HMVMXR0.cpp@ 91587

Last change on this file since 91587 was 91587, checked in by vboxsync, 3 years ago
VMM: Purge VINF_TRPM_XCPT_DISPATCHED, no longer used anywhere.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 740.6 KB

Line
1	/* $Id: HMVMXR0.cpp 91587 2021-10-06 09:37:08Z vboxsync $ */
2	/** @file
3	* HM VMX (Intel VT-x) - Host Context Ring-0.
4	*/
5
6	/*
7	* Copyright (C) 2012-2020 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18
19	/*********************************************************************************************************************************
20	* Header Files *
21	*********************************************************************************************************************************/
22	#define LOG_GROUP LOG_GROUP_HM
23	#define VMCPU_INCL_CPUM_GST_CTX
24	#include <iprt/x86.h>
25	#include <iprt/asm-amd64-x86.h>
26	#include <iprt/thread.h>
27	#include <iprt/mem.h>
28	#include <iprt/mp.h>
29
30	#include <VBox/vmm/pdmapi.h>
31	#include <VBox/vmm/dbgf.h>
32	#include <VBox/vmm/iem.h>
33	#include <VBox/vmm/iom.h>
34	#include <VBox/vmm/tm.h>
35	#include <VBox/vmm/em.h>
36	#include <VBox/vmm/gim.h>
37	#include <VBox/vmm/apic.h>
38	#include "HMInternal.h"
39	#include <VBox/vmm/vmcc.h>
40	#include <VBox/vmm/hmvmxinline.h>
41	#include "HMVMXR0.h"
42	#include "dtrace/VBoxVMM.h"
43
44	#ifdef DEBUG_ramshankar
45	# define HMVMX_ALWAYS_SAVE_GUEST_RFLAGS
46	# define HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
47	# define HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE
48	# define HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
49	# define HMVMX_ALWAYS_CLEAN_TRANSIENT
50	# define HMVMX_ALWAYS_CHECK_GUEST_STATE
51	# define HMVMX_ALWAYS_TRAP_ALL_XCPTS
52	# define HMVMX_ALWAYS_TRAP_PF
53	# define HMVMX_ALWAYS_FLUSH_TLB
54	# define HMVMX_ALWAYS_SWAP_EFER
55	#endif
56
57
58	/*********************************************************************************************************************************
59	* Defined Constants And Macros *
60	*********************************************************************************************************************************/
61	/** Use the function table. */
62	#define HMVMX_USE_FUNCTION_TABLE
63
64	/** Determine which tagged-TLB flush handler to use. */
65	#define HMVMX_FLUSH_TAGGED_TLB_EPT_VPID 0
66	#define HMVMX_FLUSH_TAGGED_TLB_EPT 1
67	#define HMVMX_FLUSH_TAGGED_TLB_VPID 2
68	#define HMVMX_FLUSH_TAGGED_TLB_NONE 3
69
70	/**
71	* Flags to skip redundant reads of some common VMCS fields that are not part of
72	* the guest-CPU or VCPU state but are needed while handling VM-exits.
73	*/
74	#define HMVMX_READ_IDT_VECTORING_INFO RT_BIT_32(0)
75	#define HMVMX_READ_IDT_VECTORING_ERROR_CODE RT_BIT_32(1)
76	#define HMVMX_READ_EXIT_QUALIFICATION RT_BIT_32(2)
77	#define HMVMX_READ_EXIT_INSTR_LEN RT_BIT_32(3)
78	#define HMVMX_READ_EXIT_INTERRUPTION_INFO RT_BIT_32(4)
79	#define HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE RT_BIT_32(5)
80	#define HMVMX_READ_EXIT_INSTR_INFO RT_BIT_32(6)
81	#define HMVMX_READ_GUEST_LINEAR_ADDR RT_BIT_32(7)
82	#define HMVMX_READ_GUEST_PHYSICAL_ADDR RT_BIT_32(8)
83	#define HMVMX_READ_GUEST_PENDING_DBG_XCPTS RT_BIT_32(9)
84
85	/** All the VMCS fields required for processing of exception/NMI VM-exits. */
86	#define HMVMX_READ_XCPT_INFO ( HMVMX_READ_EXIT_INTERRUPTION_INFO \
87	\| HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE \
88	\| HMVMX_READ_EXIT_INSTR_LEN \
89	\| HMVMX_READ_IDT_VECTORING_INFO \
90	\| HMVMX_READ_IDT_VECTORING_ERROR_CODE)
91
92	/** Assert that all the given fields have been read from the VMCS. */
93	#ifdef VBOX_STRICT
94	# define HMVMX_ASSERT_READ(a_pVmxTransient, a_fReadFields) \
95	do { \
96	uint32_t const fVmcsFieldRead = ASMAtomicUoReadU32(&pVmxTransient->fVmcsFieldsRead); \
97	Assert((fVmcsFieldRead & (a_fReadFields)) == (a_fReadFields)); \
98	} while (0)
99	#else
100	# define HMVMX_ASSERT_READ(a_pVmxTransient, a_fReadFields) do { } while (0)
101	#endif
102
103	/**
104	* Subset of the guest-CPU state that is kept by VMX R0 code while executing the
105	* guest using hardware-assisted VMX.
106	*
107	* This excludes state like GPRs (other than RSP) which are always are
108	* swapped and restored across the world-switch and also registers like EFER,
109	* MSR which cannot be modified by the guest without causing a VM-exit.
110	*/
111	#define HMVMX_CPUMCTX_EXTRN_ALL ( CPUMCTX_EXTRN_RIP \
112	\| CPUMCTX_EXTRN_RFLAGS \
113	\| CPUMCTX_EXTRN_RSP \
114	\| CPUMCTX_EXTRN_SREG_MASK \
115	\| CPUMCTX_EXTRN_TABLE_MASK \
116	\| CPUMCTX_EXTRN_KERNEL_GS_BASE \
117	\| CPUMCTX_EXTRN_SYSCALL_MSRS \
118	\| CPUMCTX_EXTRN_SYSENTER_MSRS \
119	\| CPUMCTX_EXTRN_TSC_AUX \
120	\| CPUMCTX_EXTRN_OTHER_MSRS \
121	\| CPUMCTX_EXTRN_CR0 \
122	\| CPUMCTX_EXTRN_CR3 \
123	\| CPUMCTX_EXTRN_CR4 \
124	\| CPUMCTX_EXTRN_DR7 \
125	\| CPUMCTX_EXTRN_HWVIRT \
126	\| CPUMCTX_EXTRN_HM_VMX_MASK)
127
128	/**
129	* Exception bitmap mask for real-mode guests (real-on-v86).
130	*
131	* We need to intercept all exceptions manually except:
132	* - \#AC and \#DB are always intercepted to prevent the CPU from deadlocking
133	* due to bugs in Intel CPUs.
134	* - \#PF need not be intercepted even in real-mode if we have nested paging
135	* support.
136	*/
137	#define HMVMX_REAL_MODE_XCPT_MASK ( RT_BIT(X86_XCPT_DE) /* always: \| RT_BIT(X86_XCPT_DB) */ \| RT_BIT(X86_XCPT_NMI) \
138	\| RT_BIT(X86_XCPT_BP) \| RT_BIT(X86_XCPT_OF) \| RT_BIT(X86_XCPT_BR) \
139	\| RT_BIT(X86_XCPT_UD) \| RT_BIT(X86_XCPT_NM) \| RT_BIT(X86_XCPT_DF) \
140	\| RT_BIT(X86_XCPT_CO_SEG_OVERRUN) \| RT_BIT(X86_XCPT_TS) \| RT_BIT(X86_XCPT_NP) \
141	\| RT_BIT(X86_XCPT_SS) \| RT_BIT(X86_XCPT_GP) /* RT_BIT(X86_XCPT_PF) */ \
142	\| RT_BIT(X86_XCPT_MF) /* always: \| RT_BIT(X86_XCPT_AC) */ \| RT_BIT(X86_XCPT_MC) \
143	\| RT_BIT(X86_XCPT_XF))
144
145	/** Maximum VM-instruction error number. */
146	#define HMVMX_INSTR_ERROR_MAX 28
147
148	/** Profiling macro. */
149	#ifdef HM_PROFILE_EXIT_DISPATCH
150	# define HMVMX_START_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitDispatch, ed)
151	# define HMVMX_STOP_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitDispatch, ed)
152	#else
153	# define HMVMX_START_EXIT_DISPATCH_PROF() do { } while (0)
154	# define HMVMX_STOP_EXIT_DISPATCH_PROF() do { } while (0)
155	#endif
156
157	/** Assert that preemption is disabled or covered by thread-context hooks. */
158	#define HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu) Assert( VMMR0ThreadCtxHookIsEnabled((a_pVCpu)) \
159	\|\| !RTThreadPreemptIsEnabled(NIL_RTTHREAD))
160
161	/** Assert that we haven't migrated CPUs when thread-context hooks are not
162	* used. */
163	#define HMVMX_ASSERT_CPU_SAFE(a_pVCpu) AssertMsg( VMMR0ThreadCtxHookIsEnabled((a_pVCpu)) \
164	\|\| (a_pVCpu)->hmr0.s.idEnteredCpu == RTMpCpuId(), \
165	("Illegal migration! Entered on CPU %u Current %u\n", \
166	(a_pVCpu)->hmr0.s.idEnteredCpu, RTMpCpuId()))
167
168	/** Asserts that the given CPUMCTX_EXTRN_XXX bits are present in the guest-CPU
169	* context. */
170	#define HMVMX_CPUMCTX_ASSERT(a_pVCpu, a_fExtrnMbz) AssertMsg(!((a_pVCpu)->cpum.GstCtx.fExtrn & (a_fExtrnMbz)), \
171	("fExtrn=%#RX64 fExtrnMbz=%#RX64\n", \
172	(a_pVCpu)->cpum.GstCtx.fExtrn, (a_fExtrnMbz)))
173
174	/** Log the VM-exit reason with an easily visible marker to identify it in a
175	* potential sea of logging data. */
176	#define HMVMX_LOG_EXIT(a_pVCpu, a_uExitReason) \
177	do { \
178	Log4(("VM-exit: vcpu[%RU32] %85s -v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-\n", (a_pVCpu)->idCpu, \
179	HMGetVmxExitName(a_uExitReason))); \
180	} while (0) \
181
182
183	/*********************************************************************************************************************************
184	* Structures and Typedefs *
185	*********************************************************************************************************************************/
186	/**
187	* VMX per-VCPU transient state.
188	*
189	* A state structure for holding miscellaneous information across
190	* VMX non-root operation and restored after the transition.
191	*
192	* Note: The members are ordered and aligned such that the most
193	* frequently used ones (in the guest execution loop) fall within
194	* the first cache line.
195	*/
196	typedef struct VMXTRANSIENT
197	{
198	/** Mask of currently read VMCS fields; HMVMX_READ_XXX. */
199	uint32_t fVmcsFieldsRead;
200	/** The guest's TPR value used for TPR shadowing. */
201	uint8_t u8GuestTpr;
202	uint8_t abAlignment0[3];
203
204	/** Whether the VM-exit was caused by a page-fault during delivery of an
205	* external interrupt or NMI. */
206	bool fVectoringPF;
207	/** Whether the VM-exit was caused by a page-fault during delivery of a
208	* contributory exception or a page-fault. */
209	bool fVectoringDoublePF;
210	/** Whether the VM-entry failed or not. */
211	bool fVMEntryFailed;
212	/** Whether the TSC_AUX MSR needs to be removed from the auto-load/store MSR
213	* area after VM-exit. */
214	bool fRemoveTscAuxMsr;
215	/** Whether TSC-offsetting and VMX-preemption timer was updated before VM-entry. */
216	bool fUpdatedTscOffsettingAndPreemptTimer;
217	/** Whether we are currently executing a nested-guest. */
218	bool fIsNestedGuest;
219	/** Whether the guest debug state was active at the time of VM-exit. */
220	bool fWasGuestDebugStateActive;
221	/** Whether the hyper debug state was active at the time of VM-exit. */
222	bool fWasHyperDebugStateActive;
223
224	/** The basic VM-exit reason. */
225	uint32_t uExitReason;
226	/** The VM-exit interruption error code. */
227	uint32_t uExitIntErrorCode;
228
229	/** The host's rflags/eflags. */
230	RTCCUINTREG fEFlags;
231
232	/** The VM-exit exit code qualification. */
233	uint64_t uExitQual;
234
235	/** The VMCS info. object. */
236	PVMXVMCSINFO pVmcsInfo;
237
238	/** The VM-exit interruption-information field. */
239	uint32_t uExitIntInfo;
240	/** The VM-exit instruction-length field. */
241	uint32_t cbExitInstr;
242
243	/** The VM-exit instruction-information field. */
244	VMXEXITINSTRINFO ExitInstrInfo;
245	/** IDT-vectoring information field. */
246	uint32_t uIdtVectoringInfo;
247
248	/** IDT-vectoring error code. */
249	uint32_t uIdtVectoringErrorCode;
250	uint32_t u32Alignment0;
251
252	/** The Guest-linear address. */
253	uint64_t uGuestLinearAddr;
254
255	/** The Guest-physical address. */
256	uint64_t uGuestPhysicalAddr;
257
258	/** The Guest pending-debug exceptions. */
259	uint64_t uGuestPendingDbgXcpts;
260
261	/** The VM-entry interruption-information field. */
262	uint32_t uEntryIntInfo;
263	/** The VM-entry exception error code field. */
264	uint32_t uEntryXcptErrorCode;
265
266	/** The VM-entry instruction length field. */
267	uint32_t cbEntryInstr;
268	} VMXTRANSIENT;
269	AssertCompileMemberSize(VMXTRANSIENT, ExitInstrInfo, sizeof(uint32_t));
270	AssertCompileMemberAlignment(VMXTRANSIENT, fVmcsFieldsRead, 8);
271	AssertCompileMemberAlignment(VMXTRANSIENT, fVectoringPF, 8);
272	AssertCompileMemberAlignment(VMXTRANSIENT, uExitReason, 8);
273	AssertCompileMemberAlignment(VMXTRANSIENT, fEFlags, 8);
274	AssertCompileMemberAlignment(VMXTRANSIENT, uExitQual, 8);
275	AssertCompileMemberAlignment(VMXTRANSIENT, pVmcsInfo, 8);
276	AssertCompileMemberAlignment(VMXTRANSIENT, uExitIntInfo, 8);
277	AssertCompileMemberAlignment(VMXTRANSIENT, ExitInstrInfo, 8);
278	AssertCompileMemberAlignment(VMXTRANSIENT, uIdtVectoringErrorCode, 8);
279	AssertCompileMemberAlignment(VMXTRANSIENT, uGuestLinearAddr, 8);
280	AssertCompileMemberAlignment(VMXTRANSIENT, uGuestPhysicalAddr, 8);
281	AssertCompileMemberAlignment(VMXTRANSIENT, uEntryIntInfo, 8);
282	AssertCompileMemberAlignment(VMXTRANSIENT, cbEntryInstr, 8);
283	/** Pointer to VMX transient state. */
284	typedef VMXTRANSIENT *PVMXTRANSIENT;
285	/** Pointer to a const VMX transient state. */
286	typedef const VMXTRANSIENT *PCVMXTRANSIENT;
287
288	/**
289	* VMX page allocation information.
290	*/
291	typedef struct
292	{
293	uint32_t fValid; /*< Whether to allocate this page (e.g, based on a CPU feature). /
294	uint32_t uPadding0; /*< Padding to ensure array of these structs are aligned to a multiple of 8. /
295	PRTHCPHYS pHCPhys; /*< Where to store the host-physical address of the allocation. /
296	PRTR0PTR ppVirt; /*< Where to store the host-virtual address of the allocation. /
297	} VMXPAGEALLOCINFO;
298	/** Pointer to VMX page-allocation info. */
299	typedef VMXPAGEALLOCINFO *PVMXPAGEALLOCINFO;
300	/** Pointer to a const VMX page-allocation info. */
301	typedef const VMXPAGEALLOCINFO *PCVMXPAGEALLOCINFO;
302	AssertCompileSizeAlignment(VMXPAGEALLOCINFO, 8);
303
304	/**
305	* Memory operand read or write access.
306	*/
307	typedef enum VMXMEMACCESS
308	{
309	VMXMEMACCESS_READ = 0,
310	VMXMEMACCESS_WRITE = 1
311	} VMXMEMACCESS;
312
313	/**
314	* VMX VM-exit handler.
315	*
316	* @returns Strict VBox status code (i.e. informational status codes too).
317	* @param pVCpu The cross context virtual CPU structure.
318	* @param pVmxTransient The VMX-transient structure.
319	*/
320	#ifndef HMVMX_USE_FUNCTION_TABLE
321	typedef VBOXSTRICTRC FNVMXEXITHANDLER(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
322	#else
323	typedef DECLCALLBACKTYPE(VBOXSTRICTRC, FNVMXEXITHANDLER,(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient));
324	/** Pointer to VM-exit handler. */
325	typedef FNVMXEXITHANDLER *PFNVMXEXITHANDLER;
326	#endif
327
328	/**
329	* VMX VM-exit handler, non-strict status code.
330	*
331	* This is generally the same as FNVMXEXITHANDLER, the NSRC bit is just FYI.
332	*
333	* @returns VBox status code, no informational status code returned.
334	* @param pVCpu The cross context virtual CPU structure.
335	* @param pVmxTransient The VMX-transient structure.
336	*
337	* @remarks This is not used on anything returning VERR_EM_INTERPRETER as the
338	* use of that status code will be replaced with VINF_EM_SOMETHING
339	* later when switching over to IEM.
340	*/
341	#ifndef HMVMX_USE_FUNCTION_TABLE
342	typedef int FNVMXEXITHANDLERNSRC(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
343	#else
344	typedef FNVMXEXITHANDLER FNVMXEXITHANDLERNSRC;
345	#endif
346
347
348	/*********************************************************************************************************************************
349	* Internal Functions *
350	*********************************************************************************************************************************/
351	#ifndef HMVMX_USE_FUNCTION_TABLE
352	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
353	# define HMVMX_EXIT_DECL DECLINLINE(VBOXSTRICTRC)
354	# define HMVMX_EXIT_NSRC_DECL DECLINLINE(int)
355	#else
356	# define HMVMX_EXIT_DECL static DECLCALLBACK(VBOXSTRICTRC)
357	# define HMVMX_EXIT_NSRC_DECL HMVMX_EXIT_DECL
358	#endif
359	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
360	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient);
361	#endif
362
363	static int hmR0VmxImportGuestState(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint64_t fWhat);
364
365	/** @name VM-exit handler prototypes.
366	* @{
367	*/
368	static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmi;
369	static FNVMXEXITHANDLER hmR0VmxExitExtInt;
370	static FNVMXEXITHANDLER hmR0VmxExitTripleFault;
371	static FNVMXEXITHANDLERNSRC hmR0VmxExitIntWindow;
372	static FNVMXEXITHANDLERNSRC hmR0VmxExitNmiWindow;
373	static FNVMXEXITHANDLER hmR0VmxExitTaskSwitch;
374	static FNVMXEXITHANDLER hmR0VmxExitCpuid;
375	static FNVMXEXITHANDLER hmR0VmxExitGetsec;
376	static FNVMXEXITHANDLER hmR0VmxExitHlt;
377	static FNVMXEXITHANDLERNSRC hmR0VmxExitInvd;
378	static FNVMXEXITHANDLER hmR0VmxExitInvlpg;
379	static FNVMXEXITHANDLER hmR0VmxExitRdpmc;
380	static FNVMXEXITHANDLER hmR0VmxExitVmcall;
381	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
382	static FNVMXEXITHANDLER hmR0VmxExitVmclear;
383	static FNVMXEXITHANDLER hmR0VmxExitVmlaunch;
384	static FNVMXEXITHANDLER hmR0VmxExitVmptrld;
385	static FNVMXEXITHANDLER hmR0VmxExitVmptrst;
386	static FNVMXEXITHANDLER hmR0VmxExitVmread;
387	static FNVMXEXITHANDLER hmR0VmxExitVmresume;
388	static FNVMXEXITHANDLER hmR0VmxExitVmwrite;
389	static FNVMXEXITHANDLER hmR0VmxExitVmxoff;
390	static FNVMXEXITHANDLER hmR0VmxExitVmxon;
391	static FNVMXEXITHANDLER hmR0VmxExitInvvpid;
392	#endif
393	static FNVMXEXITHANDLER hmR0VmxExitRdtsc;
394	static FNVMXEXITHANDLER hmR0VmxExitMovCRx;
395	static FNVMXEXITHANDLER hmR0VmxExitMovDRx;
396	static FNVMXEXITHANDLER hmR0VmxExitIoInstr;
397	static FNVMXEXITHANDLER hmR0VmxExitRdmsr;
398	static FNVMXEXITHANDLER hmR0VmxExitWrmsr;
399	static FNVMXEXITHANDLER hmR0VmxExitMwait;
400	static FNVMXEXITHANDLER hmR0VmxExitMtf;
401	static FNVMXEXITHANDLER hmR0VmxExitMonitor;
402	static FNVMXEXITHANDLER hmR0VmxExitPause;
403	static FNVMXEXITHANDLERNSRC hmR0VmxExitTprBelowThreshold;
404	static FNVMXEXITHANDLER hmR0VmxExitApicAccess;
405	static FNVMXEXITHANDLER hmR0VmxExitEptViolation;
406	static FNVMXEXITHANDLER hmR0VmxExitEptMisconfig;
407	static FNVMXEXITHANDLER hmR0VmxExitRdtscp;
408	static FNVMXEXITHANDLER hmR0VmxExitPreemptTimer;
409	static FNVMXEXITHANDLERNSRC hmR0VmxExitWbinvd;
410	static FNVMXEXITHANDLER hmR0VmxExitXsetbv;
411	static FNVMXEXITHANDLER hmR0VmxExitInvpcid;
412	static FNVMXEXITHANDLERNSRC hmR0VmxExitSetPendingXcptUD;
413	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrInvalidGuestState;
414	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrUnexpected;
415	/** @} */
416
417	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
418	/** @name Nested-guest VM-exit handler prototypes.
419	* @{
420	*/
421	static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmiNested;
422	static FNVMXEXITHANDLER hmR0VmxExitTripleFaultNested;
423	static FNVMXEXITHANDLERNSRC hmR0VmxExitIntWindowNested;
424	static FNVMXEXITHANDLERNSRC hmR0VmxExitNmiWindowNested;
425	static FNVMXEXITHANDLER hmR0VmxExitTaskSwitchNested;
426	static FNVMXEXITHANDLER hmR0VmxExitHltNested;
427	static FNVMXEXITHANDLER hmR0VmxExitInvlpgNested;
428	static FNVMXEXITHANDLER hmR0VmxExitRdpmcNested;
429	static FNVMXEXITHANDLER hmR0VmxExitVmreadVmwriteNested;
430	static FNVMXEXITHANDLER hmR0VmxExitRdtscNested;
431	static FNVMXEXITHANDLER hmR0VmxExitMovCRxNested;
432	static FNVMXEXITHANDLER hmR0VmxExitMovDRxNested;
433	static FNVMXEXITHANDLER hmR0VmxExitIoInstrNested;
434	static FNVMXEXITHANDLER hmR0VmxExitRdmsrNested;
435	static FNVMXEXITHANDLER hmR0VmxExitWrmsrNested;
436	static FNVMXEXITHANDLER hmR0VmxExitMwaitNested;
437	static FNVMXEXITHANDLER hmR0VmxExitMtfNested;
438	static FNVMXEXITHANDLER hmR0VmxExitMonitorNested;
439	static FNVMXEXITHANDLER hmR0VmxExitPauseNested;
440	static FNVMXEXITHANDLERNSRC hmR0VmxExitTprBelowThresholdNested;
441	static FNVMXEXITHANDLER hmR0VmxExitApicAccessNested;
442	static FNVMXEXITHANDLER hmR0VmxExitApicWriteNested;
443	static FNVMXEXITHANDLER hmR0VmxExitVirtEoiNested;
444	static FNVMXEXITHANDLER hmR0VmxExitRdtscpNested;
445	static FNVMXEXITHANDLERNSRC hmR0VmxExitWbinvdNested;
446	static FNVMXEXITHANDLER hmR0VmxExitInvpcidNested;
447	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrInvalidGuestStateNested;
448	static FNVMXEXITHANDLER hmR0VmxExitInstrNested;
449	static FNVMXEXITHANDLER hmR0VmxExitInstrWithInfoNested;
450	/** @} */
451	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
452
453
454	/*********************************************************************************************************************************
455	* Global Variables *
456	*********************************************************************************************************************************/
457	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
458	/**
459	* Array of all VMCS fields.
460	* Any fields added to the VT-x spec. should be added here.
461	*
462	* Currently only used to derive shadow VMCS fields for hardware-assisted execution
463	* of nested-guests.
464	*/
465	static const uint32_t g_aVmcsFields[] =
466	{
467	/* 16-bit control fields. */
468	VMX_VMCS16_VPID,
469	VMX_VMCS16_POSTED_INT_NOTIFY_VECTOR,
470	VMX_VMCS16_EPTP_INDEX,
471
472	/* 16-bit guest-state fields. */
473	VMX_VMCS16_GUEST_ES_SEL,
474	VMX_VMCS16_GUEST_CS_SEL,
475	VMX_VMCS16_GUEST_SS_SEL,
476	VMX_VMCS16_GUEST_DS_SEL,
477	VMX_VMCS16_GUEST_FS_SEL,
478	VMX_VMCS16_GUEST_GS_SEL,
479	VMX_VMCS16_GUEST_LDTR_SEL,
480	VMX_VMCS16_GUEST_TR_SEL,
481	VMX_VMCS16_GUEST_INTR_STATUS,
482	VMX_VMCS16_GUEST_PML_INDEX,
483
484	/* 16-bits host-state fields. */
485	VMX_VMCS16_HOST_ES_SEL,
486	VMX_VMCS16_HOST_CS_SEL,
487	VMX_VMCS16_HOST_SS_SEL,
488	VMX_VMCS16_HOST_DS_SEL,
489	VMX_VMCS16_HOST_FS_SEL,
490	VMX_VMCS16_HOST_GS_SEL,
491	VMX_VMCS16_HOST_TR_SEL,
492
493	/* 64-bit control fields. */
494	VMX_VMCS64_CTRL_IO_BITMAP_A_FULL,
495	VMX_VMCS64_CTRL_IO_BITMAP_A_HIGH,
496	VMX_VMCS64_CTRL_IO_BITMAP_B_FULL,
497	VMX_VMCS64_CTRL_IO_BITMAP_B_HIGH,
498	VMX_VMCS64_CTRL_MSR_BITMAP_FULL,
499	VMX_VMCS64_CTRL_MSR_BITMAP_HIGH,
500	VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL,
501	VMX_VMCS64_CTRL_EXIT_MSR_STORE_HIGH,
502	VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL,
503	VMX_VMCS64_CTRL_EXIT_MSR_LOAD_HIGH,
504	VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL,
505	VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_HIGH,
506	VMX_VMCS64_CTRL_EXEC_VMCS_PTR_FULL,
507	VMX_VMCS64_CTRL_EXEC_VMCS_PTR_HIGH,
508	VMX_VMCS64_CTRL_EXEC_PML_ADDR_FULL,
509	VMX_VMCS64_CTRL_EXEC_PML_ADDR_HIGH,
510	VMX_VMCS64_CTRL_TSC_OFFSET_FULL,
511	VMX_VMCS64_CTRL_TSC_OFFSET_HIGH,
512	VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL,
513	VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_HIGH,
514	VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL,
515	VMX_VMCS64_CTRL_APIC_ACCESSADDR_HIGH,
516	VMX_VMCS64_CTRL_POSTED_INTR_DESC_FULL,
517	VMX_VMCS64_CTRL_POSTED_INTR_DESC_HIGH,
518	VMX_VMCS64_CTRL_VMFUNC_CTRLS_FULL,
519	VMX_VMCS64_CTRL_VMFUNC_CTRLS_HIGH,
520	VMX_VMCS64_CTRL_EPTP_FULL,
521	VMX_VMCS64_CTRL_EPTP_HIGH,
522	VMX_VMCS64_CTRL_EOI_BITMAP_0_FULL,
523	VMX_VMCS64_CTRL_EOI_BITMAP_0_HIGH,
524	VMX_VMCS64_CTRL_EOI_BITMAP_1_FULL,
525	VMX_VMCS64_CTRL_EOI_BITMAP_1_HIGH,
526	VMX_VMCS64_CTRL_EOI_BITMAP_2_FULL,
527	VMX_VMCS64_CTRL_EOI_BITMAP_2_HIGH,
528	VMX_VMCS64_CTRL_EOI_BITMAP_3_FULL,
529	VMX_VMCS64_CTRL_EOI_BITMAP_3_HIGH,
530	VMX_VMCS64_CTRL_EPTP_LIST_FULL,
531	VMX_VMCS64_CTRL_EPTP_LIST_HIGH,
532	VMX_VMCS64_CTRL_VMREAD_BITMAP_FULL,
533	VMX_VMCS64_CTRL_VMREAD_BITMAP_HIGH,
534	VMX_VMCS64_CTRL_VMWRITE_BITMAP_FULL,
535	VMX_VMCS64_CTRL_VMWRITE_BITMAP_HIGH,
536	VMX_VMCS64_CTRL_VE_XCPT_INFO_ADDR_FULL,
537	VMX_VMCS64_CTRL_VE_XCPT_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_SPPTP_FULL,
543	VMX_VMCS64_CTRL_SPPTP_HIGH,
544	VMX_VMCS64_CTRL_TSC_MULTIPLIER_FULL,
545	VMX_VMCS64_CTRL_TSC_MULTIPLIER_HIGH,
546	VMX_VMCS64_CTRL_PROC_EXEC3_FULL,
547	VMX_VMCS64_CTRL_PROC_EXEC3_HIGH,
548	VMX_VMCS64_CTRL_ENCLV_EXITING_BITMAP_FULL,
549	VMX_VMCS64_CTRL_ENCLV_EXITING_BITMAP_HIGH,
550
551	/* 64-bit read-only data fields. */
552	VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL,
553	VMX_VMCS64_RO_GUEST_PHYS_ADDR_HIGH,
554
555	/* 64-bit guest-state fields. */
556	VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL,
557	VMX_VMCS64_GUEST_VMCS_LINK_PTR_HIGH,
558	VMX_VMCS64_GUEST_DEBUGCTL_FULL,
559	VMX_VMCS64_GUEST_DEBUGCTL_HIGH,
560	VMX_VMCS64_GUEST_PAT_FULL,
561	VMX_VMCS64_GUEST_PAT_HIGH,
562	VMX_VMCS64_GUEST_EFER_FULL,
563	VMX_VMCS64_GUEST_EFER_HIGH,
564	VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL,
565	VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_HIGH,
566	VMX_VMCS64_GUEST_PDPTE0_FULL,
567	VMX_VMCS64_GUEST_PDPTE0_HIGH,
568	VMX_VMCS64_GUEST_PDPTE1_FULL,
569	VMX_VMCS64_GUEST_PDPTE1_HIGH,
570	VMX_VMCS64_GUEST_PDPTE2_FULL,
571	VMX_VMCS64_GUEST_PDPTE2_HIGH,
572	VMX_VMCS64_GUEST_PDPTE3_FULL,
573	VMX_VMCS64_GUEST_PDPTE3_HIGH,
574	VMX_VMCS64_GUEST_BNDCFGS_FULL,
575	VMX_VMCS64_GUEST_BNDCFGS_HIGH,
576	VMX_VMCS64_GUEST_RTIT_CTL_FULL,
577	VMX_VMCS64_GUEST_RTIT_CTL_HIGH,
578	VMX_VMCS64_GUEST_PKRS_FULL,
579	VMX_VMCS64_GUEST_PKRS_HIGH,
580
581	/* 64-bit host-state fields. */
582	VMX_VMCS64_HOST_PAT_FULL,
583	VMX_VMCS64_HOST_PAT_HIGH,
584	VMX_VMCS64_HOST_EFER_FULL,
585	VMX_VMCS64_HOST_EFER_HIGH,
586	VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_FULL,
587	VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_HIGH,
588	VMX_VMCS64_HOST_PKRS_FULL,
589	VMX_VMCS64_HOST_PKRS_HIGH,
590
591	/* 32-bit control fields. */
592	VMX_VMCS32_CTRL_PIN_EXEC,
593	VMX_VMCS32_CTRL_PROC_EXEC,
594	VMX_VMCS32_CTRL_EXCEPTION_BITMAP,
595	VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK,
596	VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH,
597	VMX_VMCS32_CTRL_CR3_TARGET_COUNT,
598	VMX_VMCS32_CTRL_EXIT,
599	VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT,
600	VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT,
601	VMX_VMCS32_CTRL_ENTRY,
602	VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT,
603	VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO,
604	VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE,
605	VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH,
606	VMX_VMCS32_CTRL_TPR_THRESHOLD,
607	VMX_VMCS32_CTRL_PROC_EXEC2,
608	VMX_VMCS32_CTRL_PLE_GAP,
609	VMX_VMCS32_CTRL_PLE_WINDOW,
610
611	/* 32-bits read-only fields. */
612	VMX_VMCS32_RO_VM_INSTR_ERROR,
613	VMX_VMCS32_RO_EXIT_REASON,
614	VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO,
615	VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE,
616	VMX_VMCS32_RO_IDT_VECTORING_INFO,
617	VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE,
618	VMX_VMCS32_RO_EXIT_INSTR_LENGTH,
619	VMX_VMCS32_RO_EXIT_INSTR_INFO,
620
621	/* 32-bit guest-state fields. */
622	VMX_VMCS32_GUEST_ES_LIMIT,
623	VMX_VMCS32_GUEST_CS_LIMIT,
624	VMX_VMCS32_GUEST_SS_LIMIT,
625	VMX_VMCS32_GUEST_DS_LIMIT,
626	VMX_VMCS32_GUEST_FS_LIMIT,
627	VMX_VMCS32_GUEST_GS_LIMIT,
628	VMX_VMCS32_GUEST_LDTR_LIMIT,
629	VMX_VMCS32_GUEST_TR_LIMIT,
630	VMX_VMCS32_GUEST_GDTR_LIMIT,
631	VMX_VMCS32_GUEST_IDTR_LIMIT,
632	VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS,
633	VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS,
634	VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS,
635	VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS,
636	VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS,
637	VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS,
638	VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS,
639	VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS,
640	VMX_VMCS32_GUEST_INT_STATE,
641	VMX_VMCS32_GUEST_ACTIVITY_STATE,
642	VMX_VMCS32_GUEST_SMBASE,
643	VMX_VMCS32_GUEST_SYSENTER_CS,
644	VMX_VMCS32_PREEMPT_TIMER_VALUE,
645
646	/* 32-bit host-state fields. */
647	VMX_VMCS32_HOST_SYSENTER_CS,
648
649	/* Natural-width control fields. */
650	VMX_VMCS_CTRL_CR0_MASK,
651	VMX_VMCS_CTRL_CR4_MASK,
652	VMX_VMCS_CTRL_CR0_READ_SHADOW,
653	VMX_VMCS_CTRL_CR4_READ_SHADOW,
654	VMX_VMCS_CTRL_CR3_TARGET_VAL0,
655	VMX_VMCS_CTRL_CR3_TARGET_VAL1,
656	VMX_VMCS_CTRL_CR3_TARGET_VAL2,
657	VMX_VMCS_CTRL_CR3_TARGET_VAL3,
658
659	/* Natural-width read-only data fields. */
660	VMX_VMCS_RO_EXIT_QUALIFICATION,
661	VMX_VMCS_RO_IO_RCX,
662	VMX_VMCS_RO_IO_RSI,
663	VMX_VMCS_RO_IO_RDI,
664	VMX_VMCS_RO_IO_RIP,
665	VMX_VMCS_RO_GUEST_LINEAR_ADDR,
666
667	/* Natural-width guest-state field */
668	VMX_VMCS_GUEST_CR0,
669	VMX_VMCS_GUEST_CR3,
670	VMX_VMCS_GUEST_CR4,
671	VMX_VMCS_GUEST_ES_BASE,
672	VMX_VMCS_GUEST_CS_BASE,
673	VMX_VMCS_GUEST_SS_BASE,
674	VMX_VMCS_GUEST_DS_BASE,
675	VMX_VMCS_GUEST_FS_BASE,
676	VMX_VMCS_GUEST_GS_BASE,
677	VMX_VMCS_GUEST_LDTR_BASE,
678	VMX_VMCS_GUEST_TR_BASE,
679	VMX_VMCS_GUEST_GDTR_BASE,
680	VMX_VMCS_GUEST_IDTR_BASE,
681	VMX_VMCS_GUEST_DR7,
682	VMX_VMCS_GUEST_RSP,
683	VMX_VMCS_GUEST_RIP,
684	VMX_VMCS_GUEST_RFLAGS,
685	VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS,
686	VMX_VMCS_GUEST_SYSENTER_ESP,
687	VMX_VMCS_GUEST_SYSENTER_EIP,
688	VMX_VMCS_GUEST_S_CET,
689	VMX_VMCS_GUEST_SSP,
690	VMX_VMCS_GUEST_INTR_SSP_TABLE_ADDR,
691
692	/* Natural-width host-state fields */
693	VMX_VMCS_HOST_CR0,
694	VMX_VMCS_HOST_CR3,
695	VMX_VMCS_HOST_CR4,
696	VMX_VMCS_HOST_FS_BASE,
697	VMX_VMCS_HOST_GS_BASE,
698	VMX_VMCS_HOST_TR_BASE,
699	VMX_VMCS_HOST_GDTR_BASE,
700	VMX_VMCS_HOST_IDTR_BASE,
701	VMX_VMCS_HOST_SYSENTER_ESP,
702	VMX_VMCS_HOST_SYSENTER_EIP,
703	VMX_VMCS_HOST_RSP,
704	VMX_VMCS_HOST_RIP,
705	VMX_VMCS_HOST_S_CET,
706	VMX_VMCS_HOST_SSP,
707	VMX_VMCS_HOST_INTR_SSP_TABLE_ADDR
708	};
709	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
710
711	#ifdef VBOX_STRICT
712	static const uint32_t g_aVmcsSegBase[] =
713	{
714	VMX_VMCS_GUEST_ES_BASE,
715	VMX_VMCS_GUEST_CS_BASE,
716	VMX_VMCS_GUEST_SS_BASE,
717	VMX_VMCS_GUEST_DS_BASE,
718	VMX_VMCS_GUEST_FS_BASE,
719	VMX_VMCS_GUEST_GS_BASE
720	};
721	static const uint32_t g_aVmcsSegSel[] =
722	{
723	VMX_VMCS16_GUEST_ES_SEL,
724	VMX_VMCS16_GUEST_CS_SEL,
725	VMX_VMCS16_GUEST_SS_SEL,
726	VMX_VMCS16_GUEST_DS_SEL,
727	VMX_VMCS16_GUEST_FS_SEL,
728	VMX_VMCS16_GUEST_GS_SEL
729	};
730	static const uint32_t g_aVmcsSegLimit[] =
731	{
732	VMX_VMCS32_GUEST_ES_LIMIT,
733	VMX_VMCS32_GUEST_CS_LIMIT,
734	VMX_VMCS32_GUEST_SS_LIMIT,
735	VMX_VMCS32_GUEST_DS_LIMIT,
736	VMX_VMCS32_GUEST_FS_LIMIT,
737	VMX_VMCS32_GUEST_GS_LIMIT
738	};
739	static const uint32_t g_aVmcsSegAttr[] =
740	{
741	VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS,
742	VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS,
743	VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS,
744	VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS,
745	VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS,
746	VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS
747	};
748	AssertCompile(RT_ELEMENTS(g_aVmcsSegSel) == X86_SREG_COUNT);
749	AssertCompile(RT_ELEMENTS(g_aVmcsSegLimit) == X86_SREG_COUNT);
750	AssertCompile(RT_ELEMENTS(g_aVmcsSegBase) == X86_SREG_COUNT);
751	AssertCompile(RT_ELEMENTS(g_aVmcsSegAttr) == X86_SREG_COUNT);
752	#endif /* VBOX_STRICT */
753
754	#ifdef HMVMX_USE_FUNCTION_TABLE
755	/**
756	* VMX_EXIT dispatch table.
757	*/
758	static const struct CLANG11NOTHROWWEIRDNESS { PFNVMXEXITHANDLER pfn; } g_aVMExitHandlers[VMX_EXIT_MAX + 1] =
759	{
760	/* 0 VMX_EXIT_XCPT_OR_NMI */ { hmR0VmxExitXcptOrNmi },
761	/* 1 VMX_EXIT_EXT_INT */ { hmR0VmxExitExtInt },
762	/* 2 VMX_EXIT_TRIPLE_FAULT */ { hmR0VmxExitTripleFault },
763	/* 3 VMX_EXIT_INIT_SIGNAL */ { hmR0VmxExitErrUnexpected },
764	/* 4 VMX_EXIT_SIPI */ { hmR0VmxExitErrUnexpected },
765	/* 5 VMX_EXIT_IO_SMI */ { hmR0VmxExitErrUnexpected },
766	/* 6 VMX_EXIT_SMI */ { hmR0VmxExitErrUnexpected },
767	/* 7 VMX_EXIT_INT_WINDOW */ { hmR0VmxExitIntWindow },
768	/* 8 VMX_EXIT_NMI_WINDOW */ { hmR0VmxExitNmiWindow },
769	/* 9 VMX_EXIT_TASK_SWITCH */ { hmR0VmxExitTaskSwitch },
770	/* 10 VMX_EXIT_CPUID */ { hmR0VmxExitCpuid },
771	/* 11 VMX_EXIT_GETSEC */ { hmR0VmxExitGetsec },
772	/* 12 VMX_EXIT_HLT */ { hmR0VmxExitHlt },
773	/* 13 VMX_EXIT_INVD */ { hmR0VmxExitInvd },
774	/* 14 VMX_EXIT_INVLPG */ { hmR0VmxExitInvlpg },
775	/* 15 VMX_EXIT_RDPMC */ { hmR0VmxExitRdpmc },
776	/* 16 VMX_EXIT_RDTSC */ { hmR0VmxExitRdtsc },
777	/* 17 VMX_EXIT_RSM */ { hmR0VmxExitErrUnexpected },
778	/* 18 VMX_EXIT_VMCALL */ { hmR0VmxExitVmcall },
779	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
780	/* 19 VMX_EXIT_VMCLEAR */ { hmR0VmxExitVmclear },
781	/* 20 VMX_EXIT_VMLAUNCH */ { hmR0VmxExitVmlaunch },
782	/* 21 VMX_EXIT_VMPTRLD */ { hmR0VmxExitVmptrld },
783	/* 22 VMX_EXIT_VMPTRST */ { hmR0VmxExitVmptrst },
784	/* 23 VMX_EXIT_VMREAD */ { hmR0VmxExitVmread },
785	/* 24 VMX_EXIT_VMRESUME */ { hmR0VmxExitVmresume },
786	/* 25 VMX_EXIT_VMWRITE */ { hmR0VmxExitVmwrite },
787	/* 26 VMX_EXIT_VMXOFF */ { hmR0VmxExitVmxoff },
788	/* 27 VMX_EXIT_VMXON */ { hmR0VmxExitVmxon },
789	#else
790	/* 19 VMX_EXIT_VMCLEAR */ { hmR0VmxExitSetPendingXcptUD },
791	/* 20 VMX_EXIT_VMLAUNCH */ { hmR0VmxExitSetPendingXcptUD },
792	/* 21 VMX_EXIT_VMPTRLD */ { hmR0VmxExitSetPendingXcptUD },
793	/* 22 VMX_EXIT_VMPTRST */ { hmR0VmxExitSetPendingXcptUD },
794	/* 23 VMX_EXIT_VMREAD */ { hmR0VmxExitSetPendingXcptUD },
795	/* 24 VMX_EXIT_VMRESUME */ { hmR0VmxExitSetPendingXcptUD },
796	/* 25 VMX_EXIT_VMWRITE */ { hmR0VmxExitSetPendingXcptUD },
797	/* 26 VMX_EXIT_VMXOFF */ { hmR0VmxExitSetPendingXcptUD },
798	/* 27 VMX_EXIT_VMXON */ { hmR0VmxExitSetPendingXcptUD },
799	#endif
800	/* 28 VMX_EXIT_MOV_CRX */ { hmR0VmxExitMovCRx },
801	/* 29 VMX_EXIT_MOV_DRX */ { hmR0VmxExitMovDRx },
802	/* 30 VMX_EXIT_IO_INSTR */ { hmR0VmxExitIoInstr },
803	/* 31 VMX_EXIT_RDMSR */ { hmR0VmxExitRdmsr },
804	/* 32 VMX_EXIT_WRMSR */ { hmR0VmxExitWrmsr },
805	/* 33 VMX_EXIT_ERR_INVALID_GUEST_STATE */ { hmR0VmxExitErrInvalidGuestState },
806	/* 34 VMX_EXIT_ERR_MSR_LOAD */ { hmR0VmxExitErrUnexpected },
807	/* 35 UNDEFINED */ { hmR0VmxExitErrUnexpected },
808	/* 36 VMX_EXIT_MWAIT */ { hmR0VmxExitMwait },
809	/* 37 VMX_EXIT_MTF */ { hmR0VmxExitMtf },
810	/* 38 UNDEFINED */ { hmR0VmxExitErrUnexpected },
811	/* 39 VMX_EXIT_MONITOR */ { hmR0VmxExitMonitor },
812	/* 40 VMX_EXIT_PAUSE */ { hmR0VmxExitPause },
813	/* 41 VMX_EXIT_ERR_MACHINE_CHECK */ { hmR0VmxExitErrUnexpected },
814	/* 42 UNDEFINED */ { hmR0VmxExitErrUnexpected },
815	/* 43 VMX_EXIT_TPR_BELOW_THRESHOLD */ { hmR0VmxExitTprBelowThreshold },
816	/* 44 VMX_EXIT_APIC_ACCESS */ { hmR0VmxExitApicAccess },
817	/* 45 VMX_EXIT_VIRTUALIZED_EOI */ { hmR0VmxExitErrUnexpected },
818	/* 46 VMX_EXIT_GDTR_IDTR_ACCESS */ { hmR0VmxExitErrUnexpected },
819	/* 47 VMX_EXIT_LDTR_TR_ACCESS */ { hmR0VmxExitErrUnexpected },
820	/* 48 VMX_EXIT_EPT_VIOLATION */ { hmR0VmxExitEptViolation },
821	/* 49 VMX_EXIT_EPT_MISCONFIG */ { hmR0VmxExitEptMisconfig },
822	/* 50 VMX_EXIT_INVEPT */ { hmR0VmxExitSetPendingXcptUD },
823	/* 51 VMX_EXIT_RDTSCP */ { hmR0VmxExitRdtscp },
824	/* 52 VMX_EXIT_PREEMPT_TIMER */ { hmR0VmxExitPreemptTimer },
825	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
826	/* 53 VMX_EXIT_INVVPID */ { hmR0VmxExitInvvpid },
827	#else
828	/* 53 VMX_EXIT_INVVPID */ { hmR0VmxExitSetPendingXcptUD },
829	#endif
830	/* 54 VMX_EXIT_WBINVD */ { hmR0VmxExitWbinvd },
831	/* 55 VMX_EXIT_XSETBV */ { hmR0VmxExitXsetbv },
832	/* 56 VMX_EXIT_APIC_WRITE */ { hmR0VmxExitErrUnexpected },
833	/* 57 VMX_EXIT_RDRAND */ { hmR0VmxExitErrUnexpected },
834	/* 58 VMX_EXIT_INVPCID */ { hmR0VmxExitInvpcid },
835	/* 59 VMX_EXIT_VMFUNC */ { hmR0VmxExitErrUnexpected },
836	/* 60 VMX_EXIT_ENCLS */ { hmR0VmxExitErrUnexpected },
837	/* 61 VMX_EXIT_RDSEED */ { hmR0VmxExitErrUnexpected },
838	/* 62 VMX_EXIT_PML_FULL */ { hmR0VmxExitErrUnexpected },
839	/* 63 VMX_EXIT_XSAVES */ { hmR0VmxExitErrUnexpected },
840	/* 64 VMX_EXIT_XRSTORS */ { hmR0VmxExitErrUnexpected },
841	/* 65 UNDEFINED */ { hmR0VmxExitErrUnexpected },
842	/* 66 VMX_EXIT_SPP_EVENT */ { hmR0VmxExitErrUnexpected },
843	/* 67 VMX_EXIT_UMWAIT */ { hmR0VmxExitErrUnexpected },
844	/* 68 VMX_EXIT_TPAUSE */ { hmR0VmxExitErrUnexpected },
845	/* 69 VMX_EXIT_LOADIWKEY */ { hmR0VmxExitErrUnexpected },
846	};
847	#endif /* HMVMX_USE_FUNCTION_TABLE */
848
849	#if defined(VBOX_STRICT) && defined(LOG_ENABLED)
850	static const char * const g_apszVmxInstrErrors[HMVMX_INSTR_ERROR_MAX + 1] =
851	{
852	/* 0 */ "(Not Used)",
853	/* 1 */ "VMCALL executed in VMX root operation.",
854	/* 2 */ "VMCLEAR with invalid physical address.",
855	/* 3 */ "VMCLEAR with VMXON pointer.",
856	/* 4 */ "VMLAUNCH with non-clear VMCS.",
857	/* 5 */ "VMRESUME with non-launched VMCS.",
858	/* 6 */ "VMRESUME after VMXOFF",
859	/* 7 */ "VM-entry with invalid control fields.",
860	/* 8 */ "VM-entry with invalid host state fields.",
861	/* 9 */ "VMPTRLD with invalid physical address.",
862	/* 10 */ "VMPTRLD with VMXON pointer.",
863	/* 11 */ "VMPTRLD with incorrect revision identifier.",
864	/* 12 */ "VMREAD/VMWRITE from/to unsupported VMCS component.",
865	/* 13 */ "VMWRITE to read-only VMCS component.",
866	/* 14 */ "(Not Used)",
867	/* 15 */ "VMXON executed in VMX root operation.",
868	/* 16 */ "VM-entry with invalid executive-VMCS pointer.",
869	/* 17 */ "VM-entry with non-launched executing VMCS.",
870	/* 18 */ "VM-entry with executive-VMCS pointer not VMXON pointer.",
871	/* 19 */ "VMCALL with non-clear VMCS.",
872	/* 20 */ "VMCALL with invalid VM-exit control fields.",
873	/* 21 */ "(Not Used)",
874	/* 22 */ "VMCALL with incorrect MSEG revision identifier.",
875	/* 23 */ "VMXOFF under dual monitor treatment of SMIs and SMM.",
876	/* 24 */ "VMCALL with invalid SMM-monitor features.",
877	/* 25 */ "VM-entry with invalid VM-execution control fields in executive VMCS.",
878	/* 26 */ "VM-entry with events blocked by MOV SS.",
879	/* 27 */ "(Not Used)",
880	/* 28 */ "Invalid operand to INVEPT/INVVPID."
881	};
882	#endif /* VBOX_STRICT && LOG_ENABLED */
883
884
885	/**
886	* Checks if the given MSR is part of the lastbranch-from-IP MSR stack.
887	* @returns @c true if it's part of LBR stack, @c false otherwise.
888	*
889	* @param pVM The cross context VM structure.
890	* @param idMsr The MSR.
891	* @param pidxMsr Where to store the index of the MSR in the LBR MSR array.
892	* Optional, can be NULL.
893	*
894	* @remarks Must only be called when LBR is enabled.
895	*/
896	DECL_FORCE_INLINE(bool) hmR0VmxIsLbrBranchFromMsr(PCVMCC pVM, uint32_t idMsr, uint32_t *pidxMsr)
897	{
898	Assert(pVM->hmr0.s.vmx.fLbr);
899	Assert(pVM->hmr0.s.vmx.idLbrFromIpMsrFirst);
900	uint32_t const cLbrStack = pVM->hmr0.s.vmx.idLbrFromIpMsrLast - pVM->hmr0.s.vmx.idLbrFromIpMsrFirst + 1;
901	uint32_t const idxMsr = idMsr - pVM->hmr0.s.vmx.idLbrFromIpMsrFirst;
902	if (idxMsr < cLbrStack)
903	{
904	if (pidxMsr)
905	*pidxMsr = idxMsr;
906	return true;
907	}
908	return false;
909	}
910
911
912	/**
913	* Checks if the given MSR is part of the lastbranch-to-IP MSR stack.
914	* @returns @c true if it's part of LBR stack, @c false otherwise.
915	*
916	* @param pVM The cross context VM structure.
917	* @param idMsr The MSR.
918	* @param pidxMsr Where to store the index of the MSR in the LBR MSR array.
919	* Optional, can be NULL.
920	*
921	* @remarks Must only be called when LBR is enabled and when lastbranch-to-IP MSRs
922	* are supported by the CPU (see hmR0VmxSetupLbrMsrRange).
923	*/
924	DECL_FORCE_INLINE(bool) hmR0VmxIsLbrBranchToMsr(PCVMCC pVM, uint32_t idMsr, uint32_t *pidxMsr)
925	{
926	Assert(pVM->hmr0.s.vmx.fLbr);
927	if (pVM->hmr0.s.vmx.idLbrToIpMsrFirst)
928	{
929	uint32_t const cLbrStack = pVM->hmr0.s.vmx.idLbrToIpMsrLast - pVM->hmr0.s.vmx.idLbrToIpMsrFirst + 1;
930	uint32_t const idxMsr = idMsr - pVM->hmr0.s.vmx.idLbrToIpMsrFirst;
931	if (idxMsr < cLbrStack)
932	{
933	if (pidxMsr)
934	*pidxMsr = idxMsr;
935	return true;
936	}
937	}
938	return false;
939	}
940
941
942	/**
943	* Gets the CR0 guest/host mask.
944	*
945	* These bits typically does not change through the lifetime of a VM. Any bit set in
946	* this mask is owned by the host/hypervisor and would cause a VM-exit when modified
947	* by the guest.
948	*
949	* @returns The CR0 guest/host mask.
950	* @param pVCpu The cross context virtual CPU structure.
951	*/
952	static uint64_t hmR0VmxGetFixedCr0Mask(PCVMCPUCC pVCpu)
953	{
954	/*
955	* Modifications to CR0 bits that VT-x ignores saving/restoring (CD, ET, NW) and
956	* to CR0 bits that we require for shadow paging (PG) by the guest must cause VM-exits.
957	*
958	* Furthermore, modifications to any bits that are reserved/unspecified currently
959	* by the Intel spec. must also cause a VM-exit. This prevents unpredictable behavior
960	* when future CPUs specify and use currently reserved/unspecified bits.
961	*/
962	/** @todo Avoid intercepting CR0.PE with unrestricted guest execution. Fix PGM
963	* enmGuestMode to be in-sync with the current mode. See @bugref{6398}
964	* and @bugref{6944}. */
965	PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
966	return ( X86_CR0_PE
967	\| X86_CR0_NE
968	\| (pVM->hmr0.s.fNestedPaging ? 0 : X86_CR0_WP)
969	\| X86_CR0_PG
970	\| VMX_EXIT_HOST_CR0_IGNORE_MASK);
971	}
972
973
974	/**
975	* Gets the CR4 guest/host mask.
976	*
977	* These bits typically does not change through the lifetime of a VM. Any bit set in
978	* this mask is owned by the host/hypervisor and would cause a VM-exit when modified
979	* by the guest.
980	*
981	* @returns The CR4 guest/host mask.
982	* @param pVCpu The cross context virtual CPU structure.
983	*/
984	static uint64_t hmR0VmxGetFixedCr4Mask(PCVMCPUCC pVCpu)
985	{
986	/*
987	* We construct a mask of all CR4 bits that the guest can modify without causing
988	* a VM-exit. Then invert this mask to obtain all CR4 bits that should cause
989	* a VM-exit when the guest attempts to modify them when executing using
990	* hardware-assisted VMX.
991	*
992	* When a feature is not exposed to the guest (and may be present on the host),
993	* we want to intercept guest modifications to the bit so we can emulate proper
994	* behavior (e.g., #GP).
995	*
996	* Furthermore, only modifications to those bits that don't require immediate
997	* emulation is allowed. For e.g., PCIDE is excluded because the behavior
998	* depends on CR3 which might not always be the guest value while executing
999	* using hardware-assisted VMX.
1000	*/
1001	PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
1002	bool const fFsGsBase = pVM->cpum.ro.GuestFeatures.fFsGsBase;
1003	bool const fXSaveRstor = pVM->cpum.ro.GuestFeatures.fXSaveRstor;
1004	bool const fFxSaveRstor = pVM->cpum.ro.GuestFeatures.fFxSaveRstor;
1005
1006	/*
1007	* Paranoia.
1008	* Ensure features exposed to the guest are present on the host.
1009	*/
1010	Assert(!fFsGsBase \|\| pVM->cpum.ro.HostFeatures.fFsGsBase);
1011	Assert(!fXSaveRstor \|\| pVM->cpum.ro.HostFeatures.fXSaveRstor);
1012	Assert(!fFxSaveRstor \|\| pVM->cpum.ro.HostFeatures.fFxSaveRstor);
1013
1014	uint64_t const fGstMask = ( X86_CR4_PVI
1015	\| X86_CR4_TSD
1016	\| X86_CR4_DE
1017	\| X86_CR4_MCE
1018	\| X86_CR4_PCE
1019	\| X86_CR4_OSXMMEEXCPT
1020	\| (fFsGsBase ? X86_CR4_FSGSBASE : 0)
1021	\| (fXSaveRstor ? X86_CR4_OSXSAVE : 0)
1022	\| (fFxSaveRstor ? X86_CR4_OSFXSR : 0));
1023	return ~fGstMask;
1024	}
1025
1026
1027	/**
1028	* Gets the active (in use) VMCS info. object for the specified VCPU.
1029	*
1030	* This is either the guest or nested-guest VMCS info. and need not necessarily
1031	* pertain to the "current" VMCS (in the VMX definition of the term). For instance,
1032	* if the VM-entry failed due to an invalid-guest state, we may have "cleared" the
1033	* current VMCS while returning to ring-3. However, the VMCS info. object for that
1034	* VMCS would still be active and returned here so that we could dump the VMCS
1035	* fields to ring-3 for diagnostics. This function is thus only used to
1036	* distinguish between the nested-guest or guest VMCS.
1037	*
1038	* @returns The active VMCS information.
1039	* @param pVCpu The cross context virtual CPU structure.
1040	*
1041	* @thread EMT.
1042	* @remarks This function may be called with preemption or interrupts disabled!
1043	*/
1044	DECLINLINE(PVMXVMCSINFO) hmGetVmxActiveVmcsInfo(PVMCPUCC pVCpu)
1045	{
1046	if (!pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs)
1047	return &pVCpu->hmr0.s.vmx.VmcsInfo;
1048	return &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
1049	}
1050
1051
1052	/**
1053	* Returns whether the VM-exit MSR-store area differs from the VM-exit MSR-load
1054	* area.
1055	*
1056	* @returns @c true if it's different, @c false otherwise.
1057	* @param pVmcsInfo The VMCS info. object.
1058	*/
1059	DECL_FORCE_INLINE(bool) hmR0VmxIsSeparateExitMsrStoreAreaVmcs(PCVMXVMCSINFO pVmcsInfo)
1060	{
1061	return RT_BOOL( pVmcsInfo->pvGuestMsrStore != pVmcsInfo->pvGuestMsrLoad
1062	&& pVmcsInfo->pvGuestMsrStore);
1063	}
1064
1065
1066	/**
1067	* Sets the given Processor-based VM-execution controls.
1068	*
1069	* @param pVmxTransient The VMX-transient structure.
1070	* @param uProcCtls The Processor-based VM-execution controls to set.
1071	*/
1072	static void hmR0VmxSetProcCtlsVmcs(PVMXTRANSIENT pVmxTransient, uint32_t uProcCtls)
1073	{
1074	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1075	if ((pVmcsInfo->u32ProcCtls & uProcCtls) != uProcCtls)
1076	{
1077	pVmcsInfo->u32ProcCtls \|= uProcCtls;
1078	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
1079	AssertRC(rc);
1080	}
1081	}
1082
1083
1084	/**
1085	* Removes the given Processor-based VM-execution controls.
1086	*
1087	* @param pVCpu The cross context virtual CPU structure.
1088	* @param pVmxTransient The VMX-transient structure.
1089	* @param uProcCtls The Processor-based VM-execution controls to remove.
1090	*
1091	* @remarks When executing a nested-guest, this will not remove any of the specified
1092	* controls if the nested hypervisor has set any one of them.
1093	*/
1094	static void hmR0VmxRemoveProcCtlsVmcs(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uProcCtls)
1095	{
1096	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1097	if (pVmcsInfo->u32ProcCtls & uProcCtls)
1098	{
1099	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1100	if ( !pVmxTransient->fIsNestedGuest
1101	\|\| !CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, uProcCtls))
1102	#else
1103	NOREF(pVCpu);
1104	if (!pVmxTransient->fIsNestedGuest)
1105	#endif
1106	{
1107	pVmcsInfo->u32ProcCtls &= ~uProcCtls;
1108	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
1109	AssertRC(rc);
1110	}
1111	}
1112	}
1113
1114
1115	/**
1116	* Sets the TSC offset for the current VMCS.
1117	*
1118	* @param uTscOffset The TSC offset to set.
1119	* @param pVmcsInfo The VMCS info. object.
1120	*/
1121	static void hmR0VmxSetTscOffsetVmcs(PVMXVMCSINFO pVmcsInfo, uint64_t uTscOffset)
1122	{
1123	if (pVmcsInfo->u64TscOffset != uTscOffset)
1124	{
1125	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, uTscOffset);
1126	AssertRC(rc);
1127	pVmcsInfo->u64TscOffset = uTscOffset;
1128	}
1129	}
1130
1131
1132	/**
1133	* Adds one or more exceptions to the exception bitmap and commits it to the current
1134	* VMCS.
1135	*
1136	* @param pVmxTransient The VMX-transient structure.
1137	* @param uXcptMask The exception(s) to add.
1138	*/
1139	static void hmR0VmxAddXcptInterceptMask(PCVMXTRANSIENT pVmxTransient, uint32_t uXcptMask)
1140	{
1141	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1142	uint32_t uXcptBitmap = pVmcsInfo->u32XcptBitmap;
1143	if ((uXcptBitmap & uXcptMask) != uXcptMask)
1144	{
1145	uXcptBitmap \|= uXcptMask;
1146	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
1147	AssertRC(rc);
1148	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
1149	}
1150	}
1151
1152
1153	/**
1154	* Adds an exception to the exception bitmap and commits it to the current VMCS.
1155	*
1156	* @param pVmxTransient The VMX-transient structure.
1157	* @param uXcpt The exception to add.
1158	*/
1159	static void hmR0VmxAddXcptIntercept(PCVMXTRANSIENT pVmxTransient, uint8_t uXcpt)
1160	{
1161	Assert(uXcpt <= X86_XCPT_LAST);
1162	hmR0VmxAddXcptInterceptMask(pVmxTransient, RT_BIT_32(uXcpt));
1163	}
1164
1165
1166	/**
1167	* Remove one or more exceptions from the exception bitmap and commits it to the
1168	* current VMCS.
1169	*
1170	* This takes care of not removing the exception intercept if a nested-guest
1171	* requires the exception to be intercepted.
1172	*
1173	* @returns VBox status code.
1174	* @param pVCpu The cross context virtual CPU structure.
1175	* @param pVmxTransient The VMX-transient structure.
1176	* @param uXcptMask The exception(s) to remove.
1177	*/
1178	static int hmR0VmxRemoveXcptInterceptMask(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t uXcptMask)
1179	{
1180	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
1181	uint32_t u32XcptBitmap = pVmcsInfo->u32XcptBitmap;
1182	if (u32XcptBitmap & uXcptMask)
1183	{
1184	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1185	if (!pVmxTransient->fIsNestedGuest)
1186	{ /* likely */ }
1187	else
1188	uXcptMask &= ~pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs.u32XcptBitmap;
1189	#endif
1190	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
1191	uXcptMask &= ~( RT_BIT(X86_XCPT_BP)
1192	\| RT_BIT(X86_XCPT_DE)
1193	\| RT_BIT(X86_XCPT_NM)
1194	\| RT_BIT(X86_XCPT_TS)
1195	\| RT_BIT(X86_XCPT_UD)
1196	\| RT_BIT(X86_XCPT_NP)
1197	\| RT_BIT(X86_XCPT_SS)
1198	\| RT_BIT(X86_XCPT_GP)
1199	\| RT_BIT(X86_XCPT_PF)
1200	\| RT_BIT(X86_XCPT_MF));
1201	#elif defined(HMVMX_ALWAYS_TRAP_PF)
1202	uXcptMask &= ~RT_BIT(X86_XCPT_PF);
1203	#endif
1204	if (uXcptMask)
1205	{
1206	/* Validate we are not removing any essential exception intercepts. */
1207	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging \|\| !(uXcptMask & RT_BIT(X86_XCPT_PF)));
1208	NOREF(pVCpu);
1209	Assert(!(uXcptMask & RT_BIT(X86_XCPT_DB)));
1210	Assert(!(uXcptMask & RT_BIT(X86_XCPT_AC)));
1211
1212	/* Remove it from the exception bitmap. */
1213	u32XcptBitmap &= ~uXcptMask;
1214
1215	/* Commit and update the cache if necessary. */
1216	if (pVmcsInfo->u32XcptBitmap != u32XcptBitmap)
1217	{
1218	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
1219	AssertRC(rc);
1220	pVmcsInfo->u32XcptBitmap = u32XcptBitmap;
1221	}
1222	}
1223	}
1224	return VINF_SUCCESS;
1225	}
1226
1227
1228	/**
1229	* Remove an exceptions from the exception bitmap and commits it to the current
1230	* VMCS.
1231	*
1232	* @returns VBox status code.
1233	* @param pVCpu The cross context virtual CPU structure.
1234	* @param pVmxTransient The VMX-transient structure.
1235	* @param uXcpt The exception to remove.
1236	*/
1237	static int hmR0VmxRemoveXcptIntercept(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint8_t uXcpt)
1238	{
1239	return hmR0VmxRemoveXcptInterceptMask(pVCpu, pVmxTransient, RT_BIT(uXcpt));
1240	}
1241
1242
1243	/**
1244	* Loads the 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 hmR0VmxLoadVmcs(PVMXVMCSINFO pVmcsInfo)
1252	{
1253	Assert(pVmcsInfo->HCPhysVmcs != 0 && pVmcsInfo->HCPhysVmcs != NIL_RTHCPHYS);
1254	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1255
1256	int rc = VMXLoadVmcs(pVmcsInfo->HCPhysVmcs);
1257	if (RT_SUCCESS(rc))
1258	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_CURRENT;
1259	return rc;
1260	}
1261
1262
1263	/**
1264	* Clears the 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 hmR0VmxClearVmcs(PVMXVMCSINFO pVmcsInfo)
1272	{
1273	Assert(pVmcsInfo->HCPhysVmcs != 0 && pVmcsInfo->HCPhysVmcs != NIL_RTHCPHYS);
1274	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1275
1276	int rc = VMXClearVmcs(pVmcsInfo->HCPhysVmcs);
1277	if (RT_SUCCESS(rc))
1278	pVmcsInfo->fVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
1279	return rc;
1280	}
1281
1282
1283	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1284	/**
1285	* Loads the shadow VMCS specified by the VMCS info. object.
1286	*
1287	* @returns VBox status code.
1288	* @param pVmcsInfo The VMCS info. object.
1289	*
1290	* @remarks Can be called with interrupts disabled.
1291	*/
1292	static int hmR0VmxLoadShadowVmcs(PVMXVMCSINFO pVmcsInfo)
1293	{
1294	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1295	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
1296
1297	int rc = VMXLoadVmcs(pVmcsInfo->HCPhysShadowVmcs);
1298	if (RT_SUCCESS(rc))
1299	pVmcsInfo->fShadowVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_CURRENT;
1300	return rc;
1301	}
1302
1303
1304	/**
1305	* Clears the shadow VMCS specified by the VMCS info. object.
1306	*
1307	* @returns VBox status code.
1308	* @param pVmcsInfo The VMCS info. object.
1309	*
1310	* @remarks Can be called with interrupts disabled.
1311	*/
1312	static int hmR0VmxClearShadowVmcs(PVMXVMCSINFO pVmcsInfo)
1313	{
1314	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1315	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
1316
1317	int rc = VMXClearVmcs(pVmcsInfo->HCPhysShadowVmcs);
1318	if (RT_SUCCESS(rc))
1319	pVmcsInfo->fShadowVmcsState = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
1320	return rc;
1321	}
1322
1323
1324	/**
1325	* Switches from and to the specified VMCSes.
1326	*
1327	* @returns VBox status code.
1328	* @param pVmcsInfoFrom The VMCS info. object we are switching from.
1329	* @param pVmcsInfoTo The VMCS info. object we are switching to.
1330	*
1331	* @remarks Called with interrupts disabled.
1332	*/
1333	static int hmR0VmxSwitchVmcs(PVMXVMCSINFO pVmcsInfoFrom, PVMXVMCSINFO pVmcsInfoTo)
1334	{
1335	/*
1336	* Clear the VMCS we are switching out if it has not already been cleared.
1337	* This will sync any CPU internal data back to the VMCS.
1338	*/
1339	if (pVmcsInfoFrom->fVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
1340	{
1341	int rc = hmR0VmxClearVmcs(pVmcsInfoFrom);
1342	if (RT_SUCCESS(rc))
1343	{
1344	/*
1345	* The shadow VMCS, if any, would not be active at this point since we
1346	* would have cleared it while importing the virtual hardware-virtualization
1347	* state as part the VMLAUNCH/VMRESUME VM-exit. Hence, there's no need to
1348	* clear the shadow VMCS here, just assert for safety.
1349	*/
1350	Assert(!pVmcsInfoFrom->pvShadowVmcs \|\| pVmcsInfoFrom->fShadowVmcsState == VMX_V_VMCS_LAUNCH_STATE_CLEAR);
1351	}
1352	else
1353	return rc;
1354	}
1355
1356	/*
1357	* Clear the VMCS we are switching to if it has not already been cleared.
1358	* This will initialize the VMCS launch state to "clear" required for loading it.
1359	*
1360	* See Intel spec. 31.6 "Preparation And Launching A Virtual Machine".
1361	*/
1362	if (pVmcsInfoTo->fVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
1363	{
1364	int rc = hmR0VmxClearVmcs(pVmcsInfoTo);
1365	if (RT_SUCCESS(rc))
1366	{ /* likely */ }
1367	else
1368	return rc;
1369	}
1370
1371	/*
1372	* Finally, load the VMCS we are switching to.
1373	*/
1374	return hmR0VmxLoadVmcs(pVmcsInfoTo);
1375	}
1376
1377
1378	/**
1379	* Switches between the guest VMCS and the nested-guest VMCS as specified by the
1380	* caller.
1381	*
1382	* @returns VBox status code.
1383	* @param pVCpu The cross context virtual CPU structure.
1384	* @param fSwitchToNstGstVmcs Whether to switch to the nested-guest VMCS (pass
1385	* true) or guest VMCS (pass false).
1386	*/
1387	static int hmR0VmxSwitchToGstOrNstGstVmcs(PVMCPUCC pVCpu, bool fSwitchToNstGstVmcs)
1388	{
1389	/* Ensure we have synced everything from the guest-CPU context to the VMCS before switching. */
1390	HMVMX_CPUMCTX_ASSERT(pVCpu, HMVMX_CPUMCTX_EXTRN_ALL);
1391
1392	PVMXVMCSINFO pVmcsInfoFrom;
1393	PVMXVMCSINFO pVmcsInfoTo;
1394	if (fSwitchToNstGstVmcs)
1395	{
1396	pVmcsInfoFrom = &pVCpu->hmr0.s.vmx.VmcsInfo;
1397	pVmcsInfoTo = &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
1398	}
1399	else
1400	{
1401	pVmcsInfoFrom = &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
1402	pVmcsInfoTo = &pVCpu->hmr0.s.vmx.VmcsInfo;
1403	}
1404
1405	/*
1406	* Disable interrupts to prevent being preempted while we switch the current VMCS as the
1407	* preemption hook code path acquires the current VMCS.
1408	*/
1409	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1410
1411	int rc = hmR0VmxSwitchVmcs(pVmcsInfoFrom, pVmcsInfoTo);
1412	if (RT_SUCCESS(rc))
1413	{
1414	pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs = fSwitchToNstGstVmcs;
1415	pVCpu->hm.s.vmx.fSwitchedToNstGstVmcsCopyForRing3 = fSwitchToNstGstVmcs;
1416
1417	/*
1418	* If we are switching to a VMCS that was executed on a different host CPU or was
1419	* never executed before, flag that we need to export the host state before executing
1420	* guest/nested-guest code using hardware-assisted VMX.
1421	*
1422	* This could probably be done in a preemptible context since the preemption hook
1423	* will flag the necessary change in host context. However, since preemption is
1424	* already disabled and to avoid making assumptions about host specific code in
1425	* RTMpCpuId when called with preemption enabled, we'll do this while preemption is
1426	* disabled.
1427	*/
1428	if (pVmcsInfoTo->idHostCpuState == RTMpCpuId())
1429	{ /* likely */ }
1430	else
1431	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE);
1432
1433	ASMSetFlags(fEFlags);
1434
1435	/*
1436	* We use a different VM-exit MSR-store areas for the guest and nested-guest. Hence,
1437	* flag that we need to update the host MSR values there. Even if we decide in the
1438	* future to share the VM-exit MSR-store area page between the guest and nested-guest,
1439	* if its content differs, we would have to update the host MSRs anyway.
1440	*/
1441	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = false;
1442	}
1443	else
1444	ASMSetFlags(fEFlags);
1445	return rc;
1446	}
1447	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
1448
1449
1450	/**
1451	* Updates the VM's last error record.
1452	*
1453	* If there was a VMX instruction error, reads the error data from the VMCS and
1454	* updates VCPU's last error record as well.
1455	*
1456	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
1457	* Can be NULL if @a rc is not VERR_VMX_UNABLE_TO_START_VM or
1458	* VERR_VMX_INVALID_VMCS_FIELD.
1459	* @param rc The error code.
1460	*/
1461	static void hmR0VmxUpdateErrorRecord(PVMCPUCC pVCpu, int rc)
1462	{
1463	if ( rc == VERR_VMX_INVALID_VMCS_FIELD
1464	\|\| rc == VERR_VMX_UNABLE_TO_START_VM)
1465	{
1466	AssertPtrReturnVoid(pVCpu);
1467	VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
1468	}
1469	pVCpu->CTX_SUFF(pVM)->hm.s.ForR3.rcInit = rc;
1470	}
1471
1472
1473	#ifdef VBOX_STRICT
1474	/**
1475	* Reads the VM-entry interruption-information field from the VMCS into the VMX
1476	* transient structure.
1477	*
1478	* @param pVmxTransient The VMX-transient structure.
1479	*/
1480	DECLINLINE(void) hmR0VmxReadEntryIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
1481	{
1482	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &pVmxTransient->uEntryIntInfo);
1483	AssertRC(rc);
1484	}
1485
1486
1487	/**
1488	* Reads the VM-entry exception error code field from the VMCS into
1489	* the VMX transient structure.
1490	*
1491	* @param pVmxTransient The VMX-transient structure.
1492	*/
1493	DECLINLINE(void) hmR0VmxReadEntryXcptErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1494	{
1495	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &pVmxTransient->uEntryXcptErrorCode);
1496	AssertRC(rc);
1497	}
1498
1499
1500	/**
1501	* Reads the VM-entry exception error code field from the VMCS into
1502	* the VMX transient structure.
1503	*
1504	* @param pVmxTransient The VMX-transient structure.
1505	*/
1506	DECLINLINE(void) hmR0VmxReadEntryInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
1507	{
1508	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &pVmxTransient->cbEntryInstr);
1509	AssertRC(rc);
1510	}
1511	#endif /* VBOX_STRICT */
1512
1513
1514	/**
1515	* Reads the VM-exit interruption-information field from the VMCS into the VMX
1516	* transient structure.
1517	*
1518	* @param pVmxTransient The VMX-transient structure.
1519	*/
1520	DECLINLINE(void) hmR0VmxReadExitIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
1521	{
1522	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INTERRUPTION_INFO))
1523	{
1524	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
1525	AssertRC(rc);
1526	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INTERRUPTION_INFO;
1527	}
1528	}
1529
1530
1531	/**
1532	* Reads the VM-exit interruption error code from the VMCS into the VMX
1533	* transient structure.
1534	*
1535	* @param pVmxTransient The VMX-transient structure.
1536	*/
1537	DECLINLINE(void) hmR0VmxReadExitIntErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1538	{
1539	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE))
1540	{
1541	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
1542	AssertRC(rc);
1543	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE;
1544	}
1545	}
1546
1547
1548	/**
1549	* Reads the VM-exit instruction length field from the VMCS into the VMX
1550	* transient structure.
1551	*
1552	* @param pVmxTransient The VMX-transient structure.
1553	*/
1554	DECLINLINE(void) hmR0VmxReadExitInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
1555	{
1556	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INSTR_LEN))
1557	{
1558	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbExitInstr);
1559	AssertRC(rc);
1560	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INSTR_LEN;
1561	}
1562	}
1563
1564
1565	/**
1566	* Reads the VM-exit instruction-information field from the VMCS into
1567	* the VMX transient structure.
1568	*
1569	* @param pVmxTransient The VMX-transient structure.
1570	*/
1571	DECLINLINE(void) hmR0VmxReadExitInstrInfoVmcs(PVMXTRANSIENT pVmxTransient)
1572	{
1573	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_INSTR_INFO))
1574	{
1575	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
1576	AssertRC(rc);
1577	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_INSTR_INFO;
1578	}
1579	}
1580
1581
1582	/**
1583	* Reads the Exit Qualification from the VMCS into the VMX transient structure.
1584	*
1585	* @param pVmxTransient The VMX-transient structure.
1586	*/
1587	DECLINLINE(void) hmR0VmxReadExitQualVmcs(PVMXTRANSIENT pVmxTransient)
1588	{
1589	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_EXIT_QUALIFICATION))
1590	{
1591	int rc = VMXReadVmcsNw(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQual);
1592	AssertRC(rc);
1593	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_QUALIFICATION;
1594	}
1595	}
1596
1597
1598	/**
1599	* Reads the Guest-linear address from the VMCS into the VMX transient structure.
1600	*
1601	* @param pVmxTransient The VMX-transient structure.
1602	*/
1603	DECLINLINE(void) hmR0VmxReadGuestLinearAddrVmcs(PVMXTRANSIENT pVmxTransient)
1604	{
1605	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_LINEAR_ADDR))
1606	{
1607	int rc = VMXReadVmcsNw(VMX_VMCS_RO_GUEST_LINEAR_ADDR, &pVmxTransient->uGuestLinearAddr);
1608	AssertRC(rc);
1609	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_LINEAR_ADDR;
1610	}
1611	}
1612
1613
1614	/**
1615	* Reads the Guest-physical address from the VMCS into the VMX transient structure.
1616	*
1617	* @param pVmxTransient The VMX-transient structure.
1618	*/
1619	DECLINLINE(void) hmR0VmxReadGuestPhysicalAddrVmcs(PVMXTRANSIENT pVmxTransient)
1620	{
1621	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_PHYSICAL_ADDR))
1622	{
1623	int rc = VMXReadVmcs64(VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL, &pVmxTransient->uGuestPhysicalAddr);
1624	AssertRC(rc);
1625	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_PHYSICAL_ADDR;
1626	}
1627	}
1628
1629	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1630	/**
1631	* Reads the Guest pending-debug exceptions from the VMCS into the VMX transient
1632	* structure.
1633	*
1634	* @param pVmxTransient The VMX-transient structure.
1635	*/
1636	DECLINLINE(void) hmR0VmxReadGuestPendingDbgXctps(PVMXTRANSIENT pVmxTransient)
1637	{
1638	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_GUEST_PENDING_DBG_XCPTS))
1639	{
1640	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, &pVmxTransient->uGuestPendingDbgXcpts);
1641	AssertRC(rc);
1642	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_GUEST_PENDING_DBG_XCPTS;
1643	}
1644	}
1645	#endif
1646
1647	/**
1648	* Reads the IDT-vectoring information field from the VMCS into the VMX
1649	* transient structure.
1650	*
1651	* @param pVmxTransient The VMX-transient structure.
1652	*
1653	* @remarks No-long-jump zone!!!
1654	*/
1655	DECLINLINE(void) hmR0VmxReadIdtVectoringInfoVmcs(PVMXTRANSIENT pVmxTransient)
1656	{
1657	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_IDT_VECTORING_INFO))
1658	{
1659	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_INFO, &pVmxTransient->uIdtVectoringInfo);
1660	AssertRC(rc);
1661	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_IDT_VECTORING_INFO;
1662	}
1663	}
1664
1665
1666	/**
1667	* Reads the IDT-vectoring error code from the VMCS into the VMX
1668	* transient structure.
1669	*
1670	* @param pVmxTransient The VMX-transient structure.
1671	*/
1672	DECLINLINE(void) hmR0VmxReadIdtVectoringErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
1673	{
1674	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_READ_IDT_VECTORING_ERROR_CODE))
1675	{
1676	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
1677	AssertRC(rc);
1678	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_IDT_VECTORING_ERROR_CODE;
1679	}
1680	}
1681
1682	#ifdef HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
1683	/**
1684	* Reads all relevant read-only VMCS fields into the VMX transient structure.
1685	*
1686	* @param pVmxTransient The VMX-transient structure.
1687	*/
1688	static void hmR0VmxReadAllRoFieldsVmcs(PVMXTRANSIENT pVmxTransient)
1689	{
1690	int rc = VMXReadVmcsNw(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQual);
1691	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbExitInstr);
1692	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
1693	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_INFO, &pVmxTransient->uIdtVectoringInfo);
1694	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_IDT_VECTORING_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
1695	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
1696	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
1697	rc \|= VMXReadVmcsNw(VMX_VMCS_RO_GUEST_LINEAR_ADDR, &pVmxTransient->uGuestLinearAddr);
1698	rc \|= VMXReadVmcs64(VMX_VMCS64_RO_GUEST_PHYS_ADDR_FULL, &pVmxTransient->uGuestPhysicalAddr);
1699	AssertRC(rc);
1700	pVmxTransient->fVmcsFieldsRead \|= HMVMX_READ_EXIT_QUALIFICATION
1701	\| HMVMX_READ_EXIT_INSTR_LEN
1702	\| HMVMX_READ_EXIT_INSTR_INFO
1703	\| HMVMX_READ_IDT_VECTORING_INFO
1704	\| HMVMX_READ_IDT_VECTORING_ERROR_CODE
1705	\| HMVMX_READ_EXIT_INTERRUPTION_INFO
1706	\| HMVMX_READ_EXIT_INTERRUPTION_ERROR_CODE
1707	\| HMVMX_READ_GUEST_LINEAR_ADDR
1708	\| HMVMX_READ_GUEST_PHYSICAL_ADDR;
1709	}
1710	#endif
1711
1712	/**
1713	* Enters VMX root mode operation on the current CPU.
1714	*
1715	* @returns VBox status code.
1716	* @param pHostCpu The HM physical-CPU structure.
1717	* @param pVM The cross context VM structure. Can be
1718	* NULL, after a resume.
1719	* @param HCPhysCpuPage Physical address of the VMXON region.
1720	* @param pvCpuPage Pointer to the VMXON region.
1721	*/
1722	static int hmR0VmxEnterRootMode(PHMPHYSCPU pHostCpu, PVMCC pVM, RTHCPHYS HCPhysCpuPage, void *pvCpuPage)
1723	{
1724	Assert(pHostCpu);
1725	Assert(HCPhysCpuPage && HCPhysCpuPage != NIL_RTHCPHYS);
1726	Assert(RT_ALIGN_T(HCPhysCpuPage, _4K, RTHCPHYS) == HCPhysCpuPage);
1727	Assert(pvCpuPage);
1728	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1729
1730	if (pVM)
1731	{
1732	/* Write the VMCS revision identifier to the VMXON region. */
1733	(uint32_t )pvCpuPage = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_ID);
1734	}
1735
1736	/* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with CR4. */
1737	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1738
1739	/* Enable the VMX bit in CR4 if necessary. */
1740	RTCCUINTREG const uOldCr4 = SUPR0ChangeCR4(X86_CR4_VMXE, RTCCUINTREG_MAX);
1741
1742	/* Record whether VMXE was already prior to us enabling it above. */
1743	pHostCpu->fVmxeAlreadyEnabled = RT_BOOL(uOldCr4 & X86_CR4_VMXE);
1744
1745	/* Enter VMX root mode. */
1746	int rc = VMXEnable(HCPhysCpuPage);
1747	if (RT_FAILURE(rc))
1748	{
1749	/* Restore CR4.VMXE if it was not set prior to our attempt to set it above. */
1750	if (!pHostCpu->fVmxeAlreadyEnabled)
1751	SUPR0ChangeCR4(0 /* fOrMask */, ~(uint64_t)X86_CR4_VMXE);
1752
1753	if (pVM)
1754	pVM->hm.s.ForR3.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
1755	}
1756
1757	/* Restore interrupts. */
1758	ASMSetFlags(fEFlags);
1759	return rc;
1760	}
1761
1762
1763	/**
1764	* Exits VMX root mode operation on the current CPU.
1765	*
1766	* @returns VBox status code.
1767	* @param pHostCpu The HM physical-CPU structure.
1768	*/
1769	static int hmR0VmxLeaveRootMode(PHMPHYSCPU pHostCpu)
1770	{
1771	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1772
1773	/* Paranoid: Disable interrupts as, in theory, interrupts handlers might mess with CR4. */
1774	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
1775
1776	/* If we're for some reason not in VMX root mode, then don't leave it. */
1777	RTCCUINTREG const uHostCr4 = ASMGetCR4();
1778
1779	int rc;
1780	if (uHostCr4 & X86_CR4_VMXE)
1781	{
1782	/* Exit VMX root mode and clear the VMX bit in CR4. */
1783	VMXDisable();
1784
1785	/* Clear CR4.VMXE only if it was clear prior to use setting it. */
1786	if (!pHostCpu->fVmxeAlreadyEnabled)
1787	SUPR0ChangeCR4(0 /* fOrMask */, ~(uint64_t)X86_CR4_VMXE);
1788
1789	rc = VINF_SUCCESS;
1790	}
1791	else
1792	rc = VERR_VMX_NOT_IN_VMX_ROOT_MODE;
1793
1794	/* Restore interrupts. */
1795	ASMSetFlags(fEFlags);
1796	return rc;
1797	}
1798
1799
1800	/**
1801	* Allocates pages specified as specified by an array of VMX page allocation info
1802	* objects.
1803	*
1804	* The pages contents are zero'd after allocation.
1805	*
1806	* @returns VBox status code.
1807	* @param phMemObj Where to return the handle to the allocation.
1808	* @param paAllocInfo The pointer to the first element of the VMX
1809	* page-allocation info object array.
1810	* @param cEntries The number of elements in the @a paAllocInfo array.
1811	*/
1812	static int hmR0VmxPagesAllocZ(PRTR0MEMOBJ phMemObj, PVMXPAGEALLOCINFO paAllocInfo, uint32_t cEntries)
1813	{
1814	*phMemObj = NIL_RTR0MEMOBJ;
1815
1816	/* Figure out how many pages to allocate. */
1817	uint32_t cPages = 0;
1818	for (uint32_t iPage = 0; iPage < cEntries; iPage++)
1819	cPages += !!paAllocInfo[iPage].fValid;
1820
1821	/* Allocate the pages. */
1822	if (cPages)
1823	{
1824	size_t const cbPages = cPages << PAGE_SHIFT;
1825	int rc = RTR0MemObjAllocPage(phMemObj, cbPages, false /* fExecutable */);
1826	if (RT_FAILURE(rc))
1827	return rc;
1828
1829	/* Zero the contents and assign each page to the corresponding VMX page-allocation entry. */
1830	void pvFirstPage = RTR0MemObjAddress(phMemObj);
1831	RT_BZERO(pvFirstPage, cbPages);
1832
1833	uint32_t iPage = 0;
1834	for (uint32_t i = 0; i < cEntries; i++)
1835	if (paAllocInfo[i].fValid)
1836	{
1837	RTHCPHYS const HCPhysPage = RTR0MemObjGetPagePhysAddr(*phMemObj, iPage);
1838	void pvPage = (void )((uintptr_t)pvFirstPage + (iPage << X86_PAGE_4K_SHIFT));
1839	Assert(HCPhysPage && HCPhysPage != NIL_RTHCPHYS);
1840	AssertPtr(pvPage);
1841
1842	Assert(paAllocInfo[iPage].pHCPhys);
1843	Assert(paAllocInfo[iPage].ppVirt);
1844	*paAllocInfo[iPage].pHCPhys = HCPhysPage;
1845	*paAllocInfo[iPage].ppVirt = pvPage;
1846
1847	/* Move to next page. */
1848	++iPage;
1849	}
1850
1851	/* Make sure all valid (requested) pages have been assigned. */
1852	Assert(iPage == cPages);
1853	}
1854	return VINF_SUCCESS;
1855	}
1856
1857
1858	/**
1859	* Frees pages allocated using hmR0VmxPagesAllocZ.
1860	*
1861	* @param phMemObj Pointer to the memory object handle. Will be set to
1862	* NIL.
1863	*/
1864	DECL_FORCE_INLINE(void) hmR0VmxPagesFree(PRTR0MEMOBJ phMemObj)
1865	{
1866	/* We can cleanup wholesale since it's all one allocation. */
1867	if (*phMemObj != NIL_RTR0MEMOBJ)
1868	{
1869	RTR0MemObjFree(phMemObj, true / fFreeMappings */);
1870	*phMemObj = NIL_RTR0MEMOBJ;
1871	}
1872	}
1873
1874
1875	/**
1876	* Initializes a VMCS info. object.
1877	*
1878	* @param pVmcsInfo The VMCS info. object.
1879	* @param pVmcsInfoShared The VMCS info. object shared with ring-3.
1880	*/
1881	static void hmR0VmxVmcsInfoInit(PVMXVMCSINFO pVmcsInfo, PVMXVMCSINFOSHARED pVmcsInfoShared)
1882	{
1883	RT_ZERO(*pVmcsInfo);
1884	RT_ZERO(*pVmcsInfoShared);
1885
1886	pVmcsInfo->pShared = pVmcsInfoShared;
1887	Assert(pVmcsInfo->hMemObj == NIL_RTR0MEMOBJ);
1888	pVmcsInfo->HCPhysVmcs = NIL_RTHCPHYS;
1889	pVmcsInfo->HCPhysShadowVmcs = NIL_RTHCPHYS;
1890	pVmcsInfo->HCPhysMsrBitmap = NIL_RTHCPHYS;
1891	pVmcsInfo->HCPhysGuestMsrLoad = NIL_RTHCPHYS;
1892	pVmcsInfo->HCPhysGuestMsrStore = NIL_RTHCPHYS;
1893	pVmcsInfo->HCPhysHostMsrLoad = NIL_RTHCPHYS;
1894	pVmcsInfo->HCPhysVirtApic = NIL_RTHCPHYS;
1895	pVmcsInfo->HCPhysEPTP = NIL_RTHCPHYS;
1896	pVmcsInfo->u64VmcsLinkPtr = NIL_RTHCPHYS;
1897	pVmcsInfo->idHostCpuState = NIL_RTCPUID;
1898	pVmcsInfo->idHostCpuExec = NIL_RTCPUID;
1899	}
1900
1901
1902	/**
1903	* Frees the VT-x structures for a VMCS info. object.
1904	*
1905	* @param pVmcsInfo The VMCS info. object.
1906	* @param pVmcsInfoShared The VMCS info. object shared with ring-3.
1907	*/
1908	static void hmR0VmxVmcsInfoFree(PVMXVMCSINFO pVmcsInfo, PVMXVMCSINFOSHARED pVmcsInfoShared)
1909	{
1910	hmR0VmxPagesFree(&pVmcsInfo->hMemObj);
1911	hmR0VmxVmcsInfoInit(pVmcsInfo, pVmcsInfoShared);
1912	}
1913
1914
1915	/**
1916	* Allocates the VT-x structures for a VMCS info. object.
1917	*
1918	* @returns VBox status code.
1919	* @param pVCpu The cross context virtual CPU structure.
1920	* @param pVmcsInfo The VMCS info. object.
1921	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
1922	*
1923	* @remarks The caller is expected to take care of any and all allocation failures.
1924	* This function will not perform any cleanup for failures half-way
1925	* through.
1926	*/
1927	static int hmR0VmxAllocVmcsInfo(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
1928	{
1929	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1930
1931	bool const fMsrBitmaps = RT_BOOL(g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS);
1932	bool const fShadowVmcs = !fIsNstGstVmcs ? pVM->hmr0.s.vmx.fUseVmcsShadowing : pVM->cpum.ro.GuestFeatures.fVmxVmcsShadowing;
1933	Assert(!pVM->cpum.ro.GuestFeatures.fVmxVmcsShadowing); /* VMCS shadowing is not yet exposed to the guest. */
1934	VMXPAGEALLOCINFO aAllocInfo[] =
1935	{
1936	{ true, 0 /* Unused */, &pVmcsInfo->HCPhysVmcs, &pVmcsInfo->pvVmcs },
1937	{ true, 0 /* Unused */, &pVmcsInfo->HCPhysGuestMsrLoad, &pVmcsInfo->pvGuestMsrLoad },
1938	{ true, 0 /* Unused */, &pVmcsInfo->HCPhysHostMsrLoad, &pVmcsInfo->pvHostMsrLoad },
1939	{ fMsrBitmaps, 0 /* Unused */, &pVmcsInfo->HCPhysMsrBitmap, &pVmcsInfo->pvMsrBitmap },
1940	{ fShadowVmcs, 0 /* Unused */, &pVmcsInfo->HCPhysShadowVmcs, &pVmcsInfo->pvShadowVmcs },
1941	};
1942
1943	int rc = hmR0VmxPagesAllocZ(&pVmcsInfo->hMemObj, &aAllocInfo[0], RT_ELEMENTS(aAllocInfo));
1944	if (RT_FAILURE(rc))
1945	return rc;
1946
1947	/*
1948	* We use the same page for VM-entry MSR-load and VM-exit MSR store areas.
1949	* Because they contain a symmetric list of guest MSRs to load on VM-entry and store on VM-exit.
1950	*/
1951	AssertCompile(RT_ELEMENTS(aAllocInfo) > 0);
1952	Assert(pVmcsInfo->HCPhysGuestMsrLoad != NIL_RTHCPHYS);
1953	pVmcsInfo->pvGuestMsrStore = pVmcsInfo->pvGuestMsrLoad;
1954	pVmcsInfo->HCPhysGuestMsrStore = pVmcsInfo->HCPhysGuestMsrLoad;
1955
1956	/*
1957	* Get the virtual-APIC page rather than allocating them again.
1958	*/
1959	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW)
1960	{
1961	if (!fIsNstGstVmcs)
1962	{
1963	if (PDMHasApic(pVM))
1964	{
1965	rc = APICGetApicPageForCpu(pVCpu, &pVmcsInfo->HCPhysVirtApic, (PRTR0PTR)&pVmcsInfo->pbVirtApic, NULL /pR3Ptr/);
1966	if (RT_FAILURE(rc))
1967	return rc;
1968	Assert(pVmcsInfo->pbVirtApic);
1969	Assert(pVmcsInfo->HCPhysVirtApic && pVmcsInfo->HCPhysVirtApic != NIL_RTHCPHYS);
1970	}
1971	}
1972	else
1973	{
1974	pVmcsInfo->pbVirtApic = &pVCpu->cpum.GstCtx.hwvirt.vmx.abVirtApicPage[0];
1975	pVmcsInfo->HCPhysVirtApic = GVMMR0ConvertGVMPtr2HCPhys(pVM, pVmcsInfo->pbVirtApic);
1976	Assert(pVmcsInfo->HCPhysVirtApic && pVmcsInfo->HCPhysVirtApic != NIL_RTHCPHYS);
1977	}
1978	}
1979
1980	return VINF_SUCCESS;
1981	}
1982
1983
1984	/**
1985	* Free all VT-x structures for the VM.
1986	*
1987	* @returns IPRT status code.
1988	* @param pVM The cross context VM structure.
1989	*/
1990	static void hmR0VmxStructsFree(PVMCC pVM)
1991	{
1992	hmR0VmxPagesFree(&pVM->hmr0.s.vmx.hMemObj);
1993	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1994	if (pVM->hmr0.s.vmx.fUseVmcsShadowing)
1995	{
1996	RTMemFree(pVM->hmr0.s.vmx.paShadowVmcsFields);
1997	pVM->hmr0.s.vmx.paShadowVmcsFields = NULL;
1998	RTMemFree(pVM->hmr0.s.vmx.paShadowVmcsRoFields);
1999	pVM->hmr0.s.vmx.paShadowVmcsRoFields = NULL;
2000	}
2001	#endif
2002
2003	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
2004	{
2005	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
2006	hmR0VmxVmcsInfoFree(&pVCpu->hmr0.s.vmx.VmcsInfo, &pVCpu->hm.s.vmx.VmcsInfo);
2007	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2008	if (pVM->cpum.ro.GuestFeatures.fVmx)
2009	hmR0VmxVmcsInfoFree(&pVCpu->hmr0.s.vmx.VmcsInfoNstGst, &pVCpu->hm.s.vmx.VmcsInfoNstGst);
2010	#endif
2011	}
2012	}
2013
2014
2015	/**
2016	* Allocate all VT-x structures for the VM.
2017	*
2018	* @returns IPRT status code.
2019	* @param pVM The cross context VM structure.
2020	*
2021	* @remarks This functions will cleanup on memory allocation failures.
2022	*/
2023	static int hmR0VmxStructsAlloc(PVMCC pVM)
2024	{
2025	/*
2026	* Sanity check the VMCS size reported by the CPU as we assume 4KB allocations.
2027	* The VMCS size cannot be more than 4096 bytes.
2028	*
2029	* See Intel spec. Appendix A.1 "Basic VMX Information".
2030	*/
2031	uint32_t const cbVmcs = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_SIZE);
2032	if (cbVmcs <= X86_PAGE_4K_SIZE)
2033	{ /* likely */ }
2034	else
2035	{
2036	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_INVALID_VMCS_SIZE;
2037	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2038	}
2039
2040	/*
2041	* Allocate per-VM VT-x structures.
2042	*/
2043	bool const fVirtApicAccess = RT_BOOL(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS);
2044	bool const fUseVmcsShadowing = pVM->hmr0.s.vmx.fUseVmcsShadowing;
2045	VMXPAGEALLOCINFO aAllocInfo[] =
2046	{
2047	{ fVirtApicAccess, 0 /* Unused */, &pVM->hmr0.s.vmx.HCPhysApicAccess, (PRTR0PTR)&pVM->hmr0.s.vmx.pbApicAccess },
2048	{ fUseVmcsShadowing, 0 /* Unused */, &pVM->hmr0.s.vmx.HCPhysVmreadBitmap, &pVM->hmr0.s.vmx.pvVmreadBitmap },
2049	{ fUseVmcsShadowing, 0 /* Unused */, &pVM->hmr0.s.vmx.HCPhysVmwriteBitmap, &pVM->hmr0.s.vmx.pvVmwriteBitmap },
2050	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2051	{ true, 0 /* Unused */, &pVM->hmr0.s.vmx.HCPhysScratch, (PRTR0PTR)&pVM->hmr0.s.vmx.pbScratch },
2052	#endif
2053	};
2054
2055	int rc = hmR0VmxPagesAllocZ(&pVM->hmr0.s.vmx.hMemObj, &aAllocInfo[0], RT_ELEMENTS(aAllocInfo));
2056	if (RT_SUCCESS(rc))
2057	{
2058	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2059	/* Allocate the shadow VMCS-fields array. */
2060	if (fUseVmcsShadowing)
2061	{
2062	Assert(!pVM->hmr0.s.vmx.cShadowVmcsFields);
2063	Assert(!pVM->hmr0.s.vmx.cShadowVmcsRoFields);
2064	pVM->hmr0.s.vmx.paShadowVmcsFields = (uint32_t *)RTMemAllocZ(sizeof(g_aVmcsFields));
2065	pVM->hmr0.s.vmx.paShadowVmcsRoFields = (uint32_t *)RTMemAllocZ(sizeof(g_aVmcsFields));
2066	if (!pVM->hmr0.s.vmx.paShadowVmcsFields \|\| !pVM->hmr0.s.vmx.paShadowVmcsRoFields)
2067	rc = VERR_NO_MEMORY;
2068	}
2069	#endif
2070
2071	/*
2072	* Allocate per-VCPU VT-x structures.
2073	*/
2074	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus && RT_SUCCESS(rc); idCpu++)
2075	{
2076	/* Allocate the guest VMCS structures. */
2077	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
2078	rc = hmR0VmxAllocVmcsInfo(pVCpu, &pVCpu->hmr0.s.vmx.VmcsInfo, false /* fIsNstGstVmcs */);
2079
2080	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2081	/* Allocate the nested-guest VMCS structures, when the VMX feature is exposed to the guest. */
2082	if (pVM->cpum.ro.GuestFeatures.fVmx && RT_SUCCESS(rc))
2083	rc = hmR0VmxAllocVmcsInfo(pVCpu, &pVCpu->hmr0.s.vmx.VmcsInfoNstGst, true /* fIsNstGstVmcs */);
2084	#endif
2085	}
2086	if (RT_SUCCESS(rc))
2087	return VINF_SUCCESS;
2088	}
2089	hmR0VmxStructsFree(pVM);
2090	return rc;
2091	}
2092
2093
2094	/**
2095	* Pre-initializes non-zero fields in VMX structures that will be allocated.
2096	*
2097	* @param pVM The cross context VM structure.
2098	*/
2099	static void hmR0VmxStructsInit(PVMCC pVM)
2100	{
2101	/* Paranoia. */
2102	Assert(pVM->hmr0.s.vmx.pbApicAccess == NULL);
2103	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2104	Assert(pVM->hmr0.s.vmx.pbScratch == NULL);
2105	#endif
2106
2107	/*
2108	* Initialize members up-front so we can cleanup en masse on allocation failures.
2109	*/
2110	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2111	pVM->hmr0.s.vmx.HCPhysScratch = NIL_RTHCPHYS;
2112	#endif
2113	pVM->hmr0.s.vmx.HCPhysApicAccess = NIL_RTHCPHYS;
2114	pVM->hmr0.s.vmx.HCPhysVmreadBitmap = NIL_RTHCPHYS;
2115	pVM->hmr0.s.vmx.HCPhysVmwriteBitmap = NIL_RTHCPHYS;
2116	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
2117	{
2118	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
2119	hmR0VmxVmcsInfoInit(&pVCpu->hmr0.s.vmx.VmcsInfo, &pVCpu->hm.s.vmx.VmcsInfo);
2120	hmR0VmxVmcsInfoInit(&pVCpu->hmr0.s.vmx.VmcsInfoNstGst, &pVCpu->hm.s.vmx.VmcsInfoNstGst);
2121	}
2122	}
2123
2124	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2125	/**
2126	* Returns whether an MSR at the given MSR-bitmap offset is intercepted or not.
2127	*
2128	* @returns @c true if the MSR is intercepted, @c false otherwise.
2129	* @param pbMsrBitmap The MSR bitmap.
2130	* @param offMsr The MSR byte offset.
2131	* @param iBit The bit offset from the byte offset.
2132	*/
2133	DECLINLINE(bool) hmR0VmxIsMsrBitSet(uint8_t const *pbMsrBitmap, uint16_t offMsr, int32_t iBit)
2134	{
2135	Assert(offMsr + (iBit >> 3) <= X86_PAGE_4K_SIZE);
2136	return ASMBitTest(pbMsrBitmap + offMsr, iBit);
2137	}
2138	#endif
2139
2140	/**
2141	* Sets the permission bits for the specified MSR in the given MSR bitmap.
2142	*
2143	* If the passed VMCS is a nested-guest VMCS, this function ensures that the
2144	* read/write intercept is cleared from the MSR bitmap used for hardware-assisted
2145	* VMX execution of the nested-guest, only if nested-guest is also not intercepting
2146	* the read/write access of this MSR.
2147	*
2148	* @param pVCpu The cross context virtual CPU structure.
2149	* @param pVmcsInfo The VMCS info. object.
2150	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2151	* @param idMsr The MSR value.
2152	* @param fMsrpm The MSR permissions (see VMXMSRPM_XXX). This must
2153	* include both a read -and- a write permission!
2154	*
2155	* @sa CPUMGetVmxMsrPermission.
2156	* @remarks Can be called with interrupts disabled.
2157	*/
2158	static void hmR0VmxSetMsrPermission(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs, uint32_t idMsr, uint32_t fMsrpm)
2159	{
2160	uint8_t pbMsrBitmap = (uint8_t )pVmcsInfo->pvMsrBitmap;
2161	Assert(pbMsrBitmap);
2162	Assert(VMXMSRPM_IS_FLAG_VALID(fMsrpm));
2163
2164	/*
2165	* MSR-bitmap Layout:
2166	* Byte index MSR range Interpreted as
2167	* 0x000 - 0x3ff 0x00000000 - 0x00001fff Low MSR read bits.
2168	* 0x400 - 0x7ff 0xc0000000 - 0xc0001fff High MSR read bits.
2169	* 0x800 - 0xbff 0x00000000 - 0x00001fff Low MSR write bits.
2170	* 0xc00 - 0xfff 0xc0000000 - 0xc0001fff High MSR write bits.
2171	*
2172	* A bit corresponding to an MSR within the above range causes a VM-exit
2173	* if the bit is 1 on executions of RDMSR/WRMSR. If an MSR falls out of
2174	* the MSR range, it always cause a VM-exit.
2175	*
2176	* See Intel spec. 24.6.9 "MSR-Bitmap Address".
2177	*/
2178	uint16_t const offBitmapRead = 0;
2179	uint16_t const offBitmapWrite = 0x800;
2180	uint16_t offMsr;
2181	int32_t iBit;
2182	if (idMsr <= UINT32_C(0x00001fff))
2183	{
2184	offMsr = 0;
2185	iBit = idMsr;
2186	}
2187	else if (idMsr - UINT32_C(0xc0000000) <= UINT32_C(0x00001fff))
2188	{
2189	offMsr = 0x400;
2190	iBit = idMsr - UINT32_C(0xc0000000);
2191	}
2192	else
2193	AssertMsgFailedReturnVoid(("Invalid MSR %#RX32\n", idMsr));
2194
2195	/*
2196	* Set the MSR read permission.
2197	*/
2198	uint16_t const offMsrRead = offBitmapRead + offMsr;
2199	Assert(offMsrRead + (iBit >> 3) < offBitmapWrite);
2200	if (fMsrpm & VMXMSRPM_ALLOW_RD)
2201	{
2202	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2203	bool const fClear = !fIsNstGstVmcs ? true
2204	: !hmR0VmxIsMsrBitSet(pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap, offMsrRead, iBit);
2205	#else
2206	RT_NOREF2(pVCpu, fIsNstGstVmcs);
2207	bool const fClear = true;
2208	#endif
2209	if (fClear)
2210	ASMBitClear(pbMsrBitmap + offMsrRead, iBit);
2211	}
2212	else
2213	ASMBitSet(pbMsrBitmap + offMsrRead, iBit);
2214
2215	/*
2216	* Set the MSR write permission.
2217	*/
2218	uint16_t const offMsrWrite = offBitmapWrite + offMsr;
2219	Assert(offMsrWrite + (iBit >> 3) < X86_PAGE_4K_SIZE);
2220	if (fMsrpm & VMXMSRPM_ALLOW_WR)
2221	{
2222	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2223	bool const fClear = !fIsNstGstVmcs ? true
2224	: !hmR0VmxIsMsrBitSet(pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap, offMsrWrite, iBit);
2225	#else
2226	RT_NOREF2(pVCpu, fIsNstGstVmcs);
2227	bool const fClear = true;
2228	#endif
2229	if (fClear)
2230	ASMBitClear(pbMsrBitmap + offMsrWrite, iBit);
2231	}
2232	else
2233	ASMBitSet(pbMsrBitmap + offMsrWrite, iBit);
2234	}
2235
2236
2237	/**
2238	* Updates the VMCS with the number of effective MSRs in the auto-load/store MSR
2239	* area.
2240	*
2241	* @returns VBox status code.
2242	* @param pVCpu The cross context virtual CPU structure.
2243	* @param pVmcsInfo The VMCS info. object.
2244	* @param cMsrs The number of MSRs.
2245	*/
2246	static int hmR0VmxSetAutoLoadStoreMsrCount(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint32_t cMsrs)
2247	{
2248	/* Shouldn't ever happen but there -is- a number. We're well within the recommended 512. */
2249	uint32_t const cMaxSupportedMsrs = VMX_MISC_MAX_MSRS(g_HmMsrs.u.vmx.u64Misc);
2250	if (RT_LIKELY(cMsrs < cMaxSupportedMsrs))
2251	{
2252	/* Commit the MSR counts to the VMCS and update the cache. */
2253	if (pVmcsInfo->cEntryMsrLoad != cMsrs)
2254	{
2255	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, cMsrs); AssertRC(rc);
2256	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, cMsrs); AssertRC(rc);
2257	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, cMsrs); AssertRC(rc);
2258	pVmcsInfo->cEntryMsrLoad = cMsrs;
2259	pVmcsInfo->cExitMsrStore = cMsrs;
2260	pVmcsInfo->cExitMsrLoad = cMsrs;
2261	}
2262	return VINF_SUCCESS;
2263	}
2264
2265	LogRel(("Auto-load/store MSR count exceeded! cMsrs=%u MaxSupported=%u\n", cMsrs, cMaxSupportedMsrs));
2266	pVCpu->hm.s.u32HMError = VMX_UFC_INSUFFICIENT_GUEST_MSR_STORAGE;
2267	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2268	}
2269
2270
2271	/**
2272	* Adds a new (or updates the value of an existing) guest/host MSR
2273	* pair to be swapped during the world-switch as part of the
2274	* auto-load/store MSR area in the VMCS.
2275	*
2276	* @returns VBox status code.
2277	* @param pVCpu The cross context virtual CPU structure.
2278	* @param pVmxTransient The VMX-transient structure.
2279	* @param idMsr The MSR.
2280	* @param uGuestMsrValue Value of the guest MSR.
2281	* @param fSetReadWrite Whether to set the guest read/write access of this
2282	* MSR (thus not causing a VM-exit).
2283	* @param fUpdateHostMsr Whether to update the value of the host MSR if
2284	* necessary.
2285	*/
2286	static int hmR0VmxAddAutoLoadStoreMsr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t idMsr, uint64_t uGuestMsrValue,
2287	bool fSetReadWrite, bool fUpdateHostMsr)
2288	{
2289	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
2290	bool const fIsNstGstVmcs = pVmxTransient->fIsNestedGuest;
2291	PVMXAUTOMSR pGuestMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2292	uint32_t cMsrs = pVmcsInfo->cEntryMsrLoad;
2293	uint32_t i;
2294
2295	/* Paranoia. */
2296	Assert(pGuestMsrLoad);
2297
2298	#ifndef DEBUG_bird
2299	LogFlowFunc(("pVCpu=%p idMsr=%#RX32 uGuestMsrValue=%#RX64\n", pVCpu, idMsr, uGuestMsrValue));
2300	#endif
2301
2302	/* Check if the MSR already exists in the VM-entry MSR-load area. */
2303	for (i = 0; i < cMsrs; i++)
2304	{
2305	if (pGuestMsrLoad[i].u32Msr == idMsr)
2306	break;
2307	}
2308
2309	bool fAdded = false;
2310	if (i == cMsrs)
2311	{
2312	/* The MSR does not exist, bump the MSR count to make room for the new MSR. */
2313	++cMsrs;
2314	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, pVmcsInfo, cMsrs);
2315	AssertMsgRCReturn(rc, ("Insufficient space to add MSR to VM-entry MSR-load/store area %u\n", idMsr), rc);
2316
2317	/* Set the guest to read/write this MSR without causing VM-exits. */
2318	if ( fSetReadWrite
2319	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS))
2320	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, fIsNstGstVmcs, idMsr, VMXMSRPM_ALLOW_RD_WR);
2321
2322	Log4Func(("Added MSR %#RX32, cMsrs=%u\n", idMsr, cMsrs));
2323	fAdded = true;
2324	}
2325
2326	/* Update the MSR value for the newly added or already existing MSR. */
2327	pGuestMsrLoad[i].u32Msr = idMsr;
2328	pGuestMsrLoad[i].u64Value = uGuestMsrValue;
2329
2330	/* Create the corresponding slot in the VM-exit MSR-store area if we use a different page. */
2331	if (hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo))
2332	{
2333	PVMXAUTOMSR pGuestMsrStore = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2334	pGuestMsrStore[i].u32Msr = idMsr;
2335	pGuestMsrStore[i].u64Value = uGuestMsrValue;
2336	}
2337
2338	/* Update the corresponding slot in the host MSR area. */
2339	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2340	Assert(pHostMsr != pVmcsInfo->pvGuestMsrLoad);
2341	Assert(pHostMsr != pVmcsInfo->pvGuestMsrStore);
2342	pHostMsr[i].u32Msr = idMsr;
2343
2344	/*
2345	* Only if the caller requests to update the host MSR value AND we've newly added the
2346	* MSR to the host MSR area do we actually update the value. Otherwise, it will be
2347	* updated by hmR0VmxUpdateAutoLoadHostMsrs().
2348	*
2349	* We do this for performance reasons since reading MSRs may be quite expensive.
2350	*/
2351	if (fAdded)
2352	{
2353	if (fUpdateHostMsr)
2354	{
2355	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2356	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2357	pHostMsr[i].u64Value = ASMRdMsr(idMsr);
2358	}
2359	else
2360	{
2361	/* Someone else can do the work. */
2362	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = false;
2363	}
2364	}
2365	return VINF_SUCCESS;
2366	}
2367
2368
2369	/**
2370	* Removes a guest/host MSR pair to be swapped during the world-switch from the
2371	* auto-load/store MSR area in the VMCS.
2372	*
2373	* @returns VBox status code.
2374	* @param pVCpu The cross context virtual CPU structure.
2375	* @param pVmxTransient The VMX-transient structure.
2376	* @param idMsr The MSR.
2377	*/
2378	static int hmR0VmxRemoveAutoLoadStoreMsr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t idMsr)
2379	{
2380	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
2381	bool const fIsNstGstVmcs = pVmxTransient->fIsNestedGuest;
2382	PVMXAUTOMSR pGuestMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2383	uint32_t cMsrs = pVmcsInfo->cEntryMsrLoad;
2384
2385	#ifndef DEBUG_bird
2386	LogFlowFunc(("pVCpu=%p idMsr=%#RX32\n", pVCpu, idMsr));
2387	#endif
2388
2389	for (uint32_t i = 0; i < cMsrs; i++)
2390	{
2391	/* Find the MSR. */
2392	if (pGuestMsrLoad[i].u32Msr == idMsr)
2393	{
2394	/*
2395	* If it's the last MSR, we only need to reduce the MSR count.
2396	* If it's -not- the last MSR, copy the last MSR in place of it and reduce the MSR count.
2397	*/
2398	if (i < cMsrs - 1)
2399	{
2400	/* Remove it from the VM-entry MSR-load area. */
2401	pGuestMsrLoad[i].u32Msr = pGuestMsrLoad[cMsrs - 1].u32Msr;
2402	pGuestMsrLoad[i].u64Value = pGuestMsrLoad[cMsrs - 1].u64Value;
2403
2404	/* Remove it from the VM-exit MSR-store area if it's in a different page. */
2405	if (hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo))
2406	{
2407	PVMXAUTOMSR pGuestMsrStore = (PVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2408	Assert(pGuestMsrStore[i].u32Msr == idMsr);
2409	pGuestMsrStore[i].u32Msr = pGuestMsrStore[cMsrs - 1].u32Msr;
2410	pGuestMsrStore[i].u64Value = pGuestMsrStore[cMsrs - 1].u64Value;
2411	}
2412
2413	/* Remove it from the VM-exit MSR-load area. */
2414	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2415	Assert(pHostMsr[i].u32Msr == idMsr);
2416	pHostMsr[i].u32Msr = pHostMsr[cMsrs - 1].u32Msr;
2417	pHostMsr[i].u64Value = pHostMsr[cMsrs - 1].u64Value;
2418	}
2419
2420	/* Reduce the count to reflect the removed MSR and bail. */
2421	--cMsrs;
2422	break;
2423	}
2424	}
2425
2426	/* Update the VMCS if the count changed (meaning the MSR was found and removed). */
2427	if (cMsrs != pVmcsInfo->cEntryMsrLoad)
2428	{
2429	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, pVmcsInfo, cMsrs);
2430	AssertRCReturn(rc, rc);
2431
2432	/* We're no longer swapping MSRs during the world-switch, intercept guest read/writes to them. */
2433	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
2434	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, fIsNstGstVmcs, idMsr, VMXMSRPM_EXIT_RD \| VMXMSRPM_EXIT_WR);
2435
2436	Log4Func(("Removed MSR %#RX32, cMsrs=%u\n", idMsr, cMsrs));
2437	return VINF_SUCCESS;
2438	}
2439
2440	return VERR_NOT_FOUND;
2441	}
2442
2443
2444	/**
2445	* Checks if the specified guest MSR is part of the VM-entry MSR-load area.
2446	*
2447	* @returns @c true if found, @c false otherwise.
2448	* @param pVmcsInfo The VMCS info. object.
2449	* @param idMsr The MSR to find.
2450	*/
2451	static bool hmR0VmxIsAutoLoadGuestMsr(PCVMXVMCSINFO pVmcsInfo, uint32_t idMsr)
2452	{
2453	PCVMXAUTOMSR pMsrs = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2454	uint32_t const cMsrs = pVmcsInfo->cEntryMsrLoad;
2455	Assert(pMsrs);
2456	Assert(sizeof(pMsrs) cMsrs <= X86_PAGE_4K_SIZE);
2457	for (uint32_t i = 0; i < cMsrs; i++)
2458	{
2459	if (pMsrs[i].u32Msr == idMsr)
2460	return true;
2461	}
2462	return false;
2463	}
2464
2465
2466	/**
2467	* Updates the value of all host MSRs in the VM-exit MSR-load area.
2468	*
2469	* @param pVCpu The cross context virtual CPU structure.
2470	* @param pVmcsInfo The VMCS info. object.
2471	*
2472	* @remarks No-long-jump zone!!!
2473	*/
2474	static void hmR0VmxUpdateAutoLoadHostMsrs(PCVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
2475	{
2476	RT_NOREF(pVCpu);
2477	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2478
2479	PVMXAUTOMSR pHostMsrLoad = (PVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2480	uint32_t const cMsrs = pVmcsInfo->cExitMsrLoad;
2481	Assert(pHostMsrLoad);
2482	Assert(sizeof(pHostMsrLoad) cMsrs <= X86_PAGE_4K_SIZE);
2483	LogFlowFunc(("pVCpu=%p cMsrs=%u\n", pVCpu, cMsrs));
2484	for (uint32_t i = 0; i < cMsrs; i++)
2485	{
2486	/*
2487	* Performance hack for the host EFER MSR. We use the cached value rather than re-read it.
2488	* Strict builds will catch mismatches in hmR0VmxCheckAutoLoadStoreMsrs(). See @bugref{7368}.
2489	*/
2490	if (pHostMsrLoad[i].u32Msr == MSR_K6_EFER)
2491	pHostMsrLoad[i].u64Value = g_uHmVmxHostMsrEfer;
2492	else
2493	pHostMsrLoad[i].u64Value = ASMRdMsr(pHostMsrLoad[i].u32Msr);
2494	}
2495	}
2496
2497
2498	/**
2499	* Saves a set of host MSRs to allow read/write passthru access to the guest and
2500	* perform lazy restoration of the host MSRs while leaving VT-x.
2501	*
2502	* @param pVCpu The cross context virtual CPU structure.
2503	*
2504	* @remarks No-long-jump zone!!!
2505	*/
2506	static void hmR0VmxLazySaveHostMsrs(PVMCPUCC pVCpu)
2507	{
2508	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2509
2510	/*
2511	* Note: If you're adding MSRs here, make sure to update the MSR-bitmap accesses in hmR0VmxSetupVmcsProcCtls().
2512	*/
2513	if (!(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST))
2514	{
2515	Assert(!(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)); /* Guest MSRs better not be loaded now. */
2516	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fAllow64BitGuests)
2517	{
2518	pVCpu->hmr0.s.vmx.u64HostMsrLStar = ASMRdMsr(MSR_K8_LSTAR);
2519	pVCpu->hmr0.s.vmx.u64HostMsrStar = ASMRdMsr(MSR_K6_STAR);
2520	pVCpu->hmr0.s.vmx.u64HostMsrSfMask = ASMRdMsr(MSR_K8_SF_MASK);
2521	pVCpu->hmr0.s.vmx.u64HostMsrKernelGsBase = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
2522	}
2523	pVCpu->hmr0.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_SAVED_HOST;
2524	}
2525	}
2526
2527
2528	/**
2529	* Checks whether the MSR belongs to the set of guest MSRs that we restore
2530	* lazily while leaving VT-x.
2531	*
2532	* @returns true if it does, false otherwise.
2533	* @param pVCpu The cross context virtual CPU structure.
2534	* @param idMsr The MSR to check.
2535	*/
2536	static bool hmR0VmxIsLazyGuestMsr(PCVMCPUCC pVCpu, uint32_t idMsr)
2537	{
2538	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fAllow64BitGuests)
2539	{
2540	switch (idMsr)
2541	{
2542	case MSR_K8_LSTAR:
2543	case MSR_K6_STAR:
2544	case MSR_K8_SF_MASK:
2545	case MSR_K8_KERNEL_GS_BASE:
2546	return true;
2547	}
2548	}
2549	return false;
2550	}
2551
2552
2553	/**
2554	* Loads a set of guests MSRs to allow read/passthru to the guest.
2555	*
2556	* The name of this function is slightly confusing. This function does NOT
2557	* postpone loading, but loads the MSR right now. "hmR0VmxLazy" is simply a
2558	* common prefix for functions dealing with "lazy restoration" of the shared
2559	* MSRs.
2560	*
2561	* @param pVCpu The cross context virtual CPU structure.
2562	*
2563	* @remarks No-long-jump zone!!!
2564	*/
2565	static void hmR0VmxLazyLoadGuestMsrs(PVMCPUCC pVCpu)
2566	{
2567	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2568	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2569
2570	Assert(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
2571	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fAllow64BitGuests)
2572	{
2573	/*
2574	* If the guest MSRs are not loaded -and- if all the guest MSRs are identical
2575	* to the MSRs on the CPU (which are the saved host MSRs, see assertion above) then
2576	* we can skip a few MSR writes.
2577	*
2578	* Otherwise, it implies either 1. they're not loaded, or 2. they're loaded but the
2579	* guest MSR values in the guest-CPU context might be different to what's currently
2580	* loaded in the CPU. In either case, we need to write the new guest MSR values to the
2581	* CPU, see @bugref{8728}.
2582	*/
2583	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
2584	if ( !(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
2585	&& pCtx->msrKERNELGSBASE == pVCpu->hmr0.s.vmx.u64HostMsrKernelGsBase
2586	&& pCtx->msrLSTAR == pVCpu->hmr0.s.vmx.u64HostMsrLStar
2587	&& pCtx->msrSTAR == pVCpu->hmr0.s.vmx.u64HostMsrStar
2588	&& pCtx->msrSFMASK == pVCpu->hmr0.s.vmx.u64HostMsrSfMask)
2589	{
2590	#ifdef VBOX_STRICT
2591	Assert(ASMRdMsr(MSR_K8_KERNEL_GS_BASE) == pCtx->msrKERNELGSBASE);
2592	Assert(ASMRdMsr(MSR_K8_LSTAR) == pCtx->msrLSTAR);
2593	Assert(ASMRdMsr(MSR_K6_STAR) == pCtx->msrSTAR);
2594	Assert(ASMRdMsr(MSR_K8_SF_MASK) == pCtx->msrSFMASK);
2595	#endif
2596	}
2597	else
2598	{
2599	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pCtx->msrKERNELGSBASE);
2600	ASMWrMsr(MSR_K8_LSTAR, pCtx->msrLSTAR);
2601	ASMWrMsr(MSR_K6_STAR, pCtx->msrSTAR);
2602	/* The system call flag mask register isn't as benign and accepting of all
2603	values as the above, so mask it to avoid #GP'ing on corrupted input. */
2604	Assert(!(pCtx->msrSFMASK & ~(uint64_t)UINT32_MAX));
2605	ASMWrMsr(MSR_K8_SF_MASK, pCtx->msrSFMASK & UINT32_MAX);
2606	}
2607	}
2608	pVCpu->hmr0.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_LOADED_GUEST;
2609	}
2610
2611
2612	/**
2613	* Performs lazy restoration of the set of host MSRs if they were previously
2614	* loaded with guest MSR values.
2615	*
2616	* @param pVCpu The cross context virtual CPU structure.
2617	*
2618	* @remarks No-long-jump zone!!!
2619	* @remarks The guest MSRs should have been saved back into the guest-CPU
2620	* context by hmR0VmxImportGuestState()!!!
2621	*/
2622	static void hmR0VmxLazyRestoreHostMsrs(PVMCPUCC pVCpu)
2623	{
2624	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2625	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2626
2627	if (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
2628	{
2629	Assert(pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
2630	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fAllow64BitGuests)
2631	{
2632	ASMWrMsr(MSR_K8_LSTAR, pVCpu->hmr0.s.vmx.u64HostMsrLStar);
2633	ASMWrMsr(MSR_K6_STAR, pVCpu->hmr0.s.vmx.u64HostMsrStar);
2634	ASMWrMsr(MSR_K8_SF_MASK, pVCpu->hmr0.s.vmx.u64HostMsrSfMask);
2635	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pVCpu->hmr0.s.vmx.u64HostMsrKernelGsBase);
2636	}
2637	}
2638	pVCpu->hmr0.s.vmx.fLazyMsrs &= ~(VMX_LAZY_MSRS_LOADED_GUEST \| VMX_LAZY_MSRS_SAVED_HOST);
2639	}
2640
2641
2642	/**
2643	* Verifies that our cached values of the VMCS fields are all consistent with
2644	* what's actually present in the VMCS.
2645	*
2646	* @returns VBox status code.
2647	* @retval VINF_SUCCESS if all our caches match their respective VMCS fields.
2648	* @retval VERR_VMX_VMCS_FIELD_CACHE_INVALID if a cache field doesn't match the
2649	* VMCS content. HMCPU error-field is
2650	* updated, see VMX_VCI_XXX.
2651	* @param pVCpu The cross context virtual CPU structure.
2652	* @param pVmcsInfo The VMCS info. object.
2653	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2654	*/
2655	static int hmR0VmxCheckCachedVmcsCtls(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
2656	{
2657	const char * const pcszVmcs = fIsNstGstVmcs ? "Nested-guest VMCS" : "VMCS";
2658
2659	uint32_t u32Val;
2660	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
2661	AssertRC(rc);
2662	AssertMsgReturnStmt(pVmcsInfo->u32EntryCtls == u32Val,
2663	("%s entry controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32EntryCtls, u32Val),
2664	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_ENTRY,
2665	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2666
2667	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val);
2668	AssertRC(rc);
2669	AssertMsgReturnStmt(pVmcsInfo->u32ExitCtls == u32Val,
2670	("%s exit controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ExitCtls, u32Val),
2671	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_EXIT,
2672	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2673
2674	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val);
2675	AssertRC(rc);
2676	AssertMsgReturnStmt(pVmcsInfo->u32PinCtls == u32Val,
2677	("%s pin controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32PinCtls, u32Val),
2678	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PIN_EXEC,
2679	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2680
2681	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val);
2682	AssertRC(rc);
2683	AssertMsgReturnStmt(pVmcsInfo->u32ProcCtls == u32Val,
2684	("%s proc controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ProcCtls, u32Val),
2685	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC,
2686	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2687
2688	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
2689	{
2690	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val);
2691	AssertRC(rc);
2692	AssertMsgReturnStmt(pVmcsInfo->u32ProcCtls2 == u32Val,
2693	("%s proc2 controls mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32ProcCtls2, u32Val),
2694	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC2,
2695	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2696	}
2697
2698	uint64_t u64Val;
2699	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TERTIARY_CTLS)
2700	{
2701	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_PROC_EXEC3_FULL, &u64Val);
2702	AssertRC(rc);
2703	AssertMsgReturnStmt(pVmcsInfo->u64ProcCtls3 == u64Val,
2704	("%s proc3 controls mismatch: Cache=%#RX32 VMCS=%#RX64\n", pcszVmcs, pVmcsInfo->u64ProcCtls3, u64Val),
2705	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_PROC_EXEC3,
2706	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2707	}
2708
2709	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, &u32Val);
2710	AssertRC(rc);
2711	AssertMsgReturnStmt(pVmcsInfo->u32XcptBitmap == u32Val,
2712	("%s exception bitmap mismatch: Cache=%#RX32 VMCS=%#RX32\n", pcszVmcs, pVmcsInfo->u32XcptBitmap, u32Val),
2713	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_XCPT_BITMAP,
2714	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2715
2716	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, &u64Val);
2717	AssertRC(rc);
2718	AssertMsgReturnStmt(pVmcsInfo->u64TscOffset == u64Val,
2719	("%s TSC offset mismatch: Cache=%#RX64 VMCS=%#RX64\n", pcszVmcs, pVmcsInfo->u64TscOffset, u64Val),
2720	pVCpu->hm.s.u32HMError = VMX_VCI_CTRL_TSC_OFFSET,
2721	VERR_VMX_VMCS_FIELD_CACHE_INVALID);
2722
2723	NOREF(pcszVmcs);
2724	return VINF_SUCCESS;
2725	}
2726
2727	#ifdef VBOX_STRICT
2728
2729	/**
2730	* Verifies that our cached host EFER MSR value has not changed since we cached it.
2731	*
2732	* @param pVmcsInfo The VMCS info. object.
2733	*/
2734	static void hmR0VmxCheckHostEferMsr(PCVMXVMCSINFO pVmcsInfo)
2735	{
2736	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2737
2738	if (pVmcsInfo->u32ExitCtls & VMX_EXIT_CTLS_LOAD_EFER_MSR)
2739	{
2740	uint64_t const uHostEferMsr = ASMRdMsr(MSR_K6_EFER);
2741	uint64_t const uHostEferMsrCache = g_uHmVmxHostMsrEfer;
2742	uint64_t uVmcsEferMsrVmcs;
2743	int rc = VMXReadVmcs64(VMX_VMCS64_HOST_EFER_FULL, &uVmcsEferMsrVmcs);
2744	AssertRC(rc);
2745
2746	AssertMsgReturnVoid(uHostEferMsr == uVmcsEferMsrVmcs,
2747	("EFER Host/VMCS mismatch! host=%#RX64 vmcs=%#RX64\n", uHostEferMsr, uVmcsEferMsrVmcs));
2748	AssertMsgReturnVoid(uHostEferMsr == uHostEferMsrCache,
2749	("EFER Host/Cache mismatch! host=%#RX64 cache=%#RX64\n", uHostEferMsr, uHostEferMsrCache));
2750	}
2751	}
2752
2753
2754	/**
2755	* Verifies whether the guest/host MSR pairs in the auto-load/store area in the
2756	* VMCS are correct.
2757	*
2758	* @param pVCpu The cross context virtual CPU structure.
2759	* @param pVmcsInfo The VMCS info. object.
2760	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
2761	*/
2762	static void hmR0VmxCheckAutoLoadStoreMsrs(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
2763	{
2764	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2765
2766	/* Read the various MSR-area counts from the VMCS. */
2767	uint32_t cEntryLoadMsrs;
2768	uint32_t cExitStoreMsrs;
2769	uint32_t cExitLoadMsrs;
2770	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &cEntryLoadMsrs); AssertRC(rc);
2771	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &cExitStoreMsrs); AssertRC(rc);
2772	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &cExitLoadMsrs); AssertRC(rc);
2773
2774	/* Verify all the MSR counts are the same. */
2775	Assert(cEntryLoadMsrs == cExitStoreMsrs);
2776	Assert(cExitStoreMsrs == cExitLoadMsrs);
2777	uint32_t const cMsrs = cExitLoadMsrs;
2778
2779	/* Verify the MSR counts do not exceed the maximum count supported by the hardware. */
2780	Assert(cMsrs < VMX_MISC_MAX_MSRS(g_HmMsrs.u.vmx.u64Misc));
2781
2782	/* Verify the MSR counts are within the allocated page size. */
2783	Assert(sizeof(VMXAUTOMSR) * cMsrs <= X86_PAGE_4K_SIZE);
2784
2785	/* Verify the relevant contents of the MSR areas match. */
2786	PCVMXAUTOMSR pGuestMsrLoad = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrLoad;
2787	PCVMXAUTOMSR pGuestMsrStore = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
2788	PCVMXAUTOMSR pHostMsrLoad = (PCVMXAUTOMSR)pVmcsInfo->pvHostMsrLoad;
2789	bool const fSeparateExitMsrStorePage = hmR0VmxIsSeparateExitMsrStoreAreaVmcs(pVmcsInfo);
2790	for (uint32_t i = 0; i < cMsrs; i++)
2791	{
2792	/* Verify that the MSRs are paired properly and that the host MSR has the correct value. */
2793	if (fSeparateExitMsrStorePage)
2794	{
2795	AssertMsgReturnVoid(pGuestMsrLoad->u32Msr == pGuestMsrStore->u32Msr,
2796	("GuestMsrLoad=%#RX32 GuestMsrStore=%#RX32 cMsrs=%u\n",
2797	pGuestMsrLoad->u32Msr, pGuestMsrStore->u32Msr, cMsrs));
2798	}
2799
2800	AssertMsgReturnVoid(pHostMsrLoad->u32Msr == pGuestMsrLoad->u32Msr,
2801	("HostMsrLoad=%#RX32 GuestMsrLoad=%#RX32 cMsrs=%u\n",
2802	pHostMsrLoad->u32Msr, pGuestMsrLoad->u32Msr, cMsrs));
2803
2804	uint64_t const u64HostMsr = ASMRdMsr(pHostMsrLoad->u32Msr);
2805	AssertMsgReturnVoid(pHostMsrLoad->u64Value == u64HostMsr,
2806	("u32Msr=%#RX32 VMCS Value=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
2807	pHostMsrLoad->u32Msr, pHostMsrLoad->u64Value, u64HostMsr, cMsrs));
2808
2809	/* Verify that cached host EFER MSR matches what's loaded on the CPU. */
2810	bool const fIsEferMsr = RT_BOOL(pHostMsrLoad->u32Msr == MSR_K6_EFER);
2811	AssertMsgReturnVoid(!fIsEferMsr \|\| u64HostMsr == g_uHmVmxHostMsrEfer,
2812	("Cached=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n", g_uHmVmxHostMsrEfer, u64HostMsr, cMsrs));
2813
2814	/* Verify that the accesses are as expected in the MSR bitmap for auto-load/store MSRs. */
2815	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
2816	{
2817	uint32_t const fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, pGuestMsrLoad->u32Msr);
2818	if (fIsEferMsr)
2819	{
2820	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_EXIT_RD), ("Passthru read for EFER MSR!?\n"));
2821	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_EXIT_WR), ("Passthru write for EFER MSR!?\n"));
2822	}
2823	else
2824	{
2825	/* Verify LBR MSRs (used only for debugging) are intercepted. We don't passthru these MSRs to the guest yet. */
2826	PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
2827	if ( pVM->hmr0.s.vmx.fLbr
2828	&& ( hmR0VmxIsLbrBranchFromMsr(pVM, pGuestMsrLoad->u32Msr, NULL /* pidxMsr */)
2829	\|\| hmR0VmxIsLbrBranchToMsr(pVM, pGuestMsrLoad->u32Msr, NULL /* pidxMsr */)
2830	\|\| pGuestMsrLoad->u32Msr == pVM->hmr0.s.vmx.idLbrTosMsr))
2831	{
2832	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_MASK) == VMXMSRPM_EXIT_RD_WR,
2833	("u32Msr=%#RX32 cMsrs=%u Passthru read/write for LBR MSRs!\n",
2834	pGuestMsrLoad->u32Msr, cMsrs));
2835	}
2836	else if (!fIsNstGstVmcs)
2837	{
2838	AssertMsgReturnVoid((fMsrpm & VMXMSRPM_MASK) == VMXMSRPM_ALLOW_RD_WR,
2839	("u32Msr=%#RX32 cMsrs=%u No passthru read/write!\n", pGuestMsrLoad->u32Msr, cMsrs));
2840	}
2841	else
2842	{
2843	/*
2844	* A nested-guest VMCS must -also- allow read/write passthrough for the MSR for us to
2845	* execute a nested-guest with MSR passthrough.
2846	*
2847	* Check if the nested-guest MSR bitmap allows passthrough, and if so, assert that we
2848	* allow passthrough too.
2849	*/
2850	void const *pvMsrBitmapNstGst = pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap;
2851	Assert(pvMsrBitmapNstGst);
2852	uint32_t const fMsrpmNstGst = CPUMGetVmxMsrPermission(pvMsrBitmapNstGst, pGuestMsrLoad->u32Msr);
2853	AssertMsgReturnVoid(fMsrpm == fMsrpmNstGst,
2854	("u32Msr=%#RX32 cMsrs=%u Permission mismatch fMsrpm=%#x fMsrpmNstGst=%#x!\n",
2855	pGuestMsrLoad->u32Msr, cMsrs, fMsrpm, fMsrpmNstGst));
2856	}
2857	}
2858	}
2859
2860	/* Move to the next MSR. */
2861	pHostMsrLoad++;
2862	pGuestMsrLoad++;
2863	pGuestMsrStore++;
2864	}
2865	}
2866
2867	#endif /* VBOX_STRICT */
2868
2869	/**
2870	* Flushes the TLB using EPT.
2871	*
2872	* @returns VBox status code.
2873	* @param pVCpu The cross context virtual CPU structure of the calling
2874	* EMT. Can be NULL depending on @a enmTlbFlush.
2875	* @param pVmcsInfo The VMCS info. object. Can be NULL depending on @a
2876	* enmTlbFlush.
2877	* @param enmTlbFlush Type of flush.
2878	*
2879	* @remarks Caller is responsible for making sure this function is called only
2880	* when NestedPaging is supported and providing @a enmTlbFlush that is
2881	* supported by the CPU.
2882	* @remarks Can be called with interrupts disabled.
2883	*/
2884	static void hmR0VmxFlushEpt(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, VMXTLBFLUSHEPT enmTlbFlush)
2885	{
2886	uint64_t au64Descriptor[2];
2887	if (enmTlbFlush == VMXTLBFLUSHEPT_ALL_CONTEXTS)
2888	au64Descriptor[0] = 0;
2889	else
2890	{
2891	Assert(pVCpu);
2892	Assert(pVmcsInfo);
2893	au64Descriptor[0] = pVmcsInfo->HCPhysEPTP;
2894	}
2895	au64Descriptor[1] = 0; /* MBZ. Intel spec. 33.3 "VMX Instructions" */
2896
2897	int rc = VMXR0InvEPT(enmTlbFlush, &au64Descriptor[0]);
2898	AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvEPT %#x %#RHp failed. rc=%Rrc\n", enmTlbFlush, au64Descriptor[0], rc));
2899
2900	if ( RT_SUCCESS(rc)
2901	&& pVCpu)
2902	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushNestedPaging);
2903	}
2904
2905
2906	/**
2907	* Flushes the TLB using VPID.
2908	*
2909	* @returns VBox status code.
2910	* @param pVCpu The cross context virtual CPU structure of the calling
2911	* EMT. Can be NULL depending on @a enmTlbFlush.
2912	* @param enmTlbFlush Type of flush.
2913	* @param GCPtr Virtual address of the page to flush (can be 0 depending
2914	* on @a enmTlbFlush).
2915	*
2916	* @remarks Can be called with interrupts disabled.
2917	*/
2918	static void hmR0VmxFlushVpid(PVMCPUCC pVCpu, VMXTLBFLUSHVPID enmTlbFlush, RTGCPTR GCPtr)
2919	{
2920	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fVpid);
2921
2922	uint64_t au64Descriptor[2];
2923	if (enmTlbFlush == VMXTLBFLUSHVPID_ALL_CONTEXTS)
2924	{
2925	au64Descriptor[0] = 0;
2926	au64Descriptor[1] = 0;
2927	}
2928	else
2929	{
2930	AssertPtr(pVCpu);
2931	AssertMsg(pVCpu->hmr0.s.uCurrentAsid != 0, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hmr0.s.uCurrentAsid));
2932	AssertMsg(pVCpu->hmr0.s.uCurrentAsid <= UINT16_MAX, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hmr0.s.uCurrentAsid));
2933	au64Descriptor[0] = pVCpu->hmr0.s.uCurrentAsid;
2934	au64Descriptor[1] = GCPtr;
2935	}
2936
2937	int rc = VMXR0InvVPID(enmTlbFlush, &au64Descriptor[0]);
2938	AssertMsg(rc == VINF_SUCCESS,
2939	("VMXR0InvVPID %#x %u %RGv failed with %Rrc\n", enmTlbFlush, pVCpu ? pVCpu->hmr0.s.uCurrentAsid : 0, GCPtr, rc));
2940
2941	if ( RT_SUCCESS(rc)
2942	&& pVCpu)
2943	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid);
2944	NOREF(rc);
2945	}
2946
2947
2948	/**
2949	* Invalidates a guest page by guest virtual address. Only relevant for EPT/VPID,
2950	* otherwise there is nothing really to invalidate.
2951	*
2952	* @returns VBox status code.
2953	* @param pVCpu The cross context virtual CPU structure.
2954	* @param GCVirt Guest virtual address of the page to invalidate.
2955	*/
2956	VMMR0DECL(int) VMXR0InvalidatePage(PVMCPUCC pVCpu, RTGCPTR GCVirt)
2957	{
2958	AssertPtr(pVCpu);
2959	LogFlowFunc(("pVCpu=%p GCVirt=%RGv\n", pVCpu, GCVirt));
2960
2961	if (!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TLB_FLUSH))
2962	{
2963	/*
2964	* We must invalidate the guest TLB entry in either case, we cannot ignore it even for
2965	* the EPT case. See @bugref{6043} and @bugref{6177}.
2966	*
2967	* Set the VMCPU_FF_TLB_FLUSH force flag and flush before VM-entry in hmR0VmxFlushTLB*()
2968	* as this function maybe called in a loop with individual addresses.
2969	*/
2970	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2971	if (pVM->hmr0.s.vmx.fVpid)
2972	{
2973	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
2974	{
2975	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_INDIV_ADDR, GCVirt);
2976	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt);
2977	}
2978	else
2979	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2980	}
2981	else if (pVM->hmr0.s.fNestedPaging)
2982	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2983	}
2984
2985	return VINF_SUCCESS;
2986	}
2987
2988
2989	/**
2990	* Dummy placeholder for tagged-TLB flush handling before VM-entry. Used in the
2991	* case where neither EPT nor VPID is supported by the CPU.
2992	*
2993	* @param pHostCpu The HM physical-CPU structure.
2994	* @param pVCpu The cross context virtual CPU structure.
2995	*
2996	* @remarks Called with interrupts disabled.
2997	*/
2998	static void hmR0VmxFlushTaggedTlbNone(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu)
2999	{
3000	AssertPtr(pVCpu);
3001	AssertPtr(pHostCpu);
3002
3003	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
3004
3005	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3006	pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
3007	pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3008	pVCpu->hmr0.s.fForceTLBFlush = false;
3009	return;
3010	}
3011
3012
3013	/**
3014	* Flushes the tagged-TLB entries for EPT+VPID CPUs as necessary.
3015	*
3016	* @param pHostCpu The HM physical-CPU structure.
3017	* @param pVCpu The cross context virtual CPU structure.
3018	* @param pVmcsInfo The VMCS info. object.
3019	*
3020	* @remarks All references to "ASID" in this function pertains to "VPID" in Intel's
3021	* nomenclature. The reason is, to avoid confusion in compare statements
3022	* since the host-CPU copies are named "ASID".
3023	*
3024	* @remarks Called with interrupts disabled.
3025	*/
3026	static void hmR0VmxFlushTaggedTlbBoth(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
3027	{
3028	#ifdef VBOX_WITH_STATISTICS
3029	bool fTlbFlushed = false;
3030	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { fTlbFlushed = true; } while (0)
3031	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { \
3032	if (!fTlbFlushed) \
3033	STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); \
3034	} while (0)
3035	#else
3036	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { } while (0)
3037	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { } while (0)
3038	#endif
3039
3040	AssertPtr(pVCpu);
3041	AssertPtr(pHostCpu);
3042	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3043
3044	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3045	AssertMsg(pVM->hmr0.s.fNestedPaging && pVM->hmr0.s.vmx.fVpid,
3046	("hmR0VmxFlushTaggedTlbBoth cannot be invoked unless NestedPaging & VPID are enabled."
3047	"fNestedPaging=%RTbool fVpid=%RTbool", pVM->hmr0.s.fNestedPaging, pVM->hmr0.s.vmx.fVpid));
3048
3049	/*
3050	* Force a TLB flush for the first world-switch if the current CPU differs from the one we
3051	* ran on last. If the TLB flush count changed, another VM (VCPU rather) has hit the ASID
3052	* limit while flushing the TLB or the host CPU is online after a suspend/resume, so we
3053	* cannot reuse the current ASID anymore.
3054	*/
3055	if ( pVCpu->hmr0.s.idLastCpu != pHostCpu->idCpu
3056	\|\| pVCpu->hmr0.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3057	{
3058	++pHostCpu->uCurrentAsid;
3059	if (pHostCpu->uCurrentAsid >= g_uHmMaxAsid)
3060	{
3061	pHostCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0. */
3062	pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
3063	pHostCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
3064	}
3065
3066	pVCpu->hmr0.s.uCurrentAsid = pHostCpu->uCurrentAsid;
3067	pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
3068	pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3069
3070	/*
3071	* Flush by EPT when we get rescheduled to a new host CPU to ensure EPT-only tagged mappings are also
3072	* invalidated. We don't need to flush-by-VPID here as flushing by EPT covers it. See @bugref{6568}.
3073	*/
3074	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hmr0.s.vmx.enmTlbFlushEpt);
3075	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3076	HMVMX_SET_TAGGED_TLB_FLUSHED();
3077	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
3078	}
3079	else if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH)) /* Check for explicit TLB flushes. */
3080	{
3081	/*
3082	* Changes to the EPT paging structure by VMM requires flushing-by-EPT as the CPU
3083	* creates guest-physical (ie. only EPT-tagged) mappings while traversing the EPT
3084	* tables when EPT is in use. Flushing-by-VPID will only flush linear (only
3085	* VPID-tagged) and combined (EPT+VPID tagged) mappings but not guest-physical
3086	* mappings, see @bugref{6568}.
3087	*
3088	* See Intel spec. 28.3.2 "Creating and Using Cached Translation Information".
3089	*/
3090	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hmr0.s.vmx.enmTlbFlushEpt);
3091	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3092	HMVMX_SET_TAGGED_TLB_FLUSHED();
3093	}
3094	else if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3095	{
3096	/*
3097	* The nested-guest specifies its own guest-physical address to use as the APIC-access
3098	* address which requires flushing the TLB of EPT cached structures.
3099	*
3100	* See Intel spec. 28.3.3.4 "Guidelines for Use of the INVEPT Instruction".
3101	*/
3102	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVM->hmr0.s.vmx.enmTlbFlushEpt);
3103	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3104	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3105	HMVMX_SET_TAGGED_TLB_FLUSHED();
3106	}
3107
3108
3109	pVCpu->hmr0.s.fForceTLBFlush = false;
3110	HMVMX_UPDATE_FLUSH_SKIPPED_STAT();
3111
3112	Assert(pVCpu->hmr0.s.idLastCpu == pHostCpu->idCpu);
3113	Assert(pVCpu->hmr0.s.cTlbFlushes == pHostCpu->cTlbFlushes);
3114	AssertMsg(pVCpu->hmr0.s.cTlbFlushes == pHostCpu->cTlbFlushes,
3115	("Flush count mismatch for cpu %d (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hmr0.s.cTlbFlushes, pHostCpu->cTlbFlushes));
3116	AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < g_uHmMaxAsid,
3117	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pHostCpu->idCpu,
3118	pHostCpu->uCurrentAsid, pHostCpu->cTlbFlushes, pVCpu->hmr0.s.idLastCpu, pVCpu->hmr0.s.cTlbFlushes));
3119	AssertMsg(pVCpu->hmr0.s.uCurrentAsid >= 1 && pVCpu->hmr0.s.uCurrentAsid < g_uHmMaxAsid,
3120	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pHostCpu->idCpu, pVCpu->hmr0.s.uCurrentAsid));
3121
3122	/* Update VMCS with the VPID. */
3123	int rc = VMXWriteVmcs16(VMX_VMCS16_VPID, pVCpu->hmr0.s.uCurrentAsid);
3124	AssertRC(rc);
3125
3126	#undef HMVMX_SET_TAGGED_TLB_FLUSHED
3127	}
3128
3129
3130	/**
3131	* Flushes the tagged-TLB entries for EPT CPUs as necessary.
3132	*
3133	* @param pHostCpu The HM physical-CPU structure.
3134	* @param pVCpu The cross context virtual CPU structure.
3135	* @param pVmcsInfo The VMCS info. object.
3136	*
3137	* @remarks Called with interrupts disabled.
3138	*/
3139	static void hmR0VmxFlushTaggedTlbEpt(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
3140	{
3141	AssertPtr(pVCpu);
3142	AssertPtr(pHostCpu);
3143	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3144	AssertMsg(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked without NestedPaging."));
3145	AssertMsg(!pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fVpid, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with VPID."));
3146
3147	/*
3148	* Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
3149	* A change in the TLB flush count implies the host CPU is online after a suspend/resume.
3150	*/
3151	if ( pVCpu->hmr0.s.idLastCpu != pHostCpu->idCpu
3152	\|\| pVCpu->hmr0.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3153	{
3154	pVCpu->hmr0.s.fForceTLBFlush = true;
3155	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3156	}
3157
3158	/* Check for explicit TLB flushes. */
3159	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
3160	{
3161	pVCpu->hmr0.s.fForceTLBFlush = true;
3162	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3163	}
3164
3165	/* Check for TLB flushes while switching to/from a nested-guest. */
3166	if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3167	{
3168	pVCpu->hmr0.s.fForceTLBFlush = true;
3169	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3170	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3171	}
3172
3173	pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
3174	pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3175
3176	if (pVCpu->hmr0.s.fForceTLBFlush)
3177	{
3178	hmR0VmxFlushEpt(pVCpu, pVmcsInfo, pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.enmTlbFlushEpt);
3179	pVCpu->hmr0.s.fForceTLBFlush = false;
3180	}
3181	}
3182
3183
3184	/**
3185	* Flushes the tagged-TLB entries for VPID CPUs as necessary.
3186	*
3187	* @param pHostCpu The HM physical-CPU structure.
3188	* @param pVCpu The cross context virtual CPU structure.
3189	*
3190	* @remarks Called with interrupts disabled.
3191	*/
3192	static void hmR0VmxFlushTaggedTlbVpid(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu)
3193	{
3194	AssertPtr(pVCpu);
3195	AssertPtr(pHostCpu);
3196	Assert(pHostCpu->idCpu != NIL_RTCPUID);
3197	AssertMsg(pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fVpid, ("hmR0VmxFlushTlbVpid cannot be invoked without VPID."));
3198	AssertMsg(!pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging, ("hmR0VmxFlushTlbVpid cannot be invoked with NestedPaging"));
3199
3200	/*
3201	* Force a TLB flush for the first world switch if the current CPU differs from the one we
3202	* ran on last. If the TLB flush count changed, another VM (VCPU rather) has hit the ASID
3203	* limit while flushing the TLB or the host CPU is online after a suspend/resume, so we
3204	* cannot reuse the current ASID anymore.
3205	*/
3206	if ( pVCpu->hmr0.s.idLastCpu != pHostCpu->idCpu
3207	\|\| pVCpu->hmr0.s.cTlbFlushes != pHostCpu->cTlbFlushes)
3208	{
3209	pVCpu->hmr0.s.fForceTLBFlush = true;
3210	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
3211	}
3212
3213	/* Check for explicit TLB flushes. */
3214	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
3215	{
3216	/*
3217	* If we ever support VPID flush combinations other than ALL or SINGLE-context (see
3218	* hmR0VmxSetupTaggedTlb()) we would need to explicitly flush in this case (add an
3219	* fExplicitFlush = true here and change the pHostCpu->fFlushAsidBeforeUse check below to
3220	* include fExplicitFlush's too) - an obscure corner case.
3221	*/
3222	pVCpu->hmr0.s.fForceTLBFlush = true;
3223	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
3224	}
3225
3226	/* Check for TLB flushes while switching to/from a nested-guest. */
3227	if (pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb)
3228	{
3229	pVCpu->hmr0.s.fForceTLBFlush = true;
3230	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = false;
3231	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbNstGst);
3232	}
3233
3234	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3235	pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
3236	if (pVCpu->hmr0.s.fForceTLBFlush)
3237	{
3238	++pHostCpu->uCurrentAsid;
3239	if (pHostCpu->uCurrentAsid >= g_uHmMaxAsid)
3240	{
3241	pHostCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0 */
3242	pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
3243	pHostCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
3244	}
3245
3246	pVCpu->hmr0.s.fForceTLBFlush = false;
3247	pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
3248	pVCpu->hmr0.s.uCurrentAsid = pHostCpu->uCurrentAsid;
3249	if (pHostCpu->fFlushAsidBeforeUse)
3250	{
3251	if (pVM->hmr0.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_SINGLE_CONTEXT)
3252	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_SINGLE_CONTEXT, 0 /* GCPtr */);
3253	else if (pVM->hmr0.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_ALL_CONTEXTS)
3254	{
3255	hmR0VmxFlushVpid(pVCpu, VMXTLBFLUSHVPID_ALL_CONTEXTS, 0 /* GCPtr */);
3256	pHostCpu->fFlushAsidBeforeUse = false;
3257	}
3258	else
3259	{
3260	/* hmR0VmxSetupTaggedTlb() ensures we never get here. Paranoia. */
3261	AssertMsgFailed(("Unsupported VPID-flush context type.\n"));
3262	}
3263	}
3264	}
3265
3266	AssertMsg(pVCpu->hmr0.s.cTlbFlushes == pHostCpu->cTlbFlushes,
3267	("Flush count mismatch for cpu %d (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hmr0.s.cTlbFlushes, pHostCpu->cTlbFlushes));
3268	AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < g_uHmMaxAsid,
3269	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pHostCpu->idCpu,
3270	pHostCpu->uCurrentAsid, pHostCpu->cTlbFlushes, pVCpu->hmr0.s.idLastCpu, pVCpu->hmr0.s.cTlbFlushes));
3271	AssertMsg(pVCpu->hmr0.s.uCurrentAsid >= 1 && pVCpu->hmr0.s.uCurrentAsid < g_uHmMaxAsid,
3272	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pHostCpu->idCpu, pVCpu->hmr0.s.uCurrentAsid));
3273
3274	int rc = VMXWriteVmcs16(VMX_VMCS16_VPID, pVCpu->hmr0.s.uCurrentAsid);
3275	AssertRC(rc);
3276	}
3277
3278
3279	/**
3280	* Flushes the guest TLB entry based on CPU capabilities.
3281	*
3282	* @param pHostCpu The HM physical-CPU structure.
3283	* @param pVCpu The cross context virtual CPU structure.
3284	* @param pVmcsInfo The VMCS info. object.
3285	*
3286	* @remarks Called with interrupts disabled.
3287	*/
3288	static void hmR0VmxFlushTaggedTlb(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3289	{
3290	#ifdef HMVMX_ALWAYS_FLUSH_TLB
3291	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
3292	#endif
3293	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3294	switch (pVM->hmr0.s.vmx.enmTlbFlushType)
3295	{
3296	case VMXTLBFLUSHTYPE_EPT_VPID: hmR0VmxFlushTaggedTlbBoth(pHostCpu, pVCpu, pVmcsInfo); break;
3297	case VMXTLBFLUSHTYPE_EPT: hmR0VmxFlushTaggedTlbEpt(pHostCpu, pVCpu, pVmcsInfo); break;
3298	case VMXTLBFLUSHTYPE_VPID: hmR0VmxFlushTaggedTlbVpid(pHostCpu, pVCpu); break;
3299	case VMXTLBFLUSHTYPE_NONE: hmR0VmxFlushTaggedTlbNone(pHostCpu, pVCpu); break;
3300	default:
3301	AssertMsgFailed(("Invalid flush-tag function identifier\n"));
3302	break;
3303	}
3304	/* Don't assert that VMCPU_FF_TLB_FLUSH should no longer be pending. It can be set by other EMTs. */
3305	}
3306
3307
3308	/**
3309	* Sets up the appropriate tagged TLB-flush level and handler for flushing guest
3310	* TLB entries from the host TLB before VM-entry.
3311	*
3312	* @returns VBox status code.
3313	* @param pVM The cross context VM structure.
3314	*/
3315	static int hmR0VmxSetupTaggedTlb(PVMCC pVM)
3316	{
3317	/*
3318	* Determine optimal flush type for nested paging.
3319	* We cannot ignore EPT if no suitable flush-types is supported by the CPU as we've already setup
3320	* unrestricted guest execution (see hmR3InitFinalizeR0()).
3321	*/
3322	if (pVM->hmr0.s.fNestedPaging)
3323	{
3324	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT)
3325	{
3326	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT)
3327	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_SINGLE_CONTEXT;
3328	else if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
3329	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_ALL_CONTEXTS;
3330	else
3331	{
3332	/* Shouldn't happen. EPT is supported but no suitable flush-types supported. */
3333	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3334	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_FLUSH_TYPE_UNSUPPORTED;
3335	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3336	}
3337
3338	/* Make sure the write-back cacheable memory type for EPT is supported. */
3339	if (RT_UNLIKELY(!(g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_MEMTYPE_WB)))
3340	{
3341	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3342	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_MEM_TYPE_NOT_WB;
3343	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3344	}
3345
3346	/* EPT requires a page-walk length of 4. */
3347	if (RT_UNLIKELY(!(g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_PAGE_WALK_LENGTH_4)))
3348	{
3349	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3350	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_PAGE_WALK_LENGTH_UNSUPPORTED;
3351	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3352	}
3353	}
3354	else
3355	{
3356	/* Shouldn't happen. EPT is supported but INVEPT instruction is not supported. */
3357	pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NOT_SUPPORTED;
3358	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_EPT_INVEPT_UNAVAILABLE;
3359	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3360	}
3361	}
3362
3363	/*
3364	* Determine optimal flush type for VPID.
3365	*/
3366	if (pVM->hmr0.s.vmx.fVpid)
3367	{
3368	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID)
3369	{
3370	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT)
3371	pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_SINGLE_CONTEXT;
3372	else if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS)
3373	pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_ALL_CONTEXTS;
3374	else
3375	{
3376	/* Neither SINGLE nor ALL-context flush types for VPID is supported by the CPU. Ignore VPID capability. */
3377	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
3378	LogRelFunc(("Only INDIV_ADDR supported. Ignoring VPID.\n"));
3379	if (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
3380	LogRelFunc(("Only SINGLE_CONTEXT_RETAIN_GLOBALS supported. Ignoring VPID.\n"));
3381	pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NOT_SUPPORTED;
3382	pVM->hmr0.s.vmx.fVpid = false;
3383	}
3384	}
3385	else
3386	{
3387	/* Shouldn't happen. VPID is supported but INVVPID is not supported by the CPU. Ignore VPID capability. */
3388	Log4Func(("VPID supported without INVEPT support. Ignoring VPID.\n"));
3389	pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NOT_SUPPORTED;
3390	pVM->hmr0.s.vmx.fVpid = false;
3391	}
3392	}
3393
3394	/*
3395	* Setup the handler for flushing tagged-TLBs.
3396	*/
3397	if (pVM->hmr0.s.fNestedPaging && pVM->hmr0.s.vmx.fVpid)
3398	pVM->hmr0.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_EPT_VPID;
3399	else if (pVM->hmr0.s.fNestedPaging)
3400	pVM->hmr0.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_EPT;
3401	else if (pVM->hmr0.s.vmx.fVpid)
3402	pVM->hmr0.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_VPID;
3403	else
3404	pVM->hmr0.s.vmx.enmTlbFlushType = VMXTLBFLUSHTYPE_NONE;
3405
3406
3407	/*
3408	* Copy out the result to ring-3.
3409	*/
3410	pVM->hm.s.ForR3.vmx.fVpid = pVM->hmr0.s.vmx.fVpid;
3411	pVM->hm.s.ForR3.vmx.enmTlbFlushType = pVM->hmr0.s.vmx.enmTlbFlushType;
3412	pVM->hm.s.ForR3.vmx.enmTlbFlushEpt = pVM->hmr0.s.vmx.enmTlbFlushEpt;
3413	pVM->hm.s.ForR3.vmx.enmTlbFlushVpid = pVM->hmr0.s.vmx.enmTlbFlushVpid;
3414	return VINF_SUCCESS;
3415	}
3416
3417
3418	/**
3419	* Sets up the LBR MSR ranges based on the host CPU.
3420	*
3421	* @returns VBox status code.
3422	* @param pVM The cross context VM structure.
3423	*/
3424	static int hmR0VmxSetupLbrMsrRange(PVMCC pVM)
3425	{
3426	Assert(pVM->hmr0.s.vmx.fLbr);
3427	uint32_t idLbrFromIpMsrFirst;
3428	uint32_t idLbrFromIpMsrLast;
3429	uint32_t idLbrToIpMsrFirst;
3430	uint32_t idLbrToIpMsrLast;
3431	uint32_t idLbrTosMsr;
3432
3433	/*
3434	* Determine the LBR MSRs supported for this host CPU family and model.
3435	*
3436	* See Intel spec. 17.4.8 "LBR Stack".
3437	* See Intel "Model-Specific Registers" spec.
3438	*/
3439	uint32_t const uFamilyModel = (pVM->cpum.ro.HostFeatures.uFamily << 8)
3440	\| pVM->cpum.ro.HostFeatures.uModel;
3441	switch (uFamilyModel)
3442	{
3443	case 0x0f01: case 0x0f02:
3444	idLbrFromIpMsrFirst = MSR_P4_LASTBRANCH_0;
3445	idLbrFromIpMsrLast = MSR_P4_LASTBRANCH_3;
3446	idLbrToIpMsrFirst = 0x0;
3447	idLbrToIpMsrLast = 0x0;
3448	idLbrTosMsr = MSR_P4_LASTBRANCH_TOS;
3449	break;
3450
3451	case 0x065c: case 0x065f: case 0x064e: case 0x065e: case 0x068e:
3452	case 0x069e: case 0x0655: case 0x0666: case 0x067a: case 0x0667:
3453	case 0x066a: case 0x066c: case 0x067d: case 0x067e:
3454	idLbrFromIpMsrFirst = MSR_LASTBRANCH_0_FROM_IP;
3455	idLbrFromIpMsrLast = MSR_LASTBRANCH_31_FROM_IP;
3456	idLbrToIpMsrFirst = MSR_LASTBRANCH_0_TO_IP;
3457	idLbrToIpMsrLast = MSR_LASTBRANCH_31_TO_IP;
3458	idLbrTosMsr = MSR_LASTBRANCH_TOS;
3459	break;
3460
3461	case 0x063d: case 0x0647: case 0x064f: case 0x0656: case 0x063c:
3462	case 0x0645: case 0x0646: case 0x063f: case 0x062a: case 0x062d:
3463	case 0x063a: case 0x063e: case 0x061a: case 0x061e: case 0x061f:
3464	case 0x062e: case 0x0625: case 0x062c: case 0x062f:
3465	idLbrFromIpMsrFirst = MSR_LASTBRANCH_0_FROM_IP;
3466	idLbrFromIpMsrLast = MSR_LASTBRANCH_15_FROM_IP;
3467	idLbrToIpMsrFirst = MSR_LASTBRANCH_0_TO_IP;
3468	idLbrToIpMsrLast = MSR_LASTBRANCH_15_TO_IP;
3469	idLbrTosMsr = MSR_LASTBRANCH_TOS;
3470	break;
3471
3472	case 0x0617: case 0x061d: case 0x060f:
3473	idLbrFromIpMsrFirst = MSR_CORE2_LASTBRANCH_0_FROM_IP;
3474	idLbrFromIpMsrLast = MSR_CORE2_LASTBRANCH_3_FROM_IP;
3475	idLbrToIpMsrFirst = MSR_CORE2_LASTBRANCH_0_TO_IP;
3476	idLbrToIpMsrLast = MSR_CORE2_LASTBRANCH_3_TO_IP;
3477	idLbrTosMsr = MSR_CORE2_LASTBRANCH_TOS;
3478	break;
3479
3480	/* Atom and related microarchitectures we don't care about:
3481	case 0x0637: case 0x064a: case 0x064c: case 0x064d: case 0x065a:
3482	case 0x065d: case 0x061c: case 0x0626: case 0x0627: case 0x0635:
3483	case 0x0636: */
3484	/* All other CPUs: */
3485	default:
3486	{
3487	LogRelFunc(("Could not determine LBR stack size for the CPU model %#x\n", uFamilyModel));
3488	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_LBR_STACK_SIZE_UNKNOWN;
3489	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3490	}
3491	}
3492
3493	/*
3494	* Validate.
3495	*/
3496	uint32_t const cLbrStack = idLbrFromIpMsrLast - idLbrFromIpMsrFirst + 1;
3497	PCVMCPU pVCpu0 = VMCC_GET_CPU_0(pVM);
3498	AssertCompile( RT_ELEMENTS(pVCpu0->hm.s.vmx.VmcsInfo.au64LbrFromIpMsr)
3499	== RT_ELEMENTS(pVCpu0->hm.s.vmx.VmcsInfo.au64LbrToIpMsr));
3500	if (cLbrStack > RT_ELEMENTS(pVCpu0->hm.s.vmx.VmcsInfo.au64LbrFromIpMsr))
3501	{
3502	LogRelFunc(("LBR stack size of the CPU (%u) exceeds our buffer size\n", cLbrStack));
3503	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_LBR_STACK_SIZE_OVERFLOW;
3504	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3505	}
3506	NOREF(pVCpu0);
3507
3508	/*
3509	* Update the LBR info. to the VM struct. for use later.
3510	*/
3511	pVM->hmr0.s.vmx.idLbrTosMsr = idLbrTosMsr;
3512
3513	pVM->hm.s.ForR3.vmx.idLbrFromIpMsrFirst = pVM->hmr0.s.vmx.idLbrFromIpMsrFirst = idLbrFromIpMsrFirst;
3514	pVM->hm.s.ForR3.vmx.idLbrFromIpMsrLast = pVM->hmr0.s.vmx.idLbrFromIpMsrLast = idLbrFromIpMsrLast;
3515
3516	pVM->hm.s.ForR3.vmx.idLbrToIpMsrFirst = pVM->hmr0.s.vmx.idLbrToIpMsrFirst = idLbrToIpMsrFirst;
3517	pVM->hm.s.ForR3.vmx.idLbrToIpMsrLast = pVM->hmr0.s.vmx.idLbrToIpMsrLast = idLbrToIpMsrLast;
3518	return VINF_SUCCESS;
3519	}
3520
3521
3522	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3523	/**
3524	* Sets up the shadow VMCS fields arrays.
3525	*
3526	* This function builds arrays of VMCS fields to sync the shadow VMCS later while
3527	* executing the guest.
3528	*
3529	* @returns VBox status code.
3530	* @param pVM The cross context VM structure.
3531	*/
3532	static int hmR0VmxSetupShadowVmcsFieldsArrays(PVMCC pVM)
3533	{
3534	/*
3535	* Paranoia. Ensure we haven't exposed the VMWRITE-All VMX feature to the guest
3536	* when the host does not support it.
3537	*/
3538	bool const fGstVmwriteAll = pVM->cpum.ro.GuestFeatures.fVmxVmwriteAll;
3539	if ( !fGstVmwriteAll
3540	\|\| (g_HmMsrs.u.vmx.u64Misc & VMX_MISC_VMWRITE_ALL))
3541	{ /* likely. */ }
3542	else
3543	{
3544	LogRelFunc(("VMX VMWRITE-All feature exposed to the guest but host CPU does not support it!\n"));
3545	VMCC_GET_CPU_0(pVM)->hm.s.u32HMError = VMX_UFC_GST_HOST_VMWRITE_ALL;
3546	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3547	}
3548
3549	uint32_t const cVmcsFields = RT_ELEMENTS(g_aVmcsFields);
3550	uint32_t cRwFields = 0;
3551	uint32_t cRoFields = 0;
3552	for (uint32_t i = 0; i < cVmcsFields; i++)
3553	{
3554	VMXVMCSFIELD VmcsField;
3555	VmcsField.u = g_aVmcsFields[i];
3556
3557	/*
3558	* We will be writing "FULL" (64-bit) fields while syncing the shadow VMCS.
3559	* Therefore, "HIGH" (32-bit portion of 64-bit) fields must not be included
3560	* in the shadow VMCS fields array as they would be redundant.
3561	*
3562	* If the VMCS field depends on a CPU feature that is not exposed to the guest,
3563	* we must not include it in the shadow VMCS fields array. Guests attempting to
3564	* VMREAD/VMWRITE such VMCS fields would cause a VM-exit and we shall emulate
3565	* the required behavior.
3566	*/
3567	if ( VmcsField.n.fAccessType == VMX_VMCSFIELD_ACCESS_FULL
3568	&& CPUMIsGuestVmxVmcsFieldValid(pVM, VmcsField.u))
3569	{
3570	/*
3571	* Read-only fields are placed in a separate array so that while syncing shadow
3572	* VMCS fields later (which is more performance critical) we can avoid branches.
3573	*
3574	* However, if the guest can write to all fields (including read-only fields),
3575	* we treat it a as read/write field. Otherwise, writing to these fields would
3576	* cause a VMWRITE instruction error while syncing the shadow VMCS.
3577	*/
3578	if ( fGstVmwriteAll
3579	\|\| !VMXIsVmcsFieldReadOnly(VmcsField.u))
3580	pVM->hmr0.s.vmx.paShadowVmcsFields[cRwFields++] = VmcsField.u;
3581	else
3582	pVM->hmr0.s.vmx.paShadowVmcsRoFields[cRoFields++] = VmcsField.u;
3583	}
3584	}
3585
3586	/* Update the counts. */
3587	pVM->hmr0.s.vmx.cShadowVmcsFields = cRwFields;
3588	pVM->hmr0.s.vmx.cShadowVmcsRoFields = cRoFields;
3589	return VINF_SUCCESS;
3590	}
3591
3592
3593	/**
3594	* Sets up the VMREAD and VMWRITE bitmaps.
3595	*
3596	* @param pVM The cross context VM structure.
3597	*/
3598	static void hmR0VmxSetupVmreadVmwriteBitmaps(PVMCC pVM)
3599	{
3600	/*
3601	* By default, ensure guest attempts to access any VMCS fields cause VM-exits.
3602	*/
3603	uint32_t const cbBitmap = X86_PAGE_4K_SIZE;
3604	uint8_t pbVmreadBitmap = (uint8_t )pVM->hmr0.s.vmx.pvVmreadBitmap;
3605	uint8_t pbVmwriteBitmap = (uint8_t )pVM->hmr0.s.vmx.pvVmwriteBitmap;
3606	ASMMemFill32(pbVmreadBitmap, cbBitmap, UINT32_C(0xffffffff));
3607	ASMMemFill32(pbVmwriteBitmap, cbBitmap, UINT32_C(0xffffffff));
3608
3609	/*
3610	* Skip intercepting VMREAD/VMWRITE to guest read/write fields in the
3611	* VMREAD and VMWRITE bitmaps.
3612	*/
3613	{
3614	uint32_t const *paShadowVmcsFields = pVM->hmr0.s.vmx.paShadowVmcsFields;
3615	uint32_t const cShadowVmcsFields = pVM->hmr0.s.vmx.cShadowVmcsFields;
3616	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
3617	{
3618	uint32_t const uVmcsField = paShadowVmcsFields[i];
3619	Assert(!(uVmcsField & VMX_VMCSFIELD_RSVD_MASK));
3620	Assert(uVmcsField >> 3 < cbBitmap);
3621	ASMBitClear(pbVmreadBitmap + (uVmcsField >> 3), uVmcsField & 7);
3622	ASMBitClear(pbVmwriteBitmap + (uVmcsField >> 3), uVmcsField & 7);
3623	}
3624	}
3625
3626	/*
3627	* Skip intercepting VMREAD for guest read-only fields in the VMREAD bitmap
3628	* if the host supports VMWRITE to all supported VMCS fields.
3629	*/
3630	if (g_HmMsrs.u.vmx.u64Misc & VMX_MISC_VMWRITE_ALL)
3631	{
3632	uint32_t const *paShadowVmcsRoFields = pVM->hmr0.s.vmx.paShadowVmcsRoFields;
3633	uint32_t const cShadowVmcsRoFields = pVM->hmr0.s.vmx.cShadowVmcsRoFields;
3634	for (uint32_t i = 0; i < cShadowVmcsRoFields; i++)
3635	{
3636	uint32_t const uVmcsField = paShadowVmcsRoFields[i];
3637	Assert(!(uVmcsField & VMX_VMCSFIELD_RSVD_MASK));
3638	Assert(uVmcsField >> 3 < cbBitmap);
3639	ASMBitClear(pbVmreadBitmap + (uVmcsField >> 3), uVmcsField & 7);
3640	}
3641	}
3642	}
3643	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
3644
3645
3646	/**
3647	* Sets up the virtual-APIC page address for the VMCS.
3648	*
3649	* @param pVmcsInfo The VMCS info. object.
3650	*/
3651	DECLINLINE(void) hmR0VmxSetupVmcsVirtApicAddr(PCVMXVMCSINFO pVmcsInfo)
3652	{
3653	RTHCPHYS const HCPhysVirtApic = pVmcsInfo->HCPhysVirtApic;
3654	Assert(HCPhysVirtApic != NIL_RTHCPHYS);
3655	Assert(!(HCPhysVirtApic & 0xfff)); /* Bits 11:0 MBZ. */
3656	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL, HCPhysVirtApic);
3657	AssertRC(rc);
3658	}
3659
3660
3661	/**
3662	* Sets up the MSR-bitmap address for the VMCS.
3663	*
3664	* @param pVmcsInfo The VMCS info. object.
3665	*/
3666	DECLINLINE(void) hmR0VmxSetupVmcsMsrBitmapAddr(PCVMXVMCSINFO pVmcsInfo)
3667	{
3668	RTHCPHYS const HCPhysMsrBitmap = pVmcsInfo->HCPhysMsrBitmap;
3669	Assert(HCPhysMsrBitmap != NIL_RTHCPHYS);
3670	Assert(!(HCPhysMsrBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3671	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_MSR_BITMAP_FULL, HCPhysMsrBitmap);
3672	AssertRC(rc);
3673	}
3674
3675
3676	/**
3677	* Sets up the APIC-access page address for the VMCS.
3678	*
3679	* @param pVCpu The cross context virtual CPU structure.
3680	*/
3681	DECLINLINE(void) hmR0VmxSetupVmcsApicAccessAddr(PVMCPUCC pVCpu)
3682	{
3683	RTHCPHYS const HCPhysApicAccess = pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.HCPhysApicAccess;
3684	Assert(HCPhysApicAccess != NIL_RTHCPHYS);
3685	Assert(!(HCPhysApicAccess & 0xfff)); /* Bits 11:0 MBZ. */
3686	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, HCPhysApicAccess);
3687	AssertRC(rc);
3688	}
3689
3690	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
3691
3692	/**
3693	* Sets up the VMREAD bitmap address for the VMCS.
3694	*
3695	* @param pVCpu The cross context virtual CPU structure.
3696	*/
3697	DECLINLINE(void) hmR0VmxSetupVmcsVmreadBitmapAddr(PVMCPUCC pVCpu)
3698	{
3699	RTHCPHYS const HCPhysVmreadBitmap = pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.HCPhysVmreadBitmap;
3700	Assert(HCPhysVmreadBitmap != NIL_RTHCPHYS);
3701	Assert(!(HCPhysVmreadBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3702	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VMREAD_BITMAP_FULL, HCPhysVmreadBitmap);
3703	AssertRC(rc);
3704	}
3705
3706
3707	/**
3708	* Sets up the VMWRITE bitmap address for the VMCS.
3709	*
3710	* @param pVCpu The cross context virtual CPU structure.
3711	*/
3712	DECLINLINE(void) hmR0VmxSetupVmcsVmwriteBitmapAddr(PVMCPUCC pVCpu)
3713	{
3714	RTHCPHYS const HCPhysVmwriteBitmap = pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.HCPhysVmwriteBitmap;
3715	Assert(HCPhysVmwriteBitmap != NIL_RTHCPHYS);
3716	Assert(!(HCPhysVmwriteBitmap & 0xfff)); /* Bits 11:0 MBZ. */
3717	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_VMWRITE_BITMAP_FULL, HCPhysVmwriteBitmap);
3718	AssertRC(rc);
3719	}
3720
3721	#endif
3722
3723	/**
3724	* Sets up the VM-entry MSR load, VM-exit MSR-store and VM-exit MSR-load addresses
3725	* in the VMCS.
3726	*
3727	* @returns VBox status code.
3728	* @param pVmcsInfo The VMCS info. object.
3729	*/
3730	DECLINLINE(int) hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(PVMXVMCSINFO pVmcsInfo)
3731	{
3732	RTHCPHYS const HCPhysGuestMsrLoad = pVmcsInfo->HCPhysGuestMsrLoad;
3733	Assert(HCPhysGuestMsrLoad != NIL_RTHCPHYS);
3734	Assert(!(HCPhysGuestMsrLoad & 0xf)); /* Bits 3:0 MBZ. */
3735
3736	RTHCPHYS const HCPhysGuestMsrStore = pVmcsInfo->HCPhysGuestMsrStore;
3737	Assert(HCPhysGuestMsrStore != NIL_RTHCPHYS);
3738	Assert(!(HCPhysGuestMsrStore & 0xf)); /* Bits 3:0 MBZ. */
3739
3740	RTHCPHYS const HCPhysHostMsrLoad = pVmcsInfo->HCPhysHostMsrLoad;
3741	Assert(HCPhysHostMsrLoad != NIL_RTHCPHYS);
3742	Assert(!(HCPhysHostMsrLoad & 0xf)); /* Bits 3:0 MBZ. */
3743
3744	int rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL, HCPhysGuestMsrLoad); AssertRC(rc);
3745	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL, HCPhysGuestMsrStore); AssertRC(rc);
3746	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL, HCPhysHostMsrLoad); AssertRC(rc);
3747	return VINF_SUCCESS;
3748	}
3749
3750
3751	/**
3752	* Sets up MSR permissions in the MSR bitmap of a VMCS info. object.
3753	*
3754	* @param pVCpu The cross context virtual CPU structure.
3755	* @param pVmcsInfo The VMCS info. object.
3756	*/
3757	static void hmR0VmxSetupVmcsMsrPermissions(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3758	{
3759	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS);
3760
3761	/*
3762	* By default, ensure guest attempts to access any MSR cause VM-exits.
3763	* This shall later be relaxed for specific MSRs as necessary.
3764	*
3765	* Note: For nested-guests, the entire bitmap will be merged prior to
3766	* executing the nested-guest using hardware-assisted VMX and hence there
3767	* is no need to perform this operation. See hmR0VmxMergeMsrBitmapNested.
3768	*/
3769	Assert(pVmcsInfo->pvMsrBitmap);
3770	ASMMemFill32(pVmcsInfo->pvMsrBitmap, X86_PAGE_4K_SIZE, UINT32_C(0xffffffff));
3771
3772	/*
3773	* The guest can access the following MSRs (read, write) without causing
3774	* VM-exits; they are loaded/stored automatically using fields in the VMCS.
3775	*/
3776	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3777	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_CS, VMXMSRPM_ALLOW_RD_WR);
3778	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_ESP, VMXMSRPM_ALLOW_RD_WR);
3779	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SYSENTER_EIP, VMXMSRPM_ALLOW_RD_WR);
3780	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_GS_BASE, VMXMSRPM_ALLOW_RD_WR);
3781	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_FS_BASE, VMXMSRPM_ALLOW_RD_WR);
3782
3783	/*
3784	* The IA32_PRED_CMD and IA32_FLUSH_CMD MSRs are write-only and has no state
3785	* associated with then. We never need to intercept access (writes need to be
3786	* executed without causing a VM-exit, reads will #GP fault anyway).
3787	*
3788	* The IA32_SPEC_CTRL MSR is read/write and has state. We allow the guest to
3789	* read/write them. We swap the guest/host MSR value using the
3790	* auto-load/store MSR area.
3791	*/
3792	if (pVM->cpum.ro.GuestFeatures.fIbpb)
3793	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_PRED_CMD, VMXMSRPM_ALLOW_RD_WR);
3794	if (pVM->cpum.ro.GuestFeatures.fFlushCmd)
3795	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_FLUSH_CMD, VMXMSRPM_ALLOW_RD_WR);
3796	if (pVM->cpum.ro.GuestFeatures.fIbrs)
3797	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_IA32_SPEC_CTRL, VMXMSRPM_ALLOW_RD_WR);
3798
3799	/*
3800	* Allow full read/write access for the following MSRs (mandatory for VT-x)
3801	* required for 64-bit guests.
3802	*/
3803	if (pVM->hmr0.s.fAllow64BitGuests)
3804	{
3805	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_LSTAR, VMXMSRPM_ALLOW_RD_WR);
3806	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K6_STAR, VMXMSRPM_ALLOW_RD_WR);
3807	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_SF_MASK, VMXMSRPM_ALLOW_RD_WR);
3808	hmR0VmxSetMsrPermission(pVCpu, pVmcsInfo, false, MSR_K8_KERNEL_GS_BASE, VMXMSRPM_ALLOW_RD_WR);
3809	}
3810
3811	/*
3812	* IA32_EFER MSR is always intercepted, see @bugref{9180#c37}.
3813	*/
3814	#ifdef VBOX_STRICT
3815	Assert(pVmcsInfo->pvMsrBitmap);
3816	uint32_t const fMsrpmEfer = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, MSR_K6_EFER);
3817	Assert(fMsrpmEfer == VMXMSRPM_EXIT_RD_WR);
3818	#endif
3819	}
3820
3821
3822	/**
3823	* Sets up pin-based VM-execution controls in the VMCS.
3824	*
3825	* @returns VBox status code.
3826	* @param pVCpu The cross context virtual CPU structure.
3827	* @param pVmcsInfo The VMCS info. object.
3828	*/
3829	static int hmR0VmxSetupVmcsPinCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3830	{
3831	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3832	uint32_t fVal = g_HmMsrs.u.vmx.PinCtls.n.allowed0; /* Bits set here must always be set. */
3833	uint32_t const fZap = g_HmMsrs.u.vmx.PinCtls.n.allowed1; /* Bits cleared here must always be cleared. */
3834
3835	fVal \|= VMX_PIN_CTLS_EXT_INT_EXIT /* External interrupts cause a VM-exit. */
3836	\| VMX_PIN_CTLS_NMI_EXIT; /* Non-maskable interrupts (NMIs) cause a VM-exit. */
3837
3838	if (g_HmMsrs.u.vmx.PinCtls.n.allowed1 & VMX_PIN_CTLS_VIRT_NMI)
3839	fVal \|= VMX_PIN_CTLS_VIRT_NMI; /* Use virtual NMIs and virtual-NMI blocking features. */
3840
3841	/* Enable the VMX-preemption timer. */
3842	if (pVM->hmr0.s.vmx.fUsePreemptTimer)
3843	{
3844	Assert(g_HmMsrs.u.vmx.PinCtls.n.allowed1 & VMX_PIN_CTLS_PREEMPT_TIMER);
3845	fVal \|= VMX_PIN_CTLS_PREEMPT_TIMER;
3846	}
3847
3848	#if 0
3849	/* Enable posted-interrupt processing. */
3850	if (pVM->hm.s.fPostedIntrs)
3851	{
3852	Assert(g_HmMsrs.u.vmx.PinCtls.n.allowed1 & VMX_PIN_CTLS_POSTED_INT);
3853	Assert(g_HmMsrs.u.vmx.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_ACK_EXT_INT);
3854	fVal \|= VMX_PIN_CTLS_POSTED_INT;
3855	}
3856	#endif
3857
3858	if ((fVal & fZap) != fVal)
3859	{
3860	LogRelFunc(("Invalid pin-based VM-execution controls combo! Cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3861	g_HmMsrs.u.vmx.PinCtls.n.allowed0, fVal, fZap));
3862	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PIN_EXEC;
3863	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3864	}
3865
3866	/* Commit it to the VMCS and update our cache. */
3867	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, fVal);
3868	AssertRC(rc);
3869	pVmcsInfo->u32PinCtls = fVal;
3870
3871	return VINF_SUCCESS;
3872	}
3873
3874
3875	/**
3876	* Sets up secondary processor-based VM-execution controls in the VMCS.
3877	*
3878	* @returns VBox status code.
3879	* @param pVCpu The cross context virtual CPU structure.
3880	* @param pVmcsInfo The VMCS info. object.
3881	*/
3882	static int hmR0VmxSetupVmcsProcCtls2(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3883	{
3884	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3885	uint32_t fVal = g_HmMsrs.u.vmx.ProcCtls2.n.allowed0; /* Bits set here must be set in the VMCS. */
3886	uint32_t const fZap = g_HmMsrs.u.vmx.ProcCtls2.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3887
3888	/* WBINVD causes a VM-exit. */
3889	if (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_WBINVD_EXIT)
3890	fVal \|= VMX_PROC_CTLS2_WBINVD_EXIT;
3891
3892	/* Enable EPT (aka nested-paging). */
3893	if (pVM->hmr0.s.fNestedPaging)
3894	fVal \|= VMX_PROC_CTLS2_EPT;
3895
3896	/* Enable the INVPCID instruction if we expose it to the guest and is supported
3897	by the hardware. Without this, guest executing INVPCID would cause a #UD. */
3898	if ( pVM->cpum.ro.GuestFeatures.fInvpcid
3899	&& (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_INVPCID))
3900	fVal \|= VMX_PROC_CTLS2_INVPCID;
3901
3902	/* Enable VPID. */
3903	if (pVM->hmr0.s.vmx.fVpid)
3904	fVal \|= VMX_PROC_CTLS2_VPID;
3905
3906	/* Enable unrestricted guest execution. */
3907	if (pVM->hmr0.s.vmx.fUnrestrictedGuest)
3908	fVal \|= VMX_PROC_CTLS2_UNRESTRICTED_GUEST;
3909
3910	#if 0
3911	if (pVM->hm.s.fVirtApicRegs)
3912	{
3913	/* Enable APIC-register virtualization. */
3914	Assert(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_APIC_REG_VIRT);
3915	fVal \|= VMX_PROC_CTLS2_APIC_REG_VIRT;
3916
3917	/* Enable virtual-interrupt delivery. */
3918	Assert(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_INTR_DELIVERY);
3919	fVal \|= VMX_PROC_CTLS2_VIRT_INTR_DELIVERY;
3920	}
3921	#endif
3922
3923	/* Virtualize-APIC accesses if supported by the CPU. The virtual-APIC page is
3924	where the TPR shadow resides. */
3925	/** @todo VIRT_X2APIC support, it's mutually exclusive with this. So must be
3926	* done dynamically. */
3927	if (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
3928	{
3929	fVal \|= VMX_PROC_CTLS2_VIRT_APIC_ACCESS;
3930	hmR0VmxSetupVmcsApicAccessAddr(pVCpu);
3931	}
3932
3933	/* Enable the RDTSCP instruction if we expose it to the guest and is supported
3934	by the hardware. Without this, guest executing RDTSCP would cause a #UD. */
3935	if ( pVM->cpum.ro.GuestFeatures.fRdTscP
3936	&& (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_RDTSCP))
3937	fVal \|= VMX_PROC_CTLS2_RDTSCP;
3938
3939	/* Enable Pause-Loop exiting. */
3940	if ( (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT)
3941	&& pVM->hm.s.vmx.cPleGapTicks
3942	&& pVM->hm.s.vmx.cPleWindowTicks)
3943	{
3944	fVal \|= VMX_PROC_CTLS2_PAUSE_LOOP_EXIT;
3945
3946	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, pVM->hm.s.vmx.cPleGapTicks); AssertRC(rc);
3947	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, pVM->hm.s.vmx.cPleWindowTicks); AssertRC(rc);
3948	}
3949
3950	if ((fVal & fZap) != fVal)
3951	{
3952	LogRelFunc(("Invalid secondary processor-based VM-execution controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
3953	g_HmMsrs.u.vmx.ProcCtls2.n.allowed0, fVal, fZap));
3954	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC2;
3955	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3956	}
3957
3958	/* Commit it to the VMCS and update our cache. */
3959	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, fVal);
3960	AssertRC(rc);
3961	pVmcsInfo->u32ProcCtls2 = fVal;
3962
3963	return VINF_SUCCESS;
3964	}
3965
3966
3967	/**
3968	* Sets up processor-based VM-execution controls in the VMCS.
3969	*
3970	* @returns VBox status code.
3971	* @param pVCpu The cross context virtual CPU structure.
3972	* @param pVmcsInfo The VMCS info. object.
3973	*/
3974	static int hmR0VmxSetupVmcsProcCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
3975	{
3976	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3977	uint32_t fVal = g_HmMsrs.u.vmx.ProcCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
3978	uint32_t const fZap = g_HmMsrs.u.vmx.ProcCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3979
3980	fVal \|= VMX_PROC_CTLS_HLT_EXIT /* HLT causes a VM-exit. */
3981	\| VMX_PROC_CTLS_USE_TSC_OFFSETTING /* Use TSC-offsetting. */
3982	\| VMX_PROC_CTLS_MOV_DR_EXIT /* MOV DRx causes a VM-exit. */
3983	\| VMX_PROC_CTLS_UNCOND_IO_EXIT /* All IO instructions cause a VM-exit. */
3984	\| VMX_PROC_CTLS_RDPMC_EXIT /* RDPMC causes a VM-exit. */
3985	\| VMX_PROC_CTLS_MONITOR_EXIT /* MONITOR causes a VM-exit. */
3986	\| VMX_PROC_CTLS_MWAIT_EXIT; /* MWAIT causes a VM-exit. */
3987
3988	/* We toggle VMX_PROC_CTLS_MOV_DR_EXIT later, check if it's not -always- needed to be set or clear. */
3989	if ( !(g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_MOV_DR_EXIT)
3990	\|\| (g_HmMsrs.u.vmx.ProcCtls.n.allowed0 & VMX_PROC_CTLS_MOV_DR_EXIT))
3991	{
3992	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_MOV_DRX_EXIT;
3993	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3994	}
3995
3996	/* Without nested paging, INVLPG (also affects INVPCID) and MOV CR3 instructions should cause VM-exits. */
3997	if (!pVM->hmr0.s.fNestedPaging)
3998	{
3999	Assert(!pVM->hmr0.s.vmx.fUnrestrictedGuest);
4000	fVal \|= VMX_PROC_CTLS_INVLPG_EXIT
4001	\| VMX_PROC_CTLS_CR3_LOAD_EXIT
4002	\| VMX_PROC_CTLS_CR3_STORE_EXIT;
4003	}
4004
4005	/* Use TPR shadowing if supported by the CPU. */
4006	if ( PDMHasApic(pVM)
4007	&& (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW))
4008	{
4009	fVal \|= VMX_PROC_CTLS_USE_TPR_SHADOW; /* CR8 reads from the Virtual-APIC page. */
4010	/* CR8 writes cause a VM-exit based on TPR threshold. */
4011	Assert(!(fVal & VMX_PROC_CTLS_CR8_STORE_EXIT));
4012	Assert(!(fVal & VMX_PROC_CTLS_CR8_LOAD_EXIT));
4013	hmR0VmxSetupVmcsVirtApicAddr(pVmcsInfo);
4014	}
4015	else
4016	{
4017	/* Some 32-bit CPUs do not support CR8 load/store exiting as MOV CR8 is
4018	invalid on 32-bit Intel CPUs. Set this control only for 64-bit guests. */
4019	if (pVM->hmr0.s.fAllow64BitGuests)
4020	fVal \|= VMX_PROC_CTLS_CR8_STORE_EXIT /* CR8 reads cause a VM-exit. */
4021	\| VMX_PROC_CTLS_CR8_LOAD_EXIT; /* CR8 writes cause a VM-exit. */
4022	}
4023
4024	/* Use MSR-bitmaps if supported by the CPU. */
4025	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
4026	{
4027	fVal \|= VMX_PROC_CTLS_USE_MSR_BITMAPS;
4028	hmR0VmxSetupVmcsMsrBitmapAddr(pVmcsInfo);
4029	}
4030
4031	/* Use the secondary processor-based VM-execution controls if supported by the CPU. */
4032	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
4033	fVal \|= VMX_PROC_CTLS_USE_SECONDARY_CTLS;
4034
4035	if ((fVal & fZap) != fVal)
4036	{
4037	LogRelFunc(("Invalid processor-based VM-execution controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
4038	g_HmMsrs.u.vmx.ProcCtls.n.allowed0, fVal, fZap));
4039	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC;
4040	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
4041	}
4042
4043	/* Commit it to the VMCS and update our cache. */
4044	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, fVal);
4045	AssertRC(rc);
4046	pVmcsInfo->u32ProcCtls = fVal;
4047
4048	/* Set up MSR permissions that don't change through the lifetime of the VM. */
4049	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
4050	hmR0VmxSetupVmcsMsrPermissions(pVCpu, pVmcsInfo);
4051
4052	/* Set up secondary processor-based VM-execution controls if the CPU supports it. */
4053	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
4054	return hmR0VmxSetupVmcsProcCtls2(pVCpu, pVmcsInfo);
4055
4056	/* Sanity check, should not really happen. */
4057	if (RT_LIKELY(!pVM->hmr0.s.vmx.fUnrestrictedGuest))
4058	{ /* likely */ }
4059	else
4060	{
4061	pVCpu->hm.s.u32HMError = VMX_UFC_INVALID_UX_COMBO;
4062	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
4063	}
4064
4065	/* Old CPUs without secondary processor-based VM-execution controls would end up here. */
4066	return VINF_SUCCESS;
4067	}
4068
4069
4070	/**
4071	* Sets up miscellaneous (everything other than Pin, Processor and secondary
4072	* Processor-based VM-execution) control fields in the VMCS.
4073	*
4074	* @returns VBox status code.
4075	* @param pVCpu The cross context virtual CPU structure.
4076	* @param pVmcsInfo The VMCS info. object.
4077	*/
4078	static int hmR0VmxSetupVmcsMiscCtls(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
4079	{
4080	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4081	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUseVmcsShadowing)
4082	{
4083	hmR0VmxSetupVmcsVmreadBitmapAddr(pVCpu);
4084	hmR0VmxSetupVmcsVmwriteBitmapAddr(pVCpu);
4085	}
4086	#endif
4087
4088	Assert(pVmcsInfo->u64VmcsLinkPtr == NIL_RTHCPHYS);
4089	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
4090	AssertRC(rc);
4091
4092	rc = hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(pVmcsInfo);
4093	if (RT_SUCCESS(rc))
4094	{
4095	uint64_t const u64Cr0Mask = hmR0VmxGetFixedCr0Mask(pVCpu);
4096	uint64_t const u64Cr4Mask = hmR0VmxGetFixedCr4Mask(pVCpu);
4097
4098	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, u64Cr0Mask); AssertRC(rc);
4099	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, u64Cr4Mask); AssertRC(rc);
4100
4101	pVmcsInfo->u64Cr0Mask = u64Cr0Mask;
4102	pVmcsInfo->u64Cr4Mask = u64Cr4Mask;
4103
4104	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fLbr)
4105	{
4106	rc = VMXWriteVmcsNw(VMX_VMCS64_GUEST_DEBUGCTL_FULL, MSR_IA32_DEBUGCTL_LBR);
4107	AssertRC(rc);
4108	}
4109	return VINF_SUCCESS;
4110	}
4111	else
4112	LogRelFunc(("Failed to initialize VMCS auto-load/store MSR addresses. rc=%Rrc\n", rc));
4113	return rc;
4114	}
4115
4116
4117	/**
4118	* Sets up the initial exception bitmap in the VMCS based on static conditions.
4119	*
4120	* We shall setup those exception intercepts that don't change during the
4121	* lifetime of the VM here. The rest are done dynamically while loading the
4122	* guest state.
4123	*
4124	* @param pVCpu The cross context virtual CPU structure.
4125	* @param pVmcsInfo The VMCS info. object.
4126	*/
4127	static void hmR0VmxSetupVmcsXcptBitmap(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
4128	{
4129	/*
4130	* The following exceptions are always intercepted:
4131	*
4132	* #AC - To prevent the guest from hanging the CPU and for dealing with
4133	* split-lock detecting host configs.
4134	* #DB - To maintain the DR6 state even when intercepting DRx reads/writes and
4135	* recursive #DBs can cause a CPU hang.
4136	* #PF - To sync our shadow page tables when nested-paging is not used.
4137	*/
4138	bool const fNestedPaging = pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging;
4139	uint32_t const uXcptBitmap = RT_BIT(X86_XCPT_AC)
4140	\| RT_BIT(X86_XCPT_DB)
4141	\| (fNestedPaging ? 0 : RT_BIT(X86_XCPT_PF));
4142
4143	/* Commit it to the VMCS. */
4144	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
4145	AssertRC(rc);
4146
4147	/* Update our cache of the exception bitmap. */
4148	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
4149	}
4150
4151
4152	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4153	/**
4154	* Sets up the VMCS for executing a nested-guest using hardware-assisted VMX.
4155	*
4156	* @returns VBox status code.
4157	* @param pVmcsInfo The VMCS info. object.
4158	*/
4159	static int hmR0VmxSetupVmcsCtlsNested(PVMXVMCSINFO pVmcsInfo)
4160	{
4161	Assert(pVmcsInfo->u64VmcsLinkPtr == NIL_RTHCPHYS);
4162	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS);
4163	AssertRC(rc);
4164
4165	rc = hmR0VmxSetupVmcsAutoLoadStoreMsrAddrs(pVmcsInfo);
4166	if (RT_SUCCESS(rc))
4167	{
4168	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_MSR_BITMAPS)
4169	hmR0VmxSetupVmcsMsrBitmapAddr(pVmcsInfo);
4170
4171	/* Paranoia - We've not yet initialized these, they shall be done while merging the VMCS. */
4172	Assert(!pVmcsInfo->u64Cr0Mask);
4173	Assert(!pVmcsInfo->u64Cr4Mask);
4174	return VINF_SUCCESS;
4175	}
4176	LogRelFunc(("Failed to set up the VMCS link pointer in the nested-guest VMCS. rc=%Rrc\n", rc));
4177	return rc;
4178	}
4179	#endif
4180
4181
4182	/**
4183	* Sets pfnStartVm to the best suited variant.
4184	*
4185	* This must be called whenever anything changes relative to the hmR0VmXStartVm
4186	* variant selection:
4187	* - pVCpu->hm.s.fLoadSaveGuestXcr0
4188	* - HM_WSF_IBPB_ENTRY in pVCpu->hmr0.s.fWorldSwitcher
4189	* - HM_WSF_IBPB_EXIT in pVCpu->hmr0.s.fWorldSwitcher
4190	* - Perhaps: CPUMIsGuestFPUStateActive() (windows only)
4191	* - Perhaps: CPUMCTX.fXStateMask (windows only)
4192	*
4193	* We currently ASSUME that neither HM_WSF_IBPB_ENTRY nor HM_WSF_IBPB_EXIT
4194	* cannot be changed at runtime.
4195	*/
4196	static void hmR0VmxUpdateStartVmFunction(PVMCPUCC pVCpu)
4197	{
4198	static const struct CLANGWORKAROUND { PFNHMVMXSTARTVM pfn; } s_aHmR0VmxStartVmFunctions[] =
4199	{
4200	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_SansL1dEntry_SansMdsEntry_SansIbpbExit },
4201	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_SansL1dEntry_SansMdsEntry_SansIbpbExit },
4202	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_SansL1dEntry_SansMdsEntry_SansIbpbExit },
4203	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_SansL1dEntry_SansMdsEntry_SansIbpbExit },
4204	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_WithL1dEntry_SansMdsEntry_SansIbpbExit },
4205	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_WithL1dEntry_SansMdsEntry_SansIbpbExit },
4206	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_WithL1dEntry_SansMdsEntry_SansIbpbExit },
4207	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_WithL1dEntry_SansMdsEntry_SansIbpbExit },
4208	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_SansL1dEntry_WithMdsEntry_SansIbpbExit },
4209	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_SansL1dEntry_WithMdsEntry_SansIbpbExit },
4210	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_SansL1dEntry_WithMdsEntry_SansIbpbExit },
4211	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_SansL1dEntry_WithMdsEntry_SansIbpbExit },
4212	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_WithL1dEntry_WithMdsEntry_SansIbpbExit },
4213	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_WithL1dEntry_WithMdsEntry_SansIbpbExit },
4214	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_WithL1dEntry_WithMdsEntry_SansIbpbExit },
4215	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_WithL1dEntry_WithMdsEntry_SansIbpbExit },
4216	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_SansL1dEntry_SansMdsEntry_WithIbpbExit },
4217	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_SansL1dEntry_SansMdsEntry_WithIbpbExit },
4218	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_SansL1dEntry_SansMdsEntry_WithIbpbExit },
4219	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_SansL1dEntry_SansMdsEntry_WithIbpbExit },
4220	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_WithL1dEntry_SansMdsEntry_WithIbpbExit },
4221	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_WithL1dEntry_SansMdsEntry_WithIbpbExit },
4222	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_WithL1dEntry_SansMdsEntry_WithIbpbExit },
4223	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_WithL1dEntry_SansMdsEntry_WithIbpbExit },
4224	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_SansL1dEntry_WithMdsEntry_WithIbpbExit },
4225	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_SansL1dEntry_WithMdsEntry_WithIbpbExit },
4226	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_SansL1dEntry_WithMdsEntry_WithIbpbExit },
4227	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_SansL1dEntry_WithMdsEntry_WithIbpbExit },
4228	{ hmR0VmxStartVm_SansXcr0_SansIbpbEntry_WithL1dEntry_WithMdsEntry_WithIbpbExit },
4229	{ hmR0VmxStartVm_WithXcr0_SansIbpbEntry_WithL1dEntry_WithMdsEntry_WithIbpbExit },
4230	{ hmR0VmxStartVm_SansXcr0_WithIbpbEntry_WithL1dEntry_WithMdsEntry_WithIbpbExit },
4231	{ hmR0VmxStartVm_WithXcr0_WithIbpbEntry_WithL1dEntry_WithMdsEntry_WithIbpbExit },
4232	};
4233	uintptr_t const idx = (pVCpu->hmr0.s.fLoadSaveGuestXcr0 ? 1 : 0)
4234	\| (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_IBPB_ENTRY ? 2 : 0)
4235	\| (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_L1D_ENTRY ? 4 : 0)
4236	\| (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_MDS_ENTRY ? 8 : 0)
4237	\| (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_IBPB_EXIT ? 16 : 0);
4238	PFNHMVMXSTARTVM const pfnStartVm = s_aHmR0VmxStartVmFunctions[idx].pfn;
4239	if (pVCpu->hmr0.s.vmx.pfnStartVm != pfnStartVm)
4240	pVCpu->hmr0.s.vmx.pfnStartVm = pfnStartVm;
4241	}
4242
4243
4244	/**
4245	* Selector FNHMSVMVMRUN implementation.
4246	*/
4247	static DECLCALLBACK(int) hmR0VmxStartVmSelector(PVMXVMCSINFO pVmcsInfo, PVMCPUCC pVCpu, bool fResume)
4248	{
4249	hmR0VmxUpdateStartVmFunction(pVCpu);
4250	return pVCpu->hmr0.s.vmx.pfnStartVm(pVmcsInfo, pVCpu, fResume);
4251	}
4252
4253
4254	/**
4255	* Sets up the VMCS for executing a guest (or nested-guest) using hardware-assisted
4256	* VMX.
4257	*
4258	* @returns VBox status code.
4259	* @param pVCpu The cross context virtual CPU structure.
4260	* @param pVmcsInfo The VMCS info. object.
4261	* @param fIsNstGstVmcs Whether this is a nested-guest VMCS.
4262	*/
4263	static int hmR0VmxSetupVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, bool fIsNstGstVmcs)
4264	{
4265	Assert(pVmcsInfo->pvVmcs);
4266	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4267
4268	/* Set the CPU specified revision identifier at the beginning of the VMCS structure. */
4269	(uint32_t )pVmcsInfo->pvVmcs = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_ID);
4270	const char * const pszVmcs = fIsNstGstVmcs ? "nested-guest VMCS" : "guest VMCS";
4271
4272	LogFlowFunc(("\n"));
4273
4274	/*
4275	* Initialize the VMCS using VMCLEAR before loading the VMCS.
4276	* See Intel spec. 31.6 "Preparation And Launching A Virtual Machine".
4277	*/
4278	int rc = hmR0VmxClearVmcs(pVmcsInfo);
4279	if (RT_SUCCESS(rc))
4280	{
4281	rc = hmR0VmxLoadVmcs(pVmcsInfo);
4282	if (RT_SUCCESS(rc))
4283	{
4284	/*
4285	* Initialize the hardware-assisted VMX execution handler for guest and nested-guest VMCS.
4286	* The host is always 64-bit since we no longer support 32-bit hosts.
4287	* Currently we have just a single handler for all guest modes as well, see @bugref{6208#c73}.
4288	*/
4289	if (!fIsNstGstVmcs)
4290	{
4291	rc = hmR0VmxSetupVmcsPinCtls(pVCpu, pVmcsInfo);
4292	if (RT_SUCCESS(rc))
4293	{
4294	rc = hmR0VmxSetupVmcsProcCtls(pVCpu, pVmcsInfo);
4295	if (RT_SUCCESS(rc))
4296	{
4297	rc = hmR0VmxSetupVmcsMiscCtls(pVCpu, pVmcsInfo);
4298	if (RT_SUCCESS(rc))
4299	{
4300	hmR0VmxSetupVmcsXcptBitmap(pVCpu, pVmcsInfo);
4301	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4302	/*
4303	* If a shadow VMCS is allocated for the VMCS info. object, initialize the
4304	* VMCS revision ID and shadow VMCS indicator bit. Also, clear the VMCS
4305	* making it fit for use when VMCS shadowing is later enabled.
4306	*/
4307	if (pVmcsInfo->pvShadowVmcs)
4308	{
4309	VMXVMCSREVID VmcsRevId;
4310	VmcsRevId.u = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_ID);
4311	VmcsRevId.n.fIsShadowVmcs = 1;
4312	(uint32_t )pVmcsInfo->pvShadowVmcs = VmcsRevId.u;
4313	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
4314	if (RT_SUCCESS(rc))
4315	{ /* likely */ }
4316	else
4317	LogRelFunc(("Failed to initialize shadow VMCS. rc=%Rrc\n", rc));
4318	}
4319	#endif
4320	}
4321	else
4322	LogRelFunc(("Failed to setup miscellaneous controls. rc=%Rrc\n", rc));
4323	}
4324	else
4325	LogRelFunc(("Failed to setup processor-based VM-execution controls. rc=%Rrc\n", rc));
4326	}
4327	else
4328	LogRelFunc(("Failed to setup pin-based controls. rc=%Rrc\n", rc));
4329	}
4330	else
4331	{
4332	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4333	rc = hmR0VmxSetupVmcsCtlsNested(pVmcsInfo);
4334	if (RT_SUCCESS(rc))
4335	{ /* likely */ }
4336	else
4337	LogRelFunc(("Failed to initialize nested-guest VMCS. rc=%Rrc\n", rc));
4338	#else
4339	AssertFailed();
4340	#endif
4341	}
4342	}
4343	else
4344	LogRelFunc(("Failed to load the %s. rc=%Rrc\n", rc, pszVmcs));
4345	}
4346	else
4347	LogRelFunc(("Failed to clear the %s. rc=%Rrc\n", rc, pszVmcs));
4348
4349	/* Sync any CPU internal VMCS data back into our VMCS in memory. */
4350	if (RT_SUCCESS(rc))
4351	{
4352	rc = hmR0VmxClearVmcs(pVmcsInfo);
4353	if (RT_SUCCESS(rc))
4354	{ /* likely */ }
4355	else
4356	LogRelFunc(("Failed to clear the %s post setup. rc=%Rrc\n", rc, pszVmcs));
4357	}
4358
4359	/*
4360	* Update the last-error record both for failures and success, so we
4361	* can propagate the status code back to ring-3 for diagnostics.
4362	*/
4363	hmR0VmxUpdateErrorRecord(pVCpu, rc);
4364	NOREF(pszVmcs);
4365	return rc;
4366	}
4367
4368
4369	/**
4370	* Does global VT-x initialization (called during module initialization).
4371	*
4372	* @returns VBox status code.
4373	*/
4374	VMMR0DECL(int) VMXR0GlobalInit(void)
4375	{
4376	#ifdef HMVMX_USE_FUNCTION_TABLE
4377	AssertCompile(VMX_EXIT_MAX + 1 == RT_ELEMENTS(g_aVMExitHandlers));
4378	# ifdef VBOX_STRICT
4379	for (unsigned i = 0; i < RT_ELEMENTS(g_aVMExitHandlers); i++)
4380	Assert(g_aVMExitHandlers[i].pfn);
4381	# endif
4382	#endif
4383	return VINF_SUCCESS;
4384	}
4385
4386
4387	/**
4388	* Does global VT-x termination (called during module termination).
4389	*/
4390	VMMR0DECL(void) VMXR0GlobalTerm()
4391	{
4392	/* Nothing to do currently. */
4393	}
4394
4395
4396	/**
4397	* Sets up and activates VT-x on the current CPU.
4398	*
4399	* @returns VBox status code.
4400	* @param pHostCpu The HM physical-CPU structure.
4401	* @param pVM The cross context VM structure. Can be
4402	* NULL after a host resume operation.
4403	* @param pvCpuPage Pointer to the VMXON region (can be NULL if @a
4404	* fEnabledByHost is @c true).
4405	* @param HCPhysCpuPage Physical address of the VMXON region (can be 0 if
4406	* @a fEnabledByHost is @c true).
4407	* @param fEnabledByHost Set if SUPR0EnableVTx() or similar was used to
4408	* enable VT-x on the host.
4409	* @param pHwvirtMsrs Pointer to the hardware-virtualization MSRs.
4410	*/
4411	VMMR0DECL(int) VMXR0EnableCpu(PHMPHYSCPU pHostCpu, PVMCC pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost,
4412	PCSUPHWVIRTMSRS pHwvirtMsrs)
4413	{
4414	AssertPtr(pHostCpu);
4415	AssertPtr(pHwvirtMsrs);
4416	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4417
4418	/* Enable VT-x if it's not already enabled by the host. */
4419	if (!fEnabledByHost)
4420	{
4421	int rc = hmR0VmxEnterRootMode(pHostCpu, pVM, HCPhysCpuPage, pvCpuPage);
4422	if (RT_FAILURE(rc))
4423	return rc;
4424	}
4425
4426	/*
4427	* Flush all EPT tagged-TLB entries (in case VirtualBox or any other hypervisor have been
4428	* using EPTPs) so we don't retain any stale guest-physical mappings which won't get
4429	* invalidated when flushing by VPID.
4430	*/
4431	if (pHwvirtMsrs->u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
4432	{
4433	hmR0VmxFlushEpt(NULL /* pVCpu /, NULL / pVmcsInfo */, VMXTLBFLUSHEPT_ALL_CONTEXTS);
4434	pHostCpu->fFlushAsidBeforeUse = false;
4435	}
4436	else
4437	pHostCpu->fFlushAsidBeforeUse = true;
4438
4439	/* Ensure each VCPU scheduled on this CPU gets a new VPID on resume. See @bugref{6255}. */
4440	++pHostCpu->cTlbFlushes;
4441
4442	return VINF_SUCCESS;
4443	}
4444
4445
4446	/**
4447	* Deactivates VT-x on the current CPU.
4448	*
4449	* @returns VBox status code.
4450	* @param pHostCpu The HM physical-CPU structure.
4451	* @param pvCpuPage Pointer to the VMXON region.
4452	* @param HCPhysCpuPage Physical address of the VMXON region.
4453	*
4454	* @remarks This function should never be called when SUPR0EnableVTx() or
4455	* similar was used to enable VT-x on the host.
4456	*/
4457	VMMR0DECL(int) VMXR0DisableCpu(PHMPHYSCPU pHostCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
4458	{
4459	RT_NOREF2(pvCpuPage, HCPhysCpuPage);
4460
4461	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4462	return hmR0VmxLeaveRootMode(pHostCpu);
4463	}
4464
4465
4466	/**
4467	* Does per-VM VT-x initialization.
4468	*
4469	* @returns VBox status code.
4470	* @param pVM The cross context VM structure.
4471	*/
4472	VMMR0DECL(int) VMXR0InitVM(PVMCC pVM)
4473	{
4474	AssertPtr(pVM);
4475	LogFlowFunc(("pVM=%p\n", pVM));
4476
4477	hmR0VmxStructsInit(pVM);
4478	int rc = hmR0VmxStructsAlloc(pVM);
4479	if (RT_FAILURE(rc))
4480	{
4481	LogRelFunc(("Failed to allocated VMX structures. rc=%Rrc\n", rc));
4482	return rc;
4483	}
4484
4485	/* Setup the crash dump page. */
4486	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4487	strcpy((char *)pVM->hmr0.s.vmx.pbScratch, "SCRATCH Magic");
4488	(uint64_t )(pVM->hmr0.s.vmx.pbScratch + 16) = UINT64_C(0xdeadbeefdeadbeef);
4489	#endif
4490	return VINF_SUCCESS;
4491	}
4492
4493
4494	/**
4495	* Does per-VM VT-x termination.
4496	*
4497	* @returns VBox status code.
4498	* @param pVM The cross context VM structure.
4499	*/
4500	VMMR0DECL(int) VMXR0TermVM(PVMCC pVM)
4501	{
4502	AssertPtr(pVM);
4503	LogFlowFunc(("pVM=%p\n", pVM));
4504
4505	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
4506	if (pVM->hmr0.s.vmx.pbScratch)
4507	RT_BZERO(pVM->hmr0.s.vmx.pbScratch, X86_PAGE_4K_SIZE);
4508	#endif
4509	hmR0VmxStructsFree(pVM);
4510	return VINF_SUCCESS;
4511	}
4512
4513
4514	/**
4515	* Sets up the VM for execution using hardware-assisted VMX.
4516	* This function is only called once per-VM during initialization.
4517	*
4518	* @returns VBox status code.
4519	* @param pVM The cross context VM structure.
4520	*/
4521	VMMR0DECL(int) VMXR0SetupVM(PVMCC pVM)
4522	{
4523	AssertPtr(pVM);
4524	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4525
4526	LogFlowFunc(("pVM=%p\n", pVM));
4527
4528	/*
4529	* At least verify if VMX is enabled, since we can't check if we're in VMX root mode or not
4530	* without causing a #GP.
4531	*/
4532	RTCCUINTREG const uHostCr4 = ASMGetCR4();
4533	if (RT_LIKELY(uHostCr4 & X86_CR4_VMXE))
4534	{ /* likely */ }
4535	else
4536	return VERR_VMX_NOT_IN_VMX_ROOT_MODE;
4537
4538	/*
4539	* Check that nested paging is supported if enabled and copy over the flag to the
4540	* ring-0 only structure.
4541	*/
4542	bool const fNestedPaging = pVM->hm.s.fNestedPagingCfg;
4543	AssertReturn( !fNestedPaging
4544	\|\| (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_EPT), /** @todo use a ring-0 copy of ProcCtls2.n.allowed1 */
4545	VERR_INCOMPATIBLE_CONFIG);
4546	pVM->hmr0.s.fNestedPaging = fNestedPaging;
4547	pVM->hmr0.s.fAllow64BitGuests = pVM->hm.s.fAllow64BitGuestsCfg;
4548
4549	/*
4550	* Without unrestricted guest execution, pRealModeTSS and pNonPagingModeEPTPageTable must
4551	* always be allocated. We no longer support the highly unlikely case of unrestricted guest
4552	* without pRealModeTSS, see hmR3InitFinalizeR0Intel().
4553	*/
4554	bool const fUnrestrictedGuest = pVM->hm.s.vmx.fUnrestrictedGuestCfg;
4555	AssertReturn( !fUnrestrictedGuest
4556	\|\| ( (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST)
4557	&& fNestedPaging),
4558	VERR_INCOMPATIBLE_CONFIG);
4559	if ( !fUnrestrictedGuest
4560	&& ( !pVM->hm.s.vmx.pNonPagingModeEPTPageTable
4561	\|\| !pVM->hm.s.vmx.pRealModeTSS))
4562	{
4563	LogRelFunc(("Invalid real-on-v86 state.\n"));
4564	return VERR_INTERNAL_ERROR;
4565	}
4566	pVM->hmr0.s.vmx.fUnrestrictedGuest = fUnrestrictedGuest;
4567
4568	/* Initialize these always, see hmR3InitFinalizeR0().*/
4569	pVM->hm.s.ForR3.vmx.enmTlbFlushEpt = pVM->hmr0.s.vmx.enmTlbFlushEpt = VMXTLBFLUSHEPT_NONE;
4570	pVM->hm.s.ForR3.vmx.enmTlbFlushVpid = pVM->hmr0.s.vmx.enmTlbFlushVpid = VMXTLBFLUSHVPID_NONE;
4571
4572	/* Setup the tagged-TLB flush handlers. */
4573	int rc = hmR0VmxSetupTaggedTlb(pVM);
4574	if (RT_FAILURE(rc))
4575	{
4576	LogRelFunc(("Failed to setup tagged TLB. rc=%Rrc\n", rc));
4577	return rc;
4578	}
4579
4580	/* Determine LBR capabilities. */
4581	pVM->hmr0.s.vmx.fLbr = pVM->hm.s.vmx.fLbrCfg;
4582	if (pVM->hmr0.s.vmx.fLbr)
4583	{
4584	rc = hmR0VmxSetupLbrMsrRange(pVM);
4585	if (RT_FAILURE(rc))
4586	{
4587	LogRelFunc(("Failed to setup LBR MSR range. rc=%Rrc\n", rc));
4588	return rc;
4589	}
4590	}
4591
4592	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4593	/* Setup the shadow VMCS fields array and VMREAD/VMWRITE bitmaps. */
4594	if (pVM->hmr0.s.vmx.fUseVmcsShadowing)
4595	{
4596	rc = hmR0VmxSetupShadowVmcsFieldsArrays(pVM);
4597	if (RT_SUCCESS(rc))
4598	hmR0VmxSetupVmreadVmwriteBitmaps(pVM);
4599	else
4600	{
4601	LogRelFunc(("Failed to setup shadow VMCS fields arrays. rc=%Rrc\n", rc));
4602	return rc;
4603	}
4604	}
4605	#endif
4606
4607	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
4608	{
4609	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
4610	Log4Func(("pVCpu=%p idCpu=%RU32\n", pVCpu, pVCpu->idCpu));
4611
4612	pVCpu->hmr0.s.vmx.pfnStartVm = hmR0VmxStartVmSelector;
4613
4614	rc = hmR0VmxSetupVmcs(pVCpu, &pVCpu->hmr0.s.vmx.VmcsInfo, false /* fIsNstGstVmcs */);
4615	if (RT_SUCCESS(rc))
4616	{
4617	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4618	if (pVM->cpum.ro.GuestFeatures.fVmx)
4619	{
4620	rc = hmR0VmxSetupVmcs(pVCpu, &pVCpu->hmr0.s.vmx.VmcsInfoNstGst, true /* fIsNstGstVmcs */);
4621	if (RT_SUCCESS(rc))
4622	{ /* likely */ }
4623	else
4624	{
4625	LogRelFunc(("Nested-guest VMCS setup failed. rc=%Rrc\n", rc));
4626	return rc;
4627	}
4628	}
4629	#endif
4630	}
4631	else
4632	{
4633	LogRelFunc(("VMCS setup failed. rc=%Rrc\n", rc));
4634	return rc;
4635	}
4636	}
4637
4638	return VINF_SUCCESS;
4639	}
4640
4641
4642	/**
4643	* Saves the host control registers (CR0, CR3, CR4) into the host-state area in
4644	* the VMCS.
4645	* @returns CR4 for passing along to hmR0VmxExportHostSegmentRegs.
4646	*/
4647	static uint64_t hmR0VmxExportHostControlRegs(void)
4648	{
4649	int rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR0, ASMGetCR0()); AssertRC(rc);
4650	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR3, ASMGetCR3()); AssertRC(rc);
4651	uint64_t uHostCr4 = ASMGetCR4();
4652	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_CR4, uHostCr4); AssertRC(rc);
4653	return uHostCr4;
4654	}
4655
4656
4657	/**
4658	* Saves the host segment registers and GDTR, IDTR, (TR, GS and FS bases) into
4659	* the host-state area in the VMCS.
4660	*
4661	* @returns VBox status code.
4662	* @param pVCpu The cross context virtual CPU structure.
4663	* @param uHostCr4 The host CR4 value.
4664	*/
4665	static int hmR0VmxExportHostSegmentRegs(PVMCPUCC pVCpu, uint64_t uHostCr4)
4666	{
4667	/*
4668	* If we've executed guest code using hardware-assisted VMX, the host-state bits
4669	* will be messed up. We should -not- save the messed up state without restoring
4670	* the original host-state, see @bugref{7240}.
4671	*
4672	* This apparently can happen (most likely the FPU changes), deal with it rather than
4673	* asserting. Was observed booting Solaris 10u10 32-bit guest.
4674	*/
4675	if (pVCpu->hmr0.s.vmx.fRestoreHostFlags > VMX_RESTORE_HOST_REQUIRED)
4676	{
4677	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hmr0.s.vmx.fRestoreHostFlags,
4678	pVCpu->idCpu));
4679	VMXRestoreHostState(pVCpu->hmr0.s.vmx.fRestoreHostFlags, &pVCpu->hmr0.s.vmx.RestoreHost);
4680	pVCpu->hmr0.s.vmx.fRestoreHostFlags = 0;
4681	}
4682
4683	/*
4684	* Get all the host info.
4685	* ASSUME it is safe to use rdfsbase and friends if the CR4.FSGSBASE bit is set
4686	* without also checking the cpuid bit.
4687	*/
4688	uint32_t fRestoreHostFlags;
4689	#if RT_INLINE_ASM_EXTERNAL
4690	if (uHostCr4 & X86_CR4_FSGSBASE)
4691	{
4692	hmR0VmxExportHostSegmentRegsAsmHlp(&pVCpu->hmr0.s.vmx.RestoreHost, true /fHaveFsGsBase/);
4693	fRestoreHostFlags = VMX_RESTORE_HOST_CAN_USE_WRFSBASE_AND_WRGSBASE;
4694	}
4695	else
4696	{
4697	hmR0VmxExportHostSegmentRegsAsmHlp(&pVCpu->hmr0.s.vmx.RestoreHost, false /fHaveFsGsBase/);
4698	fRestoreHostFlags = 0;
4699	}
4700	RTSEL uSelES = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelES;
4701	RTSEL uSelDS = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelDS;
4702	RTSEL uSelFS = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelFS;
4703	RTSEL uSelGS = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelGS;
4704	#else
4705	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR = ASMGetTR();
4706	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelSS = ASMGetSS();
4707	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelCS = ASMGetCS();
4708	ASMGetGDTR((PRTGDTR)&pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr);
4709	ASMGetIDTR((PRTIDTR)&pVCpu->hmr0.s.vmx.RestoreHost.HostIdtr);
4710	if (uHostCr4 & X86_CR4_FSGSBASE)
4711	{
4712	pVCpu->hmr0.s.vmx.RestoreHost.uHostFSBase = ASMGetFSBase();
4713	pVCpu->hmr0.s.vmx.RestoreHost.uHostGSBase = ASMGetGSBase();
4714	fRestoreHostFlags = VMX_RESTORE_HOST_CAN_USE_WRFSBASE_AND_WRGSBASE;
4715	}
4716	else
4717	{
4718	pVCpu->hmr0.s.vmx.RestoreHost.uHostFSBase = ASMRdMsr(MSR_K8_FS_BASE);
4719	pVCpu->hmr0.s.vmx.RestoreHost.uHostGSBase = ASMRdMsr(MSR_K8_GS_BASE);
4720	fRestoreHostFlags = 0;
4721	}
4722	RTSEL uSelES, uSelDS, uSelFS, uSelGS;
4723	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelDS = uSelDS = ASMGetDS();
4724	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelES = uSelES = ASMGetES();
4725	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelFS = uSelFS = ASMGetFS();
4726	pVCpu->hmr0.s.vmx.RestoreHost.uHostSelGS = uSelGS = ASMGetGS();
4727	#endif
4728
4729	/*
4730	* Determine if the host segment registers are suitable for VT-x. Otherwise use zero to
4731	* gain VM-entry and restore them before we get preempted.
4732	*
4733	* See Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers".
4734	*/
4735	RTSEL const uSelAll = uSelFS \| uSelGS \| uSelES \| uSelDS;
4736	if (uSelAll & (X86_SEL_RPL \| X86_SEL_LDT))
4737	{
4738	if (!(uSelAll & X86_SEL_LDT))
4739	{
4740	#define VMXLOCAL_ADJUST_HOST_SEG(a_Seg, a_uVmcsVar) \
4741	do { \
4742	(a_uVmcsVar) = pVCpu->hmr0.s.vmx.RestoreHost.uHostSel##a_Seg; \
4743	if ((a_uVmcsVar) & X86_SEL_RPL) \
4744	{ \
4745	fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##a_Seg; \
4746	(a_uVmcsVar) = 0; \
4747	} \
4748	} while (0)
4749	VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS);
4750	VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES);
4751	VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS);
4752	VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS);
4753	#undef VMXLOCAL_ADJUST_HOST_SEG
4754	}
4755	else
4756	{
4757	#define VMXLOCAL_ADJUST_HOST_SEG(a_Seg, a_uVmcsVar) \
4758	do { \
4759	(a_uVmcsVar) = pVCpu->hmr0.s.vmx.RestoreHost.uHostSel##a_Seg; \
4760	if ((a_uVmcsVar) & (X86_SEL_RPL \| X86_SEL_LDT)) \
4761	{ \
4762	if (!((a_uVmcsVar) & X86_SEL_LDT)) \
4763	fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##a_Seg; \
4764	else \
4765	{ \
4766	uint32_t const fAttr = ASMGetSegAttr(a_uVmcsVar); \
4767	if ((fAttr & X86_DESC_P) && fAttr != UINT32_MAX) \
4768	fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##a_Seg; \
4769	} \
4770	(a_uVmcsVar) = 0; \
4771	} \
4772	} while (0)
4773	VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS);
4774	VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES);
4775	VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS);
4776	VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS);
4777	#undef VMXLOCAL_ADJUST_HOST_SEG
4778	}
4779	}
4780
4781	/* Verification based on Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers" */
4782	Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR & X86_SEL_RPL)); Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR & X86_SEL_LDT)); Assert(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR);
4783	Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelCS & X86_SEL_RPL)); Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelCS & X86_SEL_LDT)); Assert(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelCS);
4784	Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelSS & X86_SEL_RPL)); Assert(!(pVCpu->hmr0.s.vmx.RestoreHost.uHostSelSS & X86_SEL_LDT));
4785	Assert(!(uSelDS & X86_SEL_RPL)); Assert(!(uSelDS & X86_SEL_LDT));
4786	Assert(!(uSelES & X86_SEL_RPL)); Assert(!(uSelES & X86_SEL_LDT));
4787	Assert(!(uSelFS & X86_SEL_RPL)); Assert(!(uSelFS & X86_SEL_LDT));
4788	Assert(!(uSelGS & X86_SEL_RPL)); Assert(!(uSelGS & X86_SEL_LDT));
4789
4790	/*
4791	* Determine if we need to manually need to restore the GDTR and IDTR limits as VT-x zaps
4792	* them to the maximum limit (0xffff) on every VM-exit.
4793	*/
4794	if (pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.cb != 0xffff)
4795	fRestoreHostFlags \|= VMX_RESTORE_HOST_GDTR;
4796
4797	/*
4798	* IDT limit is effectively capped at 0xfff. (See Intel spec. 6.14.1 "64-Bit Mode IDT" and
4799	* Intel spec. 6.2 "Exception and Interrupt Vectors".) Therefore if the host has the limit
4800	* as 0xfff, VT-x bloating the limit to 0xffff shouldn't cause any different CPU behavior.
4801	* However, several hosts either insists on 0xfff being the limit (Windows Patch Guard) or
4802	* uses the limit for other purposes (darwin puts the CPU ID in there but botches sidt
4803	* alignment in at least one consumer). So, we're only allowing the IDTR.LIMIT to be left
4804	* at 0xffff on hosts where we are sure it won't cause trouble.
4805	*/
4806	#if defined(RT_OS_LINUX) \|\| defined(RT_OS_SOLARIS)
4807	if (pVCpu->hmr0.s.vmx.RestoreHost.HostIdtr.cb < 0x0fff)
4808	#else
4809	if (pVCpu->hmr0.s.vmx.RestoreHost.HostIdtr.cb != 0xffff)
4810	#endif
4811	fRestoreHostFlags \|= VMX_RESTORE_HOST_IDTR;
4812
4813	/*
4814	* Host TR base. Verify that TR selector doesn't point past the GDT. Masking off the TI
4815	* and RPL bits is effectively what the CPU does for "scaling by 8". TI is always 0 and
4816	* RPL should be too in most cases.
4817	*/
4818	RTSEL const uSelTR = pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR;
4819	AssertMsgReturn((uSelTR \| X86_SEL_RPL_LDT) <= pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.cb,
4820	("TR selector exceeds limit. TR=%RTsel cbGdt=%#x\n", uSelTR, pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.cb),
4821	VERR_VMX_INVALID_HOST_STATE);
4822
4823	PCX86DESCHC pDesc = (PCX86DESCHC)(pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.uAddr + (uSelTR & X86_SEL_MASK));
4824	uintptr_t const uTRBase = X86DESC64_BASE(pDesc);
4825
4826	/*
4827	* VT-x unconditionally restores the TR limit to 0x67 and type to 11 (32-bit busy TSS) on
4828	* all VM-exits. The type is the same for 64-bit busy TSS[1]. The limit needs manual
4829	* restoration if the host has something else. Task switching is not supported in 64-bit
4830	* mode[2], but the limit still matters as IOPM is supported in 64-bit mode. Restoring the
4831	* limit lazily while returning to ring-3 is safe because IOPM is not applicable in ring-0.
4832	*
4833	* [1] See Intel spec. 3.5 "System Descriptor Types".
4834	* [2] See Intel spec. 7.2.3 "TSS Descriptor in 64-bit mode".
4835	*/
4836	Assert(pDesc->System.u4Type == 11);
4837	if ( pDesc->System.u16LimitLow != 0x67
4838	\|\| pDesc->System.u4LimitHigh)
4839	{
4840	fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_TR;
4841
4842	/* If the host has made GDT read-only, we would need to temporarily toggle CR0.WP before writing the GDT. */
4843	if (g_fHmHostKernelFeatures & SUPKERNELFEATURES_GDT_READ_ONLY)
4844	fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_READ_ONLY;
4845	if (g_fHmHostKernelFeatures & SUPKERNELFEATURES_GDT_NEED_WRITABLE)
4846	{
4847	/* The GDT is read-only but the writable GDT is available. */
4848	fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_NEED_WRITABLE;
4849	pVCpu->hmr0.s.vmx.RestoreHost.HostGdtrRw.cb = pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.cb;
4850	int rc = SUPR0GetCurrentGdtRw(&pVCpu->hmr0.s.vmx.RestoreHost.HostGdtrRw.uAddr);
4851	AssertRCReturn(rc, rc);
4852	}
4853	}
4854
4855	pVCpu->hmr0.s.vmx.fRestoreHostFlags = fRestoreHostFlags;
4856
4857	/*
4858	* Do all the VMCS updates in one block to assist nested virtualization.
4859	*/
4860	int rc;
4861	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_CS_SEL, pVCpu->hmr0.s.vmx.RestoreHost.uHostSelCS); AssertRC(rc);
4862	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_SS_SEL, pVCpu->hmr0.s.vmx.RestoreHost.uHostSelSS); AssertRC(rc);
4863	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_DS_SEL, uSelDS); AssertRC(rc);
4864	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_ES_SEL, uSelES); AssertRC(rc);
4865	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_FS_SEL, uSelFS); AssertRC(rc);
4866	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_GS_SEL, uSelGS); AssertRC(rc);
4867	rc = VMXWriteVmcs16(VMX_VMCS16_HOST_TR_SEL, pVCpu->hmr0.s.vmx.RestoreHost.uHostSelTR); AssertRC(rc);
4868	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_GDTR_BASE, pVCpu->hmr0.s.vmx.RestoreHost.HostGdtr.uAddr); AssertRC(rc);
4869	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_IDTR_BASE, pVCpu->hmr0.s.vmx.RestoreHost.HostIdtr.uAddr); AssertRC(rc);
4870	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_TR_BASE, uTRBase); AssertRC(rc);
4871	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_FS_BASE, pVCpu->hmr0.s.vmx.RestoreHost.uHostFSBase); AssertRC(rc);
4872	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_GS_BASE, pVCpu->hmr0.s.vmx.RestoreHost.uHostGSBase); AssertRC(rc);
4873
4874	return VINF_SUCCESS;
4875	}
4876
4877
4878	/**
4879	* Exports certain host MSRs in the VM-exit MSR-load area and some in the
4880	* host-state area of the VMCS.
4881	*
4882	* These MSRs will be automatically restored on the host after every successful
4883	* VM-exit.
4884	*
4885	* @param pVCpu The cross context virtual CPU structure.
4886	*
4887	* @remarks No-long-jump zone!!!
4888	*/
4889	static void hmR0VmxExportHostMsrs(PVMCPUCC pVCpu)
4890	{
4891	AssertPtr(pVCpu);
4892
4893	/*
4894	* Save MSRs that we restore lazily (due to preemption or transition to ring-3)
4895	* rather than swapping them on every VM-entry.
4896	*/
4897	hmR0VmxLazySaveHostMsrs(pVCpu);
4898
4899	/*
4900	* Host Sysenter MSRs.
4901	*/
4902	int rc = VMXWriteVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, ASMRdMsr_Low(MSR_IA32_SYSENTER_CS)); AssertRC(rc);
4903	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP)); AssertRC(rc);
4904	rc = VMXWriteVmcsNw(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP)); AssertRC(rc);
4905
4906	/*
4907	* Host EFER MSR.
4908	*
4909	* If the CPU supports the newer VMCS controls for managing EFER, use it. Otherwise it's
4910	* done as part of auto-load/store MSR area in the VMCS, see hmR0VmxExportGuestMsrs().
4911	*/
4912	if (g_fHmVmxSupportsVmcsEfer)
4913	{
4914	rc = VMXWriteVmcs64(VMX_VMCS64_HOST_EFER_FULL, g_uHmVmxHostMsrEfer);
4915	AssertRC(rc);
4916	}
4917
4918	/** @todo IA32_PERF_GLOBALCTRL, IA32_PAT also see
4919	* hmR0VmxExportGuestEntryExitCtls(). */
4920	}
4921
4922
4923	/**
4924	* Figures out if we need to swap the EFER MSR which is particularly expensive.
4925	*
4926	* We check all relevant bits. For now, that's everything besides LMA/LME, as
4927	* these two bits are handled by VM-entry, see hmR0VMxExportGuestEntryExitCtls().
4928	*
4929	* @returns true if we need to load guest EFER, false otherwise.
4930	* @param pVCpu The cross context virtual CPU structure.
4931	* @param pVmxTransient The VMX-transient structure.
4932	*
4933	* @remarks Requires EFER, CR4.
4934	* @remarks No-long-jump zone!!!
4935	*/
4936	static bool hmR0VmxShouldSwapEferMsr(PCVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
4937	{
4938	#ifdef HMVMX_ALWAYS_SWAP_EFER
4939	RT_NOREF2(pVCpu, pVmxTransient);
4940	return true;
4941	#else
4942	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
4943	uint64_t const u64HostEfer = g_uHmVmxHostMsrEfer;
4944	uint64_t const u64GuestEfer = pCtx->msrEFER;
4945
4946	# ifdef VBOX_WITH_NESTED_HWVIRT_VMX
4947	/*
4948	* For nested-guests, we shall honor swapping the EFER MSR when requested by
4949	* the nested-guest.
4950	*/
4951	if ( pVmxTransient->fIsNestedGuest
4952	&& ( CPUMIsGuestVmxEntryCtlsSet(pCtx, VMX_ENTRY_CTLS_LOAD_EFER_MSR)
4953	\|\| CPUMIsGuestVmxExitCtlsSet(pCtx, VMX_EXIT_CTLS_SAVE_EFER_MSR)
4954	\|\| CPUMIsGuestVmxExitCtlsSet(pCtx, VMX_EXIT_CTLS_LOAD_EFER_MSR)))
4955	return true;
4956	# else
4957	RT_NOREF(pVmxTransient);
4958	#endif
4959
4960	/*
4961	* For 64-bit guests, if EFER.SCE bit differs, we need to swap the EFER MSR
4962	* to ensure that the guest's SYSCALL behaviour isn't broken, see @bugref{7386}.
4963	*/
4964	if ( CPUMIsGuestInLongModeEx(pCtx)
4965	&& (u64GuestEfer & MSR_K6_EFER_SCE) != (u64HostEfer & MSR_K6_EFER_SCE))
4966	return true;
4967
4968	/*
4969	* If the guest uses PAE and EFER.NXE bit differs, we need to swap the EFER MSR
4970	* as it affects guest paging. 64-bit paging implies CR4.PAE as well.
4971	*
4972	* See Intel spec. 4.5 "IA-32e Paging".
4973	* See Intel spec. 4.1.1 "Three Paging Modes".
4974	*
4975	* Verify that we always intercept CR4.PAE and CR0.PG bits, so we don't need to
4976	* import CR4 and CR0 from the VMCS here as those bits are always up to date.
4977	*/
4978	Assert(hmR0VmxGetFixedCr4Mask(pVCpu) & X86_CR4_PAE);
4979	Assert(hmR0VmxGetFixedCr0Mask(pVCpu) & X86_CR0_PG);
4980	if ( (pCtx->cr4 & X86_CR4_PAE)
4981	&& (pCtx->cr0 & X86_CR0_PG))
4982	{
4983	/*
4984	* If nested paging is not used, verify that the guest paging mode matches the
4985	* shadow paging mode which is/will be placed in the VMCS (which is what will
4986	* actually be used while executing the guest and not the CR4 shadow value).
4987	*/
4988	AssertMsg( pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging
4989	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_PAE
4990	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_PAE_NX
4991	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_AMD64
4992	\|\| pVCpu->hm.s.enmShadowMode == PGMMODE_AMD64_NX,
4993	("enmShadowMode=%u\n", pVCpu->hm.s.enmShadowMode));
4994	if ((u64GuestEfer & MSR_K6_EFER_NXE) != (u64HostEfer & MSR_K6_EFER_NXE))
4995	{
4996	/* Verify that the host is NX capable. */
4997	Assert(pVCpu->CTX_SUFF(pVM)->cpum.ro.HostFeatures.fNoExecute);
4998	return true;
4999	}
5000	}
5001
5002	return false;
5003	#endif
5004	}
5005
5006
5007	/**
5008	* Exports the guest state with appropriate VM-entry and VM-exit controls in the
5009	* VMCS.
5010	*
5011	* This is typically required when the guest changes paging mode.
5012	*
5013	* @returns VBox status code.
5014	* @param pVCpu The cross context virtual CPU structure.
5015	* @param pVmxTransient The VMX-transient structure.
5016	*
5017	* @remarks Requires EFER.
5018	* @remarks No-long-jump zone!!!
5019	*/
5020	static int hmR0VmxExportGuestEntryExitCtls(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5021	{
5022	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_ENTRY_EXIT_CTLS)
5023	{
5024	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5025	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5026
5027	/*
5028	* VM-entry controls.
5029	*/
5030	{
5031	uint32_t fVal = g_HmMsrs.u.vmx.EntryCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
5032	uint32_t const fZap = g_HmMsrs.u.vmx.EntryCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
5033
5034	/*
5035	* Load the guest debug controls (DR7 and IA32_DEBUGCTL MSR) on VM-entry.
5036	* The first VT-x capable CPUs only supported the 1-setting of this bit.
5037	*
5038	* For nested-guests, this is a mandatory VM-entry control. It's also
5039	* required because we do not want to leak host bits to the nested-guest.
5040	*/
5041	fVal \|= VMX_ENTRY_CTLS_LOAD_DEBUG;
5042
5043	/*
5044	* Set if the guest is in long mode. This will set/clear the EFER.LMA bit on VM-entry.
5045	*
5046	* For nested-guests, the "IA-32e mode guest" control we initialize with what is
5047	* required to get the nested-guest working with hardware-assisted VMX execution.
5048	* It depends on the nested-guest's IA32_EFER.LMA bit. Remember, a nested hypervisor
5049	* can skip intercepting changes to the EFER MSR. This is why it needs to be done
5050	* here rather than while merging the guest VMCS controls.
5051	*/
5052	if (CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
5053	{
5054	Assert(pVCpu->cpum.GstCtx.msrEFER & MSR_K6_EFER_LME);
5055	fVal \|= VMX_ENTRY_CTLS_IA32E_MODE_GUEST;
5056	}
5057	else
5058	Assert(!(fVal & VMX_ENTRY_CTLS_IA32E_MODE_GUEST));
5059
5060	/*
5061	* If the CPU supports the newer VMCS controls for managing guest/host EFER, use it.
5062	*
5063	* For nested-guests, we use the "load IA32_EFER" if the hardware supports it,
5064	* regardless of whether the nested-guest VMCS specifies it because we are free to
5065	* load whatever MSRs we require and we do not need to modify the guest visible copy
5066	* of the VM-entry MSR load area.
5067	*/
5068	if ( g_fHmVmxSupportsVmcsEfer
5069	&& hmR0VmxShouldSwapEferMsr(pVCpu, pVmxTransient))
5070	fVal \|= VMX_ENTRY_CTLS_LOAD_EFER_MSR;
5071	else
5072	Assert(!(fVal & VMX_ENTRY_CTLS_LOAD_EFER_MSR));
5073
5074	/*
5075	* The following should -not- be set (since we're not in SMM mode):
5076	* - VMX_ENTRY_CTLS_ENTRY_TO_SMM
5077	* - VMX_ENTRY_CTLS_DEACTIVATE_DUAL_MON
5078	*/
5079
5080	/** @todo VMX_ENTRY_CTLS_LOAD_PERF_MSR,
5081	* VMX_ENTRY_CTLS_LOAD_PAT_MSR. */
5082
5083	if ((fVal & fZap) == fVal)
5084	{ /* likely */ }
5085	else
5086	{
5087	Log4Func(("Invalid VM-entry controls combo! Cpu=%#RX32 fVal=%#RX32 fZap=%#RX32\n",
5088	g_HmMsrs.u.vmx.EntryCtls.n.allowed0, fVal, fZap));
5089	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_ENTRY;
5090	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
5091	}
5092
5093	/* Commit it to the VMCS. */
5094	if (pVmcsInfo->u32EntryCtls != fVal)
5095	{
5096	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY, fVal);
5097	AssertRC(rc);
5098	pVmcsInfo->u32EntryCtls = fVal;
5099	}
5100	}
5101
5102	/*
5103	* VM-exit controls.
5104	*/
5105	{
5106	uint32_t fVal = g_HmMsrs.u.vmx.ExitCtls.n.allowed0; /* Bits set here must be set in the VMCS. */
5107	uint32_t const fZap = g_HmMsrs.u.vmx.ExitCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
5108
5109	/*
5110	* Save debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x CPUs only
5111	* supported the 1-setting of this bit.
5112	*
5113	* For nested-guests, we set the "save debug controls" as the converse
5114	* "load debug controls" is mandatory for nested-guests anyway.
5115	*/
5116	fVal \|= VMX_EXIT_CTLS_SAVE_DEBUG;
5117
5118	/*
5119	* Set the host long mode active (EFER.LMA) bit (which Intel calls
5120	* "Host address-space size") if necessary. On VM-exit, VT-x sets both the
5121	* host EFER.LMA and EFER.LME bit to this value. See assertion in
5122	* hmR0VmxExportHostMsrs().
5123	*
5124	* For nested-guests, we always set this bit as we do not support 32-bit
5125	* hosts.
5126	*/
5127	fVal \|= VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE;
5128
5129	/*
5130	* If the VMCS EFER MSR fields are supported by the hardware, we use it.
5131	*
5132	* For nested-guests, we should use the "save IA32_EFER" control if we also
5133	* used the "load IA32_EFER" control while exporting VM-entry controls.
5134	*/
5135	if ( g_fHmVmxSupportsVmcsEfer
5136	&& hmR0VmxShouldSwapEferMsr(pVCpu, pVmxTransient))
5137	{
5138	fVal \|= VMX_EXIT_CTLS_SAVE_EFER_MSR
5139	\| VMX_EXIT_CTLS_LOAD_EFER_MSR;
5140	}
5141
5142	/*
5143	* Enable saving of the VMX-preemption timer value on VM-exit.
5144	* For nested-guests, currently not exposed/used.
5145	*/
5146	/** @todo r=bird: Measure performance hit because of this vs. always rewriting
5147	* the timer value. */
5148	if (pVM->hmr0.s.vmx.fUsePreemptTimer)
5149	{
5150	Assert(g_HmMsrs.u.vmx.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER);
5151	fVal \|= VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER;
5152	}
5153
5154	/* Don't acknowledge external interrupts on VM-exit. We want to let the host do that. */
5155	Assert(!(fVal & VMX_EXIT_CTLS_ACK_EXT_INT));
5156
5157	/** @todo VMX_EXIT_CTLS_LOAD_PERF_MSR,
5158	* VMX_EXIT_CTLS_SAVE_PAT_MSR,
5159	* VMX_EXIT_CTLS_LOAD_PAT_MSR. */
5160
5161	if ((fVal & fZap) == fVal)
5162	{ /* likely */ }
5163	else
5164	{
5165	Log4Func(("Invalid VM-exit controls combo! cpu=%#RX32 fVal=%#RX32 fZap=%R#X32\n",
5166	g_HmMsrs.u.vmx.ExitCtls.n.allowed0, fVal, fZap));
5167	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_EXIT;
5168	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
5169	}
5170
5171	/* Commit it to the VMCS. */
5172	if (pVmcsInfo->u32ExitCtls != fVal)
5173	{
5174	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT, fVal);
5175	AssertRC(rc);
5176	pVmcsInfo->u32ExitCtls = fVal;
5177	}
5178	}
5179
5180	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
5181	}
5182	return VINF_SUCCESS;
5183	}
5184
5185
5186	/**
5187	* Sets the TPR threshold in the VMCS.
5188	*
5189	* @param pVmcsInfo The VMCS info. object.
5190	* @param u32TprThreshold The TPR threshold (task-priority class only).
5191	*/
5192	DECLINLINE(void) hmR0VmxApicSetTprThreshold(PVMXVMCSINFO pVmcsInfo, uint32_t u32TprThreshold)
5193	{
5194	Assert(!(u32TprThreshold & ~VMX_TPR_THRESHOLD_MASK)); /* Bits 31:4 MBZ. */
5195	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
5196	RT_NOREF(pVmcsInfo);
5197	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
5198	AssertRC(rc);
5199	}
5200
5201
5202	/**
5203	* Exports the guest APIC TPR state into the VMCS.
5204	*
5205	* @param pVCpu The cross context virtual CPU structure.
5206	* @param pVmxTransient The VMX-transient structure.
5207	*
5208	* @remarks No-long-jump zone!!!
5209	*/
5210	static void hmR0VmxExportGuestApicTpr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5211	{
5212	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_APIC_TPR)
5213	{
5214	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_APIC_TPR);
5215
5216	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5217	if (!pVmxTransient->fIsNestedGuest)
5218	{
5219	if ( PDMHasApic(pVCpu->CTX_SUFF(pVM))
5220	&& APICIsEnabled(pVCpu))
5221	{
5222	/*
5223	* Setup TPR shadowing.
5224	*/
5225	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
5226	{
5227	bool fPendingIntr = false;
5228	uint8_t u8Tpr = 0;
5229	uint8_t u8PendingIntr = 0;
5230	int rc = APICGetTpr(pVCpu, &u8Tpr, &fPendingIntr, &u8PendingIntr);
5231	AssertRC(rc);
5232
5233	/*
5234	* If there are interrupts pending but masked by the TPR, instruct VT-x to
5235	* cause a TPR-below-threshold VM-exit when the guest lowers its TPR below the
5236	* priority of the pending interrupt so we can deliver the interrupt. If there
5237	* are no interrupts pending, set threshold to 0 to not cause any
5238	* TPR-below-threshold VM-exits.
5239	*/
5240	uint32_t u32TprThreshold = 0;
5241	if (fPendingIntr)
5242	{
5243	/* Bits 3:0 of the TPR threshold field correspond to bits 7:4 of the TPR
5244	(which is the Task-Priority Class). */
5245	const uint8_t u8PendingPriority = u8PendingIntr >> 4;
5246	const uint8_t u8TprPriority = u8Tpr >> 4;
5247	if (u8PendingPriority <= u8TprPriority)
5248	u32TprThreshold = u8PendingPriority;
5249	}
5250
5251	hmR0VmxApicSetTprThreshold(pVmcsInfo, u32TprThreshold);
5252	}
5253	}
5254	}
5255	/* else: the TPR threshold has already been updated while merging the nested-guest VMCS. */
5256	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_APIC_TPR);
5257	}
5258	}
5259
5260
5261	/**
5262	* Gets the guest interruptibility-state and updates related force-flags.
5263	*
5264	* @returns Guest's interruptibility-state.
5265	* @param pVCpu The cross context virtual CPU structure.
5266	*
5267	* @remarks No-long-jump zone!!!
5268	*/
5269	static uint32_t hmR0VmxGetGuestIntrStateAndUpdateFFs(PVMCPUCC pVCpu)
5270	{
5271	/*
5272	* Check if we should inhibit interrupt delivery due to instructions like STI and MOV SS.
5273	*/
5274	uint32_t fIntrState = 0;
5275	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
5276	{
5277	/* If inhibition is active, RIP and RFLAGS should've been imported from the VMCS already. */
5278	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS);
5279
5280	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5281	if (pCtx->rip == EMGetInhibitInterruptsPC(pVCpu))
5282	{
5283	if (pCtx->eflags.Bits.u1IF)
5284	fIntrState = VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
5285	else
5286	fIntrState = VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS;
5287	}
5288	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
5289	{
5290	/*
5291	* We can clear the inhibit force flag as even if we go back to the recompiler
5292	* without executing guest code in VT-x, the flag's condition to be cleared is
5293	* met and thus the cleared state is correct.
5294	*/
5295	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
5296	}
5297	}
5298
5299	/*
5300	* Check if we should inhibit NMI delivery.
5301	*/
5302	if (CPUMIsGuestNmiBlocking(pVCpu))
5303	fIntrState \|= VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI;
5304
5305	/*
5306	* Validate.
5307	*/
5308	#ifdef VBOX_STRICT
5309	/* We don't support block-by-SMI yet.*/
5310	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI));
5311
5312	/* Block-by-STI must not be set when interrupts are disabled. */
5313	if (fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
5314	{
5315	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
5316	Assert(pVCpu->cpum.GstCtx.eflags.u & X86_EFL_IF);
5317	}
5318	#endif
5319
5320	return fIntrState;
5321	}
5322
5323
5324	/**
5325	* Exports the exception intercepts required for guest execution in the VMCS.
5326	*
5327	* @param pVCpu The cross context virtual CPU structure.
5328	* @param pVmxTransient The VMX-transient structure.
5329	*
5330	* @remarks No-long-jump zone!!!
5331	*/
5332	static void hmR0VmxExportGuestXcptIntercepts(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5333	{
5334	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_XCPT_INTERCEPTS)
5335	{
5336	/* When executing a nested-guest, we do not need to trap GIM hypercalls by intercepting #UD. */
5337	if ( !pVmxTransient->fIsNestedGuest
5338	&& pVCpu->hm.s.fGIMTrapXcptUD)
5339	hmR0VmxAddXcptIntercept(pVmxTransient, X86_XCPT_UD);
5340	else
5341	hmR0VmxRemoveXcptIntercept(pVCpu, pVmxTransient, X86_XCPT_UD);
5342
5343	/* Other exception intercepts are handled elsewhere, e.g. while exporting guest CR0. */
5344	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_XCPT_INTERCEPTS);
5345	}
5346	}
5347
5348
5349	/**
5350	* Exports the guest's RIP into the guest-state area in the VMCS.
5351	*
5352	* @param pVCpu The cross context virtual CPU structure.
5353	*
5354	* @remarks No-long-jump zone!!!
5355	*/
5356	static void hmR0VmxExportGuestRip(PVMCPUCC pVCpu)
5357	{
5358	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RIP)
5359	{
5360	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP);
5361
5362	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RIP, pVCpu->cpum.GstCtx.rip);
5363	AssertRC(rc);
5364
5365	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RIP);
5366	Log4Func(("rip=%#RX64\n", pVCpu->cpum.GstCtx.rip));
5367	}
5368	}
5369
5370
5371	/**
5372	* Exports the guest's RSP into the guest-state area in the VMCS.
5373	*
5374	* @param pVCpu The cross context virtual CPU structure.
5375	*
5376	* @remarks No-long-jump zone!!!
5377	*/
5378	static void hmR0VmxExportGuestRsp(PVMCPUCC pVCpu)
5379	{
5380	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RSP)
5381	{
5382	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RSP);
5383
5384	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RSP, pVCpu->cpum.GstCtx.rsp);
5385	AssertRC(rc);
5386
5387	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RSP);
5388	Log4Func(("rsp=%#RX64\n", pVCpu->cpum.GstCtx.rsp));
5389	}
5390	}
5391
5392
5393	/**
5394	* Exports the guest's RFLAGS into the guest-state area in the VMCS.
5395	*
5396	* @param pVCpu The cross context virtual CPU structure.
5397	* @param pVmxTransient The VMX-transient structure.
5398	*
5399	* @remarks No-long-jump zone!!!
5400	*/
5401	static void hmR0VmxExportGuestRflags(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5402	{
5403	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_RFLAGS)
5404	{
5405	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
5406
5407	/* Intel spec. 2.3.1 "System Flags and Fields in IA-32e Mode" claims the upper 32-bits of RFLAGS are reserved (MBZ).
5408	Let us assert it as such and use 32-bit VMWRITE. */
5409	Assert(!RT_HI_U32(pVCpu->cpum.GstCtx.rflags.u64));
5410	X86EFLAGS fEFlags = pVCpu->cpum.GstCtx.eflags;
5411	Assert(fEFlags.u32 & X86_EFL_RA1_MASK);
5412	Assert(!(fEFlags.u32 & ~(X86_EFL_1 \| X86_EFL_LIVE_MASK)));
5413
5414	/*
5415	* If we're emulating real-mode using Virtual 8086 mode, save the real-mode eflags so
5416	* we can restore them on VM-exit. Modify the real-mode guest's eflags so that VT-x
5417	* can run the real-mode guest code under Virtual 8086 mode.
5418	*/
5419	PVMXVMCSINFOSHARED pVmcsInfo = pVmxTransient->pVmcsInfo->pShared;
5420	if (pVmcsInfo->RealMode.fRealOnV86Active)
5421	{
5422	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
5423	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
5424	Assert(!pVmxTransient->fIsNestedGuest);
5425	pVmcsInfo->RealMode.Eflags.u32 = fEFlags.u32; /* Save the original eflags of the real-mode guest. */
5426	fEFlags.Bits.u1VM = 1; /* Set the Virtual 8086 mode bit. */
5427	fEFlags.Bits.u2IOPL = 0; /* Change IOPL to 0, otherwise certain instructions won't fault. */
5428	}
5429
5430	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_RFLAGS, fEFlags.u32);
5431	AssertRC(rc);
5432
5433	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_RFLAGS);
5434	Log4Func(("eflags=%#RX32\n", fEFlags.u32));
5435	}
5436	}
5437
5438
5439	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5440	/**
5441	* Copies the nested-guest VMCS to the shadow VMCS.
5442	*
5443	* @returns VBox status code.
5444	* @param pVCpu The cross context virtual CPU structure.
5445	* @param pVmcsInfo The VMCS info. object.
5446	*
5447	* @remarks No-long-jump zone!!!
5448	*/
5449	static int hmR0VmxCopyNstGstToShadowVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
5450	{
5451	PVMCC const pVM = pVCpu->CTX_SUFF(pVM);
5452	PCVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
5453
5454	/*
5455	* Disable interrupts so we don't get preempted while the shadow VMCS is the
5456	* current VMCS, as we may try saving guest lazy MSRs.
5457	*
5458	* Strictly speaking the lazy MSRs are not in the VMCS, but I'd rather not risk
5459	* calling the import VMCS code which is currently performing the guest MSR reads
5460	* (on 64-bit hosts) and accessing the auto-load/store MSR area on 32-bit hosts
5461	* and the rest of the VMX leave session machinery.
5462	*/
5463	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
5464
5465	int rc = hmR0VmxLoadShadowVmcs(pVmcsInfo);
5466	if (RT_SUCCESS(rc))
5467	{
5468	/*
5469	* Copy all guest read/write VMCS fields.
5470	*
5471	* We don't check for VMWRITE failures here for performance reasons and
5472	* because they are not expected to fail, barring irrecoverable conditions
5473	* like hardware errors.
5474	*/
5475	uint32_t const cShadowVmcsFields = pVM->hmr0.s.vmx.cShadowVmcsFields;
5476	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
5477	{
5478	uint64_t u64Val;
5479	uint32_t const uVmcsField = pVM->hmr0.s.vmx.paShadowVmcsFields[i];
5480	IEMReadVmxVmcsField(pVmcsNstGst, uVmcsField, &u64Val);
5481	VMXWriteVmcs64(uVmcsField, u64Val);
5482	}
5483
5484	/*
5485	* If the host CPU supports writing all VMCS fields, copy the guest read-only
5486	* VMCS fields, so the guest can VMREAD them without causing a VM-exit.
5487	*/
5488	if (g_HmMsrs.u.vmx.u64Misc & VMX_MISC_VMWRITE_ALL)
5489	{
5490	uint32_t const cShadowVmcsRoFields = pVM->hmr0.s.vmx.cShadowVmcsRoFields;
5491	for (uint32_t i = 0; i < cShadowVmcsRoFields; i++)
5492	{
5493	uint64_t u64Val;
5494	uint32_t const uVmcsField = pVM->hmr0.s.vmx.paShadowVmcsRoFields[i];
5495	IEMReadVmxVmcsField(pVmcsNstGst, uVmcsField, &u64Val);
5496	VMXWriteVmcs64(uVmcsField, u64Val);
5497	}
5498	}
5499
5500	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
5501	rc \|= hmR0VmxLoadVmcs(pVmcsInfo);
5502	}
5503
5504	ASMSetFlags(fEFlags);
5505	return rc;
5506	}
5507
5508
5509	/**
5510	* Copies the shadow VMCS to the nested-guest VMCS.
5511	*
5512	* @returns VBox status code.
5513	* @param pVCpu The cross context virtual CPU structure.
5514	* @param pVmcsInfo The VMCS info. object.
5515	*
5516	* @remarks Called with interrupts disabled.
5517	*/
5518	static int hmR0VmxCopyShadowToNstGstVmcs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
5519	{
5520	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
5521	PVMCC const pVM = pVCpu->CTX_SUFF(pVM);
5522	PVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
5523
5524	int rc = hmR0VmxLoadShadowVmcs(pVmcsInfo);
5525	if (RT_SUCCESS(rc))
5526	{
5527	/*
5528	* Copy guest read/write fields from the shadow VMCS.
5529	* Guest read-only fields cannot be modified, so no need to copy them.
5530	*
5531	* We don't check for VMREAD failures here for performance reasons and
5532	* because they are not expected to fail, barring irrecoverable conditions
5533	* like hardware errors.
5534	*/
5535	uint32_t const cShadowVmcsFields = pVM->hmr0.s.vmx.cShadowVmcsFields;
5536	for (uint32_t i = 0; i < cShadowVmcsFields; i++)
5537	{
5538	uint64_t u64Val;
5539	uint32_t const uVmcsField = pVM->hmr0.s.vmx.paShadowVmcsFields[i];
5540	VMXReadVmcs64(uVmcsField, &u64Val);
5541	IEMWriteVmxVmcsField(pVmcsNstGst, uVmcsField, u64Val);
5542	}
5543
5544	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
5545	rc \|= hmR0VmxLoadVmcs(pVmcsInfo);
5546	}
5547	return rc;
5548	}
5549
5550
5551	/**
5552	* Enables VMCS shadowing for the given VMCS info. object.
5553	*
5554	* @param pVmcsInfo The VMCS info. object.
5555	*
5556	* @remarks No-long-jump zone!!!
5557	*/
5558	static void hmR0VmxEnableVmcsShadowing(PVMXVMCSINFO pVmcsInfo)
5559	{
5560	uint32_t uProcCtls2 = pVmcsInfo->u32ProcCtls2;
5561	if (!(uProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING))
5562	{
5563	Assert(pVmcsInfo->HCPhysShadowVmcs != 0 && pVmcsInfo->HCPhysShadowVmcs != NIL_RTHCPHYS);
5564	uProcCtls2 \|= VMX_PROC_CTLS2_VMCS_SHADOWING;
5565	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, uProcCtls2); AssertRC(rc);
5566	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, pVmcsInfo->HCPhysShadowVmcs); AssertRC(rc);
5567	pVmcsInfo->u32ProcCtls2 = uProcCtls2;
5568	pVmcsInfo->u64VmcsLinkPtr = pVmcsInfo->HCPhysShadowVmcs;
5569	Log4Func(("Enabled\n"));
5570	}
5571	}
5572
5573
5574	/**
5575	* Disables VMCS shadowing for the given VMCS info. object.
5576	*
5577	* @param pVmcsInfo The VMCS info. object.
5578	*
5579	* @remarks No-long-jump zone!!!
5580	*/
5581	static void hmR0VmxDisableVmcsShadowing(PVMXVMCSINFO pVmcsInfo)
5582	{
5583	/*
5584	* We want all VMREAD and VMWRITE instructions to cause VM-exits, so we clear the
5585	* VMCS shadowing control. However, VM-entry requires the shadow VMCS indicator bit
5586	* to match the VMCS shadowing control if the VMCS link pointer is not NIL_RTHCPHYS.
5587	* Hence, we must also reset the VMCS link pointer to ensure VM-entry does not fail.
5588	*
5589	* See Intel spec. 26.2.1.1 "VM-Execution Control Fields".
5590	* See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
5591	*/
5592	uint32_t uProcCtls2 = pVmcsInfo->u32ProcCtls2;
5593	if (uProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
5594	{
5595	uProcCtls2 &= ~VMX_PROC_CTLS2_VMCS_SHADOWING;
5596	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, uProcCtls2); AssertRC(rc);
5597	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, NIL_RTHCPHYS); AssertRC(rc);
5598	pVmcsInfo->u32ProcCtls2 = uProcCtls2;
5599	pVmcsInfo->u64VmcsLinkPtr = NIL_RTHCPHYS;
5600	Log4Func(("Disabled\n"));
5601	}
5602	}
5603	#endif
5604
5605
5606	/**
5607	* Exports the guest hardware-virtualization state.
5608	*
5609	* @returns VBox status code.
5610	* @param pVCpu The cross context virtual CPU structure.
5611	* @param pVmxTransient The VMX-transient structure.
5612	*
5613	* @remarks No-long-jump zone!!!
5614	*/
5615	static int hmR0VmxExportGuestHwvirtState(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5616	{
5617	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_HWVIRT)
5618	{
5619	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5620	/*
5621	* Check if the VMX feature is exposed to the guest and if the host CPU supports
5622	* VMCS shadowing.
5623	*/
5624	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUseVmcsShadowing)
5625	{
5626	/*
5627	* If the nested hypervisor has loaded a current VMCS and is in VMX root mode,
5628	* copy the nested hypervisor's current VMCS into the shadow VMCS and enable
5629	* VMCS shadowing to skip intercepting some or all VMREAD/VMWRITE VM-exits.
5630	*
5631	* We check for VMX root mode here in case the guest executes VMXOFF without
5632	* clearing the current VMCS pointer and our VMXOFF instruction emulation does
5633	* not clear the current VMCS pointer.
5634	*/
5635	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5636	if ( CPUMIsGuestInVmxRootMode(&pVCpu->cpum.GstCtx)
5637	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx)
5638	&& CPUMIsGuestVmxCurrentVmcsValid(&pVCpu->cpum.GstCtx))
5639	{
5640	/* Paranoia. */
5641	Assert(!pVmxTransient->fIsNestedGuest);
5642
5643	/*
5644	* For performance reasons, also check if the nested hypervisor's current VMCS
5645	* was newly loaded or modified before copying it to the shadow VMCS.
5646	*/
5647	if (!pVCpu->hm.s.vmx.fCopiedNstGstToShadowVmcs)
5648	{
5649	int rc = hmR0VmxCopyNstGstToShadowVmcs(pVCpu, pVmcsInfo);
5650	AssertRCReturn(rc, rc);
5651	pVCpu->hm.s.vmx.fCopiedNstGstToShadowVmcs = true;
5652	}
5653	hmR0VmxEnableVmcsShadowing(pVmcsInfo);
5654	}
5655	else
5656	hmR0VmxDisableVmcsShadowing(pVmcsInfo);
5657	}
5658	#else
5659	NOREF(pVmxTransient);
5660	#endif
5661	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_HWVIRT);
5662	}
5663	return VINF_SUCCESS;
5664	}
5665
5666
5667	/**
5668	* Exports the guest CR0 control register into the guest-state area in the VMCS.
5669	*
5670	* The guest FPU state is always pre-loaded hence we don't need to bother about
5671	* sharing FPU related CR0 bits between the guest and host.
5672	*
5673	* @returns VBox status code.
5674	* @param pVCpu The cross context virtual CPU structure.
5675	* @param pVmxTransient The VMX-transient structure.
5676	*
5677	* @remarks No-long-jump zone!!!
5678	*/
5679	static int hmR0VmxExportGuestCR0(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5680	{
5681	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR0)
5682	{
5683	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5684	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5685
5686	uint64_t fSetCr0 = g_HmMsrs.u.vmx.u64Cr0Fixed0;
5687	uint64_t const fZapCr0 = g_HmMsrs.u.vmx.u64Cr0Fixed1;
5688	if (pVM->hmr0.s.vmx.fUnrestrictedGuest)
5689	fSetCr0 &= ~(uint64_t)(X86_CR0_PE \| X86_CR0_PG);
5690	else
5691	Assert((fSetCr0 & (X86_CR0_PE \| X86_CR0_PG)) == (X86_CR0_PE \| X86_CR0_PG));
5692
5693	if (!pVmxTransient->fIsNestedGuest)
5694	{
5695	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
5696	uint64_t u64GuestCr0 = pVCpu->cpum.GstCtx.cr0;
5697	uint64_t const u64ShadowCr0 = u64GuestCr0;
5698	Assert(!RT_HI_U32(u64GuestCr0));
5699
5700	/*
5701	* Setup VT-x's view of the guest CR0.
5702	*/
5703	uint32_t uProcCtls = pVmcsInfo->u32ProcCtls;
5704	if (pVM->hmr0.s.fNestedPaging)
5705	{
5706	if (CPUMIsGuestPagingEnabled(pVCpu))
5707	{
5708	/* The guest has paging enabled, let it access CR3 without causing a VM-exit if supported. */
5709	uProcCtls &= ~( VMX_PROC_CTLS_CR3_LOAD_EXIT
5710	\| VMX_PROC_CTLS_CR3_STORE_EXIT);
5711	}
5712	else
5713	{
5714	/* The guest doesn't have paging enabled, make CR3 access cause a VM-exit to update our shadow. */
5715	uProcCtls \|= VMX_PROC_CTLS_CR3_LOAD_EXIT
5716	\| VMX_PROC_CTLS_CR3_STORE_EXIT;
5717	}
5718
5719	/* If we have unrestricted guest execution, we never have to intercept CR3 reads. */
5720	if (pVM->hmr0.s.vmx.fUnrestrictedGuest)
5721	uProcCtls &= ~VMX_PROC_CTLS_CR3_STORE_EXIT;
5722	}
5723	else
5724	{
5725	/* Guest CPL 0 writes to its read-only pages should cause a #PF VM-exit. */
5726	u64GuestCr0 \|= X86_CR0_WP;
5727	}
5728
5729	/*
5730	* Guest FPU bits.
5731	*
5732	* Since we pre-load the guest FPU always before VM-entry there is no need to track lazy state
5733	* using CR0.TS.
5734	*
5735	* Intel spec. 23.8 "Restrictions on VMX operation" mentions that CR0.NE bit must always be
5736	* set on the first CPUs to support VT-x and no mention of with regards to UX in VM-entry checks.
5737	*/
5738	u64GuestCr0 \|= X86_CR0_NE;
5739
5740	/* If CR0.NE isn't set, we need to intercept #MF exceptions and report them to the guest differently. */
5741	bool const fInterceptMF = !(u64ShadowCr0 & X86_CR0_NE);
5742
5743	/*
5744	* Update exception intercepts.
5745	*/
5746	uint32_t uXcptBitmap = pVmcsInfo->u32XcptBitmap;
5747	if (pVmcsInfo->pShared->RealMode.fRealOnV86Active)
5748	{
5749	Assert(PDMVmmDevHeapIsEnabled(pVM));
5750	Assert(pVM->hm.s.vmx.pRealModeTSS);
5751	uXcptBitmap \|= HMVMX_REAL_MODE_XCPT_MASK;
5752	}
5753	else
5754	{
5755	/* For now, cleared here as mode-switches can happen outside HM/VT-x. See @bugref{7626#c11}. */
5756	uXcptBitmap &= ~HMVMX_REAL_MODE_XCPT_MASK;
5757	if (fInterceptMF)
5758	uXcptBitmap \|= RT_BIT(X86_XCPT_MF);
5759	}
5760
5761	/* Additional intercepts for debugging, define these yourself explicitly. */
5762	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
5763	uXcptBitmap \|= 0
5764	\| RT_BIT(X86_XCPT_BP)
5765	\| RT_BIT(X86_XCPT_DE)
5766	\| RT_BIT(X86_XCPT_NM)
5767	\| RT_BIT(X86_XCPT_TS)
5768	\| RT_BIT(X86_XCPT_UD)
5769	\| RT_BIT(X86_XCPT_NP)
5770	\| RT_BIT(X86_XCPT_SS)
5771	\| RT_BIT(X86_XCPT_GP)
5772	\| RT_BIT(X86_XCPT_PF)
5773	\| RT_BIT(X86_XCPT_MF)
5774	;
5775	#elif defined(HMVMX_ALWAYS_TRAP_PF)
5776	uXcptBitmap \|= RT_BIT(X86_XCPT_PF);
5777	#endif
5778	if (pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv)
5779	uXcptBitmap \|= RT_BIT(X86_XCPT_GP);
5780	Assert(pVM->hmr0.s.fNestedPaging \|\| (uXcptBitmap & RT_BIT(X86_XCPT_PF)));
5781
5782	/* Apply the hardware specified CR0 fixed bits and enable caching. */
5783	u64GuestCr0 \|= fSetCr0;
5784	u64GuestCr0 &= fZapCr0;
5785	u64GuestCr0 &= ~(uint64_t)(X86_CR0_CD \| X86_CR0_NW);
5786
5787	/* Commit the CR0 and related fields to the guest VMCS. */
5788	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR0, u64GuestCr0); AssertRC(rc);
5789	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, u64ShadowCr0); AssertRC(rc);
5790	if (uProcCtls != pVmcsInfo->u32ProcCtls)
5791	{
5792	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, uProcCtls);
5793	AssertRC(rc);
5794	}
5795	if (uXcptBitmap != pVmcsInfo->u32XcptBitmap)
5796	{
5797	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, uXcptBitmap);
5798	AssertRC(rc);
5799	}
5800
5801	/* Update our caches. */
5802	pVmcsInfo->u32ProcCtls = uProcCtls;
5803	pVmcsInfo->u32XcptBitmap = uXcptBitmap;
5804
5805	Log4Func(("cr0=%#RX64 shadow=%#RX64 set=%#RX64 zap=%#RX64\n", u64GuestCr0, u64ShadowCr0, fSetCr0, fZapCr0));
5806	}
5807	else
5808	{
5809	/*
5810	* With nested-guests, we may have extended the guest/host mask here since we
5811	* merged in the outer guest's mask. Thus, the merged mask can include more bits
5812	* (to read from the nested-guest CR0 read-shadow) than the nested hypervisor
5813	* originally supplied. We must copy those bits from the nested-guest CR0 into
5814	* the nested-guest CR0 read-shadow.
5815	*/
5816	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
5817	uint64_t u64GuestCr0 = pVCpu->cpum.GstCtx.cr0;
5818	uint64_t const u64ShadowCr0 = CPUMGetGuestVmxMaskedCr0(&pVCpu->cpum.GstCtx, pVmcsInfo->u64Cr0Mask);
5819	Assert(!RT_HI_U32(u64GuestCr0));
5820	Assert(u64GuestCr0 & X86_CR0_NE);
5821
5822	/* Apply the hardware specified CR0 fixed bits and enable caching. */
5823	u64GuestCr0 \|= fSetCr0;
5824	u64GuestCr0 &= fZapCr0;
5825	u64GuestCr0 &= ~(uint64_t)(X86_CR0_CD \| X86_CR0_NW);
5826
5827	/* Commit the CR0 and CR0 read-shadow to the nested-guest VMCS. */
5828	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR0, u64GuestCr0); AssertRC(rc);
5829	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, u64ShadowCr0); AssertRC(rc);
5830
5831	Log4Func(("cr0=%#RX64 shadow=%#RX64 (set=%#RX64 zap=%#RX64)\n", u64GuestCr0, u64ShadowCr0, fSetCr0, fZapCr0));
5832	}
5833
5834	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR0);
5835	}
5836
5837	return VINF_SUCCESS;
5838	}
5839
5840
5841	/**
5842	* Exports the guest control registers (CR3, CR4) into the guest-state area
5843	* in the VMCS.
5844	*
5845	* @returns VBox strict status code.
5846	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
5847	* without unrestricted guest access and the VMMDev is not presently
5848	* mapped (e.g. EFI32).
5849	*
5850	* @param pVCpu The cross context virtual CPU structure.
5851	* @param pVmxTransient The VMX-transient structure.
5852	*
5853	* @remarks No-long-jump zone!!!
5854	*/
5855	static VBOXSTRICTRC hmR0VmxExportGuestCR3AndCR4(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
5856	{
5857	int rc = VINF_SUCCESS;
5858	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5859
5860	/*
5861	* Guest CR2.
5862	* It's always loaded in the assembler code. Nothing to do here.
5863	*/
5864
5865	/*
5866	* Guest CR3.
5867	*/
5868	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR3)
5869	{
5870	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR3);
5871
5872	if (pVM->hmr0.s.fNestedPaging)
5873	{
5874	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5875	pVmcsInfo->HCPhysEPTP = PGMGetHyperCR3(pVCpu);
5876
5877	/* Validate. See Intel spec. 28.2.2 "EPT Translation Mechanism" and 24.6.11 "Extended-Page-Table Pointer (EPTP)" */
5878	Assert(pVmcsInfo->HCPhysEPTP != NIL_RTHCPHYS);
5879	Assert(!(pVmcsInfo->HCPhysEPTP & UINT64_C(0xfff0000000000000)));
5880	Assert(!(pVmcsInfo->HCPhysEPTP & 0xfff));
5881
5882	/* VMX_EPT_MEMTYPE_WB support is already checked in hmR0VmxSetupTaggedTlb(). */
5883	pVmcsInfo->HCPhysEPTP \|= RT_BF_MAKE(VMX_BF_EPTP_MEMTYPE, VMX_EPTP_MEMTYPE_WB)
5884	\| RT_BF_MAKE(VMX_BF_EPTP_PAGE_WALK_LENGTH, VMX_EPTP_PAGE_WALK_LENGTH_4);
5885
5886	/* Validate. See Intel spec. 26.2.1 "Checks on VMX Controls" */
5887	AssertMsg( ((pVmcsInfo->HCPhysEPTP >> 3) & 0x07) == 3 /* Bits 3:5 (EPT page walk length - 1) must be 3. */
5888	&& ((pVmcsInfo->HCPhysEPTP >> 7) & 0x1f) == 0, /* Bits 7:11 MBZ. */
5889	("EPTP %#RX64\n", pVmcsInfo->HCPhysEPTP));
5890	AssertMsg( !((pVmcsInfo->HCPhysEPTP >> 6) & 0x01) /* Bit 6 (EPT accessed & dirty bit). */
5891	\|\| (g_HmMsrs.u.vmx.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_ACCESS_DIRTY),
5892	("EPTP accessed/dirty bit not supported by CPU but set %#RX64\n", pVmcsInfo->HCPhysEPTP));
5893
5894	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, pVmcsInfo->HCPhysEPTP);
5895	AssertRC(rc);
5896
5897	uint64_t u64GuestCr3;
5898	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5899	if ( pVM->hmr0.s.vmx.fUnrestrictedGuest
5900	\|\| CPUMIsGuestPagingEnabledEx(pCtx))
5901	{
5902	/* If the guest is in PAE mode, pass the PDPEs to VT-x using the VMCS fields. */
5903	if (CPUMIsGuestInPAEModeEx(pCtx))
5904	{
5905	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, pCtx->aPaePdpes[0].u); AssertRC(rc);
5906	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, pCtx->aPaePdpes[1].u); AssertRC(rc);
5907	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, pCtx->aPaePdpes[2].u); AssertRC(rc);
5908	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, pCtx->aPaePdpes[3].u); AssertRC(rc);
5909	}
5910
5911	/*
5912	* The guest's view of its CR3 is unblemished with nested paging when the
5913	* guest is using paging or we have unrestricted guest execution to handle
5914	* the guest when it's not using paging.
5915	*/
5916	u64GuestCr3 = pCtx->cr3;
5917	}
5918	else
5919	{
5920	/*
5921	* The guest is not using paging, but the CPU (VT-x) has to. While the guest
5922	* thinks it accesses physical memory directly, we use our identity-mapped
5923	* page table to map guest-linear to guest-physical addresses. EPT takes care
5924	* of translating it to host-physical addresses.
5925	*/
5926	RTGCPHYS GCPhys;
5927	Assert(pVM->hm.s.vmx.pNonPagingModeEPTPageTable);
5928
5929	/* We obtain it here every time as the guest could have relocated this PCI region. */
5930	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys);
5931	if (RT_SUCCESS(rc))
5932	{ /* likely */ }
5933	else if (rc == VERR_PDM_DEV_HEAP_R3_TO_GCPHYS)
5934	{
5935	Log4Func(("VERR_PDM_DEV_HEAP_R3_TO_GCPHYS -> VINF_EM_RESCHEDULE_REM\n"));
5936	return VINF_EM_RESCHEDULE_REM; /* We cannot execute now, switch to REM/IEM till the guest maps in VMMDev. */
5937	}
5938	else
5939	AssertMsgFailedReturn(("%Rrc\n", rc), rc);
5940
5941	u64GuestCr3 = GCPhys;
5942	}
5943
5944	Log4Func(("guest_cr3=%#RX64 (GstN)\n", u64GuestCr3));
5945	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR3, u64GuestCr3);
5946	AssertRC(rc);
5947	}
5948	else
5949	{
5950	Assert(!pVmxTransient->fIsNestedGuest);
5951	/* Non-nested paging case, just use the hypervisor's CR3. */
5952	RTHCPHYS const HCPhysGuestCr3 = PGMGetHyperCR3(pVCpu);
5953
5954	Log4Func(("guest_cr3=%#RX64 (HstN)\n", HCPhysGuestCr3));
5955	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR3, HCPhysGuestCr3);
5956	AssertRC(rc);
5957	}
5958
5959	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR3);
5960	}
5961
5962	/*
5963	* Guest CR4.
5964	* ASSUMES this is done everytime we get in from ring-3! (XCR0)
5965	*/
5966	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CR4)
5967	{
5968	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
5969	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
5970
5971	uint64_t const fSetCr4 = g_HmMsrs.u.vmx.u64Cr4Fixed0;
5972	uint64_t const fZapCr4 = g_HmMsrs.u.vmx.u64Cr4Fixed1;
5973
5974	/*
5975	* With nested-guests, we may have extended the guest/host mask here (since we
5976	* merged in the outer guest's mask, see hmR0VmxMergeVmcsNested). This means, the
5977	* mask can include more bits (to read from the nested-guest CR4 read-shadow) than
5978	* the nested hypervisor originally supplied. Thus, we should, in essence, copy
5979	* those bits from the nested-guest CR4 into the nested-guest CR4 read-shadow.
5980	*/
5981	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
5982	uint64_t u64GuestCr4 = pCtx->cr4;
5983	uint64_t const u64ShadowCr4 = !pVmxTransient->fIsNestedGuest
5984	? pCtx->cr4
5985	: CPUMGetGuestVmxMaskedCr4(pCtx, pVmcsInfo->u64Cr4Mask);
5986	Assert(!RT_HI_U32(u64GuestCr4));
5987
5988	/*
5989	* Setup VT-x's view of the guest CR4.
5990	*
5991	* If we're emulating real-mode using virtual-8086 mode, we want to redirect software
5992	* interrupts to the 8086 program interrupt handler. Clear the VME bit (the interrupt
5993	* redirection bitmap is already all 0, see hmR3InitFinalizeR0())
5994	*
5995	* See Intel spec. 20.2 "Software Interrupt Handling Methods While in Virtual-8086 Mode".
5996	*/
5997	if (pVmcsInfo->pShared->RealMode.fRealOnV86Active)
5998	{
5999	Assert(pVM->hm.s.vmx.pRealModeTSS);
6000	Assert(PDMVmmDevHeapIsEnabled(pVM));
6001	u64GuestCr4 &= ~(uint64_t)X86_CR4_VME;
6002	}
6003
6004	if (pVM->hmr0.s.fNestedPaging)
6005	{
6006	if ( !CPUMIsGuestPagingEnabledEx(pCtx)
6007	&& !pVM->hmr0.s.vmx.fUnrestrictedGuest)
6008	{
6009	/* We use 4 MB pages in our identity mapping page table when the guest doesn't have paging. */
6010	u64GuestCr4 \|= X86_CR4_PSE;
6011	/* Our identity mapping is a 32-bit page directory. */
6012	u64GuestCr4 &= ~(uint64_t)X86_CR4_PAE;
6013	}
6014	/* else use guest CR4.*/
6015	}
6016	else
6017	{
6018	Assert(!pVmxTransient->fIsNestedGuest);
6019
6020	/*
6021	* The shadow paging modes and guest paging modes are different, the shadow is in accordance with the host
6022	* paging mode and thus we need to adjust VT-x's view of CR4 depending on our shadow page tables.
6023	*/
6024	switch (pVCpu->hm.s.enmShadowMode)
6025	{
6026	case PGMMODE_REAL: /* Real-mode. */
6027	case PGMMODE_PROTECTED: /* Protected mode without paging. */
6028	case PGMMODE_32_BIT: /* 32-bit paging. */
6029	{
6030	u64GuestCr4 &= ~(uint64_t)X86_CR4_PAE;
6031	break;
6032	}
6033
6034	case PGMMODE_PAE: /* PAE paging. */
6035	case PGMMODE_PAE_NX: /* PAE paging with NX. */
6036	{
6037	u64GuestCr4 \|= X86_CR4_PAE;
6038	break;
6039	}
6040
6041	case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
6042	case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
6043	{
6044	#ifdef VBOX_WITH_64_BITS_GUESTS
6045	/* For our assumption in hmR0VmxShouldSwapEferMsr. */
6046	Assert(u64GuestCr4 & X86_CR4_PAE);
6047	break;
6048	#endif
6049	}
6050	default:
6051	AssertFailed();
6052	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
6053	}
6054	}
6055
6056	/* Apply the hardware specified CR4 fixed bits (mainly CR4.VMXE). */
6057	u64GuestCr4 \|= fSetCr4;
6058	u64GuestCr4 &= fZapCr4;
6059
6060	/* Commit the CR4 and CR4 read-shadow to the guest VMCS. */
6061	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_CR4, u64GuestCr4); AssertRC(rc);
6062	rc = VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, u64ShadowCr4); AssertRC(rc);
6063
6064	/* Whether to save/load/restore XCR0 during world switch depends on CR4.OSXSAVE and host+guest XCR0. */
6065	bool const fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
6066	if (fLoadSaveGuestXcr0 != pVCpu->hmr0.s.fLoadSaveGuestXcr0)
6067	{
6068	pVCpu->hmr0.s.fLoadSaveGuestXcr0 = fLoadSaveGuestXcr0;
6069	hmR0VmxUpdateStartVmFunction(pVCpu);
6070	}
6071
6072	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CR4);
6073
6074	Log4Func(("cr4=%#RX64 shadow=%#RX64 (set=%#RX64 zap=%#RX64)\n", u64GuestCr4, u64ShadowCr4, fSetCr4, fZapCr4));
6075	}
6076	return rc;
6077	}
6078
6079
6080	/**
6081	* Exports the guest debug registers into the guest-state area in the VMCS.
6082	* The guest debug bits are partially shared with the host (e.g. DR6, DR0-3).
6083	*
6084	* This also sets up whether \#DB and MOV DRx accesses cause VM-exits.
6085	*
6086	* @returns VBox status code.
6087	* @param pVCpu The cross context virtual CPU structure.
6088	* @param pVmxTransient The VMX-transient structure.
6089	*
6090	* @remarks No-long-jump zone!!!
6091	*/
6092	static int hmR0VmxExportSharedDebugState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
6093	{
6094	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
6095
6096	/** @todo NSTVMX: Figure out what we want to do with nested-guest instruction
6097	* stepping. */
6098	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6099	if (pVmxTransient->fIsNestedGuest)
6100	{
6101	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, CPUMGetGuestDR7(pVCpu));
6102	AssertRC(rc);
6103
6104	/*
6105	* We don't want to always intercept MOV DRx for nested-guests as it causes
6106	* problems when the nested hypervisor isn't intercepting them, see @bugref{10080}.
6107	* Instead, they are strictly only requested when the nested hypervisor intercepts
6108	* them -- handled while merging VMCS controls.
6109	*
6110	* If neither the outer nor the nested-hypervisor is intercepting MOV DRx,
6111	* then the nested-guest debug state should be actively loaded on the host so that
6112	* nested-guest reads its own debug registers without causing VM-exits.
6113	*/
6114	if ( !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MOV_DR_EXIT)
6115	&& !CPUMIsGuestDebugStateActive(pVCpu))
6116	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
6117	return VINF_SUCCESS;
6118	}
6119
6120	#ifdef VBOX_STRICT
6121	/* Validate. Intel spec. 26.3.1.1 "Checks on Guest Controls Registers, Debug Registers, MSRs" */
6122	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
6123	{
6124	/* Validate. Intel spec. 17.2 "Debug Registers", recompiler paranoia checks. */
6125	Assert((pVCpu->cpum.GstCtx.dr[7] & (X86_DR7_MBZ_MASK \| X86_DR7_RAZ_MASK)) == 0);
6126	Assert((pVCpu->cpum.GstCtx.dr[7] & X86_DR7_RA1_MASK) == X86_DR7_RA1_MASK);
6127	}
6128	#endif
6129
6130	bool fSteppingDB = false;
6131	bool fInterceptMovDRx = false;
6132	uint32_t uProcCtls = pVmcsInfo->u32ProcCtls;
6133	if (pVCpu->hm.s.fSingleInstruction)
6134	{
6135	/* If the CPU supports the monitor trap flag, use it for single stepping in DBGF and avoid intercepting #DB. */
6136	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_MONITOR_TRAP_FLAG)
6137	{
6138	uProcCtls \|= VMX_PROC_CTLS_MONITOR_TRAP_FLAG;
6139	Assert(fSteppingDB == false);
6140	}
6141	else
6142	{
6143	pVCpu->cpum.GstCtx.eflags.u32 \|= X86_EFL_TF;
6144	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_GUEST_RFLAGS;
6145	pVCpu->hmr0.s.fClearTrapFlag = true;
6146	fSteppingDB = true;
6147	}
6148	}
6149
6150	uint64_t u64GuestDr7;
6151	if ( fSteppingDB
6152	\|\| (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK))
6153	{
6154	/*
6155	* Use the combined guest and host DRx values found in the hypervisor register set
6156	* because the hypervisor debugger has breakpoints active or someone is single stepping
6157	* on the host side without a monitor trap flag.
6158	*
6159	* Note! DBGF expects a clean DR6 state before executing guest code.
6160	*/
6161	if (!CPUMIsHyperDebugStateActive(pVCpu))
6162	{
6163	CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
6164	Assert(CPUMIsHyperDebugStateActive(pVCpu));
6165	Assert(!CPUMIsGuestDebugStateActive(pVCpu));
6166	}
6167
6168	/* Update DR7 with the hypervisor value (other DRx registers are handled by CPUM one way or another). */
6169	u64GuestDr7 = CPUMGetHyperDR7(pVCpu);
6170	pVCpu->hmr0.s.fUsingHyperDR7 = true;
6171	fInterceptMovDRx = true;
6172	}
6173	else
6174	{
6175	/*
6176	* If the guest has enabled debug registers, we need to load them prior to
6177	* executing guest code so they'll trigger at the right time.
6178	*/
6179	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_DR7);
6180	if (pVCpu->cpum.GstCtx.dr[7] & (X86_DR7_ENABLED_MASK \| X86_DR7_GD))
6181	{
6182	if (!CPUMIsGuestDebugStateActive(pVCpu))
6183	{
6184	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
6185	Assert(CPUMIsGuestDebugStateActive(pVCpu));
6186	Assert(!CPUMIsHyperDebugStateActive(pVCpu));
6187	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
6188	}
6189	Assert(!fInterceptMovDRx);
6190	}
6191	else if (!CPUMIsGuestDebugStateActive(pVCpu))
6192	{
6193	/*
6194	* If no debugging enabled, we'll lazy load DR0-3. Unlike on AMD-V, we
6195	* must intercept #DB in order to maintain a correct DR6 guest value, and
6196	* because we need to intercept it to prevent nested #DBs from hanging the
6197	* CPU, we end up always having to intercept it. See hmR0VmxSetupVmcsXcptBitmap().
6198	*/
6199	fInterceptMovDRx = true;
6200	}
6201
6202	/* Update DR7 with the actual guest value. */
6203	u64GuestDr7 = pVCpu->cpum.GstCtx.dr[7];
6204	pVCpu->hmr0.s.fUsingHyperDR7 = false;
6205	}
6206
6207	if (fInterceptMovDRx)
6208	uProcCtls \|= VMX_PROC_CTLS_MOV_DR_EXIT;
6209	else
6210	uProcCtls &= ~VMX_PROC_CTLS_MOV_DR_EXIT;
6211
6212	/*
6213	* Update the processor-based VM-execution controls with the MOV-DRx intercepts and the
6214	* monitor-trap flag and update our cache.
6215	*/
6216	if (uProcCtls != pVmcsInfo->u32ProcCtls)
6217	{
6218	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, uProcCtls);
6219	AssertRC(rc);
6220	pVmcsInfo->u32ProcCtls = uProcCtls;
6221	}
6222
6223	/*
6224	* Update guest DR7.
6225	*/
6226	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, u64GuestDr7);
6227	AssertRC(rc);
6228
6229	/*
6230	* If we have forced EFLAGS.TF to be set because we're single-stepping in the hypervisor debugger,
6231	* we need to clear interrupt inhibition if any as otherwise it causes a VM-entry failure.
6232	*
6233	* See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
6234	*/
6235	if (fSteppingDB)
6236	{
6237	Assert(pVCpu->hm.s.fSingleInstruction);
6238	Assert(pVCpu->cpum.GstCtx.eflags.Bits.u1TF);
6239
6240	uint32_t fIntrState = 0;
6241	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
6242	AssertRC(rc);
6243
6244	if (fIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
6245	{
6246	fIntrState &= ~(VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
6247	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
6248	AssertRC(rc);
6249	}
6250	}
6251
6252	return VINF_SUCCESS;
6253	}
6254
6255
6256	#ifdef VBOX_STRICT
6257	/**
6258	* Strict function to validate segment registers.
6259	*
6260	* @param pVCpu The cross context virtual CPU structure.
6261	* @param pVmcsInfo The VMCS info. object.
6262	*
6263	* @remarks Will import guest CR0 on strict builds during validation of
6264	* segments.
6265	*/
6266	static void hmR0VmxValidateSegmentRegs(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo)
6267	{
6268	/*
6269	* Validate segment registers. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
6270	*
6271	* The reason we check for attribute value 0 in this function and not just the unusable bit is
6272	* because hmR0VmxExportGuestSegReg() only updates the VMCS' copy of the value with the
6273	* unusable bit and doesn't change the guest-context value.
6274	*/
6275	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6276	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6277	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR0);
6278	if ( !pVM->hmr0.s.vmx.fUnrestrictedGuest
6279	&& ( !CPUMIsGuestInRealModeEx(pCtx)
6280	&& !CPUMIsGuestInV86ModeEx(pCtx)))
6281	{
6282	/* Protected mode checks */
6283	/* CS */
6284	Assert(pCtx->cs.Attr.n.u1Present);
6285	Assert(!(pCtx->cs.Attr.u & 0xf00));
6286	Assert(!(pCtx->cs.Attr.u & 0xfffe0000));
6287	Assert( (pCtx->cs.u32Limit & 0xfff) == 0xfff
6288	\|\| !(pCtx->cs.Attr.n.u1Granularity));
6289	Assert( !(pCtx->cs.u32Limit & 0xfff00000)
6290	\|\| (pCtx->cs.Attr.n.u1Granularity));
6291	/* CS cannot be loaded with NULL in protected mode. */
6292	Assert(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE)); /** @todo is this really true even for 64-bit CS? */
6293	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
6294	Assert(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl);
6295	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
6296	Assert(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl);
6297	else
6298	AssertMsgFailed(("Invalid CS Type %#x\n", pCtx->cs.Attr.n.u2Dpl));
6299	/* SS */
6300	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
6301	Assert(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL));
6302	if ( !(pCtx->cr0 & X86_CR0_PE)
6303	\|\| pCtx->cs.Attr.n.u4Type == 3)
6304	{
6305	Assert(!pCtx->ss.Attr.n.u2Dpl);
6306	}
6307	if (pCtx->ss.Attr.u && !(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
6308	{
6309	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
6310	Assert(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7);
6311	Assert(pCtx->ss.Attr.n.u1Present);
6312	Assert(!(pCtx->ss.Attr.u & 0xf00));
6313	Assert(!(pCtx->ss.Attr.u & 0xfffe0000));
6314	Assert( (pCtx->ss.u32Limit & 0xfff) == 0xfff
6315	\|\| !(pCtx->ss.Attr.n.u1Granularity));
6316	Assert( !(pCtx->ss.u32Limit & 0xfff00000)
6317	\|\| (pCtx->ss.Attr.n.u1Granularity));
6318	}
6319	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxExportGuestSegReg(). */
6320	if (pCtx->ds.Attr.u && !(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
6321	{
6322	Assert(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6323	Assert(pCtx->ds.Attr.n.u1Present);
6324	Assert(pCtx->ds.Attr.n.u4Type > 11 \|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL));
6325	Assert(!(pCtx->ds.Attr.u & 0xf00));
6326	Assert(!(pCtx->ds.Attr.u & 0xfffe0000));
6327	Assert( (pCtx->ds.u32Limit & 0xfff) == 0xfff
6328	\|\| !(pCtx->ds.Attr.n.u1Granularity));
6329	Assert( !(pCtx->ds.u32Limit & 0xfff00000)
6330	\|\| (pCtx->ds.Attr.n.u1Granularity));
6331	Assert( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6332	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ));
6333	}
6334	if (pCtx->es.Attr.u && !(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
6335	{
6336	Assert(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6337	Assert(pCtx->es.Attr.n.u1Present);
6338	Assert(pCtx->es.Attr.n.u4Type > 11 \|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL));
6339	Assert(!(pCtx->es.Attr.u & 0xf00));
6340	Assert(!(pCtx->es.Attr.u & 0xfffe0000));
6341	Assert( (pCtx->es.u32Limit & 0xfff) == 0xfff
6342	\|\| !(pCtx->es.Attr.n.u1Granularity));
6343	Assert( !(pCtx->es.u32Limit & 0xfff00000)
6344	\|\| (pCtx->es.Attr.n.u1Granularity));
6345	Assert( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6346	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ));
6347	}
6348	if (pCtx->fs.Attr.u && !(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
6349	{
6350	Assert(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6351	Assert(pCtx->fs.Attr.n.u1Present);
6352	Assert(pCtx->fs.Attr.n.u4Type > 11 \|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL));
6353	Assert(!(pCtx->fs.Attr.u & 0xf00));
6354	Assert(!(pCtx->fs.Attr.u & 0xfffe0000));
6355	Assert( (pCtx->fs.u32Limit & 0xfff) == 0xfff
6356	\|\| !(pCtx->fs.Attr.n.u1Granularity));
6357	Assert( !(pCtx->fs.u32Limit & 0xfff00000)
6358	\|\| (pCtx->fs.Attr.n.u1Granularity));
6359	Assert( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6360	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ));
6361	}
6362	if (pCtx->gs.Attr.u && !(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
6363	{
6364	Assert(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
6365	Assert(pCtx->gs.Attr.n.u1Present);
6366	Assert(pCtx->gs.Attr.n.u4Type > 11 \|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL));
6367	Assert(!(pCtx->gs.Attr.u & 0xf00));
6368	Assert(!(pCtx->gs.Attr.u & 0xfffe0000));
6369	Assert( (pCtx->gs.u32Limit & 0xfff) == 0xfff
6370	\|\| !(pCtx->gs.Attr.n.u1Granularity));
6371	Assert( !(pCtx->gs.u32Limit & 0xfff00000)
6372	\|\| (pCtx->gs.Attr.n.u1Granularity));
6373	Assert( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
6374	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ));
6375	}
6376	/* 64-bit capable CPUs. */
6377	Assert(!RT_HI_U32(pCtx->cs.u64Base));
6378	Assert(!pCtx->ss.Attr.u \|\| !RT_HI_U32(pCtx->ss.u64Base));
6379	Assert(!pCtx->ds.Attr.u \|\| !RT_HI_U32(pCtx->ds.u64Base));
6380	Assert(!pCtx->es.Attr.u \|\| !RT_HI_U32(pCtx->es.u64Base));
6381	}
6382	else if ( CPUMIsGuestInV86ModeEx(pCtx)
6383	\|\| ( CPUMIsGuestInRealModeEx(pCtx)
6384	&& !pVM->hmr0.s.vmx.fUnrestrictedGuest))
6385	{
6386	/* Real and v86 mode checks. */
6387	/* hmR0VmxExportGuestSegReg() writes the modified in VMCS. We want what we're feeding to VT-x. */
6388	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
6389	if (pVmcsInfo->pShared->RealMode.fRealOnV86Active)
6390	{
6391	u32CSAttr = 0xf3; u32SSAttr = 0xf3; u32DSAttr = 0xf3;
6392	u32ESAttr = 0xf3; u32FSAttr = 0xf3; u32GSAttr = 0xf3;
6393	}
6394	else
6395	{
6396	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u; u32DSAttr = pCtx->ds.Attr.u;
6397	u32ESAttr = pCtx->es.Attr.u; u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
6398	}
6399
6400	/* CS */
6401	AssertMsg((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), ("CS base %#x %#x\n", pCtx->cs.u64Base, pCtx->cs.Sel));
6402	Assert(pCtx->cs.u32Limit == 0xffff);
6403	Assert(u32CSAttr == 0xf3);
6404	/* SS */
6405	Assert(pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4);
6406	Assert(pCtx->ss.u32Limit == 0xffff);
6407	Assert(u32SSAttr == 0xf3);
6408	/* DS */
6409	Assert(pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4);
6410	Assert(pCtx->ds.u32Limit == 0xffff);
6411	Assert(u32DSAttr == 0xf3);
6412	/* ES */
6413	Assert(pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4);
6414	Assert(pCtx->es.u32Limit == 0xffff);
6415	Assert(u32ESAttr == 0xf3);
6416	/* FS */
6417	Assert(pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4);
6418	Assert(pCtx->fs.u32Limit == 0xffff);
6419	Assert(u32FSAttr == 0xf3);
6420	/* GS */
6421	Assert(pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4);
6422	Assert(pCtx->gs.u32Limit == 0xffff);
6423	Assert(u32GSAttr == 0xf3);
6424	/* 64-bit capable CPUs. */
6425	Assert(!RT_HI_U32(pCtx->cs.u64Base));
6426	Assert(!u32SSAttr \|\| !RT_HI_U32(pCtx->ss.u64Base));
6427	Assert(!u32DSAttr \|\| !RT_HI_U32(pCtx->ds.u64Base));
6428	Assert(!u32ESAttr \|\| !RT_HI_U32(pCtx->es.u64Base));
6429	}
6430	}
6431	#endif /* VBOX_STRICT */
6432
6433
6434	/**
6435	* Exports a guest segment register into the guest-state area in the VMCS.
6436	*
6437	* @returns VBox status code.
6438	* @param pVCpu The cross context virtual CPU structure.
6439	* @param pVmcsInfo The VMCS info. object.
6440	* @param iSegReg The segment register number (X86_SREG_XXX).
6441	* @param pSelReg Pointer to the segment selector.
6442	*
6443	* @remarks No-long-jump zone!!!
6444	*/
6445	static int hmR0VmxExportGuestSegReg(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo, uint32_t iSegReg, PCCPUMSELREG pSelReg)
6446	{
6447	Assert(iSegReg < X86_SREG_COUNT);
6448
6449	uint32_t u32Access = pSelReg->Attr.u;
6450	if (!pVmcsInfo->pShared->RealMode.fRealOnV86Active)
6451	{
6452	/*
6453	* The way to differentiate between whether this is really a null selector or was just
6454	* a selector loaded with 0 in real-mode is using the segment attributes. A selector
6455	* loaded in real-mode with the value 0 is valid and usable in protected-mode and we
6456	* should -not- mark it as an unusable segment. Both the recompiler & VT-x ensures
6457	* NULL selectors loaded in protected-mode have their attribute as 0.
6458	*/
6459	if (u32Access)
6460	{ }
6461	else
6462	u32Access = X86DESCATTR_UNUSABLE;
6463	}
6464	else
6465	{
6466	/* VT-x requires our real-using-v86 mode hack to override the segment access-right bits. */
6467	u32Access = 0xf3;
6468	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
6469	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
6470	RT_NOREF_PV(pVCpu);
6471	}
6472
6473	/* Validate segment access rights. Refer to Intel spec. "26.3.1.2 Checks on Guest Segment Registers". */
6474	AssertMsg((u32Access & X86DESCATTR_UNUSABLE) \|\| (u32Access & X86_SEL_TYPE_ACCESSED),
6475	("Access bit not set for usable segment. %.2s sel=%#x attr %#x\n", "ESCSSSDSFSGS" + iSegReg * 2, pSelReg, pSelReg->Attr.u));
6476
6477	/*
6478	* Commit it to the VMCS.
6479	*/
6480	Assert((uint32_t)VMX_VMCS16_GUEST_SEG_SEL(iSegReg) == g_aVmcsSegSel[iSegReg]);
6481	Assert((uint32_t)VMX_VMCS32_GUEST_SEG_LIMIT(iSegReg) == g_aVmcsSegLimit[iSegReg]);
6482	Assert((uint32_t)VMX_VMCS32_GUEST_SEG_ACCESS_RIGHTS(iSegReg) == g_aVmcsSegAttr[iSegReg]);
6483	Assert((uint32_t)VMX_VMCS_GUEST_SEG_BASE(iSegReg) == g_aVmcsSegBase[iSegReg]);
6484	int rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_SEG_SEL(iSegReg), pSelReg->Sel); AssertRC(rc);
6485	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SEG_LIMIT(iSegReg), pSelReg->u32Limit); AssertRC(rc);
6486	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SEG_BASE(iSegReg), pSelReg->u64Base); AssertRC(rc);
6487	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SEG_ACCESS_RIGHTS(iSegReg), u32Access); AssertRC(rc);
6488	return VINF_SUCCESS;
6489	}
6490
6491
6492	/**
6493	* Exports the guest segment registers, GDTR, IDTR, LDTR, TR into the guest-state
6494	* area in the VMCS.
6495	*
6496	* @returns VBox status code.
6497	* @param pVCpu The cross context virtual CPU structure.
6498	* @param pVmxTransient The VMX-transient structure.
6499	*
6500	* @remarks Will import guest CR0 on strict builds during validation of
6501	* segments.
6502	* @remarks No-long-jump zone!!!
6503	*/
6504	static int hmR0VmxExportGuestSegRegsXdtr(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6505	{
6506	int rc = VERR_INTERNAL_ERROR_5;
6507	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6508	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6509	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6510	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
6511
6512	/*
6513	* Guest Segment registers: CS, SS, DS, ES, FS, GS.
6514	*/
6515	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SREG_MASK)
6516	{
6517	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_CS)
6518	{
6519	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CS);
6520	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6521	pVmcsInfoShared->RealMode.AttrCS.u = pCtx->cs.Attr.u;
6522	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_CS, &pCtx->cs);
6523	AssertRC(rc);
6524	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_CS);
6525	}
6526
6527	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SS)
6528	{
6529	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SS);
6530	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6531	pVmcsInfoShared->RealMode.AttrSS.u = pCtx->ss.Attr.u;
6532	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_SS, &pCtx->ss);
6533	AssertRC(rc);
6534	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SS);
6535	}
6536
6537	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_DS)
6538	{
6539	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_DS);
6540	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6541	pVmcsInfoShared->RealMode.AttrDS.u = pCtx->ds.Attr.u;
6542	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_DS, &pCtx->ds);
6543	AssertRC(rc);
6544	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_DS);
6545	}
6546
6547	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_ES)
6548	{
6549	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_ES);
6550	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6551	pVmcsInfoShared->RealMode.AttrES.u = pCtx->es.Attr.u;
6552	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_ES, &pCtx->es);
6553	AssertRC(rc);
6554	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_ES);
6555	}
6556
6557	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_FS)
6558	{
6559	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_FS);
6560	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6561	pVmcsInfoShared->RealMode.AttrFS.u = pCtx->fs.Attr.u;
6562	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_FS, &pCtx->fs);
6563	AssertRC(rc);
6564	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_FS);
6565	}
6566
6567	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_GS)
6568	{
6569	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GS);
6570	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
6571	pVmcsInfoShared->RealMode.AttrGS.u = pCtx->gs.Attr.u;
6572	rc = hmR0VmxExportGuestSegReg(pVCpu, pVmcsInfo, X86_SREG_GS, &pCtx->gs);
6573	AssertRC(rc);
6574	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_GS);
6575	}
6576
6577	#ifdef VBOX_STRICT
6578	hmR0VmxValidateSegmentRegs(pVCpu, pVmcsInfo);
6579	#endif
6580	Log4Func(("cs={%#04x base=%#RX64 limit=%#RX32 attr=%#RX32}\n", pCtx->cs.Sel, pCtx->cs.u64Base, pCtx->cs.u32Limit,
6581	pCtx->cs.Attr.u));
6582	}
6583
6584	/*
6585	* Guest TR.
6586	*/
6587	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_TR)
6588	{
6589	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_TR);
6590
6591	/*
6592	* Real-mode emulation using virtual-8086 mode with CR4.VME. Interrupt redirection is
6593	* achieved using the interrupt redirection bitmap (all bits cleared to let the guest
6594	* handle INT-n's) in the TSS. See hmR3InitFinalizeR0() to see how pRealModeTSS is setup.
6595	*/
6596	uint16_t u16Sel;
6597	uint32_t u32Limit;
6598	uint64_t u64Base;
6599	uint32_t u32AccessRights;
6600	if (!pVmcsInfoShared->RealMode.fRealOnV86Active)
6601	{
6602	u16Sel = pCtx->tr.Sel;
6603	u32Limit = pCtx->tr.u32Limit;
6604	u64Base = pCtx->tr.u64Base;
6605	u32AccessRights = pCtx->tr.Attr.u;
6606	}
6607	else
6608	{
6609	Assert(!pVmxTransient->fIsNestedGuest);
6610	Assert(pVM->hm.s.vmx.pRealModeTSS);
6611	Assert(PDMVmmDevHeapIsEnabled(pVM)); /* Guaranteed by HMCanExecuteGuest() -XXX- what about inner loop changes? */
6612
6613	/* We obtain it here every time as PCI regions could be reconfigured in the guest, changing the VMMDev base. */
6614	RTGCPHYS GCPhys;
6615	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys);
6616	AssertRCReturn(rc, rc);
6617
6618	X86DESCATTR DescAttr;
6619	DescAttr.u = 0;
6620	DescAttr.n.u1Present = 1;
6621	DescAttr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
6622
6623	u16Sel = 0;
6624	u32Limit = HM_VTX_TSS_SIZE;
6625	u64Base = GCPhys;
6626	u32AccessRights = DescAttr.u;
6627	}
6628
6629	/* Validate. */
6630	Assert(!(u16Sel & RT_BIT(2)));
6631	AssertMsg( (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_386_TSS_BUSY
6632	\|\| (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_286_TSS_BUSY, ("TSS is not busy!? %#x\n", u32AccessRights));
6633	AssertMsg(!(u32AccessRights & X86DESCATTR_UNUSABLE), ("TR unusable bit is not clear!? %#x\n", u32AccessRights));
6634	Assert(!(u32AccessRights & RT_BIT(4))); /* System MBZ.*/
6635	Assert(u32AccessRights & RT_BIT(7)); /* Present MB1.*/
6636	Assert(!(u32AccessRights & 0xf00)); /* 11:8 MBZ. */
6637	Assert(!(u32AccessRights & 0xfffe0000)); /* 31:17 MBZ. */
6638	Assert( (u32Limit & 0xfff) == 0xfff
6639	\|\| !(u32AccessRights & RT_BIT(15))); /* Granularity MBZ. */
6640	Assert( !(pCtx->tr.u32Limit & 0xfff00000)
6641	\|\| (u32AccessRights & RT_BIT(15))); /* Granularity MB1. */
6642
6643	rc = VMXWriteVmcs16(VMX_VMCS16_GUEST_TR_SEL, u16Sel); AssertRC(rc);
6644	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, u32Limit); AssertRC(rc);
6645	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, u32AccessRights); AssertRC(rc);
6646	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_TR_BASE, u64Base); AssertRC(rc);
6647
6648	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_TR);
6649	Log4Func(("tr base=%#RX64 limit=%#RX32\n", pCtx->tr.u64Base, pCtx->tr.u32Limit));
6650	}
6651
6652	/*
6653	* Guest GDTR.
6654	*/
6655	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_GDTR)
6656	{
6657	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GDTR);
6658
6659	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, pCtx->gdtr.cbGdt); AssertRC(rc);
6660	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, pCtx->gdtr.pGdt); AssertRC(rc);
6661
6662	/* Validate. */
6663	Assert(!(pCtx->gdtr.cbGdt & 0xffff0000)); /* Bits 31:16 MBZ. */
6664
6665	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_GDTR);
6666	Log4Func(("gdtr base=%#RX64 limit=%#RX32\n", pCtx->gdtr.pGdt, pCtx->gdtr.cbGdt));
6667	}
6668
6669	/*
6670	* Guest LDTR.
6671	*/
6672	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_LDTR)
6673	{
6674	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_LDTR);
6675
6676	/* The unusable bit is specific to VT-x, if it's a null selector mark it as an unusable segment. */
6677	uint32_t u32Access;
6678	if ( !pVmxTransient->fIsNestedGuest
6679	&& !pCtx->ldtr.Attr.u)
6680	u32Access = X86DESCATTR_UNUSABLE;
6681	else
6682	u32Access = pCtx->ldtr.Attr.u;
6683
6684	rc = VMXWriteVmcs16(VMX_VMCS16_GUEST_LDTR_SEL, pCtx->ldtr.Sel); AssertRC(rc);
6685	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, pCtx->ldtr.u32Limit); AssertRC(rc);
6686	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, u32Access); AssertRC(rc);
6687	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_LDTR_BASE, pCtx->ldtr.u64Base); AssertRC(rc);
6688
6689	/* Validate. */
6690	if (!(u32Access & X86DESCATTR_UNUSABLE))
6691	{
6692	Assert(!(pCtx->ldtr.Sel & RT_BIT(2))); /* TI MBZ. */
6693	Assert(pCtx->ldtr.Attr.n.u4Type == 2); /* Type MB2 (LDT). */
6694	Assert(!pCtx->ldtr.Attr.n.u1DescType); /* System MBZ. */
6695	Assert(pCtx->ldtr.Attr.n.u1Present == 1); /* Present MB1. */
6696	Assert(!pCtx->ldtr.Attr.n.u4LimitHigh); /* 11:8 MBZ. */
6697	Assert(!(pCtx->ldtr.Attr.u & 0xfffe0000)); /* 31:17 MBZ. */
6698	Assert( (pCtx->ldtr.u32Limit & 0xfff) == 0xfff
6699	\|\| !pCtx->ldtr.Attr.n.u1Granularity); /* Granularity MBZ. */
6700	Assert( !(pCtx->ldtr.u32Limit & 0xfff00000)
6701	\|\| pCtx->ldtr.Attr.n.u1Granularity); /* Granularity MB1. */
6702	}
6703
6704	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_LDTR);
6705	Log4Func(("ldtr base=%#RX64 limit=%#RX32\n", pCtx->ldtr.u64Base, pCtx->ldtr.u32Limit));
6706	}
6707
6708	/*
6709	* Guest IDTR.
6710	*/
6711	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_IDTR)
6712	{
6713	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_IDTR);
6714
6715	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, pCtx->idtr.cbIdt); AssertRC(rc);
6716	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, pCtx->idtr.pIdt); AssertRC(rc);
6717
6718	/* Validate. */
6719	Assert(!(pCtx->idtr.cbIdt & 0xffff0000)); /* Bits 31:16 MBZ. */
6720
6721	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_IDTR);
6722	Log4Func(("idtr base=%#RX64 limit=%#RX32\n", pCtx->idtr.pIdt, pCtx->idtr.cbIdt));
6723	}
6724
6725	return VINF_SUCCESS;
6726	}
6727
6728
6729	/**
6730	* Exports certain guest MSRs into the VM-entry MSR-load and VM-exit MSR-store
6731	* areas.
6732	*
6733	* These MSRs will automatically be loaded to the host CPU on every successful
6734	* VM-entry and stored from the host CPU on every successful VM-exit.
6735	*
6736	* We creates/updates MSR slots for the host MSRs in the VM-exit MSR-load area. The
6737	* actual host MSR values are not- updated here for performance reasons. See
6738	* hmR0VmxExportHostMsrs().
6739	*
6740	* We also exports the guest sysenter MSRs into the guest-state area in the VMCS.
6741	*
6742	* @returns VBox status code.
6743	* @param pVCpu The cross context virtual CPU structure.
6744	* @param pVmxTransient The VMX-transient structure.
6745	*
6746	* @remarks No-long-jump zone!!!
6747	*/
6748	static int hmR0VmxExportGuestMsrs(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6749	{
6750	AssertPtr(pVCpu);
6751	AssertPtr(pVmxTransient);
6752
6753	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6754	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6755
6756	/*
6757	* MSRs that we use the auto-load/store MSR area in the VMCS.
6758	* For 64-bit hosts, we load/restore them lazily, see hmR0VmxLazyLoadGuestMsrs(),
6759	* nothing to do here. The host MSR values are updated when it's safe in
6760	* hmR0VmxLazySaveHostMsrs().
6761	*
6762	* For nested-guests, the guests MSRs from the VM-entry MSR-load area are already
6763	* loaded (into the guest-CPU context) by the VMLAUNCH/VMRESUME instruction
6764	* emulation. The merged MSR permission bitmap will ensure that we get VM-exits
6765	* for any MSR that are not part of the lazy MSRs so we do not need to place
6766	* those MSRs into the auto-load/store MSR area. Nothing to do here.
6767	*/
6768	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_GUEST_AUTO_MSRS)
6769	{
6770	/* No auto-load/store MSRs currently. */
6771	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_VMX_GUEST_AUTO_MSRS);
6772	}
6773
6774	/*
6775	* Guest Sysenter MSRs.
6776	*/
6777	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_MSR_MASK)
6778	{
6779	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SYSENTER_MSRS);
6780
6781	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_CS_MSR)
6782	{
6783	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, pCtx->SysEnter.cs);
6784	AssertRC(rc);
6785	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_CS_MSR);
6786	}
6787
6788	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_EIP_MSR)
6789	{
6790	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SYSENTER_EIP, pCtx->SysEnter.eip);
6791	AssertRC(rc);
6792	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_EIP_MSR);
6793	}
6794
6795	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_SYSENTER_ESP_MSR)
6796	{
6797	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_SYSENTER_ESP, pCtx->SysEnter.esp);
6798	AssertRC(rc);
6799	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_SYSENTER_ESP_MSR);
6800	}
6801	}
6802
6803	/*
6804	* Guest/host EFER MSR.
6805	*/
6806	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_EFER_MSR)
6807	{
6808	/* Whether we are using the VMCS to swap the EFER MSR must have been
6809	determined earlier while exporting VM-entry/VM-exit controls. */
6810	Assert(!(ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_VMX_ENTRY_EXIT_CTLS));
6811	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
6812
6813	if (hmR0VmxShouldSwapEferMsr(pVCpu, pVmxTransient))
6814	{
6815	/*
6816	* EFER.LME is written by software, while EFER.LMA is set by the CPU to (CR0.PG & EFER.LME).
6817	* This means a guest can set EFER.LME=1 while CR0.PG=0 and EFER.LMA can remain 0.
6818	* VT-x requires that "IA-32e mode guest" VM-entry control must be identical to EFER.LMA
6819	* and to CR0.PG. Without unrestricted execution, CR0.PG (used for VT-x, not the shadow)
6820	* must always be 1. This forces us to effectively clear both EFER.LMA and EFER.LME until
6821	* the guest has also set CR0.PG=1. Otherwise, we would run into an invalid-guest state
6822	* during VM-entry.
6823	*/
6824	uint64_t uGuestEferMsr = pCtx->msrEFER;
6825	if (!pVM->hmr0.s.vmx.fUnrestrictedGuest)
6826	{
6827	if (!(pCtx->msrEFER & MSR_K6_EFER_LMA))
6828	uGuestEferMsr &= ~MSR_K6_EFER_LME;
6829	else
6830	Assert((pCtx->msrEFER & (MSR_K6_EFER_LMA \| MSR_K6_EFER_LME)) == (MSR_K6_EFER_LMA \| MSR_K6_EFER_LME));
6831	}
6832
6833	/*
6834	* If the CPU supports VMCS controls for swapping EFER, use it. Otherwise, we have no option
6835	* but to use the auto-load store MSR area in the VMCS for swapping EFER. See @bugref{7368}.
6836	*/
6837	if (g_fHmVmxSupportsVmcsEfer)
6838	{
6839	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_EFER_FULL, uGuestEferMsr);
6840	AssertRC(rc);
6841	}
6842	else
6843	{
6844	/*
6845	* We shall use the auto-load/store MSR area only for loading the EFER MSR but we must
6846	* continue to intercept guest read and write accesses to it, see @bugref{7386#c16}.
6847	*/
6848	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_EFER, uGuestEferMsr,
6849	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6850	AssertRCReturn(rc, rc);
6851	}
6852
6853	Log4Func(("efer=%#RX64 shadow=%#RX64\n", uGuestEferMsr, pCtx->msrEFER));
6854	}
6855	else if (!g_fHmVmxSupportsVmcsEfer)
6856	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K6_EFER);
6857
6858	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_EFER_MSR);
6859	}
6860
6861	/*
6862	* Other MSRs.
6863	*/
6864	if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_OTHER_MSRS)
6865	{
6866	/* Speculation Control (R/W). */
6867	HMVMX_CPUMCTX_ASSERT(pVCpu, HM_CHANGED_GUEST_OTHER_MSRS);
6868	if (pVM->cpum.ro.GuestFeatures.fIbrs)
6869	{
6870	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_IA32_SPEC_CTRL, CPUMGetGuestSpecCtrl(pVCpu),
6871	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6872	AssertRCReturn(rc, rc);
6873	}
6874
6875	/* Last Branch Record. */
6876	if (pVM->hmr0.s.vmx.fLbr)
6877	{
6878	PVMXVMCSINFOSHARED const pVmcsInfoShared = pVmxTransient->pVmcsInfo->pShared;
6879	uint32_t const idFromIpMsrStart = pVM->hmr0.s.vmx.idLbrFromIpMsrFirst;
6880	uint32_t const idToIpMsrStart = pVM->hmr0.s.vmx.idLbrToIpMsrFirst;
6881	uint32_t const cLbrStack = pVM->hmr0.s.vmx.idLbrFromIpMsrLast - pVM->hmr0.s.vmx.idLbrFromIpMsrFirst + 1;
6882	Assert(cLbrStack <= 32);
6883	for (uint32_t i = 0; i < cLbrStack; i++)
6884	{
6885	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, idFromIpMsrStart + i,
6886	pVmcsInfoShared->au64LbrFromIpMsr[i],
6887	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6888	AssertRCReturn(rc, rc);
6889
6890	/* Some CPUs don't have a Branch-To-IP MSR (P4 and related Xeons). */
6891	if (idToIpMsrStart != 0)
6892	{
6893	rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, idToIpMsrStart + i,
6894	pVmcsInfoShared->au64LbrToIpMsr[i],
6895	false /* fSetReadWrite /, false / fUpdateHostMsr */);
6896	AssertRCReturn(rc, rc);
6897	}
6898	}
6899
6900	/* Add LBR top-of-stack MSR (which contains the index to the most recent record). */
6901	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, pVM->hmr0.s.vmx.idLbrTosMsr,
6902	pVmcsInfoShared->u64LbrTosMsr, false /* fSetReadWrite */,
6903	false /* fUpdateHostMsr */);
6904	AssertRCReturn(rc, rc);
6905	}
6906
6907	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_OTHER_MSRS);
6908	}
6909
6910	return VINF_SUCCESS;
6911	}
6912
6913
6914	/**
6915	* Wrapper for running the guest code in VT-x.
6916	*
6917	* @returns VBox status code, no informational status codes.
6918	* @param pVCpu The cross context virtual CPU structure.
6919	* @param pVmxTransient The VMX-transient structure.
6920	*
6921	* @remarks No-long-jump zone!!!
6922	*/
6923	DECLINLINE(int) hmR0VmxRunGuest(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient)
6924	{
6925	/* Mark that HM is the keeper of all guest-CPU registers now that we're going to execute guest code. */
6926	pVCpu->cpum.GstCtx.fExtrn \|= HMVMX_CPUMCTX_EXTRN_ALL \| CPUMCTX_EXTRN_KEEPER_HM;
6927
6928	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
6929	bool const fResumeVM = RT_BOOL(pVmcsInfo->fVmcsState & VMX_V_VMCS_LAUNCH_STATE_LAUNCHED);
6930	#ifdef VBOX_WITH_STATISTICS
6931	if (fResumeVM)
6932	STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxVmResume);
6933	else
6934	STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxVmLaunch);
6935	#endif
6936	int rc = pVCpu->hmr0.s.vmx.pfnStartVm(pVmcsInfo, pVCpu, fResumeVM);
6937	AssertMsg(rc <= VINF_SUCCESS, ("%Rrc\n", rc));
6938	return rc;
6939	}
6940
6941
6942	/**
6943	* Reports world-switch error and dumps some useful debug info.
6944	*
6945	* @param pVCpu The cross context virtual CPU structure.
6946	* @param rcVMRun The return code from VMLAUNCH/VMRESUME.
6947	* @param pVmxTransient The VMX-transient structure (only
6948	* exitReason updated).
6949	*/
6950	static void hmR0VmxReportWorldSwitchError(PVMCPUCC pVCpu, int rcVMRun, PVMXTRANSIENT pVmxTransient)
6951	{
6952	Assert(pVCpu);
6953	Assert(pVmxTransient);
6954	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
6955
6956	Log4Func(("VM-entry failure: %Rrc\n", rcVMRun));
6957	switch (rcVMRun)
6958	{
6959	case VERR_VMX_INVALID_VMXON_PTR:
6960	AssertFailed();
6961	break;
6962	case VINF_SUCCESS: /* VMLAUNCH/VMRESUME succeeded but VM-entry failed... yeah, true story. */
6963	case VERR_VMX_UNABLE_TO_START_VM: /* VMLAUNCH/VMRESUME itself failed. */
6964	{
6965	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &pVCpu->hm.s.vmx.LastError.u32ExitReason);
6966	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
6967	AssertRC(rc);
6968	hmR0VmxReadExitQualVmcs(pVmxTransient);
6969
6970	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hmr0.s.idEnteredCpu;
6971	/* LastError.idCurrentCpu was already updated in hmR0VmxPreRunGuestCommitted().
6972	Cannot do it here as we may have been long preempted. */
6973
6974	#ifdef VBOX_STRICT
6975	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
6976	Log4(("uExitReason %#RX32 (VmxTransient %#RX16)\n", pVCpu->hm.s.vmx.LastError.u32ExitReason,
6977	pVmxTransient->uExitReason));
6978	Log4(("Exit Qualification %#RX64\n", pVmxTransient->uExitQual));
6979	Log4(("InstrError %#RX32\n", pVCpu->hm.s.vmx.LastError.u32InstrError));
6980	if (pVCpu->hm.s.vmx.LastError.u32InstrError <= HMVMX_INSTR_ERROR_MAX)
6981	Log4(("InstrError Desc. \"%s\"\n", g_apszVmxInstrErrors[pVCpu->hm.s.vmx.LastError.u32InstrError]));
6982	else
6983	Log4(("InstrError Desc. Range exceeded %u\n", HMVMX_INSTR_ERROR_MAX));
6984	Log4(("Entered host CPU %u\n", pVCpu->hm.s.vmx.LastError.idEnteredCpu));
6985	Log4(("Current host CPU %u\n", pVCpu->hm.s.vmx.LastError.idCurrentCpu));
6986
6987	static struct
6988	{
6989	/** Name of the field to log. */
6990	const char *pszName;
6991	/** The VMCS field. */
6992	uint32_t uVmcsField;
6993	/** Whether host support of this field needs to be checked. */
6994	bool fCheckSupport;
6995	} const s_aVmcsFields[] =
6996	{
6997	{ "VMX_VMCS32_CTRL_PIN_EXEC", VMX_VMCS32_CTRL_PIN_EXEC, false },
6998	{ "VMX_VMCS32_CTRL_PROC_EXEC", VMX_VMCS32_CTRL_PROC_EXEC, false },
6999	{ "VMX_VMCS32_CTRL_PROC_EXEC2", VMX_VMCS32_CTRL_PROC_EXEC2, true },
7000	{ "VMX_VMCS32_CTRL_ENTRY", VMX_VMCS32_CTRL_ENTRY, false },
7001	{ "VMX_VMCS32_CTRL_EXIT", VMX_VMCS32_CTRL_EXIT, false },
7002	{ "VMX_VMCS32_CTRL_CR3_TARGET_COUNT", VMX_VMCS32_CTRL_CR3_TARGET_COUNT, false },
7003	{ "VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO", VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, false },
7004	{ "VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE", VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, false },
7005	{ "VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH", VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, false },
7006	{ "VMX_VMCS32_CTRL_TPR_THRESHOLD", VMX_VMCS32_CTRL_TPR_THRESHOLD, false },
7007	{ "VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT", VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, false },
7008	{ "VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT", VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, false },
7009	{ "VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT", VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, false },
7010	{ "VMX_VMCS32_CTRL_EXCEPTION_BITMAP", VMX_VMCS32_CTRL_EXCEPTION_BITMAP, false },
7011	{ "VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK", VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, false },
7012	{ "VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH", VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, false },
7013	{ "VMX_VMCS_CTRL_CR0_MASK", VMX_VMCS_CTRL_CR0_MASK, false },
7014	{ "VMX_VMCS_CTRL_CR0_READ_SHADOW", VMX_VMCS_CTRL_CR0_READ_SHADOW, false },
7015	{ "VMX_VMCS_CTRL_CR4_MASK", VMX_VMCS_CTRL_CR4_MASK, false },
7016	{ "VMX_VMCS_CTRL_CR4_READ_SHADOW", VMX_VMCS_CTRL_CR4_READ_SHADOW, false },
7017	{ "VMX_VMCS64_CTRL_EPTP_FULL", VMX_VMCS64_CTRL_EPTP_FULL, true },
7018	{ "VMX_VMCS_GUEST_RIP", VMX_VMCS_GUEST_RIP, false },
7019	{ "VMX_VMCS_GUEST_RSP", VMX_VMCS_GUEST_RSP, false },
7020	{ "VMX_VMCS_GUEST_RFLAGS", VMX_VMCS_GUEST_RFLAGS, false },
7021	{ "VMX_VMCS16_VPID", VMX_VMCS16_VPID, true, },
7022	{ "VMX_VMCS_HOST_CR0", VMX_VMCS_HOST_CR0, false },
7023	{ "VMX_VMCS_HOST_CR3", VMX_VMCS_HOST_CR3, false },
7024	{ "VMX_VMCS_HOST_CR4", VMX_VMCS_HOST_CR4, false },
7025	/* The order of selector fields below are fixed! */
7026	{ "VMX_VMCS16_HOST_ES_SEL", VMX_VMCS16_HOST_ES_SEL, false },
7027	{ "VMX_VMCS16_HOST_CS_SEL", VMX_VMCS16_HOST_CS_SEL, false },
7028	{ "VMX_VMCS16_HOST_SS_SEL", VMX_VMCS16_HOST_SS_SEL, false },
7029	{ "VMX_VMCS16_HOST_DS_SEL", VMX_VMCS16_HOST_DS_SEL, false },
7030	{ "VMX_VMCS16_HOST_FS_SEL", VMX_VMCS16_HOST_FS_SEL, false },
7031	{ "VMX_VMCS16_HOST_GS_SEL", VMX_VMCS16_HOST_GS_SEL, false },
7032	{ "VMX_VMCS16_HOST_TR_SEL", VMX_VMCS16_HOST_TR_SEL, false },
7033	/* End of ordered selector fields. */
7034	{ "VMX_VMCS_HOST_TR_BASE", VMX_VMCS_HOST_TR_BASE, false },
7035	{ "VMX_VMCS_HOST_GDTR_BASE", VMX_VMCS_HOST_GDTR_BASE, false },
7036	{ "VMX_VMCS_HOST_IDTR_BASE", VMX_VMCS_HOST_IDTR_BASE, false },
7037	{ "VMX_VMCS32_HOST_SYSENTER_CS", VMX_VMCS32_HOST_SYSENTER_CS, false },
7038	{ "VMX_VMCS_HOST_SYSENTER_EIP", VMX_VMCS_HOST_SYSENTER_EIP, false },
7039	{ "VMX_VMCS_HOST_SYSENTER_ESP", VMX_VMCS_HOST_SYSENTER_ESP, false },
7040	{ "VMX_VMCS_HOST_RSP", VMX_VMCS_HOST_RSP, false },
7041	{ "VMX_VMCS_HOST_RIP", VMX_VMCS_HOST_RIP, false }
7042	};
7043
7044	RTGDTR HostGdtr;
7045	ASMGetGDTR(&HostGdtr);
7046
7047	uint32_t const cVmcsFields = RT_ELEMENTS(s_aVmcsFields);
7048	for (uint32_t i = 0; i < cVmcsFields; i++)
7049	{
7050	uint32_t const uVmcsField = s_aVmcsFields[i].uVmcsField;
7051
7052	bool fSupported;
7053	if (!s_aVmcsFields[i].fCheckSupport)
7054	fSupported = true;
7055	else
7056	{
7057	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7058	switch (uVmcsField)
7059	{
7060	case VMX_VMCS64_CTRL_EPTP_FULL: fSupported = pVM->hmr0.s.fNestedPaging; break;
7061	case VMX_VMCS16_VPID: fSupported = pVM->hmr0.s.vmx.fVpid; break;
7062	case VMX_VMCS32_CTRL_PROC_EXEC2:
7063	fSupported = RT_BOOL(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS);
7064	break;
7065	default:
7066	AssertMsgFailedReturnVoid(("Failed to provide VMCS field support for %#RX32\n", uVmcsField));
7067	}
7068	}
7069
7070	if (fSupported)
7071	{
7072	uint8_t const uWidth = RT_BF_GET(uVmcsField, VMX_BF_VMCSFIELD_WIDTH);
7073	switch (uWidth)
7074	{
7075	case VMX_VMCSFIELD_WIDTH_16BIT:
7076	{
7077	uint16_t u16Val;
7078	rc = VMXReadVmcs16(uVmcsField, &u16Val);
7079	AssertRC(rc);
7080	Log4(("%-40s = %#RX16\n", s_aVmcsFields[i].pszName, u16Val));
7081
7082	if ( uVmcsField >= VMX_VMCS16_HOST_ES_SEL
7083	&& uVmcsField <= VMX_VMCS16_HOST_TR_SEL)
7084	{
7085	if (u16Val < HostGdtr.cbGdt)
7086	{
7087	/* Order of selectors in s_apszSel is fixed and matches the order in s_aVmcsFields. */
7088	static const char * const s_apszSel[] = { "Host ES", "Host CS", "Host SS", "Host DS",
7089	"Host FS", "Host GS", "Host TR" };
7090	uint8_t const idxSel = RT_BF_GET(uVmcsField, VMX_BF_VMCSFIELD_INDEX);
7091	Assert(idxSel < RT_ELEMENTS(s_apszSel));
7092	PCX86DESCHC pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u16Val & X86_SEL_MASK));
7093	hmR0DumpDescriptor(pDesc, u16Val, s_apszSel[idxSel]);
7094	}
7095	else
7096	Log4((" Selector value exceeds GDT limit!\n"));
7097	}
7098	break;
7099	}
7100
7101	case VMX_VMCSFIELD_WIDTH_32BIT:
7102	{
7103	uint32_t u32Val;
7104	rc = VMXReadVmcs32(uVmcsField, &u32Val);
7105	AssertRC(rc);
7106	Log4(("%-40s = %#RX32\n", s_aVmcsFields[i].pszName, u32Val));
7107	break;
7108	}
7109
7110	case VMX_VMCSFIELD_WIDTH_64BIT:
7111	case VMX_VMCSFIELD_WIDTH_NATURAL:
7112	{
7113	uint64_t u64Val;
7114	rc = VMXReadVmcs64(uVmcsField, &u64Val);
7115	AssertRC(rc);
7116	Log4(("%-40s = %#RX64\n", s_aVmcsFields[i].pszName, u64Val));
7117	break;
7118	}
7119	}
7120	}
7121	}
7122
7123	Log4(("MSR_K6_EFER = %#RX64\n", ASMRdMsr(MSR_K6_EFER)));
7124	Log4(("MSR_K8_CSTAR = %#RX64\n", ASMRdMsr(MSR_K8_CSTAR)));
7125	Log4(("MSR_K8_LSTAR = %#RX64\n", ASMRdMsr(MSR_K8_LSTAR)));
7126	Log4(("MSR_K6_STAR = %#RX64\n", ASMRdMsr(MSR_K6_STAR)));
7127	Log4(("MSR_K8_SF_MASK = %#RX64\n", ASMRdMsr(MSR_K8_SF_MASK)));
7128	Log4(("MSR_K8_KERNEL_GS_BASE = %#RX64\n", ASMRdMsr(MSR_K8_KERNEL_GS_BASE)));
7129	#endif /* VBOX_STRICT */
7130	break;
7131	}
7132
7133	default:
7134	/* Impossible */
7135	AssertMsgFailed(("hmR0VmxReportWorldSwitchError %Rrc (%#x)\n", rcVMRun, rcVMRun));
7136	break;
7137	}
7138	}
7139
7140
7141	/**
7142	* Sets up the usage of TSC-offsetting and updates the VMCS.
7143	*
7144	* If offsetting is not possible, cause VM-exits on RDTSC(P)s. Also sets up the
7145	* VMX-preemption timer.
7146	*
7147	* @returns VBox status code.
7148	* @param pVCpu The cross context virtual CPU structure.
7149	* @param pVmxTransient The VMX-transient structure.
7150	* @param idCurrentCpu The current CPU number.
7151	*
7152	* @remarks No-long-jump zone!!!
7153	*/
7154	static void hmR0VmxUpdateTscOffsettingAndPreemptTimer(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, RTCPUID idCurrentCpu)
7155	{
7156	bool fOffsettedTsc;
7157	bool fParavirtTsc;
7158	uint64_t uTscOffset;
7159	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7160	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
7161
7162	if (pVM->hmr0.s.vmx.fUsePreemptTimer)
7163	{
7164	/* The TMCpuTickGetDeadlineAndTscOffset function is expensive (calling it on
7165	every entry slowed down the bs2-test1 CPUID testcase by ~33% (on an 10980xe). */
7166	uint64_t cTicksToDeadline;
7167	if ( idCurrentCpu == pVCpu->hmr0.s.idLastCpu
7168	&& TMVirtualSyncIsCurrentDeadlineVersion(pVM, pVCpu->hmr0.s.vmx.uTscDeadlineVersion))
7169	{
7170	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatVmxPreemptionReusingDeadline);
7171	fOffsettedTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &uTscOffset, &fParavirtTsc);
7172	cTicksToDeadline = pVCpu->hmr0.s.vmx.uTscDeadline - SUPReadTsc();
7173	if ((int64_t)cTicksToDeadline > 0)
7174	{ /* hopefully */ }
7175	else
7176	{
7177	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatVmxPreemptionReusingDeadlineExpired);
7178	cTicksToDeadline = 0;
7179	}
7180	}
7181	else
7182	{
7183	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatVmxPreemptionRecalcingDeadline);
7184	cTicksToDeadline = TMCpuTickGetDeadlineAndTscOffset(pVM, pVCpu, &uTscOffset, &fOffsettedTsc, &fParavirtTsc,
7185	&pVCpu->hmr0.s.vmx.uTscDeadline,
7186	&pVCpu->hmr0.s.vmx.uTscDeadlineVersion);
7187	pVCpu->hmr0.s.vmx.uTscDeadline += cTicksToDeadline;
7188	if (cTicksToDeadline >= 128)
7189	{ /* hopefully */ }
7190	else
7191	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatVmxPreemptionRecalcingDeadlineExpired);
7192	}
7193
7194	/* Make sure the returned values have sane upper and lower boundaries. */
7195	uint64_t const u64CpuHz = SUPGetCpuHzFromGipBySetIndex(g_pSUPGlobalInfoPage, pVCpu->iHostCpuSet);
7196	cTicksToDeadline = RT_MIN(cTicksToDeadline, u64CpuHz / 64); /* 1/64th of a second, 15.625ms. / /* @todo r=bird: Once real+virtual timers move to separate thread, we can raise the upper limit (16ms isn't much). ASSUMES working poke cpu function. */
7197	cTicksToDeadline = RT_MAX(cTicksToDeadline, u64CpuHz / 32678); /* 1/32768th of a second, ~30us. */
7198	cTicksToDeadline >>= pVM->hm.s.vmx.cPreemptTimerShift;
7199
7200	/** @todo r=ramshankar: We need to find a way to integrate nested-guest
7201	* preemption timers here. We probably need to clamp the preemption timer,
7202	* after converting the timer value to the host. */
7203	uint32_t const cPreemptionTickCount = (uint32_t)RT_MIN(cTicksToDeadline, UINT32_MAX - 16);
7204	int rc = VMXWriteVmcs32(VMX_VMCS32_PREEMPT_TIMER_VALUE, cPreemptionTickCount);
7205	AssertRC(rc);
7206	}
7207	else
7208	fOffsettedTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &uTscOffset, &fParavirtTsc);
7209
7210	if (fParavirtTsc)
7211	{
7212	/* Currently neither Hyper-V nor KVM need to update their paravirt. TSC
7213	information before every VM-entry, hence disable it for performance sake. */
7214	#if 0
7215	int rc = GIMR0UpdateParavirtTsc(pVM, 0 /* u64Offset */);
7216	AssertRC(rc);
7217	#endif
7218	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
7219	}
7220
7221	if ( fOffsettedTsc
7222	&& RT_LIKELY(!pVCpu->hmr0.s.fDebugWantRdTscExit))
7223	{
7224	if (pVmxTransient->fIsNestedGuest)
7225	uTscOffset = CPUMApplyNestedGuestTscOffset(pVCpu, uTscOffset);
7226	hmR0VmxSetTscOffsetVmcs(pVmcsInfo, uTscOffset);
7227	hmR0VmxRemoveProcCtlsVmcs(pVCpu, pVmxTransient, VMX_PROC_CTLS_RDTSC_EXIT);
7228	}
7229	else
7230	{
7231	/* We can't use TSC-offsetting (non-fixed TSC, warp drive active etc.), VM-exit on RDTSC(P). */
7232	hmR0VmxSetProcCtlsVmcs(pVmxTransient, VMX_PROC_CTLS_RDTSC_EXIT);
7233	}
7234	}
7235
7236
7237	/**
7238	* Gets the IEM exception flags for the specified vector and IDT vectoring /
7239	* VM-exit interruption info type.
7240	*
7241	* @returns The IEM exception flags.
7242	* @param uVector The event vector.
7243	* @param uVmxEventType The VMX event type.
7244	*
7245	* @remarks This function currently only constructs flags required for
7246	* IEMEvaluateRecursiveXcpt and not the complete flags (e.g, error-code
7247	* and CR2 aspects of an exception are not included).
7248	*/
7249	static uint32_t hmR0VmxGetIemXcptFlags(uint8_t uVector, uint32_t uVmxEventType)
7250	{
7251	uint32_t fIemXcptFlags;
7252	switch (uVmxEventType)
7253	{
7254	case VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT:
7255	case VMX_IDT_VECTORING_INFO_TYPE_NMI:
7256	fIemXcptFlags = IEM_XCPT_FLAGS_T_CPU_XCPT;
7257	break;
7258
7259	case VMX_IDT_VECTORING_INFO_TYPE_EXT_INT:
7260	fIemXcptFlags = IEM_XCPT_FLAGS_T_EXT_INT;
7261	break;
7262
7263	case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
7264	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT \| IEM_XCPT_FLAGS_ICEBP_INSTR;
7265	break;
7266
7267	case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT:
7268	{
7269	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
7270	if (uVector == X86_XCPT_BP)
7271	fIemXcptFlags \|= IEM_XCPT_FLAGS_BP_INSTR;
7272	else if (uVector == X86_XCPT_OF)
7273	fIemXcptFlags \|= IEM_XCPT_FLAGS_OF_INSTR;
7274	else
7275	{
7276	fIemXcptFlags = 0;
7277	AssertMsgFailed(("Unexpected vector for software exception. uVector=%#x", uVector));
7278	}
7279	break;
7280	}
7281
7282	case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
7283	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
7284	break;
7285
7286	default:
7287	fIemXcptFlags = 0;
7288	AssertMsgFailed(("Unexpected vector type! uVmxEventType=%#x uVector=%#x", uVmxEventType, uVector));
7289	break;
7290	}
7291	return fIemXcptFlags;
7292	}
7293
7294
7295	/**
7296	* Sets an event as a pending event to be injected into the guest.
7297	*
7298	* @param pVCpu The cross context virtual CPU structure.
7299	* @param u32IntInfo The VM-entry interruption-information field.
7300	* @param cbInstr The VM-entry instruction length in bytes (for
7301	* software interrupts, exceptions and privileged
7302	* software exceptions).
7303	* @param u32ErrCode The VM-entry exception error code.
7304	* @param GCPtrFaultAddress The fault-address (CR2) in case it's a
7305	* page-fault.
7306	*/
7307	DECLINLINE(void) hmR0VmxSetPendingEvent(PVMCPUCC pVCpu, uint32_t u32IntInfo, uint32_t cbInstr, uint32_t u32ErrCode,
7308	RTGCUINTPTR GCPtrFaultAddress)
7309	{
7310	Assert(!pVCpu->hm.s.Event.fPending);
7311	pVCpu->hm.s.Event.fPending = true;
7312	pVCpu->hm.s.Event.u64IntInfo = u32IntInfo;
7313	pVCpu->hm.s.Event.u32ErrCode = u32ErrCode;
7314	pVCpu->hm.s.Event.cbInstr = cbInstr;
7315	pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress;
7316	}
7317
7318
7319	/**
7320	* Sets an external interrupt as pending-for-injection into the VM.
7321	*
7322	* @param pVCpu The cross context virtual CPU structure.
7323	* @param u8Interrupt The external interrupt vector.
7324	*/
7325	DECLINLINE(void) hmR0VmxSetPendingExtInt(PVMCPUCC pVCpu, uint8_t u8Interrupt)
7326	{
7327	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, u8Interrupt)
7328	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
7329	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7330	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7331	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7332	}
7333
7334
7335	/**
7336	* Sets an NMI (\#NMI) exception as pending-for-injection into the VM.
7337	*
7338	* @param pVCpu The cross context virtual CPU structure.
7339	*/
7340	DECLINLINE(void) hmR0VmxSetPendingXcptNmi(PVMCPUCC pVCpu)
7341	{
7342	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_NMI)
7343	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_NMI)
7344	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7345	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7346	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7347	}
7348
7349
7350	/**
7351	* Sets a double-fault (\#DF) exception as pending-for-injection into the VM.
7352	*
7353	* @param pVCpu The cross context virtual CPU structure.
7354	*/
7355	DECLINLINE(void) hmR0VmxSetPendingXcptDF(PVMCPUCC pVCpu)
7356	{
7357	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DF)
7358	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7359	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7360	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7361	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7362	}
7363
7364
7365	/**
7366	* Sets an invalid-opcode (\#UD) exception as pending-for-injection into the VM.
7367	*
7368	* @param pVCpu The cross context virtual CPU structure.
7369	*/
7370	DECLINLINE(void) hmR0VmxSetPendingXcptUD(PVMCPUCC pVCpu)
7371	{
7372	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_UD)
7373	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7374	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7375	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7376	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7377	}
7378
7379
7380	/**
7381	* Sets a debug (\#DB) exception as pending-for-injection into the VM.
7382	*
7383	* @param pVCpu The cross context virtual CPU structure.
7384	*/
7385	DECLINLINE(void) hmR0VmxSetPendingXcptDB(PVMCPUCC pVCpu)
7386	{
7387	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DB)
7388	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7389	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
7390	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7391	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7392	}
7393
7394
7395	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7396	/**
7397	* Sets a general-protection (\#GP) exception as pending-for-injection into the VM.
7398	*
7399	* @param pVCpu The cross context virtual CPU structure.
7400	* @param u32ErrCode The error code for the general-protection exception.
7401	*/
7402	DECLINLINE(void) hmR0VmxSetPendingXcptGP(PVMCPUCC pVCpu, uint32_t u32ErrCode)
7403	{
7404	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_GP)
7405	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7406	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7407	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7408	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrCode, 0 / GCPtrFaultAddress */);
7409	}
7410
7411
7412	/**
7413	* Sets a stack (\#SS) exception as pending-for-injection into the VM.
7414	*
7415	* @param pVCpu The cross context virtual CPU structure.
7416	* @param u32ErrCode The error code for the stack exception.
7417	*/
7418	DECLINLINE(void) hmR0VmxSetPendingXcptSS(PVMCPUCC pVCpu, uint32_t u32ErrCode)
7419	{
7420	uint32_t const u32IntInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_SS)
7421	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
7422	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 1)
7423	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7424	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrCode, 0 / GCPtrFaultAddress */);
7425	}
7426	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
7427
7428
7429	/**
7430	* Fixes up attributes for the specified segment register.
7431	*
7432	* @param pVCpu The cross context virtual CPU structure.
7433	* @param pSelReg The segment register that needs fixing.
7434	* @param pszRegName The register name (for logging and assertions).
7435	*/
7436	static void hmR0VmxFixUnusableSegRegAttr(PVMCPUCC pVCpu, PCPUMSELREG pSelReg, const char *pszRegName)
7437	{
7438	Assert(pSelReg->Attr.u & X86DESCATTR_UNUSABLE);
7439
7440	/*
7441	* If VT-x marks the segment as unusable, most other bits remain undefined:
7442	* - For CS the L, D and G bits have meaning.
7443	* - For SS the DPL has meaning (it -is- the CPL for Intel and VBox).
7444	* - For the remaining data segments no bits are defined.
7445	*
7446	* The present bit and the unusable bit has been observed to be set at the
7447	* same time (the selector was supposed to be invalid as we started executing
7448	* a V8086 interrupt in ring-0).
7449	*
7450	* What should be important for the rest of the VBox code, is that the P bit is
7451	* cleared. Some of the other VBox code recognizes the unusable bit, but
7452	* AMD-V certainly don't, and REM doesn't really either. So, to be on the
7453	* safe side here, we'll strip off P and other bits we don't care about. If
7454	* any code breaks because Attr.u != 0 when Sel < 4, it should be fixed.
7455	*
7456	* See Intel spec. 27.3.2 "Saving Segment Registers and Descriptor-Table Registers".
7457	*/
7458	#ifdef VBOX_STRICT
7459	uint32_t const uAttr = pSelReg->Attr.u;
7460	#endif
7461
7462	/* Masking off: X86DESCATTR_P, X86DESCATTR_LIMIT_HIGH, and X86DESCATTR_AVL. The latter two are really irrelevant. */
7463	pSelReg->Attr.u &= X86DESCATTR_UNUSABLE \| X86DESCATTR_L \| X86DESCATTR_D \| X86DESCATTR_G
7464	\| X86DESCATTR_DPL \| X86DESCATTR_TYPE \| X86DESCATTR_DT;
7465
7466	#ifdef VBOX_STRICT
7467	VMMRZCallRing3Disable(pVCpu);
7468	Log4Func(("Unusable %s: sel=%#x attr=%#x -> %#x\n", pszRegName, pSelReg->Sel, uAttr, pSelReg->Attr.u));
7469	# ifdef DEBUG_bird
7470	AssertMsg((uAttr & ~X86DESCATTR_P) == pSelReg->Attr.u,
7471	("%s: %#x != %#x (sel=%#x base=%#llx limit=%#x)\n",
7472	pszRegName, uAttr, pSelReg->Attr.u, pSelReg->Sel, pSelReg->u64Base, pSelReg->u32Limit));
7473	# endif
7474	VMMRZCallRing3Enable(pVCpu);
7475	NOREF(uAttr);
7476	#endif
7477	RT_NOREF2(pVCpu, pszRegName);
7478	}
7479
7480
7481	/**
7482	* Imports a guest segment register from the current VMCS into the guest-CPU
7483	* context.
7484	*
7485	* @param pVCpu The cross context virtual CPU structure.
7486	* @param iSegReg The segment register number (X86_SREG_XXX).
7487	*
7488	* @remarks Called with interrupts and/or preemption disabled.
7489	*/
7490	static void hmR0VmxImportGuestSegReg(PVMCPUCC pVCpu, uint32_t iSegReg)
7491	{
7492	Assert(iSegReg < X86_SREG_COUNT);
7493	Assert((uint32_t)VMX_VMCS16_GUEST_SEG_SEL(iSegReg) == g_aVmcsSegSel[iSegReg]);
7494	Assert((uint32_t)VMX_VMCS32_GUEST_SEG_LIMIT(iSegReg) == g_aVmcsSegLimit[iSegReg]);
7495	Assert((uint32_t)VMX_VMCS32_GUEST_SEG_ACCESS_RIGHTS(iSegReg) == g_aVmcsSegAttr[iSegReg]);
7496	Assert((uint32_t)VMX_VMCS_GUEST_SEG_BASE(iSegReg) == g_aVmcsSegBase[iSegReg]);
7497
7498	PCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
7499
7500	uint16_t u16Sel;
7501	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_SEG_SEL(iSegReg), &u16Sel); AssertRC(rc);
7502	pSelReg->Sel = u16Sel;
7503	pSelReg->ValidSel = u16Sel;
7504
7505	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SEG_LIMIT(iSegReg), &pSelReg->u32Limit); AssertRC(rc);
7506	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SEG_BASE(iSegReg), &pSelReg->u64Base); AssertRC(rc);
7507
7508	uint32_t u32Attr;
7509	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SEG_ACCESS_RIGHTS(iSegReg), &u32Attr); AssertRC(rc);
7510	pSelReg->Attr.u = u32Attr;
7511	if (u32Attr & X86DESCATTR_UNUSABLE)
7512	hmR0VmxFixUnusableSegRegAttr(pVCpu, pSelReg, "ES\0CS\0SS\0DS\0FS\0GS" + iSegReg * 3);
7513
7514	pSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
7515	}
7516
7517
7518	/**
7519	* Imports the guest LDTR from the current VMCS into the guest-CPU context.
7520	*
7521	* @param pVCpu The cross context virtual CPU structure.
7522	*
7523	* @remarks Called with interrupts and/or preemption disabled.
7524	*/
7525	static void hmR0VmxImportGuestLdtr(PVMCPUCC pVCpu)
7526	{
7527	uint16_t u16Sel;
7528	uint64_t u64Base;
7529	uint32_t u32Limit, u32Attr;
7530	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_LDTR_SEL, &u16Sel); AssertRC(rc);
7531	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, &u32Limit); AssertRC(rc);
7532	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, &u32Attr); AssertRC(rc);
7533	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_LDTR_BASE, &u64Base); AssertRC(rc);
7534
7535	pVCpu->cpum.GstCtx.ldtr.Sel = u16Sel;
7536	pVCpu->cpum.GstCtx.ldtr.ValidSel = u16Sel;
7537	pVCpu->cpum.GstCtx.ldtr.fFlags = CPUMSELREG_FLAGS_VALID;
7538	pVCpu->cpum.GstCtx.ldtr.u32Limit = u32Limit;
7539	pVCpu->cpum.GstCtx.ldtr.u64Base = u64Base;
7540	pVCpu->cpum.GstCtx.ldtr.Attr.u = u32Attr;
7541	if (u32Attr & X86DESCATTR_UNUSABLE)
7542	hmR0VmxFixUnusableSegRegAttr(pVCpu, &pVCpu->cpum.GstCtx.ldtr, "LDTR");
7543	}
7544
7545
7546	/**
7547	* Imports the guest TR from the current VMCS into the guest-CPU context.
7548	*
7549	* @param pVCpu The cross context virtual CPU structure.
7550	*
7551	* @remarks Called with interrupts and/or preemption disabled.
7552	*/
7553	static void hmR0VmxImportGuestTr(PVMCPUCC pVCpu)
7554	{
7555	uint16_t u16Sel;
7556	uint64_t u64Base;
7557	uint32_t u32Limit, u32Attr;
7558	int rc = VMXReadVmcs16(VMX_VMCS16_GUEST_TR_SEL, &u16Sel); AssertRC(rc);
7559	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, &u32Limit); AssertRC(rc);
7560	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, &u32Attr); AssertRC(rc);
7561	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_TR_BASE, &u64Base); AssertRC(rc);
7562
7563	pVCpu->cpum.GstCtx.tr.Sel = u16Sel;
7564	pVCpu->cpum.GstCtx.tr.ValidSel = u16Sel;
7565	pVCpu->cpum.GstCtx.tr.fFlags = CPUMSELREG_FLAGS_VALID;
7566	pVCpu->cpum.GstCtx.tr.u32Limit = u32Limit;
7567	pVCpu->cpum.GstCtx.tr.u64Base = u64Base;
7568	pVCpu->cpum.GstCtx.tr.Attr.u = u32Attr;
7569	/* TR is the only selector that can never be unusable. */
7570	Assert(!(u32Attr & X86DESCATTR_UNUSABLE));
7571	}
7572
7573
7574	/**
7575	* Imports the guest RIP from the VMCS back into the guest-CPU context.
7576	*
7577	* @param pVCpu The cross context virtual CPU structure.
7578	*
7579	* @remarks Called with interrupts and/or preemption disabled, should not assert!
7580	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7581	* instead!!!
7582	*/
7583	static void hmR0VmxImportGuestRip(PVMCPUCC pVCpu)
7584	{
7585	uint64_t u64Val;
7586	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7587	if (pCtx->fExtrn & CPUMCTX_EXTRN_RIP)
7588	{
7589	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RIP, &u64Val);
7590	AssertRC(rc);
7591
7592	pCtx->rip = u64Val;
7593	EMR0HistoryUpdatePC(pVCpu, pCtx->rip, false);
7594	pCtx->fExtrn &= ~CPUMCTX_EXTRN_RIP;
7595	}
7596	}
7597
7598
7599	/**
7600	* Imports the guest RFLAGS from the VMCS back into the guest-CPU context.
7601	*
7602	* @param pVCpu The cross context virtual CPU structure.
7603	* @param pVmcsInfo The VMCS info. object.
7604	*
7605	* @remarks Called with interrupts and/or preemption disabled, should not assert!
7606	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7607	* instead!!!
7608	*/
7609	static void hmR0VmxImportGuestRFlags(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
7610	{
7611	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7612	if (pCtx->fExtrn & CPUMCTX_EXTRN_RFLAGS)
7613	{
7614	uint64_t u64Val;
7615	int rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RFLAGS, &u64Val);
7616	AssertRC(rc);
7617
7618	pCtx->rflags.u64 = u64Val;
7619	PCVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
7620	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
7621	{
7622	pCtx->eflags.Bits.u1VM = 0;
7623	pCtx->eflags.Bits.u2IOPL = pVmcsInfoShared->RealMode.Eflags.Bits.u2IOPL;
7624	}
7625	pCtx->fExtrn &= ~CPUMCTX_EXTRN_RFLAGS;
7626	}
7627	}
7628
7629
7630	/**
7631	* Imports the guest interruptibility-state from the VMCS back into the guest-CPU
7632	* context.
7633	*
7634	* @param pVCpu The cross context virtual CPU structure.
7635	* @param pVmcsInfo The VMCS info. object.
7636	*
7637	* @remarks Called with interrupts and/or preemption disabled, try not to assert and
7638	* do not log!
7639	* @remarks Do -not- call this function directly, use hmR0VmxImportGuestState()
7640	* instead!!!
7641	*/
7642	static void hmR0VmxImportGuestIntrState(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
7643	{
7644	uint32_t u32Val;
7645	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &u32Val); AssertRC(rc);
7646	if (!u32Val)
7647	{
7648	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
7649	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
7650	CPUMSetGuestNmiBlocking(pVCpu, false);
7651	}
7652	else
7653	{
7654	/*
7655	* We must import RIP here to set our EM interrupt-inhibited state.
7656	* We also import RFLAGS as our code that evaluates pending interrupts
7657	* before VM-entry requires it.
7658	*/
7659	hmR0VmxImportGuestRip(pVCpu);
7660	hmR0VmxImportGuestRFlags(pVCpu, pVmcsInfo);
7661
7662	if (u32Val & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS \| VMX_VMCS_GUEST_INT_STATE_BLOCK_STI))
7663	EMSetInhibitInterruptsPC(pVCpu, pVCpu->cpum.GstCtx.rip);
7664	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
7665	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
7666
7667	bool const fNmiBlocking = RT_BOOL(u32Val & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI);
7668	CPUMSetGuestNmiBlocking(pVCpu, fNmiBlocking);
7669	}
7670	}
7671
7672
7673	/**
7674	* Worker for VMXR0ImportStateOnDemand.
7675	*
7676	* @returns VBox status code.
7677	* @param pVCpu The cross context virtual CPU structure.
7678	* @param pVmcsInfo The VMCS info. object.
7679	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
7680	*/
7681	static int hmR0VmxImportGuestState(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint64_t fWhat)
7682	{
7683	int rc = VINF_SUCCESS;
7684	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7685	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7686	uint32_t u32Val;
7687
7688	/*
7689	* Note! This is hack to workaround a mysterious BSOD observed with release builds
7690	* on Windows 10 64-bit hosts. Profile and debug builds are not affected and
7691	* neither are other host platforms.
7692	*
7693	* Committing this temporarily as it prevents BSOD.
7694	*
7695	* Update: This is very likely a compiler optimization bug, see @bugref{9180}.
7696	*/
7697	#ifdef RT_OS_WINDOWS
7698	if (pVM == 0 \|\| pVM == (void *)(uintptr_t)-1)
7699	return VERR_HM_IPE_1;
7700	#endif
7701
7702	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatImportGuestState, x);
7703
7704	/*
7705	* We disable interrupts to make the updating of the state and in particular
7706	* the fExtrn modification atomic wrt to preemption hooks.
7707	*/
7708	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
7709
7710	fWhat &= pCtx->fExtrn;
7711	if (fWhat)
7712	{
7713	do
7714	{
7715	if (fWhat & CPUMCTX_EXTRN_RIP)
7716	hmR0VmxImportGuestRip(pVCpu);
7717
7718	if (fWhat & CPUMCTX_EXTRN_RFLAGS)
7719	hmR0VmxImportGuestRFlags(pVCpu, pVmcsInfo);
7720
7721	if (fWhat & CPUMCTX_EXTRN_HM_VMX_INT_STATE)
7722	hmR0VmxImportGuestIntrState(pVCpu, pVmcsInfo);
7723
7724	if (fWhat & CPUMCTX_EXTRN_RSP)
7725	{
7726	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RSP, &pCtx->rsp);
7727	AssertRC(rc);
7728	}
7729
7730	if (fWhat & CPUMCTX_EXTRN_SREG_MASK)
7731	{
7732	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
7733	bool const fRealOnV86Active = pVmcsInfoShared->RealMode.fRealOnV86Active;
7734	if (fWhat & CPUMCTX_EXTRN_CS)
7735	{
7736	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_CS);
7737	hmR0VmxImportGuestRip(pVCpu);
7738	if (fRealOnV86Active)
7739	pCtx->cs.Attr.u = pVmcsInfoShared->RealMode.AttrCS.u;
7740	EMR0HistoryUpdatePC(pVCpu, pCtx->cs.u64Base + pCtx->rip, true /* fFlattened */);
7741	}
7742	if (fWhat & CPUMCTX_EXTRN_SS)
7743	{
7744	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_SS);
7745	if (fRealOnV86Active)
7746	pCtx->ss.Attr.u = pVmcsInfoShared->RealMode.AttrSS.u;
7747	}
7748	if (fWhat & CPUMCTX_EXTRN_DS)
7749	{
7750	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_DS);
7751	if (fRealOnV86Active)
7752	pCtx->ds.Attr.u = pVmcsInfoShared->RealMode.AttrDS.u;
7753	}
7754	if (fWhat & CPUMCTX_EXTRN_ES)
7755	{
7756	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_ES);
7757	if (fRealOnV86Active)
7758	pCtx->es.Attr.u = pVmcsInfoShared->RealMode.AttrES.u;
7759	}
7760	if (fWhat & CPUMCTX_EXTRN_FS)
7761	{
7762	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_FS);
7763	if (fRealOnV86Active)
7764	pCtx->fs.Attr.u = pVmcsInfoShared->RealMode.AttrFS.u;
7765	}
7766	if (fWhat & CPUMCTX_EXTRN_GS)
7767	{
7768	hmR0VmxImportGuestSegReg(pVCpu, X86_SREG_GS);
7769	if (fRealOnV86Active)
7770	pCtx->gs.Attr.u = pVmcsInfoShared->RealMode.AttrGS.u;
7771	}
7772	}
7773
7774	if (fWhat & CPUMCTX_EXTRN_TABLE_MASK)
7775	{
7776	if (fWhat & CPUMCTX_EXTRN_LDTR)
7777	hmR0VmxImportGuestLdtr(pVCpu);
7778
7779	if (fWhat & CPUMCTX_EXTRN_GDTR)
7780	{
7781	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, &pCtx->gdtr.pGdt); AssertRC(rc);
7782	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val); AssertRC(rc);
7783	pCtx->gdtr.cbGdt = u32Val;
7784	}
7785
7786	/* Guest IDTR. */
7787	if (fWhat & CPUMCTX_EXTRN_IDTR)
7788	{
7789	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, &pCtx->idtr.pIdt); AssertRC(rc);
7790	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val); AssertRC(rc);
7791	pCtx->idtr.cbIdt = u32Val;
7792	}
7793
7794	/* Guest TR. */
7795	if (fWhat & CPUMCTX_EXTRN_TR)
7796	{
7797	/* Real-mode emulation using virtual-8086 mode has the fake TSS (pRealModeTSS) in TR,
7798	don't need to import that one. */
7799	if (!pVmcsInfo->pShared->RealMode.fRealOnV86Active)
7800	hmR0VmxImportGuestTr(pVCpu);
7801	}
7802	}
7803
7804	if (fWhat & CPUMCTX_EXTRN_DR7)
7805	{
7806	if (!pVCpu->hmr0.s.fUsingHyperDR7)
7807	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_DR7, &pCtx->dr[7]); AssertRC(rc);
7808	}
7809
7810	if (fWhat & CPUMCTX_EXTRN_SYSENTER_MSRS)
7811	{
7812	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SYSENTER_EIP, &pCtx->SysEnter.eip); AssertRC(rc);
7813	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_SYSENTER_ESP, &pCtx->SysEnter.esp); AssertRC(rc);
7814	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, &u32Val); AssertRC(rc);
7815	pCtx->SysEnter.cs = u32Val;
7816	}
7817
7818	if (fWhat & CPUMCTX_EXTRN_KERNEL_GS_BASE)
7819	{
7820	if ( pVM->hmr0.s.fAllow64BitGuests
7821	&& (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
7822	pCtx->msrKERNELGSBASE = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
7823	}
7824
7825	if (fWhat & CPUMCTX_EXTRN_SYSCALL_MSRS)
7826	{
7827	if ( pVM->hmr0.s.fAllow64BitGuests
7828	&& (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
7829	{
7830	pCtx->msrLSTAR = ASMRdMsr(MSR_K8_LSTAR);
7831	pCtx->msrSTAR = ASMRdMsr(MSR_K6_STAR);
7832	pCtx->msrSFMASK = ASMRdMsr(MSR_K8_SF_MASK);
7833	}
7834	}
7835
7836	if (fWhat & (CPUMCTX_EXTRN_TSC_AUX \| CPUMCTX_EXTRN_OTHER_MSRS))
7837	{
7838	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
7839	PCVMXAUTOMSR pMsrs = (PCVMXAUTOMSR)pVmcsInfo->pvGuestMsrStore;
7840	uint32_t const cMsrs = pVmcsInfo->cExitMsrStore;
7841	Assert(pMsrs);
7842	Assert(cMsrs <= VMX_MISC_MAX_MSRS(g_HmMsrs.u.vmx.u64Misc));
7843	Assert(sizeof(pMsrs) cMsrs <= X86_PAGE_4K_SIZE);
7844	for (uint32_t i = 0; i < cMsrs; i++)
7845	{
7846	uint32_t const idMsr = pMsrs[i].u32Msr;
7847	switch (idMsr)
7848	{
7849	case MSR_K8_TSC_AUX: CPUMSetGuestTscAux(pVCpu, pMsrs[i].u64Value); break;
7850	case MSR_IA32_SPEC_CTRL: CPUMSetGuestSpecCtrl(pVCpu, pMsrs[i].u64Value); break;
7851	case MSR_K6_EFER: /* Can't be changed without causing a VM-exit */ break;
7852	default:
7853	{
7854	uint32_t idxLbrMsr;
7855	if (pVM->hmr0.s.vmx.fLbr)
7856	{
7857	if (hmR0VmxIsLbrBranchFromMsr(pVM, idMsr, &idxLbrMsr))
7858	{
7859	Assert(idxLbrMsr < RT_ELEMENTS(pVmcsInfoShared->au64LbrFromIpMsr));
7860	pVmcsInfoShared->au64LbrFromIpMsr[idxLbrMsr] = pMsrs[i].u64Value;
7861	break;
7862	}
7863	if (hmR0VmxIsLbrBranchToMsr(pVM, idMsr, &idxLbrMsr))
7864	{
7865	Assert(idxLbrMsr < RT_ELEMENTS(pVmcsInfoShared->au64LbrFromIpMsr));
7866	pVmcsInfoShared->au64LbrToIpMsr[idxLbrMsr] = pMsrs[i].u64Value;
7867	break;
7868	}
7869	if (idMsr == pVM->hmr0.s.vmx.idLbrTosMsr)
7870	{
7871	pVmcsInfoShared->u64LbrTosMsr = pMsrs[i].u64Value;
7872	break;
7873	}
7874	/* Fallthru (no break) */
7875	}
7876	pCtx->fExtrn = 0;
7877	pVCpu->hm.s.u32HMError = pMsrs->u32Msr;
7878	ASMSetFlags(fEFlags);
7879	AssertMsgFailed(("Unexpected MSR in auto-load/store area. idMsr=%#RX32 cMsrs=%u\n", idMsr, cMsrs));
7880	return VERR_HM_UNEXPECTED_LD_ST_MSR;
7881	}
7882	}
7883	}
7884	}
7885
7886	if (fWhat & CPUMCTX_EXTRN_CR_MASK)
7887	{
7888	if (fWhat & CPUMCTX_EXTRN_CR0)
7889	{
7890	uint64_t u64Cr0;
7891	uint64_t u64Shadow;
7892	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64Cr0); AssertRC(rc);
7893	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, &u64Shadow); AssertRC(rc);
7894	#ifndef VBOX_WITH_NESTED_HWVIRT_VMX
7895	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7896	\| (u64Shadow & pVmcsInfo->u64Cr0Mask);
7897	#else
7898	if (!CPUMIsGuestInVmxNonRootMode(pCtx))
7899	{
7900	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7901	\| (u64Shadow & pVmcsInfo->u64Cr0Mask);
7902	}
7903	else
7904	{
7905	/*
7906	* We've merged the guest and nested-guest's CR0 guest/host mask while executing
7907	* the nested-guest using hardware-assisted VMX. Accordingly we need to
7908	* re-construct CR0. See @bugref{9180#c95} for details.
7909	*/
7910	PCVMXVMCSINFO const pVmcsInfoGst = &pVCpu->hmr0.s.vmx.VmcsInfo;
7911	PVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
7912	u64Cr0 = (u64Cr0 & ~pVmcsInfo->u64Cr0Mask)
7913	\| (pVmcsNstGst->u64GuestCr0.u & pVmcsNstGst->u64Cr0Mask.u)
7914	\| (u64Shadow & (pVmcsInfoGst->u64Cr0Mask & ~pVmcsNstGst->u64Cr0Mask.u));
7915	}
7916	#endif
7917	VMMRZCallRing3Disable(pVCpu); /* May call into PGM which has Log statements. */
7918	CPUMSetGuestCR0(pVCpu, u64Cr0);
7919	VMMRZCallRing3Enable(pVCpu);
7920	}
7921
7922	if (fWhat & CPUMCTX_EXTRN_CR4)
7923	{
7924	uint64_t u64Cr4;
7925	uint64_t u64Shadow;
7926	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR4, &u64Cr4); AssertRC(rc);
7927	rc \|= VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, &u64Shadow); AssertRC(rc);
7928	#ifndef VBOX_WITH_NESTED_HWVIRT_VMX
7929	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7930	\| (u64Shadow & pVmcsInfo->u64Cr4Mask);
7931	#else
7932	if (!CPUMIsGuestInVmxNonRootMode(pCtx))
7933	{
7934	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7935	\| (u64Shadow & pVmcsInfo->u64Cr4Mask);
7936	}
7937	else
7938	{
7939	/*
7940	* We've merged the guest and nested-guest's CR4 guest/host mask while executing
7941	* the nested-guest using hardware-assisted VMX. Accordingly we need to
7942	* re-construct CR4. See @bugref{9180#c95} for details.
7943	*/
7944	PCVMXVMCSINFO const pVmcsInfoGst = &pVCpu->hmr0.s.vmx.VmcsInfo;
7945	PVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
7946	u64Cr4 = (u64Cr4 & ~pVmcsInfo->u64Cr4Mask)
7947	\| (pVmcsNstGst->u64GuestCr4.u & pVmcsNstGst->u64Cr4Mask.u)
7948	\| (u64Shadow & (pVmcsInfoGst->u64Cr4Mask & ~pVmcsNstGst->u64Cr4Mask.u));
7949	}
7950	#endif
7951	pCtx->cr4 = u64Cr4;
7952	}
7953
7954	if (fWhat & CPUMCTX_EXTRN_CR3)
7955	{
7956	/* CR0.PG bit changes are always intercepted, so it's up to date. */
7957	if ( pVM->hmr0.s.vmx.fUnrestrictedGuest
7958	\|\| ( pVM->hmr0.s.fNestedPaging
7959	&& CPUMIsGuestPagingEnabledEx(pCtx)))
7960	{
7961	uint64_t u64Cr3;
7962	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR3, &u64Cr3); AssertRC(rc);
7963	if (pCtx->cr3 != u64Cr3)
7964	{
7965	pCtx->cr3 = u64Cr3;
7966	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
7967	}
7968
7969	/*
7970	* If the guest is in PAE mode, sync back the PDPE's into the guest state.
7971	* CR4.PAE, CR0.PG, EFER MSR changes are always intercepted, so they're up to date.
7972	*/
7973	if (CPUMIsGuestInPAEModeEx(pCtx))
7974	{
7975	X86PDPE aPaePdpes[4];
7976	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &aPaePdpes[0].u); AssertRC(rc);
7977	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &aPaePdpes[1].u); AssertRC(rc);
7978	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &aPaePdpes[2].u); AssertRC(rc);
7979	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &aPaePdpes[3].u); AssertRC(rc);
7980	if (memcmp(&aPaePdpes[0], &pCtx->aPaePdpes[0], sizeof(aPaePdpes)))
7981	{
7982	memcpy(&pCtx->aPaePdpes[0], &aPaePdpes[0], sizeof(aPaePdpes));
7983	/* PGM now updates PAE PDPTEs while updating CR3. */
7984	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
7985	}
7986	}
7987	}
7988	}
7989	}
7990
7991	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
7992	if (fWhat & CPUMCTX_EXTRN_HWVIRT)
7993	{
7994	if ( (pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
7995	&& !CPUMIsGuestInVmxNonRootMode(pCtx))
7996	{
7997	Assert(CPUMIsGuestInVmxRootMode(pCtx));
7998	rc = hmR0VmxCopyShadowToNstGstVmcs(pVCpu, pVmcsInfo);
7999	if (RT_SUCCESS(rc))
8000	{ /* likely */ }
8001	else
8002	break;
8003	}
8004	}
8005	#endif
8006	} while (0);
8007
8008	if (RT_SUCCESS(rc))
8009	{
8010	/* Update fExtrn. */
8011	pCtx->fExtrn &= ~fWhat;
8012
8013	/* If everything has been imported, clear the HM keeper bit. */
8014	if (!(pCtx->fExtrn & HMVMX_CPUMCTX_EXTRN_ALL))
8015	{
8016	pCtx->fExtrn &= ~CPUMCTX_EXTRN_KEEPER_HM;
8017	Assert(!pCtx->fExtrn);
8018	}
8019	}
8020	}
8021	else
8022	AssertMsg(!pCtx->fExtrn \|\| (pCtx->fExtrn & HMVMX_CPUMCTX_EXTRN_ALL), ("%#RX64\n", pCtx->fExtrn));
8023
8024	/*
8025	* Restore interrupts.
8026	*/
8027	ASMSetFlags(fEFlags);
8028
8029	STAM_PROFILE_ADV_STOP(& pVCpu->hm.s.StatImportGuestState, x);
8030
8031	if (RT_SUCCESS(rc))
8032	{ /* likely */ }
8033	else
8034	return rc;
8035
8036	/*
8037	* Honor any pending CR3 updates.
8038	*
8039	* Consider this scenario: VM-exit -> VMMRZCallRing3Enable() -> do stuff that causes a longjmp -> VMXR0CallRing3Callback()
8040	* -> VMMRZCallRing3Disable() -> hmR0VmxImportGuestState() -> Sets VMCPU_FF_HM_UPDATE_CR3 pending -> return from the longjmp
8041	* -> continue with VM-exit handling -> hmR0VmxImportGuestState() and here we are.
8042	*
8043	* The reason for such complicated handling is because VM-exits that call into PGM expect CR3 to be up-to-date and thus
8044	* if any CR3-saves -before- the VM-exit (longjmp) postponed the CR3 update via the force-flag, any VM-exit handler that
8045	* calls into PGM when it re-saves CR3 will end up here and we call PGMUpdateCR3(). This is why the code below should
8046	* -NOT- check if CPUMCTX_EXTRN_CR3 is set!
8047	*
8048	* The longjmp exit path can't check these CR3 force-flags and call code that takes a lock again. We cover for it here.
8049	*/
8050	if (VMMRZCallRing3IsEnabled(pVCpu))
8051	{
8052	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
8053	{
8054	Assert(!(ASMAtomicUoReadU64(&pCtx->fExtrn) & CPUMCTX_EXTRN_CR3));
8055	PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu), false /* fPdpesMapped */);
8056	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
8057	}
8058	}
8059
8060	return VINF_SUCCESS;
8061	}
8062
8063
8064	/**
8065	* Saves the guest state from the VMCS into the guest-CPU context.
8066	*
8067	* @returns VBox status code.
8068	* @param pVCpu The cross context virtual CPU structure.
8069	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
8070	*/
8071	VMMR0DECL(int) VMXR0ImportStateOnDemand(PVMCPUCC pVCpu, uint64_t fWhat)
8072	{
8073	AssertPtr(pVCpu);
8074	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8075	return hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fWhat);
8076	}
8077
8078
8079	/**
8080	* Check per-VM and per-VCPU force flag actions that require us to go back to
8081	* ring-3 for one reason or another.
8082	*
8083	* @returns Strict VBox status code (i.e. informational status codes too)
8084	* @retval VINF_SUCCESS if we don't have any actions that require going back to
8085	* ring-3.
8086	* @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync.
8087	* @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware
8088	* interrupts)
8089	* @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires
8090	* all EMTs to be in ring-3.
8091	* @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests.
8092	* @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return
8093	* to the EM loop.
8094	*
8095	* @param pVCpu The cross context virtual CPU structure.
8096	* @param pVmxTransient The VMX-transient structure.
8097	* @param fStepping Whether we are single-stepping the guest using the
8098	* hypervisor debugger.
8099	*
8100	* @remarks This might cause nested-guest VM-exits, caller must check if the guest
8101	* is no longer in VMX non-root mode.
8102	*/
8103	static VBOXSTRICTRC hmR0VmxCheckForceFlags(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, bool fStepping)
8104	{
8105	Assert(VMMRZCallRing3IsEnabled(pVCpu));
8106
8107	/*
8108	* Update pending interrupts into the APIC's IRR.
8109	*/
8110	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UPDATE_APIC))
8111	APICUpdatePendingInterrupts(pVCpu);
8112
8113	/*
8114	* Anything pending? Should be more likely than not if we're doing a good job.
8115	*/
8116	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
8117	if ( !fStepping
8118	? !VM_FF_IS_ANY_SET(pVM, VM_FF_HP_R0_PRE_HM_MASK)
8119	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HP_R0_PRE_HM_MASK)
8120	: !VM_FF_IS_ANY_SET(pVM, VM_FF_HP_R0_PRE_HM_STEP_MASK)
8121	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) )
8122	return VINF_SUCCESS;
8123
8124	/* Pending PGM C3 sync. */
8125	if (VMCPU_FF_IS_ANY_SET(pVCpu,VMCPU_FF_PGM_SYNC_CR3 \| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL))
8126	{
8127	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8128	Assert(!(ASMAtomicUoReadU64(&pCtx->fExtrn) & (CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_CR3 \| CPUMCTX_EXTRN_CR4)));
8129	VBOXSTRICTRC rcStrict = PGMSyncCR3(pVCpu, pCtx->cr0, pCtx->cr3, pCtx->cr4,
8130	VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
8131	if (rcStrict != VINF_SUCCESS)
8132	{
8133	AssertRC(VBOXSTRICTRC_VAL(rcStrict));
8134	Log4Func(("PGMSyncCR3 forcing us back to ring-3. rc2=%d\n", VBOXSTRICTRC_VAL(rcStrict)));
8135	return rcStrict;
8136	}
8137	}
8138
8139	/* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */
8140	if ( VM_FF_IS_ANY_SET(pVM, VM_FF_HM_TO_R3_MASK)
8141	\|\| VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
8142	{
8143	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
8144	int rc = RT_LIKELY(!VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_RAW_TO_R3 : VINF_EM_NO_MEMORY;
8145	Log4Func(("HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc));
8146	return rc;
8147	}
8148
8149	/* Pending VM request packets, such as hardware interrupts. */
8150	if ( VM_FF_IS_SET(pVM, VM_FF_REQUEST)
8151	\|\| VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_REQUEST))
8152	{
8153	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchVmReq);
8154	Log4Func(("Pending VM request forcing us back to ring-3\n"));
8155	return VINF_EM_PENDING_REQUEST;
8156	}
8157
8158	/* Pending PGM pool flushes. */
8159	if (VM_FF_IS_SET(pVM, VM_FF_PGM_POOL_FLUSH_PENDING))
8160	{
8161	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPgmPoolFlush);
8162	Log4Func(("PGM pool flush pending forcing us back to ring-3\n"));
8163	return VINF_PGM_POOL_FLUSH_PENDING;
8164	}
8165
8166	/* Pending DMA requests. */
8167	if (VM_FF_IS_SET(pVM, VM_FF_PDM_DMA))
8168	{
8169	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchDma);
8170	Log4Func(("Pending DMA request forcing us back to ring-3\n"));
8171	return VINF_EM_RAW_TO_R3;
8172	}
8173
8174	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8175	/*
8176	* Pending nested-guest events.
8177	*
8178	* Please note the priority of these events are specified and important.
8179	* See Intel spec. 29.4.3.2 "APIC-Write Emulation".
8180	* See Intel spec. 6.9 "Priority Among Simultaneous Exceptions And Interrupts".
8181	*/
8182	if (pVmxTransient->fIsNestedGuest)
8183	{
8184	/* Pending nested-guest APIC-write. */
8185	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_APIC_WRITE))
8186	{
8187	Log4Func(("Pending nested-guest APIC-write\n"));
8188	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitApicWrite(pVCpu);
8189	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8190	return rcStrict;
8191	}
8192
8193	/* Pending nested-guest monitor-trap flag (MTF). */
8194	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_MTF))
8195	{
8196	Log4Func(("Pending nested-guest MTF\n"));
8197	VBOXSTRICTRC rcStrict = IEMExecVmxVmexit(pVCpu, VMX_EXIT_MTF, 0 /* uExitQual */);
8198	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8199	return rcStrict;
8200	}
8201
8202	/* Pending nested-guest VMX-preemption timer expired. */
8203	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_PREEMPT_TIMER))
8204	{
8205	Log4Func(("Pending nested-guest preempt timer\n"));
8206	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitPreemptTimer(pVCpu);
8207	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
8208	return rcStrict;
8209	}
8210	}
8211	#else
8212	NOREF(pVmxTransient);
8213	#endif
8214
8215	return VINF_SUCCESS;
8216	}
8217
8218
8219	/**
8220	* Converts any TRPM trap into a pending HM event. This is typically used when
8221	* entering from ring-3 (not longjmp returns).
8222	*
8223	* @param pVCpu The cross context virtual CPU structure.
8224	*/
8225	static void hmR0VmxTrpmTrapToPendingEvent(PVMCPUCC pVCpu)
8226	{
8227	Assert(TRPMHasTrap(pVCpu));
8228	Assert(!pVCpu->hm.s.Event.fPending);
8229
8230	uint8_t uVector;
8231	TRPMEVENT enmTrpmEvent;
8232	uint32_t uErrCode;
8233	RTGCUINTPTR GCPtrFaultAddress;
8234	uint8_t cbInstr;
8235	bool fIcebp;
8236
8237	int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr, &fIcebp);
8238	AssertRC(rc);
8239
8240	uint32_t u32IntInfo;
8241	u32IntInfo = uVector \| VMX_IDT_VECTORING_INFO_VALID;
8242	u32IntInfo \|= HMTrpmEventTypeToVmxEventType(uVector, enmTrpmEvent, fIcebp);
8243
8244	rc = TRPMResetTrap(pVCpu);
8245	AssertRC(rc);
8246	Log4(("TRPM->HM event: u32IntInfo=%#RX32 enmTrpmEvent=%d cbInstr=%u uErrCode=%#RX32 GCPtrFaultAddress=%#RGv\n",
8247	u32IntInfo, enmTrpmEvent, cbInstr, uErrCode, GCPtrFaultAddress));
8248
8249	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, uErrCode, GCPtrFaultAddress);
8250	}
8251
8252
8253	/**
8254	* Converts the pending HM event into a TRPM trap.
8255	*
8256	* @param pVCpu The cross context virtual CPU structure.
8257	*/
8258	static void hmR0VmxPendingEventToTrpmTrap(PVMCPUCC pVCpu)
8259	{
8260	Assert(pVCpu->hm.s.Event.fPending);
8261
8262	/* If a trap was already pending, we did something wrong! */
8263	Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo /, NULL / pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP);
8264
8265	uint32_t const u32IntInfo = pVCpu->hm.s.Event.u64IntInfo;
8266	uint32_t const uVector = VMX_IDT_VECTORING_INFO_VECTOR(u32IntInfo);
8267	TRPMEVENT const enmTrapType = HMVmxEventTypeToTrpmEventType(u32IntInfo);
8268
8269	Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, enmTrapType));
8270
8271	int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType);
8272	AssertRC(rc);
8273
8274	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(u32IntInfo))
8275	TRPMSetErrorCode(pVCpu, pVCpu->hm.s.Event.u32ErrCode);
8276
8277	if (VMX_IDT_VECTORING_INFO_IS_XCPT_PF(u32IntInfo))
8278	TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress);
8279	else
8280	{
8281	uint8_t const uVectorType = VMX_IDT_VECTORING_INFO_TYPE(u32IntInfo);
8282	switch (uVectorType)
8283	{
8284	case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
8285	TRPMSetTrapDueToIcebp(pVCpu);
8286	RT_FALL_THRU();
8287	case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
8288	case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT:
8289	{
8290	AssertMsg( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
8291	\|\| ( uVector == X86_XCPT_BP /* INT3 */
8292	\|\| uVector == X86_XCPT_OF /* INTO */
8293	\|\| uVector == X86_XCPT_DB /* INT1 (ICEBP) */),
8294	("Invalid vector: uVector=%#x uVectorType=%#x\n", uVector, uVectorType));
8295	TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr);
8296	break;
8297	}
8298	}
8299	}
8300
8301	/* We're now done converting the pending event. */
8302	pVCpu->hm.s.Event.fPending = false;
8303	}
8304
8305
8306	/**
8307	* Sets the interrupt-window exiting control in the VMCS which instructs VT-x to
8308	* cause a VM-exit as soon as the guest is in a state to receive interrupts.
8309	*
8310	* @param pVmcsInfo The VMCS info. object.
8311	*/
8312	static void hmR0VmxSetIntWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8313	{
8314	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_INT_WINDOW_EXIT)
8315	{
8316	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT))
8317	{
8318	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_INT_WINDOW_EXIT;
8319	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8320	AssertRC(rc);
8321	}
8322	} /* else we will deliver interrupts whenever the guest Vm-exits next and is in a state to receive the interrupt. */
8323	}
8324
8325
8326	/**
8327	* Clears the interrupt-window exiting control in the VMCS.
8328	*
8329	* @param pVmcsInfo The VMCS info. object.
8330	*/
8331	DECLINLINE(void) hmR0VmxClearIntWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8332	{
8333	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT)
8334	{
8335	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_INT_WINDOW_EXIT;
8336	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8337	AssertRC(rc);
8338	}
8339	}
8340
8341
8342	/**
8343	* Sets the NMI-window exiting control in the VMCS which instructs VT-x to
8344	* cause a VM-exit as soon as the guest is in a state to receive NMIs.
8345	*
8346	* @param pVmcsInfo The VMCS info. object.
8347	*/
8348	static void hmR0VmxSetNmiWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8349	{
8350	if (g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
8351	{
8352	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))
8353	{
8354	pVmcsInfo->u32ProcCtls \|= VMX_PROC_CTLS_NMI_WINDOW_EXIT;
8355	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8356	AssertRC(rc);
8357	Log4Func(("Setup NMI-window exiting\n"));
8358	}
8359	} /* else we will deliver NMIs whenever we VM-exit next, even possibly nesting NMIs. Can't be helped on ancient CPUs. */
8360	}
8361
8362
8363	/**
8364	* Clears the NMI-window exiting control in the VMCS.
8365	*
8366	* @param pVmcsInfo The VMCS info. object.
8367	*/
8368	DECLINLINE(void) hmR0VmxClearNmiWindowExitVmcs(PVMXVMCSINFO pVmcsInfo)
8369	{
8370	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
8371	{
8372	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_NMI_WINDOW_EXIT;
8373	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
8374	AssertRC(rc);
8375	}
8376	}
8377
8378
8379	/**
8380	* Does the necessary state syncing before returning to ring-3 for any reason
8381	* (longjmp, preemption, voluntary exits to ring-3) from VT-x.
8382	*
8383	* @returns VBox status code.
8384	* @param pVCpu The cross context virtual CPU structure.
8385	* @param fImportState Whether to import the guest state from the VMCS back
8386	* to the guest-CPU context.
8387	*
8388	* @remarks No-long-jmp zone!!!
8389	*/
8390	static int hmR0VmxLeave(PVMCPUCC pVCpu, bool fImportState)
8391	{
8392	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8393	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8394
8395	RTCPUID const idCpu = RTMpCpuId();
8396	Log4Func(("HostCpuId=%u\n", idCpu));
8397
8398	/*
8399	* !!! IMPORTANT !!!
8400	* If you modify code here, check whether VMXR0CallRing3Callback() needs to be updated too.
8401	*/
8402
8403	/* Save the guest state if necessary. */
8404	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8405	if (fImportState)
8406	{
8407	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
8408	AssertRCReturn(rc, rc);
8409	}
8410
8411	/* Restore host FPU state if necessary. We will resync on next R0 reentry. */
8412	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
8413	Assert(!CPUMIsGuestFPUStateActive(pVCpu));
8414
8415	/* Restore host debug registers if necessary. We will resync on next R0 reentry. */
8416	#ifdef VBOX_STRICT
8417	if (CPUMIsHyperDebugStateActive(pVCpu))
8418	Assert(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MOV_DR_EXIT);
8419	#endif
8420	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
8421	Assert(!CPUMIsGuestDebugStateActive(pVCpu));
8422	Assert(!CPUMIsHyperDebugStateActive(pVCpu));
8423
8424	/* Restore host-state bits that VT-x only restores partially. */
8425	if (pVCpu->hmr0.s.vmx.fRestoreHostFlags > VMX_RESTORE_HOST_REQUIRED)
8426	{
8427	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hmr0.s.vmx.fRestoreHostFlags, idCpu));
8428	VMXRestoreHostState(pVCpu->hmr0.s.vmx.fRestoreHostFlags, &pVCpu->hmr0.s.vmx.RestoreHost);
8429	}
8430	pVCpu->hmr0.s.vmx.fRestoreHostFlags = 0;
8431
8432	/* Restore the lazy host MSRs as we're leaving VT-x context. */
8433	if (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
8434	{
8435	/* We shouldn't restore the host MSRs without saving the guest MSRs first. */
8436	if (!fImportState)
8437	{
8438	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_KERNEL_GS_BASE \| CPUMCTX_EXTRN_SYSCALL_MSRS);
8439	AssertRCReturn(rc, rc);
8440	}
8441	hmR0VmxLazyRestoreHostMsrs(pVCpu);
8442	Assert(!pVCpu->hmr0.s.vmx.fLazyMsrs);
8443	}
8444	else
8445	pVCpu->hmr0.s.vmx.fLazyMsrs = 0;
8446
8447	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
8448	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = false;
8449
8450	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry);
8451	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatImportGuestState);
8452	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExportGuestState);
8453	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatPreExit);
8454	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitHandling);
8455	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitIO);
8456	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitMovCRx);
8457	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitXcptNmi);
8458	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitVmentry);
8459	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
8460
8461	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
8462
8463	/** @todo This partially defeats the purpose of having preemption hooks.
8464	* The problem is, deregistering the hooks should be moved to a place that
8465	* lasts until the EMT is about to be destroyed not everytime while leaving HM
8466	* context.
8467	*/
8468	int rc = hmR0VmxClearVmcs(pVmcsInfo);
8469	AssertRCReturn(rc, rc);
8470
8471	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8472	/*
8473	* A valid shadow VMCS is made active as part of VM-entry. It is necessary to
8474	* clear a shadow VMCS before allowing that VMCS to become active on another
8475	* logical processor. We may or may not be importing guest state which clears
8476	* it, so cover for it here.
8477	*
8478	* See Intel spec. 24.11.1 "Software Use of Virtual-Machine Control Structures".
8479	*/
8480	if ( pVmcsInfo->pvShadowVmcs
8481	&& pVmcsInfo->fShadowVmcsState != VMX_V_VMCS_LAUNCH_STATE_CLEAR)
8482	{
8483	rc = hmR0VmxClearShadowVmcs(pVmcsInfo);
8484	AssertRCReturn(rc, rc);
8485	}
8486
8487	/*
8488	* Flag that we need to re-export the host state if we switch to this VMCS before
8489	* executing guest or nested-guest code.
8490	*/
8491	pVmcsInfo->idHostCpuState = NIL_RTCPUID;
8492	#endif
8493
8494	Log4Func(("Cleared Vmcs. HostCpuId=%u\n", idCpu));
8495	NOREF(idCpu);
8496	return VINF_SUCCESS;
8497	}
8498
8499
8500	/**
8501	* Leaves the VT-x session.
8502	*
8503	* @returns VBox status code.
8504	* @param pVCpu The cross context virtual CPU structure.
8505	*
8506	* @remarks No-long-jmp zone!!!
8507	*/
8508	static int hmR0VmxLeaveSession(PVMCPUCC pVCpu)
8509	{
8510	HM_DISABLE_PREEMPT(pVCpu);
8511	HMVMX_ASSERT_CPU_SAFE(pVCpu);
8512	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8513	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8514
8515	/* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before
8516	and done this from the VMXR0ThreadCtxCallback(). */
8517	if (!pVCpu->hmr0.s.fLeaveDone)
8518	{
8519	int rc2 = hmR0VmxLeave(pVCpu, true /* fImportState */);
8520	AssertRCReturnStmt(rc2, HM_RESTORE_PREEMPT(), rc2);
8521	pVCpu->hmr0.s.fLeaveDone = true;
8522	}
8523	Assert(!pVCpu->cpum.GstCtx.fExtrn);
8524
8525	/*
8526	* !!! IMPORTANT !!!
8527	* If you modify code here, make sure to check whether VMXR0CallRing3Callback() needs to be updated too.
8528	*/
8529
8530	/* Deregister hook now that we've left HM context before re-enabling preemption. */
8531	/** @todo Deregistering here means we need to VMCLEAR always
8532	* (longjmp/exit-to-r3) in VT-x which is not efficient, eliminate need
8533	* for calling VMMR0ThreadCtxHookDisable here! */
8534	VMMR0ThreadCtxHookDisable(pVCpu);
8535
8536	/* Leave HM context. This takes care of local init (term) and deregistering the longjmp-to-ring-3 callback. */
8537	int rc = HMR0LeaveCpu(pVCpu);
8538	HM_RESTORE_PREEMPT();
8539	return rc;
8540	}
8541
8542
8543	/**
8544	* Does the necessary state syncing before doing a longjmp to ring-3.
8545	*
8546	* @returns VBox status code.
8547	* @param pVCpu The cross context virtual CPU structure.
8548	*
8549	* @remarks No-long-jmp zone!!!
8550	*/
8551	DECLINLINE(int) hmR0VmxLongJmpToRing3(PVMCPUCC pVCpu)
8552	{
8553	return hmR0VmxLeaveSession(pVCpu);
8554	}
8555
8556
8557	/**
8558	* Take necessary actions before going back to ring-3.
8559	*
8560	* An action requires us to go back to ring-3. This function does the necessary
8561	* steps before we can safely return to ring-3. This is not the same as longjmps
8562	* to ring-3, this is voluntary and prepares the guest so it may continue
8563	* executing outside HM (recompiler/IEM).
8564	*
8565	* @returns VBox status code.
8566	* @param pVCpu The cross context virtual CPU structure.
8567	* @param rcExit The reason for exiting to ring-3. Can be
8568	* VINF_VMM_UNKNOWN_RING3_CALL.
8569	*/
8570	static int hmR0VmxExitToRing3(PVMCPUCC pVCpu, VBOXSTRICTRC rcExit)
8571	{
8572	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
8573
8574	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8575	if (RT_UNLIKELY(rcExit == VERR_VMX_INVALID_VMCS_PTR))
8576	{
8577	VMXGetCurrentVmcs(&pVCpu->hm.s.vmx.LastError.HCPhysCurrentVmcs);
8578	pVCpu->hm.s.vmx.LastError.u32VmcsRev = (uint32_t )pVmcsInfo->pvVmcs;
8579	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hmr0.s.idEnteredCpu;
8580	/* LastError.idCurrentCpu was updated in hmR0VmxPreRunGuestCommitted(). */
8581	}
8582
8583	/* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */
8584	VMMRZCallRing3Disable(pVCpu);
8585	Log4Func(("rcExit=%d\n", VBOXSTRICTRC_VAL(rcExit)));
8586
8587	/*
8588	* Convert any pending HM events back to TRPM due to premature exits to ring-3.
8589	* We need to do this only on returns to ring-3 and not for longjmps to ring3.
8590	*
8591	* This is because execution may continue from ring-3 and we would need to inject
8592	* the event from there (hence place it back in TRPM).
8593	*/
8594	if (pVCpu->hm.s.Event.fPending)
8595	{
8596	hmR0VmxPendingEventToTrpmTrap(pVCpu);
8597	Assert(!pVCpu->hm.s.Event.fPending);
8598
8599	/* Clear the events from the VMCS. */
8600	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, 0); AssertRC(rc);
8601	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, 0); AssertRC(rc);
8602	}
8603	#ifdef VBOX_STRICT
8604	/*
8605	* We check for rcExit here since for errors like VERR_VMX_UNABLE_TO_START_VM (which are
8606	* fatal), we don't care about verifying duplicate injection of events. Errors like
8607	* VERR_EM_INTERPRET are converted to their VINF_* counterparts -prior- to calling this
8608	* function so those should and will be checked below.
8609	*/
8610	else if (RT_SUCCESS(rcExit))
8611	{
8612	/*
8613	* Ensure we don't accidentally clear a pending HM event without clearing the VMCS.
8614	* This can be pretty hard to debug otherwise, interrupts might get injected twice
8615	* occasionally, see @bugref{9180#c42}.
8616	*
8617	* However, if the VM-entry failed, any VM entry-interruption info. field would
8618	* be left unmodified as the event would not have been injected to the guest. In
8619	* such cases, don't assert, we're not going to continue guest execution anyway.
8620	*/
8621	uint32_t uExitReason;
8622	uint32_t uEntryIntInfo;
8623	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
8624	rc \|= VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &uEntryIntInfo);
8625	AssertRC(rc);
8626	AssertMsg(VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason) \|\| !VMX_ENTRY_INT_INFO_IS_VALID(uEntryIntInfo),
8627	("uExitReason=%#RX32 uEntryIntInfo=%#RX32 rcExit=%d\n", uExitReason, uEntryIntInfo, VBOXSTRICTRC_VAL(rcExit)));
8628	}
8629	#endif
8630
8631	/*
8632	* Clear the interrupt-window and NMI-window VMCS controls as we could have got
8633	* a VM-exit with higher priority than interrupt-window or NMI-window VM-exits
8634	* (e.g. TPR below threshold).
8635	*/
8636	if (!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
8637	{
8638	hmR0VmxClearIntWindowExitVmcs(pVmcsInfo);
8639	hmR0VmxClearNmiWindowExitVmcs(pVmcsInfo);
8640	}
8641
8642	/* If we're emulating an instruction, we shouldn't have any TRPM traps pending
8643	and if we're injecting an event we should have a TRPM trap pending. */
8644	AssertMsg(rcExit != VINF_EM_RAW_INJECT_TRPM_EVENT \|\| TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
8645	#ifndef DEBUG_bird /* Triggered after firing an NMI against NT4SP1, possibly a triple fault in progress. */
8646	AssertMsg(rcExit != VINF_EM_RAW_EMULATE_INSTR \|\| !TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
8647	#endif
8648
8649	/* Save guest state and restore host state bits. */
8650	int rc = hmR0VmxLeaveSession(pVCpu);
8651	AssertRCReturn(rc, rc);
8652	STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
8653
8654	/* Thread-context hooks are unregistered at this point!!! */
8655	/* Ring-3 callback notifications are unregistered at this point!!! */
8656
8657	/* Sync recompiler state. */
8658	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3);
8659	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR
8660	\| CPUM_CHANGED_LDTR
8661	\| CPUM_CHANGED_GDTR
8662	\| CPUM_CHANGED_IDTR
8663	\| CPUM_CHANGED_TR
8664	\| CPUM_CHANGED_HIDDEN_SEL_REGS);
8665	if ( pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging
8666	&& CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx))
8667	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH);
8668
8669	Assert(!pVCpu->hmr0.s.fClearTrapFlag);
8670
8671	/* Update the exit-to-ring 3 reason. */
8672	pVCpu->hm.s.rcLastExitToR3 = VBOXSTRICTRC_VAL(rcExit);
8673
8674	/* On our way back from ring-3 reload the guest state if there is a possibility of it being changed. */
8675	if ( rcExit != VINF_EM_RAW_INTERRUPT
8676	\|\| CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
8677	{
8678	Assert(!(pVCpu->cpum.GstCtx.fExtrn & HMVMX_CPUMCTX_EXTRN_ALL));
8679	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
8680	}
8681
8682	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3);
8683	VMMRZCallRing3Enable(pVCpu);
8684	return rc;
8685	}
8686
8687
8688	/**
8689	* VMMRZCallRing3() callback wrapper which saves the guest state before we
8690	* longjump to ring-3 and possibly get preempted.
8691	*
8692	* @returns VBox status code.
8693	* @param pVCpu The cross context virtual CPU structure.
8694	* @param enmOperation The operation causing the ring-3 longjump.
8695	*/
8696	VMMR0DECL(int) VMXR0CallRing3Callback(PVMCPUCC pVCpu, VMMCALLRING3 enmOperation)
8697	{
8698	if (enmOperation == VMMCALLRING3_VM_R0_ASSERTION)
8699	{
8700	/*
8701	* !!! IMPORTANT !!!
8702	* If you modify code here, check whether hmR0VmxLeave() and hmR0VmxLeaveSession() needs to be updated too.
8703	* This is a stripped down version which gets out ASAP, trying to not trigger any further assertions.
8704	*/
8705	VMMRZCallRing3RemoveNotification(pVCpu);
8706	VMMRZCallRing3Disable(pVCpu);
8707	HM_DISABLE_PREEMPT(pVCpu);
8708
8709	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
8710	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
8711	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
8712	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
8713
8714	/* Restore host-state bits that VT-x only restores partially. */
8715	if (pVCpu->hmr0.s.vmx.fRestoreHostFlags > VMX_RESTORE_HOST_REQUIRED)
8716	VMXRestoreHostState(pVCpu->hmr0.s.vmx.fRestoreHostFlags, &pVCpu->hmr0.s.vmx.RestoreHost);
8717	pVCpu->hmr0.s.vmx.fRestoreHostFlags = 0;
8718
8719	/* Restore the lazy host MSRs as we're leaving VT-x context. */
8720	if (pVCpu->hmr0.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
8721	hmR0VmxLazyRestoreHostMsrs(pVCpu);
8722
8723	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
8724	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = false;
8725	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
8726
8727	/* Clear the current VMCS data back to memory (shadow VMCS if any would have been
8728	cleared as part of importing the guest state above. */
8729	hmR0VmxClearVmcs(pVmcsInfo);
8730
8731	/** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
8732	VMMR0ThreadCtxHookDisable(pVCpu);
8733
8734	/* Leave HM context. This takes care of local init (term). */
8735	HMR0LeaveCpu(pVCpu);
8736	HM_RESTORE_PREEMPT();
8737	return VINF_SUCCESS;
8738	}
8739
8740	Assert(pVCpu);
8741	Assert(VMMRZCallRing3IsEnabled(pVCpu));
8742	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
8743
8744	VMMRZCallRing3Disable(pVCpu);
8745
8746	Log4Func(("-> hmR0VmxLongJmpToRing3 enmOperation=%d\n", enmOperation));
8747
8748	int rc = hmR0VmxLongJmpToRing3(pVCpu);
8749	AssertRCReturn(rc, rc);
8750
8751	VMMRZCallRing3Enable(pVCpu);
8752	return VINF_SUCCESS;
8753	}
8754
8755
8756	/**
8757	* Pushes a 2-byte value onto the real-mode (in virtual-8086 mode) guest's
8758	* stack.
8759	*
8760	* @returns Strict VBox status code (i.e. informational status codes too).
8761	* @retval VINF_EM_RESET if pushing a value to the stack caused a triple-fault.
8762	* @param pVCpu The cross context virtual CPU structure.
8763	* @param uValue The value to push to the guest stack.
8764	*/
8765	static VBOXSTRICTRC hmR0VmxRealModeGuestStackPush(PVMCPUCC pVCpu, uint16_t uValue)
8766	{
8767	/*
8768	* The stack limit is 0xffff in real-on-virtual 8086 mode. Real-mode with weird stack limits cannot be run in
8769	* virtual 8086 mode in VT-x. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
8770	* See Intel Instruction reference for PUSH and Intel spec. 22.33.1 "Segment Wraparound".
8771	*/
8772	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8773	if (pCtx->sp == 1)
8774	return VINF_EM_RESET;
8775	pCtx->sp -= sizeof(uint16_t); /* May wrap around which is expected behaviour. */
8776	int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), pCtx->ss.u64Base + pCtx->sp, &uValue, sizeof(uint16_t));
8777	AssertRC(rc);
8778	return rc;
8779	}
8780
8781
8782	/**
8783	* Injects an event into the guest upon VM-entry by updating the relevant fields
8784	* in the VM-entry area in the VMCS.
8785	*
8786	* @returns Strict VBox status code (i.e. informational status codes too).
8787	* @retval VINF_SUCCESS if the event is successfully injected into the VMCS.
8788	* @retval VINF_EM_RESET if event injection resulted in a triple-fault.
8789	*
8790	* @param pVCpu The cross context virtual CPU structure.
8791	* @param pVmxTransient The VMX-transient structure.
8792	* @param pEvent The event being injected.
8793	* @param pfIntrState Pointer to the VT-x guest-interruptibility-state. This
8794	* will be updated if necessary. This cannot not be NULL.
8795	* @param fStepping Whether we're single-stepping guest execution and should
8796	* return VINF_EM_DBG_STEPPED if the event is injected
8797	* directly (registers modified by us, not by hardware on
8798	* VM-entry).
8799	*/
8800	static VBOXSTRICTRC hmR0VmxInjectEventVmcs(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, PCHMEVENT pEvent, bool fStepping,
8801	uint32_t *pfIntrState)
8802	{
8803	/* Intel spec. 24.8.3 "VM-Entry Controls for Event Injection" specifies the interruption-information field to be 32-bits. */
8804	AssertMsg(!RT_HI_U32(pEvent->u64IntInfo), ("%#RX64\n", pEvent->u64IntInfo));
8805	Assert(pfIntrState);
8806
8807	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8808	uint32_t u32IntInfo = pEvent->u64IntInfo;
8809	uint32_t const u32ErrCode = pEvent->u32ErrCode;
8810	uint32_t const cbInstr = pEvent->cbInstr;
8811	RTGCUINTPTR const GCPtrFault = pEvent->GCPtrFaultAddress;
8812	uint8_t const uVector = VMX_ENTRY_INT_INFO_VECTOR(u32IntInfo);
8813	uint32_t const uIntType = VMX_ENTRY_INT_INFO_TYPE(u32IntInfo);
8814
8815	#ifdef VBOX_STRICT
8816	/*
8817	* Validate the error-code-valid bit for hardware exceptions.
8818	* No error codes for exceptions in real-mode.
8819	*
8820	* See Intel spec. 20.1.4 "Interrupt and Exception Handling"
8821	*/
8822	if ( uIntType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT
8823	&& !CPUMIsGuestInRealModeEx(pCtx))
8824	{
8825	switch (uVector)
8826	{
8827	case X86_XCPT_PF:
8828	case X86_XCPT_DF:
8829	case X86_XCPT_TS:
8830	case X86_XCPT_NP:
8831	case X86_XCPT_SS:
8832	case X86_XCPT_GP:
8833	case X86_XCPT_AC:
8834	AssertMsg(VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(u32IntInfo),
8835	("Error-code-valid bit not set for exception that has an error code uVector=%#x\n", uVector));
8836	RT_FALL_THRU();
8837	default:
8838	break;
8839	}
8840	}
8841
8842	/* Cannot inject an NMI when block-by-MOV SS is in effect. */
8843	Assert( uIntType != VMX_EXIT_INT_INFO_TYPE_NMI
8844	\|\| !(*pfIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
8845	#endif
8846
8847	if ( uIntType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT
8848	\|\| uIntType == VMX_EXIT_INT_INFO_TYPE_NMI
8849	\|\| uIntType == VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT
8850	\|\| uIntType == VMX_EXIT_INT_INFO_TYPE_SW_XCPT)
8851	{
8852	Assert(uVector <= X86_XCPT_LAST);
8853	Assert(uIntType != VMX_EXIT_INT_INFO_TYPE_NMI \|\| uVector == X86_XCPT_NMI);
8854	Assert(uIntType != VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT \|\| uVector == X86_XCPT_DB);
8855	STAM_COUNTER_INC(&pVCpu->hm.s.aStatInjectedXcpts[uVector]);
8856	}
8857	else
8858	STAM_COUNTER_INC(&pVCpu->hm.s.aStatInjectedIrqs[uVector & MASK_INJECT_IRQ_STAT]);
8859
8860	/*
8861	* Hardware interrupts & exceptions cannot be delivered through the software interrupt
8862	* redirection bitmap to the real mode task in virtual-8086 mode. We must jump to the
8863	* interrupt handler in the (real-mode) guest.
8864	*
8865	* See Intel spec. 20.3 "Interrupt and Exception handling in Virtual-8086 Mode".
8866	* See Intel spec. 20.1.4 "Interrupt and Exception Handling" for real-mode interrupt handling.
8867	*/
8868	if (CPUMIsGuestInRealModeEx(pCtx)) /* CR0.PE bit changes are always intercepted, so it's up to date. */
8869	{
8870	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUnrestrictedGuest)
8871	{
8872	/*
8873	* For CPUs with unrestricted guest execution enabled and with the guest
8874	* in real-mode, we must not set the deliver-error-code bit.
8875	*
8876	* See Intel spec. 26.2.1.3 "VM-Entry Control Fields".
8877	*/
8878	u32IntInfo &= ~VMX_ENTRY_INT_INFO_ERROR_CODE_VALID;
8879	}
8880	else
8881	{
8882	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
8883	Assert(PDMVmmDevHeapIsEnabled(pVM));
8884	Assert(pVM->hm.s.vmx.pRealModeTSS);
8885	Assert(!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
8886
8887	/* We require RIP, RSP, RFLAGS, CS, IDTR, import them. */
8888	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
8889	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_TABLE_MASK
8890	\| CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_RFLAGS);
8891	AssertRCReturn(rc2, rc2);
8892
8893	/* Check if the interrupt handler is present in the IVT (real-mode IDT). IDT limit is (4N - 1). */
8894	size_t const cbIdtEntry = sizeof(X86IDTR16);
8895	if (uVector * cbIdtEntry + (cbIdtEntry - 1) > pCtx->idtr.cbIdt)
8896	{
8897	/* If we are trying to inject a #DF with no valid IDT entry, return a triple-fault. */
8898	if (uVector == X86_XCPT_DF)
8899	return VINF_EM_RESET;
8900
8901	/* If we're injecting a #GP with no valid IDT entry, inject a double-fault.
8902	No error codes for exceptions in real-mode. */
8903	if (uVector == X86_XCPT_GP)
8904	{
8905	uint32_t const uXcptDfInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DF)
8906	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
8907	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
8908	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
8909	HMEVENT EventXcptDf;
8910	RT_ZERO(EventXcptDf);
8911	EventXcptDf.u64IntInfo = uXcptDfInfo;
8912	return hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &EventXcptDf, fStepping, pfIntrState);
8913	}
8914
8915	/*
8916	* If we're injecting an event with no valid IDT entry, inject a #GP.
8917	* No error codes for exceptions in real-mode.
8918	*
8919	* See Intel spec. 20.1.4 "Interrupt and Exception Handling"
8920	*/
8921	uint32_t const uXcptGpInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_GP)
8922	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
8923	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID, 0)
8924	\| RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
8925	HMEVENT EventXcptGp;
8926	RT_ZERO(EventXcptGp);
8927	EventXcptGp.u64IntInfo = uXcptGpInfo;
8928	return hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &EventXcptGp, fStepping, pfIntrState);
8929	}
8930
8931	/* Software exceptions (#BP and #OF exceptions thrown as a result of INT3 or INTO) */
8932	uint16_t uGuestIp = pCtx->ip;
8933	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_SW_XCPT)
8934	{
8935	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF);
8936	/* #BP and #OF are both benign traps, we need to resume the next instruction. */
8937	uGuestIp = pCtx->ip + (uint16_t)cbInstr;
8938	}
8939	else if (uIntType == VMX_ENTRY_INT_INFO_TYPE_SW_INT)
8940	uGuestIp = pCtx->ip + (uint16_t)cbInstr;
8941
8942	/* Get the code segment selector and offset from the IDT entry for the interrupt handler. */
8943	X86IDTR16 IdtEntry;
8944	RTGCPHYS const GCPhysIdtEntry = (RTGCPHYS)pCtx->idtr.pIdt + uVector * cbIdtEntry;
8945	rc2 = PGMPhysSimpleReadGCPhys(pVM, &IdtEntry, GCPhysIdtEntry, cbIdtEntry);
8946	AssertRCReturn(rc2, rc2);
8947
8948	/* Construct the stack frame for the interrupt/exception handler. */
8949	VBOXSTRICTRC rcStrict;
8950	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, pCtx->eflags.u32);
8951	if (rcStrict == VINF_SUCCESS)
8952	{
8953	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, pCtx->cs.Sel);
8954	if (rcStrict == VINF_SUCCESS)
8955	rcStrict = hmR0VmxRealModeGuestStackPush(pVCpu, uGuestIp);
8956	}
8957
8958	/* Clear the required eflag bits and jump to the interrupt/exception handler. */
8959	if (rcStrict == VINF_SUCCESS)
8960	{
8961	pCtx->eflags.u32 &= ~(X86_EFL_IF \| X86_EFL_TF \| X86_EFL_RF \| X86_EFL_AC);
8962	pCtx->rip = IdtEntry.offSel;
8963	pCtx->cs.Sel = IdtEntry.uSel;
8964	pCtx->cs.ValidSel = IdtEntry.uSel;
8965	pCtx->cs.u64Base = IdtEntry.uSel << cbIdtEntry;
8966	if ( uIntType == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT
8967	&& uVector == X86_XCPT_PF)
8968	pCtx->cr2 = GCPtrFault;
8969
8970	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CS \| HM_CHANGED_GUEST_CR2
8971	\| HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
8972	\| HM_CHANGED_GUEST_RSP);
8973
8974	/*
8975	* If we delivered a hardware exception (other than an NMI) and if there was
8976	* block-by-STI in effect, we should clear it.
8977	*/
8978	if (*pfIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
8979	{
8980	Assert( uIntType != VMX_ENTRY_INT_INFO_TYPE_NMI
8981	&& uIntType != VMX_ENTRY_INT_INFO_TYPE_EXT_INT);
8982	Log4Func(("Clearing inhibition due to STI\n"));
8983	*pfIntrState &= ~VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
8984	}
8985
8986	Log4(("Injected real-mode: u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x Eflags=%#x CS:EIP=%04x:%04x\n",
8987	u32IntInfo, u32ErrCode, cbInstr, pCtx->eflags.u, pCtx->cs.Sel, pCtx->eip));
8988
8989	/*
8990	* The event has been truly dispatched to the guest. Mark it as no longer pending so
8991	* we don't attempt to undo it if we are returning to ring-3 before executing guest code.
8992	*/
8993	pVCpu->hm.s.Event.fPending = false;
8994
8995	/*
8996	* If we eventually support nested-guest execution without unrestricted guest execution,
8997	* we should set fInterceptEvents here.
8998	*/
8999	Assert(!pVmxTransient->fIsNestedGuest);
9000
9001	/* If we're stepping and we've changed cs:rip above, bail out of the VMX R0 execution loop. */
9002	if (fStepping)
9003	rcStrict = VINF_EM_DBG_STEPPED;
9004	}
9005	AssertMsg(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
9006	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
9007	return rcStrict;
9008	}
9009	}
9010
9011	/*
9012	* Validate.
9013	*/
9014	Assert(VMX_ENTRY_INT_INFO_IS_VALID(u32IntInfo)); /* Bit 31 (Valid bit) must be set by caller. */
9015	Assert(!(u32IntInfo & VMX_BF_ENTRY_INT_INFO_RSVD_12_30_MASK)); /* Bits 30:12 MBZ. */
9016
9017	/*
9018	* Inject the event into the VMCS.
9019	*/
9020	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, u32IntInfo);
9021	if (VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(u32IntInfo))
9022	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, u32ErrCode);
9023	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, cbInstr);
9024	AssertRC(rc);
9025
9026	/*
9027	* Update guest CR2 if this is a page-fault.
9028	*/
9029	if (VMX_ENTRY_INT_INFO_IS_XCPT_PF(u32IntInfo))
9030	pCtx->cr2 = GCPtrFault;
9031
9032	Log4(("Injecting u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x CR2=%#RX64\n", u32IntInfo, u32ErrCode, cbInstr, pCtx->cr2));
9033	return VINF_SUCCESS;
9034	}
9035
9036
9037	/**
9038	* Evaluates the event to be delivered to the guest and sets it as the pending
9039	* event.
9040	*
9041	* Toggling of interrupt force-flags here is safe since we update TRPM on premature
9042	* exits to ring-3 before executing guest code, see hmR0VmxExitToRing3(). We must
9043	* NOT restore these force-flags.
9044	*
9045	* @returns Strict VBox status code (i.e. informational status codes too).
9046	* @param pVCpu The cross context virtual CPU structure.
9047	* @param pVmxTransient The VMX-transient structure.
9048	* @param pfIntrState Where to store the VT-x guest-interruptibility state.
9049	*/
9050	static VBOXSTRICTRC hmR0VmxEvaluatePendingEvent(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t *pfIntrState)
9051	{
9052	Assert(pfIntrState);
9053	Assert(!TRPMHasTrap(pVCpu));
9054
9055	/*
9056	* Compute/update guest-interruptibility state related FFs.
9057	* The FFs will be used below while evaluating events to be injected.
9058	*/
9059	*pfIntrState = hmR0VmxGetGuestIntrStateAndUpdateFFs(pVCpu);
9060
9061	/*
9062	* Evaluate if a new event needs to be injected.
9063	* An event that's already pending has already performed all necessary checks.
9064	*/
9065	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
9066	bool const fIsNestedGuest = pVmxTransient->fIsNestedGuest;
9067	if ( !pVCpu->hm.s.Event.fPending
9068	&& !VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
9069	{
9070	/** @todo SMI. SMIs take priority over NMIs. */
9071
9072	/*
9073	* NMIs.
9074	* NMIs take priority over external interrupts.
9075	*/
9076	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
9077	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
9078	{
9079	/*
9080	* For a guest, the FF always indicates the guest's ability to receive an NMI.
9081	*
9082	* For a nested-guest, the FF always indicates the outer guest's ability to
9083	* receive an NMI while the guest-interruptibility state bit depends on whether
9084	* the nested-hypervisor is using virtual-NMIs.
9085	*/
9086	if (!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_BLOCK_NMIS))
9087	{
9088	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9089	if ( fIsNestedGuest
9090	&& CPUMIsGuestVmxPinCtlsSet(pCtx, VMX_PIN_CTLS_NMI_EXIT))
9091	return IEMExecVmxVmexitXcptNmi(pVCpu);
9092	#endif
9093	hmR0VmxSetPendingXcptNmi(pVCpu);
9094	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
9095	Log4Func(("NMI pending injection\n"));
9096
9097	/* We've injected the NMI, bail. */
9098	return VINF_SUCCESS;
9099	}
9100	else if (!fIsNestedGuest)
9101	hmR0VmxSetNmiWindowExitVmcs(pVmcsInfo);
9102	}
9103
9104	/*
9105	* External interrupts (PIC/APIC).
9106	* Once PDMGetInterrupt() returns a valid interrupt we -must- deliver it.
9107	* We cannot re-request the interrupt from the controller again.
9108	*/
9109	if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
9110	&& !pVCpu->hm.s.fSingleInstruction)
9111	{
9112	Assert(!DBGFIsStepping(pVCpu));
9113	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_RFLAGS);
9114	AssertRC(rc);
9115
9116	/*
9117	* We must not check EFLAGS directly when executing a nested-guest, use
9118	* CPUMIsGuestPhysIntrEnabled() instead as EFLAGS.IF does not control the blocking of
9119	* external interrupts when "External interrupt exiting" is set. This fixes a nasty
9120	* SMP hang while executing nested-guest VCPUs on spinlocks which aren't rescued by
9121	* other VM-exits (like a preemption timer), see @bugref{9562#c18}.
9122	*
9123	* See Intel spec. 25.4.1 "Event Blocking".
9124	*/
9125	if (CPUMIsGuestPhysIntrEnabled(pVCpu))
9126	{
9127	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9128	if ( fIsNestedGuest
9129	&& CPUMIsGuestVmxPinCtlsSet(pCtx, VMX_PIN_CTLS_EXT_INT_EXIT))
9130	{
9131	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitExtInt(pVCpu, 0 /* uVector /, true / fIntPending */);
9132	if (rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE)
9133	return rcStrict;
9134	}
9135	#endif
9136	uint8_t u8Interrupt;
9137	rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
9138	if (RT_SUCCESS(rc))
9139	{
9140	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
9141	if ( fIsNestedGuest
9142	&& CPUMIsGuestVmxPinCtlsSet(pCtx, VMX_PIN_CTLS_EXT_INT_EXIT))
9143	{
9144	VBOXSTRICTRC rcStrict = IEMExecVmxVmexitExtInt(pVCpu, u8Interrupt, false /* fIntPending */);
9145	Assert(rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE);
9146	return rcStrict;
9147	}
9148	#endif
9149	hmR0VmxSetPendingExtInt(pVCpu, u8Interrupt);
9150	Log4Func(("External interrupt (%#x) pending injection\n", u8Interrupt));
9151	}
9152	else if (rc == VERR_APIC_INTR_MASKED_BY_TPR)
9153	{
9154	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchTprMaskedIrq);
9155
9156	if ( !fIsNestedGuest
9157	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW))
9158	hmR0VmxApicSetTprThreshold(pVmcsInfo, u8Interrupt >> 4);
9159	/* else: for nested-guests, TPR threshold is picked up while merging VMCS controls. */
9160
9161	/*
9162	* If the CPU doesn't have TPR shadowing, we will always get a VM-exit on TPR changes and
9163	* APICSetTpr() will end up setting the VMCPU_FF_INTERRUPT_APIC if required, so there is no
9164	* need to re-set this force-flag here.
9165	*/
9166	}
9167	else
9168	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
9169
9170	/* We've injected the interrupt or taken necessary action, bail. */
9171	return VINF_SUCCESS;
9172	}
9173	if (!fIsNestedGuest)
9174	hmR0VmxSetIntWindowExitVmcs(pVmcsInfo);
9175	}
9176	}
9177	else if (!fIsNestedGuest)
9178	{
9179	/*
9180	* An event is being injected or we are in an interrupt shadow. Check if another event is
9181	* pending. If so, instruct VT-x to cause a VM-exit as soon as the guest is ready to accept
9182	* the pending event.
9183	*/
9184	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
9185	hmR0VmxSetNmiWindowExitVmcs(pVmcsInfo);
9186	else if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
9187	&& !pVCpu->hm.s.fSingleInstruction)
9188	hmR0VmxSetIntWindowExitVmcs(pVmcsInfo);
9189	}
9190	/* else: for nested-guests, NMI/interrupt-window exiting will be picked up when merging VMCS controls. */
9191
9192	return VINF_SUCCESS;
9193	}
9194
9195
9196	/**
9197	* Injects any pending events into the guest if the guest is in a state to
9198	* receive them.
9199	*
9200	* @returns Strict VBox status code (i.e. informational status codes too).
9201	* @param pVCpu The cross context virtual CPU structure.
9202	* @param pVmxTransient The VMX-transient structure.
9203	* @param fIntrState The VT-x guest-interruptibility state.
9204	* @param fStepping Whether we are single-stepping the guest using the
9205	* hypervisor debugger and should return
9206	* VINF_EM_DBG_STEPPED if the event was dispatched
9207	* directly.
9208	*/
9209	static VBOXSTRICTRC hmR0VmxInjectPendingEvent(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, uint32_t fIntrState, bool fStepping)
9210	{
9211	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9212	Assert(VMMRZCallRing3IsEnabled(pVCpu));
9213
9214	#ifdef VBOX_STRICT
9215	/*
9216	* Verify guest-interruptibility state.
9217	*
9218	* We put this in a scoped block so we do not accidentally use fBlockSti or fBlockMovSS,
9219	* since injecting an event may modify the interruptibility state and we must thus always
9220	* use fIntrState.
9221	*/
9222	{
9223	bool const fBlockMovSS = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS);
9224	bool const fBlockSti = RT_BOOL(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI);
9225	Assert(!fBlockSti \|\| !(ASMAtomicUoReadU64(&pVCpu->cpum.GstCtx.fExtrn) & CPUMCTX_EXTRN_RFLAGS));
9226	Assert(!fBlockSti \|\| pVCpu->cpum.GstCtx.eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
9227	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
9228	Assert(!TRPMHasTrap(pVCpu));
9229	NOREF(fBlockMovSS); NOREF(fBlockSti);
9230	}
9231	#endif
9232
9233	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
9234	if (pVCpu->hm.s.Event.fPending)
9235	{
9236	/*
9237	* Do -not- clear any interrupt-window exiting control here. We might have an interrupt
9238	* pending even while injecting an event and in this case, we want a VM-exit as soon as
9239	* the guest is ready for the next interrupt, see @bugref{6208#c45}.
9240	*
9241	* See Intel spec. 26.6.5 "Interrupt-Window Exiting and Virtual-Interrupt Delivery".
9242	*/
9243	uint32_t const uIntType = VMX_ENTRY_INT_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
9244	#ifdef VBOX_STRICT
9245	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
9246	{
9247	Assert(pVCpu->cpum.GstCtx.eflags.u32 & X86_EFL_IF);
9248	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI));
9249	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
9250	}
9251	else if (uIntType == VMX_ENTRY_INT_INFO_TYPE_NMI)
9252	{
9253	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI));
9254	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI));
9255	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
9256	}
9257	#endif
9258	Log4(("Injecting pending event vcpu[%RU32] u64IntInfo=%#RX64 Type=%#RX32\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo,
9259	uIntType));
9260
9261	/*
9262	* Inject the event and get any changes to the guest-interruptibility state.
9263	*
9264	* The guest-interruptibility state may need to be updated if we inject the event
9265	* into the guest IDT ourselves (for real-on-v86 guest injecting software interrupts).
9266	*/
9267	rcStrict = hmR0VmxInjectEventVmcs(pVCpu, pVmxTransient, &pVCpu->hm.s.Event, fStepping, &fIntrState);
9268	AssertRCReturn(VBOXSTRICTRC_VAL(rcStrict), rcStrict);
9269
9270	if (uIntType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
9271	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt);
9272	else
9273	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt);
9274	}
9275
9276	/*
9277	* Deliver any pending debug exceptions if the guest is single-stepping using EFLAGS.TF and
9278	* is an interrupt shadow (block-by-STI or block-by-MOV SS).
9279	*/
9280	if ( (fIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
9281	&& !pVmxTransient->fIsNestedGuest)
9282	{
9283	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS);
9284
9285	if (!pVCpu->hm.s.fSingleInstruction)
9286	{
9287	/*
9288	* Set or clear the BS bit depending on whether the trap flag is active or not. We need
9289	* to do both since we clear the BS bit from the VMCS while exiting to ring-3.
9290	*/
9291	Assert(!DBGFIsStepping(pVCpu));
9292	uint8_t const fTrapFlag = !!(pVCpu->cpum.GstCtx.eflags.u32 & X86_EFL_TF);
9293	int rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, fTrapFlag << VMX_BF_VMCS_PENDING_DBG_XCPT_BS_SHIFT);
9294	AssertRC(rc);
9295	}
9296	else
9297	{
9298	/*
9299	* We must not deliver a debug exception when single-stepping over STI/Mov-SS in the
9300	* hypervisor debugger using EFLAGS.TF but rather clear interrupt inhibition. However,
9301	* we take care of this case in hmR0VmxExportSharedDebugState and also the case if
9302	* we use MTF, so just make sure it's called before executing guest-code.
9303	*/
9304	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR_MASK);
9305	}
9306	}
9307	/* else: for nested-guest currently handling while merging controls. */
9308
9309	/*
9310	* Finally, update the guest-interruptibility state.
9311	*
9312	* This is required for the real-on-v86 software interrupt injection, for
9313	* pending debug exceptions as well as updates to the guest state from ring-3 (IEM).
9314	*/
9315	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
9316	AssertRC(rc);
9317
9318	/*
9319	* There's no need to clear the VM-entry interruption-information field here if we're not
9320	* injecting anything. VT-x clears the valid bit on every VM-exit.
9321	*
9322	* See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
9323	*/
9324
9325	Assert(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping));
9326	return rcStrict;
9327	}
9328
9329
9330	/**
9331	* Enters the VT-x session.
9332	*
9333	* @returns VBox status code.
9334	* @param pVCpu The cross context virtual CPU structure.
9335	*/
9336	VMMR0DECL(int) VMXR0Enter(PVMCPUCC pVCpu)
9337	{
9338	AssertPtr(pVCpu);
9339	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fSupported);
9340	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9341
9342	LogFlowFunc(("pVCpu=%p\n", pVCpu));
9343	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
9344	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
9345
9346	#ifdef VBOX_STRICT
9347	/* At least verify VMX is enabled, since we can't check if we're in VMX root mode without #GP'ing. */
9348	RTCCUINTREG uHostCr4 = ASMGetCR4();
9349	if (!(uHostCr4 & X86_CR4_VMXE))
9350	{
9351	LogRelFunc(("X86_CR4_VMXE bit in CR4 is not set!\n"));
9352	return VERR_VMX_X86_CR4_VMXE_CLEARED;
9353	}
9354	#endif
9355
9356	/*
9357	* Do the EMT scheduled L1D and MDS flush here if needed.
9358	*/
9359	if (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_L1D_SCHED)
9360	ASMWrMsr(MSR_IA32_FLUSH_CMD, MSR_IA32_FLUSH_CMD_F_L1D);
9361	else if (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_MDS_SCHED)
9362	hmR0MdsClear();
9363
9364	/*
9365	* Load the appropriate VMCS as the current and active one.
9366	*/
9367	PVMXVMCSINFO pVmcsInfo;
9368	bool const fInNestedGuestMode = CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx);
9369	if (!fInNestedGuestMode)
9370	pVmcsInfo = &pVCpu->hmr0.s.vmx.VmcsInfo;
9371	else
9372	pVmcsInfo = &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
9373	int rc = hmR0VmxLoadVmcs(pVmcsInfo);
9374	if (RT_SUCCESS(rc))
9375	{
9376	pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs = fInNestedGuestMode;
9377	pVCpu->hm.s.vmx.fSwitchedToNstGstVmcsCopyForRing3 = fInNestedGuestMode;
9378	pVCpu->hmr0.s.fLeaveDone = false;
9379	Log4Func(("Loaded Vmcs. HostCpuId=%u\n", RTMpCpuId()));
9380	}
9381	return rc;
9382	}
9383
9384
9385	/**
9386	* The thread-context callback.
9387	*
9388	* This is used together with RTThreadCtxHookCreate() on platforms which
9389	* supports it, and directly from VMMR0EmtPrepareForBlocking() and
9390	* VMMR0EmtResumeAfterBlocking() on platforms which don't.
9391	*
9392	* @param enmEvent The thread-context event.
9393	* @param pVCpu The cross context virtual CPU structure.
9394	* @param fGlobalInit Whether global VT-x/AMD-V init. was used.
9395	* @thread EMT(pVCpu)
9396	*/
9397	VMMR0DECL(void) VMXR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPUCC pVCpu, bool fGlobalInit)
9398	{
9399	AssertPtr(pVCpu);
9400	RT_NOREF1(fGlobalInit);
9401
9402	switch (enmEvent)
9403	{
9404	case RTTHREADCTXEVENT_OUT:
9405	{
9406	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9407	VMCPU_ASSERT_EMT(pVCpu);
9408
9409	/* No longjmps (logger flushes, locks) in this fragile context. */
9410	VMMRZCallRing3Disable(pVCpu);
9411	Log4Func(("Preempting: HostCpuId=%u\n", RTMpCpuId()));
9412
9413	/* Restore host-state (FPU, debug etc.) */
9414	if (!pVCpu->hmr0.s.fLeaveDone)
9415	{
9416	/*
9417	* Do -not- import the guest-state here as we might already be in the middle of importing
9418	* it, esp. bad if we're holding the PGM lock, see comment in hmR0VmxImportGuestState().
9419	*/
9420	hmR0VmxLeave(pVCpu, false /* fImportState */);
9421	pVCpu->hmr0.s.fLeaveDone = true;
9422	}
9423
9424	/* Leave HM context, takes care of local init (term). */
9425	int rc = HMR0LeaveCpu(pVCpu);
9426	AssertRC(rc);
9427
9428	/* Restore longjmp state. */
9429	VMMRZCallRing3Enable(pVCpu);
9430	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreempt);
9431	break;
9432	}
9433
9434	case RTTHREADCTXEVENT_IN:
9435	{
9436	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9437	VMCPU_ASSERT_EMT(pVCpu);
9438
9439	/* Do the EMT scheduled L1D and MDS flush here if needed. */
9440	if (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_L1D_SCHED)
9441	ASMWrMsr(MSR_IA32_FLUSH_CMD, MSR_IA32_FLUSH_CMD_F_L1D);
9442	else if (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_MDS_SCHED)
9443	hmR0MdsClear();
9444
9445	/* No longjmps here, as we don't want to trigger preemption (& its hook) while resuming. */
9446	VMMRZCallRing3Disable(pVCpu);
9447	Log4Func(("Resumed: HostCpuId=%u\n", RTMpCpuId()));
9448
9449	/* Initialize the bare minimum state required for HM. This takes care of
9450	initializing VT-x if necessary (onlined CPUs, local init etc.) */
9451	int rc = hmR0EnterCpu(pVCpu);
9452	AssertRC(rc);
9453	Assert( (pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
9454	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
9455
9456	/* Load the active VMCS as the current one. */
9457	PVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
9458	rc = hmR0VmxLoadVmcs(pVmcsInfo);
9459	AssertRC(rc);
9460	Log4Func(("Resumed: Loaded Vmcs. HostCpuId=%u\n", RTMpCpuId()));
9461	pVCpu->hmr0.s.fLeaveDone = false;
9462
9463	/* Restore longjmp state. */
9464	VMMRZCallRing3Enable(pVCpu);
9465	break;
9466	}
9467
9468	default:
9469	break;
9470	}
9471	}
9472
9473
9474	/**
9475	* Exports the host state into the VMCS host-state area.
9476	* Sets up the VM-exit MSR-load area.
9477	*
9478	* The CPU state will be loaded from these fields on every successful VM-exit.
9479	*
9480	* @returns VBox status code.
9481	* @param pVCpu The cross context virtual CPU structure.
9482	*
9483	* @remarks No-long-jump zone!!!
9484	*/
9485	static int hmR0VmxExportHostState(PVMCPUCC pVCpu)
9486	{
9487	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9488
9489	int rc = VINF_SUCCESS;
9490	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT)
9491	{
9492	uint64_t uHostCr4 = hmR0VmxExportHostControlRegs();
9493
9494	rc = hmR0VmxExportHostSegmentRegs(pVCpu, uHostCr4);
9495	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9496
9497	hmR0VmxExportHostMsrs(pVCpu);
9498
9499	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_HOST_CONTEXT;
9500	}
9501	return rc;
9502	}
9503
9504
9505	/**
9506	* Saves the host state in the VMCS host-state.
9507	*
9508	* @returns VBox status code.
9509	* @param pVCpu The cross context virtual CPU structure.
9510	*
9511	* @remarks No-long-jump zone!!!
9512	*/
9513	VMMR0DECL(int) VMXR0ExportHostState(PVMCPUCC pVCpu)
9514	{
9515	AssertPtr(pVCpu);
9516	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9517
9518	/*
9519	* Export the host state here while entering HM context.
9520	* When thread-context hooks are used, we might get preempted and have to re-save the host
9521	* state but most of the time we won't be, so do it here before we disable interrupts.
9522	*/
9523	return hmR0VmxExportHostState(pVCpu);
9524	}
9525
9526
9527	/**
9528	* Exports the guest state into the VMCS guest-state area.
9529	*
9530	* The will typically be done before VM-entry when the guest-CPU state and the
9531	* VMCS state may potentially be out of sync.
9532	*
9533	* Sets up the VM-entry MSR-load and VM-exit MSR-store areas. Sets up the
9534	* VM-entry controls.
9535	* Sets up the appropriate VMX non-root function to execute guest code based on
9536	* the guest CPU mode.
9537	*
9538	* @returns VBox strict status code.
9539	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
9540	* without unrestricted guest execution and the VMMDev is not presently
9541	* mapped (e.g. EFI32).
9542	*
9543	* @param pVCpu The cross context virtual CPU structure.
9544	* @param pVmxTransient The VMX-transient structure.
9545	*
9546	* @remarks No-long-jump zone!!!
9547	*/
9548	static VBOXSTRICTRC hmR0VmxExportGuestState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9549	{
9550	AssertPtr(pVCpu);
9551	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9552	LogFlowFunc(("pVCpu=%p\n", pVCpu));
9553
9554	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExportGuestState, x);
9555
9556	/*
9557	* Determine real-on-v86 mode.
9558	* Used when the guest is in real-mode and unrestricted guest execution is not used.
9559	*/
9560	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmxTransient->pVmcsInfo->pShared;
9561	if ( pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUnrestrictedGuest
9562	\|\| !CPUMIsGuestInRealModeEx(&pVCpu->cpum.GstCtx))
9563	pVmcsInfoShared->RealMode.fRealOnV86Active = false;
9564	else
9565	{
9566	Assert(!pVmxTransient->fIsNestedGuest);
9567	pVmcsInfoShared->RealMode.fRealOnV86Active = true;
9568	}
9569
9570	/*
9571	* Any ordering dependency among the sub-functions below must be explicitly stated using comments.
9572	* Ideally, assert that the cross-dependent bits are up-to-date at the point of using it.
9573	*/
9574	int rc = hmR0VmxExportGuestEntryExitCtls(pVCpu, pVmxTransient);
9575	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9576
9577	rc = hmR0VmxExportGuestCR0(pVCpu, pVmxTransient);
9578	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9579
9580	VBOXSTRICTRC rcStrict = hmR0VmxExportGuestCR3AndCR4(pVCpu, pVmxTransient);
9581	if (rcStrict == VINF_SUCCESS)
9582	{ /* likely */ }
9583	else
9584	{
9585	Assert(rcStrict == VINF_EM_RESCHEDULE_REM \|\| RT_FAILURE_NP(rcStrict));
9586	return rcStrict;
9587	}
9588
9589	rc = hmR0VmxExportGuestSegRegsXdtr(pVCpu, pVmxTransient);
9590	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9591
9592	rc = hmR0VmxExportGuestMsrs(pVCpu, pVmxTransient);
9593	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9594
9595	hmR0VmxExportGuestApicTpr(pVCpu, pVmxTransient);
9596	hmR0VmxExportGuestXcptIntercepts(pVCpu, pVmxTransient);
9597	hmR0VmxExportGuestRip(pVCpu);
9598	hmR0VmxExportGuestRsp(pVCpu);
9599	hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9600
9601	rc = hmR0VmxExportGuestHwvirtState(pVCpu, pVmxTransient);
9602	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc\n", rc), rc);
9603
9604	/* Clear any bits that may be set but exported unconditionally or unused/reserved bits. */
9605	ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~( (HM_CHANGED_GUEST_GPRS_MASK & ~HM_CHANGED_GUEST_RSP)
9606	\| HM_CHANGED_GUEST_CR2
9607	\| (HM_CHANGED_GUEST_DR_MASK & ~HM_CHANGED_GUEST_DR7)
9608	\| HM_CHANGED_GUEST_X87
9609	\| HM_CHANGED_GUEST_SSE_AVX
9610	\| HM_CHANGED_GUEST_OTHER_XSAVE
9611	\| HM_CHANGED_GUEST_XCRx
9612	\| HM_CHANGED_GUEST_KERNEL_GS_BASE /* Part of lazy or auto load-store MSRs. */
9613	\| HM_CHANGED_GUEST_SYSCALL_MSRS /* Part of lazy or auto load-store MSRs. */
9614	\| HM_CHANGED_GUEST_TSC_AUX
9615	\| HM_CHANGED_GUEST_OTHER_MSRS
9616	\| (HM_CHANGED_KEEPER_STATE_MASK & ~HM_CHANGED_VMX_MASK)));
9617
9618	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExportGuestState, x);
9619	return rc;
9620	}
9621
9622
9623	/**
9624	* Exports the state shared between the host and guest into the VMCS.
9625	*
9626	* @param pVCpu The cross context virtual CPU structure.
9627	* @param pVmxTransient The VMX-transient structure.
9628	*
9629	* @remarks No-long-jump zone!!!
9630	*/
9631	static void hmR0VmxExportSharedState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9632	{
9633	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
9634	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9635
9636	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_DR_MASK)
9637	{
9638	int rc = hmR0VmxExportSharedDebugState(pVCpu, pVmxTransient);
9639	AssertRC(rc);
9640	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_GUEST_DR_MASK;
9641
9642	/* Loading shared debug bits might have changed eflags.TF bit for debugging purposes. */
9643	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_RFLAGS)
9644	hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9645	}
9646
9647	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_GUEST_LAZY_MSRS)
9648	{
9649	hmR0VmxLazyLoadGuestMsrs(pVCpu);
9650	pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_VMX_GUEST_LAZY_MSRS;
9651	}
9652
9653	AssertMsg(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE),
9654	("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
9655	}
9656
9657
9658	/**
9659	* Worker for loading the guest-state bits in the inner VT-x execution loop.
9660	*
9661	* @returns Strict VBox status code (i.e. informational status codes too).
9662	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
9663	* without unrestricted guest execution and the VMMDev is not presently
9664	* mapped (e.g. EFI32).
9665	*
9666	* @param pVCpu The cross context virtual CPU structure.
9667	* @param pVmxTransient The VMX-transient structure.
9668	*
9669	* @remarks No-long-jump zone!!!
9670	*/
9671	static VBOXSTRICTRC hmR0VmxExportGuestStateOptimal(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
9672	{
9673	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
9674	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9675
9676	#ifdef HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
9677	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
9678	#endif
9679
9680	/*
9681	* For many VM-exits only RIP/RSP/RFLAGS (and HWVIRT state when executing a nested-guest)
9682	* changes. First try to export only these without going through all other changed-flag checks.
9683	*/
9684	VBOXSTRICTRC rcStrict;
9685	uint64_t const fCtxMask = HM_CHANGED_ALL_GUEST & ~HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE;
9686	uint64_t const fMinimalMask = HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT;
9687	uint64_t const fCtxChanged = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
9688
9689	/* If only RIP/RSP/RFLAGS/HWVIRT changed, export only those (quicker, happens more often).*/
9690	if ( (fCtxChanged & fMinimalMask)
9691	&& !(fCtxChanged & (fCtxMask & ~fMinimalMask)))
9692	{
9693	hmR0VmxExportGuestRip(pVCpu);
9694	hmR0VmxExportGuestRsp(pVCpu);
9695	hmR0VmxExportGuestRflags(pVCpu, pVmxTransient);
9696	rcStrict = hmR0VmxExportGuestHwvirtState(pVCpu, pVmxTransient);
9697	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportMinimal);
9698	}
9699	/* If anything else also changed, go through the full export routine and export as required. */
9700	else if (fCtxChanged & fCtxMask)
9701	{
9702	rcStrict = hmR0VmxExportGuestState(pVCpu, pVmxTransient);
9703	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
9704	{ /* likely */}
9705	else
9706	{
9707	AssertMsg(rcStrict == VINF_EM_RESCHEDULE_REM, ("Failed to export guest state! rc=%Rrc\n",
9708	VBOXSTRICTRC_VAL(rcStrict)));
9709	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
9710	return rcStrict;
9711	}
9712	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportFull);
9713	}
9714	/* Nothing changed, nothing to load here. */
9715	else
9716	rcStrict = VINF_SUCCESS;
9717
9718	#ifdef VBOX_STRICT
9719	/* All the guest state bits should be loaded except maybe the host context and/or the shared host/guest bits. */
9720	uint64_t const fCtxChangedCur = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
9721	AssertMsg(!(fCtxChangedCur & fCtxMask), ("fCtxChangedCur=%#RX64\n", fCtxChangedCur));
9722	#endif
9723	return rcStrict;
9724	}
9725
9726
9727	/**
9728	* Tries to determine what part of the guest-state VT-x has deemed as invalid
9729	* and update error record fields accordingly.
9730	*
9731	* @returns VMX_IGS_* error codes.
9732	* @retval VMX_IGS_REASON_NOT_FOUND if this function could not find anything
9733	* wrong with the guest state.
9734	*
9735	* @param pVCpu The cross context virtual CPU structure.
9736	* @param pVmcsInfo The VMCS info. object.
9737	*
9738	* @remarks This function assumes our cache of the VMCS controls
9739	* are valid, i.e. hmR0VmxCheckCachedVmcsCtls() succeeded.
9740	*/
9741	static uint32_t hmR0VmxCheckGuestState(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
9742	{
9743	#define HMVMX_ERROR_BREAK(err) { uError = (err); break; }
9744	#define HMVMX_CHECK_BREAK(expr, err) do { \
9745	if (!(expr)) { uError = (err); break; } \
9746	} while (0)
9747
9748	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
9749	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
9750	uint32_t uError = VMX_IGS_ERROR;
9751	uint32_t u32IntrState = 0;
9752	bool const fUnrestrictedGuest = pVM->hmr0.s.vmx.fUnrestrictedGuest;
9753	do
9754	{
9755	int rc;
9756
9757	/*
9758	* Guest-interruptibility state.
9759	*
9760	* Read this first so that any check that fails prior to those that actually
9761	* require the guest-interruptibility state would still reflect the correct
9762	* VMCS value and avoids causing further confusion.
9763	*/
9764	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &u32IntrState);
9765	AssertRC(rc);
9766
9767	uint32_t u32Val;
9768	uint64_t u64Val;
9769
9770	/*
9771	* CR0.
9772	*/
9773	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
9774	uint64_t fSetCr0 = (g_HmMsrs.u.vmx.u64Cr0Fixed0 & g_HmMsrs.u.vmx.u64Cr0Fixed1);
9775	uint64_t const fZapCr0 = (g_HmMsrs.u.vmx.u64Cr0Fixed0 \| g_HmMsrs.u.vmx.u64Cr0Fixed1);
9776	/* Exceptions for unrestricted guest execution for CR0 fixed bits (PE, PG).
9777	See Intel spec. 26.3.1 "Checks on Guest Control Registers, Debug Registers and MSRs." */
9778	if (fUnrestrictedGuest)
9779	fSetCr0 &= ~(uint64_t)(X86_CR0_PE \| X86_CR0_PG);
9780
9781	uint64_t u64GuestCr0;
9782	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64GuestCr0);
9783	AssertRC(rc);
9784	HMVMX_CHECK_BREAK((u64GuestCr0 & fSetCr0) == fSetCr0, VMX_IGS_CR0_FIXED1);
9785	HMVMX_CHECK_BREAK(!(u64GuestCr0 & ~fZapCr0), VMX_IGS_CR0_FIXED0);
9786	if ( !fUnrestrictedGuest
9787	&& (u64GuestCr0 & X86_CR0_PG)
9788	&& !(u64GuestCr0 & X86_CR0_PE))
9789	HMVMX_ERROR_BREAK(VMX_IGS_CR0_PG_PE_COMBO);
9790
9791	/*
9792	* CR4.
9793	*/
9794	/** @todo Why do we need to OR and AND the fixed-0 and fixed-1 bits below? */
9795	uint64_t const fSetCr4 = (g_HmMsrs.u.vmx.u64Cr4Fixed0 & g_HmMsrs.u.vmx.u64Cr4Fixed1);
9796	uint64_t const fZapCr4 = (g_HmMsrs.u.vmx.u64Cr4Fixed0 \| g_HmMsrs.u.vmx.u64Cr4Fixed1);
9797
9798	uint64_t u64GuestCr4;
9799	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR4, &u64GuestCr4);
9800	AssertRC(rc);
9801	HMVMX_CHECK_BREAK((u64GuestCr4 & fSetCr4) == fSetCr4, VMX_IGS_CR4_FIXED1);
9802	HMVMX_CHECK_BREAK(!(u64GuestCr4 & ~fZapCr4), VMX_IGS_CR4_FIXED0);
9803
9804	/*
9805	* IA32_DEBUGCTL MSR.
9806	*/
9807	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, &u64Val);
9808	AssertRC(rc);
9809	if ( (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
9810	&& (u64Val & 0xfffffe3c)) /* Bits 31:9, bits 5:2 MBZ. */
9811	{
9812	HMVMX_ERROR_BREAK(VMX_IGS_DEBUGCTL_MSR_RESERVED);
9813	}
9814	uint64_t u64DebugCtlMsr = u64Val;
9815
9816	#ifdef VBOX_STRICT
9817	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
9818	AssertRC(rc);
9819	Assert(u32Val == pVmcsInfo->u32EntryCtls);
9820	#endif
9821	bool const fLongModeGuest = RT_BOOL(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
9822
9823	/*
9824	* RIP and RFLAGS.
9825	*/
9826	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RIP, &u64Val);
9827	AssertRC(rc);
9828	/* pCtx->rip can be different than the one in the VMCS (e.g. run guest code and VM-exits that don't update it). */
9829	if ( !fLongModeGuest
9830	\|\| !pCtx->cs.Attr.n.u1Long)
9831	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffff00000000)), VMX_IGS_LONGMODE_RIP_INVALID);
9832	/** @todo If the processor supports N < 64 linear-address bits, bits 63:N
9833	* must be identical if the "IA-32e mode guest" VM-entry
9834	* control is 1 and CS.L is 1. No check applies if the
9835	* CPU supports 64 linear-address bits. */
9836
9837	/* Flags in pCtx can be different (real-on-v86 for instance). We are only concerned about the VMCS contents here. */
9838	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_RFLAGS, &u64Val);
9839	AssertRC(rc);
9840	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffc08028)), /* Bit 63:22, Bit 15, 5, 3 MBZ. */
9841	VMX_IGS_RFLAGS_RESERVED);
9842	HMVMX_CHECK_BREAK((u64Val & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
9843	uint32_t const u32Eflags = u64Val;
9844
9845	if ( fLongModeGuest
9846	\|\| ( fUnrestrictedGuest
9847	&& !(u64GuestCr0 & X86_CR0_PE)))
9848	{
9849	HMVMX_CHECK_BREAK(!(u32Eflags & X86_EFL_VM), VMX_IGS_RFLAGS_VM_INVALID);
9850	}
9851
9852	uint32_t u32EntryInfo;
9853	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo);
9854	AssertRC(rc);
9855	if (VMX_ENTRY_INT_INFO_IS_EXT_INT(u32EntryInfo))
9856	HMVMX_CHECK_BREAK(u32Eflags & X86_EFL_IF, VMX_IGS_RFLAGS_IF_INVALID);
9857
9858	/*
9859	* 64-bit checks.
9860	*/
9861	if (fLongModeGuest)
9862	{
9863	HMVMX_CHECK_BREAK(u64GuestCr0 & X86_CR0_PG, VMX_IGS_CR0_PG_LONGMODE);
9864	HMVMX_CHECK_BREAK(u64GuestCr4 & X86_CR4_PAE, VMX_IGS_CR4_PAE_LONGMODE);
9865	}
9866
9867	if ( !fLongModeGuest
9868	&& (u64GuestCr4 & X86_CR4_PCIDE))
9869	HMVMX_ERROR_BREAK(VMX_IGS_CR4_PCIDE);
9870
9871	/** @todo CR3 field must be such that bits 63:52 and bits in the range
9872	* 51:32 beyond the processor's physical-address width are 0. */
9873
9874	if ( (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
9875	&& (pCtx->dr[7] & X86_DR7_MBZ_MASK))
9876	HMVMX_ERROR_BREAK(VMX_IGS_DR7_RESERVED);
9877
9878	rc = VMXReadVmcsNw(VMX_VMCS_HOST_SYSENTER_ESP, &u64Val);
9879	AssertRC(rc);
9880	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_ESP_NOT_CANONICAL);
9881
9882	rc = VMXReadVmcsNw(VMX_VMCS_HOST_SYSENTER_EIP, &u64Val);
9883	AssertRC(rc);
9884	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_EIP_NOT_CANONICAL);
9885
9886	/*
9887	* PERF_GLOBAL MSR.
9888	*/
9889	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PERF_MSR)
9890	{
9891	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL, &u64Val);
9892	AssertRC(rc);
9893	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffff8fffffffc)),
9894	VMX_IGS_PERF_GLOBAL_MSR_RESERVED); /* Bits 63:35, bits 31:2 MBZ. */
9895	}
9896
9897	/*
9898	* PAT MSR.
9899	*/
9900	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PAT_MSR)
9901	{
9902	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PAT_FULL, &u64Val);
9903	AssertRC(rc);
9904	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0x707070707070707)), VMX_IGS_PAT_MSR_RESERVED);
9905	for (unsigned i = 0; i < 8; i++)
9906	{
9907	uint8_t u8Val = (u64Val & 0xff);
9908	if ( u8Val != 0 /* UC */
9909	&& u8Val != 1 /* WC */
9910	&& u8Val != 4 /* WT */
9911	&& u8Val != 5 /* WP */
9912	&& u8Val != 6 /* WB */
9913	&& u8Val != 7 /* UC- */)
9914	HMVMX_ERROR_BREAK(VMX_IGS_PAT_MSR_INVALID);
9915	u64Val >>= 8;
9916	}
9917	}
9918
9919	/*
9920	* EFER MSR.
9921	*/
9922	if (pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_EFER_MSR)
9923	{
9924	Assert(g_fHmVmxSupportsVmcsEfer);
9925	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_EFER_FULL, &u64Val);
9926	AssertRC(rc);
9927	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffffffffff2fe)),
9928	VMX_IGS_EFER_MSR_RESERVED); /* Bits 63:12, bit 9, bits 7:1 MBZ. */
9929	HMVMX_CHECK_BREAK(RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL( pVmcsInfo->u32EntryCtls
9930	& VMX_ENTRY_CTLS_IA32E_MODE_GUEST),
9931	VMX_IGS_EFER_LMA_GUEST_MODE_MISMATCH);
9932	/** @todo r=ramshankar: Unrestricted check here is probably wrong, see
9933	* iemVmxVmentryCheckGuestState(). */
9934	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9935	\|\| !(u64GuestCr0 & X86_CR0_PG)
9936	\|\| RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL(u64Val & MSR_K6_EFER_LME),
9937	VMX_IGS_EFER_LMA_LME_MISMATCH);
9938	}
9939
9940	/*
9941	* Segment registers.
9942	*/
9943	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9944	\|\| !(pCtx->ldtr.Sel & X86_SEL_LDT), VMX_IGS_LDTR_TI_INVALID);
9945	if (!(u32Eflags & X86_EFL_VM))
9946	{
9947	/* CS */
9948	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1Present, VMX_IGS_CS_ATTR_P_INVALID);
9949	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xf00), VMX_IGS_CS_ATTR_RESERVED);
9950	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xfffe0000), VMX_IGS_CS_ATTR_RESERVED);
9951	HMVMX_CHECK_BREAK( (pCtx->cs.u32Limit & 0xfff) == 0xfff
9952	\|\| !(pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9953	HMVMX_CHECK_BREAK( !(pCtx->cs.u32Limit & 0xfff00000)
9954	\|\| (pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9955	/* CS cannot be loaded with NULL in protected mode. */
9956	HMVMX_CHECK_BREAK(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_CS_ATTR_UNUSABLE);
9957	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1DescType, VMX_IGS_CS_ATTR_S_INVALID);
9958	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
9959	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_UNEQUAL);
9960	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
9961	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_MISMATCH);
9962	else if (fUnrestrictedGuest && pCtx->cs.Attr.n.u4Type == 3)
9963	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == 0, VMX_IGS_CS_ATTR_DPL_INVALID);
9964	else
9965	HMVMX_ERROR_BREAK(VMX_IGS_CS_ATTR_TYPE_INVALID);
9966
9967	/* SS */
9968	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9969	\|\| (pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL), VMX_IGS_SS_CS_RPL_UNEQUAL);
9970	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL), VMX_IGS_SS_ATTR_DPL_RPL_UNEQUAL);
9971	if ( !(pCtx->cr0 & X86_CR0_PE)
9972	\|\| pCtx->cs.Attr.n.u4Type == 3)
9973	HMVMX_CHECK_BREAK(!pCtx->ss.Attr.n.u2Dpl, VMX_IGS_SS_ATTR_DPL_INVALID);
9974
9975	if (!(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
9976	{
9977	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7, VMX_IGS_SS_ATTR_TYPE_INVALID);
9978	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u1Present, VMX_IGS_SS_ATTR_P_INVALID);
9979	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xf00), VMX_IGS_SS_ATTR_RESERVED);
9980	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xfffe0000), VMX_IGS_SS_ATTR_RESERVED);
9981	HMVMX_CHECK_BREAK( (pCtx->ss.u32Limit & 0xfff) == 0xfff
9982	\|\| !(pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9983	HMVMX_CHECK_BREAK( !(pCtx->ss.u32Limit & 0xfff00000)
9984	\|\| (pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9985	}
9986
9987	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxExportGuestSReg(). */
9988	if (!(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
9989	{
9990	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_DS_ATTR_A_INVALID);
9991	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u1Present, VMX_IGS_DS_ATTR_P_INVALID);
9992	HMVMX_CHECK_BREAK( fUnrestrictedGuest
9993	\|\| pCtx->ds.Attr.n.u4Type > 11
9994	\|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9995	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xf00), VMX_IGS_DS_ATTR_RESERVED);
9996	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xfffe0000), VMX_IGS_DS_ATTR_RESERVED);
9997	HMVMX_CHECK_BREAK( (pCtx->ds.u32Limit & 0xfff) == 0xfff
9998	\|\| !(pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9999	HMVMX_CHECK_BREAK( !(pCtx->ds.u32Limit & 0xfff00000)
10000	\|\| (pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
10001	HMVMX_CHECK_BREAK( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10002	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_DS_ATTR_TYPE_INVALID);
10003	}
10004	if (!(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
10005	{
10006	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_ES_ATTR_A_INVALID);
10007	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u1Present, VMX_IGS_ES_ATTR_P_INVALID);
10008	HMVMX_CHECK_BREAK( fUnrestrictedGuest
10009	\|\| pCtx->es.Attr.n.u4Type > 11
10010	\|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
10011	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xf00), VMX_IGS_ES_ATTR_RESERVED);
10012	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xfffe0000), VMX_IGS_ES_ATTR_RESERVED);
10013	HMVMX_CHECK_BREAK( (pCtx->es.u32Limit & 0xfff) == 0xfff
10014	\|\| !(pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
10015	HMVMX_CHECK_BREAK( !(pCtx->es.u32Limit & 0xfff00000)
10016	\|\| (pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
10017	HMVMX_CHECK_BREAK( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10018	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_ES_ATTR_TYPE_INVALID);
10019	}
10020	if (!(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
10021	{
10022	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_FS_ATTR_A_INVALID);
10023	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u1Present, VMX_IGS_FS_ATTR_P_INVALID);
10024	HMVMX_CHECK_BREAK( fUnrestrictedGuest
10025	\|\| pCtx->fs.Attr.n.u4Type > 11
10026	\|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL), VMX_IGS_FS_ATTR_DPL_RPL_UNEQUAL);
10027	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xf00), VMX_IGS_FS_ATTR_RESERVED);
10028	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xfffe0000), VMX_IGS_FS_ATTR_RESERVED);
10029	HMVMX_CHECK_BREAK( (pCtx->fs.u32Limit & 0xfff) == 0xfff
10030	\|\| !(pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
10031	HMVMX_CHECK_BREAK( !(pCtx->fs.u32Limit & 0xfff00000)
10032	\|\| (pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
10033	HMVMX_CHECK_BREAK( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10034	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_FS_ATTR_TYPE_INVALID);
10035	}
10036	if (!(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
10037	{
10038	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_GS_ATTR_A_INVALID);
10039	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u1Present, VMX_IGS_GS_ATTR_P_INVALID);
10040	HMVMX_CHECK_BREAK( fUnrestrictedGuest
10041	\|\| pCtx->gs.Attr.n.u4Type > 11
10042	\|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL), VMX_IGS_GS_ATTR_DPL_RPL_UNEQUAL);
10043	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xf00), VMX_IGS_GS_ATTR_RESERVED);
10044	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xfffe0000), VMX_IGS_GS_ATTR_RESERVED);
10045	HMVMX_CHECK_BREAK( (pCtx->gs.u32Limit & 0xfff) == 0xfff
10046	\|\| !(pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
10047	HMVMX_CHECK_BREAK( !(pCtx->gs.u32Limit & 0xfff00000)
10048	\|\| (pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
10049	HMVMX_CHECK_BREAK( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10050	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_GS_ATTR_TYPE_INVALID);
10051	}
10052	/* 64-bit capable CPUs. */
10053	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
10054	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
10055	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
10056	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
10057	HMVMX_CHECK_BREAK(!RT_HI_U32(pCtx->cs.u64Base), VMX_IGS_LONGMODE_CS_BASE_INVALID);
10058	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ss.u64Base),
10059	VMX_IGS_LONGMODE_SS_BASE_INVALID);
10060	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ds.u64Base),
10061	VMX_IGS_LONGMODE_DS_BASE_INVALID);
10062	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->es.u64Base),
10063	VMX_IGS_LONGMODE_ES_BASE_INVALID);
10064	}
10065	else
10066	{
10067	/* V86 mode checks. */
10068	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
10069	if (pVmcsInfo->pShared->RealMode.fRealOnV86Active)
10070	{
10071	u32CSAttr = 0xf3; u32SSAttr = 0xf3;
10072	u32DSAttr = 0xf3; u32ESAttr = 0xf3;
10073	u32FSAttr = 0xf3; u32GSAttr = 0xf3;
10074	}
10075	else
10076	{
10077	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u;
10078	u32DSAttr = pCtx->ds.Attr.u; u32ESAttr = pCtx->es.Attr.u;
10079	u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
10080	}
10081
10082	/* CS */
10083	HMVMX_CHECK_BREAK((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), VMX_IGS_V86_CS_BASE_INVALID);
10084	HMVMX_CHECK_BREAK(pCtx->cs.u32Limit == 0xffff, VMX_IGS_V86_CS_LIMIT_INVALID);
10085	HMVMX_CHECK_BREAK(u32CSAttr == 0xf3, VMX_IGS_V86_CS_ATTR_INVALID);
10086	/* SS */
10087	HMVMX_CHECK_BREAK((pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4), VMX_IGS_V86_SS_BASE_INVALID);
10088	HMVMX_CHECK_BREAK(pCtx->ss.u32Limit == 0xffff, VMX_IGS_V86_SS_LIMIT_INVALID);
10089	HMVMX_CHECK_BREAK(u32SSAttr == 0xf3, VMX_IGS_V86_SS_ATTR_INVALID);
10090	/* DS */
10091	HMVMX_CHECK_BREAK((pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4), VMX_IGS_V86_DS_BASE_INVALID);
10092	HMVMX_CHECK_BREAK(pCtx->ds.u32Limit == 0xffff, VMX_IGS_V86_DS_LIMIT_INVALID);
10093	HMVMX_CHECK_BREAK(u32DSAttr == 0xf3, VMX_IGS_V86_DS_ATTR_INVALID);
10094	/* ES */
10095	HMVMX_CHECK_BREAK((pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4), VMX_IGS_V86_ES_BASE_INVALID);
10096	HMVMX_CHECK_BREAK(pCtx->es.u32Limit == 0xffff, VMX_IGS_V86_ES_LIMIT_INVALID);
10097	HMVMX_CHECK_BREAK(u32ESAttr == 0xf3, VMX_IGS_V86_ES_ATTR_INVALID);
10098	/* FS */
10099	HMVMX_CHECK_BREAK((pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4), VMX_IGS_V86_FS_BASE_INVALID);
10100	HMVMX_CHECK_BREAK(pCtx->fs.u32Limit == 0xffff, VMX_IGS_V86_FS_LIMIT_INVALID);
10101	HMVMX_CHECK_BREAK(u32FSAttr == 0xf3, VMX_IGS_V86_FS_ATTR_INVALID);
10102	/* GS */
10103	HMVMX_CHECK_BREAK((pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4), VMX_IGS_V86_GS_BASE_INVALID);
10104	HMVMX_CHECK_BREAK(pCtx->gs.u32Limit == 0xffff, VMX_IGS_V86_GS_LIMIT_INVALID);
10105	HMVMX_CHECK_BREAK(u32GSAttr == 0xf3, VMX_IGS_V86_GS_ATTR_INVALID);
10106	/* 64-bit capable CPUs. */
10107	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
10108	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
10109	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
10110	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
10111	HMVMX_CHECK_BREAK(!RT_HI_U32(pCtx->cs.u64Base), VMX_IGS_LONGMODE_CS_BASE_INVALID);
10112	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ss.u64Base),
10113	VMX_IGS_LONGMODE_SS_BASE_INVALID);
10114	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->ds.u64Base),
10115	VMX_IGS_LONGMODE_DS_BASE_INVALID);
10116	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !RT_HI_U32(pCtx->es.u64Base),
10117	VMX_IGS_LONGMODE_ES_BASE_INVALID);
10118	}
10119
10120	/*
10121	* TR.
10122	*/
10123	HMVMX_CHECK_BREAK(!(pCtx->tr.Sel & X86_SEL_LDT), VMX_IGS_TR_TI_INVALID);
10124	/* 64-bit capable CPUs. */
10125	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->tr.u64Base), VMX_IGS_TR_BASE_NOT_CANONICAL);
10126	if (fLongModeGuest)
10127	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u4Type == 11, /* 64-bit busy TSS. */
10128	VMX_IGS_LONGMODE_TR_ATTR_TYPE_INVALID);
10129	else
10130	HMVMX_CHECK_BREAK( pCtx->tr.Attr.n.u4Type == 3 /* 16-bit busy TSS. */
10131	\|\| pCtx->tr.Attr.n.u4Type == 11, /* 32-bit busy TSS.*/
10132	VMX_IGS_TR_ATTR_TYPE_INVALID);
10133	HMVMX_CHECK_BREAK(!pCtx->tr.Attr.n.u1DescType, VMX_IGS_TR_ATTR_S_INVALID);
10134	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u1Present, VMX_IGS_TR_ATTR_P_INVALID);
10135	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & 0xf00), VMX_IGS_TR_ATTR_RESERVED); /* Bits 11:8 MBZ. */
10136	HMVMX_CHECK_BREAK( (pCtx->tr.u32Limit & 0xfff) == 0xfff
10137	\|\| !(pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
10138	HMVMX_CHECK_BREAK( !(pCtx->tr.u32Limit & 0xfff00000)
10139	\|\| (pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
10140	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_TR_ATTR_UNUSABLE);
10141
10142	/*
10143	* GDTR and IDTR (64-bit capable checks).
10144	*/
10145	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
10146	AssertRC(rc);
10147	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_GDTR_BASE_NOT_CANONICAL);
10148
10149	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
10150	AssertRC(rc);
10151	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_IDTR_BASE_NOT_CANONICAL);
10152
10153	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val);
10154	AssertRC(rc);
10155	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_GDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
10156
10157	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val);
10158	AssertRC(rc);
10159	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_IDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
10160
10161	/*
10162	* Guest Non-Register State.
10163	*/
10164	/* Activity State. */
10165	uint32_t u32ActivityState;
10166	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, &u32ActivityState);
10167	AssertRC(rc);
10168	HMVMX_CHECK_BREAK( !u32ActivityState
10169	\|\| (u32ActivityState & RT_BF_GET(g_HmMsrs.u.vmx.u64Misc, VMX_BF_MISC_ACTIVITY_STATES)),
10170	VMX_IGS_ACTIVITY_STATE_INVALID);
10171	HMVMX_CHECK_BREAK( !(pCtx->ss.Attr.n.u2Dpl)
10172	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT, VMX_IGS_ACTIVITY_STATE_HLT_INVALID);
10173
10174	if ( u32IntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS
10175	\|\| u32IntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10176	HMVMX_CHECK_BREAK(u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE, VMX_IGS_ACTIVITY_STATE_ACTIVE_INVALID);
10177
10178	/** @todo Activity state and injecting interrupts. Left as a todo since we
10179	* currently don't use activity states but ACTIVE. */
10180
10181	HMVMX_CHECK_BREAK( !(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)
10182	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT, VMX_IGS_ACTIVITY_STATE_SIPI_WAIT_INVALID);
10183
10184	/* Guest interruptibility-state. */
10185	HMVMX_CHECK_BREAK(!(u32IntrState & 0xffffffe0), VMX_IGS_INTERRUPTIBILITY_STATE_RESERVED);
10186	HMVMX_CHECK_BREAK((u32IntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
10187	!= (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI \| VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
10188	VMX_IGS_INTERRUPTIBILITY_STATE_STI_MOVSS_INVALID);
10189	HMVMX_CHECK_BREAK( (u32Eflags & X86_EFL_IF)
10190	\|\| !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI),
10191	VMX_IGS_INTERRUPTIBILITY_STATE_STI_EFL_INVALID);
10192	if (VMX_ENTRY_INT_INFO_IS_EXT_INT(u32EntryInfo))
10193	{
10194	HMVMX_CHECK_BREAK( !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10195	&& !(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
10196	VMX_IGS_INTERRUPTIBILITY_STATE_EXT_INT_INVALID);
10197	}
10198	else if (VMX_ENTRY_INT_INFO_IS_XCPT_NMI(u32EntryInfo))
10199	{
10200	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS),
10201	VMX_IGS_INTERRUPTIBILITY_STATE_MOVSS_INVALID);
10202	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI),
10203	VMX_IGS_INTERRUPTIBILITY_STATE_STI_INVALID);
10204	}
10205	/** @todo Assumes the processor is not in SMM. */
10206	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI),
10207	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_INVALID);
10208	HMVMX_CHECK_BREAK( !(pVmcsInfo->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)
10209	\|\| (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI),
10210	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_SMM_INVALID);
10211	if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
10212	&& VMX_ENTRY_INT_INFO_IS_XCPT_NMI(u32EntryInfo))
10213	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI), VMX_IGS_INTERRUPTIBILITY_STATE_NMI_INVALID);
10214
10215	/* Pending debug exceptions. */
10216	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, &u64Val);
10217	AssertRC(rc);
10218	/* Bits 63:15, Bit 13, Bits 11:4 MBZ. */
10219	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffffaff0)), VMX_IGS_LONGMODE_PENDING_DEBUG_RESERVED);
10220	u32Val = u64Val; /* For pending debug exceptions checks below. */
10221
10222	if ( (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
10223	\|\| (u32IntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS)
10224	\|\| u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT)
10225	{
10226	if ( (u32Eflags & X86_EFL_TF)
10227	&& !(u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
10228	{
10229	/* Bit 14 is PendingDebug.BS. */
10230	HMVMX_CHECK_BREAK(u32Val & RT_BIT(14), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_SET);
10231	}
10232	if ( !(u32Eflags & X86_EFL_TF)
10233	\|\| (u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
10234	{
10235	/* Bit 14 is PendingDebug.BS. */
10236	HMVMX_CHECK_BREAK(!(u32Val & RT_BIT(14)), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_CLEAR);
10237	}
10238	}
10239
10240	/* VMCS link pointer. */
10241	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, &u64Val);
10242	AssertRC(rc);
10243	if (u64Val != UINT64_C(0xffffffffffffffff))
10244	{
10245	HMVMX_CHECK_BREAK(!(u64Val & 0xfff), VMX_IGS_VMCS_LINK_PTR_RESERVED);
10246	/** @todo Bits beyond the processor's physical-address width MBZ. */
10247	/** @todo SMM checks. */
10248	Assert(pVmcsInfo->HCPhysShadowVmcs == u64Val);
10249	Assert(pVmcsInfo->pvShadowVmcs);
10250	VMXVMCSREVID VmcsRevId;
10251	VmcsRevId.u = (uint32_t )pVmcsInfo->pvShadowVmcs;
10252	HMVMX_CHECK_BREAK(VmcsRevId.n.u31RevisionId == RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_ID),
10253	VMX_IGS_VMCS_LINK_PTR_SHADOW_VMCS_ID_INVALID);
10254	HMVMX_CHECK_BREAK(VmcsRevId.n.fIsShadowVmcs == (uint32_t)!!(pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING),
10255	VMX_IGS_VMCS_LINK_PTR_NOT_SHADOW);
10256	}
10257
10258	/** @todo Checks on Guest Page-Directory-Pointer-Table Entries when guest is
10259	* not using nested paging? */
10260	if ( pVM->hmr0.s.fNestedPaging
10261	&& !fLongModeGuest
10262	&& CPUMIsGuestInPAEModeEx(pCtx))
10263	{
10264	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &u64Val);
10265	AssertRC(rc);
10266	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10267
10268	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &u64Val);
10269	AssertRC(rc);
10270	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10271
10272	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &u64Val);
10273	AssertRC(rc);
10274	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10275
10276	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &u64Val);
10277	AssertRC(rc);
10278	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10279	}
10280
10281	/* Shouldn't happen but distinguish it from AssertRCBreak() errors. */
10282	if (uError == VMX_IGS_ERROR)
10283	uError = VMX_IGS_REASON_NOT_FOUND;
10284	} while (0);
10285
10286	pVCpu->hm.s.u32HMError = uError;
10287	pVCpu->hm.s.vmx.LastError.u32GuestIntrState = u32IntrState;
10288	return uError;
10289
10290	#undef HMVMX_ERROR_BREAK
10291	#undef HMVMX_CHECK_BREAK
10292	}
10293
10294
10295	/**
10296	* Map the APIC-access page for virtualizing APIC accesses.
10297	*
10298	* This can cause a longjumps to R3 due to the acquisition of the PGM lock. Hence,
10299	* this not done as part of exporting guest state, see @bugref{8721}.
10300	*
10301	* @returns VBox status code.
10302	* @param pVCpu The cross context virtual CPU structure.
10303	*/
10304	static int hmR0VmxMapHCApicAccessPage(PVMCPUCC pVCpu)
10305	{
10306	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10307	uint64_t const u64MsrApicBase = APICGetBaseMsrNoCheck(pVCpu);
10308
10309	Assert(PDMHasApic(pVM));
10310	Assert(u64MsrApicBase);
10311
10312	RTGCPHYS const GCPhysApicBase = u64MsrApicBase & PAGE_BASE_GC_MASK;
10313	Log4Func(("Mappping HC APIC-access page at %#RGp\n", GCPhysApicBase));
10314
10315	/* Unalias the existing mapping. */
10316	int rc = PGMHandlerPhysicalReset(pVM, GCPhysApicBase);
10317	AssertRCReturn(rc, rc);
10318
10319	/* Map the HC APIC-access page in place of the MMIO page, also updates the shadow page tables if necessary. */
10320	Assert(pVM->hmr0.s.vmx.HCPhysApicAccess != NIL_RTHCPHYS);
10321	rc = IOMR0MmioMapMmioHCPage(pVM, pVCpu, GCPhysApicBase, pVM->hmr0.s.vmx.HCPhysApicAccess, X86_PTE_RW \| X86_PTE_P);
10322	AssertRCReturn(rc, rc);
10323
10324	/* Update the per-VCPU cache of the APIC base MSR. */
10325	pVCpu->hm.s.vmx.u64GstMsrApicBase = u64MsrApicBase;
10326	return VINF_SUCCESS;
10327	}
10328
10329
10330	/**
10331	* Worker function passed to RTMpOnSpecific() that is to be called on the target
10332	* CPU.
10333	*
10334	* @param idCpu The ID for the CPU the function is called on.
10335	* @param pvUser1 Null, not used.
10336	* @param pvUser2 Null, not used.
10337	*/
10338	static DECLCALLBACK(void) hmR0DispatchHostNmi(RTCPUID idCpu, void pvUser1, void pvUser2)
10339	{
10340	RT_NOREF3(idCpu, pvUser1, pvUser2);
10341	VMXDispatchHostNmi();
10342	}
10343
10344
10345	/**
10346	* Dispatching an NMI on the host CPU that received it.
10347	*
10348	* @returns VBox status code.
10349	* @param pVCpu The cross context virtual CPU structure.
10350	* @param pVmcsInfo The VMCS info. object corresponding to the VMCS that was
10351	* executing when receiving the host NMI in VMX non-root
10352	* operation.
10353	*/
10354	static int hmR0VmxExitHostNmi(PVMCPUCC pVCpu, PCVMXVMCSINFO pVmcsInfo)
10355	{
10356	RTCPUID const idCpu = pVmcsInfo->idHostCpuExec;
10357	Assert(idCpu != NIL_RTCPUID);
10358
10359	/*
10360	* We don't want to delay dispatching the NMI any more than we have to. However,
10361	* we have already chosen -not- to dispatch NMIs when interrupts were still disabled
10362	* after executing guest or nested-guest code for the following reasons:
10363	*
10364	* - We would need to perform VMREADs with interrupts disabled and is orders of
10365	* magnitude worse when we run as a nested hypervisor without VMCS shadowing
10366	* supported by the host hypervisor.
10367	*
10368	* - It affects the common VM-exit scenario and keeps interrupts disabled for a
10369	* longer period of time just for handling an edge case like host NMIs which do
10370	* not occur nearly as frequently as other VM-exits.
10371	*
10372	* Let's cover the most likely scenario first. Check if we are on the target CPU
10373	* and dispatch the NMI right away. This should be much faster than calling into
10374	* RTMpOnSpecific() machinery.
10375	*/
10376	bool fDispatched = false;
10377	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
10378	if (idCpu == RTMpCpuId())
10379	{
10380	VMXDispatchHostNmi();
10381	fDispatched = true;
10382	}
10383	ASMSetFlags(fEFlags);
10384	if (fDispatched)
10385	{
10386	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC);
10387	return VINF_SUCCESS;
10388	}
10389
10390	/*
10391	* RTMpOnSpecific() waits until the worker function has run on the target CPU. So
10392	* there should be no race or recursion even if we are unlucky enough to be preempted
10393	* (to the target CPU) without dispatching the host NMI above.
10394	*/
10395	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGCIpi);
10396	return RTMpOnSpecific(idCpu, &hmR0DispatchHostNmi, NULL /* pvUser1 /, NULL / pvUser2 */);
10397	}
10398
10399
10400	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10401	/**
10402	* Merges the guest with the nested-guest MSR bitmap in preparation of executing the
10403	* nested-guest using hardware-assisted VMX.
10404	*
10405	* @param pVCpu The cross context virtual CPU structure.
10406	* @param pVmcsInfoNstGst The nested-guest VMCS info. object.
10407	* @param pVmcsInfoGst The guest VMCS info. object.
10408	*/
10409	static void hmR0VmxMergeMsrBitmapNested(PCVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfoNstGst, PCVMXVMCSINFO pVmcsInfoGst)
10410	{
10411	uint32_t const cbMsrBitmap = X86_PAGE_4K_SIZE;
10412	uint64_t pu64MsrBitmap = (uint64_t )pVmcsInfoNstGst->pvMsrBitmap;
10413	Assert(pu64MsrBitmap);
10414
10415	/*
10416	* We merge the guest MSR bitmap with the nested-guest MSR bitmap such that any
10417	* MSR that is intercepted by the guest is also intercepted while executing the
10418	* nested-guest using hardware-assisted VMX.
10419	*
10420	* Note! If the nested-guest is not using an MSR bitmap, every MSR must cause a
10421	* nested-guest VM-exit even if the outer guest is not intercepting some
10422	* MSRs. We cannot assume the caller has initialized the nested-guest
10423	* MSR bitmap in this case.
10424	*
10425	* The nested hypervisor may also switch whether it uses MSR bitmaps for
10426	* each of its VM-entry, hence initializing it once per-VM while setting
10427	* up the nested-guest VMCS is not sufficient.
10428	*/
10429	PCVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
10430	if (pVmcsNstGst->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
10431	{
10432	uint64_t const pu64MsrBitmapNstGst = (uint64_t const )&pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap[0];
10433	uint64_t const pu64MsrBitmapGst = (uint64_t const )pVmcsInfoGst->pvMsrBitmap;
10434	Assert(pu64MsrBitmapNstGst);
10435	Assert(pu64MsrBitmapGst);
10436
10437	/** @todo Detect and use EVEX.POR? */
10438	uint32_t const cFrags = cbMsrBitmap / sizeof(uint64_t);
10439	for (uint32_t i = 0; i < cFrags; i++)
10440	pu64MsrBitmap[i] = pu64MsrBitmapNstGst[i] \| pu64MsrBitmapGst[i];
10441	}
10442	else
10443	ASMMemFill32(pu64MsrBitmap, cbMsrBitmap, UINT32_C(0xffffffff));
10444	}
10445
10446
10447	/**
10448	* Merges the guest VMCS in to the nested-guest VMCS controls in preparation of
10449	* hardware-assisted VMX execution of the nested-guest.
10450	*
10451	* For a guest, we don't modify these controls once we set up the VMCS and hence
10452	* this function is never called.
10453	*
10454	* For nested-guests since the nested hypervisor provides these controls on every
10455	* nested-guest VM-entry and could potentially change them everytime we need to
10456	* merge them before every nested-guest VM-entry.
10457	*
10458	* @returns VBox status code.
10459	* @param pVCpu The cross context virtual CPU structure.
10460	*/
10461	static int hmR0VmxMergeVmcsNested(PVMCPUCC pVCpu)
10462	{
10463	PVMCC const pVM = pVCpu->CTX_SUFF(pVM);
10464	PCVMXVMCSINFO const pVmcsInfoGst = &pVCpu->hmr0.s.vmx.VmcsInfo;
10465	PCVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
10466
10467	/*
10468	* Merge the controls with the requirements of the guest VMCS.
10469	*
10470	* We do not need to validate the nested-guest VMX features specified in the nested-guest
10471	* VMCS with the features supported by the physical CPU as it's already done by the
10472	* VMLAUNCH/VMRESUME instruction emulation.
10473	*
10474	* This is because the VMX features exposed by CPUM (through CPUID/MSRs) to the guest are
10475	* derived from the VMX features supported by the physical CPU.
10476	*/
10477
10478	/* Pin-based VM-execution controls. */
10479	uint32_t const u32PinCtls = pVmcsNstGst->u32PinCtls \| pVmcsInfoGst->u32PinCtls;
10480
10481	/* Processor-based VM-execution controls. */
10482	uint32_t u32ProcCtls = (pVmcsNstGst->u32ProcCtls & ~VMX_PROC_CTLS_USE_IO_BITMAPS)
10483	\| (pVmcsInfoGst->u32ProcCtls & ~( VMX_PROC_CTLS_INT_WINDOW_EXIT
10484	\| VMX_PROC_CTLS_NMI_WINDOW_EXIT
10485	\| VMX_PROC_CTLS_MOV_DR_EXIT
10486	\| VMX_PROC_CTLS_USE_TPR_SHADOW
10487	\| VMX_PROC_CTLS_MONITOR_TRAP_FLAG));
10488
10489	/* Secondary processor-based VM-execution controls. */
10490	uint32_t const u32ProcCtls2 = (pVmcsNstGst->u32ProcCtls2 & ~VMX_PROC_CTLS2_VPID)
10491	\| (pVmcsInfoGst->u32ProcCtls2 & ~( VMX_PROC_CTLS2_VIRT_APIC_ACCESS
10492	\| VMX_PROC_CTLS2_INVPCID
10493	\| VMX_PROC_CTLS2_VMCS_SHADOWING
10494	\| VMX_PROC_CTLS2_RDTSCP
10495	\| VMX_PROC_CTLS2_XSAVES_XRSTORS
10496	\| VMX_PROC_CTLS2_APIC_REG_VIRT
10497	\| VMX_PROC_CTLS2_VIRT_INT_DELIVERY
10498	\| VMX_PROC_CTLS2_VMFUNC));
10499
10500	/*
10501	* VM-entry controls:
10502	* These controls contains state that depends on the nested-guest state (primarily
10503	* EFER MSR) and is thus not constant between VMLAUNCH/VMRESUME and the nested-guest
10504	* VM-exit. Although the nested hypervisor cannot change it, we need to in order to
10505	* properly continue executing the nested-guest if the EFER MSR changes but does not
10506	* cause a nested-guest VM-exits.
10507	*
10508	* VM-exit controls:
10509	* These controls specify the host state on return. We cannot use the controls from
10510	* the nested hypervisor state as is as it would contain the guest state rather than
10511	* the host state. Since the host state is subject to change (e.g. preemption, trips
10512	* to ring-3, longjmp and rescheduling to a different host CPU) they are not constant
10513	* through VMLAUNCH/VMRESUME and the nested-guest VM-exit.
10514	*
10515	* VM-entry MSR-load:
10516	* The guest MSRs from the VM-entry MSR-load area are already loaded into the guest-CPU
10517	* context by the VMLAUNCH/VMRESUME instruction emulation.
10518	*
10519	* VM-exit MSR-store:
10520	* The VM-exit emulation will take care of populating the MSRs from the guest-CPU context
10521	* back into the VM-exit MSR-store area.
10522	*
10523	* VM-exit MSR-load areas:
10524	* This must contain the real host MSRs with hardware-assisted VMX execution. Hence, we
10525	* can entirely ignore what the nested hypervisor wants to load here.
10526	*/
10527
10528	/*
10529	* Exception bitmap.
10530	*
10531	* We could remove #UD from the guest bitmap and merge it with the nested-guest bitmap
10532	* here (and avoid doing anything while exporting nested-guest state), but to keep the
10533	* code more flexible if intercepting exceptions become more dynamic in the future we do
10534	* it as part of exporting the nested-guest state.
10535	*/
10536	uint32_t const u32XcptBitmap = pVmcsNstGst->u32XcptBitmap \| pVmcsInfoGst->u32XcptBitmap;
10537
10538	/*
10539	* CR0/CR4 guest/host mask.
10540	*
10541	* Modifications by the nested-guest to CR0/CR4 bits owned by the host and the guest must
10542	* cause VM-exits, so we need to merge them here.
10543	*/
10544	uint64_t const u64Cr0Mask = pVmcsNstGst->u64Cr0Mask.u \| pVmcsInfoGst->u64Cr0Mask;
10545	uint64_t const u64Cr4Mask = pVmcsNstGst->u64Cr4Mask.u \| pVmcsInfoGst->u64Cr4Mask;
10546
10547	/*
10548	* Page-fault error-code mask and match.
10549	*
10550	* Although we require unrestricted guest execution (and thereby nested-paging) for
10551	* hardware-assisted VMX execution of nested-guests and thus the outer guest doesn't
10552	* normally intercept #PFs, it might intercept them for debugging purposes.
10553	*
10554	* If the outer guest is not intercepting #PFs, we can use the nested-guest #PF filters.
10555	* If the outer guest is intercepting #PFs, we must intercept all #PFs.
10556	*/
10557	uint32_t u32XcptPFMask;
10558	uint32_t u32XcptPFMatch;
10559	if (!(pVmcsInfoGst->u32XcptBitmap & RT_BIT(X86_XCPT_PF)))
10560	{
10561	u32XcptPFMask = pVmcsNstGst->u32XcptPFMask;
10562	u32XcptPFMatch = pVmcsNstGst->u32XcptPFMatch;
10563	}
10564	else
10565	{
10566	u32XcptPFMask = 0;
10567	u32XcptPFMatch = 0;
10568	}
10569
10570	/*
10571	* Pause-Loop exiting.
10572	*/
10573	/** @todo r=bird: given that both pVM->hm.s.vmx.cPleGapTicks and
10574	* pVM->hm.s.vmx.cPleWindowTicks defaults to zero, I cannot see how
10575	* this will work... */
10576	uint32_t const cPleGapTicks = RT_MIN(pVM->hm.s.vmx.cPleGapTicks, pVmcsNstGst->u32PleGap);
10577	uint32_t const cPleWindowTicks = RT_MIN(pVM->hm.s.vmx.cPleWindowTicks, pVmcsNstGst->u32PleWindow);
10578
10579	/*
10580	* Pending debug exceptions.
10581	* Currently just copy whatever the nested-guest provides us.
10582	*/
10583	uint64_t const uPendingDbgXcpts = pVmcsNstGst->u64GuestPendingDbgXcpts.u;
10584
10585	/*
10586	* I/O Bitmap.
10587	*
10588	* We do not use the I/O bitmap that may be provided by the nested hypervisor as we always
10589	* intercept all I/O port accesses.
10590	*/
10591	Assert(u32ProcCtls & VMX_PROC_CTLS_UNCOND_IO_EXIT);
10592	Assert(!(u32ProcCtls & VMX_PROC_CTLS_USE_IO_BITMAPS));
10593
10594	/*
10595	* VMCS shadowing.
10596	*
10597	* We do not yet expose VMCS shadowing to the guest and thus VMCS shadowing should not be
10598	* enabled while executing the nested-guest.
10599	*/
10600	Assert(!(u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING));
10601
10602	/*
10603	* APIC-access page.
10604	*/
10605	RTHCPHYS HCPhysApicAccess;
10606	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10607	{
10608	Assert(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS);
10609	RTGCPHYS const GCPhysApicAccess = pVmcsNstGst->u64AddrApicAccess.u;
10610
10611	/** @todo NSTVMX: This is not really correct but currently is required to make
10612	* things work. We need to re-enable the page handler when we fallback to
10613	* IEM execution of the nested-guest! */
10614	PGMHandlerPhysicalPageTempOff(pVM, GCPhysApicAccess, GCPhysApicAccess);
10615
10616	void *pvPage;
10617	PGMPAGEMAPLOCK PgLockApicAccess;
10618	int rc = PGMPhysGCPhys2CCPtr(pVM, GCPhysApicAccess, &pvPage, &PgLockApicAccess);
10619	if (RT_SUCCESS(rc))
10620	{
10621	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysApicAccess, &HCPhysApicAccess);
10622	AssertMsgRCReturn(rc, ("Failed to get host-physical address for APIC-access page at %#RGp\n", GCPhysApicAccess), rc);
10623
10624	/** @todo Handle proper releasing of page-mapping lock later. */
10625	PGMPhysReleasePageMappingLock(pVCpu->CTX_SUFF(pVM), &PgLockApicAccess);
10626	}
10627	else
10628	return rc;
10629	}
10630	else
10631	HCPhysApicAccess = 0;
10632
10633	/*
10634	* Virtual-APIC page and TPR threshold.
10635	*/
10636	RTHCPHYS HCPhysVirtApic;
10637	uint32_t u32TprThreshold;
10638	if (u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
10639	{
10640	Assert(g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_TPR_SHADOW);
10641	RTGCPHYS const GCPhysVirtApic = pVmcsNstGst->u64AddrVirtApic.u;
10642
10643	void *pvPage;
10644	PGMPAGEMAPLOCK PgLockVirtApic;
10645	int rc = PGMPhysGCPhys2CCPtr(pVM, GCPhysVirtApic, &pvPage, &PgLockVirtApic);
10646	if (RT_SUCCESS(rc))
10647	{
10648	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysVirtApic, &HCPhysVirtApic);
10649	AssertMsgRCReturn(rc, ("Failed to get host-physical address for virtual-APIC page at %#RGp\n", GCPhysVirtApic), rc);
10650
10651	/** @todo Handle proper releasing of page-mapping lock later. */
10652	PGMPhysReleasePageMappingLock(pVCpu->CTX_SUFF(pVM), &PgLockVirtApic);
10653	}
10654	else
10655	return rc;
10656
10657	u32TprThreshold = pVmcsNstGst->u32TprThreshold;
10658	}
10659	else
10660	{
10661	HCPhysVirtApic = 0;
10662	u32TprThreshold = 0;
10663
10664	/*
10665	* We must make sure CR8 reads/write must cause VM-exits when TPR shadowing is not
10666	* used by the nested hypervisor. Preventing MMIO accesses to the physical APIC will
10667	* be taken care of by EPT/shadow paging.
10668	*/
10669	if (pVM->hmr0.s.fAllow64BitGuests)
10670	u32ProcCtls \|= VMX_PROC_CTLS_CR8_STORE_EXIT
10671	\| VMX_PROC_CTLS_CR8_LOAD_EXIT;
10672	}
10673
10674	/*
10675	* Validate basic assumptions.
10676	*/
10677	PVMXVMCSINFO pVmcsInfoNstGst = &pVCpu->hmr0.s.vmx.VmcsInfoNstGst;
10678	Assert(pVM->hmr0.s.vmx.fUnrestrictedGuest);
10679	Assert(g_HmMsrs.u.vmx.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS);
10680	Assert(hmGetVmxActiveVmcsInfo(pVCpu) == pVmcsInfoNstGst);
10681
10682	/*
10683	* Commit it to the nested-guest VMCS.
10684	*/
10685	int rc = VINF_SUCCESS;
10686	if (pVmcsInfoNstGst->u32PinCtls != u32PinCtls)
10687	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, u32PinCtls);
10688	if (pVmcsInfoNstGst->u32ProcCtls != u32ProcCtls)
10689	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, u32ProcCtls);
10690	if (pVmcsInfoNstGst->u32ProcCtls2 != u32ProcCtls2)
10691	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, u32ProcCtls2);
10692	if (pVmcsInfoNstGst->u32XcptBitmap != u32XcptBitmap)
10693	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
10694	if (pVmcsInfoNstGst->u64Cr0Mask != u64Cr0Mask)
10695	rc \|= VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, u64Cr0Mask);
10696	if (pVmcsInfoNstGst->u64Cr4Mask != u64Cr4Mask)
10697	rc \|= VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, u64Cr4Mask);
10698	if (pVmcsInfoNstGst->u32XcptPFMask != u32XcptPFMask)
10699	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, u32XcptPFMask);
10700	if (pVmcsInfoNstGst->u32XcptPFMatch != u32XcptPFMatch)
10701	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, u32XcptPFMatch);
10702	if ( !(u32ProcCtls & VMX_PROC_CTLS_PAUSE_EXIT)
10703	&& (u32ProcCtls2 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT))
10704	{
10705	Assert(g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT);
10706	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, cPleGapTicks);
10707	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, cPleWindowTicks);
10708	}
10709	if (u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
10710	{
10711	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
10712	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_VIRT_APIC_PAGEADDR_FULL, HCPhysVirtApic);
10713	}
10714	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10715	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, HCPhysApicAccess);
10716	rc \|= VMXWriteVmcsNw(VMX_VMCS_GUEST_PENDING_DEBUG_XCPTS, uPendingDbgXcpts);
10717	AssertRC(rc);
10718
10719	/*
10720	* Update the nested-guest VMCS cache.
10721	*/
10722	pVmcsInfoNstGst->u32PinCtls = u32PinCtls;
10723	pVmcsInfoNstGst->u32ProcCtls = u32ProcCtls;
10724	pVmcsInfoNstGst->u32ProcCtls2 = u32ProcCtls2;
10725	pVmcsInfoNstGst->u32XcptBitmap = u32XcptBitmap;
10726	pVmcsInfoNstGst->u64Cr0Mask = u64Cr0Mask;
10727	pVmcsInfoNstGst->u64Cr4Mask = u64Cr4Mask;
10728	pVmcsInfoNstGst->u32XcptPFMask = u32XcptPFMask;
10729	pVmcsInfoNstGst->u32XcptPFMatch = u32XcptPFMatch;
10730	pVmcsInfoNstGst->HCPhysVirtApic = HCPhysVirtApic;
10731
10732	/*
10733	* We need to flush the TLB if we are switching the APIC-access page address.
10734	* See Intel spec. 28.3.3.4 "Guidelines for Use of the INVEPT Instruction".
10735	*/
10736	if (u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10737	pVCpu->hm.s.vmx.fSwitchedNstGstFlushTlb = true;
10738
10739	/*
10740	* MSR bitmap.
10741	*
10742	* The MSR bitmap address has already been initialized while setting up the nested-guest
10743	* VMCS, here we need to merge the MSR bitmaps.
10744	*/
10745	if (u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
10746	hmR0VmxMergeMsrBitmapNested(pVCpu, pVmcsInfoNstGst, pVmcsInfoGst);
10747
10748	return VINF_SUCCESS;
10749	}
10750	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
10751
10752
10753	/**
10754	* Does the preparations before executing guest code in VT-x.
10755	*
10756	* This may cause longjmps to ring-3 and may even result in rescheduling to the
10757	* recompiler/IEM. We must be cautious what we do here regarding committing
10758	* guest-state information into the VMCS assuming we assuredly execute the
10759	* guest in VT-x mode.
10760	*
10761	* If we fall back to the recompiler/IEM after updating the VMCS and clearing
10762	* the common-state (TRPM/forceflags), we must undo those changes so that the
10763	* recompiler/IEM can (and should) use them when it resumes guest execution.
10764	* Otherwise such operations must be done when we can no longer exit to ring-3.
10765	*
10766	* @returns Strict VBox status code (i.e. informational status codes too).
10767	* @retval VINF_SUCCESS if we can proceed with running the guest, interrupts
10768	* have been disabled.
10769	* @retval VINF_VMX_VMEXIT if a nested-guest VM-exit occurs (e.g., while evaluating
10770	* pending events).
10771	* @retval VINF_EM_RESET if a triple-fault occurs while injecting a
10772	* double-fault into the guest.
10773	* @retval VINF_EM_DBG_STEPPED if @a fStepping is true and an event was
10774	* dispatched directly.
10775	* @retval VINF_* scheduling changes, we have to go back to ring-3.
10776	*
10777	* @param pVCpu The cross context virtual CPU structure.
10778	* @param pVmxTransient The VMX-transient structure.
10779	* @param fStepping Whether we are single-stepping the guest in the
10780	* hypervisor debugger. Makes us ignore some of the reasons
10781	* for returning to ring-3, and return VINF_EM_DBG_STEPPED
10782	* if event dispatching took place.
10783	*/
10784	static VBOXSTRICTRC hmR0VmxPreRunGuest(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, bool fStepping)
10785	{
10786	Assert(VMMRZCallRing3IsEnabled(pVCpu));
10787
10788	Log4Func(("fIsNested=%RTbool fStepping=%RTbool\n", pVmxTransient->fIsNestedGuest, fStepping));
10789
10790	#ifdef VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM
10791	if (pVmxTransient->fIsNestedGuest)
10792	{
10793	RT_NOREF2(pVCpu, fStepping);
10794	Log2Func(("Rescheduling to IEM due to nested-hwvirt or forced IEM exec -> VINF_EM_RESCHEDULE_REM\n"));
10795	return VINF_EM_RESCHEDULE_REM;
10796	}
10797	#endif
10798
10799	/*
10800	* Check and process force flag actions, some of which might require us to go back to ring-3.
10801	*/
10802	VBOXSTRICTRC rcStrict = hmR0VmxCheckForceFlags(pVCpu, pVmxTransient, fStepping);
10803	if (rcStrict == VINF_SUCCESS)
10804	{
10805	/* FFs don't get set all the time. */
10806	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10807	if ( pVmxTransient->fIsNestedGuest
10808	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
10809	{
10810	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
10811	return VINF_VMX_VMEXIT;
10812	}
10813	#endif
10814	}
10815	else
10816	return rcStrict;
10817
10818	/*
10819	* Virtualize memory-mapped accesses to the physical APIC (may take locks).
10820	*/
10821	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
10822	if ( !pVCpu->hm.s.vmx.u64GstMsrApicBase
10823	&& (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
10824	&& PDMHasApic(pVM))
10825	{
10826	int rc = hmR0VmxMapHCApicAccessPage(pVCpu);
10827	AssertRCReturn(rc, rc);
10828	}
10829
10830	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10831	/*
10832	* Merge guest VMCS controls with the nested-guest VMCS controls.
10833	*
10834	* Even if we have not executed the guest prior to this (e.g. when resuming from a
10835	* saved state), we should be okay with merging controls as we initialize the
10836	* guest VMCS controls as part of VM setup phase.
10837	*/
10838	if ( pVmxTransient->fIsNestedGuest
10839	&& !pVCpu->hm.s.vmx.fMergedNstGstCtls)
10840	{
10841	int rc = hmR0VmxMergeVmcsNested(pVCpu);
10842	AssertRCReturn(rc, rc);
10843	pVCpu->hm.s.vmx.fMergedNstGstCtls = true;
10844	}
10845	#endif
10846
10847	/*
10848	* Evaluate events to be injected into the guest.
10849	*
10850	* Events in TRPM can be injected without inspecting the guest state.
10851	* If any new events (interrupts/NMI) are pending currently, we try to set up the
10852	* guest to cause a VM-exit the next time they are ready to receive the event.
10853	*/
10854	if (TRPMHasTrap(pVCpu))
10855	hmR0VmxTrpmTrapToPendingEvent(pVCpu);
10856
10857	uint32_t fIntrState;
10858	rcStrict = hmR0VmxEvaluatePendingEvent(pVCpu, pVmxTransient, &fIntrState);
10859
10860	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10861	/*
10862	* While evaluating pending events if something failed (unlikely) or if we were
10863	* preparing to run a nested-guest but performed a nested-guest VM-exit, we should bail.
10864	*/
10865	if (rcStrict != VINF_SUCCESS)
10866	return rcStrict;
10867	if ( pVmxTransient->fIsNestedGuest
10868	&& !CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
10869	{
10870	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
10871	return VINF_VMX_VMEXIT;
10872	}
10873	#else
10874	Assert(rcStrict == VINF_SUCCESS);
10875	#endif
10876
10877	/*
10878	* Event injection may take locks (currently the PGM lock for real-on-v86 case) and thus
10879	* needs to be done with longjmps or interrupts + preemption enabled. Event injection might
10880	* also result in triple-faulting the VM.
10881	*
10882	* With nested-guests, the above does not apply since unrestricted guest execution is a
10883	* requirement. Regardless, we do this here to avoid duplicating code elsewhere.
10884	*/
10885	rcStrict = hmR0VmxInjectPendingEvent(pVCpu, pVmxTransient, fIntrState, fStepping);
10886	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
10887	{ /* likely */ }
10888	else
10889	{
10890	AssertMsg(rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
10891	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
10892	return rcStrict;
10893	}
10894
10895	/*
10896	* A longjump might result in importing CR3 even for VM-exits that don't necessarily
10897	* import CR3 themselves. We will need to update them here, as even as late as the above
10898	* hmR0VmxInjectPendingEvent() call may lazily import guest-CPU state on demand causing
10899	* the below force flags to be set.
10900	*/
10901	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
10902	{
10903	Assert(!(ASMAtomicUoReadU64(&pVCpu->cpum.GstCtx.fExtrn) & CPUMCTX_EXTRN_CR3));
10904	int rc2 = PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu), false /* fPdpesMapped */);
10905	AssertMsgReturn(rc2 == VINF_SUCCESS \|\| rc2 == VINF_PGM_SYNC_CR3,
10906	("%Rrc\n", rc2), RT_FAILURE_NP(rc2) ? rc2 : VERR_IPE_UNEXPECTED_INFO_STATUS);
10907	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
10908	}
10909
10910	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10911	/* Paranoia. */
10912	Assert(!pVmxTransient->fIsNestedGuest \|\| CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
10913	#endif
10914
10915	/*
10916	* No longjmps to ring-3 from this point on!!!
10917	* Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
10918	* This also disables flushing of the R0-logger instance (if any).
10919	*/
10920	VMMRZCallRing3Disable(pVCpu);
10921
10922	/*
10923	* Export the guest state bits.
10924	*
10925	* We cannot perform longjmps while loading the guest state because we do not preserve the
10926	* host/guest state (although the VMCS will be preserved) across longjmps which can cause
10927	* CPU migration.
10928	*
10929	* If we are injecting events to a real-on-v86 mode guest, we would have updated RIP and some segment
10930	* registers. Hence, exporting of the guest state needs to be done -after- injection of events.
10931	*/
10932	rcStrict = hmR0VmxExportGuestStateOptimal(pVCpu, pVmxTransient);
10933	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
10934	{ /* likely */ }
10935	else
10936	{
10937	VMMRZCallRing3Enable(pVCpu);
10938	return rcStrict;
10939	}
10940
10941	/*
10942	* We disable interrupts so that we don't miss any interrupts that would flag preemption
10943	* (IPI/timers etc.) when thread-context hooks aren't used and we've been running with
10944	* preemption disabled for a while. Since this is purely to aid the
10945	* RTThreadPreemptIsPending() code, it doesn't matter that it may temporarily reenable and
10946	* disable interrupt on NT.
10947	*
10948	* We need to check for force-flags that could've possible been altered since we last
10949	* checked them (e.g. by PDMGetInterrupt() leaving the PDM critical section,
10950	* see @bugref{6398}).
10951	*
10952	* We also check a couple of other force-flags as a last opportunity to get the EMT back
10953	* to ring-3 before executing guest code.
10954	*/
10955	pVmxTransient->fEFlags = ASMIntDisableFlags();
10956
10957	if ( ( !VM_FF_IS_ANY_SET(pVM, VM_FF_EMT_RENDEZVOUS \| VM_FF_TM_VIRTUAL_SYNC)
10958	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
10959	\|\| ( fStepping /* Optimized for the non-stepping case, so a bit of unnecessary work when stepping. */
10960	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK & ~(VMCPU_FF_TIMER \| VMCPU_FF_PDM_CRITSECT))) )
10961	{
10962	if (!RTThreadPreemptIsPending(NIL_RTTHREAD))
10963	{
10964	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
10965	/*
10966	* If we are executing a nested-guest make sure that we should intercept subsequent
10967	* events. The one we are injecting might be part of VM-entry. This is mainly to keep
10968	* the VM-exit instruction emulation happy.
10969	*/
10970	if (pVmxTransient->fIsNestedGuest)
10971	CPUMSetGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx, true);
10972	#endif
10973
10974	/*
10975	* We've injected any pending events. This is really the point of no return (to ring-3).
10976	*
10977	* Note! The caller expects to continue with interrupts & longjmps disabled on successful
10978	* returns from this function, so do -not- enable them here.
10979	*/
10980	pVCpu->hm.s.Event.fPending = false;
10981	return VINF_SUCCESS;
10982	}
10983
10984	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPendingHostIrq);
10985	rcStrict = VINF_EM_RAW_INTERRUPT;
10986	}
10987	else
10988	{
10989	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
10990	rcStrict = VINF_EM_RAW_TO_R3;
10991	}
10992
10993	ASMSetFlags(pVmxTransient->fEFlags);
10994	VMMRZCallRing3Enable(pVCpu);
10995
10996	return rcStrict;
10997	}
10998
10999
11000	/**
11001	* Final preparations before executing guest code using hardware-assisted VMX.
11002	*
11003	* We can no longer get preempted to a different host CPU and there are no returns
11004	* to ring-3. We ignore any errors that may happen from this point (e.g. VMWRITE
11005	* failures), this function is not intended to fail sans unrecoverable hardware
11006	* errors.
11007	*
11008	* @param pVCpu The cross context virtual CPU structure.
11009	* @param pVmxTransient The VMX-transient structure.
11010	*
11011	* @remarks Called with preemption disabled.
11012	* @remarks No-long-jump zone!!!
11013	*/
11014	static void hmR0VmxPreRunGuestCommitted(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
11015	{
11016	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
11017	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
11018	Assert(!pVCpu->hm.s.Event.fPending);
11019
11020	/*
11021	* Indicate start of guest execution and where poking EMT out of guest-context is recognized.
11022	*/
11023	VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
11024	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC);
11025
11026	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
11027	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11028	PHMPHYSCPU pHostCpu = hmR0GetCurrentCpu();
11029	RTCPUID const idCurrentCpu = pHostCpu->idCpu;
11030
11031	if (!CPUMIsGuestFPUStateActive(pVCpu))
11032	{
11033	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestFpuState, x);
11034	if (CPUMR0LoadGuestFPU(pVM, pVCpu) == VINF_CPUM_HOST_CR0_MODIFIED)
11035	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_HOST_CONTEXT;
11036	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestFpuState, x);
11037	STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadGuestFpu);
11038	}
11039
11040	/*
11041	* Re-export the host state bits as we may've been preempted (only happens when
11042	* thread-context hooks are used or when the VM start function changes) or if
11043	* the host CR0 is modified while loading the guest FPU state above.
11044	*
11045	* The 64-on-32 switcher saves the (64-bit) host state into the VMCS and if we
11046	* changed the switcher back to 32-bit, we must save the 32-bit host state here,
11047	* see @bugref{8432}.
11048	*
11049	* This may also happen when switching to/from a nested-guest VMCS without leaving
11050	* ring-0.
11051	*/
11052	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT)
11053	{
11054	hmR0VmxExportHostState(pVCpu);
11055	STAM_COUNTER_INC(&pVCpu->hm.s.StatExportHostState);
11056	}
11057	Assert(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_HOST_CONTEXT));
11058
11059	/*
11060	* Export the state shared between host and guest (FPU, debug, lazy MSRs).
11061	*/
11062	if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE)
11063	hmR0VmxExportSharedState(pVCpu, pVmxTransient);
11064	AssertMsg(!pVCpu->hm.s.fCtxChanged, ("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
11065
11066	/*
11067	* Store status of the shared guest/host debug state at the time of VM-entry.
11068	*/
11069	pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
11070	pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
11071
11072	/*
11073	* Always cache the TPR-shadow if the virtual-APIC page exists, thereby skipping
11074	* more than one conditional check. The post-run side of our code shall determine
11075	* if it needs to sync. the virtual APIC TPR with the TPR-shadow.
11076	*/
11077	if (pVmcsInfo->pbVirtApic)
11078	pVmxTransient->u8GuestTpr = pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR];
11079
11080	/*
11081	* Update the host MSRs values in the VM-exit MSR-load area.
11082	*/
11083	if (!pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs)
11084	{
11085	if (pVmcsInfo->cExitMsrLoad > 0)
11086	hmR0VmxUpdateAutoLoadHostMsrs(pVCpu, pVmcsInfo);
11087	pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs = true;
11088	}
11089
11090	/*
11091	* Evaluate if we need to intercept guest RDTSC/P accesses. Set up the
11092	* VMX-preemption timer based on the next virtual sync clock deadline.
11093	*/
11094	if ( !pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer
11095	\|\| idCurrentCpu != pVCpu->hmr0.s.idLastCpu)
11096	{
11097	hmR0VmxUpdateTscOffsettingAndPreemptTimer(pVCpu, pVmxTransient, idCurrentCpu);
11098	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = true;
11099	}
11100
11101	/* Record statistics of how often we use TSC offsetting as opposed to intercepting RDTSC/P. */
11102	bool const fIsRdtscIntercepted = RT_BOOL(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_RDTSC_EXIT);
11103	if (!fIsRdtscIntercepted)
11104	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
11105	else
11106	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
11107
11108	ASMAtomicUoWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB flushing, set this across the world switch. */
11109	hmR0VmxFlushTaggedTlb(pHostCpu, pVCpu, pVmcsInfo); /* Invalidate the appropriate guest entries from the TLB. */
11110	Assert(idCurrentCpu == pVCpu->hmr0.s.idLastCpu);
11111	pVCpu->hm.s.vmx.LastError.idCurrentCpu = idCurrentCpu; /* Record the error reporting info. with the current host CPU. */
11112	pVmcsInfo->idHostCpuState = idCurrentCpu; /* Record the CPU for which the host-state has been exported. */
11113	pVmcsInfo->idHostCpuExec = idCurrentCpu; /* Record the CPU on which we shall execute. */
11114
11115	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
11116
11117	TMNotifyStartOfExecution(pVM, pVCpu); /* Notify TM to resume its clocks when TSC is tied to execution,
11118	as we're about to start executing the guest. */
11119
11120	/*
11121	* Load the guest TSC_AUX MSR when we are not intercepting RDTSCP.
11122	*
11123	* This is done this late as updating the TSC offsetting/preemption timer above
11124	* figures out if we can skip intercepting RDTSCP by calculating the number of
11125	* host CPU ticks till the next virtual sync deadline (for the dynamic case).
11126	*/
11127	if ( (pVmcsInfo->u32ProcCtls2 & VMX_PROC_CTLS2_RDTSCP)
11128	&& !fIsRdtscIntercepted)
11129	{
11130	hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_TSC_AUX);
11131
11132	/* NB: Because we call hmR0VmxAddAutoLoadStoreMsr with fUpdateHostMsr=true,
11133	it's safe even after hmR0VmxUpdateAutoLoadHostMsrs has already been done. */
11134	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_TSC_AUX, CPUMGetGuestTscAux(pVCpu),
11135	true /* fSetReadWrite /, true / fUpdateHostMsr */);
11136	AssertRC(rc);
11137	Assert(!pVmxTransient->fRemoveTscAuxMsr);
11138	pVmxTransient->fRemoveTscAuxMsr = true;
11139	}
11140
11141	#ifdef VBOX_STRICT
11142	Assert(pVCpu->hmr0.s.vmx.fUpdatedHostAutoMsrs);
11143	hmR0VmxCheckAutoLoadStoreMsrs(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest);
11144	hmR0VmxCheckHostEferMsr(pVmcsInfo);
11145	AssertRC(hmR0VmxCheckCachedVmcsCtls(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest));
11146	#endif
11147
11148	#ifdef HMVMX_ALWAYS_CHECK_GUEST_STATE
11149	/** @todo r=ramshankar: We can now probably use iemVmxVmentryCheckGuestState here.
11150	* Add a PVMXMSRS parameter to it, so that IEM can look at the host MSRs,
11151	* see @bugref{9180#c54}. */
11152	uint32_t const uInvalidReason = hmR0VmxCheckGuestState(pVCpu, pVmcsInfo);
11153	if (uInvalidReason != VMX_IGS_REASON_NOT_FOUND)
11154	Log4(("hmR0VmxCheckGuestState returned %#x\n", uInvalidReason));
11155	#endif
11156	}
11157
11158
11159	/**
11160	* First C routine invoked after running guest code using hardware-assisted VMX.
11161	*
11162	* @param pVCpu The cross context virtual CPU structure.
11163	* @param pVmxTransient The VMX-transient structure.
11164	* @param rcVMRun Return code of VMLAUNCH/VMRESUME.
11165	*
11166	* @remarks Called with interrupts disabled, and returns with interrupts enabled!
11167	*
11168	* @remarks No-long-jump zone!!! This function will however re-enable longjmps
11169	* unconditionally when it is safe to do so.
11170	*/
11171	static void hmR0VmxPostRunGuest(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, int rcVMRun)
11172	{
11173	ASMAtomicUoWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB flushing. */
11174	ASMAtomicIncU32(&pVCpu->hmr0.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for EMT poking. */
11175	pVCpu->hm.s.fCtxChanged = 0; /* Exits/longjmps to ring-3 requires saving the guest state. */
11176	pVmxTransient->fVmcsFieldsRead = 0; /* Transient fields need to be read from the VMCS. */
11177	pVmxTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
11178	pVmxTransient->fVectoringDoublePF = false; /* Vectoring double page-fault needs to be determined later. */
11179
11180	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11181	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_RDTSC_EXIT))
11182	{
11183	uint64_t uGstTsc;
11184	if (!pVmxTransient->fIsNestedGuest)
11185	uGstTsc = pVCpu->hmr0.s.uTscExit + pVmcsInfo->u64TscOffset;
11186	else
11187	{
11188	uint64_t const uNstGstTsc = pVCpu->hmr0.s.uTscExit + pVmcsInfo->u64TscOffset;
11189	uGstTsc = CPUMRemoveNestedGuestTscOffset(pVCpu, uNstGstTsc);
11190	}
11191	TMCpuTickSetLastSeen(pVCpu, uGstTsc); /* Update TM with the guest TSC. */
11192	}
11193
11194	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatPreExit, x);
11195	TMNotifyEndOfExecution(pVCpu->CTX_SUFF(pVM), pVCpu, pVCpu->hmr0.s.uTscExit); /* Notify TM that the guest is no longer running. */
11196	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
11197
11198	pVCpu->hmr0.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_REQUIRED; /* Some host state messed up by VMX needs restoring. */
11199	pVmcsInfo->fVmcsState \|= VMX_V_VMCS_LAUNCH_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
11200	#ifdef VBOX_STRICT
11201	hmR0VmxCheckHostEferMsr(pVmcsInfo); /* Verify that the host EFER MSR wasn't modified. */
11202	#endif
11203	Assert(!ASMIntAreEnabled());
11204	ASMSetFlags(pVmxTransient->fEFlags); /* Enable interrupts. */
11205	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
11206
11207	#ifdef HMVMX_ALWAYS_CLEAN_TRANSIENT
11208	/*
11209	* Clean all the VMCS fields in the transient structure before reading
11210	* anything from the VMCS.
11211	*/
11212	pVmxTransient->uExitReason = 0;
11213	pVmxTransient->uExitIntErrorCode = 0;
11214	pVmxTransient->uExitQual = 0;
11215	pVmxTransient->uGuestLinearAddr = 0;
11216	pVmxTransient->uExitIntInfo = 0;
11217	pVmxTransient->cbExitInstr = 0;
11218	pVmxTransient->ExitInstrInfo.u = 0;
11219	pVmxTransient->uEntryIntInfo = 0;
11220	pVmxTransient->uEntryXcptErrorCode = 0;
11221	pVmxTransient->cbEntryInstr = 0;
11222	pVmxTransient->uIdtVectoringInfo = 0;
11223	pVmxTransient->uIdtVectoringErrorCode = 0;
11224	#endif
11225
11226	/*
11227	* Save the basic VM-exit reason and check if the VM-entry failed.
11228	* See Intel spec. 24.9.1 "Basic VM-exit Information".
11229	*/
11230	uint32_t uExitReason;
11231	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
11232	AssertRC(rc);
11233	pVmxTransient->uExitReason = VMX_EXIT_REASON_BASIC(uExitReason);
11234	pVmxTransient->fVMEntryFailed = VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason);
11235
11236	/*
11237	* Log the VM-exit before logging anything else as otherwise it might be a
11238	* tad confusing what happens before and after the world-switch.
11239	*/
11240	HMVMX_LOG_EXIT(pVCpu, uExitReason);
11241
11242	/*
11243	* Remove the TSC_AUX MSR from the auto-load/store MSR area and reset any MSR
11244	* bitmap permissions, if it was added before VM-entry.
11245	*/
11246	if (pVmxTransient->fRemoveTscAuxMsr)
11247	{
11248	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, pVmxTransient, MSR_K8_TSC_AUX);
11249	pVmxTransient->fRemoveTscAuxMsr = false;
11250	}
11251
11252	/*
11253	* Check if VMLAUNCH/VMRESUME succeeded.
11254	* If this failed, we cause a guru meditation and cease further execution.
11255	*
11256	* However, if we are executing a nested-guest we might fail if we use the
11257	* fast path rather than fully emulating VMLAUNCH/VMRESUME instruction in IEM.
11258	*/
11259	if (RT_LIKELY(rcVMRun == VINF_SUCCESS))
11260	{
11261	/*
11262	* Update the VM-exit history array here even if the VM-entry failed due to:
11263	* - Invalid guest state.
11264	* - MSR loading.
11265	* - Machine-check event.
11266	*
11267	* In any of the above cases we will still have a "valid" VM-exit reason
11268	* despite @a fVMEntryFailed being false.
11269	*
11270	* See Intel spec. 26.7 "VM-Entry failures during or after loading guest state".
11271	*
11272	* Note! We don't have CS or RIP at this point. Will probably address that later
11273	* by amending the history entry added here.
11274	*/
11275	EMHistoryAddExit(pVCpu, EMEXIT_MAKE_FT(EMEXIT_F_KIND_VMX, pVmxTransient->uExitReason & EMEXIT_F_TYPE_MASK),
11276	UINT64_MAX, pVCpu->hmr0.s.uTscExit);
11277
11278	if (RT_LIKELY(!pVmxTransient->fVMEntryFailed))
11279	{
11280	VMMRZCallRing3Enable(pVCpu);
11281	Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
11282
11283	#ifdef HMVMX_ALWAYS_SAVE_RO_GUEST_STATE
11284	hmR0VmxReadAllRoFieldsVmcs(pVmxTransient);
11285	#endif
11286
11287	/*
11288	* Import the guest-interruptibility state always as we need it while evaluating
11289	* injecting events on re-entry.
11290	*
11291	* We don't import CR0 (when unrestricted guest execution is unavailable) despite
11292	* checking for real-mode while exporting the state because all bits that cause
11293	* mode changes wrt CR0 are intercepted.
11294	*/
11295	uint64_t const fImportMask = CPUMCTX_EXTRN_HM_VMX_INT_STATE
11296	#if defined(HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE) \|\| defined(HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE)
11297	\| HMVMX_CPUMCTX_EXTRN_ALL
11298	#elif defined(HMVMX_ALWAYS_SAVE_GUEST_RFLAGS)
11299	\| CPUMCTX_EXTRN_RFLAGS
11300	#endif
11301	;
11302	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImportMask);
11303	AssertRC(rc);
11304
11305	/*
11306	* Sync the TPR shadow with our APIC state.
11307	*/
11308	if ( !pVmxTransient->fIsNestedGuest
11309	&& (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW))
11310	{
11311	Assert(pVmcsInfo->pbVirtApic);
11312	if (pVmxTransient->u8GuestTpr != pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR])
11313	{
11314	rc = APICSetTpr(pVCpu, pVmcsInfo->pbVirtApic[XAPIC_OFF_TPR]);
11315	AssertRC(rc);
11316	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
11317	}
11318	}
11319
11320	Assert(VMMRZCallRing3IsEnabled(pVCpu));
11321	Assert( pVmxTransient->fWasGuestDebugStateActive == false
11322	\|\| pVmxTransient->fWasHyperDebugStateActive == false);
11323	return;
11324	}
11325	}
11326	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11327	else if (pVmxTransient->fIsNestedGuest)
11328	AssertMsgFailed(("VMLAUNCH/VMRESUME failed but shouldn't happen when VMLAUNCH/VMRESUME was emulated in IEM!\n"));
11329	#endif
11330	else
11331	Log4Func(("VM-entry failure: rcVMRun=%Rrc fVMEntryFailed=%RTbool\n", rcVMRun, pVmxTransient->fVMEntryFailed));
11332
11333	VMMRZCallRing3Enable(pVCpu);
11334	}
11335
11336
11337	/**
11338	* Runs the guest code using hardware-assisted VMX the normal way.
11339	*
11340	* @returns VBox status code.
11341	* @param pVCpu The cross context virtual CPU structure.
11342	* @param pcLoops Pointer to the number of executed loops.
11343	*/
11344	static VBOXSTRICTRC hmR0VmxRunGuestCodeNormal(PVMCPUCC pVCpu, uint32_t *pcLoops)
11345	{
11346	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops;
11347	Assert(pcLoops);
11348	Assert(*pcLoops <= cMaxResumeLoops);
11349	Assert(!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
11350
11351	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11352	/*
11353	* Switch to the guest VMCS as we may have transitioned from executing the nested-guest
11354	* without leaving ring-0. Otherwise, if we came from ring-3 we would have loaded the
11355	* guest VMCS while entering the VMX ring-0 session.
11356	*/
11357	if (pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs)
11358	{
11359	int rc = hmR0VmxSwitchToGstOrNstGstVmcs(pVCpu, false /* fSwitchToNstGstVmcs */);
11360	if (RT_SUCCESS(rc))
11361	{ /* likely */ }
11362	else
11363	{
11364	LogRelFunc(("Failed to switch to the guest VMCS. rc=%Rrc\n", rc));
11365	return rc;
11366	}
11367	}
11368	#endif
11369
11370	VMXTRANSIENT VmxTransient;
11371	RT_ZERO(VmxTransient);
11372	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
11373
11374	/* Paranoia. */
11375	Assert(VmxTransient.pVmcsInfo == &pVCpu->hmr0.s.vmx.VmcsInfo);
11376
11377	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
11378	for (;;)
11379	{
11380	Assert(!HMR0SuspendPending());
11381	HMVMX_ASSERT_CPU_SAFE(pVCpu);
11382	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
11383
11384	/*
11385	* Preparatory work for running nested-guest code, this may force us to
11386	* return to ring-3.
11387	*
11388	* Warning! This bugger disables interrupts on VINF_SUCCESS!
11389	*/
11390	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, false /* fStepping */);
11391	if (rcStrict != VINF_SUCCESS)
11392	break;
11393
11394	/* Interrupts are disabled at this point! */
11395	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
11396	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
11397	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
11398	/* Interrupts are re-enabled at this point! */
11399
11400	/*
11401	* Check for errors with running the VM (VMLAUNCH/VMRESUME).
11402	*/
11403	if (RT_SUCCESS(rcRun))
11404	{ /* very likely */ }
11405	else
11406	{
11407	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
11408	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
11409	return rcRun;
11410	}
11411
11412	/*
11413	* Profile the VM-exit.
11414	*/
11415	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
11416	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
11417	STAM_COUNTER_INC(&pVCpu->hm.s.aStatExitReason[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
11418	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
11419	HMVMX_START_EXIT_DISPATCH_PROF();
11420
11421	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
11422
11423	/*
11424	* Handle the VM-exit.
11425	*/
11426	#ifdef HMVMX_USE_FUNCTION_TABLE
11427	rcStrict = g_aVMExitHandlers[VmxTransient.uExitReason].pfn(pVCpu, &VmxTransient);
11428	#else
11429	rcStrict = hmR0VmxHandleExit(pVCpu, &VmxTransient);
11430	#endif
11431	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
11432	if (rcStrict == VINF_SUCCESS)
11433	{
11434	if (++(*pcLoops) <= cMaxResumeLoops)
11435	continue;
11436	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
11437	rcStrict = VINF_EM_RAW_INTERRUPT;
11438	}
11439	break;
11440	}
11441
11442	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
11443	return rcStrict;
11444	}
11445
11446
11447	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
11448	/**
11449	* Runs the nested-guest code using hardware-assisted VMX.
11450	*
11451	* @returns VBox status code.
11452	* @param pVCpu The cross context virtual CPU structure.
11453	* @param pcLoops Pointer to the number of executed loops.
11454	*
11455	* @sa hmR0VmxRunGuestCodeNormal.
11456	*/
11457	static VBOXSTRICTRC hmR0VmxRunGuestCodeNested(PVMCPUCC pVCpu, uint32_t *pcLoops)
11458	{
11459	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops;
11460	Assert(pcLoops);
11461	Assert(*pcLoops <= cMaxResumeLoops);
11462	Assert(CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx));
11463
11464	/*
11465	* Switch to the nested-guest VMCS as we may have transitioned from executing the
11466	* guest without leaving ring-0. Otherwise, if we came from ring-3 we would have
11467	* loaded the nested-guest VMCS while entering the VMX ring-0 session.
11468	*/
11469	if (!pVCpu->hmr0.s.vmx.fSwitchedToNstGstVmcs)
11470	{
11471	int rc = hmR0VmxSwitchToGstOrNstGstVmcs(pVCpu, true /* fSwitchToNstGstVmcs */);
11472	if (RT_SUCCESS(rc))
11473	{ /* likely */ }
11474	else
11475	{
11476	LogRelFunc(("Failed to switch to the nested-guest VMCS. rc=%Rrc\n", rc));
11477	return rc;
11478	}
11479	}
11480
11481	VMXTRANSIENT VmxTransient;
11482	RT_ZERO(VmxTransient);
11483	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
11484	VmxTransient.fIsNestedGuest = true;
11485
11486	/* Paranoia. */
11487	Assert(VmxTransient.pVmcsInfo == &pVCpu->hmr0.s.vmx.VmcsInfoNstGst);
11488
11489	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
11490	for (;;)
11491	{
11492	Assert(!HMR0SuspendPending());
11493	HMVMX_ASSERT_CPU_SAFE(pVCpu);
11494	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
11495
11496	/*
11497	* Preparatory work for running guest code, this may force us to
11498	* return to ring-3.
11499	*
11500	* Warning! This bugger disables interrupts on VINF_SUCCESS!
11501	*/
11502	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, false /* fStepping */);
11503	if (rcStrict != VINF_SUCCESS)
11504	break;
11505
11506	/* Interrupts are disabled at this point! */
11507	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
11508	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
11509	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
11510	/* Interrupts are re-enabled at this point! */
11511
11512	/*
11513	* Check for errors with running the VM (VMLAUNCH/VMRESUME).
11514	*/
11515	if (RT_SUCCESS(rcRun))
11516	{ /* very likely */ }
11517	else
11518	{
11519	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
11520	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
11521	return rcRun;
11522	}
11523
11524	/*
11525	* Profile the VM-exit.
11526	*/
11527	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
11528	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
11529	STAM_COUNTER_INC(&pVCpu->hm.s.StatNestedExitAll);
11530	STAM_COUNTER_INC(&pVCpu->hm.s.aStatNestedExitReason[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
11531	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
11532	HMVMX_START_EXIT_DISPATCH_PROF();
11533
11534	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
11535
11536	/*
11537	* Handle the VM-exit.
11538	*/
11539	rcStrict = hmR0VmxHandleExitNested(pVCpu, &VmxTransient);
11540	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
11541	if (rcStrict == VINF_SUCCESS)
11542	{
11543	if (!CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
11544	{
11545	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
11546	rcStrict = VINF_VMX_VMEXIT;
11547	}
11548	else
11549	{
11550	if (++(*pcLoops) <= cMaxResumeLoops)
11551	continue;
11552	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
11553	rcStrict = VINF_EM_RAW_INTERRUPT;
11554	}
11555	}
11556	else
11557	Assert(rcStrict != VINF_VMX_VMEXIT);
11558	break;
11559	}
11560
11561	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
11562	return rcStrict;
11563	}
11564	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
11565
11566
11567	/** @name Execution loop for single stepping, DBGF events and expensive Dtrace
11568	* probes.
11569	*
11570	* The following few functions and associated structure contains the bloat
11571	* necessary for providing detailed debug events and dtrace probes as well as
11572	* reliable host side single stepping. This works on the principle of
11573	* "subclassing" the normal execution loop and workers. We replace the loop
11574	* method completely and override selected helpers to add necessary adjustments
11575	* to their core operation.
11576	*
11577	* The goal is to keep the "parent" code lean and mean, so as not to sacrifice
11578	* any performance for debug and analysis features.
11579	*
11580	* @{
11581	*/
11582
11583	/**
11584	* Transient per-VCPU debug state of VMCS and related info. we save/restore in
11585	* the debug run loop.
11586	*/
11587	typedef struct VMXRUNDBGSTATE
11588	{
11589	/** The RIP we started executing at. This is for detecting that we stepped. */
11590	uint64_t uRipStart;
11591	/** The CS we started executing with. */
11592	uint16_t uCsStart;
11593
11594	/** Whether we've actually modified the 1st execution control field. */
11595	bool fModifiedProcCtls : 1;
11596	/** Whether we've actually modified the 2nd execution control field. */
11597	bool fModifiedProcCtls2 : 1;
11598	/** Whether we've actually modified the exception bitmap. */
11599	bool fModifiedXcptBitmap : 1;
11600
11601	/** We desire the modified the CR0 mask to be cleared. */
11602	bool fClearCr0Mask : 1;
11603	/** We desire the modified the CR4 mask to be cleared. */
11604	bool fClearCr4Mask : 1;
11605	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC. */
11606	uint32_t fCpe1Extra;
11607	/** Stuff we do not want in VMX_VMCS32_CTRL_PROC_EXEC. */
11608	uint32_t fCpe1Unwanted;
11609	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC2. */
11610	uint32_t fCpe2Extra;
11611	/** Extra stuff we need in VMX_VMCS32_CTRL_EXCEPTION_BITMAP. */
11612	uint32_t bmXcptExtra;
11613	/** The sequence number of the Dtrace provider settings the state was
11614	* configured against. */
11615	uint32_t uDtraceSettingsSeqNo;
11616	/** VM-exits to check (one bit per VM-exit). */
11617	uint32_t bmExitsToCheck[3];
11618
11619	/** The initial VMX_VMCS32_CTRL_PROC_EXEC value (helps with restore). */
11620	uint32_t fProcCtlsInitial;
11621	/** The initial VMX_VMCS32_CTRL_PROC_EXEC2 value (helps with restore). */
11622	uint32_t fProcCtls2Initial;
11623	/** The initial VMX_VMCS32_CTRL_EXCEPTION_BITMAP value (helps with restore). */
11624	uint32_t bmXcptInitial;
11625	} VMXRUNDBGSTATE;
11626	AssertCompileMemberSize(VMXRUNDBGSTATE, bmExitsToCheck, (VMX_EXIT_MAX + 1 + 31) / 32 * 4);
11627	typedef VMXRUNDBGSTATE *PVMXRUNDBGSTATE;
11628
11629
11630	/**
11631	* Initializes the VMXRUNDBGSTATE structure.
11632	*
11633	* @param pVCpu The cross context virtual CPU structure of the
11634	* calling EMT.
11635	* @param pVmxTransient The VMX-transient structure.
11636	* @param pDbgState The debug state to initialize.
11637	*/
11638	static void hmR0VmxRunDebugStateInit(PVMCPUCC pVCpu, PCVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11639	{
11640	pDbgState->uRipStart = pVCpu->cpum.GstCtx.rip;
11641	pDbgState->uCsStart = pVCpu->cpum.GstCtx.cs.Sel;
11642
11643	pDbgState->fModifiedProcCtls = false;
11644	pDbgState->fModifiedProcCtls2 = false;
11645	pDbgState->fModifiedXcptBitmap = false;
11646	pDbgState->fClearCr0Mask = false;
11647	pDbgState->fClearCr4Mask = false;
11648	pDbgState->fCpe1Extra = 0;
11649	pDbgState->fCpe1Unwanted = 0;
11650	pDbgState->fCpe2Extra = 0;
11651	pDbgState->bmXcptExtra = 0;
11652	pDbgState->fProcCtlsInitial = pVmxTransient->pVmcsInfo->u32ProcCtls;
11653	pDbgState->fProcCtls2Initial = pVmxTransient->pVmcsInfo->u32ProcCtls2;
11654	pDbgState->bmXcptInitial = pVmxTransient->pVmcsInfo->u32XcptBitmap;
11655	}
11656
11657
11658	/**
11659	* Updates the VMSC fields with changes requested by @a pDbgState.
11660	*
11661	* This is performed after hmR0VmxPreRunGuestDebugStateUpdate as well
11662	* immediately before executing guest code, i.e. when interrupts are disabled.
11663	* We don't check status codes here as we cannot easily assert or return in the
11664	* latter case.
11665	*
11666	* @param pVCpu The cross context virtual CPU structure.
11667	* @param pVmxTransient The VMX-transient structure.
11668	* @param pDbgState The debug state.
11669	*/
11670	static void hmR0VmxPreRunGuestDebugStateApply(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11671	{
11672	/*
11673	* Ensure desired flags in VMCS control fields are set.
11674	* (Ignoring write failure here, as we're committed and it's just debug extras.)
11675	*
11676	* Note! We load the shadow CR0 & CR4 bits when we flag the clearing, so
11677	* there should be no stale data in pCtx at this point.
11678	*/
11679	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11680	if ( (pVmcsInfo->u32ProcCtls & pDbgState->fCpe1Extra) != pDbgState->fCpe1Extra
11681	\|\| (pVmcsInfo->u32ProcCtls & pDbgState->fCpe1Unwanted))
11682	{
11683	pVmcsInfo->u32ProcCtls \|= pDbgState->fCpe1Extra;
11684	pVmcsInfo->u32ProcCtls &= ~pDbgState->fCpe1Unwanted;
11685	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
11686	Log6Func(("VMX_VMCS32_CTRL_PROC_EXEC: %#RX32\n", pVmcsInfo->u32ProcCtls));
11687	pDbgState->fModifiedProcCtls = true;
11688	}
11689
11690	if ((pVmcsInfo->u32ProcCtls2 & pDbgState->fCpe2Extra) != pDbgState->fCpe2Extra)
11691	{
11692	pVmcsInfo->u32ProcCtls2 \|= pDbgState->fCpe2Extra;
11693	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pVmcsInfo->u32ProcCtls2);
11694	Log6Func(("VMX_VMCS32_CTRL_PROC_EXEC2: %#RX32\n", pVmcsInfo->u32ProcCtls2));
11695	pDbgState->fModifiedProcCtls2 = true;
11696	}
11697
11698	if ((pVmcsInfo->u32XcptBitmap & pDbgState->bmXcptExtra) != pDbgState->bmXcptExtra)
11699	{
11700	pVmcsInfo->u32XcptBitmap \|= pDbgState->bmXcptExtra;
11701	VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVmcsInfo->u32XcptBitmap);
11702	Log6Func(("VMX_VMCS32_CTRL_EXCEPTION_BITMAP: %#RX32\n", pVmcsInfo->u32XcptBitmap));
11703	pDbgState->fModifiedXcptBitmap = true;
11704	}
11705
11706	if (pDbgState->fClearCr0Mask && pVmcsInfo->u64Cr0Mask != 0)
11707	{
11708	pVmcsInfo->u64Cr0Mask = 0;
11709	VMXWriteVmcsNw(VMX_VMCS_CTRL_CR0_MASK, 0);
11710	Log6Func(("VMX_VMCS_CTRL_CR0_MASK: 0\n"));
11711	}
11712
11713	if (pDbgState->fClearCr4Mask && pVmcsInfo->u64Cr4Mask != 0)
11714	{
11715	pVmcsInfo->u64Cr4Mask = 0;
11716	VMXWriteVmcsNw(VMX_VMCS_CTRL_CR4_MASK, 0);
11717	Log6Func(("VMX_VMCS_CTRL_CR4_MASK: 0\n"));
11718	}
11719
11720	NOREF(pVCpu);
11721	}
11722
11723
11724	/**
11725	* Restores VMCS fields that were changed by hmR0VmxPreRunGuestDebugStateApply for
11726	* re-entry next time around.
11727	*
11728	* @returns Strict VBox status code (i.e. informational status codes too).
11729	* @param pVCpu The cross context virtual CPU structure.
11730	* @param pVmxTransient The VMX-transient structure.
11731	* @param pDbgState The debug state.
11732	* @param rcStrict The return code from executing the guest using single
11733	* stepping.
11734	*/
11735	static VBOXSTRICTRC hmR0VmxRunDebugStateRevert(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState,
11736	VBOXSTRICTRC rcStrict)
11737	{
11738	/*
11739	* Restore VM-exit control settings as we may not reenter this function the
11740	* next time around.
11741	*/
11742	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
11743
11744	/* We reload the initial value, trigger what we can of recalculations the
11745	next time around. From the looks of things, that's all that's required atm. */
11746	if (pDbgState->fModifiedProcCtls)
11747	{
11748	if (!(pDbgState->fProcCtlsInitial & VMX_PROC_CTLS_MOV_DR_EXIT) && CPUMIsHyperDebugStateActive(pVCpu))
11749	pDbgState->fProcCtlsInitial \|= VMX_PROC_CTLS_MOV_DR_EXIT; /* Avoid assertion in hmR0VmxLeave */
11750	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pDbgState->fProcCtlsInitial);
11751	AssertRC(rc2);
11752	pVmcsInfo->u32ProcCtls = pDbgState->fProcCtlsInitial;
11753	}
11754
11755	/* We're currently the only ones messing with this one, so just restore the
11756	cached value and reload the field. */
11757	if ( pDbgState->fModifiedProcCtls2
11758	&& pVmcsInfo->u32ProcCtls2 != pDbgState->fProcCtls2Initial)
11759	{
11760	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pDbgState->fProcCtls2Initial);
11761	AssertRC(rc2);
11762	pVmcsInfo->u32ProcCtls2 = pDbgState->fProcCtls2Initial;
11763	}
11764
11765	/* If we've modified the exception bitmap, we restore it and trigger
11766	reloading and partial recalculation the next time around. */
11767	if (pDbgState->fModifiedXcptBitmap)
11768	pVmcsInfo->u32XcptBitmap = pDbgState->bmXcptInitial;
11769
11770	return rcStrict;
11771	}
11772
11773
11774	/**
11775	* Configures VM-exit controls for current DBGF and DTrace settings.
11776	*
11777	* This updates @a pDbgState and the VMCS execution control fields to reflect
11778	* the necessary VM-exits demanded by DBGF and DTrace.
11779	*
11780	* @param pVCpu The cross context virtual CPU structure.
11781	* @param pVmxTransient The VMX-transient structure. May update
11782	* fUpdatedTscOffsettingAndPreemptTimer.
11783	* @param pDbgState The debug state.
11784	*/
11785	static void hmR0VmxPreRunGuestDebugStateUpdate(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
11786	{
11787	/*
11788	* Take down the dtrace serial number so we can spot changes.
11789	*/
11790	pDbgState->uDtraceSettingsSeqNo = VBOXVMM_GET_SETTINGS_SEQ_NO();
11791	ASMCompilerBarrier();
11792
11793	/*
11794	* We'll rebuild most of the middle block of data members (holding the
11795	* current settings) as we go along here, so start by clearing it all.
11796	*/
11797	pDbgState->bmXcptExtra = 0;
11798	pDbgState->fCpe1Extra = 0;
11799	pDbgState->fCpe1Unwanted = 0;
11800	pDbgState->fCpe2Extra = 0;
11801	for (unsigned i = 0; i < RT_ELEMENTS(pDbgState->bmExitsToCheck); i++)
11802	pDbgState->bmExitsToCheck[i] = 0;
11803
11804	/*
11805	* Software interrupts (INT XXh) - no idea how to trigger these...
11806	*/
11807	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
11808	if ( DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_INTERRUPT_SOFTWARE)
11809	\|\| VBOXVMM_INT_SOFTWARE_ENABLED())
11810	{
11811	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
11812	}
11813
11814	/*
11815	* INT3 breakpoints - triggered by #BP exceptions.
11816	*/
11817	if (pVM->dbgf.ro.cEnabledInt3Breakpoints > 0)
11818	pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
11819
11820	/*
11821	* Exception bitmap and XCPT events+probes.
11822	*/
11823	for (int iXcpt = 0; iXcpt < (DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST + 1); iXcpt++)
11824	if (DBGF_IS_EVENT_ENABLED(pVM, (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + iXcpt)))
11825	pDbgState->bmXcptExtra \|= RT_BIT_32(iXcpt);
11826
11827	if (VBOXVMM_XCPT_DE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DE);
11828	if (VBOXVMM_XCPT_DB_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DB);
11829	if (VBOXVMM_XCPT_BP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
11830	if (VBOXVMM_XCPT_OF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_OF);
11831	if (VBOXVMM_XCPT_BR_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BR);
11832	if (VBOXVMM_XCPT_UD_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_UD);
11833	if (VBOXVMM_XCPT_NM_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NM);
11834	if (VBOXVMM_XCPT_DF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DF);
11835	if (VBOXVMM_XCPT_TS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_TS);
11836	if (VBOXVMM_XCPT_NP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NP);
11837	if (VBOXVMM_XCPT_SS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SS);
11838	if (VBOXVMM_XCPT_GP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_GP);
11839	if (VBOXVMM_XCPT_PF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_PF);
11840	if (VBOXVMM_XCPT_MF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_MF);
11841	if (VBOXVMM_XCPT_AC_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_AC);
11842	if (VBOXVMM_XCPT_XF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_XF);
11843	if (VBOXVMM_XCPT_VE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_VE);
11844	if (VBOXVMM_XCPT_SX_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SX);
11845
11846	if (pDbgState->bmXcptExtra)
11847	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
11848
11849	/*
11850	* Process events and probes for VM-exits, making sure we get the wanted VM-exits.
11851	*
11852	* Note! This is the reverse of what hmR0VmxHandleExitDtraceEvents does.
11853	* So, when adding/changing/removing please don't forget to update it.
11854	*
11855	* Some of the macros are picking up local variables to save horizontal space,
11856	* (being able to see it in a table is the lesser evil here).
11857	*/
11858	#define IS_EITHER_ENABLED(a_pVM, a_EventSubName) \
11859	( DBGF_IS_EVENT_ENABLED(a_pVM, RT_CONCAT(DBGFEVENT_, a_EventSubName)) \
11860	\|\| RT_CONCAT3(VBOXVMM_, a_EventSubName, _ENABLED)() )
11861	#define SET_ONLY_XBM_IF_EITHER_EN(a_EventSubName, a_uExit) \
11862	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11863	{ AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11864	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11865	} else do { } while (0)
11866	#define SET_CPE1_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec) \
11867	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11868	{ \
11869	(pDbgState)->fCpe1Extra \|= (a_fCtrlProcExec); \
11870	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11871	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11872	} else do { } while (0)
11873	#define SET_CPEU_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fUnwantedCtrlProcExec) \
11874	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11875	{ \
11876	(pDbgState)->fCpe1Unwanted \|= (a_fUnwantedCtrlProcExec); \
11877	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11878	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11879	} else do { } while (0)
11880	#define SET_CPE2_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec2) \
11881	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
11882	{ \
11883	(pDbgState)->fCpe2Extra \|= (a_fCtrlProcExec2); \
11884	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
11885	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
11886	} else do { } while (0)
11887
11888	SET_ONLY_XBM_IF_EITHER_EN(EXIT_TASK_SWITCH, VMX_EXIT_TASK_SWITCH); /* unconditional */
11889	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_VIOLATION, VMX_EXIT_EPT_VIOLATION); /* unconditional */
11890	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_MISCONFIG, VMX_EXIT_EPT_MISCONFIG); /* unconditional (unless #VE) */
11891	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_ACCESS, VMX_EXIT_APIC_ACCESS); /* feature dependent, nothing to enable here */
11892	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_WRITE, VMX_EXIT_APIC_WRITE); /* feature dependent, nothing to enable here */
11893
11894	SET_ONLY_XBM_IF_EITHER_EN(INSTR_CPUID, VMX_EXIT_CPUID); /* unconditional */
11895	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CPUID, VMX_EXIT_CPUID);
11896	SET_ONLY_XBM_IF_EITHER_EN(INSTR_GETSEC, VMX_EXIT_GETSEC); /* unconditional */
11897	SET_ONLY_XBM_IF_EITHER_EN( EXIT_GETSEC, VMX_EXIT_GETSEC);
11898	SET_CPE1_XBM_IF_EITHER_EN(INSTR_HALT, VMX_EXIT_HLT, VMX_PROC_CTLS_HLT_EXIT); /* paranoia */
11899	SET_ONLY_XBM_IF_EITHER_EN( EXIT_HALT, VMX_EXIT_HLT);
11900	SET_ONLY_XBM_IF_EITHER_EN(INSTR_INVD, VMX_EXIT_INVD); /* unconditional */
11901	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVD, VMX_EXIT_INVD);
11902	SET_CPE1_XBM_IF_EITHER_EN(INSTR_INVLPG, VMX_EXIT_INVLPG, VMX_PROC_CTLS_INVLPG_EXIT);
11903	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVLPG, VMX_EXIT_INVLPG);
11904	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDPMC, VMX_EXIT_RDPMC, VMX_PROC_CTLS_RDPMC_EXIT);
11905	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDPMC, VMX_EXIT_RDPMC);
11906	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSC, VMX_EXIT_RDTSC, VMX_PROC_CTLS_RDTSC_EXIT);
11907	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSC, VMX_EXIT_RDTSC);
11908	SET_ONLY_XBM_IF_EITHER_EN(INSTR_RSM, VMX_EXIT_RSM); /* unconditional */
11909	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RSM, VMX_EXIT_RSM);
11910	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMM_CALL, VMX_EXIT_VMCALL); /* unconditional */
11911	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMM_CALL, VMX_EXIT_VMCALL);
11912	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMCLEAR, VMX_EXIT_VMCLEAR); /* unconditional */
11913	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMCLEAR, VMX_EXIT_VMCLEAR);
11914	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH); /* unconditional */
11915	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH);
11916	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRLD, VMX_EXIT_VMPTRLD); /* unconditional */
11917	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRLD, VMX_EXIT_VMPTRLD);
11918	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRST, VMX_EXIT_VMPTRST); /* unconditional */
11919	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRST, VMX_EXIT_VMPTRST);
11920	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMREAD, VMX_EXIT_VMREAD); /* unconditional */
11921	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMREAD, VMX_EXIT_VMREAD);
11922	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMRESUME, VMX_EXIT_VMRESUME); /* unconditional */
11923	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMRESUME, VMX_EXIT_VMRESUME);
11924	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMWRITE, VMX_EXIT_VMWRITE); /* unconditional */
11925	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMWRITE, VMX_EXIT_VMWRITE);
11926	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXOFF, VMX_EXIT_VMXOFF); /* unconditional */
11927	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXOFF, VMX_EXIT_VMXOFF);
11928	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXON, VMX_EXIT_VMXON); /* unconditional */
11929	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXON, VMX_EXIT_VMXON);
11930
11931	if ( IS_EITHER_ENABLED(pVM, INSTR_CRX_READ)
11932	\|\| IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
11933	{
11934	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_CR4
11935	\| CPUMCTX_EXTRN_APIC_TPR);
11936	AssertRC(rc);
11937
11938	#if 0 /** @todo fix me */
11939	pDbgState->fClearCr0Mask = true;
11940	pDbgState->fClearCr4Mask = true;
11941	#endif
11942	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_READ))
11943	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_CR3_STORE_EXIT \| VMX_PROC_CTLS_CR8_STORE_EXIT;
11944	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
11945	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_CR3_LOAD_EXIT \| VMX_PROC_CTLS_CR8_LOAD_EXIT;
11946	pDbgState->fCpe1Unwanted \|= VMX_PROC_CTLS_USE_TPR_SHADOW; /* risky? */
11947	/* Note! We currently don't use VMX_VMCS32_CTRL_CR3_TARGET_COUNT. It would
11948	require clearing here and in the loop if we start using it. */
11949	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_CRX);
11950	}
11951	else
11952	{
11953	if (pDbgState->fClearCr0Mask)
11954	{
11955	pDbgState->fClearCr0Mask = false;
11956	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR0);
11957	}
11958	if (pDbgState->fClearCr4Mask)
11959	{
11960	pDbgState->fClearCr4Mask = false;
11961	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR4);
11962	}
11963	}
11964	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_READ, VMX_EXIT_MOV_CRX);
11965	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_WRITE, VMX_EXIT_MOV_CRX);
11966
11967	if ( IS_EITHER_ENABLED(pVM, INSTR_DRX_READ)
11968	\|\| IS_EITHER_ENABLED(pVM, INSTR_DRX_WRITE))
11969	{
11970	/** @todo later, need to fix handler as it assumes this won't usually happen. */
11971	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_DRX);
11972	}
11973	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_READ, VMX_EXIT_MOV_DRX);
11974	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_WRITE, VMX_EXIT_MOV_DRX);
11975
11976	SET_CPEU_XBM_IF_EITHER_EN(INSTR_RDMSR, VMX_EXIT_RDMSR, VMX_PROC_CTLS_USE_MSR_BITMAPS); /* risky clearing this? */
11977	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDMSR, VMX_EXIT_RDMSR);
11978	SET_CPEU_XBM_IF_EITHER_EN(INSTR_WRMSR, VMX_EXIT_WRMSR, VMX_PROC_CTLS_USE_MSR_BITMAPS);
11979	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WRMSR, VMX_EXIT_WRMSR);
11980	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MWAIT, VMX_EXIT_MWAIT, VMX_PROC_CTLS_MWAIT_EXIT); /* paranoia */
11981	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MWAIT, VMX_EXIT_MWAIT);
11982	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MONITOR, VMX_EXIT_MONITOR, VMX_PROC_CTLS_MONITOR_EXIT); /* paranoia */
11983	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MONITOR, VMX_EXIT_MONITOR);
11984	#if 0 /** @todo too slow, fix handler. */
11985	SET_CPE1_XBM_IF_EITHER_EN(INSTR_PAUSE, VMX_EXIT_PAUSE, VMX_PROC_CTLS_PAUSE_EXIT);
11986	#endif
11987	SET_ONLY_XBM_IF_EITHER_EN( EXIT_PAUSE, VMX_EXIT_PAUSE);
11988
11989	if ( IS_EITHER_ENABLED(pVM, INSTR_SGDT)
11990	\|\| IS_EITHER_ENABLED(pVM, INSTR_SIDT)
11991	\|\| IS_EITHER_ENABLED(pVM, INSTR_LGDT)
11992	\|\| IS_EITHER_ENABLED(pVM, INSTR_LIDT))
11993	{
11994	pDbgState->fCpe2Extra \|= VMX_PROC_CTLS2_DESC_TABLE_EXIT;
11995	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_GDTR_IDTR_ACCESS);
11996	}
11997	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SGDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11998	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SIDT, VMX_EXIT_GDTR_IDTR_ACCESS);
11999	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LGDT, VMX_EXIT_GDTR_IDTR_ACCESS);
12000	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LIDT, VMX_EXIT_GDTR_IDTR_ACCESS);
12001
12002	if ( IS_EITHER_ENABLED(pVM, INSTR_SLDT)
12003	\|\| IS_EITHER_ENABLED(pVM, INSTR_STR)
12004	\|\| IS_EITHER_ENABLED(pVM, INSTR_LLDT)
12005	\|\| IS_EITHER_ENABLED(pVM, INSTR_LTR))
12006	{
12007	pDbgState->fCpe2Extra \|= VMX_PROC_CTLS2_DESC_TABLE_EXIT;
12008	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_LDTR_TR_ACCESS);
12009	}
12010	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SLDT, VMX_EXIT_LDTR_TR_ACCESS);
12011	SET_ONLY_XBM_IF_EITHER_EN( EXIT_STR, VMX_EXIT_LDTR_TR_ACCESS);
12012	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LLDT, VMX_EXIT_LDTR_TR_ACCESS);
12013	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LTR, VMX_EXIT_LDTR_TR_ACCESS);
12014
12015	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVEPT, VMX_EXIT_INVEPT); /* unconditional */
12016	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVEPT, VMX_EXIT_INVEPT);
12017	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSCP, VMX_EXIT_RDTSCP, VMX_PROC_CTLS_RDTSC_EXIT);
12018	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSCP, VMX_EXIT_RDTSCP);
12019	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVVPID, VMX_EXIT_INVVPID); /* unconditional */
12020	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVVPID, VMX_EXIT_INVVPID);
12021	SET_CPE2_XBM_IF_EITHER_EN(INSTR_WBINVD, VMX_EXIT_WBINVD, VMX_PROC_CTLS2_WBINVD_EXIT);
12022	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WBINVD, VMX_EXIT_WBINVD);
12023	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSETBV, VMX_EXIT_XSETBV); /* unconditional */
12024	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XSETBV, VMX_EXIT_XSETBV);
12025	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDRAND, VMX_EXIT_RDRAND, VMX_PROC_CTLS2_RDRAND_EXIT);
12026	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDRAND, VMX_EXIT_RDRAND);
12027	SET_CPE1_XBM_IF_EITHER_EN(INSTR_VMX_INVPCID, VMX_EXIT_INVPCID, VMX_PROC_CTLS_INVLPG_EXIT);
12028	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVPCID, VMX_EXIT_INVPCID);
12029	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMFUNC, VMX_EXIT_VMFUNC); /* unconditional for the current setup */
12030	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMFUNC, VMX_EXIT_VMFUNC);
12031	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDSEED, VMX_EXIT_RDSEED, VMX_PROC_CTLS2_RDSEED_EXIT);
12032	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDSEED, VMX_EXIT_RDSEED);
12033	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSAVES, VMX_EXIT_XSAVES); /* unconditional (enabled by host, guest cfg) */
12034	SET_ONLY_XBM_IF_EITHER_EN(EXIT_XSAVES, VMX_EXIT_XSAVES);
12035	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XRSTORS, VMX_EXIT_XRSTORS); /* unconditional (enabled by host, guest cfg) */
12036	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XRSTORS, VMX_EXIT_XRSTORS);
12037
12038	#undef IS_EITHER_ENABLED
12039	#undef SET_ONLY_XBM_IF_EITHER_EN
12040	#undef SET_CPE1_XBM_IF_EITHER_EN
12041	#undef SET_CPEU_XBM_IF_EITHER_EN
12042	#undef SET_CPE2_XBM_IF_EITHER_EN
12043
12044	/*
12045	* Sanitize the control stuff.
12046	*/
12047	pDbgState->fCpe2Extra &= g_HmMsrs.u.vmx.ProcCtls2.n.allowed1;
12048	if (pDbgState->fCpe2Extra)
12049	pDbgState->fCpe1Extra \|= VMX_PROC_CTLS_USE_SECONDARY_CTLS;
12050	pDbgState->fCpe1Extra &= g_HmMsrs.u.vmx.ProcCtls.n.allowed1;
12051	pDbgState->fCpe1Unwanted &= ~g_HmMsrs.u.vmx.ProcCtls.n.allowed0;
12052	if (pVCpu->hmr0.s.fDebugWantRdTscExit != RT_BOOL(pDbgState->fCpe1Extra & VMX_PROC_CTLS_RDTSC_EXIT))
12053	{
12054	pVCpu->hmr0.s.fDebugWantRdTscExit ^= true;
12055	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
12056	}
12057
12058	Log6(("HM: debug state: cpe1=%#RX32 cpeu=%#RX32 cpe2=%#RX32%s%s\n",
12059	pDbgState->fCpe1Extra, pDbgState->fCpe1Unwanted, pDbgState->fCpe2Extra,
12060	pDbgState->fClearCr0Mask ? " clr-cr0" : "",
12061	pDbgState->fClearCr4Mask ? " clr-cr4" : ""));
12062	}
12063
12064
12065	/**
12066	* Fires off DBGF events and dtrace probes for a VM-exit, when it's
12067	* appropriate.
12068	*
12069	* The caller has checked the VM-exit against the
12070	* VMXRUNDBGSTATE::bmExitsToCheck bitmap. The caller has checked for NMIs
12071	* already, so we don't have to do that either.
12072	*
12073	* @returns Strict VBox status code (i.e. informational status codes too).
12074	* @param pVCpu The cross context virtual CPU structure.
12075	* @param pVmxTransient The VMX-transient structure.
12076	* @param uExitReason The VM-exit reason.
12077	*
12078	* @remarks The name of this function is displayed by dtrace, so keep it short
12079	* and to the point. No longer than 33 chars long, please.
12080	*/
12081	static VBOXSTRICTRC hmR0VmxHandleExitDtraceEvents(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, uint32_t uExitReason)
12082	{
12083	/*
12084	* Translate the event into a DBGF event (enmEvent + uEventArg) and at the
12085	* same time check whether any corresponding Dtrace event is enabled (fDtrace).
12086	*
12087	* Note! This is the reverse operation of what hmR0VmxPreRunGuestDebugStateUpdate
12088	* does. Must add/change/remove both places. Same ordering, please.
12089	*
12090	* Added/removed events must also be reflected in the next section
12091	* where we dispatch dtrace events.
12092	*/
12093	bool fDtrace1 = false;
12094	bool fDtrace2 = false;
12095	DBGFEVENTTYPE enmEvent1 = DBGFEVENT_END;
12096	DBGFEVENTTYPE enmEvent2 = DBGFEVENT_END;
12097	uint32_t uEventArg = 0;
12098	#define SET_EXIT(a_EventSubName) \
12099	do { \
12100	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
12101	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
12102	} while (0)
12103	#define SET_BOTH(a_EventSubName) \
12104	do { \
12105	enmEvent1 = RT_CONCAT(DBGFEVENT_INSTR_, a_EventSubName); \
12106	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
12107	fDtrace1 = RT_CONCAT3(VBOXVMM_INSTR_, a_EventSubName, _ENABLED)(); \
12108	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
12109	} while (0)
12110	switch (uExitReason)
12111	{
12112	case VMX_EXIT_MTF:
12113	return hmR0VmxExitMtf(pVCpu, pVmxTransient);
12114
12115	case VMX_EXIT_XCPT_OR_NMI:
12116	{
12117	uint8_t const idxVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
12118	switch (VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo))
12119	{
12120	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
12121	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
12122	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
12123	if (idxVector <= (unsigned)(DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST))
12124	{
12125	if (VMX_EXIT_INT_INFO_IS_ERROR_CODE_VALID(pVmxTransient->uExitIntInfo))
12126	{
12127	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
12128	uEventArg = pVmxTransient->uExitIntErrorCode;
12129	}
12130	enmEvent1 = (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + idxVector);
12131	switch (enmEvent1)
12132	{
12133	case DBGFEVENT_XCPT_DE: fDtrace1 = VBOXVMM_XCPT_DE_ENABLED(); break;
12134	case DBGFEVENT_XCPT_DB: fDtrace1 = VBOXVMM_XCPT_DB_ENABLED(); break;
12135	case DBGFEVENT_XCPT_BP: fDtrace1 = VBOXVMM_XCPT_BP_ENABLED(); break;
12136	case DBGFEVENT_XCPT_OF: fDtrace1 = VBOXVMM_XCPT_OF_ENABLED(); break;
12137	case DBGFEVENT_XCPT_BR: fDtrace1 = VBOXVMM_XCPT_BR_ENABLED(); break;
12138	case DBGFEVENT_XCPT_UD: fDtrace1 = VBOXVMM_XCPT_UD_ENABLED(); break;
12139	case DBGFEVENT_XCPT_NM: fDtrace1 = VBOXVMM_XCPT_NM_ENABLED(); break;
12140	case DBGFEVENT_XCPT_DF: fDtrace1 = VBOXVMM_XCPT_DF_ENABLED(); break;
12141	case DBGFEVENT_XCPT_TS: fDtrace1 = VBOXVMM_XCPT_TS_ENABLED(); break;
12142	case DBGFEVENT_XCPT_NP: fDtrace1 = VBOXVMM_XCPT_NP_ENABLED(); break;
12143	case DBGFEVENT_XCPT_SS: fDtrace1 = VBOXVMM_XCPT_SS_ENABLED(); break;
12144	case DBGFEVENT_XCPT_GP: fDtrace1 = VBOXVMM_XCPT_GP_ENABLED(); break;
12145	case DBGFEVENT_XCPT_PF: fDtrace1 = VBOXVMM_XCPT_PF_ENABLED(); break;
12146	case DBGFEVENT_XCPT_MF: fDtrace1 = VBOXVMM_XCPT_MF_ENABLED(); break;
12147	case DBGFEVENT_XCPT_AC: fDtrace1 = VBOXVMM_XCPT_AC_ENABLED(); break;
12148	case DBGFEVENT_XCPT_XF: fDtrace1 = VBOXVMM_XCPT_XF_ENABLED(); break;
12149	case DBGFEVENT_XCPT_VE: fDtrace1 = VBOXVMM_XCPT_VE_ENABLED(); break;
12150	case DBGFEVENT_XCPT_SX: fDtrace1 = VBOXVMM_XCPT_SX_ENABLED(); break;
12151	default: break;
12152	}
12153	}
12154	else
12155	AssertFailed();
12156	break;
12157
12158	case VMX_EXIT_INT_INFO_TYPE_SW_INT:
12159	uEventArg = idxVector;
12160	enmEvent1 = DBGFEVENT_INTERRUPT_SOFTWARE;
12161	fDtrace1 = VBOXVMM_INT_SOFTWARE_ENABLED();
12162	break;
12163	}
12164	break;
12165	}
12166
12167	case VMX_EXIT_TRIPLE_FAULT:
12168	enmEvent1 = DBGFEVENT_TRIPLE_FAULT;
12169	//fDtrace1 = VBOXVMM_EXIT_TRIPLE_FAULT_ENABLED();
12170	break;
12171	case VMX_EXIT_TASK_SWITCH: SET_EXIT(TASK_SWITCH); break;
12172	case VMX_EXIT_EPT_VIOLATION: SET_EXIT(VMX_EPT_VIOLATION); break;
12173	case VMX_EXIT_EPT_MISCONFIG: SET_EXIT(VMX_EPT_MISCONFIG); break;
12174	case VMX_EXIT_APIC_ACCESS: SET_EXIT(VMX_VAPIC_ACCESS); break;
12175	case VMX_EXIT_APIC_WRITE: SET_EXIT(VMX_VAPIC_WRITE); break;
12176
12177	/* Instruction specific VM-exits: */
12178	case VMX_EXIT_CPUID: SET_BOTH(CPUID); break;
12179	case VMX_EXIT_GETSEC: SET_BOTH(GETSEC); break;
12180	case VMX_EXIT_HLT: SET_BOTH(HALT); break;
12181	case VMX_EXIT_INVD: SET_BOTH(INVD); break;
12182	case VMX_EXIT_INVLPG: SET_BOTH(INVLPG); break;
12183	case VMX_EXIT_RDPMC: SET_BOTH(RDPMC); break;
12184	case VMX_EXIT_RDTSC: SET_BOTH(RDTSC); break;
12185	case VMX_EXIT_RSM: SET_BOTH(RSM); break;
12186	case VMX_EXIT_VMCALL: SET_BOTH(VMM_CALL); break;
12187	case VMX_EXIT_VMCLEAR: SET_BOTH(VMX_VMCLEAR); break;
12188	case VMX_EXIT_VMLAUNCH: SET_BOTH(VMX_VMLAUNCH); break;
12189	case VMX_EXIT_VMPTRLD: SET_BOTH(VMX_VMPTRLD); break;
12190	case VMX_EXIT_VMPTRST: SET_BOTH(VMX_VMPTRST); break;
12191	case VMX_EXIT_VMREAD: SET_BOTH(VMX_VMREAD); break;
12192	case VMX_EXIT_VMRESUME: SET_BOTH(VMX_VMRESUME); break;
12193	case VMX_EXIT_VMWRITE: SET_BOTH(VMX_VMWRITE); break;
12194	case VMX_EXIT_VMXOFF: SET_BOTH(VMX_VMXOFF); break;
12195	case VMX_EXIT_VMXON: SET_BOTH(VMX_VMXON); break;
12196	case VMX_EXIT_MOV_CRX:
12197	hmR0VmxReadExitQualVmcs(pVmxTransient);
12198	if (VMX_EXIT_QUAL_CRX_ACCESS(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_CRX_ACCESS_READ)
12199	SET_BOTH(CRX_READ);
12200	else
12201	SET_BOTH(CRX_WRITE);
12202	uEventArg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
12203	break;
12204	case VMX_EXIT_MOV_DRX:
12205	hmR0VmxReadExitQualVmcs(pVmxTransient);
12206	if ( VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual)
12207	== VMX_EXIT_QUAL_DRX_DIRECTION_READ)
12208	SET_BOTH(DRX_READ);
12209	else
12210	SET_BOTH(DRX_WRITE);
12211	uEventArg = VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual);
12212	break;
12213	case VMX_EXIT_RDMSR: SET_BOTH(RDMSR); break;
12214	case VMX_EXIT_WRMSR: SET_BOTH(WRMSR); break;
12215	case VMX_EXIT_MWAIT: SET_BOTH(MWAIT); break;
12216	case VMX_EXIT_MONITOR: SET_BOTH(MONITOR); break;
12217	case VMX_EXIT_PAUSE: SET_BOTH(PAUSE); break;
12218	case VMX_EXIT_GDTR_IDTR_ACCESS:
12219	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
12220	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_BF_XDTR_INSINFO_INSTR_ID))
12221	{
12222	case VMX_XDTR_INSINFO_II_SGDT: SET_BOTH(SGDT); break;
12223	case VMX_XDTR_INSINFO_II_SIDT: SET_BOTH(SIDT); break;
12224	case VMX_XDTR_INSINFO_II_LGDT: SET_BOTH(LGDT); break;
12225	case VMX_XDTR_INSINFO_II_LIDT: SET_BOTH(LIDT); break;
12226	}
12227	break;
12228
12229	case VMX_EXIT_LDTR_TR_ACCESS:
12230	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
12231	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_BF_YYTR_INSINFO_INSTR_ID))
12232	{
12233	case VMX_YYTR_INSINFO_II_SLDT: SET_BOTH(SLDT); break;
12234	case VMX_YYTR_INSINFO_II_STR: SET_BOTH(STR); break;
12235	case VMX_YYTR_INSINFO_II_LLDT: SET_BOTH(LLDT); break;
12236	case VMX_YYTR_INSINFO_II_LTR: SET_BOTH(LTR); break;
12237	}
12238	break;
12239
12240	case VMX_EXIT_INVEPT: SET_BOTH(VMX_INVEPT); break;
12241	case VMX_EXIT_RDTSCP: SET_BOTH(RDTSCP); break;
12242	case VMX_EXIT_INVVPID: SET_BOTH(VMX_INVVPID); break;
12243	case VMX_EXIT_WBINVD: SET_BOTH(WBINVD); break;
12244	case VMX_EXIT_XSETBV: SET_BOTH(XSETBV); break;
12245	case VMX_EXIT_RDRAND: SET_BOTH(RDRAND); break;
12246	case VMX_EXIT_INVPCID: SET_BOTH(VMX_INVPCID); break;
12247	case VMX_EXIT_VMFUNC: SET_BOTH(VMX_VMFUNC); break;
12248	case VMX_EXIT_RDSEED: SET_BOTH(RDSEED); break;
12249	case VMX_EXIT_XSAVES: SET_BOTH(XSAVES); break;
12250	case VMX_EXIT_XRSTORS: SET_BOTH(XRSTORS); break;
12251
12252	/* Events that aren't relevant at this point. */
12253	case VMX_EXIT_EXT_INT:
12254	case VMX_EXIT_INT_WINDOW:
12255	case VMX_EXIT_NMI_WINDOW:
12256	case VMX_EXIT_TPR_BELOW_THRESHOLD:
12257	case VMX_EXIT_PREEMPT_TIMER:
12258	case VMX_EXIT_IO_INSTR:
12259	break;
12260
12261	/* Errors and unexpected events. */
12262	case VMX_EXIT_INIT_SIGNAL:
12263	case VMX_EXIT_SIPI:
12264	case VMX_EXIT_IO_SMI:
12265	case VMX_EXIT_SMI:
12266	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
12267	case VMX_EXIT_ERR_MSR_LOAD:
12268	case VMX_EXIT_ERR_MACHINE_CHECK:
12269	case VMX_EXIT_PML_FULL:
12270	case VMX_EXIT_VIRTUALIZED_EOI:
12271	break;
12272
12273	default:
12274	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
12275	break;
12276	}
12277	#undef SET_BOTH
12278	#undef SET_EXIT
12279
12280	/*
12281	* Dtrace tracepoints go first. We do them here at once so we don't
12282	* have to copy the guest state saving and stuff a few dozen times.
12283	* Down side is that we've got to repeat the switch, though this time
12284	* we use enmEvent since the probes are a subset of what DBGF does.
12285	*/
12286	if (fDtrace1 \|\| fDtrace2)
12287	{
12288	hmR0VmxReadExitQualVmcs(pVmxTransient);
12289	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
12290	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
12291	switch (enmEvent1)
12292	{
12293	/** @todo consider which extra parameters would be helpful for each probe. */
12294	case DBGFEVENT_END: break;
12295	case DBGFEVENT_XCPT_DE: VBOXVMM_XCPT_DE(pVCpu, pCtx); break;
12296	case DBGFEVENT_XCPT_DB: VBOXVMM_XCPT_DB(pVCpu, pCtx, pCtx->dr[6]); break;
12297	case DBGFEVENT_XCPT_BP: VBOXVMM_XCPT_BP(pVCpu, pCtx); break;
12298	case DBGFEVENT_XCPT_OF: VBOXVMM_XCPT_OF(pVCpu, pCtx); break;
12299	case DBGFEVENT_XCPT_BR: VBOXVMM_XCPT_BR(pVCpu, pCtx); break;
12300	case DBGFEVENT_XCPT_UD: VBOXVMM_XCPT_UD(pVCpu, pCtx); break;
12301	case DBGFEVENT_XCPT_NM: VBOXVMM_XCPT_NM(pVCpu, pCtx); break;
12302	case DBGFEVENT_XCPT_DF: VBOXVMM_XCPT_DF(pVCpu, pCtx); break;
12303	case DBGFEVENT_XCPT_TS: VBOXVMM_XCPT_TS(pVCpu, pCtx, uEventArg); break;
12304	case DBGFEVENT_XCPT_NP: VBOXVMM_XCPT_NP(pVCpu, pCtx, uEventArg); break;
12305	case DBGFEVENT_XCPT_SS: VBOXVMM_XCPT_SS(pVCpu, pCtx, uEventArg); break;
12306	case DBGFEVENT_XCPT_GP: VBOXVMM_XCPT_GP(pVCpu, pCtx, uEventArg); break;
12307	case DBGFEVENT_XCPT_PF: VBOXVMM_XCPT_PF(pVCpu, pCtx, uEventArg, pCtx->cr2); break;
12308	case DBGFEVENT_XCPT_MF: VBOXVMM_XCPT_MF(pVCpu, pCtx); break;
12309	case DBGFEVENT_XCPT_AC: VBOXVMM_XCPT_AC(pVCpu, pCtx); break;
12310	case DBGFEVENT_XCPT_XF: VBOXVMM_XCPT_XF(pVCpu, pCtx); break;
12311	case DBGFEVENT_XCPT_VE: VBOXVMM_XCPT_VE(pVCpu, pCtx); break;
12312	case DBGFEVENT_XCPT_SX: VBOXVMM_XCPT_SX(pVCpu, pCtx, uEventArg); break;
12313	case DBGFEVENT_INTERRUPT_SOFTWARE: VBOXVMM_INT_SOFTWARE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12314	case DBGFEVENT_INSTR_CPUID: VBOXVMM_INSTR_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
12315	case DBGFEVENT_INSTR_GETSEC: VBOXVMM_INSTR_GETSEC(pVCpu, pCtx); break;
12316	case DBGFEVENT_INSTR_HALT: VBOXVMM_INSTR_HALT(pVCpu, pCtx); break;
12317	case DBGFEVENT_INSTR_INVD: VBOXVMM_INSTR_INVD(pVCpu, pCtx); break;
12318	case DBGFEVENT_INSTR_INVLPG: VBOXVMM_INSTR_INVLPG(pVCpu, pCtx); break;
12319	case DBGFEVENT_INSTR_RDPMC: VBOXVMM_INSTR_RDPMC(pVCpu, pCtx); break;
12320	case DBGFEVENT_INSTR_RDTSC: VBOXVMM_INSTR_RDTSC(pVCpu, pCtx); break;
12321	case DBGFEVENT_INSTR_RSM: VBOXVMM_INSTR_RSM(pVCpu, pCtx); break;
12322	case DBGFEVENT_INSTR_CRX_READ: VBOXVMM_INSTR_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12323	case DBGFEVENT_INSTR_CRX_WRITE: VBOXVMM_INSTR_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12324	case DBGFEVENT_INSTR_DRX_READ: VBOXVMM_INSTR_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12325	case DBGFEVENT_INSTR_DRX_WRITE: VBOXVMM_INSTR_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12326	case DBGFEVENT_INSTR_RDMSR: VBOXVMM_INSTR_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
12327	case DBGFEVENT_INSTR_WRMSR: VBOXVMM_INSTR_WRMSR(pVCpu, pCtx, pCtx->ecx,
12328	RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
12329	case DBGFEVENT_INSTR_MWAIT: VBOXVMM_INSTR_MWAIT(pVCpu, pCtx); break;
12330	case DBGFEVENT_INSTR_MONITOR: VBOXVMM_INSTR_MONITOR(pVCpu, pCtx); break;
12331	case DBGFEVENT_INSTR_PAUSE: VBOXVMM_INSTR_PAUSE(pVCpu, pCtx); break;
12332	case DBGFEVENT_INSTR_SGDT: VBOXVMM_INSTR_SGDT(pVCpu, pCtx); break;
12333	case DBGFEVENT_INSTR_SIDT: VBOXVMM_INSTR_SIDT(pVCpu, pCtx); break;
12334	case DBGFEVENT_INSTR_LGDT: VBOXVMM_INSTR_LGDT(pVCpu, pCtx); break;
12335	case DBGFEVENT_INSTR_LIDT: VBOXVMM_INSTR_LIDT(pVCpu, pCtx); break;
12336	case DBGFEVENT_INSTR_SLDT: VBOXVMM_INSTR_SLDT(pVCpu, pCtx); break;
12337	case DBGFEVENT_INSTR_STR: VBOXVMM_INSTR_STR(pVCpu, pCtx); break;
12338	case DBGFEVENT_INSTR_LLDT: VBOXVMM_INSTR_LLDT(pVCpu, pCtx); break;
12339	case DBGFEVENT_INSTR_LTR: VBOXVMM_INSTR_LTR(pVCpu, pCtx); break;
12340	case DBGFEVENT_INSTR_RDTSCP: VBOXVMM_INSTR_RDTSCP(pVCpu, pCtx); break;
12341	case DBGFEVENT_INSTR_WBINVD: VBOXVMM_INSTR_WBINVD(pVCpu, pCtx); break;
12342	case DBGFEVENT_INSTR_XSETBV: VBOXVMM_INSTR_XSETBV(pVCpu, pCtx); break;
12343	case DBGFEVENT_INSTR_RDRAND: VBOXVMM_INSTR_RDRAND(pVCpu, pCtx); break;
12344	case DBGFEVENT_INSTR_RDSEED: VBOXVMM_INSTR_RDSEED(pVCpu, pCtx); break;
12345	case DBGFEVENT_INSTR_XSAVES: VBOXVMM_INSTR_XSAVES(pVCpu, pCtx); break;
12346	case DBGFEVENT_INSTR_XRSTORS: VBOXVMM_INSTR_XRSTORS(pVCpu, pCtx); break;
12347	case DBGFEVENT_INSTR_VMM_CALL: VBOXVMM_INSTR_VMM_CALL(pVCpu, pCtx); break;
12348	case DBGFEVENT_INSTR_VMX_VMCLEAR: VBOXVMM_INSTR_VMX_VMCLEAR(pVCpu, pCtx); break;
12349	case DBGFEVENT_INSTR_VMX_VMLAUNCH: VBOXVMM_INSTR_VMX_VMLAUNCH(pVCpu, pCtx); break;
12350	case DBGFEVENT_INSTR_VMX_VMPTRLD: VBOXVMM_INSTR_VMX_VMPTRLD(pVCpu, pCtx); break;
12351	case DBGFEVENT_INSTR_VMX_VMPTRST: VBOXVMM_INSTR_VMX_VMPTRST(pVCpu, pCtx); break;
12352	case DBGFEVENT_INSTR_VMX_VMREAD: VBOXVMM_INSTR_VMX_VMREAD(pVCpu, pCtx); break;
12353	case DBGFEVENT_INSTR_VMX_VMRESUME: VBOXVMM_INSTR_VMX_VMRESUME(pVCpu, pCtx); break;
12354	case DBGFEVENT_INSTR_VMX_VMWRITE: VBOXVMM_INSTR_VMX_VMWRITE(pVCpu, pCtx); break;
12355	case DBGFEVENT_INSTR_VMX_VMXOFF: VBOXVMM_INSTR_VMX_VMXOFF(pVCpu, pCtx); break;
12356	case DBGFEVENT_INSTR_VMX_VMXON: VBOXVMM_INSTR_VMX_VMXON(pVCpu, pCtx); break;
12357	case DBGFEVENT_INSTR_VMX_INVEPT: VBOXVMM_INSTR_VMX_INVEPT(pVCpu, pCtx); break;
12358	case DBGFEVENT_INSTR_VMX_INVVPID: VBOXVMM_INSTR_VMX_INVVPID(pVCpu, pCtx); break;
12359	case DBGFEVENT_INSTR_VMX_INVPCID: VBOXVMM_INSTR_VMX_INVPCID(pVCpu, pCtx); break;
12360	case DBGFEVENT_INSTR_VMX_VMFUNC: VBOXVMM_INSTR_VMX_VMFUNC(pVCpu, pCtx); break;
12361	default: AssertMsgFailed(("enmEvent1=%d uExitReason=%d\n", enmEvent1, uExitReason)); break;
12362	}
12363	switch (enmEvent2)
12364	{
12365	/** @todo consider which extra parameters would be helpful for each probe. */
12366	case DBGFEVENT_END: break;
12367	case DBGFEVENT_EXIT_TASK_SWITCH: VBOXVMM_EXIT_TASK_SWITCH(pVCpu, pCtx); break;
12368	case DBGFEVENT_EXIT_CPUID: VBOXVMM_EXIT_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
12369	case DBGFEVENT_EXIT_GETSEC: VBOXVMM_EXIT_GETSEC(pVCpu, pCtx); break;
12370	case DBGFEVENT_EXIT_HALT: VBOXVMM_EXIT_HALT(pVCpu, pCtx); break;
12371	case DBGFEVENT_EXIT_INVD: VBOXVMM_EXIT_INVD(pVCpu, pCtx); break;
12372	case DBGFEVENT_EXIT_INVLPG: VBOXVMM_EXIT_INVLPG(pVCpu, pCtx); break;
12373	case DBGFEVENT_EXIT_RDPMC: VBOXVMM_EXIT_RDPMC(pVCpu, pCtx); break;
12374	case DBGFEVENT_EXIT_RDTSC: VBOXVMM_EXIT_RDTSC(pVCpu, pCtx); break;
12375	case DBGFEVENT_EXIT_RSM: VBOXVMM_EXIT_RSM(pVCpu, pCtx); break;
12376	case DBGFEVENT_EXIT_CRX_READ: VBOXVMM_EXIT_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12377	case DBGFEVENT_EXIT_CRX_WRITE: VBOXVMM_EXIT_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12378	case DBGFEVENT_EXIT_DRX_READ: VBOXVMM_EXIT_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
12379	case DBGFEVENT_EXIT_DRX_WRITE: VBOXVMM_EXIT_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
12380	case DBGFEVENT_EXIT_RDMSR: VBOXVMM_EXIT_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
12381	case DBGFEVENT_EXIT_WRMSR: VBOXVMM_EXIT_WRMSR(pVCpu, pCtx, pCtx->ecx,
12382	RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
12383	case DBGFEVENT_EXIT_MWAIT: VBOXVMM_EXIT_MWAIT(pVCpu, pCtx); break;
12384	case DBGFEVENT_EXIT_MONITOR: VBOXVMM_EXIT_MONITOR(pVCpu, pCtx); break;
12385	case DBGFEVENT_EXIT_PAUSE: VBOXVMM_EXIT_PAUSE(pVCpu, pCtx); break;
12386	case DBGFEVENT_EXIT_SGDT: VBOXVMM_EXIT_SGDT(pVCpu, pCtx); break;
12387	case DBGFEVENT_EXIT_SIDT: VBOXVMM_EXIT_SIDT(pVCpu, pCtx); break;
12388	case DBGFEVENT_EXIT_LGDT: VBOXVMM_EXIT_LGDT(pVCpu, pCtx); break;
12389	case DBGFEVENT_EXIT_LIDT: VBOXVMM_EXIT_LIDT(pVCpu, pCtx); break;
12390	case DBGFEVENT_EXIT_SLDT: VBOXVMM_EXIT_SLDT(pVCpu, pCtx); break;
12391	case DBGFEVENT_EXIT_STR: VBOXVMM_EXIT_STR(pVCpu, pCtx); break;
12392	case DBGFEVENT_EXIT_LLDT: VBOXVMM_EXIT_LLDT(pVCpu, pCtx); break;
12393	case DBGFEVENT_EXIT_LTR: VBOXVMM_EXIT_LTR(pVCpu, pCtx); break;
12394	case DBGFEVENT_EXIT_RDTSCP: VBOXVMM_EXIT_RDTSCP(pVCpu, pCtx); break;
12395	case DBGFEVENT_EXIT_WBINVD: VBOXVMM_EXIT_WBINVD(pVCpu, pCtx); break;
12396	case DBGFEVENT_EXIT_XSETBV: VBOXVMM_EXIT_XSETBV(pVCpu, pCtx); break;
12397	case DBGFEVENT_EXIT_RDRAND: VBOXVMM_EXIT_RDRAND(pVCpu, pCtx); break;
12398	case DBGFEVENT_EXIT_RDSEED: VBOXVMM_EXIT_RDSEED(pVCpu, pCtx); break;
12399	case DBGFEVENT_EXIT_XSAVES: VBOXVMM_EXIT_XSAVES(pVCpu, pCtx); break;
12400	case DBGFEVENT_EXIT_XRSTORS: VBOXVMM_EXIT_XRSTORS(pVCpu, pCtx); break;
12401	case DBGFEVENT_EXIT_VMM_CALL: VBOXVMM_EXIT_VMM_CALL(pVCpu, pCtx); break;
12402	case DBGFEVENT_EXIT_VMX_VMCLEAR: VBOXVMM_EXIT_VMX_VMCLEAR(pVCpu, pCtx); break;
12403	case DBGFEVENT_EXIT_VMX_VMLAUNCH: VBOXVMM_EXIT_VMX_VMLAUNCH(pVCpu, pCtx); break;
12404	case DBGFEVENT_EXIT_VMX_VMPTRLD: VBOXVMM_EXIT_VMX_VMPTRLD(pVCpu, pCtx); break;
12405	case DBGFEVENT_EXIT_VMX_VMPTRST: VBOXVMM_EXIT_VMX_VMPTRST(pVCpu, pCtx); break;
12406	case DBGFEVENT_EXIT_VMX_VMREAD: VBOXVMM_EXIT_VMX_VMREAD(pVCpu, pCtx); break;
12407	case DBGFEVENT_EXIT_VMX_VMRESUME: VBOXVMM_EXIT_VMX_VMRESUME(pVCpu, pCtx); break;
12408	case DBGFEVENT_EXIT_VMX_VMWRITE: VBOXVMM_EXIT_VMX_VMWRITE(pVCpu, pCtx); break;
12409	case DBGFEVENT_EXIT_VMX_VMXOFF: VBOXVMM_EXIT_VMX_VMXOFF(pVCpu, pCtx); break;
12410	case DBGFEVENT_EXIT_VMX_VMXON: VBOXVMM_EXIT_VMX_VMXON(pVCpu, pCtx); break;
12411	case DBGFEVENT_EXIT_VMX_INVEPT: VBOXVMM_EXIT_VMX_INVEPT(pVCpu, pCtx); break;
12412	case DBGFEVENT_EXIT_VMX_INVVPID: VBOXVMM_EXIT_VMX_INVVPID(pVCpu, pCtx); break;
12413	case DBGFEVENT_EXIT_VMX_INVPCID: VBOXVMM_EXIT_VMX_INVPCID(pVCpu, pCtx); break;
12414	case DBGFEVENT_EXIT_VMX_VMFUNC: VBOXVMM_EXIT_VMX_VMFUNC(pVCpu, pCtx); break;
12415	case DBGFEVENT_EXIT_VMX_EPT_MISCONFIG: VBOXVMM_EXIT_VMX_EPT_MISCONFIG(pVCpu, pCtx); break;
12416	case DBGFEVENT_EXIT_VMX_EPT_VIOLATION: VBOXVMM_EXIT_VMX_EPT_VIOLATION(pVCpu, pCtx); break;
12417	case DBGFEVENT_EXIT_VMX_VAPIC_ACCESS: VBOXVMM_EXIT_VMX_VAPIC_ACCESS(pVCpu, pCtx); break;
12418	case DBGFEVENT_EXIT_VMX_VAPIC_WRITE: VBOXVMM_EXIT_VMX_VAPIC_WRITE(pVCpu, pCtx); break;
12419	default: AssertMsgFailed(("enmEvent2=%d uExitReason=%d\n", enmEvent2, uExitReason)); break;
12420	}
12421	}
12422
12423	/*
12424	* Fire of the DBGF event, if enabled (our check here is just a quick one,
12425	* the DBGF call will do a full check).
12426	*
12427	* Note! DBGF sets DBGFEVENT_INTERRUPT_SOFTWARE in the bitmap.
12428	* Note! If we have to events, we prioritize the first, i.e. the instruction
12429	* one, in order to avoid event nesting.
12430	*/
12431	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
12432	if ( enmEvent1 != DBGFEVENT_END
12433	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent1))
12434	{
12435	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12436	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent1, DBGFEVENTCTX_HM, 1, uEventArg);
12437	if (rcStrict != VINF_SUCCESS)
12438	return rcStrict;
12439	}
12440	else if ( enmEvent2 != DBGFEVENT_END
12441	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent2))
12442	{
12443	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12444	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent2, DBGFEVENTCTX_HM, 1, uEventArg);
12445	if (rcStrict != VINF_SUCCESS)
12446	return rcStrict;
12447	}
12448
12449	return VINF_SUCCESS;
12450	}
12451
12452
12453	/**
12454	* Single-stepping VM-exit filtering.
12455	*
12456	* This is preprocessing the VM-exits and deciding whether we've gotten far
12457	* enough to return VINF_EM_DBG_STEPPED already. If not, normal VM-exit
12458	* handling is performed.
12459	*
12460	* @returns Strict VBox status code (i.e. informational status codes too).
12461	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
12462	* @param pVmxTransient The VMX-transient structure.
12463	* @param pDbgState The debug state.
12464	*/
12465	DECLINLINE(VBOXSTRICTRC) hmR0VmxRunDebugHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PVMXRUNDBGSTATE pDbgState)
12466	{
12467	/*
12468	* Expensive (saves context) generic dtrace VM-exit probe.
12469	*/
12470	uint32_t const uExitReason = pVmxTransient->uExitReason;
12471	if (!VBOXVMM_R0_HMVMX_VMEXIT_ENABLED())
12472	{ /* more likely */ }
12473	else
12474	{
12475	hmR0VmxReadExitQualVmcs(pVmxTransient);
12476	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
12477	AssertRC(rc);
12478	VBOXVMM_R0_HMVMX_VMEXIT(pVCpu, &pVCpu->cpum.GstCtx, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
12479	}
12480
12481	/*
12482	* Check for host NMI, just to get that out of the way.
12483	*/
12484	if (uExitReason != VMX_EXIT_XCPT_OR_NMI)
12485	{ /* normally likely */ }
12486	else
12487	{
12488	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
12489	uint32_t const uIntType = VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo);
12490	if (uIntType == VMX_EXIT_INT_INFO_TYPE_NMI)
12491	return hmR0VmxExitHostNmi(pVCpu, pVmxTransient->pVmcsInfo);
12492	}
12493
12494	/*
12495	* Check for single stepping event if we're stepping.
12496	*/
12497	if (pVCpu->hm.s.fSingleInstruction)
12498	{
12499	switch (uExitReason)
12500	{
12501	case VMX_EXIT_MTF:
12502	return hmR0VmxExitMtf(pVCpu, pVmxTransient);
12503
12504	/* Various events: */
12505	case VMX_EXIT_XCPT_OR_NMI:
12506	case VMX_EXIT_EXT_INT:
12507	case VMX_EXIT_TRIPLE_FAULT:
12508	case VMX_EXIT_INT_WINDOW:
12509	case VMX_EXIT_NMI_WINDOW:
12510	case VMX_EXIT_TASK_SWITCH:
12511	case VMX_EXIT_TPR_BELOW_THRESHOLD:
12512	case VMX_EXIT_APIC_ACCESS:
12513	case VMX_EXIT_EPT_VIOLATION:
12514	case VMX_EXIT_EPT_MISCONFIG:
12515	case VMX_EXIT_PREEMPT_TIMER:
12516
12517	/* Instruction specific VM-exits: */
12518	case VMX_EXIT_CPUID:
12519	case VMX_EXIT_GETSEC:
12520	case VMX_EXIT_HLT:
12521	case VMX_EXIT_INVD:
12522	case VMX_EXIT_INVLPG:
12523	case VMX_EXIT_RDPMC:
12524	case VMX_EXIT_RDTSC:
12525	case VMX_EXIT_RSM:
12526	case VMX_EXIT_VMCALL:
12527	case VMX_EXIT_VMCLEAR:
12528	case VMX_EXIT_VMLAUNCH:
12529	case VMX_EXIT_VMPTRLD:
12530	case VMX_EXIT_VMPTRST:
12531	case VMX_EXIT_VMREAD:
12532	case VMX_EXIT_VMRESUME:
12533	case VMX_EXIT_VMWRITE:
12534	case VMX_EXIT_VMXOFF:
12535	case VMX_EXIT_VMXON:
12536	case VMX_EXIT_MOV_CRX:
12537	case VMX_EXIT_MOV_DRX:
12538	case VMX_EXIT_IO_INSTR:
12539	case VMX_EXIT_RDMSR:
12540	case VMX_EXIT_WRMSR:
12541	case VMX_EXIT_MWAIT:
12542	case VMX_EXIT_MONITOR:
12543	case VMX_EXIT_PAUSE:
12544	case VMX_EXIT_GDTR_IDTR_ACCESS:
12545	case VMX_EXIT_LDTR_TR_ACCESS:
12546	case VMX_EXIT_INVEPT:
12547	case VMX_EXIT_RDTSCP:
12548	case VMX_EXIT_INVVPID:
12549	case VMX_EXIT_WBINVD:
12550	case VMX_EXIT_XSETBV:
12551	case VMX_EXIT_RDRAND:
12552	case VMX_EXIT_INVPCID:
12553	case VMX_EXIT_VMFUNC:
12554	case VMX_EXIT_RDSEED:
12555	case VMX_EXIT_XSAVES:
12556	case VMX_EXIT_XRSTORS:
12557	{
12558	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12559	AssertRCReturn(rc, rc);
12560	if ( pVCpu->cpum.GstCtx.rip != pDbgState->uRipStart
12561	\|\| pVCpu->cpum.GstCtx.cs.Sel != pDbgState->uCsStart)
12562	return VINF_EM_DBG_STEPPED;
12563	break;
12564	}
12565
12566	/* Errors and unexpected events: */
12567	case VMX_EXIT_INIT_SIGNAL:
12568	case VMX_EXIT_SIPI:
12569	case VMX_EXIT_IO_SMI:
12570	case VMX_EXIT_SMI:
12571	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
12572	case VMX_EXIT_ERR_MSR_LOAD:
12573	case VMX_EXIT_ERR_MACHINE_CHECK:
12574	case VMX_EXIT_PML_FULL:
12575	case VMX_EXIT_VIRTUALIZED_EOI:
12576	case VMX_EXIT_APIC_WRITE: /* Some talk about this being fault like, so I guess we must process it? */
12577	break;
12578
12579	default:
12580	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
12581	break;
12582	}
12583	}
12584
12585	/*
12586	* Check for debugger event breakpoints and dtrace probes.
12587	*/
12588	if ( uExitReason < RT_ELEMENTS(pDbgState->bmExitsToCheck) * 32U
12589	&& ASMBitTest(pDbgState->bmExitsToCheck, uExitReason) )
12590	{
12591	VBOXSTRICTRC rcStrict = hmR0VmxHandleExitDtraceEvents(pVCpu, pVmxTransient, uExitReason);
12592	if (rcStrict != VINF_SUCCESS)
12593	return rcStrict;
12594	}
12595
12596	/*
12597	* Normal processing.
12598	*/
12599	#ifdef HMVMX_USE_FUNCTION_TABLE
12600	return g_aVMExitHandlers[uExitReason].pfn(pVCpu, pVmxTransient);
12601	#else
12602	return hmR0VmxHandleExit(pVCpu, pVmxTransient, uExitReason);
12603	#endif
12604	}
12605
12606
12607	/**
12608	* Single steps guest code using hardware-assisted VMX.
12609	*
12610	* This is -not- the same as the guest single-stepping itself (say using EFLAGS.TF)
12611	* but single-stepping through the hypervisor debugger.
12612	*
12613	* @returns Strict VBox status code (i.e. informational status codes too).
12614	* @param pVCpu The cross context virtual CPU structure.
12615	* @param pcLoops Pointer to the number of executed loops.
12616	*
12617	* @note Mostly the same as hmR0VmxRunGuestCodeNormal().
12618	*/
12619	static VBOXSTRICTRC hmR0VmxRunGuestCodeDebug(PVMCPUCC pVCpu, uint32_t *pcLoops)
12620	{
12621	uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops;
12622	Assert(pcLoops);
12623	Assert(*pcLoops <= cMaxResumeLoops);
12624
12625	VMXTRANSIENT VmxTransient;
12626	RT_ZERO(VmxTransient);
12627	VmxTransient.pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
12628
12629	/* Set HMCPU indicators. */
12630	bool const fSavedSingleInstruction = pVCpu->hm.s.fSingleInstruction;
12631	pVCpu->hm.s.fSingleInstruction = pVCpu->hm.s.fSingleInstruction \|\| DBGFIsStepping(pVCpu);
12632	pVCpu->hmr0.s.fDebugWantRdTscExit = false;
12633	pVCpu->hmr0.s.fUsingDebugLoop = true;
12634
12635	/* State we keep to help modify and later restore the VMCS fields we alter, and for detecting steps. */
12636	VMXRUNDBGSTATE DbgState;
12637	hmR0VmxRunDebugStateInit(pVCpu, &VmxTransient, &DbgState);
12638	hmR0VmxPreRunGuestDebugStateUpdate(pVCpu, &VmxTransient, &DbgState);
12639
12640	/*
12641	* The loop.
12642	*/
12643	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
12644	for (;;)
12645	{
12646	Assert(!HMR0SuspendPending());
12647	HMVMX_ASSERT_CPU_SAFE(pVCpu);
12648	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
12649	bool fStepping = pVCpu->hm.s.fSingleInstruction;
12650
12651	/* Set up VM-execution controls the next two can respond to. */
12652	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &VmxTransient, &DbgState);
12653
12654	/*
12655	* Preparatory work for running guest code, this may force us to
12656	* return to ring-3.
12657	*
12658	* Warning! This bugger disables interrupts on VINF_SUCCESS!
12659	*/
12660	rcStrict = hmR0VmxPreRunGuest(pVCpu, &VmxTransient, fStepping);
12661	if (rcStrict != VINF_SUCCESS)
12662	break;
12663
12664	/* Interrupts are disabled at this point! */
12665	hmR0VmxPreRunGuestCommitted(pVCpu, &VmxTransient);
12666
12667	/* Override any obnoxious code in the above two calls. */
12668	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &VmxTransient, &DbgState);
12669
12670	/*
12671	* Finally execute the guest.
12672	*/
12673	int rcRun = hmR0VmxRunGuest(pVCpu, &VmxTransient);
12674
12675	hmR0VmxPostRunGuest(pVCpu, &VmxTransient, rcRun);
12676	/* Interrupts are re-enabled at this point! */
12677
12678	/* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
12679	if (RT_SUCCESS(rcRun))
12680	{ /* very likely */ }
12681	else
12682	{
12683	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
12684	hmR0VmxReportWorldSwitchError(pVCpu, rcRun, &VmxTransient);
12685	return rcRun;
12686	}
12687
12688	/* Profile the VM-exit. */
12689	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
12690	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
12691	STAM_COUNTER_INC(&pVCpu->hm.s.aStatExitReason[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
12692	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
12693	HMVMX_START_EXIT_DISPATCH_PROF();
12694
12695	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, &pVCpu->cpum.GstCtx, VmxTransient.uExitReason);
12696
12697	/*
12698	* Handle the VM-exit - we quit earlier on certain VM-exits, see hmR0VmxHandleExitDebug().
12699	*/
12700	rcStrict = hmR0VmxRunDebugHandleExit(pVCpu, &VmxTransient, &DbgState);
12701	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
12702	if (rcStrict != VINF_SUCCESS)
12703	break;
12704	if (++(*pcLoops) > cMaxResumeLoops)
12705	{
12706	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
12707	rcStrict = VINF_EM_RAW_INTERRUPT;
12708	break;
12709	}
12710
12711	/*
12712	* Stepping: Did the RIP change, if so, consider it a single step.
12713	* Otherwise, make sure one of the TFs gets set.
12714	*/
12715	if (fStepping)
12716	{
12717	int rc = hmR0VmxImportGuestState(pVCpu, VmxTransient.pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
12718	AssertRC(rc);
12719	if ( pVCpu->cpum.GstCtx.rip != DbgState.uRipStart
12720	\|\| pVCpu->cpum.GstCtx.cs.Sel != DbgState.uCsStart)
12721	{
12722	rcStrict = VINF_EM_DBG_STEPPED;
12723	break;
12724	}
12725	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
12726	}
12727
12728	/*
12729	* Update when dtrace settings changes (DBGF kicks us, so no need to check).
12730	*/
12731	if (VBOXVMM_GET_SETTINGS_SEQ_NO() != DbgState.uDtraceSettingsSeqNo)
12732	hmR0VmxPreRunGuestDebugStateUpdate(pVCpu, &VmxTransient, &DbgState);
12733
12734	/* Restore all controls applied by hmR0VmxPreRunGuestDebugStateApply above. */
12735	rcStrict = hmR0VmxRunDebugStateRevert(pVCpu, &VmxTransient, &DbgState, rcStrict);
12736	Assert(rcStrict == VINF_SUCCESS);
12737	}
12738
12739	/*
12740	* Clear the X86_EFL_TF if necessary.
12741	*/
12742	if (pVCpu->hmr0.s.fClearTrapFlag)
12743	{
12744	int rc = hmR0VmxImportGuestState(pVCpu, VmxTransient.pVmcsInfo, CPUMCTX_EXTRN_RFLAGS);
12745	AssertRC(rc);
12746	pVCpu->hmr0.s.fClearTrapFlag = false;
12747	pVCpu->cpum.GstCtx.eflags.Bits.u1TF = 0;
12748	}
12749	/** @todo there seems to be issues with the resume flag when the monitor trap
12750	* flag is pending without being used. Seen early in bios init when
12751	* accessing APIC page in protected mode. */
12752
12753	/* Restore HMCPU indicators. */
12754	pVCpu->hmr0.s.fUsingDebugLoop = false;
12755	pVCpu->hmr0.s.fDebugWantRdTscExit = false;
12756	pVCpu->hm.s.fSingleInstruction = fSavedSingleInstruction;
12757
12758	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
12759	return rcStrict;
12760	}
12761
12762
12763	/** @} */
12764
12765
12766	/**
12767	* Checks if any expensive dtrace probes are enabled and we should go to the
12768	* debug loop.
12769	*
12770	* @returns true if we should use debug loop, false if not.
12771	*/
12772	static bool hmR0VmxAnyExpensiveProbesEnabled(void)
12773	{
12774	/* It's probably faster to OR the raw 32-bit counter variables together.
12775	Since the variables are in an array and the probes are next to one
12776	another (more or less), we have good locality. So, better read
12777	eight-nine cache lines ever time and only have one conditional, than
12778	128+ conditionals, right? */
12779	return ( VBOXVMM_R0_HMVMX_VMEXIT_ENABLED_RAW() /* expensive too due to context */
12780	\| VBOXVMM_XCPT_DE_ENABLED_RAW()
12781	\| VBOXVMM_XCPT_DB_ENABLED_RAW()
12782	\| VBOXVMM_XCPT_BP_ENABLED_RAW()
12783	\| VBOXVMM_XCPT_OF_ENABLED_RAW()
12784	\| VBOXVMM_XCPT_BR_ENABLED_RAW()
12785	\| VBOXVMM_XCPT_UD_ENABLED_RAW()
12786	\| VBOXVMM_XCPT_NM_ENABLED_RAW()
12787	\| VBOXVMM_XCPT_DF_ENABLED_RAW()
12788	\| VBOXVMM_XCPT_TS_ENABLED_RAW()
12789	\| VBOXVMM_XCPT_NP_ENABLED_RAW()
12790	\| VBOXVMM_XCPT_SS_ENABLED_RAW()
12791	\| VBOXVMM_XCPT_GP_ENABLED_RAW()
12792	\| VBOXVMM_XCPT_PF_ENABLED_RAW()
12793	\| VBOXVMM_XCPT_MF_ENABLED_RAW()
12794	\| VBOXVMM_XCPT_AC_ENABLED_RAW()
12795	\| VBOXVMM_XCPT_XF_ENABLED_RAW()
12796	\| VBOXVMM_XCPT_VE_ENABLED_RAW()
12797	\| VBOXVMM_XCPT_SX_ENABLED_RAW()
12798	\| VBOXVMM_INT_SOFTWARE_ENABLED_RAW()
12799	\| VBOXVMM_INT_HARDWARE_ENABLED_RAW()
12800	) != 0
12801	\|\| ( VBOXVMM_INSTR_HALT_ENABLED_RAW()
12802	\| VBOXVMM_INSTR_MWAIT_ENABLED_RAW()
12803	\| VBOXVMM_INSTR_MONITOR_ENABLED_RAW()
12804	\| VBOXVMM_INSTR_CPUID_ENABLED_RAW()
12805	\| VBOXVMM_INSTR_INVD_ENABLED_RAW()
12806	\| VBOXVMM_INSTR_WBINVD_ENABLED_RAW()
12807	\| VBOXVMM_INSTR_INVLPG_ENABLED_RAW()
12808	\| VBOXVMM_INSTR_RDTSC_ENABLED_RAW()
12809	\| VBOXVMM_INSTR_RDTSCP_ENABLED_RAW()
12810	\| VBOXVMM_INSTR_RDPMC_ENABLED_RAW()
12811	\| VBOXVMM_INSTR_RDMSR_ENABLED_RAW()
12812	\| VBOXVMM_INSTR_WRMSR_ENABLED_RAW()
12813	\| VBOXVMM_INSTR_CRX_READ_ENABLED_RAW()
12814	\| VBOXVMM_INSTR_CRX_WRITE_ENABLED_RAW()
12815	\| VBOXVMM_INSTR_DRX_READ_ENABLED_RAW()
12816	\| VBOXVMM_INSTR_DRX_WRITE_ENABLED_RAW()
12817	\| VBOXVMM_INSTR_PAUSE_ENABLED_RAW()
12818	\| VBOXVMM_INSTR_XSETBV_ENABLED_RAW()
12819	\| VBOXVMM_INSTR_SIDT_ENABLED_RAW()
12820	\| VBOXVMM_INSTR_LIDT_ENABLED_RAW()
12821	\| VBOXVMM_INSTR_SGDT_ENABLED_RAW()
12822	\| VBOXVMM_INSTR_LGDT_ENABLED_RAW()
12823	\| VBOXVMM_INSTR_SLDT_ENABLED_RAW()
12824	\| VBOXVMM_INSTR_LLDT_ENABLED_RAW()
12825	\| VBOXVMM_INSTR_STR_ENABLED_RAW()
12826	\| VBOXVMM_INSTR_LTR_ENABLED_RAW()
12827	\| VBOXVMM_INSTR_GETSEC_ENABLED_RAW()
12828	\| VBOXVMM_INSTR_RSM_ENABLED_RAW()
12829	\| VBOXVMM_INSTR_RDRAND_ENABLED_RAW()
12830	\| VBOXVMM_INSTR_RDSEED_ENABLED_RAW()
12831	\| VBOXVMM_INSTR_XSAVES_ENABLED_RAW()
12832	\| VBOXVMM_INSTR_XRSTORS_ENABLED_RAW()
12833	\| VBOXVMM_INSTR_VMM_CALL_ENABLED_RAW()
12834	\| VBOXVMM_INSTR_VMX_VMCLEAR_ENABLED_RAW()
12835	\| VBOXVMM_INSTR_VMX_VMLAUNCH_ENABLED_RAW()
12836	\| VBOXVMM_INSTR_VMX_VMPTRLD_ENABLED_RAW()
12837	\| VBOXVMM_INSTR_VMX_VMPTRST_ENABLED_RAW()
12838	\| VBOXVMM_INSTR_VMX_VMREAD_ENABLED_RAW()
12839	\| VBOXVMM_INSTR_VMX_VMRESUME_ENABLED_RAW()
12840	\| VBOXVMM_INSTR_VMX_VMWRITE_ENABLED_RAW()
12841	\| VBOXVMM_INSTR_VMX_VMXOFF_ENABLED_RAW()
12842	\| VBOXVMM_INSTR_VMX_VMXON_ENABLED_RAW()
12843	\| VBOXVMM_INSTR_VMX_VMFUNC_ENABLED_RAW()
12844	\| VBOXVMM_INSTR_VMX_INVEPT_ENABLED_RAW()
12845	\| VBOXVMM_INSTR_VMX_INVVPID_ENABLED_RAW()
12846	\| VBOXVMM_INSTR_VMX_INVPCID_ENABLED_RAW()
12847	) != 0
12848	\|\| ( VBOXVMM_EXIT_TASK_SWITCH_ENABLED_RAW()
12849	\| VBOXVMM_EXIT_HALT_ENABLED_RAW()
12850	\| VBOXVMM_EXIT_MWAIT_ENABLED_RAW()
12851	\| VBOXVMM_EXIT_MONITOR_ENABLED_RAW()
12852	\| VBOXVMM_EXIT_CPUID_ENABLED_RAW()
12853	\| VBOXVMM_EXIT_INVD_ENABLED_RAW()
12854	\| VBOXVMM_EXIT_WBINVD_ENABLED_RAW()
12855	\| VBOXVMM_EXIT_INVLPG_ENABLED_RAW()
12856	\| VBOXVMM_EXIT_RDTSC_ENABLED_RAW()
12857	\| VBOXVMM_EXIT_RDTSCP_ENABLED_RAW()
12858	\| VBOXVMM_EXIT_RDPMC_ENABLED_RAW()
12859	\| VBOXVMM_EXIT_RDMSR_ENABLED_RAW()
12860	\| VBOXVMM_EXIT_WRMSR_ENABLED_RAW()
12861	\| VBOXVMM_EXIT_CRX_READ_ENABLED_RAW()
12862	\| VBOXVMM_EXIT_CRX_WRITE_ENABLED_RAW()
12863	\| VBOXVMM_EXIT_DRX_READ_ENABLED_RAW()
12864	\| VBOXVMM_EXIT_DRX_WRITE_ENABLED_RAW()
12865	\| VBOXVMM_EXIT_PAUSE_ENABLED_RAW()
12866	\| VBOXVMM_EXIT_XSETBV_ENABLED_RAW()
12867	\| VBOXVMM_EXIT_SIDT_ENABLED_RAW()
12868	\| VBOXVMM_EXIT_LIDT_ENABLED_RAW()
12869	\| VBOXVMM_EXIT_SGDT_ENABLED_RAW()
12870	\| VBOXVMM_EXIT_LGDT_ENABLED_RAW()
12871	\| VBOXVMM_EXIT_SLDT_ENABLED_RAW()
12872	\| VBOXVMM_EXIT_LLDT_ENABLED_RAW()
12873	\| VBOXVMM_EXIT_STR_ENABLED_RAW()
12874	\| VBOXVMM_EXIT_LTR_ENABLED_RAW()
12875	\| VBOXVMM_EXIT_GETSEC_ENABLED_RAW()
12876	\| VBOXVMM_EXIT_RSM_ENABLED_RAW()
12877	\| VBOXVMM_EXIT_RDRAND_ENABLED_RAW()
12878	\| VBOXVMM_EXIT_RDSEED_ENABLED_RAW()
12879	\| VBOXVMM_EXIT_XSAVES_ENABLED_RAW()
12880	\| VBOXVMM_EXIT_XRSTORS_ENABLED_RAW()
12881	\| VBOXVMM_EXIT_VMM_CALL_ENABLED_RAW()
12882	\| VBOXVMM_EXIT_VMX_VMCLEAR_ENABLED_RAW()
12883	\| VBOXVMM_EXIT_VMX_VMLAUNCH_ENABLED_RAW()
12884	\| VBOXVMM_EXIT_VMX_VMPTRLD_ENABLED_RAW()
12885	\| VBOXVMM_EXIT_VMX_VMPTRST_ENABLED_RAW()
12886	\| VBOXVMM_EXIT_VMX_VMREAD_ENABLED_RAW()
12887	\| VBOXVMM_EXIT_VMX_VMRESUME_ENABLED_RAW()
12888	\| VBOXVMM_EXIT_VMX_VMWRITE_ENABLED_RAW()
12889	\| VBOXVMM_EXIT_VMX_VMXOFF_ENABLED_RAW()
12890	\| VBOXVMM_EXIT_VMX_VMXON_ENABLED_RAW()
12891	\| VBOXVMM_EXIT_VMX_VMFUNC_ENABLED_RAW()
12892	\| VBOXVMM_EXIT_VMX_INVEPT_ENABLED_RAW()
12893	\| VBOXVMM_EXIT_VMX_INVVPID_ENABLED_RAW()
12894	\| VBOXVMM_EXIT_VMX_INVPCID_ENABLED_RAW()
12895	\| VBOXVMM_EXIT_VMX_EPT_VIOLATION_ENABLED_RAW()
12896	\| VBOXVMM_EXIT_VMX_EPT_MISCONFIG_ENABLED_RAW()
12897	\| VBOXVMM_EXIT_VMX_VAPIC_ACCESS_ENABLED_RAW()
12898	\| VBOXVMM_EXIT_VMX_VAPIC_WRITE_ENABLED_RAW()
12899	) != 0;
12900	}
12901
12902
12903	/**
12904	* Runs the guest using hardware-assisted VMX.
12905	*
12906	* @returns Strict VBox status code (i.e. informational status codes too).
12907	* @param pVCpu The cross context virtual CPU structure.
12908	*/
12909	VMMR0DECL(VBOXSTRICTRC) VMXR0RunGuestCode(PVMCPUCC pVCpu)
12910	{
12911	AssertPtr(pVCpu);
12912	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
12913	Assert(VMMRZCallRing3IsEnabled(pVCpu));
12914	Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
12915	HMVMX_ASSERT_PREEMPT_SAFE(pVCpu);
12916
12917	VBOXSTRICTRC rcStrict;
12918	uint32_t cLoops = 0;
12919	for (;;)
12920	{
12921	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12922	bool const fInNestedGuestMode = CPUMIsGuestInVmxNonRootMode(pCtx);
12923	#else
12924	NOREF(pCtx);
12925	bool const fInNestedGuestMode = false;
12926	#endif
12927	if (!fInNestedGuestMode)
12928	{
12929	if ( !pVCpu->hm.s.fUseDebugLoop
12930	&& (!VBOXVMM_ANY_PROBES_ENABLED() \|\| !hmR0VmxAnyExpensiveProbesEnabled())
12931	&& !DBGFIsStepping(pVCpu)
12932	&& !pVCpu->CTX_SUFF(pVM)->dbgf.ro.cEnabledInt3Breakpoints)
12933	rcStrict = hmR0VmxRunGuestCodeNormal(pVCpu, &cLoops);
12934	else
12935	rcStrict = hmR0VmxRunGuestCodeDebug(pVCpu, &cLoops);
12936	}
12937	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
12938	else
12939	rcStrict = hmR0VmxRunGuestCodeNested(pVCpu, &cLoops);
12940
12941	if (rcStrict == VINF_VMX_VMLAUNCH_VMRESUME)
12942	{
12943	Assert(CPUMIsGuestInVmxNonRootMode(pCtx));
12944	continue;
12945	}
12946	if (rcStrict == VINF_VMX_VMEXIT)
12947	{
12948	Assert(!CPUMIsGuestInVmxNonRootMode(pCtx));
12949	continue;
12950	}
12951	#endif
12952	break;
12953	}
12954
12955	int const rcLoop = VBOXSTRICTRC_VAL(rcStrict);
12956	switch (rcLoop)
12957	{
12958	case VERR_EM_INTERPRETER: rcStrict = VINF_EM_RAW_EMULATE_INSTR; break;
12959	case VINF_EM_RESET: rcStrict = VINF_EM_TRIPLE_FAULT; break;
12960	}
12961
12962	int rc2 = hmR0VmxExitToRing3(pVCpu, rcStrict);
12963	if (RT_FAILURE(rc2))
12964	{
12965	pVCpu->hm.s.u32HMError = (uint32_t)VBOXSTRICTRC_VAL(rcStrict);
12966	rcStrict = rc2;
12967	}
12968	Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
12969	Assert(!VMMRZCallRing3IsNotificationSet(pVCpu));
12970	return rcStrict;
12971	}
12972
12973
12974	#ifndef HMVMX_USE_FUNCTION_TABLE
12975	/**
12976	* Handles a guest VM-exit from hardware-assisted VMX execution.
12977	*
12978	* @returns Strict VBox status code (i.e. informational status codes too).
12979	* @param pVCpu The cross context virtual CPU structure.
12980	* @param pVmxTransient The VMX-transient structure.
12981	*/
12982	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
12983	{
12984	#ifdef DEBUG_ramshankar
12985	# define VMEXIT_CALL_RET(a_fSave, a_CallExpr) \
12986	do { \
12987	if (a_fSave != 0) \
12988	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL); \
12989	VBOXSTRICTRC rcStrict = a_CallExpr; \
12990	if (a_fSave != 0) \
12991	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST); \
12992	return rcStrict; \
12993	} while (0)
12994	#else
12995	# define VMEXIT_CALL_RET(a_fSave, a_CallExpr) return a_CallExpr
12996	#endif
12997	uint32_t const uExitReason = pVmxTransient->uExitReason;
12998	switch (uExitReason)
12999	{
13000	case VMX_EXIT_EPT_MISCONFIG: VMEXIT_CALL_RET(0, hmR0VmxExitEptMisconfig(pVCpu, pVmxTransient));
13001	case VMX_EXIT_EPT_VIOLATION: VMEXIT_CALL_RET(0, hmR0VmxExitEptViolation(pVCpu, pVmxTransient));
13002	case VMX_EXIT_IO_INSTR: VMEXIT_CALL_RET(0, hmR0VmxExitIoInstr(pVCpu, pVmxTransient));
13003	case VMX_EXIT_CPUID: VMEXIT_CALL_RET(0, hmR0VmxExitCpuid(pVCpu, pVmxTransient));
13004	case VMX_EXIT_RDTSC: VMEXIT_CALL_RET(0, hmR0VmxExitRdtsc(pVCpu, pVmxTransient));
13005	case VMX_EXIT_RDTSCP: VMEXIT_CALL_RET(0, hmR0VmxExitRdtscp(pVCpu, pVmxTransient));
13006	case VMX_EXIT_APIC_ACCESS: VMEXIT_CALL_RET(0, hmR0VmxExitApicAccess(pVCpu, pVmxTransient));
13007	case VMX_EXIT_XCPT_OR_NMI: VMEXIT_CALL_RET(0, hmR0VmxExitXcptOrNmi(pVCpu, pVmxTransient));
13008	case VMX_EXIT_MOV_CRX: VMEXIT_CALL_RET(0, hmR0VmxExitMovCRx(pVCpu, pVmxTransient));
13009	case VMX_EXIT_EXT_INT: VMEXIT_CALL_RET(0, hmR0VmxExitExtInt(pVCpu, pVmxTransient));
13010	case VMX_EXIT_INT_WINDOW: VMEXIT_CALL_RET(0, hmR0VmxExitIntWindow(pVCpu, pVmxTransient));
13011	case VMX_EXIT_TPR_BELOW_THRESHOLD: VMEXIT_CALL_RET(0, hmR0VmxExitTprBelowThreshold(pVCpu, pVmxTransient));
13012	case VMX_EXIT_MWAIT: VMEXIT_CALL_RET(0, hmR0VmxExitMwait(pVCpu, pVmxTransient));
13013	case VMX_EXIT_MONITOR: VMEXIT_CALL_RET(0, hmR0VmxExitMonitor(pVCpu, pVmxTransient));
13014	case VMX_EXIT_TASK_SWITCH: VMEXIT_CALL_RET(0, hmR0VmxExitTaskSwitch(pVCpu, pVmxTransient));
13015	case VMX_EXIT_PREEMPT_TIMER: VMEXIT_CALL_RET(0, hmR0VmxExitPreemptTimer(pVCpu, pVmxTransient));
13016	case VMX_EXIT_RDMSR: VMEXIT_CALL_RET(0, hmR0VmxExitRdmsr(pVCpu, pVmxTransient));
13017	case VMX_EXIT_WRMSR: VMEXIT_CALL_RET(0, hmR0VmxExitWrmsr(pVCpu, pVmxTransient));
13018	case VMX_EXIT_VMCALL: VMEXIT_CALL_RET(0, hmR0VmxExitVmcall(pVCpu, pVmxTransient));
13019	case VMX_EXIT_MOV_DRX: VMEXIT_CALL_RET(0, hmR0VmxExitMovDRx(pVCpu, pVmxTransient));
13020	case VMX_EXIT_HLT: VMEXIT_CALL_RET(0, hmR0VmxExitHlt(pVCpu, pVmxTransient));
13021	case VMX_EXIT_INVD: VMEXIT_CALL_RET(0, hmR0VmxExitInvd(pVCpu, pVmxTransient));
13022	case VMX_EXIT_INVLPG: VMEXIT_CALL_RET(0, hmR0VmxExitInvlpg(pVCpu, pVmxTransient));
13023	case VMX_EXIT_MTF: VMEXIT_CALL_RET(0, hmR0VmxExitMtf(pVCpu, pVmxTransient));
13024	case VMX_EXIT_PAUSE: VMEXIT_CALL_RET(0, hmR0VmxExitPause(pVCpu, pVmxTransient));
13025	case VMX_EXIT_WBINVD: VMEXIT_CALL_RET(0, hmR0VmxExitWbinvd(pVCpu, pVmxTransient));
13026	case VMX_EXIT_XSETBV: VMEXIT_CALL_RET(0, hmR0VmxExitXsetbv(pVCpu, pVmxTransient));
13027	case VMX_EXIT_INVPCID: VMEXIT_CALL_RET(0, hmR0VmxExitInvpcid(pVCpu, pVmxTransient));
13028	case VMX_EXIT_GETSEC: VMEXIT_CALL_RET(0, hmR0VmxExitGetsec(pVCpu, pVmxTransient));
13029	case VMX_EXIT_RDPMC: VMEXIT_CALL_RET(0, hmR0VmxExitRdpmc(pVCpu, pVmxTransient));
13030	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13031	case VMX_EXIT_VMCLEAR: VMEXIT_CALL_RET(0, hmR0VmxExitVmclear(pVCpu, pVmxTransient));
13032	case VMX_EXIT_VMLAUNCH: VMEXIT_CALL_RET(0, hmR0VmxExitVmlaunch(pVCpu, pVmxTransient));
13033	case VMX_EXIT_VMPTRLD: VMEXIT_CALL_RET(0, hmR0VmxExitVmptrld(pVCpu, pVmxTransient));
13034	case VMX_EXIT_VMPTRST: VMEXIT_CALL_RET(0, hmR0VmxExitVmptrst(pVCpu, pVmxTransient));
13035	case VMX_EXIT_VMREAD: VMEXIT_CALL_RET(0, hmR0VmxExitVmread(pVCpu, pVmxTransient));
13036	case VMX_EXIT_VMRESUME: VMEXIT_CALL_RET(0, hmR0VmxExitVmwrite(pVCpu, pVmxTransient));
13037	case VMX_EXIT_VMWRITE: VMEXIT_CALL_RET(0, hmR0VmxExitVmresume(pVCpu, pVmxTransient));
13038	case VMX_EXIT_VMXOFF: VMEXIT_CALL_RET(0, hmR0VmxExitVmxoff(pVCpu, pVmxTransient));
13039	case VMX_EXIT_VMXON: VMEXIT_CALL_RET(0, hmR0VmxExitVmxon(pVCpu, pVmxTransient));
13040	case VMX_EXIT_INVVPID: VMEXIT_CALL_RET(0, hmR0VmxExitInvvpid(pVCpu, pVmxTransient));
13041	case VMX_EXIT_INVEPT: VMEXIT_CALL_RET(0, hmR0VmxExitSetPendingXcptUD(pVCpu, pVmxTransient));
13042	#else
13043	case VMX_EXIT_VMCLEAR:
13044	case VMX_EXIT_VMLAUNCH:
13045	case VMX_EXIT_VMPTRLD:
13046	case VMX_EXIT_VMPTRST:
13047	case VMX_EXIT_VMREAD:
13048	case VMX_EXIT_VMRESUME:
13049	case VMX_EXIT_VMWRITE:
13050	case VMX_EXIT_VMXOFF:
13051	case VMX_EXIT_VMXON:
13052	case VMX_EXIT_INVVPID:
13053	case VMX_EXIT_INVEPT:
13054	return hmR0VmxExitSetPendingXcptUD(pVCpu, pVmxTransient);
13055	#endif
13056
13057	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFault(pVCpu, pVmxTransient);
13058	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindow(pVCpu, pVmxTransient);
13059	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestState(pVCpu, pVmxTransient);
13060
13061	case VMX_EXIT_INIT_SIGNAL:
13062	case VMX_EXIT_SIPI:
13063	case VMX_EXIT_IO_SMI:
13064	case VMX_EXIT_SMI:
13065	case VMX_EXIT_ERR_MSR_LOAD:
13066	case VMX_EXIT_ERR_MACHINE_CHECK:
13067	case VMX_EXIT_PML_FULL:
13068	case VMX_EXIT_VIRTUALIZED_EOI:
13069	case VMX_EXIT_GDTR_IDTR_ACCESS:
13070	case VMX_EXIT_LDTR_TR_ACCESS:
13071	case VMX_EXIT_APIC_WRITE:
13072	case VMX_EXIT_RDRAND:
13073	case VMX_EXIT_RSM:
13074	case VMX_EXIT_VMFUNC:
13075	case VMX_EXIT_ENCLS:
13076	case VMX_EXIT_RDSEED:
13077	case VMX_EXIT_XSAVES:
13078	case VMX_EXIT_XRSTORS:
13079	case VMX_EXIT_UMWAIT:
13080	case VMX_EXIT_TPAUSE:
13081	case VMX_EXIT_LOADIWKEY:
13082	default:
13083	return hmR0VmxExitErrUnexpected(pVCpu, pVmxTransient);
13084	}
13085	#undef VMEXIT_CALL_RET
13086	}
13087	#endif /* !HMVMX_USE_FUNCTION_TABLE */
13088
13089
13090	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13091	/**
13092	* Handles a nested-guest VM-exit from hardware-assisted VMX execution.
13093	*
13094	* @returns Strict VBox status code (i.e. informational status codes too).
13095	* @param pVCpu The cross context virtual CPU structure.
13096	* @param pVmxTransient The VMX-transient structure.
13097	*/
13098	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13099	{
13100	uint32_t const uExitReason = pVmxTransient->uExitReason;
13101	switch (uExitReason)
13102	{
13103	case VMX_EXIT_EPT_MISCONFIG: return hmR0VmxExitEptMisconfig(pVCpu, pVmxTransient);
13104	case VMX_EXIT_EPT_VIOLATION: return hmR0VmxExitEptViolation(pVCpu, pVmxTransient);
13105	case VMX_EXIT_XCPT_OR_NMI: return hmR0VmxExitXcptOrNmiNested(pVCpu, pVmxTransient);
13106	case VMX_EXIT_IO_INSTR: return hmR0VmxExitIoInstrNested(pVCpu, pVmxTransient);
13107	case VMX_EXIT_HLT: return hmR0VmxExitHltNested(pVCpu, pVmxTransient);
13108
13109	/*
13110	* We shouldn't direct host physical interrupts to the nested-guest.
13111	*/
13112	case VMX_EXIT_EXT_INT:
13113	return hmR0VmxExitExtInt(pVCpu, pVmxTransient);
13114
13115	/*
13116	* Instructions that cause VM-exits unconditionally or the condition is
13117	* always is taken solely from the nested hypervisor (meaning if the VM-exit
13118	* happens, it's guaranteed to be a nested-guest VM-exit).
13119	*
13120	* - Provides VM-exit instruction length ONLY.
13121	*/
13122	case VMX_EXIT_CPUID: /* Unconditional. */
13123	case VMX_EXIT_VMCALL:
13124	case VMX_EXIT_GETSEC:
13125	case VMX_EXIT_INVD:
13126	case VMX_EXIT_XSETBV:
13127	case VMX_EXIT_VMLAUNCH:
13128	case VMX_EXIT_VMRESUME:
13129	case VMX_EXIT_VMXOFF:
13130	case VMX_EXIT_ENCLS: /* Condition specified solely by nested hypervisor. */
13131	case VMX_EXIT_VMFUNC:
13132	return hmR0VmxExitInstrNested(pVCpu, pVmxTransient);
13133
13134	/*
13135	* Instructions that cause VM-exits unconditionally or the condition is
13136	* always is taken solely from the nested hypervisor (meaning if the VM-exit
13137	* happens, it's guaranteed to be a nested-guest VM-exit).
13138	*
13139	* - Provides VM-exit instruction length.
13140	* - Provides VM-exit information.
13141	* - Optionally provides Exit qualification.
13142	*
13143	* Since Exit qualification is 0 for all VM-exits where it is not
13144	* applicable, reading and passing it to the guest should produce
13145	* defined behavior.
13146	*
13147	* See Intel spec. 27.2.1 "Basic VM-Exit Information".
13148	*/
13149	case VMX_EXIT_INVEPT: /* Unconditional. */
13150	case VMX_EXIT_INVVPID:
13151	case VMX_EXIT_VMCLEAR:
13152	case VMX_EXIT_VMPTRLD:
13153	case VMX_EXIT_VMPTRST:
13154	case VMX_EXIT_VMXON:
13155	case VMX_EXIT_GDTR_IDTR_ACCESS: /* Condition specified solely by nested hypervisor. */
13156	case VMX_EXIT_LDTR_TR_ACCESS:
13157	case VMX_EXIT_RDRAND:
13158	case VMX_EXIT_RDSEED:
13159	case VMX_EXIT_XSAVES:
13160	case VMX_EXIT_XRSTORS:
13161	case VMX_EXIT_UMWAIT:
13162	case VMX_EXIT_TPAUSE:
13163	return hmR0VmxExitInstrWithInfoNested(pVCpu, pVmxTransient);
13164
13165	case VMX_EXIT_RDTSC: return hmR0VmxExitRdtscNested(pVCpu, pVmxTransient);
13166	case VMX_EXIT_RDTSCP: return hmR0VmxExitRdtscpNested(pVCpu, pVmxTransient);
13167	case VMX_EXIT_RDMSR: return hmR0VmxExitRdmsrNested(pVCpu, pVmxTransient);
13168	case VMX_EXIT_WRMSR: return hmR0VmxExitWrmsrNested(pVCpu, pVmxTransient);
13169	case VMX_EXIT_INVLPG: return hmR0VmxExitInvlpgNested(pVCpu, pVmxTransient);
13170	case VMX_EXIT_INVPCID: return hmR0VmxExitInvpcidNested(pVCpu, pVmxTransient);
13171	case VMX_EXIT_TASK_SWITCH: return hmR0VmxExitTaskSwitchNested(pVCpu, pVmxTransient);
13172	case VMX_EXIT_WBINVD: return hmR0VmxExitWbinvdNested(pVCpu, pVmxTransient);
13173	case VMX_EXIT_MTF: return hmR0VmxExitMtfNested(pVCpu, pVmxTransient);
13174	case VMX_EXIT_APIC_ACCESS: return hmR0VmxExitApicAccessNested(pVCpu, pVmxTransient);
13175	case VMX_EXIT_APIC_WRITE: return hmR0VmxExitApicWriteNested(pVCpu, pVmxTransient);
13176	case VMX_EXIT_VIRTUALIZED_EOI: return hmR0VmxExitVirtEoiNested(pVCpu, pVmxTransient);
13177	case VMX_EXIT_MOV_CRX: return hmR0VmxExitMovCRxNested(pVCpu, pVmxTransient);
13178	case VMX_EXIT_INT_WINDOW: return hmR0VmxExitIntWindowNested(pVCpu, pVmxTransient);
13179	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindowNested(pVCpu, pVmxTransient);
13180	case VMX_EXIT_TPR_BELOW_THRESHOLD: return hmR0VmxExitTprBelowThresholdNested(pVCpu, pVmxTransient);
13181	case VMX_EXIT_MWAIT: return hmR0VmxExitMwaitNested(pVCpu, pVmxTransient);
13182	case VMX_EXIT_MONITOR: return hmR0VmxExitMonitorNested(pVCpu, pVmxTransient);
13183	case VMX_EXIT_PAUSE: return hmR0VmxExitPauseNested(pVCpu, pVmxTransient);
13184
13185	case VMX_EXIT_PREEMPT_TIMER:
13186	{
13187	/** @todo NSTVMX: Preempt timer. */
13188	return hmR0VmxExitPreemptTimer(pVCpu, pVmxTransient);
13189	}
13190
13191	case VMX_EXIT_MOV_DRX: return hmR0VmxExitMovDRxNested(pVCpu, pVmxTransient);
13192	case VMX_EXIT_RDPMC: return hmR0VmxExitRdpmcNested(pVCpu, pVmxTransient);
13193
13194	case VMX_EXIT_VMREAD:
13195	case VMX_EXIT_VMWRITE: return hmR0VmxExitVmreadVmwriteNested(pVCpu, pVmxTransient);
13196
13197	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFaultNested(pVCpu, pVmxTransient);
13198	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestStateNested(pVCpu, pVmxTransient);
13199
13200	case VMX_EXIT_INIT_SIGNAL:
13201	case VMX_EXIT_SIPI:
13202	case VMX_EXIT_IO_SMI:
13203	case VMX_EXIT_SMI:
13204	case VMX_EXIT_ERR_MSR_LOAD:
13205	case VMX_EXIT_ERR_MACHINE_CHECK:
13206	case VMX_EXIT_PML_FULL:
13207	case VMX_EXIT_RSM:
13208	default:
13209	return hmR0VmxExitErrUnexpected(pVCpu, pVmxTransient);
13210	}
13211	}
13212	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13213
13214
13215	/** @name VM-exit helpers.
13216	* @{
13217	*/
13218	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13219	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= VM-exit helpers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
13220	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13221
13222	/** Macro for VM-exits called unexpectedly. */
13223	#define HMVMX_UNEXPECTED_EXIT_RET(a_pVCpu, a_HmError) \
13224	do { \
13225	(a_pVCpu)->hm.s.u32HMError = (a_HmError); \
13226	return VERR_VMX_UNEXPECTED_EXIT; \
13227	} while (0)
13228
13229	#ifdef VBOX_STRICT
13230	/* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */
13231	# define HMVMX_ASSERT_PREEMPT_CPUID_VAR() \
13232	RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId()
13233
13234	# define HMVMX_ASSERT_PREEMPT_CPUID() \
13235	do { \
13236	RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \
13237	AssertMsg(idAssertCpu == idAssertCpuNow, ("VMX %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \
13238	} while (0)
13239
13240	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13241	do { \
13242	AssertPtr((a_pVCpu)); \
13243	AssertPtr((a_pVmxTransient)); \
13244	Assert((a_pVmxTransient)->fVMEntryFailed == false); \
13245	Assert((a_pVmxTransient)->pVmcsInfo); \
13246	Assert(ASMIntAreEnabled()); \
13247	HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu); \
13248	HMVMX_ASSERT_PREEMPT_CPUID_VAR(); \
13249	Log4Func(("vcpu[%RU32]\n", (a_pVCpu)->idCpu)); \
13250	HMVMX_ASSERT_PREEMPT_SAFE(a_pVCpu); \
13251	if (!VMMRZCallRing3IsEnabled((a_pVCpu))) \
13252	HMVMX_ASSERT_PREEMPT_CPUID(); \
13253	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
13254	} while (0)
13255
13256	# define HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13257	do { \
13258	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient); \
13259	Assert((a_pVmxTransient)->fIsNestedGuest); \
13260	} while (0)
13261
13262	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13263	do { \
13264	Log4Func(("\n")); \
13265	} while (0)
13266	#else
13267	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13268	do { \
13269	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
13270	NOREF((a_pVCpu)); NOREF((a_pVmxTransient)); \
13271	} while (0)
13272
13273	# define HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) \
13274	do { HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient); } while (0)
13275
13276	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(a_pVCpu, a_pVmxTransient) do { } while (0)
13277	#endif
13278
13279	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13280	/** Macro that does the necessary privilege checks and intercepted VM-exits for
13281	* guests that attempted to execute a VMX instruction. */
13282	# define HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(a_pVCpu, a_uExitReason) \
13283	do \
13284	{ \
13285	VBOXSTRICTRC rcStrictTmp = hmR0VmxCheckExitDueToVmxInstr((a_pVCpu), (a_uExitReason)); \
13286	if (rcStrictTmp == VINF_SUCCESS) \
13287	{ /* likely */ } \
13288	else if (rcStrictTmp == VINF_HM_PENDING_XCPT) \
13289	{ \
13290	Assert((a_pVCpu)->hm.s.Event.fPending); \
13291	Log4Func(("Privilege checks failed -> %#x\n", VMX_ENTRY_INT_INFO_VECTOR((a_pVCpu)->hm.s.Event.u64IntInfo))); \
13292	return VINF_SUCCESS; \
13293	} \
13294	else \
13295	{ \
13296	int rcTmp = VBOXSTRICTRC_VAL(rcStrictTmp); \
13297	AssertMsgFailedReturn(("Unexpected failure. rc=%Rrc", rcTmp), rcTmp); \
13298	} \
13299	} while (0)
13300
13301	/** Macro that decodes a memory operand for an VM-exit caused by an instruction. */
13302	# define HMVMX_DECODE_MEM_OPERAND(a_pVCpu, a_uExitInstrInfo, a_uExitQual, a_enmMemAccess, a_pGCPtrEffAddr) \
13303	do \
13304	{ \
13305	VBOXSTRICTRC rcStrictTmp = hmR0VmxDecodeMemOperand((a_pVCpu), (a_uExitInstrInfo), (a_uExitQual), (a_enmMemAccess), \
13306	(a_pGCPtrEffAddr)); \
13307	if (rcStrictTmp == VINF_SUCCESS) \
13308	{ /* likely */ } \
13309	else if (rcStrictTmp == VINF_HM_PENDING_XCPT) \
13310	{ \
13311	uint8_t const uXcptTmp = VMX_ENTRY_INT_INFO_VECTOR((a_pVCpu)->hm.s.Event.u64IntInfo); \
13312	Log4Func(("Memory operand decoding failed, raising xcpt %#x\n", uXcptTmp)); \
13313	NOREF(uXcptTmp); \
13314	return VINF_SUCCESS; \
13315	} \
13316	else \
13317	{ \
13318	Log4Func(("hmR0VmxDecodeMemOperand failed. rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrictTmp))); \
13319	return rcStrictTmp; \
13320	} \
13321	} while (0)
13322	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13323
13324
13325	/**
13326	* Advances the guest RIP by the specified number of bytes.
13327	*
13328	* @param pVCpu The cross context virtual CPU structure.
13329	* @param cbInstr Number of bytes to advance the RIP by.
13330	*
13331	* @remarks No-long-jump zone!!!
13332	*/
13333	DECLINLINE(void) hmR0VmxAdvanceGuestRipBy(PVMCPUCC pVCpu, uint32_t cbInstr)
13334	{
13335	/* Advance the RIP. */
13336	pVCpu->cpum.GstCtx.rip += cbInstr;
13337	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
13338
13339	/* Update interrupt inhibition. */
13340	if ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)
13341	&& pVCpu->cpum.GstCtx.rip != EMGetInhibitInterruptsPC(pVCpu))
13342	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
13343	}
13344
13345
13346	/**
13347	* Advances the guest RIP after reading it from the VMCS.
13348	*
13349	* @returns VBox status code, no informational status codes.
13350	* @param pVCpu The cross context virtual CPU structure.
13351	* @param pVmxTransient The VMX-transient structure.
13352	*
13353	* @remarks No-long-jump zone!!!
13354	*/
13355	static int hmR0VmxAdvanceGuestRip(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13356	{
13357	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13358	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS);
13359	AssertRCReturn(rc, rc);
13360
13361	hmR0VmxAdvanceGuestRipBy(pVCpu, pVmxTransient->cbExitInstr);
13362	return VINF_SUCCESS;
13363	}
13364
13365
13366	/**
13367	* Handle a condition that occurred while delivering an event through the guest or
13368	* nested-guest IDT.
13369	*
13370	* @returns Strict VBox status code (i.e. informational status codes too).
13371	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
13372	* @retval VINF_HM_DOUBLE_FAULT if a \#DF condition was detected and we ought
13373	* to continue execution of the guest which will delivery the \#DF.
13374	* @retval VINF_EM_RESET if we detected a triple-fault condition.
13375	* @retval VERR_EM_GUEST_CPU_HANG if we detected a guest CPU hang.
13376	*
13377	* @param pVCpu The cross context virtual CPU structure.
13378	* @param pVmxTransient The VMX-transient structure.
13379	*
13380	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13381	* Additionally, HMVMX_READ_EXIT_QUALIFICATION is required if the VM-exit
13382	* is due to an EPT violation, PML full or SPP-related event.
13383	*
13384	* @remarks No-long-jump zone!!!
13385	*/
13386	static VBOXSTRICTRC hmR0VmxCheckExitDueToEventDelivery(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13387	{
13388	Assert(!pVCpu->hm.s.Event.fPending);
13389	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_XCPT_INFO);
13390	if ( pVmxTransient->uExitReason == VMX_EXIT_EPT_VIOLATION
13391	\|\| pVmxTransient->uExitReason == VMX_EXIT_PML_FULL
13392	\|\| pVmxTransient->uExitReason == VMX_EXIT_SPP_EVENT)
13393	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_EXIT_QUALIFICATION);
13394
13395	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
13396	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
13397	uint32_t const uIdtVectorInfo = pVmxTransient->uIdtVectoringInfo;
13398	uint32_t const uExitIntInfo = pVmxTransient->uExitIntInfo;
13399	if (VMX_IDT_VECTORING_INFO_IS_VALID(uIdtVectorInfo))
13400	{
13401	uint32_t const uIdtVector = VMX_IDT_VECTORING_INFO_VECTOR(uIdtVectorInfo);
13402	uint32_t const uIdtVectorType = VMX_IDT_VECTORING_INFO_TYPE(uIdtVectorInfo);
13403
13404	/*
13405	* If the event was a software interrupt (generated with INT n) or a software exception
13406	* (generated by INT3/INTO) or a privileged software exception (generated by INT1), we
13407	* can handle the VM-exit and continue guest execution which will re-execute the
13408	* instruction rather than re-injecting the exception, as that can cause premature
13409	* trips to ring-3 before injection and involve TRPM which currently has no way of
13410	* storing that these exceptions were caused by these instructions (ICEBP's #DB poses
13411	* the problem).
13412	*/
13413	IEMXCPTRAISE enmRaise;
13414	IEMXCPTRAISEINFO fRaiseInfo;
13415	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
13416	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
13417	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT)
13418	{
13419	enmRaise = IEMXCPTRAISE_REEXEC_INSTR;
13420	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
13421	}
13422	else if (VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo))
13423	{
13424	uint32_t const uExitVectorType = VMX_EXIT_INT_INFO_TYPE(uExitIntInfo);
13425	uint8_t const uExitVector = VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo);
13426	Assert(uExitVectorType == VMX_EXIT_INT_INFO_TYPE_HW_XCPT);
13427
13428	uint32_t const fIdtVectorFlags = hmR0VmxGetIemXcptFlags(uIdtVector, uIdtVectorType);
13429	uint32_t const fExitVectorFlags = hmR0VmxGetIemXcptFlags(uExitVector, uExitVectorType);
13430
13431	enmRaise = IEMEvaluateRecursiveXcpt(pVCpu, fIdtVectorFlags, uIdtVector, fExitVectorFlags, uExitVector, &fRaiseInfo);
13432
13433	/* Determine a vectoring #PF condition, see comment in hmR0VmxExitXcptPF(). */
13434	if (fRaiseInfo & (IEMXCPTRAISEINFO_EXT_INT_PF \| IEMXCPTRAISEINFO_NMI_PF))
13435	{
13436	pVmxTransient->fVectoringPF = true;
13437	enmRaise = IEMXCPTRAISE_PREV_EVENT;
13438	}
13439	}
13440	else
13441	{
13442	/*
13443	* If an exception or hardware interrupt delivery caused an EPT violation/misconfig or APIC access
13444	* VM-exit, then the VM-exit interruption-information will not be valid and we end up here.
13445	* It is sufficient to reflect the original event to the guest after handling the VM-exit.
13446	*/
13447	Assert( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
13448	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
13449	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT);
13450	enmRaise = IEMXCPTRAISE_PREV_EVENT;
13451	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
13452	}
13453
13454	/*
13455	* On CPUs that support Virtual NMIs, if this VM-exit (be it an exception or EPT violation/misconfig
13456	* etc.) occurred while delivering the NMI, we need to clear the block-by-NMI field in the guest
13457	* interruptibility-state before re-delivering the NMI after handling the VM-exit. Otherwise the
13458	* subsequent VM-entry would fail, see @bugref{7445}.
13459	*
13460	* See Intel spec. 30.7.1.2 "Resuming Guest Software after Handling an Exception".
13461	*/
13462	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
13463	&& enmRaise == IEMXCPTRAISE_PREV_EVENT
13464	&& (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
13465	&& CPUMIsGuestNmiBlocking(pVCpu))
13466	{
13467	CPUMSetGuestNmiBlocking(pVCpu, false);
13468	}
13469
13470	switch (enmRaise)
13471	{
13472	case IEMXCPTRAISE_CURRENT_XCPT:
13473	{
13474	Log4Func(("IDT: Pending secondary Xcpt: idtinfo=%#RX64 exitinfo=%#RX64\n", uIdtVectorInfo, uExitIntInfo));
13475	Assert(rcStrict == VINF_SUCCESS);
13476	break;
13477	}
13478
13479	case IEMXCPTRAISE_PREV_EVENT:
13480	{
13481	uint32_t u32ErrCode;
13482	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(uIdtVectorInfo))
13483	u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
13484	else
13485	u32ErrCode = 0;
13486
13487	/* If uExitVector is #PF, CR2 value will be updated from the VMCS if it's a guest #PF, see hmR0VmxExitXcptPF(). */
13488	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflect);
13489	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(uIdtVectorInfo), 0 /* cbInstr */,
13490	u32ErrCode, pVCpu->cpum.GstCtx.cr2);
13491
13492	Log4Func(("IDT: Pending vectoring event %#RX64 Err=%#RX32\n", pVCpu->hm.s.Event.u64IntInfo,
13493	pVCpu->hm.s.Event.u32ErrCode));
13494	Assert(rcStrict == VINF_SUCCESS);
13495	break;
13496	}
13497
13498	case IEMXCPTRAISE_REEXEC_INSTR:
13499	Assert(rcStrict == VINF_SUCCESS);
13500	break;
13501
13502	case IEMXCPTRAISE_DOUBLE_FAULT:
13503	{
13504	/*
13505	* Determing a vectoring double #PF condition. Used later, when PGM evaluates the
13506	* second #PF as a guest #PF (and not a shadow #PF) and needs to be converted into a #DF.
13507	*/
13508	if (fRaiseInfo & IEMXCPTRAISEINFO_PF_PF)
13509	{
13510	pVmxTransient->fVectoringDoublePF = true;
13511	Log4Func(("IDT: Vectoring double #PF %#RX64 cr2=%#RX64\n", pVCpu->hm.s.Event.u64IntInfo,
13512	pVCpu->cpum.GstCtx.cr2));
13513	rcStrict = VINF_SUCCESS;
13514	}
13515	else
13516	{
13517	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectConvertDF);
13518	hmR0VmxSetPendingXcptDF(pVCpu);
13519	Log4Func(("IDT: Pending vectoring #DF %#RX64 uIdtVector=%#x uExitVector=%#x\n", pVCpu->hm.s.Event.u64IntInfo,
13520	uIdtVector, VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo)));
13521	rcStrict = VINF_HM_DOUBLE_FAULT;
13522	}
13523	break;
13524	}
13525
13526	case IEMXCPTRAISE_TRIPLE_FAULT:
13527	{
13528	Log4Func(("IDT: Pending vectoring triple-fault uIdt=%#x uExit=%#x\n", uIdtVector,
13529	VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo)));
13530	rcStrict = VINF_EM_RESET;
13531	break;
13532	}
13533
13534	case IEMXCPTRAISE_CPU_HANG:
13535	{
13536	Log4Func(("IDT: Bad guest! Entering CPU hang. fRaiseInfo=%#x\n", fRaiseInfo));
13537	rcStrict = VERR_EM_GUEST_CPU_HANG;
13538	break;
13539	}
13540
13541	default:
13542	{
13543	AssertMsgFailed(("IDT: vcpu[%RU32] Unexpected/invalid value! enmRaise=%#x\n", pVCpu->idCpu, enmRaise));
13544	rcStrict = VERR_VMX_IPE_2;
13545	break;
13546	}
13547	}
13548	}
13549	else if ( (pVmcsInfo->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
13550	&& !CPUMIsGuestNmiBlocking(pVCpu))
13551	{
13552	if ( VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo)
13553	&& VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo) != X86_XCPT_DF
13554	&& VMX_EXIT_INT_INFO_IS_NMI_UNBLOCK_IRET(uExitIntInfo))
13555	{
13556	/*
13557	* Execution of IRET caused a fault when NMI blocking was in effect (i.e we're in
13558	* the guest or nested-guest NMI handler). We need to set the block-by-NMI field so
13559	* that virtual NMIs remain blocked until the IRET execution is completed.
13560	*
13561	* See Intel spec. 31.7.1.2 "Resuming Guest Software After Handling An Exception".
13562	*/
13563	CPUMSetGuestNmiBlocking(pVCpu, true);
13564	Log4Func(("Set NMI blocking. uExitReason=%u\n", pVmxTransient->uExitReason));
13565	}
13566	else if ( pVmxTransient->uExitReason == VMX_EXIT_EPT_VIOLATION
13567	\|\| pVmxTransient->uExitReason == VMX_EXIT_PML_FULL
13568	\|\| pVmxTransient->uExitReason == VMX_EXIT_SPP_EVENT)
13569	{
13570	/*
13571	* Execution of IRET caused an EPT violation, page-modification log-full event or
13572	* SPP-related event VM-exit when NMI blocking was in effect (i.e. we're in the
13573	* guest or nested-guest NMI handler). We need to set the block-by-NMI field so
13574	* that virtual NMIs remain blocked until the IRET execution is completed.
13575	*
13576	* See Intel spec. 27.2.3 "Information about NMI unblocking due to IRET"
13577	*/
13578	if (VMX_EXIT_QUAL_EPT_IS_NMI_UNBLOCK_IRET(pVmxTransient->uExitQual))
13579	{
13580	CPUMSetGuestNmiBlocking(pVCpu, true);
13581	Log4Func(("Set NMI blocking. uExitReason=%u\n", pVmxTransient->uExitReason));
13582	}
13583	}
13584	}
13585
13586	Assert( rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_HM_DOUBLE_FAULT
13587	\|\| rcStrict == VINF_EM_RESET \|\| rcStrict == VERR_EM_GUEST_CPU_HANG);
13588	return rcStrict;
13589	}
13590
13591
13592	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
13593	/**
13594	* Perform the relevant VMX instruction checks for VM-exits that occurred due to the
13595	* guest attempting to execute a VMX instruction.
13596	*
13597	* @returns Strict VBox status code (i.e. informational status codes too).
13598	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
13599	* @retval VINF_HM_PENDING_XCPT if an exception was raised.
13600	*
13601	* @param pVCpu The cross context virtual CPU structure.
13602	* @param uExitReason The VM-exit reason.
13603	*
13604	* @todo NSTVMX: Document other error codes when VM-exit is implemented.
13605	* @remarks No-long-jump zone!!!
13606	*/
13607	static VBOXSTRICTRC hmR0VmxCheckExitDueToVmxInstr(PVMCPUCC pVCpu, uint32_t uExitReason)
13608	{
13609	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS
13610	\| CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_EFER);
13611
13612	/*
13613	* The physical CPU would have already checked the CPU mode/code segment.
13614	* We shall just assert here for paranoia.
13615	* See Intel spec. 25.1.1 "Relative Priority of Faults and VM Exits".
13616	*/
13617	Assert(!CPUMIsGuestInRealOrV86ModeEx(&pVCpu->cpum.GstCtx));
13618	Assert( !CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx)
13619	\|\| CPUMIsGuestIn64BitCodeEx(&pVCpu->cpum.GstCtx));
13620
13621	if (uExitReason == VMX_EXIT_VMXON)
13622	{
13623	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
13624
13625	/*
13626	* We check CR4.VMXE because it is required to be always set while in VMX operation
13627	* by physical CPUs and our CR4 read-shadow is only consulted when executing specific
13628	* instructions (CLTS, LMSW, MOV CR, and SMSW) and thus doesn't affect CPU operation
13629	* otherwise (i.e. physical CPU won't automatically #UD if Cr4Shadow.VMXE is 0).
13630	*/
13631	if (!CPUMIsGuestVmxEnabled(&pVCpu->cpum.GstCtx))
13632	{
13633	Log4Func(("CR4.VMXE is not set -> #UD\n"));
13634	hmR0VmxSetPendingXcptUD(pVCpu);
13635	return VINF_HM_PENDING_XCPT;
13636	}
13637	}
13638	else if (!CPUMIsGuestInVmxRootMode(&pVCpu->cpum.GstCtx))
13639	{
13640	/*
13641	* The guest has not entered VMX operation but attempted to execute a VMX instruction
13642	* (other than VMXON), we need to raise a #UD.
13643	*/
13644	Log4Func(("Not in VMX root mode -> #UD\n"));
13645	hmR0VmxSetPendingXcptUD(pVCpu);
13646	return VINF_HM_PENDING_XCPT;
13647	}
13648
13649	/* All other checks (including VM-exit intercepts) are handled by IEM instruction emulation. */
13650	return VINF_SUCCESS;
13651	}
13652
13653
13654	/**
13655	* Decodes the memory operand of an instruction that caused a VM-exit.
13656	*
13657	* The Exit qualification field provides the displacement field for memory
13658	* operand instructions, if any.
13659	*
13660	* @returns Strict VBox status code (i.e. informational status codes too).
13661	* @retval VINF_SUCCESS if the operand was successfully decoded.
13662	* @retval VINF_HM_PENDING_XCPT if an exception was raised while decoding the
13663	* operand.
13664	* @param pVCpu The cross context virtual CPU structure.
13665	* @param uExitInstrInfo The VM-exit instruction information field.
13666	* @param enmMemAccess The memory operand's access type (read or write).
13667	* @param GCPtrDisp The instruction displacement field, if any. For
13668	* RIP-relative addressing pass RIP + displacement here.
13669	* @param pGCPtrMem Where to store the effective destination memory address.
13670	*
13671	* @remarks Warning! This function ASSUMES the instruction cannot be used in real or
13672	* virtual-8086 mode hence skips those checks while verifying if the
13673	* segment is valid.
13674	*/
13675	static VBOXSTRICTRC hmR0VmxDecodeMemOperand(PVMCPUCC pVCpu, uint32_t uExitInstrInfo, RTGCPTR GCPtrDisp, VMXMEMACCESS enmMemAccess,
13676	PRTGCPTR pGCPtrMem)
13677	{
13678	Assert(pGCPtrMem);
13679	Assert(!CPUMIsGuestInRealOrV86Mode(pVCpu));
13680	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_EFER
13681	\| CPUMCTX_EXTRN_CR0);
13682
13683	static uint64_t const s_auAddrSizeMasks[] = { UINT64_C(0xffff), UINT64_C(0xffffffff), UINT64_C(0xffffffffffffffff) };
13684	static uint64_t const s_auAccessSizeMasks[] = { sizeof(uint16_t), sizeof(uint32_t), sizeof(uint64_t) };
13685	AssertCompile(RT_ELEMENTS(s_auAccessSizeMasks) == RT_ELEMENTS(s_auAddrSizeMasks));
13686
13687	VMXEXITINSTRINFO ExitInstrInfo;
13688	ExitInstrInfo.u = uExitInstrInfo;
13689	uint8_t const uAddrSize = ExitInstrInfo.All.u3AddrSize;
13690	uint8_t const iSegReg = ExitInstrInfo.All.iSegReg;
13691	bool const fIdxRegValid = !ExitInstrInfo.All.fIdxRegInvalid;
13692	uint8_t const iIdxReg = ExitInstrInfo.All.iIdxReg;
13693	uint8_t const uScale = ExitInstrInfo.All.u2Scaling;
13694	bool const fBaseRegValid = !ExitInstrInfo.All.fBaseRegInvalid;
13695	uint8_t const iBaseReg = ExitInstrInfo.All.iBaseReg;
13696	bool const fIsMemOperand = !ExitInstrInfo.All.fIsRegOperand;
13697	bool const fIsLongMode = CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx);
13698
13699	/*
13700	* Validate instruction information.
13701	* This shouldn't happen on real hardware but useful while testing our nested hardware-virtualization code.
13702	*/
13703	AssertLogRelMsgReturn(uAddrSize < RT_ELEMENTS(s_auAddrSizeMasks),
13704	("Invalid address size. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_1);
13705	AssertLogRelMsgReturn(iSegReg < X86_SREG_COUNT,
13706	("Invalid segment register. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_2);
13707	AssertLogRelMsgReturn(fIsMemOperand,
13708	("Expected memory operand. ExitInstrInfo=%#RX32\n", ExitInstrInfo.u), VERR_VMX_IPE_3);
13709
13710	/*
13711	* Compute the complete effective address.
13712	*
13713	* See AMD instruction spec. 1.4.2 "SIB Byte Format"
13714	* See AMD spec. 4.5.2 "Segment Registers".
13715	*/
13716	RTGCPTR GCPtrMem = GCPtrDisp;
13717	if (fBaseRegValid)
13718	GCPtrMem += pVCpu->cpum.GstCtx.aGRegs[iBaseReg].u64;
13719	if (fIdxRegValid)
13720	GCPtrMem += pVCpu->cpum.GstCtx.aGRegs[iIdxReg].u64 << uScale;
13721
13722	RTGCPTR const GCPtrOff = GCPtrMem;
13723	if ( !fIsLongMode
13724	\|\| iSegReg >= X86_SREG_FS)
13725	GCPtrMem += pVCpu->cpum.GstCtx.aSRegs[iSegReg].u64Base;
13726	GCPtrMem &= s_auAddrSizeMasks[uAddrSize];
13727
13728	/*
13729	* Validate effective address.
13730	* See AMD spec. 4.5.3 "Segment Registers in 64-Bit Mode".
13731	*/
13732	uint8_t const cbAccess = s_auAccessSizeMasks[uAddrSize];
13733	Assert(cbAccess > 0);
13734	if (fIsLongMode)
13735	{
13736	if (X86_IS_CANONICAL(GCPtrMem))
13737	{
13738	*pGCPtrMem = GCPtrMem;
13739	return VINF_SUCCESS;
13740	}
13741
13742	/** @todo r=ramshankar: We should probably raise \#SS or \#GP. See AMD spec. 4.12.2
13743	* "Data Limit Checks in 64-bit Mode". */
13744	Log4Func(("Long mode effective address is not canonical GCPtrMem=%#RX64\n", GCPtrMem));
13745	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13746	return VINF_HM_PENDING_XCPT;
13747	}
13748
13749	/*
13750	* This is a watered down version of iemMemApplySegment().
13751	* Parts that are not applicable for VMX instructions like real-or-v8086 mode
13752	* and segment CPL/DPL checks are skipped.
13753	*/
13754	RTGCPTR32 const GCPtrFirst32 = (RTGCPTR32)GCPtrOff;
13755	RTGCPTR32 const GCPtrLast32 = GCPtrFirst32 + cbAccess - 1;
13756	PCCPUMSELREG pSel = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
13757
13758	/* Check if the segment is present and usable. */
13759	if ( pSel->Attr.n.u1Present
13760	&& !pSel->Attr.n.u1Unusable)
13761	{
13762	Assert(pSel->Attr.n.u1DescType);
13763	if (!(pSel->Attr.n.u4Type & X86_SEL_TYPE_CODE))
13764	{
13765	/* Check permissions for the data segment. */
13766	if ( enmMemAccess == VMXMEMACCESS_WRITE
13767	&& !(pSel->Attr.n.u4Type & X86_SEL_TYPE_WRITE))
13768	{
13769	Log4Func(("Data segment access invalid. iSegReg=%#x Attr=%#RX32\n", iSegReg, pSel->Attr.u));
13770	hmR0VmxSetPendingXcptGP(pVCpu, iSegReg);
13771	return VINF_HM_PENDING_XCPT;
13772	}
13773
13774	/* Check limits if it's a normal data segment. */
13775	if (!(pSel->Attr.n.u4Type & X86_SEL_TYPE_DOWN))
13776	{
13777	if ( GCPtrFirst32 > pSel->u32Limit
13778	\|\| GCPtrLast32 > pSel->u32Limit)
13779	{
13780	Log4Func(("Data segment limit exceeded. "
13781	"iSegReg=%#x GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n", iSegReg, GCPtrFirst32,
13782	GCPtrLast32, pSel->u32Limit));
13783	if (iSegReg == X86_SREG_SS)
13784	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13785	else
13786	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13787	return VINF_HM_PENDING_XCPT;
13788	}
13789	}
13790	else
13791	{
13792	/* Check limits if it's an expand-down data segment.
13793	Note! The upper boundary is defined by the B bit, not the G bit! */
13794	if ( GCPtrFirst32 < pSel->u32Limit + UINT32_C(1)
13795	\|\| GCPtrLast32 > (pSel->Attr.n.u1DefBig ? UINT32_MAX : UINT32_C(0xffff)))
13796	{
13797	Log4Func(("Expand-down data segment limit exceeded. "
13798	"iSegReg=%#x GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n", iSegReg, GCPtrFirst32,
13799	GCPtrLast32, pSel->u32Limit));
13800	if (iSegReg == X86_SREG_SS)
13801	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13802	else
13803	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13804	return VINF_HM_PENDING_XCPT;
13805	}
13806	}
13807	}
13808	else
13809	{
13810	/* Check permissions for the code segment. */
13811	if ( enmMemAccess == VMXMEMACCESS_WRITE
13812	\|\| ( enmMemAccess == VMXMEMACCESS_READ
13813	&& !(pSel->Attr.n.u4Type & X86_SEL_TYPE_READ)))
13814	{
13815	Log4Func(("Code segment access invalid. Attr=%#RX32\n", pSel->Attr.u));
13816	Assert(!CPUMIsGuestInRealOrV86ModeEx(&pVCpu->cpum.GstCtx));
13817	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13818	return VINF_HM_PENDING_XCPT;
13819	}
13820
13821	/* Check limits for the code segment (normal/expand-down not applicable for code segments). */
13822	if ( GCPtrFirst32 > pSel->u32Limit
13823	\|\| GCPtrLast32 > pSel->u32Limit)
13824	{
13825	Log4Func(("Code segment limit exceeded. GCPtrFirst32=%#RX32 GCPtrLast32=%#RX32 u32Limit=%#RX32\n",
13826	GCPtrFirst32, GCPtrLast32, pSel->u32Limit));
13827	if (iSegReg == X86_SREG_SS)
13828	hmR0VmxSetPendingXcptSS(pVCpu, 0);
13829	else
13830	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13831	return VINF_HM_PENDING_XCPT;
13832	}
13833	}
13834	}
13835	else
13836	{
13837	Log4Func(("Not present or unusable segment. iSegReg=%#x Attr=%#RX32\n", iSegReg, pSel->Attr.u));
13838	hmR0VmxSetPendingXcptGP(pVCpu, 0);
13839	return VINF_HM_PENDING_XCPT;
13840	}
13841
13842	*pGCPtrMem = GCPtrMem;
13843	return VINF_SUCCESS;
13844	}
13845	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
13846
13847
13848	/**
13849	* VM-exit helper for LMSW.
13850	*/
13851	static VBOXSTRICTRC hmR0VmxExitLmsw(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint16_t uMsw, RTGCPTR GCPtrEffDst)
13852	{
13853	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13854	AssertRCReturn(rc, rc);
13855
13856	VBOXSTRICTRC rcStrict = IEMExecDecodedLmsw(pVCpu, cbInstr, uMsw, GCPtrEffDst);
13857	AssertMsg( rcStrict == VINF_SUCCESS
13858	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13859
13860	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
13861	if (rcStrict == VINF_IEM_RAISED_XCPT)
13862	{
13863	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13864	rcStrict = VINF_SUCCESS;
13865	}
13866
13867	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitLmsw);
13868	Log4Func(("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13869	return rcStrict;
13870	}
13871
13872
13873	/**
13874	* VM-exit helper for CLTS.
13875	*/
13876	static VBOXSTRICTRC hmR0VmxExitClts(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr)
13877	{
13878	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13879	AssertRCReturn(rc, rc);
13880
13881	VBOXSTRICTRC rcStrict = IEMExecDecodedClts(pVCpu, cbInstr);
13882	AssertMsg( rcStrict == VINF_SUCCESS
13883	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13884
13885	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0);
13886	if (rcStrict == VINF_IEM_RAISED_XCPT)
13887	{
13888	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13889	rcStrict = VINF_SUCCESS;
13890	}
13891
13892	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitClts);
13893	Log4Func(("rcStrict=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13894	return rcStrict;
13895	}
13896
13897
13898	/**
13899	* VM-exit helper for MOV from CRx (CRx read).
13900	*/
13901	static VBOXSTRICTRC hmR0VmxExitMovFromCrX(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint8_t iGReg, uint8_t iCrReg)
13902	{
13903	Assert(iCrReg < 16);
13904	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
13905
13906	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13907	AssertRCReturn(rc, rc);
13908
13909	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxRead(pVCpu, cbInstr, iGReg, iCrReg);
13910	AssertMsg( rcStrict == VINF_SUCCESS
13911	\|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13912
13913	if (iGReg == X86_GREG_xSP)
13914	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_RSP);
13915	else
13916	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
13917	#ifdef VBOX_WITH_STATISTICS
13918	switch (iCrReg)
13919	{
13920	case 0: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Read); break;
13921	case 2: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Read); break;
13922	case 3: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Read); break;
13923	case 4: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Read); break;
13924	case 8: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Read); break;
13925	}
13926	#endif
13927	Log4Func(("CR%d Read access rcStrict=%Rrc\n", iCrReg, VBOXSTRICTRC_VAL(rcStrict)));
13928	return rcStrict;
13929	}
13930
13931
13932	/**
13933	* VM-exit helper for MOV to CRx (CRx write).
13934	*/
13935	static VBOXSTRICTRC hmR0VmxExitMovToCrX(PVMCPUCC pVCpu, PVMXVMCSINFO pVmcsInfo, uint8_t cbInstr, uint8_t iGReg, uint8_t iCrReg)
13936	{
13937	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
13938	AssertRCReturn(rc, rc);
13939
13940	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxWrite(pVCpu, cbInstr, iCrReg, iGReg);
13941	AssertMsg( rcStrict == VINF_SUCCESS
13942	\|\| rcStrict == VINF_IEM_RAISED_XCPT
13943	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
13944
13945	switch (iCrReg)
13946	{
13947	case 0:
13948	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR0
13949	\| HM_CHANGED_GUEST_EFER_MSR \| HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
13950	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Write);
13951	Log4Func(("CR0 write. rcStrict=%Rrc CR0=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cr0));
13952	break;
13953
13954	case 2:
13955	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Write);
13956	/* Nothing to do here, CR2 it's not part of the VMCS. */
13957	break;
13958
13959	case 3:
13960	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR3);
13961	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Write);
13962	Log4Func(("CR3 write. rcStrict=%Rrc CR3=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cr3));
13963	break;
13964
13965	case 4:
13966	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_CR4);
13967	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Write);
13968	Log4Func(("CR4 write. rc=%Rrc CR4=%#RX64 fLoadSaveGuestXcr0=%u\n", VBOXSTRICTRC_VAL(rcStrict),
13969	pVCpu->cpum.GstCtx.cr4, pVCpu->hmr0.s.fLoadSaveGuestXcr0));
13970	break;
13971
13972	case 8:
13973	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged,
13974	HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_APIC_TPR);
13975	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Write);
13976	break;
13977
13978	default:
13979	AssertMsgFailed(("Invalid CRx register %#x\n", iCrReg));
13980	break;
13981	}
13982
13983	if (rcStrict == VINF_IEM_RAISED_XCPT)
13984	{
13985	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
13986	rcStrict = VINF_SUCCESS;
13987	}
13988	return rcStrict;
13989	}
13990
13991
13992	/**
13993	* VM-exit exception handler for \#PF (Page-fault exception).
13994	*
13995	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
13996	*/
13997	static VBOXSTRICTRC hmR0VmxExitXcptPF(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
13998	{
13999	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14000	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
14001	hmR0VmxReadExitQualVmcs(pVmxTransient);
14002
14003	if (!pVM->hmr0.s.fNestedPaging)
14004	{ /* likely */ }
14005	else
14006	{
14007	#if !defined(HMVMX_ALWAYS_TRAP_ALL_XCPTS) && !defined(HMVMX_ALWAYS_TRAP_PF)
14008	Assert(pVmxTransient->fIsNestedGuest \|\| pVCpu->hmr0.s.fUsingDebugLoop);
14009	#endif
14010	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
14011	if (!pVmxTransient->fVectoringDoublePF)
14012	{
14013	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */,
14014	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual);
14015	}
14016	else
14017	{
14018	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
14019	Assert(!pVmxTransient->fIsNestedGuest);
14020	hmR0VmxSetPendingXcptDF(pVCpu);
14021	Log4Func(("Pending #DF due to vectoring #PF w/ NestedPaging\n"));
14022	}
14023	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
14024	return VINF_SUCCESS;
14025	}
14026
14027	Assert(!pVmxTransient->fIsNestedGuest);
14028
14029	/* If it's a vectoring #PF, emulate injecting the original event injection as PGMTrap0eHandler() is incapable
14030	of differentiating between instruction emulation and event injection that caused a #PF. See @bugref{6607}. */
14031	if (pVmxTransient->fVectoringPF)
14032	{
14033	Assert(pVCpu->hm.s.Event.fPending);
14034	return VINF_EM_RAW_INJECT_TRPM_EVENT;
14035	}
14036
14037	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14038	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14039	AssertRCReturn(rc, rc);
14040
14041	Log4Func(("#PF: cs:rip=%#04x:%#RX64 err_code=%#RX32 exit_qual=%#RX64 cr3=%#RX64\n", pCtx->cs.Sel, pCtx->rip,
14042	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual, pCtx->cr3));
14043
14044	TRPMAssertXcptPF(pVCpu, pVmxTransient->uExitQual, (RTGCUINT)pVmxTransient->uExitIntErrorCode);
14045	rc = PGMTrap0eHandler(pVCpu, pVmxTransient->uExitIntErrorCode, CPUMCTX2CORE(pCtx), (RTGCPTR)pVmxTransient->uExitQual);
14046
14047	Log4Func(("#PF: rc=%Rrc\n", rc));
14048	if (rc == VINF_SUCCESS)
14049	{
14050	/*
14051	* This is typically a shadow page table sync or a MMIO instruction. But we may have
14052	* emulated something like LTR or a far jump. Any part of the CPU context may have changed.
14053	*/
14054	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14055	TRPMResetTrap(pVCpu);
14056	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
14057	return rc;
14058	}
14059
14060	if (rc == VINF_EM_RAW_GUEST_TRAP)
14061	{
14062	if (!pVmxTransient->fVectoringDoublePF)
14063	{
14064	/* It's a guest page fault and needs to be reflected to the guest. */
14065	uint32_t const uGstErrorCode = TRPMGetErrorCode(pVCpu);
14066	TRPMResetTrap(pVCpu);
14067	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory #PF. */
14068	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), 0 /* cbInstr */,
14069	uGstErrorCode, pVmxTransient->uExitQual);
14070	}
14071	else
14072	{
14073	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
14074	TRPMResetTrap(pVCpu);
14075	pVCpu->hm.s.Event.fPending = false; /* Clear pending #PF to replace it with #DF. */
14076	hmR0VmxSetPendingXcptDF(pVCpu);
14077	Log4Func(("#PF: Pending #DF due to vectoring #PF\n"));
14078	}
14079
14080	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
14081	return VINF_SUCCESS;
14082	}
14083
14084	TRPMResetTrap(pVCpu);
14085	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
14086	return rc;
14087	}
14088
14089
14090	/**
14091	* VM-exit exception handler for \#MF (Math Fault: floating point exception).
14092	*
14093	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14094	*/
14095	static VBOXSTRICTRC hmR0VmxExitXcptMF(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14096	{
14097	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14098	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
14099
14100	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR0);
14101	AssertRCReturn(rc, rc);
14102
14103	if (!(pVCpu->cpum.GstCtx.cr0 & X86_CR0_NE))
14104	{
14105	/* Convert a #MF into a FERR -> IRQ 13. See @bugref{6117}. */
14106	rc = PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13, 1, 0 /* uTagSrc */);
14107
14108	/** @todo r=ramshankar: The Intel spec. does -not- specify that this VM-exit
14109	* provides VM-exit instruction length. If this causes problem later,
14110	* disassemble the instruction like it's done on AMD-V. */
14111	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14112	AssertRCReturn(rc2, rc2);
14113	return rc;
14114	}
14115
14116	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo), pVmxTransient->cbExitInstr,
14117	pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14118	return VINF_SUCCESS;
14119	}
14120
14121
14122	/**
14123	* VM-exit exception handler for \#BP (Breakpoint exception).
14124	*
14125	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14126	*/
14127	static VBOXSTRICTRC hmR0VmxExitXcptBP(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14128	{
14129	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14130	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP);
14131
14132	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14133	AssertRCReturn(rc, rc);
14134
14135	if (!pVmxTransient->fIsNestedGuest)
14136	rc = DBGFTrap03Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(&pVCpu->cpum.GstCtx));
14137	else
14138	rc = VINF_EM_RAW_GUEST_TRAP;
14139
14140	if (rc == VINF_EM_RAW_GUEST_TRAP)
14141	{
14142	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14143	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14144	rc = VINF_SUCCESS;
14145	}
14146
14147	Assert(rc == VINF_SUCCESS \|\| rc == VINF_EM_DBG_BREAKPOINT);
14148	return rc;
14149	}
14150
14151
14152	/**
14153	* VM-exit exception handler for \#AC (Alignment-check exception).
14154	*
14155	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14156	*/
14157	static VBOXSTRICTRC hmR0VmxExitXcptAC(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14158	{
14159	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14160
14161	/*
14162	* Detect #ACs caused by host having enabled split-lock detection.
14163	* Emulate such instructions.
14164	*/
14165	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo,
14166	CPUMCTX_EXTRN_CR0 \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS \| CPUMCTX_EXTRN_CS);
14167	AssertRCReturn(rc, rc);
14168	/** @todo detect split lock in cpu feature? */
14169	if ( /* 1. If 486-style alignment checks aren't enabled, then this must be a split-lock exception */
14170	!(pVCpu->cpum.GstCtx.cr0 & X86_CR0_AM)
14171	/* 2. #AC cannot happen in rings 0-2 except for split-lock detection. */
14172	\|\| CPUMGetGuestCPL(pVCpu) != 3
14173	/* 3. When the EFLAGS.AC != 0 this can only be a split-lock case. */
14174	\|\| !(pVCpu->cpum.GstCtx.eflags.u & X86_EFL_AC) )
14175	{
14176	/*
14177	* Check for debug/trace events and import state accordingly.
14178	*/
14179	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitGuestACSplitLock);
14180	PVMCC pVM = pVCpu->pVMR0;
14181	if ( !DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_VMX_SPLIT_LOCK)
14182	&& !VBOXVMM_VMX_SPLIT_LOCK_ENABLED())
14183	{
14184	if (pVM->cCpus == 1)
14185	{
14186	#if 0 /** @todo r=bird: This is potentially wrong. Might have to just do a whole state sync above and mark everything changed to be safe... */
14187	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14188	#else
14189	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14190	#endif
14191	AssertRCReturn(rc, rc);
14192	}
14193	}
14194	else
14195	{
14196	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14197	AssertRCReturn(rc, rc);
14198
14199	VBOXVMM_XCPT_DF(pVCpu, &pVCpu->cpum.GstCtx);
14200
14201	if (DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_VMX_SPLIT_LOCK))
14202	{
14203	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, DBGFEVENT_VMX_SPLIT_LOCK, DBGFEVENTCTX_HM, 0);
14204	if (rcStrict != VINF_SUCCESS)
14205	return rcStrict;
14206	}
14207	}
14208
14209	/*
14210	* Emulate the instruction.
14211	*
14212	* We have to ignore the LOCK prefix here as we must not retrigger the
14213	* detection on the host. This isn't all that satisfactory, though...
14214	*/
14215	if (pVM->cCpus == 1)
14216	{
14217	Log8Func(("cs:rip=%#04x:%#RX64 rflags=%#RX64 cr0=%#RX64 split-lock #AC\n", pVCpu->cpum.GstCtx.cs.Sel,
14218	pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.rflags, pVCpu->cpum.GstCtx.cr0));
14219
14220	/** @todo For SMP configs we should do a rendezvous here. */
14221	VBOXSTRICTRC rcStrict = IEMExecOneIgnoreLock(pVCpu);
14222	if (rcStrict == VINF_SUCCESS)
14223	#if 0 /** @todo r=bird: This is potentially wrong. Might have to just do a whole state sync above and mark everything changed to be safe... */
14224	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged,
14225	HM_CHANGED_GUEST_RIP
14226	\| HM_CHANGED_GUEST_RFLAGS
14227	\| HM_CHANGED_GUEST_GPRS_MASK
14228	\| HM_CHANGED_GUEST_CS
14229	\| HM_CHANGED_GUEST_SS);
14230	#else
14231	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14232	#endif
14233	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14234	{
14235	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14236	rcStrict = VINF_SUCCESS;
14237	}
14238	return rcStrict;
14239	}
14240	Log8Func(("cs:rip=%#04x:%#RX64 rflags=%#RX64 cr0=%#RX64 split-lock #AC -> VINF_EM_EMULATE_SPLIT_LOCK\n",
14241	pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.rflags, pVCpu->cpum.GstCtx.cr0));
14242	return VINF_EM_EMULATE_SPLIT_LOCK;
14243	}
14244
14245	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC);
14246	Log8Func(("cs:rip=%#04x:%#RX64 rflags=%#RX64 cr0=%#RX64 cpl=%d -> #AC\n", pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
14247	pVCpu->cpum.GstCtx.rflags, pVCpu->cpum.GstCtx.cr0, CPUMGetGuestCPL(pVCpu) ));
14248
14249	/* Re-inject it. We'll detect any nesting before getting here. */
14250	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14251	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14252	return VINF_SUCCESS;
14253	}
14254
14255
14256	/**
14257	* VM-exit exception handler for \#DB (Debug exception).
14258	*
14259	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14260	*/
14261	static VBOXSTRICTRC hmR0VmxExitXcptDB(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14262	{
14263	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14264	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
14265
14266	/*
14267	* Get the DR6-like values from the Exit qualification and pass it to DBGF for processing.
14268	*/
14269	hmR0VmxReadExitQualVmcs(pVmxTransient);
14270
14271	/* Refer Intel spec. Table 27-1. "Exit Qualifications for debug exceptions" for the format. */
14272	uint64_t const uDR6 = X86_DR6_INIT_VAL
14273	\| (pVmxTransient->uExitQual & ( X86_DR6_B0 \| X86_DR6_B1 \| X86_DR6_B2 \| X86_DR6_B3
14274	\| X86_DR6_BD \| X86_DR6_BS));
14275
14276	int rc;
14277	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14278	if (!pVmxTransient->fIsNestedGuest)
14279	{
14280	rc = DBGFTrap01Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx), uDR6, pVCpu->hm.s.fSingleInstruction);
14281
14282	/*
14283	* Prevents stepping twice over the same instruction when the guest is stepping using
14284	* EFLAGS.TF and the hypervisor debugger is stepping using MTF.
14285	* Testcase: DOSQEMM, break (using "ba x 1") at cs:rip 0x70:0x774 and step (using "t").
14286	*/
14287	if ( rc == VINF_EM_DBG_STEPPED
14288	&& (pVmxTransient->pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_MONITOR_TRAP_FLAG))
14289	{
14290	Assert(pVCpu->hm.s.fSingleInstruction);
14291	rc = VINF_EM_RAW_GUEST_TRAP;
14292	}
14293	}
14294	else
14295	rc = VINF_EM_RAW_GUEST_TRAP;
14296	Log6Func(("rc=%Rrc\n", rc));
14297	if (rc == VINF_EM_RAW_GUEST_TRAP)
14298	{
14299	/*
14300	* The exception was for the guest. Update DR6, DR7.GD and
14301	* IA32_DEBUGCTL.LBR before forwarding it.
14302	* See Intel spec. 27.1 "Architectural State before a VM-Exit".
14303	*/
14304	VMMRZCallRing3Disable(pVCpu);
14305	HM_DISABLE_PREEMPT(pVCpu);
14306
14307	pCtx->dr[6] &= ~X86_DR6_B_MASK;
14308	pCtx->dr[6] \|= uDR6;
14309	if (CPUMIsGuestDebugStateActive(pVCpu))
14310	ASMSetDR6(pCtx->dr[6]);
14311
14312	HM_RESTORE_PREEMPT();
14313	VMMRZCallRing3Enable(pVCpu);
14314
14315	rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_DR7);
14316	AssertRCReturn(rc, rc);
14317
14318	/* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */
14319	pCtx->dr[7] &= ~(uint64_t)X86_DR7_GD;
14320
14321	/* Paranoia. */
14322	pCtx->dr[7] &= ~(uint64_t)X86_DR7_RAZ_MASK;
14323	pCtx->dr[7] \|= X86_DR7_RA1_MASK;
14324
14325	rc = VMXWriteVmcsNw(VMX_VMCS_GUEST_DR7, pCtx->dr[7]);
14326	AssertRC(rc);
14327
14328	/*
14329	* Raise #DB in the guest.
14330	*
14331	* It is important to reflect exactly what the VM-exit gave us (preserving the
14332	* interruption-type) rather than use hmR0VmxSetPendingXcptDB() as the #DB could've
14333	* been raised while executing ICEBP (INT1) and not the regular #DB. Thus it may
14334	* trigger different handling in the CPU (like skipping DPL checks), see @bugref{6398}.
14335	*
14336	* Intel re-documented ICEBP/INT1 on May 2018 previously documented as part of
14337	* Intel 386, see Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
14338	*/
14339	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14340	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14341	return VINF_SUCCESS;
14342	}
14343
14344	/*
14345	* Not a guest trap, must be a hypervisor related debug event then.
14346	* Update DR6 in case someone is interested in it.
14347	*/
14348	AssertMsg(rc == VINF_EM_DBG_STEPPED \|\| rc == VINF_EM_DBG_BREAKPOINT, ("%Rrc\n", rc));
14349	AssertReturn(pVmxTransient->fWasHyperDebugStateActive, VERR_HM_IPE_5);
14350	CPUMSetHyperDR6(pVCpu, uDR6);
14351
14352	return rc;
14353	}
14354
14355
14356	/**
14357	* Hacks its way around the lovely mesa driver's backdoor accesses.
14358	*
14359	* @sa hmR0SvmHandleMesaDrvGp.
14360	*/
14361	static int hmR0VmxHandleMesaDrvGp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PCPUMCTX pCtx)
14362	{
14363	LogFunc(("cs:rip=%#04x:%#RX64 rcx=%#RX64 rbx=%#RX64\n", pCtx->cs.Sel, pCtx->rip, pCtx->rcx, pCtx->rbx));
14364	RT_NOREF(pCtx);
14365
14366	/* For now we'll just skip the instruction. */
14367	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14368	}
14369
14370
14371	/**
14372	* Checks if the \#GP'ing instruction is the mesa driver doing it's lovely
14373	* backdoor logging w/o checking what it is running inside.
14374	*
14375	* This recognizes an "IN EAX,DX" instruction executed in flat ring-3, with the
14376	* backdoor port and magic numbers loaded in registers.
14377	*
14378	* @returns true if it is, false if it isn't.
14379	* @sa hmR0SvmIsMesaDrvGp.
14380	*/
14381	DECLINLINE(bool) hmR0VmxIsMesaDrvGp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient, PCPUMCTX pCtx)
14382	{
14383	/* 0xed: IN eAX,dx */
14384	uint8_t abInstr[1];
14385	if (pVmxTransient->cbExitInstr != sizeof(abInstr))
14386	return false;
14387
14388	/* Check that it is #GP(0). */
14389	if (pVmxTransient->uExitIntErrorCode != 0)
14390	return false;
14391
14392	/* Check magic and port. */
14393	Assert(!(pCtx->fExtrn & (CPUMCTX_EXTRN_RAX \| CPUMCTX_EXTRN_RDX \| CPUMCTX_EXTRN_RCX)));
14394	/Log(("hmR0VmxIsMesaDrvGp: rax=%RX64 rdx=%RX64\n", pCtx->rax, pCtx->rdx));/
14395	if (pCtx->rax != UINT32_C(0x564d5868))
14396	return false;
14397	if (pCtx->dx != UINT32_C(0x5658))
14398	return false;
14399
14400	/* Flat ring-3 CS. */
14401	AssertCompile(HMVMX_CPUMCTX_EXTRN_ALL & CPUMCTX_EXTRN_CS);
14402	Assert(!(pCtx->fExtrn & CPUMCTX_EXTRN_CS));
14403	/Log(("hmR0VmxIsMesaDrvGp: cs.Attr.n.u2Dpl=%d base=%Rx64\n", pCtx->cs.Attr.n.u2Dpl, pCtx->cs.u64Base));/
14404	if (pCtx->cs.Attr.n.u2Dpl != 3)
14405	return false;
14406	if (pCtx->cs.u64Base != 0)
14407	return false;
14408
14409	/* Check opcode. */
14410	AssertCompile(HMVMX_CPUMCTX_EXTRN_ALL & CPUMCTX_EXTRN_RIP);
14411	Assert(!(pCtx->fExtrn & CPUMCTX_EXTRN_RIP));
14412	int rc = PGMPhysSimpleReadGCPtr(pVCpu, abInstr, pCtx->rip, sizeof(abInstr));
14413	/Log(("hmR0VmxIsMesaDrvGp: PGMPhysSimpleReadGCPtr -> %Rrc %#x\n", rc, abInstr[0]));/
14414	if (RT_FAILURE(rc))
14415	return false;
14416	if (abInstr[0] != 0xed)
14417	return false;
14418
14419	return true;
14420	}
14421
14422
14423	/**
14424	* VM-exit exception handler for \#GP (General-protection exception).
14425	*
14426	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14427	*/
14428	static VBOXSTRICTRC hmR0VmxExitXcptGP(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14429	{
14430	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14431	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP);
14432
14433	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14434	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14435	PVMXVMCSINFOSHARED pVmcsInfoShared = pVmcsInfo->pShared;
14436	if (pVmcsInfoShared->RealMode.fRealOnV86Active)
14437	{ /* likely */ }
14438	else
14439	{
14440	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14441	Assert(pVCpu->hmr0.s.fUsingDebugLoop \|\| pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv \|\| pVmxTransient->fIsNestedGuest);
14442	#endif
14443	/*
14444	* If the guest is not in real-mode or we have unrestricted guest execution support, or if we are
14445	* executing a nested-guest, reflect #GP to the guest or nested-guest.
14446	*/
14447	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14448	AssertRCReturn(rc, rc);
14449	Log4Func(("Gst: cs:rip=%#04x:%#RX64 ErrorCode=%#x cr0=%#RX64 cpl=%u tr=%#04x\n", pCtx->cs.Sel, pCtx->rip,
14450	pVmxTransient->uExitIntErrorCode, pCtx->cr0, CPUMGetGuestCPL(pVCpu), pCtx->tr.Sel));
14451
14452	if ( pVmxTransient->fIsNestedGuest
14453	\|\| !pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv
14454	\|\| !hmR0VmxIsMesaDrvGp(pVCpu, pVmxTransient, pCtx))
14455	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14456	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14457	else
14458	rc = hmR0VmxHandleMesaDrvGp(pVCpu, pVmxTransient, pCtx);
14459	return rc;
14460	}
14461
14462	Assert(CPUMIsGuestInRealModeEx(pCtx));
14463	Assert(!pVCpu->CTX_SUFF(pVM)->hmr0.s.vmx.fUnrestrictedGuest);
14464	Assert(!pVmxTransient->fIsNestedGuest);
14465
14466	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14467	AssertRCReturn(rc, rc);
14468
14469	VBOXSTRICTRC rcStrict = IEMExecOne(pVCpu);
14470	if (rcStrict == VINF_SUCCESS)
14471	{
14472	if (!CPUMIsGuestInRealModeEx(pCtx))
14473	{
14474	/*
14475	* The guest is no longer in real-mode, check if we can continue executing the
14476	* guest using hardware-assisted VMX. Otherwise, fall back to emulation.
14477	*/
14478	pVmcsInfoShared->RealMode.fRealOnV86Active = false;
14479	if (HMCanExecuteVmxGuest(pVCpu->pVMR0, pVCpu, pCtx))
14480	{
14481	Log4Func(("Mode changed but guest still suitable for executing using hardware-assisted VMX\n"));
14482	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14483	}
14484	else
14485	{
14486	Log4Func(("Mode changed -> VINF_EM_RESCHEDULE\n"));
14487	rcStrict = VINF_EM_RESCHEDULE;
14488	}
14489	}
14490	else
14491	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14492	}
14493	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14494	{
14495	rcStrict = VINF_SUCCESS;
14496	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14497	}
14498	return VBOXSTRICTRC_VAL(rcStrict);
14499	}
14500
14501
14502	/**
14503	* VM-exit exception handler wrapper for all other exceptions that are not handled
14504	* by a specific handler.
14505	*
14506	* This simply re-injects the exception back into the VM without any special
14507	* processing.
14508	*
14509	* @remarks Requires all fields in HMVMX_READ_XCPT_INFO to be read from the VMCS.
14510	*/
14511	static VBOXSTRICTRC hmR0VmxExitXcptOthers(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14512	{
14513	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14514
14515	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14516	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14517	AssertMsg(pVCpu->hmr0.s.fUsingDebugLoop \|\| pVmcsInfo->pShared->RealMode.fRealOnV86Active \|\| pVmxTransient->fIsNestedGuest,
14518	("uVector=%#x u32XcptBitmap=%#X32\n",
14519	VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo), pVmcsInfo->u32XcptBitmap));
14520	NOREF(pVmcsInfo);
14521	#endif
14522
14523	/*
14524	* Re-inject the exception into the guest. This cannot be a double-fault condition which
14525	* would have been handled while checking exits due to event delivery.
14526	*/
14527	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14528
14529	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
14530	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_RIP);
14531	AssertRCReturn(rc, rc);
14532	Log4Func(("Reinjecting Xcpt. uVector=%#x cs:rip=%#04x:%#RX64\n", uVector, pCtx->cs.Sel, pCtx->rip));
14533	#endif
14534
14535	#ifdef VBOX_WITH_STATISTICS
14536	switch (uVector)
14537	{
14538	case X86_XCPT_DE: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE); break;
14539	case X86_XCPT_DB: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB); break;
14540	case X86_XCPT_BP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP); break;
14541	case X86_XCPT_OF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
14542	case X86_XCPT_BR: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBR); break;
14543	case X86_XCPT_UD: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD); break;
14544	case X86_XCPT_NM: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
14545	case X86_XCPT_DF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDF); break;
14546	case X86_XCPT_TS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestTS); break;
14547	case X86_XCPT_NP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNP); break;
14548	case X86_XCPT_SS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestSS); break;
14549	case X86_XCPT_GP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP); break;
14550	case X86_XCPT_PF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF); break;
14551	case X86_XCPT_MF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF); break;
14552	case X86_XCPT_AC: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC); break;
14553	case X86_XCPT_XF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXF); break;
14554	default:
14555	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXcpUnk);
14556	break;
14557	}
14558	#endif
14559
14560	/* We should never call this function for a page-fault, we'd need to pass on the fault address below otherwise. */
14561	Assert(!VMX_EXIT_INT_INFO_IS_XCPT_PF(pVmxTransient->uExitIntInfo));
14562	NOREF(uVector);
14563
14564	/* Re-inject the original exception into the guest. */
14565	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
14566	pVmxTransient->cbExitInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
14567	return VINF_SUCCESS;
14568	}
14569
14570
14571	/**
14572	* VM-exit exception handler for all exceptions (except NMIs!).
14573	*
14574	* @remarks This may be called for both guests and nested-guests. Take care to not
14575	* make assumptions and avoid doing anything that is not relevant when
14576	* executing a nested-guest (e.g., Mesa driver hacks).
14577	*/
14578	static VBOXSTRICTRC hmR0VmxExitXcpt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14579	{
14580	HMVMX_ASSERT_READ(pVmxTransient, HMVMX_READ_XCPT_INFO);
14581
14582	/*
14583	* If this VM-exit occurred while delivering an event through the guest IDT, take
14584	* action based on the return code and additional hints (e.g. for page-faults)
14585	* that will be updated in the VMX transient structure.
14586	*/
14587	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
14588	if (rcStrict == VINF_SUCCESS)
14589	{
14590	/*
14591	* If an exception caused a VM-exit due to delivery of an event, the original
14592	* event may have to be re-injected into the guest. We shall reinject it and
14593	* continue guest execution. However, page-fault is a complicated case and
14594	* needs additional processing done in hmR0VmxExitXcptPF().
14595	*/
14596	Assert(VMX_EXIT_INT_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
14597	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14598	if ( !pVCpu->hm.s.Event.fPending
14599	\|\| uVector == X86_XCPT_PF)
14600	{
14601	switch (uVector)
14602	{
14603	case X86_XCPT_PF: return hmR0VmxExitXcptPF(pVCpu, pVmxTransient);
14604	case X86_XCPT_GP: return hmR0VmxExitXcptGP(pVCpu, pVmxTransient);
14605	case X86_XCPT_MF: return hmR0VmxExitXcptMF(pVCpu, pVmxTransient);
14606	case X86_XCPT_DB: return hmR0VmxExitXcptDB(pVCpu, pVmxTransient);
14607	case X86_XCPT_BP: return hmR0VmxExitXcptBP(pVCpu, pVmxTransient);
14608	case X86_XCPT_AC: return hmR0VmxExitXcptAC(pVCpu, pVmxTransient);
14609	default:
14610	return hmR0VmxExitXcptOthers(pVCpu, pVmxTransient);
14611	}
14612	}
14613	/* else: inject pending event before resuming guest execution. */
14614	}
14615	else if (rcStrict == VINF_HM_DOUBLE_FAULT)
14616	{
14617	Assert(pVCpu->hm.s.Event.fPending);
14618	rcStrict = VINF_SUCCESS;
14619	}
14620
14621	return rcStrict;
14622	}
14623	/** @} */
14624
14625
14626	/** @name VM-exit handlers.
14627	* @{
14628	*/
14629	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
14630	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
14631	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
14632
14633	/**
14634	* VM-exit handler for external interrupts (VMX_EXIT_EXT_INT).
14635	*/
14636	HMVMX_EXIT_DECL hmR0VmxExitExtInt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14637	{
14638	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14639	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt);
14640	/* Windows hosts (32-bit and 64-bit) have DPC latency issues. See @bugref{6853}. */
14641	if (VMMR0ThreadCtxHookIsEnabled(pVCpu))
14642	return VINF_SUCCESS;
14643	return VINF_EM_RAW_INTERRUPT;
14644	}
14645
14646
14647	/**
14648	* VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI). Conditional
14649	* VM-exit.
14650	*/
14651	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmi(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14652	{
14653	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14654	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitXcptNmi, y3);
14655
14656	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
14657
14658	uint32_t const uExitIntType = VMX_EXIT_INT_INFO_TYPE(pVmxTransient->uExitIntInfo);
14659	uint8_t const uVector = VMX_EXIT_INT_INFO_VECTOR(pVmxTransient->uExitIntInfo);
14660	Assert(VMX_EXIT_INT_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
14661
14662	PCVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14663	Assert( !(pVmcsInfo->u32ExitCtls & VMX_EXIT_CTLS_ACK_EXT_INT)
14664	&& uExitIntType != VMX_EXIT_INT_INFO_TYPE_EXT_INT);
14665	NOREF(pVmcsInfo);
14666
14667	VBOXSTRICTRC rcStrict;
14668	switch (uExitIntType)
14669	{
14670	/*
14671	* Host physical NMIs:
14672	* This cannot be a guest NMI as the only way for the guest to receive an NMI is if we
14673	* injected it ourselves and anything we inject is not going to cause a VM-exit directly
14674	* for the event being injected[1]. Go ahead and dispatch the NMI to the host[2].
14675	*
14676	* See Intel spec. 27.2.3 "Information for VM Exits During Event Delivery".
14677	* See Intel spec. 27.5.5 "Updating Non-Register State".
14678	*/
14679	case VMX_EXIT_INT_INFO_TYPE_NMI:
14680	{
14681	rcStrict = hmR0VmxExitHostNmi(pVCpu, pVmcsInfo);
14682	break;
14683	}
14684
14685	/*
14686	* Privileged software exceptions (#DB from ICEBP),
14687	* Software exceptions (#BP and #OF),
14688	* Hardware exceptions:
14689	* Process the required exceptions and resume guest execution if possible.
14690	*/
14691	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
14692	Assert(uVector == X86_XCPT_DB);
14693	RT_FALL_THRU();
14694	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
14695	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF \|\| uExitIntType == VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT);
14696	RT_FALL_THRU();
14697	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
14698	{
14699	NOREF(uVector);
14700	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
14701	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14702	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
14703	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
14704
14705	rcStrict = hmR0VmxExitXcpt(pVCpu, pVmxTransient);
14706	break;
14707	}
14708
14709	default:
14710	{
14711	pVCpu->hm.s.u32HMError = pVmxTransient->uExitIntInfo;
14712	rcStrict = VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
14713	AssertMsgFailed(("Invalid/unexpected VM-exit interruption info %#x\n", pVmxTransient->uExitIntInfo));
14714	break;
14715	}
14716	}
14717
14718	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
14719	return rcStrict;
14720	}
14721
14722
14723	/**
14724	* VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
14725	*/
14726	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindow(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14727	{
14728	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14729
14730	/* Indicate that we no longer need to VM-exit when the guest is ready to receive interrupts, it is now ready. */
14731	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14732	hmR0VmxClearIntWindowExitVmcs(pVmcsInfo);
14733
14734	/* Evaluate and deliver pending events and resume guest execution. */
14735	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow);
14736	return VINF_SUCCESS;
14737	}
14738
14739
14740	/**
14741	* VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
14742	*/
14743	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindow(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14744	{
14745	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14746
14747	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14748	if (RT_UNLIKELY(!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))) /** @todo NSTVMX: Turn this into an assertion. */
14749	{
14750	AssertMsgFailed(("Unexpected NMI-window exit.\n"));
14751	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
14752	}
14753
14754	Assert(!CPUMIsGuestNmiBlocking(pVCpu));
14755
14756	/*
14757	* If block-by-STI is set when we get this VM-exit, it means the CPU doesn't block NMIs following STI.
14758	* It is therefore safe to unblock STI and deliver the NMI ourselves. See @bugref{7445}.
14759	*/
14760	uint32_t fIntrState;
14761	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState);
14762	AssertRC(rc);
14763	Assert(!(fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS));
14764	if (fIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)
14765	{
14766	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
14767	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
14768
14769	fIntrState &= ~VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
14770	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INT_STATE, fIntrState);
14771	AssertRC(rc);
14772	}
14773
14774	/* Indicate that we no longer need to VM-exit when the guest is ready to receive NMIs, it is now ready */
14775	hmR0VmxClearNmiWindowExitVmcs(pVmcsInfo);
14776
14777	/* Evaluate and deliver pending events and resume guest execution. */
14778	return VINF_SUCCESS;
14779	}
14780
14781
14782	/**
14783	* VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
14784	*/
14785	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvd(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14786	{
14787	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14788	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14789	}
14790
14791
14792	/**
14793	* VM-exit handler for INVD (VMX_EXIT_INVD). Unconditional VM-exit.
14794	*/
14795	HMVMX_EXIT_NSRC_DECL hmR0VmxExitInvd(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14796	{
14797	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14798	return hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14799	}
14800
14801
14802	/**
14803	* VM-exit handler for CPUID (VMX_EXIT_CPUID). Unconditional VM-exit.
14804	*/
14805	HMVMX_EXIT_DECL hmR0VmxExitCpuid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14806	{
14807	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14808
14809	/*
14810	* Get the state we need and update the exit history entry.
14811	*/
14812	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14813	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14814
14815	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
14816	AssertRCReturn(rc, rc);
14817
14818	VBOXSTRICTRC rcStrict;
14819	PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
14820	EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_CPUID),
14821	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
14822	if (!pExitRec)
14823	{
14824	/*
14825	* Regular CPUID instruction execution.
14826	*/
14827	rcStrict = IEMExecDecodedCpuid(pVCpu, pVmxTransient->cbExitInstr);
14828	if (rcStrict == VINF_SUCCESS)
14829	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14830	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14831	{
14832	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14833	rcStrict = VINF_SUCCESS;
14834	}
14835	}
14836	else
14837	{
14838	/*
14839	* Frequent exit or something needing probing. Get state and call EMHistoryExec.
14840	*/
14841	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
14842	AssertRCReturn(rc2, rc2);
14843
14844	Log4(("CpuIdExit/%u: %04x:%08RX64: %#x/%#x -> EMHistoryExec\n",
14845	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.eax, pVCpu->cpum.GstCtx.ecx));
14846
14847	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
14848	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
14849
14850	Log4(("CpuIdExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
14851	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
14852	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
14853	}
14854	return rcStrict;
14855	}
14856
14857
14858	/**
14859	* VM-exit handler for GETSEC (VMX_EXIT_GETSEC). Unconditional VM-exit.
14860	*/
14861	HMVMX_EXIT_DECL hmR0VmxExitGetsec(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14862	{
14863	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14864
14865	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14866	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR4);
14867	AssertRCReturn(rc, rc);
14868
14869	if (pVCpu->cpum.GstCtx.cr4 & X86_CR4_SMXE)
14870	return VINF_EM_RAW_EMULATE_INSTR;
14871
14872	AssertMsgFailed(("hmR0VmxExitGetsec: Unexpected VM-exit when CR4.SMXE is 0.\n"));
14873	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
14874	}
14875
14876
14877	/**
14878	* VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
14879	*/
14880	HMVMX_EXIT_DECL hmR0VmxExitRdtsc(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14881	{
14882	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14883
14884	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14885	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14886	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
14887	AssertRCReturn(rc, rc);
14888
14889	VBOXSTRICTRC rcStrict = IEMExecDecodedRdtsc(pVCpu, pVmxTransient->cbExitInstr);
14890	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
14891	{
14892	/* If we get a spurious VM-exit when TSC offsetting is enabled,
14893	we must reset offsetting on VM-entry. See @bugref{6634}. */
14894	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TSC_OFFSETTING)
14895	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14896	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14897	}
14898	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14899	{
14900	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14901	rcStrict = VINF_SUCCESS;
14902	}
14903	return rcStrict;
14904	}
14905
14906
14907	/**
14908	* VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
14909	*/
14910	HMVMX_EXIT_DECL hmR0VmxExitRdtscp(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14911	{
14912	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14913
14914	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14915	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
14916	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_TSC_AUX);
14917	AssertRCReturn(rc, rc);
14918
14919	VBOXSTRICTRC rcStrict = IEMExecDecodedRdtscp(pVCpu, pVmxTransient->cbExitInstr);
14920	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
14921	{
14922	/* If we get a spurious VM-exit when TSC offsetting is enabled,
14923	we must reset offsetting on VM-reentry. See @bugref{6634}. */
14924	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TSC_OFFSETTING)
14925	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
14926	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
14927	}
14928	else if (rcStrict == VINF_IEM_RAISED_XCPT)
14929	{
14930	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
14931	rcStrict = VINF_SUCCESS;
14932	}
14933	return rcStrict;
14934	}
14935
14936
14937	/**
14938	* VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
14939	*/
14940	HMVMX_EXIT_DECL hmR0VmxExitRdpmc(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14941	{
14942	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14943
14944	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14945	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_CR4 \| CPUMCTX_EXTRN_CR0
14946	\| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_SS);
14947	AssertRCReturn(rc, rc);
14948
14949	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
14950	rc = EMInterpretRdpmc(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx));
14951	if (RT_LIKELY(rc == VINF_SUCCESS))
14952	{
14953	rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14954	Assert(pVmxTransient->cbExitInstr == 2);
14955	}
14956	else
14957	{
14958	AssertMsgFailed(("hmR0VmxExitRdpmc: EMInterpretRdpmc failed with %Rrc\n", rc));
14959	rc = VERR_EM_INTERPRETER;
14960	}
14961	return rc;
14962	}
14963
14964
14965	/**
14966	* VM-exit handler for VMCALL (VMX_EXIT_VMCALL). Unconditional VM-exit.
14967	*/
14968	HMVMX_EXIT_DECL hmR0VmxExitVmcall(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
14969	{
14970	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
14971
14972	VBOXSTRICTRC rcStrict = VERR_VMX_IPE_3;
14973	if (EMAreHypercallInstructionsEnabled(pVCpu))
14974	{
14975	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
14976	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_RIP \| CPUMCTX_EXTRN_RFLAGS \| CPUMCTX_EXTRN_CR0
14977	\| CPUMCTX_EXTRN_SS \| CPUMCTX_EXTRN_CS \| CPUMCTX_EXTRN_EFER);
14978	AssertRCReturn(rc, rc);
14979
14980	/* Perform the hypercall. */
14981	rcStrict = GIMHypercall(pVCpu, &pVCpu->cpum.GstCtx);
14982	if (rcStrict == VINF_SUCCESS)
14983	{
14984	rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
14985	AssertRCReturn(rc, rc);
14986	}
14987	else
14988	Assert( rcStrict == VINF_GIM_R3_HYPERCALL
14989	\|\| rcStrict == VINF_GIM_HYPERCALL_CONTINUING
14990	\|\| RT_FAILURE(rcStrict));
14991
14992	/* If the hypercall changes anything other than guest's general-purpose registers,
14993	we would need to reload the guest changed bits here before VM-entry. */
14994	}
14995	else
14996	Log4Func(("Hypercalls not enabled\n"));
14997
14998	/* If hypercalls are disabled or the hypercall failed for some reason, raise #UD and continue. */
14999	if (RT_FAILURE(rcStrict))
15000	{
15001	hmR0VmxSetPendingXcptUD(pVCpu);
15002	rcStrict = VINF_SUCCESS;
15003	}
15004
15005	return rcStrict;
15006	}
15007
15008
15009	/**
15010	* VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
15011	*/
15012	HMVMX_EXIT_DECL hmR0VmxExitInvlpg(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15013	{
15014	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15015	Assert(!pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging \|\| pVCpu->hmr0.s.fUsingDebugLoop);
15016
15017	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15018	hmR0VmxReadExitQualVmcs(pVmxTransient);
15019	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15020	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
15021	AssertRCReturn(rc, rc);
15022
15023	VBOXSTRICTRC rcStrict = IEMExecDecodedInvlpg(pVCpu, pVmxTransient->cbExitInstr, pVmxTransient->uExitQual);
15024
15025	if (rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_PGM_SYNC_CR3)
15026	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15027	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15028	{
15029	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15030	rcStrict = VINF_SUCCESS;
15031	}
15032	else
15033	AssertMsgFailed(("Unexpected IEMExecDecodedInvlpg(%#RX64) status: %Rrc\n", pVmxTransient->uExitQual,
15034	VBOXSTRICTRC_VAL(rcStrict)));
15035	return rcStrict;
15036	}
15037
15038
15039	/**
15040	* VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
15041	*/
15042	HMVMX_EXIT_DECL hmR0VmxExitMonitor(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15043	{
15044	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15045
15046	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15047	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15048	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK \| CPUMCTX_EXTRN_DS);
15049	AssertRCReturn(rc, rc);
15050
15051	VBOXSTRICTRC rcStrict = IEMExecDecodedMonitor(pVCpu, pVmxTransient->cbExitInstr);
15052	if (rcStrict == VINF_SUCCESS)
15053	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15054	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15055	{
15056	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15057	rcStrict = VINF_SUCCESS;
15058	}
15059
15060	return rcStrict;
15061	}
15062
15063
15064	/**
15065	* VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
15066	*/
15067	HMVMX_EXIT_DECL hmR0VmxExitMwait(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15068	{
15069	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15070
15071	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15072	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15073	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
15074	AssertRCReturn(rc, rc);
15075
15076	VBOXSTRICTRC rcStrict = IEMExecDecodedMwait(pVCpu, pVmxTransient->cbExitInstr);
15077	if (RT_SUCCESS(rcStrict))
15078	{
15079	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15080	if (EMMonitorWaitShouldContinue(pVCpu, &pVCpu->cpum.GstCtx))
15081	rcStrict = VINF_SUCCESS;
15082	}
15083
15084	return rcStrict;
15085	}
15086
15087
15088	/**
15089	* VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT). Unconditional
15090	* VM-exit.
15091	*/
15092	HMVMX_EXIT_DECL hmR0VmxExitTripleFault(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15093	{
15094	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15095	return VINF_EM_RESET;
15096	}
15097
15098
15099	/**
15100	* VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
15101	*/
15102	HMVMX_EXIT_DECL hmR0VmxExitHlt(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15103	{
15104	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15105
15106	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15107	AssertRCReturn(rc, rc);
15108
15109	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RFLAGS); /* Advancing the RIP above should've imported eflags. */
15110	if (EMShouldContinueAfterHalt(pVCpu, &pVCpu->cpum.GstCtx)) /* Requires eflags. */
15111	rc = VINF_SUCCESS;
15112	else
15113	rc = VINF_EM_HALT;
15114
15115	if (rc != VINF_SUCCESS)
15116	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHltToR3);
15117	return rc;
15118	}
15119
15120
15121	/**
15122	* VM-exit handler for instructions that result in a \#UD exception delivered to
15123	* the guest.
15124	*/
15125	HMVMX_EXIT_NSRC_DECL hmR0VmxExitSetPendingXcptUD(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15126	{
15127	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15128	hmR0VmxSetPendingXcptUD(pVCpu);
15129	return VINF_SUCCESS;
15130	}
15131
15132
15133	/**
15134	* VM-exit handler for expiry of the VMX-preemption timer.
15135	*/
15136	HMVMX_EXIT_DECL hmR0VmxExitPreemptTimer(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15137	{
15138	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15139
15140	/* If the VMX-preemption timer has expired, reinitialize the preemption timer on next VM-entry. */
15141	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
15142	Log12(("hmR0VmxExitPreemptTimer:\n"));
15143
15144	/* If there are any timer events pending, fall back to ring-3, otherwise resume guest execution. */
15145	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15146	bool fTimersPending = TMTimerPollBool(pVM, pVCpu);
15147	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitPreemptTimer);
15148	return fTimersPending ? VINF_EM_RAW_TIMER_PENDING : VINF_SUCCESS;
15149	}
15150
15151
15152	/**
15153	* VM-exit handler for XSETBV (VMX_EXIT_XSETBV). Unconditional VM-exit.
15154	*/
15155	HMVMX_EXIT_DECL hmR0VmxExitXsetbv(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15156	{
15157	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15158
15159	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15160	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15161	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_CR4);
15162	AssertRCReturn(rc, rc);
15163
15164	VBOXSTRICTRC rcStrict = IEMExecDecodedXsetbv(pVCpu, pVmxTransient->cbExitInstr);
15165	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, rcStrict != VINF_IEM_RAISED_XCPT ? HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS
15166	: HM_CHANGED_RAISED_XCPT_MASK);
15167
15168	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15169	bool const fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
15170	if (fLoadSaveGuestXcr0 != pVCpu->hmr0.s.fLoadSaveGuestXcr0)
15171	{
15172	pVCpu->hmr0.s.fLoadSaveGuestXcr0 = fLoadSaveGuestXcr0;
15173	hmR0VmxUpdateStartVmFunction(pVCpu);
15174	}
15175
15176	return rcStrict;
15177	}
15178
15179
15180	/**
15181	* VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
15182	*/
15183	HMVMX_EXIT_DECL hmR0VmxExitInvpcid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15184	{
15185	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15186
15187	/** @todo Enable the new code after finding a reliably guest test-case. */
15188	#if 1
15189	return VERR_EM_INTERPRETER;
15190	#else
15191	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15192	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15193	hmR0VmxReadExitQualVmcs(pVmxTransient);
15194	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
15195	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
15196	AssertRCReturn(rc, rc);
15197
15198	/* Paranoia. Ensure this has a memory operand. */
15199	Assert(!pVmxTransient->ExitInstrInfo.Inv.u1Cleared0);
15200
15201	uint8_t const iGReg = pVmxTransient->ExitInstrInfo.VmreadVmwrite.iReg2;
15202	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
15203	uint64_t const uType = CPUMIsGuestIn64BitCode(pVCpu) ? pVCpu->cpum.GstCtx.aGRegs[iGReg].u64
15204	: pVCpu->cpum.GstCtx.aGRegs[iGReg].u32;
15205
15206	RTGCPTR GCPtrDesc;
15207	HMVMX_DECODE_MEM_OPERAND(pVCpu, pVmxTransient->ExitInstrInfo.u, pVmxTransient->uExitQual, VMXMEMACCESS_READ, &GCPtrDesc);
15208
15209	VBOXSTRICTRC rcStrict = IEMExecDecodedInvpcid(pVCpu, pVmxTransient->cbExitInstr, pVmxTransient->ExitInstrInfo.Inv.iSegReg,
15210	GCPtrDesc, uType);
15211	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
15212	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15213	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15214	{
15215	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15216	rcStrict = VINF_SUCCESS;
15217	}
15218	return rcStrict;
15219	#endif
15220	}
15221
15222
15223	/**
15224	* VM-exit handler for invalid-guest-state (VMX_EXIT_ERR_INVALID_GUEST_STATE). Error
15225	* VM-exit.
15226	*/
15227	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrInvalidGuestState(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15228	{
15229	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15230	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
15231	AssertRCReturn(rc, rc);
15232
15233	rc = hmR0VmxCheckCachedVmcsCtls(pVCpu, pVmcsInfo, pVmxTransient->fIsNestedGuest);
15234	if (RT_FAILURE(rc))
15235	return rc;
15236
15237	uint32_t const uInvalidReason = hmR0VmxCheckGuestState(pVCpu, pVmcsInfo);
15238	NOREF(uInvalidReason);
15239
15240	#ifdef VBOX_STRICT
15241	uint32_t fIntrState;
15242	uint64_t u64Val;
15243	hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
15244	hmR0VmxReadEntryXcptErrorCodeVmcs(pVmxTransient);
15245	hmR0VmxReadEntryInstrLenVmcs(pVmxTransient);
15246
15247	Log4(("uInvalidReason %u\n", uInvalidReason));
15248	Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", pVmxTransient->uEntryIntInfo));
15249	Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", pVmxTransient->uEntryXcptErrorCode));
15250	Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %#RX32\n", pVmxTransient->cbEntryInstr));
15251
15252	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INT_STATE, &fIntrState); AssertRC(rc);
15253	Log4(("VMX_VMCS32_GUEST_INT_STATE %#RX32\n", fIntrState));
15254	rc = VMXReadVmcsNw(VMX_VMCS_GUEST_CR0, &u64Val); AssertRC(rc);
15255	Log4(("VMX_VMCS_GUEST_CR0 %#RX64\n", u64Val));
15256	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_MASK, &u64Val); AssertRC(rc);
15257	Log4(("VMX_VMCS_CTRL_CR0_MASK %#RX64\n", u64Val));
15258	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR0_READ_SHADOW, &u64Val); AssertRC(rc);
15259	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RX64\n", u64Val));
15260	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_MASK, &u64Val); AssertRC(rc);
15261	Log4(("VMX_VMCS_CTRL_CR4_MASK %#RX64\n", u64Val));
15262	rc = VMXReadVmcsNw(VMX_VMCS_CTRL_CR4_READ_SHADOW, &u64Val); AssertRC(rc);
15263	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RX64\n", u64Val));
15264	if (pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging)
15265	{
15266	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
15267	Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
15268	}
15269	hmR0DumpRegs(pVCpu, HM_DUMP_REG_FLAGS_ALL);
15270	#endif
15271
15272	return VERR_VMX_INVALID_GUEST_STATE;
15273	}
15274
15275	/**
15276	* VM-exit handler for all undefined/unexpected reasons. Should never happen.
15277	*/
15278	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrUnexpected(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15279	{
15280	/*
15281	* Cumulative notes of all recognized but unexpected VM-exits.
15282	*
15283	* 1. This does -not- cover scenarios like a page-fault VM-exit occurring when
15284	* nested-paging is used.
15285	*
15286	* 2. Any instruction that causes a VM-exit unconditionally (for e.g. VMXON) must be
15287	* emulated or a #UD must be raised in the guest. Therefore, we should -not- be using
15288	* this function (and thereby stop VM execution) for handling such instructions.
15289	*
15290	*
15291	* VMX_EXIT_INIT_SIGNAL:
15292	* INIT signals are blocked in VMX root operation by VMXON and by SMI in SMM.
15293	* It is -NOT- blocked in VMX non-root operation so we can, in theory, still get these
15294	* VM-exits. However, we should not receive INIT signals VM-exit while executing a VM.
15295	*
15296	* See Intel spec. 33.14.1 Default Treatment of SMI Delivery"
15297	* See Intel spec. 29.3 "VMX Instructions" for "VMXON".
15298	* See Intel spec. "23.8 Restrictions on VMX operation".
15299	*
15300	* VMX_EXIT_SIPI:
15301	* SIPI exits can only occur in VMX non-root operation when the "wait-for-SIPI" guest
15302	* activity state is used. We don't make use of it as our guests don't have direct
15303	* access to the host local APIC.
15304	*
15305	* See Intel spec. 25.3 "Other Causes of VM-exits".
15306	*
15307	* VMX_EXIT_IO_SMI:
15308	* VMX_EXIT_SMI:
15309	* This can only happen if we support dual-monitor treatment of SMI, which can be
15310	* activated by executing VMCALL in VMX root operation. Only an STM (SMM transfer
15311	* monitor) would get this VM-exit when we (the executive monitor) execute a VMCALL in
15312	* VMX root mode or receive an SMI. If we get here, something funny is going on.
15313	*
15314	* See Intel spec. 33.15.6 "Activating the Dual-Monitor Treatment"
15315	* See Intel spec. 25.3 "Other Causes of VM-Exits"
15316	*
15317	* VMX_EXIT_ERR_MSR_LOAD:
15318	* Failures while loading MSRs are part of the VM-entry MSR-load area are unexpected
15319	* and typically indicates a bug in the hypervisor code. We thus cannot not resume
15320	* execution.
15321	*
15322	* See Intel spec. 26.7 "VM-Entry Failures During Or After Loading Guest State".
15323	*
15324	* VMX_EXIT_ERR_MACHINE_CHECK:
15325	* Machine check exceptions indicates a fatal/unrecoverable hardware condition
15326	* including but not limited to system bus, ECC, parity, cache and TLB errors. A
15327	* #MC exception abort class exception is raised. We thus cannot assume a
15328	* reasonable chance of continuing any sort of execution and we bail.
15329	*
15330	* See Intel spec. 15.1 "Machine-check Architecture".
15331	* See Intel spec. 27.1 "Architectural State Before A VM Exit".
15332	*
15333	* VMX_EXIT_PML_FULL:
15334	* VMX_EXIT_VIRTUALIZED_EOI:
15335	* VMX_EXIT_APIC_WRITE:
15336	* We do not currently support any of these features and thus they are all unexpected
15337	* VM-exits.
15338	*
15339	* VMX_EXIT_GDTR_IDTR_ACCESS:
15340	* VMX_EXIT_LDTR_TR_ACCESS:
15341	* VMX_EXIT_RDRAND:
15342	* VMX_EXIT_RSM:
15343	* VMX_EXIT_VMFUNC:
15344	* VMX_EXIT_ENCLS:
15345	* VMX_EXIT_RDSEED:
15346	* VMX_EXIT_XSAVES:
15347	* VMX_EXIT_XRSTORS:
15348	* VMX_EXIT_UMWAIT:
15349	* VMX_EXIT_TPAUSE:
15350	* VMX_EXIT_LOADIWKEY:
15351	* These VM-exits are -not- caused unconditionally by execution of the corresponding
15352	* instruction. Any VM-exit for these instructions indicate a hardware problem,
15353	* unsupported CPU modes (like SMM) or potentially corrupt VMCS controls.
15354	*
15355	* See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally".
15356	*/
15357	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15358	AssertMsgFailed(("Unexpected VM-exit %u\n", pVmxTransient->uExitReason));
15359	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
15360	}
15361
15362
15363	/**
15364	* VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
15365	*/
15366	HMVMX_EXIT_DECL hmR0VmxExitRdmsr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15367	{
15368	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15369
15370	/** @todo Optimize this: We currently drag in the whole MSR state
15371	* (CPUMCTX_EXTRN_ALL_MSRS) here. We should optimize this to only get
15372	* MSRs required. That would require changes to IEM and possibly CPUM too.
15373	* (Should probably do it lazy fashion from CPUMAllMsrs.cpp). */
15374	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15375	uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
15376	uint64_t fImport = IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK \| CPUMCTX_EXTRN_ALL_MSRS;
15377	switch (idMsr)
15378	{
15379	case MSR_K8_FS_BASE: fImport \|= CPUMCTX_EXTRN_FS; break;
15380	case MSR_K8_GS_BASE: fImport \|= CPUMCTX_EXTRN_GS; break;
15381	}
15382
15383	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15384	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImport);
15385	AssertRCReturn(rc, rc);
15386
15387	Log4Func(("ecx=%#RX32\n", idMsr));
15388
15389	#ifdef VBOX_STRICT
15390	Assert(!pVmxTransient->fIsNestedGuest);
15391	if (pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
15392	{
15393	if ( hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr)
15394	&& idMsr != MSR_K6_EFER)
15395	{
15396	AssertMsgFailed(("Unexpected RDMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n", idMsr));
15397	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15398	}
15399	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15400	{
15401	Assert(pVmcsInfo->pvMsrBitmap);
15402	uint32_t fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, idMsr);
15403	if (fMsrpm & VMXMSRPM_ALLOW_RD)
15404	{
15405	AssertMsgFailed(("Unexpected RDMSR for a passthru lazy-restore MSR. ecx=%#RX32\n", idMsr));
15406	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15407	}
15408	}
15409	}
15410	#endif
15411
15412	VBOXSTRICTRC rcStrict = IEMExecDecodedRdmsr(pVCpu, pVmxTransient->cbExitInstr);
15413	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr);
15414	if (rcStrict == VINF_SUCCESS)
15415	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15416	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15417	{
15418	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15419	rcStrict = VINF_SUCCESS;
15420	}
15421	else
15422	AssertMsg(rcStrict == VINF_CPUM_R3_MSR_READ \|\| rcStrict == VINF_EM_TRIPLE_FAULT,
15423	("Unexpected IEMExecDecodedRdmsr rc (%Rrc)\n", VBOXSTRICTRC_VAL(rcStrict)));
15424
15425	return rcStrict;
15426	}
15427
15428
15429	/**
15430	* VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
15431	*/
15432	HMVMX_EXIT_DECL hmR0VmxExitWrmsr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15433	{
15434	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15435
15436	/** @todo Optimize this: We currently drag in the whole MSR state
15437	* (CPUMCTX_EXTRN_ALL_MSRS) here. We should optimize this to only get
15438	* MSRs required. That would require changes to IEM and possibly CPUM too.
15439	* (Should probably do it lazy fashion from CPUMAllMsrs.cpp). */
15440	uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
15441	uint64_t fImport = IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK \| CPUMCTX_EXTRN_ALL_MSRS;
15442
15443	/*
15444	* The FS and GS base MSRs are not part of the above all-MSRs mask.
15445	* Although we don't need to fetch the base as it will be overwritten shortly, while
15446	* loading guest-state we would also load the entire segment register including limit
15447	* and attributes and thus we need to load them here.
15448	*/
15449	switch (idMsr)
15450	{
15451	case MSR_K8_FS_BASE: fImport \|= CPUMCTX_EXTRN_FS; break;
15452	case MSR_K8_GS_BASE: fImport \|= CPUMCTX_EXTRN_GS; break;
15453	}
15454
15455	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15456	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15457	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, fImport);
15458	AssertRCReturn(rc, rc);
15459
15460	Log4Func(("ecx=%#RX32 edx:eax=%#RX32:%#RX32\n", idMsr, pVCpu->cpum.GstCtx.edx, pVCpu->cpum.GstCtx.eax));
15461
15462	VBOXSTRICTRC rcStrict = IEMExecDecodedWrmsr(pVCpu, pVmxTransient->cbExitInstr);
15463	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr);
15464
15465	if (rcStrict == VINF_SUCCESS)
15466	{
15467	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
15468
15469	/* If this is an X2APIC WRMSR access, update the APIC state as well. */
15470	if ( idMsr == MSR_IA32_APICBASE
15471	\|\| ( idMsr >= MSR_IA32_X2APIC_START
15472	&& idMsr <= MSR_IA32_X2APIC_END))
15473	{
15474	/*
15475	* We've already saved the APIC related guest-state (TPR) in post-run phase.
15476	* When full APIC register virtualization is implemented we'll have to make
15477	* sure APIC state is saved from the VMCS before IEM changes it.
15478	*/
15479	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
15480	}
15481	else if (idMsr == MSR_IA32_TSC) /* Windows 7 does this during bootup. See @bugref{6398}. */
15482	pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
15483	else if (idMsr == MSR_K6_EFER)
15484	{
15485	/*
15486	* If the guest touches the EFER MSR we need to update the VM-Entry and VM-Exit controls
15487	* as well, even if it is -not- touching bits that cause paging mode changes (LMA/LME).
15488	* We care about the other bits as well, SCE and NXE. See @bugref{7368}.
15489	*/
15490	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_EFER_MSR \| HM_CHANGED_VMX_ENTRY_EXIT_CTLS);
15491	}
15492
15493	/* Update MSRs that are part of the VMCS and auto-load/store area when MSR-bitmaps are not used. */
15494	if (!(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS))
15495	{
15496	switch (idMsr)
15497	{
15498	case MSR_IA32_SYSENTER_CS: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_CS_MSR); break;
15499	case MSR_IA32_SYSENTER_EIP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); break;
15500	case MSR_IA32_SYSENTER_ESP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); break;
15501	case MSR_K8_FS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_FS); break;
15502	case MSR_K8_GS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_GS); break;
15503	case MSR_K6_EFER: /* Nothing to do, already handled above. */ break;
15504	default:
15505	{
15506	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15507	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_VMX_GUEST_LAZY_MSRS);
15508	else if (hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr))
15509	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
15510	break;
15511	}
15512	}
15513	}
15514	#ifdef VBOX_STRICT
15515	else
15516	{
15517	/* Paranoia. Validate that MSRs in the MSR-bitmaps with write-passthru are not intercepted. */
15518	switch (idMsr)
15519	{
15520	case MSR_IA32_SYSENTER_CS:
15521	case MSR_IA32_SYSENTER_EIP:
15522	case MSR_IA32_SYSENTER_ESP:
15523	case MSR_K8_FS_BASE:
15524	case MSR_K8_GS_BASE:
15525	{
15526	AssertMsgFailed(("Unexpected WRMSR for an MSR in the VMCS. ecx=%#RX32\n", idMsr));
15527	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15528	}
15529
15530	/* Writes to MSRs in auto-load/store area/swapped MSRs, shouldn't cause VM-exits with MSR-bitmaps. */
15531	default:
15532	{
15533	if (hmR0VmxIsAutoLoadGuestMsr(pVmcsInfo, idMsr))
15534	{
15535	/* EFER MSR writes are always intercepted. */
15536	if (idMsr != MSR_K6_EFER)
15537	{
15538	AssertMsgFailed(("Unexpected WRMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n",
15539	idMsr));
15540	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15541	}
15542	}
15543
15544	if (hmR0VmxIsLazyGuestMsr(pVCpu, idMsr))
15545	{
15546	Assert(pVmcsInfo->pvMsrBitmap);
15547	uint32_t fMsrpm = CPUMGetVmxMsrPermission(pVmcsInfo->pvMsrBitmap, idMsr);
15548	if (fMsrpm & VMXMSRPM_ALLOW_WR)
15549	{
15550	AssertMsgFailed(("Unexpected WRMSR for passthru, lazy-restore MSR. ecx=%#RX32\n", idMsr));
15551	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, idMsr);
15552	}
15553	}
15554	break;
15555	}
15556	}
15557	}
15558	#endif /* VBOX_STRICT */
15559	}
15560	else if (rcStrict == VINF_IEM_RAISED_XCPT)
15561	{
15562	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
15563	rcStrict = VINF_SUCCESS;
15564	}
15565	else
15566	AssertMsg(rcStrict == VINF_CPUM_R3_MSR_WRITE \|\| rcStrict == VINF_EM_TRIPLE_FAULT,
15567	("Unexpected IEMExecDecodedWrmsr rc (%Rrc)\n", VBOXSTRICTRC_VAL(rcStrict)));
15568
15569	return rcStrict;
15570	}
15571
15572
15573	/**
15574	* VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
15575	*/
15576	HMVMX_EXIT_DECL hmR0VmxExitPause(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15577	{
15578	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15579
15580	/** @todo The guest has likely hit a contended spinlock. We might want to
15581	* poke a schedule different guest VCPU. */
15582	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
15583	if (RT_SUCCESS(rc))
15584	return VINF_EM_RAW_INTERRUPT;
15585
15586	AssertMsgFailed(("hmR0VmxExitPause: Failed to increment RIP. rc=%Rrc\n", rc));
15587	return rc;
15588	}
15589
15590
15591	/**
15592	* VM-exit handler for when the TPR value is lowered below the specified
15593	* threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
15594	*/
15595	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThreshold(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15596	{
15597	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15598	Assert(pVmxTransient->pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
15599
15600	/*
15601	* The TPR shadow would've been synced with the APIC TPR in the post-run phase.
15602	* We'll re-evaluate pending interrupts and inject them before the next VM
15603	* entry so we can just continue execution here.
15604	*/
15605	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTprBelowThreshold);
15606	return VINF_SUCCESS;
15607	}
15608
15609
15610	/**
15611	* VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX). Conditional
15612	* VM-exit.
15613	*
15614	* @retval VINF_SUCCESS when guest execution can continue.
15615	* @retval VINF_PGM_SYNC_CR3 CR3 sync is required, back to ring-3.
15616	* @retval VERR_EM_RESCHEDULE_REM when we need to return to ring-3 due to
15617	* incompatible guest state for VMX execution (real-on-v86 case).
15618	*/
15619	HMVMX_EXIT_DECL hmR0VmxExitMovCRx(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15620	{
15621	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15622	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitMovCRx, y2);
15623
15624	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15625	hmR0VmxReadExitQualVmcs(pVmxTransient);
15626	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15627
15628	VBOXSTRICTRC rcStrict;
15629	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15630	uint64_t const uExitQual = pVmxTransient->uExitQual;
15631	uint32_t const uAccessType = VMX_EXIT_QUAL_CRX_ACCESS(uExitQual);
15632	switch (uAccessType)
15633	{
15634	/*
15635	* MOV to CRx.
15636	*/
15637	case VMX_EXIT_QUAL_CRX_ACCESS_WRITE:
15638	{
15639	/*
15640	* When PAE paging is used, the CPU will reload PAE PDPTEs from CR3 when the guest
15641	* changes certain bits even in CR0, CR4 (and not just CR3). We are currently fine
15642	* since IEM_CPUMCTX_EXTRN_MUST_MASK (used below) includes CR3 which will import
15643	* PAE PDPTEs as well.
15644	*/
15645	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
15646	AssertRCReturn(rc, rc);
15647
15648	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
15649	uint32_t const uOldCr0 = pVCpu->cpum.GstCtx.cr0;
15650	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(uExitQual);
15651	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(uExitQual);
15652
15653	/*
15654	* MOV to CR3 only cause a VM-exit when one or more of the following are true:
15655	* - When nested paging isn't used.
15656	* - If the guest doesn't have paging enabled (intercept CR3 to update shadow page tables).
15657	* - We are executing in the VM debug loop.
15658	*/
15659	Assert( iCrReg != 3
15660	\|\| !pVM->hmr0.s.fNestedPaging
15661	\|\| !CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx)
15662	\|\| pVCpu->hmr0.s.fUsingDebugLoop);
15663
15664	/* MOV to CR8 writes only cause VM-exits when TPR shadow is not used. */
15665	Assert( iCrReg != 8
15666	\|\| !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW));
15667
15668	rcStrict = hmR0VmxExitMovToCrX(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
15669	AssertMsg( rcStrict == VINF_SUCCESS
15670	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15671
15672	/*
15673	* This is a kludge for handling switches back to real mode when we try to use
15674	* V86 mode to run real mode code directly. Problem is that V86 mode cannot
15675	* deal with special selector values, so we have to return to ring-3 and run
15676	* there till the selector values are V86 mode compatible.
15677	*
15678	* Note! Using VINF_EM_RESCHEDULE_REM here rather than VINF_EM_RESCHEDULE since the
15679	* latter is an alias for VINF_IEM_RAISED_XCPT which is asserted at the end of
15680	* this function.
15681	*/
15682	if ( iCrReg == 0
15683	&& rcStrict == VINF_SUCCESS
15684	&& !pVM->hmr0.s.vmx.fUnrestrictedGuest
15685	&& CPUMIsGuestInRealModeEx(&pVCpu->cpum.GstCtx)
15686	&& (uOldCr0 & X86_CR0_PE)
15687	&& !(pVCpu->cpum.GstCtx.cr0 & X86_CR0_PE))
15688	{
15689	/** @todo Check selectors rather than returning all the time. */
15690	Assert(!pVmxTransient->fIsNestedGuest);
15691	Log4Func(("CR0 write, back to real mode -> VINF_EM_RESCHEDULE_REM\n"));
15692	rcStrict = VINF_EM_RESCHEDULE_REM;
15693	}
15694	break;
15695	}
15696
15697	/*
15698	* MOV from CRx.
15699	*/
15700	case VMX_EXIT_QUAL_CRX_ACCESS_READ:
15701	{
15702	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(uExitQual);
15703	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(uExitQual);
15704
15705	/*
15706	* MOV from CR3 only cause a VM-exit when one or more of the following are true:
15707	* - When nested paging isn't used.
15708	* - If the guest doesn't have paging enabled (pass guest's CR3 rather than our identity mapped CR3).
15709	* - We are executing in the VM debug loop.
15710	*/
15711	Assert( iCrReg != 3
15712	\|\| !pVM->hmr0.s.fNestedPaging
15713	\|\| !CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx)
15714	\|\| pVCpu->hmr0.s.fLeaveDone);
15715
15716	/* MOV from CR8 reads only cause a VM-exit when the TPR shadow feature isn't enabled. */
15717	Assert( iCrReg != 8
15718	\|\| !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW));
15719
15720	rcStrict = hmR0VmxExitMovFromCrX(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
15721	break;
15722	}
15723
15724	/*
15725	* CLTS (Clear Task-Switch Flag in CR0).
15726	*/
15727	case VMX_EXIT_QUAL_CRX_ACCESS_CLTS:
15728	{
15729	rcStrict = hmR0VmxExitClts(pVCpu, pVmcsInfo, pVmxTransient->cbExitInstr);
15730	break;
15731	}
15732
15733	/*
15734	* LMSW (Load Machine-Status Word into CR0).
15735	* LMSW cannot clear CR0.PE, so no fRealOnV86Active kludge needed here.
15736	*/
15737	case VMX_EXIT_QUAL_CRX_ACCESS_LMSW:
15738	{
15739	RTGCPTR GCPtrEffDst;
15740	uint8_t const cbInstr = pVmxTransient->cbExitInstr;
15741	uint16_t const uMsw = VMX_EXIT_QUAL_CRX_LMSW_DATA(uExitQual);
15742	bool const fMemOperand = VMX_EXIT_QUAL_CRX_LMSW_OP_MEM(uExitQual);
15743	if (fMemOperand)
15744	{
15745	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
15746	GCPtrEffDst = pVmxTransient->uGuestLinearAddr;
15747	}
15748	else
15749	GCPtrEffDst = NIL_RTGCPTR;
15750	rcStrict = hmR0VmxExitLmsw(pVCpu, pVmcsInfo, cbInstr, uMsw, GCPtrEffDst);
15751	break;
15752	}
15753
15754	default:
15755	{
15756	AssertMsgFailed(("Unrecognized Mov CRX access type %#x\n", uAccessType));
15757	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, uAccessType);
15758	}
15759	}
15760
15761	Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS))
15762	== (HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS));
15763	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
15764
15765	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitMovCRx, y2);
15766	NOREF(pVM);
15767	return rcStrict;
15768	}
15769
15770
15771	/**
15772	* VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR). Conditional
15773	* VM-exit.
15774	*/
15775	HMVMX_EXIT_DECL hmR0VmxExitIoInstr(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
15776	{
15777	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
15778	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitIO, y1);
15779
15780	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
15781	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
15782	hmR0VmxReadExitQualVmcs(pVmxTransient);
15783	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
15784	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK \| CPUMCTX_EXTRN_SREG_MASK
15785	\| CPUMCTX_EXTRN_EFER);
15786	/* EFER MSR also required for longmode checks in EMInterpretDisasCurrent(), but it's always up-to-date. */
15787	AssertRCReturn(rc, rc);
15788
15789	/* Refer Intel spec. 27-5. "Exit Qualifications for I/O Instructions" for the format. */
15790	uint32_t const uIOPort = VMX_EXIT_QUAL_IO_PORT(pVmxTransient->uExitQual);
15791	uint8_t const uIOSize = VMX_EXIT_QUAL_IO_SIZE(pVmxTransient->uExitQual);
15792	bool const fIOWrite = (VMX_EXIT_QUAL_IO_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_IO_DIRECTION_OUT);
15793	bool const fIOString = VMX_EXIT_QUAL_IO_IS_STRING(pVmxTransient->uExitQual);
15794	bool const fGstStepping = RT_BOOL(pCtx->eflags.Bits.u1TF);
15795	bool const fDbgStepping = pVCpu->hm.s.fSingleInstruction;
15796	AssertReturn(uIOSize <= 3 && uIOSize != 2, VERR_VMX_IPE_1);
15797
15798	/*
15799	* Update exit history to see if this exit can be optimized.
15800	*/
15801	VBOXSTRICTRC rcStrict;
15802	PCEMEXITREC pExitRec = NULL;
15803	if ( !fGstStepping
15804	&& !fDbgStepping)
15805	pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
15806	!fIOString
15807	? !fIOWrite
15808	? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_READ)
15809	: EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_WRITE)
15810	: !fIOWrite
15811	? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_READ)
15812	: EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_WRITE),
15813	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
15814	if (!pExitRec)
15815	{
15816	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses in bytes. */
15817	static uint32_t const s_aIOOpAnd[4] = { 0xff, 0xffff, 0, 0xffffffff }; /* AND masks for saving result in AL/AX/EAX. */
15818
15819	uint32_t const cbValue = s_aIOSizes[uIOSize];
15820	uint32_t const cbInstr = pVmxTransient->cbExitInstr;
15821	bool fUpdateRipAlready = false; /* ugly hack, should be temporary. */
15822	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
15823	if (fIOString)
15824	{
15825	/*
15826	* INS/OUTS - I/O String instruction.
15827	*
15828	* Use instruction-information if available, otherwise fall back on
15829	* interpreting the instruction.
15830	*/
15831	Log4Func(("cs:rip=%#04x:%#RX64 %#06x/%u %c str\n", pCtx->cs.Sel, pCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
15832	AssertReturn(pCtx->dx == uIOPort, VERR_VMX_IPE_2);
15833	bool const fInsOutsInfo = RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_INS_OUTS);
15834	if (fInsOutsInfo)
15835	{
15836	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
15837	AssertReturn(pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize <= 2, VERR_VMX_IPE_3);
15838	AssertCompile(IEMMODE_16BIT == 0 && IEMMODE_32BIT == 1 && IEMMODE_64BIT == 2);
15839	IEMMODE const enmAddrMode = (IEMMODE)pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize;
15840	bool const fRep = VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual);
15841	if (fIOWrite)
15842	rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, fRep, cbInstr,
15843	pVmxTransient->ExitInstrInfo.StrIo.iSegReg, true /fIoChecked/);
15844	else
15845	{
15846	/*
15847	* The segment prefix for INS cannot be overridden and is always ES. We can safely assume X86_SREG_ES.
15848	* Hence "iSegReg" field is undefined in the instruction-information field in VT-x for INS.
15849	* See Intel Instruction spec. for "INS".
15850	* See Intel spec. Table 27-8 "Format of the VM-Exit Instruction-Information Field as Used for INS and OUTS".
15851	*/
15852	rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, fRep, cbInstr, true /fIoChecked/);
15853	}
15854	}
15855	else
15856	rcStrict = IEMExecOne(pVCpu);
15857
15858	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
15859	fUpdateRipAlready = true;
15860	}
15861	else
15862	{
15863	/*
15864	* IN/OUT - I/O instruction.
15865	*/
15866	Log4Func(("cs:rip=%04x:%08RX64 %#06x/%u %c\n", pCtx->cs.Sel, pCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
15867	uint32_t const uAndVal = s_aIOOpAnd[uIOSize];
15868	Assert(!VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual));
15869	if (fIOWrite)
15870	{
15871	rcStrict = IOMIOPortWrite(pVM, pVCpu, uIOPort, pCtx->eax & uAndVal, cbValue);
15872	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite);
15873	if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
15874	&& !pCtx->eflags.Bits.u1TF)
15875	rcStrict = EMRZSetPendingIoPortWrite(pVCpu, uIOPort, cbInstr, cbValue, pCtx->eax & uAndVal);
15876	}
15877	else
15878	{
15879	uint32_t u32Result = 0;
15880	rcStrict = IOMIOPortRead(pVM, pVCpu, uIOPort, &u32Result, cbValue);
15881	if (IOM_SUCCESS(rcStrict))
15882	{
15883	/* Save result of I/O IN instr. in AL/AX/EAX. */
15884	pCtx->eax = (pCtx->eax & ~uAndVal) \| (u32Result & uAndVal);
15885	}
15886	if ( rcStrict == VINF_IOM_R3_IOPORT_READ
15887	&& !pCtx->eflags.Bits.u1TF)
15888	rcStrict = EMRZSetPendingIoPortRead(pVCpu, uIOPort, cbInstr, cbValue);
15889	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead);
15890	}
15891	}
15892
15893	if (IOM_SUCCESS(rcStrict))
15894	{
15895	if (!fUpdateRipAlready)
15896	{
15897	hmR0VmxAdvanceGuestRipBy(pVCpu, cbInstr);
15898	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP);
15899	}
15900
15901	/*
15902	* INS/OUTS with REP prefix updates RFLAGS, can be observed with triple-fault guru
15903	* while booting Fedora 17 64-bit guest.
15904	*
15905	* See Intel Instruction reference for REP/REPE/REPZ/REPNE/REPNZ.
15906	*/
15907	if (fIOString)
15908	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RFLAGS);
15909
15910	/*
15911	* If any I/O breakpoints are armed, we need to check if one triggered
15912	* and take appropriate action.
15913	* Note that the I/O breakpoint type is undefined if CR4.DE is 0.
15914	*/
15915	rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_DR7);
15916	AssertRCReturn(rc, rc);
15917
15918	/** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the
15919	* execution engines about whether hyper BPs and such are pending. */
15920	uint32_t const uDr7 = pCtx->dr[7];
15921	if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK)
15922	&& X86_DR7_ANY_RW_IO(uDr7)
15923	&& (pCtx->cr4 & X86_CR4_DE))
15924	\|\| DBGFBpIsHwIoArmed(pVM)))
15925	{
15926	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck);
15927
15928	/* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
15929	VMMRZCallRing3Disable(pVCpu);
15930	HM_DISABLE_PREEMPT(pVCpu);
15931
15932	bool fIsGuestDbgActive = CPUMR0DebugStateMaybeSaveGuest(pVCpu, true /* fDr6 */);
15933
15934	VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pCtx, uIOPort, cbValue);
15935	if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
15936	{
15937	/* Raise #DB. */
15938	if (fIsGuestDbgActive)
15939	ASMSetDR6(pCtx->dr[6]);
15940	if (pCtx->dr[7] != uDr7)
15941	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_GUEST_DR7;
15942
15943	hmR0VmxSetPendingXcptDB(pVCpu);
15944	}
15945	/* rcStrict is VINF_SUCCESS, VINF_IOM_R3_IOPORT_COMMIT_WRITE, or in [VINF_EM_FIRST..VINF_EM_LAST],
15946	however we can ditch VINF_IOM_R3_IOPORT_COMMIT_WRITE as it has VMCPU_FF_IOM as backup. */
15947	else if ( rcStrict2 != VINF_SUCCESS
15948	&& (rcStrict == VINF_SUCCESS \|\| rcStrict2 < rcStrict))
15949	rcStrict = rcStrict2;
15950	AssertCompile(VINF_EM_LAST < VINF_IOM_R3_IOPORT_COMMIT_WRITE);
15951
15952	HM_RESTORE_PREEMPT();
15953	VMMRZCallRing3Enable(pVCpu);
15954	}
15955	}
15956
15957	#ifdef VBOX_STRICT
15958	if ( rcStrict == VINF_IOM_R3_IOPORT_READ
15959	\|\| rcStrict == VINF_EM_PENDING_R3_IOPORT_READ)
15960	Assert(!fIOWrite);
15961	else if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
15962	\|\| rcStrict == VINF_IOM_R3_IOPORT_COMMIT_WRITE
15963	\|\| rcStrict == VINF_EM_PENDING_R3_IOPORT_WRITE)
15964	Assert(fIOWrite);
15965	else
15966	{
15967	# if 0 /** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST
15968	* statuses, that the VMM device and some others may return. See
15969	* IOM_SUCCESS() for guidance. */
15970	AssertMsg( RT_FAILURE(rcStrict)
15971	\|\| rcStrict == VINF_SUCCESS
15972	\|\| rcStrict == VINF_EM_RAW_EMULATE_INSTR
15973	\|\| rcStrict == VINF_EM_DBG_BREAKPOINT
15974	\|\| rcStrict == VINF_EM_RAW_GUEST_TRAP
15975	\|\| rcStrict == VINF_EM_RAW_TO_R3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
15976	# endif
15977	}
15978	#endif
15979	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitIO, y1);
15980	}
15981	else
15982	{
15983	/*
15984	* Frequent exit or something needing probing. Get state and call EMHistoryExec.
15985	*/
15986	int rc2 = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
15987	AssertRCReturn(rc2, rc2);
15988	STAM_COUNTER_INC(!fIOString ? fIOWrite ? &pVCpu->hm.s.StatExitIOWrite : &pVCpu->hm.s.StatExitIORead
15989	: fIOWrite ? &pVCpu->hm.s.StatExitIOStringWrite : &pVCpu->hm.s.StatExitIOStringRead);
15990	Log4(("IOExit/%u: %04x:%08RX64: %s%s%s %#x LB %u -> EMHistoryExec\n",
15991	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
15992	VMX_EXIT_QUAL_IO_IS_REP(pVmxTransient->uExitQual) ? "REP " : "",
15993	fIOWrite ? "OUT" : "IN", fIOString ? "S" : "", uIOPort, uIOSize));
15994
15995	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
15996	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
15997
15998	Log4(("IOExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
15999	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
16000	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
16001	}
16002	return rcStrict;
16003	}
16004
16005
16006	/**
16007	* VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH). Unconditional
16008	* VM-exit.
16009	*/
16010	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitch(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16011	{
16012	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16013
16014	/* Check if this task-switch occurred while delivery an event through the guest IDT. */
16015	hmR0VmxReadExitQualVmcs(pVmxTransient);
16016	if (VMX_EXIT_QUAL_TASK_SWITCH_TYPE(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_TASK_SWITCH_TYPE_IDT)
16017	{
16018	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16019	if (VMX_IDT_VECTORING_INFO_IS_VALID(pVmxTransient->uIdtVectoringInfo))
16020	{
16021	uint32_t uErrCode;
16022	if (VMX_IDT_VECTORING_INFO_IS_ERROR_CODE_VALID(pVmxTransient->uIdtVectoringInfo))
16023	{
16024	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16025	uErrCode = pVmxTransient->uIdtVectoringErrorCode;
16026	}
16027	else
16028	uErrCode = 0;
16029
16030	RTGCUINTPTR GCPtrFaultAddress;
16031	if (VMX_IDT_VECTORING_INFO_IS_XCPT_PF(pVmxTransient->uIdtVectoringInfo))
16032	GCPtrFaultAddress = pVCpu->cpum.GstCtx.cr2;
16033	else
16034	GCPtrFaultAddress = 0;
16035
16036	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16037
16038	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(pVmxTransient->uIdtVectoringInfo),
16039	pVmxTransient->cbExitInstr, uErrCode, GCPtrFaultAddress);
16040
16041	Log4Func(("Pending event. uIntType=%#x uVector=%#x\n", VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo),
16042	VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo)));
16043	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
16044	return VINF_EM_RAW_INJECT_TRPM_EVENT;
16045	}
16046	}
16047
16048	/* Fall back to the interpreter to emulate the task-switch. */
16049	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
16050	return VERR_EM_INTERPRETER;
16051	}
16052
16053
16054	/**
16055	* VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional VM-exit.
16056	*/
16057	HMVMX_EXIT_DECL hmR0VmxExitMtf(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16058	{
16059	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16060
16061	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16062	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_MONITOR_TRAP_FLAG;
16063	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
16064	AssertRC(rc);
16065	return VINF_EM_DBG_STEPPED;
16066	}
16067
16068
16069	/**
16070	* VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional VM-exit.
16071	*/
16072	HMVMX_EXIT_DECL hmR0VmxExitApicAccess(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16073	{
16074	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16075	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitApicAccess);
16076
16077	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16078	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16079	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16080	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16081	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16082
16083	/*
16084	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16085	*/
16086	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16087	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16088	{
16089	/* For some crazy guest, if an event delivery causes an APIC-access VM-exit, go to instruction emulation. */
16090	if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
16091	{
16092	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
16093	return VINF_EM_RAW_INJECT_TRPM_EVENT;
16094	}
16095	}
16096	else
16097	{
16098	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16099	return rcStrict;
16100	}
16101
16102	/* IOMMIOPhysHandler() below may call into IEM, save the necessary state. */
16103	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16104	hmR0VmxReadExitQualVmcs(pVmxTransient);
16105	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16106	AssertRCReturn(rc, rc);
16107
16108	/* See Intel spec. 27-6 "Exit Qualifications for APIC-access VM-exits from Linear Accesses & Guest-Phyiscal Addresses" */
16109	uint32_t const uAccessType = VMX_EXIT_QUAL_APIC_ACCESS_TYPE(pVmxTransient->uExitQual);
16110	switch (uAccessType)
16111	{
16112	case VMX_APIC_ACCESS_TYPE_LINEAR_WRITE:
16113	case VMX_APIC_ACCESS_TYPE_LINEAR_READ:
16114	{
16115	AssertMsg( !(pVmcsInfo->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
16116	\|\| VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual) != XAPIC_OFF_TPR,
16117	("hmR0VmxExitApicAccess: can't access TPR offset while using TPR shadowing.\n"));
16118
16119	RTGCPHYS GCPhys = pVCpu->hm.s.vmx.u64GstMsrApicBase; /* Always up-to-date, as it is not part of the VMCS. */
16120	GCPhys &= PAGE_BASE_GC_MASK;
16121	GCPhys += VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual);
16122	Log4Func(("Linear access uAccessType=%#x GCPhys=%#RGp Off=%#x\n", uAccessType, GCPhys,
16123	VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual)));
16124
16125	rcStrict = IOMR0MmioPhysHandler(pVCpu->CTX_SUFF(pVM), pVCpu,
16126	uAccessType == VMX_APIC_ACCESS_TYPE_LINEAR_READ ? 0 : X86_TRAP_PF_RW, GCPhys);
16127	Log4Func(("IOMMMIOPhysHandler returned %Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
16128	if ( rcStrict == VINF_SUCCESS
16129	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16130	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16131	{
16132	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS
16133	\| HM_CHANGED_GUEST_APIC_TPR);
16134	rcStrict = VINF_SUCCESS;
16135	}
16136	break;
16137	}
16138
16139	default:
16140	{
16141	Log4Func(("uAccessType=%#x\n", uAccessType));
16142	rcStrict = VINF_EM_RAW_EMULATE_INSTR;
16143	break;
16144	}
16145	}
16146
16147	if (rcStrict != VINF_SUCCESS)
16148	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchApicAccessToR3);
16149	return rcStrict;
16150	}
16151
16152
16153	/**
16154	* VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX). Conditional
16155	* VM-exit.
16156	*/
16157	HMVMX_EXIT_DECL hmR0VmxExitMovDRx(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16158	{
16159	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16160	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16161
16162	/*
16163	* We might also get this VM-exit if the nested-guest isn't intercepting MOV DRx accesses.
16164	* In such a case, rather than disabling MOV DRx intercepts and resuming execution, we
16165	* must emulate the MOV DRx access.
16166	*/
16167	if (!pVmxTransient->fIsNestedGuest)
16168	{
16169	/* We should -not- get this VM-exit if the guest's debug registers were active. */
16170	if (pVmxTransient->fWasGuestDebugStateActive)
16171	{
16172	AssertMsgFailed(("Unexpected MOV DRx exit\n"));
16173	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, pVmxTransient->uExitReason);
16174	}
16175
16176	if ( !pVCpu->hm.s.fSingleInstruction
16177	&& !pVmxTransient->fWasHyperDebugStateActive)
16178	{
16179	Assert(!DBGFIsStepping(pVCpu));
16180	Assert(pVmcsInfo->u32XcptBitmap & RT_BIT(X86_XCPT_DB));
16181
16182	/* Don't intercept MOV DRx any more. */
16183	pVmcsInfo->u32ProcCtls &= ~VMX_PROC_CTLS_MOV_DR_EXIT;
16184	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVmcsInfo->u32ProcCtls);
16185	AssertRC(rc);
16186
16187	/* We're playing with the host CPU state here, make sure we can't preempt or longjmp. */
16188	VMMRZCallRing3Disable(pVCpu);
16189	HM_DISABLE_PREEMPT(pVCpu);
16190
16191	/* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */
16192	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
16193	Assert(CPUMIsGuestDebugStateActive(pVCpu));
16194
16195	HM_RESTORE_PREEMPT();
16196	VMMRZCallRing3Enable(pVCpu);
16197
16198	#ifdef VBOX_WITH_STATISTICS
16199	hmR0VmxReadExitQualVmcs(pVmxTransient);
16200	if (VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_DRX_DIRECTION_WRITE)
16201	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
16202	else
16203	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
16204	#endif
16205	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch);
16206	return VINF_SUCCESS;
16207	}
16208	}
16209
16210	/*
16211	* EMInterpretDRx[Write\|Read]() calls CPUMIsGuestIn64BitCode() which requires EFER MSR, CS.
16212	* The EFER MSR is always up-to-date.
16213	* Update the segment registers and DR7 from the CPU.
16214	*/
16215	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16216	hmR0VmxReadExitQualVmcs(pVmxTransient);
16217	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, CPUMCTX_EXTRN_SREG_MASK \| CPUMCTX_EXTRN_DR7);
16218	AssertRCReturn(rc, rc);
16219	Log4Func(("cs:rip=%#04x:%#RX64\n", pCtx->cs.Sel, pCtx->rip));
16220
16221	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16222	if (VMX_EXIT_QUAL_DRX_DIRECTION(pVmxTransient->uExitQual) == VMX_EXIT_QUAL_DRX_DIRECTION_WRITE)
16223	{
16224	rc = EMInterpretDRxWrite(pVM, pVCpu, CPUMCTX2CORE(pCtx),
16225	VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual),
16226	VMX_EXIT_QUAL_DRX_GENREG(pVmxTransient->uExitQual));
16227	if (RT_SUCCESS(rc))
16228	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
16229	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
16230	}
16231	else
16232	{
16233	rc = EMInterpretDRxRead(pVM, pVCpu, CPUMCTX2CORE(pCtx),
16234	VMX_EXIT_QUAL_DRX_GENREG(pVmxTransient->uExitQual),
16235	VMX_EXIT_QUAL_DRX_REGISTER(pVmxTransient->uExitQual));
16236	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
16237	}
16238
16239	Assert(rc == VINF_SUCCESS \|\| rc == VERR_EM_INTERPRETER);
16240	if (RT_SUCCESS(rc))
16241	{
16242	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pVmxTransient);
16243	AssertRCReturn(rc2, rc2);
16244	return VINF_SUCCESS;
16245	}
16246	return rc;
16247	}
16248
16249
16250	/**
16251	* VM-exit handler for EPT misconfiguration (VMX_EXIT_EPT_MISCONFIG).
16252	* Conditional VM-exit.
16253	*/
16254	HMVMX_EXIT_DECL hmR0VmxExitEptMisconfig(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16255	{
16256	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16257	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
16258
16259	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16260	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16261	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16262	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16263	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16264
16265	/*
16266	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16267	*/
16268	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16269	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16270	{
16271	/*
16272	* In the unlikely case where delivering an event causes an EPT misconfig (MMIO), go back to
16273	* instruction emulation to inject the original event. Otherwise, injecting the original event
16274	* using hardware-assisted VMX would trigger the same EPT misconfig VM-exit again.
16275	*/
16276	if (!pVCpu->hm.s.Event.fPending)
16277	{ /* likely */ }
16278	else
16279	{
16280	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
16281	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16282	/** @todo NSTVMX: Think about how this should be handled. */
16283	if (pVmxTransient->fIsNestedGuest)
16284	return VERR_VMX_IPE_3;
16285	#endif
16286	return VINF_EM_RAW_INJECT_TRPM_EVENT;
16287	}
16288	}
16289	else
16290	{
16291	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16292	return rcStrict;
16293	}
16294
16295	/*
16296	* Get sufficient state and update the exit history entry.
16297	*/
16298	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16299	hmR0VmxReadGuestPhysicalAddrVmcs(pVmxTransient);
16300	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16301	AssertRCReturn(rc, rc);
16302
16303	RTGCPHYS const GCPhys = pVmxTransient->uGuestPhysicalAddr;
16304	PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
16305	EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM \| EMEXIT_F_HM, EMEXITTYPE_MMIO),
16306	pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
16307	if (!pExitRec)
16308	{
16309	/*
16310	* If we succeed, resume guest execution.
16311	* If we fail in interpreting the instruction because we couldn't get the guest physical address
16312	* of the page containing the instruction via the guest's page tables (we would invalidate the guest page
16313	* in the host TLB), resume execution which would cause a guest page fault to let the guest handle this
16314	* weird case. See @bugref{6043}.
16315	*/
16316	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16317	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16318	/** @todo bird: We can probably just go straight to IOM here and assume that
16319	* it's MMIO, then fall back on PGM if that hunch didn't work out so
16320	* well. However, we need to address that aliasing workarounds that
16321	* PGMR0Trap0eHandlerNPMisconfig implements. So, some care is needed.
16322	*
16323	* Might also be interesting to see if we can get this done more or
16324	* less locklessly inside IOM. Need to consider the lookup table
16325	* updating and use a bit more carefully first (or do all updates via
16326	* rendezvous) */
16327	rcStrict = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, PGMMODE_EPT, CPUMCTX2CORE(pCtx), GCPhys, UINT32_MAX);
16328	Log4Func(("At %#RGp RIP=%#RX64 rc=%Rrc\n", GCPhys, pCtx->rip, VBOXSTRICTRC_VAL(rcStrict)));
16329	if ( rcStrict == VINF_SUCCESS
16330	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16331	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16332	{
16333	/* Successfully handled MMIO operation. */
16334	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS
16335	\| HM_CHANGED_GUEST_APIC_TPR);
16336	rcStrict = VINF_SUCCESS;
16337	}
16338	}
16339	else
16340	{
16341	/*
16342	* Frequent exit or something needing probing. Call EMHistoryExec.
16343	*/
16344	Log4(("EptMisscfgExit/%u: %04x:%08RX64: %RGp -> EMHistoryExec\n",
16345	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, GCPhys));
16346
16347	rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
16348	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16349
16350	Log4(("EptMisscfgExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
16351	pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
16352	VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
16353	}
16354	return rcStrict;
16355	}
16356
16357
16358	/**
16359	* VM-exit handler for EPT violation (VMX_EXIT_EPT_VIOLATION). Conditional
16360	* VM-exit.
16361	*/
16362	HMVMX_EXIT_DECL hmR0VmxExitEptViolation(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16363	{
16364	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16365	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
16366
16367	hmR0VmxReadExitQualVmcs(pVmxTransient);
16368	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16369	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16370	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16371	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16372	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16373
16374	/*
16375	* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly.
16376	*/
16377	VBOXSTRICTRC rcStrict = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pVmxTransient);
16378	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16379	{
16380	/*
16381	* If delivery of an event causes an EPT violation (true nested #PF and not MMIO),
16382	* we shall resolve the nested #PF and re-inject the original event.
16383	*/
16384	if (pVCpu->hm.s.Event.fPending)
16385	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflectNPF);
16386	}
16387	else
16388	{
16389	Assert(rcStrict != VINF_HM_DOUBLE_FAULT);
16390	return rcStrict;
16391	}
16392
16393	PVMXVMCSINFO pVmcsInfo = pVmxTransient->pVmcsInfo;
16394	hmR0VmxReadGuestPhysicalAddrVmcs(pVmxTransient);
16395	int rc = hmR0VmxImportGuestState(pVCpu, pVmcsInfo, IEM_CPUMCTX_EXTRN_MUST_MASK);
16396	AssertRCReturn(rc, rc);
16397
16398	RTGCPHYS const GCPhys = pVmxTransient->uGuestPhysicalAddr;
16399	uint64_t const uExitQual = pVmxTransient->uExitQual;
16400	AssertMsg(((pVmxTransient->uExitQual >> 7) & 3) != 2, ("%#RX64", uExitQual));
16401
16402	RTGCUINT uErrorCode = 0;
16403	if (uExitQual & VMX_EXIT_QUAL_EPT_ACCESS_INSTR_FETCH)
16404	uErrorCode \|= X86_TRAP_PF_ID;
16405	if (uExitQual & VMX_EXIT_QUAL_EPT_ACCESS_WRITE)
16406	uErrorCode \|= X86_TRAP_PF_RW;
16407	if (uExitQual & (VMX_EXIT_QUAL_EPT_ENTRY_READ \| VMX_EXIT_QUAL_EPT_ENTRY_WRITE \| VMX_EXIT_QUAL_EPT_ENTRY_EXECUTE))
16408	uErrorCode \|= X86_TRAP_PF_P;
16409
16410	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
16411	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16412	Log4Func(("at %#RX64 (%#RX64 errcode=%#x) cs:rip=%#04x:%#RX64\n", GCPhys, uExitQual, uErrorCode, pCtx->cs.Sel, pCtx->rip));
16413
16414	/*
16415	* Handle the pagefault trap for the nested shadow table.
16416	*/
16417	TRPMAssertXcptPF(pVCpu, GCPhys, uErrorCode);
16418	rcStrict = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, PGMMODE_EPT, uErrorCode, CPUMCTX2CORE(pCtx), GCPhys);
16419	TRPMResetTrap(pVCpu);
16420
16421	/* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}. */
16422	if ( rcStrict == VINF_SUCCESS
16423	\|\| rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
16424	\|\| rcStrict == VERR_PAGE_NOT_PRESENT)
16425	{
16426	/* Successfully synced our nested page tables. */
16427	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf);
16428	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RSP \| HM_CHANGED_GUEST_RFLAGS);
16429	return VINF_SUCCESS;
16430	}
16431
16432	Log4Func(("EPT return to ring-3 rcStrict2=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
16433	return rcStrict;
16434	}
16435
16436
16437	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16438	/**
16439	* VM-exit handler for VMCLEAR (VMX_EXIT_VMCLEAR). Unconditional VM-exit.
16440	*/
16441	HMVMX_EXIT_DECL hmR0VmxExitVmclear(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16442	{
16443	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16444
16445	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16446	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16447	hmR0VmxReadExitQualVmcs(pVmxTransient);
16448	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16449	\| CPUMCTX_EXTRN_HWVIRT
16450	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16451	AssertRCReturn(rc, rc);
16452
16453	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16454
16455	VMXVEXITINFO ExitInfo;
16456	RT_ZERO(ExitInfo);
16457	ExitInfo.uReason = pVmxTransient->uExitReason;
16458	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16459	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16460	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16461	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16462
16463	VBOXSTRICTRC rcStrict = IEMExecDecodedVmclear(pVCpu, &ExitInfo);
16464	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16465	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16466	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16467	{
16468	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16469	rcStrict = VINF_SUCCESS;
16470	}
16471	return rcStrict;
16472	}
16473
16474
16475	/**
16476	* VM-exit handler for VMLAUNCH (VMX_EXIT_VMLAUNCH). Unconditional VM-exit.
16477	*/
16478	HMVMX_EXIT_DECL hmR0VmxExitVmlaunch(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16479	{
16480	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16481
16482	/* Import the entire VMCS state for now as we would be switching VMCS on successful VMLAUNCH,
16483	otherwise we could import just IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK. */
16484	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16485	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16486	AssertRCReturn(rc, rc);
16487
16488	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16489
16490	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
16491	VBOXSTRICTRC rcStrict = IEMExecDecodedVmlaunchVmresume(pVCpu, pVmxTransient->cbExitInstr, VMXINSTRID_VMLAUNCH);
16492	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
16493	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16494	{
16495	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16496	if (CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
16497	rcStrict = VINF_VMX_VMLAUNCH_VMRESUME;
16498	}
16499	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
16500	return rcStrict;
16501	}
16502
16503
16504	/**
16505	* VM-exit handler for VMPTRLD (VMX_EXIT_VMPTRLD). Unconditional VM-exit.
16506	*/
16507	HMVMX_EXIT_DECL hmR0VmxExitVmptrld(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16508	{
16509	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16510
16511	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16512	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16513	hmR0VmxReadExitQualVmcs(pVmxTransient);
16514	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16515	\| CPUMCTX_EXTRN_HWVIRT
16516	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16517	AssertRCReturn(rc, rc);
16518
16519	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16520
16521	VMXVEXITINFO ExitInfo;
16522	RT_ZERO(ExitInfo);
16523	ExitInfo.uReason = pVmxTransient->uExitReason;
16524	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16525	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16526	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16527	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16528
16529	VBOXSTRICTRC rcStrict = IEMExecDecodedVmptrld(pVCpu, &ExitInfo);
16530	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16531	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16532	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16533	{
16534	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16535	rcStrict = VINF_SUCCESS;
16536	}
16537	return rcStrict;
16538	}
16539
16540
16541	/**
16542	* VM-exit handler for VMPTRST (VMX_EXIT_VMPTRST). Unconditional VM-exit.
16543	*/
16544	HMVMX_EXIT_DECL hmR0VmxExitVmptrst(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16545	{
16546	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16547
16548	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16549	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16550	hmR0VmxReadExitQualVmcs(pVmxTransient);
16551	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16552	\| CPUMCTX_EXTRN_HWVIRT
16553	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16554	AssertRCReturn(rc, rc);
16555
16556	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16557
16558	VMXVEXITINFO ExitInfo;
16559	RT_ZERO(ExitInfo);
16560	ExitInfo.uReason = pVmxTransient->uExitReason;
16561	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16562	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16563	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16564	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_WRITE, &ExitInfo.GCPtrEffAddr);
16565
16566	VBOXSTRICTRC rcStrict = IEMExecDecodedVmptrst(pVCpu, &ExitInfo);
16567	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16568	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16569	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16570	{
16571	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16572	rcStrict = VINF_SUCCESS;
16573	}
16574	return rcStrict;
16575	}
16576
16577
16578	/**
16579	* VM-exit handler for VMREAD (VMX_EXIT_VMREAD). Conditional VM-exit.
16580	*/
16581	HMVMX_EXIT_DECL hmR0VmxExitVmread(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16582	{
16583	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16584
16585	/*
16586	* Strictly speaking we should not get VMREAD VM-exits for shadow VMCS fields and
16587	* thus might not need to import the shadow VMCS state, it's safer just in case
16588	* code elsewhere dares look at unsynced VMCS fields.
16589	*/
16590	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16591	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16592	hmR0VmxReadExitQualVmcs(pVmxTransient);
16593	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16594	\| CPUMCTX_EXTRN_HWVIRT
16595	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16596	AssertRCReturn(rc, rc);
16597
16598	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16599
16600	VMXVEXITINFO ExitInfo;
16601	RT_ZERO(ExitInfo);
16602	ExitInfo.uReason = pVmxTransient->uExitReason;
16603	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16604	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16605	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16606	if (!ExitInfo.InstrInfo.VmreadVmwrite.fIsRegOperand)
16607	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_WRITE, &ExitInfo.GCPtrEffAddr);
16608
16609	VBOXSTRICTRC rcStrict = IEMExecDecodedVmread(pVCpu, &ExitInfo);
16610	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16611	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16612	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16613	{
16614	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16615	rcStrict = VINF_SUCCESS;
16616	}
16617	return rcStrict;
16618	}
16619
16620
16621	/**
16622	* VM-exit handler for VMRESUME (VMX_EXIT_VMRESUME). Unconditional VM-exit.
16623	*/
16624	HMVMX_EXIT_DECL hmR0VmxExitVmresume(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16625	{
16626	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16627
16628	/* Import the entire VMCS state for now as we would be switching VMCS on successful VMRESUME,
16629	otherwise we could import just IEM_CPUMCTX_EXTRN_VMX_VMENTRY_MASK. */
16630	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16631	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16632	AssertRCReturn(rc, rc);
16633
16634	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16635
16636	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
16637	VBOXSTRICTRC rcStrict = IEMExecDecodedVmlaunchVmresume(pVCpu, pVmxTransient->cbExitInstr, VMXINSTRID_VMRESUME);
16638	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
16639	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16640	{
16641	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
16642	if (CPUMIsGuestInVmxNonRootMode(&pVCpu->cpum.GstCtx))
16643	rcStrict = VINF_VMX_VMLAUNCH_VMRESUME;
16644	}
16645	Assert(rcStrict != VINF_IEM_RAISED_XCPT);
16646	return rcStrict;
16647	}
16648
16649
16650	/**
16651	* VM-exit handler for VMWRITE (VMX_EXIT_VMWRITE). Conditional VM-exit.
16652	*/
16653	HMVMX_EXIT_DECL hmR0VmxExitVmwrite(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16654	{
16655	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16656
16657	/*
16658	* Although we should not get VMWRITE VM-exits for shadow VMCS fields, since our HM hook
16659	* gets invoked when IEM's VMWRITE instruction emulation modifies the current VMCS and it
16660	* flags re-loading the entire shadow VMCS, we should save the entire shadow VMCS here.
16661	*/
16662	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16663	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16664	hmR0VmxReadExitQualVmcs(pVmxTransient);
16665	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16666	\| CPUMCTX_EXTRN_HWVIRT
16667	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16668	AssertRCReturn(rc, rc);
16669
16670	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16671
16672	VMXVEXITINFO ExitInfo;
16673	RT_ZERO(ExitInfo);
16674	ExitInfo.uReason = pVmxTransient->uExitReason;
16675	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16676	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16677	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16678	if (!ExitInfo.InstrInfo.VmreadVmwrite.fIsRegOperand)
16679	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16680
16681	VBOXSTRICTRC rcStrict = IEMExecDecodedVmwrite(pVCpu, &ExitInfo);
16682	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16683	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16684	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16685	{
16686	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16687	rcStrict = VINF_SUCCESS;
16688	}
16689	return rcStrict;
16690	}
16691
16692
16693	/**
16694	* VM-exit handler for VMXOFF (VMX_EXIT_VMXOFF). Unconditional VM-exit.
16695	*/
16696	HMVMX_EXIT_DECL hmR0VmxExitVmxoff(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16697	{
16698	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16699
16700	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16701	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_CR4
16702	\| CPUMCTX_EXTRN_HWVIRT
16703	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
16704	AssertRCReturn(rc, rc);
16705
16706	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16707
16708	VBOXSTRICTRC rcStrict = IEMExecDecodedVmxoff(pVCpu, pVmxTransient->cbExitInstr);
16709	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16710	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_HWVIRT);
16711	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16712	{
16713	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16714	rcStrict = VINF_SUCCESS;
16715	}
16716	return rcStrict;
16717	}
16718
16719
16720	/**
16721	* VM-exit handler for VMXON (VMX_EXIT_VMXON). Unconditional VM-exit.
16722	*/
16723	HMVMX_EXIT_DECL hmR0VmxExitVmxon(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16724	{
16725	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16726
16727	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16728	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16729	hmR0VmxReadExitQualVmcs(pVmxTransient);
16730	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16731	\| CPUMCTX_EXTRN_HWVIRT
16732	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16733	AssertRCReturn(rc, rc);
16734
16735	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16736
16737	VMXVEXITINFO ExitInfo;
16738	RT_ZERO(ExitInfo);
16739	ExitInfo.uReason = pVmxTransient->uExitReason;
16740	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16741	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16742	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16743	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16744
16745	VBOXSTRICTRC rcStrict = IEMExecDecodedVmxon(pVCpu, &ExitInfo);
16746	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16747	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS \| HM_CHANGED_GUEST_HWVIRT);
16748	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16749	{
16750	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16751	rcStrict = VINF_SUCCESS;
16752	}
16753	return rcStrict;
16754	}
16755
16756
16757	/**
16758	* VM-exit handler for INVVPID (VMX_EXIT_INVVPID). Unconditional VM-exit.
16759	*/
16760	HMVMX_EXIT_DECL hmR0VmxExitInvvpid(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16761	{
16762	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16763
16764	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16765	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
16766	hmR0VmxReadExitQualVmcs(pVmxTransient);
16767	int rc = hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, CPUMCTX_EXTRN_RSP \| CPUMCTX_EXTRN_SREG_MASK
16768	\| IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
16769	AssertRCReturn(rc, rc);
16770
16771	HMVMX_CHECK_EXIT_DUE_TO_VMX_INSTR(pVCpu, pVmxTransient->uExitReason);
16772
16773	VMXVEXITINFO ExitInfo;
16774	RT_ZERO(ExitInfo);
16775	ExitInfo.uReason = pVmxTransient->uExitReason;
16776	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16777	ExitInfo.InstrInfo.u = pVmxTransient->ExitInstrInfo.u;
16778	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16779	HMVMX_DECODE_MEM_OPERAND(pVCpu, ExitInfo.InstrInfo.u, ExitInfo.u64Qual, VMXMEMACCESS_READ, &ExitInfo.GCPtrEffAddr);
16780
16781	VBOXSTRICTRC rcStrict = IEMExecDecodedInvvpid(pVCpu, &ExitInfo);
16782	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
16783	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
16784	else if (rcStrict == VINF_IEM_RAISED_XCPT)
16785	{
16786	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
16787	rcStrict = VINF_SUCCESS;
16788	}
16789	return rcStrict;
16790	}
16791	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
16792	/** @} */
16793
16794
16795	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
16796	/** @name Nested-guest VM-exit handlers.
16797	* @{
16798	*/
16799	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16800	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- Nested-guest VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16801	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
16802
16803	/**
16804	* Nested-guest VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI).
16805	* Conditional VM-exit.
16806	*/
16807	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmiNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16808	{
16809	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16810
16811	hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
16812
16813	uint64_t const uExitIntInfo = pVmxTransient->uExitIntInfo;
16814	uint32_t const uExitIntType = VMX_EXIT_INT_INFO_TYPE(uExitIntInfo);
16815	Assert(VMX_EXIT_INT_INFO_IS_VALID(uExitIntInfo));
16816
16817	switch (uExitIntType)
16818	{
16819	/*
16820	* Physical NMIs:
16821	* We shouldn't direct host physical NMIs to the nested-guest. Dispatch it to the host.
16822	*/
16823	case VMX_EXIT_INT_INFO_TYPE_NMI:
16824	return hmR0VmxExitHostNmi(pVCpu, pVmxTransient->pVmcsInfo);
16825
16826	/*
16827	* Hardware exceptions,
16828	* Software exceptions,
16829	* Privileged software exceptions:
16830	* Figure out if the exception must be delivered to the guest or the nested-guest.
16831	*/
16832	case VMX_EXIT_INT_INFO_TYPE_SW_XCPT:
16833	case VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT:
16834	case VMX_EXIT_INT_INFO_TYPE_HW_XCPT:
16835	{
16836	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
16837	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16838	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16839	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16840
16841	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
16842	bool const fIntercept = CPUMIsGuestVmxXcptInterceptSet(pCtx, VMX_EXIT_INT_INFO_VECTOR(uExitIntInfo),
16843	pVmxTransient->uExitIntErrorCode);
16844	if (fIntercept)
16845	{
16846	/* Exit qualification is required for debug and page-fault exceptions. */
16847	hmR0VmxReadExitQualVmcs(pVmxTransient);
16848
16849	/*
16850	* For VM-exits due to software exceptions (those generated by INT3 or INTO) and privileged
16851	* software exceptions (those generated by INT1/ICEBP) we need to supply the VM-exit instruction
16852	* length. However, if delivery of a software interrupt, software exception or privileged
16853	* software exception causes a VM-exit, that too provides the VM-exit instruction length.
16854	*/
16855	VMXVEXITINFO ExitInfo;
16856	RT_ZERO(ExitInfo);
16857	ExitInfo.uReason = pVmxTransient->uExitReason;
16858	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16859	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16860
16861	VMXVEXITEVENTINFO ExitEventInfo;
16862	RT_ZERO(ExitEventInfo);
16863	ExitEventInfo.uExitIntInfo = pVmxTransient->uExitIntInfo;
16864	ExitEventInfo.uExitIntErrCode = pVmxTransient->uExitIntErrorCode;
16865	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
16866	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
16867
16868	#ifdef DEBUG_ramshankar
16869	hmR0VmxImportGuestState(pVCpu, pVmxTransient->pVmcsInfo, HMVMX_CPUMCTX_EXTRN_ALL);
16870	Log4Func(("exit_int_info=%#RX32 err_code=%#RX32 exit_qual=%#RX64\n", pVmxTransient->uExitIntInfo,
16871	pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQual));
16872	if (VMX_IDT_VECTORING_INFO_IS_VALID(pVmxTransient->uIdtVectoringInfo))
16873	{
16874	Log4Func(("idt_info=%#RX32 idt_errcode=%#RX32 cr2=%#RX64\n", pVmxTransient->uIdtVectoringInfo,
16875	pVmxTransient->uIdtVectoringErrorCode, pCtx->cr2));
16876	}
16877	#endif
16878	return IEMExecVmxVmexitXcpt(pVCpu, &ExitInfo, &ExitEventInfo);
16879	}
16880
16881	/* Nested paging is currently a requirement, otherwise we would need to handle shadow #PFs in hmR0VmxExitXcptPF. */
16882	Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
16883	return hmR0VmxExitXcpt(pVCpu, pVmxTransient);
16884	}
16885
16886	/*
16887	* Software interrupts:
16888	* VM-exits cannot be caused by software interrupts.
16889	*
16890	* External interrupts:
16891	* This should only happen when "acknowledge external interrupts on VM-exit"
16892	* control is set. However, we never set this when executing a guest or
16893	* nested-guest. For nested-guests it is emulated while injecting interrupts into
16894	* the guest.
16895	*/
16896	case VMX_EXIT_INT_INFO_TYPE_SW_INT:
16897	case VMX_EXIT_INT_INFO_TYPE_EXT_INT:
16898	default:
16899	{
16900	pVCpu->hm.s.u32HMError = pVmxTransient->uExitIntInfo;
16901	return VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
16902	}
16903	}
16904	}
16905
16906
16907	/**
16908	* Nested-guest VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT).
16909	* Unconditional VM-exit.
16910	*/
16911	HMVMX_EXIT_DECL hmR0VmxExitTripleFaultNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16912	{
16913	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16914	return IEMExecVmxVmexitTripleFault(pVCpu);
16915	}
16916
16917
16918	/**
16919	* Nested-guest VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
16920	*/
16921	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindowNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16922	{
16923	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16924
16925	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INT_WINDOW_EXIT))
16926	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
16927	return hmR0VmxExitIntWindow(pVCpu, pVmxTransient);
16928	}
16929
16930
16931	/**
16932	* Nested-guest VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
16933	*/
16934	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindowNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16935	{
16936	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16937
16938	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_NMI_WINDOW_EXIT))
16939	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, 0 /* uExitQual */);
16940	return hmR0VmxExitIntWindow(pVCpu, pVmxTransient);
16941	}
16942
16943
16944	/**
16945	* Nested-guest VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH).
16946	* Unconditional VM-exit.
16947	*/
16948	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitchNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16949	{
16950	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16951
16952	hmR0VmxReadExitQualVmcs(pVmxTransient);
16953	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16954	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
16955	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
16956
16957	VMXVEXITINFO ExitInfo;
16958	RT_ZERO(ExitInfo);
16959	ExitInfo.uReason = pVmxTransient->uExitReason;
16960	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
16961	ExitInfo.u64Qual = pVmxTransient->uExitQual;
16962
16963	VMXVEXITEVENTINFO ExitEventInfo;
16964	RT_ZERO(ExitEventInfo);
16965	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
16966	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
16967	return IEMExecVmxVmexitTaskSwitch(pVCpu, &ExitInfo, &ExitEventInfo);
16968	}
16969
16970
16971	/**
16972	* Nested-guest VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
16973	*/
16974	HMVMX_EXIT_DECL hmR0VmxExitHltNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16975	{
16976	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16977
16978	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_HLT_EXIT))
16979	{
16980	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16981	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
16982	}
16983	return hmR0VmxExitHlt(pVCpu, pVmxTransient);
16984	}
16985
16986
16987	/**
16988	* Nested-guest VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
16989	*/
16990	HMVMX_EXIT_DECL hmR0VmxExitInvlpgNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
16991	{
16992	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
16993
16994	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INVLPG_EXIT))
16995	{
16996	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
16997	hmR0VmxReadExitQualVmcs(pVmxTransient);
16998
16999	VMXVEXITINFO ExitInfo;
17000	RT_ZERO(ExitInfo);
17001	ExitInfo.uReason = pVmxTransient->uExitReason;
17002	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17003	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17004	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17005	}
17006	return hmR0VmxExitInvlpg(pVCpu, pVmxTransient);
17007	}
17008
17009
17010	/**
17011	* Nested-guest VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
17012	*/
17013	HMVMX_EXIT_DECL hmR0VmxExitRdpmcNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17014	{
17015	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17016
17017	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDPMC_EXIT))
17018	{
17019	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17020	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17021	}
17022	return hmR0VmxExitRdpmc(pVCpu, pVmxTransient);
17023	}
17024
17025
17026	/**
17027	* Nested-guest VM-exit handler for VMREAD (VMX_EXIT_VMREAD) and VMWRITE
17028	* (VMX_EXIT_VMWRITE). Conditional VM-exit.
17029	*/
17030	HMVMX_EXIT_DECL hmR0VmxExitVmreadVmwriteNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17031	{
17032	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17033
17034	Assert( pVmxTransient->uExitReason == VMX_EXIT_VMREAD
17035	\|\| pVmxTransient->uExitReason == VMX_EXIT_VMWRITE);
17036
17037	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17038
17039	uint8_t const iGReg = pVmxTransient->ExitInstrInfo.VmreadVmwrite.iReg2;
17040	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
17041	uint64_t u64VmcsField = pVCpu->cpum.GstCtx.aGRegs[iGReg].u64;
17042
17043	HMVMX_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
17044	if (!CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
17045	u64VmcsField &= UINT64_C(0xffffffff);
17046
17047	if (CPUMIsGuestVmxVmreadVmwriteInterceptSet(pVCpu, pVmxTransient->uExitReason, u64VmcsField))
17048	{
17049	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17050	hmR0VmxReadExitQualVmcs(pVmxTransient);
17051
17052	VMXVEXITINFO ExitInfo;
17053	RT_ZERO(ExitInfo);
17054	ExitInfo.uReason = pVmxTransient->uExitReason;
17055	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17056	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17057	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17058	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17059	}
17060
17061	if (pVmxTransient->uExitReason == VMX_EXIT_VMREAD)
17062	return hmR0VmxExitVmread(pVCpu, pVmxTransient);
17063	return hmR0VmxExitVmwrite(pVCpu, pVmxTransient);
17064	}
17065
17066
17067	/**
17068	* Nested-guest VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
17069	*/
17070	HMVMX_EXIT_DECL hmR0VmxExitRdtscNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17071	{
17072	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17073
17074	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDTSC_EXIT))
17075	{
17076	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17077	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17078	}
17079
17080	return hmR0VmxExitRdtsc(pVCpu, pVmxTransient);
17081	}
17082
17083
17084	/**
17085	* Nested-guest VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX).
17086	* Conditional VM-exit.
17087	*/
17088	HMVMX_EXIT_DECL hmR0VmxExitMovCRxNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17089	{
17090	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17091
17092	hmR0VmxReadExitQualVmcs(pVmxTransient);
17093	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17094
17095	VBOXSTRICTRC rcStrict;
17096	uint32_t const uAccessType = VMX_EXIT_QUAL_CRX_ACCESS(pVmxTransient->uExitQual);
17097	switch (uAccessType)
17098	{
17099	case VMX_EXIT_QUAL_CRX_ACCESS_WRITE:
17100	{
17101	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
17102	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(pVmxTransient->uExitQual);
17103	Assert(iGReg < RT_ELEMENTS(pVCpu->cpum.GstCtx.aGRegs));
17104	uint64_t const uNewCrX = pVCpu->cpum.GstCtx.aGRegs[iGReg].u64;
17105
17106	bool fIntercept;
17107	switch (iCrReg)
17108	{
17109	case 0:
17110	case 4:
17111	fIntercept = CPUMIsGuestVmxMovToCr0Cr4InterceptSet(&pVCpu->cpum.GstCtx, iCrReg, uNewCrX);
17112	break;
17113
17114	case 3:
17115	fIntercept = CPUMIsGuestVmxMovToCr3InterceptSet(pVCpu, uNewCrX);
17116	break;
17117
17118	case 8:
17119	fIntercept = CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_CR8_LOAD_EXIT);
17120	break;
17121
17122	default:
17123	fIntercept = false;
17124	break;
17125	}
17126	if (fIntercept)
17127	{
17128	VMXVEXITINFO ExitInfo;
17129	RT_ZERO(ExitInfo);
17130	ExitInfo.uReason = pVmxTransient->uExitReason;
17131	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17132	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17133	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17134	}
17135	else
17136	rcStrict = hmR0VmxExitMovToCrX(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
17137	break;
17138	}
17139
17140	case VMX_EXIT_QUAL_CRX_ACCESS_READ:
17141	{
17142	/*
17143	* CR0/CR4 reads do not cause VM-exits, the read-shadow is used (subject to masking).
17144	* CR2 reads do not cause a VM-exit.
17145	* CR3 reads cause a VM-exit depending on the "CR3 store exiting" control.
17146	* CR8 reads cause a VM-exit depending on the "CR8 store exiting" control.
17147	*/
17148	uint8_t const iCrReg = VMX_EXIT_QUAL_CRX_REGISTER(pVmxTransient->uExitQual);
17149	if ( iCrReg == 3
17150	\|\| iCrReg == 8)
17151	{
17152	static const uint32_t s_auCrXReadIntercepts[] = { 0, 0, 0, VMX_PROC_CTLS_CR3_STORE_EXIT, 0,
17153	0, 0, 0, VMX_PROC_CTLS_CR8_STORE_EXIT };
17154	uint32_t const uIntercept = s_auCrXReadIntercepts[iCrReg];
17155	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, uIntercept))
17156	{
17157	VMXVEXITINFO ExitInfo;
17158	RT_ZERO(ExitInfo);
17159	ExitInfo.uReason = pVmxTransient->uExitReason;
17160	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17161	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17162	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17163	}
17164	else
17165	{
17166	uint8_t const iGReg = VMX_EXIT_QUAL_CRX_GENREG(pVmxTransient->uExitQual);
17167	rcStrict = hmR0VmxExitMovFromCrX(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, iGReg, iCrReg);
17168	}
17169	}
17170	else
17171	{
17172	AssertMsgFailed(("MOV from CR%d VM-exit must not happen\n", iCrReg));
17173	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, iCrReg);
17174	}
17175	break;
17176	}
17177
17178	case VMX_EXIT_QUAL_CRX_ACCESS_CLTS:
17179	{
17180	PCVMXVVMCS const pVmcsNstGst = &pVCpu->cpum.GstCtx.hwvirt.vmx.Vmcs;
17181	uint64_t const uGstHostMask = pVmcsNstGst->u64Cr0Mask.u;
17182	uint64_t const uReadShadow = pVmcsNstGst->u64Cr0ReadShadow.u;
17183	if ( (uGstHostMask & X86_CR0_TS)
17184	&& (uReadShadow & X86_CR0_TS))
17185	{
17186	VMXVEXITINFO ExitInfo;
17187	RT_ZERO(ExitInfo);
17188	ExitInfo.uReason = pVmxTransient->uExitReason;
17189	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17190	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17191	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17192	}
17193	else
17194	rcStrict = hmR0VmxExitClts(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr);
17195	break;
17196	}
17197
17198	case VMX_EXIT_QUAL_CRX_ACCESS_LMSW: /* LMSW (Load Machine-Status Word into CR0) */
17199	{
17200	RTGCPTR GCPtrEffDst;
17201	uint16_t const uNewMsw = VMX_EXIT_QUAL_CRX_LMSW_DATA(pVmxTransient->uExitQual);
17202	bool const fMemOperand = VMX_EXIT_QUAL_CRX_LMSW_OP_MEM(pVmxTransient->uExitQual);
17203	if (fMemOperand)
17204	{
17205	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
17206	GCPtrEffDst = pVmxTransient->uGuestLinearAddr;
17207	}
17208	else
17209	GCPtrEffDst = NIL_RTGCPTR;
17210
17211	if (CPUMIsGuestVmxLmswInterceptSet(&pVCpu->cpum.GstCtx, uNewMsw))
17212	{
17213	VMXVEXITINFO ExitInfo;
17214	RT_ZERO(ExitInfo);
17215	ExitInfo.uReason = pVmxTransient->uExitReason;
17216	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17217	ExitInfo.u64GuestLinearAddr = GCPtrEffDst;
17218	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17219	rcStrict = IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17220	}
17221	else
17222	rcStrict = hmR0VmxExitLmsw(pVCpu, pVmxTransient->pVmcsInfo, pVmxTransient->cbExitInstr, uNewMsw, GCPtrEffDst);
17223	break;
17224	}
17225
17226	default:
17227	{
17228	AssertMsgFailed(("Unrecognized Mov CRX access type %#x\n", uAccessType));
17229	HMVMX_UNEXPECTED_EXIT_RET(pVCpu, uAccessType);
17230	}
17231	}
17232
17233	if (rcStrict == VINF_IEM_RAISED_XCPT)
17234	{
17235	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
17236	rcStrict = VINF_SUCCESS;
17237	}
17238	return rcStrict;
17239	}
17240
17241
17242	/**
17243	* Nested-guest VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX).
17244	* Conditional VM-exit.
17245	*/
17246	HMVMX_EXIT_DECL hmR0VmxExitMovDRxNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17247	{
17248	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17249
17250	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MOV_DR_EXIT))
17251	{
17252	hmR0VmxReadExitQualVmcs(pVmxTransient);
17253	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17254
17255	VMXVEXITINFO ExitInfo;
17256	RT_ZERO(ExitInfo);
17257	ExitInfo.uReason = pVmxTransient->uExitReason;
17258	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17259	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17260	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17261	}
17262	return hmR0VmxExitMovDRx(pVCpu, pVmxTransient);
17263	}
17264
17265
17266	/**
17267	* Nested-guest VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR).
17268	* Conditional VM-exit.
17269	*/
17270	HMVMX_EXIT_DECL hmR0VmxExitIoInstrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17271	{
17272	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17273
17274	hmR0VmxReadExitQualVmcs(pVmxTransient);
17275
17276	uint32_t const uIOPort = VMX_EXIT_QUAL_IO_PORT(pVmxTransient->uExitQual);
17277	uint8_t const uIOSize = VMX_EXIT_QUAL_IO_SIZE(pVmxTransient->uExitQual);
17278	AssertReturn(uIOSize <= 3 && uIOSize != 2, VERR_VMX_IPE_1);
17279
17280	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses in bytes. */
17281	uint8_t const cbAccess = s_aIOSizes[uIOSize];
17282	if (CPUMIsGuestVmxIoInterceptSet(pVCpu, uIOPort, cbAccess))
17283	{
17284	/*
17285	* IN/OUT instruction:
17286	* - Provides VM-exit instruction length.
17287	*
17288	* INS/OUTS instruction:
17289	* - Provides VM-exit instruction length.
17290	* - Provides Guest-linear address.
17291	* - Optionally provides VM-exit instruction info (depends on CPU feature).
17292	*/
17293	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
17294	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17295
17296	/* Make sure we don't use stale/uninitialized VMX-transient info. below. */
17297	pVmxTransient->ExitInstrInfo.u = 0;
17298	pVmxTransient->uGuestLinearAddr = 0;
17299
17300	bool const fVmxInsOutsInfo = pVM->cpum.ro.GuestFeatures.fVmxInsOutInfo;
17301	bool const fIOString = VMX_EXIT_QUAL_IO_IS_STRING(pVmxTransient->uExitQual);
17302	if (fIOString)
17303	{
17304	hmR0VmxReadGuestLinearAddrVmcs(pVmxTransient);
17305	if (fVmxInsOutsInfo)
17306	{
17307	Assert(RT_BF_GET(g_HmMsrs.u.vmx.u64Basic, VMX_BF_BASIC_VMCS_INS_OUTS)); /* Paranoia. */
17308	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17309	}
17310	}
17311
17312	VMXVEXITINFO ExitInfo;
17313	RT_ZERO(ExitInfo);
17314	ExitInfo.uReason = pVmxTransient->uExitReason;
17315	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17316	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17317	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17318	ExitInfo.u64GuestLinearAddr = pVmxTransient->uGuestLinearAddr;
17319	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17320	}
17321	return hmR0VmxExitIoInstr(pVCpu, pVmxTransient);
17322	}
17323
17324
17325	/**
17326	* Nested-guest VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
17327	*/
17328	HMVMX_EXIT_DECL hmR0VmxExitRdmsrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17329	{
17330	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17331
17332	uint32_t fMsrpm;
17333	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_MSR_BITMAPS))
17334	fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap, pVCpu->cpum.GstCtx.ecx);
17335	else
17336	fMsrpm = VMXMSRPM_EXIT_RD;
17337
17338	if (fMsrpm & VMXMSRPM_EXIT_RD)
17339	{
17340	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17341	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17342	}
17343	return hmR0VmxExitRdmsr(pVCpu, pVmxTransient);
17344	}
17345
17346
17347	/**
17348	* Nested-guest VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
17349	*/
17350	HMVMX_EXIT_DECL hmR0VmxExitWrmsrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17351	{
17352	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17353
17354	uint32_t fMsrpm;
17355	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_MSR_BITMAPS))
17356	fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.abMsrBitmap, pVCpu->cpum.GstCtx.ecx);
17357	else
17358	fMsrpm = VMXMSRPM_EXIT_WR;
17359
17360	if (fMsrpm & VMXMSRPM_EXIT_WR)
17361	{
17362	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17363	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17364	}
17365	return hmR0VmxExitWrmsr(pVCpu, pVmxTransient);
17366	}
17367
17368
17369	/**
17370	* Nested-guest VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
17371	*/
17372	HMVMX_EXIT_DECL hmR0VmxExitMwaitNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17373	{
17374	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17375
17376	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MWAIT_EXIT))
17377	{
17378	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17379	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17380	}
17381	return hmR0VmxExitMwait(pVCpu, pVmxTransient);
17382	}
17383
17384
17385	/**
17386	* Nested-guest VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional
17387	* VM-exit.
17388	*/
17389	HMVMX_EXIT_DECL hmR0VmxExitMtfNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17390	{
17391	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17392
17393	/** @todo NSTVMX: Should consider debugging nested-guests using VM debugger. */
17394	hmR0VmxReadGuestPendingDbgXctps(pVmxTransient);
17395	VMXVEXITINFO ExitInfo;
17396	RT_ZERO(ExitInfo);
17397	ExitInfo.uReason = pVmxTransient->uExitReason;
17398	ExitInfo.u64GuestPendingDbgXcpts = pVmxTransient->uGuestPendingDbgXcpts;
17399	return IEMExecVmxVmexitTrapLike(pVCpu, &ExitInfo);
17400	}
17401
17402
17403	/**
17404	* Nested-guest VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
17405	*/
17406	HMVMX_EXIT_DECL hmR0VmxExitMonitorNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17407	{
17408	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17409
17410	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_MONITOR_EXIT))
17411	{
17412	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17413	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17414	}
17415	return hmR0VmxExitMonitor(pVCpu, pVmxTransient);
17416	}
17417
17418
17419	/**
17420	* Nested-guest VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
17421	*/
17422	HMVMX_EXIT_DECL hmR0VmxExitPauseNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17423	{
17424	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17425
17426	/** @todo NSTVMX: Think about this more. Does the outer guest need to intercept
17427	* PAUSE when executing a nested-guest? If it does not, we would not need
17428	* to check for the intercepts here. Just call VM-exit... */
17429
17430	/* The CPU would have already performed the necessary CPL checks for PAUSE-loop exiting. */
17431	if ( CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_PAUSE_EXIT)
17432	\|\| CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_PAUSE_LOOP_EXIT))
17433	{
17434	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17435	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17436	}
17437	return hmR0VmxExitPause(pVCpu, pVmxTransient);
17438	}
17439
17440
17441	/**
17442	* Nested-guest VM-exit handler for when the TPR value is lowered below the
17443	* specified threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
17444	*/
17445	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThresholdNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17446	{
17447	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17448
17449	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_USE_TPR_SHADOW))
17450	{
17451	hmR0VmxReadGuestPendingDbgXctps(pVmxTransient);
17452	VMXVEXITINFO ExitInfo;
17453	RT_ZERO(ExitInfo);
17454	ExitInfo.uReason = pVmxTransient->uExitReason;
17455	ExitInfo.u64GuestPendingDbgXcpts = pVmxTransient->uGuestPendingDbgXcpts;
17456	return IEMExecVmxVmexitTrapLike(pVCpu, &ExitInfo);
17457	}
17458	return hmR0VmxExitTprBelowThreshold(pVCpu, pVmxTransient);
17459	}
17460
17461
17462	/**
17463	* Nested-guest VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional
17464	* VM-exit.
17465	*/
17466	HMVMX_EXIT_DECL hmR0VmxExitApicAccessNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17467	{
17468	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17469
17470	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17471	hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
17472	hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
17473	hmR0VmxReadExitQualVmcs(pVmxTransient);
17474
17475	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_VIRT_APIC_ACCESS));
17476
17477	Log4Func(("at offset %#x type=%u\n", VMX_EXIT_QUAL_APIC_ACCESS_OFFSET(pVmxTransient->uExitQual),
17478	VMX_EXIT_QUAL_APIC_ACCESS_TYPE(pVmxTransient->uExitQual)));
17479
17480	VMXVEXITINFO ExitInfo;
17481	RT_ZERO(ExitInfo);
17482	ExitInfo.uReason = pVmxTransient->uExitReason;
17483	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17484	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17485
17486	VMXVEXITEVENTINFO ExitEventInfo;
17487	RT_ZERO(ExitEventInfo);
17488	ExitEventInfo.uIdtVectoringInfo = pVmxTransient->uIdtVectoringInfo;
17489	ExitEventInfo.uIdtVectoringErrCode = pVmxTransient->uIdtVectoringErrorCode;
17490	return IEMExecVmxVmexitApicAccess(pVCpu, &ExitInfo, &ExitEventInfo);
17491	}
17492
17493
17494	/**
17495	* Nested-guest VM-exit handler for APIC write emulation (VMX_EXIT_APIC_WRITE).
17496	* Conditional VM-exit.
17497	*/
17498	HMVMX_EXIT_DECL hmR0VmxExitApicWriteNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17499	{
17500	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17501
17502	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_APIC_REG_VIRT));
17503	hmR0VmxReadExitQualVmcs(pVmxTransient);
17504	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
17505	}
17506
17507
17508	/**
17509	* Nested-guest VM-exit handler for virtualized EOI (VMX_EXIT_VIRTUALIZED_EOI).
17510	* Conditional VM-exit.
17511	*/
17512	HMVMX_EXIT_DECL hmR0VmxExitVirtEoiNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17513	{
17514	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17515
17516	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_VIRT_INT_DELIVERY));
17517	hmR0VmxReadExitQualVmcs(pVmxTransient);
17518	return IEMExecVmxVmexit(pVCpu, pVmxTransient->uExitReason, pVmxTransient->uExitQual);
17519	}
17520
17521
17522	/**
17523	* Nested-guest VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
17524	*/
17525	HMVMX_EXIT_DECL hmR0VmxExitRdtscpNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17526	{
17527	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17528
17529	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_RDTSC_EXIT))
17530	{
17531	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_RDTSCP));
17532	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17533	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17534	}
17535	return hmR0VmxExitRdtscp(pVCpu, pVmxTransient);
17536	}
17537
17538
17539	/**
17540	* Nested-guest VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
17541	*/
17542	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvdNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17543	{
17544	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17545
17546	if (CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_WBINVD_EXIT))
17547	{
17548	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17549	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17550	}
17551	return hmR0VmxExitWbinvd(pVCpu, pVmxTransient);
17552	}
17553
17554
17555	/**
17556	* Nested-guest VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
17557	*/
17558	HMVMX_EXIT_DECL hmR0VmxExitInvpcidNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17559	{
17560	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17561
17562	if (CPUMIsGuestVmxProcCtlsSet(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS_INVLPG_EXIT))
17563	{
17564	Assert(CPUMIsGuestVmxProcCtls2Set(&pVCpu->cpum.GstCtx, VMX_PROC_CTLS2_INVPCID));
17565	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17566	hmR0VmxReadExitQualVmcs(pVmxTransient);
17567	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17568
17569	VMXVEXITINFO ExitInfo;
17570	RT_ZERO(ExitInfo);
17571	ExitInfo.uReason = pVmxTransient->uExitReason;
17572	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17573	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17574	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17575	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17576	}
17577	return hmR0VmxExitInvpcid(pVCpu, pVmxTransient);
17578	}
17579
17580
17581	/**
17582	* Nested-guest VM-exit handler for invalid-guest state
17583	* (VMX_EXIT_ERR_INVALID_GUEST_STATE). Error VM-exit.
17584	*/
17585	HMVMX_EXIT_DECL hmR0VmxExitErrInvalidGuestStateNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17586	{
17587	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17588
17589	/*
17590	* Currently this should never happen because we fully emulate VMLAUNCH/VMRESUME in IEM.
17591	* So if it does happen, it indicates a bug possibly in the hardware-assisted VMX code.
17592	* Handle it like it's in an invalid guest state of the outer guest.
17593	*
17594	* When the fast path is implemented, this should be changed to cause the corresponding
17595	* nested-guest VM-exit.
17596	*/
17597	return hmR0VmxExitErrInvalidGuestState(pVCpu, pVmxTransient);
17598	}
17599
17600
17601	/**
17602	* Nested-guest VM-exit handler for instructions that cause VM-exits uncondtionally
17603	* and only provide the instruction length.
17604	*
17605	* Unconditional VM-exit.
17606	*/
17607	HMVMX_EXIT_DECL hmR0VmxExitInstrNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17608	{
17609	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17610
17611	#ifdef VBOX_STRICT
17612	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
17613	switch (pVmxTransient->uExitReason)
17614	{
17615	case VMX_EXIT_ENCLS:
17616	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_ENCLS_EXIT));
17617	break;
17618
17619	case VMX_EXIT_VMFUNC:
17620	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_VMFUNC));
17621	break;
17622	}
17623	#endif
17624
17625	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17626	return IEMExecVmxVmexitInstr(pVCpu, pVmxTransient->uExitReason, pVmxTransient->cbExitInstr);
17627	}
17628
17629
17630	/**
17631	* Nested-guest VM-exit handler for instructions that provide instruction length as
17632	* well as more information.
17633	*
17634	* Unconditional VM-exit.
17635	*/
17636	HMVMX_EXIT_DECL hmR0VmxExitInstrWithInfoNested(PVMCPUCC pVCpu, PVMXTRANSIENT pVmxTransient)
17637	{
17638	HMVMX_VALIDATE_NESTED_EXIT_HANDLER_PARAMS(pVCpu, pVmxTransient);
17639
17640	#ifdef VBOX_STRICT
17641	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
17642	switch (pVmxTransient->uExitReason)
17643	{
17644	case VMX_EXIT_GDTR_IDTR_ACCESS:
17645	case VMX_EXIT_LDTR_TR_ACCESS:
17646	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_DESC_TABLE_EXIT));
17647	break;
17648
17649	case VMX_EXIT_RDRAND:
17650	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_RDRAND_EXIT));
17651	break;
17652
17653	case VMX_EXIT_RDSEED:
17654	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_RDSEED_EXIT));
17655	break;
17656
17657	case VMX_EXIT_XSAVES:
17658	case VMX_EXIT_XRSTORS:
17659	/** @todo NSTVMX: Verify XSS-bitmap. */
17660	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_XSAVES_XRSTORS));
17661	break;
17662
17663	case VMX_EXIT_UMWAIT:
17664	case VMX_EXIT_TPAUSE:
17665	Assert(CPUMIsGuestVmxProcCtlsSet(pCtx, VMX_PROC_CTLS_RDTSC_EXIT));
17666	Assert(CPUMIsGuestVmxProcCtls2Set(pCtx, VMX_PROC_CTLS2_USER_WAIT_PAUSE));
17667	break;
17668
17669	case VMX_EXIT_LOADIWKEY:
17670	Assert(CPUMIsGuestVmxProcCtls3Set(pCtx, VMX_PROC_CTLS3_LOADIWKEY_EXIT));
17671	break;
17672	}
17673	#endif
17674
17675	hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
17676	hmR0VmxReadExitQualVmcs(pVmxTransient);
17677	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
17678
17679	VMXVEXITINFO ExitInfo;
17680	RT_ZERO(ExitInfo);
17681	ExitInfo.uReason = pVmxTransient->uExitReason;
17682	ExitInfo.cbInstr = pVmxTransient->cbExitInstr;
17683	ExitInfo.u64Qual = pVmxTransient->uExitQual;
17684	ExitInfo.InstrInfo = pVmxTransient->ExitInstrInfo;
17685	return IEMExecVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
17686	}
17687
17688	/** @} */
17689	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
17690

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

source: vbox/trunk/src/VBox/VMM/VMMR0/HMVMXR0.cpp@ 91587

Download in other formats: