PGMAllBth.h@ 46422

Last change on this file since 46422 was 46420, checked in by vboxsync, 12 years ago
VMM, recompiler: Purge deprecated macros.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 209.1 KB

Line
1	/* $Id: PGMAllBth.h 46420 2013-06-06 16:27:25Z vboxsync $ */
2	/** @file
3	* VBox - Page Manager, Shadow+Guest Paging Template - All context code.
4	*
5	* @remarks The nested page tables on AMD makes use of PGM_SHW_TYPE in
6	* {PGM_TYPE_AMD64, PGM_TYPE_PAE and PGM_TYPE_32BIT} and PGM_GST_TYPE
7	* set to PGM_TYPE_PROT. Half of the code in this file is not
8	* exercised with PGM_SHW_TYPE set to PGM_TYPE_NESTED.
9	*
10	* @remarks Extended page tables (intel) are built with PGM_GST_TYPE set to
11	* PGM_TYPE_PROT (and PGM_SHW_TYPE set to PGM_TYPE_EPT).
12	*
13	* @remarks This file is one big \#ifdef-orgy!
14	*
15	*/
16
17	/*
18	* Copyright (C) 2006-2013 Oracle Corporation
19	*
20	* This file is part of VirtualBox Open Source Edition (OSE), as
21	* available from http://www.virtualbox.org. This file is free software;
22	* you can redistribute it and/or modify it under the terms of the GNU
23	* General Public License (GPL) as published by the Free Software
24	* Foundation, in version 2 as it comes in the "COPYING" file of the
25	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
26	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
27	*/
28
29	#ifdef _MSC_VER
30	/** @todo we're generating unnecessary code in nested/ept shadow mode and for
31	* real/prot-guest+RC mode. */
32	# pragma warning(disable: 4505)
33	#endif
34
35	/*******************************************************************************
36	* Internal Functions *
37	*******************************************************************************/
38	RT_C_DECLS_BEGIN
39	PGM_BTH_DECL(int, Trap0eHandler)(PVMCPU pVCpu, RTGCUINT uErr, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, bool *pfLockTaken);
40	PGM_BTH_DECL(int, InvalidatePage)(PVMCPU pVCpu, RTGCPTR GCPtrPage);
41	static int PGM_BTH_NAME(SyncPage)(PVMCPU pVCpu, GSTPDE PdeSrc, RTGCPTR GCPtrPage, unsigned cPages, unsigned uErr);
42	static int PGM_BTH_NAME(CheckDirtyPageFault)(PVMCPU pVCpu, uint32_t uErr, PSHWPDE pPdeDst, GSTPDE const *pPdeSrc, RTGCPTR GCPtrPage);
43	static int PGM_BTH_NAME(SyncPT)(PVMCPU pVCpu, unsigned iPD, PGSTPD pPDSrc, RTGCPTR GCPtrPage);
44	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
45	static void PGM_BTH_NAME(SyncPageWorker)(PVMCPU pVCpu, PSHWPTE pPteDst, GSTPDE PdeSrc, GSTPTE PteSrc, PPGMPOOLPAGE pShwPage, unsigned iPTDst);
46	# else
47	static void PGM_BTH_NAME(SyncPageWorker)(PVMCPU pVCpu, PSHWPTE pPteDst, RTGCPHYS GCPhysPage, PPGMPOOLPAGE pShwPage, unsigned iPTDst);
48	#endif
49	PGM_BTH_DECL(int, VerifyAccessSyncPage)(PVMCPU pVCpu, RTGCPTR Addr, unsigned fPage, unsigned uErr);
50	PGM_BTH_DECL(int, PrefetchPage)(PVMCPU pVCpu, RTGCPTR GCPtrPage);
51	PGM_BTH_DECL(int, SyncCR3)(PVMCPU pVCpu, uint64_t cr0, uint64_t cr3, uint64_t cr4, bool fGlobal);
52	#ifdef VBOX_STRICT
53	PGM_BTH_DECL(unsigned, AssertCR3)(PVMCPU pVCpu, uint64_t cr3, uint64_t cr4, RTGCPTR GCPtr = 0, RTGCPTR cb = ~(RTGCPTR)0);
54	#endif
55	PGM_BTH_DECL(int, MapCR3)(PVMCPU pVCpu, RTGCPHYS GCPhysCR3);
56	PGM_BTH_DECL(int, UnmapCR3)(PVMCPU pVCpu);
57	RT_C_DECLS_END
58
59
60	/*
61	* Filter out some illegal combinations of guest and shadow paging, so we can
62	* remove redundant checks inside functions.
63	*/
64	#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
65	# error "Invalid combination; PAE guest implies PAE shadow"
66	#endif
67
68	#if (PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
69	&& !(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)
70	# error "Invalid combination; real or protected mode without paging implies 32 bits or PAE shadow paging."
71	#endif
72
73	#if (PGM_GST_TYPE == PGM_TYPE_32BIT \|\| PGM_GST_TYPE == PGM_TYPE_PAE) \
74	&& !(PGM_SHW_TYPE == PGM_TYPE_32BIT \|\| PGM_SHW_TYPE == PGM_TYPE_PAE \|\| PGM_SHW_TYPE == PGM_TYPE_NESTED \|\| PGM_SHW_TYPE == PGM_TYPE_EPT)
75	# error "Invalid combination; 32 bits guest paging or PAE implies 32 bits or PAE shadow paging."
76	#endif
77
78	#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) \
79	\|\| (PGM_SHW_TYPE == PGM_TYPE_AMD64 && PGM_GST_TYPE != PGM_TYPE_AMD64 && PGM_GST_TYPE != PGM_TYPE_PROT)
80	# error "Invalid combination; AMD64 guest implies AMD64 shadow and vice versa"
81	#endif
82
83	#ifndef IN_RING3
84
85	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
86	/**
87	* Deal with a guest page fault.
88	*
89	* @returns Strict VBox status code.
90	* @retval VINF_EM_RAW_GUEST_TRAP
91	* @retval VINF_EM_RAW_EMULATE_INSTR
92	*
93	* @param pVCpu The current CPU.
94	* @param pGstWalk The guest page table walk result.
95	* @param uErr The error code.
96	*/
97	PGM_BTH_DECL(VBOXSTRICTRC, Trap0eHandlerGuestFault)(PVMCPU pVCpu, PGSTPTWALK pGstWalk, RTGCUINT uErr)
98	{
99	# if !defined(PGM_WITHOUT_MAPPINGS) && (PGM_GST_TYPE == PGM_TYPE_32BIT \|\| PGM_GST_TYPE == PGM_TYPE_PAE)
100	/*
101	* Check for write conflicts with our hypervisor mapping.
102	*
103	* If the guest happens to access a non-present page, where our hypervisor
104	* is currently mapped, then we'll create a #PF storm in the guest.
105	*/
106	if ( (uErr & (X86_TRAP_PF_P \| X86_TRAP_PF_RW)) == (X86_TRAP_PF_P \| X86_TRAP_PF_RW)
107	&& pgmMapAreMappingsEnabled(pVCpu->CTX_SUFF(pVM))
108	&& MMHyperIsInsideArea(pVCpu->CTX_SUFF(pVM), pGstWalk->Core.GCPtr))
109	{
110	/* Force a CR3 sync to check for conflicts and emulate the instruction. */
111	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
112	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2GuestTrap; });
113	return VINF_EM_RAW_EMULATE_INSTR;
114	}
115	# endif
116
117	/*
118	* Calc the error code for the guest trap.
119	*/
120	uint32_t uNewErr = GST_IS_NX_ACTIVE(pVCpu)
121	? uErr & (X86_TRAP_PF_RW \| X86_TRAP_PF_US \| X86_TRAP_PF_ID)
122	: uErr & (X86_TRAP_PF_RW \| X86_TRAP_PF_US);
123	if (pGstWalk->Core.fBadPhysAddr)
124	{
125	uNewErr \|= X86_TRAP_PF_RSVD \| X86_TRAP_PF_P;
126	Assert(!pGstWalk->Core.fNotPresent);
127	}
128	else if (!pGstWalk->Core.fNotPresent)
129	uNewErr \|= X86_TRAP_PF_P;
130	TRPMSetErrorCode(pVCpu, uNewErr);
131
132	LogFlow(("Guest trap; cr2=%RGv uErr=%RGv lvl=%d\n", pGstWalk->Core.GCPtr, uErr, pGstWalk->Core.uLevel));
133	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2GuestTrap; });
134	return VINF_EM_RAW_GUEST_TRAP;
135	}
136	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
137
138
139	/**
140	* Deal with a guest page fault.
141	*
142	* The caller has taken the PGM lock.
143	*
144	* @returns Strict VBox status code.
145	*
146	* @param pVCpu The current CPU.
147	* @param uErr The error code.
148	* @param pRegFrame The register frame.
149	* @param pvFault The fault address.
150	* @param pPage The guest page at @a pvFault.
151	* @param pGstWalk The guest page table walk result.
152	* @param pfLockTaken PGM lock taken here or not (out). This is true
153	* when we're called.
154	*/
155	static VBOXSTRICTRC PGM_BTH_NAME(Trap0eHandlerDoAccessHandlers)(PVMCPU pVCpu, RTGCUINT uErr, PCPUMCTXCORE pRegFrame,
156	RTGCPTR pvFault, PPGMPAGE pPage, bool *pfLockTaken
157	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
158	, PGSTPTWALK pGstWalk
159	# endif
160	)
161	{
162	# if !PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
163	GSTPDE const PdeSrcDummy = { X86_PDE_P \| X86_PDE_US \| X86_PDE_RW \| X86_PDE_A };
164	#endif
165	PVM pVM = pVCpu->CTX_SUFF(pVM);
166	int rc;
167
168	if (PGM_PAGE_HAS_ANY_PHYSICAL_HANDLERS(pPage))
169	{
170	/*
171	* Physical page access handler.
172	*/
173	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
174	const RTGCPHYS GCPhysFault = pGstWalk->Core.GCPhys;
175	# else
176	const RTGCPHYS GCPhysFault = PGM_A20_APPLY(pVCpu, (RTGCPHYS)pvFault);
177	# endif
178	PPGMPHYSHANDLER pCur = pgmHandlerPhysicalLookup(pVM, GCPhysFault);
179	if (pCur)
180	{
181	# ifdef PGM_SYNC_N_PAGES
182	/*
183	* If the region is write protected and we got a page not present fault, then sync
184	* the pages. If the fault was caused by a read, then restart the instruction.
185	* In case of write access continue to the GC write handler.
186	*
187	* ASSUMES that there is only one handler per page or that they have similar write properties.
188	*/
189	if ( !(uErr & X86_TRAP_PF_P)
190	&& pCur->enmType == PGMPHYSHANDLERTYPE_PHYSICAL_WRITE)
191	{
192	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
193	rc = PGM_BTH_NAME(SyncPage)(pVCpu, pGstWalk->Pde, pvFault, PGM_SYNC_NR_PAGES, uErr);
194	# else
195	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrcDummy, pvFault, PGM_SYNC_NR_PAGES, uErr);
196	# endif
197	if ( RT_FAILURE(rc)
198	\|\| !(uErr & X86_TRAP_PF_RW)
199	\|\| rc == VINF_PGM_SYNCPAGE_MODIFIED_PDE)
200	{
201	AssertRC(rc);
202	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eHandlersOutOfSync);
203	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2OutOfSyncHndPhys; });
204	return rc;
205	}
206	}
207	# endif
208	# ifdef PGM_WITH_MMIO_OPTIMIZATIONS
209	/*
210	* If the access was not thru a #PF(RSVD\|...) resync the page.
211	*/
212	if ( !(uErr & X86_TRAP_PF_RSVD)
213	&& pCur->enmType != PGMPHYSHANDLERTYPE_PHYSICAL_WRITE
214	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
215	&& pGstWalk->Core.fEffectiveRW
216	&& !pGstWalk->Core.fEffectiveUS /** @todo Remove pGstWalk->Core.fEffectiveUS and X86_PTE_US further down in the sync code. */
217	# endif
218	)
219	{
220	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
221	rc = PGM_BTH_NAME(SyncPage)(pVCpu, pGstWalk->Pde, pvFault, PGM_SYNC_NR_PAGES, uErr);
222	# else
223	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrcDummy, pvFault, PGM_SYNC_NR_PAGES, uErr);
224	# endif
225	if ( RT_FAILURE(rc)
226	\|\| rc == VINF_PGM_SYNCPAGE_MODIFIED_PDE)
227	{
228	AssertRC(rc);
229	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eHandlersOutOfSync);
230	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2OutOfSyncHndPhys; });
231	return rc;
232	}
233	}
234	# endif
235
236	AssertMsg( pCur->enmType != PGMPHYSHANDLERTYPE_PHYSICAL_WRITE
237	\|\| (pCur->enmType == PGMPHYSHANDLERTYPE_PHYSICAL_WRITE && (uErr & X86_TRAP_PF_RW)),
238	("Unexpected trap for physical handler: %08X (phys=%08x) pPage=%R[pgmpage] uErr=%X, enum=%d\n",
239	pvFault, GCPhysFault, pPage, uErr, pCur->enmType));
240	if (pCur->enmType == PGMPHYSHANDLERTYPE_PHYSICAL_WRITE)
241	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eHandlersPhysWrite);
242	else
243	{
244	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eHandlersPhysAll);
245	if (uErr & X86_TRAP_PF_RSVD) STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eHandlersPhysAllOpt);
246	}
247
248	if (pCur->CTX_SUFF(pfnHandler))
249	{
250	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
251	void *pvUser = pCur->CTX_SUFF(pvUser);
252	# ifdef IN_RING0
253	PFNPGMR0PHYSHANDLER pfnHandler = pCur->CTX_SUFF(pfnHandler);
254	# else
255	PFNPGMRCPHYSHANDLER pfnHandler = pCur->CTX_SUFF(pfnHandler);
256	# endif
257
258	STAM_PROFILE_START(&pCur->Stat, h);
259	if (pfnHandler != pPool->CTX_SUFF(pfnAccessHandler))
260	{
261	pgmUnlock(pVM);
262	*pfLockTaken = false;
263	}
264
265	rc = pfnHandler(pVM, uErr, pRegFrame, pvFault, GCPhysFault, pvUser);
266
267	# ifdef VBOX_WITH_STATISTICS
268	pgmLock(pVM);
269	pCur = pgmHandlerPhysicalLookup(pVM, GCPhysFault);
270	if (pCur)
271	STAM_PROFILE_STOP(&pCur->Stat, h);
272	pgmUnlock(pVM);
273	# endif
274	}
275	else
276	rc = VINF_EM_RAW_EMULATE_INSTR;
277
278	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2HndPhys; });
279	return rc;
280	}
281	}
282	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && !defined(IN_RING0)
283	else
284	{
285	# ifdef PGM_SYNC_N_PAGES
286	/*
287	* If the region is write protected and we got a page not present fault, then sync
288	* the pages. If the fault was caused by a read, then restart the instruction.
289	* In case of write access continue to the GC write handler.
290	*/
291	if ( PGM_PAGE_GET_HNDL_VIRT_STATE(pPage) < PGM_PAGE_HNDL_PHYS_STATE_ALL
292	&& !(uErr & X86_TRAP_PF_P))
293	{
294	rc = PGM_BTH_NAME(SyncPage)(pVCpu, pGstWalk->Pde, pvFault, PGM_SYNC_NR_PAGES, uErr);
295	if ( RT_FAILURE(rc)
296	\|\| rc == VINF_PGM_SYNCPAGE_MODIFIED_PDE
297	\|\| !(uErr & X86_TRAP_PF_RW))
298	{
299	AssertRC(rc);
300	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eHandlersOutOfSync);
301	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2OutOfSyncHndVirt; });
302	return rc;
303	}
304	}
305	# endif
306	/*
307	* Ok, it's an virtual page access handler.
308	*
309	* Since it's faster to search by address, we'll do that first
310	* and then retry by GCPhys if that fails.
311	*/
312	/** @todo r=bird: perhaps we should consider looking up by physical address directly now?
313	* r=svl: true, but lookup on virtual address should remain as a fallback as phys & virt trees might be
314	* out of sync, because the page was changed without us noticing it (not-present -> present
315	* without invlpg or mov cr3, xxx).
316	*/
317	PPGMVIRTHANDLER pCur = (PPGMVIRTHANDLER)RTAvlroGCPtrRangeGet(&pVM->pgm.s.CTX_SUFF(pTrees)->VirtHandlers, pvFault);
318	if (pCur)
319	{
320	AssertMsg(!(pvFault - pCur->Core.Key < pCur->cb)
321	\|\| ( pCur->enmType != PGMVIRTHANDLERTYPE_WRITE
322	\|\| !(uErr & X86_TRAP_PF_P)
323	\|\| (pCur->enmType == PGMVIRTHANDLERTYPE_WRITE && (uErr & X86_TRAP_PF_RW))),
324	("Unexpected trap for virtual handler: %RGv (phys=%RGp) pPage=%R[pgmpage] uErr=%X, enum=%d\n",
325	pvFault, pGstWalk->Core.GCPhys, pPage, uErr, pCur->enmType));
326
327	if ( pvFault - pCur->Core.Key < pCur->cb
328	&& ( uErr & X86_TRAP_PF_RW
329	\|\| pCur->enmType != PGMVIRTHANDLERTYPE_WRITE ) )
330	{
331	# ifdef IN_RC
332	STAM_PROFILE_START(&pCur->Stat, h);
333	RTGCPTR GCPtrStart = pCur->Core.Key;
334	CTX_MID(PFNPGM,VIRTHANDLER) pfnHandler = pCur->CTX_SUFF(pfnHandler);
335	pgmUnlock(pVM);
336	*pfLockTaken = false;
337
338	rc = pfnHandler(pVM, uErr, pRegFrame, pvFault, GCPtrStart, pvFault - GCPtrStart);
339
340	# ifdef VBOX_WITH_STATISTICS
341	pgmLock(pVM);
342	pCur = (PPGMVIRTHANDLER)RTAvlroGCPtrRangeGet(&pVM->pgm.s.CTX_SUFF(pTrees)->VirtHandlers, pvFault);
343	if (pCur)
344	STAM_PROFILE_STOP(&pCur->Stat, h);
345	pgmUnlock(pVM);
346	# endif
347	# else
348	rc = VINF_EM_RAW_EMULATE_INSTR; /** @todo for VMX */
349	# endif
350	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eHandlersVirtual);
351	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2HndVirt; });
352	return rc;
353	}
354	/* Unhandled part of a monitored page */
355	Log(("Unhandled part of monitored page %RGv\n", pvFault));
356	}
357	else
358	{
359	/* Check by physical address. */
360	unsigned iPage;
361	rc = pgmHandlerVirtualFindByPhysAddr(pVM, pGstWalk->Core.GCPhys, &pCur, &iPage);
362	Assert(RT_SUCCESS(rc) \|\| !pCur);
363	if ( pCur
364	&& ( uErr & X86_TRAP_PF_RW
365	\|\| pCur->enmType != PGMVIRTHANDLERTYPE_WRITE ) )
366	{
367	Assert((pCur->aPhysToVirt[iPage].Core.Key & X86_PTE_PAE_PG_MASK) == (pGstWalk->Core.GCPhys & X86_PTE_PAE_PG_MASK));
368	# ifdef IN_RC
369	STAM_PROFILE_START(&pCur->Stat, h);
370	RTGCPTR GCPtrStart = pCur->Core.Key;
371	CTX_MID(PFNPGM,VIRTHANDLER) pfnHandler = pCur->CTX_SUFF(pfnHandler);
372	pgmUnlock(pVM);
373	*pfLockTaken = false;
374
375	RTGCPTR off = (iPage << PAGE_SHIFT)
376	+ (pvFault & PAGE_OFFSET_MASK)
377	- (GCPtrStart & PAGE_OFFSET_MASK);
378	Assert(off < pCur->cb);
379	rc = pfnHandler(pVM, uErr, pRegFrame, pvFault, GCPtrStart, off);
380
381	# ifdef VBOX_WITH_STATISTICS
382	pgmLock(pVM);
383	pCur = (PPGMVIRTHANDLER)RTAvlroGCPtrRangeGet(&pVM->pgm.s.CTX_SUFF(pTrees)->VirtHandlers, GCPtrStart);
384	if (pCur)
385	STAM_PROFILE_STOP(&pCur->Stat, h);
386	pgmUnlock(pVM);
387	# endif
388	# else
389	rc = VINF_EM_RAW_EMULATE_INSTR; /** @todo for VMX */
390	# endif
391	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eHandlersVirtualByPhys);
392	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2HndVirt; });
393	return rc;
394	}
395	}
396	}
397	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
398
399	/*
400	* There is a handled area of the page, but this fault doesn't belong to it.
401	* We must emulate the instruction.
402	*
403	* To avoid crashing (non-fatal) in the interpreter and go back to the recompiler
404	* we first check if this was a page-not-present fault for a page with only
405	* write access handlers. Restart the instruction if it wasn't a write access.
406	*/
407	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eHandlersUnhandled);
408
409	if ( !PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage)
410	&& !(uErr & X86_TRAP_PF_P))
411	{
412	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
413	rc = PGM_BTH_NAME(SyncPage)(pVCpu, pGstWalk->Pde, pvFault, PGM_SYNC_NR_PAGES, uErr);
414	# else
415	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrcDummy, pvFault, PGM_SYNC_NR_PAGES, uErr);
416	# endif
417	if ( RT_FAILURE(rc)
418	\|\| rc == VINF_PGM_SYNCPAGE_MODIFIED_PDE
419	\|\| !(uErr & X86_TRAP_PF_RW))
420	{
421	AssertRC(rc);
422	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eHandlersOutOfSync);
423	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2OutOfSyncHndPhys; });
424	return rc;
425	}
426	}
427
428	/** @todo This particular case can cause quite a lot of overhead. E.g. early stage of kernel booting in Ubuntu 6.06
429	* It's writing to an unhandled part of the LDT page several million times.
430	*/
431	rc = VBOXSTRICTRC_TODO(PGMInterpretInstruction(pVM, pVCpu, pRegFrame, pvFault));
432	LogFlow(("PGM: PGMInterpretInstruction -> rc=%d pPage=%R[pgmpage]\n", rc, pPage));
433	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2HndUnhandled; });
434	return rc;
435	} /* if any kind of handler */
436
437
438	/**
439	* #PF Handler for raw-mode guest execution.
440	*
441	* @returns VBox status code (appropriate for trap handling and GC return).
442	*
443	* @param pVCpu Pointer to the VMCPU.
444	* @param uErr The trap error code.
445	* @param pRegFrame Trap register frame.
446	* @param pvFault The fault address.
447	* @param pfLockTaken PGM lock taken here or not (out)
448	*/
449	PGM_BTH_DECL(int, Trap0eHandler)(PVMCPU pVCpu, RTGCUINT uErr, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, bool *pfLockTaken)
450	{
451	PVM pVM = pVCpu->CTX_SUFF(pVM); NOREF(pVM);
452
453	*pfLockTaken = false;
454
455	# if ( PGM_GST_TYPE == PGM_TYPE_32BIT \|\| PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT \
456	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \|\| PGM_GST_TYPE == PGM_TYPE_AMD64) \
457	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
458	&& (PGM_SHW_TYPE != PGM_TYPE_EPT \|\| PGM_GST_TYPE == PGM_TYPE_PROT)
459	int rc;
460
461	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
462	/*
463	* Walk the guest page translation tables and check if it's a guest fault.
464	*/
465	GSTPTWALK GstWalk;
466	rc = PGM_GST_NAME(Walk)(pVCpu, pvFault, &GstWalk);
467	if (RT_FAILURE_NP(rc))
468	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(Trap0eHandlerGuestFault)(pVCpu, &GstWalk, uErr));
469
470	/* assert some GstWalk sanity. */
471	# if PGM_GST_TYPE == PGM_TYPE_AMD64
472	/AssertMsg(GstWalk.Pml4e.u == GstWalk.pPml4e->u, ("%RX64 %RX64\n", (uint64_t)GstWalk.Pml4e.u, (uint64_t)GstWalk.pPml4e->u)); - not always true with SMP guests. /
473	# endif
474	# if PGM_GST_TYPE == PGM_TYPE_AMD64 \|\| PGM_GST_TYPE == PGM_TYPE_PAE
475	/AssertMsg(GstWalk.Pdpe.u == GstWalk.pPdpe->u, ("%RX64 %RX64\n", (uint64_t)GstWalk.Pdpe.u, (uint64_t)GstWalk.pPdpe->u)); - ditto /
476	# endif
477	/AssertMsg(GstWalk.Pde.u == GstWalk.pPde->u, ("%RX64 %RX64\n", (uint64_t)GstWalk.Pde.u, (uint64_t)GstWalk.pPde->u)); - ditto /
478	/AssertMsg(GstWalk.Core.fBigPage \|\| GstWalk.Pte.u == GstWalk.pPte->u, ("%RX64 %RX64\n", (uint64_t)GstWalk.Pte.u, (uint64_t)GstWalk.pPte->u)); - ditto /
479	Assert(GstWalk.Core.fSucceeded);
480
481	if (uErr & (X86_TRAP_PF_RW \| X86_TRAP_PF_US \| X86_TRAP_PF_ID))
482	{
483	if ( ( (uErr & X86_TRAP_PF_RW)
484	&& !GstWalk.Core.fEffectiveRW
485	&& ( (uErr & X86_TRAP_PF_US)
486	\|\| CPUMIsGuestR0WriteProtEnabled(pVCpu)) )
487	\|\| ((uErr & X86_TRAP_PF_US) && !GstWalk.Core.fEffectiveUS)
488	\|\| ((uErr & X86_TRAP_PF_ID) && GstWalk.Core.fEffectiveNX)
489	)
490	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(Trap0eHandlerGuestFault)(pVCpu, &GstWalk, uErr));
491	}
492
493	/*
494	* Set the accessed and dirty flags.
495	*/
496	# if PGM_GST_TYPE == PGM_TYPE_AMD64
497	GstWalk.Pml4e.u \|= X86_PML4E_A;
498	GstWalk.pPml4e->u \|= X86_PML4E_A;
499	GstWalk.Pdpe.u \|= X86_PDPE_A;
500	GstWalk.pPdpe->u \|= X86_PDPE_A;
501	# endif
502	if (GstWalk.Core.fBigPage)
503	{
504	Assert(GstWalk.Pde.b.u1Size);
505	if (uErr & X86_TRAP_PF_RW)
506	{
507	GstWalk.Pde.u \|= X86_PDE4M_A \| X86_PDE4M_D;
508	GstWalk.pPde->u \|= X86_PDE4M_A \| X86_PDE4M_D;
509	}
510	else
511	{
512	GstWalk.Pde.u \|= X86_PDE4M_A;
513	GstWalk.pPde->u \|= X86_PDE4M_A;
514	}
515	}
516	else
517	{
518	Assert(!GstWalk.Pde.b.u1Size);
519	GstWalk.Pde.u \|= X86_PDE_A;
520	GstWalk.pPde->u \|= X86_PDE_A;
521	if (uErr & X86_TRAP_PF_RW)
522	{
523	# ifdef VBOX_WITH_STATISTICS
524	if (!GstWalk.Pte.n.u1Dirty)
525	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtiedPage));
526	else
527	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,PageAlreadyDirty));
528	# endif
529	GstWalk.Pte.u \|= X86_PTE_A \| X86_PTE_D;
530	GstWalk.pPte->u \|= X86_PTE_A \| X86_PTE_D;
531	}
532	else
533	{
534	GstWalk.Pte.u \|= X86_PTE_A;
535	GstWalk.pPte->u \|= X86_PTE_A;
536	}
537	Assert(GstWalk.Pte.u == GstWalk.pPte->u);
538	}
539	AssertMsg(GstWalk.Pde.u == GstWalk.pPde->u \|\| GstWalk.pPte->u == GstWalk.pPde->u,
540	("%RX64 %RX64 pPte=%p pPde=%p Pte=%RX64\n", (uint64_t)GstWalk.Pde.u, (uint64_t)GstWalk.pPde->u, GstWalk.pPte, GstWalk.pPde, (uint64_t)GstWalk.pPte->u));
541	# else /* !PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
542	GSTPDE const PdeSrcDummy = { X86_PDE_P \| X86_PDE_US \| X86_PDE_RW \| X86_PDE_A}; /** @todo eliminate this */
543	# endif /* !PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
544
545	/* Take the big lock now. */
546	*pfLockTaken = true;
547	pgmLock(pVM);
548
549	# ifdef PGM_WITH_MMIO_OPTIMIZATIONS
550	/*
551	* If it is a reserved bit fault we know that it is an MMIO (access
552	* handler) related fault and can skip some 200 lines of code.
553	*/
554	if (uErr & X86_TRAP_PF_RSVD)
555	{
556	Assert(uErr & X86_TRAP_PF_P);
557	PPGMPAGE pPage;
558	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
559	rc = pgmPhysGetPageEx(pVM, GstWalk.Core.GCPhys, &pPage);
560	if (RT_SUCCESS(rc) && PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage))
561	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(Trap0eHandlerDoAccessHandlers)(pVCpu, uErr, pRegFrame, pvFault, pPage,
562	pfLockTaken, &GstWalk));
563	rc = PGM_BTH_NAME(SyncPage)(pVCpu, GstWalk.Pde, pvFault, 1, uErr);
564	# else
565	rc = pgmPhysGetPageEx(pVM, PGM_A20_APPLY(pVCpu, (RTGCPHYS)pvFault), &pPage);
566	if (RT_SUCCESS(rc) && PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage))
567	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(Trap0eHandlerDoAccessHandlers)(pVCpu, uErr, pRegFrame, pvFault, pPage,
568	pfLockTaken));
569	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrcDummy, pvFault, 1, uErr);
570	# endif
571	AssertRC(rc);
572	PGM_INVL_PG(pVCpu, pvFault);
573	return rc; /* Restart with the corrected entry. */
574	}
575	# endif /* PGM_WITH_MMIO_OPTIMIZATIONS */
576
577	/*
578	* Fetch the guest PDE, PDPE and PML4E.
579	*/
580	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
581	const unsigned iPDDst = pvFault >> SHW_PD_SHIFT;
582	PX86PD pPDDst = pgmShwGet32BitPDPtr(pVCpu);
583
584	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
585	const unsigned iPDDst = (pvFault >> SHW_PD_SHIFT) & SHW_PD_MASK; /* pPDDst index, not used with the pool. */
586	PX86PDPAE pPDDst;
587	# if PGM_GST_TYPE == PGM_TYPE_PAE
588	rc = pgmShwSyncPaePDPtr(pVCpu, pvFault, GstWalk.Pdpe.u, &pPDDst);
589	# else
590	rc = pgmShwSyncPaePDPtr(pVCpu, pvFault, X86_PDPE_P, &pPDDst); /* RW, US and A are reserved in PAE mode. */
591	# endif
592	AssertMsgReturn(rc == VINF_SUCCESS, ("rc=%Rrc\n", rc), RT_FAILURE_NP(rc) ? rc : VERR_IPE_UNEXPECTED_INFO_STATUS);
593
594	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
595	const unsigned iPDDst = ((pvFault >> SHW_PD_SHIFT) & SHW_PD_MASK);
596	PX86PDPAE pPDDst;
597	# if PGM_GST_TYPE == PGM_TYPE_PROT /* (AMD-V nested paging) */
598	rc = pgmShwSyncLongModePDPtr(pVCpu, pvFault, X86_PML4E_P \| X86_PML4E_RW \| X86_PML4E_US \| X86_PML4E_A,
599	X86_PDPE_P \| X86_PDPE_RW \| X86_PDPE_US \| X86_PDPE_A, &pPDDst);
600	# else
601	rc = pgmShwSyncLongModePDPtr(pVCpu, pvFault, GstWalk.Pml4e.u, GstWalk.Pdpe.u, &pPDDst);
602	# endif
603	AssertMsgReturn(rc == VINF_SUCCESS, ("rc=%Rrc\n", rc), RT_FAILURE_NP(rc) ? rc : VERR_IPE_UNEXPECTED_INFO_STATUS);
604
605	# elif PGM_SHW_TYPE == PGM_TYPE_EPT
606	const unsigned iPDDst = ((pvFault >> SHW_PD_SHIFT) & SHW_PD_MASK);
607	PEPTPD pPDDst;
608	rc = pgmShwGetEPTPDPtr(pVCpu, pvFault, NULL, &pPDDst);
609	AssertMsgReturn(rc == VINF_SUCCESS, ("rc=%Rrc\n", rc), RT_FAILURE_NP(rc) ? rc : VERR_IPE_UNEXPECTED_INFO_STATUS);
610	# endif
611	Assert(pPDDst);
612
613	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
614	/*
615	* Dirty page handling.
616	*
617	* If we successfully correct the write protection fault due to dirty bit
618	* tracking, then return immediately.
619	*/
620	if (uErr & X86_TRAP_PF_RW) /* write fault? */
621	{
622	STAM_PROFILE_START(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyBitTracking), a);
623	rc = PGM_BTH_NAME(CheckDirtyPageFault)(pVCpu, uErr, &pPDDst->a[iPDDst], GstWalk.pPde, pvFault);
624	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyBitTracking), a);
625	if (rc == VINF_PGM_HANDLED_DIRTY_BIT_FAULT)
626	{
627	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution)
628	= rc == VINF_PGM_HANDLED_DIRTY_BIT_FAULT
629	? &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2DirtyAndAccessed
630	: &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2GuestTrap; });
631	LogBird(("Trap0eHandler: returns VINF_SUCCESS\n"));
632	return VINF_SUCCESS;
633	}
634	//AssertMsg(GstWalk.Pde.u == GstWalk.pPde->u \|\| GstWalk.pPte->u == GstWalk.pPde->u, ("%RX64 %RX64\n", (uint64_t)GstWalk.Pde.u, (uint64_t)GstWalk.pPde->u)); - triggers with smp w7 guests.
635	//AssertMsg(GstWalk.Core.fBigPage \|\| GstWalk.Pte.u == GstWalk.pPte->u, ("%RX64 %RX64\n", (uint64_t)GstWalk.Pte.u, (uint64_t)GstWalk.pPte->u)); - ditto.
636	}
637
638	# if 0 /* rarely useful; leave for debugging. */
639	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0ePD[iPDSrc]);
640	# endif
641	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
642
643	/*
644	* A common case is the not-present error caused by lazy page table syncing.
645	*
646	* It is IMPORTANT that we weed out any access to non-present shadow PDEs
647	* here so we can safely assume that the shadow PT is present when calling
648	* SyncPage later.
649	*
650	* On failure, we ASSUME that SyncPT is out of memory or detected some kind
651	* of mapping conflict and defer to SyncCR3 in R3.
652	* (Again, we do NOT support access handlers for non-present guest pages.)
653	*
654	*/
655	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
656	Assert(GstWalk.Pde.n.u1Present);
657	# endif
658	if ( !(uErr & X86_TRAP_PF_P) /* not set means page not present instead of page protection violation */
659	&& !pPDDst->a[iPDDst].n.u1Present)
660	{
661	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2SyncPT; });
662	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
663	LogFlow(("=>SyncPT %04x = %08RX64\n", (pvFault >> GST_PD_SHIFT) & GST_PD_MASK, (uint64_t)GstWalk.Pde.u));
664	rc = PGM_BTH_NAME(SyncPT)(pVCpu, (pvFault >> GST_PD_SHIFT) & GST_PD_MASK, GstWalk.pPd, pvFault);
665	# else
666	LogFlow(("=>SyncPT pvFault=%RGv\n", pvFault));
667	rc = PGM_BTH_NAME(SyncPT)(pVCpu, 0, NULL, pvFault);
668	# endif
669	if (RT_SUCCESS(rc))
670	return rc;
671	Log(("SyncPT: %RGv failed!! rc=%Rrc\n", pvFault, rc));
672	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
673	return VINF_PGM_SYNC_CR3;
674	}
675
676	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && !defined(PGM_WITHOUT_MAPPINGS)
677	/*
678	* Check if this address is within any of our mappings.
679	*
680	* This is very fast and it's gonna save us a bit of effort below and prevent
681	* us from screwing ourself with MMIO2 pages which have a GC Mapping (VRam).
682	* (BTW, it's impossible to have physical access handlers in a mapping.)
683	*/
684	if (pgmMapAreMappingsEnabled(pVM))
685	{
686	PPGMMAPPING pMapping = pVM->pgm.s.CTX_SUFF(pMappings);
687	for ( ; pMapping; pMapping = pMapping->CTX_SUFF(pNext))
688	{
689	if (pvFault < pMapping->GCPtr)
690	break;
691	if (pvFault - pMapping->GCPtr < pMapping->cb)
692	{
693	/*
694	* The first thing we check is if we've got an undetected conflict.
695	*/
696	if (pgmMapAreMappingsFloating(pVM))
697	{
698	unsigned iPT = pMapping->cb >> GST_PD_SHIFT;
699	while (iPT-- > 0)
700	if (GstWalk.pPde[iPT].n.u1Present)
701	{
702	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eConflicts);
703	Log(("Trap0e: Detected Conflict %RGv-%RGv\n", pMapping->GCPtr, pMapping->GCPtrLast));
704	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3); /** @todo no need to do global sync,right? */
705	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2Mapping; });
706	return VINF_PGM_SYNC_CR3;
707	}
708	}
709
710	/*
711	* Check if the fault address is in a virtual page access handler range.
712	*/
713	PPGMVIRTHANDLER pCur = (PPGMVIRTHANDLER)RTAvlroGCPtrRangeGet(&pVM->pgm.s.CTX_SUFF(pTrees)->HyperVirtHandlers, pvFault);
714	if ( pCur
715	&& pvFault - pCur->Core.Key < pCur->cb
716	&& uErr & X86_TRAP_PF_RW)
717	{
718	# ifdef IN_RC
719	STAM_PROFILE_START(&pCur->Stat, h);
720	pgmUnlock(pVM);
721	rc = pCur->CTX_SUFF(pfnHandler)(pVM, uErr, pRegFrame, pvFault, pCur->Core.Key, pvFault - pCur->Core.Key);
722	pgmLock(pVM);
723	STAM_PROFILE_STOP(&pCur->Stat, h);
724	# else
725	AssertFailed();
726	rc = VINF_EM_RAW_EMULATE_INSTR; /* can't happen with VMX */
727	# endif
728	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eHandlersMapping);
729	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2Mapping; });
730	return rc;
731	}
732
733	/*
734	* Pretend we're not here and let the guest handle the trap.
735	*/
736	TRPMSetErrorCode(pVCpu, uErr & ~X86_TRAP_PF_P);
737	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eGuestPFMapping);
738	LogFlow(("PGM: Mapping access -> route trap to recompiler!\n"));
739	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2Mapping; });
740	return VINF_EM_RAW_GUEST_TRAP;
741	}
742	}
743	} /* pgmAreMappingsEnabled(&pVM->pgm.s) */
744	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
745
746	/*
747	* Check if this fault address is flagged for special treatment,
748	* which means we'll have to figure out the physical address and
749	* check flags associated with it.
750	*
751	* ASSUME that we can limit any special access handling to pages
752	* in page tables which the guest believes to be present.
753	*/
754	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
755	RTGCPHYS GCPhys = GstWalk.Core.GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK;
756	# else
757	RTGCPHYS GCPhys = PGM_A20_APPLY(pVCpu, (RTGCPHYS)pvFault & ~(RTGCPHYS)PAGE_OFFSET_MASK);
758	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
759	PPGMPAGE pPage;
760	rc = pgmPhysGetPageEx(pVM, GCPhys, &pPage);
761	if (RT_FAILURE(rc))
762	{
763	/*
764	* When the guest accesses invalid physical memory (e.g. probing
765	* of RAM or accessing a remapped MMIO range), then we'll fall
766	* back to the recompiler to emulate the instruction.
767	*/
768	LogFlow(("PGM #PF: pgmPhysGetPageEx(%RGp) failed with %Rrc\n", GCPhys, rc));
769	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eHandlersInvalid);
770	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2InvalidPhys; });
771	return VINF_EM_RAW_EMULATE_INSTR;
772	}
773
774	/*
775	* Any handlers for this page?
776	*/
777	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
778	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
779	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(Trap0eHandlerDoAccessHandlers)(pVCpu, uErr, pRegFrame, pvFault, pPage, pfLockTaken,
780	&GstWalk));
781	# else
782	return VBOXSTRICTRC_TODO(PGM_BTH_NAME(Trap0eHandlerDoAccessHandlers)(pVCpu, uErr, pRegFrame, pvFault, pPage, pfLockTaken));
783	# endif
784
785	STAM_PROFILE_START(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTimeOutOfSync, c);
786
787	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && !defined(IN_RING0)
788	if (uErr & X86_TRAP_PF_P)
789	{
790	/*
791	* The page isn't marked, but it might still be monitored by a virtual page access handler.
792	* (ASSUMES no temporary disabling of virtual handlers.)
793	*/
794	/** @todo r=bird: Since the purpose is to catch out of sync pages with virtual handler(s) here,
795	* we should correct both the shadow page table and physical memory flags, and not only check for
796	* accesses within the handler region but for access to pages with virtual handlers. */
797	PPGMVIRTHANDLER pCur = (PPGMVIRTHANDLER)RTAvlroGCPtrRangeGet(&pVM->pgm.s.CTX_SUFF(pTrees)->VirtHandlers, pvFault);
798	if (pCur)
799	{
800	AssertMsg( !(pvFault - pCur->Core.Key < pCur->cb)
801	\|\| ( pCur->enmType != PGMVIRTHANDLERTYPE_WRITE
802	\|\| !(uErr & X86_TRAP_PF_P)
803	\|\| (pCur->enmType == PGMVIRTHANDLERTYPE_WRITE && (uErr & X86_TRAP_PF_RW))),
804	("Unexpected trap for virtual handler: %08X (phys=%08x) %R[pgmpage] uErr=%X, enum=%d\n", pvFault, GCPhys, pPage, uErr, pCur->enmType));
805
806	if ( pvFault - pCur->Core.Key < pCur->cb
807	&& ( uErr & X86_TRAP_PF_RW
808	\|\| pCur->enmType != PGMVIRTHANDLERTYPE_WRITE ) )
809	{
810	# ifdef IN_RC
811	STAM_PROFILE_START(&pCur->Stat, h);
812	pgmUnlock(pVM);
813	rc = pCur->CTX_SUFF(pfnHandler)(pVM, uErr, pRegFrame, pvFault, pCur->Core.Key, pvFault - pCur->Core.Key);
814	pgmLock(pVM);
815	STAM_PROFILE_STOP(&pCur->Stat, h);
816	# else
817	rc = VINF_EM_RAW_EMULATE_INSTR; /** @todo for VMX */
818	# endif
819	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2HndVirt; });
820	return rc;
821	}
822	}
823	}
824	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
825
826	/*
827	* We are here only if page is present in Guest page tables and
828	* trap is not handled by our handlers.
829	*
830	* Check it for page out-of-sync situation.
831	*/
832	if (!(uErr & X86_TRAP_PF_P))
833	{
834	/*
835	* Page is not present in our page tables. Try to sync it!
836	*/
837	if (uErr & X86_TRAP_PF_US)
838	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,PageOutOfSyncUser));
839	else /* supervisor */
840	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,PageOutOfSyncSupervisor));
841
842	if (PGM_PAGE_IS_BALLOONED(pPage))
843	{
844	/* Emulate reads from ballooned pages as they are not present in
845	our shadow page tables. (Required for e.g. Solaris guests; soft
846	ecc, random nr generator.) */
847	rc = VBOXSTRICTRC_TODO(PGMInterpretInstruction(pVM, pVCpu, pRegFrame, pvFault));
848	LogFlow(("PGM: PGMInterpretInstruction balloon -> rc=%d pPage=%R[pgmpage]\n", rc, pPage));
849	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,PageOutOfSyncBallloon));
850	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2Ballooned; });
851	return rc;
852	}
853
854	# if defined(LOG_ENABLED) && !defined(IN_RING0)
855	RTGCPHYS GCPhys2;
856	uint64_t fPageGst2;
857	PGMGstGetPage(pVCpu, pvFault, &fPageGst2, &GCPhys2);
858	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
859	Log(("Page out of sync: %RGv eip=%08x PdeSrc.US=%d fPageGst2=%08llx GCPhys2=%RGp scan=%d\n",
860	pvFault, pRegFrame->eip, GstWalk.Pde.n.u1User, fPageGst2, GCPhys2, CSAMDoesPageNeedScanning(pVM, pRegFrame->eip)));
861	# else
862	Log(("Page out of sync: %RGv eip=%08x fPageGst2=%08llx GCPhys2=%RGp scan=%d\n",
863	pvFault, pRegFrame->eip, fPageGst2, GCPhys2, CSAMDoesPageNeedScanning(pVM, pRegFrame->eip)));
864	# endif
865	# endif /* LOG_ENABLED */
866
867	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && !defined(IN_RING0)
868	if ( !GstWalk.Core.fEffectiveUS
869	&& CSAMIsEnabled(pVM)
870	&& CPUMGetGuestCPL(pVCpu) == 0)
871	{
872	/* Note: Can't check for X86_TRAP_ID bit, because that requires execute disable support on the CPU. */
873	if ( pvFault == (RTGCPTR)pRegFrame->eip
874	\|\| pvFault - pRegFrame->eip < 8 /* instruction crossing a page boundary */
875	# ifdef CSAM_DETECT_NEW_CODE_PAGES
876	\|\| ( !PATMIsPatchGCAddr(pVM, pRegFrame->eip)
877	&& CSAMDoesPageNeedScanning(pVM, pRegFrame->eip)) /* any new code we encounter here */
878	# endif /* CSAM_DETECT_NEW_CODE_PAGES */
879	)
880	{
881	LogFlow(("CSAMExecFault %RX32\n", pRegFrame->eip));
882	rc = CSAMExecFault(pVM, (RTRCPTR)pRegFrame->eip);
883	if (rc != VINF_SUCCESS)
884	{
885	/*
886	* CSAM needs to perform a job in ring 3.
887	*
888	* Sync the page before going to the host context; otherwise we'll end up in a loop if
889	* CSAM fails (e.g. instruction crosses a page boundary and the next page is not present)
890	*/
891	LogFlow(("CSAM ring 3 job\n"));
892	int rc2 = PGM_BTH_NAME(SyncPage)(pVCpu, GstWalk.Pde, pvFault, 1, uErr);
893	AssertRC(rc2);
894
895	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2CSAM; });
896	return rc;
897	}
898	}
899	# ifdef CSAM_DETECT_NEW_CODE_PAGES
900	else if ( uErr == X86_TRAP_PF_RW
901	&& pRegFrame->ecx >= 0x100 /* early check for movswd count */
902	&& pRegFrame->ecx < 0x10000)
903	{
904	/* In case of a write to a non-present supervisor shadow page, we'll take special precautions
905	* to detect loading of new code pages.
906	*/
907
908	/*
909	* Decode the instruction.
910	*/
911	PDISCPUSTATE pDis = &pVCpu->pgm.s.DisState;
912	uint32_t cbOp;
913	rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp);
914
915	/* For now we'll restrict this to rep movsw/d instructions */
916	if ( rc == VINF_SUCCESS
917	&& pDis->pCurInstr->opcode == OP_MOVSWD
918	&& (pDis->prefix & DISPREFIX_REP))
919	{
920	CSAMMarkPossibleCodePage(pVM, pvFault);
921	}
922	}
923	# endif /* CSAM_DETECT_NEW_CODE_PAGES */
924
925	/*
926	* Mark this page as safe.
927	*/
928	/** @todo not correct for pages that contain both code and data!! */
929	Log2(("CSAMMarkPage %RGv; scanned=%d\n", pvFault, true));
930	CSAMMarkPage(pVM, pvFault, true);
931	}
932	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && !defined(IN_RING0) */
933	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
934	rc = PGM_BTH_NAME(SyncPage)(pVCpu, GstWalk.Pde, pvFault, PGM_SYNC_NR_PAGES, uErr);
935	# else
936	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrcDummy, pvFault, PGM_SYNC_NR_PAGES, uErr);
937	# endif
938	if (RT_SUCCESS(rc))
939	{
940	/* The page was successfully synced, return to the guest. */
941	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2OutOfSync; });
942	return VINF_SUCCESS;
943	}
944	}
945	else /* uErr & X86_TRAP_PF_P: */
946	{
947	/*
948	* Write protected pages are made writable when the guest makes the
949	* first write to it. This happens for pages that are shared, write
950	* monitored or not yet allocated.
951	*
952	* We may also end up here when CR0.WP=0 in the guest.
953	*
954	* Also, a side effect of not flushing global PDEs are out of sync
955	* pages due to physical monitored regions, that are no longer valid.
956	* Assume for now it only applies to the read/write flag.
957	*/
958	if (uErr & X86_TRAP_PF_RW)
959	{
960	/*
961	* Check if it is a read-only page.
962	*/
963	if (PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
964	{
965	Log(("PGM #PF: Make writable: %RGp %R[pgmpage] pvFault=%RGp uErr=%#x\n", GCPhys, pPage, pvFault, uErr));
966	Assert(!PGM_PAGE_IS_ZERO(pPage));
967	AssertFatalMsg(!PGM_PAGE_IS_BALLOONED(pPage), ("Unexpected ballooned page at %RGp\n", GCPhys));
968	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2MakeWritable; });
969
970	rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhys);
971	if (rc != VINF_SUCCESS)
972	{
973	AssertMsg(rc == VINF_PGM_SYNC_CR3 \|\| RT_FAILURE(rc), ("%Rrc\n", rc));
974	return rc;
975	}
976	if (RT_UNLIKELY(VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY)))
977	return VINF_EM_NO_MEMORY;
978	}
979
980	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
981	/*
982	* Check to see if we need to emulate the instruction if CR0.WP=0.
983	*/
984	if ( !GstWalk.Core.fEffectiveRW
985	&& (CPUMGetGuestCR0(pVCpu) & (X86_CR0_WP \| X86_CR0_PG)) == X86_CR0_PG
986	&& CPUMGetGuestCPL(pVCpu) < 3)
987	{
988	Assert((uErr & (X86_TRAP_PF_RW \| X86_TRAP_PF_P)) == (X86_TRAP_PF_RW \| X86_TRAP_PF_P));
989
990	/*
991	* The Netware WP0+RO+US hack.
992	*
993	* Netware sometimes(/always?) runs with WP0. It has been observed doing
994	* excessive write accesses to pages which are mapped with US=1 and RW=0
995	* while WP=0. This causes a lot of exits and extremely slow execution.
996	* To avoid trapping and emulating every write here, we change the shadow
997	* page table entry to map it as US=0 and RW=1 until user mode tries to
998	* access it again (see further below). We count these shadow page table
999	* changes so we can avoid having to clear the page pool every time the WP
1000	* bit changes to 1 (see PGMCr0WpEnabled()).
1001	*/
1002	# if (PGM_GST_TYPE == PGM_TYPE_32BIT \|\| PGM_GST_TYPE == PGM_TYPE_PAE) && 1
1003	if ( GstWalk.Core.fEffectiveUS
1004	&& !GstWalk.Core.fEffectiveRW
1005	&& (GstWalk.Core.fBigPage \|\| GstWalk.Pde.n.u1Write)
1006	&& pVM->cCpus == 1 /* Sorry, no go on SMP. Add CFGM option? */)
1007	{
1008	Log(("PGM #PF: Netware WP0+RO+US hack: pvFault=%RGp uErr=%#x (big=%d)\n", pvFault, uErr, GstWalk.Core.fBigPage));
1009	rc = pgmShwMakePageSupervisorAndWritable(pVCpu, pvFault, GstWalk.Core.fBigPage, PGM_MK_PG_IS_WRITE_FAULT);
1010	if (rc == VINF_SUCCESS \|\| rc == VINF_PGM_SYNC_CR3)
1011	{
1012	PGM_INVL_PG(pVCpu, pvFault);
1013	pVCpu->pgm.s.cNetwareWp0Hacks++;
1014	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2Wp0RoUsHack; });
1015	return rc;
1016	}
1017	AssertMsg(RT_FAILURE_NP(rc), ("%Rrc\n", rc));
1018	Log(("pgmShwMakePageSupervisorAndWritable(%RGv) failed with rc=%Rrc - ignored\n", pvFault, rc));
1019	}
1020	# endif
1021
1022	/* Interpret the access. */
1023	rc = VBOXSTRICTRC_TODO(PGMInterpretInstruction(pVM, pVCpu, pRegFrame, pvFault));
1024	Log(("PGM #PF: WP0 emulation (pvFault=%RGp uErr=%#x cpl=%d fBig=%d fEffUs=%d)\n", pvFault, uErr, CPUMGetGuestCPL(pVCpu), GstWalk.Core.fBigPage, GstWalk.Core.fEffectiveUS));
1025	if (RT_SUCCESS(rc))
1026	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eWPEmulInRZ);
1027	else
1028	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eWPEmulToR3);
1029	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2WPEmulation; });
1030	return rc;
1031	}
1032	# endif
1033	/// @todo count the above case; else
1034	if (uErr & X86_TRAP_PF_US)
1035	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,PageOutOfSyncUserWrite));
1036	else /* supervisor */
1037	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,PageOutOfSyncSupervisorWrite));
1038
1039	/*
1040	* Sync the page.
1041	*
1042	* Note: Do NOT use PGM_SYNC_NR_PAGES here. That only works if the
1043	* page is not present, which is not true in this case.
1044	*/
1045	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1046	rc = PGM_BTH_NAME(SyncPage)(pVCpu, GstWalk.Pde, pvFault, 1, uErr);
1047	# else
1048	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrcDummy, pvFault, 1, uErr);
1049	# endif
1050	if (RT_SUCCESS(rc))
1051	{
1052	/*
1053	* Page was successfully synced, return to guest but invalidate
1054	* the TLB first as the page is very likely to be in it.
1055	*/
1056	# if PGM_SHW_TYPE == PGM_TYPE_EPT
1057	HMInvalidatePhysPage(pVM, (RTGCPHYS)pvFault);
1058	# else
1059	PGM_INVL_PG(pVCpu, pvFault);
1060	# endif
1061	# ifdef VBOX_STRICT
1062	RTGCPHYS GCPhys2;
1063	uint64_t fPageGst;
1064	if (!pVM->pgm.s.fNestedPaging)
1065	{
1066	rc = PGMGstGetPage(pVCpu, pvFault, &fPageGst, &GCPhys2);
1067	AssertMsg(RT_SUCCESS(rc) && ((fPageGst & X86_PTE_RW) \|\| ((CPUMGetGuestCR0(pVCpu) & (X86_CR0_WP \| X86_CR0_PG)) == X86_CR0_PG && CPUMGetGuestCPL(pVCpu) < 3)), ("rc=%Rrc fPageGst=%RX64\n", rc, fPageGst));
1068	LogFlow(("Obsolete physical monitor page out of sync %RGv - phys %RGp flags=%08llx\n", pvFault, GCPhys2, (uint64_t)fPageGst));
1069	}
1070	uint64_t fPageShw;
1071	rc = PGMShwGetPage(pVCpu, pvFault, &fPageShw, NULL);
1072	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 */,
1073	("rc=%Rrc fPageShw=%RX64\n", rc, fPageShw));
1074	# endif /* VBOX_STRICT */
1075	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2OutOfSyncHndObs; });
1076	return VINF_SUCCESS;
1077	}
1078	}
1079	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1080	/*
1081	* Check for Netware WP0+RO+US hack from above and undo it when user
1082	* mode accesses the page again.
1083	*/
1084	else if ( GstWalk.Core.fEffectiveUS
1085	&& !GstWalk.Core.fEffectiveRW
1086	&& (GstWalk.Core.fBigPage \|\| GstWalk.Pde.n.u1Write)
1087	&& pVCpu->pgm.s.cNetwareWp0Hacks > 0
1088	&& (CPUMGetGuestCR0(pVCpu) & (X86_CR0_WP \| X86_CR0_PG)) == X86_CR0_PG
1089	&& CPUMGetGuestCPL(pVCpu) == 3
1090	&& pVM->cCpus == 1
1091	)
1092	{
1093	Log(("PGM #PF: Undo netware WP0+RO+US hack: pvFault=%RGp uErr=%#x\n", pvFault, uErr));
1094	rc = PGM_BTH_NAME(SyncPage)(pVCpu, GstWalk.Pde, pvFault, 1, uErr);
1095	if (RT_SUCCESS(rc))
1096	{
1097	PGM_INVL_PG(pVCpu, pvFault);
1098	pVCpu->pgm.s.cNetwareWp0Hacks--;
1099	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2Wp0RoUsUnhack; });
1100	return VINF_SUCCESS;
1101	}
1102	}
1103	# endif /* PGM_WITH_PAGING */
1104
1105	/** @todo else: why are we here? */
1106
1107	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && defined(VBOX_STRICT)
1108	/*
1109	* Check for VMM page flags vs. Guest page flags consistency.
1110	* Currently only for debug purposes.
1111	*/
1112	if (RT_SUCCESS(rc))
1113	{
1114	/* Get guest page flags. */
1115	uint64_t fPageGst;
1116	rc = PGMGstGetPage(pVCpu, pvFault, &fPageGst, NULL);
1117	if (RT_SUCCESS(rc))
1118	{
1119	uint64_t fPageShw;
1120	rc = PGMShwGetPage(pVCpu, pvFault, &fPageShw, NULL);
1121
1122	/*
1123	* Compare page flags.
1124	* Note: we have AVL, A, D bits desynced.
1125	*/
1126	AssertMsg( (fPageShw & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK))
1127	== (fPageGst & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK))
1128	\|\| ( pVCpu->pgm.s.cNetwareWp0Hacks > 0
1129	&& (fPageShw & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK \| X86_PTE_RW \| X86_PTE_US))
1130	== (fPageGst & ~(X86_PTE_A \| X86_PTE_D \| X86_PTE_AVL_MASK \| X86_PTE_RW \| X86_PTE_US))
1131	&& (fPageShw & (X86_PTE_RW \| X86_PTE_US)) == X86_PTE_RW
1132	&& (fPageGst & (X86_PTE_RW \| X86_PTE_US)) == X86_PTE_US),
1133	("Page flags mismatch! pvFault=%RGv uErr=%x GCPhys=%RGp fPageShw=%RX64 fPageGst=%RX64\n",
1134	pvFault, (uint32_t)uErr, GCPhys, fPageShw, fPageGst));
1135	}
1136	else
1137	AssertMsgFailed(("PGMGstGetPage rc=%Rrc\n", rc));
1138	}
1139	else
1140	AssertMsgFailed(("PGMGCGetPage rc=%Rrc\n", rc));
1141	# endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) && VBOX_STRICT */
1142	}
1143
1144
1145	/*
1146	* If we get here it is because something failed above, i.e. most like guru
1147	* meditiation time.
1148	*/
1149	LogRel(("%s: returns rc=%Rrc pvFault=%RGv uErr=%RX64 cs:rip=%04x:%08RX64\n",
1150	__PRETTY_FUNCTION__, rc, pvFault, (uint64_t)uErr, pRegFrame->cs.Sel, pRegFrame->rip));
1151	return rc;
1152
1153	# else /* Nested paging, EPT except PGM_GST_TYPE = PROT */
1154	NOREF(uErr); NOREF(pRegFrame); NOREF(pvFault);
1155	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_GST_TYPE, PGM_SHW_TYPE));
1156	return VERR_PGM_NOT_USED_IN_MODE;
1157	# endif
1158	}
1159	#endif /* !IN_RING3 */
1160
1161
1162	/**
1163	* Emulation of the invlpg instruction.
1164	*
1165	*
1166	* @returns VBox status code.
1167	*
1168	* @param pVCpu Pointer to the VMCPU.
1169	* @param GCPtrPage Page to invalidate.
1170	*
1171	* @remark ASSUMES that the guest is updating before invalidating. This order
1172	* isn't required by the CPU, so this is speculative and could cause
1173	* trouble.
1174	* @remark No TLB shootdown is done on any other VCPU as we assume that
1175	* invlpg emulation is the only reason for calling this function.
1176	* (The guest has to shoot down TLB entries on other CPUs itself)
1177	* Currently true, but keep in mind!
1178	*
1179	* @todo Clean this up! Most of it is (or should be) no longer necessary as we catch all page table accesses.
1180	* Should only be required when PGMPOOL_WITH_OPTIMIZED_DIRTY_PT is active (PAE or AMD64 (for now))
1181	*/
1182	PGM_BTH_DECL(int, InvalidatePage)(PVMCPU pVCpu, RTGCPTR GCPtrPage)
1183	{
1184	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) \
1185	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
1186	&& PGM_SHW_TYPE != PGM_TYPE_EPT
1187	int rc;
1188	PVM pVM = pVCpu->CTX_SUFF(pVM);
1189	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1190
1191	PGM_LOCK_ASSERT_OWNER(pVM);
1192
1193	LogFlow(("InvalidatePage %RGv\n", GCPtrPage));
1194
1195	/*
1196	* Get the shadow PD entry and skip out if this PD isn't present.
1197	* (Guessing that it is frequent for a shadow PDE to not be present, do this first.)
1198	*/
1199	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
1200	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1201	PX86PDE pPdeDst = pgmShwGet32BitPDEPtr(pVCpu, GCPtrPage);
1202
1203	/* Fetch the pgm pool shadow descriptor. */
1204	PPGMPOOLPAGE pShwPde = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
1205	Assert(pShwPde);
1206
1207	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
1208	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT);
1209	PX86PDPT pPdptDst = pgmShwGetPaePDPTPtr(pVCpu);
1210
1211	/* If the shadow PDPE isn't present, then skip the invalidate. */
1212	if (!pPdptDst->a[iPdpt].n.u1Present)
1213	{
1214	Assert(!(pPdptDst->a[iPdpt].u & PGM_PLXFLAGS_MAPPING));
1215	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePageSkipped));
1216	PGM_INVL_PG(pVCpu, GCPtrPage);
1217	return VINF_SUCCESS;
1218	}
1219
1220	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1221	PPGMPOOLPAGE pShwPde = NULL;
1222	PX86PDPAE pPDDst;
1223
1224	/* Fetch the pgm pool shadow descriptor. */
1225	rc = pgmShwGetPaePoolPagePD(pVCpu, GCPtrPage, &pShwPde);
1226	AssertRCSuccessReturn(rc, rc);
1227	Assert(pShwPde);
1228
1229	pPDDst = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPde);
1230	PX86PDEPAE pPdeDst = &pPDDst->a[iPDDst];
1231
1232	# else /* PGM_SHW_TYPE == PGM_TYPE_AMD64 */
1233	/* PML4 */
1234	/const unsigned iPml4 = (GCPtrPage >> X86_PML4_SHIFT) & X86_PML4_MASK;/
1235	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
1236	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1237	PX86PDPAE pPDDst;
1238	PX86PDPT pPdptDst;
1239	PX86PML4E pPml4eDst;
1240	rc = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, &pPml4eDst, &pPdptDst, &pPDDst);
1241	if (rc != VINF_SUCCESS)
1242	{
1243	AssertMsg(rc == VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT \|\| rc == VERR_PAGE_MAP_LEVEL4_NOT_PRESENT, ("Unexpected rc=%Rrc\n", rc));
1244	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePageSkipped));
1245	PGM_INVL_PG(pVCpu, GCPtrPage);
1246	return VINF_SUCCESS;
1247	}
1248	Assert(pPDDst);
1249
1250	PX86PDEPAE pPdeDst = &pPDDst->a[iPDDst];
1251	PX86PDPE pPdpeDst = &pPdptDst->a[iPdpt];
1252
1253	if (!pPdpeDst->n.u1Present)
1254	{
1255	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePageSkipped));
1256	PGM_INVL_PG(pVCpu, GCPtrPage);
1257	return VINF_SUCCESS;
1258	}
1259
1260	/* Fetch the pgm pool shadow descriptor. */
1261	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & SHW_PDPE_PG_MASK);
1262	Assert(pShwPde);
1263
1264	# endif /* PGM_SHW_TYPE == PGM_TYPE_AMD64 */
1265
1266	const SHWPDE PdeDst = *pPdeDst;
1267	if (!PdeDst.n.u1Present)
1268	{
1269	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePageSkipped));
1270	PGM_INVL_PG(pVCpu, GCPtrPage);
1271	return VINF_SUCCESS;
1272	}
1273
1274	/*
1275	* Get the guest PD entry and calc big page.
1276	*/
1277	# if PGM_GST_TYPE == PGM_TYPE_32BIT
1278	PGSTPD pPDSrc = pgmGstGet32bitPDPtr(pVCpu);
1279	const unsigned iPDSrc = GCPtrPage >> GST_PD_SHIFT;
1280	GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
1281	# else /* PGM_GST_TYPE != PGM_TYPE_32BIT */
1282	unsigned iPDSrc = 0;
1283	# if PGM_GST_TYPE == PGM_TYPE_PAE
1284	X86PDPE PdpeSrcIgn;
1285	PX86PDPAE pPDSrc = pgmGstGetPaePDPtr(pVCpu, GCPtrPage, &iPDSrc, &PdpeSrcIgn);
1286	# else /* AMD64 */
1287	PX86PML4E pPml4eSrcIgn;
1288	X86PDPE PdpeSrcIgn;
1289	PX86PDPAE pPDSrc = pgmGstGetLongModePDPtr(pVCpu, GCPtrPage, &pPml4eSrcIgn, &PdpeSrcIgn, &iPDSrc);
1290	# endif
1291	GSTPDE PdeSrc;
1292
1293	if (pPDSrc)
1294	PdeSrc = pPDSrc->a[iPDSrc];
1295	else
1296	PdeSrc.u = 0;
1297	# endif /* PGM_GST_TYPE != PGM_TYPE_32BIT */
1298	const bool fIsBigPage = PdeSrc.b.u1Size && GST_IS_PSE_ACTIVE(pVCpu);
1299
1300	# ifdef IN_RING3
1301	/*
1302	* If a CR3 Sync is pending we may ignore the invalidate page operation
1303	* depending on the kind of sync and if it's a global page or not.
1304	* This doesn't make sense in GC/R0 so we'll skip it entirely there.
1305	*/
1306	# ifdef PGM_SKIP_GLOBAL_PAGEDIRS_ON_NONGLOBAL_FLUSH
1307	if ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3)
1308	\|\| ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL)
1309	&& fIsBigPage
1310	&& PdeSrc.b.u1Global
1311	)
1312	)
1313	# else
1314	if (VM_FF_IS_PENDING(pVM, VM_FF_PGM_SYNC_CR3 \| VM_FF_PGM_SYNC_CR3_NON_GLOBAL) )
1315	# endif
1316	{
1317	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePageSkipped));
1318	return VINF_SUCCESS;
1319	}
1320	# endif /* IN_RING3 */
1321
1322	/*
1323	* Deal with the Guest PDE.
1324	*/
1325	rc = VINF_SUCCESS;
1326	if (PdeSrc.n.u1Present)
1327	{
1328	Assert( PdeSrc.n.u1User == PdeDst.n.u1User
1329	&& (PdeSrc.n.u1Write \|\| !PdeDst.n.u1Write \|\| pVCpu->pgm.s.cNetwareWp0Hacks > 0));
1330	# ifndef PGM_WITHOUT_MAPPING
1331	if (PdeDst.u & PGM_PDFLAGS_MAPPING)
1332	{
1333	/*
1334	* Conflict - Let SyncPT deal with it to avoid duplicate code.
1335	*/
1336	Assert(pgmMapAreMappingsEnabled(pVM));
1337	Assert(PGMGetGuestMode(pVCpu) <= PGMMODE_PAE);
1338	rc = PGM_BTH_NAME(SyncPT)(pVCpu, iPDSrc, pPDSrc, GCPtrPage);
1339	}
1340	else
1341	# endif /* !PGM_WITHOUT_MAPPING */
1342	if (!fIsBigPage)
1343	{
1344	/*
1345	* 4KB - page.
1346	*/
1347	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, PdeDst.u & SHW_PDE_PG_MASK);
1348	RTGCPHYS GCPhys = GST_GET_PDE_GCPHYS(PdeSrc);
1349
1350	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
1351	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
1352	GCPhys = PGM_A20_APPLY(pVCpu, GCPhys \| ((iPDDst & 1) * (PAGE_SIZE / 2)));
1353	# endif
1354	if (pShwPage->GCPhys == GCPhys)
1355	{
1356	/* Syncing it here isn't 100% safe and it's probably not worth spending time syncing it. */
1357	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
1358
1359	PGSTPT pPTSrc;
1360	rc = PGM_GCPHYS_2_PTR_V2(pVM, pVCpu, GST_GET_PDE_GCPHYS(PdeSrc), &pPTSrc);
1361	if (RT_SUCCESS(rc))
1362	{
1363	const unsigned iPTSrc = (GCPtrPage >> GST_PT_SHIFT) & GST_PT_MASK;
1364	GSTPTE PteSrc = pPTSrc->a[iPTSrc];
1365	const unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
1366	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, PteSrc, pShwPage, iPTDst);
1367	Log2(("SyncPage: 4K %RGv PteSrc:{P=%d RW=%d U=%d raw=%08llx} PteDst=%08llx %s\n",
1368	GCPtrPage, PteSrc.n.u1Present,
1369	PteSrc.n.u1Write & PdeSrc.n.u1Write,
1370	PteSrc.n.u1User & PdeSrc.n.u1User,
1371	(uint64_t)PteSrc.u,
1372	SHW_PTE_LOG64(pPTDst->a[iPTDst]),
1373	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : ""));
1374	}
1375	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePage4KBPages));
1376	PGM_INVL_PG(pVCpu, GCPtrPage);
1377	}
1378	else
1379	{
1380	/*
1381	* The page table address changed.
1382	*/
1383	LogFlow(("InvalidatePage: Out-of-sync at %RGp PdeSrc=%RX64 PdeDst=%RX64 ShwGCPhys=%RGp iPDDst=%#x\n",
1384	GCPtrPage, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u, pShwPage->GCPhys, iPDDst));
1385	pgmPoolFree(pVM, PdeDst.u & SHW_PDE_PG_MASK, pShwPde->idx, iPDDst);
1386	ASMAtomicWriteSize(pPdeDst, 0);
1387	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePagePDOutOfSync));
1388	PGM_INVL_VCPU_TLBS(pVCpu);
1389	}
1390	}
1391	else
1392	{
1393	/*
1394	* 2/4MB - page.
1395	*/
1396	/* Before freeing the page, check if anything really changed. */
1397	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, PdeDst.u & SHW_PDE_PG_MASK);
1398	RTGCPHYS GCPhys = GST_GET_BIG_PDE_GCPHYS(pVM, PdeSrc);
1399	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
1400	/* Select the right PDE as we're emulating a 4MB page directory with two 2 MB shadow PDEs.*/
1401	GCPhys = PGM_A20_APPLY(pVCpu, GCPhys \| (GCPtrPage & (1 << X86_PD_PAE_SHIFT)));
1402	# endif
1403	if ( pShwPage->GCPhys == GCPhys
1404	&& pShwPage->enmKind == BTH_PGMPOOLKIND_PT_FOR_BIG)
1405	{
1406	/* ASSUMES a the given bits are identical for 4M and normal PDEs */
1407	/** @todo This test is wrong as it cannot check the G bit!
1408	* FIXME */
1409	if ( (PdeSrc.u & (X86_PDE_P \| X86_PDE_RW \| X86_PDE_US))
1410	== (PdeDst.u & (X86_PDE_P \| X86_PDE_RW \| X86_PDE_US))
1411	&& ( PdeSrc.b.u1Dirty /** @todo rainy day: What about read-only 4M pages? not very common, but still... */
1412	\|\| (PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY)))
1413	{
1414	LogFlow(("Skipping flush for big page containing %RGv (PD=%X .u=%RX64)-> nothing has changed!\n", GCPtrPage, iPDSrc, PdeSrc.u));
1415	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePage4MBPagesSkip));
1416	return VINF_SUCCESS;
1417	}
1418	}
1419
1420	/*
1421	* Ok, the page table is present and it's been changed in the guest.
1422	* If we're in host context, we'll just mark it as not present taking the lazy approach.
1423	* We could do this for some flushes in GC too, but we need an algorithm for
1424	* deciding which 4MB pages containing code likely to be executed very soon.
1425	*/
1426	LogFlow(("InvalidatePage: Out-of-sync PD at %RGp PdeSrc=%RX64 PdeDst=%RX64\n",
1427	GCPtrPage, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
1428	pgmPoolFree(pVM, PdeDst.u & SHW_PDE_PG_MASK, pShwPde->idx, iPDDst);
1429	ASMAtomicWriteSize(pPdeDst, 0);
1430	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePage4MBPages));
1431	PGM_INVL_BIG_PG(pVCpu, GCPtrPage);
1432	}
1433	}
1434	else
1435	{
1436	/*
1437	* Page directory is not present, mark shadow PDE not present.
1438	*/
1439	if (!(PdeDst.u & PGM_PDFLAGS_MAPPING))
1440	{
1441	pgmPoolFree(pVM, PdeDst.u & SHW_PDE_PG_MASK, pShwPde->idx, iPDDst);
1442	ASMAtomicWriteSize(pPdeDst, 0);
1443	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePagePDNPs));
1444	PGM_INVL_PG(pVCpu, GCPtrPage);
1445	}
1446	else
1447	{
1448	Assert(pgmMapAreMappingsEnabled(pVM));
1449	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePagePDMappings));
1450	}
1451	}
1452	return rc;
1453
1454	#else /* guest real and protected mode */
1455	/* There's no such thing as InvalidatePage when paging is disabled, so just ignore. */
1456	NOREF(pVCpu); NOREF(GCPtrPage);
1457	return VINF_SUCCESS;
1458	#endif
1459	}
1460
1461
1462	/**
1463	* Update the tracking of shadowed pages.
1464	*
1465	* @param pVCpu Pointer to the VMCPU.
1466	* @param pShwPage The shadow page.
1467	* @param HCPhys The physical page we is being dereferenced.
1468	* @param iPte Shadow PTE index
1469	* @param GCPhysPage Guest physical address (only valid if pShwPage->fDirty is set)
1470	*/
1471	DECLINLINE(void) PGM_BTH_NAME(SyncPageWorkerTrackDeref)(PVMCPU pVCpu, PPGMPOOLPAGE pShwPage, RTHCPHYS HCPhys, uint16_t iPte,
1472	RTGCPHYS GCPhysPage)
1473	{
1474	PVM pVM = pVCpu->CTX_SUFF(pVM);
1475
1476	# if defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT) \
1477	&& PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) \
1478	&& (PGM_GST_TYPE == PGM_TYPE_PAE \|\| PGM_GST_TYPE == PGM_TYPE_AMD64 \|\| PGM_SHW_TYPE == PGM_TYPE_PAE /* pae/32bit combo */)
1479
1480	/* Use the hint we retrieved from the cached guest PT. */
1481	if (pShwPage->fDirty)
1482	{
1483	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1484
1485	Assert(pShwPage->cPresent);
1486	Assert(pPool->cPresent);
1487	pShwPage->cPresent--;
1488	pPool->cPresent--;
1489
1490	PPGMPAGE pPhysPage = pgmPhysGetPage(pVM, GCPhysPage);
1491	AssertRelease(pPhysPage);
1492	pgmTrackDerefGCPhys(pPool, pShwPage, pPhysPage, iPte);
1493	return;
1494	}
1495	# else
1496	NOREF(GCPhysPage);
1497	# endif
1498
1499	STAM_PROFILE_START(&pVM->pgm.s.CTX_SUFF(pStats)->StatTrackDeref, a);
1500	LogFlow(("SyncPageWorkerTrackDeref: Damn HCPhys=%RHp pShwPage->idx=%#x!!!\n", HCPhys, pShwPage->idx));
1501
1502	/** @todo If this turns out to be a bottle neck (very likely) two things can be done:
1503	* 1. have a medium sized HCPhys -> GCPhys TLB (hash?)
1504	* 2. write protect all shadowed pages. I.e. implement caching.
1505	*/
1506	/** @todo duplicated in the 2nd half of pgmPoolTracDerefGCPhysHint */
1507
1508	/*
1509	* Find the guest address.
1510	*/
1511	for (PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRangesX);
1512	pRam;
1513	pRam = pRam->CTX_SUFF(pNext))
1514	{
1515	unsigned iPage = pRam->cb >> PAGE_SHIFT;
1516	while (iPage-- > 0)
1517	{
1518	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
1519	{
1520	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1521
1522	Assert(pShwPage->cPresent);
1523	Assert(pPool->cPresent);
1524	pShwPage->cPresent--;
1525	pPool->cPresent--;
1526
1527	pgmTrackDerefGCPhys(pPool, pShwPage, &pRam->aPages[iPage], iPte);
1528	STAM_PROFILE_STOP(&pVM->pgm.s.CTX_SUFF(pStats)->StatTrackDeref, a);
1529	return;
1530	}
1531	}
1532	}
1533
1534	for (;;)
1535	AssertReleaseMsgFailed(("HCPhys=%RHp wasn't found!\n", HCPhys));
1536	}
1537
1538
1539	/**
1540	* Update the tracking of shadowed pages.
1541	*
1542	* @param pVCpu Pointer to the VMCPU.
1543	* @param pShwPage The shadow page.
1544	* @param u16 The top 16-bit of the pPage->HCPhys.
1545	* @param pPage Pointer to the guest page. this will be modified.
1546	* @param iPTDst The index into the shadow table.
1547	*/
1548	DECLINLINE(void) PGM_BTH_NAME(SyncPageWorkerTrackAddref)(PVMCPU pVCpu, PPGMPOOLPAGE pShwPage, uint16_t u16, PPGMPAGE pPage, const unsigned iPTDst)
1549	{
1550	PVM pVM = pVCpu->CTX_SUFF(pVM);
1551
1552	/*
1553	* Just deal with the simple first time here.
1554	*/
1555	if (!u16)
1556	{
1557	STAM_COUNTER_INC(&pVM->pgm.s.CTX_SUFF(pStats)->StatTrackVirgin);
1558	u16 = PGMPOOL_TD_MAKE(1, pShwPage->idx);
1559	/* Save the page table index. */
1560	PGM_PAGE_SET_PTE_INDEX(pVM, pPage, iPTDst);
1561	}
1562	else
1563	u16 = pgmPoolTrackPhysExtAddref(pVM, pPage, u16, pShwPage->idx, iPTDst);
1564
1565	/* write back */
1566	Log2(("SyncPageWorkerTrackAddRef: u16=%#x->%#x iPTDst=%#x\n", u16, PGM_PAGE_GET_TRACKING(pPage), iPTDst));
1567	PGM_PAGE_SET_TRACKING(pVM, pPage, u16);
1568
1569	/* update statistics. */
1570	pVM->pgm.s.CTX_SUFF(pPool)->cPresent++;
1571	pShwPage->cPresent++;
1572	if (pShwPage->iFirstPresent > iPTDst)
1573	pShwPage->iFirstPresent = iPTDst;
1574	}
1575
1576
1577	/**
1578	* Modifies a shadow PTE to account for access handlers.
1579	*
1580	* @param pVM Pointer to the VM.
1581	* @param pPage The page in question.
1582	* @param fPteSrc The shadowed flags of the source PTE. Must include the
1583	* A (accessed) bit so it can be emulated correctly.
1584	* @param pPteDst The shadow PTE (output). This is temporary storage and
1585	* does not need to be set atomically.
1586	*/
1587	DECLINLINE(void) PGM_BTH_NAME(SyncHandlerPte)(PVM pVM, PCPGMPAGE pPage, uint64_t fPteSrc, PSHWPTE pPteDst)
1588	{
1589	NOREF(pVM);
1590	/** @todo r=bird: Are we actually handling dirty and access bits for pages with access handlers correctly? No.
1591	* Update: \#PF should deal with this before or after calling the handlers. It has all the info to do the job efficiently. */
1592	if (!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage))
1593	{
1594	LogFlow(("SyncHandlerPte: monitored page (%R[pgmpage]) -> mark read-only\n", pPage));
1595	#if PGM_SHW_TYPE == PGM_TYPE_EPT
1596	pPteDst->u = PGM_PAGE_GET_HCPHYS(pPage);
1597	pPteDst->n.u1Present = 1;
1598	pPteDst->n.u1Execute = 1;
1599	pPteDst->n.u1IgnorePAT = 1;
1600	pPteDst->n.u3EMT = VMX_EPT_MEMTYPE_WB;
1601	/* PteDst.n.u1Write = 0 && PteDst.n.u1Size = 0 */
1602	#else
1603	if (fPteSrc & X86_PTE_A)
1604	{
1605	SHW_PTE_SET(*pPteDst, fPteSrc \| PGM_PAGE_GET_HCPHYS(pPage));
1606	SHW_PTE_SET_RO(*pPteDst);
1607	}
1608	else
1609	SHW_PTE_SET(*pPteDst, 0);
1610	#endif
1611	}
1612	#ifdef PGM_WITH_MMIO_OPTIMIZATIONS
1613	# if PGM_SHW_TYPE == PGM_TYPE_EPT \|\| PGM_SHW_TYPE == PGM_TYPE_PAE \|\| PGM_SHW_TYPE == PGM_TYPE_AMD64
1614	else if ( PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage)
1615	&& ( BTH_IS_NP_ACTIVE(pVM)
1616	\|\| (fPteSrc & (X86_PTE_RW \| X86_PTE_US)) == X86_PTE_RW) /** @todo Remove X86_PTE_US here and pGstWalk->Core.fEffectiveUS before the sync page test. */
1617	# if PGM_SHW_TYPE == PGM_TYPE_AMD64
1618	&& pVM->pgm.s.fLessThan52PhysicalAddressBits
1619	# endif
1620	)
1621	{
1622	LogFlow(("SyncHandlerPte: MMIO page -> invalid \n"));
1623	# if PGM_SHW_TYPE == PGM_TYPE_EPT
1624	/* 25.2.3.1: Reserved physical address bit -> EPT Misconfiguration (exit 49) */
1625	pPteDst->u = pVM->pgm.s.HCPhysInvMmioPg;
1626	/* 25.2.3.1: bits 2:0 = 010b -> EPT Misconfiguration (exit 49) */
1627	pPteDst->n.u1Present = 0;
1628	pPteDst->n.u1Write = 1;
1629	pPteDst->n.u1Execute = 0;
1630	/* 25.2.3.1: leaf && 2:0 != 0 && u3Emt in {2, 3, 7} -> EPT Misconfiguration */
1631	pPteDst->n.u3EMT = 7;
1632	# else
1633	/* Set high page frame bits that MBZ (bankers on PAE, CPU dependent on AMD64). */
1634	SHW_PTE_SET(*pPteDst, pVM->pgm.s.HCPhysInvMmioPg \| X86_PTE_PAE_MBZ_MASK_NO_NX \| X86_PTE_P);
1635	# endif
1636	}
1637	# endif
1638	#endif /* PGM_WITH_MMIO_OPTIMIZATIONS */
1639	else
1640	{
1641	LogFlow(("SyncHandlerPte: monitored page (%R[pgmpage]) -> mark not present\n", pPage));
1642	SHW_PTE_SET(*pPteDst, 0);
1643	}
1644	/** @todo count these kinds of entries. */
1645	}
1646
1647
1648	/**
1649	* Creates a 4K shadow page for a guest page.
1650	*
1651	* For 4M pages the caller must convert the PDE4M to a PTE, this includes adjusting the
1652	* physical address. The PdeSrc argument only the flags are used. No page
1653	* structured will be mapped in this function.
1654	*
1655	* @param pVCpu Pointer to the VMCPU.
1656	* @param pPteDst Destination page table entry.
1657	* @param PdeSrc Source page directory entry (i.e. Guest OS page directory entry).
1658	* Can safely assume that only the flags are being used.
1659	* @param PteSrc Source page table entry (i.e. Guest OS page table entry).
1660	* @param pShwPage Pointer to the shadow page.
1661	* @param iPTDst The index into the shadow table.
1662	*
1663	* @remark Not used for 2/4MB pages!
1664	*/
1665	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1666	static void PGM_BTH_NAME(SyncPageWorker)(PVMCPU pVCpu, PSHWPTE pPteDst, GSTPDE PdeSrc, GSTPTE PteSrc,
1667	PPGMPOOLPAGE pShwPage, unsigned iPTDst)
1668	#else
1669	static void PGM_BTH_NAME(SyncPageWorker)(PVMCPU pVCpu, PSHWPTE pPteDst, RTGCPHYS GCPhysPage, PPGMPOOLPAGE pShwPage, unsigned iPTDst)
1670	#endif
1671	{
1672	PVM pVM = pVCpu->CTX_SUFF(pVM);
1673	RTGCPHYS GCPhysOldPage = NIL_RTGCPHYS;
1674
1675	#if defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT) \
1676	&& PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) \
1677	&& (PGM_GST_TYPE == PGM_TYPE_PAE \|\| PGM_GST_TYPE == PGM_TYPE_AMD64 \|\| PGM_SHW_TYPE == PGM_TYPE_PAE /* pae/32bit combo */)
1678
1679	if (pShwPage->fDirty)
1680	{
1681	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1682	PGSTPT pGstPT;
1683
1684	/* Note that iPTDst can be used to index the guest PT even in the pae/32bit combo as we copy only half the table; see pgmPoolAddDirtyPage. */
1685	pGstPT = (PGSTPT)&pPool->aDirtyPages[pShwPage->idxDirtyEntry].aPage[0];
1686	GCPhysOldPage = GST_GET_PTE_GCPHYS(pGstPT->a[iPTDst]);
1687	pGstPT->a[iPTDst].u = PteSrc.u;
1688	}
1689	#else
1690	Assert(!pShwPage->fDirty);
1691	#endif
1692
1693	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1694	if ( PteSrc.n.u1Present
1695	&& GST_IS_PTE_VALID(pVCpu, PteSrc))
1696	#endif
1697	{
1698	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1699	RTGCPHYS GCPhysPage = GST_GET_PTE_GCPHYS(PteSrc);
1700	# endif
1701	PGM_A20_ASSERT_MASKED(pVCpu, GCPhysPage);
1702
1703	/*
1704	* Find the ram range.
1705	*/
1706	PPGMPAGE pPage;
1707	int rc = pgmPhysGetPageEx(pVM, GCPhysPage, &pPage);
1708	if (RT_SUCCESS(rc))
1709	{
1710	/* Ignore ballooned pages.
1711	Don't return errors or use a fatal assert here as part of a
1712	shadow sync range might included ballooned pages. */
1713	if (PGM_PAGE_IS_BALLOONED(pPage))
1714	{
1715	Assert(!SHW_PTE_IS_P(pPteDst)); /* @todo user tracking needs updating if this triggers. */
1716	return;
1717	}
1718
1719	#ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
1720	/* Make the page writable if necessary. */
1721	if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM
1722	&& ( PGM_PAGE_IS_ZERO(pPage)
1723	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1724	\|\| ( PteSrc.n.u1Write
1725	# else
1726	\|\| ( 1
1727	# endif
1728	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED
1729	# ifdef VBOX_WITH_REAL_WRITE_MONITORED_PAGES
1730	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_WRITE_MONITORED
1731	# endif
1732	# ifdef VBOX_WITH_PAGE_SHARING
1733	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_SHARED
1734	# endif
1735	)
1736	)
1737	)
1738	{
1739	rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhysPage);
1740	AssertRC(rc);
1741	}
1742	#endif
1743
1744	/*
1745	* Make page table entry.
1746	*/
1747	SHWPTE PteDst;
1748	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1749	uint64_t fGstShwPteFlags = GST_GET_PTE_SHW_FLAGS(pVCpu, PteSrc);
1750	# else
1751	uint64_t fGstShwPteFlags = X86_PTE_P \| X86_PTE_RW \| X86_PTE_US \| X86_PTE_A \| X86_PTE_D;
1752	# endif
1753	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
1754	PGM_BTH_NAME(SyncHandlerPte)(pVM, pPage, fGstShwPteFlags, &PteDst);
1755	else
1756	{
1757	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1758	/*
1759	* If the page or page directory entry is not marked accessed,
1760	* we mark the page not present.
1761	*/
1762	if (!PteSrc.n.u1Accessed \|\| !PdeSrc.n.u1Accessed)
1763	{
1764	LogFlow(("SyncPageWorker: page and or page directory not accessed -> mark not present\n"));
1765	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,AccessedPage));
1766	SHW_PTE_SET(PteDst, 0);
1767	}
1768	/*
1769	* If the page is not flagged as dirty and is writable, then make it read-only, so we can set the dirty bit
1770	* when the page is modified.
1771	*/
1772	else if (!PteSrc.n.u1Dirty && (PdeSrc.n.u1Write & PteSrc.n.u1Write))
1773	{
1774	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyPage));
1775	SHW_PTE_SET(PteDst,
1776	fGstShwPteFlags
1777	\| PGM_PAGE_GET_HCPHYS(pPage)
1778	\| PGM_PTFLAGS_TRACK_DIRTY);
1779	SHW_PTE_SET_RO(PteDst);
1780	}
1781	else
1782	#endif
1783	{
1784	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyPageSkipped));
1785	#if PGM_SHW_TYPE == PGM_TYPE_EPT
1786	PteDst.u = PGM_PAGE_GET_HCPHYS(pPage);
1787	PteDst.n.u1Present = 1;
1788	PteDst.n.u1Write = 1;
1789	PteDst.n.u1Execute = 1;
1790	PteDst.n.u1IgnorePAT = 1;
1791	PteDst.n.u3EMT = VMX_EPT_MEMTYPE_WB;
1792	/* PteDst.n.u1Size = 0 */
1793	#else
1794	SHW_PTE_SET(PteDst, fGstShwPteFlags \| PGM_PAGE_GET_HCPHYS(pPage));
1795	#endif
1796	}
1797
1798	/*
1799	* Make sure only allocated pages are mapped writable.
1800	*/
1801	if ( SHW_PTE_IS_P_RW(PteDst)
1802	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
1803	{
1804	/* Still applies to shared pages. */
1805	Assert(!PGM_PAGE_IS_ZERO(pPage));
1806	SHW_PTE_SET_RO(PteDst); /** @todo this isn't quite working yet. Why, isn't it? */
1807	Log3(("SyncPageWorker: write-protecting %RGp pPage=%R[pgmpage]at iPTDst=%d\n", GCPhysPage, pPage, iPTDst));
1808	}
1809	}
1810
1811	/*
1812	* Keep user track up to date.
1813	*/
1814	if (SHW_PTE_IS_P(PteDst))
1815	{
1816	if (!SHW_PTE_IS_P(*pPteDst))
1817	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPTDst);
1818	else if (SHW_PTE_GET_HCPHYS(*pPteDst) != SHW_PTE_GET_HCPHYS(PteDst))
1819	{
1820	Log2(("SyncPageWorker: deref! pPteDst=%RX64 PteDst=%RX64\n", SHW_PTE_LOG64(pPteDst), SHW_PTE_LOG64(PteDst)));
1821	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pVCpu, pShwPage, SHW_PTE_GET_HCPHYS(*pPteDst), iPTDst, GCPhysOldPage);
1822	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPTDst);
1823	}
1824	}
1825	else if (SHW_PTE_IS_P(*pPteDst))
1826	{
1827	Log2(("SyncPageWorker: deref! pPteDst=%RX64\n", SHW_PTE_LOG64(pPteDst)));
1828	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pVCpu, pShwPage, SHW_PTE_GET_HCPHYS(*pPteDst), iPTDst, GCPhysOldPage);
1829	}
1830
1831	/*
1832	* Update statistics and commit the entry.
1833	*/
1834	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1835	if (!PteSrc.n.u1Global)
1836	pShwPage->fSeenNonGlobal = true;
1837	#endif
1838	SHW_PTE_ATOMIC_SET2(*pPteDst, PteDst);
1839	return;
1840	}
1841
1842	/** @todo count these three different kinds. */
1843	Log2(("SyncPageWorker: invalid address in Pte\n"));
1844	}
1845	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
1846	else if (!PteSrc.n.u1Present)
1847	Log2(("SyncPageWorker: page not present in Pte\n"));
1848	else
1849	Log2(("SyncPageWorker: invalid Pte\n"));
1850	#endif
1851
1852	/*
1853	* The page is not present or the PTE is bad. Replace the shadow PTE by
1854	* an empty entry, making sure to keep the user tracking up to date.
1855	*/
1856	if (SHW_PTE_IS_P(*pPteDst))
1857	{
1858	Log2(("SyncPageWorker: deref! pPteDst=%RX64\n", SHW_PTE_LOG64(pPteDst)));
1859	PGM_BTH_NAME(SyncPageWorkerTrackDeref)(pVCpu, pShwPage, SHW_PTE_GET_HCPHYS(*pPteDst), iPTDst, GCPhysOldPage);
1860	}
1861	SHW_PTE_ATOMIC_SET(*pPteDst, 0);
1862	}
1863
1864
1865	/**
1866	* Syncs a guest OS page.
1867	*
1868	* There are no conflicts at this point, neither is there any need for
1869	* page table allocations.
1870	*
1871	* When called in PAE or AMD64 guest mode, the guest PDPE shall be valid.
1872	* When called in AMD64 guest mode, the guest PML4E shall be valid.
1873	*
1874	* @returns VBox status code.
1875	* @returns VINF_PGM_SYNCPAGE_MODIFIED_PDE if it modifies the PDE in any way.
1876	* @param pVCpu Pointer to the VMCPU.
1877	* @param PdeSrc Page directory entry of the guest.
1878	* @param GCPtrPage Guest context page address.
1879	* @param cPages Number of pages to sync (PGM_SYNC_N_PAGES) (default=1).
1880	* @param uErr Fault error (X86_TRAP_PF_*).
1881	*/
1882	static int PGM_BTH_NAME(SyncPage)(PVMCPU pVCpu, GSTPDE PdeSrc, RTGCPTR GCPtrPage, unsigned cPages, unsigned uErr)
1883	{
1884	PVM pVM = pVCpu->CTX_SUFF(pVM);
1885	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
1886	LogFlow(("SyncPage: GCPtrPage=%RGv cPages=%u uErr=%#x\n", GCPtrPage, cPages, uErr));
1887
1888	PGM_LOCK_ASSERT_OWNER(pVM);
1889
1890	#if ( PGM_GST_TYPE == PGM_TYPE_32BIT \
1891	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
1892	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64) \
1893	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
1894	&& PGM_SHW_TYPE != PGM_TYPE_EPT
1895
1896	/*
1897	* Assert preconditions.
1898	*/
1899	Assert(PdeSrc.n.u1Present);
1900	Assert(cPages);
1901	# if 0 /* rarely useful; leave for debugging. */
1902	STAM_COUNTER_INC(&pVCpu->pgm.s.StatSyncPagePD[(GCPtrPage >> GST_PD_SHIFT) & GST_PD_MASK]);
1903	# endif
1904
1905	/*
1906	* Get the shadow PDE, find the shadow page table in the pool.
1907	*/
1908	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
1909	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1910	PX86PDE pPdeDst = pgmShwGet32BitPDEPtr(pVCpu, GCPtrPage);
1911
1912	/* Fetch the pgm pool shadow descriptor. */
1913	PPGMPOOLPAGE pShwPde = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
1914	Assert(pShwPde);
1915
1916	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
1917	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1918	PPGMPOOLPAGE pShwPde = NULL;
1919	PX86PDPAE pPDDst;
1920
1921	/* Fetch the pgm pool shadow descriptor. */
1922	int rc2 = pgmShwGetPaePoolPagePD(pVCpu, GCPtrPage, &pShwPde);
1923	AssertRCSuccessReturn(rc2, rc2);
1924	Assert(pShwPde);
1925
1926	pPDDst = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPde);
1927	PX86PDEPAE pPdeDst = &pPDDst->a[iPDDst];
1928
1929	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
1930	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
1931	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
1932	PX86PDPAE pPDDst = NULL; /* initialized to shut up gcc */
1933	PX86PDPT pPdptDst = NULL; /* initialized to shut up gcc */
1934
1935	int rc2 = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, NULL, &pPdptDst, &pPDDst);
1936	AssertRCSuccessReturn(rc2, rc2);
1937	Assert(pPDDst && pPdptDst);
1938	PX86PDEPAE pPdeDst = &pPDDst->a[iPDDst];
1939	# endif
1940	SHWPDE PdeDst = *pPdeDst;
1941
1942	/*
1943	* - In the guest SMP case we could have blocked while another VCPU reused
1944	* this page table.
1945	* - With W7-64 we may also take this path when the A bit is cleared on
1946	* higher level tables (PDPE/PML4E). The guest does not invalidate the
1947	* relevant TLB entries. If we're write monitoring any page mapped by
1948	* the modified entry, we may end up here with a "stale" TLB entry.
1949	*/
1950	if (!PdeDst.n.u1Present)
1951	{
1952	Log(("CPU%u: SyncPage: Pde at %RGv changed behind our back? (pPdeDst=%p/%RX64) uErr=%#x\n", pVCpu->idCpu, GCPtrPage, pPdeDst, (uint64_t)PdeDst.u, (uint32_t)uErr));
1953	AssertMsg(pVM->cCpus > 1 \|\| (uErr & (X86_TRAP_PF_P \| X86_TRAP_PF_RW)) == (X86_TRAP_PF_P \| X86_TRAP_PF_RW),
1954	("Unexpected missing PDE p=%p/%RX64 uErr=%#x\n", pPdeDst, (uint64_t)PdeDst.u, (uint32_t)uErr));
1955	if (uErr & X86_TRAP_PF_P)
1956	PGM_INVL_PG(pVCpu, GCPtrPage);
1957	return VINF_SUCCESS; /* force the instruction to be executed again. */
1958	}
1959
1960	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, PdeDst.u & SHW_PDE_PG_MASK);
1961	Assert(pShwPage);
1962
1963	# if PGM_GST_TYPE == PGM_TYPE_AMD64
1964	/* Fetch the pgm pool shadow descriptor. */
1965	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & X86_PDPE_PG_MASK);
1966	Assert(pShwPde);
1967	# endif
1968
1969	/*
1970	* Check that the page is present and that the shadow PDE isn't out of sync.
1971	*/
1972	const bool fBigPage = PdeSrc.b.u1Size && GST_IS_PSE_ACTIVE(pVCpu);
1973	const bool fPdeValid = !fBigPage ? GST_IS_PDE_VALID(pVCpu, PdeSrc) : GST_IS_BIG_PDE_VALID(pVCpu, PdeSrc);
1974	RTGCPHYS GCPhys;
1975	if (!fBigPage)
1976	{
1977	GCPhys = GST_GET_PDE_GCPHYS(PdeSrc);
1978	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
1979	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
1980	GCPhys = PGM_A20_APPLY(pVCpu, GCPhys \| ((iPDDst & 1) * (PAGE_SIZE / 2)));
1981	# endif
1982	}
1983	else
1984	{
1985	GCPhys = GST_GET_BIG_PDE_GCPHYS(pVM, PdeSrc);
1986	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
1987	/* Select the right PDE as we're emulating a 4MB page directory with two 2 MB shadow PDEs.*/
1988	GCPhys = PGM_A20_APPLY(pVCpu, GCPhys \| (GCPtrPage & (1 << X86_PD_PAE_SHIFT)));
1989	# endif
1990	}
1991	/** @todo This doesn't check the G bit of 2/4MB pages. FIXME */
1992	if ( fPdeValid
1993	&& pShwPage->GCPhys == GCPhys
1994	&& PdeSrc.n.u1Present
1995	&& PdeSrc.n.u1User == PdeDst.n.u1User
1996	&& (PdeSrc.n.u1Write == PdeDst.n.u1Write \|\| !PdeDst.n.u1Write)
1997	# if PGM_WITH_NX(PGM_GST_TYPE, PGM_SHW_TYPE)
1998	&& (PdeSrc.n.u1NoExecute == PdeDst.n.u1NoExecute \|\| !GST_IS_NX_ACTIVE(pVCpu))
1999	# endif
2000	)
2001	{
2002	/*
2003	* Check that the PDE is marked accessed already.
2004	* Since we set the accessed bit before getting here on a #PF, this
2005	* check is only meant for dealing with non-#PF'ing paths.
2006	*/
2007	if (PdeSrc.n.u1Accessed)
2008	{
2009	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
2010	if (!fBigPage)
2011	{
2012	/*
2013	* 4KB Page - Map the guest page table.
2014	*/
2015	PGSTPT pPTSrc;
2016	int rc = PGM_GCPHYS_2_PTR_V2(pVM, pVCpu, GST_GET_PDE_GCPHYS(PdeSrc), &pPTSrc);
2017	if (RT_SUCCESS(rc))
2018	{
2019	# ifdef PGM_SYNC_N_PAGES
2020	Assert(cPages == 1 \|\| !(uErr & X86_TRAP_PF_P));
2021	if ( cPages > 1
2022	&& !(uErr & X86_TRAP_PF_P)
2023	&& !VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY))
2024	{
2025	/*
2026	* This code path is currently only taken when the caller is PGMTrap0eHandler
2027	* for non-present pages!
2028	*
2029	* We're setting PGM_SYNC_NR_PAGES pages around the faulting page to sync it and
2030	* deal with locality.
2031	*/
2032	unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2033	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
2034	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
2035	const unsigned offPTSrc = ((GCPtrPage >> SHW_PD_SHIFT) & 1) * 512;
2036	# else
2037	const unsigned offPTSrc = 0;
2038	# endif
2039	const unsigned iPTDstEnd = RT_MIN(iPTDst + PGM_SYNC_NR_PAGES / 2, RT_ELEMENTS(pPTDst->a));
2040	if (iPTDst < PGM_SYNC_NR_PAGES / 2)
2041	iPTDst = 0;
2042	else
2043	iPTDst -= PGM_SYNC_NR_PAGES / 2;
2044
2045	for (; iPTDst < iPTDstEnd; iPTDst++)
2046	{
2047	const PGSTPTE pPteSrc = &pPTSrc->a[offPTSrc + iPTDst];
2048
2049	if ( pPteSrc->n.u1Present
2050	&& !SHW_PTE_IS_P(pPTDst->a[iPTDst]))
2051	{
2052	RTGCPTR GCPtrCurPage = (GCPtrPage & ~(RTGCPTR)(GST_PT_MASK << GST_PT_SHIFT)) \| ((offPTSrc + iPTDst) << PAGE_SHIFT);
2053	NOREF(GCPtrCurPage);
2054	# ifdef VBOX_WITH_RAW_MODE_NOT_R0
2055	/*
2056	* Assuming kernel code will be marked as supervisor - and not as user level
2057	* and executed using a conforming code selector - And marked as readonly.
2058	* Also assume that if we're monitoring a page, it's of no interest to CSAM.
2059	*/
2060	PPGMPAGE pPage;
2061	if ( ((PdeSrc.u & pPteSrc->u) & (X86_PTE_RW \| X86_PTE_US))
2062	\|\| iPTDst == ((GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK) /* always sync GCPtrPage */
2063	\|\| !CSAMDoesPageNeedScanning(pVM, GCPtrCurPage)
2064	\|\| ( (pPage = pgmPhysGetPage(pVM, pPteSrc->u & GST_PTE_PG_MASK))
2065	&& PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
2066	)
2067	# endif /* else: CSAM not active */
2068	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, *pPteSrc, pShwPage, iPTDst);
2069	Log2(("SyncPage: 4K+ %RGv PteSrc:{P=%d RW=%d U=%d raw=%08llx} PteDst=%08llx%s\n",
2070	GCPtrCurPage, pPteSrc->n.u1Present,
2071	pPteSrc->n.u1Write & PdeSrc.n.u1Write,
2072	pPteSrc->n.u1User & PdeSrc.n.u1User,
2073	(uint64_t)pPteSrc->u,
2074	SHW_PTE_LOG64(pPTDst->a[iPTDst]),
2075	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : ""));
2076	}
2077	}
2078	}
2079	else
2080	# endif /* PGM_SYNC_N_PAGES */
2081	{
2082	const unsigned iPTSrc = (GCPtrPage >> GST_PT_SHIFT) & GST_PT_MASK;
2083	GSTPTE PteSrc = pPTSrc->a[iPTSrc];
2084	const unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2085	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, PteSrc, pShwPage, iPTDst);
2086	Log2(("SyncPage: 4K %RGv PteSrc:{P=%d RW=%d U=%d raw=%08llx} PteDst=%08llx %s\n",
2087	GCPtrPage, PteSrc.n.u1Present,
2088	PteSrc.n.u1Write & PdeSrc.n.u1Write,
2089	PteSrc.n.u1User & PdeSrc.n.u1User,
2090	(uint64_t)PteSrc.u,
2091	SHW_PTE_LOG64(pPTDst->a[iPTDst]),
2092	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : ""));
2093	}
2094	}
2095	else /* MMIO or invalid page: emulated in #PF handler. */
2096	{
2097	LogFlow(("PGM_GCPHYS_2_PTR %RGp failed with %Rrc\n", GCPhys, rc));
2098	Assert(!SHW_PTE_IS_P(pPTDst->a[(GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK]));
2099	}
2100	}
2101	else
2102	{
2103	/*
2104	* 4/2MB page - lazy syncing shadow 4K pages.
2105	* (There are many causes of getting here, it's no longer only CSAM.)
2106	*/
2107	/* Calculate the GC physical address of this 4KB shadow page. */
2108	GCPhys = PGM_A20_APPLY(pVCpu, GST_GET_BIG_PDE_GCPHYS(pVM, PdeSrc) \| (GCPtrPage & GST_BIG_PAGE_OFFSET_MASK));
2109	/* Find ram range. */
2110	PPGMPAGE pPage;
2111	int rc = pgmPhysGetPageEx(pVM, GCPhys, &pPage);
2112	if (RT_SUCCESS(rc))
2113	{
2114	AssertFatalMsg(!PGM_PAGE_IS_BALLOONED(pPage), ("Unexpected ballooned page at %RGp\n", GCPhys));
2115
2116	# ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
2117	/* Try to make the page writable if necessary. */
2118	if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM
2119	&& ( PGM_PAGE_IS_ZERO(pPage)
2120	\|\| ( PdeSrc.n.u1Write
2121	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED
2122	# ifdef VBOX_WITH_REAL_WRITE_MONITORED_PAGES
2123	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_WRITE_MONITORED
2124	# endif
2125	# ifdef VBOX_WITH_PAGE_SHARING
2126	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_SHARED
2127	# endif
2128	)
2129	)
2130	)
2131	{
2132	rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhys);
2133	AssertRC(rc);
2134	}
2135	# endif
2136
2137	/*
2138	* Make shadow PTE entry.
2139	*/
2140	SHWPTE PteDst;
2141	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
2142	PGM_BTH_NAME(SyncHandlerPte)(pVM, pPage, GST_GET_BIG_PDE_SHW_FLAGS_4_PTE(pVCpu, PdeSrc), &PteDst);
2143	else
2144	SHW_PTE_SET(PteDst, GST_GET_BIG_PDE_SHW_FLAGS_4_PTE(pVCpu, PdeSrc) \| PGM_PAGE_GET_HCPHYS(pPage));
2145
2146	const unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2147	if ( SHW_PTE_IS_P(PteDst)
2148	&& !SHW_PTE_IS_P(pPTDst->a[iPTDst]))
2149	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPTDst);
2150
2151	/* Make sure only allocated pages are mapped writable. */
2152	if ( SHW_PTE_IS_P_RW(PteDst)
2153	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
2154	{
2155	/* Still applies to shared pages. */
2156	Assert(!PGM_PAGE_IS_ZERO(pPage));
2157	SHW_PTE_SET_RO(PteDst); /** @todo this isn't quite working yet... */
2158	Log3(("SyncPage: write-protecting %RGp pPage=%R[pgmpage] at %RGv\n", GCPhys, pPage, GCPtrPage));
2159	}
2160
2161	SHW_PTE_ATOMIC_SET2(pPTDst->a[iPTDst], PteDst);
2162
2163	/*
2164	* If the page is not flagged as dirty and is writable, then make it read-only
2165	* at PD level, so we can set the dirty bit when the page is modified.
2166	*
2167	* ASSUMES that page access handlers are implemented on page table entry level.
2168	* Thus we will first catch the dirty access and set PDE.D and restart. If
2169	* there is an access handler, we'll trap again and let it work on the problem.
2170	*/
2171	/** @todo r=bird: figure out why we need this here, SyncPT should've taken care of this already.
2172	* As for invlpg, it simply frees the whole shadow PT.
2173	* ...It's possibly because the guest clears it and the guest doesn't really tell us... */
2174	if ( !PdeSrc.b.u1Dirty
2175	&& PdeSrc.b.u1Write)
2176	{
2177	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyPageBig));
2178	PdeDst.u \|= PGM_PDFLAGS_TRACK_DIRTY;
2179	PdeDst.n.u1Write = 0;
2180	}
2181	else
2182	{
2183	PdeDst.au32[0] &= ~PGM_PDFLAGS_TRACK_DIRTY;
2184	PdeDst.n.u1Write = PdeSrc.n.u1Write;
2185	}
2186	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
2187	Log2(("SyncPage: BIG %RGv PdeSrc:{P=%d RW=%d U=%d raw=%08llx} GCPhys=%RGp%s\n",
2188	GCPtrPage, PdeSrc.n.u1Present, PdeSrc.n.u1Write, PdeSrc.n.u1User, (uint64_t)PdeSrc.u, GCPhys,
2189	PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY ? " Track-Dirty" : ""));
2190	}
2191	else
2192	{
2193	LogFlow(("PGM_GCPHYS_2_PTR %RGp (big) failed with %Rrc\n", GCPhys, rc));
2194	/** @todo must wipe the shadow page table entry in this
2195	* case. */
2196	}
2197	}
2198	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
2199	return VINF_SUCCESS;
2200	}
2201
2202	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPagePDNAs));
2203	}
2204	else if (fPdeValid)
2205	{
2206	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPagePDOutOfSync));
2207	Log2(("SyncPage: Out-Of-Sync PDE at %RGp PdeSrc=%RX64 PdeDst=%RX64 (GCPhys %RGp vs %RGp)\n",
2208	GCPtrPage, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u, pShwPage->GCPhys, GCPhys));
2209	}
2210	else
2211	{
2212	/// @todo STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_MID_Z(Stat,SyncPagePDOutOfSyncAndInvalid));
2213	Log2(("SyncPage: Bad PDE at %RGp PdeSrc=%RX64 PdeDst=%RX64 (GCPhys %RGp vs %RGp)\n",
2214	GCPtrPage, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u, pShwPage->GCPhys, GCPhys));
2215	}
2216
2217	/*
2218	* Mark the PDE not present. Restart the instruction and let #PF call SyncPT.
2219	* Yea, I'm lazy.
2220	*/
2221	pgmPoolFreeByPage(pPool, pShwPage, pShwPde->idx, iPDDst);
2222	ASMAtomicWriteSize(pPdeDst, 0);
2223
2224	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
2225	PGM_INVL_VCPU_TLBS(pVCpu);
2226	return VINF_PGM_SYNCPAGE_MODIFIED_PDE;
2227
2228
2229	#elif (PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
2230	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
2231	&& (PGM_SHW_TYPE != PGM_TYPE_EPT \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
2232	&& !defined(IN_RC)
2233	NOREF(PdeSrc);
2234
2235	# ifdef PGM_SYNC_N_PAGES
2236	/*
2237	* Get the shadow PDE, find the shadow page table in the pool.
2238	*/
2239	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
2240	X86PDE PdeDst = pgmShwGet32BitPDE(pVCpu, GCPtrPage);
2241
2242	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
2243	X86PDEPAE PdeDst = pgmShwGetPaePDE(pVCpu, GCPtrPage);
2244
2245	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
2246	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
2247	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64; NOREF(iPdpt);
2248	PX86PDPAE pPDDst = NULL; /* initialized to shut up gcc */
2249	X86PDEPAE PdeDst;
2250	PX86PDPT pPdptDst = NULL; /* initialized to shut up gcc */
2251
2252	int rc = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, NULL, &pPdptDst, &pPDDst);
2253	AssertRCSuccessReturn(rc, rc);
2254	Assert(pPDDst && pPdptDst);
2255	PdeDst = pPDDst->a[iPDDst];
2256	# elif PGM_SHW_TYPE == PGM_TYPE_EPT
2257	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
2258	PEPTPD pPDDst;
2259	EPTPDE PdeDst;
2260
2261	int rc = pgmShwGetEPTPDPtr(pVCpu, GCPtrPage, NULL, &pPDDst);
2262	if (rc != VINF_SUCCESS)
2263	{
2264	AssertRC(rc);
2265	return rc;
2266	}
2267	Assert(pPDDst);
2268	PdeDst = pPDDst->a[iPDDst];
2269	# endif
2270	/* In the guest SMP case we could have blocked while another VCPU reused this page table. */
2271	if (!PdeDst.n.u1Present)
2272	{
2273	AssertMsg(pVM->cCpus > 1, ("Unexpected missing PDE %RX64\n", (uint64_t)PdeDst.u));
2274	Log(("CPU%d: SyncPage: Pde at %RGv changed behind our back!\n", pVCpu->idCpu, GCPtrPage));
2275	return VINF_SUCCESS; /* force the instruction to be executed again. */
2276	}
2277
2278	/* Can happen in the guest SMP case; other VCPU activated this PDE while we were blocking to handle the page fault. */
2279	if (PdeDst.n.u1Size)
2280	{
2281	Assert(pVM->pgm.s.fNestedPaging);
2282	Log(("CPU%d: SyncPage: Pde (big:%RX64) at %RGv changed behind our back!\n", pVCpu->idCpu, PdeDst.u, GCPtrPage));
2283	return VINF_SUCCESS;
2284	}
2285
2286	/* Mask away the page offset. */
2287	GCPtrPage &= ~((RTGCPTR)0xfff);
2288
2289	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, PdeDst.u & SHW_PDE_PG_MASK);
2290	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
2291
2292	Assert(cPages == 1 \|\| !(uErr & X86_TRAP_PF_P));
2293	if ( cPages > 1
2294	&& !(uErr & X86_TRAP_PF_P)
2295	&& !VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY))
2296	{
2297	/*
2298	* This code path is currently only taken when the caller is PGMTrap0eHandler
2299	* for non-present pages!
2300	*
2301	* We're setting PGM_SYNC_NR_PAGES pages around the faulting page to sync it and
2302	* deal with locality.
2303	*/
2304	unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2305	const unsigned iPTDstEnd = RT_MIN(iPTDst + PGM_SYNC_NR_PAGES / 2, RT_ELEMENTS(pPTDst->a));
2306	if (iPTDst < PGM_SYNC_NR_PAGES / 2)
2307	iPTDst = 0;
2308	else
2309	iPTDst -= PGM_SYNC_NR_PAGES / 2;
2310	for (; iPTDst < iPTDstEnd; iPTDst++)
2311	{
2312	if (!SHW_PTE_IS_P(pPTDst->a[iPTDst]))
2313	{
2314	RTGCPTR GCPtrCurPage = PGM_A20_APPLY(pVCpu, (GCPtrPage & ~(RTGCPTR)(SHW_PT_MASK << SHW_PT_SHIFT))
2315	\| (iPTDst << PAGE_SHIFT));
2316
2317	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], GCPtrCurPage, pShwPage, iPTDst);
2318	Log2(("SyncPage: 4K+ %RGv PteSrc:{P=1 RW=1 U=1} PteDst=%08llx%s\n",
2319	GCPtrCurPage,
2320	SHW_PTE_LOG64(pPTDst->a[iPTDst]),
2321	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : ""));
2322
2323	if (RT_UNLIKELY(VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY)))
2324	break;
2325	}
2326	else
2327	Log4(("%RGv iPTDst=%x pPTDst->a[iPTDst] %RX64\n", (GCPtrPage & ~(RTGCPTR)(SHW_PT_MASK << SHW_PT_SHIFT)) \| (iPTDst << PAGE_SHIFT), iPTDst, SHW_PTE_LOG64(pPTDst->a[iPTDst]) ));
2328	}
2329	}
2330	else
2331	# endif /* PGM_SYNC_N_PAGES */
2332	{
2333	const unsigned iPTDst = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2334	RTGCPTR GCPtrCurPage = PGM_A20_APPLY(pVCpu, (GCPtrPage & ~(RTGCPTR)(SHW_PT_MASK << SHW_PT_SHIFT))
2335	\| (iPTDst << PAGE_SHIFT));
2336
2337	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], GCPtrCurPage, pShwPage, iPTDst);
2338
2339	Log2(("SyncPage: 4K %RGv PteSrc:{P=1 RW=1 U=1}PteDst=%08llx%s\n",
2340	GCPtrPage,
2341	SHW_PTE_LOG64(pPTDst->a[iPTDst]),
2342	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : ""));
2343	}
2344	return VINF_SUCCESS;
2345
2346	#else
2347	NOREF(PdeSrc);
2348	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_GST_TYPE, PGM_SHW_TYPE));
2349	return VERR_PGM_NOT_USED_IN_MODE;
2350	#endif
2351	}
2352
2353
2354	#if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
2355
2356	/**
2357	* CheckPageFault helper for returning a page fault indicating a non-present
2358	* (NP) entry in the page translation structures.
2359	*
2360	* @returns VINF_EM_RAW_GUEST_TRAP.
2361	* @param pVCpu Pointer to the VMCPU.
2362	* @param uErr The error code of the shadow fault. Corrections to
2363	* TRPM's copy will be made if necessary.
2364	* @param GCPtrPage For logging.
2365	* @param uPageFaultLevel For logging.
2366	*/
2367	DECLINLINE(int) PGM_BTH_NAME(CheckPageFaultReturnNP)(PVMCPU pVCpu, uint32_t uErr, RTGCPTR GCPtrPage, unsigned uPageFaultLevel)
2368	{
2369	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyTrackRealPF));
2370	AssertMsg(!(uErr & X86_TRAP_PF_P), ("%#x\n", uErr));
2371	AssertMsg(!(uErr & X86_TRAP_PF_RSVD), ("%#x\n", uErr));
2372	if (uErr & (X86_TRAP_PF_RSVD \| X86_TRAP_PF_P))
2373	TRPMSetErrorCode(pVCpu, uErr & ~(X86_TRAP_PF_RSVD \| X86_TRAP_PF_P));
2374
2375	Log(("CheckPageFault: real page fault (notp) at %RGv (%d)\n", GCPtrPage, uPageFaultLevel));
2376	return VINF_EM_RAW_GUEST_TRAP;
2377	}
2378
2379
2380	/**
2381	* CheckPageFault helper for returning a page fault indicating a reserved bit
2382	* (RSVD) error in the page translation structures.
2383	*
2384	* @returns VINF_EM_RAW_GUEST_TRAP.
2385	* @param pVCpu Pointer to the VMCPU.
2386	* @param uErr The error code of the shadow fault. Corrections to
2387	* TRPM's copy will be made if necessary.
2388	* @param GCPtrPage For logging.
2389	* @param uPageFaultLevel For logging.
2390	*/
2391	DECLINLINE(int) PGM_BTH_NAME(CheckPageFaultReturnRSVD)(PVMCPU pVCpu, uint32_t uErr, RTGCPTR GCPtrPage, unsigned uPageFaultLevel)
2392	{
2393	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyTrackRealPF));
2394	if ((uErr & (X86_TRAP_PF_RSVD \| X86_TRAP_PF_P)) != (X86_TRAP_PF_RSVD \| X86_TRAP_PF_P))
2395	TRPMSetErrorCode(pVCpu, uErr \| X86_TRAP_PF_RSVD \| X86_TRAP_PF_P);
2396
2397	Log(("CheckPageFault: real page fault (rsvd) at %RGv (%d)\n", GCPtrPage, uPageFaultLevel));
2398	return VINF_EM_RAW_GUEST_TRAP;
2399	}
2400
2401
2402	/**
2403	* CheckPageFault helper for returning a page protection fault (P).
2404	*
2405	* @returns VINF_EM_RAW_GUEST_TRAP.
2406	* @param pVCpu Pointer to the VMCPU.
2407	* @param uErr The error code of the shadow fault. Corrections to
2408	* TRPM's copy will be made if necessary.
2409	* @param GCPtrPage For logging.
2410	* @param uPageFaultLevel For logging.
2411	*/
2412	DECLINLINE(int) PGM_BTH_NAME(CheckPageFaultReturnProt)(PVMCPU pVCpu, uint32_t uErr, RTGCPTR GCPtrPage, unsigned uPageFaultLevel)
2413	{
2414	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyTrackRealPF));
2415	AssertMsg(uErr & (X86_TRAP_PF_RW \| X86_TRAP_PF_US \| X86_TRAP_PF_ID), ("%#x\n", uErr));
2416	if ((uErr & (X86_TRAP_PF_P \| X86_TRAP_PF_RSVD)) != X86_TRAP_PF_P)
2417	TRPMSetErrorCode(pVCpu, (uErr & ~X86_TRAP_PF_RSVD) \| X86_TRAP_PF_P);
2418
2419	Log(("CheckPageFault: real page fault (prot) at %RGv (%d)\n", GCPtrPage, uPageFaultLevel));
2420	return VINF_EM_RAW_GUEST_TRAP;
2421	}
2422
2423
2424	/**
2425	* Handle dirty bit tracking faults.
2426	*
2427	* @returns VBox status code.
2428	* @param pVCpu Pointer to the VMCPU.
2429	* @param uErr Page fault error code.
2430	* @param pPdeSrc Guest page directory entry.
2431	* @param pPdeDst Shadow page directory entry.
2432	* @param GCPtrPage Guest context page address.
2433	*/
2434	static int PGM_BTH_NAME(CheckDirtyPageFault)(PVMCPU pVCpu, uint32_t uErr, PSHWPDE pPdeDst, GSTPDE const *pPdeSrc,
2435	RTGCPTR GCPtrPage)
2436	{
2437	PVM pVM = pVCpu->CTX_SUFF(pVM);
2438	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2439	NOREF(uErr);
2440
2441	PGM_LOCK_ASSERT_OWNER(pVM);
2442
2443	/*
2444	* Handle big page.
2445	*/
2446	if (pPdeSrc->b.u1Size && GST_IS_PSE_ACTIVE(pVCpu))
2447	{
2448	if ( pPdeDst->n.u1Present
2449	&& (pPdeDst->u & PGM_PDFLAGS_TRACK_DIRTY))
2450	{
2451	SHWPDE PdeDst = *pPdeDst;
2452
2453	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyPageTrap));
2454	Assert(pPdeSrc->b.u1Write);
2455
2456	/* Note: No need to invalidate this entry on other VCPUs as a stale TLB entry will not harm; write access will simply
2457	* fault again and take this path to only invalidate the entry (see below).
2458	*/
2459	PdeDst.n.u1Write = 1;
2460	PdeDst.n.u1Accessed = 1;
2461	PdeDst.au32[0] &= ~PGM_PDFLAGS_TRACK_DIRTY;
2462	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
2463	PGM_INVL_BIG_PG(pVCpu, GCPtrPage);
2464	return VINF_PGM_HANDLED_DIRTY_BIT_FAULT; /* restarts the instruction. */
2465	}
2466
2467	# ifdef IN_RING0
2468	/* Check for stale TLB entry; only applies to the SMP guest case. */
2469	if ( pVM->cCpus > 1
2470	&& pPdeDst->n.u1Write
2471	&& pPdeDst->n.u1Accessed)
2472	{
2473	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, pPdeDst->u & SHW_PDE_PG_MASK);
2474	if (pShwPage)
2475	{
2476	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
2477	PSHWPTE pPteDst = &pPTDst->a[(GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK];
2478	if (SHW_PTE_IS_P_RW(*pPteDst))
2479	{
2480	/* Stale TLB entry. */
2481	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyPageStale));
2482	PGM_INVL_PG(pVCpu, GCPtrPage);
2483	return VINF_PGM_HANDLED_DIRTY_BIT_FAULT; /* restarts the instruction. */
2484	}
2485	}
2486	}
2487	# endif /* IN_RING0 */
2488	return VINF_PGM_NO_DIRTY_BIT_TRACKING;
2489	}
2490
2491	/*
2492	* Map the guest page table.
2493	*/
2494	PGSTPT pPTSrc;
2495	int rc = PGM_GCPHYS_2_PTR_V2(pVM, pVCpu, GST_GET_PDE_GCPHYS(*pPdeSrc), &pPTSrc);
2496	if (RT_FAILURE(rc))
2497	{
2498	AssertRC(rc);
2499	return rc;
2500	}
2501
2502	if (pPdeDst->n.u1Present)
2503	{
2504	GSTPTE const *pPteSrc = &pPTSrc->a[(GCPtrPage >> GST_PT_SHIFT) & GST_PT_MASK];
2505	const GSTPTE PteSrc = *pPteSrc;
2506
2507	#ifdef VBOX_WITH_RAW_MODE_NOT_R0
2508	/* Bail out here as pgmPoolGetPage will return NULL and we'll crash below.
2509	* Our individual shadow handlers will provide more information and force a fatal exit.
2510	*/
2511	if ( !HMIsEnabled(pVM)
2512	&& MMHyperIsInsideArea(pVM, (RTGCPTR)GCPtrPage))
2513	{
2514	LogRel(("CheckPageFault: write to hypervisor region %RGv\n", GCPtrPage));
2515	return VINF_PGM_NO_DIRTY_BIT_TRACKING;
2516	}
2517	#endif
2518	/*
2519	* Map shadow page table.
2520	*/
2521	PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, pPdeDst->u & SHW_PDE_PG_MASK);
2522	if (pShwPage)
2523	{
2524	PSHWPT pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
2525	PSHWPTE pPteDst = &pPTDst->a[(GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK];
2526	if (SHW_PTE_IS_P(pPteDst)) /* @todo Optimize accessed bit emulation? */
2527	{
2528	if (SHW_PTE_IS_TRACK_DIRTY(*pPteDst))
2529	{
2530	PPGMPAGE pPage = pgmPhysGetPage(pVM, GST_GET_PTE_GCPHYS(PteSrc));
2531	SHWPTE PteDst = *pPteDst;
2532
2533	LogFlow(("DIRTY page trap addr=%RGv\n", GCPtrPage));
2534	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyPageTrap));
2535
2536	Assert(PteSrc.n.u1Write);
2537
2538	/* Note: No need to invalidate this entry on other VCPUs as a stale TLB
2539	* entry will not harm; write access will simply fault again and
2540	* take this path to only invalidate the entry.
2541	*/
2542	if (RT_LIKELY(pPage))
2543	{
2544	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
2545	{
2546	//AssertMsgFailed(("%R[pgmpage] - we don't set PGM_PTFLAGS_TRACK_DIRTY for these pages\n", pPage));
2547	Assert(!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage));
2548	/* Assuming write handlers here as the PTE is present (otherwise we wouldn't be here). */
2549	SHW_PTE_SET_RO(PteDst);
2550	}
2551	else
2552	{
2553	if ( PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_WRITE_MONITORED
2554	&& PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM)
2555	{
2556	rc = pgmPhysPageMakeWritable(pVM, pPage, GST_GET_PTE_GCPHYS(PteSrc));
2557	AssertRC(rc);
2558	}
2559	if (PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED)
2560	SHW_PTE_SET_RW(PteDst);
2561	else
2562	{
2563	/* Still applies to shared pages. */
2564	Assert(!PGM_PAGE_IS_ZERO(pPage));
2565	SHW_PTE_SET_RO(PteDst);
2566	}
2567	}
2568	}
2569	else
2570	SHW_PTE_SET_RW(PteDst); /** @todo r=bird: This doesn't make sense to me. */
2571
2572	SHW_PTE_SET(PteDst, (SHW_PTE_GET_U(PteDst) \| X86_PTE_D \| X86_PTE_A) & ~(uint64_t)PGM_PTFLAGS_TRACK_DIRTY);
2573	SHW_PTE_ATOMIC_SET2(*pPteDst, PteDst);
2574	PGM_INVL_PG(pVCpu, GCPtrPage);
2575	return VINF_PGM_HANDLED_DIRTY_BIT_FAULT; /* restarts the instruction. */
2576	}
2577
2578	# ifdef IN_RING0
2579	/* Check for stale TLB entry; only applies to the SMP guest case. */
2580	if ( pVM->cCpus > 1
2581	&& SHW_PTE_IS_RW(*pPteDst)
2582	&& SHW_PTE_IS_A(*pPteDst))
2583	{
2584	/* Stale TLB entry. */
2585	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyPageStale));
2586	PGM_INVL_PG(pVCpu, GCPtrPage);
2587	return VINF_PGM_HANDLED_DIRTY_BIT_FAULT; /* restarts the instruction. */
2588	}
2589	# endif
2590	}
2591	}
2592	else
2593	AssertMsgFailed(("pgmPoolGetPageByHCPhys %RGp failed!\n", pPdeDst->u & SHW_PDE_PG_MASK));
2594	}
2595
2596	return VINF_PGM_NO_DIRTY_BIT_TRACKING;
2597	}
2598
2599	#endif /* PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE) */
2600
2601
2602	/**
2603	* Sync a shadow page table.
2604	*
2605	* The shadow page table is not present in the shadow PDE.
2606	*
2607	* Handles mapping conflicts.
2608	*
2609	* This is called by VerifyAccessSyncPage, PrefetchPage, InvalidatePage (on
2610	* conflict), and Trap0eHandler.
2611	*
2612	* A precondition for this method is that the shadow PDE is not present. The
2613	* caller must take the PGM lock before checking this and continue to hold it
2614	* when calling this method.
2615	*
2616	* @returns VBox status code.
2617	* @param pVCpu Pointer to the VMCPU.
2618	* @param iPD Page directory index.
2619	* @param pPDSrc Source page directory (i.e. Guest OS page directory).
2620	* Assume this is a temporary mapping.
2621	* @param GCPtrPage GC Pointer of the page that caused the fault
2622	*/
2623	static int PGM_BTH_NAME(SyncPT)(PVMCPU pVCpu, unsigned iPDSrc, PGSTPD pPDSrc, RTGCPTR GCPtrPage)
2624	{
2625	PVM pVM = pVCpu->CTX_SUFF(pVM);
2626	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
2627
2628	#if 0 /* rarely useful; leave for debugging. */
2629	STAM_COUNTER_INC(&pVCpu->pgm.s.StatSyncPtPD[iPDSrc]);
2630	#endif
2631	LogFlow(("SyncPT: GCPtrPage=%RGv\n", GCPtrPage));
2632
2633	PGM_LOCK_ASSERT_OWNER(pVM);
2634
2635	#if ( PGM_GST_TYPE == PGM_TYPE_32BIT \
2636	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
2637	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64) \
2638	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
2639	&& PGM_SHW_TYPE != PGM_TYPE_EPT
2640
2641	int rc = VINF_SUCCESS;
2642
2643	STAM_PROFILE_START(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPT), a);
2644
2645	/*
2646	* Some input validation first.
2647	*/
2648	AssertMsg(iPDSrc == ((GCPtrPage >> GST_PD_SHIFT) & GST_PD_MASK), ("iPDSrc=%x GCPtrPage=%RGv\n", iPDSrc, GCPtrPage));
2649
2650	/*
2651	* Get the relevant shadow PDE entry.
2652	*/
2653	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
2654	const unsigned iPDDst = GCPtrPage >> SHW_PD_SHIFT;
2655	PSHWPDE pPdeDst = pgmShwGet32BitPDEPtr(pVCpu, GCPtrPage);
2656
2657	/* Fetch the pgm pool shadow descriptor. */
2658	PPGMPOOLPAGE pShwPde = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
2659	Assert(pShwPde);
2660
2661	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
2662	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
2663	PPGMPOOLPAGE pShwPde = NULL;
2664	PX86PDPAE pPDDst;
2665	PSHWPDE pPdeDst;
2666
2667	/* Fetch the pgm pool shadow descriptor. */
2668	rc = pgmShwGetPaePoolPagePD(pVCpu, GCPtrPage, &pShwPde);
2669	AssertRCSuccessReturn(rc, rc);
2670	Assert(pShwPde);
2671
2672	pPDDst = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPde);
2673	pPdeDst = &pPDDst->a[iPDDst];
2674
2675	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
2676	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
2677	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
2678	PX86PDPAE pPDDst = NULL; /* initialized to shut up gcc */
2679	PX86PDPT pPdptDst = NULL; /* initialized to shut up gcc */
2680	rc = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, NULL, &pPdptDst, &pPDDst);
2681	AssertRCSuccessReturn(rc, rc);
2682	Assert(pPDDst);
2683	PSHWPDE pPdeDst = &pPDDst->a[iPDDst];
2684	# endif
2685	SHWPDE PdeDst = *pPdeDst;
2686
2687	# if PGM_GST_TYPE == PGM_TYPE_AMD64
2688	/* Fetch the pgm pool shadow descriptor. */
2689	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & X86_PDPE_PG_MASK);
2690	Assert(pShwPde);
2691	# endif
2692
2693	# ifndef PGM_WITHOUT_MAPPINGS
2694	/*
2695	* Check for conflicts.
2696	* RC: In case of a conflict we'll go to Ring-3 and do a full SyncCR3.
2697	* R3: Simply resolve the conflict.
2698	*/
2699	if (PdeDst.u & PGM_PDFLAGS_MAPPING)
2700	{
2701	Assert(pgmMapAreMappingsEnabled(pVM));
2702	# ifndef IN_RING3
2703	Log(("SyncPT: Conflict at %RGv\n", GCPtrPage));
2704	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPT), a);
2705	return VERR_ADDRESS_CONFLICT;
2706
2707	# else /* IN_RING3 */
2708	PPGMMAPPING pMapping = pgmGetMapping(pVM, (RTGCPTR)GCPtrPage);
2709	Assert(pMapping);
2710	# if PGM_GST_TYPE == PGM_TYPE_32BIT
2711	rc = pgmR3SyncPTResolveConflict(pVM, pMapping, pPDSrc, GCPtrPage & (GST_PD_MASK << GST_PD_SHIFT));
2712	# elif PGM_GST_TYPE == PGM_TYPE_PAE
2713	rc = pgmR3SyncPTResolveConflictPAE(pVM, pMapping, GCPtrPage & (GST_PD_MASK << GST_PD_SHIFT));
2714	# else
2715	AssertFailed(); NOREF(pMapping); /* can't happen for amd64 */
2716	# endif
2717	if (RT_FAILURE(rc))
2718	{
2719	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPT), a);
2720	return rc;
2721	}
2722	PdeDst = *pPdeDst;
2723	# endif /* IN_RING3 */
2724	}
2725	# endif /* !PGM_WITHOUT_MAPPINGS */
2726	Assert(!PdeDst.n.u1Present); /* We're only supposed to call SyncPT on PDE!P and conflicts.*/
2727
2728	/*
2729	* Sync the page directory entry.
2730	*/
2731	GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
2732	const bool fPageTable = !PdeSrc.b.u1Size \|\| !GST_IS_PSE_ACTIVE(pVCpu);
2733	if ( PdeSrc.n.u1Present
2734	&& (fPageTable ? GST_IS_PDE_VALID(pVCpu, PdeSrc) : GST_IS_BIG_PDE_VALID(pVCpu, PdeSrc)) )
2735	{
2736	/*
2737	* Allocate & map the page table.
2738	*/
2739	PSHWPT pPTDst;
2740	PPGMPOOLPAGE pShwPage;
2741	RTGCPHYS GCPhys;
2742	if (fPageTable)
2743	{
2744	GCPhys = GST_GET_PDE_GCPHYS(PdeSrc);
2745	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
2746	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
2747	GCPhys = PGM_A20_APPLY(pVCpu, GCPhys \| ((iPDDst & 1) * (PAGE_SIZE / 2)));
2748	# endif
2749	rc = pgmPoolAlloc(pVM, GCPhys, BTH_PGMPOOLKIND_PT_FOR_PT, PGMPOOLACCESS_DONTCARE, PGM_A20_IS_ENABLED(pVCpu),
2750	pShwPde->idx, iPDDst, false /fLockPage/,
2751	&pShwPage);
2752	}
2753	else
2754	{
2755	PGMPOOLACCESS enmAccess;
2756	# if PGM_WITH_NX(PGM_GST_TYPE, PGM_SHW_TYPE)
2757	const bool fNoExecute = PdeSrc.n.u1NoExecute && GST_IS_NX_ACTIVE(pVCpu);
2758	# else
2759	const bool fNoExecute = false;
2760	# endif
2761
2762	GCPhys = GST_GET_BIG_PDE_GCPHYS(pVM, PdeSrc);
2763	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
2764	/* Select the right PDE as we're emulating a 4MB page directory with two 2 MB shadow PDEs.*/
2765	GCPhys = PGM_A20_APPLY(pVCpu, GCPhys \| (GCPtrPage & (1 << X86_PD_PAE_SHIFT)));
2766	# endif
2767	/* Determine the right kind of large page to avoid incorrect cached entry reuse. */
2768	if (PdeSrc.n.u1User)
2769	{
2770	if (PdeSrc.n.u1Write)
2771	enmAccess = (fNoExecute) ? PGMPOOLACCESS_USER_RW_NX : PGMPOOLACCESS_USER_RW;
2772	else
2773	enmAccess = (fNoExecute) ? PGMPOOLACCESS_USER_R_NX : PGMPOOLACCESS_USER_R;
2774	}
2775	else
2776	{
2777	if (PdeSrc.n.u1Write)
2778	enmAccess = (fNoExecute) ? PGMPOOLACCESS_SUPERVISOR_RW_NX : PGMPOOLACCESS_SUPERVISOR_RW;
2779	else
2780	enmAccess = (fNoExecute) ? PGMPOOLACCESS_SUPERVISOR_R_NX : PGMPOOLACCESS_SUPERVISOR_R;
2781	}
2782	rc = pgmPoolAlloc(pVM, GCPhys, BTH_PGMPOOLKIND_PT_FOR_BIG, enmAccess, PGM_A20_IS_ENABLED(pVCpu),
2783	pShwPde->idx, iPDDst, false /fLockPage/,
2784	&pShwPage);
2785	}
2786	if (rc == VINF_SUCCESS)
2787	pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
2788	else if (rc == VINF_PGM_CACHED_PAGE)
2789	{
2790	/*
2791	* The PT was cached, just hook it up.
2792	*/
2793	if (fPageTable)
2794	PdeDst.u = pShwPage->Core.Key \| GST_GET_PDE_SHW_FLAGS(pVCpu, PdeSrc);
2795	else
2796	{
2797	PdeDst.u = pShwPage->Core.Key \| GST_GET_BIG_PDE_SHW_FLAGS(pVCpu, PdeSrc);
2798	/* (see explanation and assumptions further down.) */
2799	if ( !PdeSrc.b.u1Dirty
2800	&& PdeSrc.b.u1Write)
2801	{
2802	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyPageBig));
2803	PdeDst.u \|= PGM_PDFLAGS_TRACK_DIRTY;
2804	PdeDst.b.u1Write = 0;
2805	}
2806	}
2807	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
2808	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
2809	return VINF_SUCCESS;
2810	}
2811	else if (rc == VERR_PGM_POOL_FLUSHED)
2812	{
2813	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
2814	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
2815	return VINF_PGM_SYNC_CR3;
2816	}
2817	else
2818	AssertMsgFailedReturn(("rc=%Rrc\n", rc), RT_FAILURE_NP(rc) ? rc : VERR_IPE_UNEXPECTED_INFO_STATUS);
2819	/** @todo Why do we bother preserving X86_PDE_AVL_MASK here?
2820	* Both PGM_PDFLAGS_MAPPING and PGM_PDFLAGS_TRACK_DIRTY should be
2821	* irrelevant at this point. */
2822	PdeDst.u &= X86_PDE_AVL_MASK;
2823	PdeDst.u \|= pShwPage->Core.Key;
2824
2825	/*
2826	* Page directory has been accessed (this is a fault situation, remember).
2827	*/
2828	/** @todo
2829	* Well, when the caller is PrefetchPage or InvalidatePage is isn't a
2830	* fault situation. What's more, the Trap0eHandler has already set the
2831	* accessed bit. So, it's actually just VerifyAccessSyncPage which
2832	* might need setting the accessed flag.
2833	*
2834	* The best idea is to leave this change to the caller and add an
2835	* assertion that it's set already. */
2836	pPDSrc->a[iPDSrc].n.u1Accessed = 1;
2837	if (fPageTable)
2838	{
2839	/*
2840	* Page table - 4KB.
2841	*
2842	* Sync all or just a few entries depending on PGM_SYNC_N_PAGES.
2843	*/
2844	Log2(("SyncPT: 4K %RGv PdeSrc:{P=%d RW=%d U=%d raw=%08llx}\n",
2845	GCPtrPage, PdeSrc.b.u1Present, PdeSrc.b.u1Write, PdeSrc.b.u1User, (uint64_t)PdeSrc.u));
2846	PGSTPT pPTSrc;
2847	rc = PGM_GCPHYS_2_PTR(pVM, GST_GET_PDE_GCPHYS(PdeSrc), &pPTSrc);
2848	if (RT_SUCCESS(rc))
2849	{
2850	/*
2851	* Start by syncing the page directory entry so CSAM's TLB trick works.
2852	*/
2853	PdeDst.u = (PdeDst.u & (SHW_PDE_PG_MASK \| X86_PDE_AVL_MASK))
2854	\| GST_GET_PDE_SHW_FLAGS(pVCpu, PdeSrc);
2855	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
2856	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
2857
2858	/*
2859	* Directory/page user or supervisor privilege: (same goes for read/write)
2860	*
2861	* Directory Page Combined
2862	* U/S U/S U/S
2863	* 0 0 0
2864	* 0 1 0
2865	* 1 0 0
2866	* 1 1 1
2867	*
2868	* Simple AND operation. Table listed for completeness.
2869	*
2870	*/
2871	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPT4K));
2872	# ifdef PGM_SYNC_N_PAGES
2873	unsigned iPTBase = (GCPtrPage >> SHW_PT_SHIFT) & SHW_PT_MASK;
2874	unsigned iPTDst = iPTBase;
2875	const unsigned iPTDstEnd = RT_MIN(iPTDst + PGM_SYNC_NR_PAGES / 2, RT_ELEMENTS(pPTDst->a));
2876	if (iPTDst <= PGM_SYNC_NR_PAGES / 2)
2877	iPTDst = 0;
2878	else
2879	iPTDst -= PGM_SYNC_NR_PAGES / 2;
2880	# else /* !PGM_SYNC_N_PAGES */
2881	unsigned iPTDst = 0;
2882	const unsigned iPTDstEnd = RT_ELEMENTS(pPTDst->a);
2883	# endif /* !PGM_SYNC_N_PAGES */
2884	RTGCPTR GCPtrCur = (GCPtrPage & ~(RTGCPTR)((1 << SHW_PD_SHIFT) - 1))
2885	\| ((RTGCPTR)iPTDst << PAGE_SHIFT);
2886	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
2887	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
2888	const unsigned offPTSrc = ((GCPtrPage >> SHW_PD_SHIFT) & 1) * 512;
2889	# else
2890	const unsigned offPTSrc = 0;
2891	# endif
2892	for (; iPTDst < iPTDstEnd; iPTDst++, GCPtrCur += PAGE_SIZE)
2893	{
2894	const unsigned iPTSrc = iPTDst + offPTSrc;
2895	const GSTPTE PteSrc = pPTSrc->a[iPTSrc];
2896
2897	if (PteSrc.n.u1Present)
2898	{
2899	# ifdef VBOX_WITH_RAW_MODE_NOT_R0
2900	/*
2901	* Assuming kernel code will be marked as supervisor - and not as user level
2902	* and executed using a conforming code selector - And marked as readonly.
2903	* Also assume that if we're monitoring a page, it's of no interest to CSAM.
2904	*/
2905	PPGMPAGE pPage;
2906	if ( ((PdeSrc.u & pPTSrc->a[iPTSrc].u) & (X86_PTE_RW \| X86_PTE_US))
2907	\|\| !CSAMDoesPageNeedScanning(pVM, GCPtrCur)
2908	\|\| ( (pPage = pgmPhysGetPage(pVM, GST_GET_PTE_GCPHYS(PteSrc)))
2909	&& PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
2910	)
2911	# endif
2912	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], PdeSrc, PteSrc, pShwPage, iPTDst);
2913	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",
2914	GCPtrCur,
2915	PteSrc.n.u1Present,
2916	PteSrc.n.u1Write & PdeSrc.n.u1Write,
2917	PteSrc.n.u1User & PdeSrc.n.u1User,
2918	(uint64_t)PteSrc.u,
2919	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : "", SHW_PTE_LOG64(pPTDst->a[iPTDst]), iPTSrc, PdeSrc.au32[0],
2920	(RTGCPHYS)(GST_GET_PDE_GCPHYS(PdeSrc) + iPTSrc*sizeof(PteSrc)) ));
2921	}
2922	/* else: the page table was cleared by the pool */
2923	} /* for PTEs */
2924	}
2925	}
2926	else
2927	{
2928	/*
2929	* Big page - 2/4MB.
2930	*
2931	* We'll walk the ram range list in parallel and optimize lookups.
2932	* We will only sync one shadow page table at a time.
2933	*/
2934	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPT4M));
2935
2936	/**
2937	* @todo It might be more efficient to sync only a part of the 4MB
2938	* page (similar to what we do for 4KB PDs).
2939	*/
2940
2941	/*
2942	* Start by syncing the page directory entry.
2943	*/
2944	PdeDst.u = (PdeDst.u & (SHW_PDE_PG_MASK \| (X86_PDE_AVL_MASK & ~PGM_PDFLAGS_TRACK_DIRTY)))
2945	\| GST_GET_BIG_PDE_SHW_FLAGS(pVCpu, PdeSrc);
2946
2947	/*
2948	* If the page is not flagged as dirty and is writable, then make it read-only
2949	* at PD level, so we can set the dirty bit when the page is modified.
2950	*
2951	* ASSUMES that page access handlers are implemented on page table entry level.
2952	* Thus we will first catch the dirty access and set PDE.D and restart. If
2953	* there is an access handler, we'll trap again and let it work on the problem.
2954	*/
2955	/** @todo move the above stuff to a section in the PGM documentation. */
2956	Assert(!(PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY));
2957	if ( !PdeSrc.b.u1Dirty
2958	&& PdeSrc.b.u1Write)
2959	{
2960	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,DirtyPageBig));
2961	PdeDst.u \|= PGM_PDFLAGS_TRACK_DIRTY;
2962	PdeDst.b.u1Write = 0;
2963	}
2964	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
2965	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
2966
2967	/*
2968	* Fill the shadow page table.
2969	*/
2970	/* Get address and flags from the source PDE. */
2971	SHWPTE PteDstBase;
2972	SHW_PTE_SET(PteDstBase, GST_GET_BIG_PDE_SHW_FLAGS_4_PTE(pVCpu, PdeSrc));
2973
2974	/* Loop thru the entries in the shadow PT. */
2975	const RTGCPTR GCPtr = (GCPtrPage >> SHW_PD_SHIFT) << SHW_PD_SHIFT; NOREF(GCPtr);
2976	Log2(("SyncPT: BIG %RGv PdeSrc:{P=%d RW=%d U=%d raw=%08llx} Shw=%RGv GCPhys=%RGp %s\n",
2977	GCPtrPage, PdeSrc.b.u1Present, PdeSrc.b.u1Write, PdeSrc.b.u1User, (uint64_t)PdeSrc.u, GCPtr,
2978	GCPhys, PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY ? " Track-Dirty" : ""));
2979	PPGMRAMRANGE pRam = pgmPhysGetRangeAtOrAbove(pVM, GCPhys);
2980	unsigned iPTDst = 0;
2981	while ( iPTDst < RT_ELEMENTS(pPTDst->a)
2982	&& !VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY))
2983	{
2984	if (pRam && GCPhys >= pRam->GCPhys)
2985	{
2986	# ifndef PGM_WITH_A20
2987	unsigned iHCPage = (GCPhys - pRam->GCPhys) >> PAGE_SHIFT;
2988	# endif
2989	do
2990	{
2991	/* Make shadow PTE. */
2992	# ifdef PGM_WITH_A20
2993	PPGMPAGE pPage = &pRam->aPages[(GCPhys - pRam->GCPhys) >> PAGE_SHIFT];
2994	# else
2995	PPGMPAGE pPage = &pRam->aPages[iHCPage];
2996	# endif
2997	SHWPTE PteDst;
2998
2999	# ifndef VBOX_WITH_NEW_LAZY_PAGE_ALLOC
3000	/* Try to make the page writable if necessary. */
3001	if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM
3002	&& ( PGM_PAGE_IS_ZERO(pPage)
3003	\|\| ( SHW_PTE_IS_RW(PteDstBase)
3004	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED
3005	# ifdef VBOX_WITH_REAL_WRITE_MONITORED_PAGES
3006	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_WRITE_MONITORED
3007	# endif
3008	# ifdef VBOX_WITH_PAGE_SHARING
3009	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_SHARED
3010	# endif
3011	&& !PGM_PAGE_IS_BALLOONED(pPage))
3012	)
3013	)
3014	{
3015	rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhys);
3016	AssertRCReturn(rc, rc);
3017	if (VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY))
3018	break;
3019	}
3020	# endif
3021
3022	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
3023	PGM_BTH_NAME(SyncHandlerPte)(pVM, pPage, SHW_PTE_GET_U(PteDstBase), &PteDst);
3024	else if (PGM_PAGE_IS_BALLOONED(pPage))
3025	SHW_PTE_SET(PteDst, 0); /* Handle ballooned pages at #PF time. */
3026	# ifdef VBOX_WITH_RAW_MODE_NOT_R0
3027	/*
3028	* Assuming kernel code will be marked as supervisor and not as user level and executed
3029	* using a conforming code selector. Don't check for readonly, as that implies the whole
3030	* 4MB can be code or readonly data. Linux enables write access for its large pages.
3031	*/
3032	else if ( !PdeSrc.n.u1User
3033	&& CSAMDoesPageNeedScanning(pVM, GCPtr \| (iPTDst << SHW_PT_SHIFT)))
3034	SHW_PTE_SET(PteDst, 0);
3035	# endif
3036	else
3037	SHW_PTE_SET(PteDst, PGM_PAGE_GET_HCPHYS(pPage) \| SHW_PTE_GET_U(PteDstBase));
3038
3039	/* Only map writable pages writable. */
3040	if ( SHW_PTE_IS_P_RW(PteDst)
3041	&& PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED)
3042	{
3043	/* Still applies to shared pages. */
3044	Assert(!PGM_PAGE_IS_ZERO(pPage));
3045	SHW_PTE_SET_RO(PteDst); /** @todo this isn't quite working yet... */
3046	Log3(("SyncPT: write-protecting %RGp pPage=%R[pgmpage] at %RGv\n", GCPhys, pPage, (RTGCPTR)(GCPtr \| (iPTDst << SHW_PT_SHIFT))));
3047	}
3048
3049	if (SHW_PTE_IS_P(PteDst))
3050	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPage, PGM_PAGE_GET_TRACKING(pPage), pPage, iPTDst);
3051
3052	/* commit it (not atomic, new table) */
3053	pPTDst->a[iPTDst] = PteDst;
3054	Log4(("SyncPT: BIG %RGv PteDst:{P=%d RW=%d U=%d raw=%08llx}%s\n",
3055	(RTGCPTR)(GCPtr \| (iPTDst << SHW_PT_SHIFT)), SHW_PTE_IS_P(PteDst), SHW_PTE_IS_RW(PteDst), SHW_PTE_IS_US(PteDst), SHW_PTE_LOG64(PteDst),
3056	SHW_PTE_IS_TRACK_DIRTY(PteDst) ? " Track-Dirty" : ""));
3057
3058	/* advance */
3059	GCPhys += PAGE_SIZE;
3060	PGM_A20_APPLY_TO_VAR(pVCpu, GCPhys);
3061	# ifndef PGM_WITH_A20
3062	iHCPage++;
3063	# endif
3064	iPTDst++;
3065	} while ( iPTDst < RT_ELEMENTS(pPTDst->a)
3066	&& GCPhys <= pRam->GCPhysLast);
3067
3068	/* Advance ram range list. */
3069	while (pRam && GCPhys > pRam->GCPhysLast)
3070	pRam = pRam->CTX_SUFF(pNext);
3071	}
3072	else if (pRam)
3073	{
3074	Log(("Invalid pages at %RGp\n", GCPhys));
3075	do
3076	{
3077	SHW_PTE_SET(pPTDst->a[iPTDst], 0); /* Invalid page, we must handle them manually. */
3078	GCPhys += PAGE_SIZE;
3079	iPTDst++;
3080	} while ( iPTDst < RT_ELEMENTS(pPTDst->a)
3081	&& GCPhys < pRam->GCPhys);
3082	PGM_A20_APPLY_TO_VAR(pVCpu,GCPhys);
3083	}
3084	else
3085	{
3086	Log(("Invalid pages at %RGp (2)\n", GCPhys));
3087	for ( ; iPTDst < RT_ELEMENTS(pPTDst->a); iPTDst++)
3088	SHW_PTE_SET(pPTDst->a[iPTDst], 0); /* Invalid page, we must handle them manually. */
3089	}
3090	} /* while more PTEs */
3091	} /* 4KB / 4MB */
3092	}
3093	else
3094	AssertRelease(!PdeDst.n.u1Present);
3095
3096	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPT), a);
3097	if (RT_FAILURE(rc))
3098	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPTFailed));
3099	return rc;
3100
3101	#elif (PGM_GST_TYPE == PGM_TYPE_REAL \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
3102	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
3103	&& (PGM_SHW_TYPE != PGM_TYPE_EPT \|\| PGM_GST_TYPE == PGM_TYPE_PROT) \
3104	&& !defined(IN_RC)
3105	NOREF(iPDSrc); NOREF(pPDSrc);
3106
3107	STAM_PROFILE_START(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPT), a);
3108
3109	/*
3110	* Validate input a little bit.
3111	*/
3112	int rc = VINF_SUCCESS;
3113	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
3114	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
3115	PSHWPDE pPdeDst = pgmShwGet32BitPDEPtr(pVCpu, GCPtrPage);
3116
3117	/* Fetch the pgm pool shadow descriptor. */
3118	PPGMPOOLPAGE pShwPde = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
3119	Assert(pShwPde);
3120
3121	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
3122	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
3123	PPGMPOOLPAGE pShwPde = NULL; /* initialized to shut up gcc */
3124	PX86PDPAE pPDDst;
3125	PSHWPDE pPdeDst;
3126
3127	/* Fetch the pgm pool shadow descriptor. */
3128	rc = pgmShwGetPaePoolPagePD(pVCpu, GCPtrPage, &pShwPde);
3129	AssertRCSuccessReturn(rc, rc);
3130	Assert(pShwPde);
3131
3132	pPDDst = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPde);
3133	pPdeDst = &pPDDst->a[iPDDst];
3134
3135	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
3136	const unsigned iPdpt = (GCPtrPage >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
3137	const unsigned iPDDst = (GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK;
3138	PX86PDPAE pPDDst = NULL; /* initialized to shut up gcc */
3139	PX86PDPT pPdptDst= NULL; /* initialized to shut up gcc */
3140	rc = pgmShwGetLongModePDPtr(pVCpu, GCPtrPage, NULL, &pPdptDst, &pPDDst);
3141	AssertRCSuccessReturn(rc, rc);
3142	Assert(pPDDst);
3143	PSHWPDE pPdeDst = &pPDDst->a[iPDDst];
3144
3145	/* Fetch the pgm pool shadow descriptor. */
3146	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & X86_PDPE_PG_MASK);
3147	Assert(pShwPde);
3148
3149	# elif PGM_SHW_TYPE == PGM_TYPE_EPT
3150	const unsigned iPdpt = (GCPtrPage >> EPT_PDPT_SHIFT) & EPT_PDPT_MASK;
3151	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
3152	PEPTPD pPDDst;
3153	PEPTPDPT pPdptDst;
3154
3155	rc = pgmShwGetEPTPDPtr(pVCpu, GCPtrPage, &pPdptDst, &pPDDst);
3156	if (rc != VINF_SUCCESS)
3157	{
3158	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPT), a);
3159	AssertRC(rc);
3160	return rc;
3161	}
3162	Assert(pPDDst);
3163	PSHWPDE pPdeDst = &pPDDst->a[iPDDst];
3164
3165	/* Fetch the pgm pool shadow descriptor. */
3166	PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pPool, pPdptDst->a[iPdpt].u & EPT_PDPTE_PG_MASK);
3167	Assert(pShwPde);
3168	# endif
3169	SHWPDE PdeDst = *pPdeDst;
3170
3171	Assert(!(PdeDst.u & PGM_PDFLAGS_MAPPING));
3172	Assert(!PdeDst.n.u1Present); /* We're only supposed to call SyncPT on PDE!P and conflicts.*/
3173
3174	# if defined(PGM_WITH_LARGE_PAGES) && PGM_SHW_TYPE != PGM_TYPE_32BIT && PGM_SHW_TYPE != PGM_TYPE_PAE
3175	if (BTH_IS_NP_ACTIVE(pVM))
3176	{
3177	/* Check if we allocated a big page before for this 2 MB range. */
3178	PPGMPAGE pPage;
3179	rc = pgmPhysGetPageEx(pVM, PGM_A20_APPLY(pVCpu, GCPtrPage & X86_PDE2M_PAE_PG_MASK), &pPage);
3180	if (RT_SUCCESS(rc))
3181	{
3182	RTHCPHYS HCPhys = NIL_RTHCPHYS;
3183	if (PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE)
3184	{
3185	if (PGM_A20_IS_ENABLED(pVCpu))
3186	{
3187	STAM_REL_COUNTER_INC(&pVM->pgm.s.StatLargePageReused);
3188	AssertRelease(PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED);
3189	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
3190	}
3191	else
3192	{
3193	PGM_PAGE_SET_PDE_TYPE(pVM, pPage, PGM_PAGE_PDE_TYPE_PDE_DISABLED);
3194	pVM->pgm.s.cLargePagesDisabled++;
3195	}
3196	}
3197	else if ( PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE_DISABLED
3198	&& PGM_A20_IS_ENABLED(pVCpu))
3199	{
3200	/* Recheck the entire 2 MB range to see if we can use it again as a large page. */
3201	rc = pgmPhysRecheckLargePage(pVM, GCPtrPage, pPage);
3202	if (RT_SUCCESS(rc))
3203	{
3204	Assert(PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED);
3205	Assert(PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE);
3206	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
3207	}
3208	}
3209	else if ( PGMIsUsingLargePages(pVM)
3210	&& PGM_A20_IS_ENABLED(pVCpu))
3211	{
3212	rc = pgmPhysAllocLargePage(pVM, GCPtrPage);
3213	if (RT_SUCCESS(rc))
3214	{
3215	Assert(PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED);
3216	Assert(PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE);
3217	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
3218	}
3219	else
3220	LogFlow(("pgmPhysAllocLargePage failed with %Rrc\n", rc));
3221	}
3222
3223	if (HCPhys != NIL_RTHCPHYS)
3224	{
3225	PdeDst.u &= X86_PDE_AVL_MASK;
3226	PdeDst.u \|= HCPhys;
3227	PdeDst.n.u1Present = 1;
3228	PdeDst.n.u1Write = 1;
3229	PdeDst.b.u1Size = 1;
3230	# if PGM_SHW_TYPE == PGM_TYPE_EPT
3231	PdeDst.n.u1Execute = 1;
3232	PdeDst.b.u1IgnorePAT = 1;
3233	PdeDst.b.u3EMT = VMX_EPT_MEMTYPE_WB;
3234	# else
3235	PdeDst.n.u1User = 1;
3236	# endif
3237	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
3238
3239	Log(("SyncPT: Use large page at %RGp PDE=%RX64\n", GCPtrPage, PdeDst.u));
3240	/* Add a reference to the first page only. */
3241	PGM_BTH_NAME(SyncPageWorkerTrackAddref)(pVCpu, pShwPde, PGM_PAGE_GET_TRACKING(pPage), pPage, iPDDst);
3242
3243	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPT), a);
3244	return VINF_SUCCESS;
3245	}
3246	}
3247	}
3248	# endif /* HC_ARCH_BITS == 64 */
3249
3250	/*
3251	* Allocate & map the page table.
3252	*/
3253	PSHWPT pPTDst;
3254	PPGMPOOLPAGE pShwPage;
3255	RTGCPHYS GCPhys;
3256
3257	/* Virtual address = physical address */
3258	GCPhys = PGM_A20_APPLY(pVCpu, GCPtrPage & X86_PAGE_4K_BASE_MASK);
3259	rc = pgmPoolAlloc(pVM, GCPhys & ~(RT_BIT_64(SHW_PD_SHIFT) - 1), BTH_PGMPOOLKIND_PT_FOR_PT, PGMPOOLACCESS_DONTCARE,
3260	PGM_A20_IS_ENABLED(pVCpu), pShwPde->idx, iPDDst, false /fLockPage/,
3261	&pShwPage);
3262	if ( rc == VINF_SUCCESS
3263	\|\| rc == VINF_PGM_CACHED_PAGE)
3264	pPTDst = (PSHWPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
3265	else
3266	{
3267	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPT), a);
3268	AssertMsgFailedReturn(("rc=%Rrc\n", rc), RT_FAILURE_NP(rc) ? rc : VERR_IPE_UNEXPECTED_INFO_STATUS);
3269	}
3270
3271	if (rc == VINF_SUCCESS)
3272	{
3273	/* New page table; fully set it up. */
3274	Assert(pPTDst);
3275
3276	/* Mask away the page offset. */
3277	GCPtrPage &= ~(RTGCPTR)PAGE_OFFSET_MASK;
3278
3279	for (unsigned iPTDst = 0; iPTDst < RT_ELEMENTS(pPTDst->a); iPTDst++)
3280	{
3281	RTGCPTR GCPtrCurPage = PGM_A20_APPLY(pVCpu, (GCPtrPage & ~(RTGCPTR)(SHW_PT_MASK << SHW_PT_SHIFT))
3282	\| (iPTDst << PAGE_SHIFT));
3283
3284	PGM_BTH_NAME(SyncPageWorker)(pVCpu, &pPTDst->a[iPTDst], GCPtrCurPage, pShwPage, iPTDst);
3285	Log2(("SyncPage: 4K+ %RGv PteSrc:{P=1 RW=1 U=1} PteDst=%08llx%s\n",
3286	GCPtrCurPage,
3287	SHW_PTE_LOG64(pPTDst->a[iPTDst]),
3288	SHW_PTE_IS_TRACK_DIRTY(pPTDst->a[iPTDst]) ? " Track-Dirty" : ""));
3289
3290	if (RT_UNLIKELY(VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY)))
3291	break;
3292	}
3293	}
3294	else
3295	rc = VINF_SUCCESS; /* Cached entry; assume it's still fully valid. */
3296
3297	/* Save the new PDE. */
3298	PdeDst.u &= X86_PDE_AVL_MASK;
3299	PdeDst.u \|= pShwPage->Core.Key;
3300	PdeDst.n.u1Present = 1;
3301	PdeDst.n.u1Write = 1;
3302	# if PGM_SHW_TYPE == PGM_TYPE_EPT
3303	PdeDst.n.u1Execute = 1;
3304	# else
3305	PdeDst.n.u1User = 1;
3306	PdeDst.n.u1Accessed = 1;
3307	# endif
3308	ASMAtomicWriteSize(pPdeDst, PdeDst.u);
3309
3310	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPT), a);
3311	if (RT_FAILURE(rc))
3312	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncPTFailed));
3313	return rc;
3314
3315	#else
3316	NOREF(iPDSrc); NOREF(pPDSrc);
3317	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_SHW_TYPE, PGM_GST_TYPE));
3318	return VERR_PGM_NOT_USED_IN_MODE;
3319	#endif
3320	}
3321
3322
3323
3324	/**
3325	* Prefetch a page/set of pages.
3326	*
3327	* Typically used to sync commonly used pages before entering raw mode
3328	* after a CR3 reload.
3329	*
3330	* @returns VBox status code.
3331	* @param pVCpu Pointer to the VMCPU.
3332	* @param GCPtrPage Page to invalidate.
3333	*/
3334	PGM_BTH_DECL(int, PrefetchPage)(PVMCPU pVCpu, RTGCPTR GCPtrPage)
3335	{
3336	#if ( PGM_GST_TYPE == PGM_TYPE_32BIT \
3337	\|\| PGM_GST_TYPE == PGM_TYPE_REAL \
3338	\|\| PGM_GST_TYPE == PGM_TYPE_PROT \
3339	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
3340	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64 ) \
3341	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
3342	&& PGM_SHW_TYPE != PGM_TYPE_EPT
3343
3344	/*
3345	* Check that all Guest levels thru the PDE are present, getting the
3346	* PD and PDE in the processes.
3347	*/
3348	int rc = VINF_SUCCESS;
3349	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
3350	# if PGM_GST_TYPE == PGM_TYPE_32BIT
3351	const unsigned iPDSrc = GCPtrPage >> GST_PD_SHIFT;
3352	PGSTPD pPDSrc = pgmGstGet32bitPDPtr(pVCpu);
3353	# elif PGM_GST_TYPE == PGM_TYPE_PAE
3354	unsigned iPDSrc;
3355	X86PDPE PdpeSrc;
3356	PGSTPD pPDSrc = pgmGstGetPaePDPtr(pVCpu, GCPtrPage, &iPDSrc, &PdpeSrc);
3357	if (!pPDSrc)
3358	return VINF_SUCCESS; /* not present */
3359	# elif PGM_GST_TYPE == PGM_TYPE_AMD64
3360	unsigned iPDSrc;
3361	PX86PML4E pPml4eSrc;
3362	X86PDPE PdpeSrc;
3363	PGSTPD pPDSrc = pgmGstGetLongModePDPtr(pVCpu, GCPtrPage, &pPml4eSrc, &PdpeSrc, &iPDSrc);
3364	if (!pPDSrc)
3365	return VINF_SUCCESS; /* not present */
3366	# endif
3367	const GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
3368	# else
3369	PGSTPD pPDSrc = NULL;
3370	const unsigned iPDSrc = 0;
3371	GSTPDE PdeSrc;
3372
3373	PdeSrc.u = 0; /* faked so we don't have to #ifdef everything */
3374	PdeSrc.n.u1Present = 1;
3375	PdeSrc.n.u1Write = 1;
3376	PdeSrc.n.u1Accessed = 1;
3377	PdeSrc.n.u1User = 1;
3378	# endif
3379
3380	if (PdeSrc.n.u1Present && PdeSrc.n.u1Accessed)
3381	{
3382	PVM pVM = pVCpu->CTX_SUFF(pVM);
3383	pgmLock(pVM);
3384
3385	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
3386	const X86PDE PdeDst = pgmShwGet32BitPDE(pVCpu, GCPtrPage);
3387	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
3388	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
3389	PX86PDPAE pPDDst;
3390	X86PDEPAE PdeDst;
3391	# if PGM_GST_TYPE != PGM_TYPE_PAE
3392	X86PDPE PdpeSrc;
3393
3394	/* Fake PDPT entry; access control handled on the page table level, so allow everything. */
3395	PdpeSrc.u = X86_PDPE_P; /* rw/us are reserved for PAE pdpte's; accessed bit causes invalid VT-x guest state errors */
3396	# endif
3397	rc = pgmShwSyncPaePDPtr(pVCpu, GCPtrPage, PdpeSrc.u, &pPDDst);
3398	if (rc != VINF_SUCCESS)
3399	{
3400	pgmUnlock(pVM);
3401	AssertRC(rc);
3402	return rc;
3403	}
3404	Assert(pPDDst);
3405	PdeDst = pPDDst->a[iPDDst];
3406
3407	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
3408	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
3409	PX86PDPAE pPDDst;
3410	X86PDEPAE PdeDst;
3411
3412	# if PGM_GST_TYPE == PGM_TYPE_PROT
3413	/* AMD-V nested paging */
3414	X86PML4E Pml4eSrc;
3415	X86PDPE PdpeSrc;
3416	PX86PML4E pPml4eSrc = &Pml4eSrc;
3417
3418	/* Fake PML4 & PDPT entry; access control handled on the page table level, so allow everything. */
3419	Pml4eSrc.u = X86_PML4E_P \| X86_PML4E_RW \| X86_PML4E_US \| X86_PML4E_A;
3420	PdpeSrc.u = X86_PDPE_P \| X86_PDPE_RW \| X86_PDPE_US \| X86_PDPE_A;
3421	# endif
3422
3423	rc = pgmShwSyncLongModePDPtr(pVCpu, GCPtrPage, pPml4eSrc->u, PdpeSrc.u, &pPDDst);
3424	if (rc != VINF_SUCCESS)
3425	{
3426	pgmUnlock(pVM);
3427	AssertRC(rc);
3428	return rc;
3429	}
3430	Assert(pPDDst);
3431	PdeDst = pPDDst->a[iPDDst];
3432	# endif
3433	if (!(PdeDst.u & PGM_PDFLAGS_MAPPING))
3434	{
3435	if (!PdeDst.n.u1Present)
3436	{
3437	/** @todo r=bird: This guy will set the A bit on the PDE,
3438	* probably harmless. */
3439	rc = PGM_BTH_NAME(SyncPT)(pVCpu, iPDSrc, pPDSrc, GCPtrPage);
3440	}
3441	else
3442	{
3443	/* Note! We used to sync PGM_SYNC_NR_PAGES pages, which triggered assertions in CSAM, because
3444	* R/W attributes of nearby pages were reset. Not sure how that could happen. Anyway, it
3445	* makes no sense to prefetch more than one page.
3446	*/
3447	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, GCPtrPage, 1, 0);
3448	if (RT_SUCCESS(rc))
3449	rc = VINF_SUCCESS;
3450	}
3451	}
3452	pgmUnlock(pVM);
3453	}
3454	return rc;
3455
3456	#elif PGM_SHW_TYPE == PGM_TYPE_NESTED \|\| PGM_SHW_TYPE == PGM_TYPE_EPT
3457	NOREF(pVCpu); NOREF(GCPtrPage);
3458	return VINF_SUCCESS; /* ignore */
3459	#else
3460	AssertCompile(0);
3461	#endif
3462	}
3463
3464
3465
3466
3467	/**
3468	* Syncs a page during a PGMVerifyAccess() call.
3469	*
3470	* @returns VBox status code (informational included).
3471	* @param pVCpu Pointer to the VMCPU.
3472	* @param GCPtrPage The address of the page to sync.
3473	* @param fPage The effective guest page flags.
3474	* @param uErr The trap error code.
3475	* @remarks This will normally never be called on invalid guest page
3476	* translation entries.
3477	*/
3478	PGM_BTH_DECL(int, VerifyAccessSyncPage)(PVMCPU pVCpu, RTGCPTR GCPtrPage, unsigned fPage, unsigned uErr)
3479	{
3480	PVM pVM = pVCpu->CTX_SUFF(pVM); NOREF(pVM);
3481
3482	LogFlow(("VerifyAccessSyncPage: GCPtrPage=%RGv fPage=%#x uErr=%#x\n", GCPtrPage, fPage, uErr));
3483
3484	Assert(!pVM->pgm.s.fNestedPaging);
3485	#if ( PGM_GST_TYPE == PGM_TYPE_32BIT \
3486	\|\| PGM_GST_TYPE == PGM_TYPE_REAL \
3487	\|\| PGM_GST_TYPE == PGM_TYPE_PROT \
3488	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
3489	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64 ) \
3490	&& PGM_SHW_TYPE != PGM_TYPE_NESTED \
3491	&& PGM_SHW_TYPE != PGM_TYPE_EPT
3492
3493	# ifdef VBOX_WITH_RAW_MODE_NOT_R0
3494	if (!(fPage & X86_PTE_US))
3495	{
3496	/*
3497	* Mark this page as safe.
3498	*/
3499	/** @todo not correct for pages that contain both code and data!! */
3500	Log(("CSAMMarkPage %RGv; scanned=%d\n", GCPtrPage, true));
3501	CSAMMarkPage(pVM, GCPtrPage, true);
3502	}
3503	# endif
3504
3505	/*
3506	* Get guest PD and index.
3507	*/
3508	/** @todo Performance: We've done all this a jiffy ago in the
3509	* PGMGstGetPage call. */
3510	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
3511	# if PGM_GST_TYPE == PGM_TYPE_32BIT
3512	const unsigned iPDSrc = GCPtrPage >> GST_PD_SHIFT;
3513	PGSTPD pPDSrc = pgmGstGet32bitPDPtr(pVCpu);
3514
3515	# elif PGM_GST_TYPE == PGM_TYPE_PAE
3516	unsigned iPDSrc = 0;
3517	X86PDPE PdpeSrc;
3518	PGSTPD pPDSrc = pgmGstGetPaePDPtr(pVCpu, GCPtrPage, &iPDSrc, &PdpeSrc);
3519	if (RT_UNLIKELY(!pPDSrc))
3520	{
3521	Log(("PGMVerifyAccess: access violation for %RGv due to non-present PDPTR\n", GCPtrPage));
3522	return VINF_EM_RAW_GUEST_TRAP;
3523	}
3524
3525	# elif PGM_GST_TYPE == PGM_TYPE_AMD64
3526	unsigned iPDSrc = 0; /* shut up gcc */
3527	PX86PML4E pPml4eSrc = NULL; /* ditto */
3528	X86PDPE PdpeSrc;
3529	PGSTPD pPDSrc = pgmGstGetLongModePDPtr(pVCpu, GCPtrPage, &pPml4eSrc, &PdpeSrc, &iPDSrc);
3530	if (RT_UNLIKELY(!pPDSrc))
3531	{
3532	Log(("PGMVerifyAccess: access violation for %RGv due to non-present PDPTR\n", GCPtrPage));
3533	return VINF_EM_RAW_GUEST_TRAP;
3534	}
3535	# endif
3536
3537	# else /* !PGM_WITH_PAGING */
3538	PGSTPD pPDSrc = NULL;
3539	const unsigned iPDSrc = 0;
3540	# endif /* !PGM_WITH_PAGING */
3541	int rc = VINF_SUCCESS;
3542
3543	pgmLock(pVM);
3544
3545	/*
3546	* First check if the shadow pd is present.
3547	*/
3548	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
3549	PX86PDE pPdeDst = pgmShwGet32BitPDEPtr(pVCpu, GCPtrPage);
3550
3551	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
3552	PX86PDEPAE pPdeDst;
3553	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
3554	PX86PDPAE pPDDst;
3555	# if PGM_GST_TYPE != PGM_TYPE_PAE
3556	/* Fake PDPT entry; access control handled on the page table level, so allow everything. */
3557	X86PDPE PdpeSrc;
3558	PdpeSrc.u = X86_PDPE_P; /* rw/us are reserved for PAE pdpte's; accessed bit causes invalid VT-x guest state errors */
3559	# endif
3560	rc = pgmShwSyncPaePDPtr(pVCpu, GCPtrPage, PdpeSrc.u, &pPDDst);
3561	if (rc != VINF_SUCCESS)
3562	{
3563	pgmUnlock(pVM);
3564	AssertRC(rc);
3565	return rc;
3566	}
3567	Assert(pPDDst);
3568	pPdeDst = &pPDDst->a[iPDDst];
3569
3570	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
3571	const unsigned iPDDst = ((GCPtrPage >> SHW_PD_SHIFT) & SHW_PD_MASK);
3572	PX86PDPAE pPDDst;
3573	PX86PDEPAE pPdeDst;
3574
3575	# if PGM_GST_TYPE == PGM_TYPE_PROT
3576	/* AMD-V nested paging: Fake PML4 & PDPT entry; access control handled on the page table level, so allow everything. */
3577	X86PML4E Pml4eSrc;
3578	X86PDPE PdpeSrc;
3579	PX86PML4E pPml4eSrc = &Pml4eSrc;
3580	Pml4eSrc.u = X86_PML4E_P \| X86_PML4E_RW \| X86_PML4E_US \| X86_PML4E_A;
3581	PdpeSrc.u = X86_PDPE_P \| X86_PDPE_RW \| X86_PDPE_US \| X86_PDPE_A;
3582	# endif
3583
3584	rc = pgmShwSyncLongModePDPtr(pVCpu, GCPtrPage, pPml4eSrc->u, PdpeSrc.u, &pPDDst);
3585	if (rc != VINF_SUCCESS)
3586	{
3587	pgmUnlock(pVM);
3588	AssertRC(rc);
3589	return rc;
3590	}
3591	Assert(pPDDst);
3592	pPdeDst = &pPDDst->a[iPDDst];
3593	# endif
3594
3595	if (!pPdeDst->n.u1Present)
3596	{
3597	rc = PGM_BTH_NAME(SyncPT)(pVCpu, iPDSrc, pPDSrc, GCPtrPage);
3598	if (rc != VINF_SUCCESS)
3599	{
3600	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
3601	pgmUnlock(pVM);
3602	AssertRC(rc);
3603	return rc;
3604	}
3605	}
3606
3607	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
3608	/* Check for dirty bit fault */
3609	rc = PGM_BTH_NAME(CheckDirtyPageFault)(pVCpu, uErr, pPdeDst, &pPDSrc->a[iPDSrc], GCPtrPage);
3610	if (rc == VINF_PGM_HANDLED_DIRTY_BIT_FAULT)
3611	Log(("PGMVerifyAccess: success (dirty)\n"));
3612	else
3613	# endif
3614	{
3615	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
3616	GSTPDE PdeSrc = pPDSrc->a[iPDSrc];
3617	# else
3618	GSTPDE PdeSrc;
3619	PdeSrc.u = 0; /* faked so we don't have to #ifdef everything */
3620	PdeSrc.n.u1Present = 1;
3621	PdeSrc.n.u1Write = 1;
3622	PdeSrc.n.u1Accessed = 1;
3623	PdeSrc.n.u1User = 1;
3624	# endif
3625
3626	Assert(rc != VINF_EM_RAW_GUEST_TRAP);
3627	if (uErr & X86_TRAP_PF_US)
3628	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,PageOutOfSyncUser));
3629	else /* supervisor */
3630	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,PageOutOfSyncSupervisor));
3631
3632	rc = PGM_BTH_NAME(SyncPage)(pVCpu, PdeSrc, GCPtrPage, 1, 0);
3633	if (RT_SUCCESS(rc))
3634	{
3635	/* Page was successfully synced */
3636	Log2(("PGMVerifyAccess: success (sync)\n"));
3637	rc = VINF_SUCCESS;
3638	}
3639	else
3640	{
3641	Log(("PGMVerifyAccess: access violation for %RGv rc=%Rrc\n", GCPtrPage, rc));
3642	rc = VINF_EM_RAW_GUEST_TRAP;
3643	}
3644	}
3645	PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdeDst);
3646	pgmUnlock(pVM);
3647	return rc;
3648
3649	#else /* PGM_SHW_TYPE == PGM_TYPE_EPT \|\| PGM_SHW_TYPE == PGM_TYPE_NESTED */
3650
3651	AssertReleaseMsgFailed(("Shw=%d Gst=%d is not implemented!\n", PGM_GST_TYPE, PGM_SHW_TYPE));
3652	return VERR_PGM_NOT_USED_IN_MODE;
3653	#endif /* PGM_SHW_TYPE == PGM_TYPE_EPT \|\| PGM_SHW_TYPE == PGM_TYPE_NESTED */
3654	}
3655
3656
3657	/**
3658	* Syncs the paging hierarchy starting at CR3.
3659	*
3660	* @returns VBox status code, no specials.
3661	* @param pVCpu Pointer to the VMCPU.
3662	* @param cr0 Guest context CR0 register.
3663	* @param cr3 Guest context CR3 register. Not subjected to the A20
3664	* mask.
3665	* @param cr4 Guest context CR4 register.
3666	* @param fGlobal Including global page directories or not
3667	*/
3668	PGM_BTH_DECL(int, SyncCR3)(PVMCPU pVCpu, uint64_t cr0, uint64_t cr3, uint64_t cr4, bool fGlobal)
3669	{
3670	PVM pVM = pVCpu->CTX_SUFF(pVM); NOREF(pVM);
3671	NOREF(cr0); NOREF(cr3); NOREF(cr4); NOREF(fGlobal);
3672
3673	LogFlow(("SyncCR3 FF=%d fGlobal=%d\n", !!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3), fGlobal));
3674
3675	#if PGM_SHW_TYPE != PGM_TYPE_NESTED && PGM_SHW_TYPE != PGM_TYPE_EPT
3676
3677	pgmLock(pVM);
3678
3679	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
3680	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3681	if (pPool->cDirtyPages)
3682	pgmPoolResetDirtyPages(pVM);
3683	# endif
3684
3685	/*
3686	* Update page access handlers.
3687	* The virtual are always flushed, while the physical are only on demand.
3688	* WARNING: We are incorrectly not doing global flushing on Virtual Handler updates. We'll
3689	* have to look into that later because it will have a bad influence on the performance.
3690	* @note SvL: There's no need for that. Just invalidate the virtual range(s).
3691	* bird: Yes, but that won't work for aliases.
3692	*/
3693	/** @todo this MUST go away. See @bugref{1557}. */
3694	STAM_PROFILE_START(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncCR3Handlers), h);
3695	PGM_GST_NAME(HandlerVirtualUpdate)(pVM, cr4);
3696	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncCR3Handlers), h);
3697	pgmUnlock(pVM);
3698	#endif /* !NESTED && !EPT */
3699
3700	#if PGM_SHW_TYPE == PGM_TYPE_NESTED \|\| PGM_SHW_TYPE == PGM_TYPE_EPT
3701	/*
3702	* Nested / EPT - almost no work.
3703	*/
3704	Assert(!pgmMapAreMappingsEnabled(pVM));
3705	return VINF_SUCCESS;
3706
3707	#elif PGM_SHW_TYPE == PGM_TYPE_AMD64
3708	/*
3709	* AMD64 (Shw & Gst) - No need to check all paging levels; we zero
3710	* out the shadow parts when the guest modifies its tables.
3711	*/
3712	Assert(!pgmMapAreMappingsEnabled(pVM));
3713	return VINF_SUCCESS;
3714
3715	#else /* PGM_SHW_TYPE != PGM_TYPE_NESTED && PGM_SHW_TYPE != PGM_TYPE_EPT && PGM_SHW_TYPE != PGM_TYPE_AMD64 */
3716
3717	# ifndef PGM_WITHOUT_MAPPINGS
3718	/*
3719	* Check for and resolve conflicts with our guest mappings if they
3720	* are enabled and not fixed.
3721	*/
3722	if (pgmMapAreMappingsFloating(pVM))
3723	{
3724	int rc = pgmMapResolveConflicts(pVM);
3725	Assert(rc == VINF_SUCCESS \|\| rc == VINF_PGM_SYNC_CR3);
3726	if (rc == VINF_PGM_SYNC_CR3)
3727	{
3728	LogFlow(("SyncCR3: detected conflict -> VINF_PGM_SYNC_CR3\n"));
3729	return VINF_PGM_SYNC_CR3;
3730	}
3731	}
3732	# else
3733	Assert(!pgmMapAreMappingsEnabled(pVM));
3734	# endif
3735	return VINF_SUCCESS;
3736	#endif /* PGM_SHW_TYPE != PGM_TYPE_NESTED && PGM_SHW_TYPE != PGM_TYPE_EPT && PGM_SHW_TYPE != PGM_TYPE_AMD64 */
3737	}
3738
3739
3740
3741
3742	#ifdef VBOX_STRICT
3743	# ifdef IN_RC
3744	# undef AssertMsgFailed
3745	# define AssertMsgFailed Log
3746	# endif
3747
3748	/**
3749	* Checks that the shadow page table is in sync with the guest one.
3750	*
3751	* @returns The number of errors.
3752	* @param pVM The virtual machine.
3753	* @param pVCpu Pointer to the VMCPU.
3754	* @param cr3 Guest context CR3 register.
3755	* @param cr4 Guest context CR4 register.
3756	* @param GCPtr Where to start. Defaults to 0.
3757	* @param cb How much to check. Defaults to everything.
3758	*/
3759	PGM_BTH_DECL(unsigned, AssertCR3)(PVMCPU pVCpu, uint64_t cr3, uint64_t cr4, RTGCPTR GCPtr, RTGCPTR cb)
3760	{
3761	NOREF(pVCpu); NOREF(cr3); NOREF(cr4); NOREF(GCPtr); NOREF(cb);
3762	#if PGM_SHW_TYPE == PGM_TYPE_NESTED \|\| PGM_SHW_TYPE == PGM_TYPE_EPT
3763	return 0;
3764	#else
3765	unsigned cErrors = 0;
3766	PVM pVM = pVCpu->CTX_SUFF(pVM);
3767	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
3768
3769	# if PGM_GST_TYPE == PGM_TYPE_PAE
3770	/** @todo currently broken; crashes below somewhere */
3771	AssertFailed();
3772	# endif
3773
3774	# if PGM_GST_TYPE == PGM_TYPE_32BIT \
3775	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
3776	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64
3777
3778	bool fBigPagesSupported = GST_IS_PSE_ACTIVE(pVCpu);
3779	PPGMCPU pPGM = &pVCpu->pgm.s;
3780	RTGCPHYS GCPhysGst; /* page address derived from the guest page tables. */
3781	RTHCPHYS HCPhysShw; /* page address derived from the shadow page tables. */
3782	# ifndef IN_RING0
3783	RTHCPHYS HCPhys; /* general usage. */
3784	# endif
3785	int rc;
3786
3787	/*
3788	* Check that the Guest CR3 and all its mappings are correct.
3789	*/
3790	AssertMsgReturn(pPGM->GCPhysCR3 == PGM_A20_APPLY(pVCpu, cr3 & GST_CR3_PAGE_MASK),
3791	("Invalid GCPhysCR3=%RGp cr3=%RGp\n", pPGM->GCPhysCR3, (RTGCPHYS)cr3),
3792	false);
3793	# if !defined(IN_RING0) && PGM_GST_TYPE != PGM_TYPE_AMD64
3794	# if PGM_GST_TYPE == PGM_TYPE_32BIT
3795	rc = PGMShwGetPage(pVCpu, (RTRCUINTPTR)pPGM->pGst32BitPdRC, NULL, &HCPhysShw);
3796	# else
3797	rc = PGMShwGetPage(pVCpu, (RTRCUINTPTR)pPGM->pGstPaePdptRC, NULL, &HCPhysShw);
3798	# endif
3799	AssertRCReturn(rc, 1);
3800	HCPhys = NIL_RTHCPHYS;
3801	rc = pgmRamGCPhys2HCPhys(pVM, PGM_A20_APPLY(pVCpu, cr3 & GST_CR3_PAGE_MASK), &HCPhys);
3802	AssertMsgReturn(HCPhys == HCPhysShw, ("HCPhys=%RHp HCPhyswShw=%RHp (cr3)\n", HCPhys, HCPhysShw), false);
3803	# if PGM_GST_TYPE == PGM_TYPE_32BIT && defined(IN_RING3)
3804	pgmGstGet32bitPDPtr(pVCpu);
3805	RTGCPHYS GCPhys;
3806	rc = PGMR3DbgR3Ptr2GCPhys(pVM->pUVM, pPGM->pGst32BitPdR3, &GCPhys);
3807	AssertRCReturn(rc, 1);
3808	AssertMsgReturn(PGM_A20_APPLY(pVCpu, cr3 & GST_CR3_PAGE_MASK) == GCPhys, ("GCPhys=%RGp cr3=%RGp\n", GCPhys, (RTGCPHYS)cr3), false);
3809	# endif
3810	# endif /* !IN_RING0 */
3811
3812	/*
3813	* Get and check the Shadow CR3.
3814	*/
3815	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
3816	unsigned cPDEs = X86_PG_ENTRIES;
3817	unsigned cIncrement = X86_PG_ENTRIES * PAGE_SIZE;
3818	# elif PGM_SHW_TYPE == PGM_TYPE_PAE
3819	# if PGM_GST_TYPE == PGM_TYPE_32BIT
3820	unsigned cPDEs = X86_PG_PAE_ENTRIES * 4; /* treat it as a 2048 entry table. */
3821	# else
3822	unsigned cPDEs = X86_PG_PAE_ENTRIES;
3823	# endif
3824	unsigned cIncrement = X86_PG_PAE_ENTRIES * PAGE_SIZE;
3825	# elif PGM_SHW_TYPE == PGM_TYPE_AMD64
3826	unsigned cPDEs = X86_PG_PAE_ENTRIES;
3827	unsigned cIncrement = X86_PG_PAE_ENTRIES * PAGE_SIZE;
3828	# endif
3829	if (cb != ~(RTGCPTR)0)
3830	cPDEs = RT_MIN(cb >> SHW_PD_SHIFT, 1);
3831
3832	/** @todo call the other two PGMAssert() functions. /
3833
3834	# if PGM_GST_TYPE == PGM_TYPE_AMD64
3835	unsigned iPml4 = (GCPtr >> X86_PML4_SHIFT) & X86_PML4_MASK;
3836
3837	for (; iPml4 < X86_PG_PAE_ENTRIES; iPml4++)
3838	{
3839	PPGMPOOLPAGE pShwPdpt = NULL;
3840	PX86PML4E pPml4eSrc;
3841	PX86PML4E pPml4eDst;
3842	RTGCPHYS GCPhysPdptSrc;
3843
3844	pPml4eSrc = pgmGstGetLongModePML4EPtr(pVCpu, iPml4);
3845	pPml4eDst = pgmShwGetLongModePML4EPtr(pVCpu, iPml4);
3846
3847	/* Fetch the pgm pool shadow descriptor if the shadow pml4e is present. */
3848	if (!pPml4eDst->n.u1Present)
3849	{
3850	GCPtr += _2M * UINT64_C(512) * UINT64_C(512);
3851	continue;
3852	}
3853
3854	pShwPdpt = pgmPoolGetPage(pPool, pPml4eDst->u & X86_PML4E_PG_MASK);
3855	GCPhysPdptSrc = PGM_A20_APPLY(pVCpu, pPml4eSrc->u & X86_PML4E_PG_MASK);
3856
3857	if (pPml4eSrc->n.u1Present != pPml4eDst->n.u1Present)
3858	{
3859	AssertMsgFailed(("Present bit doesn't match! pPml4eDst.u=%#RX64 pPml4eSrc.u=%RX64\n", pPml4eDst->u, pPml4eSrc->u));
3860	GCPtr += _2M * UINT64_C(512) * UINT64_C(512);
3861	cErrors++;
3862	continue;
3863	}
3864
3865	if (GCPhysPdptSrc != pShwPdpt->GCPhys)
3866	{
3867	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));
3868	GCPtr += _2M * UINT64_C(512) * UINT64_C(512);
3869	cErrors++;
3870	continue;
3871	}
3872
3873	if ( pPml4eDst->n.u1User != pPml4eSrc->n.u1User
3874	\|\| pPml4eDst->n.u1Write != pPml4eSrc->n.u1Write
3875	\|\| pPml4eDst->n.u1NoExecute != pPml4eSrc->n.u1NoExecute)
3876	{
3877	AssertMsgFailed(("User/Write/NoExec bits don't match! pPml4eDst.u=%#RX64 pPml4eSrc.u=%RX64\n", pPml4eDst->u, pPml4eSrc->u));
3878	GCPtr += _2M * UINT64_C(512) * UINT64_C(512);
3879	cErrors++;
3880	continue;
3881	}
3882	# else /* PGM_GST_TYPE != PGM_TYPE_AMD64 */
3883	{
3884	# endif /* PGM_GST_TYPE != PGM_TYPE_AMD64 */
3885
3886	# if PGM_GST_TYPE == PGM_TYPE_AMD64 \|\| PGM_GST_TYPE == PGM_TYPE_PAE
3887	/*
3888	* Check the PDPTEs too.
3889	*/
3890	unsigned iPdpt = (GCPtr >> SHW_PDPT_SHIFT) & SHW_PDPT_MASK;
3891
3892	for (;iPdpt <= SHW_PDPT_MASK; iPdpt++)
3893	{
3894	unsigned iPDSrc = 0; /* initialized to shut up gcc */
3895	PPGMPOOLPAGE pShwPde = NULL;
3896	PX86PDPE pPdpeDst;
3897	RTGCPHYS GCPhysPdeSrc;
3898	X86PDPE PdpeSrc;
3899	PdpeSrc.u = 0; /* initialized to shut up gcc 4.5 */
3900	# if PGM_GST_TYPE == PGM_TYPE_PAE
3901	PGSTPD pPDSrc = pgmGstGetPaePDPtr(pVCpu, GCPtr, &iPDSrc, &PdpeSrc);
3902	PX86PDPT pPdptDst = pgmShwGetPaePDPTPtr(pVCpu);
3903	# else
3904	PX86PML4E pPml4eSrcIgn;
3905	PX86PDPT pPdptDst;
3906	PX86PDPAE pPDDst;
3907	PGSTPD pPDSrc = pgmGstGetLongModePDPtr(pVCpu, GCPtr, &pPml4eSrcIgn, &PdpeSrc, &iPDSrc);
3908
3909	rc = pgmShwGetLongModePDPtr(pVCpu, GCPtr, NULL, &pPdptDst, &pPDDst);
3910	if (rc != VINF_SUCCESS)
3911	{
3912	AssertMsg(rc == VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT, ("Unexpected rc=%Rrc\n", rc));
3913	GCPtr += 512 * _2M;
3914	continue; /* next PDPTE */
3915	}
3916	Assert(pPDDst);
3917	# endif
3918	Assert(iPDSrc == 0);
3919
3920	pPdpeDst = &pPdptDst->a[iPdpt];
3921
3922	if (!pPdpeDst->n.u1Present)
3923	{
3924	GCPtr += 512 * _2M;
3925	continue; /* next PDPTE */
3926	}
3927
3928	pShwPde = pgmPoolGetPage(pPool, pPdpeDst->u & X86_PDPE_PG_MASK);
3929	GCPhysPdeSrc = PGM_A20_APPLY(pVCpu, PdpeSrc.u & X86_PDPE_PG_MASK);
3930
3931	if (pPdpeDst->n.u1Present != PdpeSrc.n.u1Present)
3932	{
3933	AssertMsgFailed(("Present bit doesn't match! pPdpeDst.u=%#RX64 pPdpeSrc.u=%RX64\n", pPdpeDst->u, PdpeSrc.u));
3934	GCPtr += 512 * _2M;
3935	cErrors++;
3936	continue;
3937	}
3938
3939	if (GCPhysPdeSrc != pShwPde->GCPhys)
3940	{
3941	# if PGM_GST_TYPE == PGM_TYPE_AMD64
3942	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));
3943	# else
3944	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));
3945	# endif
3946	GCPtr += 512 * _2M;
3947	cErrors++;
3948	continue;
3949	}
3950
3951	# if PGM_GST_TYPE == PGM_TYPE_AMD64
3952	if ( pPdpeDst->lm.u1User != PdpeSrc.lm.u1User
3953	\|\| pPdpeDst->lm.u1Write != PdpeSrc.lm.u1Write
3954	\|\| pPdpeDst->lm.u1NoExecute != PdpeSrc.lm.u1NoExecute)
3955	{
3956	AssertMsgFailed(("User/Write/NoExec bits don't match! pPdpeDst.u=%#RX64 pPdpeSrc.u=%RX64\n", pPdpeDst->u, PdpeSrc.u));
3957	GCPtr += 512 * _2M;
3958	cErrors++;
3959	continue;
3960	}
3961	# endif
3962
3963	# else /* PGM_GST_TYPE != PGM_TYPE_AMD64 && PGM_GST_TYPE != PGM_TYPE_PAE */
3964	{
3965	# endif /* PGM_GST_TYPE != PGM_TYPE_AMD64 && PGM_GST_TYPE != PGM_TYPE_PAE */
3966	# if PGM_GST_TYPE == PGM_TYPE_32BIT
3967	GSTPD const *pPDSrc = pgmGstGet32bitPDPtr(pVCpu);
3968	# if PGM_SHW_TYPE == PGM_TYPE_32BIT
3969	PCX86PD pPDDst = pgmShwGet32BitPDPtr(pVCpu);
3970	# endif
3971	# endif /* PGM_GST_TYPE == PGM_TYPE_32BIT */
3972	/*
3973	* Iterate the shadow page directory.
3974	*/
3975	GCPtr = (GCPtr >> SHW_PD_SHIFT) << SHW_PD_SHIFT;
3976	unsigned iPDDst = (GCPtr >> SHW_PD_SHIFT) & SHW_PD_MASK;
3977
3978	for (;
3979	iPDDst < cPDEs;
3980	iPDDst++, GCPtr += cIncrement)
3981	{
3982	# if PGM_SHW_TYPE == PGM_TYPE_PAE
3983	const SHWPDE PdeDst = *pgmShwGetPaePDEPtr(pVCpu, GCPtr);
3984	# else
3985	const SHWPDE PdeDst = pPDDst->a[iPDDst];
3986	# endif
3987	if (PdeDst.u & PGM_PDFLAGS_MAPPING)
3988	{
3989	Assert(pgmMapAreMappingsEnabled(pVM));
3990	if ((PdeDst.u & X86_PDE_AVL_MASK) != PGM_PDFLAGS_MAPPING)
3991	{
3992	AssertMsgFailed(("Mapping shall only have PGM_PDFLAGS_MAPPING set! PdeDst.u=%#RX64\n", (uint64_t)PdeDst.u));
3993	cErrors++;
3994	continue;
3995	}
3996	}
3997	else if ( (PdeDst.u & X86_PDE_P)
3998	\|\| ((PdeDst.u & (X86_PDE_P \| PGM_PDFLAGS_TRACK_DIRTY)) == (X86_PDE_P \| PGM_PDFLAGS_TRACK_DIRTY))
3999	)
4000	{
4001	HCPhysShw = PdeDst.u & SHW_PDE_PG_MASK;
4002	PPGMPOOLPAGE pPoolPage = pgmPoolGetPage(pPool, HCPhysShw);
4003	if (!pPoolPage)
4004	{
4005	AssertMsgFailed(("Invalid page table address %RHp at %RGv! PdeDst=%#RX64\n",
4006	HCPhysShw, GCPtr, (uint64_t)PdeDst.u));
4007	cErrors++;
4008	continue;
4009	}
4010	const SHWPT pPTDst = (const SHWPT )PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pPoolPage);
4011
4012	if (PdeDst.u & (X86_PDE4M_PWT \| X86_PDE4M_PCD))
4013	{
4014	AssertMsgFailed(("PDE flags PWT and/or PCD is set at %RGv! These flags are not virtualized! PdeDst=%#RX64\n",
4015	GCPtr, (uint64_t)PdeDst.u));
4016	cErrors++;
4017	}
4018
4019	if (PdeDst.u & (X86_PDE4M_G \| X86_PDE4M_D))
4020	{
4021	AssertMsgFailed(("4K PDE reserved flags at %RGv! PdeDst=%#RX64\n",
4022	GCPtr, (uint64_t)PdeDst.u));
4023	cErrors++;
4024	}
4025
4026	const GSTPDE PdeSrc = pPDSrc->a[(iPDDst >> (GST_PD_SHIFT - SHW_PD_SHIFT)) & GST_PD_MASK];
4027	if (!PdeSrc.n.u1Present)
4028	{
4029	AssertMsgFailed(("Guest PDE at %RGv is not present! PdeDst=%#RX64 PdeSrc=%#RX64\n",
4030	GCPtr, (uint64_t)PdeDst.u, (uint64_t)PdeSrc.u));
4031	cErrors++;
4032	continue;
4033	}
4034
4035	if ( !PdeSrc.b.u1Size
4036	\|\| !fBigPagesSupported)
4037	{
4038	GCPhysGst = GST_GET_PDE_GCPHYS(PdeSrc);
4039	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
4040	GCPhysGst = PGM_A20_APPLY(pVCpu, GCPhysGst \| ((iPDDst & 1) * (PAGE_SIZE / 2)));
4041	# endif
4042	}
4043	else
4044	{
4045	# if PGM_GST_TYPE == PGM_TYPE_32BIT
4046	if (PdeSrc.u & X86_PDE4M_PG_HIGH_MASK)
4047	{
4048	AssertMsgFailed(("Guest PDE at %RGv is using PSE36 or similar! PdeSrc=%#RX64\n",
4049	GCPtr, (uint64_t)PdeSrc.u));
4050	cErrors++;
4051	continue;
4052	}
4053	# endif
4054	GCPhysGst = GST_GET_BIG_PDE_GCPHYS(pVM, PdeSrc);
4055	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
4056	GCPhysGst = PGM_A20_APPLY(pVCpu, GCPhysGst \| (GCPtr & RT_BIT(X86_PAGE_2M_SHIFT)));
4057	# endif
4058	}
4059
4060	if ( pPoolPage->enmKind
4061	!= (!PdeSrc.b.u1Size \|\| !fBigPagesSupported ? BTH_PGMPOOLKIND_PT_FOR_PT : BTH_PGMPOOLKIND_PT_FOR_BIG))
4062	{
4063	AssertMsgFailed(("Invalid shadow page table kind %d at %RGv! PdeSrc=%#RX64\n",
4064	pPoolPage->enmKind, GCPtr, (uint64_t)PdeSrc.u));
4065	cErrors++;
4066	}
4067
4068	PPGMPAGE pPhysPage = pgmPhysGetPage(pVM, GCPhysGst);
4069	if (!pPhysPage)
4070	{
4071	AssertMsgFailed(("Cannot find guest physical address %RGp in the PDE at %RGv! PdeSrc=%#RX64\n",
4072	GCPhysGst, GCPtr, (uint64_t)PdeSrc.u));
4073	cErrors++;
4074	continue;
4075	}
4076
4077	if (GCPhysGst != pPoolPage->GCPhys)
4078	{
4079	AssertMsgFailed(("GCPhysGst=%RGp != pPage->GCPhys=%RGp at %RGv\n",
4080	GCPhysGst, pPoolPage->GCPhys, GCPtr));
4081	cErrors++;
4082	continue;
4083	}
4084
4085	if ( !PdeSrc.b.u1Size
4086	\|\| !fBigPagesSupported)
4087	{
4088	/*
4089	* Page Table.
4090	*/
4091	const GSTPT *pPTSrc;
4092	rc = PGM_GCPHYS_2_PTR_V2(pVM, pVCpu, PGM_A20_APPLY(pVCpu, GCPhysGst & ~(RTGCPHYS)(PAGE_SIZE - 1)),
4093	&pPTSrc);
4094	if (RT_FAILURE(rc))
4095	{
4096	AssertMsgFailed(("Cannot map/convert guest physical address %RGp in the PDE at %RGv! PdeSrc=%#RX64\n",
4097	GCPhysGst, GCPtr, (uint64_t)PdeSrc.u));
4098	cErrors++;
4099	continue;
4100	}
4101	if ( (PdeSrc.u & (X86_PDE_P \| X86_PDE_US \| X86_PDE_RW/* \| X86_PDE_A*/))
4102	!= (PdeDst.u & (X86_PDE_P \| X86_PDE_US \| X86_PDE_RW/* \| X86_PDE_A*/)))
4103	{
4104	/// @todo We get here a lot on out-of-sync CR3 entries. The access handler should zap them to avoid false alarms here!
4105	// (This problem will go away when/if we shadow multiple CR3s.)
4106	AssertMsgFailed(("4K PDE flags mismatch at %RGv! PdeSrc=%#RX64 PdeDst=%#RX64\n",
4107	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4108	cErrors++;
4109	continue;
4110	}
4111	if (PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY)
4112	{
4113	AssertMsgFailed(("4K PDEs cannot have PGM_PDFLAGS_TRACK_DIRTY set! GCPtr=%RGv PdeDst=%#RX64\n",
4114	GCPtr, (uint64_t)PdeDst.u));
4115	cErrors++;
4116	continue;
4117	}
4118
4119	/* iterate the page table. */
4120	# if PGM_SHW_TYPE == PGM_TYPE_PAE && PGM_GST_TYPE == PGM_TYPE_32BIT
4121	/* Select the right PDE as we're emulating a 4kb page table with 2 shadow page tables. */
4122	const unsigned offPTSrc = ((GCPtr >> SHW_PD_SHIFT) & 1) * 512;
4123	# else
4124	const unsigned offPTSrc = 0;
4125	# endif
4126	for (unsigned iPT = 0, off = 0;
4127	iPT < RT_ELEMENTS(pPTDst->a);
4128	iPT++, off += PAGE_SIZE)
4129	{
4130	const SHWPTE PteDst = pPTDst->a[iPT];
4131
4132	/* skip not-present and dirty tracked entries. */
4133	if (!(SHW_PTE_GET_U(PteDst) & (X86_PTE_P \| PGM_PTFLAGS_TRACK_DIRTY))) /** @todo deal with ALL handlers and CSAM !P pages! */
4134	continue;
4135	Assert(SHW_PTE_IS_P(PteDst));
4136
4137	const GSTPTE PteSrc = pPTSrc->a[iPT + offPTSrc];
4138	if (!PteSrc.n.u1Present)
4139	{
4140	# ifdef IN_RING3
4141	PGMAssertHandlerAndFlagsInSync(pVM);
4142	DBGFR3PagingDumpEx(pVM->pUVM, pVCpu->idCpu, DBGFPGDMP_FLAGS_CURRENT_CR3 \| DBGFPGDMP_FLAGS_CURRENT_MODE
4143	\| DBGFPGDMP_FLAGS_GUEST \| DBGFPGDMP_FLAGS_HEADER \| DBGFPGDMP_FLAGS_PRINT_CR3,
4144	0, 0, UINT64_MAX, 99, NULL);
4145	# endif
4146	AssertMsgFailed(("Out of sync (!P) PTE at %RGv! PteSrc=%#RX64 PteDst=%#RX64 pPTSrc=%RGv iPTSrc=%x PdeSrc=%x physpte=%RGp\n",
4147	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst), pPTSrc, iPT + offPTSrc, PdeSrc.au32[0],
4148	(uint64_t)GST_GET_PDE_GCPHYS(PdeSrc) + (iPT + offPTSrc) * sizeof(PteSrc)));
4149	cErrors++;
4150	continue;
4151	}
4152
4153	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;
4154	# if 1 /** @todo sync accessed bit properly... */
4155	fIgnoreFlags \|= X86_PTE_A;
4156	# endif
4157
4158	/* match the physical addresses */
4159	HCPhysShw = SHW_PTE_GET_HCPHYS(PteDst);
4160	GCPhysGst = GST_GET_PTE_GCPHYS(PteSrc);
4161
4162	# ifdef IN_RING3
4163	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysGst, &HCPhys);
4164	if (RT_FAILURE(rc))
4165	{
4166	if (HCPhysShw != MMR3PageDummyHCPhys(pVM)) /** @todo this is wrong. */
4167	{
4168	AssertMsgFailed(("Cannot find guest physical address %RGp at %RGv! PteSrc=%#RX64 PteDst=%#RX64\n",
4169	GCPhysGst, GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4170	cErrors++;
4171	continue;
4172	}
4173	}
4174	else if (HCPhysShw != (HCPhys & SHW_PTE_PG_MASK))
4175	{
4176	AssertMsgFailed(("Out of sync (phys) at %RGv! HCPhysShw=%RHp HCPhys=%RHp GCPhysGst=%RGp PteSrc=%#RX64 PteDst=%#RX64\n",
4177	GCPtr + off, HCPhysShw, HCPhys, GCPhysGst, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4178	cErrors++;
4179	continue;
4180	}
4181	# endif
4182
4183	pPhysPage = pgmPhysGetPage(pVM, GCPhysGst);
4184	if (!pPhysPage)
4185	{
4186	# ifdef IN_RING3 /** @todo make MMR3PageDummyHCPhys an 'All' function! */
4187	if (HCPhysShw != MMR3PageDummyHCPhys(pVM)) /** @todo this is wrong. */
4188	{
4189	AssertMsgFailed(("Cannot find guest physical address %RGp at %RGv! PteSrc=%#RX64 PteDst=%#RX64\n",
4190	GCPhysGst, GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4191	cErrors++;
4192	continue;
4193	}
4194	# endif
4195	if (SHW_PTE_IS_RW(PteDst))
4196	{
4197	AssertMsgFailed(("Invalid guest page at %RGv is writable! GCPhysGst=%RGp PteSrc=%#RX64 PteDst=%#RX64\n",
4198	GCPtr + off, GCPhysGst, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4199	cErrors++;
4200	}
4201	fIgnoreFlags \|= X86_PTE_RW;
4202	}
4203	else if (HCPhysShw != PGM_PAGE_GET_HCPHYS(pPhysPage))
4204	{
4205	AssertMsgFailed(("Out of sync (phys) at %RGv! HCPhysShw=%RHp pPhysPage:%R[pgmpage] GCPhysGst=%RGp PteSrc=%#RX64 PteDst=%#RX64\n",
4206	GCPtr + off, HCPhysShw, pPhysPage, GCPhysGst, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4207	cErrors++;
4208	continue;
4209	}
4210
4211	/* flags */
4212	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPhysPage))
4213	{
4214	if (!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPhysPage))
4215	{
4216	if (SHW_PTE_IS_RW(PteDst))
4217	{
4218	AssertMsgFailed(("WRITE access flagged at %RGv but the page is writable! pPhysPage=%R[pgmpage] PteSrc=%#RX64 PteDst=%#RX64\n",
4219	GCPtr + off, pPhysPage, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4220	cErrors++;
4221	continue;
4222	}
4223	fIgnoreFlags \|= X86_PTE_RW;
4224	}
4225	else
4226	{
4227	if ( SHW_PTE_IS_P(PteDst)
4228	# if PGM_SHW_TYPE == PGM_TYPE_EPT \|\| PGM_SHW_TYPE == PGM_TYPE_PAE \|\| PGM_SHW_TYPE == PGM_TYPE_AMD64
4229	&& !PGM_PAGE_IS_MMIO(pPhysPage)
4230	# endif
4231	)
4232	{
4233	AssertMsgFailed(("ALL access flagged at %RGv but the page is present! pPhysPage=%R[pgmpage] PteSrc=%#RX64 PteDst=%#RX64\n",
4234	GCPtr + off, pPhysPage, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4235	cErrors++;
4236	continue;
4237	}
4238	fIgnoreFlags \|= X86_PTE_P;
4239	}
4240	}
4241	else
4242	{
4243	if (!PteSrc.n.u1Dirty && PteSrc.n.u1Write)
4244	{
4245	if (SHW_PTE_IS_RW(PteDst))
4246	{
4247	AssertMsgFailed(("!DIRTY page at %RGv is writable! PteSrc=%#RX64 PteDst=%#RX64\n",
4248	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4249	cErrors++;
4250	continue;
4251	}
4252	if (!SHW_PTE_IS_TRACK_DIRTY(PteDst))
4253	{
4254	AssertMsgFailed(("!DIRTY page at %RGv is not marked TRACK_DIRTY! PteSrc=%#RX64 PteDst=%#RX64\n",
4255	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4256	cErrors++;
4257	continue;
4258	}
4259	if (SHW_PTE_IS_D(PteDst))
4260	{
4261	AssertMsgFailed(("!DIRTY page at %RGv is marked DIRTY! PteSrc=%#RX64 PteDst=%#RX64\n",
4262	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4263	cErrors++;
4264	}
4265	# if 0 /** @todo sync access bit properly... */
4266	if (PteDst.n.u1Accessed != PteSrc.n.u1Accessed)
4267	{
4268	AssertMsgFailed(("!DIRTY page at %RGv is has mismatching accessed bit! PteSrc=%#RX64 PteDst=%#RX64\n",
4269	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4270	cErrors++;
4271	}
4272	fIgnoreFlags \|= X86_PTE_RW;
4273	# else
4274	fIgnoreFlags \|= X86_PTE_RW \| X86_PTE_A;
4275	# endif
4276	}
4277	else if (SHW_PTE_IS_TRACK_DIRTY(PteDst))
4278	{
4279	/* access bit emulation (not implemented). */
4280	if (PteSrc.n.u1Accessed \|\| SHW_PTE_IS_P(PteDst))
4281	{
4282	AssertMsgFailed(("PGM_PTFLAGS_TRACK_DIRTY set at %RGv but no accessed bit emulation! PteSrc=%#RX64 PteDst=%#RX64\n",
4283	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4284	cErrors++;
4285	continue;
4286	}
4287	if (!SHW_PTE_IS_A(PteDst))
4288	{
4289	AssertMsgFailed(("!ACCESSED page at %RGv is has the accessed bit set! PteSrc=%#RX64 PteDst=%#RX64\n",
4290	GCPtr + off, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4291	cErrors++;
4292	}
4293	fIgnoreFlags \|= X86_PTE_P;
4294	}
4295	# ifdef DEBUG_sandervl
4296	fIgnoreFlags \|= X86_PTE_D \| X86_PTE_A;
4297	# endif
4298	}
4299
4300	if ( (PteSrc.u & ~fIgnoreFlags) != (SHW_PTE_GET_U(PteDst) & ~fIgnoreFlags)
4301	&& (PteSrc.u & ~(fIgnoreFlags \| X86_PTE_RW)) != (SHW_PTE_GET_U(PteDst) & ~fIgnoreFlags)
4302	)
4303	{
4304	AssertMsgFailed(("Flags mismatch at %RGv! %#RX64 != %#RX64 fIgnoreFlags=%#RX64 PteSrc=%#RX64 PteDst=%#RX64\n",
4305	GCPtr + off, (uint64_t)PteSrc.u & ~fIgnoreFlags, SHW_PTE_LOG64(PteDst) & ~fIgnoreFlags,
4306	fIgnoreFlags, (uint64_t)PteSrc.u, SHW_PTE_LOG64(PteDst)));
4307	cErrors++;
4308	continue;
4309	}
4310	} /* foreach PTE */
4311	}
4312	else
4313	{
4314	/*
4315	* Big Page.
4316	*/
4317	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;
4318	if (!PdeSrc.b.u1Dirty && PdeSrc.b.u1Write)
4319	{
4320	if (PdeDst.n.u1Write)
4321	{
4322	AssertMsgFailed(("!DIRTY page at %RGv is writable! PdeSrc=%#RX64 PdeDst=%#RX64\n",
4323	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4324	cErrors++;
4325	continue;
4326	}
4327	if (!(PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY))
4328	{
4329	AssertMsgFailed(("!DIRTY page at %RGv is not marked TRACK_DIRTY! PteSrc=%#RX64 PteDst=%#RX64\n",
4330	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4331	cErrors++;
4332	continue;
4333	}
4334	# if 0 /** @todo sync access bit properly... */
4335	if (PdeDst.n.u1Accessed != PdeSrc.b.u1Accessed)
4336	{
4337	AssertMsgFailed(("!DIRTY page at %RGv is has mismatching accessed bit! PteSrc=%#RX64 PteDst=%#RX64\n",
4338	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4339	cErrors++;
4340	}
4341	fIgnoreFlags \|= X86_PTE_RW;
4342	# else
4343	fIgnoreFlags \|= X86_PTE_RW \| X86_PTE_A;
4344	# endif
4345	}
4346	else if (PdeDst.u & PGM_PDFLAGS_TRACK_DIRTY)
4347	{
4348	/* access bit emulation (not implemented). */
4349	if (PdeSrc.b.u1Accessed \|\| PdeDst.n.u1Present)
4350	{
4351	AssertMsgFailed(("PGM_PDFLAGS_TRACK_DIRTY set at %RGv but no accessed bit emulation! PdeSrc=%#RX64 PdeDst=%#RX64\n",
4352	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4353	cErrors++;
4354	continue;
4355	}
4356	if (!PdeDst.n.u1Accessed)
4357	{
4358	AssertMsgFailed(("!ACCESSED page at %RGv is has the accessed bit set! PdeSrc=%#RX64 PdeDst=%#RX64\n",
4359	GCPtr, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4360	cErrors++;
4361	}
4362	fIgnoreFlags \|= X86_PTE_P;
4363	}
4364
4365	if ((PdeSrc.u & ~fIgnoreFlags) != (PdeDst.u & ~fIgnoreFlags))
4366	{
4367	AssertMsgFailed(("Flags mismatch (B) at %RGv! %#RX64 != %#RX64 fIgnoreFlags=%#RX64 PdeSrc=%#RX64 PdeDst=%#RX64\n",
4368	GCPtr, (uint64_t)PdeSrc.u & ~fIgnoreFlags, (uint64_t)PdeDst.u & ~fIgnoreFlags,
4369	fIgnoreFlags, (uint64_t)PdeSrc.u, (uint64_t)PdeDst.u));
4370	cErrors++;
4371	}
4372
4373	/* iterate the page table. */
4374	for (unsigned iPT = 0, off = 0;
4375	iPT < RT_ELEMENTS(pPTDst->a);
4376	iPT++, off += PAGE_SIZE, GCPhysGst = PGM_A20_APPLY(pVCpu, GCPhysGst + PAGE_SIZE))
4377	{
4378	const SHWPTE PteDst = pPTDst->a[iPT];
4379
4380	if (SHW_PTE_IS_TRACK_DIRTY(PteDst))
4381	{
4382	AssertMsgFailed(("The PTE at %RGv emulating a 2/4M page is marked TRACK_DIRTY! PdeSrc=%#RX64 PteDst=%#RX64\n",
4383	GCPtr + off, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
4384	cErrors++;
4385	}
4386
4387	/* skip not-present entries. */
4388	if (!SHW_PTE_IS_P(PteDst)) /** @todo deal with ALL handlers and CSAM !P pages! */
4389	continue;
4390
4391	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;
4392
4393	/* match the physical addresses */
4394	HCPhysShw = SHW_PTE_GET_HCPHYS(PteDst);
4395
4396	# ifdef IN_RING3
4397	rc = PGMPhysGCPhys2HCPhys(pVM, GCPhysGst, &HCPhys);
4398	if (RT_FAILURE(rc))
4399	{
4400	if (HCPhysShw != MMR3PageDummyHCPhys(pVM)) /** @todo this is wrong. */
4401	{
4402	AssertMsgFailed(("Cannot find guest physical address %RGp at %RGv! PdeSrc=%#RX64 PteDst=%#RX64\n",
4403	GCPhysGst, GCPtr + off, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
4404	cErrors++;
4405	}
4406	}
4407	else if (HCPhysShw != (HCPhys & X86_PTE_PAE_PG_MASK))
4408	{
4409	AssertMsgFailed(("Out of sync (phys) at %RGv! HCPhysShw=%RHp HCPhys=%RHp GCPhysGst=%RGp PdeSrc=%#RX64 PteDst=%#RX64\n",
4410	GCPtr + off, HCPhysShw, HCPhys, GCPhysGst, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
4411	cErrors++;
4412	continue;
4413	}
4414	# endif
4415	pPhysPage = pgmPhysGetPage(pVM, GCPhysGst);
4416	if (!pPhysPage)
4417	{
4418	# ifdef IN_RING3 /** @todo make MMR3PageDummyHCPhys an 'All' function! */
4419	if (HCPhysShw != MMR3PageDummyHCPhys(pVM)) /** @todo this is wrong. */
4420	{
4421	AssertMsgFailed(("Cannot find guest physical address %RGp at %RGv! PdeSrc=%#RX64 PteDst=%#RX64\n",
4422	GCPhysGst, GCPtr + off, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
4423	cErrors++;
4424	continue;
4425	}
4426	# endif
4427	if (SHW_PTE_IS_RW(PteDst))
4428	{
4429	AssertMsgFailed(("Invalid guest page at %RGv is writable! GCPhysGst=%RGp PdeSrc=%#RX64 PteDst=%#RX64\n",
4430	GCPtr + off, GCPhysGst, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
4431	cErrors++;
4432	}
4433	fIgnoreFlags \|= X86_PTE_RW;
4434	}
4435	else if (HCPhysShw != PGM_PAGE_GET_HCPHYS(pPhysPage))
4436	{
4437	AssertMsgFailed(("Out of sync (phys) at %RGv! HCPhysShw=%RHp pPhysPage=%R[pgmpage] GCPhysGst=%RGp PdeSrc=%#RX64 PteDst=%#RX64\n",
4438	GCPtr + off, HCPhysShw, pPhysPage, GCPhysGst, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
4439	cErrors++;
4440	continue;
4441	}
4442
4443	/* flags */
4444	if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPhysPage))
4445	{
4446	if (!PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPhysPage))
4447	{
4448	if (PGM_PAGE_GET_HNDL_PHYS_STATE(pPhysPage) != PGM_PAGE_HNDL_PHYS_STATE_DISABLED)
4449	{
4450	if (SHW_PTE_IS_RW(PteDst))
4451	{
4452	AssertMsgFailed(("WRITE access flagged at %RGv but the page is writable! pPhysPage=%R[pgmpage] PdeSrc=%#RX64 PteDst=%#RX64\n",
4453	GCPtr + off, pPhysPage, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
4454	cErrors++;
4455	continue;
4456	}
4457	fIgnoreFlags \|= X86_PTE_RW;
4458	}
4459	}
4460	else
4461	{
4462	if ( SHW_PTE_IS_P(PteDst)
4463	# if PGM_SHW_TYPE == PGM_TYPE_EPT \|\| PGM_SHW_TYPE == PGM_TYPE_PAE \|\| PGM_SHW_TYPE == PGM_TYPE_AMD64
4464	&& !PGM_PAGE_IS_MMIO(pPhysPage)
4465	# endif
4466	)
4467	{
4468	AssertMsgFailed(("ALL access flagged at %RGv but the page is present! pPhysPage=%R[pgmpage] PdeSrc=%#RX64 PteDst=%#RX64\n",
4469	GCPtr + off, pPhysPage, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
4470	cErrors++;
4471	continue;
4472	}
4473	fIgnoreFlags \|= X86_PTE_P;
4474	}
4475	}
4476
4477	if ( (PdeSrc.u & ~fIgnoreFlags) != (SHW_PTE_GET_U(PteDst) & ~fIgnoreFlags)
4478	&& (PdeSrc.u & ~(fIgnoreFlags \| X86_PTE_RW)) != (SHW_PTE_GET_U(PteDst) & ~fIgnoreFlags) /* lazy phys handler dereg. */
4479	)
4480	{
4481	AssertMsgFailed(("Flags mismatch (BT) at %RGv! %#RX64 != %#RX64 fIgnoreFlags=%#RX64 PdeSrc=%#RX64 PteDst=%#RX64\n",
4482	GCPtr + off, (uint64_t)PdeSrc.u & ~fIgnoreFlags, SHW_PTE_LOG64(PteDst) & ~fIgnoreFlags,
4483	fIgnoreFlags, (uint64_t)PdeSrc.u, SHW_PTE_LOG64(PteDst)));
4484	cErrors++;
4485	continue;
4486	}
4487	} /* for each PTE */
4488	}
4489	}
4490	/* not present */
4491
4492	} /* for each PDE */
4493
4494	} /* for each PDPTE */
4495
4496	} /* for each PML4E */
4497
4498	# ifdef DEBUG
4499	if (cErrors)
4500	LogFlow(("AssertCR3: cErrors=%d\n", cErrors));
4501	# endif
4502	# endif /* GST is in {32BIT, PAE, AMD64} */
4503	return cErrors;
4504	#endif /* PGM_SHW_TYPE != PGM_TYPE_NESTED && PGM_SHW_TYPE != PGM_TYPE_EPT */
4505	}
4506	#endif /* VBOX_STRICT */
4507
4508
4509	/**
4510	* Sets up the CR3 for shadow paging
4511	*
4512	* @returns Strict VBox status code.
4513	* @retval VINF_SUCCESS.
4514	*
4515	* @param pVCpu Pointer to the VMCPU.
4516	* @param GCPhysCR3 The physical address in the CR3 register. (A20
4517	* mask already applied.)
4518	*/
4519	PGM_BTH_DECL(int, MapCR3)(PVMCPU pVCpu, RTGCPHYS GCPhysCR3)
4520	{
4521	PVM pVM = pVCpu->CTX_SUFF(pVM); NOREF(pVM);
4522
4523	/* Update guest paging info. */
4524	#if PGM_GST_TYPE == PGM_TYPE_32BIT \
4525	\|\| PGM_GST_TYPE == PGM_TYPE_PAE \
4526	\|\| PGM_GST_TYPE == PGM_TYPE_AMD64
4527
4528	LogFlow(("MapCR3: %RGp\n", GCPhysCR3));
4529	PGM_A20_ASSERT_MASKED(pVCpu, GCPhysCR3);
4530
4531	/*
4532	* Map the page CR3 points at.
4533	*/
4534	RTHCPTR HCPtrGuestCR3;
4535	RTHCPHYS HCPhysGuestCR3;
4536	pgmLock(pVM);
4537	PPGMPAGE pPageCR3 = pgmPhysGetPage(pVM, GCPhysCR3);
4538	AssertReturn(pPageCR3, VERR_PGM_INVALID_CR3_ADDR);
4539	HCPhysGuestCR3 = PGM_PAGE_GET_HCPHYS(pPageCR3);
4540	/** @todo this needs some reworking wrt. locking? */
4541	# if defined(IN_RC) \|\| defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0)
4542	HCPtrGuestCR3 = NIL_RTHCPTR;
4543	int rc = VINF_SUCCESS;
4544	# else
4545	int rc = pgmPhysGCPhys2CCPtrInternalDepr(pVM, pPageCR3, GCPhysCR3 & GST_CR3_PAGE_MASK, (void )&HCPtrGuestCR3); / @todo r=bird: This GCPhysCR3 masking isn't necessary. */
4546	# endif
4547	pgmUnlock(pVM);
4548	if (RT_SUCCESS(rc))
4549	{
4550	rc = PGMMap(pVM, (RTGCPTR)pVM->pgm.s.GCPtrCR3Mapping, HCPhysGuestCR3, PAGE_SIZE, 0);
4551	if (RT_SUCCESS(rc))
4552	{
4553	# ifdef IN_RC
4554	PGM_INVL_PG(pVCpu, pVM->pgm.s.GCPtrCR3Mapping);
4555	# endif
4556	# if PGM_GST_TYPE == PGM_TYPE_32BIT
4557	pVCpu->pgm.s.pGst32BitPdR3 = (R3PTRTYPE(PX86PD))HCPtrGuestCR3;
4558	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4559	pVCpu->pgm.s.pGst32BitPdR0 = (R0PTRTYPE(PX86PD))HCPtrGuestCR3;
4560	# endif
4561	pVCpu->pgm.s.pGst32BitPdRC = (RCPTRTYPE(PX86PD))(RTRCUINTPTR)pVM->pgm.s.GCPtrCR3Mapping;
4562
4563	# elif PGM_GST_TYPE == PGM_TYPE_PAE
4564	unsigned off = GCPhysCR3 & GST_CR3_PAGE_MASK & PAGE_OFFSET_MASK;
4565	pVCpu->pgm.s.pGstPaePdptR3 = (R3PTRTYPE(PX86PDPT))HCPtrGuestCR3;
4566	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4567	pVCpu->pgm.s.pGstPaePdptR0 = (R0PTRTYPE(PX86PDPT))HCPtrGuestCR3;
4568	# endif
4569	pVCpu->pgm.s.pGstPaePdptRC = (RCPTRTYPE(PX86PDPT))((RTRCUINTPTR)pVM->pgm.s.GCPtrCR3Mapping + off);
4570	LogFlow(("Cached mapping %RRv\n", pVCpu->pgm.s.pGstPaePdptRC));
4571
4572	/*
4573	* Map the 4 PDs too.
4574	*/
4575	PX86PDPT pGuestPDPT = pgmGstGetPaePDPTPtr(pVCpu);
4576	RTGCPTR GCPtr = pVM->pgm.s.GCPtrCR3Mapping + PAGE_SIZE;
4577	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++, GCPtr += PAGE_SIZE)
4578	{
4579	pVCpu->pgm.s.aGstPaePdpeRegs[i].u = pGuestPDPT->a[i].u;
4580	if (pGuestPDPT->a[i].n.u1Present)
4581	{
4582	RTHCPTR HCPtr;
4583	RTHCPHYS HCPhys;
4584	RTGCPHYS GCPhys = PGM_A20_APPLY(pVCpu, pGuestPDPT->a[i].u & X86_PDPE_PG_MASK);
4585	pgmLock(pVM);
4586	PPGMPAGE pPage = pgmPhysGetPage(pVM, GCPhys);
4587	AssertReturn(pPage, VERR_PGM_INVALID_PDPE_ADDR);
4588	HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
4589	# if defined(IN_RC) \|\| defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0)
4590	HCPtr = NIL_RTHCPTR;
4591	int rc2 = VINF_SUCCESS;
4592	# else
4593	int rc2 = pgmPhysGCPhys2CCPtrInternalDepr(pVM, pPage, GCPhys, (void **)&HCPtr);
4594	# endif
4595	pgmUnlock(pVM);
4596	if (RT_SUCCESS(rc2))
4597	{
4598	rc = PGMMap(pVM, GCPtr, HCPhys, PAGE_SIZE, 0);
4599	AssertRCReturn(rc, rc);
4600
4601	pVCpu->pgm.s.apGstPaePDsR3[i] = (R3PTRTYPE(PX86PDPAE))HCPtr;
4602	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4603	pVCpu->pgm.s.apGstPaePDsR0[i] = (R0PTRTYPE(PX86PDPAE))HCPtr;
4604	# endif
4605	pVCpu->pgm.s.apGstPaePDsRC[i] = (RCPTRTYPE(PX86PDPAE))(RTRCUINTPTR)GCPtr;
4606	pVCpu->pgm.s.aGCPhysGstPaePDs[i] = GCPhys;
4607	# ifdef IN_RC
4608	PGM_INVL_PG(pVCpu, GCPtr);
4609	# endif
4610	continue;
4611	}
4612	AssertMsgFailed(("pgmR3Gst32BitMapCR3: rc2=%d GCPhys=%RGp i=%d\n", rc2, GCPhys, i));
4613	}
4614
4615	pVCpu->pgm.s.apGstPaePDsR3[i] = 0;
4616	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4617	pVCpu->pgm.s.apGstPaePDsR0[i] = 0;
4618	# endif
4619	pVCpu->pgm.s.apGstPaePDsRC[i] = 0;
4620	pVCpu->pgm.s.aGCPhysGstPaePDs[i] = NIL_RTGCPHYS;
4621	# ifdef IN_RC
4622	PGM_INVL_PG(pVCpu, GCPtr); /** @todo this shouldn't be necessary? */
4623	# endif
4624	}
4625
4626	# elif PGM_GST_TYPE == PGM_TYPE_AMD64
4627	pVCpu->pgm.s.pGstAmd64Pml4R3 = (R3PTRTYPE(PX86PML4))HCPtrGuestCR3;
4628	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4629	pVCpu->pgm.s.pGstAmd64Pml4R0 = (R0PTRTYPE(PX86PML4))HCPtrGuestCR3;
4630	# endif
4631	# endif
4632	}
4633	else
4634	AssertMsgFailed(("rc=%Rrc GCPhysGuestPD=%RGp\n", rc, GCPhysCR3));
4635	}
4636	else
4637	AssertMsgFailed(("rc=%Rrc GCPhysGuestPD=%RGp\n", rc, GCPhysCR3));
4638
4639	#else /* prot/real stub */
4640	int rc = VINF_SUCCESS;
4641	#endif
4642
4643	/* Update shadow paging info for guest modes with paging (32, pae, 64). */
4644	# if ( ( PGM_SHW_TYPE == PGM_TYPE_32BIT \
4645	\|\| PGM_SHW_TYPE == PGM_TYPE_PAE \
4646	\|\| PGM_SHW_TYPE == PGM_TYPE_AMD64) \
4647	&& ( PGM_GST_TYPE != PGM_TYPE_REAL \
4648	&& PGM_GST_TYPE != PGM_TYPE_PROT))
4649
4650	Assert(!pVM->pgm.s.fNestedPaging);
4651	PGM_A20_ASSERT_MASKED(pVCpu, GCPhysCR3);
4652
4653	/*
4654	* Update the shadow root page as well since that's not fixed.
4655	*/
4656	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4657	PPGMPOOLPAGE pOldShwPageCR3 = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
4658	PPGMPOOLPAGE pNewShwPageCR3;
4659
4660	pgmLock(pVM);
4661
4662	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
4663	if (pPool->cDirtyPages)
4664	pgmPoolResetDirtyPages(pVM);
4665	# endif
4666
4667	Assert(!(GCPhysCR3 >> (PAGE_SHIFT + 32)));
4668	rc = pgmPoolAlloc(pVM, GCPhysCR3 & GST_CR3_PAGE_MASK, BTH_PGMPOOLKIND_ROOT, PGMPOOLACCESS_DONTCARE, PGM_A20_IS_ENABLED(pVCpu),
4669	NIL_PGMPOOL_IDX, UINT32_MAX, true /fLockPage/,
4670	&pNewShwPageCR3);
4671	AssertFatalRC(rc);
4672	rc = VINF_SUCCESS;
4673
4674	# ifdef IN_RC
4675	/*
4676	* WARNING! We can't deal with jumps to ring 3 in the code below as the
4677	* state will be inconsistent! Flush important things now while
4678	* we still can and then make sure there are no ring-3 calls.
4679	*/
4680	# ifdef VBOX_WITH_REM
4681	REMNotifyHandlerPhysicalFlushIfAlmostFull(pVM, pVCpu);
4682	# endif
4683	VMMRZCallRing3Disable(pVCpu);
4684	# endif
4685
4686	pVCpu->pgm.s.CTX_SUFF(pShwPageCR3) = pNewShwPageCR3;
4687	# ifdef IN_RING0
4688	pVCpu->pgm.s.pShwPageCR3R3 = MMHyperCCToR3(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4689	pVCpu->pgm.s.pShwPageCR3RC = MMHyperCCToRC(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4690	# elif defined(IN_RC)
4691	pVCpu->pgm.s.pShwPageCR3R3 = MMHyperCCToR3(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4692	pVCpu->pgm.s.pShwPageCR3R0 = MMHyperCCToR0(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4693	# else
4694	pVCpu->pgm.s.pShwPageCR3R0 = MMHyperCCToR0(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4695	pVCpu->pgm.s.pShwPageCR3RC = MMHyperCCToRC(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4696	# endif
4697
4698	# ifndef PGM_WITHOUT_MAPPINGS
4699	/*
4700	* Apply all hypervisor mappings to the new CR3.
4701	* Note that SyncCR3 will be executed in case CR3 is changed in a guest paging mode; this will
4702	* make sure we check for conflicts in the new CR3 root.
4703	*/
4704	# if PGM_WITH_PAGING(PGM_GST_TYPE, PGM_SHW_TYPE)
4705	Assert(VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL) \|\| VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
4706	# endif
4707	rc = pgmMapActivateCR3(pVM, pNewShwPageCR3);
4708	AssertRCReturn(rc, rc);
4709	# endif
4710
4711	/* Set the current hypervisor CR3. */
4712	CPUMSetHyperCR3(pVCpu, PGMGetHyperCR3(pVCpu));
4713	SELMShadowCR3Changed(pVM, pVCpu);
4714
4715	# ifdef IN_RC
4716	/* NOTE: The state is consistent again. */
4717	VMMRZCallRing3Enable(pVCpu);
4718	# endif
4719
4720	/* Clean up the old CR3 root. */
4721	if ( pOldShwPageCR3
4722	&& pOldShwPageCR3 != pNewShwPageCR3 /* @todo can happen due to incorrect syncing between REM & PGM; find the real cause */)
4723	{
4724	Assert(pOldShwPageCR3->enmKind != PGMPOOLKIND_FREE);
4725	# ifndef PGM_WITHOUT_MAPPINGS
4726	/* Remove the hypervisor mappings from the shadow page table. */
4727	pgmMapDeactivateCR3(pVM, pOldShwPageCR3);
4728	# endif
4729	/* Mark the page as unlocked; allow flushing again. */
4730	pgmPoolUnlockPage(pPool, pOldShwPageCR3);
4731
4732	pgmPoolFreeByPage(pPool, pOldShwPageCR3, NIL_PGMPOOL_IDX, UINT32_MAX);
4733	}
4734	pgmUnlock(pVM);
4735	# else
4736	NOREF(GCPhysCR3);
4737	# endif
4738
4739	return rc;
4740	}
4741
4742	/**
4743	* Unmaps the shadow CR3.
4744	*
4745	* @returns VBox status, no specials.
4746	* @param pVCpu Pointer to the VMCPU.
4747	*/
4748	PGM_BTH_DECL(int, UnmapCR3)(PVMCPU pVCpu)
4749	{
4750	LogFlow(("UnmapCR3\n"));
4751
4752	int rc = VINF_SUCCESS;
4753	PVM pVM = pVCpu->CTX_SUFF(pVM); NOREF(pVM);
4754
4755	/*
4756	* Update guest paging info.
4757	*/
4758	#if PGM_GST_TYPE == PGM_TYPE_32BIT
4759	pVCpu->pgm.s.pGst32BitPdR3 = 0;
4760	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4761	pVCpu->pgm.s.pGst32BitPdR0 = 0;
4762	# endif
4763	pVCpu->pgm.s.pGst32BitPdRC = 0;
4764
4765	#elif PGM_GST_TYPE == PGM_TYPE_PAE
4766	pVCpu->pgm.s.pGstPaePdptR3 = 0;
4767	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4768	pVCpu->pgm.s.pGstPaePdptR0 = 0;
4769	# endif
4770	pVCpu->pgm.s.pGstPaePdptRC = 0;
4771	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
4772	{
4773	pVCpu->pgm.s.apGstPaePDsR3[i] = 0;
4774	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4775	pVCpu->pgm.s.apGstPaePDsR0[i] = 0;
4776	# endif
4777	pVCpu->pgm.s.apGstPaePDsRC[i] = 0;
4778	pVCpu->pgm.s.aGCPhysGstPaePDs[i] = NIL_RTGCPHYS;
4779	}
4780
4781	#elif PGM_GST_TYPE == PGM_TYPE_AMD64
4782	pVCpu->pgm.s.pGstAmd64Pml4R3 = 0;
4783	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
4784	pVCpu->pgm.s.pGstAmd64Pml4R0 = 0;
4785	# endif
4786
4787	#else /* prot/real mode stub */
4788	/* nothing to do */
4789	#endif
4790
4791	#if !defined(IN_RC) /* In RC we rely on MapCR3 to do the shadow part for us at a safe time */
4792	/*
4793	* Update shadow paging info.
4794	*/
4795	# if ( ( PGM_SHW_TYPE == PGM_TYPE_32BIT \
4796	\|\| PGM_SHW_TYPE == PGM_TYPE_PAE \
4797	\|\| PGM_SHW_TYPE == PGM_TYPE_AMD64))
4798
4799	# if PGM_GST_TYPE != PGM_TYPE_REAL
4800	Assert(!pVM->pgm.s.fNestedPaging);
4801	# endif
4802
4803	pgmLock(pVM);
4804
4805	# ifndef PGM_WITHOUT_MAPPINGS
4806	if (pVCpu->pgm.s.CTX_SUFF(pShwPageCR3))
4807	/* Remove the hypervisor mappings from the shadow page table. */
4808	pgmMapDeactivateCR3(pVM, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4809	# endif
4810
4811	if (pVCpu->pgm.s.CTX_SUFF(pShwPageCR3))
4812	{
4813	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4814
4815	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
4816	if (pPool->cDirtyPages)
4817	pgmPoolResetDirtyPages(pVM);
4818	# endif
4819
4820	/* Mark the page as unlocked; allow flushing again. */
4821	pgmPoolUnlockPage(pPool, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
4822
4823	pgmPoolFreeByPage(pPool, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3), NIL_PGMPOOL_IDX, UINT32_MAX);
4824	pVCpu->pgm.s.pShwPageCR3R3 = 0;
4825	pVCpu->pgm.s.pShwPageCR3R0 = 0;
4826	pVCpu->pgm.s.pShwPageCR3RC = 0;
4827	}
4828	pgmUnlock(pVM);
4829	# endif
4830	#endif /* !IN_RC*/
4831
4832	return rc;
4833	}

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllBth.h@ 46422

Download in other formats: