PGMAllPool.cpp@ 48749

Last change on this file since 48749 was 48184, checked in by vboxsync, 11 years ago
PGMAllPool.cpp: Try work around assertion in pgmPoolMonitorChainChanging caused by cbWrite being larger than 8.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 210.5 KB

Line
1	/* $Id: PGMAllPool.cpp 48184 2013-08-30 09:57:05Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
4	*/
5
6	/*
7	* Copyright (C) 2006-2013 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18
19	/*******************************************************************************
20	* Header Files *
21	*******************************************************************************/
22	#define LOG_GROUP LOG_GROUP_PGM_POOL
23	#include <VBox/vmm/pgm.h>
24	#include <VBox/vmm/mm.h>
25	#include <VBox/vmm/em.h>
26	#include <VBox/vmm/cpum.h>
27	#ifdef IN_RC
28	# include <VBox/vmm/patm.h>
29	#endif
30	#include "PGMInternal.h"
31	#include <VBox/vmm/vm.h>
32	#include "PGMInline.h"
33	#include <VBox/disopcode.h>
34	#include <VBox/vmm/hm_vmx.h>
35
36	#include <VBox/log.h>
37	#include <VBox/err.h>
38	#include <iprt/asm.h>
39	#include <iprt/asm-amd64-x86.h>
40	#include <iprt/string.h>
41
42
43	/*******************************************************************************
44	* Internal Functions *
45	*******************************************************************************/
46	RT_C_DECLS_BEGIN
47	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind);
48	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind);
49	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
50	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable);
51	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
52	#ifndef IN_RING3
53	DECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser);
54	#endif
55	#if defined(LOG_ENABLED) \|\| defined(VBOX_STRICT)
56	static const char *pgmPoolPoolKindToStr(uint8_t enmKind);
57	#endif
58	#if 0 /defined(VBOX_STRICT) && defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT)/
59	static void pgmPoolTrackCheckPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PTPAE pGstPT);
60	#endif
61
62	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage);
63	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt);
64	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt);
65	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt);
66
67	RT_C_DECLS_END
68
69
70	/**
71	* Checks if the specified page pool kind is for a 4MB or 2MB guest page.
72	*
73	* @returns true if it's the shadow of a 4MB or 2MB guest page, otherwise false.
74	* @param enmKind The page kind.
75	*/
76	DECLINLINE(bool) pgmPoolIsBigPage(PGMPOOLKIND enmKind)
77	{
78	switch (enmKind)
79	{
80	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
81	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
82	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
83	return true;
84	default:
85	return false;
86	}
87	}
88
89
90	/**
91	* Flushes a chain of pages sharing the same access monitor.
92	*
93	* @returns VBox status code suitable for scheduling.
94	* @param pPool The pool.
95	* @param pPage A page in the chain.
96	* @todo VBOXSTRICTRC
97	*/
98	int pgmPoolMonitorChainFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
99	{
100	LogFlow(("pgmPoolMonitorChainFlush: Flush page %RGp type=%d\n", pPage->GCPhys, pPage->enmKind));
101
102	/*
103	* Find the list head.
104	*/
105	uint16_t idx = pPage->idx;
106	if (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
107	{
108	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
109	{
110	idx = pPage->iMonitoredPrev;
111	Assert(idx != pPage->idx);
112	pPage = &pPool->aPages[idx];
113	}
114	}
115
116	/*
117	* Iterate the list flushing each shadow page.
118	*/
119	int rc = VINF_SUCCESS;
120	for (;;)
121	{
122	idx = pPage->iMonitoredNext;
123	Assert(idx != pPage->idx);
124	if (pPage->idx >= PGMPOOL_IDX_FIRST)
125	{
126	int rc2 = pgmPoolFlushPage(pPool, pPage);
127	AssertRC(rc2);
128	}
129	/* next */
130	if (idx == NIL_PGMPOOL_IDX)
131	break;
132	pPage = &pPool->aPages[idx];
133	}
134	return rc;
135	}
136
137
138	/**
139	* Wrapper for getting the current context pointer to the entry being modified.
140	*
141	* @returns VBox status code suitable for scheduling.
142	* @param pVM Pointer to the VM.
143	* @param pvDst Destination address
144	* @param pvSrc Source guest virtual address.
145	* @param GCPhysSrc The source guest physical address.
146	* @param cb Size of data to read
147	*/
148	DECLINLINE(int) pgmPoolPhysSimpleReadGCPhys(PVM pVM, void *pvDst, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvSrc,
149	RTGCPHYS GCPhysSrc, size_t cb)
150	{
151	#if defined(IN_RING3)
152	NOREF(pVM); NOREF(GCPhysSrc);
153	memcpy(pvDst, (RTHCPTR)((uintptr_t)pvSrc & ~(RTHCUINTPTR)(cb - 1)), cb);
154	return VINF_SUCCESS;
155	#else
156	/* @todo in RC we could attempt to use the virtual address, although this can cause many faults (PAE Windows XP guest). */
157	NOREF(pvSrc);
158	return PGMPhysSimpleReadGCPhys(pVM, pvDst, GCPhysSrc & ~(RTGCPHYS)(cb - 1), cb);
159	#endif
160	}
161
162
163	/**
164	* Process shadow entries before they are changed by the guest.
165	*
166	* For PT entries we will clear them. For PD entries, we'll simply check
167	* for mapping conflicts and set the SyncCR3 FF if found.
168	*
169	* @param pVCpu Pointer to the VMCPU.
170	* @param pPool The pool.
171	* @param pPage The head page.
172	* @param GCPhysFault The guest physical fault address.
173	* @param uAddress In R0 and GC this is the guest context fault address (flat).
174	* In R3 this is the host context 'fault' address.
175	* @param cbWrite Write size; might be zero if the caller knows we're not crossing entry boundaries
176	*/
177	void pgmPoolMonitorChainChanging(PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhysFault,
178	CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvAddress, unsigned cbWrite)
179	{
180	AssertMsg(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX, ("%u (idx=%u)\n", pPage->iMonitoredPrev, pPage->idx));
181	const unsigned off = GCPhysFault & PAGE_OFFSET_MASK;
182	PVM pVM = pPool->CTX_SUFF(pVM);
183	NOREF(pVCpu);
184
185	LogFlow(("pgmPoolMonitorChainChanging: %RGv phys=%RGp cbWrite=%d\n", (RTGCPTR)(CTXTYPE(RTGCPTR, uintptr_t, RTGCPTR))pvAddress, GCPhysFault, cbWrite));
186
187	for (;;)
188	{
189	union
190	{
191	void *pv;
192	PX86PT pPT;
193	PPGMSHWPTPAE pPTPae;
194	PX86PD pPD;
195	PX86PDPAE pPDPae;
196	PX86PDPT pPDPT;
197	PX86PML4 pPML4;
198	} uShw;
199
200	LogFlow(("pgmPoolMonitorChainChanging: page idx=%d phys=%RGp (next=%d) kind=%s\n", pPage->idx, pPage->GCPhys, pPage->iMonitoredNext, pgmPoolPoolKindToStr(pPage->enmKind), cbWrite));
201
202	uShw.pv = NULL;
203	switch (pPage->enmKind)
204	{
205	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
206	{
207	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPT));
208	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
209	const unsigned iShw = off / sizeof(X86PTE);
210	LogFlow(("PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT iShw=%x\n", iShw));
211	if (uShw.pPT->a[iShw].n.u1Present)
212	{
213	X86PTE GstPte;
214
215	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
216	AssertRC(rc);
217	Log4(("pgmPoolMonitorChainChanging 32_32: deref %016RX64 GCPhys %08RX32\n", uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PG_MASK));
218	pgmPoolTracDerefGCPhysHint(pPool, pPage,
219	uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK,
220	GstPte.u & X86_PTE_PG_MASK,
221	iShw);
222	ASMAtomicWriteU32(&uShw.pPT->a[iShw].u, 0);
223	}
224	break;
225	}
226
227	/* page/2 sized */
228	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
229	{
230	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPT));
231	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
232	if (!((off ^ pPage->GCPhys) & (PAGE_SIZE / 2)))
233	{
234	const unsigned iShw = (off / sizeof(X86PTE)) & (X86_PG_PAE_ENTRIES - 1);
235	LogFlow(("PGMPOOLKIND_PAE_PT_FOR_32BIT_PT iShw=%x\n", iShw));
236	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw]))
237	{
238	X86PTE GstPte;
239	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
240	AssertRC(rc);
241
242	Log4(("pgmPoolMonitorChainChanging pae_32: deref %016RX64 GCPhys %08RX32\n", uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PG_MASK));
243	pgmPoolTracDerefGCPhysHint(pPool, pPage,
244	PGMSHWPTEPAE_GET_HCPHYS(uShw.pPTPae->a[iShw]),
245	GstPte.u & X86_PTE_PG_MASK,
246	iShw);
247	PGMSHWPTEPAE_ATOMIC_SET(uShw.pPTPae->a[iShw], 0);
248	}
249	}
250	break;
251	}
252
253	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
254	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
255	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
256	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
257	{
258	unsigned iGst = off / sizeof(X86PDE);
259	unsigned iShwPdpt = iGst / 256;
260	unsigned iShw = (iGst % 256) * 2;
261	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
262
263	LogFlow(("pgmPoolMonitorChainChanging PAE for 32 bits: iGst=%x iShw=%x idx = %d page idx=%d\n", iGst, iShw, iShwPdpt, pPage->enmKind - PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD));
264	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPD));
265	if (iShwPdpt == pPage->enmKind - (unsigned)PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD)
266	{
267	for (unsigned i = 0; i < 2; i++)
268	{
269	# ifdef VBOX_WITH_RAW_MODE_NOT_R0
270	if ((uShw.pPDPae->a[iShw + i].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
271	{
272	Assert(pgmMapAreMappingsEnabled(pVM));
273	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
274	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw=%#x!\n", iShwPdpt, iShw+i));
275	break;
276	}
277	# endif /* VBOX_WITH_RAW_MODE_NOT_R0 */
278	if (uShw.pPDPae->a[iShw+i].n.u1Present)
279	{
280	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw+i, uShw.pPDPae->a[iShw+i].u));
281	pgmPoolFree(pVM,
282	uShw.pPDPae->a[iShw+i].u & X86_PDE_PAE_PG_MASK,
283	pPage->idx,
284	iShw + i);
285	ASMAtomicWriteU64(&uShw.pPDPae->a[iShw+i].u, 0);
286	}
287
288	/* paranoia / a bit assumptive. */
289	if ( (off & 3)
290	&& (off & 3) + cbWrite > 4)
291	{
292	const unsigned iShw2 = iShw + 2 + i;
293	if (iShw2 < RT_ELEMENTS(uShw.pPDPae->a))
294	{
295	# ifdef VBOX_WITH_RAW_MODE_NOT_R0
296	if ((uShw.pPDPae->a[iShw2].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
297	{
298	Assert(pgmMapAreMappingsEnabled(pVM));
299	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
300	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw2=%#x!\n", iShwPdpt, iShw2));
301	break;
302	}
303	# endif /* VBOX_WITH_RAW_MODE_NOT_R0 */
304	if (uShw.pPDPae->a[iShw2].n.u1Present)
305	{
306	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
307	pgmPoolFree(pVM,
308	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
309	pPage->idx,
310	iShw2);
311	ASMAtomicWriteU64(&uShw.pPDPae->a[iShw2].u, 0);
312	}
313	}
314	}
315	}
316	}
317	break;
318	}
319
320	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
321	{
322	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
323	const unsigned iShw = off / sizeof(X86PTEPAE);
324	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPT));
325	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw]))
326	{
327	X86PTEPAE GstPte;
328	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
329	AssertRC(rc);
330
331	Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", PGMSHWPTEPAE_GET_HCPHYS(uShw.pPTPae->a[iShw]), GstPte.u & X86_PTE_PAE_PG_MASK));
332	pgmPoolTracDerefGCPhysHint(pPool, pPage,
333	PGMSHWPTEPAE_GET_HCPHYS(uShw.pPTPae->a[iShw]),
334	GstPte.u & X86_PTE_PAE_PG_MASK,
335	iShw);
336	PGMSHWPTEPAE_ATOMIC_SET(uShw.pPTPae->a[iShw], 0);
337	}
338
339	/* paranoia / a bit assumptive. */
340	if ( (off & 7)
341	&& (off & 7) + cbWrite > sizeof(X86PTEPAE))
342	{
343	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTEPAE);
344	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPTPae->a));
345
346	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw2]))
347	{
348	X86PTEPAE GstPte;
349	# ifdef IN_RING3
350	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, (RTHCPTR)((RTHCUINTPTR)pvAddress + sizeof(GstPte)), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
351	# else
352	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress + sizeof(GstPte), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
353	# endif
354	AssertRC(rc);
355	Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", PGMSHWPTEPAE_GET_HCPHYS(uShw.pPTPae->a[iShw2]), GstPte.u & X86_PTE_PAE_PG_MASK));
356	pgmPoolTracDerefGCPhysHint(pPool, pPage,
357	PGMSHWPTEPAE_GET_HCPHYS(uShw.pPTPae->a[iShw2]),
358	GstPte.u & X86_PTE_PAE_PG_MASK,
359	iShw2);
360	PGMSHWPTEPAE_ATOMIC_SET(uShw.pPTPae->a[iShw2], 0);
361	}
362	}
363	break;
364	}
365
366	case PGMPOOLKIND_32BIT_PD:
367	{
368	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
369	const unsigned iShw = off / sizeof(X86PTE); // ASSUMING 32-bit guest paging!
370
371	LogFlow(("pgmPoolMonitorChainChanging: PGMPOOLKIND_32BIT_PD %x\n", iShw));
372	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPD));
373	# ifdef VBOX_WITH_RAW_MODE_NOT_R0
374	if (uShw.pPD->a[iShw].u & PGM_PDFLAGS_MAPPING)
375	{
376	Assert(pgmMapAreMappingsEnabled(pVM));
377	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
378	STAM_COUNTER_INC(&(pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZGuestCR3WriteConflict));
379	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
380	break;
381	}
382	else
383	# endif /* VBOX_WITH_RAW_MODE_NOT_R0 */
384	{
385	if (uShw.pPD->a[iShw].n.u1Present)
386	{
387	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
388	pgmPoolFree(pVM,
389	uShw.pPD->a[iShw].u & X86_PDE_PAE_PG_MASK,
390	pPage->idx,
391	iShw);
392	ASMAtomicWriteU32(&uShw.pPD->a[iShw].u, 0);
393	}
394	}
395	/* paranoia / a bit assumptive. */
396	if ( (off & 3)
397	&& (off & 3) + cbWrite > sizeof(X86PTE))
398	{
399	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTE);
400	if ( iShw2 != iShw
401	&& iShw2 < RT_ELEMENTS(uShw.pPD->a))
402	{
403	# ifdef VBOX_WITH_RAW_MODE_NOT_R0
404	if (uShw.pPD->a[iShw2].u & PGM_PDFLAGS_MAPPING)
405	{
406	Assert(pgmMapAreMappingsEnabled(pVM));
407	STAM_COUNTER_INC(&(pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZGuestCR3WriteConflict));
408	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
409	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
410	break;
411	}
412	# endif /* VBOX_WITH_RAW_MODE_NOT_R0 */
413	if (uShw.pPD->a[iShw2].n.u1Present)
414	{
415	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPD->a[iShw2].u));
416	pgmPoolFree(pVM,
417	uShw.pPD->a[iShw2].u & X86_PDE_PAE_PG_MASK,
418	pPage->idx,
419	iShw2);
420	ASMAtomicWriteU32(&uShw.pPD->a[iShw2].u, 0);
421	}
422	}
423	}
424	#if 0 /* useful when running PGMAssertCR3(), a bit too troublesome for general use (TLBs). */
425	if ( uShw.pPD->a[iShw].n.u1Present
426	&& !VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
427	{
428	LogFlow(("pgmPoolMonitorChainChanging: iShw=%#x: %RX32 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
429	# ifdef IN_RC /* TLB load - we're pushing things a bit... */
430	ASMProbeReadByte(pvAddress);
431	# endif
432	pgmPoolFree(pVM, uShw.pPD->a[iShw].u & X86_PDE_PG_MASK, pPage->idx, iShw);
433	ASMAtomicWriteU32(&uShw.pPD->a[iShw].u, 0);
434	}
435	#endif
436	break;
437	}
438
439	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
440	{
441	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
442	const unsigned iShw = off / sizeof(X86PDEPAE);
443	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPD));
444	#ifdef VBOX_WITH_RAW_MODE_NOT_R0
445	if (uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING)
446	{
447	Assert(pgmMapAreMappingsEnabled(pVM));
448	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
449	STAM_COUNTER_INC(&(pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZGuestCR3WriteConflict));
450	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
451	break;
452	}
453	#endif /* VBOX_WITH_RAW_MODE_NOT_R0 */
454	/*
455	* Causes trouble when the guest uses a PDE to refer to the whole page table level
456	* structure. (Invalidate here; faults later on when it tries to change the page
457	* table entries -> recheck; probably only applies to the RC case.)
458	*/
459	#ifdef VBOX_WITH_RAW_MODE_NOT_R0
460	else
461	#endif
462	{
463	if (uShw.pPDPae->a[iShw].n.u1Present)
464	{
465	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
466	pgmPoolFree(pVM,
467	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
468	pPage->idx,
469	iShw);
470	ASMAtomicWriteU64(&uShw.pPDPae->a[iShw].u, 0);
471	}
472	}
473	/* paranoia / a bit assumptive. */
474	if ( (off & 7)
475	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
476	{
477	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
478	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
479
480	#ifdef VBOX_WITH_RAW_MODE_NOT_R0
481	if ( iShw2 != iShw
482	&& uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING)
483	{
484	Assert(pgmMapAreMappingsEnabled(pVM));
485	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
486	STAM_COUNTER_INC(&(pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZGuestCR3WriteConflict));
487	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
488	break;
489	}
490	else
491	#endif /* VBOX_WITH_RAW_MODE_NOT_R0 */
492	if (uShw.pPDPae->a[iShw2].n.u1Present)
493	{
494	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
495	pgmPoolFree(pVM,
496	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
497	pPage->idx,
498	iShw2);
499	ASMAtomicWriteU64(&uShw.pPDPae->a[iShw2].u, 0);
500	}
501	}
502	break;
503	}
504
505	case PGMPOOLKIND_PAE_PDPT:
506	{
507	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPDPT));
508	/*
509	* Hopefully this doesn't happen very often:
510	* - touching unused parts of the page
511	* - messing with the bits of pd pointers without changing the physical address
512	*/
513	/* PDPT roots are not page aligned; 32 byte only! */
514	const unsigned offPdpt = GCPhysFault - pPage->GCPhys;
515
516	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
517	const unsigned iShw = offPdpt / sizeof(X86PDPE);
518	if (iShw < X86_PG_PAE_PDPE_ENTRIES) /* don't use RT_ELEMENTS(uShw.pPDPT->a), because that's for long mode only */
519	{
520	# ifdef VBOX_WITH_RAW_MODE_NOT_R0
521	if (uShw.pPDPT->a[iShw].u & PGM_PLXFLAGS_MAPPING)
522	{
523	Assert(pgmMapAreMappingsEnabled(pVM));
524	STAM_COUNTER_INC(&(pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZGuestCR3WriteConflict));
525	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
526	LogFlow(("pgmPoolMonitorChainChanging: Detected pdpt conflict at iShw=%#x!\n", iShw));
527	break;
528	}
529	else
530	# endif /* VBOX_WITH_RAW_MODE_NOT_R0 */
531	if (uShw.pPDPT->a[iShw].n.u1Present)
532	{
533	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
534	pgmPoolFree(pVM,
535	uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK,
536	pPage->idx,
537	iShw);
538	ASMAtomicWriteU64(&uShw.pPDPT->a[iShw].u, 0);
539	}
540
541	/* paranoia / a bit assumptive. */
542	if ( (offPdpt & 7)
543	&& (offPdpt & 7) + cbWrite > sizeof(X86PDPE))
544	{
545	const unsigned iShw2 = (offPdpt + cbWrite - 1) / sizeof(X86PDPE);
546	if ( iShw2 != iShw
547	&& iShw2 < X86_PG_PAE_PDPE_ENTRIES)
548	{
549	# ifdef VBOX_WITH_RAW_MODE_NOT_R0
550	if (uShw.pPDPT->a[iShw2].u & PGM_PLXFLAGS_MAPPING)
551	{
552	Assert(pgmMapAreMappingsEnabled(pVM));
553	STAM_COUNTER_INC(&(pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZGuestCR3WriteConflict));
554	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
555	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
556	break;
557	}
558	else
559	# endif /* VBOX_WITH_RAW_MODE_NOT_R0 */
560	if (uShw.pPDPT->a[iShw2].n.u1Present)
561	{
562	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
563	pgmPoolFree(pVM,
564	uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK,
565	pPage->idx,
566	iShw2);
567	ASMAtomicWriteU64(&uShw.pPDPT->a[iShw2].u, 0);
568	}
569	}
570	}
571	}
572	break;
573	}
574
575	#ifndef IN_RC
576	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
577	{
578	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPD));
579	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
580	const unsigned iShw = off / sizeof(X86PDEPAE);
581	Assert(!(uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING));
582	if (uShw.pPDPae->a[iShw].n.u1Present)
583	{
584	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
585	pgmPoolFree(pVM,
586	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
587	pPage->idx,
588	iShw);
589	ASMAtomicWriteU64(&uShw.pPDPae->a[iShw].u, 0);
590	}
591	/* paranoia / a bit assumptive. */
592	if ( (off & 7)
593	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
594	{
595	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
596	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
597
598	Assert(!(uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING));
599	if (uShw.pPDPae->a[iShw2].n.u1Present)
600	{
601	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
602	pgmPoolFree(pVM,
603	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
604	pPage->idx,
605	iShw2);
606	ASMAtomicWriteU64(&uShw.pPDPae->a[iShw2].u, 0);
607	}
608	}
609	break;
610	}
611
612	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
613	{
614	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPDPT));
615	/*
616	* Hopefully this doesn't happen very often:
617	* - messing with the bits of pd pointers without changing the physical address
618	*/
619	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
620	const unsigned iShw = off / sizeof(X86PDPE);
621	if (uShw.pPDPT->a[iShw].n.u1Present)
622	{
623	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
624	pgmPoolFree(pVM, uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK, pPage->idx, iShw);
625	ASMAtomicWriteU64(&uShw.pPDPT->a[iShw].u, 0);
626	}
627	/* paranoia / a bit assumptive. */
628	if ( (off & 7)
629	&& (off & 7) + cbWrite > sizeof(X86PDPE))
630	{
631	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDPE);
632	if (uShw.pPDPT->a[iShw2].n.u1Present)
633	{
634	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
635	pgmPoolFree(pVM, uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK, pPage->idx, iShw2);
636	ASMAtomicWriteU64(&uShw.pPDPT->a[iShw2].u, 0);
637	}
638	}
639	break;
640	}
641
642	case PGMPOOLKIND_64BIT_PML4:
643	{
644	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPML4));
645	/*
646	* Hopefully this doesn't happen very often:
647	* - messing with the bits of pd pointers without changing the physical address
648	*/
649	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
650	const unsigned iShw = off / sizeof(X86PDPE);
651	if (uShw.pPML4->a[iShw].n.u1Present)
652	{
653	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPML4->a[iShw].u));
654	pgmPoolFree(pVM, uShw.pPML4->a[iShw].u & X86_PML4E_PG_MASK, pPage->idx, iShw);
655	ASMAtomicWriteU64(&uShw.pPML4->a[iShw].u, 0);
656	}
657	/* paranoia / a bit assumptive. */
658	if ( (off & 7)
659	&& (off & 7) + cbWrite > sizeof(X86PDPE))
660	{
661	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PML4E);
662	if (uShw.pPML4->a[iShw2].n.u1Present)
663	{
664	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPML4->a[iShw2].u));
665	pgmPoolFree(pVM, uShw.pPML4->a[iShw2].u & X86_PML4E_PG_MASK, pPage->idx, iShw2);
666	ASMAtomicWriteU64(&uShw.pPML4->a[iShw2].u, 0);
667	}
668	}
669	break;
670	}
671	#endif /* IN_RING0 */
672
673	default:
674	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
675	}
676	PGM_DYNMAP_UNUSED_HINT_VM(pVM, uShw.pv);
677
678	/* next */
679	if (pPage->iMonitoredNext == NIL_PGMPOOL_IDX)
680	return;
681	pPage = &pPool->aPages[pPage->iMonitoredNext];
682	}
683	}
684
685	# ifndef IN_RING3
686
687	/**
688	* Checks if a access could be a fork operation in progress.
689	*
690	* Meaning, that the guest is setting up the parent process for Copy-On-Write.
691	*
692	* @returns true if it's likely that we're forking, otherwise false.
693	* @param pPool The pool.
694	* @param pDis The disassembled instruction.
695	* @param offFault The access offset.
696	*/
697	DECLINLINE(bool) pgmPoolMonitorIsForking(PPGMPOOL pPool, PDISCPUSTATE pDis, unsigned offFault)
698	{
699	/*
700	* i386 linux is using btr to clear X86_PTE_RW.
701	* The functions involved are (2.6.16 source inspection):
702	* clear_bit
703	* ptep_set_wrprotect
704	* copy_one_pte
705	* copy_pte_range
706	* copy_pmd_range
707	* copy_pud_range
708	* copy_page_range
709	* dup_mmap
710	* dup_mm
711	* copy_mm
712	* copy_process
713	* do_fork
714	*/
715	if ( pDis->pCurInstr->uOpcode == OP_BTR
716	&& !(offFault & 4)
717	/** @todo Validate that the bit index is X86_PTE_RW. */
718	)
719	{
720	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,Fork));
721	return true;
722	}
723	return false;
724	}
725
726
727	/**
728	* Determine whether the page is likely to have been reused.
729	*
730	* @returns true if we consider the page as being reused for a different purpose.
731	* @returns false if we consider it to still be a paging page.
732	* @param pVM Pointer to the VM.
733	* @param pVCpu Pointer to the VMCPU.
734	* @param pRegFrame Trap register frame.
735	* @param pDis The disassembly info for the faulting instruction.
736	* @param pvFault The fault address.
737	*
738	* @remark The REP prefix check is left to the caller because of STOSD/W.
739	*/
740	DECLINLINE(bool) pgmPoolMonitorIsReused(PVM pVM, PVMCPU pVCpu, PCPUMCTXCORE pRegFrame, PDISCPUSTATE pDis, RTGCPTR pvFault)
741	{
742	#ifndef IN_RC
743	/** @todo could make this general, faulting close to rsp should be a safe reuse heuristic. */
744	if ( HMHasPendingIrq(pVM)
745	&& (pRegFrame->rsp - pvFault) < 32)
746	{
747	/* Fault caused by stack writes while trying to inject an interrupt event. */
748	Log(("pgmPoolMonitorIsReused: reused %RGv for interrupt stack (rsp=%RGv).\n", pvFault, pRegFrame->rsp));
749	return true;
750	}
751	#else
752	NOREF(pVM); NOREF(pvFault);
753	#endif
754
755	LogFlow(("Reused instr %RGv %d at %RGv param1.fUse=%llx param1.reg=%d\n", pRegFrame->rip, pDis->pCurInstr->uOpcode, pvFault, pDis->Param1.fUse, pDis->Param1.Base.idxGenReg));
756
757	/* Non-supervisor mode write means it's used for something else. */
758	if (CPUMGetGuestCPL(pVCpu) == 3)
759	return true;
760
761	switch (pDis->pCurInstr->uOpcode)
762	{
763	/* call implies the actual push of the return address faulted */
764	case OP_CALL:
765	Log4(("pgmPoolMonitorIsReused: CALL\n"));
766	return true;
767	case OP_PUSH:
768	Log4(("pgmPoolMonitorIsReused: PUSH\n"));
769	return true;
770	case OP_PUSHF:
771	Log4(("pgmPoolMonitorIsReused: PUSHF\n"));
772	return true;
773	case OP_PUSHA:
774	Log4(("pgmPoolMonitorIsReused: PUSHA\n"));
775	return true;
776	case OP_FXSAVE:
777	Log4(("pgmPoolMonitorIsReused: FXSAVE\n"));
778	return true;
779	case OP_MOVNTI: /* solaris - block_zero_no_xmm */
780	Log4(("pgmPoolMonitorIsReused: MOVNTI\n"));
781	return true;
782	case OP_MOVNTDQ: /* solaris - hwblkclr & hwblkpagecopy */
783	Log4(("pgmPoolMonitorIsReused: MOVNTDQ\n"));
784	return true;
785	case OP_MOVSWD:
786	case OP_STOSWD:
787	if ( pDis->fPrefix == (DISPREFIX_REP\|DISPREFIX_REX)
788	&& pRegFrame->rcx >= 0x40
789	)
790	{
791	Assert(pDis->uCpuMode == DISCPUMODE_64BIT);
792
793	Log(("pgmPoolMonitorIsReused: OP_STOSQ\n"));
794	return true;
795	}
796	return false;
797	}
798	if ( ( (pDis->Param1.fUse & DISUSE_REG_GEN32)
799	\|\| (pDis->Param1.fUse & DISUSE_REG_GEN64))
800	&& (pDis->Param1.Base.idxGenReg == DISGREG_ESP))
801	{
802	Log4(("pgmPoolMonitorIsReused: ESP\n"));
803	return true;
804	}
805
806	return false;
807	}
808
809
810	/**
811	* Flushes the page being accessed.
812	*
813	* @returns VBox status code suitable for scheduling.
814	* @param pVM Pointer to the VM.
815	* @param pVCpu Pointer to the VMCPU.
816	* @param pPool The pool.
817	* @param pPage The pool page (head).
818	* @param pDis The disassembly of the write instruction.
819	* @param pRegFrame The trap register frame.
820	* @param GCPhysFault The fault address as guest physical address.
821	* @param pvFault The fault address.
822	* @todo VBOXSTRICTRC
823	*/
824	static int pgmPoolAccessHandlerFlush(PVM pVM, PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pDis,
825	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
826	{
827	NOREF(GCPhysFault);
828
829	/*
830	* First, do the flushing.
831	*/
832	int rc = pgmPoolMonitorChainFlush(pPool, pPage);
833
834	/*
835	* Emulate the instruction (xp/w2k problem, requires pc/cr2/sp detection).
836	* Must do this in raw mode (!); XP boot will fail otherwise.
837	*/
838	VBOXSTRICTRC rc2 = EMInterpretInstructionDisasState(pVCpu, pDis, pRegFrame, pvFault, EMCODETYPE_ALL);
839	if (rc2 == VINF_SUCCESS)
840	{ /* do nothing */ }
841	#ifdef VBOX_WITH_IEM
842	else if (rc2 == VINF_EM_RESCHEDULE)
843	{
844	if (rc == VINF_SUCCESS)
845	rc = VBOXSTRICTRC_VAL(rc2);
846	# ifndef IN_RING3
847	VMCPU_FF_SET(pVCpu, VMCPU_FF_TO_R3);
848	# endif
849	}
850	#endif
851	else if (rc2 == VERR_EM_INTERPRETER)
852	{
853	#ifdef IN_RC
854	if (PATMIsPatchGCAddr(pVM, pRegFrame->eip))
855	{
856	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for patch code %04x:%RGv, ignoring.\n",
857	pRegFrame->cs.Sel, (RTGCPTR)pRegFrame->eip));
858	rc = VINF_SUCCESS;
859	STAM_COUNTER_INC(&pPool->StatMonitorRZIntrFailPatch2);
860	}
861	else
862	#endif
863	{
864	rc = VINF_EM_RAW_EMULATE_INSTR;
865	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
866	}
867	}
868	else if (RT_FAILURE_NP(rc2))
869	rc = VBOXSTRICTRC_VAL(rc2);
870	else
871	AssertMsgFailed(("%Rrc\n", VBOXSTRICTRC_VAL(rc2))); /* ASSUMES no complicated stuff here. */
872
873	LogFlow(("pgmPoolAccessHandlerPT: returns %Rrc (flushed)\n", rc));
874	return rc;
875	}
876
877
878	/**
879	* Handles the STOSD write accesses.
880	*
881	* @returns VBox status code suitable for scheduling.
882	* @param pVM Pointer to the VM.
883	* @param pPool The pool.
884	* @param pPage The pool page (head).
885	* @param pDis The disassembly of the write instruction.
886	* @param pRegFrame The trap register frame.
887	* @param GCPhysFault The fault address as guest physical address.
888	* @param pvFault The fault address.
889	*/
890	DECLINLINE(int) pgmPoolAccessHandlerSTOSD(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pDis,
891	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
892	{
893	unsigned uIncrement = pDis->Param1.cb;
894	NOREF(pVM);
895
896	Assert(pDis->uCpuMode == DISCPUMODE_32BIT \|\| pDis->uCpuMode == DISCPUMODE_64BIT);
897	Assert(pRegFrame->rcx <= 0x20);
898
899	#ifdef VBOX_STRICT
900	if (pDis->uOpMode == DISCPUMODE_32BIT)
901	Assert(uIncrement == 4);
902	else
903	Assert(uIncrement == 8);
904	#endif
905
906	Log3(("pgmPoolAccessHandlerSTOSD\n"));
907
908	/*
909	* Increment the modification counter and insert it into the list
910	* of modified pages the first time.
911	*/
912	if (!pPage->cModifications++)
913	pgmPoolMonitorModifiedInsert(pPool, pPage);
914
915	/*
916	* Execute REP STOSD.
917	*
918	* This ASSUMES that we're not invoked by Trap0e on in a out-of-sync
919	* write situation, meaning that it's safe to write here.
920	*/
921	PVMCPU pVCpu = VMMGetCpu(pPool->CTX_SUFF(pVM));
922	RTGCUINTPTR pu32 = (RTGCUINTPTR)pvFault;
923	while (pRegFrame->rcx)
924	{
925	#if defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0) \|\| defined(IN_RC)
926	uint32_t iPrevSubset = PGMRZDynMapPushAutoSubset(pVCpu);
927	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, (RTGCPTR)pu32, uIncrement);
928	PGMRZDynMapPopAutoSubset(pVCpu, iPrevSubset);
929	#else
930	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, (RTGCPTR)pu32, uIncrement);
931	#endif
932	#ifdef IN_RC
933	(uint32_t )(uintptr_t)pu32 = pRegFrame->eax;
934	#else
935	PGMPhysSimpleWriteGCPhys(pVM, GCPhysFault, &pRegFrame->rax, uIncrement);
936	#endif
937	pu32 += uIncrement;
938	GCPhysFault += uIncrement;
939	pRegFrame->rdi += uIncrement;
940	pRegFrame->rcx--;
941	}
942	pRegFrame->rip += pDis->cbInstr;
943
944	LogFlow(("pgmPoolAccessHandlerSTOSD: returns\n"));
945	return VINF_SUCCESS;
946	}
947
948
949	/**
950	* Handles the simple write accesses.
951	*
952	* @returns VBox status code suitable for scheduling.
953	* @param pVM Pointer to the VM.
954	* @param pVCpu Pointer to the VMCPU.
955	* @param pPool The pool.
956	* @param pPage The pool page (head).
957	* @param pDis The disassembly of the write instruction.
958	* @param pRegFrame The trap register frame.
959	* @param GCPhysFault The fault address as guest physical address.
960	* @param pvFault The fault address.
961	* @param pfReused Reused state (in/out)
962	*/
963	DECLINLINE(int) pgmPoolAccessHandlerSimple(PVM pVM, PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pDis,
964	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault, bool *pfReused)
965	{
966	Log3(("pgmPoolAccessHandlerSimple\n"));
967	NOREF(pfReused); /* initialized by caller */
968
969	/*
970	* Increment the modification counter and insert it into the list
971	* of modified pages the first time.
972	*/
973	if (!pPage->cModifications++)
974	pgmPoolMonitorModifiedInsert(pPool, pPage);
975
976	/*
977	* Clear all the pages. ASSUMES that pvFault is readable.
978	*/
979	#if defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0) \|\| defined(IN_RC)
980	uint32_t iPrevSubset = PGMRZDynMapPushAutoSubset(pVCpu);
981	#endif
982
983	uint32_t cbWrite = DISGetParamSize(pDis, &pDis->Param1);
984	if (cbWrite <= 8)
985	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, pvFault, cbWrite);
986	else
987	{
988	Assert(cbWrite <= 16);
989	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, pvFault, 8);
990	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault + 8, pvFault + 8, cbWrite - 8);
991	}
992
993	#if defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0) \|\| defined(IN_RC)
994	PGMRZDynMapPopAutoSubset(pVCpu, iPrevSubset);
995	#endif
996
997	/*
998	* Interpret the instruction.
999	*/
1000	VBOXSTRICTRC rc = EMInterpretInstructionDisasState(pVCpu, pDis, pRegFrame, pvFault, EMCODETYPE_ALL);
1001	if (RT_SUCCESS(rc))
1002	AssertMsg(rc == VINF_SUCCESS, ("%Rrc\n", VBOXSTRICTRC_VAL(rc))); /* ASSUMES no complicated stuff here. */
1003	else if (rc == VERR_EM_INTERPRETER)
1004	{
1005	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for %04x:%RGv - opcode=%d\n",
1006	pRegFrame->cs.Sel, (RTGCPTR)pRegFrame->rip, pDis->pCurInstr->uOpcode));
1007	rc = VINF_EM_RAW_EMULATE_INSTR;
1008	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
1009	}
1010
1011	#if 0 /* experimental code */
1012	if (rc == VINF_SUCCESS)
1013	{
1014	switch (pPage->enmKind)
1015	{
1016	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1017	{
1018	X86PTEPAE GstPte;
1019	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvFault, GCPhysFault, sizeof(GstPte));
1020	AssertRC(rc);
1021
1022	/* Check the new value written by the guest. If present and with a bogus physical address, then
1023	* it's fairly safe to assume the guest is reusing the PT.
1024	*/
1025	if (GstPte.n.u1Present)
1026	{
1027	RTHCPHYS HCPhys = -1;
1028	int rc = PGMPhysGCPhys2HCPhys(pVM, GstPte.u & X86_PTE_PAE_PG_MASK, &HCPhys);
1029	if (rc != VINF_SUCCESS)
1030	{
1031	*pfReused = true;
1032	STAM_COUNTER_INC(&pPool->StatForceFlushReused);
1033	}
1034	}
1035	break;
1036	}
1037	}
1038	}
1039	#endif
1040
1041	LogFlow(("pgmPoolAccessHandlerSimple: returns %Rrc\n", VBOXSTRICTRC_VAL(rc)));
1042	return VBOXSTRICTRC_VAL(rc);
1043	}
1044
1045
1046	/**
1047	* \#PF Handler callback for PT write accesses.
1048	*
1049	* @returns VBox status code (appropriate for GC return).
1050	* @param pVM Pointer to the VM.
1051	* @param uErrorCode CPU Error code.
1052	* @param pRegFrame Trap register frame.
1053	* NULL on DMA and other non CPU access.
1054	* @param pvFault The fault address (cr2).
1055	* @param GCPhysFault The GC physical address corresponding to pvFault.
1056	* @param pvUser User argument.
1057	*/
1058	DECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault,
1059	RTGCPHYS GCPhysFault, void *pvUser)
1060	{
1061	STAM_PROFILE_START(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), a);
1062	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1063	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
1064	PVMCPU pVCpu = VMMGetCpu(pVM);
1065	unsigned cMaxModifications;
1066	bool fForcedFlush = false;
1067	NOREF(uErrorCode);
1068
1069	LogFlow(("pgmPoolAccessHandler: pvFault=%RGv pPage=%p:{.idx=%d} GCPhysFault=%RGp\n", pvFault, pPage, pPage->idx, GCPhysFault));
1070
1071	pgmLock(pVM);
1072	if (PHYS_PAGE_ADDRESS(GCPhysFault) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
1073	{
1074	/* Pool page changed while we were waiting for the lock; ignore. */
1075	Log(("CPU%d: pgmPoolAccessHandler pgm pool page for %RGp changed (to %RGp) while waiting!\n", pVCpu->idCpu, PHYS_PAGE_ADDRESS(GCPhysFault), PHYS_PAGE_ADDRESS(pPage->GCPhys)));
1076	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,Handled), a);
1077	pgmUnlock(pVM);
1078	return VINF_SUCCESS;
1079	}
1080	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1081	if (pPage->fDirty)
1082	{
1083	Assert(VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TLB_FLUSH));
1084	pgmUnlock(pVM);
1085	return VINF_SUCCESS; /* SMP guest case where we were blocking on the pgm lock while the same page was being marked dirty. */
1086	}
1087	#endif
1088
1089	#if 0 /* test code defined(VBOX_STRICT) && defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT) */
1090	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1091	{
1092	void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
1093	void *pvGst;
1094	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
1095	pgmPoolTrackCheckPTPaePae(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PTPAE)pvGst);
1096	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvGst);
1097	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvShw);
1098	}
1099	#endif
1100
1101	/*
1102	* Disassemble the faulting instruction.
1103	*/
1104	PDISCPUSTATE pDis = &pVCpu->pgm.s.DisState;
1105	int rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, NULL);
1106	if (RT_UNLIKELY(rc != VINF_SUCCESS))
1107	{
1108	AssertMsg(rc == VERR_PAGE_NOT_PRESENT \|\| rc == VERR_PAGE_TABLE_NOT_PRESENT, ("Unexpected rc %d\n", rc));
1109	pgmUnlock(pVM);
1110	return rc;
1111	}
1112
1113	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
1114
1115	/*
1116	* We should ALWAYS have the list head as user parameter. This
1117	* is because we use that page to record the changes.
1118	*/
1119	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1120
1121	#ifdef IN_RING0
1122	/* Maximum nr of modifications depends on the page type. */
1123	if ( pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT
1124	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_32BIT_PT)
1125	cMaxModifications = 4;
1126	else
1127	cMaxModifications = 24;
1128	#else
1129	cMaxModifications = 48;
1130	#endif
1131
1132	/*
1133	* Incremental page table updates should weigh more than random ones.
1134	* (Only applies when started from offset 0)
1135	*/
1136	pVCpu->pgm.s.cPoolAccessHandler++;
1137	if ( pPage->GCPtrLastAccessHandlerRip >= pRegFrame->rip - 0x40 /* observed loops in Windows 7 x64 */
1138	&& pPage->GCPtrLastAccessHandlerRip < pRegFrame->rip + 0x40
1139	&& pvFault == (pPage->GCPtrLastAccessHandlerFault + pDis->Param1.cb)
1140	&& pVCpu->pgm.s.cPoolAccessHandler == pPage->cLastAccessHandler + 1)
1141	{
1142	Log(("Possible page reuse cMods=%d -> %d (locked=%d type=%s)\n", pPage->cModifications, pPage->cModifications * 2, pgmPoolIsPageLocked(pPage), pgmPoolPoolKindToStr(pPage->enmKind)));
1143	Assert(pPage->cModifications < 32000);
1144	pPage->cModifications = pPage->cModifications * 2;
1145	pPage->GCPtrLastAccessHandlerFault = pvFault;
1146	pPage->cLastAccessHandler = pVCpu->pgm.s.cPoolAccessHandler;
1147	if (pPage->cModifications >= cMaxModifications)
1148	{
1149	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FlushReinit));
1150	fForcedFlush = true;
1151	}
1152	}
1153
1154	if (pPage->cModifications >= cMaxModifications)
1155	Log(("Mod overflow %RGv cMods=%d (locked=%d type=%s)\n", pvFault, pPage->cModifications, pgmPoolIsPageLocked(pPage), pgmPoolPoolKindToStr(pPage->enmKind)));
1156
1157	/*
1158	* Check if it's worth dealing with.
1159	*/
1160	bool fReused = false;
1161	bool fNotReusedNotForking = false;
1162	if ( ( pPage->cModifications < cMaxModifications /** @todo #define / /* @todo need to check that it's not mapping EIP. / /* @todo adjust this! */
1163	\|\| pgmPoolIsPageLocked(pPage)
1164	)
1165	&& !(fReused = pgmPoolMonitorIsReused(pVM, pVCpu, pRegFrame, pDis, pvFault))
1166	&& !pgmPoolMonitorIsForking(pPool, pDis, GCPhysFault & PAGE_OFFSET_MASK))
1167	{
1168	/*
1169	* Simple instructions, no REP prefix.
1170	*/
1171	if (!(pDis->fPrefix & (DISPREFIX_REP \| DISPREFIX_REPNE)))
1172	{
1173	rc = pgmPoolAccessHandlerSimple(pVM, pVCpu, pPool, pPage, pDis, pRegFrame, GCPhysFault, pvFault, &fReused);
1174	if (fReused)
1175	goto flushPage;
1176
1177	/* A mov instruction to change the first page table entry will be remembered so we can detect
1178	* full page table changes early on. This will reduce the amount of unnecessary traps we'll take.
1179	*/
1180	if ( rc == VINF_SUCCESS
1181	&& !pPage->cLocked /* only applies to unlocked pages as we can't free locked ones (e.g. cr3 root). */
1182	&& pDis->pCurInstr->uOpcode == OP_MOV
1183	&& (pvFault & PAGE_OFFSET_MASK) == 0)
1184	{
1185	pPage->GCPtrLastAccessHandlerFault = pvFault;
1186	pPage->cLastAccessHandler = pVCpu->pgm.s.cPoolAccessHandler;
1187	pPage->GCPtrLastAccessHandlerRip = pRegFrame->rip;
1188	/* Make sure we don't kick out a page too quickly. */
1189	if (pPage->cModifications > 8)
1190	pPage->cModifications = 2;
1191	}
1192	else if (pPage->GCPtrLastAccessHandlerFault == pvFault)
1193	{
1194	/* ignore the 2nd write to this page table entry. */
1195	pPage->cLastAccessHandler = pVCpu->pgm.s.cPoolAccessHandler;
1196	}
1197	else
1198	{
1199	pPage->GCPtrLastAccessHandlerFault = NIL_RTGCPTR;
1200	pPage->GCPtrLastAccessHandlerRip = 0;
1201	}
1202
1203	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,Handled), a);
1204	pgmUnlock(pVM);
1205	return rc;
1206	}
1207
1208	/*
1209	* Windows is frequently doing small memset() operations (netio test 4k+).
1210	* We have to deal with these or we'll kill the cache and performance.
1211	*/
1212	if ( pDis->pCurInstr->uOpcode == OP_STOSWD
1213	&& !pRegFrame->eflags.Bits.u1DF
1214	&& pDis->uOpMode == pDis->uCpuMode
1215	&& pDis->uAddrMode == pDis->uCpuMode)
1216	{
1217	bool fValidStosd = false;
1218
1219	if ( pDis->uCpuMode == DISCPUMODE_32BIT
1220	&& pDis->fPrefix == DISPREFIX_REP
1221	&& pRegFrame->ecx <= 0x20
1222	&& pRegFrame->ecx * 4 <= PAGE_SIZE - ((uintptr_t)pvFault & PAGE_OFFSET_MASK)
1223	&& !((uintptr_t)pvFault & 3)
1224	&& (pRegFrame->eax == 0 \|\| pRegFrame->eax == 0x80) /* the two values observed. */
1225	)
1226	{
1227	fValidStosd = true;
1228	pRegFrame->rcx &= 0xffffffff; /* paranoia */
1229	}
1230	else
1231	if ( pDis->uCpuMode == DISCPUMODE_64BIT
1232	&& pDis->fPrefix == (DISPREFIX_REP \| DISPREFIX_REX)
1233	&& pRegFrame->rcx <= 0x20
1234	&& pRegFrame->rcx * 8 <= PAGE_SIZE - ((uintptr_t)pvFault & PAGE_OFFSET_MASK)
1235	&& !((uintptr_t)pvFault & 7)
1236	&& (pRegFrame->rax == 0 \|\| pRegFrame->rax == 0x80) /* the two values observed. */
1237	)
1238	{
1239	fValidStosd = true;
1240	}
1241
1242	if (fValidStosd)
1243	{
1244	rc = pgmPoolAccessHandlerSTOSD(pVM, pPool, pPage, pDis, pRegFrame, GCPhysFault, pvFault);
1245	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,RepStosd), a);
1246	pgmUnlock(pVM);
1247	return rc;
1248	}
1249	}
1250
1251	/* REP prefix, don't bother. */
1252	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,RepPrefix));
1253	Log4(("pgmPoolAccessHandler: eax=%#x ecx=%#x edi=%#x esi=%#x rip=%RGv opcode=%d prefix=%#x\n",
1254	pRegFrame->eax, pRegFrame->ecx, pRegFrame->edi, pRegFrame->esi, (RTGCPTR)pRegFrame->rip, pDis->pCurInstr->uOpcode, pDis->fPrefix));
1255	fNotReusedNotForking = true;
1256	}
1257
1258	#if defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT) && defined(IN_RING0)
1259	/* E.g. Windows 7 x64 initializes page tables and touches some pages in the table during the process. This
1260	* leads to pgm pool trashing and an excessive amount of write faults due to page monitoring.
1261	*/
1262	if ( pPage->cModifications >= cMaxModifications
1263	&& !fForcedFlush
1264	&& (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT \|\| pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_32BIT_PT)
1265	&& ( fNotReusedNotForking
1266	\|\| ( !pgmPoolMonitorIsReused(pVM, pVCpu, pRegFrame, pDis, pvFault)
1267	&& !pgmPoolMonitorIsForking(pPool, pDis, GCPhysFault & PAGE_OFFSET_MASK))
1268	)
1269	)
1270	{
1271	Assert(!pgmPoolIsPageLocked(pPage));
1272	Assert(pPage->fDirty == false);
1273
1274	/* Flush any monitored duplicates as we will disable write protection. */
1275	if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
1276	\|\| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1277	{
1278	PPGMPOOLPAGE pPageHead = pPage;
1279
1280	/* Find the monitor head. */
1281	while (pPageHead->iMonitoredPrev != NIL_PGMPOOL_IDX)
1282	pPageHead = &pPool->aPages[pPageHead->iMonitoredPrev];
1283
1284	while (pPageHead)
1285	{
1286	unsigned idxNext = pPageHead->iMonitoredNext;
1287
1288	if (pPageHead != pPage)
1289	{
1290	STAM_COUNTER_INC(&pPool->StatDirtyPageDupFlush);
1291	Log(("Flush duplicate page idx=%d GCPhys=%RGp type=%s\n", pPageHead->idx, pPageHead->GCPhys, pgmPoolPoolKindToStr(pPageHead->enmKind)));
1292	int rc2 = pgmPoolFlushPage(pPool, pPageHead);
1293	AssertRC(rc2);
1294	}
1295
1296	if (idxNext == NIL_PGMPOOL_IDX)
1297	break;
1298
1299	pPageHead = &pPool->aPages[idxNext];
1300	}
1301	}
1302
1303	/* The flushing above might fail for locked pages, so double check. */
1304	if ( pPage->iMonitoredNext == NIL_PGMPOOL_IDX
1305	&& pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
1306	{
1307	pgmPoolAddDirtyPage(pVM, pPool, pPage);
1308
1309	/* Temporarily allow write access to the page table again. */
1310	rc = PGMHandlerPhysicalPageTempOff(pVM, pPage->GCPhys & PAGE_BASE_GC_MASK, pPage->GCPhys & PAGE_BASE_GC_MASK);
1311	if (rc == VINF_SUCCESS)
1312	{
1313	rc = PGMShwMakePageWritable(pVCpu, pvFault, PGM_MK_PG_IS_WRITE_FAULT);
1314	AssertMsg(rc == VINF_SUCCESS
1315	/* In the SMP case the page table might be removed while we wait for the PGM lock in the trap handler. */
1316	\|\| rc == VERR_PAGE_TABLE_NOT_PRESENT
1317	\|\| rc == VERR_PAGE_NOT_PRESENT,
1318	("PGMShwModifyPage -> GCPtr=%RGv rc=%d\n", pvFault, rc));
1319	# ifdef VBOX_STRICT
1320	pPage->GCPtrDirtyFault = pvFault;
1321	# endif
1322
1323	STAM_PROFILE_STOP(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), a);
1324	pgmUnlock(pVM);
1325	return rc;
1326	}
1327	}
1328	}
1329	#endif /* PGMPOOL_WITH_OPTIMIZED_DIRTY_PT */
1330
1331	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FlushModOverflow));
1332	flushPage:
1333	/*
1334	* Not worth it, so flush it.
1335	*
1336	* If we considered it to be reused, don't go back to ring-3
1337	* to emulate failed instructions since we usually cannot
1338	* interpret then. This may be a bit risky, in which case
1339	* the reuse detection must be fixed.
1340	*/
1341	rc = pgmPoolAccessHandlerFlush(pVM, pVCpu, pPool, pPage, pDis, pRegFrame, GCPhysFault, pvFault);
1342	if ( rc == VINF_EM_RAW_EMULATE_INSTR
1343	&& fReused)
1344	{
1345	/* Make sure that the current instruction still has shadow page backing, otherwise we'll end up in a loop. */
1346	if (PGMShwGetPage(pVCpu, pRegFrame->rip, NULL, NULL) == VINF_SUCCESS)
1347	rc = VINF_SUCCESS; /* safe to restart the instruction. */
1348	}
1349	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,FlushPage), a);
1350	pgmUnlock(pVM);
1351	return rc;
1352	}
1353
1354	# endif /* !IN_RING3 */
1355
1356	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1357
1358	# if defined(VBOX_STRICT) && !defined(IN_RING3)
1359
1360	/**
1361	* Check references to guest physical memory in a PAE / PAE page table.
1362	*
1363	* @param pPool The pool.
1364	* @param pPage The page.
1365	* @param pShwPT The shadow page table (mapping of the page).
1366	* @param pGstPT The guest page table.
1367	*/
1368	static void pgmPoolTrackCheckPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PTPAE pGstPT)
1369	{
1370	unsigned cErrors = 0;
1371	int LastRc = -1; /* initialized to shut up gcc */
1372	unsigned LastPTE = ~0U; /* initialized to shut up gcc */
1373	RTHCPHYS LastHCPhys = NIL_RTHCPHYS; /* initialized to shut up gcc */
1374	PVM pVM = pPool->CTX_SUFF(pVM);
1375
1376	#ifdef VBOX_STRICT
1377	for (unsigned i = 0; i < RT_MIN(RT_ELEMENTS(pShwPT->a), pPage->iFirstPresent); i++)
1378	AssertMsg(!PGMSHWPTEPAE_IS_P(pShwPT->a[i]), ("Unexpected PTE: idx=%d %RX64 (first=%d)\n", i, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), pPage->iFirstPresent));
1379	#endif
1380	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
1381	{
1382	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
1383	{
1384	RTHCPHYS HCPhys = NIL_RTHCPHYS;
1385	int rc = PGMPhysGCPhys2HCPhys(pVM, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK, &HCPhys);
1386	if ( rc != VINF_SUCCESS
1387	\|\| PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]) != HCPhys)
1388	{
1389	Log(("rc=%d idx=%d guest %RX64 shw=%RX64 vs %RHp\n", rc, i, pGstPT->a[i].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), HCPhys));
1390	LastPTE = i;
1391	LastRc = rc;
1392	LastHCPhys = HCPhys;
1393	cErrors++;
1394
1395	RTHCPHYS HCPhysPT = NIL_RTHCPHYS;
1396	rc = PGMPhysGCPhys2HCPhys(pVM, pPage->GCPhys, &HCPhysPT);
1397	AssertRC(rc);
1398
1399	for (unsigned iPage = 0; iPage < pPool->cCurPages; iPage++)
1400	{
1401	PPGMPOOLPAGE pTempPage = &pPool->aPages[iPage];
1402
1403	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1404	{
1405	PPGMSHWPTPAE pShwPT2 = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pVM, pTempPage);
1406
1407	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
1408	{
1409	if ( PGMSHWPTEPAE_IS_P_RW(pShwPT2->a[j])
1410	&& PGMSHWPTEPAE_GET_HCPHYS(pShwPT2->a[j]) == HCPhysPT)
1411	{
1412	Log(("GCPhys=%RGp idx=%d %RX64 vs %RX64\n", pTempPage->GCPhys, j, PGMSHWPTEPAE_GET_LOG(pShwPT->a[j]), PGMSHWPTEPAE_GET_LOG(pShwPT2->a[j])));
1413	}
1414	}
1415
1416	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pShwPT2);
1417	}
1418	}
1419	}
1420	}
1421	}
1422	AssertMsg(!cErrors, ("cErrors=%d: last rc=%d idx=%d guest %RX64 shw=%RX64 vs %RHp\n", cErrors, LastRc, LastPTE, pGstPT->a[LastPTE].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[LastPTE]), LastHCPhys));
1423	}
1424
1425
1426	/**
1427	* Check references to guest physical memory in a PAE / 32-bit page table.
1428	*
1429	* @param pPool The pool.
1430	* @param pPage The page.
1431	* @param pShwPT The shadow page table (mapping of the page).
1432	* @param pGstPT The guest page table.
1433	*/
1434	static void pgmPoolTrackCheckPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PT pGstPT)
1435	{
1436	unsigned cErrors = 0;
1437	int LastRc = -1; /* initialized to shut up gcc */
1438	unsigned LastPTE = ~0U; /* initialized to shut up gcc */
1439	RTHCPHYS LastHCPhys = NIL_RTHCPHYS; /* initialized to shut up gcc */
1440	PVM pVM = pPool->CTX_SUFF(pVM);
1441
1442	#ifdef VBOX_STRICT
1443	for (unsigned i = 0; i < RT_MIN(RT_ELEMENTS(pShwPT->a), pPage->iFirstPresent); i++)
1444	AssertMsg(!PGMSHWPTEPAE_IS_P(pShwPT->a[i]), ("Unexpected PTE: idx=%d %RX64 (first=%d)\n", i, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), pPage->iFirstPresent));
1445	#endif
1446	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
1447	{
1448	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
1449	{
1450	RTHCPHYS HCPhys = NIL_RTHCPHYS;
1451	int rc = PGMPhysGCPhys2HCPhys(pVM, pGstPT->a[i].u & X86_PTE_PG_MASK, &HCPhys);
1452	if ( rc != VINF_SUCCESS
1453	\|\| PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]) != HCPhys)
1454	{
1455	Log(("rc=%d idx=%d guest %x shw=%RX64 vs %RHp\n", rc, i, pGstPT->a[i].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), HCPhys));
1456	LastPTE = i;
1457	LastRc = rc;
1458	LastHCPhys = HCPhys;
1459	cErrors++;
1460
1461	RTHCPHYS HCPhysPT = NIL_RTHCPHYS;
1462	rc = PGMPhysGCPhys2HCPhys(pVM, pPage->GCPhys, &HCPhysPT);
1463	AssertRC(rc);
1464
1465	for (unsigned iPage = 0; iPage < pPool->cCurPages; iPage++)
1466	{
1467	PPGMPOOLPAGE pTempPage = &pPool->aPages[iPage];
1468
1469	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_32BIT_PT)
1470	{
1471	PPGMSHWPTPAE pShwPT2 = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pVM, pTempPage);
1472
1473	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
1474	{
1475	if ( PGMSHWPTEPAE_IS_P_RW(pShwPT2->a[j])
1476	&& PGMSHWPTEPAE_GET_HCPHYS(pShwPT2->a[j]) == HCPhysPT)
1477	{
1478	Log(("GCPhys=%RGp idx=%d %RX64 vs %RX64\n", pTempPage->GCPhys, j, PGMSHWPTEPAE_GET_LOG(pShwPT->a[j]), PGMSHWPTEPAE_GET_LOG(pShwPT2->a[j])));
1479	}
1480	}
1481
1482	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pShwPT2);
1483	}
1484	}
1485	}
1486	}
1487	}
1488	AssertMsg(!cErrors, ("cErrors=%d: last rc=%d idx=%d guest %x shw=%RX64 vs %RHp\n", cErrors, LastRc, LastPTE, pGstPT->a[LastPTE].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[LastPTE]), LastHCPhys));
1489	}
1490
1491	# endif /* VBOX_STRICT && !IN_RING3 */
1492
1493	/**
1494	* Clear references to guest physical memory in a PAE / PAE page table.
1495	*
1496	* @returns nr of changed PTEs
1497	* @param pPool The pool.
1498	* @param pPage The page.
1499	* @param pShwPT The shadow page table (mapping of the page).
1500	* @param pGstPT The guest page table.
1501	* @param pOldGstPT The old cached guest page table.
1502	* @param fAllowRemoval Bail out as soon as we encounter an invalid PTE
1503	* @param pfFlush Flush reused page table (out)
1504	*/
1505	DECLINLINE(unsigned) pgmPoolTrackFlushPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PTPAE pGstPT,
1506	PCX86PTPAE pOldGstPT, bool fAllowRemoval, bool *pfFlush)
1507	{
1508	unsigned cChanged = 0;
1509
1510	#ifdef VBOX_STRICT
1511	for (unsigned i = 0; i < RT_MIN(RT_ELEMENTS(pShwPT->a), pPage->iFirstPresent); i++)
1512	AssertMsg(!PGMSHWPTEPAE_IS_P(pShwPT->a[i]), ("Unexpected PTE: idx=%d %RX64 (first=%d)\n", i, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), pPage->iFirstPresent));
1513	#endif
1514	*pfFlush = false;
1515
1516	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
1517	{
1518	/* Check the new value written by the guest. If present and with a bogus physical address, then
1519	* it's fairly safe to assume the guest is reusing the PT.
1520	*/
1521	if ( fAllowRemoval
1522	&& pGstPT->a[i].n.u1Present)
1523	{
1524	if (!PGMPhysIsGCPhysValid(pPool->CTX_SUFF(pVM), pGstPT->a[i].u & X86_PTE_PAE_PG_MASK))
1525	{
1526	*pfFlush = true;
1527	return ++cChanged;
1528	}
1529	}
1530	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
1531	{
1532	/* If the old cached PTE is identical, then there's no need to flush the shadow copy. */
1533	if ((pGstPT->a[i].u & X86_PTE_PAE_PG_MASK) == (pOldGstPT->a[i].u & X86_PTE_PAE_PG_MASK))
1534	{
1535	#ifdef VBOX_STRICT
1536	RTHCPHYS HCPhys = NIL_RTGCPHYS;
1537	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[i].u & X86_PTE_PAE_PG_MASK, &HCPhys);
1538	AssertMsg(rc == VINF_SUCCESS && PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]) == HCPhys, ("rc=%d guest %RX64 old %RX64 shw=%RX64 vs %RHp\n", rc, pGstPT->a[i].u, pOldGstPT->a[i].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), HCPhys));
1539	#endif
1540	uint64_t uHostAttr = PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & (X86_PTE_P \| X86_PTE_US \| X86_PTE_A \| X86_PTE_D \| X86_PTE_G \| X86_PTE_PAE_NX);
1541	bool fHostRW = !!(PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & X86_PTE_RW);
1542	uint64_t uGuestAttr = pGstPT->a[i].u & (X86_PTE_P \| X86_PTE_US \| X86_PTE_A \| X86_PTE_D \| X86_PTE_G \| X86_PTE_PAE_NX);
1543	bool fGuestRW = !!(pGstPT->a[i].u & X86_PTE_RW);
1544
1545	if ( uHostAttr == uGuestAttr
1546	&& fHostRW <= fGuestRW)
1547	continue;
1548	}
1549	cChanged++;
1550	/* Something was changed, so flush it. */
1551	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX64 hint=%RX64\n",
1552	i, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pOldGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
1553	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pOldGstPT->a[i].u & X86_PTE_PAE_PG_MASK, i);
1554	PGMSHWPTEPAE_ATOMIC_SET(pShwPT->a[i], 0);
1555	}
1556	}
1557	return cChanged;
1558	}
1559
1560
1561	/**
1562	* Clear references to guest physical memory in a PAE / PAE page table.
1563	*
1564	* @returns nr of changed PTEs
1565	* @param pPool The pool.
1566	* @param pPage The page.
1567	* @param pShwPT The shadow page table (mapping of the page).
1568	* @param pGstPT The guest page table.
1569	* @param pOldGstPT The old cached guest page table.
1570	* @param fAllowRemoval Bail out as soon as we encounter an invalid PTE
1571	* @param pfFlush Flush reused page table (out)
1572	*/
1573	DECLINLINE(unsigned) pgmPoolTrackFlushPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PT pGstPT,
1574	PCX86PT pOldGstPT, bool fAllowRemoval, bool *pfFlush)
1575	{
1576	unsigned cChanged = 0;
1577
1578	#ifdef VBOX_STRICT
1579	for (unsigned i = 0; i < RT_MIN(RT_ELEMENTS(pShwPT->a), pPage->iFirstPresent); i++)
1580	AssertMsg(!PGMSHWPTEPAE_IS_P(pShwPT->a[i]), ("Unexpected PTE: idx=%d %RX64 (first=%d)\n", i, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), pPage->iFirstPresent));
1581	#endif
1582	*pfFlush = false;
1583
1584	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
1585	{
1586	/* Check the new value written by the guest. If present and with a bogus physical address, then
1587	* it's fairly safe to assume the guest is reusing the PT.
1588	*/
1589	if ( fAllowRemoval
1590	&& pGstPT->a[i].n.u1Present)
1591	{
1592	if (!PGMPhysIsGCPhysValid(pPool->CTX_SUFF(pVM), pGstPT->a[i].u & X86_PTE_PG_MASK))
1593	{
1594	*pfFlush = true;
1595	return ++cChanged;
1596	}
1597	}
1598	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
1599	{
1600	/* If the old cached PTE is identical, then there's no need to flush the shadow copy. */
1601	if ((pGstPT->a[i].u & X86_PTE_PG_MASK) == (pOldGstPT->a[i].u & X86_PTE_PG_MASK))
1602	{
1603	#ifdef VBOX_STRICT
1604	RTHCPHYS HCPhys = NIL_RTGCPHYS;
1605	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[i].u & X86_PTE_PG_MASK, &HCPhys);
1606	AssertMsg(rc == VINF_SUCCESS && PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]) == HCPhys, ("rc=%d guest %x old %x shw=%RX64 vs %RHp\n", rc, pGstPT->a[i].u, pOldGstPT->a[i].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), HCPhys));
1607	#endif
1608	uint64_t uHostAttr = PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & (X86_PTE_P \| X86_PTE_US \| X86_PTE_A \| X86_PTE_D \| X86_PTE_G);
1609	bool fHostRW = !!(PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & X86_PTE_RW);
1610	uint64_t uGuestAttr = pGstPT->a[i].u & (X86_PTE_P \| X86_PTE_US \| X86_PTE_A \| X86_PTE_D \| X86_PTE_G);
1611	bool fGuestRW = !!(pGstPT->a[i].u & X86_PTE_RW);
1612
1613	if ( uHostAttr == uGuestAttr
1614	&& fHostRW <= fGuestRW)
1615	continue;
1616	}
1617	cChanged++;
1618	/* Something was changed, so flush it. */
1619	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX64 hint=%x\n",
1620	i, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pOldGstPT->a[i].u & X86_PTE_PG_MASK));
1621	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pOldGstPT->a[i].u & X86_PTE_PG_MASK, i);
1622	PGMSHWPTEPAE_ATOMIC_SET(pShwPT->a[i], 0);
1623	}
1624	}
1625	return cChanged;
1626	}
1627
1628
1629	/**
1630	* Flush a dirty page
1631	*
1632	* @param pVM Pointer to the VM.
1633	* @param pPool The pool.
1634	* @param idxSlot Dirty array slot index
1635	* @param fAllowRemoval Allow a reused page table to be removed
1636	*/
1637	static void pgmPoolFlushDirtyPage(PVM pVM, PPGMPOOL pPool, unsigned idxSlot, bool fAllowRemoval = false)
1638	{
1639	PPGMPOOLPAGE pPage;
1640	unsigned idxPage;
1641
1642	Assert(idxSlot < RT_ELEMENTS(pPool->aDirtyPages));
1643	if (pPool->aDirtyPages[idxSlot].uIdx == NIL_PGMPOOL_IDX)
1644	return;
1645
1646	idxPage = pPool->aDirtyPages[idxSlot].uIdx;
1647	AssertRelease(idxPage != NIL_PGMPOOL_IDX);
1648	pPage = &pPool->aPages[idxPage];
1649	Assert(pPage->idx == idxPage);
1650	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1651
1652	AssertMsg(pPage->fDirty, ("Page %RGp (slot=%d) not marked dirty!", pPage->GCPhys, idxSlot));
1653	Log(("Flush dirty page %RGp cMods=%d\n", pPage->GCPhys, pPage->cModifications));
1654
1655	#if defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0) \|\| defined(IN_RC)
1656	PVMCPU pVCpu = VMMGetCpu(pVM);
1657	uint32_t iPrevSubset = PGMRZDynMapPushAutoSubset(pVCpu);
1658	#endif
1659
1660	/* First write protect the page again to catch all write accesses. (before checking for changes -> SMP) */
1661	int rc = PGMHandlerPhysicalReset(pVM, pPage->GCPhys & PAGE_BASE_GC_MASK);
1662	Assert(rc == VINF_SUCCESS);
1663	pPage->fDirty = false;
1664
1665	#ifdef VBOX_STRICT
1666	uint64_t fFlags = 0;
1667	RTHCPHYS HCPhys;
1668	rc = PGMShwGetPage(VMMGetCpu(pVM), pPage->GCPtrDirtyFault, &fFlags, &HCPhys);
1669	AssertMsg( ( rc == VINF_SUCCESS
1670	&& (!(fFlags & X86_PTE_RW) \|\| HCPhys != pPage->Core.Key))
1671	/* In the SMP case the page table might be removed while we wait for the PGM lock in the trap handler. */
1672	\|\| rc == VERR_PAGE_TABLE_NOT_PRESENT
1673	\|\| rc == VERR_PAGE_NOT_PRESENT,
1674	("PGMShwGetPage -> GCPtr=%RGv rc=%d flags=%RX64\n", pPage->GCPtrDirtyFault, rc, fFlags));
1675	#endif
1676
1677	/* Flush those PTEs that have changed. */
1678	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
1679	void *pvShw = PGMPOOL_PAGE_2_PTR(pVM, pPage);
1680	void *pvGst;
1681	rc = PGM_GCPHYS_2_PTR_EX(pVM, pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
1682	bool fFlush;
1683	unsigned cChanges;
1684
1685	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1686	cChanges = pgmPoolTrackFlushPTPaePae(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PTPAE)pvGst,
1687	(PCX86PTPAE)&pPool->aDirtyPages[idxSlot].aPage[0], fAllowRemoval, &fFlush);
1688	else
1689	cChanges = pgmPoolTrackFlushPTPae32Bit(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PT)pvGst,
1690	(PCX86PT)&pPool->aDirtyPages[idxSlot].aPage[0], fAllowRemoval, &fFlush);
1691
1692	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvGst);
1693	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvShw);
1694	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
1695	/* Note: we might want to consider keeping the dirty page active in case there were many changes. */
1696
1697	/* This page is likely to be modified again, so reduce the nr of modifications just a bit here. */
1698	Assert(pPage->cModifications);
1699	if (cChanges < 4)
1700	pPage->cModifications = 1; /* must use > 0 here */
1701	else
1702	pPage->cModifications = RT_MAX(1, pPage->cModifications / 2);
1703
1704	STAM_COUNTER_INC(&pPool->StatResetDirtyPages);
1705	if (pPool->cDirtyPages == RT_ELEMENTS(pPool->aDirtyPages))
1706	pPool->idxFreeDirtyPage = idxSlot;
1707
1708	pPool->cDirtyPages--;
1709	pPool->aDirtyPages[idxSlot].uIdx = NIL_PGMPOOL_IDX;
1710	Assert(pPool->cDirtyPages <= RT_ELEMENTS(pPool->aDirtyPages));
1711	if (fFlush)
1712	{
1713	Assert(fAllowRemoval);
1714	Log(("Flush reused page table!\n"));
1715	pgmPoolFlushPage(pPool, pPage);
1716	STAM_COUNTER_INC(&pPool->StatForceFlushReused);
1717	}
1718	else
1719	Log(("Removed dirty page %RGp cMods=%d cChanges=%d\n", pPage->GCPhys, pPage->cModifications, cChanges));
1720
1721	#if defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0) \|\| defined(IN_RC)
1722	PGMRZDynMapPopAutoSubset(pVCpu, iPrevSubset);
1723	#endif
1724	}
1725
1726
1727	# ifndef IN_RING3
1728	/**
1729	* Add a new dirty page
1730	*
1731	* @param pVM Pointer to the VM.
1732	* @param pPool The pool.
1733	* @param pPage The page.
1734	*/
1735	void pgmPoolAddDirtyPage(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1736	{
1737	unsigned idxFree;
1738
1739	PGM_LOCK_ASSERT_OWNER(pVM);
1740	AssertCompile(RT_ELEMENTS(pPool->aDirtyPages) == 8 \|\| RT_ELEMENTS(pPool->aDirtyPages) == 16);
1741	Assert(!pPage->fDirty);
1742
1743	idxFree = pPool->idxFreeDirtyPage;
1744	Assert(idxFree < RT_ELEMENTS(pPool->aDirtyPages));
1745	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1746
1747	if (pPool->cDirtyPages >= RT_ELEMENTS(pPool->aDirtyPages))
1748	{
1749	STAM_COUNTER_INC(&pPool->StatDirtyPageOverFlowFlush);
1750	pgmPoolFlushDirtyPage(pVM, pPool, idxFree, true /* allow removal of reused page tables*/);
1751	}
1752	Assert(pPool->cDirtyPages < RT_ELEMENTS(pPool->aDirtyPages));
1753	AssertMsg(pPool->aDirtyPages[idxFree].uIdx == NIL_PGMPOOL_IDX, ("idxFree=%d cDirtyPages=%d\n", idxFree, pPool->cDirtyPages));
1754
1755	Log(("Add dirty page %RGp (slot=%d)\n", pPage->GCPhys, idxFree));
1756
1757	/*
1758	* Make a copy of the guest page table as we require valid GCPhys addresses
1759	* when removing references to physical pages.
1760	* (The HCPhys linear lookup is extremely expensive!)
1761	*/
1762	void *pvGst;
1763	int rc = PGM_GCPHYS_2_PTR_EX(pVM, pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
1764	memcpy(&pPool->aDirtyPages[idxFree].aPage[0], pvGst, (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT) ? PAGE_SIZE : PAGE_SIZE/2);
1765	# ifdef VBOX_STRICT
1766	void *pvShw = PGMPOOL_PAGE_2_PTR(pVM, pPage);
1767	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1768	pgmPoolTrackCheckPTPaePae(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PTPAE)pvGst);
1769	else
1770	pgmPoolTrackCheckPTPae32Bit(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PT)pvGst);
1771	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvShw);
1772	# endif
1773	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvGst);
1774
1775	STAM_COUNTER_INC(&pPool->StatDirtyPage);
1776	pPage->fDirty = true;
1777	pPage->idxDirtyEntry = (uint8_t)idxFree; Assert(pPage->idxDirtyEntry == idxFree);
1778	pPool->aDirtyPages[idxFree].uIdx = pPage->idx;
1779	pPool->cDirtyPages++;
1780
1781	pPool->idxFreeDirtyPage = (pPool->idxFreeDirtyPage + 1) & (RT_ELEMENTS(pPool->aDirtyPages) - 1);
1782	if ( pPool->cDirtyPages < RT_ELEMENTS(pPool->aDirtyPages)
1783	&& pPool->aDirtyPages[pPool->idxFreeDirtyPage].uIdx != NIL_PGMPOOL_IDX)
1784	{
1785	unsigned i;
1786	for (i = 1; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
1787	{
1788	idxFree = (pPool->idxFreeDirtyPage + i) & (RT_ELEMENTS(pPool->aDirtyPages) - 1);
1789	if (pPool->aDirtyPages[idxFree].uIdx == NIL_PGMPOOL_IDX)
1790	{
1791	pPool->idxFreeDirtyPage = idxFree;
1792	break;
1793	}
1794	}
1795	Assert(i != RT_ELEMENTS(pPool->aDirtyPages));
1796	}
1797
1798	Assert(pPool->cDirtyPages == RT_ELEMENTS(pPool->aDirtyPages) \|\| pPool->aDirtyPages[pPool->idxFreeDirtyPage].uIdx == NIL_PGMPOOL_IDX);
1799	return;
1800	}
1801	# endif /* !IN_RING3 */
1802
1803
1804	/**
1805	* Check if the specified page is dirty (not write monitored)
1806	*
1807	* @return dirty or not
1808	* @param pVM Pointer to the VM.
1809	* @param GCPhys Guest physical address
1810	*/
1811	bool pgmPoolIsDirtyPage(PVM pVM, RTGCPHYS GCPhys)
1812	{
1813	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1814	PGM_LOCK_ASSERT_OWNER(pVM);
1815	if (!pPool->cDirtyPages)
1816	return false;
1817
1818	GCPhys = GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK;
1819
1820	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
1821	{
1822	if (pPool->aDirtyPages[i].uIdx != NIL_PGMPOOL_IDX)
1823	{
1824	PPGMPOOLPAGE pPage;
1825	unsigned idxPage = pPool->aDirtyPages[i].uIdx;
1826
1827	pPage = &pPool->aPages[idxPage];
1828	if (pPage->GCPhys == GCPhys)
1829	return true;
1830	}
1831	}
1832	return false;
1833	}
1834
1835
1836	/**
1837	* Reset all dirty pages by reinstating page monitoring.
1838	*
1839	* @param pVM Pointer to the VM.
1840	*/
1841	void pgmPoolResetDirtyPages(PVM pVM)
1842	{
1843	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1844	PGM_LOCK_ASSERT_OWNER(pVM);
1845	Assert(pPool->cDirtyPages <= RT_ELEMENTS(pPool->aDirtyPages));
1846
1847	if (!pPool->cDirtyPages)
1848	return;
1849
1850	Log(("pgmPoolResetDirtyPages\n"));
1851	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
1852	pgmPoolFlushDirtyPage(pVM, pPool, i, true /* allow removal of reused page tables*/);
1853
1854	pPool->idxFreeDirtyPage = 0;
1855	if ( pPool->cDirtyPages != RT_ELEMENTS(pPool->aDirtyPages)
1856	&& pPool->aDirtyPages[pPool->idxFreeDirtyPage].uIdx != NIL_PGMPOOL_IDX)
1857	{
1858	unsigned i;
1859	for (i = 1; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
1860	{
1861	if (pPool->aDirtyPages[i].uIdx == NIL_PGMPOOL_IDX)
1862	{
1863	pPool->idxFreeDirtyPage = i;
1864	break;
1865	}
1866	}
1867	AssertMsg(i != RT_ELEMENTS(pPool->aDirtyPages), ("cDirtyPages %d", pPool->cDirtyPages));
1868	}
1869
1870	Assert(pPool->aDirtyPages[pPool->idxFreeDirtyPage].uIdx == NIL_PGMPOOL_IDX \|\| pPool->cDirtyPages == RT_ELEMENTS(pPool->aDirtyPages));
1871	return;
1872	}
1873
1874
1875	/**
1876	* Invalidate the PT entry for the specified page
1877	*
1878	* @param pVM Pointer to the VM.
1879	* @param GCPtrPage Guest page to invalidate
1880	*/
1881	void pgmPoolResetDirtyPage(PVM pVM, RTGCPTR GCPtrPage)
1882	{
1883	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1884	PGM_LOCK_ASSERT_OWNER(pVM);
1885	Assert(pPool->cDirtyPages <= RT_ELEMENTS(pPool->aDirtyPages));
1886
1887	if (!pPool->cDirtyPages)
1888	return;
1889
1890	Log(("pgmPoolResetDirtyPage %RGv\n", GCPtrPage));
1891	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
1892	{
1893	}
1894	}
1895
1896
1897	/**
1898	* Reset all dirty pages by reinstating page monitoring.
1899	*
1900	* @param pVM Pointer to the VM.
1901	* @param GCPhysPT Physical address of the page table
1902	*/
1903	void pgmPoolInvalidateDirtyPage(PVM pVM, RTGCPHYS GCPhysPT)
1904	{
1905	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1906	PGM_LOCK_ASSERT_OWNER(pVM);
1907	Assert(pPool->cDirtyPages <= RT_ELEMENTS(pPool->aDirtyPages));
1908	unsigned idxDirtyPage = RT_ELEMENTS(pPool->aDirtyPages);
1909
1910	if (!pPool->cDirtyPages)
1911	return;
1912
1913	GCPhysPT = GCPhysPT & ~(RTGCPHYS)PAGE_OFFSET_MASK;
1914
1915	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
1916	{
1917	if (pPool->aDirtyPages[i].uIdx != NIL_PGMPOOL_IDX)
1918	{
1919	unsigned idxPage = pPool->aDirtyPages[i].uIdx;
1920
1921	PPGMPOOLPAGE pPage = &pPool->aPages[idxPage];
1922	if (pPage->GCPhys == GCPhysPT)
1923	{
1924	idxDirtyPage = i;
1925	break;
1926	}
1927	}
1928	}
1929
1930	if (idxDirtyPage != RT_ELEMENTS(pPool->aDirtyPages))
1931	{
1932	pgmPoolFlushDirtyPage(pVM, pPool, idxDirtyPage, true /* allow removal of reused page tables*/);
1933	if ( pPool->cDirtyPages != RT_ELEMENTS(pPool->aDirtyPages)
1934	&& pPool->aDirtyPages[pPool->idxFreeDirtyPage].uIdx != NIL_PGMPOOL_IDX)
1935	{
1936	unsigned i;
1937	for (i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
1938	{
1939	if (pPool->aDirtyPages[i].uIdx == NIL_PGMPOOL_IDX)
1940	{
1941	pPool->idxFreeDirtyPage = i;
1942	break;
1943	}
1944	}
1945	AssertMsg(i != RT_ELEMENTS(pPool->aDirtyPages), ("cDirtyPages %d", pPool->cDirtyPages));
1946	}
1947	}
1948	}
1949
1950	# endif /* PGMPOOL_WITH_OPTIMIZED_DIRTY_PT */
1951
1952	/**
1953	* Inserts a page into the GCPhys hash table.
1954	*
1955	* @param pPool The pool.
1956	* @param pPage The page.
1957	*/
1958	DECLINLINE(void) pgmPoolHashInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1959	{
1960	Log3(("pgmPoolHashInsert: %RGp\n", pPage->GCPhys));
1961	Assert(pPage->GCPhys != NIL_RTGCPHYS); Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1962	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1963	pPage->iNext = pPool->aiHash[iHash];
1964	pPool->aiHash[iHash] = pPage->idx;
1965	}
1966
1967
1968	/**
1969	* Removes a page from the GCPhys hash table.
1970	*
1971	* @param pPool The pool.
1972	* @param pPage The page.
1973	*/
1974	DECLINLINE(void) pgmPoolHashRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1975	{
1976	Log3(("pgmPoolHashRemove: %RGp\n", pPage->GCPhys));
1977	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1978	if (pPool->aiHash[iHash] == pPage->idx)
1979	pPool->aiHash[iHash] = pPage->iNext;
1980	else
1981	{
1982	uint16_t iPrev = pPool->aiHash[iHash];
1983	for (;;)
1984	{
1985	const int16_t i = pPool->aPages[iPrev].iNext;
1986	if (i == pPage->idx)
1987	{
1988	pPool->aPages[iPrev].iNext = pPage->iNext;
1989	break;
1990	}
1991	if (i == NIL_PGMPOOL_IDX)
1992	{
1993	AssertReleaseMsgFailed(("GCPhys=%RGp idx=%d\n", pPage->GCPhys, pPage->idx));
1994	break;
1995	}
1996	iPrev = i;
1997	}
1998	}
1999	pPage->iNext = NIL_PGMPOOL_IDX;
2000	}
2001
2002
2003	/**
2004	* Frees up one cache page.
2005	*
2006	* @returns VBox status code.
2007	* @retval VINF_SUCCESS on success.
2008	* @param pPool The pool.
2009	* @param iUser The user index.
2010	*/
2011	static int pgmPoolCacheFreeOne(PPGMPOOL pPool, uint16_t iUser)
2012	{
2013	#ifndef IN_RC
2014	const PVM pVM = pPool->CTX_SUFF(pVM);
2015	#endif
2016	Assert(pPool->iAgeHead != pPool->iAgeTail); /* We shouldn't be here if there < 2 cached entries! */
2017	STAM_COUNTER_INC(&pPool->StatCacheFreeUpOne);
2018
2019	/*
2020	* Select one page from the tail of the age list.
2021	*/
2022	PPGMPOOLPAGE pPage;
2023	for (unsigned iLoop = 0; ; iLoop++)
2024	{
2025	uint16_t iToFree = pPool->iAgeTail;
2026	if (iToFree == iUser && iUser != NIL_PGMPOOL_IDX)
2027	iToFree = pPool->aPages[iToFree].iAgePrev;
2028	/* This is the alternative to the SyncCR3 pgmPoolCacheUsed calls.
2029	if (pPool->aPages[iToFree].iUserHead != NIL_PGMPOOL_USER_INDEX)
2030	{
2031	uint16_t i = pPool->aPages[iToFree].iAgePrev;
2032	for (unsigned j = 0; j < 10 && i != NIL_PGMPOOL_USER_INDEX; j++, i = pPool->aPages[i].iAgePrev)
2033	{
2034	if (pPool->aPages[iToFree].iUserHead == NIL_PGMPOOL_USER_INDEX)
2035	continue;
2036	iToFree = i;
2037	break;
2038	}
2039	}
2040	*/
2041	Assert(iToFree != iUser);
2042	AssertRelease(iToFree != NIL_PGMPOOL_IDX);
2043	pPage = &pPool->aPages[iToFree];
2044
2045	/*
2046	* Reject any attempts at flushing the currently active shadow CR3 mapping.
2047	* Call pgmPoolCacheUsed to move the page to the head of the age list.
2048	*/
2049	if ( !pgmPoolIsPageLocked(pPage)
2050	&& pPage->idx >= PGMPOOL_IDX_FIRST /* paranoia (#6349) */)
2051	break;
2052	LogFlow(("pgmPoolCacheFreeOne: refuse CR3 mapping\n"));
2053	pgmPoolCacheUsed(pPool, pPage);
2054	AssertLogRelReturn(iLoop < 8192, VERR_PGM_POOL_TOO_MANY_LOOPS);
2055	}
2056
2057	/*
2058	* Found a usable page, flush it and return.
2059	*/
2060	int rc = pgmPoolFlushPage(pPool, pPage);
2061	/* This flush was initiated by us and not the guest, so explicitly flush the TLB. */
2062	/* todo: find out why this is necessary; pgmPoolFlushPage should trigger a flush if one is really needed. */
2063	if (rc == VINF_SUCCESS)
2064	PGM_INVL_ALL_VCPU_TLBS(pVM);
2065	return rc;
2066	}
2067
2068
2069	/**
2070	* Checks if a kind mismatch is really a page being reused
2071	* or if it's just normal remappings.
2072	*
2073	* @returns true if reused and the cached page (enmKind1) should be flushed
2074	* @returns false if not reused.
2075	* @param enmKind1 The kind of the cached page.
2076	* @param enmKind2 The kind of the requested page.
2077	*/
2078	static bool pgmPoolCacheReusedByKind(PGMPOOLKIND enmKind1, PGMPOOLKIND enmKind2)
2079	{
2080	switch (enmKind1)
2081	{
2082	/*
2083	* Never reuse them. There is no remapping in non-paging mode.
2084	*/
2085	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2086	case PGMPOOLKIND_32BIT_PD_PHYS:
2087	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2088	case PGMPOOLKIND_PAE_PD_PHYS:
2089	case PGMPOOLKIND_PAE_PDPT_PHYS:
2090	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2091	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2092	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2093	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2094	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2095	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT: /* never reuse them for other types */
2096	return false;
2097
2098	/*
2099	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
2100	*/
2101	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2102	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2103	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2104	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2105	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2106	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2107	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2108	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2109	case PGMPOOLKIND_32BIT_PD:
2110	case PGMPOOLKIND_PAE_PDPT:
2111	switch (enmKind2)
2112	{
2113	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2114	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2115	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2116	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2117	case PGMPOOLKIND_64BIT_PML4:
2118	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2119	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2120	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2121	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2122	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2123	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2124	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2125	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2126	return true;
2127	default:
2128	return false;
2129	}
2130
2131	/*
2132	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
2133	*/
2134	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2135	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2136	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2137	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2138	case PGMPOOLKIND_64BIT_PML4:
2139	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2140	switch (enmKind2)
2141	{
2142	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2143	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2144	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2145	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2146	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2147	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2148	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2149	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2150	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2151	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2152	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2153	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2154	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2155	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2156	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2157	return true;
2158	default:
2159	return false;
2160	}
2161
2162	/*
2163	* These cannot be flushed, and it's common to reuse the PDs as PTs.
2164	*/
2165	case PGMPOOLKIND_ROOT_NESTED:
2166	return false;
2167
2168	default:
2169	AssertFatalMsgFailed(("enmKind1=%d\n", enmKind1));
2170	}
2171	}
2172
2173
2174	/**
2175	* Attempts to satisfy a pgmPoolAlloc request from the cache.
2176	*
2177	* @returns VBox status code.
2178	* @retval VINF_PGM_CACHED_PAGE on success.
2179	* @retval VERR_FILE_NOT_FOUND if not found.
2180	* @param pPool The pool.
2181	* @param GCPhys The GC physical address of the page we're gonna shadow.
2182	* @param enmKind The kind of mapping.
2183	* @param enmAccess Access type for the mapping (only relevant for big pages)
2184	* @param fA20Enabled Whether the CPU has the A20 gate enabled.
2185	* @param iUser The shadow page pool index of the user table. This is
2186	* NIL_PGMPOOL_IDX for root pages.
2187	* @param iUserTable The index into the user table (shadowed). Ignored if
2188	* root page
2189	* @param ppPage Where to store the pointer to the page.
2190	*/
2191	static int pgmPoolCacheAlloc(PPGMPOOL pPool, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, PGMPOOLACCESS enmAccess, bool fA20Enabled,
2192	uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
2193	{
2194	/*
2195	* Look up the GCPhys in the hash.
2196	*/
2197	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
2198	Log3(("pgmPoolCacheAlloc: %RGp kind %s iUser=%d iUserTable=%x SLOT=%d\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable, i));
2199	if (i != NIL_PGMPOOL_IDX)
2200	{
2201	do
2202	{
2203	PPGMPOOLPAGE pPage = &pPool->aPages[i];
2204	Log4(("pgmPoolCacheAlloc: slot %d found page %RGp\n", i, pPage->GCPhys));
2205	if (pPage->GCPhys == GCPhys)
2206	{
2207	if ( (PGMPOOLKIND)pPage->enmKind == enmKind
2208	&& (PGMPOOLACCESS)pPage->enmAccess == enmAccess
2209	&& pPage->fA20Enabled == fA20Enabled)
2210	{
2211	/* Put it at the start of the use list to make sure pgmPoolTrackAddUser
2212	* doesn't flush it in case there are no more free use records.
2213	*/
2214	pgmPoolCacheUsed(pPool, pPage);
2215
2216	int rc = VINF_SUCCESS;
2217	if (iUser != NIL_PGMPOOL_IDX)
2218	rc = pgmPoolTrackAddUser(pPool, pPage, iUser, iUserTable);
2219	if (RT_SUCCESS(rc))
2220	{
2221	Assert((PGMPOOLKIND)pPage->enmKind == enmKind);
2222	*ppPage = pPage;
2223	if (pPage->cModifications)
2224	pPage->cModifications = 1; /* reset counter (can't use 0, or else it will be reinserted in the modified list) */
2225	STAM_COUNTER_INC(&pPool->StatCacheHits);
2226	return VINF_PGM_CACHED_PAGE;
2227	}
2228	return rc;
2229	}
2230
2231	if ((PGMPOOLKIND)pPage->enmKind != enmKind)
2232	{
2233	/*
2234	* The kind is different. In some cases we should now flush the page
2235	* as it has been reused, but in most cases this is normal remapping
2236	* of PDs as PT or big pages using the GCPhys field in a slightly
2237	* different way than the other kinds.
2238	*/
2239	if (pgmPoolCacheReusedByKind((PGMPOOLKIND)pPage->enmKind, enmKind))
2240	{
2241	STAM_COUNTER_INC(&pPool->StatCacheKindMismatches);
2242	pgmPoolFlushPage(pPool, pPage);
2243	break;
2244	}
2245	}
2246	}
2247
2248	/* next */
2249	i = pPage->iNext;
2250	} while (i != NIL_PGMPOOL_IDX);
2251	}
2252
2253	Log3(("pgmPoolCacheAlloc: Missed GCPhys=%RGp enmKind=%s\n", GCPhys, pgmPoolPoolKindToStr(enmKind)));
2254	STAM_COUNTER_INC(&pPool->StatCacheMisses);
2255	return VERR_FILE_NOT_FOUND;
2256	}
2257
2258
2259	/**
2260	* Inserts a page into the cache.
2261	*
2262	* @param pPool The pool.
2263	* @param pPage The cached page.
2264	* @param fCanBeCached Set if the page is fit for caching from the caller's point of view.
2265	*/
2266	static void pgmPoolCacheInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fCanBeCached)
2267	{
2268	/*
2269	* Insert into the GCPhys hash if the page is fit for that.
2270	*/
2271	Assert(!pPage->fCached);
2272	if (fCanBeCached)
2273	{
2274	pPage->fCached = true;
2275	pgmPoolHashInsert(pPool, pPage);
2276	Log3(("pgmPoolCacheInsert: Caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
2277	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
2278	STAM_COUNTER_INC(&pPool->StatCacheCacheable);
2279	}
2280	else
2281	{
2282	Log3(("pgmPoolCacheInsert: Not caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
2283	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
2284	STAM_COUNTER_INC(&pPool->StatCacheUncacheable);
2285	}
2286
2287	/*
2288	* Insert at the head of the age list.
2289	*/
2290	pPage->iAgePrev = NIL_PGMPOOL_IDX;
2291	pPage->iAgeNext = pPool->iAgeHead;
2292	if (pPool->iAgeHead != NIL_PGMPOOL_IDX)
2293	pPool->aPages[pPool->iAgeHead].iAgePrev = pPage->idx;
2294	else
2295	pPool->iAgeTail = pPage->idx;
2296	pPool->iAgeHead = pPage->idx;
2297	}
2298
2299
2300	/**
2301	* Flushes a cached page.
2302	*
2303	* @param pPool The pool.
2304	* @param pPage The cached page.
2305	*/
2306	static void pgmPoolCacheFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2307	{
2308	Log3(("pgmPoolCacheFlushPage: %RGp\n", pPage->GCPhys));
2309
2310	/*
2311	* Remove the page from the hash.
2312	*/
2313	if (pPage->fCached)
2314	{
2315	pPage->fCached = false;
2316	pgmPoolHashRemove(pPool, pPage);
2317	}
2318	else
2319	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
2320
2321	/*
2322	* Remove it from the age list.
2323	*/
2324	if (pPage->iAgeNext != NIL_PGMPOOL_IDX)
2325	pPool->aPages[pPage->iAgeNext].iAgePrev = pPage->iAgePrev;
2326	else
2327	pPool->iAgeTail = pPage->iAgePrev;
2328	if (pPage->iAgePrev != NIL_PGMPOOL_IDX)
2329	pPool->aPages[pPage->iAgePrev].iAgeNext = pPage->iAgeNext;
2330	else
2331	pPool->iAgeHead = pPage->iAgeNext;
2332	pPage->iAgeNext = NIL_PGMPOOL_IDX;
2333	pPage->iAgePrev = NIL_PGMPOOL_IDX;
2334	}
2335
2336
2337	/**
2338	* Looks for pages sharing the monitor.
2339	*
2340	* @returns Pointer to the head page.
2341	* @returns NULL if not found.
2342	* @param pPool The Pool
2343	* @param pNewPage The page which is going to be monitored.
2344	*/
2345	static PPGMPOOLPAGE pgmPoolMonitorGetPageByGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pNewPage)
2346	{
2347	/*
2348	* Look up the GCPhys in the hash.
2349	*/
2350	RTGCPHYS GCPhys = pNewPage->GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK;
2351	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
2352	if (i == NIL_PGMPOOL_IDX)
2353	return NULL;
2354	do
2355	{
2356	PPGMPOOLPAGE pPage = &pPool->aPages[i];
2357	if ( pPage->GCPhys - GCPhys < PAGE_SIZE
2358	&& pPage != pNewPage)
2359	{
2360	switch (pPage->enmKind)
2361	{
2362	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2363	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2364	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2365	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2366	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2367	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2368	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2369	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2370	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2371	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2372	case PGMPOOLKIND_64BIT_PML4:
2373	case PGMPOOLKIND_32BIT_PD:
2374	case PGMPOOLKIND_PAE_PDPT:
2375	{
2376	/* find the head */
2377	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
2378	{
2379	Assert(pPage->iMonitoredPrev != pPage->idx);
2380	pPage = &pPool->aPages[pPage->iMonitoredPrev];
2381	}
2382	return pPage;
2383	}
2384
2385	/* ignore, no monitoring. */
2386	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2387	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2388	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2389	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2390	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2391	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2392	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2393	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2394	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2395	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2396	case PGMPOOLKIND_ROOT_NESTED:
2397	case PGMPOOLKIND_PAE_PD_PHYS:
2398	case PGMPOOLKIND_PAE_PDPT_PHYS:
2399	case PGMPOOLKIND_32BIT_PD_PHYS:
2400	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2401	break;
2402	default:
2403	AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
2404	}
2405	}
2406
2407	/* next */
2408	i = pPage->iNext;
2409	} while (i != NIL_PGMPOOL_IDX);
2410	return NULL;
2411	}
2412
2413
2414	/**
2415	* Enabled write monitoring of a guest page.
2416	*
2417	* @returns VBox status code.
2418	* @retval VINF_SUCCESS on success.
2419	* @param pPool The pool.
2420	* @param pPage The cached page.
2421	*/
2422	static int pgmPoolMonitorInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2423	{
2424	LogFlow(("pgmPoolMonitorInsert %RGp\n", pPage->GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK));
2425
2426	/*
2427	* Filter out the relevant kinds.
2428	*/
2429	switch (pPage->enmKind)
2430	{
2431	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2432	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2433	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2434	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2435	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2436	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2437	case PGMPOOLKIND_64BIT_PML4:
2438	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2439	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2440	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2441	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2442	case PGMPOOLKIND_32BIT_PD:
2443	case PGMPOOLKIND_PAE_PDPT:
2444	break;
2445
2446	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2447	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2448	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2449	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2450	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2451	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2452	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2453	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2454	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2455	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2456	case PGMPOOLKIND_ROOT_NESTED:
2457	/* Nothing to monitor here. */
2458	return VINF_SUCCESS;
2459
2460	case PGMPOOLKIND_32BIT_PD_PHYS:
2461	case PGMPOOLKIND_PAE_PDPT_PHYS:
2462	case PGMPOOLKIND_PAE_PD_PHYS:
2463	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2464	/* Nothing to monitor here. */
2465	return VINF_SUCCESS;
2466	default:
2467	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
2468	}
2469
2470	/*
2471	* Install handler.
2472	*/
2473	int rc;
2474	PPGMPOOLPAGE pPageHead = pgmPoolMonitorGetPageByGCPhys(pPool, pPage);
2475	if (pPageHead)
2476	{
2477	Assert(pPageHead != pPage); Assert(pPageHead->iMonitoredNext != pPage->idx);
2478	Assert(pPageHead->iMonitoredPrev != pPage->idx);
2479
2480	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
2481	if (pPageHead->fDirty)
2482	pgmPoolFlushDirtyPage(pPool->CTX_SUFF(pVM), pPool, pPageHead->idxDirtyEntry, false /* do not remove */);
2483	#endif
2484
2485	pPage->iMonitoredPrev = pPageHead->idx;
2486	pPage->iMonitoredNext = pPageHead->iMonitoredNext;
2487	if (pPageHead->iMonitoredNext != NIL_PGMPOOL_IDX)
2488	pPool->aPages[pPageHead->iMonitoredNext].iMonitoredPrev = pPage->idx;
2489	pPageHead->iMonitoredNext = pPage->idx;
2490	rc = VINF_SUCCESS;
2491	}
2492	else
2493	{
2494	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX); Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
2495	PVM pVM = pPool->CTX_SUFF(pVM);
2496	const RTGCPHYS GCPhysPage = pPage->GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK;
2497	rc = PGMHandlerPhysicalRegisterEx(pVM, PGMPHYSHANDLERTYPE_PHYSICAL_WRITE,
2498	GCPhysPage, GCPhysPage + PAGE_OFFSET_MASK,
2499	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
2500	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
2501	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
2502	pPool->pszAccessHandler);
2503	/** @todo we should probably deal with out-of-memory conditions here, but for now increasing
2504	* the heap size should suffice. */
2505	AssertFatalMsgRC(rc, ("PGMHandlerPhysicalRegisterEx %RGp failed with %Rrc\n", GCPhysPage, rc));
2506	PVMCPU pVCpu = VMMGetCpu(pVM);
2507	AssertFatalMsg(!(pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3), ("fSyncFlags=%x syncff=%d\n", pVCpu->pgm.s.fSyncFlags, VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3)));
2508	}
2509	pPage->fMonitored = true;
2510	return rc;
2511	}
2512
2513
2514	/**
2515	* Disables write monitoring of a guest page.
2516	*
2517	* @returns VBox status code.
2518	* @retval VINF_SUCCESS on success.
2519	* @param pPool The pool.
2520	* @param pPage The cached page.
2521	*/
2522	static int pgmPoolMonitorFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2523	{
2524	/*
2525	* Filter out the relevant kinds.
2526	*/
2527	switch (pPage->enmKind)
2528	{
2529	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2530	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2531	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2532	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2533	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2534	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2535	case PGMPOOLKIND_64BIT_PML4:
2536	case PGMPOOLKIND_32BIT_PD:
2537	case PGMPOOLKIND_PAE_PDPT:
2538	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2539	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2540	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2541	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2542	break;
2543
2544	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2545	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2546	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2547	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2548	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2549	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2550	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2551	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2552	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2553	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2554	case PGMPOOLKIND_ROOT_NESTED:
2555	case PGMPOOLKIND_PAE_PD_PHYS:
2556	case PGMPOOLKIND_PAE_PDPT_PHYS:
2557	case PGMPOOLKIND_32BIT_PD_PHYS:
2558	/* Nothing to monitor here. */
2559	Assert(!pPage->fMonitored);
2560	return VINF_SUCCESS;
2561
2562	default:
2563	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
2564	}
2565	Assert(pPage->fMonitored);
2566
2567	/*
2568	* Remove the page from the monitored list or uninstall it if last.
2569	*/
2570	const PVM pVM = pPool->CTX_SUFF(pVM);
2571	int rc;
2572	if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
2573	\|\| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
2574	{
2575	if (pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
2576	{
2577	PPGMPOOLPAGE pNewHead = &pPool->aPages[pPage->iMonitoredNext];
2578	pNewHead->iMonitoredPrev = NIL_PGMPOOL_IDX;
2579	rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK,
2580	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pNewHead),
2581	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pNewHead),
2582	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pNewHead),
2583	pPool->pszAccessHandler);
2584	AssertFatalRCSuccess(rc);
2585	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
2586	}
2587	else
2588	{
2589	pPool->aPages[pPage->iMonitoredPrev].iMonitoredNext = pPage->iMonitoredNext;
2590	if (pPage->iMonitoredNext != NIL_PGMPOOL_IDX)
2591	{
2592	pPool->aPages[pPage->iMonitoredNext].iMonitoredPrev = pPage->iMonitoredPrev;
2593	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
2594	}
2595	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
2596	rc = VINF_SUCCESS;
2597	}
2598	}
2599	else
2600	{
2601	rc = PGMHandlerPhysicalDeregister(pVM, pPage->GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK);
2602	AssertFatalRC(rc);
2603	PVMCPU pVCpu = VMMGetCpu(pVM);
2604	AssertFatalMsg(!(pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3),
2605	("%#x %#x\n", pVCpu->pgm.s.fSyncFlags, pVM->fGlobalForcedActions));
2606	}
2607	pPage->fMonitored = false;
2608
2609	/*
2610	* Remove it from the list of modified pages (if in it).
2611	*/
2612	pgmPoolMonitorModifiedRemove(pPool, pPage);
2613
2614	return rc;
2615	}
2616
2617
2618	/**
2619	* Inserts the page into the list of modified pages.
2620	*
2621	* @param pPool The pool.
2622	* @param pPage The page.
2623	*/
2624	void pgmPoolMonitorModifiedInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2625	{
2626	Log3(("pgmPoolMonitorModifiedInsert: idx=%d\n", pPage->idx));
2627	AssertMsg( pPage->iModifiedNext == NIL_PGMPOOL_IDX
2628	&& pPage->iModifiedPrev == NIL_PGMPOOL_IDX
2629	&& pPool->iModifiedHead != pPage->idx,
2630	("Next=%d Prev=%d idx=%d cModifications=%d Head=%d cModifiedPages=%d\n",
2631	pPage->iModifiedNext, pPage->iModifiedPrev, pPage->idx, pPage->cModifications,
2632	pPool->iModifiedHead, pPool->cModifiedPages));
2633
2634	pPage->iModifiedNext = pPool->iModifiedHead;
2635	if (pPool->iModifiedHead != NIL_PGMPOOL_IDX)
2636	pPool->aPages[pPool->iModifiedHead].iModifiedPrev = pPage->idx;
2637	pPool->iModifiedHead = pPage->idx;
2638	pPool->cModifiedPages++;
2639	#ifdef VBOX_WITH_STATISTICS
2640	if (pPool->cModifiedPages > pPool->cModifiedPagesHigh)
2641	pPool->cModifiedPagesHigh = pPool->cModifiedPages;
2642	#endif
2643	}
2644
2645
2646	/**
2647	* Removes the page from the list of modified pages and resets the
2648	* modification counter.
2649	*
2650	* @param pPool The pool.
2651	* @param pPage The page which is believed to be in the list of modified pages.
2652	*/
2653	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2654	{
2655	Log3(("pgmPoolMonitorModifiedRemove: idx=%d cModifications=%d\n", pPage->idx, pPage->cModifications));
2656	if (pPool->iModifiedHead == pPage->idx)
2657	{
2658	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
2659	pPool->iModifiedHead = pPage->iModifiedNext;
2660	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
2661	{
2662	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = NIL_PGMPOOL_IDX;
2663	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2664	}
2665	pPool->cModifiedPages--;
2666	}
2667	else if (pPage->iModifiedPrev != NIL_PGMPOOL_IDX)
2668	{
2669	pPool->aPages[pPage->iModifiedPrev].iModifiedNext = pPage->iModifiedNext;
2670	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
2671	{
2672	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = pPage->iModifiedPrev;
2673	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2674	}
2675	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2676	pPool->cModifiedPages--;
2677	}
2678	else
2679	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
2680	pPage->cModifications = 0;
2681	}
2682
2683
2684	/**
2685	* Zaps the list of modified pages, resetting their modification counters in the process.
2686	*
2687	* @param pVM Pointer to the VM.
2688	*/
2689	static void pgmPoolMonitorModifiedClearAll(PVM pVM)
2690	{
2691	pgmLock(pVM);
2692	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2693	LogFlow(("pgmPoolMonitorModifiedClearAll: cModifiedPages=%d\n", pPool->cModifiedPages));
2694
2695	unsigned cPages = 0; NOREF(cPages);
2696
2697	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
2698	pgmPoolResetDirtyPages(pVM);
2699	#endif
2700
2701	uint16_t idx = pPool->iModifiedHead;
2702	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
2703	while (idx != NIL_PGMPOOL_IDX)
2704	{
2705	PPGMPOOLPAGE pPage = &pPool->aPages[idx];
2706	idx = pPage->iModifiedNext;
2707	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2708	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2709	pPage->cModifications = 0;
2710	Assert(++cPages);
2711	}
2712	AssertMsg(cPages == pPool->cModifiedPages, ("%d != %d\n", cPages, pPool->cModifiedPages));
2713	pPool->cModifiedPages = 0;
2714	pgmUnlock(pVM);
2715	}
2716
2717
2718	/**
2719	* Handle SyncCR3 pool tasks
2720	*
2721	* @returns VBox status code.
2722	* @retval VINF_SUCCESS if successfully added.
2723	* @retval VINF_PGM_SYNC_CR3 is it needs to be deferred to ring 3 (GC only)
2724	* @param pVCpu Pointer to the VMCPU.
2725	* @remark Should only be used when monitoring is available, thus placed in
2726	* the PGMPOOL_WITH_MONITORING #ifdef.
2727	*/
2728	int pgmPoolSyncCR3(PVMCPU pVCpu)
2729	{
2730	PVM pVM = pVCpu->CTX_SUFF(pVM);
2731	LogFlow(("pgmPoolSyncCR3 fSyncFlags=%x\n", pVCpu->pgm.s.fSyncFlags));
2732
2733	/*
2734	* When monitoring shadowed pages, we reset the modification counters on CR3 sync.
2735	* Occasionally we will have to clear all the shadow page tables because we wanted
2736	* to monitor a page which was mapped by too many shadowed page tables. This operation
2737	* sometimes referred to as a 'lightweight flush'.
2738	*/
2739	# ifdef IN_RING3 /* Don't flush in ring-0 or raw mode, it's taking too long. */
2740	if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
2741	pgmR3PoolClearAll(pVM, false /fFlushRemTlb/);
2742	# else /* !IN_RING3 */
2743	if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
2744	{
2745	Log(("SyncCR3: PGM_SYNC_CLEAR_PGM_POOL is set -> VINF_PGM_SYNC_CR3\n"));
2746	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
2747
2748	/* Make sure all other VCPUs return to ring 3. */
2749	if (pVM->cCpus > 1)
2750	{
2751	VM_FF_SET(pVM, VM_FF_PGM_POOL_FLUSH_PENDING);
2752	PGM_INVL_ALL_VCPU_TLBS(pVM);
2753	}
2754	return VINF_PGM_SYNC_CR3;
2755	}
2756	# endif /* !IN_RING3 */
2757	else
2758	{
2759	pgmPoolMonitorModifiedClearAll(pVM);
2760
2761	/* pgmPoolMonitorModifiedClearAll can cause a pgm pool flush (dirty page clearing), so make sure we handle this! */
2762	if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
2763	{
2764	Log(("pgmPoolMonitorModifiedClearAll caused a pgm flush -> call pgmPoolSyncCR3 again!\n"));
2765	return pgmPoolSyncCR3(pVCpu);
2766	}
2767	}
2768	return VINF_SUCCESS;
2769	}
2770
2771
2772	/**
2773	* Frees up at least one user entry.
2774	*
2775	* @returns VBox status code.
2776	* @retval VINF_SUCCESS if successfully added.
2777	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2778	* @param pPool The pool.
2779	* @param iUser The user index.
2780	*/
2781	static int pgmPoolTrackFreeOneUser(PPGMPOOL pPool, uint16_t iUser)
2782	{
2783	STAM_COUNTER_INC(&pPool->StatTrackFreeUpOneUser);
2784	/*
2785	* Just free cached pages in a braindead fashion.
2786	*/
2787	/** @todo walk the age list backwards and free the first with usage. */
2788	int rc = VINF_SUCCESS;
2789	do
2790	{
2791	int rc2 = pgmPoolCacheFreeOne(pPool, iUser);
2792	if (RT_FAILURE(rc2) && rc == VINF_SUCCESS)
2793	rc = rc2;
2794	} while (pPool->iUserFreeHead == NIL_PGMPOOL_USER_INDEX);
2795	return rc;
2796	}
2797
2798
2799	/**
2800	* Inserts a page into the cache.
2801	*
2802	* This will create user node for the page, insert it into the GCPhys
2803	* hash, and insert it into the age list.
2804	*
2805	* @returns VBox status code.
2806	* @retval VINF_SUCCESS if successfully added.
2807	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2808	* @param pPool The pool.
2809	* @param pPage The cached page.
2810	* @param GCPhys The GC physical address of the page we're gonna shadow.
2811	* @param iUser The user index.
2812	* @param iUserTable The user table index.
2813	*/
2814	DECLINLINE(int) pgmPoolTrackInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhys, uint16_t iUser, uint32_t iUserTable)
2815	{
2816	int rc = VINF_SUCCESS;
2817	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2818
2819	LogFlow(("pgmPoolTrackInsert GCPhys=%RGp iUser=%d iUserTable=%x\n", GCPhys, iUser, iUserTable));
2820
2821	if (iUser != NIL_PGMPOOL_IDX)
2822	{
2823	#ifdef VBOX_STRICT
2824	/*
2825	* Check that the entry doesn't already exists.
2826	*/
2827	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2828	{
2829	uint16_t i = pPage->iUserHead;
2830	do
2831	{
2832	Assert(i < pPool->cMaxUsers);
2833	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2834	i = paUsers[i].iNext;
2835	} while (i != NIL_PGMPOOL_USER_INDEX);
2836	}
2837	#endif
2838
2839	/*
2840	* Find free a user node.
2841	*/
2842	uint16_t i = pPool->iUserFreeHead;
2843	if (i == NIL_PGMPOOL_USER_INDEX)
2844	{
2845	rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2846	if (RT_FAILURE(rc))
2847	return rc;
2848	i = pPool->iUserFreeHead;
2849	}
2850
2851	/*
2852	* Unlink the user node from the free list,
2853	* initialize and insert it into the user list.
2854	*/
2855	pPool->iUserFreeHead = paUsers[i].iNext;
2856	paUsers[i].iNext = NIL_PGMPOOL_USER_INDEX;
2857	paUsers[i].iUser = iUser;
2858	paUsers[i].iUserTable = iUserTable;
2859	pPage->iUserHead = i;
2860	}
2861	else
2862	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
2863
2864
2865	/*
2866	* Insert into cache and enable monitoring of the guest page if enabled.
2867	*
2868	* Until we implement caching of all levels, including the CR3 one, we'll
2869	* have to make sure we don't try monitor & cache any recursive reuse of
2870	* a monitored CR3 page. Because all windows versions are doing this we'll
2871	* have to be able to do combined access monitoring, CR3 + PT and
2872	* PD + PT (guest PAE).
2873	*
2874	* Update:
2875	* We're now cooperating with the CR3 monitor if an uncachable page is found.
2876	*/
2877	const bool fCanBeMonitored = true;
2878	pgmPoolCacheInsert(pPool, pPage, fCanBeMonitored); /* This can be expanded. */
2879	if (fCanBeMonitored)
2880	{
2881	rc = pgmPoolMonitorInsert(pPool, pPage);
2882	AssertRC(rc);
2883	}
2884	return rc;
2885	}
2886
2887
2888	/**
2889	* Adds a user reference to a page.
2890	*
2891	* This will move the page to the head of the
2892	*
2893	* @returns VBox status code.
2894	* @retval VINF_SUCCESS if successfully added.
2895	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2896	* @param pPool The pool.
2897	* @param pPage The cached page.
2898	* @param iUser The user index.
2899	* @param iUserTable The user table.
2900	*/
2901	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2902	{
2903	Log3(("pgmPoolTrackAddUser: GCPhys=%RGp iUser=%%x iUserTable=%x\n", pPage->GCPhys, iUser, iUserTable));
2904	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2905	Assert(iUser != NIL_PGMPOOL_IDX);
2906
2907	# ifdef VBOX_STRICT
2908	/*
2909	* Check that the entry doesn't already exists. We only allow multiple
2910	* users of top-level paging structures (SHW_POOL_ROOT_IDX).
2911	*/
2912	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2913	{
2914	uint16_t i = pPage->iUserHead;
2915	do
2916	{
2917	Assert(i < pPool->cMaxUsers);
2918	/** @todo this assertion looks odd... Shouldn't it be && here? */
2919	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2920	i = paUsers[i].iNext;
2921	} while (i != NIL_PGMPOOL_USER_INDEX);
2922	}
2923	# endif
2924
2925	/*
2926	* Allocate a user node.
2927	*/
2928	uint16_t i = pPool->iUserFreeHead;
2929	if (i == NIL_PGMPOOL_USER_INDEX)
2930	{
2931	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2932	if (RT_FAILURE(rc))
2933	return rc;
2934	i = pPool->iUserFreeHead;
2935	}
2936	pPool->iUserFreeHead = paUsers[i].iNext;
2937
2938	/*
2939	* Initialize the user node and insert it.
2940	*/
2941	paUsers[i].iNext = pPage->iUserHead;
2942	paUsers[i].iUser = iUser;
2943	paUsers[i].iUserTable = iUserTable;
2944	pPage->iUserHead = i;
2945
2946	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
2947	if (pPage->fDirty)
2948	pgmPoolFlushDirtyPage(pPool->CTX_SUFF(pVM), pPool, pPage->idxDirtyEntry, false /* do not remove */);
2949	# endif
2950
2951	/*
2952	* Tell the cache to update its replacement stats for this page.
2953	*/
2954	pgmPoolCacheUsed(pPool, pPage);
2955	return VINF_SUCCESS;
2956	}
2957
2958
2959	/**
2960	* Frees a user record associated with a page.
2961	*
2962	* This does not clear the entry in the user table, it simply replaces the
2963	* user record to the chain of free records.
2964	*
2965	* @param pPool The pool.
2966	* @param HCPhys The HC physical address of the shadow page.
2967	* @param iUser The shadow page pool index of the user table.
2968	* @param iUserTable The index into the user table (shadowed).
2969	*
2970	* @remarks Don't call this for root pages.
2971	*/
2972	static void pgmPoolTrackFreeUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2973	{
2974	Log3(("pgmPoolTrackFreeUser %RGp %x %x\n", pPage->GCPhys, iUser, iUserTable));
2975	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2976	Assert(iUser != NIL_PGMPOOL_IDX);
2977
2978	/*
2979	* Unlink and free the specified user entry.
2980	*/
2981
2982	/* Special: For PAE and 32-bit paging, there is usually no more than one user. */
2983	uint16_t i = pPage->iUserHead;
2984	if ( i != NIL_PGMPOOL_USER_INDEX
2985	&& paUsers[i].iUser == iUser
2986	&& paUsers[i].iUserTable == iUserTable)
2987	{
2988	pPage->iUserHead = paUsers[i].iNext;
2989
2990	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2991	paUsers[i].iNext = pPool->iUserFreeHead;
2992	pPool->iUserFreeHead = i;
2993	return;
2994	}
2995
2996	/* General: Linear search. */
2997	uint16_t iPrev = NIL_PGMPOOL_USER_INDEX;
2998	while (i != NIL_PGMPOOL_USER_INDEX)
2999	{
3000	if ( paUsers[i].iUser == iUser
3001	&& paUsers[i].iUserTable == iUserTable)
3002	{
3003	if (iPrev != NIL_PGMPOOL_USER_INDEX)
3004	paUsers[iPrev].iNext = paUsers[i].iNext;
3005	else
3006	pPage->iUserHead = paUsers[i].iNext;
3007
3008	paUsers[i].iUser = NIL_PGMPOOL_IDX;
3009	paUsers[i].iNext = pPool->iUserFreeHead;
3010	pPool->iUserFreeHead = i;
3011	return;
3012	}
3013	iPrev = i;
3014	i = paUsers[i].iNext;
3015	}
3016
3017	/* Fatal: didn't find it */
3018	AssertFatalMsgFailed(("Didn't find the user entry! iUser=%d iUserTable=%#x GCPhys=%RGp\n",
3019	iUser, iUserTable, pPage->GCPhys));
3020	}
3021
3022
3023	/**
3024	* Gets the entry size of a shadow table.
3025	*
3026	* @param enmKind The kind of page.
3027	*
3028	* @returns The size of the entry in bytes. That is, 4 or 8.
3029	* @returns If the kind is not for a table, an assertion is raised and 0 is
3030	* returned.
3031	*/
3032	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind)
3033	{
3034	switch (enmKind)
3035	{
3036	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3037	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3038	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3039	case PGMPOOLKIND_32BIT_PD:
3040	case PGMPOOLKIND_32BIT_PD_PHYS:
3041	return 4;
3042
3043	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3044	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3045	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3046	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3047	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3048	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3049	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3050	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3051	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3052	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3053	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3054	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3055	case PGMPOOLKIND_64BIT_PML4:
3056	case PGMPOOLKIND_PAE_PDPT:
3057	case PGMPOOLKIND_ROOT_NESTED:
3058	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3059	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3060	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3061	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3062	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3063	case PGMPOOLKIND_PAE_PD_PHYS:
3064	case PGMPOOLKIND_PAE_PDPT_PHYS:
3065	return 8;
3066
3067	default:
3068	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
3069	}
3070	}
3071
3072
3073	/**
3074	* Gets the entry size of a guest table.
3075	*
3076	* @param enmKind The kind of page.
3077	*
3078	* @returns The size of the entry in bytes. That is, 0, 4 or 8.
3079	* @returns If the kind is not for a table, an assertion is raised and 0 is
3080	* returned.
3081	*/
3082	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind)
3083	{
3084	switch (enmKind)
3085	{
3086	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3087	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3088	case PGMPOOLKIND_32BIT_PD:
3089	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3090	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3091	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3092	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3093	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3094	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3095	return 4;
3096
3097	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3098	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3099	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3100	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3101	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3102	case PGMPOOLKIND_64BIT_PML4:
3103	case PGMPOOLKIND_PAE_PDPT:
3104	return 8;
3105
3106	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3107	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3108	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3109	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3110	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3111	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3112	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3113	case PGMPOOLKIND_ROOT_NESTED:
3114	case PGMPOOLKIND_PAE_PD_PHYS:
3115	case PGMPOOLKIND_PAE_PDPT_PHYS:
3116	case PGMPOOLKIND_32BIT_PD_PHYS:
3117	/** @todo can we return 0? (nobody is calling this...) */
3118	AssertFailed();
3119	return 0;
3120
3121	default:
3122	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
3123	}
3124	}
3125
3126
3127	/**
3128	* Checks one shadow page table entry for a mapping of a physical page.
3129	*
3130	* @returns true / false indicating removal of all relevant PTEs
3131	*
3132	* @param pVM Pointer to the VM.
3133	* @param pPhysPage The guest page in question.
3134	* @param fFlushPTEs Flush PTEs or allow them to be updated (e.g. in case of an RW bit change)
3135	* @param iShw The shadow page table.
3136	* @param iPte Page table entry or NIL_PGMPOOL_PHYSEXT_IDX_PTE if unknown
3137	*/
3138	static bool pgmPoolTrackFlushGCPhysPTInt(PVM pVM, PCPGMPAGE pPhysPage, bool fFlushPTEs, uint16_t iShw, uint16_t iPte)
3139	{
3140	LogFlow(("pgmPoolTrackFlushGCPhysPTInt: pPhysPage=%RHp iShw=%d iPte=%d\n", PGM_PAGE_GET_HCPHYS(pPhysPage), iShw, iPte));
3141	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3142	bool fRet = false;
3143
3144	/*
3145	* Assert sanity.
3146	*/
3147	Assert(iPte != NIL_PGMPOOL_PHYSEXT_IDX_PTE);
3148	AssertFatalMsg(iShw < pPool->cCurPages && iShw != NIL_PGMPOOL_IDX, ("iShw=%d\n", iShw));
3149	PPGMPOOLPAGE pPage = &pPool->aPages[iShw];
3150
3151	/*
3152	* Then, clear the actual mappings to the page in the shadow PT.
3153	*/
3154	switch (pPage->enmKind)
3155	{
3156	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3157	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3158	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3159	{
3160	const uint32_t u32 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
3161	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3162	uint32_t u32AndMask = 0;
3163	uint32_t u32OrMask = 0;
3164
3165	if (!fFlushPTEs)
3166	{
3167	switch (PGM_PAGE_GET_HNDL_PHYS_STATE(pPhysPage))
3168	{
3169	case PGM_PAGE_HNDL_PHYS_STATE_NONE: /** No handler installed. */
3170	case PGM_PAGE_HNDL_PHYS_STATE_DISABLED: /** Monitoring is temporarily disabled. */
3171	u32OrMask = X86_PTE_RW;
3172	u32AndMask = UINT32_MAX;
3173	fRet = true;
3174	STAM_COUNTER_INC(&pPool->StatTrackFlushEntryKeep);
3175	break;
3176
3177	case PGM_PAGE_HNDL_PHYS_STATE_WRITE: /** Write access is monitored. */
3178	u32OrMask = 0;
3179	u32AndMask = ~X86_PTE_RW;
3180	fRet = true;
3181	STAM_COUNTER_INC(&pPool->StatTrackFlushEntryKeep);
3182	break;
3183	default:
3184	/* (shouldn't be here, will assert below) */
3185	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3186	break;
3187	}
3188	}
3189	else
3190	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3191
3192	/* Update the counter if we're removing references. */
3193	if (!u32AndMask)
3194	{
3195	Assert(pPage->cPresent);
3196	Assert(pPool->cPresent);
3197	pPage->cPresent--;
3198	pPool->cPresent--;
3199	}
3200
3201	if ((pPT->a[iPte].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
3202	{
3203	X86PTE Pte;
3204
3205	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX32\n", iPte, pPT->a[iPte]));
3206	Pte.u = (pPT->a[iPte].u & u32AndMask) \| u32OrMask;
3207	if (Pte.u & PGM_PTFLAGS_TRACK_DIRTY)
3208	Pte.n.u1Write = 0; /* need to disallow writes when dirty bit tracking is still active. */
3209
3210	ASMAtomicWriteU32(&pPT->a[iPte].u, Pte.u);
3211	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);
3212	return fRet;
3213	}
3214	#ifdef LOG_ENABLED
3215	Log(("iFirstPresent=%d cPresent=%d\n", pPage->iFirstPresent, pPage->cPresent));
3216	for (unsigned i = 0, cFound = 0; i < RT_ELEMENTS(pPT->a); i++)
3217	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
3218	{
3219	Log(("i=%d cFound=%d\n", i, ++cFound));
3220	}
3221	#endif
3222	AssertFatalMsgFailed(("iFirstPresent=%d cPresent=%d u32=%RX32 poolkind=%x\n", pPage->iFirstPresent, pPage->cPresent, u32, pPage->enmKind));
3223	/PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);/
3224	break;
3225	}
3226
3227	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3228	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3229	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3230	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3231	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3232	case PGMPOOLKIND_EPT_PT_FOR_PHYS: /* physical mask the same as PAE; RW bit as well; be careful! */
3233	{
3234	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
3235	PPGMSHWPTPAE pPT = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3236	uint64_t u64OrMask = 0;
3237	uint64_t u64AndMask = 0;
3238
3239	if (!fFlushPTEs)
3240	{
3241	switch (PGM_PAGE_GET_HNDL_PHYS_STATE(pPhysPage))
3242	{
3243	case PGM_PAGE_HNDL_PHYS_STATE_NONE: /* No handler installed. */
3244	case PGM_PAGE_HNDL_PHYS_STATE_DISABLED: /* Monitoring is temporarily disabled. */
3245	u64OrMask = X86_PTE_RW;
3246	u64AndMask = UINT64_MAX;
3247	fRet = true;
3248	STAM_COUNTER_INC(&pPool->StatTrackFlushEntryKeep);
3249	break;
3250
3251	case PGM_PAGE_HNDL_PHYS_STATE_WRITE: /* Write access is monitored. */
3252	u64OrMask = 0;
3253	u64AndMask = ~(uint64_t)X86_PTE_RW;
3254	fRet = true;
3255	STAM_COUNTER_INC(&pPool->StatTrackFlushEntryKeep);
3256	break;
3257
3258	default:
3259	/* (shouldn't be here, will assert below) */
3260	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3261	break;
3262	}
3263	}
3264	else
3265	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3266
3267	/* Update the counter if we're removing references. */
3268	if (!u64AndMask)
3269	{
3270	Assert(pPage->cPresent);
3271	Assert(pPool->cPresent);
3272	pPage->cPresent--;
3273	pPool->cPresent--;
3274	}
3275
3276	if ((PGMSHWPTEPAE_GET_U(pPT->a[iPte]) & (X86_PTE_PAE_PG_MASK \| X86_PTE_P \| X86_PTE_PAE_MBZ_MASK_NX)) == u64)
3277	{
3278	X86PTEPAE Pte;
3279
3280	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64\n", iPte, PGMSHWPTEPAE_GET_LOG(pPT->a[iPte])));
3281	Pte.u = (PGMSHWPTEPAE_GET_U(pPT->a[iPte]) & u64AndMask) \| u64OrMask;
3282	if (Pte.u & PGM_PTFLAGS_TRACK_DIRTY)
3283	Pte.n.u1Write = 0; /* need to disallow writes when dirty bit tracking is still active. */
3284
3285	PGMSHWPTEPAE_ATOMIC_SET(pPT->a[iPte], Pte.u);
3286	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);
3287	return fRet;
3288	}
3289	#ifdef LOG_ENABLED
3290	Log(("iFirstPresent=%d cPresent=%d\n", pPage->iFirstPresent, pPage->cPresent));
3291	Log(("Found %RX64 expected %RX64\n", PGMSHWPTEPAE_GET_U(pPT->a[iPte]) & (X86_PTE_PAE_PG_MASK \| X86_PTE_P \| X86_PTE_PAE_MBZ_MASK_NX), u64));
3292	for (unsigned i = 0, cFound = 0; i < RT_ELEMENTS(pPT->a); i++)
3293	if ((PGMSHWPTEPAE_GET_U(pPT->a[i]) & (X86_PTE_PAE_PG_MASK \| X86_PTE_P \| X86_PTE_PAE_MBZ_MASK_NX)) == u64)
3294	Log(("i=%d cFound=%d\n", i, ++cFound));
3295	#endif
3296	AssertFatalMsgFailed(("iFirstPresent=%d cPresent=%d u64=%RX64 poolkind=%x iPte=%d PT=%RX64\n", pPage->iFirstPresent, pPage->cPresent, u64, pPage->enmKind, iPte, PGMSHWPTEPAE_GET_LOG(pPT->a[iPte])));
3297	/PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);/
3298	break;
3299	}
3300
3301	#ifdef PGM_WITH_LARGE_PAGES
3302	/* Large page case only. */
3303	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3304	{
3305	Assert(pVM->pgm.s.fNestedPaging);
3306
3307	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PDE4M_P \| X86_PDE4M_PS;
3308	PEPTPD pPD = (PEPTPD)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3309
3310	if ((pPD->a[iPte].u & (EPT_PDE2M_PG_MASK \| X86_PDE4M_P \| X86_PDE4M_PS)) == u64)
3311	{
3312	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pde=%RX64\n", iPte, pPD->a[iPte]));
3313	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3314	pPD->a[iPte].u = 0;
3315	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPD);
3316
3317	/* Update the counter as we're removing references. */
3318	Assert(pPage->cPresent);
3319	Assert(pPool->cPresent);
3320	pPage->cPresent--;
3321	pPool->cPresent--;
3322
3323	return fRet;
3324	}
3325	# ifdef LOG_ENABLED
3326	Log(("iFirstPresent=%d cPresent=%d\n", pPage->iFirstPresent, pPage->cPresent));
3327	for (unsigned i = 0, cFound = 0; i < RT_ELEMENTS(pPD->a); i++)
3328	if ((pPD->a[i].u & (EPT_PDE2M_PG_MASK \| X86_PDE4M_P \| X86_PDE4M_PS)) == u64)
3329	Log(("i=%d cFound=%d\n", i, ++cFound));
3330	# endif
3331	AssertFatalMsgFailed(("iFirstPresent=%d cPresent=%d\n", pPage->iFirstPresent, pPage->cPresent));
3332	/PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPD);/
3333	break;
3334	}
3335
3336	/* AMD-V nested paging / /* @todo merge with EPT as we only check the parts that are identical. */
3337	case PGMPOOLKIND_PAE_PD_PHYS:
3338	{
3339	Assert(pVM->pgm.s.fNestedPaging);
3340
3341	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PDE4M_P \| X86_PDE4M_PS;
3342	PX86PD pPD = (PX86PD)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3343
3344	if ((pPD->a[iPte].u & (X86_PDE2M_PAE_PG_MASK \| X86_PDE4M_P \| X86_PDE4M_PS)) == u64)
3345	{
3346	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pde=%RX64\n", iPte, pPD->a[iPte]));
3347	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3348	pPD->a[iPte].u = 0;
3349	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPD);
3350
3351	/* Update the counter as we're removing references. */
3352	Assert(pPage->cPresent);
3353	Assert(pPool->cPresent);
3354	pPage->cPresent--;
3355	pPool->cPresent--;
3356	return fRet;
3357	}
3358	# ifdef LOG_ENABLED
3359	Log(("iFirstPresent=%d cPresent=%d\n", pPage->iFirstPresent, pPage->cPresent));
3360	for (unsigned i = 0, cFound = 0; i < RT_ELEMENTS(pPD->a); i++)
3361	if ((pPD->a[i].u & (X86_PDE2M_PAE_PG_MASK \| X86_PDE4M_P \| X86_PDE4M_PS)) == u64)
3362	Log(("i=%d cFound=%d\n", i, ++cFound));
3363	# endif
3364	AssertFatalMsgFailed(("iFirstPresent=%d cPresent=%d\n", pPage->iFirstPresent, pPage->cPresent));
3365	/PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPD);/
3366	break;
3367	}
3368	#endif /* PGM_WITH_LARGE_PAGES */
3369
3370	default:
3371	AssertFatalMsgFailed(("enmKind=%d iShw=%d\n", pPage->enmKind, iShw));
3372	}
3373
3374	/* not reached. */
3375	#ifndef _MSC_VER
3376	return fRet;
3377	#endif
3378	}
3379
3380
3381	/**
3382	* Scans one shadow page table for mappings of a physical page.
3383	*
3384	* @param pVM Pointer to the VM.
3385	* @param pPhysPage The guest page in question.
3386	* @param fFlushPTEs Flush PTEs or allow them to be updated (e.g. in case of an RW bit change)
3387	* @param iShw The shadow page table.
3388	*/
3389	static void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, bool fFlushPTEs, uint16_t iShw)
3390	{
3391	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
3392
3393	/* We should only come here with when there's only one reference to this physical page. */
3394	Assert(PGMPOOL_TD_GET_CREFS(PGM_PAGE_GET_TRACKING(pPhysPage)) == 1);
3395
3396	Log2(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%RHp iShw=%d\n", PGM_PAGE_GET_HCPHYS(pPhysPage), iShw));
3397	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPT, f);
3398	bool fKeptPTEs = pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, fFlushPTEs, iShw, PGM_PAGE_GET_PTE_INDEX(pPhysPage));
3399	if (!fKeptPTEs)
3400	PGM_PAGE_SET_TRACKING(pVM, pPhysPage, 0);
3401	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPT, f);
3402	}
3403
3404
3405	/**
3406	* Flushes a list of shadow page tables mapping the same physical page.
3407	*
3408	* @param pVM Pointer to the VM.
3409	* @param pPhysPage The guest page in question.
3410	* @param fFlushPTEs Flush PTEs or allow them to be updated (e.g. in case of an RW bit change)
3411	* @param iPhysExt The physical cross reference extent list to flush.
3412	*/
3413	static void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, bool fFlushPTEs, uint16_t iPhysExt)
3414	{
3415	PGM_LOCK_ASSERT_OWNER(pVM);
3416	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3417	bool fKeepList = false;
3418
3419	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTs, f);
3420	Log2(("pgmPoolTrackFlushGCPhysPTs: pPhysPage=%RHp iPhysExt\n", PGM_PAGE_GET_HCPHYS(pPhysPage), iPhysExt));
3421
3422	const uint16_t iPhysExtStart = iPhysExt;
3423	PPGMPOOLPHYSEXT pPhysExt;
3424	do
3425	{
3426	Assert(iPhysExt < pPool->cMaxPhysExts);
3427	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3428	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3429	{
3430	if (pPhysExt->aidx[i] != NIL_PGMPOOL_IDX)
3431	{
3432	bool fKeptPTEs = pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, fFlushPTEs, pPhysExt->aidx[i], pPhysExt->apte[i]);
3433	if (!fKeptPTEs)
3434	{
3435	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3436	pPhysExt->apte[i] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
3437	}
3438	else
3439	fKeepList = true;
3440	}
3441	}
3442	/* next */
3443	iPhysExt = pPhysExt->iNext;
3444	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3445
3446	if (!fKeepList)
3447	{
3448	/* insert the list into the free list and clear the ram range entry. */
3449	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3450	pPool->iPhysExtFreeHead = iPhysExtStart;
3451	/* Invalidate the tracking data. */
3452	PGM_PAGE_SET_TRACKING(pVM, pPhysPage, 0);
3453	}
3454
3455	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTs, f);
3456	}
3457
3458
3459	/**
3460	* Flushes all shadow page table mappings of the given guest page.
3461	*
3462	* This is typically called when the host page backing the guest one has been
3463	* replaced or when the page protection was changed due to a guest access
3464	* caught by the monitoring.
3465	*
3466	* @returns VBox status code.
3467	* @retval VINF_SUCCESS if all references has been successfully cleared.
3468	* @retval VINF_PGM_SYNC_CR3 if we're better off with a CR3 sync and a page
3469	* pool cleaning. FF and sync flags are set.
3470	*
3471	* @param pVM Pointer to the VM.
3472	* @param GCPhysPage GC physical address of the page in question
3473	* @param pPhysPage The guest page in question.
3474	* @param fFlushPTEs Flush PTEs or allow them to be updated (e.g. in case of an RW bit change)
3475	* @param pfFlushTLBs This is set to @a true if the shadow TLBs should be
3476	* flushed, it is NOT touched if this isn't necessary.
3477	* The caller MUST initialized this to @a false.
3478	*/
3479	int pgmPoolTrackUpdateGCPhys(PVM pVM, RTGCPHYS GCPhysPage, PPGMPAGE pPhysPage, bool fFlushPTEs, bool *pfFlushTLBs)
3480	{
3481	PVMCPU pVCpu = VMMGetCpu(pVM);
3482	pgmLock(pVM);
3483	int rc = VINF_SUCCESS;
3484
3485	#ifdef PGM_WITH_LARGE_PAGES
3486	/* Is this page part of a large page? */
3487	if (PGM_PAGE_GET_PDE_TYPE(pPhysPage) == PGM_PAGE_PDE_TYPE_PDE)
3488	{
3489	RTGCPHYS GCPhysBase = GCPhysPage & X86_PDE2M_PAE_PG_MASK;
3490	GCPhysPage &= X86_PDE_PAE_PG_MASK;
3491
3492	/* Fetch the large page base. */
3493	PPGMPAGE pLargePage;
3494	if (GCPhysBase != GCPhysPage)
3495	{
3496	pLargePage = pgmPhysGetPage(pVM, GCPhysBase);
3497	AssertFatal(pLargePage);
3498	}
3499	else
3500	pLargePage = pPhysPage;
3501
3502	Log(("pgmPoolTrackUpdateGCPhys: update large page PDE for %RGp (%RGp)\n", GCPhysBase, GCPhysPage));
3503
3504	if (PGM_PAGE_GET_PDE_TYPE(pLargePage) == PGM_PAGE_PDE_TYPE_PDE)
3505	{
3506	/* Mark the large page as disabled as we need to break it up to change a single page in the 2 MB range. */
3507	PGM_PAGE_SET_PDE_TYPE(pVM, pLargePage, PGM_PAGE_PDE_TYPE_PDE_DISABLED);
3508	pVM->pgm.s.cLargePagesDisabled++;
3509
3510	/* Update the base as that only that one has a reference and there's only one PDE to clear. */
3511	rc = pgmPoolTrackUpdateGCPhys(pVM, GCPhysBase, pLargePage, fFlushPTEs, pfFlushTLBs);
3512
3513	*pfFlushTLBs = true;
3514	pgmUnlock(pVM);
3515	return rc;
3516	}
3517	}
3518	#else
3519	NOREF(GCPhysPage);
3520	#endif /* PGM_WITH_LARGE_PAGES */
3521
3522	const uint16_t u16 = PGM_PAGE_GET_TRACKING(pPhysPage);
3523	if (u16)
3524	{
3525	/*
3526	* The zero page is currently screwing up the tracking and we'll
3527	* have to flush the whole shebang. Unless VBOX_WITH_NEW_LAZY_PAGE_ALLOC
3528	* is defined, zero pages won't normally be mapped. Some kind of solution
3529	* will be needed for this problem of course, but it will have to wait...
3530	*/
3531	if ( PGM_PAGE_IS_ZERO(pPhysPage)
3532	\|\| PGM_PAGE_IS_BALLOONED(pPhysPage))
3533	rc = VINF_PGM_GCPHYS_ALIASED;
3534	else
3535	{
3536	# if defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0) \|\| defined(IN_RC) /** @todo we can drop this now. */
3537	/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow and
3538	pgmPoolTrackFlushGCPhysPTs will/may kill the pool otherwise. */
3539	uint32_t iPrevSubset = PGMRZDynMapPushAutoSubset(pVCpu);
3540	# endif
3541
3542	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
3543	{
3544	Assert(PGMPOOL_TD_GET_CREFS(u16) == 1);
3545	pgmPoolTrackFlushGCPhysPT(pVM,
3546	pPhysPage,
3547	fFlushPTEs,
3548	PGMPOOL_TD_GET_IDX(u16));
3549	}
3550	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
3551	pgmPoolTrackFlushGCPhysPTs(pVM, pPhysPage, fFlushPTEs, PGMPOOL_TD_GET_IDX(u16));
3552	else
3553	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
3554	*pfFlushTLBs = true;
3555
3556	# if defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0) \|\| defined(IN_RC)
3557	PGMRZDynMapPopAutoSubset(pVCpu, iPrevSubset);
3558	# endif
3559	}
3560	}
3561
3562	if (rc == VINF_PGM_GCPHYS_ALIASED)
3563	{
3564	pVCpu->pgm.s.fSyncFlags \|= PGM_SYNC_CLEAR_PGM_POOL;
3565	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
3566	rc = VINF_PGM_SYNC_CR3;
3567	}
3568	pgmUnlock(pVM);
3569	return rc;
3570	}
3571
3572
3573	/**
3574	* Scans all shadow page tables for mappings of a physical page.
3575	*
3576	* This may be slow, but it's most likely more efficient than cleaning
3577	* out the entire page pool / cache.
3578	*
3579	* @returns VBox status code.
3580	* @retval VINF_SUCCESS if all references has been successfully cleared.
3581	* @retval VINF_PGM_GCPHYS_ALIASED if we're better off with a CR3 sync and
3582	* a page pool cleaning.
3583	*
3584	* @param pVM Pointer to the VM.
3585	* @param pPhysPage The guest page in question.
3586	*/
3587	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage)
3588	{
3589	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3590	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTsSlow, s);
3591	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: cUsedPages=%d cPresent=%d pPhysPage=%R[pgmpage]\n",
3592	pPool->cUsedPages, pPool->cPresent, pPhysPage));
3593
3594	/*
3595	* There is a limit to what makes sense.
3596	*/
3597	if ( pPool->cPresent > 1024
3598	&& pVM->cCpus == 1)
3599	{
3600	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
3601	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
3602	return VINF_PGM_GCPHYS_ALIASED;
3603	}
3604
3605	/*
3606	* Iterate all the pages until we've encountered all that in use.
3607	* This is simple but not quite optimal solution.
3608	*/
3609	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P; /** @todo drop X86_PTE_P here as we always test if present separately, anyway. */
3610	const uint32_t u32 = u64; /** @todo move into the 32BIT_PT_xx case */
3611	unsigned cLeft = pPool->cUsedPages;
3612	unsigned iPage = pPool->cCurPages;
3613	while (--iPage >= PGMPOOL_IDX_FIRST)
3614	{
3615	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
3616	if ( pPage->GCPhys != NIL_RTGCPHYS
3617	&& pPage->cPresent)
3618	{
3619	switch (pPage->enmKind)
3620	{
3621	/*
3622	* We only care about shadow page tables.
3623	*/
3624	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3625	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3626	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3627	{
3628	unsigned cPresent = pPage->cPresent;
3629	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3630	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
3631	if (pPT->a[i].n.u1Present)
3632	{
3633	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
3634	{
3635	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX32\n", iPage, i, pPT->a[i]));
3636	pPT->a[i].u = 0;
3637
3638	/* Update the counter as we're removing references. */
3639	Assert(pPage->cPresent);
3640	Assert(pPool->cPresent);
3641	pPage->cPresent--;
3642	pPool->cPresent--;
3643	}
3644	if (!--cPresent)
3645	break;
3646	}
3647	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);
3648	break;
3649	}
3650
3651	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3652	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3653	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3654	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3655	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3656	{
3657	unsigned cPresent = pPage->cPresent;
3658	PPGMSHWPTPAE pPT = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3659	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
3660	if (PGMSHWPTEPAE_IS_P(pPT->a[i]))
3661	{
3662	if ((PGMSHWPTEPAE_GET_U(pPT->a[i]) & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
3663	{
3664	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
3665	PGMSHWPTEPAE_SET(pPT->a[i], 0); /// @todo why not atomic?
3666
3667	/* Update the counter as we're removing references. */
3668	Assert(pPage->cPresent);
3669	Assert(pPool->cPresent);
3670	pPage->cPresent--;
3671	pPool->cPresent--;
3672	}
3673	if (!--cPresent)
3674	break;
3675	}
3676	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);
3677	break;
3678	}
3679	#ifndef IN_RC
3680	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3681	{
3682	unsigned cPresent = pPage->cPresent;
3683	PEPTPT pPT = (PEPTPT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3684	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
3685	if (pPT->a[i].n.u1Present)
3686	{
3687	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
3688	{
3689	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
3690	pPT->a[i].u = 0;
3691
3692	/* Update the counter as we're removing references. */
3693	Assert(pPage->cPresent);
3694	Assert(pPool->cPresent);
3695	pPage->cPresent--;
3696	pPool->cPresent--;
3697	}
3698	if (!--cPresent)
3699	break;
3700	}
3701	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);
3702	break;
3703	}
3704	#endif
3705	}
3706	if (!--cLeft)
3707	break;
3708	}
3709	}
3710
3711	PGM_PAGE_SET_TRACKING(pVM, pPhysPage, 0);
3712	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
3713
3714	/*
3715	* There is a limit to what makes sense. The above search is very expensive, so force a pgm pool flush.
3716	*/
3717	if (pPool->cPresent > 1024)
3718	{
3719	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
3720	return VINF_PGM_GCPHYS_ALIASED;
3721	}
3722
3723	return VINF_SUCCESS;
3724	}
3725
3726
3727	/**
3728	* Clears the user entry in a user table.
3729	*
3730	* This is used to remove all references to a page when flushing it.
3731	*/
3732	static void pgmPoolTrackClearPageUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PCPGMPOOLUSER pUser)
3733	{
3734	Assert(pUser->iUser != NIL_PGMPOOL_IDX);
3735	Assert(pUser->iUser < pPool->cCurPages);
3736	uint32_t iUserTable = pUser->iUserTable;
3737
3738	/*
3739	* Map the user page. Ignore references made by fictitious pages.
3740	*/
3741	PPGMPOOLPAGE pUserPage = &pPool->aPages[pUser->iUser];
3742	LogFlow(("pgmPoolTrackClearPageUser: clear %x in %s (%RGp) (flushing %s)\n", iUserTable, pgmPoolPoolKindToStr(pUserPage->enmKind), pUserPage->Core.Key, pgmPoolPoolKindToStr(pPage->enmKind)));
3743	union
3744	{
3745	uint64_t *pau64;
3746	uint32_t *pau32;
3747	} u;
3748	if (pUserPage->idx < PGMPOOL_IDX_FIRST)
3749	{
3750	Assert(!pUserPage->pvPageR3);
3751	return;
3752	}
3753	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pUserPage);
3754
3755
3756	/* Safety precaution in case we change the paging for other modes too in the future. */
3757	Assert(!pgmPoolIsPageLocked(pPage));
3758
3759	#ifdef VBOX_STRICT
3760	/*
3761	* Some sanity checks.
3762	*/
3763	switch (pUserPage->enmKind)
3764	{
3765	case PGMPOOLKIND_32BIT_PD:
3766	case PGMPOOLKIND_32BIT_PD_PHYS:
3767	Assert(iUserTable < X86_PG_ENTRIES);
3768	break;
3769	case PGMPOOLKIND_PAE_PDPT:
3770	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
3771	case PGMPOOLKIND_PAE_PDPT_PHYS:
3772	Assert(iUserTable < 4);
3773	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
3774	break;
3775	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3776	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3777	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3778	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3779	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3780	case PGMPOOLKIND_PAE_PD_PHYS:
3781	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3782	break;
3783	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3784	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3785	Assert(!(u.pau64[iUserTable] & PGM_PDFLAGS_MAPPING));
3786	break;
3787	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3788	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3789	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
3790	break;
3791	case PGMPOOLKIND_64BIT_PML4:
3792	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
3793	/* GCPhys >> PAGE_SHIFT is the index here */
3794	break;
3795	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3796	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3797	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3798	break;
3799
3800	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3801	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3802	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3803	break;
3804
3805	case PGMPOOLKIND_ROOT_NESTED:
3806	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3807	break;
3808
3809	default:
3810	AssertMsgFailed(("enmKind=%d\n", pUserPage->enmKind));
3811	break;
3812	}
3813	#endif /* VBOX_STRICT */
3814
3815	/*
3816	* Clear the entry in the user page.
3817	*/
3818	switch (pUserPage->enmKind)
3819	{
3820	/* 32-bit entries */
3821	case PGMPOOLKIND_32BIT_PD:
3822	case PGMPOOLKIND_32BIT_PD_PHYS:
3823	ASMAtomicWriteU32(&u.pau32[iUserTable], 0);
3824	break;
3825
3826	/* 64-bit entries */
3827	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3828	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3829	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3830	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3831	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3832	#ifdef IN_RC
3833	/*
3834	* In 32 bits PAE mode we must invalidate the TLB when changing a
3835	* PDPT entry; the CPU fetches them only during cr3 load, so any
3836	* non-present PDPT will continue to cause page faults.
3837	*/
3838	ASMReloadCR3();
3839	/* no break */
3840	#endif
3841	case PGMPOOLKIND_PAE_PD_PHYS:
3842	case PGMPOOLKIND_PAE_PDPT_PHYS:
3843	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3844	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3845	case PGMPOOLKIND_64BIT_PML4:
3846	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3847	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3848	case PGMPOOLKIND_PAE_PDPT:
3849	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
3850	case PGMPOOLKIND_ROOT_NESTED:
3851	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3852	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3853	ASMAtomicWriteU64(&u.pau64[iUserTable], 0);
3854	break;
3855
3856	default:
3857	AssertFatalMsgFailed(("enmKind=%d iUser=%d iUserTable=%#x\n", pUserPage->enmKind, pUser->iUser, pUser->iUserTable));
3858	}
3859	PGM_DYNMAP_UNUSED_HINT_VM(pPool->CTX_SUFF(pVM), u.pau64);
3860	}
3861
3862
3863	/**
3864	* Clears all users of a page.
3865	*/
3866	static void pgmPoolTrackClearPageUsers(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3867	{
3868	/*
3869	* Free all the user records.
3870	*/
3871	LogFlow(("pgmPoolTrackClearPageUsers %RGp\n", pPage->GCPhys));
3872
3873	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
3874	uint16_t i = pPage->iUserHead;
3875	while (i != NIL_PGMPOOL_USER_INDEX)
3876	{
3877	/* Clear enter in user table. */
3878	pgmPoolTrackClearPageUser(pPool, pPage, &paUsers[i]);
3879
3880	/* Free it. */
3881	const uint16_t iNext = paUsers[i].iNext;
3882	paUsers[i].iUser = NIL_PGMPOOL_IDX;
3883	paUsers[i].iNext = pPool->iUserFreeHead;
3884	pPool->iUserFreeHead = i;
3885
3886	/* Next. */
3887	i = iNext;
3888	}
3889	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
3890	}
3891
3892
3893	/**
3894	* Allocates a new physical cross reference extent.
3895	*
3896	* @returns Pointer to the allocated extent on success. NULL if we're out of them.
3897	* @param pVM Pointer to the VM.
3898	* @param piPhysExt Where to store the phys ext index.
3899	*/
3900	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt)
3901	{
3902	PGM_LOCK_ASSERT_OWNER(pVM);
3903	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3904	uint16_t iPhysExt = pPool->iPhysExtFreeHead;
3905	if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
3906	{
3907	STAM_COUNTER_INC(&pPool->StamTrackPhysExtAllocFailures);
3908	return NULL;
3909	}
3910	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3911	pPool->iPhysExtFreeHead = pPhysExt->iNext;
3912	pPhysExt->iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
3913	*piPhysExt = iPhysExt;
3914	return pPhysExt;
3915	}
3916
3917
3918	/**
3919	* Frees a physical cross reference extent.
3920	*
3921	* @param pVM Pointer to the VM.
3922	* @param iPhysExt The extent to free.
3923	*/
3924	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt)
3925	{
3926	PGM_LOCK_ASSERT_OWNER(pVM);
3927	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3928	Assert(iPhysExt < pPool->cMaxPhysExts);
3929	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3930	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3931	{
3932	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3933	pPhysExt->apte[i] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
3934	}
3935	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3936	pPool->iPhysExtFreeHead = iPhysExt;
3937	}
3938
3939
3940	/**
3941	* Frees a physical cross reference extent.
3942	*
3943	* @param pVM Pointer to the VM.
3944	* @param iPhysExt The extent to free.
3945	*/
3946	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt)
3947	{
3948	PGM_LOCK_ASSERT_OWNER(pVM);
3949	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3950
3951	const uint16_t iPhysExtStart = iPhysExt;
3952	PPGMPOOLPHYSEXT pPhysExt;
3953	do
3954	{
3955	Assert(iPhysExt < pPool->cMaxPhysExts);
3956	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3957	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3958	{
3959	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3960	pPhysExt->apte[i] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
3961	}
3962
3963	/* next */
3964	iPhysExt = pPhysExt->iNext;
3965	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3966
3967	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3968	pPool->iPhysExtFreeHead = iPhysExtStart;
3969	}
3970
3971
3972	/**
3973	* Insert a reference into a list of physical cross reference extents.
3974	*
3975	* @returns The new tracking data for PGMPAGE.
3976	*
3977	* @param pVM Pointer to the VM.
3978	* @param iPhysExt The physical extent index of the list head.
3979	* @param iShwPT The shadow page table index.
3980	* @param iPte Page table entry
3981	*
3982	*/
3983	static uint16_t pgmPoolTrackPhysExtInsert(PVM pVM, uint16_t iPhysExt, uint16_t iShwPT, uint16_t iPte)
3984	{
3985	PGM_LOCK_ASSERT_OWNER(pVM);
3986	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3987	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3988
3989	/*
3990	* Special common cases.
3991	*/
3992	if (paPhysExts[iPhysExt].aidx[1] == NIL_PGMPOOL_IDX)
3993	{
3994	paPhysExts[iPhysExt].aidx[1] = iShwPT;
3995	paPhysExts[iPhysExt].apte[1] = iPte;
3996	STAM_COUNTER_INC(&pVM->pgm.s.CTX_SUFF(pStats)->StatTrackAliasedMany);
3997	LogFlow(("pgmPoolTrackPhysExtInsert: %d:{,%d pte %d,}\n", iPhysExt, iShwPT, iPte));
3998	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3999	}
4000	if (paPhysExts[iPhysExt].aidx[2] == NIL_PGMPOOL_IDX)
4001	{
4002	paPhysExts[iPhysExt].aidx[2] = iShwPT;
4003	paPhysExts[iPhysExt].apte[2] = iPte;
4004	STAM_COUNTER_INC(&pVM->pgm.s.CTX_SUFF(pStats)->StatTrackAliasedMany);
4005	LogFlow(("pgmPoolTrackPhysExtInsert: %d:{,,%d pte %d}\n", iPhysExt, iShwPT, iPte));
4006	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
4007	}
4008	AssertCompile(RT_ELEMENTS(paPhysExts[iPhysExt].aidx) == 3);
4009
4010	/*
4011	* General treatment.
4012	*/
4013	const uint16_t iPhysExtStart = iPhysExt;
4014	unsigned cMax = 15;
4015	for (;;)
4016	{
4017	Assert(iPhysExt < pPool->cMaxPhysExts);
4018	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
4019	if (paPhysExts[iPhysExt].aidx[i] == NIL_PGMPOOL_IDX)
4020	{
4021	paPhysExts[iPhysExt].aidx[i] = iShwPT;
4022	paPhysExts[iPhysExt].apte[i] = iPte;
4023	STAM_COUNTER_INC(&pVM->pgm.s.CTX_SUFF(pStats)->StatTrackAliasedMany);
4024	LogFlow(("pgmPoolTrackPhysExtInsert: %d:{%d pte %d} i=%d cMax=%d\n", iPhysExt, iShwPT, iPte, i, cMax));
4025	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtStart);
4026	}
4027	if (!--cMax)
4028	{
4029	STAM_COUNTER_INC(&pVM->pgm.s.CTX_SUFF(pStats)->StatTrackOverflows);
4030	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
4031	LogFlow(("pgmPoolTrackPhysExtInsert: overflow (1) iShwPT=%d\n", iShwPT));
4032	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
4033	}
4034
4035	/* advance */
4036	iPhysExt = paPhysExts[iPhysExt].iNext;
4037	if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
4038	break;
4039	}
4040
4041	/*
4042	* Add another extent to the list.
4043	*/
4044	PPGMPOOLPHYSEXT pNew = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
4045	if (!pNew)
4046	{
4047	STAM_COUNTER_INC(&pVM->pgm.s.CTX_SUFF(pStats)->StatTrackNoExtentsLeft);
4048	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
4049	LogFlow(("pgmPoolTrackPhysExtInsert: pgmPoolTrackPhysExtAlloc failed iShwPT=%d\n", iShwPT));
4050	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
4051	}
4052	pNew->iNext = iPhysExtStart;
4053	pNew->aidx[0] = iShwPT;
4054	pNew->apte[0] = iPte;
4055	LogFlow(("pgmPoolTrackPhysExtInsert: added new extent %d:{%d pte %d}->%d\n", iPhysExt, iShwPT, iPte, iPhysExtStart));
4056	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
4057	}
4058
4059
4060	/**
4061	* Add a reference to guest physical page where extents are in use.
4062	*
4063	* @returns The new tracking data for PGMPAGE.
4064	*
4065	* @param pVM Pointer to the VM.
4066	* @param pPhysPage Pointer to the aPages entry in the ram range.
4067	* @param u16 The ram range flags (top 16-bits).
4068	* @param iShwPT The shadow page table index.
4069	* @param iPte Page table entry
4070	*/
4071	uint16_t pgmPoolTrackPhysExtAddref(PVM pVM, PPGMPAGE pPhysPage, uint16_t u16, uint16_t iShwPT, uint16_t iPte)
4072	{
4073	pgmLock(pVM);
4074	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
4075	{
4076	/*
4077	* Convert to extent list.
4078	*/
4079	Assert(PGMPOOL_TD_GET_CREFS(u16) == 1);
4080	uint16_t iPhysExt;
4081	PPGMPOOLPHYSEXT pPhysExt = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
4082	if (pPhysExt)
4083	{
4084	LogFlow(("pgmPoolTrackPhysExtAddref: new extent: %d:{%d, %d}\n", iPhysExt, PGMPOOL_TD_GET_IDX(u16), iShwPT));
4085	STAM_COUNTER_INC(&pVM->pgm.s.CTX_SUFF(pStats)->StatTrackAliased);
4086	pPhysExt->aidx[0] = PGMPOOL_TD_GET_IDX(u16);
4087	pPhysExt->apte[0] = PGM_PAGE_GET_PTE_INDEX(pPhysPage);
4088	pPhysExt->aidx[1] = iShwPT;
4089	pPhysExt->apte[1] = iPte;
4090	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
4091	}
4092	else
4093	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
4094	}
4095	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
4096	{
4097	/*
4098	* Insert into the extent list.
4099	*/
4100	u16 = pgmPoolTrackPhysExtInsert(pVM, PGMPOOL_TD_GET_IDX(u16), iShwPT, iPte);
4101	}
4102	else
4103	STAM_COUNTER_INC(&pVM->pgm.s.CTX_SUFF(pStats)->StatTrackAliasedLots);
4104	pgmUnlock(pVM);
4105	return u16;
4106	}
4107
4108
4109	/**
4110	* Clear references to guest physical memory.
4111	*
4112	* @param pPool The pool.
4113	* @param pPage The page.
4114	* @param pPhysPage Pointer to the aPages entry in the ram range.
4115	* @param iPte Shadow PTE index
4116	*/
4117	void pgmPoolTrackPhysExtDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMPAGE pPhysPage, uint16_t iPte)
4118	{
4119	PVM pVM = pPool->CTX_SUFF(pVM);
4120	const unsigned cRefs = PGM_PAGE_GET_TD_CREFS(pPhysPage);
4121	AssertFatalMsg(cRefs == PGMPOOL_TD_CREFS_PHYSEXT, ("cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
4122
4123	uint16_t iPhysExt = PGM_PAGE_GET_TD_IDX(pPhysPage);
4124	if (iPhysExt != PGMPOOL_TD_IDX_OVERFLOWED)
4125	{
4126	pgmLock(pVM);
4127
4128	uint16_t iPhysExtPrev = NIL_PGMPOOL_PHYSEXT_INDEX;
4129	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
4130	do
4131	{
4132	Assert(iPhysExt < pPool->cMaxPhysExts);
4133
4134	/*
4135	* Look for the shadow page and check if it's all freed.
4136	*/
4137	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
4138	{
4139	if ( paPhysExts[iPhysExt].aidx[i] == pPage->idx
4140	&& paPhysExts[iPhysExt].apte[i] == iPte)
4141	{
4142	paPhysExts[iPhysExt].aidx[i] = NIL_PGMPOOL_IDX;
4143	paPhysExts[iPhysExt].apte[i] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
4144
4145	for (i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
4146	if (paPhysExts[iPhysExt].aidx[i] != NIL_PGMPOOL_IDX)
4147	{
4148	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
4149	pgmUnlock(pVM);
4150	return;
4151	}
4152
4153	/* we can free the node. */
4154	const uint16_t iPhysExtNext = paPhysExts[iPhysExt].iNext;
4155	if ( iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX
4156	&& iPhysExtNext == NIL_PGMPOOL_PHYSEXT_INDEX)
4157	{
4158	/* lonely node */
4159	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
4160	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d lonely\n", pPhysPage, pPage->idx));
4161	PGM_PAGE_SET_TRACKING(pVM, pPhysPage, 0);
4162	}
4163	else if (iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX)
4164	{
4165	/* head */
4166	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d head\n", pPhysPage, pPage->idx));
4167	PGM_PAGE_SET_TRACKING(pVM, pPhysPage, PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtNext));
4168	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
4169	}
4170	else
4171	{
4172	/* in list */
4173	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d in list\n", pPhysPage, pPage->idx));
4174	paPhysExts[iPhysExtPrev].iNext = iPhysExtNext;
4175	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
4176	}
4177	iPhysExt = iPhysExtNext;
4178	pgmUnlock(pVM);
4179	return;
4180	}
4181	}
4182
4183	/* next */
4184	iPhysExtPrev = iPhysExt;
4185	iPhysExt = paPhysExts[iPhysExt].iNext;
4186	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
4187
4188	pgmUnlock(pVM);
4189	AssertFatalMsgFailed(("not-found! cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
4190	}
4191	else /* nothing to do */
4192	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage]\n", pPhysPage));
4193	}
4194
4195	/**
4196	* Clear references to guest physical memory.
4197	*
4198	* This is the same as pgmPoolTracDerefGCPhysHint except that the guest
4199	* physical address is assumed to be correct, so the linear search can be
4200	* skipped and we can assert at an earlier point.
4201	*
4202	* @param pPool The pool.
4203	* @param pPage The page.
4204	* @param HCPhys The host physical address corresponding to the guest page.
4205	* @param GCPhys The guest physical address corresponding to HCPhys.
4206	* @param iPte Shadow PTE index
4207	*/
4208	static void pgmPoolTracDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhys, uint16_t iPte)
4209	{
4210	/*
4211	* Lookup the page and check if it checks out before derefing it.
4212	*/
4213	PVM pVM = pPool->CTX_SUFF(pVM);
4214	PPGMPAGE pPhysPage = pgmPhysGetPage(pVM, GCPhys);
4215	if (pPhysPage)
4216	{
4217	Assert(PGM_PAGE_GET_HCPHYS(pPhysPage));
4218	#ifdef LOG_ENABLED
4219	RTHCPHYS HCPhysPage = PGM_PAGE_GET_HCPHYS(pPhysPage);
4220	Log2(("pgmPoolTracDerefGCPhys %RHp vs %RHp\n", HCPhysPage, HCPhys));
4221	#endif
4222	if (PGM_PAGE_GET_HCPHYS(pPhysPage) == HCPhys)
4223	{
4224	Assert(pPage->cPresent);
4225	Assert(pPool->cPresent);
4226	pPage->cPresent--;
4227	pPool->cPresent--;
4228	pgmTrackDerefGCPhys(pPool, pPage, pPhysPage, iPte);
4229	return;
4230	}
4231
4232	AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp; found page has HCPhys=%RHp\n",
4233	HCPhys, GCPhys, PGM_PAGE_GET_HCPHYS(pPhysPage)));
4234	}
4235	AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp\n", HCPhys, GCPhys));
4236	}
4237
4238
4239	/**
4240	* Clear references to guest physical memory.
4241	*
4242	* @param pPool The pool.
4243	* @param pPage The page.
4244	* @param HCPhys The host physical address corresponding to the guest page.
4245	* @param GCPhysHint The guest physical address which may corresponding to HCPhys.
4246	* @param iPte Shadow pte index
4247	*/
4248	void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint, uint16_t iPte)
4249	{
4250	Log4(("pgmPoolTracDerefGCPhysHint %RHp %RGp\n", HCPhys, GCPhysHint));
4251
4252	/*
4253	* Try the hint first.
4254	*/
4255	RTHCPHYS HCPhysHinted;
4256	PVM pVM = pPool->CTX_SUFF(pVM);
4257	PPGMPAGE pPhysPage = pgmPhysGetPage(pVM, GCPhysHint);
4258	if (pPhysPage)
4259	{
4260	HCPhysHinted = PGM_PAGE_GET_HCPHYS(pPhysPage);
4261	Assert(HCPhysHinted);
4262	if (HCPhysHinted == HCPhys)
4263	{
4264	Assert(pPage->cPresent);
4265	Assert(pPool->cPresent);
4266	pPage->cPresent--;
4267	pPool->cPresent--;
4268	pgmTrackDerefGCPhys(pPool, pPage, pPhysPage, iPte);
4269	return;
4270	}
4271	}
4272	else
4273	HCPhysHinted = UINT64_C(0xdeadbeefdeadbeef);
4274
4275	/*
4276	* Damn, the hint didn't work. We'll have to do an expensive linear search.
4277	*/
4278	STAM_COUNTER_INC(&pPool->StatTrackLinearRamSearches);
4279	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRangesX);
4280	while (pRam)
4281	{
4282	unsigned iPage = pRam->cb >> PAGE_SHIFT;
4283	while (iPage-- > 0)
4284	{
4285	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
4286	{
4287	Log4(("pgmPoolTracDerefGCPhysHint: Linear HCPhys=%RHp GCPhysHint=%RGp GCPhysReal=%RGp\n",
4288	HCPhys, GCPhysHint, pRam->GCPhys + (iPage << PAGE_SHIFT)));
4289	Assert(pPage->cPresent);
4290	Assert(pPool->cPresent);
4291	pPage->cPresent--;
4292	pPool->cPresent--;
4293	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage], iPte);
4294	return;
4295	}
4296	}
4297	pRam = pRam->CTX_SUFF(pNext);
4298	}
4299
4300	AssertFatalMsgFailed(("HCPhys=%RHp GCPhysHint=%RGp (Hinted page has HCPhys = %RHp)\n", HCPhys, GCPhysHint, HCPhysHinted));
4301	}
4302
4303
4304	/**
4305	* Clear references to guest physical memory in a 32-bit / 32-bit page table.
4306	*
4307	* @param pPool The pool.
4308	* @param pPage The page.
4309	* @param pShwPT The shadow page table (mapping of the page).
4310	* @param pGstPT The guest page table.
4311	*/
4312	DECLINLINE(void) pgmPoolTrackDerefPT32Bit32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT, PCX86PT pGstPT)
4313	{
4314	RTGCPHYS32 const fPgMask = pPage->fA20Enabled ? X86_PTE_PG_MASK : X86_PTE_PG_MASK & ~RT_BIT_32(20);
4315	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
4316	{
4317	Assert(!(pShwPT->a[i].u & RT_BIT_32(10)));
4318	if (pShwPT->a[i].n.u1Present)
4319	{
4320	Log4(("pgmPoolTrackDerefPT32Bit32Bit: i=%d pte=%RX32 hint=%RX32\n",
4321	i, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
4322	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & fPgMask, i);
4323	if (!pPage->cPresent)
4324	break;
4325	}
4326	}
4327	}
4328
4329
4330	/**
4331	* Clear references to guest physical memory in a PAE / 32-bit page table.
4332	*
4333	* @param pPool The pool.
4334	* @param pPage The page.
4335	* @param pShwPT The shadow page table (mapping of the page).
4336	* @param pGstPT The guest page table (just a half one).
4337	*/
4338	DECLINLINE(void) pgmPoolTrackDerefPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PT pGstPT)
4339	{
4340	RTGCPHYS32 const fPgMask = pPage->fA20Enabled ? X86_PTE_PG_MASK : X86_PTE_PG_MASK & ~RT_BIT_32(20);
4341	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
4342	{
4343	Assert( (PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & UINT64_C(0x7ff0000000000400)) == 0
4344	\|\| (PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & UINT64_C(0x7ff0000000000400)) == UINT64_C(0x7ff0000000000000));
4345	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
4346	{
4347	Log4(("pgmPoolTrackDerefPTPae32Bit: i=%d pte=%RX64 hint=%RX32\n",
4348	i, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pGstPT->a[i].u & X86_PTE_PG_MASK));
4349	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pGstPT->a[i].u & fPgMask, i);
4350	if (!pPage->cPresent)
4351	break;
4352	}
4353	}
4354	}
4355
4356
4357	/**
4358	* Clear references to guest physical memory in a PAE / PAE page table.
4359	*
4360	* @param pPool The pool.
4361	* @param pPage The page.
4362	* @param pShwPT The shadow page table (mapping of the page).
4363	* @param pGstPT The guest page table.
4364	*/
4365	DECLINLINE(void) pgmPoolTrackDerefPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PTPAE pGstPT)
4366	{
4367	RTGCPHYS const fPgMask = pPage->fA20Enabled ? X86_PTE_PAE_PG_MASK : X86_PTE_PAE_PG_MASK & ~RT_BIT_64(20);
4368	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
4369	{
4370	Assert( (PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & UINT64_C(0x7ff0000000000400)) == 0
4371	\|\| (PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & UINT64_C(0x7ff0000000000400)) == UINT64_C(0x7ff0000000000000));
4372	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
4373	{
4374	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX32 hint=%RX32\n",
4375	i, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
4376	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pGstPT->a[i].u & fPgMask, i);
4377	if (!pPage->cPresent)
4378	break;
4379	}
4380	}
4381	}
4382
4383
4384	/**
4385	* Clear references to guest physical memory in a 32-bit / 4MB page table.
4386	*
4387	* @param pPool The pool.
4388	* @param pPage The page.
4389	* @param pShwPT The shadow page table (mapping of the page).
4390	*/
4391	DECLINLINE(void) pgmPoolTrackDerefPT32Bit4MB(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT)
4392	{
4393	RTGCPHYS const GCPhysA20Mask = pPage->fA20Enabled ? UINT64_MAX : ~RT_BIT_64(20);
4394	RTGCPHYS GCPhys = pPage->GCPhys + PAGE_SIZE * pPage->iFirstPresent;
4395	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
4396	{
4397	Assert(!(pShwPT->a[i].u & RT_BIT_32(10)));
4398	if (pShwPT->a[i].n.u1Present)
4399	{
4400	Log4(("pgmPoolTrackDerefPT32Bit4MB: i=%d pte=%RX32 GCPhys=%RGp\n",
4401	i, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys));
4402	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys & GCPhysA20Mask, i);
4403	if (!pPage->cPresent)
4404	break;
4405	}
4406	}
4407	}
4408
4409
4410	/**
4411	* Clear references to guest physical memory in a PAE / 2/4MB page table.
4412	*
4413	* @param pPool The pool.
4414	* @param pPage The page.
4415	* @param pShwPT The shadow page table (mapping of the page).
4416	*/
4417	DECLINLINE(void) pgmPoolTrackDerefPTPaeBig(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT)
4418	{
4419	RTGCPHYS const GCPhysA20Mask = pPage->fA20Enabled ? UINT64_MAX : ~RT_BIT_64(20);
4420	RTGCPHYS GCPhys = pPage->GCPhys + PAGE_SIZE * pPage->iFirstPresent;
4421	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
4422	{
4423	Assert( (PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & UINT64_C(0x7ff0000000000400)) == 0
4424	\|\| (PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & UINT64_C(0x7ff0000000000400)) == UINT64_C(0x7ff0000000000000));
4425	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
4426	{
4427	Log4(("pgmPoolTrackDerefPTPaeBig: i=%d pte=%RX64 hint=%RGp\n",
4428	i, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), GCPhys));
4429	pgmPoolTracDerefGCPhys(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), GCPhys & GCPhysA20Mask, i);
4430	if (!pPage->cPresent)
4431	break;
4432	}
4433	}
4434	}
4435
4436
4437	/**
4438	* Clear references to shadowed pages in an EPT page table.
4439	*
4440	* @param pPool The pool.
4441	* @param pPage The page.
4442	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
4443	*/
4444	DECLINLINE(void) pgmPoolTrackDerefPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT)
4445	{
4446	RTGCPHYS const GCPhysA20Mask = pPage->fA20Enabled ? UINT64_MAX : ~RT_BIT_64(20);
4447	RTGCPHYS GCPhys = pPage->GCPhys + PAGE_SIZE * pPage->iFirstPresent;
4448	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
4449	{
4450	Assert((pShwPT->a[i].u & UINT64_C(0xfff0000000000f80)) == 0);
4451	if (pShwPT->a[i].n.u1Present)
4452	{
4453	Log4(("pgmPoolTrackDerefPTEPT: i=%d pte=%RX64 GCPhys=%RX64\n",
4454	i, pShwPT->a[i].u & EPT_PTE_PG_MASK, pPage->GCPhys));
4455	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & EPT_PTE_PG_MASK, GCPhys & GCPhysA20Mask, i);
4456	if (!pPage->cPresent)
4457	break;
4458	}
4459	}
4460	}
4461
4462
4463	/**
4464	* Clear references to shadowed pages in a 32 bits page directory.
4465	*
4466	* @param pPool The pool.
4467	* @param pPage The page.
4468	* @param pShwPD The shadow page directory (mapping of the page).
4469	*/
4470	DECLINLINE(void) pgmPoolTrackDerefPD(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PD pShwPD)
4471	{
4472	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
4473	{
4474	Assert(!(pShwPD->a[i].u & RT_BIT_32(9)));
4475	if ( pShwPD->a[i].n.u1Present
4476	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
4477	)
4478	{
4479	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PG_MASK);
4480	if (pSubPage)
4481	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4482	else
4483	AssertFatalMsgFailed(("%x\n", pShwPD->a[i].u & X86_PDE_PG_MASK));
4484	}
4485	}
4486	}
4487
4488
4489	/**
4490	* Clear references to shadowed pages in a PAE (legacy or 64 bits) page directory.
4491	*
4492	* @param pPool The pool.
4493	* @param pPage The page.
4494	* @param pShwPD The shadow page directory (mapping of the page).
4495	*/
4496	DECLINLINE(void) pgmPoolTrackDerefPDPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPAE pShwPD)
4497	{
4498	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
4499	{
4500	if ( pShwPD->a[i].n.u1Present
4501	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING))
4502	{
4503	#ifdef PGM_WITH_LARGE_PAGES
4504	if (pShwPD->a[i].b.u1Size)
4505	{
4506	Log4(("pgmPoolTrackDerefPDPae: i=%d pde=%RX64 GCPhys=%RX64\n",
4507	i, pShwPD->a[i].u & X86_PDE2M_PAE_PG_MASK, pPage->GCPhys));
4508	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPD->a[i].u & X86_PDE2M_PAE_PG_MASK,
4509	pPage->GCPhys + i * 2 * _1M /* pPage->GCPhys = base address of the memory described by the PD */,
4510	i);
4511	}
4512	else
4513	#endif
4514	{
4515	Assert((pShwPD->a[i].u & (X86_PDE_PAE_MBZ_MASK_NX \| UINT64_C(0x7ff0000000000200))) == 0);
4516	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PAE_PG_MASK);
4517	if (pSubPage)
4518	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4519	else
4520	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & X86_PDE_PAE_PG_MASK));
4521	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4522	}
4523	}
4524	}
4525	}
4526
4527
4528	/**
4529	* Clear references to shadowed pages in a PAE page directory pointer table.
4530	*
4531	* @param pPool The pool.
4532	* @param pPage The page.
4533	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
4534	*/
4535	DECLINLINE(void) pgmPoolTrackDerefPDPTPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
4536	{
4537	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
4538	{
4539	Assert((pShwPDPT->a[i].u & (X86_PDPE_PAE_MBZ_MASK \| UINT64_C(0x7ff0000000000200))) == 0);
4540	if ( pShwPDPT->a[i].n.u1Present
4541	&& !(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING)
4542	)
4543	{
4544	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
4545	if (pSubPage)
4546	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4547	else
4548	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
4549	}
4550	}
4551	}
4552
4553
4554	/**
4555	* Clear references to shadowed pages in a 64-bit page directory pointer table.
4556	*
4557	* @param pPool The pool.
4558	* @param pPage The page.
4559	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
4560	*/
4561	DECLINLINE(void) pgmPoolTrackDerefPDPT64Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
4562	{
4563	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
4564	{
4565	Assert((pShwPDPT->a[i].u & (X86_PDPE_LM_MBZ_MASK_NX \| UINT64_C(0x7ff0000000000200))) == 0);
4566	if (pShwPDPT->a[i].n.u1Present)
4567	{
4568	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
4569	if (pSubPage)
4570	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4571	else
4572	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
4573	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4574	}
4575	}
4576	}
4577
4578
4579	/**
4580	* Clear references to shadowed pages in a 64-bit level 4 page table.
4581	*
4582	* @param pPool The pool.
4583	* @param pPage The page.
4584	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
4585	*/
4586	DECLINLINE(void) pgmPoolTrackDerefPML464Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PML4 pShwPML4)
4587	{
4588	for (unsigned i = 0; i < RT_ELEMENTS(pShwPML4->a); i++)
4589	{
4590	Assert((pShwPML4->a[i].u & (X86_PML4E_MBZ_MASK_NX \| UINT64_C(0x7ff0000000000200))) == 0);
4591	if (pShwPML4->a[i].n.u1Present)
4592	{
4593	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPML4->a[i].u & X86_PDPE_PG_MASK);
4594	if (pSubPage)
4595	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4596	else
4597	AssertFatalMsgFailed(("%RX64\n", pShwPML4->a[i].u & X86_PML4E_PG_MASK));
4598	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4599	}
4600	}
4601	}
4602
4603
4604	/**
4605	* Clear references to shadowed pages in an EPT page directory.
4606	*
4607	* @param pPool The pool.
4608	* @param pPage The page.
4609	* @param pShwPD The shadow page directory (mapping of the page).
4610	*/
4611	DECLINLINE(void) pgmPoolTrackDerefPDEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPD pShwPD)
4612	{
4613	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
4614	{
4615	Assert((pShwPD->a[i].u & UINT64_C(0xfff0000000000f80)) == 0);
4616	if (pShwPD->a[i].n.u1Present)
4617	{
4618	#ifdef PGM_WITH_LARGE_PAGES
4619	if (pShwPD->a[i].b.u1Size)
4620	{
4621	Log4(("pgmPoolTrackDerefPDEPT: i=%d pde=%RX64 GCPhys=%RX64\n",
4622	i, pShwPD->a[i].u & X86_PDE2M_PAE_PG_MASK, pPage->GCPhys));
4623	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPD->a[i].u & X86_PDE2M_PAE_PG_MASK,
4624	pPage->GCPhys + i * 2 * _1M /* pPage->GCPhys = base address of the memory described by the PD */,
4625	i);
4626	}
4627	else
4628	#endif
4629	{
4630	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & EPT_PDE_PG_MASK);
4631	if (pSubPage)
4632	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4633	else
4634	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & EPT_PDE_PG_MASK));
4635	}
4636	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4637	}
4638	}
4639	}
4640
4641
4642	/**
4643	* Clear references to shadowed pages in an EPT page directory pointer table.
4644	*
4645	* @param pPool The pool.
4646	* @param pPage The page.
4647	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
4648	*/
4649	DECLINLINE(void) pgmPoolTrackDerefPDPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPDPT pShwPDPT)
4650	{
4651	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
4652	{
4653	Assert((pShwPDPT->a[i].u & UINT64_C(0xfff0000000000f80)) == 0);
4654	if (pShwPDPT->a[i].n.u1Present)
4655	{
4656	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK);
4657	if (pSubPage)
4658	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4659	else
4660	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK));
4661	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4662	}
4663	}
4664	}
4665
4666
4667	/**
4668	* Clears all references made by this page.
4669	*
4670	* This includes other shadow pages and GC physical addresses.
4671	*
4672	* @param pPool The pool.
4673	* @param pPage The page.
4674	*/
4675	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
4676	{
4677	/*
4678	* Map the shadow page and take action according to the page kind.
4679	*/
4680	PVM pVM = pPool->CTX_SUFF(pVM);
4681	void *pvShw = PGMPOOL_PAGE_2_PTR(pVM, pPage);
4682	switch (pPage->enmKind)
4683	{
4684	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
4685	{
4686	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
4687	void *pvGst;
4688	int rc = PGM_GCPHYS_2_PTR(pVM, pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
4689	pgmPoolTrackDerefPT32Bit32Bit(pPool, pPage, (PX86PT)pvShw, (PCX86PT)pvGst);
4690	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvGst);
4691	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
4692	break;
4693	}
4694
4695	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
4696	{
4697	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
4698	void *pvGst;
4699	int rc = PGM_GCPHYS_2_PTR_EX(pVM, pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
4700	pgmPoolTrackDerefPTPae32Bit(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PT)pvGst);
4701	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvGst);
4702	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
4703	break;
4704	}
4705
4706	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
4707	{
4708	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
4709	void *pvGst;
4710	int rc = PGM_GCPHYS_2_PTR(pVM, pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
4711	pgmPoolTrackDerefPTPaePae(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PTPAE)pvGst);
4712	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvGst);
4713	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
4714	break;
4715	}
4716
4717	case PGMPOOLKIND_32BIT_PT_FOR_PHYS: /* treat it like a 4 MB page */
4718	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
4719	{
4720	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
4721	pgmPoolTrackDerefPT32Bit4MB(pPool, pPage, (PX86PT)pvShw);
4722	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
4723	break;
4724	}
4725
4726	case PGMPOOLKIND_PAE_PT_FOR_PHYS: /* treat it like a 2 MB page */
4727	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
4728	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
4729	{
4730	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
4731	pgmPoolTrackDerefPTPaeBig(pPool, pPage, (PPGMSHWPTPAE)pvShw);
4732	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
4733	break;
4734	}
4735
4736	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
4737	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
4738	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
4739	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
4740	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
4741	case PGMPOOLKIND_PAE_PD_PHYS:
4742	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
4743	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
4744	pgmPoolTrackDerefPDPae(pPool, pPage, (PX86PDPAE)pvShw);
4745	break;
4746
4747	case PGMPOOLKIND_32BIT_PD_PHYS:
4748	case PGMPOOLKIND_32BIT_PD:
4749	pgmPoolTrackDerefPD(pPool, pPage, (PX86PD)pvShw);
4750	break;
4751
4752	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
4753	case PGMPOOLKIND_PAE_PDPT:
4754	case PGMPOOLKIND_PAE_PDPT_PHYS:
4755	pgmPoolTrackDerefPDPTPae(pPool, pPage, (PX86PDPT)pvShw);
4756	break;
4757
4758	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
4759	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
4760	pgmPoolTrackDerefPDPT64Bit(pPool, pPage, (PX86PDPT)pvShw);
4761	break;
4762
4763	case PGMPOOLKIND_64BIT_PML4:
4764	pgmPoolTrackDerefPML464Bit(pPool, pPage, (PX86PML4)pvShw);
4765	break;
4766
4767	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
4768	pgmPoolTrackDerefPTEPT(pPool, pPage, (PEPTPT)pvShw);
4769	break;
4770
4771	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
4772	pgmPoolTrackDerefPDEPT(pPool, pPage, (PEPTPD)pvShw);
4773	break;
4774
4775	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
4776	pgmPoolTrackDerefPDPTEPT(pPool, pPage, (PEPTPDPT)pvShw);
4777	break;
4778
4779	default:
4780	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
4781	}
4782
4783	/* paranoia, clear the shadow page. Remove this laser (i.e. let Alloc and ClearAll do it). */
4784	STAM_PROFILE_START(&pPool->StatZeroPage, z);
4785	ASMMemZeroPage(pvShw);
4786	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
4787	pPage->fZeroed = true;
4788	Assert(!pPage->cPresent);
4789	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvShw);
4790	}
4791
4792
4793	/**
4794	* Flushes a pool page.
4795	*
4796	* This moves the page to the free list after removing all user references to it.
4797	*
4798	* @returns VBox status code.
4799	* @retval VINF_SUCCESS on success.
4800	* @param pPool The pool.
4801	* @param HCPhys The HC physical address of the shadow page.
4802	* @param fFlush Flush the TLBS when required (should only be false in very specific use cases!!)
4803	*/
4804	int pgmPoolFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fFlush)
4805	{
4806	PVM pVM = pPool->CTX_SUFF(pVM);
4807	bool fFlushRequired = false;
4808
4809	int rc = VINF_SUCCESS;
4810	STAM_PROFILE_START(&pPool->StatFlushPage, f);
4811	LogFlow(("pgmPoolFlushPage: pPage=%p:{.Key=%RHp, .idx=%d, .enmKind=%s, .GCPhys=%RGp}\n",
4812	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
4813
4814	/*
4815	* Reject any attempts at flushing any of the special root pages (shall
4816	* not happen).
4817	*/
4818	AssertMsgReturn(pPage->idx >= PGMPOOL_IDX_FIRST,
4819	("pgmPoolFlushPage: special root page, rejected. enmKind=%s idx=%d\n",
4820	pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx),
4821	VINF_SUCCESS);
4822
4823	pgmLock(pVM);
4824
4825	/*
4826	* Quietly reject any attempts at flushing the currently active shadow CR3 mapping
4827	*/
4828	if (pgmPoolIsPageLocked(pPage))
4829	{
4830	AssertMsg( pPage->enmKind == PGMPOOLKIND_64BIT_PML4
4831	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT
4832	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT_FOR_32BIT
4833	\|\| pPage->enmKind == PGMPOOLKIND_32BIT_PD
4834	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD
4835	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD
4836	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD
4837	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD
4838	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD
4839	\|\| pPage->enmKind == PGMPOOLKIND_ROOT_NESTED,
4840	("Can't free the shadow CR3! (%RHp vs %RHp kind=%d\n", PGMGetHyperCR3(VMMGetCpu(pVM)), pPage->Core.Key, pPage->enmKind));
4841	Log(("pgmPoolFlushPage: current active shadow CR3, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
4842	pgmUnlock(pVM);
4843	return VINF_SUCCESS;
4844	}
4845
4846	#if defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0) \|\| defined(IN_RC)
4847	/* Start a subset so we won't run out of mapping space. */
4848	PVMCPU pVCpu = VMMGetCpu(pVM);
4849	uint32_t iPrevSubset = PGMRZDynMapPushAutoSubset(pVCpu);
4850	#endif
4851
4852	/*
4853	* Mark the page as being in need of an ASMMemZeroPage().
4854	*/
4855	pPage->fZeroed = false;
4856
4857	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
4858	if (pPage->fDirty)
4859	pgmPoolFlushDirtyPage(pVM, pPool, pPage->idxDirtyEntry, false /* do not remove */);
4860	#endif
4861
4862	/* If there are any users of this table, then we must issue a tlb flush on all VCPUs. */
4863	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
4864	fFlushRequired = true;
4865
4866	/*
4867	* Clear the page.
4868	*/
4869	pgmPoolTrackClearPageUsers(pPool, pPage);
4870	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
4871	pgmPoolTrackDeref(pPool, pPage);
4872	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
4873
4874	/*
4875	* Flush it from the cache.
4876	*/
4877	pgmPoolCacheFlushPage(pPool, pPage);
4878
4879	#if defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0) \|\| defined(IN_RC)
4880	/* Heavy stuff done. */
4881	PGMRZDynMapPopAutoSubset(pVCpu, iPrevSubset);
4882	#endif
4883
4884	/*
4885	* Deregistering the monitoring.
4886	*/
4887	if (pPage->fMonitored)
4888	rc = pgmPoolMonitorFlush(pPool, pPage);
4889
4890	/*
4891	* Free the page.
4892	*/
4893	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
4894	pPage->iNext = pPool->iFreeHead;
4895	pPool->iFreeHead = pPage->idx;
4896	pPage->enmKind = PGMPOOLKIND_FREE;
4897	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
4898	pPage->GCPhys = NIL_RTGCPHYS;
4899	pPage->fReusedFlushPending = false;
4900
4901	pPool->cUsedPages--;
4902
4903	/* Flush the TLBs of all VCPUs if required. */
4904	if ( fFlushRequired
4905	&& fFlush)
4906	{
4907	PGM_INVL_ALL_VCPU_TLBS(pVM);
4908	}
4909
4910	pgmUnlock(pVM);
4911	STAM_PROFILE_STOP(&pPool->StatFlushPage, f);
4912	return rc;
4913	}
4914
4915
4916	/**
4917	* Frees a usage of a pool page.
4918	*
4919	* The caller is responsible to updating the user table so that it no longer
4920	* references the shadow page.
4921	*
4922	* @param pPool The pool.
4923	* @param HCPhys The HC physical address of the shadow page.
4924	* @param iUser The shadow page pool index of the user table.
4925	* NIL_PGMPOOL_IDX for root pages.
4926	* @param iUserTable The index into the user table (shadowed). Ignored if
4927	* root page.
4928	*/
4929	void pgmPoolFreeByPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
4930	{
4931	PVM pVM = pPool->CTX_SUFF(pVM);
4932
4933	STAM_PROFILE_START(&pPool->StatFree, a);
4934	LogFlow(("pgmPoolFreeByPage: pPage=%p:{.Key=%RHp, .idx=%d, enmKind=%s} iUser=%d iUserTable=%#x\n",
4935	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), iUser, iUserTable));
4936	AssertReturnVoid(pPage->idx >= PGMPOOL_IDX_FIRST); /* paranoia (#6349) */
4937
4938	pgmLock(pVM);
4939	if (iUser != NIL_PGMPOOL_IDX)
4940	pgmPoolTrackFreeUser(pPool, pPage, iUser, iUserTable);
4941	if (!pPage->fCached)
4942	pgmPoolFlushPage(pPool, pPage);
4943	pgmUnlock(pVM);
4944	STAM_PROFILE_STOP(&pPool->StatFree, a);
4945	}
4946
4947
4948	/**
4949	* Makes one or more free page free.
4950	*
4951	* @returns VBox status code.
4952	* @retval VINF_SUCCESS on success.
4953	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
4954	*
4955	* @param pPool The pool.
4956	* @param enmKind Page table kind
4957	* @param iUser The user of the page.
4958	*/
4959	static int pgmPoolMakeMoreFreePages(PPGMPOOL pPool, PGMPOOLKIND enmKind, uint16_t iUser)
4960	{
4961	PVM pVM = pPool->CTX_SUFF(pVM);
4962	LogFlow(("pgmPoolMakeMoreFreePages: enmKind=%d iUser=%d\n", enmKind, iUser));
4963	NOREF(enmKind);
4964
4965	/*
4966	* If the pool isn't full grown yet, expand it.
4967	*/
4968	if ( pPool->cCurPages < pPool->cMaxPages
4969	#if defined(IN_RC)
4970	/* Hack alert: we can't deal with jumps to ring 3 when called from MapCR3 and allocating pages for PAE PDs. */
4971	&& enmKind != PGMPOOLKIND_PAE_PD_FOR_PAE_PD
4972	&& (enmKind < PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD \|\| enmKind > PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD)
4973	#endif
4974	)
4975	{
4976	STAM_PROFILE_ADV_SUSPEND(&pPool->StatAlloc, a);
4977	#ifdef IN_RING3
4978	int rc = PGMR3PoolGrow(pVM);
4979	#else
4980	int rc = VMMRZCallRing3NoCpu(pVM, VMMCALLRING3_PGM_POOL_GROW, 0);
4981	#endif
4982	if (RT_FAILURE(rc))
4983	return rc;
4984	STAM_PROFILE_ADV_RESUME(&pPool->StatAlloc, a);
4985	if (pPool->iFreeHead != NIL_PGMPOOL_IDX)
4986	return VINF_SUCCESS;
4987	}
4988
4989	/*
4990	* Free one cached page.
4991	*/
4992	return pgmPoolCacheFreeOne(pPool, iUser);
4993	}
4994
4995
4996	/**
4997	* Allocates a page from the pool.
4998	*
4999	* This page may actually be a cached page and not in need of any processing
5000	* on the callers part.
5001	*
5002	* @returns VBox status code.
5003	* @retval VINF_SUCCESS if a NEW page was allocated.
5004	* @retval VINF_PGM_CACHED_PAGE if a CACHED page was returned.
5005	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
5006	*
5007	* @param pVM Pointer to the VM.
5008	* @param GCPhys The GC physical address of the page we're gonna shadow.
5009	* For 4MB and 2MB PD entries, it's the first address the
5010	* shadow PT is covering.
5011	* @param enmKind The kind of mapping.
5012	* @param enmAccess Access type for the mapping (only relevant for big pages)
5013	* @param fA20Enabled Whether the A20 gate is enabled or not.
5014	* @param iUser The shadow page pool index of the user table. Root
5015	* pages should pass NIL_PGMPOOL_IDX.
5016	* @param iUserTable The index into the user table (shadowed). Ignored for
5017	* root pages (iUser == NIL_PGMPOOL_IDX).
5018	* @param fLockPage Lock the page
5019	* @param ppPage Where to store the pointer to the page. NULL is stored here on failure.
5020	*/
5021	int pgmPoolAlloc(PVM pVM, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, PGMPOOLACCESS enmAccess, bool fA20Enabled,
5022	uint16_t iUser, uint32_t iUserTable, bool fLockPage, PPPGMPOOLPAGE ppPage)
5023	{
5024	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
5025	STAM_PROFILE_ADV_START(&pPool->StatAlloc, a);
5026	LogFlow(("pgmPoolAlloc: GCPhys=%RGp enmKind=%s iUser=%d iUserTable=%#x\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable));
5027	*ppPage = NULL;
5028	/** @todo CSAM/PGMPrefetchPage messes up here during CSAMR3CheckGates
5029	* (TRPMR3SyncIDT) because of FF priority. Try fix that?
5030	* Assert(!(pVM->pgm.s.fGlobalSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)); */
5031
5032	pgmLock(pVM);
5033
5034	if (pPool->fCacheEnabled)
5035	{
5036	int rc2 = pgmPoolCacheAlloc(pPool, GCPhys, enmKind, enmAccess, fA20Enabled, iUser, iUserTable, ppPage);
5037	if (RT_SUCCESS(rc2))
5038	{
5039	if (fLockPage)
5040	pgmPoolLockPage(pPool, *ppPage);
5041	pgmUnlock(pVM);
5042	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
5043	LogFlow(("pgmPoolAlloc: cached returns %Rrc ppPage=%p:{.Key=%RHp, .idx=%d}\n", rc2, ppPage, (ppPage)->Core.Key, (ppPage)->idx));
5044	return rc2;
5045	}
5046	}
5047
5048	/*
5049	* Allocate a new one.
5050	*/
5051	int rc = VINF_SUCCESS;
5052	uint16_t iNew = pPool->iFreeHead;
5053	if (iNew == NIL_PGMPOOL_IDX)
5054	{
5055	rc = pgmPoolMakeMoreFreePages(pPool, enmKind, iUser);
5056	if (RT_FAILURE(rc))
5057	{
5058	pgmUnlock(pVM);
5059	Log(("pgmPoolAlloc: returns %Rrc (Free)\n", rc));
5060	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
5061	return rc;
5062	}
5063	iNew = pPool->iFreeHead;
5064	AssertReleaseReturn(iNew != NIL_PGMPOOL_IDX, VERR_PGM_POOL_IPE);
5065	}
5066
5067	/* unlink the free head */
5068	PPGMPOOLPAGE pPage = &pPool->aPages[iNew];
5069	pPool->iFreeHead = pPage->iNext;
5070	pPage->iNext = NIL_PGMPOOL_IDX;
5071
5072	/*
5073	* Initialize it.
5074	*/
5075	pPool->cUsedPages++; /* physical handler registration / pgmPoolTrackFlushGCPhysPTsSlow requirement. */
5076	pPage->enmKind = enmKind;
5077	pPage->enmAccess = enmAccess;
5078	pPage->GCPhys = GCPhys;
5079	pPage->fA20Enabled = fA20Enabled;
5080	pPage->fSeenNonGlobal = false; /* Set this to 'true' to disable this feature. */
5081	pPage->fMonitored = false;
5082	pPage->fCached = false;
5083	pPage->fDirty = false;
5084	pPage->fReusedFlushPending = false;
5085	pPage->cModifications = 0;
5086	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
5087	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
5088	pPage->cPresent = 0;
5089	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
5090	pPage->idxDirtyEntry = 0;
5091	pPage->GCPtrLastAccessHandlerFault = NIL_RTGCPTR;
5092	pPage->GCPtrLastAccessHandlerRip = NIL_RTGCPTR;
5093	pPage->cLastAccessHandler = 0;
5094	pPage->cLocked = 0;
5095	# ifdef VBOX_STRICT
5096	pPage->GCPtrDirtyFault = NIL_RTGCPTR;
5097	# endif
5098
5099	/*
5100	* Insert into the tracking and cache. If this fails, free the page.
5101	*/
5102	int rc3 = pgmPoolTrackInsert(pPool, pPage, GCPhys, iUser, iUserTable);
5103	if (RT_FAILURE(rc3))
5104	{
5105	pPool->cUsedPages--;
5106	pPage->enmKind = PGMPOOLKIND_FREE;
5107	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
5108	pPage->GCPhys = NIL_RTGCPHYS;
5109	pPage->iNext = pPool->iFreeHead;
5110	pPool->iFreeHead = pPage->idx;
5111	pgmUnlock(pVM);
5112	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
5113	Log(("pgmPoolAlloc: returns %Rrc (Insert)\n", rc3));
5114	return rc3;
5115	}
5116
5117	/*
5118	* Commit the allocation, clear the page and return.
5119	*/
5120	#ifdef VBOX_WITH_STATISTICS
5121	if (pPool->cUsedPages > pPool->cUsedPagesHigh)
5122	pPool->cUsedPagesHigh = pPool->cUsedPages;
5123	#endif
5124
5125	if (!pPage->fZeroed)
5126	{
5127	STAM_PROFILE_START(&pPool->StatZeroPage, z);
5128	void *pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
5129	ASMMemZeroPage(pv);
5130	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
5131	}
5132
5133	*ppPage = pPage;
5134	if (fLockPage)
5135	pgmPoolLockPage(pPool, pPage);
5136	pgmUnlock(pVM);
5137	LogFlow(("pgmPoolAlloc: returns %Rrc *ppPage=%p:{.Key=%RHp, .idx=%d, .fCached=%RTbool, .fMonitored=%RTbool}\n",
5138	rc, pPage, pPage->Core.Key, pPage->idx, pPage->fCached, pPage->fMonitored));
5139	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
5140	return rc;
5141	}
5142
5143
5144	/**
5145	* Frees a usage of a pool page.
5146	*
5147	* @param pVM Pointer to the VM.
5148	* @param HCPhys The HC physical address of the shadow page.
5149	* @param iUser The shadow page pool index of the user table.
5150	* NIL_PGMPOOL_IDX if root page.
5151	* @param iUserTable The index into the user table (shadowed). Ignored if
5152	* root page.
5153	*/
5154	void pgmPoolFree(PVM pVM, RTHCPHYS HCPhys, uint16_t iUser, uint32_t iUserTable)
5155	{
5156	LogFlow(("pgmPoolFree: HCPhys=%RHp iUser=%d iUserTable=%#x\n", HCPhys, iUser, iUserTable));
5157	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
5158	pgmPoolFreeByPage(pPool, pgmPoolGetPage(pPool, HCPhys), iUser, iUserTable);
5159	}
5160
5161
5162	/**
5163	* Internal worker for finding a 'in-use' shadow page give by it's physical address.
5164	*
5165	* @returns Pointer to the shadow page structure.
5166	* @param pPool The pool.
5167	* @param HCPhys The HC physical address of the shadow page.
5168	*/
5169	PPGMPOOLPAGE pgmPoolGetPage(PPGMPOOL pPool, RTHCPHYS HCPhys)
5170	{
5171	PGM_LOCK_ASSERT_OWNER(pPool->CTX_SUFF(pVM));
5172
5173	/*
5174	* Look up the page.
5175	*/
5176	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, HCPhys & X86_PTE_PAE_PG_MASK);
5177
5178	AssertFatalMsg(pPage && pPage->enmKind != PGMPOOLKIND_FREE, ("HCPhys=%RHp pPage=%p idx=%d\n", HCPhys, pPage, (pPage) ? pPage->idx : 0));
5179	return pPage;
5180	}
5181
5182
5183	/**
5184	* Internal worker for finding a page for debugging purposes, no assertions.
5185	*
5186	* @returns Pointer to the shadow page structure. NULL on if not found.
5187	* @param pPool The pool.
5188	* @param HCPhys The HC physical address of the shadow page.
5189	*/
5190	PPGMPOOLPAGE pgmPoolQueryPageForDbg(PPGMPOOL pPool, RTHCPHYS HCPhys)
5191	{
5192	PGM_LOCK_ASSERT_OWNER(pPool->CTX_SUFF(pVM));
5193	return (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, HCPhys & X86_PTE_PAE_PG_MASK);
5194	}
5195
5196	#ifdef IN_RING3 /* currently only used in ring 3; save some space in the R0 & GC modules (left it here as we might need it elsewhere later on) */
5197
5198	/**
5199	* Flush the specified page if present
5200	*
5201	* @param pVM Pointer to the VM.
5202	* @param GCPhys Guest physical address of the page to flush
5203	*/
5204	void pgmPoolFlushPageByGCPhys(PVM pVM, RTGCPHYS GCPhys)
5205	{
5206	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
5207
5208	VM_ASSERT_EMT(pVM);
5209
5210	/*
5211	* Look up the GCPhys in the hash.
5212	*/
5213	GCPhys = GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK;
5214	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
5215	if (i == NIL_PGMPOOL_IDX)
5216	return;
5217
5218	do
5219	{
5220	PPGMPOOLPAGE pPage = &pPool->aPages[i];
5221	if (pPage->GCPhys - GCPhys < PAGE_SIZE)
5222	{
5223	switch (pPage->enmKind)
5224	{
5225	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
5226	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
5227	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
5228	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
5229	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
5230	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
5231	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
5232	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
5233	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
5234	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
5235	case PGMPOOLKIND_64BIT_PML4:
5236	case PGMPOOLKIND_32BIT_PD:
5237	case PGMPOOLKIND_PAE_PDPT:
5238	{
5239	Log(("PGMPoolFlushPage: found pgm pool pages for %RGp\n", GCPhys));
5240	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
5241	if (pPage->fDirty)
5242	STAM_COUNTER_INC(&pPool->StatForceFlushDirtyPage);
5243	else
5244	#endif
5245	STAM_COUNTER_INC(&pPool->StatForceFlushPage);
5246	Assert(!pgmPoolIsPageLocked(pPage));
5247	pgmPoolMonitorChainFlush(pPool, pPage);
5248	return;
5249	}
5250
5251	/* ignore, no monitoring. */
5252	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
5253	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
5254	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
5255	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
5256	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
5257	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
5258	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
5259	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
5260	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
5261	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
5262	case PGMPOOLKIND_ROOT_NESTED:
5263	case PGMPOOLKIND_PAE_PD_PHYS:
5264	case PGMPOOLKIND_PAE_PDPT_PHYS:
5265	case PGMPOOLKIND_32BIT_PD_PHYS:
5266	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
5267	break;
5268
5269	default:
5270	AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
5271	}
5272	}
5273
5274	/* next */
5275	i = pPage->iNext;
5276	} while (i != NIL_PGMPOOL_IDX);
5277	return;
5278	}
5279
5280	#endif /* IN_RING3 */
5281	#ifdef IN_RING3
5282
5283	/**
5284	* Reset CPU on hot plugging.
5285	*
5286	* @param pVM Pointer to the VM.
5287	* @param pVCpu The virtual CPU.
5288	*/
5289	void pgmR3PoolResetUnpluggedCpu(PVM pVM, PVMCPU pVCpu)
5290	{
5291	pgmR3ExitShadowModeBeforePoolFlush(pVCpu);
5292
5293	pgmR3ReEnterShadowModeAfterPoolFlush(pVM, pVCpu);
5294	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
5295	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
5296	}
5297
5298
5299	/**
5300	* Flushes the entire cache.
5301	*
5302	* It will assert a global CR3 flush (FF) and assumes the caller is aware of
5303	* this and execute this CR3 flush.
5304	*
5305	* @param pPool The pool.
5306	*/
5307	void pgmR3PoolReset(PVM pVM)
5308	{
5309	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
5310
5311	PGM_LOCK_ASSERT_OWNER(pVM);
5312	STAM_PROFILE_START(&pPool->StatR3Reset, a);
5313	LogFlow(("pgmR3PoolReset:\n"));
5314
5315	/*
5316	* If there are no pages in the pool, there is nothing to do.
5317	*/
5318	if (pPool->cCurPages <= PGMPOOL_IDX_FIRST)
5319	{
5320	STAM_PROFILE_STOP(&pPool->StatR3Reset, a);
5321	return;
5322	}
5323
5324	/*
5325	* Exit the shadow mode since we're going to clear everything,
5326	* including the root page.
5327	*/
5328	for (VMCPUID i = 0; i < pVM->cCpus; i++)
5329	pgmR3ExitShadowModeBeforePoolFlush(&pVM->aCpus[i]);
5330
5331	/*
5332	* Nuke the free list and reinsert all pages into it.
5333	*/
5334	for (unsigned i = pPool->cCurPages - 1; i >= PGMPOOL_IDX_FIRST; i--)
5335	{
5336	PPGMPOOLPAGE pPage = &pPool->aPages[i];
5337
5338	Assert(pPage->Core.Key == MMPage2Phys(pVM, pPage->pvPageR3));
5339	if (pPage->fMonitored)
5340	pgmPoolMonitorFlush(pPool, pPage);
5341	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
5342	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
5343	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
5344	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
5345	pPage->cModifications = 0;
5346	pPage->GCPhys = NIL_RTGCPHYS;
5347	pPage->enmKind = PGMPOOLKIND_FREE;
5348	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
5349	Assert(pPage->idx == i);
5350	pPage->iNext = i + 1;
5351	pPage->fA20Enabled = true;
5352	pPage->fZeroed = false; /* This could probably be optimized, but better safe than sorry. */
5353	pPage->fSeenNonGlobal = false;
5354	pPage->fMonitored = false;
5355	pPage->fDirty = false;
5356	pPage->fCached = false;
5357	pPage->fReusedFlushPending = false;
5358	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
5359	pPage->iAgeNext = NIL_PGMPOOL_IDX;
5360	pPage->iAgePrev = NIL_PGMPOOL_IDX;
5361	pPage->GCPtrLastAccessHandlerRip = NIL_RTGCPTR;
5362	pPage->GCPtrLastAccessHandlerFault = NIL_RTGCPTR;
5363	pPage->cLastAccessHandler = 0;
5364	pPage->cLocked = 0;
5365	#ifdef VBOX_STRICT
5366	pPage->GCPtrDirtyFault = NIL_RTGCPTR;
5367	#endif
5368	}
5369	pPool->aPages[pPool->cCurPages - 1].iNext = NIL_PGMPOOL_IDX;
5370	pPool->iFreeHead = PGMPOOL_IDX_FIRST;
5371	pPool->cUsedPages = 0;
5372
5373	/*
5374	* Zap and reinitialize the user records.
5375	*/
5376	pPool->cPresent = 0;
5377	pPool->iUserFreeHead = 0;
5378	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
5379	const unsigned cMaxUsers = pPool->cMaxUsers;
5380	for (unsigned i = 0; i < cMaxUsers; i++)
5381	{
5382	paUsers[i].iNext = i + 1;
5383	paUsers[i].iUser = NIL_PGMPOOL_IDX;
5384	paUsers[i].iUserTable = 0xfffffffe;
5385	}
5386	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
5387
5388	/*
5389	* Clear all the GCPhys links and rebuild the phys ext free list.
5390	*/
5391	for (PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRangesX);
5392	pRam;
5393	pRam = pRam->CTX_SUFF(pNext))
5394	{
5395	unsigned iPage = pRam->cb >> PAGE_SHIFT;
5396	while (iPage-- > 0)
5397	PGM_PAGE_SET_TRACKING(pVM, &pRam->aPages[iPage], 0);
5398	}
5399
5400	pPool->iPhysExtFreeHead = 0;
5401	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
5402	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
5403	for (unsigned i = 0; i < cMaxPhysExts; i++)
5404	{
5405	paPhysExts[i].iNext = i + 1;
5406	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
5407	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
5408	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
5409	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
5410	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
5411	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
5412	}
5413	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
5414
5415	/*
5416	* Just zap the modified list.
5417	*/
5418	pPool->cModifiedPages = 0;
5419	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
5420
5421	/*
5422	* Clear the GCPhys hash and the age list.
5423	*/
5424	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
5425	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
5426	pPool->iAgeHead = NIL_PGMPOOL_IDX;
5427	pPool->iAgeTail = NIL_PGMPOOL_IDX;
5428
5429	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
5430	/* Clear all dirty pages. */
5431	pPool->idxFreeDirtyPage = 0;
5432	pPool->cDirtyPages = 0;
5433	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
5434	pPool->aDirtyPages[i].uIdx = NIL_PGMPOOL_IDX;
5435	#endif
5436
5437	/*
5438	* Reinsert active pages into the hash and ensure monitoring chains are correct.
5439	*/
5440	for (VMCPUID i = 0; i < pVM->cCpus; i++)
5441	{
5442	/*
5443	* Re-enter the shadowing mode and assert Sync CR3 FF.
5444	*/
5445	PVMCPU pVCpu = &pVM->aCpus[i];
5446	pgmR3ReEnterShadowModeAfterPoolFlush(pVM, pVCpu);
5447	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
5448	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
5449	}
5450
5451	STAM_PROFILE_STOP(&pPool->StatR3Reset, a);
5452	}
5453
5454	#endif /* IN_RING3 */
5455
5456	#if defined(LOG_ENABLED) \|\| defined(VBOX_STRICT)
5457	/**
5458	* Stringifies a PGMPOOLKIND value.
5459	*/
5460	static const char *pgmPoolPoolKindToStr(uint8_t enmKind)
5461	{
5462	switch ((PGMPOOLKIND)enmKind)
5463	{
5464	case PGMPOOLKIND_INVALID:
5465	return "PGMPOOLKIND_INVALID";
5466	case PGMPOOLKIND_FREE:
5467	return "PGMPOOLKIND_FREE";
5468	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
5469	return "PGMPOOLKIND_32BIT_PT_FOR_PHYS";
5470	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
5471	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT";
5472	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
5473	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB";
5474	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
5475	return "PGMPOOLKIND_PAE_PT_FOR_PHYS";
5476	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
5477	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_PT";
5478	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
5479	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB";
5480	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
5481	return "PGMPOOLKIND_PAE_PT_FOR_PAE_PT";
5482	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
5483	return "PGMPOOLKIND_PAE_PT_FOR_PAE_2MB";
5484	case PGMPOOLKIND_32BIT_PD:
5485	return "PGMPOOLKIND_32BIT_PD";
5486	case PGMPOOLKIND_32BIT_PD_PHYS:
5487	return "PGMPOOLKIND_32BIT_PD_PHYS";
5488	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
5489	return "PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD";
5490	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
5491	return "PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD";
5492	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
5493	return "PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD";
5494	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
5495	return "PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD";
5496	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
5497	return "PGMPOOLKIND_PAE_PD_FOR_PAE_PD";
5498	case PGMPOOLKIND_PAE_PD_PHYS:
5499	return "PGMPOOLKIND_PAE_PD_PHYS";
5500	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
5501	return "PGMPOOLKIND_PAE_PDPT_FOR_32BIT";
5502	case PGMPOOLKIND_PAE_PDPT:
5503	return "PGMPOOLKIND_PAE_PDPT";
5504	case PGMPOOLKIND_PAE_PDPT_PHYS:
5505	return "PGMPOOLKIND_PAE_PDPT_PHYS";
5506	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
5507	return "PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT";
5508	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
5509	return "PGMPOOLKIND_64BIT_PDPT_FOR_PHYS";
5510	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
5511	return "PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD";
5512	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
5513	return "PGMPOOLKIND_64BIT_PD_FOR_PHYS";
5514	case PGMPOOLKIND_64BIT_PML4:
5515	return "PGMPOOLKIND_64BIT_PML4";
5516	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
5517	return "PGMPOOLKIND_EPT_PDPT_FOR_PHYS";
5518	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
5519	return "PGMPOOLKIND_EPT_PD_FOR_PHYS";
5520	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
5521	return "PGMPOOLKIND_EPT_PT_FOR_PHYS";
5522	case PGMPOOLKIND_ROOT_NESTED:
5523	return "PGMPOOLKIND_ROOT_NESTED";
5524	}
5525	return "Unknown kind!";
5526	}
5527	#endif /* LOG_ENABLED \|\| VBOX_STRICT */
5528

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllPool.cpp@ 48749

Download in other formats: