PGMAllPool.cpp@ 18645

Last change on this file since 18645 was 18355, checked in by vboxsync, 16 years ago
pgmPoolCacheFreeOne: Rewrote recursion to a finit loop; overflowing the stack is bad both in RC (#DF) and R0 (panic/corruption).
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 159.4 KB

Line
1	/* $Id: PGMAllPool.cpp 18355 2009-03-26 22:18:55Z 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	__BEGIN_DECLS
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	__END_DECLS
71
72
73	/**
74	* Checks if the specified page pool kind is for a 4MB or 2MB guest page.
75	*
76	* @returns true if it's the shadow of a 4MB or 2MB guest page, otherwise false.
77	* @param enmKind The page kind.
78	*/
79	DECLINLINE(bool) pgmPoolIsBigPage(PGMPOOLKIND enmKind)
80	{
81	switch (enmKind)
82	{
83	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
84	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
85	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
86	return true;
87	default:
88	return false;
89	}
90	}
91
92	/** @def PGMPOOL_PAGE_2_LOCKED_PTR
93	* Maps a pool page pool into the current context and lock it (RC only).
94	*
95	* @returns VBox status code.
96	* @param pVM The VM handle.
97	* @param pPage The pool page.
98	*
99	* @remark In RC this uses PGMGCDynMapHCPage(), so it will consume of the
100	* small page window employeed by that function. Be careful.
101	* @remark There is no need to assert on the result.
102	*/
103	#if defined(IN_RC)
104	DECLINLINE(void *) PGMPOOL_PAGE_2_LOCKED_PTR(PVM pVM, PPGMPOOLPAGE pPage)
105	{
106	void *pv = pgmPoolMapPageInlined(&pVM->pgm.s, pPage);
107
108	/* Make sure the dynamic mapping will not be reused. */
109	if (pv)
110	PGMDynLockHCPage(pVM, (uint8_t *)pv);
111
112	return pv;
113	}
114	#else
115	# define PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage) PGMPOOL_PAGE_2_PTR(pVM, pPage)
116	#endif
117
118	/** @def PGMPOOL_UNLOCK_PTR
119	* Unlock a previously locked dynamic caching (RC only).
120	*
121	* @returns VBox status code.
122	* @param pVM The VM handle.
123	* @param pPage The pool page.
124	*
125	* @remark In RC this uses PGMGCDynMapHCPage(), so it will consume of the
126	* small page window employeed by that function. Be careful.
127	* @remark There is no need to assert on the result.
128	*/
129	#if defined(IN_RC)
130	DECLINLINE(void) PGMPOOL_UNLOCK_PTR(PVM pVM, void *pvPage)
131	{
132	if (pvPage)
133	PGMDynUnlockHCPage(pVM, (uint8_t *)pvPage);
134	}
135	#else
136	# define PGMPOOL_UNLOCK_PTR(pVM, pPage) do {} while (0)
137	#endif
138
139
140	#ifdef PGMPOOL_WITH_MONITORING
141	/**
142	* Determin the size of a write instruction.
143	* @returns number of bytes written.
144	* @param pDis The disassembler state.
145	*/
146	static unsigned pgmPoolDisasWriteSize(PDISCPUSTATE pDis)
147	{
148	/*
149	* This is very crude and possibly wrong for some opcodes,
150	* but since it's not really supposed to be called we can
151	* probably live with that.
152	*/
153	return DISGetParamSize(pDis, &pDis->param1);
154	}
155
156
157	/**
158	* Flushes a chain of pages sharing the same access monitor.
159	*
160	* @returns VBox status code suitable for scheduling.
161	* @param pPool The pool.
162	* @param pPage A page in the chain.
163	*/
164	int pgmPoolMonitorChainFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
165	{
166	LogFlow(("pgmPoolMonitorChainFlush: Flush page %RGp type=%d\n", pPage->GCPhys, pPage->enmKind));
167
168	/*
169	* Find the list head.
170	*/
171	uint16_t idx = pPage->idx;
172	if (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
173	{
174	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
175	{
176	idx = pPage->iMonitoredPrev;
177	Assert(idx != pPage->idx);
178	pPage = &pPool->aPages[idx];
179	}
180	}
181
182	/*
183	* Iterate the list flushing each shadow page.
184	*/
185	int rc = VINF_SUCCESS;
186	for (;;)
187	{
188	idx = pPage->iMonitoredNext;
189	Assert(idx != pPage->idx);
190	if (pPage->idx >= PGMPOOL_IDX_FIRST)
191	{
192	int rc2 = pgmPoolFlushPage(pPool, pPage);
193	AssertRC(rc2);
194	}
195	/* next */
196	if (idx == NIL_PGMPOOL_IDX)
197	break;
198	pPage = &pPool->aPages[idx];
199	}
200	return rc;
201	}
202
203
204	/**
205	* Wrapper for getting the current context pointer to the entry being modified.
206	*
207	* @returns VBox status code suitable for scheduling.
208	* @param pVM VM Handle.
209	* @param pvDst Destination address
210	* @param pvSrc Source guest virtual address.
211	* @param GCPhysSrc The source guest physical address.
212	* @param cb Size of data to read
213	*/
214	DECLINLINE(int) pgmPoolPhysSimpleReadGCPhys(PVM pVM, void *pvDst, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvSrc, RTGCPHYS GCPhysSrc, size_t cb)
215	{
216	#if defined(IN_RING3)
217	memcpy(pvDst, (RTHCPTR)((uintptr_t)pvSrc & ~(RTHCUINTPTR)(cb - 1)), cb);
218	return VINF_SUCCESS;
219	#else
220	/* @todo in RC we could attempt to use the virtual address, although this can cause many faults (PAE Windows XP guest). */
221	return PGMPhysSimpleReadGCPhys(pVM, pvDst, GCPhysSrc & ~(RTGCPHYS)(cb - 1), cb);
222	#endif
223	}
224
225	/**
226	* Process shadow entries before they are changed by the guest.
227	*
228	* For PT entries we will clear them. For PD entries, we'll simply check
229	* for mapping conflicts and set the SyncCR3 FF if found.
230	*
231	* @param pPool The pool.
232	* @param pPage The head page.
233	* @param GCPhysFault The guest physical fault address.
234	* @param uAddress In R0 and GC this is the guest context fault address (flat).
235	* In R3 this is the host context 'fault' address.
236	* @param pCpu The disassembler state for figuring out the write size.
237	* This need not be specified if the caller knows we won't do cross entry accesses.
238	*/
239	void pgmPoolMonitorChainChanging(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhysFault, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvAddress, PDISCPUSTATE pCpu)
240	{
241	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
242	const unsigned off = GCPhysFault & PAGE_OFFSET_MASK;
243	const unsigned cbWrite = (pCpu) ? pgmPoolDisasWriteSize(pCpu) : 0;
244	PVM pVM = pPool->CTX_SUFF(pVM);
245
246	LogFlow(("pgmPoolMonitorChainChanging: %RGv phys=%RGp kind=%s cbWrite=%d\n", (RTGCPTR)pvAddress, GCPhysFault, pgmPoolPoolKindToStr(pPage->enmKind), cbWrite));
247	for (;;)
248	{
249	union
250	{
251	void *pv;
252	PX86PT pPT;
253	PX86PTPAE pPTPae;
254	PX86PD pPD;
255	PX86PDPAE pPDPae;
256	PX86PDPT pPDPT;
257	PX86PML4 pPML4;
258	} uShw;
259
260	uShw.pv = NULL;
261	switch (pPage->enmKind)
262	{
263	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
264	{
265	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
266	const unsigned iShw = off / sizeof(X86PTE);
267	LogFlow(("PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT iShw=%x\n", iShw));
268	if (uShw.pPT->a[iShw].n.u1Present)
269	{
270	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
271	X86PTE GstPte;
272
273	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
274	AssertRC(rc);
275	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));
276	pgmPoolTracDerefGCPhysHint(pPool, pPage,
277	uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK,
278	GstPte.u & X86_PTE_PG_MASK);
279	# endif
280	uShw.pPT->a[iShw].u = 0;
281	}
282	break;
283	}
284
285	/* page/2 sized */
286	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
287	{
288	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
289	if (!((off ^ pPage->GCPhys) & (PAGE_SIZE / 2)))
290	{
291	const unsigned iShw = (off / sizeof(X86PTE)) & (X86_PG_PAE_ENTRIES - 1);
292	LogFlow(("PGMPOOLKIND_PAE_PT_FOR_32BIT_PT iShw=%x\n", iShw));
293	if (uShw.pPTPae->a[iShw].n.u1Present)
294	{
295	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
296	X86PTE GstPte;
297	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
298	AssertRC(rc);
299
300	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));
301	pgmPoolTracDerefGCPhysHint(pPool, pPage,
302	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
303	GstPte.u & X86_PTE_PG_MASK);
304	# endif
305	uShw.pPTPae->a[iShw].u = 0;
306	}
307	}
308	break;
309	}
310
311	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
312	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
313	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
314	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
315	{
316	unsigned iGst = off / sizeof(X86PDE);
317	unsigned iShwPdpt = iGst / 256;
318	unsigned iShw = (iGst % 256) * 2;
319	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
320
321	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));
322	if (iShwPdpt == pPage->enmKind - (unsigned)PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD)
323	{
324	for (unsigned i = 0; i < 2; i++)
325	{
326	# ifndef IN_RING0
327	if ((uShw.pPDPae->a[iShw + i].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
328	{
329	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
330	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3);
331	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw=%#x!\n", iShwPdpt, iShw+i));
332	break;
333	}
334	else
335	# endif /* !IN_RING0 */
336	if (uShw.pPDPae->a[iShw+i].n.u1Present)
337	{
338	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw+i, uShw.pPDPae->a[iShw+i].u));
339	pgmPoolFree(pVM,
340	uShw.pPDPae->a[iShw+i].u & X86_PDE_PAE_PG_MASK,
341	pPage->idx,
342	iShw + i);
343	uShw.pPDPae->a[iShw+i].u = 0;
344	}
345
346	/* paranoia / a bit assumptive. */
347	if ( pCpu
348	&& (off & 3)
349	&& (off & 3) + cbWrite > 4)
350	{
351	const unsigned iShw2 = iShw + 2 + i;
352	if (iShw2 < RT_ELEMENTS(uShw.pPDPae->a))
353	{
354	# ifndef IN_RING0
355	if ((uShw.pPDPae->a[iShw2].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
356	{
357	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
358	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3);
359	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw2=%#x!\n", iShwPdpt, iShw2));
360	break;
361	}
362	else
363	# endif /* !IN_RING0 */
364	if (uShw.pPDPae->a[iShw2].n.u1Present)
365	{
366	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
367	pgmPoolFree(pVM,
368	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
369	pPage->idx,
370	iShw2);
371	uShw.pPDPae->a[iShw2].u = 0;
372	}
373	}
374	}
375	}
376	}
377	break;
378	}
379
380	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
381	{
382	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
383	const unsigned iShw = off / sizeof(X86PTEPAE);
384	if (uShw.pPTPae->a[iShw].n.u1Present)
385	{
386	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
387	X86PTEPAE GstPte;
388	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
389	AssertRC(rc);
390
391	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));
392	pgmPoolTracDerefGCPhysHint(pPool, pPage,
393	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
394	GstPte.u & X86_PTE_PAE_PG_MASK);
395	# endif
396	uShw.pPTPae->a[iShw].u = 0;
397	}
398
399	/* paranoia / a bit assumptive. */
400	if ( pCpu
401	&& (off & 7)
402	&& (off & 7) + cbWrite > sizeof(X86PTEPAE))
403	{
404	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTEPAE);
405	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPTPae->a));
406
407	if (uShw.pPTPae->a[iShw2].n.u1Present)
408	{
409	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
410	X86PTEPAE GstPte;
411	# ifdef IN_RING3
412	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, (RTHCPTR)((RTHCUINTPTR)pvAddress + sizeof(GstPte)), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
413	# else
414	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress + sizeof(GstPte), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
415	# endif
416	AssertRC(rc);
417	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));
418	pgmPoolTracDerefGCPhysHint(pPool, pPage,
419	uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK,
420	GstPte.u & X86_PTE_PAE_PG_MASK);
421	# endif
422	uShw.pPTPae->a[iShw2].u = 0;
423	}
424	}
425	break;
426	}
427
428	case PGMPOOLKIND_32BIT_PD:
429	{
430	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
431	const unsigned iShw = off / sizeof(X86PTE); // ASSUMING 32-bit guest paging!
432
433	LogFlow(("pgmPoolMonitorChainChanging: PGMPOOLKIND_32BIT_PD %x\n", iShw));
434	# ifndef IN_RING0
435	if (uShw.pPD->a[iShw].u & PGM_PDFLAGS_MAPPING)
436	{
437	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
438	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3);
439	STAM_COUNTER_INC(&(pVM->pgm.s.StatRZGuestCR3WriteConflict));
440	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
441	break;
442	}
443	# endif /* !IN_RING0 */
444	# ifndef IN_RING0
445	else
446	# endif /* !IN_RING0 */
447	{
448	if (uShw.pPD->a[iShw].n.u1Present)
449	{
450	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
451	pgmPoolFree(pVM,
452	uShw.pPD->a[iShw].u & X86_PDE_PAE_PG_MASK,
453	pPage->idx,
454	iShw);
455	uShw.pPD->a[iShw].u = 0;
456	}
457	}
458	/* paranoia / a bit assumptive. */
459	if ( pCpu
460	&& (off & 3)
461	&& (off & 3) + cbWrite > sizeof(X86PTE))
462	{
463	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTE);
464	if ( iShw2 != iShw
465	&& iShw2 < RT_ELEMENTS(uShw.pPD->a))
466	{
467	# ifndef IN_RING0
468	if (uShw.pPD->a[iShw2].u & PGM_PDFLAGS_MAPPING)
469	{
470	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
471	STAM_COUNTER_INC(&(pVM->pgm.s.StatRZGuestCR3WriteConflict));
472	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3);
473	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
474	break;
475	}
476	# endif /* !IN_RING0 */
477	# ifndef IN_RING0
478	else
479	# endif /* !IN_RING0 */
480	{
481	if (uShw.pPD->a[iShw2].n.u1Present)
482	{
483	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPD->a[iShw2].u));
484	pgmPoolFree(pVM,
485	uShw.pPD->a[iShw2].u & X86_PDE_PAE_PG_MASK,
486	pPage->idx,
487	iShw2);
488	uShw.pPD->a[iShw2].u = 0;
489	}
490	}
491	}
492	}
493	#if 0 /* useful when running PGMAssertCR3(), a bit too troublesome for general use (TLBs). */
494	if ( uShw.pPD->a[iShw].n.u1Present
495	&& !VM_FF_ISSET(pVM, VM_FF_PGM_SYNC_CR3))
496	{
497	LogFlow(("pgmPoolMonitorChainChanging: iShw=%#x: %RX32 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
498	# ifdef IN_RC /* TLB load - we're pushing things a bit... */
499	ASMProbeReadByte(pvAddress);
500	# endif
501	pgmPoolFree(pVM, uShw.pPD->a[iShw].u & X86_PDE_PG_MASK, pPage->idx, iShw);
502	uShw.pPD->a[iShw].u = 0;
503	}
504	#endif
505	break;
506	}
507
508	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
509	{
510	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
511	const unsigned iShw = off / sizeof(X86PDEPAE);
512	#ifndef IN_RING0
513	if (uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING)
514	{
515	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
516	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3);
517	STAM_COUNTER_INC(&(pVM->pgm.s.StatRZGuestCR3WriteConflict));
518	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
519	break;
520	}
521	#endif /* !IN_RING0 */
522	/*
523	* Causes trouble when the guest uses a PDE to refer to the whole page table level
524	* structure. (Invalidate here; faults later on when it tries to change the page
525	* table entries -> recheck; probably only applies to the RC case.)
526	*/
527	# ifndef IN_RING0
528	else
529	# endif /* !IN_RING0 */
530	{
531	if (uShw.pPDPae->a[iShw].n.u1Present)
532	{
533	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
534	pgmPoolFree(pVM,
535	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
536	pPage->idx,
537	iShw);
538	uShw.pPDPae->a[iShw].u = 0;
539	}
540	}
541	/* paranoia / a bit assumptive. */
542	if ( pCpu
543	&& (off & 7)
544	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
545	{
546	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
547	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
548
549	#ifndef IN_RING0
550	if ( iShw2 != iShw
551	&& uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING)
552	{
553	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
554	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3);
555	STAM_COUNTER_INC(&(pVM->pgm.s.StatRZGuestCR3WriteConflict));
556	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
557	break;
558	}
559	#endif /* !IN_RING0 */
560	# ifndef IN_RING0
561	else
562	# endif /* !IN_RING0 */
563	if (uShw.pPDPae->a[iShw2].n.u1Present)
564	{
565	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
566	pgmPoolFree(pVM,
567	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
568	pPage->idx,
569	iShw2);
570	uShw.pPDPae->a[iShw2].u = 0;
571	}
572	}
573	break;
574	}
575
576	case PGMPOOLKIND_PAE_PDPT:
577	{
578	/*
579	* Hopefully this doesn't happen very often:
580	* - touching unused parts of the page
581	* - messing with the bits of pd pointers without changing the physical address
582	*/
583	/* PDPT roots are not page aligned; 32 byte only! */
584	const unsigned offPdpt = GCPhysFault - pPage->GCPhys;
585
586	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
587	const unsigned iShw = offPdpt / sizeof(X86PDPE);
588	if (iShw < X86_PG_PAE_PDPE_ENTRIES) /* don't use RT_ELEMENTS(uShw.pPDPT->a), because that's for long mode only */
589	{
590	# ifndef IN_RING0
591	if (uShw.pPDPT->a[iShw].u & PGM_PLXFLAGS_MAPPING)
592	{
593	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
594	STAM_COUNTER_INC(&(pVM->pgm.s.StatRZGuestCR3WriteConflict));
595	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3);
596	LogFlow(("pgmPoolMonitorChainChanging: Detected pdpt conflict at iShw=%#x!\n", iShw));
597	break;
598	}
599	# endif /* !IN_RING0 */
600	# ifndef IN_RING0
601	else
602	# endif /* !IN_RING0 */
603	if (uShw.pPDPT->a[iShw].n.u1Present)
604	{
605	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
606	pgmPoolFree(pVM,
607	uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK,
608	pPage->idx,
609	iShw);
610	uShw.pPDPT->a[iShw].u = 0;
611	}
612
613	/* paranoia / a bit assumptive. */
614	if ( pCpu
615	&& (offPdpt & 7)
616	&& (offPdpt & 7) + cbWrite > sizeof(X86PDPE))
617	{
618	const unsigned iShw2 = (offPdpt + cbWrite - 1) / sizeof(X86PDPE);
619	if ( iShw2 != iShw
620	&& iShw2 < X86_PG_PAE_PDPE_ENTRIES)
621	{
622	# ifndef IN_RING0
623	if (uShw.pPDPT->a[iShw2].u & PGM_PLXFLAGS_MAPPING)
624	{
625	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
626	STAM_COUNTER_INC(&(pVM->pgm.s.StatRZGuestCR3WriteConflict));
627	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3);
628	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
629	break;
630	}
631	# endif /* !IN_RING0 */
632	# ifndef IN_RING0
633	else
634	# endif /* !IN_RING0 */
635	if (uShw.pPDPT->a[iShw2].n.u1Present)
636	{
637	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
638	pgmPoolFree(pVM,
639	uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK,
640	pPage->idx,
641	iShw2);
642	uShw.pPDPT->a[iShw2].u = 0;
643	}
644	}
645	}
646	}
647	break;
648	}
649
650	#ifndef IN_RC
651	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
652	{
653	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
654	const unsigned iShw = off / sizeof(X86PDEPAE);
655	Assert(!(uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING));
656	if (uShw.pPDPae->a[iShw].n.u1Present)
657	{
658	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
659	pgmPoolFree(pVM,
660	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
661	pPage->idx,
662	iShw);
663	uShw.pPDPae->a[iShw].u = 0;
664	}
665	/* paranoia / a bit assumptive. */
666	if ( pCpu
667	&& (off & 7)
668	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
669	{
670	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
671	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
672
673	Assert(!(uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING));
674	if (uShw.pPDPae->a[iShw2].n.u1Present)
675	{
676	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
677	pgmPoolFree(pVM,
678	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
679	pPage->idx,
680	iShw2);
681	uShw.pPDPae->a[iShw2].u = 0;
682	}
683	}
684	break;
685	}
686
687	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
688	{
689	/*
690	* Hopefully this doesn't happen very often:
691	* - messing with the bits of pd pointers without changing the physical address
692	*/
693	if (!VM_FF_ISSET(pVM, VM_FF_PGM_SYNC_CR3))
694	{
695	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
696	const unsigned iShw = off / sizeof(X86PDPE);
697	if (uShw.pPDPT->a[iShw].n.u1Present)
698	{
699	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
700	pgmPoolFree(pVM, uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK, pPage->idx, iShw);
701	uShw.pPDPT->a[iShw].u = 0;
702	}
703	/* paranoia / a bit assumptive. */
704	if ( pCpu
705	&& (off & 7)
706	&& (off & 7) + cbWrite > sizeof(X86PDPE))
707	{
708	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDPE);
709	if (uShw.pPDPT->a[iShw2].n.u1Present)
710	{
711	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
712	pgmPoolFree(pVM, uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK, pPage->idx, iShw2);
713	uShw.pPDPT->a[iShw2].u = 0;
714	}
715	}
716	}
717	break;
718	}
719
720	case PGMPOOLKIND_64BIT_PML4:
721	{
722	/*
723	* Hopefully this doesn't happen very often:
724	* - messing with the bits of pd pointers without changing the physical address
725	*/
726	if (!VM_FF_ISSET(pVM, VM_FF_PGM_SYNC_CR3))
727	{
728	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
729	const unsigned iShw = off / sizeof(X86PDPE);
730	if (uShw.pPML4->a[iShw].n.u1Present)
731	{
732	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPML4->a[iShw].u));
733	pgmPoolFree(pVM, uShw.pPML4->a[iShw].u & X86_PML4E_PG_MASK, pPage->idx, iShw);
734	uShw.pPML4->a[iShw].u = 0;
735	}
736	/* paranoia / a bit assumptive. */
737	if ( pCpu
738	&& (off & 7)
739	&& (off & 7) + cbWrite > sizeof(X86PDPE))
740	{
741	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PML4E);
742	if (uShw.pPML4->a[iShw2].n.u1Present)
743	{
744	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPML4->a[iShw2].u));
745	pgmPoolFree(pVM, uShw.pPML4->a[iShw2].u & X86_PML4E_PG_MASK, pPage->idx, iShw2);
746	uShw.pPML4->a[iShw2].u = 0;
747	}
748	}
749	}
750	break;
751	}
752	#endif /* IN_RING0 */
753
754	default:
755	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
756	}
757	PGMPOOL_UNLOCK_PTR(pVM, uShw.pv);
758
759	/* next */
760	if (pPage->iMonitoredNext == NIL_PGMPOOL_IDX)
761	return;
762	pPage = &pPool->aPages[pPage->iMonitoredNext];
763	}
764	}
765
766	# ifndef IN_RING3
767	/**
768	* Checks if a access could be a fork operation in progress.
769	*
770	* Meaning, that the guest is setuping up the parent process for Copy-On-Write.
771	*
772	* @returns true if it's likly that we're forking, otherwise false.
773	* @param pPool The pool.
774	* @param pCpu The disassembled instruction.
775	* @param offFault The access offset.
776	*/
777	DECLINLINE(bool) pgmPoolMonitorIsForking(PPGMPOOL pPool, PDISCPUSTATE pCpu, unsigned offFault)
778	{
779	/*
780	* i386 linux is using btr to clear X86_PTE_RW.
781	* The functions involved are (2.6.16 source inspection):
782	* clear_bit
783	* ptep_set_wrprotect
784	* copy_one_pte
785	* copy_pte_range
786	* copy_pmd_range
787	* copy_pud_range
788	* copy_page_range
789	* dup_mmap
790	* dup_mm
791	* copy_mm
792	* copy_process
793	* do_fork
794	*/
795	if ( pCpu->pCurInstr->opcode == OP_BTR
796	&& !(offFault & 4)
797	/** @todo Validate that the bit index is X86_PTE_RW. */
798	)
799	{
800	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,Fork));
801	return true;
802	}
803	return false;
804	}
805
806
807	/**
808	* Determine whether the page is likely to have been reused.
809	*
810	* @returns true if we consider the page as being reused for a different purpose.
811	* @returns false if we consider it to still be a paging page.
812	* @param pVM VM Handle.
813	* @param pPage The page in question.
814	* @param pRegFrame Trap register frame.
815	* @param pCpu The disassembly info for the faulting instruction.
816	* @param pvFault The fault address.
817	*
818	* @remark The REP prefix check is left to the caller because of STOSD/W.
819	*/
820	DECLINLINE(bool) pgmPoolMonitorIsReused(PVM pVM, PPGMPOOLPAGE pPage, PCPUMCTXCORE pRegFrame, PDISCPUSTATE pCpu, RTGCPTR pvFault)
821	{
822	#ifndef IN_RC
823	/** @todo could make this general, faulting close to rsp should be safe reuse heuristic. */
824	if ( HWACCMHasPendingIrq(pVM)
825	&& (pRegFrame->rsp - pvFault) < 32)
826	{
827	/* Fault caused by stack writes while trying to inject an interrupt event. */
828	Log(("pgmPoolMonitorIsReused: reused %RGv for interrupt stack (rsp=%RGv).\n", pvFault, pRegFrame->rsp));
829	return true;
830	}
831	#else
832	NOREF(pVM); NOREF(pvFault);
833	#endif
834
835	switch (pCpu->pCurInstr->opcode)
836	{
837	/* call implies the actual push of the return address faulted */
838	case OP_CALL:
839	Log4(("pgmPoolMonitorIsReused: CALL\n"));
840	return true;
841	case OP_PUSH:
842	Log4(("pgmPoolMonitorIsReused: PUSH\n"));
843	return true;
844	case OP_PUSHF:
845	Log4(("pgmPoolMonitorIsReused: PUSHF\n"));
846	return true;
847	case OP_PUSHA:
848	Log4(("pgmPoolMonitorIsReused: PUSHA\n"));
849	return true;
850	case OP_FXSAVE:
851	Log4(("pgmPoolMonitorIsReused: FXSAVE\n"));
852	return true;
853	case OP_MOVNTI: /* solaris - block_zero_no_xmm */
854	Log4(("pgmPoolMonitorIsReused: MOVNTI\n"));
855	return true;
856	case OP_MOVNTDQ: /* solaris - hwblkclr & hwblkpagecopy */
857	Log4(("pgmPoolMonitorIsReused: MOVNTDQ\n"));
858	return true;
859	case OP_MOVSWD:
860	case OP_STOSWD:
861	if ( pCpu->prefix == (PREFIX_REP\|PREFIX_REX)
862	&& pRegFrame->rcx >= 0x40
863	)
864	{
865	Assert(pCpu->mode == CPUMODE_64BIT);
866
867	Log(("pgmPoolMonitorIsReused: OP_STOSQ\n"));
868	return true;
869	}
870	return false;
871	}
872	if ( (pCpu->param1.flags & USE_REG_GEN32)
873	&& (pCpu->param1.base.reg_gen == USE_REG_ESP))
874	{
875	Log4(("pgmPoolMonitorIsReused: ESP\n"));
876	return true;
877	}
878
879	//if (pPage->fCR3Mix)
880	// return false;
881	return false;
882	}
883
884
885	/**
886	* Flushes the page being accessed.
887	*
888	* @returns VBox status code suitable for scheduling.
889	* @param pVM The VM handle.
890	* @param pPool The pool.
891	* @param pPage The pool page (head).
892	* @param pCpu The disassembly of the write instruction.
893	* @param pRegFrame The trap register frame.
894	* @param GCPhysFault The fault address as guest physical address.
895	* @param pvFault The fault address.
896	*/
897	static int pgmPoolAccessHandlerFlush(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
898	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
899	{
900	/*
901	* First, do the flushing.
902	*/
903	int rc = pgmPoolMonitorChainFlush(pPool, pPage);
904
905	/*
906	* Emulate the instruction (xp/w2k problem, requires pc/cr2/sp detection).
907	*/
908	uint32_t cbWritten;
909	int rc2 = EMInterpretInstructionCPU(pVM, pCpu, pRegFrame, pvFault, &cbWritten);
910	if (RT_SUCCESS(rc2))
911	pRegFrame->rip += pCpu->opsize;
912	else if (rc2 == VERR_EM_INTERPRETER)
913	{
914	#ifdef IN_RC
915	if (PATMIsPatchGCAddr(pVM, (RTRCPTR)pRegFrame->eip))
916	{
917	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for patch code %04x:%RGv, ignoring.\n",
918	pRegFrame->cs, (RTGCPTR)pRegFrame->eip));
919	rc = VINF_SUCCESS;
920	STAM_COUNTER_INC(&pPool->StatMonitorRZIntrFailPatch2);
921	}
922	else
923	#endif
924	{
925	rc = VINF_EM_RAW_EMULATE_INSTR;
926	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
927	}
928	}
929	else
930	rc = rc2;
931
932	/* See use in pgmPoolAccessHandlerSimple(). */
933	PGM_INVL_GUEST_TLBS();
934
935	LogFlow(("pgmPoolAccessHandlerPT: returns %Rrc (flushed)\n", rc));
936	return rc;
937
938	}
939
940
941	/**
942	* Handles the STOSD write accesses.
943	*
944	* @returns VBox status code suitable for scheduling.
945	* @param pVM The VM handle.
946	* @param pPool The pool.
947	* @param pPage The pool page (head).
948	* @param pCpu The disassembly of the write instruction.
949	* @param pRegFrame The trap register frame.
950	* @param GCPhysFault The fault address as guest physical address.
951	* @param pvFault The fault address.
952	*/
953	DECLINLINE(int) pgmPoolAccessHandlerSTOSD(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
954	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
955	{
956	Assert(pCpu->mode == CPUMODE_32BIT);
957
958	Log3(("pgmPoolAccessHandlerSTOSD\n"));
959
960	/*
961	* Increment the modification counter and insert it into the list
962	* of modified pages the first time.
963	*/
964	if (!pPage->cModifications++)
965	pgmPoolMonitorModifiedInsert(pPool, pPage);
966
967	/*
968	* Execute REP STOSD.
969	*
970	* This ASSUMES that we're not invoked by Trap0e on in a out-of-sync
971	* write situation, meaning that it's safe to write here.
972	*/
973	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
974	PVMCPU pVCpu = VMMGetCpu(pPool->CTX_SUFF(pVM));
975	#endif
976	RTGCUINTPTR pu32 = (RTGCUINTPTR)pvFault;
977	while (pRegFrame->ecx)
978	{
979	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
980	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
981	pgmPoolMonitorChainChanging(pPool, pPage, GCPhysFault, (RTGCPTR)pu32, NULL);
982	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
983	#else
984	pgmPoolMonitorChainChanging(pPool, pPage, GCPhysFault, (RTGCPTR)pu32, NULL);
985	#endif
986	#ifdef IN_RC
987	(uint32_t )pu32 = pRegFrame->eax;
988	#else
989	PGMPhysSimpleWriteGCPhys(pVM, GCPhysFault, &pRegFrame->eax, 4);
990	#endif
991	pu32 += 4;
992	GCPhysFault += 4;
993	pRegFrame->edi += 4;
994	pRegFrame->ecx--;
995	}
996	pRegFrame->rip += pCpu->opsize;
997
998	#ifdef IN_RC
999	/* See use in pgmPoolAccessHandlerSimple(). */
1000	PGM_INVL_GUEST_TLBS();
1001	#endif
1002
1003	LogFlow(("pgmPoolAccessHandlerSTOSD: returns\n"));
1004	return VINF_SUCCESS;
1005	}
1006
1007
1008	/**
1009	* Handles the simple write accesses.
1010	*
1011	* @returns VBox status code suitable for scheduling.
1012	* @param pVM The VM handle.
1013	* @param pPool The pool.
1014	* @param pPage The pool page (head).
1015	* @param pCpu The disassembly of the write instruction.
1016	* @param pRegFrame The trap register frame.
1017	* @param GCPhysFault The fault address as guest physical address.
1018	* @param pvFault The fault address.
1019	*/
1020	DECLINLINE(int) pgmPoolAccessHandlerSimple(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
1021	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
1022	{
1023	Log3(("pgmPoolAccessHandlerSimple\n"));
1024	/*
1025	* Increment the modification counter and insert it into the list
1026	* of modified pages the first time.
1027	*/
1028	if (!pPage->cModifications++)
1029	pgmPoolMonitorModifiedInsert(pPool, pPage);
1030
1031	/*
1032	* Clear all the pages. ASSUMES that pvFault is readable.
1033	*/
1034	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
1035	PVMCPU pVCpu = VMMGetCpu(pPool->CTX_SUFF(pVM));
1036	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
1037	pgmPoolMonitorChainChanging(pPool, pPage, GCPhysFault, pvFault, pCpu);
1038	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
1039	#else
1040	pgmPoolMonitorChainChanging(pPool, pPage, GCPhysFault, pvFault, pCpu);
1041	#endif
1042
1043	/*
1044	* Interpret the instruction.
1045	*/
1046	uint32_t cb;
1047	int rc = EMInterpretInstructionCPU(pVM, pCpu, pRegFrame, pvFault, &cb);
1048	if (RT_SUCCESS(rc))
1049	pRegFrame->rip += pCpu->opsize;
1050	else if (rc == VERR_EM_INTERPRETER)
1051	{
1052	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for %04x:%RGv - opcode=%d\n",
1053	pRegFrame->cs, (RTGCPTR)pRegFrame->rip, pCpu->pCurInstr->opcode));
1054	rc = VINF_EM_RAW_EMULATE_INSTR;
1055	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
1056	}
1057
1058	#ifdef IN_RC
1059	/*
1060	* Quick hack, with logging enabled we're getting stale
1061	* code TLBs but no data TLB for EIP and crash in EMInterpretDisasOne.
1062	* Flushing here is BAD and expensive, I think EMInterpretDisasOne will
1063	* have to be fixed to support this. But that'll have to wait till next week.
1064	*
1065	* An alternative is to keep track of the changed PTEs together with the
1066	* GCPhys from the guest PT. This may proove expensive though.
1067	*
1068	* At the moment, it's VITAL that it's done AFTER the instruction interpreting
1069	* because we need the stale TLBs in some cases (XP boot). This MUST be fixed properly!
1070	*/
1071	PGM_INVL_GUEST_TLBS();
1072	#endif
1073
1074	LogFlow(("pgmPoolAccessHandlerSimple: returns %Rrc cb=%d\n", rc, cb));
1075	return rc;
1076	}
1077
1078	/**
1079	* \#PF Handler callback for PT write accesses.
1080	*
1081	* @returns VBox status code (appropriate for GC return).
1082	* @param pVM VM Handle.
1083	* @param uErrorCode CPU Error code.
1084	* @param pRegFrame Trap register frame.
1085	* NULL on DMA and other non CPU access.
1086	* @param pvFault The fault address (cr2).
1087	* @param GCPhysFault The GC physical address corresponding to pvFault.
1088	* @param pvUser User argument.
1089	*/
1090	DECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser)
1091	{
1092	STAM_PROFILE_START(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), a);
1093	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1094	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
1095	LogFlow(("pgmPoolAccessHandler: pvFault=%RGv pPage=%p:{.idx=%d} GCPhysFault=%RGp\n", pvFault, pPage, pPage->idx, GCPhysFault));
1096
1097	/*
1098	* We should ALWAYS have the list head as user parameter. This
1099	* is because we use that page to record the changes.
1100	*/
1101	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1102
1103	/*
1104	* Disassemble the faulting instruction.
1105	*/
1106	DISCPUSTATE Cpu;
1107	int rc = EMInterpretDisasOne(pVM, pRegFrame, &Cpu, NULL);
1108	AssertRCReturn(rc, rc);
1109
1110	/*
1111	* Check if it's worth dealing with.
1112	*/
1113	bool fReused = false;
1114	if ( ( pPage->cModifications < 48 /** @todo #define / /* @todo need to check that it's not mapping EIP. / /* @todo adjust this! */
1115	\|\| pgmPoolIsPageLocked(&pVM->pgm.s, pPage)
1116	)
1117	&& !(fReused = pgmPoolMonitorIsReused(pVM, pPage, pRegFrame, &Cpu, pvFault))
1118	&& !pgmPoolMonitorIsForking(pPool, &Cpu, GCPhysFault & PAGE_OFFSET_MASK))
1119	{
1120	/*
1121	* Simple instructions, no REP prefix.
1122	*/
1123	if (!(Cpu.prefix & (PREFIX_REP \| PREFIX_REPNE)))
1124	{
1125	rc = pgmPoolAccessHandlerSimple(pVM, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
1126	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,Handled), a);
1127	return rc;
1128	}
1129
1130	/*
1131	* Windows is frequently doing small memset() operations (netio test 4k+).
1132	* We have to deal with these or we'll kill the cache and performance.
1133	*/
1134	if ( Cpu.pCurInstr->opcode == OP_STOSWD
1135	&& CPUMGetGuestCPL(pVM, pRegFrame) == 0
1136	&& pRegFrame->ecx <= 0x20
1137	&& pRegFrame->ecx * 4 <= PAGE_SIZE - ((uintptr_t)pvFault & PAGE_OFFSET_MASK)
1138	&& !((uintptr_t)pvFault & 3)
1139	&& (pRegFrame->eax == 0 \|\| pRegFrame->eax == 0x80) /* the two values observed. */
1140	&& Cpu.mode == CPUMODE_32BIT
1141	&& Cpu.opmode == CPUMODE_32BIT
1142	&& Cpu.addrmode == CPUMODE_32BIT
1143	&& Cpu.prefix == PREFIX_REP
1144	&& !pRegFrame->eflags.Bits.u1DF
1145	)
1146	{
1147	rc = pgmPoolAccessHandlerSTOSD(pVM, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
1148	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,RepStosd), a);
1149	return rc;
1150	}
1151
1152	/* REP prefix, don't bother. */
1153	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,RepPrefix));
1154	Log4(("pgmPoolAccessHandler: eax=%#x ecx=%#x edi=%#x esi=%#x rip=%RGv opcode=%d prefix=%#x\n",
1155	pRegFrame->eax, pRegFrame->ecx, pRegFrame->edi, pRegFrame->esi, (RTGCPTR)pRegFrame->rip, Cpu.pCurInstr->opcode, Cpu.prefix));
1156	}
1157
1158	/*
1159	* Not worth it, so flush it.
1160	*
1161	* If we considered it to be reused, don't to back to ring-3
1162	* to emulate failed instructions since we usually cannot
1163	* interpret then. This may be a bit risky, in which case
1164	* the reuse detection must be fixed.
1165	*/
1166	rc = pgmPoolAccessHandlerFlush(pVM, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
1167	if (rc == VINF_EM_RAW_EMULATE_INSTR && fReused)
1168	rc = VINF_SUCCESS;
1169	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,FlushPage), a);
1170	return rc;
1171	}
1172
1173	# endif /* !IN_RING3 */
1174	#endif /* PGMPOOL_WITH_MONITORING */
1175
1176	#ifdef PGMPOOL_WITH_CACHE
1177
1178	/**
1179	* Inserts a page into the GCPhys hash table.
1180	*
1181	* @param pPool The pool.
1182	* @param pPage The page.
1183	*/
1184	DECLINLINE(void) pgmPoolHashInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1185	{
1186	Log3(("pgmPoolHashInsert: %RGp\n", pPage->GCPhys));
1187	Assert(pPage->GCPhys != NIL_RTGCPHYS); Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1188	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1189	pPage->iNext = pPool->aiHash[iHash];
1190	pPool->aiHash[iHash] = pPage->idx;
1191	}
1192
1193
1194	/**
1195	* Removes a page from the GCPhys hash table.
1196	*
1197	* @param pPool The pool.
1198	* @param pPage The page.
1199	*/
1200	DECLINLINE(void) pgmPoolHashRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1201	{
1202	Log3(("pgmPoolHashRemove: %RGp\n", pPage->GCPhys));
1203	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1204	if (pPool->aiHash[iHash] == pPage->idx)
1205	pPool->aiHash[iHash] = pPage->iNext;
1206	else
1207	{
1208	uint16_t iPrev = pPool->aiHash[iHash];
1209	for (;;)
1210	{
1211	const int16_t i = pPool->aPages[iPrev].iNext;
1212	if (i == pPage->idx)
1213	{
1214	pPool->aPages[iPrev].iNext = pPage->iNext;
1215	break;
1216	}
1217	if (i == NIL_PGMPOOL_IDX)
1218	{
1219	AssertReleaseMsgFailed(("GCPhys=%RGp idx=%#x\n", pPage->GCPhys, pPage->idx));
1220	break;
1221	}
1222	iPrev = i;
1223	}
1224	}
1225	pPage->iNext = NIL_PGMPOOL_IDX;
1226	}
1227
1228
1229	/**
1230	* Frees up one cache page.
1231	*
1232	* @returns VBox status code.
1233	* @retval VINF_SUCCESS on success.
1234	* @param pPool The pool.
1235	* @param iUser The user index.
1236	*/
1237	static int pgmPoolCacheFreeOne(PPGMPOOL pPool, uint16_t iUser)
1238	{
1239	#ifndef IN_RC
1240	const PVM pVM = pPool->CTX_SUFF(pVM);
1241	#endif
1242	Assert(pPool->iAgeHead != pPool->iAgeTail); /* We shouldn't be here if there < 2 cached entries! */
1243	STAM_COUNTER_INC(&pPool->StatCacheFreeUpOne);
1244
1245	/*
1246	* Select one page from the tail of the age list.
1247	*/
1248	PPGMPOOLPAGE pPage;
1249	for (unsigned iLoop = 0; ; iLoop++)
1250	{
1251	uint16_t iToFree = pPool->iAgeTail;
1252	if (iToFree == iUser)
1253	iToFree = pPool->aPages[iToFree].iAgePrev;
1254	/* This is the alternative to the SyncCR3 pgmPoolCacheUsed calls.
1255	if (pPool->aPages[iToFree].iUserHead != NIL_PGMPOOL_USER_INDEX)
1256	{
1257	uint16_t i = pPool->aPages[iToFree].iAgePrev;
1258	for (unsigned j = 0; j < 10 && i != NIL_PGMPOOL_USER_INDEX; j++, i = pPool->aPages[i].iAgePrev)
1259	{
1260	if (pPool->aPages[iToFree].iUserHead == NIL_PGMPOOL_USER_INDEX)
1261	continue;
1262	iToFree = i;
1263	break;
1264	}
1265	}
1266	*/
1267	Assert(iToFree != iUser);
1268	AssertRelease(iToFree != NIL_PGMPOOL_IDX);
1269	pPage = &pPool->aPages[iToFree];
1270
1271	/*
1272	* Reject any attempts at flushing the currently active shadow CR3 mapping.
1273	* Call pgmPoolCacheUsed to move the page to the head of the age list.
1274	*/
1275	if (!pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage))
1276	break;
1277	LogFlow(("pgmPoolCacheFreeOne: refuse CR3 mapping\n"));
1278	pgmPoolCacheUsed(pPool, pPage);
1279	AssertLogRelReturn(iLoop < 8192, VERR_INTERNAL_ERROR);
1280	}
1281
1282	/*
1283	* Found a usable page, flush it and return.
1284	*/
1285	int rc = pgmPoolFlushPage(pPool, pPage);
1286	if (rc == VINF_SUCCESS)
1287	PGM_INVL_GUEST_TLBS(); /* see PT handler. */
1288	return rc;
1289	}
1290
1291
1292	/**
1293	* Checks if a kind mismatch is really a page being reused
1294	* or if it's just normal remappings.
1295	*
1296	* @returns true if reused and the cached page (enmKind1) should be flushed
1297	* @returns false if not reused.
1298	* @param enmKind1 The kind of the cached page.
1299	* @param enmKind2 The kind of the requested page.
1300	*/
1301	static bool pgmPoolCacheReusedByKind(PGMPOOLKIND enmKind1, PGMPOOLKIND enmKind2)
1302	{
1303	switch (enmKind1)
1304	{
1305	/*
1306	* Never reuse them. There is no remapping in non-paging mode.
1307	*/
1308	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1309	case PGMPOOLKIND_32BIT_PD_PHYS:
1310	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1311	case PGMPOOLKIND_PAE_PD_PHYS:
1312	case PGMPOOLKIND_PAE_PDPT_PHYS:
1313	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1314	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1315	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1316	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1317	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1318	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT: /* never reuse them for other types */
1319	return false;
1320
1321	/*
1322	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1323	*/
1324	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1325	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1326	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1327	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1328	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1329	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1330	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1331	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1332	case PGMPOOLKIND_32BIT_PD:
1333	case PGMPOOLKIND_PAE_PDPT:
1334	switch (enmKind2)
1335	{
1336	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1337	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1338	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1339	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1340	case PGMPOOLKIND_64BIT_PML4:
1341	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1342	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1343	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1344	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1345	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1346	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1347	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1348	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1349	return true;
1350	default:
1351	return false;
1352	}
1353
1354	/*
1355	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1356	*/
1357	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1358	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1359	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1360	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1361	case PGMPOOLKIND_64BIT_PML4:
1362	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1363	switch (enmKind2)
1364	{
1365	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1366	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1367	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1368	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1369	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1370	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1371	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1372	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1373	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1374	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1375	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1376	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1377	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1378	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1379	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1380	return true;
1381	default:
1382	return false;
1383	}
1384
1385	/*
1386	* These cannot be flushed, and it's common to reuse the PDs as PTs.
1387	*/
1388	case PGMPOOLKIND_ROOT_NESTED:
1389	return false;
1390
1391	default:
1392	AssertFatalMsgFailed(("enmKind1=%d\n", enmKind1));
1393	}
1394	}
1395
1396
1397	/**
1398	* Attempts to satisfy a pgmPoolAlloc request from the cache.
1399	*
1400	* @returns VBox status code.
1401	* @retval VINF_PGM_CACHED_PAGE on success.
1402	* @retval VERR_FILE_NOT_FOUND if not found.
1403	* @param pPool The pool.
1404	* @param GCPhys The GC physical address of the page we're gonna shadow.
1405	* @param enmKind The kind of mapping.
1406	* @param iUser The shadow page pool index of the user table.
1407	* @param iUserTable The index into the user table (shadowed).
1408	* @param ppPage Where to store the pointer to the page.
1409	*/
1410	static int pgmPoolCacheAlloc(PPGMPOOL pPool, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
1411	{
1412	#ifndef IN_RC
1413	const PVM pVM = pPool->CTX_SUFF(pVM);
1414	#endif
1415	/*
1416	* Look up the GCPhys in the hash.
1417	*/
1418	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
1419	Log3(("pgmPoolCacheAlloc: %RGp kind %s iUser=%x iUserTable=%x SLOT=%d\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable, i));
1420	if (i != NIL_PGMPOOL_IDX)
1421	{
1422	do
1423	{
1424	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1425	Log4(("pgmPoolCacheAlloc: slot %d found page %RGp\n", i, pPage->GCPhys));
1426	if (pPage->GCPhys == GCPhys)
1427	{
1428	if ((PGMPOOLKIND)pPage->enmKind == enmKind)
1429	{
1430	/* Put it at the start of the use list to make sure pgmPoolTrackAddUser
1431	* doesn't flush it in case there are no more free use records.
1432	*/
1433	pgmPoolCacheUsed(pPool, pPage);
1434
1435	int rc = pgmPoolTrackAddUser(pPool, pPage, iUser, iUserTable);
1436	if (RT_SUCCESS(rc))
1437	{
1438	Assert((PGMPOOLKIND)pPage->enmKind == enmKind);
1439	*ppPage = pPage;
1440	STAM_COUNTER_INC(&pPool->StatCacheHits);
1441	return VINF_PGM_CACHED_PAGE;
1442	}
1443	return rc;
1444	}
1445
1446	/*
1447	* The kind is different. In some cases we should now flush the page
1448	* as it has been reused, but in most cases this is normal remapping
1449	* of PDs as PT or big pages using the GCPhys field in a slightly
1450	* different way than the other kinds.
1451	*/
1452	if (pgmPoolCacheReusedByKind((PGMPOOLKIND)pPage->enmKind, enmKind))
1453	{
1454	STAM_COUNTER_INC(&pPool->StatCacheKindMismatches);
1455	pgmPoolFlushPage(pPool, pPage);
1456	PGM_INVL_GUEST_TLBS(); /* see PT handler. */
1457	break;
1458	}
1459	}
1460
1461	/* next */
1462	i = pPage->iNext;
1463	} while (i != NIL_PGMPOOL_IDX);
1464	}
1465
1466	Log3(("pgmPoolCacheAlloc: Missed GCPhys=%RGp enmKind=%s\n", GCPhys, pgmPoolPoolKindToStr(enmKind)));
1467	STAM_COUNTER_INC(&pPool->StatCacheMisses);
1468	return VERR_FILE_NOT_FOUND;
1469	}
1470
1471
1472	/**
1473	* Inserts a page into the cache.
1474	*
1475	* @param pPool The pool.
1476	* @param pPage The cached page.
1477	* @param fCanBeCached Set if the page is fit for caching from the caller's point of view.
1478	*/
1479	static void pgmPoolCacheInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fCanBeCached)
1480	{
1481	/*
1482	* Insert into the GCPhys hash if the page is fit for that.
1483	*/
1484	Assert(!pPage->fCached);
1485	if (fCanBeCached)
1486	{
1487	pPage->fCached = true;
1488	pgmPoolHashInsert(pPool, pPage);
1489	Log3(("pgmPoolCacheInsert: Caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
1490	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
1491	STAM_COUNTER_INC(&pPool->StatCacheCacheable);
1492	}
1493	else
1494	{
1495	Log3(("pgmPoolCacheInsert: Not caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
1496	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
1497	STAM_COUNTER_INC(&pPool->StatCacheUncacheable);
1498	}
1499
1500	/*
1501	* Insert at the head of the age list.
1502	*/
1503	pPage->iAgePrev = NIL_PGMPOOL_IDX;
1504	pPage->iAgeNext = pPool->iAgeHead;
1505	if (pPool->iAgeHead != NIL_PGMPOOL_IDX)
1506	pPool->aPages[pPool->iAgeHead].iAgePrev = pPage->idx;
1507	else
1508	pPool->iAgeTail = pPage->idx;
1509	pPool->iAgeHead = pPage->idx;
1510	}
1511
1512
1513	/**
1514	* Flushes a cached page.
1515	*
1516	* @param pPool The pool.
1517	* @param pPage The cached page.
1518	*/
1519	static void pgmPoolCacheFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1520	{
1521	Log3(("pgmPoolCacheFlushPage: %RGp\n", pPage->GCPhys));
1522
1523	/*
1524	* Remove the page from the hash.
1525	*/
1526	if (pPage->fCached)
1527	{
1528	pPage->fCached = false;
1529	pgmPoolHashRemove(pPool, pPage);
1530	}
1531	else
1532	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1533
1534	/*
1535	* Remove it from the age list.
1536	*/
1537	if (pPage->iAgeNext != NIL_PGMPOOL_IDX)
1538	pPool->aPages[pPage->iAgeNext].iAgePrev = pPage->iAgePrev;
1539	else
1540	pPool->iAgeTail = pPage->iAgePrev;
1541	if (pPage->iAgePrev != NIL_PGMPOOL_IDX)
1542	pPool->aPages[pPage->iAgePrev].iAgeNext = pPage->iAgeNext;
1543	else
1544	pPool->iAgeHead = pPage->iAgeNext;
1545	pPage->iAgeNext = NIL_PGMPOOL_IDX;
1546	pPage->iAgePrev = NIL_PGMPOOL_IDX;
1547	}
1548
1549	#endif /* PGMPOOL_WITH_CACHE */
1550	#ifdef PGMPOOL_WITH_MONITORING
1551
1552	/**
1553	* Looks for pages sharing the monitor.
1554	*
1555	* @returns Pointer to the head page.
1556	* @returns NULL if not found.
1557	* @param pPool The Pool
1558	* @param pNewPage The page which is going to be monitored.
1559	*/
1560	static PPGMPOOLPAGE pgmPoolMonitorGetPageByGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pNewPage)
1561	{
1562	#ifdef PGMPOOL_WITH_CACHE
1563	/*
1564	* Look up the GCPhys in the hash.
1565	*/
1566	RTGCPHYS GCPhys = pNewPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1567	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
1568	if (i == NIL_PGMPOOL_IDX)
1569	return NULL;
1570	do
1571	{
1572	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1573	if ( pPage->GCPhys - GCPhys < PAGE_SIZE
1574	&& pPage != pNewPage)
1575	{
1576	switch (pPage->enmKind)
1577	{
1578	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1579	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1580	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1581	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1582	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1583	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1584	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1585	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1586	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1587	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1588	case PGMPOOLKIND_64BIT_PML4:
1589	case PGMPOOLKIND_32BIT_PD:
1590	case PGMPOOLKIND_PAE_PDPT:
1591	{
1592	/* find the head */
1593	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1594	{
1595	Assert(pPage->iMonitoredPrev != pPage->idx);
1596	pPage = &pPool->aPages[pPage->iMonitoredPrev];
1597	}
1598	return pPage;
1599	}
1600
1601	/* ignore, no monitoring. */
1602	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1603	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1604	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1605	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1606	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1607	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1608	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1609	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1610	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1611	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1612	case PGMPOOLKIND_ROOT_NESTED:
1613	case PGMPOOLKIND_PAE_PD_PHYS:
1614	case PGMPOOLKIND_PAE_PDPT_PHYS:
1615	case PGMPOOLKIND_32BIT_PD_PHYS:
1616	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
1617	break;
1618	default:
1619	AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
1620	}
1621	}
1622
1623	/* next */
1624	i = pPage->iNext;
1625	} while (i != NIL_PGMPOOL_IDX);
1626	#endif
1627	return NULL;
1628	}
1629
1630
1631	/**
1632	* Enabled write monitoring of a guest page.
1633	*
1634	* @returns VBox status code.
1635	* @retval VINF_SUCCESS on success.
1636	* @param pPool The pool.
1637	* @param pPage The cached page.
1638	*/
1639	static int pgmPoolMonitorInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1640	{
1641	LogFlow(("pgmPoolMonitorInsert %RGp\n", pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1)));
1642
1643	/*
1644	* Filter out the relevant kinds.
1645	*/
1646	switch (pPage->enmKind)
1647	{
1648	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1649	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1650	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1651	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1652	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1653	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1654	case PGMPOOLKIND_64BIT_PML4:
1655	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1656	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1657	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1658	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1659	case PGMPOOLKIND_32BIT_PD:
1660	case PGMPOOLKIND_PAE_PDPT:
1661	break;
1662
1663	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1664	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1665	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1666	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1667	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1668	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1669	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1670	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1671	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1672	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1673	case PGMPOOLKIND_ROOT_NESTED:
1674	/* Nothing to monitor here. */
1675	return VINF_SUCCESS;
1676
1677	case PGMPOOLKIND_32BIT_PD_PHYS:
1678	case PGMPOOLKIND_PAE_PDPT_PHYS:
1679	case PGMPOOLKIND_PAE_PD_PHYS:
1680	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
1681	/* Nothing to monitor here. */
1682	return VINF_SUCCESS;
1683	#ifdef PGMPOOL_WITH_MIXED_PT_CR3
1684	break;
1685	#else
1686	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1687	#endif
1688	default:
1689	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
1690	}
1691
1692	/*
1693	* Install handler.
1694	*/
1695	int rc;
1696	PPGMPOOLPAGE pPageHead = pgmPoolMonitorGetPageByGCPhys(pPool, pPage);
1697	if (pPageHead)
1698	{
1699	Assert(pPageHead != pPage); Assert(pPageHead->iMonitoredNext != pPage->idx);
1700	Assert(pPageHead->iMonitoredPrev != pPage->idx);
1701	pPage->iMonitoredPrev = pPageHead->idx;
1702	pPage->iMonitoredNext = pPageHead->iMonitoredNext;
1703	if (pPageHead->iMonitoredNext != NIL_PGMPOOL_IDX)
1704	pPool->aPages[pPageHead->iMonitoredNext].iMonitoredPrev = pPage->idx;
1705	pPageHead->iMonitoredNext = pPage->idx;
1706	rc = VINF_SUCCESS;
1707	}
1708	else
1709	{
1710	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX); Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1711	PVM pVM = pPool->CTX_SUFF(pVM);
1712	const RTGCPHYS GCPhysPage = pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1713	rc = PGMHandlerPhysicalRegisterEx(pVM, PGMPHYSHANDLERTYPE_PHYSICAL_WRITE,
1714	GCPhysPage, GCPhysPage + (PAGE_SIZE - 1),
1715	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
1716	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
1717	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
1718	pPool->pszAccessHandler);
1719	/** @todo we should probably deal with out-of-memory conditions here, but for now increasing
1720	* the heap size should suffice. */
1721	AssertFatalRC(rc);
1722	Assert(!(pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VM_FF_ISSET(pVM, VM_FF_PGM_SYNC_CR3));
1723	}
1724	pPage->fMonitored = true;
1725	return rc;
1726	}
1727
1728
1729	/**
1730	* Disables write monitoring of a guest page.
1731	*
1732	* @returns VBox status code.
1733	* @retval VINF_SUCCESS on success.
1734	* @param pPool The pool.
1735	* @param pPage The cached page.
1736	*/
1737	static int pgmPoolMonitorFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1738	{
1739	/*
1740	* Filter out the relevant kinds.
1741	*/
1742	switch (pPage->enmKind)
1743	{
1744	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1745	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1746	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1747	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1748	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1749	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1750	case PGMPOOLKIND_64BIT_PML4:
1751	case PGMPOOLKIND_32BIT_PD:
1752	case PGMPOOLKIND_PAE_PDPT:
1753	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1754	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1755	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1756	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1757	break;
1758
1759	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1760	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1761	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1762	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1763	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1764	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1765	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1766	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1767	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1768	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1769	case PGMPOOLKIND_ROOT_NESTED:
1770	case PGMPOOLKIND_PAE_PD_PHYS:
1771	case PGMPOOLKIND_PAE_PDPT_PHYS:
1772	case PGMPOOLKIND_32BIT_PD_PHYS:
1773	/* Nothing to monitor here. */
1774	return VINF_SUCCESS;
1775
1776	#ifdef PGMPOOL_WITH_MIXED_PT_CR3
1777	break;
1778	#endif
1779	default:
1780	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
1781	}
1782
1783	/*
1784	* Remove the page from the monitored list or uninstall it if last.
1785	*/
1786	const PVM pVM = pPool->CTX_SUFF(pVM);
1787	int rc;
1788	if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
1789	\|\| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1790	{
1791	if (pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
1792	{
1793	PPGMPOOLPAGE pNewHead = &pPool->aPages[pPage->iMonitoredNext];
1794	pNewHead->iMonitoredPrev = NIL_PGMPOOL_IDX;
1795	rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
1796	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pNewHead),
1797	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pNewHead),
1798	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pNewHead),
1799	pPool->pszAccessHandler);
1800	AssertFatalRCSuccess(rc);
1801	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
1802	}
1803	else
1804	{
1805	pPool->aPages[pPage->iMonitoredPrev].iMonitoredNext = pPage->iMonitoredNext;
1806	if (pPage->iMonitoredNext != NIL_PGMPOOL_IDX)
1807	{
1808	pPool->aPages[pPage->iMonitoredNext].iMonitoredPrev = pPage->iMonitoredPrev;
1809	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
1810	}
1811	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
1812	rc = VINF_SUCCESS;
1813	}
1814	}
1815	else
1816	{
1817	rc = PGMHandlerPhysicalDeregister(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1));
1818	AssertFatalRC(rc);
1819	AssertMsg(!(pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VM_FF_ISSET(pVM, VM_FF_PGM_SYNC_CR3),
1820	("%#x %#x\n", pVM->pgm.s.fSyncFlags, pVM->fForcedActions));
1821	}
1822	pPage->fMonitored = false;
1823
1824	/*
1825	* Remove it from the list of modified pages (if in it).
1826	*/
1827	pgmPoolMonitorModifiedRemove(pPool, pPage);
1828
1829	return rc;
1830	}
1831
1832
1833	/**
1834	* Inserts the page into the list of modified pages.
1835	*
1836	* @param pPool The pool.
1837	* @param pPage The page.
1838	*/
1839	void pgmPoolMonitorModifiedInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1840	{
1841	Log3(("pgmPoolMonitorModifiedInsert: idx=%d\n", pPage->idx));
1842	AssertMsg( pPage->iModifiedNext == NIL_PGMPOOL_IDX
1843	&& pPage->iModifiedPrev == NIL_PGMPOOL_IDX
1844	&& pPool->iModifiedHead != pPage->idx,
1845	("Next=%d Prev=%d idx=%d cModifications=%d Head=%d cModifiedPages=%d\n",
1846	pPage->iModifiedNext, pPage->iModifiedPrev, pPage->idx, pPage->cModifications,
1847	pPool->iModifiedHead, pPool->cModifiedPages));
1848
1849	pPage->iModifiedNext = pPool->iModifiedHead;
1850	if (pPool->iModifiedHead != NIL_PGMPOOL_IDX)
1851	pPool->aPages[pPool->iModifiedHead].iModifiedPrev = pPage->idx;
1852	pPool->iModifiedHead = pPage->idx;
1853	pPool->cModifiedPages++;
1854	#ifdef VBOX_WITH_STATISTICS
1855	if (pPool->cModifiedPages > pPool->cModifiedPagesHigh)
1856	pPool->cModifiedPagesHigh = pPool->cModifiedPages;
1857	#endif
1858	}
1859
1860
1861	/**
1862	* Removes the page from the list of modified pages and resets the
1863	* moficiation counter.
1864	*
1865	* @param pPool The pool.
1866	* @param pPage The page which is believed to be in the list of modified pages.
1867	*/
1868	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1869	{
1870	Log3(("pgmPoolMonitorModifiedRemove: idx=%d cModifications=%d\n", pPage->idx, pPage->cModifications));
1871	if (pPool->iModifiedHead == pPage->idx)
1872	{
1873	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1874	pPool->iModifiedHead = pPage->iModifiedNext;
1875	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1876	{
1877	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = NIL_PGMPOOL_IDX;
1878	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1879	}
1880	pPool->cModifiedPages--;
1881	}
1882	else if (pPage->iModifiedPrev != NIL_PGMPOOL_IDX)
1883	{
1884	pPool->aPages[pPage->iModifiedPrev].iModifiedNext = pPage->iModifiedNext;
1885	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1886	{
1887	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = pPage->iModifiedPrev;
1888	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1889	}
1890	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1891	pPool->cModifiedPages--;
1892	}
1893	else
1894	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1895	pPage->cModifications = 0;
1896	}
1897
1898
1899	/**
1900	* Zaps the list of modified pages, resetting their modification counters in the process.
1901	*
1902	* @param pVM The VM handle.
1903	*/
1904	void pgmPoolMonitorModifiedClearAll(PVM pVM)
1905	{
1906	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1907	LogFlow(("pgmPoolMonitorModifiedClearAll: cModifiedPages=%d\n", pPool->cModifiedPages));
1908
1909	unsigned cPages = 0; NOREF(cPages);
1910	uint16_t idx = pPool->iModifiedHead;
1911	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
1912	while (idx != NIL_PGMPOOL_IDX)
1913	{
1914	PPGMPOOLPAGE pPage = &pPool->aPages[idx];
1915	idx = pPage->iModifiedNext;
1916	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1917	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1918	pPage->cModifications = 0;
1919	Assert(++cPages);
1920	}
1921	AssertMsg(cPages == pPool->cModifiedPages, ("%d != %d\n", cPages, pPool->cModifiedPages));
1922	pPool->cModifiedPages = 0;
1923	}
1924
1925
1926	#ifdef IN_RING3
1927	/**
1928	* Clear all shadow pages and clear all modification counters.
1929	*
1930	* @param pVM The VM handle.
1931	* @remark Should only be used when monitoring is available, thus placed in
1932	* the PGMPOOL_WITH_MONITORING #ifdef.
1933	*/
1934	void pgmPoolClearAll(PVM pVM)
1935	{
1936	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1937	STAM_PROFILE_START(&pPool->StatClearAll, c);
1938	LogFlow(("pgmPoolClearAll: cUsedPages=%d\n", pPool->cUsedPages));
1939
1940	/*
1941	* Iterate all the pages until we've encountered all that in use.
1942	* This is simple but not quite optimal solution.
1943	*/
1944	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
1945	unsigned cLeft = pPool->cUsedPages;
1946	unsigned iPage = pPool->cCurPages;
1947	while (--iPage >= PGMPOOL_IDX_FIRST)
1948	{
1949	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
1950	if (pPage->GCPhys != NIL_RTGCPHYS)
1951	{
1952	switch (pPage->enmKind)
1953	{
1954	/*
1955	* We only care about shadow page tables.
1956	*/
1957	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1958	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1959	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1960	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1961	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1962	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1963	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1964	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1965	{
1966	#ifdef PGMPOOL_WITH_USER_TRACKING
1967	if (pPage->cPresent)
1968	#endif
1969	{
1970	void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
1971	STAM_PROFILE_START(&pPool->StatZeroPage, z);
1972	ASMMemZeroPage(pvShw);
1973	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
1974	#ifdef PGMPOOL_WITH_USER_TRACKING
1975	pPage->cPresent = 0;
1976	pPage->iFirstPresent = ~0;
1977	#endif
1978	}
1979	}
1980	/* fall thru */
1981
1982	default:
1983	Assert(!pPage->cModifications \|\| ++cModifiedPages);
1984	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
1985	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
1986	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1987	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1988	pPage->cModifications = 0;
1989	break;
1990
1991	}
1992	if (!--cLeft)
1993	break;
1994	}
1995	}
1996
1997	/* swipe the special pages too. */
1998	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
1999	{
2000	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2001	if (pPage->GCPhys != NIL_RTGCPHYS)
2002	{
2003	Assert(!pPage->cModifications \|\| ++cModifiedPages);
2004	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2005	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2006	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2007	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2008	pPage->cModifications = 0;
2009	}
2010	}
2011
2012	#ifndef DEBUG_michael
2013	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
2014	#endif
2015	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
2016	pPool->cModifiedPages = 0;
2017
2018	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2019	/*
2020	* Clear all the GCPhys links and rebuild the phys ext free list.
2021	*/
2022	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
2023	pRam;
2024	pRam = pRam->CTX_SUFF(pNext))
2025	{
2026	unsigned iPage = pRam->cb >> PAGE_SHIFT;
2027	while (iPage-- > 0)
2028	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
2029	}
2030
2031	pPool->iPhysExtFreeHead = 0;
2032	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
2033	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
2034	for (unsigned i = 0; i < cMaxPhysExts; i++)
2035	{
2036	paPhysExts[i].iNext = i + 1;
2037	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
2038	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
2039	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
2040	}
2041	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
2042	#endif
2043
2044
2045	pPool->cPresent = 0;
2046	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
2047	}
2048	#endif /* IN_RING3 */
2049
2050
2051	/**
2052	* Handle SyncCR3 pool tasks
2053	*
2054	* @returns VBox status code.
2055	* @retval VINF_SUCCESS if successfully added.
2056	* @retval VINF_PGM_SYNC_CR3 is it needs to be deferred to ring 3 (GC only)
2057	* @param pVM The VM handle.
2058	* @remark Should only be used when monitoring is available, thus placed in
2059	* the PGMPOOL_WITH_MONITORING #ifdef.
2060	*/
2061	int pgmPoolSyncCR3(PVM pVM)
2062	{
2063	LogFlow(("pgmPoolSyncCR3\n"));
2064	/*
2065	* When monitoring shadowed pages, we reset the modification counters on CR3 sync.
2066	* Occasionally we will have to clear all the shadow page tables because we wanted
2067	* to monitor a page which was mapped by too many shadowed page tables. This operation
2068	* sometimes refered to as a 'lightweight flush'.
2069	*/
2070	if (!(pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL))
2071	pgmPoolMonitorModifiedClearAll(pVM);
2072	else
2073	{
2074	# ifdef IN_RING3 /* Don't flush in ring-0 or raw mode, it's taking too long. */
2075	pVM->pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
2076	pgmPoolClearAll(pVM);
2077	# else /* !IN_RING3 */
2078	LogFlow(("SyncCR3: PGM_SYNC_CLEAR_PGM_POOL is set -> VINF_PGM_SYNC_CR3\n"));
2079	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
2080	return VINF_PGM_SYNC_CR3;
2081	# endif /* !IN_RING3 */
2082	}
2083	return VINF_SUCCESS;
2084	}
2085
2086	#endif /* PGMPOOL_WITH_MONITORING */
2087	#ifdef PGMPOOL_WITH_USER_TRACKING
2088
2089	/**
2090	* Frees up at least one user entry.
2091	*
2092	* @returns VBox status code.
2093	* @retval VINF_SUCCESS if successfully added.
2094	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2095	* @param pPool The pool.
2096	* @param iUser The user index.
2097	*/
2098	static int pgmPoolTrackFreeOneUser(PPGMPOOL pPool, uint16_t iUser)
2099	{
2100	STAM_COUNTER_INC(&pPool->StatTrackFreeUpOneUser);
2101	#ifdef PGMPOOL_WITH_CACHE
2102	/*
2103	* Just free cached pages in a braindead fashion.
2104	*/
2105	/** @todo walk the age list backwards and free the first with usage. */
2106	int rc = VINF_SUCCESS;
2107	do
2108	{
2109	int rc2 = pgmPoolCacheFreeOne(pPool, iUser);
2110	if (RT_FAILURE(rc2) && rc == VINF_SUCCESS)
2111	rc = rc2;
2112	} while (pPool->iUserFreeHead == NIL_PGMPOOL_USER_INDEX);
2113	return rc;
2114	#else
2115	/*
2116	* Lazy approach.
2117	*/
2118	/* @todo This path no longer works (CR3 root pages will be flushed)!! */
2119	AssertCompileFailed();
2120	Assert(!CPUMIsGuestInLongMode(pVM));
2121	pgmPoolFlushAllInt(pPool);
2122	return VERR_PGM_POOL_FLUSHED;
2123	#endif
2124	}
2125
2126
2127	/**
2128	* Inserts a page into the cache.
2129	*
2130	* This will create user node for the page, insert it into the GCPhys
2131	* hash, and insert it into the age list.
2132	*
2133	* @returns VBox status code.
2134	* @retval VINF_SUCCESS if successfully added.
2135	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2136	* @param pPool The pool.
2137	* @param pPage The cached page.
2138	* @param GCPhys The GC physical address of the page we're gonna shadow.
2139	* @param iUser The user index.
2140	* @param iUserTable The user table index.
2141	*/
2142	DECLINLINE(int) pgmPoolTrackInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhys, uint16_t iUser, uint32_t iUserTable)
2143	{
2144	int rc = VINF_SUCCESS;
2145	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2146
2147	LogFlow(("pgmPoolTrackInsert GCPhys=%RGp iUser %x iUserTable %x\n", GCPhys, iUser, iUserTable));
2148
2149	#ifdef VBOX_STRICT
2150	/*
2151	* Check that the entry doesn't already exists.
2152	*/
2153	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2154	{
2155	uint16_t i = pPage->iUserHead;
2156	do
2157	{
2158	Assert(i < pPool->cMaxUsers);
2159	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2160	i = paUsers[i].iNext;
2161	} while (i != NIL_PGMPOOL_USER_INDEX);
2162	}
2163	#endif
2164
2165	/*
2166	* Find free a user node.
2167	*/
2168	uint16_t i = pPool->iUserFreeHead;
2169	if (i == NIL_PGMPOOL_USER_INDEX)
2170	{
2171	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2172	if (RT_FAILURE(rc))
2173	return rc;
2174	i = pPool->iUserFreeHead;
2175	}
2176
2177	/*
2178	* Unlink the user node from the free list,
2179	* initialize and insert it into the user list.
2180	*/
2181	pPool->iUserFreeHead = paUsers[i].iNext;
2182	paUsers[i].iNext = NIL_PGMPOOL_USER_INDEX;
2183	paUsers[i].iUser = iUser;
2184	paUsers[i].iUserTable = iUserTable;
2185	pPage->iUserHead = i;
2186
2187	/*
2188	* Insert into cache and enable monitoring of the guest page if enabled.
2189	*
2190	* Until we implement caching of all levels, including the CR3 one, we'll
2191	* have to make sure we don't try monitor & cache any recursive reuse of
2192	* a monitored CR3 page. Because all windows versions are doing this we'll
2193	* have to be able to do combined access monitoring, CR3 + PT and
2194	* PD + PT (guest PAE).
2195	*
2196	* Update:
2197	* We're now cooperating with the CR3 monitor if an uncachable page is found.
2198	*/
2199	#if defined(PGMPOOL_WITH_MONITORING) \|\| defined(PGMPOOL_WITH_CACHE)
2200	# ifdef PGMPOOL_WITH_MIXED_PT_CR3
2201	const bool fCanBeMonitored = true;
2202	# else
2203	bool fCanBeMonitored = pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored == NIL_RTGCPHYS
2204	\|\| (GCPhys & X86_PTE_PAE_PG_MASK) != (pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored & X86_PTE_PAE_PG_MASK)
2205	\|\| pgmPoolIsBigPage((PGMPOOLKIND)pPage->enmKind);
2206	# endif
2207	# ifdef PGMPOOL_WITH_CACHE
2208	pgmPoolCacheInsert(pPool, pPage, fCanBeMonitored); /* This can be expanded. */
2209	# endif
2210	if (fCanBeMonitored)
2211	{
2212	# ifdef PGMPOOL_WITH_MONITORING
2213	rc = pgmPoolMonitorInsert(pPool, pPage);
2214	AssertRC(rc);
2215	}
2216	# endif
2217	#endif /* PGMPOOL_WITH_MONITORING */
2218	return rc;
2219	}
2220
2221
2222	# ifdef PGMPOOL_WITH_CACHE /* (only used when the cache is enabled.) */
2223	/**
2224	* Adds a user reference to a page.
2225	*
2226	* This will move the page to the head of the
2227	*
2228	* @returns VBox status code.
2229	* @retval VINF_SUCCESS if successfully added.
2230	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2231	* @param pPool The pool.
2232	* @param pPage The cached page.
2233	* @param iUser The user index.
2234	* @param iUserTable The user table.
2235	*/
2236	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2237	{
2238	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2239
2240	Log3(("pgmPoolTrackAddUser GCPhys = %RGp iUser %x iUserTable %x\n", pPage->GCPhys, iUser, iUserTable));
2241	# ifdef VBOX_STRICT
2242	/*
2243	* Check that the entry doesn't already exists.
2244	*/
2245	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2246	{
2247	uint16_t i = pPage->iUserHead;
2248	do
2249	{
2250	Assert(i < pPool->cMaxUsers);
2251	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2252	i = paUsers[i].iNext;
2253	} while (i != NIL_PGMPOOL_USER_INDEX);
2254	}
2255	# endif
2256
2257	/*
2258	* Allocate a user node.
2259	*/
2260	uint16_t i = pPool->iUserFreeHead;
2261	if (i == NIL_PGMPOOL_USER_INDEX)
2262	{
2263	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2264	if (RT_FAILURE(rc))
2265	return rc;
2266	i = pPool->iUserFreeHead;
2267	}
2268	pPool->iUserFreeHead = paUsers[i].iNext;
2269
2270	/*
2271	* Initialize the user node and insert it.
2272	*/
2273	paUsers[i].iNext = pPage->iUserHead;
2274	paUsers[i].iUser = iUser;
2275	paUsers[i].iUserTable = iUserTable;
2276	pPage->iUserHead = i;
2277
2278	# ifdef PGMPOOL_WITH_CACHE
2279	/*
2280	* Tell the cache to update its replacement stats for this page.
2281	*/
2282	pgmPoolCacheUsed(pPool, pPage);
2283	# endif
2284	return VINF_SUCCESS;
2285	}
2286	# endif /* PGMPOOL_WITH_CACHE */
2287
2288
2289	/**
2290	* Frees a user record associated with a page.
2291	*
2292	* This does not clear the entry in the user table, it simply replaces the
2293	* user record to the chain of free records.
2294	*
2295	* @param pPool The pool.
2296	* @param HCPhys The HC physical address of the shadow page.
2297	* @param iUser The shadow page pool index of the user table.
2298	* @param iUserTable The index into the user table (shadowed).
2299	*/
2300	static void pgmPoolTrackFreeUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2301	{
2302	/*
2303	* Unlink and free the specified user entry.
2304	*/
2305	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2306
2307	Log3(("pgmPoolTrackFreeUser %RGp %x %x\n", pPage->GCPhys, iUser, iUserTable));
2308	/* Special: For PAE and 32-bit paging, there is usually no more than one user. */
2309	uint16_t i = pPage->iUserHead;
2310	if ( i != NIL_PGMPOOL_USER_INDEX
2311	&& paUsers[i].iUser == iUser
2312	&& paUsers[i].iUserTable == iUserTable)
2313	{
2314	pPage->iUserHead = paUsers[i].iNext;
2315
2316	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2317	paUsers[i].iNext = pPool->iUserFreeHead;
2318	pPool->iUserFreeHead = i;
2319	return;
2320	}
2321
2322	/* General: Linear search. */
2323	uint16_t iPrev = NIL_PGMPOOL_USER_INDEX;
2324	while (i != NIL_PGMPOOL_USER_INDEX)
2325	{
2326	if ( paUsers[i].iUser == iUser
2327	&& paUsers[i].iUserTable == iUserTable)
2328	{
2329	if (iPrev != NIL_PGMPOOL_USER_INDEX)
2330	paUsers[iPrev].iNext = paUsers[i].iNext;
2331	else
2332	pPage->iUserHead = paUsers[i].iNext;
2333
2334	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2335	paUsers[i].iNext = pPool->iUserFreeHead;
2336	pPool->iUserFreeHead = i;
2337	return;
2338	}
2339	iPrev = i;
2340	i = paUsers[i].iNext;
2341	}
2342
2343	/* Fatal: didn't find it */
2344	AssertFatalMsgFailed(("Didn't find the user entry! iUser=%#x iUserTable=%#x GCPhys=%RGp\n",
2345	iUser, iUserTable, pPage->GCPhys));
2346	}
2347
2348
2349	/**
2350	* Gets the entry size of a shadow table.
2351	*
2352	* @param enmKind The kind of page.
2353	*
2354	* @returns The size of the entry in bytes. That is, 4 or 8.
2355	* @returns If the kind is not for a table, an assertion is raised and 0 is
2356	* returned.
2357	*/
2358	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind)
2359	{
2360	switch (enmKind)
2361	{
2362	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2363	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2364	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2365	case PGMPOOLKIND_32BIT_PD:
2366	case PGMPOOLKIND_32BIT_PD_PHYS:
2367	return 4;
2368
2369	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2370	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2371	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2372	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2373	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2374	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2375	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2376	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2377	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2378	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2379	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2380	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2381	case PGMPOOLKIND_64BIT_PML4:
2382	case PGMPOOLKIND_PAE_PDPT:
2383	case PGMPOOLKIND_ROOT_NESTED:
2384	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2385	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2386	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2387	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2388	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2389	case PGMPOOLKIND_PAE_PD_PHYS:
2390	case PGMPOOLKIND_PAE_PDPT_PHYS:
2391	return 8;
2392
2393	default:
2394	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2395	}
2396	}
2397
2398
2399	/**
2400	* Gets the entry size of a guest table.
2401	*
2402	* @param enmKind The kind of page.
2403	*
2404	* @returns The size of the entry in bytes. That is, 0, 4 or 8.
2405	* @returns If the kind is not for a table, an assertion is raised and 0 is
2406	* returned.
2407	*/
2408	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind)
2409	{
2410	switch (enmKind)
2411	{
2412	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2413	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2414	case PGMPOOLKIND_32BIT_PD:
2415	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2416	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2417	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2418	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2419	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2420	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2421	return 4;
2422
2423	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2424	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2425	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2426	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2427	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2428	case PGMPOOLKIND_64BIT_PML4:
2429	case PGMPOOLKIND_PAE_PDPT:
2430	return 8;
2431
2432	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2433	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2434	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2435	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2436	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2437	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2438	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2439	case PGMPOOLKIND_ROOT_NESTED:
2440	case PGMPOOLKIND_PAE_PD_PHYS:
2441	case PGMPOOLKIND_PAE_PDPT_PHYS:
2442	case PGMPOOLKIND_32BIT_PD_PHYS:
2443	/** @todo can we return 0? (nobody is calling this...) */
2444	AssertFailed();
2445	return 0;
2446
2447	default:
2448	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2449	}
2450	}
2451
2452	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2453
2454	/**
2455	* Scans one shadow page table for mappings of a physical page.
2456	*
2457	* @param pVM The VM handle.
2458	* @param pPhysPage The guest page in question.
2459	* @param iShw The shadow page table.
2460	* @param cRefs The number of references made in that PT.
2461	*/
2462	static void pgmPoolTrackFlushGCPhysPTInt(PVM pVM, PCPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2463	{
2464	LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%R[pgmpage] iShw=%d cRefs=%d\n", pPhysPage, iShw, cRefs));
2465	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2466
2467	/*
2468	* Assert sanity.
2469	*/
2470	Assert(cRefs == 1);
2471	AssertFatalMsg(iShw < pPool->cCurPages && iShw != NIL_PGMPOOL_IDX, ("iShw=%d\n", iShw));
2472	PPGMPOOLPAGE pPage = &pPool->aPages[iShw];
2473
2474	/*
2475	* Then, clear the actual mappings to the page in the shadow PT.
2476	*/
2477	switch (pPage->enmKind)
2478	{
2479	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2480	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2481	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2482	{
2483	const uint32_t u32 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2484	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2485	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2486	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2487	{
2488	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX32 cRefs=%#x\n", i, pPT->a[i], cRefs));
2489	pPT->a[i].u = 0;
2490	cRefs--;
2491	if (!cRefs)
2492	return;
2493	}
2494	#ifdef LOG_ENABLED
2495	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2496	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2497	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2498	{
2499	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2500	pPT->a[i].u = 0;
2501	}
2502	#endif
2503	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2504	break;
2505	}
2506
2507	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2508	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2509	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2510	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2511	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2512	{
2513	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2514	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2515	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2516	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2517	{
2518	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2519	pPT->a[i].u = 0;
2520	cRefs--;
2521	if (!cRefs)
2522	return;
2523	}
2524	#ifdef LOG_ENABLED
2525	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2526	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2527	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2528	{
2529	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2530	pPT->a[i].u = 0;
2531	}
2532	#endif
2533	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d u64=%RX64\n", cRefs, pPage->iFirstPresent, pPage->cPresent, u64));
2534	break;
2535	}
2536
2537	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2538	{
2539	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2540	PEPTPT pPT = (PEPTPT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2541	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2542	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2543	{
2544	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2545	pPT->a[i].u = 0;
2546	cRefs--;
2547	if (!cRefs)
2548	return;
2549	}
2550	#ifdef LOG_ENABLED
2551	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2552	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2553	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2554	{
2555	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2556	pPT->a[i].u = 0;
2557	}
2558	#endif
2559	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2560	break;
2561	}
2562
2563	default:
2564	AssertFatalMsgFailed(("enmKind=%d iShw=%d\n", pPage->enmKind, iShw));
2565	}
2566	}
2567
2568
2569	/**
2570	* Scans one shadow page table for mappings of a physical page.
2571	*
2572	* @param pVM The VM handle.
2573	* @param pPhysPage The guest page in question.
2574	* @param iShw The shadow page table.
2575	* @param cRefs The number of references made in that PT.
2576	*/
2577	void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2578	{
2579	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
2580	LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%R[pgmpage] iShw=%d cRefs=%d\n", pPhysPage, iShw, cRefs));
2581	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPT, f);
2582	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, iShw, cRefs);
2583	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2584	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPT, f);
2585	}
2586
2587
2588	/**
2589	* Flushes a list of shadow page tables mapping the same physical page.
2590	*
2591	* @param pVM The VM handle.
2592	* @param pPhysPage The guest page in question.
2593	* @param iPhysExt The physical cross reference extent list to flush.
2594	*/
2595	void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, uint16_t iPhysExt)
2596	{
2597	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2598	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTs, f);
2599	LogFlow(("pgmPoolTrackFlushGCPhysPTs: pPhysPage=%R[pgmpage] iPhysExt\n", pPhysPage, iPhysExt));
2600
2601	const uint16_t iPhysExtStart = iPhysExt;
2602	PPGMPOOLPHYSEXT pPhysExt;
2603	do
2604	{
2605	Assert(iPhysExt < pPool->cMaxPhysExts);
2606	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2607	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
2608	if (pPhysExt->aidx[i] != NIL_PGMPOOL_IDX)
2609	{
2610	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, pPhysExt->aidx[i], 1);
2611	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
2612	}
2613
2614	/* next */
2615	iPhysExt = pPhysExt->iNext;
2616	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
2617
2618	/* insert the list into the free list and clear the ram range entry. */
2619	pPhysExt->iNext = pPool->iPhysExtFreeHead;
2620	pPool->iPhysExtFreeHead = iPhysExtStart;
2621	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2622
2623	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTs, f);
2624	}
2625
2626	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
2627
2628	/**
2629	* Flushes all shadow page table mappings of the given guest page.
2630	*
2631	* This is typically called when the host page backing the guest one has been
2632	* replaced or when the page protection was changed due to an access handler.
2633	*
2634	* @returns VBox status code.
2635	* @retval VINF_SUCCESS if all references has been successfully cleared.
2636	* @retval VINF_PGM_SYNC_CR3 if we're better off with a CR3 sync and a page
2637	* pool cleaning. FF and sync flags are set.
2638	*
2639	* @param pVM The VM handle.
2640	* @param pPhysPage The guest page in question.
2641	* @param pfFlushTLBs This is set to @a true if the shadow TLBs should be
2642	* flushed, it is NOT touched if this isn't necessary.
2643	* The caller MUST initialized this to @a false.
2644	*/
2645	int pgmPoolTrackFlushGCPhys(PVM pVM, PPGMPAGE pPhysPage, bool *pfFlushTLBs)
2646	{
2647	int rc = VINF_SUCCESS;
2648	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2649	const uint16_t u16 = PGM_PAGE_GET_TRACKING(pPhysPage);
2650	if (u16)
2651	{
2652	# ifdef VBOX_WITH_NEW_PHYS_CODE
2653	/*
2654	* The zero page is currently screwing up the tracking and we'll
2655	* have to flush the whole shebang. Unless VBOX_WITH_NEW_LAZY_PAGE_ALLOC
2656	* is defined, zero pages won't normally be mapped. Some kind of solution
2657	* will be needed for this problem of course, but it will have to wait...
2658	*/
2659	if (PGM_PAGE_IS_ZERO(pPhysPage))
2660	rc = VINF_PGM_GCPHYS_ALIASED;
2661	else
2662	# endif
2663	{
2664	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2665	/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow and
2666	pgmPoolTrackFlushGCPhysPTs will/may kill the pool otherwise. */
2667	PVMCPU pVCpu = VMMGetCpu(pVM);
2668	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
2669	# endif
2670
2671	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
2672	pgmPoolTrackFlushGCPhysPT(pVM,
2673	pPhysPage,
2674	PGMPOOL_TD_GET_IDX(u16),
2675	PGMPOOL_TD_GET_CREFS(u16));
2676	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
2677	pgmPoolTrackFlushGCPhysPTs(pVM, pPhysPage, PGMPOOL_TD_GET_IDX(u16));
2678	else
2679	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
2680	*pfFlushTLBs = true;
2681
2682	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2683	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
2684	# endif
2685	}
2686	}
2687
2688	#elif defined(PGMPOOL_WITH_CACHE)
2689	# ifdef VBOX_WITH_NEW_PHYS_CODE
2690	if (PGM_PAGE_IS_ZERO(pPhysPage))
2691	rc = VINF_PGM_GCPHYS_ALIASED;
2692	else
2693	# endif
2694	{
2695	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2696	/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow kill the pool otherwise. */
2697	PVMCPU pVCpu = VMMGetCpu(pVM);
2698	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
2699	# endif
2700	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
2701	if (rc == VINF_SUCCESS)
2702	*pfFlushTLBs = true;
2703	}
2704
2705	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2706	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
2707	# endif
2708
2709	#else
2710	rc = VINF_PGM_GCPHYS_ALIASED;
2711	#endif
2712
2713	if (rc == VINF_PGM_GCPHYS_ALIASED)
2714	{
2715	pVM->pgm.s.fSyncFlags \|= PGM_SYNC_CLEAR_PGM_POOL;
2716	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3);
2717	rc = VINF_PGM_SYNC_CR3;
2718	}
2719
2720	return rc;
2721	}
2722
2723
2724	/**
2725	* Scans all shadow page tables for mappings of a physical page.
2726	*
2727	* This may be slow, but it's most likely more efficient than cleaning
2728	* out the entire page pool / cache.
2729	*
2730	* @returns VBox status code.
2731	* @retval VINF_SUCCESS if all references has been successfully cleared.
2732	* @retval VINF_PGM_GCPHYS_ALIASED if we're better off with a CR3 sync and
2733	* a page pool cleaning.
2734	*
2735	* @param pVM The VM handle.
2736	* @param pPhysPage The guest page in question.
2737	*/
2738	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage)
2739	{
2740	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2741	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2742	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: cUsedPages=%d cPresent=%d pPhysPage=%R[pgmpage]\n",
2743	pPool->cUsedPages, pPool->cPresent, pPhysPage));
2744
2745	#if 1
2746	/*
2747	* There is a limit to what makes sense.
2748	*/
2749	if (pPool->cPresent > 1024)
2750	{
2751	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
2752	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2753	return VINF_PGM_GCPHYS_ALIASED;
2754	}
2755	#endif
2756
2757	/*
2758	* Iterate all the pages until we've encountered all that in use.
2759	* This is simple but not quite optimal solution.
2760	*/
2761	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2762	const uint32_t u32 = u64;
2763	unsigned cLeft = pPool->cUsedPages;
2764	unsigned iPage = pPool->cCurPages;
2765	while (--iPage >= PGMPOOL_IDX_FIRST)
2766	{
2767	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2768	if (pPage->GCPhys != NIL_RTGCPHYS)
2769	{
2770	switch (pPage->enmKind)
2771	{
2772	/*
2773	* We only care about shadow page tables.
2774	*/
2775	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2776	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2777	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2778	{
2779	unsigned cPresent = pPage->cPresent;
2780	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2781	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2782	if (pPT->a[i].n.u1Present)
2783	{
2784	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2785	{
2786	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX32\n", iPage, i, pPT->a[i]));
2787	pPT->a[i].u = 0;
2788	}
2789	if (!--cPresent)
2790	break;
2791	}
2792	break;
2793	}
2794
2795	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2796	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2797	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2798	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2799	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2800	{
2801	unsigned cPresent = pPage->cPresent;
2802	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2803	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2804	if (pPT->a[i].n.u1Present)
2805	{
2806	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2807	{
2808	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
2809	pPT->a[i].u = 0;
2810	}
2811	if (!--cPresent)
2812	break;
2813	}
2814	break;
2815	}
2816	}
2817	if (!--cLeft)
2818	break;
2819	}
2820	}
2821
2822	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2823	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2824	return VINF_SUCCESS;
2825	}
2826
2827
2828	/**
2829	* Clears the user entry in a user table.
2830	*
2831	* This is used to remove all references to a page when flushing it.
2832	*/
2833	static void pgmPoolTrackClearPageUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PCPGMPOOLUSER pUser)
2834	{
2835	Assert(pUser->iUser != NIL_PGMPOOL_IDX);
2836	Assert(pUser->iUser < pPool->cCurPages);
2837	uint32_t iUserTable = pUser->iUserTable;
2838
2839	/*
2840	* Map the user page.
2841	*/
2842	PPGMPOOLPAGE pUserPage = &pPool->aPages[pUser->iUser];
2843	union
2844	{
2845	uint64_t *pau64;
2846	uint32_t *pau32;
2847	} u;
2848	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pUserPage);
2849
2850	LogFlow(("pgmPoolTrackClearPageUser: clear %x in %s (%RGp) (flushing %s)\n", iUserTable, pgmPoolPoolKindToStr(pUserPage->enmKind), pUserPage->Core.Key, pgmPoolPoolKindToStr(pPage->enmKind)));
2851
2852	/* Safety precaution in case we change the paging for other modes too in the future. */
2853	Assert(!pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage));
2854
2855	#ifdef VBOX_STRICT
2856	/*
2857	* Some sanity checks.
2858	*/
2859	switch (pUserPage->enmKind)
2860	{
2861	case PGMPOOLKIND_32BIT_PD:
2862	case PGMPOOLKIND_32BIT_PD_PHYS:
2863	Assert(iUserTable < X86_PG_ENTRIES);
2864	break;
2865	case PGMPOOLKIND_PAE_PDPT:
2866	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2867	case PGMPOOLKIND_PAE_PDPT_PHYS:
2868	Assert(iUserTable < 4);
2869	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2870	break;
2871	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2872	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2873	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2874	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2875	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2876	case PGMPOOLKIND_PAE_PD_PHYS:
2877	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2878	break;
2879	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2880	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2881	Assert(!(u.pau64[iUserTable] & PGM_PDFLAGS_MAPPING));
2882	break;
2883	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2884	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2885	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2886	break;
2887	case PGMPOOLKIND_64BIT_PML4:
2888	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2889	/* GCPhys >> PAGE_SHIFT is the index here */
2890	break;
2891	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2892	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2893	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2894	break;
2895
2896	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2897	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2898	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2899	break;
2900
2901	case PGMPOOLKIND_ROOT_NESTED:
2902	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2903	break;
2904
2905	default:
2906	AssertMsgFailed(("enmKind=%d\n", pUserPage->enmKind));
2907	break;
2908	}
2909	#endif /* VBOX_STRICT */
2910
2911	/*
2912	* Clear the entry in the user page.
2913	*/
2914	switch (pUserPage->enmKind)
2915	{
2916	/* 32-bit entries */
2917	case PGMPOOLKIND_32BIT_PD:
2918	case PGMPOOLKIND_32BIT_PD_PHYS:
2919	u.pau32[iUserTable] = 0;
2920	break;
2921
2922	/* 64-bit entries */
2923	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2924	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2925	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2926	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2927	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2928	#if defined(IN_RC)
2929	/* 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
2930	* non-present PDPT will continue to cause page faults.
2931	*/
2932	ASMReloadCR3();
2933	#endif
2934	/* no break */
2935	case PGMPOOLKIND_PAE_PD_PHYS:
2936	case PGMPOOLKIND_PAE_PDPT_PHYS:
2937	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2938	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2939	case PGMPOOLKIND_64BIT_PML4:
2940	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2941	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2942	case PGMPOOLKIND_PAE_PDPT:
2943	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2944	case PGMPOOLKIND_ROOT_NESTED:
2945	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2946	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2947	u.pau64[iUserTable] = 0;
2948	break;
2949
2950	default:
2951	AssertFatalMsgFailed(("enmKind=%d iUser=%#x iUserTable=%#x\n", pUserPage->enmKind, pUser->iUser, pUser->iUserTable));
2952	}
2953	}
2954
2955
2956	/**
2957	* Clears all users of a page.
2958	*/
2959	static void pgmPoolTrackClearPageUsers(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2960	{
2961	/*
2962	* Free all the user records.
2963	*/
2964	LogFlow(("pgmPoolTrackClearPageUsers %RGp\n", pPage->GCPhys));
2965
2966	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2967	uint16_t i = pPage->iUserHead;
2968	while (i != NIL_PGMPOOL_USER_INDEX)
2969	{
2970	/* Clear enter in user table. */
2971	pgmPoolTrackClearPageUser(pPool, pPage, &paUsers[i]);
2972
2973	/* Free it. */
2974	const uint16_t iNext = paUsers[i].iNext;
2975	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2976	paUsers[i].iNext = pPool->iUserFreeHead;
2977	pPool->iUserFreeHead = i;
2978
2979	/* Next. */
2980	i = iNext;
2981	}
2982	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
2983	}
2984
2985	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2986
2987	/**
2988	* Allocates a new physical cross reference extent.
2989	*
2990	* @returns Pointer to the allocated extent on success. NULL if we're out of them.
2991	* @param pVM The VM handle.
2992	* @param piPhysExt Where to store the phys ext index.
2993	*/
2994	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt)
2995	{
2996	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2997	uint16_t iPhysExt = pPool->iPhysExtFreeHead;
2998	if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
2999	{
3000	STAM_COUNTER_INC(&pPool->StamTrackPhysExtAllocFailures);
3001	return NULL;
3002	}
3003	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3004	pPool->iPhysExtFreeHead = pPhysExt->iNext;
3005	pPhysExt->iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
3006	*piPhysExt = iPhysExt;
3007	return pPhysExt;
3008	}
3009
3010
3011	/**
3012	* Frees a physical cross reference extent.
3013	*
3014	* @param pVM The VM handle.
3015	* @param iPhysExt The extent to free.
3016	*/
3017	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt)
3018	{
3019	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3020	Assert(iPhysExt < pPool->cMaxPhysExts);
3021	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3022	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3023	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3024	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3025	pPool->iPhysExtFreeHead = iPhysExt;
3026	}
3027
3028
3029	/**
3030	* Frees a physical cross reference extent.
3031	*
3032	* @param pVM The VM handle.
3033	* @param iPhysExt The extent to free.
3034	*/
3035	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt)
3036	{
3037	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3038
3039	const uint16_t iPhysExtStart = iPhysExt;
3040	PPGMPOOLPHYSEXT pPhysExt;
3041	do
3042	{
3043	Assert(iPhysExt < pPool->cMaxPhysExts);
3044	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3045	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3046	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3047
3048	/* next */
3049	iPhysExt = pPhysExt->iNext;
3050	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3051
3052	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3053	pPool->iPhysExtFreeHead = iPhysExtStart;
3054	}
3055
3056
3057	/**
3058	* Insert a reference into a list of physical cross reference extents.
3059	*
3060	* @returns The new tracking data for PGMPAGE.
3061	*
3062	* @param pVM The VM handle.
3063	* @param iPhysExt The physical extent index of the list head.
3064	* @param iShwPT The shadow page table index.
3065	*
3066	*/
3067	static uint16_t pgmPoolTrackPhysExtInsert(PVM pVM, uint16_t iPhysExt, uint16_t iShwPT)
3068	{
3069	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3070	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3071
3072	/* special common case. */
3073	if (paPhysExts[iPhysExt].aidx[2] == NIL_PGMPOOL_IDX)
3074	{
3075	paPhysExts[iPhysExt].aidx[2] = iShwPT;
3076	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
3077	LogFlow(("pgmPoolTrackPhysExtAddref: %d:{,,%d}\n", iPhysExt, iShwPT));
3078	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3079	}
3080
3081	/* general treatment. */
3082	const uint16_t iPhysExtStart = iPhysExt;
3083	unsigned cMax = 15;
3084	for (;;)
3085	{
3086	Assert(iPhysExt < pPool->cMaxPhysExts);
3087	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3088	if (paPhysExts[iPhysExt].aidx[i] == NIL_PGMPOOL_IDX)
3089	{
3090	paPhysExts[iPhysExt].aidx[i] = iShwPT;
3091	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
3092	LogFlow(("pgmPoolTrackPhysExtAddref: %d:{%d} i=%d cMax=%d\n", iPhysExt, iShwPT, i, cMax));
3093	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtStart);
3094	}
3095	if (!--cMax)
3096	{
3097	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
3098	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
3099	LogFlow(("pgmPoolTrackPhysExtAddref: overflow (1) iShwPT=%d\n", iShwPT));
3100	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3101	}
3102	}
3103
3104	/* add another extent to the list. */
3105	PPGMPOOLPHYSEXT pNew = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3106	if (!pNew)
3107	{
3108	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
3109	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
3110	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3111	}
3112	pNew->iNext = iPhysExtStart;
3113	pNew->aidx[0] = iShwPT;
3114	LogFlow(("pgmPoolTrackPhysExtAddref: added new extent %d:{%d}->%d\n", iPhysExt, iShwPT, iPhysExtStart));
3115	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3116	}
3117
3118
3119	/**
3120	* Add a reference to guest physical page where extents are in use.
3121	*
3122	* @returns The new tracking data for PGMPAGE.
3123	*
3124	* @param pVM The VM handle.
3125	* @param u16 The ram range flags (top 16-bits).
3126	* @param iShwPT The shadow page table index.
3127	*/
3128	uint16_t pgmPoolTrackPhysExtAddref(PVM pVM, uint16_t u16, uint16_t iShwPT)
3129	{
3130	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
3131	{
3132	/*
3133	* Convert to extent list.
3134	*/
3135	Assert(PGMPOOL_TD_GET_CREFS(u16) == 1);
3136	uint16_t iPhysExt;
3137	PPGMPOOLPHYSEXT pPhysExt = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3138	if (pPhysExt)
3139	{
3140	LogFlow(("pgmPoolTrackPhysExtAddref: new extent: %d:{%d, %d}\n", iPhysExt, PGMPOOL_TD_GET_IDX(u16), iShwPT));
3141	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliased);
3142	pPhysExt->aidx[0] = PGMPOOL_TD_GET_IDX(u16);
3143	pPhysExt->aidx[1] = iShwPT;
3144	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3145	}
3146	else
3147	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3148	}
3149	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
3150	{
3151	/*
3152	* Insert into the extent list.
3153	*/
3154	u16 = pgmPoolTrackPhysExtInsert(pVM, PGMPOOL_TD_GET_IDX(u16), iShwPT);
3155	}
3156	else
3157	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedLots);
3158	return u16;
3159	}
3160
3161
3162	/**
3163	* Clear references to guest physical memory.
3164	*
3165	* @param pPool The pool.
3166	* @param pPage The page.
3167	* @param pPhysPage Pointer to the aPages entry in the ram range.
3168	*/
3169	void pgmPoolTrackPhysExtDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMPAGE pPhysPage)
3170	{
3171	const unsigned cRefs = PGM_PAGE_GET_TD_CREFS(pPhysPage);
3172	AssertFatalMsg(cRefs == PGMPOOL_TD_CREFS_PHYSEXT, ("cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
3173
3174	uint16_t iPhysExt = PGM_PAGE_GET_TD_IDX(pPhysPage);
3175	if (iPhysExt != PGMPOOL_TD_IDX_OVERFLOWED)
3176	{
3177	uint16_t iPhysExtPrev = NIL_PGMPOOL_PHYSEXT_INDEX;
3178	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3179	do
3180	{
3181	Assert(iPhysExt < pPool->cMaxPhysExts);
3182
3183	/*
3184	* Look for the shadow page and check if it's all freed.
3185	*/
3186	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3187	{
3188	if (paPhysExts[iPhysExt].aidx[i] == pPage->idx)
3189	{
3190	paPhysExts[iPhysExt].aidx[i] = NIL_PGMPOOL_IDX;
3191
3192	for (i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3193	if (paPhysExts[iPhysExt].aidx[i] != NIL_PGMPOOL_IDX)
3194	{
3195	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
3196	return;
3197	}
3198
3199	/* we can free the node. */
3200	PVM pVM = pPool->CTX_SUFF(pVM);
3201	const uint16_t iPhysExtNext = paPhysExts[iPhysExt].iNext;
3202	if ( iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX
3203	&& iPhysExtNext == NIL_PGMPOOL_PHYSEXT_INDEX)
3204	{
3205	/* lonely node */
3206	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3207	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d lonely\n", pPhysPage, pPage->idx));
3208	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
3209	}
3210	else if (iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX)
3211	{
3212	/* head */
3213	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d head\n", pPhysPage, pPage->idx));
3214	PGM_PAGE_SET_TRACKING(pPhysPage, PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtNext));
3215	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3216	}
3217	else
3218	{
3219	/* in list */
3220	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
3221	paPhysExts[iPhysExtPrev].iNext = iPhysExtNext;
3222	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3223	}
3224	iPhysExt = iPhysExtNext;
3225	return;
3226	}
3227	}
3228
3229	/* next */
3230	iPhysExtPrev = iPhysExt;
3231	iPhysExt = paPhysExts[iPhysExt].iNext;
3232	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3233
3234	AssertFatalMsgFailed(("not-found! cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
3235	}
3236	else /* nothing to do */
3237	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage]\n", pPhysPage));
3238	}
3239
3240
3241	/**
3242	* Clear references to guest physical memory.
3243	*
3244	* This is the same as pgmPoolTracDerefGCPhys except that the guest physical address
3245	* is assumed to be correct, so the linear search can be skipped and we can assert
3246	* at an earlier point.
3247	*
3248	* @param pPool The pool.
3249	* @param pPage The page.
3250	* @param HCPhys The host physical address corresponding to the guest page.
3251	* @param GCPhys The guest physical address corresponding to HCPhys.
3252	*/
3253	static void pgmPoolTracDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhys)
3254	{
3255	/*
3256	* Walk range list.
3257	*/
3258	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3259	while (pRam)
3260	{
3261	RTGCPHYS off = GCPhys - pRam->GCPhys;
3262	if (off < pRam->cb)
3263	{
3264	/* does it match? */
3265	const unsigned iPage = off >> PAGE_SHIFT;
3266	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3267	#ifdef LOG_ENABLED
3268	RTHCPHYS HCPhysPage = PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]);
3269	Log2(("pgmPoolTracDerefGCPhys %RHp vs %RHp\n", HCPhysPage, HCPhys));
3270	#endif
3271	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3272	{
3273	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3274	return;
3275	}
3276	break;
3277	}
3278	pRam = pRam->CTX_SUFF(pNext);
3279	}
3280	AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp\n", HCPhys, GCPhys));
3281	}
3282
3283
3284	/**
3285	* Clear references to guest physical memory.
3286	*
3287	* @param pPool The pool.
3288	* @param pPage The page.
3289	* @param HCPhys The host physical address corresponding to the guest page.
3290	* @param GCPhysHint The guest physical address which may corresponding to HCPhys.
3291	*/
3292	static void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint)
3293	{
3294	Log4(("pgmPoolTracDerefGCPhysHint %RHp %RGp\n", HCPhys, GCPhysHint));
3295
3296	/*
3297	* Walk range list.
3298	*/
3299	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3300	while (pRam)
3301	{
3302	RTGCPHYS off = GCPhysHint - pRam->GCPhys;
3303	if (off < pRam->cb)
3304	{
3305	/* does it match? */
3306	const unsigned iPage = off >> PAGE_SHIFT;
3307	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3308	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3309	{
3310	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3311	return;
3312	}
3313	break;
3314	}
3315	pRam = pRam->CTX_SUFF(pNext);
3316	}
3317
3318	/*
3319	* Damn, the hint didn't work. We'll have to do an expensive linear search.
3320	*/
3321	STAM_COUNTER_INC(&pPool->StatTrackLinearRamSearches);
3322	pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3323	while (pRam)
3324	{
3325	unsigned iPage = pRam->cb >> PAGE_SHIFT;
3326	while (iPage-- > 0)
3327	{
3328	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3329	{
3330	Log4(("pgmPoolTracDerefGCPhysHint: Linear HCPhys=%RHp GCPhysHint=%RGp GCPhysReal=%RGp\n",
3331	HCPhys, GCPhysHint, pRam->GCPhys + (iPage << PAGE_SHIFT)));
3332	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3333	return;
3334	}
3335	}
3336	pRam = pRam->CTX_SUFF(pNext);
3337	}
3338
3339	AssertFatalMsgFailed(("HCPhys=%RHp GCPhysHint=%RGp\n", HCPhys, GCPhysHint));
3340	}
3341
3342
3343	/**
3344	* Clear references to guest physical memory in a 32-bit / 32-bit page table.
3345	*
3346	* @param pPool The pool.
3347	* @param pPage The page.
3348	* @param pShwPT The shadow page table (mapping of the page).
3349	* @param pGstPT The guest page table.
3350	*/
3351	DECLINLINE(void) pgmPoolTrackDerefPT32Bit32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT, PCX86PT pGstPT)
3352	{
3353	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
3354	if (pShwPT->a[i].n.u1Present)
3355	{
3356	Log4(("pgmPoolTrackDerefPT32Bit32Bit: i=%d pte=%RX32 hint=%RX32\n",
3357	i, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3358	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3359	if (!--pPage->cPresent)
3360	break;
3361	}
3362	}
3363
3364
3365	/**
3366	* Clear references to guest physical memory in a PAE / 32-bit page table.
3367	*
3368	* @param pPool The pool.
3369	* @param pPage The page.
3370	* @param pShwPT The shadow page table (mapping of the page).
3371	* @param pGstPT The guest page table (just a half one).
3372	*/
3373	DECLINLINE(void) pgmPoolTrackDerefPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PT pGstPT)
3374	{
3375	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3376	if (pShwPT->a[i].n.u1Present)
3377	{
3378	Log4(("pgmPoolTrackDerefPTPae32Bit: i=%d pte=%RX64 hint=%RX32\n",
3379	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3380	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3381	}
3382	}
3383
3384
3385	/**
3386	* Clear references to guest physical memory in a PAE / PAE page table.
3387	*
3388	* @param pPool The pool.
3389	* @param pPage The page.
3390	* @param pShwPT The shadow page table (mapping of the page).
3391	* @param pGstPT The guest page table.
3392	*/
3393	DECLINLINE(void) pgmPoolTrackDerefPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PTPAE pGstPT)
3394	{
3395	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3396	if (pShwPT->a[i].n.u1Present)
3397	{
3398	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX32 hint=%RX32\n",
3399	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
3400	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK);
3401	}
3402	}
3403
3404
3405	/**
3406	* Clear references to guest physical memory in a 32-bit / 4MB page table.
3407	*
3408	* @param pPool The pool.
3409	* @param pPage The page.
3410	* @param pShwPT The shadow page table (mapping of the page).
3411	*/
3412	DECLINLINE(void) pgmPoolTrackDerefPT32Bit4MB(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT)
3413	{
3414	RTGCPHYS GCPhys = pPage->GCPhys;
3415	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3416	if (pShwPT->a[i].n.u1Present)
3417	{
3418	Log4(("pgmPoolTrackDerefPT32Bit4MB: i=%d pte=%RX32 GCPhys=%RGp\n",
3419	i, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys));
3420	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys);
3421	}
3422	}
3423
3424
3425	/**
3426	* Clear references to guest physical memory in a PAE / 2/4MB page table.
3427	*
3428	* @param pPool The pool.
3429	* @param pPage The page.
3430	* @param pShwPT The shadow page table (mapping of the page).
3431	*/
3432	DECLINLINE(void) pgmPoolTrackDerefPTPaeBig(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT)
3433	{
3434	RTGCPHYS GCPhys = pPage->GCPhys;
3435	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3436	if (pShwPT->a[i].n.u1Present)
3437	{
3438	Log4(("pgmPoolTrackDerefPTPaeBig: i=%d pte=%RX64 hint=%RGp\n",
3439	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys));
3440	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys);
3441	}
3442	}
3443
3444	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
3445
3446
3447	/**
3448	* Clear references to shadowed pages in a 32 bits page directory.
3449	*
3450	* @param pPool The pool.
3451	* @param pPage The page.
3452	* @param pShwPD The shadow page directory (mapping of the page).
3453	*/
3454	DECLINLINE(void) pgmPoolTrackDerefPD(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PD pShwPD)
3455	{
3456	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3457	{
3458	if ( pShwPD->a[i].n.u1Present
3459	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
3460	)
3461	{
3462	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PG_MASK);
3463	if (pSubPage)
3464	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3465	else
3466	AssertFatalMsgFailed(("%x\n", pShwPD->a[i].u & X86_PDE_PG_MASK));
3467	}
3468	}
3469	}
3470
3471	/**
3472	* Clear references to shadowed pages in a PAE (legacy or 64 bits) page directory.
3473	*
3474	* @param pPool The pool.
3475	* @param pPage The page.
3476	* @param pShwPD The shadow page directory (mapping of the page).
3477	*/
3478	DECLINLINE(void) pgmPoolTrackDerefPDPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPAE pShwPD)
3479	{
3480	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3481	{
3482	if ( pShwPD->a[i].n.u1Present
3483	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
3484	)
3485	{
3486	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PAE_PG_MASK);
3487	if (pSubPage)
3488	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3489	else
3490	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & X86_PDE_PAE_PG_MASK));
3491	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3492	}
3493	}
3494	}
3495
3496	/**
3497	* Clear references to shadowed pages in a PAE page directory pointer table.
3498	*
3499	* @param pPool The pool.
3500	* @param pPage The page.
3501	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3502	*/
3503	DECLINLINE(void) pgmPoolTrackDerefPDPTPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
3504	{
3505	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
3506	{
3507	if ( pShwPDPT->a[i].n.u1Present
3508	&& !(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING)
3509	)
3510	{
3511	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
3512	if (pSubPage)
3513	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3514	else
3515	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
3516	}
3517	}
3518	}
3519
3520
3521	/**
3522	* Clear references to shadowed pages in a 64-bit page directory pointer table.
3523	*
3524	* @param pPool The pool.
3525	* @param pPage The page.
3526	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3527	*/
3528	DECLINLINE(void) pgmPoolTrackDerefPDPT64Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
3529	{
3530	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3531	{
3532	Assert(!(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING));
3533	if (pShwPDPT->a[i].n.u1Present)
3534	{
3535	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
3536	if (pSubPage)
3537	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3538	else
3539	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
3540	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3541	}
3542	}
3543	}
3544
3545
3546	/**
3547	* Clear references to shadowed pages in a 64-bit level 4 page table.
3548	*
3549	* @param pPool The pool.
3550	* @param pPage The page.
3551	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3552	*/
3553	DECLINLINE(void) pgmPoolTrackDerefPML464Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PML4 pShwPML4)
3554	{
3555	for (unsigned i = 0; i < RT_ELEMENTS(pShwPML4->a); i++)
3556	{
3557	if (pShwPML4->a[i].n.u1Present)
3558	{
3559	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPML4->a[i].u & X86_PDPE_PG_MASK);
3560	if (pSubPage)
3561	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3562	else
3563	AssertFatalMsgFailed(("%RX64\n", pShwPML4->a[i].u & X86_PML4E_PG_MASK));
3564	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3565	}
3566	}
3567	}
3568
3569
3570	/**
3571	* Clear references to shadowed pages in an EPT page table.
3572	*
3573	* @param pPool The pool.
3574	* @param pPage The page.
3575	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3576	*/
3577	DECLINLINE(void) pgmPoolTrackDerefPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT)
3578	{
3579	RTGCPHYS GCPhys = pPage->GCPhys;
3580	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3581	if (pShwPT->a[i].n.u1Present)
3582	{
3583	Log4(("pgmPoolTrackDerefPTEPT: i=%d pte=%RX64 GCPhys=%RX64\n",
3584	i, pShwPT->a[i].u & EPT_PTE_PG_MASK, pPage->GCPhys));
3585	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & EPT_PTE_PG_MASK, GCPhys);
3586	}
3587	}
3588
3589
3590	/**
3591	* Clear references to shadowed pages in an EPT page directory.
3592	*
3593	* @param pPool The pool.
3594	* @param pPage The page.
3595	* @param pShwPD The shadow page directory (mapping of the page).
3596	*/
3597	DECLINLINE(void) pgmPoolTrackDerefPDEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPD pShwPD)
3598	{
3599	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3600	{
3601	if (pShwPD->a[i].n.u1Present)
3602	{
3603	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & EPT_PDE_PG_MASK);
3604	if (pSubPage)
3605	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3606	else
3607	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & EPT_PDE_PG_MASK));
3608	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3609	}
3610	}
3611	}
3612
3613
3614	/**
3615	* Clear references to shadowed pages in an EPT page directory pointer table.
3616	*
3617	* @param pPool The pool.
3618	* @param pPage The page.
3619	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3620	*/
3621	DECLINLINE(void) pgmPoolTrackDerefPDPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPDPT pShwPDPT)
3622	{
3623	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3624	{
3625	if (pShwPDPT->a[i].n.u1Present)
3626	{
3627	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK);
3628	if (pSubPage)
3629	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3630	else
3631	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK));
3632	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3633	}
3634	}
3635	}
3636
3637
3638	/**
3639	* Clears all references made by this page.
3640	*
3641	* This includes other shadow pages and GC physical addresses.
3642	*
3643	* @param pPool The pool.
3644	* @param pPage The page.
3645	*/
3646	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3647	{
3648	/*
3649	* Map the shadow page and take action according to the page kind.
3650	*/
3651	void *pvShw = PGMPOOL_PAGE_2_LOCKED_PTR(pPool->CTX_SUFF(pVM), pPage);
3652	switch (pPage->enmKind)
3653	{
3654	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3655	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3656	{
3657	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3658	void *pvGst;
3659	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3660	pgmPoolTrackDerefPT32Bit32Bit(pPool, pPage, (PX86PT)pvShw, (PCX86PT)pvGst);
3661	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3662	break;
3663	}
3664
3665	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3666	{
3667	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3668	void *pvGst;
3669	int rc = PGM_GCPHYS_2_PTR_EX(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3670	pgmPoolTrackDerefPTPae32Bit(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PT)pvGst);
3671	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3672	break;
3673	}
3674
3675	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3676	{
3677	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3678	void *pvGst;
3679	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3680	pgmPoolTrackDerefPTPaePae(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PTPAE)pvGst);
3681	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3682	break;
3683	}
3684
3685	case PGMPOOLKIND_32BIT_PT_FOR_PHYS: /* treat it like a 4 MB page */
3686	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3687	{
3688	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3689	pgmPoolTrackDerefPT32Bit4MB(pPool, pPage, (PX86PT)pvShw);
3690	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3691	break;
3692	}
3693
3694	case PGMPOOLKIND_PAE_PT_FOR_PHYS: /* treat it like a 2 MB page */
3695	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3696	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3697	{
3698	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3699	pgmPoolTrackDerefPTPaeBig(pPool, pPage, (PX86PTPAE)pvShw);
3700	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3701	break;
3702	}
3703
3704	#else /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3705	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3706	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3707	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3708	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3709	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3710	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3711	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3712	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3713	break;
3714	#endif /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3715
3716	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3717	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3718	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3719	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3720	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3721	case PGMPOOLKIND_PAE_PD_PHYS:
3722	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3723	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3724	pgmPoolTrackDerefPDPae(pPool, pPage, (PX86PDPAE)pvShw);
3725	break;
3726
3727	case PGMPOOLKIND_32BIT_PD_PHYS:
3728	case PGMPOOLKIND_32BIT_PD:
3729	pgmPoolTrackDerefPD(pPool, pPage, (PX86PD)pvShw);
3730	break;
3731
3732	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
3733	case PGMPOOLKIND_PAE_PDPT:
3734	case PGMPOOLKIND_PAE_PDPT_PHYS:
3735	pgmPoolTrackDerefPDPTPae(pPool, pPage, (PX86PDPT)pvShw);
3736	break;
3737
3738	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3739	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3740	pgmPoolTrackDerefPDPT64Bit(pPool, pPage, (PX86PDPT)pvShw);
3741	break;
3742
3743	case PGMPOOLKIND_64BIT_PML4:
3744	pgmPoolTrackDerefPML464Bit(pPool, pPage, (PX86PML4)pvShw);
3745	break;
3746
3747	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3748	pgmPoolTrackDerefPTEPT(pPool, pPage, (PEPTPT)pvShw);
3749	break;
3750
3751	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3752	pgmPoolTrackDerefPDEPT(pPool, pPage, (PEPTPD)pvShw);
3753	break;
3754
3755	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3756	pgmPoolTrackDerefPDPTEPT(pPool, pPage, (PEPTPDPT)pvShw);
3757	break;
3758
3759	default:
3760	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
3761	}
3762
3763	/* paranoia, clear the shadow page. Remove this laser (i.e. let Alloc and ClearAll do it). */
3764	STAM_PROFILE_START(&pPool->StatZeroPage, z);
3765	ASMMemZeroPage(pvShw);
3766	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
3767	pPage->fZeroed = true;
3768	PGMPOOL_UNLOCK_PTR(pPool->CTX_SUFF(pVM), pvShw);
3769	}
3770
3771	#endif /* PGMPOOL_WITH_USER_TRACKING */
3772	#ifdef IN_RING3
3773
3774	/**
3775	* Flushes all the special root pages as part of a pgmPoolFlushAllInt operation.
3776	*
3777	* @param pPool The pool.
3778	*/
3779	static void pgmPoolFlushAllSpecialRoots(PPGMPOOL pPool)
3780	{
3781	/*
3782	* These special pages are all mapped into the indexes 1..PGMPOOL_IDX_FIRST.
3783	*/
3784	Assert(NIL_PGMPOOL_IDX == 0);
3785
3786	/*
3787	* Paranoia (to be removed), flag a global CR3 sync.
3788	*/
3789	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
3790	}
3791
3792
3793	/**
3794	* Flushes the entire cache.
3795	*
3796	* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
3797	* and execute this CR3 flush.
3798	*
3799	* @param pPool The pool.
3800	*/
3801	static void pgmPoolFlushAllInt(PPGMPOOL pPool)
3802	{
3803	PVM pVM = pPool->CTX_SUFF(pVM);
3804
3805	STAM_PROFILE_START(&pPool->StatFlushAllInt, a);
3806	LogFlow(("pgmPoolFlushAllInt:\n"));
3807
3808	/*
3809	* If there are no pages in the pool, there is nothing to do.
3810	*/
3811	if (pPool->cCurPages <= PGMPOOL_IDX_FIRST)
3812	{
3813	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
3814	return;
3815	}
3816
3817	/* Unmap the old CR3 value before flushing everything. */
3818	int rc = PGM_BTH_PFN(UnmapCR3, pVM)(pVM);
3819	AssertRC(rc);
3820
3821	/* Exit the current shadow paging mode as well; nested paging and EPT use a root CR3 which will get flushed here. */
3822	rc = PGM_SHW_PFN(Exit, pVM)(pVM);
3823	AssertRC(rc);
3824
3825	/*
3826	* Nuke the free list and reinsert all pages into it.
3827	*/
3828	for (unsigned i = pPool->cCurPages - 1; i >= PGMPOOL_IDX_FIRST; i--)
3829	{
3830	PPGMPOOLPAGE pPage = &pPool->aPages[i];
3831
3832	#ifdef IN_RING3
3833	Assert(pPage->Core.Key == MMPage2Phys(pVM, pPage->pvPageR3));
3834	#endif
3835	#ifdef PGMPOOL_WITH_MONITORING
3836	if (pPage->fMonitored)
3837	pgmPoolMonitorFlush(pPool, pPage);
3838	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
3839	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
3840	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
3841	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
3842	pPage->cModifications = 0;
3843	#endif
3844	pPage->GCPhys = NIL_RTGCPHYS;
3845	pPage->enmKind = PGMPOOLKIND_FREE;
3846	Assert(pPage->idx == i);
3847	pPage->iNext = i + 1;
3848	pPage->fZeroed = false; /* This could probably be optimized, but better safe than sorry. */
3849	pPage->fSeenNonGlobal = false;
3850	pPage->fMonitored= false;
3851	pPage->fCached = false;
3852	pPage->fReusedFlushPending = false;
3853	#ifdef PGMPOOL_WITH_USER_TRACKING
3854	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
3855	#else
3856	pPage->fCR3Mix = false;
3857	#endif
3858	#ifdef PGMPOOL_WITH_CACHE
3859	pPage->iAgeNext = NIL_PGMPOOL_IDX;
3860	pPage->iAgePrev = NIL_PGMPOOL_IDX;
3861	#endif
3862	pPage->fLocked = false;
3863	}
3864	pPool->aPages[pPool->cCurPages - 1].iNext = NIL_PGMPOOL_IDX;
3865	pPool->iFreeHead = PGMPOOL_IDX_FIRST;
3866	pPool->cUsedPages = 0;
3867
3868	#ifdef PGMPOOL_WITH_USER_TRACKING
3869	/*
3870	* Zap and reinitialize the user records.
3871	*/
3872	pPool->cPresent = 0;
3873	pPool->iUserFreeHead = 0;
3874	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
3875	const unsigned cMaxUsers = pPool->cMaxUsers;
3876	for (unsigned i = 0; i < cMaxUsers; i++)
3877	{
3878	paUsers[i].iNext = i + 1;
3879	paUsers[i].iUser = NIL_PGMPOOL_IDX;
3880	paUsers[i].iUserTable = 0xfffffffe;
3881	}
3882	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
3883	#endif
3884
3885	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3886	/*
3887	* Clear all the GCPhys links and rebuild the phys ext free list.
3888	*/
3889	for (PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRanges);
3890	pRam;
3891	pRam = pRam->CTX_SUFF(pNext))
3892	{
3893	unsigned iPage = pRam->cb >> PAGE_SHIFT;
3894	while (iPage-- > 0)
3895	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
3896	}
3897
3898	pPool->iPhysExtFreeHead = 0;
3899	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3900	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
3901	for (unsigned i = 0; i < cMaxPhysExts; i++)
3902	{
3903	paPhysExts[i].iNext = i + 1;
3904	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
3905	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
3906	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
3907	}
3908	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
3909	#endif
3910
3911	#ifdef PGMPOOL_WITH_MONITORING
3912	/*
3913	* Just zap the modified list.
3914	*/
3915	pPool->cModifiedPages = 0;
3916	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
3917	#endif
3918
3919	#ifdef PGMPOOL_WITH_CACHE
3920	/*
3921	* Clear the GCPhys hash and the age list.
3922	*/
3923	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
3924	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
3925	pPool->iAgeHead = NIL_PGMPOOL_IDX;
3926	pPool->iAgeTail = NIL_PGMPOOL_IDX;
3927	#endif
3928
3929	/*
3930	* Flush all the special root pages.
3931	* Reinsert active pages into the hash and ensure monitoring chains are correct.
3932	*/
3933	pgmPoolFlushAllSpecialRoots(pPool);
3934	for (unsigned i = PGMPOOL_IDX_FIRST_SPECIAL; i < PGMPOOL_IDX_FIRST; i++)
3935	{
3936	PPGMPOOLPAGE pPage = &pPool->aPages[i];
3937	pPage->iNext = NIL_PGMPOOL_IDX;
3938	#ifdef PGMPOOL_WITH_MONITORING
3939	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
3940	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
3941	pPage->cModifications = 0;
3942	/* ASSUMES that we're not sharing with any of the other special pages (safe for now). */
3943	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
3944	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
3945	if (pPage->fMonitored)
3946	{
3947	int rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
3948	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
3949	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
3950	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
3951	pPool->pszAccessHandler);
3952	AssertFatalRCSuccess(rc);
3953	# ifdef PGMPOOL_WITH_CACHE
3954	pgmPoolHashInsert(pPool, pPage);
3955	# endif
3956	}
3957	#endif
3958	#ifdef PGMPOOL_WITH_USER_TRACKING
3959	Assert(pPage->iUserHead == NIL_PGMPOOL_USER_INDEX); /* for now */
3960	#endif
3961	#ifdef PGMPOOL_WITH_CACHE
3962	Assert(pPage->iAgeNext == NIL_PGMPOOL_IDX);
3963	Assert(pPage->iAgePrev == NIL_PGMPOOL_IDX);
3964	#endif
3965	}
3966
3967	/*
3968	* Force a shadow mode reinit (necessary for nested paging and ept).
3969	* Reinit the current shadow paging mode as well; nested paging and
3970	* EPT use a root CR3 which will get flushed here.
3971	*/
3972	pVM->pgm.s.enmShadowMode = PGMMODE_INVALID;
3973	rc = PGMR3ChangeMode(pVM, PGMGetGuestMode(pVM));
3974	AssertRC(rc);
3975
3976	/*
3977	* Finally, assert the FF.
3978	*/
3979	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3);
3980
3981	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
3982	}
3983
3984	#endif /* IN_RING3 */
3985
3986	/**
3987	* Flushes a pool page.
3988	*
3989	* This moves the page to the free list after removing all user references to it.
3990	* In GC this will cause a CR3 reload if the page is traced back to an active root page.
3991	*
3992	* @returns VBox status code.
3993	* @retval VINF_SUCCESS on success.
3994	* @param pPool The pool.
3995	* @param HCPhys The HC physical address of the shadow page.
3996	*/
3997	int pgmPoolFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3998	{
3999	int rc = VINF_SUCCESS;
4000	STAM_PROFILE_START(&pPool->StatFlushPage, f);
4001	LogFlow(("pgmPoolFlushPage: pPage=%p:{.Key=%RHp, .idx=%d, .enmKind=%s, .GCPhys=%RGp}\n",
4002	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
4003
4004	/*
4005	* Quietly reject any attempts at flushing any of the special root pages.
4006	*/
4007	if (pPage->idx < PGMPOOL_IDX_FIRST)
4008	{
4009	AssertFailed(); /* can no longer happen */
4010	Log(("pgmPoolFlushPage: special root page, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
4011	return VINF_SUCCESS;
4012	}
4013
4014	/*
4015	* Quietly reject any attempts at flushing the currently active shadow CR3 mapping
4016	*/
4017	if (pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage))
4018	{
4019	AssertMsg( pPage->enmKind == PGMPOOLKIND_64BIT_PML4
4020	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT
4021	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT_FOR_32BIT
4022	\|\| pPage->enmKind == PGMPOOLKIND_32BIT_PD
4023	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD
4024	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD
4025	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD
4026	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD
4027	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD,
4028	("Can't free the shadow CR3! (%RHp vs %RHp kind=%d\n", PGMGetHyperCR3(pPool->CTX_SUFF(pVM)), pPage->Core.Key, pPage->enmKind));
4029	Log(("pgmPoolFlushPage: current active shadow CR3, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
4030	return VINF_SUCCESS;
4031	}
4032
4033	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
4034	/* Start a subset so we won't run out of mapping space. */
4035	PVMCPU pVCpu = VMMGetCpu(pPool->CTX_SUFF(pVM));
4036	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
4037	#endif
4038
4039	/*
4040	* Mark the page as being in need of a ASMMemZeroPage().
4041	*/
4042	pPage->fZeroed = false;
4043
4044	#ifdef PGMPOOL_WITH_USER_TRACKING
4045	/*
4046	* Clear the page.
4047	*/
4048	pgmPoolTrackClearPageUsers(pPool, pPage);
4049	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
4050	pgmPoolTrackDeref(pPool, pPage);
4051	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
4052	#endif
4053
4054	#ifdef PGMPOOL_WITH_CACHE
4055	/*
4056	* Flush it from the cache.
4057	*/
4058	pgmPoolCacheFlushPage(pPool, pPage);
4059	#endif /* PGMPOOL_WITH_CACHE */
4060
4061	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
4062	/* Heavy stuff done. */
4063	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
4064	#endif
4065
4066	#ifdef PGMPOOL_WITH_MONITORING
4067	/*
4068	* Deregistering the monitoring.
4069	*/
4070	if (pPage->fMonitored)
4071	rc = pgmPoolMonitorFlush(pPool, pPage);
4072	#endif
4073
4074	/*
4075	* Free the page.
4076	*/
4077	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
4078	pPage->iNext = pPool->iFreeHead;
4079	pPool->iFreeHead = pPage->idx;
4080	pPage->enmKind = PGMPOOLKIND_FREE;
4081	pPage->GCPhys = NIL_RTGCPHYS;
4082	pPage->fReusedFlushPending = false;
4083
4084	pPool->cUsedPages--;
4085	STAM_PROFILE_STOP(&pPool->StatFlushPage, f);
4086	return rc;
4087	}
4088
4089
4090	/**
4091	* Frees a usage of a pool page.
4092	*
4093	* The caller is responsible to updating the user table so that it no longer
4094	* references the shadow page.
4095	*
4096	* @param pPool The pool.
4097	* @param HCPhys The HC physical address of the shadow page.
4098	* @param iUser The shadow page pool index of the user table.
4099	* @param iUserTable The index into the user table (shadowed).
4100	*/
4101	void pgmPoolFreeByPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
4102	{
4103	STAM_PROFILE_START(&pPool->StatFree, a);
4104	LogFlow(("pgmPoolFreeByPage: pPage=%p:{.Key=%RHp, .idx=%d, enmKind=%s} iUser=%#x iUserTable=%#x\n",
4105	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), iUser, iUserTable));
4106	Assert(pPage->idx >= PGMPOOL_IDX_FIRST);
4107	#ifdef PGMPOOL_WITH_USER_TRACKING
4108	pgmPoolTrackFreeUser(pPool, pPage, iUser, iUserTable);
4109	#endif
4110	#ifdef PGMPOOL_WITH_CACHE
4111	if (!pPage->fCached)
4112	#endif
4113	pgmPoolFlushPage(pPool, pPage);
4114	STAM_PROFILE_STOP(&pPool->StatFree, a);
4115	}
4116
4117
4118	/**
4119	* Makes one or more free page free.
4120	*
4121	* @returns VBox status code.
4122	* @retval VINF_SUCCESS on success.
4123	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
4124	*
4125	* @param pPool The pool.
4126	* @param enmKind Page table kind
4127	* @param iUser The user of the page.
4128	*/
4129	static int pgmPoolMakeMoreFreePages(PPGMPOOL pPool, PGMPOOLKIND enmKind, uint16_t iUser)
4130	{
4131	LogFlow(("pgmPoolMakeMoreFreePages: iUser=%#x\n", iUser));
4132
4133	/*
4134	* If the pool isn't full grown yet, expand it.
4135	*/
4136	if ( pPool->cCurPages < pPool->cMaxPages
4137	#if defined(IN_RC)
4138	/* Hack alert: we can't deal with jumps to ring 3 when called from MapCR3 and allocating pages for PAE PDs. */
4139	&& enmKind != PGMPOOLKIND_PAE_PD_FOR_PAE_PD
4140	&& (enmKind < PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD \|\| enmKind > PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD)
4141	#endif
4142	)
4143	{
4144	STAM_PROFILE_ADV_SUSPEND(&pPool->StatAlloc, a);
4145	#ifdef IN_RING3
4146	int rc = PGMR3PoolGrow(pPool->pVMR3);
4147	#else
4148	int rc = CTXALLMID(VMM, CallHost)(pPool->CTX_SUFF(pVM), VMMCALLHOST_PGM_POOL_GROW, 0);
4149	#endif
4150	if (RT_FAILURE(rc))
4151	return rc;
4152	STAM_PROFILE_ADV_RESUME(&pPool->StatAlloc, a);
4153	if (pPool->iFreeHead != NIL_PGMPOOL_IDX)
4154	return VINF_SUCCESS;
4155	}
4156
4157	#ifdef PGMPOOL_WITH_CACHE
4158	/*
4159	* Free one cached page.
4160	*/
4161	return pgmPoolCacheFreeOne(pPool, iUser);
4162	#else
4163	/*
4164	* Flush the pool.
4165	*
4166	* If we have tracking enabled, it should be possible to come up with
4167	* a cheap replacement strategy...
4168	*/
4169	/* @todo This path no longer works (CR3 root pages will be flushed)!! */
4170	AssertCompileFailed();
4171	Assert(!CPUMIsGuestInLongMode(pVM));
4172	pgmPoolFlushAllInt(pPool);
4173	return VERR_PGM_POOL_FLUSHED;
4174	#endif
4175	}
4176
4177
4178	/**
4179	* Allocates a page from the pool.
4180	*
4181	* This page may actually be a cached page and not in need of any processing
4182	* on the callers part.
4183	*
4184	* @returns VBox status code.
4185	* @retval VINF_SUCCESS if a NEW page was allocated.
4186	* @retval VINF_PGM_CACHED_PAGE if a CACHED page was returned.
4187	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
4188	* @param pVM The VM handle.
4189	* @param GCPhys The GC physical address of the page we're gonna shadow.
4190	* For 4MB and 2MB PD entries, it's the first address the
4191	* shadow PT is covering.
4192	* @param enmKind The kind of mapping.
4193	* @param iUser The shadow page pool index of the user table.
4194	* @param iUserTable The index into the user table (shadowed).
4195	* @param ppPage Where to store the pointer to the page. NULL is stored here on failure.
4196	*/
4197	int pgmPoolAlloc(PVM pVM, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
4198	{
4199	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4200	STAM_PROFILE_ADV_START(&pPool->StatAlloc, a);
4201	LogFlow(("pgmPoolAlloc: GCPhys=%RGp enmKind=%s iUser=%#x iUserTable=%#x\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable));
4202	*ppPage = NULL;
4203	/** @todo CSAM/PGMPrefetchPage messes up here during CSAMR3CheckGates
4204	* (TRPMR3SyncIDT) because of FF priority. Try fix that?
4205	* Assert(!(pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)); */
4206
4207	#ifdef PGMPOOL_WITH_CACHE
4208	if (pPool->fCacheEnabled)
4209	{
4210	int rc2 = pgmPoolCacheAlloc(pPool, GCPhys, enmKind, iUser, iUserTable, ppPage);
4211	if (RT_SUCCESS(rc2))
4212	{
4213	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4214	LogFlow(("pgmPoolAlloc: cached returns %Rrc ppPage=%p:{.Key=%RHp, .idx=%d}\n", rc2, ppPage, (ppPage)->Core.Key, (ppPage)->idx));
4215	return rc2;
4216	}
4217	}
4218	#endif
4219
4220	/*
4221	* Allocate a new one.
4222	*/
4223	int rc = VINF_SUCCESS;
4224	uint16_t iNew = pPool->iFreeHead;
4225	if (iNew == NIL_PGMPOOL_IDX)
4226	{
4227	rc = pgmPoolMakeMoreFreePages(pPool, enmKind, iUser);
4228	if (RT_FAILURE(rc))
4229	{
4230	Log(("pgmPoolAlloc: returns %Rrc (Free)\n", rc));
4231	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4232	return rc;
4233	}
4234	iNew = pPool->iFreeHead;
4235	AssertReleaseReturn(iNew != NIL_PGMPOOL_IDX, VERR_INTERNAL_ERROR);
4236	}
4237
4238	/* unlink the free head */
4239	PPGMPOOLPAGE pPage = &pPool->aPages[iNew];
4240	pPool->iFreeHead = pPage->iNext;
4241	pPage->iNext = NIL_PGMPOOL_IDX;
4242
4243	/*
4244	* Initialize it.
4245	*/
4246	pPool->cUsedPages++; /* physical handler registration / pgmPoolTrackFlushGCPhysPTsSlow requirement. */
4247	pPage->enmKind = enmKind;
4248	pPage->GCPhys = GCPhys;
4249	pPage->fSeenNonGlobal = false; /* Set this to 'true' to disable this feature. */
4250	pPage->fMonitored = false;
4251	pPage->fCached = false;
4252	pPage->fReusedFlushPending = false;
4253	#ifdef PGMPOOL_WITH_MONITORING
4254	pPage->cModifications = 0;
4255	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4256	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4257	#else
4258	pPage->fCR3Mix = false;
4259	#endif
4260	#ifdef PGMPOOL_WITH_USER_TRACKING
4261	pPage->cPresent = 0;
4262	pPage->iFirstPresent = ~0;
4263
4264	/*
4265	* Insert into the tracking and cache. If this fails, free the page.
4266	*/
4267	int rc3 = pgmPoolTrackInsert(pPool, pPage, GCPhys, iUser, iUserTable);
4268	if (RT_FAILURE(rc3))
4269	{
4270	pPool->cUsedPages--;
4271	pPage->enmKind = PGMPOOLKIND_FREE;
4272	pPage->GCPhys = NIL_RTGCPHYS;
4273	pPage->iNext = pPool->iFreeHead;
4274	pPool->iFreeHead = pPage->idx;
4275	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4276	Log(("pgmPoolAlloc: returns %Rrc (Insert)\n", rc3));
4277	return rc3;
4278	}
4279	#endif /* PGMPOOL_WITH_USER_TRACKING */
4280
4281	/*
4282	* Commit the allocation, clear the page and return.
4283	*/
4284	#ifdef VBOX_WITH_STATISTICS
4285	if (pPool->cUsedPages > pPool->cUsedPagesHigh)
4286	pPool->cUsedPagesHigh = pPool->cUsedPages;
4287	#endif
4288
4289	if (!pPage->fZeroed)
4290	{
4291	STAM_PROFILE_START(&pPool->StatZeroPage, z);
4292	void *pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
4293	ASMMemZeroPage(pv);
4294	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
4295	}
4296
4297	*ppPage = pPage;
4298	LogFlow(("pgmPoolAlloc: returns %Rrc *ppPage=%p:{.Key=%RHp, .idx=%d, .fCached=%RTbool, .fMonitored=%RTbool}\n",
4299	rc, pPage, pPage->Core.Key, pPage->idx, pPage->fCached, pPage->fMonitored));
4300	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4301	return rc;
4302	}
4303
4304
4305	/**
4306	* Frees a usage of a pool page.
4307	*
4308	* @param pVM The VM handle.
4309	* @param HCPhys The HC physical address of the shadow page.
4310	* @param iUser The shadow page pool index of the user table.
4311	* @param iUserTable The index into the user table (shadowed).
4312	*/
4313	void pgmPoolFree(PVM pVM, RTHCPHYS HCPhys, uint16_t iUser, uint32_t iUserTable)
4314	{
4315	LogFlow(("pgmPoolFree: HCPhys=%RHp iUser=%#x iUserTable=%#x\n", HCPhys, iUser, iUserTable));
4316	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4317	pgmPoolFreeByPage(pPool, pgmPoolGetPage(pPool, HCPhys), iUser, iUserTable);
4318	}
4319
4320
4321	/**
4322	* Gets a in-use page in the pool by it's physical address.
4323	*
4324	* @returns Pointer to the page.
4325	* @param pVM The VM handle.
4326	* @param HCPhys The HC physical address of the shadow page.
4327	* @remark This function will NEVER return NULL. It will assert if HCPhys is invalid.
4328	*/
4329	PPGMPOOLPAGE pgmPoolGetPageByHCPhys(PVM pVM, RTHCPHYS HCPhys)
4330	{
4331	/** @todo profile this! */
4332	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4333	PPGMPOOLPAGE pPage = pgmPoolGetPage(pPool, HCPhys);
4334	#ifndef DEBUG_bird /* extremely noisy */
4335	Log5(("pgmPoolGetPageByHCPhys: HCPhys=%RHp -> %p:{.idx=%d .GCPhys=%RGp .enmKind=%s}\n",
4336	HCPhys, pPage, pPage->idx, pPage->GCPhys, pgmPoolPoolKindToStr(pPage->enmKind)));
4337	#endif
4338	return pPage;
4339	}
4340
4341	#ifdef IN_RING3
4342	/**
4343	* Flushes the entire cache.
4344	*
4345	* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
4346	* and execute this CR3 flush.
4347	*
4348	* @param pPool The pool.
4349	*/
4350	void pgmPoolFlushAll(PVM pVM)
4351	{
4352	LogFlow(("pgmPoolFlushAll:\n"));
4353	pgmPoolFlushAllInt(pVM->pgm.s.CTX_SUFF(pPool));
4354	}
4355	#endif /* IN_RING3 */
4356
4357	#ifdef LOG_ENABLED
4358	static const char *pgmPoolPoolKindToStr(uint8_t enmKind)
4359	{
4360	switch(enmKind)
4361	{
4362	case PGMPOOLKIND_INVALID:
4363	return "PGMPOOLKIND_INVALID";
4364	case PGMPOOLKIND_FREE:
4365	return "PGMPOOLKIND_FREE";
4366	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
4367	return "PGMPOOLKIND_32BIT_PT_FOR_PHYS";
4368	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
4369	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT";
4370	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
4371	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB";
4372	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
4373	return "PGMPOOLKIND_PAE_PT_FOR_PHYS";
4374	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
4375	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_PT";
4376	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
4377	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB";
4378	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
4379	return "PGMPOOLKIND_PAE_PT_FOR_PAE_PT";
4380	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
4381	return "PGMPOOLKIND_PAE_PT_FOR_PAE_2MB";
4382	case PGMPOOLKIND_32BIT_PD:
4383	return "PGMPOOLKIND_32BIT_PD";
4384	case PGMPOOLKIND_32BIT_PD_PHYS:
4385	return "PGMPOOLKIND_32BIT_PD_PHYS";
4386	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
4387	return "PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD";
4388	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
4389	return "PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD";
4390	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
4391	return "PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD";
4392	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
4393	return "PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD";
4394	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
4395	return "PGMPOOLKIND_PAE_PD_FOR_PAE_PD";
4396	case PGMPOOLKIND_PAE_PD_PHYS:
4397	return "PGMPOOLKIND_PAE_PD_PHYS";
4398	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
4399	return "PGMPOOLKIND_PAE_PDPT_FOR_32BIT";
4400	case PGMPOOLKIND_PAE_PDPT:
4401	return "PGMPOOLKIND_PAE_PDPT";
4402	case PGMPOOLKIND_PAE_PDPT_PHYS:
4403	return "PGMPOOLKIND_PAE_PDPT_PHYS";
4404	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
4405	return "PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT";
4406	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
4407	return "PGMPOOLKIND_64BIT_PDPT_FOR_PHYS";
4408	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
4409	return "PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD";
4410	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
4411	return "PGMPOOLKIND_64BIT_PD_FOR_PHYS";
4412	case PGMPOOLKIND_64BIT_PML4:
4413	return "PGMPOOLKIND_64BIT_PML4";
4414	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
4415	return "PGMPOOLKIND_EPT_PDPT_FOR_PHYS";
4416	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
4417	return "PGMPOOLKIND_EPT_PD_FOR_PHYS";
4418	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
4419	return "PGMPOOLKIND_EPT_PT_FOR_PHYS";
4420	case PGMPOOLKIND_ROOT_NESTED:
4421	return "PGMPOOLKIND_ROOT_NESTED";
4422	}
4423	return "Unknown kind!";
4424	}
4425	#endif /* LOG_ENABLED*/

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllPool.cpp@ 18645

Download in other formats: