PGMAllPool.cpp@ 29287

Last change on this file since 29287 was 29250, checked in by vboxsync, 15 years ago
iprt/asm.h: split out asm-math.h, don't include asm-.h from asm.h, don't include asm.h from sup.h. Fixed a couple file headers.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 197.4 KB

Line
1	/* $Id: PGMAllPool.cpp 29250 2010-05-09 17:53:58Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Oracle Corporation
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
18
19	/*******************************************************************************
20	* Header Files *
21	*******************************************************************************/
22	#define LOG_GROUP LOG_GROUP_PGM_POOL
23	#include <VBox/pgm.h>
24	#include <VBox/mm.h>
25	#include <VBox/em.h>
26	#include <VBox/cpum.h>
27	#ifdef IN_RC
28	# include <VBox/patm.h>
29	#endif
30	#include "../PGMInternal.h"
31	#include <VBox/vm.h>
32	#include "../PGMInline.h"
33	#include <VBox/disopcode.h>
34	#include <VBox/hwacc_vmx.h>
35
36	#include <VBox/log.h>
37	#include <VBox/err.h>
38	#include <iprt/asm.h>
39	#include <iprt/asm-amd64-x86.h>
40	#include <iprt/string.h>
41
42
43	/*******************************************************************************
44	* Internal Functions *
45	*******************************************************************************/
46	RT_C_DECLS_BEGIN
47	static void pgmPoolFlushAllInt(PPGMPOOL pPool);
48	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind);
49	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind);
50	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
51	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable);
52	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
53	#ifndef IN_RING3
54	DECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser);
55	#endif
56	#ifdef LOG_ENABLED
57	static const char *pgmPoolPoolKindToStr(uint8_t enmKind);
58	#endif
59	#if defined(VBOX_STRICT) && defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT)
60	static void pgmPoolTrackCheckPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PTPAE pGstPT);
61	#endif
62
63	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage);
64	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt);
65	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt);
66	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt);
67
68	RT_C_DECLS_END
69
70
71	/**
72	* Checks if the specified page pool kind is for a 4MB or 2MB guest page.
73	*
74	* @returns true if it's the shadow of a 4MB or 2MB guest page, otherwise false.
75	* @param enmKind The page kind.
76	*/
77	DECLINLINE(bool) pgmPoolIsBigPage(PGMPOOLKIND enmKind)
78	{
79	switch (enmKind)
80	{
81	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
82	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
83	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
84	return true;
85	default:
86	return false;
87	}
88	}
89
90	/** @def PGMPOOL_PAGE_2_LOCKED_PTR
91	* Maps a pool page pool into the current context and lock it (RC only).
92	*
93	* @returns VBox status code.
94	* @param pVM The VM handle.
95	* @param pPage The pool page.
96	*
97	* @remark In RC this uses PGMGCDynMapHCPage(), so it will consume of the
98	* small page window employeed by that function. Be careful.
99	* @remark There is no need to assert on the result.
100	*/
101	#if defined(IN_RC)
102	DECLINLINE(void *) PGMPOOL_PAGE_2_LOCKED_PTR(PVM pVM, PPGMPOOLPAGE pPage)
103	{
104	void *pv = pgmPoolMapPageInlined(&pVM->pgm.s, pPage);
105
106	/* Make sure the dynamic mapping will not be reused. */
107	if (pv)
108	PGMDynLockHCPage(pVM, (uint8_t *)pv);
109
110	return pv;
111	}
112	#else
113	# define PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage) PGMPOOL_PAGE_2_PTR(pVM, pPage)
114	#endif
115
116	/** @def PGMPOOL_UNLOCK_PTR
117	* Unlock a previously locked dynamic caching (RC only).
118	*
119	* @returns VBox status code.
120	* @param pVM The VM handle.
121	* @param pPage The pool page.
122	*
123	* @remark In RC this uses PGMGCDynMapHCPage(), so it will consume of the
124	* small page window employeed by that function. Be careful.
125	* @remark There is no need to assert on the result.
126	*/
127	#if defined(IN_RC)
128	DECLINLINE(void) PGMPOOL_UNLOCK_PTR(PVM pVM, void *pvPage)
129	{
130	if (pvPage)
131	PGMDynUnlockHCPage(pVM, (uint8_t *)pvPage);
132	}
133	#else
134	# define PGMPOOL_UNLOCK_PTR(pVM, pPage) do {} while (0)
135	#endif
136
137
138	/**
139	* Flushes a chain of pages sharing the same access monitor.
140	*
141	* @returns VBox status code suitable for scheduling.
142	* @param pPool The pool.
143	* @param pPage A page in the chain.
144	*/
145	int pgmPoolMonitorChainFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
146	{
147	LogFlow(("pgmPoolMonitorChainFlush: Flush page %RGp type=%d\n", pPage->GCPhys, pPage->enmKind));
148
149	/*
150	* Find the list head.
151	*/
152	uint16_t idx = pPage->idx;
153	if (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
154	{
155	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
156	{
157	idx = pPage->iMonitoredPrev;
158	Assert(idx != pPage->idx);
159	pPage = &pPool->aPages[idx];
160	}
161	}
162
163	/*
164	* Iterate the list flushing each shadow page.
165	*/
166	int rc = VINF_SUCCESS;
167	for (;;)
168	{
169	idx = pPage->iMonitoredNext;
170	Assert(idx != pPage->idx);
171	if (pPage->idx >= PGMPOOL_IDX_FIRST)
172	{
173	int rc2 = pgmPoolFlushPage(pPool, pPage);
174	AssertRC(rc2);
175	}
176	/* next */
177	if (idx == NIL_PGMPOOL_IDX)
178	break;
179	pPage = &pPool->aPages[idx];
180	}
181	return rc;
182	}
183
184
185	/**
186	* Wrapper for getting the current context pointer to the entry being modified.
187	*
188	* @returns VBox status code suitable for scheduling.
189	* @param pVM VM Handle.
190	* @param pvDst Destination address
191	* @param pvSrc Source guest virtual address.
192	* @param GCPhysSrc The source guest physical address.
193	* @param cb Size of data to read
194	*/
195	DECLINLINE(int) pgmPoolPhysSimpleReadGCPhys(PVM pVM, void *pvDst, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvSrc, RTGCPHYS GCPhysSrc, size_t cb)
196	{
197	#if defined(IN_RING3)
198	memcpy(pvDst, (RTHCPTR)((uintptr_t)pvSrc & ~(RTHCUINTPTR)(cb - 1)), cb);
199	return VINF_SUCCESS;
200	#else
201	/* @todo in RC we could attempt to use the virtual address, although this can cause many faults (PAE Windows XP guest). */
202	return PGMPhysSimpleReadGCPhys(pVM, pvDst, GCPhysSrc & ~(RTGCPHYS)(cb - 1), cb);
203	#endif
204	}
205
206	/**
207	* Process shadow entries before they are changed by the guest.
208	*
209	* For PT entries we will clear them. For PD entries, we'll simply check
210	* for mapping conflicts and set the SyncCR3 FF if found.
211	*
212	* @param pVCpu VMCPU handle
213	* @param pPool The pool.
214	* @param pPage The head page.
215	* @param GCPhysFault The guest physical fault address.
216	* @param uAddress In R0 and GC this is the guest context fault address (flat).
217	* In R3 this is the host context 'fault' address.
218	* @param cbWrite Write size; might be zero if the caller knows we're not crossing entry boundaries
219	*/
220	void pgmPoolMonitorChainChanging(PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhysFault, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvAddress, unsigned cbWrite)
221	{
222	AssertMsg(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX, ("%#x (idx=%#x)\n", pPage->iMonitoredPrev, pPage->idx));
223	const unsigned off = GCPhysFault & PAGE_OFFSET_MASK;
224	PVM pVM = pPool->CTX_SUFF(pVM);
225
226	LogFlow(("pgmPoolMonitorChainChanging: %RGv phys=%RGp cbWrite=%d\n", (RTGCPTR)(CTXTYPE(RTGCPTR, uintptr_t, RTGCPTR))pvAddress, GCPhysFault, cbWrite));
227
228	for (;;)
229	{
230	union
231	{
232	void *pv;
233	PX86PT pPT;
234	PX86PTPAE pPTPae;
235	PX86PD pPD;
236	PX86PDPAE pPDPae;
237	PX86PDPT pPDPT;
238	PX86PML4 pPML4;
239	} uShw;
240
241	LogFlow(("pgmPoolMonitorChainChanging: page idx=%d phys=%RGp (next=%d) kind=%s\n", pPage->idx, pPage->GCPhys, pPage->iMonitoredNext, pgmPoolPoolKindToStr(pPage->enmKind), cbWrite));
242
243	uShw.pv = NULL;
244	switch (pPage->enmKind)
245	{
246	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
247	{
248	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPT));
249	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
250	const unsigned iShw = off / sizeof(X86PTE);
251	LogFlow(("PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT iShw=%x\n", iShw));
252	if (uShw.pPT->a[iShw].n.u1Present)
253	{
254	X86PTE GstPte;
255
256	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
257	AssertRC(rc);
258	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));
259	pgmPoolTracDerefGCPhysHint(pPool, pPage,
260	uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK,
261	GstPte.u & X86_PTE_PG_MASK,
262	iShw);
263	ASMAtomicWriteSize(&uShw.pPT->a[iShw], 0);
264	}
265	break;
266	}
267
268	/* page/2 sized */
269	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
270	{
271	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPT));
272	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
273	if (!((off ^ pPage->GCPhys) & (PAGE_SIZE / 2)))
274	{
275	const unsigned iShw = (off / sizeof(X86PTE)) & (X86_PG_PAE_ENTRIES - 1);
276	LogFlow(("PGMPOOLKIND_PAE_PT_FOR_32BIT_PT iShw=%x\n", iShw));
277	if (uShw.pPTPae->a[iShw].n.u1Present)
278	{
279	X86PTE GstPte;
280	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
281	AssertRC(rc);
282
283	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));
284	pgmPoolTracDerefGCPhysHint(pPool, pPage,
285	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
286	GstPte.u & X86_PTE_PG_MASK,
287	iShw);
288	ASMAtomicWriteSize(&uShw.pPTPae->a[iShw], 0);
289	}
290	}
291	break;
292	}
293
294	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
295	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
296	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
297	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
298	{
299	unsigned iGst = off / sizeof(X86PDE);
300	unsigned iShwPdpt = iGst / 256;
301	unsigned iShw = (iGst % 256) * 2;
302	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
303
304	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));
305	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPD));
306	if (iShwPdpt == pPage->enmKind - (unsigned)PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD)
307	{
308	for (unsigned i = 0; i < 2; i++)
309	{
310	# ifndef IN_RING0
311	if ((uShw.pPDPae->a[iShw + i].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
312	{
313	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
314	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
315	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw=%#x!\n", iShwPdpt, iShw+i));
316	break;
317	}
318	else
319	# endif /* !IN_RING0 */
320	if (uShw.pPDPae->a[iShw+i].n.u1Present)
321	{
322	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw+i, uShw.pPDPae->a[iShw+i].u));
323	pgmPoolFree(pVM,
324	uShw.pPDPae->a[iShw+i].u & X86_PDE_PAE_PG_MASK,
325	pPage->idx,
326	iShw + i);
327	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw+i], 0);
328	}
329
330	/* paranoia / a bit assumptive. */
331	if ( (off & 3)
332	&& (off & 3) + cbWrite > 4)
333	{
334	const unsigned iShw2 = iShw + 2 + i;
335	if (iShw2 < RT_ELEMENTS(uShw.pPDPae->a))
336	{
337	# ifndef IN_RING0
338	if ((uShw.pPDPae->a[iShw2].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
339	{
340	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
341	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
342	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw2=%#x!\n", iShwPdpt, iShw2));
343	break;
344	}
345	else
346	# endif /* !IN_RING0 */
347	if (uShw.pPDPae->a[iShw2].n.u1Present)
348	{
349	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
350	pgmPoolFree(pVM,
351	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
352	pPage->idx,
353	iShw2);
354	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw2].u, 0);
355	}
356	}
357	}
358	}
359	}
360	break;
361	}
362
363	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
364	{
365	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
366	const unsigned iShw = off / sizeof(X86PTEPAE);
367	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPT));
368	if (uShw.pPTPae->a[iShw].n.u1Present)
369	{
370	X86PTEPAE GstPte;
371	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
372	AssertRC(rc);
373
374	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));
375	pgmPoolTracDerefGCPhysHint(pPool, pPage,
376	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
377	GstPte.u & X86_PTE_PAE_PG_MASK,
378	iShw);
379	ASMAtomicWriteSize(&uShw.pPTPae->a[iShw].u, 0);
380	}
381
382	/* paranoia / a bit assumptive. */
383	if ( (off & 7)
384	&& (off & 7) + cbWrite > sizeof(X86PTEPAE))
385	{
386	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTEPAE);
387	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPTPae->a));
388
389	if (uShw.pPTPae->a[iShw2].n.u1Present)
390	{
391	X86PTEPAE GstPte;
392	# ifdef IN_RING3
393	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, (RTHCPTR)((RTHCUINTPTR)pvAddress + sizeof(GstPte)), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
394	# else
395	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress + sizeof(GstPte), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
396	# endif
397	AssertRC(rc);
398	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));
399	pgmPoolTracDerefGCPhysHint(pPool, pPage,
400	uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK,
401	GstPte.u & X86_PTE_PAE_PG_MASK,
402	iShw2);
403	ASMAtomicWriteSize(&uShw.pPTPae->a[iShw2].u ,0);
404	}
405	}
406	break;
407	}
408
409	case PGMPOOLKIND_32BIT_PD:
410	{
411	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
412	const unsigned iShw = off / sizeof(X86PTE); // ASSUMING 32-bit guest paging!
413
414	LogFlow(("pgmPoolMonitorChainChanging: PGMPOOLKIND_32BIT_PD %x\n", iShw));
415	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPD));
416	# ifndef IN_RING0
417	if (uShw.pPD->a[iShw].u & PGM_PDFLAGS_MAPPING)
418	{
419	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
420	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
421	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
422	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
423	break;
424	}
425	# endif /* !IN_RING0 */
426	# ifndef IN_RING0
427	else
428	# endif /* !IN_RING0 */
429	{
430	if (uShw.pPD->a[iShw].n.u1Present)
431	{
432	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
433	pgmPoolFree(pVM,
434	uShw.pPD->a[iShw].u & X86_PDE_PAE_PG_MASK,
435	pPage->idx,
436	iShw);
437	ASMAtomicWriteSize(&uShw.pPD->a[iShw].u, 0);
438	}
439	}
440	/* paranoia / a bit assumptive. */
441	if ( (off & 3)
442	&& (off & 3) + cbWrite > sizeof(X86PTE))
443	{
444	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTE);
445	if ( iShw2 != iShw
446	&& iShw2 < RT_ELEMENTS(uShw.pPD->a))
447	{
448	# ifndef IN_RING0
449	if (uShw.pPD->a[iShw2].u & PGM_PDFLAGS_MAPPING)
450	{
451	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
452	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
453	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
454	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
455	break;
456	}
457	# endif /* !IN_RING0 */
458	# ifndef IN_RING0
459	else
460	# endif /* !IN_RING0 */
461	{
462	if (uShw.pPD->a[iShw2].n.u1Present)
463	{
464	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPD->a[iShw2].u));
465	pgmPoolFree(pVM,
466	uShw.pPD->a[iShw2].u & X86_PDE_PAE_PG_MASK,
467	pPage->idx,
468	iShw2);
469	ASMAtomicWriteSize(&uShw.pPD->a[iShw2].u, 0);
470	}
471	}
472	}
473	}
474	#if 0 /* useful when running PGMAssertCR3(), a bit too troublesome for general use (TLBs). */
475	if ( uShw.pPD->a[iShw].n.u1Present
476	&& !VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
477	{
478	LogFlow(("pgmPoolMonitorChainChanging: iShw=%#x: %RX32 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
479	# ifdef IN_RC /* TLB load - we're pushing things a bit... */
480	ASMProbeReadByte(pvAddress);
481	# endif
482	pgmPoolFree(pVM, uShw.pPD->a[iShw].u & X86_PDE_PG_MASK, pPage->idx, iShw);
483	ASMAtomicWriteSize(&uShw.pPD->a[iShw].u, 0);
484	}
485	#endif
486	break;
487	}
488
489	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
490	{
491	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
492	const unsigned iShw = off / sizeof(X86PDEPAE);
493	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPD));
494	#ifndef IN_RING0
495	if (uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING)
496	{
497	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
498	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
499	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
500	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
501	break;
502	}
503	#endif /* !IN_RING0 */
504	/*
505	* Causes trouble when the guest uses a PDE to refer to the whole page table level
506	* structure. (Invalidate here; faults later on when it tries to change the page
507	* table entries -> recheck; probably only applies to the RC case.)
508	*/
509	# ifndef IN_RING0
510	else
511	# endif /* !IN_RING0 */
512	{
513	if (uShw.pPDPae->a[iShw].n.u1Present)
514	{
515	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
516	pgmPoolFree(pVM,
517	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
518	pPage->idx,
519	iShw);
520	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw].u, 0);
521	}
522	}
523	/* paranoia / a bit assumptive. */
524	if ( (off & 7)
525	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
526	{
527	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
528	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
529
530	#ifndef IN_RING0
531	if ( iShw2 != iShw
532	&& uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING)
533	{
534	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
535	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
536	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
537	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
538	break;
539	}
540	#endif /* !IN_RING0 */
541	# ifndef IN_RING0
542	else
543	# endif /* !IN_RING0 */
544	if (uShw.pPDPae->a[iShw2].n.u1Present)
545	{
546	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
547	pgmPoolFree(pVM,
548	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
549	pPage->idx,
550	iShw2);
551	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw2].u, 0);
552	}
553	}
554	break;
555	}
556
557	case PGMPOOLKIND_PAE_PDPT:
558	{
559	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPDPT));
560	/*
561	* Hopefully this doesn't happen very often:
562	* - touching unused parts of the page
563	* - messing with the bits of pd pointers without changing the physical address
564	*/
565	/* PDPT roots are not page aligned; 32 byte only! */
566	const unsigned offPdpt = GCPhysFault - pPage->GCPhys;
567
568	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
569	const unsigned iShw = offPdpt / sizeof(X86PDPE);
570	if (iShw < X86_PG_PAE_PDPE_ENTRIES) /* don't use RT_ELEMENTS(uShw.pPDPT->a), because that's for long mode only */
571	{
572	# ifndef IN_RING0
573	if (uShw.pPDPT->a[iShw].u & PGM_PLXFLAGS_MAPPING)
574	{
575	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
576	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
577	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
578	LogFlow(("pgmPoolMonitorChainChanging: Detected pdpt conflict at iShw=%#x!\n", iShw));
579	break;
580	}
581	# endif /* !IN_RING0 */
582	# ifndef IN_RING0
583	else
584	# endif /* !IN_RING0 */
585	if (uShw.pPDPT->a[iShw].n.u1Present)
586	{
587	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
588	pgmPoolFree(pVM,
589	uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK,
590	pPage->idx,
591	iShw);
592	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw].u, 0);
593	}
594
595	/* paranoia / a bit assumptive. */
596	if ( (offPdpt & 7)
597	&& (offPdpt & 7) + cbWrite > sizeof(X86PDPE))
598	{
599	const unsigned iShw2 = (offPdpt + cbWrite - 1) / sizeof(X86PDPE);
600	if ( iShw2 != iShw
601	&& iShw2 < X86_PG_PAE_PDPE_ENTRIES)
602	{
603	# ifndef IN_RING0
604	if (uShw.pPDPT->a[iShw2].u & PGM_PLXFLAGS_MAPPING)
605	{
606	Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
607	STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
608	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
609	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
610	break;
611	}
612	# endif /* !IN_RING0 */
613	# ifndef IN_RING0
614	else
615	# endif /* !IN_RING0 */
616	if (uShw.pPDPT->a[iShw2].n.u1Present)
617	{
618	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
619	pgmPoolFree(pVM,
620	uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK,
621	pPage->idx,
622	iShw2);
623	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw2].u, 0);
624	}
625	}
626	}
627	}
628	break;
629	}
630
631	#ifndef IN_RC
632	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
633	{
634	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPD));
635	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
636	const unsigned iShw = off / sizeof(X86PDEPAE);
637	Assert(!(uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING));
638	if (uShw.pPDPae->a[iShw].n.u1Present)
639	{
640	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
641	pgmPoolFree(pVM,
642	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
643	pPage->idx,
644	iShw);
645	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw].u, 0);
646	}
647	/* paranoia / a bit assumptive. */
648	if ( (off & 7)
649	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
650	{
651	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
652	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
653
654	Assert(!(uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING));
655	if (uShw.pPDPae->a[iShw2].n.u1Present)
656	{
657	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
658	pgmPoolFree(pVM,
659	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
660	pPage->idx,
661	iShw2);
662	ASMAtomicWriteSize(&uShw.pPDPae->a[iShw2].u, 0);
663	}
664	}
665	break;
666	}
667
668	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
669	{
670	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPDPT));
671	/*
672	* Hopefully this doesn't happen very often:
673	* - messing with the bits of pd pointers without changing the physical address
674	*/
675	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
676	const unsigned iShw = off / sizeof(X86PDPE);
677	if (uShw.pPDPT->a[iShw].n.u1Present)
678	{
679	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
680	pgmPoolFree(pVM, uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK, pPage->idx, iShw);
681	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw].u, 0);
682	}
683	/* paranoia / a bit assumptive. */
684	if ( (off & 7)
685	&& (off & 7) + cbWrite > sizeof(X86PDPE))
686	{
687	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDPE);
688	if (uShw.pPDPT->a[iShw2].n.u1Present)
689	{
690	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
691	pgmPoolFree(pVM, uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK, pPage->idx, iShw2);
692	ASMAtomicWriteSize(&uShw.pPDPT->a[iShw2].u, 0);
693	}
694	}
695	break;
696	}
697
698	case PGMPOOLKIND_64BIT_PML4:
699	{
700	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPML4));
701	/*
702	* Hopefully this doesn't happen very often:
703	* - messing with the bits of pd pointers without changing the physical address
704	*/
705	uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
706	const unsigned iShw = off / sizeof(X86PDPE);
707	if (uShw.pPML4->a[iShw].n.u1Present)
708	{
709	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPML4->a[iShw].u));
710	pgmPoolFree(pVM, uShw.pPML4->a[iShw].u & X86_PML4E_PG_MASK, pPage->idx, iShw);
711	ASMAtomicWriteSize(&uShw.pPML4->a[iShw].u, 0);
712	}
713	/* paranoia / a bit assumptive. */
714	if ( (off & 7)
715	&& (off & 7) + cbWrite > sizeof(X86PDPE))
716	{
717	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PML4E);
718	if (uShw.pPML4->a[iShw2].n.u1Present)
719	{
720	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPML4->a[iShw2].u));
721	pgmPoolFree(pVM, uShw.pPML4->a[iShw2].u & X86_PML4E_PG_MASK, pPage->idx, iShw2);
722	ASMAtomicWriteSize(&uShw.pPML4->a[iShw2].u, 0);
723	}
724	}
725	break;
726	}
727	#endif /* IN_RING0 */
728
729	default:
730	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
731	}
732	PGMPOOL_UNLOCK_PTR(pVM, uShw.pv);
733
734	/* next */
735	if (pPage->iMonitoredNext == NIL_PGMPOOL_IDX)
736	return;
737	pPage = &pPool->aPages[pPage->iMonitoredNext];
738	}
739	}
740
741	# ifndef IN_RING3
742	/**
743	* Checks if a access could be a fork operation in progress.
744	*
745	* Meaning, that the guest is setting up the parent process for Copy-On-Write.
746	*
747	* @returns true if it's likly that we're forking, otherwise false.
748	* @param pPool The pool.
749	* @param pDis The disassembled instruction.
750	* @param offFault The access offset.
751	*/
752	DECLINLINE(bool) pgmPoolMonitorIsForking(PPGMPOOL pPool, PDISCPUSTATE pDis, unsigned offFault)
753	{
754	/*
755	* i386 linux is using btr to clear X86_PTE_RW.
756	* The functions involved are (2.6.16 source inspection):
757	* clear_bit
758	* ptep_set_wrprotect
759	* copy_one_pte
760	* copy_pte_range
761	* copy_pmd_range
762	* copy_pud_range
763	* copy_page_range
764	* dup_mmap
765	* dup_mm
766	* copy_mm
767	* copy_process
768	* do_fork
769	*/
770	if ( pDis->pCurInstr->opcode == OP_BTR
771	&& !(offFault & 4)
772	/** @todo Validate that the bit index is X86_PTE_RW. */
773	)
774	{
775	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,Fork));
776	return true;
777	}
778	return false;
779	}
780
781
782	/**
783	* Determine whether the page is likely to have been reused.
784	*
785	* @returns true if we consider the page as being reused for a different purpose.
786	* @returns false if we consider it to still be a paging page.
787	* @param pVM VM Handle.
788	* @param pVCpu VMCPU Handle.
789	* @param pRegFrame Trap register frame.
790	* @param pDis The disassembly info for the faulting instruction.
791	* @param pvFault The fault address.
792	*
793	* @remark The REP prefix check is left to the caller because of STOSD/W.
794	*/
795	DECLINLINE(bool) pgmPoolMonitorIsReused(PVM pVM, PVMCPU pVCpu, PCPUMCTXCORE pRegFrame, PDISCPUSTATE pDis, RTGCPTR pvFault)
796	{
797	#ifndef IN_RC
798	/** @todo could make this general, faulting close to rsp should be a safe reuse heuristic. */
799	if ( HWACCMHasPendingIrq(pVM)
800	&& (pRegFrame->rsp - pvFault) < 32)
801	{
802	/* Fault caused by stack writes while trying to inject an interrupt event. */
803	Log(("pgmPoolMonitorIsReused: reused %RGv for interrupt stack (rsp=%RGv).\n", pvFault, pRegFrame->rsp));
804	return true;
805	}
806	#else
807	NOREF(pVM); NOREF(pvFault);
808	#endif
809
810	LogFlow(("Reused instr %RGv %d at %RGv param1.flags=%x param1.reg=%d\n", pRegFrame->rip, pDis->pCurInstr->opcode, pvFault, pDis->param1.flags, pDis->param1.base.reg_gen));
811
812	/* Non-supervisor mode write means it's used for something else. */
813	if (CPUMGetGuestCPL(pVCpu, pRegFrame) != 0)
814	return true;
815
816	switch (pDis->pCurInstr->opcode)
817	{
818	/* call implies the actual push of the return address faulted */
819	case OP_CALL:
820	Log4(("pgmPoolMonitorIsReused: CALL\n"));
821	return true;
822	case OP_PUSH:
823	Log4(("pgmPoolMonitorIsReused: PUSH\n"));
824	return true;
825	case OP_PUSHF:
826	Log4(("pgmPoolMonitorIsReused: PUSHF\n"));
827	return true;
828	case OP_PUSHA:
829	Log4(("pgmPoolMonitorIsReused: PUSHA\n"));
830	return true;
831	case OP_FXSAVE:
832	Log4(("pgmPoolMonitorIsReused: FXSAVE\n"));
833	return true;
834	case OP_MOVNTI: /* solaris - block_zero_no_xmm */
835	Log4(("pgmPoolMonitorIsReused: MOVNTI\n"));
836	return true;
837	case OP_MOVNTDQ: /* solaris - hwblkclr & hwblkpagecopy */
838	Log4(("pgmPoolMonitorIsReused: MOVNTDQ\n"));
839	return true;
840	case OP_MOVSWD:
841	case OP_STOSWD:
842	if ( pDis->prefix == (PREFIX_REP\|PREFIX_REX)
843	&& pRegFrame->rcx >= 0x40
844	)
845	{
846	Assert(pDis->mode == CPUMODE_64BIT);
847
848	Log(("pgmPoolMonitorIsReused: OP_STOSQ\n"));
849	return true;
850	}
851	return false;
852	}
853	if ( ( (pDis->param1.flags & USE_REG_GEN32)
854	\|\| (pDis->param1.flags & USE_REG_GEN64))
855	&& (pDis->param1.base.reg_gen == USE_REG_ESP))
856	{
857	Log4(("pgmPoolMonitorIsReused: ESP\n"));
858	return true;
859	}
860
861	return false;
862	}
863
864	/**
865	* Flushes the page being accessed.
866	*
867	* @returns VBox status code suitable for scheduling.
868	* @param pVM The VM handle.
869	* @param pVCpu The VMCPU handle.
870	* @param pPool The pool.
871	* @param pPage The pool page (head).
872	* @param pDis The disassembly of the write instruction.
873	* @param pRegFrame The trap register frame.
874	* @param GCPhysFault The fault address as guest physical address.
875	* @param pvFault The fault address.
876	*/
877	static int pgmPoolAccessHandlerFlush(PVM pVM, PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pDis,
878	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
879	{
880	/*
881	* First, do the flushing.
882	*/
883	int rc = pgmPoolMonitorChainFlush(pPool, pPage);
884
885	/*
886	* Emulate the instruction (xp/w2k problem, requires pc/cr2/sp detection). Must do this in raw mode (!); XP boot will fail otherwise
887	*/
888	uint32_t cbWritten;
889	int rc2 = EMInterpretInstructionCPUEx(pVM, pVCpu, pDis, pRegFrame, pvFault, &cbWritten, EMCODETYPE_ALL);
890	if (RT_SUCCESS(rc2))
891	pRegFrame->rip += pDis->opsize;
892	else if (rc2 == VERR_EM_INTERPRETER)
893	{
894	#ifdef IN_RC
895	if (PATMIsPatchGCAddr(pVM, pRegFrame->eip))
896	{
897	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for patch code %04x:%RGv, ignoring.\n",
898	pRegFrame->cs, (RTGCPTR)pRegFrame->eip));
899	rc = VINF_SUCCESS;
900	STAM_COUNTER_INC(&pPool->StatMonitorRZIntrFailPatch2);
901	}
902	else
903	#endif
904	{
905	rc = VINF_EM_RAW_EMULATE_INSTR;
906	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
907	}
908	}
909	else
910	rc = rc2;
911
912	LogFlow(("pgmPoolAccessHandlerPT: returns %Rrc (flushed)\n", rc));
913	return rc;
914	}
915
916	/**
917	* Handles the STOSD write accesses.
918	*
919	* @returns VBox status code suitable for scheduling.
920	* @param pVM The VM handle.
921	* @param pPool The pool.
922	* @param pPage The pool page (head).
923	* @param pDis The disassembly of the write instruction.
924	* @param pRegFrame The trap register frame.
925	* @param GCPhysFault The fault address as guest physical address.
926	* @param pvFault The fault address.
927	*/
928	DECLINLINE(int) pgmPoolAccessHandlerSTOSD(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pDis,
929	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
930	{
931	unsigned uIncrement = pDis->param1.size;
932
933	Assert(pDis->mode == CPUMODE_32BIT \|\| pDis->mode == CPUMODE_64BIT);
934	Assert(pRegFrame->rcx <= 0x20);
935
936	#ifdef VBOX_STRICT
937	if (pDis->opmode == CPUMODE_32BIT)
938	Assert(uIncrement == 4);
939	else
940	Assert(uIncrement == 8);
941	#endif
942
943	Log3(("pgmPoolAccessHandlerSTOSD\n"));
944
945	/*
946	* Increment the modification counter and insert it into the list
947	* of modified pages the first time.
948	*/
949	if (!pPage->cModifications++)
950	pgmPoolMonitorModifiedInsert(pPool, pPage);
951
952	/*
953	* Execute REP STOSD.
954	*
955	* This ASSUMES that we're not invoked by Trap0e on in a out-of-sync
956	* write situation, meaning that it's safe to write here.
957	*/
958	PVMCPU pVCpu = VMMGetCpu(pPool->CTX_SUFF(pVM));
959	RTGCUINTPTR pu32 = (RTGCUINTPTR)pvFault;
960	while (pRegFrame->rcx)
961	{
962	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
963	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
964	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, (RTGCPTR)pu32, uIncrement);
965	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
966	#else
967	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, (RTGCPTR)pu32, uIncrement);
968	#endif
969	#ifdef IN_RC
970	(uint32_t )(uintptr_t)pu32 = pRegFrame->eax;
971	#else
972	PGMPhysSimpleWriteGCPhys(pVM, GCPhysFault, &pRegFrame->rax, uIncrement);
973	#endif
974	pu32 += uIncrement;
975	GCPhysFault += uIncrement;
976	pRegFrame->rdi += uIncrement;
977	pRegFrame->rcx--;
978	}
979	pRegFrame->rip += pDis->opsize;
980
981	LogFlow(("pgmPoolAccessHandlerSTOSD: returns\n"));
982	return VINF_SUCCESS;
983	}
984
985
986	/**
987	* Handles the simple write accesses.
988	*
989	* @returns VBox status code suitable for scheduling.
990	* @param pVM The VM handle.
991	* @param pVCpu The VMCPU handle.
992	* @param pPool The pool.
993	* @param pPage The pool page (head).
994	* @param pDis The disassembly of the write instruction.
995	* @param pRegFrame The trap register frame.
996	* @param GCPhysFault The fault address as guest physical address.
997	* @param pvFault The fault address.
998	* @param pfReused Reused state (out)
999	*/
1000	DECLINLINE(int) pgmPoolAccessHandlerSimple(PVM pVM, PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pDis,
1001	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault, bool *pfReused)
1002	{
1003	Log3(("pgmPoolAccessHandlerSimple\n"));
1004	/*
1005	* Increment the modification counter and insert it into the list
1006	* of modified pages the first time.
1007	*/
1008	if (!pPage->cModifications++)
1009	pgmPoolMonitorModifiedInsert(pPool, pPage);
1010
1011	/*
1012	* Clear all the pages. ASSUMES that pvFault is readable.
1013	*/
1014	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
1015	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
1016	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, pvFault, DISGetParamSize(pDis, &pDis->param1));
1017	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
1018	#else
1019	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, pvFault, DISGetParamSize(pDis, &pDis->param1));
1020	#endif
1021
1022	/*
1023	* Interpret the instruction.
1024	*/
1025	uint32_t cb;
1026	int rc = EMInterpretInstructionCPUEx(pVM, pVCpu, pDis, pRegFrame, pvFault, &cb, EMCODETYPE_ALL);
1027	if (RT_SUCCESS(rc))
1028	pRegFrame->rip += pDis->opsize;
1029	else if (rc == VERR_EM_INTERPRETER)
1030	{
1031	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for %04x:%RGv - opcode=%d\n",
1032	pRegFrame->cs, (RTGCPTR)pRegFrame->rip, pDis->pCurInstr->opcode));
1033	rc = VINF_EM_RAW_EMULATE_INSTR;
1034	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
1035	}
1036
1037	#if 0 /* experimental code */
1038	if (rc == VINF_SUCCESS)
1039	{
1040	switch (pPage->enmKind)
1041	{
1042	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1043	{
1044	X86PTEPAE GstPte;
1045	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvFault, GCPhysFault, sizeof(GstPte));
1046	AssertRC(rc);
1047
1048	/* Check the new value written by the guest. If present and with a bogus physical address, then
1049	* it's fairly safe to assume the guest is reusing the PT.
1050	*/
1051	if (GstPte.n.u1Present)
1052	{
1053	RTHCPHYS HCPhys = -1;
1054	int rc = PGMPhysGCPhys2HCPhys(pVM, GstPte.u & X86_PTE_PAE_PG_MASK, &HCPhys);
1055	if (rc != VINF_SUCCESS)
1056	{
1057	*pfReused = true;
1058	STAM_COUNTER_INC(&pPool->StatForceFlushReused);
1059	}
1060	}
1061	break;
1062	}
1063	}
1064	}
1065	#endif
1066
1067	LogFlow(("pgmPoolAccessHandlerSimple: returns %Rrc cb=%d\n", rc, cb));
1068	return rc;
1069	}
1070
1071	/**
1072	* \#PF Handler callback for PT write accesses.
1073	*
1074	* @returns VBox status code (appropriate for GC return).
1075	* @param pVM VM Handle.
1076	* @param uErrorCode CPU Error code.
1077	* @param pRegFrame Trap register frame.
1078	* NULL on DMA and other non CPU access.
1079	* @param pvFault The fault address (cr2).
1080	* @param GCPhysFault The GC physical address corresponding to pvFault.
1081	* @param pvUser User argument.
1082	*/
1083	DECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser)
1084	{
1085	STAM_PROFILE_START(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), a);
1086	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1087	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
1088	PVMCPU pVCpu = VMMGetCpu(pVM);
1089	unsigned cMaxModifications;
1090	bool fForcedFlush = false;
1091
1092	LogFlow(("pgmPoolAccessHandler: pvFault=%RGv pPage=%p:{.idx=%d} GCPhysFault=%RGp\n", pvFault, pPage, pPage->idx, GCPhysFault));
1093
1094	pgmLock(pVM);
1095	if (PHYS_PAGE_ADDRESS(GCPhysFault) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
1096	{
1097	/* Pool page changed while we were waiting for the lock; ignore. */
1098	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)));
1099	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,Handled), a);
1100	pgmUnlock(pVM);
1101	return VINF_SUCCESS;
1102	}
1103	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1104	if (pPage->fDirty)
1105	{
1106	Assert(VMCPU_FF_ISSET(pVCpu, VMCPU_FF_TLB_FLUSH));
1107	pgmUnlock(pVM);
1108	return VINF_SUCCESS; /* SMP guest case where we were blocking on the pgm lock while the same page was being marked dirty. */
1109	}
1110	#endif
1111
1112	#if 0 /* test code defined(VBOX_STRICT) && defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT) */
1113	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1114	{
1115	void *pvShw = PGMPOOL_PAGE_2_LOCKED_PTR(pPool->CTX_SUFF(pVM), pPage);
1116	void *pvGst;
1117	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
1118	pgmPoolTrackCheckPTPaePae(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PTPAE)pvGst);
1119	}
1120	#endif
1121
1122	/*
1123	* Disassemble the faulting instruction.
1124	*/
1125	PDISCPUSTATE pDis = &pVCpu->pgm.s.DisState;
1126	int rc = EMInterpretDisasOne(pVM, pVCpu, pRegFrame, pDis, NULL);
1127	if (RT_UNLIKELY(rc != VINF_SUCCESS))
1128	{
1129	AssertMsg(rc == VERR_PAGE_NOT_PRESENT \|\| rc == VERR_PAGE_TABLE_NOT_PRESENT, ("Unexpected rc %d\n", rc));
1130	pgmUnlock(pVM);
1131	return rc;
1132	}
1133
1134	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
1135
1136	/*
1137	* We should ALWAYS have the list head as user parameter. This
1138	* is because we use that page to record the changes.
1139	*/
1140	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1141
1142	#ifdef IN_RING0
1143	/* Maximum nr of modifications depends on the page type. */
1144	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1145	cMaxModifications = 4;
1146	else
1147	cMaxModifications = 24;
1148	#else
1149	cMaxModifications = 48;
1150	#endif
1151
1152	/*
1153	* Incremental page table updates should weigh more than random ones.
1154	* (Only applies when started from offset 0)
1155	*/
1156	pVCpu->pgm.s.cPoolAccessHandler++;
1157	if ( pPage->pvLastAccessHandlerRip >= pRegFrame->rip - 0x40 /* observed loops in Windows 7 x64 */
1158	&& pPage->pvLastAccessHandlerRip < pRegFrame->rip + 0x40
1159	&& pvFault == (pPage->pvLastAccessHandlerFault + pDis->param1.size)
1160	&& pVCpu->pgm.s.cPoolAccessHandler == (pPage->cLastAccessHandlerCount + 1))
1161	{
1162	Log(("Possible page reuse cMods=%d -> %d (locked=%d type=%s)\n", pPage->cModifications, pPage->cModifications * 2, pgmPoolIsPageLocked(&pVM->pgm.s, pPage), pgmPoolPoolKindToStr(pPage->enmKind)));
1163	Assert(pPage->cModifications < 32000);
1164	pPage->cModifications = pPage->cModifications * 2;
1165	pPage->pvLastAccessHandlerFault = pvFault;
1166	pPage->cLastAccessHandlerCount = pVCpu->pgm.s.cPoolAccessHandler;
1167	if (pPage->cModifications >= cMaxModifications)
1168	{
1169	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FlushReinit));
1170	fForcedFlush = true;
1171	}
1172	}
1173
1174	if (pPage->cModifications >= cMaxModifications)
1175	Log(("Mod overflow %RGv cMods=%d (locked=%d type=%s)\n", pvFault, pPage->cModifications, pgmPoolIsPageLocked(&pVM->pgm.s, pPage), pgmPoolPoolKindToStr(pPage->enmKind)));
1176
1177	/*
1178	* Check if it's worth dealing with.
1179	*/
1180	bool fReused = false;
1181	bool fNotReusedNotForking = false;
1182	if ( ( pPage->cModifications < cMaxModifications /** @todo #define / /* @todo need to check that it's not mapping EIP. / /* @todo adjust this! */
1183	\|\| pgmPoolIsPageLocked(&pVM->pgm.s, pPage)
1184	)
1185	&& !(fReused = pgmPoolMonitorIsReused(pVM, pVCpu, pRegFrame, pDis, pvFault))
1186	&& !pgmPoolMonitorIsForking(pPool, pDis, GCPhysFault & PAGE_OFFSET_MASK))
1187	{
1188	/*
1189	* Simple instructions, no REP prefix.
1190	*/
1191	if (!(pDis->prefix & (PREFIX_REP \| PREFIX_REPNE)))
1192	{
1193	rc = pgmPoolAccessHandlerSimple(pVM, pVCpu, pPool, pPage, pDis, pRegFrame, GCPhysFault, pvFault, &fReused);
1194	if (fReused)
1195	goto flushPage;
1196
1197	/* A mov instruction to change the first page table entry will be remembered so we can detect
1198	* full page table changes early on. This will reduce the amount of unnecessary traps we'll take.
1199	*/
1200	if ( rc == VINF_SUCCESS
1201	&& !pPage->cLocked /* only applies to unlocked pages as we can't free locked ones (e.g. cr3 root). */
1202	&& pDis->pCurInstr->opcode == OP_MOV
1203	&& (pvFault & PAGE_OFFSET_MASK) == 0)
1204	{
1205	pPage->pvLastAccessHandlerFault = pvFault;
1206	pPage->cLastAccessHandlerCount = pVCpu->pgm.s.cPoolAccessHandler;
1207	pPage->pvLastAccessHandlerRip = pRegFrame->rip;
1208	/* Make sure we don't kick out a page too quickly. */
1209	if (pPage->cModifications > 8)
1210	pPage->cModifications = 2;
1211	}
1212	else
1213	if (pPage->pvLastAccessHandlerFault == pvFault)
1214	{
1215	/* ignore the 2nd write to this page table entry. */
1216	pPage->cLastAccessHandlerCount = pVCpu->pgm.s.cPoolAccessHandler;
1217	}
1218	else
1219	{
1220	pPage->pvLastAccessHandlerFault = 0;
1221	pPage->pvLastAccessHandlerRip = 0;
1222	}
1223
1224	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,Handled), a);
1225	pgmUnlock(pVM);
1226	return rc;
1227	}
1228
1229	/*
1230	* Windows is frequently doing small memset() operations (netio test 4k+).
1231	* We have to deal with these or we'll kill the cache and performance.
1232	*/
1233	if ( pDis->pCurInstr->opcode == OP_STOSWD
1234	&& !pRegFrame->eflags.Bits.u1DF
1235	&& pDis->opmode == pDis->mode
1236	&& pDis->addrmode == pDis->mode)
1237	{
1238	bool fValidStosd = false;
1239
1240	if ( pDis->mode == CPUMODE_32BIT
1241	&& pDis->prefix == PREFIX_REP
1242	&& pRegFrame->ecx <= 0x20
1243	&& pRegFrame->ecx * 4 <= PAGE_SIZE - ((uintptr_t)pvFault & PAGE_OFFSET_MASK)
1244	&& !((uintptr_t)pvFault & 3)
1245	&& (pRegFrame->eax == 0 \|\| pRegFrame->eax == 0x80) /* the two values observed. */
1246	)
1247	{
1248	fValidStosd = true;
1249	pRegFrame->rcx &= 0xffffffff; /* paranoia */
1250	}
1251	else
1252	if ( pDis->mode == CPUMODE_64BIT
1253	&& pDis->prefix == (PREFIX_REP \| PREFIX_REX)
1254	&& pRegFrame->rcx <= 0x20
1255	&& pRegFrame->rcx * 8 <= PAGE_SIZE - ((uintptr_t)pvFault & PAGE_OFFSET_MASK)
1256	&& !((uintptr_t)pvFault & 7)
1257	&& (pRegFrame->rax == 0 \|\| pRegFrame->rax == 0x80) /* the two values observed. */
1258	)
1259	{
1260	fValidStosd = true;
1261	}
1262
1263	if (fValidStosd)
1264	{
1265	rc = pgmPoolAccessHandlerSTOSD(pVM, pPool, pPage, pDis, pRegFrame, GCPhysFault, pvFault);
1266	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,RepStosd), a);
1267	pgmUnlock(pVM);
1268	return rc;
1269	}
1270	}
1271
1272	/* REP prefix, don't bother. */
1273	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,RepPrefix));
1274	Log4(("pgmPoolAccessHandler: eax=%#x ecx=%#x edi=%#x esi=%#x rip=%RGv opcode=%d prefix=%#x\n",
1275	pRegFrame->eax, pRegFrame->ecx, pRegFrame->edi, pRegFrame->esi, (RTGCPTR)pRegFrame->rip, pDis->pCurInstr->opcode, pDis->prefix));
1276	fNotReusedNotForking = true;
1277	}
1278
1279	#if defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT) && defined(IN_RING0)
1280	/* E.g. Windows 7 x64 initializes page tables and touches some pages in the table during the process. This
1281	* leads to pgm pool trashing and an excessive amount of write faults due to page monitoring.
1282	*/
1283	if ( pPage->cModifications >= cMaxModifications
1284	&& !fForcedFlush
1285	&& pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT
1286	&& ( fNotReusedNotForking
1287	\|\| ( !pgmPoolMonitorIsReused(pVM, pVCpu, pRegFrame, pDis, pvFault)
1288	&& !pgmPoolMonitorIsForking(pPool, pDis, GCPhysFault & PAGE_OFFSET_MASK))
1289	)
1290	)
1291	{
1292	Assert(!pgmPoolIsPageLocked(&pVM->pgm.s, pPage));
1293	Assert(pPage->fDirty == false);
1294
1295	/* Flush any monitored duplicates as we will disable write protection. */
1296	if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
1297	\|\| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1298	{
1299	PPGMPOOLPAGE pPageHead = pPage;
1300
1301	/* Find the monitor head. */
1302	while (pPageHead->iMonitoredPrev != NIL_PGMPOOL_IDX)
1303	pPageHead = &pPool->aPages[pPageHead->iMonitoredPrev];
1304
1305	while (pPageHead)
1306	{
1307	unsigned idxNext = pPageHead->iMonitoredNext;
1308
1309	if (pPageHead != pPage)
1310	{
1311	STAM_COUNTER_INC(&pPool->StatDirtyPageDupFlush);
1312	Log(("Flush duplicate page idx=%d GCPhys=%RGp type=%s\n", pPageHead->idx, pPageHead->GCPhys, pgmPoolPoolKindToStr(pPageHead->enmKind)));
1313	int rc2 = pgmPoolFlushPage(pPool, pPageHead);
1314	AssertRC(rc2);
1315	}
1316
1317	if (idxNext == NIL_PGMPOOL_IDX)
1318	break;
1319
1320	pPageHead = &pPool->aPages[idxNext];
1321	}
1322	}
1323
1324	/* The flushing above might fail for locked pages, so double check. */
1325	if ( pPage->iMonitoredNext == NIL_PGMPOOL_IDX
1326	&& pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
1327	{
1328	pgmPoolAddDirtyPage(pVM, pPool, pPage);
1329
1330	/* Temporarily allow write access to the page table again. */
1331	rc = PGMHandlerPhysicalPageTempOff(pVM, pPage->GCPhys, pPage->GCPhys);
1332	if (rc == VINF_SUCCESS)
1333	{
1334	rc = PGMShwModifyPage(pVCpu, pvFault, 1, X86_PTE_RW, ~(uint64_t)X86_PTE_RW);
1335	AssertMsg(rc == VINF_SUCCESS
1336	/* In the SMP case the page table might be removed while we wait for the PGM lock in the trap handler. */
1337	\|\| rc == VERR_PAGE_TABLE_NOT_PRESENT
1338	\|\| rc == VERR_PAGE_NOT_PRESENT,
1339	("PGMShwModifyPage -> GCPtr=%RGv rc=%d\n", pvFault, rc));
1340
1341	pPage->pvDirtyFault = pvFault;
1342
1343	STAM_PROFILE_STOP(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), a);
1344	pgmUnlock(pVM);
1345	return rc;
1346	}
1347	}
1348	}
1349	#endif /* PGMPOOL_WITH_OPTIMIZED_DIRTY_PT */
1350
1351	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FlushModOverflow));
1352	flushPage:
1353	/*
1354	* Not worth it, so flush it.
1355	*
1356	* If we considered it to be reused, don't go back to ring-3
1357	* to emulate failed instructions since we usually cannot
1358	* interpret then. This may be a bit risky, in which case
1359	* the reuse detection must be fixed.
1360	*/
1361	rc = pgmPoolAccessHandlerFlush(pVM, pVCpu, pPool, pPage, pDis, pRegFrame, GCPhysFault, pvFault);
1362	if ( rc == VINF_EM_RAW_EMULATE_INSTR
1363	&& fReused)
1364	{
1365	/* Make sure that the current instruction still has shadow page backing, otherwise we'll end up in a loop. */
1366	if (PGMShwGetPage(pVCpu, pRegFrame->rip, NULL, NULL) == VINF_SUCCESS)
1367	rc = VINF_SUCCESS; /* safe to restart the instruction. */
1368	}
1369	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,FlushPage), a);
1370	pgmUnlock(pVM);
1371	return rc;
1372	}
1373
1374	# endif /* !IN_RING3 */
1375
1376	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1377
1378	# ifdef VBOX_STRICT
1379	/**
1380	* Check references to guest physical memory in a PAE / PAE page table.
1381	*
1382	* @param pPool The pool.
1383	* @param pPage The page.
1384	* @param pShwPT The shadow page table (mapping of the page).
1385	* @param pGstPT The guest page table.
1386	*/
1387	static void pgmPoolTrackCheckPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PTPAE pGstPT)
1388	{
1389	unsigned cErrors = 0;
1390	int LastRc = -1; /* initialized to shut up gcc */
1391	unsigned LastPTE = ~0U; /* initialized to shut up gcc */
1392	RTHCPHYS LastHCPhys = NIL_RTHCPHYS; /* initialized to shut up gcc */
1393
1394	#ifdef VBOX_STRICT
1395	for (unsigned i = 0; i < RT_MIN(RT_ELEMENTS(pShwPT->a), pPage->iFirstPresent); i++)
1396	AssertMsg(!pShwPT->a[i].n.u1Present, ("Unexpected PTE: idx=%d %RX64 (first=%d)\n", i, pShwPT->a[i].u, pPage->iFirstPresent));
1397	#endif
1398	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
1399	{
1400	if (pShwPT->a[i].n.u1Present)
1401	{
1402	RTHCPHYS HCPhys = NIL_RTHCPHYS;
1403	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[i].u & X86_PTE_PAE_PG_MASK, &HCPhys);
1404	if ( rc != VINF_SUCCESS
1405	\|\| (pShwPT->a[i].u & X86_PTE_PAE_PG_MASK) != HCPhys)
1406	{
1407	Log(("rc=%d idx=%d guest %RX64 shw=%RX64 vs %RHp\n", rc, i, pGstPT->a[i].u, pShwPT->a[i].u, HCPhys));
1408	LastPTE = i;
1409	LastRc = rc;
1410	LastHCPhys = HCPhys;
1411	cErrors++;
1412
1413	RTHCPHYS HCPhysPT = NIL_RTHCPHYS;
1414	rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pPage->GCPhys, &HCPhysPT);
1415	AssertRC(rc);
1416
1417	for (unsigned iPage = 0; iPage < pPool->cCurPages; iPage++)
1418	{
1419	PPGMPOOLPAGE pTempPage = &pPool->aPages[iPage];
1420
1421	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1422	{
1423	PX86PTPAE pShwPT2 = (PX86PTPAE)PGMPOOL_PAGE_2_LOCKED_PTR(pPool->CTX_SUFF(pVM), pTempPage);
1424
1425	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
1426	{
1427	if ( pShwPT2->a[j].n.u1Present
1428	&& pShwPT2->a[j].n.u1Write
1429	&& ((pShwPT2->a[j].u & X86_PTE_PAE_PG_MASK) == HCPhysPT))
1430	{
1431	Log(("GCPhys=%RGp idx=%d %RX64 vs %RX64\n", pTempPage->GCPhys, j, pShwPT->a[j].u, pShwPT2->a[j].u));
1432	}
1433	}
1434	}
1435	}
1436	}
1437	}
1438	}
1439	AssertMsg(!cErrors, ("cErrors=%d: last rc=%d idx=%d guest %RX64 shw=%RX64 vs %RHp\n", cErrors, LastRc, LastPTE, pGstPT->a[LastPTE].u, pShwPT->a[LastPTE].u, LastHCPhys));
1440	}
1441	# endif /* VBOX_STRICT */
1442
1443	/**
1444	* Clear references to guest physical memory in a PAE / PAE page table.
1445	*
1446	* @returns nr of changed PTEs
1447	* @param pPool The pool.
1448	* @param pPage The page.
1449	* @param pShwPT The shadow page table (mapping of the page).
1450	* @param pGstPT The guest page table.
1451	* @param pOldGstPT The old cached guest page table.
1452	* @param fAllowRemoval Bail out as soon as we encounter an invalid PTE
1453	* @param pfFlush Flush reused page table (out)
1454	*/
1455	DECLINLINE(unsigned) pgmPoolTrackFlushPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PTPAE pGstPT, PCX86PTPAE pOldGstPT, bool fAllowRemoval, bool *pfFlush)
1456	{
1457	unsigned cChanged = 0;
1458
1459	#ifdef VBOX_STRICT
1460	for (unsigned i = 0; i < RT_MIN(RT_ELEMENTS(pShwPT->a), pPage->iFirstPresent); i++)
1461	AssertMsg(!pShwPT->a[i].n.u1Present, ("Unexpected PTE: idx=%d %RX64 (first=%d)\n", i, pShwPT->a[i].u, pPage->iFirstPresent));
1462	#endif
1463	*pfFlush = false;
1464
1465	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
1466	{
1467	/* Check the new value written by the guest. If present and with a bogus physical address, then
1468	* it's fairly safe to assume the guest is reusing the PT.
1469	*/
1470	if ( fAllowRemoval
1471	&& pGstPT->a[i].n.u1Present)
1472	{
1473	if (!PGMPhysIsGCPhysValid(pPool->CTX_SUFF(pVM), pGstPT->a[i].u & X86_PTE_PAE_PG_MASK))
1474	{
1475	*pfFlush = true;
1476	return ++cChanged;
1477	}
1478	}
1479	if (pShwPT->a[i].n.u1Present)
1480	{
1481	/* If the old cached PTE is identical, then there's no need to flush the shadow copy. */
1482	if ((pGstPT->a[i].u & X86_PTE_PAE_PG_MASK) == (pOldGstPT->a[i].u & X86_PTE_PAE_PG_MASK))
1483	{
1484	#ifdef VBOX_STRICT
1485	RTHCPHYS HCPhys = NIL_RTGCPHYS;
1486	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[i].u & X86_PTE_PAE_PG_MASK, &HCPhys);
1487	AssertMsg(rc == VINF_SUCCESS && (pShwPT->a[i].u & X86_PTE_PAE_PG_MASK) == HCPhys, ("rc=%d guest %RX64 old %RX64 shw=%RX64 vs %RHp\n", rc, pGstPT->a[i].u, pOldGstPT->a[i].u, pShwPT->a[i].u, HCPhys));
1488	#endif
1489	uint64_t uHostAttr = pShwPT->a[i].u & (X86_PTE_P \| X86_PTE_US \| X86_PTE_A \| X86_PTE_D \| X86_PTE_G \| X86_PTE_PAE_NX);
1490	bool fHostRW = !!(pShwPT->a[i].u & X86_PTE_RW);
1491	uint64_t uGuestAttr = pGstPT->a[i].u & (X86_PTE_P \| X86_PTE_US \| X86_PTE_A \| X86_PTE_D \| X86_PTE_G \| X86_PTE_PAE_NX);
1492	bool fGuestRW = !!(pGstPT->a[i].u & X86_PTE_RW);
1493
1494	if ( uHostAttr == uGuestAttr
1495	&& fHostRW <= fGuestRW)
1496	continue;
1497	}
1498	cChanged++;
1499	/* Something was changed, so flush it. */
1500	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX64 hint=%RX64\n",
1501	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pOldGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
1502	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pOldGstPT->a[i].u & X86_PTE_PAE_PG_MASK, i);
1503	ASMAtomicWriteSize(&pShwPT->a[i].u, 0);
1504	}
1505	}
1506	return cChanged;
1507	}
1508
1509
1510	/**
1511	* Flush a dirty page
1512	*
1513	* @param pVM VM Handle.
1514	* @param pPool The pool.
1515	* @param idxSlot Dirty array slot index
1516	* @param fAllowRemoval Allow a reused page table to be removed
1517	*/
1518	static void pgmPoolFlushDirtyPage(PVM pVM, PPGMPOOL pPool, unsigned idxSlot, bool fAllowRemoval = false)
1519	{
1520	PPGMPOOLPAGE pPage;
1521	unsigned idxPage;
1522
1523	Assert(idxSlot < RT_ELEMENTS(pPool->aIdxDirtyPages));
1524	if (pPool->aIdxDirtyPages[idxSlot] == NIL_PGMPOOL_IDX)
1525	return;
1526
1527	idxPage = pPool->aIdxDirtyPages[idxSlot];
1528	AssertRelease(idxPage != NIL_PGMPOOL_IDX);
1529	pPage = &pPool->aPages[idxPage];
1530	Assert(pPage->idx == idxPage);
1531	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1532
1533	AssertMsg(pPage->fDirty, ("Page %RGp (slot=%d) not marked dirty!", pPage->GCPhys, idxSlot));
1534	Log(("Flush dirty page %RGp cMods=%d\n", pPage->GCPhys, pPage->cModifications));
1535
1536	/* First write protect the page again to catch all write accesses. (before checking for changes -> SMP) */
1537	int rc = PGMHandlerPhysicalReset(pVM, pPage->GCPhys);
1538	Assert(rc == VINF_SUCCESS);
1539	pPage->fDirty = false;
1540
1541	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
1542	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(VMMGetCpu(pVM));
1543	#endif
1544
1545	#ifdef VBOX_STRICT
1546	uint64_t fFlags = 0;
1547	RTHCPHYS HCPhys;
1548	rc = PGMShwGetPage(VMMGetCpu(pVM), pPage->pvDirtyFault, &fFlags, &HCPhys);
1549	AssertMsg( ( rc == VINF_SUCCESS
1550	&& (!(fFlags & X86_PTE_RW) \|\| HCPhys != pPage->Core.Key))
1551	/* In the SMP case the page table might be removed while we wait for the PGM lock in the trap handler. */
1552	\|\| rc == VERR_PAGE_TABLE_NOT_PRESENT
1553	\|\| rc == VERR_PAGE_NOT_PRESENT,
1554	("PGMShwGetPage -> GCPtr=%RGv rc=%d flags=%RX64\n", pPage->pvDirtyFault, rc, fFlags));
1555	#endif
1556
1557	/* Flush those PTEs that have changed. */
1558	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
1559	void *pvShw = PGMPOOL_PAGE_2_LOCKED_PTR(pPool->CTX_SUFF(pVM), pPage);
1560	void *pvGst;
1561	bool fFlush;
1562	rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
1563	unsigned cChanges = pgmPoolTrackFlushPTPaePae(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PTPAE)pvGst, (PCX86PTPAE)&pPool->aDirtyPages[idxSlot][0], fAllowRemoval, &fFlush);
1564	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
1565	/** Note: we might want to consider keeping the dirty page active in case there were many changes. */
1566
1567	/* This page is likely to be modified again, so reduce the nr of modifications just a bit here. */
1568	Assert(pPage->cModifications);
1569	if (cChanges < 4)
1570	pPage->cModifications = 1; /* must use > 0 here */
1571	else
1572	pPage->cModifications = RT_MAX(1, pPage->cModifications / 2);
1573
1574	STAM_COUNTER_INC(&pPool->StatResetDirtyPages);
1575	if (pPool->cDirtyPages == RT_ELEMENTS(pPool->aIdxDirtyPages))
1576	pPool->idxFreeDirtyPage = idxSlot;
1577
1578	pPool->cDirtyPages--;
1579	pPool->aIdxDirtyPages[idxSlot] = NIL_PGMPOOL_IDX;
1580	Assert(pPool->cDirtyPages <= RT_ELEMENTS(pPool->aIdxDirtyPages));
1581	if (fFlush)
1582	{
1583	Assert(fAllowRemoval);
1584	Log(("Flush reused page table!\n"));
1585	pgmPoolFlushPage(pPool, pPage);
1586	STAM_COUNTER_INC(&pPool->StatForceFlushReused);
1587	}
1588	else
1589	Log(("Removed dirty page %RGp cMods=%d cChanges=%d\n", pPage->GCPhys, pPage->cModifications, cChanges));
1590
1591	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
1592	PGMDynMapPopAutoSubset(VMMGetCpu(pVM), iPrevSubset);
1593	#endif
1594	}
1595
1596	# ifndef IN_RING3
1597	/**
1598	* Add a new dirty page
1599	*
1600	* @param pVM VM Handle.
1601	* @param pPool The pool.
1602	* @param pPage The page.
1603	*/
1604	void pgmPoolAddDirtyPage(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1605	{
1606	unsigned idxFree;
1607
1608	Assert(PGMIsLocked(pVM));
1609	AssertCompile(RT_ELEMENTS(pPool->aIdxDirtyPages) == 8 \|\| RT_ELEMENTS(pPool->aIdxDirtyPages) == 16);
1610	Assert(!pPage->fDirty);
1611
1612	idxFree = pPool->idxFreeDirtyPage;
1613	Assert(idxFree < RT_ELEMENTS(pPool->aIdxDirtyPages));
1614	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1615
1616	if (pPool->cDirtyPages >= RT_ELEMENTS(pPool->aIdxDirtyPages))
1617	{
1618	STAM_COUNTER_INC(&pPool->StatDirtyPageOverFlowFlush);
1619	pgmPoolFlushDirtyPage(pVM, pPool, idxFree, true /* allow removal of reused page tables*/);
1620	}
1621	Assert(pPool->cDirtyPages < RT_ELEMENTS(pPool->aIdxDirtyPages));
1622	AssertMsg(pPool->aIdxDirtyPages[idxFree] == NIL_PGMPOOL_IDX, ("idxFree=%d cDirtyPages=%d\n", idxFree, pPool->cDirtyPages));
1623
1624	Log(("Add dirty page %RGp (slot=%d)\n", pPage->GCPhys, idxFree));
1625
1626	/* Make a copy of the guest page table as we require valid GCPhys addresses when removing
1627	* references to physical pages. (the HCPhys linear lookup is extremely expensive!)
1628	*/
1629	void *pvShw = PGMPOOL_PAGE_2_LOCKED_PTR(pPool->CTX_SUFF(pVM), pPage);
1630	void *pvGst;
1631	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
1632	memcpy(&pPool->aDirtyPages[idxFree][0], pvGst, PAGE_SIZE);
1633	#ifdef VBOX_STRICT
1634	pgmPoolTrackCheckPTPaePae(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PTPAE)pvGst);
1635	#endif
1636
1637	STAM_COUNTER_INC(&pPool->StatDirtyPage);
1638	pPage->fDirty = true;
1639	pPage->idxDirty = idxFree;
1640	pPool->aIdxDirtyPages[idxFree] = pPage->idx;
1641	pPool->cDirtyPages++;
1642
1643	pPool->idxFreeDirtyPage = (pPool->idxFreeDirtyPage + 1) & (RT_ELEMENTS(pPool->aIdxDirtyPages) - 1);
1644	if ( pPool->cDirtyPages < RT_ELEMENTS(pPool->aIdxDirtyPages)
1645	&& pPool->aIdxDirtyPages[pPool->idxFreeDirtyPage] != NIL_PGMPOOL_IDX)
1646	{
1647	unsigned i;
1648	for (i = 1; i < RT_ELEMENTS(pPool->aIdxDirtyPages); i++)
1649	{
1650	idxFree = (pPool->idxFreeDirtyPage + i) & (RT_ELEMENTS(pPool->aIdxDirtyPages) - 1);
1651	if (pPool->aIdxDirtyPages[idxFree] == NIL_PGMPOOL_IDX)
1652	{
1653	pPool->idxFreeDirtyPage = idxFree;
1654	break;
1655	}
1656	}
1657	Assert(i != RT_ELEMENTS(pPool->aIdxDirtyPages));
1658	}
1659
1660	Assert(pPool->cDirtyPages == RT_ELEMENTS(pPool->aIdxDirtyPages) \|\| pPool->aIdxDirtyPages[pPool->idxFreeDirtyPage] == NIL_PGMPOOL_IDX);
1661	return;
1662	}
1663	# endif /* !IN_RING3 */
1664
1665	/**
1666	* Check if the specified page is dirty (not write monitored)
1667	*
1668	* @return dirty or not
1669	* @param pVM VM Handle.
1670	* @param GCPhys Guest physical address
1671	*/
1672	bool pgmPoolIsDirtyPage(PVM pVM, RTGCPHYS GCPhys)
1673	{
1674	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1675	Assert(PGMIsLocked(pVM));
1676	if (!pPool->cDirtyPages)
1677	return false;
1678
1679	GCPhys = GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1680
1681	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aIdxDirtyPages); i++)
1682	{
1683	if (pPool->aIdxDirtyPages[i] != NIL_PGMPOOL_IDX)
1684	{
1685	PPGMPOOLPAGE pPage;
1686	unsigned idxPage = pPool->aIdxDirtyPages[i];
1687
1688	pPage = &pPool->aPages[idxPage];
1689	if (pPage->GCPhys == GCPhys)
1690	return true;
1691	}
1692	}
1693	return false;
1694	}
1695
1696	/**
1697	* Reset all dirty pages by reinstating page monitoring.
1698	*
1699	* @param pVM VM Handle.
1700	*/
1701	void pgmPoolResetDirtyPages(PVM pVM)
1702	{
1703	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1704	Assert(PGMIsLocked(pVM));
1705	Assert(pPool->cDirtyPages <= RT_ELEMENTS(pPool->aIdxDirtyPages));
1706
1707	if (!pPool->cDirtyPages)
1708	return;
1709
1710	Log(("pgmPoolResetDirtyPages\n"));
1711	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aIdxDirtyPages); i++)
1712	pgmPoolFlushDirtyPage(pVM, pPool, i, true /* allow removal of reused page tables*/);
1713
1714	pPool->idxFreeDirtyPage = 0;
1715	if ( pPool->cDirtyPages != RT_ELEMENTS(pPool->aIdxDirtyPages)
1716	&& pPool->aIdxDirtyPages[pPool->idxFreeDirtyPage] != NIL_PGMPOOL_IDX)
1717	{
1718	unsigned i;
1719	for (i = 1; i < RT_ELEMENTS(pPool->aIdxDirtyPages); i++)
1720	{
1721	if (pPool->aIdxDirtyPages[i] == NIL_PGMPOOL_IDX)
1722	{
1723	pPool->idxFreeDirtyPage = i;
1724	break;
1725	}
1726	}
1727	AssertMsg(i != RT_ELEMENTS(pPool->aIdxDirtyPages), ("cDirtyPages %d", pPool->cDirtyPages));
1728	}
1729
1730	Assert(pPool->aIdxDirtyPages[pPool->idxFreeDirtyPage] == NIL_PGMPOOL_IDX \|\| pPool->cDirtyPages == RT_ELEMENTS(pPool->aIdxDirtyPages));
1731	return;
1732	}
1733
1734	/**
1735	* Reset all dirty pages by reinstating page monitoring.
1736	*
1737	* @param pVM VM Handle.
1738	* @param GCPhysPT Physical address of the page table
1739	*/
1740	void pgmPoolInvalidateDirtyPage(PVM pVM, RTGCPHYS GCPhysPT)
1741	{
1742	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1743	Assert(PGMIsLocked(pVM));
1744	Assert(pPool->cDirtyPages <= RT_ELEMENTS(pPool->aIdxDirtyPages));
1745	unsigned idxDirtyPage = RT_ELEMENTS(pPool->aIdxDirtyPages);
1746
1747	if (!pPool->cDirtyPages)
1748	return;
1749
1750	GCPhysPT = GCPhysPT & ~(RTGCPHYS)(PAGE_SIZE - 1);
1751
1752	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aIdxDirtyPages); i++)
1753	{
1754	if (pPool->aIdxDirtyPages[i] != NIL_PGMPOOL_IDX)
1755	{
1756	unsigned idxPage = pPool->aIdxDirtyPages[i];
1757
1758	PPGMPOOLPAGE pPage = &pPool->aPages[idxPage];
1759	if (pPage->GCPhys == GCPhysPT)
1760	{
1761	idxDirtyPage = i;
1762	break;
1763	}
1764	}
1765	}
1766
1767	if (idxDirtyPage != RT_ELEMENTS(pPool->aIdxDirtyPages))
1768	{
1769	pgmPoolFlushDirtyPage(pVM, pPool, idxDirtyPage, true /* allow removal of reused page tables*/);
1770	if ( pPool->cDirtyPages != RT_ELEMENTS(pPool->aIdxDirtyPages)
1771	&& pPool->aIdxDirtyPages[pPool->idxFreeDirtyPage] != NIL_PGMPOOL_IDX)
1772	{
1773	unsigned i;
1774	for (i = 0; i < RT_ELEMENTS(pPool->aIdxDirtyPages); i++)
1775	{
1776	if (pPool->aIdxDirtyPages[i] == NIL_PGMPOOL_IDX)
1777	{
1778	pPool->idxFreeDirtyPage = i;
1779	break;
1780	}
1781	}
1782	AssertMsg(i != RT_ELEMENTS(pPool->aIdxDirtyPages), ("cDirtyPages %d", pPool->cDirtyPages));
1783	}
1784	}
1785	}
1786
1787	# endif /* PGMPOOL_WITH_OPTIMIZED_DIRTY_PT */
1788
1789	/**
1790	* Inserts a page into the GCPhys hash table.
1791	*
1792	* @param pPool The pool.
1793	* @param pPage The page.
1794	*/
1795	DECLINLINE(void) pgmPoolHashInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1796	{
1797	Log3(("pgmPoolHashInsert: %RGp\n", pPage->GCPhys));
1798	Assert(pPage->GCPhys != NIL_RTGCPHYS); Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1799	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1800	pPage->iNext = pPool->aiHash[iHash];
1801	pPool->aiHash[iHash] = pPage->idx;
1802	}
1803
1804
1805	/**
1806	* Removes a page from the GCPhys hash table.
1807	*
1808	* @param pPool The pool.
1809	* @param pPage The page.
1810	*/
1811	DECLINLINE(void) pgmPoolHashRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1812	{
1813	Log3(("pgmPoolHashRemove: %RGp\n", pPage->GCPhys));
1814	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1815	if (pPool->aiHash[iHash] == pPage->idx)
1816	pPool->aiHash[iHash] = pPage->iNext;
1817	else
1818	{
1819	uint16_t iPrev = pPool->aiHash[iHash];
1820	for (;;)
1821	{
1822	const int16_t i = pPool->aPages[iPrev].iNext;
1823	if (i == pPage->idx)
1824	{
1825	pPool->aPages[iPrev].iNext = pPage->iNext;
1826	break;
1827	}
1828	if (i == NIL_PGMPOOL_IDX)
1829	{
1830	AssertReleaseMsgFailed(("GCPhys=%RGp idx=%#x\n", pPage->GCPhys, pPage->idx));
1831	break;
1832	}
1833	iPrev = i;
1834	}
1835	}
1836	pPage->iNext = NIL_PGMPOOL_IDX;
1837	}
1838
1839
1840	/**
1841	* Frees up one cache page.
1842	*
1843	* @returns VBox status code.
1844	* @retval VINF_SUCCESS on success.
1845	* @param pPool The pool.
1846	* @param iUser The user index.
1847	*/
1848	static int pgmPoolCacheFreeOne(PPGMPOOL pPool, uint16_t iUser)
1849	{
1850	#ifndef IN_RC
1851	const PVM pVM = pPool->CTX_SUFF(pVM);
1852	#endif
1853	Assert(pPool->iAgeHead != pPool->iAgeTail); /* We shouldn't be here if there < 2 cached entries! */
1854	STAM_COUNTER_INC(&pPool->StatCacheFreeUpOne);
1855
1856	/*
1857	* Select one page from the tail of the age list.
1858	*/
1859	PPGMPOOLPAGE pPage;
1860	for (unsigned iLoop = 0; ; iLoop++)
1861	{
1862	uint16_t iToFree = pPool->iAgeTail;
1863	if (iToFree == iUser)
1864	iToFree = pPool->aPages[iToFree].iAgePrev;
1865	/* This is the alternative to the SyncCR3 pgmPoolCacheUsed calls.
1866	if (pPool->aPages[iToFree].iUserHead != NIL_PGMPOOL_USER_INDEX)
1867	{
1868	uint16_t i = pPool->aPages[iToFree].iAgePrev;
1869	for (unsigned j = 0; j < 10 && i != NIL_PGMPOOL_USER_INDEX; j++, i = pPool->aPages[i].iAgePrev)
1870	{
1871	if (pPool->aPages[iToFree].iUserHead == NIL_PGMPOOL_USER_INDEX)
1872	continue;
1873	iToFree = i;
1874	break;
1875	}
1876	}
1877	*/
1878	Assert(iToFree != iUser);
1879	AssertRelease(iToFree != NIL_PGMPOOL_IDX);
1880	pPage = &pPool->aPages[iToFree];
1881
1882	/*
1883	* Reject any attempts at flushing the currently active shadow CR3 mapping.
1884	* Call pgmPoolCacheUsed to move the page to the head of the age list.
1885	*/
1886	if (!pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage))
1887	break;
1888	LogFlow(("pgmPoolCacheFreeOne: refuse CR3 mapping\n"));
1889	pgmPoolCacheUsed(pPool, pPage);
1890	AssertLogRelReturn(iLoop < 8192, VERR_INTERNAL_ERROR);
1891	}
1892
1893	/*
1894	* Found a usable page, flush it and return.
1895	*/
1896	int rc = pgmPoolFlushPage(pPool, pPage);
1897	/* This flush was initiated by us and not the guest, so explicitly flush the TLB. */
1898	/* todo: find out why this is necessary; pgmPoolFlushPage should trigger a flush if one is really needed. */
1899	if (rc == VINF_SUCCESS)
1900	PGM_INVL_ALL_VCPU_TLBS(pVM);
1901	return rc;
1902	}
1903
1904
1905	/**
1906	* Checks if a kind mismatch is really a page being reused
1907	* or if it's just normal remappings.
1908	*
1909	* @returns true if reused and the cached page (enmKind1) should be flushed
1910	* @returns false if not reused.
1911	* @param enmKind1 The kind of the cached page.
1912	* @param enmKind2 The kind of the requested page.
1913	*/
1914	static bool pgmPoolCacheReusedByKind(PGMPOOLKIND enmKind1, PGMPOOLKIND enmKind2)
1915	{
1916	switch (enmKind1)
1917	{
1918	/*
1919	* Never reuse them. There is no remapping in non-paging mode.
1920	*/
1921	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1922	case PGMPOOLKIND_32BIT_PD_PHYS:
1923	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1924	case PGMPOOLKIND_PAE_PD_PHYS:
1925	case PGMPOOLKIND_PAE_PDPT_PHYS:
1926	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1927	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1928	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1929	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1930	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1931	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT: /* never reuse them for other types */
1932	return false;
1933
1934	/*
1935	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1936	*/
1937	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1938	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1939	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1940	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1941	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1942	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1943	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1944	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1945	case PGMPOOLKIND_32BIT_PD:
1946	case PGMPOOLKIND_PAE_PDPT:
1947	switch (enmKind2)
1948	{
1949	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1950	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1951	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1952	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1953	case PGMPOOLKIND_64BIT_PML4:
1954	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1955	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1956	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1957	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1958	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1959	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1960	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1961	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1962	return true;
1963	default:
1964	return false;
1965	}
1966
1967	/*
1968	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1969	*/
1970	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1971	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1972	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1973	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1974	case PGMPOOLKIND_64BIT_PML4:
1975	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1976	switch (enmKind2)
1977	{
1978	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1979	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1980	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1981	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1982	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
1983	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
1984	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
1985	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
1986	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1987	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1988	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1989	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1990	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1991	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1992	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1993	return true;
1994	default:
1995	return false;
1996	}
1997
1998	/*
1999	* These cannot be flushed, and it's common to reuse the PDs as PTs.
2000	*/
2001	case PGMPOOLKIND_ROOT_NESTED:
2002	return false;
2003
2004	default:
2005	AssertFatalMsgFailed(("enmKind1=%d\n", enmKind1));
2006	}
2007	}
2008
2009
2010	/**
2011	* Attempts to satisfy a pgmPoolAlloc request from the cache.
2012	*
2013	* @returns VBox status code.
2014	* @retval VINF_PGM_CACHED_PAGE on success.
2015	* @retval VERR_FILE_NOT_FOUND if not found.
2016	* @param pPool The pool.
2017	* @param GCPhys The GC physical address of the page we're gonna shadow.
2018	* @param enmKind The kind of mapping.
2019	* @param enmAccess Access type for the mapping (only relevant for big pages)
2020	* @param iUser The shadow page pool index of the user table.
2021	* @param iUserTable The index into the user table (shadowed).
2022	* @param ppPage Where to store the pointer to the page.
2023	*/
2024	static int pgmPoolCacheAlloc(PPGMPOOL pPool, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, PGMPOOLACCESS enmAccess, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
2025	{
2026	#ifndef IN_RC
2027	const PVM pVM = pPool->CTX_SUFF(pVM);
2028	#endif
2029	/*
2030	* Look up the GCPhys in the hash.
2031	*/
2032	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
2033	Log3(("pgmPoolCacheAlloc: %RGp kind %s iUser=%x iUserTable=%x SLOT=%d\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable, i));
2034	if (i != NIL_PGMPOOL_IDX)
2035	{
2036	do
2037	{
2038	PPGMPOOLPAGE pPage = &pPool->aPages[i];
2039	Log4(("pgmPoolCacheAlloc: slot %d found page %RGp\n", i, pPage->GCPhys));
2040	if (pPage->GCPhys == GCPhys)
2041	{
2042	if ( (PGMPOOLKIND)pPage->enmKind == enmKind
2043	&& (PGMPOOLACCESS)pPage->enmAccess == enmAccess)
2044	{
2045	/* Put it at the start of the use list to make sure pgmPoolTrackAddUser
2046	* doesn't flush it in case there are no more free use records.
2047	*/
2048	pgmPoolCacheUsed(pPool, pPage);
2049
2050	int rc = pgmPoolTrackAddUser(pPool, pPage, iUser, iUserTable);
2051	if (RT_SUCCESS(rc))
2052	{
2053	Assert((PGMPOOLKIND)pPage->enmKind == enmKind);
2054	*ppPage = pPage;
2055	if (pPage->cModifications)
2056	pPage->cModifications = 1; /* reset counter (can't use 0, or else it will be reinserted in the modified list) */
2057	STAM_COUNTER_INC(&pPool->StatCacheHits);
2058	return VINF_PGM_CACHED_PAGE;
2059	}
2060	return rc;
2061	}
2062
2063	if ((PGMPOOLKIND)pPage->enmKind != enmKind)
2064	{
2065	/*
2066	* The kind is different. In some cases we should now flush the page
2067	* as it has been reused, but in most cases this is normal remapping
2068	* of PDs as PT or big pages using the GCPhys field in a slightly
2069	* different way than the other kinds.
2070	*/
2071	if (pgmPoolCacheReusedByKind((PGMPOOLKIND)pPage->enmKind, enmKind))
2072	{
2073	STAM_COUNTER_INC(&pPool->StatCacheKindMismatches);
2074	pgmPoolFlushPage(pPool, pPage);
2075	break;
2076	}
2077	}
2078	}
2079
2080	/* next */
2081	i = pPage->iNext;
2082	} while (i != NIL_PGMPOOL_IDX);
2083	}
2084
2085	Log3(("pgmPoolCacheAlloc: Missed GCPhys=%RGp enmKind=%s\n", GCPhys, pgmPoolPoolKindToStr(enmKind)));
2086	STAM_COUNTER_INC(&pPool->StatCacheMisses);
2087	return VERR_FILE_NOT_FOUND;
2088	}
2089
2090
2091	/**
2092	* Inserts a page into the cache.
2093	*
2094	* @param pPool The pool.
2095	* @param pPage The cached page.
2096	* @param fCanBeCached Set if the page is fit for caching from the caller's point of view.
2097	*/
2098	static void pgmPoolCacheInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fCanBeCached)
2099	{
2100	/*
2101	* Insert into the GCPhys hash if the page is fit for that.
2102	*/
2103	Assert(!pPage->fCached);
2104	if (fCanBeCached)
2105	{
2106	pPage->fCached = true;
2107	pgmPoolHashInsert(pPool, pPage);
2108	Log3(("pgmPoolCacheInsert: Caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
2109	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
2110	STAM_COUNTER_INC(&pPool->StatCacheCacheable);
2111	}
2112	else
2113	{
2114	Log3(("pgmPoolCacheInsert: Not caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
2115	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
2116	STAM_COUNTER_INC(&pPool->StatCacheUncacheable);
2117	}
2118
2119	/*
2120	* Insert at the head of the age list.
2121	*/
2122	pPage->iAgePrev = NIL_PGMPOOL_IDX;
2123	pPage->iAgeNext = pPool->iAgeHead;
2124	if (pPool->iAgeHead != NIL_PGMPOOL_IDX)
2125	pPool->aPages[pPool->iAgeHead].iAgePrev = pPage->idx;
2126	else
2127	pPool->iAgeTail = pPage->idx;
2128	pPool->iAgeHead = pPage->idx;
2129	}
2130
2131
2132	/**
2133	* Flushes a cached page.
2134	*
2135	* @param pPool The pool.
2136	* @param pPage The cached page.
2137	*/
2138	static void pgmPoolCacheFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2139	{
2140	Log3(("pgmPoolCacheFlushPage: %RGp\n", pPage->GCPhys));
2141
2142	/*
2143	* Remove the page from the hash.
2144	*/
2145	if (pPage->fCached)
2146	{
2147	pPage->fCached = false;
2148	pgmPoolHashRemove(pPool, pPage);
2149	}
2150	else
2151	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
2152
2153	/*
2154	* Remove it from the age list.
2155	*/
2156	if (pPage->iAgeNext != NIL_PGMPOOL_IDX)
2157	pPool->aPages[pPage->iAgeNext].iAgePrev = pPage->iAgePrev;
2158	else
2159	pPool->iAgeTail = pPage->iAgePrev;
2160	if (pPage->iAgePrev != NIL_PGMPOOL_IDX)
2161	pPool->aPages[pPage->iAgePrev].iAgeNext = pPage->iAgeNext;
2162	else
2163	pPool->iAgeHead = pPage->iAgeNext;
2164	pPage->iAgeNext = NIL_PGMPOOL_IDX;
2165	pPage->iAgePrev = NIL_PGMPOOL_IDX;
2166	}
2167
2168
2169	/**
2170	* Looks for pages sharing the monitor.
2171	*
2172	* @returns Pointer to the head page.
2173	* @returns NULL if not found.
2174	* @param pPool The Pool
2175	* @param pNewPage The page which is going to be monitored.
2176	*/
2177	static PPGMPOOLPAGE pgmPoolMonitorGetPageByGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pNewPage)
2178	{
2179	/*
2180	* Look up the GCPhys in the hash.
2181	*/
2182	RTGCPHYS GCPhys = pNewPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
2183	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
2184	if (i == NIL_PGMPOOL_IDX)
2185	return NULL;
2186	do
2187	{
2188	PPGMPOOLPAGE pPage = &pPool->aPages[i];
2189	if ( pPage->GCPhys - GCPhys < PAGE_SIZE
2190	&& pPage != pNewPage)
2191	{
2192	switch (pPage->enmKind)
2193	{
2194	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2195	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2196	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2197	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2198	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2199	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2200	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2201	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2202	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2203	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2204	case PGMPOOLKIND_64BIT_PML4:
2205	case PGMPOOLKIND_32BIT_PD:
2206	case PGMPOOLKIND_PAE_PDPT:
2207	{
2208	/* find the head */
2209	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
2210	{
2211	Assert(pPage->iMonitoredPrev != pPage->idx);
2212	pPage = &pPool->aPages[pPage->iMonitoredPrev];
2213	}
2214	return pPage;
2215	}
2216
2217	/* ignore, no monitoring. */
2218	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2219	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2220	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2221	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2222	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2223	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2224	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2225	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2226	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2227	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2228	case PGMPOOLKIND_ROOT_NESTED:
2229	case PGMPOOLKIND_PAE_PD_PHYS:
2230	case PGMPOOLKIND_PAE_PDPT_PHYS:
2231	case PGMPOOLKIND_32BIT_PD_PHYS:
2232	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2233	break;
2234	default:
2235	AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
2236	}
2237	}
2238
2239	/* next */
2240	i = pPage->iNext;
2241	} while (i != NIL_PGMPOOL_IDX);
2242	return NULL;
2243	}
2244
2245
2246	/**
2247	* Enabled write monitoring of a guest page.
2248	*
2249	* @returns VBox status code.
2250	* @retval VINF_SUCCESS on success.
2251	* @param pPool The pool.
2252	* @param pPage The cached page.
2253	*/
2254	static int pgmPoolMonitorInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2255	{
2256	LogFlow(("pgmPoolMonitorInsert %RGp\n", pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1)));
2257
2258	/*
2259	* Filter out the relevant kinds.
2260	*/
2261	switch (pPage->enmKind)
2262	{
2263	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2264	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2265	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2266	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2267	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2268	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2269	case PGMPOOLKIND_64BIT_PML4:
2270	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2271	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2272	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2273	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2274	case PGMPOOLKIND_32BIT_PD:
2275	case PGMPOOLKIND_PAE_PDPT:
2276	break;
2277
2278	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2279	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2280	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2281	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2282	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2283	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2284	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2285	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2286	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2287	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2288	case PGMPOOLKIND_ROOT_NESTED:
2289	/* Nothing to monitor here. */
2290	return VINF_SUCCESS;
2291
2292	case PGMPOOLKIND_32BIT_PD_PHYS:
2293	case PGMPOOLKIND_PAE_PDPT_PHYS:
2294	case PGMPOOLKIND_PAE_PD_PHYS:
2295	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2296	/* Nothing to monitor here. */
2297	return VINF_SUCCESS;
2298	default:
2299	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
2300	}
2301
2302	/*
2303	* Install handler.
2304	*/
2305	int rc;
2306	PPGMPOOLPAGE pPageHead = pgmPoolMonitorGetPageByGCPhys(pPool, pPage);
2307	if (pPageHead)
2308	{
2309	Assert(pPageHead != pPage); Assert(pPageHead->iMonitoredNext != pPage->idx);
2310	Assert(pPageHead->iMonitoredPrev != pPage->idx);
2311
2312	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
2313	if (pPageHead->fDirty)
2314	pgmPoolFlushDirtyPage(pPool->CTX_SUFF(pVM), pPool, pPageHead->idxDirty, false /* do not remove */);
2315	#endif
2316
2317	pPage->iMonitoredPrev = pPageHead->idx;
2318	pPage->iMonitoredNext = pPageHead->iMonitoredNext;
2319	if (pPageHead->iMonitoredNext != NIL_PGMPOOL_IDX)
2320	pPool->aPages[pPageHead->iMonitoredNext].iMonitoredPrev = pPage->idx;
2321	pPageHead->iMonitoredNext = pPage->idx;
2322	rc = VINF_SUCCESS;
2323	}
2324	else
2325	{
2326	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX); Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
2327	PVM pVM = pPool->CTX_SUFF(pVM);
2328	const RTGCPHYS GCPhysPage = pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
2329	rc = PGMHandlerPhysicalRegisterEx(pVM, PGMPHYSHANDLERTYPE_PHYSICAL_WRITE,
2330	GCPhysPage, GCPhysPage + (PAGE_SIZE - 1),
2331	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
2332	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
2333	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
2334	pPool->pszAccessHandler);
2335	/** @todo we should probably deal with out-of-memory conditions here, but for now increasing
2336	* the heap size should suffice. */
2337	AssertFatalMsgRC(rc, ("PGMHandlerPhysicalRegisterEx %RGp failed with %Rrc\n", GCPhysPage, rc));
2338	PVMCPU pVCpu = VMMGetCpu(pVM);
2339	AssertFatalMsg(!(pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3), ("fSyncFlags=%x syncff=%d\n", pVCpu->pgm.s.fSyncFlags, VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3)));
2340	}
2341	pPage->fMonitored = true;
2342	return rc;
2343	}
2344
2345
2346	/**
2347	* Disables write monitoring of a guest page.
2348	*
2349	* @returns VBox status code.
2350	* @retval VINF_SUCCESS on success.
2351	* @param pPool The pool.
2352	* @param pPage The cached page.
2353	*/
2354	static int pgmPoolMonitorFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2355	{
2356	/*
2357	* Filter out the relevant kinds.
2358	*/
2359	switch (pPage->enmKind)
2360	{
2361	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2362	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2363	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2364	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2365	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2366	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2367	case PGMPOOLKIND_64BIT_PML4:
2368	case PGMPOOLKIND_32BIT_PD:
2369	case PGMPOOLKIND_PAE_PDPT:
2370	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2371	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2372	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2373	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2374	break;
2375
2376	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2377	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2378	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2379	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2380	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2381	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2382	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2383	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2384	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2385	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2386	case PGMPOOLKIND_ROOT_NESTED:
2387	case PGMPOOLKIND_PAE_PD_PHYS:
2388	case PGMPOOLKIND_PAE_PDPT_PHYS:
2389	case PGMPOOLKIND_32BIT_PD_PHYS:
2390	/* Nothing to monitor here. */
2391	Assert(!pPage->fMonitored);
2392	return VINF_SUCCESS;
2393
2394	default:
2395	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
2396	}
2397	Assert(pPage->fMonitored);
2398
2399	/*
2400	* Remove the page from the monitored list or uninstall it if last.
2401	*/
2402	const PVM pVM = pPool->CTX_SUFF(pVM);
2403	int rc;
2404	if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
2405	\|\| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
2406	{
2407	if (pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
2408	{
2409	PPGMPOOLPAGE pNewHead = &pPool->aPages[pPage->iMonitoredNext];
2410	pNewHead->iMonitoredPrev = NIL_PGMPOOL_IDX;
2411	rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
2412	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pNewHead),
2413	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pNewHead),
2414	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pNewHead),
2415	pPool->pszAccessHandler);
2416	AssertFatalRCSuccess(rc);
2417	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
2418	}
2419	else
2420	{
2421	pPool->aPages[pPage->iMonitoredPrev].iMonitoredNext = pPage->iMonitoredNext;
2422	if (pPage->iMonitoredNext != NIL_PGMPOOL_IDX)
2423	{
2424	pPool->aPages[pPage->iMonitoredNext].iMonitoredPrev = pPage->iMonitoredPrev;
2425	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
2426	}
2427	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
2428	rc = VINF_SUCCESS;
2429	}
2430	}
2431	else
2432	{
2433	rc = PGMHandlerPhysicalDeregister(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1));
2434	AssertFatalRC(rc);
2435	PVMCPU pVCpu = VMMGetCpu(pVM);
2436	AssertFatalMsg(!(pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3),
2437	("%#x %#x\n", pVCpu->pgm.s.fSyncFlags, pVM->fGlobalForcedActions));
2438	}
2439	pPage->fMonitored = false;
2440
2441	/*
2442	* Remove it from the list of modified pages (if in it).
2443	*/
2444	pgmPoolMonitorModifiedRemove(pPool, pPage);
2445
2446	return rc;
2447	}
2448
2449
2450	/**
2451	* Inserts the page into the list of modified pages.
2452	*
2453	* @param pPool The pool.
2454	* @param pPage The page.
2455	*/
2456	void pgmPoolMonitorModifiedInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2457	{
2458	Log3(("pgmPoolMonitorModifiedInsert: idx=%d\n", pPage->idx));
2459	AssertMsg( pPage->iModifiedNext == NIL_PGMPOOL_IDX
2460	&& pPage->iModifiedPrev == NIL_PGMPOOL_IDX
2461	&& pPool->iModifiedHead != pPage->idx,
2462	("Next=%d Prev=%d idx=%d cModifications=%d Head=%d cModifiedPages=%d\n",
2463	pPage->iModifiedNext, pPage->iModifiedPrev, pPage->idx, pPage->cModifications,
2464	pPool->iModifiedHead, pPool->cModifiedPages));
2465
2466	pPage->iModifiedNext = pPool->iModifiedHead;
2467	if (pPool->iModifiedHead != NIL_PGMPOOL_IDX)
2468	pPool->aPages[pPool->iModifiedHead].iModifiedPrev = pPage->idx;
2469	pPool->iModifiedHead = pPage->idx;
2470	pPool->cModifiedPages++;
2471	#ifdef VBOX_WITH_STATISTICS
2472	if (pPool->cModifiedPages > pPool->cModifiedPagesHigh)
2473	pPool->cModifiedPagesHigh = pPool->cModifiedPages;
2474	#endif
2475	}
2476
2477
2478	/**
2479	* Removes the page from the list of modified pages and resets the
2480	* moficiation counter.
2481	*
2482	* @param pPool The pool.
2483	* @param pPage The page which is believed to be in the list of modified pages.
2484	*/
2485	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2486	{
2487	Log3(("pgmPoolMonitorModifiedRemove: idx=%d cModifications=%d\n", pPage->idx, pPage->cModifications));
2488	if (pPool->iModifiedHead == pPage->idx)
2489	{
2490	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
2491	pPool->iModifiedHead = pPage->iModifiedNext;
2492	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
2493	{
2494	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = NIL_PGMPOOL_IDX;
2495	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2496	}
2497	pPool->cModifiedPages--;
2498	}
2499	else if (pPage->iModifiedPrev != NIL_PGMPOOL_IDX)
2500	{
2501	pPool->aPages[pPage->iModifiedPrev].iModifiedNext = pPage->iModifiedNext;
2502	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
2503	{
2504	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = pPage->iModifiedPrev;
2505	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2506	}
2507	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2508	pPool->cModifiedPages--;
2509	}
2510	else
2511	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
2512	pPage->cModifications = 0;
2513	}
2514
2515
2516	/**
2517	* Zaps the list of modified pages, resetting their modification counters in the process.
2518	*
2519	* @param pVM The VM handle.
2520	*/
2521	static void pgmPoolMonitorModifiedClearAll(PVM pVM)
2522	{
2523	pgmLock(pVM);
2524	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2525	LogFlow(("pgmPoolMonitorModifiedClearAll: cModifiedPages=%d\n", pPool->cModifiedPages));
2526
2527	unsigned cPages = 0; NOREF(cPages);
2528
2529	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
2530	pgmPoolResetDirtyPages(pVM);
2531	#endif
2532
2533	uint16_t idx = pPool->iModifiedHead;
2534	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
2535	while (idx != NIL_PGMPOOL_IDX)
2536	{
2537	PPGMPOOLPAGE pPage = &pPool->aPages[idx];
2538	idx = pPage->iModifiedNext;
2539	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2540	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2541	pPage->cModifications = 0;
2542	Assert(++cPages);
2543	}
2544	AssertMsg(cPages == pPool->cModifiedPages, ("%d != %d\n", cPages, pPool->cModifiedPages));
2545	pPool->cModifiedPages = 0;
2546	pgmUnlock(pVM);
2547	}
2548
2549
2550	/**
2551	* Handle SyncCR3 pool tasks
2552	*
2553	* @returns VBox status code.
2554	* @retval VINF_SUCCESS if successfully added.
2555	* @retval VINF_PGM_SYNC_CR3 is it needs to be deferred to ring 3 (GC only)
2556	* @param pVCpu The VMCPU handle.
2557	* @remark Should only be used when monitoring is available, thus placed in
2558	* the PGMPOOL_WITH_MONITORING #ifdef.
2559	*/
2560	int pgmPoolSyncCR3(PVMCPU pVCpu)
2561	{
2562	PVM pVM = pVCpu->CTX_SUFF(pVM);
2563	LogFlow(("pgmPoolSyncCR3\n"));
2564
2565	/*
2566	* When monitoring shadowed pages, we reset the modification counters on CR3 sync.
2567	* Occasionally we will have to clear all the shadow page tables because we wanted
2568	* to monitor a page which was mapped by too many shadowed page tables. This operation
2569	* sometimes refered to as a 'lightweight flush'.
2570	*/
2571	# ifdef IN_RING3 /* Don't flush in ring-0 or raw mode, it's taking too long. */
2572	if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
2573	pgmR3PoolClearAll(pVM);
2574	# else /* !IN_RING3 */
2575	if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
2576	{
2577	Log(("SyncCR3: PGM_SYNC_CLEAR_PGM_POOL is set -> VINF_PGM_SYNC_CR3\n"));
2578	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
2579
2580	/* Make sure all other VCPUs return to ring 3. */
2581	if (pVM->cCpus > 1)
2582	{
2583	VM_FF_SET(pVM, VM_FF_PGM_POOL_FLUSH_PENDING);
2584	PGM_INVL_ALL_VCPU_TLBS(pVM);
2585	}
2586	return VINF_PGM_SYNC_CR3;
2587	}
2588	# endif /* !IN_RING3 */
2589	else
2590	{
2591	pgmPoolMonitorModifiedClearAll(pVM);
2592
2593	/* pgmPoolMonitorModifiedClearAll can cause a pgm pool flush (dirty page clearing), so make sure we handle this! */
2594	if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
2595	{
2596	Log(("pgmPoolMonitorModifiedClearAll caused a pgm flush -> call pgmPoolSyncCR3 again!\n"));
2597	return pgmPoolSyncCR3(pVCpu);
2598	}
2599	}
2600	return VINF_SUCCESS;
2601	}
2602
2603
2604	/**
2605	* Frees up at least one user entry.
2606	*
2607	* @returns VBox status code.
2608	* @retval VINF_SUCCESS if successfully added.
2609	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2610	* @param pPool The pool.
2611	* @param iUser The user index.
2612	*/
2613	static int pgmPoolTrackFreeOneUser(PPGMPOOL pPool, uint16_t iUser)
2614	{
2615	STAM_COUNTER_INC(&pPool->StatTrackFreeUpOneUser);
2616	/*
2617	* Just free cached pages in a braindead fashion.
2618	*/
2619	/** @todo walk the age list backwards and free the first with usage. */
2620	int rc = VINF_SUCCESS;
2621	do
2622	{
2623	int rc2 = pgmPoolCacheFreeOne(pPool, iUser);
2624	if (RT_FAILURE(rc2) && rc == VINF_SUCCESS)
2625	rc = rc2;
2626	} while (pPool->iUserFreeHead == NIL_PGMPOOL_USER_INDEX);
2627	return rc;
2628	}
2629
2630
2631	/**
2632	* Inserts a page into the cache.
2633	*
2634	* This will create user node for the page, insert it into the GCPhys
2635	* hash, and insert it into the age list.
2636	*
2637	* @returns VBox status code.
2638	* @retval VINF_SUCCESS if successfully added.
2639	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2640	* @param pPool The pool.
2641	* @param pPage The cached page.
2642	* @param GCPhys The GC physical address of the page we're gonna shadow.
2643	* @param iUser The user index.
2644	* @param iUserTable The user table index.
2645	*/
2646	DECLINLINE(int) pgmPoolTrackInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhys, uint16_t iUser, uint32_t iUserTable)
2647	{
2648	int rc = VINF_SUCCESS;
2649	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2650
2651	LogFlow(("pgmPoolTrackInsert GCPhys=%RGp iUser %x iUserTable %x\n", GCPhys, iUser, iUserTable));
2652
2653	#ifdef VBOX_STRICT
2654	/*
2655	* Check that the entry doesn't already exists.
2656	*/
2657	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2658	{
2659	uint16_t i = pPage->iUserHead;
2660	do
2661	{
2662	Assert(i < pPool->cMaxUsers);
2663	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2664	i = paUsers[i].iNext;
2665	} while (i != NIL_PGMPOOL_USER_INDEX);
2666	}
2667	#endif
2668
2669	/*
2670	* Find free a user node.
2671	*/
2672	uint16_t i = pPool->iUserFreeHead;
2673	if (i == NIL_PGMPOOL_USER_INDEX)
2674	{
2675	rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2676	if (RT_FAILURE(rc))
2677	return rc;
2678	i = pPool->iUserFreeHead;
2679	}
2680
2681	/*
2682	* Unlink the user node from the free list,
2683	* initialize and insert it into the user list.
2684	*/
2685	pPool->iUserFreeHead = paUsers[i].iNext;
2686	paUsers[i].iNext = NIL_PGMPOOL_USER_INDEX;
2687	paUsers[i].iUser = iUser;
2688	paUsers[i].iUserTable = iUserTable;
2689	pPage->iUserHead = i;
2690
2691	/*
2692	* Insert into cache and enable monitoring of the guest page if enabled.
2693	*
2694	* Until we implement caching of all levels, including the CR3 one, we'll
2695	* have to make sure we don't try monitor & cache any recursive reuse of
2696	* a monitored CR3 page. Because all windows versions are doing this we'll
2697	* have to be able to do combined access monitoring, CR3 + PT and
2698	* PD + PT (guest PAE).
2699	*
2700	* Update:
2701	* We're now cooperating with the CR3 monitor if an uncachable page is found.
2702	*/
2703	const bool fCanBeMonitored = true;
2704	pgmPoolCacheInsert(pPool, pPage, fCanBeMonitored); /* This can be expanded. */
2705	if (fCanBeMonitored)
2706	{
2707	rc = pgmPoolMonitorInsert(pPool, pPage);
2708	AssertRC(rc);
2709	}
2710	return rc;
2711	}
2712
2713
2714	/**
2715	* Adds a user reference to a page.
2716	*
2717	* This will move the page to the head of the
2718	*
2719	* @returns VBox status code.
2720	* @retval VINF_SUCCESS if successfully added.
2721	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2722	* @param pPool The pool.
2723	* @param pPage The cached page.
2724	* @param iUser The user index.
2725	* @param iUserTable The user table.
2726	*/
2727	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2728	{
2729	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2730
2731	Log3(("pgmPoolTrackAddUser GCPhys = %RGp iUser %x iUserTable %x\n", pPage->GCPhys, iUser, iUserTable));
2732
2733	# ifdef VBOX_STRICT
2734	/*
2735	* Check that the entry doesn't already exists. We only allow multiple users of top-level paging structures (SHW_POOL_ROOT_IDX).
2736	*/
2737	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2738	{
2739	uint16_t i = pPage->iUserHead;
2740	do
2741	{
2742	Assert(i < pPool->cMaxUsers);
2743	AssertMsg(iUser != PGMPOOL_IDX_PD \|\| iUser != PGMPOOL_IDX_PDPT \|\| iUser != PGMPOOL_IDX_NESTED_ROOT \|\| iUser != PGMPOOL_IDX_AMD64_CR3 \|\|
2744	paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2745	i = paUsers[i].iNext;
2746	} while (i != NIL_PGMPOOL_USER_INDEX);
2747	}
2748	# endif
2749
2750	/*
2751	* Allocate a user node.
2752	*/
2753	uint16_t i = pPool->iUserFreeHead;
2754	if (i == NIL_PGMPOOL_USER_INDEX)
2755	{
2756	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2757	if (RT_FAILURE(rc))
2758	return rc;
2759	i = pPool->iUserFreeHead;
2760	}
2761	pPool->iUserFreeHead = paUsers[i].iNext;
2762
2763	/*
2764	* Initialize the user node and insert it.
2765	*/
2766	paUsers[i].iNext = pPage->iUserHead;
2767	paUsers[i].iUser = iUser;
2768	paUsers[i].iUserTable = iUserTable;
2769	pPage->iUserHead = i;
2770
2771	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
2772	if (pPage->fDirty)
2773	pgmPoolFlushDirtyPage(pPool->CTX_SUFF(pVM), pPool, pPage->idxDirty, false /* do not remove */);
2774	# endif
2775
2776	/*
2777	* Tell the cache to update its replacement stats for this page.
2778	*/
2779	pgmPoolCacheUsed(pPool, pPage);
2780	return VINF_SUCCESS;
2781	}
2782
2783
2784	/**
2785	* Frees a user record associated with a page.
2786	*
2787	* This does not clear the entry in the user table, it simply replaces the
2788	* user record to the chain of free records.
2789	*
2790	* @param pPool The pool.
2791	* @param HCPhys The HC physical address of the shadow page.
2792	* @param iUser The shadow page pool index of the user table.
2793	* @param iUserTable The index into the user table (shadowed).
2794	*/
2795	static void pgmPoolTrackFreeUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2796	{
2797	/*
2798	* Unlink and free the specified user entry.
2799	*/
2800	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2801
2802	Log3(("pgmPoolTrackFreeUser %RGp %x %x\n", pPage->GCPhys, iUser, iUserTable));
2803	/* Special: For PAE and 32-bit paging, there is usually no more than one user. */
2804	uint16_t i = pPage->iUserHead;
2805	if ( i != NIL_PGMPOOL_USER_INDEX
2806	&& paUsers[i].iUser == iUser
2807	&& paUsers[i].iUserTable == iUserTable)
2808	{
2809	pPage->iUserHead = paUsers[i].iNext;
2810
2811	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2812	paUsers[i].iNext = pPool->iUserFreeHead;
2813	pPool->iUserFreeHead = i;
2814	return;
2815	}
2816
2817	/* General: Linear search. */
2818	uint16_t iPrev = NIL_PGMPOOL_USER_INDEX;
2819	while (i != NIL_PGMPOOL_USER_INDEX)
2820	{
2821	if ( paUsers[i].iUser == iUser
2822	&& paUsers[i].iUserTable == iUserTable)
2823	{
2824	if (iPrev != NIL_PGMPOOL_USER_INDEX)
2825	paUsers[iPrev].iNext = paUsers[i].iNext;
2826	else
2827	pPage->iUserHead = paUsers[i].iNext;
2828
2829	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2830	paUsers[i].iNext = pPool->iUserFreeHead;
2831	pPool->iUserFreeHead = i;
2832	return;
2833	}
2834	iPrev = i;
2835	i = paUsers[i].iNext;
2836	}
2837
2838	/* Fatal: didn't find it */
2839	AssertFatalMsgFailed(("Didn't find the user entry! iUser=%#x iUserTable=%#x GCPhys=%RGp\n",
2840	iUser, iUserTable, pPage->GCPhys));
2841	}
2842
2843
2844	/**
2845	* Gets the entry size of a shadow table.
2846	*
2847	* @param enmKind The kind of page.
2848	*
2849	* @returns The size of the entry in bytes. That is, 4 or 8.
2850	* @returns If the kind is not for a table, an assertion is raised and 0 is
2851	* returned.
2852	*/
2853	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind)
2854	{
2855	switch (enmKind)
2856	{
2857	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2858	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2859	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2860	case PGMPOOLKIND_32BIT_PD:
2861	case PGMPOOLKIND_32BIT_PD_PHYS:
2862	return 4;
2863
2864	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2865	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2866	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2867	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2868	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2869	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2870	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2871	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2872	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2873	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2874	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2875	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2876	case PGMPOOLKIND_64BIT_PML4:
2877	case PGMPOOLKIND_PAE_PDPT:
2878	case PGMPOOLKIND_ROOT_NESTED:
2879	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2880	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2881	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2882	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2883	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2884	case PGMPOOLKIND_PAE_PD_PHYS:
2885	case PGMPOOLKIND_PAE_PDPT_PHYS:
2886	return 8;
2887
2888	default:
2889	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2890	}
2891	}
2892
2893
2894	/**
2895	* Gets the entry size of a guest table.
2896	*
2897	* @param enmKind The kind of page.
2898	*
2899	* @returns The size of the entry in bytes. That is, 0, 4 or 8.
2900	* @returns If the kind is not for a table, an assertion is raised and 0 is
2901	* returned.
2902	*/
2903	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind)
2904	{
2905	switch (enmKind)
2906	{
2907	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2908	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2909	case PGMPOOLKIND_32BIT_PD:
2910	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2911	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2912	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2913	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2914	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2915	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2916	return 4;
2917
2918	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2919	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2920	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2921	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2922	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2923	case PGMPOOLKIND_64BIT_PML4:
2924	case PGMPOOLKIND_PAE_PDPT:
2925	return 8;
2926
2927	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2928	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2929	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2930	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2931	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2932	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2933	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2934	case PGMPOOLKIND_ROOT_NESTED:
2935	case PGMPOOLKIND_PAE_PD_PHYS:
2936	case PGMPOOLKIND_PAE_PDPT_PHYS:
2937	case PGMPOOLKIND_32BIT_PD_PHYS:
2938	/** @todo can we return 0? (nobody is calling this...) */
2939	AssertFailed();
2940	return 0;
2941
2942	default:
2943	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2944	}
2945	}
2946
2947
2948	/**
2949	* Scans one shadow page table for mappings of a physical page.
2950	*
2951	* @returns true/false indicating removal of all relevant PTEs
2952	* @param pVM The VM handle.
2953	* @param pPhysPage The guest page in question.
2954	* @param fFlushPTEs Flush PTEs or allow them to be updated (e.g. in case of an RW bit change)
2955	* @param iShw The shadow page table.
2956	* @param iPte Page table entry or NIL_PGMPOOL_PHYSEXT_IDX_PTE if unknown
2957	* @param cRefs The number of references made in that PT.
2958	*/
2959	static bool pgmPoolTrackFlushGCPhysPTInt(PVM pVM, PCPGMPAGE pPhysPage, bool fFlushPTEs, uint16_t iShw, uint16_t iPte, uint16_t cRefs)
2960	{
2961	LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%RHp iShw=%d iPte=%d cRefs=%d\n", PGM_PAGE_GET_HCPHYS(pPhysPage), iShw, iPte, cRefs));
2962	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2963	bool bRet = false;
2964
2965	/*
2966	* Assert sanity.
2967	*/
2968	Assert(cRefs == 1);
2969	Assert(iPte != NIL_PGMPOOL_PHYSEXT_IDX_PTE);
2970	AssertFatalMsg(iShw < pPool->cCurPages && iShw != NIL_PGMPOOL_IDX, ("iShw=%d\n", iShw));
2971	PPGMPOOLPAGE pPage = &pPool->aPages[iShw];
2972
2973	/*
2974	* Then, clear the actual mappings to the page in the shadow PT.
2975	*/
2976	switch (pPage->enmKind)
2977	{
2978	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2979	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2980	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2981	{
2982	const uint32_t u32 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2983	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2984	uint32_t u32AndMask, u32OrMask;
2985
2986	u32AndMask = 0;
2987	u32OrMask = 0;
2988
2989	if (!fFlushPTEs)
2990	{
2991	switch (PGM_PAGE_GET_HNDL_PHYS_STATE(pPhysPage))
2992	{
2993	case PGM_PAGE_HNDL_PHYS_STATE_NONE: /** No handler installed. */
2994	case PGM_PAGE_HNDL_PHYS_STATE_DISABLED: /** Monitoring is temporarily disabled. */
2995	u32OrMask = X86_PTE_RW;
2996	u32AndMask = UINT32_MAX;
2997	bRet = true;
2998	STAM_COUNTER_INC(&pPool->StatTrackFlushEntryKeep);
2999	break;
3000
3001	case PGM_PAGE_HNDL_PHYS_STATE_WRITE: /** Write access is monitored. */
3002	u32OrMask = 0;
3003	u32AndMask = ~X86_PTE_RW;
3004	bRet = true;
3005	STAM_COUNTER_INC(&pPool->StatTrackFlushEntryKeep);
3006	break;
3007	default:
3008	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3009	break;
3010	}
3011	}
3012	else
3013	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3014
3015	/* Update the counter if we're removing references. */
3016	if (!u32AndMask)
3017	{
3018	Assert(pPage->cPresent >= cRefs);
3019	Assert(pPool->cPresent >= cRefs);
3020	pPage->cPresent -= cRefs;
3021	pPool->cPresent -= cRefs;
3022	}
3023
3024	if ((pPT->a[iPte].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
3025	{
3026	X86PTE Pte;
3027
3028	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX32 cRefs=%#x\n", iPte, pPT->a[iPte], cRefs));
3029	Pte.u = (pPT->a[iPte].u & u32AndMask) \| u32OrMask;
3030	if (Pte.u & PGM_PTFLAGS_TRACK_DIRTY)
3031	Pte.n.u1Write = 0; /* need to disallow writes when dirty bit tracking is still active. */
3032
3033	ASMAtomicWriteSize(&pPT->a[iPte].u, Pte.u);
3034	return bRet;
3035	}
3036	#ifdef LOG_ENABLED
3037	Log(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
3038	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
3039	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
3040	{
3041	Log(("i=%d cRefs=%d\n", i, cRefs--));
3042	}
3043	#endif
3044	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d u32=%RX32 poolkind=%x\n", cRefs, pPage->iFirstPresent, pPage->cPresent, u32, pPage->enmKind));
3045	break;
3046	}
3047
3048	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3049	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3050	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3051	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3052	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3053	case PGMPOOLKIND_EPT_PT_FOR_PHYS: /* physical mask the same as PAE; RW bit as well; be careful! */
3054	{
3055	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
3056	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3057	uint64_t u64AndMask, u64OrMask;
3058
3059	u64OrMask = 0;
3060	u64AndMask = 0;
3061	if (!fFlushPTEs)
3062	{
3063	switch (PGM_PAGE_GET_HNDL_PHYS_STATE(pPhysPage))
3064	{
3065	case PGM_PAGE_HNDL_PHYS_STATE_NONE: /** No handler installed. */
3066	case PGM_PAGE_HNDL_PHYS_STATE_DISABLED: /** Monitoring is temporarily disabled. */
3067	u64OrMask = X86_PTE_RW;
3068	u64AndMask = UINT64_MAX;
3069	bRet = true;
3070	STAM_COUNTER_INC(&pPool->StatTrackFlushEntryKeep);
3071	break;
3072
3073	case PGM_PAGE_HNDL_PHYS_STATE_WRITE: /** Write access is monitored. */
3074	u64OrMask = 0;
3075	u64AndMask = ~((uint64_t)X86_PTE_RW);
3076	bRet = true;
3077	STAM_COUNTER_INC(&pPool->StatTrackFlushEntryKeep);
3078	break;
3079
3080	default:
3081	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3082	break;
3083	}
3084	}
3085	else
3086	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3087
3088	/* Update the counter if we're removing references. */
3089	if (!u64AndMask)
3090	{
3091	Assert(pPage->cPresent >= cRefs);
3092	Assert(pPool->cPresent >= cRefs);
3093	pPage->cPresent -= cRefs;
3094	pPool->cPresent -= cRefs;
3095	}
3096
3097	if ((pPT->a[iPte].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
3098	{
3099	X86PTEPAE Pte;
3100
3101	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", iPte, pPT->a[iPte], cRefs));
3102	Pte.u = (pPT->a[iPte].u & u64AndMask) \| u64OrMask;
3103	if (Pte.u & PGM_PTFLAGS_TRACK_DIRTY)
3104	Pte.n.u1Write = 0; /* need to disallow writes when dirty bit tracking is still active. */
3105
3106	ASMAtomicWriteSize(&pPT->a[iPte].u, Pte.u);
3107	return bRet;
3108	}
3109	#ifdef LOG_ENABLED
3110	Log(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
3111	Log(("Found %RX64 expected %RX64\n", pPT->a[iPte].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P), u64));
3112	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
3113	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
3114	{
3115	Log(("i=%d cRefs=%d\n", i, cRefs--));
3116	}
3117	#endif
3118	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d u64=%RX64 poolkind=%x\n", cRefs, pPage->iFirstPresent, pPage->cPresent, u64, pPage->enmKind));
3119	break;
3120	}
3121
3122	#ifdef PGM_WITH_LARGE_PAGES
3123	/* Large page case only. */
3124	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3125	{
3126	Assert(HWACCMIsNestedPagingActive(pVM));
3127
3128	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PDE4M_P \| X86_PDE4M_PS;
3129	PEPTPD pPD = (PEPTPD)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3130
3131	if ((pPD->a[iPte].u & (EPT_PDE2M_PG_MASK \| X86_PDE4M_P \| X86_PDE4M_PS)) == u64)
3132	{
3133	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pde=%RX64 cRefs=%#x\n", iPte, pPD->a[iPte], cRefs));
3134	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3135	pPD->a[iPte].u = 0;
3136
3137	/* Update the counter as we're removing references. */
3138	Assert(pPage->cPresent);
3139	Assert(pPool->cPresent);
3140	pPage->cPresent--;
3141	pPool->cPresent--;
3142
3143	return bRet;
3144	}
3145	# ifdef LOG_ENABLED
3146	Log(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
3147	for (unsigned i = 0; i < RT_ELEMENTS(pPD->a); i++)
3148	if ((pPD->a[i].u & (EPT_PDE2M_PG_MASK \| X86_PDE4M_P \| X86_PDE4M_PS)) == u64)
3149	{
3150	Log(("i=%d cRefs=%d\n", i, cRefs--));
3151	}
3152	# endif
3153	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
3154	break;
3155	}
3156
3157	/* AMD-V nested paging - @todo merge with EPT as we only check the parts that are identical. */
3158	case PGMPOOLKIND_PAE_PD_PHYS:
3159	{
3160	Assert(HWACCMIsNestedPagingActive(pVM));
3161
3162	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PDE4M_P \| X86_PDE4M_PS;
3163	PX86PD pPD = (PX86PD)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3164
3165	if ((pPD->a[iPte].u & (X86_PDE2M_PAE_PG_MASK \| X86_PDE4M_P \| X86_PDE4M_PS)) == u64)
3166	{
3167	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pde=%RX64 cRefs=%#x\n", iPte, pPD->a[iPte], cRefs));
3168	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3169	pPD->a[iPte].u = 0;
3170
3171	/* Update the counter as we're removing references. */
3172	Assert(pPage->cPresent);
3173	Assert(pPool->cPresent);
3174	pPage->cPresent--;
3175	pPool->cPresent--;
3176	return bRet;
3177	}
3178	# ifdef LOG_ENABLED
3179	Log(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
3180	for (unsigned i = 0; i < RT_ELEMENTS(pPD->a); i++)
3181	if ((pPD->a[i].u & (X86_PDE2M_PAE_PG_MASK \| X86_PDE4M_P \| X86_PDE4M_PS)) == u64)
3182	{
3183	Log(("i=%d cRefs=%d\n", i, cRefs--));
3184	}
3185	# endif
3186	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
3187	break;
3188	}
3189	#endif /* PGM_WITH_LARGE_PAGES */
3190
3191	default:
3192	AssertFatalMsgFailed(("enmKind=%d iShw=%d\n", pPage->enmKind, iShw));
3193	}
3194	return bRet;
3195	}
3196
3197
3198	/**
3199	* Scans one shadow page table for mappings of a physical page.
3200	*
3201	* @param pVM The VM handle.
3202	* @param pPhysPage The guest page in question.
3203	* @param fFlushPTEs Flush PTEs or allow them to be updated (e.g. in case of an RW bit change)
3204	* @param iShw The shadow page table.
3205	* @param cRefs The number of references made in that PT.
3206	*/
3207	static void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, bool fFlushPTEs, uint16_t iShw, uint16_t cRefs)
3208	{
3209	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
3210
3211	/* We should only come here with when there's only one reference to this physical page. */
3212	Assert(PGMPOOL_TD_GET_CREFS(PGM_PAGE_GET_TRACKING(pPhysPage)) == 1);
3213	Assert(cRefs == 1);
3214
3215	Log2(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%RHp iShw=%d cRefs=%d\n", PGM_PAGE_GET_HCPHYS(pPhysPage), iShw, cRefs));
3216	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPT, f);
3217	bool fKeptPTEs = pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, fFlushPTEs, iShw, PGM_PAGE_GET_PTE_INDEX(pPhysPage), cRefs);
3218	if (!fKeptPTEs)
3219	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
3220	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPT, f);
3221	}
3222
3223
3224	/**
3225	* Flushes a list of shadow page tables mapping the same physical page.
3226	*
3227	* @param pVM The VM handle.
3228	* @param pPhysPage The guest page in question.
3229	* @param fFlushPTEs Flush PTEs or allow them to be updated (e.g. in case of an RW bit change)
3230	* @param iPhysExt The physical cross reference extent list to flush.
3231	*/
3232	static void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, bool fFlushPTEs, uint16_t iPhysExt)
3233	{
3234	Assert(PGMIsLockOwner(pVM));
3235	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3236	bool fKeepList = false;
3237
3238	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTs, f);
3239	Log2(("pgmPoolTrackFlushGCPhysPTs: pPhysPage=%RHp iPhysExt\n", PGM_PAGE_GET_HCPHYS(pPhysPage), iPhysExt));
3240
3241	const uint16_t iPhysExtStart = iPhysExt;
3242	PPGMPOOLPHYSEXT pPhysExt;
3243	do
3244	{
3245	Assert(iPhysExt < pPool->cMaxPhysExts);
3246	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3247	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3248	{
3249	if (pPhysExt->aidx[i] != NIL_PGMPOOL_IDX)
3250	{
3251	bool fKeptPTEs = pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, fFlushPTEs, pPhysExt->aidx[i], pPhysExt->apte[i], 1);
3252	if (!fKeptPTEs)
3253	{
3254	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3255	pPhysExt->apte[i] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
3256	}
3257	else
3258	fKeepList = true;
3259	}
3260	}
3261	/* next */
3262	iPhysExt = pPhysExt->iNext;
3263	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3264
3265	if (!fKeepList)
3266	{
3267	/* insert the list into the free list and clear the ram range entry. */
3268	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3269	pPool->iPhysExtFreeHead = iPhysExtStart;
3270	/* Invalidate the tracking data. */
3271	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
3272	}
3273
3274	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTs, f);
3275	}
3276
3277
3278	/**
3279	* Flushes all shadow page table mappings of the given guest page.
3280	*
3281	* This is typically called when the host page backing the guest one has been
3282	* replaced or when the page protection was changed due to an access handler.
3283	*
3284	* @returns VBox status code.
3285	* @retval VINF_SUCCESS if all references has been successfully cleared.
3286	* @retval VINF_PGM_SYNC_CR3 if we're better off with a CR3 sync and a page
3287	* pool cleaning. FF and sync flags are set.
3288	*
3289	* @param pVM The VM handle.
3290	* @param GCPhysPage GC physical address of the page in question
3291	* @param pPhysPage The guest page in question.
3292	* @param fFlushPTEs Flush PTEs or allow them to be updated (e.g. in case of an RW bit change)
3293	* @param pfFlushTLBs This is set to @a true if the shadow TLBs should be
3294	* flushed, it is NOT touched if this isn't necessary.
3295	* The caller MUST initialized this to @a false.
3296	*/
3297	int pgmPoolTrackUpdateGCPhys(PVM pVM, RTGCPHYS GCPhysPage, PPGMPAGE pPhysPage, bool fFlushPTEs, bool *pfFlushTLBs)
3298	{
3299	PVMCPU pVCpu = VMMGetCpu(pVM);
3300	pgmLock(pVM);
3301	int rc = VINF_SUCCESS;
3302
3303	#ifdef PGM_WITH_LARGE_PAGES
3304	/* Is this page part of a large page? */
3305	if (PGM_PAGE_GET_PDE_TYPE(pPhysPage) == PGM_PAGE_PDE_TYPE_PDE)
3306	{
3307	PPGMPAGE pPhysBase;
3308	RTGCPHYS GCPhysBase = GCPhysPage & X86_PDE2M_PAE_PG_MASK;
3309
3310	GCPhysPage &= X86_PDE_PAE_PG_MASK;
3311
3312	/* Fetch the large page base. */
3313	if (GCPhysBase != GCPhysPage)
3314	{
3315	pPhysBase = pgmPhysGetPage(&pVM->pgm.s, GCPhysBase);
3316	AssertFatal(pPhysBase);
3317	}
3318	else
3319	pPhysBase = pPhysPage;
3320
3321	Log(("pgmPoolTrackUpdateGCPhys: update large page PDE for %RGp (%RGp)\n", GCPhysBase, GCPhysPage));
3322
3323	if (PGM_PAGE_GET_PDE_TYPE(pPhysBase) == PGM_PAGE_PDE_TYPE_PDE)
3324	{
3325	/* Mark the large page as disabled as we need to break it up to change a single page in the 2 MB range. */
3326	PGM_PAGE_SET_PDE_TYPE(pPhysBase, PGM_PAGE_PDE_TYPE_PDE_DISABLED);
3327
3328	/* Update the base as that only that one has a reference and there's only one PDE to clear. */
3329	rc = pgmPoolTrackUpdateGCPhys(pVM, GCPhysBase, pPhysBase, fFlushPTEs, pfFlushTLBs);
3330
3331	*pfFlushTLBs = true;
3332	pgmUnlock(pVM);
3333	return rc;
3334	}
3335	}
3336	#else
3337	NOREF(GCPhysPage);
3338	#endif /* PGM_WITH_LARGE_PAGES */
3339
3340	const uint16_t u16 = PGM_PAGE_GET_TRACKING(pPhysPage);
3341	if (u16)
3342	{
3343	/*
3344	* The zero page is currently screwing up the tracking and we'll
3345	* have to flush the whole shebang. Unless VBOX_WITH_NEW_LAZY_PAGE_ALLOC
3346	* is defined, zero pages won't normally be mapped. Some kind of solution
3347	* will be needed for this problem of course, but it will have to wait...
3348	*/
3349	if ( PGM_PAGE_IS_ZERO(pPhysPage)
3350	\|\| PGM_PAGE_IS_BALLOONED(pPhysPage))
3351	rc = VINF_PGM_GCPHYS_ALIASED;
3352	else
3353	{
3354	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
3355	/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow and
3356	pgmPoolTrackFlushGCPhysPTs will/may kill the pool otherwise. */
3357	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
3358	# endif
3359
3360	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
3361	pgmPoolTrackFlushGCPhysPT(pVM,
3362	pPhysPage,
3363	fFlushPTEs,
3364	PGMPOOL_TD_GET_IDX(u16),
3365	PGMPOOL_TD_GET_CREFS(u16));
3366	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
3367	pgmPoolTrackFlushGCPhysPTs(pVM, pPhysPage, fFlushPTEs, PGMPOOL_TD_GET_IDX(u16));
3368	else
3369	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
3370	*pfFlushTLBs = true;
3371
3372	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
3373	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
3374	# endif
3375	}
3376	}
3377
3378	if (rc == VINF_PGM_GCPHYS_ALIASED)
3379	{
3380	pVCpu->pgm.s.fSyncFlags \|= PGM_SYNC_CLEAR_PGM_POOL;
3381	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
3382	rc = VINF_PGM_SYNC_CR3;
3383	}
3384	pgmUnlock(pVM);
3385	return rc;
3386	}
3387
3388
3389	/**
3390	* Scans all shadow page tables for mappings of a physical page.
3391	*
3392	* This may be slow, but it's most likely more efficient than cleaning
3393	* out the entire page pool / cache.
3394	*
3395	* @returns VBox status code.
3396	* @retval VINF_SUCCESS if all references has been successfully cleared.
3397	* @retval VINF_PGM_GCPHYS_ALIASED if we're better off with a CR3 sync and
3398	* a page pool cleaning.
3399	*
3400	* @param pVM The VM handle.
3401	* @param pPhysPage The guest page in question.
3402	*/
3403	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage)
3404	{
3405	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3406	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTsSlow, s);
3407	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: cUsedPages=%d cPresent=%d pPhysPage=%R[pgmpage]\n",
3408	pPool->cUsedPages, pPool->cPresent, pPhysPage));
3409
3410	/*
3411	* There is a limit to what makes sense.
3412	*/
3413	if ( pPool->cPresent > 1024
3414	&& pVM->cCpus == 1)
3415	{
3416	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
3417	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
3418	return VINF_PGM_GCPHYS_ALIASED;
3419	}
3420
3421	/*
3422	* Iterate all the pages until we've encountered all that in use.
3423	* This is simple but not quite optimal solution.
3424	*/
3425	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
3426	const uint32_t u32 = u64;
3427	unsigned cLeft = pPool->cUsedPages;
3428	unsigned iPage = pPool->cCurPages;
3429	while (--iPage >= PGMPOOL_IDX_FIRST)
3430	{
3431	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
3432	if ( pPage->GCPhys != NIL_RTGCPHYS
3433	&& pPage->cPresent)
3434	{
3435	switch (pPage->enmKind)
3436	{
3437	/*
3438	* We only care about shadow page tables.
3439	*/
3440	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3441	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3442	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3443	{
3444	unsigned cPresent = pPage->cPresent;
3445	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3446	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
3447	if (pPT->a[i].n.u1Present)
3448	{
3449	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
3450	{
3451	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX32\n", iPage, i, pPT->a[i]));
3452	pPT->a[i].u = 0;
3453
3454	/* Update the counter as we're removing references. */
3455	Assert(pPage->cPresent);
3456	Assert(pPool->cPresent);
3457	pPage->cPresent--;
3458	pPool->cPresent--;
3459	}
3460	if (!--cPresent)
3461	break;
3462	}
3463	break;
3464	}
3465
3466	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3467	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3468	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3469	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3470	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3471	{
3472	unsigned cPresent = pPage->cPresent;
3473	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3474	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
3475	if (pPT->a[i].n.u1Present)
3476	{
3477	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
3478	{
3479	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
3480	pPT->a[i].u = 0;
3481
3482	/* Update the counter as we're removing references. */
3483	Assert(pPage->cPresent);
3484	Assert(pPool->cPresent);
3485	pPage->cPresent--;
3486	pPool->cPresent--;
3487	}
3488	if (!--cPresent)
3489	break;
3490	}
3491	break;
3492	}
3493	#ifndef IN_RC
3494	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3495	{
3496	unsigned cPresent = pPage->cPresent;
3497	PEPTPT pPT = (PEPTPT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3498	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
3499	if (pPT->a[i].n.u1Present)
3500	{
3501	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
3502	{
3503	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
3504	pPT->a[i].u = 0;
3505
3506	/* Update the counter as we're removing references. */
3507	Assert(pPage->cPresent);
3508	Assert(pPool->cPresent);
3509	pPage->cPresent--;
3510	pPool->cPresent--;
3511	}
3512	if (!--cPresent)
3513	break;
3514	}
3515	break;
3516	}
3517	#endif
3518	}
3519	if (!--cLeft)
3520	break;
3521	}
3522	}
3523
3524	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
3525	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
3526
3527	/*
3528	* There is a limit to what makes sense. The above search is very expensive, so force a pgm pool flush.
3529	*/
3530	if (pPool->cPresent > 1024)
3531	{
3532	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
3533	return VINF_PGM_GCPHYS_ALIASED;
3534	}
3535
3536	return VINF_SUCCESS;
3537	}
3538
3539
3540	/**
3541	* Clears the user entry in a user table.
3542	*
3543	* This is used to remove all references to a page when flushing it.
3544	*/
3545	static void pgmPoolTrackClearPageUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PCPGMPOOLUSER pUser)
3546	{
3547	Assert(pUser->iUser != NIL_PGMPOOL_IDX);
3548	Assert(pUser->iUser < pPool->cCurPages);
3549	uint32_t iUserTable = pUser->iUserTable;
3550
3551	/*
3552	* Map the user page.
3553	*/
3554	PPGMPOOLPAGE pUserPage = &pPool->aPages[pUser->iUser];
3555	union
3556	{
3557	uint64_t *pau64;
3558	uint32_t *pau32;
3559	} u;
3560	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pUserPage);
3561
3562	LogFlow(("pgmPoolTrackClearPageUser: clear %x in %s (%RGp) (flushing %s)\n", iUserTable, pgmPoolPoolKindToStr(pUserPage->enmKind), pUserPage->Core.Key, pgmPoolPoolKindToStr(pPage->enmKind)));
3563
3564	/* Safety precaution in case we change the paging for other modes too in the future. */
3565	Assert(!pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage));
3566
3567	#ifdef VBOX_STRICT
3568	/*
3569	* Some sanity checks.
3570	*/
3571	switch (pUserPage->enmKind)
3572	{
3573	case PGMPOOLKIND_32BIT_PD:
3574	case PGMPOOLKIND_32BIT_PD_PHYS:
3575	Assert(iUserTable < X86_PG_ENTRIES);
3576	break;
3577	case PGMPOOLKIND_PAE_PDPT:
3578	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
3579	case PGMPOOLKIND_PAE_PDPT_PHYS:
3580	Assert(iUserTable < 4);
3581	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
3582	break;
3583	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3584	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3585	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3586	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3587	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3588	case PGMPOOLKIND_PAE_PD_PHYS:
3589	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3590	break;
3591	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3592	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3593	Assert(!(u.pau64[iUserTable] & PGM_PDFLAGS_MAPPING));
3594	break;
3595	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3596	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3597	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
3598	break;
3599	case PGMPOOLKIND_64BIT_PML4:
3600	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
3601	/* GCPhys >> PAGE_SHIFT is the index here */
3602	break;
3603	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3604	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3605	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3606	break;
3607
3608	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3609	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3610	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3611	break;
3612
3613	case PGMPOOLKIND_ROOT_NESTED:
3614	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3615	break;
3616
3617	default:
3618	AssertMsgFailed(("enmKind=%d\n", pUserPage->enmKind));
3619	break;
3620	}
3621	#endif /* VBOX_STRICT */
3622
3623	/*
3624	* Clear the entry in the user page.
3625	*/
3626	switch (pUserPage->enmKind)
3627	{
3628	/* 32-bit entries */
3629	case PGMPOOLKIND_32BIT_PD:
3630	case PGMPOOLKIND_32BIT_PD_PHYS:
3631	ASMAtomicWriteSize(&u.pau32[iUserTable], 0);
3632	break;
3633
3634	/* 64-bit entries */
3635	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3636	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3637	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3638	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3639	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3640	#if defined(IN_RC)
3641	/* 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
3642	* non-present PDPT will continue to cause page faults.
3643	*/
3644	ASMReloadCR3();
3645	#endif
3646	/* no break */
3647	case PGMPOOLKIND_PAE_PD_PHYS:
3648	case PGMPOOLKIND_PAE_PDPT_PHYS:
3649	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3650	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3651	case PGMPOOLKIND_64BIT_PML4:
3652	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3653	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3654	case PGMPOOLKIND_PAE_PDPT:
3655	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
3656	case PGMPOOLKIND_ROOT_NESTED:
3657	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3658	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3659	ASMAtomicWriteSize(&u.pau64[iUserTable], 0);
3660	break;
3661
3662	default:
3663	AssertFatalMsgFailed(("enmKind=%d iUser=%#x iUserTable=%#x\n", pUserPage->enmKind, pUser->iUser, pUser->iUserTable));
3664	}
3665	}
3666
3667
3668	/**
3669	* Clears all users of a page.
3670	*/
3671	static void pgmPoolTrackClearPageUsers(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3672	{
3673	/*
3674	* Free all the user records.
3675	*/
3676	LogFlow(("pgmPoolTrackClearPageUsers %RGp\n", pPage->GCPhys));
3677
3678	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
3679	uint16_t i = pPage->iUserHead;
3680	while (i != NIL_PGMPOOL_USER_INDEX)
3681	{
3682	/* Clear enter in user table. */
3683	pgmPoolTrackClearPageUser(pPool, pPage, &paUsers[i]);
3684
3685	/* Free it. */
3686	const uint16_t iNext = paUsers[i].iNext;
3687	paUsers[i].iUser = NIL_PGMPOOL_IDX;
3688	paUsers[i].iNext = pPool->iUserFreeHead;
3689	pPool->iUserFreeHead = i;
3690
3691	/* Next. */
3692	i = iNext;
3693	}
3694	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
3695	}
3696
3697
3698	/**
3699	* Allocates a new physical cross reference extent.
3700	*
3701	* @returns Pointer to the allocated extent on success. NULL if we're out of them.
3702	* @param pVM The VM handle.
3703	* @param piPhysExt Where to store the phys ext index.
3704	*/
3705	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt)
3706	{
3707	Assert(PGMIsLockOwner(pVM));
3708	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3709	uint16_t iPhysExt = pPool->iPhysExtFreeHead;
3710	if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
3711	{
3712	STAM_COUNTER_INC(&pPool->StamTrackPhysExtAllocFailures);
3713	return NULL;
3714	}
3715	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3716	pPool->iPhysExtFreeHead = pPhysExt->iNext;
3717	pPhysExt->iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
3718	*piPhysExt = iPhysExt;
3719	return pPhysExt;
3720	}
3721
3722
3723	/**
3724	* Frees a physical cross reference extent.
3725	*
3726	* @param pVM The VM handle.
3727	* @param iPhysExt The extent to free.
3728	*/
3729	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt)
3730	{
3731	Assert(PGMIsLockOwner(pVM));
3732	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3733	Assert(iPhysExt < pPool->cMaxPhysExts);
3734	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3735	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3736	{
3737	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3738	pPhysExt->apte[i] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
3739	}
3740	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3741	pPool->iPhysExtFreeHead = iPhysExt;
3742	}
3743
3744
3745	/**
3746	* Frees a physical cross reference extent.
3747	*
3748	* @param pVM The VM handle.
3749	* @param iPhysExt The extent to free.
3750	*/
3751	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt)
3752	{
3753	Assert(PGMIsLockOwner(pVM));
3754	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3755
3756	const uint16_t iPhysExtStart = iPhysExt;
3757	PPGMPOOLPHYSEXT pPhysExt;
3758	do
3759	{
3760	Assert(iPhysExt < pPool->cMaxPhysExts);
3761	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3762	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3763	{
3764	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3765	pPhysExt->apte[i] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
3766	}
3767
3768	/* next */
3769	iPhysExt = pPhysExt->iNext;
3770	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3771
3772	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3773	pPool->iPhysExtFreeHead = iPhysExtStart;
3774	}
3775
3776
3777	/**
3778	* Insert a reference into a list of physical cross reference extents.
3779	*
3780	* @returns The new tracking data for PGMPAGE.
3781	*
3782	* @param pVM The VM handle.
3783	* @param iPhysExt The physical extent index of the list head.
3784	* @param iShwPT The shadow page table index.
3785	* @param iPte Page table entry
3786	*
3787	*/
3788	static uint16_t pgmPoolTrackPhysExtInsert(PVM pVM, uint16_t iPhysExt, uint16_t iShwPT, uint16_t iPte)
3789	{
3790	Assert(PGMIsLockOwner(pVM));
3791	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3792	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3793
3794	/* special common case. */
3795	if (paPhysExts[iPhysExt].aidx[2] == NIL_PGMPOOL_IDX)
3796	{
3797	paPhysExts[iPhysExt].aidx[2] = iShwPT;
3798	paPhysExts[iPhysExt].apte[2] = iPte;
3799	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
3800	LogFlow(("pgmPoolTrackPhysExtInsert: %d:{,,%d pte %d}\n", iPhysExt, iShwPT, iPte));
3801	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3802	}
3803
3804	/* general treatment. */
3805	const uint16_t iPhysExtStart = iPhysExt;
3806	unsigned cMax = 15;
3807	for (;;)
3808	{
3809	Assert(iPhysExt < pPool->cMaxPhysExts);
3810	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3811	if (paPhysExts[iPhysExt].aidx[i] == NIL_PGMPOOL_IDX)
3812	{
3813	paPhysExts[iPhysExt].aidx[i] = iShwPT;
3814	paPhysExts[iPhysExt].apte[i] = iPte;
3815	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
3816	LogFlow(("pgmPoolTrackPhysExtInsert: %d:{%d pte %d} i=%d cMax=%d\n", iPhysExt, iShwPT, iPte, i, cMax));
3817	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtStart);
3818	}
3819	if (!--cMax)
3820	{
3821	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
3822	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
3823	LogFlow(("pgmPoolTrackPhysExtInsert: overflow (1) iShwPT=%d\n", iShwPT));
3824	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3825	}
3826	}
3827
3828	/* add another extent to the list. */
3829	PPGMPOOLPHYSEXT pNew = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3830	if (!pNew)
3831	{
3832	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackNoExtentsLeft);
3833	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
3834	LogFlow(("pgmPoolTrackPhysExtInsert: pgmPoolTrackPhysExtAlloc failed iShwPT=%d\n", iShwPT));
3835	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3836	}
3837	pNew->iNext = iPhysExtStart;
3838	pNew->aidx[0] = iShwPT;
3839	pNew->apte[0] = iPte;
3840	LogFlow(("pgmPoolTrackPhysExtInsert: added new extent %d:{%d pte %d}->%d\n", iPhysExt, iShwPT, iPte, iPhysExtStart));
3841	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3842	}
3843
3844
3845	/**
3846	* Add a reference to guest physical page where extents are in use.
3847	*
3848	* @returns The new tracking data for PGMPAGE.
3849	*
3850	* @param pVM The VM handle.
3851	* @param pPhysPage Pointer to the aPages entry in the ram range.
3852	* @param u16 The ram range flags (top 16-bits).
3853	* @param iShwPT The shadow page table index.
3854	* @param iPte Page table entry
3855	*/
3856	uint16_t pgmPoolTrackPhysExtAddref(PVM pVM, PPGMPAGE pPhysPage, uint16_t u16, uint16_t iShwPT, uint16_t iPte)
3857	{
3858	pgmLock(pVM);
3859	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
3860	{
3861	/*
3862	* Convert to extent list.
3863	*/
3864	Assert(PGMPOOL_TD_GET_CREFS(u16) == 1);
3865	uint16_t iPhysExt;
3866	PPGMPOOLPHYSEXT pPhysExt = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3867	if (pPhysExt)
3868	{
3869	LogFlow(("pgmPoolTrackPhysExtAddref: new extent: %d:{%d, %d}\n", iPhysExt, PGMPOOL_TD_GET_IDX(u16), iShwPT));
3870	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliased);
3871	pPhysExt->aidx[0] = PGMPOOL_TD_GET_IDX(u16);
3872	pPhysExt->apte[0] = PGM_PAGE_GET_PTE_INDEX(pPhysPage);
3873	pPhysExt->aidx[1] = iShwPT;
3874	pPhysExt->apte[1] = iPte;
3875	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
3876	}
3877	else
3878	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
3879	}
3880	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
3881	{
3882	/*
3883	* Insert into the extent list.
3884	*/
3885	u16 = pgmPoolTrackPhysExtInsert(pVM, PGMPOOL_TD_GET_IDX(u16), iShwPT, iPte);
3886	}
3887	else
3888	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedLots);
3889	pgmUnlock(pVM);
3890	return u16;
3891	}
3892
3893	/**
3894	* Clear references to guest physical memory.
3895	*
3896	* @param pPool The pool.
3897	* @param pPage The page.
3898	* @param pPhysPage Pointer to the aPages entry in the ram range.
3899	* @param iPte Shadow PTE index
3900	*/
3901	void pgmPoolTrackPhysExtDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMPAGE pPhysPage, uint16_t iPte)
3902	{
3903	const unsigned cRefs = PGM_PAGE_GET_TD_CREFS(pPhysPage);
3904	AssertFatalMsg(cRefs == PGMPOOL_TD_CREFS_PHYSEXT, ("cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
3905
3906	uint16_t iPhysExt = PGM_PAGE_GET_TD_IDX(pPhysPage);
3907	if (iPhysExt != PGMPOOL_TD_IDX_OVERFLOWED)
3908	{
3909	PVM pVM = pPool->CTX_SUFF(pVM);
3910	pgmLock(pVM);
3911
3912	uint16_t iPhysExtPrev = NIL_PGMPOOL_PHYSEXT_INDEX;
3913	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3914	do
3915	{
3916	Assert(iPhysExt < pPool->cMaxPhysExts);
3917
3918	/*
3919	* Look for the shadow page and check if it's all freed.
3920	*/
3921	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3922	{
3923	if ( paPhysExts[iPhysExt].aidx[i] == pPage->idx
3924	&& paPhysExts[iPhysExt].apte[i] == iPte)
3925	{
3926	paPhysExts[iPhysExt].aidx[i] = NIL_PGMPOOL_IDX;
3927	paPhysExts[iPhysExt].apte[i] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
3928
3929	for (i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3930	if (paPhysExts[iPhysExt].aidx[i] != NIL_PGMPOOL_IDX)
3931	{
3932	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
3933	pgmUnlock(pVM);
3934	return;
3935	}
3936
3937	/* we can free the node. */
3938	const uint16_t iPhysExtNext = paPhysExts[iPhysExt].iNext;
3939	if ( iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX
3940	&& iPhysExtNext == NIL_PGMPOOL_PHYSEXT_INDEX)
3941	{
3942	/* lonely node */
3943	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3944	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d lonely\n", pPhysPage, pPage->idx));
3945	PGM_PAGE_SET_TRACKING(pPhysPage, 0);
3946	}
3947	else if (iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX)
3948	{
3949	/* head */
3950	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d head\n", pPhysPage, pPage->idx));
3951	PGM_PAGE_SET_TRACKING(pPhysPage, PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtNext));
3952	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3953	}
3954	else
3955	{
3956	/* in list */
3957	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d in list\n", pPhysPage, pPage->idx));
3958	paPhysExts[iPhysExtPrev].iNext = iPhysExtNext;
3959	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3960	}
3961	iPhysExt = iPhysExtNext;
3962	pgmUnlock(pVM);
3963	return;
3964	}
3965	}
3966
3967	/* next */
3968	iPhysExtPrev = iPhysExt;
3969	iPhysExt = paPhysExts[iPhysExt].iNext;
3970	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3971
3972	pgmUnlock(pVM);
3973	AssertFatalMsgFailed(("not-found! cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
3974	}
3975	else /* nothing to do */
3976	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage]\n", pPhysPage));
3977	}
3978
3979	/**
3980	* Clear references to guest physical memory.
3981	*
3982	* This is the same as pgmPoolTracDerefGCPhys except that the guest physical address
3983	* is assumed to be correct, so the linear search can be skipped and we can assert
3984	* at an earlier point.
3985	*
3986	* @param pPool The pool.
3987	* @param pPage The page.
3988	* @param HCPhys The host physical address corresponding to the guest page.
3989	* @param GCPhys The guest physical address corresponding to HCPhys.
3990	* @param iPte Shadow PTE index
3991	*/
3992	static void pgmPoolTracDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhys, uint16_t iPte)
3993	{
3994	/*
3995	* Walk range list.
3996	*/
3997	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3998	while (pRam)
3999	{
4000	RTGCPHYS off = GCPhys - pRam->GCPhys;
4001	if (off < pRam->cb)
4002	{
4003	/* does it match? */
4004	const unsigned iPage = off >> PAGE_SHIFT;
4005	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
4006	#ifdef LOG_ENABLED
4007	RTHCPHYS HCPhysPage = PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]);
4008	Log2(("pgmPoolTracDerefGCPhys %RHp vs %RHp\n", HCPhysPage, HCPhys));
4009	#endif
4010	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
4011	{
4012	Assert(pPage->cPresent);
4013	Assert(pPool->cPresent);
4014	pPage->cPresent--;
4015	pPool->cPresent--;
4016	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage], iPte);
4017	return;
4018	}
4019	break;
4020	}
4021	pRam = pRam->CTX_SUFF(pNext);
4022	}
4023	AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp\n", HCPhys, GCPhys));
4024	}
4025
4026
4027	/**
4028	* Clear references to guest physical memory.
4029	*
4030	* @param pPool The pool.
4031	* @param pPage The page.
4032	* @param HCPhys The host physical address corresponding to the guest page.
4033	* @param GCPhysHint The guest physical address which may corresponding to HCPhys.
4034	* @param iPte Shadow pte index
4035	*/
4036	void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint, uint16_t iPte)
4037	{
4038	Log4(("pgmPoolTracDerefGCPhysHint %RHp %RGp\n", HCPhys, GCPhysHint));
4039
4040	/*
4041	* Walk range list.
4042	*/
4043	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
4044	while (pRam)
4045	{
4046	RTGCPHYS off = GCPhysHint - pRam->GCPhys;
4047	if (off < pRam->cb)
4048	{
4049	/* does it match? */
4050	const unsigned iPage = off >> PAGE_SHIFT;
4051	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
4052	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
4053	{
4054	Assert(pPage->cPresent);
4055	Assert(pPool->cPresent);
4056	pPage->cPresent--;
4057	pPool->cPresent--;
4058	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage], iPte);
4059	return;
4060	}
4061	break;
4062	}
4063	pRam = pRam->CTX_SUFF(pNext);
4064	}
4065
4066	/*
4067	* Damn, the hint didn't work. We'll have to do an expensive linear search.
4068	*/
4069	STAM_COUNTER_INC(&pPool->StatTrackLinearRamSearches);
4070	pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
4071	while (pRam)
4072	{
4073	unsigned iPage = pRam->cb >> PAGE_SHIFT;
4074	while (iPage-- > 0)
4075	{
4076	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
4077	{
4078	Log4(("pgmPoolTracDerefGCPhysHint: Linear HCPhys=%RHp GCPhysHint=%RGp GCPhysReal=%RGp\n",
4079	HCPhys, GCPhysHint, pRam->GCPhys + (iPage << PAGE_SHIFT)));
4080	Assert(pPage->cPresent);
4081	Assert(pPool->cPresent);
4082	pPage->cPresent--;
4083	pPool->cPresent--;
4084	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage], iPte);
4085	return;
4086	}
4087	}
4088	pRam = pRam->CTX_SUFF(pNext);
4089	}
4090
4091	AssertFatalMsgFailed(("HCPhys=%RHp GCPhysHint=%RGp\n", HCPhys, GCPhysHint));
4092	}
4093
4094
4095	/**
4096	* Clear references to guest physical memory in a 32-bit / 32-bit page table.
4097	*
4098	* @param pPool The pool.
4099	* @param pPage The page.
4100	* @param pShwPT The shadow page table (mapping of the page).
4101	* @param pGstPT The guest page table.
4102	*/
4103	DECLINLINE(void) pgmPoolTrackDerefPT32Bit32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT, PCX86PT pGstPT)
4104	{
4105	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
4106	if (pShwPT->a[i].n.u1Present)
4107	{
4108	Log4(("pgmPoolTrackDerefPT32Bit32Bit: i=%d pte=%RX32 hint=%RX32\n",
4109	i, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
4110	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK, i);
4111	if (!pPage->cPresent)
4112	break;
4113	}
4114	}
4115
4116
4117	/**
4118	* Clear references to guest physical memory in a PAE / 32-bit page table.
4119	*
4120	* @param pPool The pool.
4121	* @param pPage The page.
4122	* @param pShwPT The shadow page table (mapping of the page).
4123	* @param pGstPT The guest page table (just a half one).
4124	*/
4125	DECLINLINE(void) pgmPoolTrackDerefPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PT pGstPT)
4126	{
4127	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
4128	if (pShwPT->a[i].n.u1Present)
4129	{
4130	Log4(("pgmPoolTrackDerefPTPae32Bit: i=%d pte=%RX64 hint=%RX32\n",
4131	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
4132	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK, i);
4133	if (!pPage->cPresent)
4134	break;
4135	}
4136	}
4137
4138
4139	/**
4140	* Clear references to guest physical memory in a PAE / PAE page table.
4141	*
4142	* @param pPool The pool.
4143	* @param pPage The page.
4144	* @param pShwPT The shadow page table (mapping of the page).
4145	* @param pGstPT The guest page table.
4146	*/
4147	DECLINLINE(void) pgmPoolTrackDerefPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PTPAE pGstPT)
4148	{
4149	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
4150	if (pShwPT->a[i].n.u1Present)
4151	{
4152	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX32 hint=%RX32\n",
4153	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
4154	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK, i);
4155	if (!pPage->cPresent)
4156	break;
4157	}
4158	}
4159
4160
4161	/**
4162	* Clear references to guest physical memory in a 32-bit / 4MB page table.
4163	*
4164	* @param pPool The pool.
4165	* @param pPage The page.
4166	* @param pShwPT The shadow page table (mapping of the page).
4167	*/
4168	DECLINLINE(void) pgmPoolTrackDerefPT32Bit4MB(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT)
4169	{
4170	RTGCPHYS GCPhys = pPage->GCPhys + PAGE_SIZE * pPage->iFirstPresent;
4171	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
4172	if (pShwPT->a[i].n.u1Present)
4173	{
4174	Log4(("pgmPoolTrackDerefPT32Bit4MB: i=%d pte=%RX32 GCPhys=%RGp\n",
4175	i, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys));
4176	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys, i);
4177	if (!pPage->cPresent)
4178	break;
4179	}
4180	}
4181
4182
4183	/**
4184	* Clear references to guest physical memory in a PAE / 2/4MB page table.
4185	*
4186	* @param pPool The pool.
4187	* @param pPage The page.
4188	* @param pShwPT The shadow page table (mapping of the page).
4189	*/
4190	DECLINLINE(void) pgmPoolTrackDerefPTPaeBig(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT)
4191	{
4192	RTGCPHYS GCPhys = pPage->GCPhys + PAGE_SIZE * pPage->iFirstPresent;
4193	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
4194	if (pShwPT->a[i].n.u1Present)
4195	{
4196	Log4(("pgmPoolTrackDerefPTPaeBig: i=%d pte=%RX64 hint=%RGp\n",
4197	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys));
4198	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys, i);
4199	if (!pPage->cPresent)
4200	break;
4201	}
4202	}
4203
4204
4205	/**
4206	* Clear references to shadowed pages in an EPT page table.
4207	*
4208	* @param pPool The pool.
4209	* @param pPage The page.
4210	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
4211	*/
4212	DECLINLINE(void) pgmPoolTrackDerefPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT)
4213	{
4214	RTGCPHYS GCPhys = pPage->GCPhys + PAGE_SIZE * pPage->iFirstPresent;
4215	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
4216	if (pShwPT->a[i].n.u1Present)
4217	{
4218	Log4(("pgmPoolTrackDerefPTEPT: i=%d pte=%RX64 GCPhys=%RX64\n",
4219	i, pShwPT->a[i].u & EPT_PTE_PG_MASK, pPage->GCPhys));
4220	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & EPT_PTE_PG_MASK, GCPhys, i);
4221	if (!pPage->cPresent)
4222	break;
4223	}
4224	}
4225
4226
4227
4228	/**
4229	* Clear references to shadowed pages in a 32 bits page directory.
4230	*
4231	* @param pPool The pool.
4232	* @param pPage The page.
4233	* @param pShwPD The shadow page directory (mapping of the page).
4234	*/
4235	DECLINLINE(void) pgmPoolTrackDerefPD(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PD pShwPD)
4236	{
4237	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
4238	{
4239	if ( pShwPD->a[i].n.u1Present
4240	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
4241	)
4242	{
4243	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PG_MASK);
4244	if (pSubPage)
4245	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4246	else
4247	AssertFatalMsgFailed(("%x\n", pShwPD->a[i].u & X86_PDE_PG_MASK));
4248	}
4249	}
4250	}
4251
4252	/**
4253	* Clear references to shadowed pages in a PAE (legacy or 64 bits) page directory.
4254	*
4255	* @param pPool The pool.
4256	* @param pPage The page.
4257	* @param pShwPD The shadow page directory (mapping of the page).
4258	*/
4259	DECLINLINE(void) pgmPoolTrackDerefPDPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPAE pShwPD)
4260	{
4261	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
4262	{
4263	if ( pShwPD->a[i].n.u1Present
4264	&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
4265	)
4266	{
4267	#ifdef PGM_WITH_LARGE_PAGES
4268	if (pShwPD->a[i].b.u1Size)
4269	{
4270	Log4(("pgmPoolTrackDerefPDPae: i=%d pde=%RX64 GCPhys=%RX64\n",
4271	i, pShwPD->a[i].u & X86_PDE2M_PAE_PG_MASK, pPage->GCPhys));
4272	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPD->a[i].u & X86_PDE2M_PAE_PG_MASK, pPage->GCPhys /* == base of 2 MB page */, i);
4273	}
4274	else
4275	#endif
4276	{
4277	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PAE_PG_MASK);
4278	if (pSubPage)
4279	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4280	else
4281	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & X86_PDE_PAE_PG_MASK));
4282	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4283	}
4284	}
4285	}
4286	}
4287
4288	/**
4289	* Clear references to shadowed pages in a PAE page directory pointer table.
4290	*
4291	* @param pPool The pool.
4292	* @param pPage The page.
4293	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
4294	*/
4295	DECLINLINE(void) pgmPoolTrackDerefPDPTPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
4296	{
4297	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
4298	{
4299	if ( pShwPDPT->a[i].n.u1Present
4300	&& !(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING)
4301	)
4302	{
4303	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
4304	if (pSubPage)
4305	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4306	else
4307	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
4308	}
4309	}
4310	}
4311
4312
4313	/**
4314	* Clear references to shadowed pages in a 64-bit page directory pointer table.
4315	*
4316	* @param pPool The pool.
4317	* @param pPage The page.
4318	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
4319	*/
4320	DECLINLINE(void) pgmPoolTrackDerefPDPT64Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
4321	{
4322	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
4323	{
4324	Assert(!(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING));
4325	if (pShwPDPT->a[i].n.u1Present)
4326	{
4327	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
4328	if (pSubPage)
4329	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4330	else
4331	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
4332	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4333	}
4334	}
4335	}
4336
4337
4338	/**
4339	* Clear references to shadowed pages in a 64-bit level 4 page table.
4340	*
4341	* @param pPool The pool.
4342	* @param pPage The page.
4343	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
4344	*/
4345	DECLINLINE(void) pgmPoolTrackDerefPML464Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PML4 pShwPML4)
4346	{
4347	for (unsigned i = 0; i < RT_ELEMENTS(pShwPML4->a); i++)
4348	{
4349	if (pShwPML4->a[i].n.u1Present)
4350	{
4351	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPML4->a[i].u & X86_PDPE_PG_MASK);
4352	if (pSubPage)
4353	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4354	else
4355	AssertFatalMsgFailed(("%RX64\n", pShwPML4->a[i].u & X86_PML4E_PG_MASK));
4356	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4357	}
4358	}
4359	}
4360
4361
4362	/**
4363	* Clear references to shadowed pages in an EPT page directory.
4364	*
4365	* @param pPool The pool.
4366	* @param pPage The page.
4367	* @param pShwPD The shadow page directory (mapping of the page).
4368	*/
4369	DECLINLINE(void) pgmPoolTrackDerefPDEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPD pShwPD)
4370	{
4371	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
4372	{
4373	if (pShwPD->a[i].n.u1Present)
4374	{
4375	#ifdef PGM_WITH_LARGE_PAGES
4376	if (pShwPD->a[i].b.u1Size)
4377	{
4378	Log4(("pgmPoolTrackDerefPDEPT: i=%d pde=%RX64 GCPhys=%RX64\n",
4379	i, pShwPD->a[i].u & X86_PDE2M_PAE_PG_MASK, pPage->GCPhys));
4380	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPD->a[i].u & X86_PDE2M_PAE_PG_MASK, pPage->GCPhys /* == base of 2 MB page */, i);
4381	}
4382	else
4383	#endif
4384	{
4385	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & EPT_PDE_PG_MASK);
4386	if (pSubPage)
4387	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4388	else
4389	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & EPT_PDE_PG_MASK));
4390	}
4391	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4392	}
4393	}
4394	}
4395
4396
4397	/**
4398	* Clear references to shadowed pages in an EPT page directory pointer table.
4399	*
4400	* @param pPool The pool.
4401	* @param pPage The page.
4402	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
4403	*/
4404	DECLINLINE(void) pgmPoolTrackDerefPDPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPDPT pShwPDPT)
4405	{
4406	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
4407	{
4408	if (pShwPDPT->a[i].n.u1Present)
4409	{
4410	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK);
4411	if (pSubPage)
4412	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4413	else
4414	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK));
4415	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4416	}
4417	}
4418	}
4419
4420
4421	/**
4422	* Clears all references made by this page.
4423	*
4424	* This includes other shadow pages and GC physical addresses.
4425	*
4426	* @param pPool The pool.
4427	* @param pPage The page.
4428	*/
4429	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
4430	{
4431	/*
4432	* Map the shadow page and take action according to the page kind.
4433	*/
4434	void *pvShw = PGMPOOL_PAGE_2_LOCKED_PTR(pPool->CTX_SUFF(pVM), pPage);
4435	switch (pPage->enmKind)
4436	{
4437	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
4438	{
4439	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
4440	void *pvGst;
4441	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
4442	pgmPoolTrackDerefPT32Bit32Bit(pPool, pPage, (PX86PT)pvShw, (PCX86PT)pvGst);
4443	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
4444	break;
4445	}
4446
4447	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
4448	{
4449	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
4450	void *pvGst;
4451	int rc = PGM_GCPHYS_2_PTR_EX(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
4452	pgmPoolTrackDerefPTPae32Bit(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PT)pvGst);
4453	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
4454	break;
4455	}
4456
4457	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
4458	{
4459	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
4460	void *pvGst;
4461	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
4462	pgmPoolTrackDerefPTPaePae(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PTPAE)pvGst);
4463	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
4464	break;
4465	}
4466
4467	case PGMPOOLKIND_32BIT_PT_FOR_PHYS: /* treat it like a 4 MB page */
4468	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
4469	{
4470	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
4471	pgmPoolTrackDerefPT32Bit4MB(pPool, pPage, (PX86PT)pvShw);
4472	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
4473	break;
4474	}
4475
4476	case PGMPOOLKIND_PAE_PT_FOR_PHYS: /* treat it like a 2 MB page */
4477	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
4478	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
4479	{
4480	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
4481	pgmPoolTrackDerefPTPaeBig(pPool, pPage, (PX86PTPAE)pvShw);
4482	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
4483	break;
4484	}
4485
4486	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
4487	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
4488	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
4489	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
4490	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
4491	case PGMPOOLKIND_PAE_PD_PHYS:
4492	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
4493	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
4494	pgmPoolTrackDerefPDPae(pPool, pPage, (PX86PDPAE)pvShw);
4495	break;
4496
4497	case PGMPOOLKIND_32BIT_PD_PHYS:
4498	case PGMPOOLKIND_32BIT_PD:
4499	pgmPoolTrackDerefPD(pPool, pPage, (PX86PD)pvShw);
4500	break;
4501
4502	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
4503	case PGMPOOLKIND_PAE_PDPT:
4504	case PGMPOOLKIND_PAE_PDPT_PHYS:
4505	pgmPoolTrackDerefPDPTPae(pPool, pPage, (PX86PDPT)pvShw);
4506	break;
4507
4508	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
4509	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
4510	pgmPoolTrackDerefPDPT64Bit(pPool, pPage, (PX86PDPT)pvShw);
4511	break;
4512
4513	case PGMPOOLKIND_64BIT_PML4:
4514	pgmPoolTrackDerefPML464Bit(pPool, pPage, (PX86PML4)pvShw);
4515	break;
4516
4517	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
4518	pgmPoolTrackDerefPTEPT(pPool, pPage, (PEPTPT)pvShw);
4519	break;
4520
4521	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
4522	pgmPoolTrackDerefPDEPT(pPool, pPage, (PEPTPD)pvShw);
4523	break;
4524
4525	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
4526	pgmPoolTrackDerefPDPTEPT(pPool, pPage, (PEPTPDPT)pvShw);
4527	break;
4528
4529	default:
4530	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
4531	}
4532
4533	/* paranoia, clear the shadow page. Remove this laser (i.e. let Alloc and ClearAll do it). */
4534	STAM_PROFILE_START(&pPool->StatZeroPage, z);
4535	ASMMemZeroPage(pvShw);
4536	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
4537	pPage->fZeroed = true;
4538	PGMPOOL_UNLOCK_PTR(pPool->CTX_SUFF(pVM), pvShw);
4539	Assert(!pPage->cPresent);
4540	}
4541
4542	/**
4543	* Flushes a pool page.
4544	*
4545	* This moves the page to the free list after removing all user references to it.
4546	*
4547	* @returns VBox status code.
4548	* @retval VINF_SUCCESS on success.
4549	* @param pPool The pool.
4550	* @param HCPhys The HC physical address of the shadow page.
4551	* @param fFlush Flush the TLBS when required (should only be false in very specific use cases!!)
4552	*/
4553	int pgmPoolFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fFlush)
4554	{
4555	PVM pVM = pPool->CTX_SUFF(pVM);
4556	bool fFlushRequired = false;
4557
4558	int rc = VINF_SUCCESS;
4559	STAM_PROFILE_START(&pPool->StatFlushPage, f);
4560	LogFlow(("pgmPoolFlushPage: pPage=%p:{.Key=%RHp, .idx=%d, .enmKind=%s, .GCPhys=%RGp}\n",
4561	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
4562
4563	/*
4564	* Quietly reject any attempts at flushing any of the special root pages.
4565	*/
4566	if (pPage->idx < PGMPOOL_IDX_FIRST)
4567	{
4568	AssertFailed(); /* can no longer happen */
4569	Log(("pgmPoolFlushPage: special root page, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
4570	return VINF_SUCCESS;
4571	}
4572
4573	pgmLock(pVM);
4574
4575	/*
4576	* Quietly reject any attempts at flushing the currently active shadow CR3 mapping
4577	*/
4578	if (pgmPoolIsPageLocked(&pVM->pgm.s, pPage))
4579	{
4580	AssertMsg( pPage->enmKind == PGMPOOLKIND_64BIT_PML4
4581	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT
4582	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT_FOR_32BIT
4583	\|\| pPage->enmKind == PGMPOOLKIND_32BIT_PD
4584	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD
4585	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD
4586	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD
4587	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD
4588	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD,
4589	("Can't free the shadow CR3! (%RHp vs %RHp kind=%d\n", PGMGetHyperCR3(VMMGetCpu(pVM)), pPage->Core.Key, pPage->enmKind));
4590	Log(("pgmPoolFlushPage: current active shadow CR3, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
4591	pgmUnlock(pVM);
4592	return VINF_SUCCESS;
4593	}
4594
4595	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
4596	/* Start a subset so we won't run out of mapping space. */
4597	PVMCPU pVCpu = VMMGetCpu(pVM);
4598	uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
4599	#endif
4600
4601	/*
4602	* Mark the page as being in need of an ASMMemZeroPage().
4603	*/
4604	pPage->fZeroed = false;
4605
4606	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
4607	if (pPage->fDirty)
4608	pgmPoolFlushDirtyPage(pVM, pPool, pPage->idxDirty, false /* do not remove */);
4609	#endif
4610
4611	/* If there are any users of this table, then we must issue a tlb flush on all VCPUs. */
4612	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
4613	fFlushRequired = true;
4614
4615	/*
4616	* Clear the page.
4617	*/
4618	pgmPoolTrackClearPageUsers(pPool, pPage);
4619	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
4620	pgmPoolTrackDeref(pPool, pPage);
4621	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
4622
4623	/*
4624	* Flush it from the cache.
4625	*/
4626	pgmPoolCacheFlushPage(pPool, pPage);
4627
4628	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
4629	/* Heavy stuff done. */
4630	PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
4631	#endif
4632
4633	/*
4634	* Deregistering the monitoring.
4635	*/
4636	if (pPage->fMonitored)
4637	rc = pgmPoolMonitorFlush(pPool, pPage);
4638
4639	/*
4640	* Free the page.
4641	*/
4642	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
4643	pPage->iNext = pPool->iFreeHead;
4644	pPool->iFreeHead = pPage->idx;
4645	pPage->enmKind = PGMPOOLKIND_FREE;
4646	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
4647	pPage->GCPhys = NIL_RTGCPHYS;
4648	pPage->fReusedFlushPending = false;
4649
4650	pPool->cUsedPages--;
4651
4652	/* Flush the TLBs of all VCPUs if required. */
4653	if ( fFlushRequired
4654	&& fFlush)
4655	{
4656	PGM_INVL_ALL_VCPU_TLBS(pVM);
4657	}
4658
4659	pgmUnlock(pVM);
4660	STAM_PROFILE_STOP(&pPool->StatFlushPage, f);
4661	return rc;
4662	}
4663
4664
4665	/**
4666	* Frees a usage of a pool page.
4667	*
4668	* The caller is responsible to updating the user table so that it no longer
4669	* references the shadow page.
4670	*
4671	* @param pPool The pool.
4672	* @param HCPhys The HC physical address of the shadow page.
4673	* @param iUser The shadow page pool index of the user table.
4674	* @param iUserTable The index into the user table (shadowed).
4675	*/
4676	void pgmPoolFreeByPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
4677	{
4678	PVM pVM = pPool->CTX_SUFF(pVM);
4679
4680	STAM_PROFILE_START(&pPool->StatFree, a);
4681	LogFlow(("pgmPoolFreeByPage: pPage=%p:{.Key=%RHp, .idx=%d, enmKind=%s} iUser=%#x iUserTable=%#x\n",
4682	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), iUser, iUserTable));
4683	Assert(pPage->idx >= PGMPOOL_IDX_FIRST);
4684	pgmLock(pVM);
4685	pgmPoolTrackFreeUser(pPool, pPage, iUser, iUserTable);
4686	if (!pPage->fCached)
4687	pgmPoolFlushPage(pPool, pPage);
4688	pgmUnlock(pVM);
4689	STAM_PROFILE_STOP(&pPool->StatFree, a);
4690	}
4691
4692
4693	/**
4694	* Makes one or more free page free.
4695	*
4696	* @returns VBox status code.
4697	* @retval VINF_SUCCESS on success.
4698	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
4699	*
4700	* @param pPool The pool.
4701	* @param enmKind Page table kind
4702	* @param iUser The user of the page.
4703	*/
4704	static int pgmPoolMakeMoreFreePages(PPGMPOOL pPool, PGMPOOLKIND enmKind, uint16_t iUser)
4705	{
4706	PVM pVM = pPool->CTX_SUFF(pVM);
4707
4708	LogFlow(("pgmPoolMakeMoreFreePages: iUser=%#x\n", iUser));
4709
4710	/*
4711	* If the pool isn't full grown yet, expand it.
4712	*/
4713	if ( pPool->cCurPages < pPool->cMaxPages
4714	#if defined(IN_RC)
4715	/* Hack alert: we can't deal with jumps to ring 3 when called from MapCR3 and allocating pages for PAE PDs. */
4716	&& enmKind != PGMPOOLKIND_PAE_PD_FOR_PAE_PD
4717	&& (enmKind < PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD \|\| enmKind > PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD)
4718	#endif
4719	)
4720	{
4721	STAM_PROFILE_ADV_SUSPEND(&pPool->StatAlloc, a);
4722	#ifdef IN_RING3
4723	int rc = PGMR3PoolGrow(pVM);
4724	#else
4725	int rc = VMMRZCallRing3NoCpu(pVM, VMMCALLRING3_PGM_POOL_GROW, 0);
4726	#endif
4727	if (RT_FAILURE(rc))
4728	return rc;
4729	STAM_PROFILE_ADV_RESUME(&pPool->StatAlloc, a);
4730	if (pPool->iFreeHead != NIL_PGMPOOL_IDX)
4731	return VINF_SUCCESS;
4732	}
4733
4734	/*
4735	* Free one cached page.
4736	*/
4737	return pgmPoolCacheFreeOne(pPool, iUser);
4738	}
4739
4740	/**
4741	* Allocates a page from the pool.
4742	*
4743	* This page may actually be a cached page and not in need of any processing
4744	* on the callers part.
4745	*
4746	* @returns VBox status code.
4747	* @retval VINF_SUCCESS if a NEW page was allocated.
4748	* @retval VINF_PGM_CACHED_PAGE if a CACHED page was returned.
4749	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
4750	* @param pVM The VM handle.
4751	* @param GCPhys The GC physical address of the page we're gonna shadow.
4752	* For 4MB and 2MB PD entries, it's the first address the
4753	* shadow PT is covering.
4754	* @param enmKind The kind of mapping.
4755	* @param enmAccess Access type for the mapping (only relevant for big pages)
4756	* @param iUser The shadow page pool index of the user table.
4757	* @param iUserTable The index into the user table (shadowed).
4758	* @param ppPage Where to store the pointer to the page. NULL is stored here on failure.
4759	* @param fLockPage Lock the page
4760	*/
4761	int pgmPoolAllocEx(PVM pVM, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, PGMPOOLACCESS enmAccess, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage, bool fLockPage)
4762	{
4763	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4764	STAM_PROFILE_ADV_START(&pPool->StatAlloc, a);
4765	LogFlow(("pgmPoolAlloc: GCPhys=%RGp enmKind=%s iUser=%#x iUserTable=%#x\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable));
4766	*ppPage = NULL;
4767	/** @todo CSAM/PGMPrefetchPage messes up here during CSAMR3CheckGates
4768	* (TRPMR3SyncIDT) because of FF priority. Try fix that?
4769	* Assert(!(pVM->pgm.s.fGlobalSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)); */
4770
4771	pgmLock(pVM);
4772
4773	if (pPool->fCacheEnabled)
4774	{
4775	int rc2 = pgmPoolCacheAlloc(pPool, GCPhys, enmKind, enmAccess, iUser, iUserTable, ppPage);
4776	if (RT_SUCCESS(rc2))
4777	{
4778	if (fLockPage)
4779	pgmPoolLockPage(pPool, *ppPage);
4780	pgmUnlock(pVM);
4781	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4782	LogFlow(("pgmPoolAlloc: cached returns %Rrc ppPage=%p:{.Key=%RHp, .idx=%d}\n", rc2, ppPage, (ppPage)->Core.Key, (ppPage)->idx));
4783	return rc2;
4784	}
4785	}
4786
4787	/*
4788	* Allocate a new one.
4789	*/
4790	int rc = VINF_SUCCESS;
4791	uint16_t iNew = pPool->iFreeHead;
4792	if (iNew == NIL_PGMPOOL_IDX)
4793	{
4794	rc = pgmPoolMakeMoreFreePages(pPool, enmKind, iUser);
4795	if (RT_FAILURE(rc))
4796	{
4797	pgmUnlock(pVM);
4798	Log(("pgmPoolAlloc: returns %Rrc (Free)\n", rc));
4799	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4800	return rc;
4801	}
4802	iNew = pPool->iFreeHead;
4803	AssertReleaseReturn(iNew != NIL_PGMPOOL_IDX, VERR_INTERNAL_ERROR);
4804	}
4805
4806	/* unlink the free head */
4807	PPGMPOOLPAGE pPage = &pPool->aPages[iNew];
4808	pPool->iFreeHead = pPage->iNext;
4809	pPage->iNext = NIL_PGMPOOL_IDX;
4810
4811	/*
4812	* Initialize it.
4813	*/
4814	pPool->cUsedPages++; /* physical handler registration / pgmPoolTrackFlushGCPhysPTsSlow requirement. */
4815	pPage->enmKind = enmKind;
4816	pPage->enmAccess = enmAccess;
4817	pPage->GCPhys = GCPhys;
4818	pPage->fSeenNonGlobal = false; /* Set this to 'true' to disable this feature. */
4819	pPage->fMonitored = false;
4820	pPage->fCached = false;
4821	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
4822	pPage->fDirty = false;
4823	#endif
4824	pPage->fReusedFlushPending = false;
4825	pPage->cModifications = 0;
4826	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4827	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4828	pPage->cLocked = 0;
4829	pPage->cPresent = 0;
4830	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
4831	pPage->pvLastAccessHandlerFault = 0;
4832	pPage->cLastAccessHandlerCount = 0;
4833	pPage->pvLastAccessHandlerRip = 0;
4834
4835	/*
4836	* Insert into the tracking and cache. If this fails, free the page.
4837	*/
4838	int rc3 = pgmPoolTrackInsert(pPool, pPage, GCPhys, iUser, iUserTable);
4839	if (RT_FAILURE(rc3))
4840	{
4841	pPool->cUsedPages--;
4842	pPage->enmKind = PGMPOOLKIND_FREE;
4843	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
4844	pPage->GCPhys = NIL_RTGCPHYS;
4845	pPage->iNext = pPool->iFreeHead;
4846	pPool->iFreeHead = pPage->idx;
4847	pgmUnlock(pVM);
4848	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4849	Log(("pgmPoolAlloc: returns %Rrc (Insert)\n", rc3));
4850	return rc3;
4851	}
4852
4853	/*
4854	* Commit the allocation, clear the page and return.
4855	*/
4856	#ifdef VBOX_WITH_STATISTICS
4857	if (pPool->cUsedPages > pPool->cUsedPagesHigh)
4858	pPool->cUsedPagesHigh = pPool->cUsedPages;
4859	#endif
4860
4861	if (!pPage->fZeroed)
4862	{
4863	STAM_PROFILE_START(&pPool->StatZeroPage, z);
4864	void *pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
4865	ASMMemZeroPage(pv);
4866	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
4867	}
4868
4869	*ppPage = pPage;
4870	if (fLockPage)
4871	pgmPoolLockPage(pPool, pPage);
4872	pgmUnlock(pVM);
4873	LogFlow(("pgmPoolAlloc: returns %Rrc *ppPage=%p:{.Key=%RHp, .idx=%d, .fCached=%RTbool, .fMonitored=%RTbool}\n",
4874	rc, pPage, pPage->Core.Key, pPage->idx, pPage->fCached, pPage->fMonitored));
4875	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4876	return rc;
4877	}
4878
4879
4880	/**
4881	* Frees a usage of a pool page.
4882	*
4883	* @param pVM The VM handle.
4884	* @param HCPhys The HC physical address of the shadow page.
4885	* @param iUser The shadow page pool index of the user table.
4886	* @param iUserTable The index into the user table (shadowed).
4887	*/
4888	void pgmPoolFree(PVM pVM, RTHCPHYS HCPhys, uint16_t iUser, uint32_t iUserTable)
4889	{
4890	LogFlow(("pgmPoolFree: HCPhys=%RHp iUser=%#x iUserTable=%#x\n", HCPhys, iUser, iUserTable));
4891	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4892	pgmPoolFreeByPage(pPool, pgmPoolGetPage(pPool, HCPhys), iUser, iUserTable);
4893	}
4894
4895	/**
4896	* Internal worker for finding a 'in-use' shadow page give by it's physical address.
4897	*
4898	* @returns Pointer to the shadow page structure.
4899	* @param pPool The pool.
4900	* @param HCPhys The HC physical address of the shadow page.
4901	*/
4902	PPGMPOOLPAGE pgmPoolGetPage(PPGMPOOL pPool, RTHCPHYS HCPhys)
4903	{
4904	PVM pVM = pPool->CTX_SUFF(pVM);
4905
4906	Assert(PGMIsLockOwner(pVM));
4907
4908	/*
4909	* Look up the page.
4910	*/
4911	pgmLock(pVM);
4912	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, HCPhys & X86_PTE_PAE_PG_MASK);
4913	pgmUnlock(pVM);
4914
4915	AssertFatalMsg(pPage && pPage->enmKind != PGMPOOLKIND_FREE, ("HCPhys=%RHp pPage=%p idx=%d\n", HCPhys, pPage, (pPage) ? pPage->idx : 0));
4916	return pPage;
4917	}
4918
4919	#ifdef IN_RING3 /* currently only used in ring 3; save some space in the R0 & GC modules (left it here as we might need it elsewhere later on) */
4920	/**
4921	* Flush the specified page if present
4922	*
4923	* @param pVM The VM handle.
4924	* @param GCPhys Guest physical address of the page to flush
4925	*/
4926	void pgmPoolFlushPageByGCPhys(PVM pVM, RTGCPHYS GCPhys)
4927	{
4928	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4929
4930	VM_ASSERT_EMT(pVM);
4931
4932	/*
4933	* Look up the GCPhys in the hash.
4934	*/
4935	GCPhys = GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
4936	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
4937	if (i == NIL_PGMPOOL_IDX)
4938	return;
4939
4940	do
4941	{
4942	PPGMPOOLPAGE pPage = &pPool->aPages[i];
4943	if (pPage->GCPhys - GCPhys < PAGE_SIZE)
4944	{
4945	switch (pPage->enmKind)
4946	{
4947	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
4948	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
4949	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
4950	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
4951	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
4952	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
4953	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
4954	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
4955	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
4956	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
4957	case PGMPOOLKIND_64BIT_PML4:
4958	case PGMPOOLKIND_32BIT_PD:
4959	case PGMPOOLKIND_PAE_PDPT:
4960	{
4961	Log(("PGMPoolFlushPage: found pgm pool pages for %RGp\n", GCPhys));
4962	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
4963	if (pPage->fDirty)
4964	STAM_COUNTER_INC(&pPool->StatForceFlushDirtyPage);
4965	else
4966	#endif
4967	STAM_COUNTER_INC(&pPool->StatForceFlushPage);
4968	Assert(!pgmPoolIsPageLocked(&pVM->pgm.s, pPage));
4969	pgmPoolMonitorChainFlush(pPool, pPage);
4970	return;
4971	}
4972
4973	/* ignore, no monitoring. */
4974	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
4975	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
4976	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
4977	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
4978	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
4979	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
4980	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
4981	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
4982	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
4983	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
4984	case PGMPOOLKIND_ROOT_NESTED:
4985	case PGMPOOLKIND_PAE_PD_PHYS:
4986	case PGMPOOLKIND_PAE_PDPT_PHYS:
4987	case PGMPOOLKIND_32BIT_PD_PHYS:
4988	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
4989	break;
4990
4991	default:
4992	AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
4993	}
4994	}
4995
4996	/* next */
4997	i = pPage->iNext;
4998	} while (i != NIL_PGMPOOL_IDX);
4999	return;
5000	}
5001	#endif /* IN_RING3 */
5002
5003	#ifdef IN_RING3
5004
5005
5006	/**
5007	* Reset CPU on hot plugging.
5008	*
5009	* @param pVM The VM handle.
5010	* @param pVCpu The virtual CPU.
5011	*/
5012	void pgmR3PoolResetUnpluggedCpu(PVM pVM, PVMCPU pVCpu)
5013	{
5014	pgmR3ExitShadowModeBeforePoolFlush(pVM, pVCpu);
5015
5016	pgmR3ReEnterShadowModeAfterPoolFlush(pVM, pVCpu);
5017	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
5018	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
5019	}
5020
5021
5022	/**
5023	* Flushes the entire cache.
5024	*
5025	* It will assert a global CR3 flush (FF) and assumes the caller is aware of
5026	* this and execute this CR3 flush.
5027	*
5028	* @param pPool The pool.
5029	*/
5030	void pgmR3PoolReset(PVM pVM)
5031	{
5032	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
5033
5034	Assert(PGMIsLockOwner(pVM));
5035	STAM_PROFILE_START(&pPool->StatR3Reset, a);
5036	LogFlow(("pgmR3PoolReset:\n"));
5037
5038	/*
5039	* If there are no pages in the pool, there is nothing to do.
5040	*/
5041	if (pPool->cCurPages <= PGMPOOL_IDX_FIRST)
5042	{
5043	STAM_PROFILE_STOP(&pPool->StatR3Reset, a);
5044	return;
5045	}
5046
5047	/*
5048	* Exit the shadow mode since we're going to clear everything,
5049	* including the root page.
5050	*/
5051	for (VMCPUID i = 0; i < pVM->cCpus; i++)
5052	{
5053	PVMCPU pVCpu = &pVM->aCpus[i];
5054	pgmR3ExitShadowModeBeforePoolFlush(pVM, pVCpu);
5055	}
5056
5057	/*
5058	* Nuke the free list and reinsert all pages into it.
5059	*/
5060	for (unsigned i = pPool->cCurPages - 1; i >= PGMPOOL_IDX_FIRST; i--)
5061	{
5062	PPGMPOOLPAGE pPage = &pPool->aPages[i];
5063
5064	Assert(pPage->Core.Key == MMPage2Phys(pVM, pPage->pvPageR3));
5065	if (pPage->fMonitored)
5066	pgmPoolMonitorFlush(pPool, pPage);
5067	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
5068	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
5069	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
5070	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
5071	pPage->cModifications = 0;
5072	pPage->GCPhys = NIL_RTGCPHYS;
5073	pPage->enmKind = PGMPOOLKIND_FREE;
5074	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
5075	Assert(pPage->idx == i);
5076	pPage->iNext = i + 1;
5077	pPage->fZeroed = false; /* This could probably be optimized, but better safe than sorry. */
5078	pPage->fSeenNonGlobal = false;
5079	pPage->fMonitored = false;
5080	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
5081	pPage->fDirty = false;
5082	#endif
5083	pPage->fCached = false;
5084	pPage->fReusedFlushPending = false;
5085	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
5086	pPage->iAgeNext = NIL_PGMPOOL_IDX;
5087	pPage->iAgePrev = NIL_PGMPOOL_IDX;
5088	pPage->cLocked = 0;
5089	}
5090	pPool->aPages[pPool->cCurPages - 1].iNext = NIL_PGMPOOL_IDX;
5091	pPool->iFreeHead = PGMPOOL_IDX_FIRST;
5092	pPool->cUsedPages = 0;
5093
5094	/*
5095	* Zap and reinitialize the user records.
5096	*/
5097	pPool->cPresent = 0;
5098	pPool->iUserFreeHead = 0;
5099	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
5100	const unsigned cMaxUsers = pPool->cMaxUsers;
5101	for (unsigned i = 0; i < cMaxUsers; i++)
5102	{
5103	paUsers[i].iNext = i + 1;
5104	paUsers[i].iUser = NIL_PGMPOOL_IDX;
5105	paUsers[i].iUserTable = 0xfffffffe;
5106	}
5107	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
5108
5109	/*
5110	* Clear all the GCPhys links and rebuild the phys ext free list.
5111	*/
5112	for (PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRanges);
5113	pRam;
5114	pRam = pRam->CTX_SUFF(pNext))
5115	{
5116	unsigned iPage = pRam->cb >> PAGE_SHIFT;
5117	while (iPage-- > 0)
5118	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
5119	}
5120
5121	pPool->iPhysExtFreeHead = 0;
5122	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
5123	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
5124	for (unsigned i = 0; i < cMaxPhysExts; i++)
5125	{
5126	paPhysExts[i].iNext = i + 1;
5127	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
5128	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
5129	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
5130	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
5131	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
5132	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
5133	}
5134	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
5135
5136	/*
5137	* Just zap the modified list.
5138	*/
5139	pPool->cModifiedPages = 0;
5140	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
5141
5142	/*
5143	* Clear the GCPhys hash and the age list.
5144	*/
5145	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
5146	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
5147	pPool->iAgeHead = NIL_PGMPOOL_IDX;
5148	pPool->iAgeTail = NIL_PGMPOOL_IDX;
5149
5150	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
5151	/* Clear all dirty pages. */
5152	pPool->idxFreeDirtyPage = 0;
5153	pPool->cDirtyPages = 0;
5154	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aIdxDirtyPages); i++)
5155	pPool->aIdxDirtyPages[i] = NIL_PGMPOOL_IDX;
5156	#endif
5157
5158	/*
5159	* Reinsert active pages into the hash and ensure monitoring chains are correct.
5160	*/
5161	for (unsigned i = PGMPOOL_IDX_FIRST_SPECIAL; i < PGMPOOL_IDX_FIRST; i++)
5162	{
5163	PPGMPOOLPAGE pPage = &pPool->aPages[i];
5164	pPage->iNext = NIL_PGMPOOL_IDX;
5165	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
5166	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
5167	pPage->cModifications = 0;
5168	/* ASSUMES that we're not sharing with any of the other special pages (safe for now). */
5169	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
5170	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
5171	if (pPage->fMonitored)
5172	{
5173	int rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
5174	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
5175	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
5176	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
5177	pPool->pszAccessHandler);
5178	AssertFatalRCSuccess(rc);
5179	pgmPoolHashInsert(pPool, pPage);
5180	}
5181	Assert(pPage->iUserHead == NIL_PGMPOOL_USER_INDEX); /* for now */
5182	Assert(pPage->iAgeNext == NIL_PGMPOOL_IDX);
5183	Assert(pPage->iAgePrev == NIL_PGMPOOL_IDX);
5184	}
5185
5186	for (VMCPUID i = 0; i < pVM->cCpus; i++)
5187	{
5188	/*
5189	* Re-enter the shadowing mode and assert Sync CR3 FF.
5190	*/
5191	PVMCPU pVCpu = &pVM->aCpus[i];
5192	pgmR3ReEnterShadowModeAfterPoolFlush(pVM, pVCpu);
5193	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
5194	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
5195	}
5196
5197	STAM_PROFILE_STOP(&pPool->StatR3Reset, a);
5198	}
5199	#endif /* IN_RING3 */
5200
5201	#ifdef LOG_ENABLED
5202	static const char *pgmPoolPoolKindToStr(uint8_t enmKind)
5203	{
5204	switch(enmKind)
5205	{
5206	case PGMPOOLKIND_INVALID:
5207	return "PGMPOOLKIND_INVALID";
5208	case PGMPOOLKIND_FREE:
5209	return "PGMPOOLKIND_FREE";
5210	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
5211	return "PGMPOOLKIND_32BIT_PT_FOR_PHYS";
5212	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
5213	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT";
5214	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
5215	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB";
5216	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
5217	return "PGMPOOLKIND_PAE_PT_FOR_PHYS";
5218	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
5219	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_PT";
5220	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
5221	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB";
5222	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
5223	return "PGMPOOLKIND_PAE_PT_FOR_PAE_PT";
5224	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
5225	return "PGMPOOLKIND_PAE_PT_FOR_PAE_2MB";
5226	case PGMPOOLKIND_32BIT_PD:
5227	return "PGMPOOLKIND_32BIT_PD";
5228	case PGMPOOLKIND_32BIT_PD_PHYS:
5229	return "PGMPOOLKIND_32BIT_PD_PHYS";
5230	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
5231	return "PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD";
5232	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
5233	return "PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD";
5234	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
5235	return "PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD";
5236	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
5237	return "PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD";
5238	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
5239	return "PGMPOOLKIND_PAE_PD_FOR_PAE_PD";
5240	case PGMPOOLKIND_PAE_PD_PHYS:
5241	return "PGMPOOLKIND_PAE_PD_PHYS";
5242	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
5243	return "PGMPOOLKIND_PAE_PDPT_FOR_32BIT";
5244	case PGMPOOLKIND_PAE_PDPT:
5245	return "PGMPOOLKIND_PAE_PDPT";
5246	case PGMPOOLKIND_PAE_PDPT_PHYS:
5247	return "PGMPOOLKIND_PAE_PDPT_PHYS";
5248	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
5249	return "PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT";
5250	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
5251	return "PGMPOOLKIND_64BIT_PDPT_FOR_PHYS";
5252	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
5253	return "PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD";
5254	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
5255	return "PGMPOOLKIND_64BIT_PD_FOR_PHYS";
5256	case PGMPOOLKIND_64BIT_PML4:
5257	return "PGMPOOLKIND_64BIT_PML4";
5258	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
5259	return "PGMPOOLKIND_EPT_PDPT_FOR_PHYS";
5260	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
5261	return "PGMPOOLKIND_EPT_PD_FOR_PHYS";
5262	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
5263	return "PGMPOOLKIND_EPT_PT_FOR_PHYS";
5264	case PGMPOOLKIND_ROOT_NESTED:
5265	return "PGMPOOLKIND_ROOT_NESTED";
5266	}
5267	return "Unknown kind!";
5268	}
5269	#endif /* LOG_ENABLED*/

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllPool.cpp@ 29287

Download in other formats: