HMSVMR0.cpp@ 70721

Last change on this file since 70721 was 70721, checked in by vboxsync, 7 years ago
VMM/HMSVMR0: Nested Hw.virt: Exclude VINTR intercept of the outer guest while executing the nested-guest. It will remain in the outer guest's VMCB and thus should be used when switching to the outer guest.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 306.4 KB

Line
1	/* $Id: HMSVMR0.cpp 70721 2018-01-24 09:00:50Z vboxsync $ */
2	/** @file
3	* HM SVM (AMD-V) - Host Context Ring-0.
4	*/
5
6	/*
7	* Copyright (C) 2013-2017 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/asm-amd64-x86.h>
25	#include <iprt/thread.h>
26
27	#include <VBox/vmm/pdmapi.h>
28	#include <VBox/vmm/dbgf.h>
29	#include <VBox/vmm/iem.h>
30	#include <VBox/vmm/iom.h>
31	#include <VBox/vmm/tm.h>
32	#include <VBox/vmm/gim.h>
33	#include <VBox/vmm/apic.h>
34	#include "HMInternal.h"
35	#include <VBox/vmm/vm.h>
36	#include "HMSVMR0.h"
37	#include "dtrace/VBoxVMM.h"
38
39	#define HMSVM_USE_IEM_EVENT_REFLECTION
40	#ifdef DEBUG_ramshankar
41	# define HMSVM_SYNC_FULL_GUEST_STATE
42	# define HMSVM_SYNC_FULL_NESTED_GUEST_STATE
43	# define HMSVM_ALWAYS_TRAP_ALL_XCPTS
44	# define HMSVM_ALWAYS_TRAP_PF
45	# define HMSVM_ALWAYS_TRAP_TASK_SWITCH
46	#endif
47
48
49	/*********************************************************************************************************************************
50	* Defined Constants And Macros *
51	*********************************************************************************************************************************/
52	#ifdef VBOX_WITH_STATISTICS
53	# define HMSVM_EXITCODE_STAM_COUNTER_INC(u64ExitCode) do { \
54	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll); \
55	if ((u64ExitCode) == SVM_EXIT_NPF) \
56	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf); \
57	else \
58	STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[(u64ExitCode) & MASK_EXITREASON_STAT]); \
59	} while (0)
60	#else
61	# define HMSVM_EXITCODE_STAM_COUNTER_INC(u64ExitCode) do { } while (0)
62	#endif
63
64	/** If we decide to use a function table approach this can be useful to
65	* switch to a "static DECLCALLBACK(int)". */
66	#define HMSVM_EXIT_DECL static int
67
68	/** Macro for checking and returning from the using function for
69	* \#VMEXIT intercepts that maybe caused during delivering of another
70	* event in the guest. */
71	#ifdef VBOX_WITH_NESTED_HWVIRT
72	# define HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY() \
73	do \
74	{ \
75	int rc = hmR0SvmCheckExitDueToEventDelivery(pVCpu, pCtx, pSvmTransient); \
76	if (RT_LIKELY(rc == VINF_SUCCESS)) { /* continue #VMEXIT handling */ } \
77	else if ( rc == VINF_HM_DOUBLE_FAULT) { return VINF_SUCCESS; } \
78	else if ( rc == VINF_EM_RESET \
79	&& HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_SHUTDOWN)) \
80	return VBOXSTRICTRC_TODO(IEMExecSvmVmexit(pVCpu, SVM_EXIT_SHUTDOWN, 0, 0)); \
81	else \
82	return rc; \
83	} while (0)
84	#else
85	# define HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY() \
86	do \
87	{ \
88	int rc = hmR0SvmCheckExitDueToEventDelivery(pVCpu, pCtx, pSvmTransient); \
89	if (RT_LIKELY(rc == VINF_SUCCESS)) { /* continue #VMEXIT handling */ } \
90	else if ( rc == VINF_HM_DOUBLE_FAULT) { return VINF_SUCCESS; } \
91	else \
92	return rc; \
93	} while (0)
94	#endif
95
96	/**
97	* Updates interrupt shadow for the current RIP.
98	*/
99	#define HMSVM_UPDATE_INTR_SHADOW(pVCpu, pCtx) \
100	do { \
101	/* Update interrupt shadow. */ \
102	if ( VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS) \
103	&& pCtx->rip != EMGetInhibitInterruptsPC(pVCpu)) \
104	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS); \
105	} while (0)
106
107	/** Macro for upgrading a @a a_rc to VINF_EM_DBG_STEPPED after emulating an
108	* instruction that exited. */
109	#define HMSVM_CHECK_SINGLE_STEP(a_pVCpu, a_rc) \
110	do { \
111	if ((a_pVCpu)->hm.s.fSingleInstruction && (a_rc) == VINF_SUCCESS) \
112	(a_rc) = VINF_EM_DBG_STEPPED; \
113	} while (0)
114
115	/** Assert that preemption is disabled or covered by thread-context hooks. */
116	#define HMSVM_ASSERT_PREEMPT_SAFE() Assert( VMMR0ThreadCtxHookIsEnabled(pVCpu) \
117	\|\| !RTThreadPreemptIsEnabled(NIL_RTTHREAD));
118
119	/** Assert that we haven't migrated CPUs when thread-context hooks are not
120	* used. */
121	#define HMSVM_ASSERT_CPU_SAFE() AssertMsg( VMMR0ThreadCtxHookIsEnabled(pVCpu) \
122	\|\| pVCpu->hm.s.idEnteredCpu == RTMpCpuId(), \
123	("Illegal migration! Entered on CPU %u Current %u\n", \
124	pVCpu->hm.s.idEnteredCpu, RTMpCpuId()));
125
126	/** Assert that we're not executing a nested-guest. */
127	#ifdef VBOX_WITH_NESTED_HWVIRT
128	# define HMSVM_ASSERT_NOT_IN_NESTED_GUEST(a_pCtx) Assert(!CPUMIsGuestInSvmNestedHwVirtMode((a_pCtx)))
129	#else
130	# define HMSVM_ASSERT_NOT_IN_NESTED_GUEST(a_pCtx) do { NOREF((a_pCtx)); } while (0)
131	#endif
132
133	/** Assert that we're executing a nested-guest. */
134	#ifdef VBOX_WITH_NESTED_HWVIRT
135	# define HMSVM_ASSERT_IN_NESTED_GUEST(a_pCtx) Assert(CPUMIsGuestInSvmNestedHwVirtMode((a_pCtx)))
136	#else
137	# define HMSVM_ASSERT_IN_NESTED_GUEST(a_pCtx) do { NOREF((a_pCtx)); } while (0)
138	#endif
139
140	/** Validate segment descriptor granularity bit. */
141	#ifdef VBOX_STRICT
142	# define HMSVM_ASSERT_SEG_GRANULARITY(reg) \
143	AssertMsg( !pMixedCtx->reg.Attr.n.u1Present \
144	\|\| ( pMixedCtx->reg.Attr.n.u1Granularity \
145	? (pMixedCtx->reg.u32Limit & 0xfff) == 0xfff \
146	: pMixedCtx->reg.u32Limit <= UINT32_C(0xfffff)), \
147	("Invalid Segment Attributes Limit=%#RX32 Attr=%#RX32 Base=%#RX64\n", pMixedCtx->reg.u32Limit, \
148	pMixedCtx->reg.Attr.u, pMixedCtx->reg.u64Base))
149	#else
150	# define HMSVM_ASSERT_SEG_GRANULARITY(reg) do { } while (0)
151	#endif
152
153	/**
154	* Exception bitmap mask for all contributory exceptions.
155	*
156	* Page fault is deliberately excluded here as it's conditional as to whether
157	* it's contributory or benign. Page faults are handled separately.
158	*/
159	#define HMSVM_CONTRIBUTORY_XCPT_MASK ( RT_BIT(X86_XCPT_GP) \| RT_BIT(X86_XCPT_NP) \| RT_BIT(X86_XCPT_SS) \| RT_BIT(X86_XCPT_TS) \
160	\| RT_BIT(X86_XCPT_DE))
161
162	/**
163	* Mandatory/unconditional guest control intercepts.
164	*
165	* SMIs can and do happen in normal operation. We need not intercept them
166	* while executing the guest or nested-guest.
167	*/
168	#define HMSVM_MANDATORY_GUEST_CTRL_INTERCEPTS ( SVM_CTRL_INTERCEPT_INTR \
169	\| SVM_CTRL_INTERCEPT_NMI \
170	\| SVM_CTRL_INTERCEPT_INIT \
171	\| SVM_CTRL_INTERCEPT_RDPMC \
172	\| SVM_CTRL_INTERCEPT_CPUID \
173	\| SVM_CTRL_INTERCEPT_RSM \
174	\| SVM_CTRL_INTERCEPT_HLT \
175	\| SVM_CTRL_INTERCEPT_IOIO_PROT \
176	\| SVM_CTRL_INTERCEPT_MSR_PROT \
177	\| SVM_CTRL_INTERCEPT_INVLPGA \
178	\| SVM_CTRL_INTERCEPT_SHUTDOWN \
179	\| SVM_CTRL_INTERCEPT_FERR_FREEZE \
180	\| SVM_CTRL_INTERCEPT_VMRUN \
181	\| SVM_CTRL_INTERCEPT_VMMCALL \
182	\| SVM_CTRL_INTERCEPT_SKINIT \
183	\| SVM_CTRL_INTERCEPT_WBINVD \
184	\| SVM_CTRL_INTERCEPT_MONITOR \
185	\| SVM_CTRL_INTERCEPT_MWAIT \
186	\| SVM_CTRL_INTERCEPT_XSETBV)
187
188	/** @name VMCB Clean Bits.
189	*
190	* These flags are used for VMCB-state caching. A set VMCB Clean bit indicates
191	* AMD-V doesn't need to reload the corresponding value(s) from the VMCB in
192	* memory.
193	*
194	* @{ */
195	/** All intercepts vectors, TSC offset, PAUSE filter counter. */
196	#define HMSVM_VMCB_CLEAN_INTERCEPTS RT_BIT(0)
197	/** I/O permission bitmap, MSR permission bitmap. */
198	#define HMSVM_VMCB_CLEAN_IOPM_MSRPM RT_BIT(1)
199	/** ASID. */
200	#define HMSVM_VMCB_CLEAN_ASID RT_BIT(2)
201	/** TRP: V_TPR, V_IRQ, V_INTR_PRIO, V_IGN_TPR, V_INTR_MASKING,
202	V_INTR_VECTOR. */
203	#define HMSVM_VMCB_CLEAN_TPR RT_BIT(3)
204	/** Nested Paging: Nested CR3 (nCR3), PAT. */
205	#define HMSVM_VMCB_CLEAN_NP RT_BIT(4)
206	/** Control registers (CR0, CR3, CR4, EFER). */
207	#define HMSVM_VMCB_CLEAN_CRX_EFER RT_BIT(5)
208	/** Debug registers (DR6, DR7). */
209	#define HMSVM_VMCB_CLEAN_DRX RT_BIT(6)
210	/** GDT, IDT limit and base. */
211	#define HMSVM_VMCB_CLEAN_DT RT_BIT(7)
212	/** Segment register: CS, SS, DS, ES limit and base. */
213	#define HMSVM_VMCB_CLEAN_SEG RT_BIT(8)
214	/** CR2.*/
215	#define HMSVM_VMCB_CLEAN_CR2 RT_BIT(9)
216	/** Last-branch record (DbgCtlMsr, br_from, br_to, lastint_from, lastint_to) */
217	#define HMSVM_VMCB_CLEAN_LBR RT_BIT(10)
218	/** AVIC (AVIC APIC_BAR; AVIC APIC_BACKING_PAGE, AVIC
219	PHYSICAL_TABLE and AVIC LOGICAL_TABLE Pointers). */
220	#define HMSVM_VMCB_CLEAN_AVIC RT_BIT(11)
221	/** Mask of all valid VMCB Clean bits. */
222	#define HMSVM_VMCB_CLEAN_ALL ( HMSVM_VMCB_CLEAN_INTERCEPTS \
223	\| HMSVM_VMCB_CLEAN_IOPM_MSRPM \
224	\| HMSVM_VMCB_CLEAN_ASID \
225	\| HMSVM_VMCB_CLEAN_TPR \
226	\| HMSVM_VMCB_CLEAN_NP \
227	\| HMSVM_VMCB_CLEAN_CRX_EFER \
228	\| HMSVM_VMCB_CLEAN_DRX \
229	\| HMSVM_VMCB_CLEAN_DT \
230	\| HMSVM_VMCB_CLEAN_SEG \
231	\| HMSVM_VMCB_CLEAN_CR2 \
232	\| HMSVM_VMCB_CLEAN_LBR \
233	\| HMSVM_VMCB_CLEAN_AVIC)
234	/** @} */
235
236	/** @name SVM transient.
237	*
238	* A state structure for holding miscellaneous information across AMD-V
239	* VMRUN/\#VMEXIT operation, restored after the transition.
240	*
241	* @{ */
242	typedef struct SVMTRANSIENT
243	{
244	/** The host's rflags/eflags. */
245	RTCCUINTREG fEFlags;
246	#if HC_ARCH_BITS == 32
247	uint32_t u32Alignment0;
248	#endif
249
250	/** The \#VMEXIT exit code (the EXITCODE field in the VMCB). */
251	uint64_t u64ExitCode;
252	/** The guest's TPR value used for TPR shadowing. */
253	uint8_t u8GuestTpr;
254	/** Alignment. */
255	uint8_t abAlignment0[7];
256
257	/** Whether the guest FPU state was active at the time of \#VMEXIT. */
258	bool fWasGuestFPUStateActive;
259	/** Whether the guest debug state was active at the time of \#VMEXIT. */
260	bool fWasGuestDebugStateActive;
261	/** Whether the hyper debug state was active at the time of \#VMEXIT. */
262	bool fWasHyperDebugStateActive;
263	/** Whether the TSC offset mode needs to be updated. */
264	bool fUpdateTscOffsetting;
265	/** Whether the TSC_AUX MSR needs restoring on \#VMEXIT. */
266	bool fRestoreTscAuxMsr;
267	/** Whether the \#VMEXIT was caused by a page-fault during delivery of a
268	* contributary exception or a page-fault. */
269	bool fVectoringDoublePF;
270	/** Whether the \#VMEXIT was caused by a page-fault during delivery of an
271	* external interrupt or NMI. */
272	bool fVectoringPF;
273	} SVMTRANSIENT, *PSVMTRANSIENT;
274	AssertCompileMemberAlignment(SVMTRANSIENT, u64ExitCode, sizeof(uint64_t));
275	AssertCompileMemberAlignment(SVMTRANSIENT, fWasGuestFPUStateActive, sizeof(uint64_t));
276	/** @} */
277
278	/**
279	* MSRPM (MSR permission bitmap) read permissions (for guest RDMSR).
280	*/
281	typedef enum SVMMSREXITREAD
282	{
283	/** Reading this MSR causes a \#VMEXIT. */
284	SVMMSREXIT_INTERCEPT_READ = 0xb,
285	/** Reading this MSR does not cause a \#VMEXIT. */
286	SVMMSREXIT_PASSTHRU_READ
287	} SVMMSREXITREAD;
288
289	/**
290	* MSRPM (MSR permission bitmap) write permissions (for guest WRMSR).
291	*/
292	typedef enum SVMMSREXITWRITE
293	{
294	/** Writing to this MSR causes a \#VMEXIT. */
295	SVMMSREXIT_INTERCEPT_WRITE = 0xd,
296	/** Writing to this MSR does not cause a \#VMEXIT. */
297	SVMMSREXIT_PASSTHRU_WRITE
298	} SVMMSREXITWRITE;
299
300	/**
301	* SVM \#VMEXIT handler.
302	*
303	* @returns VBox status code.
304	* @param pVCpu The cross context virtual CPU structure.
305	* @param pMixedCtx Pointer to the guest-CPU context.
306	* @param pSvmTransient Pointer to the SVM-transient structure.
307	*/
308	typedef int FNSVMEXITHANDLER(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient);
309
310
311	/*********************************************************************************************************************************
312	* Internal Functions *
313	*********************************************************************************************************************************/
314	static void hmR0SvmSetMsrPermission(PSVMVMCB pVmcb, uint8_t *pbMsrBitmap, unsigned uMsr, SVMMSREXITREAD enmRead,
315	SVMMSREXITWRITE enmWrite);
316	static void hmR0SvmPendingEventToTrpmTrap(PVMCPU pVCpu);
317	static void hmR0SvmLeave(PVMCPU pVCpu);
318
319	/** @name \#VMEXIT handlers.
320	* @{
321	*/
322	static FNSVMEXITHANDLER hmR0SvmExitIntr;
323	static FNSVMEXITHANDLER hmR0SvmExitWbinvd;
324	static FNSVMEXITHANDLER hmR0SvmExitInvd;
325	static FNSVMEXITHANDLER hmR0SvmExitCpuid;
326	static FNSVMEXITHANDLER hmR0SvmExitRdtsc;
327	static FNSVMEXITHANDLER hmR0SvmExitRdtscp;
328	static FNSVMEXITHANDLER hmR0SvmExitRdpmc;
329	static FNSVMEXITHANDLER hmR0SvmExitInvlpg;
330	static FNSVMEXITHANDLER hmR0SvmExitHlt;
331	static FNSVMEXITHANDLER hmR0SvmExitMonitor;
332	static FNSVMEXITHANDLER hmR0SvmExitMwait;
333	static FNSVMEXITHANDLER hmR0SvmExitShutdown;
334	static FNSVMEXITHANDLER hmR0SvmExitUnexpected;
335	static FNSVMEXITHANDLER hmR0SvmExitReadCRx;
336	static FNSVMEXITHANDLER hmR0SvmExitWriteCRx;
337	static FNSVMEXITHANDLER hmR0SvmExitMsr;
338	static FNSVMEXITHANDLER hmR0SvmExitReadDRx;
339	static FNSVMEXITHANDLER hmR0SvmExitWriteDRx;
340	static FNSVMEXITHANDLER hmR0SvmExitXsetbv;
341	static FNSVMEXITHANDLER hmR0SvmExitIOInstr;
342	static FNSVMEXITHANDLER hmR0SvmExitNestedPF;
343	static FNSVMEXITHANDLER hmR0SvmExitVIntr;
344	static FNSVMEXITHANDLER hmR0SvmExitTaskSwitch;
345	static FNSVMEXITHANDLER hmR0SvmExitVmmCall;
346	static FNSVMEXITHANDLER hmR0SvmExitPause;
347	static FNSVMEXITHANDLER hmR0SvmExitIret;
348	static FNSVMEXITHANDLER hmR0SvmExitXcptPF;
349	static FNSVMEXITHANDLER hmR0SvmExitXcptNM;
350	static FNSVMEXITHANDLER hmR0SvmExitXcptUD;
351	static FNSVMEXITHANDLER hmR0SvmExitXcptMF;
352	static FNSVMEXITHANDLER hmR0SvmExitXcptDB;
353	static FNSVMEXITHANDLER hmR0SvmExitXcptAC;
354	static FNSVMEXITHANDLER hmR0SvmExitXcptBP;
355	#if defined(HMSVM_ALWAYS_TRAP_ALL_XCPTS) \|\| defined(VBOX_WITH_NESTED_HWVIRT)
356	static FNSVMEXITHANDLER hmR0SvmExitXcptGeneric;
357	#endif
358	#ifdef VBOX_WITH_NESTED_HWVIRT
359	static FNSVMEXITHANDLER hmR0SvmExitXcptPFNested;
360	static FNSVMEXITHANDLER hmR0SvmExitClgi;
361	static FNSVMEXITHANDLER hmR0SvmExitStgi;
362	static FNSVMEXITHANDLER hmR0SvmExitVmload;
363	static FNSVMEXITHANDLER hmR0SvmExitVmsave;
364	static FNSVMEXITHANDLER hmR0SvmExitInvlpga;
365	static FNSVMEXITHANDLER hmR0SvmExitVmrun;
366	static FNSVMEXITHANDLER hmR0SvmNestedExitXcptDB;
367	static FNSVMEXITHANDLER hmR0SvmNestedExitXcptBP;
368	#endif
369	/** @} */
370
371	static int hmR0SvmHandleExit(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PSVMTRANSIENT pSvmTransient);
372	#ifdef VBOX_WITH_NESTED_HWVIRT
373	static int hmR0SvmHandleExitNested(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient);
374	#endif
375
376
377	/*********************************************************************************************************************************
378	* Global Variables *
379	*********************************************************************************************************************************/
380	/** Ring-0 memory object for the IO bitmap. */
381	RTR0MEMOBJ g_hMemObjIOBitmap = NIL_RTR0MEMOBJ;
382	/** Physical address of the IO bitmap. */
383	RTHCPHYS g_HCPhysIOBitmap = 0;
384	/** Pointer to the IO bitmap. */
385	R0PTRTYPE(void *) g_pvIOBitmap = NULL;
386
387	#ifdef VBOX_WITH_NESTED_HWVIRT
388	/** Ring-0 memory object for the nested-guest MSRPM bitmap. */
389	RTR0MEMOBJ g_hMemObjNstGstMsrBitmap = NIL_RTR0MEMOBJ;
390	/** Physical address of the nested-guest MSRPM bitmap. */
391	RTHCPHYS g_HCPhysNstGstMsrBitmap = 0;
392	/** Pointer to the nested-guest MSRPM bitmap. */
393	R0PTRTYPE(void *) g_pvNstGstMsrBitmap = NULL;
394	#endif
395
396
397	#ifdef VBOX_STRICT
398	# define HMSVM_LOG_CS RT_BIT_32(0)
399	# define HMSVM_LOG_SS RT_BIT_32(1)
400	# define HMSVM_LOG_FS RT_BIT_32(2)
401	# define HMSVM_LOG_GS RT_BIT_32(3)
402	# define HMSVM_LOG_LBR RT_BIT_32(4)
403	# define HMSVM_LOG_ALL ( HMSVM_LOG_CS \
404	\| HMSVM_LOG_SS \
405	\| HMSVM_LOG_FS \
406	\| HMSVM_LOG_GS \
407	\| HMSVM_LOG_LBR)
408
409	/**
410	* Dumps CPU state and additional info. to the logger for diagnostics.
411	*
412	* @param pVCpu The cross context virtual CPU structure.
413	* @param pVmcb Pointer to the VM control block.
414	* @param pCtx Pointer to the guest-CPU context.
415	* @param pszPrefix Log prefix.
416	* @param fFlags Log flags, see HMSVM_LOG_XXX.
417	* @param uVerbose The verbosity level, currently unused.
418	*/
419	static void hmR0SvmLogState(PVMCPU pVCpu, PCSVMVMCB pVmcb, PCPUMCTX pCtx, const char *pszPrefix, uint32_t fFlags,
420	uint8_t uVerbose)
421	{
422	RT_NOREF2(pVCpu, uVerbose);
423
424	Log4(("%s: cs:rip=%04x:%RX64 efl=%#RX32 cr0=%#RX32 cr3=%#RX32 cr4=%#RX32\n", pszPrefix, pCtx->cs.Sel, pCtx->rip,
425	pCtx->eflags.u, pCtx->cr0, pCtx->cr3, pCtx->cr4));
426	Log4(("%s: rsp=%#RX64 rbp=%#RX64 rdi=%#RX64\n", pszPrefix, pCtx->rsp, pCtx->rbp, pCtx->rdi));
427	if (fFlags & HMSVM_LOG_CS)
428	{
429	Log4(("%s: cs={%04x base=%016RX64 limit=%08x flags=%08x}\n", pszPrefix, pCtx->cs.Sel, pCtx->cs.u64Base,
430	pCtx->cs.u32Limit, pCtx->cs.Attr.u));
431	}
432	if (fFlags & HMSVM_LOG_SS)
433	{
434	Log4(("%s: ss={%04x base=%016RX64 limit=%08x flags=%08x}\n", pszPrefix, pCtx->ss.Sel, pCtx->ss.u64Base,
435	pCtx->ss.u32Limit, pCtx->ss.Attr.u));
436	}
437	if (fFlags & HMSVM_LOG_FS)
438	{
439	Log4(("%s: fs={%04x base=%016RX64 limit=%08x flags=%08x}\n", pszPrefix, pCtx->fs.Sel, pCtx->fs.u64Base,
440	pCtx->fs.u32Limit, pCtx->fs.Attr.u));
441	}
442	if (fFlags & HMSVM_LOG_GS)
443	{
444	Log4(("%s: gs={%04x base=%016RX64 limit=%08x flags=%08x}\n", pszPrefix, pCtx->gs.Sel, pCtx->gs.u64Base,
445	pCtx->gs.u32Limit, pCtx->gs.Attr.u));
446	}
447
448	PCSVMVMCBSTATESAVE pVmcbGuest = &pVmcb->guest;
449	if (fFlags & HMSVM_LOG_LBR)
450	{
451	Log4(("%s: br_from=%#RX64 br_to=%#RX64 lastxcpt_from=%#RX64 lastxcpt_to=%#RX64\n", pszPrefix, pVmcbGuest->u64BR_FROM,
452	pVmcbGuest->u64BR_TO, pVmcbGuest->u64LASTEXCPFROM, pVmcbGuest->u64LASTEXCPTO));
453	}
454	NOREF(pVmcbGuest);
455	}
456	#endif
457
458
459	/**
460	* Sets up and activates AMD-V on the current CPU.
461	*
462	* @returns VBox status code.
463	* @param pCpu Pointer to the CPU info struct.
464	* @param pVM The cross context VM structure. Can be
465	* NULL after a resume!
466	* @param pvCpuPage Pointer to the global CPU page.
467	* @param HCPhysCpuPage Physical address of the global CPU page.
468	* @param fEnabledByHost Whether the host OS has already initialized AMD-V.
469	* @param pvArg Unused on AMD-V.
470	*/
471	VMMR0DECL(int) SVMR0EnableCpu(PHMGLOBALCPUINFO pCpu, PVM pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost,
472	void *pvArg)
473	{
474	Assert(!fEnabledByHost);
475	Assert(HCPhysCpuPage && HCPhysCpuPage != NIL_RTHCPHYS);
476	Assert(RT_ALIGN_T(HCPhysCpuPage, _4K, RTHCPHYS) == HCPhysCpuPage);
477	Assert(pvCpuPage); NOREF(pvCpuPage);
478	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
479
480	NOREF(pvArg);
481	NOREF(fEnabledByHost);
482
483	/* Paranoid: Disable interrupt as, in theory, interrupt handlers might mess with EFER. */
484	RTCCUINTREG fEFlags = ASMIntDisableFlags();
485
486	/*
487	* We must turn on AMD-V and setup the host state physical address, as those MSRs are per CPU.
488	*/
489	uint64_t u64HostEfer = ASMRdMsr(MSR_K6_EFER);
490	if (u64HostEfer & MSR_K6_EFER_SVME)
491	{
492	/* If the VBOX_HWVIRTEX_IGNORE_SVM_IN_USE is active, then we blindly use AMD-V. */
493	if ( pVM
494	&& pVM->hm.s.svm.fIgnoreInUseError)
495	{
496	pCpu->fIgnoreAMDVInUseError = true;
497	}
498
499	if (!pCpu->fIgnoreAMDVInUseError)
500	{
501	ASMSetFlags(fEFlags);
502	return VERR_SVM_IN_USE;
503	}
504	}
505
506	/* Turn on AMD-V in the EFER MSR. */
507	ASMWrMsr(MSR_K6_EFER, u64HostEfer \| MSR_K6_EFER_SVME);
508
509	/* Write the physical page address where the CPU will store the host state while executing the VM. */
510	ASMWrMsr(MSR_K8_VM_HSAVE_PA, HCPhysCpuPage);
511
512	/* Restore interrupts. */
513	ASMSetFlags(fEFlags);
514
515	/*
516	* Theoretically, other hypervisors may have used ASIDs, ideally we should flush all non-zero ASIDs
517	* when enabling SVM. AMD doesn't have an SVM instruction to flush all ASIDs (flushing is done
518	* upon VMRUN). Therefore, flag that we need to flush the TLB entirely with before executing any
519	* guest code.
520	*/
521	pCpu->fFlushAsidBeforeUse = true;
522
523	/*
524	* Ensure each VCPU scheduled on this CPU gets a new ASID on resume. See @bugref{6255}.
525	*/
526	++pCpu->cTlbFlushes;
527
528	return VINF_SUCCESS;
529	}
530
531
532	/**
533	* Deactivates AMD-V on the current CPU.
534	*
535	* @returns VBox status code.
536	* @param pCpu Pointer to the CPU info struct.
537	* @param pvCpuPage Pointer to the global CPU page.
538	* @param HCPhysCpuPage Physical address of the global CPU page.
539	*/
540	VMMR0DECL(int) SVMR0DisableCpu(PHMGLOBALCPUINFO pCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
541	{
542	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
543	AssertReturn( HCPhysCpuPage
544	&& HCPhysCpuPage != NIL_RTHCPHYS, VERR_INVALID_PARAMETER);
545	AssertReturn(pvCpuPage, VERR_INVALID_PARAMETER);
546	NOREF(pCpu);
547
548	/* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with EFER. */
549	RTCCUINTREG fEFlags = ASMIntDisableFlags();
550
551	/* Turn off AMD-V in the EFER MSR. */
552	uint64_t u64HostEfer = ASMRdMsr(MSR_K6_EFER);
553	ASMWrMsr(MSR_K6_EFER, u64HostEfer & ~MSR_K6_EFER_SVME);
554
555	/* Invalidate host state physical address. */
556	ASMWrMsr(MSR_K8_VM_HSAVE_PA, 0);
557
558	/* Restore interrupts. */
559	ASMSetFlags(fEFlags);
560
561	return VINF_SUCCESS;
562	}
563
564
565	/**
566	* Does global AMD-V initialization (called during module initialization).
567	*
568	* @returns VBox status code.
569	*/
570	VMMR0DECL(int) SVMR0GlobalInit(void)
571	{
572	/*
573	* Allocate 12 KB for the IO bitmap. Since this is non-optional and we always intercept all IO accesses, it's done
574	* once globally here instead of per-VM.
575	*/
576	Assert(g_hMemObjIOBitmap == NIL_RTR0MEMOBJ);
577	int rc = RTR0MemObjAllocCont(&g_hMemObjIOBitmap, SVM_IOPM_PAGES << X86_PAGE_4K_SHIFT, false /* fExecutable */);
578	if (RT_FAILURE(rc))
579	return rc;
580
581	g_pvIOBitmap = RTR0MemObjAddress(g_hMemObjIOBitmap);
582	g_HCPhysIOBitmap = RTR0MemObjGetPagePhysAddr(g_hMemObjIOBitmap, 0 /* iPage */);
583
584	/* Set all bits to intercept all IO accesses. */
585	ASMMemFill32(g_pvIOBitmap, SVM_IOPM_PAGES << X86_PAGE_4K_SHIFT, UINT32_C(0xffffffff));
586
587	#ifdef VBOX_WITH_NESTED_HWVIRT
588	/*
589	* Allocate 8 KB for the MSR permission bitmap for the nested-guest.
590	*/
591	Assert(g_hMemObjNstGstMsrBitmap == NIL_RTR0MEMOBJ);
592	rc = RTR0MemObjAllocCont(&g_hMemObjNstGstMsrBitmap, SVM_MSRPM_PAGES << X86_PAGE_4K_SHIFT, false /* fExecutable */);
593	if (RT_FAILURE(rc))
594	return rc;
595
596	g_pvNstGstMsrBitmap = RTR0MemObjAddress(g_hMemObjNstGstMsrBitmap);
597	g_HCPhysNstGstMsrBitmap = RTR0MemObjGetPagePhysAddr(g_hMemObjNstGstMsrBitmap, 0 /* iPage */);
598
599	/* Set all bits to intercept all MSR accesses. */
600	ASMMemFill32(g_pvNstGstMsrBitmap, SVM_MSRPM_PAGES << X86_PAGE_4K_SHIFT, UINT32_C(0xffffffff));
601	#endif
602
603	return VINF_SUCCESS;
604	}
605
606
607	/**
608	* Does global AMD-V termination (called during module termination).
609	*/
610	VMMR0DECL(void) SVMR0GlobalTerm(void)
611	{
612	if (g_hMemObjIOBitmap != NIL_RTR0MEMOBJ)
613	{
614	RTR0MemObjFree(g_hMemObjIOBitmap, true /* fFreeMappings */);
615	g_pvIOBitmap = NULL;
616	g_HCPhysIOBitmap = 0;
617	g_hMemObjIOBitmap = NIL_RTR0MEMOBJ;
618	}
619
620	#ifdef VBOX_WITH_NESTED_HWVIRT
621	if (g_hMemObjNstGstMsrBitmap != NIL_RTR0MEMOBJ)
622	{
623	RTR0MemObjFree(g_hMemObjNstGstMsrBitmap, true /* fFreeMappings */);
624	g_pvNstGstMsrBitmap = NULL;
625	g_HCPhysNstGstMsrBitmap = 0;
626	g_hMemObjNstGstMsrBitmap = NIL_RTR0MEMOBJ;
627	}
628	#endif
629	}
630
631
632	/**
633	* Frees any allocated per-VCPU structures for a VM.
634	*
635	* @param pVM The cross context VM structure.
636	*/
637	DECLINLINE(void) hmR0SvmFreeStructs(PVM pVM)
638	{
639	for (uint32_t i = 0; i < pVM->cCpus; i++)
640	{
641	PVMCPU pVCpu = &pVM->aCpus[i];
642	AssertPtr(pVCpu);
643
644	if (pVCpu->hm.s.svm.hMemObjVmcbHost != NIL_RTR0MEMOBJ)
645	{
646	RTR0MemObjFree(pVCpu->hm.s.svm.hMemObjVmcbHost, false);
647	pVCpu->hm.s.svm.HCPhysVmcbHost = 0;
648	pVCpu->hm.s.svm.hMemObjVmcbHost = NIL_RTR0MEMOBJ;
649	}
650
651	if (pVCpu->hm.s.svm.hMemObjVmcb != NIL_RTR0MEMOBJ)
652	{
653	RTR0MemObjFree(pVCpu->hm.s.svm.hMemObjVmcb, false);
654	pVCpu->hm.s.svm.pVmcb = NULL;
655	pVCpu->hm.s.svm.HCPhysVmcb = 0;
656	pVCpu->hm.s.svm.hMemObjVmcb = NIL_RTR0MEMOBJ;
657	}
658
659	if (pVCpu->hm.s.svm.hMemObjMsrBitmap != NIL_RTR0MEMOBJ)
660	{
661	RTR0MemObjFree(pVCpu->hm.s.svm.hMemObjMsrBitmap, false);
662	pVCpu->hm.s.svm.pvMsrBitmap = NULL;
663	pVCpu->hm.s.svm.HCPhysMsrBitmap = 0;
664	pVCpu->hm.s.svm.hMemObjMsrBitmap = NIL_RTR0MEMOBJ;
665	}
666	}
667	}
668
669
670	/**
671	* Does per-VM AMD-V initialization.
672	*
673	* @returns VBox status code.
674	* @param pVM The cross context VM structure.
675	*/
676	VMMR0DECL(int) SVMR0InitVM(PVM pVM)
677	{
678	int rc = VERR_INTERNAL_ERROR_5;
679
680	/*
681	* Check for an AMD CPU erratum which requires us to flush the TLB before every world-switch.
682	*/
683	uint32_t u32Family;
684	uint32_t u32Model;
685	uint32_t u32Stepping;
686	if (HMAmdIsSubjectToErratum170(&u32Family, &u32Model, &u32Stepping))
687	{
688	Log4(("SVMR0InitVM: AMD cpu with erratum 170 family %#x model %#x stepping %#x\n", u32Family, u32Model, u32Stepping));
689	pVM->hm.s.svm.fAlwaysFlushTLB = true;
690	}
691
692	/*
693	* Initialize the R0 memory objects up-front so we can properly cleanup on allocation failures.
694	*/
695	for (VMCPUID i = 0; i < pVM->cCpus; i++)
696	{
697	PVMCPU pVCpu = &pVM->aCpus[i];
698	pVCpu->hm.s.svm.hMemObjVmcbHost = NIL_RTR0MEMOBJ;
699	pVCpu->hm.s.svm.hMemObjVmcb = NIL_RTR0MEMOBJ;
700	pVCpu->hm.s.svm.hMemObjMsrBitmap = NIL_RTR0MEMOBJ;
701	}
702
703	for (VMCPUID i = 0; i < pVM->cCpus; i++)
704	{
705	PVMCPU pVCpu = &pVM->aCpus[i];
706
707	/*
708	* Allocate one page for the host-context VM control block (VMCB). This is used for additional host-state (such as
709	* FS, GS, Kernel GS Base, etc.) apart from the host-state save area specified in MSR_K8_VM_HSAVE_PA.
710	*/
711	rc = RTR0MemObjAllocCont(&pVCpu->hm.s.svm.hMemObjVmcbHost, SVM_VMCB_PAGES << PAGE_SHIFT, false /* fExecutable */);
712	if (RT_FAILURE(rc))
713	goto failure_cleanup;
714
715	void *pvVmcbHost = RTR0MemObjAddress(pVCpu->hm.s.svm.hMemObjVmcbHost);
716	pVCpu->hm.s.svm.HCPhysVmcbHost = RTR0MemObjGetPagePhysAddr(pVCpu->hm.s.svm.hMemObjVmcbHost, 0 /* iPage */);
717	Assert(pVCpu->hm.s.svm.HCPhysVmcbHost < _4G);
718	ASMMemZeroPage(pvVmcbHost);
719
720	/*
721	* Allocate one page for the guest-state VMCB.
722	*/
723	rc = RTR0MemObjAllocCont(&pVCpu->hm.s.svm.hMemObjVmcb, SVM_VMCB_PAGES << PAGE_SHIFT, false /* fExecutable */);
724	if (RT_FAILURE(rc))
725	goto failure_cleanup;
726
727	pVCpu->hm.s.svm.pVmcb = (PSVMVMCB)RTR0MemObjAddress(pVCpu->hm.s.svm.hMemObjVmcb);
728	pVCpu->hm.s.svm.HCPhysVmcb = RTR0MemObjGetPagePhysAddr(pVCpu->hm.s.svm.hMemObjVmcb, 0 /* iPage */);
729	Assert(pVCpu->hm.s.svm.HCPhysVmcb < _4G);
730	ASMMemZeroPage(pVCpu->hm.s.svm.pVmcb);
731
732	/*
733	* Allocate two pages (8 KB) for the MSR permission bitmap. There doesn't seem to be a way to convince
734	* SVM to not require one.
735	*/
736	rc = RTR0MemObjAllocCont(&pVCpu->hm.s.svm.hMemObjMsrBitmap, SVM_MSRPM_PAGES << X86_PAGE_4K_SHIFT,
737	false /* fExecutable */);
738	if (RT_FAILURE(rc))
739	goto failure_cleanup;
740
741	pVCpu->hm.s.svm.pvMsrBitmap = RTR0MemObjAddress(pVCpu->hm.s.svm.hMemObjMsrBitmap);
742	pVCpu->hm.s.svm.HCPhysMsrBitmap = RTR0MemObjGetPagePhysAddr(pVCpu->hm.s.svm.hMemObjMsrBitmap, 0 /* iPage */);
743	/* Set all bits to intercept all MSR accesses (changed later on). */
744	ASMMemFill32(pVCpu->hm.s.svm.pvMsrBitmap, SVM_MSRPM_PAGES << X86_PAGE_4K_SHIFT, UINT32_C(0xffffffff));
745	}
746
747	return VINF_SUCCESS;
748
749	failure_cleanup:
750	hmR0SvmFreeStructs(pVM);
751	return rc;
752	}
753
754
755	/**
756	* Does per-VM AMD-V termination.
757	*
758	* @returns VBox status code.
759	* @param pVM The cross context VM structure.
760	*/
761	VMMR0DECL(int) SVMR0TermVM(PVM pVM)
762	{
763	hmR0SvmFreeStructs(pVM);
764	return VINF_SUCCESS;
765	}
766
767
768	/**
769	* Returns whether the VMCB Clean Bits feature is supported.
770	*
771	* @return @c true if supported, @c false otherwise.
772	* @param pVCpu The cross context virtual CPU structure.
773	* @param pCtx Pointer to the guest-CPU context.
774	*/
775	DECLINLINE(bool) hmR0SvmSupportsVmcbCleanBits(PVMCPU pVCpu, PCPUMCTX pCtx)
776	{
777	PVM pVM = pVCpu->CTX_SUFF(pVM);
778	#ifdef VBOX_WITH_NESTED_HWVIRT
779	if (CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
780	{
781	return (pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_VMCB_CLEAN)
782	&& pVM->cpum.ro.GuestFeatures.fSvmVmcbClean;
783	}
784	#else
785	RT_NOREF(pCtx);
786	#endif
787	return RT_BOOL(pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_VMCB_CLEAN);
788	}
789
790
791	/**
792	* Returns whether the decode assists feature is supported.
793	*
794	* @return @c true if supported, @c false otherwise.
795	* @param pVCpu The cross context virtual CPU structure.
796	* @param pCtx Pointer to the guest-CPU context.
797	*/
798	DECLINLINE(bool) hmR0SvmSupportsDecodeAssists(PVMCPU pVCpu, PCPUMCTX pCtx)
799	{
800	PVM pVM = pVCpu->CTX_SUFF(pVM);
801	#ifdef VBOX_WITH_NESTED_HWVIRT
802	if (CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
803	{
804	return (pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_DECODE_ASSISTS)
805	&& pVM->cpum.ro.GuestFeatures.fSvmDecodeAssists;
806	}
807	#else
808	RT_NOREF(pCtx);
809	#endif
810	return RT_BOOL(pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_DECODE_ASSISTS);
811	}
812
813
814	/**
815	* Returns whether the NRIP_SAVE feature is supported.
816	*
817	* @return @c true if supported, @c false otherwise.
818	* @param pVCpu The cross context virtual CPU structure.
819	* @param pCtx Pointer to the guest-CPU context.
820	*/
821	DECLINLINE(bool) hmR0SvmSupportsNextRipSave(PVMCPU pVCpu, PCPUMCTX pCtx)
822	{
823	PVM pVM = pVCpu->CTX_SUFF(pVM);
824	#ifdef VBOX_WITH_NESTED_HWVIRT
825	if (CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
826	{
827	return (pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_NRIP_SAVE)
828	&& pVM->cpum.ro.GuestFeatures.fSvmNextRipSave;
829	}
830	#else
831	RT_NOREF(pCtx);
832	#endif
833	return RT_BOOL(pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_NRIP_SAVE);
834	}
835
836
837	/**
838	* Sets the permission bits for the specified MSR in the MSRPM.
839	*
840	* @param pVmcb Pointer to the VM control block.
841	* @param pbMsrBitmap Pointer to the MSR bitmap.
842	* @param uMsr The MSR for which the access permissions are being set.
843	* @param enmRead MSR read permissions.
844	* @param enmWrite MSR write permissions.
845	*/
846	static void hmR0SvmSetMsrPermission(PSVMVMCB pVmcb, uint8_t *pbMsrBitmap, unsigned uMsr, SVMMSREXITREAD enmRead,
847	SVMMSREXITWRITE enmWrite)
848	{
849	uint16_t offMsrpm;
850	uint32_t uMsrpmBit;
851	int rc = HMSvmGetMsrpmOffsetAndBit(uMsr, &offMsrpm, &uMsrpmBit);
852	AssertRC(rc);
853
854	Assert(uMsrpmBit < 0x3fff);
855	Assert(offMsrpm < SVM_MSRPM_PAGES << X86_PAGE_4K_SHIFT);
856
857	pbMsrBitmap += offMsrpm;
858	if (enmRead == SVMMSREXIT_INTERCEPT_READ)
859	ASMBitSet(pbMsrBitmap, uMsrpmBit);
860	else
861	ASMBitClear(pbMsrBitmap, uMsrpmBit);
862
863	if (enmWrite == SVMMSREXIT_INTERCEPT_WRITE)
864	ASMBitSet(pbMsrBitmap, uMsrpmBit + 1);
865	else
866	ASMBitClear(pbMsrBitmap, uMsrpmBit + 1);
867
868	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_IOPM_MSRPM;
869	}
870
871
872	/**
873	* Sets up AMD-V for the specified VM.
874	* This function is only called once per-VM during initalization.
875	*
876	* @returns VBox status code.
877	* @param pVM The cross context VM structure.
878	*/
879	VMMR0DECL(int) SVMR0SetupVM(PVM pVM)
880	{
881	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
882	AssertReturn(pVM, VERR_INVALID_PARAMETER);
883	Assert(pVM->hm.s.svm.fSupported);
884
885	bool const fPauseFilter = RT_BOOL(pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_PAUSE_FILTER);
886	bool const fPauseFilterThreshold = RT_BOOL(pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_PAUSE_FILTER_THRESHOLD);
887	bool const fUsePauseFilter = fPauseFilter && pVM->hm.s.svm.cPauseFilter && pVM->hm.s.svm.cPauseFilterThresholdTicks;
888
889	bool const fLbrVirt = RT_BOOL(pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_LBR_VIRT);
890	bool const fUseLbrVirt = fLbrVirt; /** @todo CFGM, IEM implementation etc. */
891
892	#ifdef VBOX_WITH_NESTED_HWVIRT
893	bool const fVirtVmsaveVmload = RT_BOOL(pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_VIRT_VMSAVE_VMLOAD);
894	bool const fUseVirtVmsaveVmload = fVirtVmsaveVmload && pVM->hm.s.svm.fVirtVmsaveVmload && pVM->hm.s.fNestedPaging;
895
896	bool const fVGif = RT_BOOL(pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_VGIF);
897	bool const fUseVGif = fVGif && pVM->hm.s.svm.fVGif;
898	#endif
899
900	for (VMCPUID i = 0; i < pVM->cCpus; i++)
901	{
902	PVMCPU pVCpu = &pVM->aCpus[i];
903	PSVMVMCB pVmcb = pVM->aCpus[i].hm.s.svm.pVmcb;
904
905	AssertMsgReturn(pVmcb, ("Invalid pVmcb for vcpu[%u]\n", i), VERR_SVM_INVALID_PVMCB);
906
907	/* Initialize the #VMEXIT history array with end-of-array markers (UINT16_MAX). */
908	Assert(!pVCpu->hm.s.idxExitHistoryFree);
909	HMCPU_EXIT_HISTORY_RESET(pVCpu);
910
911	/* Always trap #AC for reasons of security. */
912	pVmcb->ctrl.u32InterceptXcpt \|= RT_BIT_32(X86_XCPT_AC);
913
914	/* Always trap #DB for reasons of security. */
915	pVmcb->ctrl.u32InterceptXcpt \|= RT_BIT_32(X86_XCPT_DB);
916
917	/* Trap exceptions unconditionally (debug purposes). */
918	#ifdef HMSVM_ALWAYS_TRAP_PF
919	pVmcb->ctrl.u32InterceptXcpt \|= RT_BIT(X86_XCPT_PF);
920	#endif
921	#ifdef HMSVM_ALWAYS_TRAP_ALL_XCPTS
922	/* If you add any exceptions here, make sure to update hmR0SvmHandleExit(). */
923	pVmcb->ctrl.u32InterceptXcpt \|= 0
924	\| RT_BIT(X86_XCPT_BP)
925	\| RT_BIT(X86_XCPT_DE)
926	\| RT_BIT(X86_XCPT_NM)
927	\| RT_BIT(X86_XCPT_UD)
928	\| RT_BIT(X86_XCPT_NP)
929	\| RT_BIT(X86_XCPT_SS)
930	\| RT_BIT(X86_XCPT_GP)
931	\| RT_BIT(X86_XCPT_PF)
932	\| RT_BIT(X86_XCPT_MF)
933	;
934	#endif
935
936	/* Set up unconditional intercepts and conditions. */
937	pVmcb->ctrl.u64InterceptCtrl = HMSVM_MANDATORY_GUEST_CTRL_INTERCEPTS;
938
939	/* CR0, CR4 reads must be intercepted, our shadow values are not necessarily the same as the guest's. */
940	pVmcb->ctrl.u16InterceptRdCRx = RT_BIT(0) \| RT_BIT(4);
941
942	/* CR0, CR4 writes must be intercepted for the same reasons as above. */
943	pVmcb->ctrl.u16InterceptWrCRx = RT_BIT(0) \| RT_BIT(4);
944
945	/* Intercept all DRx reads and writes by default. Changed later on. */
946	pVmcb->ctrl.u16InterceptRdDRx = 0xffff;
947	pVmcb->ctrl.u16InterceptWrDRx = 0xffff;
948
949	/* Virtualize masking of INTR interrupts. (reads/writes from/to CR8 go to the V_TPR register) */
950	pVmcb->ctrl.IntCtrl.n.u1VIntrMasking = 1;
951
952	/* Ignore the priority in the virtual TPR. This is necessary for delivering PIC style (ExtInt) interrupts
953	and we currently deliver both PIC and APIC interrupts alike. See hmR0SvmInjectPendingEvent() */
954	pVmcb->ctrl.IntCtrl.n.u1IgnoreTPR = 1;
955
956	/* Set IO and MSR bitmap permission bitmap physical addresses. */
957	pVmcb->ctrl.u64IOPMPhysAddr = g_HCPhysIOBitmap;
958	pVmcb->ctrl.u64MSRPMPhysAddr = pVCpu->hm.s.svm.HCPhysMsrBitmap;
959
960	/* LBR virtualization. */
961	if (fUseLbrVirt)
962	{
963	pVmcb->ctrl.LbrVirt.n.u1LbrVirt = fUseLbrVirt;
964	pVmcb->guest.u64DBGCTL = MSR_IA32_DEBUGCTL_LBR;
965	}
966	else
967	Assert(pVmcb->ctrl.LbrVirt.n.u1LbrVirt == 0);
968
969	#ifdef VBOX_WITH_NESTED_HWVIRT
970	/* Virtualized VMSAVE/VMLOAD. */
971	pVmcb->ctrl.LbrVirt.n.u1VirtVmsaveVmload = fUseVirtVmsaveVmload;
972	if (!fUseVirtVmsaveVmload)
973	{
974	pVmcb->ctrl.u64InterceptCtrl \|= SVM_CTRL_INTERCEPT_VMSAVE
975	\| SVM_CTRL_INTERCEPT_VMLOAD;
976	}
977
978	/* Virtual GIF. */
979	pVmcb->ctrl.IntCtrl.n.u1VGifEnable = fUseVGif;
980	if (!fUseVGif)
981	{
982	pVmcb->ctrl.u64InterceptCtrl \|= SVM_CTRL_INTERCEPT_CLGI
983	\| SVM_CTRL_INTERCEPT_STGI;
984	}
985	#endif
986
987	/* Initially all VMCB clean bits MBZ indicating that everything should be loaded from the VMCB in memory. */
988	Assert(pVmcb->ctrl.u32VmcbCleanBits == 0);
989
990	/* The host ASID MBZ, for the guest start with 1. */
991	pVmcb->ctrl.TLBCtrl.n.u32ASID = 1;
992
993	/*
994	* Setup the PAT MSR (applicable for Nested Paging only).
995	* The default value should be 0x0007040600070406ULL, but we want to treat all guest memory as WB,
996	* so choose type 6 for all PAT slots.
997	*/
998	pVmcb->guest.u64GPAT = UINT64_C(0x0006060606060606);
999
1000	/* Setup Nested Paging. This doesn't change throughout the execution time of the VM. */
1001	pVmcb->ctrl.NestedPaging.n.u1NestedPaging = pVM->hm.s.fNestedPaging;
1002
1003	/* Without Nested Paging, we need additionally intercepts. */
1004	if (!pVM->hm.s.fNestedPaging)
1005	{
1006	/* CR3 reads/writes must be intercepted; our shadow values differ from the guest values. */
1007	pVmcb->ctrl.u16InterceptRdCRx \|= RT_BIT(3);
1008	pVmcb->ctrl.u16InterceptWrCRx \|= RT_BIT(3);
1009
1010	/* Intercept INVLPG and task switches (may change CR3, EFLAGS, LDT). */
1011	pVmcb->ctrl.u64InterceptCtrl \|= SVM_CTRL_INTERCEPT_INVLPG
1012	\| SVM_CTRL_INTERCEPT_TASK_SWITCH;
1013
1014	/* Page faults must be intercepted to implement shadow paging. */
1015	pVmcb->ctrl.u32InterceptXcpt \|= RT_BIT(X86_XCPT_PF);
1016	}
1017
1018	#ifdef HMSVM_ALWAYS_TRAP_TASK_SWITCH
1019	pVmcb->ctrl.u64InterceptCtrl \|= SVM_CTRL_INTERCEPT_TASK_SWITCH;
1020	#endif
1021
1022	/* Apply the exceptions intercepts needed by the GIM provider. */
1023	if (pVCpu->hm.s.fGIMTrapXcptUD)
1024	pVmcb->ctrl.u32InterceptXcpt \|= RT_BIT(X86_XCPT_UD);
1025
1026	/* Setup Pause Filter for guest pause-loop (spinlock) exiting. */
1027	if (fUsePauseFilter)
1028	{
1029	pVmcb->ctrl.u16PauseFilterCount = pVM->hm.s.svm.cPauseFilter;
1030	if (fPauseFilterThreshold)
1031	pVmcb->ctrl.u16PauseFilterThreshold = pVM->hm.s.svm.cPauseFilterThresholdTicks;
1032	}
1033
1034	/*
1035	* The following MSRs are saved/restored automatically during the world-switch.
1036	* Don't intercept guest read/write accesses to these MSRs.
1037	*/
1038	uint8_t pbMsrBitmap = (uint8_t )pVCpu->hm.s.svm.pvMsrBitmap;
1039	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_K8_LSTAR, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1040	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_K8_CSTAR, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1041	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_K6_STAR, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1042	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_K8_SF_MASK, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1043	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_K8_FS_BASE, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1044	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_K8_GS_BASE, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1045	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_K8_KERNEL_GS_BASE, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1046	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_IA32_SYSENTER_CS, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1047	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_IA32_SYSENTER_ESP, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1048	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_IA32_SYSENTER_EIP, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1049	}
1050
1051	return VINF_SUCCESS;
1052	}
1053
1054
1055	/**
1056	* Gets a pointer to the currently active guest or nested-guest VMCB.
1057	*
1058	* @returns Pointer to the current context VMCB.
1059	* @param pVCpu The cross context virtual CPU structure.
1060	* @param pCtx Pointer to the guest-CPU context.
1061	*/
1062	DECLINLINE(PSVMVMCB) hmR0SvmGetCurrentVmcb(PVMCPU pVCpu, PCPUMCTX pCtx)
1063	{
1064	#ifdef VBOX_WITH_NESTED_HWVIRT
1065	if (CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
1066	return pCtx->hwvirt.svm.CTX_SUFF(pVmcb);
1067	#else
1068	RT_NOREF(pCtx);
1069	#endif
1070	return pVCpu->hm.s.svm.pVmcb;
1071	}
1072
1073
1074	/**
1075	* Invalidates a guest page by guest virtual address.
1076	*
1077	* @returns VBox status code.
1078	* @param pVM The cross context VM structure.
1079	* @param pVCpu The cross context virtual CPU structure.
1080	* @param GCVirt Guest virtual address of the page to invalidate.
1081	*/
1082	VMMR0DECL(int) SVMR0InvalidatePage(PVM pVM, PVMCPU pVCpu, RTGCPTR GCVirt)
1083	{
1084	AssertReturn(pVM, VERR_INVALID_PARAMETER);
1085	Assert(pVM->hm.s.svm.fSupported);
1086
1087	bool fFlushPending = pVM->hm.s.svm.fAlwaysFlushTLB \|\| VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_FLUSH);
1088
1089	/* Skip it if a TLB flush is already pending. */
1090	if (!fFlushPending)
1091	{
1092	Log4(("SVMR0InvalidatePage %RGv\n", GCVirt));
1093
1094	PCPUMCTX pCtx = CPUMQueryGuestCtxPtr(pVCpu);
1095	PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
1096	AssertMsgReturn(pVmcb, ("Invalid pVmcb!\n"), VERR_SVM_INVALID_PVMCB);
1097
1098	#if HC_ARCH_BITS == 32
1099	/* If we get a flush in 64-bit guest mode, then force a full TLB flush. INVLPGA takes only 32-bit addresses. */
1100	if (CPUMIsGuestInLongMode(pVCpu))
1101	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1102	else
1103	#endif
1104	{
1105	SVMR0InvlpgA(GCVirt, pVmcb->ctrl.TLBCtrl.n.u32ASID);
1106	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt);
1107	}
1108	}
1109	return VINF_SUCCESS;
1110	}
1111
1112
1113	/**
1114	* Flushes the appropriate tagged-TLB entries.
1115	*
1116	* @param pVCpu The cross context virtual CPU structure.
1117	* @param pCtx Pointer to the guest-CPU or nested-guest-CPU context.
1118	* @param pVmcb Pointer to the VM control block.
1119	*/
1120	static void hmR0SvmFlushTaggedTlb(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMVMCB pVmcb)
1121	{
1122	#ifndef VBOX_WITH_NESTED_HWVIRT
1123	RT_NOREF(pCtx);
1124	#endif
1125
1126	PVM pVM = pVCpu->CTX_SUFF(pVM);
1127	PHMGLOBALCPUINFO pCpu = hmR0GetCurrentCpu();
1128
1129	/*
1130	* Force a TLB flush for the first world switch if the current CPU differs from the one we ran on last.
1131	* This can happen both for start & resume due to long jumps back to ring-3.
1132	*
1133	* We also force a TLB flush every time when executing a nested-guest VCPU as there is no correlation
1134	* between it and the physical CPU.
1135	*
1136	* If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB,
1137	* so we cannot reuse the ASIDs without flushing.
1138	*/
1139	bool fNewAsid = false;
1140	Assert(pCpu->idCpu != NIL_RTCPUID);
1141	if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
1142	\|\| pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes
1143	#ifdef VBOX_WITH_NESTED_HWVIRT
1144	\|\| CPUMIsGuestInSvmNestedHwVirtMode(pCtx)
1145	#endif
1146	)
1147	{
1148	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
1149	pVCpu->hm.s.fForceTLBFlush = true;
1150	fNewAsid = true;
1151	}
1152
1153	/* Set TLB flush state as checked until we return from the world switch. */
1154	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true);
1155
1156	/* Check for explicit TLB flushes. */
1157	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
1158	{
1159	pVCpu->hm.s.fForceTLBFlush = true;
1160	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
1161	}
1162
1163	/*
1164	* If the AMD CPU erratum 170, We need to flush the entire TLB for each world switch. Sad.
1165	* This Host CPU requirement takes precedence.
1166	*/
1167	if (pVM->hm.s.svm.fAlwaysFlushTLB)
1168	{
1169	pCpu->uCurrentAsid = 1;
1170	pVCpu->hm.s.uCurrentAsid = 1;
1171	pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
1172	pVmcb->ctrl.TLBCtrl.n.u8TLBFlush = SVM_TLB_FLUSH_ENTIRE;
1173
1174	/* Clear the VMCB Clean Bit for NP while flushing the TLB. See @bugref{7152}. */
1175	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_NP;
1176
1177	/* Keep track of last CPU ID even when flushing all the time. */
1178	if (fNewAsid)
1179	pVCpu->hm.s.idLastCpu = pCpu->idCpu;
1180	}
1181	else
1182	{
1183	pVmcb->ctrl.TLBCtrl.n.u8TLBFlush = SVM_TLB_FLUSH_NOTHING;
1184	if (pVCpu->hm.s.fForceTLBFlush)
1185	{
1186	/* Clear the VMCB Clean Bit for NP while flushing the TLB. See @bugref{7152}. */
1187	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_NP;
1188
1189	if (fNewAsid)
1190	{
1191	++pCpu->uCurrentAsid;
1192
1193	bool fHitASIDLimit = false;
1194	if (pCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
1195	{
1196	pCpu->uCurrentAsid = 1; /* Wraparound at 1; host uses 0 */
1197	pCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new ASID. */
1198	fHitASIDLimit = true;
1199	}
1200
1201	if ( fHitASIDLimit
1202	\|\| pCpu->fFlushAsidBeforeUse)
1203	{
1204	pVmcb->ctrl.TLBCtrl.n.u8TLBFlush = SVM_TLB_FLUSH_ENTIRE;
1205	pCpu->fFlushAsidBeforeUse = false;
1206	}
1207
1208	pVCpu->hm.s.uCurrentAsid = pCpu->uCurrentAsid;
1209	pVCpu->hm.s.idLastCpu = pCpu->idCpu;
1210	pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
1211	}
1212	else
1213	{
1214	if (pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_FLUSH_BY_ASID)
1215	pVmcb->ctrl.TLBCtrl.n.u8TLBFlush = SVM_TLB_FLUSH_SINGLE_CONTEXT;
1216	else
1217	pVmcb->ctrl.TLBCtrl.n.u8TLBFlush = SVM_TLB_FLUSH_ENTIRE;
1218	}
1219
1220	pVCpu->hm.s.fForceTLBFlush = false;
1221	}
1222	}
1223
1224	/* Update VMCB with the ASID. */
1225	if (pVmcb->ctrl.TLBCtrl.n.u32ASID != pVCpu->hm.s.uCurrentAsid)
1226	{
1227	pVmcb->ctrl.TLBCtrl.n.u32ASID = pVCpu->hm.s.uCurrentAsid;
1228	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_ASID;
1229	}
1230
1231	AssertMsg(pVCpu->hm.s.idLastCpu == pCpu->idCpu,
1232	("vcpu idLastCpu=%u pcpu idCpu=%u\n", pVCpu->hm.s.idLastCpu, pCpu->idCpu));
1233	AssertMsg(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes,
1234	("Flush count mismatch for cpu %u (%u vs %u)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes));
1235	AssertMsg(pCpu->uCurrentAsid >= 1 && pCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
1236	("cpu%d uCurrentAsid = %x\n", pCpu->idCpu, pCpu->uCurrentAsid));
1237	AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
1238	("cpu%d VM uCurrentAsid = %x\n", pCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
1239
1240	#ifdef VBOX_WITH_STATISTICS
1241	if (pVmcb->ctrl.TLBCtrl.n.u8TLBFlush == SVM_TLB_FLUSH_NOTHING)
1242	STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch);
1243	else if ( pVmcb->ctrl.TLBCtrl.n.u8TLBFlush == SVM_TLB_FLUSH_SINGLE_CONTEXT
1244	\|\| pVmcb->ctrl.TLBCtrl.n.u8TLBFlush == SVM_TLB_FLUSH_SINGLE_CONTEXT_RETAIN_GLOBALS)
1245	{
1246	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid);
1247	}
1248	else
1249	{
1250	Assert(pVmcb->ctrl.TLBCtrl.n.u8TLBFlush == SVM_TLB_FLUSH_ENTIRE);
1251	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushEntire);
1252	}
1253	#endif
1254	}
1255
1256
1257	/** @name 64-bit guest on 32-bit host OS helper functions.
1258	*
1259	* The host CPU is still 64-bit capable but the host OS is running in 32-bit
1260	* mode (code segment, paging). These wrappers/helpers perform the necessary
1261	* bits for the 32->64 switcher.
1262	*
1263	* @{ */
1264	#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
1265	/**
1266	* Prepares for and executes VMRUN (64-bit guests on a 32-bit host).
1267	*
1268	* @returns VBox status code.
1269	* @param HCPhysVmcbHost Physical address of host VMCB.
1270	* @param HCPhysVmcb Physical address of the VMCB.
1271	* @param pCtx Pointer to the guest-CPU context.
1272	* @param pVM The cross context VM structure.
1273	* @param pVCpu The cross context virtual CPU structure.
1274	*/
1275	DECLASM(int) SVMR0VMSwitcherRun64(RTHCPHYS HCPhysVmcbHost, RTHCPHYS HCPhysVmcb, PCPUMCTX pCtx, PVM pVM, PVMCPU pVCpu)
1276	{
1277	uint32_t aParam[8];
1278	aParam[0] = RT_LO_U32(HCPhysVmcbHost); /* Param 1: HCPhysVmcbHost - Lo. */
1279	aParam[1] = RT_HI_U32(HCPhysVmcbHost); /* Param 1: HCPhysVmcbHost - Hi. */
1280	aParam[2] = RT_LO_U32(HCPhysVmcb); /* Param 2: HCPhysVmcb - Lo. */
1281	aParam[3] = RT_HI_U32(HCPhysVmcb); /* Param 2: HCPhysVmcb - Hi. */
1282	aParam[4] = VM_RC_ADDR(pVM, pVM);
1283	aParam[5] = 0;
1284	aParam[6] = VM_RC_ADDR(pVM, pVCpu);
1285	aParam[7] = 0;
1286
1287	return SVMR0Execute64BitsHandler(pVM, pVCpu, pCtx, HM64ON32OP_SVMRCVMRun64, RT_ELEMENTS(aParam), &aParam[0]);
1288	}
1289
1290
1291	/**
1292	* Executes the specified VMRUN handler in 64-bit mode.
1293	*
1294	* @returns VBox status code.
1295	* @param pVM The cross context VM structure.
1296	* @param pVCpu The cross context virtual CPU structure.
1297	* @param pCtx Pointer to the guest-CPU context.
1298	* @param enmOp The operation to perform.
1299	* @param cParams Number of parameters.
1300	* @param paParam Array of 32-bit parameters.
1301	*/
1302	VMMR0DECL(int) SVMR0Execute64BitsHandler(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, HM64ON32OP enmOp,
1303	uint32_t cParams, uint32_t *paParam)
1304	{
1305	AssertReturn(pVM->hm.s.pfnHost32ToGuest64R0, VERR_HM_NO_32_TO_64_SWITCHER);
1306	Assert(enmOp > HM64ON32OP_INVALID && enmOp < HM64ON32OP_END);
1307
1308	NOREF(pCtx);
1309
1310	/* Disable interrupts. */
1311	RTHCUINTREG uOldEFlags = ASMIntDisableFlags();
1312
1313	#ifdef VBOX_WITH_VMMR0_DISABLE_LAPIC_NMI
1314	RTCPUID idHostCpu = RTMpCpuId();
1315	CPUMR0SetLApic(pVCpu, idHostCpu);
1316	#endif
1317
1318	CPUMSetHyperESP(pVCpu, VMMGetStackRC(pVCpu));
1319	CPUMSetHyperEIP(pVCpu, enmOp);
1320	for (int i = (int)cParams - 1; i >= 0; i--)
1321	CPUMPushHyper(pVCpu, paParam[i]);
1322
1323	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatWorldSwitch3264, z);
1324	/* Call the switcher. */
1325	int rc = pVM->hm.s.pfnHost32ToGuest64R0(pVM, RT_OFFSETOF(VM, aCpus[pVCpu->idCpu].cpum) - RT_OFFSETOF(VM, cpum));
1326	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatWorldSwitch3264, z);
1327
1328	/* Restore interrupts. */
1329	ASMSetFlags(uOldEFlags);
1330	return rc;
1331	}
1332
1333	#endif /* HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) */
1334	/** @} */
1335
1336
1337	/**
1338	* Adds an exception to the intercept exception bitmap in the VMCB and updates
1339	* the corresponding VMCB Clean bit.
1340	*
1341	* @param pVmcb Pointer to the VM control block.
1342	* @param u32Xcpt The value of the exception (X86_XCPT_*).
1343	*/
1344	DECLINLINE(void) hmR0SvmAddXcptIntercept(PSVMVMCB pVmcb, uint32_t u32Xcpt)
1345	{
1346	if (!(pVmcb->ctrl.u32InterceptXcpt & RT_BIT(u32Xcpt)))
1347	{
1348	pVmcb->ctrl.u32InterceptXcpt \|= RT_BIT(u32Xcpt);
1349	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
1350	}
1351	}
1352
1353
1354	/**
1355	* Removes an exception from the intercept-exception bitmap in the VMCB and
1356	* updates the corresponding VMCB Clean bit.
1357	*
1358	* @param pVCpu The cross context virtual CPU structure.
1359	* @param pCtx Pointer to the guest-CPU context.
1360	* @param pVmcb Pointer to the VM control block.
1361	* @param u32Xcpt The value of the exception (X86_XCPT_*).
1362	*
1363	* @remarks This takes into account if we're executing a nested-guest and only
1364	* removes the exception intercept if both the guest -and- nested-guest
1365	* are not intercepting it.
1366	*/
1367	DECLINLINE(void) hmR0SvmRemoveXcptIntercept(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMVMCB pVmcb, uint32_t u32Xcpt)
1368	{
1369	Assert(u32Xcpt != X86_XCPT_DB);
1370	Assert(u32Xcpt != X86_XCPT_AC);
1371	#ifndef HMSVM_ALWAYS_TRAP_ALL_XCPTS
1372	if (pVmcb->ctrl.u32InterceptXcpt & RT_BIT(u32Xcpt))
1373	{
1374	bool fRemoveXcpt = true;
1375	#ifdef VBOX_WITH_NESTED_HWVIRT
1376	/* Only remove the intercept if the nested-guest is also not intercepting it! */
1377	if (CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
1378	{
1379	Assert(pCtx->hwvirt.svm.fHMCachedVmcb); NOREF(pCtx);
1380	PCSVMNESTEDVMCBCACHE pVmcbNstGstCache = &pVCpu->hm.s.svm.NstGstVmcbCache;
1381	fRemoveXcpt = !(pVmcbNstGstCache->u32InterceptXcpt & RT_BIT(u32Xcpt));
1382	}
1383	#else
1384	RT_NOREF2(pVCpu, pCtx);
1385	#endif
1386	if (fRemoveXcpt)
1387	{
1388	pVmcb->ctrl.u32InterceptXcpt &= ~RT_BIT(u32Xcpt);
1389	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
1390	}
1391	}
1392	#else
1393	RT_NOREF3(pVCpu, pCtx, pVmcb);
1394	#endif
1395	}
1396
1397
1398	/**
1399	* Loads the guest (or nested-guest) CR0 control register into the guest-state
1400	* area in the VMCB.
1401	*
1402	* Although the guest CR0 is a separate field in the VMCB we have to consider
1403	* the FPU state itself which is shared between the host and the guest.
1404	*
1405	* @returns VBox status code.
1406	* @param pVCpu The cross context virtual CPU structure.
1407	* @param pVmcb Pointer to the VM control block.
1408	* @param pCtx Pointer to the guest-CPU context.
1409	*
1410	* @remarks No-long-jump zone!!!
1411	*/
1412	static void hmR0SvmLoadSharedCR0(PVMCPU pVCpu, PSVMVMCB pVmcb, PCPUMCTX pCtx)
1413	{
1414	uint64_t u64GuestCR0 = pCtx->cr0;
1415
1416	/* Always enable caching. */
1417	u64GuestCR0 &= ~(X86_CR0_CD \| X86_CR0_NW);
1418
1419	/*
1420	* When Nested Paging is not available use shadow page tables and intercept #PFs (the latter done in SVMR0SetupVM()).
1421	*/
1422	if (!pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging)
1423	{
1424	u64GuestCR0 \|= X86_CR0_PG; /* When Nested Paging is not available, use shadow page tables. */
1425	u64GuestCR0 \|= X86_CR0_WP; /* Guest CPL 0 writes to its read-only pages should cause a #PF #VMEXIT. */
1426	}
1427
1428	/*
1429	* Guest FPU bits.
1430	*/
1431	bool fInterceptNM = false;
1432	bool fInterceptMF = false;
1433	u64GuestCR0 \|= X86_CR0_NE; /* Use internal x87 FPU exceptions handling rather than external interrupts. */
1434	if (CPUMIsGuestFPUStateActive(pVCpu))
1435	{
1436	/* Catch floating point exceptions if we need to report them to the guest in a different way. */
1437	if (!(pCtx->cr0 & X86_CR0_NE))
1438	{
1439	Log4(("hmR0SvmLoadSharedCR0: Intercepting Guest CR0.MP Old-style FPU handling!!!\n"));
1440	fInterceptMF = true;
1441	}
1442	}
1443	else
1444	{
1445	fInterceptNM = true; /* Guest FPU inactive, #VMEXIT on #NM for lazy FPU loading. */
1446	u64GuestCR0 \|= X86_CR0_TS /* Guest can task switch quickly and do lazy FPU syncing. */
1447	\| X86_CR0_MP; /* FWAIT/WAIT should not ignore CR0.TS and should generate #NM. */
1448	}
1449
1450	/*
1451	* Update the exception intercept bitmap.
1452	*/
1453	if (fInterceptNM)
1454	hmR0SvmAddXcptIntercept(pVmcb, X86_XCPT_NM);
1455	else
1456	hmR0SvmRemoveXcptIntercept(pVCpu, pCtx, pVmcb, X86_XCPT_NM);
1457
1458	if (fInterceptMF)
1459	hmR0SvmAddXcptIntercept(pVmcb, X86_XCPT_MF);
1460	else
1461	hmR0SvmRemoveXcptIntercept(pVCpu, pCtx, pVmcb, X86_XCPT_MF);
1462
1463	pVmcb->guest.u64CR0 = u64GuestCR0;
1464	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_CRX_EFER;
1465	}
1466
1467
1468	/**
1469	* Loads the guest/nested-guest control registers (CR2, CR3, CR4) into the VMCB.
1470	*
1471	* @returns VBox status code.
1472	* @param pVCpu The cross context virtual CPU structure.
1473	* @param pVmcb Pointer to the VM control block.
1474	* @param pCtx Pointer to the guest-CPU context.
1475	*
1476	* @remarks No-long-jump zone!!!
1477	*/
1478	static int hmR0SvmLoadGuestControlRegs(PVMCPU pVCpu, PSVMVMCB pVmcb, PCPUMCTX pCtx)
1479	{
1480	PVM pVM = pVCpu->CTX_SUFF(pVM);
1481
1482	/*
1483	* Guest CR2.
1484	*/
1485	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR2))
1486	{
1487	pVmcb->guest.u64CR2 = pCtx->cr2;
1488	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_CR2;
1489	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR2);
1490	}
1491
1492	/*
1493	* Guest CR3.
1494	*/
1495	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR3))
1496	{
1497	if (pVM->hm.s.fNestedPaging)
1498	{
1499	PGMMODE enmShwPagingMode;
1500	#if HC_ARCH_BITS == 32
1501	if (CPUMIsGuestInLongModeEx(pCtx))
1502	enmShwPagingMode = PGMMODE_AMD64_NX;
1503	else
1504	#endif
1505	enmShwPagingMode = PGMGetHostMode(pVM);
1506
1507	pVmcb->ctrl.u64NestedPagingCR3 = PGMGetNestedCR3(pVCpu, enmShwPagingMode);
1508	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_NP;
1509	Assert(pVmcb->ctrl.u64NestedPagingCR3);
1510	pVmcb->guest.u64CR3 = pCtx->cr3;
1511	}
1512	else
1513	{
1514	pVmcb->guest.u64CR3 = PGMGetHyperCR3(pVCpu);
1515	Log4(("hmR0SvmLoadGuestControlRegs: CR3=%#RX64 (HyperCR3=%#RX64)\n", pCtx->cr3, pVmcb->guest.u64CR3));
1516	}
1517
1518	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_CRX_EFER;
1519	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR3);
1520	}
1521
1522	/*
1523	* Guest CR4.
1524	* ASSUMES this is done everytime we get in from ring-3! (XCR0)
1525	*/
1526	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR4))
1527	{
1528	uint64_t u64GuestCR4 = pCtx->cr4;
1529	Assert(RT_HI_U32(u64GuestCR4) == 0);
1530	if (!pVM->hm.s.fNestedPaging)
1531	{
1532	switch (pVCpu->hm.s.enmShadowMode)
1533	{
1534	case PGMMODE_REAL:
1535	case PGMMODE_PROTECTED: /* Protected mode, no paging. */
1536	AssertFailed();
1537	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
1538
1539	case PGMMODE_32_BIT: /* 32-bit paging. */
1540	u64GuestCR4 &= ~X86_CR4_PAE;
1541	break;
1542
1543	case PGMMODE_PAE: /* PAE paging. */
1544	case PGMMODE_PAE_NX: /* PAE paging with NX enabled. */
1545	/** Must use PAE paging as we could use physical memory > 4 GB */
1546	u64GuestCR4 \|= X86_CR4_PAE;
1547	break;
1548
1549	case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
1550	case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
1551	#ifdef VBOX_ENABLE_64_BITS_GUESTS
1552	break;
1553	#else
1554	AssertFailed();
1555	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
1556	#endif
1557
1558	default: /* shut up gcc */
1559	AssertFailed();
1560	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
1561	}
1562	}
1563
1564	pVmcb->guest.u64CR4 = u64GuestCR4;
1565	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_CRX_EFER;
1566
1567	/* Whether to save/load/restore XCR0 during world switch depends on CR4.OSXSAVE and host+guest XCR0. */
1568	pVCpu->hm.s.fLoadSaveGuestXcr0 = (u64GuestCR4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
1569
1570	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR4);
1571	}
1572
1573	return VINF_SUCCESS;
1574	}
1575
1576
1577	/**
1578	* Loads the guest (or nested-guest) segment registers into the VMCB.
1579	*
1580	* @returns VBox status code.
1581	* @param pVCpu The cross context virtual CPU structure.
1582	* @param pVmcb Pointer to the VM control block.
1583	* @param pCtx Pointer to the guest-CPU context.
1584	*
1585	* @remarks No-long-jump zone!!!
1586	*/
1587	static void hmR0SvmLoadGuestSegmentRegs(PVMCPU pVCpu, PSVMVMCB pVmcb, PCPUMCTX pCtx)
1588	{
1589	/* Guest Segment registers: CS, SS, DS, ES, FS, GS. */
1590	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS))
1591	{
1592	HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, CS, cs);
1593	HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, SS, ss);
1594	HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, DS, ds);
1595	HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, ES, es);
1596	HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, FS, fs);
1597	HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, GS, gs);
1598
1599	pVmcb->guest.u8CPL = pCtx->ss.Attr.n.u2Dpl;
1600	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_SEG;
1601	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS);
1602	}
1603
1604	/* Guest TR. */
1605	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_TR))
1606	{
1607	HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, TR, tr);
1608	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_TR);
1609	}
1610
1611	/* Guest LDTR. */
1612	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LDTR))
1613	{
1614	HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, LDTR, ldtr);
1615	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_LDTR);
1616	}
1617
1618	/* Guest GDTR. */
1619	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_GDTR))
1620	{
1621	pVmcb->guest.GDTR.u32Limit = pCtx->gdtr.cbGdt;
1622	pVmcb->guest.GDTR.u64Base = pCtx->gdtr.pGdt;
1623	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_DT;
1624	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_GDTR);
1625	}
1626
1627	/* Guest IDTR. */
1628	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_IDTR))
1629	{
1630	pVmcb->guest.IDTR.u32Limit = pCtx->idtr.cbIdt;
1631	pVmcb->guest.IDTR.u64Base = pCtx->idtr.pIdt;
1632	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_DT;
1633	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_IDTR);
1634	}
1635	}
1636
1637
1638	/**
1639	* Loads the guest (or nested-guest) MSRs into the VMCB.
1640	*
1641	* @param pVCpu The cross context virtual CPU structure.
1642	* @param pVmcb Pointer to the VM control block.
1643	* @param pCtx Pointer to the guest-CPU context.
1644	*
1645	* @remarks No-long-jump zone!!!
1646	*/
1647	static void hmR0SvmLoadGuestMsrs(PVMCPU pVCpu, PSVMVMCB pVmcb, PCPUMCTX pCtx)
1648	{
1649	/* Guest Sysenter MSRs. */
1650	pVmcb->guest.u64SysEnterCS = pCtx->SysEnter.cs;
1651	pVmcb->guest.u64SysEnterEIP = pCtx->SysEnter.eip;
1652	pVmcb->guest.u64SysEnterESP = pCtx->SysEnter.esp;
1653
1654	/*
1655	* Guest EFER MSR.
1656	* AMD-V requires guest EFER.SVME to be set. Weird.
1657	* See AMD spec. 15.5.1 "Basic Operation" \| "Canonicalization and Consistency Checks".
1658	*/
1659	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_EFER_MSR))
1660	{
1661	pVmcb->guest.u64EFER = pCtx->msrEFER \| MSR_K6_EFER_SVME;
1662	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_CRX_EFER;
1663	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_EFER_MSR);
1664	}
1665
1666	/* 64-bit MSRs. */
1667	if (CPUMIsGuestInLongModeEx(pCtx))
1668	{
1669	/* Load these always as the guest may modify FS/GS base using MSRs in 64-bit mode which we don't intercept. */
1670	pVmcb->guest.FS.u64Base = pCtx->fs.u64Base;
1671	pVmcb->guest.GS.u64Base = pCtx->gs.u64Base;
1672	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_SEG;
1673	}
1674	else
1675	{
1676	/* If the guest isn't in 64-bit mode, clear MSR_K6_LME bit from guest EFER otherwise AMD-V expects amd64 shadow paging. */
1677	if (pCtx->msrEFER & MSR_K6_EFER_LME)
1678	{
1679	pVmcb->guest.u64EFER &= ~MSR_K6_EFER_LME;
1680	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_CRX_EFER;
1681	}
1682	}
1683
1684	/** @todo The following are used in 64-bit only (SYSCALL/SYSRET) but they might
1685	* be writable in 32-bit mode. Clarify with AMD spec. */
1686	pVmcb->guest.u64STAR = pCtx->msrSTAR;
1687	pVmcb->guest.u64LSTAR = pCtx->msrLSTAR;
1688	pVmcb->guest.u64CSTAR = pCtx->msrCSTAR;
1689	pVmcb->guest.u64SFMASK = pCtx->msrSFMASK;
1690	pVmcb->guest.u64KernelGSBase = pCtx->msrKERNELGSBASE;
1691	}
1692
1693
1694	/**
1695	* Loads the guest (or nested-guest) debug state into the VMCB and programs the
1696	* necessary intercepts accordingly.
1697	*
1698	* @param pVCpu The cross context virtual CPU structure.
1699	* @param pVmcb Pointer to the VM control block.
1700	* @param pCtx Pointer to the guest-CPU context.
1701	*
1702	* @remarks No-long-jump zone!!!
1703	* @remarks Requires EFLAGS to be up-to-date in the VMCB!
1704	*/
1705	static void hmR0SvmLoadSharedDebugState(PVMCPU pVCpu, PSVMVMCB pVmcb, PCPUMCTX pCtx)
1706	{
1707	bool fInterceptMovDRx = false;
1708
1709	/*
1710	* Anyone single stepping on the host side? If so, we'll have to use the
1711	* trap flag in the guest EFLAGS since AMD-V doesn't have a trap flag on
1712	* the VMM level like the VT-x implementations does.
1713	*/
1714	bool const fStepping = pVCpu->hm.s.fSingleInstruction \|\| DBGFIsStepping(pVCpu);
1715	if (fStepping)
1716	{
1717	pVCpu->hm.s.fClearTrapFlag = true;
1718	pVmcb->guest.u64RFlags \|= X86_EFL_TF;
1719	fInterceptMovDRx = true; /* Need clean DR6, no guest mess. */
1720	}
1721
1722	if ( fStepping
1723	\|\| (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK))
1724	{
1725	/*
1726	* Use the combined guest and host DRx values found in the hypervisor
1727	* register set because the debugger has breakpoints active or someone
1728	* is single stepping on the host side.
1729	*
1730	* Note! DBGF expects a clean DR6 state before executing guest code.
1731	*/
1732	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
1733	if ( CPUMIsGuestInLongModeEx(pCtx)
1734	&& !CPUMIsHyperDebugStateActivePending(pVCpu))
1735	{
1736	CPUMR0LoadHyperDebugState(pVCpu, false /* include DR6 */);
1737	Assert(!CPUMIsGuestDebugStateActivePending(pVCpu));
1738	Assert(CPUMIsHyperDebugStateActivePending(pVCpu));
1739	}
1740	else
1741	#endif
1742	if (!CPUMIsHyperDebugStateActive(pVCpu))
1743	{
1744	CPUMR0LoadHyperDebugState(pVCpu, false /* include DR6 */);
1745	Assert(!CPUMIsGuestDebugStateActive(pVCpu));
1746	Assert(CPUMIsHyperDebugStateActive(pVCpu));
1747	}
1748
1749	/* Update DR6 & DR7. (The other DRx values are handled by CPUM one way or the other.) */
1750	if ( pVmcb->guest.u64DR6 != X86_DR6_INIT_VAL
1751	\|\| pVmcb->guest.u64DR7 != CPUMGetHyperDR7(pVCpu))
1752	{
1753	pVmcb->guest.u64DR7 = CPUMGetHyperDR7(pVCpu);
1754	pVmcb->guest.u64DR6 = X86_DR6_INIT_VAL;
1755	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_DRX;
1756	pVCpu->hm.s.fUsingHyperDR7 = true;
1757	}
1758
1759	/** @todo If we cared, we could optimize to allow the guest to read registers
1760	* with the same values. */
1761	fInterceptMovDRx = true;
1762	Log5(("hmR0SvmLoadSharedDebugState: Loaded hyper DRx\n"));
1763	}
1764	else
1765	{
1766	/*
1767	* Update DR6, DR7 with the guest values if necessary.
1768	*/
1769	if ( pVmcb->guest.u64DR7 != pCtx->dr[7]
1770	\|\| pVmcb->guest.u64DR6 != pCtx->dr[6])
1771	{
1772	pVmcb->guest.u64DR7 = pCtx->dr[7];
1773	pVmcb->guest.u64DR6 = pCtx->dr[6];
1774	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_DRX;
1775	pVCpu->hm.s.fUsingHyperDR7 = false;
1776	}
1777
1778	/*
1779	* If the guest has enabled debug registers, we need to load them prior to
1780	* executing guest code so they'll trigger at the right time.
1781	*/
1782	if (pCtx->dr[7] & (X86_DR7_ENABLED_MASK \| X86_DR7_GD)) /** @todo Why GD? */
1783	{
1784	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
1785	if ( CPUMIsGuestInLongModeEx(pCtx)
1786	&& !CPUMIsGuestDebugStateActivePending(pVCpu))
1787	{
1788	CPUMR0LoadGuestDebugState(pVCpu, false /* include DR6 */);
1789	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
1790	Assert(!CPUMIsHyperDebugStateActivePending(pVCpu));
1791	Assert(CPUMIsGuestDebugStateActivePending(pVCpu));
1792	}
1793	else
1794	#endif
1795	if (!CPUMIsGuestDebugStateActive(pVCpu))
1796	{
1797	CPUMR0LoadGuestDebugState(pVCpu, false /* include DR6 */);
1798	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
1799	Assert(!CPUMIsHyperDebugStateActive(pVCpu));
1800	Assert(CPUMIsGuestDebugStateActive(pVCpu));
1801	}
1802	Log5(("hmR0SvmLoadSharedDebugState: Loaded guest DRx\n"));
1803	}
1804	/*
1805	* If no debugging enabled, we'll lazy load DR0-3. We don't need to
1806	* intercept #DB as DR6 is updated in the VMCB.
1807	*
1808	* Note! If we cared and dared, we could skip intercepting \#DB here.
1809	* However, \#DB shouldn't be performance critical, so we'll play safe
1810	* and keep the code similar to the VT-x code and always intercept it.
1811	*/
1812	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
1813	else if ( !CPUMIsGuestDebugStateActivePending(pVCpu)
1814	&& !CPUMIsGuestDebugStateActive(pVCpu))
1815	#else
1816	else if (!CPUMIsGuestDebugStateActive(pVCpu))
1817	#endif
1818	{
1819	fInterceptMovDRx = true;
1820	}
1821	}
1822
1823	Assert(pVmcb->ctrl.u32InterceptXcpt & RT_BIT_32(X86_XCPT_DB));
1824	if (fInterceptMovDRx)
1825	{
1826	if ( pVmcb->ctrl.u16InterceptRdDRx != 0xffff
1827	\|\| pVmcb->ctrl.u16InterceptWrDRx != 0xffff)
1828	{
1829	pVmcb->ctrl.u16InterceptRdDRx = 0xffff;
1830	pVmcb->ctrl.u16InterceptWrDRx = 0xffff;
1831	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
1832	}
1833	}
1834	else
1835	{
1836	if ( pVmcb->ctrl.u16InterceptRdDRx
1837	\|\| pVmcb->ctrl.u16InterceptWrDRx)
1838	{
1839	pVmcb->ctrl.u16InterceptRdDRx = 0;
1840	pVmcb->ctrl.u16InterceptWrDRx = 0;
1841	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
1842	}
1843	}
1844	Log4(("hmR0SvmLoadSharedDebugState: DR6=%#RX64 DR7=%#RX64\n", pCtx->dr[6], pCtx->dr[7]));
1845	}
1846
1847
1848	#ifdef VBOX_WITH_NESTED_HWVIRT
1849	/**
1850	* Loads the nested-guest APIC state (currently just the TPR).
1851	*
1852	* @param pVCpu The cross context virtual CPU structure.
1853	* @param pVmcbNstGst Pointer to the nested-guest VM control block.
1854	*/
1855	static void hmR0SvmLoadGuestApicStateNested(PVMCPU pVCpu, PSVMVMCB pVmcbNstGst)
1856	{
1857	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_SVM_GUEST_APIC_STATE))
1858	{
1859	/* Always enable V_INTR_MASKING as we do not want to allow access to the physical APIC TPR. */
1860	pVmcbNstGst->ctrl.IntCtrl.n.u1VIntrMasking = 1;
1861	pVCpu->hm.s.svm.fSyncVTpr = false;
1862	pVmcbNstGst->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_TPR;
1863
1864	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_SVM_GUEST_APIC_STATE);
1865	}
1866	}
1867	#endif
1868
1869	/**
1870	* Loads the guest APIC state (currently just the TPR).
1871	*
1872	* @returns VBox status code.
1873	* @param pVCpu The cross context virtual CPU structure.
1874	* @param pVmcb Pointer to the VM control block.
1875	* @param pCtx Pointer to the guest-CPU context.
1876	*/
1877	static int hmR0SvmLoadGuestApicState(PVMCPU pVCpu, PSVMVMCB pVmcb, PCPUMCTX pCtx)
1878	{
1879	if (!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_SVM_GUEST_APIC_STATE))
1880	return VINF_SUCCESS;
1881
1882	int rc = VINF_SUCCESS;
1883	PVM pVM = pVCpu->CTX_SUFF(pVM);
1884	if ( PDMHasApic(pVM)
1885	&& APICIsEnabled(pVCpu))
1886	{
1887	bool fPendingIntr;
1888	uint8_t u8Tpr;
1889	rc = APICGetTpr(pVCpu, &u8Tpr, &fPendingIntr, NULL /* pu8PendingIrq */);
1890	AssertRCReturn(rc, rc);
1891
1892	/* Assume that we need to trap all TPR accesses and thus need not check on
1893	every #VMEXIT if we should update the TPR. */
1894	Assert(pVmcb->ctrl.IntCtrl.n.u1VIntrMasking);
1895	pVCpu->hm.s.svm.fSyncVTpr = false;
1896
1897	/* 32-bit guests uses LSTAR MSR for patching guest code which touches the TPR. */
1898	if (pVM->hm.s.fTPRPatchingActive)
1899	{
1900	pCtx->msrLSTAR = u8Tpr;
1901	uint8_t pbMsrBitmap = (uint8_t )pVCpu->hm.s.svm.pvMsrBitmap;
1902
1903	/* If there are interrupts pending, intercept LSTAR writes, otherwise don't intercept reads or writes. */
1904	if (fPendingIntr)
1905	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_K8_LSTAR, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_INTERCEPT_WRITE);
1906	else
1907	{
1908	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_K8_LSTAR, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1909	pVCpu->hm.s.svm.fSyncVTpr = true;
1910	}
1911	}
1912	else
1913	{
1914	/* Bits 3-0 of the VTPR field correspond to bits 7-4 of the TPR (which is the Task-Priority Class). */
1915	pVmcb->ctrl.IntCtrl.n.u8VTPR = (u8Tpr >> 4);
1916
1917	/* If there are interrupts pending, intercept CR8 writes to evaluate ASAP if we can deliver the interrupt to the guest. */
1918	if (fPendingIntr)
1919	pVmcb->ctrl.u16InterceptWrCRx \|= RT_BIT(8);
1920	else
1921	{
1922	pVmcb->ctrl.u16InterceptWrCRx &= ~RT_BIT(8);
1923	pVCpu->hm.s.svm.fSyncVTpr = true;
1924	}
1925
1926	pVmcb->ctrl.u32VmcbCleanBits &= ~(HMSVM_VMCB_CLEAN_INTERCEPTS \| HMSVM_VMCB_CLEAN_TPR);
1927	}
1928	}
1929
1930	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_SVM_GUEST_APIC_STATE);
1931	return rc;
1932	}
1933
1934
1935	/**
1936	* Loads the exception interrupts required for guest (or nested-guest) execution in
1937	* the VMCB.
1938	*
1939	* @param pVCpu The cross context virtual CPU structure.
1940	* @param pVmcb Pointer to the VM control block.
1941	* @param pCtx Pointer to the guest-CPU context.
1942	*/
1943	static void hmR0SvmLoadGuestXcptIntercepts(PVMCPU pVCpu, PSVMVMCB pVmcb, PCPUMCTX pCtx)
1944	{
1945	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS))
1946	{
1947	/* Trap #UD for GIM provider (e.g. for hypercalls). */
1948	if (pVCpu->hm.s.fGIMTrapXcptUD)
1949	hmR0SvmAddXcptIntercept(pVmcb, X86_XCPT_UD);
1950	else
1951	hmR0SvmRemoveXcptIntercept(pVCpu, pCtx, pVmcb, X86_XCPT_UD);
1952
1953	/* Trap #BP for INT3 debug breakpoints set by the VM debugger. */
1954	if (pVCpu->CTX_SUFF(pVM)->dbgf.ro.cEnabledInt3Breakpoints)
1955	hmR0SvmAddXcptIntercept(pVmcb, X86_XCPT_BP);
1956	else
1957	hmR0SvmRemoveXcptIntercept(pVCpu, pCtx, pVmcb, X86_XCPT_BP);
1958
1959	/* The remaining intercepts are handled elsewhere, e.g. in hmR0SvmLoadSharedCR0(). */
1960	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
1961	}
1962	}
1963
1964
1965	#ifdef VBOX_WITH_NESTED_HWVIRT
1966	/**
1967	* Loads the intercepts required for nested-guest execution in the VMCB.
1968	*
1969	* This merges the guest and nested-guest intercepts in a way that if the outer
1970	* guest intercepts an exception we need to intercept it in the nested-guest as
1971	* well and handle it accordingly.
1972	*
1973	* @param pVCpu The cross context virtual CPU structure.
1974	* @param pVmcbNstGst Pointer to the nested-guest VM control block.
1975	* @param pCtx Pointer to the guest-CPU context.
1976	*/
1977	static void hmR0SvmLoadGuestXcptInterceptsNested(PVMCPU pVCpu, PSVMVMCB pVmcbNstGst, PCPUMCTX pCtx)
1978	{
1979	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS))
1980	{
1981	/* First, load the guest exception intercepts into the guest VMCB. */
1982	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
1983	hmR0SvmLoadGuestXcptIntercepts(pVCpu, pVmcb, pCtx);
1984
1985	/* Next, merge the intercepts into the nested-guest VMCB. */
1986	pVmcbNstGst->ctrl.u16InterceptRdCRx \|= pVmcb->ctrl.u16InterceptRdCRx;
1987	pVmcbNstGst->ctrl.u16InterceptWrCRx \|= pVmcb->ctrl.u16InterceptWrCRx;
1988
1989	/* Always intercept CR0, CR4 reads and writes as we alter them. */
1990	pVmcbNstGst->ctrl.u16InterceptRdCRx \|= RT_BIT(0) \| RT_BIT(4);
1991	pVmcbNstGst->ctrl.u16InterceptWrCRx \|= RT_BIT(0) \| RT_BIT(4);
1992
1993	/* Always intercept CR3 reads and writes without nested-paging as we load shadow page tables. */
1994	if (!pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging)
1995	{
1996	pVmcbNstGst->ctrl.u16InterceptRdCRx \|= RT_BIT(3);
1997	pVmcbNstGst->ctrl.u16InterceptWrCRx \|= RT_BIT(3);
1998	}
1999
2000	/** @todo Figure out debugging with nested-guests, till then just intercept
2001	* all DR[0-15] accesses. */
2002	pVmcbNstGst->ctrl.u16InterceptRdDRx \|= 0xffff;
2003	pVmcbNstGst->ctrl.u16InterceptWrDRx \|= 0xffff;
2004
2005	/* Exclude the VINTR intercept of the outer guest as we don't need to cause VINTR #VMEXITs
2006	that belong to the nested-guest to the outer guest. */
2007	pVmcbNstGst->ctrl.u32InterceptXcpt \|= pVmcb->ctrl.u32InterceptXcpt;
2008	pVmcbNstGst->ctrl.u64InterceptCtrl \|= (pVmcb->ctrl.u64InterceptCtrl & ~SVM_CTRL_INTERCEPT_VINTR)
2009	\| HMSVM_MANDATORY_GUEST_CTRL_INTERCEPTS;
2010
2011	/*
2012	* Adjust control intercepts while executing the nested-guest that differ
2013	* from the outer guest intercepts.
2014	*
2015	* VMMCALL when not intercepted raises a \#UD exception in the guest. However,
2016	* other SVM instructions like VMSAVE when not intercept can cause havoc on the
2017	* host as they can write to any location in physical memory, hence they always
2018	* need to be intercepted (see below).
2019	*/
2020	Assert( (pVmcbNstGst->ctrl.u64InterceptCtrl & HMSVM_MANDATORY_GUEST_CTRL_INTERCEPTS)
2021	== HMSVM_MANDATORY_GUEST_CTRL_INTERCEPTS);
2022	pVmcbNstGst->ctrl.u64InterceptCtrl &= ~SVM_CTRL_INTERCEPT_VMMCALL;
2023
2024	/*
2025	* If we don't expose Virtualized-VMSAVE/VMLOAD feature to the outer guest, we
2026	* need to intercept VMSAVE/VMLOAD instructions executed by the nested-guest.
2027	*/
2028	if (!pVCpu->CTX_SUFF(pVM)->cpum.ro.GuestFeatures.fSvmVirtVmsaveVmload)
2029	{
2030	pVmcbNstGst->ctrl.u64InterceptCtrl \|= SVM_CTRL_INTERCEPT_VMSAVE
2031	\| SVM_CTRL_INTERCEPT_VMLOAD;
2032	}
2033
2034	/*
2035	* If we don't expose Virtual GIF feature to the outer guest, we need to intercept
2036	* CLGI/STGI instructions executed by the nested-guest.
2037	*/
2038	if (!pVCpu->CTX_SUFF(pVM)->cpum.ro.GuestFeatures.fSvmVGif)
2039	{
2040	pVmcbNstGst->ctrl.u64InterceptCtrl \|= SVM_CTRL_INTERCEPT_CLGI
2041	\| SVM_CTRL_INTERCEPT_STGI;
2042	}
2043
2044	/* Finally, update the VMCB clean bits. */
2045	pVmcbNstGst->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
2046
2047	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS));
2048	}
2049	}
2050	#endif
2051
2052
2053	/**
2054	* Sets up the appropriate function to run guest code.
2055	*
2056	* @returns VBox status code.
2057	* @param pVCpu The cross context virtual CPU structure.
2058	*
2059	* @remarks No-long-jump zone!!!
2060	*/
2061	static int hmR0SvmSetupVMRunHandler(PVMCPU pVCpu)
2062	{
2063	if (CPUMIsGuestInLongMode(pVCpu))
2064	{
2065	#ifndef VBOX_ENABLE_64_BITS_GUESTS
2066	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
2067	#endif
2068	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests); /* Guaranteed by hmR3InitFinalizeR0(). */
2069	#if HC_ARCH_BITS == 32
2070	/* 32-bit host. We need to switch to 64-bit before running the 64-bit guest. */
2071	pVCpu->hm.s.svm.pfnVMRun = SVMR0VMSwitcherRun64;
2072	#else
2073	/* 64-bit host or hybrid host. */
2074	pVCpu->hm.s.svm.pfnVMRun = SVMR0VMRun64;
2075	#endif
2076	}
2077	else
2078	{
2079	/* Guest is not in long mode, use the 32-bit handler. */
2080	pVCpu->hm.s.svm.pfnVMRun = SVMR0VMRun;
2081	}
2082	return VINF_SUCCESS;
2083	}
2084
2085
2086	/**
2087	* Enters the AMD-V session.
2088	*
2089	* @returns VBox status code.
2090	* @param pVM The cross context VM structure.
2091	* @param pVCpu The cross context virtual CPU structure.
2092	* @param pCpu Pointer to the CPU info struct.
2093	*/
2094	VMMR0DECL(int) SVMR0Enter(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2095	{
2096	AssertPtr(pVM);
2097	AssertPtr(pVCpu);
2098	Assert(pVM->hm.s.svm.fSupported);
2099	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2100	NOREF(pVM); NOREF(pCpu);
2101
2102	LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
2103	Assert(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_HOST_GUEST_SHARED_STATE));
2104
2105	pVCpu->hm.s.fLeaveDone = false;
2106	return VINF_SUCCESS;
2107	}
2108
2109
2110	/**
2111	* Thread-context callback for AMD-V.
2112	*
2113	* @param enmEvent The thread-context event.
2114	* @param pVCpu The cross context virtual CPU structure.
2115	* @param fGlobalInit Whether global VT-x/AMD-V init. is used.
2116	* @thread EMT(pVCpu)
2117	*/
2118	VMMR0DECL(void) SVMR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPU pVCpu, bool fGlobalInit)
2119	{
2120	NOREF(fGlobalInit);
2121
2122	switch (enmEvent)
2123	{
2124	case RTTHREADCTXEVENT_OUT:
2125	{
2126	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2127	Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
2128	VMCPU_ASSERT_EMT(pVCpu);
2129
2130	/* No longjmps (log-flush, locks) in this fragile context. */
2131	VMMRZCallRing3Disable(pVCpu);
2132
2133	if (!pVCpu->hm.s.fLeaveDone)
2134	{
2135	hmR0SvmLeave(pVCpu);
2136	pVCpu->hm.s.fLeaveDone = true;
2137	}
2138
2139	/* Leave HM context, takes care of local init (term). */
2140	int rc = HMR0LeaveCpu(pVCpu);
2141	AssertRC(rc); NOREF(rc);
2142
2143	/* Restore longjmp state. */
2144	VMMRZCallRing3Enable(pVCpu);
2145	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreempt);
2146	break;
2147	}
2148
2149	case RTTHREADCTXEVENT_IN:
2150	{
2151	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2152	Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
2153	VMCPU_ASSERT_EMT(pVCpu);
2154
2155	/* No longjmps (log-flush, locks) in this fragile context. */
2156	VMMRZCallRing3Disable(pVCpu);
2157
2158	/*
2159	* Initialize the bare minimum state required for HM. This takes care of
2160	* initializing AMD-V if necessary (onlined CPUs, local init etc.)
2161	*/
2162	int rc = HMR0EnterCpu(pVCpu);
2163	AssertRC(rc); NOREF(rc);
2164	Assert(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_HOST_GUEST_SHARED_STATE));
2165
2166	pVCpu->hm.s.fLeaveDone = false;
2167
2168	/* Restore longjmp state. */
2169	VMMRZCallRing3Enable(pVCpu);
2170	break;
2171	}
2172
2173	default:
2174	break;
2175	}
2176	}
2177
2178
2179	/**
2180	* Saves the host state.
2181	*
2182	* @returns VBox status code.
2183	* @param pVM The cross context VM structure.
2184	* @param pVCpu The cross context virtual CPU structure.
2185	*
2186	* @remarks No-long-jump zone!!!
2187	*/
2188	VMMR0DECL(int) SVMR0SaveHostState(PVM pVM, PVMCPU pVCpu)
2189	{
2190	NOREF(pVM);
2191	NOREF(pVCpu);
2192	/* Nothing to do here. AMD-V does this for us automatically during the world-switch. */
2193	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_HOST_CONTEXT);
2194	return VINF_SUCCESS;
2195	}
2196
2197
2198	/**
2199	* Loads the guest state into the VMCB.
2200	*
2201	* The CPU state will be loaded from these fields on every successful VM-entry.
2202	* Also sets up the appropriate VMRUN function to execute guest code based on
2203	* the guest CPU mode.
2204	*
2205	* @returns VBox status code.
2206	* @param pVM The cross context VM structure.
2207	* @param pVCpu The cross context virtual CPU structure.
2208	* @param pCtx Pointer to the guest-CPU context.
2209	*
2210	* @remarks No-long-jump zone!!!
2211	*/
2212	static int hmR0SvmLoadGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
2213	{
2214	HMSVM_ASSERT_NOT_IN_NESTED_GUEST(pCtx);
2215
2216	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
2217	AssertMsgReturn(pVmcb, ("Invalid pVmcb\n"), VERR_SVM_INVALID_PVMCB);
2218
2219	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestState, x);
2220
2221	int rc = hmR0SvmLoadGuestControlRegs(pVCpu, pVmcb, pCtx);
2222	AssertLogRelMsgRCReturn(rc, ("hmR0SvmLoadGuestControlRegs! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
2223
2224	hmR0SvmLoadGuestSegmentRegs(pVCpu, pVmcb, pCtx);
2225	hmR0SvmLoadGuestMsrs(pVCpu, pVmcb, pCtx);
2226
2227	pVmcb->guest.u64RIP = pCtx->rip;
2228	pVmcb->guest.u64RSP = pCtx->rsp;
2229	pVmcb->guest.u64RFlags = pCtx->eflags.u32;
2230	pVmcb->guest.u64RAX = pCtx->rax;
2231
2232	#ifdef VBOX_WITH_NESTED_HWVIRT
2233	if (pVmcb->ctrl.IntCtrl.n.u1VGifEnable == 1)
2234	{
2235	Assert(pVM->hm.s.svm.fVGif);
2236	pVmcb->ctrl.IntCtrl.n.u1VGif = pCtx->hwvirt.svm.fGif;
2237	}
2238	#endif
2239
2240	rc = hmR0SvmLoadGuestApicState(pVCpu, pVmcb, pCtx);
2241	AssertLogRelMsgRCReturn(rc, ("hmR0SvmLoadGuestApicState! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
2242
2243	hmR0SvmLoadGuestXcptIntercepts(pVCpu, pVmcb, pCtx);
2244
2245	rc = hmR0SvmSetupVMRunHandler(pVCpu);
2246	AssertLogRelMsgRCReturn(rc, ("hmR0SvmSetupVMRunHandler! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
2247
2248	/* Clear any unused and reserved bits. */
2249	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RIP /* Unused (loaded unconditionally). */
2250	\| HM_CHANGED_GUEST_RSP
2251	\| HM_CHANGED_GUEST_RFLAGS
2252	\| HM_CHANGED_GUEST_SYSENTER_CS_MSR
2253	\| HM_CHANGED_GUEST_SYSENTER_EIP_MSR
2254	\| HM_CHANGED_GUEST_SYSENTER_ESP_MSR
2255	\| HM_CHANGED_GUEST_LAZY_MSRS /* Unused. */
2256	\| HM_CHANGED_SVM_RESERVED1 /* Reserved. */
2257	\| HM_CHANGED_SVM_RESERVED2
2258	\| HM_CHANGED_SVM_RESERVED3
2259	\| HM_CHANGED_SVM_RESERVED4);
2260
2261	/* All the guest state bits should be loaded except maybe the host context and/or shared host/guest bits. */
2262	AssertMsg( !HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_ALL_GUEST)
2263	\|\| HMCPU_CF_IS_PENDING_ONLY(pVCpu, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_HOST_GUEST_SHARED_STATE),
2264	("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
2265
2266	Log4(("hmR0SvmLoadGuestState: CS:RIP=%04x:%RX64 EFL=%#x CR0=%#RX32 CR3=%#RX32 CR4=%#RX32 ESP=%#RX32 EBP=%#RX32\n",
2267	pCtx->cs.Sel, pCtx->rip, pCtx->eflags.u, pCtx->cr0, pCtx->cr3, pCtx->cr4, pCtx->esp, pCtx->ebp));
2268	Log4(("hmR0SvmLoadGuestState: SS={%04x base=%016RX64 limit=%08x flags=%08x}\n", pCtx->ss.Sel, pCtx->ss.u64Base,
2269	pCtx->ss.u32Limit, pCtx->ss.Attr.u));
2270	Log4(("hmR0SvmLoadGuestState: FS={%04x base=%016RX64 limit=%08x flags=%08x}\n", pCtx->fs.Sel, pCtx->fs.u64Base,
2271	pCtx->fs.u32Limit, pCtx->fs.Attr.u));
2272	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestState, x);
2273	return rc;
2274	}
2275
2276
2277	#ifdef VBOX_WITH_NESTED_HWVIRT
2278	/**
2279	* Caches the nested-guest VMCB fields before we modify them for execution using
2280	* hardware-assisted SVM.
2281	*
2282	* @returns true if the VMCB was previously already cached, false otherwise.
2283	* @param pCtx Pointer to the guest-CPU context.
2284	*
2285	* @sa HMSvmNstGstVmExitNotify.
2286	*/
2287	static bool hmR0SvmVmRunCacheVmcb(PVMCPU pVCpu, PCPUMCTX pCtx)
2288	{
2289	PSVMVMCB pVmcbNstGst = pCtx->hwvirt.svm.CTX_SUFF(pVmcb);
2290	PCSVMVMCBCTRL pVmcbNstGstCtrl = &pVmcbNstGst->ctrl;
2291	PCSVMVMCBSTATESAVE pVmcbNstGstState = &pVmcbNstGst->guest;
2292	PSVMNESTEDVMCBCACHE pNstGstVmcbCache = &pVCpu->hm.s.svm.NstGstVmcbCache;
2293
2294	/*
2295	* Cache the nested-guest programmed VMCB fields if we have not cached it yet.
2296	* Otherwise we risk re-caching the values we may have modified, see @bugref{7243#c44}.
2297	*
2298	* Nested-paging CR3 is not saved back into the VMCB on #VMEXIT, hence no need to
2299	* cache and restore it, see AMD spec. 15.25.4 "Nested Paging and VMRUN/#VMEXIT".
2300	*/
2301	bool const fWasCached = pCtx->hwvirt.svm.fHMCachedVmcb;
2302	if (!fWasCached)
2303	{
2304	pNstGstVmcbCache->u16InterceptRdCRx = pVmcbNstGstCtrl->u16InterceptRdCRx;
2305	pNstGstVmcbCache->u16InterceptWrCRx = pVmcbNstGstCtrl->u16InterceptWrCRx;
2306	pNstGstVmcbCache->u16InterceptRdDRx = pVmcbNstGstCtrl->u16InterceptRdDRx;
2307	pNstGstVmcbCache->u16InterceptWrDRx = pVmcbNstGstCtrl->u16InterceptWrDRx;
2308	pNstGstVmcbCache->u32InterceptXcpt = pVmcbNstGstCtrl->u32InterceptXcpt;
2309	pNstGstVmcbCache->u64InterceptCtrl = pVmcbNstGstCtrl->u64InterceptCtrl;
2310	pNstGstVmcbCache->u64CR0 = pVmcbNstGstState->u64CR0;
2311	pNstGstVmcbCache->u64CR3 = pVmcbNstGstState->u64CR3;
2312	pNstGstVmcbCache->u64CR4 = pVmcbNstGstState->u64CR4;
2313	pNstGstVmcbCache->u64EFER = pVmcbNstGstState->u64EFER;
2314	pNstGstVmcbCache->u64DBGCTL = pVmcbNstGstState->u64DBGCTL;
2315	pNstGstVmcbCache->u64IOPMPhysAddr = pVmcbNstGstCtrl->u64IOPMPhysAddr;
2316	pNstGstVmcbCache->u64MSRPMPhysAddr = pVmcbNstGstCtrl->u64MSRPMPhysAddr;
2317	pNstGstVmcbCache->u64TSCOffset = pVmcbNstGstCtrl->u64TSCOffset;
2318	pNstGstVmcbCache->u32VmcbCleanBits = pVmcbNstGstCtrl->u32VmcbCleanBits;
2319	pNstGstVmcbCache->fVIntrMasking = pVmcbNstGstCtrl->IntCtrl.n.u1VIntrMasking;
2320	pNstGstVmcbCache->TLBCtrl = pVmcbNstGstCtrl->TLBCtrl;
2321	pNstGstVmcbCache->u1NestedPaging = pVmcbNstGstCtrl->NestedPaging.n.u1NestedPaging;
2322	pNstGstVmcbCache->u1LbrVirt = pVmcbNstGstCtrl->LbrVirt.n.u1LbrVirt;
2323	pCtx->hwvirt.svm.fHMCachedVmcb = true;
2324	Log4(("hmR0SvmVmRunCacheVmcb: Cached VMCB fields\n"));
2325	}
2326
2327	return fWasCached;
2328	}
2329
2330
2331	/**
2332	* Sets up the nested-guest VMCB for execution using hardware-assisted SVM.
2333	*
2334	* @param pVCpu The cross context virtual CPU structure.
2335	* @param pCtx Pointer to the guest-CPU context.
2336	*/
2337	static void hmR0SvmVmRunSetupVmcb(PVMCPU pVCpu, PCPUMCTX pCtx)
2338	{
2339	PSVMVMCB pVmcbNstGst = pCtx->hwvirt.svm.CTX_SUFF(pVmcb);
2340	PSVMVMCBCTRL pVmcbNstGstCtrl = &pVmcbNstGst->ctrl;
2341
2342	/*
2343	* First cache the nested-guest VMCB fields we may potentially modify.
2344	*/
2345	bool const fVmcbCached = hmR0SvmVmRunCacheVmcb(pVCpu, pCtx);
2346	if (!fVmcbCached)
2347	{
2348	/*
2349	* The IOPM of the nested-guest can be ignored because the the guest always
2350	* intercepts all IO port accesses. Thus, we'll swap to the guest IOPM rather
2351	* into the nested-guest one and swap it back on the #VMEXIT.
2352	*/
2353	pVmcbNstGstCtrl->u64IOPMPhysAddr = g_HCPhysIOBitmap;
2354
2355	/*
2356	* Load the host-physical address into the MSRPM rather than the nested-guest
2357	* physical address (currently we trap all MSRs in the nested-guest).
2358	*/
2359	pVmcbNstGstCtrl->u64MSRPMPhysAddr = g_HCPhysNstGstMsrBitmap;
2360
2361	/*
2362	* Use the same nested-paging as the "outer" guest. We can't dynamically
2363	* switch off nested-paging suddenly while executing a VM (see assertion at the
2364	* end of Trap0eHandler in PGMAllBth.h).
2365	*/
2366	pVmcbNstGstCtrl->NestedPaging.n.u1NestedPaging = pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging;
2367
2368	/* For now copy the LBR info. from outer guest VMCB. */
2369	/** @todo fix this later. */
2370	PCSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
2371	pVmcbNstGstCtrl->LbrVirt.n.u1LbrVirt = pVmcb->ctrl.LbrVirt.n.u1LbrVirt;
2372	pVmcbNstGst->guest.u64DBGCTL = pVmcb->guest.u64DBGCTL;
2373	}
2374	else
2375	{
2376	Assert(pVmcbNstGstCtrl->u64IOPMPhysAddr == g_HCPhysIOBitmap);
2377	Assert(pVmcbNstGstCtrl->u64MSRPMPhysAddr = g_HCPhysNstGstMsrBitmap);
2378	Assert(RT_BOOL(pVmcbNstGstCtrl->NestedPaging.n.u1NestedPaging) == pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
2379	}
2380	}
2381
2382
2383	/**
2384	* Loads the nested-guest state into the VMCB.
2385	*
2386	* @returns VBox status code.
2387	* @param pVCpu The cross context virtual CPU structure.
2388	* @param pCtx Pointer to the guest-CPU context.
2389	*
2390	* @remarks No-long-jump zone!!!
2391	*/
2392	static int hmR0SvmLoadGuestStateNested(PVMCPU pVCpu, PCPUMCTX pCtx)
2393	{
2394	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestState, x);
2395
2396	PSVMVMCB pVmcbNstGst = pCtx->hwvirt.svm.CTX_SUFF(pVmcb);
2397	Assert(pVmcbNstGst);
2398
2399	hmR0SvmVmRunSetupVmcb(pVCpu, pCtx);
2400
2401	int rc = hmR0SvmLoadGuestControlRegs(pVCpu, pVmcbNstGst, pCtx);
2402	AssertRCReturn(rc, rc);
2403
2404	/*
2405	* We need to load the entire state (including FS, GS etc.) as we could be continuing
2406	* to execute the nested-guest at any point (not just immediately after VMRUN) and thus
2407	* the VMCB can possibly be out-of-sync with the actual nested-guest state if it was
2408	* executed in IEM.
2409	*/
2410	hmR0SvmLoadGuestSegmentRegs(pVCpu, pVmcbNstGst, pCtx);
2411	hmR0SvmLoadGuestMsrs(pVCpu, pVmcbNstGst, pCtx);
2412	hmR0SvmLoadGuestApicStateNested(pVCpu, pVmcbNstGst);
2413
2414	pVmcbNstGst->guest.u64RIP = pCtx->rip;
2415	pVmcbNstGst->guest.u64RSP = pCtx->rsp;
2416	pVmcbNstGst->guest.u64RFlags = pCtx->eflags.u32;
2417	pVmcbNstGst->guest.u64RAX = pCtx->rax;
2418
2419	#ifdef VBOX_WITH_NESTED_HWVIRT
2420	Assert(pVmcbNstGst->ctrl.IntCtrl.n.u1VGifEnable == 0); /* Nested VGIF not supported yet. */
2421	#endif
2422
2423	hmR0SvmLoadGuestXcptInterceptsNested(pVCpu, pVmcbNstGst, pCtx);
2424
2425	rc = hmR0SvmSetupVMRunHandler(pVCpu);
2426	AssertRCReturn(rc, rc);
2427
2428	/* Clear any unused and reserved bits. */
2429	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RIP /* Unused (loaded unconditionally). */
2430	\| HM_CHANGED_GUEST_RSP
2431	\| HM_CHANGED_GUEST_RFLAGS
2432	\| HM_CHANGED_GUEST_SYSENTER_CS_MSR
2433	\| HM_CHANGED_GUEST_SYSENTER_EIP_MSR
2434	\| HM_CHANGED_GUEST_SYSENTER_ESP_MSR
2435	\| HM_CHANGED_GUEST_LAZY_MSRS /* Unused. */
2436	\| HM_CHANGED_SVM_RESERVED1 /* Reserved. */
2437	\| HM_CHANGED_SVM_RESERVED2
2438	\| HM_CHANGED_SVM_RESERVED3
2439	\| HM_CHANGED_SVM_RESERVED4);
2440
2441	/* All the guest state bits should be loaded except maybe the host context and/or shared host/guest bits. */
2442	AssertMsg( !HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_ALL_GUEST)
2443	\|\| HMCPU_CF_IS_PENDING_ONLY(pVCpu, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_HOST_GUEST_SHARED_STATE),
2444	("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
2445
2446	#ifdef VBOX_STRICT
2447	hmR0SvmLogState(pVCpu, pVmcbNstGst, pCtx, "hmR0SvmLoadGuestStateNested", HMSVM_LOG_ALL, 0 /* uVerbose */);
2448	#endif
2449	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestState, x);
2450	return rc;
2451	}
2452	#endif
2453
2454
2455	/**
2456	* Loads the state shared between the host and guest or nested-guest into the
2457	* VMCB.
2458	*
2459	* @param pVCpu The cross context virtual CPU structure.
2460	* @param pVmcb Pointer to the VM control block.
2461	* @param pCtx Pointer to the guest-CPU context.
2462	*
2463	* @remarks No-long-jump zone!!!
2464	*/
2465	static void hmR0SvmLoadSharedState(PVMCPU pVCpu, PSVMVMCB pVmcb, PCPUMCTX pCtx)
2466	{
2467	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2468	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2469
2470	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0))
2471	{
2472	hmR0SvmLoadSharedCR0(pVCpu, pVmcb, pCtx);
2473	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR0);
2474	}
2475
2476	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_DEBUG))
2477	{
2478	/** @todo Figure out stepping with nested-guest. */
2479	if (!CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
2480	hmR0SvmLoadSharedDebugState(pVCpu, pVmcb, pCtx);
2481	else
2482	{
2483	pVmcb->guest.u64DR6 = pCtx->dr[6];
2484	pVmcb->guest.u64DR7 = pCtx->dr[7];
2485	Log4(("hmR0SvmLoadSharedState: DR6=%#RX64 DR7=%#RX64\n", pCtx->dr[6], pCtx->dr[7]));
2486	}
2487
2488	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_DEBUG);
2489	}
2490
2491	/* Unused on AMD-V. */
2492	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS);
2493
2494	AssertMsg(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE),
2495	("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
2496	}
2497
2498
2499	/**
2500	* Saves the guest (or nested-guest) state from the VMCB into the guest-CPU
2501	* context.
2502	*
2503	* Currently there is no residual state left in the CPU that is not updated in the
2504	* VMCB.
2505	*
2506	* @returns VBox status code.
2507	* @param pVCpu The cross context virtual CPU structure.
2508	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
2509	* out-of-sync. Make sure to update the required fields
2510	* before using them.
2511	* @param pVmcb Pointer to the VM control block.
2512	*/
2513	static void hmR0SvmSaveGuestState(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PCSVMVMCB pVmcb)
2514	{
2515	Assert(VMMRZCallRing3IsEnabled(pVCpu));
2516
2517	pMixedCtx->rip = pVmcb->guest.u64RIP;
2518	pMixedCtx->rsp = pVmcb->guest.u64RSP;
2519	pMixedCtx->eflags.u32 = pVmcb->guest.u64RFlags;
2520	pMixedCtx->rax = pVmcb->guest.u64RAX;
2521
2522	#ifdef VBOX_WITH_NESTED_HWVIRT
2523	/*
2524	* Guest Virtual GIF (Global Interrupt Flag).
2525	*/
2526	if ( pVmcb->ctrl.IntCtrl.n.u1VGifEnable == 1
2527	&& !CPUMIsGuestInSvmNestedHwVirtMode(pMixedCtx))
2528	{
2529	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.svm.fVGif);
2530	pMixedCtx->hwvirt.svm.fGif = pVmcb->ctrl.IntCtrl.n.u1VGif;
2531	}
2532	#endif
2533
2534	/*
2535	* Guest interrupt shadow.
2536	*/
2537	if (pVmcb->ctrl.IntShadow.n.u1IntShadow)
2538	EMSetInhibitInterruptsPC(pVCpu, pMixedCtx->rip);
2539	else if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
2540	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
2541
2542	/*
2543	* Guest Control registers: CR2, CR3 (handled at the end) - accesses to other control registers are always intercepted.
2544	*/
2545	pMixedCtx->cr2 = pVmcb->guest.u64CR2;
2546
2547	/*
2548	* Guest MSRs.
2549	*/
2550	pMixedCtx->msrSTAR = pVmcb->guest.u64STAR; /* legacy syscall eip, cs & ss */
2551	pMixedCtx->msrLSTAR = pVmcb->guest.u64LSTAR; /* 64-bit mode syscall rip */
2552	pMixedCtx->msrCSTAR = pVmcb->guest.u64CSTAR; /* compatibility mode syscall rip */
2553	pMixedCtx->msrSFMASK = pVmcb->guest.u64SFMASK; /* syscall flag mask */
2554	pMixedCtx->msrKERNELGSBASE = pVmcb->guest.u64KernelGSBase; /* swapgs exchange value */
2555	pMixedCtx->SysEnter.cs = pVmcb->guest.u64SysEnterCS;
2556	pMixedCtx->SysEnter.eip = pVmcb->guest.u64SysEnterEIP;
2557	pMixedCtx->SysEnter.esp = pVmcb->guest.u64SysEnterESP;
2558
2559	/*
2560	* Guest segment registers (includes FS, GS base MSRs for 64-bit guests).
2561	*/
2562	HMSVM_SEG_REG_COPY_FROM_VMCB(pMixedCtx, &pVmcb->guest, CS, cs);
2563	HMSVM_SEG_REG_COPY_FROM_VMCB(pMixedCtx, &pVmcb->guest, SS, ss);
2564	HMSVM_SEG_REG_COPY_FROM_VMCB(pMixedCtx, &pVmcb->guest, DS, ds);
2565	HMSVM_SEG_REG_COPY_FROM_VMCB(pMixedCtx, &pVmcb->guest, ES, es);
2566	HMSVM_SEG_REG_COPY_FROM_VMCB(pMixedCtx, &pVmcb->guest, FS, fs);
2567	HMSVM_SEG_REG_COPY_FROM_VMCB(pMixedCtx, &pVmcb->guest, GS, gs);
2568
2569	/*
2570	* Correct the hidden CS granularity bit. Haven't seen it being wrong in any other
2571	* register (yet).
2572	*/
2573	/** @todo SELM might need to be fixed as it too should not care about the
2574	* granularity bit. See @bugref{6785}. */
2575	if ( !pMixedCtx->cs.Attr.n.u1Granularity
2576	&& pMixedCtx->cs.Attr.n.u1Present
2577	&& pMixedCtx->cs.u32Limit > UINT32_C(0xfffff))
2578	{
2579	Assert((pMixedCtx->cs.u32Limit & 0xfff) == 0xfff);
2580	pMixedCtx->cs.Attr.n.u1Granularity = 1;
2581	}
2582
2583	HMSVM_ASSERT_SEG_GRANULARITY(cs);
2584	HMSVM_ASSERT_SEG_GRANULARITY(ss);
2585	HMSVM_ASSERT_SEG_GRANULARITY(ds);
2586	HMSVM_ASSERT_SEG_GRANULARITY(es);
2587	HMSVM_ASSERT_SEG_GRANULARITY(fs);
2588	HMSVM_ASSERT_SEG_GRANULARITY(gs);
2589
2590	/*
2591	* Sync the hidden SS DPL field. AMD CPUs have a separate CPL field in the VMCB and uses that
2592	* and thus it's possible that when the CPL changes during guest execution that the SS DPL
2593	* isn't updated by AMD-V. Observed on some AMD Fusion CPUs with 64-bit guests.
2594	* See AMD spec. 15.5.1 "Basic operation".
2595	*/
2596	Assert(!(pVmcb->guest.u8CPL & ~0x3));
2597	uint8_t const uCpl = pVmcb->guest.u8CPL;
2598	if (pMixedCtx->ss.Attr.n.u2Dpl != uCpl)
2599	{
2600	Log4(("hmR0SvmSaveGuestState: CPL differs. SS.DPL=%u, CPL=%u, overwriting SS.DPL!\n", pMixedCtx->ss.Attr.n.u2Dpl, uCpl));
2601	pMixedCtx->ss.Attr.n.u2Dpl = pVmcb->guest.u8CPL & 0x3;
2602	}
2603
2604	/*
2605	* Guest TR.
2606	* Fixup TR attributes so it's compatible with Intel. Important when saved-states are used
2607	* between Intel and AMD. See @bugref{6208#c39}.
2608	* ASSUME that it's normally correct and that we're in 32-bit or 64-bit mode.
2609	*/
2610	HMSVM_SEG_REG_COPY_FROM_VMCB(pMixedCtx, &pVmcb->guest, TR, tr);
2611	if (pMixedCtx->tr.Attr.n.u4Type != X86_SEL_TYPE_SYS_386_TSS_BUSY)
2612	{
2613	if ( pMixedCtx->tr.Attr.n.u4Type == X86_SEL_TYPE_SYS_386_TSS_AVAIL
2614	\|\| CPUMIsGuestInLongModeEx(pMixedCtx))
2615	pMixedCtx->tr.Attr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
2616	else if (pMixedCtx->tr.Attr.n.u4Type == X86_SEL_TYPE_SYS_286_TSS_AVAIL)
2617	pMixedCtx->tr.Attr.n.u4Type = X86_SEL_TYPE_SYS_286_TSS_BUSY;
2618	}
2619
2620	/*
2621	* Guest Descriptor-Table registers (GDTR, IDTR, LDTR).
2622	*/
2623	HMSVM_SEG_REG_COPY_FROM_VMCB(pMixedCtx, &pVmcb->guest, LDTR, ldtr);
2624	pMixedCtx->gdtr.cbGdt = pVmcb->guest.GDTR.u32Limit;
2625	pMixedCtx->gdtr.pGdt = pVmcb->guest.GDTR.u64Base;
2626
2627	pMixedCtx->idtr.cbIdt = pVmcb->guest.IDTR.u32Limit;
2628	pMixedCtx->idtr.pIdt = pVmcb->guest.IDTR.u64Base;
2629
2630	/*
2631	* Guest Debug registers.
2632	*/
2633	if (!pVCpu->hm.s.fUsingHyperDR7)
2634	{
2635	pMixedCtx->dr[6] = pVmcb->guest.u64DR6;
2636	pMixedCtx->dr[7] = pVmcb->guest.u64DR7;
2637	}
2638	else
2639	{
2640	Assert(pVmcb->guest.u64DR7 == CPUMGetHyperDR7(pVCpu));
2641	CPUMSetHyperDR6(pVCpu, pVmcb->guest.u64DR6);
2642	}
2643
2644	/*
2645	* With Nested Paging, CR3 changes are not intercepted. Therefore, sync. it now.
2646	* This is done as the very last step of syncing the guest state, as PGMUpdateCR3() may cause longjmp's to ring-3.
2647	*/
2648	if ( pVmcb->ctrl.NestedPaging.n.u1NestedPaging
2649	&& pMixedCtx->cr3 != pVmcb->guest.u64CR3)
2650	{
2651	CPUMSetGuestCR3(pVCpu, pVmcb->guest.u64CR3);
2652	PGMUpdateCR3(pVCpu, pVmcb->guest.u64CR3);
2653	}
2654
2655	#ifdef VBOX_STRICT
2656	if (CPUMIsGuestInSvmNestedHwVirtMode(pMixedCtx))
2657	hmR0SvmLogState(pVCpu, pVmcb, pMixedCtx, "hmR0SvmSaveGuestStateNested", HMSVM_LOG_ALL & ~HMSVM_LOG_LBR, 0 /* uVerbose */);
2658	#endif
2659	}
2660
2661
2662	/**
2663	* Does the necessary state syncing before returning to ring-3 for any reason
2664	* (longjmp, preemption, voluntary exits to ring-3) from AMD-V.
2665	*
2666	* @param pVCpu The cross context virtual CPU structure.
2667	*
2668	* @remarks No-long-jmp zone!!!
2669	*/
2670	static void hmR0SvmLeave(PVMCPU pVCpu)
2671	{
2672	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2673	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2674	Assert(VMMR0IsLogFlushDisabled(pVCpu));
2675
2676	/*
2677	* !!! IMPORTANT !!!
2678	* If you modify code here, make sure to check whether hmR0SvmCallRing3Callback() needs to be updated too.
2679	*/
2680
2681	/* Restore host FPU state if necessary and resync on next R0 reentry .*/
2682	if (CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu))
2683	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0); /** @todo r=ramshankar: This shouldn't be necessary, it's set in HMR0EnterCpu. */
2684
2685	/*
2686	* Restore host debug registers if necessary and resync on next R0 reentry.
2687	*/
2688	#ifdef VBOX_STRICT
2689	if (CPUMIsHyperDebugStateActive(pVCpu))
2690	{
2691	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb; /** @todo nested-guest. */
2692	Assert(pVmcb->ctrl.u16InterceptRdDRx == 0xffff);
2693	Assert(pVmcb->ctrl.u16InterceptWrDRx == 0xffff);
2694	}
2695	#endif
2696	if (CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, false /* save DR6 */))
2697	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);/** @todo r=ramshankar: This shouldn't be necessary, it's set in HMR0EnterCpu. */
2698
2699	Assert(!CPUMIsHyperDebugStateActive(pVCpu));
2700	Assert(!CPUMIsGuestDebugStateActive(pVCpu));
2701
2702	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry);
2703	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatLoadGuestState);
2704	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExit1);
2705	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExit2);
2706	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
2707
2708	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
2709	}
2710
2711
2712	/**
2713	* Leaves the AMD-V session.
2714	*
2715	* @returns VBox status code.
2716	* @param pVCpu The cross context virtual CPU structure.
2717	*/
2718	static int hmR0SvmLeaveSession(PVMCPU pVCpu)
2719	{
2720	HM_DISABLE_PREEMPT();
2721	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2722	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2723
2724	/* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before
2725	and done this from the SVMR0ThreadCtxCallback(). */
2726	if (!pVCpu->hm.s.fLeaveDone)
2727	{
2728	hmR0SvmLeave(pVCpu);
2729	pVCpu->hm.s.fLeaveDone = true;
2730	}
2731
2732	/*
2733	* !!! IMPORTANT !!!
2734	* If you modify code here, make sure to check whether hmR0SvmCallRing3Callback() needs to be updated too.
2735	*/
2736
2737	/** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
2738	/* Deregister hook now that we've left HM context before re-enabling preemption. */
2739	VMMR0ThreadCtxHookDisable(pVCpu);
2740
2741	/* Leave HM context. This takes care of local init (term). */
2742	int rc = HMR0LeaveCpu(pVCpu);
2743
2744	HM_RESTORE_PREEMPT();
2745	return rc;
2746	}
2747
2748
2749	/**
2750	* Does the necessary state syncing before doing a longjmp to ring-3.
2751	*
2752	* @returns VBox status code.
2753	* @param pVCpu The cross context virtual CPU structure.
2754	*
2755	* @remarks No-long-jmp zone!!!
2756	*/
2757	static int hmR0SvmLongJmpToRing3(PVMCPU pVCpu)
2758	{
2759	return hmR0SvmLeaveSession(pVCpu);
2760	}
2761
2762
2763	/**
2764	* VMMRZCallRing3() callback wrapper which saves the guest state (or restores
2765	* any remaining host state) before we longjump to ring-3 and possibly get
2766	* preempted.
2767	*
2768	* @param pVCpu The cross context virtual CPU structure.
2769	* @param enmOperation The operation causing the ring-3 longjump.
2770	* @param pvUser The user argument (pointer to the possibly
2771	* out-of-date guest-CPU context).
2772	*/
2773	static DECLCALLBACK(int) hmR0SvmCallRing3Callback(PVMCPU pVCpu, VMMCALLRING3 enmOperation, void *pvUser)
2774	{
2775	RT_NOREF_PV(pvUser);
2776
2777	if (enmOperation == VMMCALLRING3_VM_R0_ASSERTION)
2778	{
2779	/*
2780	* !!! IMPORTANT !!!
2781	* If you modify code here, make sure to check whether hmR0SvmLeave() and hmR0SvmLeaveSession() needs
2782	* to be updated too. This is a stripped down version which gets out ASAP trying to not trigger any assertion.
2783	*/
2784	VMMRZCallRing3RemoveNotification(pVCpu);
2785	VMMRZCallRing3Disable(pVCpu);
2786	HM_DISABLE_PREEMPT();
2787
2788	/* Restore host FPU state if necessary and resync on next R0 reentry. */
2789	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
2790
2791	/* Restore host debug registers if necessary and resync on next R0 reentry. */
2792	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, false /* save DR6 */);
2793
2794	/* Deregister the hook now that we've left HM context before re-enabling preemption. */
2795	/** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
2796	VMMR0ThreadCtxHookDisable(pVCpu);
2797
2798	/* Leave HM context. This takes care of local init (term). */
2799	HMR0LeaveCpu(pVCpu);
2800
2801	HM_RESTORE_PREEMPT();
2802	return VINF_SUCCESS;
2803	}
2804
2805	Assert(pVCpu);
2806	Assert(pvUser);
2807	Assert(VMMRZCallRing3IsEnabled(pVCpu));
2808	HMSVM_ASSERT_PREEMPT_SAFE();
2809
2810	VMMRZCallRing3Disable(pVCpu);
2811	Assert(VMMR0IsLogFlushDisabled(pVCpu));
2812
2813	Log4(("hmR0SvmCallRing3Callback->hmR0SvmLongJmpToRing3\n"));
2814	int rc = hmR0SvmLongJmpToRing3(pVCpu);
2815	AssertRCReturn(rc, rc);
2816
2817	VMMRZCallRing3Enable(pVCpu);
2818	return VINF_SUCCESS;
2819	}
2820
2821
2822	/**
2823	* Take necessary actions before going back to ring-3.
2824	*
2825	* An action requires us to go back to ring-3. This function does the necessary
2826	* steps before we can safely return to ring-3. This is not the same as longjmps
2827	* to ring-3, this is voluntary.
2828	*
2829	* @returns VBox status code.
2830	* @param pVM The cross context VM structure.
2831	* @param pVCpu The cross context virtual CPU structure.
2832	* @param pCtx Pointer to the guest-CPU context.
2833	* @param rcExit The reason for exiting to ring-3. Can be
2834	* VINF_VMM_UNKNOWN_RING3_CALL.
2835	*/
2836	static int hmR0SvmExitToRing3(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, int rcExit)
2837	{
2838	Assert(pVM);
2839	Assert(pVCpu);
2840	Assert(pCtx);
2841	HMSVM_ASSERT_PREEMPT_SAFE();
2842
2843	/* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */
2844	VMMRZCallRing3Disable(pVCpu);
2845	Log4(("hmR0SvmExitToRing3: VCPU[%u]: rcExit=%d LocalFF=%#RX32 GlobalFF=%#RX32\n", pVCpu->idCpu, rcExit,
2846	pVCpu->fLocalForcedActions, pVM->fGlobalForcedActions));
2847
2848	/* We need to do this only while truly exiting the "inner loop" back to ring-3 and -not- for any longjmp to ring3. */
2849	if (pVCpu->hm.s.Event.fPending)
2850	{
2851	hmR0SvmPendingEventToTrpmTrap(pVCpu);
2852	Assert(!pVCpu->hm.s.Event.fPending);
2853	}
2854
2855	/* Sync. the necessary state for going back to ring-3. */
2856	hmR0SvmLeaveSession(pVCpu);
2857	STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
2858
2859	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3);
2860	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR
2861	\| CPUM_CHANGED_LDTR
2862	\| CPUM_CHANGED_GDTR
2863	\| CPUM_CHANGED_IDTR
2864	\| CPUM_CHANGED_TR
2865	\| CPUM_CHANGED_HIDDEN_SEL_REGS);
2866	if ( pVM->hm.s.fNestedPaging
2867	&& CPUMIsGuestPagingEnabledEx(pCtx))
2868	{
2869	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH);
2870	}
2871
2872	/* On our way back from ring-3 reload the guest state if there is a possibility of it being changed. */
2873	if (rcExit != VINF_EM_RAW_INTERRUPT)
2874	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
2875
2876	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3);
2877
2878	/* We do -not- want any longjmp notifications after this! We must return to ring-3 ASAP. */
2879	VMMRZCallRing3RemoveNotification(pVCpu);
2880	VMMRZCallRing3Enable(pVCpu);
2881
2882	/*
2883	* If we're emulating an instruction, we shouldn't have any TRPM traps pending
2884	* and if we're injecting an event we should have a TRPM trap pending.
2885	*/
2886	AssertReturnStmt(rcExit != VINF_EM_RAW_INJECT_TRPM_EVENT \|\| TRPMHasTrap(pVCpu),
2887	pVCpu->hm.s.u32HMError = rcExit,
2888	VERR_SVM_IPE_5);
2889	AssertReturnStmt(rcExit != VINF_EM_RAW_EMULATE_INSTR \|\| !TRPMHasTrap(pVCpu),
2890	pVCpu->hm.s.u32HMError = rcExit,
2891	VERR_SVM_IPE_4);
2892
2893	return rcExit;
2894	}
2895
2896
2897	#ifdef VBOX_WITH_NESTED_HWVIRT
2898	/**
2899	* Updates the use of TSC offsetting mode for the CPU and adjusts the necessary
2900	* intercepts for the nested-guest.
2901	*
2902	* @param pVM The cross context VM structure.
2903	* @param pVCpu The cross context virtual CPU structure.
2904	* @param pCtx Pointer to the nested guest-CPU context.
2905	* @param pVmcbNstGst Pointer to the nested-guest VM control block.
2906	*
2907	* @remarks No-long-jump zone!!!
2908	*/
2909	static void hmR0SvmUpdateTscOffsettingNested(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, PSVMVMCB pVmcbNstGst)
2910	{
2911	Assert(CPUMIsGuestInSvmNestedHwVirtMode(pCtx));
2912
2913	bool fParavirtTsc;
2914	uint64_t uTscOffset;
2915	bool fCanUseRealTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &uTscOffset, &fParavirtTsc);
2916
2917	PCSVMNESTEDVMCBCACHE pVmcbNstGstCache = &pVCpu->hm.s.svm.NstGstVmcbCache;
2918	PSVMVMCBCTRL pVmcbNstGstCtrl = &pVmcbNstGst->ctrl;
2919
2920	/*
2921	* Only avoid intercepting if we determined the host TSC (++) is stable enough
2922	* to not intercept -and- the nested-hypervisor itself does not want to intercept it.
2923	*/
2924	if ( fCanUseRealTsc
2925	&& !(pVmcbNstGstCache->u64InterceptCtrl & (SVM_CTRL_INTERCEPT_RDTSC \| SVM_CTRL_INTERCEPT_RDTSCP)))
2926	{
2927	pVmcbNstGstCtrl->u64InterceptCtrl &= ~SVM_CTRL_INTERCEPT_RDTSC;
2928	pVmcbNstGstCtrl->u64InterceptCtrl &= ~SVM_CTRL_INTERCEPT_RDTSCP;
2929	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
2930	}
2931	else
2932	{
2933	pVmcbNstGstCtrl->u64InterceptCtrl \|= SVM_CTRL_INTERCEPT_RDTSC;
2934	pVmcbNstGstCtrl->u64InterceptCtrl \|= SVM_CTRL_INTERCEPT_RDTSCP;
2935	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
2936	}
2937
2938	/* Apply the nested-guest VMCB's TSC offset over the guest one. */
2939	uTscOffset = CPUMApplyNestedGuestTscOffset(pVCpu, uTscOffset);
2940
2941	/* Update the nested-guest VMCB with the combined TSC offset (of guest and nested-guest). */
2942	pVmcbNstGstCtrl->u64TSCOffset = uTscOffset;
2943
2944	/* Finally update the VMCB clean bits since we touched the intercepts as well as the TSC offset. */
2945	pVmcbNstGstCtrl->u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
2946
2947	if (fParavirtTsc)
2948	{
2949	/* Currently neither Hyper-V nor KVM need to update their paravirt. TSC
2950	information before every VM-entry, hence disable it for performance sake. */
2951	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
2952	}
2953	}
2954	#endif
2955
2956
2957	/**
2958	* Updates the use of TSC offsetting mode for the CPU and adjusts the necessary
2959	* intercepts.
2960	*
2961	* @param pVM The cross context VM structure.
2962	* @param pVCpu The cross context virtual CPU structure.
2963	* @param pVmcb Pointer to the VM control block.
2964	*
2965	* @remarks No-long-jump zone!!!
2966	*/
2967	static void hmR0SvmUpdateTscOffsetting(PVM pVM, PVMCPU pVCpu, PSVMVMCB pVmcb)
2968	{
2969	bool fParavirtTsc;
2970	bool fCanUseRealTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &pVmcb->ctrl.u64TSCOffset, &fParavirtTsc);
2971	if (fCanUseRealTsc)
2972	{
2973	pVmcb->ctrl.u64InterceptCtrl &= ~SVM_CTRL_INTERCEPT_RDTSC;
2974	pVmcb->ctrl.u64InterceptCtrl &= ~SVM_CTRL_INTERCEPT_RDTSCP;
2975	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
2976	}
2977	else
2978	{
2979	pVmcb->ctrl.u64InterceptCtrl \|= SVM_CTRL_INTERCEPT_RDTSC;
2980	pVmcb->ctrl.u64InterceptCtrl \|= SVM_CTRL_INTERCEPT_RDTSCP;
2981	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
2982	}
2983	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
2984
2985	/** @todo later optimize this to be done elsewhere and not before every
2986	* VM-entry. */
2987	if (fParavirtTsc)
2988	{
2989	/* Currently neither Hyper-V nor KVM need to update their paravirt. TSC
2990	information before every VM-entry, hence disable it for performance sake. */
2991	#if 0
2992	int rc = GIMR0UpdateParavirtTsc(pVM, 0 /* u64Offset */);
2993	AssertRC(rc);
2994	#endif
2995	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
2996	}
2997	}
2998
2999
3000	/**
3001	* Sets an event as a pending event to be injected into the guest.
3002	*
3003	* @param pVCpu The cross context virtual CPU structure.
3004	* @param pEvent Pointer to the SVM event.
3005	* @param GCPtrFaultAddress The fault-address (CR2) in case it's a
3006	* page-fault.
3007	*
3008	* @remarks Statistics counter assumes this is a guest event being reflected to
3009	* the guest i.e. 'StatInjectPendingReflect' is incremented always.
3010	*/
3011	DECLINLINE(void) hmR0SvmSetPendingEvent(PVMCPU pVCpu, PSVMEVENT pEvent, RTGCUINTPTR GCPtrFaultAddress)
3012	{
3013	Assert(!pVCpu->hm.s.Event.fPending);
3014	Assert(pEvent->n.u1Valid);
3015
3016	pVCpu->hm.s.Event.u64IntInfo = pEvent->u;
3017	pVCpu->hm.s.Event.fPending = true;
3018	pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress;
3019
3020	Log4(("hmR0SvmSetPendingEvent: u=%#RX64 u8Vector=%#x Type=%#x ErrorCodeValid=%RTbool ErrorCode=%#RX32\n", pEvent->u,
3021	pEvent->n.u8Vector, (uint8_t)pEvent->n.u3Type, !!pEvent->n.u1ErrorCodeValid, pEvent->n.u32ErrorCode));
3022	}
3023
3024
3025	/**
3026	* Sets an invalid-opcode (\#UD) exception as pending-for-injection into the VM.
3027	*
3028	* @param pVCpu The cross context virtual CPU structure.
3029	*/
3030	DECLINLINE(void) hmR0SvmSetPendingXcptUD(PVMCPU pVCpu)
3031	{
3032	SVMEVENT Event;
3033	Event.u = 0;
3034	Event.n.u1Valid = 1;
3035	Event.n.u3Type = SVM_EVENT_EXCEPTION;
3036	Event.n.u8Vector = X86_XCPT_UD;
3037	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3038	}
3039
3040
3041	/**
3042	* Sets a debug (\#DB) exception as pending-for-injection into the VM.
3043	*
3044	* @param pVCpu The cross context virtual CPU structure.
3045	*/
3046	DECLINLINE(void) hmR0SvmSetPendingXcptDB(PVMCPU pVCpu)
3047	{
3048	SVMEVENT Event;
3049	Event.u = 0;
3050	Event.n.u1Valid = 1;
3051	Event.n.u3Type = SVM_EVENT_EXCEPTION;
3052	Event.n.u8Vector = X86_XCPT_DB;
3053	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3054	}
3055
3056
3057	/**
3058	* Sets a page fault (\#PF) exception as pending-for-injection into the VM.
3059	*
3060	* @param pVCpu The cross context virtual CPU structure.
3061	* @param pCtx Pointer to the guest-CPU context.
3062	* @param u32ErrCode The error-code for the page-fault.
3063	* @param uFaultAddress The page fault address (CR2).
3064	*
3065	* @remarks This updates the guest CR2 with @a uFaultAddress!
3066	*/
3067	DECLINLINE(void) hmR0SvmSetPendingXcptPF(PVMCPU pVCpu, PCPUMCTX pCtx, uint32_t u32ErrCode, RTGCUINTPTR uFaultAddress)
3068	{
3069	SVMEVENT Event;
3070	Event.u = 0;
3071	Event.n.u1Valid = 1;
3072	Event.n.u3Type = SVM_EVENT_EXCEPTION;
3073	Event.n.u8Vector = X86_XCPT_PF;
3074	Event.n.u1ErrorCodeValid = 1;
3075	Event.n.u32ErrorCode = u32ErrCode;
3076
3077	/* Update CR2 of the guest. */
3078	if (pCtx->cr2 != uFaultAddress)
3079	{
3080	pCtx->cr2 = uFaultAddress;
3081	/* The VMCB clean bit for CR2 will be updated while re-loading the guest state. */
3082	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR2);
3083	}
3084
3085	hmR0SvmSetPendingEvent(pVCpu, &Event, uFaultAddress);
3086	}
3087
3088
3089	/**
3090	* Sets a device-not-available (\#NM) exception as pending-for-injection into
3091	* the VM.
3092	*
3093	* @param pVCpu The cross context virtual CPU structure.
3094	*/
3095	DECLINLINE(void) hmR0SvmSetPendingXcptNM(PVMCPU pVCpu)
3096	{
3097	SVMEVENT Event;
3098	Event.u = 0;
3099	Event.n.u1Valid = 1;
3100	Event.n.u3Type = SVM_EVENT_EXCEPTION;
3101	Event.n.u8Vector = X86_XCPT_NM;
3102	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3103	}
3104
3105
3106	/**
3107	* Sets a math-fault (\#MF) exception as pending-for-injection into the VM.
3108	*
3109	* @param pVCpu The cross context virtual CPU structure.
3110	*/
3111	DECLINLINE(void) hmR0SvmSetPendingXcptMF(PVMCPU pVCpu)
3112	{
3113	SVMEVENT Event;
3114	Event.u = 0;
3115	Event.n.u1Valid = 1;
3116	Event.n.u3Type = SVM_EVENT_EXCEPTION;
3117	Event.n.u8Vector = X86_XCPT_MF;
3118	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3119	}
3120
3121
3122	/**
3123	* Sets a double fault (\#DF) exception as pending-for-injection into the VM.
3124	*
3125	* @param pVCpu The cross context virtual CPU structure.
3126	*/
3127	DECLINLINE(void) hmR0SvmSetPendingXcptDF(PVMCPU pVCpu)
3128	{
3129	SVMEVENT Event;
3130	Event.u = 0;
3131	Event.n.u1Valid = 1;
3132	Event.n.u3Type = SVM_EVENT_EXCEPTION;
3133	Event.n.u8Vector = X86_XCPT_DF;
3134	Event.n.u1ErrorCodeValid = 1;
3135	Event.n.u32ErrorCode = 0;
3136	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3137	}
3138
3139
3140	/**
3141	* Injects an event into the guest upon VMRUN by updating the relevant field
3142	* in the VMCB.
3143	*
3144	* @param pVCpu The cross context virtual CPU structure.
3145	* @param pVmcb Pointer to the guest VM control block.
3146	* @param pCtx Pointer to the guest-CPU context.
3147	* @param pEvent Pointer to the event.
3148	*
3149	* @remarks No-long-jump zone!!!
3150	* @remarks Requires CR0!
3151	*/
3152	DECLINLINE(void) hmR0SvmInjectEventVmcb(PVMCPU pVCpu, PSVMVMCB pVmcb, PCPUMCTX pCtx, PSVMEVENT pEvent)
3153	{
3154	NOREF(pVCpu); NOREF(pCtx);
3155
3156	Assert(!pVmcb->ctrl.EventInject.n.u1Valid);
3157	pVmcb->ctrl.EventInject.u = pEvent->u;
3158	STAM_COUNTER_INC(&pVCpu->hm.s.paStatInjectedIrqsR0[pEvent->n.u8Vector & MASK_INJECT_IRQ_STAT]);
3159
3160	Log4(("hmR0SvmInjectEventVmcb: u=%#RX64 u8Vector=%#x Type=%#x ErrorCodeValid=%RTbool ErrorCode=%#RX32\n", pEvent->u,
3161	pEvent->n.u8Vector, (uint8_t)pEvent->n.u3Type, !!pEvent->n.u1ErrorCodeValid, pEvent->n.u32ErrorCode));
3162	}
3163
3164
3165
3166	/**
3167	* Converts any TRPM trap into a pending HM event. This is typically used when
3168	* entering from ring-3 (not longjmp returns).
3169	*
3170	* @param pVCpu The cross context virtual CPU structure.
3171	*/
3172	static void hmR0SvmTrpmTrapToPendingEvent(PVMCPU pVCpu)
3173	{
3174	Assert(TRPMHasTrap(pVCpu));
3175	Assert(!pVCpu->hm.s.Event.fPending);
3176
3177	uint8_t uVector;
3178	TRPMEVENT enmTrpmEvent;
3179	RTGCUINT uErrCode;
3180	RTGCUINTPTR GCPtrFaultAddress;
3181	uint8_t cbInstr;
3182
3183	int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr);
3184	AssertRC(rc);
3185
3186	SVMEVENT Event;
3187	Event.u = 0;
3188	Event.n.u1Valid = 1;
3189	Event.n.u8Vector = uVector;
3190
3191	/* Refer AMD spec. 15.20 "Event Injection" for the format. */
3192	if (enmTrpmEvent == TRPM_TRAP)
3193	{
3194	Event.n.u3Type = SVM_EVENT_EXCEPTION;
3195	switch (uVector)
3196	{
3197	case X86_XCPT_NMI:
3198	{
3199	Event.n.u3Type = SVM_EVENT_NMI;
3200	break;
3201	}
3202
3203	case X86_XCPT_PF:
3204	case X86_XCPT_DF:
3205	case X86_XCPT_TS:
3206	case X86_XCPT_NP:
3207	case X86_XCPT_SS:
3208	case X86_XCPT_GP:
3209	case X86_XCPT_AC:
3210	{
3211	Event.n.u1ErrorCodeValid = 1;
3212	Event.n.u32ErrorCode = uErrCode;
3213	break;
3214	}
3215	}
3216	}
3217	else if (enmTrpmEvent == TRPM_HARDWARE_INT)
3218	Event.n.u3Type = SVM_EVENT_EXTERNAL_IRQ;
3219	else if (enmTrpmEvent == TRPM_SOFTWARE_INT)
3220	Event.n.u3Type = SVM_EVENT_SOFTWARE_INT;
3221	else
3222	AssertMsgFailed(("Invalid TRPM event type %d\n", enmTrpmEvent));
3223
3224	rc = TRPMResetTrap(pVCpu);
3225	AssertRC(rc);
3226
3227	Log4(("TRPM->HM event: u=%#RX64 u8Vector=%#x uErrorCodeValid=%RTbool uErrorCode=%#RX32\n", Event.u, Event.n.u8Vector,
3228	!!Event.n.u1ErrorCodeValid, Event.n.u32ErrorCode));
3229
3230	hmR0SvmSetPendingEvent(pVCpu, &Event, GCPtrFaultAddress);
3231	}
3232
3233
3234	/**
3235	* Converts any pending SVM event into a TRPM trap. Typically used when leaving
3236	* AMD-V to execute any instruction.
3237	*
3238	* @param pVCpu The cross context virtual CPU structure.
3239	*/
3240	static void hmR0SvmPendingEventToTrpmTrap(PVMCPU pVCpu)
3241	{
3242	Assert(pVCpu->hm.s.Event.fPending);
3243	Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo /, NULL / pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP);
3244
3245	SVMEVENT Event;
3246	Event.u = pVCpu->hm.s.Event.u64IntInfo;
3247
3248	uint8_t uVector = Event.n.u8Vector;
3249	uint8_t uVectorType = Event.n.u3Type;
3250	TRPMEVENT enmTrapType = HMSvmEventToTrpmEventType(&Event);
3251
3252	Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, uVectorType));
3253
3254	int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType);
3255	AssertRC(rc);
3256
3257	if (Event.n.u1ErrorCodeValid)
3258	TRPMSetErrorCode(pVCpu, Event.n.u32ErrorCode);
3259
3260	if ( uVectorType == SVM_EVENT_EXCEPTION
3261	&& uVector == X86_XCPT_PF)
3262	{
3263	TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress);
3264	Assert(pVCpu->hm.s.Event.GCPtrFaultAddress == CPUMGetGuestCR2(pVCpu));
3265	}
3266	else if (uVectorType == SVM_EVENT_SOFTWARE_INT)
3267	{
3268	AssertMsg( uVectorType == SVM_EVENT_SOFTWARE_INT
3269	\|\| (uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF),
3270	("Invalid vector: uVector=%#x uVectorType=%#x\n", uVector, uVectorType));
3271	TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr);
3272	}
3273	pVCpu->hm.s.Event.fPending = false;
3274	}
3275
3276
3277	/**
3278	* Checks if the guest (or nested-guest) has an interrupt shadow active right
3279	* now.
3280	*
3281	* @returns @c true if the interrupt shadow is active, @c false otherwise.
3282	* @param pVCpu The cross context virtual CPU structure.
3283	* @param pCtx Pointer to the guest-CPU context.
3284	*
3285	* @remarks No-long-jump zone!!!
3286	* @remarks Has side-effects with VMCPU_FF_INHIBIT_INTERRUPTS force-flag.
3287	*/
3288	DECLINLINE(bool) hmR0SvmIsIntrShadowActive(PVMCPU pVCpu, PCPUMCTX pCtx)
3289	{
3290	/*
3291	* Instructions like STI and MOV SS inhibit interrupts till the next instruction completes. Check if we should
3292	* inhibit interrupts or clear any existing interrupt-inhibition.
3293	*/
3294	if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
3295	{
3296	if (pCtx->rip != EMGetInhibitInterruptsPC(pVCpu))
3297	{
3298	/*
3299	* We can clear the inhibit force flag as even if we go back to the recompiler without executing guest code in
3300	* AMD-V, the flag's condition to be cleared is met and thus the cleared state is correct.
3301	*/
3302	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
3303	return false;
3304	}
3305	return true;
3306	}
3307	return false;
3308	}
3309
3310
3311	/**
3312	* Sets the virtual interrupt intercept control in the VMCB.
3313	*
3314	* @param pVmcb Pointer to the VM control block.
3315	*/
3316	DECLINLINE(void) hmR0SvmSetVirtIntrIntercept(PSVMVMCB pVmcb)
3317	{
3318	/*
3319	* When AVIC isn't supported, indicate that a virtual interrupt is pending and to
3320	* cause a #VMEXIT when the guest is ready to accept interrupts. At #VMEXIT, we
3321	* then get the interrupt from the APIC (updating ISR at the right time) and
3322	* inject the interrupt.
3323	*
3324	* With AVIC is supported, we could make use of the asynchronously delivery without
3325	* #VMEXIT and we would be passing the AVIC page to SVM.
3326	*/
3327	if (!(pVmcb->ctrl.u64InterceptCtrl & SVM_CTRL_INTERCEPT_VINTR))
3328	{
3329	Assert(pVmcb->ctrl.IntCtrl.n.u1VIrqPending == 0);
3330	pVmcb->ctrl.IntCtrl.n.u1VIrqPending = 1;
3331	pVmcb->ctrl.u64InterceptCtrl \|= SVM_CTRL_INTERCEPT_VINTR;
3332	pVmcb->ctrl.u32VmcbCleanBits &= ~(HMSVM_VMCB_CLEAN_INTERCEPTS \| HMSVM_VMCB_CLEAN_TPR);
3333	Log4(("Set VINTR intercept\n"));
3334	}
3335	}
3336
3337
3338	/**
3339	* Clears the virtual interrupt intercept control in the VMCB as
3340	* we are figured the guest is unable process any interrupts
3341	* at this point of time.
3342	*
3343	* @param pVmcb Pointer to the VM control block.
3344	*/
3345	DECLINLINE(void) hmR0SvmClearVirtIntrIntercept(PSVMVMCB pVmcb)
3346	{
3347	if (pVmcb->ctrl.u64InterceptCtrl & SVM_CTRL_INTERCEPT_VINTR)
3348	{
3349	Assert(pVmcb->ctrl.IntCtrl.n.u1VIrqPending == 1);
3350	pVmcb->ctrl.IntCtrl.n.u1VIrqPending = 0;
3351	pVmcb->ctrl.u64InterceptCtrl &= ~SVM_CTRL_INTERCEPT_VINTR;
3352	pVmcb->ctrl.u32VmcbCleanBits &= ~(HMSVM_VMCB_CLEAN_INTERCEPTS \| HMSVM_VMCB_CLEAN_TPR);
3353	Log4(("Cleared VINTR intercept\n"));
3354	}
3355	}
3356
3357
3358	/**
3359	* Sets the IRET intercept control in the VMCB which instructs AMD-V to cause a
3360	* \#VMEXIT as soon as a guest starts executing an IRET. This is used to unblock
3361	* virtual NMIs.
3362	*
3363	* @param pVmcb Pointer to the VM control block.
3364	*/
3365	DECLINLINE(void) hmR0SvmSetIretIntercept(PSVMVMCB pVmcb)
3366	{
3367	if (!(pVmcb->ctrl.u64InterceptCtrl & SVM_CTRL_INTERCEPT_IRET))
3368	{
3369	pVmcb->ctrl.u64InterceptCtrl \|= SVM_CTRL_INTERCEPT_IRET;
3370	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
3371
3372	Log4(("Setting IRET intercept\n"));
3373	}
3374	}
3375
3376
3377	/**
3378	* Clears the IRET intercept control in the VMCB.
3379	*
3380	* @param pVmcb Pointer to the VM control block.
3381	*/
3382	DECLINLINE(void) hmR0SvmClearIretIntercept(PSVMVMCB pVmcb)
3383	{
3384	if (pVmcb->ctrl.u64InterceptCtrl & SVM_CTRL_INTERCEPT_IRET)
3385	{
3386	pVmcb->ctrl.u64InterceptCtrl &= ~SVM_CTRL_INTERCEPT_IRET;
3387	pVmcb->ctrl.u32VmcbCleanBits &= ~(HMSVM_VMCB_CLEAN_INTERCEPTS);
3388
3389	Log4(("Clearing IRET intercept\n"));
3390	}
3391	}
3392
3393	#ifdef VBOX_WITH_NESTED_HWVIRT
3394
3395
3396	/**
3397	* Evaluates the event to be delivered to the nested-guest and sets it as the
3398	* pending event.
3399	*
3400	* @returns VBox strict status code.
3401	* @param pVCpu The cross context virtual CPU structure.
3402	* @param pCtx Pointer to the guest-CPU context.
3403	*/
3404	static VBOXSTRICTRC hmR0SvmEvaluatePendingEventNested(PVMCPU pVCpu, PCPUMCTX pCtx)
3405	{
3406	Log4Func(("\n"));
3407
3408	Assert(!pVCpu->hm.s.Event.fPending);
3409
3410	bool const fGif = pCtx->hwvirt.svm.fGif;
3411	if (fGif)
3412	{
3413	PSVMVMCB pVmcbNstGst = pCtx->hwvirt.svm.CTX_SUFF(pVmcb);
3414
3415	bool const fIntShadow = hmR0SvmIsIntrShadowActive(pVCpu, pCtx);
3416
3417	/*
3418	* Check if the nested-guest can receive NMIs.
3419	* NMIs are higher priority than regular interrupts.
3420	*/
3421	/** @todo SMI. SMIs take priority over NMIs. */
3422	if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INTERRUPT_NMI))
3423	{
3424	bool const fBlockNmi = VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS);
3425	if (fBlockNmi)
3426	hmR0SvmSetIretIntercept(pVmcbNstGst);
3427	else if (fIntShadow)
3428	{
3429	/** @todo Figure this out, how we shall manage virt. intercept if the
3430	* nested-guest already has one set and/or if we really need it? */
3431	//hmR0SvmSetVirtIntrIntercept(pVmcbNstGst);
3432	}
3433	else
3434	{
3435	Log4(("Pending NMI\n"));
3436
3437	SVMEVENT Event;
3438	Event.u = 0;
3439	Event.n.u1Valid = 1;
3440	Event.n.u8Vector = X86_XCPT_NMI;
3441	Event.n.u3Type = SVM_EVENT_NMI;
3442
3443	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3444	hmR0SvmSetIretIntercept(pVmcbNstGst);
3445	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
3446	return VINF_SUCCESS;
3447	}
3448	}
3449
3450	/*
3451	* Check if the nested-guest can receive external interrupts (generated by
3452	* the guest's PIC/APIC).
3453	*
3454	* External intercepts, NMI, SMI etc. from the physical CPU are -always- intercepted
3455	* when executing using hardware-assisted SVM, see HMSVM_MANDATORY_GUEST_CTRL_INTERCEPTS.
3456	*
3457	* External interrupts that are generated for the outer guest may be intercepted
3458	* depending on how the nested-guest VMCB was programmed by guest software.
3459	*
3460	* Physical interrupts always take priority over virtual interrupts,
3461	* see AMD spec. 15.21.4 "Injecting Virtual (INTR) Interrupts".
3462	*/
3463	if (!fIntShadow)
3464	{
3465	PCSVMNESTEDVMCBCACHE pVmcbNstGstCache = &pVCpu->hm.s.svm.NstGstVmcbCache;
3466	if ( VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
3467	&& !pVCpu->hm.s.fSingleInstruction
3468	&& CPUMCanSvmNstGstTakePhysIntr(pVCpu, pCtx))
3469	{
3470	if (pVmcbNstGstCache->u64InterceptCtrl & SVM_CTRL_INTERCEPT_INTR)
3471	{
3472	Log4(("Intercepting external interrupt -> #VMEXIT\n"));
3473	return IEMExecSvmVmexit(pVCpu, SVM_EXIT_INTR, 0, 0);
3474	}
3475
3476	uint8_t u8Interrupt;
3477	int rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
3478	if (RT_SUCCESS(rc))
3479	{
3480	Log4(("Injecting external interrupt u8Interrupt=%#x\n", u8Interrupt));
3481
3482	SVMEVENT Event;
3483	Event.u = 0;
3484	Event.n.u1Valid = 1;
3485	Event.n.u8Vector = u8Interrupt;
3486	Event.n.u3Type = SVM_EVENT_EXTERNAL_IRQ;
3487
3488	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3489	}
3490	else if (rc == VERR_APIC_INTR_MASKED_BY_TPR)
3491	{
3492	/*
3493	* AMD-V has no TPR thresholding feature. TPR and the force-flag will be
3494	* updated eventually when the TPR is written by the guest.
3495	*/
3496	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchTprMaskedIrq);
3497	}
3498	else
3499	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
3500	}
3501
3502	/*
3503	* Check if the nested-guest is intercepting virtual (using V_IRQ and related fields)
3504	* interrupt injection. The virtual interrupt injection itself, if any, will be done
3505	* by the physical CPU.
3506	*/
3507	#if 0
3508	if ( VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INTERRUPT_NESTED_GUEST)
3509	&& (pVmcbNstGstCache->u64InterceptCtrl & SVM_CTRL_INTERCEPT_VINTR)
3510	&& CPUMCanSvmNstGstTakeVirtIntr(pCtx))
3511	{
3512	Log4(("Intercepting virtual interrupt -> #VMEXIT\n"));
3513	return IEMExecSvmVmexit(pVCpu, SVM_EXIT_VINTR, 0, 0);
3514	}
3515	#endif
3516	}
3517	}
3518
3519	return VINF_SUCCESS;
3520	}
3521	#endif
3522
3523
3524	/**
3525	* Evaluates the event to be delivered to the guest and sets it as the pending
3526	* event.
3527	*
3528	* @param pVCpu The cross context virtual CPU structure.
3529	* @param pCtx Pointer to the guest-CPU context.
3530	*
3531	* @remarks Don't use this function when we are actively executing a
3532	* nested-guest, use hmR0SvmEvaluatePendingEventNested instead.
3533	*/
3534	static void hmR0SvmEvaluatePendingEvent(PVMCPU pVCpu, PCPUMCTX pCtx)
3535	{
3536	HMSVM_ASSERT_NOT_IN_NESTED_GUEST(pCtx);
3537	Assert(!pVCpu->hm.s.Event.fPending);
3538
3539	#ifdef VBOX_WITH_NESTED_HWVIRT
3540	bool const fGif = pCtx->hwvirt.svm.fGif;
3541	#else
3542	bool const fGif = true;
3543	#endif
3544	Log4Func(("fGif=%RTbool\n", fGif));
3545
3546	/*
3547	* If the global interrupt flag (GIF) isn't set, even NMIs and other events are blocked.
3548	* See AMD spec. Table 15-10. "Effect of the GIF on Interrupt Handling".
3549	*/
3550	if (fGif)
3551	{
3552	bool const fIntShadow = hmR0SvmIsIntrShadowActive(pVCpu, pCtx);
3553	bool const fBlockInt = !(pCtx->eflags.u32 & X86_EFL_IF);
3554	bool const fBlockNmi = VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS);
3555	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
3556
3557	Log4Func(("fBlockInt=%RTbool fIntShadow=%RTbool APIC/PIC_Pending=%RTbool\n", fBlockInt, fIntShadow,
3558	VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)));
3559
3560	/** @todo SMI. SMIs take priority over NMIs. */
3561	if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INTERRUPT_NMI)) /* NMI. NMIs take priority over regular interrupts. */
3562	{
3563	if (fBlockNmi)
3564	hmR0SvmSetIretIntercept(pVmcb);
3565	else if (fIntShadow)
3566	hmR0SvmSetVirtIntrIntercept(pVmcb);
3567	else
3568	{
3569	Log4(("Pending NMI\n"));
3570
3571	SVMEVENT Event;
3572	Event.u = 0;
3573	Event.n.u1Valid = 1;
3574	Event.n.u8Vector = X86_XCPT_NMI;
3575	Event.n.u3Type = SVM_EVENT_NMI;
3576
3577	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3578	hmR0SvmSetIretIntercept(pVmcb);
3579	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
3580	return;
3581	}
3582	}
3583	else if ( VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
3584	&& !pVCpu->hm.s.fSingleInstruction)
3585	{
3586	/*
3587	* Check if the guest can receive external interrupts (PIC/APIC). Once PDMGetInterrupt() returns
3588	* a valid interrupt we -must- deliver the interrupt. We can no longer re-request it from the APIC.
3589	*/
3590	if ( !fBlockInt
3591	&& !fIntShadow)
3592	{
3593	uint8_t u8Interrupt;
3594	int rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
3595	if (RT_SUCCESS(rc))
3596	{
3597	Log4(("Injecting external interrupt u8Interrupt=%#x\n", u8Interrupt));
3598
3599	SVMEVENT Event;
3600	Event.u = 0;
3601	Event.n.u1Valid = 1;
3602	Event.n.u8Vector = u8Interrupt;
3603	Event.n.u3Type = SVM_EVENT_EXTERNAL_IRQ;
3604
3605	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3606	}
3607	else if (rc == VERR_APIC_INTR_MASKED_BY_TPR)
3608	{
3609	/*
3610	* AMD-V has no TPR thresholding feature. TPR and the force-flag will be
3611	* updated eventually when the TPR is written by the guest.
3612	*/
3613	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchTprMaskedIrq);
3614	}
3615	else
3616	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
3617	}
3618	else
3619	hmR0SvmSetVirtIntrIntercept(pVmcb);
3620	}
3621	}
3622	}
3623
3624
3625	/**
3626	* Injects any pending events into the guest or nested-guest.
3627	*
3628	* @param pVCpu The cross context virtual CPU structure.
3629	* @param pCtx Pointer to the guest-CPU context.
3630	* @param pVmcb Pointer to the VM control block.
3631	*/
3632	static void hmR0SvmInjectPendingEvent(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMVMCB pVmcb)
3633	{
3634	Assert(!TRPMHasTrap(pVCpu));
3635	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
3636
3637	bool const fIntShadow = hmR0SvmIsIntrShadowActive(pVCpu, pCtx);
3638	#ifdef VBOX_STRICT
3639	bool const fGif = pCtx->hwvirt.svm.fGif;
3640	bool fAllowInt = fGif;
3641	if (fGif)
3642	{
3643	if (CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
3644	fAllowInt = CPUMCanSvmNstGstTakePhysIntr(pVCpu, pCtx);
3645	else
3646	fAllowInt = RT_BOOL(pCtx->eflags.u32 & X86_EFL_IF);
3647	}
3648	#endif
3649
3650	if (pVCpu->hm.s.Event.fPending)
3651	{
3652	SVMEVENT Event;
3653	Event.u = pVCpu->hm.s.Event.u64IntInfo;
3654	Assert(Event.n.u1Valid);
3655
3656	/*
3657	* Validate event injection pre-conditions.
3658	*/
3659	if (Event.n.u3Type == SVM_EVENT_EXTERNAL_IRQ)
3660	{
3661	Assert(fAllowInt);
3662	Assert(!fIntShadow);
3663	}
3664	else if (Event.n.u3Type == SVM_EVENT_NMI)
3665	{
3666	Assert(fGif);
3667	Assert(!fIntShadow);
3668	}
3669
3670	/*
3671	* Inject it (update VMCB for injection by the hardware).
3672	*/
3673	Log4(("Injecting pending HM event\n"));
3674	hmR0SvmInjectEventVmcb(pVCpu, pVmcb, pCtx, &Event);
3675	pVCpu->hm.s.Event.fPending = false;
3676
3677	if (Event.n.u3Type == SVM_EVENT_EXTERNAL_IRQ)
3678	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt);
3679	else
3680	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt);
3681	}
3682	else
3683	{
3684	#ifdef VBOX_WITH_NESTED_HWVIRT
3685	/*
3686	* If IEM emulated VMRUN and injected an event, it would not clear the EVENTINJ::Valid bit
3687	* as a physical CPU clears it in the VMCB as part of the #VMEXIT (if the AMD spec. is to
3688	* believed, real behavior might differ). Regardless, IEM does it only on #VMEXIT for now
3689	* and since we are continuing nested-guest execution using hardware-assisted SVM, we need
3690	* to clear this field otherwise we will inject the event twice, see @bugref{7243#78}.
3691	*/
3692	if (CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
3693	pVmcb->ctrl.EventInject.n.u1Valid = 0;
3694	#endif
3695	Assert(pVmcb->ctrl.EventInject.n.u1Valid == 0);
3696	}
3697
3698	/*
3699	* Update the guest interrupt shadow in the guest or nested-guest VMCB.
3700	*
3701	* For nested-guests: We need to update it too for the scenario where IEM executes
3702	* the nested-guest but execution later continues here with an interrupt shadow active.
3703	*/
3704	pVmcb->ctrl.IntShadow.n.u1IntShadow = fIntShadow;
3705	}
3706
3707
3708	/**
3709	* Reports world-switch error and dumps some useful debug info.
3710	*
3711	* @param pVM The cross context VM structure.
3712	* @param pVCpu The cross context virtual CPU structure.
3713	* @param rcVMRun The return code from VMRUN (or
3714	* VERR_SVM_INVALID_GUEST_STATE for invalid
3715	* guest-state).
3716	* @param pCtx Pointer to the guest-CPU context.
3717	*/
3718	static void hmR0SvmReportWorldSwitchError(PVM pVM, PVMCPU pVCpu, int rcVMRun, PCPUMCTX pCtx)
3719	{
3720	NOREF(pCtx);
3721	HMSVM_ASSERT_PREEMPT_SAFE();
3722	HMSVM_ASSERT_NOT_IN_NESTED_GUEST(pCtx);
3723	PCSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
3724
3725	if (rcVMRun == VERR_SVM_INVALID_GUEST_STATE)
3726	{
3727	hmR0DumpRegs(pVM, pVCpu, pCtx); NOREF(pVM);
3728	#ifdef VBOX_STRICT
3729	Log4(("ctrl.u32VmcbCleanBits %#RX32\n", pVmcb->ctrl.u32VmcbCleanBits));
3730	Log4(("ctrl.u16InterceptRdCRx %#x\n", pVmcb->ctrl.u16InterceptRdCRx));
3731	Log4(("ctrl.u16InterceptWrCRx %#x\n", pVmcb->ctrl.u16InterceptWrCRx));
3732	Log4(("ctrl.u16InterceptRdDRx %#x\n", pVmcb->ctrl.u16InterceptRdDRx));
3733	Log4(("ctrl.u16InterceptWrDRx %#x\n", pVmcb->ctrl.u16InterceptWrDRx));
3734	Log4(("ctrl.u32InterceptXcpt %#x\n", pVmcb->ctrl.u32InterceptXcpt));
3735	Log4(("ctrl.u64InterceptCtrl %#RX64\n", pVmcb->ctrl.u64InterceptCtrl));
3736	Log4(("ctrl.u64IOPMPhysAddr %#RX64\n", pVmcb->ctrl.u64IOPMPhysAddr));
3737	Log4(("ctrl.u64MSRPMPhysAddr %#RX64\n", pVmcb->ctrl.u64MSRPMPhysAddr));
3738	Log4(("ctrl.u64TSCOffset %#RX64\n", pVmcb->ctrl.u64TSCOffset));
3739
3740	Log4(("ctrl.TLBCtrl.u32ASID %#x\n", pVmcb->ctrl.TLBCtrl.n.u32ASID));
3741	Log4(("ctrl.TLBCtrl.u8TLBFlush %#x\n", pVmcb->ctrl.TLBCtrl.n.u8TLBFlush));
3742	Log4(("ctrl.TLBCtrl.u24Reserved %#x\n", pVmcb->ctrl.TLBCtrl.n.u24Reserved));
3743
3744	Log4(("ctrl.IntCtrl.u8VTPR %#x\n", pVmcb->ctrl.IntCtrl.n.u8VTPR));
3745	Log4(("ctrl.IntCtrl.u1VIrqPending %#x\n", pVmcb->ctrl.IntCtrl.n.u1VIrqPending));
3746	Log4(("ctrl.IntCtrl.u1VGif %#x\n", pVmcb->ctrl.IntCtrl.n.u1VGif));
3747	Log4(("ctrl.IntCtrl.u6Reserved0 %#x\n", pVmcb->ctrl.IntCtrl.n.u6Reserved0));
3748	Log4(("ctrl.IntCtrl.u4VIntrPrio %#x\n", pVmcb->ctrl.IntCtrl.n.u4VIntrPrio));
3749	Log4(("ctrl.IntCtrl.u1IgnoreTPR %#x\n", pVmcb->ctrl.IntCtrl.n.u1IgnoreTPR));
3750	Log4(("ctrl.IntCtrl.u3Reserved %#x\n", pVmcb->ctrl.IntCtrl.n.u3Reserved));
3751	Log4(("ctrl.IntCtrl.u1VIntrMasking %#x\n", pVmcb->ctrl.IntCtrl.n.u1VIntrMasking));
3752	Log4(("ctrl.IntCtrl.u1VGifEnable %#x\n", pVmcb->ctrl.IntCtrl.n.u1VGifEnable));
3753	Log4(("ctrl.IntCtrl.u5Reserved1 %#x\n", pVmcb->ctrl.IntCtrl.n.u5Reserved1));
3754	Log4(("ctrl.IntCtrl.u8VIntrVector %#x\n", pVmcb->ctrl.IntCtrl.n.u8VIntrVector));
3755	Log4(("ctrl.IntCtrl.u24Reserved %#x\n", pVmcb->ctrl.IntCtrl.n.u24Reserved));
3756
3757	Log4(("ctrl.IntShadow.u1IntShadow %#x\n", pVmcb->ctrl.IntShadow.n.u1IntShadow));
3758	Log4(("ctrl.IntShadow.u1GuestIntMask %#x\n", pVmcb->ctrl.IntShadow.n.u1GuestIntMask));
3759	Log4(("ctrl.u64ExitCode %#RX64\n", pVmcb->ctrl.u64ExitCode));
3760	Log4(("ctrl.u64ExitInfo1 %#RX64\n", pVmcb->ctrl.u64ExitInfo1));
3761	Log4(("ctrl.u64ExitInfo2 %#RX64\n", pVmcb->ctrl.u64ExitInfo2));
3762	Log4(("ctrl.ExitIntInfo.u8Vector %#x\n", pVmcb->ctrl.ExitIntInfo.n.u8Vector));
3763	Log4(("ctrl.ExitIntInfo.u3Type %#x\n", pVmcb->ctrl.ExitIntInfo.n.u3Type));
3764	Log4(("ctrl.ExitIntInfo.u1ErrorCodeValid %#x\n", pVmcb->ctrl.ExitIntInfo.n.u1ErrorCodeValid));
3765	Log4(("ctrl.ExitIntInfo.u19Reserved %#x\n", pVmcb->ctrl.ExitIntInfo.n.u19Reserved));
3766	Log4(("ctrl.ExitIntInfo.u1Valid %#x\n", pVmcb->ctrl.ExitIntInfo.n.u1Valid));
3767	Log4(("ctrl.ExitIntInfo.u32ErrorCode %#x\n", pVmcb->ctrl.ExitIntInfo.n.u32ErrorCode));
3768	Log4(("ctrl.NestedPaging.u1NestedPaging %#x\n", pVmcb->ctrl.NestedPaging.n.u1NestedPaging));
3769	Log4(("ctrl.NestedPaging.u1Sev %#x\n", pVmcb->ctrl.NestedPaging.n.u1Sev));
3770	Log4(("ctrl.NestedPaging.u1SevEs %#x\n", pVmcb->ctrl.NestedPaging.n.u1SevEs));
3771	Log4(("ctrl.EventInject.u8Vector %#x\n", pVmcb->ctrl.EventInject.n.u8Vector));
3772	Log4(("ctrl.EventInject.u3Type %#x\n", pVmcb->ctrl.EventInject.n.u3Type));
3773	Log4(("ctrl.EventInject.u1ErrorCodeValid %#x\n", pVmcb->ctrl.EventInject.n.u1ErrorCodeValid));
3774	Log4(("ctrl.EventInject.u19Reserved %#x\n", pVmcb->ctrl.EventInject.n.u19Reserved));
3775	Log4(("ctrl.EventInject.u1Valid %#x\n", pVmcb->ctrl.EventInject.n.u1Valid));
3776	Log4(("ctrl.EventInject.u32ErrorCode %#x\n", pVmcb->ctrl.EventInject.n.u32ErrorCode));
3777
3778	Log4(("ctrl.u64NestedPagingCR3 %#RX64\n", pVmcb->ctrl.u64NestedPagingCR3));
3779
3780	Log4(("ctrl.LbrVirt.u1LbrVirt %#x\n", pVmcb->ctrl.LbrVirt.n.u1LbrVirt));
3781	Log4(("ctrl.LbrVirt.u1VirtVmsaveVmload %#x\n", pVmcb->ctrl.LbrVirt.n.u1VirtVmsaveVmload));
3782
3783	Log4(("guest.CS.u16Sel %RTsel\n", pVmcb->guest.CS.u16Sel));
3784	Log4(("guest.CS.u16Attr %#x\n", pVmcb->guest.CS.u16Attr));
3785	Log4(("guest.CS.u32Limit %#RX32\n", pVmcb->guest.CS.u32Limit));
3786	Log4(("guest.CS.u64Base %#RX64\n", pVmcb->guest.CS.u64Base));
3787	Log4(("guest.DS.u16Sel %#RTsel\n", pVmcb->guest.DS.u16Sel));
3788	Log4(("guest.DS.u16Attr %#x\n", pVmcb->guest.DS.u16Attr));
3789	Log4(("guest.DS.u32Limit %#RX32\n", pVmcb->guest.DS.u32Limit));
3790	Log4(("guest.DS.u64Base %#RX64\n", pVmcb->guest.DS.u64Base));
3791	Log4(("guest.ES.u16Sel %RTsel\n", pVmcb->guest.ES.u16Sel));
3792	Log4(("guest.ES.u16Attr %#x\n", pVmcb->guest.ES.u16Attr));
3793	Log4(("guest.ES.u32Limit %#RX32\n", pVmcb->guest.ES.u32Limit));
3794	Log4(("guest.ES.u64Base %#RX64\n", pVmcb->guest.ES.u64Base));
3795	Log4(("guest.FS.u16Sel %RTsel\n", pVmcb->guest.FS.u16Sel));
3796	Log4(("guest.FS.u16Attr %#x\n", pVmcb->guest.FS.u16Attr));
3797	Log4(("guest.FS.u32Limit %#RX32\n", pVmcb->guest.FS.u32Limit));
3798	Log4(("guest.FS.u64Base %#RX64\n", pVmcb->guest.FS.u64Base));
3799	Log4(("guest.GS.u16Sel %RTsel\n", pVmcb->guest.GS.u16Sel));
3800	Log4(("guest.GS.u16Attr %#x\n", pVmcb->guest.GS.u16Attr));
3801	Log4(("guest.GS.u32Limit %#RX32\n", pVmcb->guest.GS.u32Limit));
3802	Log4(("guest.GS.u64Base %#RX64\n", pVmcb->guest.GS.u64Base));
3803
3804	Log4(("guest.GDTR.u32Limit %#RX32\n", pVmcb->guest.GDTR.u32Limit));
3805	Log4(("guest.GDTR.u64Base %#RX64\n", pVmcb->guest.GDTR.u64Base));
3806
3807	Log4(("guest.LDTR.u16Sel %RTsel\n", pVmcb->guest.LDTR.u16Sel));
3808	Log4(("guest.LDTR.u16Attr %#x\n", pVmcb->guest.LDTR.u16Attr));
3809	Log4(("guest.LDTR.u32Limit %#RX32\n", pVmcb->guest.LDTR.u32Limit));
3810	Log4(("guest.LDTR.u64Base %#RX64\n", pVmcb->guest.LDTR.u64Base));
3811
3812	Log4(("guest.IDTR.u32Limit %#RX32\n", pVmcb->guest.IDTR.u32Limit));
3813	Log4(("guest.IDTR.u64Base %#RX64\n", pVmcb->guest.IDTR.u64Base));
3814
3815	Log4(("guest.TR.u16Sel %RTsel\n", pVmcb->guest.TR.u16Sel));
3816	Log4(("guest.TR.u16Attr %#x\n", pVmcb->guest.TR.u16Attr));
3817	Log4(("guest.TR.u32Limit %#RX32\n", pVmcb->guest.TR.u32Limit));
3818	Log4(("guest.TR.u64Base %#RX64\n", pVmcb->guest.TR.u64Base));
3819
3820	Log4(("guest.u8CPL %#x\n", pVmcb->guest.u8CPL));
3821	Log4(("guest.u64CR0 %#RX64\n", pVmcb->guest.u64CR0));
3822	Log4(("guest.u64CR2 %#RX64\n", pVmcb->guest.u64CR2));
3823	Log4(("guest.u64CR3 %#RX64\n", pVmcb->guest.u64CR3));
3824	Log4(("guest.u64CR4 %#RX64\n", pVmcb->guest.u64CR4));
3825	Log4(("guest.u64DR6 %#RX64\n", pVmcb->guest.u64DR6));
3826	Log4(("guest.u64DR7 %#RX64\n", pVmcb->guest.u64DR7));
3827
3828	Log4(("guest.u64RIP %#RX64\n", pVmcb->guest.u64RIP));
3829	Log4(("guest.u64RSP %#RX64\n", pVmcb->guest.u64RSP));
3830	Log4(("guest.u64RAX %#RX64\n", pVmcb->guest.u64RAX));
3831	Log4(("guest.u64RFlags %#RX64\n", pVmcb->guest.u64RFlags));
3832
3833	Log4(("guest.u64SysEnterCS %#RX64\n", pVmcb->guest.u64SysEnterCS));
3834	Log4(("guest.u64SysEnterEIP %#RX64\n", pVmcb->guest.u64SysEnterEIP));
3835	Log4(("guest.u64SysEnterESP %#RX64\n", pVmcb->guest.u64SysEnterESP));
3836
3837	Log4(("guest.u64EFER %#RX64\n", pVmcb->guest.u64EFER));
3838	Log4(("guest.u64STAR %#RX64\n", pVmcb->guest.u64STAR));
3839	Log4(("guest.u64LSTAR %#RX64\n", pVmcb->guest.u64LSTAR));
3840	Log4(("guest.u64CSTAR %#RX64\n", pVmcb->guest.u64CSTAR));
3841	Log4(("guest.u64SFMASK %#RX64\n", pVmcb->guest.u64SFMASK));
3842	Log4(("guest.u64KernelGSBase %#RX64\n", pVmcb->guest.u64KernelGSBase));
3843	Log4(("guest.u64GPAT %#RX64\n", pVmcb->guest.u64GPAT));
3844	Log4(("guest.u64DBGCTL %#RX64\n", pVmcb->guest.u64DBGCTL));
3845	Log4(("guest.u64BR_FROM %#RX64\n", pVmcb->guest.u64BR_FROM));
3846	Log4(("guest.u64BR_TO %#RX64\n", pVmcb->guest.u64BR_TO));
3847	Log4(("guest.u64LASTEXCPFROM %#RX64\n", pVmcb->guest.u64LASTEXCPFROM));
3848	Log4(("guest.u64LASTEXCPTO %#RX64\n", pVmcb->guest.u64LASTEXCPTO));
3849	#endif /* VBOX_STRICT */
3850	}
3851	else
3852	Log4(("hmR0SvmReportWorldSwitchError: rcVMRun=%d\n", rcVMRun));
3853
3854	NOREF(pVmcb);
3855	}
3856
3857
3858	/**
3859	* Check per-VM and per-VCPU force flag actions that require us to go back to
3860	* ring-3 for one reason or another.
3861	*
3862	* @returns VBox status code (information status code included).
3863	* @retval VINF_SUCCESS if we don't have any actions that require going back to
3864	* ring-3.
3865	* @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync.
3866	* @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware
3867	* interrupts)
3868	* @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires
3869	* all EMTs to be in ring-3.
3870	* @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests.
3871	* @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return
3872	* to the EM loop.
3873	*
3874	* @param pVM The cross context VM structure.
3875	* @param pVCpu The cross context virtual CPU structure.
3876	* @param pCtx Pointer to the guest-CPU context.
3877	*/
3878	static int hmR0SvmCheckForceFlags(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
3879	{
3880	Assert(VMMRZCallRing3IsEnabled(pVCpu));
3881
3882	/* On AMD-V we don't need to update CR3, PAE PDPES lazily. See hmR0SvmSaveGuestState(). */
3883	Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
3884	Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
3885
3886	/* Update pending interrupts into the APIC's IRR. */
3887	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UPDATE_APIC))
3888	APICUpdatePendingInterrupts(pVCpu);
3889
3890	if ( VM_FF_IS_PENDING(pVM, !pVCpu->hm.s.fSingleInstruction
3891	? VM_FF_HP_R0_PRE_HM_MASK : VM_FF_HP_R0_PRE_HM_STEP_MASK)
3892	\|\| VMCPU_FF_IS_PENDING(pVCpu, !pVCpu->hm.s.fSingleInstruction
3893	? VMCPU_FF_HP_R0_PRE_HM_MASK : VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) )
3894	{
3895	/* Pending PGM C3 sync. */
3896	if (VMCPU_FF_IS_PENDING(pVCpu,VMCPU_FF_PGM_SYNC_CR3 \| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL))
3897	{
3898	int rc = PGMSyncCR3(pVCpu, pCtx->cr0, pCtx->cr3, pCtx->cr4, VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
3899	if (rc != VINF_SUCCESS)
3900	{
3901	Log4(("hmR0SvmCheckForceFlags: PGMSyncCR3 forcing us back to ring-3. rc=%d\n", rc));
3902	return rc;
3903	}
3904	}
3905
3906	/* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */
3907	/* -XXX- what was that about single stepping? */
3908	if ( VM_FF_IS_PENDING(pVM, VM_FF_HM_TO_R3_MASK)
3909	\|\| VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
3910	{
3911	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
3912	int rc = RT_UNLIKELY(VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_NO_MEMORY : VINF_EM_RAW_TO_R3;
3913	Log4(("hmR0SvmCheckForceFlags: HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc));
3914	return rc;
3915	}
3916
3917	/* Pending VM request packets, such as hardware interrupts. */
3918	if ( VM_FF_IS_PENDING(pVM, VM_FF_REQUEST)
3919	\|\| VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_REQUEST))
3920	{
3921	Log4(("hmR0SvmCheckForceFlags: Pending VM request forcing us back to ring-3\n"));
3922	return VINF_EM_PENDING_REQUEST;
3923	}
3924
3925	/* Pending PGM pool flushes. */
3926	if (VM_FF_IS_PENDING(pVM, VM_FF_PGM_POOL_FLUSH_PENDING))
3927	{
3928	Log4(("hmR0SvmCheckForceFlags: PGM pool flush pending forcing us back to ring-3\n"));
3929	return VINF_PGM_POOL_FLUSH_PENDING;
3930	}
3931
3932	/* Pending DMA requests. */
3933	if (VM_FF_IS_PENDING(pVM, VM_FF_PDM_DMA))
3934	{
3935	Log4(("hmR0SvmCheckForceFlags: Pending DMA request forcing us back to ring-3\n"));
3936	return VINF_EM_RAW_TO_R3;
3937	}
3938	}
3939
3940	return VINF_SUCCESS;
3941	}
3942
3943
3944	#ifdef VBOX_WITH_NESTED_HWVIRT
3945	/**
3946	* Does the preparations before executing nested-guest code in AMD-V.
3947	*
3948	* @returns VBox status code (informational status codes included).
3949	* @retval VINF_SUCCESS if we can proceed with running the guest.
3950	* @retval VINF_* scheduling changes, we have to go back to ring-3.
3951	*
3952	* @param pVM The cross context VM structure.
3953	* @param pVCpu The cross context virtual CPU structure.
3954	* @param pCtx Pointer to the guest-CPU context.
3955	* @param pSvmTransient Pointer to the SVM transient structure.
3956	*
3957	* @remarks Same caveats regarding longjumps as hmR0SvmPreRunGuest applies.
3958	* @sa hmR0SvmPreRunGuest.
3959	*/
3960	static int hmR0SvmPreRunGuestNested(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
3961	{
3962	HMSVM_ASSERT_PREEMPT_SAFE();
3963
3964	if (CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
3965	{
3966	#ifdef VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM
3967	Log2(("hmR0SvmPreRunGuest: Rescheduling to IEM due to nested-hwvirt or forced IEM exec -> VINF_EM_RESCHEDULE_REM\n"));
3968	return VINF_EM_RESCHEDULE_REM;
3969	#endif
3970	}
3971	else
3972	return VINF_SVM_VMEXIT;
3973
3974	/* Check force flag actions that might require us to go back to ring-3. */
3975	int rc = hmR0SvmCheckForceFlags(pVM, pVCpu, pCtx);
3976	if (rc != VINF_SUCCESS)
3977	return rc;
3978
3979	if (TRPMHasTrap(pVCpu))
3980	hmR0SvmTrpmTrapToPendingEvent(pVCpu);
3981	else if (!pVCpu->hm.s.Event.fPending)
3982	{
3983	VBOXSTRICTRC rcStrict = hmR0SvmEvaluatePendingEventNested(pVCpu, pCtx);
3984	if (rcStrict != VINF_SUCCESS)
3985	return VBOXSTRICTRC_VAL(rcStrict);
3986	if (!CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
3987	return VINF_SVM_VMEXIT;
3988	}
3989
3990	/*
3991	* On the oldest AMD-V systems, we may not get enough information to reinject an NMI.
3992	* Just do it in software, see @bugref{8411}.
3993	* NB: If we could continue a task switch exit we wouldn't need to do this.
3994	*/
3995	if (RT_UNLIKELY( !pVM->hm.s.svm.u32Features
3996	&& pVCpu->hm.s.Event.fPending
3997	&& SVM_EVENT_GET_TYPE(pVCpu->hm.s.Event.u64IntInfo) == SVM_EVENT_NMI))
3998	{
3999	return VINF_EM_RAW_INJECT_TRPM_EVENT;
4000	}
4001
4002	#ifdef HMSVM_SYNC_FULL_NESTED_GUEST_STATE
4003	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
4004	#endif
4005
4006	/*
4007	* Load the nested-guest state.
4008	*/
4009	rc = hmR0SvmLoadGuestStateNested(pVCpu, pCtx);
4010	AssertRCReturn(rc, rc);
4011	STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadFull); /** @todo Get new STAM counter for this? */
4012
4013	/* Ensure we've cached (and hopefully modified) the VMCB for execution using hardware SVM. */
4014	Assert(pCtx->hwvirt.svm.fHMCachedVmcb);
4015
4016	/*
4017	* No longjmps to ring-3 from this point on!!!
4018	* Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
4019	* This also disables flushing of the R0-logger instance (if any).
4020	*/
4021	VMMRZCallRing3Disable(pVCpu);
4022
4023	/*
4024	* We disable interrupts so that we don't miss any interrupts that would flag preemption (IPI/timers etc.)
4025	* when thread-context hooks aren't used and we've been running with preemption disabled for a while.
4026	*
4027	* We need to check for force-flags that could've possible been altered since we last checked them (e.g.
4028	* by PDMGetInterrupt() leaving the PDM critical section, see @bugref{6398}).
4029	*
4030	* We also check a couple of other force-flags as a last opportunity to get the EMT back to ring-3 before
4031	* executing guest code.
4032	*/
4033	pSvmTransient->fEFlags = ASMIntDisableFlags();
4034	if ( VM_FF_IS_PENDING(pVM, VM_FF_EMT_RENDEZVOUS \| VM_FF_TM_VIRTUAL_SYNC)
4035	\|\| VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
4036	{
4037	ASMSetFlags(pSvmTransient->fEFlags);
4038	VMMRZCallRing3Enable(pVCpu);
4039	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
4040	return VINF_EM_RAW_TO_R3;
4041	}
4042	if (RTThreadPreemptIsPending(NIL_RTTHREAD))
4043	{
4044	ASMSetFlags(pSvmTransient->fEFlags);
4045	VMMRZCallRing3Enable(pVCpu);
4046	STAM_COUNTER_INC(&pVCpu->hm.s.StatPendingHostIrq);
4047	return VINF_EM_RAW_INTERRUPT;
4048	}
4049
4050	/*
4051	* If we are injecting an NMI, we must set VMCPU_FF_BLOCK_NMIS only when we are going to execute
4052	* guest code for certain (no exits to ring-3). Otherwise, we could re-read the flag on re-entry into
4053	* AMD-V and conclude that NMI inhibition is active when we have not even delivered the NMI.
4054	*
4055	* With VT-x, this is handled by the Guest interruptibility information VMCS field which will set the
4056	* VMCS field after actually delivering the NMI which we read on VM-exit to determine the state.
4057	*/
4058	if (pVCpu->hm.s.Event.fPending)
4059	{
4060	SVMEVENT Event;
4061	Event.u = pVCpu->hm.s.Event.u64IntInfo;
4062	if ( Event.n.u1Valid
4063	&& Event.n.u3Type == SVM_EVENT_NMI
4064	&& Event.n.u8Vector == X86_XCPT_NMI
4065	&& !VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
4066	{
4067	VMCPU_FF_SET(pVCpu, VMCPU_FF_BLOCK_NMIS);
4068	}
4069	}
4070
4071	return VINF_SUCCESS;
4072	}
4073	#endif
4074
4075
4076	/**
4077	* Does the preparations before executing guest code in AMD-V.
4078	*
4079	* This may cause longjmps to ring-3 and may even result in rescheduling to the
4080	* recompiler. We must be cautious what we do here regarding committing
4081	* guest-state information into the VMCB assuming we assuredly execute the guest
4082	* in AMD-V. If we fall back to the recompiler after updating the VMCB and
4083	* clearing the common-state (TRPM/forceflags), we must undo those changes so
4084	* that the recompiler can (and should) use them when it resumes guest
4085	* execution. Otherwise such operations must be done when we can no longer
4086	* exit to ring-3.
4087	*
4088	* @returns VBox status code (informational status codes included).
4089	* @retval VINF_SUCCESS if we can proceed with running the guest.
4090	* @retval VINF_* scheduling changes, we have to go back to ring-3.
4091	*
4092	* @param pVM The cross context VM structure.
4093	* @param pVCpu The cross context virtual CPU structure.
4094	* @param pCtx Pointer to the guest-CPU context.
4095	* @param pSvmTransient Pointer to the SVM transient structure.
4096	*/
4097	static int hmR0SvmPreRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
4098	{
4099	HMSVM_ASSERT_PREEMPT_SAFE();
4100	HMSVM_ASSERT_NOT_IN_NESTED_GUEST(pCtx);
4101
4102	/* Check force flag actions that might require us to go back to ring-3. */
4103	int rc = hmR0SvmCheckForceFlags(pVM, pVCpu, pCtx);
4104	if (rc != VINF_SUCCESS)
4105	return rc;
4106
4107	if (TRPMHasTrap(pVCpu))
4108	hmR0SvmTrpmTrapToPendingEvent(pVCpu);
4109	else if (!pVCpu->hm.s.Event.fPending)
4110	hmR0SvmEvaluatePendingEvent(pVCpu, pCtx);
4111
4112	/*
4113	* On the oldest AMD-V systems, we may not get enough information to reinject an NMI.
4114	* Just do it in software, see @bugref{8411}.
4115	* NB: If we could continue a task switch exit we wouldn't need to do this.
4116	*/
4117	if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending && (((pVCpu->hm.s.Event.u64IntInfo >> 8) & 7) == SVM_EVENT_NMI)))
4118	if (RT_UNLIKELY(!pVM->hm.s.svm.u32Features))
4119	return VINF_EM_RAW_INJECT_TRPM_EVENT;
4120
4121	#ifdef HMSVM_SYNC_FULL_GUEST_STATE
4122	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
4123	#endif
4124
4125	/* Load the guest bits that are not shared with the host in any way since we can longjmp or get preempted. */
4126	rc = hmR0SvmLoadGuestState(pVM, pVCpu, pCtx);
4127	AssertRCReturn(rc, rc);
4128	STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadFull);
4129
4130	/*
4131	* If we're not intercepting TPR changes in the guest, save the guest TPR before the world-switch
4132	* so we can update it on the way back if the guest changed the TPR.
4133	*/
4134	if (pVCpu->hm.s.svm.fSyncVTpr)
4135	{
4136	if (pVM->hm.s.fTPRPatchingActive)
4137	pSvmTransient->u8GuestTpr = pCtx->msrLSTAR;
4138	else
4139	{
4140	PCSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
4141	pSvmTransient->u8GuestTpr = pVmcb->ctrl.IntCtrl.n.u8VTPR;
4142	}
4143	}
4144
4145	/*
4146	* No longjmps to ring-3 from this point on!!!
4147	* Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
4148	* This also disables flushing of the R0-logger instance (if any).
4149	*/
4150	VMMRZCallRing3Disable(pVCpu);
4151
4152	/*
4153	* We disable interrupts so that we don't miss any interrupts that would flag preemption (IPI/timers etc.)
4154	* when thread-context hooks aren't used and we've been running with preemption disabled for a while.
4155	*
4156	* We need to check for force-flags that could've possible been altered since we last checked them (e.g.
4157	* by PDMGetInterrupt() leaving the PDM critical section, see @bugref{6398}).
4158	*
4159	* We also check a couple of other force-flags as a last opportunity to get the EMT back to ring-3 before
4160	* executing guest code.
4161	*/
4162	pSvmTransient->fEFlags = ASMIntDisableFlags();
4163	if ( VM_FF_IS_PENDING(pVM, VM_FF_EMT_RENDEZVOUS \| VM_FF_TM_VIRTUAL_SYNC)
4164	\|\| VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
4165	{
4166	ASMSetFlags(pSvmTransient->fEFlags);
4167	VMMRZCallRing3Enable(pVCpu);
4168	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
4169	return VINF_EM_RAW_TO_R3;
4170	}
4171	if (RTThreadPreemptIsPending(NIL_RTTHREAD))
4172	{
4173	ASMSetFlags(pSvmTransient->fEFlags);
4174	VMMRZCallRing3Enable(pVCpu);
4175	STAM_COUNTER_INC(&pVCpu->hm.s.StatPendingHostIrq);
4176	return VINF_EM_RAW_INTERRUPT;
4177	}
4178
4179	/*
4180	* If we are injecting an NMI, we must set VMCPU_FF_BLOCK_NMIS only when we are going to execute
4181	* guest code for certain (no exits to ring-3). Otherwise, we could re-read the flag on re-entry into
4182	* AMD-V and conclude that NMI inhibition is active when we have not even delivered the NMI.
4183	*
4184	* With VT-x, this is handled by the Guest interruptibility information VMCS field which will set the
4185	* VMCS field after actually delivering the NMI which we read on VM-exit to determine the state.
4186	*/
4187	if (pVCpu->hm.s.Event.fPending)
4188	{
4189	SVMEVENT Event;
4190	Event.u = pVCpu->hm.s.Event.u64IntInfo;
4191	if ( Event.n.u1Valid
4192	&& Event.n.u3Type == SVM_EVENT_NMI
4193	&& Event.n.u8Vector == X86_XCPT_NMI
4194	&& !VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
4195	{
4196	VMCPU_FF_SET(pVCpu, VMCPU_FF_BLOCK_NMIS);
4197	}
4198	}
4199
4200	return VINF_SUCCESS;
4201	}
4202
4203
4204	#ifdef VBOX_WITH_NESTED_HWVIRT
4205	/**
4206	* Prepares to run nested-guest code in AMD-V and we've committed to doing so. This
4207	* means there is no backing out to ring-3 or anywhere else at this point.
4208	*
4209	* @param pVM The cross context VM structure.
4210	* @param pVCpu The cross context virtual CPU structure.
4211	* @param pCtx Pointer to the guest-CPU context.
4212	* @param pSvmTransient Pointer to the SVM transient structure.
4213	*
4214	* @remarks Called with preemption disabled.
4215	* @remarks No-long-jump zone!!!
4216	*/
4217	static void hmR0SvmPreRunGuestCommittedNested(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
4218	{
4219	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
4220	Assert(VMMR0IsLogFlushDisabled(pVCpu));
4221	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4222	HMSVM_ASSERT_IN_NESTED_GUEST(pCtx);
4223
4224	VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
4225	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC); /* Indicate the start of guest execution. */
4226
4227	PSVMVMCB pVmcbNstGst = pCtx->hwvirt.svm.CTX_SUFF(pVmcb);
4228	hmR0SvmInjectPendingEvent(pVCpu, pCtx, pVmcbNstGst);
4229
4230	if ( pVCpu->hm.s.fPreloadGuestFpu
4231	&& !CPUMIsGuestFPUStateActive(pVCpu))
4232	{
4233	CPUMR0LoadGuestFPU(pVM, pVCpu); /* (Ignore rc, no need to set HM_CHANGED_HOST_CONTEXT for SVM.) */
4234	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
4235	}
4236
4237	/* Load the state shared between host and nested-guest (FPU, debug). */
4238	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE))
4239	hmR0SvmLoadSharedState(pVCpu, pVmcbNstGst, pCtx);
4240
4241	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_HOST_CONTEXT); /* Preemption might set this, nothing to do on AMD-V. */
4242	AssertMsg(!HMCPU_CF_VALUE(pVCpu), ("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
4243
4244	/* Setup TSC offsetting. */
4245	RTCPUID idCurrentCpu = hmR0GetCurrentCpu()->idCpu;
4246	if ( pSvmTransient->fUpdateTscOffsetting
4247	\|\| idCurrentCpu != pVCpu->hm.s.idLastCpu)
4248	{
4249	hmR0SvmUpdateTscOffsettingNested(pVM, pVCpu, pCtx, pVmcbNstGst);
4250	pSvmTransient->fUpdateTscOffsetting = false;
4251	}
4252
4253	/* If we've migrating CPUs, mark the VMCB Clean bits as dirty. */
4254	if (idCurrentCpu != pVCpu->hm.s.idLastCpu)
4255	pVmcbNstGst->ctrl.u32VmcbCleanBits = 0;
4256
4257	/* Store status of the shared guest-host state at the time of VMRUN. */
4258	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
4259	if (CPUMIsGuestInLongModeEx(pCtx))
4260	{
4261	pSvmTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActivePending(pVCpu);
4262	pSvmTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActivePending(pVCpu);
4263	}
4264	else
4265	#endif
4266	{
4267	pSvmTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
4268	pSvmTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
4269	}
4270	pSvmTransient->fWasGuestFPUStateActive = CPUMIsGuestFPUStateActive(pVCpu);
4271
4272	/* The TLB flushing would've already been setup by the nested-hypervisor. */
4273	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB flushing, set this across the world switch. */
4274	hmR0SvmFlushTaggedTlb(pVCpu, pCtx, pVmcbNstGst);
4275	Assert(hmR0GetCurrentCpu()->idCpu == pVCpu->hm.s.idLastCpu);
4276
4277	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
4278
4279	TMNotifyStartOfExecution(pVCpu); /* Finally, notify TM to resume its clocks as we're about
4280	to start executing. */
4281
4282	/*
4283	* Save the current Host TSC_AUX and write the guest TSC_AUX to the host, so that
4284	* RDTSCPs (that don't cause exits) reads the guest MSR. See @bugref{3324}.
4285	*
4286	* This should be done -after- any RDTSCPs for obtaining the host timestamp (TM, STAM etc).
4287	*/
4288	uint8_t pbMsrBitmap = (uint8_t )pCtx->hwvirt.svm.CTX_SUFF(pvMsrBitmap);
4289	if ( (pVM->hm.s.cpuid.u32AMDFeatureEDX & X86_CPUID_EXT_FEATURE_EDX_RDTSCP)
4290	&& !(pVmcbNstGst->ctrl.u64InterceptCtrl & SVM_CTRL_INTERCEPT_RDTSCP))
4291	{
4292	hmR0SvmSetMsrPermission(pVmcbNstGst, pbMsrBitmap, MSR_K8_TSC_AUX, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
4293	pVCpu->hm.s.u64HostTscAux = ASMRdMsr(MSR_K8_TSC_AUX);
4294	uint64_t u64GuestTscAux = CPUMR0GetGuestTscAux(pVCpu);
4295	if (u64GuestTscAux != pVCpu->hm.s.u64HostTscAux)
4296	ASMWrMsr(MSR_K8_TSC_AUX, u64GuestTscAux);
4297	pSvmTransient->fRestoreTscAuxMsr = true;
4298	}
4299	else
4300	{
4301	hmR0SvmSetMsrPermission(pVmcbNstGst, pbMsrBitmap, MSR_K8_TSC_AUX, SVMMSREXIT_INTERCEPT_READ, SVMMSREXIT_INTERCEPT_WRITE);
4302	pSvmTransient->fRestoreTscAuxMsr = false;
4303	}
4304
4305	/*
4306	* If VMCB Clean bits isn't supported by the CPU or exposed by the guest,
4307	* mark all state-bits as dirty indicating to the CPU to re-load from VMCB.
4308	*/
4309	bool const fSupportsVmcbCleanBits = hmR0SvmSupportsVmcbCleanBits(pVCpu, pCtx);
4310	if (!fSupportsVmcbCleanBits)
4311	pVmcbNstGst->ctrl.u32VmcbCleanBits = 0;
4312	}
4313	#endif
4314
4315
4316	/**
4317	* Prepares to run guest code in AMD-V and we've committed to doing so. This
4318	* means there is no backing out to ring-3 or anywhere else at this
4319	* point.
4320	*
4321	* @param pVM The cross context VM structure.
4322	* @param pVCpu The cross context virtual CPU structure.
4323	* @param pCtx Pointer to the guest-CPU context.
4324	* @param pSvmTransient Pointer to the SVM transient structure.
4325	*
4326	* @remarks Called with preemption disabled.
4327	* @remarks No-long-jump zone!!!
4328	*/
4329	static void hmR0SvmPreRunGuestCommitted(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
4330	{
4331	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
4332	Assert(VMMR0IsLogFlushDisabled(pVCpu));
4333	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4334	HMSVM_ASSERT_NOT_IN_NESTED_GUEST(pCtx);
4335
4336	VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
4337	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC); /* Indicate the start of guest execution. */
4338
4339	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
4340	hmR0SvmInjectPendingEvent(pVCpu, pCtx, pVmcb);
4341
4342	if ( pVCpu->hm.s.fPreloadGuestFpu
4343	&& !CPUMIsGuestFPUStateActive(pVCpu))
4344	{
4345	CPUMR0LoadGuestFPU(pVM, pVCpu); /* (Ignore rc, no need to set HM_CHANGED_HOST_CONTEXT for SVM.) */
4346	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
4347	}
4348
4349	/* Load the state shared between host and guest (FPU, debug). */
4350	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE))
4351	hmR0SvmLoadSharedState(pVCpu, pVmcb, pCtx);
4352
4353	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_HOST_CONTEXT); /* Preemption might set this, nothing to do on AMD-V. */
4354	AssertMsg(!HMCPU_CF_VALUE(pVCpu), ("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
4355
4356	/* Setup TSC offsetting. */
4357	RTCPUID idCurrentCpu = hmR0GetCurrentCpu()->idCpu;
4358	if ( pSvmTransient->fUpdateTscOffsetting
4359	\|\| idCurrentCpu != pVCpu->hm.s.idLastCpu)
4360	{
4361	hmR0SvmUpdateTscOffsetting(pVM, pVCpu, pVmcb);
4362	pSvmTransient->fUpdateTscOffsetting = false;
4363	}
4364
4365	/* If we've migrating CPUs, mark the VMCB Clean bits as dirty. */
4366	if (idCurrentCpu != pVCpu->hm.s.idLastCpu)
4367	pVmcb->ctrl.u32VmcbCleanBits = 0;
4368
4369	/* Store status of the shared guest-host state at the time of VMRUN. */
4370	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
4371	if (CPUMIsGuestInLongModeEx(pCtx))
4372	{
4373	pSvmTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActivePending(pVCpu);
4374	pSvmTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActivePending(pVCpu);
4375	}
4376	else
4377	#endif
4378	{
4379	pSvmTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
4380	pSvmTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
4381	}
4382	pSvmTransient->fWasGuestFPUStateActive = CPUMIsGuestFPUStateActive(pVCpu);
4383
4384	/* Flush the appropriate tagged-TLB entries. */
4385	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB flushing, set this across the world switch. */
4386	hmR0SvmFlushTaggedTlb(pVCpu, pCtx, pVmcb);
4387	Assert(hmR0GetCurrentCpu()->idCpu == pVCpu->hm.s.idLastCpu);
4388
4389	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
4390
4391	TMNotifyStartOfExecution(pVCpu); /* Finally, notify TM to resume its clocks as we're about
4392	to start executing. */
4393
4394	/*
4395	* Save the current Host TSC_AUX and write the guest TSC_AUX to the host, so that
4396	* RDTSCPs (that don't cause exits) reads the guest MSR. See @bugref{3324}.
4397	*
4398	* This should be done -after- any RDTSCPs for obtaining the host timestamp (TM, STAM etc).
4399	*/
4400	uint8_t pbMsrBitmap = (uint8_t )pVCpu->hm.s.svm.pvMsrBitmap;
4401	if ( (pVM->hm.s.cpuid.u32AMDFeatureEDX & X86_CPUID_EXT_FEATURE_EDX_RDTSCP)
4402	&& !(pVmcb->ctrl.u64InterceptCtrl & SVM_CTRL_INTERCEPT_RDTSCP))
4403	{
4404	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_K8_TSC_AUX, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
4405	pVCpu->hm.s.u64HostTscAux = ASMRdMsr(MSR_K8_TSC_AUX);
4406	uint64_t u64GuestTscAux = CPUMR0GetGuestTscAux(pVCpu);
4407	if (u64GuestTscAux != pVCpu->hm.s.u64HostTscAux)
4408	ASMWrMsr(MSR_K8_TSC_AUX, u64GuestTscAux);
4409	pSvmTransient->fRestoreTscAuxMsr = true;
4410	}
4411	else
4412	{
4413	hmR0SvmSetMsrPermission(pVmcb, pbMsrBitmap, MSR_K8_TSC_AUX, SVMMSREXIT_INTERCEPT_READ, SVMMSREXIT_INTERCEPT_WRITE);
4414	pSvmTransient->fRestoreTscAuxMsr = false;
4415	}
4416
4417	/* If VMCB Clean bits isn't supported by the CPU, simply mark all state-bits as dirty, indicating (re)load-from-VMCB. */
4418	bool const fSupportsVmcbCleanBits = hmR0SvmSupportsVmcbCleanBits(pVCpu, pCtx);
4419	if (!fSupportsVmcbCleanBits)
4420	pVmcb->ctrl.u32VmcbCleanBits = 0;
4421	}
4422
4423
4424	/**
4425	* Wrapper for running the guest code in AMD-V.
4426	*
4427	* @returns VBox strict status code.
4428	* @param pVM The cross context VM structure.
4429	* @param pVCpu The cross context virtual CPU structure.
4430	* @param pCtx Pointer to the guest-CPU context.
4431	*
4432	* @remarks No-long-jump zone!!!
4433	*/
4434	DECLINLINE(int) hmR0SvmRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
4435	{
4436	/*
4437	* 64-bit Windows uses XMM registers in the kernel as the Microsoft compiler expresses floating-point operations
4438	* using SSE instructions. Some XMM registers (XMM6-XMM15) are callee-saved and thus the need for this XMM wrapper.
4439	* Refer MSDN docs. "Configuring Programs for 64-bit / x64 Software Conventions / Register Usage" for details.
4440	*/
4441	#ifdef VBOX_WITH_KERNEL_USING_XMM
4442	return hmR0SVMRunWrapXMM(pVCpu->hm.s.svm.HCPhysVmcbHost, pVCpu->hm.s.svm.HCPhysVmcb, pCtx, pVM, pVCpu,
4443	pVCpu->hm.s.svm.pfnVMRun);
4444	#else
4445	return pVCpu->hm.s.svm.pfnVMRun(pVCpu->hm.s.svm.HCPhysVmcbHost, pVCpu->hm.s.svm.HCPhysVmcb, pCtx, pVM, pVCpu);
4446	#endif
4447	}
4448
4449
4450	#ifdef VBOX_WITH_NESTED_HWVIRT
4451	/**
4452	* Wrapper for running the nested-guest code in AMD-V.
4453	*
4454	* @returns VBox strict status code.
4455	* @param pVM The cross context VM structure.
4456	* @param pVCpu The cross context virtual CPU structure.
4457	* @param pCtx Pointer to the guest-CPU context.
4458	*
4459	* @remarks No-long-jump zone!!!
4460	*/
4461	DECLINLINE(int) hmR0SvmRunGuestNested(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
4462	{
4463	/*
4464	* 64-bit Windows uses XMM registers in the kernel as the Microsoft compiler expresses floating-point operations
4465	* using SSE instructions. Some XMM registers (XMM6-XMM15) are callee-saved and thus the need for this XMM wrapper.
4466	* Refer MSDN docs. "Configuring Programs for 64-bit / x64 Software Conventions / Register Usage" for details.
4467	*/
4468	#ifdef VBOX_WITH_KERNEL_USING_XMM
4469	return hmR0SVMRunWrapXMM(pVCpu->hm.s.svm.HCPhysVmcbHost, pCtx->hwvirt.svm.HCPhysVmcb, pCtx, pVM, pVCpu,
4470	pVCpu->hm.s.svm.pfnVMRun);
4471	#else
4472	return pVCpu->hm.s.svm.pfnVMRun(pVCpu->hm.s.svm.HCPhysVmcbHost, pCtx->hwvirt.svm.HCPhysVmcb, pCtx, pVM, pVCpu);
4473	#endif
4474	}
4475
4476
4477	/**
4478	* Performs some essential restoration of state after running nested-guest code in
4479	* AMD-V.
4480	*
4481	* @param pVM The cross context VM structure.
4482	* @param pVCpu The cross context virtual CPU structure.
4483	* @param pMixedCtx Pointer to the nested-guest-CPU context. The data maybe
4484	* out-of-sync. Make sure to update the required fields
4485	* before using them.
4486	* @param pSvmTransient Pointer to the SVM transient structure.
4487	* @param rcVMRun Return code of VMRUN.
4488	*
4489	* @remarks Called with interrupts disabled.
4490	* @remarks No-long-jump zone!!! This function will however re-enable longjmps
4491	* unconditionally when it is safe to do so.
4492	*/
4493	static void hmR0SvmPostRunGuestNested(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PSVMTRANSIENT pSvmTransient, int rcVMRun)
4494	{
4495	RT_NOREF(pVM);
4496	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
4497
4498	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB flushing. */
4499	ASMAtomicIncU32(&pVCpu->hm.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for EMT poking. */
4500
4501	/* TSC read must be done early for maximum accuracy. */
4502	PSVMVMCB pVmcbNstGst = pMixedCtx->hwvirt.svm.CTX_SUFF(pVmcb);
4503	PSVMVMCBCTRL pVmcbNstGstCtrl = &pVmcbNstGst->ctrl;
4504	PCSVMNESTEDVMCBCACHE pVmcbNstGstCache = &pVCpu->hm.s.svm.NstGstVmcbCache;
4505	if (!(pVmcbNstGstCtrl->u64InterceptCtrl & SVM_CTRL_INTERCEPT_RDTSC))
4506	{
4507	/*
4508	* Undo what we did in hmR0SvmUpdateTscOffsettingNested() but don't restore the
4509	* nested-guest VMCB TSC offset here. It shall eventually be restored on #VMEXIT
4510	* later by HMSvmNstGstVmExitNotify().
4511	*/
4512	TMCpuTickSetLastSeen(pVCpu, ASMReadTSC() + pVmcbNstGstCtrl->u64TSCOffset - pVmcbNstGstCache->u64TSCOffset);
4513	}
4514
4515	if (pSvmTransient->fRestoreTscAuxMsr)
4516	{
4517	uint64_t u64GuestTscAuxMsr = ASMRdMsr(MSR_K8_TSC_AUX);
4518	CPUMR0SetGuestTscAux(pVCpu, u64GuestTscAuxMsr);
4519	if (u64GuestTscAuxMsr != pVCpu->hm.s.u64HostTscAux)
4520	ASMWrMsr(MSR_K8_TSC_AUX, pVCpu->hm.s.u64HostTscAux);
4521	}
4522
4523	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatExit1, x);
4524	TMNotifyEndOfExecution(pVCpu); /* Notify TM that the guest is no longer running. */
4525	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
4526
4527	Assert(!(ASMGetFlags() & X86_EFL_IF));
4528	ASMSetFlags(pSvmTransient->fEFlags); /* Enable interrupts. */
4529	VMMRZCallRing3Enable(pVCpu); /* It is now safe to do longjmps to ring-3!!! */
4530
4531	/* Mark the VMCB-state cache as unmodified by VMM. */
4532	pVmcbNstGstCtrl->u32VmcbCleanBits = HMSVM_VMCB_CLEAN_ALL;
4533
4534	/* If VMRUN failed, we can bail out early. This does -not- cover SVM_EXIT_INVALID. */
4535	if (RT_UNLIKELY(rcVMRun != VINF_SUCCESS))
4536	{
4537	Log4(("VMRUN failure: rcVMRun=%Rrc\n", rcVMRun));
4538	return;
4539	}
4540
4541	pSvmTransient->u64ExitCode = pVmcbNstGstCtrl->u64ExitCode; /* Save the #VMEXIT reason. */
4542	HMCPU_EXIT_HISTORY_ADD(pVCpu, pVmcbNstGstCtrl->u64ExitCode);/* Update the #VMEXIT history array. */
4543	pSvmTransient->fVectoringDoublePF = false; /* Vectoring double page-fault needs to be determined later. */
4544	pSvmTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
4545
4546	Assert(!pVCpu->hm.s.svm.fSyncVTpr);
4547	hmR0SvmSaveGuestState(pVCpu, pMixedCtx, pVmcbNstGst); /* Save the nested-guest state from the VMCB to the
4548	guest-CPU context. */
4549	}
4550	#endif
4551
4552	/**
4553	* Performs some essential restoration of state after running guest code in
4554	* AMD-V.
4555	*
4556	* @param pVM The cross context VM structure.
4557	* @param pVCpu The cross context virtual CPU structure.
4558	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
4559	* out-of-sync. Make sure to update the required fields
4560	* before using them.
4561	* @param pSvmTransient Pointer to the SVM transient structure.
4562	* @param rcVMRun Return code of VMRUN.
4563	*
4564	* @remarks Called with interrupts disabled.
4565	* @remarks No-long-jump zone!!! This function will however re-enable longjmps
4566	* unconditionally when it is safe to do so.
4567	*/
4568	static void hmR0SvmPostRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PSVMTRANSIENT pSvmTransient, int rcVMRun)
4569	{
4570	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
4571
4572	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB flushing. */
4573	ASMAtomicIncU32(&pVCpu->hm.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for EMT poking. */
4574
4575	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
4576	pVmcb->ctrl.u32VmcbCleanBits = HMSVM_VMCB_CLEAN_ALL; /* Mark the VMCB-state cache as unmodified by VMM. */
4577
4578	/* TSC read must be done early for maximum accuracy. */
4579	if (!(pVmcb->ctrl.u64InterceptCtrl & SVM_CTRL_INTERCEPT_RDTSC))
4580	TMCpuTickSetLastSeen(pVCpu, ASMReadTSC() + pVmcb->ctrl.u64TSCOffset);
4581
4582	if (pSvmTransient->fRestoreTscAuxMsr)
4583	{
4584	uint64_t u64GuestTscAuxMsr = ASMRdMsr(MSR_K8_TSC_AUX);
4585	CPUMR0SetGuestTscAux(pVCpu, u64GuestTscAuxMsr);
4586	if (u64GuestTscAuxMsr != pVCpu->hm.s.u64HostTscAux)
4587	ASMWrMsr(MSR_K8_TSC_AUX, pVCpu->hm.s.u64HostTscAux);
4588	}
4589
4590	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatExit1, x);
4591	TMNotifyEndOfExecution(pVCpu); /* Notify TM that the guest is no longer running. */
4592	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
4593
4594	Assert(!(ASMGetFlags() & X86_EFL_IF));
4595	ASMSetFlags(pSvmTransient->fEFlags); /* Enable interrupts. */
4596	VMMRZCallRing3Enable(pVCpu); /* It is now safe to do longjmps to ring-3!!! */
4597
4598	/* If VMRUN failed, we can bail out early. This does -not- cover SVM_EXIT_INVALID. */
4599	if (RT_UNLIKELY(rcVMRun != VINF_SUCCESS))
4600	{
4601	Log4(("VMRUN failure: rcVMRun=%Rrc\n", rcVMRun));
4602	return;
4603	}
4604
4605	pSvmTransient->u64ExitCode = pVmcb->ctrl.u64ExitCode; /* Save the #VMEXIT reason. */
4606	HMCPU_EXIT_HISTORY_ADD(pVCpu, pVmcb->ctrl.u64ExitCode); /* Update the #VMEXIT history array. */
4607	pSvmTransient->fVectoringDoublePF = false; /* Vectoring double page-fault needs to be determined later. */
4608	pSvmTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
4609
4610	hmR0SvmSaveGuestState(pVCpu, pMixedCtx, pVmcb); /* Save the guest state from the VMCB to the guest-CPU context. */
4611
4612	if (RT_LIKELY(pSvmTransient->u64ExitCode != SVM_EXIT_INVALID))
4613	{
4614	if (pVCpu->hm.s.svm.fSyncVTpr)
4615	{
4616	/* TPR patching (for 32-bit guests) uses LSTAR MSR for holding the TPR value, otherwise uses the VTPR. */
4617	if ( pVM->hm.s.fTPRPatchingActive
4618	&& (pMixedCtx->msrLSTAR & 0xff) != pSvmTransient->u8GuestTpr)
4619	{
4620	int rc = APICSetTpr(pVCpu, pMixedCtx->msrLSTAR & 0xff);
4621	AssertRC(rc);
4622	HMCPU_CF_SET(pVCpu, HM_CHANGED_SVM_GUEST_APIC_STATE);
4623	}
4624	else if (pSvmTransient->u8GuestTpr != pVmcb->ctrl.IntCtrl.n.u8VTPR)
4625	{
4626	int rc = APICSetTpr(pVCpu, pVmcb->ctrl.IntCtrl.n.u8VTPR << 4);
4627	AssertRC(rc);
4628	HMCPU_CF_SET(pVCpu, HM_CHANGED_SVM_GUEST_APIC_STATE);
4629	}
4630	}
4631	}
4632	}
4633
4634
4635	/**
4636	* Runs the guest code using AMD-V.
4637	*
4638	* @returns VBox status code.
4639	* @param pVM The cross context VM structure.
4640	* @param pVCpu The cross context virtual CPU structure.
4641	* @param pCtx Pointer to the guest-CPU context.
4642	* @param pcLoops Pointer to the number of executed loops.
4643	*/
4644	static int hmR0SvmRunGuestCodeNormal(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, uint32_t *pcLoops)
4645	{
4646	uint32_t const cMaxResumeLoops = pVM->hm.s.cMaxResumeLoops;
4647	Assert(pcLoops);
4648	Assert(*pcLoops <= cMaxResumeLoops);
4649
4650	SVMTRANSIENT SvmTransient;
4651	SvmTransient.fUpdateTscOffsetting = true;
4652
4653	int rc = VERR_INTERNAL_ERROR_5;
4654	for (;;)
4655	{
4656	Assert(!HMR0SuspendPending());
4657	HMSVM_ASSERT_CPU_SAFE();
4658
4659	/* Preparatory work for running guest code, this may force us to return
4660	to ring-3. This bugger disables interrupts on VINF_SUCCESS! */
4661	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
4662	rc = hmR0SvmPreRunGuest(pVM, pVCpu, pCtx, &SvmTransient);
4663	if (rc != VINF_SUCCESS)
4664	break;
4665
4666	/*
4667	* No longjmps to ring-3 from this point on!!!
4668	* Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
4669	* This also disables flushing of the R0-logger instance (if any).
4670	*/
4671	hmR0SvmPreRunGuestCommitted(pVM, pVCpu, pCtx, &SvmTransient);
4672	rc = hmR0SvmRunGuest(pVM, pVCpu, pCtx);
4673
4674	/* Restore any residual host-state and save any bits shared between host
4675	and guest into the guest-CPU state. Re-enables interrupts! */
4676	hmR0SvmPostRunGuest(pVM, pVCpu, pCtx, &SvmTransient, rc);
4677
4678	if (RT_UNLIKELY( rc != VINF_SUCCESS /* Check for VMRUN errors. */
4679	\|\| SvmTransient.u64ExitCode == SVM_EXIT_INVALID)) /* Check for invalid guest-state errors. */
4680	{
4681	if (rc == VINF_SUCCESS)
4682	rc = VERR_SVM_INVALID_GUEST_STATE;
4683	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit1, x);
4684	hmR0SvmReportWorldSwitchError(pVM, pVCpu, rc, pCtx);
4685	break;
4686	}
4687
4688	/* Handle the #VMEXIT. */
4689	HMSVM_EXITCODE_STAM_COUNTER_INC(SvmTransient.u64ExitCode);
4690	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x);
4691	VBOXVMM_R0_HMSVM_VMEXIT(pVCpu, pCtx, SvmTransient.u64ExitCode, pVCpu->hm.s.svm.pVmcb);
4692	rc = hmR0SvmHandleExit(pVCpu, pCtx, &SvmTransient);
4693	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x);
4694	if (rc != VINF_SUCCESS)
4695	break;
4696	if (++(*pcLoops) >= cMaxResumeLoops)
4697	{
4698	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
4699	rc = VINF_EM_RAW_INTERRUPT;
4700	break;
4701	}
4702	}
4703
4704	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
4705	return rc;
4706	}
4707
4708
4709	/**
4710	* Runs the guest code using AMD-V in single step mode.
4711	*
4712	* @returns VBox status code.
4713	* @param pVM The cross context VM structure.
4714	* @param pVCpu The cross context virtual CPU structure.
4715	* @param pCtx Pointer to the guest-CPU context.
4716	* @param pcLoops Pointer to the number of executed loops.
4717	*/
4718	static int hmR0SvmRunGuestCodeStep(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, uint32_t *pcLoops)
4719	{
4720	uint32_t const cMaxResumeLoops = pVM->hm.s.cMaxResumeLoops;
4721	Assert(pcLoops);
4722	Assert(*pcLoops <= cMaxResumeLoops);
4723
4724	SVMTRANSIENT SvmTransient;
4725	SvmTransient.fUpdateTscOffsetting = true;
4726
4727	uint16_t uCsStart = pCtx->cs.Sel;
4728	uint64_t uRipStart = pCtx->rip;
4729
4730	int rc = VERR_INTERNAL_ERROR_5;
4731	for (;;)
4732	{
4733	Assert(!HMR0SuspendPending());
4734	AssertMsg(pVCpu->hm.s.idEnteredCpu == RTMpCpuId(),
4735	("Illegal migration! Entered on CPU %u Current %u cLoops=%u\n", (unsigned)pVCpu->hm.s.idEnteredCpu,
4736	(unsigned)RTMpCpuId(), *pcLoops));
4737
4738	/* Preparatory work for running guest code, this may force us to return
4739	to ring-3. This bugger disables interrupts on VINF_SUCCESS! */
4740	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
4741	rc = hmR0SvmPreRunGuest(pVM, pVCpu, pCtx, &SvmTransient);
4742	if (rc != VINF_SUCCESS)
4743	break;
4744
4745	/*
4746	* No longjmps to ring-3 from this point on!!!
4747	* Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
4748	* This also disables flushing of the R0-logger instance (if any).
4749	*/
4750	VMMRZCallRing3Disable(pVCpu);
4751	VMMRZCallRing3RemoveNotification(pVCpu);
4752	hmR0SvmPreRunGuestCommitted(pVM, pVCpu, pCtx, &SvmTransient);
4753
4754	rc = hmR0SvmRunGuest(pVM, pVCpu, pCtx);
4755
4756	/*
4757	* Restore any residual host-state and save any bits shared between host and guest into the guest-CPU state.
4758	* This will also re-enable longjmps to ring-3 when it has reached a safe point!!!
4759	*/
4760	hmR0SvmPostRunGuest(pVM, pVCpu, pCtx, &SvmTransient, rc);
4761	if (RT_UNLIKELY( rc != VINF_SUCCESS /* Check for VMRUN errors. */
4762	\|\| SvmTransient.u64ExitCode == SVM_EXIT_INVALID)) /* Check for invalid guest-state errors. */
4763	{
4764	if (rc == VINF_SUCCESS)
4765	rc = VERR_SVM_INVALID_GUEST_STATE;
4766	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit1, x);
4767	hmR0SvmReportWorldSwitchError(pVM, pVCpu, rc, pCtx);
4768	return rc;
4769	}
4770
4771	/* Handle the #VMEXIT. */
4772	HMSVM_EXITCODE_STAM_COUNTER_INC(SvmTransient.u64ExitCode);
4773	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x);
4774	VBOXVMM_R0_HMSVM_VMEXIT(pVCpu, pCtx, SvmTransient.u64ExitCode, pVCpu->hm.s.svm.pVmcb);
4775	rc = hmR0SvmHandleExit(pVCpu, pCtx, &SvmTransient);
4776	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x);
4777	if (rc != VINF_SUCCESS)
4778	break;
4779	if (++(*pcLoops) >= cMaxResumeLoops)
4780	{
4781	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
4782	rc = VINF_EM_RAW_INTERRUPT;
4783	break;
4784	}
4785
4786	/*
4787	* Did the RIP change, if so, consider it a single step.
4788	* Otherwise, make sure one of the TFs gets set.
4789	*/
4790	if ( pCtx->rip != uRipStart
4791	\|\| pCtx->cs.Sel != uCsStart)
4792	{
4793	rc = VINF_EM_DBG_STEPPED;
4794	break;
4795	}
4796	pVCpu->hm.s.fContextUseFlags \|= HM_CHANGED_GUEST_DEBUG;
4797	}
4798
4799	/*
4800	* Clear the X86_EFL_TF if necessary.
4801	*/
4802	if (pVCpu->hm.s.fClearTrapFlag)
4803	{
4804	pVCpu->hm.s.fClearTrapFlag = false;
4805	pCtx->eflags.Bits.u1TF = 0;
4806	}
4807
4808	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
4809	return rc;
4810	}
4811
4812	#ifdef VBOX_WITH_NESTED_HWVIRT
4813	/**
4814	* Runs the nested-guest code using AMD-V.
4815	*
4816	* @returns VBox status code.
4817	* @param pVM The cross context VM structure.
4818	* @param pVCpu The cross context virtual CPU structure.
4819	* @param pCtx Pointer to the guest-CPU context.
4820	* @param pcLoops Pointer to the number of executed loops. If we're switching
4821	* from the guest-code execution loop to this nested-guest
4822	* execution loop pass the remainder value, else pass 0.
4823	*/
4824	static int hmR0SvmRunGuestCodeNested(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, uint32_t *pcLoops)
4825	{
4826	HMSVM_ASSERT_IN_NESTED_GUEST(pCtx);
4827	Assert(pcLoops);
4828	Assert(*pcLoops <= pVM->hm.s.cMaxResumeLoops);
4829
4830	SVMTRANSIENT SvmTransient;
4831	SvmTransient.fUpdateTscOffsetting = true;
4832
4833	int rc = VERR_INTERNAL_ERROR_4;
4834	for (;;)
4835	{
4836	Assert(!HMR0SuspendPending());
4837	HMSVM_ASSERT_CPU_SAFE();
4838
4839	/* Preparatory work for running nested-guest code, this may force us to return
4840	to ring-3. This bugger disables interrupts on VINF_SUCCESS! */
4841	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
4842	rc = hmR0SvmPreRunGuestNested(pVM, pVCpu, pCtx, &SvmTransient);
4843	if ( rc != VINF_SUCCESS
4844	\|\| !CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
4845	{
4846	break;
4847	}
4848
4849	/*
4850	* No longjmps to ring-3 from this point on!!!
4851	* Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
4852	* This also disables flushing of the R0-logger instance (if any).
4853	*/
4854	hmR0SvmPreRunGuestCommittedNested(pVM, pVCpu, pCtx, &SvmTransient);
4855
4856	rc = hmR0SvmRunGuestNested(pVM, pVCpu, pCtx);
4857
4858	/* Restore any residual host-state and save any bits shared between host
4859	and guest into the guest-CPU state. Re-enables interrupts! */
4860	hmR0SvmPostRunGuestNested(pVM, pVCpu, pCtx, &SvmTransient, rc);
4861
4862	if (RT_LIKELY( rc == VINF_SUCCESS
4863	&& SvmTransient.u64ExitCode != SVM_EXIT_INVALID))
4864	{ /* extremely likely */ }
4865	else
4866	{
4867	/* VMRUN failed, shouldn't really happen, Guru. */
4868	if (rc != VINF_SUCCESS)
4869	break;
4870
4871	/* Invalid nested-guest state. Cause a #VMEXIT but assert on strict builds. */
4872	AssertMsgFailed(("Invalid nested-guest state. rc=%Rrc u64ExitCode=%#RX64\n", rc, SvmTransient.u64ExitCode));
4873	rc = VBOXSTRICTRC_TODO(IEMExecSvmVmexit(pVCpu, SVM_EXIT_INVALID, 0, 0));
4874	break;
4875	}
4876
4877	/* Handle the #VMEXIT. */
4878	HMSVM_EXITCODE_STAM_COUNTER_INC(SvmTransient.u64ExitCode);
4879	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x);
4880	VBOXVMM_R0_HMSVM_VMEXIT(pVCpu, pCtx, SvmTransient.u64ExitCode, pCtx->hwvirt.svm.CTX_SUFF(pVmcb));
4881	rc = hmR0SvmHandleExitNested(pVCpu, pCtx, &SvmTransient);
4882	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x);
4883	if (rc != VINF_SUCCESS)
4884	break;
4885	if (++(*pcLoops) >= pVM->hm.s.cMaxResumeLoops)
4886	{
4887	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
4888	rc = VINF_EM_RAW_INTERRUPT;
4889	break;
4890	}
4891
4892	/** @todo handle single-stepping */
4893	}
4894
4895	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
4896	return rc;
4897	}
4898	#endif
4899
4900
4901	/**
4902	* Runs the guest code using AMD-V.
4903	*
4904	* @returns Strict VBox status code.
4905	* @param pVM The cross context VM structure.
4906	* @param pVCpu The cross context virtual CPU structure.
4907	* @param pCtx Pointer to the guest-CPU context.
4908	*/
4909	VMMR0DECL(VBOXSTRICTRC) SVMR0RunGuestCode(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
4910	{
4911	Assert(VMMRZCallRing3IsEnabled(pVCpu));
4912	HMSVM_ASSERT_PREEMPT_SAFE();
4913	VMMRZCallRing3SetNotification(pVCpu, hmR0SvmCallRing3Callback, pCtx);
4914
4915	uint32_t cLoops = 0;
4916	int rc;
4917	#ifdef VBOX_WITH_NESTED_HWVIRT
4918	if (!CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
4919	#endif
4920	{
4921	if (!pVCpu->hm.s.fSingleInstruction)
4922	rc = hmR0SvmRunGuestCodeNormal(pVM, pVCpu, pCtx, &cLoops);
4923	else
4924	rc = hmR0SvmRunGuestCodeStep(pVM, pVCpu, pCtx, &cLoops);
4925	}
4926	#ifdef VBOX_WITH_NESTED_HWVIRT
4927	else
4928	{
4929	rc = VINF_SVM_VMRUN;
4930	}
4931
4932	/* Re-check the nested-guest condition here as we may be transitioning from the normal
4933	execution loop into the nested-guest, hence this is not placed in the 'else' part above. */
4934	if (rc == VINF_SVM_VMRUN)
4935	{
4936	rc = hmR0SvmRunGuestCodeNested(pVM, pVCpu, pCtx, &cLoops);
4937	if (rc == VINF_SVM_VMEXIT)
4938	rc = VINF_SUCCESS;
4939	}
4940	#endif
4941
4942	/* Fixup error codes. */
4943	if (rc == VERR_EM_INTERPRETER)
4944	rc = VINF_EM_RAW_EMULATE_INSTR;
4945	else if (rc == VINF_EM_RESET)
4946	rc = VINF_EM_TRIPLE_FAULT;
4947
4948	/* Prepare to return to ring-3. This will remove longjmp notifications. */
4949	rc = hmR0SvmExitToRing3(pVM, pVCpu, pCtx, rc);
4950	Assert(!VMMRZCallRing3IsNotificationSet(pVCpu));
4951	return rc;
4952	}
4953
4954
4955	#ifdef VBOX_WITH_NESTED_HWVIRT
4956	/**
4957	* Determines whether an IOIO intercept is active for the nested-guest or not.
4958	*
4959	* @param pvIoBitmap Pointer to the nested-guest IO bitmap.
4960	* @param pIoExitInfo Pointer to the SVMIOIOEXITINFO.
4961	*/
4962	static bool hmR0SvmIsIoInterceptActive(void *pvIoBitmap, PSVMIOIOEXITINFO pIoExitInfo)
4963	{
4964	const uint16_t u16Port = pIoExitInfo->n.u16Port;
4965	const SVMIOIOTYPE enmIoType = (SVMIOIOTYPE)pIoExitInfo->n.u1Type;
4966	const uint8_t cbReg = (pIoExitInfo->u >> SVM_IOIO_OP_SIZE_SHIFT) & 7;
4967	const uint8_t cAddrSizeBits = ((pIoExitInfo->u >> SVM_IOIO_ADDR_SIZE_SHIFT) & 7) << 4;
4968	const uint8_t iEffSeg = pIoExitInfo->n.u3SEG;
4969	const bool fRep = pIoExitInfo->n.u1REP;
4970	const bool fStrIo = pIoExitInfo->n.u1STR;
4971
4972	return HMSvmIsIOInterceptActive(pvIoBitmap, u16Port, enmIoType, cbReg, cAddrSizeBits, iEffSeg, fRep, fStrIo,
4973	NULL /* pIoExitInfo */);
4974	}
4975
4976
4977	/**
4978	* Handles a nested-guest \#VMEXIT (for all EXITCODE values except
4979	* SVM_EXIT_INVALID).
4980	*
4981	* @returns VBox status code (informational status codes included).
4982	* @param pVCpu The cross context virtual CPU structure.
4983	* @param pCtx Pointer to the guest-CPU context.
4984	* @param pSvmTransient Pointer to the SVM transient structure.
4985	*/
4986	static int hmR0SvmHandleExitNested(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
4987	{
4988	HMSVM_ASSERT_IN_NESTED_GUEST(pCtx);
4989	Assert(pSvmTransient->u64ExitCode != SVM_EXIT_INVALID);
4990	Assert(pSvmTransient->u64ExitCode <= SVM_EXIT_MAX);
4991
4992	#define HM_SVM_VMEXIT_NESTED(a_pVCpu, a_uExitCode, a_uExitInfo1, a_uExitInfo2) \
4993	VBOXSTRICTRC_TODO(IEMExecSvmVmexit(a_pVCpu, a_uExitCode, a_uExitInfo1, a_uExitInfo2))
4994
4995	/*
4996	* For all the #VMEXITs here we primarily figure out if the #VMEXIT is expected
4997	* by the nested-guest. If it isn't, it should be handled by the (outer) guest.
4998	*/
4999	PSVMVMCB pVmcbNstGst = pCtx->hwvirt.svm.CTX_SUFF(pVmcb);
5000	PSVMVMCBCTRL pVmcbNstGstCtrl = &pVmcbNstGst->ctrl;
5001	uint64_t const uExitCode = pVmcbNstGstCtrl->u64ExitCode;
5002	uint64_t const uExitInfo1 = pVmcbNstGstCtrl->u64ExitInfo1;
5003	uint64_t const uExitInfo2 = pVmcbNstGstCtrl->u64ExitInfo2;
5004
5005	Assert(uExitCode == pVmcbNstGstCtrl->u64ExitCode);
5006	switch (uExitCode)
5007	{
5008	case SVM_EXIT_CPUID:
5009	{
5010	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_CPUID))
5011	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5012	return hmR0SvmExitCpuid(pVCpu, pCtx, pSvmTransient);
5013	}
5014
5015	case SVM_EXIT_RDTSC:
5016	{
5017	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_RDTSC))
5018	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5019	return hmR0SvmExitRdtsc(pVCpu, pCtx, pSvmTransient);
5020	}
5021
5022	case SVM_EXIT_RDTSCP:
5023	{
5024	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_RDTSCP))
5025	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5026	return hmR0SvmExitRdtscp(pVCpu, pCtx, pSvmTransient);
5027	}
5028
5029
5030	case SVM_EXIT_MONITOR:
5031	{
5032	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_MONITOR))
5033	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5034	return hmR0SvmExitMonitor(pVCpu, pCtx, pSvmTransient);
5035	}
5036
5037	case SVM_EXIT_MWAIT:
5038	{
5039	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_MWAIT))
5040	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5041	return hmR0SvmExitMwait(pVCpu, pCtx, pSvmTransient);
5042	}
5043
5044	case SVM_EXIT_HLT:
5045	{
5046	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_HLT))
5047	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5048	return hmR0SvmExitHlt(pVCpu, pCtx, pSvmTransient);
5049	}
5050
5051	case SVM_EXIT_MSR:
5052	{
5053	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_MSR_PROT))
5054	{
5055	uint32_t const idMsr = pCtx->ecx;
5056	uint16_t offMsrpm;
5057	uint32_t uMsrpmBit;
5058	int rc = HMSvmGetMsrpmOffsetAndBit(idMsr, &offMsrpm, &uMsrpmBit);
5059	if (RT_SUCCESS(rc))
5060	{
5061	void const *pvMsrBitmap = pCtx->hwvirt.svm.CTX_SUFF(pvMsrBitmap);
5062	bool const fInterceptRead = ASMBitTest(pvMsrBitmap, (offMsrpm << 3) + uMsrpmBit);
5063	bool const fInterceptWrite = ASMBitTest(pvMsrBitmap, (offMsrpm << 3) + uMsrpmBit + 1);
5064
5065	if ( (fInterceptWrite && pVmcbNstGstCtrl->u64ExitInfo1 == SVM_EXIT1_MSR_WRITE)
5066	\|\| (fInterceptRead && pVmcbNstGstCtrl->u64ExitInfo1 == SVM_EXIT1_MSR_READ))
5067	{
5068	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5069	}
5070	}
5071	else
5072	{
5073	/*
5074	* MSRs not covered by the MSRPM automatically cause an #VMEXIT.
5075	* See AMD-V spec. "15.11 MSR Intercepts".
5076	*/
5077	Assert(rc == VERR_OUT_OF_RANGE);
5078	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5079	}
5080	}
5081	return hmR0SvmExitMsr(pVCpu, pCtx, pSvmTransient);
5082	}
5083
5084	case SVM_EXIT_IOIO:
5085	{
5086	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_IOIO_PROT))
5087	{
5088	void *pvIoBitmap = pCtx->hwvirt.svm.CTX_SUFF(pvIoBitmap);
5089	SVMIOIOEXITINFO IoExitInfo;
5090	IoExitInfo.u = pVmcbNstGst->ctrl.u64ExitInfo1;
5091	bool const fIntercept = hmR0SvmIsIoInterceptActive(pvIoBitmap, &IoExitInfo);
5092	if (fIntercept)
5093	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5094	}
5095	return hmR0SvmExitIOInstr(pVCpu, pCtx, pSvmTransient);
5096	}
5097
5098	case SVM_EXIT_EXCEPTION_14: /* X86_XCPT_PF */
5099	{
5100	PVM pVM = pVCpu->CTX_SUFF(pVM);
5101	if (pVM->hm.s.fNestedPaging)
5102	{
5103	uint32_t const u32ErrCode = pVmcbNstGstCtrl->u64ExitInfo1;
5104	uint64_t const uFaultAddress = pVmcbNstGstCtrl->u64ExitInfo2;
5105
5106	/* If the nested-guest is intercepting #PFs, cause a #PF #VMEXIT. */
5107	if (HMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_PF))
5108	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, u32ErrCode, uFaultAddress);
5109
5110	/* If the nested-guest is not intercepting #PFs, forward the #PF to the nested-guest. */
5111	hmR0SvmSetPendingXcptPF(pVCpu, pCtx, u32ErrCode, uFaultAddress);
5112	return VINF_SUCCESS;
5113	}
5114	return hmR0SvmExitXcptPFNested(pVCpu, pCtx,pSvmTransient);
5115	}
5116
5117	case SVM_EXIT_EXCEPTION_7: /* X86_XCPT_NM */
5118	{
5119	if (HMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_NM))
5120	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5121	hmR0SvmSetPendingXcptNM(pVCpu);
5122	return VINF_SUCCESS;
5123	}
5124
5125	case SVM_EXIT_EXCEPTION_6: /* X86_XCPT_UD */
5126	{
5127	if (HMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_UD))
5128	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5129	hmR0SvmSetPendingXcptUD(pVCpu);
5130	return VINF_SUCCESS;
5131	}
5132
5133	case SVM_EXIT_EXCEPTION_16: /* X86_XCPT_MF */
5134	{
5135	if (HMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_MF))
5136	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5137	hmR0SvmSetPendingXcptMF(pVCpu);
5138	return VINF_SUCCESS;
5139	}
5140
5141	case SVM_EXIT_EXCEPTION_1: /* X86_XCPT_DB */
5142	{
5143	if (HMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_DB))
5144	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5145	return hmR0SvmNestedExitXcptDB(pVCpu, pCtx, pSvmTransient);
5146	}
5147
5148	case SVM_EXIT_EXCEPTION_17: /* X86_XCPT_AC */
5149	{
5150	if (HMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_AC))
5151	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5152	return hmR0SvmExitXcptAC(pVCpu, pCtx, pSvmTransient);
5153	}
5154
5155	case SVM_EXIT_EXCEPTION_3: /* X86_XCPT_BP */
5156	{
5157	if (HMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_BP))
5158	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5159	return hmR0SvmNestedExitXcptBP(pVCpu, pCtx, pSvmTransient);
5160	}
5161
5162	case SVM_EXIT_READ_CR0:
5163	case SVM_EXIT_READ_CR3:
5164	case SVM_EXIT_READ_CR4:
5165	{
5166	uint8_t const uCr = uExitCode - SVM_EXIT_READ_CR0;
5167	if (HMIsGuestSvmReadCRxInterceptSet(pVCpu, pCtx, uCr))
5168	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5169	return hmR0SvmExitReadCRx(pVCpu, pCtx, pSvmTransient);
5170	}
5171
5172	case SVM_EXIT_WRITE_CR0:
5173	case SVM_EXIT_WRITE_CR3:
5174	case SVM_EXIT_WRITE_CR4:
5175	case SVM_EXIT_WRITE_CR8: /** @todo Shouldn't writes to CR8 go to V_TPR instead since we run with V_INTR_MASKING set?? */
5176	{
5177	uint8_t const uCr = uExitCode - SVM_EXIT_WRITE_CR0;
5178	Log4(("hmR0SvmHandleExitNested: Write CR%u: uExitInfo1=%#RX64 uExitInfo2=%#RX64\n", uCr, uExitInfo1, uExitInfo2));
5179
5180	if (HMIsGuestSvmWriteCRxInterceptSet(pVCpu, pCtx, uCr))
5181	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5182	return hmR0SvmExitWriteCRx(pVCpu, pCtx, pSvmTransient);
5183	}
5184
5185	case SVM_EXIT_PAUSE:
5186	{
5187	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_PAUSE))
5188	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5189	return hmR0SvmExitPause(pVCpu, pCtx, pSvmTransient);
5190	}
5191
5192	case SVM_EXIT_VINTR:
5193	{
5194	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_VINTR))
5195	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5196	return hmR0SvmExitUnexpected(pVCpu, pCtx, pSvmTransient);
5197	}
5198
5199	case SVM_EXIT_INTR:
5200	case SVM_EXIT_NMI:
5201	case SVM_EXIT_SMI:
5202	{
5203	/*
5204	* We shouldn't direct physical interrupts, NMIs, SMIs to the nested-guest.
5205	*
5206	* Although we don't intercept SMIs, the nested-guest might. Therefore, we
5207	* might get an SMI #VMEXIT here so simply ignore rather than causing a
5208	* corresponding nested-guest #VMEXIT.
5209	*/
5210	return hmR0SvmExitIntr(pVCpu, pCtx, pSvmTransient);
5211	}
5212
5213	case SVM_EXIT_FERR_FREEZE:
5214	{
5215	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_FERR_FREEZE))
5216	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5217	return hmR0SvmExitIntr(pVCpu, pCtx, pSvmTransient);
5218	}
5219
5220	case SVM_EXIT_INVLPG:
5221	{
5222	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_INVLPG))
5223	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5224	return hmR0SvmExitInvlpg(pVCpu, pCtx, pSvmTransient);
5225	}
5226
5227	case SVM_EXIT_WBINVD:
5228	{
5229	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_WBINVD))
5230	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5231	return hmR0SvmExitWbinvd(pVCpu, pCtx, pSvmTransient);
5232	}
5233
5234	case SVM_EXIT_INVD:
5235	{
5236	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_INVD))
5237	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5238	return hmR0SvmExitInvd(pVCpu, pCtx, pSvmTransient);
5239	}
5240
5241	case SVM_EXIT_RDPMC:
5242	{
5243	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_RDPMC))
5244	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5245	return hmR0SvmExitRdpmc(pVCpu, pCtx, pSvmTransient);
5246	}
5247
5248	default:
5249	{
5250	switch (uExitCode)
5251	{
5252	case SVM_EXIT_READ_DR0: case SVM_EXIT_READ_DR1: case SVM_EXIT_READ_DR2: case SVM_EXIT_READ_DR3:
5253	case SVM_EXIT_READ_DR6: case SVM_EXIT_READ_DR7: case SVM_EXIT_READ_DR8: case SVM_EXIT_READ_DR9:
5254	case SVM_EXIT_READ_DR10: case SVM_EXIT_READ_DR11: case SVM_EXIT_READ_DR12: case SVM_EXIT_READ_DR13:
5255	case SVM_EXIT_READ_DR14: case SVM_EXIT_READ_DR15:
5256	{
5257	uint8_t const uDr = uExitCode - SVM_EXIT_READ_DR0;
5258	if (HMIsGuestSvmReadDRxInterceptSet(pVCpu, pCtx, uDr))
5259	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5260	return hmR0SvmExitReadDRx(pVCpu, pCtx, pSvmTransient);
5261	}
5262
5263	case SVM_EXIT_WRITE_DR0: case SVM_EXIT_WRITE_DR1: case SVM_EXIT_WRITE_DR2: case SVM_EXIT_WRITE_DR3:
5264	case SVM_EXIT_WRITE_DR6: case SVM_EXIT_WRITE_DR7: case SVM_EXIT_WRITE_DR8: case SVM_EXIT_WRITE_DR9:
5265	case SVM_EXIT_WRITE_DR10: case SVM_EXIT_WRITE_DR11: case SVM_EXIT_WRITE_DR12: case SVM_EXIT_WRITE_DR13:
5266	case SVM_EXIT_WRITE_DR14: case SVM_EXIT_WRITE_DR15:
5267	{
5268	uint8_t const uDr = uExitCode - SVM_EXIT_WRITE_DR0;
5269	if (HMIsGuestSvmWriteDRxInterceptSet(pVCpu, pCtx, uDr))
5270	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5271	return hmR0SvmExitWriteDRx(pVCpu, pCtx, pSvmTransient);
5272	}
5273
5274	/* The exceptions not handled here are already handled individually above (as they occur more frequently). */
5275	case SVM_EXIT_EXCEPTION_0: /case SVM_EXIT_EXCEPTION_1:/ case SVM_EXIT_EXCEPTION_2:
5276	/case SVM_EXIT_EXCEPTION_3:/ case SVM_EXIT_EXCEPTION_4: case SVM_EXIT_EXCEPTION_5:
5277	/case SVM_EXIT_EXCEPTION_6:/ /case SVM_EXIT_EXCEPTION_7:/ case SVM_EXIT_EXCEPTION_8:
5278	case SVM_EXIT_EXCEPTION_9: case SVM_EXIT_EXCEPTION_10: case SVM_EXIT_EXCEPTION_11:
5279	case SVM_EXIT_EXCEPTION_12: case SVM_EXIT_EXCEPTION_13: /case SVM_EXIT_EXCEPTION_14:/
5280	case SVM_EXIT_EXCEPTION_15: case SVM_EXIT_EXCEPTION_16: /case SVM_EXIT_EXCEPTION_17:/
5281	case SVM_EXIT_EXCEPTION_18: case SVM_EXIT_EXCEPTION_19: case SVM_EXIT_EXCEPTION_20:
5282	case SVM_EXIT_EXCEPTION_21: case SVM_EXIT_EXCEPTION_22: case SVM_EXIT_EXCEPTION_23:
5283	case SVM_EXIT_EXCEPTION_24: case SVM_EXIT_EXCEPTION_25: case SVM_EXIT_EXCEPTION_26:
5284	case SVM_EXIT_EXCEPTION_27: case SVM_EXIT_EXCEPTION_28: case SVM_EXIT_EXCEPTION_29:
5285	case SVM_EXIT_EXCEPTION_30: case SVM_EXIT_EXCEPTION_31:
5286	{
5287	uint8_t const uVector = uExitCode - SVM_EXIT_EXCEPTION_0;
5288	if (HMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, uVector))
5289	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5290	return hmR0SvmExitXcptGeneric(pVCpu, pCtx, pSvmTransient);
5291	}
5292
5293	case SVM_EXIT_XSETBV:
5294	{
5295	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_XSETBV))
5296	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5297	return hmR0SvmExitXsetbv(pVCpu, pCtx, pSvmTransient);
5298	}
5299
5300	case SVM_EXIT_TASK_SWITCH:
5301	{
5302	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_TASK_SWITCH))
5303	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5304	return hmR0SvmExitTaskSwitch(pVCpu, pCtx, pSvmTransient);
5305	}
5306
5307	case SVM_EXIT_IRET:
5308	{
5309	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_IRET))
5310	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5311	return hmR0SvmExitIret(pVCpu, pCtx, pSvmTransient);
5312	}
5313
5314	case SVM_EXIT_SHUTDOWN:
5315	{
5316	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_SHUTDOWN))
5317	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5318	return hmR0SvmExitShutdown(pVCpu, pCtx, pSvmTransient);
5319	}
5320
5321	case SVM_EXIT_VMMCALL:
5322	{
5323	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_VMMCALL))
5324	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5325	return hmR0SvmExitVmmCall(pVCpu, pCtx, pSvmTransient);
5326	}
5327
5328	case SVM_EXIT_CLGI:
5329	{
5330	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_CLGI))
5331	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5332	return hmR0SvmExitClgi(pVCpu, pCtx, pSvmTransient);
5333	}
5334
5335	case SVM_EXIT_STGI:
5336	{
5337	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_STGI))
5338	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5339	return hmR0SvmExitStgi(pVCpu, pCtx, pSvmTransient);
5340	}
5341
5342	case SVM_EXIT_VMLOAD:
5343	{
5344	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_VMLOAD))
5345	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5346	return hmR0SvmExitVmload(pVCpu, pCtx, pSvmTransient);
5347	}
5348
5349	case SVM_EXIT_VMSAVE:
5350	{
5351	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_VMSAVE))
5352	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5353	return hmR0SvmExitVmsave(pVCpu, pCtx, pSvmTransient);
5354	}
5355
5356	case SVM_EXIT_INVLPGA:
5357	{
5358	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_INVLPGA))
5359	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5360	return hmR0SvmExitInvlpga(pVCpu, pCtx, pSvmTransient);
5361	}
5362
5363	case SVM_EXIT_VMRUN:
5364	{
5365	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_VMRUN))
5366	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5367	return hmR0SvmExitVmrun(pVCpu, pCtx, pSvmTransient);
5368	}
5369
5370	case SVM_EXIT_RSM:
5371	{
5372	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_RSM))
5373	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5374	hmR0SvmSetPendingXcptUD(pVCpu);
5375	return VINF_SUCCESS;
5376	}
5377
5378	case SVM_EXIT_SKINIT:
5379	{
5380	if (HMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_SKINIT))
5381	return HM_SVM_VMEXIT_NESTED(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5382	hmR0SvmSetPendingXcptUD(pVCpu);
5383	return VINF_SUCCESS;
5384	}
5385
5386	case SVM_EXIT_INIT: /* We shouldn't get INIT signals while executing a nested-guest. */
5387	case SVM_EXIT_NPF: /* We don't yet support nested-paging for nested-guests, so this should never happen. */
5388	{
5389	return hmR0SvmExitUnexpected(pVCpu, pCtx, pSvmTransient);
5390	}
5391
5392	default:
5393	{
5394	AssertMsgFailed(("hmR0SvmHandleExitNested: Unknown exit code %#x\n", pSvmTransient->u64ExitCode));
5395	pVCpu->hm.s.u32HMError = pSvmTransient->u64ExitCode;
5396	return VERR_SVM_UNKNOWN_EXIT;
5397	}
5398	}
5399	}
5400	}
5401	/* not reached */
5402
5403	#undef HM_SVM_VMEXIT_NESTED
5404	}
5405	#endif
5406
5407
5408	/**
5409	* Handles a guest \#VMEXIT (for all EXITCODE values except SVM_EXIT_INVALID).
5410	*
5411	* @returns VBox status code (informational status codes included).
5412	* @param pVCpu The cross context virtual CPU structure.
5413	* @param pCtx Pointer to the guest-CPU context.
5414	* @param pSvmTransient Pointer to the SVM transient structure.
5415	*/
5416	static int hmR0SvmHandleExit(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
5417	{
5418	Assert(pSvmTransient->u64ExitCode != SVM_EXIT_INVALID);
5419	Assert(pSvmTransient->u64ExitCode <= SVM_EXIT_MAX);
5420
5421	/*
5422	* The ordering of the case labels is based on most-frequently-occurring #VMEXITs for most guests under
5423	* normal workloads (for some definition of "normal").
5424	*/
5425	uint64_t const uExitCode = pSvmTransient->u64ExitCode;
5426	switch (uExitCode)
5427	{
5428	case SVM_EXIT_NPF:
5429	return hmR0SvmExitNestedPF(pVCpu, pCtx, pSvmTransient);
5430
5431	case SVM_EXIT_IOIO:
5432	return hmR0SvmExitIOInstr(pVCpu, pCtx, pSvmTransient);
5433
5434	case SVM_EXIT_RDTSC:
5435	return hmR0SvmExitRdtsc(pVCpu, pCtx, pSvmTransient);
5436
5437	case SVM_EXIT_RDTSCP:
5438	return hmR0SvmExitRdtscp(pVCpu, pCtx, pSvmTransient);
5439
5440	case SVM_EXIT_CPUID:
5441	return hmR0SvmExitCpuid(pVCpu, pCtx, pSvmTransient);
5442
5443	case SVM_EXIT_EXCEPTION_14: /* X86_XCPT_PF */
5444	return hmR0SvmExitXcptPF(pVCpu, pCtx, pSvmTransient);
5445
5446	case SVM_EXIT_EXCEPTION_7: /* X86_XCPT_NM */
5447	return hmR0SvmExitXcptNM(pVCpu, pCtx, pSvmTransient);
5448
5449	case SVM_EXIT_EXCEPTION_6: /* X86_XCPT_UD */
5450	return hmR0SvmExitXcptUD(pVCpu, pCtx, pSvmTransient);
5451
5452	case SVM_EXIT_EXCEPTION_16: /* X86_XCPT_MF */
5453	return hmR0SvmExitXcptMF(pVCpu, pCtx, pSvmTransient);
5454
5455	case SVM_EXIT_EXCEPTION_1: /* X86_XCPT_DB */
5456	return hmR0SvmExitXcptDB(pVCpu, pCtx, pSvmTransient);
5457
5458	case SVM_EXIT_EXCEPTION_17: /* X86_XCPT_AC */
5459	return hmR0SvmExitXcptAC(pVCpu, pCtx, pSvmTransient);
5460
5461	case SVM_EXIT_EXCEPTION_3: /* X86_XCPT_BP */
5462	return hmR0SvmExitXcptBP(pVCpu, pCtx, pSvmTransient);
5463
5464	case SVM_EXIT_MONITOR:
5465	return hmR0SvmExitMonitor(pVCpu, pCtx, pSvmTransient);
5466
5467	case SVM_EXIT_MWAIT:
5468	return hmR0SvmExitMwait(pVCpu, pCtx, pSvmTransient);
5469
5470	case SVM_EXIT_HLT:
5471	return hmR0SvmExitHlt(pVCpu, pCtx, pSvmTransient);
5472
5473	case SVM_EXIT_READ_CR0:
5474	case SVM_EXIT_READ_CR3:
5475	case SVM_EXIT_READ_CR4:
5476	return hmR0SvmExitReadCRx(pVCpu, pCtx, pSvmTransient);
5477
5478	case SVM_EXIT_WRITE_CR0:
5479	case SVM_EXIT_WRITE_CR3:
5480	case SVM_EXIT_WRITE_CR4:
5481	case SVM_EXIT_WRITE_CR8:
5482	{
5483	uint8_t const uCr = uExitCode - SVM_EXIT_WRITE_CR0;
5484	Log4(("hmR0SvmHandleExitNested: Write CR%u\n", uCr)); NOREF(uCr);
5485	return hmR0SvmExitWriteCRx(pVCpu, pCtx, pSvmTransient);
5486	}
5487
5488	case SVM_EXIT_PAUSE:
5489	return hmR0SvmExitPause(pVCpu, pCtx, pSvmTransient);
5490
5491	case SVM_EXIT_VMMCALL:
5492	return hmR0SvmExitVmmCall(pVCpu, pCtx, pSvmTransient);
5493
5494	case SVM_EXIT_VINTR:
5495	return hmR0SvmExitVIntr(pVCpu, pCtx, pSvmTransient);
5496
5497	case SVM_EXIT_INTR:
5498	case SVM_EXIT_FERR_FREEZE:
5499	case SVM_EXIT_NMI:
5500	return hmR0SvmExitIntr(pVCpu, pCtx, pSvmTransient);
5501
5502	case SVM_EXIT_MSR:
5503	return hmR0SvmExitMsr(pVCpu, pCtx, pSvmTransient);
5504
5505	case SVM_EXIT_INVLPG:
5506	return hmR0SvmExitInvlpg(pVCpu, pCtx, pSvmTransient);
5507
5508	case SVM_EXIT_WBINVD:
5509	return hmR0SvmExitWbinvd(pVCpu, pCtx, pSvmTransient);
5510
5511	case SVM_EXIT_INVD:
5512	return hmR0SvmExitInvd(pVCpu, pCtx, pSvmTransient);
5513
5514	case SVM_EXIT_RDPMC:
5515	return hmR0SvmExitRdpmc(pVCpu, pCtx, pSvmTransient);
5516
5517	default:
5518	{
5519	switch (pSvmTransient->u64ExitCode)
5520	{
5521	case SVM_EXIT_READ_DR0: case SVM_EXIT_READ_DR1: case SVM_EXIT_READ_DR2: case SVM_EXIT_READ_DR3:
5522	case SVM_EXIT_READ_DR6: case SVM_EXIT_READ_DR7: case SVM_EXIT_READ_DR8: case SVM_EXIT_READ_DR9:
5523	case SVM_EXIT_READ_DR10: case SVM_EXIT_READ_DR11: case SVM_EXIT_READ_DR12: case SVM_EXIT_READ_DR13:
5524	case SVM_EXIT_READ_DR14: case SVM_EXIT_READ_DR15:
5525	return hmR0SvmExitReadDRx(pVCpu, pCtx, pSvmTransient);
5526
5527	case SVM_EXIT_WRITE_DR0: case SVM_EXIT_WRITE_DR1: case SVM_EXIT_WRITE_DR2: case SVM_EXIT_WRITE_DR3:
5528	case SVM_EXIT_WRITE_DR6: case SVM_EXIT_WRITE_DR7: case SVM_EXIT_WRITE_DR8: case SVM_EXIT_WRITE_DR9:
5529	case SVM_EXIT_WRITE_DR10: case SVM_EXIT_WRITE_DR11: case SVM_EXIT_WRITE_DR12: case SVM_EXIT_WRITE_DR13:
5530	case SVM_EXIT_WRITE_DR14: case SVM_EXIT_WRITE_DR15:
5531	return hmR0SvmExitWriteDRx(pVCpu, pCtx, pSvmTransient);
5532
5533	case SVM_EXIT_XSETBV:
5534	return hmR0SvmExitXsetbv(pVCpu, pCtx, pSvmTransient);
5535
5536	case SVM_EXIT_TASK_SWITCH:
5537	return hmR0SvmExitTaskSwitch(pVCpu, pCtx, pSvmTransient);
5538
5539	case SVM_EXIT_IRET:
5540	return hmR0SvmExitIret(pVCpu, pCtx, pSvmTransient);
5541
5542	case SVM_EXIT_SHUTDOWN:
5543	return hmR0SvmExitShutdown(pVCpu, pCtx, pSvmTransient);
5544
5545	case SVM_EXIT_SMI:
5546	case SVM_EXIT_INIT:
5547	{
5548	/*
5549	* We don't intercept SMIs. As for INIT signals, it really shouldn't ever happen here.
5550	* If it ever does, we want to know about it so log the exit code and bail.
5551	*/
5552	return hmR0SvmExitUnexpected(pVCpu, pCtx, pSvmTransient);
5553	}
5554
5555	#ifdef VBOX_WITH_NESTED_HWVIRT
5556	case SVM_EXIT_CLGI: return hmR0SvmExitClgi(pVCpu, pCtx, pSvmTransient);
5557	case SVM_EXIT_STGI: return hmR0SvmExitStgi(pVCpu, pCtx, pSvmTransient);
5558	case SVM_EXIT_VMLOAD: return hmR0SvmExitVmload(pVCpu, pCtx, pSvmTransient);
5559	case SVM_EXIT_VMSAVE: return hmR0SvmExitVmsave(pVCpu, pCtx, pSvmTransient);
5560	case SVM_EXIT_INVLPGA: return hmR0SvmExitInvlpga(pVCpu, pCtx, pSvmTransient);
5561	case SVM_EXIT_VMRUN: return hmR0SvmExitVmrun(pVCpu, pCtx, pSvmTransient);
5562	#else
5563	case SVM_EXIT_CLGI:
5564	case SVM_EXIT_STGI:
5565	case SVM_EXIT_VMLOAD:
5566	case SVM_EXIT_VMSAVE:
5567	case SVM_EXIT_INVLPGA:
5568	case SVM_EXIT_VMRUN:
5569	#endif
5570	case SVM_EXIT_RSM:
5571	case SVM_EXIT_SKINIT:
5572	{
5573	hmR0SvmSetPendingXcptUD(pVCpu);
5574	return VINF_SUCCESS;
5575	}
5576
5577	#ifdef HMSVM_ALWAYS_TRAP_ALL_XCPTS
5578	case SVM_EXIT_EXCEPTION_0: /* X86_XCPT_DE */
5579	/* SVM_EXIT_EXCEPTION_1: / / X86_XCPT_DB - Handled above. */
5580	case SVM_EXIT_EXCEPTION_2: /* X86_XCPT_NMI */
5581	/* SVM_EXIT_EXCEPTION_3: / / X86_XCPT_BP - Handled above. */
5582	case SVM_EXIT_EXCEPTION_4: /* X86_XCPT_OF */
5583	case SVM_EXIT_EXCEPTION_5: /* X86_XCPT_BR */
5584	/* SVM_EXIT_EXCEPTION_6: / / X86_XCPT_UD - Handled above. */
5585	/* SVM_EXIT_EXCEPTION_7: / / X86_XCPT_NM - Handled above. */
5586	case SVM_EXIT_EXCEPTION_8: /* X86_XCPT_DF */
5587	case SVM_EXIT_EXCEPTION_9: /* X86_XCPT_CO_SEG_OVERRUN */
5588	case SVM_EXIT_EXCEPTION_10: /* X86_XCPT_TS */
5589	case SVM_EXIT_EXCEPTION_11: /* X86_XCPT_NP */
5590	case SVM_EXIT_EXCEPTION_12: /* X86_XCPT_SS */
5591	case SVM_EXIT_EXCEPTION_13: /* X86_XCPT_GP */
5592	/* SVM_EXIT_EXCEPTION_14: / / X86_XCPT_PF - Handled above. */
5593	case SVM_EXIT_EXCEPTION_15: /* Reserved. */
5594	/* SVM_EXIT_EXCEPTION_16: / / X86_XCPT_MF - Handled above. */
5595	/* SVM_EXIT_EXCEPTION_17: / / X86_XCPT_AC - Handled above. */
5596	case SVM_EXIT_EXCEPTION_18: /* X86_XCPT_MC */
5597	case SVM_EXIT_EXCEPTION_19: /* X86_XCPT_XF */
5598	case SVM_EXIT_EXCEPTION_20: case SVM_EXIT_EXCEPTION_21: case SVM_EXIT_EXCEPTION_22:
5599	case SVM_EXIT_EXCEPTION_23: case SVM_EXIT_EXCEPTION_24: case SVM_EXIT_EXCEPTION_25:
5600	case SVM_EXIT_EXCEPTION_26: case SVM_EXIT_EXCEPTION_27: case SVM_EXIT_EXCEPTION_28:
5601	case SVM_EXIT_EXCEPTION_29: case SVM_EXIT_EXCEPTION_30: case SVM_EXIT_EXCEPTION_31:
5602	return hmR0SvmExitXcptGeneric(pVCpu, pCtx, pSvmTransient);
5603	#endif /* HMSVM_ALWAYS_TRAP_ALL_XCPTS */
5604
5605	default:
5606	{
5607	AssertMsgFailed(("hmR0SvmHandleExit: Unknown exit code %#RX64\n", uExitCode));
5608	pVCpu->hm.s.u32HMError = uExitCode;
5609	return VERR_SVM_UNKNOWN_EXIT;
5610	}
5611	}
5612	}
5613	}
5614	/* not reached */
5615	}
5616
5617
5618	#ifdef DEBUG
5619	/* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */
5620	# define HMSVM_ASSERT_PREEMPT_CPUID_VAR() \
5621	RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId()
5622
5623	# define HMSVM_ASSERT_PREEMPT_CPUID() \
5624	do \
5625	{ \
5626	RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \
5627	AssertMsg(idAssertCpu == idAssertCpuNow, ("SVM %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \
5628	} while (0)
5629
5630	# define HMSVM_VALIDATE_EXIT_HANDLER_PARAMS() \
5631	do { \
5632	AssertPtr(pVCpu); \
5633	AssertPtr(pCtx); \
5634	AssertPtr(pSvmTransient); \
5635	Assert(ASMIntAreEnabled()); \
5636	HMSVM_ASSERT_PREEMPT_SAFE(); \
5637	HMSVM_ASSERT_PREEMPT_CPUID_VAR(); \
5638	Log4Func(("vcpu[%u] -v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-\n", (uint32_t)pVCpu->idCpu)); \
5639	HMSVM_ASSERT_PREEMPT_SAFE(); \
5640	if (VMMR0IsLogFlushDisabled(pVCpu)) \
5641	HMSVM_ASSERT_PREEMPT_CPUID(); \
5642	} while (0)
5643	#else /* Release builds */
5644	# define HMSVM_VALIDATE_EXIT_HANDLER_PARAMS() do { NOREF(pVCpu); NOREF(pCtx); NOREF(pSvmTransient); } while (0)
5645	#endif
5646
5647
5648	/**
5649	* Worker for hmR0SvmInterpretInvlpg().
5650	*
5651	* @return VBox status code.
5652	* @param pVCpu The cross context virtual CPU structure.
5653	* @param pCpu Pointer to the disassembler state.
5654	* @param pCtx The guest CPU context.
5655	*/
5656	static int hmR0SvmInterpretInvlPgEx(PVMCPU pVCpu, PDISCPUSTATE pCpu, PCPUMCTX pCtx)
5657	{
5658	DISQPVPARAMVAL Param1;
5659	RTGCPTR GCPtrPage;
5660
5661	int rc = DISQueryParamVal(CPUMCTX2CORE(pCtx), pCpu, &pCpu->Param1, &Param1, DISQPVWHICH_SRC);
5662	if (RT_FAILURE(rc))
5663	return VERR_EM_INTERPRETER;
5664
5665	if ( Param1.type == DISQPV_TYPE_IMMEDIATE
5666	\|\| Param1.type == DISQPV_TYPE_ADDRESS)
5667	{
5668	if (!(Param1.flags & (DISQPV_FLAG_32 \| DISQPV_FLAG_64)))
5669	return VERR_EM_INTERPRETER;
5670
5671	GCPtrPage = Param1.val.val64;
5672	VBOXSTRICTRC rc2 = EMInterpretInvlpg(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx), GCPtrPage);
5673	rc = VBOXSTRICTRC_VAL(rc2);
5674	}
5675	else
5676	{
5677	Log4(("hmR0SvmInterpretInvlPgEx invalid parameter type %#x\n", Param1.type));
5678	rc = VERR_EM_INTERPRETER;
5679	}
5680
5681	return rc;
5682	}
5683
5684
5685	/**
5686	* Interprets INVLPG.
5687	*
5688	* @returns VBox status code.
5689	* @retval VINF_* Scheduling instructions.
5690	* @retval VERR_EM_INTERPRETER Something we can't cope with.
5691	* @retval VERR_* Fatal errors.
5692	*
5693	* @param pVM The cross context VM structure.
5694	* @param pVCpu The cross context virtual CPU structure.
5695	* @param pCtx The guest CPU context.
5696	*
5697	* @remarks Updates the RIP if the instruction was executed successfully.
5698	*/
5699	static int hmR0SvmInterpretInvlpg(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
5700	{
5701	/* Only allow 32 & 64 bit code. */
5702	if (CPUMGetGuestCodeBits(pVCpu) != 16)
5703	{
5704	PDISSTATE pDis = &pVCpu->hm.s.DisState;
5705	int rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, NULL /* pcbInstr */);
5706	if ( RT_SUCCESS(rc)
5707	&& pDis->pCurInstr->uOpcode == OP_INVLPG)
5708	{
5709	rc = hmR0SvmInterpretInvlPgEx(pVCpu, pDis, pCtx);
5710	if (RT_SUCCESS(rc))
5711	pCtx->rip += pDis->cbInstr;
5712	return rc;
5713	}
5714	else
5715	Log4(("hmR0SvmInterpretInvlpg: EMInterpretDisasCurrent returned %Rrc uOpCode=%#x\n", rc, pDis->pCurInstr->uOpcode));
5716	}
5717	return VERR_EM_INTERPRETER;
5718	}
5719
5720
5721	#ifdef HMSVM_USE_IEM_EVENT_REFLECTION
5722	/**
5723	* Gets the IEM exception flags for the specified SVM event.
5724	*
5725	* @returns The IEM exception flags.
5726	* @param pEvent Pointer to the SVM event.
5727	*
5728	* @remarks This function currently only constructs flags required for
5729	* IEMEvaluateRecursiveXcpt and not the complete flags (e.g. error-code
5730	* and CR2 aspects of an exception are not included).
5731	*/
5732	static uint32_t hmR0SvmGetIemXcptFlags(PCSVMEVENT pEvent)
5733	{
5734	uint8_t const uEventType = pEvent->n.u3Type;
5735	uint32_t fIemXcptFlags;
5736	switch (uEventType)
5737	{
5738	case SVM_EVENT_EXCEPTION:
5739	/*
5740	* Only INT3 and INTO instructions can raise #BP and #OF exceptions.
5741	* See AMD spec. Table 8-1. "Interrupt Vector Source and Cause".
5742	*/
5743	if (pEvent->n.u8Vector == X86_XCPT_BP)
5744	{
5745	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT \| IEM_XCPT_FLAGS_BP_INSTR;
5746	break;
5747	}
5748	if (pEvent->n.u8Vector == X86_XCPT_OF)
5749	{
5750	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT \| IEM_XCPT_FLAGS_OF_INSTR;
5751	break;
5752	}
5753	/** @todo How do we distinguish ICEBP \#DB from the regular one? */
5754	RT_FALL_THRU();
5755	case SVM_EVENT_NMI:
5756	fIemXcptFlags = IEM_XCPT_FLAGS_T_CPU_XCPT;
5757	break;
5758
5759	case SVM_EVENT_EXTERNAL_IRQ:
5760	fIemXcptFlags = IEM_XCPT_FLAGS_T_EXT_INT;
5761	break;
5762
5763	case SVM_EVENT_SOFTWARE_INT:
5764	fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
5765	break;
5766
5767	default:
5768	fIemXcptFlags = 0;
5769	AssertMsgFailed(("Unexpected event type! uEventType=%#x uVector=%#x", uEventType, pEvent->n.u8Vector));
5770	break;
5771	}
5772	return fIemXcptFlags;
5773	}
5774
5775	#else
5776	/**
5777	* Determines if an exception is a contributory exception.
5778	*
5779	* Contributory exceptions are ones which can cause double-faults unless the
5780	* original exception was a benign exception. Page-fault is intentionally not
5781	* included here as it's a conditional contributory exception.
5782	*
5783	* @returns @c true if the exception is contributory, @c false otherwise.
5784	* @param uVector The exception vector.
5785	*/
5786	DECLINLINE(bool) hmR0SvmIsContributoryXcpt(const uint32_t uVector)
5787	{
5788	switch (uVector)
5789	{
5790	case X86_XCPT_GP:
5791	case X86_XCPT_SS:
5792	case X86_XCPT_NP:
5793	case X86_XCPT_TS:
5794	case X86_XCPT_DE:
5795	return true;
5796	default:
5797	break;
5798	}
5799	return false;
5800	}
5801	#endif /* HMSVM_USE_IEM_EVENT_REFLECTION */
5802
5803
5804	/**
5805	* Handle a condition that occurred while delivering an event through the guest
5806	* IDT.
5807	*
5808	* @returns VBox status code (informational error codes included).
5809	* @retval VINF_SUCCESS if we should continue handling the \#VMEXIT.
5810	* @retval VINF_HM_DOUBLE_FAULT if a \#DF condition was detected and we ought to
5811	* continue execution of the guest which will delivery the \#DF.
5812	* @retval VINF_EM_RESET if we detected a triple-fault condition.
5813	* @retval VERR_EM_GUEST_CPU_HANG if we detected a guest CPU hang.
5814	*
5815	* @param pVCpu The cross context virtual CPU structure.
5816	* @param pCtx Pointer to the guest-CPU context.
5817	* @param pSvmTransient Pointer to the SVM transient structure.
5818	*
5819	* @remarks No-long-jump zone!!!
5820	*/
5821	static int hmR0SvmCheckExitDueToEventDelivery(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
5822	{
5823	int rc = VINF_SUCCESS;
5824	PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
5825
5826	Log4(("EXITINTINFO: Pending vectoring event %#RX64 Valid=%RTbool ErrValid=%RTbool Err=%#RX32 Type=%u Vector=%u\n",
5827	pVmcb->ctrl.ExitIntInfo.u, !!pVmcb->ctrl.ExitIntInfo.n.u1Valid, !!pVmcb->ctrl.ExitIntInfo.n.u1ErrorCodeValid,
5828	pVmcb->ctrl.ExitIntInfo.n.u32ErrorCode, pVmcb->ctrl.ExitIntInfo.n.u3Type, pVmcb->ctrl.ExitIntInfo.n.u8Vector));
5829
5830	/* See AMD spec. 15.7.3 "EXITINFO Pseudo-Code". The EXITINTINFO (if valid) contains the prior exception (IDT vector)
5831	* that was trying to be delivered to the guest which caused a #VMEXIT which was intercepted (Exit vector). */
5832	if (pVmcb->ctrl.ExitIntInfo.n.u1Valid)
5833	{
5834	#ifdef HMSVM_USE_IEM_EVENT_REFLECTION
5835	IEMXCPTRAISE enmRaise;
5836	IEMXCPTRAISEINFO fRaiseInfo;
5837	bool const fExitIsHwXcpt = pSvmTransient->u64ExitCode - SVM_EXIT_EXCEPTION_0 <= SVM_EXIT_EXCEPTION_31;
5838	uint8_t const uIdtVector = pVmcb->ctrl.ExitIntInfo.n.u8Vector;
5839	if (fExitIsHwXcpt)
5840	{
5841	uint8_t const uExitVector = pSvmTransient->u64ExitCode - SVM_EXIT_EXCEPTION_0;
5842	uint32_t const fIdtVectorFlags = hmR0SvmGetIemXcptFlags(&pVmcb->ctrl.ExitIntInfo);
5843	uint32_t const fExitVectorFlags = IEM_XCPT_FLAGS_T_CPU_XCPT;
5844	enmRaise = IEMEvaluateRecursiveXcpt(pVCpu, fIdtVectorFlags, uIdtVector, fExitVectorFlags, uExitVector, &fRaiseInfo);
5845	}
5846	else
5847	{
5848	/*
5849	* If delivery of an event caused a #VMEXIT that is not an exception (e.g. #NPF) then we
5850	* end up here.
5851	*
5852	* If the event was:
5853	* - a software interrupt, we can re-execute the instruction which will regenerate
5854	* the event.
5855	* - an NMI, we need to clear NMI blocking and re-inject the NMI.
5856	* - a hardware exception or external interrupt, we re-inject it.
5857	*/
5858	fRaiseInfo = IEMXCPTRAISEINFO_NONE;
5859	if (pVmcb->ctrl.ExitIntInfo.n.u3Type == SVM_EVENT_SOFTWARE_INT)
5860	enmRaise = IEMXCPTRAISE_REEXEC_INSTR;
5861	else
5862	enmRaise = IEMXCPTRAISE_PREV_EVENT;
5863	}
5864
5865	switch (enmRaise)
5866	{
5867	case IEMXCPTRAISE_CURRENT_XCPT:
5868	case IEMXCPTRAISE_PREV_EVENT:
5869	{
5870	/* For software interrupts, we shall re-execute the instruction. */
5871	if (!(fRaiseInfo & IEMXCPTRAISEINFO_SOFT_INT_XCPT))
5872	{
5873	RTGCUINTPTR GCPtrFaultAddress = 0;
5874
5875	/* If we are re-injecting an NMI, clear NMI blocking. */
5876	if (pVmcb->ctrl.ExitIntInfo.n.u3Type == SVM_EVENT_NMI)
5877	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
5878
5879	/* Determine a vectoring #PF condition, see comment in hmR0SvmExitXcptPF(). */
5880	if (fRaiseInfo & (IEMXCPTRAISEINFO_EXT_INT_PF \| IEMXCPTRAISEINFO_NMI_PF))
5881	pSvmTransient->fVectoringPF = true;
5882	else if ( pVmcb->ctrl.ExitIntInfo.n.u3Type == SVM_EVENT_EXCEPTION
5883	&& uIdtVector == X86_XCPT_PF)
5884	{
5885	/*
5886	* If the previous exception was a #PF, we need to recover the CR2 value.
5887	* This can't happen with shadow paging.
5888	*/
5889	GCPtrFaultAddress = pCtx->cr2;
5890	}
5891
5892	/*
5893	* Without nested paging, when uExitVector is #PF, CR2 value will be updated from the VMCB's
5894	* exit info. fields, if it's a guest #PF, see hmR0SvmExitXcptPF().
5895	*/
5896	Assert(pVmcb->ctrl.ExitIntInfo.n.u3Type != SVM_EVENT_SOFTWARE_INT);
5897	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingReflect);
5898	hmR0SvmSetPendingEvent(pVCpu, &pVmcb->ctrl.ExitIntInfo, GCPtrFaultAddress);
5899
5900	Log4(("IDT: Pending vectoring event %#RX64 ErrValid=%RTbool Err=%#RX32 GCPtrFaultAddress=%#RX64\n",
5901	pVmcb->ctrl.ExitIntInfo.u, RT_BOOL(pVmcb->ctrl.ExitIntInfo.n.u1ErrorCodeValid),
5902	pVmcb->ctrl.ExitIntInfo.n.u32ErrorCode, GCPtrFaultAddress));
5903	}
5904	break;
5905	}
5906
5907	case IEMXCPTRAISE_REEXEC_INSTR:
5908	{
5909	Assert(rc == VINF_SUCCESS);
5910	break;
5911	}
5912
5913	case IEMXCPTRAISE_DOUBLE_FAULT:
5914	{
5915	/*
5916	* Determing a vectoring double #PF condition. Used later, when PGM evaluates the
5917	* second #PF as a guest #PF (and not a shadow #PF) and needs to be converted into a #DF.
5918	*/
5919	if (fRaiseInfo & IEMXCPTRAISEINFO_PF_PF)
5920	{
5921	pSvmTransient->fVectoringDoublePF = true;
5922	Assert(rc == VINF_SUCCESS);
5923	}
5924	else
5925	{
5926	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingReflect);
5927	hmR0SvmSetPendingXcptDF(pVCpu);
5928	rc = VINF_HM_DOUBLE_FAULT;
5929	}
5930	break;
5931	}
5932
5933	case IEMXCPTRAISE_TRIPLE_FAULT:
5934	{
5935	rc = VINF_EM_RESET;
5936	break;
5937	}
5938
5939	case IEMXCPTRAISE_CPU_HANG:
5940	{
5941	rc = VERR_EM_GUEST_CPU_HANG;
5942	break;
5943	}
5944
5945	default:
5946	{
5947	AssertMsgFailed(("hmR0SvmExitCpuid: EMInterpretCpuId failed with %Rrc\n", rc));
5948	rc = VERR_SVM_IPE_2;
5949	break;
5950	}
5951	}
5952	#else
5953	uint8_t uIdtVector = pVmcb->ctrl.ExitIntInfo.n.u8Vector;
5954
5955	typedef enum
5956	{
5957	SVMREFLECTXCPT_XCPT, /* Reflect the exception to the guest or for further evaluation by VMM. */
5958	SVMREFLECTXCPT_DF, /* Reflect the exception as a double-fault to the guest. */
5959	SVMREFLECTXCPT_TF, /* Indicate a triple faulted state to the VMM. */
5960	SVMREFLECTXCPT_HANG, /* Indicate bad VM trying to deadlock the CPU. */
5961	SVMREFLECTXCPT_NONE /* Nothing to reflect. */
5962	} SVMREFLECTXCPT;
5963
5964	SVMREFLECTXCPT enmReflect = SVMREFLECTXCPT_NONE;
5965	bool fReflectingNmi = false;
5966	if (pVmcb->ctrl.ExitIntInfo.n.u3Type == SVM_EVENT_EXCEPTION)
5967	{
5968	if (pSvmTransient->u64ExitCode - SVM_EXIT_EXCEPTION_0 <= SVM_EXIT_EXCEPTION_31)
5969	{
5970	uint8_t uExitVector = (uint8_t)(pSvmTransient->u64ExitCode - SVM_EXIT_EXCEPTION_0);
5971
5972	#ifdef VBOX_STRICT
5973	if ( hmR0SvmIsContributoryXcpt(uIdtVector)
5974	&& uExitVector == X86_XCPT_PF)
5975	{
5976	Log4(("IDT: Contributory #PF idCpu=%u uCR2=%#RX64\n", pVCpu->idCpu, pCtx->cr2));
5977	}
5978	#endif
5979
5980	if ( uIdtVector == X86_XCPT_BP
5981	\|\| uIdtVector == X86_XCPT_OF)
5982	{
5983	/* Ignore INT3/INTO, just re-execute. See @bugref{8357}. */
5984	}
5985	else if ( uExitVector == X86_XCPT_PF
5986	&& uIdtVector == X86_XCPT_PF)
5987	{
5988	pSvmTransient->fVectoringDoublePF = true;
5989	Log4(("IDT: Vectoring double #PF uCR2=%#RX64\n", pCtx->cr2));
5990	}
5991	else if ( uExitVector == X86_XCPT_AC
5992	&& uIdtVector == X86_XCPT_AC)
5993	{
5994	enmReflect = SVMREFLECTXCPT_HANG;
5995	Log4(("IDT: Nested #AC - Bad guest\n"));
5996	}
5997	else if ( (pVmcb->ctrl.u32InterceptXcpt & HMSVM_CONTRIBUTORY_XCPT_MASK)
5998	&& hmR0SvmIsContributoryXcpt(uExitVector)
5999	&& ( hmR0SvmIsContributoryXcpt(uIdtVector)
6000	\|\| uIdtVector == X86_XCPT_PF))
6001	{
6002	enmReflect = SVMREFLECTXCPT_DF;
6003	Log4(("IDT: Pending vectoring #DF %#RX64 uIdtVector=%#x uExitVector=%#x\n", pVCpu->hm.s.Event.u64IntInfo,
6004	uIdtVector, uExitVector));
6005	}
6006	else if (uIdtVector == X86_XCPT_DF)
6007	{
6008	enmReflect = SVMREFLECTXCPT_TF;
6009	Log4(("IDT: Pending vectoring triple-fault %#RX64 uIdtVector=%#x uExitVector=%#x\n",
6010	pVCpu->hm.s.Event.u64IntInfo, uIdtVector, uExitVector));
6011	}
6012	else
6013	enmReflect = SVMREFLECTXCPT_XCPT;
6014	}
6015	else
6016	{
6017	/*
6018	* If event delivery caused an #VMEXIT that is not an exception (e.g. #NPF) then reflect the original
6019	* exception to the guest after handling the #VMEXIT.
6020	*/
6021	enmReflect = SVMREFLECTXCPT_XCPT;
6022	}
6023	}
6024	else if ( pVmcb->ctrl.ExitIntInfo.n.u3Type == SVM_EVENT_EXTERNAL_IRQ
6025	\|\| pVmcb->ctrl.ExitIntInfo.n.u3Type == SVM_EVENT_NMI)
6026	{
6027	enmReflect = SVMREFLECTXCPT_XCPT;
6028	fReflectingNmi = RT_BOOL(pVmcb->ctrl.ExitIntInfo.n.u3Type == SVM_EVENT_NMI);
6029
6030	if (pSvmTransient->u64ExitCode - SVM_EXIT_EXCEPTION_0 <= SVM_EXIT_EXCEPTION_31)
6031	{
6032	uint8_t uExitVector = (uint8_t)(pSvmTransient->u64ExitCode - SVM_EXIT_EXCEPTION_0);
6033	if (uExitVector == X86_XCPT_PF)
6034	{
6035	pSvmTransient->fVectoringPF = true;
6036	Log4(("IDT: Vectoring #PF due to Ext-Int/NMI. uCR2=%#RX64\n", pCtx->cr2));
6037	}
6038	}
6039	}
6040	/* else: Ignore software interrupts (INT n) as they reoccur when restarting the instruction. */
6041
6042	switch (enmReflect)
6043	{
6044	case SVMREFLECTXCPT_XCPT:
6045	{
6046	/* If we are re-injecting the NMI, clear NMI blocking. */
6047	if (fReflectingNmi)
6048	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
6049
6050	Assert(pVmcb->ctrl.ExitIntInfo.n.u3Type != SVM_EVENT_SOFTWARE_INT);
6051	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingReflect);
6052	hmR0SvmSetPendingEvent(pVCpu, &pVmcb->ctrl.ExitIntInfo, 0 /* GCPtrFaultAddress */);
6053
6054	/* If uExitVector is #PF, CR2 value will be updated from the VMCB if it's a guest #PF. See hmR0SvmExitXcptPF(). */
6055	Log4(("IDT: Pending vectoring event %#RX64 ErrValid=%RTbool Err=%#RX32\n", pVmcb->ctrl.ExitIntInfo.u,
6056	!!pVmcb->ctrl.ExitIntInfo.n.u1ErrorCodeValid, pVmcb->ctrl.ExitIntInfo.n.u32ErrorCode));
6057	break;
6058	}
6059
6060	case SVMREFLECTXCPT_DF:
6061	{
6062	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingReflect);
6063	hmR0SvmSetPendingXcptDF(pVCpu);
6064	rc = VINF_HM_DOUBLE_FAULT;
6065	break;
6066	}
6067
6068	case SVMREFLECTXCPT_TF:
6069	{
6070	rc = VINF_EM_RESET;
6071	break;
6072	}
6073
6074	case SVMREFLECTXCPT_HANG:
6075	{
6076	rc = VERR_EM_GUEST_CPU_HANG;
6077	break;
6078	}
6079
6080	default:
6081	Assert(rc == VINF_SUCCESS);
6082	break;
6083	}
6084	#endif /* HMSVM_USE_IEM_EVENT_REFLECTION */
6085	}
6086	Assert(rc == VINF_SUCCESS \|\| rc == VINF_HM_DOUBLE_FAULT \|\| rc == VINF_EM_RESET \|\| rc == VERR_EM_GUEST_CPU_HANG);
6087	NOREF(pCtx);
6088	return rc;
6089	}
6090
6091
6092	/**
6093	* Advances the guest RIP making use of the CPU's NRIP_SAVE feature if
6094	* supported, otherwise advances the RIP by the number of bytes specified in
6095	* @a cb.
6096	*
6097	* @param pVCpu The cross context virtual CPU structure.
6098	* @param pCtx Pointer to the guest-CPU context.
6099	* @param cb RIP increment value in bytes.
6100	*
6101	* @remarks Use this function only from \#VMEXIT's where the NRIP value is valid
6102	* when NRIP_SAVE is supported by the CPU, otherwise use
6103	* hmR0SvmAdvanceRipDumb!
6104	*/
6105	DECLINLINE(void) hmR0SvmAdvanceRipHwAssist(PVMCPU pVCpu, PCPUMCTX pCtx, uint32_t cb)
6106	{
6107	bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu, pCtx);
6108	if (fSupportsNextRipSave)
6109	{
6110	PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
6111	Assert(pVmcb->ctrl.u64NextRIP);
6112	AssertRelease(pVmcb->ctrl.u64NextRIP - pCtx->rip == cb); /* temporary, remove later */
6113	pCtx->rip = pVmcb->ctrl.u64NextRIP;
6114	}
6115	else
6116	pCtx->rip += cb;
6117
6118	HMSVM_UPDATE_INTR_SHADOW(pVCpu, pCtx);
6119	}
6120
6121
6122	#ifdef VBOX_WITH_NESTED_HWVIRT
6123	/**
6124	* Gets the length of the current instruction if the CPU supports the NRIP_SAVE
6125	* feature. Otherwise, returns the value in @a cbLikely.
6126	*
6127	* @param pVCpu The cross context virtual CPU structure.
6128	* @param pCtx Pointer to the guest-CPU context.
6129	* @param cbLikely The likely instruction length.
6130	*/
6131	DECLINLINE(uint8_t) hmR0SvmGetInstrLengthHwAssist(PVMCPU pVCpu, PCPUMCTX pCtx, uint8_t cbLikely)
6132	{
6133	Assert(cbLikely <= 15); /* See Intel spec. 2.3.11 "AVX Instruction Length" */
6134	bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu, pCtx);
6135	if (fSupportsNextRipSave)
6136	{
6137	PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
6138	uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pCtx->rip;
6139	Assert(cbInstr == cbLikely);
6140	return cbInstr;
6141	}
6142	return cbLikely;
6143	}
6144	#endif
6145
6146
6147	/**
6148	* Advances the guest RIP by the number of bytes specified in @a cb. This does
6149	* not make use of any hardware features to determine the instruction length.
6150	*
6151	* @param pVCpu The cross context virtual CPU structure.
6152	* @param pCtx Pointer to the guest-CPU context.
6153	* @param cb RIP increment value in bytes.
6154	*/
6155	DECLINLINE(void) hmR0SvmAdvanceRipDumb(PVMCPU pVCpu, PCPUMCTX pCtx, uint32_t cb)
6156	{
6157	pCtx->rip += cb;
6158	HMSVM_UPDATE_INTR_SHADOW(pVCpu, pCtx);
6159	}
6160	#undef HMSVM_UPDATE_INTR_SHADOW
6161
6162
6163	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
6164	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- #VMEXIT handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
6165	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
6166
6167	/** @name \#VMEXIT handlers.
6168	* @{
6169	*/
6170
6171	/**
6172	* \#VMEXIT handler for external interrupts, NMIs, FPU assertion freeze and INIT
6173	* signals (SVM_EXIT_INTR, SVM_EXIT_NMI, SVM_EXIT_FERR_FREEZE, SVM_EXIT_INIT).
6174	*/
6175	HMSVM_EXIT_DECL hmR0SvmExitIntr(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6176	{
6177	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6178
6179	if (pSvmTransient->u64ExitCode == SVM_EXIT_NMI)
6180	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC);
6181	else if (pSvmTransient->u64ExitCode == SVM_EXIT_INTR)
6182	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt);
6183
6184	/*
6185	* AMD-V has no preemption timer and the generic periodic preemption timer has no way to signal -before- the timer
6186	* fires if the current interrupt is our own timer or a some other host interrupt. We also cannot examine what
6187	* interrupt it is until the host actually take the interrupt.
6188	*
6189	* Going back to executing guest code here unconditionally causes random scheduling problems (observed on an
6190	* AMD Phenom 9850 Quad-Core on Windows 64-bit host).
6191	*/
6192	return VINF_EM_RAW_INTERRUPT;
6193	}
6194
6195
6196	/**
6197	* \#VMEXIT handler for WBINVD (SVM_EXIT_WBINVD). Conditional \#VMEXIT.
6198	*/
6199	HMSVM_EXIT_DECL hmR0SvmExitWbinvd(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6200	{
6201	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6202
6203	hmR0SvmAdvanceRipHwAssist(pVCpu, pCtx, 2);
6204	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWbinvd);
6205	int rc = VINF_SUCCESS;
6206	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6207	return rc;
6208	}
6209
6210
6211	/**
6212	* \#VMEXIT handler for INVD (SVM_EXIT_INVD). Unconditional \#VMEXIT.
6213	*/
6214	HMSVM_EXIT_DECL hmR0SvmExitInvd(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6215	{
6216	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6217
6218	hmR0SvmAdvanceRipHwAssist(pVCpu, pCtx, 2);
6219	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvd);
6220	int rc = VINF_SUCCESS;
6221	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6222	return rc;
6223	}
6224
6225
6226	/**
6227	* \#VMEXIT handler for INVD (SVM_EXIT_CPUID). Conditional \#VMEXIT.
6228	*/
6229	HMSVM_EXIT_DECL hmR0SvmExitCpuid(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6230	{
6231	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6232	PVM pVM = pVCpu->CTX_SUFF(pVM);
6233	int rc = EMInterpretCpuId(pVM, pVCpu, CPUMCTX2CORE(pCtx));
6234	if (RT_LIKELY(rc == VINF_SUCCESS))
6235	{
6236	hmR0SvmAdvanceRipHwAssist(pVCpu, pCtx, 2);
6237	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6238	}
6239	else
6240	{
6241	AssertMsgFailed(("hmR0SvmExitCpuid: EMInterpretCpuId failed with %Rrc\n", rc));
6242	rc = VERR_EM_INTERPRETER;
6243	}
6244	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCpuid);
6245	return rc;
6246	}
6247
6248
6249	/**
6250	* \#VMEXIT handler for RDTSC (SVM_EXIT_RDTSC). Conditional \#VMEXIT.
6251	*/
6252	HMSVM_EXIT_DECL hmR0SvmExitRdtsc(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6253	{
6254	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6255	PVM pVM = pVCpu->CTX_SUFF(pVM);
6256	int rc = EMInterpretRdtsc(pVM, pVCpu, CPUMCTX2CORE(pCtx));
6257	if (RT_LIKELY(rc == VINF_SUCCESS))
6258	{
6259	pSvmTransient->fUpdateTscOffsetting = true;
6260	hmR0SvmAdvanceRipHwAssist(pVCpu, pCtx, 2);
6261	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6262	}
6263	else
6264	{
6265	AssertMsgFailed(("hmR0SvmExitRdtsc: EMInterpretRdtsc failed with %Rrc\n", rc));
6266	rc = VERR_EM_INTERPRETER;
6267	}
6268	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdtsc);
6269	return rc;
6270	}
6271
6272
6273	/**
6274	* \#VMEXIT handler for RDTSCP (SVM_EXIT_RDTSCP). Conditional \#VMEXIT.
6275	*/
6276	HMSVM_EXIT_DECL hmR0SvmExitRdtscp(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6277	{
6278	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6279	int rc = EMInterpretRdtscp(pVCpu->CTX_SUFF(pVM), pVCpu, pCtx);
6280	if (RT_LIKELY(rc == VINF_SUCCESS))
6281	{
6282	pSvmTransient->fUpdateTscOffsetting = true;
6283	hmR0SvmAdvanceRipHwAssist(pVCpu, pCtx, 3);
6284	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6285	}
6286	else
6287	{
6288	AssertMsgFailed(("hmR0SvmExitRdtsc: EMInterpretRdtscp failed with %Rrc\n", rc));
6289	rc = VERR_EM_INTERPRETER;
6290	}
6291	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdtscp);
6292	return rc;
6293	}
6294
6295
6296	/**
6297	* \#VMEXIT handler for RDPMC (SVM_EXIT_RDPMC). Conditional \#VMEXIT.
6298	*/
6299	HMSVM_EXIT_DECL hmR0SvmExitRdpmc(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6300	{
6301	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6302	int rc = EMInterpretRdpmc(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx));
6303	if (RT_LIKELY(rc == VINF_SUCCESS))
6304	{
6305	hmR0SvmAdvanceRipHwAssist(pVCpu, pCtx, 2);
6306	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6307	}
6308	else
6309	{
6310	AssertMsgFailed(("hmR0SvmExitRdpmc: EMInterpretRdpmc failed with %Rrc\n", rc));
6311	rc = VERR_EM_INTERPRETER;
6312	}
6313	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdpmc);
6314	return rc;
6315	}
6316
6317
6318	/**
6319	* \#VMEXIT handler for INVLPG (SVM_EXIT_INVLPG). Conditional \#VMEXIT.
6320	*/
6321	HMSVM_EXIT_DECL hmR0SvmExitInvlpg(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6322	{
6323	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6324	PVM pVM = pVCpu->CTX_SUFF(pVM);
6325	Assert(!pVM->hm.s.fNestedPaging);
6326	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvlpg);
6327
6328	bool const fSupportsDecodeAssists = hmR0SvmSupportsDecodeAssists(pVCpu, pCtx);
6329	bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu, pCtx);
6330	if ( fSupportsDecodeAssists
6331	&& fSupportsNextRipSave)
6332	{
6333	PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
6334	uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pCtx->rip;
6335	RTGCPTR const GCPtrPage = pVmcb->ctrl.u64ExitInfo1;
6336	VBOXSTRICTRC rcStrict = IEMExecDecodedInvlpg(pVCpu, cbInstr, GCPtrPage);
6337	HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
6338	return VBOXSTRICTRC_VAL(rcStrict);
6339	}
6340
6341	int rc = hmR0SvmInterpretInvlpg(pVM, pVCpu, pCtx); /* Updates RIP if successful. */
6342	Assert(rc == VINF_SUCCESS \|\| rc == VERR_EM_INTERPRETER);
6343	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6344	return rc;
6345	}
6346
6347
6348	/**
6349	* \#VMEXIT handler for HLT (SVM_EXIT_HLT). Conditional \#VMEXIT.
6350	*/
6351	HMSVM_EXIT_DECL hmR0SvmExitHlt(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6352	{
6353	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6354
6355	hmR0SvmAdvanceRipHwAssist(pVCpu, pCtx, 1);
6356	int rc = EMShouldContinueAfterHalt(pVCpu, pCtx) ? VINF_SUCCESS : VINF_EM_HALT;
6357	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6358	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitHlt);
6359	if (rc != VINF_SUCCESS)
6360	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHltToR3);
6361	return rc;
6362	}
6363
6364
6365	/**
6366	* \#VMEXIT handler for MONITOR (SVM_EXIT_MONITOR). Conditional \#VMEXIT.
6367	*/
6368	HMSVM_EXIT_DECL hmR0SvmExitMonitor(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6369	{
6370	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6371	int rc = EMInterpretMonitor(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx));
6372	if (RT_LIKELY(rc == VINF_SUCCESS))
6373	{
6374	hmR0SvmAdvanceRipHwAssist(pVCpu, pCtx, 3);
6375	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6376	}
6377	else
6378	{
6379	AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0SvmExitMonitor: EMInterpretMonitor failed with %Rrc\n", rc));
6380	rc = VERR_EM_INTERPRETER;
6381	}
6382	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMonitor);
6383	return rc;
6384	}
6385
6386
6387	/**
6388	* \#VMEXIT handler for MWAIT (SVM_EXIT_MWAIT). Conditional \#VMEXIT.
6389	*/
6390	HMSVM_EXIT_DECL hmR0SvmExitMwait(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6391	{
6392	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6393	VBOXSTRICTRC rc2 = EMInterpretMWait(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx));
6394	int rc = VBOXSTRICTRC_VAL(rc2);
6395	if ( rc == VINF_EM_HALT
6396	\|\| rc == VINF_SUCCESS)
6397	{
6398	hmR0SvmAdvanceRipHwAssist(pVCpu, pCtx, 3);
6399
6400	if ( rc == VINF_EM_HALT
6401	&& EMMonitorWaitShouldContinue(pVCpu, pCtx))
6402	{
6403	rc = VINF_SUCCESS;
6404	}
6405	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6406	}
6407	else
6408	{
6409	AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0SvmExitMwait: EMInterpretMWait failed with %Rrc\n", rc));
6410	rc = VERR_EM_INTERPRETER;
6411	}
6412	AssertMsg(rc == VINF_SUCCESS \|\| rc == VINF_EM_HALT \|\| rc == VERR_EM_INTERPRETER,
6413	("hmR0SvmExitMwait: EMInterpretMWait failed rc=%Rrc\n", rc));
6414	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMwait);
6415	return rc;
6416	}
6417
6418
6419	/**
6420	* \#VMEXIT handler for shutdown (triple-fault) (SVM_EXIT_SHUTDOWN). Conditional
6421	* \#VMEXIT.
6422	*/
6423	HMSVM_EXIT_DECL hmR0SvmExitShutdown(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6424	{
6425	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6426	return VINF_EM_RESET;
6427	}
6428
6429
6430	/**
6431	* \#VMEXIT handler for unexpected exits. Conditional \#VMEXIT.
6432	*/
6433	HMSVM_EXIT_DECL hmR0SvmExitUnexpected(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6434	{
6435	RT_NOREF(pCtx);
6436	AssertMsgFailed(("hmR0SvmExitUnexpected: ExitCode=%#RX64\n", pSvmTransient->u64ExitCode));
6437	pVCpu->hm.s.u32HMError = (uint32_t)pSvmTransient->u64ExitCode;
6438	return VERR_SVM_UNEXPECTED_EXIT;
6439	}
6440
6441
6442	/**
6443	* \#VMEXIT handler for CRx reads (SVM_EXIT_READ_CR*). Conditional \#VMEXIT.
6444	*/
6445	HMSVM_EXIT_DECL hmR0SvmExitReadCRx(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6446	{
6447	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6448
6449	Log4(("hmR0SvmExitReadCRx: CS:RIP=%04x:%#RX64\n", pCtx->cs.Sel, pCtx->rip));
6450	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxRead[pSvmTransient->u64ExitCode - SVM_EXIT_READ_CR0]);
6451
6452	bool const fSupportsDecodeAssists = hmR0SvmSupportsDecodeAssists(pVCpu, pCtx);
6453	bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu, pCtx);
6454	if ( fSupportsDecodeAssists
6455	&& fSupportsNextRipSave)
6456	{
6457	PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
6458	bool const fMovCRx = RT_BOOL(pVmcb->ctrl.u64ExitInfo1 & SVM_EXIT1_MOV_CRX_MASK);
6459	if (fMovCRx)
6460	{
6461	uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pCtx->rip;
6462	uint8_t const iCrReg = pSvmTransient->u64ExitCode - SVM_EXIT_READ_CR0;
6463	uint8_t const iGReg = pVmcb->ctrl.u64ExitInfo1 & SVM_EXIT1_MOV_CRX_GPR_NUMBER;
6464	VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxRead(pVCpu, cbInstr, iGReg, iCrReg);
6465	HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
6466	return VBOXSTRICTRC_VAL(rcStrict);
6467	}
6468	/* else: SMSW instruction, fall back below to IEM for this. */
6469	}
6470
6471	VBOXSTRICTRC rc2 = EMInterpretInstruction(pVCpu, CPUMCTX2CORE(pCtx), 0 /* pvFault */);
6472	int rc = VBOXSTRICTRC_VAL(rc2);
6473	AssertMsg(rc == VINF_SUCCESS \|\| rc == VERR_EM_INTERPRETER \|\| rc == VINF_PGM_CHANGE_MODE \|\| rc == VINF_PGM_SYNC_CR3,
6474	("hmR0SvmExitReadCRx: EMInterpretInstruction failed rc=%Rrc\n", rc));
6475	Assert((pSvmTransient->u64ExitCode - SVM_EXIT_READ_CR0) <= 15);
6476	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6477	return rc;
6478	}
6479
6480
6481	/**
6482	* \#VMEXIT handler for CRx writes (SVM_EXIT_WRITE_CR*). Conditional \#VMEXIT.
6483	*/
6484	HMSVM_EXIT_DECL hmR0SvmExitWriteCRx(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6485	{
6486	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6487
6488	uint8_t const iCrReg = pSvmTransient->u64ExitCode - SVM_EXIT_WRITE_CR0;
6489	Assert(iCrReg <= 15);
6490
6491	VBOXSTRICTRC rcStrict = VERR_SVM_IPE_5;
6492	bool fDecodedInstr = false;
6493	bool const fSupportsDecodeAssists = hmR0SvmSupportsDecodeAssists(pVCpu, pCtx);
6494	bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu, pCtx);
6495	if ( fSupportsDecodeAssists
6496	&& fSupportsNextRipSave)
6497	{
6498	PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
6499	bool const fMovCRx = RT_BOOL(pVmcb->ctrl.u64ExitInfo1 & SVM_EXIT1_MOV_CRX_MASK);
6500	if (fMovCRx)
6501	{
6502	uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pCtx->rip;
6503	uint8_t const iGReg = pVmcb->ctrl.u64ExitInfo1 & SVM_EXIT1_MOV_CRX_GPR_NUMBER;
6504	Log4(("hmR0SvmExitWriteCRx: Mov CR%u w/ iGReg=%#x\n", iCrReg, iGReg));
6505	rcStrict = IEMExecDecodedMovCRxWrite(pVCpu, cbInstr, iCrReg, iGReg);
6506	fDecodedInstr = true;
6507	}
6508	/* else: LMSW or CLTS instruction, fall back below to IEM for this. */
6509	}
6510
6511	if (!fDecodedInstr)
6512	{
6513	Log4(("hmR0SvmExitWriteCRx: iCrReg=%#x\n", iCrReg));
6514	rcStrict = IEMExecOneBypassEx(pVCpu, CPUMCTX2CORE(pCtx), NULL);
6515	if (RT_UNLIKELY( rcStrict == VERR_IEM_ASPECT_NOT_IMPLEMENTED
6516	\|\| rcStrict == VERR_IEM_INSTR_NOT_IMPLEMENTED))
6517	rcStrict = VERR_EM_INTERPRETER;
6518	}
6519
6520	if (rcStrict == VINF_SUCCESS)
6521	{
6522	switch (iCrReg)
6523	{
6524	case 0: /* CR0. */
6525	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
6526	break;
6527
6528	case 3: /* CR3. */
6529	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR3);
6530	break;
6531
6532	case 4: /* CR4. */
6533	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR4);
6534	break;
6535
6536	case 8: /* CR8 (TPR). */
6537	HMCPU_CF_SET(pVCpu, HM_CHANGED_SVM_GUEST_APIC_STATE);
6538	break;
6539
6540	default:
6541	AssertMsgFailed(("hmR0SvmExitWriteCRx: Invalid/Unexpected Write-CRx exit. u64ExitCode=%#RX64 %#x\n",
6542	pSvmTransient->u64ExitCode, iCrReg));
6543	break;
6544	}
6545	HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
6546	}
6547	else
6548	Assert(rcStrict == VERR_EM_INTERPRETER \|\| rcStrict == VINF_PGM_CHANGE_MODE \|\| rcStrict == VINF_PGM_SYNC_CR3);
6549	return VBOXSTRICTRC_TODO(rcStrict);
6550	}
6551
6552
6553	/**
6554	* \#VMEXIT handler for MSR read and writes (SVM_EXIT_MSR). Conditional
6555	* \#VMEXIT.
6556	*/
6557	HMSVM_EXIT_DECL hmR0SvmExitMsr(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6558	{
6559	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6560	PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
6561	PVM pVM = pVCpu->CTX_SUFF(pVM);
6562
6563	int rc;
6564	if (pVmcb->ctrl.u64ExitInfo1 == SVM_EXIT1_MSR_WRITE)
6565	{
6566	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr);
6567	Log4(("MSR Write: idMsr=%#RX32\n", pCtx->ecx));
6568
6569	/* Handle TPR patching; intercepted LSTAR write. */
6570	if ( pVM->hm.s.fTPRPatchingActive
6571	&& pCtx->ecx == MSR_K8_LSTAR)
6572	{
6573	if ((pCtx->eax & 0xff) != pSvmTransient->u8GuestTpr)
6574	{
6575	/* Our patch code uses LSTAR for TPR caching for 32-bit guests. */
6576	int rc2 = APICSetTpr(pVCpu, pCtx->eax & 0xff);
6577	AssertRC(rc2);
6578	HMCPU_CF_SET(pVCpu, HM_CHANGED_SVM_GUEST_APIC_STATE);
6579	}
6580	rc = VINF_SUCCESS;
6581	hmR0SvmAdvanceRipHwAssist(pVCpu, pCtx, 2);
6582	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6583	return rc;
6584	}
6585
6586	bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu, pCtx);
6587	if (fSupportsNextRipSave)
6588	{
6589	rc = EMInterpretWrmsr(pVM, pVCpu, CPUMCTX2CORE(pCtx));
6590	if (RT_LIKELY(rc == VINF_SUCCESS))
6591	{
6592	pCtx->rip = pVmcb->ctrl.u64NextRIP;
6593	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6594	}
6595	else
6596	AssertMsg( rc == VERR_EM_INTERPRETER
6597	\|\| rc == VINF_CPUM_R3_MSR_WRITE, ("hmR0SvmExitMsr: EMInterpretWrmsr failed rc=%Rrc\n", rc));
6598	}
6599	else
6600	{
6601	rc = VBOXSTRICTRC_TODO(EMInterpretInstruction(pVCpu, CPUMCTX2CORE(pCtx), 0 /* pvFault */));
6602	if (RT_LIKELY(rc == VINF_SUCCESS))
6603	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc); /* RIP updated by EMInterpretInstruction(). */
6604	else
6605	AssertMsg( rc == VERR_EM_INTERPRETER
6606	\|\| rc == VINF_CPUM_R3_MSR_WRITE, ("hmR0SvmExitMsr: WrMsr. EMInterpretInstruction failed rc=%Rrc\n", rc));
6607	}
6608
6609	if (rc == VINF_SUCCESS)
6610	{
6611	/* If this is an X2APIC WRMSR access, update the APIC state as well. */
6612	if ( pCtx->ecx >= MSR_IA32_X2APIC_START
6613	&& pCtx->ecx <= MSR_IA32_X2APIC_END)
6614	{
6615	/*
6616	* We've already saved the APIC related guest-state (TPR) in hmR0SvmPostRunGuest(). When full APIC register
6617	* virtualization is implemented we'll have to make sure APIC state is saved from the VMCB before
6618	* EMInterpretWrmsr() changes it.
6619	*/
6620	HMCPU_CF_SET(pVCpu, HM_CHANGED_SVM_GUEST_APIC_STATE);
6621	}
6622	else
6623	{
6624	switch (pCtx->ecx)
6625	{
6626	case MSR_K6_EFER: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_EFER_MSR); break;
6627	case MSR_IA32_TSC: pSvmTransient->fUpdateTscOffsetting = true; break;
6628	case MSR_K8_FS_BASE:
6629	case MSR_K8_GS_BASE: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS); break;
6630	case MSR_IA32_SYSENTER_CS: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR); break;
6631	case MSR_IA32_SYSENTER_EIP: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); break;
6632	case MSR_IA32_SYSENTER_ESP: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); break;
6633	}
6634	}
6635	}
6636	}
6637	else
6638	{
6639	/* MSR Read access. */
6640	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr);
6641	Assert(pVmcb->ctrl.u64ExitInfo1 == SVM_EXIT1_MSR_READ);
6642	Log4(("MSR Read: idMsr=%#RX32\n", pCtx->ecx));
6643
6644	bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu, pCtx);
6645	if (fSupportsNextRipSave)
6646	{
6647	rc = EMInterpretRdmsr(pVM, pVCpu, CPUMCTX2CORE(pCtx));
6648	if (RT_LIKELY(rc == VINF_SUCCESS))
6649	{
6650	pCtx->rip = pVmcb->ctrl.u64NextRIP;
6651	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6652	}
6653	else
6654	AssertMsg( rc == VERR_EM_INTERPRETER
6655	\|\| rc == VINF_CPUM_R3_MSR_READ, ("hmR0SvmExitMsr: EMInterpretRdmsr failed rc=%Rrc\n", rc));
6656	}
6657	else
6658	{
6659	rc = VBOXSTRICTRC_TODO(EMInterpretInstruction(pVCpu, CPUMCTX2CORE(pCtx), 0));
6660	if (RT_UNLIKELY(rc != VINF_SUCCESS))
6661	{
6662	AssertMsg( rc == VERR_EM_INTERPRETER
6663	\|\| rc == VINF_CPUM_R3_MSR_READ, ("hmR0SvmExitMsr: RdMsr. EMInterpretInstruction failed rc=%Rrc\n", rc));
6664	}
6665	/* RIP updated by EMInterpretInstruction(). */
6666	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6667	}
6668	}
6669
6670	/* RIP has been updated by EMInterpret[Rd\|Wr]msr() or EMInterpretInstruction(). */
6671	return rc;
6672	}
6673
6674
6675	/**
6676	* \#VMEXIT handler for DRx read (SVM_EXIT_READ_DRx). Conditional \#VMEXIT.
6677	*/
6678	HMSVM_EXIT_DECL hmR0SvmExitReadDRx(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6679	{
6680	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6681	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
6682
6683	/** @todo Stepping with nested-guest. */
6684	if (!CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
6685	{
6686	/* We should -not- get this #VMEXIT if the guest's debug registers were active. */
6687	if (pSvmTransient->fWasGuestDebugStateActive)
6688	{
6689	AssertMsgFailed(("hmR0SvmExitReadDRx: Unexpected exit %#RX32\n", (uint32_t)pSvmTransient->u64ExitCode));
6690	pVCpu->hm.s.u32HMError = (uint32_t)pSvmTransient->u64ExitCode;
6691	return VERR_SVM_UNEXPECTED_EXIT;
6692	}
6693
6694	/*
6695	* Lazy DR0-3 loading.
6696	*/
6697	if (!pSvmTransient->fWasHyperDebugStateActive)
6698	{
6699	Assert(!DBGFIsStepping(pVCpu)); Assert(!pVCpu->hm.s.fSingleInstruction);
6700	Log5(("hmR0SvmExitReadDRx: Lazy loading guest debug registers\n"));
6701
6702	/* Don't intercept DRx read and writes. */
6703	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
6704	pVmcb->ctrl.u16InterceptRdDRx = 0;
6705	pVmcb->ctrl.u16InterceptWrDRx = 0;
6706	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
6707
6708	/* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
6709	VMMRZCallRing3Disable(pVCpu);
6710	HM_DISABLE_PREEMPT();
6711
6712	/* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */
6713	CPUMR0LoadGuestDebugState(pVCpu, false /* include DR6 */);
6714	Assert(CPUMIsGuestDebugStateActive(pVCpu) \|\| HC_ARCH_BITS == 32);
6715
6716	HM_RESTORE_PREEMPT();
6717	VMMRZCallRing3Enable(pVCpu);
6718
6719	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch);
6720	return VINF_SUCCESS;
6721	}
6722	}
6723
6724	/*
6725	* Interpret the read/writing of DRx.
6726	*/
6727	/** @todo Decode assist. */
6728	VBOXSTRICTRC rc = EMInterpretInstruction(pVCpu, CPUMCTX2CORE(pCtx), 0 /* pvFault */);
6729	Log5(("hmR0SvmExitReadDRx: Emulated DRx access: rc=%Rrc\n", VBOXSTRICTRC_VAL(rc)));
6730	if (RT_LIKELY(rc == VINF_SUCCESS))
6731	{
6732	/* Not necessary for read accesses but whatever doesn't hurt for now, will be fixed with decode assist. */
6733	/** @todo CPUM should set this flag! */
6734	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
6735	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
6736	}
6737	else
6738	Assert(rc == VERR_EM_INTERPRETER);
6739	return VBOXSTRICTRC_TODO(rc);
6740	}
6741
6742
6743	/**
6744	* \#VMEXIT handler for DRx write (SVM_EXIT_WRITE_DRx). Conditional \#VMEXIT.
6745	*/
6746	HMSVM_EXIT_DECL hmR0SvmExitWriteDRx(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6747	{
6748	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6749	/* For now it's the same since we interpret the instruction anyway. Will change when using of Decode Assist is implemented. */
6750	int rc = hmR0SvmExitReadDRx(pVCpu, pCtx, pSvmTransient);
6751	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
6752	STAM_COUNTER_DEC(&pVCpu->hm.s.StatExitDRxRead);
6753	return rc;
6754	}
6755
6756
6757	/**
6758	* \#VMEXIT handler for XCRx write (SVM_EXIT_XSETBV). Conditional \#VMEXIT.
6759	*/
6760	HMSVM_EXIT_DECL hmR0SvmExitXsetbv(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6761	{
6762	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6763
6764	/** @todo decode assists... */
6765	VBOXSTRICTRC rcStrict = IEMExecOne(pVCpu);
6766	if (rcStrict == VINF_IEM_RAISED_XCPT)
6767	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
6768
6769	pVCpu->hm.s.fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
6770	Log4(("hmR0SvmExitXsetbv: New XCR0=%#RX64 fLoadSaveGuestXcr0=%d (cr4=%RX64) rcStrict=%Rrc\n",
6771	pCtx->aXcr[0], pVCpu->hm.s.fLoadSaveGuestXcr0, pCtx->cr4, VBOXSTRICTRC_VAL(rcStrict)));
6772
6773	HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
6774	return VBOXSTRICTRC_TODO(rcStrict);
6775	}
6776
6777
6778	/**
6779	* \#VMEXIT handler for I/O instructions (SVM_EXIT_IOIO). Conditional \#VMEXIT.
6780	*/
6781	HMSVM_EXIT_DECL hmR0SvmExitIOInstr(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
6782	{
6783	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
6784
6785	/* I/O operation lookup arrays. */
6786	static uint32_t const s_aIOSize[8] = { 0, 1, 2, 0, 4, 0, 0, 0 }; /* Size of the I/O accesses in bytes. */
6787	static uint32_t const s_aIOOpAnd[8] = { 0, 0xff, 0xffff, 0, 0xffffffff, 0, 0, 0 }; /* AND masks for saving
6788	the result (in AL/AX/EAX). */
6789	Log4(("hmR0SvmExitIOInstr: CS:RIP=%04x:%#RX64\n", pCtx->cs.Sel, pCtx->rip));
6790
6791	PVM pVM = pVCpu->CTX_SUFF(pVM);
6792	PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
6793
6794	/* Refer AMD spec. 15.10.2 "IN and OUT Behaviour" and Figure 15-2. "EXITINFO1 for IOIO Intercept" for the format. */
6795	SVMIOIOEXITINFO IoExitInfo;
6796	IoExitInfo.u = (uint32_t)pVmcb->ctrl.u64ExitInfo1;
6797	uint32_t uIOWidth = (IoExitInfo.u >> 4) & 0x7;
6798	uint32_t cbValue = s_aIOSize[uIOWidth];
6799	uint32_t uAndVal = s_aIOOpAnd[uIOWidth];
6800
6801	if (RT_UNLIKELY(!cbValue))
6802	{
6803	AssertMsgFailed(("hmR0SvmExitIOInstr: Invalid IO operation. uIOWidth=%u\n", uIOWidth));
6804	return VERR_EM_INTERPRETER;
6805	}
6806
6807	VBOXSTRICTRC rcStrict;
6808	bool fUpdateRipAlready = false;
6809	if (IoExitInfo.n.u1STR)
6810	{
6811	#ifdef VBOX_WITH_2ND_IEM_STEP
6812	/* INS/OUTS - I/O String instruction. */
6813	/** @todo Huh? why can't we use the segment prefix information given by AMD-V
6814	* in EXITINFO1? Investigate once this thing is up and running. */
6815	Log4(("CS:RIP=%04x:%08RX64 %#06x/%u %c str\n", pCtx->cs.Sel, pCtx->rip, IoExitInfo.n.u16Port, cbValue,
6816	IoExitInfo.n.u1Type == SVM_IOIO_WRITE ? 'w' : 'r'));
6817	AssertReturn(pCtx->dx == IoExitInfo.n.u16Port, VERR_SVM_IPE_2);
6818	static IEMMODE const s_aenmAddrMode[8] =
6819	{
6820	(IEMMODE)-1, IEMMODE_16BIT, IEMMODE_32BIT, (IEMMODE)-1, IEMMODE_64BIT, (IEMMODE)-1, (IEMMODE)-1, (IEMMODE)-1
6821	};
6822	IEMMODE enmAddrMode = s_aenmAddrMode[(IoExitInfo.u >> 7) & 0x7];
6823	if (enmAddrMode != (IEMMODE)-1)
6824	{
6825	uint64_t cbInstr = pVmcb->ctrl.u64ExitInfo2 - pCtx->rip;
6826	if (cbInstr <= 15 && cbInstr >= 1)
6827	{
6828	Assert(cbInstr >= 1U + IoExitInfo.n.u1REP);
6829	if (IoExitInfo.n.u1Type == SVM_IOIO_WRITE)
6830	{
6831	/* Don't know exactly how to detect whether u3SEG is valid, currently
6832	only enabling it for Bulldozer and later with NRIP. OS/2 broke on
6833	2384 Opterons when only checking NRIP. */
6834	bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu, pCtx);
6835	if ( fSupportsNextRipSave
6836	&& pVM->cpum.ro.GuestFeatures.enmMicroarch >= kCpumMicroarch_AMD_15h_First)
6837	{
6838	AssertMsg(IoExitInfo.n.u3SEG == X86_SREG_DS \|\| cbInstr > 1U + IoExitInfo.n.u1REP,
6839	("u32Seg=%d cbInstr=%d u1REP=%d", IoExitInfo.n.u3SEG, cbInstr, IoExitInfo.n.u1REP));
6840	rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, IoExitInfo.n.u1REP, (uint8_t)cbInstr,
6841	IoExitInfo.n.u3SEG, true /fIoChecked/);
6842	}
6843	else if (cbInstr == 1U + IoExitInfo.n.u1REP)
6844	rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, IoExitInfo.n.u1REP, (uint8_t)cbInstr,
6845	X86_SREG_DS, true /fIoChecked/);
6846	else
6847	rcStrict = IEMExecOne(pVCpu);
6848	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringWrite);
6849	}
6850	else
6851	{
6852	AssertMsg(IoExitInfo.n.u3SEG == X86_SREG_ES /=0/, ("%#x\n", IoExitInfo.n.u3SEG));
6853	rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, IoExitInfo.n.u1REP, (uint8_t)cbInstr,
6854	true /fIoChecked/);
6855	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringRead);
6856	}
6857	}
6858	else
6859	{
6860	AssertMsgFailed(("rip=%RX64 nrip=%#RX64 cbInstr=%#RX64\n", pCtx->rip, pVmcb->ctrl.u64ExitInfo2, cbInstr));
6861	rcStrict = IEMExecOne(pVCpu);
6862	}
6863	}
6864	else
6865	{
6866	AssertMsgFailed(("IoExitInfo=%RX64\n", IoExitInfo.u));
6867	rcStrict = IEMExecOne(pVCpu);
6868	}
6869	fUpdateRipAlready = true;
6870
6871	#else
6872	/* INS/OUTS - I/O String instruction. */
6873	PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
6874
6875	/** @todo Huh? why can't we use the segment prefix information given by AMD-V
6876	* in EXITINFO1? Investigate once this thing is up and running. */
6877
6878	rcStrict = EMInterpretDisasCurrent(pVM, pVCpu, pDis, NULL);
6879	if (rcStrict == VINF_SUCCESS)
6880	{
6881	if (IoExitInfo.n.u1Type == SVM_IOIO_WRITE)
6882	{
6883	rcStrict = IOMInterpretOUTSEx(pVM, pVCpu, CPUMCTX2CORE(pCtx), IoExitInfo.n.u16Port, pDis->fPrefix,
6884	(DISCPUMODE)pDis->uAddrMode, cbValue);
6885	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringWrite);
6886	}
6887	else
6888	{
6889	rcStrict = IOMInterpretINSEx(pVM, pVCpu, CPUMCTX2CORE(pCtx), IoExitInfo.n.u16Port, pDis->fPrefix,
6890	(DISCPUMODE)pDis->uAddrMode, cbValue);
6891	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringRead);
6892	}
6893	}
6894	else
6895	rcStrict = VINF_EM_RAW_EMULATE_INSTR;
6896	#endif
6897	}
6898	else
6899	{
6900	/* IN/OUT - I/O instruction. */
6901	Assert(!IoExitInfo.n.u1REP);
6902
6903	if (IoExitInfo.n.u1Type == SVM_IOIO_WRITE)
6904	{
6905	rcStrict = IOMIOPortWrite(pVM, pVCpu, IoExitInfo.n.u16Port, pCtx->eax & uAndVal, cbValue);
6906	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite);
6907	}
6908	else
6909	{
6910	uint32_t u32Val = 0;
6911	rcStrict = IOMIOPortRead(pVM, pVCpu, IoExitInfo.n.u16Port, &u32Val, cbValue);
6912	if (IOM_SUCCESS(rcStrict))
6913	{
6914	/* Save result of I/O IN instr. in AL/AX/EAX. */
6915	/** @todo r=bird: 32-bit op size should clear high bits of rax! */
6916	pCtx->eax = (pCtx->eax & ~uAndVal) \| (u32Val & uAndVal);
6917	}
6918	else if (rcStrict == VINF_IOM_R3_IOPORT_READ)
6919	HMR0SavePendingIOPortRead(pVCpu, pCtx->rip, pVmcb->ctrl.u64ExitInfo2, IoExitInfo.n.u16Port, uAndVal, cbValue);
6920
6921	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead);
6922	}
6923	}
6924
6925	if (IOM_SUCCESS(rcStrict))
6926	{
6927	/* AMD-V saves the RIP of the instruction following the IO instruction in EXITINFO2. */
6928	if (!fUpdateRipAlready)
6929	pCtx->rip = pVmcb->ctrl.u64ExitInfo2;
6930
6931	/*
6932	* If any I/O breakpoints are armed, we need to check if one triggered
6933	* and take appropriate action.
6934	* Note that the I/O breakpoint type is undefined if CR4.DE is 0.
6935	*/
6936	/** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the
6937	* execution engines about whether hyper BPs and such are pending. */
6938	uint32_t const uDr7 = pCtx->dr[7];
6939	if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK)
6940	&& X86_DR7_ANY_RW_IO(uDr7)
6941	&& (pCtx->cr4 & X86_CR4_DE))
6942	\|\| DBGFBpIsHwIoArmed(pVM)))
6943	{
6944	/* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
6945	VMMRZCallRing3Disable(pVCpu);
6946	HM_DISABLE_PREEMPT();
6947
6948	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck);
6949	CPUMR0DebugStateMaybeSaveGuest(pVCpu, false /fDr6/);
6950
6951	VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pCtx, IoExitInfo.n.u16Port, cbValue);
6952	if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
6953	{
6954	/* Raise #DB. */
6955	pVmcb->guest.u64DR6 = pCtx->dr[6];
6956	pVmcb->guest.u64DR7 = pCtx->dr[7];
6957	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_DRX;
6958	hmR0SvmSetPendingXcptDB(pVCpu);
6959	}
6960	/* rcStrict is VINF_SUCCESS, VINF_IOM_R3_IOPORT_COMMIT_WRITE, or in [VINF_EM_FIRST..VINF_EM_LAST],
6961	however we can ditch VINF_IOM_R3_IOPORT_COMMIT_WRITE as it has VMCPU_FF_IOM as backup. */
6962	else if ( rcStrict2 != VINF_SUCCESS
6963	&& (rcStrict == VINF_SUCCESS \|\| rcStrict2 < rcStrict))
6964	rcStrict = rcStrict2;
6965	AssertCompile(VINF_EM_LAST < VINF_IOM_R3_IOPORT_COMMIT_WRITE);
6966
6967	HM_RESTORE_PREEMPT();
6968	VMMRZCallRing3Enable(pVCpu);
6969	}
6970
6971	HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
6972	}
6973
6974	#ifdef VBOX_STRICT
6975	if (rcStrict == VINF_IOM_R3_IOPORT_READ)
6976	Assert(IoExitInfo.n.u1Type == SVM_IOIO_READ);
6977	else if (rcStrict == VINF_IOM_R3_IOPORT_WRITE \|\| rcStrict == VINF_IOM_R3_IOPORT_COMMIT_WRITE)
6978	Assert(IoExitInfo.n.u1Type == SVM_IOIO_WRITE);
6979	else
6980	{
6981	/** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST
6982	* statuses, that the VMM device and some others may return. See
6983	* IOM_SUCCESS() for guidance. */
6984	AssertMsg( RT_FAILURE(rcStrict)
6985	\|\| rcStrict == VINF_SUCCESS
6986	\|\| rcStrict == VINF_EM_RAW_EMULATE_INSTR
6987	\|\| rcStrict == VINF_EM_DBG_BREAKPOINT
6988	\|\| rcStrict == VINF_EM_RAW_GUEST_TRAP
6989	\|\| rcStrict == VINF_EM_RAW_TO_R3
6990	\|\| rcStrict == VINF_TRPM_XCPT_DISPATCHED
6991	\|\| rcStrict == VINF_EM_TRIPLE_FAULT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
6992	}
6993	#endif
6994	return VBOXSTRICTRC_TODO(rcStrict);
6995	}
6996
6997
6998	/**
6999	* \#VMEXIT handler for Nested Page-faults (SVM_EXIT_NPF). Conditional \#VMEXIT.
7000	*/
7001	HMSVM_EXIT_DECL hmR0SvmExitNestedPF(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7002	{
7003	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7004	HMSVM_ASSERT_NOT_IN_NESTED_GUEST(pCtx);
7005
7006	PVM pVM = pVCpu->CTX_SUFF(pVM);
7007	Assert(pVM->hm.s.fNestedPaging);
7008
7009	HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY();
7010
7011	/* See AMD spec. 15.25.6 "Nested versus Guest Page Faults, Fault Ordering" for VMCB details for #NPF. */
7012	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
7013	uint32_t u32ErrCode = pVmcb->ctrl.u64ExitInfo1;
7014	RTGCPHYS GCPhysFaultAddr = pVmcb->ctrl.u64ExitInfo2;
7015
7016	Log4(("#NPF at CS:RIP=%04x:%#RX64 faultaddr=%RGp errcode=%#x \n", pCtx->cs.Sel, pCtx->rip, GCPhysFaultAddr, u32ErrCode));
7017
7018	#ifdef VBOX_HM_WITH_GUEST_PATCHING
7019	/* TPR patching for 32-bit guests, using the reserved bit in the page tables for MMIO regions. */
7020	if ( pVM->hm.s.fTprPatchingAllowed
7021	&& (GCPhysFaultAddr & PAGE_OFFSET_MASK) == XAPIC_OFF_TPR
7022	&& ( !(u32ErrCode & X86_TRAP_PF_P) /* Not present */
7023	\|\| (u32ErrCode & (X86_TRAP_PF_P \| X86_TRAP_PF_RSVD)) == (X86_TRAP_PF_P \| X86_TRAP_PF_RSVD)) /* MMIO page. */
7024	&& !CPUMIsGuestInLongModeEx(pCtx)
7025	&& !CPUMGetGuestCPL(pVCpu)
7026	&& pVM->hm.s.cPatches < RT_ELEMENTS(pVM->hm.s.aPatches))
7027	{
7028	RTGCPHYS GCPhysApicBase = APICGetBaseMsrNoCheck(pVCpu);
7029	GCPhysApicBase &= PAGE_BASE_GC_MASK;
7030
7031	if (GCPhysFaultAddr == GCPhysApicBase + XAPIC_OFF_TPR)
7032	{
7033	/* Only attempt to patch the instruction once. */
7034	PHMTPRPATCH pPatch = (PHMTPRPATCH)RTAvloU32Get(&pVM->hm.s.PatchTree, (AVLOU32KEY)pCtx->eip);
7035	if (!pPatch)
7036	return VINF_EM_HM_PATCH_TPR_INSTR;
7037	}
7038	}
7039	#endif
7040
7041	/*
7042	* Determine the nested paging mode.
7043	*/
7044	PGMMODE enmNestedPagingMode;
7045	#if HC_ARCH_BITS == 32
7046	if (CPUMIsGuestInLongModeEx(pCtx))
7047	enmNestedPagingMode = PGMMODE_AMD64_NX;
7048	else
7049	#endif
7050	enmNestedPagingMode = PGMGetHostMode(pVM);
7051
7052	/*
7053	* MMIO optimization using the reserved (RSVD) bit in the guest page tables for MMIO pages.
7054	*/
7055	int rc;
7056	Assert((u32ErrCode & (X86_TRAP_PF_RSVD \| X86_TRAP_PF_P)) != X86_TRAP_PF_RSVD);
7057	if ((u32ErrCode & (X86_TRAP_PF_RSVD \| X86_TRAP_PF_P)) == (X86_TRAP_PF_RSVD \| X86_TRAP_PF_P))
7058	{
7059	/* If event delivery causes an MMIO #NPF, go back to instruction emulation as
7060	otherwise injecting the original pending event would most likely cause the same MMIO #NPF. */
7061	if (pVCpu->hm.s.Event.fPending)
7062	return VINF_EM_RAW_INJECT_TRPM_EVENT;
7063
7064	VBOXSTRICTRC rc2 = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, enmNestedPagingMode, CPUMCTX2CORE(pCtx), GCPhysFaultAddr,
7065	u32ErrCode);
7066	rc = VBOXSTRICTRC_VAL(rc2);
7067
7068	/*
7069	* If we succeed, resume guest execution.
7070	* If we fail in interpreting the instruction because we couldn't get the guest physical address
7071	* of the page containing the instruction via the guest's page tables (we would invalidate the guest page
7072	* in the host TLB), resume execution which would cause a guest page fault to let the guest handle this
7073	* weird case. See @bugref{6043}.
7074	*/
7075	if ( rc == VINF_SUCCESS
7076	\|\| rc == VERR_PAGE_TABLE_NOT_PRESENT
7077	\|\| rc == VERR_PAGE_NOT_PRESENT)
7078	{
7079	/* Successfully handled MMIO operation. */
7080	HMCPU_CF_SET(pVCpu, HM_CHANGED_SVM_GUEST_APIC_STATE);
7081	rc = VINF_SUCCESS;
7082	}
7083	return rc;
7084	}
7085
7086	TRPMAssertXcptPF(pVCpu, GCPhysFaultAddr, u32ErrCode);
7087	rc = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, enmNestedPagingMode, u32ErrCode, CPUMCTX2CORE(pCtx), GCPhysFaultAddr);
7088	TRPMResetTrap(pVCpu);
7089
7090	Log4(("#NPF: PGMR0Trap0eHandlerNestedPaging returned %Rrc CS:RIP=%04x:%#RX64\n", rc, pCtx->cs.Sel, pCtx->rip));
7091
7092	/*
7093	* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}.
7094	*/
7095	if ( rc == VINF_SUCCESS
7096	\|\| rc == VERR_PAGE_TABLE_NOT_PRESENT
7097	\|\| rc == VERR_PAGE_NOT_PRESENT)
7098	{
7099	/* We've successfully synced our shadow page tables. */
7100	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
7101	rc = VINF_SUCCESS;
7102	}
7103
7104	return rc;
7105	}
7106
7107
7108	/**
7109	* \#VMEXIT handler for virtual interrupt (SVM_EXIT_VINTR). Conditional
7110	* \#VMEXIT.
7111	*/
7112	HMSVM_EXIT_DECL hmR0SvmExitVIntr(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7113	{
7114	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7115	HMSVM_ASSERT_NOT_IN_NESTED_GUEST(pCtx);
7116
7117	/* Indicate that we no longer need to #VMEXIT when the guest is ready to receive NMIs, it is now ready. */
7118	PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
7119	hmR0SvmClearVirtIntrIntercept(pVmcb);
7120
7121	/* Deliver the pending interrupt via hmR0SvmEvaluatePendingEvent() and resume guest execution. */
7122	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow);
7123	return VINF_SUCCESS;
7124	}
7125
7126
7127	/**
7128	* \#VMEXIT handler for task switches (SVM_EXIT_TASK_SWITCH). Conditional
7129	* \#VMEXIT.
7130	*/
7131	HMSVM_EXIT_DECL hmR0SvmExitTaskSwitch(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7132	{
7133	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7134
7135	HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY();
7136
7137	#ifndef HMSVM_ALWAYS_TRAP_TASK_SWITCH
7138	Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
7139	#endif
7140
7141	/* Check if this task-switch occurred while delivering an event through the guest IDT. */
7142	if (pVCpu->hm.s.Event.fPending) /* Can happen with exceptions/NMI. See @bugref{8411}. */
7143	{
7144	/*
7145	* AMD-V provides us with the exception which caused the TS; we collect
7146	* the information in the call to hmR0SvmCheckExitDueToEventDelivery.
7147	*/
7148	Log4(("hmR0SvmExitTaskSwitch: TS occurred during event delivery.\n"));
7149	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
7150	return VINF_EM_RAW_INJECT_TRPM_EVENT;
7151	}
7152
7153	/** @todo Emulate task switch someday, currently just going back to ring-3 for
7154	* emulation. */
7155	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
7156	return VERR_EM_INTERPRETER;
7157	}
7158
7159
7160	/**
7161	* \#VMEXIT handler for VMMCALL (SVM_EXIT_VMMCALL). Conditional \#VMEXIT.
7162	*/
7163	HMSVM_EXIT_DECL hmR0SvmExitVmmCall(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7164	{
7165	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7166	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitVmcall);
7167
7168	bool fRipUpdated;
7169	VBOXSTRICTRC rcStrict = HMSvmVmmcall(pVCpu, pCtx, &fRipUpdated);
7170	if (RT_SUCCESS(rcStrict))
7171	{
7172	/* Only update the RIP if we're continuing guest execution and not
7173	in the case of say VINF_GIM_R3_HYPERCALL. */
7174	if ( rcStrict == VINF_SUCCESS
7175	&& !fRipUpdated)
7176	{
7177	hmR0SvmAdvanceRipHwAssist(pVCpu, pCtx, 3 /* cbInstr */);
7178	}
7179
7180	/* If the hypercall or TPR patching changes anything other than guest's general-purpose registers,
7181	we would need to reload the guest changed bits here before VM-entry. */
7182	return VBOXSTRICTRC_VAL(rcStrict);
7183	}
7184
7185	hmR0SvmSetPendingXcptUD(pVCpu);
7186	return VINF_SUCCESS;
7187	}
7188
7189
7190	/**
7191	* \#VMEXIT handler for VMMCALL (SVM_EXIT_VMMCALL). Conditional \#VMEXIT.
7192	*/
7193	HMSVM_EXIT_DECL hmR0SvmExitPause(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7194	{
7195	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7196	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPause);
7197	return VINF_EM_RAW_INTERRUPT;
7198	}
7199
7200
7201	/**
7202	* \#VMEXIT handler for IRET (SVM_EXIT_IRET). Conditional \#VMEXIT.
7203	*/
7204	HMSVM_EXIT_DECL hmR0SvmExitIret(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7205	{
7206	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7207
7208	/* Clear NMI blocking. */
7209	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
7210
7211	/* Indicate that we no longer need to #VMEXIT when the guest is ready to receive NMIs, it is now ready. */
7212	PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
7213	hmR0SvmClearIretIntercept(pVmcb);
7214
7215	/* Deliver the pending NMI via hmR0SvmEvaluatePendingEvent() and resume guest execution. */
7216	return VINF_SUCCESS;
7217	}
7218
7219
7220	/**
7221	* \#VMEXIT handler for page-fault exceptions (SVM_EXIT_EXCEPTION_14).
7222	* Conditional \#VMEXIT.
7223	*/
7224	HMSVM_EXIT_DECL hmR0SvmExitXcptPF(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7225	{
7226	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7227	HMSVM_ASSERT_NOT_IN_NESTED_GUEST(pCtx);
7228
7229	HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY();
7230
7231	/* See AMD spec. 15.12.15 "#PF (Page Fault)". */
7232	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
7233	uint32_t u32ErrCode = pVmcb->ctrl.u64ExitInfo1;
7234	RTGCUINTPTR uFaultAddress = pVmcb->ctrl.u64ExitInfo2;
7235	PVM pVM = pVCpu->CTX_SUFF(pVM);
7236
7237	#if defined(HMSVM_ALWAYS_TRAP_ALL_XCPTS) \|\| defined(HMSVM_ALWAYS_TRAP_PF)
7238	if (pVM->hm.s.fNestedPaging)
7239	{
7240	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
7241	if (!pSvmTransient->fVectoringDoublePF)
7242	{
7243	/* A genuine guest #PF, reflect it to the guest. */
7244	hmR0SvmSetPendingXcptPF(pVCpu, pCtx, u32ErrCode, uFaultAddress);
7245	Log4(("#PF: Guest page fault at %04X:%RGv FaultAddr=%RGv ErrCode=%#x\n", pCtx->cs.Sel, (RTGCPTR)pCtx->rip,
7246	uFaultAddress, u32ErrCode));
7247	}
7248	else
7249	{
7250	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
7251	hmR0SvmSetPendingXcptDF(pVCpu);
7252	Log4(("Pending #DF due to vectoring #PF. NP\n"));
7253	}
7254	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
7255	return VINF_SUCCESS;
7256	}
7257	#endif
7258
7259	Assert(!pVM->hm.s.fNestedPaging);
7260
7261	#ifdef VBOX_HM_WITH_GUEST_PATCHING
7262	/* Shortcut for APIC TPR reads and writes; only applicable to 32-bit guests. */
7263	if ( pVM->hm.s.fTprPatchingAllowed
7264	&& (uFaultAddress & 0xfff) == XAPIC_OFF_TPR
7265	&& !(u32ErrCode & X86_TRAP_PF_P) /* Not present. */
7266	&& !CPUMIsGuestInLongModeEx(pCtx)
7267	&& !CPUMGetGuestCPL(pVCpu)
7268	&& pVM->hm.s.cPatches < RT_ELEMENTS(pVM->hm.s.aPatches))
7269	{
7270	RTGCPHYS GCPhysApicBase;
7271	GCPhysApicBase = APICGetBaseMsrNoCheck(pVCpu);
7272	GCPhysApicBase &= PAGE_BASE_GC_MASK;
7273
7274	/* Check if the page at the fault-address is the APIC base. */
7275	RTGCPHYS GCPhysPage;
7276	int rc2 = PGMGstGetPage(pVCpu, (RTGCPTR)uFaultAddress, NULL /* pfFlags */, &GCPhysPage);
7277	if ( rc2 == VINF_SUCCESS
7278	&& GCPhysPage == GCPhysApicBase)
7279	{
7280	/* Only attempt to patch the instruction once. */
7281	PHMTPRPATCH pPatch = (PHMTPRPATCH)RTAvloU32Get(&pVM->hm.s.PatchTree, (AVLOU32KEY)pCtx->eip);
7282	if (!pPatch)
7283	return VINF_EM_HM_PATCH_TPR_INSTR;
7284	}
7285	}
7286	#endif
7287
7288	Log4(("#PF: uFaultAddress=%#RX64 CS:RIP=%#04x:%#RX64 u32ErrCode %#RX32 cr3=%#RX64\n", uFaultAddress, pCtx->cs.Sel,
7289	pCtx->rip, u32ErrCode, pCtx->cr3));
7290
7291	/* If it's a vectoring #PF, emulate injecting the original event injection as PGMTrap0eHandler() is incapable
7292	of differentiating between instruction emulation and event injection that caused a #PF. See @bugref{6607}. */
7293	if (pSvmTransient->fVectoringPF)
7294	{
7295	Assert(pVCpu->hm.s.Event.fPending);
7296	return VINF_EM_RAW_INJECT_TRPM_EVENT;
7297	}
7298
7299	TRPMAssertXcptPF(pVCpu, uFaultAddress, u32ErrCode);
7300	int rc = PGMTrap0eHandler(pVCpu, u32ErrCode, CPUMCTX2CORE(pCtx), (RTGCPTR)uFaultAddress);
7301
7302	Log4(("#PF rc=%Rrc\n", rc));
7303
7304	if (rc == VINF_SUCCESS)
7305	{
7306	/* Successfully synced shadow pages tables or emulated an MMIO instruction. */
7307	TRPMResetTrap(pVCpu);
7308	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
7309	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
7310	return rc;
7311	}
7312	else if (rc == VINF_EM_RAW_GUEST_TRAP)
7313	{
7314	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
7315
7316	if (!pSvmTransient->fVectoringDoublePF)
7317	{
7318	/* It's a guest page fault and needs to be reflected to the guest. */
7319	u32ErrCode = TRPMGetErrorCode(pVCpu); /* The error code might have been changed. */
7320	TRPMResetTrap(pVCpu);
7321	hmR0SvmSetPendingXcptPF(pVCpu, pCtx, u32ErrCode, uFaultAddress);
7322	}
7323	else
7324	{
7325	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
7326	TRPMResetTrap(pVCpu);
7327	hmR0SvmSetPendingXcptDF(pVCpu);
7328	Log4(("#PF: Pending #DF due to vectoring #PF\n"));
7329	}
7330
7331	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
7332	return VINF_SUCCESS;
7333	}
7334
7335	TRPMResetTrap(pVCpu);
7336	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
7337	return rc;
7338	}
7339
7340
7341	/**
7342	* \#VMEXIT handler for device-not-available exceptions (SVM_EXIT_EXCEPTION_7).
7343	* Conditional \#VMEXIT.
7344	*/
7345	HMSVM_EXIT_DECL hmR0SvmExitXcptNM(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7346	{
7347	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7348
7349	/* Paranoia; Ensure we cannot be called as a result of event delivery. */
7350	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
7351	Assert(!pVmcb->ctrl.ExitIntInfo.n.u1Valid); NOREF(pVmcb);
7352
7353	/* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
7354	VMMRZCallRing3Disable(pVCpu);
7355	HM_DISABLE_PREEMPT();
7356
7357	int rc;
7358	/* If the guest FPU was active at the time of the #NM exit, then it's a guest fault. */
7359	if (pSvmTransient->fWasGuestFPUStateActive)
7360	{
7361	rc = VINF_EM_RAW_GUEST_TRAP;
7362	Assert(CPUMIsGuestFPUStateActive(pVCpu) \|\| HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0));
7363	}
7364	else
7365	{
7366	#ifndef HMSVM_ALWAYS_TRAP_ALL_XCPTS
7367	Assert(!pSvmTransient->fWasGuestFPUStateActive);
7368	#endif
7369	rc = CPUMR0Trap07Handler(pVCpu->CTX_SUFF(pVM), pVCpu); /* (No need to set HM_CHANGED_HOST_CONTEXT for SVM.) */
7370	Assert( rc == VINF_EM_RAW_GUEST_TRAP
7371	\|\| ((rc == VINF_SUCCESS \|\| rc == VINF_CPUM_HOST_CR0_MODIFIED) && CPUMIsGuestFPUStateActive(pVCpu)));
7372	}
7373
7374	HM_RESTORE_PREEMPT();
7375	VMMRZCallRing3Enable(pVCpu);
7376
7377	if (rc == VINF_SUCCESS \|\| rc == VINF_CPUM_HOST_CR0_MODIFIED)
7378	{
7379	/* Guest FPU state was activated, we'll want to change CR0 FPU intercepts before the next VM-reentry. */
7380	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
7381	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowNM);
7382	pVCpu->hm.s.fPreloadGuestFpu = true;
7383	}
7384	else
7385	{
7386	/* Forward #NM to the guest. */
7387	Assert(rc == VINF_EM_RAW_GUEST_TRAP);
7388	hmR0SvmSetPendingXcptNM(pVCpu);
7389	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNM);
7390	}
7391	return VINF_SUCCESS;
7392	}
7393
7394
7395	/**
7396	* \#VMEXIT handler for undefined opcode (SVM_EXIT_EXCEPTION_6).
7397	* Conditional \#VMEXIT.
7398	*/
7399	HMSVM_EXIT_DECL hmR0SvmExitXcptUD(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7400	{
7401	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7402
7403	/* Paranoia; Ensure we cannot be called as a result of event delivery. */
7404	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
7405	Assert(!pVmcb->ctrl.ExitIntInfo.n.u1Valid); NOREF(pVmcb);
7406
7407	int rc = VERR_SVM_UNEXPECTED_XCPT_EXIT;
7408	if (pVCpu->hm.s.fGIMTrapXcptUD)
7409	{
7410	uint8_t cbInstr = 0;
7411	VBOXSTRICTRC rcStrict = GIMXcptUD(pVCpu, pCtx, NULL /* pDis */, &cbInstr);
7412	if (rcStrict == VINF_SUCCESS)
7413	{
7414	/* #UD #VMEXIT does not have valid NRIP information, manually advance RIP. See @bugref{7270#c170}. */
7415	hmR0SvmAdvanceRipDumb(pVCpu, pCtx, cbInstr);
7416	rc = VINF_SUCCESS;
7417	HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
7418	}
7419	else if (rcStrict == VINF_GIM_HYPERCALL_CONTINUING)
7420	rc = VINF_SUCCESS;
7421	else if (rcStrict == VINF_GIM_R3_HYPERCALL)
7422	rc = VINF_GIM_R3_HYPERCALL;
7423	else
7424	Assert(RT_FAILURE(VBOXSTRICTRC_VAL(rcStrict)));
7425	}
7426
7427	/* If the GIM #UD exception handler didn't succeed for some reason or wasn't needed, raise #UD. */
7428	if (RT_FAILURE(rc))
7429	{
7430	hmR0SvmSetPendingXcptUD(pVCpu);
7431	rc = VINF_SUCCESS;
7432	}
7433
7434	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD);
7435	return rc;
7436	}
7437
7438
7439	/**
7440	* \#VMEXIT handler for math-fault exceptions (SVM_EXIT_EXCEPTION_16).
7441	* Conditional \#VMEXIT.
7442	*/
7443	HMSVM_EXIT_DECL hmR0SvmExitXcptMF(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7444	{
7445	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7446
7447	/* Paranoia; Ensure we cannot be called as a result of event delivery. */
7448	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
7449	Assert(!pVmcb->ctrl.ExitIntInfo.n.u1Valid); NOREF(pVmcb);
7450
7451	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
7452
7453	if (!(pCtx->cr0 & X86_CR0_NE))
7454	{
7455	PVM pVM = pVCpu->CTX_SUFF(pVM);
7456	PDISSTATE pDis = &pVCpu->hm.s.DisState;
7457	unsigned cbOp;
7458	int rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp);
7459	if (RT_SUCCESS(rc))
7460	{
7461	/* Convert a #MF into a FERR -> IRQ 13. See @bugref{6117}. */
7462	/** @todo FERR intercept when in nested-guest mode? */
7463	rc = PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13, 1, 0 /* uTagSrc */);
7464	if (RT_SUCCESS(rc))
7465	pCtx->rip += cbOp;
7466	}
7467	else
7468	Log4(("hmR0SvmExitXcptMF: EMInterpretDisasCurrent returned %Rrc uOpCode=%#x\n", rc, pDis->pCurInstr->uOpcode));
7469	return rc;
7470	}
7471
7472	hmR0SvmSetPendingXcptMF(pVCpu);
7473	return VINF_SUCCESS;
7474	}
7475
7476
7477	/**
7478	* \#VMEXIT handler for debug exceptions (SVM_EXIT_EXCEPTION_1). Conditional
7479	* \#VMEXIT.
7480	*/
7481	HMSVM_EXIT_DECL hmR0SvmExitXcptDB(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7482	{
7483	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7484
7485	/* If this #DB is the result of delivering an event, go back to the interpreter. */
7486	HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY();
7487	if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
7488	{
7489	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingInterpret);
7490	return VINF_EM_RAW_INJECT_TRPM_EVENT;
7491	}
7492
7493	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
7494
7495	/* This can be a fault-type #DB (instruction breakpoint) or a trap-type #DB (data breakpoint). However, for both cases
7496	DR6 and DR7 are updated to what the exception handler expects. See AMD spec. 15.12.2 "#DB (Debug)". */
7497	PVM pVM = pVCpu->CTX_SUFF(pVM);
7498	PSVMVMCB pVmcb = pVCpu->hm.s.svm.pVmcb;
7499	int rc = DBGFRZTrap01Handler(pVM, pVCpu, CPUMCTX2CORE(pCtx), pVmcb->guest.u64DR6, pVCpu->hm.s.fSingleInstruction);
7500	if (rc == VINF_EM_RAW_GUEST_TRAP)
7501	{
7502	Log5(("hmR0SvmExitXcptDB: DR6=%#RX64 -> guest trap\n", pVmcb->guest.u64DR6));
7503	if (CPUMIsHyperDebugStateActive(pVCpu))
7504	CPUMSetGuestDR6(pVCpu, CPUMGetGuestDR6(pVCpu) \| pVmcb->guest.u64DR6);
7505
7506	/* Reflect the exception back to the guest. */
7507	hmR0SvmSetPendingXcptDB(pVCpu);
7508	rc = VINF_SUCCESS;
7509	}
7510
7511	/*
7512	* Update DR6.
7513	*/
7514	if (CPUMIsHyperDebugStateActive(pVCpu))
7515	{
7516	Log5(("hmR0SvmExitXcptDB: DR6=%#RX64 -> %Rrc\n", pVmcb->guest.u64DR6, rc));
7517	pVmcb->guest.u64DR6 = X86_DR6_INIT_VAL;
7518	pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_DRX;
7519	}
7520	else
7521	{
7522	AssertMsg(rc == VINF_SUCCESS, ("rc=%Rrc\n", rc));
7523	Assert(!pVCpu->hm.s.fSingleInstruction && !DBGFIsStepping(pVCpu));
7524	}
7525
7526	return rc;
7527	}
7528
7529
7530	/**
7531	* \#VMEXIT handler for alignment check exceptions (SVM_EXIT_EXCEPTION_17).
7532	* Conditional \#VMEXIT.
7533	*/
7534	HMSVM_EXIT_DECL hmR0SvmExitXcptAC(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7535	{
7536	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7537
7538	/** @todo if triple-fault is returned in nested-guest scenario convert to a
7539	* shutdown VMEXIT. */
7540	HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY();
7541
7542	SVMEVENT Event;
7543	Event.u = 0;
7544	Event.n.u1Valid = 1;
7545	Event.n.u3Type = SVM_EVENT_EXCEPTION;
7546	Event.n.u8Vector = X86_XCPT_AC;
7547	Event.n.u1ErrorCodeValid = 1;
7548	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
7549	return VINF_SUCCESS;
7550	}
7551
7552
7553	/**
7554	* \#VMEXIT handler for breakpoint exceptions (SVM_EXIT_EXCEPTION_3).
7555	* Conditional \#VMEXIT.
7556	*/
7557	HMSVM_EXIT_DECL hmR0SvmExitXcptBP(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7558	{
7559	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7560
7561	HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY();
7562
7563	int rc = DBGFRZTrap03Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pCtx));
7564	if (rc == VINF_EM_RAW_GUEST_TRAP)
7565	{
7566	SVMEVENT Event;
7567	Event.u = 0;
7568	Event.n.u1Valid = 1;
7569	Event.n.u3Type = SVM_EVENT_EXCEPTION;
7570	Event.n.u8Vector = X86_XCPT_BP;
7571	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
7572	}
7573
7574	Assert(rc == VINF_SUCCESS \|\| rc == VINF_EM_RAW_GUEST_TRAP \|\| rc == VINF_EM_DBG_BREAKPOINT);
7575	return rc;
7576	}
7577
7578
7579	#if defined(HMSVM_ALWAYS_TRAP_ALL_XCPTS) \|\| defined(VBOX_WITH_NESTED_HWVIRT)
7580	/**
7581	* \#VMEXIT handler for generic exceptions. Conditional \#VMEXIT.
7582	*/
7583	HMSVM_EXIT_DECL hmR0SvmExitXcptGeneric(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7584	{
7585	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7586
7587	HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY();
7588
7589	PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
7590	uint8_t const uVector = pVmcb->ctrl.u64ExitCode - SVM_EXIT_EXCEPTION_0;
7591	uint32_t const uErrCode = pVmcb->ctrl.u64ExitInfo1;
7592	Assert(pSvmTransient->u64ExitCode == pVmcb->ctrl.u64ExitCode);
7593	Assert(uVector <= X86_XCPT_LAST);
7594	Log4(("hmR0SvmExitXcptGeneric: uVector=%#x uErrCode=%u\n", uVector, uErrCode));
7595
7596	SVMEVENT Event;
7597	Event.u = 0;
7598	Event.n.u1Valid = 1;
7599	Event.n.u3Type = SVM_EVENT_EXCEPTION;
7600	Event.n.u8Vector = uVector;
7601	switch (uVector)
7602	{
7603	/* Shouldn't be here for reflecting #PFs (among other things, the fault address isn't passed along). */
7604	case X86_XCPT_PF: AssertMsgFailed(("hmR0SvmExitXcptGeneric: Unexpected exception")); return VERR_SVM_IPE_5;
7605	case X86_XCPT_DF:
7606	case X86_XCPT_TS:
7607	case X86_XCPT_NP:
7608	case X86_XCPT_SS:
7609	case X86_XCPT_GP:
7610	case X86_XCPT_AC:
7611	{
7612	Event.n.u1ErrorCodeValid = 1;
7613	Event.n.u32ErrorCode = uErrCode;
7614	break;
7615	}
7616	}
7617
7618	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
7619	return VINF_SUCCESS;
7620	}
7621	#endif
7622
7623	#ifdef VBOX_WITH_NESTED_HWVIRT
7624	/**
7625	* \#VMEXIT handler for #PF occuring while in nested-guest execution
7626	* (SVM_EXIT_EXCEPTION_14). Conditional \#VMEXIT.
7627	*/
7628	HMSVM_EXIT_DECL hmR0SvmExitXcptPFNested(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7629	{
7630	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7631
7632	HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY();
7633
7634	/* See AMD spec. 15.12.15 "#PF (Page Fault)". */
7635	PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
7636	uint32_t u32ErrCode = pVmcb->ctrl.u64ExitInfo1;
7637	uint64_t const uFaultAddress = pVmcb->ctrl.u64ExitInfo2;
7638
7639	Log4(("#PFNested: uFaultAddress=%#RX64 CS:RIP=%#04x:%#RX64 u32ErrCode=%#RX32 CR3=%#RX64\n", uFaultAddress, pCtx->cs.Sel,
7640	pCtx->rip, u32ErrCode, pCtx->cr3));
7641
7642	/* If it's a vectoring #PF, emulate injecting the original event injection as PGMTrap0eHandler() is incapable
7643	of differentiating between instruction emulation and event injection that caused a #PF. See @bugref{6607}. */
7644	if (pSvmTransient->fVectoringPF)
7645	{
7646	Assert(pVCpu->hm.s.Event.fPending);
7647	return VINF_EM_RAW_INJECT_TRPM_EVENT;
7648	}
7649
7650	Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
7651
7652	TRPMAssertXcptPF(pVCpu, uFaultAddress, u32ErrCode);
7653	int rc = PGMTrap0eHandler(pVCpu, u32ErrCode, CPUMCTX2CORE(pCtx), (RTGCPTR)uFaultAddress);
7654
7655	Log4(("#PFNested: rc=%Rrc\n", rc));
7656
7657	if (rc == VINF_SUCCESS)
7658	{
7659	/* Successfully synced shadow pages tables or emulated an MMIO instruction. */
7660	TRPMResetTrap(pVCpu);
7661	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
7662	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
7663	return rc;
7664	}
7665
7666	if (rc == VINF_EM_RAW_GUEST_TRAP)
7667	{
7668	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
7669
7670	if (!pSvmTransient->fVectoringDoublePF)
7671	{
7672	/* It's a nested-guest page fault and needs to be reflected to the nested-guest. */
7673	u32ErrCode = TRPMGetErrorCode(pVCpu); /* The error code might have been changed. */
7674	TRPMResetTrap(pVCpu);
7675	hmR0SvmSetPendingXcptPF(pVCpu, pCtx, u32ErrCode, uFaultAddress);
7676	}
7677	else
7678	{
7679	/* A nested-guest page-fault occurred during delivery of a page-fault. Inject #DF. */
7680	TRPMResetTrap(pVCpu);
7681	hmR0SvmSetPendingXcptDF(pVCpu);
7682	Log4(("#PF: Pending #DF due to vectoring #PF\n"));
7683	}
7684
7685	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
7686	return VINF_SUCCESS;
7687	}
7688
7689	TRPMResetTrap(pVCpu);
7690	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
7691	return rc;
7692	}
7693
7694
7695	/**
7696	* \#VMEXIT handler for CLGI (SVM_EXIT_CLGI). Conditional \#VMEXIT.
7697	*/
7698	HMSVM_EXIT_DECL hmR0SvmExitClgi(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7699	{
7700	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7701
7702	#ifdef VBOX_STRICT
7703	PCSVMVMCB pVmcbTmp = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
7704	Assert(pVmcbTmp);
7705	Assert(!pVmcbTmp->ctrl.IntCtrl.n.u1VGifEnable);
7706	RT_NOREF(pVmcbTmp);
7707	#endif
7708
7709	/** @todo Stat. */
7710	/* STAM_COUNTER_INC(&pVCpu->hm.s.StatExitClgi); */
7711	uint8_t const cbInstr = hmR0SvmGetInstrLengthHwAssist(pVCpu, pCtx, 3);
7712	VBOXSTRICTRC rcStrict = IEMExecDecodedClgi(pVCpu, cbInstr);
7713	return VBOXSTRICTRC_VAL(rcStrict);
7714	}
7715
7716
7717	/**
7718	* \#VMEXIT handler for STGI (SVM_EXIT_STGI). Conditional \#VMEXIT.
7719	*/
7720	HMSVM_EXIT_DECL hmR0SvmExitStgi(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7721	{
7722	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7723
7724	#ifdef VBOX_STRICT
7725	PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
7726	Assert(pVmcb);
7727	Assert(!pVmcb->ctrl.IntCtrl.n.u1VGifEnable);
7728	RT_NOREF(pVmcb);
7729	#endif
7730
7731	/** @todo Stat. */
7732	/* STAM_COUNTER_INC(&pVCpu->hm.s.StatExitStgi); */
7733	uint8_t const cbInstr = hmR0SvmGetInstrLengthHwAssist(pVCpu, pCtx, 3);
7734	VBOXSTRICTRC rcStrict = IEMExecDecodedStgi(pVCpu, cbInstr);
7735	return VBOXSTRICTRC_VAL(rcStrict);
7736	}
7737
7738
7739	/**
7740	* \#VMEXIT handler for VMLOAD (SVM_EXIT_VMLOAD). Conditional \#VMEXIT.
7741	*/
7742	HMSVM_EXIT_DECL hmR0SvmExitVmload(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7743	{
7744	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7745
7746	#ifdef VBOX_STRICT
7747	PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
7748	Assert(pVmcb);
7749	Assert(!pVmcb->ctrl.LbrVirt.n.u1VirtVmsaveVmload);
7750	RT_NOREF(pVmcb);
7751	#endif
7752
7753	/** @todo Stat. */
7754	/* STAM_COUNTER_INC(&pVCpu->hm.s.StatExitVmload); */
7755	uint8_t const cbInstr = hmR0SvmGetInstrLengthHwAssist(pVCpu, pCtx, 3);
7756	VBOXSTRICTRC rcStrict = IEMExecDecodedVmload(pVCpu, cbInstr);
7757	if (rcStrict == VINF_SUCCESS)
7758	{
7759	/* We skip flagging changes made to LSTAR, STAR, SFMASK and other MSRs as they are always re-loaded. */
7760	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS
7761	\| HM_CHANGED_GUEST_TR
7762	\| HM_CHANGED_GUEST_LDTR);
7763	}
7764	return VBOXSTRICTRC_VAL(rcStrict);
7765	}
7766
7767
7768	/**
7769	* \#VMEXIT handler for VMSAVE (SVM_EXIT_VMSAVE). Conditional \#VMEXIT.
7770	*/
7771	HMSVM_EXIT_DECL hmR0SvmExitVmsave(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7772	{
7773	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7774
7775	#ifdef VBOX_STRICT
7776	PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu, pCtx);
7777	Assert(pVmcb);
7778	Assert(!pVmcb->ctrl.LbrVirt.n.u1VirtVmsaveVmload);
7779	RT_NOREF(pVmcb);
7780	#endif
7781
7782	/** @todo Stat. */
7783	/* STAM_COUNTER_INC(&pVCpu->hm.s.StatExitVmsave); */
7784	uint8_t const cbInstr = hmR0SvmGetInstrLengthHwAssist(pVCpu, pCtx, 3);
7785	VBOXSTRICTRC rcStrict = IEMExecDecodedVmsave(pVCpu, cbInstr);
7786	return VBOXSTRICTRC_VAL(rcStrict);
7787	}
7788
7789
7790	/**
7791	* \#VMEXIT handler for INVLPGA (SVM_EXIT_INVLPGA). Conditional \#VMEXIT.
7792	*/
7793	HMSVM_EXIT_DECL hmR0SvmExitInvlpga(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7794	{
7795	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7796	/** @todo Stat. */
7797	/* STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvlpga); */
7798	uint8_t const cbInstr = hmR0SvmGetInstrLengthHwAssist(pVCpu, pCtx, 3);
7799	VBOXSTRICTRC rcStrict = IEMExecDecodedInvlpga(pVCpu, cbInstr);
7800	return VBOXSTRICTRC_VAL(rcStrict);
7801	}
7802
7803
7804	/**
7805	* \#VMEXIT handler for STGI (SVM_EXIT_VMRUN). Conditional \#VMEXIT.
7806	*/
7807	HMSVM_EXIT_DECL hmR0SvmExitVmrun(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7808	{
7809	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7810	/** @todo Stat. */
7811	/* STAM_COUNTER_INC(&pVCpu->hm.s.StatExitVmrun); */
7812	#if 0
7813	VBOXSTRICTRC rcStrict;
7814	uint8_t const cbInstr = hmR0SvmGetInstrLengthHwAssist(pVCpu, pCtx, 3);
7815	rcStrict = IEMExecDecodedVmrun(pVCpu, cbInstr);
7816	Log4(("IEMExecDecodedVmrun: returned %d\n", VBOXSTRICTRC_VAL(rcStrict)));
7817	if (rcStrict == VINF_SUCCESS)
7818	{
7819	rcStrict = VINF_SVM_VMRUN;
7820	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
7821	}
7822	return VBOXSTRICTRC_VAL(rcStrict);
7823	#endif
7824	return VERR_EM_INTERPRETER;
7825	}
7826
7827
7828	/**
7829	* Nested-guest \#VMEXIT handler for debug exceptions (SVM_EXIT_EXCEPTION_1).
7830	* Unconditional \#VMEXIT.
7831	*/
7832	HMSVM_EXIT_DECL hmR0SvmNestedExitXcptDB(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7833	{
7834	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7835
7836	/* If this #DB is the result of delivering an event, go back to the interpreter. */
7837	/** @todo if triple-fault is returned in nested-guest scenario convert to a
7838	* shutdown VMEXIT. */
7839	HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY();
7840	if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
7841	{
7842	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingInterpret);
7843	return VINF_EM_RAW_INJECT_TRPM_EVENT;
7844	}
7845
7846	hmR0SvmSetPendingXcptDB(pVCpu);
7847	return VINF_SUCCESS;
7848	}
7849
7850
7851	/**
7852	* Nested-guest \#VMEXIT handler for breakpoint exceptions (SVM_EXIT_EXCEPTION_3).
7853	* Conditional \#VMEXIT.
7854	*/
7855	HMSVM_EXIT_DECL hmR0SvmNestedExitXcptBP(PVMCPU pVCpu, PCPUMCTX pCtx, PSVMTRANSIENT pSvmTransient)
7856	{
7857	HMSVM_VALIDATE_EXIT_HANDLER_PARAMS();
7858
7859	/** @todo if triple-fault is returned in nested-guest scenario convert to a
7860	* shutdown VMEXIT. */
7861	HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY();
7862
7863	SVMEVENT Event;
7864	Event.u = 0;
7865	Event.n.u1Valid = 1;
7866	Event.n.u3Type = SVM_EVENT_EXCEPTION;
7867	Event.n.u8Vector = X86_XCPT_BP;
7868	hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
7869	return VINF_SUCCESS;
7870	}
7871
7872	#endif /* VBOX_WITH_NESTED_HWVIRT */
7873
7874
7875	/** @} */
7876

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

source: vbox/trunk/src/VBox/VMM/VMMR0/HMSVMR0.cpp@ 70721

Download in other formats: