PGMAllPool.cpp@ 22348

Last change on this file since 22348 was 22348, checked in by vboxsync, 15 years ago
Tweak the defaults a bit to prevent kicking out pages too quickly
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 166.5 KB

Line
1	/* $Id: PGMAllPool.cpp 22348 2009-08-19 14:04:03Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Sun Microsystems, Inc.
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	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
18	* Clara, CA 95054 USA or visit http://www.sun.com if you need
19	* additional information or have any questions.
20	*/
21
22
23	/*******************************************************************************
24	* Header Files *
25	*******************************************************************************/
26	#define LOG_GROUP LOG_GROUP_PGM_POOL
27	#include <VBox/pgm.h>
28	#include <VBox/mm.h>
29	#include <VBox/em.h>
30	#include <VBox/cpum.h>
31	#ifdef IN_RC
32	# include <VBox/patm.h>
33	#endif
34	#include "PGMInternal.h"
35	#include <VBox/vm.h>
36	#include <VBox/disopcode.h>
37	#include <VBox/hwacc_vmx.h>
38
39	#include <VBox/log.h>
40	#include <VBox/err.h>
41	#include <iprt/asm.h>
42	#include <iprt/string.h>
43
44
45	/*******************************************************************************
46	* Internal Functions *
47	*******************************************************************************/
48	RT_C_DECLS_BEGIN
49	static void pgmPoolFlushAllInt(PPGMPOOL pPool);
50	#ifdef PGMPOOL_WITH_USER_TRACKING
51	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind);
52	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind);
53	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
54	#endif
55	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
56	static void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint);
57	#endif
58	#ifdef PGMPOOL_WITH_CACHE
59	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable);
60	#endif
61	#ifdef PGMPOOL_WITH_MONITORING
62	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
63	#endif
64	#ifndef IN_RING3
65	DECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser);
66	#endif
67	#ifdef LOG_ENABLED
68	static const char *pgmPoolPoolKindToStr(uint8_t enmKind);
69	#endif
70
71	void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs);
72	void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, uint16_t iPhysExt);
73	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage);
74	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt);
75	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt);
76	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt);
77
78	RT_C_DECLS_END
79
80
81	/**
82	* Checks if the specified page pool kind is for a 4MB or 2MB guest page.
83	*
84	* @returns true if it's the shadow of a 4MB or 2MB guest page, otherwise false.
85	* @param enmKind The page kind.
86	*/
87	DECLINLINE(bool) pgmPoolIsBigPage(PGMPOOLKIND enmKind)
88	{
89	switch (enmKind)
90	{
91	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
92	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
93	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
94	return true;
95	default:
96	return false;
97	}
98	}
99
100	/** @def PGMPOOL_PAGE_2_LOCKED_PTR
101	* Maps a pool page pool into the current context and lock it (RC only).
102	*
103	* @returns VBox status code.
104	* @param pVM The VM handle.
105	* @param pPage The pool page.
106	*
107	* @remark In RC this uses PGMGCDynMapHCPage(), so it will consume of the
108	* small page window employeed by that function. Be careful.
109	* @remark There is no need to assert on the result.
110	*/
111	#if defined(IN_RC)
112	DECLINLINE(void *) PGMPOOL_PAGE_2_LOCKED_PTR(PVM pVM, PPGMPOOLPAGE pPage)
113	{
114	void *pv = pgmPoolMapPageInlined(&pVM->pgm.s, pPage);
115
116	/* Make sure the dynamic mapping will not be reused. */
117	if (pv)
118	PGMDynLockHCPage(pVM, (uint8_t *)pv);
119
120	return pv;
121	}
122	#else
123	# define PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage) PGMPOOL_PAGE_2_PTR(pVM, pPage)
124	#endif
125
126	/** @def PGMPOOL_UNLOCK_PTR
127	* Unlock a previously locked dynamic caching (RC only).
128	*
129	* @returns VBox status code.
130	* @param pVM The VM handle.
131	* @param pPage The pool page.
132	*
133	* @remark In RC this uses PGMGCDynMapHCPage(), so it will consume of the
134	* small page window employeed by that function. Be careful.
135	* @remark There is no need to assert on the result.
136	*/
137	#if defined(IN_RC)
138	DECLINLINE(void) PGMPOOL_UNLOCK_PTR(PVM pVM, void *pvPage)
139	{
140	if (pvPage)
141	PGMDynUnlockHCPage(pVM, (uint8_t *)pvPage);
142	}
143	#else
144	# define PGMPOOL_UNLOCK_PTR(pVM, pPage) do {} while (0)
145	#endif
146
147
148	#ifdef PGMPOOL_WITH_MONITORING
149	/**
150	* Determin the size of a write instruction.
151	* @returns number of bytes written.
152	* @param pDis The disassembler state.
153	*/
154	static unsigned pgmPoolDisasWriteSize(PDISCPUSTATE pDis)
155	{
156	/*
157	* This is very crude and possibly wrong for some opcodes,
158	* but since it's not really supposed to be called we can
159	* probably live with that.
160	*/
161	return DISGetParamSize(pDis, &pDis->param1);
162	}
163
164
165	/**
166	* Flushes a chain of pages sharing the same access monitor.
167	*
168	* @returns VBox status code suitable for scheduling.
169	* @param pPool The pool.
170	* @param pPage A page in the chain.
171	*/
172	int pgmPoolMonitorChainFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
173	{
174	LogFlow(("pgmPoolMonitorChainFlush: Flush page %RGp type=%d\n", pPage->GCPhys, pPage->enmKind));
175
176	/*
177	* Find the list head.
178	*/
179	uint16_t idx = pPage->idx;
180	if (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
181	{
182	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
183	{
184	idx = pPage->iMonitoredPrev;
185	Assert(idx != pPage->idx);
186	pPage = &pPool->aPages[idx];
187	}
188	}
189
190	/*
191	* Iterate the list flushing each shadow page.
192	*/
193	int rc = VINF_SUCCESS;
194	for (;;)
195	{
196	idx = pPage->iMonitoredNext;
197	Assert(idx != pPage->idx);
198	if (pPage->idx >= PGMPOOL_IDX_FIRST)
199	{
200	int rc2 = pgmPoolFlushPage(pPool, pPage);
201	AssertRC(rc2);
202	}
203	/* next */
204	if (idx == NIL_PGMPOOL_IDX)
205	break;
206	pPage = &pPool->aPages[idx];
207	}
208	return rc;
209	}
210
211
212	/**
213	* Wrapper for getting the current context pointer to the entry being modified.
214	*
215	* @returns VBox status code suitable for scheduling.
216	* @param pVM VM Handle.
217	* @param pvDst Destination address
218	* @param pvSrc Source guest virtual address.
219	* @param GCPhysSrc The source guest physical address.
220	* @param cb Size of data to read
221	*/
222	DECLINLINE(int) pgmPoolPhysSimpleReadGCPhys(PVM pVM, void *pvDst, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvSrc, RTGCPHYS GCPhysSrc, size_t cb)
223	{
224	#if defined(IN_RING3)
225	memcpy(pvDst, (RTHCPTR)((uintptr_t)pvSrc & ~(RTHCUINTPTR)(cb - 1)), cb);
226	return VINF_SUCCESS;
227	#else
228	/* @todo in RC we could attempt to use the virtual address, although this can cause many faults (PAE Windows XP guest). */
229	return PGMPhysSimpleReadGCPhys(pVM, pvDst, GCPhysSrc & ~(RTGCPHYS)(cb - 1), cb);
230	#endif
231	}
232
233	/**
234	* Process shadow entries before they are changed by the guest.
235	*
236	* For PT entries we will clear them. For PD entries, we'll simply check
237	* for mapping conflicts and set the SyncCR3 FF if found.
238	*
239	* @param pVCpu VMCPU handle
240	* @param pPool The pool.
241	* @param pPage The head page.
242	* @param GCPhysFault The guest physical fault address.
243	* @param uAddress In R0 and GC this is the guest context fault address (flat).
244	* In R3 this is the host context 'fault' address.
245	* @param pDis The disassembler state for figuring out the write size.
246	* This need not be specified if the caller knows we won't do cross entry accesses.
247	*/
248	void pgmPoolMonitorChainChanging(PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhysFault, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvAddress, PDISCPUSTATE pDis)
249	{
250	AssertMsg(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX, ("%#x (idx=%#x)\n", pPage->iMonitoredPrev, pPage->idx));
251	const unsigned off = GCPhysFault & PAGE_OFFSET_MASK;
252	const unsigned cbWrite = pDis ? pgmPoolDisasWriteSize(pDis) : 0;
253	PVM pVM = pPool->CTX_SUFF(pVM);
254
255	LogFlow(("pgmPoolMonitorChainChanging: %RGv phys=%RGp kind=%s cbWrite=%d\n", (RTGCPTR)pvAddress, GCPhysFault, pgmPoolPoolKindToStr(pPage->enmKind), cbWrite));
256	for (;;)
257	{
258	union
259	{
260	void *pv;
261	PX86PT pPT;
262	PX86PTPAE pPTPae;
263	PX86PD pPD;
264	PX86PDPAE pPDPae;
265	PX86PDPT pPDPT;
266	PX86PML4 pPML4;
267	} uShw;
268
269	uShw.pv = NULL;
270	switch (pPage->enmKind)
271	{
272	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
273	{
274	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
275	const unsigned iShw = off / sizeof(X86PTE);
276	LogFlow(("PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT iShw=%x\n", iShw));
277	if (uShw.pPT->a[iShw].n.u1Present)
278	{
279	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
280	X86PTE GstPte;
281
282	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
283	AssertRC(rc);
284	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));
285	pgmPoolTracDerefGCPhysHint(pPool, pPage,
286	uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK,
287	GstPte.u & X86_PTE_PG_MASK);
288	# endif
289	ASMAtomicWriteSize(&uShw.pPT->a[iShw], 0);
290	}
291	break;
292	}
293
294	/* page/2 sized */
295	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
296	{
297	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
298	if (!((off ^ pPage->GCPhys) & (PAGE_SIZE / 2)))
299	{
300	const unsigned iShw = (off / sizeof(X86PTE)) & (X86_PG_PAE_ENTRIES - 1);
301	LogFlow(("PGMPOOLKIND_PAE_PT_FOR_32BIT_PT iShw=%x\n", iShw));
302	if (uShw.pPTPae->a[iShw].n.u1Present)
303	{
304	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
305	X86PTE GstPte;
306	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
307	AssertRC(rc);
308
309	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));
310	pgmPoolTracDerefGCPhysHint(pPool, pPage,
311	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
312	GstPte.u & X86_PTE_PG_MASK);
313	# endif
314	ASMAtomicWriteSize(&uShw.pPTPae->a[iShw], 0);
315	}
316	}
317	break;
318	}
319
320	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
321	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
322	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
323	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
324	{
325	unsigned iGst = off / sizeof(X86PDE);
326	unsigned iShwPdpt = iGst / 256;
327	unsigned iShw = (iGst % 256) * 2;
328	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
329
330	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));
331	if (iShwPdpt == pPage->enmKind - (unsigned)PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD)
332	{
333	for (unsigned i = 0; i < 2; i++)
334	{
335	# ifndef IN_RING0
336	if ((uShw.pPDPae->a[iShw + i].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
337	{
338	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
339	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
340	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw=%#x!\n", iShwPdpt, iShw+i));
341	break;
342	}
343	else
344	# endif /* !IN_RING0 */
345	if (uShw.pPDPae->a[iShw+i].n.u1Present)
346	{
347	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw+i, uShw.pPDPae->a[iShw+i].u));
348	pgmPoolFree(pVM,
349	uShw.pPDPae->a[iShw+i].u & X86_PDE_PAE_PG_MASK,
350	pPage->idx,
351	iShw + i);
352	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw+i], 0);
353	}
354
355	/* paranoia / a bit assumptive. */
356	if ( pDis
357	&& (off & 3)
358	&& (off & 3) + cbWrite > 4)
359	{
360	const unsigned iShw2 = iShw + 2 + i;
361	if (iShw2 < RT_ELEMENTS(uShw.pPDPae->a))
362	{
363	# ifndef IN_RING0
364	if ((uShw.pPDPae->a[iShw2].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
365	{
366	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
367	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
368	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw2=%#x!\n", iShwPdpt, iShw2));
369	break;
370	}
371	else
372	# endif /* !IN_RING0 */
373	if (uShw.pPDPae->a[iShw2].n.u1Present)
374	{
375	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
376	pgmPoolFree(pVM,
377	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
378	pPage->idx,
379	iShw2);
380	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw2].u, 0);
381	}
382	}
383	}
384	}
385	}
386	break;
387	}
388
389	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
390	{
391	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
392	const unsigned iShw = off / sizeof(X86PTEPAE);
393	if (uShw.pPTPae->a[iShw].n.u1Present)
394	{
395	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
396	X86PTEPAE GstPte;
397	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
398	AssertRC(rc);
399
400	Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PAE_PG_MASK));
401	pgmPoolTracDerefGCPhysHint(pPool, pPage,
402	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
403	GstPte.u & X86_PTE_PAE_PG_MASK);
404	# endif
405	ASMAtomicWriteSize(&uShw.pPTPae->a[iShw].u, 0);
406	}
407
408	/* paranoia / a bit assumptive. */
409	if ( pDis
410	&& (off & 7)
411	&& (off & 7) + cbWrite > sizeof(X86PTEPAE))
412	{
413	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTEPAE);
414	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPTPae->a));
415
416	if (uShw.pPTPae->a[iShw2].n.u1Present)
417	{
418	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
419	X86PTEPAE GstPte;
420	# ifdef IN_RING3
421	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, (RTHCPTR)((RTHCUINTPTR)pvAddress + sizeof(GstPte)), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
422	# else
423	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress + sizeof(GstPte), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
424	# endif
425	AssertRC(rc);
426	Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PAE_PG_MASK));
427	pgmPoolTracDerefGCPhysHint(pPool, pPage,
428	uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK,
429	GstPte.u & X86_PTE_PAE_PG_MASK);
430	# endif
431	ASMAtomicWriteSize(&uShw.pPTPae->a[iShw2].u ,0);
432	}
433	}
434	break;
435	}
436
437	case PGMPOOLKIND_32BIT_PD:
438	{
439	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
440	const unsigned iShw = off / sizeof(X86PTE); // ASSUMING 32-bit guest paging!
441
442	LogFlow(("pgmPoolMonitorChainChanging: PGMPOOLKIND_32BIT_PD %x\n", iShw));
443	# ifndef IN_RING0
444	if (uShw.pPD->a[iShw].u & PGM_PDFLAGS_MAPPING)
445	{
446	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
447	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
448	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
449	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
450	break;
451	}
452	# endif /* !IN_RING0 */
453	# ifndef IN_RING0
454	else
455	# endif /* !IN_RING0 */
456	{
457	if (uShw.pPD->a[iShw].n.u1Present)
458	{
459	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
460	pgmPoolFree(pVM,
461	uShw.pPD->a[iShw].u & X86_PDE_PAE_PG_MASK,
462	pPage->idx,
463	iShw);
464	ASMAtomicWriteSize(&uShw.pPD->a[iShw].u, 0);
465	}
466	}
467	/* paranoia / a bit assumptive. */
468	if ( pDis
469	&& (off & 3)
470	&& (off & 3) + cbWrite > sizeof(X86PTE))
471	{
472	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTE);
473	if ( iShw2 != iShw
474	&& iShw2 < RT_ELEMENTS(uShw.pPD->a))
475	{
476	# ifndef IN_RING0
477	if (uShw.pPD->a[iShw2].u & PGM_PDFLAGS_MAPPING)
478	{
479	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
480	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
481	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
482	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
483	break;
484	}
485	# endif /* !IN_RING0 */
486	# ifndef IN_RING0
487	else
488	# endif /* !IN_RING0 */
489	{
490	if (uShw.pPD->a[iShw2].n.u1Present)
491	{
492	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPD->a[iShw2].u));
493	pgmPoolFree(pVM,
494	uShw.pPD->a[iShw2].u & X86_PDE_PAE_PG_MASK,
495	pPage->idx,
496	iShw2);
497	ASMAtomicWriteSize(&uShw.pPD->a[iShw2].u, 0);
498	}
499	}
500	}
501	}
502	#if 0 /* useful when running PGMAssertCR3(), a bit too troublesome for general use (TLBs). */
503	if ( uShw.pPD->a[iShw].n.u1Present
504	&& !VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
505	{
506	LogFlow(("pgmPoolMonitorChainChanging: iShw=%#x: %RX32 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
507	# ifdef IN_RC /* TLB load - we're pushing things a bit... */
508	ASMProbeReadByte(pvAddress);
509	# endif
510	pgmPoolFree(pVM, uShw.pPD->a[iShw].u & X86_PDE_PG_MASK, pPage->idx, iShw);
511	ASMAtomicWriteSize(&uShw.pPD->a[iShw].u, 0);
512	}
513	#endif
514	break;
515	}
516
517	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
518	{
519	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
520	const unsigned iShw = off / sizeof(X86PDEPAE);
521	#ifndef IN_RING0
522	if (uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING)
523	{
524	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
525	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
526	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
527	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
528	break;
529	}
530	#endif /* !IN_RING0 */
531	/*
532	* Causes trouble when the guest uses a PDE to refer to the whole page table level
533	* structure. (Invalidate here; faults later on when it tries to change the page
534	* table entries -> recheck; probably only applies to the RC case.)
535	*/
536	# ifndef IN_RING0
537	else
538	# endif /* !IN_RING0 */
539	{
540	if (uShw.pPDPae->a[iShw].n.u1Present)
541	{
542	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
543	pgmPoolFree(pVM,
544	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
545	pPage->idx,
546	iShw);
547	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw].u, 0);
548	}
549	}
550	/* paranoia / a bit assumptive. */
551	if ( pDis
552	&& (off & 7)
553	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
554	{
555	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
556	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
557
558	#ifndef IN_RING0
559	if ( iShw2 != iShw
560	&& uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING)
561	{
562	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
563	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
564	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
565	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
566	break;
567	}
568	#endif /* !IN_RING0 */
569	# ifndef IN_RING0
570	else
571	# endif /* !IN_RING0 */
572	if (uShw.pPDPae->a[iShw2].n.u1Present)
573	{
574	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
575	pgmPoolFree(pVM,
576	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
577	pPage->idx,
578	iShw2);
579	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw2].u, 0);
580	}
581	}
582	break;
583	}
584
585	case PGMPOOLKIND_PAE_PDPT:
586	{
587	/*
588	* Hopefully this doesn't happen very often:
589	* - touching unused parts of the page
590	* - messing with the bits of pd pointers without changing the physical address
591	*/
592	/* PDPT roots are not page aligned; 32 byte only! */
593	const unsigned offPdpt = GCPhysFault - pPage->GCPhys;
594
595	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
596	const unsigned iShw = offPdpt / sizeof(X86PDPE);
597	if (iShw < X86_PG_PAE_PDPE_ENTRIES) /* don't use RT_ELEMENTS(uShw.pPDPT->a), because that's for long mode only */
598	{
599	# ifndef IN_RING0
600	if (uShw.pPDPT->a[iShw].u & PGM_PLXFLAGS_MAPPING)
601	{
602	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
603	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
604	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
605	LogFlow(("pgmPoolMonitorChainChanging: Detected pdpt conflict at iShw=%#x!\n", iShw));
606	break;
607	}
608	# endif /* !IN_RING0 */
609	# ifndef IN_RING0
610	else
611	# endif /* !IN_RING0 */
612	if (uShw.pPDPT->a[iShw].n.u1Present)
613	{
614	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
615	pgmPoolFree(pVM,
616	uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK,
617	pPage->idx,
618	iShw);
619	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw].u, 0);
620	}
621
622	/* paranoia / a bit assumptive. */
623	if ( pDis
624	&& (offPdpt & 7)
625	&& (offPdpt & 7) + cbWrite > sizeof(X86PDPE))
626	{
627	const unsigned iShw2 = (offPdpt + cbWrite - 1) / sizeof(X86PDPE);
628	if ( iShw2 != iShw
629	&& iShw2 < X86_PG_PAE_PDPE_ENTRIES)
630	{
631	# ifndef IN_RING0
632	if (uShw.pPDPT->a[iShw2].u & PGM_PLXFLAGS_MAPPING)
633	{
634	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
635	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
636	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
637	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
638	break;
639	}
640	# endif /* !IN_RING0 */
641	# ifndef IN_RING0
642	else
643	# endif /* !IN_RING0 */
644	if (uShw.pPDPT->a[iShw2].n.u1Present)
645	{
646	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
647	pgmPoolFree(pVM,
648	uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK,
649	pPage->idx,
650	iShw2);
651	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw2].u, 0);
652	}
653	}
654	}
655	}
656	break;
657	}
658
659	#ifndef IN_RC
660	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
661	{
662	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
663	const unsigned iShw = off / sizeof(X86PDEPAE);
664	Assert(!(uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING));
665	if (uShw.pPDPae->a[iShw].n.u1Present)
666	{
667	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
668	pgmPoolFree(pVM,
669	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
670	pPage->idx,
671	iShw);
672	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw].u, 0);
673	}
674	/* paranoia / a bit assumptive. */
675	if ( pDis
676	&& (off & 7)
677	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
678	{
679	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
680	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
681
682	Assert(!(uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING));
683	if (uShw.pPDPae->a[iShw2].n.u1Present)
684	{
685	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
686	pgmPoolFree(pVM,
687	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
688	pPage->idx,
689	iShw2);
690	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw2].u, 0);
691	}
692	}
693	break;
694	}
695
696	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
697	{
698	/*
699	* Hopefully this doesn't happen very often:
700	* - messing with the bits of pd pointers without changing the physical address
701	*/
702	if (!VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
703	{
704	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
705	const unsigned iShw = off / sizeof(X86PDPE);
706	if (uShw.pPDPT->a[iShw].n.u1Present)
707	{
708	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
709	pgmPoolFree(pVM, uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK, pPage->idx, iShw);
710	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw].u, 0);
711	}
712	/* paranoia / a bit assumptive. */
713	if ( pDis
714	&& (off & 7)
715	&& (off & 7) + cbWrite > sizeof(X86PDPE))
716	{
717	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDPE);
718	if (uShw.pPDPT->a[iShw2].n.u1Present)
719	{
720	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
721	pgmPoolFree(pVM, uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK, pPage->idx, iShw2);
722	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw2].u, 0);
723	}
724	}
725	}
726	break;
727	}
728
729	case PGMPOOLKIND_64BIT_PML4:
730	{
731	/*
732	* Hopefully this doesn't happen very often:
733	* - messing with the bits of pd pointers without changing the physical address
734	*/
735	if (!VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
736	{
737	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
738	const unsigned iShw = off / sizeof(X86PDPE);
739	if (uShw.pPML4->a[iShw].n.u1Present)
740	{
741	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPML4->a[iShw].u));
742	pgmPoolFree(pVM, uShw.pPML4->a[iShw].u & X86_PML4E_PG_MASK, pPage->idx, iShw);
743	ASMAtomicWriteSize(&uShw.pPML4->a[iShw].u, 0);
744	}
745	/* paranoia / a bit assumptive. */
746	if ( pDis
747	&& (off & 7)
748	&& (off & 7) + cbWrite > sizeof(X86PDPE))
749	{
750	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PML4E);
751	if (uShw.pPML4->a[iShw2].n.u1Present)
752	{
753	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPML4->a[iShw2].u));
754	pgmPoolFree(pVM, uShw.pPML4->a[iShw2].u & X86_PML4E_PG_MASK, pPage->idx, iShw2);
755	ASMAtomicWriteSize(&uShw.pPML4->a[iShw2].u, 0);
756	}
757	}
758	}
759	break;
760	}
761	#endif /* IN_RING0 */
762
763	default:
764	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
765	}
766	PGMPOOL_UNLOCK_PTR(pVM, uShw.pv);
767
768	/* next */
769	if (pPage->iMonitoredNext == NIL_PGMPOOL_IDX)
770	return;
771	pPage = &pPool->aPages[pPage->iMonitoredNext];
772	}
773	}
774
775	# ifndef IN_RING3
776	/**
777	* Checks if a access could be a fork operation in progress.
778	*
779	* Meaning, that the guest is setuping up the parent process for Copy-On-Write.
780	*
781	* @returns true if it's likly that we're forking, otherwise false.
782	* @param pPool The pool.
783	* @param pDis The disassembled instruction.
784	* @param offFault The access offset.
785	*/
786	DECLINLINE(bool) pgmPoolMonitorIsForking(PPGMPOOL pPool, PDISCPUSTATE pDis, unsigned offFault)
787	{
788	/*
789	* i386 linux is using btr to clear X86_PTE_RW.
790	* The functions involved are (2.6.16 source inspection):
791	* clear_bit
792	* ptep_set_wrprotect
793	* copy_one_pte
794	* copy_pte_range
795	* copy_pmd_range
796	* copy_pud_range
797	* copy_page_range
798	* dup_mmap
799	* dup_mm
800	* copy_mm
801	* copy_process
802	* do_fork
803	*/
804	if ( pDis->pCurInstr->opcode == OP_BTR
805	&& !(offFault & 4)
806	/** @todo Validate that the bit index is X86_PTE_RW. */
807	)
808	{
809	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,Fork));
810	return true;
811	}
812	return false;
813	}
814
815
816	/**
817	* Determine whether the page is likely to have been reused.
818	*
819	* @returns true if we consider the page as being reused for a different purpose.
820	* @returns false if we consider it to still be a paging page.
821	* @param pVM VM Handle.
822	* @param pVCpu VMCPU Handle.
823	* @param pRegFrame Trap register frame.
824	* @param pDis The disassembly info for the faulting instruction.
825	* @param pvFault The fault address.
826	*
827	* @remark The REP prefix check is left to the caller because of STOSD/W.
828	*/
829	DECLINLINE(bool) pgmPoolMonitorIsReused(PVM pVM, PVMCPU pVCpu, PCPUMCTXCORE pRegFrame, PDISCPUSTATE pDis, RTGCPTR pvFault)
830	{
831	#ifndef IN_RC
832	/** @todo could make this general, faulting close to rsp should be safe reuse heuristic. */
833	if ( HWACCMHasPendingIrq(pVM)
834	&& (pRegFrame->rsp - pvFault) < 32)
835	{
836	/* Fault caused by stack writes while trying to inject an interrupt event. */
837	Log(("pgmPoolMonitorIsReused: reused %RGv for interrupt stack (rsp=%RGv).\n", pvFault, pRegFrame->rsp));
838	return true;
839	}
840	#else
841	NOREF(pVM); NOREF(pvFault);
842	#endif
843
844	LogFlow(("Reused instr %RGv %d at %RGv param1.flags=%x param1.reg=%d\n", pRegFrame->rip, pDis->pCurInstr->opcode, pvFault, pDis->param1.flags, pDis->param1.base.reg_gen));
845
846	/* Non-supervisor mode write means it's used for something else. */
847	if (CPUMGetGuestCPL(pVCpu, pRegFrame) != 0)
848	return true;
849
850	switch (pDis->pCurInstr->opcode)
851	{
852	/* call implies the actual push of the return address faulted */
853	case OP_CALL:
854	Log4(("pgmPoolMonitorIsReused: CALL\n"));
855	return true;
856	case OP_PUSH:
857	Log4(("pgmPoolMonitorIsReused: PUSH\n"));
858	return true;
859	case OP_PUSHF:
860	Log4(("pgmPoolMonitorIsReused: PUSHF\n"));
861	return true;
862	case OP_PUSHA:
863	Log4(("pgmPoolMonitorIsReused: PUSHA\n"));
864	return true;
865	case OP_FXSAVE:
866	Log4(("pgmPoolMonitorIsReused: FXSAVE\n"));
867	return true;
868	case OP_MOVNTI: /* solaris - block_zero_no_xmm */
869	Log4(("pgmPoolMonitorIsReused: MOVNTI\n"));
870	return true;
871	case OP_MOVNTDQ: /* solaris - hwblkclr & hwblkpagecopy */
872	Log4(("pgmPoolMonitorIsReused: MOVNTDQ\n"));
873	return true;
874	case OP_MOVSWD:
875	case OP_STOSWD:
876	if ( pDis->prefix == (PREFIX_REP\|PREFIX_REX)
877	&& pRegFrame->rcx >= 0x40
878	)
879	{
880	Assert(pDis->mode == CPUMODE_64BIT);
881
882	Log(("pgmPoolMonitorIsReused: OP_STOSQ\n"));
883	return true;
884	}
885	return false;
886	}
887	if ( (pDis->param1.flags & USE_REG_GEN32)
888	&& (pDis->param1.base.reg_gen == USE_REG_ESP))
889	{
890	Log4(("pgmPoolMonitorIsReused: ESP\n"));
891	return true;
892	}
893
894	return false;
895	}
896
897
898	/**
899	* Flushes the page being accessed.
900	*
901	* @returns VBox status code suitable for scheduling.
902	* @param pVM The VM handle.
903	* @param pVCpu The VMCPU handle.
904	* @param pPool The pool.
905	* @param pPage The pool page (head).
906	* @param pDis The disassembly of the write instruction.
907	* @param pRegFrame The trap register frame.
908	* @param GCPhysFault The fault address as guest physical address.
909	* @param pvFault The fault address.
910	*/
911	static int pgmPoolAccessHandlerFlush(PVM pVM, PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pDis,
912	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
913	{
914	/*
915	* First, do the flushing.
916	*/
917	int rc = pgmPoolMonitorChainFlush(pPool, pPage);
918
919	/*
920	* Emulate the instruction (xp/w2k problem, requires pc/cr2/sp detection).
921	*/
922	uint32_t cbWritten;
923	int rc2 = EMInterpretInstructionCPU(pVM, pVCpu, pDis, pRegFrame, pvFault, &cbWritten);
924	if (RT_SUCCESS(rc2))
925	pRegFrame->rip += pDis->opsize;
926	else if (rc2 == VERR_EM_INTERPRETER)
927	{
928	#ifdef IN_RC
929	if (PATMIsPatchGCAddr(pVM, (RTRCPTR)pRegFrame->eip))
930	{
931	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for patch code %04x:%RGv, ignoring.\n",
932	pRegFrame->cs, (RTGCPTR)pRegFrame->eip));
933	rc = VINF_SUCCESS;
934	STAM_COUNTER_INC(&pPool->StatMonitorRZIntrFailPatch2);
935	}
936	else
937	#endif
938	{
939	rc = VINF_EM_RAW_EMULATE_INSTR;
940	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
941	}
942	}
943	else
944	rc = rc2;
945
946	/* See use in pgmPoolAccessHandlerSimple(). */
947	PGM_INVL_VCPU_TLBS(pVCpu);
948
949	LogFlow(("pgmPoolAccessHandlerPT: returns %Rrc (flushed)\n", rc));
950	return rc;
951
952	}
953
954
955	/**
956	* Handles the STOSD write accesses.
957	*
958	* @returns VBox status code suitable for scheduling.
959	* @param pVM The VM handle.
960	* @param pPool The pool.
961	* @param pPage The pool page (head).
962	* @param pDis The disassembly of the write instruction.
963	* @param pRegFrame The trap register frame.
964	* @param GCPhysFault The fault address as guest physical address.
965	* @param pvFault The fault address.
966	*/
967	DECLINLINE(int) pgmPoolAccessHandlerSTOSD(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pDis,
968	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
969	{
970	unsigned uIncrement = pDis->param1.size;
971
972	Assert(pDis->mode == CPUMODE_32BIT \|\| pDis->mode == CPUMODE_64BIT);
973	Assert(pRegFrame->rcx <= 0x20);
974
975	#ifdef VBOX_STRICT
976	if (pDis->opmode == CPUMODE_32BIT)
977	Assert(uIncrement == 4);
978	else
979	Assert(uIncrement == 8);
980	#endif
981
982	Log3(("pgmPoolAccessHandlerSTOSD\n"));
983
984	/*
985	* Increment the modification counter and insert it into the list
986	* of modified pages the first time.
987	*/
988	if (!pPage->cModifications++)
989	pgmPoolMonitorModifiedInsert(pPool, pPage);
990
991	/*
992	* Execute REP STOSD.
993	*
994	* This ASSUMES that we're not invoked by Trap0e on in a out-of-sync
995	* write situation, meaning that it's safe to write here.
996	*/
997	PVMCPU pVCpu = VMMGetCpu(pPool->CTX_SUFF(pVM));
998	RTGCUINTPTR pu32 = (RTGCUINTPTR)pvFault;
999	while (pRegFrame->rcx)
1000	{
1001	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
1002	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
1003	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, (RTGCPTR)pu32, NULL);
1004	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
1005	#else
1006	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, (RTGCPTR)pu32, NULL);
1007	#endif
1008	#ifdef IN_RC
1009	(uint32_t )pu32 = pRegFrame->eax;
1010	#else
1011	PGMPhysSimpleWriteGCPhys(pVM, GCPhysFault, &pRegFrame->rax, uIncrement);
1012	#endif
1013	pu32 += uIncrement;
1014	GCPhysFault += uIncrement;
1015	pRegFrame->rdi += uIncrement;
1016	pRegFrame->rcx--;
1017	}
1018	pRegFrame->rip += pDis->opsize;
1019
1020	#ifdef IN_RC
1021	/* See use in pgmPoolAccessHandlerSimple(). */
1022	PGM_INVL_VCPU_TLBS(pVCpu);
1023	#endif
1024
1025	LogFlow(("pgmPoolAccessHandlerSTOSD: returns\n"));
1026	return VINF_SUCCESS;
1027	}
1028
1029
1030	/**
1031	* Handles the simple write accesses.
1032	*
1033	* @returns VBox status code suitable for scheduling.
1034	* @param pVM The VM handle.
1035	* @param pVCpu The VMCPU handle.
1036	* @param pPool The pool.
1037	* @param pPage The pool page (head).
1038	* @param pDis The disassembly of the write instruction.
1039	* @param pRegFrame The trap register frame.
1040	* @param GCPhysFault The fault address as guest physical address.
1041	* @param pvFault The fault address.
1042	*/
1043	DECLINLINE(int) pgmPoolAccessHandlerSimple(PVM pVM, PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pDis,
1044	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
1045	{
1046	Log3(("pgmPoolAccessHandlerSimple\n"));
1047	/*
1048	* Increment the modification counter and insert it into the list
1049	* of modified pages the first time.
1050	*/
1051	if (!pPage->cModifications++)
1052	pgmPoolMonitorModifiedInsert(pPool, pPage);
1053
1054	/*
1055	* Clear all the pages. ASSUMES that pvFault is readable.
1056	*/
1057	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
1058	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
1059	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, pvFault, pDis);
1060	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
1061	#else
1062	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, pvFault, pDis);
1063	#endif
1064
1065	/*
1066	* Interpret the instruction.
1067	*/
1068	uint32_t cb;
1069	int rc = EMInterpretInstructionCPU(pVM, pVCpu, pDis, pRegFrame, pvFault, &cb);
1070	if (RT_SUCCESS(rc))
1071	pRegFrame->rip += pDis->opsize;
1072	else if (rc == VERR_EM_INTERPRETER)
1073	{
1074	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for %04x:%RGv - opcode=%d\n",
1075	pRegFrame->cs, (RTGCPTR)pRegFrame->rip, pDis->pCurInstr->opcode));
1076	rc = VINF_EM_RAW_EMULATE_INSTR;
1077	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
1078	}
1079
1080	#ifdef IN_RC
1081	/*
1082	* Quick hack, with logging enabled we're getting stale
1083	* code TLBs but no data TLB for EIP and crash in EMInterpretDisasOne.
1084	* Flushing here is BAD and expensive, I think EMInterpretDisasOne will
1085	* have to be fixed to support this. But that'll have to wait till next week.
1086	*
1087	* An alternative is to keep track of the changed PTEs together with the
1088	* GCPhys from the guest PT. This may proove expensive though.
1089	*
1090	* At the moment, it's VITAL that it's done AFTER the instruction interpreting
1091	* because we need the stale TLBs in some cases (XP boot). This MUST be fixed properly!
1092	*/
1093	PGM_INVL_VCPU_TLBS(pVCpu);
1094	#endif
1095
1096	LogFlow(("pgmPoolAccessHandlerSimple: returns %Rrc cb=%d\n", rc, cb));
1097	return rc;
1098	}
1099
1100	/**
1101	* \#PF Handler callback for PT write accesses.
1102	*
1103	* @returns VBox status code (appropriate for GC return).
1104	* @param pVM VM Handle.
1105	* @param uErrorCode CPU Error code.
1106	* @param pRegFrame Trap register frame.
1107	* NULL on DMA and other non CPU access.
1108	* @param pvFault The fault address (cr2).
1109	* @param GCPhysFault The GC physical address corresponding to pvFault.
1110	* @param pvUser User argument.
1111	*/
1112	DECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser)
1113	{
1114	STAM_PROFILE_START(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), a);
1115	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1116	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
1117	PVMCPU pVCpu = VMMGetCpu(pVM);
1118	unsigned cMaxModifications;
1119
1120	LogFlow(("pgmPoolAccessHandler: pvFault=%RGv pPage=%p:{.idx=%d} GCPhysFault=%RGp\n", pvFault, pPage, pPage->idx, GCPhysFault));
1121
1122	pgmLock(pVM);
1123	if (PHYS_PAGE_ADDRESS(GCPhysFault) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
1124	{
1125	/* Pool page changed while we were waiting for the lock; ignore. */
1126	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)));
1127	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,Handled), a);
1128	pgmUnlock(pVM);
1129	return VINF_SUCCESS;
1130	}
1131
1132	/*
1133	* Disassemble the faulting instruction.
1134	*/
1135	PDISCPUSTATE pDis = &pVCpu->pgm.s.DisState;
1136	int rc = EMInterpretDisasOne(pVM, pVCpu, pRegFrame, pDis, NULL);
1137	AssertReturnStmt(rc == VINF_SUCCESS, pgmUnlock(pVM), rc);
1138
1139	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
1140
1141	/*
1142	* We should ALWAYS have the list head as user parameter. This
1143	* is because we use that page to record the changes.
1144	*/
1145	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1146
1147	/* Maximum nr of modifications depends on the guest mode. */
1148	if (pDis->mode == CPUMODE_32BIT)
1149	cMaxModifications = 48;
1150	else
1151	cMaxModifications = 24;
1152
1153	/*
1154	* Incremental page table updates should weight more than random ones.
1155	* (Only applies when started from offset 0)
1156	*/
1157	pVCpu->pgm.s.cPoolAccessHandler++;
1158	if ( pPage->pvLastAccessHandlerRip >= pRegFrame->rip - 0x40 /* observed loops in Windows 7 x64 */
1159	&& pPage->pvLastAccessHandlerRip < pRegFrame->rip + 0x40
1160	&& pvFault == (pPage->pvLastAccessHandlerFault + pDis->param1.size)
1161	&& pVCpu->pgm.s.cPoolAccessHandler == (pPage->cLastAccessHandlerCount + 1))
1162	{
1163	Log(("Possible page reuse cMods=%d -> %d (locked=%d type=%s)\n", pPage->cModifications, pPage->cModifications * 2, pgmPoolIsPageLocked(&pVM->pgm.s, pPage), pgmPoolPoolKindToStr(pPage->enmKind)));
1164	pPage->cModifications = pPage->cModifications * 2;
1165	pPage->pvLastAccessHandlerFault = pvFault;
1166	pPage->cLastAccessHandlerCount = pVCpu->pgm.s.cPoolAccessHandler;
1167	if (pPage->cModifications > cMaxModifications)
1168	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FlushReinit));
1169	}
1170
1171	/*
1172	* Check if it's worth dealing with.
1173	*/
1174	bool fReused = false;
1175	if ( ( pPage->cModifications < cMaxModifications /** @todo #define / /* @todo need to check that it's not mapping EIP. / /* @todo adjust this! */
1176	\|\| pgmPoolIsPageLocked(&pVM->pgm.s, pPage)
1177	)
1178	&& !(fReused = pgmPoolMonitorIsReused(pVM, pVCpu, pRegFrame, pDis, pvFault))
1179	&& !pgmPoolMonitorIsForking(pPool, pDis, GCPhysFault & PAGE_OFFSET_MASK))
1180	{
1181	/*
1182	* Simple instructions, no REP prefix.
1183	*/
1184	if (!(pDis->prefix & (PREFIX_REP \| PREFIX_REPNE)))
1185	{
1186	rc = pgmPoolAccessHandlerSimple(pVM, pVCpu, pPool, pPage, pDis, pRegFrame, GCPhysFault, pvFault);
1187
1188	/* A mov instruction to change the first page table entry will be remembered so we can detect
1189	* full page table changes early on. This will reduce the amount of unnecessary traps we'll take.
1190	*/
1191	if ( rc == VINF_SUCCESS
1192	&& pDis->pCurInstr->opcode == OP_MOV
1193	&& (pvFault & PAGE_OFFSET_MASK) == 0)
1194	{
1195	pPage->pvLastAccessHandlerFault = pvFault;
1196	pPage->cLastAccessHandlerCount = pVCpu->pgm.s.cPoolAccessHandler;
1197	pPage->pvLastAccessHandlerRip = pRegFrame->rip;
1198	}
1199	else
1200	if (pPage->pvLastAccessHandlerFault == pvFault)
1201	{
1202	/* ignore the 2nd write to this page table entry. */
1203	pPage->cLastAccessHandlerCount = pVCpu->pgm.s.cPoolAccessHandler;
1204	}
1205	else
1206	{
1207	pPage->pvLastAccessHandlerFault = 0;
1208	pPage->pvLastAccessHandlerRip = 0;
1209	}
1210
1211	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,Handled), a);
1212	pgmUnlock(pVM);
1213	return rc;
1214	}
1215
1216	/*
1217	* Windows is frequently doing small memset() operations (netio test 4k+).
1218	* We have to deal with these or we'll kill the cache and performance.
1219	*/
1220	if ( pDis->pCurInstr->opcode == OP_STOSWD
1221	&& !pRegFrame->eflags.Bits.u1DF
1222	&& pDis->opmode == pDis->mode
1223	&& pDis->addrmode == pDis->mode)
1224	{
1225	bool fValidStosd = false;
1226
1227	if ( pDis->mode == CPUMODE_32BIT
1228	&& pDis->prefix == PREFIX_REP
1229	&& pRegFrame->ecx <= 0x20
1230	&& pRegFrame->ecx * 4 <= PAGE_SIZE - ((uintptr_t)pvFault & PAGE_OFFSET_MASK)
1231	&& !((uintptr_t)pvFault & 3)
1232	&& (pRegFrame->eax == 0 \|\| pRegFrame->eax == 0x80) /* the two values observed. */
1233	)
1234	{
1235	fValidStosd = true;
1236	pRegFrame->rcx &= 0xffffffff; /* paranoia */
1237	}
1238	else
1239	if ( pDis->mode == CPUMODE_64BIT
1240	&& pDis->prefix == (PREFIX_REP \| PREFIX_REX)
1241	&& pRegFrame->rcx <= 0x20
1242	&& pRegFrame->rcx * 8 <= PAGE_SIZE - ((uintptr_t)pvFault & PAGE_OFFSET_MASK)
1243	&& !((uintptr_t)pvFault & 7)
1244	&& (pRegFrame->rax == 0 \|\| pRegFrame->rax == 0x80) /* the two values observed. */
1245	)
1246	{
1247	fValidStosd = true;
1248	}
1249
1250	if (fValidStosd)
1251	{
1252	rc = pgmPoolAccessHandlerSTOSD(pVM, pPool, pPage, pDis, pRegFrame, GCPhysFault, pvFault);
1253	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,RepStosd), a);
1254	pgmUnlock(pVM);
1255	return rc;
1256	}
1257	}
1258
1259	/* REP prefix, don't bother. */
1260	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,RepPrefix));
1261	Log4(("pgmPoolAccessHandler: eax=%#x ecx=%#x edi=%#x esi=%#x rip=%RGv opcode=%d prefix=%#x\n",
1262	pRegFrame->eax, pRegFrame->ecx, pRegFrame->edi, pRegFrame->esi, (RTGCPTR)pRegFrame->rip, pDis->pCurInstr->opcode, pDis->prefix));
1263	}
1264
1265	/*
1266	* Not worth it, so flush it.
1267	*
1268	* If we considered it to be reused, don't go back to ring-3
1269	* to emulate failed instructions since we usually cannot
1270	* interpret then. This may be a bit risky, in which case
1271	* the reuse detection must be fixed.
1272	*/
1273	rc = pgmPoolAccessHandlerFlush(pVM, pVCpu, pPool, pPage, pDis, pRegFrame, GCPhysFault, pvFault);
1274	if (rc == VINF_EM_RAW_EMULATE_INSTR && fReused)
1275	rc = VINF_SUCCESS;
1276	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,FlushPage), a);
1277	pgmUnlock(pVM);
1278	return rc;
1279	}
1280
1281	# endif /* !IN_RING3 */
1282	#endif /* PGMPOOL_WITH_MONITORING */
1283
1284	#ifdef PGMPOOL_WITH_CACHE
1285
1286	/**
1287	* Inserts a page into the GCPhys hash table.
1288	*
1289	* @param pPool The pool.
1290	* @param pPage The page.
1291	*/
1292	DECLINLINE(void) pgmPoolHashInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1293	{
1294	Log3(("pgmPoolHashInsert: %RGp\n", pPage->GCPhys));
1295	Assert(pPage->GCPhys != NIL_RTGCPHYS); Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1296	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1297	pPage->iNext = pPool->aiHash[iHash];
1298	pPool->aiHash[iHash] = pPage->idx;
1299	}
1300
1301
1302	/**
1303	* Removes a page from the GCPhys hash table.
1304	*
1305	* @param pPool The pool.
1306	* @param pPage The page.
1307	*/
1308	DECLINLINE(void) pgmPoolHashRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1309	{
1310	Log3(("pgmPoolHashRemove: %RGp\n", pPage->GCPhys));
1311	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1312	if (pPool->aiHash[iHash] == pPage->idx)
1313	pPool->aiHash[iHash] = pPage->iNext;
1314	else
1315	{
1316	uint16_t iPrev = pPool->aiHash[iHash];
1317	for (;;)
1318	{
1319	const int16_t i = pPool->aPages[iPrev].iNext;
1320	if (i == pPage->idx)
1321	{
1322	pPool->aPages[iPrev].iNext = pPage->iNext;
1323	break;
1324	}
1325	if (i == NIL_PGMPOOL_IDX)
1326	{
1327	AssertReleaseMsgFailed(("GCPhys=%RGp idx=%#x\n", pPage->GCPhys, pPage->idx));
1328	break;
1329	}
1330	iPrev = i;
1331	}
1332	}
1333	pPage->iNext = NIL_PGMPOOL_IDX;
1334	}
1335
1336
1337	/**
1338	* Frees up one cache page.
1339	*
1340	* @returns VBox status code.
1341	* @retval VINF_SUCCESS on success.
1342	* @param pPool The pool.
1343	* @param iUser The user index.
1344	*/
1345	static int pgmPoolCacheFreeOne(PPGMPOOL pPool, uint16_t iUser)
1346	{
1347	#ifndef IN_RC
1348	const PVM pVM = pPool->CTX_SUFF(pVM);
1349	#endif
1350	Assert(pPool->iAgeHead != pPool->iAgeTail); /* We shouldn't be here if there < 2 cached entries! */
1351	STAM_COUNTER_INC(&pPool->StatCacheFreeUpOne);
1352
1353	/*
1354	* Select one page from the tail of the age list.
1355	*/
1356	PPGMPOOLPAGE pPage;
1357	for (unsigned iLoop = 0; ; iLoop++)
1358	{
1359	uint16_t iToFree = pPool->iAgeTail;
1360	if (iToFree == iUser)
1361	iToFree = pPool->aPages[iToFree].iAgePrev;
1362	/* This is the alternative to the SyncCR3 pgmPoolCacheUsed calls.
1363	if (pPool->aPages[iToFree].iUserHead != NIL_PGMPOOL_USER_INDEX)
1364	{
1365	uint16_t i = pPool->aPages[iToFree].iAgePrev;
1366	for (unsigned j = 0; j < 10 && i != NIL_PGMPOOL_USER_INDEX; j++, i = pPool->aPages[i].iAgePrev)
1367	{
1368	if (pPool->aPages[iToFree].iUserHead == NIL_PGMPOOL_USER_INDEX)
1369	continue;
1370	iToFree = i;
1371	break;
1372	}
1373	}
1374	*/
1375	Assert(iToFree != iUser);
1376	AssertRelease(iToFree != NIL_PGMPOOL_IDX);
1377	pPage = &pPool->aPages[iToFree];
1378
1379	/*
1380	* Reject any attempts at flushing the currently active shadow CR3 mapping.
1381	* Call pgmPoolCacheUsed to move the page to the head of the age list.
1382	*/
1383	if (!pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage))
1384	break;
1385	LogFlow(("pgmPoolCacheFreeOne: refuse CR3 mapping\n"));
1386	pgmPoolCacheUsed(pPool, pPage);
1387	AssertLogRelReturn(iLoop < 8192, VERR_INTERNAL_ERROR);
1388	}
1389
1390	/*
1391	* Found a usable page, flush it and return.
1392	*/
1393	int rc = pgmPoolFlushPage(pPool, pPage);
1394	/* This flush was initiated by us and not the guest, so explicitly flush the TLB. */
1395	if (rc == VINF_SUCCESS)
1396	PGM_INVL_ALL_VCPU_TLBS(pVM);
1397	return rc;
1398	}
1399
1400
1401	/**
1402	* Checks if a kind mismatch is really a page being reused
1403	* or if it's just normal remappings.
1404	*
1405	* @returns true if reused and the cached page (enmKind1) should be flushed
1406	* @returns false if not reused.
1407	* @param enmKind1 The kind of the cached page.
1408	* @param enmKind2 The kind of the requested page.
1409	*/
1410	static bool pgmPoolCacheReusedByKind(PGMPOOLKIND enmKind1, PGMPOOLKIND enmKind2)
1411	{
1412	switch (enmKind1)
1413	{
1414	/*
1415	* Never reuse them. There is no remapping in non-paging mode.
1416	*/
1417	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1418	case PGMPOOLKIND_32BIT_PD_PHYS:
1419	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1420	case PGMPOOLKIND_PAE_PD_PHYS:
1421	case PGMPOOLKIND_PAE_PDPT_PHYS:
1422	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1423	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1424	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1425	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1426	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1427	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT: /* never reuse them for other types */
1428	return false;
1429
1430	/*
1431	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1432	*/
1433	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1434	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1435	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1436	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1437	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1438	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1439	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1440	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1441	case PGMPOOLKIND_32BIT_PD:
1442	case PGMPOOLKIND_PAE_PDPT:
1443	switch (enmKind2)
1444	{
1445	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1446	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1447	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1448	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1449	case PGMPOOLKIND_64BIT_PML4:
1450	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1451	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1452	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1453	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1454	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1455	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1456	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1457	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1458	return true;
1459	default:
1460	return false;
1461	}
1462
1463	/*
1464	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1465	*/
1466	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1467	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1468	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1469	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1470	case PGMPOOLKIND_64BIT_PML4:
1471	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1472	switch (enmKind2)
1473	{
1474	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1475	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1476	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1477	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1478	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1479	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1480	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1481	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1482	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1483	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1484	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1485	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1486	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1487	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1488	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1489	return true;
1490	default:
1491	return false;
1492	}
1493
1494	/*
1495	* These cannot be flushed, and it's common to reuse the PDs as PTs.
1496	*/
1497	case PGMPOOLKIND_ROOT_NESTED:
1498	return false;
1499
1500	default:
1501	AssertFatalMsgFailed(("enmKind1=%d\n", enmKind1));
1502	}
1503	}
1504
1505
1506	/**
1507	* Attempts to satisfy a pgmPoolAlloc request from the cache.
1508	*
1509	* @returns VBox status code.
1510	* @retval VINF_PGM_CACHED_PAGE on success.
1511	* @retval VERR_FILE_NOT_FOUND if not found.
1512	* @param pPool The pool.
1513	* @param GCPhys The GC physical address of the page we're gonna shadow.
1514	* @param enmKind The kind of mapping.
1515	* @param enmAccess Access type for the mapping (only relevant for big pages)
1516	* @param iUser The shadow page pool index of the user table.
1517	* @param iUserTable The index into the user table (shadowed).
1518	* @param ppPage Where to store the pointer to the page.
1519	*/
1520	static int pgmPoolCacheAlloc(PPGMPOOL pPool, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, PGMPOOLACCESS enmAccess, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
1521	{
1522	#ifndef IN_RC
1523	const PVM pVM = pPool->CTX_SUFF(pVM);
1524	#endif
1525	/*
1526	* Look up the GCPhys in the hash.
1527	*/
1528	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
1529	Log3(("pgmPoolCacheAlloc: %RGp kind %s iUser=%x iUserTable=%x SLOT=%d\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable, i));
1530	if (i != NIL_PGMPOOL_IDX)
1531	{
1532	do
1533	{
1534	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1535	Log4(("pgmPoolCacheAlloc: slot %d found page %RGp\n", i, pPage->GCPhys));
1536	if (pPage->GCPhys == GCPhys)
1537	{
1538	if ( (PGMPOOLKIND)pPage->enmKind == enmKind
1539	&& (PGMPOOLACCESS)pPage->enmAccess == enmAccess)
1540	{
1541	/* Put it at the start of the use list to make sure pgmPoolTrackAddUser
1542	* doesn't flush it in case there are no more free use records.
1543	*/
1544	pgmPoolCacheUsed(pPool, pPage);
1545
1546	int rc = pgmPoolTrackAddUser(pPool, pPage, iUser, iUserTable);
1547	if (RT_SUCCESS(rc))
1548	{
1549	Assert((PGMPOOLKIND)pPage->enmKind == enmKind);
1550	*ppPage = pPage;
1551	if (pPage->cModifications)
1552	pPage->cModifications = 1; /* reset counter (can't use 0, or else it will be reinserted in the modified list) */
1553	STAM_COUNTER_INC(&pPool->StatCacheHits);
1554	return VINF_PGM_CACHED_PAGE;
1555	}
1556	return rc;
1557	}
1558
1559	if ((PGMPOOLKIND)pPage->enmKind != enmKind)
1560	{
1561	/*
1562	* The kind is different. In some cases we should now flush the page
1563	* as it has been reused, but in most cases this is normal remapping
1564	* of PDs as PT or big pages using the GCPhys field in a slightly
1565	* different way than the other kinds.
1566	*/
1567	if (pgmPoolCacheReusedByKind((PGMPOOLKIND)pPage->enmKind, enmKind))
1568	{
1569	STAM_COUNTER_INC(&pPool->StatCacheKindMismatches);
1570	pgmPoolFlushPage(pPool, pPage);
1571	PGM_INVL_VCPU_TLBS(VMMGetCpu(pVM)); /* see PT handler. */
1572	break;
1573	}
1574	}
1575	}
1576
1577	/* next */
1578	i = pPage->iNext;
1579	} while (i != NIL_PGMPOOL_IDX);
1580	}
1581
1582	Log3(("pgmPoolCacheAlloc: Missed GCPhys=%RGp enmKind=%s\n", GCPhys, pgmPoolPoolKindToStr(enmKind)));
1583	STAM_COUNTER_INC(&pPool->StatCacheMisses);
1584	return VERR_FILE_NOT_FOUND;
1585	}
1586
1587
1588	/**
1589	* Inserts a page into the cache.
1590	*
1591	* @param pPool The pool.
1592	* @param pPage The cached page.
1593	* @param fCanBeCached Set if the page is fit for caching from the caller's point of view.
1594	*/
1595	static void pgmPoolCacheInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fCanBeCached)
1596	{
1597	/*
1598	* Insert into the GCPhys hash if the page is fit for that.
1599	*/
1600	Assert(!pPage->fCached);
1601	if (fCanBeCached)
1602	{
1603	pPage->fCached = true;
1604	pgmPoolHashInsert(pPool, pPage);
1605	Log3(("pgmPoolCacheInsert: Caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
1606	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
1607	STAM_COUNTER_INC(&pPool->StatCacheCacheable);
1608	}
1609	else
1610	{
1611	Log3(("pgmPoolCacheInsert: Not caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
1612	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
1613	STAM_COUNTER_INC(&pPool->StatCacheUncacheable);
1614	}
1615
1616	/*
1617	* Insert at the head of the age list.
1618	*/
1619	pPage->iAgePrev = NIL_PGMPOOL_IDX;
1620	pPage->iAgeNext = pPool->iAgeHead;
1621	if (pPool->iAgeHead != NIL_PGMPOOL_IDX)
1622	pPool->aPages[pPool->iAgeHead].iAgePrev = pPage->idx;
1623	else
1624	pPool->iAgeTail = pPage->idx;
1625	pPool->iAgeHead = pPage->idx;
1626	}
1627
1628
1629	/**
1630	* Flushes a cached page.
1631	*
1632	* @param pPool The pool.
1633	* @param pPage The cached page.
1634	*/
1635	static void pgmPoolCacheFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1636	{
1637	Log3(("pgmPoolCacheFlushPage: %RGp\n", pPage->GCPhys));
1638
1639	/*
1640	* Remove the page from the hash.
1641	*/
1642	if (pPage->fCached)
1643	{
1644	pPage->fCached = false;
1645	pgmPoolHashRemove(pPool, pPage);
1646	}
1647	else
1648	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1649
1650	/*
1651	* Remove it from the age list.
1652	*/
1653	if (pPage->iAgeNext != NIL_PGMPOOL_IDX)
1654	pPool->aPages[pPage->iAgeNext].iAgePrev = pPage->iAgePrev;
1655	else
1656	pPool->iAgeTail = pPage->iAgePrev;
1657	if (pPage->iAgePrev != NIL_PGMPOOL_IDX)
1658	pPool->aPages[pPage->iAgePrev].iAgeNext = pPage->iAgeNext;
1659	else
1660	pPool->iAgeHead = pPage->iAgeNext;
1661	pPage->iAgeNext = NIL_PGMPOOL_IDX;
1662	pPage->iAgePrev = NIL_PGMPOOL_IDX;
1663	}
1664
1665	#endif /* PGMPOOL_WITH_CACHE */
1666	#ifdef PGMPOOL_WITH_MONITORING
1667
1668	/**
1669	* Looks for pages sharing the monitor.
1670	*
1671	* @returns Pointer to the head page.
1672	* @returns NULL if not found.
1673	* @param pPool The Pool
1674	* @param pNewPage The page which is going to be monitored.
1675	*/
1676	static PPGMPOOLPAGE pgmPoolMonitorGetPageByGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pNewPage)
1677	{
1678	#ifdef PGMPOOL_WITH_CACHE
1679	/*
1680	* Look up the GCPhys in the hash.
1681	*/
1682	RTGCPHYS GCPhys = pNewPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1683	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
1684	if (i == NIL_PGMPOOL_IDX)
1685	return NULL;
1686	do
1687	{
1688	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1689	if ( pPage->GCPhys - GCPhys < PAGE_SIZE
1690	&& pPage != pNewPage)
1691	{
1692	switch (pPage->enmKind)
1693	{
1694	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1695	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1696	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1697	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1698	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1699	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1700	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1701	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1702	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1703	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1704	case PGMPOOLKIND_64BIT_PML4:
1705	case PGMPOOLKIND_32BIT_PD:
1706	case PGMPOOLKIND_PAE_PDPT:
1707	{
1708	/* find the head */
1709	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1710	{
1711	Assert(pPage->iMonitoredPrev != pPage->idx);
1712	pPage = &pPool->aPages[pPage->iMonitoredPrev];
1713	}
1714	return pPage;
1715	}
1716
1717	/* ignore, no monitoring. */
1718	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1719	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1720	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1721	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1722	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1723	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1724	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1725	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1726	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1727	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1728	case PGMPOOLKIND_ROOT_NESTED:
1729	case PGMPOOLKIND_PAE_PD_PHYS:
1730	case PGMPOOLKIND_PAE_PDPT_PHYS:
1731	case PGMPOOLKIND_32BIT_PD_PHYS:
1732	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
1733	break;
1734	default:
1735	AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
1736	}
1737	}
1738
1739	/* next */
1740	i = pPage->iNext;
1741	} while (i != NIL_PGMPOOL_IDX);
1742	#endif
1743	return NULL;
1744	}
1745
1746
1747	/**
1748	* Enabled write monitoring of a guest page.
1749	*
1750	* @returns VBox status code.
1751	* @retval VINF_SUCCESS on success.
1752	* @param pPool The pool.
1753	* @param pPage The cached page.
1754	*/
1755	static int pgmPoolMonitorInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1756	{
1757	LogFlow(("pgmPoolMonitorInsert %RGp\n", pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1)));
1758
1759	/*
1760	* Filter out the relevant kinds.
1761	*/
1762	switch (pPage->enmKind)
1763	{
1764	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1765	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1766	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1767	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1768	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1769	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1770	case PGMPOOLKIND_64BIT_PML4:
1771	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1772	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1773	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1774	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1775	case PGMPOOLKIND_32BIT_PD:
1776	case PGMPOOLKIND_PAE_PDPT:
1777	break;
1778
1779	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1780	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1781	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1782	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1783	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1784	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1785	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1786	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1787	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1788	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1789	case PGMPOOLKIND_ROOT_NESTED:
1790	/* Nothing to monitor here. */
1791	return VINF_SUCCESS;
1792
1793	case PGMPOOLKIND_32BIT_PD_PHYS:
1794	case PGMPOOLKIND_PAE_PDPT_PHYS:
1795	case PGMPOOLKIND_PAE_PD_PHYS:
1796	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
1797	/* Nothing to monitor here. */
1798	return VINF_SUCCESS;
1799	#ifdef PGMPOOL_WITH_MIXED_PT_CR3
1800	break;
1801	#else
1802	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1803	#endif
1804	default:
1805	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
1806	}
1807
1808	/*
1809	* Install handler.
1810	*/
1811	int rc;
1812	PPGMPOOLPAGE pPageHead = pgmPoolMonitorGetPageByGCPhys(pPool, pPage);
1813	if (pPageHead)
1814	{
1815	Assert(pPageHead != pPage); Assert(pPageHead->iMonitoredNext != pPage->idx);
1816	Assert(pPageHead->iMonitoredPrev != pPage->idx);
1817	pPage->iMonitoredPrev = pPageHead->idx;
1818	pPage->iMonitoredNext = pPageHead->iMonitoredNext;
1819	if (pPageHead->iMonitoredNext != NIL_PGMPOOL_IDX)
1820	pPool->aPages[pPageHead->iMonitoredNext].iMonitoredPrev = pPage->idx;
1821	pPageHead->iMonitoredNext = pPage->idx;
1822	rc = VINF_SUCCESS;
1823	}
1824	else
1825	{
1826	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX); Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1827	PVM pVM = pPool->CTX_SUFF(pVM);
1828	const RTGCPHYS GCPhysPage = pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1829	rc = PGMHandlerPhysicalRegisterEx(pVM, PGMPHYSHANDLERTYPE_PHYSICAL_WRITE,
1830	GCPhysPage, GCPhysPage + (PAGE_SIZE - 1),
1831	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
1832	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
1833	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
1834	pPool->pszAccessHandler);
1835	/** @todo we should probably deal with out-of-memory conditions here, but for now increasing
1836	* the heap size should suffice. */
1837	AssertFatalMsgRC(rc, ("PGMHandlerPhysicalRegisterEx %RGp failed with %Rrc\n", GCPhysPage, rc));
1838	Assert(!(VMMGetCpu(pVM)->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VMCPU_FF_ISSET(VMMGetCpu(pVM), VMCPU_FF_PGM_SYNC_CR3));
1839	}
1840	pPage->fMonitored = true;
1841	return rc;
1842	}
1843
1844
1845	/**
1846	* Disables write monitoring of a guest page.
1847	*
1848	* @returns VBox status code.
1849	* @retval VINF_SUCCESS on success.
1850	* @param pPool The pool.
1851	* @param pPage The cached page.
1852	*/
1853	static int pgmPoolMonitorFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1854	{
1855	/*
1856	* Filter out the relevant kinds.
1857	*/
1858	switch (pPage->enmKind)
1859	{
1860	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1861	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1862	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1863	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1864	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1865	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1866	case PGMPOOLKIND_64BIT_PML4:
1867	case PGMPOOLKIND_32BIT_PD:
1868	case PGMPOOLKIND_PAE_PDPT:
1869	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1870	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1871	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1872	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1873	break;
1874
1875	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1876	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1877	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1878	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1879	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1880	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1881	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1882	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1883	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1884	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1885	case PGMPOOLKIND_ROOT_NESTED:
1886	case PGMPOOLKIND_PAE_PD_PHYS:
1887	case PGMPOOLKIND_PAE_PDPT_PHYS:
1888	case PGMPOOLKIND_32BIT_PD_PHYS:
1889	/* Nothing to monitor here. */
1890	return VINF_SUCCESS;
1891
1892	#ifdef PGMPOOL_WITH_MIXED_PT_CR3
1893	break;
1894	#endif
1895	default:
1896	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
1897	}
1898
1899	/*
1900	* Remove the page from the monitored list or uninstall it if last.
1901	*/
1902	const PVM pVM = pPool->CTX_SUFF(pVM);
1903	int rc;
1904	if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
1905	\|\| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1906	{
1907	if (pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
1908	{
1909	PPGMPOOLPAGE pNewHead = &pPool->aPages[pPage->iMonitoredNext];
1910	pNewHead->iMonitoredPrev = NIL_PGMPOOL_IDX;
1911	rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
1912	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pNewHead),
1913	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pNewHead),
1914	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pNewHead),
1915	pPool->pszAccessHandler);
1916	AssertFatalRCSuccess(rc);
1917	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
1918	}
1919	else
1920	{
1921	pPool->aPages[pPage->iMonitoredPrev].iMonitoredNext = pPage->iMonitoredNext;
1922	if (pPage->iMonitoredNext != NIL_PGMPOOL_IDX)
1923	{
1924	pPool->aPages[pPage->iMonitoredNext].iMonitoredPrev = pPage->iMonitoredPrev;
1925	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
1926	}
1927	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
1928	rc = VINF_SUCCESS;
1929	}
1930	}
1931	else
1932	{
1933	rc = PGMHandlerPhysicalDeregister(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1));
1934	AssertFatalRC(rc);
1935	#ifdef VBOX_STRICT
1936	PVMCPU pVCpu = VMMGetCpu(pVM);
1937	#endif
1938	AssertMsg(!(pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3),
1939	("%#x %#x\n", pVCpu->pgm.s.fSyncFlags, pVM->fGlobalForcedActions));
1940	}
1941	pPage->fMonitored = false;
1942
1943	/*
1944	* Remove it from the list of modified pages (if in it).
1945	*/
1946	pgmPoolMonitorModifiedRemove(pPool, pPage);
1947
1948	return rc;
1949	}
1950
1951
1952	/**
1953	* Inserts the page into the list of modified pages.
1954	*
1955	* @param pPool The pool.
1956	* @param pPage The page.
1957	*/
1958	void pgmPoolMonitorModifiedInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1959	{
1960	Log3(("pgmPoolMonitorModifiedInsert: idx=%d\n", pPage->idx));
1961	AssertMsg( pPage->iModifiedNext == NIL_PGMPOOL_IDX
1962	&& pPage->iModifiedPrev == NIL_PGMPOOL_IDX
1963	&& pPool->iModifiedHead != pPage->idx,
1964	("Next=%d Prev=%d idx=%d cModifications=%d Head=%d cModifiedPages=%d\n",
1965	pPage->iModifiedNext, pPage->iModifiedPrev, pPage->idx, pPage->cModifications,
1966	pPool->iModifiedHead, pPool->cModifiedPages));
1967
1968	pPage->iModifiedNext = pPool->iModifiedHead;
1969	if (pPool->iModifiedHead != NIL_PGMPOOL_IDX)
1970	pPool->aPages[pPool->iModifiedHead].iModifiedPrev = pPage->idx;
1971	pPool->iModifiedHead = pPage->idx;
1972	pPool->cModifiedPages++;
1973	#ifdef VBOX_WITH_STATISTICS
1974	if (pPool->cModifiedPages > pPool->cModifiedPagesHigh)
1975	pPool->cModifiedPagesHigh = pPool->cModifiedPages;
1976	#endif
1977	}
1978
1979
1980	/**
1981	* Removes the page from the list of modified pages and resets the
1982	* moficiation counter.
1983	*
1984	* @param pPool The pool.
1985	* @param pPage The page which is believed to be in the list of modified pages.
1986	*/
1987	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1988	{
1989	Log3(("pgmPoolMonitorModifiedRemove: idx=%d cModifications=%d\n", pPage->idx, pPage->cModifications));
1990	if (pPool->iModifiedHead == pPage->idx)
1991	{
1992	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1993	pPool->iModifiedHead = pPage->iModifiedNext;
1994	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1995	{
1996	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = NIL_PGMPOOL_IDX;
1997	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1998	}
1999	pPool->cModifiedPages--;
2000	}
2001	else if (pPage->iModifiedPrev != NIL_PGMPOOL_IDX)
2002	{
2003	pPool->aPages[pPage->iModifiedPrev].iModifiedNext = pPage->iModifiedNext;
2004	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
2005	{
2006	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = pPage->iModifiedPrev;
2007	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2008	}
2009	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2010	pPool->cModifiedPages--;
2011	}
2012	else
2013	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
2014	pPage->cModifications = 0;
2015	}
2016
2017
2018	/**
2019	* Zaps the list of modified pages, resetting their modification counters in the process.
2020	*
2021	* @param pVM The VM handle.
2022	*/
2023	void pgmPoolMonitorModifiedClearAll(PVM pVM)
2024	{
2025	pgmLock(pVM);
2026	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2027	LogFlow(("pgmPoolMonitorModifiedClearAll: cModifiedPages=%d\n", pPool->cModifiedPages));
2028
2029	unsigned cPages = 0; NOREF(cPages);
2030	uint16_t idx = pPool->iModifiedHead;
2031	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
2032	while (idx != NIL_PGMPOOL_IDX)
2033	{
2034	PPGMPOOLPAGE pPage = &pPool->aPages[idx];
2035	idx = pPage->iModifiedNext;
2036	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2037	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2038	pPage->cModifications = 0;
2039	Assert(++cPages);
2040	}
2041	AssertMsg(cPages == pPool->cModifiedPages, ("%d != %d\n", cPages, pPool->cModifiedPages));
2042	pPool->cModifiedPages = 0;
2043	pgmUnlock(pVM);
2044	}
2045
2046
2047	#ifdef IN_RING3
2048	/**
2049	* Callback to clear all shadow pages and clear all modification counters.
2050	*
2051	* @returns VBox status code.
2052	* @param pVM The VM handle.
2053	* @param pVCpu The VMCPU for the EMT we're being called on. Unused.
2054	* @param pvUser Unused parameter.
2055	*
2056	* @remark Should only be used when monitoring is available, thus placed in
2057	* the PGMPOOL_WITH_MONITORING \#ifdef.
2058	*/
2059	DECLCALLBACK(int) pgmPoolClearAll(PVM pVM, PVMCPU pVCpu, void *pvUser)
2060	{
2061	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2062	STAM_PROFILE_START(&pPool->StatClearAll, c);
2063	LogFlow(("pgmPoolClearAll: cUsedPages=%d\n", pPool->cUsedPages));
2064	NOREF(pvUser); NOREF(pVCpu);
2065
2066	pgmLock(pVM);
2067
2068	/*
2069	* Iterate all the pages until we've encountered all that in use.
2070	* This is simple but not quite optimal solution.
2071	*/
2072	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
2073	unsigned cLeft = pPool->cUsedPages;
2074	unsigned iPage = pPool->cCurPages;
2075	while (--iPage >= PGMPOOL_IDX_FIRST)
2076	{
2077	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2078	if (pPage->GCPhys != NIL_RTGCPHYS)
2079	{
2080	switch (pPage->enmKind)
2081	{
2082	/*
2083	* We only care about shadow page tables.
2084	*/
2085	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2086	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2087	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2088	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2089	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2090	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2091	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2092	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2093	{
2094	#ifdef PGMPOOL_WITH_USER_TRACKING
2095	if (pPage->cPresent)
2096	#endif
2097	{
2098	void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
2099	STAM_PROFILE_START(&pPool->StatZeroPage, z);
2100	ASMMemZeroPage(pvShw);
2101	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
2102	#ifdef PGMPOOL_WITH_USER_TRACKING
2103	pPage->cPresent = 0;
2104	pPage->iFirstPresent = ~0;
2105	#endif
2106	}
2107	}
2108	/* fall thru */
2109
2110	default:
2111	Assert(!pPage->cModifications \|\| ++cModifiedPages);
2112	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2113	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2114	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2115	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2116	pPage->cModifications = 0;
2117	break;
2118
2119	}
2120	if (!--cLeft)
2121	break;
2122	}
2123	}
2124
2125	/* swipe the special pages too. */
2126	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
2127	{
2128	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2129	if (pPage->GCPhys != NIL_RTGCPHYS)
2130	{
2131	Assert(!pPage->cModifications \|\| ++cModifiedPages);
2132	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2133	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2134	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2135	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2136	pPage->cModifications = 0;
2137	}
2138	}
2139
2140	#ifndef DEBUG_michael
2141	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
2142	#endif
2143	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
2144	pPool->cModifiedPages = 0;
2145
2146	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2147	/*
2148	* Clear all the GCPhys links and rebuild the phys ext free list.
2149	*/
2150	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
2151	pRam;
2152	pRam = pRam->CTX_SUFF(pNext))
2153	{
2154	unsigned iPage = pRam->cb >> PAGE_SHIFT;
2155	while (iPage-- > 0)
2156	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
2157	}
2158
2159	pPool->iPhysExtFreeHead = 0;
2160	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
2161	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
2162	for (unsigned i = 0; i < cMaxPhysExts; i++)
2163	{
2164	paPhysExts[i].iNext = i + 1;
2165	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
2166	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
2167	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
2168	}
2169	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
2170	#endif
2171
2172	/* Clear the PGM_SYNC_CLEAR_PGM_POOL flag on all VCPUs to prevent redundant flushes. */
2173	for (unsigned idCpu = 0; idCpu < pVM->cCPUs; idCpu++)
2174	{
2175	PVMCPU pVCpu = &pVM->aCpus[idCpu];
2176
2177	pVCpu->pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
2178	}
2179
2180	pPool->cPresent = 0;
2181	pgmUnlock(pVM);
2182	PGM_INVL_ALL_VCPU_TLBS(pVM);
2183	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
2184	return VINF_SUCCESS;
2185	}
2186	#endif /* IN_RING3 */
2187
2188
2189	/**
2190	* Handle SyncCR3 pool tasks
2191	*
2192	* @returns VBox status code.
2193	* @retval VINF_SUCCESS if successfully added.
2194	* @retval VINF_PGM_SYNC_CR3 is it needs to be deferred to ring 3 (GC only)
2195	* @param pVCpu The VMCPU handle.
2196	* @remark Should only be used when monitoring is available, thus placed in
2197	* the PGMPOOL_WITH_MONITORING #ifdef.
2198	*/
2199	int pgmPoolSyncCR3(PVMCPU pVCpu)
2200	{
2201	PVM pVM = pVCpu->CTX_SUFF(pVM);
2202	LogFlow(("pgmPoolSyncCR3\n"));
2203
2204	/*
2205	* When monitoring shadowed pages, we reset the modification counters on CR3 sync.
2206	* Occasionally we will have to clear all the shadow page tables because we wanted
2207	* to monitor a page which was mapped by too many shadowed page tables. This operation
2208	* sometimes refered to as a 'lightweight flush'.
2209	*/
2210	# ifdef IN_RING3 /* Don't flush in ring-0 or raw mode, it's taking too long. */
2211	if (ASMBitTestAndClear(&pVCpu->pgm.s.fSyncFlags, PGM_SYNC_CLEAR_PGM_POOL_BIT))
2212	{
2213	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmPoolClearAll, NULL);
2214	AssertRC(rc);
2215	}
2216	# else /* !IN_RING3 */
2217	if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
2218	{
2219	LogFlow(("SyncCR3: PGM_SYNC_CLEAR_PGM_POOL is set -> VINF_PGM_SYNC_CR3\n"));
2220	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
2221	return VINF_PGM_SYNC_CR3;
2222	}
2223	# endif /* !IN_RING3 */
2224	else
2225	pgmPoolMonitorModifiedClearAll(pVM);
2226
2227	return VINF_SUCCESS;
2228	}
2229
2230	#endif /* PGMPOOL_WITH_MONITORING */
2231	#ifdef PGMPOOL_WITH_USER_TRACKING
2232
2233	/**
2234	* Frees up at least one user entry.
2235	*
2236	* @returns VBox status code.
2237	* @retval VINF_SUCCESS if successfully added.
2238	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2239	* @param pPool The pool.
2240	* @param iUser The user index.
2241	*/
2242	static int pgmPoolTrackFreeOneUser(PPGMPOOL pPool, uint16_t iUser)
2243	{
2244	STAM_COUNTER_INC(&pPool->StatTrackFreeUpOneUser);
2245	#ifdef PGMPOOL_WITH_CACHE
2246	/*
2247	* Just free cached pages in a braindead fashion.
2248	*/
2249	/** @todo walk the age list backwards and free the first with usage. */
2250	int rc = VINF_SUCCESS;
2251	do
2252	{
2253	int rc2 = pgmPoolCacheFreeOne(pPool, iUser);
2254	if (RT_FAILURE(rc2) && rc == VINF_SUCCESS)
2255	rc = rc2;
2256	} while (pPool->iUserFreeHead == NIL_PGMPOOL_USER_INDEX);
2257	return rc;
2258	#else
2259	/*
2260	* Lazy approach.
2261	*/
2262	/* @todo This path no longer works (CR3 root pages will be flushed)!! */
2263	AssertCompileFailed();
2264	Assert(!CPUMIsGuestInLongMode(pVM));
2265	pgmPoolFlushAllInt(pPool);
2266	return VERR_PGM_POOL_FLUSHED;
2267	#endif
2268	}
2269
2270
2271	/**
2272	* Inserts a page into the cache.
2273	*
2274	* This will create user node for the page, insert it into the GCPhys
2275	* hash, and insert it into the age list.
2276	*
2277	* @returns VBox status code.
2278	* @retval VINF_SUCCESS if successfully added.
2279	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2280	* @param pPool The pool.
2281	* @param pPage The cached page.
2282	* @param GCPhys The GC physical address of the page we're gonna shadow.
2283	* @param iUser The user index.
2284	* @param iUserTable The user table index.
2285	*/
2286	DECLINLINE(int) pgmPoolTrackInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhys, uint16_t iUser, uint32_t iUserTable)
2287	{
2288	int rc = VINF_SUCCESS;
2289	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2290
2291	LogFlow(("pgmPoolTrackInsert GCPhys=%RGp iUser %x iUserTable %x\n", GCPhys, iUser, iUserTable));
2292
2293	#ifdef VBOX_STRICT
2294	/*
2295	* Check that the entry doesn't already exists.
2296	*/
2297	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2298	{
2299	uint16_t i = pPage->iUserHead;
2300	do
2301	{
2302	Assert(i < pPool->cMaxUsers);
2303	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2304	i = paUsers[i].iNext;
2305	} while (i != NIL_PGMPOOL_USER_INDEX);
2306	}
2307	#endif
2308
2309	/*
2310	* Find free a user node.
2311	*/
2312	uint16_t i = pPool->iUserFreeHead;
2313	if (i == NIL_PGMPOOL_USER_INDEX)
2314	{
2315	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2316	if (RT_FAILURE(rc))
2317	return rc;
2318	i = pPool->iUserFreeHead;
2319	}
2320
2321	/*
2322	* Unlink the user node from the free list,
2323	* initialize and insert it into the user list.
2324	*/
2325	pPool->iUserFreeHead = paUsers[i].iNext;
2326	paUsers[i].iNext = NIL_PGMPOOL_USER_INDEX;
2327	paUsers[i].iUser = iUser;
2328	paUsers[i].iUserTable = iUserTable;
2329	pPage->iUserHead = i;
2330
2331	/*
2332	* Insert into cache and enable monitoring of the guest page if enabled.
2333	*
2334	* Until we implement caching of all levels, including the CR3 one, we'll
2335	* have to make sure we don't try monitor & cache any recursive reuse of
2336	* a monitored CR3 page. Because all windows versions are doing this we'll
2337	* have to be able to do combined access monitoring, CR3 + PT and
2338	* PD + PT (guest PAE).
2339	*
2340	* Update:
2341	* We're now cooperating with the CR3 monitor if an uncachable page is found.
2342	*/
2343	#if defined(PGMPOOL_WITH_MONITORING) \|\| defined(PGMPOOL_WITH_CACHE)
2344	# ifdef PGMPOOL_WITH_MIXED_PT_CR3
2345	const bool fCanBeMonitored = true;
2346	# else
2347	bool fCanBeMonitored = pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored == NIL_RTGCPHYS
2348	\|\| (GCPhys & X86_PTE_PAE_PG_MASK) != (pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored & X86_PTE_PAE_PG_MASK)
2349	\|\| pgmPoolIsBigPage((PGMPOOLKIND)pPage->enmKind);
2350	# endif
2351	# ifdef PGMPOOL_WITH_CACHE
2352	pgmPoolCacheInsert(pPool, pPage, fCanBeMonitored); /* This can be expanded. */
2353	# endif
2354	if (fCanBeMonitored)
2355	{
2356	# ifdef PGMPOOL_WITH_MONITORING
2357	rc = pgmPoolMonitorInsert(pPool, pPage);
2358	AssertRC(rc);
2359	}
2360	# endif
2361	#endif /* PGMPOOL_WITH_MONITORING */
2362	return rc;
2363	}
2364
2365
2366	# ifdef PGMPOOL_WITH_CACHE /* (only used when the cache is enabled.) */
2367	/**
2368	* Adds a user reference to a page.
2369	*
2370	* This will move the page to the head of the
2371	*
2372	* @returns VBox status code.
2373	* @retval VINF_SUCCESS if successfully added.
2374	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2375	* @param pPool The pool.
2376	* @param pPage The cached page.
2377	* @param iUser The user index.
2378	* @param iUserTable The user table.
2379	*/
2380	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2381	{
2382	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2383
2384	Log3(("pgmPoolTrackAddUser GCPhys = %RGp iUser %x iUserTable %x\n", pPage->GCPhys, iUser, iUserTable));
2385
2386	# ifdef VBOX_STRICT
2387	/*
2388	* Check that the entry doesn't already exists. We only allow multiple users of top-level paging structures (SHW_POOL_ROOT_IDX).
2389	*/
2390	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2391	{
2392	uint16_t i = pPage->iUserHead;
2393	do
2394	{
2395	Assert(i < pPool->cMaxUsers);
2396	AssertMsg(iUser != PGMPOOL_IDX_PD \|\| iUser != PGMPOOL_IDX_PDPT \|\| iUser != PGMPOOL_IDX_NESTED_ROOT \|\| iUser != PGMPOOL_IDX_AMD64_CR3 \|\|
2397	paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2398	i = paUsers[i].iNext;
2399	} while (i != NIL_PGMPOOL_USER_INDEX);
2400	}
2401	# endif
2402
2403	/*
2404	* Allocate a user node.
2405	*/
2406	uint16_t i = pPool->iUserFreeHead;
2407	if (i == NIL_PGMPOOL_USER_INDEX)
2408	{
2409	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2410	if (RT_FAILURE(rc))
2411	return rc;
2412	i = pPool->iUserFreeHead;
2413	}
2414	pPool->iUserFreeHead = paUsers[i].iNext;
2415
2416	/*
2417	* Initialize the user node and insert it.
2418	*/
2419	paUsers[i].iNext = pPage->iUserHead;
2420	paUsers[i].iUser = iUser;
2421	paUsers[i].iUserTable = iUserTable;
2422	pPage->iUserHead = i;
2423
2424	# ifdef PGMPOOL_WITH_CACHE
2425	/*
2426	* Tell the cache to update its replacement stats for this page.
2427	*/
2428	pgmPoolCacheUsed(pPool, pPage);
2429	# endif
2430	return VINF_SUCCESS;
2431	}
2432	# endif /* PGMPOOL_WITH_CACHE */
2433
2434
2435	/**
2436	* Frees a user record associated with a page.
2437	*
2438	* This does not clear the entry in the user table, it simply replaces the
2439	* user record to the chain of free records.
2440	*
2441	* @param pPool The pool.
2442	* @param HCPhys The HC physical address of the shadow page.
2443	* @param iUser The shadow page pool index of the user table.
2444	* @param iUserTable The index into the user table (shadowed).
2445	*/
2446	static void pgmPoolTrackFreeUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2447	{
2448	/*
2449	* Unlink and free the specified user entry.
2450	*/
2451	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2452
2453	Log3(("pgmPoolTrackFreeUser %RGp %x %x\n", pPage->GCPhys, iUser, iUserTable));
2454	/* Special: For PAE and 32-bit paging, there is usually no more than one user. */
2455	uint16_t i = pPage->iUserHead;
2456	if ( i != NIL_PGMPOOL_USER_INDEX
2457	&& paUsers[i].iUser == iUser
2458	&& paUsers[i].iUserTable == iUserTable)
2459	{
2460	pPage->iUserHead = paUsers[i].iNext;
2461
2462	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2463	paUsers[i].iNext = pPool->iUserFreeHead;
2464	pPool->iUserFreeHead = i;
2465	return;
2466	}
2467
2468	/* General: Linear search. */
2469	uint16_t iPrev = NIL_PGMPOOL_USER_INDEX;
2470	while (i != NIL_PGMPOOL_USER_INDEX)
2471	{
2472	if ( paUsers[i].iUser == iUser
2473	&& paUsers[i].iUserTable == iUserTable)
2474	{
2475	if (iPrev != NIL_PGMPOOL_USER_INDEX)
2476	paUsers[iPrev].iNext = paUsers[i].iNext;
2477	else
2478	pPage->iUserHead = paUsers[i].iNext;
2479
2480	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2481	paUsers[i].iNext = pPool->iUserFreeHead;
2482	pPool->iUserFreeHead = i;
2483	return;
2484	}
2485	iPrev = i;
2486	i = paUsers[i].iNext;
2487	}
2488
2489	/* Fatal: didn't find it */
2490	AssertFatalMsgFailed(("Didn't find the user entry! iUser=%#x iUserTable=%#x GCPhys=%RGp\n",
2491	iUser, iUserTable, pPage->GCPhys));
2492	}
2493
2494
2495	/**
2496	* Gets the entry size of a shadow table.
2497	*
2498	* @param enmKind The kind of page.
2499	*
2500	* @returns The size of the entry in bytes. That is, 4 or 8.
2501	* @returns If the kind is not for a table, an assertion is raised and 0 is
2502	* returned.
2503	*/
2504	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind)
2505	{
2506	switch (enmKind)
2507	{
2508	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2509	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2510	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2511	case PGMPOOLKIND_32BIT_PD:
2512	case PGMPOOLKIND_32BIT_PD_PHYS:
2513	return 4;
2514
2515	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2516	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2517	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2518	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2519	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2520	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2521	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2522	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2523	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2524	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2525	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2526	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2527	case PGMPOOLKIND_64BIT_PML4:
2528	case PGMPOOLKIND_PAE_PDPT:
2529	case PGMPOOLKIND_ROOT_NESTED:
2530	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2531	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2532	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2533	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2534	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2535	case PGMPOOLKIND_PAE_PD_PHYS:
2536	case PGMPOOLKIND_PAE_PDPT_PHYS:
2537	return 8;
2538
2539	default:
2540	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2541	}
2542	}
2543
2544
2545	/**
2546	* Gets the entry size of a guest table.
2547	*
2548	* @param enmKind The kind of page.
2549	*
2550	* @returns The size of the entry in bytes. That is, 0, 4 or 8.
2551	* @returns If the kind is not for a table, an assertion is raised and 0 is
2552	* returned.
2553	*/
2554	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind)
2555	{
2556	switch (enmKind)
2557	{
2558	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2559	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2560	case PGMPOOLKIND_32BIT_PD:
2561	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2562	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2563	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2564	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2565	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2566	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2567	return 4;
2568
2569	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2570	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2571	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2572	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2573	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2574	case PGMPOOLKIND_64BIT_PML4:
2575	case PGMPOOLKIND_PAE_PDPT:
2576	return 8;
2577
2578	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2579	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2580	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2581	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2582	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2583	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2584	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2585	case PGMPOOLKIND_ROOT_NESTED:
2586	case PGMPOOLKIND_PAE_PD_PHYS:
2587	case PGMPOOLKIND_PAE_PDPT_PHYS:
2588	case PGMPOOLKIND_32BIT_PD_PHYS:
2589	/** @todo can we return 0? (nobody is calling this...) */
2590	AssertFailed();
2591	return 0;
2592
2593	default:
2594	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2595	}
2596	}
2597
2598	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2599
2600	/**
2601	* Scans one shadow page table for mappings of a physical page.
2602	*
2603	* @param pVM The VM handle.
2604	* @param pPhysPage The guest page in question.
2605	* @param iShw The shadow page table.
2606	* @param cRefs The number of references made in that PT.
2607	*/
2608	static void pgmPoolTrackFlushGCPhysPTInt(PVM pVM, PCPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2609	{
2610	LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%R[pgmpage] iShw=%d cRefs=%d\n", pPhysPage, iShw, cRefs));
2611	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2612
2613	/*
2614	* Assert sanity.
2615	*/
2616	Assert(cRefs == 1);
2617	AssertFatalMsg(iShw < pPool->cCurPages && iShw != NIL_PGMPOOL_IDX, ("iShw=%d\n", iShw));
2618	PPGMPOOLPAGE pPage = &pPool->aPages[iShw];
2619
2620	/*
2621	* Then, clear the actual mappings to the page in the shadow PT.
2622	*/
2623	switch (pPage->enmKind)
2624	{
2625	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2626	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2627	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2628	{
2629	const uint32_t u32 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2630	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2631	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2632	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2633	{
2634	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX32 cRefs=%#x\n", i, pPT->a[i], cRefs));
2635	pPT->a[i].u = 0;
2636	cRefs--;
2637	if (!cRefs)
2638	return;
2639	}
2640	#ifdef LOG_ENABLED
2641	Log(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2642	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2643	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2644	{
2645	Log(("i=%d cRefs=%d\n", i, cRefs--));
2646	}
2647	#endif
2648	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2649	break;
2650	}
2651
2652	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2653	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2654	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2655	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2656	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2657	{
2658	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2659	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2660	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2661	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2662	{
2663	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2664	pPT->a[i].u = 0;
2665	cRefs--;
2666	if (!cRefs)
2667	return;
2668	}
2669	#ifdef LOG_ENABLED
2670	Log(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2671	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2672	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2673	{
2674	Log(("i=%d cRefs=%d\n", i, cRefs--));
2675	}
2676	#endif
2677	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d u64=%RX64\n", cRefs, pPage->iFirstPresent, pPage->cPresent, u64));
2678	break;
2679	}
2680
2681	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2682	{
2683	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2684	PEPTPT pPT = (PEPTPT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2685	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2686	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2687	{
2688	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2689	pPT->a[i].u = 0;
2690	cRefs--;
2691	if (!cRefs)
2692	return;
2693	}
2694	#ifdef LOG_ENABLED
2695	Log(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2696	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2697	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2698	{
2699	Log(("i=%d cRefs=%d\n", i, cRefs--));
2700	}
2701	#endif
2702	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2703	break;
2704	}
2705
2706	default:
2707	AssertFatalMsgFailed(("enmKind=%d iShw=%d\n", pPage->enmKind, iShw));
2708	}
2709	}
2710
2711
2712	/**
2713	* Scans one shadow page table for mappings of a physical page.
2714	*
2715	* @param pVM The VM handle.
2716	* @param pPhysPage The guest page in question.
2717	* @param iShw The shadow page table.
2718	* @param cRefs The number of references made in that PT.
2719	*/
2720	void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2721	{
2722	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
2723	LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%R[pgmpage] iShw=%d cRefs=%d\n", pPhysPage, iShw, cRefs));
2724	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPT, f);
2725	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, iShw, cRefs);
2726	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2727	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPT, f);
2728	}
2729
2730
2731	/**
2732	* Flushes a list of shadow page tables mapping the same physical page.
2733	*
2734	* @param pVM The VM handle.
2735	* @param pPhysPage The guest page in question.
2736	* @param iPhysExt The physical cross reference extent list to flush.
2737	*/
2738	void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, uint16_t iPhysExt)
2739	{
2740	Assert(PGMIsLockOwner(pVM));
2741	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2742	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTs, f);
2743	LogFlow(("pgmPoolTrackFlushGCPhysPTs: pPhysPage=%R[pgmpage] iPhysExt\n", pPhysPage, iPhysExt));
2744
2745	const uint16_t iPhysExtStart = iPhysExt;
2746	PPGMPOOLPHYSEXT pPhysExt;
2747	do
2748	{
2749	Assert(iPhysExt < pPool->cMaxPhysExts);
2750	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2751	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
2752	if (pPhysExt->aidx[i] != NIL_PGMPOOL_IDX)
2753	{
2754	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, pPhysExt->aidx[i], 1);
2755	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
2756	}
2757
2758	/* next */
2759	iPhysExt = pPhysExt->iNext;
2760	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
2761
2762	/* insert the list into the free list and clear the ram range entry. */
2763	pPhysExt->iNext = pPool->iPhysExtFreeHead;
2764	pPool->iPhysExtFreeHead = iPhysExtStart;
2765	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2766
2767	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTs, f);
2768	}
2769
2770	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
2771
2772	/**
2773	* Flushes all shadow page table mappings of the given guest page.
2774	*
2775	* This is typically called when the host page backing the guest one has been
2776	* replaced or when the page protection was changed due to an access handler.
2777	*
2778	* @returns VBox status code.
2779	* @retval VINF_SUCCESS if all references has been successfully cleared.
2780	* @retval VINF_PGM_SYNC_CR3 if we're better off with a CR3 sync and a page
2781	* pool cleaning. FF and sync flags are set.
2782	*
2783	* @param pVM The VM handle.
2784	* @param pPhysPage The guest page in question.
2785	* @param pfFlushTLBs This is set to @a true if the shadow TLBs should be
2786	* flushed, it is NOT touched if this isn't necessary.
2787	* The caller MUST initialized this to @a false.
2788	*/
2789	int pgmPoolTrackFlushGCPhys(PVM pVM, PPGMPAGE pPhysPage, bool *pfFlushTLBs)
2790	{
2791	PVMCPU pVCpu = VMMGetCpu(pVM);
2792	pgmLock(pVM);
2793	int rc = VINF_SUCCESS;
2794	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2795	const uint16_t u16 = PGM_PAGE_GET_TRACKING(pPhysPage);
2796	if (u16)
2797	{
2798	/*
2799	* The zero page is currently screwing up the tracking and we'll
2800	* have to flush the whole shebang. Unless VBOX_WITH_NEW_LAZY_PAGE_ALLOC
2801	* is defined, zero pages won't normally be mapped. Some kind of solution
2802	* will be needed for this problem of course, but it will have to wait...
2803	*/
2804	if (PGM_PAGE_IS_ZERO(pPhysPage))
2805	rc = VINF_PGM_GCPHYS_ALIASED;
2806	else
2807	{
2808	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2809	/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow and
2810	pgmPoolTrackFlushGCPhysPTs will/may kill the pool otherwise. */
2811	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
2812	# endif
2813
2814	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
2815	pgmPoolTrackFlushGCPhysPT(pVM,
2816	pPhysPage,
2817	PGMPOOL_TD_GET_IDX(u16),
2818	PGMPOOL_TD_GET_CREFS(u16));
2819	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
2820	pgmPoolTrackFlushGCPhysPTs(pVM, pPhysPage, PGMPOOL_TD_GET_IDX(u16));
2821	else
2822	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
2823	*pfFlushTLBs = true;
2824
2825	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2826	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
2827	# endif
2828	}
2829	}
2830
2831	#elif defined(PGMPOOL_WITH_CACHE)
2832	if (PGM_PAGE_IS_ZERO(pPhysPage))
2833	rc = VINF_PGM_GCPHYS_ALIASED;
2834	else
2835	{
2836	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2837	/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow kill the pool otherwise. */
2838	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
2839	# endif
2840	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
2841	if (rc == VINF_SUCCESS)
2842	*pfFlushTLBs = true;
2843	}
2844
2845	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2846	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
2847	# endif
2848
2849	#else
2850	rc = VINF_PGM_GCPHYS_ALIASED;
2851	#endif
2852
2853	if (rc == VINF_PGM_GCPHYS_ALIASED)
2854	{
2855	pVCpu->pgm.s.fSyncFlags \|= PGM_SYNC_CLEAR_PGM_POOL;
2856	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
2857	rc = VINF_PGM_SYNC_CR3;
2858	}
2859	pgmUnlock(pVM);
2860	return rc;
2861	}
2862
2863
2864	/**
2865	* Scans all shadow page tables for mappings of a physical page.
2866	*
2867	* This may be slow, but it's most likely more efficient than cleaning
2868	* out the entire page pool / cache.
2869	*
2870	* @returns VBox status code.
2871	* @retval VINF_SUCCESS if all references has been successfully cleared.
2872	* @retval VINF_PGM_GCPHYS_ALIASED if we're better off with a CR3 sync and
2873	* a page pool cleaning.
2874	*
2875	* @param pVM The VM handle.
2876	* @param pPhysPage The guest page in question.
2877	*/
2878	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage)
2879	{
2880	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2881	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2882	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: cUsedPages=%d cPresent=%d pPhysPage=%R[pgmpage]\n",
2883	pPool->cUsedPages, pPool->cPresent, pPhysPage));
2884
2885	#if 1
2886	/*
2887	* There is a limit to what makes sense.
2888	*/
2889	if (pPool->cPresent > 1024)
2890	{
2891	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
2892	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2893	return VINF_PGM_GCPHYS_ALIASED;
2894	}
2895	#endif
2896
2897	/*
2898	* Iterate all the pages until we've encountered all that in use.
2899	* This is simple but not quite optimal solution.
2900	*/
2901	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2902	const uint32_t u32 = u64;
2903	unsigned cLeft = pPool->cUsedPages;
2904	unsigned iPage = pPool->cCurPages;
2905	while (--iPage >= PGMPOOL_IDX_FIRST)
2906	{
2907	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2908	if (pPage->GCPhys != NIL_RTGCPHYS)
2909	{
2910	switch (pPage->enmKind)
2911	{
2912	/*
2913	* We only care about shadow page tables.
2914	*/
2915	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2916	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2917	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2918	{
2919	unsigned cPresent = pPage->cPresent;
2920	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2921	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2922	if (pPT->a[i].n.u1Present)
2923	{
2924	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2925	{
2926	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX32\n", iPage, i, pPT->a[i]));
2927	pPT->a[i].u = 0;
2928	}
2929	if (!--cPresent)
2930	break;
2931	}
2932	break;
2933	}
2934
2935	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2936	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2937	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2938	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2939	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2940	{
2941	unsigned cPresent = pPage->cPresent;
2942	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2943	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2944	if (pPT->a[i].n.u1Present)
2945	{
2946	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2947	{
2948	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
2949	pPT->a[i].u = 0;
2950	}
2951	if (!--cPresent)
2952	break;
2953	}
2954	break;
2955	}
2956	}
2957	if (!--cLeft)
2958	break;
2959	}
2960	}
2961
2962	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2963	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2964	return VINF_SUCCESS;
2965	}
2966
2967
2968	/**
2969	* Clears the user entry in a user table.
2970	*
2971	* This is used to remove all references to a page when flushing it.
2972	*/
2973	static void pgmPoolTrackClearPageUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PCPGMPOOLUSER pUser)
2974	{
2975	Assert(pUser->iUser != NIL_PGMPOOL_IDX);
2976	Assert(pUser->iUser < pPool->cCurPages);
2977	uint32_t iUserTable = pUser->iUserTable;
2978
2979	/*
2980	* Map the user page.
2981	*/
2982	PPGMPOOLPAGE pUserPage = &pPool->aPages[pUser->iUser];
2983	union
2984	{
2985	uint64_t *pau64;
2986	uint32_t *pau32;
2987	} u;
2988	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pUserPage);
2989
2990	LogFlow(("pgmPoolTrackClearPageUser: clear %x in %s (%RGp) (flushing %s)\n", iUserTable, pgmPoolPoolKindToStr(pUserPage->enmKind), pUserPage->Core.Key, pgmPoolPoolKindToStr(pPage->enmKind)));
2991
2992	/* Safety precaution in case we change the paging for other modes too in the future. */
2993	Assert(!pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage));
2994
2995	#ifdef VBOX_STRICT
2996	/*
2997	* Some sanity checks.
2998	*/
2999	switch (pUserPage->enmKind)
3000	{
3001	case PGMPOOLKIND_32BIT_PD:
3002	case PGMPOOLKIND_32BIT_PD_PHYS:
3003	Assert(iUserTable < X86_PG_ENTRIES);
3004	break;
3005	case PGMPOOLKIND_PAE_PDPT:
3006	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
3007	case PGMPOOLKIND_PAE_PDPT_PHYS:
3008	Assert(iUserTable < 4);
3009	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
3010	break;
3011	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3012	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3013	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3014	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3015	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3016	case PGMPOOLKIND_PAE_PD_PHYS:
3017	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3018	break;
3019	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3020	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3021	Assert(!(u.pau64[iUserTable] & PGM_PDFLAGS_MAPPING));
3022	break;
3023	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3024	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3025	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
3026	break;
3027	case PGMPOOLKIND_64BIT_PML4:
3028	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
3029	/* GCPhys >> PAGE_SHIFT is the index here */
3030	break;
3031	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3032	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3033	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3034	break;
3035
3036	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3037	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3038	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3039	break;
3040
3041	case PGMPOOLKIND_ROOT_NESTED:
3042	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3043	break;
3044
3045	default:
3046	AssertMsgFailed(("enmKind=%d\n", pUserPage->enmKind));
3047	break;
3048	}
3049	#endif /* VBOX_STRICT */
3050
3051	/*
3052	* Clear the entry in the user page.
3053	*/
3054	switch (pUserPage->enmKind)
3055	{
3056	/* 32-bit entries */
3057	case PGMPOOLKIND_32BIT_PD:
3058	case PGMPOOLKIND_32BIT_PD_PHYS:
3059	u.pau32[iUserTable] = 0;
3060	break;
3061
3062	/* 64-bit entries */
3063	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3064	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3065	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3066	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3067	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3068	#if defined(IN_RC)
3069	/* In 32 bits PAE mode we must invalidate the TLB when changing a PDPT entry; the CPU fetches them only during cr3 load, so any
3070	* non-present PDPT will continue to cause page faults.
3071	*/
3072	ASMReloadCR3();
3073	#endif
3074	/* no break */
3075	case PGMPOOLKIND_PAE_PD_PHYS:
3076	case PGMPOOLKIND_PAE_PDPT_PHYS:
3077	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3078	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3079	case PGMPOOLKIND_64BIT_PML4:
3080	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3081	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3082	case PGMPOOLKIND_PAE_PDPT:
3083	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
3084	case PGMPOOLKIND_ROOT_NESTED:
3085	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3086	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3087	u.pau64[iUserTable] = 0;
3088	break;
3089
3090	default:
3091	AssertFatalMsgFailed(("enmKind=%d iUser=%#x iUserTable=%#x\n", pUserPage->enmKind, pUser->iUser, pUser->iUserTable));
3092	}
3093	}
3094
3095
3096	/**
3097	* Clears all users of a page.
3098	*/
3099	static void pgmPoolTrackClearPageUsers(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3100	{
3101	/*
3102	* Free all the user records.
3103	*/
3104	LogFlow(("pgmPoolTrackClearPageUsers %RGp\n", pPage->GCPhys));
3105
3106	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
3107	uint16_t i = pPage->iUserHead;
3108	while (i != NIL_PGMPOOL_USER_INDEX)
3109	{
3110	/* Clear enter in user table. */
3111	pgmPoolTrackClearPageUser(pPool, pPage, &paUsers[i]);
3112
3113	/* Free it. */
3114	const uint16_t iNext = paUsers[i].iNext;
3115	paUsers[i].iUser = NIL_PGMPOOL_IDX;
3116	paUsers[i].iNext = pPool->iUserFreeHead;
3117	pPool->iUserFreeHead = i;
3118
3119	/* Next. */
3120	i = iNext;
3121	}
3122	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
3123	}
3124
3125	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3126
3127	/**
3128	* Allocates a new physical cross reference extent.
3129	*
3130	* @returns Pointer to the allocated extent on success. NULL if we're out of them.
3131	* @param pVM The VM handle.
3132	* @param piPhysExt Where to store the phys ext index.
3133	*/
3134	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt)
3135	{
3136	Assert(PGMIsLockOwner(pVM));
3137	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3138	uint16_t iPhysExt = pPool->iPhysExtFreeHead;
3139	if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
3140	{
3141	STAM_COUNTER_INC(&pPool->StamTrackPhysExtAllocFailures);
3142	return NULL;
3143	}
3144	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3145	pPool->iPhysExtFreeHead = pPhysExt->iNext;
3146	pPhysExt->iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
3147	*piPhysExt = iPhysExt;
3148	return pPhysExt;
3149	}
3150
3151
3152	/**
3153	* Frees a physical cross reference extent.
3154	*
3155	* @param pVM The VM handle.
3156	* @param iPhysExt The extent to free.
3157	*/
3158	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt)
3159	{
3160	Assert(PGMIsLockOwner(pVM));
3161	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3162	Assert(iPhysExt < pPool->cMaxPhysExts);
3163	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3164	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3165	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3166	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3167	pPool->iPhysExtFreeHead = iPhysExt;
3168	}
3169
3170
3171	/**
3172	* Frees a physical cross reference extent.
3173	*
3174	* @param pVM The VM handle.
3175	* @param iPhysExt The extent to free.
3176	*/
3177	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt)
3178	{
3179	Assert(PGMIsLockOwner(pVM));
3180	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3181
3182	const uint16_t iPhysExtStart = iPhysExt;
3183	PPGMPOOLPHYSEXT pPhysExt;
3184	do
3185	{
3186	Assert(iPhysExt < pPool->cMaxPhysExts);
3187	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3188	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3189	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3190
3191	/* next */
3192	iPhysExt = pPhysExt->iNext;
3193	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3194
3195	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3196	pPool->iPhysExtFreeHead = iPhysExtStart;
3197	}
3198
3199
3200	/**
3201	* Insert a reference into a list of physical cross reference extents.
3202	*
3203	* @returns The new tracking data for PGMPAGE.
3204	*
3205	* @param pVM The VM handle.
3206	* @param iPhysExt The physical extent index of the list head.
3207	* @param iShwPT The shadow page table index.
3208	*
3209	*/
3210	static uint16_t pgmPoolTrackPhysExtInsert(PVM pVM, uint16_t iPhysExt, uint16_t iShwPT)
3211	{
3212	Assert(PGMIsLockOwner(pVM));
3213	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3214	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3215
3216	/* special common case. */
3217	if (paPhysExts[iPhysExt].aidx[2] == NIL_PGMPOOL_IDX)
3218	{
3219	paPhysExts[iPhysExt].aidx[2] = iShwPT;
3220	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
3221	LogFlow(("pgmPoolTrackPhysExtInsert: %d:{,,%d}\n", iPhysExt, iShwPT));
3222	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3223	}
3224
3225	/* general treatment. */
3226	const uint16_t iPhysExtStart = iPhysExt;
3227	unsigned cMax = 15;
3228	for (;;)
3229	{
3230	Assert(iPhysExt < pPool->cMaxPhysExts);
3231	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3232	if (paPhysExts[iPhysExt].aidx[i] == NIL_PGMPOOL_IDX)
3233	{
3234	paPhysExts[iPhysExt].aidx[i] = iShwPT;
3235	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
3236	LogFlow(("pgmPoolTrackPhysExtInsert: %d:{%d} i=%d cMax=%d\n", iPhysExt, iShwPT, i, cMax));
3237	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtStart);
3238	}
3239	if (!--cMax)
3240	{
3241	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
3242	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
3243	LogFlow(("pgmPoolTrackPhysExtInsert: overflow (1) iShwPT=%d\n", iShwPT));
3244	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3245	}
3246	}
3247
3248	/* add another extent to the list. */
3249	PPGMPOOLPHYSEXT pNew = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3250	if (!pNew)
3251	{
3252	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
3253	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
3254	LogFlow(("pgmPoolTrackPhysExtInsert: pgmPoolTrackPhysExtAlloc failed iShwPT=%d\n", iShwPT));
3255	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3256	}
3257	pNew->iNext = iPhysExtStart;
3258	pNew->aidx[0] = iShwPT;
3259	LogFlow(("pgmPoolTrackPhysExtInsert: added new extent %d:{%d}->%d\n", iPhysExt, iShwPT, iPhysExtStart));
3260	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3261	}
3262
3263
3264	/**
3265	* Add a reference to guest physical page where extents are in use.
3266	*
3267	* @returns The new tracking data for PGMPAGE.
3268	*
3269	* @param pVM The VM handle.
3270	* @param u16 The ram range flags (top 16-bits).
3271	* @param iShwPT The shadow page table index.
3272	*/
3273	uint16_t pgmPoolTrackPhysExtAddref(PVM pVM, uint16_t u16, uint16_t iShwPT)
3274	{
3275	pgmLock(pVM);
3276	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
3277	{
3278	/*
3279	* Convert to extent list.
3280	*/
3281	Assert(PGMPOOL_TD_GET_CREFS(u16) == 1);
3282	uint16_t iPhysExt;
3283	PPGMPOOLPHYSEXT pPhysExt = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3284	if (pPhysExt)
3285	{
3286	LogFlow(("pgmPoolTrackPhysExtAddref: new extent: %d:{%d, %d}\n", iPhysExt, PGMPOOL_TD_GET_IDX(u16), iShwPT));
3287	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliased);
3288	pPhysExt->aidx[0] = PGMPOOL_TD_GET_IDX(u16);
3289	pPhysExt->aidx[1] = iShwPT;
3290	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3291	}
3292	else
3293	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3294	}
3295	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
3296	{
3297	/*
3298	* Insert into the extent list.
3299	*/
3300	u16 = pgmPoolTrackPhysExtInsert(pVM, PGMPOOL_TD_GET_IDX(u16), iShwPT);
3301	}
3302	else
3303	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedLots);
3304	pgmUnlock(pVM);
3305	return u16;
3306	}
3307
3308
3309	/**
3310	* Clear references to guest physical memory.
3311	*
3312	* @param pPool The pool.
3313	* @param pPage The page.
3314	* @param pPhysPage Pointer to the aPages entry in the ram range.
3315	*/
3316	void pgmPoolTrackPhysExtDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMPAGE pPhysPage)
3317	{
3318	const unsigned cRefs = PGM_PAGE_GET_TD_CREFS(pPhysPage);
3319	AssertFatalMsg(cRefs == PGMPOOL_TD_CREFS_PHYSEXT, ("cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
3320
3321	uint16_t iPhysExt = PGM_PAGE_GET_TD_IDX(pPhysPage);
3322	if (iPhysExt != PGMPOOL_TD_IDX_OVERFLOWED)
3323	{
3324	PVM pVM = pPool->CTX_SUFF(pVM);
3325	pgmLock(pVM);
3326
3327	uint16_t iPhysExtPrev = NIL_PGMPOOL_PHYSEXT_INDEX;
3328	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3329	do
3330	{
3331	Assert(iPhysExt < pPool->cMaxPhysExts);
3332
3333	/*
3334	* Look for the shadow page and check if it's all freed.
3335	*/
3336	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3337	{
3338	if (paPhysExts[iPhysExt].aidx[i] == pPage->idx)
3339	{
3340	paPhysExts[iPhysExt].aidx[i] = NIL_PGMPOOL_IDX;
3341
3342	for (i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3343	if (paPhysExts[iPhysExt].aidx[i] != NIL_PGMPOOL_IDX)
3344	{
3345	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
3346	pgmUnlock(pVM);
3347	return;
3348	}
3349
3350	/* we can free the node. */
3351	const uint16_t iPhysExtNext = paPhysExts[iPhysExt].iNext;
3352	if ( iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX
3353	&& iPhysExtNext == NIL_PGMPOOL_PHYSEXT_INDEX)
3354	{
3355	/* lonely node */
3356	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3357	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d lonely\n", pPhysPage, pPage->idx));
3358	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
3359	}
3360	else if (iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX)
3361	{
3362	/* head */
3363	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d head\n", pPhysPage, pPage->idx));
3364	PGM_PAGE_SET_TRACKING(pPhysPage, PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtNext));
3365	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3366	}
3367	else
3368	{
3369	/* in list */
3370	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
3371	paPhysExts[iPhysExtPrev].iNext = iPhysExtNext;
3372	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3373	}
3374	iPhysExt = iPhysExtNext;
3375	pgmUnlock(pVM);
3376	return;
3377	}
3378	}
3379
3380	/* next */
3381	iPhysExtPrev = iPhysExt;
3382	iPhysExt = paPhysExts[iPhysExt].iNext;
3383	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3384
3385	pgmUnlock(pVM);
3386	AssertFatalMsgFailed(("not-found! cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
3387	}
3388	else /* nothing to do */
3389	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage]\n", pPhysPage));
3390	}
3391
3392
3393	/**
3394	* Clear references to guest physical memory.
3395	*
3396	* This is the same as pgmPoolTracDerefGCPhys except that the guest physical address
3397	* is assumed to be correct, so the linear search can be skipped and we can assert
3398	* at an earlier point.
3399	*
3400	* @param pPool The pool.
3401	* @param pPage The page.
3402	* @param HCPhys The host physical address corresponding to the guest page.
3403	* @param GCPhys The guest physical address corresponding to HCPhys.
3404	*/
3405	static void pgmPoolTracDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhys)
3406	{
3407	/*
3408	* Walk range list.
3409	*/
3410	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3411	while (pRam)
3412	{
3413	RTGCPHYS off = GCPhys - pRam->GCPhys;
3414	if (off < pRam->cb)
3415	{
3416	/* does it match? */
3417	const unsigned iPage = off >> PAGE_SHIFT;
3418	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3419	#ifdef LOG_ENABLED
3420	RTHCPHYS HCPhysPage = PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]);
3421	Log2(("pgmPoolTracDerefGCPhys %RHp vs %RHp\n", HCPhysPage, HCPhys));
3422	#endif
3423	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3424	{
3425	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3426	return;
3427	}
3428	break;
3429	}
3430	pRam = pRam->CTX_SUFF(pNext);
3431	}
3432	AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp\n", HCPhys, GCPhys));
3433	}
3434
3435
3436	/**
3437	* Clear references to guest physical memory.
3438	*
3439	* @param pPool The pool.
3440	* @param pPage The page.
3441	* @param HCPhys The host physical address corresponding to the guest page.
3442	* @param GCPhysHint The guest physical address which may corresponding to HCPhys.
3443	*/
3444	static void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint)
3445	{
3446	Log4(("pgmPoolTracDerefGCPhysHint %RHp %RGp\n", HCPhys, GCPhysHint));
3447
3448	/*
3449	* Walk range list.
3450	*/
3451	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3452	while (pRam)
3453	{
3454	RTGCPHYS off = GCPhysHint - pRam->GCPhys;
3455	if (off < pRam->cb)
3456	{
3457	/* does it match? */
3458	const unsigned iPage = off >> PAGE_SHIFT;
3459	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3460	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3461	{
3462	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3463	return;
3464	}
3465	break;
3466	}
3467	pRam = pRam->CTX_SUFF(pNext);
3468	}
3469
3470	/*
3471	* Damn, the hint didn't work. We'll have to do an expensive linear search.
3472	*/
3473	STAM_COUNTER_INC(&pPool->StatTrackLinearRamSearches);
3474	pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3475	while (pRam)
3476	{
3477	unsigned iPage = pRam->cb >> PAGE_SHIFT;
3478	while (iPage-- > 0)
3479	{
3480	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3481	{
3482	Log4(("pgmPoolTracDerefGCPhysHint: Linear HCPhys=%RHp GCPhysHint=%RGp GCPhysReal=%RGp\n",
3483	HCPhys, GCPhysHint, pRam->GCPhys + (iPage << PAGE_SHIFT)));
3484	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3485	return;
3486	}
3487	}
3488	pRam = pRam->CTX_SUFF(pNext);
3489	}
3490
3491	AssertFatalMsgFailed(("HCPhys=%RHp GCPhysHint=%RGp\n", HCPhys, GCPhysHint));
3492	}
3493
3494
3495	/**
3496	* Clear references to guest physical memory in a 32-bit / 32-bit page table.
3497	*
3498	* @param pPool The pool.
3499	* @param pPage The page.
3500	* @param pShwPT The shadow page table (mapping of the page).
3501	* @param pGstPT The guest page table.
3502	*/
3503	DECLINLINE(void) pgmPoolTrackDerefPT32Bit32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT, PCX86PT pGstPT)
3504	{
3505	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
3506	if (pShwPT->a[i].n.u1Present)
3507	{
3508	Log4(("pgmPoolTrackDerefPT32Bit32Bit: i=%d pte=%RX32 hint=%RX32\n",
3509	i, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3510	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3511	if (!--pPage->cPresent)
3512	break;
3513	}
3514	}
3515
3516
3517	/**
3518	* Clear references to guest physical memory in a PAE / 32-bit page table.
3519	*
3520	* @param pPool The pool.
3521	* @param pPage The page.
3522	* @param pShwPT The shadow page table (mapping of the page).
3523	* @param pGstPT The guest page table (just a half one).
3524	*/
3525	DECLINLINE(void) pgmPoolTrackDerefPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PT pGstPT)
3526	{
3527	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3528	if (pShwPT->a[i].n.u1Present)
3529	{
3530	Log4(("pgmPoolTrackDerefPTPae32Bit: i=%d pte=%RX64 hint=%RX32\n",
3531	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3532	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3533	}
3534	}
3535
3536
3537	/**
3538	* Clear references to guest physical memory in a PAE / PAE page table.
3539	*
3540	* @param pPool The pool.
3541	* @param pPage The page.
3542	* @param pShwPT The shadow page table (mapping of the page).
3543	* @param pGstPT The guest page table.
3544	*/
3545	DECLINLINE(void) pgmPoolTrackDerefPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PTPAE pGstPT)
3546	{
3547	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3548	if (pShwPT->a[i].n.u1Present)
3549	{
3550	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX32 hint=%RX32\n",
3551	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
3552	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK);
3553	}
3554	}
3555
3556
3557	/**
3558	* Clear references to guest physical memory in a 32-bit / 4MB page table.
3559	*
3560	* @param pPool The pool.
3561	* @param pPage The page.
3562	* @param pShwPT The shadow page table (mapping of the page).
3563	*/
3564	DECLINLINE(void) pgmPoolTrackDerefPT32Bit4MB(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT)
3565	{
3566	RTGCPHYS GCPhys = pPage->GCPhys;
3567	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3568	if (pShwPT->a[i].n.u1Present)
3569	{
3570	Log4(("pgmPoolTrackDerefPT32Bit4MB: i=%d pte=%RX32 GCPhys=%RGp\n",
3571	i, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys));
3572	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys);
3573	}
3574	}
3575
3576
3577	/**
3578	* Clear references to guest physical memory in a PAE / 2/4MB page table.
3579	*
3580	* @param pPool The pool.
3581	* @param pPage The page.
3582	* @param pShwPT The shadow page table (mapping of the page).
3583	*/
3584	DECLINLINE(void) pgmPoolTrackDerefPTPaeBig(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT)
3585	{
3586	RTGCPHYS GCPhys = pPage->GCPhys;
3587	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3588	if (pShwPT->a[i].n.u1Present)
3589	{
3590	Log4(("pgmPoolTrackDerefPTPaeBig: i=%d pte=%RX64 hint=%RGp\n",
3591	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys));
3592	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys);
3593	}
3594	}
3595
3596	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
3597
3598
3599	/**
3600	* Clear references to shadowed pages in a 32 bits page directory.
3601	*
3602	* @param pPool The pool.
3603	* @param pPage The page.
3604	* @param pShwPD The shadow page directory (mapping of the page).
3605	*/
3606	DECLINLINE(void) pgmPoolTrackDerefPD(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PD pShwPD)
3607	{
3608	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3609	{
3610	if ( pShwPD->a[i].n.u1Present
3611	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
3612	)
3613	{
3614	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PG_MASK);
3615	if (pSubPage)
3616	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3617	else
3618	AssertFatalMsgFailed(("%x\n", pShwPD->a[i].u & X86_PDE_PG_MASK));
3619	}
3620	}
3621	}
3622
3623	/**
3624	* Clear references to shadowed pages in a PAE (legacy or 64 bits) page directory.
3625	*
3626	* @param pPool The pool.
3627	* @param pPage The page.
3628	* @param pShwPD The shadow page directory (mapping of the page).
3629	*/
3630	DECLINLINE(void) pgmPoolTrackDerefPDPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPAE pShwPD)
3631	{
3632	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3633	{
3634	if ( pShwPD->a[i].n.u1Present
3635	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
3636	)
3637	{
3638	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PAE_PG_MASK);
3639	if (pSubPage)
3640	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3641	else
3642	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & X86_PDE_PAE_PG_MASK));
3643	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3644	}
3645	}
3646	}
3647
3648	/**
3649	* Clear references to shadowed pages in a PAE page directory pointer table.
3650	*
3651	* @param pPool The pool.
3652	* @param pPage The page.
3653	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3654	*/
3655	DECLINLINE(void) pgmPoolTrackDerefPDPTPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
3656	{
3657	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
3658	{
3659	if ( pShwPDPT->a[i].n.u1Present
3660	&& !(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING)
3661	)
3662	{
3663	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
3664	if (pSubPage)
3665	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3666	else
3667	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
3668	}
3669	}
3670	}
3671
3672
3673	/**
3674	* Clear references to shadowed pages in a 64-bit page directory pointer table.
3675	*
3676	* @param pPool The pool.
3677	* @param pPage The page.
3678	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3679	*/
3680	DECLINLINE(void) pgmPoolTrackDerefPDPT64Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
3681	{
3682	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3683	{
3684	Assert(!(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING));
3685	if (pShwPDPT->a[i].n.u1Present)
3686	{
3687	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
3688	if (pSubPage)
3689	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3690	else
3691	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
3692	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3693	}
3694	}
3695	}
3696
3697
3698	/**
3699	* Clear references to shadowed pages in a 64-bit level 4 page table.
3700	*
3701	* @param pPool The pool.
3702	* @param pPage The page.
3703	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3704	*/
3705	DECLINLINE(void) pgmPoolTrackDerefPML464Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PML4 pShwPML4)
3706	{
3707	for (unsigned i = 0; i < RT_ELEMENTS(pShwPML4->a); i++)
3708	{
3709	if (pShwPML4->a[i].n.u1Present)
3710	{
3711	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPML4->a[i].u & X86_PDPE_PG_MASK);
3712	if (pSubPage)
3713	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3714	else
3715	AssertFatalMsgFailed(("%RX64\n", pShwPML4->a[i].u & X86_PML4E_PG_MASK));
3716	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3717	}
3718	}
3719	}
3720
3721
3722	/**
3723	* Clear references to shadowed pages in an EPT page table.
3724	*
3725	* @param pPool The pool.
3726	* @param pPage The page.
3727	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3728	*/
3729	DECLINLINE(void) pgmPoolTrackDerefPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT)
3730	{
3731	RTGCPHYS GCPhys = pPage->GCPhys;
3732	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3733	if (pShwPT->a[i].n.u1Present)
3734	{
3735	Log4(("pgmPoolTrackDerefPTEPT: i=%d pte=%RX64 GCPhys=%RX64\n",
3736	i, pShwPT->a[i].u & EPT_PTE_PG_MASK, pPage->GCPhys));
3737	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & EPT_PTE_PG_MASK, GCPhys);
3738	}
3739	}
3740
3741
3742	/**
3743	* Clear references to shadowed pages in an EPT page directory.
3744	*
3745	* @param pPool The pool.
3746	* @param pPage The page.
3747	* @param pShwPD The shadow page directory (mapping of the page).
3748	*/
3749	DECLINLINE(void) pgmPoolTrackDerefPDEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPD pShwPD)
3750	{
3751	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3752	{
3753	if (pShwPD->a[i].n.u1Present)
3754	{
3755	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & EPT_PDE_PG_MASK);
3756	if (pSubPage)
3757	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3758	else
3759	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & EPT_PDE_PG_MASK));
3760	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3761	}
3762	}
3763	}
3764
3765
3766	/**
3767	* Clear references to shadowed pages in an EPT page directory pointer table.
3768	*
3769	* @param pPool The pool.
3770	* @param pPage The page.
3771	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3772	*/
3773	DECLINLINE(void) pgmPoolTrackDerefPDPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPDPT pShwPDPT)
3774	{
3775	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3776	{
3777	if (pShwPDPT->a[i].n.u1Present)
3778	{
3779	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK);
3780	if (pSubPage)
3781	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3782	else
3783	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK));
3784	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3785	}
3786	}
3787	}
3788
3789
3790	/**
3791	* Clears all references made by this page.
3792	*
3793	* This includes other shadow pages and GC physical addresses.
3794	*
3795	* @param pPool The pool.
3796	* @param pPage The page.
3797	*/
3798	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3799	{
3800	/*
3801	* Map the shadow page and take action according to the page kind.
3802	*/
3803	void *pvShw = PGMPOOL_PAGE_2_LOCKED_PTR(pPool->CTX_SUFF(pVM), pPage);
3804	switch (pPage->enmKind)
3805	{
3806	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3807	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3808	{
3809	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3810	void *pvGst;
3811	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3812	pgmPoolTrackDerefPT32Bit32Bit(pPool, pPage, (PX86PT)pvShw, (PCX86PT)pvGst);
3813	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3814	break;
3815	}
3816
3817	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3818	{
3819	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3820	void *pvGst;
3821	int rc = PGM_GCPHYS_2_PTR_EX(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3822	pgmPoolTrackDerefPTPae32Bit(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PT)pvGst);
3823	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3824	break;
3825	}
3826
3827	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3828	{
3829	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3830	void *pvGst;
3831	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3832	pgmPoolTrackDerefPTPaePae(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PTPAE)pvGst);
3833	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3834	break;
3835	}
3836
3837	case PGMPOOLKIND_32BIT_PT_FOR_PHYS: /* treat it like a 4 MB page */
3838	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3839	{
3840	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3841	pgmPoolTrackDerefPT32Bit4MB(pPool, pPage, (PX86PT)pvShw);
3842	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3843	break;
3844	}
3845
3846	case PGMPOOLKIND_PAE_PT_FOR_PHYS: /* treat it like a 2 MB page */
3847	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3848	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3849	{
3850	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3851	pgmPoolTrackDerefPTPaeBig(pPool, pPage, (PX86PTPAE)pvShw);
3852	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3853	break;
3854	}
3855
3856	#else /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3857	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3858	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3859	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3860	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3861	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3862	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3863	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3864	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3865	break;
3866	#endif /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3867
3868	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3869	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3870	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3871	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3872	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3873	case PGMPOOLKIND_PAE_PD_PHYS:
3874	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3875	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3876	pgmPoolTrackDerefPDPae(pPool, pPage, (PX86PDPAE)pvShw);
3877	break;
3878
3879	case PGMPOOLKIND_32BIT_PD_PHYS:
3880	case PGMPOOLKIND_32BIT_PD:
3881	pgmPoolTrackDerefPD(pPool, pPage, (PX86PD)pvShw);
3882	break;
3883
3884	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
3885	case PGMPOOLKIND_PAE_PDPT:
3886	case PGMPOOLKIND_PAE_PDPT_PHYS:
3887	pgmPoolTrackDerefPDPTPae(pPool, pPage, (PX86PDPT)pvShw);
3888	break;
3889
3890	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3891	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3892	pgmPoolTrackDerefPDPT64Bit(pPool, pPage, (PX86PDPT)pvShw);
3893	break;
3894
3895	case PGMPOOLKIND_64BIT_PML4:
3896	pgmPoolTrackDerefPML464Bit(pPool, pPage, (PX86PML4)pvShw);
3897	break;
3898
3899	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3900	pgmPoolTrackDerefPTEPT(pPool, pPage, (PEPTPT)pvShw);
3901	break;
3902
3903	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3904	pgmPoolTrackDerefPDEPT(pPool, pPage, (PEPTPD)pvShw);
3905	break;
3906
3907	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3908	pgmPoolTrackDerefPDPTEPT(pPool, pPage, (PEPTPDPT)pvShw);
3909	break;
3910
3911	default:
3912	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
3913	}
3914
3915	/* paranoia, clear the shadow page. Remove this laser (i.e. let Alloc and ClearAll do it). */
3916	STAM_PROFILE_START(&pPool->StatZeroPage, z);
3917	ASMMemZeroPage(pvShw);
3918	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
3919	pPage->fZeroed = true;
3920	PGMPOOL_UNLOCK_PTR(pPool->CTX_SUFF(pVM), pvShw);
3921	}
3922	#endif /* PGMPOOL_WITH_USER_TRACKING */
3923
3924	/**
3925	* Flushes a pool page.
3926	*
3927	* This moves the page to the free list after removing all user references to it.
3928	*
3929	* @returns VBox status code.
3930	* @retval VINF_SUCCESS on success.
3931	* @param pPool The pool.
3932	* @param HCPhys The HC physical address of the shadow page.
3933	*/
3934	int pgmPoolFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3935	{
3936	PVM pVM = pPool->CTX_SUFF(pVM);
3937
3938	int rc = VINF_SUCCESS;
3939	STAM_PROFILE_START(&pPool->StatFlushPage, f);
3940	LogFlow(("pgmPoolFlushPage: pPage=%p:{.Key=%RHp, .idx=%d, .enmKind=%s, .GCPhys=%RGp}\n",
3941	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
3942
3943	/*
3944	* Quietly reject any attempts at flushing any of the special root pages.
3945	*/
3946	if (pPage->idx < PGMPOOL_IDX_FIRST)
3947	{
3948	AssertFailed(); /* can no longer happen */
3949	Log(("pgmPoolFlushPage: special root page, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
3950	return VINF_SUCCESS;
3951	}
3952
3953	pgmLock(pVM);
3954
3955	/*
3956	* Quietly reject any attempts at flushing the currently active shadow CR3 mapping
3957	*/
3958	if (pgmPoolIsPageLocked(&pVM->pgm.s, pPage))
3959	{
3960	AssertMsg( pPage->enmKind == PGMPOOLKIND_64BIT_PML4
3961	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT
3962	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT_FOR_32BIT
3963	\|\| pPage->enmKind == PGMPOOLKIND_32BIT_PD
3964	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD
3965	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD
3966	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD
3967	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD
3968	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD,
3969	("Can't free the shadow CR3! (%RHp vs %RHp kind=%d\n", PGMGetHyperCR3(VMMGetCpu(pVM)), pPage->Core.Key, pPage->enmKind));
3970	Log(("pgmPoolFlushPage: current active shadow CR3, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
3971	pgmUnlock(pVM);
3972	return VINF_SUCCESS;
3973	}
3974
3975	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
3976	/* Start a subset so we won't run out of mapping space. */
3977	PVMCPU pVCpu = VMMGetCpu(pVM);
3978	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
3979	#endif
3980
3981	/*
3982	* Mark the page as being in need of a ASMMemZeroPage().
3983	*/
3984	pPage->fZeroed = false;
3985
3986	#ifdef PGMPOOL_WITH_USER_TRACKING
3987	/*
3988	* Clear the page.
3989	*/
3990	pgmPoolTrackClearPageUsers(pPool, pPage);
3991	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
3992	pgmPoolTrackDeref(pPool, pPage);
3993	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
3994	#endif
3995
3996	#ifdef PGMPOOL_WITH_CACHE
3997	/*
3998	* Flush it from the cache.
3999	*/
4000	pgmPoolCacheFlushPage(pPool, pPage);
4001	#endif /* PGMPOOL_WITH_CACHE */
4002
4003	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
4004	/* Heavy stuff done. */
4005	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
4006	#endif
4007
4008	#ifdef PGMPOOL_WITH_MONITORING
4009	/*
4010	* Deregistering the monitoring.
4011	*/
4012	if (pPage->fMonitored)
4013	rc = pgmPoolMonitorFlush(pPool, pPage);
4014	#endif
4015
4016	/*
4017	* Free the page.
4018	*/
4019	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
4020	pPage->iNext = pPool->iFreeHead;
4021	pPool->iFreeHead = pPage->idx;
4022	pPage->enmKind = PGMPOOLKIND_FREE;
4023	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
4024	pPage->GCPhys = NIL_RTGCPHYS;
4025	pPage->fReusedFlushPending = false;
4026
4027	pPool->cUsedPages--;
4028	pgmUnlock(pVM);
4029	STAM_PROFILE_STOP(&pPool->StatFlushPage, f);
4030	return rc;
4031	}
4032
4033
4034	/**
4035	* Frees a usage of a pool page.
4036	*
4037	* The caller is responsible to updating the user table so that it no longer
4038	* references the shadow page.
4039	*
4040	* @param pPool The pool.
4041	* @param HCPhys The HC physical address of the shadow page.
4042	* @param iUser The shadow page pool index of the user table.
4043	* @param iUserTable The index into the user table (shadowed).
4044	*/
4045	void pgmPoolFreeByPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
4046	{
4047	PVM pVM = pPool->CTX_SUFF(pVM);
4048
4049	STAM_PROFILE_START(&pPool->StatFree, a);
4050	LogFlow(("pgmPoolFreeByPage: pPage=%p:{.Key=%RHp, .idx=%d, enmKind=%s} iUser=%#x iUserTable=%#x\n",
4051	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), iUser, iUserTable));
4052	Assert(pPage->idx >= PGMPOOL_IDX_FIRST);
4053	pgmLock(pVM);
4054	#ifdef PGMPOOL_WITH_USER_TRACKING
4055	pgmPoolTrackFreeUser(pPool, pPage, iUser, iUserTable);
4056	#endif
4057	#ifdef PGMPOOL_WITH_CACHE
4058	if (!pPage->fCached)
4059	#endif
4060	pgmPoolFlushPage(pPool, pPage);
4061	pgmUnlock(pVM);
4062	STAM_PROFILE_STOP(&pPool->StatFree, a);
4063	}
4064
4065
4066	/**
4067	* Makes one or more free page free.
4068	*
4069	* @returns VBox status code.
4070	* @retval VINF_SUCCESS on success.
4071	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
4072	*
4073	* @param pPool The pool.
4074	* @param enmKind Page table kind
4075	* @param iUser The user of the page.
4076	*/
4077	static int pgmPoolMakeMoreFreePages(PPGMPOOL pPool, PGMPOOLKIND enmKind, uint16_t iUser)
4078	{
4079	PVM pVM = pPool->CTX_SUFF(pVM);
4080
4081	LogFlow(("pgmPoolMakeMoreFreePages: iUser=%#x\n", iUser));
4082
4083	/*
4084	* If the pool isn't full grown yet, expand it.
4085	*/
4086	if ( pPool->cCurPages < pPool->cMaxPages
4087	#if defined(IN_RC)
4088	/* Hack alert: we can't deal with jumps to ring 3 when called from MapCR3 and allocating pages for PAE PDs. */
4089	&& enmKind != PGMPOOLKIND_PAE_PD_FOR_PAE_PD
4090	&& (enmKind < PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD \|\| enmKind > PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD)
4091	#endif
4092	)
4093	{
4094	STAM_PROFILE_ADV_SUSPEND(&pPool->StatAlloc, a);
4095	#ifdef IN_RING3
4096	int rc = PGMR3PoolGrow(pVM);
4097	#else
4098	int rc = VMMRZCallRing3NoCpu(pVM, VMMCALLRING3_PGM_POOL_GROW, 0);
4099	#endif
4100	if (RT_FAILURE(rc))
4101	return rc;
4102	STAM_PROFILE_ADV_RESUME(&pPool->StatAlloc, a);
4103	if (pPool->iFreeHead != NIL_PGMPOOL_IDX)
4104	return VINF_SUCCESS;
4105	}
4106
4107	#ifdef PGMPOOL_WITH_CACHE
4108	/*
4109	* Free one cached page.
4110	*/
4111	return pgmPoolCacheFreeOne(pPool, iUser);
4112	#else
4113	/*
4114	* Flush the pool.
4115	*
4116	* If we have tracking enabled, it should be possible to come up with
4117	* a cheap replacement strategy...
4118	*/
4119	/* @todo This path no longer works (CR3 root pages will be flushed)!! */
4120	AssertCompileFailed();
4121	Assert(!CPUMIsGuestInLongMode(pVM));
4122	pgmPoolFlushAllInt(pPool);
4123	return VERR_PGM_POOL_FLUSHED;
4124	#endif
4125	}
4126
4127	/**
4128	* Allocates a page from the pool.
4129	*
4130	* This page may actually be a cached page and not in need of any processing
4131	* on the callers part.
4132	*
4133	* @returns VBox status code.
4134	* @retval VINF_SUCCESS if a NEW page was allocated.
4135	* @retval VINF_PGM_CACHED_PAGE if a CACHED page was returned.
4136	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
4137	* @param pVM The VM handle.
4138	* @param GCPhys The GC physical address of the page we're gonna shadow.
4139	* For 4MB and 2MB PD entries, it's the first address the
4140	* shadow PT is covering.
4141	* @param enmKind The kind of mapping.
4142	* @param enmAccess Access type for the mapping (only relevant for big pages)
4143	* @param iUser The shadow page pool index of the user table.
4144	* @param iUserTable The index into the user table (shadowed).
4145	* @param ppPage Where to store the pointer to the page. NULL is stored here on failure.
4146	* @param fLockPage Lock the page
4147	*/
4148	int pgmPoolAllocEx(PVM pVM, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, PGMPOOLACCESS enmAccess, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage, bool fLockPage)
4149	{
4150	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4151	STAM_PROFILE_ADV_START(&pPool->StatAlloc, a);
4152	LogFlow(("pgmPoolAlloc: GCPhys=%RGp enmKind=%s iUser=%#x iUserTable=%#x\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable));
4153	*ppPage = NULL;
4154	/** @todo CSAM/PGMPrefetchPage messes up here during CSAMR3CheckGates
4155	* (TRPMR3SyncIDT) because of FF priority. Try fix that?
4156	* Assert(!(pVM->pgm.s.fGlobalSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)); */
4157
4158	pgmLock(pVM);
4159
4160	#ifdef PGMPOOL_WITH_CACHE
4161	if (pPool->fCacheEnabled)
4162	{
4163	int rc2 = pgmPoolCacheAlloc(pPool, GCPhys, enmKind, enmAccess, iUser, iUserTable, ppPage);
4164	if (RT_SUCCESS(rc2))
4165	{
4166	if (fLockPage)
4167	pgmPoolLockPage(pPool, *ppPage);
4168	pgmUnlock(pVM);
4169	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4170	LogFlow(("pgmPoolAlloc: cached returns %Rrc ppPage=%p:{.Key=%RHp, .idx=%d}\n", rc2, ppPage, (ppPage)->Core.Key, (ppPage)->idx));
4171	return rc2;
4172	}
4173	}
4174	#endif
4175
4176	/*
4177	* Allocate a new one.
4178	*/
4179	int rc = VINF_SUCCESS;
4180	uint16_t iNew = pPool->iFreeHead;
4181	if (iNew == NIL_PGMPOOL_IDX)
4182	{
4183	rc = pgmPoolMakeMoreFreePages(pPool, enmKind, iUser);
4184	if (RT_FAILURE(rc))
4185	{
4186	pgmUnlock(pVM);
4187	Log(("pgmPoolAlloc: returns %Rrc (Free)\n", rc));
4188	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4189	return rc;
4190	}
4191	iNew = pPool->iFreeHead;
4192	AssertReleaseReturn(iNew != NIL_PGMPOOL_IDX, VERR_INTERNAL_ERROR);
4193	}
4194
4195	/* unlink the free head */
4196	PPGMPOOLPAGE pPage = &pPool->aPages[iNew];
4197	pPool->iFreeHead = pPage->iNext;
4198	pPage->iNext = NIL_PGMPOOL_IDX;
4199
4200	/*
4201	* Initialize it.
4202	*/
4203	pPool->cUsedPages++; /* physical handler registration / pgmPoolTrackFlushGCPhysPTsSlow requirement. */
4204	pPage->enmKind = enmKind;
4205	pPage->enmAccess = enmAccess;
4206	pPage->GCPhys = GCPhys;
4207	pPage->fSeenNonGlobal = false; /* Set this to 'true' to disable this feature. */
4208	pPage->fMonitored = false;
4209	pPage->fCached = false;
4210	pPage->fReusedFlushPending = false;
4211	#ifdef PGMPOOL_WITH_MONITORING
4212	pPage->cModifications = 0;
4213	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4214	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4215	#else
4216	pPage->fCR3Mix = false;
4217	#endif
4218	#ifdef PGMPOOL_WITH_USER_TRACKING
4219	pPage->cPresent = 0;
4220	pPage->iFirstPresent = ~0;
4221	pPage->pvLastAccessHandlerFault = 0;
4222	pPage->cLastAccessHandlerCount = 0;
4223	pPage->pvLastAccessHandlerRip = 0;
4224
4225	/*
4226	* Insert into the tracking and cache. If this fails, free the page.
4227	*/
4228	int rc3 = pgmPoolTrackInsert(pPool, pPage, GCPhys, iUser, iUserTable);
4229	if (RT_FAILURE(rc3))
4230	{
4231	pPool->cUsedPages--;
4232	pPage->enmKind = PGMPOOLKIND_FREE;
4233	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
4234	pPage->GCPhys = NIL_RTGCPHYS;
4235	pPage->iNext = pPool->iFreeHead;
4236	pPool->iFreeHead = pPage->idx;
4237	pgmUnlock(pVM);
4238	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4239	Log(("pgmPoolAlloc: returns %Rrc (Insert)\n", rc3));
4240	return rc3;
4241	}
4242	#endif /* PGMPOOL_WITH_USER_TRACKING */
4243
4244	/*
4245	* Commit the allocation, clear the page and return.
4246	*/
4247	#ifdef VBOX_WITH_STATISTICS
4248	if (pPool->cUsedPages > pPool->cUsedPagesHigh)
4249	pPool->cUsedPagesHigh = pPool->cUsedPages;
4250	#endif
4251
4252	if (!pPage->fZeroed)
4253	{
4254	STAM_PROFILE_START(&pPool->StatZeroPage, z);
4255	void *pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
4256	ASMMemZeroPage(pv);
4257	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
4258	}
4259
4260	*ppPage = pPage;
4261	if (fLockPage)
4262	pgmPoolLockPage(pPool, pPage);
4263	pgmUnlock(pVM);
4264	LogFlow(("pgmPoolAlloc: returns %Rrc *ppPage=%p:{.Key=%RHp, .idx=%d, .fCached=%RTbool, .fMonitored=%RTbool}\n",
4265	rc, pPage, pPage->Core.Key, pPage->idx, pPage->fCached, pPage->fMonitored));
4266	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4267	return rc;
4268	}
4269
4270
4271	/**
4272	* Frees a usage of a pool page.
4273	*
4274	* @param pVM The VM handle.
4275	* @param HCPhys The HC physical address of the shadow page.
4276	* @param iUser The shadow page pool index of the user table.
4277	* @param iUserTable The index into the user table (shadowed).
4278	*/
4279	void pgmPoolFree(PVM pVM, RTHCPHYS HCPhys, uint16_t iUser, uint32_t iUserTable)
4280	{
4281	LogFlow(("pgmPoolFree: HCPhys=%RHp iUser=%#x iUserTable=%#x\n", HCPhys, iUser, iUserTable));
4282	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4283	pgmPoolFreeByPage(pPool, pgmPoolGetPage(pPool, HCPhys), iUser, iUserTable);
4284	}
4285
4286	/**
4287	* Internal worker for finding a 'in-use' shadow page give by it's physical address.
4288	*
4289	* @returns Pointer to the shadow page structure.
4290	* @param pPool The pool.
4291	* @param HCPhys The HC physical address of the shadow page.
4292	*/
4293	PPGMPOOLPAGE pgmPoolGetPage(PPGMPOOL pPool, RTHCPHYS HCPhys)
4294	{
4295	PVM pVM = pPool->CTX_SUFF(pVM);
4296
4297	Assert(PGMIsLockOwner(pVM));
4298
4299	/*
4300	* Look up the page.
4301	*/
4302	pgmLock(pVM);
4303	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, HCPhys & X86_PTE_PAE_PG_MASK);
4304	pgmUnlock(pVM);
4305
4306	AssertFatalMsg(pPage && pPage->enmKind != PGMPOOLKIND_FREE, ("HCPhys=%RHp pPage=%p idx=%d\n", HCPhys, pPage, (pPage) ? pPage->idx : 0));
4307	return pPage;
4308	}
4309
4310
4311	#ifdef IN_RING3
4312	/**
4313	* Flushes the entire cache.
4314	*
4315	* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
4316	* and execute this CR3 flush.
4317	*
4318	* @param pPool The pool.
4319	*/
4320	void pgmR3PoolReset(PVM pVM)
4321	{
4322	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4323
4324	Assert(PGMIsLockOwner(pVM));
4325	STAM_PROFILE_START(&pPool->StatFlushAllInt, a);
4326	LogFlow(("pgmPoolFlushAllInt:\n"));
4327
4328	/*
4329	* If there are no pages in the pool, there is nothing to do.
4330	*/
4331	if (pPool->cCurPages <= PGMPOOL_IDX_FIRST)
4332	{
4333	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
4334	return;
4335	}
4336
4337	/*
4338	* Exit the shadow mode since we're going to clear everything,
4339	* including the root page.
4340	*/
4341	for (unsigned i=0;i<pVM->cCPUs;i++)
4342	{
4343	PVMCPU pVCpu = &pVM->aCpus[i];
4344	pgmR3ExitShadowModeBeforePoolFlush(pVM, pVCpu);
4345	}
4346
4347	/*
4348	* Nuke the free list and reinsert all pages into it.
4349	*/
4350	for (unsigned i = pPool->cCurPages - 1; i >= PGMPOOL_IDX_FIRST; i--)
4351	{
4352	PPGMPOOLPAGE pPage = &pPool->aPages[i];
4353
4354	Assert(pPage->Core.Key == MMPage2Phys(pVM, pPage->pvPageR3));
4355	#ifdef PGMPOOL_WITH_MONITORING
4356	if (pPage->fMonitored)
4357	pgmPoolMonitorFlush(pPool, pPage);
4358	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4359	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4360	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
4361	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
4362	pPage->cModifications = 0;
4363	#endif
4364	pPage->GCPhys = NIL_RTGCPHYS;
4365	pPage->enmKind = PGMPOOLKIND_FREE;
4366	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
4367	Assert(pPage->idx == i);
4368	pPage->iNext = i + 1;
4369	pPage->fZeroed = false; /* This could probably be optimized, but better safe than sorry. */
4370	pPage->fSeenNonGlobal = false;
4371	pPage->fMonitored = false;
4372	pPage->fCached = false;
4373	pPage->fReusedFlushPending = false;
4374	#ifdef PGMPOOL_WITH_USER_TRACKING
4375	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
4376	#else
4377	pPage->fCR3Mix = false;
4378	#endif
4379	#ifdef PGMPOOL_WITH_CACHE
4380	pPage->iAgeNext = NIL_PGMPOOL_IDX;
4381	pPage->iAgePrev = NIL_PGMPOOL_IDX;
4382	#endif
4383	pPage->cLocked = 0;
4384	}
4385	pPool->aPages[pPool->cCurPages - 1].iNext = NIL_PGMPOOL_IDX;
4386	pPool->iFreeHead = PGMPOOL_IDX_FIRST;
4387	pPool->cUsedPages = 0;
4388
4389	#ifdef PGMPOOL_WITH_USER_TRACKING
4390	/*
4391	* Zap and reinitialize the user records.
4392	*/
4393	pPool->cPresent = 0;
4394	pPool->iUserFreeHead = 0;
4395	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
4396	const unsigned cMaxUsers = pPool->cMaxUsers;
4397	for (unsigned i = 0; i < cMaxUsers; i++)
4398	{
4399	paUsers[i].iNext = i + 1;
4400	paUsers[i].iUser = NIL_PGMPOOL_IDX;
4401	paUsers[i].iUserTable = 0xfffffffe;
4402	}
4403	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
4404	#endif
4405
4406	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
4407	/*
4408	* Clear all the GCPhys links and rebuild the phys ext free list.
4409	*/
4410	for (PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRanges);
4411	pRam;
4412	pRam = pRam->CTX_SUFF(pNext))
4413	{
4414	unsigned iPage = pRam->cb >> PAGE_SHIFT;
4415	while (iPage-- > 0)
4416	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
4417	}
4418
4419	pPool->iPhysExtFreeHead = 0;
4420	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
4421	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
4422	for (unsigned i = 0; i < cMaxPhysExts; i++)
4423	{
4424	paPhysExts[i].iNext = i + 1;
4425	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
4426	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
4427	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
4428	}
4429	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
4430	#endif
4431
4432	#ifdef PGMPOOL_WITH_MONITORING
4433	/*
4434	* Just zap the modified list.
4435	*/
4436	pPool->cModifiedPages = 0;
4437	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
4438	#endif
4439
4440	#ifdef PGMPOOL_WITH_CACHE
4441	/*
4442	* Clear the GCPhys hash and the age list.
4443	*/
4444	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
4445	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
4446	pPool->iAgeHead = NIL_PGMPOOL_IDX;
4447	pPool->iAgeTail = NIL_PGMPOOL_IDX;
4448	#endif
4449
4450	/*
4451	* Reinsert active pages into the hash and ensure monitoring chains are correct.
4452	*/
4453	for (unsigned i = PGMPOOL_IDX_FIRST_SPECIAL; i < PGMPOOL_IDX_FIRST; i++)
4454	{
4455	PPGMPOOLPAGE pPage = &pPool->aPages[i];
4456	pPage->iNext = NIL_PGMPOOL_IDX;
4457	#ifdef PGMPOOL_WITH_MONITORING
4458	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4459	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4460	pPage->cModifications = 0;
4461	/* ASSUMES that we're not sharing with any of the other special pages (safe for now). */
4462	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
4463	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
4464	if (pPage->fMonitored)
4465	{
4466	int rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
4467	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
4468	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
4469	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
4470	pPool->pszAccessHandler);
4471	AssertFatalRCSuccess(rc);
4472	# ifdef PGMPOOL_WITH_CACHE
4473	pgmPoolHashInsert(pPool, pPage);
4474	# endif
4475	}
4476	#endif
4477	#ifdef PGMPOOL_WITH_USER_TRACKING
4478	Assert(pPage->iUserHead == NIL_PGMPOOL_USER_INDEX); /* for now */
4479	#endif
4480	#ifdef PGMPOOL_WITH_CACHE
4481	Assert(pPage->iAgeNext == NIL_PGMPOOL_IDX);
4482	Assert(pPage->iAgePrev == NIL_PGMPOOL_IDX);
4483	#endif
4484	}
4485
4486	for (unsigned i=0;i<pVM->cCPUs;i++)
4487	{
4488	PVMCPU pVCpu = &pVM->aCpus[i];
4489	/*
4490	* Re-enter the shadowing mode and assert Sync CR3 FF.
4491	*/
4492	pgmR3ReEnterShadowModeAfterPoolFlush(pVM, pVCpu);
4493	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
4494	}
4495
4496	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
4497	}
4498	#endif /* IN_RING3 */
4499
4500	#ifdef LOG_ENABLED
4501	static const char *pgmPoolPoolKindToStr(uint8_t enmKind)
4502	{
4503	switch(enmKind)
4504	{
4505	case PGMPOOLKIND_INVALID:
4506	return "PGMPOOLKIND_INVALID";
4507	case PGMPOOLKIND_FREE:
4508	return "PGMPOOLKIND_FREE";
4509	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
4510	return "PGMPOOLKIND_32BIT_PT_FOR_PHYS";
4511	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
4512	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT";
4513	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
4514	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB";
4515	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
4516	return "PGMPOOLKIND_PAE_PT_FOR_PHYS";
4517	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
4518	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_PT";
4519	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
4520	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB";
4521	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
4522	return "PGMPOOLKIND_PAE_PT_FOR_PAE_PT";
4523	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
4524	return "PGMPOOLKIND_PAE_PT_FOR_PAE_2MB";
4525	case PGMPOOLKIND_32BIT_PD:
4526	return "PGMPOOLKIND_32BIT_PD";
4527	case PGMPOOLKIND_32BIT_PD_PHYS:
4528	return "PGMPOOLKIND_32BIT_PD_PHYS";
4529	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
4530	return "PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD";
4531	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
4532	return "PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD";
4533	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
4534	return "PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD";
4535	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
4536	return "PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD";
4537	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
4538	return "PGMPOOLKIND_PAE_PD_FOR_PAE_PD";
4539	case PGMPOOLKIND_PAE_PD_PHYS:
4540	return "PGMPOOLKIND_PAE_PD_PHYS";
4541	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
4542	return "PGMPOOLKIND_PAE_PDPT_FOR_32BIT";
4543	case PGMPOOLKIND_PAE_PDPT:
4544	return "PGMPOOLKIND_PAE_PDPT";
4545	case PGMPOOLKIND_PAE_PDPT_PHYS:
4546	return "PGMPOOLKIND_PAE_PDPT_PHYS";
4547	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
4548	return "PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT";
4549	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
4550	return "PGMPOOLKIND_64BIT_PDPT_FOR_PHYS";
4551	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
4552	return "PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD";
4553	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
4554	return "PGMPOOLKIND_64BIT_PD_FOR_PHYS";
4555	case PGMPOOLKIND_64BIT_PML4:
4556	return "PGMPOOLKIND_64BIT_PML4";
4557	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
4558	return "PGMPOOLKIND_EPT_PDPT_FOR_PHYS";
4559	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
4560	return "PGMPOOLKIND_EPT_PD_FOR_PHYS";
4561	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
4562	return "PGMPOOLKIND_EPT_PT_FOR_PHYS";
4563	case PGMPOOLKIND_ROOT_NESTED:
4564	return "PGMPOOLKIND_ROOT_NESTED";
4565	}
4566	return "Unknown kind!";
4567	}
4568	#endif /* LOG_ENABLED*/

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllPool.cpp@ 22348

Download in other formats: