTransient->uIdtVectoringInfo))
15731	GCPtrFaultAddress = pVCpu->cpum.GstCtx.cr2;
15732	else
15733	GCPtrFaultAddress = 0;
15734
15735	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15736
15737	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(pVmxTransient->uIdtVectoringInfo),
15738	pVmxTransient->cbExitInstr, uErrCode, GCPtrFaultAddress);
15739
15740	Log4Func(("Pending event. uIntType=%#x uVector=%#x\n", VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo),
15741	VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo)));
15742	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
15743	return VINF_EM_RAW_INJECT_TRPM_EVENT;
15744	}
15745	}
15746
15747	/* Fall back to the interpreter to emulate the task-switch. */
15748	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
15749	return VERR_EM_INTERPRETER;
15750	}
15751
15752
15753	/**
15754	* VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional VM-exit.
15755	*/
15756	HMVMX_EXIT_DECL hmR0VmxExitMtf(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15757	{
15758	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15759
15760	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15761	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_MONITOR_TRAP_FLAG;
15762	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
15763	AssertRC(rc);
15764	return VINF_EM_DBG_STEPPED;
15765	}
15766
15767
15768	/**
15769	* VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional VM-exit.
15770	*/
15771	HMVMX_EXIT_DECL hmR0VmxExitApicAccess(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15772	{
15773	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15774	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitApicAccess);
15775
15776	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
15777	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
15778	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15779	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
15780	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
15781
15782	/*
15783	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
15784	*/
15785	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
15786	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15787	{
15788	/* For some crazy guest, if an event delivery causes an APIC-access VM-exit, go to instruction emulation. */
15789	if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
15790	{
15791	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
15792	return VINF_EM_RAW_INJECT_TRPM_EVENT;
15793	}
15794	}
15795	else
15796	{
15797	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
15798	return rcStrict;
15799	}
15800
15801	/* IOMMIOPhysHandler() below may call into IEM, save the necessary state. */
15802	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15803	hmR0VmxReadExitQualVmcs(pVmxTransient);
15804	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
15805	AssertRCReturn(rc, rc);
15806
15807	/* See Intel spec. 27-6 "Exit Qualifications for APIC-access VM-exits from Linear Accesses & Guest-Phyiscal Addresses" */
15808	uint32_t const uAccessType = VMX_EXIT_QUAL_APIC_ACCESS_TYPE(pVmxTransient->uExitQual);
15809	switch (uAccessType)
15810	{
15811	case VMX_APIC_ACCESS_TYPE_LINEAR_WRITE:
15812	case VMX_APIC_ACCESS_TYPE_LINEAR_READ:
15813	{
15814	AssertMsg( !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
15815	\|\| VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual) != XAPIC_OFF_TPR,
15816	("hmR0VmxExitApicAccess: can't access TPR offset while using TPR shadowing.\n"));
15817
15818	RTGCPHYS GCPhys = pVCpu->hm.s.vmx.u64GstMsrApicBase; /* Always up-to-date, as it is not part of the VMCS. */
15819	GCPhys &= PAGE_BASE_GC_MASK;
15820	GCPhys += VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual);
15821	Log4Func(("Linear access uAccessType=%#x GCPhys=%#RGp Off=%#x\n", uAccessType, GCPhys,
15822	VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual)));
15823
15824	rcStrict = IOMR0MmioPhysHandler(pVCpu->CTX_SUFF(pVM), pVCpu,
15825	uAccessType == VMX_APIC_ACCESS_TYPE_LINEAR_READ ? 0 : X86_TRAP_PF_RW, GCPhys);
15826	Log4Func(("IOMMMIOPhysHandler returned %Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15827	if ( rcStrict == VINF_SUCCESS
15828	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
15829	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
15830	{
15831	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS
15832	\| HM_CHANGED_GUEST_APIC_TPR);
15833	rcStrict = VINF_SUCCESS;
15834	}
15835	break;
15836	}
15837
15838	default:
15839	{
15840	Log4Func(("uAccessType=%#x\n", uAccessType));
15841	rcStrict = VINF_EM_RAW_EMULATE_INSTR;
15842	break;
15843	}
15844	}
15845
15846	if (rcStrict != VINF_SUCCESS)
15847	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchApicAccessToR3);
15848	return rcStrict;
15849	}
15850
15851
15852	/**
15853	* VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX). Conditional
15854	* VM-exit.
15855	*/
15856	HMVMX_EXIT_DECL hmR0VmxExitMovDRx(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15857	{
15858	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15859	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15860
15861	/* We might get this VM-exit if the nested-guest is not intercepting MOV DRx accesses. */
15862	if (!pVmxTransient->fIsNestedGuest)
15863	{
15864	/* We should -not- get this VM-exit if the guest's debug registers were active. */
15865	if (pVmxTransient->fWasGuestDebugStateActive)
15866	{
15867	AssertMsgFailed(("Unexpected MOV DRx exit\n"));
15868	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
15869	}
15870
15871	if ( !pVCpu->hm.s.fSingleInstruction
15872	&& !pVmxTransient->fWasHyperDebugStateActive)
15873	{
15874	Assert(!DBGFIsStepping(pVCpu));
15875	Assert(pVmcsInfo->u32XcptBitmap & RT_BIT(X86_XCPT_DB));
15876
15877	/* Don't intercept MOV DRx any more. */
15878	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_MOV_DR_EXIT;
15879	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
15880	AssertRC(rc);
15881
15882	/* We're playing with the host CPU state here, make sure we can't preempt or longjmp. */
15883	VMMRZCallRing3Disable(pVCpu);
15884	HM_DISABLE_PREEMPT(pVCpu);
15885
15886	/* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */
15887	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
15888	Assert(CPUMIsGuestDebugStateActive(pVCpu));
15889
15890	HM_RESTORE_PREEMPT();
15891	VMMRZCallRing3Enable(pVCpu);
15892
15893	#ifdef VBOX_WITH_STATISTICS
15894	hmR0VmxReadExitQualVmcs(pVmxTransient);
15895	if (VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_DRX_DIRECTION_WRITE)
15896	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
15897	else
15898	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
15899	#endif
15900	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch);
15901	return VINF_SUCCESS;
15902	}
15903	}
15904
15905	/*
15906	* EMInterpretDRx[Write\|Read]() calls CPUMIsGuestIn64BitCode() which requires EFER MSR, CS.
15907	* The EFER MSR is always up-to-date.
15908	* Update the segment registers and DR7 from the CPU.
15909	*/
15910	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15911	hmR0VmxReadExitQualVmcs(pVmxTransient);
15912	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_DR7);
15913	AssertRCReturn(rc, rc);
15914	Log4Func(("cs:rip=%#04x:%#RX64\n", pCtx->cs.Sel, pCtx->rip));
15915
15916	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15917	if (VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_DRX_DIRECTION_WRITE)
15918	{
15919	rc = EMInterpretDRxWrite(pVM, pVCpu, CPUMCTX2CORE(pCtx),
15920	VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual),
15921	VMX_EXIT_QUAL_DRX_GENREG(pVmxTransient->uExitQual));
15922	if (RT_SUCCESS(rc))
15923	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
15924	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
15925	}
15926	else
15927	{
15928	rc = EMInterpretDRxRead(pVM, pVCpu, CPUMCTX2CORE(pCtx),
15929	VMX_EXIT_QUAL_DRX_GENREG(pVmxTransient->uExitQual),
15930	VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual));
15931	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
15932	}
15933
15934	Assert(rc == VINF_SUCCESS \|\| rc == VERR_EM_INTERPRETER);
15935	if (RT_SUCCESS(rc))
15936	{
15937	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15938	AssertRCReturn(rc2, rc2);
15939	return VINF_SUCCESS;
15940	}
15941	return rc;
15942	}
15943
15944
15945	/**
15946	* VM-exit handler for EPT misconfiguration (VMX_EXIT_EPT_MISCONFIG).
15947	* Conditional VM-exit.
15948	*/
15949	HMVMX_EXIT_DECL hmR0VmxExitEptMisconfig(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15950	{
15951	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15952	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
15953
15954	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
15955	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
15956	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15957	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
15958	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
15959
15960	/*
15961	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
15962	*/
15963	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
15964	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15965	{
15966	/*
15967	* In the unlikely case where delivering an event causes an EPT misconfig (MMIO), go back to
15968	* instruction emulation to inject the original event. Otherwise, injecting the original event
15969	* using hardware-assisted VMX would trigger the same EPT misconfig VM-exit again.
15970	*/
15971	if (!pVCpu->hm.s.Event.fPending)
15972	{ /* likely */ }
15973	else
15974	{
15975	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
15976	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
15977	/** @todo NSTVMX: Think about how this should be handled. */
15978	if (pVmxTransient->fIsNestedGuest)
15979	return VERR_VMX_IPE_3;
15980	#endif
15981	return VINF_EM_RAW_INJECT_TRPM_EVENT;
15982	}
15983	}
15984	else
15985	{
15986	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
15987	return rcStrict;
15988	}
15989
15990	/*
15991	* Get sufficient state and update the exit history entry.
15992	*/
15993	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15994	hmR0VmxReadGuestPhysicalAddrVmcs(pVmxTransient);
15995	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
15996	AssertRCReturn(rc, rc);
15997
15998	RTGCPHYS const GCPhys = pVmxTransient->uGuestPhysicalAddr;
15999	PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
16000	EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_MMIO),
16001	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
16002	if (!pExitRec)
16003	{
16004	/*
16005	* If we succeed, resume guest execution.
16006	* If we fail in interpreting the instruction because we couldn't get the guest physical address
16007	* of the page containing the instruction via the guest's page tables (we would invalidate the guest page
16008	* in the host TLB), resume execution which would cause a guest page fault to let the guest handle this
16009	* weird case. See @bugref{6043}.
16010	*/
16011	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16012	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16013	/** @todo bird: We can probably just go straight to IOM here and assume that
16014	* it's MMIO, then fall back on PGM if that hunch didn't work out so
16015	* well. However, we need to address that aliasing workarounds that
16016	* PGMR0Trap0eHandlerNPMisconfig implements. So, some care is needed.
16017	*
16018	* Might also be interesting to see if we can get this done more or
16019	* less locklessly inside IOM. Need to consider the lookup table
16020	* updating and use a bit more carefully first (or do all updates via
16021	* rendezvous) */
16022	rcStrict = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, PGMMODE_EPT, CPUMCTX2CORE(pCtx), GCPhys, UINT32_MAX);
16023	Log4Func(("At %#RGp RIP=%#RX64 rc=%Rrc\n", GCPhys, pCtx->rip, VBOXSTRICTRC_VAL(rcStrict)));
16024	if ( rcStrict == VINF_SUCCESS
16025	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16026	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16027	{
16028	/* Successfully handled MMIO operation. */
16029	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS
16030	\| HM_CHANGED_GUEST_APIC_TPR);
16031	rcStrict = VINF_SUCCESS;
16032	}
16033	}
16034	else
16035	{
16036	/*
16037	* Frequent exit or something needing probing. Call EMHistoryExec.
16038	*/
16039	Log4(("EptMisscfgExit/%u: %04x:%08RX64: %RGp -> EMHistoryExec\n",
16040	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, GCPhys));
16041
16042	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
16043	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16044
16045	Log4(("EptMisscfgExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
16046	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
16047	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
16048	}
16049	return rcStrict;
16050	}
16051
16052
16053	/**
16054	* VM-exit handler for EPT violation (VMX_EXIT_EPT_VIOLATION). Conditional
16055	* VM-exit.
16056	*/
16057	HMVMX_EXIT_DECL hmR0VmxExitEptViolation(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16058	{
16059	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16060	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
16061
16062	hmR0VmxReadExitQualVmcs(pVmxTransient);
16063	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16064	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16065	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16066	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16067	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16068
16069	/*
16070	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16071	*/
16072	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16073	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16074	{
16075	/*
16076	* If delivery of an event causes an EPT violation (true nested #PF and not MMIO),
16077	* we shall resolve the nested #PF and re-inject the original event.
16078	*/
16079	if (pVCpu->hm.s.Event.fPending)
16080	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflectNPF);
16081	}
16082	else
16083	{
16084	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16085	return rcStrict;
16086	}
16087
16088	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16089	hmR0VmxReadGuestPhysicalAddrVmcs(pVmxTransient);
16090	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16091	AssertRCReturn(rc, rc);
16092
16093	RTGCPHYS const GCPhys = pVmxTransient->uGuestPhysicalAddr;
16094	uint64_t const uExitQual = pVmxTransient->uExitQual;
16095	AssertMsg(((pVmxTransient->uExitQual >> 7) & 3) != 2, ("%#RX64", uExitQual));
16096
16097	RTGCUINT uErrorCode = 0;
16098	if (uExitQual & VMX_EXIT_QUAL_EPT_INSTR_FETCH)
16099	uErrorCode \|= X86_TRAP_PF_ID;
16100	if (uExitQual & VMX_EXIT_QUAL_EPT_DATA_WRITE)
16101	uErrorCode \|= X86_TRAP_PF_RW;
16102	if (uExitQual & VMX_EXIT_QUAL_EPT_ENTRY_PRESENT)
16103	uErrorCode \|= X86_TRAP_PF_P;
16104
16105	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16106	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16107	Log4Func(("at %#RX64 (%#RX64 errcode=%#x) cs:rip=%#04x:%#RX64\n", GCPhys, uExitQual, uErrorCode, pCtx->cs.Sel, pCtx->rip));
16108
16109	/*
16110	* Handle the pagefault trap for the nested shadow table.
16111	*/
16112	TRPMAssertXcptPF(pVCpu, GCPhys, uErrorCode);
16113	rcStrict = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, PGMMODE_EPT, uErrorCode, CPUMCTX2CORE(pCtx), GCPhys);
16114	TRPMResetTrap(pVCpu);
16115
16116	/* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}. */
16117	if ( rcStrict == VINF_SUCCESS
16118	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16119	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16120	{
16121	/* Successfully synced our nested page tables. */
16122	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf);
16123	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS);
16124	return VINF_SUCCESS;
16125	}
16126
16127	Log4Func(("EPT return to ring-3 rcStrict2=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
16128	return rcStrict;
16129	}
16130
16131
16132	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16133	/**
16134	* VM-exit handler for VMCLEAR (VMX_EXIT_VMCLEAR). Unconditional VM-exit.
16135	*/
16136	HMVMX_EXIT_DECL hmR0VmxExitVmclear(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16137	{
16138	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16139
16140	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16141	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16142	hmR0VmxReadExitQualVmcs(pVmxTransient);
16143	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16144	\| CPUMCTX_EXTRN_HWVIRT
16145	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16146	AssertRCReturn(rc, rc);
16147
16148	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16149
16150	VMXVEXITINFO ExitInfo;
16151	RT_ZERO(ExitInfo);
16152	ExitInfo.uReason = pVmxTransient->uExitReason;
16153	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16154	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16155	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16156	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16157
16158	VBOXSTRICTRC rcStrict = IEMExecDecodedVmclear(pVCpu, &ExitInfo);
16159	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16160	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16161	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16162	{
16163	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16164	rcStrict = VINF_SUCCESS;
16165	}
16166	return rcStrict;
16167	}
16168
16169
16170	/**
16171	* VM-exit handler for VMLAUNCH (VMX_EXIT_VMLAUNCH). Unconditional VM-exit.
16172	*/
16173	HMVMX_EXIT_DECL hmR0VmxExitVmlaunch(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16174	{
16175	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16176
16177	/* Import the entire VMCS state for now as we would be switching VMCS on successful VMLAUNCH,
16178	otherwise we could import just IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK. */
16179	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16180	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16181	AssertRCReturn(rc, rc);
16182
16183	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16184
16185	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
16186	VBOXSTRICTRC rcStrict = IEMExecDecodedVmlaunchVmresume(pVCpu, pVmxTransient->cbExitInstr, VMXINSTRID_VMLAUNCH);
16187	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
16188	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16189	{
16190	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16191	if (CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
16192	rcStrict = VINF_VMX_VMLAUNCH_VMRESUME;
16193	}
16194	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
16195	return rcStrict;
16196	}
16197
16198
16199	/**
16200	* VM-exit handler for VMPTRLD (VMX_EXIT_VMPTRLD). Unconditional VM-exit.
16201	*/
16202	HMVMX_EXIT_DECL hmR0VmxExitVmptrld(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16203	{
16204	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16205
16206	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16207	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16208	hmR0VmxReadExitQualVmcs(pVmxTransient);
16209	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16210	\| CPUMCTX_EXTRN_HWVIRT
16211	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16212	AssertRCReturn(rc, rc);
16213
16214	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16215
16216	VMXVEXITINFO ExitInfo;
16217	RT_ZERO(ExitInfo);
16218	ExitInfo.uReason = pVmxTransient->uExitReason;
16219	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16220	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16221	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16222	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16223
16224	VBOXSTRICTRC rcStrict = IEMExecDecodedVmptrld(pVCpu, &ExitInfo);
16225	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16226	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16227	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16228	{
16229	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16230	rcStrict = VINF_SUCCESS;
16231	}
16232	return rcStrict;
16233	}
16234
16235
16236	/**
16237	* VM-exit handler for VMPTRST (VMX_EXIT_VMPTRST). Unconditional VM-exit.
16238	*/
16239	HMVMX_EXIT_DECL hmR0VmxExitVmptrst(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16240	{
16241	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16242
16243	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16244	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16245	hmR0VmxReadExitQualVmcs(pVmxTransient);
16246	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16247	\| CPUMCTX_EXTRN_HWVIRT
16248	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16249	AssertRCReturn(rc, rc);
16250
16251	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16252
16253	VMXVEXITINFO ExitInfo;
16254	RT_ZERO(ExitInfo);
16255	ExitInfo.uReason = pVmxTransient->uExitReason;
16256	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16257	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16258	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16259	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_WRITE, &ExitInfo.GCPtrEffAddr);
16260
16261	VBOXSTRICTRC rcStrict = IEMExecDecodedVmptrst(pVCpu, &ExitInfo);
16262	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16263	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16264	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16265	{
16266	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16267	rcStrict = VINF_SUCCESS;
16268	}
16269	return rcStrict;
16270	}
16271
16272
16273	/**
16274	* VM-exit handler for VMREAD (VMX_EXIT_VMREAD). Conditional VM-exit.
16275	*/
16276	HMVMX_EXIT_DECL hmR0VmxExitVmread(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16277	{
16278	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16279
16280	/*
16281	* Strictly speaking we should not get VMREAD VM-exits for shadow VMCS fields and
16282	* thus might not need to import the shadow VMCS state, it's safer just in case
16283	* code elsewhere dares look at unsynced VMCS fields.
16284	*/
16285	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16286	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16287	hmR0VmxReadExitQualVmcs(pVmxTransient);
16288	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16289	\| CPUMCTX_EXTRN_HWVIRT
16290	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16291	AssertRCReturn(rc, rc);
16292
16293	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16294
16295	VMXVEXITINFO ExitInfo;
16296	RT_ZERO(ExitInfo);
16297	ExitInfo.uReason = pVmxTransient->uExitReason;
16298	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16299	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16300	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16301	if (!ExitInfo.InstrInfo.VmreadVmwrite.fIsRegOperand)
16302	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_WRITE, &ExitInfo.GCPtrEffAddr);
16303
16304	VBOXSTRICTRC rcStrict = IEMExecDecodedVmread(pVCpu, &ExitInfo);
16305	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16306	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16307	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16308	{
16309	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16310	rcStrict = VINF_SUCCESS;
16311	}
16312	return rcStrict;
16313	}
16314
16315
16316	/**
16317	* VM-exit handler for VMRESUME (VMX_EXIT_VMRESUME). Unconditional VM-exit.
16318	*/
16319	HMVMX_EXIT_DECL hmR0VmxExitVmresume(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16320	{
16321	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16322
16323	/* Import the entire VMCS state for now as we would be switching VMCS on successful VMRESUME,
16324	otherwise we could import just IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK. */
16325	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16326	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16327	AssertRCReturn(rc, rc);
16328
16329	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16330
16331	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
16332	VBOXSTRICTRC rcStrict = IEMExecDecodedVmlaunchVmresume(pVCpu, pVmxTransient->cbExitInstr, VMXINSTRID_VMRESUME);
16333	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
16334	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16335	{
16336	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16337	if (CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
16338	rcStrict = VINF_VMX_VMLAUNCH_VMRESUME;
16339	}
16340	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
16341	return rcStrict;
16342	}
16343
16344
16345	/**
16346	* VM-exit handler for VMWRITE (VMX_EXIT_VMWRITE). Conditional VM-exit.
16347	*/
16348	HMVMX_EXIT_DECL hmR0VmxExitVmwrite(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16349	{
16350	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16351
16352	/*
16353	* Although we should not get VMWRITE VM-exits for shadow VMCS fields, since our HM hook
16354	* gets invoked when IEM's VMWRITE instruction emulation modifies the current VMCS and it
16355	* flags re-loading the entire shadow VMCS, we should save the entire shadow VMCS here.
16356	*/
16357	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16358	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16359	hmR0VmxReadExitQualVmcs(pVmxTransient);
16360	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16361	\| CPUMCTX_EXTRN_HWVIRT
16362	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16363	AssertRCReturn(rc, rc);
16364
16365	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16366
16367	VMXVEXITINFO ExitInfo;
16368	RT_ZERO(ExitInfo);
16369	ExitInfo.uReason = pVmxTransient->uExitReason;
16370	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16371	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16372	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16373	if (!ExitInfo.InstrInfo.VmreadVmwrite.fIsRegOperand)
16374	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16375
16376	VBOXSTRICTRC rcStrict = IEMExecDecodedVmwrite(pVCpu, &ExitInfo);
16377	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16378	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16379	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16380	{
16381	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16382	rcStrict = VINF_SUCCESS;
16383	}
16384	return rcStrict;
16385	}
16386
16387
16388	/**
16389	* VM-exit handler for VMXOFF (VMX_EXIT_VMXOFF). Unconditional VM-exit.
16390	*/
16391	HMVMX_EXIT_DECL hmR0VmxExitVmxoff(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16392	{
16393	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16394
16395	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16396	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR4
16397	\| CPUMCTX_EXTRN_HWVIRT
16398	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
16399	AssertRCReturn(rc, rc);
16400
16401	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16402
16403	VBOXSTRICTRC rcStrict = IEMExecDecodedVmxoff(pVCpu, pVmxTransient->cbExitInstr);
16404	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16405	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_HWVIRT);
16406	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16407	{
16408	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16409	rcStrict = VINF_SUCCESS;
16410	}
16411	return rcStrict;
16412	}
16413
16414
16415	/**
16416	* VM-exit handler for VMXON (VMX_EXIT_VMXON). Unconditional VM-exit.
16417	*/
16418	HMVMX_EXIT_DECL hmR0VmxExitVmxon(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16419	{
16420	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16421
16422	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16423	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16424	hmR0VmxReadExitQualVmcs(pVmxTransient);
16425	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16426	\| CPUMCTX_EXTRN_HWVIRT
16427	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16428	AssertRCReturn(rc, rc);
16429
16430	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16431
16432	VMXVEXITINFO ExitInfo;
16433	RT_ZERO(ExitInfo);
16434	ExitInfo.uReason = pVmxTransient->uExitReason;
16435	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16436	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16437	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16438	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16439
16440	VBOXSTRICTRC rcStrict = IEMExecDecodedVmxon(pVCpu, &ExitInfo);
16441	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16442	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16443	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16444	{
16445	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16446	rcStrict = VINF_SUCCESS;
16447	}
16448	return rcStrict;
16449	}
16450
16451
16452	/**
16453	* VM-exit handler for INVVPID (VMX_EXIT_INVVPID). Unconditional VM-exit.
16454	*/
16455	HMVMX_EXIT_DECL hmR0VmxExitInvvpid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16456	{
16457	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16458
16459	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16460	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16461	hmR0VmxReadExitQualVmcs(pVmxTransient);
16462	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16463	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16464	AssertRCReturn(rc, rc);
16465
16466	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16467
16468	VMXVEXITINFO ExitInfo;
16469	RT_ZERO(ExitInfo);
16470	ExitInfo.uReason = pVmxTransient->uExitReason;
16471	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16472	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16473	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16474	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16475
16476	VBOXSTRICTRC rcStrict = IEMExecDecodedInvvpid(pVCpu, &ExitInfo);
16477	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16478	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16479	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16480	{
16481	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16482	rcStrict = VINF_SUCCESS;
16483	}
16484	return rcStrict;
16485	}
16486	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
16487	/** @} */
16488
16489
16490	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16491	/** @name Nested-guest VM-exit handlers.
16492	* @{
16493	*/
16494	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16495	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Nested-guest VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16496	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16497
16498	/**
16499	* Nested-guest VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI).
16500	* Conditional VM-exit.
16501	*/
16502	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmiNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16503	{
16504	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16505
16506	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16507
16508	uint64_t const uExitIntInfo = pVmxTransient->uExitIntInfo;
16509	uint32_t const uExitIntType = VMX_EXIT_INT_INFO_TYPE(uExitIntInfo);
16510	Assert(VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo));
16511
16512	switch (uExitIntType)
16513	{
16514	/*
16515	* Physical NMIs:
16516	* We shouldn't direct host physical NMIs to the nested-guest. Dispatch it to the host.
16517	*/
16518	case VMX_EXIT_INT_INFO_TYPE_NMI:
16519	return hmR0VmxExitHostNmi(pVCpu, pVmxTransient->pVmcsInfo);
16520
16521	/*
16522	* Hardware exceptions,
16523	* Software exceptions,
16524	* Privileged software exceptions:
16525	* Figure out if the exception must be delivered to the guest or the nested-guest.
16526	*/
16527	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
16528	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
16529	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
16530	{
16531	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16532	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16533	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16534	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16535
16536	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16537	bool const fIntercept = CPUMIsGuestVmxXcptInterceptSet(pCtx, VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo),
16538	pVmxTransient->uExitIntErrorCode);
16539	if (fIntercept)
16540	{
16541	/* Exit qualification is required for debug and page-fault exceptions. */
16542	hmR0VmxReadExitQualVmcs(pVmxTransient);
16543
16544	/*
16545	* For VM-exits due to software exceptions (those generated by INT3 or INTO) and privileged
16546	* software exceptions (those generated by INT1/ICEBP) we need to supply the VM-exit instruction
16547	* length. However, if delivery of a software interrupt, software exception or privileged
16548	* software exception causes a VM-exit, that too provides the VM-exit instruction length.
16549	*/
16550	VMXVEXITINFO ExitInfo;
16551	RT_ZERO(ExitInfo);
16552	ExitInfo.uReason = pVmxTransient->uExitReason;
16553	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16554	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16555
16556	VMXVEXITEVENTINFO ExitEventInfo;
16557	RT_ZERO(ExitEventInfo);
16558	ExitEventInfo.uExitIntInfo = pVmxTransient->uExitIntInfo;
16559	ExitEventInfo.uExitIntErrCode = pVmxTransient->uExitIntErrorCode;
16560	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
16561	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
16562
16563	#ifdef DEBUG_ramshankar
16564	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16565	Log4Func(("exit_int_info=%#RX32 err_code=%#RX32 exit_qual=%#RX64\n", pVmxTransient->uExitIntInfo,
16566	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual));
16567	if (VMX_IDT_VECTORING_INFO_IS_VALID(pVmxTransient->uIdtVectoringInfo))
16568	{
16569	Log4Func(("idt_info=%#RX32 idt_errcode=%#RX32 cr2=%#RX64\n", pVmxTransient->uIdtVectoringInfo,
16570	pVmxTransient->uIdtVectoringErrorCode, pCtx->cr2));
16571	}
16572	#endif
16573	return IEMExecVmxVmexitXcpt(pVCpu, &ExitInfo, &ExitEventInfo);
16574	}
16575
16576	/* Nested paging is currently a requirement, otherwise we would need to handle shadow #PFs in hmR0VmxExitXcptPF. */
16577	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
16578	return hmR0VmxExitXcpt(pVCpu, pVmxTransient);
16579	}
16580
16581	/*
16582	* Software interrupts:
16583	* VM-exits cannot be caused by software interrupts.
16584	*
16585	* External interrupts:
16586	* This should only happen when "acknowledge external interrupts on VM-exit"
16587	* control is set. However, we never set this when executing a guest or
16588	* nested-guest. For nested-guests it is emulated while injecting interrupts into
16589	* the guest.
16590	*/
16591	case VMX_EXIT_INT_INFO_TYPE_SW_INT:
16592	case VMX_EXIT_INT_INFO_TYPE_EXT_INT:
16593	default:
16594	{
16595	pVCpu->hm.s.u32HMError = pVmxTransient->uExitIntInfo;
16596	return VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
16597	}
16598	}
16599	}
16600
16601
16602	/**
16603	* Nested-guest VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT).
16604	* Unconditional VM-exit.
16605	*/
16606	HMVMX_EXIT_DECL hmR0VmxExitTripleFaultNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16607	{
16608	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16609	return IEMExecVmxVmexitTripleFault(pVCpu);
16610	}
16611
16612
16613	/**
16614	* Nested-guest VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
16615	*/
16616	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindowNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16617	{
16618	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16619
16620	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INT_WINDOW_EXIT))
16621	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
16622	return hmR0VmxExitIntWindow(pVCpu, pVmxTransient);
16623	}
16624
16625
16626	/**
16627	* Nested-guest VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
16628	*/
16629	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindowNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16630	{
16631	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16632
16633	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_NMI_WINDOW_EXIT))
16634	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
16635	return hmR0VmxExitIntWindow(pVCpu, pVmxTransient);
16636	}
16637
16638
16639	/**
16640	* Nested-guest VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH).
16641	* Unconditional VM-exit.
16642	*/
16643	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitchNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16644	{
16645	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16646
16647	hmR0VmxReadExitQualVmcs(pVmxTransient);
16648	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16649	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16650	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16651
16652	VMXVEXITINFO ExitInfo;
16653	RT_ZERO(ExitInfo);
16654	ExitInfo.uReason = pVmxTransient->uExitReason;
16655	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16656	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16657
16658	VMXVEXITEVENTINFO ExitEventInfo;
16659	RT_ZERO(ExitEventInfo);
16660	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
16661	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
16662	return IEMExecVmxVmexitTaskSwitch(pVCpu, &ExitInfo, &ExitEventInfo);
16663	}
16664
16665
16666	/**
16667	* Nested-guest VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
16668	*/
16669	HMVMX_EXIT_DECL hmR0VmxExitHltNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16670	{
16671	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16672
16673	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_HLT_EXIT))
16674	{
16675	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16676	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16677	}
16678	return hmR0VmxExitHlt(pVCpu, pVmxTransient);
16679	}
16680
16681
16682	/**
16683	* Nested-guest VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
16684	*/
16685	HMVMX_EXIT_DECL hmR0VmxExitInvlpgNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16686	{
16687	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16688
16689	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INVLPG_EXIT))
16690	{
16691	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16692	hmR0VmxReadExitQualVmcs(pVmxTransient);
16693
16694	VMXVEXITINFO ExitInfo;
16695	RT_ZERO(ExitInfo);
16696	ExitInfo.uReason = pVmxTransient->uExitReason;
16697	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16698	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16699	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16700	}
16701	return hmR0VmxExitInvlpg(pVCpu, pVmxTransient);
16702	}
16703
16704
16705	/**
16706	* Nested-guest VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
16707	*/
16708	HMVMX_EXIT_DECL hmR0VmxExitRdpmcNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16709	{
16710	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16711
16712	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDPMC_EXIT))
16713	{
16714	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16715	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16716	}
16717	return hmR0VmxExitRdpmc(pVCpu, pVmxTransient);
16718	}
16719
16720
16721	/**
16722	* Nested-guest VM-exit handler for VMREAD (VMX_EXIT_VMREAD) and VMWRITE
16723	* (VMX_EXIT_VMWRITE). Conditional VM-exit.
16724	*/
16725	HMVMX_EXIT_DECL hmR0VmxExitVmreadVmwriteNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16726	{
16727	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16728
16729	Assert( pVmxTransient->uExitReason == VMX_EXIT_VMREAD
16730	\|\| pVmxTransient->uExitReason == VMX_EXIT_VMWRITE);
16731
16732	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16733
16734	uint8_t const iGReg = pVmxTransient->ExitInstrInfo.VmreadVmwrite.iReg2;
16735	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
16736	uint64_t u64VmcsField = pVCpu->cpum.GstCtx.aGRegs[iGReg].u64;
16737
16738	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
16739	if (!CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
16740	u64VmcsField &= UINT64_C(0xffffffff);
16741
16742	if (CPUMIsGuestVmxVmreadVmwriteInterceptSet(pVCpu, pVmxTransient->uExitReason, u64VmcsField))
16743	{
16744	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16745	hmR0VmxReadExitQualVmcs(pVmxTransient);
16746
16747	VMXVEXITINFO ExitInfo;
16748	RT_ZERO(ExitInfo);
16749	ExitInfo.uReason = pVmxTransient->uExitReason;
16750	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16751	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16752	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
16753	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16754	}
16755
16756	if (pVmxTransient->uExitReason == VMX_EXIT_VMREAD)
16757	return hmR0VmxExitVmread(pVCpu, pVmxTransient);
16758	return hmR0VmxExitVmwrite(pVCpu, pVmxTransient);
16759	}
16760
16761
16762	/**
16763	* Nested-guest VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
16764	*/
16765	HMVMX_EXIT_DECL hmR0VmxExitRdtscNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16766	{
16767	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16768
16769	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDTSC_EXIT))
16770	{
16771	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16772	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16773	}
16774
16775	return hmR0VmxExitRdtsc(pVCpu, pVmxTransient);
16776	}
16777
16778
16779	/**
16780	* Nested-guest VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX).
16781	* Conditional VM-exit.
16782	*/
16783	HMVMX_EXIT_DECL hmR0VmxExitMovCRxNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16784	{
16785	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16786
16787	hmR0VmxReadExitQualVmcs(pVmxTransient);
16788	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16789
16790	VBOXSTRICTRC rcStrict;
16791	uint32_t const uAccessType = VMX_EXIT_QUAL_CRX_ACCESS(pVmxTransient->uExitQual);
16792	switch (uAccessType)
16793	{
16794	case VMX_EXIT_QUAL_CRX_ACCESS_WRITE:
16795	{
16796	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
16797	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(pVmxTransient->uExitQual);
16798	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
16799	uint64_t const uNewCrX = pVCpu->cpum.GstCtx.aGRegs[iGReg].u64;
16800
16801	bool fIntercept;
16802	switch (iCrReg)
16803	{
16804	case 0:
16805	case 4:
16806	fIntercept = CPUMIsGuestVmxMovToCr0Cr4InterceptSet(&pVCpu->cpum.GstCtx, iCrReg, uNewCrX);
16807	break;
16808
16809	case 3:
16810	fIntercept = CPUMIsGuestVmxMovToCr3InterceptSet(pVCpu, uNewCrX);
16811	break;
16812
16813	case 8:
16814	fIntercept = CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_CR8_LOAD_EXIT);
16815	break;
16816
16817	default:
16818	fIntercept = false;
16819	break;
16820	}
16821	if (fIntercept)
16822	{
16823	VMXVEXITINFO ExitInfo;
16824	RT_ZERO(ExitInfo);
16825	ExitInfo.uReason = pVmxTransient->uExitReason;
16826	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16827	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16828	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16829	}
16830	else
16831	rcStrict = hmR0VmxExitMovToCrX(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
16832	break;
16833	}
16834
16835	case VMX_EXIT_QUAL_CRX_ACCESS_READ:
16836	{
16837	/*
16838	* CR0/CR4 reads do not cause VM-exits, the read-shadow is used (subject to masking).
16839	* CR2 reads do not cause a VM-exit.
16840	* CR3 reads cause a VM-exit depending on the "CR3 store exiting" control.
16841	* CR8 reads cause a VM-exit depending on the "CR8 store exiting" control.
16842	*/
16843	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
16844	if ( iCrReg == 3
16845	\|\| iCrReg == 8)
16846	{
16847	static const uint32_t s_auCrXReadIntercepts[] = { 0, 0, 0, VMX_PROC_CTLS_CR3_STORE_EXIT, 0,
16848	0, 0, 0, VMX_PROC_CTLS_CR8_STORE_EXIT };
16849	uint32_t const uIntercept = s_auCrXReadIntercepts[iCrReg];
16850	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, uIntercept))
16851	{
16852	VMXVEXITINFO ExitInfo;
16853	RT_ZERO(ExitInfo);
16854	ExitInfo.uReason = pVmxTransient->uExitReason;
16855	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16856	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16857	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16858	}
16859	else
16860	{
16861	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(pVmxTransient->uExitQual);
16862	rcStrict = hmR0VmxExitMovFromCrX(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
16863	}
16864	}
16865	else
16866	{
16867	AssertMsgFailed(("MOV from CR%d VM-exit must not happen\n", iCrReg));
16868	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, iCrReg);
16869	}
16870	break;
16871	}
16872
16873	case VMX_EXIT_QUAL_CRX_ACCESS_CLTS:
16874	{
16875	PCVMXVVMCS pVmcsNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
16876	Assert(pVmcsNstGst);
16877	uint64_t const uGstHostMask = pVmcsNstGst->u64Cr0Mask.u;
16878	uint64_t const uReadShadow = pVmcsNstGst->u64Cr0ReadShadow.u;
16879	if ( (uGstHostMask & X86_CR0_TS)
16880	&& (uReadShadow & X86_CR0_TS))
16881	{
16882	VMXVEXITINFO ExitInfo;
16883	RT_ZERO(ExitInfo);
16884	ExitInfo.uReason = pVmxTransient->uExitReason;
16885	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16886	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16887	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16888	}
16889	else
16890	rcStrict = hmR0VmxExitClts(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr);
16891	break;
16892	}
16893
16894	case VMX_EXIT_QUAL_CRX_ACCESS_LMSW: /* LMSW (Load Machine-Status Word into CR0) */
16895	{
16896	RTGCPTR GCPtrEffDst;
16897	uint16_t const uNewMsw = VMX_EXIT_QUAL_CRX_LMSW_DATA(pVmxTransient->uExitQual);
16898	bool const fMemOperand = VMX_EXIT_QUAL_CRX_LMSW_OP_MEM(pVmxTransient->uExitQual);
16899	if (fMemOperand)
16900	{
16901	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
16902	GCPtrEffDst = pVmxTransient->uGuestLinearAddr;
16903	}
16904	else
16905	GCPtrEffDst = NIL_RTGCPTR;
16906
16907	if (CPUMIsGuestVmxLmswInterceptSet(&pVCpu->cpum.GstCtx, uNewMsw))
16908	{
16909	VMXVEXITINFO ExitInfo;
16910	RT_ZERO(ExitInfo);
16911	ExitInfo.uReason = pVmxTransient->uExitReason;
16912	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16913	ExitInfo.u64GuestLinearAddr = GCPtrEffDst;
16914	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16915	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16916	}
16917	else
16918	rcStrict = hmR0VmxExitLmsw(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, uNewMsw, GCPtrEffDst);
16919	break;
16920	}
16921
16922	default:
16923	{
16924	AssertMsgFailed(("Unrecognized Mov CRX access type %#x\n", uAccessType));
16925	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, uAccessType);
16926	}
16927	}
16928
16929	if (rcStrict == VINF_IEM_RAISED_XCPT)
16930	{
16931	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16932	rcStrict = VINF_SUCCESS;
16933	}
16934	return rcStrict;
16935	}
16936
16937
16938	/**
16939	* Nested-guest VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX).
16940	* Conditional VM-exit.
16941	*/
16942	HMVMX_EXIT_DECL hmR0VmxExitMovDRxNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16943	{
16944	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16945
16946	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MOV_DR_EXIT))
16947	{
16948	hmR0VmxReadExitQualVmcs(pVmxTransient);
16949	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16950
16951	VMXVEXITINFO ExitInfo;
16952	RT_ZERO(ExitInfo);
16953	ExitInfo.uReason = pVmxTransient->uExitReason;
16954	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16955	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16956	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
16957	}
16958	return hmR0VmxExitMovDRx(pVCpu, pVmxTransient);
16959	}
16960
16961
16962	/**
16963	* Nested-guest VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR).
16964	* Conditional VM-exit.
16965	*/
16966	HMVMX_EXIT_DECL hmR0VmxExitIoInstrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16967	{
16968	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16969
16970	hmR0VmxReadExitQualVmcs(pVmxTransient);
16971
16972	uint32_t const uIOPort = VMX_EXIT_QUAL_IO_PORT(pVmxTransient->uExitQual);
16973	uint8_t const uIOSize = VMX_EXIT_QUAL_IO_SIZE(pVmxTransient->uExitQual);
16974	AssertReturn(uIOSize <= 3 && uIOSize != 2, VERR_VMX_IPE_1);
16975
16976	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses in bytes. */
16977	uint8_t const cbAccess = s_aIOSizes[uIOSize];
16978	if (CPUMIsGuestVmxIoInterceptSet(pVCpu, uIOPort, cbAccess))
16979	{
16980	/*
16981	* IN/OUT instruction:
16982	* - Provides VM-exit instruction length.
16983	*
16984	* INS/OUTS instruction:
16985	* - Provides VM-exit instruction length.
16986	* - Provides Guest-linear address.
16987	* - Optionally provides VM-exit instruction info (depends on CPU feature).
16988	*/
16989	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16990	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16991
16992	/* Make sure we don't use stale/uninitialized VMX-transient info. below. */
16993	pVmxTransient->ExitInstrInfo.u = 0;
16994	pVmxTransient->uGuestLinearAddr = 0;
16995
16996	bool const fVmxInsOutsInfo = pVM->cpum.ro.GuestFeatures.fVmxInsOutInfo;
16997	bool const fIOString = VMX_EXIT_QUAL_IO_IS_STRING(pVmxTransient->uExitQual);
16998	if (fIOString)
16999	{
17000	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
17001	if (fVmxInsOutsInfo)
17002	{
17003	Assert(RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_VMCS_INS_OUTS)); /* Paranoia. */
17004	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17005	}
17006	}
17007
17008	VMXVEXITINFO ExitInfo;
17009	RT_ZERO(ExitInfo);
17010	ExitInfo.uReason = pVmxTransient->uExitReason;
17011	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17012	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17013	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17014	ExitInfo.u64GuestLinearAddr = pVmxTransient->uGuestLinearAddr;
17015	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17016	}
17017	return hmR0VmxExitIoInstr(pVCpu, pVmxTransient);
17018	}
17019
17020
17021	/**
17022	* Nested-guest VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
17023	*/
17024	HMVMX_EXIT_DECL hmR0VmxExitRdmsrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17025	{
17026	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17027
17028	uint32_t fMsrpm;
17029	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_MSR_BITMAPS))
17030	fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), pVCpu->cpum.GstCtx.ecx);
17031	else
17032	fMsrpm = VMXMSRPM_EXIT_RD;
17033
17034	if (fMsrpm & VMXMSRPM_EXIT_RD)
17035	{
17036	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17037	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17038	}
17039	return hmR0VmxExitRdmsr(pVCpu, pVmxTransient);
17040	}
17041
17042
17043	/**
17044	* Nested-guest VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
17045	*/
17046	HMVMX_EXIT_DECL hmR0VmxExitWrmsrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17047	{
17048	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17049
17050	uint32_t fMsrpm;
17051	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_MSR_BITMAPS))
17052	fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), pVCpu->cpum.GstCtx.ecx);
17053	else
17054	fMsrpm = VMXMSRPM_EXIT_WR;
17055
17056	if (fMsrpm & VMXMSRPM_EXIT_WR)
17057	{
17058	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17059	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17060	}
17061	return hmR0VmxExitWrmsr(pVCpu, pVmxTransient);
17062	}
17063
17064
17065	/**
17066	* Nested-guest VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
17067	*/
17068	HMVMX_EXIT_DECL hmR0VmxExitMwaitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17069	{
17070	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17071
17072	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MWAIT_EXIT))
17073	{
17074	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17075	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17076	}
17077	return hmR0VmxExitMwait(pVCpu, pVmxTransient);
17078	}
17079
17080
17081	/**
17082	* Nested-guest VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional
17083	* VM-exit.
17084	*/
17085	HMVMX_EXIT_DECL hmR0VmxExitMtfNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17086	{
17087	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17088
17089	/** @todo NSTVMX: Should consider debugging nested-guests using VM debugger. */
17090	hmR0VmxReadGuestPendingDbgXctps(pVmxTransient);
17091	VMXVEXITINFO ExitInfo;
17092	RT_ZERO(ExitInfo);
17093	ExitInfo.uReason = pVmxTransient->uExitReason;
17094	ExitInfo.u64GuestPendingDbgXcpts = pVmxTransient->uGuestPendingDbgXcpts;
17095	return IEMExecVmxVmexitTrapLike(pVCpu, &ExitInfo);
17096	}
17097
17098
17099	/**
17100	* Nested-guest VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
17101	*/
17102	HMVMX_EXIT_DECL hmR0VmxExitMonitorNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17103	{
17104	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17105
17106	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MONITOR_EXIT))
17107	{
17108	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17109	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17110	}
17111	return hmR0VmxExitMonitor(pVCpu, pVmxTransient);
17112	}
17113
17114
17115	/**
17116	* Nested-guest VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
17117	*/
17118	HMVMX_EXIT_DECL hmR0VmxExitPauseNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17119	{
17120	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17121
17122	/** @todo NSTVMX: Think about this more. Does the outer guest need to intercept
17123	* PAUSE when executing a nested-guest? If it does not, we would not need
17124	* to check for the intercepts here. Just call VM-exit... */
17125
17126	/* The CPU would have already performed the necessary CPL checks for PAUSE-loop exiting. */
17127	if ( CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_PAUSE_EXIT)
17128	\|\| CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_PAUSE_LOOP_EXIT))
17129	{
17130	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17131	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17132	}
17133	return hmR0VmxExitPause(pVCpu, pVmxTransient);
17134	}
17135
17136
17137	/**
17138	* Nested-guest VM-exit handler for when the TPR value is lowered below the
17139	* specified threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
17140	*/
17141	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThresholdNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17142	{
17143	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17144
17145	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_TPR_SHADOW))
17146	{
17147	hmR0VmxReadGuestPendingDbgXctps(pVmxTransient);
17148	VMXVEXITINFO ExitInfo;
17149	RT_ZERO(ExitInfo);
17150	ExitInfo.uReason = pVmxTransient->uExitReason;
17151	ExitInfo.u64GuestPendingDbgXcpts = pVmxTransient->uGuestPendingDbgXcpts;
17152	return IEMExecVmxVmexitTrapLike(pVCpu, &ExitInfo);
17153	}
17154	return hmR0VmxExitTprBelowThreshold(pVCpu, pVmxTransient);
17155	}
17156
17157
17158	/**
17159	* Nested-guest VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional
17160	* VM-exit.
17161	*/
17162	HMVMX_EXIT_DECL hmR0VmxExitApicAccessNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17163	{
17164	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17165
17166	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17167	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
17168	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
17169	hmR0VmxReadExitQualVmcs(pVmxTransient);
17170
17171	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_VIRT_APIC_ACCESS));
17172
17173	Log4Func(("at offset %#x type=%u\n", VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual),
17174	VMX_EXIT_QUAL_APIC_ACCESS_TYPE(pVmxTransient->uExitQual)));
17175
17176	VMXVEXITINFO ExitInfo;
17177	RT_ZERO(ExitInfo);
17178	ExitInfo.uReason = pVmxTransient->uExitReason;
17179	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17180	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17181
17182	VMXVEXITEVENTINFO ExitEventInfo;
17183	RT_ZERO(ExitEventInfo);
17184	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
17185	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
17186	return IEMExecVmxVmexitApicAccess(pVCpu, &ExitInfo, &ExitEventInfo);
17187	}
17188
17189
17190	/**
17191	* Nested-guest VM-exit handler for APIC write emulation (VMX_EXIT_APIC_WRITE).
17192	* Conditional VM-exit.
17193	*/
17194	HMVMX_EXIT_DECL hmR0VmxExitApicWriteNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17195	{
17196	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17197
17198	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_APIC_REG_VIRT));
17199	hmR0VmxReadExitQualVmcs(pVmxTransient);
17200	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
17201	}
17202
17203
17204	/**
17205	* Nested-guest VM-exit handler for virtualized EOI (VMX_EXIT_VIRTUALIZED_EOI).
17206	* Conditional VM-exit.
17207	*/
17208	HMVMX_EXIT_DECL hmR0VmxExitVirtEoiNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17209	{
17210	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17211
17212	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_VIRT_INT_DELIVERY));
17213	hmR0VmxReadExitQualVmcs(pVmxTransient);
17214	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
17215	}
17216
17217
17218	/**
17219	* Nested-guest VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
17220	*/
17221	HMVMX_EXIT_DECL hmR0VmxExitRdtscpNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17222	{
17223	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17224
17225	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDTSC_EXIT))
17226	{
17227	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_RDTSCP));
17228	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17229	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17230	}
17231	return hmR0VmxExitRdtscp(pVCpu, pVmxTransient);
17232	}
17233
17234
17235	/**
17236	* Nested-guest VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
17237	*/
17238	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvdNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17239	{
17240	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17241
17242	if (CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_WBINVD_EXIT))
17243	{
17244	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17245	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17246	}
17247	return hmR0VmxExitWbinvd(pVCpu, pVmxTransient);
17248	}
17249
17250
17251	/**
17252	* Nested-guest VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
17253	*/
17254	HMVMX_EXIT_DECL hmR0VmxExitInvpcidNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17255	{
17256	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17257
17258	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INVLPG_EXIT))
17259	{
17260	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_INVPCID));
17261	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17262	hmR0VmxReadExitQualVmcs(pVmxTransient);
17263	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17264
17265	VMXVEXITINFO ExitInfo;
17266	RT_ZERO(ExitInfo);
17267	ExitInfo.uReason = pVmxTransient->uExitReason;
17268	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17269	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17270	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17271	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17272	}
17273	return hmR0VmxExitInvpcid(pVCpu, pVmxTransient);
17274	}
17275
17276
17277	/**
17278	* Nested-guest VM-exit handler for invalid-guest state
17279	* (VMX_EXIT_ERR_INVALID_GUEST_STATE). Error VM-exit.
17280	*/
17281	HMVMX_EXIT_DECL hmR0VmxExitErrInvalidGuestStateNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17282	{
17283	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17284
17285	/*
17286	* Currently this should never happen because we fully emulate VMLAUNCH/VMRESUME in IEM.
17287	* So if it does happen, it indicates a bug possibly in the hardware-assisted VMX code.
17288	* Handle it like it's in an invalid guest state of the outer guest.
17289	*
17290	* When the fast path is implemented, this should be changed to cause the corresponding
17291	* nested-guest VM-exit.
17292	*/
17293	return hmR0VmxExitErrInvalidGuestState(pVCpu, pVmxTransient);
17294	}
17295
17296
17297	/**
17298	* Nested-guest VM-exit handler for instructions that cause VM-exits uncondtionally
17299	* and only provide the instruction length.
17300	*
17301	* Unconditional VM-exit.
17302	*/
17303	HMVMX_EXIT_DECL hmR0VmxExitInstrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17304	{
17305	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17306
17307	#ifdef VBOX_STRICT
17308	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
17309	switch (pVmxTransient->uExitReason)
17310	{
17311	case VMX_EXIT_ENCLS:
17312	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_ENCLS_EXIT));
17313	break;
17314
17315	case VMX_EXIT_VMFUNC:
17316	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_VMFUNC));
17317	break;
17318	}
17319	#endif
17320
17321	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17322	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17323	}
17324
17325
17326	/**
17327	* Nested-guest VM-exit handler for instructions that provide instruction length as
17328	* well as more information.
17329	*
17330	* Unconditional VM-exit.
17331	*/
17332	HMVMX_EXIT_DECL hmR0VmxExitInstrWithInfoNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17333	{
17334	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17335
17336	#ifdef VBOX_STRICT
17337	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
17338	switch (pVmxTransient->uExitReason)
17339	{
17340	case VMX_EXIT_GDTR_IDTR_ACCESS:
17341	case VMX_EXIT_LDTR_TR_ACCESS:
17342	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_DESC_TABLE_EXIT));
17343	break;
17344
17345	case VMX_EXIT_RDRAND:
17346	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_RDRAND_EXIT));
17347	break;
17348
17349	case VMX_EXIT_RDSEED:
17350	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_RDSEED_EXIT));
17351	break;
17352
17353	case VMX_EXIT_XSAVES:
17354	case VMX_EXIT_XRSTORS:
17355	/** @todo NSTVMX: Verify XSS-bitmap. */
17356	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_XSAVES_XRSTORS));
17357	break;
17358
17359	case VMX_EXIT_UMWAIT:
17360	case VMX_EXIT_TPAUSE:
17361	Assert(CPUMIsGuestVmxProcCtlsSet(pCtx, VMX_PROC_CTLS_RDTSC_EXIT));
17362	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_USER_WAIT_PAUSE));
17363	break;
17364	}
17365	#endif
17366
17367	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17368	hmR0VmxReadExitQualVmcs(pVmxTransient);
17369	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17370
17371	VMXVEXITINFO ExitInfo;
17372	RT_ZERO(ExitInfo);
17373	ExitInfo.uReason = pVmxTransient->uExitReason;
17374	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17375	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17376	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17377	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17378	}
17379
17380	/** @} */
17381	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
17382

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

source: vbox/trunk/src/VBox/VMM/VMMR0/HMVMXR0.cpp@ 85765

Download in other formats: