PGMAllPool.cpp@ 18215

Last change on this file since 18215 was 18215, checked in by vboxsync, 16 years ago
pgmPoolMonitorFlush: details.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 158.9 KB

Line
1	/* $Id: PGMAllPool.cpp 18215 2009-03-24 18:39:39Z 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	uint16_t iToFree = pPool->iAgeTail;
1249	if (iToFree == iUser)
1250	iToFree = pPool->aPages[iToFree].iAgePrev;
1251	/* This is the alternative to the SyncCR3 pgmPoolCacheUsed calls.
1252	if (pPool->aPages[iToFree].iUserHead != NIL_PGMPOOL_USER_INDEX)
1253	{
1254	uint16_t i = pPool->aPages[iToFree].iAgePrev;
1255	for (unsigned j = 0; j < 10 && i != NIL_PGMPOOL_USER_INDEX; j++, i = pPool->aPages[i].iAgePrev)
1256	{
1257	if (pPool->aPages[iToFree].iUserHead == NIL_PGMPOOL_USER_INDEX)
1258	continue;
1259	iToFree = i;
1260	break;
1261	}
1262	}
1263	*/
1264
1265	Assert(iToFree != iUser);
1266	AssertRelease(iToFree != NIL_PGMPOOL_IDX);
1267
1268	PPGMPOOLPAGE pPage = &pPool->aPages[iToFree];
1269
1270	/*
1271	* Reject any attempts at flushing the currently active shadow CR3 mapping
1272	*/
1273	if (pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage))
1274	{
1275	/* Refresh the cr3 mapping by putting it at the head of the age list. */
1276	LogFlow(("pgmPoolCacheFreeOne refuse CR3 mapping\n"));
1277	pgmPoolCacheUsed(pPool, pPage);
1278	return pgmPoolCacheFreeOne(pPool, iUser);
1279	}
1280
1281	int rc = pgmPoolFlushPage(pPool, pPage);
1282	if (rc == VINF_SUCCESS)
1283	PGM_INVL_GUEST_TLBS(); /* see PT handler. */
1284	return rc;
1285	}
1286
1287
1288	/**
1289	* Checks if a kind mismatch is really a page being reused
1290	* or if it's just normal remappings.
1291	*
1292	* @returns true if reused and the cached page (enmKind1) should be flushed
1293	* @returns false if not reused.
1294	* @param enmKind1 The kind of the cached page.
1295	* @param enmKind2 The kind of the requested page.
1296	*/
1297	static bool pgmPoolCacheReusedByKind(PGMPOOLKIND enmKind1, PGMPOOLKIND enmKind2)
1298	{
1299	switch (enmKind1)
1300	{
1301	/*
1302	* Never reuse them. There is no remapping in non-paging mode.
1303	*/
1304	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1305	case PGMPOOLKIND_32BIT_PD_PHYS:
1306	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1307	case PGMPOOLKIND_PAE_PD_PHYS:
1308	case PGMPOOLKIND_PAE_PDPT_PHYS:
1309	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1310	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1311	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1312	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1313	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1314	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT: /* never reuse them for other types */
1315	return false;
1316
1317	/*
1318	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1319	*/
1320	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1321	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1322	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1323	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1324	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1325	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1326	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1327	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1328	case PGMPOOLKIND_32BIT_PD:
1329	case PGMPOOLKIND_PAE_PDPT:
1330	switch (enmKind2)
1331	{
1332	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1333	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1334	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1335	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1336	case PGMPOOLKIND_64BIT_PML4:
1337	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1338	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1339	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1340	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1341	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1342	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1343	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1344	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1345	return true;
1346	default:
1347	return false;
1348	}
1349
1350	/*
1351	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1352	*/
1353	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1354	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1355	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1356	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1357	case PGMPOOLKIND_64BIT_PML4:
1358	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1359	switch (enmKind2)
1360	{
1361	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1362	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1363	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1364	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1365	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1366	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1367	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1368	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1369	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1370	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1371	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1372	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1373	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1374	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1375	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1376	return true;
1377	default:
1378	return false;
1379	}
1380
1381	/*
1382	* These cannot be flushed, and it's common to reuse the PDs as PTs.
1383	*/
1384	case PGMPOOLKIND_ROOT_NESTED:
1385	return false;
1386
1387	default:
1388	AssertFatalMsgFailed(("enmKind1=%d\n", enmKind1));
1389	}
1390	}
1391
1392
1393	/**
1394	* Attempts to satisfy a pgmPoolAlloc request from the cache.
1395	*
1396	* @returns VBox status code.
1397	* @retval VINF_PGM_CACHED_PAGE on success.
1398	* @retval VERR_FILE_NOT_FOUND if not found.
1399	* @param pPool The pool.
1400	* @param GCPhys The GC physical address of the page we're gonna shadow.
1401	* @param enmKind The kind of mapping.
1402	* @param iUser The shadow page pool index of the user table.
1403	* @param iUserTable The index into the user table (shadowed).
1404	* @param ppPage Where to store the pointer to the page.
1405	*/
1406	static int pgmPoolCacheAlloc(PPGMPOOL pPool, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
1407	{
1408	#ifndef IN_RC
1409	const PVM pVM = pPool->CTX_SUFF(pVM);
1410	#endif
1411	/*
1412	* Look up the GCPhys in the hash.
1413	*/
1414	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
1415	Log3(("pgmPoolCacheAlloc: %RGp kind %s iUser=%x iUserTable=%x SLOT=%d\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable, i));
1416	if (i != NIL_PGMPOOL_IDX)
1417	{
1418	do
1419	{
1420	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1421	Log4(("pgmPoolCacheAlloc: slot %d found page %RGp\n", i, pPage->GCPhys));
1422	if (pPage->GCPhys == GCPhys)
1423	{
1424	if ((PGMPOOLKIND)pPage->enmKind == enmKind)
1425	{
1426	/* Put it at the start of the use list to make sure pgmPoolTrackAddUser
1427	* doesn't flush it in case there are no more free use records.
1428	*/
1429	pgmPoolCacheUsed(pPool, pPage);
1430
1431	int rc = pgmPoolTrackAddUser(pPool, pPage, iUser, iUserTable);
1432	if (RT_SUCCESS(rc))
1433	{
1434	Assert((PGMPOOLKIND)pPage->enmKind == enmKind);
1435	*ppPage = pPage;
1436	STAM_COUNTER_INC(&pPool->StatCacheHits);
1437	return VINF_PGM_CACHED_PAGE;
1438	}
1439	return rc;
1440	}
1441
1442	/*
1443	* The kind is different. In some cases we should now flush the page
1444	* as it has been reused, but in most cases this is normal remapping
1445	* of PDs as PT or big pages using the GCPhys field in a slightly
1446	* different way than the other kinds.
1447	*/
1448	if (pgmPoolCacheReusedByKind((PGMPOOLKIND)pPage->enmKind, enmKind))
1449	{
1450	STAM_COUNTER_INC(&pPool->StatCacheKindMismatches);
1451	pgmPoolFlushPage(pPool, pPage);
1452	PGM_INVL_GUEST_TLBS(); /* see PT handler. */
1453	break;
1454	}
1455	}
1456
1457	/* next */
1458	i = pPage->iNext;
1459	} while (i != NIL_PGMPOOL_IDX);
1460	}
1461
1462	Log3(("pgmPoolCacheAlloc: Missed GCPhys=%RGp enmKind=%s\n", GCPhys, pgmPoolPoolKindToStr(enmKind)));
1463	STAM_COUNTER_INC(&pPool->StatCacheMisses);
1464	return VERR_FILE_NOT_FOUND;
1465	}
1466
1467
1468	/**
1469	* Inserts a page into the cache.
1470	*
1471	* @param pPool The pool.
1472	* @param pPage The cached page.
1473	* @param fCanBeCached Set if the page is fit for caching from the caller's point of view.
1474	*/
1475	static void pgmPoolCacheInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fCanBeCached)
1476	{
1477	/*
1478	* Insert into the GCPhys hash if the page is fit for that.
1479	*/
1480	Assert(!pPage->fCached);
1481	if (fCanBeCached)
1482	{
1483	pPage->fCached = true;
1484	pgmPoolHashInsert(pPool, pPage);
1485	Log3(("pgmPoolCacheInsert: Caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
1486	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
1487	STAM_COUNTER_INC(&pPool->StatCacheCacheable);
1488	}
1489	else
1490	{
1491	Log3(("pgmPoolCacheInsert: Not caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
1492	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
1493	STAM_COUNTER_INC(&pPool->StatCacheUncacheable);
1494	}
1495
1496	/*
1497	* Insert at the head of the age list.
1498	*/
1499	pPage->iAgePrev = NIL_PGMPOOL_IDX;
1500	pPage->iAgeNext = pPool->iAgeHead;
1501	if (pPool->iAgeHead != NIL_PGMPOOL_IDX)
1502	pPool->aPages[pPool->iAgeHead].iAgePrev = pPage->idx;
1503	else
1504	pPool->iAgeTail = pPage->idx;
1505	pPool->iAgeHead = pPage->idx;
1506	}
1507
1508
1509	/**
1510	* Flushes a cached page.
1511	*
1512	* @param pPool The pool.
1513	* @param pPage The cached page.
1514	*/
1515	static void pgmPoolCacheFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1516	{
1517	Log3(("pgmPoolCacheFlushPage: %RGp\n", pPage->GCPhys));
1518
1519	/*
1520	* Remove the page from the hash.
1521	*/
1522	if (pPage->fCached)
1523	{
1524	pPage->fCached = false;
1525	pgmPoolHashRemove(pPool, pPage);
1526	}
1527	else
1528	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1529
1530	/*
1531	* Remove it from the age list.
1532	*/
1533	if (pPage->iAgeNext != NIL_PGMPOOL_IDX)
1534	pPool->aPages[pPage->iAgeNext].iAgePrev = pPage->iAgePrev;
1535	else
1536	pPool->iAgeTail = pPage->iAgePrev;
1537	if (pPage->iAgePrev != NIL_PGMPOOL_IDX)
1538	pPool->aPages[pPage->iAgePrev].iAgeNext = pPage->iAgeNext;
1539	else
1540	pPool->iAgeHead = pPage->iAgeNext;
1541	pPage->iAgeNext = NIL_PGMPOOL_IDX;
1542	pPage->iAgePrev = NIL_PGMPOOL_IDX;
1543	}
1544
1545	#endif /* PGMPOOL_WITH_CACHE */
1546	#ifdef PGMPOOL_WITH_MONITORING
1547
1548	/**
1549	* Looks for pages sharing the monitor.
1550	*
1551	* @returns Pointer to the head page.
1552	* @returns NULL if not found.
1553	* @param pPool The Pool
1554	* @param pNewPage The page which is going to be monitored.
1555	*/
1556	static PPGMPOOLPAGE pgmPoolMonitorGetPageByGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pNewPage)
1557	{
1558	#ifdef PGMPOOL_WITH_CACHE
1559	/*
1560	* Look up the GCPhys in the hash.
1561	*/
1562	RTGCPHYS GCPhys = pNewPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1563	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
1564	if (i == NIL_PGMPOOL_IDX)
1565	return NULL;
1566	do
1567	{
1568	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1569	if ( pPage->GCPhys - GCPhys < PAGE_SIZE
1570	&& pPage != pNewPage)
1571	{
1572	switch (pPage->enmKind)
1573	{
1574	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1575	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1576	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1577	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1578	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1579	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1580	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1581	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1582	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1583	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1584	case PGMPOOLKIND_64BIT_PML4:
1585	case PGMPOOLKIND_32BIT_PD:
1586	case PGMPOOLKIND_PAE_PDPT:
1587	{
1588	/* find the head */
1589	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1590	{
1591	Assert(pPage->iMonitoredPrev != pPage->idx);
1592	pPage = &pPool->aPages[pPage->iMonitoredPrev];
1593	}
1594	return pPage;
1595	}
1596
1597	/* ignore, no monitoring. */
1598	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1599	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1600	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1601	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1602	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1603	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1604	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1605	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1606	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1607	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1608	case PGMPOOLKIND_ROOT_NESTED:
1609	case PGMPOOLKIND_PAE_PD_PHYS:
1610	case PGMPOOLKIND_PAE_PDPT_PHYS:
1611	case PGMPOOLKIND_32BIT_PD_PHYS:
1612	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
1613	break;
1614	default:
1615	AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
1616	}
1617	}
1618
1619	/* next */
1620	i = pPage->iNext;
1621	} while (i != NIL_PGMPOOL_IDX);
1622	#endif
1623	return NULL;
1624	}
1625
1626
1627	/**
1628	* Enabled write monitoring of a guest page.
1629	*
1630	* @returns VBox status code.
1631	* @retval VINF_SUCCESS on success.
1632	* @param pPool The pool.
1633	* @param pPage The cached page.
1634	*/
1635	static int pgmPoolMonitorInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1636	{
1637	LogFlow(("pgmPoolMonitorInsert %RGp\n", pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1)));
1638
1639	/*
1640	* Filter out the relevant kinds.
1641	*/
1642	switch (pPage->enmKind)
1643	{
1644	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1645	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1646	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1647	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1648	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1649	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1650	case PGMPOOLKIND_64BIT_PML4:
1651	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1652	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1653	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1654	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1655	case PGMPOOLKIND_32BIT_PD:
1656	case PGMPOOLKIND_PAE_PDPT:
1657	break;
1658
1659	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1660	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1661	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1662	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1663	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1664	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1665	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1666	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1667	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1668	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1669	case PGMPOOLKIND_ROOT_NESTED:
1670	/* Nothing to monitor here. */
1671	return VINF_SUCCESS;
1672
1673	case PGMPOOLKIND_32BIT_PD_PHYS:
1674	case PGMPOOLKIND_PAE_PDPT_PHYS:
1675	case PGMPOOLKIND_PAE_PD_PHYS:
1676	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
1677	/* Nothing to monitor here. */
1678	return VINF_SUCCESS;
1679	#ifdef PGMPOOL_WITH_MIXED_PT_CR3
1680	break;
1681	#else
1682	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1683	#endif
1684	default:
1685	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
1686	}
1687
1688	/*
1689	* Install handler.
1690	*/
1691	int rc;
1692	PPGMPOOLPAGE pPageHead = pgmPoolMonitorGetPageByGCPhys(pPool, pPage);
1693	if (pPageHead)
1694	{
1695	Assert(pPageHead != pPage); Assert(pPageHead->iMonitoredNext != pPage->idx);
1696	Assert(pPageHead->iMonitoredPrev != pPage->idx);
1697	pPage->iMonitoredPrev = pPageHead->idx;
1698	pPage->iMonitoredNext = pPageHead->iMonitoredNext;
1699	if (pPageHead->iMonitoredNext != NIL_PGMPOOL_IDX)
1700	pPool->aPages[pPageHead->iMonitoredNext].iMonitoredPrev = pPage->idx;
1701	pPageHead->iMonitoredNext = pPage->idx;
1702	rc = VINF_SUCCESS;
1703	}
1704	else
1705	{
1706	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX); Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1707	PVM pVM = pPool->CTX_SUFF(pVM);
1708	const RTGCPHYS GCPhysPage = pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1709	rc = PGMHandlerPhysicalRegisterEx(pVM, PGMPHYSHANDLERTYPE_PHYSICAL_WRITE,
1710	GCPhysPage, GCPhysPage + (PAGE_SIZE - 1),
1711	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
1712	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
1713	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
1714	pPool->pszAccessHandler);
1715	/** @todo we should probably deal with out-of-memory conditions here, but for now increasing
1716	* the heap size should suffice. */
1717	AssertFatalRC(rc);
1718	Assert(!(pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VM_FF_ISSET(pVM, VM_FF_PGM_SYNC_CR3));
1719	}
1720	pPage->fMonitored = true;
1721	return rc;
1722	}
1723
1724
1725	/**
1726	* Disables write monitoring of a guest page.
1727	*
1728	* @returns VBox status code.
1729	* @retval VINF_SUCCESS on success.
1730	* @param pPool The pool.
1731	* @param pPage The cached page.
1732	*/
1733	static int pgmPoolMonitorFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1734	{
1735	/*
1736	* Filter out the relevant kinds.
1737	*/
1738	switch (pPage->enmKind)
1739	{
1740	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1741	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1742	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1743	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1744	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1745	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1746	case PGMPOOLKIND_64BIT_PML4:
1747	case PGMPOOLKIND_32BIT_PD:
1748	case PGMPOOLKIND_PAE_PDPT:
1749	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1750	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1751	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1752	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1753	break;
1754
1755	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1756	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1757	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1758	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1759	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1760	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1761	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1762	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1763	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1764	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1765	case PGMPOOLKIND_ROOT_NESTED:
1766	case PGMPOOLKIND_PAE_PD_PHYS:
1767	case PGMPOOLKIND_PAE_PDPT_PHYS:
1768	case PGMPOOLKIND_32BIT_PD_PHYS:
1769	/* Nothing to monitor here. */
1770	return VINF_SUCCESS;
1771
1772	#ifdef PGMPOOL_WITH_MIXED_PT_CR3
1773	break;
1774	#endif
1775	default:
1776	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
1777	}
1778
1779	/*
1780	* Remove the page from the monitored list or uninstall it if last.
1781	*/
1782	const PVM pVM = pPool->CTX_SUFF(pVM);
1783	int rc;
1784	if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
1785	\|\| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1786	{
1787	if (pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
1788	{
1789	PPGMPOOLPAGE pNewHead = &pPool->aPages[pPage->iMonitoredNext];
1790	pNewHead->iMonitoredPrev = NIL_PGMPOOL_IDX;
1791	rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
1792	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pNewHead),
1793	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pNewHead),
1794	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pNewHead),
1795	pPool->pszAccessHandler);
1796	AssertFatalRCSuccess(rc);
1797	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
1798	}
1799	else
1800	{
1801	pPool->aPages[pPage->iMonitoredPrev].iMonitoredNext = pPage->iMonitoredNext;
1802	if (pPage->iMonitoredNext != NIL_PGMPOOL_IDX)
1803	{
1804	pPool->aPages[pPage->iMonitoredNext].iMonitoredPrev = pPage->iMonitoredPrev;
1805	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
1806	}
1807	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
1808	rc = VINF_SUCCESS;
1809	}
1810	}
1811	else
1812	{
1813	rc = PGMHandlerPhysicalDeregister(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1));
1814	AssertFatalRC(rc);
1815	AssertMsg(!(pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VM_FF_ISSET(pVM, VM_FF_PGM_SYNC_CR3),
1816	("%#x %#x\n", pVM->pgm.s.fSyncFlags, pVM->fForcedActions));
1817	}
1818	pPage->fMonitored = false;
1819
1820	/*
1821	* Remove it from the list of modified pages (if in it).
1822	*/
1823	pgmPoolMonitorModifiedRemove(pPool, pPage);
1824
1825	return rc;
1826	}
1827
1828
1829	/**
1830	* Inserts the page into the list of modified pages.
1831	*
1832	* @param pPool The pool.
1833	* @param pPage The page.
1834	*/
1835	void pgmPoolMonitorModifiedInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1836	{
1837	Log3(("pgmPoolMonitorModifiedInsert: idx=%d\n", pPage->idx));
1838	AssertMsg( pPage->iModifiedNext == NIL_PGMPOOL_IDX
1839	&& pPage->iModifiedPrev == NIL_PGMPOOL_IDX
1840	&& pPool->iModifiedHead != pPage->idx,
1841	("Next=%d Prev=%d idx=%d cModifications=%d Head=%d cModifiedPages=%d\n",
1842	pPage->iModifiedNext, pPage->iModifiedPrev, pPage->idx, pPage->cModifications,
1843	pPool->iModifiedHead, pPool->cModifiedPages));
1844
1845	pPage->iModifiedNext = pPool->iModifiedHead;
1846	if (pPool->iModifiedHead != NIL_PGMPOOL_IDX)
1847	pPool->aPages[pPool->iModifiedHead].iModifiedPrev = pPage->idx;
1848	pPool->iModifiedHead = pPage->idx;
1849	pPool->cModifiedPages++;
1850	#ifdef VBOX_WITH_STATISTICS
1851	if (pPool->cModifiedPages > pPool->cModifiedPagesHigh)
1852	pPool->cModifiedPagesHigh = pPool->cModifiedPages;
1853	#endif
1854	}
1855
1856
1857	/**
1858	* Removes the page from the list of modified pages and resets the
1859	* moficiation counter.
1860	*
1861	* @param pPool The pool.
1862	* @param pPage The page which is believed to be in the list of modified pages.
1863	*/
1864	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1865	{
1866	Log3(("pgmPoolMonitorModifiedRemove: idx=%d cModifications=%d\n", pPage->idx, pPage->cModifications));
1867	if (pPool->iModifiedHead == pPage->idx)
1868	{
1869	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1870	pPool->iModifiedHead = pPage->iModifiedNext;
1871	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1872	{
1873	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = NIL_PGMPOOL_IDX;
1874	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1875	}
1876	pPool->cModifiedPages--;
1877	}
1878	else if (pPage->iModifiedPrev != NIL_PGMPOOL_IDX)
1879	{
1880	pPool->aPages[pPage->iModifiedPrev].iModifiedNext = pPage->iModifiedNext;
1881	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1882	{
1883	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = pPage->iModifiedPrev;
1884	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1885	}
1886	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1887	pPool->cModifiedPages--;
1888	}
1889	else
1890	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1891	pPage->cModifications = 0;
1892	}
1893
1894
1895	/**
1896	* Zaps the list of modified pages, resetting their modification counters in the process.
1897	*
1898	* @param pVM The VM handle.
1899	*/
1900	void pgmPoolMonitorModifiedClearAll(PVM pVM)
1901	{
1902	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1903	LogFlow(("pgmPoolMonitorModifiedClearAll: cModifiedPages=%d\n", pPool->cModifiedPages));
1904
1905	unsigned cPages = 0; NOREF(cPages);
1906	uint16_t idx = pPool->iModifiedHead;
1907	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
1908	while (idx != NIL_PGMPOOL_IDX)
1909	{
1910	PPGMPOOLPAGE pPage = &pPool->aPages[idx];
1911	idx = pPage->iModifiedNext;
1912	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1913	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1914	pPage->cModifications = 0;
1915	Assert(++cPages);
1916	}
1917	AssertMsg(cPages == pPool->cModifiedPages, ("%d != %d\n", cPages, pPool->cModifiedPages));
1918	pPool->cModifiedPages = 0;
1919	}
1920
1921
1922	#ifdef IN_RING3
1923	/**
1924	* Clear all shadow pages and clear all modification counters.
1925	*
1926	* @param pVM The VM handle.
1927	* @remark Should only be used when monitoring is available, thus placed in
1928	* the PGMPOOL_WITH_MONITORING #ifdef.
1929	*/
1930	void pgmPoolClearAll(PVM pVM)
1931	{
1932	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1933	STAM_PROFILE_START(&pPool->StatClearAll, c);
1934	LogFlow(("pgmPoolClearAll: cUsedPages=%d\n", pPool->cUsedPages));
1935
1936	/*
1937	* Iterate all the pages until we've encountered all that in use.
1938	* This is simple but not quite optimal solution.
1939	*/
1940	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
1941	unsigned cLeft = pPool->cUsedPages;
1942	unsigned iPage = pPool->cCurPages;
1943	while (--iPage >= PGMPOOL_IDX_FIRST)
1944	{
1945	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
1946	if (pPage->GCPhys != NIL_RTGCPHYS)
1947	{
1948	switch (pPage->enmKind)
1949	{
1950	/*
1951	* We only care about shadow page tables.
1952	*/
1953	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1954	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1955	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1956	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1957	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1958	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1959	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1960	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1961	{
1962	#ifdef PGMPOOL_WITH_USER_TRACKING
1963	if (pPage->cPresent)
1964	#endif
1965	{
1966	void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
1967	STAM_PROFILE_START(&pPool->StatZeroPage, z);
1968	ASMMemZeroPage(pvShw);
1969	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
1970	#ifdef PGMPOOL_WITH_USER_TRACKING
1971	pPage->cPresent = 0;
1972	pPage->iFirstPresent = ~0;
1973	#endif
1974	}
1975	}
1976	/* fall thru */
1977
1978	default:
1979	Assert(!pPage->cModifications \|\| ++cModifiedPages);
1980	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
1981	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
1982	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1983	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1984	pPage->cModifications = 0;
1985	break;
1986
1987	}
1988	if (!--cLeft)
1989	break;
1990	}
1991	}
1992
1993	/* swipe the special pages too. */
1994	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
1995	{
1996	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
1997	if (pPage->GCPhys != NIL_RTGCPHYS)
1998	{
1999	Assert(!pPage->cModifications \|\| ++cModifiedPages);
2000	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2001	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2002	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2003	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2004	pPage->cModifications = 0;
2005	}
2006	}
2007
2008	#ifndef DEBUG_michael
2009	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
2010	#endif
2011	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
2012	pPool->cModifiedPages = 0;
2013
2014	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2015	/*
2016	* Clear all the GCPhys links and rebuild the phys ext free list.
2017	*/
2018	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
2019	pRam;
2020	pRam = pRam->CTX_SUFF(pNext))
2021	{
2022	unsigned iPage = pRam->cb >> PAGE_SHIFT;
2023	while (iPage-- > 0)
2024	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
2025	}
2026
2027	pPool->iPhysExtFreeHead = 0;
2028	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
2029	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
2030	for (unsigned i = 0; i < cMaxPhysExts; i++)
2031	{
2032	paPhysExts[i].iNext = i + 1;
2033	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
2034	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
2035	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
2036	}
2037	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
2038	#endif
2039
2040
2041	pPool->cPresent = 0;
2042	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
2043	}
2044	#endif /* IN_RING3 */
2045
2046
2047	/**
2048	* Handle SyncCR3 pool tasks
2049	*
2050	* @returns VBox status code.
2051	* @retval VINF_SUCCESS if successfully added.
2052	* @retval VINF_PGM_SYNC_CR3 is it needs to be deferred to ring 3 (GC only)
2053	* @param pVM The VM handle.
2054	* @remark Should only be used when monitoring is available, thus placed in
2055	* the PGMPOOL_WITH_MONITORING #ifdef.
2056	*/
2057	int pgmPoolSyncCR3(PVM pVM)
2058	{
2059	LogFlow(("pgmPoolSyncCR3\n"));
2060	/*
2061	* When monitoring shadowed pages, we reset the modification counters on CR3 sync.
2062	* Occasionally we will have to clear all the shadow page tables because we wanted
2063	* to monitor a page which was mapped by too many shadowed page tables. This operation
2064	* sometimes refered to as a 'lightweight flush'.
2065	*/
2066	if (!(pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL))
2067	pgmPoolMonitorModifiedClearAll(pVM);
2068	else
2069	{
2070	# ifdef IN_RING3 /* Don't flush in ring-0 or raw mode, it's taking too long. */
2071	pVM->pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
2072	pgmPoolClearAll(pVM);
2073	# else /* !IN_RING3 */
2074	LogFlow(("SyncCR3: PGM_SYNC_CLEAR_PGM_POOL is set -> VINF_PGM_SYNC_CR3\n"));
2075	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
2076	return VINF_PGM_SYNC_CR3;
2077	# endif /* !IN_RING3 */
2078	}
2079	return VINF_SUCCESS;
2080	}
2081
2082	#endif /* PGMPOOL_WITH_MONITORING */
2083	#ifdef PGMPOOL_WITH_USER_TRACKING
2084
2085	/**
2086	* Frees up at least one user entry.
2087	*
2088	* @returns VBox status code.
2089	* @retval VINF_SUCCESS if successfully added.
2090	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2091	* @param pPool The pool.
2092	* @param iUser The user index.
2093	*/
2094	static int pgmPoolTrackFreeOneUser(PPGMPOOL pPool, uint16_t iUser)
2095	{
2096	STAM_COUNTER_INC(&pPool->StatTrackFreeUpOneUser);
2097	#ifdef PGMPOOL_WITH_CACHE
2098	/*
2099	* Just free cached pages in a braindead fashion.
2100	*/
2101	/** @todo walk the age list backwards and free the first with usage. */
2102	int rc = VINF_SUCCESS;
2103	do
2104	{
2105	int rc2 = pgmPoolCacheFreeOne(pPool, iUser);
2106	if (RT_FAILURE(rc2) && rc == VINF_SUCCESS)
2107	rc = rc2;
2108	} while (pPool->iUserFreeHead == NIL_PGMPOOL_USER_INDEX);
2109	return rc;
2110	#else
2111	/*
2112	* Lazy approach.
2113	*/
2114	/* @todo This path no longer works (CR3 root pages will be flushed)!! */
2115	AssertCompileFailed();
2116	Assert(!CPUMIsGuestInLongMode(pVM));
2117	pgmPoolFlushAllInt(pPool);
2118	return VERR_PGM_POOL_FLUSHED;
2119	#endif
2120	}
2121
2122
2123	/**
2124	* Inserts a page into the cache.
2125	*
2126	* This will create user node for the page, insert it into the GCPhys
2127	* hash, and insert it into the age list.
2128	*
2129	* @returns VBox status code.
2130	* @retval VINF_SUCCESS if successfully added.
2131	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2132	* @param pPool The pool.
2133	* @param pPage The cached page.
2134	* @param GCPhys The GC physical address of the page we're gonna shadow.
2135	* @param iUser The user index.
2136	* @param iUserTable The user table index.
2137	*/
2138	DECLINLINE(int) pgmPoolTrackInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhys, uint16_t iUser, uint32_t iUserTable)
2139	{
2140	int rc = VINF_SUCCESS;
2141	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2142
2143	LogFlow(("pgmPoolTrackInsert GCPhys=%RGp iUser %x iUserTable %x\n", GCPhys, iUser, iUserTable));
2144
2145	#ifdef VBOX_STRICT
2146	/*
2147	* Check that the entry doesn't already exists.
2148	*/
2149	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2150	{
2151	uint16_t i = pPage->iUserHead;
2152	do
2153	{
2154	Assert(i < pPool->cMaxUsers);
2155	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2156	i = paUsers[i].iNext;
2157	} while (i != NIL_PGMPOOL_USER_INDEX);
2158	}
2159	#endif
2160
2161	/*
2162	* Find free a user node.
2163	*/
2164	uint16_t i = pPool->iUserFreeHead;
2165	if (i == NIL_PGMPOOL_USER_INDEX)
2166	{
2167	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2168	if (RT_FAILURE(rc))
2169	return rc;
2170	i = pPool->iUserFreeHead;
2171	}
2172
2173	/*
2174	* Unlink the user node from the free list,
2175	* initialize and insert it into the user list.
2176	*/
2177	pPool->iUserFreeHead = paUsers[i].iNext;
2178	paUsers[i].iNext = NIL_PGMPOOL_USER_INDEX;
2179	paUsers[i].iUser = iUser;
2180	paUsers[i].iUserTable = iUserTable;
2181	pPage->iUserHead = i;
2182
2183	/*
2184	* Insert into cache and enable monitoring of the guest page if enabled.
2185	*
2186	* Until we implement caching of all levels, including the CR3 one, we'll
2187	* have to make sure we don't try monitor & cache any recursive reuse of
2188	* a monitored CR3 page. Because all windows versions are doing this we'll
2189	* have to be able to do combined access monitoring, CR3 + PT and
2190	* PD + PT (guest PAE).
2191	*
2192	* Update:
2193	* We're now cooperating with the CR3 monitor if an uncachable page is found.
2194	*/
2195	#if defined(PGMPOOL_WITH_MONITORING) \|\| defined(PGMPOOL_WITH_CACHE)
2196	# ifdef PGMPOOL_WITH_MIXED_PT_CR3
2197	const bool fCanBeMonitored = true;
2198	# else
2199	bool fCanBeMonitored = pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored == NIL_RTGCPHYS
2200	\|\| (GCPhys & X86_PTE_PAE_PG_MASK) != (pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored & X86_PTE_PAE_PG_MASK)
2201	\|\| pgmPoolIsBigPage((PGMPOOLKIND)pPage->enmKind);
2202	# endif
2203	# ifdef PGMPOOL_WITH_CACHE
2204	pgmPoolCacheInsert(pPool, pPage, fCanBeMonitored); /* This can be expanded. */
2205	# endif
2206	if (fCanBeMonitored)
2207	{
2208	# ifdef PGMPOOL_WITH_MONITORING
2209	rc = pgmPoolMonitorInsert(pPool, pPage);
2210	AssertRC(rc);
2211	}
2212	# endif
2213	#endif /* PGMPOOL_WITH_MONITORING */
2214	return rc;
2215	}
2216
2217
2218	# ifdef PGMPOOL_WITH_CACHE /* (only used when the cache is enabled.) */
2219	/**
2220	* Adds a user reference to a page.
2221	*
2222	* This will move the page to the head of the
2223	*
2224	* @returns VBox status code.
2225	* @retval VINF_SUCCESS if successfully added.
2226	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2227	* @param pPool The pool.
2228	* @param pPage The cached page.
2229	* @param iUser The user index.
2230	* @param iUserTable The user table.
2231	*/
2232	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2233	{
2234	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2235
2236	Log3(("pgmPoolTrackAddUser GCPhys = %RGp iUser %x iUserTable %x\n", pPage->GCPhys, iUser, iUserTable));
2237	# ifdef VBOX_STRICT
2238	/*
2239	* Check that the entry doesn't already exists.
2240	*/
2241	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2242	{
2243	uint16_t i = pPage->iUserHead;
2244	do
2245	{
2246	Assert(i < pPool->cMaxUsers);
2247	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2248	i = paUsers[i].iNext;
2249	} while (i != NIL_PGMPOOL_USER_INDEX);
2250	}
2251	# endif
2252
2253	/*
2254	* Allocate a user node.
2255	*/
2256	uint16_t i = pPool->iUserFreeHead;
2257	if (i == NIL_PGMPOOL_USER_INDEX)
2258	{
2259	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2260	if (RT_FAILURE(rc))
2261	return rc;
2262	i = pPool->iUserFreeHead;
2263	}
2264	pPool->iUserFreeHead = paUsers[i].iNext;
2265
2266	/*
2267	* Initialize the user node and insert it.
2268	*/
2269	paUsers[i].iNext = pPage->iUserHead;
2270	paUsers[i].iUser = iUser;
2271	paUsers[i].iUserTable = iUserTable;
2272	pPage->iUserHead = i;
2273
2274	# ifdef PGMPOOL_WITH_CACHE
2275	/*
2276	* Tell the cache to update its replacement stats for this page.
2277	*/
2278	pgmPoolCacheUsed(pPool, pPage);
2279	# endif
2280	return VINF_SUCCESS;
2281	}
2282	# endif /* PGMPOOL_WITH_CACHE */
2283
2284
2285	/**
2286	* Frees a user record associated with a page.
2287	*
2288	* This does not clear the entry in the user table, it simply replaces the
2289	* user record to the chain of free records.
2290	*
2291	* @param pPool The pool.
2292	* @param HCPhys The HC physical address of the shadow page.
2293	* @param iUser The shadow page pool index of the user table.
2294	* @param iUserTable The index into the user table (shadowed).
2295	*/
2296	static void pgmPoolTrackFreeUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2297	{
2298	/*
2299	* Unlink and free the specified user entry.
2300	*/
2301	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2302
2303	Log3(("pgmPoolTrackFreeUser %RGp %x %x\n", pPage->GCPhys, iUser, iUserTable));
2304	/* Special: For PAE and 32-bit paging, there is usually no more than one user. */
2305	uint16_t i = pPage->iUserHead;
2306	if ( i != NIL_PGMPOOL_USER_INDEX
2307	&& paUsers[i].iUser == iUser
2308	&& paUsers[i].iUserTable == iUserTable)
2309	{
2310	pPage->iUserHead = paUsers[i].iNext;
2311
2312	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2313	paUsers[i].iNext = pPool->iUserFreeHead;
2314	pPool->iUserFreeHead = i;
2315	return;
2316	}
2317
2318	/* General: Linear search. */
2319	uint16_t iPrev = NIL_PGMPOOL_USER_INDEX;
2320	while (i != NIL_PGMPOOL_USER_INDEX)
2321	{
2322	if ( paUsers[i].iUser == iUser
2323	&& paUsers[i].iUserTable == iUserTable)
2324	{
2325	if (iPrev != NIL_PGMPOOL_USER_INDEX)
2326	paUsers[iPrev].iNext = paUsers[i].iNext;
2327	else
2328	pPage->iUserHead = paUsers[i].iNext;
2329
2330	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2331	paUsers[i].iNext = pPool->iUserFreeHead;
2332	pPool->iUserFreeHead = i;
2333	return;
2334	}
2335	iPrev = i;
2336	i = paUsers[i].iNext;
2337	}
2338
2339	/* Fatal: didn't find it */
2340	AssertFatalMsgFailed(("Didn't find the user entry! iUser=%#x iUserTable=%#x GCPhys=%RGp\n",
2341	iUser, iUserTable, pPage->GCPhys));
2342	}
2343
2344
2345	/**
2346	* Gets the entry size of a shadow table.
2347	*
2348	* @param enmKind The kind of page.
2349	*
2350	* @returns The size of the entry in bytes. That is, 4 or 8.
2351	* @returns If the kind is not for a table, an assertion is raised and 0 is
2352	* returned.
2353	*/
2354	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind)
2355	{
2356	switch (enmKind)
2357	{
2358	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2359	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2360	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2361	case PGMPOOLKIND_32BIT_PD:
2362	case PGMPOOLKIND_32BIT_PD_PHYS:
2363	return 4;
2364
2365	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2366	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2367	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2368	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2369	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2370	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2371	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2372	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2373	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2374	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2375	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2376	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2377	case PGMPOOLKIND_64BIT_PML4:
2378	case PGMPOOLKIND_PAE_PDPT:
2379	case PGMPOOLKIND_ROOT_NESTED:
2380	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2381	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2382	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2383	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2384	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2385	case PGMPOOLKIND_PAE_PD_PHYS:
2386	case PGMPOOLKIND_PAE_PDPT_PHYS:
2387	return 8;
2388
2389	default:
2390	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2391	}
2392	}
2393
2394
2395	/**
2396	* Gets the entry size of a guest table.
2397	*
2398	* @param enmKind The kind of page.
2399	*
2400	* @returns The size of the entry in bytes. That is, 0, 4 or 8.
2401	* @returns If the kind is not for a table, an assertion is raised and 0 is
2402	* returned.
2403	*/
2404	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind)
2405	{
2406	switch (enmKind)
2407	{
2408	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2409	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2410	case PGMPOOLKIND_32BIT_PD:
2411	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2412	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2413	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2414	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2415	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2416	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2417	return 4;
2418
2419	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2420	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2421	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2422	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2423	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2424	case PGMPOOLKIND_64BIT_PML4:
2425	case PGMPOOLKIND_PAE_PDPT:
2426	return 8;
2427
2428	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2429	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2430	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2431	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2432	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2433	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2434	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2435	case PGMPOOLKIND_ROOT_NESTED:
2436	case PGMPOOLKIND_PAE_PD_PHYS:
2437	case PGMPOOLKIND_PAE_PDPT_PHYS:
2438	case PGMPOOLKIND_32BIT_PD_PHYS:
2439	/** @todo can we return 0? (nobody is calling this...) */
2440	AssertFailed();
2441	return 0;
2442
2443	default:
2444	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2445	}
2446	}
2447
2448	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2449
2450	/**
2451	* Scans one shadow page table for mappings of a physical page.
2452	*
2453	* @param pVM The VM handle.
2454	* @param pPhysPage The guest page in question.
2455	* @param iShw The shadow page table.
2456	* @param cRefs The number of references made in that PT.
2457	*/
2458	static void pgmPoolTrackFlushGCPhysPTInt(PVM pVM, PCPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2459	{
2460	LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%R[pgmpage] iShw=%d cRefs=%d\n", pPhysPage, iShw, cRefs));
2461	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2462
2463	/*
2464	* Assert sanity.
2465	*/
2466	Assert(cRefs == 1);
2467	AssertFatalMsg(iShw < pPool->cCurPages && iShw != NIL_PGMPOOL_IDX, ("iShw=%d\n", iShw));
2468	PPGMPOOLPAGE pPage = &pPool->aPages[iShw];
2469
2470	/*
2471	* Then, clear the actual mappings to the page in the shadow PT.
2472	*/
2473	switch (pPage->enmKind)
2474	{
2475	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2476	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2477	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2478	{
2479	const uint32_t u32 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2480	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2481	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2482	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2483	{
2484	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX32 cRefs=%#x\n", i, pPT->a[i], cRefs));
2485	pPT->a[i].u = 0;
2486	cRefs--;
2487	if (!cRefs)
2488	return;
2489	}
2490	#ifdef LOG_ENABLED
2491	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2492	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2493	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2494	{
2495	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2496	pPT->a[i].u = 0;
2497	}
2498	#endif
2499	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2500	break;
2501	}
2502
2503	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2504	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2505	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2506	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2507	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2508	{
2509	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2510	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2511	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2512	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2513	{
2514	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2515	pPT->a[i].u = 0;
2516	cRefs--;
2517	if (!cRefs)
2518	return;
2519	}
2520	#ifdef LOG_ENABLED
2521	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2522	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2523	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2524	{
2525	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2526	pPT->a[i].u = 0;
2527	}
2528	#endif
2529	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d u64=%RX64\n", cRefs, pPage->iFirstPresent, pPage->cPresent, u64));
2530	break;
2531	}
2532
2533	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2534	{
2535	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2536	PEPTPT pPT = (PEPTPT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2537	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2538	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2539	{
2540	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2541	pPT->a[i].u = 0;
2542	cRefs--;
2543	if (!cRefs)
2544	return;
2545	}
2546	#ifdef LOG_ENABLED
2547	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2548	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2549	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2550	{
2551	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2552	pPT->a[i].u = 0;
2553	}
2554	#endif
2555	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2556	break;
2557	}
2558
2559	default:
2560	AssertFatalMsgFailed(("enmKind=%d iShw=%d\n", pPage->enmKind, iShw));
2561	}
2562	}
2563
2564
2565	/**
2566	* Scans one shadow page table for mappings of a physical page.
2567	*
2568	* @param pVM The VM handle.
2569	* @param pPhysPage The guest page in question.
2570	* @param iShw The shadow page table.
2571	* @param cRefs The number of references made in that PT.
2572	*/
2573	void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2574	{
2575	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
2576	LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%R[pgmpage] iShw=%d cRefs=%d\n", pPhysPage, iShw, cRefs));
2577	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPT, f);
2578	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, iShw, cRefs);
2579	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2580	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPT, f);
2581	}
2582
2583
2584	/**
2585	* Flushes a list of shadow page tables mapping the same physical page.
2586	*
2587	* @param pVM The VM handle.
2588	* @param pPhysPage The guest page in question.
2589	* @param iPhysExt The physical cross reference extent list to flush.
2590	*/
2591	void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, uint16_t iPhysExt)
2592	{
2593	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2594	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTs, f);
2595	LogFlow(("pgmPoolTrackFlushGCPhysPTs: pPhysPage=%R[pgmpage] iPhysExt\n", pPhysPage, iPhysExt));
2596
2597	const uint16_t iPhysExtStart = iPhysExt;
2598	PPGMPOOLPHYSEXT pPhysExt;
2599	do
2600	{
2601	Assert(iPhysExt < pPool->cMaxPhysExts);
2602	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2603	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
2604	if (pPhysExt->aidx[i] != NIL_PGMPOOL_IDX)
2605	{
2606	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, pPhysExt->aidx[i], 1);
2607	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
2608	}
2609
2610	/* next */
2611	iPhysExt = pPhysExt->iNext;
2612	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
2613
2614	/* insert the list into the free list and clear the ram range entry. */
2615	pPhysExt->iNext = pPool->iPhysExtFreeHead;
2616	pPool->iPhysExtFreeHead = iPhysExtStart;
2617	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2618
2619	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTs, f);
2620	}
2621
2622	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
2623
2624	/**
2625	* Flushes all shadow page table mappings of the given guest page.
2626	*
2627	* This is typically called when the host page backing the guest one has been
2628	* replaced or when the page protection was changed due to an access handler.
2629	*
2630	* @returns VBox status code.
2631	* @retval VINF_SUCCESS if all references has been successfully cleared.
2632	* @retval VINF_PGM_GCPHYS_ALIASED if we're better off with a CR3 sync and
2633	* a page pool cleaning.
2634	*
2635	* @param pVM The VM handle.
2636	* @param pPhysPage The guest page in question.
2637	* @param pfFlushTLBs This is set to @a true if the shadow TLBs should be
2638	* flushed, it is NOT touched if this isn't necessary.
2639	* The caller MUST initialized this to @a false.
2640	*/
2641	int pgmPoolTrackFlushGCPhys(PVM pVM, PPGMPAGE pPhysPage, bool *pfFlushTLBs)
2642	{
2643	int rc = VINF_SUCCESS;
2644	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2645	const uint16_t u16 = PGM_PAGE_GET_TRACKING(pPhysPage);
2646	if (u16)
2647	{
2648	# ifdef VBOX_WITH_NEW_PHYS_CODE
2649	/*
2650	* The zero page is currently screwing up the tracking and we'll
2651	* have to flush the whole shebang. Unless VBOX_WITH_NEW_LAZY_PAGE_ALLOC
2652	* is defined, zero pages won't normally be mapped. Some kind of solution
2653	* will be needed for this problem of course, but it will have to wait...
2654	*/
2655	if (PGM_PAGE_IS_ZERO(pPhysPage))
2656	rc = VINF_PGM_GCPHYS_ALIASED;
2657	else
2658	# endif
2659	{
2660	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2661	/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow and
2662	pgmPoolTrackFlushGCPhysPTs will/may kill the pool otherwise. */
2663	PVMCPU pVCpu = VMMGetCpu(pVM);
2664	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
2665	# endif
2666
2667	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
2668	pgmPoolTrackFlushGCPhysPT(pVM,
2669	pPhysPage,
2670	PGMPOOL_TD_GET_IDX(u16),
2671	PGMPOOL_TD_GET_CREFS(u16));
2672	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
2673	pgmPoolTrackFlushGCPhysPTs(pVM, pPhysPage, PGMPOOL_TD_GET_IDX(u16));
2674	else
2675	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
2676	*pfFlushTLBs = true;
2677
2678	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2679	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
2680	# endif
2681	}
2682	}
2683
2684	#elif defined(PGMPOOL_WITH_CACHE)
2685	# ifdef VBOX_WITH_NEW_PHYS_CODE
2686	if (PGM_PAGE_IS_ZERO(pPhysPage))
2687	rc = VINF_PGM_GCPHYS_ALIASED;
2688	else
2689	# endif
2690	{
2691	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2692	/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow kill the pool otherwise. */
2693	PVMCPU pVCpu = VMMGetCpu(pVM);
2694	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
2695	# endif
2696	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
2697	if (rc == VINF_SUCCESS)
2698	*pfFlushTLBs = true;
2699	}
2700
2701	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2702	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
2703	# endif
2704
2705	#else
2706	rc = VINF_PGM_GCPHYS_ALIASED;
2707	#endif
2708
2709	return rc;
2710	}
2711
2712
2713	/**
2714	* Scans all shadow page tables for mappings of a physical page.
2715	*
2716	* This may be slow, but it's most likely more efficient than cleaning
2717	* out the entire page pool / cache.
2718	*
2719	* @returns VBox status code.
2720	* @retval VINF_SUCCESS if all references has been successfully cleared.
2721	* @retval VINF_PGM_GCPHYS_ALIASED if we're better off with a CR3 sync and
2722	* a page pool cleaning.
2723	*
2724	* @param pVM The VM handle.
2725	* @param pPhysPage The guest page in question.
2726	*/
2727	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage)
2728	{
2729	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2730	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2731	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: cUsedPages=%d cPresent=%d pPhysPage=%R[pgmpage]\n",
2732	pPool->cUsedPages, pPool->cPresent, pPhysPage));
2733
2734	#if 1
2735	/*
2736	* There is a limit to what makes sense.
2737	*/
2738	if (pPool->cPresent > 1024)
2739	{
2740	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
2741	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2742	return VINF_PGM_GCPHYS_ALIASED;
2743	}
2744	#endif
2745
2746	/*
2747	* Iterate all the pages until we've encountered all that in use.
2748	* This is simple but not quite optimal solution.
2749	*/
2750	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2751	const uint32_t u32 = u64;
2752	unsigned cLeft = pPool->cUsedPages;
2753	unsigned iPage = pPool->cCurPages;
2754	while (--iPage >= PGMPOOL_IDX_FIRST)
2755	{
2756	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2757	if (pPage->GCPhys != NIL_RTGCPHYS)
2758	{
2759	switch (pPage->enmKind)
2760	{
2761	/*
2762	* We only care about shadow page tables.
2763	*/
2764	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2765	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2766	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2767	{
2768	unsigned cPresent = pPage->cPresent;
2769	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2770	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2771	if (pPT->a[i].n.u1Present)
2772	{
2773	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2774	{
2775	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX32\n", iPage, i, pPT->a[i]));
2776	pPT->a[i].u = 0;
2777	}
2778	if (!--cPresent)
2779	break;
2780	}
2781	break;
2782	}
2783
2784	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2785	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2786	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2787	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2788	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2789	{
2790	unsigned cPresent = pPage->cPresent;
2791	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2792	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2793	if (pPT->a[i].n.u1Present)
2794	{
2795	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2796	{
2797	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
2798	pPT->a[i].u = 0;
2799	}
2800	if (!--cPresent)
2801	break;
2802	}
2803	break;
2804	}
2805	}
2806	if (!--cLeft)
2807	break;
2808	}
2809	}
2810
2811	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2812	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2813	return VINF_SUCCESS;
2814	}
2815
2816
2817	/**
2818	* Clears the user entry in a user table.
2819	*
2820	* This is used to remove all references to a page when flushing it.
2821	*/
2822	static void pgmPoolTrackClearPageUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PCPGMPOOLUSER pUser)
2823	{
2824	Assert(pUser->iUser != NIL_PGMPOOL_IDX);
2825	Assert(pUser->iUser < pPool->cCurPages);
2826	uint32_t iUserTable = pUser->iUserTable;
2827
2828	/*
2829	* Map the user page.
2830	*/
2831	PPGMPOOLPAGE pUserPage = &pPool->aPages[pUser->iUser];
2832	union
2833	{
2834	uint64_t *pau64;
2835	uint32_t *pau32;
2836	} u;
2837	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pUserPage);
2838
2839	LogFlow(("pgmPoolTrackClearPageUser: clear %x in %s (%RGp) (flushing %s)\n", iUserTable, pgmPoolPoolKindToStr(pUserPage->enmKind), pUserPage->Core.Key, pgmPoolPoolKindToStr(pPage->enmKind)));
2840
2841	/* Safety precaution in case we change the paging for other modes too in the future. */
2842	Assert(!pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage));
2843
2844	#ifdef VBOX_STRICT
2845	/*
2846	* Some sanity checks.
2847	*/
2848	switch (pUserPage->enmKind)
2849	{
2850	case PGMPOOLKIND_32BIT_PD:
2851	case PGMPOOLKIND_32BIT_PD_PHYS:
2852	Assert(iUserTable < X86_PG_ENTRIES);
2853	break;
2854	case PGMPOOLKIND_PAE_PDPT:
2855	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2856	case PGMPOOLKIND_PAE_PDPT_PHYS:
2857	Assert(iUserTable < 4);
2858	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2859	break;
2860	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2861	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2862	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2863	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2864	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2865	case PGMPOOLKIND_PAE_PD_PHYS:
2866	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2867	break;
2868	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2869	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2870	Assert(!(u.pau64[iUserTable] & PGM_PDFLAGS_MAPPING));
2871	break;
2872	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2873	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2874	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2875	break;
2876	case PGMPOOLKIND_64BIT_PML4:
2877	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2878	/* GCPhys >> PAGE_SHIFT is the index here */
2879	break;
2880	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2881	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2882	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2883	break;
2884
2885	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2886	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2887	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2888	break;
2889
2890	case PGMPOOLKIND_ROOT_NESTED:
2891	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2892	break;
2893
2894	default:
2895	AssertMsgFailed(("enmKind=%d\n", pUserPage->enmKind));
2896	break;
2897	}
2898	#endif /* VBOX_STRICT */
2899
2900	/*
2901	* Clear the entry in the user page.
2902	*/
2903	switch (pUserPage->enmKind)
2904	{
2905	/* 32-bit entries */
2906	case PGMPOOLKIND_32BIT_PD:
2907	case PGMPOOLKIND_32BIT_PD_PHYS:
2908	u.pau32[iUserTable] = 0;
2909	break;
2910
2911	/* 64-bit entries */
2912	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2913	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2914	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2915	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2916	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2917	#if defined(IN_RC)
2918	/* 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
2919	* non-present PDPT will continue to cause page faults.
2920	*/
2921	ASMReloadCR3();
2922	#endif
2923	/* no break */
2924	case PGMPOOLKIND_PAE_PD_PHYS:
2925	case PGMPOOLKIND_PAE_PDPT_PHYS:
2926	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2927	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2928	case PGMPOOLKIND_64BIT_PML4:
2929	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2930	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2931	case PGMPOOLKIND_PAE_PDPT:
2932	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2933	case PGMPOOLKIND_ROOT_NESTED:
2934	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2935	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2936	u.pau64[iUserTable] = 0;
2937	break;
2938
2939	default:
2940	AssertFatalMsgFailed(("enmKind=%d iUser=%#x iUserTable=%#x\n", pUserPage->enmKind, pUser->iUser, pUser->iUserTable));
2941	}
2942	}
2943
2944
2945	/**
2946	* Clears all users of a page.
2947	*/
2948	static void pgmPoolTrackClearPageUsers(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2949	{
2950	/*
2951	* Free all the user records.
2952	*/
2953	LogFlow(("pgmPoolTrackClearPageUsers %RGp\n", pPage->GCPhys));
2954
2955	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2956	uint16_t i = pPage->iUserHead;
2957	while (i != NIL_PGMPOOL_USER_INDEX)
2958	{
2959	/* Clear enter in user table. */
2960	pgmPoolTrackClearPageUser(pPool, pPage, &paUsers[i]);
2961
2962	/* Free it. */
2963	const uint16_t iNext = paUsers[i].iNext;
2964	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2965	paUsers[i].iNext = pPool->iUserFreeHead;
2966	pPool->iUserFreeHead = i;
2967
2968	/* Next. */
2969	i = iNext;
2970	}
2971	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
2972	}
2973
2974	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2975
2976	/**
2977	* Allocates a new physical cross reference extent.
2978	*
2979	* @returns Pointer to the allocated extent on success. NULL if we're out of them.
2980	* @param pVM The VM handle.
2981	* @param piPhysExt Where to store the phys ext index.
2982	*/
2983	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt)
2984	{
2985	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2986	uint16_t iPhysExt = pPool->iPhysExtFreeHead;
2987	if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
2988	{
2989	STAM_COUNTER_INC(&pPool->StamTrackPhysExtAllocFailures);
2990	return NULL;
2991	}
2992	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2993	pPool->iPhysExtFreeHead = pPhysExt->iNext;
2994	pPhysExt->iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
2995	*piPhysExt = iPhysExt;
2996	return pPhysExt;
2997	}
2998
2999
3000	/**
3001	* Frees a physical cross reference extent.
3002	*
3003	* @param pVM The VM handle.
3004	* @param iPhysExt The extent to free.
3005	*/
3006	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt)
3007	{
3008	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3009	Assert(iPhysExt < pPool->cMaxPhysExts);
3010	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3011	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3012	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3013	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3014	pPool->iPhysExtFreeHead = iPhysExt;
3015	}
3016
3017
3018	/**
3019	* Frees a physical cross reference extent.
3020	*
3021	* @param pVM The VM handle.
3022	* @param iPhysExt The extent to free.
3023	*/
3024	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt)
3025	{
3026	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3027
3028	const uint16_t iPhysExtStart = iPhysExt;
3029	PPGMPOOLPHYSEXT pPhysExt;
3030	do
3031	{
3032	Assert(iPhysExt < pPool->cMaxPhysExts);
3033	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3034	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3035	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3036
3037	/* next */
3038	iPhysExt = pPhysExt->iNext;
3039	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3040
3041	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3042	pPool->iPhysExtFreeHead = iPhysExtStart;
3043	}
3044
3045
3046	/**
3047	* Insert a reference into a list of physical cross reference extents.
3048	*
3049	* @returns The new tracking data for PGMPAGE.
3050	*
3051	* @param pVM The VM handle.
3052	* @param iPhysExt The physical extent index of the list head.
3053	* @param iShwPT The shadow page table index.
3054	*
3055	*/
3056	static uint16_t pgmPoolTrackPhysExtInsert(PVM pVM, uint16_t iPhysExt, uint16_t iShwPT)
3057	{
3058	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3059	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3060
3061	/* special common case. */
3062	if (paPhysExts[iPhysExt].aidx[2] == NIL_PGMPOOL_IDX)
3063	{
3064	paPhysExts[iPhysExt].aidx[2] = iShwPT;
3065	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
3066	LogFlow(("pgmPoolTrackPhysExtAddref: %d:{,,%d}\n", iPhysExt, iShwPT));
3067	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3068	}
3069
3070	/* general treatment. */
3071	const uint16_t iPhysExtStart = iPhysExt;
3072	unsigned cMax = 15;
3073	for (;;)
3074	{
3075	Assert(iPhysExt < pPool->cMaxPhysExts);
3076	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3077	if (paPhysExts[iPhysExt].aidx[i] == NIL_PGMPOOL_IDX)
3078	{
3079	paPhysExts[iPhysExt].aidx[i] = iShwPT;
3080	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
3081	LogFlow(("pgmPoolTrackPhysExtAddref: %d:{%d} i=%d cMax=%d\n", iPhysExt, iShwPT, i, cMax));
3082	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtStart);
3083	}
3084	if (!--cMax)
3085	{
3086	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
3087	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
3088	LogFlow(("pgmPoolTrackPhysExtAddref: overflow (1) iShwPT=%d\n", iShwPT));
3089	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3090	}
3091	}
3092
3093	/* add another extent to the list. */
3094	PPGMPOOLPHYSEXT pNew = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3095	if (!pNew)
3096	{
3097	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
3098	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
3099	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3100	}
3101	pNew->iNext = iPhysExtStart;
3102	pNew->aidx[0] = iShwPT;
3103	LogFlow(("pgmPoolTrackPhysExtAddref: added new extent %d:{%d}->%d\n", iPhysExt, iShwPT, iPhysExtStart));
3104	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3105	}
3106
3107
3108	/**
3109	* Add a reference to guest physical page where extents are in use.
3110	*
3111	* @returns The new tracking data for PGMPAGE.
3112	*
3113	* @param pVM The VM handle.
3114	* @param u16 The ram range flags (top 16-bits).
3115	* @param iShwPT The shadow page table index.
3116	*/
3117	uint16_t pgmPoolTrackPhysExtAddref(PVM pVM, uint16_t u16, uint16_t iShwPT)
3118	{
3119	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
3120	{
3121	/*
3122	* Convert to extent list.
3123	*/
3124	Assert(PGMPOOL_TD_GET_CREFS(u16) == 1);
3125	uint16_t iPhysExt;
3126	PPGMPOOLPHYSEXT pPhysExt = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3127	if (pPhysExt)
3128	{
3129	LogFlow(("pgmPoolTrackPhysExtAddref: new extent: %d:{%d, %d}\n", iPhysExt, PGMPOOL_TD_GET_IDX(u16), iShwPT));
3130	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliased);
3131	pPhysExt->aidx[0] = PGMPOOL_TD_GET_IDX(u16);
3132	pPhysExt->aidx[1] = iShwPT;
3133	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3134	}
3135	else
3136	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3137	}
3138	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
3139	{
3140	/*
3141	* Insert into the extent list.
3142	*/
3143	u16 = pgmPoolTrackPhysExtInsert(pVM, PGMPOOL_TD_GET_IDX(u16), iShwPT);
3144	}
3145	else
3146	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedLots);
3147	return u16;
3148	}
3149
3150
3151	/**
3152	* Clear references to guest physical memory.
3153	*
3154	* @param pPool The pool.
3155	* @param pPage The page.
3156	* @param pPhysPage Pointer to the aPages entry in the ram range.
3157	*/
3158	void pgmPoolTrackPhysExtDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMPAGE pPhysPage)
3159	{
3160	const unsigned cRefs = PGM_PAGE_GET_TD_CREFS(pPhysPage);
3161	AssertFatalMsg(cRefs == PGMPOOL_TD_CREFS_PHYSEXT, ("cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
3162
3163	uint16_t iPhysExt = PGM_PAGE_GET_TD_IDX(pPhysPage);
3164	if (iPhysExt != PGMPOOL_TD_IDX_OVERFLOWED)
3165	{
3166	uint16_t iPhysExtPrev = NIL_PGMPOOL_PHYSEXT_INDEX;
3167	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3168	do
3169	{
3170	Assert(iPhysExt < pPool->cMaxPhysExts);
3171
3172	/*
3173	* Look for the shadow page and check if it's all freed.
3174	*/
3175	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3176	{
3177	if (paPhysExts[iPhysExt].aidx[i] == pPage->idx)
3178	{
3179	paPhysExts[iPhysExt].aidx[i] = NIL_PGMPOOL_IDX;
3180
3181	for (i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3182	if (paPhysExts[iPhysExt].aidx[i] != NIL_PGMPOOL_IDX)
3183	{
3184	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
3185	return;
3186	}
3187
3188	/* we can free the node. */
3189	PVM pVM = pPool->CTX_SUFF(pVM);
3190	const uint16_t iPhysExtNext = paPhysExts[iPhysExt].iNext;
3191	if ( iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX
3192	&& iPhysExtNext == NIL_PGMPOOL_PHYSEXT_INDEX)
3193	{
3194	/* lonely node */
3195	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3196	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d lonely\n", pPhysPage, pPage->idx));
3197	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
3198	}
3199	else if (iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX)
3200	{
3201	/* head */
3202	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d head\n", pPhysPage, pPage->idx));
3203	PGM_PAGE_SET_TRACKING(pPhysPage, PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtNext));
3204	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3205	}
3206	else
3207	{
3208	/* in list */
3209	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
3210	paPhysExts[iPhysExtPrev].iNext = iPhysExtNext;
3211	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3212	}
3213	iPhysExt = iPhysExtNext;
3214	return;
3215	}
3216	}
3217
3218	/* next */
3219	iPhysExtPrev = iPhysExt;
3220	iPhysExt = paPhysExts[iPhysExt].iNext;
3221	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3222
3223	AssertFatalMsgFailed(("not-found! cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
3224	}
3225	else /* nothing to do */
3226	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage]\n", pPhysPage));
3227	}
3228
3229
3230	/**
3231	* Clear references to guest physical memory.
3232	*
3233	* This is the same as pgmPoolTracDerefGCPhys except that the guest physical address
3234	* is assumed to be correct, so the linear search can be skipped and we can assert
3235	* at an earlier point.
3236	*
3237	* @param pPool The pool.
3238	* @param pPage The page.
3239	* @param HCPhys The host physical address corresponding to the guest page.
3240	* @param GCPhys The guest physical address corresponding to HCPhys.
3241	*/
3242	static void pgmPoolTracDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhys)
3243	{
3244	/*
3245	* Walk range list.
3246	*/
3247	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3248	while (pRam)
3249	{
3250	RTGCPHYS off = GCPhys - pRam->GCPhys;
3251	if (off < pRam->cb)
3252	{
3253	/* does it match? */
3254	const unsigned iPage = off >> PAGE_SHIFT;
3255	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3256	#ifdef LOG_ENABLED
3257	RTHCPHYS HCPhysPage = PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]);
3258	Log2(("pgmPoolTracDerefGCPhys %RHp vs %RHp\n", HCPhysPage, HCPhys));
3259	#endif
3260	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3261	{
3262	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3263	return;
3264	}
3265	break;
3266	}
3267	pRam = pRam->CTX_SUFF(pNext);
3268	}
3269	AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp\n", HCPhys, GCPhys));
3270	}
3271
3272
3273	/**
3274	* Clear references to guest physical memory.
3275	*
3276	* @param pPool The pool.
3277	* @param pPage The page.
3278	* @param HCPhys The host physical address corresponding to the guest page.
3279	* @param GCPhysHint The guest physical address which may corresponding to HCPhys.
3280	*/
3281	static void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint)
3282	{
3283	Log4(("pgmPoolTracDerefGCPhysHint %RHp %RGp\n", HCPhys, GCPhysHint));
3284
3285	/*
3286	* Walk range list.
3287	*/
3288	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3289	while (pRam)
3290	{
3291	RTGCPHYS off = GCPhysHint - pRam->GCPhys;
3292	if (off < pRam->cb)
3293	{
3294	/* does it match? */
3295	const unsigned iPage = off >> PAGE_SHIFT;
3296	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3297	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3298	{
3299	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3300	return;
3301	}
3302	break;
3303	}
3304	pRam = pRam->CTX_SUFF(pNext);
3305	}
3306
3307	/*
3308	* Damn, the hint didn't work. We'll have to do an expensive linear search.
3309	*/
3310	STAM_COUNTER_INC(&pPool->StatTrackLinearRamSearches);
3311	pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3312	while (pRam)
3313	{
3314	unsigned iPage = pRam->cb >> PAGE_SHIFT;
3315	while (iPage-- > 0)
3316	{
3317	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3318	{
3319	Log4(("pgmPoolTracDerefGCPhysHint: Linear HCPhys=%RHp GCPhysHint=%RGp GCPhysReal=%RGp\n",
3320	HCPhys, GCPhysHint, pRam->GCPhys + (iPage << PAGE_SHIFT)));
3321	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3322	return;
3323	}
3324	}
3325	pRam = pRam->CTX_SUFF(pNext);
3326	}
3327
3328	AssertFatalMsgFailed(("HCPhys=%RHp GCPhysHint=%RGp\n", HCPhys, GCPhysHint));
3329	}
3330
3331
3332	/**
3333	* Clear references to guest physical memory in a 32-bit / 32-bit page table.
3334	*
3335	* @param pPool The pool.
3336	* @param pPage The page.
3337	* @param pShwPT The shadow page table (mapping of the page).
3338	* @param pGstPT The guest page table.
3339	*/
3340	DECLINLINE(void) pgmPoolTrackDerefPT32Bit32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT, PCX86PT pGstPT)
3341	{
3342	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
3343	if (pShwPT->a[i].n.u1Present)
3344	{
3345	Log4(("pgmPoolTrackDerefPT32Bit32Bit: i=%d pte=%RX32 hint=%RX32\n",
3346	i, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3347	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3348	if (!--pPage->cPresent)
3349	break;
3350	}
3351	}
3352
3353
3354	/**
3355	* Clear references to guest physical memory in a PAE / 32-bit page table.
3356	*
3357	* @param pPool The pool.
3358	* @param pPage The page.
3359	* @param pShwPT The shadow page table (mapping of the page).
3360	* @param pGstPT The guest page table (just a half one).
3361	*/
3362	DECLINLINE(void) pgmPoolTrackDerefPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PT pGstPT)
3363	{
3364	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3365	if (pShwPT->a[i].n.u1Present)
3366	{
3367	Log4(("pgmPoolTrackDerefPTPae32Bit: i=%d pte=%RX64 hint=%RX32\n",
3368	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3369	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3370	}
3371	}
3372
3373
3374	/**
3375	* Clear references to guest physical memory in a PAE / PAE page table.
3376	*
3377	* @param pPool The pool.
3378	* @param pPage The page.
3379	* @param pShwPT The shadow page table (mapping of the page).
3380	* @param pGstPT The guest page table.
3381	*/
3382	DECLINLINE(void) pgmPoolTrackDerefPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PTPAE pGstPT)
3383	{
3384	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3385	if (pShwPT->a[i].n.u1Present)
3386	{
3387	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX32 hint=%RX32\n",
3388	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
3389	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK);
3390	}
3391	}
3392
3393
3394	/**
3395	* Clear references to guest physical memory in a 32-bit / 4MB page table.
3396	*
3397	* @param pPool The pool.
3398	* @param pPage The page.
3399	* @param pShwPT The shadow page table (mapping of the page).
3400	*/
3401	DECLINLINE(void) pgmPoolTrackDerefPT32Bit4MB(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT)
3402	{
3403	RTGCPHYS GCPhys = pPage->GCPhys;
3404	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3405	if (pShwPT->a[i].n.u1Present)
3406	{
3407	Log4(("pgmPoolTrackDerefPT32Bit4MB: i=%d pte=%RX32 GCPhys=%RGp\n",
3408	i, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys));
3409	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys);
3410	}
3411	}
3412
3413
3414	/**
3415	* Clear references to guest physical memory in a PAE / 2/4MB page table.
3416	*
3417	* @param pPool The pool.
3418	* @param pPage The page.
3419	* @param pShwPT The shadow page table (mapping of the page).
3420	*/
3421	DECLINLINE(void) pgmPoolTrackDerefPTPaeBig(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT)
3422	{
3423	RTGCPHYS GCPhys = pPage->GCPhys;
3424	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3425	if (pShwPT->a[i].n.u1Present)
3426	{
3427	Log4(("pgmPoolTrackDerefPTPaeBig: i=%d pte=%RX64 hint=%RGp\n",
3428	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys));
3429	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys);
3430	}
3431	}
3432
3433	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
3434
3435
3436	/**
3437	* Clear references to shadowed pages in a 32 bits page directory.
3438	*
3439	* @param pPool The pool.
3440	* @param pPage The page.
3441	* @param pShwPD The shadow page directory (mapping of the page).
3442	*/
3443	DECLINLINE(void) pgmPoolTrackDerefPD(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PD pShwPD)
3444	{
3445	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3446	{
3447	if ( pShwPD->a[i].n.u1Present
3448	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
3449	)
3450	{
3451	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PG_MASK);
3452	if (pSubPage)
3453	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3454	else
3455	AssertFatalMsgFailed(("%x\n", pShwPD->a[i].u & X86_PDE_PG_MASK));
3456	}
3457	}
3458	}
3459
3460	/**
3461	* Clear references to shadowed pages in a PAE (legacy or 64 bits) page directory.
3462	*
3463	* @param pPool The pool.
3464	* @param pPage The page.
3465	* @param pShwPD The shadow page directory (mapping of the page).
3466	*/
3467	DECLINLINE(void) pgmPoolTrackDerefPDPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPAE pShwPD)
3468	{
3469	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3470	{
3471	if ( pShwPD->a[i].n.u1Present
3472	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
3473	)
3474	{
3475	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PAE_PG_MASK);
3476	if (pSubPage)
3477	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3478	else
3479	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & X86_PDE_PAE_PG_MASK));
3480	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3481	}
3482	}
3483	}
3484
3485	/**
3486	* Clear references to shadowed pages in a PAE page directory pointer table.
3487	*
3488	* @param pPool The pool.
3489	* @param pPage The page.
3490	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3491	*/
3492	DECLINLINE(void) pgmPoolTrackDerefPDPTPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
3493	{
3494	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
3495	{
3496	if ( pShwPDPT->a[i].n.u1Present
3497	&& !(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING)
3498	)
3499	{
3500	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
3501	if (pSubPage)
3502	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3503	else
3504	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
3505	}
3506	}
3507	}
3508
3509
3510	/**
3511	* Clear references to shadowed pages in a 64-bit page directory pointer table.
3512	*
3513	* @param pPool The pool.
3514	* @param pPage The page.
3515	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3516	*/
3517	DECLINLINE(void) pgmPoolTrackDerefPDPT64Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
3518	{
3519	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3520	{
3521	Assert(!(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING));
3522	if (pShwPDPT->a[i].n.u1Present)
3523	{
3524	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
3525	if (pSubPage)
3526	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3527	else
3528	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
3529	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3530	}
3531	}
3532	}
3533
3534
3535	/**
3536	* Clear references to shadowed pages in a 64-bit level 4 page table.
3537	*
3538	* @param pPool The pool.
3539	* @param pPage The page.
3540	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3541	*/
3542	DECLINLINE(void) pgmPoolTrackDerefPML464Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PML4 pShwPML4)
3543	{
3544	for (unsigned i = 0; i < RT_ELEMENTS(pShwPML4->a); i++)
3545	{
3546	if (pShwPML4->a[i].n.u1Present)
3547	{
3548	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPML4->a[i].u & X86_PDPE_PG_MASK);
3549	if (pSubPage)
3550	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3551	else
3552	AssertFatalMsgFailed(("%RX64\n", pShwPML4->a[i].u & X86_PML4E_PG_MASK));
3553	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3554	}
3555	}
3556	}
3557
3558
3559	/**
3560	* Clear references to shadowed pages in an EPT page table.
3561	*
3562	* @param pPool The pool.
3563	* @param pPage The page.
3564	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3565	*/
3566	DECLINLINE(void) pgmPoolTrackDerefPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT)
3567	{
3568	RTGCPHYS GCPhys = pPage->GCPhys;
3569	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3570	if (pShwPT->a[i].n.u1Present)
3571	{
3572	Log4(("pgmPoolTrackDerefPTEPT: i=%d pte=%RX64 GCPhys=%RX64\n",
3573	i, pShwPT->a[i].u & EPT_PTE_PG_MASK, pPage->GCPhys));
3574	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & EPT_PTE_PG_MASK, GCPhys);
3575	}
3576	}
3577
3578
3579	/**
3580	* Clear references to shadowed pages in an EPT page directory.
3581	*
3582	* @param pPool The pool.
3583	* @param pPage The page.
3584	* @param pShwPD The shadow page directory (mapping of the page).
3585	*/
3586	DECLINLINE(void) pgmPoolTrackDerefPDEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPD pShwPD)
3587	{
3588	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3589	{
3590	if (pShwPD->a[i].n.u1Present)
3591	{
3592	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & EPT_PDE_PG_MASK);
3593	if (pSubPage)
3594	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3595	else
3596	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & EPT_PDE_PG_MASK));
3597	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3598	}
3599	}
3600	}
3601
3602
3603	/**
3604	* Clear references to shadowed pages in an EPT page directory pointer table.
3605	*
3606	* @param pPool The pool.
3607	* @param pPage The page.
3608	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3609	*/
3610	DECLINLINE(void) pgmPoolTrackDerefPDPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPDPT pShwPDPT)
3611	{
3612	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3613	{
3614	if (pShwPDPT->a[i].n.u1Present)
3615	{
3616	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK);
3617	if (pSubPage)
3618	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3619	else
3620	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK));
3621	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3622	}
3623	}
3624	}
3625
3626
3627	/**
3628	* Clears all references made by this page.
3629	*
3630	* This includes other shadow pages and GC physical addresses.
3631	*
3632	* @param pPool The pool.
3633	* @param pPage The page.
3634	*/
3635	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3636	{
3637	/*
3638	* Map the shadow page and take action according to the page kind.
3639	*/
3640	void *pvShw = PGMPOOL_PAGE_2_LOCKED_PTR(pPool->CTX_SUFF(pVM), pPage);
3641	switch (pPage->enmKind)
3642	{
3643	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3644	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3645	{
3646	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3647	void *pvGst;
3648	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3649	pgmPoolTrackDerefPT32Bit32Bit(pPool, pPage, (PX86PT)pvShw, (PCX86PT)pvGst);
3650	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3651	break;
3652	}
3653
3654	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3655	{
3656	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3657	void *pvGst;
3658	int rc = PGM_GCPHYS_2_PTR_EX(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3659	pgmPoolTrackDerefPTPae32Bit(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PT)pvGst);
3660	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3661	break;
3662	}
3663
3664	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3665	{
3666	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3667	void *pvGst;
3668	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3669	pgmPoolTrackDerefPTPaePae(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PTPAE)pvGst);
3670	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3671	break;
3672	}
3673
3674	case PGMPOOLKIND_32BIT_PT_FOR_PHYS: /* treat it like a 4 MB page */
3675	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3676	{
3677	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3678	pgmPoolTrackDerefPT32Bit4MB(pPool, pPage, (PX86PT)pvShw);
3679	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3680	break;
3681	}
3682
3683	case PGMPOOLKIND_PAE_PT_FOR_PHYS: /* treat it like a 2 MB page */
3684	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3685	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3686	{
3687	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3688	pgmPoolTrackDerefPTPaeBig(pPool, pPage, (PX86PTPAE)pvShw);
3689	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3690	break;
3691	}
3692
3693	#else /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3694	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3695	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3696	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3697	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3698	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3699	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3700	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3701	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3702	break;
3703	#endif /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3704
3705	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3706	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3707	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3708	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3709	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3710	case PGMPOOLKIND_PAE_PD_PHYS:
3711	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3712	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3713	pgmPoolTrackDerefPDPae(pPool, pPage, (PX86PDPAE)pvShw);
3714	break;
3715
3716	case PGMPOOLKIND_32BIT_PD_PHYS:
3717	case PGMPOOLKIND_32BIT_PD:
3718	pgmPoolTrackDerefPD(pPool, pPage, (PX86PD)pvShw);
3719	break;
3720
3721	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
3722	case PGMPOOLKIND_PAE_PDPT:
3723	case PGMPOOLKIND_PAE_PDPT_PHYS:
3724	pgmPoolTrackDerefPDPTPae(pPool, pPage, (PX86PDPT)pvShw);
3725	break;
3726
3727	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3728	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3729	pgmPoolTrackDerefPDPT64Bit(pPool, pPage, (PX86PDPT)pvShw);
3730	break;
3731
3732	case PGMPOOLKIND_64BIT_PML4:
3733	pgmPoolTrackDerefPML464Bit(pPool, pPage, (PX86PML4)pvShw);
3734	break;
3735
3736	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3737	pgmPoolTrackDerefPTEPT(pPool, pPage, (PEPTPT)pvShw);
3738	break;
3739
3740	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3741	pgmPoolTrackDerefPDEPT(pPool, pPage, (PEPTPD)pvShw);
3742	break;
3743
3744	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3745	pgmPoolTrackDerefPDPTEPT(pPool, pPage, (PEPTPDPT)pvShw);
3746	break;
3747
3748	default:
3749	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
3750	}
3751
3752	/* paranoia, clear the shadow page. Remove this laser (i.e. let Alloc and ClearAll do it). */
3753	STAM_PROFILE_START(&pPool->StatZeroPage, z);
3754	ASMMemZeroPage(pvShw);
3755	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
3756	pPage->fZeroed = true;
3757	PGMPOOL_UNLOCK_PTR(pPool->CTX_SUFF(pVM), pvShw);
3758	}
3759
3760	#endif /* PGMPOOL_WITH_USER_TRACKING */
3761	#ifdef IN_RING3
3762
3763	/**
3764	* Flushes all the special root pages as part of a pgmPoolFlushAllInt operation.
3765	*
3766	* @param pPool The pool.
3767	*/
3768	static void pgmPoolFlushAllSpecialRoots(PPGMPOOL pPool)
3769	{
3770	/*
3771	* These special pages are all mapped into the indexes 1..PGMPOOL_IDX_FIRST.
3772	*/
3773	Assert(NIL_PGMPOOL_IDX == 0);
3774
3775	/*
3776	* Paranoia (to be removed), flag a global CR3 sync.
3777	*/
3778	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
3779	}
3780
3781
3782	/**
3783	* Flushes the entire cache.
3784	*
3785	* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
3786	* and execute this CR3 flush.
3787	*
3788	* @param pPool The pool.
3789	*/
3790	static void pgmPoolFlushAllInt(PPGMPOOL pPool)
3791	{
3792	PVM pVM = pPool->CTX_SUFF(pVM);
3793
3794	STAM_PROFILE_START(&pPool->StatFlushAllInt, a);
3795	LogFlow(("pgmPoolFlushAllInt:\n"));
3796
3797	/*
3798	* If there are no pages in the pool, there is nothing to do.
3799	*/
3800	if (pPool->cCurPages <= PGMPOOL_IDX_FIRST)
3801	{
3802	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
3803	return;
3804	}
3805
3806	/* Unmap the old CR3 value before flushing everything. */
3807	int rc = PGM_BTH_PFN(UnmapCR3, pVM)(pVM);
3808	AssertRC(rc);
3809
3810	/* Exit the current shadow paging mode as well; nested paging and EPT use a root CR3 which will get flushed here. */
3811	rc = PGM_SHW_PFN(Exit, pVM)(pVM);
3812	AssertRC(rc);
3813
3814	/*
3815	* Nuke the free list and reinsert all pages into it.
3816	*/
3817	for (unsigned i = pPool->cCurPages - 1; i >= PGMPOOL_IDX_FIRST; i--)
3818	{
3819	PPGMPOOLPAGE pPage = &pPool->aPages[i];
3820
3821	#ifdef IN_RING3
3822	Assert(pPage->Core.Key == MMPage2Phys(pVM, pPage->pvPageR3));
3823	#endif
3824	#ifdef PGMPOOL_WITH_MONITORING
3825	if (pPage->fMonitored)
3826	pgmPoolMonitorFlush(pPool, pPage);
3827	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
3828	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
3829	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
3830	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
3831	pPage->cModifications = 0;
3832	#endif
3833	pPage->GCPhys = NIL_RTGCPHYS;
3834	pPage->enmKind = PGMPOOLKIND_FREE;
3835	Assert(pPage->idx == i);
3836	pPage->iNext = i + 1;
3837	pPage->fZeroed = false; /* This could probably be optimized, but better safe than sorry. */
3838	pPage->fSeenNonGlobal = false;
3839	pPage->fMonitored= false;
3840	pPage->fCached = false;
3841	pPage->fReusedFlushPending = false;
3842	#ifdef PGMPOOL_WITH_USER_TRACKING
3843	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
3844	#else
3845	pPage->fCR3Mix = false;
3846	#endif
3847	#ifdef PGMPOOL_WITH_CACHE
3848	pPage->iAgeNext = NIL_PGMPOOL_IDX;
3849	pPage->iAgePrev = NIL_PGMPOOL_IDX;
3850	#endif
3851	pPage->fLocked = false;
3852	}
3853	pPool->aPages[pPool->cCurPages - 1].iNext = NIL_PGMPOOL_IDX;
3854	pPool->iFreeHead = PGMPOOL_IDX_FIRST;
3855	pPool->cUsedPages = 0;
3856
3857	#ifdef PGMPOOL_WITH_USER_TRACKING
3858	/*
3859	* Zap and reinitialize the user records.
3860	*/
3861	pPool->cPresent = 0;
3862	pPool->iUserFreeHead = 0;
3863	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
3864	const unsigned cMaxUsers = pPool->cMaxUsers;
3865	for (unsigned i = 0; i < cMaxUsers; i++)
3866	{
3867	paUsers[i].iNext = i + 1;
3868	paUsers[i].iUser = NIL_PGMPOOL_IDX;
3869	paUsers[i].iUserTable = 0xfffffffe;
3870	}
3871	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
3872	#endif
3873
3874	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3875	/*
3876	* Clear all the GCPhys links and rebuild the phys ext free list.
3877	*/
3878	for (PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRanges);
3879	pRam;
3880	pRam = pRam->CTX_SUFF(pNext))
3881	{
3882	unsigned iPage = pRam->cb >> PAGE_SHIFT;
3883	while (iPage-- > 0)
3884	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
3885	}
3886
3887	pPool->iPhysExtFreeHead = 0;
3888	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3889	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
3890	for (unsigned i = 0; i < cMaxPhysExts; i++)
3891	{
3892	paPhysExts[i].iNext = i + 1;
3893	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
3894	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
3895	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
3896	}
3897	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
3898	#endif
3899
3900	#ifdef PGMPOOL_WITH_MONITORING
3901	/*
3902	* Just zap the modified list.
3903	*/
3904	pPool->cModifiedPages = 0;
3905	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
3906	#endif
3907
3908	#ifdef PGMPOOL_WITH_CACHE
3909	/*
3910	* Clear the GCPhys hash and the age list.
3911	*/
3912	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
3913	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
3914	pPool->iAgeHead = NIL_PGMPOOL_IDX;
3915	pPool->iAgeTail = NIL_PGMPOOL_IDX;
3916	#endif
3917
3918	/*
3919	* Flush all the special root pages.
3920	* Reinsert active pages into the hash and ensure monitoring chains are correct.
3921	*/
3922	pgmPoolFlushAllSpecialRoots(pPool);
3923	for (unsigned i = PGMPOOL_IDX_FIRST_SPECIAL; i < PGMPOOL_IDX_FIRST; i++)
3924	{
3925	PPGMPOOLPAGE pPage = &pPool->aPages[i];
3926	pPage->iNext = NIL_PGMPOOL_IDX;
3927	#ifdef PGMPOOL_WITH_MONITORING
3928	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
3929	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
3930	pPage->cModifications = 0;
3931	/* ASSUMES that we're not sharing with any of the other special pages (safe for now). */
3932	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
3933	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
3934	if (pPage->fMonitored)
3935	{
3936	int rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
3937	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
3938	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
3939	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
3940	pPool->pszAccessHandler);
3941	AssertFatalRCSuccess(rc);
3942	# ifdef PGMPOOL_WITH_CACHE
3943	pgmPoolHashInsert(pPool, pPage);
3944	# endif
3945	}
3946	#endif
3947	#ifdef PGMPOOL_WITH_USER_TRACKING
3948	Assert(pPage->iUserHead == NIL_PGMPOOL_USER_INDEX); /* for now */
3949	#endif
3950	#ifdef PGMPOOL_WITH_CACHE
3951	Assert(pPage->iAgeNext == NIL_PGMPOOL_IDX);
3952	Assert(pPage->iAgePrev == NIL_PGMPOOL_IDX);
3953	#endif
3954	}
3955
3956	/* Force a shadow mode reinit (necessary for nested paging and ept). */
3957	pVM->pgm.s.enmShadowMode = PGMMODE_INVALID;
3958
3959	/* Reinit the current shadow paging mode as well; nested paging and EPT use a root CR3 which will get flushed here. */
3960	rc = PGMR3ChangeMode(pVM, PGMGetGuestMode(pVM));
3961	AssertRC(rc);
3962
3963	/*
3964	* Finally, assert the FF.
3965	*/
3966	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3);
3967
3968	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
3969	}
3970
3971	#endif /* IN_RING3 */
3972
3973	/**
3974	* Flushes a pool page.
3975	*
3976	* This moves the page to the free list after removing all user references to it.
3977	* In GC this will cause a CR3 reload if the page is traced back to an active root page.
3978	*
3979	* @returns VBox status code.
3980	* @retval VINF_SUCCESS on success.
3981	* @param pPool The pool.
3982	* @param HCPhys The HC physical address of the shadow page.
3983	*/
3984	int pgmPoolFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3985	{
3986	int rc = VINF_SUCCESS;
3987	STAM_PROFILE_START(&pPool->StatFlushPage, f);
3988	LogFlow(("pgmPoolFlushPage: pPage=%p:{.Key=%RHp, .idx=%d, .enmKind=%s, .GCPhys=%RGp}\n",
3989	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
3990
3991	/*
3992	* Quietly reject any attempts at flushing any of the special root pages.
3993	*/
3994	if (pPage->idx < PGMPOOL_IDX_FIRST)
3995	{
3996	AssertFailed(); /* can no longer happen */
3997	Log(("pgmPoolFlushPage: special root page, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
3998	return VINF_SUCCESS;
3999	}
4000
4001	/*
4002	* Quietly reject any attempts at flushing the currently active shadow CR3 mapping
4003	*/
4004	if (pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage))
4005	{
4006	AssertMsg( pPage->enmKind == PGMPOOLKIND_64BIT_PML4
4007	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT
4008	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT_FOR_32BIT
4009	\|\| pPage->enmKind == PGMPOOLKIND_32BIT_PD
4010	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD
4011	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD
4012	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD
4013	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD
4014	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD,
4015	("Can't free the shadow CR3! (%RHp vs %RHp kind=%d\n", PGMGetHyperCR3(pPool->CTX_SUFF(pVM)), pPage->Core.Key, pPage->enmKind));
4016	Log(("pgmPoolFlushPage: current active shadow CR3, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
4017	return VINF_SUCCESS;
4018	}
4019
4020	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
4021	/* Start a subset so we won't run out of mapping space. */
4022	PVMCPU pVCpu = VMMGetCpu(pPool->CTX_SUFF(pVM));
4023	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
4024	#endif
4025
4026	/*
4027	* Mark the page as being in need of a ASMMemZeroPage().
4028	*/
4029	pPage->fZeroed = false;
4030
4031	#ifdef PGMPOOL_WITH_USER_TRACKING
4032	/*
4033	* Clear the page.
4034	*/
4035	pgmPoolTrackClearPageUsers(pPool, pPage);
4036	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
4037	pgmPoolTrackDeref(pPool, pPage);
4038	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
4039	#endif
4040
4041	#ifdef PGMPOOL_WITH_CACHE
4042	/*
4043	* Flush it from the cache.
4044	*/
4045	pgmPoolCacheFlushPage(pPool, pPage);
4046	#endif /* PGMPOOL_WITH_CACHE */
4047
4048	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
4049	/* Heavy stuff done. */
4050	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
4051	#endif
4052
4053	#ifdef PGMPOOL_WITH_MONITORING
4054	/*
4055	* Deregistering the monitoring.
4056	*/
4057	if (pPage->fMonitored)
4058	rc = pgmPoolMonitorFlush(pPool, pPage);
4059	#endif
4060
4061	/*
4062	* Free the page.
4063	*/
4064	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
4065	pPage->iNext = pPool->iFreeHead;
4066	pPool->iFreeHead = pPage->idx;
4067	pPage->enmKind = PGMPOOLKIND_FREE;
4068	pPage->GCPhys = NIL_RTGCPHYS;
4069	pPage->fReusedFlushPending = false;
4070
4071	pPool->cUsedPages--;
4072	STAM_PROFILE_STOP(&pPool->StatFlushPage, f);
4073	return rc;
4074	}
4075
4076
4077	/**
4078	* Frees a usage of a pool page.
4079	*
4080	* The caller is responsible to updating the user table so that it no longer
4081	* references the shadow page.
4082	*
4083	* @param pPool The pool.
4084	* @param HCPhys The HC physical address of the shadow page.
4085	* @param iUser The shadow page pool index of the user table.
4086	* @param iUserTable The index into the user table (shadowed).
4087	*/
4088	void pgmPoolFreeByPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
4089	{
4090	STAM_PROFILE_START(&pPool->StatFree, a);
4091	LogFlow(("pgmPoolFreeByPage: pPage=%p:{.Key=%RHp, .idx=%d, enmKind=%s} iUser=%#x iUserTable=%#x\n",
4092	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), iUser, iUserTable));
4093	Assert(pPage->idx >= PGMPOOL_IDX_FIRST);
4094	#ifdef PGMPOOL_WITH_USER_TRACKING
4095	pgmPoolTrackFreeUser(pPool, pPage, iUser, iUserTable);
4096	#endif
4097	#ifdef PGMPOOL_WITH_CACHE
4098	if (!pPage->fCached)
4099	#endif
4100	pgmPoolFlushPage(pPool, pPage);
4101	STAM_PROFILE_STOP(&pPool->StatFree, a);
4102	}
4103
4104
4105	/**
4106	* Makes one or more free page free.
4107	*
4108	* @returns VBox status code.
4109	* @retval VINF_SUCCESS on success.
4110	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
4111	*
4112	* @param pPool The pool.
4113	* @param enmKind Page table kind
4114	* @param iUser The user of the page.
4115	*/
4116	static int pgmPoolMakeMoreFreePages(PPGMPOOL pPool, PGMPOOLKIND enmKind, uint16_t iUser)
4117	{
4118	LogFlow(("pgmPoolMakeMoreFreePages: iUser=%#x\n", iUser));
4119
4120	/*
4121	* If the pool isn't full grown yet, expand it.
4122	*/
4123	if ( pPool->cCurPages < pPool->cMaxPages
4124	#if defined(IN_RC)
4125	/* Hack alert: we can't deal with jumps to ring 3 when called from MapCR3 and allocating pages for PAE PDs. */
4126	&& enmKind != PGMPOOLKIND_PAE_PD_FOR_PAE_PD
4127	&& (enmKind < PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD \|\| enmKind > PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD)
4128	#endif
4129	)
4130	{
4131	STAM_PROFILE_ADV_SUSPEND(&pPool->StatAlloc, a);
4132	#ifdef IN_RING3
4133	int rc = PGMR3PoolGrow(pPool->pVMR3);
4134	#else
4135	int rc = CTXALLMID(VMM, CallHost)(pPool->CTX_SUFF(pVM), VMMCALLHOST_PGM_POOL_GROW, 0);
4136	#endif
4137	if (RT_FAILURE(rc))
4138	return rc;
4139	STAM_PROFILE_ADV_RESUME(&pPool->StatAlloc, a);
4140	if (pPool->iFreeHead != NIL_PGMPOOL_IDX)
4141	return VINF_SUCCESS;
4142	}
4143
4144	#ifdef PGMPOOL_WITH_CACHE
4145	/*
4146	* Free one cached page.
4147	*/
4148	return pgmPoolCacheFreeOne(pPool, iUser);
4149	#else
4150	/*
4151	* Flush the pool.
4152	*
4153	* If we have tracking enabled, it should be possible to come up with
4154	* a cheap replacement strategy...
4155	*/
4156	/* @todo This path no longer works (CR3 root pages will be flushed)!! */
4157	AssertCompileFailed();
4158	Assert(!CPUMIsGuestInLongMode(pVM));
4159	pgmPoolFlushAllInt(pPool);
4160	return VERR_PGM_POOL_FLUSHED;
4161	#endif
4162	}
4163
4164
4165	/**
4166	* Allocates a page from the pool.
4167	*
4168	* This page may actually be a cached page and not in need of any processing
4169	* on the callers part.
4170	*
4171	* @returns VBox status code.
4172	* @retval VINF_SUCCESS if a NEW page was allocated.
4173	* @retval VINF_PGM_CACHED_PAGE if a CACHED page was returned.
4174	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
4175	* @param pVM The VM handle.
4176	* @param GCPhys The GC physical address of the page we're gonna shadow.
4177	* For 4MB and 2MB PD entries, it's the first address the
4178	* shadow PT is covering.
4179	* @param enmKind The kind of mapping.
4180	* @param iUser The shadow page pool index of the user table.
4181	* @param iUserTable The index into the user table (shadowed).
4182	* @param ppPage Where to store the pointer to the page. NULL is stored here on failure.
4183	*/
4184	int pgmPoolAlloc(PVM pVM, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
4185	{
4186	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4187	STAM_PROFILE_ADV_START(&pPool->StatAlloc, a);
4188	LogFlow(("pgmPoolAlloc: GCPhys=%RGp enmKind=%s iUser=%#x iUserTable=%#x\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable));
4189	*ppPage = NULL;
4190	/** @todo CSAM/PGMPrefetchPage messes up here during CSAMR3CheckGates
4191	* (TRPMR3SyncIDT) because of FF priority. Try fix that?
4192	* Assert(!(pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)); */
4193
4194	#ifdef PGMPOOL_WITH_CACHE
4195	if (pPool->fCacheEnabled)
4196	{
4197	int rc2 = pgmPoolCacheAlloc(pPool, GCPhys, enmKind, iUser, iUserTable, ppPage);
4198	if (RT_SUCCESS(rc2))
4199	{
4200	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4201	LogFlow(("pgmPoolAlloc: cached returns %Rrc ppPage=%p:{.Key=%RHp, .idx=%d}\n", rc2, ppPage, (ppPage)->Core.Key, (ppPage)->idx));
4202	return rc2;
4203	}
4204	}
4205	#endif
4206
4207	/*
4208	* Allocate a new one.
4209	*/
4210	int rc = VINF_SUCCESS;
4211	uint16_t iNew = pPool->iFreeHead;
4212	if (iNew == NIL_PGMPOOL_IDX)
4213	{
4214	rc = pgmPoolMakeMoreFreePages(pPool, enmKind, iUser);
4215	if (RT_FAILURE(rc))
4216	{
4217	Log(("pgmPoolAlloc: returns %Rrc (Free)\n", rc));
4218	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4219	return rc;
4220	}
4221	iNew = pPool->iFreeHead;
4222	AssertReleaseReturn(iNew != NIL_PGMPOOL_IDX, VERR_INTERNAL_ERROR);
4223	}
4224
4225	/* unlink the free head */
4226	PPGMPOOLPAGE pPage = &pPool->aPages[iNew];
4227	pPool->iFreeHead = pPage->iNext;
4228	pPage->iNext = NIL_PGMPOOL_IDX;
4229
4230	/*
4231	* Initialize it.
4232	*/
4233	pPool->cUsedPages++; /* physical handler registration / pgmPoolTrackFlushGCPhysPTsSlow requirement. */
4234	pPage->enmKind = enmKind;
4235	pPage->GCPhys = GCPhys;
4236	pPage->fSeenNonGlobal = false; /* Set this to 'true' to disable this feature. */
4237	pPage->fMonitored = false;
4238	pPage->fCached = false;
4239	pPage->fReusedFlushPending = false;
4240	#ifdef PGMPOOL_WITH_MONITORING
4241	pPage->cModifications = 0;
4242	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4243	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4244	#else
4245	pPage->fCR3Mix = false;
4246	#endif
4247	#ifdef PGMPOOL_WITH_USER_TRACKING
4248	pPage->cPresent = 0;
4249	pPage->iFirstPresent = ~0;
4250
4251	/*
4252	* Insert into the tracking and cache. If this fails, free the page.
4253	*/
4254	int rc3 = pgmPoolTrackInsert(pPool, pPage, GCPhys, iUser, iUserTable);
4255	if (RT_FAILURE(rc3))
4256	{
4257	pPool->cUsedPages--;
4258	pPage->enmKind = PGMPOOLKIND_FREE;
4259	pPage->GCPhys = NIL_RTGCPHYS;
4260	pPage->iNext = pPool->iFreeHead;
4261	pPool->iFreeHead = pPage->idx;
4262	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4263	Log(("pgmPoolAlloc: returns %Rrc (Insert)\n", rc3));
4264	return rc3;
4265	}
4266	#endif /* PGMPOOL_WITH_USER_TRACKING */
4267
4268	/*
4269	* Commit the allocation, clear the page and return.
4270	*/
4271	#ifdef VBOX_WITH_STATISTICS
4272	if (pPool->cUsedPages > pPool->cUsedPagesHigh)
4273	pPool->cUsedPagesHigh = pPool->cUsedPages;
4274	#endif
4275
4276	if (!pPage->fZeroed)
4277	{
4278	STAM_PROFILE_START(&pPool->StatZeroPage, z);
4279	void *pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
4280	ASMMemZeroPage(pv);
4281	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
4282	}
4283
4284	*ppPage = pPage;
4285	LogFlow(("pgmPoolAlloc: returns %Rrc *ppPage=%p:{.Key=%RHp, .idx=%d, .fCached=%RTbool, .fMonitored=%RTbool}\n",
4286	rc, pPage, pPage->Core.Key, pPage->idx, pPage->fCached, pPage->fMonitored));
4287	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4288	return rc;
4289	}
4290
4291
4292	/**
4293	* Frees a usage of a pool page.
4294	*
4295	* @param pVM The VM handle.
4296	* @param HCPhys The HC physical address of the shadow page.
4297	* @param iUser The shadow page pool index of the user table.
4298	* @param iUserTable The index into the user table (shadowed).
4299	*/
4300	void pgmPoolFree(PVM pVM, RTHCPHYS HCPhys, uint16_t iUser, uint32_t iUserTable)
4301	{
4302	LogFlow(("pgmPoolFree: HCPhys=%RHp iUser=%#x iUserTable=%#x\n", HCPhys, iUser, iUserTable));
4303	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4304	pgmPoolFreeByPage(pPool, pgmPoolGetPage(pPool, HCPhys), iUser, iUserTable);
4305	}
4306
4307
4308	/**
4309	* Gets a in-use page in the pool by it's physical address.
4310	*
4311	* @returns Pointer to the page.
4312	* @param pVM The VM handle.
4313	* @param HCPhys The HC physical address of the shadow page.
4314	* @remark This function will NEVER return NULL. It will assert if HCPhys is invalid.
4315	*/
4316	PPGMPOOLPAGE pgmPoolGetPageByHCPhys(PVM pVM, RTHCPHYS HCPhys)
4317	{
4318	/** @todo profile this! */
4319	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4320	PPGMPOOLPAGE pPage = pgmPoolGetPage(pPool, HCPhys);
4321	Log4(("pgmPoolGetPageByHCPhys: HCPhys=%RHp -> %p:{.idx=%d .GCPhys=%RGp .enmKind=%s}\n",
4322	HCPhys, pPage, pPage->idx, pPage->GCPhys, pgmPoolPoolKindToStr(pPage->enmKind)));
4323	return pPage;
4324	}
4325
4326	#ifdef IN_RING3
4327	/**
4328	* Flushes the entire cache.
4329	*
4330	* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
4331	* and execute this CR3 flush.
4332	*
4333	* @param pPool The pool.
4334	*/
4335	void pgmPoolFlushAll(PVM pVM)
4336	{
4337	LogFlow(("pgmPoolFlushAll:\n"));
4338	pgmPoolFlushAllInt(pVM->pgm.s.CTX_SUFF(pPool));
4339	}
4340	#endif /* IN_RING3 */
4341
4342	#ifdef LOG_ENABLED
4343	static const char *pgmPoolPoolKindToStr(uint8_t enmKind)
4344	{
4345	switch(enmKind)
4346	{
4347	case PGMPOOLKIND_INVALID:
4348	return "PGMPOOLKIND_INVALID";
4349	case PGMPOOLKIND_FREE:
4350	return "PGMPOOLKIND_FREE";
4351	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
4352	return "PGMPOOLKIND_32BIT_PT_FOR_PHYS";
4353	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
4354	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT";
4355	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
4356	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB";
4357	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
4358	return "PGMPOOLKIND_PAE_PT_FOR_PHYS";
4359	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
4360	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_PT";
4361	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
4362	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB";
4363	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
4364	return "PGMPOOLKIND_PAE_PT_FOR_PAE_PT";
4365	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
4366	return "PGMPOOLKIND_PAE_PT_FOR_PAE_2MB";
4367	case PGMPOOLKIND_32BIT_PD:
4368	return "PGMPOOLKIND_32BIT_PD";
4369	case PGMPOOLKIND_32BIT_PD_PHYS:
4370	return "PGMPOOLKIND_32BIT_PD_PHYS";
4371	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
4372	return "PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD";
4373	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
4374	return "PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD";
4375	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
4376	return "PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD";
4377	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
4378	return "PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD";
4379	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
4380	return "PGMPOOLKIND_PAE_PD_FOR_PAE_PD";
4381	case PGMPOOLKIND_PAE_PD_PHYS:
4382	return "PGMPOOLKIND_PAE_PD_PHYS";
4383	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
4384	return "PGMPOOLKIND_PAE_PDPT_FOR_32BIT";
4385	case PGMPOOLKIND_PAE_PDPT:
4386	return "PGMPOOLKIND_PAE_PDPT";
4387	case PGMPOOLKIND_PAE_PDPT_PHYS:
4388	return "PGMPOOLKIND_PAE_PDPT_PHYS";
4389	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
4390	return "PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT";
4391	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
4392	return "PGMPOOLKIND_64BIT_PDPT_FOR_PHYS";
4393	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
4394	return "PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD";
4395	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
4396	return "PGMPOOLKIND_64BIT_PD_FOR_PHYS";
4397	case PGMPOOLKIND_64BIT_PML4:
4398	return "PGMPOOLKIND_64BIT_PML4";
4399	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
4400	return "PGMPOOLKIND_EPT_PDPT_FOR_PHYS";
4401	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
4402	return "PGMPOOLKIND_EPT_PD_FOR_PHYS";
4403	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
4404	return "PGMPOOLKIND_EPT_PT_FOR_PHYS";
4405	case PGMPOOLKIND_ROOT_NESTED:
4406	return "PGMPOOLKIND_ROOT_NESTED";
4407	}
4408	return "Unknown kind!";
4409	}
4410	#endif /* LOG_ENABLED*/

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllPool.cpp@ 18215

Download in other formats: