PGMAllPool.cpp@ 14809

Last change on this file since 14809 was 14809, checked in by vboxsync, 16 years ago
PGMAllPool: RTLogPrintf exist in ring-0 now.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 149.7 KB

Line
1	/* $Id: PGMAllPool.cpp 14809 2008-11-29 23:22:01Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Sun Microsystems, Inc.
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*
17	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
18	* Clara, CA 95054 USA or visit http://www.sun.com if you need
19	* additional information or have any questions.
20	*/
21
22
23	/*******************************************************************************
24	* Header Files *
25	*******************************************************************************/
26	#define LOG_GROUP LOG_GROUP_PGM_POOL
27	#include <VBox/pgm.h>
28	#include <VBox/mm.h>
29	#include <VBox/em.h>
30	#include <VBox/cpum.h>
31	#ifdef IN_RC
32	# include <VBox/patm.h>
33	#endif
34	#include "PGMInternal.h"
35	#include <VBox/vm.h>
36	#include <VBox/disopcode.h>
37	#include <VBox/hwacc_vmx.h>
38
39	#include <VBox/log.h>
40	#include <VBox/err.h>
41	#include <iprt/asm.h>
42
43
44	/*******************************************************************************
45	* Internal Functions *
46	*******************************************************************************/
47	__BEGIN_DECLS
48	static void pgmPoolFlushAllInt(PPGMPOOL pPool);
49	#ifdef PGMPOOL_WITH_USER_TRACKING
50	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind);
51	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind);
52	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
53	#endif
54	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
55	static void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint);
56	#endif
57	#ifdef PGMPOOL_WITH_CACHE
58	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable);
59	#endif
60	#ifdef PGMPOOL_WITH_MONITORING
61	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
62	#endif
63	#ifndef IN_RING3
64	DECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser);
65	#endif
66	__END_DECLS
67
68
69	/**
70	* Checks if the specified page pool kind is for a 4MB or 2MB guest page.
71	*
72	* @returns true if it's the shadow of a 4MB or 2MB guest page, otherwise false.
73	* @param enmKind The page kind.
74	*/
75	DECLINLINE(bool) pgmPoolIsBigPage(PGMPOOLKIND enmKind)
76	{
77	switch (enmKind)
78	{
79	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
80	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
81	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
82	return true;
83	default:
84	return false;
85	}
86	}
87
88
89	#if defined(IN_RC) \|\| defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0)
90	/**
91	* Maps a pool page into the current context.
92	*
93	* @returns Pointer to the mapping.
94	* @param pVM The VM handle.
95	* @param pPage The page to map.
96	*/
97	void *pgmPoolMapPage(PVM pVM, PPGMPOOLPAGE pPage)
98	{
99	/* general pages. */
100	if (pPage->idx >= PGMPOOL_IDX_FIRST)
101	{
102	Assert(pPage->idx < pVM->pgm.s.CTX_SUFF(pPool)->cCurPages);
103	void *pv;
104	int rc = PGMDynMapHCPage(pVM, pPage->Core.Key, &pv);
105	AssertReleaseRC(rc);
106	return pv;
107	}
108
109	/* special pages. */
110	# ifdef IN_RC
111	switch (pPage->idx)
112	{
113	case PGMPOOL_IDX_PD:
114	return pVM->pgm.s.pShw32BitPdRC;
115	case PGMPOOL_IDX_PAE_PD:
116	case PGMPOOL_IDX_PAE_PD_0:
117	return pVM->pgm.s.apShwPaePDsRC[0];
118	case PGMPOOL_IDX_PAE_PD_1:
119	return pVM->pgm.s.apShwPaePDsRC[1];
120	case PGMPOOL_IDX_PAE_PD_2:
121	return pVM->pgm.s.apShwPaePDsRC[2];
122	case PGMPOOL_IDX_PAE_PD_3:
123	return pVM->pgm.s.apShwPaePDsRC[3];
124	case PGMPOOL_IDX_PDPT:
125	return pVM->pgm.s.pShwPaePdptRC;
126	default:
127	AssertReleaseMsgFailed(("Invalid index %d\n", pPage->idx));
128	return NULL;
129	}
130
131	# else /* VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0 */
132	RTHCPHYS HCPhys;
133	switch (pPage->idx)
134	{
135	case PGMPOOL_IDX_PD:
136	HCPhys = pVM->pgm.s.HCPhysShw32BitPD;
137	break;
138	case PGMPOOL_IDX_PAE_PD_0:
139	HCPhys = pVM->pgm.s.aHCPhysPaePDs[0];
140	break;
141	case PGMPOOL_IDX_PAE_PD_1:
142	HCPhys = pVM->pgm.s.aHCPhysPaePDs[1];
143	break;
144	case PGMPOOL_IDX_PAE_PD_2:
145	HCPhys = pVM->pgm.s.aHCPhysPaePDs[2];
146	break;
147	case PGMPOOL_IDX_PAE_PD_3:
148	HCPhys = pVM->pgm.s.aHCPhysPaePDs[3];
149	break;
150	case PGMPOOL_IDX_PDPT:
151	HCPhys = pVM->pgm.s.HCPhysShwPaePdpt;
152	break;
153	case PGMPOOL_IDX_PAE_PD:
154	AssertReleaseMsgFailed(("PGMPOOL_IDX_PAE_PD is not usable in VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0 context\n"));
155	return NULL;
156	default:
157	AssertReleaseMsgFailed(("Invalid index %d\n", pPage->idx));
158	return NULL;
159	}
160	void *pv;
161	int rc = PGMDynMapHCPage(pVM, HCPhys, &pv);
162	AssertReleaseRC(rc);
163	return pv;
164	# endif /* VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0 */
165	}
166	#endif /* IN_RC \|\| VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0 */
167
168
169	#ifdef PGMPOOL_WITH_MONITORING
170	/**
171	* Determin the size of a write instruction.
172	* @returns number of bytes written.
173	* @param pDis The disassembler state.
174	*/
175	static unsigned pgmPoolDisasWriteSize(PDISCPUSTATE pDis)
176	{
177	/*
178	* This is very crude and possibly wrong for some opcodes,
179	* but since it's not really supposed to be called we can
180	* probably live with that.
181	*/
182	return DISGetParamSize(pDis, &pDis->param1);
183	}
184
185
186	/**
187	* Flushes a chain of pages sharing the same access monitor.
188	*
189	* @returns VBox status code suitable for scheduling.
190	* @param pPool The pool.
191	* @param pPage A page in the chain.
192	*/
193	int pgmPoolMonitorChainFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
194	{
195	LogFlow(("pgmPoolMonitorChainFlush: Flush page %RGp type=%d\n", pPage->GCPhys, pPage->enmKind));
196
197	/*
198	* Find the list head.
199	*/
200	uint16_t idx = pPage->idx;
201	if (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
202	{
203	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
204	{
205	idx = pPage->iMonitoredPrev;
206	Assert(idx != pPage->idx);
207	pPage = &pPool->aPages[idx];
208	}
209	}
210
211	/*
212	* Iterate the list flushing each shadow page.
213	*/
214	int rc = VINF_SUCCESS;
215	for (;;)
216	{
217	idx = pPage->iMonitoredNext;
218	Assert(idx != pPage->idx);
219	if (pPage->idx >= PGMPOOL_IDX_FIRST)
220	{
221	int rc2 = pgmPoolFlushPage(pPool, pPage);
222	if (rc2 == VERR_PGM_POOL_CLEARED && rc == VINF_SUCCESS)
223	rc = VINF_PGM_SYNC_CR3;
224	}
225	/* next */
226	if (idx == NIL_PGMPOOL_IDX)
227	break;
228	pPage = &pPool->aPages[idx];
229	}
230	return rc;
231	}
232
233
234	/**
235	* Wrapper for getting the current context pointer to the entry being modified.
236	*
237	* @returns Pointer to the current context mapping of the entry.
238	* @param pPool The pool.
239	* @param pvFault The fault virtual address.
240	* @param GCPhysFault The fault physical address.
241	* @param cbEntry The entry size.
242	*/
243	#ifdef IN_RING3
244	DECLINLINE(const void *) pgmPoolMonitorGCPtr2CCPtr(PPGMPOOL pPool, RTHCPTR pvFault, RTGCPHYS GCPhysFault, const unsigned cbEntry)
245	#else
246	DECLINLINE(const void *) pgmPoolMonitorGCPtr2CCPtr(PPGMPOOL pPool, RTGCPTR pvFault, RTGCPHYS GCPhysFault, const unsigned cbEntry)
247	#endif
248	{
249	#ifdef IN_RC
250	return (const void *)((RTGCUINTPTR)pvFault & ~(RTGCUINTPTR)(cbEntry - 1));
251
252	#elif defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0)
253	void *pvRet;
254	int rc = PGMDynMapGCPageOff(pPool->pVMR0, GCPhysFault & ~(RTGCPHYS)(cbEntry - 1), &pvRet);
255	AssertFatalRCSuccess(rc);
256	return pvRet;
257
258	#elif defined(IN_RING0)
259	void *pvRet;
260	int rc = pgmRamGCPhys2HCPtr(&pPool->pVMR0->pgm.s, GCPhysFault & ~(RTGCPHYS)(cbEntry - 1), &pvRet);
261	AssertFatalRCSuccess(rc);
262	return pvRet;
263
264	#elif defined(IN_RING3)
265	return (RTHCPTR)((uintptr_t)pvFault & ~(RTHCUINTPTR)(cbEntry - 1));
266	#else
267	# error "huh?"
268	#endif
269	}
270
271
272	/**
273	* Process shadow entries before they are changed by the guest.
274	*
275	* For PT entries we will clear them. For PD entries, we'll simply check
276	* for mapping conflicts and set the SyncCR3 FF if found.
277	*
278	* @param pPool The pool.
279	* @param pPage The head page.
280	* @param GCPhysFault The guest physical fault address.
281	* @param uAddress In R0 and GC this is the guest context fault address (flat).
282	* In R3 this is the host context 'fault' address.
283	* @param pCpu The disassembler state for figuring out the write size.
284	* This need not be specified if the caller knows we won't do cross entry accesses.
285	*/
286	#ifdef IN_RING3
287	void pgmPoolMonitorChainChanging(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhysFault, RTHCPTR pvAddress, PDISCPUSTATE pCpu)
288	#else
289	void pgmPoolMonitorChainChanging(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhysFault, RTGCPTR pvAddress, PDISCPUSTATE pCpu)
290	#endif
291	{
292	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
293	const unsigned off = GCPhysFault & PAGE_OFFSET_MASK;
294	const unsigned cbWrite = (pCpu) ? pgmPoolDisasWriteSize(pCpu) : 0;
295
296	LogFlow(("pgmPoolMonitorChainChanging: %RGv phys=%RGp kind=%d cbWrite=%d\n", pvAddress, GCPhysFault, pPage->enmKind, cbWrite));
297
298	for (;;)
299	{
300	union
301	{
302	void *pv;
303	PX86PT pPT;
304	PX86PTPAE pPTPae;
305	PX86PD pPD;
306	PX86PDPAE pPDPae;
307	PX86PDPT pPDPT;
308	PX86PML4 pPML4;
309	} uShw;
310
311	switch (pPage->enmKind)
312	{
313	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
314	{
315	uShw.pv = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
316	const unsigned iShw = off / sizeof(X86PTE);
317	if (uShw.pPT->a[iShw].n.u1Present)
318	{
319	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
320	PCX86PTE pGstPte = (PCX86PTE)pgmPoolMonitorGCPtr2CCPtr(pPool, pvAddress, GCPhysFault, sizeof(*pGstPte));
321	Log4(("pgmPoolMonitorChainChanging 32_32: deref %016RX64 GCPhys %08RX32\n", uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK, pGstPte->u & X86_PTE_PG_MASK));
322	pgmPoolTracDerefGCPhysHint(pPool, pPage,
323	uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK,
324	pGstPte->u & X86_PTE_PG_MASK);
325	# endif
326	uShw.pPT->a[iShw].u = 0;
327	}
328	break;
329	}
330
331	/* page/2 sized */
332	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
333	uShw.pv = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
334	if (!((off ^ pPage->GCPhys) & (PAGE_SIZE / 2)))
335	{
336	const unsigned iShw = (off / sizeof(X86PTE)) & (X86_PG_PAE_ENTRIES - 1);
337	if (uShw.pPTPae->a[iShw].n.u1Present)
338	{
339	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
340	PCX86PTE pGstPte = (PCX86PTE)pgmPoolMonitorGCPtr2CCPtr(pPool, pvAddress, GCPhysFault, sizeof(*pGstPte));
341	Log4(("pgmPoolMonitorChainChanging pae_32: deref %016RX64 GCPhys %08RX32\n", uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK, pGstPte->u & X86_PTE_PG_MASK));
342	pgmPoolTracDerefGCPhysHint(pPool, pPage,
343	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
344	pGstPte->u & X86_PTE_PG_MASK);
345	# endif
346	uShw.pPTPae->a[iShw].u = 0;
347	}
348	}
349	break;
350
351	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
352	{
353	uShw.pv = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
354	const unsigned iShw = off / sizeof(X86PTEPAE);
355	if (uShw.pPTPae->a[iShw].n.u1Present)
356	{
357	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
358	PCX86PTEPAE pGstPte = (PCX86PTEPAE)pgmPoolMonitorGCPtr2CCPtr(pPool, pvAddress, GCPhysFault, sizeof(*pGstPte));
359	Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK, pGstPte->u & X86_PTE_PAE_PG_MASK));
360	pgmPoolTracDerefGCPhysHint(pPool, pPage,
361	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
362	pGstPte->u & X86_PTE_PAE_PG_MASK);
363	# endif
364	uShw.pPTPae->a[iShw].u = 0;
365	}
366
367	/* paranoia / a bit assumptive. */
368	if ( pCpu
369	&& (off & 7)
370	&& (off & 7) + cbWrite > sizeof(X86PTEPAE))
371	{
372	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTEPAE);
373	AssertReturnVoid(iShw2 < RT_ELEMENTS(uShw.pPTPae->a));
374
375	if (uShw.pPTPae->a[iShw2].n.u1Present)
376	{
377	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
378	PCX86PTEPAE pGstPte = (PCX86PTEPAE)pgmPoolMonitorGCPtr2CCPtr(pPool, pvAddress, GCPhysFault, sizeof(*pGstPte));
379	Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK, pGstPte->u & X86_PTE_PAE_PG_MASK));
380	pgmPoolTracDerefGCPhysHint(pPool, pPage,
381	uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK,
382	pGstPte->u & X86_PTE_PAE_PG_MASK);
383	# endif
384	uShw.pPTPae->a[iShw2].u = 0;
385	}
386	}
387
388	break;
389	}
390
391	case PGMPOOLKIND_ROOT_32BIT_PD:
392	{
393	uShw.pv = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
394	const unsigned iShw = off / sizeof(X86PTE); // ASSUMING 32-bit guest paging!
395	if (uShw.pPD->a[iShw].u & PGM_PDFLAGS_MAPPING)
396	{
397	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
398	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
399	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
400	}
401	/* paranoia / a bit assumptive. */
402	else if ( pCpu
403	&& (off & 3)
404	&& (off & 3) + cbWrite > sizeof(X86PTE))
405	{
406	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTE);
407	if ( iShw2 != iShw
408	&& iShw2 < RT_ELEMENTS(uShw.pPD->a)
409	&& uShw.pPD->a[iShw2].u & PGM_PDFLAGS_MAPPING)
410	{
411	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
412	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
413	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
414	}
415	}
416	#if 0 /* useful when running PGMAssertCR3(), a bit too troublesome for general use (TLBs). */
417	if ( uShw.pPD->a[iShw].n.u1Present
418	&& !VM_FF_ISSET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3))
419	{
420	LogFlow(("pgmPoolMonitorChainChanging: iShw=%#x: %RX32 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
421	# ifdef IN_RC /* TLB load - we're pushing things a bit... */
422	ASMProbeReadByte(pvAddress);
423	# endif
424	pgmPoolFree(pPool->CTX_SUFF(pVM), uShw.pPD->a[iShw].u & X86_PDE_PG_MASK, pPage->idx, iShw);
425	uShw.pPD->a[iShw].u = 0;
426	}
427	#endif
428	break;
429	}
430
431	case PGMPOOLKIND_ROOT_PAE_PD:
432	{
433	unsigned iGst = off / sizeof(X86PDE); // ASSUMING 32-bit guest paging!
434	unsigned iShwPdpt = iGst / 256;
435	unsigned iShw = (iGst % 256) * 2;
436	Assert(pPage->idx == PGMPOOL_IDX_PAE_PD);
437	PPGMPOOLPAGE pPage2 = pPage + 1 + iShwPdpt;
438	Assert(pPage2->idx == PGMPOOL_IDX_PAE_PD_0 + iShwPdpt);
439	uShw.pv = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage2);
440	for (unsigned i = 0; i < 2; i++, iShw++)
441	{
442	if ((uShw.pPDPae->a[iShw].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
443	{
444	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
445	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
446	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw=%#x!\n", iShwPdpt, iShw));
447	}
448	/* paranoia / a bit assumptive. */
449	else if ( pCpu
450	&& (off & 3)
451	&& (off & 3) + cbWrite > 4)
452	{
453	const unsigned iShw2 = iShw + 2;
454	if ( iShw2 < RT_ELEMENTS(uShw.pPDPae->a) /** @todo was completely wrong, it's better now after #1865 but still wrong from cross PD. */
455	&& (uShw.pPDPae->a[iShw2].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
456	{
457	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
458	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
459	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw2=%#x!\n", iShwPdpt, iShw2));
460	}
461	}
462	#if 0 /* useful when running PGMAssertCR3(), a bit too troublesome for general use (TLBs). */
463	if ( uShw.pPDPae->a[iShw].n.u1Present
464	&& !VM_FF_ISSET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3))
465	{
466	LogFlow(("pgmPoolMonitorChainChanging: iShwPdpt=%#x iShw=%#x: %RX64 -> freeing it!\n", iShwPdpt, iShw, uShw.pPDPae->a[iShw].u));
467	# ifdef IN_RC /* TLB load - we're pushing things a bit... */
468	ASMProbeReadByte(pvAddress);
469	# endif
470	pgmPoolFree(pPool->CTX_SUFF(pVM), uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK, pPage->idx, iShw + iShwPdpt * X86_PG_PAE_ENTRIES);
471	uShw.pPDPae->a[iShw].u = 0;
472	}
473	#endif
474	}
475	break;
476	}
477
478	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
479	{
480	uShw.pv = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
481	const unsigned iShw = off / sizeof(X86PDEPAE);
482	if (uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING)
483	{
484	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
485	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
486	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
487	}
488	#ifdef PGMPOOL_INVALIDATE_UPPER_SHADOW_TABLE_ENTRIES
489	/*
490	* Causes trouble when the guest uses a PDE to refer to the whole page table level
491	* structure. (Invalidate here; faults later on when it tries to change the page
492	* table entries -> recheck; probably only applies to the RC case.)
493	*/
494	else
495	{
496	if (uShw.pPDPae->a[iShw].n.u1Present)
497	{
498	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
499	pgmPoolFree(pPool->CTX_SUFF(pVM),
500	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
501	/* Note: hardcoded PAE implementation dependency */
502	(pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD) ? PGMPOOL_IDX_PAE_PD : pPage->idx,
503	(pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD) ? iShw + (pPage->idx - PGMPOOL_IDX_PAE_PD_0) * X86_PG_PAE_ENTRIES : iShw);
504	uShw.pPDPae->a[iShw].u = 0;
505	}
506	}
507	#endif
508	/* paranoia / a bit assumptive. */
509	if ( pCpu
510	&& (off & 7)
511	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
512	{
513	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
514	AssertReturnVoid(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
515
516	if ( iShw2 != iShw
517	&& uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING)
518	{
519	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
520	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
521	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
522	}
523	#ifdef PGMPOOL_INVALIDATE_UPPER_SHADOW_TABLE_ENTRIES
524	else if (uShw.pPDPae->a[iShw2].n.u1Present)
525	{
526	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
527	pgmPoolFree(pPool->CTX_SUFF(pVM),
528	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
529	/* Note: hardcoded PAE implementation dependency */
530	(pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD) ? PGMPOOL_IDX_PAE_PD : pPage->idx,
531	(pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD) ? iShw2 + (pPage->idx - PGMPOOL_IDX_PAE_PD_0) * X86_PG_PAE_ENTRIES : iShw2);
532	uShw.pPDPae->a[iShw2].u = 0;
533	}
534	#endif
535	}
536	break;
537	}
538
539	case PGMPOOLKIND_ROOT_PDPT:
540	{
541	/*
542	* Hopefully this doesn't happen very often:
543	* - touching unused parts of the page
544	* - messing with the bits of pd pointers without changing the physical address
545	*/
546	uShw.pv = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
547	const unsigned iShw = off / sizeof(X86PDPE);
548	if (iShw < X86_PG_PAE_PDPE_ENTRIES) /* don't use RT_ELEMENTS(uShw.pPDPT->a), because that's for long mode only */
549	{
550	if (uShw.pPDPT->a[iShw].u & PGM_PLXFLAGS_MAPPING)
551	{
552	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
553	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
554	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
555	}
556	/* paranoia / a bit assumptive. */
557	else if ( pCpu
558	&& (off & 7)
559	&& (off & 7) + cbWrite > sizeof(X86PDPE))
560	{
561	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDPE);
562	if ( iShw2 != iShw
563	&& iShw2 < X86_PG_PAE_PDPE_ENTRIES
564	&& uShw.pPDPT->a[iShw2].u & PGM_PLXFLAGS_MAPPING)
565	{
566	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
567	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
568	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
569	}
570	}
571	}
572	break;
573	}
574
575	#ifndef IN_RC
576	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
577	{
578	Assert(pPage->enmKind == PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD);
579
580	uShw.pv = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
581	const unsigned iShw = off / sizeof(X86PDEPAE);
582	if (uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING)
583	{
584	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
585	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
586	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
587	}
588	else
589	{
590	if (uShw.pPDPae->a[iShw].n.u1Present)
591	{
592	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
593	pgmPoolFree(pPool->CTX_SUFF(pVM),
594	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
595	pPage->idx,
596	iShw);
597	uShw.pPDPae->a[iShw].u = 0;
598	}
599	}
600	/* paranoia / a bit assumptive. */
601	if ( pCpu
602	&& (off & 7)
603	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
604	{
605	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
606	AssertReturnVoid(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
607
608	if ( iShw2 != iShw
609	&& uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING)
610	{
611	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
612	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
613	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
614	}
615	else
616	if (uShw.pPDPae->a[iShw2].n.u1Present)
617	{
618	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
619	pgmPoolFree(pPool->CTX_SUFF(pVM),
620	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
621	pPage->idx,
622	iShw2);
623	uShw.pPDPae->a[iShw2].u = 0;
624	}
625	}
626	break;
627	}
628
629	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
630	{
631	/*
632	* Hopefully this doesn't happen very often:
633	* - messing with the bits of pd pointers without changing the physical address
634	*/
635	if (!VM_FF_ISSET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3))
636	{
637	uShw.pv = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
638	const unsigned iShw = off / sizeof(X86PDPE);
639	if (uShw.pPDPT->a[iShw].n.u1Present)
640	{
641	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
642	pgmPoolFree(pPool->CTX_SUFF(pVM), uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK, pPage->idx, iShw);
643	uShw.pPDPT->a[iShw].u = 0;
644	}
645	/* paranoia / a bit assumptive. */
646	if ( pCpu
647	&& (off & 7)
648	&& (off & 7) + cbWrite > sizeof(X86PDPE))
649	{
650	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDPE);
651	if (uShw.pPDPT->a[iShw2].n.u1Present)
652	{
653	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
654	pgmPoolFree(pPool->CTX_SUFF(pVM), uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK, pPage->idx, iShw2);
655	uShw.pPDPT->a[iShw2].u = 0;
656	}
657	}
658	}
659	break;
660	}
661
662	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
663	{
664	/*
665	* Hopefully this doesn't happen very often:
666	* - messing with the bits of pd pointers without changing the physical address
667	*/
668	if (!VM_FF_ISSET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3))
669	{
670	uShw.pv = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
671	const unsigned iShw = off / sizeof(X86PDPE);
672	if (uShw.pPML4->a[iShw].n.u1Present)
673	{
674	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPML4->a[iShw].u));
675	pgmPoolFree(pPool->CTX_SUFF(pVM), uShw.pPML4->a[iShw].u & X86_PML4E_PG_MASK, pPage->idx, iShw);
676	uShw.pPML4->a[iShw].u = 0;
677	}
678	/* paranoia / a bit assumptive. */
679	if ( pCpu
680	&& (off & 7)
681	&& (off & 7) + cbWrite > sizeof(X86PDPE))
682	{
683	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PML4E);
684	if (uShw.pPML4->a[iShw2].n.u1Present)
685	{
686	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPML4->a[iShw2].u));
687	pgmPoolFree(pPool->CTX_SUFF(pVM), uShw.pPML4->a[iShw2].u & X86_PML4E_PG_MASK, pPage->idx, iShw2);
688	uShw.pPML4->a[iShw2].u = 0;
689	}
690	}
691	}
692	break;
693	}
694	#endif /* IN_RING0 */
695
696	default:
697	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
698	}
699
700	/* next */
701	if (pPage->iMonitoredNext == NIL_PGMPOOL_IDX)
702	return;
703	pPage = &pPool->aPages[pPage->iMonitoredNext];
704	}
705	}
706
707
708	# ifndef IN_RING3
709	/**
710	* Checks if a access could be a fork operation in progress.
711	*
712	* Meaning, that the guest is setuping up the parent process for Copy-On-Write.
713	*
714	* @returns true if it's likly that we're forking, otherwise false.
715	* @param pPool The pool.
716	* @param pCpu The disassembled instruction.
717	* @param offFault The access offset.
718	*/
719	DECLINLINE(bool) pgmPoolMonitorIsForking(PPGMPOOL pPool, PDISCPUSTATE pCpu, unsigned offFault)
720	{
721	/*
722	* i386 linux is using btr to clear X86_PTE_RW.
723	* The functions involved are (2.6.16 source inspection):
724	* clear_bit
725	* ptep_set_wrprotect
726	* copy_one_pte
727	* copy_pte_range
728	* copy_pmd_range
729	* copy_pud_range
730	* copy_page_range
731	* dup_mmap
732	* dup_mm
733	* copy_mm
734	* copy_process
735	* do_fork
736	*/
737	if ( pCpu->pCurInstr->opcode == OP_BTR
738	&& !(offFault & 4)
739	/** @todo Validate that the bit index is X86_PTE_RW. */
740	)
741	{
742	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,Fork));
743	return true;
744	}
745	return false;
746	}
747
748
749	/**
750	* Determine whether the page is likely to have been reused.
751	*
752	* @returns true if we consider the page as being reused for a different purpose.
753	* @returns false if we consider it to still be a paging page.
754	* @param pVM VM Handle.
755	* @param pPage The page in question.
756	* @param pRegFrame Trap register frame.
757	* @param pCpu The disassembly info for the faulting instruction.
758	* @param pvFault The fault address.
759	*
760	* @remark The REP prefix check is left to the caller because of STOSD/W.
761	*/
762	DECLINLINE(bool) pgmPoolMonitorIsReused(PVM pVM, PPGMPOOLPAGE pPage, PCPUMCTXCORE pRegFrame, PDISCPUSTATE pCpu, RTGCPTR pvFault)
763	{
764	#ifndef IN_RC
765	/** @todo could make this general, faulting close to rsp should be safe reuse heuristic. */
766	if ( HWACCMHasPendingIrq(pVM)
767	&& (pRegFrame->rsp - pvFault) < 32)
768	{
769	/* Fault caused by stack writes while trying to inject an interrupt event. */
770	Log(("pgmPoolMonitorIsReused: reused %RGv for interrupt stack (rsp=%RGv).\n", pvFault, pRegFrame->rsp));
771	return true;
772	}
773	#else
774	NOREF(pVM); NOREF(pvFault);
775	#endif
776
777	switch (pCpu->pCurInstr->opcode)
778	{
779	/* call implies the actual push of the return address faulted */
780	case OP_CALL:
781	Log4(("pgmPoolMonitorIsReused: CALL\n"));
782	return true;
783	case OP_PUSH:
784	Log4(("pgmPoolMonitorIsReused: PUSH\n"));
785	return true;
786	case OP_PUSHF:
787	Log4(("pgmPoolMonitorIsReused: PUSHF\n"));
788	return true;
789	case OP_PUSHA:
790	Log4(("pgmPoolMonitorIsReused: PUSHA\n"));
791	return true;
792	case OP_FXSAVE:
793	Log4(("pgmPoolMonitorIsReused: FXSAVE\n"));
794	return true;
795	case OP_MOVNTI: /* solaris - block_zero_no_xmm */
796	Log4(("pgmPoolMonitorIsReused: MOVNTI\n"));
797	return true;
798	case OP_MOVNTDQ: /* solaris - hwblkclr & hwblkpagecopy */
799	Log4(("pgmPoolMonitorIsReused: MOVNTDQ\n"));
800	return true;
801	case OP_MOVSWD:
802	case OP_STOSWD:
803	if ( pCpu->prefix == (PREFIX_REP\|PREFIX_REX)
804	&& pRegFrame->rcx >= 0x40
805	)
806	{
807	Assert(pCpu->mode == CPUMODE_64BIT);
808
809	Log(("pgmPoolMonitorIsReused: OP_STOSQ\n"));
810	return true;
811	}
812	return false;
813	}
814	if ( (pCpu->param1.flags & USE_REG_GEN32)
815	&& (pCpu->param1.base.reg_gen == USE_REG_ESP))
816	{
817	Log4(("pgmPoolMonitorIsReused: ESP\n"));
818	return true;
819	}
820
821	//if (pPage->fCR3Mix)
822	// return false;
823	return false;
824	}
825
826
827	/**
828	* Flushes the page being accessed.
829	*
830	* @returns VBox status code suitable for scheduling.
831	* @param pVM The VM handle.
832	* @param pPool The pool.
833	* @param pPage The pool page (head).
834	* @param pCpu The disassembly of the write instruction.
835	* @param pRegFrame The trap register frame.
836	* @param GCPhysFault The fault address as guest physical address.
837	* @param pvFault The fault address.
838	*/
839	static int pgmPoolAccessHandlerFlush(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
840	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
841	{
842	/*
843	* First, do the flushing.
844	*/
845	int rc = pgmPoolMonitorChainFlush(pPool, pPage);
846
847	/*
848	* Emulate the instruction (xp/w2k problem, requires pc/cr2/sp detection).
849	*/
850	uint32_t cbWritten;
851	int rc2 = EMInterpretInstructionCPU(pVM, pCpu, pRegFrame, pvFault, &cbWritten);
852	if (RT_SUCCESS(rc2))
853	pRegFrame->rip += pCpu->opsize;
854	else if (rc2 == VERR_EM_INTERPRETER)
855	{
856	#ifdef IN_RC
857	if (PATMIsPatchGCAddr(pVM, (RTRCPTR)pRegFrame->eip))
858	{
859	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for patch code %04x:%RGv, ignoring.\n",
860	pRegFrame->cs, (RTGCPTR)pRegFrame->eip));
861	rc = VINF_SUCCESS;
862	STAM_COUNTER_INC(&pPool->StatMonitorRZIntrFailPatch2);
863	}
864	else
865	#endif
866	{
867	rc = VINF_EM_RAW_EMULATE_INSTR;
868	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
869	}
870	}
871	else
872	rc = rc2;
873
874	/* See use in pgmPoolAccessHandlerSimple(). */
875	PGM_INVL_GUEST_TLBS();
876
877	LogFlow(("pgmPoolAccessHandlerPT: returns %Rrc (flushed)\n", rc));
878	return rc;
879
880	}
881
882
883	/**
884	* Handles the STOSD write accesses.
885	*
886	* @returns VBox status code suitable for scheduling.
887	* @param pVM The VM handle.
888	* @param pPool The pool.
889	* @param pPage The pool page (head).
890	* @param pCpu The disassembly of the write instruction.
891	* @param pRegFrame The trap register frame.
892	* @param GCPhysFault The fault address as guest physical address.
893	* @param pvFault The fault address.
894	*/
895	DECLINLINE(int) pgmPoolAccessHandlerSTOSD(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
896	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
897	{
898	Assert(pCpu->mode == CPUMODE_32BIT);
899
900	/*
901	* Increment the modification counter and insert it into the list
902	* of modified pages the first time.
903	*/
904	if (!pPage->cModifications++)
905	pgmPoolMonitorModifiedInsert(pPool, pPage);
906
907	/*
908	* Execute REP STOSD.
909	*
910	* This ASSUMES that we're not invoked by Trap0e on in a out-of-sync
911	* write situation, meaning that it's safe to write here.
912	*/
913	RTGCUINTPTR pu32 = (RTGCUINTPTR)pvFault;
914	while (pRegFrame->ecx)
915	{
916	pgmPoolMonitorChainChanging(pPool, pPage, GCPhysFault, (RTGCPTR)pu32, NULL);
917	#ifdef IN_RC
918	(uint32_t )pu32 = pRegFrame->eax;
919	#else
920	PGMPhysSimpleWriteGCPhys(pVM, GCPhysFault, &pRegFrame->eax, 4);
921	#endif
922	pu32 += 4;
923	GCPhysFault += 4;
924	pRegFrame->edi += 4;
925	pRegFrame->ecx--;
926	}
927	pRegFrame->rip += pCpu->opsize;
928
929	/* See use in pgmPoolAccessHandlerSimple(). */
930	PGM_INVL_GUEST_TLBS();
931
932	LogFlow(("pgmPoolAccessHandlerSTOSD: returns\n"));
933	return VINF_SUCCESS;
934	}
935
936
937	/**
938	* Handles the simple write accesses.
939	*
940	* @returns VBox status code suitable for scheduling.
941	* @param pVM The VM handle.
942	* @param pPool The pool.
943	* @param pPage The pool page (head).
944	* @param pCpu The disassembly of the write instruction.
945	* @param pRegFrame The trap register frame.
946	* @param GCPhysFault The fault address as guest physical address.
947	* @param pvFault The fault address.
948	*/
949	DECLINLINE(int) pgmPoolAccessHandlerSimple(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
950	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
951	{
952	/*
953	* Increment the modification counter and insert it into the list
954	* of modified pages the first time.
955	*/
956	if (!pPage->cModifications++)
957	pgmPoolMonitorModifiedInsert(pPool, pPage);
958
959	/*
960	* Clear all the pages. ASSUMES that pvFault is readable.
961	*/
962	pgmPoolMonitorChainChanging(pPool, pPage, GCPhysFault, pvFault, pCpu);
963
964	/*
965	* Interpret the instruction.
966	*/
967	uint32_t cb;
968	int rc = EMInterpretInstructionCPU(pVM, pCpu, pRegFrame, pvFault, &cb);
969	if (RT_SUCCESS(rc))
970	pRegFrame->rip += pCpu->opsize;
971	else if (rc == VERR_EM_INTERPRETER)
972	{
973	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for %04x:%RGv - opcode=%d\n",
974	pRegFrame->cs, (RTGCPTR)pRegFrame->rip, pCpu->pCurInstr->opcode));
975	rc = VINF_EM_RAW_EMULATE_INSTR;
976	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
977	}
978
979	/*
980	* Quick hack, with logging enabled we're getting stale
981	* code TLBs but no data TLB for EIP and crash in EMInterpretDisasOne.
982	* Flushing here is BAD and expensive, I think EMInterpretDisasOne will
983	* have to be fixed to support this. But that'll have to wait till next week.
984	*
985	* An alternative is to keep track of the changed PTEs together with the
986	* GCPhys from the guest PT. This may proove expensive though.
987	*
988	* At the moment, it's VITAL that it's done AFTER the instruction interpreting
989	* because we need the stale TLBs in some cases (XP boot). This MUST be fixed properly!
990	*/
991	PGM_INVL_GUEST_TLBS();
992
993	LogFlow(("pgmPoolAccessHandlerSimple: returns %Rrc cb=%d\n", rc, cb));
994	return rc;
995	}
996
997
998	/**
999	* \#PF Handler callback for PT write accesses.
1000	*
1001	* @returns VBox status code (appropriate for GC return).
1002	* @param pVM VM Handle.
1003	* @param uErrorCode CPU Error code.
1004	* @param pRegFrame Trap register frame.
1005	* NULL on DMA and other non CPU access.
1006	* @param pvFault The fault address (cr2).
1007	* @param GCPhysFault The GC physical address corresponding to pvFault.
1008	* @param pvUser User argument.
1009	*/
1010	DECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser)
1011	{
1012	STAM_PROFILE_START(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), a);
1013	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1014	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
1015	LogFlow(("pgmPoolAccessHandler: pvFault=%RGv pPage=%p:{.idx=%d} GCPhysFault=%RGp\n", pvFault, pPage, pPage->idx, GCPhysFault));
1016
1017	/*
1018	* We should ALWAYS have the list head as user parameter. This
1019	* is because we use that page to record the changes.
1020	*/
1021	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1022
1023	/*
1024	* Disassemble the faulting instruction.
1025	*/
1026	DISCPUSTATE Cpu;
1027	int rc = EMInterpretDisasOne(pVM, pRegFrame, &Cpu, NULL);
1028	AssertRCReturn(rc, rc);
1029
1030	/*
1031	* Check if it's worth dealing with.
1032	*/
1033	bool fReused = false;
1034	if ( ( pPage->cModifications < 48 /** @todo #define / /* @todo need to check that it's not mapping EIP. / /* @todo adjust this! */
1035	\|\| pPage->fCR3Mix)
1036	&& !(fReused = pgmPoolMonitorIsReused(pVM, pPage, pRegFrame, &Cpu, pvFault))
1037	&& !pgmPoolMonitorIsForking(pPool, &Cpu, GCPhysFault & PAGE_OFFSET_MASK))
1038	{
1039	/*
1040	* Simple instructions, no REP prefix.
1041	*/
1042	if (!(Cpu.prefix & (PREFIX_REP \| PREFIX_REPNE)))
1043	{
1044	rc = pgmPoolAccessHandlerSimple(pVM, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
1045	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,Handled), a);
1046	return rc;
1047	}
1048
1049	/*
1050	* Windows is frequently doing small memset() operations (netio test 4k+).
1051	* We have to deal with these or we'll kill the cache and performance.
1052	*/
1053	if ( Cpu.pCurInstr->opcode == OP_STOSWD
1054	&& CPUMGetGuestCPL(pVM, pRegFrame) == 0
1055	&& pRegFrame->ecx <= 0x20
1056	&& pRegFrame->ecx * 4 <= PAGE_SIZE - ((uintptr_t)pvFault & PAGE_OFFSET_MASK)
1057	&& !((uintptr_t)pvFault & 3)
1058	&& (pRegFrame->eax == 0 \|\| pRegFrame->eax == 0x80) /* the two values observed. */
1059	&& Cpu.mode == CPUMODE_32BIT
1060	&& Cpu.opmode == CPUMODE_32BIT
1061	&& Cpu.addrmode == CPUMODE_32BIT
1062	&& Cpu.prefix == PREFIX_REP
1063	&& !pRegFrame->eflags.Bits.u1DF
1064	)
1065	{
1066	rc = pgmPoolAccessHandlerSTOSD(pVM, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
1067	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,RepStosd), a);
1068	return rc;
1069	}
1070
1071	/* REP prefix, don't bother. */
1072	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,RepPrefix));
1073	Log4(("pgmPoolAccessHandler: eax=%#x ecx=%#x edi=%#x esi=%#x rip=%RGv opcode=%d prefix=%#x\n",
1074	pRegFrame->eax, pRegFrame->ecx, pRegFrame->edi, pRegFrame->esi, (RTGCPTR)pRegFrame->rip, Cpu.pCurInstr->opcode, Cpu.prefix));
1075	}
1076
1077	/*
1078	* Not worth it, so flush it.
1079	*
1080	* If we considered it to be reused, don't to back to ring-3
1081	* to emulate failed instructions since we usually cannot
1082	* interpret then. This may be a bit risky, in which case
1083	* the reuse detection must be fixed.
1084	*/
1085	rc = pgmPoolAccessHandlerFlush(pVM, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
1086	if (rc == VINF_EM_RAW_EMULATE_INSTR && fReused)
1087	rc = VINF_SUCCESS;
1088	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,FlushPage), a);
1089	return rc;
1090	}
1091
1092	# endif /* !IN_RING3 */
1093	#endif /* PGMPOOL_WITH_MONITORING */
1094
1095	#ifdef PGMPOOL_WITH_CACHE
1096
1097	/**
1098	* Inserts a page into the GCPhys hash table.
1099	*
1100	* @param pPool The pool.
1101	* @param pPage The page.
1102	*/
1103	DECLINLINE(void) pgmPoolHashInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1104	{
1105	Log3(("pgmPoolHashInsert: %RGp\n", pPage->GCPhys));
1106	Assert(pPage->GCPhys != NIL_RTGCPHYS); Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1107	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1108	pPage->iNext = pPool->aiHash[iHash];
1109	pPool->aiHash[iHash] = pPage->idx;
1110	}
1111
1112
1113	/**
1114	* Removes a page from the GCPhys hash table.
1115	*
1116	* @param pPool The pool.
1117	* @param pPage The page.
1118	*/
1119	DECLINLINE(void) pgmPoolHashRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1120	{
1121	Log3(("pgmPoolHashRemove: %RGp\n", pPage->GCPhys));
1122	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1123	if (pPool->aiHash[iHash] == pPage->idx)
1124	pPool->aiHash[iHash] = pPage->iNext;
1125	else
1126	{
1127	uint16_t iPrev = pPool->aiHash[iHash];
1128	for (;;)
1129	{
1130	const int16_t i = pPool->aPages[iPrev].iNext;
1131	if (i == pPage->idx)
1132	{
1133	pPool->aPages[iPrev].iNext = pPage->iNext;
1134	break;
1135	}
1136	if (i == NIL_PGMPOOL_IDX)
1137	{
1138	AssertReleaseMsgFailed(("GCPhys=%RGp idx=%#x\n", pPage->GCPhys, pPage->idx));
1139	break;
1140	}
1141	iPrev = i;
1142	}
1143	}
1144	pPage->iNext = NIL_PGMPOOL_IDX;
1145	}
1146
1147
1148	/**
1149	* Frees up one cache page.
1150	*
1151	* @returns VBox status code.
1152	* @retval VINF_SUCCESS on success.
1153	* @retval VERR_PGM_POOL_CLEARED if the deregistration of a physical handler will cause a light weight pool flush.
1154	* @param pPool The pool.
1155	* @param iUser The user index.
1156	*/
1157	static int pgmPoolCacheFreeOne(PPGMPOOL pPool, uint16_t iUser)
1158	{
1159	#ifndef IN_RC
1160	const PVM pVM = pPool->CTX_SUFF(pVM);
1161	#endif
1162	Assert(pPool->iAgeHead != pPool->iAgeTail); /* We shouldn't be here if there < 2 cached entries! */
1163	STAM_COUNTER_INC(&pPool->StatCacheFreeUpOne);
1164
1165	/*
1166	* Select one page from the tail of the age list.
1167	*/
1168	uint16_t iToFree = pPool->iAgeTail;
1169	if (iToFree == iUser)
1170	iToFree = pPool->aPages[iToFree].iAgePrev;
1171	/* This is the alternative to the SyncCR3 pgmPoolCacheUsed calls.
1172	if (pPool->aPages[iToFree].iUserHead != NIL_PGMPOOL_USER_INDEX)
1173	{
1174	uint16_t i = pPool->aPages[iToFree].iAgePrev;
1175	for (unsigned j = 0; j < 10 && i != NIL_PGMPOOL_USER_INDEX; j++, i = pPool->aPages[i].iAgePrev)
1176	{
1177	if (pPool->aPages[iToFree].iUserHead == NIL_PGMPOOL_USER_INDEX)
1178	continue;
1179	iToFree = i;
1180	break;
1181	}
1182	}
1183	*/
1184
1185	Assert(iToFree != iUser);
1186	AssertRelease(iToFree != NIL_PGMPOOL_IDX);
1187
1188	PPGMPOOLPAGE pPage = &pPool->aPages[iToFree];
1189
1190	/*
1191	* Reject any attempts at flushing the currently active shadow CR3 mapping
1192	*/
1193	if (PGMGetHyperCR3(pPool->CTX_SUFF(pVM)) == pPage->Core.Key)
1194	{
1195	/* Refresh the cr3 mapping by putting it at the head of the age list. */
1196	pgmPoolCacheUsed(pPool, pPage);
1197	return pgmPoolCacheFreeOne(pPool, iUser);
1198	}
1199
1200	int rc = pgmPoolFlushPage(pPool, pPage);
1201	if (rc == VINF_SUCCESS)
1202	PGM_INVL_GUEST_TLBS(); /* see PT handler. */
1203	return rc;
1204	}
1205
1206
1207	/**
1208	* Checks if a kind mismatch is really a page being reused
1209	* or if it's just normal remappings.
1210	*
1211	* @returns true if reused and the cached page (enmKind1) should be flushed
1212	* @returns false if not reused.
1213	* @param enmKind1 The kind of the cached page.
1214	* @param enmKind2 The kind of the requested page.
1215	*/
1216	static bool pgmPoolCacheReusedByKind(PGMPOOLKIND enmKind1, PGMPOOLKIND enmKind2)
1217	{
1218	switch (enmKind1)
1219	{
1220	/*
1221	* Never reuse them. There is no remapping in non-paging mode.
1222	*/
1223	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1224	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1225	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1226	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1227	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1228	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1229	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1230	return true;
1231
1232	/*
1233	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1234	*/
1235	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1236	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1237	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1238	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1239	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
1240	switch (enmKind2)
1241	{
1242	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1243	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1244	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1245	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1246	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
1247	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1248	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1249	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1250	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1251	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1252	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1253	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1254	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1255	return true;
1256	default:
1257	return false;
1258	}
1259
1260	/*
1261	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1262	*/
1263	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1264	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1265	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1266	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1267	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
1268	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1269	switch (enmKind2)
1270	{
1271	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1272	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1273	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1274	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1275	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
1276	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1277	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1278	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1279	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1280	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1281	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1282	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1283	return true;
1284	default:
1285	return false;
1286	}
1287
1288	/*
1289	* These cannot be flushed, and it's common to reuse the PDs as PTs.
1290	*/
1291	case PGMPOOLKIND_ROOT_32BIT_PD:
1292	case PGMPOOLKIND_ROOT_PAE_PD:
1293	case PGMPOOLKIND_ROOT_PDPT:
1294	case PGMPOOLKIND_ROOT_NESTED:
1295	return false;
1296
1297	default:
1298	AssertFatalMsgFailed(("enmKind1=%d\n", enmKind1));
1299	}
1300	}
1301
1302
1303	/**
1304	* Attempts to satisfy a pgmPoolAlloc request from the cache.
1305	*
1306	* @returns VBox status code.
1307	* @retval VINF_PGM_CACHED_PAGE on success.
1308	* @retval VERR_FILE_NOT_FOUND if not found.
1309	* @param pPool The pool.
1310	* @param GCPhys The GC physical address of the page we're gonna shadow.
1311	* @param enmKind The kind of mapping.
1312	* @param iUser The shadow page pool index of the user table.
1313	* @param iUserTable The index into the user table (shadowed).
1314	* @param ppPage Where to store the pointer to the page.
1315	*/
1316	static int pgmPoolCacheAlloc(PPGMPOOL pPool, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
1317	{
1318	#ifndef IN_RC
1319	const PVM pVM = pPool->CTX_SUFF(pVM);
1320	#endif
1321	/*
1322	* Look up the GCPhys in the hash.
1323	*/
1324	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
1325	Log3(("pgmPoolCacheAlloc: %RGp kind %d iUser=%d iUserTable=%x SLOT=%d\n", GCPhys, enmKind, iUser, iUserTable, i));
1326	if (i != NIL_PGMPOOL_IDX)
1327	{
1328	do
1329	{
1330	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1331	Log3(("pgmPoolCacheAlloc: slot %d found page %RGp\n", i, pPage->GCPhys));
1332	if (pPage->GCPhys == GCPhys)
1333	{
1334	if ((PGMPOOLKIND)pPage->enmKind == enmKind)
1335	{
1336	int rc = pgmPoolTrackAddUser(pPool, pPage, iUser, iUserTable);
1337	if (RT_SUCCESS(rc))
1338	{
1339	*ppPage = pPage;
1340	STAM_COUNTER_INC(&pPool->StatCacheHits);
1341	return VINF_PGM_CACHED_PAGE;
1342	}
1343	return rc;
1344	}
1345
1346	/*
1347	* The kind is different. In some cases we should now flush the page
1348	* as it has been reused, but in most cases this is normal remapping
1349	* of PDs as PT or big pages using the GCPhys field in a slightly
1350	* different way than the other kinds.
1351	*/
1352	if (pgmPoolCacheReusedByKind((PGMPOOLKIND)pPage->enmKind, enmKind))
1353	{
1354	STAM_COUNTER_INC(&pPool->StatCacheKindMismatches);
1355	pgmPoolFlushPage(pPool, pPage); /* ASSUMES that VERR_PGM_POOL_CLEARED will be returned by pgmPoolTracInsert. */
1356	PGM_INVL_GUEST_TLBS(); /* see PT handler. */
1357	break;
1358	}
1359	}
1360
1361	/* next */
1362	i = pPage->iNext;
1363	} while (i != NIL_PGMPOOL_IDX);
1364	}
1365
1366	Log3(("pgmPoolCacheAlloc: Missed GCPhys=%RGp enmKind=%d\n", GCPhys, enmKind));
1367	STAM_COUNTER_INC(&pPool->StatCacheMisses);
1368	return VERR_FILE_NOT_FOUND;
1369	}
1370
1371
1372	/**
1373	* Inserts a page into the cache.
1374	*
1375	* @param pPool The pool.
1376	* @param pPage The cached page.
1377	* @param fCanBeCached Set if the page is fit for caching from the caller's point of view.
1378	*/
1379	static void pgmPoolCacheInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fCanBeCached)
1380	{
1381	/*
1382	* Insert into the GCPhys hash if the page is fit for that.
1383	*/
1384	Assert(!pPage->fCached);
1385	if (fCanBeCached)
1386	{
1387	pPage->fCached = true;
1388	pgmPoolHashInsert(pPool, pPage);
1389	Log3(("pgmPoolCacheInsert: Caching %p:{.Core=%RHp, .idx=%d, .enmKind=%d, GCPhys=%RGp}\n",
1390	pPage, pPage->Core.Key, pPage->idx, pPage->enmKind, pPage->GCPhys));
1391	STAM_COUNTER_INC(&pPool->StatCacheCacheable);
1392	}
1393	else
1394	{
1395	Log3(("pgmPoolCacheInsert: Not caching %p:{.Core=%RHp, .idx=%d, .enmKind=%d, GCPhys=%RGp}\n",
1396	pPage, pPage->Core.Key, pPage->idx, pPage->enmKind, pPage->GCPhys));
1397	STAM_COUNTER_INC(&pPool->StatCacheUncacheable);
1398	}
1399
1400	/*
1401	* Insert at the head of the age list.
1402	*/
1403	pPage->iAgePrev = NIL_PGMPOOL_IDX;
1404	pPage->iAgeNext = pPool->iAgeHead;
1405	if (pPool->iAgeHead != NIL_PGMPOOL_IDX)
1406	pPool->aPages[pPool->iAgeHead].iAgePrev = pPage->idx;
1407	else
1408	pPool->iAgeTail = pPage->idx;
1409	pPool->iAgeHead = pPage->idx;
1410	}
1411
1412
1413	/**
1414	* Flushes a cached page.
1415	*
1416	* @param pPool The pool.
1417	* @param pPage The cached page.
1418	*/
1419	static void pgmPoolCacheFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1420	{
1421	Log3(("pgmPoolCacheFlushPage: %RGp\n", pPage->GCPhys));
1422
1423	/*
1424	* Remove the page from the hash.
1425	*/
1426	if (pPage->fCached)
1427	{
1428	pPage->fCached = false;
1429	pgmPoolHashRemove(pPool, pPage);
1430	}
1431	else
1432	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1433
1434	/*
1435	* Remove it from the age list.
1436	*/
1437	if (pPage->iAgeNext != NIL_PGMPOOL_IDX)
1438	pPool->aPages[pPage->iAgeNext].iAgePrev = pPage->iAgePrev;
1439	else
1440	pPool->iAgeTail = pPage->iAgePrev;
1441	if (pPage->iAgePrev != NIL_PGMPOOL_IDX)
1442	pPool->aPages[pPage->iAgePrev].iAgeNext = pPage->iAgeNext;
1443	else
1444	pPool->iAgeHead = pPage->iAgeNext;
1445	pPage->iAgeNext = NIL_PGMPOOL_IDX;
1446	pPage->iAgePrev = NIL_PGMPOOL_IDX;
1447	}
1448
1449	#endif /* PGMPOOL_WITH_CACHE */
1450	#ifdef PGMPOOL_WITH_MONITORING
1451
1452	/**
1453	* Looks for pages sharing the monitor.
1454	*
1455	* @returns Pointer to the head page.
1456	* @returns NULL if not found.
1457	* @param pPool The Pool
1458	* @param pNewPage The page which is going to be monitored.
1459	*/
1460	static PPGMPOOLPAGE pgmPoolMonitorGetPageByGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pNewPage)
1461	{
1462	#ifdef PGMPOOL_WITH_CACHE
1463	/*
1464	* Look up the GCPhys in the hash.
1465	*/
1466	RTGCPHYS GCPhys = pNewPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1467	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
1468	if (i == NIL_PGMPOOL_IDX)
1469	return NULL;
1470	do
1471	{
1472	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1473	if ( pPage->GCPhys - GCPhys < PAGE_SIZE
1474	&& pPage != pNewPage)
1475	{
1476	switch (pPage->enmKind)
1477	{
1478	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1479	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1480	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1481	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
1482	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1483	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1484	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1485	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
1486	case PGMPOOLKIND_ROOT_32BIT_PD:
1487	case PGMPOOLKIND_ROOT_PAE_PD:
1488	case PGMPOOLKIND_ROOT_PDPT:
1489	{
1490	/* find the head */
1491	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1492	{
1493	Assert(pPage->iMonitoredPrev != pPage->idx);
1494	pPage = &pPool->aPages[pPage->iMonitoredPrev];
1495	}
1496	return pPage;
1497	}
1498
1499	/* ignore, no monitoring. */
1500	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1501	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1502	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1503	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1504	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1505	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1506	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1507	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1508	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1509	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1510	case PGMPOOLKIND_ROOT_NESTED:
1511	break;
1512	default:
1513	AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
1514	}
1515	}
1516
1517	/* next */
1518	i = pPage->iNext;
1519	} while (i != NIL_PGMPOOL_IDX);
1520	#endif
1521	return NULL;
1522	}
1523
1524
1525	/**
1526	* Enabled write monitoring of a guest page.
1527	*
1528	* @returns VBox status code.
1529	* @retval VINF_SUCCESS on success.
1530	* @retval VERR_PGM_POOL_CLEARED if the registration of the physical handler will cause a light weight pool flush.
1531	* @param pPool The pool.
1532	* @param pPage The cached page.
1533	*/
1534	static int pgmPoolMonitorInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1535	{
1536	LogFlow(("pgmPoolMonitorInsert %RGp\n", pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1)));
1537
1538	/*
1539	* Filter out the relevant kinds.
1540	*/
1541	switch (pPage->enmKind)
1542	{
1543	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1544	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1545	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1546	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1547	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1548	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1549	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
1550	case PGMPOOLKIND_ROOT_PDPT:
1551	break;
1552
1553	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1554	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1555	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1556	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1557	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1558	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1559	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1560	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1561	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1562	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1563	case PGMPOOLKIND_ROOT_NESTED:
1564	/* Nothing to monitor here. */
1565	return VINF_SUCCESS;
1566
1567	case PGMPOOLKIND_ROOT_32BIT_PD:
1568	case PGMPOOLKIND_ROOT_PAE_PD:
1569	#ifdef PGMPOOL_WITH_MIXED_PT_CR3
1570	break;
1571	#endif
1572	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
1573	default:
1574	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
1575	}
1576
1577	/*
1578	* Install handler.
1579	*/
1580	int rc;
1581	PPGMPOOLPAGE pPageHead = pgmPoolMonitorGetPageByGCPhys(pPool, pPage);
1582	if (pPageHead)
1583	{
1584	Assert(pPageHead != pPage); Assert(pPageHead->iMonitoredNext != pPage->idx);
1585	Assert(pPageHead->iMonitoredPrev != pPage->idx);
1586	pPage->iMonitoredPrev = pPageHead->idx;
1587	pPage->iMonitoredNext = pPageHead->iMonitoredNext;
1588	if (pPageHead->iMonitoredNext != NIL_PGMPOOL_IDX)
1589	pPool->aPages[pPageHead->iMonitoredNext].iMonitoredPrev = pPage->idx;
1590	pPageHead->iMonitoredNext = pPage->idx;
1591	rc = VINF_SUCCESS;
1592	}
1593	else
1594	{
1595	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX); Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1596	PVM pVM = pPool->CTX_SUFF(pVM);
1597	const RTGCPHYS GCPhysPage = pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1598	rc = PGMHandlerPhysicalRegisterEx(pVM, PGMPHYSHANDLERTYPE_PHYSICAL_WRITE,
1599	GCPhysPage, GCPhysPage + (PAGE_SIZE - 1),
1600	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
1601	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
1602	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
1603	pPool->pszAccessHandler);
1604	/** @todo we should probably deal with out-of-memory conditions here, but for now increasing
1605	* the heap size should suffice. */
1606	AssertFatalRC(rc);
1607	if (pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
1608	rc = VERR_PGM_POOL_CLEARED;
1609	}
1610	pPage->fMonitored = true;
1611	return rc;
1612	}
1613
1614
1615	/**
1616	* Disables write monitoring of a guest page.
1617	*
1618	* @returns VBox status code.
1619	* @retval VINF_SUCCESS on success.
1620	* @retval VERR_PGM_POOL_CLEARED if the deregistration of the physical handler will cause a light weight pool flush.
1621	* @param pPool The pool.
1622	* @param pPage The cached page.
1623	*/
1624	static int pgmPoolMonitorFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1625	{
1626	/*
1627	* Filter out the relevant kinds.
1628	*/
1629	switch (pPage->enmKind)
1630	{
1631	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1632	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1633	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1634	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1635	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1636	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1637	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
1638	case PGMPOOLKIND_ROOT_PDPT:
1639	break;
1640
1641	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1642	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1643	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1644	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1645	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1646	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1647	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1648	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1649	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1650	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1651	case PGMPOOLKIND_ROOT_NESTED:
1652	/* Nothing to monitor here. */
1653	return VINF_SUCCESS;
1654
1655	case PGMPOOLKIND_ROOT_32BIT_PD:
1656	case PGMPOOLKIND_ROOT_PAE_PD:
1657	#ifdef PGMPOOL_WITH_MIXED_PT_CR3
1658	break;
1659	#endif
1660	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
1661	default:
1662	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
1663	}
1664
1665	/*
1666	* Remove the page from the monitored list or uninstall it if last.
1667	*/
1668	const PVM pVM = pPool->CTX_SUFF(pVM);
1669	int rc;
1670	if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
1671	\|\| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1672	{
1673	if (pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
1674	{
1675	PPGMPOOLPAGE pNewHead = &pPool->aPages[pPage->iMonitoredNext];
1676	pNewHead->iMonitoredPrev = NIL_PGMPOOL_IDX;
1677	pNewHead->fCR3Mix = pPage->fCR3Mix;
1678	rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
1679	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pNewHead),
1680	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pNewHead),
1681	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pNewHead),
1682	pPool->pszAccessHandler);
1683	AssertFatalRCSuccess(rc);
1684	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
1685	}
1686	else
1687	{
1688	pPool->aPages[pPage->iMonitoredPrev].iMonitoredNext = pPage->iMonitoredNext;
1689	if (pPage->iMonitoredNext != NIL_PGMPOOL_IDX)
1690	{
1691	pPool->aPages[pPage->iMonitoredNext].iMonitoredPrev = pPage->iMonitoredPrev;
1692	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
1693	}
1694	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
1695	rc = VINF_SUCCESS;
1696	}
1697	}
1698	else
1699	{
1700	rc = PGMHandlerPhysicalDeregister(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1));
1701	AssertFatalRC(rc);
1702	if (pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
1703	rc = VERR_PGM_POOL_CLEARED;
1704	}
1705	pPage->fMonitored = false;
1706
1707	/*
1708	* Remove it from the list of modified pages (if in it).
1709	*/
1710	pgmPoolMonitorModifiedRemove(pPool, pPage);
1711
1712	return rc;
1713	}
1714
1715	# ifdef PGMPOOL_WITH_MIXED_PT_CR3
1716
1717	/**
1718	* Set or clear the fCR3Mix attribute in a chain of monitored pages.
1719	*
1720	* @param pPool The Pool.
1721	* @param pPage A page in the chain.
1722	* @param fCR3Mix The new fCR3Mix value.
1723	*/
1724	static void pgmPoolMonitorChainChangeCR3Mix(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fCR3Mix)
1725	{
1726	/* current */
1727	pPage->fCR3Mix = fCR3Mix;
1728
1729	/* before */
1730	int16_t idx = pPage->iMonitoredPrev;
1731	while (idx != NIL_PGMPOOL_IDX)
1732	{
1733	pPool->aPages[idx].fCR3Mix = fCR3Mix;
1734	idx = pPool->aPages[idx].iMonitoredPrev;
1735	}
1736
1737	/* after */
1738	idx = pPage->iMonitoredNext;
1739	while (idx != NIL_PGMPOOL_IDX)
1740	{
1741	pPool->aPages[idx].fCR3Mix = fCR3Mix;
1742	idx = pPool->aPages[idx].iMonitoredNext;
1743	}
1744	}
1745
1746
1747	/**
1748	* Installs or modifies monitoring of a CR3 page (special).
1749	*
1750	* We're pretending the CR3 page is shadowed by the pool so we can use the
1751	* generic mechanisms in detecting chained monitoring. (This also gives us a
1752	* tast of what code changes are required to really pool CR3 shadow pages.)
1753	*
1754	* @returns VBox status code.
1755	* @param pPool The pool.
1756	* @param idxRoot The CR3 (root) page index.
1757	* @param GCPhysCR3 The (new) CR3 value.
1758	*/
1759	int pgmPoolMonitorMonitorCR3(PPGMPOOL pPool, uint16_t idxRoot, RTGCPHYS GCPhysCR3)
1760	{
1761	Assert(idxRoot != NIL_PGMPOOL_IDX && idxRoot < PGMPOOL_IDX_FIRST);
1762	PPGMPOOLPAGE pPage = &pPool->aPages[idxRoot];
1763	LogFlow(("pgmPoolMonitorMonitorCR3: idxRoot=%d pPage=%p:{.GCPhys=%RGp, .fMonitored=%d} GCPhysCR3=%RGp\n",
1764	idxRoot, pPage, pPage->GCPhys, pPage->fMonitored, GCPhysCR3));
1765
1766	/*
1767	* The unlikely case where it already matches.
1768	*/
1769	if (pPage->GCPhys == GCPhysCR3)
1770	{
1771	Assert(pPage->fMonitored);
1772	return VINF_SUCCESS;
1773	}
1774
1775	/*
1776	* Flush the current monitoring and remove it from the hash.
1777	*/
1778	int rc = VINF_SUCCESS;
1779	if (pPage->fMonitored)
1780	{
1781	pgmPoolMonitorChainChangeCR3Mix(pPool, pPage, false);
1782	rc = pgmPoolMonitorFlush(pPool, pPage);
1783	if (rc == VERR_PGM_POOL_CLEARED)
1784	rc = VINF_SUCCESS;
1785	else
1786	AssertFatalRC(rc);
1787	pgmPoolHashRemove(pPool, pPage);
1788	}
1789
1790	/*
1791	* Monitor the page at the new location and insert it into the hash.
1792	*/
1793	pPage->GCPhys = GCPhysCR3;
1794	int rc2 = pgmPoolMonitorInsert(pPool, pPage);
1795	if (rc2 != VERR_PGM_POOL_CLEARED)
1796	{
1797	AssertFatalRC(rc2);
1798	if (rc2 != VINF_SUCCESS && rc == VINF_SUCCESS)
1799	rc = rc2;
1800	}
1801	pgmPoolHashInsert(pPool, pPage);
1802	pgmPoolMonitorChainChangeCR3Mix(pPool, pPage, true);
1803	return rc;
1804	}
1805
1806
1807	/**
1808	* Removes the monitoring of a CR3 page (special).
1809	*
1810	* @returns VBox status code.
1811	* @param pPool The pool.
1812	* @param idxRoot The CR3 (root) page index.
1813	*/
1814	int pgmPoolMonitorUnmonitorCR3(PPGMPOOL pPool, uint16_t idxRoot)
1815	{
1816	Assert(idxRoot != NIL_PGMPOOL_IDX && idxRoot < PGMPOOL_IDX_FIRST);
1817	PPGMPOOLPAGE pPage = &pPool->aPages[idxRoot];
1818	LogFlow(("pgmPoolMonitorUnmonitorCR3: idxRoot=%d pPage=%p:{.GCPhys=%RGp, .fMonitored=%d}\n",
1819	idxRoot, pPage, pPage->GCPhys, pPage->fMonitored));
1820
1821	if (!pPage->fMonitored)
1822	return VINF_SUCCESS;
1823
1824	pgmPoolMonitorChainChangeCR3Mix(pPool, pPage, false);
1825	int rc = pgmPoolMonitorFlush(pPool, pPage);
1826	if (rc != VERR_PGM_POOL_CLEARED)
1827	AssertFatalRC(rc);
1828	else
1829	rc = VINF_SUCCESS;
1830	pgmPoolHashRemove(pPool, pPage);
1831	Assert(!pPage->fMonitored);
1832	pPage->GCPhys = NIL_RTGCPHYS;
1833	return rc;
1834	}
1835
1836	# endif /* PGMPOOL_WITH_MIXED_PT_CR3 */
1837
1838	/**
1839	* Inserts the page into the list of modified pages.
1840	*
1841	* @param pPool The pool.
1842	* @param pPage The page.
1843	*/
1844	void pgmPoolMonitorModifiedInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1845	{
1846	Log3(("pgmPoolMonitorModifiedInsert: idx=%d\n", pPage->idx));
1847	AssertMsg( pPage->iModifiedNext == NIL_PGMPOOL_IDX
1848	&& pPage->iModifiedPrev == NIL_PGMPOOL_IDX
1849	&& pPool->iModifiedHead != pPage->idx,
1850	("Next=%d Prev=%d idx=%d cModifications=%d Head=%d cModifiedPages=%d\n",
1851	pPage->iModifiedNext, pPage->iModifiedPrev, pPage->idx, pPage->cModifications,
1852	pPool->iModifiedHead, pPool->cModifiedPages));
1853
1854	pPage->iModifiedNext = pPool->iModifiedHead;
1855	if (pPool->iModifiedHead != NIL_PGMPOOL_IDX)
1856	pPool->aPages[pPool->iModifiedHead].iModifiedPrev = pPage->idx;
1857	pPool->iModifiedHead = pPage->idx;
1858	pPool->cModifiedPages++;
1859	#ifdef VBOX_WITH_STATISTICS
1860	if (pPool->cModifiedPages > pPool->cModifiedPagesHigh)
1861	pPool->cModifiedPagesHigh = pPool->cModifiedPages;
1862	#endif
1863	}
1864
1865
1866	/**
1867	* Removes the page from the list of modified pages and resets the
1868	* moficiation counter.
1869	*
1870	* @param pPool The pool.
1871	* @param pPage The page which is believed to be in the list of modified pages.
1872	*/
1873	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1874	{
1875	Log3(("pgmPoolMonitorModifiedRemove: idx=%d cModifications=%d\n", pPage->idx, pPage->cModifications));
1876	if (pPool->iModifiedHead == pPage->idx)
1877	{
1878	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1879	pPool->iModifiedHead = pPage->iModifiedNext;
1880	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1881	{
1882	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = NIL_PGMPOOL_IDX;
1883	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1884	}
1885	pPool->cModifiedPages--;
1886	}
1887	else if (pPage->iModifiedPrev != NIL_PGMPOOL_IDX)
1888	{
1889	pPool->aPages[pPage->iModifiedPrev].iModifiedNext = pPage->iModifiedNext;
1890	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1891	{
1892	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = pPage->iModifiedPrev;
1893	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1894	}
1895	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1896	pPool->cModifiedPages--;
1897	}
1898	else
1899	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1900	pPage->cModifications = 0;
1901	}
1902
1903
1904	/**
1905	* Zaps the list of modified pages, resetting their modification counters in the process.
1906	*
1907	* @param pVM The VM handle.
1908	*/
1909	void pgmPoolMonitorModifiedClearAll(PVM pVM)
1910	{
1911	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1912	LogFlow(("pgmPoolMonitorModifiedClearAll: cModifiedPages=%d\n", pPool->cModifiedPages));
1913
1914	unsigned cPages = 0; NOREF(cPages);
1915	uint16_t idx = pPool->iModifiedHead;
1916	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
1917	while (idx != NIL_PGMPOOL_IDX)
1918	{
1919	PPGMPOOLPAGE pPage = &pPool->aPages[idx];
1920	idx = pPage->iModifiedNext;
1921	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1922	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1923	pPage->cModifications = 0;
1924	Assert(++cPages);
1925	}
1926	AssertMsg(cPages == pPool->cModifiedPages, ("%d != %d\n", cPages, pPool->cModifiedPages));
1927	pPool->cModifiedPages = 0;
1928	}
1929
1930
1931	/**
1932	* Clear all shadow pages and clear all modification counters.
1933	*
1934	* @param pVM The VM handle.
1935	* @remark Should only be used when monitoring is available, thus placed in
1936	* the PGMPOOL_WITH_MONITORING #ifdef.
1937	*/
1938	void pgmPoolClearAll(PVM pVM)
1939	{
1940	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1941	STAM_PROFILE_START(&pPool->StatClearAll, c);
1942	LogFlow(("pgmPoolClearAll: cUsedPages=%d\n", pPool->cUsedPages));
1943
1944	/*
1945	* Iterate all the pages until we've encountered all that in use.
1946	* This is simple but not quite optimal solution.
1947	*/
1948	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
1949	unsigned cLeft = pPool->cUsedPages;
1950	unsigned iPage = pPool->cCurPages;
1951	while (--iPage >= PGMPOOL_IDX_FIRST)
1952	{
1953	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
1954	if (pPage->GCPhys != NIL_RTGCPHYS)
1955	{
1956	switch (pPage->enmKind)
1957	{
1958	/*
1959	* We only care about shadow page tables.
1960	*/
1961	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1962	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1963	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1964	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1965	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1966	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1967	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1968	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1969	{
1970	#ifdef PGMPOOL_WITH_USER_TRACKING
1971	if (pPage->cPresent)
1972	#endif
1973	{
1974	void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
1975	STAM_PROFILE_START(&pPool->StatZeroPage, z);
1976	ASMMemZeroPage(pvShw);
1977	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
1978	#ifdef PGMPOOL_WITH_USER_TRACKING
1979	pPage->cPresent = 0;
1980	pPage->iFirstPresent = ~0;
1981	#endif
1982	}
1983	}
1984	/* fall thru */
1985
1986	default:
1987	Assert(!pPage->cModifications \|\| ++cModifiedPages);
1988	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
1989	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
1990	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1991	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1992	pPage->cModifications = 0;
1993	break;
1994
1995	}
1996	if (!--cLeft)
1997	break;
1998	}
1999	}
2000
2001	/* swipe the special pages too. */
2002	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
2003	{
2004	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2005	if (pPage->GCPhys != NIL_RTGCPHYS)
2006	{
2007	Assert(!pPage->cModifications \|\| ++cModifiedPages);
2008	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2009	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
2010	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2011	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2012	pPage->cModifications = 0;
2013	}
2014	}
2015
2016	#ifndef DEBUG_michael
2017	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
2018	#endif
2019	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
2020	pPool->cModifiedPages = 0;
2021
2022	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2023	/*
2024	* Clear all the GCPhys links and rebuild the phys ext free list.
2025	*/
2026	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
2027	pRam;
2028	pRam = pRam->CTX_SUFF(pNext))
2029	{
2030	unsigned iPage = pRam->cb >> PAGE_SHIFT;
2031	while (iPage-- > 0)
2032	pRam->aPages[iPage].HCPhys &= MM_RAM_FLAGS_NO_REFS_MASK; /** @todo PAGE FLAGS */
2033	}
2034
2035	pPool->iPhysExtFreeHead = 0;
2036	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
2037	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
2038	for (unsigned i = 0; i < cMaxPhysExts; i++)
2039	{
2040	paPhysExts[i].iNext = i + 1;
2041	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
2042	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
2043	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
2044	}
2045	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
2046	#endif
2047
2048
2049	pPool->cPresent = 0;
2050	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
2051	}
2052
2053
2054	/**
2055	* Handle SyncCR3 pool tasks
2056	*
2057	* @returns VBox status code.
2058	* @retval VINF_SUCCESS if successfully added.
2059	* @retval VINF_PGM_SYNC_CR3 is it needs to be deferred to ring 3 (GC only)
2060	* @param pVM The VM handle.
2061	* @remark Should only be used when monitoring is available, thus placed in
2062	* the PGMPOOL_WITH_MONITORING #ifdef.
2063	*/
2064	int pgmPoolSyncCR3(PVM pVM)
2065	{
2066	/*
2067	* When monitoring shadowed pages, we reset the modification counters on CR3 sync.
2068	* Occasionally we will have to clear all the shadow page tables because we wanted
2069	* to monitor a page which was mapped by too many shadowed page tables. This operation
2070	* sometimes refered to as a 'lightweight flush'.
2071	*/
2072	if (!(pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL))
2073	pgmPoolMonitorModifiedClearAll(pVM);
2074	else
2075	{
2076	# ifndef IN_RC
2077	pVM->pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
2078	pgmPoolClearAll(pVM);
2079	# else
2080	LogFlow(("SyncCR3: PGM_SYNC_CLEAR_PGM_POOL is set -> VINF_PGM_SYNC_CR3\n"));
2081	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
2082	return VINF_PGM_SYNC_CR3;
2083	# endif
2084	}
2085	return VINF_SUCCESS;
2086	}
2087
2088	#endif /* PGMPOOL_WITH_MONITORING */
2089	#ifdef PGMPOOL_WITH_USER_TRACKING
2090
2091	/**
2092	* Frees up at least one user entry.
2093	*
2094	* @returns VBox status code.
2095	* @retval VINF_SUCCESS if successfully added.
2096	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2097	* @param pPool The pool.
2098	* @param iUser The user index.
2099	*/
2100	static int pgmPoolTrackFreeOneUser(PPGMPOOL pPool, uint16_t iUser)
2101	{
2102	STAM_COUNTER_INC(&pPool->StatTrackFreeUpOneUser);
2103	#ifdef PGMPOOL_WITH_CACHE
2104	/*
2105	* Just free cached pages in a braindead fashion.
2106	*/
2107	/** @todo walk the age list backwards and free the first with usage. */
2108	int rc = VINF_SUCCESS;
2109	do
2110	{
2111	int rc2 = pgmPoolCacheFreeOne(pPool, iUser);
2112	if (RT_FAILURE(rc2) && rc == VINF_SUCCESS)
2113	rc = rc2;
2114	} while (pPool->iUserFreeHead == NIL_PGMPOOL_USER_INDEX);
2115	return rc;
2116	#else
2117	/*
2118	* Lazy approach.
2119	*/
2120	/* @todo incompatible with long mode paging (cr3 root will be flushed) */
2121	Assert(!CPUMIsGuestInLongMode(pVM));
2122	pgmPoolFlushAllInt(pPool);
2123	return VERR_PGM_POOL_FLUSHED;
2124	#endif
2125	}
2126
2127
2128	/**
2129	* Inserts a page into the cache.
2130	*
2131	* This will create user node for the page, insert it into the GCPhys
2132	* hash, and insert it into the age list.
2133	*
2134	* @returns VBox status code.
2135	* @retval VINF_SUCCESS if successfully added.
2136	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2137	* @retval VERR_PGM_POOL_CLEARED if the deregistration of the physical handler will cause a light weight pool flush.
2138	* @param pPool The pool.
2139	* @param pPage The cached page.
2140	* @param GCPhys The GC physical address of the page we're gonna shadow.
2141	* @param iUser The user index.
2142	* @param iUserTable The user table index.
2143	*/
2144	DECLINLINE(int) pgmPoolTrackInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhys, uint16_t iUser, uint32_t iUserTable)
2145	{
2146	int rc = VINF_SUCCESS;
2147	PPGMPOOLUSER pUser = pPool->CTX_SUFF(paUsers);
2148
2149	LogFlow(("pgmPoolTrackInsert iUser %d iUserTable %d\n", iUser, iUserTable));
2150
2151	/*
2152	* Find free a user node.
2153	*/
2154	uint16_t i = pPool->iUserFreeHead;
2155	if (i == NIL_PGMPOOL_USER_INDEX)
2156	{
2157	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2158	if (RT_FAILURE(rc))
2159	return rc;
2160	i = pPool->iUserFreeHead;
2161	}
2162
2163	/*
2164	* Unlink the user node from the free list,
2165	* initialize and insert it into the user list.
2166	*/
2167	pPool->iUserFreeHead = pUser[i].iNext;
2168	pUser[i].iNext = NIL_PGMPOOL_USER_INDEX;
2169	pUser[i].iUser = iUser;
2170	pUser[i].iUserTable = iUserTable;
2171	pPage->iUserHead = i;
2172
2173	/*
2174	* Insert into cache and enable monitoring of the guest page if enabled.
2175	*
2176	* Until we implement caching of all levels, including the CR3 one, we'll
2177	* have to make sure we don't try monitor & cache any recursive reuse of
2178	* a monitored CR3 page. Because all windows versions are doing this we'll
2179	* have to be able to do combined access monitoring, CR3 + PT and
2180	* PD + PT (guest PAE).
2181	*
2182	* Update:
2183	* We're now cooperating with the CR3 monitor if an uncachable page is found.
2184	*/
2185	#if defined(PGMPOOL_WITH_MONITORING) \|\| defined(PGMPOOL_WITH_CACHE)
2186	# ifdef PGMPOOL_WITH_MIXED_PT_CR3
2187	const bool fCanBeMonitored = true;
2188	# else
2189	bool fCanBeMonitored = pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored == NIL_RTGCPHYS
2190	\|\| (GCPhys & X86_PTE_PAE_PG_MASK) != (pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored & X86_PTE_PAE_PG_MASK)
2191	\|\| pgmPoolIsBigPage((PGMPOOLKIND)pPage->enmKind);
2192	# endif
2193	# ifdef PGMPOOL_WITH_CACHE
2194	pgmPoolCacheInsert(pPool, pPage, fCanBeMonitored); /* This can be expanded. */
2195	# endif
2196	if (fCanBeMonitored)
2197	{
2198	# ifdef PGMPOOL_WITH_MONITORING
2199	rc = pgmPoolMonitorInsert(pPool, pPage);
2200	if (rc == VERR_PGM_POOL_CLEARED)
2201	{
2202	/* 'Failed' - free the usage, and keep it in the cache (if enabled). */
2203	# ifndef PGMPOOL_WITH_CACHE
2204	pgmPoolMonitorFlush(pPool, pPage);
2205	rc = VERR_PGM_POOL_FLUSHED;
2206	# endif
2207	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
2208	pUser[i].iNext = pPool->iUserFreeHead;
2209	pUser[i].iUser = NIL_PGMPOOL_IDX;
2210	pPool->iUserFreeHead = i;
2211	}
2212	}
2213	# endif
2214	#endif /* PGMPOOL_WITH_MONITORING */
2215	return rc;
2216	}
2217
2218
2219	# ifdef PGMPOOL_WITH_CACHE /* (only used when the cache is enabled.) */
2220	/**
2221	* Adds a user reference to a page.
2222	*
2223	* This will
2224	* This will move the page to the head of the
2225	*
2226	* @returns VBox status code.
2227	* @retval VINF_SUCCESS if successfully added.
2228	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2229	* @param pPool The pool.
2230	* @param pPage The cached page.
2231	* @param iUser The user index.
2232	* @param iUserTable The user table.
2233	*/
2234	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2235	{
2236	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2237
2238	LogFlow(("pgmPoolTrackAddUser iUser %d iUserTable %d\n", iUser, iUserTable));
2239	# ifdef VBOX_STRICT
2240	/*
2241	* Check that the entry doesn't already exists.
2242	*/
2243	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2244	{
2245	uint16_t i = pPage->iUserHead;
2246	do
2247	{
2248	Assert(i < pPool->cMaxUsers);
2249	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2250	i = paUsers[i].iNext;
2251	} while (i != NIL_PGMPOOL_USER_INDEX);
2252	}
2253	# endif
2254
2255	/*
2256	* Allocate a user node.
2257	*/
2258	uint16_t i = pPool->iUserFreeHead;
2259	if (i == NIL_PGMPOOL_USER_INDEX)
2260	{
2261	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2262	if (RT_FAILURE(rc))
2263	return rc;
2264	i = pPool->iUserFreeHead;
2265	}
2266	pPool->iUserFreeHead = paUsers[i].iNext;
2267
2268	/*
2269	* Initialize the user node and insert it.
2270	*/
2271	paUsers[i].iNext = pPage->iUserHead;
2272	paUsers[i].iUser = iUser;
2273	paUsers[i].iUserTable = iUserTable;
2274	pPage->iUserHead = i;
2275
2276	# ifdef PGMPOOL_WITH_CACHE
2277	/*
2278	* Tell the cache to update its replacement stats for this page.
2279	*/
2280	pgmPoolCacheUsed(pPool, pPage);
2281	# endif
2282	return VINF_SUCCESS;
2283	}
2284	# endif /* PGMPOOL_WITH_CACHE */
2285
2286
2287	/**
2288	* Frees a user record associated with a page.
2289	*
2290	* This does not clear the entry in the user table, it simply replaces the
2291	* user record to the chain of free records.
2292	*
2293	* @param pPool The pool.
2294	* @param HCPhys The HC physical address of the shadow page.
2295	* @param iUser The shadow page pool index of the user table.
2296	* @param iUserTable The index into the user table (shadowed).
2297	*/
2298	static void pgmPoolTrackFreeUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2299	{
2300	/*
2301	* Unlink and free the specified user entry.
2302	*/
2303	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2304
2305	/* Special: For PAE and 32-bit paging, there is usually no more than one user. */
2306	uint16_t i = pPage->iUserHead;
2307	if ( i != NIL_PGMPOOL_USER_INDEX
2308	&& paUsers[i].iUser == iUser
2309	&& paUsers[i].iUserTable == iUserTable)
2310	{
2311	pPage->iUserHead = paUsers[i].iNext;
2312
2313	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2314	paUsers[i].iNext = pPool->iUserFreeHead;
2315	pPool->iUserFreeHead = i;
2316	return;
2317	}
2318
2319	/* General: Linear search. */
2320	uint16_t iPrev = NIL_PGMPOOL_USER_INDEX;
2321	while (i != NIL_PGMPOOL_USER_INDEX)
2322	{
2323	if ( paUsers[i].iUser == iUser
2324	&& paUsers[i].iUserTable == iUserTable)
2325	{
2326	if (iPrev != NIL_PGMPOOL_USER_INDEX)
2327	paUsers[iPrev].iNext = paUsers[i].iNext;
2328	else
2329	pPage->iUserHead = paUsers[i].iNext;
2330
2331	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2332	paUsers[i].iNext = pPool->iUserFreeHead;
2333	pPool->iUserFreeHead = i;
2334	return;
2335	}
2336	iPrev = i;
2337	i = paUsers[i].iNext;
2338	}
2339
2340	/* Fatal: didn't find it */
2341	AssertFatalMsgFailed(("Didn't find the user entry! iUser=%#x iUserTable=%#x GCPhys=%RGp\n",
2342	iUser, iUserTable, pPage->GCPhys));
2343	}
2344
2345
2346	/**
2347	* Gets the entry size of a shadow table.
2348	*
2349	* @param enmKind The kind of page.
2350	*
2351	* @returns The size of the entry in bytes. That is, 4 or 8.
2352	* @returns If the kind is not for a table, an assertion is raised and 0 is
2353	* returned.
2354	*/
2355	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind)
2356	{
2357	switch (enmKind)
2358	{
2359	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2360	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2361	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2362	case PGMPOOLKIND_ROOT_32BIT_PD:
2363	return 4;
2364
2365	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2366	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2367	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2368	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2369	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2370	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
2371	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2372	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2373	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2374	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
2375	case PGMPOOLKIND_ROOT_PAE_PD:
2376	case PGMPOOLKIND_ROOT_PDPT:
2377	case PGMPOOLKIND_ROOT_NESTED:
2378	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2379	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2380	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2381	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2382	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2383	return 8;
2384
2385	default:
2386	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2387	}
2388	}
2389
2390
2391	/**
2392	* Gets the entry size of a guest table.
2393	*
2394	* @param enmKind The kind of page.
2395	*
2396	* @returns The size of the entry in bytes. That is, 0, 4 or 8.
2397	* @returns If the kind is not for a table, an assertion is raised and 0 is
2398	* returned.
2399	*/
2400	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind)
2401	{
2402	switch (enmKind)
2403	{
2404	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2405	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2406	case PGMPOOLKIND_ROOT_32BIT_PD:
2407	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2408	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2409	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
2410	return 4;
2411
2412	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2413	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2414	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2415	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2416	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2417	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
2418	case PGMPOOLKIND_ROOT_PAE_PD:
2419	case PGMPOOLKIND_ROOT_PDPT:
2420	return 8;
2421
2422	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2423	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2424	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2425	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2426	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2427	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2428	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2429	case PGMPOOLKIND_ROOT_NESTED:
2430	/** @todo can we return 0? (nobody is calling this...) */
2431	AssertFailed();
2432	return 0;
2433
2434	default:
2435	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2436	}
2437	}
2438
2439	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2440
2441	/**
2442	* Scans one shadow page table for mappings of a physical page.
2443	*
2444	* @param pVM The VM handle.
2445	* @param pPhysPage The guest page in question.
2446	* @param iShw The shadow page table.
2447	* @param cRefs The number of references made in that PT.
2448	*/
2449	static void pgmPoolTrackFlushGCPhysPTInt(PVM pVM, PCPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2450	{
2451	LogFlow(("pgmPoolTrackFlushGCPhysPT: HCPhys=%RHp iShw=%d cRefs=%d\n", pPhysPage->HCPhys, iShw, cRefs));
2452	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2453
2454	/*
2455	* Assert sanity.
2456	*/
2457	Assert(cRefs == 1);
2458	AssertFatalMsg(iShw < pPool->cCurPages && iShw != NIL_PGMPOOL_IDX, ("iShw=%d\n", iShw));
2459	PPGMPOOLPAGE pPage = &pPool->aPages[iShw];
2460
2461	/*
2462	* Then, clear the actual mappings to the page in the shadow PT.
2463	*/
2464	switch (pPage->enmKind)
2465	{
2466	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2467	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2468	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2469	{
2470	const uint32_t u32 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2471	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2472	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2473	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2474	{
2475	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX32 cRefs=%#x\n", i, pPT->a[i], cRefs));
2476	pPT->a[i].u = 0;
2477	cRefs--;
2478	if (!cRefs)
2479	return;
2480	}
2481	#ifdef LOG_ENABLED
2482	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2483	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2484	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2485	{
2486	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2487	pPT->a[i].u = 0;
2488	}
2489	#endif
2490	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2491	break;
2492	}
2493
2494	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2495	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2496	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2497	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2498	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2499	{
2500	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2501	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2502	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2503	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2504	{
2505	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2506	pPT->a[i].u = 0;
2507	cRefs--;
2508	if (!cRefs)
2509	return;
2510	}
2511	#ifdef LOG_ENABLED
2512	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2513	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2514	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2515	{
2516	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2517	pPT->a[i].u = 0;
2518	}
2519	#endif
2520	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d u64=%RX64\n", cRefs, pPage->iFirstPresent, pPage->cPresent, u64));
2521	break;
2522	}
2523
2524	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2525	{
2526	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2527	PEPTPT pPT = (PEPTPT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2528	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2529	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2530	{
2531	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2532	pPT->a[i].u = 0;
2533	cRefs--;
2534	if (!cRefs)
2535	return;
2536	}
2537	#ifdef LOG_ENABLED
2538	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2539	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2540	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2541	{
2542	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2543	pPT->a[i].u = 0;
2544	}
2545	#endif
2546	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2547	break;
2548	}
2549
2550	default:
2551	AssertFatalMsgFailed(("enmKind=%d iShw=%d\n", pPage->enmKind, iShw));
2552	}
2553	}
2554
2555
2556	/**
2557	* Scans one shadow page table for mappings of a physical page.
2558	*
2559	* @param pVM The VM handle.
2560	* @param pPhysPage The guest page in question.
2561	* @param iShw The shadow page table.
2562	* @param cRefs The number of references made in that PT.
2563	*/
2564	void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2565	{
2566	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
2567	LogFlow(("pgmPoolTrackFlushGCPhysPT: HCPhys=%RHp iShw=%d cRefs=%d\n", pPhysPage->HCPhys, iShw, cRefs));
2568	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPT, f);
2569	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, iShw, cRefs);
2570	pPhysPage->HCPhys &= MM_RAM_FLAGS_NO_REFS_MASK; /** @todo PAGE FLAGS */
2571	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPT, f);
2572	}
2573
2574
2575	/**
2576	* Flushes a list of shadow page tables mapping the same physical page.
2577	*
2578	* @param pVM The VM handle.
2579	* @param pPhysPage The guest page in question.
2580	* @param iPhysExt The physical cross reference extent list to flush.
2581	*/
2582	void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, uint16_t iPhysExt)
2583	{
2584	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2585	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTs, f);
2586	LogFlow(("pgmPoolTrackFlushGCPhysPTs: HCPhys=%RHp iPhysExt\n", pPhysPage->HCPhys, iPhysExt));
2587
2588	const uint16_t iPhysExtStart = iPhysExt;
2589	PPGMPOOLPHYSEXT pPhysExt;
2590	do
2591	{
2592	Assert(iPhysExt < pPool->cMaxPhysExts);
2593	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2594	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
2595	if (pPhysExt->aidx[i] != NIL_PGMPOOL_IDX)
2596	{
2597	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, pPhysExt->aidx[i], 1);
2598	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
2599	}
2600
2601	/* next */
2602	iPhysExt = pPhysExt->iNext;
2603	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
2604
2605	/* insert the list into the free list and clear the ram range entry. */
2606	pPhysExt->iNext = pPool->iPhysExtFreeHead;
2607	pPool->iPhysExtFreeHead = iPhysExtStart;
2608	pPhysPage->HCPhys &= MM_RAM_FLAGS_NO_REFS_MASK; /** @todo PAGE FLAGS */
2609
2610	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTs, f);
2611	}
2612
2613	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
2614
2615	/**
2616	* Scans all shadow page tables for mappings of a physical page.
2617	*
2618	* This may be slow, but it's most likely more efficient than cleaning
2619	* out the entire page pool / cache.
2620	*
2621	* @returns VBox status code.
2622	* @retval VINF_SUCCESS if all references has been successfully cleared.
2623	* @retval VINF_PGM_GCPHYS_ALIASED if we're better off with a CR3 sync and
2624	* a page pool cleaning.
2625	*
2626	* @param pVM The VM handle.
2627	* @param pPhysPage The guest page in question.
2628	*/
2629	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage)
2630	{
2631	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2632	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2633	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: cUsedPages=%d cPresent=%d HCPhys=%RHp\n",
2634	pPool->cUsedPages, pPool->cPresent, pPhysPage->HCPhys));
2635
2636	#if 1
2637	/*
2638	* There is a limit to what makes sense.
2639	*/
2640	if (pPool->cPresent > 1024)
2641	{
2642	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
2643	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2644	return VINF_PGM_GCPHYS_ALIASED;
2645	}
2646	#endif
2647
2648	/*
2649	* Iterate all the pages until we've encountered all that in use.
2650	* This is simple but not quite optimal solution.
2651	*/
2652	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2653	const uint32_t u32 = u64;
2654	unsigned cLeft = pPool->cUsedPages;
2655	unsigned iPage = pPool->cCurPages;
2656	while (--iPage >= PGMPOOL_IDX_FIRST)
2657	{
2658	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2659	if (pPage->GCPhys != NIL_RTGCPHYS)
2660	{
2661	switch (pPage->enmKind)
2662	{
2663	/*
2664	* We only care about shadow page tables.
2665	*/
2666	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2667	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2668	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2669	{
2670	unsigned cPresent = pPage->cPresent;
2671	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2672	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2673	if (pPT->a[i].n.u1Present)
2674	{
2675	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2676	{
2677	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX32\n", iPage, i, pPT->a[i]));
2678	pPT->a[i].u = 0;
2679	}
2680	if (!--cPresent)
2681	break;
2682	}
2683	break;
2684	}
2685
2686	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2687	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2688	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2689	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2690	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2691	{
2692	unsigned cPresent = pPage->cPresent;
2693	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2694	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2695	if (pPT->a[i].n.u1Present)
2696	{
2697	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2698	{
2699	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
2700	pPT->a[i].u = 0;
2701	}
2702	if (!--cPresent)
2703	break;
2704	}
2705	break;
2706	}
2707	}
2708	if (!--cLeft)
2709	break;
2710	}
2711	}
2712
2713	pPhysPage->HCPhys &= MM_RAM_FLAGS_NO_REFS_MASK; /** @todo PAGE FLAGS */
2714	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2715	return VINF_SUCCESS;
2716	}
2717
2718
2719	/**
2720	* Clears the user entry in a user table.
2721	*
2722	* This is used to remove all references to a page when flushing it.
2723	*/
2724	static void pgmPoolTrackClearPageUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PCPGMPOOLUSER pUser)
2725	{
2726	Assert(pUser->iUser != NIL_PGMPOOL_IDX);
2727	Assert(pUser->iUser < pPool->cCurPages);
2728	uint32_t iUserTable = pUser->iUserTable;
2729
2730	/*
2731	* Map the user page.
2732	*/
2733	PPGMPOOLPAGE pUserPage = &pPool->aPages[pUser->iUser];
2734	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2735	if (pUserPage->enmKind == PGMPOOLKIND_ROOT_PAE_PD)
2736	{
2737	/* Must translate the fake 2048 entry PD to a 512 PD one since the R0 mapping is not linear. */
2738	Assert(pUser->iUser == PGMPOOL_IDX_PAE_PD);
2739	uint32_t iPdpt = iUserTable / X86_PG_PAE_ENTRIES;
2740	iUserTable %= X86_PG_PAE_ENTRIES;
2741	pUserPage = &pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + iPdpt];
2742	Assert(pUserPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD);
2743	}
2744	#endif
2745	union
2746	{
2747	uint64_t *pau64;
2748	uint32_t *pau32;
2749	} u;
2750	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pUserPage);
2751
2752	/* Safety precaution in case we change the paging for other modes too in the future. */
2753	Assert(PGMGetHyperCR3(pPool->CTX_SUFF(pVM)) != pPage->Core.Key);
2754
2755	#ifdef VBOX_STRICT
2756	/*
2757	* Some sanity checks.
2758	*/
2759	switch (pUserPage->enmKind)
2760	{
2761	case PGMPOOLKIND_ROOT_32BIT_PD:
2762	Assert(iUserTable < X86_PG_ENTRIES);
2763	Assert(!(u.pau32[iUserTable] & PGM_PDFLAGS_MAPPING));
2764	break;
2765	# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2766	case PGMPOOLKIND_ROOT_PAE_PD:
2767	Assert(iUserTable < 2048 && pUser->iUser == PGMPOOL_IDX_PAE_PD);
2768	AssertMsg(!(u.pau64[iUserTable] & PGM_PDFLAGS_MAPPING), ("%llx %d\n", u.pau64[iUserTable], iUserTable));
2769	break;
2770	# endif
2771	case PGMPOOLKIND_ROOT_PDPT:
2772	Assert(iUserTable < 4);
2773	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2774	break;
2775	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
2776	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2777	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2778	break;
2779	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2780	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2781	Assert(!(u.pau64[iUserTable] & PGM_PDFLAGS_MAPPING));
2782	break;
2783	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2784	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2785	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2786	break;
2787	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
2788	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
2789	/* GCPhys >> PAGE_SHIFT is the index here */
2790	break;
2791	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2792	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2793	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2794	break;
2795
2796	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2797	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2798	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2799	break;
2800
2801	case PGMPOOLKIND_ROOT_NESTED:
2802	Assert(iUserTable < X86_PG_PAE_ENTRIES);
2803	break;
2804
2805	default:
2806	AssertMsgFailed(("enmKind=%d\n", pUserPage->enmKind));
2807	break;
2808	}
2809	#endif /* VBOX_STRICT */
2810
2811	/*
2812	* Clear the entry in the user page.
2813	*/
2814	switch (pUserPage->enmKind)
2815	{
2816	/* 32-bit entries */
2817	case PGMPOOLKIND_ROOT_32BIT_PD:
2818	u.pau32[iUserTable] = 0;
2819	break;
2820
2821	/* 64-bit entries */
2822	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
2823	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2824	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2825	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2826	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
2827	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2828	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2829	#ifndef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
2830	case PGMPOOLKIND_ROOT_PAE_PD:
2831	#endif
2832	case PGMPOOLKIND_ROOT_PDPT:
2833	case PGMPOOLKIND_ROOT_NESTED:
2834	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2835	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2836	u.pau64[iUserTable] = 0;
2837	break;
2838
2839	default:
2840	AssertFatalMsgFailed(("enmKind=%d iUser=%#x iUserTable=%#x\n", pUserPage->enmKind, pUser->iUser, pUser->iUserTable));
2841	}
2842	}
2843
2844
2845	/**
2846	* Clears all users of a page.
2847	*/
2848	static void pgmPoolTrackClearPageUsers(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2849	{
2850	/*
2851	* Free all the user records.
2852	*/
2853	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2854	uint16_t i = pPage->iUserHead;
2855	while (i != NIL_PGMPOOL_USER_INDEX)
2856	{
2857	/* Clear enter in user table. */
2858	pgmPoolTrackClearPageUser(pPool, pPage, &paUsers[i]);
2859
2860	/* Free it. */
2861	const uint16_t iNext = paUsers[i].iNext;
2862	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2863	paUsers[i].iNext = pPool->iUserFreeHead;
2864	pPool->iUserFreeHead = i;
2865
2866	/* Next. */
2867	i = iNext;
2868	}
2869	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
2870	}
2871
2872	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2873
2874	/**
2875	* Allocates a new physical cross reference extent.
2876	*
2877	* @returns Pointer to the allocated extent on success. NULL if we're out of them.
2878	* @param pVM The VM handle.
2879	* @param piPhysExt Where to store the phys ext index.
2880	*/
2881	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt)
2882	{
2883	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2884	uint16_t iPhysExt = pPool->iPhysExtFreeHead;
2885	if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
2886	{
2887	STAM_COUNTER_INC(&pPool->StamTrackPhysExtAllocFailures);
2888	return NULL;
2889	}
2890	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2891	pPool->iPhysExtFreeHead = pPhysExt->iNext;
2892	pPhysExt->iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
2893	*piPhysExt = iPhysExt;
2894	return pPhysExt;
2895	}
2896
2897
2898	/**
2899	* Frees a physical cross reference extent.
2900	*
2901	* @param pVM The VM handle.
2902	* @param iPhysExt The extent to free.
2903	*/
2904	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt)
2905	{
2906	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2907	Assert(iPhysExt < pPool->cMaxPhysExts);
2908	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2909	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
2910	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
2911	pPhysExt->iNext = pPool->iPhysExtFreeHead;
2912	pPool->iPhysExtFreeHead = iPhysExt;
2913	}
2914
2915
2916	/**
2917	* Frees a physical cross reference extent.
2918	*
2919	* @param pVM The VM handle.
2920	* @param iPhysExt The extent to free.
2921	*/
2922	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt)
2923	{
2924	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2925
2926	const uint16_t iPhysExtStart = iPhysExt;
2927	PPGMPOOLPHYSEXT pPhysExt;
2928	do
2929	{
2930	Assert(iPhysExt < pPool->cMaxPhysExts);
2931	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2932	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
2933	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
2934
2935	/* next */
2936	iPhysExt = pPhysExt->iNext;
2937	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
2938
2939	pPhysExt->iNext = pPool->iPhysExtFreeHead;
2940	pPool->iPhysExtFreeHead = iPhysExtStart;
2941	}
2942
2943
2944	/**
2945	* Insert a reference into a list of physical cross reference extents.
2946	*
2947	* @returns The new ram range flags (top 16-bits).
2948	*
2949	* @param pVM The VM handle.
2950	* @param iPhysExt The physical extent index of the list head.
2951	* @param iShwPT The shadow page table index.
2952	*
2953	*/
2954	static uint16_t pgmPoolTrackPhysExtInsert(PVM pVM, uint16_t iPhysExt, uint16_t iShwPT)
2955	{
2956	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2957	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
2958
2959	/* special common case. */
2960	if (paPhysExts[iPhysExt].aidx[2] == NIL_PGMPOOL_IDX)
2961	{
2962	paPhysExts[iPhysExt].aidx[2] = iShwPT;
2963	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
2964	LogFlow(("pgmPoolTrackPhysExtAddref: %d:{,,%d}\n", iPhysExt, iShwPT));
2965	return iPhysExt \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
2966	}
2967
2968	/* general treatment. */
2969	const uint16_t iPhysExtStart = iPhysExt;
2970	unsigned cMax = 15;
2971	for (;;)
2972	{
2973	Assert(iPhysExt < pPool->cMaxPhysExts);
2974	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
2975	if (paPhysExts[iPhysExt].aidx[i] == NIL_PGMPOOL_IDX)
2976	{
2977	paPhysExts[iPhysExt].aidx[i] = iShwPT;
2978	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
2979	LogFlow(("pgmPoolTrackPhysExtAddref: %d:{%d} i=%d cMax=%d\n", iPhysExt, iShwPT, i, cMax));
2980	return iPhysExtStart \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
2981	}
2982	if (!--cMax)
2983	{
2984	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
2985	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
2986	LogFlow(("pgmPoolTrackPhysExtAddref: overflow (1) iShwPT=%d\n", iShwPT));
2987	return MM_RAM_FLAGS_IDX_OVERFLOWED \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
2988	}
2989	}
2990
2991	/* add another extent to the list. */
2992	PPGMPOOLPHYSEXT pNew = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
2993	if (!pNew)
2994	{
2995	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
2996	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
2997	return MM_RAM_FLAGS_IDX_OVERFLOWED \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
2998	}
2999	pNew->iNext = iPhysExtStart;
3000	pNew->aidx[0] = iShwPT;
3001	LogFlow(("pgmPoolTrackPhysExtAddref: added new extent %d:{%d}->%d\n", iPhysExt, iShwPT, iPhysExtStart));
3002	return iPhysExt \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
3003	}
3004
3005
3006	/**
3007	* Add a reference to guest physical page where extents are in use.
3008	*
3009	* @returns The new ram range flags (top 16-bits).
3010	*
3011	* @param pVM The VM handle.
3012	* @param u16 The ram range flags (top 16-bits).
3013	* @param iShwPT The shadow page table index.
3014	*/
3015	uint16_t pgmPoolTrackPhysExtAddref(PVM pVM, uint16_t u16, uint16_t iShwPT)
3016	{
3017	if ((u16 >> (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT)) != MM_RAM_FLAGS_CREFS_PHYSEXT)
3018	{
3019	/*
3020	* Convert to extent list.
3021	*/
3022	Assert((u16 >> (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT)) == 1);
3023	uint16_t iPhysExt;
3024	PPGMPOOLPHYSEXT pPhysExt = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
3025	if (pPhysExt)
3026	{
3027	LogFlow(("pgmPoolTrackPhysExtAddref: new extent: %d:{%d, %d}\n", iPhysExt, u16 & MM_RAM_FLAGS_IDX_MASK, iShwPT));
3028	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliased);
3029	pPhysExt->aidx[0] = u16 & MM_RAM_FLAGS_IDX_MASK;
3030	pPhysExt->aidx[1] = iShwPT;
3031	u16 = iPhysExt \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
3032	}
3033	else
3034	u16 = MM_RAM_FLAGS_IDX_OVERFLOWED \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
3035	}
3036	else if (u16 != (MM_RAM_FLAGS_IDX_OVERFLOWED \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT))))
3037	{
3038	/*
3039	* Insert into the extent list.
3040	*/
3041	u16 = pgmPoolTrackPhysExtInsert(pVM, u16 & MM_RAM_FLAGS_IDX_MASK, iShwPT);
3042	}
3043	else
3044	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedLots);
3045	return u16;
3046	}
3047
3048
3049	/**
3050	* Clear references to guest physical memory.
3051	*
3052	* @param pPool The pool.
3053	* @param pPage The page.
3054	* @param pPhysPage Pointer to the aPages entry in the ram range.
3055	*/
3056	void pgmPoolTrackPhysExtDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMPAGE pPhysPage)
3057	{
3058	const unsigned cRefs = pPhysPage->HCPhys >> MM_RAM_FLAGS_CREFS_SHIFT; /** @todo PAGE FLAGS */
3059	AssertFatalMsg(cRefs == MM_RAM_FLAGS_CREFS_PHYSEXT, ("cRefs=%d HCPhys=%RHp pPage=%p:{.idx=%d}\n", cRefs, pPhysPage->HCPhys, pPage, pPage->idx));
3060
3061	uint16_t iPhysExt = (pPhysPage->HCPhys >> MM_RAM_FLAGS_IDX_SHIFT) & MM_RAM_FLAGS_IDX_MASK;
3062	if (iPhysExt != MM_RAM_FLAGS_IDX_OVERFLOWED)
3063	{
3064	uint16_t iPhysExtPrev = NIL_PGMPOOL_PHYSEXT_INDEX;
3065	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3066	do
3067	{
3068	Assert(iPhysExt < pPool->cMaxPhysExts);
3069
3070	/*
3071	* Look for the shadow page and check if it's all freed.
3072	*/
3073	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3074	{
3075	if (paPhysExts[iPhysExt].aidx[i] == pPage->idx)
3076	{
3077	paPhysExts[iPhysExt].aidx[i] = NIL_PGMPOOL_IDX;
3078
3079	for (i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3080	if (paPhysExts[iPhysExt].aidx[i] != NIL_PGMPOOL_IDX)
3081	{
3082	LogFlow(("pgmPoolTrackPhysExtDerefGCPhys: HCPhys=%RX64 idx=%d\n", pPhysPage->HCPhys, pPage->idx));
3083	return;
3084	}
3085
3086	/* we can free the node. */
3087	PVM pVM = pPool->CTX_SUFF(pVM);
3088	const uint16_t iPhysExtNext = paPhysExts[iPhysExt].iNext;
3089	if ( iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX
3090	&& iPhysExtNext == NIL_PGMPOOL_PHYSEXT_INDEX)
3091	{
3092	/* lonely node */
3093	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3094	LogFlow(("pgmPoolTrackPhysExtDerefGCPhys: HCPhys=%RX64 idx=%d lonely\n", pPhysPage->HCPhys, pPage->idx));
3095	pPhysPage->HCPhys &= MM_RAM_FLAGS_NO_REFS_MASK; /** @todo PAGE FLAGS */
3096	}
3097	else if (iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX)
3098	{
3099	/* head */
3100	LogFlow(("pgmPoolTrackPhysExtDerefGCPhys: HCPhys=%RX64 idx=%d head\n", pPhysPage->HCPhys, pPage->idx));
3101	pPhysPage->HCPhys = (pPhysPage->HCPhys & MM_RAM_FLAGS_NO_REFS_MASK) /** @todo PAGE FLAGS */
3102	\| ((uint64_t)MM_RAM_FLAGS_CREFS_PHYSEXT << MM_RAM_FLAGS_CREFS_SHIFT)
3103	\| ((uint64_t)iPhysExtNext << MM_RAM_FLAGS_IDX_SHIFT);
3104	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3105	}
3106	else
3107	{
3108	/* in list */
3109	LogFlow(("pgmPoolTrackPhysExtDerefGCPhys: HCPhys=%RX64 idx=%d\n", pPhysPage->HCPhys, pPage->idx));
3110	paPhysExts[iPhysExtPrev].iNext = iPhysExtNext;
3111	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3112	}
3113	iPhysExt = iPhysExtNext;
3114	return;
3115	}
3116	}
3117
3118	/* next */
3119	iPhysExtPrev = iPhysExt;
3120	iPhysExt = paPhysExts[iPhysExt].iNext;
3121	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3122
3123	AssertFatalMsgFailed(("not-found! cRefs=%d HCPhys=%RHp pPage=%p:{.idx=%d}\n", cRefs, pPhysPage->HCPhys, pPage, pPage->idx));
3124	}
3125	else /* nothing to do */
3126	LogFlow(("pgmPoolTrackPhysExtDerefGCPhys: HCPhys=%RX64\n", pPhysPage->HCPhys));
3127	}
3128
3129
3130	/**
3131	* Clear references to guest physical memory.
3132	*
3133	* This is the same as pgmPoolTracDerefGCPhys except that the guest physical address
3134	* is assumed to be correct, so the linear search can be skipped and we can assert
3135	* at an earlier point.
3136	*
3137	* @param pPool The pool.
3138	* @param pPage The page.
3139	* @param HCPhys The host physical address corresponding to the guest page.
3140	* @param GCPhys The guest physical address corresponding to HCPhys.
3141	*/
3142	static void pgmPoolTracDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhys)
3143	{
3144	/*
3145	* Walk range list.
3146	*/
3147	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3148	while (pRam)
3149	{
3150	RTGCPHYS off = GCPhys - pRam->GCPhys;
3151	if (off < pRam->cb)
3152	{
3153	/* does it match? */
3154	const unsigned iPage = off >> PAGE_SHIFT;
3155	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3156	#ifdef LOG_ENABLED
3157	RTHCPHYS HCPhysPage = PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]);
3158	Log(("pgmPoolTracDerefGCPhys %RHp vs %RHp\n", HCPhysPage, HCPhys));
3159	#endif
3160	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3161	{
3162	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3163	return;
3164	}
3165	break;
3166	}
3167	pRam = pRam->CTX_SUFF(pNext);
3168	}
3169	AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp\n", HCPhys, GCPhys));
3170	}
3171
3172
3173	/**
3174	* Clear references to guest physical memory.
3175	*
3176	* @param pPool The pool.
3177	* @param pPage The page.
3178	* @param HCPhys The host physical address corresponding to the guest page.
3179	* @param GCPhysHint The guest physical address which may corresponding to HCPhys.
3180	*/
3181	static void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint)
3182	{
3183	/*
3184	* Walk range list.
3185	*/
3186	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3187	while (pRam)
3188	{
3189	RTGCPHYS off = GCPhysHint - pRam->GCPhys;
3190	if (off < pRam->cb)
3191	{
3192	/* does it match? */
3193	const unsigned iPage = off >> PAGE_SHIFT;
3194	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3195	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3196	{
3197	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3198	return;
3199	}
3200	break;
3201	}
3202	pRam = pRam->CTX_SUFF(pNext);
3203	}
3204
3205	/*
3206	* Damn, the hint didn't work. We'll have to do an expensive linear search.
3207	*/
3208	STAM_COUNTER_INC(&pPool->StatTrackLinearRamSearches);
3209	pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3210	while (pRam)
3211	{
3212	unsigned iPage = pRam->cb >> PAGE_SHIFT;
3213	while (iPage-- > 0)
3214	{
3215	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3216	{
3217	Log4(("pgmPoolTracDerefGCPhysHint: Linear HCPhys=%RHp GCPhysHint=%RGp GCPhysReal=%RGp\n",
3218	HCPhys, GCPhysHint, pRam->GCPhys + (iPage << PAGE_SHIFT)));
3219	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3220	return;
3221	}
3222	}
3223	pRam = pRam->CTX_SUFF(pNext);
3224	}
3225
3226	AssertFatalMsgFailed(("HCPhys=%RHp GCPhysHint=%RGp\n", HCPhys, GCPhysHint));
3227	}
3228
3229
3230	/**
3231	* Clear references to guest physical memory in a 32-bit / 32-bit page table.
3232	*
3233	* @param pPool The pool.
3234	* @param pPage The page.
3235	* @param pShwPT The shadow page table (mapping of the page).
3236	* @param pGstPT The guest page table.
3237	*/
3238	DECLINLINE(void) pgmPoolTrackDerefPT32Bit32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT, PCX86PT pGstPT)
3239	{
3240	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
3241	if (pShwPT->a[i].n.u1Present)
3242	{
3243	Log4(("pgmPoolTrackDerefPT32Bit32Bit: i=%d pte=%RX32 hint=%RX32\n",
3244	i, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3245	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3246	if (!--pPage->cPresent)
3247	break;
3248	}
3249	}
3250
3251
3252	/**
3253	* Clear references to guest physical memory in a PAE / 32-bit page table.
3254	*
3255	* @param pPool The pool.
3256	* @param pPage The page.
3257	* @param pShwPT The shadow page table (mapping of the page).
3258	* @param pGstPT The guest page table (just a half one).
3259	*/
3260	DECLINLINE(void) pgmPoolTrackDerefPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PT pGstPT)
3261	{
3262	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3263	if (pShwPT->a[i].n.u1Present)
3264	{
3265	Log4(("pgmPoolTrackDerefPTPae32Bit: i=%d pte=%RX32 hint=%RX32\n",
3266	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3267	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3268	}
3269	}
3270
3271
3272	/**
3273	* Clear references to guest physical memory in a PAE / PAE page table.
3274	*
3275	* @param pPool The pool.
3276	* @param pPage The page.
3277	* @param pShwPT The shadow page table (mapping of the page).
3278	* @param pGstPT The guest page table.
3279	*/
3280	DECLINLINE(void) pgmPoolTrackDerefPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PTPAE pGstPT)
3281	{
3282	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3283	if (pShwPT->a[i].n.u1Present)
3284	{
3285	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX32 hint=%RX32\n",
3286	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
3287	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK);
3288	}
3289	}
3290
3291
3292	/**
3293	* Clear references to guest physical memory in a 32-bit / 4MB page table.
3294	*
3295	* @param pPool The pool.
3296	* @param pPage The page.
3297	* @param pShwPT The shadow page table (mapping of the page).
3298	*/
3299	DECLINLINE(void) pgmPoolTrackDerefPT32Bit4MB(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT)
3300	{
3301	RTGCPHYS GCPhys = pPage->GCPhys;
3302	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3303	if (pShwPT->a[i].n.u1Present)
3304	{
3305	Log4(("pgmPoolTrackDerefPT32Bit4MB: i=%d pte=%RX32 GCPhys=%RGp\n",
3306	i, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys));
3307	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys);
3308	}
3309	}
3310
3311
3312	/**
3313	* Clear references to guest physical memory in a PAE / 2/4MB page table.
3314	*
3315	* @param pPool The pool.
3316	* @param pPage The page.
3317	* @param pShwPT The shadow page table (mapping of the page).
3318	*/
3319	DECLINLINE(void) pgmPoolTrackDerefPTPaeBig(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT)
3320	{
3321	RTGCPHYS GCPhys = pPage->GCPhys;
3322	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3323	if (pShwPT->a[i].n.u1Present)
3324	{
3325	Log4(("pgmPoolTrackDerefPTPaeBig: i=%d pte=%RX64 hint=%RGp\n",
3326	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys));
3327	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys);
3328	}
3329	}
3330
3331	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
3332
3333	/**
3334	* Clear references to shadowed pages in a PAE (legacy or 64 bits) page directory.
3335	*
3336	* @param pPool The pool.
3337	* @param pPage The page.
3338	* @param pShwPD The shadow page directory (mapping of the page).
3339	*/
3340	DECLINLINE(void) pgmPoolTrackDerefPDPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPAE pShwPD)
3341	{
3342	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3343	{
3344	if (pShwPD->a[i].n.u1Present)
3345	{
3346	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PAE_PG_MASK);
3347	if (pSubPage)
3348	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3349	else
3350	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & X86_PDE_PAE_PG_MASK));
3351	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3352	}
3353	}
3354	}
3355
3356
3357	/**
3358	* Clear references to shadowed pages in a 64-bit page directory pointer table.
3359	*
3360	* @param pPool The pool.
3361	* @param pPage The page.
3362	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3363	*/
3364	DECLINLINE(void) pgmPoolTrackDerefPDPT64Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
3365	{
3366	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3367	{
3368	if (pShwPDPT->a[i].n.u1Present)
3369	{
3370	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
3371	if (pSubPage)
3372	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3373	else
3374	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
3375	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3376	}
3377	}
3378	}
3379
3380
3381	/**
3382	* Clear references to shadowed pages in a 64-bit level 4 page table.
3383	*
3384	* @param pPool The pool.
3385	* @param pPage The page.
3386	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3387	*/
3388	DECLINLINE(void) pgmPoolTrackDerefPML464Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PML4 pShwPML4)
3389	{
3390	for (unsigned i = 0; i < RT_ELEMENTS(pShwPML4->a); i++)
3391	{
3392	if (pShwPML4->a[i].n.u1Present)
3393	{
3394	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPML4->a[i].u & X86_PDPE_PG_MASK);
3395	if (pSubPage)
3396	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3397	else
3398	AssertFatalMsgFailed(("%RX64\n", pShwPML4->a[i].u & X86_PML4E_PG_MASK));
3399	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3400	}
3401	}
3402	}
3403
3404
3405	/**
3406	* Clear references to shadowed pages in an EPT page table.
3407	*
3408	* @param pPool The pool.
3409	* @param pPage The page.
3410	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3411	*/
3412	DECLINLINE(void) pgmPoolTrackDerefPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT)
3413	{
3414	RTGCPHYS GCPhys = pPage->GCPhys;
3415	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3416	if (pShwPT->a[i].n.u1Present)
3417	{
3418	Log4(("pgmPoolTrackDerefPTEPT: i=%d pte=%RX64 GCPhys=%RX64\n",
3419	i, pShwPT->a[i].u & EPT_PTE_PG_MASK, pPage->GCPhys));
3420	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & EPT_PTE_PG_MASK, GCPhys);
3421	}
3422	}
3423
3424
3425	/**
3426	* Clear references to shadowed pages in an EPT page directory.
3427	*
3428	* @param pPool The pool.
3429	* @param pPage The page.
3430	* @param pShwPD The shadow page directory (mapping of the page).
3431	*/
3432	DECLINLINE(void) pgmPoolTrackDerefPDEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPD pShwPD)
3433	{
3434	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3435	{
3436	if (pShwPD->a[i].n.u1Present)
3437	{
3438	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & EPT_PDE_PG_MASK);
3439	if (pSubPage)
3440	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3441	else
3442	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & EPT_PDE_PG_MASK));
3443	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3444	}
3445	}
3446	}
3447
3448
3449	/**
3450	* Clear references to shadowed pages in an EPT page directory pointer table.
3451	*
3452	* @param pPool The pool.
3453	* @param pPage The page.
3454	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3455	*/
3456	DECLINLINE(void) pgmPoolTrackDerefPDPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPDPT pShwPDPT)
3457	{
3458	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3459	{
3460	if (pShwPDPT->a[i].n.u1Present)
3461	{
3462	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK);
3463	if (pSubPage)
3464	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3465	else
3466	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK));
3467	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3468	}
3469	}
3470	}
3471
3472
3473	/**
3474	* Clears all references made by this page.
3475	*
3476	* This includes other shadow pages and GC physical addresses.
3477	*
3478	* @param pPool The pool.
3479	* @param pPage The page.
3480	*/
3481	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3482	{
3483	/*
3484	* Map the shadow page and take action according to the page kind.
3485	*/
3486	void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
3487	switch (pPage->enmKind)
3488	{
3489	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3490	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3491	{
3492	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3493	void *pvGst;
3494	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3495	pgmPoolTrackDerefPT32Bit32Bit(pPool, pPage, (PX86PT)pvShw, (PCX86PT)pvGst);
3496	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3497	break;
3498	}
3499
3500	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3501	{
3502	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3503	void *pvGst;
3504	int rc = PGM_GCPHYS_2_PTR_EX(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3505	pgmPoolTrackDerefPTPae32Bit(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PT)pvGst);
3506	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3507	break;
3508	}
3509
3510	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3511	{
3512	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3513	void *pvGst;
3514	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3515	pgmPoolTrackDerefPTPaePae(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PTPAE)pvGst);
3516	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3517	break;
3518	}
3519
3520	case PGMPOOLKIND_32BIT_PT_FOR_PHYS: /* treat it like a 4 MB page */
3521	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3522	{
3523	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3524	pgmPoolTrackDerefPT32Bit4MB(pPool, pPage, (PX86PT)pvShw);
3525	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3526	break;
3527	}
3528
3529	case PGMPOOLKIND_PAE_PT_FOR_PHYS: /* treat it like a 2 MB page */
3530	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3531	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3532	{
3533	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3534	pgmPoolTrackDerefPTPaeBig(pPool, pPage, (PX86PTPAE)pvShw);
3535	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3536	break;
3537	}
3538
3539	#else /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3540	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3541	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3542	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3543	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3544	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3545	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3546	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3547	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3548	break;
3549	#endif /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3550
3551	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
3552	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3553	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3554	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3555	pgmPoolTrackDerefPDPae(pPool, pPage, (PX86PDPAE)pvShw);
3556	break;
3557
3558	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3559	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3560	pgmPoolTrackDerefPDPT64Bit(pPool, pPage, (PX86PDPT)pvShw);
3561	break;
3562
3563	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
3564	pgmPoolTrackDerefPML464Bit(pPool, pPage, (PX86PML4)pvShw);
3565	break;
3566
3567	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3568	pgmPoolTrackDerefPTEPT(pPool, pPage, (PEPTPT)pvShw);
3569	break;
3570
3571	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3572	pgmPoolTrackDerefPDEPT(pPool, pPage, (PEPTPD)pvShw);
3573	break;
3574
3575	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3576	pgmPoolTrackDerefPDPTEPT(pPool, pPage, (PEPTPDPT)pvShw);
3577	break;
3578
3579	default:
3580	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
3581	}
3582
3583	/* paranoia, clear the shadow page. Remove this laser (i.e. let Alloc and ClearAll do it). */
3584	STAM_PROFILE_START(&pPool->StatZeroPage, z);
3585	ASMMemZeroPage(pvShw);
3586	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
3587	pPage->fZeroed = true;
3588	}
3589
3590	#endif /* PGMPOOL_WITH_USER_TRACKING */
3591
3592	/**
3593	* Flushes all the special root pages as part of a pgmPoolFlushAllInt operation.
3594	*
3595	* @param pPool The pool.
3596	*/
3597	static void pgmPoolFlushAllSpecialRoots(PPGMPOOL pPool)
3598	{
3599	/*
3600	* These special pages are all mapped into the indexes 1..PGMPOOL_IDX_FIRST.
3601	*/
3602	Assert(NIL_PGMPOOL_IDX == 0);
3603	for (unsigned i = 1; i < PGMPOOL_IDX_FIRST; i++)
3604	{
3605	/*
3606	* Get the page address.
3607	*/
3608	PPGMPOOLPAGE pPage = &pPool->aPages[i];
3609	union
3610	{
3611	uint64_t *pau64;
3612	uint32_t *pau32;
3613	} u;
3614
3615	/*
3616	* Mark stuff not present.
3617	*/
3618	switch (pPage->enmKind)
3619	{
3620	case PGMPOOLKIND_ROOT_32BIT_PD:
3621	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
3622	for (unsigned iPage = 0; iPage < X86_PG_ENTRIES; iPage++)
3623	if ((u.pau32[iPage] & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == X86_PDE_P)
3624	u.pau32[iPage] = 0;
3625	break;
3626
3627	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3628	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
3629	for (unsigned iPage = 0; iPage < X86_PG_PAE_ENTRIES; iPage++)
3630	if ((u.pau64[iPage] & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == X86_PDE_P)
3631	u.pau64[iPage] = 0;
3632	break;
3633
3634	case PGMPOOLKIND_ROOT_PDPT:
3635	/* Not root of shadowed pages currently, ignore it. */
3636	break;
3637
3638	case PGMPOOLKIND_ROOT_NESTED:
3639	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
3640	ASMMemZero32(u.pau64, PAGE_SIZE);
3641	break;
3642	}
3643	}
3644
3645	/*
3646	* Paranoia (to be removed), flag a global CR3 sync.
3647	*/
3648	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
3649	}
3650
3651
3652	/**
3653	* Flushes the entire cache.
3654	*
3655	* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
3656	* and execute this CR3 flush.
3657	*
3658	* @param pPool The pool.
3659	*/
3660	static void pgmPoolFlushAllInt(PPGMPOOL pPool)
3661	{
3662	STAM_PROFILE_START(&pPool->StatFlushAllInt, a);
3663	LogFlow(("pgmPoolFlushAllInt:\n"));
3664
3665	/*
3666	* If there are no pages in the pool, there is nothing to do.
3667	*/
3668	if (pPool->cCurPages <= PGMPOOL_IDX_FIRST)
3669	{
3670	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
3671	return;
3672	}
3673
3674	/*
3675	* Nuke the free list and reinsert all pages into it.
3676	*/
3677	for (unsigned i = pPool->cCurPages - 1; i >= PGMPOOL_IDX_FIRST; i--)
3678	{
3679	PPGMPOOLPAGE pPage = &pPool->aPages[i];
3680
3681	#ifdef IN_RING3
3682	Assert(pPage->Core.Key == MMPage2Phys(pPool->pVMR3, pPage->pvPageR3));
3683	#endif
3684	#ifdef PGMPOOL_WITH_MONITORING
3685	if (pPage->fMonitored)
3686	pgmPoolMonitorFlush(pPool, pPage);
3687	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
3688	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
3689	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
3690	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
3691	pPage->cModifications = 0;
3692	#endif
3693	pPage->GCPhys = NIL_RTGCPHYS;
3694	pPage->enmKind = PGMPOOLKIND_FREE;
3695	Assert(pPage->idx == i);
3696	pPage->iNext = i + 1;
3697	pPage->fZeroed = false; /* This could probably be optimized, but better safe than sorry. */
3698	pPage->fSeenNonGlobal = false;
3699	pPage->fMonitored= false;
3700	pPage->fCached = false;
3701	pPage->fReusedFlushPending = false;
3702	pPage->fCR3Mix = false;
3703	#ifdef PGMPOOL_WITH_USER_TRACKING
3704	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
3705	#endif
3706	#ifdef PGMPOOL_WITH_CACHE
3707	pPage->iAgeNext = NIL_PGMPOOL_IDX;
3708	pPage->iAgePrev = NIL_PGMPOOL_IDX;
3709	#endif
3710	}
3711	pPool->aPages[pPool->cCurPages - 1].iNext = NIL_PGMPOOL_IDX;
3712	pPool->iFreeHead = PGMPOOL_IDX_FIRST;
3713	pPool->cUsedPages = 0;
3714
3715	#ifdef PGMPOOL_WITH_USER_TRACKING
3716	/*
3717	* Zap and reinitialize the user records.
3718	*/
3719	pPool->cPresent = 0;
3720	pPool->iUserFreeHead = 0;
3721	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
3722	const unsigned cMaxUsers = pPool->cMaxUsers;
3723	for (unsigned i = 0; i < cMaxUsers; i++)
3724	{
3725	paUsers[i].iNext = i + 1;
3726	paUsers[i].iUser = NIL_PGMPOOL_IDX;
3727	paUsers[i].iUserTable = 0xfffffffe;
3728	}
3729	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
3730	#endif
3731
3732	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3733	/*
3734	* Clear all the GCPhys links and rebuild the phys ext free list.
3735	*/
3736	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3737	pRam;
3738	pRam = pRam->CTX_SUFF(pNext))
3739	{
3740	unsigned iPage = pRam->cb >> PAGE_SHIFT;
3741	while (iPage-- > 0)
3742	pRam->aPages[iPage].HCPhys &= MM_RAM_FLAGS_NO_REFS_MASK; /** @todo PAGE FLAGS */
3743	}
3744
3745	pPool->iPhysExtFreeHead = 0;
3746	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3747	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
3748	for (unsigned i = 0; i < cMaxPhysExts; i++)
3749	{
3750	paPhysExts[i].iNext = i + 1;
3751	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
3752	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
3753	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
3754	}
3755	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
3756	#endif
3757
3758	#ifdef PGMPOOL_WITH_MONITORING
3759	/*
3760	* Just zap the modified list.
3761	*/
3762	pPool->cModifiedPages = 0;
3763	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
3764	#endif
3765
3766	#ifdef PGMPOOL_WITH_CACHE
3767	/*
3768	* Clear the GCPhys hash and the age list.
3769	*/
3770	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
3771	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
3772	pPool->iAgeHead = NIL_PGMPOOL_IDX;
3773	pPool->iAgeTail = NIL_PGMPOOL_IDX;
3774	#endif
3775
3776	/*
3777	* Flush all the special root pages.
3778	* Reinsert active pages into the hash and ensure monitoring chains are correct.
3779	*/
3780	pgmPoolFlushAllSpecialRoots(pPool);
3781	for (unsigned i = PGMPOOL_IDX_FIRST_SPECIAL; i < PGMPOOL_IDX_FIRST; i++)
3782	{
3783	PPGMPOOLPAGE pPage = &pPool->aPages[i];
3784	pPage->iNext = NIL_PGMPOOL_IDX;
3785	#ifdef PGMPOOL_WITH_MONITORING
3786	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
3787	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
3788	pPage->cModifications = 0;
3789	/* ASSUMES that we're not sharing with any of the other special pages (safe for now). */
3790	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
3791	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
3792	if (pPage->fMonitored)
3793	{
3794	PVM pVM = pPool->CTX_SUFF(pVM);
3795	int rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
3796	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
3797	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
3798	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
3799	pPool->pszAccessHandler);
3800	AssertFatalRCSuccess(rc);
3801	# ifdef PGMPOOL_WITH_CACHE
3802	pgmPoolHashInsert(pPool, pPage);
3803	# endif
3804	}
3805	#endif
3806	#ifdef PGMPOOL_WITH_USER_TRACKING
3807	Assert(pPage->iUserHead == NIL_PGMPOOL_USER_INDEX); /* for now */
3808	#endif
3809	#ifdef PGMPOOL_WITH_CACHE
3810	Assert(pPage->iAgeNext == NIL_PGMPOOL_IDX);
3811	Assert(pPage->iAgePrev == NIL_PGMPOOL_IDX);
3812	#endif
3813	}
3814
3815	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
3816	}
3817
3818
3819	/**
3820	* Flushes a pool page.
3821	*
3822	* This moves the page to the free list after removing all user references to it.
3823	* In GC this will cause a CR3 reload if the page is traced back to an active root page.
3824	*
3825	* @returns VBox status code.
3826	* @retval VINF_SUCCESS on success.
3827	* @retval VERR_PGM_POOL_CLEARED if the deregistration of the physical handler will cause a light weight pool flush.
3828	* @param pPool The pool.
3829	* @param HCPhys The HC physical address of the shadow page.
3830	*/
3831	int pgmPoolFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3832	{
3833	int rc = VINF_SUCCESS;
3834	STAM_PROFILE_START(&pPool->StatFlushPage, f);
3835	LogFlow(("pgmPoolFlushPage: pPage=%p:{.Key=%RHp, .idx=%d, .enmKind=%d, .GCPhys=%RGp}\n",
3836	pPage, pPage->Core.Key, pPage->idx, pPage->enmKind, pPage->GCPhys));
3837
3838	/*
3839	* Quietly reject any attempts at flushing any of the special root pages.
3840	*/
3841	if (pPage->idx < PGMPOOL_IDX_FIRST)
3842	{
3843	Log(("pgmPoolFlushPage: special root page, rejected. enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
3844	return VINF_SUCCESS;
3845	}
3846
3847	/*
3848	* Quietly reject any attempts at flushing the currently active shadow CR3 mapping
3849	*/
3850	if (PGMGetHyperCR3(pPool->CTX_SUFF(pVM)) == pPage->Core.Key)
3851	{
3852	AssertMsg(pPage->enmKind == PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4,
3853	("Can't free the shadow CR3! (%RHp vs %RHp kind=%d\n", PGMGetHyperCR3(pPool->CTX_SUFF(pVM)), pPage->Core.Key, pPage->enmKind));
3854	Log(("pgmPoolFlushPage: current active shadow CR3, rejected. enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
3855	return VINF_SUCCESS;
3856	}
3857
3858	/*
3859	* Mark the page as being in need of a ASMMemZeroPage().
3860	*/
3861	pPage->fZeroed = false;
3862
3863	#ifdef PGMPOOL_WITH_USER_TRACKING
3864	/*
3865	* Clear the page.
3866	*/
3867	pgmPoolTrackClearPageUsers(pPool, pPage);
3868	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
3869	pgmPoolTrackDeref(pPool, pPage);
3870	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
3871	#endif
3872
3873	#ifdef PGMPOOL_WITH_CACHE
3874	/*
3875	* Flush it from the cache.
3876	*/
3877	pgmPoolCacheFlushPage(pPool, pPage);
3878	#endif /* PGMPOOL_WITH_CACHE */
3879
3880	#ifdef PGMPOOL_WITH_MONITORING
3881	/*
3882	* Deregistering the monitoring.
3883	*/
3884	if (pPage->fMonitored)
3885	rc = pgmPoolMonitorFlush(pPool, pPage);
3886	#endif
3887
3888	/*
3889	* Free the page.
3890	*/
3891	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
3892	pPage->iNext = pPool->iFreeHead;
3893	pPool->iFreeHead = pPage->idx;
3894	pPage->enmKind = PGMPOOLKIND_FREE;
3895	pPage->GCPhys = NIL_RTGCPHYS;
3896	pPage->fReusedFlushPending = false;
3897
3898	pPool->cUsedPages--;
3899	STAM_PROFILE_STOP(&pPool->StatFlushPage, f);
3900	return rc;
3901	}
3902
3903
3904	/**
3905	* Frees a usage of a pool page.
3906	*
3907	* The caller is responsible to updating the user table so that it no longer
3908	* references the shadow page.
3909	*
3910	* @param pPool The pool.
3911	* @param HCPhys The HC physical address of the shadow page.
3912	* @param iUser The shadow page pool index of the user table.
3913	* @param iUserTable The index into the user table (shadowed).
3914	*/
3915	void pgmPoolFreeByPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
3916	{
3917	STAM_PROFILE_START(&pPool->StatFree, a);
3918	LogFlow(("pgmPoolFreeByPage: pPage=%p:{.Key=%RHp, .idx=%d, enmKind=%d} iUser=%#x iUserTable=%#x\n",
3919	pPage, pPage->Core.Key, pPage->idx, pPage->enmKind, iUser, iUserTable));
3920	Assert(pPage->idx >= PGMPOOL_IDX_FIRST);
3921	#ifdef PGMPOOL_WITH_USER_TRACKING
3922	pgmPoolTrackFreeUser(pPool, pPage, iUser, iUserTable);
3923	#endif
3924	#ifdef PGMPOOL_WITH_CACHE
3925	if (!pPage->fCached)
3926	#endif
3927	pgmPoolFlushPage(pPool, pPage); /* ASSUMES that VERR_PGM_POOL_CLEARED can be ignored here. */
3928	STAM_PROFILE_STOP(&pPool->StatFree, a);
3929	}
3930
3931
3932	/**
3933	* Makes one or more free page free.
3934	*
3935	* @returns VBox status code.
3936	* @retval VINF_SUCCESS on success.
3937	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
3938	*
3939	* @param pPool The pool.
3940	* @param iUser The user of the page.
3941	*/
3942	static int pgmPoolMakeMoreFreePages(PPGMPOOL pPool, uint16_t iUser)
3943	{
3944	LogFlow(("pgmPoolMakeMoreFreePages: iUser=%#x\n", iUser));
3945
3946	/*
3947	* If the pool isn't full grown yet, expand it.
3948	*/
3949	if (pPool->cCurPages < pPool->cMaxPages)
3950	{
3951	STAM_PROFILE_ADV_SUSPEND(&pPool->StatAlloc, a);
3952	#ifdef IN_RING3
3953	int rc = PGMR3PoolGrow(pPool->pVMR3);
3954	#else
3955	int rc = CTXALLMID(VMM, CallHost)(pPool->CTX_SUFF(pVM), VMMCALLHOST_PGM_POOL_GROW, 0);
3956	#endif
3957	if (RT_FAILURE(rc))
3958	return rc;
3959	STAM_PROFILE_ADV_RESUME(&pPool->StatAlloc, a);
3960	if (pPool->iFreeHead != NIL_PGMPOOL_IDX)
3961	return VINF_SUCCESS;
3962	}
3963
3964	#ifdef PGMPOOL_WITH_CACHE
3965	/*
3966	* Free one cached page.
3967	*/
3968	return pgmPoolCacheFreeOne(pPool, iUser);
3969	#else
3970	/*
3971	* Flush the pool.
3972	* If we have tracking enabled, it should be possible to come up with
3973	* a cheap replacement strategy...
3974	*/
3975	/* @todo incompatible with long mode paging (cr3 root will be flushed) */
3976	Assert(!CPUMIsGuestInLongMode(pVM));
3977	pgmPoolFlushAllInt(pPool);
3978	return VERR_PGM_POOL_FLUSHED;
3979	#endif
3980	}
3981
3982
3983	/**
3984	* Allocates a page from the pool.
3985	*
3986	* This page may actually be a cached page and not in need of any processing
3987	* on the callers part.
3988	*
3989	* @returns VBox status code.
3990	* @retval VINF_SUCCESS if a NEW page was allocated.
3991	* @retval VINF_PGM_CACHED_PAGE if a CACHED page was returned.
3992	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
3993	* @param pVM The VM handle.
3994	* @param GCPhys The GC physical address of the page we're gonna shadow.
3995	* For 4MB and 2MB PD entries, it's the first address the
3996	* shadow PT is covering.
3997	* @param enmKind The kind of mapping.
3998	* @param iUser The shadow page pool index of the user table.
3999	* @param iUserTable The index into the user table (shadowed).
4000	* @param ppPage Where to store the pointer to the page. NULL is stored here on failure.
4001	*/
4002	int pgmPoolAlloc(PVM pVM, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
4003	{
4004	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4005	STAM_PROFILE_ADV_START(&pPool->StatAlloc, a);
4006	LogFlow(("pgmPoolAlloc: GCPhys=%RGp enmKind=%d iUser=%#x iUserTable=%#x\n", GCPhys, enmKind, iUser, iUserTable));
4007	*ppPage = NULL;
4008
4009	#ifdef PGMPOOL_WITH_CACHE
4010	if (pPool->fCacheEnabled)
4011	{
4012	int rc2 = pgmPoolCacheAlloc(pPool, GCPhys, enmKind, iUser, iUserTable, ppPage);
4013	if (RT_SUCCESS(rc2))
4014	{
4015	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4016	LogFlow(("pgmPoolAlloc: cached returns %Rrc ppPage=%p:{.Key=%RHp, .idx=%d}\n", rc2, ppPage, (ppPage)->Core.Key, (ppPage)->idx));
4017	return rc2;
4018	}
4019	}
4020	#endif
4021
4022	/*
4023	* Allocate a new one.
4024	*/
4025	int rc = VINF_SUCCESS;
4026	uint16_t iNew = pPool->iFreeHead;
4027	if (iNew == NIL_PGMPOOL_IDX)
4028	{
4029	rc = pgmPoolMakeMoreFreePages(pPool, iUser);
4030	if (RT_FAILURE(rc))
4031	{
4032	if (rc != VERR_PGM_POOL_CLEARED)
4033	{
4034	Log(("pgmPoolAlloc: returns %Rrc (Free)\n", rc));
4035	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4036	return rc;
4037	}
4038	Log(("pgmPoolMakeMoreFreePages failed with %Rrc -> return VERR_PGM_POOL_FLUSHED\n", rc));
4039	rc = VERR_PGM_POOL_FLUSHED;
4040	}
4041	iNew = pPool->iFreeHead;
4042	AssertReleaseReturn(iNew != NIL_PGMPOOL_IDX, VERR_INTERNAL_ERROR);
4043	}
4044
4045	/* unlink the free head */
4046	PPGMPOOLPAGE pPage = &pPool->aPages[iNew];
4047	pPool->iFreeHead = pPage->iNext;
4048	pPage->iNext = NIL_PGMPOOL_IDX;
4049
4050	/*
4051	* Initialize it.
4052	*/
4053	pPool->cUsedPages++; /* physical handler registration / pgmPoolTrackFlushGCPhysPTsSlow requirement. */
4054	pPage->enmKind = enmKind;
4055	pPage->GCPhys = GCPhys;
4056	pPage->fSeenNonGlobal = false; /* Set this to 'true' to disable this feature. */
4057	pPage->fMonitored = false;
4058	pPage->fCached = false;
4059	pPage->fReusedFlushPending = false;
4060	pPage->fCR3Mix = false;
4061	#ifdef PGMPOOL_WITH_MONITORING
4062	pPage->cModifications = 0;
4063	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4064	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4065	#endif
4066	#ifdef PGMPOOL_WITH_USER_TRACKING
4067	pPage->cPresent = 0;
4068	pPage->iFirstPresent = ~0;
4069
4070	/*
4071	* Insert into the tracking and cache. If this fails, free the page.
4072	*/
4073	int rc3 = pgmPoolTrackInsert(pPool, pPage, GCPhys, iUser, iUserTable);
4074	if (RT_FAILURE(rc3))
4075	{
4076	if (rc3 != VERR_PGM_POOL_CLEARED)
4077	{
4078	pPool->cUsedPages--;
4079	pPage->enmKind = PGMPOOLKIND_FREE;
4080	pPage->GCPhys = NIL_RTGCPHYS;
4081	pPage->iNext = pPool->iFreeHead;
4082	pPool->iFreeHead = pPage->idx;
4083	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4084	Log(("pgmPoolAlloc: returns %Rrc (Insert)\n", rc3));
4085	return rc3;
4086	}
4087	Log(("pgmPoolTrackInsert failed with %Rrc -> return VERR_PGM_POOL_FLUSHED\n", rc3));
4088	rc = VERR_PGM_POOL_FLUSHED;
4089	}
4090	#endif /* PGMPOOL_WITH_USER_TRACKING */
4091
4092	/*
4093	* Commit the allocation, clear the page and return.
4094	*/
4095	#ifdef VBOX_WITH_STATISTICS
4096	if (pPool->cUsedPages > pPool->cUsedPagesHigh)
4097	pPool->cUsedPagesHigh = pPool->cUsedPages;
4098	#endif
4099
4100	if (!pPage->fZeroed)
4101	{
4102	STAM_PROFILE_START(&pPool->StatZeroPage, z);
4103	void *pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
4104	ASMMemZeroPage(pv);
4105	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
4106	}
4107
4108	*ppPage = pPage;
4109	LogFlow(("pgmPoolAlloc: returns %Rrc *ppPage=%p:{.Key=%RHp, .idx=%d, .fCached=%RTbool, .fMonitored=%RTbool}\n",
4110	rc, pPage, pPage->Core.Key, pPage->idx, pPage->fCached, pPage->fMonitored));
4111	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4112	return rc;
4113	}
4114
4115
4116	/**
4117	* Frees a usage of a pool page.
4118	*
4119	* @param pVM The VM handle.
4120	* @param HCPhys The HC physical address of the shadow page.
4121	* @param iUser The shadow page pool index of the user table.
4122	* @param iUserTable The index into the user table (shadowed).
4123	*/
4124	void pgmPoolFree(PVM pVM, RTHCPHYS HCPhys, uint16_t iUser, uint32_t iUserTable)
4125	{
4126	LogFlow(("pgmPoolFree: HCPhys=%RHp iUser=%#x iUserTable=%#x\n", HCPhys, iUser, iUserTable));
4127	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4128	pgmPoolFreeByPage(pPool, pgmPoolGetPage(pPool, HCPhys), iUser, iUserTable);
4129	}
4130
4131
4132	/**
4133	* Gets a in-use page in the pool by it's physical address.
4134	*
4135	* @returns Pointer to the page.
4136	* @param pVM The VM handle.
4137	* @param HCPhys The HC physical address of the shadow page.
4138	* @remark This function will NEVER return NULL. It will assert if HCPhys is invalid.
4139	*/
4140	PPGMPOOLPAGE pgmPoolGetPageByHCPhys(PVM pVM, RTHCPHYS HCPhys)
4141	{
4142	/** @todo profile this! */
4143	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4144	PPGMPOOLPAGE pPage = pgmPoolGetPage(pPool, HCPhys);
4145	Log3(("pgmPoolGetPageByHCPhys: HCPhys=%RHp -> %p:{.idx=%d .GCPhys=%RGp .enmKind=%d}\n",
4146	HCPhys, pPage, pPage->idx, pPage->GCPhys, pPage->enmKind));
4147	return pPage;
4148	}
4149
4150
4151	/**
4152	* Flushes the entire cache.
4153	*
4154	* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
4155	* and execute this CR3 flush.
4156	*
4157	* @param pPool The pool.
4158	*/
4159	void pgmPoolFlushAll(PVM pVM)
4160	{
4161	LogFlow(("pgmPoolFlushAll:\n"));
4162	pgmPoolFlushAllInt(pVM->pgm.s.CTX_SUFF(pPool));
4163	}
4164

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllPool.cpp@ 14809

Download in other formats: