PGMAllBth.h@ 29888

Last change on this file since 29888 was 29676, checked in by vboxsync, 15 years ago
Too noisy
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 199.9 KB

Line
1	/* $Id: PGMAllBth.h 29676 2010-05-20 09:54:57Z vboxsync $ */
2	/** @file
3	* VBox - Page Manager, Shadow+Guest Paging Template - All context code.
4	*
5	* This file is a big challenge!
6	*/
7
8	/*
9	* Copyright (C) 2006-2007 Oracle Corporation
10	*
11	* This file is part of VirtualBox Open Source Edition (OSE), as
12	* available from http://www.virtualbox.org. This file is free software;
13	* you can redistribute it and/or modify it under the terms of the GNU
14	* General Public License (GPL) as published by the Free Software
15	* Foundation, in version 2 as it comes in the "COPYING" file of the
16	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
17	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
18	*/
19
20	/*******************************************************************************
21	* Internal Functions *
22	*******************************************************************************/
23	RT_C_DECLS_BEGIN
24	PGM_BTH_DECL(int, Trap0eHandler)(PVMCPU pVCpu, RTGCUINT uErr, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, bool *pfLockTaken);
25	PGM_BTH_DECL(int, InvalidatePage)(PVMCPU pVCpu, RTGCPTR GCPtrPage);
26	PGM_BTH_DECL(int, SyncPage)(PVMCPU pVCpu, GSTPDE PdeSrc, RTGCPTR GCPtrPage, unsigned cPages, unsigned uErr);
27	PGM_BTH_DECL(int, CheckPageFault)(PVMCPU pVCpu, uint32_t uErr, PGSTPDE pPdeSrc, RTGCPTR GCPtrPage);
28	PGM_BTH_DECL(int, CheckDirtyPageFault)(PVMCPU pVCpu, uint32_t uErr, PSHWPDE pPdeDst, PGSTPDE pPdeSrc, RTGCPTR GCPtrPage);
29	PGM_BTH_DECL(int, SyncPT)(PVMCPU pVCpu, unsigned iPD, PGSTPD pPDSrc, RTGCPTR GCPtrPage);
30	PGM_BTH_DECL(int, VerifyAccessSyncPage)(PVMCPU pVCpu, RTGCPTR Addr, unsigned fPage, unsigned uErr);
31	PGM_BTH_DECL(int, PrefetchPage)(PVMCPU pVCpu, RTGCPTR GCPtrPage);
32	PGM_BTH_DECL(int, SyncCR3)(PVMCPU pVCpu, uint64_t cr0, uint64_t cr3, uint64_t cr4, bool fGlobal);
33	#ifdef VBOX_STRICT
34	PGM_BTH_DECL(unsigned, AssertCR3)(PVMCPU pVCpu, uint64_t cr3, uint64_t cr4, RTGCPTR GCPtr = 0, RTGCPTR cb = ~(RTGCPTR)0);
35	#endif
36	DECLINLINE(void) PGM_BTH_NAME(SyncPageWorkerTrackDeref)(PVMCPU pVCpu, PPGMPOOLPAGE pShwPage, RTHCPHYS HCPhys, uint16_t iPte);
37	PGM_BTH_DECL(int, MapCR3)(PVMCPU pVCpu, RTGCPHYS GCPhysCR3);
38	PGM_BTH_DECL(int, UnmapCR3)(PVMCPU pVCpu);
39	RT_C_DECLS_END
40
41
42	/* Filter out some illegal combinations of guest and shadow paging, so we can remove redundant checks inside functions. */
43	#if PGM_GST_TYPE == PGM_TYPE_PAE && PGM_SHW_TYPE != PGM_TYPE_PAE && PGM_SHW_TYPE != PGM_TYPE_NESTED && PGM_SHW_TYPE != PGM_TYPE_EPT
44	# error "Invalid combination; PAE guest implies PAE shadow"
45	#endif
46
47	#if (PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
48	&& !(PGM_SHW_TYPE == PGM_TYPE_32BIT \|\| PGM_SHW_TYPE == PGM_TYPE_PAE \|\| PGM_SHW_TYPE == PGM_TYPE_AMD64 \|\| PGM_SHW_TYPE == PGM_TYPE_NESTED \|\| PGM_SHW_TYPE == PGM_TYPE_EPT)
49	# error "Invalid combination; real or protected mode without paging implies 32 bits or PAE shadow paging."
50	#endif
51
52	#if (PGM_GST_TYPE == PGM_TYPE_32BIT \|\| PGM_GST_TYPE == PGM_TYPE_PAE) \
53	&& !(PGM_SHW_TYPE == PGM_TYPE_32BIT \|\| PGM_SHW_TYPE == PGM_TYPE_PAE \|\| PGM_SHW_TYPE == PGM_TYPE_NESTED \|\| PGM_SHW_TYPE == PGM_TYPE_EPT)
54	# error "Invalid combination; 32 bits guest paging or PAE implies 32 bits or PAE shadow paging."
55	#endif
56
57	#if (PGM_GST_TYPE == PGM_TYPE_AMD64 && PGM_SHW_TYPE != PGM_TYPE_AMD64 && PGM_SHW_TYPE != PGM_TYPE_NESTED && PGM_SHW_TYPE != PGM_TYPE_EPT) \
58	\|\| (PGM_SHW_TYPE == PGM_TYPE_AMD64 && PGM_GST_TYPE != PGM_TYPE_AMD64 && PGM_GST_TYPE != PGM_TYPE_PROT)
59	# error "Invalid combination; AMD64 guest implies AMD64 shadow and vice versa"
60	#endif
61
62
63	#ifndef IN_RING3
64	/**
65	* #PF Handler for raw-mode guest execution.
66	*
67	* @returns VBox status code (appropriate for trap handling and GC return).
68	*
69	* @param pVCpu VMCPU Handle.
70	* @param uErr The trap error code.
71	* @param pRegFrame Trap register frame.
72	* @param pvFault The fault address.
73	* @param pfLockTaken PGM lock taken here or not (out)
74	*/
75	PGM_BTH_DECL(int, Trap0eHandler)(PVMCPU pVCpu, RTGCUINT uErr, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, bool *pfLockTaken)
76	{
77	PVM pVM = pVCpu->CTX_SUFF(pVM);
78
79	*pfLockTaken = false;
80
81	# if defined(IN_RC) && defined(VBOX_STRICT)
82	PGMDynCheckLocks(pVM);
83	# endif
84
85	# if (PGM_GST_TYPE == PGM_TYPE_32BIT \|\| PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT \|\| PGM_GST_TYPE == PGM_TYPE_PAE \|\| PGM_GST_TYPE == PGM_TYPE_AMD64) \
86	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
87	&& (PGM_SHW_TYPE != PGM_TYPE_EPT \|\| PGM_GST_TYPE == PGM_TYPE_PROT)
88
89	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE != PGM_TYPE_PAE
90	/*
91	* Hide the instruction fetch trap indicator for now.
92	*/
93	/** @todo NXE will change this and we must fix NXE in the switcher too! */
94	if (uErr & X86_TRAP_PF_ID)
95	{
96	uErr &= ~X86_TRAP_PF_ID;
97	TRPMSetErrorCode(pVCpu, uErr);
98	}
99	# endif
100
101	/*
102	* Get PDs.
103	*/
104	int rc;
105	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
106	# if PGM_GST_TYPE == PGM_TYPE_32BIT
107	const unsigned iPDSrc = pvFault >> GST_PD_SHIFT;
108	PGSTPD pPDSrc = pgmGstGet32bitPDPtr(&pVCpu->pgm.s);
109
110	# elif PGM_GST_TYPE == PGM_TYPE_PAE \|\| PGM_GST_TYPE == PGM_TYPE_AMD64
111
112	# if PGM_GST_TYPE == PGM_TYPE_PAE
113	unsigned iPDSrc = 0; /* initialized to shut up gcc */
114	X86PDPE PdpeSrc;
115	PGSTPD pPDSrc = pgmGstGetPaePDPtr(&pVCpu->pgm.s, pvFault, &iPDSrc, &PdpeSrc);
116
117	# elif PGM_GST_TYPE == PGM_TYPE_AMD64
118	unsigned iPDSrc = 0; /* initialized to shut up gcc */
119	PX86PML4E pPml4eSrc;
120	X86PDPE PdpeSrc;
121	PGSTPD pPDSrc;
122
123	pPDSrc = pgmGstGetLongModePDPtr(&pVCpu->pgm.s, pvFault, &pPml4eSrc, &PdpeSrc, &iPDSrc);
124	Assert(pPml4eSrc);
125	# endif
126
127	/* Quick check for a valid guest trap. (PAE & AMD64) */
128	if (!pPDSrc)
129	{
130	# if PGM_GST_TYPE == PGM_TYPE_AMD64 && GC_ARCH_BITS == 64
131	LogFlow(("Trap0eHandler: guest PML4 %d not present CR3=%RGp\n", (int)((pvFault >> X86_PML4_SHIFT) & X86_PML4_MASK), CPUMGetGuestCR3(pVCpu) & X86_CR3_PAGE_MASK));
132	# else
133	LogFlow(("Trap0eHandler: guest iPDSrc=%u not present CR3=%RGp\n", iPDSrc, CPUMGetGuestCR3(pVCpu) & X86_CR3_PAGE_MASK));
134	# endif
135	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2GuestTrap; });
136	TRPMSetErrorCode(pVCpu, uErr);
137	return VINF_EM_RAW_GUEST_TRAP;
138	}
139	# endif
140
141	# else /* !PGM_WITH_PAGING */
142	PGSTPD pPDSrc = NULL;
143	const unsigned iPDSrc = 0;
144	# endif /* !PGM_WITH_PAGING */
145
146	# if !defined(PGM_WITHOUT_MAPPINGS) && ((PGM_GST_TYPE == PGM_TYPE_32BIT) \|\| (PGM_GST_TYPE == PGM_TYPE_PAE))
147	/*
148	* Check for write conflicts with our hypervisor mapping early on. If the guest happens to access a non-present page,
149	* where our hypervisor is currently mapped, then we'll create a #PF storm in the guest.
150	*/
151	if ( (uErr & (X86_TRAP_PF_P \| X86_TRAP_PF_RW)) == (X86_TRAP_PF_P \| X86_TRAP_PF_RW)
152	&& MMHyperIsInsideArea(pVM, pvFault))
153	{
154	/* Force a CR3 sync to check for conflicts and emulate the instruction. */
155	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
156	return VINF_EM_RAW_EMULATE_INSTR;
157	}
158	# endif
159
160	/* First check for a genuine guest page fault. */
161	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
162	STAM_PROFILE_START(&pVCpu->pgm.s.StatRZTrap0eTimeCheckPageFault, e);
163	rc = PGM_BTH_NAME(CheckPageFault)(pVCpu, uErr, &pPDSrc->a[iPDSrc], pvFault);
164	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeCheckPageFault, e);
165	if (rc == VINF_EM_RAW_GUEST_TRAP)
166	{
167	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution)
168	= &pVCpu->pgm.s.StatRZTrap0eTime2GuestTrap; });
169	return rc;
170	}
171	# endif /* PGM_WITH_PAGING */
172
173	/* Take the big lock now. */
174	*pfLockTaken = true;
175	pgmLock(pVM);
176
177	/* Fetch the guest PDE */
178	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
179	GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
180	# else
181	GSTPDE PdeSrc;
182	PdeSrc.u = 0; /* faked so we don't have to #ifdef everything */
183	PdeSrc.n.u1Present = 1;
184	PdeSrc.n.u1Write = 1;
185	PdeSrc.n.u1Accessed = 1;
186	PdeSrc.n.u1User = 1;
187	# endif
188
189	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
190	const unsigned iPDDst = pvFault >> SHW_PD_SHIFT;
191	PX86PD pPDDst = pgmShwGet32BitPDPtr(&pVCpu->pgm.s);
192
193	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
194	const unsigned iPDDst = (pvFault >> SHW_PD_SHIFT) & SHW_PD_MASK; /* pPDDst index, not used with the pool. */
195
196	PX86PDPAE pPDDst;
197	# if PGM_GST_TYPE != PGM_TYPE_PAE
198	X86PDPE PdpeSrc;
199
200	/* Fake PDPT entry; access control handled on the page table level, so allow everything. */
201	PdpeSrc.u = X86_PDPE_P; /* rw/us are reserved for PAE pdpte's; accessed bit causes invalid VT-x guest state errors */
202	# endif
203	rc = pgmShwSyncPaePDPtr(pVCpu, pvFault, &PdpeSrc, &pPDDst);
204	if (rc != VINF_SUCCESS)
205	{
206	AssertRC(rc);
207	return rc;
208	}
209	Assert(pPDDst);
210
211	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
212	const unsigned iPDDst = ((pvFault >> SHW_PD_SHIFT) & SHW_PD_MASK);
213	PX86PDPAE pPDDst;
214	# if PGM_GST_TYPE == PGM_TYPE_PROT
215	/* AMD-V nested paging */
216	X86PML4E Pml4eSrc;
217	X86PDPE PdpeSrc;
218	PX86PML4E pPml4eSrc = &Pml4eSrc;
219
220	/* Fake PML4 & PDPT entry; access control handled on the page table level, so allow everything. */
221	Pml4eSrc.u = X86_PML4E_P \| X86_PML4E_RW \| X86_PML4E_US \| X86_PML4E_A;
222	PdpeSrc.u = X86_PDPE_P \| X86_PDPE_RW \| X86_PDPE_US \| X86_PDPE_A;
223	# endif
224
225	rc = pgmShwSyncLongModePDPtr(pVCpu, pvFault, pPml4eSrc, &PdpeSrc, &pPDDst);
226	if (rc != VINF_SUCCESS)
227	{
228	AssertRC(rc);
229	return rc;
230	}
231	Assert(pPDDst);
232
233	# elif PGM_SHW_TYPE == PGM_TYPE_EPT
234	const unsigned iPDDst = ((pvFault >> SHW_PD_SHIFT) & SHW_PD_MASK);
235	PEPTPD pPDDst;
236
237	rc = pgmShwGetEPTPDPtr(pVCpu, pvFault, NULL, &pPDDst);
238	if (rc != VINF_SUCCESS)
239	{
240	AssertRC(rc);
241	return rc;
242	}
243	Assert(pPDDst);
244	# endif
245
246	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
247	/* Dirty page handling. */
248	if (uErr & X86_TRAP_PF_RW) /* write fault? */
249	{
250	/*
251	* If we successfully correct the write protection fault due to dirty bit
252	* tracking, then return immediately.
253	*/
254	STAM_PROFILE_START(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyBitTracking), a);
255	rc = PGM_BTH_NAME(CheckDirtyPageFault)(pVCpu, uErr, &pPDDst->a[iPDDst], &pPDSrc->a[iPDSrc], pvFault);
256	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyBitTracking), a);
257	if (rc == VINF_PGM_HANDLED_DIRTY_BIT_FAULT)
258	{
259	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution)
260	= rc == VINF_PGM_HANDLED_DIRTY_BIT_FAULT ? &pVCpu->pgm.s.StatRZTrap0eTime2DirtyAndAccessed : &pVCpu->pgm.s.StatRZTrap0eTime2GuestTrap; });
261	LogBird(("Trap0eHandler: returns VINF_SUCCESS\n"));
262	return VINF_SUCCESS;
263	}
264	}
265
266	# if 0 /* rarely useful; leave for debugging. */
267	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0ePD[iPDSrc]);
268	# endif
269	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
270
271	/*
272	* A common case is the not-present error caused by lazy page table syncing.
273	*
274	* It is IMPORTANT that we weed out any access to non-present shadow PDEs here
275	* so we can safely assume that the shadow PT is present when calling SyncPage later.
276	*
277	* On failure, we ASSUME that SyncPT is out of memory or detected some kind
278	* of mapping conflict and defer to SyncCR3 in R3.
279	* (Again, we do NOT support access handlers for non-present guest pages.)
280	*
281	*/
282	Assert(PdeSrc.n.u1Present);
283	if ( !(uErr & X86_TRAP_PF_P) /* not set means page not present instead of page protection violation */
284	&& !pPDDst->a[iPDDst].n.u1Present
285	)
286	{
287	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2SyncPT; });
288	STAM_PROFILE_START(&pVCpu->pgm.s.StatRZTrap0eTimeSyncPT, f);
289	LogFlow(("=>SyncPT %04x = %08x\n", iPDSrc, PdeSrc.au32[0]));
290	rc = PGM_BTH_NAME(SyncPT)(pVCpu, iPDSrc, pPDSrc, pvFault);
291	if (RT_SUCCESS(rc))
292	{
293	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeSyncPT, f);
294	return rc;
295	}
296	Log(("SyncPT: %d failed!! rc=%d\n", iPDSrc, rc));
297	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
298	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeSyncPT, f);
299	return VINF_PGM_SYNC_CR3;
300	}
301
302	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && !defined(PGM_WITHOUT_MAPPINGS)
303	/*
304	* Check if this address is within any of our mappings.
305	*
306	* This is very fast and it's gonna save us a bit of effort below and prevent
307	* us from screwing ourself with MMIO2 pages which have a GC Mapping (VRam).
308	* (BTW, it's impossible to have physical access handlers in a mapping.)
309	*/
310	if (pgmMapAreMappingsEnabled(&pVM->pgm.s))
311	{
312	STAM_PROFILE_START(&pVCpu->pgm.s.StatRZTrap0eTimeMapping, a);
313	PPGMMAPPING pMapping = pVM->pgm.s.CTX_SUFF(pMappings);
314	for ( ; pMapping; pMapping = pMapping->CTX_SUFF(pNext))
315	{
316	if (pvFault < pMapping->GCPtr)
317	break;
318	if (pvFault - pMapping->GCPtr < pMapping->cb)
319	{
320	/*
321	* The first thing we check is if we've got an undetected conflict.
322	*/
323	if (pgmMapAreMappingsFloating(&pVM->pgm.s))
324	{
325	unsigned iPT = pMapping->cb >> GST_PD_SHIFT;
326	while (iPT-- > 0)
327	if (pPDSrc->a[iPDSrc + iPT].n.u1Present)
328	{
329	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eConflicts);
330	Log(("Trap0e: Detected Conflict %RGv-%RGv\n", pMapping->GCPtr, pMapping->GCPtrLast));
331	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3); /** @todo no need to do global sync,right? */
332	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeMapping, a);
333	return VINF_PGM_SYNC_CR3;
334	}
335	}
336
337	/*
338	* Check if the fault address is in a virtual page access handler range.
339	*/
340	PPGMVIRTHANDLER pCur = (PPGMVIRTHANDLER)RTAvlroGCPtrRangeGet(&pVM->pgm.s.CTX_SUFF(pTrees)->HyperVirtHandlers, pvFault);
341	if ( pCur
342	&& pvFault - pCur->Core.Key < pCur->cb
343	&& uErr & X86_TRAP_PF_RW)
344	{
345	# ifdef IN_RC
346	STAM_PROFILE_START(&pCur->Stat, h);
347	pgmUnlock(pVM);
348	rc = pCur->CTX_SUFF(pfnHandler)(pVM, uErr, pRegFrame, pvFault, pCur->Core.Key, pvFault - pCur->Core.Key);
349	pgmLock(pVM);
350	STAM_PROFILE_STOP(&pCur->Stat, h);
351	# else
352	AssertFailed();
353	rc = VINF_EM_RAW_EMULATE_INSTR; /* can't happen with VMX */
354	# endif
355	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eHandlersMapping);
356	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeMapping, a);
357	return rc;
358	}
359
360	/*
361	* Pretend we're not here and let the guest handle the trap.
362	*/
363	TRPMSetErrorCode(pVCpu, uErr & ~X86_TRAP_PF_P);
364	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eGuestPFMapping);
365	LogFlow(("PGM: Mapping access -> route trap to recompiler!\n"));
366	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeMapping, a);
367	return VINF_EM_RAW_GUEST_TRAP;
368	}
369	}
370	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeMapping, a);
371	} /* pgmAreMappingsEnabled(&pVM->pgm.s) */
372	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
373
374	/*
375	* Check if this fault address is flagged for special treatment,
376	* which means we'll have to figure out the physical address and
377	* check flags associated with it.
378	*
379	* ASSUME that we can limit any special access handling to pages
380	* in page tables which the guest believes to be present.
381	*/
382	Assert(PdeSrc.n.u1Present);
383	{
384	RTGCPHYS GCPhys = NIL_RTGCPHYS;
385
386	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
387	if ( PdeSrc.b.u1Size
388	# if PGM_GST_TYPE == PGM_TYPE_32BIT
389	&& CPUMIsGuestPageSizeExtEnabled(pVCpu)
390	# endif
391	)
392	GCPhys = GST_GET_PDE_BIG_PG_GCPHYS(PdeSrc)
393	\| ((RTGCPHYS)pvFault & (GST_BIG_PAGE_OFFSET_MASK ^ PAGE_OFFSET_MASK));
394	else
395	{
396	PGSTPT pPTSrc;
397	rc = PGM_GCPHYS_2_PTR(pVM, PdeSrc.u & GST_PDE_PG_MASK, &pPTSrc);
398	if (RT_SUCCESS(rc))
399	{
400	unsigned iPTESrc = (pvFault >> GST_PT_SHIFT) & GST_PT_MASK;
401	if (pPTSrc->a[iPTESrc].n.u1Present)
402	GCPhys = pPTSrc->a[iPTESrc].u & GST_PTE_PG_MASK;
403	}
404	}
405	# else
406	/* No paging so the fault address is the physical address */
407	GCPhys = (RTGCPHYS)(pvFault & ~PAGE_OFFSET_MASK);
408	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
409
410	/*
411	* If we have a GC address we'll check if it has any flags set.
412	*/
413	if (GCPhys != NIL_RTGCPHYS)
414	{
415	STAM_PROFILE_START(&pVCpu->pgm.s.StatRZTrap0eTimeHandlers, b);
416
417	PPGMPAGE pPage;
418	rc = pgmPhysGetPageEx(&pVM->pgm.s, GCPhys, &pPage);
419	if (RT_SUCCESS(rc)) /** just handle the failure immediate (it returns) and make things easier to read. */
420	{
421	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
422	{
423	if (PGM_PAGE_HAS_ANY_PHYSICAL_HANDLERS(pPage))
424	{
425	/*
426	* Physical page access handler.
427	*/
428	const RTGCPHYS GCPhysFault = GCPhys \| (pvFault & PAGE_OFFSET_MASK);
429	PPGMPHYSHANDLER pCur = (PPGMPHYSHANDLER)RTAvlroGCPhysRangeGet(&pVM->pgm.s.CTX_SUFF(pTrees)->PhysHandlers, GCPhysFault);
430	if (pCur)
431	{
432	# ifdef PGM_SYNC_N_PAGES
433	/*
434	* If the region is write protected and we got a page not present fault, then sync
435	* the pages. If the fault was caused by a read, then restart the instruction.
436	* In case of write access continue to the GC write handler.
437	*
438	* ASSUMES that there is only one handler per page or that they have similar write properties.
439	*/
440	if ( pCur->enmType == PGMPHYSHANDLERTYPE_PHYSICAL_WRITE
441	&& !(uErr & X86_TRAP_PF_P))
442	{
443	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, pvFault, PGM_SYNC_NR_PAGES, uErr);
444	if ( RT_FAILURE(rc)
445	\|\| !(uErr & X86_TRAP_PF_RW)
446	\|\| rc == VINF_PGM_SYNCPAGE_MODIFIED_PDE)
447	{
448	AssertRC(rc);
449	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eHandlersOutOfSync);
450	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeHandlers, b);
451	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2OutOfSyncHndPhys; });
452	return rc;
453	}
454	}
455	# endif
456
457	AssertMsg( pCur->enmType != PGMPHYSHANDLERTYPE_PHYSICAL_WRITE
458	\|\| (pCur->enmType == PGMPHYSHANDLERTYPE_PHYSICAL_WRITE && (uErr & X86_TRAP_PF_RW)),
459	("Unexpected trap for physical handler: %08X (phys=%08x) pPage=%R[pgmpage] uErr=%X, enum=%d\n", pvFault, GCPhys, pPage, uErr, pCur->enmType));
460
461	# if defined(IN_RC) \|\| defined(IN_RING0)
462	if (pCur->CTX_SUFF(pfnHandler))
463	{
464	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
465	# ifdef IN_RING0
466	PFNPGMR0PHYSHANDLER pfnHandler = pCur->CTX_SUFF(pfnHandler);
467	# else
468	PFNPGMRCPHYSHANDLER pfnHandler = pCur->CTX_SUFF(pfnHandler);
469	# endif
470	bool fLeaveLock = (pfnHandler != pPool->CTX_SUFF(pfnAccessHandler));
471	void *pvUser = pCur->CTX_SUFF(pvUser);
472
473	STAM_PROFILE_START(&pCur->Stat, h);
474	if (fLeaveLock)
475	pgmUnlock(pVM); /* @todo: Not entirely safe. */
476
477	rc = pfnHandler(pVM, uErr, pRegFrame, pvFault, GCPhysFault, pvUser);
478	if (fLeaveLock)
479	pgmLock(pVM);
480	# ifdef VBOX_WITH_STATISTICS
481	pCur = (PPGMPHYSHANDLER)RTAvlroGCPhysRangeGet(&pVM->pgm.s.CTX_SUFF(pTrees)->PhysHandlers, GCPhysFault);
482	if (pCur)
483	STAM_PROFILE_STOP(&pCur->Stat, h);
484	# else
485	pCur = NULL; /* might be invalid by now. */
486	# endif
487
488	}
489	else
490	# endif
491	rc = VINF_EM_RAW_EMULATE_INSTR;
492
493	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eHandlersPhysical);
494	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeHandlers, b);
495	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2HndPhys; });
496	return rc;
497	}
498	}
499	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
500	else
501	{
502	# ifdef PGM_SYNC_N_PAGES
503	/*
504	* If the region is write protected and we got a page not present fault, then sync
505	* the pages. If the fault was caused by a read, then restart the instruction.
506	* In case of write access continue to the GC write handler.
507	*/
508	if ( PGM_PAGE_GET_HNDL_VIRT_STATE(pPage) < PGM_PAGE_HNDL_PHYS_STATE_ALL
509	&& !(uErr & X86_TRAP_PF_P))
510	{
511	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, pvFault, PGM_SYNC_NR_PAGES, uErr);
512	if ( RT_FAILURE(rc)
513	\|\| rc == VINF_PGM_SYNCPAGE_MODIFIED_PDE
514	\|\| !(uErr & X86_TRAP_PF_RW))
515	{
516	AssertRC(rc);
517	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eHandlersOutOfSync);
518	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeHandlers, b);
519	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2OutOfSyncHndVirt; });
520	return rc;
521	}
522	}
523	# endif
524	/*
525	* Ok, it's an virtual page access handler.
526	*
527	* Since it's faster to search by address, we'll do that first
528	* and then retry by GCPhys if that fails.
529	*/
530	/** @todo r=bird: perhaps we should consider looking up by physical address directly now? */
531	/** @note r=svl: true, but lookup on virtual address should remain as a fallback as phys & virt trees might be out of sync, because the
532	* page was changed without us noticing it (not-present -> present without invlpg or mov cr3, xxx)
533	*/
534	PPGMVIRTHANDLER pCur = (PPGMVIRTHANDLER)RTAvlroGCPtrRangeGet(&pVM->pgm.s.CTX_SUFF(pTrees)->VirtHandlers, pvFault);
535	if (pCur)
536	{
537	AssertMsg(!(pvFault - pCur->Core.Key < pCur->cb)
538	\|\| ( pCur->enmType != PGMVIRTHANDLERTYPE_WRITE
539	\|\| !(uErr & X86_TRAP_PF_P)
540	\|\| (pCur->enmType == PGMVIRTHANDLERTYPE_WRITE && (uErr & X86_TRAP_PF_RW))),
541	("Unexpected trap for virtual handler: %RGv (phys=%RGp) pPage=%R[pgmpage] uErr=%X, enum=%d\n", pvFault, GCPhys, pPage, uErr, pCur->enmType));
542
543	if ( pvFault - pCur->Core.Key < pCur->cb
544	&& ( uErr & X86_TRAP_PF_RW
545	\|\| pCur->enmType != PGMVIRTHANDLERTYPE_WRITE ) )
546	{
547	# ifdef IN_RC
548	STAM_PROFILE_START(&pCur->Stat, h);
549	pgmUnlock(pVM);
550	rc = pCur->CTX_SUFF(pfnHandler)(pVM, uErr, pRegFrame, pvFault, pCur->Core.Key, pvFault - pCur->Core.Key);
551	pgmLock(pVM);
552	STAM_PROFILE_STOP(&pCur->Stat, h);
553	# else
554	rc = VINF_EM_RAW_EMULATE_INSTR; /** @todo for VMX */
555	# endif
556	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eHandlersVirtual);
557	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeHandlers, b);
558	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2HndVirt; });
559	return rc;
560	}
561	/* Unhandled part of a monitored page */
562	}
563	else
564	{
565	/* Check by physical address. */
566	unsigned iPage;
567	rc = pgmHandlerVirtualFindByPhysAddr(pVM, GCPhys + (pvFault & PAGE_OFFSET_MASK),
568	&pCur, &iPage);
569	Assert(RT_SUCCESS(rc) \|\| !pCur);
570	if ( pCur
571	&& ( uErr & X86_TRAP_PF_RW
572	\|\| pCur->enmType != PGMVIRTHANDLERTYPE_WRITE ) )
573	{
574	Assert((pCur->aPhysToVirt[iPage].Core.Key & X86_PTE_PAE_PG_MASK) == GCPhys);
575	# ifdef IN_RC
576	RTGCPTR off = (iPage << PAGE_SHIFT) + (pvFault & PAGE_OFFSET_MASK) - (pCur->Core.Key & PAGE_OFFSET_MASK);
577	Assert(off < pCur->cb);
578	STAM_PROFILE_START(&pCur->Stat, h);
579	pgmUnlock(pVM);
580	rc = pCur->CTX_SUFF(pfnHandler)(pVM, uErr, pRegFrame, pvFault, pCur->Core.Key, off);
581	pgmLock(pVM);
582	STAM_PROFILE_STOP(&pCur->Stat, h);
583	# else
584	rc = VINF_EM_RAW_EMULATE_INSTR; /** @todo for VMX */
585	# endif
586	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eHandlersVirtualByPhys);
587	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeHandlers, b);
588	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2HndVirt; });
589	return rc;
590	}
591	}
592	}
593	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
594
595	/*
596	* There is a handled area of the page, but this fault doesn't belong to it.
597	* We must emulate the instruction.
598	*
599	* To avoid crashing (non-fatal) in the interpreter and go back to the recompiler
600	* we first check if this was a page-not-present fault for a page with only
601	* write access handlers. Restart the instruction if it wasn't a write access.
602	*/
603	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eHandlersUnhandled);
604
605	if ( !PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage)
606	&& !(uErr & X86_TRAP_PF_P))
607	{
608	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, pvFault, PGM_SYNC_NR_PAGES, uErr);
609	if ( RT_FAILURE(rc)
610	\|\| rc == VINF_PGM_SYNCPAGE_MODIFIED_PDE
611	\|\| !(uErr & X86_TRAP_PF_RW))
612	{
613	AssertRC(rc);
614	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eHandlersOutOfSync);
615	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeHandlers, b);
616	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2OutOfSyncHndPhys; });
617	return rc;
618	}
619	}
620
621	/** @todo This particular case can cause quite a lot of overhead. E.g. early stage of kernel booting in Ubuntu 6.06
622	* It's writing to an unhandled part of the LDT page several million times.
623	*/
624	rc = PGMInterpretInstruction(pVM, pVCpu, pRegFrame, pvFault);
625	LogFlow(("PGM: PGMInterpretInstruction -> rc=%d pPage=%R[pgmpage]\n", rc, pPage));
626	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeHandlers, b);
627	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2HndUnhandled; });
628	return rc;
629	} /* if any kind of handler */
630
631	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
632	if (uErr & X86_TRAP_PF_P)
633	{
634	/*
635	* The page isn't marked, but it might still be monitored by a virtual page access handler.
636	* (ASSUMES no temporary disabling of virtual handlers.)
637	*/
638	/** @todo r=bird: Since the purpose is to catch out of sync pages with virtual handler(s) here,
639	* we should correct both the shadow page table and physical memory flags, and not only check for
640	* accesses within the handler region but for access to pages with virtual handlers. */
641	PPGMVIRTHANDLER pCur = (PPGMVIRTHANDLER)RTAvlroGCPtrRangeGet(&pVM->pgm.s.CTX_SUFF(pTrees)->VirtHandlers, pvFault);
642	if (pCur)
643	{
644	AssertMsg( !(pvFault - pCur->Core.Key < pCur->cb)
645	\|\| ( pCur->enmType != PGMVIRTHANDLERTYPE_WRITE
646	\|\| !(uErr & X86_TRAP_PF_P)
647	\|\| (pCur->enmType == PGMVIRTHANDLERTYPE_WRITE && (uErr & X86_TRAP_PF_RW))),
648	("Unexpected trap for virtual handler: %08X (phys=%08x) %R[pgmpage] uErr=%X, enum=%d\n", pvFault, GCPhys, pPage, uErr, pCur->enmType));
649
650	if ( pvFault - pCur->Core.Key < pCur->cb
651	&& ( uErr & X86_TRAP_PF_RW
652	\|\| pCur->enmType != PGMVIRTHANDLERTYPE_WRITE ) )
653	{
654	# ifdef IN_RC
655	STAM_PROFILE_START(&pCur->Stat, h);
656	pgmUnlock(pVM);
657	rc = pCur->CTX_SUFF(pfnHandler)(pVM, uErr, pRegFrame, pvFault, pCur->Core.Key, pvFault - pCur->Core.Key);
658	pgmLock(pVM);
659	STAM_PROFILE_STOP(&pCur->Stat, h);
660	# else
661	rc = VINF_EM_RAW_EMULATE_INSTR; /** @todo for VMX */
662	# endif
663	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eHandlersVirtualUnmarked);
664	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeHandlers, b);
665	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2HndVirt; });
666	return rc;
667	}
668	}
669	}
670	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
671	}
672	else
673	{
674	/*
675	* When the guest accesses invalid physical memory (e.g. probing
676	* of RAM or accessing a remapped MMIO range), then we'll fall
677	* back to the recompiler to emulate the instruction.
678	*/
679	LogFlow(("PGM #PF: pgmPhysGetPageEx(%RGp) failed with %Rrc\n", GCPhys, rc));
680	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eHandlersInvalid);
681	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeHandlers, b);
682	return VINF_EM_RAW_EMULATE_INSTR;
683	}
684
685	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeHandlers, b);
686
687	# ifdef PGM_OUT_OF_SYNC_IN_GC /** @todo remove this bugger. */
688	/*
689	* We are here only if page is present in Guest page tables and
690	* trap is not handled by our handlers.
691	*
692	* Check it for page out-of-sync situation.
693	*/
694	STAM_PROFILE_START(&pVCpu->pgm.s.StatRZTrap0eTimeOutOfSync, c);
695
696	if (!(uErr & X86_TRAP_PF_P))
697	{
698	/*
699	* Page is not present in our page tables.
700	* Try to sync it!
701	* BTW, fPageShw is invalid in this branch!
702	*/
703	if (uErr & X86_TRAP_PF_US)
704	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,PageOutOfSyncUser));
705	else /* supervisor */
706	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,PageOutOfSyncSupervisor));
707
708	if (PGM_PAGE_IS_BALLOONED(pPage))
709	{
710	/* Emulate reads from ballooned pages as they are not present in our shadow page tables. (required for e.g. Solaris guests; soft ecc, random nr generator) */
711	rc = PGMInterpretInstruction(pVM, pVCpu, pRegFrame, pvFault);
712	LogFlow(("PGM: PGMInterpretInstruction balloon -> rc=%d pPage=%R[pgmpage]\n", rc, pPage));
713	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,PageOutOfSyncBallloon));
714	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeHandlers, b);
715	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2HndUnhandled; });
716	return rc;
717	}
718	# if defined(LOG_ENABLED) && !defined(IN_RING0)
719	RTGCPHYS GCPhys2;
720	uint64_t fPageGst2;
721	PGMGstGetPage(pVCpu, pvFault, &fPageGst2, &GCPhys2);
722	Log(("Page out of sync: %RGv eip=%08x PdeSrc.n.u1User=%d fPageGst2=%08llx GCPhys2=%RGp scan=%d\n",
723	pvFault, pRegFrame->eip, PdeSrc.n.u1User, fPageGst2, GCPhys2, CSAMDoesPageNeedScanning(pVM, pRegFrame->eip)));
724	# endif /* LOG_ENABLED */
725
726	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && !defined(IN_RING0)
727	if (CPUMGetGuestCPL(pVCpu, pRegFrame) == 0)
728	{
729	uint64_t fPageGst;
730	rc = PGMGstGetPage(pVCpu, pvFault, &fPageGst, NULL);
731	if ( RT_SUCCESS(rc)
732	&& !(fPageGst & X86_PTE_US))
733	{
734	/* Note: can't check for X86_TRAP_ID bit, because that requires execute disable support on the CPU */
735	if ( pvFault == (RTGCPTR)pRegFrame->eip
736	\|\| pvFault - pRegFrame->eip < 8 /* instruction crossing a page boundary */
737	# ifdef CSAM_DETECT_NEW_CODE_PAGES
738	\|\| ( !PATMIsPatchGCAddr(pVM, pRegFrame->eip)
739	&& CSAMDoesPageNeedScanning(pVM, pRegFrame->eip)) /* any new code we encounter here */
740	# endif /* CSAM_DETECT_NEW_CODE_PAGES */
741	)
742	{
743	LogFlow(("CSAMExecFault %RX32\n", pRegFrame->eip));
744	rc = CSAMExecFault(pVM, (RTRCPTR)pRegFrame->eip);
745	if (rc != VINF_SUCCESS)
746	{
747	/*
748	* CSAM needs to perform a job in ring 3.
749	*
750	* Sync the page before going to the host context; otherwise we'll end up in a loop if
751	* CSAM fails (e.g. instruction crosses a page boundary and the next page is not present)
752	*/
753	LogFlow(("CSAM ring 3 job\n"));
754	int rc2 = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, pvFault, 1, uErr);
755	AssertRC(rc2);
756
757	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeOutOfSync, c);
758	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2CSAM; });
759	return rc;
760	}
761	}
762	# ifdef CSAM_DETECT_NEW_CODE_PAGES
763	else if ( uErr == X86_TRAP_PF_RW
764	&& pRegFrame->ecx >= 0x100 /* early check for movswd count */
765	&& pRegFrame->ecx < 0x10000)
766	{
767	/* In case of a write to a non-present supervisor shadow page, we'll take special precautions
768	* to detect loading of new code pages.
769	*/
770
771	/*
772	* Decode the instruction.
773	*/
774	RTGCPTR PC;
775	rc = SELMValidateAndConvertCSAddr(pVM, pRegFrame->eflags, pRegFrame->ss, pRegFrame->cs, &pRegFrame->csHid, (RTGCPTR)pRegFrame->eip, &PC);
776	if (rc == VINF_SUCCESS)
777	{
778	PDISCPUSTATE pDis = &pVCpu->pgm.s.DisState;
779	uint32_t cbOp;
780	rc = EMInterpretDisasOneEx(pVM, pVCpu, PC, pRegFrame, pDis, &cbOp);
781
782	/* For now we'll restrict this to rep movsw/d instructions */
783	if ( rc == VINF_SUCCESS
784	&& pDis->pCurInstr->opcode == OP_MOVSWD
785	&& (pDis->prefix & PREFIX_REP))
786	{
787	CSAMMarkPossibleCodePage(pVM, pvFault);
788	}
789	}
790	}
791	# endif /* CSAM_DETECT_NEW_CODE_PAGES */
792
793	/*
794	* Mark this page as safe.
795	*/
796	/** @todo not correct for pages that contain both code and data!! */
797	Log2(("CSAMMarkPage %RGv; scanned=%d\n", pvFault, true));
798	CSAMMarkPage(pVM, pvFault, true);
799	}
800	}
801	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && !defined(IN_RING0) */
802	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, pvFault, PGM_SYNC_NR_PAGES, uErr);
803	if (RT_SUCCESS(rc))
804	{
805	/* The page was successfully synced, return to the guest. */
806	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeOutOfSync, c);
807	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2OutOfSync; });
808	return VINF_SUCCESS;
809	}
810	}
811	else /* uErr & X86_TRAP_PF_P: */
812	{
813	/*
814	* Write protected pages are make writable when the guest makes the first
815	* write to it. This happens for pages that are shared, write monitored
816	* and not yet allocated.
817	*
818	* Also, a side effect of not flushing global PDEs are out of sync pages due
819	* to physical monitored regions, that are no longer valid.
820	* Assume for now it only applies to the read/write flag.
821	*/
822	if ( RT_SUCCESS(rc)
823	&& (uErr & X86_TRAP_PF_RW))
824	{
825	if (PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
826	{
827	Log(("PGM #PF: Make writable: %RGp %R[pgmpage] pvFault=%RGp uErr=%#x\n", GCPhys, pPage, pvFault, uErr));
828	Assert(!PGM_PAGE_IS_ZERO(pPage));
829	AssertFatalMsg(!PGM_PAGE_IS_BALLOONED(pPage), ("Unexpected ballooned page at %RGp\n", GCPhys));
830
831	rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhys);
832	if (rc != VINF_SUCCESS)
833	{
834	AssertMsg(rc == VINF_PGM_SYNC_CR3 \|\| RT_FAILURE(rc), ("%Rrc\n", rc));
835	return rc;
836	}
837	if (RT_UNLIKELY(VM_FF_ISPENDING(pVM, VM_FF_PGM_NO_MEMORY)))
838	return VINF_EM_NO_MEMORY;
839	}
840
841	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
842	/* Check to see if we need to emulate the instruction as X86_CR0_WP has been cleared. */
843	if ( CPUMGetGuestCPL(pVCpu, pRegFrame) == 0
844	&& ((CPUMGetGuestCR0(pVCpu) & (X86_CR0_WP \| X86_CR0_PG)) == X86_CR0_PG))
845	{
846	Assert((uErr & (X86_TRAP_PF_RW \| X86_TRAP_PF_P)) == (X86_TRAP_PF_RW \| X86_TRAP_PF_P));
847	uint64_t fPageGst;
848	rc = PGMGstGetPage(pVCpu, pvFault, &fPageGst, NULL);
849	if ( RT_SUCCESS(rc)
850	&& !(fPageGst & X86_PTE_RW))
851	{
852	rc = PGMInterpretInstruction(pVM, pVCpu, pRegFrame, pvFault);
853	if (RT_SUCCESS(rc))
854	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eWPEmulInRZ);
855	else
856	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eWPEmulToR3);
857	return rc;
858	}
859	AssertMsg(RT_SUCCESS(rc), ("Unexpected r/w page %RGv flag=%x rc=%Rrc\n", pvFault, (uint32_t)fPageGst, rc));
860	}
861	# endif
862	/// @todo count the above case; else
863	if (uErr & X86_TRAP_PF_US)
864	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,PageOutOfSyncUserWrite));
865	else /* supervisor */
866	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,PageOutOfSyncSupervisorWrite));
867
868	/*
869	* Note: Do NOT use PGM_SYNC_NR_PAGES here. That only works if the
870	* page is not present, which is not true in this case.
871	*/
872	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, pvFault, 1, uErr);
873	if (RT_SUCCESS(rc))
874	{
875	/*
876	* Page was successfully synced, return to guest.
877	* First invalidate the page as it might be in the TLB.
878	*/
879	# if PGM_SHW_TYPE == PGM_TYPE_EPT
880	HWACCMInvalidatePhysPage(pVM, (RTGCPHYS)pvFault);
881	# else
882	PGM_INVL_PG(pVCpu, pvFault);
883	# endif
884	# ifdef VBOX_STRICT
885	RTGCPHYS GCPhys2;
886	uint64_t fPageGst;
887	if (!HWACCMIsNestedPagingActive(pVM))
888	{
889	rc = PGMGstGetPage(pVCpu, pvFault, &fPageGst, &GCPhys2);
890	AssertMsg(RT_SUCCESS(rc) && (fPageGst & X86_PTE_RW), ("rc=%d fPageGst=%RX64\n"));
891	LogFlow(("Obsolete physical monitor page out of sync %RGv - phys %RGp flags=%08llx\n", pvFault, GCPhys2, (uint64_t)fPageGst));
892	}
893	uint64_t fPageShw;
894	rc = PGMShwGetPage(pVCpu, pvFault, &fPageShw, NULL);
895	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 */, ("rc=%Rrc fPageShw=%RX64\n", rc, fPageShw));
896	# endif /* VBOX_STRICT */
897	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeOutOfSync, c);
898	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2OutOfSyncHndObs; });
899	return VINF_SUCCESS;
900	}
901	}
902
903	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
904	# ifdef VBOX_STRICT
905	/*
906	* Check for VMM page flags vs. Guest page flags consistency.
907	* Currently only for debug purposes.
908	*/
909	if (RT_SUCCESS(rc))
910	{
911	/* Get guest page flags. */
912	uint64_t fPageGst;
913	rc = PGMGstGetPage(pVCpu, pvFault, &fPageGst, NULL);
914	if (RT_SUCCESS(rc))
915	{
916	uint64_t fPageShw;
917	rc = PGMShwGetPage(pVCpu, pvFault, &fPageShw, NULL);
918
919	/*
920	* Compare page flags.
921	* Note: we have AVL, A, D bits desynched.
922	*/
923	AssertMsg((fPageShw & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK)) == (fPageGst & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK)),
924	("Page flags mismatch! pvFault=%RGv uErr=%x GCPhys=%RGp fPageShw=%RX64 fPageGst=%RX64\n", pvFault, (uint32_t)uErr, GCPhys, fPageShw, fPageGst));
925	}
926	else
927	AssertMsgFailed(("PGMGstGetPage rc=%Rrc\n", rc));
928	}
929	else
930	AssertMsgFailed(("PGMGCGetPage rc=%Rrc\n", rc));
931	# endif /* VBOX_STRICT */
932	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
933	}
934	STAM_PROFILE_STOP(&pVCpu->pgm.s.StatRZTrap0eTimeOutOfSync, c);
935	# endif /* PGM_OUT_OF_SYNC_IN_GC */
936	}
937	else /* GCPhys == NIL_RTGCPHYS */
938	{
939	/*
940	* Page not present in Guest OS or invalid page table address.
941	* This is potential virtual page access handler food.
942	*
943	* For the present we'll say that our access handlers don't
944	* work for this case - we've already discarded the page table
945	* not present case which is identical to this.
946	*
947	* When we perchance find we need this, we will probably have AVL
948	* trees (offset based) to operate on and we can measure their speed
949	* agains mapping a page table and probably rearrange this handling
950	* a bit. (Like, searching virtual ranges before checking the
951	* physical address.)
952	*/
953	}
954	}
955
956	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
957	/*
958	* Conclusion, this is a guest trap.
959	*/
960	LogFlow(("PGM: Unhandled #PF -> route trap to recompiler!\n"));
961	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eGuestPFUnh);
962	return VINF_EM_RAW_GUEST_TRAP;
963	# else
964	/* present, but not a monitored page; perhaps the guest is probing physical memory */
965	return VINF_EM_RAW_EMULATE_INSTR;
966	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
967
968
969	# else /* PGM_GST_TYPE != PGM_TYPE_32BIT */
970
971	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_GST_TYPE, PGM_SHW_TYPE));
972	return VERR_INTERNAL_ERROR;
973	# endif /* PGM_GST_TYPE != PGM_TYPE_32BIT */
974	}
975	#endif /* !IN_RING3 */
976
977
978	/**
979	* Emulation of the invlpg instruction.
980	*
981	*
982	* @returns VBox status code.
983	*
984	* @param pVCpu The VMCPU handle.
985	* @param GCPtrPage Page to invalidate.
986	*
987	* @remark ASSUMES that the guest is updating before invalidating. This order
988	* isn't required by the CPU, so this is speculative and could cause
989	* trouble.
990	* @remark No TLB shootdown is done on any other VCPU as we assume that
991	* invlpg emulation is the only reason for calling this function.
992	* (The guest has to shoot down TLB entries on other CPUs itself)
993	* Currently true, but keep in mind!
994	*
995	* @todo Clean this up! Most of it is (or should be) no longer necessary as we catch all page table accesses.
996	*/
997	PGM_BTH_DECL(int, InvalidatePage)(PVMCPU pVCpu, RTGCPTR GCPtrPage)
998	{
999	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) \
1000	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
1001	&& PGM_SHW_TYPE != PGM_TYPE_EPT
1002	int rc;
1003	PVM pVM = pVCpu->CTX_SUFF(pVM);
1004	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1005
1006	Assert(PGMIsLockOwner(pVM));
1007
1008	LogFlow(("InvalidatePage %RGv\n", GCPtrPage));
1009
1010	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1011	if (pPool->cDirtyPages)
1012	pgmPoolResetDirtyPages(pVM);
1013	# endif
1014
1015	/*
1016	* Get the shadow PD entry and skip out if this PD isn't present.
1017	* (Guessing that it is frequent for a shadow PDE to not be present, do this first.)
1018	*/
1019	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
1020	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1021	PX86PDE pPdeDst = pgmShwGet32BitPDEPtr(&pVCpu->pgm.s, GCPtrPage);
1022
1023	/* Fetch the pgm pool shadow descriptor. */
1024	PPGMPOOLPAGE pShwPde = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
1025	Assert(pShwPde);
1026
1027	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
1028	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT);
1029	PX86PDPT pPdptDst = pgmShwGetPaePDPTPtr(&pVCpu->pgm.s);
1030
1031	/* If the shadow PDPE isn't present, then skip the invalidate. */
1032	if (!pPdptDst->a[iPdpt].n.u1Present)
1033	{
1034	Assert(!(pPdptDst->a[iPdpt].u & PGM_PLXFLAGS_MAPPING));
1035	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,InvalidatePageSkipped));
1036	return VINF_SUCCESS;
1037	}
1038
1039	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1040	PPGMPOOLPAGE pShwPde = NULL;
1041	PX86PDPAE pPDDst;
1042
1043	/* Fetch the pgm pool shadow descriptor. */
1044	rc = pgmShwGetPaePoolPagePD(&pVCpu->pgm.s, GCPtrPage, &pShwPde);
1045	AssertRCSuccessReturn(rc, rc);
1046	Assert(pShwPde);
1047
1048	pPDDst = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_BY_PGM(&pVM->pgm.s, pShwPde);
1049	PX86PDEPAE pPdeDst = &pPDDst->a[iPDDst];
1050
1051	# else /* PGM_SHW_TYPE == PGM_TYPE_AMD64 */
1052	/* PML4 */
1053	const unsigned iPml4 = (GCPtrPage >> X86_PML4_SHIFT) & X86_PML4_MASK;
1054	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
1055	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1056	PX86PDPAE pPDDst;
1057	PX86PDPT pPdptDst;
1058	PX86PML4E pPml4eDst;
1059	rc = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, &pPml4eDst, &pPdptDst, &pPDDst);
1060	if (rc != VINF_SUCCESS)
1061	{
1062	AssertMsg(rc == VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT \|\| rc == VERR_PAGE_MAP_LEVEL4_NOT_PRESENT, ("Unexpected rc=%Rrc\n", rc));
1063	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,InvalidatePageSkipped));
1064	return VINF_SUCCESS;
1065	}
1066	Assert(pPDDst);
1067
1068	PX86PDEPAE pPdeDst = &pPDDst->a[iPDDst];
1069	PX86PDPE pPdpeDst = &pPdptDst->a[iPdpt];
1070
1071	if (!pPdpeDst->n.u1Present)
1072	{
1073	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,InvalidatePageSkipped));
1074	return VINF_SUCCESS;
1075	}
1076
1077	/* Fetch the pgm pool shadow descriptor. */
1078	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & SHW_PDPE_PG_MASK);
1079	Assert(pShwPde);
1080
1081	# endif /* PGM_SHW_TYPE == PGM_TYPE_AMD64 */
1082
1083	const SHWPDE PdeDst = *pPdeDst;
1084	if (!PdeDst.n.u1Present)
1085	{
1086	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,InvalidatePageSkipped));
1087	return VINF_SUCCESS;
1088	}
1089
1090	# if defined(IN_RC)
1091	/* Make sure the dynamic pPdeDst mapping will not be reused during this function. */
1092	PGMDynLockHCPage(pVM, (uint8_t *)pPdeDst);
1093	# endif
1094
1095	/*
1096	* Get the guest PD entry and calc big page.
1097	*/
1098	# if PGM_GST_TYPE == PGM_TYPE_32BIT
1099	PGSTPD pPDSrc = pgmGstGet32bitPDPtr(&pVCpu->pgm.s);
1100	const unsigned iPDSrc = GCPtrPage >> GST_PD_SHIFT;
1101	GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
1102	# else /* PGM_GST_TYPE != PGM_TYPE_32BIT */
1103	unsigned iPDSrc = 0;
1104	# if PGM_GST_TYPE == PGM_TYPE_PAE
1105	X86PDPE PdpeSrc;
1106	PX86PDPAE pPDSrc = pgmGstGetPaePDPtr(&pVCpu->pgm.s, GCPtrPage, &iPDSrc, &PdpeSrc);
1107	# else /* AMD64 */
1108	PX86PML4E pPml4eSrc;
1109	X86PDPE PdpeSrc;
1110	PX86PDPAE pPDSrc = pgmGstGetLongModePDPtr(&pVCpu->pgm.s, GCPtrPage, &pPml4eSrc, &PdpeSrc, &iPDSrc);
1111	# endif
1112	GSTPDE PdeSrc;
1113
1114	if (pPDSrc)
1115	PdeSrc = pPDSrc->a[iPDSrc];
1116	else
1117	PdeSrc.u = 0;
1118	# endif /* PGM_GST_TYPE != PGM_TYPE_32BIT */
1119
1120	# if PGM_GST_TYPE == PGM_TYPE_32BIT
1121	const bool fIsBigPage = PdeSrc.b.u1Size && CPUMIsGuestPageSizeExtEnabled(pVCpu);
1122	# else
1123	const bool fIsBigPage = PdeSrc.b.u1Size;
1124	# endif
1125
1126	# ifdef IN_RING3
1127	/*
1128	* If a CR3 Sync is pending we may ignore the invalidate page operation
1129	* depending on the kind of sync and if it's a global page or not.
1130	* This doesn't make sense in GC/R0 so we'll skip it entirely there.
1131	*/
1132	# ifdef PGM_SKIP_GLOBAL_PAGEDIRS_ON_NONGLOBAL_FLUSH
1133	if ( VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3)
1134	\|\| ( VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL)
1135	&& fIsBigPage
1136	&& PdeSrc.b.u1Global
1137	)
1138	)
1139	# else
1140	if (VM_FF_ISPENDING(pVM, VM_FF_PGM_SYNC_CR3 \| VM_FF_PGM_SYNC_CR3_NON_GLOBAL) )
1141	# endif
1142	{
1143	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,InvalidatePageSkipped));
1144	return VINF_SUCCESS;
1145	}
1146	# endif /* IN_RING3 */
1147
1148	/*
1149	* Deal with the Guest PDE.
1150	*/
1151	rc = VINF_SUCCESS;
1152	if (PdeSrc.n.u1Present)
1153	{
1154	Assert( PdeSrc.n.u1User == PdeDst.n.u1User
1155	&& (PdeSrc.n.u1Write \|\| !PdeDst.n.u1Write));
1156	# ifndef PGM_WITHOUT_MAPPING
1157	if (PdeDst.u & PGM_PDFLAGS_MAPPING)
1158	{
1159	/*
1160	* Conflict - Let SyncPT deal with it to avoid duplicate code.
1161	*/
1162	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
1163	Assert(PGMGetGuestMode(pVCpu) <= PGMMODE_PAE);
1164	pgmLock(pVM);
1165	rc = PGM_BTH_NAME(SyncPT)(pVCpu, iPDSrc, pPDSrc, GCPtrPage);
1166	pgmUnlock(pVM);
1167	}
1168	else
1169	# endif /* !PGM_WITHOUT_MAPPING */
1170	if (!fIsBigPage)
1171	{
1172	/*
1173	* 4KB - page.
1174	*/
1175	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, PdeDst.u & SHW_PDE_PG_MASK);
1176	RTGCPHYS GCPhys = PdeSrc.u & GST_PDE_PG_MASK;
1177
1178	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1179	/* Reset the modification counter (OpenSolaris trashes tlb entries very often) */
1180	if (pShwPage->cModifications)
1181	pShwPage->cModifications = 1;
1182	# endif
1183
1184	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
1185	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
1186	GCPhys \|= (iPDDst & 1) * (PAGE_SIZE/2);
1187	# endif
1188	if (pShwPage->GCPhys == GCPhys)
1189	{
1190	# if 0 /* likely cause of a major performance regression; must be SyncPageWorkerTrackDeref then */
1191	const unsigned iPTEDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
1192	PSHWPT pPT = (PSHWPT)PGMPOOL_PAGE_2_PTR(pVM, pShwPage);
1193	if (pPT->a[iPTEDst].n.u1Present)
1194	{
1195	/* This is very unlikely with caching/monitoring enabled. */
1196	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pShwPage, pPT->a[iPTEDst].u & SHW_PTE_PG_MASK, iPTEDst);
1197	ASMAtomicWriteSize(&pPT->a[iPTEDst], 0);
1198	}
1199	# else /* Syncing it here isn't 100% safe and it's probably not worth spending time syncing it. */
1200	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, GCPtrPage, 1, 0);
1201	if (RT_SUCCESS(rc))
1202	rc = VINF_SUCCESS;
1203	# endif
1204	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,InvalidatePage4KBPages));
1205	PGM_INVL_PG(pVCpu, GCPtrPage);
1206	}
1207	else
1208	{
1209	/*
1210	* The page table address changed.
1211	*/
1212	LogFlow(("InvalidatePage: Out-of-sync at %RGp PdeSrc=%RX64 PdeDst=%RX64 ShwGCPhys=%RGp iPDDst=%#x\n",
1213	GCPtrPage, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u, pShwPage->GCPhys, iPDDst));
1214	pgmPoolFree(pVM, PdeDst.u & SHW_PDE_PG_MASK, pShwPde->idx, iPDDst);
1215	ASMAtomicWriteSize(pPdeDst, 0);
1216	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,InvalidatePagePDOutOfSync));
1217	PGM_INVL_VCPU_TLBS(pVCpu);
1218	}
1219	}
1220	else
1221	{
1222	/*
1223	* 2/4MB - page.
1224	*/
1225	/* Before freeing the page, check if anything really changed. */
1226	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, PdeDst.u & SHW_PDE_PG_MASK);
1227	RTGCPHYS GCPhys = GST_GET_PDE_BIG_PG_GCPHYS(PdeSrc);
1228	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
1229	/* Select the right PDE as we're emulating a 4MB page directory with two 2 MB shadow PDEs.*/
1230	GCPhys \|= GCPtrPage & (1 << X86_PD_PAE_SHIFT);
1231	# endif
1232	if ( pShwPage->GCPhys == GCPhys
1233	&& pShwPage->enmKind == BTH_PGMPOOLKIND_PT_FOR_BIG)
1234	{
1235	/* ASSUMES a the given bits are identical for 4M and normal PDEs */
1236	/** @todo PAT */
1237	if ( (PdeSrc.u & (X86_PDE_P \| X86_PDE_RW \| X86_PDE_US \| X86_PDE_PWT \| X86_PDE_PCD))
1238	== (PdeDst.u & (X86_PDE_P \| X86_PDE_RW \| X86_PDE_US \| X86_PDE_PWT \| X86_PDE_PCD))
1239	&& ( PdeSrc.b.u1Dirty /** @todo rainy day: What about read-only 4M pages? not very common, but still... */
1240	\|\| (PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY)))
1241	{
1242	LogFlow(("Skipping flush for big page containing %RGv (PD=%X .u=%RX64)-> nothing has changed!\n", GCPtrPage, iPDSrc, PdeSrc.u));
1243	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,InvalidatePage4MBPagesSkip));
1244	# if defined(IN_RC)
1245	/* Make sure the dynamic pPdeDst mapping will not be reused during this function. */
1246	PGMDynUnlockHCPage(pVM, (uint8_t *)pPdeDst);
1247	# endif
1248	return VINF_SUCCESS;
1249	}
1250	}
1251
1252	/*
1253	* Ok, the page table is present and it's been changed in the guest.
1254	* If we're in host context, we'll just mark it as not present taking the lazy approach.
1255	* We could do this for some flushes in GC too, but we need an algorithm for
1256	* deciding which 4MB pages containing code likely to be executed very soon.
1257	*/
1258	LogFlow(("InvalidatePage: Out-of-sync PD at %RGp PdeSrc=%RX64 PdeDst=%RX64\n",
1259	GCPtrPage, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
1260	pgmPoolFree(pVM, PdeDst.u & SHW_PDE_PG_MASK, pShwPde->idx, iPDDst);
1261	ASMAtomicWriteSize(pPdeDst, 0);
1262	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,InvalidatePage4MBPages));
1263	PGM_INVL_BIG_PG(pVCpu, GCPtrPage);
1264	}
1265	}
1266	else
1267	{
1268	/*
1269	* Page directory is not present, mark shadow PDE not present.
1270	*/
1271	if (!(PdeDst.u & PGM_PDFLAGS_MAPPING))
1272	{
1273	pgmPoolFree(pVM, PdeDst.u & SHW_PDE_PG_MASK, pShwPde->idx, iPDDst);
1274	ASMAtomicWriteSize(pPdeDst, 0);
1275	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,InvalidatePagePDNPs));
1276	PGM_INVL_PG(pVCpu, GCPtrPage);
1277	}
1278	else
1279	{
1280	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
1281	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,InvalidatePagePDMappings));
1282	}
1283	}
1284	# if defined(IN_RC)
1285	/* Make sure the dynamic pPdeDst mapping will not be reused during this function. */
1286	PGMDynUnlockHCPage(pVM, (uint8_t *)pPdeDst);
1287	# endif
1288	return rc;
1289
1290	#else /* guest real and protected mode */
1291	/* There's no such thing as InvalidatePage when paging is disabled, so just ignore. */
1292	return VINF_SUCCESS;
1293	#endif
1294	}
1295
1296
1297	/**
1298	* Update the tracking of shadowed pages.
1299	*
1300	* @param pVCpu The VMCPU handle.
1301	* @param pShwPage The shadow page.
1302	* @param HCPhys The physical page we is being dereferenced.
1303	* @param iPte Shadow PTE index
1304	*/
1305	DECLINLINE(void) PGM_BTH_NAME(SyncPageWorkerTrackDeref)(PVMCPU pVCpu, PPGMPOOLPAGE pShwPage, RTHCPHYS HCPhys, uint16_t iPte)
1306	{
1307	PVM pVM = pVCpu->CTX_SUFF(pVM);
1308
1309	STAM_PROFILE_START(&pVM->pgm.s.StatTrackDeref, a);
1310	LogFlow(("SyncPageWorkerTrackDeref: Damn HCPhys=%RHp pShwPage->idx=%#x!!!\n", HCPhys, pShwPage->idx));
1311
1312	/** @todo If this turns out to be a bottle neck (very likely) two things can be done:
1313	* 1. have a medium sized HCPhys -> GCPhys TLB (hash?)
1314	* 2. write protect all shadowed pages. I.e. implement caching.
1315	*/
1316	/** @todo duplicated in the 2nd half of pgmPoolTracDerefGCPhysHint */
1317
1318	/*
1319	* Find the guest address.
1320	*/
1321	for (PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRanges);
1322	pRam;
1323	pRam = pRam->CTX_SUFF(pNext))
1324	{
1325	unsigned iPage = pRam->cb >> PAGE_SHIFT;
1326	while (iPage-- > 0)
1327	{
1328	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
1329	{
1330	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1331
1332	Assert(pShwPage->cPresent);
1333	Assert(pPool->cPresent);
1334	pShwPage->cPresent--;
1335	pPool->cPresent--;
1336
1337	pgmTrackDerefGCPhys(pPool, pShwPage, &pRam->aPages[iPage], iPte);
1338	STAM_PROFILE_STOP(&pVM->pgm.s.StatTrackDeref, a);
1339	return;
1340	}
1341	}
1342	}
1343
1344	for (;;)
1345	AssertReleaseMsgFailed(("HCPhys=%RHp wasn't found!\n", HCPhys));
1346	}
1347
1348
1349	/**
1350	* Update the tracking of shadowed pages.
1351	*
1352	* @param pVCpu The VMCPU handle.
1353	* @param pShwPage The shadow page.
1354	* @param u16 The top 16-bit of the pPage->HCPhys.
1355	* @param pPage Pointer to the guest page. this will be modified.
1356	* @param iPTDst The index into the shadow table.
1357	*/
1358	DECLINLINE(void) PGM_BTH_NAME(SyncPageWorkerTrackAddref)(PVMCPU pVCpu, PPGMPOOLPAGE pShwPage, uint16_t u16, PPGMPAGE pPage, const unsigned iPTDst)
1359	{
1360	PVM pVM = pVCpu->CTX_SUFF(pVM);
1361	/*
1362	* Just deal with the simple first time here.
1363	*/
1364	if (!u16)
1365	{
1366	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackVirgin);
1367	u16 = PGMPOOL_TD_MAKE(1, pShwPage->idx);
1368	/* Save the page table index. */
1369	PGM_PAGE_SET_PTE_INDEX(pPage, iPTDst);
1370	}
1371	else
1372	u16 = pgmPoolTrackPhysExtAddref(pVM, pPage, u16, pShwPage->idx, iPTDst);
1373
1374	/* write back */
1375	Log2(("SyncPageWorkerTrackAddRef: u16=%#x->%#x iPTDst=%#x\n", u16, PGM_PAGE_GET_TRACKING(pPage), iPTDst));
1376	PGM_PAGE_SET_TRACKING(pPage, u16);
1377
1378	/* update statistics. */
1379	pVM->pgm.s.CTX_SUFF(pPool)->cPresent++;
1380	pShwPage->cPresent++;
1381	if (pShwPage->iFirstPresent > iPTDst)
1382	pShwPage->iFirstPresent = iPTDst;
1383	}
1384
1385
1386	/**
1387	* Creates a 4K shadow page for a guest page.
1388	*
1389	* For 4M pages the caller must convert the PDE4M to a PTE, this includes adjusting the
1390	* physical address. The PdeSrc argument only the flags are used. No page structured
1391	* will be mapped in this function.
1392	*
1393	* @param pVCpu The VMCPU handle.
1394	* @param pPteDst Destination page table entry.
1395	* @param PdeSrc Source page directory entry (i.e. Guest OS page directory entry).
1396	* Can safely assume that only the flags are being used.
1397	* @param PteSrc Source page table entry (i.e. Guest OS page table entry).
1398	* @param pShwPage Pointer to the shadow page.
1399	* @param iPTDst The index into the shadow table.
1400	*
1401	* @remark Not used for 2/4MB pages!
1402	*/
1403	DECLINLINE(void) PGM_BTH_NAME(SyncPageWorker)(PVMCPU pVCpu, PSHWPTE pPteDst, GSTPDE PdeSrc, GSTPTE PteSrc, PPGMPOOLPAGE pShwPage, unsigned iPTDst)
1404	{
1405	if (PteSrc.n.u1Present)
1406	{
1407	PVM pVM = pVCpu->CTX_SUFF(pVM);
1408
1409	# if defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT) \
1410	&& PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) \
1411	&& (PGM_GST_TYPE == PGM_TYPE_PAE \|\| PGM_GST_TYPE == PGM_TYPE_AMD64)
1412	if (pShwPage->fDirty)
1413	{
1414	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1415	PX86PTPAE pGstPT;
1416
1417	pGstPT = (PX86PTPAE)&pPool->aDirtyPages[pShwPage->idxDirty][0];
1418	pGstPT->a[iPTDst].u = PteSrc.u;
1419	}
1420	# endif
1421	/*
1422	* Find the ram range.
1423	*/
1424	PPGMPAGE pPage;
1425	int rc = pgmPhysGetPageEx(&pVM->pgm.s, PteSrc.u & GST_PTE_PG_MASK, &pPage);
1426	if (RT_SUCCESS(rc))
1427	{
1428	/* Ignore ballooned pages. Don't return errors or use a fatal assert here as part of a shadow sync range might included ballooned pages. */
1429	if (PGM_PAGE_IS_BALLOONED(pPage))
1430	return;
1431
1432	#ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
1433	/* Try to make the page writable if necessary. */
1434	if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM
1435	&& ( PGM_PAGE_IS_ZERO(pPage)
1436	\|\| ( PteSrc.n.u1Write
1437	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED
1438	# ifdef VBOX_WITH_REAL_WRITE_MONITORED_PAGES
1439	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_WRITE_MONITORED
1440	# endif
1441	# ifdef VBOX_WITH_PAGE_SHARING
1442	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_SHARED
1443	# endif
1444	)
1445	)
1446	)
1447	{
1448	rc = pgmPhysPageMakeWritable(pVM, pPage, PteSrc.u & GST_PTE_PG_MASK);
1449	AssertRC(rc);
1450	}
1451	#endif
1452
1453	/** @todo investiage PWT, PCD and PAT. */
1454	/*
1455	* Make page table entry.
1456	*/
1457	SHWPTE PteDst;
1458	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
1459	{
1460	/** @todo r=bird: Are we actually handling dirty and access bits for pages with access handlers correctly? No. */
1461	if (!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage))
1462	{
1463	#if PGM_SHW_TYPE == PGM_TYPE_EPT
1464	PteDst.u = PGM_PAGE_GET_HCPHYS(pPage);
1465	PteDst.n.u1Present = 1;
1466	PteDst.n.u1Execute = 1;
1467	PteDst.n.u1IgnorePAT = 1;
1468	PteDst.n.u3EMT = VMX_EPT_MEMTYPE_WB;
1469	/* PteDst.n.u1Write = 0 && PteDst.n.u1Size = 0 */
1470	#else
1471	PteDst.u = (PteSrc.u & ~(X86_PTE_PAE_PG_MASK \| X86_PTE_AVL_MASK \| X86_PTE_PAT \| X86_PTE_PCD \| X86_PTE_PWT \| X86_PTE_RW))
1472	\| PGM_PAGE_GET_HCPHYS(pPage);
1473	#endif
1474	}
1475	else
1476	{
1477	LogFlow(("SyncPageWorker: monitored page (%RHp) -> mark not present\n", PGM_PAGE_GET_HCPHYS(pPage)));
1478	PteDst.u = 0;
1479	}
1480	/** @todo count these two kinds. */
1481	}
1482	else
1483	{
1484	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1485	/*
1486	* If the page or page directory entry is not marked accessed,
1487	* we mark the page not present.
1488	*/
1489	if (!PteSrc.n.u1Accessed \|\| !PdeSrc.n.u1Accessed)
1490	{
1491	LogFlow(("SyncPageWorker: page and or page directory not accessed -> mark not present\n"));
1492	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,AccessedPage));
1493	PteDst.u = 0;
1494	}
1495	else
1496	/*
1497	* If the page is not flagged as dirty and is writable, then make it read-only, so we can set the dirty bit
1498	* when the page is modified.
1499	*/
1500	if (!PteSrc.n.u1Dirty && (PdeSrc.n.u1Write & PteSrc.n.u1Write))
1501	{
1502	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyPage));
1503	PteDst.u = (PteSrc.u & ~(X86_PTE_PAE_PG_MASK \| X86_PTE_AVL_MASK \| X86_PTE_PAT \| X86_PTE_PCD \| X86_PTE_PWT \| X86_PTE_RW))
1504	\| PGM_PAGE_GET_HCPHYS(pPage)
1505	\| PGM_PTFLAGS_TRACK_DIRTY;
1506	}
1507	else
1508	#endif
1509	{
1510	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyPageSkipped));
1511	#if PGM_SHW_TYPE == PGM_TYPE_EPT
1512	PteDst.u = PGM_PAGE_GET_HCPHYS(pPage);
1513	PteDst.n.u1Present = 1;
1514	PteDst.n.u1Write = 1;
1515	PteDst.n.u1Execute = 1;
1516	PteDst.n.u1IgnorePAT = 1;
1517	PteDst.n.u3EMT = VMX_EPT_MEMTYPE_WB;
1518	/* PteDst.n.u1Size = 0 */
1519	#else
1520	PteDst.u = (PteSrc.u & ~(X86_PTE_PAE_PG_MASK \| X86_PTE_AVL_MASK \| X86_PTE_PAT \| X86_PTE_PCD \| X86_PTE_PWT))
1521	\| PGM_PAGE_GET_HCPHYS(pPage);
1522	#endif
1523	}
1524	}
1525
1526	/*
1527	* Make sure only allocated pages are mapped writable.
1528	*/
1529	if ( PteDst.n.u1Write
1530	&& PteDst.n.u1Present
1531	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
1532	{
1533	/* Still applies to shared pages. */
1534	Assert(!PGM_PAGE_IS_ZERO(pPage));
1535	PteDst.n.u1Write = 0; /** @todo this isn't quite working yet. */
1536	Log3(("SyncPageWorker: write-protecting %RGp pPage=%R[pgmpage]at iPTDst=%d\n", (RTGCPHYS)(PteSrc.u & X86_PTE_PAE_PG_MASK), pPage, iPTDst));
1537	}
1538
1539	/*
1540	* Keep user track up to date.
1541	*/
1542	if (PteDst.n.u1Present)
1543	{
1544	if (!pPteDst->n.u1Present)
1545	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPTDst);
1546	else if ((pPteDst->u & SHW_PTE_PG_MASK) != (PteDst.u & SHW_PTE_PG_MASK))
1547	{
1548	Log2(("SyncPageWorker: deref! *pPteDst=%RX64 PteDst=%RX64\n", (uint64_t)pPteDst->u, (uint64_t)PteDst.u));
1549	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pVCpu, pShwPage, pPteDst->u & SHW_PTE_PG_MASK, iPTDst);
1550	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPTDst);
1551	}
1552	}
1553	else if (pPteDst->n.u1Present)
1554	{
1555	Log2(("SyncPageWorker: deref! *pPteDst=%RX64\n", (uint64_t)pPteDst->u));
1556	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pVCpu, pShwPage, pPteDst->u & SHW_PTE_PG_MASK, iPTDst);
1557	}
1558
1559	/*
1560	* Update statistics and commit the entry.
1561	*/
1562	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1563	if (!PteSrc.n.u1Global)
1564	pShwPage->fSeenNonGlobal = true;
1565	#endif
1566	ASMAtomicWriteSize(pPteDst, PteDst.u);
1567	}
1568	/* else MMIO or invalid page, we must handle them manually in the #PF handler. */
1569	/** @todo count these. */
1570	}
1571	else
1572	{
1573	/*
1574	* Page not-present.
1575	*/
1576	Log2(("SyncPageWorker: page not present in Pte\n"));
1577	/* Keep user track up to date. */
1578	if (pPteDst->n.u1Present)
1579	{
1580	Log2(("SyncPageWorker: deref! *pPteDst=%RX64\n", (uint64_t)pPteDst->u));
1581	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pVCpu, pShwPage, pPteDst->u & SHW_PTE_PG_MASK, iPTDst);
1582	}
1583	ASMAtomicWriteSize(pPteDst, 0);
1584	/** @todo count these. */
1585	}
1586	}
1587
1588
1589	/**
1590	* Syncs a guest OS page.
1591	*
1592	* There are no conflicts at this point, neither is there any need for
1593	* page table allocations.
1594	*
1595	* @returns VBox status code.
1596	* @returns VINF_PGM_SYNCPAGE_MODIFIED_PDE if it modifies the PDE in any way.
1597	* @param pVCpu The VMCPU handle.
1598	* @param PdeSrc Page directory entry of the guest.
1599	* @param GCPtrPage Guest context page address.
1600	* @param cPages Number of pages to sync (PGM_SYNC_N_PAGES) (default=1).
1601	* @param uErr Fault error (X86_TRAP_PF_*).
1602	*/
1603	PGM_BTH_DECL(int, SyncPage)(PVMCPU pVCpu, GSTPDE PdeSrc, RTGCPTR GCPtrPage, unsigned cPages, unsigned uErr)
1604	{
1605	PVM pVM = pVCpu->CTX_SUFF(pVM);
1606	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1607	LogFlow(("SyncPage: GCPtrPage=%RGv cPages=%u uErr=%#x\n", GCPtrPage, cPages, uErr));
1608
1609	Assert(PGMIsLockOwner(pVM));
1610
1611	#if ( PGM_GST_TYPE == PGM_TYPE_32BIT \
1612	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
1613	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64) \
1614	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
1615	&& PGM_SHW_TYPE != PGM_TYPE_EPT
1616
1617	/*
1618	* Assert preconditions.
1619	*/
1620	Assert(PdeSrc.n.u1Present);
1621	Assert(cPages);
1622	# if 0 /* rarely useful; leave for debugging. */
1623	STAM_COUNTER_INC(&pVCpu->pgm.s.StatSyncPagePD[(GCPtrPage >> GST_PD_SHIFT) & GST_PD_MASK]);
1624	# endif
1625
1626	/*
1627	* Get the shadow PDE, find the shadow page table in the pool.
1628	*/
1629	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
1630	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1631	PX86PDE pPdeDst = pgmShwGet32BitPDEPtr(&pVCpu->pgm.s, GCPtrPage);
1632
1633	/* Fetch the pgm pool shadow descriptor. */
1634	PPGMPOOLPAGE pShwPde = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
1635	Assert(pShwPde);
1636
1637	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
1638	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1639	PPGMPOOLPAGE pShwPde = NULL;
1640	PX86PDPAE pPDDst;
1641
1642	/* Fetch the pgm pool shadow descriptor. */
1643	int rc2 = pgmShwGetPaePoolPagePD(&pVCpu->pgm.s, GCPtrPage, &pShwPde);
1644	AssertRCSuccessReturn(rc2, rc2);
1645	Assert(pShwPde);
1646
1647	pPDDst = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_BY_PGM(&pVM->pgm.s, pShwPde);
1648	PX86PDEPAE pPdeDst = &pPDDst->a[iPDDst];
1649
1650	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
1651	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1652	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
1653	PX86PDPAE pPDDst = NULL; /* initialized to shut up gcc */
1654	PX86PDPT pPdptDst = NULL; /* initialized to shut up gcc */
1655
1656	int rc2 = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, NULL, &pPdptDst, &pPDDst);
1657	AssertRCSuccessReturn(rc2, rc2);
1658	Assert(pPDDst && pPdptDst);
1659	PX86PDEPAE pPdeDst = &pPDDst->a[iPDDst];
1660	# endif
1661	SHWPDE PdeDst = *pPdeDst;
1662
1663	/* In the guest SMP case we could have blocked while another VCPU reused this page table. */
1664	if (!PdeDst.n.u1Present)
1665	{
1666	AssertMsg(pVM->cCpus > 1, ("Unexpected missing PDE p=%p/%RX64\n", pPdeDst, (uint64_t)PdeDst.u));
1667	Log(("CPU%d: SyncPage: Pde at %RGv changed behind our back!\n", pVCpu->idCpu, GCPtrPage));
1668	return VINF_SUCCESS; /* force the instruction to be executed again. */
1669	}
1670
1671	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, PdeDst.u & SHW_PDE_PG_MASK);
1672	Assert(pShwPage);
1673
1674	# if PGM_GST_TYPE == PGM_TYPE_AMD64
1675	/* Fetch the pgm pool shadow descriptor. */
1676	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & X86_PDPE_PG_MASK);
1677	Assert(pShwPde);
1678	# endif
1679
1680	# if defined(IN_RC)
1681	/* Make sure the dynamic pPdeDst mapping will not be reused during this function. */
1682	PGMDynLockHCPage(pVM, (uint8_t *)pPdeDst);
1683	# endif
1684
1685	/*
1686	* Check that the page is present and that the shadow PDE isn't out of sync.
1687	*/
1688	# if PGM_GST_TYPE == PGM_TYPE_32BIT
1689	const bool fBigPage = PdeSrc.b.u1Size && CPUMIsGuestPageSizeExtEnabled(pVCpu);
1690	# else
1691	const bool fBigPage = PdeSrc.b.u1Size;
1692	# endif
1693	RTGCPHYS GCPhys;
1694	if (!fBigPage)
1695	{
1696	GCPhys = PdeSrc.u & GST_PDE_PG_MASK;
1697	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
1698	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
1699	GCPhys \|= (iPDDst & 1) * (PAGE_SIZE / 2);
1700	# endif
1701	}
1702	else
1703	{
1704	GCPhys = GST_GET_PDE_BIG_PG_GCPHYS(PdeSrc);
1705	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
1706	/* Select the right PDE as we're emulating a 4MB page directory with two 2 MB shadow PDEs.*/
1707	GCPhys \|= GCPtrPage & (1 << X86_PD_PAE_SHIFT);
1708	# endif
1709	}
1710	if ( pShwPage->GCPhys == GCPhys
1711	&& PdeSrc.n.u1Present
1712	&& (PdeSrc.n.u1User == PdeDst.n.u1User)
1713	&& (PdeSrc.n.u1Write == PdeDst.n.u1Write \|\| !PdeDst.n.u1Write)
1714	# if PGM_WITH_NX(PGM_GST_TYPE, PGM_SHW_TYPE)
1715	&& (PdeSrc.n.u1NoExecute == PdeDst.n.u1NoExecute \|\| !CPUMIsGuestNXEnabled(pVCpu))
1716	# endif
1717	)
1718	{
1719	/*
1720	* Check that the PDE is marked accessed already.
1721	* Since we set the accessed bit before getting here on a #PF, this
1722	* check is only meant for dealing with non-#PF'ing paths.
1723	*/
1724	if (PdeSrc.n.u1Accessed)
1725	{
1726	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR(pVM, pShwPage);
1727	if (!fBigPage)
1728	{
1729	/*
1730	* 4KB Page - Map the guest page table.
1731	*/
1732	PGSTPT pPTSrc;
1733	int rc = PGM_GCPHYS_2_PTR(pVM, PdeSrc.u & GST_PDE_PG_MASK, &pPTSrc);
1734	if (RT_SUCCESS(rc))
1735	{
1736	# ifdef PGM_SYNC_N_PAGES
1737	Assert(cPages == 1 \|\| !(uErr & X86_TRAP_PF_P));
1738	if ( cPages > 1
1739	&& !(uErr & X86_TRAP_PF_P)
1740	&& !VM_FF_ISPENDING(pVM, VM_FF_PGM_NO_MEMORY))
1741	{
1742	/*
1743	* This code path is currently only taken when the caller is PGMTrap0eHandler
1744	* for non-present pages!
1745	*
1746	* We're setting PGM_SYNC_NR_PAGES pages around the faulting page to sync it and
1747	* deal with locality.
1748	*/
1749	unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
1750	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
1751	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
1752	const unsigned offPTSrc = ((GCPtrPage >> SHW_PD_SHIFT) & 1) * 512;
1753	# else
1754	const unsigned offPTSrc = 0;
1755	# endif
1756	const unsigned iPTDstEnd = RT_MIN(iPTDst + PGM_SYNC_NR_PAGES / 2, RT_ELEMENTS(pPTDst->a));
1757	if (iPTDst < PGM_SYNC_NR_PAGES / 2)
1758	iPTDst = 0;
1759	else
1760	iPTDst -= PGM_SYNC_NR_PAGES / 2;
1761	for (; iPTDst < iPTDstEnd; iPTDst++)
1762	{
1763	if (!pPTDst->a[iPTDst].n.u1Present)
1764	{
1765	GSTPTE PteSrc = pPTSrc->a[offPTSrc + iPTDst];
1766	RTGCPTR GCPtrCurPage = (GCPtrPage & ~(RTGCPTR)(GST_PT_MASK << GST_PT_SHIFT)) \| ((offPTSrc + iPTDst) << PAGE_SHIFT);
1767	NOREF(GCPtrCurPage);
1768	#ifndef IN_RING0
1769	/*
1770	* Assuming kernel code will be marked as supervisor - and not as user level
1771	* and executed using a conforming code selector - And marked as readonly.
1772	* Also assume that if we're monitoring a page, it's of no interest to CSAM.
1773	*/
1774	PPGMPAGE pPage;
1775	if ( ((PdeSrc.u & PteSrc.u) & (X86_PTE_RW \| X86_PTE_US))
1776	\|\| iPTDst == ((GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK) /* always sync GCPtrPage */
1777	\|\| !CSAMDoesPageNeedScanning(pVM, GCPtrCurPage)
1778	\|\| ( (pPage = pgmPhysGetPage(&pVM->pgm.s, PteSrc.u & GST_PTE_PG_MASK))
1779	&& PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
1780	)
1781	#endif /* else: CSAM not active */
1782	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, PteSrc, pShwPage, iPTDst);
1783	Log2(("SyncPage: 4K+ %RGv PteSrc:{P=%d RW=%d U=%d raw=%08llx} PteDst=%08llx%s\n",
1784	GCPtrCurPage, PteSrc.n.u1Present,
1785	PteSrc.n.u1Write & PdeSrc.n.u1Write,
1786	PteSrc.n.u1User & PdeSrc.n.u1User,
1787	(uint64_t)PteSrc.u,
1788	(uint64_t)pPTDst->a[iPTDst].u,
1789	pPTDst->a[iPTDst].u & PGM_PTFLAGS_TRACK_DIRTY ? " Track-Dirty" : ""));
1790	}
1791	}
1792	}
1793	else
1794	# endif /* PGM_SYNC_N_PAGES */
1795	{
1796	const unsigned iPTSrc = (GCPtrPage >> GST_PT_SHIFT) & GST_PT_MASK;
1797	GSTPTE PteSrc = pPTSrc->a[iPTSrc];
1798	const unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
1799	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, PteSrc, pShwPage, iPTDst);
1800	Log2(("SyncPage: 4K %RGv PteSrc:{P=%d RW=%d U=%d raw=%08llx} PteDst=%08llx %s\n",
1801	GCPtrPage, PteSrc.n.u1Present,
1802	PteSrc.n.u1Write & PdeSrc.n.u1Write,
1803	PteSrc.n.u1User & PdeSrc.n.u1User,
1804	(uint64_t)PteSrc.u,
1805	(uint64_t)pPTDst->a[iPTDst].u,
1806	pPTDst->a[iPTDst].u & PGM_PTFLAGS_TRACK_DIRTY ? " Track-Dirty" : ""));
1807	}
1808	}
1809	else /* MMIO or invalid page: emulated in #PF handler. */
1810	{
1811	LogFlow(("PGM_GCPHYS_2_PTR %RGp failed with %Rrc\n", GCPhys, rc));
1812	Assert(!pPTDst->a[(GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK].n.u1Present);
1813	}
1814	}
1815	else
1816	{
1817	/*
1818	* 4/2MB page - lazy syncing shadow 4K pages.
1819	* (There are many causes of getting here, it's no longer only CSAM.)
1820	*/
1821	/* Calculate the GC physical address of this 4KB shadow page. */
1822	GCPhys = GST_GET_PDE_BIG_PG_GCPHYS(PdeSrc) \| (GCPtrPage & GST_BIG_PAGE_OFFSET_MASK);
1823	/* Find ram range. */
1824	PPGMPAGE pPage;
1825	int rc = pgmPhysGetPageEx(&pVM->pgm.s, GCPhys, &pPage);
1826	if (RT_SUCCESS(rc))
1827	{
1828	AssertFatalMsg(!PGM_PAGE_IS_BALLOONED(pPage), ("Unexpected ballooned page at %RGp\n", GCPhys));
1829
1830	# ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
1831	/* Try to make the page writable if necessary. */
1832	if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM
1833	&& ( PGM_PAGE_IS_ZERO(pPage)
1834	\|\| ( PdeSrc.n.u1Write
1835	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED
1836	# ifdef VBOX_WITH_REAL_WRITE_MONITORED_PAGES
1837	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_WRITE_MONITORED
1838	# endif
1839	# ifdef VBOX_WITH_PAGE_SHARING
1840	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_SHARED
1841	# endif
1842	)
1843	)
1844	)
1845	{
1846	rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhys);
1847	AssertRC(rc);
1848	}
1849	# endif
1850
1851	/*
1852	* Make shadow PTE entry.
1853	*/
1854	SHWPTE PteDst;
1855	PteDst.u = (PdeSrc.u & ~(X86_PTE_PAE_PG_MASK \| X86_PTE_AVL_MASK \| X86_PTE_PAT \| X86_PTE_PCD \| X86_PTE_PWT))
1856	\| PGM_PAGE_GET_HCPHYS(pPage);
1857	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
1858	{
1859	if (!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage))
1860	PteDst.n.u1Write = 0;
1861	else
1862	PteDst.u = 0;
1863	}
1864
1865	const unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
1866	if ( PteDst.n.u1Present
1867	&& !pPTDst->a[iPTDst].n.u1Present)
1868	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPTDst);
1869
1870	/* Make sure only allocated pages are mapped writable. */
1871	if ( PteDst.n.u1Write
1872	&& PteDst.n.u1Present
1873	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
1874	{
1875	/* Still applies to shared pages. */
1876	Assert(!PGM_PAGE_IS_ZERO(pPage));
1877	PteDst.n.u1Write = 0; /** @todo this isn't quite working yet... */
1878	Log3(("SyncPage: write-protecting %RGp pPage=%R[pgmpage] at %RGv\n", GCPhys, pPage, GCPtrPage));
1879	}
1880
1881	ASMAtomicWriteSize(&pPTDst->a[iPTDst], PteDst.u);
1882
1883	/*
1884	* If the page is not flagged as dirty and is writable, then make it read-only
1885	* at PD level, so we can set the dirty bit when the page is modified.
1886	*
1887	* ASSUMES that page access handlers are implemented on page table entry level.
1888	* Thus we will first catch the dirty access and set PDE.D and restart. If
1889	* there is an access handler, we'll trap again and let it work on the problem.
1890	*/
1891	/** @todo r=bird: figure out why we need this here, SyncPT should've taken care of this already.
1892	* As for invlpg, it simply frees the whole shadow PT.
1893	* ...It's possibly because the guest clears it and the guest doesn't really tell us... */
1894	if ( !PdeSrc.b.u1Dirty
1895	&& PdeSrc.b.u1Write)
1896	{
1897	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyPageBig));
1898	PdeDst.u \|= PGM_PDFLAGS_TRACK_DIRTY;
1899	PdeDst.n.u1Write = 0;
1900	}
1901	else
1902	{
1903	PdeDst.au32[0] &= ~PGM_PDFLAGS_TRACK_DIRTY;
1904	PdeDst.n.u1Write = PdeSrc.n.u1Write;
1905	}
1906	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
1907	Log2(("SyncPage: BIG %RGv PdeSrc:{P=%d RW=%d U=%d raw=%08llx} GCPhys=%RGp%s\n",
1908	GCPtrPage, PdeSrc.n.u1Present, PdeSrc.n.u1Write, PdeSrc.n.u1User, (uint64_t)PdeSrc.u, GCPhys,
1909	PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY ? " Track-Dirty" : ""));
1910	}
1911	else
1912	LogFlow(("PGM_GCPHYS_2_PTR %RGp (big) failed with %Rrc\n", GCPhys, rc));
1913	}
1914	# if defined(IN_RC)
1915	/* Make sure the dynamic pPdeDst mapping will not be reused during this function. */
1916	PGMDynUnlockHCPage(pVM, (uint8_t *)pPdeDst);
1917	# endif
1918	return VINF_SUCCESS;
1919	}
1920	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPagePDNAs));
1921	}
1922	else
1923	{
1924	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPagePDOutOfSync));
1925	Log2(("SyncPage: Out-Of-Sync PDE at %RGp PdeSrc=%RX64 PdeDst=%RX64 (GCPhys %RGp vs %RGp)\n",
1926	GCPtrPage, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u, pShwPage->GCPhys, GCPhys));
1927	}
1928
1929	/*
1930	* Mark the PDE not present. Restart the instruction and let #PF call SyncPT.
1931	* Yea, I'm lazy.
1932	*/
1933	pgmPoolFreeByPage(pPool, pShwPage, pShwPde->idx, iPDDst);
1934	ASMAtomicWriteSize(pPdeDst, 0);
1935
1936	# if defined(IN_RC)
1937	/* Make sure the dynamic pPdeDst mapping will not be reused during this function. */
1938	PGMDynUnlockHCPage(pVM, (uint8_t *)pPdeDst);
1939	# endif
1940	PGM_INVL_VCPU_TLBS(pVCpu);
1941	return VINF_PGM_SYNCPAGE_MODIFIED_PDE;
1942
1943	#elif (PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
1944	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
1945	&& (PGM_SHW_TYPE != PGM_TYPE_EPT \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
1946	&& !defined(IN_RC)
1947
1948	# ifdef PGM_SYNC_N_PAGES
1949	/*
1950	* Get the shadow PDE, find the shadow page table in the pool.
1951	*/
1952	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
1953	X86PDE PdeDst = pgmShwGet32BitPDE(&pVCpu->pgm.s, GCPtrPage);
1954
1955	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
1956	X86PDEPAE PdeDst = pgmShwGetPaePDE(&pVCpu->pgm.s, GCPtrPage);
1957
1958	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
1959	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
1960	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64; NOREF(iPdpt);
1961	PX86PDPAE pPDDst = NULL; /* initialized to shut up gcc */
1962	X86PDEPAE PdeDst;
1963	PX86PDPT pPdptDst = NULL; /* initialized to shut up gcc */
1964
1965	int rc = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, NULL, &pPdptDst, &pPDDst);
1966	AssertRCSuccessReturn(rc, rc);
1967	Assert(pPDDst && pPdptDst);
1968	PdeDst = pPDDst->a[iPDDst];
1969	# elif PGM_SHW_TYPE == PGM_TYPE_EPT
1970	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
1971	PEPTPD pPDDst;
1972	EPTPDE PdeDst;
1973
1974	int rc = pgmShwGetEPTPDPtr(pVCpu, GCPtrPage, NULL, &pPDDst);
1975	if (rc != VINF_SUCCESS)
1976	{
1977	AssertRC(rc);
1978	return rc;
1979	}
1980	Assert(pPDDst);
1981	PdeDst = pPDDst->a[iPDDst];
1982	# endif
1983	/* In the guest SMP case we could have blocked while another VCPU reused this page table. */
1984	if (!PdeDst.n.u1Present)
1985	{
1986	AssertMsg(pVM->cCpus > 1, ("Unexpected missing PDE %RX64\n", (uint64_t)PdeDst.u));
1987	Log(("CPU%d: SyncPage: Pde at %RGv changed behind our back!\n", pVCpu->idCpu, GCPtrPage));
1988	return VINF_SUCCESS; /* force the instruction to be executed again. */
1989	}
1990
1991	/* Can happen in the guest SMP case; other VCPU activated this PDE while we were blocking to handle the page fault. */
1992	if (PdeDst.n.u1Size)
1993	{
1994	Assert(HWACCMIsNestedPagingActive(pVM));
1995	Log(("CPU%d: SyncPage: Pde (big:%RX64) at %RGv changed behind our back!\n", pVCpu->idCpu, PdeDst.u, GCPtrPage));
1996	return VINF_SUCCESS;
1997	}
1998
1999	/* Mask away the page offset. */
2000	GCPtrPage &= ~((RTGCPTR)0xfff);
2001
2002	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, PdeDst.u & SHW_PDE_PG_MASK);
2003	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR(pVM, pShwPage);
2004
2005	Assert(cPages == 1 \|\| !(uErr & X86_TRAP_PF_P));
2006	if ( cPages > 1
2007	&& !(uErr & X86_TRAP_PF_P)
2008	&& !VM_FF_ISPENDING(pVM, VM_FF_PGM_NO_MEMORY))
2009	{
2010	/*
2011	* This code path is currently only taken when the caller is PGMTrap0eHandler
2012	* for non-present pages!
2013	*
2014	* We're setting PGM_SYNC_NR_PAGES pages around the faulting page to sync it and
2015	* deal with locality.
2016	*/
2017	unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2018	const unsigned iPTDstEnd = RT_MIN(iPTDst + PGM_SYNC_NR_PAGES / 2, RT_ELEMENTS(pPTDst->a));
2019	if (iPTDst < PGM_SYNC_NR_PAGES / 2)
2020	iPTDst = 0;
2021	else
2022	iPTDst -= PGM_SYNC_NR_PAGES / 2;
2023	for (; iPTDst < iPTDstEnd; iPTDst++)
2024	{
2025	if (!pPTDst->a[iPTDst].n.u1Present)
2026	{
2027	GSTPTE PteSrc;
2028
2029	RTGCPTR GCPtrCurPage = (GCPtrPage & ~(RTGCPTR)(SHW_PT_MASK << SHW_PT_SHIFT)) \| (iPTDst << PAGE_SHIFT);
2030
2031	/* Fake the page table entry */
2032	PteSrc.u = GCPtrCurPage;
2033	PteSrc.n.u1Present = 1;
2034	PteSrc.n.u1Dirty = 1;
2035	PteSrc.n.u1Accessed = 1;
2036	PteSrc.n.u1Write = 1;
2037	PteSrc.n.u1User = 1;
2038
2039	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, PteSrc, pShwPage, iPTDst);
2040
2041	Log2(("SyncPage: 4K+ %RGv PteSrc:{P=%d RW=%d U=%d raw=%08llx} PteDst=%08llx%s\n",
2042	GCPtrCurPage, PteSrc.n.u1Present,
2043	PteSrc.n.u1Write & PdeSrc.n.u1Write,
2044	PteSrc.n.u1User & PdeSrc.n.u1User,
2045	(uint64_t)PteSrc.u,
2046	(uint64_t)pPTDst->a[iPTDst].u,
2047	pPTDst->a[iPTDst].u & PGM_PTFLAGS_TRACK_DIRTY ? " Track-Dirty" : ""));
2048
2049	if (RT_UNLIKELY(VM_FF_ISPENDING(pVM, VM_FF_PGM_NO_MEMORY)))
2050	break;
2051	}
2052	else
2053	Log4(("%RGv iPTDst=%x pPTDst->a[iPTDst] %RX64\n", (GCPtrPage & ~(RTGCPTR)(SHW_PT_MASK << SHW_PT_SHIFT)) \| (iPTDst << PAGE_SHIFT), iPTDst, pPTDst->a[iPTDst].u));
2054	}
2055	}
2056	else
2057	# endif /* PGM_SYNC_N_PAGES */
2058	{
2059	GSTPTE PteSrc;
2060	const unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2061	RTGCPTR GCPtrCurPage = (GCPtrPage & ~(RTGCPTR)(SHW_PT_MASK << SHW_PT_SHIFT)) \| (iPTDst << PAGE_SHIFT);
2062
2063	/* Fake the page table entry */
2064	PteSrc.u = GCPtrCurPage;
2065	PteSrc.n.u1Present = 1;
2066	PteSrc.n.u1Dirty = 1;
2067	PteSrc.n.u1Accessed = 1;
2068	PteSrc.n.u1Write = 1;
2069	PteSrc.n.u1User = 1;
2070	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, PteSrc, pShwPage, iPTDst);
2071
2072	Log2(("SyncPage: 4K %RGv PteSrc:{P=%d RW=%d U=%d raw=%08llx}PteDst=%08llx%s\n",
2073	GCPtrPage, PteSrc.n.u1Present,
2074	PteSrc.n.u1Write & PdeSrc.n.u1Write,
2075	PteSrc.n.u1User & PdeSrc.n.u1User,
2076	(uint64_t)PteSrc.u,
2077	(uint64_t)pPTDst->a[iPTDst].u,
2078	pPTDst->a[iPTDst].u & PGM_PTFLAGS_TRACK_DIRTY ? " Track-Dirty" : ""));
2079	}
2080	return VINF_SUCCESS;
2081
2082	#else
2083	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_GST_TYPE, PGM_SHW_TYPE));
2084	return VERR_INTERNAL_ERROR;
2085	#endif
2086	}
2087
2088
2089	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
2090	/**
2091	* Investigate page fault and handle write protection page faults caused by
2092	* dirty bit tracking.
2093	*
2094	* @returns VBox status code.
2095	* @param pVCpu The VMCPU handle.
2096	* @param uErr Page fault error code.
2097	* @param pPdeSrc Guest page directory entry.
2098	* @param GCPtrPage Guest context page address.
2099	*/
2100	PGM_BTH_DECL(int, CheckPageFault)(PVMCPU pVCpu, uint32_t uErr, PGSTPDE pPdeSrc, RTGCPTR GCPtrPage)
2101	{
2102	bool fUserLevelFault = !!(uErr & X86_TRAP_PF_US);
2103	bool fWriteFault = !!(uErr & X86_TRAP_PF_RW);
2104	bool fMaybeWriteProtFault = fWriteFault && (fUserLevelFault \|\| CPUMIsGuestR0WriteProtEnabled(pVCpu));
2105	# if PGM_WITH_NX(PGM_GST_TYPE, PGM_SHW_TYPE)
2106	bool fMaybeNXEFault = (uErr & X86_TRAP_PF_ID) && CPUMIsGuestNXEnabled(pVCpu);
2107	# endif
2108	unsigned uPageFaultLevel;
2109	int rc;
2110	PVM pVM = pVCpu->CTX_SUFF(pVM);
2111
2112	LogFlow(("CheckPageFault: GCPtrPage=%RGv uErr=%#x PdeSrc=%08x\n", GCPtrPage, uErr, pPdeSrc->u));
2113
2114	# if PGM_GST_TYPE == PGM_TYPE_PAE \
2115	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64
2116
2117	# if PGM_GST_TYPE == PGM_TYPE_AMD64
2118	PX86PML4E pPml4eSrc;
2119	PX86PDPE pPdpeSrc;
2120
2121	pPdpeSrc = pgmGstGetLongModePDPTPtr(&pVCpu->pgm.s, GCPtrPage, &pPml4eSrc);
2122	Assert(pPml4eSrc);
2123
2124	/*
2125	* Real page fault? (PML4E level)
2126	*/
2127	if ( (uErr & X86_TRAP_PF_RSVD)
2128	\|\| !pPml4eSrc->n.u1Present
2129	\|\| (fMaybeWriteProtFault && !pPml4eSrc->n.u1Write)
2130	\|\| (fMaybeNXEFault && pPml4eSrc->n.u1NoExecute)
2131	\|\| (fUserLevelFault && !pPml4eSrc->n.u1User)
2132	)
2133	{
2134	uPageFaultLevel = 0;
2135	goto l_UpperLevelPageFault;
2136	}
2137	Assert(pPdpeSrc);
2138
2139	# else /* PAE */
2140	PX86PDPE pPdpeSrc = pgmGstGetPaePDPEPtr(&pVCpu->pgm.s, GCPtrPage);
2141	# endif /* PAE */
2142
2143	/*
2144	* Real page fault? (PDPE level)
2145	*/
2146	if ( (uErr & X86_TRAP_PF_RSVD)
2147	\|\| !pPdpeSrc->n.u1Present
2148	# if PGM_GST_TYPE == PGM_TYPE_AMD64 /* NX, r/w, u/s bits in the PDPE are long mode only */
2149	\|\| (fMaybeWriteProtFault && !pPdpeSrc->lm.u1Write)
2150	\|\| (fMaybeNXEFault && pPdpeSrc->lm.u1NoExecute)
2151	\|\| (fUserLevelFault && !pPdpeSrc->lm.u1User)
2152	# endif
2153	)
2154	{
2155	uPageFaultLevel = 1;
2156	goto l_UpperLevelPageFault;
2157	}
2158	# endif
2159
2160	/*
2161	* Real page fault? (PDE level)
2162	*/
2163	if ( (uErr & X86_TRAP_PF_RSVD)
2164	\|\| !pPdeSrc->n.u1Present
2165	\|\| (fMaybeWriteProtFault && !pPdeSrc->n.u1Write)
2166	# if PGM_WITH_NX(PGM_GST_TYPE, PGM_SHW_TYPE)
2167	\|\| (fMaybeNXEFault && pPdeSrc->n.u1NoExecute)
2168	# endif
2169	\|\| (fUserLevelFault && !pPdeSrc->n.u1User) )
2170	{
2171	uPageFaultLevel = 2;
2172	goto l_UpperLevelPageFault;
2173	}
2174
2175	/*
2176	* First check the easy case where the page directory has been marked read-only to track
2177	* the dirty bit of an emulated BIG page
2178	*/
2179	if ( pPdeSrc->b.u1Size
2180	# if PGM_GST_TYPE == PGM_TYPE_32BIT
2181	&& CPUMIsGuestPageSizeExtEnabled(pVCpu)
2182	# endif
2183	)
2184	{
2185	/* Mark guest page directory as accessed */
2186	# if PGM_GST_TYPE == PGM_TYPE_AMD64
2187	pPml4eSrc->n.u1Accessed = 1;
2188	pPdpeSrc->lm.u1Accessed = 1;
2189	# endif
2190	pPdeSrc->b.u1Accessed = 1;
2191
2192	/*
2193	* Only write protection page faults are relevant here.
2194	*/
2195	if (fWriteFault)
2196	{
2197	/* Mark guest page directory as dirty (BIG page only). */
2198	pPdeSrc->b.u1Dirty = 1;
2199	}
2200	return VINF_SUCCESS;
2201	}
2202	/* else: 4KB page table */
2203
2204	/*
2205	* Map the guest page table.
2206	*/
2207	PGSTPT pPTSrc;
2208	rc = PGM_GCPHYS_2_PTR(pVM, pPdeSrc->u & GST_PDE_PG_MASK, &pPTSrc);
2209	if (RT_SUCCESS(rc))
2210	{
2211	/*
2212	* Real page fault?
2213	*/
2214	PGSTPTE pPteSrc = &pPTSrc->a[(GCPtrPage >> GST_PT_SHIFT) & GST_PT_MASK];
2215	const GSTPTE PteSrc = *pPteSrc;
2216	if ( !PteSrc.n.u1Present
2217	\|\| (fMaybeWriteProtFault && !PteSrc.n.u1Write)
2218	# if PGM_WITH_NX(PGM_GST_TYPE, PGM_SHW_TYPE)
2219	\|\| (fMaybeNXEFault && PteSrc.n.u1NoExecute)
2220	# endif
2221	\|\| (fUserLevelFault && !PteSrc.n.u1User)
2222	)
2223	{
2224	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyTrackRealPF));
2225	LogFlow(("CheckPageFault: real page fault at %RGv PteSrc.u=%08x (2)\n", GCPtrPage, PteSrc.u));
2226
2227	/* Check the present bit as the shadow tables can cause different error codes by being out of sync.
2228	* See the 2nd case above as well.
2229	*/
2230	if (pPdeSrc->n.u1Present && pPteSrc->n.u1Present)
2231	TRPMSetErrorCode(pVCpu, uErr \| X86_TRAP_PF_P); /* page-level protection violation */
2232
2233	return VINF_EM_RAW_GUEST_TRAP;
2234	}
2235	LogFlow(("CheckPageFault: page fault at %RGv PteSrc.u=%08x\n", GCPtrPage, PteSrc.u));
2236
2237	/*
2238	* Set the accessed bits in the page directory and the page table.
2239	*/
2240	# if PGM_GST_TYPE == PGM_TYPE_AMD64
2241	pPml4eSrc->n.u1Accessed = 1;
2242	pPdpeSrc->lm.u1Accessed = 1;
2243	# endif
2244	pPdeSrc->n.u1Accessed = 1;
2245	pPteSrc->n.u1Accessed = 1;
2246
2247	/*
2248	* Only write protection page faults are relevant here.
2249	*/
2250	if (fWriteFault)
2251	{
2252	/* Write access, so mark guest entry as dirty. */
2253	# ifdef VBOX_WITH_STATISTICS
2254	if (!pPteSrc->n.u1Dirty)
2255	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtiedPage));
2256	else
2257	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,PageAlreadyDirty));
2258	# endif
2259
2260	pPteSrc->n.u1Dirty = 1;
2261	}
2262	return VINF_SUCCESS;
2263	}
2264	AssertRC(rc);
2265	return rc;
2266
2267
2268	l_UpperLevelPageFault:
2269	/*
2270	* Pagefault detected while checking the PML4E, PDPE or PDE.
2271	* Single exit handler to get rid of duplicate code paths.
2272	*/
2273	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyTrackRealPF));
2274	Log(("CheckPageFault: real page fault at %RGv (%d)\n", GCPtrPage, uPageFaultLevel));
2275
2276	if ( 1
2277	# if PGM_GST_TYPE == PGM_TYPE_AMD64
2278	&& pPml4eSrc->n.u1Present
2279	# endif
2280	# if PGM_GST_TYPE == PGM_TYPE_AMD64 \|\| PGM_GST_TYPE == PGM_TYPE_PAE
2281	&& pPdpeSrc->n.u1Present
2282	# endif
2283	&& pPdeSrc->n.u1Present)
2284	{
2285	/* Check the present bit as the shadow tables can cause different error codes by being out of sync. */
2286	if ( pPdeSrc->b.u1Size
2287	# if PGM_GST_TYPE == PGM_TYPE_32BIT
2288	&& CPUMIsGuestPageSizeExtEnabled(pVCpu)
2289	# endif
2290	)
2291	{
2292	TRPMSetErrorCode(pVCpu, uErr \| X86_TRAP_PF_P); /* page-level protection violation */
2293	}
2294	else
2295	{
2296	/*
2297	* Map the guest page table.
2298	*/
2299	PGSTPT pPTSrc2;
2300	rc = PGM_GCPHYS_2_PTR(pVM, pPdeSrc->u & GST_PDE_PG_MASK, &pPTSrc2);
2301	if (RT_SUCCESS(rc))
2302	{
2303	PGSTPTE pPteSrc = &pPTSrc2->a[(GCPtrPage >> GST_PT_SHIFT) & GST_PT_MASK];
2304	if (pPteSrc->n.u1Present)
2305	TRPMSetErrorCode(pVCpu, uErr \| X86_TRAP_PF_P); /* page-level protection violation */
2306	}
2307	AssertRC(rc);
2308	}
2309	}
2310	return VINF_EM_RAW_GUEST_TRAP;
2311	}
2312
2313	/**
2314	* Handle dirty bit tracking faults.
2315	*
2316	* @returns VBox status code.
2317	* @param pVCpu The VMCPU handle.
2318	* @param uErr Page fault error code.
2319	* @param pPdeSrc Guest page directory entry.
2320	* @param pPdeDst Shadow page directory entry.
2321	* @param GCPtrPage Guest context page address.
2322	*/
2323	PGM_BTH_DECL(int, CheckDirtyPageFault)(PVMCPU pVCpu, uint32_t uErr, PSHWPDE pPdeDst, PGSTPDE pPdeSrc, RTGCPTR GCPtrPage)
2324	{
2325	# if PGM_GST_TYPE == PGM_TYPE_32BIT
2326	const bool fBigPagesSupported = CPUMIsGuestPageSizeExtEnabled(pVCpu);
2327	# else
2328	const bool fBigPagesSupported = true;
2329	# endif
2330	PVM pVM = pVCpu->CTX_SUFF(pVM);
2331	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2332
2333	Assert(PGMIsLockOwner(pVM));
2334
2335	if (pPdeSrc->b.u1Size && fBigPagesSupported)
2336	{
2337	if ( pPdeDst->n.u1Present
2338	&& (pPdeDst->u & PGM_PDFLAGS_TRACK_DIRTY))
2339	{
2340	SHWPDE PdeDst = *pPdeDst;
2341
2342	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyPageTrap));
2343	Assert(pPdeSrc->b.u1Write);
2344
2345	/* Note: No need to invalidate this entry on other VCPUs as a stale TLB entry will not harm; write access will simply
2346	* fault again and take this path to only invalidate the entry.
2347	*/
2348	PdeDst.n.u1Write = 1;
2349	PdeDst.n.u1Accessed = 1;
2350	PdeDst.au32[0] &= ~PGM_PDFLAGS_TRACK_DIRTY;
2351	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
2352	PGM_INVL_BIG_PG(pVCpu, GCPtrPage);
2353	return VINF_PGM_HANDLED_DIRTY_BIT_FAULT; /* restarts the instruction. */
2354	}
2355	# ifdef IN_RING0
2356	else
2357	/* Check for stale TLB entry; only applies to the SMP guest case. */
2358	if ( pVM->cCpus > 1
2359	&& pPdeDst->n.u1Write
2360	&& pPdeDst->n.u1Accessed)
2361	{
2362	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, pPdeDst->u & SHW_PDE_PG_MASK);
2363	if (pShwPage)
2364	{
2365	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR(pVM, pShwPage);
2366	PSHWPTE pPteDst = &pPTDst->a[(GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK];
2367	if ( pPteDst->n.u1Present
2368	&& pPteDst->n.u1Write)
2369	{
2370	/* Stale TLB entry. */
2371	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyPageStale));
2372	PGM_INVL_PG(pVCpu, GCPtrPage);
2373	return VINF_PGM_HANDLED_DIRTY_BIT_FAULT; /* restarts the instruction. */
2374	}
2375	}
2376	}
2377	# endif /* IN_RING0 */
2378	return VINF_PGM_NO_DIRTY_BIT_TRACKING;
2379	}
2380
2381	/*
2382	* Map the guest page table.
2383	*/
2384	PGSTPT pPTSrc;
2385	int rc = PGM_GCPHYS_2_PTR(pVM, pPdeSrc->u & GST_PDE_PG_MASK, &pPTSrc);
2386	if (RT_SUCCESS(rc))
2387	{
2388	if (pPdeDst->n.u1Present)
2389	{
2390	PGSTPTE pPteSrc = &pPTSrc->a[(GCPtrPage >> GST_PT_SHIFT) & GST_PT_MASK];
2391	const GSTPTE PteSrc = *pPteSrc;
2392	#ifndef IN_RING0
2393	/* Bail out here as pgmPoolGetPageByHCPhys will return NULL and we'll crash below.
2394	* Our individual shadow handlers will provide more information and force a fatal exit.
2395	*/
2396	if (MMHyperIsInsideArea(pVM, (RTGCPTR)GCPtrPage))
2397	{
2398	LogRel(("CheckPageFault: write to hypervisor region %RGv\n", GCPtrPage));
2399	return VINF_PGM_NO_DIRTY_BIT_TRACKING;
2400	}
2401	#endif
2402	/*
2403	* Map shadow page table.
2404	*/
2405	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, pPdeDst->u & SHW_PDE_PG_MASK);
2406	if (pShwPage)
2407	{
2408	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR(pVM, pShwPage);
2409	PSHWPTE pPteDst = &pPTDst->a[(GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK];
2410	if (pPteDst->n.u1Present) /** @todo Optimize accessed bit emulation? */
2411	{
2412	if (pPteDst->u & PGM_PTFLAGS_TRACK_DIRTY)
2413	{
2414	PPGMPAGE pPage = pgmPhysGetPage(&pVM->pgm.s, pPteSrc->u & GST_PTE_PG_MASK);
2415	SHWPTE PteDst = *pPteDst;
2416
2417	LogFlow(("DIRTY page trap addr=%RGv\n", GCPtrPage));
2418	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyPageTrap));
2419
2420	Assert(pPteSrc->n.u1Write);
2421
2422	/* Note: No need to invalidate this entry on other VCPUs as a stale TLB entry will not harm; write access will simply
2423	* fault again and take this path to only invalidate the entry.
2424	*/
2425	if (RT_LIKELY(pPage))
2426	{
2427	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
2428	{
2429	/* Assuming write handlers here as the PTE is present (otherwise we wouldn't be here). */
2430	PteDst.n.u1Write = 0;
2431	}
2432	else
2433	{
2434	if ( PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_WRITE_MONITORED
2435	&& PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM)
2436	{
2437	rc = pgmPhysPageMakeWritable(pVM, pPage, pPteSrc->u & GST_PTE_PG_MASK);
2438	AssertRC(rc);
2439	}
2440	if (PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED)
2441	{
2442	PteDst.n.u1Write = 1;
2443	}
2444	else
2445	{
2446	/* Still applies to shared pages. */
2447	Assert(!PGM_PAGE_IS_ZERO(pPage));
2448	PteDst.n.u1Write = 0;
2449	}
2450	}
2451	}
2452	else
2453	PteDst.n.u1Write = 1;
2454
2455	PteDst.n.u1Dirty = 1;
2456	PteDst.n.u1Accessed = 1;
2457	PteDst.au32[0] &= ~PGM_PTFLAGS_TRACK_DIRTY;
2458	ASMAtomicWriteSize(pPteDst, PteDst.u);
2459	PGM_INVL_PG(pVCpu, GCPtrPage);
2460	return VINF_PGM_HANDLED_DIRTY_BIT_FAULT; /* restarts the instruction. */
2461	}
2462	# ifdef IN_RING0
2463	else
2464	/* Check for stale TLB entry; only applies to the SMP guest case. */
2465	if ( pVM->cCpus > 1
2466	&& pPteDst->n.u1Write == 1
2467	&& pPteDst->n.u1Accessed == 1)
2468	{
2469	/* Stale TLB entry. */
2470	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyPageStale));
2471	PGM_INVL_PG(pVCpu, GCPtrPage);
2472	return VINF_PGM_HANDLED_DIRTY_BIT_FAULT; /* restarts the instruction. */
2473	}
2474	# endif
2475	}
2476	}
2477	else
2478	AssertMsgFailed(("pgmPoolGetPageByHCPhys %RGp failed!\n", pPdeDst->u & SHW_PDE_PG_MASK));
2479	}
2480	return VINF_PGM_NO_DIRTY_BIT_TRACKING;
2481	}
2482	AssertRC(rc);
2483	return rc;
2484	}
2485	#endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
2486
2487
2488	/**
2489	* Sync a shadow page table.
2490	*
2491	* The shadow page table is not present. This includes the case where
2492	* there is a conflict with a mapping.
2493	*
2494	* @returns VBox status code.
2495	* @param pVCpu The VMCPU handle.
2496	* @param iPD Page directory index.
2497	* @param pPDSrc Source page directory (i.e. Guest OS page directory).
2498	* Assume this is a temporary mapping.
2499	* @param GCPtrPage GC Pointer of the page that caused the fault
2500	*/
2501	PGM_BTH_DECL(int, SyncPT)(PVMCPU pVCpu, unsigned iPDSrc, PGSTPD pPDSrc, RTGCPTR GCPtrPage)
2502	{
2503	PVM pVM = pVCpu->CTX_SUFF(pVM);
2504	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2505
2506	STAM_PROFILE_START(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPT), a);
2507	#if 0 /* rarely useful; leave for debugging. */
2508	STAM_COUNTER_INC(&pVCpu->pgm.s.StatSyncPtPD[iPDSrc]);
2509	#endif
2510	LogFlow(("SyncPT: GCPtrPage=%RGv\n", GCPtrPage));
2511
2512	Assert(PGMIsLocked(pVM));
2513
2514	#if ( PGM_GST_TYPE == PGM_TYPE_32BIT \
2515	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
2516	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64) \
2517	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
2518	&& PGM_SHW_TYPE != PGM_TYPE_EPT
2519
2520	int rc = VINF_SUCCESS;
2521
2522	/*
2523	* Validate input a little bit.
2524	*/
2525	AssertMsg(iPDSrc == ((GCPtrPage >> GST_PD_SHIFT) & GST_PD_MASK), ("iPDSrc=%x GCPtrPage=%RGv\n", iPDSrc, GCPtrPage));
2526	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
2527	const unsigned iPDDst = GCPtrPage >> SHW_PD_SHIFT;
2528	PSHWPDE pPdeDst = pgmShwGet32BitPDEPtr(&pVCpu->pgm.s, GCPtrPage);
2529
2530	/* Fetch the pgm pool shadow descriptor. */
2531	PPGMPOOLPAGE pShwPde = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
2532	Assert(pShwPde);
2533
2534	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
2535	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
2536	PPGMPOOLPAGE pShwPde = NULL;
2537	PX86PDPAE pPDDst;
2538	PSHWPDE pPdeDst;
2539
2540	/* Fetch the pgm pool shadow descriptor. */
2541	rc = pgmShwGetPaePoolPagePD(&pVCpu->pgm.s, GCPtrPage, &pShwPde);
2542	AssertRCSuccessReturn(rc, rc);
2543	Assert(pShwPde);
2544
2545	pPDDst = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_BY_PGM(&pVM->pgm.s, pShwPde);
2546	pPdeDst = &pPDDst->a[iPDDst];
2547
2548	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
2549	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
2550	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
2551	PX86PDPAE pPDDst = NULL; /* initialized to shut up gcc */
2552	PX86PDPT pPdptDst = NULL; /* initialized to shut up gcc */
2553	rc = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, NULL, &pPdptDst, &pPDDst);
2554	AssertRCSuccessReturn(rc, rc);
2555	Assert(pPDDst);
2556	PSHWPDE pPdeDst = &pPDDst->a[iPDDst];
2557	# endif
2558	SHWPDE PdeDst = *pPdeDst;
2559
2560	# if PGM_GST_TYPE == PGM_TYPE_AMD64
2561	/* Fetch the pgm pool shadow descriptor. */
2562	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & X86_PDPE_PG_MASK);
2563	Assert(pShwPde);
2564	# endif
2565
2566	# ifndef PGM_WITHOUT_MAPPINGS
2567	/*
2568	* Check for conflicts.
2569	* GC: In case of a conflict we'll go to Ring-3 and do a full SyncCR3.
2570	* HC: Simply resolve the conflict.
2571	*/
2572	if (PdeDst.u & PGM_PDFLAGS_MAPPING)
2573	{
2574	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
2575	# ifndef IN_RING3
2576	Log(("SyncPT: Conflict at %RGv\n", GCPtrPage));
2577	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPT), a);
2578	return VERR_ADDRESS_CONFLICT;
2579	# else
2580	PPGMMAPPING pMapping = pgmGetMapping(pVM, (RTGCPTR)GCPtrPage);
2581	Assert(pMapping);
2582	# if PGM_GST_TYPE == PGM_TYPE_32BIT
2583	rc = pgmR3SyncPTResolveConflict(pVM, pMapping, pPDSrc, GCPtrPage & (GST_PD_MASK << GST_PD_SHIFT));
2584	# elif PGM_GST_TYPE == PGM_TYPE_PAE
2585	rc = pgmR3SyncPTResolveConflictPAE(pVM, pMapping, GCPtrPage & (GST_PD_MASK << GST_PD_SHIFT));
2586	# else
2587	AssertFailed(); /* can't happen for amd64 */
2588	# endif
2589	if (RT_FAILURE(rc))
2590	{
2591	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPT), a);
2592	return rc;
2593	}
2594	PdeDst = *pPdeDst;
2595	# endif
2596	}
2597	# endif /* !PGM_WITHOUT_MAPPINGS */
2598	Assert(!PdeDst.n.u1Present); /* We're only supposed to call SyncPT on PDE!P and conflicts.*/
2599
2600	# if defined(IN_RC)
2601	/* Make sure the dynamic pPdeDst mapping will not be reused during this function. */
2602	PGMDynLockHCPage(pVM, (uint8_t *)pPdeDst);
2603	# endif
2604
2605	/*
2606	* Sync page directory entry.
2607	*/
2608	GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
2609	if (PdeSrc.n.u1Present)
2610	{
2611	/*
2612	* Allocate & map the page table.
2613	*/
2614	PSHWPT pPTDst;
2615	# if PGM_GST_TYPE == PGM_TYPE_32BIT
2616	const bool fPageTable = !PdeSrc.b.u1Size \|\| !CPUMIsGuestPageSizeExtEnabled(pVCpu);
2617	# else
2618	const bool fPageTable = !PdeSrc.b.u1Size;
2619	# endif
2620	PPGMPOOLPAGE pShwPage;
2621	RTGCPHYS GCPhys;
2622	if (fPageTable)
2623	{
2624	GCPhys = PdeSrc.u & GST_PDE_PG_MASK;
2625	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
2626	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
2627	GCPhys \|= (iPDDst & 1) * (PAGE_SIZE / 2);
2628	# endif
2629	rc = pgmPoolAlloc(pVM, GCPhys, BTH_PGMPOOLKIND_PT_FOR_PT, pShwPde->idx, iPDDst, &pShwPage);
2630	}
2631	else
2632	{
2633	PGMPOOLACCESS enmAccess;
2634	# if PGM_WITH_NX(PGM_GST_TYPE, PGM_SHW_TYPE)
2635	const bool fNoExecute = PdeSrc.n.u1NoExecute && CPUMIsGuestNXEnabled(pVCpu);
2636	# else
2637	const bool fNoExecute = false;
2638	# endif
2639
2640	GCPhys = GST_GET_PDE_BIG_PG_GCPHYS(PdeSrc);
2641	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
2642	/* Select the right PDE as we're emulating a 4MB page directory with two 2 MB shadow PDEs.*/
2643	GCPhys \|= GCPtrPage & (1 << X86_PD_PAE_SHIFT);
2644	# endif
2645	/* Determine the right kind of large page to avoid incorrect cached entry reuse. */
2646	if (PdeSrc.n.u1User)
2647	{
2648	if (PdeSrc.n.u1Write)
2649	enmAccess = (fNoExecute) ? PGMPOOLACCESS_USER_RW_NX : PGMPOOLACCESS_USER_RW;
2650	else
2651	enmAccess = (fNoExecute) ? PGMPOOLACCESS_USER_R_NX : PGMPOOLACCESS_USER_R;
2652	}
2653	else
2654	{
2655	if (PdeSrc.n.u1Write)
2656	enmAccess = (fNoExecute) ? PGMPOOLACCESS_SUPERVISOR_RW_NX : PGMPOOLACCESS_SUPERVISOR_RW;
2657	else
2658	enmAccess = (fNoExecute) ? PGMPOOLACCESS_SUPERVISOR_R_NX : PGMPOOLACCESS_SUPERVISOR_R;
2659	}
2660	rc = pgmPoolAllocEx(pVM, GCPhys, BTH_PGMPOOLKIND_PT_FOR_BIG, enmAccess, pShwPde->idx, iPDDst, &pShwPage);
2661	}
2662	if (rc == VINF_SUCCESS)
2663	pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR(pVM, pShwPage);
2664	else if (rc == VINF_PGM_CACHED_PAGE)
2665	{
2666	/*
2667	* The PT was cached, just hook it up.
2668	*/
2669	if (fPageTable)
2670	PdeDst.u = pShwPage->Core.Key
2671	\| (PdeSrc.u & ~(GST_PDE_PG_MASK \| X86_PDE_AVL_MASK \| X86_PDE_PCD \| X86_PDE_PWT \| X86_PDE_PS \| X86_PDE4M_G \| X86_PDE4M_D));
2672	else
2673	{
2674	PdeDst.u = pShwPage->Core.Key
2675	\| (PdeSrc.u & ~(GST_PDE_PG_MASK \| X86_PDE_AVL_MASK \| X86_PDE_PCD \| X86_PDE_PWT \| X86_PDE_PS \| X86_PDE4M_G \| X86_PDE4M_D));
2676	/* (see explanation and assumptions further down.) */
2677	if ( !PdeSrc.b.u1Dirty
2678	&& PdeSrc.b.u1Write)
2679	{
2680	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyPageBig));
2681	PdeDst.u \|= PGM_PDFLAGS_TRACK_DIRTY;
2682	PdeDst.b.u1Write = 0;
2683	}
2684	}
2685	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
2686	# if defined(IN_RC)
2687	PGMDynUnlockHCPage(pVM, (uint8_t *)pPdeDst);
2688	# endif
2689	return VINF_SUCCESS;
2690	}
2691	else if (rc == VERR_PGM_POOL_FLUSHED)
2692	{
2693	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
2694	# if defined(IN_RC)
2695	PGMDynUnlockHCPage(pVM, (uint8_t *)pPdeDst);
2696	# endif
2697	return VINF_PGM_SYNC_CR3;
2698	}
2699	else
2700	AssertMsgFailedReturn(("rc=%Rrc\n", rc), VERR_INTERNAL_ERROR);
2701	PdeDst.u &= X86_PDE_AVL_MASK;
2702	PdeDst.u \|= pShwPage->Core.Key;
2703
2704	/*
2705	* Page directory has been accessed (this is a fault situation, remember).
2706	*/
2707	pPDSrc->a[iPDSrc].n.u1Accessed = 1;
2708	if (fPageTable)
2709	{
2710	/*
2711	* Page table - 4KB.
2712	*
2713	* Sync all or just a few entries depending on PGM_SYNC_N_PAGES.
2714	*/
2715	Log2(("SyncPT: 4K %RGv PdeSrc:{P=%d RW=%d U=%d raw=%08llx}\n",
2716	GCPtrPage, PdeSrc.b.u1Present, PdeSrc.b.u1Write, PdeSrc.b.u1User, (uint64_t)PdeSrc.u));
2717	PGSTPT pPTSrc;
2718	rc = PGM_GCPHYS_2_PTR(pVM, PdeSrc.u & GST_PDE_PG_MASK, &pPTSrc);
2719	if (RT_SUCCESS(rc))
2720	{
2721	/*
2722	* Start by syncing the page directory entry so CSAM's TLB trick works.
2723	*/
2724	PdeDst.u = (PdeDst.u & (SHW_PDE_PG_MASK \| X86_PDE_AVL_MASK))
2725	\| (PdeSrc.u & ~(GST_PDE_PG_MASK \| X86_PDE_AVL_MASK \| X86_PDE_PCD \| X86_PDE_PWT \| X86_PDE_PS \| X86_PDE4M_G \| X86_PDE4M_D));
2726	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
2727	# if defined(IN_RC)
2728	PGMDynUnlockHCPage(pVM, (uint8_t *)pPdeDst);
2729	# endif
2730
2731	/*
2732	* Directory/page user or supervisor privilege: (same goes for read/write)
2733	*
2734	* Directory Page Combined
2735	* U/S U/S U/S
2736	* 0 0 0
2737	* 0 1 0
2738	* 1 0 0
2739	* 1 1 1
2740	*
2741	* Simple AND operation. Table listed for completeness.
2742	*
2743	*/
2744	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPT4K));
2745	# ifdef PGM_SYNC_N_PAGES
2746	unsigned iPTBase = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2747	unsigned iPTDst = iPTBase;
2748	const unsigned iPTDstEnd = RT_MIN(iPTDst + PGM_SYNC_NR_PAGES / 2, RT_ELEMENTS(pPTDst->a));
2749	if (iPTDst <= PGM_SYNC_NR_PAGES / 2)
2750	iPTDst = 0;
2751	else
2752	iPTDst -= PGM_SYNC_NR_PAGES / 2;
2753	# else /* !PGM_SYNC_N_PAGES */
2754	unsigned iPTDst = 0;
2755	const unsigned iPTDstEnd = RT_ELEMENTS(pPTDst->a);
2756	# endif /* !PGM_SYNC_N_PAGES */
2757	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
2758	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
2759	const unsigned offPTSrc = ((GCPtrPage >> SHW_PD_SHIFT) & 1) * 512;
2760	# else
2761	const unsigned offPTSrc = 0;
2762	# endif
2763	for (; iPTDst < iPTDstEnd; iPTDst++)
2764	{
2765	const unsigned iPTSrc = iPTDst + offPTSrc;
2766	const GSTPTE PteSrc = pPTSrc->a[iPTSrc];
2767
2768	if (PteSrc.n.u1Present) /* we've already cleared it above */
2769	{
2770	# ifndef IN_RING0
2771	/*
2772	* Assuming kernel code will be marked as supervisor - and not as user level
2773	* and executed using a conforming code selector - And marked as readonly.
2774	* Also assume that if we're monitoring a page, it's of no interest to CSAM.
2775	*/
2776	PPGMPAGE pPage;
2777	if ( ((PdeSrc.u & pPTSrc->a[iPTSrc].u) & (X86_PTE_RW \| X86_PTE_US))
2778	\|\| !CSAMDoesPageNeedScanning(pVM, (iPDSrc << GST_PD_SHIFT) \| (iPTSrc << PAGE_SHIFT))
2779	\|\| ( (pPage = pgmPhysGetPage(&pVM->pgm.s, PteSrc.u & GST_PTE_PG_MASK))
2780	&& PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
2781	)
2782	# endif
2783	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, PteSrc, pShwPage, iPTDst);
2784	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",
2785	(RTGCPTR)(((RTGCPTR)iPDSrc << GST_PD_SHIFT) \| ((RTGCPTR)iPTSrc << PAGE_SHIFT)),
2786	PteSrc.n.u1Present,
2787	PteSrc.n.u1Write & PdeSrc.n.u1Write,
2788	PteSrc.n.u1User & PdeSrc.n.u1User,
2789	(uint64_t)PteSrc.u,
2790	pPTDst->a[iPTDst].u & PGM_PTFLAGS_TRACK_DIRTY ? " Track-Dirty" : "", pPTDst->a[iPTDst].u, iPTSrc, PdeSrc.au32[0],
2791	(RTGCPHYS)((PdeSrc.u & GST_PDE_PG_MASK) + iPTSrc*sizeof(PteSrc)) ));
2792	}
2793	} /* for PTEs */
2794	}
2795	}
2796	else
2797	{
2798	/*
2799	* Big page - 2/4MB.
2800	*
2801	* We'll walk the ram range list in parallel and optimize lookups.
2802	* We will only sync on shadow page table at a time.
2803	*/
2804	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPT4M));
2805
2806	/**
2807	* @todo It might be more efficient to sync only a part of the 4MB page (similar to what we do for 4kb PDs).
2808	*/
2809
2810	/*
2811	* Start by syncing the page directory entry.
2812	*/
2813	PdeDst.u = (PdeDst.u & (SHW_PDE_PG_MASK \| (X86_PDE_AVL_MASK & ~PGM_PDFLAGS_TRACK_DIRTY)))
2814	\| (PdeSrc.u & ~(GST_PDE_PG_MASK \| X86_PDE_AVL_MASK \| X86_PDE_PCD \| X86_PDE_PWT \| X86_PDE_PS \| X86_PDE4M_G \| X86_PDE4M_D));
2815
2816	/*
2817	* If the page is not flagged as dirty and is writable, then make it read-only
2818	* at PD level, so we can set the dirty bit when the page is modified.
2819	*
2820	* ASSUMES that page access handlers are implemented on page table entry level.
2821	* Thus we will first catch the dirty access and set PDE.D and restart. If
2822	* there is an access handler, we'll trap again and let it work on the problem.
2823	*/
2824	/** @todo move the above stuff to a section in the PGM documentation. */
2825	Assert(!(PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY));
2826	if ( !PdeSrc.b.u1Dirty
2827	&& PdeSrc.b.u1Write)
2828	{
2829	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,DirtyPageBig));
2830	PdeDst.u \|= PGM_PDFLAGS_TRACK_DIRTY;
2831	PdeDst.b.u1Write = 0;
2832	}
2833	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
2834	# if defined(IN_RC)
2835	PGMDynUnlockHCPage(pVM, (uint8_t *)pPdeDst);
2836	# endif
2837
2838	/*
2839	* Fill the shadow page table.
2840	*/
2841	/* Get address and flags from the source PDE. */
2842	SHWPTE PteDstBase;
2843	PteDstBase.u = PdeSrc.u & ~(GST_PDE_PG_MASK \| X86_PTE_AVL_MASK \| X86_PTE_PAT \| X86_PTE_PCD \| X86_PTE_PWT);
2844
2845	/* Loop thru the entries in the shadow PT. */
2846	const RTGCPTR GCPtr = (GCPtrPage >> SHW_PD_SHIFT) << SHW_PD_SHIFT; NOREF(GCPtr);
2847	Log2(("SyncPT: BIG %RGv PdeSrc:{P=%d RW=%d U=%d raw=%08llx} Shw=%RGv GCPhys=%RGp %s\n",
2848	GCPtrPage, PdeSrc.b.u1Present, PdeSrc.b.u1Write, PdeSrc.b.u1User, (uint64_t)PdeSrc.u, GCPtr,
2849	GCPhys, PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY ? " Track-Dirty" : ""));
2850	PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRanges);
2851	unsigned iPTDst = 0;
2852	while ( iPTDst < RT_ELEMENTS(pPTDst->a)
2853	&& !VM_FF_ISPENDING(pVM, VM_FF_PGM_NO_MEMORY))
2854	{
2855	/* Advance ram range list. */
2856	while (pRam && GCPhys > pRam->GCPhysLast)
2857	pRam = pRam->CTX_SUFF(pNext);
2858	if (pRam && GCPhys >= pRam->GCPhys)
2859	{
2860	unsigned iHCPage = (GCPhys - pRam->GCPhys) >> PAGE_SHIFT;
2861	do
2862	{
2863	/* Make shadow PTE. */
2864	PPGMPAGE pPage = &pRam->aPages[iHCPage];
2865	SHWPTE PteDst;
2866
2867	# ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
2868	/* Try to make the page writable if necessary. */
2869	if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM
2870	&& ( PGM_PAGE_IS_ZERO(pPage)
2871	\|\| ( PteDstBase.n.u1Write
2872	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED
2873	# ifdef VBOX_WITH_REAL_WRITE_MONITORED_PAGES
2874	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_WRITE_MONITORED
2875	# endif
2876	# ifdef VBOX_WITH_PAGE_SHARING
2877	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_SHARED
2878	# endif
2879	&& !PGM_PAGE_IS_BALLOONED(pPage))
2880	)
2881	)
2882	{
2883	rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhys);
2884	AssertRCReturn(rc, rc);
2885	if (VM_FF_ISPENDING(pVM, VM_FF_PGM_NO_MEMORY))
2886	break;
2887	}
2888	# endif
2889
2890	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
2891	{
2892	if (!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage))
2893	{
2894	PteDst.u = PGM_PAGE_GET_HCPHYS(pPage) \| PteDstBase.u;
2895	PteDst.n.u1Write = 0;
2896	}
2897	else
2898	PteDst.u = 0;
2899	}
2900	else
2901	if (PGM_PAGE_IS_BALLOONED(pPage))
2902	{
2903	/* Skip ballooned pages. */
2904	PteDst.u = 0;
2905	}
2906	# ifndef IN_RING0
2907	/*
2908	* Assuming kernel code will be marked as supervisor and not as user level and executed
2909	* using a conforming code selector. Don't check for readonly, as that implies the whole
2910	* 4MB can be code or readonly data. Linux enables write access for its large pages.
2911	*/
2912	else if ( !PdeSrc.n.u1User
2913	&& CSAMDoesPageNeedScanning(pVM, GCPtr \| (iPTDst << SHW_PT_SHIFT)))
2914	PteDst.u = 0;
2915	# endif
2916	else
2917	PteDst.u = PGM_PAGE_GET_HCPHYS(pPage) \| PteDstBase.u;
2918
2919	/* Only map writable pages writable. */
2920	if ( PteDst.n.u1Write
2921	&& PteDst.n.u1Present
2922	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
2923	{
2924	/* Still applies to shared pages. */
2925	Assert(!PGM_PAGE_IS_ZERO(pPage));
2926	PteDst.n.u1Write = 0; /** @todo this isn't quite working yet... */
2927	Log3(("SyncPT: write-protecting %RGp pPage=%R[pgmpage] at %RGv\n", GCPhys, pPage, (RTGCPTR)(GCPtr \| (iPTDst << SHW_PT_SHIFT))));
2928	}
2929
2930	if (PteDst.n.u1Present)
2931	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPTDst);
2932
2933	/* commit it */
2934	pPTDst->a[iPTDst] = PteDst;
2935	Log4(("SyncPT: BIG %RGv PteDst:{P=%d RW=%d U=%d raw=%08llx}%s\n",
2936	(RTGCPTR)(GCPtr \| (iPTDst << SHW_PT_SHIFT)), PteDst.n.u1Present, PteDst.n.u1Write, PteDst.n.u1User, (uint64_t)PteDst.u,
2937	PteDst.u & PGM_PTFLAGS_TRACK_DIRTY ? " Track-Dirty" : ""));
2938
2939	/* advance */
2940	GCPhys += PAGE_SIZE;
2941	iHCPage++;
2942	iPTDst++;
2943	} while ( iPTDst < RT_ELEMENTS(pPTDst->a)
2944	&& GCPhys <= pRam->GCPhysLast);
2945	}
2946	else if (pRam)
2947	{
2948	Log(("Invalid pages at %RGp\n", GCPhys));
2949	do
2950	{
2951	pPTDst->a[iPTDst].u = 0; /* MMIO or invalid page, we must handle them manually. */
2952	GCPhys += PAGE_SIZE;
2953	iPTDst++;
2954	} while ( iPTDst < RT_ELEMENTS(pPTDst->a)
2955	&& GCPhys < pRam->GCPhys);
2956	}
2957	else
2958	{
2959	Log(("Invalid pages at %RGp (2)\n", GCPhys));
2960	for ( ; iPTDst < RT_ELEMENTS(pPTDst->a); iPTDst++)
2961	pPTDst->a[iPTDst].u = 0; /* MMIO or invalid page, we must handle them manually. */
2962	}
2963	} /* while more PTEs */
2964	} /* 4KB / 4MB */
2965	}
2966	else
2967	AssertRelease(!PdeDst.n.u1Present);
2968
2969	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPT), a);
2970	if (RT_FAILURE(rc))
2971	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPTFailed));
2972	return rc;
2973
2974	#elif (PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
2975	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
2976	&& (PGM_SHW_TYPE != PGM_TYPE_EPT \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
2977	&& !defined(IN_RC)
2978
2979	/*
2980	* Validate input a little bit.
2981	*/
2982	int rc = VINF_SUCCESS;
2983	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
2984	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
2985	PSHWPDE pPdeDst = pgmShwGet32BitPDEPtr(&pVCpu->pgm.s, GCPtrPage);
2986
2987	/* Fetch the pgm pool shadow descriptor. */
2988	PPGMPOOLPAGE pShwPde = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
2989	Assert(pShwPde);
2990
2991	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
2992	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
2993	PPGMPOOLPAGE pShwPde = NULL; /* initialized to shut up gcc */
2994	PX86PDPAE pPDDst;
2995	PSHWPDE pPdeDst;
2996
2997	/* Fetch the pgm pool shadow descriptor. */
2998	rc = pgmShwGetPaePoolPagePD(&pVCpu->pgm.s, GCPtrPage, &pShwPde);
2999	AssertRCSuccessReturn(rc, rc);
3000	Assert(pShwPde);
3001
3002	pPDDst = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_BY_PGM(&pVM->pgm.s, pShwPde);
3003	pPdeDst = &pPDDst->a[iPDDst];
3004
3005	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
3006	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
3007	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
3008	PX86PDPAE pPDDst = NULL; /* initialized to shut up gcc */
3009	PX86PDPT pPdptDst= NULL; /* initialized to shut up gcc */
3010	rc = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, NULL, &pPdptDst, &pPDDst);
3011	AssertRCSuccessReturn(rc, rc);
3012	Assert(pPDDst);
3013	PSHWPDE pPdeDst = &pPDDst->a[iPDDst];
3014
3015	/* Fetch the pgm pool shadow descriptor. */
3016	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & X86_PDPE_PG_MASK);
3017	Assert(pShwPde);
3018
3019	# elif PGM_SHW_TYPE == PGM_TYPE_EPT
3020	const unsigned iPdpt = (GCPtrPage >> EPT_PDPT_SHIFT) & EPT_PDPT_MASK;
3021	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
3022	PEPTPD pPDDst;
3023	PEPTPDPT pPdptDst;
3024
3025	rc = pgmShwGetEPTPDPtr(pVCpu, GCPtrPage, &pPdptDst, &pPDDst);
3026	if (rc != VINF_SUCCESS)
3027	{
3028	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPT), a);
3029	AssertRC(rc);
3030	return rc;
3031	}
3032	Assert(pPDDst);
3033	PSHWPDE pPdeDst = &pPDDst->a[iPDDst];
3034
3035	/* Fetch the pgm pool shadow descriptor. */
3036	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & EPT_PDPTE_PG_MASK);
3037	Assert(pShwPde);
3038	# endif
3039	SHWPDE PdeDst = *pPdeDst;
3040
3041	Assert(!(PdeDst.u & PGM_PDFLAGS_MAPPING));
3042	Assert(!PdeDst.n.u1Present); /* We're only supposed to call SyncPT on PDE!P and conflicts.*/
3043
3044	# if defined(PGM_WITH_LARGE_PAGES) && (PGM_SHW_TYPE != PGM_TYPE_32BIT && PGM_SHW_TYPE != PGM_TYPE_PAE)
3045	# if (PGM_SHW_TYPE != PGM_TYPE_EPT) /* PGM_TYPE_EPT implies nested paging */
3046	if (HWACCMIsNestedPagingActive(pVM))
3047	# endif
3048	{
3049	PPGMPAGE pPage;
3050
3051	/* Check if we allocated a big page before for this 2 MB range. */
3052	rc = pgmPhysGetPageEx(&pVM->pgm.s, GCPtrPage & X86_PDE2M_PAE_PG_MASK, &pPage);
3053	if (RT_SUCCESS(rc))
3054	{
3055	RTHCPHYS HCPhys = NIL_RTHCPHYS;
3056
3057	if (PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE)
3058	{
3059	STAM_REL_COUNTER_INC(&pVM->pgm.s.StatLargePageReused);
3060	AssertRelease(PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED);
3061	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
3062	}
3063	else
3064	if (PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE_DISABLED)
3065	{
3066	/* Recheck the entire 2 MB range to see if we can use it again as a large page. */
3067	rc = pgmPhysIsValidLargePage(pVM, GCPtrPage, pPage);
3068	if (RT_SUCCESS(rc))
3069	{
3070	Assert(PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED);
3071	Assert(PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE);
3072	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
3073	}
3074	}
3075	else
3076	if (PGMIsUsingLargePages(pVM))
3077	{
3078	rc = pgmPhysAllocLargePage(pVM, GCPtrPage);
3079	if (RT_SUCCESS(rc))
3080	{
3081	Assert(PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED);
3082	Assert(PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE);
3083	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
3084	}
3085	else
3086	LogFlow(("pgmPhysAllocLargePage failed with %Rrc\n", rc));
3087	}
3088
3089	if (HCPhys != NIL_RTHCPHYS)
3090	{
3091	PdeDst.u &= X86_PDE_AVL_MASK;
3092	PdeDst.u \|= HCPhys;
3093	PdeDst.n.u1Present = 1;
3094	PdeDst.n.u1Write = 1;
3095	PdeDst.b.u1Size = 1;
3096	# if PGM_SHW_TYPE == PGM_TYPE_EPT
3097	PdeDst.n.u1Execute = 1;
3098	PdeDst.b.u1IgnorePAT = 1;
3099	PdeDst.b.u3EMT = VMX_EPT_MEMTYPE_WB;
3100	# else
3101	PdeDst.n.u1User = 1;
3102	# endif
3103	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
3104
3105	Log(("SyncPT: Use large page at %RGp PDE=%RX64\n", GCPtrPage, PdeDst.u));
3106	/* Add a reference to the first page only. */
3107	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPde, PGM_PAGE_GET_TRACKING(pPage), pPage, iPDDst);
3108
3109	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPT), a);
3110	return VINF_SUCCESS;
3111	}
3112	}
3113	}
3114	# endif /* HC_ARCH_BITS == 64 */
3115
3116	GSTPDE PdeSrc;
3117	PdeSrc.u = 0; /* faked so we don't have to #ifdef everything */
3118	PdeSrc.n.u1Present = 1;
3119	PdeSrc.n.u1Write = 1;
3120	PdeSrc.n.u1Accessed = 1;
3121	PdeSrc.n.u1User = 1;
3122
3123	/*
3124	* Allocate & map the page table.
3125	*/
3126	PSHWPT pPTDst;
3127	PPGMPOOLPAGE pShwPage;
3128	RTGCPHYS GCPhys;
3129
3130	/* Virtual address = physical address */
3131	GCPhys = GCPtrPage & X86_PAGE_4K_BASE_MASK;
3132	rc = pgmPoolAlloc(pVM, GCPhys & ~(RT_BIT_64(SHW_PD_SHIFT) - 1), BTH_PGMPOOLKIND_PT_FOR_PT, pShwPde->idx, iPDDst, &pShwPage);
3133
3134	if ( rc == VINF_SUCCESS
3135	\|\| rc == VINF_PGM_CACHED_PAGE)
3136	pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR(pVM, pShwPage);
3137	else
3138	AssertMsgFailedReturn(("rc=%Rrc\n", rc), VERR_INTERNAL_ERROR);
3139
3140	PdeDst.u &= X86_PDE_AVL_MASK;
3141	PdeDst.u \|= pShwPage->Core.Key;
3142	PdeDst.n.u1Present = 1;
3143	PdeDst.n.u1Write = 1;
3144	# if PGM_SHW_TYPE == PGM_TYPE_EPT
3145	PdeDst.n.u1Execute = 1;
3146	# else
3147	PdeDst.n.u1User = 1;
3148	PdeDst.n.u1Accessed = 1;
3149	# endif
3150	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
3151
3152	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, GCPtrPage, PGM_SYNC_NR_PAGES, 0 /* page not present */);
3153	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPT), a);
3154	return rc;
3155
3156	#else
3157	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_SHW_TYPE, PGM_GST_TYPE));
3158	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPT), a);
3159	return VERR_INTERNAL_ERROR;
3160	#endif
3161	}
3162
3163
3164
3165	/**
3166	* Prefetch a page/set of pages.
3167	*
3168	* Typically used to sync commonly used pages before entering raw mode
3169	* after a CR3 reload.
3170	*
3171	* @returns VBox status code.
3172	* @param pVCpu The VMCPU handle.
3173	* @param GCPtrPage Page to invalidate.
3174	*/
3175	PGM_BTH_DECL(int, PrefetchPage)(PVMCPU pVCpu, RTGCPTR GCPtrPage)
3176	{
3177	#if (PGM_GST_TYPE == PGM_TYPE_32BIT \|\| PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT \|\| PGM_GST_TYPE == PGM_TYPE_PAE \|\| PGM_GST_TYPE == PGM_TYPE_AMD64) \
3178	&& PGM_SHW_TYPE != PGM_TYPE_NESTED && PGM_SHW_TYPE != PGM_TYPE_EPT
3179	/*
3180	* Check that all Guest levels thru the PDE are present, getting the
3181	* PD and PDE in the processes.
3182	*/
3183	int rc = VINF_SUCCESS;
3184	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
3185	# if PGM_GST_TYPE == PGM_TYPE_32BIT
3186	const unsigned iPDSrc = GCPtrPage >> GST_PD_SHIFT;
3187	PGSTPD pPDSrc = pgmGstGet32bitPDPtr(&pVCpu->pgm.s);
3188	# elif PGM_GST_TYPE == PGM_TYPE_PAE
3189	unsigned iPDSrc;
3190	X86PDPE PdpeSrc;
3191	PGSTPD pPDSrc = pgmGstGetPaePDPtr(&pVCpu->pgm.s, GCPtrPage, &iPDSrc, &PdpeSrc);
3192	if (!pPDSrc)
3193	return VINF_SUCCESS; /* not present */
3194	# elif PGM_GST_TYPE == PGM_TYPE_AMD64
3195	unsigned iPDSrc;
3196	PX86PML4E pPml4eSrc;
3197	X86PDPE PdpeSrc;
3198	PGSTPD pPDSrc = pgmGstGetLongModePDPtr(&pVCpu->pgm.s, GCPtrPage, &pPml4eSrc, &PdpeSrc, &iPDSrc);
3199	if (!pPDSrc)
3200	return VINF_SUCCESS; /* not present */
3201	# endif
3202	const GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
3203	# else
3204	PGSTPD pPDSrc = NULL;
3205	const unsigned iPDSrc = 0;
3206	GSTPDE PdeSrc;
3207
3208	PdeSrc.u = 0; /* faked so we don't have to #ifdef everything */
3209	PdeSrc.n.u1Present = 1;
3210	PdeSrc.n.u1Write = 1;
3211	PdeSrc.n.u1Accessed = 1;
3212	PdeSrc.n.u1User = 1;
3213	# endif
3214
3215	if (PdeSrc.n.u1Present && PdeSrc.n.u1Accessed)
3216	{
3217	PVM pVM = pVCpu->CTX_SUFF(pVM);
3218	pgmLock(pVM);
3219
3220	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
3221	const X86PDE PdeDst = pgmShwGet32BitPDE(&pVCpu->pgm.s, GCPtrPage);
3222	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
3223	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
3224	PX86PDPAE pPDDst;
3225	X86PDEPAE PdeDst;
3226	# if PGM_GST_TYPE != PGM_TYPE_PAE
3227	X86PDPE PdpeSrc;
3228
3229	/* Fake PDPT entry; access control handled on the page table level, so allow everything. */
3230	PdpeSrc.u = X86_PDPE_P; /* rw/us are reserved for PAE pdpte's; accessed bit causes invalid VT-x guest state errors */
3231	# endif
3232	rc = pgmShwSyncPaePDPtr(pVCpu, GCPtrPage, &PdpeSrc, &pPDDst);
3233	if (rc != VINF_SUCCESS)
3234	{
3235	pgmUnlock(pVM);
3236	AssertRC(rc);
3237	return rc;
3238	}
3239	Assert(pPDDst);
3240	PdeDst = pPDDst->a[iPDDst];
3241
3242	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
3243	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
3244	PX86PDPAE pPDDst;
3245	X86PDEPAE PdeDst;
3246
3247	# if PGM_GST_TYPE == PGM_TYPE_PROT
3248	/* AMD-V nested paging */
3249	X86PML4E Pml4eSrc;
3250	X86PDPE PdpeSrc;
3251	PX86PML4E pPml4eSrc = &Pml4eSrc;
3252
3253	/* Fake PML4 & PDPT entry; access control handled on the page table level, so allow everything. */
3254	Pml4eSrc.u = X86_PML4E_P \| X86_PML4E_RW \| X86_PML4E_US \| X86_PML4E_A;
3255	PdpeSrc.u = X86_PDPE_P \| X86_PDPE_RW \| X86_PDPE_US \| X86_PDPE_A;
3256	# endif
3257
3258	rc = pgmShwSyncLongModePDPtr(pVCpu, GCPtrPage, pPml4eSrc, &PdpeSrc, &pPDDst);
3259	if (rc != VINF_SUCCESS)
3260	{
3261	pgmUnlock(pVM);
3262	AssertRC(rc);
3263	return rc;
3264	}
3265	Assert(pPDDst);
3266	PdeDst = pPDDst->a[iPDDst];
3267	# endif
3268	if (!(PdeDst.u & PGM_PDFLAGS_MAPPING))
3269	{
3270	if (!PdeDst.n.u1Present)
3271	{
3272	/** r=bird: This guy will set the A bit on the PDE, probably harmless. */
3273	rc = PGM_BTH_NAME(SyncPT)(pVCpu, iPDSrc, pPDSrc, GCPtrPage);
3274	}
3275	else
3276	{
3277	/** @note We used to sync PGM_SYNC_NR_PAGES pages, which triggered assertions in CSAM, because
3278	* R/W attributes of nearby pages were reset. Not sure how that could happen. Anyway, it
3279	* makes no sense to prefetch more than one page.
3280	*/
3281	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, GCPtrPage, 1, 0);
3282	if (RT_SUCCESS(rc))
3283	rc = VINF_SUCCESS;
3284	}
3285	}
3286	pgmUnlock(pVM);
3287	}
3288	return rc;
3289
3290	#elif PGM_SHW_TYPE == PGM_TYPE_NESTED \|\| PGM_SHW_TYPE == PGM_TYPE_EPT
3291	return VINF_SUCCESS; /* ignore */
3292	#endif
3293	}
3294
3295
3296
3297
3298	/**
3299	* Syncs a page during a PGMVerifyAccess() call.
3300	*
3301	* @returns VBox status code (informational included).
3302	* @param pVCpu The VMCPU handle.
3303	* @param GCPtrPage The address of the page to sync.
3304	* @param fPage The effective guest page flags.
3305	* @param uErr The trap error code.
3306	*/
3307	PGM_BTH_DECL(int, VerifyAccessSyncPage)(PVMCPU pVCpu, RTGCPTR GCPtrPage, unsigned fPage, unsigned uErr)
3308	{
3309	PVM pVM = pVCpu->CTX_SUFF(pVM);
3310
3311	LogFlow(("VerifyAccessSyncPage: GCPtrPage=%RGv fPage=%#x uErr=%#x\n", GCPtrPage, fPage, uErr));
3312
3313	Assert(!HWACCMIsNestedPagingActive(pVM));
3314	#if (PGM_GST_TYPE == PGM_TYPE_32BIT \|\| PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT \|\| PGM_GST_TYPE == PGM_TYPE_PAE \|\| PGM_TYPE_AMD64) \
3315	&& PGM_SHW_TYPE != PGM_TYPE_NESTED && PGM_SHW_TYPE != PGM_TYPE_EPT
3316
3317	# ifndef IN_RING0
3318	if (!(fPage & X86_PTE_US))
3319	{
3320	/*
3321	* Mark this page as safe.
3322	*/
3323	/** @todo not correct for pages that contain both code and data!! */
3324	Log(("CSAMMarkPage %RGv; scanned=%d\n", GCPtrPage, true));
3325	CSAMMarkPage(pVM, GCPtrPage, true);
3326	}
3327	# endif
3328
3329	/*
3330	* Get guest PD and index.
3331	*/
3332	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
3333	# if PGM_GST_TYPE == PGM_TYPE_32BIT
3334	const unsigned iPDSrc = GCPtrPage >> GST_PD_SHIFT;
3335	PGSTPD pPDSrc = pgmGstGet32bitPDPtr(&pVCpu->pgm.s);
3336	# elif PGM_GST_TYPE == PGM_TYPE_PAE
3337	unsigned iPDSrc = 0;
3338	X86PDPE PdpeSrc;
3339	PGSTPD pPDSrc = pgmGstGetPaePDPtr(&pVCpu->pgm.s, GCPtrPage, &iPDSrc, &PdpeSrc);
3340
3341	if (pPDSrc)
3342	{
3343	Log(("PGMVerifyAccess: access violation for %RGv due to non-present PDPTR\n", GCPtrPage));
3344	return VINF_EM_RAW_GUEST_TRAP;
3345	}
3346	# elif PGM_GST_TYPE == PGM_TYPE_AMD64
3347	unsigned iPDSrc;
3348	PX86PML4E pPml4eSrc;
3349	X86PDPE PdpeSrc;
3350	PGSTPD pPDSrc = pgmGstGetLongModePDPtr(&pVCpu->pgm.s, GCPtrPage, &pPml4eSrc, &PdpeSrc, &iPDSrc);
3351	if (!pPDSrc)
3352	{
3353	Log(("PGMVerifyAccess: access violation for %RGv due to non-present PDPTR\n", GCPtrPage));
3354	return VINF_EM_RAW_GUEST_TRAP;
3355	}
3356	# endif
3357	# else
3358	PGSTPD pPDSrc = NULL;
3359	const unsigned iPDSrc = 0;
3360	# endif
3361	int rc = VINF_SUCCESS;
3362
3363	pgmLock(pVM);
3364
3365	/*
3366	* First check if the shadow pd is present.
3367	*/
3368	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
3369	PX86PDE pPdeDst = pgmShwGet32BitPDEPtr(&pVCpu->pgm.s, GCPtrPage);
3370	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
3371	PX86PDEPAE pPdeDst;
3372	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
3373	PX86PDPAE pPDDst;
3374	# if PGM_GST_TYPE != PGM_TYPE_PAE
3375	X86PDPE PdpeSrc;
3376
3377	/* Fake PDPT entry; access control handled on the page table level, so allow everything. */
3378	PdpeSrc.u = X86_PDPE_P; /* rw/us are reserved for PAE pdpte's; accessed bit causes invalid VT-x guest state errors */
3379	# endif
3380	rc = pgmShwSyncPaePDPtr(pVCpu, GCPtrPage, &PdpeSrc, &pPDDst);
3381	if (rc != VINF_SUCCESS)
3382	{
3383	pgmUnlock(pVM);
3384	AssertRC(rc);
3385	return rc;
3386	}
3387	Assert(pPDDst);
3388	pPdeDst = &pPDDst->a[iPDDst];
3389
3390	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
3391	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
3392	PX86PDPAE pPDDst;
3393	PX86PDEPAE pPdeDst;
3394
3395	# if PGM_GST_TYPE == PGM_TYPE_PROT
3396	/* AMD-V nested paging */
3397	X86PML4E Pml4eSrc;
3398	X86PDPE PdpeSrc;
3399	PX86PML4E pPml4eSrc = &Pml4eSrc;
3400
3401	/* Fake PML4 & PDPT entry; access control handled on the page table level, so allow everything. */
3402	Pml4eSrc.u = X86_PML4E_P \| X86_PML4E_RW \| X86_PML4E_US \| X86_PML4E_A;
3403	PdpeSrc.u = X86_PDPE_P \| X86_PDPE_RW \| X86_PDPE_US \| X86_PDPE_A;
3404	# endif
3405
3406	rc = pgmShwSyncLongModePDPtr(pVCpu, GCPtrPage, pPml4eSrc, &PdpeSrc, &pPDDst);
3407	if (rc != VINF_SUCCESS)
3408	{
3409	pgmUnlock(pVM);
3410	AssertRC(rc);
3411	return rc;
3412	}
3413	Assert(pPDDst);
3414	pPdeDst = &pPDDst->a[iPDDst];
3415	# endif
3416
3417	# if defined(IN_RC)
3418	/* Make sure the dynamic pPdeDst mapping will not be reused during this function. */
3419	PGMDynLockHCPage(pVM, (uint8_t *)pPdeDst);
3420	# endif
3421
3422	if (!pPdeDst->n.u1Present)
3423	{
3424	rc = PGM_BTH_NAME(SyncPT)(pVCpu, iPDSrc, pPDSrc, GCPtrPage);
3425	if (rc != VINF_SUCCESS)
3426	{
3427	# if defined(IN_RC)
3428	/* Make sure the dynamic pPdeDst mapping will not be reused during this function. */
3429	PGMDynUnlockHCPage(pVM, (uint8_t *)pPdeDst);
3430	# endif
3431	pgmUnlock(pVM);
3432	AssertRC(rc);
3433	return rc;
3434	}
3435	}
3436
3437	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
3438	/* Check for dirty bit fault */
3439	rc = PGM_BTH_NAME(CheckDirtyPageFault)(pVCpu, uErr, pPdeDst, &pPDSrc->a[iPDSrc], GCPtrPage);
3440	if (rc == VINF_PGM_HANDLED_DIRTY_BIT_FAULT)
3441	Log(("PGMVerifyAccess: success (dirty)\n"));
3442	else
3443	{
3444	GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
3445	# else
3446	{
3447	GSTPDE PdeSrc;
3448	PdeSrc.u = 0; /* faked so we don't have to #ifdef everything */
3449	PdeSrc.n.u1Present = 1;
3450	PdeSrc.n.u1Write = 1;
3451	PdeSrc.n.u1Accessed = 1;
3452	PdeSrc.n.u1User = 1;
3453
3454	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
3455	Assert(rc != VINF_EM_RAW_GUEST_TRAP);
3456	if (uErr & X86_TRAP_PF_US)
3457	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,PageOutOfSyncUser));
3458	else /* supervisor */
3459	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,PageOutOfSyncSupervisor));
3460
3461	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, GCPtrPage, 1, 0);
3462	if (RT_SUCCESS(rc))
3463	{
3464	/* Page was successfully synced */
3465	Log2(("PGMVerifyAccess: success (sync)\n"));
3466	rc = VINF_SUCCESS;
3467	}
3468	else
3469	{
3470	Log(("PGMVerifyAccess: access violation for %RGv rc=%d\n", GCPtrPage, rc));
3471	rc = VINF_EM_RAW_GUEST_TRAP;
3472	}
3473	}
3474	# if defined(IN_RC)
3475	/* Make sure the dynamic pPdeDst mapping will not be reused during this function. */
3476	PGMDynUnlockHCPage(pVM, (uint8_t *)pPdeDst);
3477	# endif
3478	pgmUnlock(pVM);
3479	return rc;
3480
3481	#else /* PGM_GST_TYPE != PGM_TYPE_32BIT */
3482
3483	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_GST_TYPE, PGM_SHW_TYPE));
3484	return VERR_INTERNAL_ERROR;
3485	#endif /* PGM_GST_TYPE != PGM_TYPE_32BIT */
3486	}
3487
3488
3489	/**
3490	* Syncs the paging hierarchy starting at CR3.
3491	*
3492	* @returns VBox status code, no specials.
3493	* @param pVCpu The VMCPU handle.
3494	* @param cr0 Guest context CR0 register
3495	* @param cr3 Guest context CR3 register
3496	* @param cr4 Guest context CR4 register
3497	* @param fGlobal Including global page directories or not
3498	*/
3499	PGM_BTH_DECL(int, SyncCR3)(PVMCPU pVCpu, uint64_t cr0, uint64_t cr3, uint64_t cr4, bool fGlobal)
3500	{
3501	PVM pVM = pVCpu->CTX_SUFF(pVM);
3502
3503	LogFlow(("SyncCR3 fGlobal=%d\n", !!VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3)));
3504
3505	#if PGM_SHW_TYPE != PGM_TYPE_NESTED && PGM_SHW_TYPE != PGM_TYPE_EPT
3506
3507	pgmLock(pVM);
3508
3509	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
3510	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3511	if (pPool->cDirtyPages)
3512	pgmPoolResetDirtyPages(pVM);
3513	# endif
3514
3515	/*
3516	* Update page access handlers.
3517	* The virtual are always flushed, while the physical are only on demand.
3518	* WARNING: We are incorrectly not doing global flushing on Virtual Handler updates. We'll
3519	* have to look into that later because it will have a bad influence on the performance.
3520	* @note SvL: There's no need for that. Just invalidate the virtual range(s).
3521	* bird: Yes, but that won't work for aliases.
3522	*/
3523	/** @todo this MUST go away. See #1557. */
3524	STAM_PROFILE_START(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncCR3Handlers), h);
3525	PGM_GST_NAME(HandlerVirtualUpdate)(pVM, cr4);
3526	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncCR3Handlers), h);
3527	pgmUnlock(pVM);
3528	#endif /* !NESTED && !EPT */
3529
3530	#if PGM_SHW_TYPE == PGM_TYPE_NESTED \|\| PGM_SHW_TYPE == PGM_TYPE_EPT
3531	/*
3532	* Nested / EPT - almost no work.
3533	*/
3534	Assert(!pgmMapAreMappingsEnabled(&pVM->pgm.s));
3535	return VINF_SUCCESS;
3536
3537	#elif PGM_SHW_TYPE == PGM_TYPE_AMD64
3538	/*
3539	* AMD64 (Shw & Gst) - No need to check all paging levels; we zero
3540	* out the shadow parts when the guest modifies its tables.
3541	*/
3542	Assert(!pgmMapAreMappingsEnabled(&pVM->pgm.s));
3543	return VINF_SUCCESS;
3544
3545	#else /* PGM_SHW_TYPE != PGM_TYPE_NESTED && PGM_SHW_TYPE != PGM_TYPE_EPT && PGM_SHW_TYPE != PGM_TYPE_AMD64 */
3546
3547	# ifndef PGM_WITHOUT_MAPPINGS
3548	/*
3549	* Check for and resolve conflicts with our guest mappings if they
3550	* are enabled and not fixed.
3551	*/
3552	if (pgmMapAreMappingsFloating(&pVM->pgm.s))
3553	{
3554	int rc = pgmMapResolveConflicts(pVM);
3555	Assert(rc == VINF_SUCCESS \|\| rc == VINF_PGM_SYNC_CR3);
3556	if (rc == VINF_PGM_SYNC_CR3)
3557	{
3558	LogFlow(("SyncCR3: detected conflict -> VINF_PGM_SYNC_CR3\n"));
3559	return VINF_PGM_SYNC_CR3;
3560	}
3561	}
3562	# else
3563	Assert(!pgmMapAreMappingsEnabled(&pVM->pgm.s));
3564	# endif
3565	return VINF_SUCCESS;
3566	#endif /* PGM_SHW_TYPE != PGM_TYPE_NESTED && PGM_SHW_TYPE != PGM_TYPE_EPT && PGM_SHW_TYPE != PGM_TYPE_AMD64 */
3567	}
3568
3569
3570
3571
3572	#ifdef VBOX_STRICT
3573	#ifdef IN_RC
3574	# undef AssertMsgFailed
3575	# define AssertMsgFailed Log
3576	#endif
3577	#ifdef IN_RING3
3578	# include <VBox/dbgf.h>
3579
3580	/**
3581	* Dumps a page table hierarchy use only physical addresses and cr4/lm flags.
3582	*
3583	* @returns VBox status code (VINF_SUCCESS).
3584	* @param cr3 The root of the hierarchy.
3585	* @param crr The cr4, only PAE and PSE is currently used.
3586	* @param fLongMode Set if long mode, false if not long mode.
3587	* @param cMaxDepth Number of levels to dump.
3588	* @param pHlp Pointer to the output functions.
3589	*/
3590	RT_C_DECLS_BEGIN
3591	VMMR3DECL(int) PGMR3DumpHierarchyHC(PVM pVM, uint32_t cr3, uint32_t cr4, bool fLongMode, unsigned cMaxDepth, PCDBGFINFOHLP pHlp);
3592	RT_C_DECLS_END
3593
3594	#endif
3595
3596	/**
3597	* Checks that the shadow page table is in sync with the guest one.
3598	*
3599	* @returns The number of errors.
3600	* @param pVM The virtual machine.
3601	* @param pVCpu The VMCPU handle.
3602	* @param cr3 Guest context CR3 register
3603	* @param cr4 Guest context CR4 register
3604	* @param GCPtr Where to start. Defaults to 0.
3605	* @param cb How much to check. Defaults to everything.
3606	*/
3607	PGM_BTH_DECL(unsigned, AssertCR3)(PVMCPU pVCpu, uint64_t cr3, uint64_t cr4, RTGCPTR GCPtr, RTGCPTR cb)
3608	{
3609	#if PGM_SHW_TYPE == PGM_TYPE_NESTED \|\| PGM_SHW_TYPE == PGM_TYPE_EPT
3610	return 0;
3611	#else
3612	unsigned cErrors = 0;
3613	PVM pVM = pVCpu->CTX_SUFF(pVM);
3614	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3615
3616	#if PGM_GST_TYPE == PGM_TYPE_PAE
3617	/** @todo currently broken; crashes below somewhere */
3618	AssertFailed();
3619	#endif
3620
3621	#if PGM_GST_TYPE == PGM_TYPE_32BIT \
3622	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
3623	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64
3624
3625	# if PGM_GST_TYPE == PGM_TYPE_32BIT
3626	bool fBigPagesSupported = CPUMIsGuestPageSizeExtEnabled(pVCpu);
3627	# else
3628	bool fBigPagesSupported = true;
3629	# endif
3630	PPGMCPU pPGM = &pVCpu->pgm.s;
3631	RTGCPHYS GCPhysGst; /* page address derived from the guest page tables. */
3632	RTHCPHYS HCPhysShw; /* page address derived from the shadow page tables. */
3633	# ifndef IN_RING0
3634	RTHCPHYS HCPhys; /* general usage. */
3635	# endif
3636	int rc;
3637
3638	/*
3639	* Check that the Guest CR3 and all its mappings are correct.
3640	*/
3641	AssertMsgReturn(pPGM->GCPhysCR3 == (cr3 & GST_CR3_PAGE_MASK),
3642	("Invalid GCPhysCR3=%RGp cr3=%RGp\n", pPGM->GCPhysCR3, (RTGCPHYS)cr3),
3643	false);
3644	# if !defined(IN_RING0) && PGM_GST_TYPE != PGM_TYPE_AMD64
3645	# if PGM_GST_TYPE == PGM_TYPE_32BIT
3646	rc = PGMShwGetPage(pVCpu, (RTRCUINTPTR)pPGM->pGst32BitPdRC, NULL, &HCPhysShw);
3647	# else
3648	rc = PGMShwGetPage(pVCpu, (RTRCUINTPTR)pPGM->pGstPaePdptRC, NULL, &HCPhysShw);
3649	# endif
3650	AssertRCReturn(rc, 1);
3651	HCPhys = NIL_RTHCPHYS;
3652	rc = pgmRamGCPhys2HCPhys(&pVM->pgm.s, cr3 & GST_CR3_PAGE_MASK, &HCPhys);
3653	AssertMsgReturn(HCPhys == HCPhysShw, ("HCPhys=%RHp HCPhyswShw=%RHp (cr3)\n", HCPhys, HCPhysShw), false);
3654	# if PGM_GST_TYPE == PGM_TYPE_32BIT && defined(IN_RING3)
3655	pgmGstGet32bitPDPtr(pPGM);
3656	RTGCPHYS GCPhys;
3657	rc = PGMR3DbgR3Ptr2GCPhys(pVM, pPGM->pGst32BitPdR3, &GCPhys);
3658	AssertRCReturn(rc, 1);
3659	AssertMsgReturn((cr3 & GST_CR3_PAGE_MASK) == GCPhys, ("GCPhys=%RGp cr3=%RGp\n", GCPhys, (RTGCPHYS)cr3), false);
3660	# endif
3661	# endif /* !IN_RING0 */
3662
3663	/*
3664	* Get and check the Shadow CR3.
3665	*/
3666	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
3667	unsigned cPDEs = X86_PG_ENTRIES;
3668	unsigned cIncrement = X86_PG_ENTRIES * PAGE_SIZE;
3669	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
3670	# if PGM_GST_TYPE == PGM_TYPE_32BIT
3671	unsigned cPDEs = X86_PG_PAE_ENTRIES * 4; /* treat it as a 2048 entry table. */
3672	# else
3673	unsigned cPDEs = X86_PG_PAE_ENTRIES;
3674	# endif
3675	unsigned cIncrement = X86_PG_PAE_ENTRIES * PAGE_SIZE;
3676	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
3677	unsigned cPDEs = X86_PG_PAE_ENTRIES;
3678	unsigned cIncrement = X86_PG_PAE_ENTRIES * PAGE_SIZE;
3679	# endif
3680	if (cb != ~(RTGCPTR)0)
3681	cPDEs = RT_MIN(cb >> SHW_PD_SHIFT, 1);
3682
3683	/** @todo call the other two PGMAssert() functions. /
3684
3685	# if PGM_GST_TYPE == PGM_TYPE_AMD64
3686	unsigned iPml4 = (GCPtr >> X86_PML4_SHIFT) & X86_PML4_MASK;
3687
3688	for (; iPml4 < X86_PG_PAE_ENTRIES; iPml4++)
3689	{
3690	PPGMPOOLPAGE pShwPdpt = NULL;
3691	PX86PML4E pPml4eSrc;
3692	PX86PML4E pPml4eDst;
3693	RTGCPHYS GCPhysPdptSrc;
3694
3695	pPml4eSrc = pgmGstGetLongModePML4EPtr(&pVCpu->pgm.s, iPml4);
3696	pPml4eDst = pgmShwGetLongModePML4EPtr(&pVCpu->pgm.s, iPml4);
3697
3698	/* Fetch the pgm pool shadow descriptor if the shadow pml4e is present. */
3699	if (!pPml4eDst->n.u1Present)
3700	{
3701	GCPtr += _2M * UINT64_C(512) * UINT64_C(512);
3702	continue;
3703	}
3704
3705	pShwPdpt = pgmPoolGetPage(pPool, pPml4eDst->u & X86_PML4E_PG_MASK);
3706	GCPhysPdptSrc = pPml4eSrc->u & X86_PML4E_PG_MASK_FULL;
3707
3708	if (pPml4eSrc->n.u1Present != pPml4eDst->n.u1Present)
3709	{
3710	AssertMsgFailed(("Present bit doesn't match! pPml4eDst.u=%#RX64 pPml4eSrc.u=%RX64\n", pPml4eDst->u, pPml4eSrc->u));
3711	GCPtr += _2M * UINT64_C(512) * UINT64_C(512);
3712	cErrors++;
3713	continue;
3714	}
3715
3716	if (GCPhysPdptSrc != pShwPdpt->GCPhys)
3717	{
3718	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));
3719	GCPtr += _2M * UINT64_C(512) * UINT64_C(512);
3720	cErrors++;
3721	continue;
3722	}
3723
3724	if ( pPml4eDst->n.u1User != pPml4eSrc->n.u1User
3725	\|\| pPml4eDst->n.u1Write != pPml4eSrc->n.u1Write
3726	\|\| pPml4eDst->n.u1NoExecute != pPml4eSrc->n.u1NoExecute)
3727	{
3728	AssertMsgFailed(("User/Write/NoExec bits don't match! pPml4eDst.u=%#RX64 pPml4eSrc.u=%RX64\n", pPml4eDst->u, pPml4eSrc->u));
3729	GCPtr += _2M * UINT64_C(512) * UINT64_C(512);
3730	cErrors++;
3731	continue;
3732	}
3733	# else /* PGM_GST_TYPE != PGM_TYPE_AMD64 */
3734	{
3735	# endif /* PGM_GST_TYPE != PGM_TYPE_AMD64 */
3736
3737	# if PGM_GST_TYPE == PGM_TYPE_AMD64 \|\| PGM_GST_TYPE == PGM_TYPE_PAE
3738	/*
3739	* Check the PDPTEs too.
3740	*/
3741	unsigned iPdpt = (GCPtr >> SHW_PDPT_SHIFT) & SHW_PDPT_MASK;
3742
3743	for (;iPdpt <= SHW_PDPT_MASK; iPdpt++)
3744	{
3745	unsigned iPDSrc = 0; /* initialized to shut up gcc */
3746	PPGMPOOLPAGE pShwPde = NULL;
3747	PX86PDPE pPdpeDst;
3748	RTGCPHYS GCPhysPdeSrc;
3749	# if PGM_GST_TYPE == PGM_TYPE_PAE
3750	X86PDPE PdpeSrc;
3751	PGSTPD pPDSrc = pgmGstGetPaePDPtr(&pVCpu->pgm.s, GCPtr, &iPDSrc, &PdpeSrc);
3752	PX86PDPT pPdptDst = pgmShwGetPaePDPTPtr(&pVCpu->pgm.s);
3753	# else
3754	PX86PML4E pPml4eSrcIgn;
3755	X86PDPE PdpeSrc;
3756	PX86PDPT pPdptDst;
3757	PX86PDPAE pPDDst;
3758	PGSTPD pPDSrc = pgmGstGetLongModePDPtr(&pVCpu->pgm.s, GCPtr, &pPml4eSrcIgn, &PdpeSrc, &iPDSrc);
3759
3760	rc = pgmShwGetLongModePDPtr(pVCpu, GCPtr, NULL, &pPdptDst, &pPDDst);
3761	if (rc != VINF_SUCCESS)
3762	{
3763	AssertMsg(rc == VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT, ("Unexpected rc=%Rrc\n", rc));
3764	GCPtr += 512 * _2M;
3765	continue; /* next PDPTE */
3766	}
3767	Assert(pPDDst);
3768	# endif
3769	Assert(iPDSrc == 0);
3770
3771	pPdpeDst = &pPdptDst->a[iPdpt];
3772
3773	if (!pPdpeDst->n.u1Present)
3774	{
3775	GCPtr += 512 * _2M;
3776	continue; /* next PDPTE */
3777	}
3778
3779	pShwPde = pgmPoolGetPage(pPool, pPdpeDst->u & X86_PDPE_PG_MASK);
3780	GCPhysPdeSrc = PdpeSrc.u & X86_PDPE_PG_MASK;
3781
3782	if (pPdpeDst->n.u1Present != PdpeSrc.n.u1Present)
3783	{
3784	AssertMsgFailed(("Present bit doesn't match! pPdpeDst.u=%#RX64 pPdpeSrc.u=%RX64\n", pPdpeDst->u, PdpeSrc.u));
3785	GCPtr += 512 * _2M;
3786	cErrors++;
3787	continue;
3788	}
3789
3790	if (GCPhysPdeSrc != pShwPde->GCPhys)
3791	{
3792	# if PGM_GST_TYPE == PGM_TYPE_AMD64
3793	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));
3794	# else
3795	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));
3796	# endif
3797	GCPtr += 512 * _2M;
3798	cErrors++;
3799	continue;
3800	}
3801
3802	# if PGM_GST_TYPE == PGM_TYPE_AMD64
3803	if ( pPdpeDst->lm.u1User != PdpeSrc.lm.u1User
3804	\|\| pPdpeDst->lm.u1Write != PdpeSrc.lm.u1Write
3805	\|\| pPdpeDst->lm.u1NoExecute != PdpeSrc.lm.u1NoExecute)
3806	{
3807	AssertMsgFailed(("User/Write/NoExec bits don't match! pPdpeDst.u=%#RX64 pPdpeSrc.u=%RX64\n", pPdpeDst->u, PdpeSrc.u));
3808	GCPtr += 512 * _2M;
3809	cErrors++;
3810	continue;
3811	}
3812	# endif
3813
3814	# else /* PGM_GST_TYPE != PGM_TYPE_AMD64 && PGM_GST_TYPE != PGM_TYPE_PAE */
3815	{
3816	# endif /* PGM_GST_TYPE != PGM_TYPE_AMD64 && PGM_GST_TYPE != PGM_TYPE_PAE */
3817	# if PGM_GST_TYPE == PGM_TYPE_32BIT
3818	GSTPD const *pPDSrc = pgmGstGet32bitPDPtr(&pVCpu->pgm.s);
3819	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
3820	PCX86PD pPDDst = pgmShwGet32BitPDPtr(&pVCpu->pgm.s);
3821	# endif
3822	# endif /* PGM_GST_TYPE == PGM_TYPE_32BIT */
3823	/*
3824	* Iterate the shadow page directory.
3825	*/
3826	GCPtr = (GCPtr >> SHW_PD_SHIFT) << SHW_PD_SHIFT;
3827	unsigned iPDDst = (GCPtr >> SHW_PD_SHIFT) & SHW_PD_MASK;
3828
3829	for (;
3830	iPDDst < cPDEs;
3831	iPDDst++, GCPtr += cIncrement)
3832	{
3833	# if PGM_SHW_TYPE == PGM_TYPE_PAE
3834	const SHWPDE PdeDst = *pgmShwGetPaePDEPtr(pPGM, GCPtr);
3835	# else
3836	const SHWPDE PdeDst = pPDDst->a[iPDDst];
3837	# endif
3838	if (PdeDst.u & PGM_PDFLAGS_MAPPING)
3839	{
3840	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
3841	if ((PdeDst.u & X86_PDE_AVL_MASK) != PGM_PDFLAGS_MAPPING)
3842	{
3843	AssertMsgFailed(("Mapping shall only have PGM_PDFLAGS_MAPPING set! PdeDst.u=%#RX64\n", (uint64_t)PdeDst.u));
3844	cErrors++;
3845	continue;
3846	}
3847	}
3848	else if ( (PdeDst.u & X86_PDE_P)
3849	\|\| ((PdeDst.u & (X86_PDE_P \| PGM_PDFLAGS_TRACK_DIRTY)) == (X86_PDE_P \| PGM_PDFLAGS_TRACK_DIRTY))
3850	)
3851	{
3852	HCPhysShw = PdeDst.u & SHW_PDE_PG_MASK;
3853	PPGMPOOLPAGE pPoolPage = pgmPoolGetPage(pPool, HCPhysShw);
3854	if (!pPoolPage)
3855	{
3856	AssertMsgFailed(("Invalid page table address %RHp at %RGv! PdeDst=%#RX64\n",
3857	HCPhysShw, GCPtr, (uint64_t)PdeDst.u));
3858	cErrors++;
3859	continue;
3860	}
3861	const SHWPT pPTDst = (const SHWPT )PGMPOOL_PAGE_2_PTR(pVM, pPoolPage);
3862
3863	if (PdeDst.u & (X86_PDE4M_PWT \| X86_PDE4M_PCD))
3864	{
3865	AssertMsgFailed(("PDE flags PWT and/or PCD is set at %RGv! These flags are not virtualized! PdeDst=%#RX64\n",
3866	GCPtr, (uint64_t)PdeDst.u));
3867	cErrors++;
3868	}
3869
3870	if (PdeDst.u & (X86_PDE4M_G \| X86_PDE4M_D))
3871	{
3872	AssertMsgFailed(("4K PDE reserved flags at %RGv! PdeDst=%#RX64\n",
3873	GCPtr, (uint64_t)PdeDst.u));
3874	cErrors++;
3875	}
3876
3877	const GSTPDE PdeSrc = pPDSrc->a[(iPDDst >> (GST_PD_SHIFT - SHW_PD_SHIFT)) & GST_PD_MASK];
3878	if (!PdeSrc.n.u1Present)
3879	{
3880	AssertMsgFailed(("Guest PDE at %RGv is not present! PdeDst=%#RX64 PdeSrc=%#RX64\n",
3881	GCPtr, (uint64_t)PdeDst.u, (uint64_t)PdeSrc.u));
3882	cErrors++;
3883	continue;
3884	}
3885
3886	if ( !PdeSrc.b.u1Size
3887	\|\| !fBigPagesSupported)
3888	{
3889	GCPhysGst = PdeSrc.u & GST_PDE_PG_MASK;
3890	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
3891	GCPhysGst \|= (iPDDst & 1) * (PAGE_SIZE / 2);
3892	# endif
3893	}
3894	else
3895	{
3896	# if PGM_GST_TYPE == PGM_TYPE_32BIT
3897	if (PdeSrc.u & X86_PDE4M_PG_HIGH_MASK)
3898	{
3899	AssertMsgFailed(("Guest PDE at %RGv is using PSE36 or similar! PdeSrc=%#RX64\n",
3900	GCPtr, (uint64_t)PdeSrc.u));
3901	cErrors++;
3902	continue;
3903	}
3904	# endif
3905	GCPhysGst = GST_GET_PDE_BIG_PG_GCPHYS(PdeSrc);
3906	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
3907	GCPhysGst \|= GCPtr & RT_BIT(X86_PAGE_2M_SHIFT);
3908	# endif
3909	}
3910
3911	if ( pPoolPage->enmKind
3912	!= (!PdeSrc.b.u1Size \|\| !fBigPagesSupported ? BTH_PGMPOOLKIND_PT_FOR_PT : BTH_PGMPOOLKIND_PT_FOR_BIG))
3913	{
3914	AssertMsgFailed(("Invalid shadow page table kind %d at %RGv! PdeSrc=%#RX64\n",
3915	pPoolPage->enmKind, GCPtr, (uint64_t)PdeSrc.u));
3916	cErrors++;
3917	}
3918
3919	PPGMPAGE pPhysPage = pgmPhysGetPage(&pVM->pgm.s, GCPhysGst);
3920	if (!pPhysPage)
3921	{
3922	AssertMsgFailed(("Cannot find guest physical address %RGp in the PDE at %RGv! PdeSrc=%#RX64\n",
3923	GCPhysGst, GCPtr, (uint64_t)PdeSrc.u));
3924	cErrors++;
3925	continue;
3926	}
3927
3928	if (GCPhysGst != pPoolPage->GCPhys)
3929	{
3930	AssertMsgFailed(("GCPhysGst=%RGp != pPage->GCPhys=%RGp at %RGv\n",
3931	GCPhysGst, pPoolPage->GCPhys, GCPtr));
3932	cErrors++;
3933	continue;
3934	}
3935
3936	if ( !PdeSrc.b.u1Size
3937	\|\| !fBigPagesSupported)
3938	{
3939	/*
3940	* Page Table.
3941	*/
3942	const GSTPT *pPTSrc;
3943	rc = PGM_GCPHYS_2_PTR(pVM, GCPhysGst & ~(RTGCPHYS)(PAGE_SIZE - 1), &pPTSrc);
3944	if (RT_FAILURE(rc))
3945	{
3946	AssertMsgFailed(("Cannot map/convert guest physical address %RGp in the PDE at %RGv! PdeSrc=%#RX64\n",
3947	GCPhysGst, GCPtr, (uint64_t)PdeSrc.u));
3948	cErrors++;
3949	continue;
3950	}
3951	if ( (PdeSrc.u & (X86_PDE_P \| X86_PDE_US \| X86_PDE_RW/* \| X86_PDE_A*/))
3952	!= (PdeDst.u & (X86_PDE_P \| X86_PDE_US \| X86_PDE_RW/* \| X86_PDE_A*/)))
3953	{
3954	/// @todo We get here a lot on out-of-sync CR3 entries. The access handler should zap them to avoid false alarms here!
3955	// (This problem will go away when/if we shadow multiple CR3s.)
3956	AssertMsgFailed(("4K PDE flags mismatch at %RGv! PdeSrc=%#RX64 PdeDst=%#RX64\n",
3957	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
3958	cErrors++;
3959	continue;
3960	}
3961	if (PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY)
3962	{
3963	AssertMsgFailed(("4K PDEs cannot have PGM_PDFLAGS_TRACK_DIRTY set! GCPtr=%RGv PdeDst=%#RX64\n",
3964	GCPtr, (uint64_t)PdeDst.u));
3965	cErrors++;
3966	continue;
3967	}
3968
3969	/* iterate the page table. */
3970	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
3971	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
3972	const unsigned offPTSrc = ((GCPtr >> SHW_PD_SHIFT) & 1) * 512;
3973	# else
3974	const unsigned offPTSrc = 0;
3975	# endif
3976	for (unsigned iPT = 0, off = 0;
3977	iPT < RT_ELEMENTS(pPTDst->a);
3978	iPT++, off += PAGE_SIZE)
3979	{
3980	const SHWPTE PteDst = pPTDst->a[iPT];
3981
3982	/* skip not-present entries. */
3983	if (!(PteDst.u & (X86_PTE_P \| PGM_PTFLAGS_TRACK_DIRTY))) /** @todo deal with ALL handlers and CSAM !P pages! */
3984	continue;
3985	Assert(PteDst.n.u1Present);
3986
3987	const GSTPTE PteSrc = pPTSrc->a[iPT + offPTSrc];
3988	if (!PteSrc.n.u1Present)
3989	{
3990	# ifdef IN_RING3
3991	PGMAssertHandlerAndFlagsInSync(pVM);
3992	PGMR3DumpHierarchyGC(pVM, cr3, cr4, (PdeSrc.u & GST_PDE_PG_MASK));
3993	# endif
3994	AssertMsgFailed(("Out of sync (!P) PTE at %RGv! PteSrc=%#RX64 PteDst=%#RX64 pPTSrc=%RGv iPTSrc=%x PdeSrc=%x physpte=%RGp\n",
3995	GCPtr + off, (uint64_t)PteSrc.u, (uint64_t)PteDst.u, pPTSrc, iPT + offPTSrc, PdeSrc.au32[0],
3996	(PdeSrc.u & GST_PDE_PG_MASK) + (iPT + offPTSrc)*sizeof(PteSrc)));
3997	cErrors++;
3998	continue;
3999	}
4000
4001	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;
4002	# if 1 /** @todo sync accessed bit properly... */
4003	fIgnoreFlags \|= X86_PTE_A;
4004	# endif
4005
4006	/* match the physical addresses */
4007	HCPhysShw = PteDst.u & SHW_PTE_PG_MASK;
4008	GCPhysGst = PteSrc.u & GST_PTE_PG_MASK;
4009
4010	# ifdef IN_RING3
4011	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysGst, &HCPhys);
4012	if (RT_FAILURE(rc))
4013	{
4014	if (HCPhysShw != MMR3PageDummyHCPhys(pVM)) /** @todo this is wrong. */
4015	{
4016	AssertMsgFailed(("Cannot find guest physical address %RGp at %RGv! PteSrc=%#RX64 PteDst=%#RX64\n",
4017	GCPhysGst, GCPtr + off, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4018	cErrors++;
4019	continue;
4020	}
4021	}
4022	else if (HCPhysShw != (HCPhys & SHW_PTE_PG_MASK))
4023	{
4024	AssertMsgFailed(("Out of sync (phys) at %RGv! HCPhysShw=%RHp HCPhys=%RHp GCPhysGst=%RGp PteSrc=%#RX64 PteDst=%#RX64\n",
4025	GCPtr + off, HCPhysShw, HCPhys, GCPhysGst, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4026	cErrors++;
4027	continue;
4028	}
4029	# endif
4030
4031	pPhysPage = pgmPhysGetPage(&pVM->pgm.s, GCPhysGst);
4032	if (!pPhysPage)
4033	{
4034	# ifdef IN_RING3 /** @todo make MMR3PageDummyHCPhys an 'All' function! */
4035	if (HCPhysShw != MMR3PageDummyHCPhys(pVM)) /** @todo this is wrong. */
4036	{
4037	AssertMsgFailed(("Cannot find guest physical address %RGp at %RGv! PteSrc=%#RX64 PteDst=%#RX64\n",
4038	GCPhysGst, GCPtr + off, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4039	cErrors++;
4040	continue;
4041	}
4042	# endif
4043	if (PteDst.n.u1Write)
4044	{
4045	AssertMsgFailed(("Invalid guest page at %RGv is writable! GCPhysGst=%RGp PteSrc=%#RX64 PteDst=%#RX64\n",
4046	GCPtr + off, GCPhysGst, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4047	cErrors++;
4048	}
4049	fIgnoreFlags \|= X86_PTE_RW;
4050	}
4051	else if (HCPhysShw != PGM_PAGE_GET_HCPHYS(pPhysPage))
4052	{
4053	AssertMsgFailed(("Out of sync (phys) at %RGv! HCPhysShw=%RHp pPhysPage:%R[pgmpage] GCPhysGst=%RGp PteSrc=%#RX64 PteDst=%#RX64\n",
4054	GCPtr + off, HCPhysShw, pPhysPage, GCPhysGst, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4055	cErrors++;
4056	continue;
4057	}
4058
4059	/* flags */
4060	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPhysPage))
4061	{
4062	if (!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPhysPage))
4063	{
4064	if (PteDst.n.u1Write)
4065	{
4066	AssertMsgFailed(("WRITE access flagged at %RGv but the page is writable! pPhysPage=%R[pgmpage] PteSrc=%#RX64 PteDst=%#RX64\n",
4067	GCPtr + off, pPhysPage, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4068	cErrors++;
4069	continue;
4070	}
4071	fIgnoreFlags \|= X86_PTE_RW;
4072	}
4073	else
4074	{
4075	if (PteDst.n.u1Present)
4076	{
4077	AssertMsgFailed(("ALL access flagged at %RGv but the page is present! pPhysPage=%R[pgmpage] PteSrc=%#RX64 PteDst=%#RX64\n",
4078	GCPtr + off, pPhysPage, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4079	cErrors++;
4080	continue;
4081	}
4082	fIgnoreFlags \|= X86_PTE_P;
4083	}
4084	}
4085	else
4086	{
4087	if (!PteSrc.n.u1Dirty && PteSrc.n.u1Write)
4088	{
4089	if (PteDst.n.u1Write)
4090	{
4091	AssertMsgFailed(("!DIRTY page at %RGv is writable! PteSrc=%#RX64 PteDst=%#RX64\n",
4092	GCPtr + off, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4093	cErrors++;
4094	continue;
4095	}
4096	if (!(PteDst.u & PGM_PTFLAGS_TRACK_DIRTY))
4097	{
4098	AssertMsgFailed(("!DIRTY page at %RGv is not marked TRACK_DIRTY! PteSrc=%#RX64 PteDst=%#RX64\n",
4099	GCPtr + off, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4100	cErrors++;
4101	continue;
4102	}
4103	if (PteDst.n.u1Dirty)
4104	{
4105	AssertMsgFailed(("!DIRTY page at %RGv is marked DIRTY! PteSrc=%#RX64 PteDst=%#RX64\n",
4106	GCPtr + off, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4107	cErrors++;
4108	}
4109	# if 0 /** @todo sync access bit properly... */
4110	if (PteDst.n.u1Accessed != PteSrc.n.u1Accessed)
4111	{
4112	AssertMsgFailed(("!DIRTY page at %RGv is has mismatching accessed bit! PteSrc=%#RX64 PteDst=%#RX64\n",
4113	GCPtr + off, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4114	cErrors++;
4115	}
4116	fIgnoreFlags \|= X86_PTE_RW;
4117	# else
4118	fIgnoreFlags \|= X86_PTE_RW \| X86_PTE_A;
4119	# endif
4120	}
4121	else if (PteDst.u & PGM_PTFLAGS_TRACK_DIRTY)
4122	{
4123	/* access bit emulation (not implemented). */
4124	if (PteSrc.n.u1Accessed \|\| PteDst.n.u1Present)
4125	{
4126	AssertMsgFailed(("PGM_PTFLAGS_TRACK_DIRTY set at %RGv but no accessed bit emulation! PteSrc=%#RX64 PteDst=%#RX64\n",
4127	GCPtr + off, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4128	cErrors++;
4129	continue;
4130	}
4131	if (!PteDst.n.u1Accessed)
4132	{
4133	AssertMsgFailed(("!ACCESSED page at %RGv is has the accessed bit set! PteSrc=%#RX64 PteDst=%#RX64\n",
4134	GCPtr + off, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4135	cErrors++;
4136	}
4137	fIgnoreFlags \|= X86_PTE_P;
4138	}
4139	# ifdef DEBUG_sandervl
4140	fIgnoreFlags \|= X86_PTE_D \| X86_PTE_A;
4141	# endif
4142	}
4143
4144	if ( (PteSrc.u & ~fIgnoreFlags) != (PteDst.u & ~fIgnoreFlags)
4145	&& (PteSrc.u & ~(fIgnoreFlags \| X86_PTE_RW)) != (PteDst.u & ~fIgnoreFlags)
4146	)
4147	{
4148	AssertMsgFailed(("Flags mismatch at %RGv! %#RX64 != %#RX64 fIgnoreFlags=%#RX64 PteSrc=%#RX64 PteDst=%#RX64\n",
4149	GCPtr + off, (uint64_t)PteSrc.u & ~fIgnoreFlags, (uint64_t)PteDst.u & ~fIgnoreFlags,
4150	fIgnoreFlags, (uint64_t)PteSrc.u, (uint64_t)PteDst.u));
4151	cErrors++;
4152	continue;
4153	}
4154	} /* foreach PTE */
4155	}
4156	else
4157	{
4158	/*
4159	* Big Page.
4160	*/
4161	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;
4162	if (!PdeSrc.b.u1Dirty && PdeSrc.b.u1Write)
4163	{
4164	if (PdeDst.n.u1Write)
4165	{
4166	AssertMsgFailed(("!DIRTY page at %RGv is writable! PdeSrc=%#RX64 PdeDst=%#RX64\n",
4167	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4168	cErrors++;
4169	continue;
4170	}
4171	if (!(PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY))
4172	{
4173	AssertMsgFailed(("!DIRTY page at %RGv is not marked TRACK_DIRTY! PteSrc=%#RX64 PteDst=%#RX64\n",
4174	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4175	cErrors++;
4176	continue;
4177	}
4178	# if 0 /** @todo sync access bit properly... */
4179	if (PdeDst.n.u1Accessed != PdeSrc.b.u1Accessed)
4180	{
4181	AssertMsgFailed(("!DIRTY page at %RGv is has mismatching accessed bit! PteSrc=%#RX64 PteDst=%#RX64\n",
4182	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4183	cErrors++;
4184	}
4185	fIgnoreFlags \|= X86_PTE_RW;
4186	# else
4187	fIgnoreFlags \|= X86_PTE_RW \| X86_PTE_A;
4188	# endif
4189	}
4190	else if (PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY)
4191	{
4192	/* access bit emulation (not implemented). */
4193	if (PdeSrc.b.u1Accessed \|\| PdeDst.n.u1Present)
4194	{
4195	AssertMsgFailed(("PGM_PDFLAGS_TRACK_DIRTY set at %RGv but no accessed bit emulation! PdeSrc=%#RX64 PdeDst=%#RX64\n",
4196	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4197	cErrors++;
4198	continue;
4199	}
4200	if (!PdeDst.n.u1Accessed)
4201	{
4202	AssertMsgFailed(("!ACCESSED page at %RGv is has the accessed bit set! PdeSrc=%#RX64 PdeDst=%#RX64\n",
4203	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4204	cErrors++;
4205	}
4206	fIgnoreFlags \|= X86_PTE_P;
4207	}
4208
4209	if ((PdeSrc.u & ~fIgnoreFlags) != (PdeDst.u & ~fIgnoreFlags))
4210	{
4211	AssertMsgFailed(("Flags mismatch (B) at %RGv! %#RX64 != %#RX64 fIgnoreFlags=%#RX64 PdeSrc=%#RX64 PdeDst=%#RX64\n",
4212	GCPtr, (uint64_t)PdeSrc.u & ~fIgnoreFlags, (uint64_t)PdeDst.u & ~fIgnoreFlags,
4213	fIgnoreFlags, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4214	cErrors++;
4215	}
4216
4217	/* iterate the page table. */
4218	for (unsigned iPT = 0, off = 0;
4219	iPT < RT_ELEMENTS(pPTDst->a);
4220	iPT++, off += PAGE_SIZE, GCPhysGst += PAGE_SIZE)
4221	{
4222	const SHWPTE PteDst = pPTDst->a[iPT];
4223
4224	if (PteDst.u & PGM_PTFLAGS_TRACK_DIRTY)
4225	{
4226	AssertMsgFailed(("The PTE at %RGv emulating a 2/4M page is marked TRACK_DIRTY! PdeSrc=%#RX64 PteDst=%#RX64\n",
4227	GCPtr + off, (uint64_t)PdeSrc.u, (uint64_t)PteDst.u));
4228	cErrors++;
4229	}
4230
4231	/* skip not-present entries. */
4232	if (!PteDst.n.u1Present) /** @todo deal with ALL handlers and CSAM !P pages! */
4233	continue;
4234
4235	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;
4236
4237	/* match the physical addresses */
4238	HCPhysShw = PteDst.u & X86_PTE_PAE_PG_MASK;
4239
4240	# ifdef IN_RING3
4241	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysGst, &HCPhys);
4242	if (RT_FAILURE(rc))
4243	{
4244	if (HCPhysShw != MMR3PageDummyHCPhys(pVM)) /** @todo this is wrong. */
4245	{
4246	AssertMsgFailed(("Cannot find guest physical address %RGp at %RGv! PdeSrc=%#RX64 PteDst=%#RX64\n",
4247	GCPhysGst, GCPtr + off, (uint64_t)PdeSrc.u, (uint64_t)PteDst.u));
4248	cErrors++;
4249	}
4250	}
4251	else if (HCPhysShw != (HCPhys & X86_PTE_PAE_PG_MASK))
4252	{
4253	AssertMsgFailed(("Out of sync (phys) at %RGv! HCPhysShw=%RHp HCPhys=%RHp GCPhysGst=%RGp PdeSrc=%#RX64 PteDst=%#RX64\n",
4254	GCPtr + off, HCPhysShw, HCPhys, GCPhysGst, (uint64_t)PdeSrc.u, (uint64_t)PteDst.u));
4255	cErrors++;
4256	continue;
4257	}
4258	# endif
4259	pPhysPage = pgmPhysGetPage(&pVM->pgm.s, GCPhysGst);
4260	if (!pPhysPage)
4261	{
4262	# ifdef IN_RING3 /** @todo make MMR3PageDummyHCPhys an 'All' function! */
4263	if (HCPhysShw != MMR3PageDummyHCPhys(pVM)) /** @todo this is wrong. */
4264	{
4265	AssertMsgFailed(("Cannot find guest physical address %RGp at %RGv! PdeSrc=%#RX64 PteDst=%#RX64\n",
4266	GCPhysGst, GCPtr + off, (uint64_t)PdeSrc.u, (uint64_t)PteDst.u));
4267	cErrors++;
4268	continue;
4269	}
4270	# endif
4271	if (PteDst.n.u1Write)
4272	{
4273	AssertMsgFailed(("Invalid guest page at %RGv is writable! GCPhysGst=%RGp PdeSrc=%#RX64 PteDst=%#RX64\n",
4274	GCPtr + off, GCPhysGst, (uint64_t)PdeSrc.u, (uint64_t)PteDst.u));
4275	cErrors++;
4276	}
4277	fIgnoreFlags \|= X86_PTE_RW;
4278	}
4279	else if (HCPhysShw != PGM_PAGE_GET_HCPHYS(pPhysPage))
4280	{
4281	AssertMsgFailed(("Out of sync (phys) at %RGv! HCPhysShw=%RHp pPhysPage=%R[pgmpage] GCPhysGst=%RGp PdeSrc=%#RX64 PteDst=%#RX64\n",
4282	GCPtr + off, HCPhysShw, pPhysPage, GCPhysGst, (uint64_t)PdeSrc.u, (uint64_t)PteDst.u));
4283	cErrors++;
4284	continue;
4285	}
4286
4287	/* flags */
4288	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPhysPage))
4289	{
4290	if (!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPhysPage))
4291	{
4292	if (PGM_PAGE_GET_HNDL_PHYS_STATE(pPhysPage) != PGM_PAGE_HNDL_PHYS_STATE_DISABLED)
4293	{
4294	if (PteDst.n.u1Write)
4295	{
4296	AssertMsgFailed(("WRITE access flagged at %RGv but the page is writable! pPhysPage=%R[pgmpage] PdeSrc=%#RX64 PteDst=%#RX64\n",
4297	GCPtr + off, pPhysPage, (uint64_t)PdeSrc.u, (uint64_t)PteDst.u));
4298	cErrors++;
4299	continue;
4300	}
4301	fIgnoreFlags \|= X86_PTE_RW;
4302	}
4303	}
4304	else
4305	{
4306	if (PteDst.n.u1Present)
4307	{
4308	AssertMsgFailed(("ALL access flagged at %RGv but the page is present! pPhysPage=%R[pgmpage] PdeSrc=%#RX64 PteDst=%#RX64\n",
4309	GCPtr + off, pPhysPage, (uint64_t)PdeSrc.u, (uint64_t)PteDst.u));
4310	cErrors++;
4311	continue;
4312	}
4313	fIgnoreFlags \|= X86_PTE_P;
4314	}
4315	}
4316
4317	if ( (PdeSrc.u & ~fIgnoreFlags) != (PteDst.u & ~fIgnoreFlags)
4318	&& (PdeSrc.u & ~(fIgnoreFlags \| X86_PTE_RW)) != (PteDst.u & ~fIgnoreFlags) /* lazy phys handler dereg. */
4319	)
4320	{
4321	AssertMsgFailed(("Flags mismatch (BT) at %RGv! %#RX64 != %#RX64 fIgnoreFlags=%#RX64 PdeSrc=%#RX64 PteDst=%#RX64\n",
4322	GCPtr + off, (uint64_t)PdeSrc.u & ~fIgnoreFlags, (uint64_t)PteDst.u & ~fIgnoreFlags,
4323	fIgnoreFlags, (uint64_t)PdeSrc.u, (uint64_t)PteDst.u));
4324	cErrors++;
4325	continue;
4326	}
4327	} /* for each PTE */
4328	}
4329	}
4330	/* not present */
4331
4332	} /* for each PDE */
4333
4334	} /* for each PDPTE */
4335
4336	} /* for each PML4E */
4337
4338	# ifdef DEBUG
4339	if (cErrors)
4340	LogFlow(("AssertCR3: cErrors=%d\n", cErrors));
4341	# endif
4342
4343	#endif /* GST == 32BIT, PAE or AMD64 */
4344	return cErrors;
4345
4346	#endif /* PGM_SHW_TYPE != PGM_TYPE_NESTED && PGM_SHW_TYPE != PGM_TYPE_EPT */
4347	}
4348	#endif /* VBOX_STRICT */
4349
4350
4351	/**
4352	* Sets up the CR3 for shadow paging
4353	*
4354	* @returns Strict VBox status code.
4355	* @retval VINF_SUCCESS.
4356	*
4357	* @param pVCpu The VMCPU handle.
4358	* @param GCPhysCR3 The physical address in the CR3 register.
4359	*/
4360	PGM_BTH_DECL(int, MapCR3)(PVMCPU pVCpu, RTGCPHYS GCPhysCR3)
4361	{
4362	PVM pVM = pVCpu->CTX_SUFF(pVM);
4363
4364	/* Update guest paging info. */
4365	#if PGM_GST_TYPE == PGM_TYPE_32BIT \
4366	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
4367	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64
4368
4369	LogFlow(("MapCR3: %RGp\n", GCPhysCR3));
4370
4371	/*
4372	* Map the page CR3 points at.
4373	*/
4374	RTHCPTR HCPtrGuestCR3;
4375	RTHCPHYS HCPhysGuestCR3;
4376	pgmLock(pVM);
4377	PPGMPAGE pPageCR3 = pgmPhysGetPage(&pVM->pgm.s, GCPhysCR3);
4378	AssertReturn(pPageCR3, VERR_INTERNAL_ERROR_2);
4379	HCPhysGuestCR3 = PGM_PAGE_GET_HCPHYS(pPageCR3);
4380	/** @todo this needs some reworking wrt. locking. */
4381	# if defined(IN_RC) \|\| defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0)
4382	HCPtrGuestCR3 = NIL_RTHCPTR;
4383	int rc = VINF_SUCCESS;
4384	# else
4385	int rc = pgmPhysGCPhys2CCPtrInternal(pVM, pPageCR3, GCPhysCR3 & GST_CR3_PAGE_MASK, (void )&HCPtrGuestCR3); / @todo r=bird: This GCPhysCR3 masking isn't necessary. */
4386	# endif
4387	pgmUnlock(pVM);
4388	if (RT_SUCCESS(rc))
4389	{
4390	rc = PGMMap(pVM, (RTGCPTR)pVM->pgm.s.GCPtrCR3Mapping, HCPhysGuestCR3, PAGE_SIZE, 0);
4391	if (RT_SUCCESS(rc))
4392	{
4393	# ifdef IN_RC
4394	PGM_INVL_PG(pVCpu, pVM->pgm.s.GCPtrCR3Mapping);
4395	# endif
4396	# if PGM_GST_TYPE == PGM_TYPE_32BIT
4397	pVCpu->pgm.s.pGst32BitPdR3 = (R3PTRTYPE(PX86PD))HCPtrGuestCR3;
4398	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4399	pVCpu->pgm.s.pGst32BitPdR0 = (R0PTRTYPE(PX86PD))HCPtrGuestCR3;
4400	# endif
4401	pVCpu->pgm.s.pGst32BitPdRC = (RCPTRTYPE(PX86PD))(RTRCUINTPTR)pVM->pgm.s.GCPtrCR3Mapping;
4402
4403	# elif PGM_GST_TYPE == PGM_TYPE_PAE
4404	unsigned off = GCPhysCR3 & GST_CR3_PAGE_MASK & PAGE_OFFSET_MASK;
4405	pVCpu->pgm.s.pGstPaePdptR3 = (R3PTRTYPE(PX86PDPT))HCPtrGuestCR3;
4406	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4407	pVCpu->pgm.s.pGstPaePdptR0 = (R0PTRTYPE(PX86PDPT))HCPtrGuestCR3;
4408	# endif
4409	pVCpu->pgm.s.pGstPaePdptRC = (RCPTRTYPE(PX86PDPT))((RTRCUINTPTR)pVM->pgm.s.GCPtrCR3Mapping + off);
4410	LogFlow(("Cached mapping %RRv\n", pVCpu->pgm.s.pGstPaePdptRC));
4411
4412	/*
4413	* Map the 4 PDs too.
4414	*/
4415	PX86PDPT pGuestPDPT = pgmGstGetPaePDPTPtr(&pVCpu->pgm.s);
4416	RTGCPTR GCPtr = pVM->pgm.s.GCPtrCR3Mapping + PAGE_SIZE;
4417	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++, GCPtr += PAGE_SIZE)
4418	{
4419	if (pGuestPDPT->a[i].n.u1Present)
4420	{
4421	RTHCPTR HCPtr;
4422	RTHCPHYS HCPhys;
4423	RTGCPHYS GCPhys = pGuestPDPT->a[i].u & X86_PDPE_PG_MASK;
4424	pgmLock(pVM);
4425	PPGMPAGE pPage = pgmPhysGetPage(&pVM->pgm.s, GCPhys);
4426	AssertReturn(pPage, VERR_INTERNAL_ERROR_2);
4427	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
4428	# if defined(IN_RC) \|\| defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0)
4429	HCPtr = NIL_RTHCPTR;
4430	int rc2 = VINF_SUCCESS;
4431	# else
4432	int rc2 = pgmPhysGCPhys2CCPtrInternal(pVM, pPage, GCPhys, (void **)&HCPtr);
4433	# endif
4434	pgmUnlock(pVM);
4435	if (RT_SUCCESS(rc2))
4436	{
4437	rc = PGMMap(pVM, GCPtr, HCPhys, PAGE_SIZE, 0);
4438	AssertRCReturn(rc, rc);
4439
4440	pVCpu->pgm.s.apGstPaePDsR3[i] = (R3PTRTYPE(PX86PDPAE))HCPtr;
4441	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4442	pVCpu->pgm.s.apGstPaePDsR0[i] = (R0PTRTYPE(PX86PDPAE))HCPtr;
4443	# endif
4444	pVCpu->pgm.s.apGstPaePDsRC[i] = (RCPTRTYPE(PX86PDPAE))(RTRCUINTPTR)GCPtr;
4445	pVCpu->pgm.s.aGCPhysGstPaePDs[i] = GCPhys;
4446	# ifdef IN_RC
4447	PGM_INVL_PG(pVCpu, GCPtr);
4448	# endif
4449	continue;
4450	}
4451	AssertMsgFailed(("pgmR3Gst32BitMapCR3: rc2=%d GCPhys=%RGp i=%d\n", rc2, GCPhys, i));
4452	}
4453
4454	pVCpu->pgm.s.apGstPaePDsR3[i] = 0;
4455	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4456	pVCpu->pgm.s.apGstPaePDsR0[i] = 0;
4457	# endif
4458	pVCpu->pgm.s.apGstPaePDsRC[i] = 0;
4459	pVCpu->pgm.s.aGCPhysGstPaePDs[i] = NIL_RTGCPHYS;
4460	# ifdef IN_RC
4461	PGM_INVL_PG(pVCpu, GCPtr); /** @todo this shouldn't be necessary? */
4462	# endif
4463	}
4464
4465	# elif PGM_GST_TYPE == PGM_TYPE_AMD64
4466	pVCpu->pgm.s.pGstAmd64Pml4R3 = (R3PTRTYPE(PX86PML4))HCPtrGuestCR3;
4467	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4468	pVCpu->pgm.s.pGstAmd64Pml4R0 = (R0PTRTYPE(PX86PML4))HCPtrGuestCR3;
4469	# endif
4470	# endif
4471	}
4472	else
4473	AssertMsgFailed(("rc=%Rrc GCPhysGuestPD=%RGp\n", rc, GCPhysCR3));
4474	}
4475	else
4476	AssertMsgFailed(("rc=%Rrc GCPhysGuestPD=%RGp\n", rc, GCPhysCR3));
4477
4478	#else /* prot/real stub */
4479	int rc = VINF_SUCCESS;
4480	#endif
4481
4482	/* Update shadow paging info for guest modes with paging (32, pae, 64). */
4483	# if ( ( PGM_SHW_TYPE == PGM_TYPE_32BIT \
4484	\|\| PGM_SHW_TYPE == PGM_TYPE_PAE \
4485	\|\| PGM_SHW_TYPE == PGM_TYPE_AMD64) \
4486	&& ( PGM_GST_TYPE != PGM_TYPE_REAL \
4487	&& PGM_GST_TYPE != PGM_TYPE_PROT))
4488
4489	Assert(!HWACCMIsNestedPagingActive(pVM));
4490
4491	/*
4492	* Update the shadow root page as well since that's not fixed.
4493	*/
4494	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4495	PPGMPOOLPAGE pOldShwPageCR3 = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
4496	uint32_t iOldShwUserTable = pVCpu->pgm.s.iShwUserTable;
4497	uint32_t iOldShwUser = pVCpu->pgm.s.iShwUser;
4498	PPGMPOOLPAGE pNewShwPageCR3;
4499
4500	pgmLock(pVM);
4501
4502	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
4503	if (pPool->cDirtyPages)
4504	pgmPoolResetDirtyPages(pVM);
4505	# endif
4506
4507	Assert(!(GCPhysCR3 >> (PAGE_SHIFT + 32)));
4508	rc = pgmPoolAlloc(pVM, GCPhysCR3 & GST_CR3_PAGE_MASK, BTH_PGMPOOLKIND_ROOT, SHW_POOL_ROOT_IDX, GCPhysCR3 >> PAGE_SHIFT, &pNewShwPageCR3, true /* lock page */);
4509	AssertFatalRC(rc);
4510	rc = VINF_SUCCESS;
4511
4512	# ifdef IN_RC
4513	/*
4514	* WARNING! We can't deal with jumps to ring 3 in the code below as the
4515	* state will be inconsistent! Flush important things now while
4516	* we still can and then make sure there are no ring-3 calls.
4517	*/
4518	REMNotifyHandlerPhysicalFlushIfAlmostFull(pVM, pVCpu);
4519	VMMRZCallRing3Disable(pVCpu);
4520	# endif
4521
4522	pVCpu->pgm.s.iShwUser = SHW_POOL_ROOT_IDX;
4523	pVCpu->pgm.s.iShwUserTable = GCPhysCR3 >> PAGE_SHIFT;
4524	pVCpu->pgm.s.CTX_SUFF(pShwPageCR3) = pNewShwPageCR3;
4525	# ifdef IN_RING0
4526	pVCpu->pgm.s.pShwPageCR3R3 = MMHyperCCToR3(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4527	pVCpu->pgm.s.pShwPageCR3RC = MMHyperCCToRC(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4528	# elif defined(IN_RC)
4529	pVCpu->pgm.s.pShwPageCR3R3 = MMHyperCCToR3(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4530	pVCpu->pgm.s.pShwPageCR3R0 = MMHyperCCToR0(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4531	# else
4532	pVCpu->pgm.s.pShwPageCR3R0 = MMHyperCCToR0(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4533	pVCpu->pgm.s.pShwPageCR3RC = MMHyperCCToRC(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4534	# endif
4535
4536	# ifndef PGM_WITHOUT_MAPPINGS
4537	/*
4538	* Apply all hypervisor mappings to the new CR3.
4539	* Note that SyncCR3 will be executed in case CR3 is changed in a guest paging mode; this will
4540	* make sure we check for conflicts in the new CR3 root.
4541	*/
4542	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
4543	Assert(VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL) \|\| VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
4544	# endif
4545	rc = pgmMapActivateCR3(pVM, pNewShwPageCR3);
4546	AssertRCReturn(rc, rc);
4547	# endif
4548
4549	/* Set the current hypervisor CR3. */
4550	CPUMSetHyperCR3(pVCpu, PGMGetHyperCR3(pVCpu));
4551	SELMShadowCR3Changed(pVM, pVCpu);
4552
4553	# ifdef IN_RC
4554	/* NOTE: The state is consistent again. */
4555	VMMRZCallRing3Enable(pVCpu);
4556	# endif
4557
4558	/* Clean up the old CR3 root. */
4559	if ( pOldShwPageCR3
4560	&& pOldShwPageCR3 != pNewShwPageCR3 /* @todo can happen due to incorrect syncing between REM & PGM; find the real cause */)
4561	{
4562	Assert(pOldShwPageCR3->enmKind != PGMPOOLKIND_FREE);
4563	# ifndef PGM_WITHOUT_MAPPINGS
4564	/* Remove the hypervisor mappings from the shadow page table. */
4565	pgmMapDeactivateCR3(pVM, pOldShwPageCR3);
4566	# endif
4567	/* Mark the page as unlocked; allow flushing again. */
4568	pgmPoolUnlockPage(pPool, pOldShwPageCR3);
4569
4570	pgmPoolFreeByPage(pPool, pOldShwPageCR3, iOldShwUser, iOldShwUserTable);
4571	}
4572	pgmUnlock(pVM);
4573	# endif
4574
4575	return rc;
4576	}
4577
4578	/**
4579	* Unmaps the shadow CR3.
4580	*
4581	* @returns VBox status, no specials.
4582	* @param pVCpu The VMCPU handle.
4583	*/
4584	PGM_BTH_DECL(int, UnmapCR3)(PVMCPU pVCpu)
4585	{
4586	LogFlow(("UnmapCR3\n"));
4587
4588	int rc = VINF_SUCCESS;
4589	PVM pVM = pVCpu->CTX_SUFF(pVM);
4590
4591	/*
4592	* Update guest paging info.
4593	*/
4594	#if PGM_GST_TYPE == PGM_TYPE_32BIT
4595	pVCpu->pgm.s.pGst32BitPdR3 = 0;
4596	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4597	pVCpu->pgm.s.pGst32BitPdR0 = 0;
4598	# endif
4599	pVCpu->pgm.s.pGst32BitPdRC = 0;
4600
4601	#elif PGM_GST_TYPE == PGM_TYPE_PAE
4602	pVCpu->pgm.s.pGstPaePdptR3 = 0;
4603	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4604	pVCpu->pgm.s.pGstPaePdptR0 = 0;
4605	# endif
4606	pVCpu->pgm.s.pGstPaePdptRC = 0;
4607	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
4608	{
4609	pVCpu->pgm.s.apGstPaePDsR3[i] = 0;
4610	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4611	pVCpu->pgm.s.apGstPaePDsR0[i] = 0;
4612	# endif
4613	pVCpu->pgm.s.apGstPaePDsRC[i] = 0;
4614	pVCpu->pgm.s.aGCPhysGstPaePDs[i] = NIL_RTGCPHYS;
4615	}
4616
4617	#elif PGM_GST_TYPE == PGM_TYPE_AMD64
4618	pVCpu->pgm.s.pGstAmd64Pml4R3 = 0;
4619	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4620	pVCpu->pgm.s.pGstAmd64Pml4R0 = 0;
4621	# endif
4622
4623	#else /* prot/real mode stub */
4624	/* nothing to do */
4625	#endif
4626
4627	#if !defined(IN_RC) /* In RC we rely on MapCR3 to do the shadow part for us at a safe time */
4628	/*
4629	* Update shadow paging info.
4630	*/
4631	# if ( ( PGM_SHW_TYPE == PGM_TYPE_32BIT \
4632	\|\| PGM_SHW_TYPE == PGM_TYPE_PAE \
4633	\|\| PGM_SHW_TYPE == PGM_TYPE_AMD64))
4634
4635	# if PGM_GST_TYPE != PGM_TYPE_REAL
4636	Assert(!HWACCMIsNestedPagingActive(pVM));
4637	# endif
4638
4639	pgmLock(pVM);
4640
4641	# ifndef PGM_WITHOUT_MAPPINGS
4642	if (pVCpu->pgm.s.CTX_SUFF(pShwPageCR3))
4643	/* Remove the hypervisor mappings from the shadow page table. */
4644	pgmMapDeactivateCR3(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4645	# endif
4646
4647	if (pVCpu->pgm.s.CTX_SUFF(pShwPageCR3))
4648	{
4649	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4650
4651	Assert(pVCpu->pgm.s.iShwUser != PGMPOOL_IDX_NESTED_ROOT);
4652
4653	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
4654	if (pPool->cDirtyPages)
4655	pgmPoolResetDirtyPages(pVM);
4656	# endif
4657
4658	/* Mark the page as unlocked; allow flushing again. */
4659	pgmPoolUnlockPage(pPool, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4660
4661	pgmPoolFreeByPage(pPool, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3), pVCpu->pgm.s.iShwUser, pVCpu->pgm.s.iShwUserTable);
4662	pVCpu->pgm.s.pShwPageCR3R3 = 0;
4663	pVCpu->pgm.s.pShwPageCR3R0 = 0;
4664	pVCpu->pgm.s.pShwPageCR3RC = 0;
4665	pVCpu->pgm.s.iShwUser = 0;
4666	pVCpu->pgm.s.iShwUserTable = 0;
4667	}
4668	pgmUnlock(pVM);
4669	# endif
4670	#endif /* !IN_RC*/
4671
4672	return rc;
4673	}

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllBth.h@ 29888

Download in other formats: