PGMAllPool.cpp@ 20563

Last change on this file since 20563 was 20530, checked in by vboxsync, 16 years ago
VMM: remove DISCPUSTATE from the stack.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 162.1 KB

Line
1	/* $Id: PGMAllPool.cpp 20530 2009-06-13 20:53:44Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Sun Microsystems, Inc.
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*
17	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
18	* Clara, CA 95054 USA or visit http://www.sun.com if you need
19	* additional information or have any questions.
20	*/
21
22
23	/*******************************************************************************
24	* Header Files *
25	*******************************************************************************/
26	#define LOG_GROUP LOG_GROUP_PGM_POOL
27	#include <VBox/pgm.h>
28	#include <VBox/mm.h>
29	#include <VBox/em.h>
30	#include <VBox/cpum.h>
31	#ifdef IN_RC
32	# include <VBox/patm.h>
33	#endif
34	#include "PGMInternal.h"
35	#include <VBox/vm.h>
36	#include <VBox/disopcode.h>
37	#include <VBox/hwacc_vmx.h>
38
39	#include <VBox/log.h>
40	#include <VBox/err.h>
41	#include <iprt/asm.h>
42	#include <iprt/string.h>
43
44
45	/*******************************************************************************
46	* Internal Functions *
47	*******************************************************************************/
48	RT_C_DECLS_BEGIN
49	static void pgmPoolFlushAllInt(PPGMPOOL pPool);
50	#ifdef PGMPOOL_WITH_USER_TRACKING
51	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind);
52	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind);
53	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
54	#endif
55	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
56	static void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint);
57	#endif
58	#ifdef PGMPOOL_WITH_CACHE
59	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable);
60	#endif
61	#ifdef PGMPOOL_WITH_MONITORING
62	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
63	#endif
64	#ifndef IN_RING3
65	DECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser);
66	#endif
67	#ifdef LOG_ENABLED
68	static const char *pgmPoolPoolKindToStr(uint8_t enmKind);
69	#endif
70
71	void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs);
72	void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, uint16_t iPhysExt);
73	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage);
74	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt);
75	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt);
76	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt);
77
78	RT_C_DECLS_END
79
80
81	/**
82	* Checks if the specified page pool kind is for a 4MB or 2MB guest page.
83	*
84	* @returns true if it's the shadow of a 4MB or 2MB guest page, otherwise false.
85	* @param enmKind The page kind.
86	*/
87	DECLINLINE(bool) pgmPoolIsBigPage(PGMPOOLKIND enmKind)
88	{
89	switch (enmKind)
90	{
91	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
92	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
93	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
94	return true;
95	default:
96	return false;
97	}
98	}
99
100	/** @def PGMPOOL_PAGE_2_LOCKED_PTR
101	* Maps a pool page pool into the current context and lock it (RC only).
102	*
103	* @returns VBox status code.
104	* @param pVM The VM handle.
105	* @param pPage The pool page.
106	*
107	* @remark In RC this uses PGMGCDynMapHCPage(), so it will consume of the
108	* small page window employeed by that function. Be careful.
109	* @remark There is no need to assert on the result.
110	*/
111	#if defined(IN_RC)
112	DECLINLINE(void *) PGMPOOL_PAGE_2_LOCKED_PTR(PVM pVM, PPGMPOOLPAGE pPage)
113	{
114	void *pv = pgmPoolMapPageInlined(&pVM->pgm.s, pPage);
115
116	/* Make sure the dynamic mapping will not be reused. */
117	if (pv)
118	PGMDynLockHCPage(pVM, (uint8_t *)pv);
119
120	return pv;
121	}
122	#else
123	# define PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage) PGMPOOL_PAGE_2_PTR(pVM, pPage)
124	#endif
125
126	/** @def PGMPOOL_UNLOCK_PTR
127	* Unlock a previously locked dynamic caching (RC only).
128	*
129	* @returns VBox status code.
130	* @param pVM The VM handle.
131	* @param pPage The pool page.
132	*
133	* @remark In RC this uses PGMGCDynMapHCPage(), so it will consume of the
134	* small page window employeed by that function. Be careful.
135	* @remark There is no need to assert on the result.
136	*/
137	#if defined(IN_RC)
138	DECLINLINE(void) PGMPOOL_UNLOCK_PTR(PVM pVM, void *pvPage)
139	{
140	if (pvPage)
141	PGMDynUnlockHCPage(pVM, (uint8_t *)pvPage);
142	}
143	#else
144	# define PGMPOOL_UNLOCK_PTR(pVM, pPage) do {} while (0)
145	#endif
146
147
148	#ifdef PGMPOOL_WITH_MONITORING
149	/**
150	* Determin the size of a write instruction.
151	* @returns number of bytes written.
152	* @param pDis The disassembler state.
153	*/
154	static unsigned pgmPoolDisasWriteSize(PDISCPUSTATE pDis)
155	{
156	/*
157	* This is very crude and possibly wrong for some opcodes,
158	* but since it's not really supposed to be called we can
159	* probably live with that.
160	*/
161	return DISGetParamSize(pDis, &pDis->param1);
162	}
163
164
165	/**
166	* Flushes a chain of pages sharing the same access monitor.
167	*
168	* @returns VBox status code suitable for scheduling.
169	* @param pPool The pool.
170	* @param pPage A page in the chain.
171	*/
172	int pgmPoolMonitorChainFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
173	{
174	LogFlow(("pgmPoolMonitorChainFlush: Flush page %RGp type=%d\n", pPage->GCPhys, pPage->enmKind));
175
176	/*
177	* Find the list head.
178	*/
179	uint16_t idx = pPage->idx;
180	if (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
181	{
182	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
183	{
184	idx = pPage->iMonitoredPrev;
185	Assert(idx != pPage->idx);
186	pPage = &pPool->aPages[idx];
187	}
188	}
189
190	/*
191	* Iterate the list flushing each shadow page.
192	*/
193	int rc = VINF_SUCCESS;
194	for (;;)
195	{
196	idx = pPage->iMonitoredNext;
197	Assert(idx != pPage->idx);
198	if (pPage->idx >= PGMPOOL_IDX_FIRST)
199	{
200	int rc2 = pgmPoolFlushPage(pPool, pPage);
201	AssertRC(rc2);
202	}
203	/* next */
204	if (idx == NIL_PGMPOOL_IDX)
205	break;
206	pPage = &pPool->aPages[idx];
207	}
208	return rc;
209	}
210
211
212	/**
213	* Wrapper for getting the current context pointer to the entry being modified.
214	*
215	* @returns VBox status code suitable for scheduling.
216	* @param pVM VM Handle.
217	* @param pvDst Destination address
218	* @param pvSrc Source guest virtual address.
219	* @param GCPhysSrc The source guest physical address.
220	* @param cb Size of data to read
221	*/
222	DECLINLINE(int) pgmPoolPhysSimpleReadGCPhys(PVM pVM, void *pvDst, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvSrc, RTGCPHYS GCPhysSrc, size_t cb)
223	{
224	#if defined(IN_RING3)
225	memcpy(pvDst, (RTHCPTR)((uintptr_t)pvSrc & ~(RTHCUINTPTR)(cb - 1)), cb);
226	return VINF_SUCCESS;
227	#else
228	/* @todo in RC we could attempt to use the virtual address, although this can cause many faults (PAE Windows XP guest). */
229	return PGMPhysSimpleReadGCPhys(pVM, pvDst, GCPhysSrc & ~(RTGCPHYS)(cb - 1), cb);
230	#endif
231	}
232
233	/**
234	* Process shadow entries before they are changed by the guest.
235	*
236	* For PT entries we will clear them. For PD entries, we'll simply check
237	* for mapping conflicts and set the SyncCR3 FF if found.
238	*
239	* @param pVCpu VMCPU handle
240	* @param pPool The pool.
241	* @param pPage The head page.
242	* @param GCPhysFault The guest physical fault address.
243	* @param uAddress In R0 and GC this is the guest context fault address (flat).
244	* In R3 this is the host context 'fault' address.
245	* @param pDis The disassembler state for figuring out the write size.
246	* This need not be specified if the caller knows we won't do cross entry accesses.
247	*/
248	void pgmPoolMonitorChainChanging(PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhysFault, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvAddress, PDISCPUSTATE pDis)
249	{
250	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
251	const unsigned off = GCPhysFault & PAGE_OFFSET_MASK;
252	const unsigned cbWrite = pDis ? pgmPoolDisasWriteSize(pDis) : 0;
253	PVM pVM = pPool->CTX_SUFF(pVM);
254
255	LogFlow(("pgmPoolMonitorChainChanging: %RGv phys=%RGp kind=%s cbWrite=%d\n", (RTGCPTR)pvAddress, GCPhysFault, pgmPoolPoolKindToStr(pPage->enmKind), cbWrite));
256	for (;;)
257	{
258	union
259	{
260	void *pv;
261	PX86PT pPT;
262	PX86PTPAE pPTPae;
263	PX86PD pPD;
264	PX86PDPAE pPDPae;
265	PX86PDPT pPDPT;
266	PX86PML4 pPML4;
267	} uShw;
268
269	uShw.pv = NULL;
270	switch (pPage->enmKind)
271	{
272	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
273	{
274	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
275	const unsigned iShw = off / sizeof(X86PTE);
276	LogFlow(("PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT iShw=%x\n", iShw));
277	if (uShw.pPT->a[iShw].n.u1Present)
278	{
279	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
280	X86PTE GstPte;
281
282	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
283	AssertRC(rc);
284	Log4(("pgmPoolMonitorChainChanging 32_32: deref %016RX64 GCPhys %08RX32\n", uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PG_MASK));
285	pgmPoolTracDerefGCPhysHint(pPool, pPage,
286	uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK,
287	GstPte.u & X86_PTE_PG_MASK);
288	# endif
289	ASMAtomicWriteSize(&uShw.pPT->a[iShw], 0);
290	}
291	break;
292	}
293
294	/* page/2 sized */
295	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
296	{
297	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
298	if (!((off ^ pPage->GCPhys) & (PAGE_SIZE / 2)))
299	{
300	const unsigned iShw = (off / sizeof(X86PTE)) & (X86_PG_PAE_ENTRIES - 1);
301	LogFlow(("PGMPOOLKIND_PAE_PT_FOR_32BIT_PT iShw=%x\n", iShw));
302	if (uShw.pPTPae->a[iShw].n.u1Present)
303	{
304	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
305	X86PTE GstPte;
306	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
307	AssertRC(rc);
308
309	Log4(("pgmPoolMonitorChainChanging pae_32: deref %016RX64 GCPhys %08RX32\n", uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PG_MASK));
310	pgmPoolTracDerefGCPhysHint(pPool, pPage,
311	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
312	GstPte.u & X86_PTE_PG_MASK);
313	# endif
314	ASMAtomicWriteSize(&uShw.pPTPae->a[iShw], 0);
315	}
316	}
317	break;
318	}
319
320	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
321	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
322	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
323	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
324	{
325	unsigned iGst = off / sizeof(X86PDE);
326	unsigned iShwPdpt = iGst / 256;
327	unsigned iShw = (iGst % 256) * 2;
328	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
329
330	LogFlow(("pgmPoolMonitorChainChanging PAE for 32 bits: iGst=%x iShw=%x idx = %d page idx=%d\n", iGst, iShw, iShwPdpt, pPage->enmKind - PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD));
331	if (iShwPdpt == pPage->enmKind - (unsigned)PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD)
332	{
333	for (unsigned i = 0; i < 2; i++)
334	{
335	# ifndef IN_RING0
336	if ((uShw.pPDPae->a[iShw + i].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
337	{
338	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
339	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
340	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw=%#x!\n", iShwPdpt, iShw+i));
341	break;
342	}
343	else
344	# endif /* !IN_RING0 */
345	if (uShw.pPDPae->a[iShw+i].n.u1Present)
346	{
347	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw+i, uShw.pPDPae->a[iShw+i].u));
348	pgmPoolFree(pVM,
349	uShw.pPDPae->a[iShw+i].u & X86_PDE_PAE_PG_MASK,
350	pPage->idx,
351	iShw + i);
352	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw+i], 0);
353	}
354
355	/* paranoia / a bit assumptive. */
356	if ( pDis
357	&& (off & 3)
358	&& (off & 3) + cbWrite > 4)
359	{
360	const unsigned iShw2 = iShw + 2 + i;
361	if (iShw2 < RT_ELEMENTS(uShw.pPDPae->a))
362	{
363	# ifndef IN_RING0
364	if ((uShw.pPDPae->a[iShw2].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
365	{
366	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
367	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
368	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw2=%#x!\n", iShwPdpt, iShw2));
369	break;
370	}
371	else
372	# endif /* !IN_RING0 */
373	if (uShw.pPDPae->a[iShw2].n.u1Present)
374	{
375	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
376	pgmPoolFree(pVM,
377	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
378	pPage->idx,
379	iShw2);
380	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw2].u, 0);
381	}
382	}
383	}
384	}
385	}
386	break;
387	}
388
389	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
390	{
391	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
392	const unsigned iShw = off / sizeof(X86PTEPAE);
393	if (uShw.pPTPae->a[iShw].n.u1Present)
394	{
395	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
396	X86PTEPAE GstPte;
397	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
398	AssertRC(rc);
399
400	Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PAE_PG_MASK));
401	pgmPoolTracDerefGCPhysHint(pPool, pPage,
402	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
403	GstPte.u & X86_PTE_PAE_PG_MASK);
404	# endif
405	ASMAtomicWriteSize(&uShw.pPTPae->a[iShw].u, 0);
406	}
407
408	/* paranoia / a bit assumptive. */
409	if ( pDis
410	&& (off & 7)
411	&& (off & 7) + cbWrite > sizeof(X86PTEPAE))
412	{
413	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTEPAE);
414	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPTPae->a));
415
416	if (uShw.pPTPae->a[iShw2].n.u1Present)
417	{
418	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
419	X86PTEPAE GstPte;
420	# ifdef IN_RING3
421	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, (RTHCPTR)((RTHCUINTPTR)pvAddress + sizeof(GstPte)), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
422	# else
423	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress + sizeof(GstPte), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
424	# endif
425	AssertRC(rc);
426	Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PAE_PG_MASK));
427	pgmPoolTracDerefGCPhysHint(pPool, pPage,
428	uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK,
429	GstPte.u & X86_PTE_PAE_PG_MASK);
430	# endif
431	ASMAtomicWriteSize(&uShw.pPTPae->a[iShw2].u ,0);
432	}
433	}
434	break;
435	}
436
437	case PGMPOOLKIND_32BIT_PD:
438	{
439	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
440	const unsigned iShw = off / sizeof(X86PTE); // ASSUMING 32-bit guest paging!
441
442	LogFlow(("pgmPoolMonitorChainChanging: PGMPOOLKIND_32BIT_PD %x\n", iShw));
443	# ifndef IN_RING0
444	if (uShw.pPD->a[iShw].u & PGM_PDFLAGS_MAPPING)
445	{
446	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
447	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
448	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
449	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
450	break;
451	}
452	# endif /* !IN_RING0 */
453	# ifndef IN_RING0
454	else
455	# endif /* !IN_RING0 */
456	{
457	if (uShw.pPD->a[iShw].n.u1Present)
458	{
459	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
460	pgmPoolFree(pVM,
461	uShw.pPD->a[iShw].u & X86_PDE_PAE_PG_MASK,
462	pPage->idx,
463	iShw);
464	ASMAtomicWriteSize(&uShw.pPD->a[iShw].u, 0);
465	}
466	}
467	/* paranoia / a bit assumptive. */
468	if ( pDis
469	&& (off & 3)
470	&& (off & 3) + cbWrite > sizeof(X86PTE))
471	{
472	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTE);
473	if ( iShw2 != iShw
474	&& iShw2 < RT_ELEMENTS(uShw.pPD->a))
475	{
476	# ifndef IN_RING0
477	if (uShw.pPD->a[iShw2].u & PGM_PDFLAGS_MAPPING)
478	{
479	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
480	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
481	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
482	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
483	break;
484	}
485	# endif /* !IN_RING0 */
486	# ifndef IN_RING0
487	else
488	# endif /* !IN_RING0 */
489	{
490	if (uShw.pPD->a[iShw2].n.u1Present)
491	{
492	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPD->a[iShw2].u));
493	pgmPoolFree(pVM,
494	uShw.pPD->a[iShw2].u & X86_PDE_PAE_PG_MASK,
495	pPage->idx,
496	iShw2);
497	ASMAtomicWriteSize(&uShw.pPD->a[iShw2].u, 0);
498	}
499	}
500	}
501	}
502	#if 0 /* useful when running PGMAssertCR3(), a bit too troublesome for general use (TLBs). */
503	if ( uShw.pPD->a[iShw].n.u1Present
504	&& !VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
505	{
506	LogFlow(("pgmPoolMonitorChainChanging: iShw=%#x: %RX32 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
507	# ifdef IN_RC /* TLB load - we're pushing things a bit... */
508	ASMProbeReadByte(pvAddress);
509	# endif
510	pgmPoolFree(pVM, uShw.pPD->a[iShw].u & X86_PDE_PG_MASK, pPage->idx, iShw);
511	ASMAtomicWriteSize(&uShw.pPD->a[iShw].u, 0);
512	}
513	#endif
514	break;
515	}
516
517	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
518	{
519	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
520	const unsigned iShw = off / sizeof(X86PDEPAE);
521	#ifndef IN_RING0
522	if (uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING)
523	{
524	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
525	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
526	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
527	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
528	break;
529	}
530	#endif /* !IN_RING0 */
531	/*
532	* Causes trouble when the guest uses a PDE to refer to the whole page table level
533	* structure. (Invalidate here; faults later on when it tries to change the page
534	* table entries -> recheck; probably only applies to the RC case.)
535	*/
536	# ifndef IN_RING0
537	else
538	# endif /* !IN_RING0 */
539	{
540	if (uShw.pPDPae->a[iShw].n.u1Present)
541	{
542	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
543	pgmPoolFree(pVM,
544	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
545	pPage->idx,
546	iShw);
547	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw].u, 0);
548	}
549	}
550	/* paranoia / a bit assumptive. */
551	if ( pDis
552	&& (off & 7)
553	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
554	{
555	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
556	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
557
558	#ifndef IN_RING0
559	if ( iShw2 != iShw
560	&& uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING)
561	{
562	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
563	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
564	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
565	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
566	break;
567	}
568	#endif /* !IN_RING0 */
569	# ifndef IN_RING0
570	else
571	# endif /* !IN_RING0 */
572	if (uShw.pPDPae->a[iShw2].n.u1Present)
573	{
574	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
575	pgmPoolFree(pVM,
576	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
577	pPage->idx,
578	iShw2);
579	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw2].u, 0);
580	}
581	}
582	break;
583	}
584
585	case PGMPOOLKIND_PAE_PDPT:
586	{
587	/*
588	* Hopefully this doesn't happen very often:
589	* - touching unused parts of the page
590	* - messing with the bits of pd pointers without changing the physical address
591	*/
592	/* PDPT roots are not page aligned; 32 byte only! */
593	const unsigned offPdpt = GCPhysFault - pPage->GCPhys;
594
595	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
596	const unsigned iShw = offPdpt / sizeof(X86PDPE);
597	if (iShw < X86_PG_PAE_PDPE_ENTRIES) /* don't use RT_ELEMENTS(uShw.pPDPT->a), because that's for long mode only */
598	{
599	# ifndef IN_RING0
600	if (uShw.pPDPT->a[iShw].u & PGM_PLXFLAGS_MAPPING)
601	{
602	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
603	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
604	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
605	LogFlow(("pgmPoolMonitorChainChanging: Detected pdpt conflict at iShw=%#x!\n", iShw));
606	break;
607	}
608	# endif /* !IN_RING0 */
609	# ifndef IN_RING0
610	else
611	# endif /* !IN_RING0 */
612	if (uShw.pPDPT->a[iShw].n.u1Present)
613	{
614	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
615	pgmPoolFree(pVM,
616	uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK,
617	pPage->idx,
618	iShw);
619	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw].u, 0);
620	}
621
622	/* paranoia / a bit assumptive. */
623	if ( pDis
624	&& (offPdpt & 7)
625	&& (offPdpt & 7) + cbWrite > sizeof(X86PDPE))
626	{
627	const unsigned iShw2 = (offPdpt + cbWrite - 1) / sizeof(X86PDPE);
628	if ( iShw2 != iShw
629	&& iShw2 < X86_PG_PAE_PDPE_ENTRIES)
630	{
631	# ifndef IN_RING0
632	if (uShw.pPDPT->a[iShw2].u & PGM_PLXFLAGS_MAPPING)
633	{
634	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
635	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
636	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
637	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
638	break;
639	}
640	# endif /* !IN_RING0 */
641	# ifndef IN_RING0
642	else
643	# endif /* !IN_RING0 */
644	if (uShw.pPDPT->a[iShw2].n.u1Present)
645	{
646	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
647	pgmPoolFree(pVM,
648	uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK,
649	pPage->idx,
650	iShw2);
651	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw2].u, 0);
652	}
653	}
654	}
655	}
656	break;
657	}
658
659	#ifndef IN_RC
660	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
661	{
662	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
663	const unsigned iShw = off / sizeof(X86PDEPAE);
664	Assert(!(uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING));
665	if (uShw.pPDPae->a[iShw].n.u1Present)
666	{
667	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
668	pgmPoolFree(pVM,
669	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
670	pPage->idx,
671	iShw);
672	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw].u, 0);
673	}
674	/* paranoia / a bit assumptive. */
675	if ( pDis
676	&& (off & 7)
677	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
678	{
679	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
680	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
681
682	Assert(!(uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING));
683	if (uShw.pPDPae->a[iShw2].n.u1Present)
684	{
685	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
686	pgmPoolFree(pVM,
687	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
688	pPage->idx,
689	iShw2);
690	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw2].u, 0);
691	}
692	}
693	break;
694	}
695
696	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
697	{
698	/*
699	* Hopefully this doesn't happen very often:
700	* - messing with the bits of pd pointers without changing the physical address
701	*/
702	if (!VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
703	{
704	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
705	const unsigned iShw = off / sizeof(X86PDPE);
706	if (uShw.pPDPT->a[iShw].n.u1Present)
707	{
708	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
709	pgmPoolFree(pVM, uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK, pPage->idx, iShw);
710	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw].u, 0);
711	}
712	/* paranoia / a bit assumptive. */
713	if ( pDis
714	&& (off & 7)
715	&& (off & 7) + cbWrite > sizeof(X86PDPE))
716	{
717	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDPE);
718	if (uShw.pPDPT->a[iShw2].n.u1Present)
719	{
720	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
721	pgmPoolFree(pVM, uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK, pPage->idx, iShw2);
722	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw2].u, 0);
723	}
724	}
725	}
726	break;
727	}
728
729	case PGMPOOLKIND_64BIT_PML4:
730	{
731	/*
732	* Hopefully this doesn't happen very often:
733	* - messing with the bits of pd pointers without changing the physical address
734	*/
735	if (!VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
736	{
737	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
738	const unsigned iShw = off / sizeof(X86PDPE);
739	if (uShw.pPML4->a[iShw].n.u1Present)
740	{
741	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPML4->a[iShw].u));
742	pgmPoolFree(pVM, uShw.pPML4->a[iShw].u & X86_PML4E_PG_MASK, pPage->idx, iShw);
743	ASMAtomicWriteSize(&uShw.pPML4->a[iShw].u, 0);
744	}
745	/* paranoia / a bit assumptive. */
746	if ( pDis
747	&& (off & 7)
748	&& (off & 7) + cbWrite > sizeof(X86PDPE))
749	{
750	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PML4E);
751	if (uShw.pPML4->a[iShw2].n.u1Present)
752	{
753	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPML4->a[iShw2].u));
754	pgmPoolFree(pVM, uShw.pPML4->a[iShw2].u & X86_PML4E_PG_MASK, pPage->idx, iShw2);
755	ASMAtomicWriteSize(&uShw.pPML4->a[iShw2].u, 0);
756	}
757	}
758	}
759	break;
760	}
761	#endif /* IN_RING0 */
762
763	default:
764	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
765	}
766	PGMPOOL_UNLOCK_PTR(pVM, uShw.pv);
767
768	/* next */
769	if (pPage->iMonitoredNext == NIL_PGMPOOL_IDX)
770	return;
771	pPage = &pPool->aPages[pPage->iMonitoredNext];
772	}
773	}
774
775	# ifndef IN_RING3
776	/**
777	* Checks if a access could be a fork operation in progress.
778	*
779	* Meaning, that the guest is setuping up the parent process for Copy-On-Write.
780	*
781	* @returns true if it's likly that we're forking, otherwise false.
782	* @param pPool The pool.
783	* @param pDis The disassembled instruction.
784	* @param offFault The access offset.
785	*/
786	DECLINLINE(bool) pgmPoolMonitorIsForking(PPGMPOOL pPool, PDISCPUSTATE pDis, unsigned offFault)
787	{
788	/*
789	* i386 linux is using btr to clear X86_PTE_RW.
790	* The functions involved are (2.6.16 source inspection):
791	* clear_bit
792	* ptep_set_wrprotect
793	* copy_one_pte
794	* copy_pte_range
795	* copy_pmd_range
796	* copy_pud_range
797	* copy_page_range
798	* dup_mmap
799	* dup_mm
800	* copy_mm
801	* copy_process
802	* do_fork
803	*/
804	if ( pDis->pCurInstr->opcode == OP_BTR
805	&& !(offFault & 4)
806	/** @todo Validate that the bit index is X86_PTE_RW. */
807	)
808	{
809	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,Fork));
810	return true;
811	}
812	return false;
813	}
814
815
816	/**
817	* Determine whether the page is likely to have been reused.
818	*
819	* @returns true if we consider the page as being reused for a different purpose.
820	* @returns false if we consider it to still be a paging page.
821	* @param pVM VM Handle.
822	* @param pRegFrame Trap register frame.
823	* @param pDis 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 pDis, 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 (pDis->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 ( pDis->prefix == (PREFIX_REP\|PREFIX_REX)
870	&& pRegFrame->rcx >= 0x40
871	)
872	{
873	Assert(pDis->mode == CPUMODE_64BIT);
874
875	Log(("pgmPoolMonitorIsReused: OP_STOSQ\n"));
876	return true;
877	}
878	return false;
879	}
880	if ( (pDis->param1.flags & USE_REG_GEN32)
881	&& (pDis->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 pDis 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 pDis,
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, pDis, pRegFrame, pvFault, &cbWritten);
917	if (RT_SUCCESS(rc2))
918	pRegFrame->rip += pDis->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 pDis 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 pDis,
961	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
962	{
963	Assert(pDis->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 += pDis->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 pDis 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 pDis,
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, pDis);
1043	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
1044	#else
1045	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, pvFault, pDis);
1046	#endif
1047
1048	/*
1049	* Interpret the instruction.
1050	*/
1051	uint32_t cb;
1052	int rc = EMInterpretInstructionCPU(pVM, pVCpu, pDis, pRegFrame, pvFault, &cb);
1053	if (RT_SUCCESS(rc))
1054	pRegFrame->rip += pDis->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, pDis->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	PDISCPUSTATE pDis = &pVCpu->pgm.s.DisState;
1114	int rc = EMInterpretDisasOne(pVM, pVCpu, pRegFrame, pDis, 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, pDis, pvFault))
1135	&& !pgmPoolMonitorIsForking(pPool, pDis, GCPhysFault & PAGE_OFFSET_MASK))
1136	{
1137	/*
1138	* Simple instructions, no REP prefix.
1139	*/
1140	if (!(pDis->prefix & (PREFIX_REP \| PREFIX_REPNE)))
1141	{
1142	rc = pgmPoolAccessHandlerSimple(pVM, pVCpu, pPool, pPage, pDis, 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 ( pDis->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	&& pDis->mode == CPUMODE_32BIT
1159	&& pDis->opmode == CPUMODE_32BIT
1160	&& pDis->addrmode == CPUMODE_32BIT
1161	&& pDis->prefix == PREFIX_REP
1162	&& !pRegFrame->eflags.Bits.u1DF
1163	)
1164	{
1165	rc = pgmPoolAccessHandlerSTOSD(pVM, pPool, pPage, pDis, 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, pDis->pCurInstr->opcode, pDis->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, pDis, 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(!(VMMGetCpu(pVM)->pgm.s.fSyncFlags & PGM_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	#ifdef VBOX_STRICT
1846	PVMCPU pVCpu = VMMGetCpu(pVM);
1847	#endif
1848	AssertMsg(!(pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3),
1849	("%#x %#x\n", pVCpu->pgm.s.fSyncFlags, pVM->fGlobalForcedActions));
1850	}
1851	pPage->fMonitored = false;
1852
1853	/*
1854	* Remove it from the list of modified pages (if in it).
1855	*/
1856	pgmPoolMonitorModifiedRemove(pPool, pPage);
1857
1858	return rc;
1859	}
1860
1861
1862	/**
1863	* Inserts the page into the list of modified pages.
1864	*
1865	* @param pPool The pool.
1866	* @param pPage The page.
1867	*/
1868	void pgmPoolMonitorModifiedInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1869	{
1870	Log3(("pgmPoolMonitorModifiedInsert: idx=%d\n", pPage->idx));
1871	AssertMsg( pPage->iModifiedNext == NIL_PGMPOOL_IDX
1872	&& pPage->iModifiedPrev == NIL_PGMPOOL_IDX
1873	&& pPool->iModifiedHead != pPage->idx,
1874	("Next=%d Prev=%d idx=%d cModifications=%d Head=%d cModifiedPages=%d\n",
1875	pPage->iModifiedNext, pPage->iModifiedPrev, pPage->idx, pPage->cModifications,
1876	pPool->iModifiedHead, pPool->cModifiedPages));
1877
1878	pPage->iModifiedNext = pPool->iModifiedHead;
1879	if (pPool->iModifiedHead != NIL_PGMPOOL_IDX)
1880	pPool->aPages[pPool->iModifiedHead].iModifiedPrev = pPage->idx;
1881	pPool->iModifiedHead = pPage->idx;
1882	pPool->cModifiedPages++;
1883	#ifdef VBOX_WITH_STATISTICS
1884	if (pPool->cModifiedPages > pPool->cModifiedPagesHigh)
1885	pPool->cModifiedPagesHigh = pPool->cModifiedPages;
1886	#endif
1887	}
1888
1889
1890	/**
1891	* Removes the page from the list of modified pages and resets the
1892	* moficiation counter.
1893	*
1894	* @param pPool The pool.
1895	* @param pPage The page which is believed to be in the list of modified pages.
1896	*/
1897	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1898	{
1899	Log3(("pgmPoolMonitorModifiedRemove: idx=%d cModifications=%d\n", pPage->idx, pPage->cModifications));
1900	if (pPool->iModifiedHead == pPage->idx)
1901	{
1902	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1903	pPool->iModifiedHead = pPage->iModifiedNext;
1904	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1905	{
1906	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = NIL_PGMPOOL_IDX;
1907	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1908	}
1909	pPool->cModifiedPages--;
1910	}
1911	else if (pPage->iModifiedPrev != NIL_PGMPOOL_IDX)
1912	{
1913	pPool->aPages[pPage->iModifiedPrev].iModifiedNext = pPage->iModifiedNext;
1914	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1915	{
1916	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = pPage->iModifiedPrev;
1917	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1918	}
1919	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1920	pPool->cModifiedPages--;
1921	}
1922	else
1923	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1924	pPage->cModifications = 0;
1925	}
1926
1927
1928	/**
1929	* Zaps the list of modified pages, resetting their modification counters in the process.
1930	*
1931	* @param pVM The VM handle.
1932	*/
1933	void pgmPoolMonitorModifiedClearAll(PVM pVM)
1934	{
1935	pgmLock(pVM);
1936	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1937	LogFlow(("pgmPoolMonitorModifiedClearAll: cModifiedPages=%d\n", pPool->cModifiedPages));
1938
1939	unsigned cPages = 0; NOREF(cPages);
1940	uint16_t idx = pPool->iModifiedHead;
1941	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
1942	while (idx != NIL_PGMPOOL_IDX)
1943	{
1944	PPGMPOOLPAGE pPage = &pPool->aPages[idx];
1945	idx = pPage->iModifiedNext;
1946	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1947	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1948	pPage->cModifications = 0;
1949	Assert(++cPages);
1950	}
1951	AssertMsg(cPages == pPool->cModifiedPages, ("%d != %d\n", cPages, pPool->cModifiedPages));
1952	pPool->cModifiedPages = 0;
1953	pgmUnlock(pVM);
1954	}
1955
1956
1957	#ifdef IN_RING3
1958	/**
1959	* Callback to clear all shadow pages and clear all modification counters.
1960	*
1961	* @returns VBox status code.
1962	* @param pVM The VM handle.
1963	* @param pvUser Unused parameter
1964	* @remark Should only be used when monitoring is available, thus placed in
1965	* the PGMPOOL_WITH_MONITORING #ifdef.
1966	*/
1967	DECLCALLBACK(int) pgmPoolClearAll(PVM pVM, void *pvUser)
1968	{
1969	NOREF(pvUser);
1970	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1971	STAM_PROFILE_START(&pPool->StatClearAll, c);
1972	LogFlow(("pgmPoolClearAll: cUsedPages=%d\n", pPool->cUsedPages));
1973
1974	pgmLock(pVM);
1975
1976	/*
1977	* Iterate all the pages until we've encountered all that in use.
1978	* This is simple but not quite optimal solution.
1979	*/
1980	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
1981	unsigned cLeft = pPool->cUsedPages;
1982	unsigned iPage = pPool->cCurPages;
1983	while (--iPage >= PGMPOOL_IDX_FIRST)
1984	{
1985	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
1986	if (pPage->GCPhys != NIL_RTGCPHYS)
1987	{
1988	switch (pPage->enmKind)
1989	{
1990	/*
1991	* We only care about shadow page tables.
1992	*/
1993	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1994	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1995	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1996	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1997	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1998	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1999	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2000	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2001	{
2002	#ifdef PGMPOOL_WITH_USER_TRACKING
2003	if (pPage->cPresent)
2004	#endif
2005	{
2006	void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
2007	STAM_PROFILE_START(&pPool->StatZeroPage, z);
2008	ASMMemZeroPage(pvShw);
2009	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
2010	#ifdef PGMPOOL_WITH_USER_TRACKING
2011	pPage->cPresent = 0;
2012	pPage->iFirstPresent = ~0;
2013	#endif
2014	}
2015	}
2016	/* fall thru */
2017
2018	default:
2019	Assert(!pPage->cModifications \|\| ++cModifiedPages);
2020	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2021	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2022	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2023	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2024	pPage->cModifications = 0;
2025	break;
2026
2027	}
2028	if (!--cLeft)
2029	break;
2030	}
2031	}
2032
2033	/* swipe the special pages too. */
2034	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
2035	{
2036	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2037	if (pPage->GCPhys != NIL_RTGCPHYS)
2038	{
2039	Assert(!pPage->cModifications \|\| ++cModifiedPages);
2040	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2041	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2042	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2043	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2044	pPage->cModifications = 0;
2045	}
2046	}
2047
2048	#ifndef DEBUG_michael
2049	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
2050	#endif
2051	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
2052	pPool->cModifiedPages = 0;
2053
2054	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2055	/*
2056	* Clear all the GCPhys links and rebuild the phys ext free list.
2057	*/
2058	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
2059	pRam;
2060	pRam = pRam->CTX_SUFF(pNext))
2061	{
2062	unsigned iPage = pRam->cb >> PAGE_SHIFT;
2063	while (iPage-- > 0)
2064	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
2065	}
2066
2067	pPool->iPhysExtFreeHead = 0;
2068	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
2069	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
2070	for (unsigned i = 0; i < cMaxPhysExts; i++)
2071	{
2072	paPhysExts[i].iNext = i + 1;
2073	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
2074	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
2075	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
2076	}
2077	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
2078	#endif
2079
2080	/* Clear the PGM_SYNC_CLEAR_PGM_POOL flag on all VCPUs to prevent redundant flushes. */
2081	for (unsigned idCpu = 0; idCpu < pVM->cCPUs; idCpu++)
2082	{
2083	PVMCPU pVCpu = &pVM->aCpus[idCpu];
2084
2085	pVCpu->pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
2086	}
2087
2088	pPool->cPresent = 0;
2089	pgmUnlock(pVM);
2090	PGM_INVL_ALL_VCPU_TLBS(pVM);
2091	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
2092	return VINF_SUCCESS;
2093	}
2094	#endif /* IN_RING3 */
2095
2096
2097	/**
2098	* Handle SyncCR3 pool tasks
2099	*
2100	* @returns VBox status code.
2101	* @retval VINF_SUCCESS if successfully added.
2102	* @retval VINF_PGM_SYNC_CR3 is it needs to be deferred to ring 3 (GC only)
2103	* @param pVCpu The VMCPU handle.
2104	* @remark Should only be used when monitoring is available, thus placed in
2105	* the PGMPOOL_WITH_MONITORING #ifdef.
2106	*/
2107	int pgmPoolSyncCR3(PVMCPU pVCpu)
2108	{
2109	PVM pVM = pVCpu->CTX_SUFF(pVM);
2110	LogFlow(("pgmPoolSyncCR3\n"));
2111	/*
2112	* When monitoring shadowed pages, we reset the modification counters on CR3 sync.
2113	* Occasionally we will have to clear all the shadow page tables because we wanted
2114	* to monitor a page which was mapped by too many shadowed page tables. This operation
2115	* sometimes refered to as a 'lightweight flush'.
2116	*/
2117	# ifdef IN_RING3 /* Don't flush in ring-0 or raw mode, it's taking too long. */
2118	if (ASMBitTestAndClear(&pVCpu->pgm.s.fSyncFlags, PGM_SYNC_CLEAR_PGM_POOL_BIT))
2119	{
2120	VMMR3AtomicExecuteHandler(pVM, pgmPoolClearAll, NULL);
2121	# else /* !IN_RING3 */
2122	if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
2123	{
2124	LogFlow(("SyncCR3: PGM_SYNC_CLEAR_PGM_POOL is set -> VINF_PGM_SYNC_CR3\n"));
2125	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
2126	return VINF_PGM_SYNC_CR3;
2127	# endif /* !IN_RING3 */
2128	}
2129	else
2130	pgmPoolMonitorModifiedClearAll(pVM);
2131
2132	return VINF_SUCCESS;
2133	}
2134
2135	#endif /* PGMPOOL_WITH_MONITORING */
2136	#ifdef PGMPOOL_WITH_USER_TRACKING
2137
2138	/**
2139	* Frees up at least one user entry.
2140	*
2141	* @returns VBox status code.
2142	* @retval VINF_SUCCESS if successfully added.
2143	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2144	* @param pPool The pool.
2145	* @param iUser The user index.
2146	*/
2147	static int pgmPoolTrackFreeOneUser(PPGMPOOL pPool, uint16_t iUser)
2148	{
2149	STAM_COUNTER_INC(&pPool->StatTrackFreeUpOneUser);
2150	#ifdef PGMPOOL_WITH_CACHE
2151	/*
2152	* Just free cached pages in a braindead fashion.
2153	*/
2154	/** @todo walk the age list backwards and free the first with usage. */
2155	int rc = VINF_SUCCESS;
2156	do
2157	{
2158	int rc2 = pgmPoolCacheFreeOne(pPool, iUser);
2159	if (RT_FAILURE(rc2) && rc == VINF_SUCCESS)
2160	rc = rc2;
2161	} while (pPool->iUserFreeHead == NIL_PGMPOOL_USER_INDEX);
2162	return rc;
2163	#else
2164	/*
2165	* Lazy approach.
2166	*/
2167	/* @todo This path no longer works (CR3 root pages will be flushed)!! */
2168	AssertCompileFailed();
2169	Assert(!CPUMIsGuestInLongMode(pVM));
2170	pgmPoolFlushAllInt(pPool);
2171	return VERR_PGM_POOL_FLUSHED;
2172	#endif
2173	}
2174
2175
2176	/**
2177	* Inserts a page into the cache.
2178	*
2179	* This will create user node for the page, insert it into the GCPhys
2180	* hash, and insert it into the age list.
2181	*
2182	* @returns VBox status code.
2183	* @retval VINF_SUCCESS if successfully added.
2184	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2185	* @param pPool The pool.
2186	* @param pPage The cached page.
2187	* @param GCPhys The GC physical address of the page we're gonna shadow.
2188	* @param iUser The user index.
2189	* @param iUserTable The user table index.
2190	*/
2191	DECLINLINE(int) pgmPoolTrackInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhys, uint16_t iUser, uint32_t iUserTable)
2192	{
2193	int rc = VINF_SUCCESS;
2194	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2195
2196	LogFlow(("pgmPoolTrackInsert GCPhys=%RGp iUser %x iUserTable %x\n", GCPhys, iUser, iUserTable));
2197
2198	#ifdef VBOX_STRICT
2199	/*
2200	* Check that the entry doesn't already exists.
2201	*/
2202	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2203	{
2204	uint16_t i = pPage->iUserHead;
2205	do
2206	{
2207	Assert(i < pPool->cMaxUsers);
2208	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2209	i = paUsers[i].iNext;
2210	} while (i != NIL_PGMPOOL_USER_INDEX);
2211	}
2212	#endif
2213
2214	/*
2215	* Find free a user node.
2216	*/
2217	uint16_t i = pPool->iUserFreeHead;
2218	if (i == NIL_PGMPOOL_USER_INDEX)
2219	{
2220	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2221	if (RT_FAILURE(rc))
2222	return rc;
2223	i = pPool->iUserFreeHead;
2224	}
2225
2226	/*
2227	* Unlink the user node from the free list,
2228	* initialize and insert it into the user list.
2229	*/
2230	pPool->iUserFreeHead = paUsers[i].iNext;
2231	paUsers[i].iNext = NIL_PGMPOOL_USER_INDEX;
2232	paUsers[i].iUser = iUser;
2233	paUsers[i].iUserTable = iUserTable;
2234	pPage->iUserHead = i;
2235
2236	/*
2237	* Insert into cache and enable monitoring of the guest page if enabled.
2238	*
2239	* Until we implement caching of all levels, including the CR3 one, we'll
2240	* have to make sure we don't try monitor & cache any recursive reuse of
2241	* a monitored CR3 page. Because all windows versions are doing this we'll
2242	* have to be able to do combined access monitoring, CR3 + PT and
2243	* PD + PT (guest PAE).
2244	*
2245	* Update:
2246	* We're now cooperating with the CR3 monitor if an uncachable page is found.
2247	*/
2248	#if defined(PGMPOOL_WITH_MONITORING) \|\| defined(PGMPOOL_WITH_CACHE)
2249	# ifdef PGMPOOL_WITH_MIXED_PT_CR3
2250	const bool fCanBeMonitored = true;
2251	# else
2252	bool fCanBeMonitored = pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored == NIL_RTGCPHYS
2253	\|\| (GCPhys & X86_PTE_PAE_PG_MASK) != (pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored & X86_PTE_PAE_PG_MASK)
2254	\|\| pgmPoolIsBigPage((PGMPOOLKIND)pPage->enmKind);
2255	# endif
2256	# ifdef PGMPOOL_WITH_CACHE
2257	pgmPoolCacheInsert(pPool, pPage, fCanBeMonitored); /* This can be expanded. */
2258	# endif
2259	if (fCanBeMonitored)
2260	{
2261	# ifdef PGMPOOL_WITH_MONITORING
2262	rc = pgmPoolMonitorInsert(pPool, pPage);
2263	AssertRC(rc);
2264	}
2265	# endif
2266	#endif /* PGMPOOL_WITH_MONITORING */
2267	return rc;
2268	}
2269
2270
2271	# ifdef PGMPOOL_WITH_CACHE /* (only used when the cache is enabled.) */
2272	/**
2273	* Adds a user reference to a page.
2274	*
2275	* This will move the page to the head of the
2276	*
2277	* @returns VBox status code.
2278	* @retval VINF_SUCCESS if successfully added.
2279	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2280	* @param pPool The pool.
2281	* @param pPage The cached page.
2282	* @param iUser The user index.
2283	* @param iUserTable The user table.
2284	*/
2285	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2286	{
2287	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2288
2289	Log3(("pgmPoolTrackAddUser GCPhys = %RGp iUser %x iUserTable %x\n", pPage->GCPhys, iUser, iUserTable));
2290
2291	# ifdef VBOX_STRICT
2292	/*
2293	* Check that the entry doesn't already exists. We only allow multiple users of top-level paging structures (SHW_POOL_ROOT_IDX).
2294	*/
2295	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2296	{
2297	uint16_t i = pPage->iUserHead;
2298	do
2299	{
2300	Assert(i < pPool->cMaxUsers);
2301	AssertMsg(iUser != PGMPOOL_IDX_PD \|\| iUser != PGMPOOL_IDX_PDPT \|\| iUser != PGMPOOL_IDX_NESTED_ROOT \|\| iUser != PGMPOOL_IDX_AMD64_CR3 \|\|
2302	paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2303	i = paUsers[i].iNext;
2304	} while (i != NIL_PGMPOOL_USER_INDEX);
2305	}
2306	# endif
2307
2308	/*
2309	* Allocate a user node.
2310	*/
2311	uint16_t i = pPool->iUserFreeHead;
2312	if (i == NIL_PGMPOOL_USER_INDEX)
2313	{
2314	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2315	if (RT_FAILURE(rc))
2316	return rc;
2317	i = pPool->iUserFreeHead;
2318	}
2319	pPool->iUserFreeHead = paUsers[i].iNext;
2320
2321	/*
2322	* Initialize the user node and insert it.
2323	*/
2324	paUsers[i].iNext = pPage->iUserHead;
2325	paUsers[i].iUser = iUser;
2326	paUsers[i].iUserTable = iUserTable;
2327	pPage->iUserHead = i;
2328
2329	# ifdef PGMPOOL_WITH_CACHE
2330	/*
2331	* Tell the cache to update its replacement stats for this page.
2332	*/
2333	pgmPoolCacheUsed(pPool, pPage);
2334	# endif
2335	return VINF_SUCCESS;
2336	}
2337	# endif /* PGMPOOL_WITH_CACHE */
2338
2339
2340	/**
2341	* Frees a user record associated with a page.
2342	*
2343	* This does not clear the entry in the user table, it simply replaces the
2344	* user record to the chain of free records.
2345	*
2346	* @param pPool The pool.
2347	* @param HCPhys The HC physical address of the shadow page.
2348	* @param iUser The shadow page pool index of the user table.
2349	* @param iUserTable The index into the user table (shadowed).
2350	*/
2351	static void pgmPoolTrackFreeUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2352	{
2353	/*
2354	* Unlink and free the specified user entry.
2355	*/
2356	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2357
2358	Log3(("pgmPoolTrackFreeUser %RGp %x %x\n", pPage->GCPhys, iUser, iUserTable));
2359	/* Special: For PAE and 32-bit paging, there is usually no more than one user. */
2360	uint16_t i = pPage->iUserHead;
2361	if ( i != NIL_PGMPOOL_USER_INDEX
2362	&& paUsers[i].iUser == iUser
2363	&& paUsers[i].iUserTable == iUserTable)
2364	{
2365	pPage->iUserHead = paUsers[i].iNext;
2366
2367	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2368	paUsers[i].iNext = pPool->iUserFreeHead;
2369	pPool->iUserFreeHead = i;
2370	return;
2371	}
2372
2373	/* General: Linear search. */
2374	uint16_t iPrev = NIL_PGMPOOL_USER_INDEX;
2375	while (i != NIL_PGMPOOL_USER_INDEX)
2376	{
2377	if ( paUsers[i].iUser == iUser
2378	&& paUsers[i].iUserTable == iUserTable)
2379	{
2380	if (iPrev != NIL_PGMPOOL_USER_INDEX)
2381	paUsers[iPrev].iNext = paUsers[i].iNext;
2382	else
2383	pPage->iUserHead = paUsers[i].iNext;
2384
2385	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2386	paUsers[i].iNext = pPool->iUserFreeHead;
2387	pPool->iUserFreeHead = i;
2388	return;
2389	}
2390	iPrev = i;
2391	i = paUsers[i].iNext;
2392	}
2393
2394	/* Fatal: didn't find it */
2395	AssertFatalMsgFailed(("Didn't find the user entry! iUser=%#x iUserTable=%#x GCPhys=%RGp\n",
2396	iUser, iUserTable, pPage->GCPhys));
2397	}
2398
2399
2400	/**
2401	* Gets the entry size of a shadow table.
2402	*
2403	* @param enmKind The kind of page.
2404	*
2405	* @returns The size of the entry in bytes. That is, 4 or 8.
2406	* @returns If the kind is not for a table, an assertion is raised and 0 is
2407	* returned.
2408	*/
2409	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind)
2410	{
2411	switch (enmKind)
2412	{
2413	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2414	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2415	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2416	case PGMPOOLKIND_32BIT_PD:
2417	case PGMPOOLKIND_32BIT_PD_PHYS:
2418	return 4;
2419
2420	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2421	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2422	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2423	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2424	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2425	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2426	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2427	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2428	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2429	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2430	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2431	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2432	case PGMPOOLKIND_64BIT_PML4:
2433	case PGMPOOLKIND_PAE_PDPT:
2434	case PGMPOOLKIND_ROOT_NESTED:
2435	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2436	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2437	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2438	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2439	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2440	case PGMPOOLKIND_PAE_PD_PHYS:
2441	case PGMPOOLKIND_PAE_PDPT_PHYS:
2442	return 8;
2443
2444	default:
2445	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2446	}
2447	}
2448
2449
2450	/**
2451	* Gets the entry size of a guest table.
2452	*
2453	* @param enmKind The kind of page.
2454	*
2455	* @returns The size of the entry in bytes. That is, 0, 4 or 8.
2456	* @returns If the kind is not for a table, an assertion is raised and 0 is
2457	* returned.
2458	*/
2459	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind)
2460	{
2461	switch (enmKind)
2462	{
2463	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2464	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2465	case PGMPOOLKIND_32BIT_PD:
2466	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2467	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2468	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2469	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2470	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2471	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2472	return 4;
2473
2474	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2475	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2476	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2477	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2478	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2479	case PGMPOOLKIND_64BIT_PML4:
2480	case PGMPOOLKIND_PAE_PDPT:
2481	return 8;
2482
2483	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2484	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2485	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2486	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2487	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2488	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2489	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2490	case PGMPOOLKIND_ROOT_NESTED:
2491	case PGMPOOLKIND_PAE_PD_PHYS:
2492	case PGMPOOLKIND_PAE_PDPT_PHYS:
2493	case PGMPOOLKIND_32BIT_PD_PHYS:
2494	/** @todo can we return 0? (nobody is calling this...) */
2495	AssertFailed();
2496	return 0;
2497
2498	default:
2499	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2500	}
2501	}
2502
2503	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2504
2505	/**
2506	* Scans one shadow page table for mappings of a physical page.
2507	*
2508	* @param pVM The VM handle.
2509	* @param pPhysPage The guest page in question.
2510	* @param iShw The shadow page table.
2511	* @param cRefs The number of references made in that PT.
2512	*/
2513	static void pgmPoolTrackFlushGCPhysPTInt(PVM pVM, PCPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2514	{
2515	LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%R[pgmpage] iShw=%d cRefs=%d\n", pPhysPage, iShw, cRefs));
2516	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2517
2518	/*
2519	* Assert sanity.
2520	*/
2521	Assert(cRefs == 1);
2522	AssertFatalMsg(iShw < pPool->cCurPages && iShw != NIL_PGMPOOL_IDX, ("iShw=%d\n", iShw));
2523	PPGMPOOLPAGE pPage = &pPool->aPages[iShw];
2524
2525	/*
2526	* Then, clear the actual mappings to the page in the shadow PT.
2527	*/
2528	switch (pPage->enmKind)
2529	{
2530	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2531	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2532	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2533	{
2534	const uint32_t u32 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2535	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2536	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2537	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2538	{
2539	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX32 cRefs=%#x\n", i, pPT->a[i], cRefs));
2540	pPT->a[i].u = 0;
2541	cRefs--;
2542	if (!cRefs)
2543	return;
2544	}
2545	#ifdef LOG_ENABLED
2546	Log(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2547	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2548	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2549	{
2550	Log(("i=%d cRefs=%d\n", i, cRefs--));
2551	}
2552	#endif
2553	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2554	break;
2555	}
2556
2557	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2558	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2559	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2560	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2561	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2562	{
2563	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2564	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2565	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2566	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2567	{
2568	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2569	pPT->a[i].u = 0;
2570	cRefs--;
2571	if (!cRefs)
2572	return;
2573	}
2574	#ifdef LOG_ENABLED
2575	Log(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2576	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2577	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2578	{
2579	Log(("i=%d cRefs=%d\n", i, cRefs--));
2580	}
2581	#endif
2582	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d u64=%RX64\n", cRefs, pPage->iFirstPresent, pPage->cPresent, u64));
2583	break;
2584	}
2585
2586	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2587	{
2588	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2589	PEPTPT pPT = (PEPTPT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2590	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2591	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2592	{
2593	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2594	pPT->a[i].u = 0;
2595	cRefs--;
2596	if (!cRefs)
2597	return;
2598	}
2599	#ifdef LOG_ENABLED
2600	Log(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2601	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2602	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2603	{
2604	Log(("i=%d cRefs=%d\n", i, cRefs--));
2605	}
2606	#endif
2607	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2608	break;
2609	}
2610
2611	default:
2612	AssertFatalMsgFailed(("enmKind=%d iShw=%d\n", pPage->enmKind, iShw));
2613	}
2614	}
2615
2616
2617	/**
2618	* Scans one shadow page table for mappings of a physical page.
2619	*
2620	* @param pVM The VM handle.
2621	* @param pPhysPage The guest page in question.
2622	* @param iShw The shadow page table.
2623	* @param cRefs The number of references made in that PT.
2624	*/
2625	void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2626	{
2627	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
2628	LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%R[pgmpage] iShw=%d cRefs=%d\n", pPhysPage, iShw, cRefs));
2629	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPT, f);
2630	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, iShw, cRefs);
2631	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2632	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPT, f);
2633	}
2634
2635
2636	/**
2637	* Flushes a list of shadow page tables mapping the same physical page.
2638	*
2639	* @param pVM The VM handle.
2640	* @param pPhysPage The guest page in question.
2641	* @param iPhysExt The physical cross reference extent list to flush.
2642	*/
2643	void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, uint16_t iPhysExt)
2644	{
2645	Assert(PGMIsLockOwner(pVM));
2646	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2647	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTs, f);
2648	LogFlow(("pgmPoolTrackFlushGCPhysPTs: pPhysPage=%R[pgmpage] iPhysExt\n", pPhysPage, iPhysExt));
2649
2650	const uint16_t iPhysExtStart = iPhysExt;
2651	PPGMPOOLPHYSEXT pPhysExt;
2652	do
2653	{
2654	Assert(iPhysExt < pPool->cMaxPhysExts);
2655	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2656	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
2657	if (pPhysExt->aidx[i] != NIL_PGMPOOL_IDX)
2658	{
2659	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, pPhysExt->aidx[i], 1);
2660	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
2661	}
2662
2663	/* next */
2664	iPhysExt = pPhysExt->iNext;
2665	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
2666
2667	/* insert the list into the free list and clear the ram range entry. */
2668	pPhysExt->iNext = pPool->iPhysExtFreeHead;
2669	pPool->iPhysExtFreeHead = iPhysExtStart;
2670	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2671
2672	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTs, f);
2673	}
2674
2675	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
2676
2677	/**
2678	* Flushes all shadow page table mappings of the given guest page.
2679	*
2680	* This is typically called when the host page backing the guest one has been
2681	* replaced or when the page protection was changed due to an access handler.
2682	*
2683	* @returns VBox status code.
2684	* @retval VINF_SUCCESS if all references has been successfully cleared.
2685	* @retval VINF_PGM_SYNC_CR3 if we're better off with a CR3 sync and a page
2686	* pool cleaning. FF and sync flags are set.
2687	*
2688	* @param pVM The VM handle.
2689	* @param pPhysPage The guest page in question.
2690	* @param pfFlushTLBs This is set to @a true if the shadow TLBs should be
2691	* flushed, it is NOT touched if this isn't necessary.
2692	* The caller MUST initialized this to @a false.
2693	*/
2694	int pgmPoolTrackFlushGCPhys(PVM pVM, PPGMPAGE pPhysPage, bool *pfFlushTLBs)
2695	{
2696	PVMCPU pVCpu = VMMGetCpu(pVM);
2697	pgmLock(pVM);
2698	int rc = VINF_SUCCESS;
2699	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2700	const uint16_t u16 = PGM_PAGE_GET_TRACKING(pPhysPage);
2701	if (u16)
2702	{
2703	/*
2704	* The zero page is currently screwing up the tracking and we'll
2705	* have to flush the whole shebang. Unless VBOX_WITH_NEW_LAZY_PAGE_ALLOC
2706	* is defined, zero pages won't normally be mapped. Some kind of solution
2707	* will be needed for this problem of course, but it will have to wait...
2708	*/
2709	if (PGM_PAGE_IS_ZERO(pPhysPage))
2710	rc = VINF_PGM_GCPHYS_ALIASED;
2711	else
2712	{
2713	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2714	/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow and
2715	pgmPoolTrackFlushGCPhysPTs will/may kill the pool otherwise. */
2716	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
2717	# endif
2718
2719	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
2720	pgmPoolTrackFlushGCPhysPT(pVM,
2721	pPhysPage,
2722	PGMPOOL_TD_GET_IDX(u16),
2723	PGMPOOL_TD_GET_CREFS(u16));
2724	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
2725	pgmPoolTrackFlushGCPhysPTs(pVM, pPhysPage, PGMPOOL_TD_GET_IDX(u16));
2726	else
2727	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
2728	*pfFlushTLBs = true;
2729
2730	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2731	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
2732	# endif
2733	}
2734	}
2735
2736	#elif defined(PGMPOOL_WITH_CACHE)
2737	if (PGM_PAGE_IS_ZERO(pPhysPage))
2738	rc = VINF_PGM_GCPHYS_ALIASED;
2739	else
2740	{
2741	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2742	/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow kill the pool otherwise. */
2743	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
2744	# endif
2745	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
2746	if (rc == VINF_SUCCESS)
2747	*pfFlushTLBs = true;
2748	}
2749
2750	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2751	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
2752	# endif
2753
2754	#else
2755	rc = VINF_PGM_GCPHYS_ALIASED;
2756	#endif
2757
2758	if (rc == VINF_PGM_GCPHYS_ALIASED)
2759	{
2760	pVCpu->pgm.s.fSyncFlags \|= PGM_SYNC_CLEAR_PGM_POOL;
2761	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
2762	rc = VINF_PGM_SYNC_CR3;
2763	}
2764	pgmUnlock(pVM);
2765	return rc;
2766	}
2767
2768
2769	/**
2770	* Scans all shadow page tables for mappings of a physical page.
2771	*
2772	* This may be slow, but it's most likely more efficient than cleaning
2773	* out the entire page pool / cache.
2774	*
2775	* @returns VBox status code.
2776	* @retval VINF_SUCCESS if all references has been successfully cleared.
2777	* @retval VINF_PGM_GCPHYS_ALIASED if we're better off with a CR3 sync and
2778	* a page pool cleaning.
2779	*
2780	* @param pVM The VM handle.
2781	* @param pPhysPage The guest page in question.
2782	*/
2783	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage)
2784	{
2785	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2786	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2787	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: cUsedPages=%d cPresent=%d pPhysPage=%R[pgmpage]\n",
2788	pPool->cUsedPages, pPool->cPresent, pPhysPage));
2789
2790	#if 1
2791	/*
2792	* There is a limit to what makes sense.
2793	*/
2794	if (pPool->cPresent > 1024)
2795	{
2796	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
2797	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2798	return VINF_PGM_GCPHYS_ALIASED;
2799	}
2800	#endif
2801
2802	/*
2803	* Iterate all the pages until we've encountered all that in use.
2804	* This is simple but not quite optimal solution.
2805	*/
2806	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2807	const uint32_t u32 = u64;
2808	unsigned cLeft = pPool->cUsedPages;
2809	unsigned iPage = pPool->cCurPages;
2810	while (--iPage >= PGMPOOL_IDX_FIRST)
2811	{
2812	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2813	if (pPage->GCPhys != NIL_RTGCPHYS)
2814	{
2815	switch (pPage->enmKind)
2816	{
2817	/*
2818	* We only care about shadow page tables.
2819	*/
2820	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2821	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2822	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2823	{
2824	unsigned cPresent = pPage->cPresent;
2825	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2826	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2827	if (pPT->a[i].n.u1Present)
2828	{
2829	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2830	{
2831	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX32\n", iPage, i, pPT->a[i]));
2832	pPT->a[i].u = 0;
2833	}
2834	if (!--cPresent)
2835	break;
2836	}
2837	break;
2838	}
2839
2840	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2841	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2842	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2843	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2844	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2845	{
2846	unsigned cPresent = pPage->cPresent;
2847	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2848	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2849	if (pPT->a[i].n.u1Present)
2850	{
2851	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2852	{
2853	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
2854	pPT->a[i].u = 0;
2855	}
2856	if (!--cPresent)
2857	break;
2858	}
2859	break;
2860	}
2861	}
2862	if (!--cLeft)
2863	break;
2864	}
2865	}
2866
2867	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
2868	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2869	return VINF_SUCCESS;
2870	}
2871
2872
2873	/**
2874	* Clears the user entry in a user table.
2875	*
2876	* This is used to remove all references to a page when flushing it.
2877	*/
2878	static void pgmPoolTrackClearPageUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PCPGMPOOLUSER pUser)
2879	{
2880	Assert(pUser->iUser != NIL_PGMPOOL_IDX);
2881	Assert(pUser->iUser < pPool->cCurPages);
2882	uint32_t iUserTable = pUser->iUserTable;
2883
2884	/*
2885	* Map the user page.
2886	*/
2887	PPGMPOOLPAGE pUserPage = &pPool->aPages[pUser->iUser];
2888	union
2889	{
2890	uint64_t *pau64;
2891	uint32_t *pau32;
2892	} u;
2893	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pUserPage);
2894
2895	LogFlow(("pgmPoolTrackClearPageUser: clear %x in %s (%RGp) (flushing %s)\n", iUserTable, pgmPoolPoolKindToStr(pUserPage->enmKind), pUserPage->Core.Key, pgmPoolPoolKindToStr(pPage->enmKind)));
2896
2897	/* Safety precaution in case we change the paging for other modes too in the future. */
2898	Assert(!pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage));
2899
2900	#ifdef VBOX_STRICT
2901	/*
2902	* Some sanity checks.
2903	*/
2904	switch (pUserPage->enmKind)
2905	{
2906	case PGMPOOLKIND_32BIT_PD:
2907	case PGMPOOLKIND_32BIT_PD_PHYS:
2908	Assert(iUserTable < X86_PG_ENTRIES);
2909	break;
2910	case PGMPOOLKIND_PAE_PDPT:
2911	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2912	case PGMPOOLKIND_PAE_PDPT_PHYS:
2913	Assert(iUserTable < 4);
2914	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2915	break;
2916	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2917	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2918	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2919	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2920	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2921	case PGMPOOLKIND_PAE_PD_PHYS:
2922	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2923	break;
2924	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2925	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2926	Assert(!(u.pau64[iUserTable] & PGM_PDFLAGS_MAPPING));
2927	break;
2928	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2929	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2930	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2931	break;
2932	case PGMPOOLKIND_64BIT_PML4:
2933	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2934	/* GCPhys >> PAGE_SHIFT is the index here */
2935	break;
2936	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2937	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2938	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2939	break;
2940
2941	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2942	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2943	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2944	break;
2945
2946	case PGMPOOLKIND_ROOT_NESTED:
2947	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2948	break;
2949
2950	default:
2951	AssertMsgFailed(("enmKind=%d\n", pUserPage->enmKind));
2952	break;
2953	}
2954	#endif /* VBOX_STRICT */
2955
2956	/*
2957	* Clear the entry in the user page.
2958	*/
2959	switch (pUserPage->enmKind)
2960	{
2961	/* 32-bit entries */
2962	case PGMPOOLKIND_32BIT_PD:
2963	case PGMPOOLKIND_32BIT_PD_PHYS:
2964	u.pau32[iUserTable] = 0;
2965	break;
2966
2967	/* 64-bit entries */
2968	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2969	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2970	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2971	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2972	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2973	#if defined(IN_RC)
2974	/* 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
2975	* non-present PDPT will continue to cause page faults.
2976	*/
2977	ASMReloadCR3();
2978	#endif
2979	/* no break */
2980	case PGMPOOLKIND_PAE_PD_PHYS:
2981	case PGMPOOLKIND_PAE_PDPT_PHYS:
2982	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2983	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2984	case PGMPOOLKIND_64BIT_PML4:
2985	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2986	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2987	case PGMPOOLKIND_PAE_PDPT:
2988	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2989	case PGMPOOLKIND_ROOT_NESTED:
2990	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2991	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2992	u.pau64[iUserTable] = 0;
2993	break;
2994
2995	default:
2996	AssertFatalMsgFailed(("enmKind=%d iUser=%#x iUserTable=%#x\n", pUserPage->enmKind, pUser->iUser, pUser->iUserTable));
2997	}
2998	}
2999
3000
3001	/**
3002	* Clears all users of a page.
3003	*/
3004	static void pgmPoolTrackClearPageUsers(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3005	{
3006	/*
3007	* Free all the user records.
3008	*/
3009	LogFlow(("pgmPoolTrackClearPageUsers %RGp\n", pPage->GCPhys));
3010
3011	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
3012	uint16_t i = pPage->iUserHead;
3013	while (i != NIL_PGMPOOL_USER_INDEX)
3014	{
3015	/* Clear enter in user table. */
3016	pgmPoolTrackClearPageUser(pPool, pPage, &paUsers[i]);
3017
3018	/* Free it. */
3019	const uint16_t iNext = paUsers[i].iNext;
3020	paUsers[i].iUser = NIL_PGMPOOL_IDX;
3021	paUsers[i].iNext = pPool->iUserFreeHead;
3022	pPool->iUserFreeHead = i;
3023
3024	/* Next. */
3025	i = iNext;
3026	}
3027	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
3028	}
3029
3030	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3031
3032	/**
3033	* Allocates a new physical cross reference extent.
3034	*
3035	* @returns Pointer to the allocated extent on success. NULL if we're out of them.
3036	* @param pVM The VM handle.
3037	* @param piPhysExt Where to store the phys ext index.
3038	*/
3039	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt)
3040	{
3041	Assert(PGMIsLockOwner(pVM));
3042	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3043	uint16_t iPhysExt = pPool->iPhysExtFreeHead;
3044	if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
3045	{
3046	STAM_COUNTER_INC(&pPool->StamTrackPhysExtAllocFailures);
3047	return NULL;
3048	}
3049	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3050	pPool->iPhysExtFreeHead = pPhysExt->iNext;
3051	pPhysExt->iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
3052	*piPhysExt = iPhysExt;
3053	return pPhysExt;
3054	}
3055
3056
3057	/**
3058	* Frees a physical cross reference extent.
3059	*
3060	* @param pVM The VM handle.
3061	* @param iPhysExt The extent to free.
3062	*/
3063	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt)
3064	{
3065	Assert(PGMIsLockOwner(pVM));
3066	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3067	Assert(iPhysExt < pPool->cMaxPhysExts);
3068	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3069	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3070	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3071	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3072	pPool->iPhysExtFreeHead = iPhysExt;
3073	}
3074
3075
3076	/**
3077	* Frees a physical cross reference extent.
3078	*
3079	* @param pVM The VM handle.
3080	* @param iPhysExt The extent to free.
3081	*/
3082	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt)
3083	{
3084	Assert(PGMIsLockOwner(pVM));
3085	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3086
3087	const uint16_t iPhysExtStart = iPhysExt;
3088	PPGMPOOLPHYSEXT pPhysExt;
3089	do
3090	{
3091	Assert(iPhysExt < pPool->cMaxPhysExts);
3092	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3093	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3094	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3095
3096	/* next */
3097	iPhysExt = pPhysExt->iNext;
3098	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3099
3100	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3101	pPool->iPhysExtFreeHead = iPhysExtStart;
3102	}
3103
3104
3105	/**
3106	* Insert a reference into a list of physical cross reference extents.
3107	*
3108	* @returns The new tracking data for PGMPAGE.
3109	*
3110	* @param pVM The VM handle.
3111	* @param iPhysExt The physical extent index of the list head.
3112	* @param iShwPT The shadow page table index.
3113	*
3114	*/
3115	static uint16_t pgmPoolTrackPhysExtInsert(PVM pVM, uint16_t iPhysExt, uint16_t iShwPT)
3116	{
3117	Assert(PGMIsLockOwner(pVM));
3118	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3119	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3120
3121	/* special common case. */
3122	if (paPhysExts[iPhysExt].aidx[2] == NIL_PGMPOOL_IDX)
3123	{
3124	paPhysExts[iPhysExt].aidx[2] = iShwPT;
3125	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
3126	LogFlow(("pgmPoolTrackPhysExtInsert: %d:{,,%d}\n", iPhysExt, iShwPT));
3127	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3128	}
3129
3130	/* general treatment. */
3131	const uint16_t iPhysExtStart = iPhysExt;
3132	unsigned cMax = 15;
3133	for (;;)
3134	{
3135	Assert(iPhysExt < pPool->cMaxPhysExts);
3136	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3137	if (paPhysExts[iPhysExt].aidx[i] == NIL_PGMPOOL_IDX)
3138	{
3139	paPhysExts[iPhysExt].aidx[i] = iShwPT;
3140	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
3141	LogFlow(("pgmPoolTrackPhysExtInsert: %d:{%d} i=%d cMax=%d\n", iPhysExt, iShwPT, i, cMax));
3142	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtStart);
3143	}
3144	if (!--cMax)
3145	{
3146	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
3147	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
3148	LogFlow(("pgmPoolTrackPhysExtInsert: overflow (1) iShwPT=%d\n", iShwPT));
3149	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3150	}
3151	}
3152
3153	/* add another extent to the list. */
3154	PPGMPOOLPHYSEXT pNew = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3155	if (!pNew)
3156	{
3157	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
3158	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
3159	LogFlow(("pgmPoolTrackPhysExtInsert: pgmPoolTrackPhysExtAlloc failed iShwPT=%d\n", iShwPT));
3160	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3161	}
3162	pNew->iNext = iPhysExtStart;
3163	pNew->aidx[0] = iShwPT;
3164	LogFlow(("pgmPoolTrackPhysExtInsert: added new extent %d:{%d}->%d\n", iPhysExt, iShwPT, iPhysExtStart));
3165	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3166	}
3167
3168
3169	/**
3170	* Add a reference to guest physical page where extents are in use.
3171	*
3172	* @returns The new tracking data for PGMPAGE.
3173	*
3174	* @param pVM The VM handle.
3175	* @param u16 The ram range flags (top 16-bits).
3176	* @param iShwPT The shadow page table index.
3177	*/
3178	uint16_t pgmPoolTrackPhysExtAddref(PVM pVM, uint16_t u16, uint16_t iShwPT)
3179	{
3180	pgmLock(pVM);
3181	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
3182	{
3183	/*
3184	* Convert to extent list.
3185	*/
3186	Assert(PGMPOOL_TD_GET_CREFS(u16) == 1);
3187	uint16_t iPhysExt;
3188	PPGMPOOLPHYSEXT pPhysExt = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3189	if (pPhysExt)
3190	{
3191	LogFlow(("pgmPoolTrackPhysExtAddref: new extent: %d:{%d, %d}\n", iPhysExt, PGMPOOL_TD_GET_IDX(u16), iShwPT));
3192	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliased);
3193	pPhysExt->aidx[0] = PGMPOOL_TD_GET_IDX(u16);
3194	pPhysExt->aidx[1] = iShwPT;
3195	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3196	}
3197	else
3198	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3199	}
3200	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
3201	{
3202	/*
3203	* Insert into the extent list.
3204	*/
3205	u16 = pgmPoolTrackPhysExtInsert(pVM, PGMPOOL_TD_GET_IDX(u16), iShwPT);
3206	}
3207	else
3208	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedLots);
3209	pgmUnlock(pVM);
3210	return u16;
3211	}
3212
3213
3214	/**
3215	* Clear references to guest physical memory.
3216	*
3217	* @param pPool The pool.
3218	* @param pPage The page.
3219	* @param pPhysPage Pointer to the aPages entry in the ram range.
3220	*/
3221	void pgmPoolTrackPhysExtDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMPAGE pPhysPage)
3222	{
3223	const unsigned cRefs = PGM_PAGE_GET_TD_CREFS(pPhysPage);
3224	AssertFatalMsg(cRefs == PGMPOOL_TD_CREFS_PHYSEXT, ("cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
3225
3226	uint16_t iPhysExt = PGM_PAGE_GET_TD_IDX(pPhysPage);
3227	if (iPhysExt != PGMPOOL_TD_IDX_OVERFLOWED)
3228	{
3229	PVM pVM = pPool->CTX_SUFF(pVM);
3230	pgmLock(pVM);
3231
3232	uint16_t iPhysExtPrev = NIL_PGMPOOL_PHYSEXT_INDEX;
3233	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3234	do
3235	{
3236	Assert(iPhysExt < pPool->cMaxPhysExts);
3237
3238	/*
3239	* Look for the shadow page and check if it's all freed.
3240	*/
3241	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3242	{
3243	if (paPhysExts[iPhysExt].aidx[i] == pPage->idx)
3244	{
3245	paPhysExts[iPhysExt].aidx[i] = NIL_PGMPOOL_IDX;
3246
3247	for (i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3248	if (paPhysExts[iPhysExt].aidx[i] != NIL_PGMPOOL_IDX)
3249	{
3250	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
3251	pgmUnlock(pVM);
3252	return;
3253	}
3254
3255	/* we can free the node. */
3256	const uint16_t iPhysExtNext = paPhysExts[iPhysExt].iNext;
3257	if ( iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX
3258	&& iPhysExtNext == NIL_PGMPOOL_PHYSEXT_INDEX)
3259	{
3260	/* lonely node */
3261	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3262	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d lonely\n", pPhysPage, pPage->idx));
3263	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
3264	}
3265	else if (iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX)
3266	{
3267	/* head */
3268	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d head\n", pPhysPage, pPage->idx));
3269	PGM_PAGE_SET_TRACKING(pPhysPage, PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtNext));
3270	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3271	}
3272	else
3273	{
3274	/* in list */
3275	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
3276	paPhysExts[iPhysExtPrev].iNext = iPhysExtNext;
3277	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3278	}
3279	iPhysExt = iPhysExtNext;
3280	pgmUnlock(pVM);
3281	return;
3282	}
3283	}
3284
3285	/* next */
3286	iPhysExtPrev = iPhysExt;
3287	iPhysExt = paPhysExts[iPhysExt].iNext;
3288	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3289
3290	pgmUnlock(pVM);
3291	AssertFatalMsgFailed(("not-found! cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
3292	}
3293	else /* nothing to do */
3294	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage]\n", pPhysPage));
3295	}
3296
3297
3298	/**
3299	* Clear references to guest physical memory.
3300	*
3301	* This is the same as pgmPoolTracDerefGCPhys except that the guest physical address
3302	* is assumed to be correct, so the linear search can be skipped and we can assert
3303	* at an earlier point.
3304	*
3305	* @param pPool The pool.
3306	* @param pPage The page.
3307	* @param HCPhys The host physical address corresponding to the guest page.
3308	* @param GCPhys The guest physical address corresponding to HCPhys.
3309	*/
3310	static void pgmPoolTracDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhys)
3311	{
3312	/*
3313	* Walk range list.
3314	*/
3315	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3316	while (pRam)
3317	{
3318	RTGCPHYS off = GCPhys - pRam->GCPhys;
3319	if (off < pRam->cb)
3320	{
3321	/* does it match? */
3322	const unsigned iPage = off >> PAGE_SHIFT;
3323	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3324	#ifdef LOG_ENABLED
3325	RTHCPHYS HCPhysPage = PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]);
3326	Log2(("pgmPoolTracDerefGCPhys %RHp vs %RHp\n", HCPhysPage, HCPhys));
3327	#endif
3328	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3329	{
3330	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3331	return;
3332	}
3333	break;
3334	}
3335	pRam = pRam->CTX_SUFF(pNext);
3336	}
3337	AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp\n", HCPhys, GCPhys));
3338	}
3339
3340
3341	/**
3342	* Clear references to guest physical memory.
3343	*
3344	* @param pPool The pool.
3345	* @param pPage The page.
3346	* @param HCPhys The host physical address corresponding to the guest page.
3347	* @param GCPhysHint The guest physical address which may corresponding to HCPhys.
3348	*/
3349	static void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint)
3350	{
3351	Log4(("pgmPoolTracDerefGCPhysHint %RHp %RGp\n", HCPhys, GCPhysHint));
3352
3353	/*
3354	* Walk range list.
3355	*/
3356	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3357	while (pRam)
3358	{
3359	RTGCPHYS off = GCPhysHint - pRam->GCPhys;
3360	if (off < pRam->cb)
3361	{
3362	/* does it match? */
3363	const unsigned iPage = off >> PAGE_SHIFT;
3364	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3365	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3366	{
3367	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3368	return;
3369	}
3370	break;
3371	}
3372	pRam = pRam->CTX_SUFF(pNext);
3373	}
3374
3375	/*
3376	* Damn, the hint didn't work. We'll have to do an expensive linear search.
3377	*/
3378	STAM_COUNTER_INC(&pPool->StatTrackLinearRamSearches);
3379	pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3380	while (pRam)
3381	{
3382	unsigned iPage = pRam->cb >> PAGE_SHIFT;
3383	while (iPage-- > 0)
3384	{
3385	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3386	{
3387	Log4(("pgmPoolTracDerefGCPhysHint: Linear HCPhys=%RHp GCPhysHint=%RGp GCPhysReal=%RGp\n",
3388	HCPhys, GCPhysHint, pRam->GCPhys + (iPage << PAGE_SHIFT)));
3389	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3390	return;
3391	}
3392	}
3393	pRam = pRam->CTX_SUFF(pNext);
3394	}
3395
3396	AssertFatalMsgFailed(("HCPhys=%RHp GCPhysHint=%RGp\n", HCPhys, GCPhysHint));
3397	}
3398
3399
3400	/**
3401	* Clear references to guest physical memory in a 32-bit / 32-bit page table.
3402	*
3403	* @param pPool The pool.
3404	* @param pPage The page.
3405	* @param pShwPT The shadow page table (mapping of the page).
3406	* @param pGstPT The guest page table.
3407	*/
3408	DECLINLINE(void) pgmPoolTrackDerefPT32Bit32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT, PCX86PT pGstPT)
3409	{
3410	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
3411	if (pShwPT->a[i].n.u1Present)
3412	{
3413	Log4(("pgmPoolTrackDerefPT32Bit32Bit: i=%d pte=%RX32 hint=%RX32\n",
3414	i, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3415	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3416	if (!--pPage->cPresent)
3417	break;
3418	}
3419	}
3420
3421
3422	/**
3423	* Clear references to guest physical memory in a PAE / 32-bit page table.
3424	*
3425	* @param pPool The pool.
3426	* @param pPage The page.
3427	* @param pShwPT The shadow page table (mapping of the page).
3428	* @param pGstPT The guest page table (just a half one).
3429	*/
3430	DECLINLINE(void) pgmPoolTrackDerefPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PT pGstPT)
3431	{
3432	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3433	if (pShwPT->a[i].n.u1Present)
3434	{
3435	Log4(("pgmPoolTrackDerefPTPae32Bit: i=%d pte=%RX64 hint=%RX32\n",
3436	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3437	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3438	}
3439	}
3440
3441
3442	/**
3443	* Clear references to guest physical memory in a PAE / PAE page table.
3444	*
3445	* @param pPool The pool.
3446	* @param pPage The page.
3447	* @param pShwPT The shadow page table (mapping of the page).
3448	* @param pGstPT The guest page table.
3449	*/
3450	DECLINLINE(void) pgmPoolTrackDerefPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PTPAE pGstPT)
3451	{
3452	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3453	if (pShwPT->a[i].n.u1Present)
3454	{
3455	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX32 hint=%RX32\n",
3456	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
3457	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK);
3458	}
3459	}
3460
3461
3462	/**
3463	* Clear references to guest physical memory in a 32-bit / 4MB page table.
3464	*
3465	* @param pPool The pool.
3466	* @param pPage The page.
3467	* @param pShwPT The shadow page table (mapping of the page).
3468	*/
3469	DECLINLINE(void) pgmPoolTrackDerefPT32Bit4MB(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT)
3470	{
3471	RTGCPHYS GCPhys = pPage->GCPhys;
3472	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3473	if (pShwPT->a[i].n.u1Present)
3474	{
3475	Log4(("pgmPoolTrackDerefPT32Bit4MB: i=%d pte=%RX32 GCPhys=%RGp\n",
3476	i, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys));
3477	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys);
3478	}
3479	}
3480
3481
3482	/**
3483	* Clear references to guest physical memory in a PAE / 2/4MB page table.
3484	*
3485	* @param pPool The pool.
3486	* @param pPage The page.
3487	* @param pShwPT The shadow page table (mapping of the page).
3488	*/
3489	DECLINLINE(void) pgmPoolTrackDerefPTPaeBig(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT)
3490	{
3491	RTGCPHYS GCPhys = pPage->GCPhys;
3492	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3493	if (pShwPT->a[i].n.u1Present)
3494	{
3495	Log4(("pgmPoolTrackDerefPTPaeBig: i=%d pte=%RX64 hint=%RGp\n",
3496	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys));
3497	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys);
3498	}
3499	}
3500
3501	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
3502
3503
3504	/**
3505	* Clear references to shadowed pages in a 32 bits page directory.
3506	*
3507	* @param pPool The pool.
3508	* @param pPage The page.
3509	* @param pShwPD The shadow page directory (mapping of the page).
3510	*/
3511	DECLINLINE(void) pgmPoolTrackDerefPD(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PD pShwPD)
3512	{
3513	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3514	{
3515	if ( pShwPD->a[i].n.u1Present
3516	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
3517	)
3518	{
3519	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PG_MASK);
3520	if (pSubPage)
3521	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3522	else
3523	AssertFatalMsgFailed(("%x\n", pShwPD->a[i].u & X86_PDE_PG_MASK));
3524	}
3525	}
3526	}
3527
3528	/**
3529	* Clear references to shadowed pages in a PAE (legacy or 64 bits) page directory.
3530	*
3531	* @param pPool The pool.
3532	* @param pPage The page.
3533	* @param pShwPD The shadow page directory (mapping of the page).
3534	*/
3535	DECLINLINE(void) pgmPoolTrackDerefPDPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPAE pShwPD)
3536	{
3537	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3538	{
3539	if ( pShwPD->a[i].n.u1Present
3540	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
3541	)
3542	{
3543	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PAE_PG_MASK);
3544	if (pSubPage)
3545	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3546	else
3547	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & X86_PDE_PAE_PG_MASK));
3548	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3549	}
3550	}
3551	}
3552
3553	/**
3554	* Clear references to shadowed pages in a PAE page directory pointer table.
3555	*
3556	* @param pPool The pool.
3557	* @param pPage The page.
3558	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3559	*/
3560	DECLINLINE(void) pgmPoolTrackDerefPDPTPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
3561	{
3562	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
3563	{
3564	if ( pShwPDPT->a[i].n.u1Present
3565	&& !(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING)
3566	)
3567	{
3568	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
3569	if (pSubPage)
3570	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3571	else
3572	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
3573	}
3574	}
3575	}
3576
3577
3578	/**
3579	* Clear references to shadowed pages in a 64-bit page directory pointer table.
3580	*
3581	* @param pPool The pool.
3582	* @param pPage The page.
3583	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3584	*/
3585	DECLINLINE(void) pgmPoolTrackDerefPDPT64Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
3586	{
3587	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3588	{
3589	Assert(!(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING));
3590	if (pShwPDPT->a[i].n.u1Present)
3591	{
3592	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
3593	if (pSubPage)
3594	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3595	else
3596	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
3597	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3598	}
3599	}
3600	}
3601
3602
3603	/**
3604	* Clear references to shadowed pages in a 64-bit level 4 page table.
3605	*
3606	* @param pPool The pool.
3607	* @param pPage The page.
3608	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3609	*/
3610	DECLINLINE(void) pgmPoolTrackDerefPML464Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PML4 pShwPML4)
3611	{
3612	for (unsigned i = 0; i < RT_ELEMENTS(pShwPML4->a); i++)
3613	{
3614	if (pShwPML4->a[i].n.u1Present)
3615	{
3616	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPML4->a[i].u & X86_PDPE_PG_MASK);
3617	if (pSubPage)
3618	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3619	else
3620	AssertFatalMsgFailed(("%RX64\n", pShwPML4->a[i].u & X86_PML4E_PG_MASK));
3621	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3622	}
3623	}
3624	}
3625
3626
3627	/**
3628	* Clear references to shadowed pages in an EPT page table.
3629	*
3630	* @param pPool The pool.
3631	* @param pPage The page.
3632	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3633	*/
3634	DECLINLINE(void) pgmPoolTrackDerefPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT)
3635	{
3636	RTGCPHYS GCPhys = pPage->GCPhys;
3637	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3638	if (pShwPT->a[i].n.u1Present)
3639	{
3640	Log4(("pgmPoolTrackDerefPTEPT: i=%d pte=%RX64 GCPhys=%RX64\n",
3641	i, pShwPT->a[i].u & EPT_PTE_PG_MASK, pPage->GCPhys));
3642	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & EPT_PTE_PG_MASK, GCPhys);
3643	}
3644	}
3645
3646
3647	/**
3648	* Clear references to shadowed pages in an EPT page directory.
3649	*
3650	* @param pPool The pool.
3651	* @param pPage The page.
3652	* @param pShwPD The shadow page directory (mapping of the page).
3653	*/
3654	DECLINLINE(void) pgmPoolTrackDerefPDEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPD pShwPD)
3655	{
3656	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3657	{
3658	if (pShwPD->a[i].n.u1Present)
3659	{
3660	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & EPT_PDE_PG_MASK);
3661	if (pSubPage)
3662	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3663	else
3664	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & EPT_PDE_PG_MASK));
3665	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3666	}
3667	}
3668	}
3669
3670
3671	/**
3672	* Clear references to shadowed pages in an EPT page directory pointer table.
3673	*
3674	* @param pPool The pool.
3675	* @param pPage The page.
3676	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3677	*/
3678	DECLINLINE(void) pgmPoolTrackDerefPDPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPDPT pShwPDPT)
3679	{
3680	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3681	{
3682	if (pShwPDPT->a[i].n.u1Present)
3683	{
3684	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK);
3685	if (pSubPage)
3686	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3687	else
3688	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK));
3689	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3690	}
3691	}
3692	}
3693
3694
3695	/**
3696	* Clears all references made by this page.
3697	*
3698	* This includes other shadow pages and GC physical addresses.
3699	*
3700	* @param pPool The pool.
3701	* @param pPage The page.
3702	*/
3703	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3704	{
3705	/*
3706	* Map the shadow page and take action according to the page kind.
3707	*/
3708	void *pvShw = PGMPOOL_PAGE_2_LOCKED_PTR(pPool->CTX_SUFF(pVM), pPage);
3709	switch (pPage->enmKind)
3710	{
3711	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3712	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3713	{
3714	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3715	void *pvGst;
3716	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3717	pgmPoolTrackDerefPT32Bit32Bit(pPool, pPage, (PX86PT)pvShw, (PCX86PT)pvGst);
3718	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3719	break;
3720	}
3721
3722	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3723	{
3724	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3725	void *pvGst;
3726	int rc = PGM_GCPHYS_2_PTR_EX(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3727	pgmPoolTrackDerefPTPae32Bit(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PT)pvGst);
3728	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3729	break;
3730	}
3731
3732	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3733	{
3734	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3735	void *pvGst;
3736	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3737	pgmPoolTrackDerefPTPaePae(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PTPAE)pvGst);
3738	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3739	break;
3740	}
3741
3742	case PGMPOOLKIND_32BIT_PT_FOR_PHYS: /* treat it like a 4 MB page */
3743	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3744	{
3745	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3746	pgmPoolTrackDerefPT32Bit4MB(pPool, pPage, (PX86PT)pvShw);
3747	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3748	break;
3749	}
3750
3751	case PGMPOOLKIND_PAE_PT_FOR_PHYS: /* treat it like a 2 MB page */
3752	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3753	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3754	{
3755	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3756	pgmPoolTrackDerefPTPaeBig(pPool, pPage, (PX86PTPAE)pvShw);
3757	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3758	break;
3759	}
3760
3761	#else /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3762	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3763	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3764	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3765	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3766	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3767	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3768	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3769	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3770	break;
3771	#endif /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3772
3773	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3774	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3775	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3776	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3777	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3778	case PGMPOOLKIND_PAE_PD_PHYS:
3779	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3780	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3781	pgmPoolTrackDerefPDPae(pPool, pPage, (PX86PDPAE)pvShw);
3782	break;
3783
3784	case PGMPOOLKIND_32BIT_PD_PHYS:
3785	case PGMPOOLKIND_32BIT_PD:
3786	pgmPoolTrackDerefPD(pPool, pPage, (PX86PD)pvShw);
3787	break;
3788
3789	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
3790	case PGMPOOLKIND_PAE_PDPT:
3791	case PGMPOOLKIND_PAE_PDPT_PHYS:
3792	pgmPoolTrackDerefPDPTPae(pPool, pPage, (PX86PDPT)pvShw);
3793	break;
3794
3795	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3796	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3797	pgmPoolTrackDerefPDPT64Bit(pPool, pPage, (PX86PDPT)pvShw);
3798	break;
3799
3800	case PGMPOOLKIND_64BIT_PML4:
3801	pgmPoolTrackDerefPML464Bit(pPool, pPage, (PX86PML4)pvShw);
3802	break;
3803
3804	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3805	pgmPoolTrackDerefPTEPT(pPool, pPage, (PEPTPT)pvShw);
3806	break;
3807
3808	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3809	pgmPoolTrackDerefPDEPT(pPool, pPage, (PEPTPD)pvShw);
3810	break;
3811
3812	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3813	pgmPoolTrackDerefPDPTEPT(pPool, pPage, (PEPTPDPT)pvShw);
3814	break;
3815
3816	default:
3817	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
3818	}
3819
3820	/* paranoia, clear the shadow page. Remove this laser (i.e. let Alloc and ClearAll do it). */
3821	STAM_PROFILE_START(&pPool->StatZeroPage, z);
3822	ASMMemZeroPage(pvShw);
3823	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
3824	pPage->fZeroed = true;
3825	PGMPOOL_UNLOCK_PTR(pPool->CTX_SUFF(pVM), pvShw);
3826	}
3827	#endif /* PGMPOOL_WITH_USER_TRACKING */
3828
3829	/**
3830	* Flushes a pool page.
3831	*
3832	* This moves the page to the free list after removing all user references to it.
3833	*
3834	* @returns VBox status code.
3835	* @retval VINF_SUCCESS on success.
3836	* @param pPool The pool.
3837	* @param HCPhys The HC physical address of the shadow page.
3838	*/
3839	int pgmPoolFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3840	{
3841	PVM pVM = pPool->CTX_SUFF(pVM);
3842
3843	int rc = VINF_SUCCESS;
3844	STAM_PROFILE_START(&pPool->StatFlushPage, f);
3845	LogFlow(("pgmPoolFlushPage: pPage=%p:{.Key=%RHp, .idx=%d, .enmKind=%s, .GCPhys=%RGp}\n",
3846	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
3847
3848	/*
3849	* Quietly reject any attempts at flushing any of the special root pages.
3850	*/
3851	if (pPage->idx < PGMPOOL_IDX_FIRST)
3852	{
3853	AssertFailed(); /* can no longer happen */
3854	Log(("pgmPoolFlushPage: special root page, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
3855	return VINF_SUCCESS;
3856	}
3857
3858	pgmLock(pVM);
3859
3860	/*
3861	* Quietly reject any attempts at flushing the currently active shadow CR3 mapping
3862	*/
3863	if (pgmPoolIsPageLocked(&pVM->pgm.s, pPage))
3864	{
3865	AssertMsg( pPage->enmKind == PGMPOOLKIND_64BIT_PML4
3866	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT
3867	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT_FOR_32BIT
3868	\|\| pPage->enmKind == PGMPOOLKIND_32BIT_PD
3869	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD
3870	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD
3871	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD
3872	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD
3873	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD,
3874	("Can't free the shadow CR3! (%RHp vs %RHp kind=%d\n", PGMGetHyperCR3(VMMGetCpu(pVM)), pPage->Core.Key, pPage->enmKind));
3875	Log(("pgmPoolFlushPage: current active shadow CR3, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
3876	pgmUnlock(pVM);
3877	return VINF_SUCCESS;
3878	}
3879
3880	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
3881	/* Start a subset so we won't run out of mapping space. */
3882	PVMCPU pVCpu = VMMGetCpu(pVM);
3883	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
3884	#endif
3885
3886	/*
3887	* Mark the page as being in need of a ASMMemZeroPage().
3888	*/
3889	pPage->fZeroed = false;
3890
3891	#ifdef PGMPOOL_WITH_USER_TRACKING
3892	/*
3893	* Clear the page.
3894	*/
3895	pgmPoolTrackClearPageUsers(pPool, pPage);
3896	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
3897	pgmPoolTrackDeref(pPool, pPage);
3898	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
3899	#endif
3900
3901	#ifdef PGMPOOL_WITH_CACHE
3902	/*
3903	* Flush it from the cache.
3904	*/
3905	pgmPoolCacheFlushPage(pPool, pPage);
3906	#endif /* PGMPOOL_WITH_CACHE */
3907
3908	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
3909	/* Heavy stuff done. */
3910	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
3911	#endif
3912
3913	#ifdef PGMPOOL_WITH_MONITORING
3914	/*
3915	* Deregistering the monitoring.
3916	*/
3917	if (pPage->fMonitored)
3918	rc = pgmPoolMonitorFlush(pPool, pPage);
3919	#endif
3920
3921	/*
3922	* Free the page.
3923	*/
3924	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
3925	pPage->iNext = pPool->iFreeHead;
3926	pPool->iFreeHead = pPage->idx;
3927	pPage->enmKind = PGMPOOLKIND_FREE;
3928	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
3929	pPage->GCPhys = NIL_RTGCPHYS;
3930	pPage->fReusedFlushPending = false;
3931
3932	pPool->cUsedPages--;
3933	pgmUnlock(pVM);
3934	STAM_PROFILE_STOP(&pPool->StatFlushPage, f);
3935	return rc;
3936	}
3937
3938
3939	/**
3940	* Frees a usage of a pool page.
3941	*
3942	* The caller is responsible to updating the user table so that it no longer
3943	* references the shadow page.
3944	*
3945	* @param pPool The pool.
3946	* @param HCPhys The HC physical address of the shadow page.
3947	* @param iUser The shadow page pool index of the user table.
3948	* @param iUserTable The index into the user table (shadowed).
3949	*/
3950	void pgmPoolFreeByPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
3951	{
3952	PVM pVM = pPool->CTX_SUFF(pVM);
3953
3954	STAM_PROFILE_START(&pPool->StatFree, a);
3955	LogFlow(("pgmPoolFreeByPage: pPage=%p:{.Key=%RHp, .idx=%d, enmKind=%s} iUser=%#x iUserTable=%#x\n",
3956	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), iUser, iUserTable));
3957	Assert(pPage->idx >= PGMPOOL_IDX_FIRST);
3958	pgmLock(pVM);
3959	#ifdef PGMPOOL_WITH_USER_TRACKING
3960	pgmPoolTrackFreeUser(pPool, pPage, iUser, iUserTable);
3961	#endif
3962	#ifdef PGMPOOL_WITH_CACHE
3963	if (!pPage->fCached)
3964	#endif
3965	pgmPoolFlushPage(pPool, pPage);
3966	pgmUnlock(pVM);
3967	STAM_PROFILE_STOP(&pPool->StatFree, a);
3968	}
3969
3970
3971	/**
3972	* Makes one or more free page free.
3973	*
3974	* @returns VBox status code.
3975	* @retval VINF_SUCCESS on success.
3976	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
3977	*
3978	* @param pPool The pool.
3979	* @param enmKind Page table kind
3980	* @param iUser The user of the page.
3981	*/
3982	static int pgmPoolMakeMoreFreePages(PPGMPOOL pPool, PGMPOOLKIND enmKind, uint16_t iUser)
3983	{
3984	PVM pVM = pPool->CTX_SUFF(pVM);
3985
3986	LogFlow(("pgmPoolMakeMoreFreePages: iUser=%#x\n", iUser));
3987
3988	/*
3989	* If the pool isn't full grown yet, expand it.
3990	*/
3991	if ( pPool->cCurPages < pPool->cMaxPages
3992	#if defined(IN_RC)
3993	/* Hack alert: we can't deal with jumps to ring 3 when called from MapCR3 and allocating pages for PAE PDs. */
3994	&& enmKind != PGMPOOLKIND_PAE_PD_FOR_PAE_PD
3995	&& (enmKind < PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD \|\| enmKind > PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD)
3996	#endif
3997	)
3998	{
3999	STAM_PROFILE_ADV_SUSPEND(&pPool->StatAlloc, a);
4000	#ifdef IN_RING3
4001	int rc = PGMR3PoolGrow(pVM);
4002	#else
4003	int rc = CTXALLMID(VMM, CallHost)(pVM, VMMCALLHOST_PGM_POOL_GROW, 0);
4004	#endif
4005	if (RT_FAILURE(rc))
4006	return rc;
4007	STAM_PROFILE_ADV_RESUME(&pPool->StatAlloc, a);
4008	if (pPool->iFreeHead != NIL_PGMPOOL_IDX)
4009	return VINF_SUCCESS;
4010	}
4011
4012	#ifdef PGMPOOL_WITH_CACHE
4013	/*
4014	* Free one cached page.
4015	*/
4016	return pgmPoolCacheFreeOne(pPool, iUser);
4017	#else
4018	/*
4019	* Flush the pool.
4020	*
4021	* If we have tracking enabled, it should be possible to come up with
4022	* a cheap replacement strategy...
4023	*/
4024	/* @todo This path no longer works (CR3 root pages will be flushed)!! */
4025	AssertCompileFailed();
4026	Assert(!CPUMIsGuestInLongMode(pVM));
4027	pgmPoolFlushAllInt(pPool);
4028	return VERR_PGM_POOL_FLUSHED;
4029	#endif
4030	}
4031
4032	/**
4033	* Allocates a page from the pool.
4034	*
4035	* This page may actually be a cached page and not in need of any processing
4036	* on the callers part.
4037	*
4038	* @returns VBox status code.
4039	* @retval VINF_SUCCESS if a NEW page was allocated.
4040	* @retval VINF_PGM_CACHED_PAGE if a CACHED page was returned.
4041	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
4042	* @param pVM The VM handle.
4043	* @param GCPhys The GC physical address of the page we're gonna shadow.
4044	* For 4MB and 2MB PD entries, it's the first address the
4045	* shadow PT is covering.
4046	* @param enmKind The kind of mapping.
4047	* @param enmAccess Access type for the mapping (only relevant for big pages)
4048	* @param iUser The shadow page pool index of the user table.
4049	* @param iUserTable The index into the user table (shadowed).
4050	* @param ppPage Where to store the pointer to the page. NULL is stored here on failure.
4051	*/
4052	int pgmPoolAllocEx(PVM pVM, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, PGMPOOLACCESS enmAccess, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
4053	{
4054	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4055	STAM_PROFILE_ADV_START(&pPool->StatAlloc, a);
4056	LogFlow(("pgmPoolAlloc: GCPhys=%RGp enmKind=%s iUser=%#x iUserTable=%#x\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable));
4057	*ppPage = NULL;
4058	/** @todo CSAM/PGMPrefetchPage messes up here during CSAMR3CheckGates
4059	* (TRPMR3SyncIDT) because of FF priority. Try fix that?
4060	* Assert(!(pVM->pgm.s.fGlobalSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)); */
4061
4062	pgmLock(pVM);
4063
4064	#ifdef PGMPOOL_WITH_CACHE
4065	if (pPool->fCacheEnabled)
4066	{
4067	int rc2 = pgmPoolCacheAlloc(pPool, GCPhys, enmKind, enmAccess, iUser, iUserTable, ppPage);
4068	if (RT_SUCCESS(rc2))
4069	{
4070	pgmUnlock(pVM);
4071	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4072	LogFlow(("pgmPoolAlloc: cached returns %Rrc ppPage=%p:{.Key=%RHp, .idx=%d}\n", rc2, ppPage, (ppPage)->Core.Key, (ppPage)->idx));
4073	return rc2;
4074	}
4075	}
4076	#endif
4077
4078	/*
4079	* Allocate a new one.
4080	*/
4081	int rc = VINF_SUCCESS;
4082	uint16_t iNew = pPool->iFreeHead;
4083	if (iNew == NIL_PGMPOOL_IDX)
4084	{
4085	rc = pgmPoolMakeMoreFreePages(pPool, enmKind, iUser);
4086	if (RT_FAILURE(rc))
4087	{
4088	pgmUnlock(pVM);
4089	Log(("pgmPoolAlloc: returns %Rrc (Free)\n", rc));
4090	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4091	return rc;
4092	}
4093	iNew = pPool->iFreeHead;
4094	AssertReleaseReturn(iNew != NIL_PGMPOOL_IDX, VERR_INTERNAL_ERROR);
4095	}
4096
4097	/* unlink the free head */
4098	PPGMPOOLPAGE pPage = &pPool->aPages[iNew];
4099	pPool->iFreeHead = pPage->iNext;
4100	pPage->iNext = NIL_PGMPOOL_IDX;
4101
4102	/*
4103	* Initialize it.
4104	*/
4105	pPool->cUsedPages++; /* physical handler registration / pgmPoolTrackFlushGCPhysPTsSlow requirement. */
4106	pPage->enmKind = enmKind;
4107	pPage->enmAccess = enmAccess;
4108	pPage->GCPhys = GCPhys;
4109	pPage->fSeenNonGlobal = false; /* Set this to 'true' to disable this feature. */
4110	pPage->fMonitored = false;
4111	pPage->fCached = false;
4112	pPage->fReusedFlushPending = false;
4113	#ifdef PGMPOOL_WITH_MONITORING
4114	pPage->cModifications = 0;
4115	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4116	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4117	#else
4118	pPage->fCR3Mix = false;
4119	#endif
4120	#ifdef PGMPOOL_WITH_USER_TRACKING
4121	pPage->cPresent = 0;
4122	pPage->iFirstPresent = ~0;
4123
4124	/*
4125	* Insert into the tracking and cache. If this fails, free the page.
4126	*/
4127	int rc3 = pgmPoolTrackInsert(pPool, pPage, GCPhys, iUser, iUserTable);
4128	if (RT_FAILURE(rc3))
4129	{
4130	pPool->cUsedPages--;
4131	pPage->enmKind = PGMPOOLKIND_FREE;
4132	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
4133	pPage->GCPhys = NIL_RTGCPHYS;
4134	pPage->iNext = pPool->iFreeHead;
4135	pPool->iFreeHead = pPage->idx;
4136	pgmUnlock(pVM);
4137	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4138	Log(("pgmPoolAlloc: returns %Rrc (Insert)\n", rc3));
4139	return rc3;
4140	}
4141	#endif /* PGMPOOL_WITH_USER_TRACKING */
4142
4143	/*
4144	* Commit the allocation, clear the page and return.
4145	*/
4146	#ifdef VBOX_WITH_STATISTICS
4147	if (pPool->cUsedPages > pPool->cUsedPagesHigh)
4148	pPool->cUsedPagesHigh = pPool->cUsedPages;
4149	#endif
4150
4151	if (!pPage->fZeroed)
4152	{
4153	STAM_PROFILE_START(&pPool->StatZeroPage, z);
4154	void *pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
4155	ASMMemZeroPage(pv);
4156	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
4157	}
4158
4159	*ppPage = pPage;
4160	pgmUnlock(pVM);
4161	LogFlow(("pgmPoolAlloc: returns %Rrc *ppPage=%p:{.Key=%RHp, .idx=%d, .fCached=%RTbool, .fMonitored=%RTbool}\n",
4162	rc, pPage, pPage->Core.Key, pPage->idx, pPage->fCached, pPage->fMonitored));
4163	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4164	return rc;
4165	}
4166
4167
4168	/**
4169	* Frees a usage of a pool page.
4170	*
4171	* @param pVM The VM handle.
4172	* @param HCPhys The HC physical address of the shadow page.
4173	* @param iUser The shadow page pool index of the user table.
4174	* @param iUserTable The index into the user table (shadowed).
4175	*/
4176	void pgmPoolFree(PVM pVM, RTHCPHYS HCPhys, uint16_t iUser, uint32_t iUserTable)
4177	{
4178	LogFlow(("pgmPoolFree: HCPhys=%RHp iUser=%#x iUserTable=%#x\n", HCPhys, iUser, iUserTable));
4179	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4180	pgmPoolFreeByPage(pPool, pgmPoolGetPage(pPool, HCPhys), iUser, iUserTable);
4181	}
4182
4183	/**
4184	* Internal worker for finding a 'in-use' shadow page give by it's physical address.
4185	*
4186	* @returns Pointer to the shadow page structure.
4187	* @param pPool The pool.
4188	* @param HCPhys The HC physical address of the shadow page.
4189	*/
4190	PPGMPOOLPAGE pgmPoolGetPage(PPGMPOOL pPool, RTHCPHYS HCPhys)
4191	{
4192	PVM pVM = pPool->CTX_SUFF(pVM);
4193
4194	/*
4195	* Look up the page.
4196	*/
4197	pgmLock(pVM);
4198	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, HCPhys & X86_PTE_PAE_PG_MASK);
4199	pgmUnlock(pVM);
4200
4201	AssertFatalMsg(pPage && pPage->enmKind != PGMPOOLKIND_FREE, ("HCPhys=%RHp pPage=%p idx=%d\n", HCPhys, pPage, (pPage) ? pPage->idx : 0));
4202	return pPage;
4203	}
4204
4205
4206	#ifdef IN_RING3
4207	/**
4208	* Flushes the entire cache.
4209	*
4210	* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
4211	* and execute this CR3 flush.
4212	*
4213	* @param pPool The pool.
4214	*/
4215	void pgmR3PoolReset(PVM pVM)
4216	{
4217	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4218
4219	Assert(PGMIsLockOwner(pVM));
4220	STAM_PROFILE_START(&pPool->StatFlushAllInt, a);
4221	LogFlow(("pgmPoolFlushAllInt:\n"));
4222
4223	/*
4224	* If there are no pages in the pool, there is nothing to do.
4225	*/
4226	if (pPool->cCurPages <= PGMPOOL_IDX_FIRST)
4227	{
4228	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
4229	return;
4230	}
4231
4232	/*
4233	* Exit the shadow mode since we're going to clear everything,
4234	* including the root page.
4235	*/
4236	for (unsigned i=0;i<pVM->cCPUs;i++)
4237	{
4238	PVMCPU pVCpu = &pVM->aCpus[i];
4239	pgmR3ExitShadowModeBeforePoolFlush(pVM, pVCpu);
4240	}
4241
4242	/*
4243	* Nuke the free list and reinsert all pages into it.
4244	*/
4245	for (unsigned i = pPool->cCurPages - 1; i >= PGMPOOL_IDX_FIRST; i--)
4246	{
4247	PPGMPOOLPAGE pPage = &pPool->aPages[i];
4248
4249	Assert(pPage->Core.Key == MMPage2Phys(pVM, pPage->pvPageR3));
4250	#ifdef PGMPOOL_WITH_MONITORING
4251	if (pPage->fMonitored)
4252	pgmPoolMonitorFlush(pPool, pPage);
4253	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4254	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4255	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
4256	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
4257	pPage->cModifications = 0;
4258	#endif
4259	pPage->GCPhys = NIL_RTGCPHYS;
4260	pPage->enmKind = PGMPOOLKIND_FREE;
4261	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
4262	Assert(pPage->idx == i);
4263	pPage->iNext = i + 1;
4264	pPage->fZeroed = false; /* This could probably be optimized, but better safe than sorry. */
4265	pPage->fSeenNonGlobal = false;
4266	pPage->fMonitored = false;
4267	pPage->fCached = false;
4268	pPage->fReusedFlushPending = false;
4269	#ifdef PGMPOOL_WITH_USER_TRACKING
4270	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
4271	#else
4272	pPage->fCR3Mix = false;
4273	#endif
4274	#ifdef PGMPOOL_WITH_CACHE
4275	pPage->iAgeNext = NIL_PGMPOOL_IDX;
4276	pPage->iAgePrev = NIL_PGMPOOL_IDX;
4277	#endif
4278	pPage->cLocked = 0;
4279	}
4280	pPool->aPages[pPool->cCurPages - 1].iNext = NIL_PGMPOOL_IDX;
4281	pPool->iFreeHead = PGMPOOL_IDX_FIRST;
4282	pPool->cUsedPages = 0;
4283
4284	#ifdef PGMPOOL_WITH_USER_TRACKING
4285	/*
4286	* Zap and reinitialize the user records.
4287	*/
4288	pPool->cPresent = 0;
4289	pPool->iUserFreeHead = 0;
4290	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
4291	const unsigned cMaxUsers = pPool->cMaxUsers;
4292	for (unsigned i = 0; i < cMaxUsers; i++)
4293	{
4294	paUsers[i].iNext = i + 1;
4295	paUsers[i].iUser = NIL_PGMPOOL_IDX;
4296	paUsers[i].iUserTable = 0xfffffffe;
4297	}
4298	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
4299	#endif
4300
4301	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
4302	/*
4303	* Clear all the GCPhys links and rebuild the phys ext free list.
4304	*/
4305	for (PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRanges);
4306	pRam;
4307	pRam = pRam->CTX_SUFF(pNext))
4308	{
4309	unsigned iPage = pRam->cb >> PAGE_SHIFT;
4310	while (iPage-- > 0)
4311	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
4312	}
4313
4314	pPool->iPhysExtFreeHead = 0;
4315	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
4316	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
4317	for (unsigned i = 0; i < cMaxPhysExts; i++)
4318	{
4319	paPhysExts[i].iNext = i + 1;
4320	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
4321	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
4322	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
4323	}
4324	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
4325	#endif
4326
4327	#ifdef PGMPOOL_WITH_MONITORING
4328	/*
4329	* Just zap the modified list.
4330	*/
4331	pPool->cModifiedPages = 0;
4332	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
4333	#endif
4334
4335	#ifdef PGMPOOL_WITH_CACHE
4336	/*
4337	* Clear the GCPhys hash and the age list.
4338	*/
4339	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
4340	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
4341	pPool->iAgeHead = NIL_PGMPOOL_IDX;
4342	pPool->iAgeTail = NIL_PGMPOOL_IDX;
4343	#endif
4344
4345	/*
4346	* Reinsert active pages into the hash and ensure monitoring chains are correct.
4347	*/
4348	for (unsigned i = PGMPOOL_IDX_FIRST_SPECIAL; i < PGMPOOL_IDX_FIRST; i++)
4349	{
4350	PPGMPOOLPAGE pPage = &pPool->aPages[i];
4351	pPage->iNext = NIL_PGMPOOL_IDX;
4352	#ifdef PGMPOOL_WITH_MONITORING
4353	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4354	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4355	pPage->cModifications = 0;
4356	/* ASSUMES that we're not sharing with any of the other special pages (safe for now). */
4357	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
4358	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
4359	if (pPage->fMonitored)
4360	{
4361	int rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
4362	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
4363	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
4364	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
4365	pPool->pszAccessHandler);
4366	AssertFatalRCSuccess(rc);
4367	# ifdef PGMPOOL_WITH_CACHE
4368	pgmPoolHashInsert(pPool, pPage);
4369	# endif
4370	}
4371	#endif
4372	#ifdef PGMPOOL_WITH_USER_TRACKING
4373	Assert(pPage->iUserHead == NIL_PGMPOOL_USER_INDEX); /* for now */
4374	#endif
4375	#ifdef PGMPOOL_WITH_CACHE
4376	Assert(pPage->iAgeNext == NIL_PGMPOOL_IDX);
4377	Assert(pPage->iAgePrev == NIL_PGMPOOL_IDX);
4378	#endif
4379	}
4380
4381	for (unsigned i=0;i<pVM->cCPUs;i++)
4382	{
4383	PVMCPU pVCpu = &pVM->aCpus[i];
4384	/*
4385	* Re-enter the shadowing mode and assert Sync CR3 FF.
4386	*/
4387	pgmR3ReEnterShadowModeAfterPoolFlush(pVM, pVCpu);
4388	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
4389	}
4390
4391	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
4392	}
4393	#endif /* IN_RING3 */
4394
4395	#ifdef LOG_ENABLED
4396	static const char *pgmPoolPoolKindToStr(uint8_t enmKind)
4397	{
4398	switch(enmKind)
4399	{
4400	case PGMPOOLKIND_INVALID:
4401	return "PGMPOOLKIND_INVALID";
4402	case PGMPOOLKIND_FREE:
4403	return "PGMPOOLKIND_FREE";
4404	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
4405	return "PGMPOOLKIND_32BIT_PT_FOR_PHYS";
4406	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
4407	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT";
4408	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
4409	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB";
4410	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
4411	return "PGMPOOLKIND_PAE_PT_FOR_PHYS";
4412	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
4413	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_PT";
4414	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
4415	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB";
4416	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
4417	return "PGMPOOLKIND_PAE_PT_FOR_PAE_PT";
4418	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
4419	return "PGMPOOLKIND_PAE_PT_FOR_PAE_2MB";
4420	case PGMPOOLKIND_32BIT_PD:
4421	return "PGMPOOLKIND_32BIT_PD";
4422	case PGMPOOLKIND_32BIT_PD_PHYS:
4423	return "PGMPOOLKIND_32BIT_PD_PHYS";
4424	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
4425	return "PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD";
4426	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
4427	return "PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD";
4428	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
4429	return "PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD";
4430	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
4431	return "PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD";
4432	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
4433	return "PGMPOOLKIND_PAE_PD_FOR_PAE_PD";
4434	case PGMPOOLKIND_PAE_PD_PHYS:
4435	return "PGMPOOLKIND_PAE_PD_PHYS";
4436	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
4437	return "PGMPOOLKIND_PAE_PDPT_FOR_32BIT";
4438	case PGMPOOLKIND_PAE_PDPT:
4439	return "PGMPOOLKIND_PAE_PDPT";
4440	case PGMPOOLKIND_PAE_PDPT_PHYS:
4441	return "PGMPOOLKIND_PAE_PDPT_PHYS";
4442	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
4443	return "PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT";
4444	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
4445	return "PGMPOOLKIND_64BIT_PDPT_FOR_PHYS";
4446	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
4447	return "PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD";
4448	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
4449	return "PGMPOOLKIND_64BIT_PD_FOR_PHYS";
4450	case PGMPOOLKIND_64BIT_PML4:
4451	return "PGMPOOLKIND_64BIT_PML4";
4452	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
4453	return "PGMPOOLKIND_EPT_PDPT_FOR_PHYS";
4454	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
4455	return "PGMPOOLKIND_EPT_PD_FOR_PHYS";
4456	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
4457	return "PGMPOOLKIND_EPT_PT_FOR_PHYS";
4458	case PGMPOOLKIND_ROOT_NESTED:
4459	return "PGMPOOLKIND_ROOT_NESTED";
4460	}
4461	return "Unknown kind!";
4462	}
4463	#endif /* LOG_ENABLED*/

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllPool.cpp@ 20563

Download in other formats: