PGMAllPool.cpp@ 20140

Last change on this file since 20140 was 20140, checked in by vboxsync, 16 years ago
No need to inform all VCPUs about a pending pool flush. The current VCPU will take care of it.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 161.8 KB

Line
1	/* $Id: PGMAllPool.cpp 20140 2009-05-29 09:50:33Z 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.215389.xyz. 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
71	void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs);
72	void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, uint16_t iPhysExt);
73	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage);
74	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt);
75	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt);
76	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt);
77
78	__END_DECLS
79
80
81	/**
82	* Checks if the specified page pool kind is for a 4MB or 2MB guest page.
83	*
84	* @returns true if it's the shadow of a 4MB or 2MB guest page, otherwise false.
85	* @param enmKind The page kind.
86	*/
87	DECLINLINE(bool) pgmPoolIsBigPage(PGMPOOLKIND enmKind)
88	{
89	switch (enmKind)
90	{
91	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
92	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
93	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
94	return true;
95	default:
96	return false;
97	}
98	}
99
100	/** @def PGMPOOL_PAGE_2_LOCKED_PTR
101	* Maps a pool page pool into the current context and lock it (RC only).
102	*
103	* @returns VBox status code.
104	* @param pVM The VM handle.
105	* @param pPage The pool page.
106	*
107	* @remark In RC this uses PGMGCDynMapHCPage(), so it will consume of the
108	* small page window employeed by that function. Be careful.
109	* @remark There is no need to assert on the result.
110	*/
111	#if defined(IN_RC)
112	DECLINLINE(void *) PGMPOOL_PAGE_2_LOCKED_PTR(PVM pVM, PPGMPOOLPAGE pPage)
113	{
114	void *pv = pgmPoolMapPageInlined(&pVM->pgm.s, pPage);
115
116	/* Make sure the dynamic mapping will not be reused. */
117	if (pv)
118	PGMDynLockHCPage(pVM, (uint8_t *)pv);
119
120	return pv;
121	}
122	#else
123	# define PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage) PGMPOOL_PAGE_2_PTR(pVM, pPage)
124	#endif
125
126	/** @def PGMPOOL_UNLOCK_PTR
127	* Unlock a previously locked dynamic caching (RC only).
128	*
129	* @returns VBox status code.
130	* @param pVM The VM handle.
131	* @param pPage The pool page.
132	*
133	* @remark In RC this uses PGMGCDynMapHCPage(), so it will consume of the
134	* small page window employeed by that function. Be careful.
135	* @remark There is no need to assert on the result.
136	*/
137	#if defined(IN_RC)
138	DECLINLINE(void) PGMPOOL_UNLOCK_PTR(PVM pVM, void *pvPage)
139	{
140	if (pvPage)
141	PGMDynUnlockHCPage(pVM, (uint8_t *)pvPage);
142	}
143	#else
144	# define PGMPOOL_UNLOCK_PTR(pVM, pPage) do {} while (0)
145	#endif
146
147
148	#ifdef PGMPOOL_WITH_MONITORING
149	/**
150	* Determin the size of a write instruction.
151	* @returns number of bytes written.
152	* @param pDis The disassembler state.
153	*/
154	static unsigned pgmPoolDisasWriteSize(PDISCPUSTATE pDis)
155	{
156	/*
157	* This is very crude and possibly wrong for some opcodes,
158	* but since it's not really supposed to be called we can
159	* probably live with that.
160	*/
161	return DISGetParamSize(pDis, &pDis->param1);
162	}
163
164
165	/**
166	* Flushes a chain of pages sharing the same access monitor.
167	*
168	* @returns VBox status code suitable for scheduling.
169	* @param pPool The pool.
170	* @param pPage A page in the chain.
171	*/
172	int pgmPoolMonitorChainFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
173	{
174	LogFlow(("pgmPoolMonitorChainFlush: Flush page %RGp type=%d\n", pPage->GCPhys, pPage->enmKind));
175
176	/*
177	* Find the list head.
178	*/
179	uint16_t idx = pPage->idx;
180	if (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
181	{
182	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
183	{
184	idx = pPage->iMonitoredPrev;
185	Assert(idx != pPage->idx);
186	pPage = &pPool->aPages[idx];
187	}
188	}
189
190	/*
191	* Iterate the list flushing each shadow page.
192	*/
193	int rc = VINF_SUCCESS;
194	for (;;)
195	{
196	idx = pPage->iMonitoredNext;
197	Assert(idx != pPage->idx);
198	if (pPage->idx >= PGMPOOL_IDX_FIRST)
199	{
200	int rc2 = pgmPoolFlushPage(pPool, pPage);
201	AssertRC(rc2);
202	}
203	/* next */
204	if (idx == NIL_PGMPOOL_IDX)
205	break;
206	pPage = &pPool->aPages[idx];
207	}
208	return rc;
209	}
210
211
212	/**
213	* Wrapper for getting the current context pointer to the entry being modified.
214	*
215	* @returns VBox status code suitable for scheduling.
216	* @param pVM VM Handle.
217	* @param pvDst Destination address
218	* @param pvSrc Source guest virtual address.
219	* @param GCPhysSrc The source guest physical address.
220	* @param cb Size of data to read
221	*/
222	DECLINLINE(int) pgmPoolPhysSimpleReadGCPhys(PVM pVM, void *pvDst, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvSrc, RTGCPHYS GCPhysSrc, size_t cb)
223	{
224	#if defined(IN_RING3)
225	memcpy(pvDst, (RTHCPTR)((uintptr_t)pvSrc & ~(RTHCUINTPTR)(cb - 1)), cb);
226	return VINF_SUCCESS;
227	#else
228	/* @todo in RC we could attempt to use the virtual address, although this can cause many faults (PAE Windows XP guest). */
229	return PGMPhysSimpleReadGCPhys(pVM, pvDst, GCPhysSrc & ~(RTGCPHYS)(cb - 1), cb);
230	#endif
231	}
232
233	/**
234	* Process shadow entries before they are changed by the guest.
235	*
236	* For PT entries we will clear them. For PD entries, we'll simply check
237	* for mapping conflicts and set the SyncCR3 FF if found.
238	*
239	* @param pVCpu VMCPU handle
240	* @param pPool The pool.
241	* @param pPage The head page.
242	* @param GCPhysFault The guest physical fault address.
243	* @param uAddress In R0 and GC this is the guest context fault address (flat).
244	* In R3 this is the host context 'fault' address.
245	* @param pCpu The disassembler state for figuring out the write size.
246	* This need not be specified if the caller knows we won't do cross entry accesses.
247	*/
248	void pgmPoolMonitorChainChanging(PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhysFault, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvAddress, PDISCPUSTATE pCpu)
249	{
250	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
251	const unsigned off = GCPhysFault & PAGE_OFFSET_MASK;
252	const unsigned cbWrite = (pCpu) ? pgmPoolDisasWriteSize(pCpu) : 0;
253	PVM pVM = pPool->CTX_SUFF(pVM);
254
255	LogFlow(("pgmPoolMonitorChainChanging: %RGv phys=%RGp kind=%s cbWrite=%d\n", (RTGCPTR)pvAddress, GCPhysFault, pgmPoolPoolKindToStr(pPage->enmKind), cbWrite));
256	for (;;)
257	{
258	union
259	{
260	void *pv;
261	PX86PT pPT;
262	PX86PTPAE pPTPae;
263	PX86PD pPD;
264	PX86PDPAE pPDPae;
265	PX86PDPT pPDPT;
266	PX86PML4 pPML4;
267	} uShw;
268
269	uShw.pv = NULL;
270	switch (pPage->enmKind)
271	{
272	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
273	{
274	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
275	const unsigned iShw = off / sizeof(X86PTE);
276	LogFlow(("PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT iShw=%x\n", iShw));
277	if (uShw.pPT->a[iShw].n.u1Present)
278	{
279	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
280	X86PTE GstPte;
281
282	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
283	AssertRC(rc);
284	Log4(("pgmPoolMonitorChainChanging 32_32: deref %016RX64 GCPhys %08RX32\n", uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PG_MASK));
285	pgmPoolTracDerefGCPhysHint(pPool, pPage,
286	uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK,
287	GstPte.u & X86_PTE_PG_MASK);
288	# endif
289	ASMAtomicWriteSize(&uShw.pPT->a[iShw], 0);
290	}
291	break;
292	}
293
294	/* page/2 sized */
295	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
296	{
297	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
298	if (!((off ^ pPage->GCPhys) & (PAGE_SIZE / 2)))
299	{
300	const unsigned iShw = (off / sizeof(X86PTE)) & (X86_PG_PAE_ENTRIES - 1);
301	LogFlow(("PGMPOOLKIND_PAE_PT_FOR_32BIT_PT iShw=%x\n", iShw));
302	if (uShw.pPTPae->a[iShw].n.u1Present)
303	{
304	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
305	X86PTE GstPte;
306	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
307	AssertRC(rc);
308
309	Log4(("pgmPoolMonitorChainChanging pae_32: deref %016RX64 GCPhys %08RX32\n", uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PG_MASK));
310	pgmPoolTracDerefGCPhysHint(pPool, pPage,
311	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
312	GstPte.u & X86_PTE_PG_MASK);
313	# endif
314	ASMAtomicWriteSize(&uShw.pPTPae->a[iShw], 0);
315	}
316	}
317	break;
318	}
319
320	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
321	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
322	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
323	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
324	{
325	unsigned iGst = off / sizeof(X86PDE);
326	unsigned iShwPdpt = iGst / 256;
327	unsigned iShw = (iGst % 256) * 2;
328	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
329
330	LogFlow(("pgmPoolMonitorChainChanging PAE for 32 bits: iGst=%x iShw=%x idx = %d page idx=%d\n", iGst, iShw, iShwPdpt, pPage->enmKind - PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD));
331	if (iShwPdpt == pPage->enmKind - (unsigned)PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD)
332	{
333	for (unsigned i = 0; i < 2; i++)
334	{
335	# ifndef IN_RING0
336	if ((uShw.pPDPae->a[iShw + i].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
337	{
338	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
339	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
340	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw=%#x!\n", iShwPdpt, iShw+i));
341	break;
342	}
343	else
344	# endif /* !IN_RING0 */
345	if (uShw.pPDPae->a[iShw+i].n.u1Present)
346	{
347	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw+i, uShw.pPDPae->a[iShw+i].u));
348	pgmPoolFree(pVM,
349	uShw.pPDPae->a[iShw+i].u & X86_PDE_PAE_PG_MASK,
350	pPage->idx,
351	iShw + i);
352	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw+i], 0);
353	}
354
355	/* paranoia / a bit assumptive. */
356	if ( pCpu
357	&& (off & 3)
358	&& (off & 3) + cbWrite > 4)
359	{
360	const unsigned iShw2 = iShw + 2 + i;
361	if (iShw2 < RT_ELEMENTS(uShw.pPDPae->a))
362	{
363	# ifndef IN_RING0
364	if ((uShw.pPDPae->a[iShw2].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
365	{
366	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
367	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
368	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw2=%#x!\n", iShwPdpt, iShw2));
369	break;
370	}
371	else
372	# endif /* !IN_RING0 */
373	if (uShw.pPDPae->a[iShw2].n.u1Present)
374	{
375	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
376	pgmPoolFree(pVM,
377	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
378	pPage->idx,
379	iShw2);
380	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw2].u, 0);
381	}
382	}
383	}
384	}
385	}
386	break;
387	}
388
389	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
390	{
391	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
392	const unsigned iShw = off / sizeof(X86PTEPAE);
393	if (uShw.pPTPae->a[iShw].n.u1Present)
394	{
395	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
396	X86PTEPAE GstPte;
397	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
398	AssertRC(rc);
399
400	Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PAE_PG_MASK));
401	pgmPoolTracDerefGCPhysHint(pPool, pPage,
402	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
403	GstPte.u & X86_PTE_PAE_PG_MASK);
404	# endif
405	ASMAtomicWriteSize(&uShw.pPTPae->a[iShw].u, 0);
406	}
407
408	/* paranoia / a bit assumptive. */
409	if ( pCpu
410	&& (off & 7)
411	&& (off & 7) + cbWrite > sizeof(X86PTEPAE))
412	{
413	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTEPAE);
414	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPTPae->a));
415
416	if (uShw.pPTPae->a[iShw2].n.u1Present)
417	{
418	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
419	X86PTEPAE GstPte;
420	# ifdef IN_RING3
421	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, (RTHCPTR)((RTHCUINTPTR)pvAddress + sizeof(GstPte)), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
422	# else
423	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress + sizeof(GstPte), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
424	# endif
425	AssertRC(rc);
426	Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PAE_PG_MASK));
427	pgmPoolTracDerefGCPhysHint(pPool, pPage,
428	uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK,
429	GstPte.u & X86_PTE_PAE_PG_MASK);
430	# endif
431	ASMAtomicWriteSize(&uShw.pPTPae->a[iShw2].u ,0);
432	}
433	}
434	break;
435	}
436
437	case PGMPOOLKIND_32BIT_PD:
438	{
439	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
440	const unsigned iShw = off / sizeof(X86PTE); // ASSUMING 32-bit guest paging!
441
442	LogFlow(("pgmPoolMonitorChainChanging: PGMPOOLKIND_32BIT_PD %x\n", iShw));
443	# ifndef IN_RING0
444	if (uShw.pPD->a[iShw].u & PGM_PDFLAGS_MAPPING)
445	{
446	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
447	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
448	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
449	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
450	break;
451	}
452	# endif /* !IN_RING0 */
453	# ifndef IN_RING0
454	else
455	# endif /* !IN_RING0 */
456	{
457	if (uShw.pPD->a[iShw].n.u1Present)
458	{
459	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
460	pgmPoolFree(pVM,
461	uShw.pPD->a[iShw].u & X86_PDE_PAE_PG_MASK,
462	pPage->idx,
463	iShw);
464	ASMAtomicWriteSize(&uShw.pPD->a[iShw].u, 0);
465	}
466	}
467	/* paranoia / a bit assumptive. */
468	if ( pCpu
469	&& (off & 3)
470	&& (off & 3) + cbWrite > sizeof(X86PTE))
471	{
472	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTE);
473	if ( iShw2 != iShw
474	&& iShw2 < RT_ELEMENTS(uShw.pPD->a))
475	{
476	# ifndef IN_RING0
477	if (uShw.pPD->a[iShw2].u & PGM_PDFLAGS_MAPPING)
478	{
479	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
480	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
481	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
482	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
483	break;
484	}
485	# endif /* !IN_RING0 */
486	# ifndef IN_RING0
487	else
488	# endif /* !IN_RING0 */
489	{
490	if (uShw.pPD->a[iShw2].n.u1Present)
491	{
492	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPD->a[iShw2].u));
493	pgmPoolFree(pVM,
494	uShw.pPD->a[iShw2].u & X86_PDE_PAE_PG_MASK,
495	pPage->idx,
496	iShw2);
497	ASMAtomicWriteSize(&uShw.pPD->a[iShw2].u, 0);
498	}
499	}
500	}
501	}
502	#if 0 /* useful when running PGMAssertCR3(), a bit too troublesome for general use (TLBs). */
503	if ( uShw.pPD->a[iShw].n.u1Present
504	&& !VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
505	{
506	LogFlow(("pgmPoolMonitorChainChanging: iShw=%#x: %RX32 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
507	# ifdef IN_RC /* TLB load - we're pushing things a bit... */
508	ASMProbeReadByte(pvAddress);
509	# endif
510	pgmPoolFree(pVM, uShw.pPD->a[iShw].u & X86_PDE_PG_MASK, pPage->idx, iShw);
511	ASMAtomicWriteSize(&uShw.pPD->a[iShw].u, 0);
512	}
513	#endif
514	break;
515	}
516
517	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
518	{
519	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
520	const unsigned iShw = off / sizeof(X86PDEPAE);
521	#ifndef IN_RING0
522	if (uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING)
523	{
524	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
525	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
526	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
527	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
528	break;
529	}
530	#endif /* !IN_RING0 */
531	/*
532	* Causes trouble when the guest uses a PDE to refer to the whole page table level
533	* structure. (Invalidate here; faults later on when it tries to change the page
534	* table entries -> recheck; probably only applies to the RC case.)
535	*/
536	# ifndef IN_RING0
537	else
538	# endif /* !IN_RING0 */
539	{
540	if (uShw.pPDPae->a[iShw].n.u1Present)
541	{
542	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
543	pgmPoolFree(pVM,
544	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
545	pPage->idx,
546	iShw);
547	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw].u, 0);
548	}
549	}
550	/* paranoia / a bit assumptive. */
551	if ( pCpu
552	&& (off & 7)
553	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
554	{
555	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
556	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
557
558	#ifndef IN_RING0
559	if ( iShw2 != iShw
560	&& uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING)
561	{
562	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
563	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
564	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
565	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
566	break;
567	}
568	#endif /* !IN_RING0 */
569	# ifndef IN_RING0
570	else
571	# endif /* !IN_RING0 */
572	if (uShw.pPDPae->a[iShw2].n.u1Present)
573	{
574	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
575	pgmPoolFree(pVM,
576	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
577	pPage->idx,
578	iShw2);
579	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw2].u, 0);
580	}
581	}
582	break;
583	}
584
585	case PGMPOOLKIND_PAE_PDPT:
586	{
587	/*
588	* Hopefully this doesn't happen very often:
589	* - touching unused parts of the page
590	* - messing with the bits of pd pointers without changing the physical address
591	*/
592	/* PDPT roots are not page aligned; 32 byte only! */
593	const unsigned offPdpt = GCPhysFault - pPage->GCPhys;
594
595	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
596	const unsigned iShw = offPdpt / sizeof(X86PDPE);
597	if (iShw < X86_PG_PAE_PDPE_ENTRIES) /* don't use RT_ELEMENTS(uShw.pPDPT->a), because that's for long mode only */
598	{
599	# ifndef IN_RING0
600	if (uShw.pPDPT->a[iShw].u & PGM_PLXFLAGS_MAPPING)
601	{
602	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
603	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
604	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
605	LogFlow(("pgmPoolMonitorChainChanging: Detected pdpt conflict at iShw=%#x!\n", iShw));
606	break;
607	}
608	# endif /* !IN_RING0 */
609	# ifndef IN_RING0
610	else
611	# endif /* !IN_RING0 */
612	if (uShw.pPDPT->a[iShw].n.u1Present)
613	{
614	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
615	pgmPoolFree(pVM,
616	uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK,
617	pPage->idx,
618	iShw);
619	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw].u, 0);
620	}
621
622	/* paranoia / a bit assumptive. */
623	if ( pCpu
624	&& (offPdpt & 7)
625	&& (offPdpt & 7) + cbWrite > sizeof(X86PDPE))
626	{
627	const unsigned iShw2 = (offPdpt + cbWrite - 1) / sizeof(X86PDPE);
628	if ( iShw2 != iShw
629	&& iShw2 < X86_PG_PAE_PDPE_ENTRIES)
630	{
631	# ifndef IN_RING0
632	if (uShw.pPDPT->a[iShw2].u & PGM_PLXFLAGS_MAPPING)
633	{
634	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
635	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
636	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
637	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
638	break;
639	}
640	# endif /* !IN_RING0 */
641	# ifndef IN_RING0
642	else
643	# endif /* !IN_RING0 */
644	if (uShw.pPDPT->a[iShw2].n.u1Present)
645	{
646	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
647	pgmPoolFree(pVM,
648	uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK,
649	pPage->idx,
650	iShw2);
651	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw2].u, 0);
652	}
653	}
654	}
655	}
656	break;
657	}
658
659	#ifndef IN_RC
660	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
661	{
662	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
663	const unsigned iShw = off / sizeof(X86PDEPAE);
664	Assert(!(uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING));
665	if (uShw.pPDPae->a[iShw].n.u1Present)
666	{
667	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
668	pgmPoolFree(pVM,
669	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
670	pPage->idx,
671	iShw);
672	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw].u, 0);
673	}
674	/* paranoia / a bit assumptive. */
675	if ( pCpu
676	&& (off & 7)
677	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
678	{
679	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
680	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
681
682	Assert(!(uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING));
683	if (uShw.pPDPae->a[iShw2].n.u1Present)
684	{
685	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
686	pgmPoolFree(pVM,
687	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
688	pPage->idx,
689	iShw2);
690	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw2].u, 0);
691	}
692	}
693	break;
694	}
695
696	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
697	{
698	/*
699	* Hopefully this doesn't happen very often:
700	* - messing with the bits of pd pointers without changing the physical address
701	*/
702	if (!VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
703	{
704	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
705	const unsigned iShw = off / sizeof(X86PDPE);
706	if (uShw.pPDPT->a[iShw].n.u1Present)
707	{
708	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
709	pgmPoolFree(pVM, uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK, pPage->idx, iShw);
710	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw].u, 0);
711	}
712	/* paranoia / a bit assumptive. */
713	if ( pCpu
714	&& (off & 7)
715	&& (off & 7) + cbWrite > sizeof(X86PDPE))
716	{
717	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDPE);
718	if (uShw.pPDPT->a[iShw2].n.u1Present)
719	{
720	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
721	pgmPoolFree(pVM, uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK, pPage->idx, iShw2);
722	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw2].u, 0);
723	}
724	}
725	}
726	break;
727	}
728
729	case PGMPOOLKIND_64BIT_PML4:
730	{
731	/*
732	* Hopefully this doesn't happen very often:
733	* - messing with the bits of pd pointers without changing the physical address
734	*/
735	if (!VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
736	{
737	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
738	const unsigned iShw = off / sizeof(X86PDPE);
739	if (uShw.pPML4->a[iShw].n.u1Present)
740	{
741	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPML4->a[iShw].u));
742	pgmPoolFree(pVM, uShw.pPML4->a[iShw].u & X86_PML4E_PG_MASK, pPage->idx, iShw);
743	ASMAtomicWriteSize(&uShw.pPML4->a[iShw].u, 0);
744	}
745	/* paranoia / a bit assumptive. */
746	if ( pCpu
747	&& (off & 7)
748	&& (off & 7) + cbWrite > sizeof(X86PDPE))
749	{
750	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PML4E);
751	if (uShw.pPML4->a[iShw2].n.u1Present)
752	{
753	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPML4->a[iShw2].u));
754	pgmPoolFree(pVM, uShw.pPML4->a[iShw2].u & X86_PML4E_PG_MASK, pPage->idx, iShw2);
755	ASMAtomicWriteSize(&uShw.pPML4->a[iShw2].u, 0);
756	}
757	}
758	}
759	break;
760	}
761	#endif /* IN_RING0 */
762
763	default:
764	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
765	}
766	PGMPOOL_UNLOCK_PTR(pVM, uShw.pv);
767
768	/* next */
769	if (pPage->iMonitoredNext == NIL_PGMPOOL_IDX)
770	return;
771	pPage = &pPool->aPages[pPage->iMonitoredNext];
772	}
773	}
774
775	# ifndef IN_RING3
776	/**
777	* Checks if a access could be a fork operation in progress.
778	*
779	* Meaning, that the guest is setuping up the parent process for Copy-On-Write.
780	*
781	* @returns true if it's likly that we're forking, otherwise false.
782	* @param pPool The pool.
783	* @param pCpu The disassembled instruction.
784	* @param offFault The access offset.
785	*/
786	DECLINLINE(bool) pgmPoolMonitorIsForking(PPGMPOOL pPool, PDISCPUSTATE pCpu, unsigned offFault)
787	{
788	/*
789	* i386 linux is using btr to clear X86_PTE_RW.
790	* The functions involved are (2.6.16 source inspection):
791	* clear_bit
792	* ptep_set_wrprotect
793	* copy_one_pte
794	* copy_pte_range
795	* copy_pmd_range
796	* copy_pud_range
797	* copy_page_range
798	* dup_mmap
799	* dup_mm
800	* copy_mm
801	* copy_process
802	* do_fork
803	*/
804	if ( pCpu->pCurInstr->opcode == OP_BTR
805	&& !(offFault & 4)
806	/** @todo Validate that the bit index is X86_PTE_RW. */
807	)
808	{
809	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,Fork));
810	return true;
811	}
812	return false;
813	}
814
815
816	/**
817	* Determine whether the page is likely to have been reused.
818	*
819	* @returns true if we consider the page as being reused for a different purpose.
820	* @returns false if we consider it to still be a paging page.
821	* @param pVM VM Handle.
822	* @param pRegFrame Trap register frame.
823	* @param pCpu The disassembly info for the faulting instruction.
824	* @param pvFault The fault address.
825	*
826	* @remark The REP prefix check is left to the caller because of STOSD/W.
827	*/
828	DECLINLINE(bool) pgmPoolMonitorIsReused(PVM pVM, PCPUMCTXCORE pRegFrame, PDISCPUSTATE pCpu, RTGCPTR pvFault)
829	{
830	#ifndef IN_RC
831	/** @todo could make this general, faulting close to rsp should be safe reuse heuristic. */
832	if ( HWACCMHasPendingIrq(pVM)
833	&& (pRegFrame->rsp - pvFault) < 32)
834	{
835	/* Fault caused by stack writes while trying to inject an interrupt event. */
836	Log(("pgmPoolMonitorIsReused: reused %RGv for interrupt stack (rsp=%RGv).\n", pvFault, pRegFrame->rsp));
837	return true;
838	}
839	#else
840	NOREF(pVM); NOREF(pvFault);
841	#endif
842
843	switch (pCpu->pCurInstr->opcode)
844	{
845	/* call implies the actual push of the return address faulted */
846	case OP_CALL:
847	Log4(("pgmPoolMonitorIsReused: CALL\n"));
848	return true;
849	case OP_PUSH:
850	Log4(("pgmPoolMonitorIsReused: PUSH\n"));
851	return true;
852	case OP_PUSHF:
853	Log4(("pgmPoolMonitorIsReused: PUSHF\n"));
854	return true;
855	case OP_PUSHA:
856	Log4(("pgmPoolMonitorIsReused: PUSHA\n"));
857	return true;
858	case OP_FXSAVE:
859	Log4(("pgmPoolMonitorIsReused: FXSAVE\n"));
860	return true;
861	case OP_MOVNTI: /* solaris - block_zero_no_xmm */
862	Log4(("pgmPoolMonitorIsReused: MOVNTI\n"));
863	return true;
864	case OP_MOVNTDQ: /* solaris - hwblkclr & hwblkpagecopy */
865	Log4(("pgmPoolMonitorIsReused: MOVNTDQ\n"));
866	return true;
867	case OP_MOVSWD:
868	case OP_STOSWD:
869	if ( pCpu->prefix == (PREFIX_REP\|PREFIX_REX)
870	&& pRegFrame->rcx >= 0x40
871	)
872	{
873	Assert(pCpu->mode == CPUMODE_64BIT);
874
875	Log(("pgmPoolMonitorIsReused: OP_STOSQ\n"));
876	return true;
877	}
878	return false;
879	}
880	if ( (pCpu->param1.flags & USE_REG_GEN32)
881	&& (pCpu->param1.base.reg_gen == USE_REG_ESP))
882	{
883	Log4(("pgmPoolMonitorIsReused: ESP\n"));
884	return true;
885	}
886
887	return false;
888	}
889
890
891	/**
892	* Flushes the page being accessed.
893	*
894	* @returns VBox status code suitable for scheduling.
895	* @param pVM The VM handle.
896	* @param pVCpu The VMCPU handle.
897	* @param pPool The pool.
898	* @param pPage The pool page (head).
899	* @param pCpu The disassembly of the write instruction.
900	* @param pRegFrame The trap register frame.
901	* @param GCPhysFault The fault address as guest physical address.
902	* @param pvFault The fault address.
903	*/
904	static int pgmPoolAccessHandlerFlush(PVM pVM, PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
905	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
906	{
907	/*
908	* First, do the flushing.
909	*/
910	int rc = pgmPoolMonitorChainFlush(pPool, pPage);
911
912	/*
913	* Emulate the instruction (xp/w2k problem, requires pc/cr2/sp detection).
914	*/
915	uint32_t cbWritten;
916	int rc2 = EMInterpretInstructionCPU(pVM, pVCpu, pCpu, pRegFrame, pvFault, &cbWritten);
917	if (RT_SUCCESS(rc2))
918	pRegFrame->rip += pCpu->opsize;
919	else if (rc2 == VERR_EM_INTERPRETER)
920	{
921	#ifdef IN_RC
922	if (PATMIsPatchGCAddr(pVM, (RTRCPTR)pRegFrame->eip))
923	{
924	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for patch code %04x:%RGv, ignoring.\n",
925	pRegFrame->cs, (RTGCPTR)pRegFrame->eip));
926	rc = VINF_SUCCESS;
927	STAM_COUNTER_INC(&pPool->StatMonitorRZIntrFailPatch2);
928	}
929	else
930	#endif
931	{
932	rc = VINF_EM_RAW_EMULATE_INSTR;
933	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
934	}
935	}
936	else
937	rc = rc2;
938
939	/* See use in pgmPoolAccessHandlerSimple(). */
940	PGM_INVL_VCPU_TLBS(pVCpu);
941
942	LogFlow(("pgmPoolAccessHandlerPT: returns %Rrc (flushed)\n", rc));
943	return rc;
944
945	}
946
947
948	/**
949	* Handles the STOSD write accesses.
950	*
951	* @returns VBox status code suitable for scheduling.
952	* @param pVM The VM handle.
953	* @param pPool The pool.
954	* @param pPage The pool page (head).
955	* @param pCpu The disassembly of the write instruction.
956	* @param pRegFrame The trap register frame.
957	* @param GCPhysFault The fault address as guest physical address.
958	* @param pvFault The fault address.
959	*/
960	DECLINLINE(int) pgmPoolAccessHandlerSTOSD(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
961	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
962	{
963	Assert(pCpu->mode == CPUMODE_32BIT);
964
965	Log3(("pgmPoolAccessHandlerSTOSD\n"));
966
967	/*
968	* Increment the modification counter and insert it into the list
969	* of modified pages the first time.
970	*/
971	if (!pPage->cModifications++)
972	pgmPoolMonitorModifiedInsert(pPool, pPage);
973
974	/*
975	* Execute REP STOSD.
976	*
977	* This ASSUMES that we're not invoked by Trap0e on in a out-of-sync
978	* write situation, meaning that it's safe to write here.
979	*/
980	PVMCPU pVCpu = VMMGetCpu(pPool->CTX_SUFF(pVM));
981	RTGCUINTPTR pu32 = (RTGCUINTPTR)pvFault;
982	while (pRegFrame->ecx)
983	{
984	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
985	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
986	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, (RTGCPTR)pu32, NULL);
987	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
988	#else
989	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, (RTGCPTR)pu32, NULL);
990	#endif
991	#ifdef IN_RC
992	(uint32_t )pu32 = pRegFrame->eax;
993	#else
994	PGMPhysSimpleWriteGCPhys(pVM, GCPhysFault, &pRegFrame->eax, 4);
995	#endif
996	pu32 += 4;
997	GCPhysFault += 4;
998	pRegFrame->edi += 4;
999	pRegFrame->ecx--;
1000	}
1001	pRegFrame->rip += pCpu->opsize;
1002
1003	#ifdef IN_RC
1004	/* See use in pgmPoolAccessHandlerSimple(). */
1005	PGM_INVL_VCPU_TLBS(pVCpu);
1006	#endif
1007
1008	LogFlow(("pgmPoolAccessHandlerSTOSD: returns\n"));
1009	return VINF_SUCCESS;
1010	}
1011
1012
1013	/**
1014	* Handles the simple write accesses.
1015	*
1016	* @returns VBox status code suitable for scheduling.
1017	* @param pVM The VM handle.
1018	* @param pVCpu The VMCPU handle.
1019	* @param pPool The pool.
1020	* @param pPage The pool page (head).
1021	* @param pCpu The disassembly of the write instruction.
1022	* @param pRegFrame The trap register frame.
1023	* @param GCPhysFault The fault address as guest physical address.
1024	* @param pvFault The fault address.
1025	*/
1026	DECLINLINE(int) pgmPoolAccessHandlerSimple(PVM pVM, PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
1027	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
1028	{
1029	Log3(("pgmPoolAccessHandlerSimple\n"));
1030	/*
1031	* Increment the modification counter and insert it into the list
1032	* of modified pages the first time.
1033	*/
1034	if (!pPage->cModifications++)
1035	pgmPoolMonitorModifiedInsert(pPool, pPage);
1036
1037	/*
1038	* Clear all the pages. ASSUMES that pvFault is readable.
1039	*/
1040	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
1041	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
1042	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, pvFault, pCpu);
1043	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
1044	#else
1045	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, pvFault, pCpu);
1046	#endif
1047
1048	/*
1049	* Interpret the instruction.
1050	*/
1051	uint32_t cb;
1052	int rc = EMInterpretInstructionCPU(pVM, pVCpu, pCpu, pRegFrame, pvFault, &cb);
1053	if (RT_SUCCESS(rc))
1054	pRegFrame->rip += pCpu->opsize;
1055	else if (rc == VERR_EM_INTERPRETER)
1056	{
1057	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for %04x:%RGv - opcode=%d\n",
1058	pRegFrame->cs, (RTGCPTR)pRegFrame->rip, pCpu->pCurInstr->opcode));
1059	rc = VINF_EM_RAW_EMULATE_INSTR;
1060	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
1061	}
1062
1063	#ifdef IN_RC
1064	/*
1065	* Quick hack, with logging enabled we're getting stale
1066	* code TLBs but no data TLB for EIP and crash in EMInterpretDisasOne.
1067	* Flushing here is BAD and expensive, I think EMInterpretDisasOne will
1068	* have to be fixed to support this. But that'll have to wait till next week.
1069	*
1070	* An alternative is to keep track of the changed PTEs together with the
1071	* GCPhys from the guest PT. This may proove expensive though.
1072	*
1073	* At the moment, it's VITAL that it's done AFTER the instruction interpreting
1074	* because we need the stale TLBs in some cases (XP boot). This MUST be fixed properly!
1075	*/
1076	PGM_INVL_VCPU_TLBS(pVCpu);
1077	#endif
1078
1079	LogFlow(("pgmPoolAccessHandlerSimple: returns %Rrc cb=%d\n", rc, cb));
1080	return rc;
1081	}
1082
1083	/**
1084	* \#PF Handler callback for PT write accesses.
1085	*
1086	* @returns VBox status code (appropriate for GC return).
1087	* @param pVM VM Handle.
1088	* @param uErrorCode CPU Error code.
1089	* @param pRegFrame Trap register frame.
1090	* NULL on DMA and other non CPU access.
1091	* @param pvFault The fault address (cr2).
1092	* @param GCPhysFault The GC physical address corresponding to pvFault.
1093	* @param pvUser User argument.
1094	*/
1095	DECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser)
1096	{
1097	STAM_PROFILE_START(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), a);
1098	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1099	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
1100	PVMCPU pVCpu = VMMGetCpu(pVM);
1101
1102	LogFlow(("pgmPoolAccessHandler: pvFault=%RGv pPage=%p:{.idx=%d} GCPhysFault=%RGp\n", pvFault, pPage, pPage->idx, GCPhysFault));
1103
1104	/*
1105	* We should ALWAYS have the list head as user parameter. This
1106	* is because we use that page to record the changes.
1107	*/
1108	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1109
1110	/*
1111	* Disassemble the faulting instruction.
1112	*/
1113	DISCPUSTATE Cpu;
1114	int rc = EMInterpretDisasOne(pVM, pVCpu, pRegFrame, &Cpu, NULL);
1115	AssertRCReturn(rc, rc);
1116
1117	pgmLock(pVM);
1118	if (PHYS_PAGE_ADDRESS(GCPhysFault) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
1119	{
1120	/* Pool page changed while we were waiting for the lock; ignore. */
1121	Log(("CPU%d: pgmPoolAccessHandler pgm pool page for %RGp changed (to %RGp) while waiting!\n", pVCpu->idCpu, PHYS_PAGE_ADDRESS(GCPhysFault), PHYS_PAGE_ADDRESS(pPage->GCPhys)));
1122	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,Handled), a);
1123	pgmUnlock(pVM);
1124	return VINF_SUCCESS;
1125	}
1126
1127	/*
1128	* Check if it's worth dealing with.
1129	*/
1130	bool fReused = false;
1131	if ( ( pPage->cModifications < 48 /** @todo #define / /* @todo need to check that it's not mapping EIP. / /* @todo adjust this! */
1132	\|\| pgmPoolIsPageLocked(&pVM->pgm.s, pPage)
1133	)
1134	&& !(fReused = pgmPoolMonitorIsReused(pVM, pRegFrame, &Cpu, pvFault))
1135	&& !pgmPoolMonitorIsForking(pPool, &Cpu, GCPhysFault & PAGE_OFFSET_MASK))
1136	{
1137	/*
1138	* Simple instructions, no REP prefix.
1139	*/
1140	if (!(Cpu.prefix & (PREFIX_REP \| PREFIX_REPNE)))
1141	{
1142	rc = pgmPoolAccessHandlerSimple(pVM, pVCpu, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
1143	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,Handled), a);
1144	pgmUnlock(pVM);
1145	return rc;
1146	}
1147
1148	/*
1149	* Windows is frequently doing small memset() operations (netio test 4k+).
1150	* We have to deal with these or we'll kill the cache and performance.
1151	*/
1152	if ( Cpu.pCurInstr->opcode == OP_STOSWD
1153	&& CPUMGetGuestCPL(pVCpu, pRegFrame) == 0
1154	&& pRegFrame->ecx <= 0x20
1155	&& pRegFrame->ecx * 4 <= PAGE_SIZE - ((uintptr_t)pvFault & PAGE_OFFSET_MASK)
1156	&& !((uintptr_t)pvFault & 3)
1157	&& (pRegFrame->eax == 0 \|\| pRegFrame->eax == 0x80) /* the two values observed. */
1158	&& Cpu.mode == CPUMODE_32BIT
1159	&& Cpu.opmode == CPUMODE_32BIT
1160	&& Cpu.addrmode == CPUMODE_32BIT
1161	&& Cpu.prefix == PREFIX_REP
1162	&& !pRegFrame->eflags.Bits.u1DF
1163	)
1164	{
1165	rc = pgmPoolAccessHandlerSTOSD(pVM, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
1166	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,RepStosd), a);
1167	pgmUnlock(pVM);
1168	return rc;
1169	}
1170
1171	/* REP prefix, don't bother. */
1172	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,RepPrefix));
1173	Log4(("pgmPoolAccessHandler: eax=%#x ecx=%#x edi=%#x esi=%#x rip=%RGv opcode=%d prefix=%#x\n",
1174	pRegFrame->eax, pRegFrame->ecx, pRegFrame->edi, pRegFrame->esi, (RTGCPTR)pRegFrame->rip, Cpu.pCurInstr->opcode, Cpu.prefix));
1175	}
1176
1177	/*
1178	* Not worth it, so flush it.
1179	*
1180	* If we considered it to be reused, don't go back to ring-3
1181	* to emulate failed instructions since we usually cannot
1182	* interpret then. This may be a bit risky, in which case
1183	* the reuse detection must be fixed.
1184	*/
1185	rc = pgmPoolAccessHandlerFlush(pVM, pVCpu, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
1186	if (rc == VINF_EM_RAW_EMULATE_INSTR && fReused)
1187	rc = VINF_SUCCESS;
1188	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,FlushPage), a);
1189	pgmUnlock(pVM);
1190	return rc;
1191	}
1192
1193	# endif /* !IN_RING3 */
1194	#endif /* PGMPOOL_WITH_MONITORING */
1195
1196	#ifdef PGMPOOL_WITH_CACHE
1197
1198	/**
1199	* Inserts a page into the GCPhys hash table.
1200	*
1201	* @param pPool The pool.
1202	* @param pPage The page.
1203	*/
1204	DECLINLINE(void) pgmPoolHashInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1205	{
1206	Log3(("pgmPoolHashInsert: %RGp\n", pPage->GCPhys));
1207	Assert(pPage->GCPhys != NIL_RTGCPHYS); Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1208	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1209	pPage->iNext = pPool->aiHash[iHash];
1210	pPool->aiHash[iHash] = pPage->idx;
1211	}
1212
1213
1214	/**
1215	* Removes a page from the GCPhys hash table.
1216	*
1217	* @param pPool The pool.
1218	* @param pPage The page.
1219	*/
1220	DECLINLINE(void) pgmPoolHashRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1221	{
1222	Log3(("pgmPoolHashRemove: %RGp\n", pPage->GCPhys));
1223	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1224	if (pPool->aiHash[iHash] == pPage->idx)
1225	pPool->aiHash[iHash] = pPage->iNext;
1226	else
1227	{
1228	uint16_t iPrev = pPool->aiHash[iHash];
1229	for (;;)
1230	{
1231	const int16_t i = pPool->aPages[iPrev].iNext;
1232	if (i == pPage->idx)
1233	{
1234	pPool->aPages[iPrev].iNext = pPage->iNext;
1235	break;
1236	}
1237	if (i == NIL_PGMPOOL_IDX)
1238	{
1239	AssertReleaseMsgFailed(("GCPhys=%RGp idx=%#x\n", pPage->GCPhys, pPage->idx));
1240	break;
1241	}
1242	iPrev = i;
1243	}
1244	}
1245	pPage->iNext = NIL_PGMPOOL_IDX;
1246	}
1247
1248
1249	/**
1250	* Frees up one cache page.
1251	*
1252	* @returns VBox status code.
1253	* @retval VINF_SUCCESS on success.
1254	* @param pPool The pool.
1255	* @param iUser The user index.
1256	*/
1257	static int pgmPoolCacheFreeOne(PPGMPOOL pPool, uint16_t iUser)
1258	{
1259	#ifndef IN_RC
1260	const PVM pVM = pPool->CTX_SUFF(pVM);
1261	#endif
1262	Assert(pPool->iAgeHead != pPool->iAgeTail); /* We shouldn't be here if there < 2 cached entries! */
1263	STAM_COUNTER_INC(&pPool->StatCacheFreeUpOne);
1264
1265	/*
1266	* Select one page from the tail of the age list.
1267	*/
1268	PPGMPOOLPAGE pPage;
1269	for (unsigned iLoop = 0; ; iLoop++)
1270	{
1271	uint16_t iToFree = pPool->iAgeTail;
1272	if (iToFree == iUser)
1273	iToFree = pPool->aPages[iToFree].iAgePrev;
1274	/* This is the alternative to the SyncCR3 pgmPoolCacheUsed calls.
1275	if (pPool->aPages[iToFree].iUserHead != NIL_PGMPOOL_USER_INDEX)
1276	{
1277	uint16_t i = pPool->aPages[iToFree].iAgePrev;
1278	for (unsigned j = 0; j < 10 && i != NIL_PGMPOOL_USER_INDEX; j++, i = pPool->aPages[i].iAgePrev)
1279	{
1280	if (pPool->aPages[iToFree].iUserHead == NIL_PGMPOOL_USER_INDEX)
1281	continue;
1282	iToFree = i;
1283	break;
1284	}
1285	}
1286	*/
1287	Assert(iToFree != iUser);
1288	AssertRelease(iToFree != NIL_PGMPOOL_IDX);
1289	pPage = &pPool->aPages[iToFree];
1290
1291	/*
1292	* Reject any attempts at flushing the currently active shadow CR3 mapping.
1293	* Call pgmPoolCacheUsed to move the page to the head of the age list.
1294	*/
1295	if (!pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage))
1296	break;
1297	LogFlow(("pgmPoolCacheFreeOne: refuse CR3 mapping\n"));
1298	pgmPoolCacheUsed(pPool, pPage);
1299	AssertLogRelReturn(iLoop < 8192, VERR_INTERNAL_ERROR);
1300	}
1301
1302	/*
1303	* Found a usable page, flush it and return.
1304	*/
1305	int rc = pgmPoolFlushPage(pPool, pPage);
1306	/* This flush was initiated by us and not the guest, so explicitly flush the TLB. */
1307	if (rc == VINF_SUCCESS)
1308	PGM_INVL_ALL_VCPU_TLBS(pVM);
1309	return rc;
1310	}
1311
1312
1313	/**
1314	* Checks if a kind mismatch is really a page being reused
1315	* or if it's just normal remappings.
1316	*
1317	* @returns true if reused and the cached page (enmKind1) should be flushed
1318	* @returns false if not reused.
1319	* @param enmKind1 The kind of the cached page.
1320	* @param enmKind2 The kind of the requested page.
1321	*/
1322	static bool pgmPoolCacheReusedByKind(PGMPOOLKIND enmKind1, PGMPOOLKIND enmKind2)
1323	{
1324	switch (enmKind1)
1325	{
1326	/*
1327	* Never reuse them. There is no remapping in non-paging mode.
1328	*/
1329	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1330	case PGMPOOLKIND_32BIT_PD_PHYS:
1331	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1332	case PGMPOOLKIND_PAE_PD_PHYS:
1333	case PGMPOOLKIND_PAE_PDPT_PHYS:
1334	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1335	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1336	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1337	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1338	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1339	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT: /* never reuse them for other types */
1340	return false;
1341
1342	/*
1343	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1344	*/
1345	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1346	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1347	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1348	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1349	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1350	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1351	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1352	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1353	case PGMPOOLKIND_32BIT_PD:
1354	case PGMPOOLKIND_PAE_PDPT:
1355	switch (enmKind2)
1356	{
1357	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1358	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1359	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1360	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1361	case PGMPOOLKIND_64BIT_PML4:
1362	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1363	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1364	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1365	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1366	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1367	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1368	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1369	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1370	return true;
1371	default:
1372	return false;
1373	}
1374
1375	/*
1376	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1377	*/
1378	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1379	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1380	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1381	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1382	case PGMPOOLKIND_64BIT_PML4:
1383	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1384	switch (enmKind2)
1385	{
1386	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1387	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1388	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1389	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1390	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1391	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1392	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1393	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1394	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1395	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1396	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1397	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1398	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1399	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1400	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1401	return true;
1402	default:
1403	return false;
1404	}
1405
1406	/*
1407	* These cannot be flushed, and it's common to reuse the PDs as PTs.
1408	*/
1409	case PGMPOOLKIND_ROOT_NESTED:
1410	return false;
1411
1412	default:
1413	AssertFatalMsgFailed(("enmKind1=%d\n", enmKind1));
1414	}
1415	}
1416
1417
1418	/**
1419	* Attempts to satisfy a pgmPoolAlloc request from the cache.
1420	*
1421	* @returns VBox status code.
1422	* @retval VINF_PGM_CACHED_PAGE on success.
1423	* @retval VERR_FILE_NOT_FOUND if not found.
1424	* @param pPool The pool.
1425	* @param GCPhys The GC physical address of the page we're gonna shadow.
1426	* @param enmKind The kind of mapping.
1427	* @param enmAccess Access type for the mapping (only relevant for big pages)
1428	* @param iUser The shadow page pool index of the user table.
1429	* @param iUserTable The index into the user table (shadowed).
1430	* @param ppPage Where to store the pointer to the page.
1431	*/
1432	static int pgmPoolCacheAlloc(PPGMPOOL pPool, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, PGMPOOLACCESS enmAccess, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
1433	{
1434	#ifndef IN_RC
1435	const PVM pVM = pPool->CTX_SUFF(pVM);
1436	#endif
1437	/*
1438	* Look up the GCPhys in the hash.
1439	*/
1440	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
1441	Log3(("pgmPoolCacheAlloc: %RGp kind %s iUser=%x iUserTable=%x SLOT=%d\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable, i));
1442	if (i != NIL_PGMPOOL_IDX)
1443	{
1444	do
1445	{
1446	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1447	Log4(("pgmPoolCacheAlloc: slot %d found page %RGp\n", i, pPage->GCPhys));
1448	if (pPage->GCPhys == GCPhys)
1449	{
1450	if ( (PGMPOOLKIND)pPage->enmKind == enmKind
1451	&& (PGMPOOLACCESS)pPage->enmAccess == enmAccess)
1452	{
1453	/* Put it at the start of the use list to make sure pgmPoolTrackAddUser
1454	* doesn't flush it in case there are no more free use records.
1455	*/
1456	pgmPoolCacheUsed(pPool, pPage);
1457
1458	int rc = pgmPoolTrackAddUser(pPool, pPage, iUser, iUserTable);
1459	if (RT_SUCCESS(rc))
1460	{
1461	Assert((PGMPOOLKIND)pPage->enmKind == enmKind);
1462	*ppPage = pPage;
1463	STAM_COUNTER_INC(&pPool->StatCacheHits);
1464	return VINF_PGM_CACHED_PAGE;
1465	}
1466	return rc;
1467	}
1468
1469	if ((PGMPOOLKIND)pPage->enmKind != enmKind)
1470	{
1471	/*
1472	* The kind is different. In some cases we should now flush the page
1473	* as it has been reused, but in most cases this is normal remapping
1474	* of PDs as PT or big pages using the GCPhys field in a slightly
1475	* different way than the other kinds.
1476	*/
1477	if (pgmPoolCacheReusedByKind((PGMPOOLKIND)pPage->enmKind, enmKind))
1478	{
1479	STAM_COUNTER_INC(&pPool->StatCacheKindMismatches);
1480	pgmPoolFlushPage(pPool, pPage);
1481	PGM_INVL_VCPU_TLBS(VMMGetCpu(pVM)); /* see PT handler. */
1482	break;
1483	}
1484	}
1485	}
1486
1487	/* next */
1488	i = pPage->iNext;
1489	} while (i != NIL_PGMPOOL_IDX);
1490	}
1491
1492	Log3(("pgmPoolCacheAlloc: Missed GCPhys=%RGp enmKind=%s\n", GCPhys, pgmPoolPoolKindToStr(enmKind)));
1493	STAM_COUNTER_INC(&pPool->StatCacheMisses);
1494	return VERR_FILE_NOT_FOUND;
1495	}
1496
1497
1498	/**
1499	* Inserts a page into the cache.
1500	*
1501	* @param pPool The pool.
1502	* @param pPage The cached page.
1503	* @param fCanBeCached Set if the page is fit for caching from the caller's point of view.
1504	*/
1505	static void pgmPoolCacheInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fCanBeCached)
1506	{
1507	/*
1508	* Insert into the GCPhys hash if the page is fit for that.
1509	*/
1510	Assert(!pPage->fCached);
1511	if (fCanBeCached)
1512	{
1513	pPage->fCached = true;
1514	pgmPoolHashInsert(pPool, pPage);
1515	Log3(("pgmPoolCacheInsert: Caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
1516	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
1517	STAM_COUNTER_INC(&pPool->StatCacheCacheable);
1518	}
1519	else
1520	{
1521	Log3(("pgmPoolCacheInsert: Not caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
1522	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
1523	STAM_COUNTER_INC(&pPool->StatCacheUncacheable);
1524	}
1525
1526	/*
1527	* Insert at the head of the age list.
1528	*/
1529	pPage->iAgePrev = NIL_PGMPOOL_IDX;
1530	pPage->iAgeNext = pPool->iAgeHead;
1531	if (pPool->iAgeHead != NIL_PGMPOOL_IDX)
1532	pPool->aPages[pPool->iAgeHead].iAgePrev = pPage->idx;
1533	else
1534	pPool->iAgeTail = pPage->idx;
1535	pPool->iAgeHead = pPage->idx;
1536	}
1537
1538
1539	/**
1540	* Flushes a cached page.
1541	*
1542	* @param pPool The pool.
1543	* @param pPage The cached page.
1544	*/
1545	static void pgmPoolCacheFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1546	{
1547	Log3(("pgmPoolCacheFlushPage: %RGp\n", pPage->GCPhys));
1548
1549	/*
1550	* Remove the page from the hash.
1551	*/
1552	if (pPage->fCached)
1553	{
1554	pPage->fCached = false;
1555	pgmPoolHashRemove(pPool, pPage);
1556	}
1557	else
1558	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1559
1560	/*
1561	* Remove it from the age list.
1562	*/
1563	if (pPage->iAgeNext != NIL_PGMPOOL_IDX)
1564	pPool->aPages[pPage->iAgeNext].iAgePrev = pPage->iAgePrev;
1565	else
1566	pPool->iAgeTail = pPage->iAgePrev;
1567	if (pPage->iAgePrev != NIL_PGMPOOL_IDX)
1568	pPool->aPages[pPage->iAgePrev].iAgeNext = pPage->iAgeNext;
1569	else
1570	pPool->iAgeHead = pPage->iAgeNext;
1571	pPage->iAgeNext = NIL_PGMPOOL_IDX;
1572	pPage->iAgePrev = NIL_PGMPOOL_IDX;
1573	}
1574
1575	#endif /* PGMPOOL_WITH_CACHE */
1576	#ifdef PGMPOOL_WITH_MONITORING
1577
1578	/**
1579	* Looks for pages sharing the monitor.
1580	*
1581	* @returns Pointer to the head page.
1582	* @returns NULL if not found.
1583	* @param pPool The Pool
1584	* @param pNewPage The page which is going to be monitored.
1585	*/
1586	static PPGMPOOLPAGE pgmPoolMonitorGetPageByGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pNewPage)
1587	{
1588	#ifdef PGMPOOL_WITH_CACHE
1589	/*
1590	* Look up the GCPhys in the hash.
1591	*/
1592	RTGCPHYS GCPhys = pNewPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1593	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
1594	if (i == NIL_PGMPOOL_IDX)
1595	return NULL;
1596	do
1597	{
1598	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1599	if ( pPage->GCPhys - GCPhys < PAGE_SIZE
1600	&& pPage != pNewPage)
1601	{
1602	switch (pPage->enmKind)
1603	{
1604	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1605	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1606	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1607	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1608	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1609	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1610	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1611	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1612	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1613	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1614	case PGMPOOLKIND_64BIT_PML4:
1615	case PGMPOOLKIND_32BIT_PD:
1616	case PGMPOOLKIND_PAE_PDPT:
1617	{
1618	/* find the head */
1619	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1620	{
1621	Assert(pPage->iMonitoredPrev != pPage->idx);
1622	pPage = &pPool->aPages[pPage->iMonitoredPrev];
1623	}
1624	return pPage;
1625	}
1626
1627	/* ignore, no monitoring. */
1628	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1629	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1630	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1631	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1632	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1633	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1634	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1635	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1636	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1637	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1638	case PGMPOOLKIND_ROOT_NESTED:
1639	case PGMPOOLKIND_PAE_PD_PHYS:
1640	case PGMPOOLKIND_PAE_PDPT_PHYS:
1641	case PGMPOOLKIND_32BIT_PD_PHYS:
1642	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
1643	break;
1644	default:
1645	AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
1646	}
1647	}
1648
1649	/* next */
1650	i = pPage->iNext;
1651	} while (i != NIL_PGMPOOL_IDX);
1652	#endif
1653	return NULL;
1654	}
1655
1656
1657	/**
1658	* Enabled write monitoring of a guest page.
1659	*
1660	* @returns VBox status code.
1661	* @retval VINF_SUCCESS on success.
1662	* @param pPool The pool.
1663	* @param pPage The cached page.
1664	*/
1665	static int pgmPoolMonitorInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1666	{
1667	LogFlow(("pgmPoolMonitorInsert %RGp\n", pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1)));
1668
1669	/*
1670	* Filter out the relevant kinds.
1671	*/
1672	switch (pPage->enmKind)
1673	{
1674	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1675	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1676	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1677	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1678	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1679	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1680	case PGMPOOLKIND_64BIT_PML4:
1681	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1682	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1683	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1684	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1685	case PGMPOOLKIND_32BIT_PD:
1686	case PGMPOOLKIND_PAE_PDPT:
1687	break;
1688
1689	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1690	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1691	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1692	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1693	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1694	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1695	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1696	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1697	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1698	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1699	case PGMPOOLKIND_ROOT_NESTED:
1700	/* Nothing to monitor here. */
1701	return VINF_SUCCESS;
1702
1703	case PGMPOOLKIND_32BIT_PD_PHYS:
1704	case PGMPOOLKIND_PAE_PDPT_PHYS:
1705	case PGMPOOLKIND_PAE_PD_PHYS:
1706	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
1707	/* Nothing to monitor here. */
1708	return VINF_SUCCESS;
1709	#ifdef PGMPOOL_WITH_MIXED_PT_CR3
1710	break;
1711	#else
1712	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1713	#endif
1714	default:
1715	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
1716	}
1717
1718	/*
1719	* Install handler.
1720	*/
1721	int rc;
1722	PPGMPOOLPAGE pPageHead = pgmPoolMonitorGetPageByGCPhys(pPool, pPage);
1723	if (pPageHead)
1724	{
1725	Assert(pPageHead != pPage); Assert(pPageHead->iMonitoredNext != pPage->idx);
1726	Assert(pPageHead->iMonitoredPrev != pPage->idx);
1727	pPage->iMonitoredPrev = pPageHead->idx;
1728	pPage->iMonitoredNext = pPageHead->iMonitoredNext;
1729	if (pPageHead->iMonitoredNext != NIL_PGMPOOL_IDX)
1730	pPool->aPages[pPageHead->iMonitoredNext].iMonitoredPrev = pPage->idx;
1731	pPageHead->iMonitoredNext = pPage->idx;
1732	rc = VINF_SUCCESS;
1733	}
1734	else
1735	{
1736	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX); Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1737	PVM pVM = pPool->CTX_SUFF(pVM);
1738	const RTGCPHYS GCPhysPage = pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1739	rc = PGMHandlerPhysicalRegisterEx(pVM, PGMPHYSHANDLERTYPE_PHYSICAL_WRITE,
1740	GCPhysPage, GCPhysPage + (PAGE_SIZE - 1),
1741	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
1742	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
1743	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
1744	pPool->pszAccessHandler);
1745	/** @todo we should probably deal with out-of-memory conditions here, but for now increasing
1746	* the heap size should suffice. */
1747	AssertFatalMsgRC(rc, ("PGMHandlerPhysicalRegisterEx %RGp failed with %Rrc\n", GCPhysPage, rc));
1748	Assert(!(pVM->pgm.s.fGlobalSyncFlags & PGM_GLOBAL_SYNC_CLEAR_PGM_POOL) \|\| VMCPU_FF_ISSET(VMMGetCpu(pVM), VMCPU_FF_PGM_SYNC_CR3));
1749	}
1750	pPage->fMonitored = true;
1751	return rc;
1752	}
1753
1754
1755	/**
1756	* Disables write monitoring of a guest page.
1757	*
1758	* @returns VBox status code.
1759	* @retval VINF_SUCCESS on success.
1760	* @param pPool The pool.
1761	* @param pPage The cached page.
1762	*/
1763	static int pgmPoolMonitorFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1764	{
1765	/*
1766	* Filter out the relevant kinds.
1767	*/
1768	switch (pPage->enmKind)
1769	{
1770	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1771	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1772	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1773	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1774	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1775	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1776	case PGMPOOLKIND_64BIT_PML4:
1777	case PGMPOOLKIND_32BIT_PD:
1778	case PGMPOOLKIND_PAE_PDPT:
1779	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1780	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1781	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1782	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1783	break;
1784
1785	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1786	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1787	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1788	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1789	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1790	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1791	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1792	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1793	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1794	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1795	case PGMPOOLKIND_ROOT_NESTED:
1796	case PGMPOOLKIND_PAE_PD_PHYS:
1797	case PGMPOOLKIND_PAE_PDPT_PHYS:
1798	case PGMPOOLKIND_32BIT_PD_PHYS:
1799	/* Nothing to monitor here. */
1800	return VINF_SUCCESS;
1801
1802	#ifdef PGMPOOL_WITH_MIXED_PT_CR3
1803	break;
1804	#endif
1805	default:
1806	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
1807	}
1808
1809	/*
1810	* Remove the page from the monitored list or uninstall it if last.
1811	*/
1812	const PVM pVM = pPool->CTX_SUFF(pVM);
1813	int rc;
1814	if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
1815	\|\| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1816	{
1817	if (pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
1818	{
1819	PPGMPOOLPAGE pNewHead = &pPool->aPages[pPage->iMonitoredNext];
1820	pNewHead->iMonitoredPrev = NIL_PGMPOOL_IDX;
1821	rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
1822	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pNewHead),
1823	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pNewHead),
1824	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pNewHead),
1825	pPool->pszAccessHandler);
1826	AssertFatalRCSuccess(rc);
1827	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
1828	}
1829	else
1830	{
1831	pPool->aPages[pPage->iMonitoredPrev].iMonitoredNext = pPage->iMonitoredNext;
1832	if (pPage->iMonitoredNext != NIL_PGMPOOL_IDX)
1833	{
1834	pPool->aPages[pPage->iMonitoredNext].iMonitoredPrev = pPage->iMonitoredPrev;
1835	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
1836	}
1837	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
1838	rc = VINF_SUCCESS;
1839	}
1840	}
1841	else
1842	{
1843	rc = PGMHandlerPhysicalDeregister(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1));
1844	AssertFatalRC(rc);
1845	AssertMsg(!(pVM->pgm.s.fGlobalSyncFlags & PGM_GLOBAL_SYNC_CLEAR_PGM_POOL) \|\| VMCPU_FF_ISSET(VMMGetCpu(pVM), VMCPU_FF_PGM_SYNC_CR3),
1846	("%#x %#x\n", pVM->pgm.s.fGlobalSyncFlags, pVM->fGlobalForcedActions));
1847	}
1848	pPage->fMonitored = false;
1849
1850	/*
1851	* Remove it from the list of modified pages (if in it).
1852	*/
1853	pgmPoolMonitorModifiedRemove(pPool, pPage);
1854
1855	return rc;
1856	}
1857
1858
1859	/**
1860	* Inserts the page into the list of modified pages.
1861	*
1862	* @param pPool The pool.
1863	* @param pPage The page.
1864	*/
1865	void pgmPoolMonitorModifiedInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1866	{
1867	Log3(("pgmPoolMonitorModifiedInsert: idx=%d\n", pPage->idx));
1868	AssertMsg( pPage->iModifiedNext == NIL_PGMPOOL_IDX
1869	&& pPage->iModifiedPrev == NIL_PGMPOOL_IDX
1870	&& pPool->iModifiedHead != pPage->idx,
1871	("Next=%d Prev=%d idx=%d cModifications=%d Head=%d cModifiedPages=%d\n",
1872	pPage->iModifiedNext, pPage->iModifiedPrev, pPage->idx, pPage->cModifications,
1873	pPool->iModifiedHead, pPool->cModifiedPages));
1874
1875	pPage->iModifiedNext = pPool->iModifiedHead;
1876	if (pPool->iModifiedHead != NIL_PGMPOOL_IDX)
1877	pPool->aPages[pPool->iModifiedHead].iModifiedPrev = pPage->idx;
1878	pPool->iModifiedHead = pPage->idx;
1879	pPool->cModifiedPages++;
1880	#ifdef VBOX_WITH_STATISTICS
1881	if (pPool->cModifiedPages > pPool->cModifiedPagesHigh)
1882	pPool->cModifiedPagesHigh = pPool->cModifiedPages;
1883	#endif
1884	}
1885
1886
1887	/**
1888	* Removes the page from the list of modified pages and resets the
1889	* moficiation counter.
1890	*
1891	* @param pPool The pool.
1892	* @param pPage The page which is believed to be in the list of modified pages.
1893	*/
1894	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1895	{
1896	Log3(("pgmPoolMonitorModifiedRemove: idx=%d cModifications=%d\n", pPage->idx, pPage->cModifications));
1897	if (pPool->iModifiedHead == pPage->idx)
1898	{
1899	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1900	pPool->iModifiedHead = pPage->iModifiedNext;
1901	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1902	{
1903	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = NIL_PGMPOOL_IDX;
1904	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1905	}
1906	pPool->cModifiedPages--;
1907	}
1908	else if (pPage->iModifiedPrev != NIL_PGMPOOL_IDX)
1909	{
1910	pPool->aPages[pPage->iModifiedPrev].iModifiedNext = pPage->iModifiedNext;
1911	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1912	{
1913	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = pPage->iModifiedPrev;
1914	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1915	}
1916	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1917	pPool->cModifiedPages--;
1918	}
1919	else
1920	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1921	pPage->cModifications = 0;
1922	}
1923
1924
1925	/**
1926	* Zaps the list of modified pages, resetting their modification counters in the process.
1927	*
1928	* @param pVM The VM handle.
1929	*/
1930	void pgmPoolMonitorModifiedClearAll(PVM pVM)
1931	{
1932	pgmLock(pVM);
1933	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1934	LogFlow(("pgmPoolMonitorModifiedClearAll: cModifiedPages=%d\n", pPool->cModifiedPages));
1935
1936	unsigned cPages = 0; NOREF(cPages);
1937	uint16_t idx = pPool->iModifiedHead;
1938	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
1939	while (idx != NIL_PGMPOOL_IDX)
1940	{
1941	PPGMPOOLPAGE pPage = &pPool->aPages[idx];
1942	idx = pPage->iModifiedNext;
1943	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1944	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1945	pPage->cModifications = 0;
1946	Assert(++cPages);
1947	}
1948	AssertMsg(cPages == pPool->cModifiedPages, ("%d != %d\n", cPages, pPool->cModifiedPages));
1949	pPool->cModifiedPages = 0;
1950	pgmUnlock(pVM);
1951	}
1952
1953
1954	#ifdef IN_RING3
1955	/**
1956	* Callback to clear all shadow pages and clear all modification counters.
1957	*
1958	* @returns VBox status code.
1959	* @param pVM The VM handle.
1960	* @param pvUser Unused parameter
1961	* @remark Should only be used when monitoring is available, thus placed in
1962	* the PGMPOOL_WITH_MONITORING #ifdef.
1963	*/
1964	DECLCALLBACK(int) pgmPoolClearAll(PVM pVM, void *pvUser)
1965	{
1966	NOREF(pvUser);
1967	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1968	STAM_PROFILE_START(&pPool->StatClearAll, c);
1969	LogFlow(("pgmPoolClearAll: cUsedPages=%d\n", pPool->cUsedPages));
1970
1971	pgmLock(pVM);
1972
1973	/*
1974	* Iterate all the pages until we've encountered all that in use.
1975	* This is simple but not quite optimal solution.
1976	*/
1977	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
1978	unsigned cLeft = pPool->cUsedPages;
1979	unsigned iPage = pPool->cCurPages;
1980	while (--iPage >= PGMPOOL_IDX_FIRST)
1981	{
1982	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
1983	if (pPage->GCPhys != NIL_RTGCPHYS)
1984	{
1985	switch (pPage->enmKind)
1986	{
1987	/*
1988	* We only care about shadow page tables.
1989	*/
1990	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1991	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1992	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1993	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1994	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1995	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1996	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1997	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1998	{
1999	#ifdef PGMPOOL_WITH_USER_TRACKING
2000	if (pPage->cPresent)
2001	#endif
2002	{
2003	void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
2004	STAM_PROFILE_START(&pPool->StatZeroPage, z);
2005	ASMMemZeroPage(pvShw);
2006	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
2007	#ifdef PGMPOOL_WITH_USER_TRACKING
2008	pPage->cPresent = 0;
2009	pPage->iFirstPresent = ~0;
2010	#endif
2011	}
2012	}
2013	/* fall thru */
2014
2015	default:
2016	Assert(!pPage->cModifications \|\| ++cModifiedPages);
2017	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2018	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2019	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2020	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2021	pPage->cModifications = 0;
2022	break;
2023
2024	}
2025	if (!--cLeft)
2026	break;
2027	}
2028	}
2029
2030	/* swipe the special pages too. */
2031	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
2032	{
2033	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2034	if (pPage->GCPhys != NIL_RTGCPHYS)
2035	{
2036	Assert(!pPage->cModifications \|\| ++cModifiedPages);
2037	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2038	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2039	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2040	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2041	pPage->cModifications = 0;
2042	}
2043	}
2044
2045	#ifndef DEBUG_michael
2046	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
2047	#endif
2048	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
2049	pPool->cModifiedPages = 0;
2050
2051	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2052	/*
2053	* Clear all the GCPhys links and rebuild the phys ext free list.
2054	*/
2055	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
2056	pRam;
2057	pRam = pRam->CTX_SUFF(pNext))
2058	{
2059	unsigned iPage = pRam->cb >> PAGE_SHIFT;
2060	while (iPage-- > 0)
2061	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
2062	}
2063
2064	pPool->iPhysExtFreeHead = 0;
2065	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
2066	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
2067	for (unsigned i = 0; i < cMaxPhysExts; i++)
2068	{
2069	paPhysExts[i].iNext = i + 1;
2070	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
2071	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
2072	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
2073	}
2074	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
2075	#endif
2076
2077	pPool->cPresent = 0;
2078	pgmUnlock(pVM);
2079	PGM_INVL_ALL_VCPU_TLBS(pVM);
2080	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
2081	return VINF_SUCCESS;
2082	}
2083	#endif /* IN_RING3 */
2084
2085
2086	/**
2087	* Handle SyncCR3 pool tasks
2088	*
2089	* @returns VBox status code.
2090	* @retval VINF_SUCCESS if successfully added.
2091	* @retval VINF_PGM_SYNC_CR3 is it needs to be deferred to ring 3 (GC only)
2092	* @param pVM The VM handle.
2093	* @remark Should only be used when monitoring is available, thus placed in
2094	* the PGMPOOL_WITH_MONITORING #ifdef.
2095	*/
2096	int pgmPoolSyncCR3(PVM pVM)
2097	{
2098	LogFlow(("pgmPoolSyncCR3\n"));
2099	/*
2100	* When monitoring shadowed pages, we reset the modification counters on CR3 sync.
2101	* Occasionally we will have to clear all the shadow page tables because we wanted
2102	* to monitor a page which was mapped by too many shadowed page tables. This operation
2103	* sometimes refered to as a 'lightweight flush'.
2104	*/
2105	# ifdef IN_RING3 /* Don't flush in ring-0 or raw mode, it's taking too long. */
2106	if (ASMBitTestAndClear(&pVM->pgm.s.fGlobalSyncFlags, PGM_GLOBAL_SYNC_CLEAR_PGM_POOL_BIT))
2107	{
2108	VMMR3AtomicExecuteHandler(pVM, pgmPoolClearAll, NULL);
2109	# else /* !IN_RING3 */
2110	if (pVM->pgm.s.fGlobalSyncFlags & PGM_GLOBAL_SYNC_CLEAR_PGM_POOL)
2111	{
2112	LogFlow(("SyncCR3: PGM_GLOBAL_SYNC_CLEAR_PGM_POOL is set -> VINF_PGM_SYNC_CR3\n"));
2113	VMCPU_FF_SET(VMMGetCpu(pVM), VMCPU_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
2114	return VINF_PGM_SYNC_CR3;
2115	# endif /* !IN_RING3 */
2116	}
2117	else
2118	pgmPoolMonitorModifiedClearAll(pVM);
2119
2120	return VINF_SUCCESS;
2121	}
2122
2123	#endif /* PGMPOOL_WITH_MONITORING */
2124	#ifdef PGMPOOL_WITH_USER_TRACKING
2125
2126	/**
2127	* Frees up at least one user entry.
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 iUser The user index.
2134	*/
2135	static int pgmPoolTrackFreeOneUser(PPGMPOOL pPool, uint16_t iUser)
2136	{
2137	STAM_COUNTER_INC(&pPool->StatTrackFreeUpOneUser);
2138	#ifdef PGMPOOL_WITH_CACHE
2139	/*
2140	* Just free cached pages in a braindead fashion.
2141	*/
2142	/** @todo walk the age list backwards and free the first with usage. */
2143	int rc = VINF_SUCCESS;
2144	do
2145	{
2146	int rc2 = pgmPoolCacheFreeOne(pPool, iUser);
2147	if (RT_FAILURE(rc2) && rc == VINF_SUCCESS)
2148	rc = rc2;
2149	} while (pPool->iUserFreeHead == NIL_PGMPOOL_USER_INDEX);
2150	return rc;
2151	#else
2152	/*
2153	* Lazy approach.
2154	*/
2155	/* @todo This path no longer works (CR3 root pages will be flushed)!! */
2156	AssertCompileFailed();
2157	Assert(!CPUMIsGuestInLongMode(pVM));
2158	pgmPoolFlushAllInt(pPool);
2159	return VERR_PGM_POOL_FLUSHED;
2160	#endif
2161	}
2162
2163
2164	/**
2165	* Inserts a page into the cache.
2166	*
2167	* This will create user node for the page, insert it into the GCPhys
2168	* hash, and insert it into the age list.
2169	*
2170	* @returns VBox status code.
2171	* @retval VINF_SUCCESS if successfully added.
2172	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2173	* @param pPool The pool.
2174	* @param pPage The cached page.
2175	* @param GCPhys The GC physical address of the page we're gonna shadow.
2176	* @param iUser The user index.
2177	* @param iUserTable The user table index.
2178	*/
2179	DECLINLINE(int) pgmPoolTrackInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhys, uint16_t iUser, uint32_t iUserTable)
2180	{
2181	int rc = VINF_SUCCESS;
2182	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2183
2184	LogFlow(("pgmPoolTrackInsert GCPhys=%RGp iUser %x iUserTable %x\n", GCPhys, iUser, iUserTable));
2185
2186	#ifdef VBOX_STRICT
2187	/*
2188	* Check that the entry doesn't already exists.
2189	*/
2190	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2191	{
2192	uint16_t i = pPage->iUserHead;
2193	do
2194	{
2195	Assert(i < pPool->cMaxUsers);
2196	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2197	i = paUsers[i].iNext;
2198	} while (i != NIL_PGMPOOL_USER_INDEX);
2199	}
2200	#endif
2201
2202	/*
2203	* Find free a user node.
2204	*/
2205	uint16_t i = pPool->iUserFreeHead;
2206	if (i == NIL_PGMPOOL_USER_INDEX)
2207	{
2208	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2209	if (RT_FAILURE(rc))
2210	return rc;
2211	i = pPool->iUserFreeHead;
2212	}
2213
2214	/*
2215	* Unlink the user node from the free list,
2216	* initialize and insert it into the user list.
2217	*/
2218	pPool->iUserFreeHead = paUsers[i].iNext;
2219	paUsers[i].iNext = NIL_PGMPOOL_USER_INDEX;
2220	paUsers[i].iUser = iUser;
2221	paUsers[i].iUserTable = iUserTable;
2222	pPage->iUserHead = i;
2223
2224	/*
2225	* Insert into cache and enable monitoring of the guest page if enabled.
2226	*
2227	* Until we implement caching of all levels, including the CR3 one, we'll
2228	* have to make sure we don't try monitor & cache any recursive reuse of
2229	* a monitored CR3 page. Because all windows versions are doing this we'll
2230	* have to be able to do combined access monitoring, CR3 + PT and
2231	* PD + PT (guest PAE).
2232	*
2233	* Update:
2234	* We're now cooperating with the CR3 monitor if an uncachable page is found.
2235	*/
2236	#if defined(PGMPOOL_WITH_MONITORING) \|\| defined(PGMPOOL_WITH_CACHE)
2237	# ifdef PGMPOOL_WITH_MIXED_PT_CR3
2238	const bool fCanBeMonitored = true;
2239	# else
2240	bool fCanBeMonitored = pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored == NIL_RTGCPHYS
2241	\|\| (GCPhys & X86_PTE_PAE_PG_MASK) != (pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored & X86_PTE_PAE_PG_MASK)
2242	\|\| pgmPoolIsBigPage((PGMPOOLKIND)pPage->enmKind);
2243	# endif
2244	# ifdef PGMPOOL_WITH_CACHE
2245	pgmPoolCacheInsert(pPool, pPage, fCanBeMonitored); /* This can be expanded. */
2246	# endif
2247	if (fCanBeMonitored)
2248	{
2249	# ifdef PGMPOOL_WITH_MONITORING
2250	rc = pgmPoolMonitorInsert(pPool, pPage);
2251	AssertRC(rc);
2252	}
2253	# endif
2254	#endif /* PGMPOOL_WITH_MONITORING */
2255	return rc;
2256	}
2257
2258
2259	# ifdef PGMPOOL_WITH_CACHE /* (only used when the cache is enabled.) */
2260	/**
2261	* Adds a user reference to a page.
2262	*
2263	* This will move the page to the head of the
2264	*
2265	* @returns VBox status code.
2266	* @retval VINF_SUCCESS if successfully added.
2267	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2268	* @param pPool The pool.
2269	* @param pPage The cached page.
2270	* @param iUser The user index.
2271	* @param iUserTable The user table.
2272	*/
2273	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2274	{
2275	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2276
2277	Log3(("pgmPoolTrackAddUser GCPhys = %RGp iUser %x iUserTable %x\n", pPage->GCPhys, iUser, iUserTable));
2278
2279	# ifdef VBOX_STRICT
2280	/*
2281	* Check that the entry doesn't already exists. We only allow multiple users of top-level paging structures (SHW_POOL_ROOT_IDX).
2282	*/
2283	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2284	{
2285	uint16_t i = pPage->iUserHead;
2286	do
2287	{
2288	Assert(i < pPool->cMaxUsers);
2289	AssertMsg(iUser != PGMPOOL_IDX_PD \|\| iUser != PGMPOOL_IDX_PDPT \|\| iUser != PGMPOOL_IDX_NESTED_ROOT \|\| iUser != PGMPOOL_IDX_AMD64_CR3 \|\|
2290	paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2291	i = paUsers[i].iNext;
2292	} while (i != NIL_PGMPOOL_USER_INDEX);
2293	}
2294	# endif
2295
2296	/*
2297	* Allocate a user node.
2298	*/
2299	uint16_t i = pPool->iUserFreeHead;
2300	if (i == NIL_PGMPOOL_USER_INDEX)
2301	{
2302	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2303	if (RT_FAILURE(rc))
2304	return rc;
2305	i = pPool->iUserFreeHead;
2306	}
2307	pPool->iUserFreeHead = paUsers[i].iNext;
2308
2309	/*
2310	* Initialize the user node and insert it.
2311	*/
2312	paUsers[i].iNext = pPage->iUserHead;
2313	paUsers[i].iUser = iUser;
2314	paUsers[i].iUserTable = iUserTable;
2315	pPage->iUserHead = i;
2316
2317	# ifdef PGMPOOL_WITH_CACHE
2318	/*
2319	* Tell the cache to update its replacement stats for this page.
2320	*/
2321	pgmPoolCacheUsed(pPool, pPage);
2322	# endif
2323	return VINF_SUCCESS;
2324	}
2325	# endif /* PGMPOOL_WITH_CACHE */
2326
2327
2328	/**
2329	* Frees a user record associated with a page.
2330	*
2331	* This does not clear the entry in the user table, it simply replaces the
2332	* user record to the chain of free records.
2333	*
2334	* @param pPool The pool.
2335	* @param HCPhys The HC physical address of the shadow page.
2336	* @param iUser The shadow page pool index of the user table.
2337	* @param iUserTable The index into the user table (shadowed).
2338	*/
2339	static void pgmPoolTrackFreeUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2340	{
2341	/*
2342	* Unlink and free the specified user entry.
2343	*/
2344	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2345
2346	Log3(("pgmPoolTrackFreeUser %RGp %x %x\n", pPage->GCPhys, iUser, iUserTable));
2347	/* Special: For PAE and 32-bit paging, there is usually no more than one user. */
2348	uint16_t i = pPage->iUserHead;
2349	if ( i != NIL_PGMPOOL_USER_INDEX
2350	&& paUsers[i].iUser == iUser
2351	&& paUsers[i].iUserTable == iUserTable)
2352	{
2353	pPage->iUserHead = paUsers[i].iNext;
2354
2355	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2356	paUsers[i].iNext = pPool->iUserFreeHead;
2357	pPool->iUserFreeHead = i;
2358	return;
2359	}
2360
2361	/* General: Linear search. */
2362	uint16_t iPrev = NIL_PGMPOOL_USER_INDEX;
2363	while (i != NIL_PGMPOOL_USER_INDEX)
2364	{
2365	if ( paUsers[i].iUser == iUser
2366	&& paUsers[i].iUserTable == iUserTable)
2367	{
2368	if (iPrev != NIL_PGMPOOL_USER_INDEX)
2369	paUsers[iPrev].iNext = paUsers[i].iNext;
2370	else
2371	pPage->iUserHead = paUsers[i].iNext;
2372
2373	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2374	paUsers[i].iNext = pPool->iUserFreeHead;
2375	pPool->iUserFreeHead = i;
2376	return;
2377	}
2378	iPrev = i;
2379	i = paUsers[i].iNext;
2380	}
2381
2382	/* Fatal: didn't find it */
2383	AssertFatalMsgFailed(("Didn't find the user entry! iUser=%#x iUserTable=%#x GCPhys=%RGp\n",
2384	iUser, iUserTable, pPage->GCPhys));
2385	}
2386
2387
2388	/**
2389	* Gets the entry size of a shadow table.
2390	*
2391	* @param enmKind The kind of page.
2392	*
2393	* @returns The size of the entry in bytes. That is, 4 or 8.
2394	* @returns If the kind is not for a table, an assertion is raised and 0 is
2395	* returned.
2396	*/
2397	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind)
2398	{
2399	switch (enmKind)
2400	{
2401	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2402	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2403	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2404	case PGMPOOLKIND_32BIT_PD:
2405	case PGMPOOLKIND_32BIT_PD_PHYS:
2406	return 4;
2407
2408	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2409	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2410	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2411	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2412	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
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	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2418	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2419	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2420	case PGMPOOLKIND_64BIT_PML4:
2421	case PGMPOOLKIND_PAE_PDPT:
2422	case PGMPOOLKIND_ROOT_NESTED:
2423	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2424	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2425	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2426	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2427	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2428	case PGMPOOLKIND_PAE_PD_PHYS:
2429	case PGMPOOLKIND_PAE_PDPT_PHYS:
2430	return 8;
2431
2432	default:
2433	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2434	}
2435	}
2436
2437
2438	/**
2439	* Gets the entry size of a guest table.
2440	*
2441	* @param enmKind The kind of page.
2442	*
2443	* @returns The size of the entry in bytes. That is, 0, 4 or 8.
2444	* @returns If the kind is not for a table, an assertion is raised and 0 is
2445	* returned.
2446	*/
2447	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind)
2448	{
2449	switch (enmKind)
2450	{
2451	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2452	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2453	case PGMPOOLKIND_32BIT_PD:
2454	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2455	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2456	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2457	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2458	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2459	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2460	return 4;
2461
2462	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2463	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2464	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2465	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2466	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2467	case PGMPOOLKIND_64BIT_PML4:
2468	case PGMPOOLKIND_PAE_PDPT:
2469	return 8;
2470
2471	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2472	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2473	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2474	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2475	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2476	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2477	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2478	case PGMPOOLKIND_ROOT_NESTED:
2479	case PGMPOOLKIND_PAE_PD_PHYS:
2480	case PGMPOOLKIND_PAE_PDPT_PHYS:
2481	case PGMPOOLKIND_32BIT_PD_PHYS:
2482	/** @todo can we return 0? (nobody is calling this...) */
2483	AssertFailed();
2484	return 0;
2485
2486	default:
2487	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2488	}
2489	}
2490
2491	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2492
2493	/**
2494	* Scans one shadow page table for mappings of a physical page.
2495	*
2496	* @param pVM The VM handle.
2497	* @param pPhysPage The guest page in question.
2498	* @param iShw The shadow page table.
2499	* @param cRefs The number of references made in that PT.
2500	*/
2501	static void pgmPoolTrackFlushGCPhysPTInt(PVM pVM, PCPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2502	{
2503	LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%R[pgmpage] iShw=%d cRefs=%d\n", pPhysPage, iShw, cRefs));
2504	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2505
2506	/*
2507	* Assert sanity.
2508	*/
2509	Assert(cRefs == 1);
2510	AssertFatalMsg(iShw < pPool->cCurPages && iShw != NIL_PGMPOOL_IDX, ("iShw=%d\n", iShw));
2511	PPGMPOOLPAGE pPage = &pPool->aPages[iShw];
2512
2513	/*
2514	* Then, clear the actual mappings to the page in the shadow PT.
2515	*/
2516	switch (pPage->enmKind)
2517	{
2518	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2519	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2520	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2521	{
2522	const uint32_t u32 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2523	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2524	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2525	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2526	{
2527	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX32 cRefs=%#x\n", i, pPT->a[i], cRefs));
2528	pPT->a[i].u = 0;
2529	cRefs--;
2530	if (!cRefs)
2531	return;
2532	}
2533	#ifdef LOG_ENABLED
2534	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2535	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2536	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2537	{
2538	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2539	pPT->a[i].u = 0;
2540	}
2541	#endif
2542	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2543	break;
2544	}
2545
2546	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2547	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2548	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2549	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2550	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2551	{
2552	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2553	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2554	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2555	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2556	{
2557	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2558	pPT->a[i].u = 0;
2559	cRefs--;
2560	if (!cRefs)
2561	return;
2562	}
2563	#ifdef LOG_ENABLED
2564	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2565	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2566	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2567	{
2568	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2569	pPT->a[i].u = 0;
2570	}
2571	#endif
2572	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d u64=%RX64\n", cRefs, pPage->iFirstPresent, pPage->cPresent, u64));
2573	break;
2574	}
2575
2576	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2577	{
2578	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2579	PEPTPT pPT = (PEPTPT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2580	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2581	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2582	{
2583	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2584	pPT->a[i].u = 0;
2585	cRefs--;
2586	if (!cRefs)
2587	return;
2588	}
2589	#ifdef LOG_ENABLED
2590	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2591	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2592	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2593	{
2594	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2595	pPT->a[i].u = 0;
2596	}
2597	#endif
2598	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2599	break;
2600	}
2601
2602	default:
2603	AssertFatalMsgFailed(("enmKind=%d iShw=%d\n", pPage->enmKind, iShw));
2604	}
2605	}
2606
2607
2608	/**
2609	* Scans one shadow page table for mappings of a physical page.
2610	*
2611	* @param pVM The VM handle.
2612	* @param pPhysPage The guest page in question.
2613	* @param iShw The shadow page table.
2614	* @param cRefs The number of references made in that PT.
2615	*/
2616	void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2617	{
2618	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
2619	LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%R[pgmpage] iShw=%d cRefs=%d\n", pPhysPage, iShw, cRefs));
2620	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPT, f);
2621	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, iShw, cRefs);
2622	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2623	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPT, f);
2624	}
2625
2626
2627	/**
2628	* Flushes a list of shadow page tables mapping the same physical page.
2629	*
2630	* @param pVM The VM handle.
2631	* @param pPhysPage The guest page in question.
2632	* @param iPhysExt The physical cross reference extent list to flush.
2633	*/
2634	void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, uint16_t iPhysExt)
2635	{
2636	Assert(PGMIsLockOwner(pVM));
2637	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2638	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTs, f);
2639	LogFlow(("pgmPoolTrackFlushGCPhysPTs: pPhysPage=%R[pgmpage] iPhysExt\n", pPhysPage, iPhysExt));
2640
2641	const uint16_t iPhysExtStart = iPhysExt;
2642	PPGMPOOLPHYSEXT pPhysExt;
2643	do
2644	{
2645	Assert(iPhysExt < pPool->cMaxPhysExts);
2646	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2647	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
2648	if (pPhysExt->aidx[i] != NIL_PGMPOOL_IDX)
2649	{
2650	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, pPhysExt->aidx[i], 1);
2651	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
2652	}
2653
2654	/* next */
2655	iPhysExt = pPhysExt->iNext;
2656	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
2657
2658	/* insert the list into the free list and clear the ram range entry. */
2659	pPhysExt->iNext = pPool->iPhysExtFreeHead;
2660	pPool->iPhysExtFreeHead = iPhysExtStart;
2661	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2662
2663	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTs, f);
2664	}
2665
2666	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
2667
2668	/**
2669	* Flushes all shadow page table mappings of the given guest page.
2670	*
2671	* This is typically called when the host page backing the guest one has been
2672	* replaced or when the page protection was changed due to an access handler.
2673	*
2674	* @returns VBox status code.
2675	* @retval VINF_SUCCESS if all references has been successfully cleared.
2676	* @retval VINF_PGM_SYNC_CR3 if we're better off with a CR3 sync and a page
2677	* pool cleaning. FF and sync flags are set.
2678	*
2679	* @param pVM The VM handle.
2680	* @param pPhysPage The guest page in question.
2681	* @param pfFlushTLBs This is set to @a true if the shadow TLBs should be
2682	* flushed, it is NOT touched if this isn't necessary.
2683	* The caller MUST initialized this to @a false.
2684	*/
2685	int pgmPoolTrackFlushGCPhys(PVM pVM, PPGMPAGE pPhysPage, bool *pfFlushTLBs)
2686	{
2687	PVMCPU pVCpu = VMMGetCpu(pVM);
2688	pgmLock(pVM);
2689	int rc = VINF_SUCCESS;
2690	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2691	const uint16_t u16 = PGM_PAGE_GET_TRACKING(pPhysPage);
2692	if (u16)
2693	{
2694	/*
2695	* The zero page is currently screwing up the tracking and we'll
2696	* have to flush the whole shebang. Unless VBOX_WITH_NEW_LAZY_PAGE_ALLOC
2697	* is defined, zero pages won't normally be mapped. Some kind of solution
2698	* will be needed for this problem of course, but it will have to wait...
2699	*/
2700	if (PGM_PAGE_IS_ZERO(pPhysPage))
2701	rc = VINF_PGM_GCPHYS_ALIASED;
2702	else
2703	{
2704	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2705	/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow and
2706	pgmPoolTrackFlushGCPhysPTs will/may kill the pool otherwise. */
2707	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
2708	# endif
2709
2710	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
2711	pgmPoolTrackFlushGCPhysPT(pVM,
2712	pPhysPage,
2713	PGMPOOL_TD_GET_IDX(u16),
2714	PGMPOOL_TD_GET_CREFS(u16));
2715	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
2716	pgmPoolTrackFlushGCPhysPTs(pVM, pPhysPage, PGMPOOL_TD_GET_IDX(u16));
2717	else
2718	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
2719	*pfFlushTLBs = true;
2720
2721	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2722	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
2723	# endif
2724	}
2725	}
2726
2727	#elif defined(PGMPOOL_WITH_CACHE)
2728	if (PGM_PAGE_IS_ZERO(pPhysPage))
2729	rc = VINF_PGM_GCPHYS_ALIASED;
2730	else
2731	{
2732	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2733	/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow kill the pool otherwise. */
2734	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
2735	# endif
2736	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
2737	if (rc == VINF_SUCCESS)
2738	*pfFlushTLBs = true;
2739	}
2740
2741	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2742	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
2743	# endif
2744
2745	#else
2746	rc = VINF_PGM_GCPHYS_ALIASED;
2747	#endif
2748
2749	if (rc == VINF_PGM_GCPHYS_ALIASED)
2750	{
2751	pVM->pgm.s.fGlobalSyncFlags \|= PGM_GLOBAL_SYNC_CLEAR_PGM_POOL;
2752	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
2753	rc = VINF_PGM_SYNC_CR3;
2754	}
2755	pgmUnlock(pVM);
2756	return rc;
2757	}
2758
2759
2760	/**
2761	* Scans all shadow page tables for mappings of a physical page.
2762	*
2763	* This may be slow, but it's most likely more efficient than cleaning
2764	* out the entire page pool / cache.
2765	*
2766	* @returns VBox status code.
2767	* @retval VINF_SUCCESS if all references has been successfully cleared.
2768	* @retval VINF_PGM_GCPHYS_ALIASED if we're better off with a CR3 sync and
2769	* a page pool cleaning.
2770	*
2771	* @param pVM The VM handle.
2772	* @param pPhysPage The guest page in question.
2773	*/
2774	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage)
2775	{
2776	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2777	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2778	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: cUsedPages=%d cPresent=%d pPhysPage=%R[pgmpage]\n",
2779	pPool->cUsedPages, pPool->cPresent, pPhysPage));
2780
2781	#if 1
2782	/*
2783	* There is a limit to what makes sense.
2784	*/
2785	if (pPool->cPresent > 1024)
2786	{
2787	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
2788	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2789	return VINF_PGM_GCPHYS_ALIASED;
2790	}
2791	#endif
2792
2793	/*
2794	* Iterate all the pages until we've encountered all that in use.
2795	* This is simple but not quite optimal solution.
2796	*/
2797	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2798	const uint32_t u32 = u64;
2799	unsigned cLeft = pPool->cUsedPages;
2800	unsigned iPage = pPool->cCurPages;
2801	while (--iPage >= PGMPOOL_IDX_FIRST)
2802	{
2803	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2804	if (pPage->GCPhys != NIL_RTGCPHYS)
2805	{
2806	switch (pPage->enmKind)
2807	{
2808	/*
2809	* We only care about shadow page tables.
2810	*/
2811	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2812	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2813	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2814	{
2815	unsigned cPresent = pPage->cPresent;
2816	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2817	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2818	if (pPT->a[i].n.u1Present)
2819	{
2820	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2821	{
2822	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX32\n", iPage, i, pPT->a[i]));
2823	pPT->a[i].u = 0;
2824	}
2825	if (!--cPresent)
2826	break;
2827	}
2828	break;
2829	}
2830
2831	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2832	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2833	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2834	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2835	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2836	{
2837	unsigned cPresent = pPage->cPresent;
2838	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2839	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2840	if (pPT->a[i].n.u1Present)
2841	{
2842	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2843	{
2844	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
2845	pPT->a[i].u = 0;
2846	}
2847	if (!--cPresent)
2848	break;
2849	}
2850	break;
2851	}
2852	}
2853	if (!--cLeft)
2854	break;
2855	}
2856	}
2857
2858	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2859	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2860	return VINF_SUCCESS;
2861	}
2862
2863
2864	/**
2865	* Clears the user entry in a user table.
2866	*
2867	* This is used to remove all references to a page when flushing it.
2868	*/
2869	static void pgmPoolTrackClearPageUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PCPGMPOOLUSER pUser)
2870	{
2871	Assert(pUser->iUser != NIL_PGMPOOL_IDX);
2872	Assert(pUser->iUser < pPool->cCurPages);
2873	uint32_t iUserTable = pUser->iUserTable;
2874
2875	/*
2876	* Map the user page.
2877	*/
2878	PPGMPOOLPAGE pUserPage = &pPool->aPages[pUser->iUser];
2879	union
2880	{
2881	uint64_t *pau64;
2882	uint32_t *pau32;
2883	} u;
2884	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pUserPage);
2885
2886	LogFlow(("pgmPoolTrackClearPageUser: clear %x in %s (%RGp) (flushing %s)\n", iUserTable, pgmPoolPoolKindToStr(pUserPage->enmKind), pUserPage->Core.Key, pgmPoolPoolKindToStr(pPage->enmKind)));
2887
2888	/* Safety precaution in case we change the paging for other modes too in the future. */
2889	Assert(!pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage));
2890
2891	#ifdef VBOX_STRICT
2892	/*
2893	* Some sanity checks.
2894	*/
2895	switch (pUserPage->enmKind)
2896	{
2897	case PGMPOOLKIND_32BIT_PD:
2898	case PGMPOOLKIND_32BIT_PD_PHYS:
2899	Assert(iUserTable < X86_PG_ENTRIES);
2900	break;
2901	case PGMPOOLKIND_PAE_PDPT:
2902	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2903	case PGMPOOLKIND_PAE_PDPT_PHYS:
2904	Assert(iUserTable < 4);
2905	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2906	break;
2907	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2908	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2909	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2910	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2911	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2912	case PGMPOOLKIND_PAE_PD_PHYS:
2913	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2914	break;
2915	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2916	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2917	Assert(!(u.pau64[iUserTable] & PGM_PDFLAGS_MAPPING));
2918	break;
2919	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2920	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2921	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2922	break;
2923	case PGMPOOLKIND_64BIT_PML4:
2924	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2925	/* GCPhys >> PAGE_SHIFT is the index here */
2926	break;
2927	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2928	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2929	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2930	break;
2931
2932	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2933	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2934	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2935	break;
2936
2937	case PGMPOOLKIND_ROOT_NESTED:
2938	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2939	break;
2940
2941	default:
2942	AssertMsgFailed(("enmKind=%d\n", pUserPage->enmKind));
2943	break;
2944	}
2945	#endif /* VBOX_STRICT */
2946
2947	/*
2948	* Clear the entry in the user page.
2949	*/
2950	switch (pUserPage->enmKind)
2951	{
2952	/* 32-bit entries */
2953	case PGMPOOLKIND_32BIT_PD:
2954	case PGMPOOLKIND_32BIT_PD_PHYS:
2955	u.pau32[iUserTable] = 0;
2956	break;
2957
2958	/* 64-bit entries */
2959	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2960	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2961	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2962	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2963	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2964	#if defined(IN_RC)
2965	/* 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
2966	* non-present PDPT will continue to cause page faults.
2967	*/
2968	ASMReloadCR3();
2969	#endif
2970	/* no break */
2971	case PGMPOOLKIND_PAE_PD_PHYS:
2972	case PGMPOOLKIND_PAE_PDPT_PHYS:
2973	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2974	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2975	case PGMPOOLKIND_64BIT_PML4:
2976	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2977	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2978	case PGMPOOLKIND_PAE_PDPT:
2979	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2980	case PGMPOOLKIND_ROOT_NESTED:
2981	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2982	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2983	u.pau64[iUserTable] = 0;
2984	break;
2985
2986	default:
2987	AssertFatalMsgFailed(("enmKind=%d iUser=%#x iUserTable=%#x\n", pUserPage->enmKind, pUser->iUser, pUser->iUserTable));
2988	}
2989	}
2990
2991
2992	/**
2993	* Clears all users of a page.
2994	*/
2995	static void pgmPoolTrackClearPageUsers(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2996	{
2997	/*
2998	* Free all the user records.
2999	*/
3000	LogFlow(("pgmPoolTrackClearPageUsers %RGp\n", pPage->GCPhys));
3001
3002	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
3003	uint16_t i = pPage->iUserHead;
3004	while (i != NIL_PGMPOOL_USER_INDEX)
3005	{
3006	/* Clear enter in user table. */
3007	pgmPoolTrackClearPageUser(pPool, pPage, &paUsers[i]);
3008
3009	/* Free it. */
3010	const uint16_t iNext = paUsers[i].iNext;
3011	paUsers[i].iUser = NIL_PGMPOOL_IDX;
3012	paUsers[i].iNext = pPool->iUserFreeHead;
3013	pPool->iUserFreeHead = i;
3014
3015	/* Next. */
3016	i = iNext;
3017	}
3018	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
3019	}
3020
3021	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3022
3023	/**
3024	* Allocates a new physical cross reference extent.
3025	*
3026	* @returns Pointer to the allocated extent on success. NULL if we're out of them.
3027	* @param pVM The VM handle.
3028	* @param piPhysExt Where to store the phys ext index.
3029	*/
3030	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt)
3031	{
3032	Assert(PGMIsLockOwner(pVM));
3033	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3034	uint16_t iPhysExt = pPool->iPhysExtFreeHead;
3035	if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
3036	{
3037	STAM_COUNTER_INC(&pPool->StamTrackPhysExtAllocFailures);
3038	return NULL;
3039	}
3040	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3041	pPool->iPhysExtFreeHead = pPhysExt->iNext;
3042	pPhysExt->iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
3043	*piPhysExt = iPhysExt;
3044	return pPhysExt;
3045	}
3046
3047
3048	/**
3049	* Frees a physical cross reference extent.
3050	*
3051	* @param pVM The VM handle.
3052	* @param iPhysExt The extent to free.
3053	*/
3054	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt)
3055	{
3056	Assert(PGMIsLockOwner(pVM));
3057	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3058	Assert(iPhysExt < pPool->cMaxPhysExts);
3059	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3060	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3061	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3062	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3063	pPool->iPhysExtFreeHead = iPhysExt;
3064	}
3065
3066
3067	/**
3068	* Frees a physical cross reference extent.
3069	*
3070	* @param pVM The VM handle.
3071	* @param iPhysExt The extent to free.
3072	*/
3073	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt)
3074	{
3075	Assert(PGMIsLockOwner(pVM));
3076	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3077
3078	const uint16_t iPhysExtStart = iPhysExt;
3079	PPGMPOOLPHYSEXT pPhysExt;
3080	do
3081	{
3082	Assert(iPhysExt < pPool->cMaxPhysExts);
3083	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3084	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3085	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3086
3087	/* next */
3088	iPhysExt = pPhysExt->iNext;
3089	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3090
3091	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3092	pPool->iPhysExtFreeHead = iPhysExtStart;
3093	}
3094
3095
3096	/**
3097	* Insert a reference into a list of physical cross reference extents.
3098	*
3099	* @returns The new tracking data for PGMPAGE.
3100	*
3101	* @param pVM The VM handle.
3102	* @param iPhysExt The physical extent index of the list head.
3103	* @param iShwPT The shadow page table index.
3104	*
3105	*/
3106	static uint16_t pgmPoolTrackPhysExtInsert(PVM pVM, uint16_t iPhysExt, uint16_t iShwPT)
3107	{
3108	Assert(PGMIsLockOwner(pVM));
3109	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3110	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3111
3112	/* special common case. */
3113	if (paPhysExts[iPhysExt].aidx[2] == NIL_PGMPOOL_IDX)
3114	{
3115	paPhysExts[iPhysExt].aidx[2] = iShwPT;
3116	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
3117	LogFlow(("pgmPoolTrackPhysExtAddref: %d:{,,%d}\n", iPhysExt, iShwPT));
3118	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3119	}
3120
3121	/* general treatment. */
3122	const uint16_t iPhysExtStart = iPhysExt;
3123	unsigned cMax = 15;
3124	for (;;)
3125	{
3126	Assert(iPhysExt < pPool->cMaxPhysExts);
3127	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3128	if (paPhysExts[iPhysExt].aidx[i] == NIL_PGMPOOL_IDX)
3129	{
3130	paPhysExts[iPhysExt].aidx[i] = iShwPT;
3131	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
3132	LogFlow(("pgmPoolTrackPhysExtAddref: %d:{%d} i=%d cMax=%d\n", iPhysExt, iShwPT, i, cMax));
3133	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtStart);
3134	}
3135	if (!--cMax)
3136	{
3137	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
3138	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
3139	LogFlow(("pgmPoolTrackPhysExtAddref: overflow (1) iShwPT=%d\n", iShwPT));
3140	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3141	}
3142	}
3143
3144	/* add another extent to the list. */
3145	PPGMPOOLPHYSEXT pNew = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3146	if (!pNew)
3147	{
3148	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
3149	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
3150	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3151	}
3152	pNew->iNext = iPhysExtStart;
3153	pNew->aidx[0] = iShwPT;
3154	LogFlow(("pgmPoolTrackPhysExtAddref: added new extent %d:{%d}->%d\n", iPhysExt, iShwPT, iPhysExtStart));
3155	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3156	}
3157
3158
3159	/**
3160	* Add a reference to guest physical page where extents are in use.
3161	*
3162	* @returns The new tracking data for PGMPAGE.
3163	*
3164	* @param pVM The VM handle.
3165	* @param u16 The ram range flags (top 16-bits).
3166	* @param iShwPT The shadow page table index.
3167	*/
3168	uint16_t pgmPoolTrackPhysExtAddref(PVM pVM, uint16_t u16, uint16_t iShwPT)
3169	{
3170	pgmLock(pVM);
3171	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
3172	{
3173	/*
3174	* Convert to extent list.
3175	*/
3176	Assert(PGMPOOL_TD_GET_CREFS(u16) == 1);
3177	uint16_t iPhysExt;
3178	PPGMPOOLPHYSEXT pPhysExt = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3179	if (pPhysExt)
3180	{
3181	LogFlow(("pgmPoolTrackPhysExtAddref: new extent: %d:{%d, %d}\n", iPhysExt, PGMPOOL_TD_GET_IDX(u16), iShwPT));
3182	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliased);
3183	pPhysExt->aidx[0] = PGMPOOL_TD_GET_IDX(u16);
3184	pPhysExt->aidx[1] = iShwPT;
3185	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3186	}
3187	else
3188	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3189	}
3190	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
3191	{
3192	/*
3193	* Insert into the extent list.
3194	*/
3195	u16 = pgmPoolTrackPhysExtInsert(pVM, PGMPOOL_TD_GET_IDX(u16), iShwPT);
3196	}
3197	else
3198	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedLots);
3199	pgmUnlock(pVM);
3200	return u16;
3201	}
3202
3203
3204	/**
3205	* Clear references to guest physical memory.
3206	*
3207	* @param pPool The pool.
3208	* @param pPage The page.
3209	* @param pPhysPage Pointer to the aPages entry in the ram range.
3210	*/
3211	void pgmPoolTrackPhysExtDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMPAGE pPhysPage)
3212	{
3213	const unsigned cRefs = PGM_PAGE_GET_TD_CREFS(pPhysPage);
3214	AssertFatalMsg(cRefs == PGMPOOL_TD_CREFS_PHYSEXT, ("cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
3215
3216	uint16_t iPhysExt = PGM_PAGE_GET_TD_IDX(pPhysPage);
3217	if (iPhysExt != PGMPOOL_TD_IDX_OVERFLOWED)
3218	{
3219	PVM pVM = pPool->CTX_SUFF(pVM);
3220	pgmLock(pVM);
3221
3222	uint16_t iPhysExtPrev = NIL_PGMPOOL_PHYSEXT_INDEX;
3223	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3224	do
3225	{
3226	Assert(iPhysExt < pPool->cMaxPhysExts);
3227
3228	/*
3229	* Look for the shadow page and check if it's all freed.
3230	*/
3231	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3232	{
3233	if (paPhysExts[iPhysExt].aidx[i] == pPage->idx)
3234	{
3235	paPhysExts[iPhysExt].aidx[i] = NIL_PGMPOOL_IDX;
3236
3237	for (i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3238	if (paPhysExts[iPhysExt].aidx[i] != NIL_PGMPOOL_IDX)
3239	{
3240	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
3241	pgmUnlock(pVM);
3242	return;
3243	}
3244
3245	/* we can free the node. */
3246	const uint16_t iPhysExtNext = paPhysExts[iPhysExt].iNext;
3247	if ( iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX
3248	&& iPhysExtNext == NIL_PGMPOOL_PHYSEXT_INDEX)
3249	{
3250	/* lonely node */
3251	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3252	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d lonely\n", pPhysPage, pPage->idx));
3253	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
3254	}
3255	else if (iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX)
3256	{
3257	/* head */
3258	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d head\n", pPhysPage, pPage->idx));
3259	PGM_PAGE_SET_TRACKING(pPhysPage, PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtNext));
3260	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3261	}
3262	else
3263	{
3264	/* in list */
3265	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
3266	paPhysExts[iPhysExtPrev].iNext = iPhysExtNext;
3267	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3268	}
3269	iPhysExt = iPhysExtNext;
3270	pgmUnlock(pVM);
3271	return;
3272	}
3273	}
3274
3275	/* next */
3276	iPhysExtPrev = iPhysExt;
3277	iPhysExt = paPhysExts[iPhysExt].iNext;
3278	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3279
3280	pgmUnlock(pVM);
3281	AssertFatalMsgFailed(("not-found! cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
3282	}
3283	else /* nothing to do */
3284	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage]\n", pPhysPage));
3285	}
3286
3287
3288	/**
3289	* Clear references to guest physical memory.
3290	*
3291	* This is the same as pgmPoolTracDerefGCPhys except that the guest physical address
3292	* is assumed to be correct, so the linear search can be skipped and we can assert
3293	* at an earlier point.
3294	*
3295	* @param pPool The pool.
3296	* @param pPage The page.
3297	* @param HCPhys The host physical address corresponding to the guest page.
3298	* @param GCPhys The guest physical address corresponding to HCPhys.
3299	*/
3300	static void pgmPoolTracDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhys)
3301	{
3302	/*
3303	* Walk range list.
3304	*/
3305	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3306	while (pRam)
3307	{
3308	RTGCPHYS off = GCPhys - pRam->GCPhys;
3309	if (off < pRam->cb)
3310	{
3311	/* does it match? */
3312	const unsigned iPage = off >> PAGE_SHIFT;
3313	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3314	#ifdef LOG_ENABLED
3315	RTHCPHYS HCPhysPage = PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]);
3316	Log2(("pgmPoolTracDerefGCPhys %RHp vs %RHp\n", HCPhysPage, HCPhys));
3317	#endif
3318	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3319	{
3320	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3321	return;
3322	}
3323	break;
3324	}
3325	pRam = pRam->CTX_SUFF(pNext);
3326	}
3327	AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp\n", HCPhys, GCPhys));
3328	}
3329
3330
3331	/**
3332	* Clear references to guest physical memory.
3333	*
3334	* @param pPool The pool.
3335	* @param pPage The page.
3336	* @param HCPhys The host physical address corresponding to the guest page.
3337	* @param GCPhysHint The guest physical address which may corresponding to HCPhys.
3338	*/
3339	static void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint)
3340	{
3341	Log4(("pgmPoolTracDerefGCPhysHint %RHp %RGp\n", HCPhys, GCPhysHint));
3342
3343	/*
3344	* Walk range list.
3345	*/
3346	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3347	while (pRam)
3348	{
3349	RTGCPHYS off = GCPhysHint - pRam->GCPhys;
3350	if (off < pRam->cb)
3351	{
3352	/* does it match? */
3353	const unsigned iPage = off >> PAGE_SHIFT;
3354	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3355	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3356	{
3357	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3358	return;
3359	}
3360	break;
3361	}
3362	pRam = pRam->CTX_SUFF(pNext);
3363	}
3364
3365	/*
3366	* Damn, the hint didn't work. We'll have to do an expensive linear search.
3367	*/
3368	STAM_COUNTER_INC(&pPool->StatTrackLinearRamSearches);
3369	pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3370	while (pRam)
3371	{
3372	unsigned iPage = pRam->cb >> PAGE_SHIFT;
3373	while (iPage-- > 0)
3374	{
3375	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3376	{
3377	Log4(("pgmPoolTracDerefGCPhysHint: Linear HCPhys=%RHp GCPhysHint=%RGp GCPhysReal=%RGp\n",
3378	HCPhys, GCPhysHint, pRam->GCPhys + (iPage << PAGE_SHIFT)));
3379	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3380	return;
3381	}
3382	}
3383	pRam = pRam->CTX_SUFF(pNext);
3384	}
3385
3386	AssertFatalMsgFailed(("HCPhys=%RHp GCPhysHint=%RGp\n", HCPhys, GCPhysHint));
3387	}
3388
3389
3390	/**
3391	* Clear references to guest physical memory in a 32-bit / 32-bit page table.
3392	*
3393	* @param pPool The pool.
3394	* @param pPage The page.
3395	* @param pShwPT The shadow page table (mapping of the page).
3396	* @param pGstPT The guest page table.
3397	*/
3398	DECLINLINE(void) pgmPoolTrackDerefPT32Bit32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT, PCX86PT pGstPT)
3399	{
3400	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
3401	if (pShwPT->a[i].n.u1Present)
3402	{
3403	Log4(("pgmPoolTrackDerefPT32Bit32Bit: i=%d pte=%RX32 hint=%RX32\n",
3404	i, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3405	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3406	if (!--pPage->cPresent)
3407	break;
3408	}
3409	}
3410
3411
3412	/**
3413	* Clear references to guest physical memory in a PAE / 32-bit page table.
3414	*
3415	* @param pPool The pool.
3416	* @param pPage The page.
3417	* @param pShwPT The shadow page table (mapping of the page).
3418	* @param pGstPT The guest page table (just a half one).
3419	*/
3420	DECLINLINE(void) pgmPoolTrackDerefPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PT pGstPT)
3421	{
3422	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3423	if (pShwPT->a[i].n.u1Present)
3424	{
3425	Log4(("pgmPoolTrackDerefPTPae32Bit: i=%d pte=%RX64 hint=%RX32\n",
3426	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3427	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3428	}
3429	}
3430
3431
3432	/**
3433	* Clear references to guest physical memory in a PAE / PAE page table.
3434	*
3435	* @param pPool The pool.
3436	* @param pPage The page.
3437	* @param pShwPT The shadow page table (mapping of the page).
3438	* @param pGstPT The guest page table.
3439	*/
3440	DECLINLINE(void) pgmPoolTrackDerefPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PTPAE pGstPT)
3441	{
3442	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3443	if (pShwPT->a[i].n.u1Present)
3444	{
3445	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX32 hint=%RX32\n",
3446	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
3447	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK);
3448	}
3449	}
3450
3451
3452	/**
3453	* Clear references to guest physical memory in a 32-bit / 4MB page table.
3454	*
3455	* @param pPool The pool.
3456	* @param pPage The page.
3457	* @param pShwPT The shadow page table (mapping of the page).
3458	*/
3459	DECLINLINE(void) pgmPoolTrackDerefPT32Bit4MB(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT)
3460	{
3461	RTGCPHYS GCPhys = pPage->GCPhys;
3462	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3463	if (pShwPT->a[i].n.u1Present)
3464	{
3465	Log4(("pgmPoolTrackDerefPT32Bit4MB: i=%d pte=%RX32 GCPhys=%RGp\n",
3466	i, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys));
3467	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys);
3468	}
3469	}
3470
3471
3472	/**
3473	* Clear references to guest physical memory in a PAE / 2/4MB page table.
3474	*
3475	* @param pPool The pool.
3476	* @param pPage The page.
3477	* @param pShwPT The shadow page table (mapping of the page).
3478	*/
3479	DECLINLINE(void) pgmPoolTrackDerefPTPaeBig(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT)
3480	{
3481	RTGCPHYS GCPhys = pPage->GCPhys;
3482	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3483	if (pShwPT->a[i].n.u1Present)
3484	{
3485	Log4(("pgmPoolTrackDerefPTPaeBig: i=%d pte=%RX64 hint=%RGp\n",
3486	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys));
3487	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys);
3488	}
3489	}
3490
3491	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
3492
3493
3494	/**
3495	* Clear references to shadowed pages in a 32 bits page directory.
3496	*
3497	* @param pPool The pool.
3498	* @param pPage The page.
3499	* @param pShwPD The shadow page directory (mapping of the page).
3500	*/
3501	DECLINLINE(void) pgmPoolTrackDerefPD(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PD pShwPD)
3502	{
3503	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3504	{
3505	if ( pShwPD->a[i].n.u1Present
3506	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
3507	)
3508	{
3509	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PG_MASK);
3510	if (pSubPage)
3511	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3512	else
3513	AssertFatalMsgFailed(("%x\n", pShwPD->a[i].u & X86_PDE_PG_MASK));
3514	}
3515	}
3516	}
3517
3518	/**
3519	* Clear references to shadowed pages in a PAE (legacy or 64 bits) page directory.
3520	*
3521	* @param pPool The pool.
3522	* @param pPage The page.
3523	* @param pShwPD The shadow page directory (mapping of the page).
3524	*/
3525	DECLINLINE(void) pgmPoolTrackDerefPDPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPAE pShwPD)
3526	{
3527	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3528	{
3529	if ( pShwPD->a[i].n.u1Present
3530	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
3531	)
3532	{
3533	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PAE_PG_MASK);
3534	if (pSubPage)
3535	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3536	else
3537	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & X86_PDE_PAE_PG_MASK));
3538	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3539	}
3540	}
3541	}
3542
3543	/**
3544	* Clear references to shadowed pages in a PAE page directory pointer table.
3545	*
3546	* @param pPool The pool.
3547	* @param pPage The page.
3548	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3549	*/
3550	DECLINLINE(void) pgmPoolTrackDerefPDPTPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
3551	{
3552	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
3553	{
3554	if ( pShwPDPT->a[i].n.u1Present
3555	&& !(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING)
3556	)
3557	{
3558	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
3559	if (pSubPage)
3560	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3561	else
3562	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
3563	}
3564	}
3565	}
3566
3567
3568	/**
3569	* Clear references to shadowed pages in a 64-bit page directory pointer table.
3570	*
3571	* @param pPool The pool.
3572	* @param pPage The page.
3573	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3574	*/
3575	DECLINLINE(void) pgmPoolTrackDerefPDPT64Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
3576	{
3577	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3578	{
3579	Assert(!(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING));
3580	if (pShwPDPT->a[i].n.u1Present)
3581	{
3582	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
3583	if (pSubPage)
3584	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3585	else
3586	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
3587	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3588	}
3589	}
3590	}
3591
3592
3593	/**
3594	* Clear references to shadowed pages in a 64-bit level 4 page table.
3595	*
3596	* @param pPool The pool.
3597	* @param pPage The page.
3598	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3599	*/
3600	DECLINLINE(void) pgmPoolTrackDerefPML464Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PML4 pShwPML4)
3601	{
3602	for (unsigned i = 0; i < RT_ELEMENTS(pShwPML4->a); i++)
3603	{
3604	if (pShwPML4->a[i].n.u1Present)
3605	{
3606	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPML4->a[i].u & X86_PDPE_PG_MASK);
3607	if (pSubPage)
3608	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3609	else
3610	AssertFatalMsgFailed(("%RX64\n", pShwPML4->a[i].u & X86_PML4E_PG_MASK));
3611	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3612	}
3613	}
3614	}
3615
3616
3617	/**
3618	* Clear references to shadowed pages in an EPT page table.
3619	*
3620	* @param pPool The pool.
3621	* @param pPage The page.
3622	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3623	*/
3624	DECLINLINE(void) pgmPoolTrackDerefPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT)
3625	{
3626	RTGCPHYS GCPhys = pPage->GCPhys;
3627	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3628	if (pShwPT->a[i].n.u1Present)
3629	{
3630	Log4(("pgmPoolTrackDerefPTEPT: i=%d pte=%RX64 GCPhys=%RX64\n",
3631	i, pShwPT->a[i].u & EPT_PTE_PG_MASK, pPage->GCPhys));
3632	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & EPT_PTE_PG_MASK, GCPhys);
3633	}
3634	}
3635
3636
3637	/**
3638	* Clear references to shadowed pages in an EPT page directory.
3639	*
3640	* @param pPool The pool.
3641	* @param pPage The page.
3642	* @param pShwPD The shadow page directory (mapping of the page).
3643	*/
3644	DECLINLINE(void) pgmPoolTrackDerefPDEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPD pShwPD)
3645	{
3646	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3647	{
3648	if (pShwPD->a[i].n.u1Present)
3649	{
3650	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & EPT_PDE_PG_MASK);
3651	if (pSubPage)
3652	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3653	else
3654	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & EPT_PDE_PG_MASK));
3655	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3656	}
3657	}
3658	}
3659
3660
3661	/**
3662	* Clear references to shadowed pages in an EPT page directory pointer table.
3663	*
3664	* @param pPool The pool.
3665	* @param pPage The page.
3666	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3667	*/
3668	DECLINLINE(void) pgmPoolTrackDerefPDPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPDPT pShwPDPT)
3669	{
3670	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3671	{
3672	if (pShwPDPT->a[i].n.u1Present)
3673	{
3674	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK);
3675	if (pSubPage)
3676	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3677	else
3678	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK));
3679	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3680	}
3681	}
3682	}
3683
3684
3685	/**
3686	* Clears all references made by this page.
3687	*
3688	* This includes other shadow pages and GC physical addresses.
3689	*
3690	* @param pPool The pool.
3691	* @param pPage The page.
3692	*/
3693	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3694	{
3695	/*
3696	* Map the shadow page and take action according to the page kind.
3697	*/
3698	void *pvShw = PGMPOOL_PAGE_2_LOCKED_PTR(pPool->CTX_SUFF(pVM), pPage);
3699	switch (pPage->enmKind)
3700	{
3701	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3702	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3703	{
3704	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3705	void *pvGst;
3706	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3707	pgmPoolTrackDerefPT32Bit32Bit(pPool, pPage, (PX86PT)pvShw, (PCX86PT)pvGst);
3708	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3709	break;
3710	}
3711
3712	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3713	{
3714	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3715	void *pvGst;
3716	int rc = PGM_GCPHYS_2_PTR_EX(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3717	pgmPoolTrackDerefPTPae32Bit(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PT)pvGst);
3718	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3719	break;
3720	}
3721
3722	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3723	{
3724	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3725	void *pvGst;
3726	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3727	pgmPoolTrackDerefPTPaePae(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PTPAE)pvGst);
3728	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3729	break;
3730	}
3731
3732	case PGMPOOLKIND_32BIT_PT_FOR_PHYS: /* treat it like a 4 MB page */
3733	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3734	{
3735	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3736	pgmPoolTrackDerefPT32Bit4MB(pPool, pPage, (PX86PT)pvShw);
3737	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3738	break;
3739	}
3740
3741	case PGMPOOLKIND_PAE_PT_FOR_PHYS: /* treat it like a 2 MB page */
3742	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3743	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3744	{
3745	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3746	pgmPoolTrackDerefPTPaeBig(pPool, pPage, (PX86PTPAE)pvShw);
3747	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3748	break;
3749	}
3750
3751	#else /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3752	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3753	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3754	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3755	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3756	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3757	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3758	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3759	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3760	break;
3761	#endif /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3762
3763	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3764	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3765	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3766	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3767	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3768	case PGMPOOLKIND_PAE_PD_PHYS:
3769	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3770	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3771	pgmPoolTrackDerefPDPae(pPool, pPage, (PX86PDPAE)pvShw);
3772	break;
3773
3774	case PGMPOOLKIND_32BIT_PD_PHYS:
3775	case PGMPOOLKIND_32BIT_PD:
3776	pgmPoolTrackDerefPD(pPool, pPage, (PX86PD)pvShw);
3777	break;
3778
3779	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
3780	case PGMPOOLKIND_PAE_PDPT:
3781	case PGMPOOLKIND_PAE_PDPT_PHYS:
3782	pgmPoolTrackDerefPDPTPae(pPool, pPage, (PX86PDPT)pvShw);
3783	break;
3784
3785	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3786	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3787	pgmPoolTrackDerefPDPT64Bit(pPool, pPage, (PX86PDPT)pvShw);
3788	break;
3789
3790	case PGMPOOLKIND_64BIT_PML4:
3791	pgmPoolTrackDerefPML464Bit(pPool, pPage, (PX86PML4)pvShw);
3792	break;
3793
3794	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3795	pgmPoolTrackDerefPTEPT(pPool, pPage, (PEPTPT)pvShw);
3796	break;
3797
3798	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3799	pgmPoolTrackDerefPDEPT(pPool, pPage, (PEPTPD)pvShw);
3800	break;
3801
3802	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3803	pgmPoolTrackDerefPDPTEPT(pPool, pPage, (PEPTPDPT)pvShw);
3804	break;
3805
3806	default:
3807	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
3808	}
3809
3810	/* paranoia, clear the shadow page. Remove this laser (i.e. let Alloc and ClearAll do it). */
3811	STAM_PROFILE_START(&pPool->StatZeroPage, z);
3812	ASMMemZeroPage(pvShw);
3813	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
3814	pPage->fZeroed = true;
3815	PGMPOOL_UNLOCK_PTR(pPool->CTX_SUFF(pVM), pvShw);
3816	}
3817	#endif /* PGMPOOL_WITH_USER_TRACKING */
3818
3819	/**
3820	* Flushes a pool page.
3821	*
3822	* This moves the page to the free list after removing all user references to it.
3823	*
3824	* @returns VBox status code.
3825	* @retval VINF_SUCCESS on success.
3826	* @param pPool The pool.
3827	* @param HCPhys The HC physical address of the shadow page.
3828	*/
3829	int pgmPoolFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3830	{
3831	PVM pVM = pPool->CTX_SUFF(pVM);
3832
3833	int rc = VINF_SUCCESS;
3834	STAM_PROFILE_START(&pPool->StatFlushPage, f);
3835	LogFlow(("pgmPoolFlushPage: pPage=%p:{.Key=%RHp, .idx=%d, .enmKind=%s, .GCPhys=%RGp}\n",
3836	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
3837
3838	/*
3839	* Quietly reject any attempts at flushing any of the special root pages.
3840	*/
3841	if (pPage->idx < PGMPOOL_IDX_FIRST)
3842	{
3843	AssertFailed(); /* can no longer happen */
3844	Log(("pgmPoolFlushPage: special root page, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
3845	return VINF_SUCCESS;
3846	}
3847
3848	pgmLock(pVM);
3849
3850	/*
3851	* Quietly reject any attempts at flushing the currently active shadow CR3 mapping
3852	*/
3853	if (pgmPoolIsPageLocked(&pVM->pgm.s, pPage))
3854	{
3855	AssertMsg( pPage->enmKind == PGMPOOLKIND_64BIT_PML4
3856	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT
3857	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT_FOR_32BIT
3858	\|\| pPage->enmKind == PGMPOOLKIND_32BIT_PD
3859	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD
3860	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD
3861	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD
3862	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD
3863	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD,
3864	("Can't free the shadow CR3! (%RHp vs %RHp kind=%d\n", PGMGetHyperCR3(VMMGetCpu(pVM)), pPage->Core.Key, pPage->enmKind));
3865	Log(("pgmPoolFlushPage: current active shadow CR3, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
3866	pgmUnlock(pVM);
3867	return VINF_SUCCESS;
3868	}
3869
3870	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
3871	/* Start a subset so we won't run out of mapping space. */
3872	PVMCPU pVCpu = VMMGetCpu(pVM);
3873	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
3874	#endif
3875
3876	/*
3877	* Mark the page as being in need of a ASMMemZeroPage().
3878	*/
3879	pPage->fZeroed = false;
3880
3881	#ifdef PGMPOOL_WITH_USER_TRACKING
3882	/*
3883	* Clear the page.
3884	*/
3885	pgmPoolTrackClearPageUsers(pPool, pPage);
3886	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
3887	pgmPoolTrackDeref(pPool, pPage);
3888	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
3889	#endif
3890
3891	#ifdef PGMPOOL_WITH_CACHE
3892	/*
3893	* Flush it from the cache.
3894	*/
3895	pgmPoolCacheFlushPage(pPool, pPage);
3896	#endif /* PGMPOOL_WITH_CACHE */
3897
3898	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
3899	/* Heavy stuff done. */
3900	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
3901	#endif
3902
3903	#ifdef PGMPOOL_WITH_MONITORING
3904	/*
3905	* Deregistering the monitoring.
3906	*/
3907	if (pPage->fMonitored)
3908	rc = pgmPoolMonitorFlush(pPool, pPage);
3909	#endif
3910
3911	/*
3912	* Free the page.
3913	*/
3914	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
3915	pPage->iNext = pPool->iFreeHead;
3916	pPool->iFreeHead = pPage->idx;
3917	pPage->enmKind = PGMPOOLKIND_FREE;
3918	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
3919	pPage->GCPhys = NIL_RTGCPHYS;
3920	pPage->fReusedFlushPending = false;
3921
3922	pPool->cUsedPages--;
3923	pgmUnlock(pVM);
3924	STAM_PROFILE_STOP(&pPool->StatFlushPage, f);
3925	return rc;
3926	}
3927
3928
3929	/**
3930	* Frees a usage of a pool page.
3931	*
3932	* The caller is responsible to updating the user table so that it no longer
3933	* references the shadow page.
3934	*
3935	* @param pPool The pool.
3936	* @param HCPhys The HC physical address of the shadow page.
3937	* @param iUser The shadow page pool index of the user table.
3938	* @param iUserTable The index into the user table (shadowed).
3939	*/
3940	void pgmPoolFreeByPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
3941	{
3942	PVM pVM = pPool->CTX_SUFF(pVM);
3943
3944	STAM_PROFILE_START(&pPool->StatFree, a);
3945	LogFlow(("pgmPoolFreeByPage: pPage=%p:{.Key=%RHp, .idx=%d, enmKind=%s} iUser=%#x iUserTable=%#x\n",
3946	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), iUser, iUserTable));
3947	Assert(pPage->idx >= PGMPOOL_IDX_FIRST);
3948	pgmLock(pVM);
3949	#ifdef PGMPOOL_WITH_USER_TRACKING
3950	pgmPoolTrackFreeUser(pPool, pPage, iUser, iUserTable);
3951	#endif
3952	#ifdef PGMPOOL_WITH_CACHE
3953	if (!pPage->fCached)
3954	#endif
3955	pgmPoolFlushPage(pPool, pPage);
3956	pgmUnlock(pVM);
3957	STAM_PROFILE_STOP(&pPool->StatFree, a);
3958	}
3959
3960
3961	/**
3962	* Makes one or more free page free.
3963	*
3964	* @returns VBox status code.
3965	* @retval VINF_SUCCESS on success.
3966	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
3967	*
3968	* @param pPool The pool.
3969	* @param enmKind Page table kind
3970	* @param iUser The user of the page.
3971	*/
3972	static int pgmPoolMakeMoreFreePages(PPGMPOOL pPool, PGMPOOLKIND enmKind, uint16_t iUser)
3973	{
3974	PVM pVM = pPool->CTX_SUFF(pVM);
3975
3976	LogFlow(("pgmPoolMakeMoreFreePages: iUser=%#x\n", iUser));
3977
3978	/*
3979	* If the pool isn't full grown yet, expand it.
3980	*/
3981	if ( pPool->cCurPages < pPool->cMaxPages
3982	#if defined(IN_RC)
3983	/* Hack alert: we can't deal with jumps to ring 3 when called from MapCR3 and allocating pages for PAE PDs. */
3984	&& enmKind != PGMPOOLKIND_PAE_PD_FOR_PAE_PD
3985	&& (enmKind < PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD \|\| enmKind > PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD)
3986	#endif
3987	)
3988	{
3989	STAM_PROFILE_ADV_SUSPEND(&pPool->StatAlloc, a);
3990	#ifdef IN_RING3
3991	int rc = PGMR3PoolGrow(pVM);
3992	#else
3993	int rc = CTXALLMID(VMM, CallHost)(pVM, VMMCALLHOST_PGM_POOL_GROW, 0);
3994	#endif
3995	if (RT_FAILURE(rc))
3996	return rc;
3997	STAM_PROFILE_ADV_RESUME(&pPool->StatAlloc, a);
3998	if (pPool->iFreeHead != NIL_PGMPOOL_IDX)
3999	return VINF_SUCCESS;
4000	}
4001
4002	#ifdef PGMPOOL_WITH_CACHE
4003	/*
4004	* Free one cached page.
4005	*/
4006	return pgmPoolCacheFreeOne(pPool, iUser);
4007	#else
4008	/*
4009	* Flush the pool.
4010	*
4011	* If we have tracking enabled, it should be possible to come up with
4012	* a cheap replacement strategy...
4013	*/
4014	/* @todo This path no longer works (CR3 root pages will be flushed)!! */
4015	AssertCompileFailed();
4016	Assert(!CPUMIsGuestInLongMode(pVM));
4017	pgmPoolFlushAllInt(pPool);
4018	return VERR_PGM_POOL_FLUSHED;
4019	#endif
4020	}
4021
4022	/**
4023	* Allocates a page from the pool.
4024	*
4025	* This page may actually be a cached page and not in need of any processing
4026	* on the callers part.
4027	*
4028	* @returns VBox status code.
4029	* @retval VINF_SUCCESS if a NEW page was allocated.
4030	* @retval VINF_PGM_CACHED_PAGE if a CACHED page was returned.
4031	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
4032	* @param pVM The VM handle.
4033	* @param GCPhys The GC physical address of the page we're gonna shadow.
4034	* For 4MB and 2MB PD entries, it's the first address the
4035	* shadow PT is covering.
4036	* @param enmKind The kind of mapping.
4037	* @param enmAccess Access type for the mapping (only relevant for big pages)
4038	* @param iUser The shadow page pool index of the user table.
4039	* @param iUserTable The index into the user table (shadowed).
4040	* @param ppPage Where to store the pointer to the page. NULL is stored here on failure.
4041	*/
4042	int pgmPoolAllocEx(PVM pVM, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, PGMPOOLACCESS enmAccess, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
4043	{
4044	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4045	STAM_PROFILE_ADV_START(&pPool->StatAlloc, a);
4046	LogFlow(("pgmPoolAlloc: GCPhys=%RGp enmKind=%s iUser=%#x iUserTable=%#x\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable));
4047	*ppPage = NULL;
4048	/** @todo CSAM/PGMPrefetchPage messes up here during CSAMR3CheckGates
4049	* (TRPMR3SyncIDT) because of FF priority. Try fix that?
4050	* Assert(!(pVM->pgm.s.fGlobalSyncFlags & PGM_GLOBAL_SYNC_CLEAR_PGM_POOL)); */
4051
4052	pgmLock(pVM);
4053
4054	#ifdef PGMPOOL_WITH_CACHE
4055	if (pPool->fCacheEnabled)
4056	{
4057	int rc2 = pgmPoolCacheAlloc(pPool, GCPhys, enmKind, enmAccess, iUser, iUserTable, ppPage);
4058	if (RT_SUCCESS(rc2))
4059	{
4060	pgmUnlock(pVM);
4061	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4062	LogFlow(("pgmPoolAlloc: cached returns %Rrc ppPage=%p:{.Key=%RHp, .idx=%d}\n", rc2, ppPage, (ppPage)->Core.Key, (ppPage)->idx));
4063	return rc2;
4064	}
4065	}
4066	#endif
4067
4068	/*
4069	* Allocate a new one.
4070	*/
4071	int rc = VINF_SUCCESS;
4072	uint16_t iNew = pPool->iFreeHead;
4073	if (iNew == NIL_PGMPOOL_IDX)
4074	{
4075	rc = pgmPoolMakeMoreFreePages(pPool, enmKind, iUser);
4076	if (RT_FAILURE(rc))
4077	{
4078	pgmUnlock(pVM);
4079	Log(("pgmPoolAlloc: returns %Rrc (Free)\n", rc));
4080	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4081	return rc;
4082	}
4083	iNew = pPool->iFreeHead;
4084	AssertReleaseReturn(iNew != NIL_PGMPOOL_IDX, VERR_INTERNAL_ERROR);
4085	}
4086
4087	/* unlink the free head */
4088	PPGMPOOLPAGE pPage = &pPool->aPages[iNew];
4089	pPool->iFreeHead = pPage->iNext;
4090	pPage->iNext = NIL_PGMPOOL_IDX;
4091
4092	/*
4093	* Initialize it.
4094	*/
4095	pPool->cUsedPages++; /* physical handler registration / pgmPoolTrackFlushGCPhysPTsSlow requirement. */
4096	pPage->enmKind = enmKind;
4097	pPage->enmAccess = enmAccess;
4098	pPage->GCPhys = GCPhys;
4099	pPage->fSeenNonGlobal = false; /* Set this to 'true' to disable this feature. */
4100	pPage->fMonitored = false;
4101	pPage->fCached = false;
4102	pPage->fReusedFlushPending = false;
4103	#ifdef PGMPOOL_WITH_MONITORING
4104	pPage->cModifications = 0;
4105	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4106	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4107	#else
4108	pPage->fCR3Mix = false;
4109	#endif
4110	#ifdef PGMPOOL_WITH_USER_TRACKING
4111	pPage->cPresent = 0;
4112	pPage->iFirstPresent = ~0;
4113
4114	/*
4115	* Insert into the tracking and cache. If this fails, free the page.
4116	*/
4117	int rc3 = pgmPoolTrackInsert(pPool, pPage, GCPhys, iUser, iUserTable);
4118	if (RT_FAILURE(rc3))
4119	{
4120	pPool->cUsedPages--;
4121	pPage->enmKind = PGMPOOLKIND_FREE;
4122	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
4123	pPage->GCPhys = NIL_RTGCPHYS;
4124	pPage->iNext = pPool->iFreeHead;
4125	pPool->iFreeHead = pPage->idx;
4126	pgmUnlock(pVM);
4127	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4128	Log(("pgmPoolAlloc: returns %Rrc (Insert)\n", rc3));
4129	return rc3;
4130	}
4131	#endif /* PGMPOOL_WITH_USER_TRACKING */
4132
4133	/*
4134	* Commit the allocation, clear the page and return.
4135	*/
4136	#ifdef VBOX_WITH_STATISTICS
4137	if (pPool->cUsedPages > pPool->cUsedPagesHigh)
4138	pPool->cUsedPagesHigh = pPool->cUsedPages;
4139	#endif
4140
4141	if (!pPage->fZeroed)
4142	{
4143	STAM_PROFILE_START(&pPool->StatZeroPage, z);
4144	void *pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
4145	ASMMemZeroPage(pv);
4146	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
4147	}
4148
4149	*ppPage = pPage;
4150	pgmUnlock(pVM);
4151	LogFlow(("pgmPoolAlloc: returns %Rrc *ppPage=%p:{.Key=%RHp, .idx=%d, .fCached=%RTbool, .fMonitored=%RTbool}\n",
4152	rc, pPage, pPage->Core.Key, pPage->idx, pPage->fCached, pPage->fMonitored));
4153	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4154	return rc;
4155	}
4156
4157
4158	/**
4159	* Frees a usage of a pool page.
4160	*
4161	* @param pVM The VM handle.
4162	* @param HCPhys The HC physical address of the shadow page.
4163	* @param iUser The shadow page pool index of the user table.
4164	* @param iUserTable The index into the user table (shadowed).
4165	*/
4166	void pgmPoolFree(PVM pVM, RTHCPHYS HCPhys, uint16_t iUser, uint32_t iUserTable)
4167	{
4168	LogFlow(("pgmPoolFree: HCPhys=%RHp iUser=%#x iUserTable=%#x\n", HCPhys, iUser, iUserTable));
4169	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4170	pgmPoolFreeByPage(pPool, pgmPoolGetPage(pPool, HCPhys), iUser, iUserTable);
4171	}
4172
4173	/**
4174	* Internal worker for finding a 'in-use' shadow page give by it's physical address.
4175	*
4176	* @returns Pointer to the shadow page structure.
4177	* @param pPool The pool.
4178	* @param HCPhys The HC physical address of the shadow page.
4179	*/
4180	PPGMPOOLPAGE pgmPoolGetPage(PPGMPOOL pPool, RTHCPHYS HCPhys)
4181	{
4182	PVM pVM = pPool->CTX_SUFF(pVM);
4183
4184	/*
4185	* Look up the page.
4186	*/
4187	pgmLock(pVM);
4188	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, HCPhys & X86_PTE_PAE_PG_MASK);
4189	pgmUnlock(pVM);
4190
4191	AssertFatalMsg(pPage && pPage->enmKind != PGMPOOLKIND_FREE, ("HCPhys=%RHp pPage=%p idx=%d\n", HCPhys, pPage, (pPage) ? pPage->idx : 0));
4192	return pPage;
4193	}
4194
4195
4196	#ifdef IN_RING3
4197	/**
4198	* Flushes the entire cache.
4199	*
4200	* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
4201	* and execute this CR3 flush.
4202	*
4203	* @param pPool The pool.
4204	*/
4205	void pgmR3PoolReset(PVM pVM)
4206	{
4207	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4208
4209	Assert(PGMIsLockOwner(pVM));
4210	STAM_PROFILE_START(&pPool->StatFlushAllInt, a);
4211	LogFlow(("pgmPoolFlushAllInt:\n"));
4212
4213	/*
4214	* If there are no pages in the pool, there is nothing to do.
4215	*/
4216	if (pPool->cCurPages <= PGMPOOL_IDX_FIRST)
4217	{
4218	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
4219	return;
4220	}
4221
4222	/*
4223	* Exit the shadow mode since we're going to clear everything,
4224	* including the root page.
4225	*/
4226	for (unsigned i=0;i<pVM->cCPUs;i++)
4227	{
4228	PVMCPU pVCpu = &pVM->aCpus[i];
4229	pgmR3ExitShadowModeBeforePoolFlush(pVM, pVCpu);
4230	}
4231
4232	/*
4233	* Nuke the free list and reinsert all pages into it.
4234	*/
4235	for (unsigned i = pPool->cCurPages - 1; i >= PGMPOOL_IDX_FIRST; i--)
4236	{
4237	PPGMPOOLPAGE pPage = &pPool->aPages[i];
4238
4239	Assert(pPage->Core.Key == MMPage2Phys(pVM, pPage->pvPageR3));
4240	#ifdef PGMPOOL_WITH_MONITORING
4241	if (pPage->fMonitored)
4242	pgmPoolMonitorFlush(pPool, pPage);
4243	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4244	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4245	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
4246	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
4247	pPage->cModifications = 0;
4248	#endif
4249	pPage->GCPhys = NIL_RTGCPHYS;
4250	pPage->enmKind = PGMPOOLKIND_FREE;
4251	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
4252	Assert(pPage->idx == i);
4253	pPage->iNext = i + 1;
4254	pPage->fZeroed = false; /* This could probably be optimized, but better safe than sorry. */
4255	pPage->fSeenNonGlobal = false;
4256	pPage->fMonitored = false;
4257	pPage->fCached = false;
4258	pPage->fReusedFlushPending = false;
4259	#ifdef PGMPOOL_WITH_USER_TRACKING
4260	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
4261	#else
4262	pPage->fCR3Mix = false;
4263	#endif
4264	#ifdef PGMPOOL_WITH_CACHE
4265	pPage->iAgeNext = NIL_PGMPOOL_IDX;
4266	pPage->iAgePrev = NIL_PGMPOOL_IDX;
4267	#endif
4268	pPage->cLocked = 0;
4269	}
4270	pPool->aPages[pPool->cCurPages - 1].iNext = NIL_PGMPOOL_IDX;
4271	pPool->iFreeHead = PGMPOOL_IDX_FIRST;
4272	pPool->cUsedPages = 0;
4273
4274	#ifdef PGMPOOL_WITH_USER_TRACKING
4275	/*
4276	* Zap and reinitialize the user records.
4277	*/
4278	pPool->cPresent = 0;
4279	pPool->iUserFreeHead = 0;
4280	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
4281	const unsigned cMaxUsers = pPool->cMaxUsers;
4282	for (unsigned i = 0; i < cMaxUsers; i++)
4283	{
4284	paUsers[i].iNext = i + 1;
4285	paUsers[i].iUser = NIL_PGMPOOL_IDX;
4286	paUsers[i].iUserTable = 0xfffffffe;
4287	}
4288	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
4289	#endif
4290
4291	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
4292	/*
4293	* Clear all the GCPhys links and rebuild the phys ext free list.
4294	*/
4295	for (PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRanges);
4296	pRam;
4297	pRam = pRam->CTX_SUFF(pNext))
4298	{
4299	unsigned iPage = pRam->cb >> PAGE_SHIFT;
4300	while (iPage-- > 0)
4301	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
4302	}
4303
4304	pPool->iPhysExtFreeHead = 0;
4305	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
4306	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
4307	for (unsigned i = 0; i < cMaxPhysExts; i++)
4308	{
4309	paPhysExts[i].iNext = i + 1;
4310	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
4311	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
4312	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
4313	}
4314	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
4315	#endif
4316
4317	#ifdef PGMPOOL_WITH_MONITORING
4318	/*
4319	* Just zap the modified list.
4320	*/
4321	pPool->cModifiedPages = 0;
4322	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
4323	#endif
4324
4325	#ifdef PGMPOOL_WITH_CACHE
4326	/*
4327	* Clear the GCPhys hash and the age list.
4328	*/
4329	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
4330	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
4331	pPool->iAgeHead = NIL_PGMPOOL_IDX;
4332	pPool->iAgeTail = NIL_PGMPOOL_IDX;
4333	#endif
4334
4335	/*
4336	* Reinsert active pages into the hash and ensure monitoring chains are correct.
4337	*/
4338	for (unsigned i = PGMPOOL_IDX_FIRST_SPECIAL; i < PGMPOOL_IDX_FIRST; i++)
4339	{
4340	PPGMPOOLPAGE pPage = &pPool->aPages[i];
4341	pPage->iNext = NIL_PGMPOOL_IDX;
4342	#ifdef PGMPOOL_WITH_MONITORING
4343	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4344	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4345	pPage->cModifications = 0;
4346	/* ASSUMES that we're not sharing with any of the other special pages (safe for now). */
4347	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
4348	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
4349	if (pPage->fMonitored)
4350	{
4351	int rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
4352	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
4353	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
4354	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
4355	pPool->pszAccessHandler);
4356	AssertFatalRCSuccess(rc);
4357	# ifdef PGMPOOL_WITH_CACHE
4358	pgmPoolHashInsert(pPool, pPage);
4359	# endif
4360	}
4361	#endif
4362	#ifdef PGMPOOL_WITH_USER_TRACKING
4363	Assert(pPage->iUserHead == NIL_PGMPOOL_USER_INDEX); /* for now */
4364	#endif
4365	#ifdef PGMPOOL_WITH_CACHE
4366	Assert(pPage->iAgeNext == NIL_PGMPOOL_IDX);
4367	Assert(pPage->iAgePrev == NIL_PGMPOOL_IDX);
4368	#endif
4369	}
4370
4371	for (unsigned i=0;i<pVM->cCPUs;i++)
4372	{
4373	PVMCPU pVCpu = &pVM->aCpus[i];
4374	/*
4375	* Re-enter the shadowing mode and assert Sync CR3 FF.
4376	*/
4377	pgmR3ReEnterShadowModeAfterPoolFlush(pVM, pVCpu);
4378	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
4379	}
4380
4381	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
4382	}
4383	#endif /* IN_RING3 */
4384
4385	#ifdef LOG_ENABLED
4386	static const char *pgmPoolPoolKindToStr(uint8_t enmKind)
4387	{
4388	switch(enmKind)
4389	{
4390	case PGMPOOLKIND_INVALID:
4391	return "PGMPOOLKIND_INVALID";
4392	case PGMPOOLKIND_FREE:
4393	return "PGMPOOLKIND_FREE";
4394	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
4395	return "PGMPOOLKIND_32BIT_PT_FOR_PHYS";
4396	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
4397	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT";
4398	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
4399	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB";
4400	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
4401	return "PGMPOOLKIND_PAE_PT_FOR_PHYS";
4402	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
4403	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_PT";
4404	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
4405	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB";
4406	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
4407	return "PGMPOOLKIND_PAE_PT_FOR_PAE_PT";
4408	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
4409	return "PGMPOOLKIND_PAE_PT_FOR_PAE_2MB";
4410	case PGMPOOLKIND_32BIT_PD:
4411	return "PGMPOOLKIND_32BIT_PD";
4412	case PGMPOOLKIND_32BIT_PD_PHYS:
4413	return "PGMPOOLKIND_32BIT_PD_PHYS";
4414	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
4415	return "PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD";
4416	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
4417	return "PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD";
4418	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
4419	return "PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD";
4420	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
4421	return "PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD";
4422	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
4423	return "PGMPOOLKIND_PAE_PD_FOR_PAE_PD";
4424	case PGMPOOLKIND_PAE_PD_PHYS:
4425	return "PGMPOOLKIND_PAE_PD_PHYS";
4426	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
4427	return "PGMPOOLKIND_PAE_PDPT_FOR_32BIT";
4428	case PGMPOOLKIND_PAE_PDPT:
4429	return "PGMPOOLKIND_PAE_PDPT";
4430	case PGMPOOLKIND_PAE_PDPT_PHYS:
4431	return "PGMPOOLKIND_PAE_PDPT_PHYS";
4432	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
4433	return "PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT";
4434	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
4435	return "PGMPOOLKIND_64BIT_PDPT_FOR_PHYS";
4436	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
4437	return "PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD";
4438	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
4439	return "PGMPOOLKIND_64BIT_PD_FOR_PHYS";
4440	case PGMPOOLKIND_64BIT_PML4:
4441	return "PGMPOOLKIND_64BIT_PML4";
4442	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
4443	return "PGMPOOLKIND_EPT_PDPT_FOR_PHYS";
4444	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
4445	return "PGMPOOLKIND_EPT_PD_FOR_PHYS";
4446	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
4447	return "PGMPOOLKIND_EPT_PT_FOR_PHYS";
4448	case PGMPOOLKIND_ROOT_NESTED:
4449	return "PGMPOOLKIND_ROOT_NESTED";
4450	}
4451	return "Unknown kind!";
4452	}
4453	#endif /* LOG_ENABLED*/

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllPool.cpp@ 20140

Download in other formats: