PGMAllBth.h@ 107044

Last change on this file since 107044 was 106061, checked in by vboxsync, 2 months ago
Copyright year updates by scm.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 236.3 KB

Line
1	/* $Id: PGMAllBth.h 106061 2024-09-16 14:03:52Z vboxsync $ */
2	/** @file
3	* VBox - Page Manager, Shadow+Guest Paging Template - All context code.
4	*
5	* @remarks Extended page tables (intel) are built with PGM_GST_TYPE set to
6	* PGM_TYPE_PROT (and PGM_SHW_TYPE set to PGM_TYPE_EPT).
7	* bird: WTF does this mean these days? Looking at PGMAll.cpp it's
8	*
9	* @remarks This file is one big \#ifdef-orgy!
10	*
11	*/
12
13	/*
14	* Copyright (C) 2006-2024 Oracle and/or its affiliates.
15	*
16	* This file is part of VirtualBox base platform packages, as
17	* available from https://www.virtualbox.org.
18	*
19	* This program is free software; you can redistribute it and/or
20	* modify it under the terms of the GNU General Public License
21	* as published by the Free Software Foundation, in version 3 of the
22	* License.
23	*
24	* This program is distributed in the hope that it will be useful, but
25	* WITHOUT ANY WARRANTY; without even the implied warranty of
26	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
27	* General Public License for more details.
28	*
29	* You should have received a copy of the GNU General Public License
30	* along with this program; if not, see <https://www.gnu.org/licenses>.
31	*
32	* SPDX-License-Identifier: GPL-3.0-only
33	*/
34
35	#ifdef _MSC_VER
36	/** @todo we're generating unnecessary code in nested/ept shadow mode and for
37	* real/prot-guest+RC mode. */
38	# pragma warning(disable: 4505)
39	#endif
40
41
42	/*********************************************************************************************************************************
43	* Internal Functions *
44	*********************************************************************************************************************************/
45	RT_C_DECLS_BEGIN
46	PGM_BTH_DECL(int, Enter)(PVMCPUCC pVCpu, RTGCPHYS GCPhysCR3);
47	#ifndef IN_RING3
48	PGM_BTH_DECL(int, Trap0eHandler)(PVMCPUCC pVCpu, RTGCUINT uErr, PCPUMCTX pCtx, RTGCPTR pvFault, bool *pfLockTaken);
49	PGM_BTH_DECL(int, NestedTrap0eHandler)(PVMCPUCC pVCpu, RTGCUINT uErr, PCPUMCTX pCtx, RTGCPHYS GCPhysNestedFault,
50	bool fIsLinearAddrValid, RTGCPTR GCPtrNestedFault, PPGMPTWALK pWalk, bool *pfLockTaken);
51	# if defined(VBOX_WITH_NESTED_HWVIRT_VMX_EPT) && PGM_SHW_TYPE == PGM_TYPE_EPT
52	static void PGM_BTH_NAME(NestedSyncPageWorker)(PVMCPUCC pVCpu, PSHWPTE pPte, RTGCPHYS GCPhysPage, PPGMPOOLPAGE pShwPage,
53	unsigned iPte, SLATPTE GstSlatPte);
54	static int PGM_BTH_NAME(NestedSyncPage)(PVMCPUCC pVCpu, RTGCPHYS GCPhysNestedPage, RTGCPHYS GCPhysPage, unsigned cPages,
55	uint32_t uErr, PPGMPTWALKGST pGstWalkAll);
56	static int PGM_BTH_NAME(NestedSyncPT)(PVMCPUCC pVCpu, RTGCPHYS GCPhysNestedPage, RTGCPHYS GCPhysPage, PPGMPTWALKGST pGstWalkAll);
57	# endif /* VBOX_WITH_NESTED_HWVIRT_VMX_EPT */
58	#endif
59	PGM_BTH_DECL(int, InvalidatePage)(PVMCPUCC pVCpu, RTGCPTR GCPtrPage);
60	static int PGM_BTH_NAME(SyncPage)(PVMCPUCC pVCpu, GSTPDE PdeSrc, RTGCPTR GCPtrPage, unsigned cPages, unsigned uErr);
61	static int PGM_BTH_NAME(CheckDirtyPageFault)(PVMCPUCC pVCpu, uint32_t uErr, PSHWPDE pPdeDst, GSTPDE const *pPdeSrc, RTGCPTR GCPtrPage);
62	static int PGM_BTH_NAME(SyncPT)(PVMCPUCC pVCpu, unsigned iPD, PGSTPD pPDSrc, RTGCPTR GCPtrPage);
63	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
64	static void PGM_BTH_NAME(SyncPageWorker)(PVMCPUCC pVCpu, PSHWPTE pPteDst, GSTPDE PdeSrc, GSTPTE PteSrc, PPGMPOOLPAGE pShwPage, unsigned iPTDst);
65	#else
66	static void PGM_BTH_NAME(SyncPageWorker)(PVMCPUCC pVCpu, PSHWPTE pPteDst, RTGCPHYS GCPhysPage, PPGMPOOLPAGE pShwPage, unsigned iPTDst);
67	#endif
68	PGM_BTH_DECL(int, VerifyAccessSyncPage)(PVMCPUCC pVCpu, RTGCPTR Addr, unsigned fPage, unsigned uErr);
69	PGM_BTH_DECL(int, PrefetchPage)(PVMCPUCC pVCpu, RTGCPTR GCPtrPage);
70	PGM_BTH_DECL(int, SyncCR3)(PVMCPUCC pVCpu, uint64_t cr0, uint64_t cr3, uint64_t cr4, bool fGlobal);
71	#ifdef VBOX_STRICT
72	PGM_BTH_DECL(unsigned, AssertCR3)(PVMCPUCC pVCpu, uint64_t cr3, uint64_t cr4, RTGCPTR GCPtr = 0, RTGCPTR cb = ~(RTGCPTR)0);
73	#endif
74	PGM_BTH_DECL(int, MapCR3)(PVMCPUCC pVCpu, RTGCPHYS GCPhysCR3);
75	PGM_BTH_DECL(int, UnmapCR3)(PVMCPUCC pVCpu);
76
77	#ifdef IN_RING3
78	PGM_BTH_DECL(int, Relocate)(PVMCPUCC pVCpu, RTGCPTR offDelta);
79	#endif
80	RT_C_DECLS_END
81
82
83
84
85	/*
86	* Filter out some illegal combinations of guest and shadow paging, so we can
87	* remove redundant checks inside functions.
88	*/
89	#if PGM_GST_TYPE == PGM_TYPE_PAE && PGM_SHW_TYPE != PGM_TYPE_PAE \
90	&& !PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) && PGM_SHW_TYPE != PGM_TYPE_NONE
91	# error "Invalid combination; PAE guest implies PAE shadow"
92	#endif
93
94	#if (PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
95	&& !( PGM_SHW_TYPE == PGM_TYPE_32BIT \|\| PGM_SHW_TYPE == PGM_TYPE_PAE \|\| PGM_SHW_TYPE == PGM_TYPE_AMD64 \
96	\|\| PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) \|\| PGM_SHW_TYPE == PGM_TYPE_NONE)
97	# error "Invalid combination; real or protected mode without paging implies 32 bits or PAE shadow paging."
98	#endif
99
100	#if (PGM_GST_TYPE == PGM_TYPE_32BIT \|\| PGM_GST_TYPE == PGM_TYPE_PAE) \
101	&& !( PGM_SHW_TYPE == PGM_TYPE_32BIT \|\| PGM_SHW_TYPE == PGM_TYPE_PAE \
102	\|\| PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) \|\| PGM_SHW_TYPE == PGM_TYPE_NONE)
103	# error "Invalid combination; 32 bits guest paging or PAE implies 32 bits or PAE shadow paging."
104	#endif
105
106	#if (PGM_GST_TYPE == PGM_TYPE_AMD64 && PGM_SHW_TYPE != PGM_TYPE_AMD64 && !PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) && PGM_SHW_TYPE != PGM_TYPE_NONE) \
107	\|\| (PGM_SHW_TYPE == PGM_TYPE_AMD64 && PGM_GST_TYPE != PGM_TYPE_AMD64 && PGM_GST_TYPE != PGM_TYPE_PROT)
108	# error "Invalid combination; AMD64 guest implies AMD64 shadow and vice versa"
109	#endif
110
111
112	/**
113	* Enters the shadow+guest mode.
114	*
115	* @returns VBox status code.
116	* @param pVCpu The cross context virtual CPU structure.
117	* @param GCPhysCR3 The physical address from the CR3 register.
118	*/
119	PGM_BTH_DECL(int, Enter)(PVMCPUCC pVCpu, RTGCPHYS GCPhysCR3)
120	{
121	/* Here we deal with allocation of the root shadow page table for real and protected mode during mode switches;
122	* Other modes rely on MapCR3/UnmapCR3 to setup the shadow root page tables.
123	*/
124	#if ( ( PGM_SHW_TYPE == PGM_TYPE_32BIT \
125	\|\| PGM_SHW_TYPE == PGM_TYPE_PAE \
126	\|\| PGM_SHW_TYPE == PGM_TYPE_AMD64) \
127	&& ( PGM_GST_TYPE == PGM_TYPE_REAL \
128	\|\| PGM_GST_TYPE == PGM_TYPE_PROT))
129
130	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
131
132	Assert(!pVM->pgm.s.fNestedPaging);
133
134	PGM_LOCK_VOID(pVM);
135	/* Note: we only really need shadow paging in real and protected mode for VT-x and AMD-V (excluding nested paging/EPT modes),
136	* but any calls to GC need a proper shadow page setup as well.
137	*/
138	/* Free the previous root mapping if still active. */
139	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
140	PPGMPOOLPAGE pOldShwPageCR3 = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
141	if (pOldShwPageCR3)
142	{
143	Assert(pOldShwPageCR3->enmKind != PGMPOOLKIND_FREE);
144
145	/* Mark the page as unlocked; allow flushing again. */
146	pgmPoolUnlockPage(pPool, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
147
148	pgmPoolFreeByPage(pPool, pOldShwPageCR3, NIL_PGMPOOL_IDX, UINT32_MAX);
149	pVCpu->pgm.s.pShwPageCR3R3 = NIL_RTR3PTR;
150	pVCpu->pgm.s.pShwPageCR3R0 = NIL_RTR0PTR;
151	}
152
153	/* construct a fake address. */
154	GCPhysCR3 = RT_BIT_64(63);
155	PPGMPOOLPAGE pNewShwPageCR3;
156	int rc = pgmPoolAlloc(pVM, GCPhysCR3, BTH_PGMPOOLKIND_ROOT, PGMPOOLACCESS_DONTCARE, PGM_A20_IS_ENABLED(pVCpu),
157	NIL_PGMPOOL_IDX, UINT32_MAX, false /fLockPage/,
158	&pNewShwPageCR3);
159	AssertRCReturn(rc, rc);
160
161	pVCpu->pgm.s.pShwPageCR3R3 = pgmPoolConvertPageToR3(pPool, pNewShwPageCR3);
162	pVCpu->pgm.s.pShwPageCR3R0 = pgmPoolConvertPageToR0(pPool, pNewShwPageCR3);
163
164	/* Mark the page as locked; disallow flushing. */
165	pgmPoolLockPage(pPool, pNewShwPageCR3);
166
167	/* Set the current hypervisor CR3. */
168	CPUMSetHyperCR3(pVCpu, PGMGetHyperCR3(pVCpu));
169
170	PGM_UNLOCK(pVM);
171	return rc;
172	#else
173	NOREF(pVCpu); NOREF(GCPhysCR3);
174	return VINF_SUCCESS;
175	#endif
176	}
177
178
179	#ifndef IN_RING3
180
181	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
182	/**
183	* Deal with a guest page fault.
184	*
185	* @returns Strict VBox status code.
186	* @retval VINF_EM_RAW_GUEST_TRAP
187	* @retval VINF_EM_RAW_EMULATE_INSTR
188	*
189	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
190	* @param pWalk The guest page table walk result.
191	* @param uErr The error code.
192	*/
193	PGM_BTH_DECL(VBOXSTRICTRC, Trap0eHandlerGuestFault)(PVMCPUCC pVCpu, PPGMPTWALK pWalk, RTGCUINT uErr)
194	{
195	/*
196	* Calc the error code for the guest trap.
197	*/
198	uint32_t uNewErr = GST_IS_NX_ACTIVE(pVCpu)
199	? uErr & (X86_TRAP_PF_RW \| X86_TRAP_PF_US \| X86_TRAP_PF_ID)
200	: uErr & (X86_TRAP_PF_RW \| X86_TRAP_PF_US);
201	if ( pWalk->fRsvdError
202	\|\| pWalk->fBadPhysAddr)
203	{
204	uNewErr \|= X86_TRAP_PF_RSVD \| X86_TRAP_PF_P;
205	Assert(!pWalk->fNotPresent);
206	}
207	else if (!pWalk->fNotPresent)
208	uNewErr \|= X86_TRAP_PF_P;
209	TRPMSetErrorCode(pVCpu, uNewErr);
210
211	LogFlow(("Guest trap; cr2=%RGv uErr=%RGv lvl=%d\n", pWalk->GCPtr, uErr, pWalk->uLevel));
212	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2GuestTrap; });
213	return VINF_EM_RAW_GUEST_TRAP;
214	}
215	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
216
217
218	#if !PGM_TYPE_IS_NESTED(PGM_SHW_TYPE) && PGM_SHW_TYPE != PGM_TYPE_NONE
219	/**
220	* Deal with a guest page fault.
221	*
222	* The caller has taken the PGM lock.
223	*
224	* @returns Strict VBox status code.
225	*
226	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
227	* @param uErr The error code.
228	* @param pCtx Pointer to the register context for the CPU.
229	* @param pvFault The fault address.
230	* @param pPage The guest page at @a pvFault.
231	* @param pWalk The guest page table walk result.
232	* @param pGstWalk The guest paging-mode specific walk information.
233	* @param pfLockTaken PGM lock taken here or not (out). This is true
234	* when we're called.
235	*/
236	static VBOXSTRICTRC PGM_BTH_NAME(Trap0eHandlerDoAccessHandlers)(PVMCPUCC pVCpu, RTGCUINT uErr, PCPUMCTX pCtx,
237	RTGCPTR pvFault, PPGMPAGE pPage, bool *pfLockTaken
238	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) \|\| defined(DOXYGEN_RUNNING)
239	, PPGMPTWALK pWalk
240	, PGSTPTWALK pGstWalk
241	# endif
242	)
243	{
244	# if !PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
245	GSTPDE const PdeSrcDummy = { X86_PDE_P \| X86_PDE_US \| X86_PDE_RW \| X86_PDE_A };
246	# endif
247	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
248	VBOXSTRICTRC rcStrict;
249
250	if (PGM_PAGE_HAS_ANY_PHYSICAL_HANDLERS(pPage))
251	{
252	/*
253	* Physical page access handler.
254	*/
255	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
256	const RTGCPHYS GCPhysFault = pWalk->GCPhys;
257	# else
258	const RTGCPHYS GCPhysFault = PGM_A20_APPLY(pVCpu, (RTGCPHYS)pvFault);
259	# endif
260	PPGMPHYSHANDLER pCur;
261	rcStrict = pgmHandlerPhysicalLookup(pVM, GCPhysFault, &pCur);
262	if (RT_SUCCESS(rcStrict))
263	{
264	PCPGMPHYSHANDLERTYPEINT const pCurType = PGMPHYSHANDLER_GET_TYPE(pVM, pCur);
265
266	# ifdef PGM_SYNC_N_PAGES
267	/*
268	* If the region is write protected and we got a page not present fault, then sync
269	* the pages. If the fault was caused by a read, then restart the instruction.
270	* In case of write access continue to the GC write handler.
271	*
272	* ASSUMES that there is only one handler per page or that they have similar write properties.
273	*/
274	if ( !(uErr & X86_TRAP_PF_P)
275	&& pCurType->enmKind == PGMPHYSHANDLERKIND_WRITE)
276	{
277	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
278	rcStrict = PGM_BTH_NAME(SyncPage)(pVCpu, pGstWalk->Pde, pvFault, PGM_SYNC_NR_PAGES, uErr);
279	# else
280	rcStrict = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrcDummy, pvFault, PGM_SYNC_NR_PAGES, uErr);
281	# endif
282	if ( RT_FAILURE(rcStrict)
283	\|\| !(uErr & X86_TRAP_PF_RW)
284	\|\| rcStrict == VINF_PGM_SYNCPAGE_MODIFIED_PDE)
285	{
286	AssertMsgRC(rcStrict, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
287	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersOutOfSync);
288	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2OutOfSyncHndPhys; });
289	return rcStrict;
290	}
291	}
292	# endif
293	# ifdef PGM_WITH_MMIO_OPTIMIZATIONS
294	/*
295	* If the access was not thru a #PF(RSVD\|...) resync the page.
296	*/
297	if ( !(uErr & X86_TRAP_PF_RSVD)
298	&& pCurType->enmKind != PGMPHYSHANDLERKIND_WRITE
299	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
300	&& (pWalk->fEffective & (PGM_PTATTRS_W_MASK \| PGM_PTATTRS_US_MASK))
301	== PGM_PTATTRS_W_MASK /** @todo Remove pGstWalk->Core.fEffectiveUS and X86_PTE_US further down in the sync code. */
302	# endif
303	)
304	{
305	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
306	rcStrict = PGM_BTH_NAME(SyncPage)(pVCpu, pGstWalk->Pde, pvFault, PGM_SYNC_NR_PAGES, uErr);
307	# else
308	rcStrict = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrcDummy, pvFault, PGM_SYNC_NR_PAGES, uErr);
309	# endif
310	if ( RT_FAILURE(rcStrict)
311	\|\| rcStrict == VINF_PGM_SYNCPAGE_MODIFIED_PDE)
312	{
313	AssertMsgRC(rcStrict, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
314	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersOutOfSync);
315	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2OutOfSyncHndPhys; });
316	return rcStrict;
317	}
318	}
319	# endif
320
321	AssertMsg( pCurType->enmKind != PGMPHYSHANDLERKIND_WRITE
322	\|\| (pCurType->enmKind == PGMPHYSHANDLERKIND_WRITE && (uErr & X86_TRAP_PF_RW)),
323	("Unexpected trap for physical handler: %08X (phys=%08x) pPage=%R[pgmpage] uErr=%X, enmKind=%d\n",
324	pvFault, GCPhysFault, pPage, uErr, pCurType->enmKind));
325	if (pCurType->enmKind == PGMPHYSHANDLERKIND_WRITE)
326	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersPhysWrite);
327	else
328	{
329	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersPhysAll);
330	if (uErr & X86_TRAP_PF_RSVD) STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersPhysAllOpt);
331	}
332
333	if (pCurType->pfnPfHandler)
334	{
335	STAM_PROFILE_START(&pCur->Stat, h);
336
337	if (pCurType->fKeepPgmLock)
338	{
339	rcStrict = pCurType->pfnPfHandler(pVM, pVCpu, uErr, pCtx, pvFault, GCPhysFault,
340	!pCurType->fRing0DevInsIdx ? pCur->uUser
341	: (uintptr_t)PDMDeviceRing0IdxToInstance(pVM, pCur->uUser));
342
343	STAM_PROFILE_STOP(&pCur->Stat, h); /* no locking needed, entry is unlikely reused before we get here. */
344	}
345	else
346	{
347	uint64_t const uUser = !pCurType->fRing0DevInsIdx ? pCur->uUser
348	: (uintptr_t)PDMDeviceRing0IdxToInstance(pVM, pCur->uUser);
349	PGM_UNLOCK(pVM);
350	*pfLockTaken = false;
351
352	rcStrict = pCurType->pfnPfHandler(pVM, pVCpu, uErr, pCtx, pvFault, GCPhysFault, uUser);
353
354	STAM_PROFILE_STOP(&pCur->Stat, h); /* no locking needed, entry is unlikely reused before we get here. */
355	}
356	}
357	else
358	rcStrict = VINF_EM_RAW_EMULATE_INSTR;
359
360	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2HndPhys; });
361	return rcStrict;
362	}
363	AssertMsgReturn(rcStrict == VERR_NOT_FOUND, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)), rcStrict);
364	}
365
366	/*
367	* There is a handled area of the page, but this fault doesn't belong to it.
368	* We must emulate the instruction.
369	*
370	* To avoid crashing (non-fatal) in the interpreter and go back to the recompiler
371	* we first check if this was a page-not-present fault for a page with only
372	* write access handlers. Restart the instruction if it wasn't a write access.
373	*/
374	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersUnhandled);
375
376	if ( !PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage)
377	&& !(uErr & X86_TRAP_PF_P))
378	{
379	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
380	rcStrict = PGM_BTH_NAME(SyncPage)(pVCpu, pGstWalk->Pde, pvFault, PGM_SYNC_NR_PAGES, uErr);
381	# else
382	rcStrict = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrcDummy, pvFault, PGM_SYNC_NR_PAGES, uErr);
383	# endif
384	if ( RT_FAILURE(rcStrict)
385	\|\| rcStrict == VINF_PGM_SYNCPAGE_MODIFIED_PDE
386	\|\| !(uErr & X86_TRAP_PF_RW))
387	{
388	AssertMsgRC(rcStrict, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
389	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersOutOfSync);
390	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2OutOfSyncHndPhys; });
391	return rcStrict;
392	}
393	}
394
395	/** @todo This particular case can cause quite a lot of overhead. E.g. early stage of kernel booting in Ubuntu 6.06
396	* It's writing to an unhandled part of the LDT page several million times.
397	*/
398	rcStrict = PGMInterpretInstruction(pVCpu, pvFault);
399	LogFlow(("PGM: PGMInterpretInstruction -> rcStrict=%d pPage=%R[pgmpage]\n", VBOXSTRICTRC_VAL(rcStrict), pPage));
400	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2HndUnhandled; });
401	return rcStrict;
402	} /* if any kind of handler */
403	# endif /* !PGM_TYPE_IS_NESTED(PGM_SHW_TYPE) && PGM_SHW_TYPE != PGM_TYPE_NONE*/
404
405
406	/**
407	* \#PF Handler for raw-mode guest execution.
408	*
409	* @returns VBox status code (appropriate for trap handling and GC return).
410	*
411	* @param pVCpu The cross context virtual CPU structure.
412	* @param uErr The trap error code.
413	* @param pCtx Pointer to the register context for the CPU.
414	* @param pvFault The fault address.
415	* @param pfLockTaken PGM lock taken here or not (out)
416	*/
417	PGM_BTH_DECL(int, Trap0eHandler)(PVMCPUCC pVCpu, RTGCUINT uErr, PCPUMCTX pCtx, RTGCPTR pvFault, bool *pfLockTaken)
418	{
419	PVMCC pVM = pVCpu->CTX_SUFF(pVM); NOREF(pVM);
420
421	*pfLockTaken = false;
422
423	# if ( PGM_GST_TYPE == PGM_TYPE_32BIT \|\| PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT \
424	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \|\| PGM_GST_TYPE == PGM_TYPE_AMD64) \
425	&& !PGM_TYPE_IS_NESTED(PGM_SHW_TYPE) \
426	&& (PGM_SHW_TYPE != PGM_TYPE_EPT \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
427	&& PGM_SHW_TYPE != PGM_TYPE_NONE
428	int rc;
429
430	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
431	/*
432	* Walk the guest page translation tables and check if it's a guest fault.
433	*/
434	PGMPTWALK Walk;
435	GSTPTWALK GstWalk;
436	rc = PGM_GST_NAME(Walk)(pVCpu, pvFault, &Walk, &GstWalk);
437	if (RT_FAILURE_NP(rc))
438	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(Trap0eHandlerGuestFault)(pVCpu, &Walk, uErr));
439
440	/* assert some GstWalk sanity. */
441	# if PGM_GST_TYPE == PGM_TYPE_AMD64
442	/AssertMsg(GstWalk.Pml4e.u == GstWalk.pPml4e->u, ("%RX64 %RX64\n", (uint64_t)GstWalk.Pml4e.u, (uint64_t)GstWalk.pPml4e->u)); - not always true with SMP guests. /
443	# endif
444	# if PGM_GST_TYPE == PGM_TYPE_AMD64 \|\| PGM_GST_TYPE == PGM_TYPE_PAE
445	/AssertMsg(GstWalk.Pdpe.u == GstWalk.pPdpe->u, ("%RX64 %RX64\n", (uint64_t)GstWalk.Pdpe.u, (uint64_t)GstWalk.pPdpe->u)); - ditto /
446	# endif
447	/AssertMsg(GstWalk.Pde.u == GstWalk.pPde->u, ("%RX64 %RX64\n", (uint64_t)GstWalk.Pde.u, (uint64_t)GstWalk.pPde->u)); - ditto /
448	/AssertMsg(GstWalk.Core.fBigPage \|\| GstWalk.Pte.u == GstWalk.pPte->u, ("%RX64 %RX64\n", (uint64_t)GstWalk.Pte.u, (uint64_t)GstWalk.pPte->u)); - ditto /
449	Assert(Walk.fSucceeded);
450	Assert(Walk.fEffective & PGM_PTATTRS_R_MASK);
451
452	if (uErr & (X86_TRAP_PF_RW \| X86_TRAP_PF_US \| X86_TRAP_PF_ID))
453	{
454	if ( ( (uErr & X86_TRAP_PF_RW)
455	&& !(Walk.fEffective & PGM_PTATTRS_W_MASK)
456	&& ( (uErr & X86_TRAP_PF_US)
457	\|\| CPUMIsGuestR0WriteProtEnabled(pVCpu)) )
458	\|\| ((uErr & X86_TRAP_PF_US) && !(Walk.fEffective & PGM_PTATTRS_US_MASK))
459	\|\| ((uErr & X86_TRAP_PF_ID) && (Walk.fEffective & PGM_PTATTRS_NX_MASK))
460	)
461	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(Trap0eHandlerGuestFault)(pVCpu, &Walk, uErr));
462	}
463
464	/* Take the big lock now before we update flags. */
465	*pfLockTaken = true;
466	PGM_LOCK_VOID(pVM);
467
468	/*
469	* Set the accessed and dirty flags.
470	*/
471	/** @todo Should probably use cmpxchg logic here as we're potentially racing
472	* other CPUs in SMP configs. (the lock isn't enough, since we take it
473	* after walking and the page tables could be stale already) */
474	# if PGM_GST_TYPE == PGM_TYPE_AMD64
475	if (!(GstWalk.Pml4e.u & X86_PML4E_A))
476	{
477	GstWalk.Pml4e.u \|= X86_PML4E_A;
478	GST_ATOMIC_OR(&GstWalk.pPml4e->u, X86_PML4E_A);
479	}
480	if (!(GstWalk.Pdpe.u & X86_PDPE_A))
481	{
482	GstWalk.Pdpe.u \|= X86_PDPE_A;
483	GST_ATOMIC_OR(&GstWalk.pPdpe->u, X86_PDPE_A);
484	}
485	# endif
486	if (Walk.fBigPage)
487	{
488	Assert(GstWalk.Pde.u & X86_PDE_PS);
489	if (uErr & X86_TRAP_PF_RW)
490	{
491	if ((GstWalk.Pde.u & (X86_PDE4M_A \| X86_PDE4M_D)) != (X86_PDE4M_A \| X86_PDE4M_D))
492	{
493	GstWalk.Pde.u \|= X86_PDE4M_A \| X86_PDE4M_D;
494	GST_ATOMIC_OR(&GstWalk.pPde->u, X86_PDE4M_A \| X86_PDE4M_D);
495	}
496	}
497	else
498	{
499	if (!(GstWalk.Pde.u & X86_PDE4M_A))
500	{
501	GstWalk.Pde.u \|= X86_PDE4M_A;
502	GST_ATOMIC_OR(&GstWalk.pPde->u, X86_PDE4M_A);
503	}
504	}
505	}
506	else
507	{
508	Assert(!(GstWalk.Pde.u & X86_PDE_PS));
509	if (!(GstWalk.Pde.u & X86_PDE_A))
510	{
511	GstWalk.Pde.u \|= X86_PDE_A;
512	GST_ATOMIC_OR(&GstWalk.pPde->u, X86_PDE_A);
513	}
514
515	if (uErr & X86_TRAP_PF_RW)
516	{
517	# ifdef VBOX_WITH_STATISTICS
518	if (GstWalk.Pte.u & X86_PTE_D)
519	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,PageAlreadyDirty));
520	else
521	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,DirtiedPage));
522	# endif
523	if ((GstWalk.Pte.u & (X86_PTE_A \| X86_PTE_D)) != (X86_PTE_A \| X86_PTE_D))
524	{
525	GstWalk.Pte.u \|= X86_PTE_A \| X86_PTE_D;
526	GST_ATOMIC_OR(&GstWalk.pPte->u, X86_PTE_A \| X86_PTE_D);
527	}
528	}
529	else
530	{
531	if (!(GstWalk.Pte.u & X86_PTE_A))
532	{
533	GstWalk.Pte.u \|= X86_PTE_A;
534	GST_ATOMIC_OR(&GstWalk.pPte->u, X86_PTE_A);
535	}
536	}
537	Assert(GstWalk.Pte.u == GstWalk.pPte->u);
538	}
539	#if 0
540	/* Disabling this since it's not reliable for SMP, see @bugref{10092#c22}. */
541	AssertMsg(GstWalk.Pde.u == GstWalk.pPde->u \|\| GstWalk.pPte->u == GstWalk.pPde->u,
542	("%RX64 %RX64 pPte=%p pPde=%p Pte=%RX64\n", (uint64_t)GstWalk.Pde.u, (uint64_t)GstWalk.pPde->u, GstWalk.pPte, GstWalk.pPde, (uint64_t)GstWalk.pPte->u));
543	#endif
544
545	# else /* !PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
546	GSTPDE const PdeSrcDummy = { X86_PDE_P \| X86_PDE_US \| X86_PDE_RW \| X86_PDE_A}; /** @todo eliminate this */
547
548	/* Take the big lock now. */
549	*pfLockTaken = true;
550	PGM_LOCK_VOID(pVM);
551	# endif /* !PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
552
553	# ifdef PGM_WITH_MMIO_OPTIMIZATIONS
554	/*
555	* If it is a reserved bit fault we know that it is an MMIO (access
556	* handler) related fault and can skip some 200 lines of code.
557	*/
558	if (uErr & X86_TRAP_PF_RSVD)
559	{
560	Assert(uErr & X86_TRAP_PF_P);
561	PPGMPAGE pPage;
562	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
563	rc = pgmPhysGetPageEx(pVM, Walk.GCPhys, &pPage);
564	if (RT_SUCCESS(rc) && PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage))
565	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(Trap0eHandlerDoAccessHandlers)(pVCpu, uErr, pCtx, pvFault, pPage,
566	pfLockTaken, &Walk, &GstWalk));
567	rc = PGM_BTH_NAME(SyncPage)(pVCpu, GstWalk.Pde, pvFault, 1, uErr);
568	# else
569	rc = pgmPhysGetPageEx(pVM, PGM_A20_APPLY(pVCpu, (RTGCPHYS)pvFault), &pPage);
570	if (RT_SUCCESS(rc) && PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage))
571	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(Trap0eHandlerDoAccessHandlers)(pVCpu, uErr, pCtx, pvFault, pPage, pfLockTaken));
572	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrcDummy, pvFault, 1, uErr);
573	# endif
574	AssertRC(rc);
575	PGM_INVL_PG(pVCpu, pvFault);
576	return rc; /* Restart with the corrected entry. */
577	}
578	# endif /* PGM_WITH_MMIO_OPTIMIZATIONS */
579
580	/*
581	* Fetch the guest PDE, PDPE and PML4E.
582	*/
583	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
584	const unsigned iPDDst = pvFault >> SHW_PD_SHIFT;
585	PX86PD pPDDst = pgmShwGet32BitPDPtr(pVCpu);
586
587	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
588	const unsigned iPDDst = (pvFault >> SHW_PD_SHIFT) & SHW_PD_MASK; /* pPDDst index, not used with the pool. */
589	PX86PDPAE pPDDst;
590	# if PGM_GST_TYPE == PGM_TYPE_PAE
591	rc = pgmShwSyncPaePDPtr(pVCpu, pvFault, GstWalk.Pdpe.u, &pPDDst);
592	# else
593	rc = pgmShwSyncPaePDPtr(pVCpu, pvFault, X86_PDPE_P, &pPDDst); /* RW, US and A are reserved in PAE mode. */
594	# endif
595	AssertMsgReturn(rc == VINF_SUCCESS, ("rc=%Rrc\n", rc), RT_FAILURE_NP(rc) ? rc : VERR_IPE_UNEXPECTED_INFO_STATUS);
596
597	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
598	const unsigned iPDDst = ((pvFault >> SHW_PD_SHIFT) & SHW_PD_MASK);
599	PX86PDPAE pPDDst;
600	# if PGM_GST_TYPE == PGM_TYPE_PROT /* (AMD-V nested paging) */
601	rc = pgmShwSyncLongModePDPtr(pVCpu, pvFault, X86_PML4E_P \| X86_PML4E_RW \| X86_PML4E_US \| X86_PML4E_A,
602	X86_PDPE_P \| X86_PDPE_RW \| X86_PDPE_US \| X86_PDPE_A, &pPDDst);
603	# else
604	rc = pgmShwSyncLongModePDPtr(pVCpu, pvFault, GstWalk.Pml4e.u, GstWalk.Pdpe.u, &pPDDst);
605	# endif
606	AssertMsgReturn(rc == VINF_SUCCESS, ("rc=%Rrc\n", rc), RT_FAILURE_NP(rc) ? rc : VERR_IPE_UNEXPECTED_INFO_STATUS);
607
608	# elif PGM_SHW_TYPE == PGM_TYPE_EPT
609	const unsigned iPDDst = ((pvFault >> SHW_PD_SHIFT) & SHW_PD_MASK);
610	PEPTPD pPDDst;
611	rc = pgmShwGetEPTPDPtr(pVCpu, pvFault, NULL, &pPDDst);
612	AssertMsgReturn(rc == VINF_SUCCESS, ("rc=%Rrc\n", rc), RT_FAILURE_NP(rc) ? rc : VERR_IPE_UNEXPECTED_INFO_STATUS);
613	# endif
614	Assert(pPDDst);
615
616	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
617	/*
618	* Dirty page handling.
619	*
620	* If we successfully correct the write protection fault due to dirty bit
621	* tracking, then return immediately.
622	*/
623	if (uErr & X86_TRAP_PF_RW) /* write fault? */
624	{
625	STAM_PROFILE_START(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,DirtyBitTracking), a);
626	rc = PGM_BTH_NAME(CheckDirtyPageFault)(pVCpu, uErr, &pPDDst->a[iPDDst], GstWalk.pPde, pvFault);
627	STAM_PROFILE_STOP(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,DirtyBitTracking), a);
628	if (rc == VINF_PGM_HANDLED_DIRTY_BIT_FAULT)
629	{
630	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0
631	= rc == VINF_PGM_HANDLED_DIRTY_BIT_FAULT
632	? &pVCpu->pgm.s.Stats.StatRZTrap0eTime2DirtyAndAccessed
633	: &pVCpu->pgm.s.Stats.StatRZTrap0eTime2GuestTrap; });
634	Log8(("Trap0eHandler: returns VINF_SUCCESS\n"));
635	return VINF_SUCCESS;
636	}
637	#ifdef DEBUG_bird
638	AssertMsg(GstWalk.Pde.u == GstWalk.pPde->u \|\| GstWalk.pPte->u == GstWalk.pPde->u \|\| pVM->cCpus > 1, ("%RX64 %RX64\n", (uint64_t)GstWalk.Pde.u, (uint64_t)GstWalk.pPde->u)); // - triggers with smp w7 guests.
639	AssertMsg(Walk.fBigPage \|\| GstWalk.Pte.u == GstWalk.pPte->u \|\| pVM->cCpus > 1, ("%RX64 %RX64\n", (uint64_t)GstWalk.Pte.u, (uint64_t)GstWalk.pPte->u)); // - ditto.
640	#endif
641	}
642
643	# if 0 /* rarely useful; leave for debugging. */
644	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0ePD[iPDSrc]);
645	# endif
646	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
647
648	/*
649	* A common case is the not-present error caused by lazy page table syncing.
650	*
651	* It is IMPORTANT that we weed out any access to non-present shadow PDEs
652	* here so we can safely assume that the shadow PT is present when calling
653	* SyncPage later.
654	*
655	* On failure, we ASSUME that SyncPT is out of memory or detected some kind
656	* of mapping conflict and defer to SyncCR3 in R3.
657	* (Again, we do NOT support access handlers for non-present guest pages.)
658	*
659	*/
660	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
661	Assert(GstWalk.Pde.u & X86_PDE_P);
662	# endif
663	if ( !(uErr & X86_TRAP_PF_P) /* not set means page not present instead of page protection violation */
664	&& !SHW_PDE_IS_P(pPDDst->a[iPDDst]))
665	{
666	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2SyncPT; });
667	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
668	LogFlow(("=>SyncPT %04x = %08RX64\n", (pvFault >> GST_PD_SHIFT) & GST_PD_MASK, (uint64_t)GstWalk.Pde.u));
669	rc = PGM_BTH_NAME(SyncPT)(pVCpu, (pvFault >> GST_PD_SHIFT) & GST_PD_MASK, GstWalk.pPd, pvFault);
670	# else
671	LogFlow(("=>SyncPT pvFault=%RGv\n", pvFault));
672	rc = PGM_BTH_NAME(SyncPT)(pVCpu, 0, NULL, pvFault);
673	# endif
674	if (RT_SUCCESS(rc))
675	return rc;
676	Log(("SyncPT: %RGv failed!! rc=%Rrc\n", pvFault, rc));
677	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
678	return VINF_PGM_SYNC_CR3;
679	}
680
681	/*
682	* Check if this fault address is flagged for special treatment,
683	* which means we'll have to figure out the physical address and
684	* check flags associated with it.
685	*
686	* ASSUME that we can limit any special access handling to pages
687	* in page tables which the guest believes to be present.
688	*/
689	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
690	RTGCPHYS GCPhys = Walk.GCPhys & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK;
691	# else
692	RTGCPHYS GCPhys = PGM_A20_APPLY(pVCpu, (RTGCPHYS)pvFault & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK);
693	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
694	PPGMPAGE pPage;
695	rc = pgmPhysGetPageEx(pVM, GCPhys, &pPage);
696	if (RT_FAILURE(rc))
697	{
698	/*
699	* When the guest accesses invalid physical memory (e.g. probing
700	* of RAM or accessing a remapped MMIO range), then we'll fall
701	* back to the recompiler to emulate the instruction.
702	*/
703	LogFlow(("PGM #PF: pgmPhysGetPageEx(%RGp) failed with %Rrc\n", GCPhys, rc));
704	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersInvalid);
705	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2InvalidPhys; });
706	return VINF_EM_RAW_EMULATE_INSTR;
707	}
708
709	/*
710	* Any handlers for this page?
711	*/
712	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage) && !PGM_PAGE_IS_HNDL_PHYS_NOT_IN_HM(pPage))
713	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
714	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(Trap0eHandlerDoAccessHandlers)(pVCpu, uErr, pCtx, pvFault, pPage, pfLockTaken,
715	&Walk, &GstWalk));
716	# else
717	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(Trap0eHandlerDoAccessHandlers)(pVCpu, uErr, pCtx, pvFault, pPage, pfLockTaken));
718	# endif
719
720	/*
721	* We are here only if page is present in Guest page tables and
722	* trap is not handled by our handlers.
723	*
724	* Check it for page out-of-sync situation.
725	*/
726	if (!(uErr & X86_TRAP_PF_P))
727	{
728	/*
729	* Page is not present in our page tables. Try to sync it!
730	*/
731	if (uErr & X86_TRAP_PF_US)
732	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,PageOutOfSyncUser));
733	else /* supervisor */
734	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,PageOutOfSyncSupervisor));
735
736	if (PGM_PAGE_IS_BALLOONED(pPage))
737	{
738	/* Emulate reads from ballooned pages as they are not present in
739	our shadow page tables. (Required for e.g. Solaris guests; soft
740	ecc, random nr generator.) */
741	rc = VBOXSTRICTRC_TODO(PGMInterpretInstruction(pVCpu, pvFault));
742	LogFlow(("PGM: PGMInterpretInstruction balloon -> rc=%d pPage=%R[pgmpage]\n", rc, pPage));
743	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,PageOutOfSyncBallloon));
744	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2Ballooned; });
745	return rc;
746	}
747
748	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
749	rc = PGM_BTH_NAME(SyncPage)(pVCpu, GstWalk.Pde, pvFault, PGM_SYNC_NR_PAGES, uErr);
750	# else
751	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrcDummy, pvFault, PGM_SYNC_NR_PAGES, uErr);
752	# endif
753	if (RT_SUCCESS(rc))
754	{
755	/* The page was successfully synced, return to the guest. */
756	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2OutOfSync; });
757	return VINF_SUCCESS;
758	}
759	}
760	else /* uErr & X86_TRAP_PF_P: */
761	{
762	/*
763	* Write protected pages are made writable when the guest makes the
764	* first write to it. This happens for pages that are shared, write
765	* monitored or not yet allocated.
766	*
767	* We may also end up here when CR0.WP=0 in the guest.
768	*
769	* Also, a side effect of not flushing global PDEs are out of sync
770	* pages due to physical monitored regions, that are no longer valid.
771	* Assume for now it only applies to the read/write flag.
772	*/
773	if (uErr & X86_TRAP_PF_RW)
774	{
775	/*
776	* Check if it is a read-only page.
777	*/
778	if (PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
779	{
780	Log(("PGM #PF: Make writable: %RGp %R[pgmpage] pvFault=%RGp uErr=%#x\n", GCPhys, pPage, pvFault, uErr));
781	# ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
782	Assert(!PGM_PAGE_IS_ZERO(pPage));
783	# endif
784	AssertFatalMsg(!PGM_PAGE_IS_BALLOONED(pPage), ("Unexpected ballooned page at %RGp\n", GCPhys));
785	# ifdef PGM_WITH_PAGE_ZEROING_DETECTION
786	if ( PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ZERO
787	&& (pvFault & X86_PAGE_OFFSET_MASK) == 0
788	&& pgmHandlePageZeroingCode(pVCpu, pCtx))
789	{
790	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2PageZeroing; });
791	return VINF_SUCCESS;
792	}
793	# endif
794	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2MakeWritable; });
795
796	rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhys);
797	if (rc != VINF_SUCCESS)
798	{
799	AssertMsg(rc == VINF_PGM_SYNC_CR3 \|\| RT_FAILURE(rc), ("%Rrc\n", rc));
800	return rc;
801	}
802	if (RT_UNLIKELY(VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)))
803	return VINF_EM_NO_MEMORY;
804	}
805
806	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
807	/*
808	* Check to see if we need to emulate the instruction if CR0.WP=0.
809	*/
810	if ( !(Walk.fEffective & PGM_PTATTRS_W_MASK)
811	&& (CPUMGetGuestCR0(pVCpu) & (X86_CR0_WP \| X86_CR0_PG)) == X86_CR0_PG
812	&& CPUMGetGuestCPL(pVCpu) < 3)
813	{
814	Assert((uErr & (X86_TRAP_PF_RW \| X86_TRAP_PF_P)) == (X86_TRAP_PF_RW \| X86_TRAP_PF_P));
815
816	/*
817	* The Netware WP0+RO+US hack.
818	*
819	* Netware sometimes(/always?) runs with WP0. It has been observed doing
820	* excessive write accesses to pages which are mapped with US=1 and RW=0
821	* while WP=0. This causes a lot of exits and extremely slow execution.
822	* To avoid trapping and emulating every write here, we change the shadow
823	* page table entry to map it as US=0 and RW=1 until user mode tries to
824	* access it again (see further below). We count these shadow page table
825	* changes so we can avoid having to clear the page pool every time the WP
826	* bit changes to 1 (see PGMCr0WpEnabled()).
827	*/
828	# if (PGM_GST_TYPE == PGM_TYPE_32BIT \|\| PGM_GST_TYPE == PGM_TYPE_PAE) && 1
829	if ( (Walk.fEffective & (PGM_PTATTRS_W_MASK \| PGM_PTATTRS_US_MASK)) == PGM_PTATTRS_US_MASK
830	&& (Walk.fBigPage \|\| (GstWalk.Pde.u & X86_PDE_RW))
831	&& pVM->cCpus == 1 /* Sorry, no go on SMP. Add CFGM option? */)
832	{
833	Log(("PGM #PF: Netware WP0+RO+US hack: pvFault=%RGp uErr=%#x (big=%d)\n", pvFault, uErr, Walk.fBigPage));
834	rc = pgmShwMakePageSupervisorAndWritable(pVCpu, pvFault, Walk.fBigPage, PGM_MK_PG_IS_WRITE_FAULT);
835	if (rc == VINF_SUCCESS \|\| rc == VINF_PGM_SYNC_CR3)
836	{
837	PGM_INVL_PG(pVCpu, pvFault);
838	pVCpu->pgm.s.cNetwareWp0Hacks++;
839	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2Wp0RoUsHack; });
840	return rc;
841	}
842	AssertMsg(RT_FAILURE_NP(rc), ("%Rrc\n", rc));
843	Log(("pgmShwMakePageSupervisorAndWritable(%RGv) failed with rc=%Rrc - ignored\n", pvFault, rc));
844	}
845	# endif
846
847	/* Interpret the access. */
848	rc = VBOXSTRICTRC_TODO(PGMInterpretInstruction(pVCpu, pvFault));
849	Log(("PGM #PF: WP0 emulation (pvFault=%RGp uErr=%#x cpl=%d fBig=%d fEffUs=%d)\n", pvFault, uErr, CPUMGetGuestCPL(pVCpu), Walk.fBigPage, !!(Walk.fEffective & PGM_PTATTRS_US_MASK)));
850	if (RT_SUCCESS(rc))
851	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eWPEmulInRZ);
852	else
853	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eWPEmulToR3);
854	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2WPEmulation; });
855	return rc;
856	}
857	# endif
858	/// @todo count the above case; else
859	if (uErr & X86_TRAP_PF_US)
860	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,PageOutOfSyncUserWrite));
861	else /* supervisor */
862	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,PageOutOfSyncSupervisorWrite));
863
864	/*
865	* Sync the page.
866	*
867	* Note: Do NOT use PGM_SYNC_NR_PAGES here. That only works if the
868	* page is not present, which is not true in this case.
869	*/
870	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
871	rc = PGM_BTH_NAME(SyncPage)(pVCpu, GstWalk.Pde, pvFault, 1, uErr);
872	# else
873	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrcDummy, pvFault, 1, uErr);
874	# endif
875	if (RT_SUCCESS(rc))
876	{
877	/*
878	* Page was successfully synced, return to guest but invalidate
879	* the TLB first as the page is very likely to be in it.
880	*/
881	# if PGM_SHW_TYPE == PGM_TYPE_EPT
882	HMInvalidatePhysPage(pVM, (RTGCPHYS)pvFault);
883	# else
884	PGM_INVL_PG(pVCpu, pvFault);
885	# endif
886	# ifdef VBOX_STRICT
887	PGMPTWALK GstPageWalk;
888	GstPageWalk.GCPhys = RTGCPHYS_MAX;
889	if (!pVM->pgm.s.fNestedPaging)
890	{
891	rc = PGMGstGetPage(pVCpu, pvFault, &GstPageWalk);
892	AssertMsg(RT_SUCCESS(rc) && ((GstPageWalk.fEffective & X86_PTE_RW) \|\| ((CPUMGetGuestCR0(pVCpu) & (X86_CR0_WP \| X86_CR0_PG)) == X86_CR0_PG && CPUMGetGuestCPL(pVCpu) < 3)), ("rc=%Rrc fPageGst=%RX64\n", rc, GstPageWalk.fEffective));
893	LogFlow(("Obsolete physical monitor page out of sync %RGv - phys %RGp flags=%08llx\n", pvFault, GstPageWalk.GCPhys, GstPageWalk.fEffective));
894	}
895	# if 0 /* Bogus! Triggers incorrectly with w7-64 and later for the SyncPage case: "Pde at %RGv changed behind our back?" */
896	uint64_t fPageShw = 0;
897	rc = PGMShwGetPage(pVCpu, pvFault, &fPageShw, NULL);
898	AssertMsg((RT_SUCCESS(rc) && (fPageShw & X86_PTE_RW)) \|\| pVM->cCpus > 1 /* new monitor can be installed/page table flushed between the trap exit and PGMTrap0eHandler */,
899	("rc=%Rrc fPageShw=%RX64 GCPhys2=%RGp fPageGst=%RX64 pvFault=%RGv\n", rc, fPageShw, GstPageWalk.GCPhys, fPageGst, pvFault));
900	# endif
901	# endif /* VBOX_STRICT */
902	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2OutOfSyncHndObs; });
903	return VINF_SUCCESS;
904	}
905	}
906	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
907	/*
908	* Check for Netware WP0+RO+US hack from above and undo it when user
909	* mode accesses the page again.
910	*/
911	else if ( (Walk.fEffective & (PGM_PTATTRS_W_MASK \| PGM_PTATTRS_US_MASK)) == PGM_PTATTRS_US_MASK
912	&& (Walk.fBigPage \|\| (GstWalk.Pde.u & X86_PDE_RW))
913	&& pVCpu->pgm.s.cNetwareWp0Hacks > 0
914	&& (CPUMGetGuestCR0(pVCpu) & (X86_CR0_WP \| X86_CR0_PG)) == X86_CR0_PG
915	&& CPUMGetGuestCPL(pVCpu) == 3
916	&& pVM->cCpus == 1
917	)
918	{
919	Log(("PGM #PF: Undo netware WP0+RO+US hack: pvFault=%RGp uErr=%#x\n", pvFault, uErr));
920	rc = PGM_BTH_NAME(SyncPage)(pVCpu, GstWalk.Pde, pvFault, 1, uErr);
921	if (RT_SUCCESS(rc))
922	{
923	PGM_INVL_PG(pVCpu, pvFault);
924	pVCpu->pgm.s.cNetwareWp0Hacks--;
925	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2Wp0RoUsUnhack; });
926	return VINF_SUCCESS;
927	}
928	}
929	# endif /* PGM_WITH_PAGING */
930
931	/** @todo else: why are we here? */
932
933	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && defined(VBOX_STRICT)
934	/*
935	* Check for VMM page flags vs. Guest page flags consistency.
936	* Currently only for debug purposes.
937	*/
938	if (RT_SUCCESS(rc))
939	{
940	/* Get guest page flags. */
941	PGMPTWALK GstPageWalk;
942	int rc2 = PGMGstGetPage(pVCpu, pvFault, &GstPageWalk);
943	if (RT_SUCCESS(rc2))
944	{
945	uint64_t fPageShw = 0;
946	rc2 = PGMShwGetPage(pVCpu, pvFault, &fPageShw, NULL);
947
948	#if 0
949	/*
950	* Compare page flags.
951	* Note: we have AVL, A, D bits desynced.
952	*/
953	AssertMsg( (fPageShw & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK))
954	== (fPageGst & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK))
955	\|\| ( pVCpu->pgm.s.cNetwareWp0Hacks > 0
956	&& (fPageShw & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK \| X86_PTE_RW \| X86_PTE_US))
957	== (fPageGst & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK \| X86_PTE_RW \| X86_PTE_US))
958	&& (fPageShw & (X86_PTE_RW \| X86_PTE_US)) == X86_PTE_RW
959	&& (fPageGst & (X86_PTE_RW \| X86_PTE_US)) == X86_PTE_US),
960	("Page flags mismatch! pvFault=%RGv uErr=%x GCPhys=%RGp fPageShw=%RX64 fPageGst=%RX64 rc=%d\n",
961	pvFault, (uint32_t)uErr, GCPhys, fPageShw, fPageGst, rc));
962	01:01:15.623511 00:08:43.266063 Expression: (fPageShw & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK)) == (fPageGst & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK)) \|\| ( pVCpu->pgm.s.cNetwareWp0Hacks > 0 && (fPageShw & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK \| X86_PTE_RW \| X86_PTE_US)) == (fPageGst & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK \| X86_PTE_RW \| X86_PTE_US)) && (fPageShw & (X86_PTE_RW \| X86_PTE_US)) == X86_PTE_RW && (fPageGst & (X86_PTE_RW \| X86_PTE_US)) == X86_PTE_US)
963	01:01:15.623511 00:08:43.266064 Location : e:\vbox\svn\trunk\srcPage flags mismatch! pvFault=fffff801b0d7b000 uErr=11 GCPhys=0000000019b52000 fPageShw=0 fPageGst=77b0000000000121 rc=0
964
965	01:01:15.625516 00:08:43.268051 Expression: (fPageShw & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK)) == (fPageGst & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK)) \|\| ( pVCpu->pgm.s.cNetwareWp0Hacks > 0 && (fPageShw & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK \| X86_PTE_RW \| X86_PTE_US)) == (fPageGst & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK \| X86_PTE_RW \| X86_PTE_US)) && (fPageShw & (X86_PTE_RW \| X86_PTE_US)) == X86_PTE_RW && (fPageGst & (X86_PTE_RW \| X86_PTE_US)) == X86_PTE_US)
966	01:01:15.625516 00:08:43.268051 Location :
967	e:\vbox\svn\trunk\srcPage flags mismatch!
968	pvFault=fffff801b0d7b000
969	uErr=11 X86_TRAP_PF_ID \| X86_TRAP_PF_P
970	GCPhys=0000000019b52000
971	fPageShw=0
972	fPageGst=77b0000000000121
973	rc=0
974	#endif
975
976	}
977	else
978	AssertMsgFailed(("PGMGstGetPage rc=%Rrc\n", rc));
979	}
980	else
981	AssertMsgFailed(("PGMGCGetPage rc=%Rrc\n", rc));
982	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && VBOX_STRICT */
983	}
984
985
986	/*
987	* If we get here it is because something failed above, i.e. most like guru
988	* meditiation time.
989	*/
990	LogRel(("%s: returns rc=%Rrc pvFault=%RGv uErr=%RX64 cs:rip=%04x:%08RX64\n",
991	__PRETTY_FUNCTION__, rc, pvFault, (uint64_t)uErr, pCtx->cs.Sel, pCtx->rip));
992	return rc;
993
994	# else /* Nested paging, EPT except PGM_GST_TYPE = PROT, NONE. */
995	NOREF(uErr); NOREF(pCtx); NOREF(pvFault);
996	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_SHW_TYPE, PGM_GST_TYPE));
997	return VERR_PGM_NOT_USED_IN_MODE;
998	# endif
999	}
1000
1001
1002	# if defined(VBOX_WITH_NESTED_HWVIRT_VMX_EPT)
1003	/**
1004	* Deals with a nested-guest \#PF fault for a guest-physical page with a handler.
1005	*
1006	* @returns Strict VBox status code.
1007	* @param pVCpu The cross context virtual CPU structure.
1008	* @param uErr The error code.
1009	* @param pCtx Pointer to the register context for the CPU.
1010	* @param GCPhysNestedFault The nested-guest physical address of the fault.
1011	* @param pPage The guest page at @a GCPhysNestedFault.
1012	* @param GCPhysFault The guest-physical address of the fault.
1013	* @param pGstWalkAll The guest page walk result.
1014	* @param pfLockTaken Where to store whether the PGM is still held when
1015	* this function completes.
1016	*
1017	* @note The caller has taken the PGM lock.
1018	*/
1019	static VBOXSTRICTRC PGM_BTH_NAME(NestedTrap0eHandlerDoAccessHandlers)(PVMCPUCC pVCpu, RTGCUINT uErr, PCPUMCTX pCtx,
1020	RTGCPHYS GCPhysNestedFault, PPGMPAGE pPage,
1021	RTGCPHYS GCPhysFault, PPGMPTWALKGST pGstWalkAll,
1022	bool *pfLockTaken)
1023	{
1024	# if PGM_GST_TYPE == PGM_TYPE_PROT \
1025	&& PGM_SHW_TYPE == PGM_TYPE_EPT
1026
1027	/** @todo Assert uErr isn't X86_TRAP_PF_RSVD and remove release checks. */
1028	PGM_A20_ASSERT_MASKED(pVCpu, GCPhysFault);
1029	AssertMsgReturn(PGM_PAGE_HAS_ANY_PHYSICAL_HANDLERS(pPage), ("%RGp %RGp uErr=%u\n", GCPhysNestedFault, GCPhysFault, uErr),
1030	VERR_PGM_HANDLER_IPE_1);
1031
1032	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1033	RTGCPHYS const GCPhysNestedPage = GCPhysNestedFault & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK;
1034	RTGCPHYS const GCPhysPage = GCPhysFault & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK;
1035
1036	/*
1037	* Physical page access handler.
1038	*/
1039	PPGMPHYSHANDLER pCur;
1040	VBOXSTRICTRC rcStrict = pgmHandlerPhysicalLookup(pVM, GCPhysPage, &pCur);
1041	AssertRCReturn(VBOXSTRICTRC_VAL(rcStrict), rcStrict);
1042
1043	PCPGMPHYSHANDLERTYPEINT const pCurType = PGMPHYSHANDLER_GET_TYPE(pVM, pCur);
1044	Assert(pCurType);
1045
1046	/*
1047	* If the region is write protected and we got a page not present fault, then sync
1048	* the pages. If the fault was caused by a read, then restart the instruction.
1049	* In case of write access continue to the GC write handler.
1050	*/
1051	if ( !(uErr & X86_TRAP_PF_P)
1052	&& pCurType->enmKind == PGMPHYSHANDLERKIND_WRITE)
1053	{
1054	Log7Func(("Syncing Monitored: GCPhysNestedPage=%RGp GCPhysPage=%RGp uErr=%#x\n", GCPhysNestedPage, GCPhysPage, uErr));
1055	rcStrict = PGM_BTH_NAME(NestedSyncPage)(pVCpu, GCPhysNestedPage, GCPhysPage, 1 /cPages/, uErr, pGstWalkAll);
1056	Assert(rcStrict != VINF_PGM_SYNCPAGE_MODIFIED_PDE);
1057	if ( RT_FAILURE(rcStrict)
1058	\|\| !(uErr & X86_TRAP_PF_RW))
1059	{
1060	AssertMsgRC(rcStrict, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
1061	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersOutOfSync);
1062	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2OutOfSyncHndPhys; });
1063	return rcStrict;
1064	}
1065	}
1066	else if ( !(uErr & X86_TRAP_PF_RSVD)
1067	&& pCurType->enmKind != PGMPHYSHANDLERKIND_WRITE)
1068	{
1069	/*
1070	* If the access was NOT through an EPT misconfig (i.e. RSVD), sync the page.
1071	* This can happen for the VMX APIC-access page.
1072	*/
1073	Log7Func(("Syncing MMIO: GCPhysNestedPage=%RGp GCPhysPage=%RGp\n", GCPhysNestedPage, GCPhysPage));
1074	rcStrict = PGM_BTH_NAME(NestedSyncPage)(pVCpu, GCPhysNestedPage, GCPhysPage, 1 /cPages/, uErr, pGstWalkAll);
1075	Assert(rcStrict != VINF_PGM_SYNCPAGE_MODIFIED_PDE);
1076	if (RT_FAILURE(rcStrict))
1077	{
1078	AssertMsgRC(rcStrict, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
1079	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersOutOfSync);
1080	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2OutOfSyncHndPhys; });
1081	return rcStrict;
1082	}
1083	}
1084
1085	AssertMsg( pCurType->enmKind != PGMPHYSHANDLERKIND_WRITE
1086	\|\| (pCurType->enmKind == PGMPHYSHANDLERKIND_WRITE && (uErr & X86_TRAP_PF_RW)),
1087	("Unexpected trap for physical handler: %08X (phys=%08x) pPage=%R[pgmpage] uErr=%X, enmKind=%d\n",
1088	GCPhysNestedFault, GCPhysFault, pPage, uErr, pCurType->enmKind));
1089	if (pCurType->enmKind == PGMPHYSHANDLERKIND_WRITE)
1090	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersPhysWrite);
1091	else
1092	{
1093	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersPhysAll);
1094	if (uErr & X86_TRAP_PF_RSVD)
1095	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersPhysAllOpt);
1096	}
1097
1098	if (pCurType->pfnPfHandler)
1099	{
1100	STAM_PROFILE_START(&pCur->Stat, h);
1101	uint64_t const uUser = !pCurType->fRing0DevInsIdx ? pCur->uUser
1102	: (uintptr_t)PDMDeviceRing0IdxToInstance(pVM, pCur->uUser);
1103
1104	if (pCurType->fKeepPgmLock)
1105	{
1106	rcStrict = pCurType->pfnPfHandler(pVM, pVCpu, uErr, pCtx, GCPhysNestedFault, GCPhysFault, uUser);
1107	STAM_PROFILE_STOP(&pCur->Stat, h);
1108	}
1109	else
1110	{
1111	PGM_UNLOCK(pVM);
1112	*pfLockTaken = false;
1113	rcStrict = pCurType->pfnPfHandler(pVM, pVCpu, uErr, pCtx, GCPhysNestedFault, GCPhysFault, uUser);
1114	STAM_PROFILE_STOP(&pCur->Stat, h); /* no locking needed, entry is unlikely reused before we get here. */
1115	}
1116	}
1117	else
1118	{
1119	AssertMsgFailed(("What's going on here!? Fault falls outside handler range!?\n"));
1120	rcStrict = VINF_EM_RAW_EMULATE_INSTR;
1121	}
1122
1123	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2HndPhys; });
1124	return rcStrict;
1125
1126	# else
1127	RT_NOREF8(pVCpu, uErr, pCtx, GCPhysNestedFault, pPage, GCPhysFault, pGstWalkAll, pfLockTaken);
1128	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_SHW_TYPE, PGM_GST_TYPE));
1129	return VERR_PGM_NOT_USED_IN_MODE;
1130	# endif
1131	}
1132	# endif /* VBOX_WITH_NESTED_HWVIRT_VMX_EPT */
1133
1134
1135	/**
1136	* Nested \#PF handler for nested-guest hardware-assisted execution using nested
1137	* paging.
1138	*
1139	* @returns VBox status code (appropriate for trap handling and GC return).
1140	* @param pVCpu The cross context virtual CPU structure.
1141	* @param uErr The fault error (X86_TRAP_PF_*).
1142	* @param pCtx Pointer to the register context for the CPU.
1143	* @param GCPhysNestedFault The nested-guest physical address of the fault.
1144	* @param fIsLinearAddrValid Whether translation of a nested-guest linear address
1145	* caused this fault. If @c false, GCPtrNestedFault
1146	* must be 0.
1147	* @param GCPtrNestedFault The nested-guest linear address of this fault.
1148	* @param pWalk The guest page table walk result.
1149	* @param pfLockTaken Where to store whether the PGM lock is still held
1150	* when this function completes.
1151	*/
1152	PGM_BTH_DECL(int, NestedTrap0eHandler)(PVMCPUCC pVCpu, RTGCUINT uErr, PCPUMCTX pCtx, RTGCPHYS GCPhysNestedFault,
1153	bool fIsLinearAddrValid, RTGCPTR GCPtrNestedFault, PPGMPTWALK pWalk, bool *pfLockTaken)
1154	{
1155	*pfLockTaken = false;
1156	# if defined(VBOX_WITH_NESTED_HWVIRT_VMX_EPT) \
1157	&& PGM_GST_TYPE == PGM_TYPE_PROT \
1158	&& PGM_SHW_TYPE == PGM_TYPE_EPT
1159
1160	Assert(CPUMIsGuestVmxEptPagingEnabled(pVCpu));
1161	Assert(PGM_A20_IS_ENABLED(pVCpu));
1162
1163	/* We don't support mode-based execute control for EPT yet. */
1164	Assert(!pVCpu->CTX_SUFF(pVM)->cpum.ro.GuestFeatures.fVmxModeBasedExecuteEpt);
1165	Assert(!(uErr & X86_TRAP_PF_US));
1166
1167	/* Take the big lock now. */
1168	*pfLockTaken = true;
1169	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1170	PGM_LOCK_VOID(pVM);
1171
1172	/*
1173	* Walk the guest EPT tables and check if it's an EPT violation or misconfiguration.
1174	*/
1175	if (fIsLinearAddrValid)
1176	Log7Func(("cs:rip=%04x:%#08RX64 GCPhysNestedFault=%RGp uErr=%#x GCPtrNestedFault=%RGv\n",
1177	pCtx->cs.Sel, pCtx->rip, GCPhysNestedFault, uErr, GCPtrNestedFault));
1178	else
1179	Log7Func(("cs:rip=%04x:%#08RX64 GCPhysNestedFault=%RGp uErr=%#x\n",
1180	pCtx->cs.Sel, pCtx->rip, GCPhysNestedFault, uErr));
1181	PGMPTWALKGST GstWalkAll;
1182	int rc = pgmGstSlatWalk(pVCpu, GCPhysNestedFault, fIsLinearAddrValid, GCPtrNestedFault, pWalk, &GstWalkAll);
1183	if (RT_FAILURE(rc))
1184	return rc;
1185
1186	Assert(GstWalkAll.enmType == PGMPTWALKGSTTYPE_EPT);
1187	Assert(pWalk->fSucceeded);
1188	Assert(pWalk->fEffective & (PGM_PTATTRS_EPT_R_MASK \| PGM_PTATTRS_EPT_W_MASK \| PGM_PTATTRS_EPT_X_SUPER_MASK));
1189	Assert(pWalk->fIsSlat);
1190
1191	# ifdef DEBUG_ramshankar
1192	/* Paranoia. */
1193	Assert(RT_BOOL(pWalk->fEffective & PGM_PTATTRS_R_MASK) == RT_BOOL(pWalk->fEffective & PGM_PTATTRS_EPT_R_MASK));
1194	Assert(RT_BOOL(pWalk->fEffective & PGM_PTATTRS_W_MASK) == RT_BOOL(pWalk->fEffective & PGM_PTATTRS_EPT_W_MASK));
1195	Assert(RT_BOOL(pWalk->fEffective & PGM_PTATTRS_NX_MASK) == !RT_BOOL(pWalk->fEffective & PGM_PTATTRS_EPT_X_SUPER_MASK));
1196	# endif
1197
1198	Log7Func(("SLAT: GCPhysNestedFault=%RGp -> GCPhys=%#RGp\n", GCPhysNestedFault, pWalk->GCPhys));
1199
1200	/*
1201	* Check page-access permissions.
1202	*/
1203	if ( ((uErr & X86_TRAP_PF_RW) && !(pWalk->fEffective & PGM_PTATTRS_W_MASK))
1204	\|\| ((uErr & X86_TRAP_PF_ID) && (pWalk->fEffective & PGM_PTATTRS_NX_MASK)))
1205	{
1206	Log7Func(("Permission failed! GCPtrNested=%RGv GCPhysNested=%RGp uErr=%#x fEffective=%#RX64\n", GCPtrNestedFault,
1207	GCPhysNestedFault, uErr, pWalk->fEffective));
1208	pWalk->fFailed = PGM_WALKFAIL_EPT_VIOLATION;
1209	return VERR_ACCESS_DENIED;
1210	}
1211
1212	PGM_A20_ASSERT_MASKED(pVCpu, pWalk->GCPhys);
1213	RTGCPHYS const GCPhysPage = pWalk->GCPhys & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK;
1214	RTGCPHYS const GCPhysNestedPage = GCPhysNestedFault & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK;
1215
1216	/*
1217	* If we were called via an EPT misconfig, it should've already resulted in a nested-guest VM-exit.
1218	*/
1219	AssertMsgReturn(!(uErr & X86_TRAP_PF_RSVD),
1220	("Unexpected EPT misconfig VM-exit. GCPhysPage=%RGp GCPhysNestedPage=%RGp\n", GCPhysPage, GCPhysNestedPage),
1221	VERR_PGM_MAPPING_IPE);
1222
1223	/*
1224	* Fetch and sync the nested-guest EPT page directory pointer.
1225	*/
1226	PEPTPD pEptPd;
1227	rc = pgmShwGetNestedEPTPDPtr(pVCpu, GCPhysNestedPage, NULL /ppPdpt/, &pEptPd, &GstWalkAll);
1228	AssertRCReturn(rc, rc);
1229	Assert(pEptPd);
1230
1231	/*
1232	* A common case is the not-present error caused by lazy page table syncing.
1233	*
1234	* It is IMPORTANT that we weed out any access to non-present shadow PDEs
1235	* here so we can safely assume that the shadow PT is present when calling
1236	* NestedSyncPage later.
1237	*
1238	* NOTE: It's possible we will be syncing the VMX APIC-access page here.
1239	* In that case, we would sync the page but will NOT go ahead with emulating
1240	* the APIC-access VM-exit through IEM. However, once the page is mapped in
1241	* the shadow tables, subsequent APIC-access VM-exits for the nested-guest
1242	* will be triggered by hardware. Maybe calling the IEM #PF handler can be
1243	* considered as an optimization later.
1244	*/
1245	unsigned const iPde = (GCPhysNestedPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1246	if ( !(uErr & X86_TRAP_PF_P)
1247	&& !(pEptPd->a[iPde].u & EPT_PRESENT_MASK))
1248	{
1249	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2SyncPT; });
1250	Log7Func(("NestedSyncPT: Lazy. GCPhysNestedPage=%RGp GCPhysPage=%RGp\n", GCPhysNestedPage, GCPhysPage));
1251	rc = PGM_BTH_NAME(NestedSyncPT)(pVCpu, GCPhysNestedPage, GCPhysPage, &GstWalkAll);
1252	if (RT_SUCCESS(rc))
1253	return rc;
1254	AssertMsgFailedReturn(("NestedSyncPT: %RGv failed! rc=%Rrc\n", GCPhysNestedPage, rc), VERR_PGM_MAPPING_IPE);
1255	}
1256
1257	/*
1258	* Check if this fault address is flagged for special treatment.
1259	* This handles faults on an MMIO or write-monitored page.
1260	*
1261	* If this happens to be the VMX APIC-access page, we don't treat is as MMIO
1262	* but rather sync it further below (as a regular guest page) which lets
1263	* hardware-assisted execution trigger the APIC-access VM-exits of the
1264	* nested-guest directly.
1265	*/
1266	PPGMPAGE pPage;
1267	rc = pgmPhysGetPageEx(pVM, GCPhysPage, &pPage);
1268	if (RT_FAILURE(rc))
1269	{
1270	/*
1271	* We failed to get the physical page which means it's a reserved/invalid
1272	* page address (not MMIO even). This can typically be observed with
1273	* Microsoft Hyper-V enabled Windows guests. We must fall back to emulating
1274	* the instruction, see @bugref{10318#c7}.
1275	*/
1276	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.StatRZTrap0eHandlersInvalid);
1277	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2InvalidPhys; });
1278	return VINF_EM_RAW_EMULATE_INSTR;
1279	}
1280	/* Check if this is an MMIO page and NOT the VMX APIC-access page. */
1281	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage) && !PGM_PAGE_IS_HNDL_PHYS_NOT_IN_HM(pPage))
1282	{
1283	Log7Func(("MMIO: Calling NestedTrap0eHandlerDoAccessHandlers for GCPhys %RGp\n", GCPhysPage));
1284	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(NestedTrap0eHandlerDoAccessHandlers)(pVCpu, uErr, pCtx, GCPhysNestedFault,
1285	pPage, pWalk->GCPhys, &GstWalkAll,
1286	pfLockTaken));
1287	}
1288
1289	/*
1290	* We are here only if page is present in nested-guest page tables but the
1291	* trap is not handled by our handlers. Check for page out-of-sync situation.
1292	*/
1293	if (!(uErr & X86_TRAP_PF_P))
1294	{
1295	Assert(!PGM_PAGE_IS_BALLOONED(pPage));
1296	Assert(!(uErr & X86_TRAP_PF_US)); /* Mode-based execute not supported yet. */
1297	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,PageOutOfSyncSupervisor));
1298
1299	Log7Func(("SyncPage: Not-Present: GCPhysNestedPage=%RGp GCPhysPage=%RGp\n", GCPhysNestedFault, GCPhysPage));
1300	rc = PGM_BTH_NAME(NestedSyncPage)(pVCpu, GCPhysNestedPage, GCPhysPage, PGM_SYNC_NR_PAGES, uErr, &GstWalkAll);
1301	if (RT_SUCCESS(rc))
1302	{
1303	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2OutOfSync; });
1304	return VINF_SUCCESS;
1305	}
1306	}
1307	else if (uErr & X86_TRAP_PF_RW)
1308	{
1309	/*
1310	* Write protected pages are made writable when the guest makes the
1311	* first write to it. This happens for pages that are shared, write
1312	* monitored or not yet allocated.
1313	*
1314	* We may also end up here when CR0.WP=0 in the guest.
1315	*
1316	* Also, a side effect of not flushing global PDEs are out of sync
1317	* pages due to physical monitored regions, that are no longer valid.
1318	* Assume for now it only applies to the read/write flag.
1319	*/
1320	if (PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
1321	{
1322	/* This is a read-only page. */
1323	AssertFatalMsg(!PGM_PAGE_IS_BALLOONED(pPage), ("Unexpected ballooned page at %RGp\n", GCPhysPage));
1324	#ifdef PGM_WITH_PAGE_ZEROING_DETECTION
1325	if ( PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ZERO
1326	&& (GCPhysNestedFault & X86_PAGE_OFFSET_MASK) == 0
1327	&& pgmHandlePageZeroingCode(pVCpu, pCtx))
1328	{
1329	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2PageZeroing; });
1330	return VINF_SUCCESS;
1331	}
1332	#endif
1333	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2MakeWritable; });
1334
1335	Log7Func(("Calling pgmPhysPageMakeWritable for GCPhysPage=%RGp\n", GCPhysPage));
1336	rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhysPage);
1337	if (rc != VINF_SUCCESS)
1338	{
1339	AssertMsg(rc == VINF_PGM_SYNC_CR3 \|\| RT_FAILURE(rc), ("%Rrc\n", rc));
1340	return rc;
1341	}
1342	if (RT_UNLIKELY(VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)))
1343	return VINF_EM_NO_MEMORY;
1344	}
1345
1346	Assert(!(uErr & X86_TRAP_PF_US)); /* Mode-based execute not supported yet. */
1347	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,PageOutOfSyncSupervisorWrite));
1348
1349	/*
1350	* Sync the write-protected page.
1351	* Note: Do NOT use PGM_SYNC_NR_PAGES here. That only works if the
1352	* page is not present, which is not true in this case.
1353	*/
1354	Log7Func(("SyncPage: RW: cs:rip=%04x:%#RX64 GCPhysNestedPage=%RGp uErr=%#RX32 GCPhysPage=%RGp WalkGCPhys=%RGp\n",
1355	pCtx->cs.Sel, pCtx->rip, GCPhysNestedPage, (uint32_t)uErr, GCPhysPage, pWalk->GCPhys));
1356	rc = PGM_BTH_NAME(NestedSyncPage)(pVCpu, GCPhysNestedPage, GCPhysPage, 1 /* cPages */, uErr, &GstWalkAll);
1357	if (RT_SUCCESS(rc))
1358	{
1359	HMInvalidatePhysPage(pVM, GCPhysPage);
1360	STAM_STATS({ pVCpu->pgmr0.s.pStatTrap0eAttributionR0 = &pVCpu->pgm.s.Stats.StatRZTrap0eTime2OutOfSyncHndObs; });
1361	return VINF_SUCCESS;
1362	}
1363	}
1364
1365	/*
1366	* If we get here it is because something failed above => guru meditation time?
1367	*/
1368	LogRelMaxFunc(32, ("rc=%Rrc GCPhysNestedFault=%#RGp (%#RGp) uErr=%#RX32 cs:rip=%04x:%08RX64\n",
1369	rc, GCPhysNestedFault, GCPhysPage, (uint32_t)uErr, pCtx->cs.Sel, pCtx->rip));
1370	return VERR_PGM_MAPPING_IPE;
1371
1372	# else /* !VBOX_WITH_NESTED_HWVIRT_VMX_EPT \|\| PGM_GST_TYPE != PGM_TYPE_PROT \|\| PGM_SHW_TYPE != PGM_TYPE_EPT */
1373	RT_NOREF7(pVCpu, uErr, pCtx, GCPhysNestedFault, fIsLinearAddrValid, GCPtrNestedFault, pWalk);
1374	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_SHW_TYPE, PGM_GST_TYPE));
1375	return VERR_PGM_NOT_USED_IN_MODE;
1376	# endif
1377	}
1378
1379	#endif /* !IN_RING3 */
1380
1381
1382	/**
1383	* Emulation of the invlpg instruction.
1384	*
1385	*
1386	* @returns VBox status code.
1387	*
1388	* @param pVCpu The cross context virtual CPU structure.
1389	* @param GCPtrPage Page to invalidate.
1390	*
1391	* @remark ASSUMES that the guest is updating before invalidating. This order
1392	* isn't required by the CPU, so this is speculative and could cause
1393	* trouble.
1394	* @remark No TLB shootdown is done on any other VCPU as we assume that
1395	* invlpg emulation is the only reason for calling this function.
1396	* (The guest has to shoot down TLB entries on other CPUs itself)
1397	* Currently true, but keep in mind!
1398	*
1399	* @todo Clean this up! Most of it is (or should be) no longer necessary as we catch all page table accesses.
1400	* Should only be required when PGMPOOL_WITH_OPTIMIZED_DIRTY_PT is active (PAE or AMD64 (for now))
1401	*/
1402	PGM_BTH_DECL(int, InvalidatePage)(PVMCPUCC pVCpu, RTGCPTR GCPtrPage)
1403	{
1404	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) \
1405	&& !PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) \
1406	&& PGM_SHW_TYPE != PGM_TYPE_NONE
1407	int rc;
1408	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1409	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1410
1411	PGM_LOCK_ASSERT_OWNER(pVM);
1412
1413	LogFlow(("InvalidatePage %RGv\n", GCPtrPage));
1414
1415	/*
1416	* Get the shadow PD entry and skip out if this PD isn't present.
1417	* (Guessing that it is frequent for a shadow PDE to not be present, do this first.)
1418	*/
1419	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
1420	const unsigned iPDDst = (uint32_t)GCPtrPage >> SHW_PD_SHIFT;
1421	PX86PDE pPdeDst = pgmShwGet32BitPDEPtr(pVCpu, GCPtrPage);
1422	AssertReturn(pPdeDst, VERR_INTERNAL_ERROR_3);
1423
1424	/* Fetch the pgm pool shadow descriptor. */
1425	PPGMPOOLPAGE pShwPde = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
1426	# ifdef IN_RING3 /* Possible we didn't resync yet when called from REM. */
1427	if (!pShwPde)
1428	{
1429	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,InvalidatePageSkipped));
1430	return VINF_SUCCESS;
1431	}
1432	# else
1433	Assert(pShwPde);
1434	# endif
1435
1436	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
1437	const unsigned iPdpt = (uint32_t)GCPtrPage >> X86_PDPT_SHIFT;
1438	PX86PDPT pPdptDst = pgmShwGetPaePDPTPtr(pVCpu);
1439
1440	/* If the shadow PDPE isn't present, then skip the invalidate. */
1441	# ifdef IN_RING3 /* Possible we didn't resync yet when called from REM. */
1442	if (!pPdptDst \|\| !(pPdptDst->a[iPdpt].u & X86_PDPE_P))
1443	# else
1444	if (!(pPdptDst->a[iPdpt].u & X86_PDPE_P))
1445	# endif
1446	{
1447	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,InvalidatePageSkipped));
1448	PGM_INVL_PG(pVCpu, GCPtrPage);
1449	return VINF_SUCCESS;
1450	}
1451
1452	/* Fetch the pgm pool shadow descriptor. */
1453	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & X86_PDPE_PG_MASK);
1454	AssertReturn(pShwPde, VERR_PGM_POOL_GET_PAGE_FAILED);
1455
1456	PX86PDPAE pPDDst = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPde);
1457	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1458	PX86PDEPAE pPdeDst = &pPDDst->a[iPDDst];
1459
1460	# else /* PGM_SHW_TYPE == PGM_TYPE_AMD64 */
1461	/* PML4 */
1462	/const unsigned iPml4 = (GCPtrPage >> X86_PML4_SHIFT) & X86_PML4_MASK;/
1463	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
1464	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1465	PX86PDPAE pPDDst;
1466	PX86PDPT pPdptDst;
1467	PX86PML4E pPml4eDst;
1468	rc = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, &pPml4eDst, &pPdptDst, &pPDDst);
1469	if (rc != VINF_SUCCESS)
1470	{
1471	AssertMsg(rc == VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT \|\| rc == VERR_PAGE_MAP_LEVEL4_NOT_PRESENT, ("Unexpected rc=%Rrc\n", rc));
1472	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,InvalidatePageSkipped));
1473	PGM_INVL_PG(pVCpu, GCPtrPage);
1474	return VINF_SUCCESS;
1475	}
1476	PX86PDEPAE pPdeDst = &pPDDst->a[iPDDst];
1477	Assert(pPDDst);
1478	Assert(pPdptDst->a[iPdpt].u & X86_PDPE_P);
1479
1480	/* Fetch the pgm pool shadow descriptor. */
1481	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & SHW_PDPE_PG_MASK);
1482	Assert(pShwPde);
1483
1484	# endif /* PGM_SHW_TYPE == PGM_TYPE_AMD64 */
1485
1486	const SHWPDE PdeDst = *pPdeDst;
1487	if (!(PdeDst.u & X86_PDE_P))
1488	{
1489	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,InvalidatePageSkipped));
1490	PGM_INVL_PG(pVCpu, GCPtrPage);
1491	return VINF_SUCCESS;
1492	}
1493
1494	/*
1495	* Get the guest PD entry and calc big page.
1496	*/
1497	# if PGM_GST_TYPE == PGM_TYPE_32BIT
1498	PGSTPD pPDSrc = pgmGstGet32bitPDPtr(pVCpu);
1499	const unsigned iPDSrc = (uint32_t)GCPtrPage >> GST_PD_SHIFT;
1500	GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
1501	# else /* PGM_GST_TYPE != PGM_TYPE_32BIT */
1502	unsigned iPDSrc = 0;
1503	# if PGM_GST_TYPE == PGM_TYPE_PAE
1504	X86PDPE PdpeSrcIgn;
1505	PX86PDPAE pPDSrc = pgmGstGetPaePDPtr(pVCpu, GCPtrPage, &iPDSrc, &PdpeSrcIgn);
1506	# else /* AMD64 */
1507	PX86PML4E pPml4eSrcIgn;
1508	X86PDPE PdpeSrcIgn;
1509	PX86PDPAE pPDSrc = pgmGstGetLongModePDPtr(pVCpu, GCPtrPage, &pPml4eSrcIgn, &PdpeSrcIgn, &iPDSrc);
1510	# endif
1511	GSTPDE PdeSrc;
1512
1513	if (pPDSrc)
1514	PdeSrc = pPDSrc->a[iPDSrc];
1515	else
1516	PdeSrc.u = 0;
1517	# endif /* PGM_GST_TYPE != PGM_TYPE_32BIT */
1518	const bool fWasBigPage = RT_BOOL(PdeDst.u & PGM_PDFLAGS_BIG_PAGE);
1519	const bool fIsBigPage = (PdeSrc.u & X86_PDE_PS) && GST_IS_PSE_ACTIVE(pVCpu);
1520	if (fWasBigPage != fIsBigPage)
1521	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,InvalidatePageSkipped));
1522
1523	# ifdef IN_RING3
1524	/*
1525	* If a CR3 Sync is pending we may ignore the invalidate page operation
1526	* depending on the kind of sync and if it's a global page or not.
1527	* This doesn't make sense in GC/R0 so we'll skip it entirely there.
1528	*/
1529	# ifdef PGM_SKIP_GLOBAL_PAGEDIRS_ON_NONGLOBAL_FLUSH
1530	if ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3)
1531	\|\| ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL)
1532	&& fIsBigPage
1533	&& (PdeSrc.u & X86_PDE4M_G)
1534	)
1535	)
1536	# else
1537	if (VM_FF_IS_ANY_SET(pVM, VM_FF_PGM_SYNC_CR3 \| VM_FF_PGM_SYNC_CR3_NON_GLOBAL) )
1538	# endif
1539	{
1540	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,InvalidatePageSkipped));
1541	return VINF_SUCCESS;
1542	}
1543	# endif /* IN_RING3 */
1544
1545	/*
1546	* Deal with the Guest PDE.
1547	*/
1548	rc = VINF_SUCCESS;
1549	if (PdeSrc.u & X86_PDE_P)
1550	{
1551	Assert( (PdeSrc.u & X86_PDE_US) == (PdeDst.u & X86_PDE_US)
1552	&& ((PdeSrc.u & X86_PDE_RW) \|\| !(PdeDst.u & X86_PDE_RW) \|\| pVCpu->pgm.s.cNetwareWp0Hacks > 0));
1553	if (!fIsBigPage)
1554	{
1555	/*
1556	* 4KB - page.
1557	*/
1558	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, PdeDst.u & SHW_PDE_PG_MASK);
1559	RTGCPHYS GCPhys = GST_GET_PDE_GCPHYS(PdeSrc);
1560
1561	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
1562	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
1563	GCPhys = PGM_A20_APPLY(pVCpu, GCPhys \| ((iPDDst & 1) * (GUEST_PAGE_SIZE / 2)));
1564	# endif
1565	if (pShwPage->GCPhys == GCPhys)
1566	{
1567	/* Syncing it here isn't 100% safe and it's probably not worth spending time syncing it. */
1568	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
1569
1570	PGSTPT pPTSrc;
1571	rc = PGM_GCPHYS_2_PTR_V2(pVM, pVCpu, GST_GET_PDE_GCPHYS(PdeSrc), &pPTSrc);
1572	if (RT_SUCCESS(rc))
1573	{
1574	const unsigned iPTSrc = (GCPtrPage >> GST_PT_SHIFT) & GST_PT_MASK;
1575	GSTPTE PteSrc = pPTSrc->a[iPTSrc];
1576	const unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
1577	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, PteSrc, pShwPage, iPTDst);
1578	Log2(("SyncPage: 4K %RGv PteSrc:{P=%d RW=%d U=%d raw=%08llx} PteDst=%08llx %s\n",
1579	GCPtrPage, PteSrc.u & X86_PTE_P,
1580	(PteSrc.u & PdeSrc.u & X86_PTE_RW),
1581	(PteSrc.u & PdeSrc.u & X86_PTE_US),
1582	(uint64_t)PteSrc.u,
1583	SHW_PTE_LOG64(pPTDst->a[iPTDst]),
1584	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : ""));
1585	}
1586	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,InvalidatePage4KBPages));
1587	PGM_INVL_PG(pVCpu, GCPtrPage);
1588	}
1589	else
1590	{
1591	/*
1592	* The page table address changed.
1593	*/
1594	LogFlow(("InvalidatePage: Out-of-sync at %RGp PdeSrc=%RX64 PdeDst=%RX64 ShwGCPhys=%RGp iPDDst=%#x\n",
1595	GCPtrPage, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u, pShwPage->GCPhys, iPDDst));
1596	pgmPoolFree(pVM, PdeDst.u & SHW_PDE_PG_MASK, pShwPde->idx, iPDDst);
1597	SHW_PDE_ATOMIC_SET(*pPdeDst, 0);
1598	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,InvalidatePagePDOutOfSync));
1599	PGM_INVL_VCPU_TLBS(pVCpu);
1600	}
1601	}
1602	else
1603	{
1604	/*
1605	* 2/4MB - page.
1606	*/
1607	/* Before freeing the page, check if anything really changed. */
1608	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, PdeDst.u & SHW_PDE_PG_MASK);
1609	RTGCPHYS GCPhys = GST_GET_BIG_PDE_GCPHYS(pVM, PdeSrc);
1610	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
1611	/* Select the right PDE as we're emulating a 4MB page directory with two 2 MB shadow PDEs.*/
1612	GCPhys = PGM_A20_APPLY(pVCpu, GCPhys \| (GCPtrPage & (1 << X86_PD_PAE_SHIFT)));
1613	# endif
1614	if ( pShwPage->GCPhys == GCPhys
1615	&& pShwPage->enmKind == BTH_PGMPOOLKIND_PT_FOR_BIG)
1616	{
1617	/* ASSUMES a the given bits are identical for 4M and normal PDEs */
1618	/** @todo This test is wrong as it cannot check the G bit!
1619	* FIXME */
1620	if ( (PdeSrc.u & (X86_PDE_P \| X86_PDE_RW \| X86_PDE_US))
1621	== (PdeDst.u & (X86_PDE_P \| X86_PDE_RW \| X86_PDE_US))
1622	&& ( (PdeSrc.u & X86_PDE4M_D) /** @todo rainy day: What about read-only 4M pages? not very common, but still... */
1623	\|\| (PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY)))
1624	{
1625	LogFlow(("Skipping flush for big page containing %RGv (PD=%X .u=%RX64)-> nothing has changed!\n", GCPtrPage, iPDSrc, PdeSrc.u));
1626	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,InvalidatePage4MBPagesSkip));
1627	return VINF_SUCCESS;
1628	}
1629	}
1630
1631	/*
1632	* Ok, the page table is present and it's been changed in the guest.
1633	* If we're in host context, we'll just mark it as not present taking the lazy approach.
1634	* We could do this for some flushes in GC too, but we need an algorithm for
1635	* deciding which 4MB pages containing code likely to be executed very soon.
1636	*/
1637	LogFlow(("InvalidatePage: Out-of-sync PD at %RGp PdeSrc=%RX64 PdeDst=%RX64\n",
1638	GCPtrPage, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
1639	pgmPoolFree(pVM, PdeDst.u & SHW_PDE_PG_MASK, pShwPde->idx, iPDDst);
1640	SHW_PDE_ATOMIC_SET(*pPdeDst, 0);
1641	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,InvalidatePage4MBPages));
1642	PGM_INVL_BIG_PG(pVCpu, GCPtrPage);
1643	}
1644	}
1645	else
1646	{
1647	/*
1648	* Page directory is not present, mark shadow PDE not present.
1649	*/
1650	pgmPoolFree(pVM, PdeDst.u & SHW_PDE_PG_MASK, pShwPde->idx, iPDDst);
1651	SHW_PDE_ATOMIC_SET(*pPdeDst, 0);
1652	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,InvalidatePagePDNPs));
1653	PGM_INVL_PG(pVCpu, GCPtrPage);
1654	}
1655	return rc;
1656
1657	#else /* guest real and protected mode, nested + ept, none. */
1658	/* There's no such thing as InvalidatePage when paging is disabled, so just ignore. */
1659	NOREF(pVCpu); NOREF(GCPtrPage);
1660	return VINF_SUCCESS;
1661	#endif
1662	}
1663
1664	#if PGM_SHW_TYPE != PGM_TYPE_NONE
1665
1666	/**
1667	* Update the tracking of shadowed pages.
1668	*
1669	* @param pVCpu The cross context virtual CPU structure.
1670	* @param pShwPage The shadow page.
1671	* @param HCPhys The physical page we is being dereferenced.
1672	* @param iPte Shadow PTE index
1673	* @param GCPhysPage Guest physical address (only valid if pShwPage->fDirty is set)
1674	*/
1675	DECLINLINE(void) PGM_BTH_NAME(SyncPageWorkerTrackDeref)(PVMCPUCC pVCpu, PPGMPOOLPAGE pShwPage, RTHCPHYS HCPhys, uint16_t iPte,
1676	RTGCPHYS GCPhysPage)
1677	{
1678	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1679
1680	# if defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT) \
1681	&& PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) \
1682	&& (PGM_GST_TYPE == PGM_TYPE_PAE \|\| PGM_GST_TYPE == PGM_TYPE_AMD64 \|\| PGM_SHW_TYPE == PGM_TYPE_PAE /* pae/32bit combo */)
1683
1684	/* Use the hint we retrieved from the cached guest PT. */
1685	if (pShwPage->fDirty)
1686	{
1687	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1688
1689	Assert(pShwPage->cPresent);
1690	Assert(pPool->cPresent);
1691	pShwPage->cPresent--;
1692	pPool->cPresent--;
1693
1694	PPGMPAGE pPhysPage = pgmPhysGetPage(pVM, GCPhysPage);
1695	AssertRelease(pPhysPage);
1696	pgmTrackDerefGCPhys(pPool, pShwPage, pPhysPage, iPte);
1697	return;
1698	}
1699	# else
1700	NOREF(GCPhysPage);
1701	# endif
1702
1703	/** @todo If this turns out to be a bottle neck (very likely) two things can be done:
1704	* 1. have a medium sized HCPhys -> GCPhys TLB (hash?)
1705	* 2. write protect all shadowed pages. I.e. implement caching.
1706	*
1707	* 2023-08-24 bird: If we allow the ZeroPg to enter the shadow page tables,
1708	* this becomes a common occurence and we screw up. A better to the above would
1709	* be to have a parallel table that records the guest physical addresses of the
1710	* pages mapped by the shadow page table... For nested page tables,
1711	* we can easily correleate a table entry to a page entry, so it won't be
1712	* needed for those.
1713	*/
1714	# if PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) \|\| !PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1715	/*
1716	* For non-paged guest tables, EPT and nested tables we can figure out the
1717	* physical page corresponding to the entry and dereference it.
1718	* (This ASSUMES that shadow PTs won't be used ever be used out of place.)
1719	*/
1720	if ( pShwPage->enmKind == PGMPOOLKIND_EPT_PT_FOR_PHYS
1721	\|\| pShwPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PHYS
1722	\|\| pShwPage->enmKind == PGMPOOLKIND_32BIT_PT_FOR_PHYS)
1723	{
1724	RTGCPHYS GCPhysNestedEntry = pShwPage->GCPhys + ((uint32_t)iPte << X86_PAGE_SHIFT);
1725	if (!pShwPage->fA20Enabled)
1726	GCPhysNestedEntry &= ~(uint64_t)RT_BIT_64(20);
1727	PPGMPAGE const pPhysPage = pgmPhysGetPage(pVM, GCPhysNestedEntry);
1728	AssertRelease(pPhysPage);
1729	pgmTrackDerefGCPhys(pVM->pgm.s.CTX_SUFF(pPool), pShwPage, pPhysPage, iPte);
1730	}
1731	else
1732	AssertMsgFailed(("enmKind=%d GCPhys=%RGp\n", pShwPage->enmKind, pShwPage->GCPhys));
1733	# endif
1734
1735	/** @todo duplicated in the 2nd half of pgmPoolTracDerefGCPhysHint */
1736
1737	/*
1738	* Find the guest address.
1739	*/
1740	STAM_PROFILE_START(&pVM->pgm.s.Stats.StatTrackDeref, a);
1741	LogFlow(("SyncPageWorkerTrackDeref(%d,%d): Damn HCPhys=%RHp pShwPage->idx=%#x!!!\n",
1742	PGM_SHW_TYPE, PGM_GST_TYPE, HCPhys, pShwPage->idx));
1743	uint32_t const idRamRangeMax = RT_MIN(pVM->pgm.s.idRamRangeMax, RT_ELEMENTS(pVM->pgm.s.apRamRanges) - 1U);
1744	Assert(pVM->pgm.s.apRamRanges[0] == NULL);
1745	for (uint32_t idx = 1; idx <= idRamRangeMax; idx++)
1746	{
1747	PPGMRAMRANGE const pRam = pVM->CTX_EXPR(pgm, pgmr0, pgm).s.apRamRanges[idx];
1748	AssertContinue(pRam);
1749	unsigned iPage = pRam->cb >> GUEST_PAGE_SHIFT;
1750	while (iPage-- > 0)
1751	{
1752	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
1753	{
1754	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1755
1756	Assert(pShwPage->cPresent);
1757	Assert(pPool->cPresent);
1758	pShwPage->cPresent--;
1759	pPool->cPresent--;
1760
1761	pgmTrackDerefGCPhys(pPool, pShwPage, &pRam->aPages[iPage], iPte);
1762	STAM_PROFILE_STOP(&pVM->pgm.s.Stats.StatTrackDeref, a);
1763	return;
1764	}
1765	}
1766	}
1767
1768	for (;;)
1769	AssertReleaseMsgFailed(("HCPhys=%RHp wasn't found!\n", HCPhys));
1770	}
1771
1772
1773	/**
1774	* Update the tracking of shadowed pages.
1775	*
1776	* @param pVCpu The cross context virtual CPU structure.
1777	* @param pShwPage The shadow page.
1778	* @param u16 The top 16-bit of the pPage->HCPhys.
1779	* @param pPage Pointer to the guest page. this will be modified.
1780	* @param iPTDst The index into the shadow table.
1781	*/
1782	DECLINLINE(void) PGM_BTH_NAME(SyncPageWorkerTrackAddref)(PVMCPUCC pVCpu, PPGMPOOLPAGE pShwPage, uint16_t u16,
1783	PPGMPAGE pPage, const unsigned iPTDst)
1784	{
1785	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1786
1787	/*
1788	* Just deal with the simple first time here.
1789	*/
1790	if (!u16)
1791	{
1792	STAM_COUNTER_INC(&pVM->pgm.s.Stats.StatTrackVirgin);
1793	u16 = PGMPOOL_TD_MAKE(1, pShwPage->idx);
1794	/* Save the page table index. */
1795	PGM_PAGE_SET_PTE_INDEX(pVM, pPage, iPTDst);
1796	}
1797	else
1798	u16 = pgmPoolTrackPhysExtAddref(pVM, pPage, u16, pShwPage->idx, iPTDst);
1799
1800	/* write back */
1801	Log2(("SyncPageWorkerTrackAddRef: u16=%#x->%#x iPTDst=%#x pPage=%p\n", u16, PGM_PAGE_GET_TRACKING(pPage), iPTDst, pPage));
1802	PGM_PAGE_SET_TRACKING(pVM, pPage, u16);
1803
1804	/* update statistics. */
1805	pVM->pgm.s.CTX_SUFF(pPool)->cPresent++;
1806	pShwPage->cPresent++;
1807	if (pShwPage->iFirstPresent > iPTDst)
1808	pShwPage->iFirstPresent = iPTDst;
1809	}
1810
1811
1812	/**
1813	* Modifies a shadow PTE to account for access handlers.
1814	*
1815	* @param pVM The cross context VM structure.
1816	* @param pVCpu The cross context virtual CPU structure.
1817	* @param pPage The page in question.
1818	* @param GCPhysPage The guest-physical address of the page.
1819	* @param fPteSrc The shadowed flags of the source PTE. Must include the
1820	* A (accessed) bit so it can be emulated correctly.
1821	* @param pPteDst The shadow PTE (output). This is temporary storage and
1822	* does not need to be set atomically.
1823	*/
1824	DECLINLINE(void) PGM_BTH_NAME(SyncHandlerPte)(PVMCC pVM, PVMCPUCC pVCpu, PCPGMPAGE pPage, RTGCPHYS GCPhysPage, uint64_t fPteSrc,
1825	PSHWPTE pPteDst)
1826	{
1827	RT_NOREF_PV(pVM); RT_NOREF_PV(fPteSrc); RT_NOREF_PV(pVCpu); RT_NOREF_PV(GCPhysPage);
1828
1829	/** @todo r=bird: Are we actually handling dirty and access bits for pages with access handlers correctly? No.
1830	* Update: \#PF should deal with this before or after calling the handlers. It has all the info to do the job efficiently. */
1831	if (!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage))
1832	{
1833	LogFlow(("SyncHandlerPte: monitored page (%R[pgmpage]) -> mark read-only\n", pPage));
1834	# if PGM_SHW_TYPE == PGM_TYPE_EPT
1835	pPteDst->u = PGM_PAGE_GET_HCPHYS(pPage) \| EPT_E_READ \| EPT_E_EXECUTE \| EPT_E_MEMTYPE_WB \| EPT_E_IGNORE_PAT;
1836	# else
1837	if (fPteSrc & X86_PTE_A)
1838	{
1839	SHW_PTE_SET(*pPteDst, fPteSrc \| PGM_PAGE_GET_HCPHYS(pPage));
1840	SHW_PTE_SET_RO(*pPteDst);
1841	}
1842	else
1843	SHW_PTE_SET(*pPteDst, 0);
1844	# endif
1845	}
1846	# ifdef PGM_WITH_MMIO_OPTIMIZATIONS
1847	# if PGM_SHW_TYPE == PGM_TYPE_EPT \|\| PGM_SHW_TYPE == PGM_TYPE_PAE \|\| PGM_SHW_TYPE == PGM_TYPE_AMD64
1848	else if ( PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage)
1849	&& ( BTH_IS_NP_ACTIVE(pVM)
1850	\|\| (fPteSrc & (X86_PTE_RW \| X86_PTE_US)) == X86_PTE_RW) /** @todo Remove X86_PTE_US here and pGstWalk->Core.fEffectiveUS before the sync page test. */
1851	# if PGM_SHW_TYPE == PGM_TYPE_AMD64
1852	&& pVM->pgm.s.fLessThan52PhysicalAddressBits
1853	# endif
1854	)
1855	{
1856	LogFlow(("SyncHandlerPte: MMIO page -> invalid \n"));
1857	# if PGM_SHW_TYPE == PGM_TYPE_EPT
1858	/* 25.2.3.1: Reserved physical address bit -> EPT Misconfiguration (exit 49) */
1859	pPteDst->u = pVM->pgm.s.HCPhysInvMmioPg
1860	/* 25.2.3.1: bits 2:0 = 010b -> EPT Misconfiguration (exit 49) */
1861	\| EPT_E_WRITE
1862	/* 25.2.3.1: leaf && 2:0 != 0 && u3Emt in {2, 3, 7} -> EPT Misconfiguration */
1863	\| EPT_E_MEMTYPE_INVALID_3;
1864	# else
1865	/* Set high page frame bits that MBZ (bankers on PAE, CPU dependent on AMD64). */
1866	SHW_PTE_SET(*pPteDst, pVM->pgm.s.HCPhysInvMmioPg \| X86_PTE_PAE_MBZ_MASK_NO_NX \| X86_PTE_P);
1867	# endif
1868	}
1869	# endif
1870	# endif /* PGM_WITH_MMIO_OPTIMIZATIONS */
1871	else
1872	{
1873	LogFlow(("SyncHandlerPte: monitored page (%R[pgmpage]) -> mark not present\n", pPage));
1874	SHW_PTE_SET(*pPteDst, 0);
1875	}
1876	/** @todo count these kinds of entries. */
1877	}
1878
1879
1880	/**
1881	* Creates a 4K shadow page for a guest page.
1882	*
1883	* For 4M pages the caller must convert the PDE4M to a PTE, this includes adjusting the
1884	* physical address. The PdeSrc argument only the flags are used. No page
1885	* structured will be mapped in this function.
1886	*
1887	* @param pVCpu The cross context virtual CPU structure.
1888	* @param pPteDst Destination page table entry.
1889	* @param PdeSrc Source page directory entry (i.e. Guest OS page directory entry).
1890	* Can safely assume that only the flags are being used.
1891	* @param PteSrc Source page table entry (i.e. Guest OS page table entry).
1892	* @param pShwPage Pointer to the shadow page.
1893	* @param iPTDst The index into the shadow table.
1894	*
1895	* @remark Not used for 2/4MB pages!
1896	*/
1897	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) \|\| defined(DOXYGEN_RUNNING)
1898	static void PGM_BTH_NAME(SyncPageWorker)(PVMCPUCC pVCpu, PSHWPTE pPteDst, GSTPDE PdeSrc, GSTPTE PteSrc,
1899	PPGMPOOLPAGE pShwPage, unsigned iPTDst)
1900	# else
1901	static void PGM_BTH_NAME(SyncPageWorker)(PVMCPUCC pVCpu, PSHWPTE pPteDst, RTGCPHYS GCPhysPage,
1902	PPGMPOOLPAGE pShwPage, unsigned iPTDst)
1903	# endif
1904	{
1905	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1906	RTGCPHYS GCPhysOldPage = NIL_RTGCPHYS;
1907
1908	# if defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT) \
1909	&& PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) \
1910	&& (PGM_GST_TYPE == PGM_TYPE_PAE \|\| PGM_GST_TYPE == PGM_TYPE_AMD64 \|\| PGM_SHW_TYPE == PGM_TYPE_PAE /* pae/32bit combo */)
1911
1912	if (pShwPage->fDirty)
1913	{
1914	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1915	PGSTPT pGstPT;
1916
1917	/* Note that iPTDst can be used to index the guest PT even in the pae/32bit combo as we copy only half the table; see pgmPoolAddDirtyPage. */
1918	pGstPT = (PGSTPT)&pPool->aDirtyPages[pShwPage->idxDirtyEntry].aPage[0];
1919	GCPhysOldPage = GST_GET_PTE_GCPHYS(pGstPT->a[iPTDst]);
1920	pGstPT->a[iPTDst].u = PteSrc.u;
1921	}
1922	# else
1923	Assert(!pShwPage->fDirty);
1924	# endif
1925
1926	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1927	if ( (PteSrc.u & X86_PTE_P)
1928	&& GST_IS_PTE_VALID(pVCpu, PteSrc))
1929	# endif
1930	{
1931	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1932	RTGCPHYS GCPhysPage = GST_GET_PTE_GCPHYS(PteSrc);
1933	# endif
1934	PGM_A20_ASSERT_MASKED(pVCpu, GCPhysPage);
1935
1936	/*
1937	* Find the ram range.
1938	*/
1939	PPGMPAGE pPage;
1940	int rc = pgmPhysGetPageEx(pVM, GCPhysPage, &pPage);
1941	if (RT_SUCCESS(rc))
1942	{
1943	/* Ignore ballooned pages.
1944	Don't return errors or use a fatal assert here as part of a
1945	shadow sync range might included ballooned pages. */
1946	if (PGM_PAGE_IS_BALLOONED(pPage))
1947	{
1948	Assert(!SHW_PTE_IS_P(pPteDst)); /* @todo user tracking needs updating if this triggers. */
1949	return;
1950	}
1951
1952	# ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
1953	/* Make the page writable if necessary. */
1954	if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM
1955	&& ( PGM_PAGE_IS_ZERO(pPage)
1956	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1957	\|\| ( (PteSrc.u & X86_PTE_RW)
1958	# else
1959	\|\| ( 1
1960	# endif
1961	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED
1962	# ifdef VBOX_WITH_REAL_WRITE_MONITORED_PAGES
1963	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_WRITE_MONITORED
1964	# endif
1965	# ifdef VBOX_WITH_PAGE_SHARING
1966	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_SHARED
1967	# endif
1968	)
1969	)
1970	)
1971	{
1972	rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhysPage);
1973	AssertRC(rc);
1974	}
1975	# endif
1976
1977	/*
1978	* Make page table entry.
1979	*/
1980	SHWPTE PteDst;
1981	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1982	uint64_t fGstShwPteFlags = GST_GET_PTE_SHW_FLAGS(pVCpu, PteSrc);
1983	# else
1984	uint64_t fGstShwPteFlags = X86_PTE_P \| X86_PTE_RW \| X86_PTE_US \| X86_PTE_A \| X86_PTE_D;
1985	# endif
1986	if (!PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage) \|\| PGM_PAGE_IS_HNDL_PHYS_NOT_IN_HM(pPage))
1987	{
1988	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1989	/*
1990	* If the page or page directory entry is not marked accessed,
1991	* we mark the page not present.
1992	*/
1993	if (!(PteSrc.u & X86_PTE_A) \|\| !(PdeSrc.u & X86_PDE_A))
1994	{
1995	LogFlow(("SyncPageWorker: page and or page directory not accessed -> mark not present\n"));
1996	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,AccessedPage));
1997	SHW_PTE_SET(PteDst, 0);
1998	}
1999	/*
2000	* If the page is not flagged as dirty and is writable, then make it read-only, so we can set the dirty bit
2001	* when the page is modified.
2002	*/
2003	else if (!(PteSrc.u & X86_PTE_D) && (PdeSrc.u & PteSrc.u & X86_PTE_RW))
2004	{
2005	AssertCompile(X86_PTE_RW == X86_PDE_RW);
2006	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,DirtyPage));
2007	SHW_PTE_SET(PteDst,
2008	fGstShwPteFlags
2009	\| PGM_PAGE_GET_HCPHYS(pPage)
2010	\| PGM_PTFLAGS_TRACK_DIRTY);
2011	SHW_PTE_SET_RO(PteDst);
2012	}
2013	else
2014	# endif
2015	{
2016	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,DirtyPageSkipped));
2017	# if PGM_SHW_TYPE == PGM_TYPE_EPT
2018	PteDst.u = PGM_PAGE_GET_HCPHYS(pPage)
2019	\| EPT_E_READ \| EPT_E_WRITE \| EPT_E_EXECUTE \| EPT_E_MEMTYPE_WB \| EPT_E_IGNORE_PAT;
2020	# else
2021	SHW_PTE_SET(PteDst, fGstShwPteFlags \| PGM_PAGE_GET_HCPHYS(pPage));
2022	# endif
2023	}
2024
2025	/*
2026	* Make sure only allocated pages are mapped writable.
2027	*/
2028	if ( SHW_PTE_IS_P_RW(PteDst)
2029	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
2030	{
2031	# ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
2032	/* Still applies to shared pages. */
2033	Assert(!PGM_PAGE_IS_ZERO(pPage));
2034	# endif
2035	SHW_PTE_SET_RO(PteDst); /** @todo this isn't quite working yet. Why, isn't it? */
2036	Log3(("SyncPageWorker: write-protecting %RGp pPage=%R[pgmpage]at iPTDst=%d\n", GCPhysPage, pPage, iPTDst));
2037	}
2038	}
2039	else
2040	PGM_BTH_NAME(SyncHandlerPte)(pVM, pVCpu, pPage, GCPhysPage, fGstShwPteFlags, &PteDst);
2041
2042	/*
2043	* Keep user track up to date.
2044	*/
2045	if (SHW_PTE_IS_P(PteDst))
2046	{
2047	if (!SHW_PTE_IS_P(*pPteDst))
2048	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPTDst);
2049	else if (SHW_PTE_GET_HCPHYS(*pPteDst) != SHW_PTE_GET_HCPHYS(PteDst))
2050	{
2051	Log2(("SyncPageWorker: deref! pPteDst=%RX64 PteDst=%RX64\n", SHW_PTE_LOG64(pPteDst), SHW_PTE_LOG64(PteDst)));
2052	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pVCpu, pShwPage, SHW_PTE_GET_HCPHYS(*pPteDst), iPTDst, GCPhysOldPage);
2053	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPTDst);
2054	}
2055	}
2056	else if (SHW_PTE_IS_P(*pPteDst))
2057	{
2058	Log2(("SyncPageWorker: deref! pPteDst=%RX64\n", SHW_PTE_LOG64(pPteDst)));
2059	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pVCpu, pShwPage, SHW_PTE_GET_HCPHYS(*pPteDst), iPTDst, GCPhysOldPage);
2060	}
2061
2062	/*
2063	* Update statistics and commit the entry.
2064	*/
2065	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
2066	if (!(PteSrc.u & X86_PTE_G))
2067	pShwPage->fSeenNonGlobal = true;
2068	# endif
2069	SHW_PTE_ATOMIC_SET2(*pPteDst, PteDst);
2070	return;
2071	}
2072
2073	/** @todo count these three different kinds. */
2074	Log2(("SyncPageWorker: invalid address in Pte\n"));
2075	}
2076	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
2077	else if (!(PteSrc.u & X86_PTE_P))
2078	Log2(("SyncPageWorker: page not present in Pte\n"));
2079	else
2080	Log2(("SyncPageWorker: invalid Pte\n"));
2081	# endif
2082
2083	/*
2084	* The page is not present or the PTE is bad. Replace the shadow PTE by
2085	* an empty entry, making sure to keep the user tracking up to date.
2086	*/
2087	if (SHW_PTE_IS_P(*pPteDst))
2088	{
2089	Log2(("SyncPageWorker: deref! pPteDst=%RX64\n", SHW_PTE_LOG64(pPteDst)));
2090	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pVCpu, pShwPage, SHW_PTE_GET_HCPHYS(*pPteDst), iPTDst, GCPhysOldPage);
2091	}
2092	SHW_PTE_ATOMIC_SET(*pPteDst, 0);
2093	}
2094
2095
2096	/**
2097	* Syncs a guest OS page.
2098	*
2099	* There are no conflicts at this point, neither is there any need for
2100	* page table allocations.
2101	*
2102	* When called in PAE or AMD64 guest mode, the guest PDPE shall be valid.
2103	* When called in AMD64 guest mode, the guest PML4E shall be valid.
2104	*
2105	* @returns VBox status code.
2106	* @returns VINF_PGM_SYNCPAGE_MODIFIED_PDE if it modifies the PDE in any way.
2107	* @param pVCpu The cross context virtual CPU structure.
2108	* @param PdeSrc Page directory entry of the guest.
2109	* @param GCPtrPage Guest context page address.
2110	* @param cPages Number of pages to sync (PGM_SYNC_N_PAGES) (default=1).
2111	* @param uErr Fault error (X86_TRAP_PF_*).
2112	*/
2113	static int PGM_BTH_NAME(SyncPage)(PVMCPUCC pVCpu, GSTPDE PdeSrc, RTGCPTR GCPtrPage, unsigned cPages, unsigned uErr)
2114	{
2115	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2116	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
2117	LogFlow(("SyncPage: GCPtrPage=%RGv cPages=%u uErr=%#x\n", GCPtrPage, cPages, uErr));
2118	RT_NOREF_PV(uErr); RT_NOREF_PV(cPages); RT_NOREF_PV(GCPtrPage);
2119
2120	PGM_LOCK_ASSERT_OWNER(pVM);
2121
2122	# if ( PGM_GST_TYPE == PGM_TYPE_32BIT \
2123	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
2124	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64) \
2125	&& !PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE)
2126
2127	/*
2128	* Assert preconditions.
2129	*/
2130	Assert(PdeSrc.u & X86_PDE_P);
2131	Assert(cPages);
2132	# if 0 /* rarely useful; leave for debugging. */
2133	STAM_COUNTER_INC(&pVCpu->pgm.s.StatSyncPagePD[(GCPtrPage >> GST_PD_SHIFT) & GST_PD_MASK]);
2134	# endif
2135
2136	/*
2137	* Get the shadow PDE, find the shadow page table in the pool.
2138	*/
2139	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
2140	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
2141	PX86PDE pPdeDst = pgmShwGet32BitPDEPtr(pVCpu, GCPtrPage);
2142	AssertReturn(pPdeDst, VERR_INTERNAL_ERROR_3);
2143
2144	/* Fetch the pgm pool shadow descriptor. */
2145	PPGMPOOLPAGE pShwPde = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
2146	Assert(pShwPde);
2147
2148	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
2149	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
2150	PPGMPOOLPAGE pShwPde = NULL;
2151	PX86PDPAE pPDDst;
2152
2153	/* Fetch the pgm pool shadow descriptor. */
2154	int rc2 = pgmShwGetPaePoolPagePD(pVCpu, GCPtrPage, &pShwPde);
2155	AssertRCSuccessReturn(rc2, rc2);
2156	Assert(pShwPde);
2157
2158	pPDDst = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPde);
2159	PX86PDEPAE pPdeDst = &pPDDst->a[iPDDst];
2160
2161	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
2162	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
2163	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
2164	PX86PDPAE pPDDst = NULL; /* initialized to shut up gcc */
2165	PX86PDPT pPdptDst = NULL; /* initialized to shut up gcc */
2166
2167	int rc2 = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, NULL, &pPdptDst, &pPDDst);
2168	AssertRCSuccessReturn(rc2, rc2);
2169	Assert(pPDDst && pPdptDst);
2170	PX86PDEPAE pPdeDst = &pPDDst->a[iPDDst];
2171	# endif
2172	SHWPDE PdeDst = *pPdeDst;
2173
2174	/*
2175	* - In the guest SMP case we could have blocked while another VCPU reused
2176	* this page table.
2177	* - With W7-64 we may also take this path when the A bit is cleared on
2178	* higher level tables (PDPE/PML4E). The guest does not invalidate the
2179	* relevant TLB entries. If we're write monitoring any page mapped by
2180	* the modified entry, we may end up here with a "stale" TLB entry.
2181	*/
2182	if (!(PdeDst.u & X86_PDE_P))
2183	{
2184	Log(("CPU%u: SyncPage: Pde at %RGv changed behind our back? (pPdeDst=%p/%RX64) uErr=%#x\n", pVCpu->idCpu, GCPtrPage, pPdeDst, (uint64_t)PdeDst.u, (uint32_t)uErr));
2185	AssertMsg(pVM->cCpus > 1 \|\| (uErr & (X86_TRAP_PF_P \| X86_TRAP_PF_RW)) == (X86_TRAP_PF_P \| X86_TRAP_PF_RW),
2186	("Unexpected missing PDE p=%p/%RX64 uErr=%#x\n", pPdeDst, (uint64_t)PdeDst.u, (uint32_t)uErr));
2187	if (uErr & X86_TRAP_PF_P)
2188	PGM_INVL_PG(pVCpu, GCPtrPage);
2189	return VINF_SUCCESS; /* force the instruction to be executed again. */
2190	}
2191
2192	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, PdeDst.u & SHW_PDE_PG_MASK);
2193	Assert(pShwPage);
2194
2195	# if PGM_GST_TYPE == PGM_TYPE_AMD64
2196	/* Fetch the pgm pool shadow descriptor. */
2197	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & X86_PDPE_PG_MASK);
2198	Assert(pShwPde);
2199	# endif
2200
2201	/*
2202	* Check that the page is present and that the shadow PDE isn't out of sync.
2203	*/
2204	const bool fBigPage = (PdeSrc.u & X86_PDE_PS) && GST_IS_PSE_ACTIVE(pVCpu);
2205	const bool fPdeValid = !fBigPage ? GST_IS_PDE_VALID(pVCpu, PdeSrc) : GST_IS_BIG_PDE_VALID(pVCpu, PdeSrc);
2206	RTGCPHYS GCPhys;
2207	if (!fBigPage)
2208	{
2209	GCPhys = GST_GET_PDE_GCPHYS(PdeSrc);
2210	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
2211	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
2212	GCPhys = PGM_A20_APPLY(pVCpu, GCPhys \| ((iPDDst & 1) * (GUEST_PAGE_SIZE / 2)));
2213	# endif
2214	}
2215	else
2216	{
2217	GCPhys = GST_GET_BIG_PDE_GCPHYS(pVM, PdeSrc);
2218	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
2219	/* Select the right PDE as we're emulating a 4MB page directory with two 2 MB shadow PDEs.*/
2220	GCPhys = PGM_A20_APPLY(pVCpu, GCPhys \| (GCPtrPage & (1 << X86_PD_PAE_SHIFT)));
2221	# endif
2222	}
2223	/** @todo This doesn't check the G bit of 2/4MB pages. FIXME */
2224	if ( fPdeValid
2225	&& pShwPage->GCPhys == GCPhys
2226	&& (PdeSrc.u & X86_PDE_P)
2227	&& (PdeSrc.u & X86_PDE_US) == (PdeDst.u & X86_PDE_US)
2228	&& ((PdeSrc.u & X86_PDE_RW) == (PdeDst.u & X86_PDE_RW) \|\| !(PdeDst.u & X86_PDE_RW))
2229	# if PGM_WITH_NX(PGM_GST_TYPE, PGM_SHW_TYPE)
2230	&& ((PdeSrc.u & X86_PDE_PAE_NX) == (PdeDst.u & X86_PDE_PAE_NX) \|\| !GST_IS_NX_ACTIVE(pVCpu))
2231	# endif
2232	)
2233	{
2234	/*
2235	* Check that the PDE is marked accessed already.
2236	* Since we set the accessed bit before getting here on a #PF, this
2237	* check is only meant for dealing with non-#PF'ing paths.
2238	*/
2239	if (PdeSrc.u & X86_PDE_A)
2240	{
2241	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
2242	if (!fBigPage)
2243	{
2244	/*
2245	* 4KB Page - Map the guest page table.
2246	*/
2247	PGSTPT pPTSrc;
2248	int rc = PGM_GCPHYS_2_PTR_V2(pVM, pVCpu, GST_GET_PDE_GCPHYS(PdeSrc), &pPTSrc);
2249	if (RT_SUCCESS(rc))
2250	{
2251	# ifdef PGM_SYNC_N_PAGES
2252	Assert(cPages == 1 \|\| !(uErr & X86_TRAP_PF_P));
2253	if ( cPages > 1
2254	&& !(uErr & X86_TRAP_PF_P)
2255	&& !VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY))
2256	{
2257	/*
2258	* This code path is currently only taken when the caller is PGMTrap0eHandler
2259	* for non-present pages!
2260	*
2261	* We're setting PGM_SYNC_NR_PAGES pages around the faulting page to sync it and
2262	* deal with locality.
2263	*/
2264	unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2265	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
2266	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
2267	const unsigned offPTSrc = ((GCPtrPage >> SHW_PD_SHIFT) & 1) * 512;
2268	# else
2269	const unsigned offPTSrc = 0;
2270	# endif
2271	const unsigned iPTDstEnd = RT_MIN(iPTDst + PGM_SYNC_NR_PAGES / 2, RT_ELEMENTS(pPTDst->a));
2272	if (iPTDst < PGM_SYNC_NR_PAGES / 2)
2273	iPTDst = 0;
2274	else
2275	iPTDst -= PGM_SYNC_NR_PAGES / 2;
2276
2277	for (; iPTDst < iPTDstEnd; iPTDst++)
2278	{
2279	const PGSTPTE pPteSrc = &pPTSrc->a[offPTSrc + iPTDst];
2280
2281	if ( (pPteSrc->u & X86_PTE_P)
2282	&& !SHW_PTE_IS_P(pPTDst->a[iPTDst]))
2283	{
2284	RTGCPTR GCPtrCurPage = (GCPtrPage & ~(RTGCPTR)(GST_PT_MASK << GST_PT_SHIFT))
2285	\| ((offPTSrc + iPTDst) << GUEST_PAGE_SHIFT);
2286	NOREF(GCPtrCurPage);
2287	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, *pPteSrc, pShwPage, iPTDst);
2288	Log2(("SyncPage: 4K+ %RGv PteSrc:{P=%d RW=%d U=%d raw=%08llx} PteDst=%08llx%s\n",
2289	GCPtrCurPage, pPteSrc->u & X86_PTE_P,
2290	!!(pPteSrc->u & PdeSrc.u & X86_PTE_RW),
2291	!!(pPteSrc->u & PdeSrc.u & X86_PTE_US),
2292	(uint64_t)pPteSrc->u,
2293	SHW_PTE_LOG64(pPTDst->a[iPTDst]),
2294	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : ""));
2295	}
2296	}
2297	}
2298	else
2299	# endif /* PGM_SYNC_N_PAGES */
2300	{
2301	const unsigned iPTSrc = (GCPtrPage >> GST_PT_SHIFT) & GST_PT_MASK;
2302	GSTPTE PteSrc = pPTSrc->a[iPTSrc];
2303	const unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2304	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, PteSrc, pShwPage, iPTDst);
2305	Log2(("SyncPage: 4K %RGv PteSrc:{P=%d RW=%d U=%d raw=%08llx} PteDst=%08llx %s\n",
2306	GCPtrPage, PteSrc.u & X86_PTE_P,
2307	!!(PteSrc.u & PdeSrc.u & X86_PTE_RW),
2308	!!(PteSrc.u & PdeSrc.u & X86_PTE_US),
2309	(uint64_t)PteSrc.u,
2310	SHW_PTE_LOG64(pPTDst->a[iPTDst]),
2311	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : ""));
2312	}
2313	}
2314	else /* MMIO or invalid page: emulated in #PF handler. */
2315	{
2316	LogFlow(("PGM_GCPHYS_2_PTR_V2 %RGp failed with %Rrc\n", GCPhys, rc));
2317	Assert(!SHW_PTE_IS_P(pPTDst->a[(GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK]));
2318	}
2319	}
2320	else
2321	{
2322	/*
2323	* 4/2MB page - lazy syncing shadow 4K pages.
2324	* (There are many causes of getting here, it's no longer only CSAM.)
2325	*/
2326	/* Calculate the GC physical address of this 4KB shadow page. */
2327	GCPhys = PGM_A20_APPLY(pVCpu, GST_GET_BIG_PDE_GCPHYS(pVM, PdeSrc) \| (GCPtrPage & GST_BIG_PAGE_OFFSET_MASK));
2328	/* Find ram range. */
2329	PPGMPAGE pPage;
2330	int rc = pgmPhysGetPageEx(pVM, GCPhys, &pPage);
2331	if (RT_SUCCESS(rc))
2332	{
2333	AssertFatalMsg(!PGM_PAGE_IS_BALLOONED(pPage), ("Unexpected ballooned page at %RGp\n", GCPhys));
2334
2335	# ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
2336	/* Try to make the page writable if necessary. */
2337	if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM
2338	&& ( PGM_PAGE_IS_ZERO(pPage)
2339	\|\| ( (PdeSrc.u & X86_PDE_RW)
2340	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED
2341	# ifdef VBOX_WITH_REAL_WRITE_MONITORED_PAGES
2342	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_WRITE_MONITORED
2343	# endif
2344	# ifdef VBOX_WITH_PAGE_SHARING
2345	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_SHARED
2346	# endif
2347	)
2348	)
2349	)
2350	{
2351	rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhys);
2352	AssertRC(rc);
2353	}
2354	# endif
2355
2356	/*
2357	* Make shadow PTE entry.
2358	*/
2359	SHWPTE PteDst;
2360	if (!PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage) \|\| PGM_PAGE_IS_HNDL_PHYS_NOT_IN_HM(pPage))
2361	SHW_PTE_SET(PteDst, GST_GET_BIG_PDE_SHW_FLAGS_4_PTE(pVCpu, PdeSrc) \| PGM_PAGE_GET_HCPHYS(pPage));
2362	else
2363	PGM_BTH_NAME(SyncHandlerPte)(pVM, pVCpu, pPage, GCPhys, GST_GET_BIG_PDE_SHW_FLAGS_4_PTE(pVCpu, PdeSrc), &PteDst);
2364
2365	const unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2366	if ( SHW_PTE_IS_P(PteDst)
2367	&& !SHW_PTE_IS_P(pPTDst->a[iPTDst]))
2368	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPTDst);
2369
2370	/* Make sure only allocated pages are mapped writable. */
2371	if ( SHW_PTE_IS_P_RW(PteDst)
2372	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
2373	{
2374	# ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
2375	/* Still applies to shared pages. */
2376	Assert(!PGM_PAGE_IS_ZERO(pPage));
2377	# endif
2378	SHW_PTE_SET_RO(PteDst); /** @todo this isn't quite working yet... */
2379	Log3(("SyncPage: write-protecting %RGp pPage=%R[pgmpage] at %RGv\n", GCPhys, pPage, GCPtrPage));
2380	}
2381
2382	SHW_PTE_ATOMIC_SET2(pPTDst->a[iPTDst], PteDst);
2383
2384	/*
2385	* If the page is not flagged as dirty and is writable, then make it read-only
2386	* at PD level, so we can set the dirty bit when the page is modified.
2387	*
2388	* ASSUMES that page access handlers are implemented on page table entry level.
2389	* Thus we will first catch the dirty access and set PDE.D and restart. If
2390	* there is an access handler, we'll trap again and let it work on the problem.
2391	*/
2392	/** @todo r=bird: figure out why we need this here, SyncPT should've taken care of this already.
2393	* As for invlpg, it simply frees the whole shadow PT.
2394	* ...It's possibly because the guest clears it and the guest doesn't really tell us... */
2395	if ((PdeSrc.u & (X86_PDE4M_D \| X86_PDE_RW)) == X86_PDE_RW)
2396	{
2397	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,DirtyPageBig));
2398	PdeDst.u \|= PGM_PDFLAGS_TRACK_DIRTY;
2399	PdeDst.u &= ~(SHWUINT)X86_PDE_RW;
2400	}
2401	else
2402	{
2403	PdeDst.u &= ~(SHWUINT)(PGM_PDFLAGS_TRACK_DIRTY \| X86_PDE_RW);
2404	PdeDst.u \|= PdeSrc.u & X86_PDE_RW;
2405	}
2406	SHW_PDE_ATOMIC_SET2(*pPdeDst, PdeDst);
2407	Log2(("SyncPage: BIG %RGv PdeSrc:{P=%d RW=%d U=%d raw=%08llx} GCPhys=%RGp%s\n",
2408	GCPtrPage, PdeSrc.u & X86_PDE_P, !!(PdeSrc.u & X86_PDE_RW), !!(PdeSrc.u & X86_PDE_US),
2409	(uint64_t)PdeSrc.u, GCPhys, PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY ? " Track-Dirty" : ""));
2410	}
2411	else
2412	{
2413	LogFlow(("pgmPhysGetPageEx %RGp (big) failed with %Rrc\n", GCPhys, rc));
2414	/** @todo must wipe the shadow page table entry in this
2415	* case. */
2416	}
2417	}
2418	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
2419	return VINF_SUCCESS;
2420	}
2421
2422	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPagePDNAs));
2423	}
2424	else if (fPdeValid)
2425	{
2426	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPagePDOutOfSync));
2427	Log2(("SyncPage: Out-Of-Sync PDE at %RGp PdeSrc=%RX64 PdeDst=%RX64 (GCPhys %RGp vs %RGp)\n",
2428	GCPtrPage, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u, pShwPage->GCPhys, GCPhys));
2429	}
2430	else
2431	{
2432	/// @todo STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPagePDOutOfSyncAndInvalid));
2433	Log2(("SyncPage: Bad PDE at %RGp PdeSrc=%RX64 PdeDst=%RX64 (GCPhys %RGp vs %RGp)\n",
2434	GCPtrPage, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u, pShwPage->GCPhys, GCPhys));
2435	}
2436
2437	/*
2438	* Mark the PDE not present. Restart the instruction and let #PF call SyncPT.
2439	* Yea, I'm lazy.
2440	*/
2441	pgmPoolFreeByPage(pPool, pShwPage, pShwPde->idx, iPDDst);
2442	SHW_PDE_ATOMIC_SET(*pPdeDst, 0);
2443
2444	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
2445	PGM_INVL_VCPU_TLBS(pVCpu);
2446	return VINF_PGM_SYNCPAGE_MODIFIED_PDE;
2447
2448
2449	# elif (PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
2450	&& !PGM_TYPE_IS_NESTED(PGM_SHW_TYPE) \
2451	&& (PGM_SHW_TYPE != PGM_TYPE_EPT \|\| PGM_GST_TYPE == PGM_TYPE_PROT)
2452	NOREF(PdeSrc);
2453
2454	# ifdef PGM_SYNC_N_PAGES
2455	/*
2456	* Get the shadow PDE, find the shadow page table in the pool.
2457	*/
2458	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
2459	X86PDE PdeDst = pgmShwGet32BitPDE(pVCpu, GCPtrPage);
2460
2461	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
2462	X86PDEPAE PdeDst = pgmShwGetPaePDE(pVCpu, GCPtrPage);
2463
2464	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
2465	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
2466	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64; NOREF(iPdpt);
2467	PX86PDPAE pPDDst = NULL; /* initialized to shut up gcc */
2468	X86PDEPAE PdeDst;
2469	PX86PDPT pPdptDst = NULL; /* initialized to shut up gcc */
2470
2471	int rc = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, NULL, &pPdptDst, &pPDDst);
2472	AssertRCSuccessReturn(rc, rc);
2473	Assert(pPDDst && pPdptDst);
2474	PdeDst = pPDDst->a[iPDDst];
2475
2476	# elif PGM_SHW_TYPE == PGM_TYPE_EPT
2477	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
2478	PEPTPD pPDDst;
2479	EPTPDE PdeDst;
2480
2481	int rc = pgmShwGetEPTPDPtr(pVCpu, GCPtrPage, NULL, &pPDDst);
2482	if (rc != VINF_SUCCESS)
2483	{
2484	AssertRC(rc);
2485	return rc;
2486	}
2487	Assert(pPDDst);
2488	PdeDst = pPDDst->a[iPDDst];
2489	# endif
2490	/* In the guest SMP case we could have blocked while another VCPU reused this page table. */
2491	if (!SHW_PDE_IS_P(PdeDst))
2492	{
2493	AssertMsg(pVM->cCpus > 1, ("Unexpected missing PDE %RX64\n", (uint64_t)PdeDst.u));
2494	Log(("CPU%d: SyncPage: Pde at %RGv changed behind our back!\n", pVCpu->idCpu, GCPtrPage));
2495	return VINF_SUCCESS; /* force the instruction to be executed again. */
2496	}
2497
2498	/* Can happen in the guest SMP case; other VCPU activated this PDE while we were blocking to handle the page fault. */
2499	if (SHW_PDE_IS_BIG(PdeDst))
2500	{
2501	Assert(pVM->pgm.s.fNestedPaging);
2502	Log(("CPU%d: SyncPage: Pde (big:%RX64) at %RGv changed behind our back!\n", pVCpu->idCpu, PdeDst.u, GCPtrPage));
2503	return VINF_SUCCESS;
2504	}
2505
2506	/* Mask away the page offset. */
2507	GCPtrPage &= ~((RTGCPTR)0xfff);
2508
2509	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, PdeDst.u & SHW_PDE_PG_MASK);
2510	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
2511
2512	Assert(cPages == 1 \|\| !(uErr & X86_TRAP_PF_P));
2513	if ( cPages > 1
2514	&& !(uErr & X86_TRAP_PF_P)
2515	&& !VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY))
2516	{
2517	/*
2518	* This code path is currently only taken when the caller is PGMTrap0eHandler
2519	* for non-present pages!
2520	*
2521	* We're setting PGM_SYNC_NR_PAGES pages around the faulting page to sync it and
2522	* deal with locality.
2523	*/
2524	unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2525	const unsigned iPTDstEnd = RT_MIN(iPTDst + PGM_SYNC_NR_PAGES / 2, RT_ELEMENTS(pPTDst->a));
2526	if (iPTDst < PGM_SYNC_NR_PAGES / 2)
2527	iPTDst = 0;
2528	else
2529	iPTDst -= PGM_SYNC_NR_PAGES / 2;
2530	for (; iPTDst < iPTDstEnd; iPTDst++)
2531	{
2532	if (!SHW_PTE_IS_P(pPTDst->a[iPTDst]))
2533	{
2534	RTGCPTR GCPtrCurPage = PGM_A20_APPLY(pVCpu, (GCPtrPage & ~(RTGCPTR)(SHW_PT_MASK << SHW_PT_SHIFT))
2535	\| (iPTDst << GUEST_PAGE_SHIFT));
2536
2537	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], GCPtrCurPage, pShwPage, iPTDst);
2538	Log2(("SyncPage: 4K+ %RGv PteSrc:{P=1 RW=1 U=1} PteDst=%08llx%s\n",
2539	GCPtrCurPage,
2540	SHW_PTE_LOG64(pPTDst->a[iPTDst]),
2541	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : ""));
2542
2543	if (RT_UNLIKELY(VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)))
2544	break;
2545	}
2546	else
2547	Log4(("%RGv iPTDst=%x pPTDst->a[iPTDst] %RX64\n",
2548	(GCPtrPage & ~(RTGCPTR)(SHW_PT_MASK << SHW_PT_SHIFT)) \| (iPTDst << GUEST_PAGE_SHIFT), iPTDst, SHW_PTE_LOG64(pPTDst->a[iPTDst]) ));
2549	}
2550	}
2551	else
2552	# endif /* PGM_SYNC_N_PAGES */
2553	{
2554	const unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2555	RTGCPTR GCPtrCurPage = PGM_A20_APPLY(pVCpu, (GCPtrPage & ~(RTGCPTR)(SHW_PT_MASK << SHW_PT_SHIFT))
2556	\| (iPTDst << GUEST_PAGE_SHIFT));
2557
2558	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], GCPtrCurPage, pShwPage, iPTDst);
2559
2560	Log2(("SyncPage: 4K %RGv PteSrc:{P=1 RW=1 U=1}PteDst=%08llx%s\n",
2561	GCPtrPage,
2562	SHW_PTE_LOG64(pPTDst->a[iPTDst]),
2563	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : ""));
2564	}
2565	return VINF_SUCCESS;
2566
2567	# else
2568	NOREF(PdeSrc);
2569	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_GST_TYPE, PGM_SHW_TYPE));
2570	return VERR_PGM_NOT_USED_IN_MODE;
2571	# endif
2572	}
2573
2574	#endif /* PGM_SHW_TYPE != PGM_TYPE_NONE */
2575
2576	#if !defined(IN_RING3) && defined(VBOX_WITH_NESTED_HWVIRT_VMX_EPT) && PGM_SHW_TYPE == PGM_TYPE_EPT
2577
2578	/**
2579	* Sync a shadow page for a nested-guest page.
2580	*
2581	* @param pVCpu The cross context virtual CPU structure.
2582	* @param pPte The shadow page table entry.
2583	* @param GCPhysPage The guest-physical address of the page.
2584	* @param pShwPage The shadow page of the page table.
2585	* @param iPte The index of the page table entry.
2586	* @param pGstSlatPte The guest SLAT page table entry.
2587	*
2588	* @note Not to be used for 2/4MB pages!
2589	*/
2590	static void PGM_BTH_NAME(NestedSyncPageWorker)(PVMCPUCC pVCpu, PSHWPTE pPte, RTGCPHYS GCPhysPage, PPGMPOOLPAGE pShwPage,
2591	unsigned iPte, SLATPTE GstSlatPte)
2592	{
2593	PGM_A20_ASSERT_MASKED(pVCpu, GCPhysPage);
2594	Assert(PGMPOOL_PAGE_IS_NESTED(pShwPage));
2595	Assert(!pShwPage->fDirty);
2596	Assert(pVCpu->pgm.s.enmGuestSlatMode == PGMSLAT_EPT);
2597	AssertMsg(!(GstSlatPte.u & EPT_E_LEAF), ("Large page unexpected: %RX64\n", GstSlatPte.u));
2598	AssertMsg((GstSlatPte.u & EPT_PTE_PG_MASK) == GCPhysPage,
2599	("PTE address mismatch. GCPhysPage=%RGp Pte=%RX64\n", GCPhysPage, GstSlatPte.u & EPT_PTE_PG_MASK));
2600
2601	/*
2602	* Find the ram range.
2603	*/
2604	PPGMPAGE pPage;
2605	int rc = pgmPhysGetPageEx(pVCpu->CTX_SUFF(pVM), GCPhysPage, &pPage);
2606	if (RT_SUCCESS(rc))
2607	{ /* likely */ }
2608	else
2609	{
2610	/*
2611	* This is a RAM hole/invalid/reserved address (not MMIO).
2612	* Nested Microsoft Hyper-V maps addresses like 0xf0220000 as RW WB memory.
2613	* Shadow a not-present page similar to MMIO, see @bugref{10318#c7}.
2614	*/
2615	Assert(rc == VERR_PGM_INVALID_GC_PHYSICAL_ADDRESS);
2616	if (SHW_PTE_IS_P(*pPte))
2617	{
2618	Log2(("NestedSyncPageWorker: deref! pPte=%RX64\n", SHW_PTE_LOG64(pPte)));
2619	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pVCpu, pShwPage, SHW_PTE_GET_HCPHYS(*pPte), iPte, NIL_RTGCPHYS);
2620	}
2621	Log7Func(("RAM hole/reserved %RGp -> ShwPte=0\n", GCPhysPage));
2622	SHW_PTE_ATOMIC_SET(*pPte, 0);
2623	return;
2624	}
2625
2626	Assert(!PGM_PAGE_IS_BALLOONED(pPage));
2627
2628	/*
2629	* Make page table entry.
2630	*/
2631	SHWPTE Pte;
2632	uint64_t const fGstShwPteFlags = (GstSlatPte.u & pVCpu->pgm.s.fGstEptShadowedPteMask)
2633	\| EPT_E_MEMTYPE_WB \| EPT_E_IGNORE_PAT;
2634	if (!PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage) \|\| PGM_PAGE_IS_HNDL_PHYS_NOT_IN_HM(pPage))
2635	{
2636	# ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
2637	/* If it's the zero page or write to an unallocated page, allocate it to make it writable. */
2638	if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM
2639	&& ( PGM_PAGE_IS_ZERO(pPage)
2640	\|\| ( (GstSlatPte.u & EPT_E_WRITE)
2641	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED
2642	# ifdef VBOX_WITH_REAL_WRITE_MONITORED_PAGES
2643	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_WRITE_MONITORED
2644	# endif
2645	# ifdef VBOX_WITH_PAGE_SHARING
2646	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_SHARED
2647	# endif
2648	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_BALLOONED
2649	)
2650	)
2651	)
2652	{
2653	rc = pgmPhysPageMakeWritable(pVCpu->CTX_SUFF(pVM), pPage, GCPhysPage);
2654	AssertRC(rc);
2655	Log7Func(("made writable (%R[pgmpage]) at %RGp\n", pPage, GCPhysPage));
2656	}
2657	# endif
2658	/** @todo access bit. */
2659	Pte.u = PGM_PAGE_GET_HCPHYS(pPage) \| fGstShwPteFlags;
2660	Log7Func(("regular page (%R[pgmpage]) at %RGp -> %RX64\n", pPage, GCPhysPage, Pte.u));
2661
2662	/* Make sure only allocated pages are mapped writable. */
2663	if ( (fGstShwPteFlags & EPT_E_WRITE)
2664	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
2665	{
2666	Pte.u &= ~EPT_E_WRITE;
2667	Log7Func(("write-protecting page (%R[pgmpage]) at %RGp -> %RX64\n", pPage, GCPhysPage, Pte.u));
2668	}
2669	}
2670	else if (!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage))
2671	{
2672	/** @todo access bit. */
2673	Pte.u = PGM_PAGE_GET_HCPHYS(pPage) \| (fGstShwPteFlags & ~EPT_E_WRITE);
2674	Log7Func(("monitored page (%R[pgmpage]) at %RGp -> %RX64\n", pPage, GCPhysPage, Pte.u));
2675	}
2676	else
2677	{
2678	/** @todo Do MMIO optimizations here too? */
2679	Log7Func(("mmio/all page (%R[pgmpage]) at %RGp -> 0\n", pPage, GCPhysPage));
2680	Pte.u = 0;
2681	}
2682
2683	/* Make sure only allocated pages are mapped writable. */
2684	Assert(!SHW_PTE_IS_P_RW(Pte) \|\| PGM_PAGE_IS_ALLOCATED(pPage));
2685
2686	/*
2687	* Keep user track up to date.
2688	*/
2689	if (SHW_PTE_IS_P(Pte))
2690	{
2691	if (!SHW_PTE_IS_P(*pPte))
2692	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPte);
2693	else if (SHW_PTE_GET_HCPHYS(*pPte) != SHW_PTE_GET_HCPHYS(Pte))
2694	{
2695	Log2(("NestedSyncPageWorker: deref! pPte=%RX64 Pte=%RX64\n", SHW_PTE_LOG64(pPte), SHW_PTE_LOG64(Pte)));
2696	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pVCpu, pShwPage, SHW_PTE_GET_HCPHYS(*pPte), iPte, NIL_RTGCPHYS);
2697	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPte);
2698	}
2699	}
2700	else if (SHW_PTE_IS_P(*pPte))
2701	{
2702	Log2(("NestedSyncPageWorker: deref! pPte=%RX64\n", SHW_PTE_LOG64(pPte)));
2703	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pVCpu, pShwPage, SHW_PTE_GET_HCPHYS(*pPte), iPte, NIL_RTGCPHYS);
2704	}
2705
2706	/*
2707	* Commit the entry.
2708	*/
2709	SHW_PTE_ATOMIC_SET2(*pPte, Pte);
2710	return;
2711	}
2712
2713
2714	/**
2715	* Syncs a nested-guest page.
2716	*
2717	* There are no conflicts at this point, neither is there any need for
2718	* page table allocations.
2719	*
2720	* @returns VBox status code.
2721	* @param pVCpu The cross context virtual CPU structure.
2722	* @param GCPhysNestedPage The nested-guest physical address of the page being
2723	* synced.
2724	* @param GCPhysPage The guest-physical address of the page being synced.
2725	* @param cPages Number of pages to sync (PGM_SYNC_N_PAGES) (default=1).
2726	* @param uErr The page fault error (X86_TRAP_PF_XXX).
2727	* @param pGstWalkAll The guest page table walk result.
2728	*/
2729	static int PGM_BTH_NAME(NestedSyncPage)(PVMCPUCC pVCpu, RTGCPHYS GCPhysNestedPage, RTGCPHYS GCPhysPage, unsigned cPages,
2730	uint32_t uErr, PPGMPTWALKGST pGstWalkAll)
2731	{
2732	PGM_A20_ASSERT_MASKED(pVCpu, GCPhysPage);
2733	Assert(!(GCPhysNestedPage & GUEST_PAGE_OFFSET_MASK));
2734	Assert(!(GCPhysPage & GUEST_PAGE_OFFSET_MASK));
2735
2736	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2737	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
2738	Log7Func(("GCPhysNestedPage=%RGv GCPhysPage=%RGp cPages=%u uErr=%#x\n", GCPhysNestedPage, GCPhysPage, cPages, uErr));
2739	RT_NOREF_PV(uErr); RT_NOREF_PV(cPages);
2740
2741	PGM_LOCK_ASSERT_OWNER(pVM);
2742
2743	/*
2744	* Get the shadow PDE, find the shadow page table in the pool.
2745	*/
2746	unsigned const iPde = ((GCPhysNestedPage >> EPT_PD_SHIFT) & EPT_PD_MASK);
2747	PEPTPD pPd;
2748	int rc = pgmShwGetNestedEPTPDPtr(pVCpu, GCPhysNestedPage, NULL, &pPd, pGstWalkAll);
2749	if (RT_SUCCESS(rc))
2750	{ /* likely */ }
2751	else
2752	{
2753	Log(("Failed to fetch EPT PD for %RGp (%RGp) rc=%Rrc\n", GCPhysNestedPage, GCPhysPage, rc));
2754	return rc;
2755	}
2756	Assert(pPd);
2757	EPTPDE Pde = pPd->a[iPde];
2758
2759	/* In the guest SMP case we could have blocked while another VCPU reused this page table. */
2760	if (!SHW_PDE_IS_P(Pde))
2761	{
2762	AssertMsg(pVM->cCpus > 1, ("Unexpected missing PDE %RX64\n", (uint64_t)Pde.u));
2763	Log7Func(("CPU%d: SyncPage: Pde at %RGp changed behind our back!\n", pVCpu->idCpu, GCPhysNestedPage));
2764	return VINF_SUCCESS; /* force the instruction to be executed again. */
2765	}
2766
2767	/* Can happen in the guest SMP case; other VCPU activated this PDE while we were blocking to handle the page fault. */
2768	if (SHW_PDE_IS_BIG(Pde))
2769	{
2770	Log7Func(("CPU%d: SyncPage: %RGp changed behind our back!\n", pVCpu->idCpu, GCPhysNestedPage));
2771	return VINF_SUCCESS;
2772	}
2773
2774	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, Pde.u & EPT_PDE_PG_MASK);
2775	PEPTPT pPt = (PEPTPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
2776
2777	/*
2778	* If we've shadowed a guest EPT PDE that maps a 2M page using a 4K table,
2779	* then sync the 4K sub-page in the 2M range.
2780	*/
2781	if (pGstWalkAll->u.Ept.Pde.u & EPT_E_LEAF)
2782	{
2783	Assert(!SHW_PDE_IS_BIG(Pde));
2784
2785	Assert(pGstWalkAll->u.Ept.Pte.u == 0);
2786	Assert((Pde.u & EPT_PRESENT_MASK) == (pGstWalkAll->u.Ept.Pde.u & EPT_PRESENT_MASK));
2787	Assert(pShwPage->GCPhys == (pGstWalkAll->u.Ept.Pde.u & EPT_PDE2M_PG_MASK));
2788
2789	#if defined(VBOX_STRICT) && defined(DEBUG_ramshankar)
2790	PPGMPAGE pPage;
2791	rc = pgmPhysGetPageEx(pVM, GCPhysPage, &pPage); AssertRC(rc);
2792	Assert(PGM_PAGE_GET_PDE_TYPE(pPage) != PGM_PAGE_PDE_TYPE_PDE);
2793	Assert(pShwPage->enmKind == PGMPOOLKIND_EPT_PT_FOR_EPT_2MB);
2794	#endif
2795	uint64_t const fGstShwPteFlags = (pGstWalkAll->u.Ept.Pde.u & pVCpu->pgm.s.fGstEptShadowedBigPdeMask & ~EPT_E_LEAF)
2796	\| EPT_E_MEMTYPE_WB \| EPT_E_IGNORE_PAT;
2797	SLATPTE GstSlatPte;
2798	GstSlatPte.u = GCPhysPage \| fGstShwPteFlags;
2799
2800	unsigned const iPte = (GCPhysNestedPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2801	PGM_BTH_NAME(NestedSyncPageWorker)(pVCpu, &pPt->a[iPte], GCPhysPage, pShwPage, iPte, GstSlatPte);
2802	Log7Func(("4K: GCPhysPage=%RGp iPte=%u ShwPte=%08llx\n", GCPhysPage, iPte, SHW_PTE_LOG64(pPt->a[iPte])));
2803	return VINF_SUCCESS;
2804	}
2805
2806	Assert(cPages == 1 \|\| !(uErr & X86_TRAP_PF_P));
2807	# ifdef PGM_SYNC_N_PAGES
2808	if ( cPages > 1
2809	&& !(uErr & X86_TRAP_PF_P)
2810	&& !VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY))
2811	{
2812	/*
2813	* This code path is currently only taken for non-present pages!
2814	*
2815	* We're setting PGM_SYNC_NR_PAGES pages around the faulting page to sync it and
2816	* deal with locality.
2817	*/
2818	unsigned iPte = (GCPhysNestedPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2819	unsigned const iPteEnd = RT_MIN(iPte + PGM_SYNC_NR_PAGES / 2, RT_ELEMENTS(pPt->a));
2820	if (iPte < PGM_SYNC_NR_PAGES / 2)
2821	iPte = 0;
2822	else
2823	iPte -= PGM_SYNC_NR_PAGES / 2;
2824	for (; iPte < iPteEnd; iPte++)
2825	{
2826	if (!SHW_PTE_IS_P(pPt->a[iPte]))
2827	{
2828	PGMPTWALKGST GstWalkPt;
2829	PGMPTWALK WalkPt;
2830	GCPhysNestedPage &= ~(SHW_PT_MASK << SHW_PT_SHIFT);
2831	GCPhysNestedPage \|= (iPte << GUEST_PAGE_SHIFT);
2832	rc = pgmGstSlatWalk(pVCpu, GCPhysNestedPage, false /fIsLinearAddrValid/, 0 /GCPtrNested/, &WalkPt,
2833	&GstWalkPt);
2834	if (RT_SUCCESS(rc))
2835	PGM_BTH_NAME(NestedSyncPageWorker)(pVCpu, &pPt->a[iPte], WalkPt.GCPhys, pShwPage, iPte, GstWalkPt.u.Ept.Pte);
2836	else
2837	{
2838	/*
2839	* This could be MMIO pages reserved by the nested-hypevisor or genuinely not-present pages.
2840	* Ensure the shadow tables entry is not-present.
2841	*/
2842	/** @todo Potential room for optimization (explained in NestedSyncPT). */
2843	AssertMsg(!pPt->a[iPte].u, ("%RX64\n", pPt->a[iPte].u));
2844	}
2845	Log7Func(("Many: %RGp iPte=%u ShwPte=%RX64\n", GCPhysNestedPage, iPte, SHW_PTE_LOG64(pPt->a[iPte])));
2846	if (RT_UNLIKELY(VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)))
2847	break;
2848	}
2849	else
2850	{
2851	# ifdef VBOX_STRICT
2852	/* Paranoia - Verify address of the page is what it should be. */
2853	PGMPTWALKGST GstWalkPt;
2854	PGMPTWALK WalkPt;
2855	GCPhysNestedPage &= ~(SHW_PT_MASK << SHW_PT_SHIFT);
2856	GCPhysNestedPage \|= (iPte << GUEST_PAGE_SHIFT);
2857	rc = pgmGstSlatWalk(pVCpu, GCPhysNestedPage, false /fIsLinearAddrValid/, 0 /GCPtrNested/, &WalkPt, &GstWalkPt);
2858	AssertRC(rc);
2859	PPGMPAGE pPage;
2860	rc = pgmPhysGetPageEx(pVM, WalkPt.GCPhys, &pPage);
2861	AssertRC(rc);
2862	AssertMsg(PGM_PAGE_GET_HCPHYS(pPage) == SHW_PTE_GET_HCPHYS(pPt->a[iPte]),
2863	("PGM page and shadow PTE address conflict. GCPhysNestedPage=%RGp GCPhysPage=%RGp HCPhys=%RHp Shw=%RHp\n",
2864	GCPhysNestedPage, WalkPt.GCPhys, PGM_PAGE_GET_HCPHYS(pPage), SHW_PTE_GET_HCPHYS(pPt->a[iPte])));
2865	# endif
2866	Log7Func(("Many3: %RGp iPte=%u ShwPte=%RX64\n", GCPhysNestedPage, iPte, SHW_PTE_LOG64(pPt->a[iPte])));
2867	}
2868	}
2869	}
2870	else
2871	# endif /* PGM_SYNC_N_PAGES */
2872	{
2873	unsigned const iPte = (GCPhysNestedPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2874	PGM_BTH_NAME(NestedSyncPageWorker)(pVCpu, &pPt->a[iPte], GCPhysPage, pShwPage, iPte, pGstWalkAll->u.Ept.Pte);
2875	Log7Func(("4K: GCPhysPage=%RGp iPte=%u ShwPte=%08llx\n", GCPhysPage, iPte, SHW_PTE_LOG64(pPt->a[iPte])));
2876	}
2877
2878	return VINF_SUCCESS;
2879	}
2880
2881
2882	/**
2883	* Sync a shadow page table for a nested-guest page table.
2884	*
2885	* The shadow page table is not present in the shadow PDE.
2886	*
2887	* Handles mapping conflicts.
2888	*
2889	* A precondition for this method is that the shadow PDE is not present. The
2890	* caller must take the PGM lock before checking this and continue to hold it
2891	* when calling this method.
2892	*
2893	* @returns VBox status code.
2894	* @param pVCpu The cross context virtual CPU structure.
2895	* @param GCPhysNestedPage The nested-guest physical page address of the page
2896	* being synced.
2897	* @param GCPhysPage The guest-physical address of the page being synced.
2898	* @param pGstWalkAll The guest page table walk result.
2899	*/
2900	static int PGM_BTH_NAME(NestedSyncPT)(PVMCPUCC pVCpu, RTGCPHYS GCPhysNestedPage, RTGCPHYS GCPhysPage, PPGMPTWALKGST pGstWalkAll)
2901	{
2902	PGM_A20_ASSERT_MASKED(pVCpu, GCPhysPage);
2903	Assert(!(GCPhysNestedPage & GUEST_PAGE_OFFSET_MASK));
2904	Assert(!(GCPhysPage & GUEST_PAGE_OFFSET_MASK));
2905
2906	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2907	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2908
2909	Log7Func(("GCPhysNestedPage=%RGp GCPhysPage=%RGp\n", GCPhysNestedPage, GCPhysPage));
2910
2911	PGM_LOCK_ASSERT_OWNER(pVM);
2912	STAM_PROFILE_START(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
2913
2914	PEPTPD pPd;
2915	PEPTPDPT pPdpt;
2916	unsigned const iPde = (GCPhysNestedPage >> EPT_PD_SHIFT) & EPT_PD_MASK;
2917	int rc = pgmShwGetNestedEPTPDPtr(pVCpu, GCPhysNestedPage, &pPdpt, &pPd, pGstWalkAll);
2918	if (RT_SUCCESS(rc))
2919	{ /* likely */ }
2920	else
2921	{
2922	STAM_PROFILE_STOP(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
2923	AssertRC(rc);
2924	return rc;
2925	}
2926	Assert(pPd);
2927	PSHWPDE pPde = &pPd->a[iPde];
2928
2929	unsigned const iPdpt = (GCPhysNestedPage >> EPT_PDPT_SHIFT) & EPT_PDPT_MASK;
2930	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdpt->a[iPdpt].u & EPT_PDPTE_PG_MASK);
2931	Assert(pShwPde->enmKind == PGMPOOLKIND_EPT_PD_FOR_EPT_PD);
2932
2933	SHWPDE Pde = *pPde;
2934	Assert(!SHW_PDE_IS_P(Pde)); /* We're only supposed to call SyncPT on PDE!P and conflicts. */
2935
2936	# ifdef PGM_WITH_LARGE_PAGES
2937	Assert(BTH_IS_NP_ACTIVE(pVM));
2938
2939	/*
2940	* Check if the guest is mapping a 2M page.
2941	*/
2942	if (pGstWalkAll->u.Ept.Pde.u & EPT_E_LEAF)
2943	{
2944	PPGMPAGE pPage;
2945	rc = pgmPhysGetPageEx(pVM, GCPhysPage & X86_PDE2M_PAE_PG_MASK, &pPage);
2946	AssertRCReturn(rc, rc);
2947
2948	/* A20 is always enabled in VMX root and non-root operation. */
2949	Assert(PGM_A20_IS_ENABLED(pVCpu));
2950
2951	/*
2952	* Check if we have or can get a 2M backing page here.
2953	*/
2954	RTHCPHYS HCPhys = NIL_RTHCPHYS;
2955	if (PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE)
2956	{
2957	STAM_REL_COUNTER_INC(&pVM->pgm.s.StatLargePageReused);
2958	AssertRelease(PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED);
2959	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
2960	}
2961	else if (PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE_DISABLED)
2962	{
2963	/* Recheck the entire 2 MB range to see if we can use it again as a large page. */
2964	rc = pgmPhysRecheckLargePage(pVM, GCPhysPage, pPage);
2965	if (RT_SUCCESS(rc))
2966	{
2967	Assert(PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED);
2968	Assert(PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE);
2969	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
2970	}
2971	}
2972	else if (PGMIsUsingLargePages(pVM))
2973	{
2974	rc = pgmPhysAllocLargePage(pVM, GCPhysPage);
2975	if (RT_SUCCESS(rc))
2976	{
2977	Assert(PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED);
2978	Assert(PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE);
2979	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
2980	}
2981	}
2982
2983	/*
2984	* If we have a 2M backing page, we can map the guest's 2M page right away.
2985	*/
2986	uint64_t const fGstShwBigPdeFlags = (pGstWalkAll->u.Ept.Pde.u & pVCpu->pgm.s.fGstEptShadowedBigPdeMask)
2987	\| EPT_E_MEMTYPE_WB \| EPT_E_IGNORE_PAT;
2988	if (HCPhys != NIL_RTHCPHYS)
2989	{
2990	Pde.u = HCPhys \| fGstShwBigPdeFlags;
2991	Assert(!(Pde.u & pVCpu->pgm.s.fGstEptMbzBigPdeMask));
2992	Assert(Pde.u & EPT_E_LEAF);
2993	SHW_PDE_ATOMIC_SET2(*pPde, Pde);
2994
2995	/* Add a reference to the first page only. */
2996	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPde, PGM_PAGE_GET_TRACKING(pPage), pPage, iPde);
2997
2998	Assert(PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_WRITE_MONITORED);
2999
3000	STAM_PROFILE_STOP(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
3001	Log7Func(("GstPde=%RGp ShwPde=%RX64 [2M]\n", pGstWalkAll->u.Ept.Pde.u, Pde.u));
3002	return VINF_SUCCESS;
3003	}
3004
3005	/*
3006	* We didn't get a perfect 2M fit. Split the 2M page into 4K pages.
3007	* The page ought not to be marked as a big (2M) page at this point.
3008	*/
3009	Assert(PGM_PAGE_GET_PDE_TYPE(pPage) != PGM_PAGE_PDE_TYPE_PDE);
3010
3011	/* Determine the right kind of large page to avoid incorrect cached entry reuse. */
3012	PGMPOOLACCESS enmAccess;
3013	{
3014	/*
3015	* Mode-based execute control for EPT not supported.
3016	*
3017	* However, Windows 10 with Hyper-V enabled sets the EPT_E_USER_EXECUTE bit but does
3018	* not enable "mode-based execute control for EPT" in the VT-x secondary VM-execution
3019	* controls. The CPU ignores this bit when the control isn't set. Hence, the assertion
3020	* below is commented out.
3021	*/
3022	/* Assert(!(pGstWalkAll->u.Ept.Pde.u & EPT_E_USER_EXECUTE)); */
3023	Assert(!pVCpu->CTX_SUFF(pVM)->cpum.ro.GuestFeatures.fVmxModeBasedExecuteEpt);
3024	bool const fNoExecute = !(pGstWalkAll->u.Ept.Pde.u & EPT_E_EXECUTE);
3025	if (pGstWalkAll->u.Ept.Pde.u & EPT_E_WRITE)
3026	enmAccess = fNoExecute ? PGMPOOLACCESS_SUPERVISOR_RW_NX : PGMPOOLACCESS_SUPERVISOR_RW;
3027	else
3028	enmAccess = fNoExecute ? PGMPOOLACCESS_SUPERVISOR_R_NX : PGMPOOLACCESS_SUPERVISOR_R;
3029	}
3030
3031	/*
3032	* Allocate & map a 4K shadow table to cover the 2M guest page.
3033	*/
3034	PPGMPOOLPAGE pShwPage;
3035	RTGCPHYS const GCPhysPt = pGstWalkAll->u.Ept.Pde.u & EPT_PDE2M_PG_MASK;
3036	rc = pgmPoolAlloc(pVM, GCPhysPt, PGMPOOLKIND_EPT_PT_FOR_EPT_2MB, enmAccess, PGM_A20_IS_ENABLED(pVCpu),
3037	pShwPde->idx, iPde, false /fLockPage/, &pShwPage);
3038	if ( rc == VINF_SUCCESS
3039	\|\| rc == VINF_PGM_CACHED_PAGE)
3040	{ /* likely */ }
3041	else
3042	{
3043	STAM_PROFILE_STOP(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
3044	AssertMsgFailedReturn(("rc=%Rrc\n", rc), RT_FAILURE_NP(rc) ? rc : VERR_IPE_UNEXPECTED_INFO_STATUS);
3045	}
3046
3047	PSHWPT pPt = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
3048	Assert(pPt);
3049	Assert(PGMPOOL_PAGE_IS_NESTED(pShwPage));
3050	if (rc == VINF_SUCCESS)
3051	{
3052	/* The 4K PTEs shall inherit the flags of the 2M PDE page sans the leaf bit. */
3053	uint64_t const fGstShwPteFlags = fGstShwBigPdeFlags & ~EPT_E_LEAF;
3054
3055	/* Sync each 4K pages in the 2M range. */
3056	for (unsigned iPte = 0; iPte < RT_ELEMENTS(pPt->a); iPte++)
3057	{
3058	RTGCPHYS const GCPhysSubPage = GCPhysPt \| (iPte << GUEST_PAGE_SHIFT);
3059	SLATPTE GstSlatPte;
3060	GstSlatPte.u = GCPhysSubPage \| fGstShwPteFlags;
3061	Assert(!(GstSlatPte.u & pVCpu->pgm.s.fGstEptMbzPteMask));
3062	PGM_BTH_NAME(NestedSyncPageWorker)(pVCpu, &pPt->a[iPte], GCPhysSubPage, pShwPage, iPte, GstSlatPte);
3063	Log7Func(("GstPte=%RGp ShwPte=%RX64 iPte=%u [2M->4K]\n", pGstWalkAll->u.Ept.Pte, pPt->a[iPte].u, iPte));
3064	if (RT_UNLIKELY(VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)))
3065	break;
3066	}
3067	}
3068	else
3069	{
3070	Assert(rc == VINF_PGM_CACHED_PAGE);
3071	# if defined(VBOX_STRICT) && defined(DEBUG_ramshankar)
3072	/* Paranoia - Verify address of each of the subpages are what they should be. */
3073	RTGCPHYS GCPhysSubPage = GCPhysPt;
3074	for (unsigned iPte = 0; iPte < RT_ELEMENTS(pPt->a); iPte++, GCPhysSubPage += GUEST_PAGE_SIZE)
3075	{
3076	PPGMPAGE pSubPage;
3077	rc = pgmPhysGetPageEx(pVM, GCPhysSubPage, &pSubPage);
3078	AssertRC(rc);
3079	AssertMsg( PGM_PAGE_GET_HCPHYS(pSubPage) == SHW_PTE_GET_HCPHYS(pPt->a[iPte])
3080	\|\| !SHW_PTE_IS_P(pPt->a[iPte]),
3081	("PGM 2M page and shadow PTE conflict. GCPhysSubPage=%RGp Page=%RHp Shw=%RHp\n",
3082	GCPhysSubPage, PGM_PAGE_GET_HCPHYS(pSubPage), SHW_PTE_GET_HCPHYS(pPt->a[iPte])));
3083	}
3084	# endif
3085	rc = VINF_SUCCESS; /* Cached entry; assume it's still fully valid. */
3086	}
3087
3088	/* Save the new PDE. */
3089	uint64_t const fShwPdeFlags = pGstWalkAll->u.Ept.Pde.u & pVCpu->pgm.s.fGstEptShadowedPdeMask;
3090	Pde.u = pShwPage->Core.Key \| fShwPdeFlags;
3091	Assert(!(Pde.u & EPT_E_LEAF));
3092	Assert(!(Pde.u & pVCpu->pgm.s.fGstEptMbzPdeMask));
3093	SHW_PDE_ATOMIC_SET2(*pPde, Pde);
3094	STAM_PROFILE_STOP(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
3095	Log7Func(("GstPde=%RGp ShwPde=%RX64 iPde=%u\n", pGstWalkAll->u.Ept.Pde.u, pPde->u, iPde));
3096	return rc;
3097	}
3098	# endif /* PGM_WITH_LARGE_PAGES */
3099
3100	/*
3101	* Allocate & map the shadow page table.
3102	*/
3103	PSHWPT pPt;
3104	PPGMPOOLPAGE pShwPage;
3105
3106	RTGCPHYS const GCPhysPt = pGstWalkAll->u.Ept.Pde.u & EPT_PDE_PG_MASK;
3107	rc = pgmPoolAlloc(pVM, GCPhysPt, PGMPOOLKIND_EPT_PT_FOR_EPT_PT, PGMPOOLACCESS_DONTCARE,
3108	PGM_A20_IS_ENABLED(pVCpu), pShwPde->idx, iPde, false /fLockPage/, &pShwPage);
3109	if ( rc == VINF_SUCCESS
3110	\|\| rc == VINF_PGM_CACHED_PAGE)
3111	{ /* likely */ }
3112	else
3113	{
3114	STAM_PROFILE_STOP(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
3115	AssertMsgFailedReturn(("rc=%Rrc\n", rc), RT_FAILURE_NP(rc) ? rc : VERR_IPE_UNEXPECTED_INFO_STATUS);
3116	}
3117
3118	pPt = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
3119	Assert(pPt);
3120	Assert(PGMPOOL_PAGE_IS_NESTED(pShwPage));
3121
3122	if (rc == VINF_SUCCESS)
3123	{
3124	/* Sync the page we've already translated through SLAT. */
3125	const unsigned iPte = (GCPhysNestedPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
3126	PGM_BTH_NAME(NestedSyncPageWorker)(pVCpu, &pPt->a[iPte], GCPhysPage, pShwPage, iPte, pGstWalkAll->u.Ept.Pte);
3127	Log7Func(("GstPte=%RGp ShwPte=%RX64 iPte=%u\n", pGstWalkAll->u.Ept.Pte.u, pPt->a[iPte].u, iPte));
3128
3129	/* Sync the rest of page table (expensive but might be cheaper than nested-guest VM-exits in hardware). */
3130	for (unsigned iPteCur = 0; iPteCur < RT_ELEMENTS(pPt->a); iPteCur++)
3131	{
3132	if (iPteCur != iPte)
3133	{
3134	PGMPTWALKGST GstWalkPt;
3135	PGMPTWALK WalkPt;
3136	GCPhysNestedPage &= ~(SHW_PT_MASK << SHW_PT_SHIFT);
3137	GCPhysNestedPage \|= (iPteCur << GUEST_PAGE_SHIFT);
3138	int const rc2 = pgmGstSlatWalk(pVCpu, GCPhysNestedPage, false /fIsLinearAddrValid/, 0 /GCPtrNested/,
3139	&WalkPt, &GstWalkPt);
3140	if (RT_SUCCESS(rc2))
3141	{
3142	PGM_BTH_NAME(NestedSyncPageWorker)(pVCpu, &pPt->a[iPteCur], WalkPt.GCPhys, pShwPage, iPteCur,
3143	GstWalkPt.u.Ept.Pte);
3144	Log7Func(("GstPte=%RGp ShwPte=%RX64 iPte=%u\n", GstWalkPt.u.Ept.Pte.u, pPt->a[iPteCur].u, iPteCur));
3145	}
3146	else
3147	{
3148	/*
3149	* This could be MMIO pages reserved by the nested-hypevisor or genuinely not-present pages.
3150	* Ensure the shadow tables entry is not-present.
3151	*/
3152	/** @todo We currently don't configure these to cause EPT misconfigs but rather trap
3153	* them using EPT violations and walk the guest EPT tables to determine
3154	* whether they are EPT misconfigs VM-exits for the nested-hypervisor. We
3155	* could optimize this by using a specific combination of reserved bits
3156	* which we could immediately identify as EPT misconfigs of the
3157	* nested-hypervisor without having to walk its EPT tables. However, tracking
3158	* non-present entries might be tricky...
3159	*/
3160	AssertMsg(!pPt->a[iPteCur].u, ("%RX64\n", pPt->a[iPteCur].u));
3161	}
3162	if (RT_UNLIKELY(VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)))
3163	break;
3164	}
3165	}
3166	}
3167	else
3168	{
3169	Assert(rc == VINF_PGM_CACHED_PAGE);
3170	# if defined(VBOX_STRICT) && defined(DEBUG_ramshankar)
3171	/* Paranoia - Verify address of the page is what it should be. */
3172	PPGMPAGE pPage;
3173	rc = pgmPhysGetPageEx(pVM, GCPhysPage, &pPage);
3174	AssertRC(rc);
3175	const unsigned iPte = (GCPhysNestedPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
3176	AssertMsg(PGM_PAGE_GET_HCPHYS(pPage) == SHW_PTE_GET_HCPHYS(pPt->a[iPte]) \|\| !SHW_PTE_IS_P(pPt->a[iPte]),
3177	("PGM page and shadow PTE address conflict. GCPhysNestedPage=%RGp GCPhysPage=%RGp Page=%RHp Shw=%RHp\n",
3178	GCPhysNestedPage, GCPhysPage, PGM_PAGE_GET_HCPHYS(pPage), SHW_PTE_GET_HCPHYS(pPt->a[iPte])));
3179	Log7Func(("GstPte=%RGp ShwPte=%RX64 iPte=%u [cache]\n", pGstWalkAll->u.Ept.Pte.u, pPt->a[iPte].u, iPte));
3180	# endif
3181	rc = VINF_SUCCESS; /* Cached entry; assume it's still fully valid. */
3182	}
3183
3184	/* Save the new PDE. */
3185	uint64_t const fShwPdeFlags = pGstWalkAll->u.Ept.Pde.u & pVCpu->pgm.s.fGstEptShadowedPdeMask;
3186	Assert(!(pGstWalkAll->u.Ept.Pde.u & EPT_E_LEAF));
3187	Assert(!(pGstWalkAll->u.Ept.Pde.u & pVCpu->pgm.s.fGstEptMbzPdeMask));
3188	Pde.u = pShwPage->Core.Key \| fShwPdeFlags;
3189	SHW_PDE_ATOMIC_SET2(*pPde, Pde);
3190	Log7Func(("GstPde=%RGp ShwPde=%RX64 iPde=%u\n", pGstWalkAll->u.Ept.Pde.u, pPde->u, iPde));
3191
3192	STAM_PROFILE_STOP(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
3193	return rc;
3194	}
3195
3196	#endif /* !IN_RING3 && VBOX_WITH_NESTED_HWVIRT_VMX_EPT && PGM_SHW_TYPE == PGM_TYPE_EPT*/
3197	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && PGM_SHW_TYPE != PGM_TYPE_NONE
3198
3199	/**
3200	* Handle dirty bit tracking faults.
3201	*
3202	* @returns VBox status code.
3203	* @param pVCpu The cross context virtual CPU structure.
3204	* @param uErr Page fault error code.
3205	* @param pPdeSrc Guest page directory entry.
3206	* @param pPdeDst Shadow page directory entry.
3207	* @param GCPtrPage Guest context page address.
3208	*/
3209	static int PGM_BTH_NAME(CheckDirtyPageFault)(PVMCPUCC pVCpu, uint32_t uErr, PSHWPDE pPdeDst, GSTPDE const *pPdeSrc,
3210	RTGCPTR GCPtrPage)
3211	{
3212	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3213	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3214	NOREF(uErr);
3215
3216	PGM_LOCK_ASSERT_OWNER(pVM);
3217
3218	/*
3219	* Handle big page.
3220	*/
3221	if ((pPdeSrc->u & X86_PDE_PS) && GST_IS_PSE_ACTIVE(pVCpu))
3222	{
3223	if ((pPdeDst->u & (X86_PDE_P \| PGM_PDFLAGS_TRACK_DIRTY)) == (X86_PDE_P \| PGM_PDFLAGS_TRACK_DIRTY))
3224	{
3225	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,DirtyPageTrap));
3226	Assert(pPdeSrc->u & X86_PDE_RW);
3227
3228	/* Note: No need to invalidate this entry on other VCPUs as a stale TLB entry will not harm; write access will simply
3229	* fault again and take this path to only invalidate the entry (see below). */
3230	SHWPDE PdeDst = *pPdeDst;
3231	PdeDst.u &= ~(SHWUINT)PGM_PDFLAGS_TRACK_DIRTY;
3232	PdeDst.u \|= X86_PDE_RW \| X86_PDE_A;
3233	SHW_PDE_ATOMIC_SET2(*pPdeDst, PdeDst);
3234	PGM_INVL_BIG_PG(pVCpu, GCPtrPage);
3235	return VINF_PGM_HANDLED_DIRTY_BIT_FAULT; /* restarts the instruction. */
3236	}
3237
3238	# ifdef IN_RING0
3239	/* Check for stale TLB entry; only applies to the SMP guest case. */
3240	if ( pVM->cCpus > 1
3241	&& (pPdeDst->u & (X86_PDE_P \| X86_PDE_RW \| X86_PDE_A)) == (X86_PDE_P \| X86_PDE_RW \| X86_PDE_A))
3242	{
3243	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, pPdeDst->u & SHW_PDE_PG_MASK);
3244	if (pShwPage)
3245	{
3246	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
3247	PSHWPTE pPteDst = &pPTDst->a[(GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK];
3248	if (SHW_PTE_IS_P_RW(*pPteDst))
3249	{
3250	/* Stale TLB entry. */
3251	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,DirtyPageStale));
3252	PGM_INVL_PG(pVCpu, GCPtrPage);
3253	return VINF_PGM_HANDLED_DIRTY_BIT_FAULT; /* restarts the instruction. */
3254	}
3255	}
3256	}
3257	# endif /* IN_RING0 */
3258	return VINF_PGM_NO_DIRTY_BIT_TRACKING;
3259	}
3260
3261	/*
3262	* Map the guest page table.
3263	*/
3264	PGSTPT pPTSrc;
3265	int rc = PGM_GCPHYS_2_PTR_V2(pVM, pVCpu, GST_GET_PDE_GCPHYS(*pPdeSrc), &pPTSrc);
3266	AssertRCReturn(rc, rc);
3267
3268	if (SHW_PDE_IS_P(*pPdeDst))
3269	{
3270	GSTPTE const *pPteSrc = &pPTSrc->a[(GCPtrPage >> GST_PT_SHIFT) & GST_PT_MASK];
3271	const GSTPTE PteSrc = *pPteSrc;
3272
3273	/*
3274	* Map shadow page table.
3275	*/
3276	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, pPdeDst->u & SHW_PDE_PG_MASK);
3277	if (pShwPage)
3278	{
3279	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
3280	PSHWPTE pPteDst = &pPTDst->a[(GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK];
3281	if (SHW_PTE_IS_P(pPteDst)) /* @todo Optimize accessed bit emulation? */
3282	{
3283	if (SHW_PTE_IS_TRACK_DIRTY(*pPteDst))
3284	{
3285	PPGMPAGE pPage = pgmPhysGetPage(pVM, GST_GET_PTE_GCPHYS(PteSrc));
3286	SHWPTE PteDst = *pPteDst;
3287
3288	LogFlow(("DIRTY page trap addr=%RGv\n", GCPtrPage));
3289	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,DirtyPageTrap));
3290
3291	Assert(PteSrc.u & X86_PTE_RW);
3292
3293	/* Note: No need to invalidate this entry on other VCPUs as a stale TLB
3294	* entry will not harm; write access will simply fault again and
3295	* take this path to only invalidate the entry.
3296	*/
3297	if (RT_LIKELY(pPage))
3298	{
3299	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage) && !PGM_PAGE_IS_HNDL_PHYS_NOT_IN_HM(pPage))
3300	{
3301	//AssertMsgFailed(("%R[pgmpage] - we don't set PGM_PTFLAGS_TRACK_DIRTY for these pages\n", pPage));
3302	Assert(!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage));
3303	/* Assuming write handlers here as the PTE is present (otherwise we wouldn't be here). */
3304	SHW_PTE_SET_RO(PteDst);
3305	}
3306	else
3307	{
3308	if ( PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_WRITE_MONITORED
3309	&& PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM)
3310	{
3311	rc = pgmPhysPageMakeWritable(pVM, pPage, GST_GET_PTE_GCPHYS(PteSrc));
3312	AssertRC(rc);
3313	}
3314	if (PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED)
3315	SHW_PTE_SET_RW(PteDst);
3316	else
3317	{
3318	/* Still applies to shared pages. */
3319	Assert(!PGM_PAGE_IS_ZERO(pPage));
3320	SHW_PTE_SET_RO(PteDst);
3321	}
3322	}
3323	}
3324	else
3325	SHW_PTE_SET_RW(PteDst); /** @todo r=bird: This doesn't make sense to me. */
3326
3327	SHW_PTE_SET(PteDst, (SHW_PTE_GET_U(PteDst) \| X86_PTE_D \| X86_PTE_A) & ~(uint64_t)PGM_PTFLAGS_TRACK_DIRTY);
3328	SHW_PTE_ATOMIC_SET2(*pPteDst, PteDst);
3329	PGM_INVL_PG(pVCpu, GCPtrPage);
3330	return VINF_PGM_HANDLED_DIRTY_BIT_FAULT; /* restarts the instruction. */
3331	}
3332
3333	# ifdef IN_RING0
3334	/* Check for stale TLB entry; only applies to the SMP guest case. */
3335	if ( pVM->cCpus > 1
3336	&& SHW_PTE_IS_RW(*pPteDst)
3337	&& SHW_PTE_IS_A(*pPteDst))
3338	{
3339	/* Stale TLB entry. */
3340	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,DirtyPageStale));
3341	PGM_INVL_PG(pVCpu, GCPtrPage);
3342	return VINF_PGM_HANDLED_DIRTY_BIT_FAULT; /* restarts the instruction. */
3343	}
3344	# endif
3345	}
3346	}
3347	else
3348	AssertMsgFailed(("pgmPoolGetPageByHCPhys %RGp failed!\n", pPdeDst->u & SHW_PDE_PG_MASK));
3349	}
3350
3351	return VINF_PGM_NO_DIRTY_BIT_TRACKING;
3352	}
3353
3354	#endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && PGM_SHW_TYPE != PGM_TYPE_NONE */
3355
3356	/**
3357	* Sync a shadow page table.
3358	*
3359	* The shadow page table is not present in the shadow PDE.
3360	*
3361	* Handles mapping conflicts.
3362	*
3363	* This is called by VerifyAccessSyncPage, PrefetchPage, InvalidatePage (on
3364	* conflict), and Trap0eHandler.
3365	*
3366	* A precondition for this method is that the shadow PDE is not present. The
3367	* caller must take the PGM lock before checking this and continue to hold it
3368	* when calling this method.
3369	*
3370	* @returns VBox status code.
3371	* @param pVCpu The cross context virtual CPU structure.
3372	* @param iPDSrc Page directory index.
3373	* @param pPDSrc Source page directory (i.e. Guest OS page directory).
3374	* Assume this is a temporary mapping.
3375	* @param GCPtrPage GC Pointer of the page that caused the fault
3376	*/
3377	static int PGM_BTH_NAME(SyncPT)(PVMCPUCC pVCpu, unsigned iPDSrc, PGSTPD pPDSrc, RTGCPTR GCPtrPage)
3378	{
3379	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3380	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
3381
3382	#if 0 /* rarely useful; leave for debugging. */
3383	STAM_COUNTER_INC(&pVCpu->pgm.s.StatSyncPtPD[iPDSrc]);
3384	#endif
3385	LogFlow(("SyncPT: GCPtrPage=%RGv\n", GCPtrPage)); RT_NOREF_PV(GCPtrPage);
3386
3387	PGM_LOCK_ASSERT_OWNER(pVM);
3388
3389	#if ( PGM_GST_TYPE == PGM_TYPE_32BIT \
3390	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
3391	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64) \
3392	&& !PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) \
3393	&& PGM_SHW_TYPE != PGM_TYPE_NONE
3394	int rc = VINF_SUCCESS;
3395
3396	STAM_PROFILE_START(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
3397
3398	/*
3399	* Some input validation first.
3400	*/
3401	AssertMsg(iPDSrc == ((GCPtrPage >> GST_PD_SHIFT) & GST_PD_MASK), ("iPDSrc=%x GCPtrPage=%RGv\n", iPDSrc, GCPtrPage));
3402
3403	/*
3404	* Get the relevant shadow PDE entry.
3405	*/
3406	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
3407	const unsigned iPDDst = GCPtrPage >> SHW_PD_SHIFT;
3408	PSHWPDE pPdeDst = pgmShwGet32BitPDEPtr(pVCpu, GCPtrPage);
3409	AssertReturn(pPdeDst, VERR_INTERNAL_ERROR_3);
3410
3411	/* Fetch the pgm pool shadow descriptor. */
3412	PPGMPOOLPAGE pShwPde = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
3413	Assert(pShwPde);
3414
3415	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
3416	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
3417	PPGMPOOLPAGE pShwPde = NULL;
3418	PX86PDPAE pPDDst;
3419	PSHWPDE pPdeDst;
3420
3421	/* Fetch the pgm pool shadow descriptor. */
3422	rc = pgmShwGetPaePoolPagePD(pVCpu, GCPtrPage, &pShwPde);
3423	AssertRCSuccessReturn(rc, rc);
3424	Assert(pShwPde);
3425
3426	pPDDst = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPde);
3427	pPdeDst = &pPDDst->a[iPDDst];
3428
3429	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
3430	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
3431	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
3432	PX86PDPAE pPDDst = NULL; /* initialized to shut up gcc */
3433	PX86PDPT pPdptDst = NULL; /* initialized to shut up gcc */
3434	rc = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, NULL, &pPdptDst, &pPDDst);
3435	AssertRCSuccessReturn(rc, rc);
3436	Assert(pPDDst);
3437	PSHWPDE pPdeDst = &pPDDst->a[iPDDst];
3438
3439	# endif
3440	SHWPDE PdeDst = *pPdeDst;
3441
3442	# if PGM_GST_TYPE == PGM_TYPE_AMD64
3443	/* Fetch the pgm pool shadow descriptor. */
3444	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & X86_PDPE_PG_MASK);
3445	Assert(pShwPde);
3446	# endif
3447
3448	Assert(!SHW_PDE_IS_P(PdeDst)); /* We're only supposed to call SyncPT on PDE!P.*/
3449
3450	/*
3451	* Sync the page directory entry.
3452	*/
3453	GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
3454	const bool fPageTable = !(PdeSrc.u & X86_PDE_PS) \|\| !GST_IS_PSE_ACTIVE(pVCpu);
3455	if ( (PdeSrc.u & X86_PDE_P)
3456	&& (fPageTable ? GST_IS_PDE_VALID(pVCpu, PdeSrc) : GST_IS_BIG_PDE_VALID(pVCpu, PdeSrc)) )
3457	{
3458	/*
3459	* Allocate & map the page table.
3460	*/
3461	PSHWPT pPTDst;
3462	PPGMPOOLPAGE pShwPage;
3463	RTGCPHYS GCPhys;
3464	if (fPageTable)
3465	{
3466	GCPhys = GST_GET_PDE_GCPHYS(PdeSrc);
3467	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
3468	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
3469	GCPhys = PGM_A20_APPLY(pVCpu, GCPhys \| ((iPDDst & 1) * (GUEST_PAGE_SIZE / 2)));
3470	# endif
3471	rc = pgmPoolAlloc(pVM, GCPhys, BTH_PGMPOOLKIND_PT_FOR_PT, PGMPOOLACCESS_DONTCARE, PGM_A20_IS_ENABLED(pVCpu),
3472	pShwPde->idx, iPDDst, false /fLockPage/,
3473	&pShwPage);
3474	}
3475	else
3476	{
3477	PGMPOOLACCESS enmAccess;
3478	# if PGM_WITH_NX(PGM_GST_TYPE, PGM_SHW_TYPE)
3479	const bool fNoExecute = (PdeSrc.u & X86_PDE_PAE_NX) && GST_IS_NX_ACTIVE(pVCpu);
3480	# else
3481	const bool fNoExecute = false;
3482	# endif
3483
3484	GCPhys = GST_GET_BIG_PDE_GCPHYS(pVM, PdeSrc);
3485	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
3486	/* Select the right PDE as we're emulating a 4MB page directory with two 2 MB shadow PDEs.*/
3487	GCPhys = PGM_A20_APPLY(pVCpu, GCPhys \| (GCPtrPage & (1 << X86_PD_PAE_SHIFT)));
3488	# endif
3489	/* Determine the right kind of large page to avoid incorrect cached entry reuse. */
3490	if (PdeSrc.u & X86_PDE_US)
3491	{
3492	if (PdeSrc.u & X86_PDE_RW)
3493	enmAccess = (fNoExecute) ? PGMPOOLACCESS_USER_RW_NX : PGMPOOLACCESS_USER_RW;
3494	else
3495	enmAccess = (fNoExecute) ? PGMPOOLACCESS_USER_R_NX : PGMPOOLACCESS_USER_R;
3496	}
3497	else
3498	{
3499	if (PdeSrc.u & X86_PDE_RW)
3500	enmAccess = (fNoExecute) ? PGMPOOLACCESS_SUPERVISOR_RW_NX : PGMPOOLACCESS_SUPERVISOR_RW;
3501	else
3502	enmAccess = (fNoExecute) ? PGMPOOLACCESS_SUPERVISOR_R_NX : PGMPOOLACCESS_SUPERVISOR_R;
3503	}
3504	rc = pgmPoolAlloc(pVM, GCPhys, BTH_PGMPOOLKIND_PT_FOR_BIG, enmAccess, PGM_A20_IS_ENABLED(pVCpu),
3505	pShwPde->idx, iPDDst, false /fLockPage/,
3506	&pShwPage);
3507	}
3508	if (rc == VINF_SUCCESS)
3509	pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
3510	else if (rc == VINF_PGM_CACHED_PAGE)
3511	{
3512	/*
3513	* The PT was cached, just hook it up.
3514	*/
3515	if (fPageTable)
3516	PdeDst.u = pShwPage->Core.Key \| GST_GET_PDE_SHW_FLAGS(pVCpu, PdeSrc);
3517	else
3518	{
3519	PdeDst.u = pShwPage->Core.Key \| GST_GET_BIG_PDE_SHW_FLAGS(pVCpu, PdeSrc);
3520	/* (see explanation and assumptions further down.) */
3521	if ((PdeSrc.u & (X86_PDE_RW \| X86_PDE4M_D)) == X86_PDE_RW)
3522	{
3523	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,DirtyPageBig));
3524	PdeDst.u \|= PGM_PDFLAGS_TRACK_DIRTY;
3525	PdeDst.u &= ~(SHWUINT)X86_PDE_RW;
3526	}
3527	}
3528	SHW_PDE_ATOMIC_SET2(*pPdeDst, PdeDst);
3529	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
3530	return VINF_SUCCESS;
3531	}
3532	else
3533	AssertMsgFailedReturn(("rc=%Rrc\n", rc), RT_FAILURE_NP(rc) ? rc : VERR_IPE_UNEXPECTED_INFO_STATUS);
3534	/** @todo Why do we bother preserving X86_PDE_AVL_MASK here?
3535	* Both PGM_PDFLAGS_MAPPING and PGM_PDFLAGS_TRACK_DIRTY should be
3536	* irrelevant at this point. */
3537	PdeDst.u &= X86_PDE_AVL_MASK;
3538	PdeDst.u \|= pShwPage->Core.Key;
3539
3540	/*
3541	* Page directory has been accessed (this is a fault situation, remember).
3542	*/
3543	/** @todo
3544	* Well, when the caller is PrefetchPage or InvalidatePage is isn't a
3545	* fault situation. What's more, the Trap0eHandler has already set the
3546	* accessed bit. So, it's actually just VerifyAccessSyncPage which
3547	* might need setting the accessed flag.
3548	*
3549	* The best idea is to leave this change to the caller and add an
3550	* assertion that it's set already. */
3551	pPDSrc->a[iPDSrc].u \|= X86_PDE_A;
3552	if (fPageTable)
3553	{
3554	/*
3555	* Page table - 4KB.
3556	*
3557	* Sync all or just a few entries depending on PGM_SYNC_N_PAGES.
3558	*/
3559	Log2(("SyncPT: 4K %RGv PdeSrc:{P=%d RW=%d U=%d raw=%08llx}\n",
3560	GCPtrPage, PdeSrc.u & X86_PTE_P, !!(PdeSrc.u & X86_PTE_RW), !!(PdeSrc.u & X86_PDE_US), (uint64_t)PdeSrc.u));
3561	PGSTPT pPTSrc;
3562	rc = PGM_GCPHYS_2_PTR(pVM, GST_GET_PDE_GCPHYS(PdeSrc), &pPTSrc);
3563	if (RT_SUCCESS(rc))
3564	{
3565	/*
3566	* Start by syncing the page directory entry so CSAM's TLB trick works.
3567	*/
3568	PdeDst.u = (PdeDst.u & (SHW_PDE_PG_MASK \| X86_PDE_AVL_MASK))
3569	\| GST_GET_PDE_SHW_FLAGS(pVCpu, PdeSrc);
3570	SHW_PDE_ATOMIC_SET2(*pPdeDst, PdeDst);
3571	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
3572
3573	/*
3574	* Directory/page user or supervisor privilege: (same goes for read/write)
3575	*
3576	* Directory Page Combined
3577	* U/S U/S U/S
3578	* 0 0 0
3579	* 0 1 0
3580	* 1 0 0
3581	* 1 1 1
3582	*
3583	* Simple AND operation. Table listed for completeness.
3584	*
3585	*/
3586	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT4K));
3587	# ifdef PGM_SYNC_N_PAGES
3588	unsigned iPTBase = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
3589	unsigned iPTDst = iPTBase;
3590	const unsigned iPTDstEnd = RT_MIN(iPTDst + PGM_SYNC_NR_PAGES / 2, RT_ELEMENTS(pPTDst->a));
3591	if (iPTDst <= PGM_SYNC_NR_PAGES / 2)
3592	iPTDst = 0;
3593	else
3594	iPTDst -= PGM_SYNC_NR_PAGES / 2;
3595	# else /* !PGM_SYNC_N_PAGES */
3596	unsigned iPTDst = 0;
3597	const unsigned iPTDstEnd = RT_ELEMENTS(pPTDst->a);
3598	# endif /* !PGM_SYNC_N_PAGES */
3599	RTGCPTR GCPtrCur = (GCPtrPage & ~(RTGCPTR)((1 << SHW_PD_SHIFT) - 1))
3600	\| ((RTGCPTR)iPTDst << GUEST_PAGE_SHIFT);
3601	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
3602	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
3603	const unsigned offPTSrc = ((GCPtrPage >> SHW_PD_SHIFT) & 1) * 512;
3604	# else
3605	const unsigned offPTSrc = 0;
3606	# endif
3607	for (; iPTDst < iPTDstEnd; iPTDst++, GCPtrCur += GUEST_PAGE_SIZE)
3608	{
3609	const unsigned iPTSrc = iPTDst + offPTSrc;
3610	const GSTPTE PteSrc = pPTSrc->a[iPTSrc];
3611	if (PteSrc.u & X86_PTE_P)
3612	{
3613	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, PteSrc, pShwPage, iPTDst);
3614	Log2(("SyncPT: 4K+ %RGv PteSrc:{P=%d RW=%d U=%d raw=%08llx}%s dst.raw=%08llx iPTSrc=%x PdeSrc.u=%x physpte=%RGp\n",
3615	GCPtrCur,
3616	PteSrc.u & X86_PTE_P,
3617	!!(PteSrc.u & PdeSrc.u & X86_PTE_RW),
3618	!!(PteSrc.u & PdeSrc.u & X86_PTE_US),
3619	(uint64_t)PteSrc.u,
3620	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : "", SHW_PTE_LOG64(pPTDst->a[iPTDst]), iPTSrc, PdeSrc.au32[0],
3621	(RTGCPHYS)(GST_GET_PDE_GCPHYS(PdeSrc) + iPTSrc*sizeof(PteSrc)) ));
3622	}
3623	/* else: the page table was cleared by the pool */
3624	} /* for PTEs */
3625	}
3626	}
3627	else
3628	{
3629	/*
3630	* Big page - 2/4MB.
3631	*
3632	* We'll walk the ram range list in parallel and optimize lookups.
3633	* We will only sync one shadow page table at a time.
3634	*/
3635	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT4M));
3636
3637	/**
3638	* @todo It might be more efficient to sync only a part of the 4MB
3639	* page (similar to what we do for 4KB PDs).
3640	*/
3641
3642	/*
3643	* Start by syncing the page directory entry.
3644	*/
3645	PdeDst.u = (PdeDst.u & (SHW_PDE_PG_MASK \| (X86_PDE_AVL_MASK & ~PGM_PDFLAGS_TRACK_DIRTY)))
3646	\| GST_GET_BIG_PDE_SHW_FLAGS(pVCpu, PdeSrc);
3647
3648	/*
3649	* If the page is not flagged as dirty and is writable, then make it read-only
3650	* at PD level, so we can set the dirty bit when the page is modified.
3651	*
3652	* ASSUMES that page access handlers are implemented on page table entry level.
3653	* Thus we will first catch the dirty access and set PDE.D and restart. If
3654	* there is an access handler, we'll trap again and let it work on the problem.
3655	*/
3656	/** @todo move the above stuff to a section in the PGM documentation. */
3657	Assert(!(PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY));
3658	if ((PdeSrc.u & (X86_PDE_RW \| X86_PDE4M_D)) == X86_PDE_RW)
3659	{
3660	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,DirtyPageBig));
3661	PdeDst.u \|= PGM_PDFLAGS_TRACK_DIRTY;
3662	PdeDst.u &= ~(SHWUINT)X86_PDE_RW;
3663	}
3664	SHW_PDE_ATOMIC_SET2(*pPdeDst, PdeDst);
3665	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
3666
3667	/*
3668	* Fill the shadow page table.
3669	*/
3670	/* Get address and flags from the source PDE. */
3671	SHWPTE PteDstBase;
3672	SHW_PTE_SET(PteDstBase, GST_GET_BIG_PDE_SHW_FLAGS_4_PTE(pVCpu, PdeSrc));
3673
3674	/* Loop thru the entries in the shadow PT. */
3675	const RTGCPTR GCPtr = (GCPtrPage >> SHW_PD_SHIFT) << SHW_PD_SHIFT; NOREF(GCPtr);
3676	Log2(("SyncPT: BIG %RGv PdeSrc:{P=%d RW=%d U=%d raw=%08llx} Shw=%RGv GCPhys=%RGp %s\n",
3677	GCPtrPage, PdeSrc.u & X86_PDE_P, !!(PdeSrc.u & X86_PDE_RW), !!(PdeSrc.u & X86_PDE_US), (uint64_t)PdeSrc.u, GCPtr,
3678	GCPhys, PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY ? " Track-Dirty" : ""));
3679	unsigned iPTDst = 0;
3680	while ( iPTDst < RT_ELEMENTS(pPTDst->a)
3681	&& !VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY))
3682	{
3683	PPGMRAMRANGE const pRam = pgmPhysGetRangeAtOrAbove(pVM, GCPhys);
3684	if (pRam && GCPhys >= pRam->GCPhys)
3685	{
3686	# ifndef PGM_WITH_A20
3687	unsigned iHCPage = (GCPhys - pRam->GCPhys) >> GUEST_PAGE_SHIFT;
3688	# endif
3689	do
3690	{
3691	/* Make shadow PTE. */
3692	# ifdef PGM_WITH_A20
3693	PPGMPAGE pPage = &pRam->aPages[(GCPhys - pRam->GCPhys) >> GUEST_PAGE_SHIFT];
3694	# else
3695	PPGMPAGE pPage = &pRam->aPages[iHCPage];
3696	# endif
3697	SHWPTE PteDst;
3698
3699	# ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
3700	/* Try to make the page writable if necessary. */
3701	if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM
3702	&& ( PGM_PAGE_IS_ZERO(pPage)
3703	\|\| ( SHW_PTE_IS_RW(PteDstBase)
3704	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED
3705	# ifdef VBOX_WITH_REAL_WRITE_MONITORED_PAGES
3706	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_WRITE_MONITORED
3707	# endif
3708	# ifdef VBOX_WITH_PAGE_SHARING
3709	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_SHARED
3710	# endif
3711	&& !PGM_PAGE_IS_BALLOONED(pPage))
3712	)
3713	)
3714	{
3715	rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhys);
3716	AssertRCReturn(rc, rc);
3717	if (VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY))
3718	break;
3719	}
3720	# endif
3721
3722	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage) && !PGM_PAGE_IS_HNDL_PHYS_NOT_IN_HM(pPage))
3723	PGM_BTH_NAME(SyncHandlerPte)(pVM, pVCpu, pPage, GCPhys, SHW_PTE_GET_U(PteDstBase), &PteDst);
3724	else if (PGM_PAGE_IS_BALLOONED(pPage))
3725	SHW_PTE_SET(PteDst, 0); /* Handle ballooned pages at #PF time. */
3726	else
3727	SHW_PTE_SET(PteDst, PGM_PAGE_GET_HCPHYS(pPage) \| SHW_PTE_GET_U(PteDstBase));
3728
3729	/* Only map writable pages writable. */
3730	if ( SHW_PTE_IS_P_RW(PteDst)
3731	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
3732	{
3733	# ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
3734	/* Still applies to shared pages. */
3735	Assert(!PGM_PAGE_IS_ZERO(pPage));
3736	# endif
3737	SHW_PTE_SET_RO(PteDst); /** @todo this isn't quite working yet... */
3738	Log3(("SyncPT: write-protecting %RGp pPage=%R[pgmpage] at %RGv\n", GCPhys, pPage, (RTGCPTR)(GCPtr \| (iPTDst << SHW_PT_SHIFT))));
3739	}
3740
3741	if (SHW_PTE_IS_P(PteDst))
3742	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPTDst);
3743
3744	/* commit it (not atomic, new table) */
3745	pPTDst->a[iPTDst] = PteDst;
3746	Log4(("SyncPT: BIG %RGv PteDst:{P=%d RW=%d U=%d raw=%08llx}%s\n",
3747	(RTGCPTR)(GCPtr \| (iPTDst << SHW_PT_SHIFT)), SHW_PTE_IS_P(PteDst), SHW_PTE_IS_RW(PteDst), SHW_PTE_IS_US(PteDst), SHW_PTE_LOG64(PteDst),
3748	SHW_PTE_IS_TRACK_DIRTY(PteDst) ? " Track-Dirty" : ""));
3749
3750	/* advance */
3751	GCPhys += GUEST_PAGE_SIZE;
3752	PGM_A20_APPLY_TO_VAR(pVCpu, GCPhys);
3753	# ifndef PGM_WITH_A20
3754	iHCPage++;
3755	# endif
3756	iPTDst++;
3757	} while ( iPTDst < RT_ELEMENTS(pPTDst->a)
3758	&& GCPhys <= pRam->GCPhysLast);
3759	}
3760	else if (pRam)
3761	{
3762	Log(("Invalid pages at %RGp\n", GCPhys));
3763	do
3764	{
3765	SHW_PTE_SET(pPTDst->a[iPTDst], 0); /* Invalid page, we must handle them manually. */
3766	GCPhys += GUEST_PAGE_SIZE;
3767	iPTDst++;
3768	} while ( iPTDst < RT_ELEMENTS(pPTDst->a)
3769	&& GCPhys < pRam->GCPhys);
3770	PGM_A20_APPLY_TO_VAR(pVCpu,GCPhys);
3771	}
3772	else
3773	{
3774	Log(("Invalid pages at %RGp (2)\n", GCPhys));
3775	for ( ; iPTDst < RT_ELEMENTS(pPTDst->a); iPTDst++)
3776	SHW_PTE_SET(pPTDst->a[iPTDst], 0); /* Invalid page, we must handle them manually. */
3777	}
3778	} /* while more PTEs */
3779	} /* 4KB / 4MB */
3780	}
3781	else
3782	AssertRelease(!SHW_PDE_IS_P(PdeDst));
3783
3784	STAM_PROFILE_STOP(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
3785	if (RT_FAILURE(rc))
3786	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPTFailed));
3787	return rc;
3788
3789	#elif (PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
3790	&& !PGM_TYPE_IS_NESTED(PGM_SHW_TYPE) \
3791	&& (PGM_SHW_TYPE != PGM_TYPE_EPT \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
3792	&& PGM_SHW_TYPE != PGM_TYPE_NONE
3793	NOREF(iPDSrc); NOREF(pPDSrc);
3794
3795	STAM_PROFILE_START(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
3796
3797	/*
3798	* Validate input a little bit.
3799	*/
3800	int rc = VINF_SUCCESS;
3801	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
3802	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
3803	PSHWPDE pPdeDst = pgmShwGet32BitPDEPtr(pVCpu, GCPtrPage);
3804	AssertReturn(pPdeDst, VERR_INTERNAL_ERROR_3);
3805
3806	/* Fetch the pgm pool shadow descriptor. */
3807	PPGMPOOLPAGE pShwPde = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
3808	Assert(pShwPde);
3809
3810	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
3811	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
3812	PPGMPOOLPAGE pShwPde = NULL; /* initialized to shut up gcc */
3813	PX86PDPAE pPDDst;
3814	PSHWPDE pPdeDst;
3815
3816	/* Fetch the pgm pool shadow descriptor. */
3817	rc = pgmShwGetPaePoolPagePD(pVCpu, GCPtrPage, &pShwPde);
3818	AssertRCSuccessReturn(rc, rc);
3819	Assert(pShwPde);
3820
3821	pPDDst = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPde);
3822	pPdeDst = &pPDDst->a[iPDDst];
3823
3824	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
3825	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
3826	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
3827	PX86PDPAE pPDDst = NULL; /* initialized to shut up gcc */
3828	PX86PDPT pPdptDst= NULL; /* initialized to shut up gcc */
3829	rc = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, NULL, &pPdptDst, &pPDDst);
3830	AssertRCSuccessReturn(rc, rc);
3831	Assert(pPDDst);
3832	PSHWPDE pPdeDst = &pPDDst->a[iPDDst];
3833
3834	/* Fetch the pgm pool shadow descriptor. */
3835	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & X86_PDPE_PG_MASK);
3836	Assert(pShwPde);
3837
3838	# elif PGM_SHW_TYPE == PGM_TYPE_EPT
3839	const unsigned iPdpt = (GCPtrPage >> EPT_PDPT_SHIFT) & EPT_PDPT_MASK;
3840	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
3841	PEPTPD pPDDst;
3842	PEPTPDPT pPdptDst;
3843
3844	rc = pgmShwGetEPTPDPtr(pVCpu, GCPtrPage, &pPdptDst, &pPDDst);
3845	if (rc != VINF_SUCCESS)
3846	{
3847	STAM_PROFILE_STOP(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
3848	AssertRC(rc);
3849	return rc;
3850	}
3851	Assert(pPDDst);
3852	PSHWPDE pPdeDst = &pPDDst->a[iPDDst];
3853
3854	/* Fetch the pgm pool shadow descriptor. */
3855	/** @todo r=bird: didn't pgmShwGetEPTPDPtr just do this lookup already? */
3856	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & EPT_PDPTE_PG_MASK);
3857	Assert(pShwPde);
3858	# endif
3859	SHWPDE PdeDst = *pPdeDst;
3860
3861	Assert(!SHW_PDE_IS_P(PdeDst)); /* We're only supposed to call SyncPT on PDE!P and conflicts.*/
3862
3863	# if defined(PGM_WITH_LARGE_PAGES) && PGM_SHW_TYPE != PGM_TYPE_32BIT && PGM_SHW_TYPE != PGM_TYPE_PAE
3864	if (BTH_IS_NP_ACTIVE(pVM))
3865	{
3866	Assert(!VM_IS_NEM_ENABLED(pVM));
3867
3868	/* Check if we allocated a big page before for this 2 MB range. */
3869	PPGMPAGE pPage;
3870	rc = pgmPhysGetPageEx(pVM, PGM_A20_APPLY(pVCpu, GCPtrPage & X86_PDE2M_PAE_PG_MASK), &pPage);
3871	if (RT_SUCCESS(rc))
3872	{
3873	RTHCPHYS HCPhys = NIL_RTHCPHYS;
3874	if (PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE)
3875	{
3876	if (PGM_A20_IS_ENABLED(pVCpu))
3877	{
3878	STAM_REL_COUNTER_INC(&pVM->pgm.s.StatLargePageReused);
3879	AssertRelease(PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED);
3880	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
3881	}
3882	else
3883	{
3884	PGM_PAGE_SET_PDE_TYPE(pVM, pPage, PGM_PAGE_PDE_TYPE_PDE_DISABLED);
3885	pVM->pgm.s.cLargePagesDisabled++;
3886	}
3887	}
3888	else if ( PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE_DISABLED
3889	&& PGM_A20_IS_ENABLED(pVCpu))
3890	{
3891	/* Recheck the entire 2 MB range to see if we can use it again as a large page. */
3892	rc = pgmPhysRecheckLargePage(pVM, GCPtrPage, pPage);
3893	if (RT_SUCCESS(rc))
3894	{
3895	Assert(PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED);
3896	Assert(PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE);
3897	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
3898	}
3899	}
3900	# if !defined(VBOX_WITH_NEW_LAZY_PAGE_ALLOC) && !defined(PGM_WITH_PAGE_ZEROING_DETECTION) /* This code is too aggresive! */
3901	else if ( PGMIsUsingLargePages(pVM)
3902	&& PGM_A20_IS_ENABLED(pVCpu))
3903	{
3904	rc = pgmPhysAllocLargePage(pVM, GCPtrPage);
3905	if (RT_SUCCESS(rc))
3906	{
3907	Assert(PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED);
3908	Assert(PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE);
3909	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
3910	}
3911	else
3912	LogFlow(("pgmPhysAllocLargePage failed with %Rrc\n", rc));
3913	}
3914	# endif
3915
3916	if (HCPhys != NIL_RTHCPHYS)
3917	{
3918	# if PGM_SHW_TYPE == PGM_TYPE_EPT
3919	PdeDst.u = HCPhys \| EPT_E_READ \| EPT_E_WRITE \| EPT_E_EXECUTE \| EPT_E_LEAF \| EPT_E_IGNORE_PAT \| EPT_E_MEMTYPE_WB
3920	\| (PdeDst.u & X86_PDE_AVL_MASK) /** @todo do we need this? */;
3921	# else
3922	PdeDst.u = HCPhys \| X86_PDE_P \| X86_PDE_RW \| X86_PDE_US \| X86_PDE_PS
3923	\| (PdeDst.u & X86_PDE_AVL_MASK) /** @todo PGM_PD_FLAGS? */;
3924	# endif
3925	SHW_PDE_ATOMIC_SET2(*pPdeDst, PdeDst);
3926
3927	Log(("SyncPT: Use large page at %RGp PDE=%RX64\n", GCPtrPage, PdeDst.u));
3928	/* Add a reference to the first page only. */
3929	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPde, PGM_PAGE_GET_TRACKING(pPage), pPage, iPDDst);
3930
3931	STAM_PROFILE_STOP(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
3932	return VINF_SUCCESS;
3933	}
3934	}
3935	}
3936	# endif /* defined(PGM_WITH_LARGE_PAGES) && PGM_SHW_TYPE != PGM_TYPE_32BIT && PGM_SHW_TYPE != PGM_TYPE_PAE */
3937
3938	/*
3939	* Allocate & map the page table.
3940	*/
3941	PSHWPT pPTDst;
3942	PPGMPOOLPAGE pShwPage;
3943	RTGCPHYS GCPhys;
3944
3945	/* Virtual address = physical address */
3946	GCPhys = PGM_A20_APPLY(pVCpu, GCPtrPage & X86_PAGE_4K_BASE_MASK);
3947	rc = pgmPoolAlloc(pVM, GCPhys & ~(RT_BIT_64(SHW_PD_SHIFT) - 1), BTH_PGMPOOLKIND_PT_FOR_PT, PGMPOOLACCESS_DONTCARE,
3948	PGM_A20_IS_ENABLED(pVCpu), pShwPde->idx, iPDDst, false /fLockPage/,
3949	&pShwPage);
3950	if ( rc == VINF_SUCCESS
3951	\|\| rc == VINF_PGM_CACHED_PAGE)
3952	pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
3953	else
3954	{
3955	STAM_PROFILE_STOP(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
3956	AssertMsgFailedReturn(("rc=%Rrc\n", rc), RT_FAILURE_NP(rc) ? rc : VERR_IPE_UNEXPECTED_INFO_STATUS);
3957	}
3958
3959	if (rc == VINF_SUCCESS)
3960	{
3961	/* New page table; fully set it up. */
3962	Assert(pPTDst);
3963
3964	/* Mask away the page offset. */
3965	GCPtrPage &= ~(RTGCPTR)GUEST_PAGE_OFFSET_MASK;
3966
3967	for (unsigned iPTDst = 0; iPTDst < RT_ELEMENTS(pPTDst->a); iPTDst++)
3968	{
3969	RTGCPTR GCPtrCurPage = PGM_A20_APPLY(pVCpu, (GCPtrPage & ~(RTGCPTR)(SHW_PT_MASK << SHW_PT_SHIFT))
3970	\| (iPTDst << GUEST_PAGE_SHIFT));
3971
3972	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], GCPtrCurPage, pShwPage, iPTDst);
3973	Log2(("SyncPage: 4K+ %RGv PteSrc:{P=1 RW=1 U=1} PteDst=%08llx%s\n",
3974	GCPtrCurPage,
3975	SHW_PTE_LOG64(pPTDst->a[iPTDst]),
3976	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : ""));
3977
3978	if (RT_UNLIKELY(VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)))
3979	break;
3980	}
3981	}
3982	else
3983	rc = VINF_SUCCESS; /* Cached entry; assume it's still fully valid. */
3984
3985	/* Save the new PDE. */
3986	# if PGM_SHW_TYPE == PGM_TYPE_EPT
3987	PdeDst.u = pShwPage->Core.Key \| EPT_E_READ \| EPT_E_WRITE \| EPT_E_EXECUTE
3988	\| (PdeDst.u & X86_PDE_AVL_MASK /** @todo do we really need this? */);
3989	# else
3990	PdeDst.u = pShwPage->Core.Key \| X86_PDE_P \| X86_PDE_RW \| X86_PDE_US \| X86_PDE_A
3991	\| (PdeDst.u & X86_PDE_AVL_MASK /** @todo use a PGM_PD_FLAGS define */);
3992	# endif
3993	SHW_PDE_ATOMIC_SET2(*pPdeDst, PdeDst);
3994
3995	STAM_PROFILE_STOP(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPT), a);
3996	if (RT_FAILURE(rc))
3997	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,SyncPTFailed));
3998	return rc;
3999
4000	#else
4001	NOREF(iPDSrc); NOREF(pPDSrc);
4002	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_SHW_TYPE, PGM_GST_TYPE));
4003	return VERR_PGM_NOT_USED_IN_MODE;
4004	#endif
4005	}
4006
4007
4008
4009	/**
4010	* Prefetch a page/set of pages.
4011	*
4012	* Typically used to sync commonly used pages before entering raw mode
4013	* after a CR3 reload.
4014	*
4015	* @returns VBox status code.
4016	* @param pVCpu The cross context virtual CPU structure.
4017	* @param GCPtrPage Page to invalidate.
4018	*/
4019	PGM_BTH_DECL(int, PrefetchPage)(PVMCPUCC pVCpu, RTGCPTR GCPtrPage)
4020	{
4021	#if ( PGM_GST_TYPE == PGM_TYPE_32BIT \
4022	\|\| PGM_GST_TYPE == PGM_TYPE_REAL \
4023	\|\| PGM_GST_TYPE == PGM_TYPE_PROT \
4024	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
4025	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64 ) \
4026	&& !PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) \
4027	&& PGM_SHW_TYPE != PGM_TYPE_NONE
4028	/*
4029	* Check that all Guest levels thru the PDE are present, getting the
4030	* PD and PDE in the processes.
4031	*/
4032	int rc = VINF_SUCCESS;
4033	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
4034	# if PGM_GST_TYPE == PGM_TYPE_32BIT
4035	const unsigned iPDSrc = (uint32_t)GCPtrPage >> GST_PD_SHIFT;
4036	PGSTPD pPDSrc = pgmGstGet32bitPDPtr(pVCpu);
4037	# elif PGM_GST_TYPE == PGM_TYPE_PAE
4038	unsigned iPDSrc;
4039	X86PDPE PdpeSrc;
4040	PGSTPD pPDSrc = pgmGstGetPaePDPtr(pVCpu, GCPtrPage, &iPDSrc, &PdpeSrc);
4041	if (!pPDSrc)
4042	return VINF_SUCCESS; /* not present */
4043	# elif PGM_GST_TYPE == PGM_TYPE_AMD64
4044	unsigned iPDSrc;
4045	PX86PML4E pPml4eSrc;
4046	X86PDPE PdpeSrc;
4047	PGSTPD pPDSrc = pgmGstGetLongModePDPtr(pVCpu, GCPtrPage, &pPml4eSrc, &PdpeSrc, &iPDSrc);
4048	if (!pPDSrc)
4049	return VINF_SUCCESS; /* not present */
4050	# endif
4051	const GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
4052	# else
4053	PGSTPD pPDSrc = NULL;
4054	const unsigned iPDSrc = 0;
4055	GSTPDE const PdeSrc = { X86_PDE_P \| X86_PDE_RW \| X86_PDE_US \| X86_PDE_A }; /* faked so we don't have to #ifdef everything */
4056	# endif
4057
4058	if ((PdeSrc.u & (X86_PDE_P \| X86_PDE_A)) == (X86_PDE_P \| X86_PDE_A))
4059	{
4060	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4061	PGM_LOCK_VOID(pVM);
4062
4063	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
4064	const X86PDE PdeDst = pgmShwGet32BitPDE(pVCpu, GCPtrPage);
4065	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
4066	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
4067	PX86PDPAE pPDDst;
4068	X86PDEPAE PdeDst;
4069	# if PGM_GST_TYPE != PGM_TYPE_PAE
4070	X86PDPE PdpeSrc;
4071
4072	/* Fake PDPT entry; access control handled on the page table level, so allow everything. */
4073	PdpeSrc.u = X86_PDPE_P; /* rw/us are reserved for PAE pdpte's; accessed bit causes invalid VT-x guest state errors */
4074	# endif
4075	rc = pgmShwSyncPaePDPtr(pVCpu, GCPtrPage, PdpeSrc.u, &pPDDst);
4076	if (rc != VINF_SUCCESS)
4077	{
4078	PGM_UNLOCK(pVM);
4079	AssertRC(rc);
4080	return rc;
4081	}
4082	Assert(pPDDst);
4083	PdeDst = pPDDst->a[iPDDst];
4084
4085	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
4086	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
4087	PX86PDPAE pPDDst;
4088	X86PDEPAE PdeDst;
4089
4090	# if PGM_GST_TYPE == PGM_TYPE_PROT
4091	/* AMD-V nested paging */
4092	X86PML4E Pml4eSrc;
4093	X86PDPE PdpeSrc;
4094	PX86PML4E pPml4eSrc = &Pml4eSrc;
4095
4096	/* Fake PML4 & PDPT entry; access control handled on the page table level, so allow everything. */
4097	Pml4eSrc.u = X86_PML4E_P \| X86_PML4E_RW \| X86_PML4E_US \| X86_PML4E_A;
4098	PdpeSrc.u = X86_PDPE_P \| X86_PDPE_RW \| X86_PDPE_US \| X86_PDPE_A;
4099	# endif
4100
4101	rc = pgmShwSyncLongModePDPtr(pVCpu, GCPtrPage, pPml4eSrc->u, PdpeSrc.u, &pPDDst);
4102	if (rc != VINF_SUCCESS)
4103	{
4104	PGM_UNLOCK(pVM);
4105	AssertRC(rc);
4106	return rc;
4107	}
4108	Assert(pPDDst);
4109	PdeDst = pPDDst->a[iPDDst];
4110	# endif
4111	if (!(PdeDst.u & X86_PDE_P))
4112	{
4113	/** @todo r=bird: This guy will set the A bit on the PDE,
4114	* probably harmless. */
4115	rc = PGM_BTH_NAME(SyncPT)(pVCpu, iPDSrc, pPDSrc, GCPtrPage);
4116	}
4117	else
4118	{
4119	/* Note! We used to sync PGM_SYNC_NR_PAGES pages, which triggered assertions in CSAM, because
4120	* R/W attributes of nearby pages were reset. Not sure how that could happen. Anyway, it
4121	* makes no sense to prefetch more than one page.
4122	*/
4123	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, GCPtrPage, 1, 0);
4124	if (RT_SUCCESS(rc))
4125	rc = VINF_SUCCESS;
4126	}
4127	PGM_UNLOCK(pVM);
4128	}
4129	return rc;
4130
4131	#elif PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) \|\| PGM_SHW_TYPE == PGM_TYPE_NONE
4132	NOREF(pVCpu); NOREF(GCPtrPage);
4133	return VINF_SUCCESS; /* ignore */
4134	#else
4135	AssertCompile(0);
4136	#endif
4137	}
4138
4139
4140
4141
4142	/**
4143	* Syncs a page during a PGMVerifyAccess() call.
4144	*
4145	* @returns VBox status code (informational included).
4146	* @param pVCpu The cross context virtual CPU structure.
4147	* @param GCPtrPage The address of the page to sync.
4148	* @param fPage The effective guest page flags.
4149	* @param uErr The trap error code.
4150	* @remarks This will normally never be called on invalid guest page
4151	* translation entries.
4152	*/
4153	PGM_BTH_DECL(int, VerifyAccessSyncPage)(PVMCPUCC pVCpu, RTGCPTR GCPtrPage, unsigned fPage, unsigned uErr)
4154	{
4155	PVMCC pVM = pVCpu->CTX_SUFF(pVM); NOREF(pVM);
4156
4157	LogFlow(("VerifyAccessSyncPage: GCPtrPage=%RGv fPage=%#x uErr=%#x\n", GCPtrPage, fPage, uErr));
4158	RT_NOREF_PV(GCPtrPage); RT_NOREF_PV(fPage); RT_NOREF_PV(uErr);
4159
4160	Assert(!pVM->pgm.s.fNestedPaging);
4161	#if ( PGM_GST_TYPE == PGM_TYPE_32BIT \
4162	\|\| PGM_GST_TYPE == PGM_TYPE_REAL \
4163	\|\| PGM_GST_TYPE == PGM_TYPE_PROT \
4164	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
4165	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64 ) \
4166	&& !PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) \
4167	&& PGM_SHW_TYPE != PGM_TYPE_NONE
4168
4169	/*
4170	* Get guest PD and index.
4171	*/
4172	/** @todo Performance: We've done all this a jiffy ago in the
4173	* PGMGstGetPage call. */
4174	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
4175	# if PGM_GST_TYPE == PGM_TYPE_32BIT
4176	const unsigned iPDSrc = (uint32_t)GCPtrPage >> GST_PD_SHIFT;
4177	PGSTPD pPDSrc = pgmGstGet32bitPDPtr(pVCpu);
4178
4179	# elif PGM_GST_TYPE == PGM_TYPE_PAE
4180	unsigned iPDSrc = 0;
4181	X86PDPE PdpeSrc;
4182	PGSTPD pPDSrc = pgmGstGetPaePDPtr(pVCpu, GCPtrPage, &iPDSrc, &PdpeSrc);
4183	if (RT_UNLIKELY(!pPDSrc))
4184	{
4185	Log(("PGMVerifyAccess: access violation for %RGv due to non-present PDPTR\n", GCPtrPage));
4186	return VINF_EM_RAW_GUEST_TRAP;
4187	}
4188
4189	# elif PGM_GST_TYPE == PGM_TYPE_AMD64
4190	unsigned iPDSrc = 0; /* shut up gcc */
4191	PX86PML4E pPml4eSrc = NULL; /* ditto */
4192	X86PDPE PdpeSrc;
4193	PGSTPD pPDSrc = pgmGstGetLongModePDPtr(pVCpu, GCPtrPage, &pPml4eSrc, &PdpeSrc, &iPDSrc);
4194	if (RT_UNLIKELY(!pPDSrc))
4195	{
4196	Log(("PGMVerifyAccess: access violation for %RGv due to non-present PDPTR\n", GCPtrPage));
4197	return VINF_EM_RAW_GUEST_TRAP;
4198	}
4199	# endif
4200
4201	# else /* !PGM_WITH_PAGING */
4202	PGSTPD pPDSrc = NULL;
4203	const unsigned iPDSrc = 0;
4204	# endif /* !PGM_WITH_PAGING */
4205	int rc = VINF_SUCCESS;
4206
4207	PGM_LOCK_VOID(pVM);
4208
4209	/*
4210	* First check if the shadow pd is present.
4211	*/
4212	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
4213	PX86PDE pPdeDst = pgmShwGet32BitPDEPtr(pVCpu, GCPtrPage);
4214	AssertReturn(pPdeDst, VERR_INTERNAL_ERROR_3);
4215
4216	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
4217	PX86PDEPAE pPdeDst;
4218	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
4219	PX86PDPAE pPDDst;
4220	# if PGM_GST_TYPE != PGM_TYPE_PAE
4221	/* Fake PDPT entry; access control handled on the page table level, so allow everything. */
4222	X86PDPE PdpeSrc;
4223	PdpeSrc.u = X86_PDPE_P; /* rw/us are reserved for PAE pdpte's; accessed bit causes invalid VT-x guest state errors */
4224	# endif
4225	rc = pgmShwSyncPaePDPtr(pVCpu, GCPtrPage, PdpeSrc.u, &pPDDst);
4226	if (rc != VINF_SUCCESS)
4227	{
4228	PGM_UNLOCK(pVM);
4229	AssertRC(rc);
4230	return rc;
4231	}
4232	Assert(pPDDst);
4233	pPdeDst = &pPDDst->a[iPDDst];
4234
4235	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
4236	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
4237	PX86PDPAE pPDDst;
4238	PX86PDEPAE pPdeDst;
4239
4240	# if PGM_GST_TYPE == PGM_TYPE_PROT
4241	/* AMD-V nested paging: Fake PML4 & PDPT entry; access control handled on the page table level, so allow everything. */
4242	X86PML4E Pml4eSrc;
4243	X86PDPE PdpeSrc;
4244	PX86PML4E pPml4eSrc = &Pml4eSrc;
4245	Pml4eSrc.u = X86_PML4E_P \| X86_PML4E_RW \| X86_PML4E_US \| X86_PML4E_A;
4246	PdpeSrc.u = X86_PDPE_P \| X86_PDPE_RW \| X86_PDPE_US \| X86_PDPE_A;
4247	# endif
4248
4249	rc = pgmShwSyncLongModePDPtr(pVCpu, GCPtrPage, pPml4eSrc->u, PdpeSrc.u, &pPDDst);
4250	if (rc != VINF_SUCCESS)
4251	{
4252	PGM_UNLOCK(pVM);
4253	AssertRC(rc);
4254	return rc;
4255	}
4256	Assert(pPDDst);
4257	pPdeDst = &pPDDst->a[iPDDst];
4258	# endif
4259
4260	if (!(pPdeDst->u & X86_PDE_P))
4261	{
4262	rc = PGM_BTH_NAME(SyncPT)(pVCpu, iPDSrc, pPDSrc, GCPtrPage);
4263	if (rc != VINF_SUCCESS)
4264	{
4265	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
4266	PGM_UNLOCK(pVM);
4267	AssertRC(rc);
4268	return rc;
4269	}
4270	}
4271
4272	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
4273	/* Check for dirty bit fault */
4274	rc = PGM_BTH_NAME(CheckDirtyPageFault)(pVCpu, uErr, pPdeDst, &pPDSrc->a[iPDSrc], GCPtrPage);
4275	if (rc == VINF_PGM_HANDLED_DIRTY_BIT_FAULT)
4276	Log(("PGMVerifyAccess: success (dirty)\n"));
4277	else
4278	# endif
4279	{
4280	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
4281	GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
4282	# else
4283	GSTPDE const PdeSrc = { X86_PDE_P \| X86_PDE_RW \| X86_PDE_US \| X86_PDE_A }; /* faked so we don't have to #ifdef everything */
4284	# endif
4285
4286	Assert(rc != VINF_EM_RAW_GUEST_TRAP);
4287	if (uErr & X86_TRAP_PF_US)
4288	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,PageOutOfSyncUser));
4289	else /* supervisor */
4290	STAM_COUNTER_INC(&pVCpu->pgm.s.Stats.CTX_MID_Z(Stat,PageOutOfSyncSupervisor));
4291
4292	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, GCPtrPage, 1, 0);
4293	if (RT_SUCCESS(rc))
4294	{
4295	/* Page was successfully synced */
4296	Log2(("PGMVerifyAccess: success (sync)\n"));
4297	rc = VINF_SUCCESS;
4298	}
4299	else
4300	{
4301	Log(("PGMVerifyAccess: access violation for %RGv rc=%Rrc\n", GCPtrPage, rc));
4302	rc = VINF_EM_RAW_GUEST_TRAP;
4303	}
4304	}
4305	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
4306	PGM_UNLOCK(pVM);
4307	return rc;
4308
4309	#else /* PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) */
4310
4311	AssertLogRelMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_GST_TYPE, PGM_SHW_TYPE));
4312	return VERR_PGM_NOT_USED_IN_MODE;
4313	#endif /* PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) */
4314	}
4315
4316
4317	/**
4318	* Syncs the paging hierarchy starting at CR3.
4319	*
4320	* @returns VBox status code, R0/RC may return VINF_PGM_SYNC_CR3, no other
4321	* informational status codes.
4322	* @retval VERR_PGM_NO_HYPERVISOR_ADDRESS in raw-mode when we're unable to map
4323	* the VMM into guest context.
4324	* @param pVCpu The cross context virtual CPU structure.
4325	* @param cr0 Guest context CR0 register.
4326	* @param cr3 Guest context CR3 register. Not subjected to the A20
4327	* mask.
4328	* @param cr4 Guest context CR4 register.
4329	* @param fGlobal Including global page directories or not
4330	*/
4331	PGM_BTH_DECL(int, SyncCR3)(PVMCPUCC pVCpu, uint64_t cr0, uint64_t cr3, uint64_t cr4, bool fGlobal)
4332	{
4333	PVMCC pVM = pVCpu->CTX_SUFF(pVM); NOREF(pVM);
4334	NOREF(cr0); NOREF(cr3); NOREF(cr4); NOREF(fGlobal);
4335
4336	LogFlow(("SyncCR3 FF=%d fGlobal=%d\n", !!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3), fGlobal));
4337
4338	#if !PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) && PGM_SHW_TYPE != PGM_TYPE_NONE
4339	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
4340	PGM_LOCK_VOID(pVM);
4341	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4342	if (pPool->cDirtyPages)
4343	pgmPoolResetDirtyPages(pVM);
4344	PGM_UNLOCK(pVM);
4345	# endif
4346	#endif /* !NESTED && !EPT */
4347
4348	#if PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) \|\| PGM_SHW_TYPE == PGM_TYPE_NONE
4349	/*
4350	* Nested / EPT / None - No work.
4351	*/
4352	return VINF_SUCCESS;
4353
4354	#elif PGM_SHW_TYPE == PGM_TYPE_AMD64
4355	/*
4356	* AMD64 (Shw & Gst) - No need to check all paging levels; we zero
4357	* out the shadow parts when the guest modifies its tables.
4358	*/
4359	return VINF_SUCCESS;
4360
4361	#else /* !PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) && PGM_SHW_TYPE != PGM_TYPE_AMD64 */
4362
4363	return VINF_SUCCESS;
4364	#endif /* !PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) && PGM_SHW_TYPE != PGM_TYPE_AMD64 */
4365	}
4366
4367
4368
4369
4370	#ifdef VBOX_STRICT
4371
4372	/**
4373	* Checks that the shadow page table is in sync with the guest one.
4374	*
4375	* @returns The number of errors.
4376	* @param pVCpu The cross context virtual CPU structure.
4377	* @param cr3 Guest context CR3 register.
4378	* @param cr4 Guest context CR4 register.
4379	* @param GCPtr Where to start. Defaults to 0.
4380	* @param cb How much to check. Defaults to everything.
4381	*/
4382	PGM_BTH_DECL(unsigned, AssertCR3)(PVMCPUCC pVCpu, uint64_t cr3, uint64_t cr4, RTGCPTR GCPtr, RTGCPTR cb)
4383	{
4384	NOREF(pVCpu); NOREF(cr3); NOREF(cr4); NOREF(GCPtr); NOREF(cb);
4385	#if PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) \|\| PGM_SHW_TYPE == PGM_TYPE_NONE
4386	return 0;
4387	#else
4388	unsigned cErrors = 0;
4389	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4390	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
4391
4392	# if PGM_GST_TYPE == PGM_TYPE_PAE
4393	/** @todo currently broken; crashes below somewhere */
4394	AssertFailed();
4395	# endif
4396
4397	# if PGM_GST_TYPE == PGM_TYPE_32BIT \
4398	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
4399	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64
4400
4401	bool fBigPagesSupported = GST_IS_PSE_ACTIVE(pVCpu);
4402	PPGMCPU pPGM = &pVCpu->pgm.s;
4403	RTGCPHYS GCPhysGst; /* page address derived from the guest page tables. */
4404	RTHCPHYS HCPhysShw; /* page address derived from the shadow page tables. */
4405	# ifndef IN_RING0
4406	RTHCPHYS HCPhys; /* general usage. */
4407	# endif
4408	int rc;
4409
4410	/*
4411	* Check that the Guest CR3 and all its mappings are correct.
4412	*/
4413	AssertMsgReturn(pPGM->GCPhysCR3 == PGM_A20_APPLY(pVCpu, cr3 & GST_CR3_PAGE_MASK),
4414	("Invalid GCPhysCR3=%RGp cr3=%RGp\n", pPGM->GCPhysCR3, (RTGCPHYS)cr3),
4415	false);
4416	# if !defined(IN_RING0) && PGM_GST_TYPE != PGM_TYPE_AMD64
4417	# if 0
4418	# if PGM_GST_TYPE == PGM_TYPE_32BIT
4419	rc = PGMShwGetPage(pVCpu, (RTRCUINTPTR)pPGM->pGst32BitPdRC, NULL, &HCPhysShw);
4420	# else
4421	rc = PGMShwGetPage(pVCpu, (RTRCUINTPTR)pPGM->pGstPaePdptRC, NULL, &HCPhysShw);
4422	# endif
4423	AssertRCReturn(rc, 1);
4424	HCPhys = NIL_RTHCPHYS;
4425	rc = pgmRamGCPhys2HCPhys(pVM, PGM_A20_APPLY(pVCpu, cr3 & GST_CR3_PAGE_MASK), &HCPhys);
4426	AssertMsgReturn(HCPhys == HCPhysShw, ("HCPhys=%RHp HCPhyswShw=%RHp (cr3)\n", HCPhys, HCPhysShw), false);
4427	# endif
4428	# if PGM_GST_TYPE == PGM_TYPE_32BIT && defined(IN_RING3)
4429	pgmGstGet32bitPDPtr(pVCpu);
4430	RTGCPHYS GCPhys;
4431	rc = PGMR3DbgR3Ptr2GCPhys(pVM->pUVM, pPGM->pGst32BitPdR3, &GCPhys);
4432	AssertRCReturn(rc, 1);
4433	AssertMsgReturn(PGM_A20_APPLY(pVCpu, cr3 & GST_CR3_PAGE_MASK) == GCPhys, ("GCPhys=%RGp cr3=%RGp\n", GCPhys, (RTGCPHYS)cr3), false);
4434	# endif
4435	# endif /* !IN_RING0 */
4436
4437	/*
4438	* Get and check the Shadow CR3.
4439	*/
4440	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
4441	unsigned cPDEs = X86_PG_ENTRIES;
4442	unsigned cIncrement = X86_PG_ENTRIES * GUEST_PAGE_SIZE;
4443	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
4444	# if PGM_GST_TYPE == PGM_TYPE_32BIT
4445	unsigned cPDEs = X86_PG_PAE_ENTRIES * 4; /* treat it as a 2048 entry table. */
4446	# else
4447	unsigned cPDEs = X86_PG_PAE_ENTRIES;
4448	# endif
4449	unsigned cIncrement = X86_PG_PAE_ENTRIES * GUEST_PAGE_SIZE;
4450	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
4451	unsigned cPDEs = X86_PG_PAE_ENTRIES;
4452	unsigned cIncrement = X86_PG_PAE_ENTRIES * GUEST_PAGE_SIZE;
4453	# endif
4454	if (cb != ~(RTGCPTR)0)
4455	cPDEs = RT_MIN(cb >> SHW_PD_SHIFT, 1);
4456
4457	/** @todo call the other two PGMAssert() functions. /
4458
4459	# if PGM_GST_TYPE == PGM_TYPE_AMD64
4460	unsigned iPml4 = (GCPtr >> X86_PML4_SHIFT) & X86_PML4_MASK;
4461
4462	for (; iPml4 < X86_PG_PAE_ENTRIES; iPml4++)
4463	{
4464	PX86PML4E const pPml4eSrc = pgmGstGetLongModePML4EPtr(pVCpu, iPml4);
4465	AssertContinueStmt(pPml4eSrc, cErrors++);
4466
4467	PX86PML4E const pPml4eDst = pgmShwGetLongModePML4EPtr(pVCpu, iPml4);
4468	AssertContinueStmt(pPml4eDst, cErrors++);
4469
4470	/* Fetch the pgm pool shadow descriptor if the shadow pml4e is present. */
4471	if (!(pPml4eDst->u & X86_PML4E_P))
4472	{
4473	GCPtr += _2M * UINT64_C(512) * UINT64_C(512);
4474	continue;
4475	}
4476
4477	PPGMPOOLPAGE pShwPdpt = pgmPoolGetPage(pPool, pPml4eDst->u & X86_PML4E_PG_MASK);
4478	RTGCPHYS GCPhysPdptSrc = PGM_A20_APPLY(pVCpu, pPml4eSrc->u & X86_PML4E_PG_MASK);
4479
4480	if ((pPml4eSrc->u & X86_PML4E_P) != (pPml4eDst->u & X86_PML4E_P))
4481	{
4482	AssertMsgFailed(("Present bit doesn't match! pPml4eDst.u=%#RX64 pPml4eSrc.u=%RX64\n", pPml4eDst->u, pPml4eSrc->u));
4483	GCPtr += _2M * UINT64_C(512) * UINT64_C(512);
4484	cErrors++;
4485	continue;
4486	}
4487
4488	if (GCPhysPdptSrc != pShwPdpt->GCPhys)
4489	{
4490	AssertMsgFailed(("Physical address doesn't match! iPml4 %d pPml4eDst.u=%#RX64 pPml4eSrc.u=%RX64 Phys %RX64 vs %RX64\n", iPml4, pPml4eDst->u, pPml4eSrc->u, pShwPdpt->GCPhys, GCPhysPdptSrc));
4491	GCPtr += _2M * UINT64_C(512) * UINT64_C(512);
4492	cErrors++;
4493	continue;
4494	}
4495
4496	if ( (pPml4eDst->u & (X86_PML4E_US \| X86_PML4E_RW \| X86_PML4E_NX))
4497	!= (pPml4eSrc->u & (X86_PML4E_US \| X86_PML4E_RW \| X86_PML4E_NX)))
4498	{
4499	AssertMsgFailed(("User/Write/NoExec bits don't match! pPml4eDst.u=%#RX64 pPml4eSrc.u=%RX64\n", pPml4eDst->u, pPml4eSrc->u));
4500	GCPtr += _2M * UINT64_C(512) * UINT64_C(512);
4501	cErrors++;
4502	continue;
4503	}
4504	# else /* PGM_GST_TYPE != PGM_TYPE_AMD64 */
4505	{
4506	# endif /* PGM_GST_TYPE != PGM_TYPE_AMD64 */
4507
4508	# if PGM_GST_TYPE == PGM_TYPE_AMD64 \|\| PGM_GST_TYPE == PGM_TYPE_PAE
4509	/*
4510	* Check the PDPTEs too.
4511	*/
4512	unsigned iPdpt = (GCPtr >> SHW_PDPT_SHIFT) & SHW_PDPT_MASK;
4513
4514	for (;iPdpt <= SHW_PDPT_MASK; iPdpt++)
4515	{
4516	unsigned iPDSrc = 0; /* initialized to shut up gcc */
4517	PPGMPOOLPAGE pShwPde = NULL;
4518	PX86PDPE pPdpeDst;
4519	RTGCPHYS GCPhysPdeSrc;
4520	X86PDPE PdpeSrc;
4521	PdpeSrc.u = 0; /* initialized to shut up gcc 4.5 */
4522	# if PGM_GST_TYPE == PGM_TYPE_PAE
4523	PGSTPD pPDSrc = pgmGstGetPaePDPtr(pVCpu, GCPtr, &iPDSrc, &PdpeSrc);
4524	PX86PDPT pPdptDst = pgmShwGetPaePDPTPtr(pVCpu);
4525	# else
4526	PX86PML4E pPml4eSrcIgn;
4527	PX86PDPT pPdptDst;
4528	PX86PDPAE pPDDst;
4529	PGSTPD pPDSrc = pgmGstGetLongModePDPtr(pVCpu, GCPtr, &pPml4eSrcIgn, &PdpeSrc, &iPDSrc);
4530
4531	rc = pgmShwGetLongModePDPtr(pVCpu, GCPtr, NULL, &pPdptDst, &pPDDst);
4532	if (rc != VINF_SUCCESS)
4533	{
4534	AssertMsg(rc == VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT, ("Unexpected rc=%Rrc\n", rc));
4535	GCPtr += 512 * _2M;
4536	continue; /* next PDPTE */
4537	}
4538	Assert(pPDDst);
4539	# endif
4540	Assert(iPDSrc == 0);
4541
4542	pPdpeDst = &pPdptDst->a[iPdpt];
4543
4544	if (!(pPdpeDst->u & X86_PDPE_P))
4545	{
4546	GCPtr += 512 * _2M;
4547	continue; /* next PDPTE */
4548	}
4549
4550	pShwPde = pgmPoolGetPage(pPool, pPdpeDst->u & X86_PDPE_PG_MASK);
4551	GCPhysPdeSrc = PGM_A20_APPLY(pVCpu, PdpeSrc.u & X86_PDPE_PG_MASK);
4552
4553	if ((pPdpeDst->u & X86_PDPE_P) != (PdpeSrc.u & X86_PDPE_P))
4554	{
4555	AssertMsgFailed(("Present bit doesn't match! pPdpeDst.u=%#RX64 pPdpeSrc.u=%RX64\n", pPdpeDst->u, PdpeSrc.u));
4556	GCPtr += 512 * _2M;
4557	cErrors++;
4558	continue;
4559	}
4560
4561	if (GCPhysPdeSrc != pShwPde->GCPhys)
4562	{
4563	# if PGM_GST_TYPE == PGM_TYPE_AMD64
4564	AssertMsgFailed(("Physical address doesn't match! iPml4 %d iPdpt %d pPdpeDst.u=%#RX64 pPdpeSrc.u=%RX64 Phys %RX64 vs %RX64\n", iPml4, iPdpt, pPdpeDst->u, PdpeSrc.u, pShwPde->GCPhys, GCPhysPdeSrc));
4565	# else
4566	AssertMsgFailed(("Physical address doesn't match! iPdpt %d pPdpeDst.u=%#RX64 pPdpeSrc.u=%RX64 Phys %RX64 vs %RX64\n", iPdpt, pPdpeDst->u, PdpeSrc.u, pShwPde->GCPhys, GCPhysPdeSrc));
4567	# endif
4568	GCPtr += 512 * _2M;
4569	cErrors++;
4570	continue;
4571	}
4572
4573	# if PGM_GST_TYPE == PGM_TYPE_AMD64
4574	if ( (pPdpeDst->u & (X86_PDPE_US \| X86_PDPE_RW \| X86_PDPE_LM_NX))
4575	!= (PdpeSrc.u & (X86_PDPE_US \| X86_PDPE_RW \| X86_PDPE_LM_NX)))
4576	{
4577	AssertMsgFailed(("User/Write/NoExec bits don't match! pPdpeDst.u=%#RX64 pPdpeSrc.u=%RX64\n", pPdpeDst->u, PdpeSrc.u));
4578	GCPtr += 512 * _2M;
4579	cErrors++;
4580	continue;
4581	}
4582	# endif
4583
4584	# else /* PGM_GST_TYPE != PGM_TYPE_AMD64 && PGM_GST_TYPE != PGM_TYPE_PAE */
4585	{
4586	# endif /* PGM_GST_TYPE != PGM_TYPE_AMD64 && PGM_GST_TYPE != PGM_TYPE_PAE */
4587	# if PGM_GST_TYPE == PGM_TYPE_32BIT
4588	GSTPD const *pPDSrc = pgmGstGet32bitPDPtr(pVCpu);
4589	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
4590	PCX86PD pPDDst = pgmShwGet32BitPDPtr(pVCpu);
4591	# endif
4592	# endif /* PGM_GST_TYPE == PGM_TYPE_32BIT */
4593	/*
4594	* Iterate the shadow page directory.
4595	*/
4596	GCPtr = (GCPtr >> SHW_PD_SHIFT) << SHW_PD_SHIFT;
4597	unsigned iPDDst = (GCPtr >> SHW_PD_SHIFT) & SHW_PD_MASK;
4598
4599	for (;
4600	iPDDst < cPDEs;
4601	iPDDst++, GCPtr += cIncrement)
4602	{
4603	# if PGM_SHW_TYPE == PGM_TYPE_PAE
4604	const SHWPDE PdeDst = *pgmShwGetPaePDEPtr(pVCpu, GCPtr);
4605	# else
4606	const SHWPDE PdeDst = pPDDst->a[iPDDst];
4607	# endif
4608	if ( (PdeDst.u & X86_PDE_P)
4609	\|\| ((PdeDst.u & (X86_PDE_P \| PGM_PDFLAGS_TRACK_DIRTY)) == (X86_PDE_P \| PGM_PDFLAGS_TRACK_DIRTY)) )
4610	{
4611	HCPhysShw = PdeDst.u & SHW_PDE_PG_MASK;
4612	PPGMPOOLPAGE pPoolPage = pgmPoolGetPage(pPool, HCPhysShw);
4613	if (!pPoolPage)
4614	{
4615	AssertMsgFailed(("Invalid page table address %RHp at %RGv! PdeDst=%#RX64\n",
4616	HCPhysShw, GCPtr, (uint64_t)PdeDst.u));
4617	cErrors++;
4618	continue;
4619	}
4620	const SHWPT pPTDst = (const SHWPT )PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pPoolPage);
4621
4622	if (PdeDst.u & (X86_PDE4M_PWT \| X86_PDE4M_PCD))
4623	{
4624	AssertMsgFailed(("PDE flags PWT and/or PCD is set at %RGv! These flags are not virtualized! PdeDst=%#RX64\n",
4625	GCPtr, (uint64_t)PdeDst.u));
4626	cErrors++;
4627	}
4628
4629	if (PdeDst.u & (X86_PDE4M_G \| X86_PDE4M_D))
4630	{
4631	AssertMsgFailed(("4K PDE reserved flags at %RGv! PdeDst=%#RX64\n",
4632	GCPtr, (uint64_t)PdeDst.u));
4633	cErrors++;
4634	}
4635
4636	const GSTPDE PdeSrc = pPDSrc->a[(iPDDst >> (GST_PD_SHIFT - SHW_PD_SHIFT)) & GST_PD_MASK];
4637	if (!(PdeSrc.u & X86_PDE_P))
4638	{
4639	AssertMsgFailed(("Guest PDE at %RGv is not present! PdeDst=%#RX64 PdeSrc=%#RX64\n",
4640	GCPtr, (uint64_t)PdeDst.u, (uint64_t)PdeSrc.u));
4641	cErrors++;
4642	continue;
4643	}
4644
4645	if ( !(PdeSrc.u & X86_PDE_PS)
4646	\|\| !fBigPagesSupported)
4647	{
4648	GCPhysGst = GST_GET_PDE_GCPHYS(PdeSrc);
4649	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
4650	GCPhysGst = PGM_A20_APPLY(pVCpu, GCPhysGst \| ((iPDDst & 1) * (GUEST_PAGE_SIZE / 2)));
4651	# endif
4652	}
4653	else
4654	{
4655	# if PGM_GST_TYPE == PGM_TYPE_32BIT
4656	if (PdeSrc.u & X86_PDE4M_PG_HIGH_MASK)
4657	{
4658	AssertMsgFailed(("Guest PDE at %RGv is using PSE36 or similar! PdeSrc=%#RX64\n",
4659	GCPtr, (uint64_t)PdeSrc.u));
4660	cErrors++;
4661	continue;
4662	}
4663	# endif
4664	GCPhysGst = GST_GET_BIG_PDE_GCPHYS(pVM, PdeSrc);
4665	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
4666	GCPhysGst = PGM_A20_APPLY(pVCpu, GCPhysGst \| (GCPtr & RT_BIT(X86_PAGE_2M_SHIFT)));
4667	# endif
4668	}
4669
4670	if ( pPoolPage->enmKind
4671	!= (!(PdeSrc.u & X86_PDE_PS) \|\| !fBigPagesSupported ? BTH_PGMPOOLKIND_PT_FOR_PT : BTH_PGMPOOLKIND_PT_FOR_BIG))
4672	{
4673	AssertMsgFailed(("Invalid shadow page table kind %d at %RGv! PdeSrc=%#RX64\n",
4674	pPoolPage->enmKind, GCPtr, (uint64_t)PdeSrc.u));
4675	cErrors++;
4676	}
4677
4678	PPGMPAGE pPhysPage = pgmPhysGetPage(pVM, GCPhysGst);
4679	if (!pPhysPage)
4680	{
4681	AssertMsgFailed(("Cannot find guest physical address %RGp in the PDE at %RGv! PdeSrc=%#RX64\n",
4682	GCPhysGst, GCPtr, (uint64_t)PdeSrc.u));
4683	cErrors++;
4684	continue;
4685	}
4686
4687	if (GCPhysGst != pPoolPage->GCPhys)
4688	{
4689	AssertMsgFailed(("GCPhysGst=%RGp != pPage->GCPhys=%RGp at %RGv\n",
4690	GCPhysGst, pPoolPage->GCPhys, GCPtr));
4691	cErrors++;
4692	continue;
4693	}
4694
4695	if ( !(PdeSrc.u & X86_PDE_PS)
4696	\|\| !fBigPagesSupported)
4697	{
4698	/*
4699	* Page Table.
4700	*/
4701	const GSTPT *pPTSrc;
4702	rc = PGM_GCPHYS_2_PTR_V2(pVM, pVCpu, PGM_A20_APPLY(pVCpu, GCPhysGst & ~(RTGCPHYS)(GUEST_PAGE_SIZE - 1)),
4703	&pPTSrc);
4704	if (RT_FAILURE(rc))
4705	{
4706	AssertMsgFailed(("Cannot map/convert guest physical address %RGp in the PDE at %RGv! PdeSrc=%#RX64\n",
4707	GCPhysGst, GCPtr, (uint64_t)PdeSrc.u));
4708	cErrors++;
4709	continue;
4710	}
4711	if ( (PdeSrc.u & (X86_PDE_P \| X86_PDE_US \| X86_PDE_RW/* \| X86_PDE_A*/))
4712	!= (PdeDst.u & (X86_PDE_P \| X86_PDE_US \| X86_PDE_RW/* \| X86_PDE_A*/)))
4713	{
4714	/// @todo We get here a lot on out-of-sync CR3 entries. The access handler should zap them to avoid false alarms here!
4715	// (This problem will go away when/if we shadow multiple CR3s.)
4716	AssertMsgFailed(("4K PDE flags mismatch at %RGv! PdeSrc=%#RX64 PdeDst=%#RX64\n",
4717	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4718	cErrors++;
4719	continue;
4720	}
4721	if (PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY)
4722	{
4723	AssertMsgFailed(("4K PDEs cannot have PGM_PDFLAGS_TRACK_DIRTY set! GCPtr=%RGv PdeDst=%#RX64\n",
4724	GCPtr, (uint64_t)PdeDst.u));
4725	cErrors++;
4726	continue;
4727	}
4728
4729	/* iterate the page table. */
4730	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
4731	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
4732	const unsigned offPTSrc = ((GCPtr >> SHW_PD_SHIFT) & 1) * 512;
4733	# else
4734	const unsigned offPTSrc = 0;
4735	# endif
4736	for (unsigned iPT = 0, off = 0;
4737	iPT < RT_ELEMENTS(pPTDst->a);
4738	iPT++, off += GUEST_PAGE_SIZE)
4739	{
4740	const SHWPTE PteDst = pPTDst->a[iPT];
4741
4742	/* skip not-present and dirty tracked entries. */
4743	if (!(SHW_PTE_GET_U(PteDst) & (X86_PTE_P \| PGM_PTFLAGS_TRACK_DIRTY))) /** @todo deal with ALL handlers and CSAM !P pages! */
4744	continue;
4745	Assert(SHW_PTE_IS_P(PteDst));
4746
4747	const GSTPTE PteSrc = pPTSrc->a[iPT + offPTSrc];
4748	if (!(PteSrc.u & X86_PTE_P))
4749	{
4750	# ifdef IN_RING3
4751	PGMAssertHandlerAndFlagsInSync(pVM);
4752	DBGFR3PagingDumpEx(pVM->pUVM, pVCpu->idCpu, DBGFPGDMP_FLAGS_CURRENT_CR3 \| DBGFPGDMP_FLAGS_CURRENT_MODE
4753	\| DBGFPGDMP_FLAGS_GUEST \| DBGFPGDMP_FLAGS_HEADER \| DBGFPGDMP_FLAGS_PRINT_CR3,
4754	0, 0, UINT64_MAX, 99, NULL);
4755	# endif
4756	AssertMsgFailed(("Out of sync (!P) PTE at %RGv! PteSrc=%#RX64 PteDst=%#RX64 pPTSrc=%RGv iPTSrc=%x PdeSrc=%x physpte=%RGp\n",
4757	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst), pPTSrc, iPT + offPTSrc, PdeSrc.au32[0],
4758	(uint64_t)GST_GET_PDE_GCPHYS(PdeSrc) + (iPT + offPTSrc) * sizeof(PteSrc)));
4759	cErrors++;
4760	continue;
4761	}
4762
4763	uint64_t fIgnoreFlags = GST_PTE_PG_MASK \| X86_PTE_AVL_MASK \| X86_PTE_G \| X86_PTE_D \| X86_PTE_PWT \| X86_PTE_PCD \| X86_PTE_PAT;
4764	# if 1 /** @todo sync accessed bit properly... */
4765	fIgnoreFlags \|= X86_PTE_A;
4766	# endif
4767
4768	/* match the physical addresses */
4769	HCPhysShw = SHW_PTE_GET_HCPHYS(PteDst);
4770	GCPhysGst = GST_GET_PTE_GCPHYS(PteSrc);
4771
4772	# ifdef IN_RING3
4773	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysGst, &HCPhys);
4774	if (RT_FAILURE(rc))
4775	{
4776	# if 0
4777	if (HCPhysShw != MMR3PageDummyHCPhys(pVM)) /** @todo this is wrong. */
4778	{
4779	AssertMsgFailed(("Cannot find guest physical address %RGp at %RGv! PteSrc=%#RX64 PteDst=%#RX64\n",
4780	GCPhysGst, GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4781	cErrors++;
4782	continue;
4783	}
4784	# endif
4785	}
4786	else if (HCPhysShw != (HCPhys & SHW_PTE_PG_MASK))
4787	{
4788	AssertMsgFailed(("Out of sync (phys) at %RGv! HCPhysShw=%RHp HCPhys=%RHp GCPhysGst=%RGp PteSrc=%#RX64 PteDst=%#RX64\n",
4789	GCPtr + off, HCPhysShw, HCPhys, GCPhysGst, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4790	cErrors++;
4791	continue;
4792	}
4793	# endif
4794
4795	pPhysPage = pgmPhysGetPage(pVM, GCPhysGst);
4796	if (!pPhysPage)
4797	{
4798	# if 0
4799	if (HCPhysShw != MMR3PageDummyHCPhys(pVM)) /** @todo this is wrong. */
4800	{
4801	AssertMsgFailed(("Cannot find guest physical address %RGp at %RGv! PteSrc=%#RX64 PteDst=%#RX64\n",
4802	GCPhysGst, GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4803	cErrors++;
4804	continue;
4805	}
4806	# endif
4807	if (SHW_PTE_IS_RW(PteDst))
4808	{
4809	AssertMsgFailed(("Invalid guest page at %RGv is writable! GCPhysGst=%RGp PteSrc=%#RX64 PteDst=%#RX64\n",
4810	GCPtr + off, GCPhysGst, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4811	cErrors++;
4812	}
4813	fIgnoreFlags \|= X86_PTE_RW;
4814	}
4815	else if (HCPhysShw != PGM_PAGE_GET_HCPHYS(pPhysPage))
4816	{
4817	AssertMsgFailed(("Out of sync (phys) at %RGv! HCPhysShw=%RHp pPhysPage:%R[pgmpage] GCPhysGst=%RGp PteSrc=%#RX64 PteDst=%#RX64\n",
4818	GCPtr + off, HCPhysShw, pPhysPage, GCPhysGst, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4819	cErrors++;
4820	continue;
4821	}
4822
4823	/* flags */
4824	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPhysPage) && !PGM_PAGE_IS_HNDL_PHYS_NOT_IN_HM(pPhysPage))
4825	{
4826	if (!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPhysPage))
4827	{
4828	if (SHW_PTE_IS_RW(PteDst))
4829	{
4830	AssertMsgFailed(("WRITE access flagged at %RGv but the page is writable! pPhysPage=%R[pgmpage] PteSrc=%#RX64 PteDst=%#RX64\n",
4831	GCPtr + off, pPhysPage, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4832	cErrors++;
4833	continue;
4834	}
4835	fIgnoreFlags \|= X86_PTE_RW;
4836	}
4837	else
4838	{
4839	if ( SHW_PTE_IS_P(PteDst)
4840	# if PGM_SHW_TYPE == PGM_TYPE_EPT \|\| PGM_SHW_TYPE == PGM_TYPE_PAE \|\| PGM_SHW_TYPE == PGM_TYPE_AMD64
4841	&& !PGM_PAGE_IS_MMIO(pPhysPage)
4842	# endif
4843	)
4844	{
4845	AssertMsgFailed(("ALL access flagged at %RGv but the page is present! pPhysPage=%R[pgmpage] PteSrc=%#RX64 PteDst=%#RX64\n",
4846	GCPtr + off, pPhysPage, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4847	cErrors++;
4848	continue;
4849	}
4850	fIgnoreFlags \|= X86_PTE_P;
4851	}
4852	}
4853	else
4854	{
4855	if ((PteSrc.u & (X86_PTE_RW \| X86_PTE_D)) == X86_PTE_RW)
4856	{
4857	if (SHW_PTE_IS_RW(PteDst))
4858	{
4859	AssertMsgFailed(("!DIRTY page at %RGv is writable! PteSrc=%#RX64 PteDst=%#RX64\n",
4860	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4861	cErrors++;
4862	continue;
4863	}
4864	if (!SHW_PTE_IS_TRACK_DIRTY(PteDst))
4865	{
4866	AssertMsgFailed(("!DIRTY page at %RGv is not marked TRACK_DIRTY! PteSrc=%#RX64 PteDst=%#RX64\n",
4867	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4868	cErrors++;
4869	continue;
4870	}
4871	if (SHW_PTE_IS_D(PteDst))
4872	{
4873	AssertMsgFailed(("!DIRTY page at %RGv is marked DIRTY! PteSrc=%#RX64 PteDst=%#RX64\n",
4874	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4875	cErrors++;
4876	}
4877	# if 0 /** @todo sync access bit properly... */
4878	if (PteDst.n.u1Accessed != PteSrc.n.u1Accessed)
4879	{
4880	AssertMsgFailed(("!DIRTY page at %RGv is has mismatching accessed bit! PteSrc=%#RX64 PteDst=%#RX64\n",
4881	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4882	cErrors++;
4883	}
4884	fIgnoreFlags \|= X86_PTE_RW;
4885	# else
4886	fIgnoreFlags \|= X86_PTE_RW \| X86_PTE_A;
4887	# endif
4888	}
4889	else if (SHW_PTE_IS_TRACK_DIRTY(PteDst))
4890	{
4891	/* access bit emulation (not implemented). */
4892	if ((PteSrc.u & X86_PTE_A) \|\| SHW_PTE_IS_P(PteDst))
4893	{
4894	AssertMsgFailed(("PGM_PTFLAGS_TRACK_DIRTY set at %RGv but no accessed bit emulation! PteSrc=%#RX64 PteDst=%#RX64\n",
4895	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4896	cErrors++;
4897	continue;
4898	}
4899	if (!SHW_PTE_IS_A(PteDst))
4900	{
4901	AssertMsgFailed(("!ACCESSED page at %RGv is has the accessed bit set! PteSrc=%#RX64 PteDst=%#RX64\n",
4902	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4903	cErrors++;
4904	}
4905	fIgnoreFlags \|= X86_PTE_P;
4906	}
4907	# ifdef DEBUG_sandervl
4908	fIgnoreFlags \|= X86_PTE_D \| X86_PTE_A;
4909	# endif
4910	}
4911
4912	if ( (PteSrc.u & ~fIgnoreFlags) != (SHW_PTE_GET_U(PteDst) & ~fIgnoreFlags)
4913	&& (PteSrc.u & ~(fIgnoreFlags \| X86_PTE_RW)) != (SHW_PTE_GET_U(PteDst) & ~fIgnoreFlags)
4914	)
4915	{
4916	AssertMsgFailed(("Flags mismatch at %RGv! %#RX64 != %#RX64 fIgnoreFlags=%#RX64 PteSrc=%#RX64 PteDst=%#RX64\n",
4917	GCPtr + off, (uint64_t)PteSrc.u & ~fIgnoreFlags, SHW_PTE_LOG64(PteDst) & ~fIgnoreFlags,
4918	fIgnoreFlags, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4919	cErrors++;
4920	continue;
4921	}
4922	} /* foreach PTE */
4923	}
4924	else
4925	{
4926	/*
4927	* Big Page.
4928	*/
4929	uint64_t fIgnoreFlags = X86_PDE_AVL_MASK \| GST_PDE_PG_MASK \| X86_PDE4M_G \| X86_PDE4M_D \| X86_PDE4M_PS \| X86_PDE4M_PWT \| X86_PDE4M_PCD;
4930	if ((PdeSrc.u & (X86_PDE_RW \| X86_PDE4M_D)) == X86_PDE_RW)
4931	{
4932	if (PdeDst.u & X86_PDE_RW)
4933	{
4934	AssertMsgFailed(("!DIRTY page at %RGv is writable! PdeSrc=%#RX64 PdeDst=%#RX64\n",
4935	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4936	cErrors++;
4937	continue;
4938	}
4939	if (!(PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY))
4940	{
4941	AssertMsgFailed(("!DIRTY page at %RGv is not marked TRACK_DIRTY! PteSrc=%#RX64 PteDst=%#RX64\n",
4942	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4943	cErrors++;
4944	continue;
4945	}
4946	# if 0 /** @todo sync access bit properly... */
4947	if (PdeDst.n.u1Accessed != PdeSrc.b.u1Accessed)
4948	{
4949	AssertMsgFailed(("!DIRTY page at %RGv is has mismatching accessed bit! PteSrc=%#RX64 PteDst=%#RX64\n",
4950	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4951	cErrors++;
4952	}
4953	fIgnoreFlags \|= X86_PTE_RW;
4954	# else
4955	fIgnoreFlags \|= X86_PTE_RW \| X86_PTE_A;
4956	# endif
4957	}
4958	else if (PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY)
4959	{
4960	/* access bit emulation (not implemented). */
4961	if ((PdeSrc.u & X86_PDE_A) \|\| SHW_PDE_IS_P(PdeDst))
4962	{
4963	AssertMsgFailed(("PGM_PDFLAGS_TRACK_DIRTY set at %RGv but no accessed bit emulation! PdeSrc=%#RX64 PdeDst=%#RX64\n",
4964	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4965	cErrors++;
4966	continue;
4967	}
4968	if (!SHW_PDE_IS_A(PdeDst))
4969	{
4970	AssertMsgFailed(("!ACCESSED page at %RGv is has the accessed bit set! PdeSrc=%#RX64 PdeDst=%#RX64\n",
4971	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4972	cErrors++;
4973	}
4974	fIgnoreFlags \|= X86_PTE_P;
4975	}
4976
4977	if ((PdeSrc.u & ~fIgnoreFlags) != (PdeDst.u & ~fIgnoreFlags))
4978	{
4979	AssertMsgFailed(("Flags mismatch (B) at %RGv! %#RX64 != %#RX64 fIgnoreFlags=%#RX64 PdeSrc=%#RX64 PdeDst=%#RX64\n",
4980	GCPtr, (uint64_t)PdeSrc.u & ~fIgnoreFlags, (uint64_t)PdeDst.u & ~fIgnoreFlags,
4981	fIgnoreFlags, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4982	cErrors++;
4983	}
4984
4985	/* iterate the page table. */
4986	for (unsigned iPT = 0, off = 0;
4987	iPT < RT_ELEMENTS(pPTDst->a);
4988	iPT++, off += GUEST_PAGE_SIZE, GCPhysGst = PGM_A20_APPLY(pVCpu, GCPhysGst + GUEST_PAGE_SIZE))
4989	{
4990	const SHWPTE PteDst = pPTDst->a[iPT];
4991
4992	if (SHW_PTE_IS_TRACK_DIRTY(PteDst))
4993	{
4994	AssertMsgFailed(("The PTE at %RGv emulating a 2/4M page is marked TRACK_DIRTY! PdeSrc=%#RX64 PteDst=%#RX64\n",
4995	GCPtr + off, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
4996	cErrors++;
4997	}
4998
4999	/* skip not-present entries. */
5000	if (!SHW_PTE_IS_P(PteDst)) /** @todo deal with ALL handlers and CSAM !P pages! */
5001	continue;
5002
5003	fIgnoreFlags = X86_PTE_PAE_PG_MASK \| X86_PTE_AVL_MASK \| X86_PTE_PWT \| X86_PTE_PCD \| X86_PTE_PAT \| X86_PTE_D \| X86_PTE_A \| X86_PTE_G \| X86_PTE_PAE_NX;
5004
5005	/* match the physical addresses */
5006	HCPhysShw = SHW_PTE_GET_HCPHYS(PteDst);
5007
5008	# ifdef IN_RING3
5009	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysGst, &HCPhys);
5010	if (RT_FAILURE(rc))
5011	{
5012	# if 0
5013	if (HCPhysShw != MMR3PageDummyHCPhys(pVM)) /** @todo this is wrong. */
5014	{
5015	AssertMsgFailed(("Cannot find guest physical address %RGp at %RGv! PdeSrc=%#RX64 PteDst=%#RX64\n",
5016	GCPhysGst, GCPtr + off, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
5017	cErrors++;
5018	}
5019	# endif
5020	}
5021	else if (HCPhysShw != (HCPhys & X86_PTE_PAE_PG_MASK))
5022	{
5023	AssertMsgFailed(("Out of sync (phys) at %RGv! HCPhysShw=%RHp HCPhys=%RHp GCPhysGst=%RGp PdeSrc=%#RX64 PteDst=%#RX64\n",
5024	GCPtr + off, HCPhysShw, HCPhys, GCPhysGst, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
5025	cErrors++;
5026	continue;
5027	}
5028	# endif
5029	pPhysPage = pgmPhysGetPage(pVM, GCPhysGst);
5030	if (!pPhysPage)
5031	{
5032	# if 0 /** @todo make MMR3PageDummyHCPhys an 'All' function! */
5033	if (HCPhysShw != MMR3PageDummyHCPhys(pVM)) /** @todo this is wrong. */
5034	{
5035	AssertMsgFailed(("Cannot find guest physical address %RGp at %RGv! PdeSrc=%#RX64 PteDst=%#RX64\n",
5036	GCPhysGst, GCPtr + off, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
5037	cErrors++;
5038	continue;
5039	}
5040	# endif
5041	if (SHW_PTE_IS_RW(PteDst))
5042	{
5043	AssertMsgFailed(("Invalid guest page at %RGv is writable! GCPhysGst=%RGp PdeSrc=%#RX64 PteDst=%#RX64\n",
5044	GCPtr + off, GCPhysGst, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
5045	cErrors++;
5046	}
5047	fIgnoreFlags \|= X86_PTE_RW;
5048	}
5049	else if (HCPhysShw != PGM_PAGE_GET_HCPHYS(pPhysPage))
5050	{
5051	AssertMsgFailed(("Out of sync (phys) at %RGv! HCPhysShw=%RHp pPhysPage=%R[pgmpage] GCPhysGst=%RGp PdeSrc=%#RX64 PteDst=%#RX64\n",
5052	GCPtr + off, HCPhysShw, pPhysPage, GCPhysGst, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
5053	cErrors++;
5054	continue;
5055	}
5056
5057	/* flags */
5058	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPhysPage))
5059	{
5060	if (!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPhysPage))
5061	{
5062	if (PGM_PAGE_GET_HNDL_PHYS_STATE(pPhysPage) != PGM_PAGE_HNDL_PHYS_STATE_DISABLED)
5063	{
5064	if ( SHW_PTE_IS_RW(PteDst)
5065	&& !PGM_PAGE_IS_HNDL_PHYS_NOT_IN_HM(pPhysPage))
5066	{
5067	AssertMsgFailed(("WRITE access flagged at %RGv but the page is writable! pPhysPage=%R[pgmpage] PdeSrc=%#RX64 PteDst=%#RX64\n",
5068	GCPtr + off, pPhysPage, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
5069	cErrors++;
5070	continue;
5071	}
5072	fIgnoreFlags \|= X86_PTE_RW;
5073	}
5074	}
5075	else
5076	{
5077	if ( SHW_PTE_IS_P(PteDst)
5078	&& !PGM_PAGE_IS_HNDL_PHYS_NOT_IN_HM(pPhysPage)
5079	# if PGM_SHW_TYPE == PGM_TYPE_EPT \|\| PGM_SHW_TYPE == PGM_TYPE_PAE \|\| PGM_SHW_TYPE == PGM_TYPE_AMD64
5080	&& !PGM_PAGE_IS_MMIO(pPhysPage)
5081	# endif
5082	)
5083	{
5084	AssertMsgFailed(("ALL access flagged at %RGv but the page is present! pPhysPage=%R[pgmpage] PdeSrc=%#RX64 PteDst=%#RX64\n",
5085	GCPtr + off, pPhysPage, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
5086	cErrors++;
5087	continue;
5088	}
5089	fIgnoreFlags \|= X86_PTE_P;
5090	}
5091	}
5092
5093	if ( (PdeSrc.u & ~fIgnoreFlags) != (SHW_PTE_GET_U(PteDst) & ~fIgnoreFlags)
5094	&& (PdeSrc.u & ~(fIgnoreFlags \| X86_PTE_RW)) != (SHW_PTE_GET_U(PteDst) & ~fIgnoreFlags) /* lazy phys handler dereg. */
5095	)
5096	{
5097	AssertMsgFailed(("Flags mismatch (BT) at %RGv! %#RX64 != %#RX64 fIgnoreFlags=%#RX64 PdeSrc=%#RX64 PteDst=%#RX64\n",
5098	GCPtr + off, (uint64_t)PdeSrc.u & ~fIgnoreFlags, SHW_PTE_LOG64(PteDst) & ~fIgnoreFlags,
5099	fIgnoreFlags, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
5100	cErrors++;
5101	continue;
5102	}
5103	} /* for each PTE */
5104	}
5105	}
5106	/* not present */
5107
5108	} /* for each PDE */
5109
5110	} /* for each PDPTE */
5111
5112	} /* for each PML4E */
5113
5114	# ifdef DEBUG
5115	if (cErrors)
5116	LogFlow(("AssertCR3: cErrors=%d\n", cErrors));
5117	# endif
5118	# endif /* GST is in {32BIT, PAE, AMD64} */
5119	return cErrors;
5120	#endif /* !PGM_TYPE_IS_NESTED_OR_EPT(PGM_SHW_TYPE) && PGM_SHW_TYPE != PGM_TYPE_NONE */
5121	}
5122	#endif /* VBOX_STRICT */
5123
5124
5125	/**
5126	* Sets up the CR3 for shadow paging
5127	*
5128	* @returns Strict VBox status code.
5129	* @retval VINF_SUCCESS.
5130	*
5131	* @param pVCpu The cross context virtual CPU structure.
5132	* @param GCPhysCR3 The physical address in the CR3 register. (A20 mask
5133	* already applied.)
5134	*/
5135	PGM_BTH_DECL(int, MapCR3)(PVMCPUCC pVCpu, RTGCPHYS GCPhysCR3)
5136	{
5137	PVMCC pVM = pVCpu->CTX_SUFF(pVM); NOREF(pVM);
5138	int rc = VINF_SUCCESS;
5139
5140	/* Update guest paging info. */
5141	#if PGM_GST_TYPE == PGM_TYPE_32BIT \
5142	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
5143	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64
5144
5145	LogFlow(("MapCR3: %RGp\n", GCPhysCR3));
5146	PGM_A20_ASSERT_MASKED(pVCpu, GCPhysCR3);
5147
5148	# if PGM_GST_TYPE == PGM_TYPE_PAE
5149	if ( !pVCpu->pgm.s.CTX_SUFF(fPaePdpesAndCr3Mapped)
5150	\|\| pVCpu->pgm.s.GCPhysPaeCR3 != GCPhysCR3)
5151	# endif
5152	{
5153	/*
5154	* Map the page CR3 points at.
5155	*/
5156	RTHCPTR HCPtrGuestCR3;
5157	rc = pgmGstMapCr3(pVCpu, GCPhysCR3, &HCPtrGuestCR3);
5158	if (RT_SUCCESS(rc))
5159	{
5160	# if PGM_GST_TYPE == PGM_TYPE_32BIT
5161	# ifdef IN_RING3
5162	pVCpu->pgm.s.pGst32BitPdR3 = (PX86PD)HCPtrGuestCR3;
5163	pVCpu->pgm.s.pGst32BitPdR0 = NIL_RTR0PTR;
5164	# else
5165	pVCpu->pgm.s.pGst32BitPdR3 = NIL_RTR3PTR;
5166	pVCpu->pgm.s.pGst32BitPdR0 = (PX86PD)HCPtrGuestCR3;
5167	# endif
5168
5169	# elif PGM_GST_TYPE == PGM_TYPE_PAE
5170	# ifdef IN_RING3
5171	pVCpu->pgm.s.pGstPaePdptR3 = (PX86PDPT)HCPtrGuestCR3;
5172	pVCpu->pgm.s.pGstPaePdptR0 = NIL_RTR0PTR;
5173	# else
5174	pVCpu->pgm.s.pGstPaePdptR3 = NIL_RTR3PTR;
5175	pVCpu->pgm.s.pGstPaePdptR0 = (PX86PDPT)HCPtrGuestCR3;
5176	# endif
5177
5178	X86PDPE aGstPaePdpes[X86_PG_PAE_PDPE_ENTRIES];
5179	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
5180	/*
5181	* When EPT is enabled by the nested-hypervisor and the nested-guest is in PAE mode,
5182	* the guest-CPU context would've already been updated with the 4 PAE PDPEs specified
5183	* in the virtual VMCS. The PDPEs can differ from those in guest memory referenced by
5184	* the translated nested-guest CR3. We -MUST- use the PDPEs provided in the virtual VMCS
5185	* rather than those in guest memory.
5186	*
5187	* See Intel spec. 26.3.2.4 "Loading Page-Directory-Pointer-Table Entries".
5188	*/
5189	if (pVCpu->pgm.s.enmGuestSlatMode == PGMSLAT_EPT)
5190	CPUMGetGuestPaePdpes(pVCpu, &aGstPaePdpes[0]);
5191	else
5192	#endif
5193	{
5194	/* Update CPUM with the PAE PDPEs referenced by CR3. */
5195	memcpy(&aGstPaePdpes, HCPtrGuestCR3, sizeof(aGstPaePdpes));
5196	CPUMSetGuestPaePdpes(pVCpu, &aGstPaePdpes[0]);
5197	}
5198
5199	/*
5200	* Map the 4 PAE PDPEs.
5201	*/
5202	rc = PGMGstMapPaePdpes(pVCpu, &aGstPaePdpes[0]);
5203	if (RT_SUCCESS(rc))
5204	{
5205	# ifdef IN_RING3
5206	pVCpu->pgm.s.fPaePdpesAndCr3MappedR3 = true;
5207	pVCpu->pgm.s.fPaePdpesAndCr3MappedR0 = false;
5208	# else
5209	pVCpu->pgm.s.fPaePdpesAndCr3MappedR3 = false;
5210	pVCpu->pgm.s.fPaePdpesAndCr3MappedR0 = true;
5211	# endif
5212	pVCpu->pgm.s.GCPhysPaeCR3 = GCPhysCR3;
5213	}
5214
5215	# elif PGM_GST_TYPE == PGM_TYPE_AMD64
5216	# ifdef IN_RING3
5217	pVCpu->pgm.s.pGstAmd64Pml4R3 = (PX86PML4)HCPtrGuestCR3;
5218	pVCpu->pgm.s.pGstAmd64Pml4R0 = NIL_RTR0PTR;
5219	# else
5220	pVCpu->pgm.s.pGstAmd64Pml4R3 = NIL_RTR3PTR;
5221	pVCpu->pgm.s.pGstAmd64Pml4R0 = (PX86PML4)HCPtrGuestCR3;
5222	# endif
5223	# endif
5224	}
5225	else
5226	AssertMsgFailed(("rc=%Rrc GCPhysGuestPD=%RGp\n", rc, GCPhysCR3));
5227	}
5228	#endif
5229
5230	/*
5231	* Update shadow paging info for guest modes with paging (32-bit, PAE, AMD64).
5232	*/
5233	# if ( ( PGM_SHW_TYPE == PGM_TYPE_32BIT \
5234	\|\| PGM_SHW_TYPE == PGM_TYPE_PAE \
5235	\|\| PGM_SHW_TYPE == PGM_TYPE_AMD64) \
5236	&& ( PGM_GST_TYPE != PGM_TYPE_REAL \
5237	&& PGM_GST_TYPE != PGM_TYPE_PROT))
5238
5239	Assert(!pVM->pgm.s.fNestedPaging);
5240	PGM_A20_ASSERT_MASKED(pVCpu, GCPhysCR3);
5241
5242	/*
5243	* Update the shadow root page as well since that's not fixed.
5244	*/
5245	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
5246	PPGMPOOLPAGE pOldShwPageCR3 = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
5247	PPGMPOOLPAGE pNewShwPageCR3;
5248
5249	PGM_LOCK_VOID(pVM);
5250
5251	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
5252	if (pPool->cDirtyPages)
5253	pgmPoolResetDirtyPages(pVM);
5254	# endif
5255
5256	Assert(!(GCPhysCR3 >> (GUEST_PAGE_SHIFT + 32))); /** @todo what is this for? */
5257	int const rc2 = pgmPoolAlloc(pVM, GCPhysCR3 & GST_CR3_PAGE_MASK, BTH_PGMPOOLKIND_ROOT, PGMPOOLACCESS_DONTCARE,
5258	PGM_A20_IS_ENABLED(pVCpu), NIL_PGMPOOL_IDX, UINT32_MAX, true /fLockPage/, &pNewShwPageCR3);
5259	AssertFatalRC(rc2);
5260
5261	pVCpu->pgm.s.pShwPageCR3R3 = pgmPoolConvertPageToR3(pPool, pNewShwPageCR3);
5262	pVCpu->pgm.s.pShwPageCR3R0 = pgmPoolConvertPageToR0(pPool, pNewShwPageCR3);
5263
5264	/* Set the current hypervisor CR3. */
5265	CPUMSetHyperCR3(pVCpu, PGMGetHyperCR3(pVCpu));
5266
5267	/* Clean up the old CR3 root. */
5268	if ( pOldShwPageCR3
5269	&& pOldShwPageCR3 != pNewShwPageCR3 /* @todo can happen due to incorrect syncing between REM & PGM; find the real cause */)
5270	{
5271	Assert(pOldShwPageCR3->enmKind != PGMPOOLKIND_FREE);
5272
5273	/* Mark the page as unlocked; allow flushing again. */
5274	pgmPoolUnlockPage(pPool, pOldShwPageCR3);
5275
5276	pgmPoolFreeByPage(pPool, pOldShwPageCR3, NIL_PGMPOOL_IDX, UINT32_MAX);
5277	}
5278	PGM_UNLOCK(pVM);
5279	# else
5280	NOREF(GCPhysCR3);
5281	# endif
5282
5283	return rc;
5284	}
5285
5286	/**
5287	* Unmaps the shadow CR3.
5288	*
5289	* @returns VBox status, no specials.
5290	* @param pVCpu The cross context virtual CPU structure.
5291	*/
5292	PGM_BTH_DECL(int, UnmapCR3)(PVMCPUCC pVCpu)
5293	{
5294	LogFlow(("UnmapCR3\n"));
5295
5296	int rc = VINF_SUCCESS;
5297	PVMCC pVM = pVCpu->CTX_SUFF(pVM); NOREF(pVM);
5298
5299	/*
5300	* Update guest paging info.
5301	*/
5302	#if PGM_GST_TYPE == PGM_TYPE_32BIT
5303	pVCpu->pgm.s.pGst32BitPdR3 = 0;
5304	pVCpu->pgm.s.pGst32BitPdR0 = 0;
5305
5306	#elif PGM_GST_TYPE == PGM_TYPE_PAE
5307	pVCpu->pgm.s.pGstPaePdptR3 = 0;
5308	pVCpu->pgm.s.pGstPaePdptR0 = 0;
5309	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
5310	{
5311	pVCpu->pgm.s.apGstPaePDsR3[i] = 0;
5312	pVCpu->pgm.s.apGstPaePDsR0[i] = 0;
5313	pVCpu->pgm.s.aGCPhysGstPaePDs[i] = NIL_RTGCPHYS;
5314	}
5315
5316	#elif PGM_GST_TYPE == PGM_TYPE_AMD64
5317	pVCpu->pgm.s.pGstAmd64Pml4R3 = 0;
5318	pVCpu->pgm.s.pGstAmd64Pml4R0 = 0;
5319
5320	#else /* prot/real mode stub */
5321	/* nothing to do */
5322	#endif
5323
5324	/*
5325	* PAE PDPEs (and CR3) might have been mapped via PGMGstMapPaePdpesAtCr3()
5326	* prior to switching to PAE in pfnMapCr3(), so we need to clear them here.
5327	*/
5328	pVCpu->pgm.s.fPaePdpesAndCr3MappedR3 = false;
5329	pVCpu->pgm.s.fPaePdpesAndCr3MappedR0 = false;
5330	pVCpu->pgm.s.GCPhysPaeCR3 = NIL_RTGCPHYS;
5331
5332	/*
5333	* Update shadow paging info.
5334	*/
5335	#if ( ( PGM_SHW_TYPE == PGM_TYPE_32BIT \
5336	\|\| PGM_SHW_TYPE == PGM_TYPE_PAE \
5337	\|\| PGM_SHW_TYPE == PGM_TYPE_AMD64))
5338	# if PGM_GST_TYPE != PGM_TYPE_REAL
5339	Assert(!pVM->pgm.s.fNestedPaging);
5340	# endif
5341	PGM_LOCK_VOID(pVM);
5342
5343	if (pVCpu->pgm.s.CTX_SUFF(pShwPageCR3))
5344	{
5345	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
5346
5347	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
5348	if (pPool->cDirtyPages)
5349	pgmPoolResetDirtyPages(pVM);
5350	# endif
5351
5352	/* Mark the page as unlocked; allow flushing again. */
5353	pgmPoolUnlockPage(pPool, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
5354
5355	pgmPoolFreeByPage(pPool, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3), NIL_PGMPOOL_IDX, UINT32_MAX);
5356	pVCpu->pgm.s.pShwPageCR3R3 = 0;
5357	pVCpu->pgm.s.pShwPageCR3R0 = 0;
5358	}
5359
5360	PGM_UNLOCK(pVM);
5361	#endif
5362
5363	return rc;
5364	}
5365

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllBth.h@ 107044

Download in other formats: