PGMAllPool.cpp@ 14082

Last change on this file since 14082 was 14038, checked in by vboxsync, 16 years ago
#1865: PGM - and another one.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 147.8 KB

Line
1	/* $Id: PGMAllPool.cpp 14038 2008-11-10 18:23:15Z 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.pGC32BitPD;
115	case PGMPOOL_IDX_PAE_PD:
116	case PGMPOOL_IDX_PAE_PD_0:
117	return pVM->pgm.s.apGCPaePDs[0];
118	case PGMPOOL_IDX_PAE_PD_1:
119	return pVM->pgm.s.apGCPaePDs[1];
120	case PGMPOOL_IDX_PAE_PD_2:
121	return pVM->pgm.s.apGCPaePDs[2];
122	case PGMPOOL_IDX_PAE_PD_3:
123	return pVM->pgm.s.apGCPaePDs[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.HCPhys32BitPD;
137	break;
138	case PGMPOOL_IDX_PAE_PD:
139	case PGMPOOL_IDX_PAE_PD_0:
140	HCPhys = pVM->pgm.s.aHCPhysPaePDs[0];
141	break;
142	case PGMPOOL_IDX_PAE_PD_1:
143	HCPhys = pVM->pgm.s.aHCPhysPaePDs[1];
144	break;
145	case PGMPOOL_IDX_PAE_PD_2:
146	HCPhys = pVM->pgm.s.aHCPhysPaePDs[2];
147	break;
148	case PGMPOOL_IDX_PAE_PD_3:
149	HCPhys = pVM->pgm.s.aHCPhysPaePDs[3];
150	break;
151	case PGMPOOL_IDX_PDPT:
152	HCPhys = pVM->pgm.s.HCPhysPaePDPT;
153	break;
154	default:
155	AssertReleaseMsgFailed(("Invalid index %d\n", pPage->idx));
156	return NULL;
157	}
158	void *pv;
159	int rc = PGMDynMapHCPage(pVM, pPage->Core.Key, &pv);
160	AssertReleaseRC(rc);
161	return pv;
162	# endif /* VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0 */
163	}
164	#endif /* IN_RC \|\| VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0 */
165
166
167	#ifdef PGMPOOL_WITH_MONITORING
168	/**
169	* Determin the size of a write instruction.
170	* @returns number of bytes written.
171	* @param pDis The disassembler state.
172	*/
173	static unsigned pgmPoolDisasWriteSize(PDISCPUSTATE pDis)
174	{
175	/*
176	* This is very crude and possibly wrong for some opcodes,
177	* but since it's not really supposed to be called we can
178	* probably live with that.
179	*/
180	return DISGetParamSize(pDis, &pDis->param1);
181	}
182
183
184	/**
185	* Flushes a chain of pages sharing the same access monitor.
186	*
187	* @returns VBox status code suitable for scheduling.
188	* @param pPool The pool.
189	* @param pPage A page in the chain.
190	*/
191	int pgmPoolMonitorChainFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
192	{
193	LogFlow(("pgmPoolMonitorChainFlush: Flush page %RGp type=%d\n", pPage->GCPhys, pPage->enmKind));
194
195	/*
196	* Find the list head.
197	*/
198	uint16_t idx = pPage->idx;
199	if (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
200	{
201	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
202	{
203	idx = pPage->iMonitoredPrev;
204	Assert(idx != pPage->idx);
205	pPage = &pPool->aPages[idx];
206	}
207	}
208
209	/*
210	* Iterate the list flushing each shadow page.
211	*/
212	int rc = VINF_SUCCESS;
213	for (;;)
214	{
215	idx = pPage->iMonitoredNext;
216	Assert(idx != pPage->idx);
217	if (pPage->idx >= PGMPOOL_IDX_FIRST)
218	{
219	int rc2 = pgmPoolFlushPage(pPool, pPage);
220	if (rc2 == VERR_PGM_POOL_CLEARED && rc == VINF_SUCCESS)
221	rc = VINF_PGM_SYNC_CR3;
222	}
223	/* next */
224	if (idx == NIL_PGMPOOL_IDX)
225	break;
226	pPage = &pPool->aPages[idx];
227	}
228	return rc;
229	}
230
231
232	/**
233	* Wrapper for getting the current context pointer to the entry being modified.
234	*
235	* @returns Pointer to the current context mapping of the entry.
236	* @param pPool The pool.
237	* @param pvFault The fault virtual address.
238	* @param GCPhysFault The fault physical address.
239	* @param cbEntry The entry size.
240	*/
241	#ifdef IN_RING3
242	DECLINLINE(const void *) pgmPoolMonitorGCPtr2CCPtr(PPGMPOOL pPool, RTHCPTR pvFault, RTGCPHYS GCPhysFault, const unsigned cbEntry)
243	#else
244	DECLINLINE(const void *) pgmPoolMonitorGCPtr2CCPtr(PPGMPOOL pPool, RTGCPTR pvFault, RTGCPHYS GCPhysFault, const unsigned cbEntry)
245	#endif
246	{
247	#ifdef IN_RC
248	return (const void *)((RTGCUINTPTR)pvFault & ~(RTGCUINTPTR)(cbEntry - 1));
249
250	#elif defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0)
251	void *pvRet;
252	int rc = PGMDynMapGCPageOff(pPool->pVMR0, GCPhysFault & ~(RTGCPHYS)(cbEntry - 1), &pvRet);
253	AssertFatalRCSuccess(rc);
254	return pvRet;
255
256	#elif defined(IN_RING0)
257	void *pvRet;
258	int rc = pgmRamGCPhys2HCPtr(&pPool->pVMR0->pgm.s, GCPhysFault & ~(RTGCPHYS)(cbEntry - 1), &pvRet);
259	AssertFatalRCSuccess(rc);
260	return pvRet;
261
262	#elif defined(IN_RING3)
263	return (RTHCPTR)((uintptr_t)pvFault & ~(RTHCUINTPTR)(cbEntry - 1));
264	#else
265	# error "huh?"
266	#endif
267	}
268
269
270	/**
271	* Process shadow entries before they are changed by the guest.
272	*
273	* For PT entries we will clear them. For PD entries, we'll simply check
274	* for mapping conflicts and set the SyncCR3 FF if found.
275	*
276	* @param pPool The pool.
277	* @param pPage The head page.
278	* @param GCPhysFault The guest physical fault address.
279	* @param uAddress In R0 and GC this is the guest context fault address (flat).
280	* In R3 this is the host context 'fault' address.
281	* @param pCpu The disassembler state for figuring out the write size.
282	* This need not be specified if the caller knows we won't do cross entry accesses.
283	*/
284	#ifdef IN_RING3
285	void pgmPoolMonitorChainChanging(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhysFault, RTHCPTR pvAddress, PDISCPUSTATE pCpu)
286	#else
287	void pgmPoolMonitorChainChanging(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhysFault, RTGCPTR pvAddress, PDISCPUSTATE pCpu)
288	#endif
289	{
290	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
291	const unsigned off = GCPhysFault & PAGE_OFFSET_MASK;
292	const unsigned cbWrite = (pCpu) ? pgmPoolDisasWriteSize(pCpu) : 0;
293
294	LogFlow(("pgmPoolMonitorChainChanging: %RGv phys=%RGp kind=%d cbWrite=%d\n", pvAddress, GCPhysFault, pPage->enmKind, cbWrite));
295
296	for (;;)
297	{
298	union
299	{
300	void *pv;
301	PX86PT pPT;
302	PX86PTPAE pPTPae;
303	PX86PD pPD;
304	PX86PDPAE pPDPae;
305	PX86PDPT pPDPT;
306	PX86PML4 pPML4;
307	} uShw;
308	uShw.pv = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
309
310	switch (pPage->enmKind)
311	{
312	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
313	{
314	const unsigned iShw = off / sizeof(X86PTE);
315	if (uShw.pPT->a[iShw].n.u1Present)
316	{
317	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
318	PCX86PTE pGstPte = (PCX86PTE)pgmPoolMonitorGCPtr2CCPtr(pPool, pvAddress, GCPhysFault, sizeof(*pGstPte));
319	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));
320	pgmPoolTracDerefGCPhysHint(pPool, pPage,
321	uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK,
322	pGstPte->u & X86_PTE_PG_MASK);
323	# endif
324	uShw.pPT->a[iShw].u = 0;
325	}
326	break;
327	}
328
329	/* page/2 sized */
330	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
331	if (!((off ^ pPage->GCPhys) & (PAGE_SIZE / 2)))
332	{
333	const unsigned iShw = (off / sizeof(X86PTE)) & (X86_PG_PAE_ENTRIES - 1);
334	if (uShw.pPTPae->a[iShw].n.u1Present)
335	{
336	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
337	PCX86PTE pGstPte = (PCX86PTE)pgmPoolMonitorGCPtr2CCPtr(pPool, pvAddress, GCPhysFault, sizeof(*pGstPte));
338	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));
339	pgmPoolTracDerefGCPhysHint(pPool, pPage,
340	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
341	pGstPte->u & X86_PTE_PG_MASK);
342	# endif
343	uShw.pPTPae->a[iShw].u = 0;
344	}
345	}
346	break;
347
348	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
349	{
350	const unsigned iShw = off / sizeof(X86PTEPAE);
351	if (uShw.pPTPae->a[iShw].n.u1Present)
352	{
353	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
354	PCX86PTEPAE pGstPte = (PCX86PTEPAE)pgmPoolMonitorGCPtr2CCPtr(pPool, pvAddress, GCPhysFault, sizeof(*pGstPte));
355	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));
356	pgmPoolTracDerefGCPhysHint(pPool, pPage,
357	uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
358	pGstPte->u & X86_PTE_PAE_PG_MASK);
359	# endif
360	uShw.pPTPae->a[iShw].u = 0;
361	}
362
363	/* paranoia / a bit assumptive. */
364	if ( pCpu
365	&& (off & 7)
366	&& (off & 7) + cbWrite > sizeof(X86PTEPAE))
367	{
368	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTEPAE);
369	AssertReturnVoid(iShw2 < RT_ELEMENTS(uShw.pPTPae->a));
370
371	if (uShw.pPTPae->a[iShw2].n.u1Present)
372	{
373	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
374	PCX86PTEPAE pGstPte = (PCX86PTEPAE)pgmPoolMonitorGCPtr2CCPtr(pPool, pvAddress, GCPhysFault, sizeof(*pGstPte));
375	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));
376	pgmPoolTracDerefGCPhysHint(pPool, pPage,
377	uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK,
378	pGstPte->u & X86_PTE_PAE_PG_MASK);
379	# endif
380	uShw.pPTPae->a[iShw2].u = 0;
381	}
382	}
383
384	break;
385	}
386
387	case PGMPOOLKIND_ROOT_32BIT_PD:
388	{
389	const unsigned iShw = off / sizeof(X86PTE); // ASSUMING 32-bit guest paging!
390	if (uShw.pPD->a[iShw].u & PGM_PDFLAGS_MAPPING)
391	{
392	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
393	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
394	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
395	}
396	/* paranoia / a bit assumptive. */
397	else if ( pCpu
398	&& (off & 3)
399	&& (off & 3) + cbWrite > sizeof(X86PTE))
400	{
401	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTE);
402	if ( iShw2 != iShw
403	&& iShw2 < RT_ELEMENTS(uShw.pPD->a)
404	&& uShw.pPD->a[iShw2].u & PGM_PDFLAGS_MAPPING)
405	{
406	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
407	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
408	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
409	}
410	}
411	#if 0 /* useful when running PGMAssertCR3(), a bit too troublesome for general use (TLBs). */
412	if ( uShw.pPD->a[iShw].n.u1Present
413	&& !VM_FF_ISSET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3))
414	{
415	LogFlow(("pgmPoolMonitorChainChanging: iShw=%#x: %RX32 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
416	# ifdef IN_RC /* TLB load - we're pushing things a bit... */
417	ASMProbeReadByte(pvAddress);
418	# endif
419	pgmPoolFree(pPool->CTX_SUFF(pVM), uShw.pPD->a[iShw].u & X86_PDE_PG_MASK, pPage->idx, iShw);
420	uShw.pPD->a[iShw].u = 0;
421	}
422	#endif
423	break;
424	}
425
426	case PGMPOOLKIND_ROOT_PAE_PD:
427	{
428	unsigned iShw = (off / sizeof(X86PTE)) * 2; // ASSUMING 32-bit guest paging!
429	for (unsigned i = 0; i < 2; i++, iShw++)
430	{
431	if ((uShw.pPDPae->a[iShw].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
432	{
433	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
434	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
435	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
436	}
437	/* paranoia / a bit assumptive. */
438	else if ( pCpu
439	&& (off & 3)
440	&& (off & 3) + cbWrite > 4)
441	{
442	const unsigned iShw2 = iShw + 2;
443	if ( iShw2 < RT_ELEMENTS(uShw.pPDPae->a)
444	&& (uShw.pPDPae->a[iShw2].u & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == (PGM_PDFLAGS_MAPPING \| X86_PDE_P))
445	{
446	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
447	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
448	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
449	}
450	}
451	#if 0 /* useful when running PGMAssertCR3(), a bit too troublesome for general use (TLBs). */
452	if ( uShw.pPDPae->a[iShw].n.u1Present
453	&& !VM_FF_ISSET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3))
454	{
455	LogFlow(("pgmPoolMonitorChainChanging: iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
456	# ifdef IN_RC /* TLB load - we're pushing things a bit... */
457	ASMProbeReadByte(pvAddress);
458	# endif
459	pgmPoolFree(pPool->CTX_SUFF(pVM), uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK, pPage->idx, iShw);
460	uShw.pPDPae->a[iShw].u = 0;
461	}
462	#endif
463	}
464	break;
465	}
466
467	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
468	{
469	const unsigned iShw = off / sizeof(X86PDEPAE);
470	if (uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING)
471	{
472	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
473	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
474	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
475	}
476	#ifdef PGMPOOL_INVALIDATE_UPPER_SHADOW_TABLE_ENTRIES
477	/*
478	* Causes trouble when the guest uses a PDE to refer to the whole page table level
479	* structure. (Invalidate here; faults later on when it tries to change the page
480	* table entries -> recheck; probably only applies to the RC case.)
481	*/
482	else
483	{
484	if (uShw.pPDPae->a[iShw].n.u1Present)
485	{
486	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
487	pgmPoolFree(pPool->CTX_SUFF(pVM),
488	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
489	/* Note: hardcoded PAE implementation dependency */
490	(pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD) ? PGMPOOL_IDX_PAE_PD : pPage->idx,
491	(pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD) ? iShw + (pPage->idx - PGMPOOL_IDX_PAE_PD_0) * X86_PG_PAE_ENTRIES : iShw);
492	uShw.pPDPae->a[iShw].u = 0;
493	}
494	}
495	#endif
496	/* paranoia / a bit assumptive. */
497	if ( pCpu
498	&& (off & 7)
499	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
500	{
501	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
502	AssertReturnVoid(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
503
504	if ( iShw2 != iShw
505	&& uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING)
506	{
507	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
508	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
509	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
510	}
511	#ifdef PGMPOOL_INVALIDATE_UPPER_SHADOW_TABLE_ENTRIES
512	else if (uShw.pPDPae->a[iShw2].n.u1Present)
513	{
514	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
515	pgmPoolFree(pPool->CTX_SUFF(pVM),
516	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
517	/* Note: hardcoded PAE implementation dependency */
518	(pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD) ? PGMPOOL_IDX_PAE_PD : pPage->idx,
519	(pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD) ? iShw2 + (pPage->idx - PGMPOOL_IDX_PAE_PD_0) * X86_PG_PAE_ENTRIES : iShw2);
520	uShw.pPDPae->a[iShw2].u = 0;
521	}
522	#endif
523	}
524	break;
525	}
526
527	case PGMPOOLKIND_ROOT_PDPT:
528	{
529	/*
530	* Hopefully this doesn't happen very often:
531	* - touching unused parts of the page
532	* - messing with the bits of pd pointers without changing the physical address
533	*/
534	const unsigned iShw = off / sizeof(X86PDPE);
535	if (iShw < X86_PG_PAE_PDPE_ENTRIES) /* don't use RT_ELEMENTS(uShw.pPDPT->a), because that's for long mode only */
536	{
537	if (uShw.pPDPT->a[iShw].u & PGM_PLXFLAGS_MAPPING)
538	{
539	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
540	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
541	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
542	}
543	/* paranoia / a bit assumptive. */
544	else if ( pCpu
545	&& (off & 7)
546	&& (off & 7) + cbWrite > sizeof(X86PDPE))
547	{
548	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDPE);
549	if ( iShw2 != iShw
550	&& iShw2 < X86_PG_PAE_PDPE_ENTRIES
551	&& uShw.pPDPT->a[iShw2].u & PGM_PLXFLAGS_MAPPING)
552	{
553	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
554	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
555	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
556	}
557	}
558	}
559	break;
560	}
561
562	#ifndef IN_RC
563	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
564	{
565	Assert(pPage->enmKind == PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD);
566
567	const unsigned iShw = off / sizeof(X86PDEPAE);
568	if (uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING)
569	{
570	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
571	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
572	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
573	}
574	else
575	{
576	if (uShw.pPDPae->a[iShw].n.u1Present)
577	{
578	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
579	pgmPoolFree(pPool->CTX_SUFF(pVM),
580	uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
581	pPage->idx,
582	iShw);
583	uShw.pPDPae->a[iShw].u = 0;
584	}
585	}
586	/* paranoia / a bit assumptive. */
587	if ( pCpu
588	&& (off & 7)
589	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
590	{
591	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
592	AssertReturnVoid(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
593
594	if ( iShw2 != iShw
595	&& uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING)
596	{
597	Assert(pgmMapAreMappingsEnabled(&pPool->CTX_SUFF(pVM)->pgm.s));
598	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
599	LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
600	}
601	else
602	if (uShw.pPDPae->a[iShw2].n.u1Present)
603	{
604	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
605	pgmPoolFree(pPool->CTX_SUFF(pVM),
606	uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
607	pPage->idx,
608	iShw2);
609	uShw.pPDPae->a[iShw2].u = 0;
610	}
611	}
612	break;
613	}
614
615	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
616	{
617	/*
618	* Hopefully this doesn't happen very often:
619	* - messing with the bits of pd pointers without changing the physical address
620	*/
621	if (!VM_FF_ISSET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3))
622	{
623	const unsigned iShw = off / sizeof(X86PDPE);
624	if (uShw.pPDPT->a[iShw].n.u1Present)
625	{
626	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
627	pgmPoolFree(pPool->CTX_SUFF(pVM), uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK, pPage->idx, iShw);
628	uShw.pPDPT->a[iShw].u = 0;
629	}
630	/* paranoia / a bit assumptive. */
631	if ( pCpu
632	&& (off & 7)
633	&& (off & 7) + cbWrite > sizeof(X86PDPE))
634	{
635	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDPE);
636	if (uShw.pPDPT->a[iShw2].n.u1Present)
637	{
638	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
639	pgmPoolFree(pPool->CTX_SUFF(pVM), uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK, pPage->idx, iShw2);
640	uShw.pPDPT->a[iShw2].u = 0;
641	}
642	}
643	}
644	break;
645	}
646
647	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
648	{
649	/*
650	* Hopefully this doesn't happen very often:
651	* - messing with the bits of pd pointers without changing the physical address
652	*/
653	if (!VM_FF_ISSET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3))
654	{
655	const unsigned iShw = off / sizeof(X86PDPE);
656	if (uShw.pPML4->a[iShw].n.u1Present)
657	{
658	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPML4->a[iShw].u));
659	pgmPoolFree(pPool->CTX_SUFF(pVM), uShw.pPML4->a[iShw].u & X86_PML4E_PG_MASK, pPage->idx, iShw);
660	uShw.pPML4->a[iShw].u = 0;
661	}
662	/* paranoia / a bit assumptive. */
663	if ( pCpu
664	&& (off & 7)
665	&& (off & 7) + cbWrite > sizeof(X86PDPE))
666	{
667	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PML4E);
668	if (uShw.pPML4->a[iShw2].n.u1Present)
669	{
670	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPML4->a[iShw2].u));
671	pgmPoolFree(pPool->CTX_SUFF(pVM), uShw.pPML4->a[iShw2].u & X86_PML4E_PG_MASK, pPage->idx, iShw2);
672	uShw.pPML4->a[iShw2].u = 0;
673	}
674	}
675	}
676	break;
677	}
678	#endif /* IN_RING0 */
679
680	default:
681	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
682	}
683
684	/* next */
685	if (pPage->iMonitoredNext == NIL_PGMPOOL_IDX)
686	return;
687	pPage = &pPool->aPages[pPage->iMonitoredNext];
688	}
689	}
690
691
692	# ifndef IN_RING3
693	/**
694	* Checks if a access could be a fork operation in progress.
695	*
696	* Meaning, that the guest is setuping up the parent process for Copy-On-Write.
697	*
698	* @returns true if it's likly that we're forking, otherwise false.
699	* @param pPool The pool.
700	* @param pCpu The disassembled instruction.
701	* @param offFault The access offset.
702	*/
703	DECLINLINE(bool) pgmPoolMonitorIsForking(PPGMPOOL pPool, PDISCPUSTATE pCpu, unsigned offFault)
704	{
705	/*
706	* i386 linux is using btr to clear X86_PTE_RW.
707	* The functions involved are (2.6.16 source inspection):
708	* clear_bit
709	* ptep_set_wrprotect
710	* copy_one_pte
711	* copy_pte_range
712	* copy_pmd_range
713	* copy_pud_range
714	* copy_page_range
715	* dup_mmap
716	* dup_mm
717	* copy_mm
718	* copy_process
719	* do_fork
720	*/
721	if ( pCpu->pCurInstr->opcode == OP_BTR
722	&& !(offFault & 4)
723	/** @todo Validate that the bit index is X86_PTE_RW. */
724	)
725	{
726	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,Fork));
727	return true;
728	}
729	return false;
730	}
731
732
733	/**
734	* Determine whether the page is likely to have been reused.
735	*
736	* @returns true if we consider the page as being reused for a different purpose.
737	* @returns false if we consider it to still be a paging page.
738	* @param pVM VM Handle.
739	* @param pPage The page in question.
740	* @param pRegFrame Trap register frame.
741	* @param pCpu The disassembly info for the faulting instruction.
742	* @param pvFault The fault address.
743	*
744	* @remark The REP prefix check is left to the caller because of STOSD/W.
745	*/
746	DECLINLINE(bool) pgmPoolMonitorIsReused(PVM pVM, PPGMPOOLPAGE pPage, PCPUMCTXCORE pRegFrame, PDISCPUSTATE pCpu, RTGCPTR pvFault)
747	{
748	#ifndef IN_RC
749	/** @todo could make this general, faulting close to rsp should be safe reuse heuristic. */
750	if ( HWACCMHasPendingIrq(pVM)
751	&& (pRegFrame->rsp - pvFault) < 32)
752	{
753	/* Fault caused by stack writes while trying to inject an interrupt event. */
754	Log(("pgmPoolMonitorIsReused: reused %RGv for interrupt stack (rsp=%RGv).\n", pvFault, pRegFrame->rsp));
755	return true;
756	}
757	#else
758	NOREF(pVM); NOREF(pvFault);
759	#endif
760
761	switch (pCpu->pCurInstr->opcode)
762	{
763	/* call implies the actual push of the return address faulted */
764	case OP_CALL:
765	Log4(("pgmPoolMonitorIsReused: CALL\n"));
766	return true;
767	case OP_PUSH:
768	Log4(("pgmPoolMonitorIsReused: PUSH\n"));
769	return true;
770	case OP_PUSHF:
771	Log4(("pgmPoolMonitorIsReused: PUSHF\n"));
772	return true;
773	case OP_PUSHA:
774	Log4(("pgmPoolMonitorIsReused: PUSHA\n"));
775	return true;
776	case OP_FXSAVE:
777	Log4(("pgmPoolMonitorIsReused: FXSAVE\n"));
778	return true;
779	case OP_MOVNTI: /* solaris - block_zero_no_xmm */
780	Log4(("pgmPoolMonitorIsReused: MOVNTI\n"));
781	return true;
782	case OP_MOVNTDQ: /* solaris - hwblkclr & hwblkpagecopy */
783	Log4(("pgmPoolMonitorIsReused: MOVNTDQ\n"));
784	return true;
785	case OP_MOVSWD:
786	case OP_STOSWD:
787	if ( pCpu->prefix == (PREFIX_REP\|PREFIX_REX)
788	&& pRegFrame->rcx >= 0x40
789	)
790	{
791	Assert(pCpu->mode == CPUMODE_64BIT);
792
793	Log(("pgmPoolMonitorIsReused: OP_STOSQ\n"));
794	return true;
795	}
796	return false;
797	}
798	if ( (pCpu->param1.flags & USE_REG_GEN32)
799	&& (pCpu->param1.base.reg_gen == USE_REG_ESP))
800	{
801	Log4(("pgmPoolMonitorIsReused: ESP\n"));
802	return true;
803	}
804
805	//if (pPage->fCR3Mix)
806	// return false;
807	return false;
808	}
809
810
811	/**
812	* Flushes the page being accessed.
813	*
814	* @returns VBox status code suitable for scheduling.
815	* @param pVM The VM handle.
816	* @param pPool The pool.
817	* @param pPage The pool page (head).
818	* @param pCpu The disassembly of the write instruction.
819	* @param pRegFrame The trap register frame.
820	* @param GCPhysFault The fault address as guest physical address.
821	* @param pvFault The fault address.
822	*/
823	static int pgmPoolAccessHandlerFlush(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
824	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
825	{
826	/*
827	* First, do the flushing.
828	*/
829	int rc = pgmPoolMonitorChainFlush(pPool, pPage);
830
831	/*
832	* Emulate the instruction (xp/w2k problem, requires pc/cr2/sp detection).
833	*/
834	uint32_t cbWritten;
835	int rc2 = EMInterpretInstructionCPU(pVM, pCpu, pRegFrame, pvFault, &cbWritten);
836	if (RT_SUCCESS(rc2))
837	pRegFrame->rip += pCpu->opsize;
838	else if (rc2 == VERR_EM_INTERPRETER)
839	{
840	#ifdef IN_RC
841	if (PATMIsPatchGCAddr(pVM, (RTRCPTR)pRegFrame->eip))
842	{
843	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for patch code %04x:%RGv, ignoring.\n",
844	pRegFrame->cs, (RTGCPTR)pRegFrame->eip));
845	rc = VINF_SUCCESS;
846	STAM_COUNTER_INC(&pPool->StatMonitorRZIntrFailPatch2);
847	}
848	else
849	#endif
850	{
851	rc = VINF_EM_RAW_EMULATE_INSTR;
852	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
853	}
854	}
855	else
856	rc = rc2;
857
858	/* See use in pgmPoolAccessHandlerSimple(). */
859	PGM_INVL_GUEST_TLBS();
860
861	LogFlow(("pgmPoolAccessHandlerPT: returns %Rrc (flushed)\n", rc));
862	return rc;
863
864	}
865
866
867	/**
868	* Handles the STOSD write accesses.
869	*
870	* @returns VBox status code suitable for scheduling.
871	* @param pVM The VM handle.
872	* @param pPool The pool.
873	* @param pPage The pool page (head).
874	* @param pCpu The disassembly of the write instruction.
875	* @param pRegFrame The trap register frame.
876	* @param GCPhysFault The fault address as guest physical address.
877	* @param pvFault The fault address.
878	*/
879	DECLINLINE(int) pgmPoolAccessHandlerSTOSD(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
880	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
881	{
882	Assert(pCpu->mode == CPUMODE_32BIT);
883
884	/*
885	* Increment the modification counter and insert it into the list
886	* of modified pages the first time.
887	*/
888	if (!pPage->cModifications++)
889	pgmPoolMonitorModifiedInsert(pPool, pPage);
890
891	/*
892	* Execute REP STOSD.
893	*
894	* This ASSUMES that we're not invoked by Trap0e on in a out-of-sync
895	* write situation, meaning that it's safe to write here.
896	*/
897	RTGCUINTPTR pu32 = (RTGCUINTPTR)pvFault;
898	while (pRegFrame->ecx)
899	{
900	pgmPoolMonitorChainChanging(pPool, pPage, GCPhysFault, (RTGCPTR)pu32, NULL);
901	#ifdef IN_RC
902	(uint32_t )pu32 = pRegFrame->eax;
903	#else
904	PGMPhysSimpleWriteGCPhys(pVM, GCPhysFault, &pRegFrame->eax, 4);
905	#endif
906	pu32 += 4;
907	GCPhysFault += 4;
908	pRegFrame->edi += 4;
909	pRegFrame->ecx--;
910	}
911	pRegFrame->rip += pCpu->opsize;
912
913	/* See use in pgmPoolAccessHandlerSimple(). */
914	PGM_INVL_GUEST_TLBS();
915
916	LogFlow(("pgmPoolAccessHandlerSTOSD: returns\n"));
917	return VINF_SUCCESS;
918	}
919
920
921	/**
922	* Handles the simple write accesses.
923	*
924	* @returns VBox status code suitable for scheduling.
925	* @param pVM The VM handle.
926	* @param pPool The pool.
927	* @param pPage The pool page (head).
928	* @param pCpu The disassembly of the write instruction.
929	* @param pRegFrame The trap register frame.
930	* @param GCPhysFault The fault address as guest physical address.
931	* @param pvFault The fault address.
932	*/
933	DECLINLINE(int) pgmPoolAccessHandlerSimple(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
934	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
935	{
936	/*
937	* Increment the modification counter and insert it into the list
938	* of modified pages the first time.
939	*/
940	if (!pPage->cModifications++)
941	pgmPoolMonitorModifiedInsert(pPool, pPage);
942
943	/*
944	* Clear all the pages. ASSUMES that pvFault is readable.
945	*/
946	pgmPoolMonitorChainChanging(pPool, pPage, GCPhysFault, pvFault, pCpu);
947
948	/*
949	* Interpret the instruction.
950	*/
951	uint32_t cb;
952	int rc = EMInterpretInstructionCPU(pVM, pCpu, pRegFrame, pvFault, &cb);
953	if (RT_SUCCESS(rc))
954	pRegFrame->rip += pCpu->opsize;
955	else if (rc == VERR_EM_INTERPRETER)
956	{
957	LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for %04x:%RGv - opcode=%d\n",
958	pRegFrame->cs, (RTGCPTR)pRegFrame->rip, pCpu->pCurInstr->opcode));
959	rc = VINF_EM_RAW_EMULATE_INSTR;
960	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
961	}
962
963	/*
964	* Quick hack, with logging enabled we're getting stale
965	* code TLBs but no data TLB for EIP and crash in EMInterpretDisasOne.
966	* Flushing here is BAD and expensive, I think EMInterpretDisasOne will
967	* have to be fixed to support this. But that'll have to wait till next week.
968	*
969	* An alternative is to keep track of the changed PTEs together with the
970	* GCPhys from the guest PT. This may proove expensive though.
971	*
972	* At the moment, it's VITAL that it's done AFTER the instruction interpreting
973	* because we need the stale TLBs in some cases (XP boot). This MUST be fixed properly!
974	*/
975	PGM_INVL_GUEST_TLBS();
976
977	LogFlow(("pgmPoolAccessHandlerSimple: returns %Rrc cb=%d\n", rc, cb));
978	return rc;
979	}
980
981
982	/**
983	* \#PF Handler callback for PT write accesses.
984	*
985	* @returns VBox status code (appropriate for GC return).
986	* @param pVM VM Handle.
987	* @param uErrorCode CPU Error code.
988	* @param pRegFrame Trap register frame.
989	* NULL on DMA and other non CPU access.
990	* @param pvFault The fault address (cr2).
991	* @param GCPhysFault The GC physical address corresponding to pvFault.
992	* @param pvUser User argument.
993	*/
994	DECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser)
995	{
996	STAM_PROFILE_START(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), a);
997	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
998	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
999	LogFlow(("pgmPoolAccessHandler: pvFault=%RGv pPage=%p:{.idx=%d} GCPhysFault=%RGp\n", pvFault, pPage, pPage->idx, GCPhysFault));
1000
1001	/*
1002	* We should ALWAYS have the list head as user parameter. This
1003	* is because we use that page to record the changes.
1004	*/
1005	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1006
1007	/*
1008	* Disassemble the faulting instruction.
1009	*/
1010	DISCPUSTATE Cpu;
1011	int rc = EMInterpretDisasOne(pVM, pRegFrame, &Cpu, NULL);
1012	AssertRCReturn(rc, rc);
1013
1014	/*
1015	* Check if it's worth dealing with.
1016	*/
1017	bool fReused = false;
1018	if ( ( pPage->cModifications < 48 /** @todo #define / /* @todo need to check that it's not mapping EIP. / /* @todo adjust this! */
1019	\|\| pPage->fCR3Mix)
1020	&& !(fReused = pgmPoolMonitorIsReused(pVM, pPage, pRegFrame, &Cpu, pvFault))
1021	&& !pgmPoolMonitorIsForking(pPool, &Cpu, GCPhysFault & PAGE_OFFSET_MASK))
1022	{
1023	/*
1024	* Simple instructions, no REP prefix.
1025	*/
1026	if (!(Cpu.prefix & (PREFIX_REP \| PREFIX_REPNE)))
1027	{
1028	rc = pgmPoolAccessHandlerSimple(pVM, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
1029	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,Handled), a);
1030	return rc;
1031	}
1032
1033	/*
1034	* Windows is frequently doing small memset() operations (netio test 4k+).
1035	* We have to deal with these or we'll kill the cache and performance.
1036	*/
1037	if ( Cpu.pCurInstr->opcode == OP_STOSWD
1038	&& CPUMGetGuestCPL(pVM, pRegFrame) == 0
1039	&& pRegFrame->ecx <= 0x20
1040	&& pRegFrame->ecx * 4 <= PAGE_SIZE - ((uintptr_t)pvFault & PAGE_OFFSET_MASK)
1041	&& !((uintptr_t)pvFault & 3)
1042	&& (pRegFrame->eax == 0 \|\| pRegFrame->eax == 0x80) /* the two values observed. */
1043	&& Cpu.mode == CPUMODE_32BIT
1044	&& Cpu.opmode == CPUMODE_32BIT
1045	&& Cpu.addrmode == CPUMODE_32BIT
1046	&& Cpu.prefix == PREFIX_REP
1047	&& !pRegFrame->eflags.Bits.u1DF
1048	)
1049	{
1050	rc = pgmPoolAccessHandlerSTOSD(pVM, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
1051	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,RepStosd), a);
1052	return rc;
1053	}
1054
1055	/* REP prefix, don't bother. */
1056	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,RepPrefix));
1057	Log4(("pgmPoolAccessHandler: eax=%#x ecx=%#x edi=%#x esi=%#x rip=%RGv opcode=%d prefix=%#x\n",
1058	pRegFrame->eax, pRegFrame->ecx, pRegFrame->edi, pRegFrame->esi, (RTGCPTR)pRegFrame->rip, Cpu.pCurInstr->opcode, Cpu.prefix));
1059	}
1060
1061	/*
1062	* Not worth it, so flush it.
1063	*
1064	* If we considered it to be reused, don't to back to ring-3
1065	* to emulate failed instructions since we usually cannot
1066	* interpret then. This may be a bit risky, in which case
1067	* the reuse detection must be fixed.
1068	*/
1069	rc = pgmPoolAccessHandlerFlush(pVM, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
1070	if (rc == VINF_EM_RAW_EMULATE_INSTR && fReused)
1071	rc = VINF_SUCCESS;
1072	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,FlushPage), a);
1073	return rc;
1074	}
1075
1076	# endif /* !IN_RING3 */
1077	#endif /* PGMPOOL_WITH_MONITORING */
1078
1079	#ifdef PGMPOOL_WITH_CACHE
1080
1081	/**
1082	* Inserts a page into the GCPhys hash table.
1083	*
1084	* @param pPool The pool.
1085	* @param pPage The page.
1086	*/
1087	DECLINLINE(void) pgmPoolHashInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1088	{
1089	Log3(("pgmPoolHashInsert: %RGp\n", pPage->GCPhys));
1090	Assert(pPage->GCPhys != NIL_RTGCPHYS); Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1091	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1092	pPage->iNext = pPool->aiHash[iHash];
1093	pPool->aiHash[iHash] = pPage->idx;
1094	}
1095
1096
1097	/**
1098	* Removes a page from the GCPhys hash table.
1099	*
1100	* @param pPool The pool.
1101	* @param pPage The page.
1102	*/
1103	DECLINLINE(void) pgmPoolHashRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1104	{
1105	Log3(("pgmPoolHashRemove: %RGp\n", pPage->GCPhys));
1106	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
1107	if (pPool->aiHash[iHash] == pPage->idx)
1108	pPool->aiHash[iHash] = pPage->iNext;
1109	else
1110	{
1111	uint16_t iPrev = pPool->aiHash[iHash];
1112	for (;;)
1113	{
1114	const int16_t i = pPool->aPages[iPrev].iNext;
1115	if (i == pPage->idx)
1116	{
1117	pPool->aPages[iPrev].iNext = pPage->iNext;
1118	break;
1119	}
1120	if (i == NIL_PGMPOOL_IDX)
1121	{
1122	AssertReleaseMsgFailed(("GCPhys=%RGp idx=%#x\n", pPage->GCPhys, pPage->idx));
1123	break;
1124	}
1125	iPrev = i;
1126	}
1127	}
1128	pPage->iNext = NIL_PGMPOOL_IDX;
1129	}
1130
1131
1132	/**
1133	* Frees up one cache page.
1134	*
1135	* @returns VBox status code.
1136	* @retval VINF_SUCCESS on success.
1137	* @retval VERR_PGM_POOL_CLEARED if the deregistration of a physical handler will cause a light weight pool flush.
1138	* @param pPool The pool.
1139	* @param iUser The user index.
1140	*/
1141	static int pgmPoolCacheFreeOne(PPGMPOOL pPool, uint16_t iUser)
1142	{
1143	#ifndef IN_RC
1144	const PVM pVM = pPool->CTX_SUFF(pVM);
1145	#endif
1146	Assert(pPool->iAgeHead != pPool->iAgeTail); /* We shouldn't be here if there < 2 cached entries! */
1147	STAM_COUNTER_INC(&pPool->StatCacheFreeUpOne);
1148
1149	/*
1150	* Select one page from the tail of the age list.
1151	*/
1152	uint16_t iToFree = pPool->iAgeTail;
1153	if (iToFree == iUser)
1154	iToFree = pPool->aPages[iToFree].iAgePrev;
1155	/* This is the alternative to the SyncCR3 pgmPoolCacheUsed calls.
1156	if (pPool->aPages[iToFree].iUserHead != NIL_PGMPOOL_USER_INDEX)
1157	{
1158	uint16_t i = pPool->aPages[iToFree].iAgePrev;
1159	for (unsigned j = 0; j < 10 && i != NIL_PGMPOOL_USER_INDEX; j++, i = pPool->aPages[i].iAgePrev)
1160	{
1161	if (pPool->aPages[iToFree].iUserHead == NIL_PGMPOOL_USER_INDEX)
1162	continue;
1163	iToFree = i;
1164	break;
1165	}
1166	}
1167	*/
1168
1169	Assert(iToFree != iUser);
1170	AssertRelease(iToFree != NIL_PGMPOOL_IDX);
1171
1172	PPGMPOOLPAGE pPage = &pPool->aPages[iToFree];
1173
1174	/*
1175	* Reject any attempts at flushing the currently active shadow CR3 mapping
1176	*/
1177	if (PGMGetHyperCR3(pPool->CTX_SUFF(pVM)) == pPage->Core.Key)
1178	{
1179	/* Refresh the cr3 mapping by putting it at the head of the age list. */
1180	pgmPoolCacheUsed(pPool, pPage);
1181	return pgmPoolCacheFreeOne(pPool, iUser);
1182	}
1183
1184	int rc = pgmPoolFlushPage(pPool, pPage);
1185	if (rc == VINF_SUCCESS)
1186	PGM_INVL_GUEST_TLBS(); /* see PT handler. */
1187	return rc;
1188	}
1189
1190
1191	/**
1192	* Checks if a kind mismatch is really a page being reused
1193	* or if it's just normal remappings.
1194	*
1195	* @returns true if reused and the cached page (enmKind1) should be flushed
1196	* @returns false if not reused.
1197	* @param enmKind1 The kind of the cached page.
1198	* @param enmKind2 The kind of the requested page.
1199	*/
1200	static bool pgmPoolCacheReusedByKind(PGMPOOLKIND enmKind1, PGMPOOLKIND enmKind2)
1201	{
1202	switch (enmKind1)
1203	{
1204	/*
1205	* Never reuse them. There is no remapping in non-paging mode.
1206	*/
1207	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1208	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1209	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1210	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1211	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1212	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1213	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1214	return true;
1215
1216	/*
1217	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1218	*/
1219	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1220	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1221	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1222	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1223	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
1224	switch (enmKind2)
1225	{
1226	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1227	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1228	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1229	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1230	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
1231	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1232	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1233	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1234	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1235	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1236	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1237	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1238	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1239	return true;
1240	default:
1241	return false;
1242	}
1243
1244	/*
1245	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
1246	*/
1247	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1248	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1249	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1250	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1251	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
1252	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1253	switch (enmKind2)
1254	{
1255	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1256	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1257	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1258	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1259	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
1260	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1261	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1262	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1263	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1264	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1265	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1266	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1267	return true;
1268	default:
1269	return false;
1270	}
1271
1272	/*
1273	* These cannot be flushed, and it's common to reuse the PDs as PTs.
1274	*/
1275	case PGMPOOLKIND_ROOT_32BIT_PD:
1276	case PGMPOOLKIND_ROOT_PAE_PD:
1277	case PGMPOOLKIND_ROOT_PDPT:
1278	case PGMPOOLKIND_ROOT_NESTED:
1279	return false;
1280
1281	default:
1282	AssertFatalMsgFailed(("enmKind1=%d\n", enmKind1));
1283	}
1284	}
1285
1286
1287	/**
1288	* Attempts to satisfy a pgmPoolAlloc request from the cache.
1289	*
1290	* @returns VBox status code.
1291	* @retval VINF_PGM_CACHED_PAGE on success.
1292	* @retval VERR_FILE_NOT_FOUND if not found.
1293	* @param pPool The pool.
1294	* @param GCPhys The GC physical address of the page we're gonna shadow.
1295	* @param enmKind The kind of mapping.
1296	* @param iUser The shadow page pool index of the user table.
1297	* @param iUserTable The index into the user table (shadowed).
1298	* @param ppPage Where to store the pointer to the page.
1299	*/
1300	static int pgmPoolCacheAlloc(PPGMPOOL pPool, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
1301	{
1302	#ifndef IN_RC
1303	const PVM pVM = pPool->CTX_SUFF(pVM);
1304	#endif
1305	/*
1306	* Look up the GCPhys in the hash.
1307	*/
1308	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
1309	Log3(("pgmPoolCacheAlloc: %RGp kind %d iUser=%d iUserTable=%x SLOT=%d\n", GCPhys, enmKind, iUser, iUserTable, i));
1310	if (i != NIL_PGMPOOL_IDX)
1311	{
1312	do
1313	{
1314	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1315	Log3(("pgmPoolCacheAlloc: slot %d found page %RGp\n", i, pPage->GCPhys));
1316	if (pPage->GCPhys == GCPhys)
1317	{
1318	if ((PGMPOOLKIND)pPage->enmKind == enmKind)
1319	{
1320	int rc = pgmPoolTrackAddUser(pPool, pPage, iUser, iUserTable);
1321	if (RT_SUCCESS(rc))
1322	{
1323	*ppPage = pPage;
1324	STAM_COUNTER_INC(&pPool->StatCacheHits);
1325	return VINF_PGM_CACHED_PAGE;
1326	}
1327	return rc;
1328	}
1329
1330	/*
1331	* The kind is different. In some cases we should now flush the page
1332	* as it has been reused, but in most cases this is normal remapping
1333	* of PDs as PT or big pages using the GCPhys field in a slightly
1334	* different way than the other kinds.
1335	*/
1336	if (pgmPoolCacheReusedByKind((PGMPOOLKIND)pPage->enmKind, enmKind))
1337	{
1338	STAM_COUNTER_INC(&pPool->StatCacheKindMismatches);
1339	pgmPoolFlushPage(pPool, pPage); /* ASSUMES that VERR_PGM_POOL_CLEARED will be returned by pgmPoolTracInsert. */
1340	PGM_INVL_GUEST_TLBS(); /* see PT handler. */
1341	break;
1342	}
1343	}
1344
1345	/* next */
1346	i = pPage->iNext;
1347	} while (i != NIL_PGMPOOL_IDX);
1348	}
1349
1350	Log3(("pgmPoolCacheAlloc: Missed GCPhys=%RGp enmKind=%d\n", GCPhys, enmKind));
1351	STAM_COUNTER_INC(&pPool->StatCacheMisses);
1352	return VERR_FILE_NOT_FOUND;
1353	}
1354
1355
1356	/**
1357	* Inserts a page into the cache.
1358	*
1359	* @param pPool The pool.
1360	* @param pPage The cached page.
1361	* @param fCanBeCached Set if the page is fit for caching from the caller's point of view.
1362	*/
1363	static void pgmPoolCacheInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fCanBeCached)
1364	{
1365	/*
1366	* Insert into the GCPhys hash if the page is fit for that.
1367	*/
1368	Assert(!pPage->fCached);
1369	if (fCanBeCached)
1370	{
1371	pPage->fCached = true;
1372	pgmPoolHashInsert(pPool, pPage);
1373	Log3(("pgmPoolCacheInsert: Caching %p:{.Core=%RHp, .idx=%d, .enmKind=%d, GCPhys=%RGp}\n",
1374	pPage, pPage->Core.Key, pPage->idx, pPage->enmKind, pPage->GCPhys));
1375	STAM_COUNTER_INC(&pPool->StatCacheCacheable);
1376	}
1377	else
1378	{
1379	Log3(("pgmPoolCacheInsert: Not caching %p:{.Core=%RHp, .idx=%d, .enmKind=%d, GCPhys=%RGp}\n",
1380	pPage, pPage->Core.Key, pPage->idx, pPage->enmKind, pPage->GCPhys));
1381	STAM_COUNTER_INC(&pPool->StatCacheUncacheable);
1382	}
1383
1384	/*
1385	* Insert at the head of the age list.
1386	*/
1387	pPage->iAgePrev = NIL_PGMPOOL_IDX;
1388	pPage->iAgeNext = pPool->iAgeHead;
1389	if (pPool->iAgeHead != NIL_PGMPOOL_IDX)
1390	pPool->aPages[pPool->iAgeHead].iAgePrev = pPage->idx;
1391	else
1392	pPool->iAgeTail = pPage->idx;
1393	pPool->iAgeHead = pPage->idx;
1394	}
1395
1396
1397	/**
1398	* Flushes a cached page.
1399	*
1400	* @param pPool The pool.
1401	* @param pPage The cached page.
1402	*/
1403	static void pgmPoolCacheFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1404	{
1405	Log3(("pgmPoolCacheFlushPage: %RGp\n", pPage->GCPhys));
1406
1407	/*
1408	* Remove the page from the hash.
1409	*/
1410	if (pPage->fCached)
1411	{
1412	pPage->fCached = false;
1413	pgmPoolHashRemove(pPool, pPage);
1414	}
1415	else
1416	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
1417
1418	/*
1419	* Remove it from the age list.
1420	*/
1421	if (pPage->iAgeNext != NIL_PGMPOOL_IDX)
1422	pPool->aPages[pPage->iAgeNext].iAgePrev = pPage->iAgePrev;
1423	else
1424	pPool->iAgeTail = pPage->iAgePrev;
1425	if (pPage->iAgePrev != NIL_PGMPOOL_IDX)
1426	pPool->aPages[pPage->iAgePrev].iAgeNext = pPage->iAgeNext;
1427	else
1428	pPool->iAgeHead = pPage->iAgeNext;
1429	pPage->iAgeNext = NIL_PGMPOOL_IDX;
1430	pPage->iAgePrev = NIL_PGMPOOL_IDX;
1431	}
1432
1433	#endif /* PGMPOOL_WITH_CACHE */
1434	#ifdef PGMPOOL_WITH_MONITORING
1435
1436	/**
1437	* Looks for pages sharing the monitor.
1438	*
1439	* @returns Pointer to the head page.
1440	* @returns NULL if not found.
1441	* @param pPool The Pool
1442	* @param pNewPage The page which is going to be monitored.
1443	*/
1444	static PPGMPOOLPAGE pgmPoolMonitorGetPageByGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pNewPage)
1445	{
1446	#ifdef PGMPOOL_WITH_CACHE
1447	/*
1448	* Look up the GCPhys in the hash.
1449	*/
1450	RTGCPHYS GCPhys = pNewPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1451	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
1452	if (i == NIL_PGMPOOL_IDX)
1453	return NULL;
1454	do
1455	{
1456	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1457	if ( pPage->GCPhys - GCPhys < PAGE_SIZE
1458	&& pPage != pNewPage)
1459	{
1460	switch (pPage->enmKind)
1461	{
1462	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1463	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1464	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1465	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
1466	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1467	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1468	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1469	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
1470	case PGMPOOLKIND_ROOT_32BIT_PD:
1471	case PGMPOOLKIND_ROOT_PAE_PD:
1472	case PGMPOOLKIND_ROOT_PDPT:
1473	{
1474	/* find the head */
1475	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1476	{
1477	Assert(pPage->iMonitoredPrev != pPage->idx);
1478	pPage = &pPool->aPages[pPage->iMonitoredPrev];
1479	}
1480	return pPage;
1481	}
1482
1483	/* ignore, no monitoring. */
1484	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1485	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1486	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1487	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1488	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1489	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1490	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1491	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1492	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1493	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1494	case PGMPOOLKIND_ROOT_NESTED:
1495	break;
1496	default:
1497	AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
1498	}
1499	}
1500
1501	/* next */
1502	i = pPage->iNext;
1503	} while (i != NIL_PGMPOOL_IDX);
1504	#endif
1505	return NULL;
1506	}
1507
1508
1509	/**
1510	* Enabled write monitoring of a guest page.
1511	*
1512	* @returns VBox status code.
1513	* @retval VINF_SUCCESS on success.
1514	* @retval VERR_PGM_POOL_CLEARED if the registration of the physical handler will cause a light weight pool flush.
1515	* @param pPool The pool.
1516	* @param pPage The cached page.
1517	*/
1518	static int pgmPoolMonitorInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1519	{
1520	LogFlow(("pgmPoolMonitorInsert %RGp\n", pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1)));
1521
1522	/*
1523	* Filter out the relevant kinds.
1524	*/
1525	switch (pPage->enmKind)
1526	{
1527	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1528	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1529	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1530	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1531	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1532	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1533	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
1534	case PGMPOOLKIND_ROOT_PDPT:
1535	break;
1536
1537	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1538	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1539	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1540	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1541	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1542	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1543	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1544	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1545	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1546	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1547	case PGMPOOLKIND_ROOT_NESTED:
1548	/* Nothing to monitor here. */
1549	return VINF_SUCCESS;
1550
1551	case PGMPOOLKIND_ROOT_32BIT_PD:
1552	case PGMPOOLKIND_ROOT_PAE_PD:
1553	#ifdef PGMPOOL_WITH_MIXED_PT_CR3
1554	break;
1555	#endif
1556	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
1557	default:
1558	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
1559	}
1560
1561	/*
1562	* Install handler.
1563	*/
1564	int rc;
1565	PPGMPOOLPAGE pPageHead = pgmPoolMonitorGetPageByGCPhys(pPool, pPage);
1566	if (pPageHead)
1567	{
1568	Assert(pPageHead != pPage); Assert(pPageHead->iMonitoredNext != pPage->idx);
1569	Assert(pPageHead->iMonitoredPrev != pPage->idx);
1570	pPage->iMonitoredPrev = pPageHead->idx;
1571	pPage->iMonitoredNext = pPageHead->iMonitoredNext;
1572	if (pPageHead->iMonitoredNext != NIL_PGMPOOL_IDX)
1573	pPool->aPages[pPageHead->iMonitoredNext].iMonitoredPrev = pPage->idx;
1574	pPageHead->iMonitoredNext = pPage->idx;
1575	rc = VINF_SUCCESS;
1576	}
1577	else
1578	{
1579	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX); Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1580	PVM pVM = pPool->CTX_SUFF(pVM);
1581	const RTGCPHYS GCPhysPage = pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
1582	rc = PGMHandlerPhysicalRegisterEx(pVM, PGMPHYSHANDLERTYPE_PHYSICAL_WRITE,
1583	GCPhysPage, GCPhysPage + (PAGE_SIZE - 1),
1584	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
1585	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
1586	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
1587	pPool->pszAccessHandler);
1588	/** @todo we should probably deal with out-of-memory conditions here, but for now increasing
1589	* the heap size should suffice. */
1590	AssertFatalRC(rc);
1591	if (pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
1592	rc = VERR_PGM_POOL_CLEARED;
1593	}
1594	pPage->fMonitored = true;
1595	return rc;
1596	}
1597
1598
1599	/**
1600	* Disables write monitoring of a guest page.
1601	*
1602	* @returns VBox status code.
1603	* @retval VINF_SUCCESS on success.
1604	* @retval VERR_PGM_POOL_CLEARED if the deregistration of the physical handler will cause a light weight pool flush.
1605	* @param pPool The pool.
1606	* @param pPage The cached page.
1607	*/
1608	static int pgmPoolMonitorFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1609	{
1610	/*
1611	* Filter out the relevant kinds.
1612	*/
1613	switch (pPage->enmKind)
1614	{
1615	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1616	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1617	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
1618	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1619	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
1620	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
1621	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
1622	case PGMPOOLKIND_ROOT_PDPT:
1623	break;
1624
1625	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1626	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1627	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1628	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1629	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1630	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
1631	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
1632	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
1633	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
1634	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
1635	case PGMPOOLKIND_ROOT_NESTED:
1636	/* Nothing to monitor here. */
1637	return VINF_SUCCESS;
1638
1639	case PGMPOOLKIND_ROOT_32BIT_PD:
1640	case PGMPOOLKIND_ROOT_PAE_PD:
1641	#ifdef PGMPOOL_WITH_MIXED_PT_CR3
1642	break;
1643	#endif
1644	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
1645	default:
1646	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
1647	}
1648
1649	/*
1650	* Remove the page from the monitored list or uninstall it if last.
1651	*/
1652	const PVM pVM = pPool->CTX_SUFF(pVM);
1653	int rc;
1654	if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
1655	\|\| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1656	{
1657	if (pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
1658	{
1659	PPGMPOOLPAGE pNewHead = &pPool->aPages[pPage->iMonitoredNext];
1660	pNewHead->iMonitoredPrev = NIL_PGMPOOL_IDX;
1661	pNewHead->fCR3Mix = pPage->fCR3Mix;
1662	rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
1663	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pNewHead),
1664	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pNewHead),
1665	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pNewHead),
1666	pPool->pszAccessHandler);
1667	AssertFatalRCSuccess(rc);
1668	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
1669	}
1670	else
1671	{
1672	pPool->aPages[pPage->iMonitoredPrev].iMonitoredNext = pPage->iMonitoredNext;
1673	if (pPage->iMonitoredNext != NIL_PGMPOOL_IDX)
1674	{
1675	pPool->aPages[pPage->iMonitoredNext].iMonitoredPrev = pPage->iMonitoredPrev;
1676	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
1677	}
1678	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
1679	rc = VINF_SUCCESS;
1680	}
1681	}
1682	else
1683	{
1684	rc = PGMHandlerPhysicalDeregister(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1));
1685	AssertFatalRC(rc);
1686	if (pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
1687	rc = VERR_PGM_POOL_CLEARED;
1688	}
1689	pPage->fMonitored = false;
1690
1691	/*
1692	* Remove it from the list of modified pages (if in it).
1693	*/
1694	pgmPoolMonitorModifiedRemove(pPool, pPage);
1695
1696	return rc;
1697	}
1698
1699	# ifdef PGMPOOL_WITH_MIXED_PT_CR3
1700
1701	/**
1702	* Set or clear the fCR3Mix attribute in a chain of monitored pages.
1703	*
1704	* @param pPool The Pool.
1705	* @param pPage A page in the chain.
1706	* @param fCR3Mix The new fCR3Mix value.
1707	*/
1708	static void pgmPoolMonitorChainChangeCR3Mix(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fCR3Mix)
1709	{
1710	/* current */
1711	pPage->fCR3Mix = fCR3Mix;
1712
1713	/* before */
1714	int16_t idx = pPage->iMonitoredPrev;
1715	while (idx != NIL_PGMPOOL_IDX)
1716	{
1717	pPool->aPages[idx].fCR3Mix = fCR3Mix;
1718	idx = pPool->aPages[idx].iMonitoredPrev;
1719	}
1720
1721	/* after */
1722	idx = pPage->iMonitoredNext;
1723	while (idx != NIL_PGMPOOL_IDX)
1724	{
1725	pPool->aPages[idx].fCR3Mix = fCR3Mix;
1726	idx = pPool->aPages[idx].iMonitoredNext;
1727	}
1728	}
1729
1730
1731	/**
1732	* Installs or modifies monitoring of a CR3 page (special).
1733	*
1734	* We're pretending the CR3 page is shadowed by the pool so we can use the
1735	* generic mechanisms in detecting chained monitoring. (This also gives us a
1736	* tast of what code changes are required to really pool CR3 shadow pages.)
1737	*
1738	* @returns VBox status code.
1739	* @param pPool The pool.
1740	* @param idxRoot The CR3 (root) page index.
1741	* @param GCPhysCR3 The (new) CR3 value.
1742	*/
1743	int pgmPoolMonitorMonitorCR3(PPGMPOOL pPool, uint16_t idxRoot, RTGCPHYS GCPhysCR3)
1744	{
1745	Assert(idxRoot != NIL_PGMPOOL_IDX && idxRoot < PGMPOOL_IDX_FIRST);
1746	PPGMPOOLPAGE pPage = &pPool->aPages[idxRoot];
1747	LogFlow(("pgmPoolMonitorMonitorCR3: idxRoot=%d pPage=%p:{.GCPhys=%RGp, .fMonitored=%d} GCPhysCR3=%RGp\n",
1748	idxRoot, pPage, pPage->GCPhys, pPage->fMonitored, GCPhysCR3));
1749
1750	/*
1751	* The unlikely case where it already matches.
1752	*/
1753	if (pPage->GCPhys == GCPhysCR3)
1754	{
1755	Assert(pPage->fMonitored);
1756	return VINF_SUCCESS;
1757	}
1758
1759	/*
1760	* Flush the current monitoring and remove it from the hash.
1761	*/
1762	int rc = VINF_SUCCESS;
1763	if (pPage->fMonitored)
1764	{
1765	pgmPoolMonitorChainChangeCR3Mix(pPool, pPage, false);
1766	rc = pgmPoolMonitorFlush(pPool, pPage);
1767	if (rc == VERR_PGM_POOL_CLEARED)
1768	rc = VINF_SUCCESS;
1769	else
1770	AssertFatalRC(rc);
1771	pgmPoolHashRemove(pPool, pPage);
1772	}
1773
1774	/*
1775	* Monitor the page at the new location and insert it into the hash.
1776	*/
1777	pPage->GCPhys = GCPhysCR3;
1778	int rc2 = pgmPoolMonitorInsert(pPool, pPage);
1779	if (rc2 != VERR_PGM_POOL_CLEARED)
1780	{
1781	AssertFatalRC(rc2);
1782	if (rc2 != VINF_SUCCESS && rc == VINF_SUCCESS)
1783	rc = rc2;
1784	}
1785	pgmPoolHashInsert(pPool, pPage);
1786	pgmPoolMonitorChainChangeCR3Mix(pPool, pPage, true);
1787	return rc;
1788	}
1789
1790
1791	/**
1792	* Removes the monitoring of a CR3 page (special).
1793	*
1794	* @returns VBox status code.
1795	* @param pPool The pool.
1796	* @param idxRoot The CR3 (root) page index.
1797	*/
1798	int pgmPoolMonitorUnmonitorCR3(PPGMPOOL pPool, uint16_t idxRoot)
1799	{
1800	Assert(idxRoot != NIL_PGMPOOL_IDX && idxRoot < PGMPOOL_IDX_FIRST);
1801	PPGMPOOLPAGE pPage = &pPool->aPages[idxRoot];
1802	LogFlow(("pgmPoolMonitorUnmonitorCR3: idxRoot=%d pPage=%p:{.GCPhys=%RGp, .fMonitored=%d}\n",
1803	idxRoot, pPage, pPage->GCPhys, pPage->fMonitored));
1804
1805	if (!pPage->fMonitored)
1806	return VINF_SUCCESS;
1807
1808	pgmPoolMonitorChainChangeCR3Mix(pPool, pPage, false);
1809	int rc = pgmPoolMonitorFlush(pPool, pPage);
1810	if (rc != VERR_PGM_POOL_CLEARED)
1811	AssertFatalRC(rc);
1812	else
1813	rc = VINF_SUCCESS;
1814	pgmPoolHashRemove(pPool, pPage);
1815	Assert(!pPage->fMonitored);
1816	pPage->GCPhys = NIL_RTGCPHYS;
1817	return rc;
1818	}
1819
1820	# endif /* PGMPOOL_WITH_MIXED_PT_CR3 */
1821
1822	/**
1823	* Inserts the page into the list of modified pages.
1824	*
1825	* @param pPool The pool.
1826	* @param pPage The page.
1827	*/
1828	void pgmPoolMonitorModifiedInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1829	{
1830	Log3(("pgmPoolMonitorModifiedInsert: idx=%d\n", pPage->idx));
1831	AssertMsg( pPage->iModifiedNext == NIL_PGMPOOL_IDX
1832	&& pPage->iModifiedPrev == NIL_PGMPOOL_IDX
1833	&& pPool->iModifiedHead != pPage->idx,
1834	("Next=%d Prev=%d idx=%d cModifications=%d Head=%d cModifiedPages=%d\n",
1835	pPage->iModifiedNext, pPage->iModifiedPrev, pPage->idx, pPage->cModifications,
1836	pPool->iModifiedHead, pPool->cModifiedPages));
1837
1838	pPage->iModifiedNext = pPool->iModifiedHead;
1839	if (pPool->iModifiedHead != NIL_PGMPOOL_IDX)
1840	pPool->aPages[pPool->iModifiedHead].iModifiedPrev = pPage->idx;
1841	pPool->iModifiedHead = pPage->idx;
1842	pPool->cModifiedPages++;
1843	#ifdef VBOX_WITH_STATISTICS
1844	if (pPool->cModifiedPages > pPool->cModifiedPagesHigh)
1845	pPool->cModifiedPagesHigh = pPool->cModifiedPages;
1846	#endif
1847	}
1848
1849
1850	/**
1851	* Removes the page from the list of modified pages and resets the
1852	* moficiation counter.
1853	*
1854	* @param pPool The pool.
1855	* @param pPage The page which is believed to be in the list of modified pages.
1856	*/
1857	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1858	{
1859	Log3(("pgmPoolMonitorModifiedRemove: idx=%d cModifications=%d\n", pPage->idx, pPage->cModifications));
1860	if (pPool->iModifiedHead == pPage->idx)
1861	{
1862	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1863	pPool->iModifiedHead = pPage->iModifiedNext;
1864	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1865	{
1866	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = NIL_PGMPOOL_IDX;
1867	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1868	}
1869	pPool->cModifiedPages--;
1870	}
1871	else if (pPage->iModifiedPrev != NIL_PGMPOOL_IDX)
1872	{
1873	pPool->aPages[pPage->iModifiedPrev].iModifiedNext = pPage->iModifiedNext;
1874	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
1875	{
1876	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = pPage->iModifiedPrev;
1877	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1878	}
1879	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1880	pPool->cModifiedPages--;
1881	}
1882	else
1883	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
1884	pPage->cModifications = 0;
1885	}
1886
1887
1888	/**
1889	* Zaps the list of modified pages, resetting their modification counters in the process.
1890	*
1891	* @param pVM The VM handle.
1892	*/
1893	void pgmPoolMonitorModifiedClearAll(PVM pVM)
1894	{
1895	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1896	LogFlow(("pgmPoolMonitorModifiedClearAll: cModifiedPages=%d\n", pPool->cModifiedPages));
1897
1898	unsigned cPages = 0; NOREF(cPages);
1899	uint16_t idx = pPool->iModifiedHead;
1900	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
1901	while (idx != NIL_PGMPOOL_IDX)
1902	{
1903	PPGMPOOLPAGE pPage = &pPool->aPages[idx];
1904	idx = pPage->iModifiedNext;
1905	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1906	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1907	pPage->cModifications = 0;
1908	Assert(++cPages);
1909	}
1910	AssertMsg(cPages == pPool->cModifiedPages, ("%d != %d\n", cPages, pPool->cModifiedPages));
1911	pPool->cModifiedPages = 0;
1912	}
1913
1914
1915	/**
1916	* Clear all shadow pages and clear all modification counters.
1917	*
1918	* @param pVM The VM handle.
1919	* @remark Should only be used when monitoring is available, thus placed in
1920	* the PGMPOOL_WITH_MONITORING #ifdef.
1921	*/
1922	void pgmPoolClearAll(PVM pVM)
1923	{
1924	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1925	STAM_PROFILE_START(&pPool->StatClearAll, c);
1926	LogFlow(("pgmPoolClearAll: cUsedPages=%d\n", pPool->cUsedPages));
1927
1928	/*
1929	* Iterate all the pages until we've encountered all that in use.
1930	* This is simple but not quite optimal solution.
1931	*/
1932	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
1933	unsigned cLeft = pPool->cUsedPages;
1934	unsigned iPage = pPool->cCurPages;
1935	while (--iPage >= PGMPOOL_IDX_FIRST)
1936	{
1937	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
1938	if (pPage->GCPhys != NIL_RTGCPHYS)
1939	{
1940	switch (pPage->enmKind)
1941	{
1942	/*
1943	* We only care about shadow page tables.
1944	*/
1945	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1946	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1947	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
1948	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
1949	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1950	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
1951	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1952	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
1953	{
1954	#ifdef PGMPOOL_WITH_USER_TRACKING
1955	if (pPage->cPresent)
1956	#endif
1957	{
1958	void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
1959	STAM_PROFILE_START(&pPool->StatZeroPage, z);
1960	ASMMemZeroPage(pvShw);
1961	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
1962	#ifdef PGMPOOL_WITH_USER_TRACKING
1963	pPage->cPresent = 0;
1964	pPage->iFirstPresent = ~0;
1965	#endif
1966	}
1967	}
1968	/* fall thru */
1969
1970	default:
1971	Assert(!pPage->cModifications \|\| ++cModifiedPages);
1972	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
1973	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
1974	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1975	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1976	pPage->cModifications = 0;
1977	break;
1978
1979	}
1980	if (!--cLeft)
1981	break;
1982	}
1983	}
1984
1985	/* swipe the special pages too. */
1986	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
1987	{
1988	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
1989	if (pPage->GCPhys != NIL_RTGCPHYS)
1990	{
1991	Assert(!pPage->cModifications \|\| ++cModifiedPages);
1992	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
1993	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
1994	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
1995	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
1996	pPage->cModifications = 0;
1997	}
1998	}
1999
2000	#ifndef DEBUG_michael
2001	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
2002	#endif
2003	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
2004	pPool->cModifiedPages = 0;
2005
2006	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2007	/*
2008	* Clear all the GCPhys links and rebuild the phys ext free list.
2009	*/
2010	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
2011	pRam;
2012	pRam = pRam->CTX_SUFF(pNext))
2013	{
2014	unsigned iPage = pRam->cb >> PAGE_SHIFT;
2015	while (iPage-- > 0)
2016	pRam->aPages[iPage].HCPhys &= MM_RAM_FLAGS_NO_REFS_MASK; /** @todo PAGE FLAGS */
2017	}
2018
2019	pPool->iPhysExtFreeHead = 0;
2020	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
2021	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
2022	for (unsigned i = 0; i < cMaxPhysExts; i++)
2023	{
2024	paPhysExts[i].iNext = i + 1;
2025	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
2026	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
2027	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
2028	}
2029	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
2030	#endif
2031
2032
2033	pPool->cPresent = 0;
2034	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
2035	}
2036
2037
2038	/**
2039	* Handle SyncCR3 pool tasks
2040	*
2041	* @returns VBox status code.
2042	* @retval VINF_SUCCESS if successfully added.
2043	* @retval VINF_PGM_SYNC_CR3 is it needs to be deferred to ring 3 (GC only)
2044	* @param pVM The VM handle.
2045	* @remark Should only be used when monitoring is available, thus placed in
2046	* the PGMPOOL_WITH_MONITORING #ifdef.
2047	*/
2048	int pgmPoolSyncCR3(PVM pVM)
2049	{
2050	/*
2051	* When monitoring shadowed pages, we reset the modification counters on CR3 sync.
2052	* Occasionally we will have to clear all the shadow page tables because we wanted
2053	* to monitor a page which was mapped by too many shadowed page tables. This operation
2054	* sometimes refered to as a 'lightweight flush'.
2055	*/
2056	if (!(pVM->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL))
2057	pgmPoolMonitorModifiedClearAll(pVM);
2058	else
2059	{
2060	# ifndef IN_RC
2061	pVM->pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
2062	pgmPoolClearAll(pVM);
2063	# else
2064	LogFlow(("SyncCR3: PGM_SYNC_CLEAR_PGM_POOL is set -> VINF_PGM_SYNC_CR3\n"));
2065	VM_FF_SET(pVM, VM_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
2066	return VINF_PGM_SYNC_CR3;
2067	# endif
2068	}
2069	return VINF_SUCCESS;
2070	}
2071
2072	#endif /* PGMPOOL_WITH_MONITORING */
2073	#ifdef PGMPOOL_WITH_USER_TRACKING
2074
2075	/**
2076	* Frees up at least one user entry.
2077	*
2078	* @returns VBox status code.
2079	* @retval VINF_SUCCESS if successfully added.
2080	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2081	* @param pPool The pool.
2082	* @param iUser The user index.
2083	*/
2084	static int pgmPoolTrackFreeOneUser(PPGMPOOL pPool, uint16_t iUser)
2085	{
2086	STAM_COUNTER_INC(&pPool->StatTrackFreeUpOneUser);
2087	#ifdef PGMPOOL_WITH_CACHE
2088	/*
2089	* Just free cached pages in a braindead fashion.
2090	*/
2091	/** @todo walk the age list backwards and free the first with usage. */
2092	int rc = VINF_SUCCESS;
2093	do
2094	{
2095	int rc2 = pgmPoolCacheFreeOne(pPool, iUser);
2096	if (RT_FAILURE(rc2) && rc == VINF_SUCCESS)
2097	rc = rc2;
2098	} while (pPool->iUserFreeHead == NIL_PGMPOOL_USER_INDEX);
2099	return rc;
2100	#else
2101	/*
2102	* Lazy approach.
2103	*/
2104	/* @todo incompatible with long mode paging (cr3 root will be flushed) */
2105	Assert(!CPUMIsGuestInLongMode(pVM));
2106	pgmPoolFlushAllInt(pPool);
2107	return VERR_PGM_POOL_FLUSHED;
2108	#endif
2109	}
2110
2111
2112	/**
2113	* Inserts a page into the cache.
2114	*
2115	* This will create user node for the page, insert it into the GCPhys
2116	* hash, and insert it into the age list.
2117	*
2118	* @returns VBox status code.
2119	* @retval VINF_SUCCESS if successfully added.
2120	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2121	* @retval VERR_PGM_POOL_CLEARED if the deregistration of the physical handler will cause a light weight pool flush.
2122	* @param pPool The pool.
2123	* @param pPage The cached page.
2124	* @param GCPhys The GC physical address of the page we're gonna shadow.
2125	* @param iUser The user index.
2126	* @param iUserTable The user table index.
2127	*/
2128	DECLINLINE(int) pgmPoolTrackInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhys, uint16_t iUser, uint32_t iUserTable)
2129	{
2130	int rc = VINF_SUCCESS;
2131	PPGMPOOLUSER pUser = pPool->CTX_SUFF(paUsers);
2132
2133	LogFlow(("pgmPoolTrackInsert iUser %d iUserTable %d\n", iUser, iUserTable));
2134
2135	/*
2136	* Find free a user node.
2137	*/
2138	uint16_t i = pPool->iUserFreeHead;
2139	if (i == NIL_PGMPOOL_USER_INDEX)
2140	{
2141	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2142	if (RT_FAILURE(rc))
2143	return rc;
2144	i = pPool->iUserFreeHead;
2145	}
2146
2147	/*
2148	* Unlink the user node from the free list,
2149	* initialize and insert it into the user list.
2150	*/
2151	pPool->iUserFreeHead = pUser[i].iNext;
2152	pUser[i].iNext = NIL_PGMPOOL_USER_INDEX;
2153	pUser[i].iUser = iUser;
2154	pUser[i].iUserTable = iUserTable;
2155	pPage->iUserHead = i;
2156
2157	/*
2158	* Insert into cache and enable monitoring of the guest page if enabled.
2159	*
2160	* Until we implement caching of all levels, including the CR3 one, we'll
2161	* have to make sure we don't try monitor & cache any recursive reuse of
2162	* a monitored CR3 page. Because all windows versions are doing this we'll
2163	* have to be able to do combined access monitoring, CR3 + PT and
2164	* PD + PT (guest PAE).
2165	*
2166	* Update:
2167	* We're now cooperating with the CR3 monitor if an uncachable page is found.
2168	*/
2169	#if defined(PGMPOOL_WITH_MONITORING) \|\| defined(PGMPOOL_WITH_CACHE)
2170	# ifdef PGMPOOL_WITH_MIXED_PT_CR3
2171	const bool fCanBeMonitored = true;
2172	# else
2173	bool fCanBeMonitored = pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored == NIL_RTGCPHYS
2174	\|\| (GCPhys & X86_PTE_PAE_PG_MASK) != (pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored & X86_PTE_PAE_PG_MASK)
2175	\|\| pgmPoolIsBigPage((PGMPOOLKIND)pPage->enmKind);
2176	# endif
2177	# ifdef PGMPOOL_WITH_CACHE
2178	pgmPoolCacheInsert(pPool, pPage, fCanBeMonitored); /* This can be expanded. */
2179	# endif
2180	if (fCanBeMonitored)
2181	{
2182	# ifdef PGMPOOL_WITH_MONITORING
2183	rc = pgmPoolMonitorInsert(pPool, pPage);
2184	if (rc == VERR_PGM_POOL_CLEARED)
2185	{
2186	/* 'Failed' - free the usage, and keep it in the cache (if enabled). */
2187	# ifndef PGMPOOL_WITH_CACHE
2188	pgmPoolMonitorFlush(pPool, pPage);
2189	rc = VERR_PGM_POOL_FLUSHED;
2190	# endif
2191	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
2192	pUser[i].iNext = pPool->iUserFreeHead;
2193	pUser[i].iUser = NIL_PGMPOOL_IDX;
2194	pPool->iUserFreeHead = i;
2195	}
2196	}
2197	# endif
2198	#endif /* PGMPOOL_WITH_MONITORING */
2199	return rc;
2200	}
2201
2202
2203	# ifdef PGMPOOL_WITH_CACHE /* (only used when the cache is enabled.) */
2204	/**
2205	* Adds a user reference to a page.
2206	*
2207	* This will
2208	* This will move the page to the head of the
2209	*
2210	* @returns VBox status code.
2211	* @retval VINF_SUCCESS if successfully added.
2212	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
2213	* @param pPool The pool.
2214	* @param pPage The cached page.
2215	* @param iUser The user index.
2216	* @param iUserTable The user table.
2217	*/
2218	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2219	{
2220	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2221
2222	LogFlow(("pgmPoolTrackAddUser iUser %d iUserTable %d\n", iUser, iUserTable));
2223	# ifdef VBOX_STRICT
2224	/*
2225	* Check that the entry doesn't already exists.
2226	*/
2227	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
2228	{
2229	uint16_t i = pPage->iUserHead;
2230	do
2231	{
2232	Assert(i < pPool->cMaxUsers);
2233	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
2234	i = paUsers[i].iNext;
2235	} while (i != NIL_PGMPOOL_USER_INDEX);
2236	}
2237	# endif
2238
2239	/*
2240	* Allocate a user node.
2241	*/
2242	uint16_t i = pPool->iUserFreeHead;
2243	if (i == NIL_PGMPOOL_USER_INDEX)
2244	{
2245	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
2246	if (RT_FAILURE(rc))
2247	return rc;
2248	i = pPool->iUserFreeHead;
2249	}
2250	pPool->iUserFreeHead = paUsers[i].iNext;
2251
2252	/*
2253	* Initialize the user node and insert it.
2254	*/
2255	paUsers[i].iNext = pPage->iUserHead;
2256	paUsers[i].iUser = iUser;
2257	paUsers[i].iUserTable = iUserTable;
2258	pPage->iUserHead = i;
2259
2260	# ifdef PGMPOOL_WITH_CACHE
2261	/*
2262	* Tell the cache to update its replacement stats for this page.
2263	*/
2264	pgmPoolCacheUsed(pPool, pPage);
2265	# endif
2266	return VINF_SUCCESS;
2267	}
2268	# endif /* PGMPOOL_WITH_CACHE */
2269
2270
2271	/**
2272	* Frees a user record associated with a page.
2273	*
2274	* This does not clear the entry in the user table, it simply replaces the
2275	* user record to the chain of free records.
2276	*
2277	* @param pPool The pool.
2278	* @param HCPhys The HC physical address of the shadow page.
2279	* @param iUser The shadow page pool index of the user table.
2280	* @param iUserTable The index into the user table (shadowed).
2281	*/
2282	static void pgmPoolTrackFreeUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
2283	{
2284	/*
2285	* Unlink and free the specified user entry.
2286	*/
2287	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2288
2289	/* Special: For PAE and 32-bit paging, there is usually no more than one user. */
2290	uint16_t i = pPage->iUserHead;
2291	if ( i != NIL_PGMPOOL_USER_INDEX
2292	&& paUsers[i].iUser == iUser
2293	&& paUsers[i].iUserTable == iUserTable)
2294	{
2295	pPage->iUserHead = paUsers[i].iNext;
2296
2297	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2298	paUsers[i].iNext = pPool->iUserFreeHead;
2299	pPool->iUserFreeHead = i;
2300	return;
2301	}
2302
2303	/* General: Linear search. */
2304	uint16_t iPrev = NIL_PGMPOOL_USER_INDEX;
2305	while (i != NIL_PGMPOOL_USER_INDEX)
2306	{
2307	if ( paUsers[i].iUser == iUser
2308	&& paUsers[i].iUserTable == iUserTable)
2309	{
2310	if (iPrev != NIL_PGMPOOL_USER_INDEX)
2311	paUsers[iPrev].iNext = paUsers[i].iNext;
2312	else
2313	pPage->iUserHead = paUsers[i].iNext;
2314
2315	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2316	paUsers[i].iNext = pPool->iUserFreeHead;
2317	pPool->iUserFreeHead = i;
2318	return;
2319	}
2320	iPrev = i;
2321	i = paUsers[i].iNext;
2322	}
2323
2324	/* Fatal: didn't find it */
2325	AssertFatalMsgFailed(("Didn't find the user entry! iUser=%#x iUserTable=%#x GCPhys=%RGp\n",
2326	iUser, iUserTable, pPage->GCPhys));
2327	}
2328
2329
2330	/**
2331	* Gets the entry size of a shadow table.
2332	*
2333	* @param enmKind The kind of page.
2334	*
2335	* @returns The size of the entry in bytes. That is, 4 or 8.
2336	* @returns If the kind is not for a table, an assertion is raised and 0 is
2337	* returned.
2338	*/
2339	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind)
2340	{
2341	switch (enmKind)
2342	{
2343	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2344	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2345	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2346	case PGMPOOLKIND_ROOT_32BIT_PD:
2347	return 4;
2348
2349	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2350	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2351	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2352	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2353	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2354	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
2355	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2356	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2357	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2358	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
2359	case PGMPOOLKIND_ROOT_PAE_PD:
2360	case PGMPOOLKIND_ROOT_PDPT:
2361	case PGMPOOLKIND_ROOT_NESTED:
2362	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2363	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2364	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2365	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2366	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2367	return 8;
2368
2369	default:
2370	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2371	}
2372	}
2373
2374
2375	/**
2376	* Gets the entry size of a guest table.
2377	*
2378	* @param enmKind The kind of page.
2379	*
2380	* @returns The size of the entry in bytes. That is, 0, 4 or 8.
2381	* @returns If the kind is not for a table, an assertion is raised and 0 is
2382	* returned.
2383	*/
2384	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind)
2385	{
2386	switch (enmKind)
2387	{
2388	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2389	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2390	case PGMPOOLKIND_ROOT_32BIT_PD:
2391	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2392	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2393	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
2394	return 4;
2395
2396	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2397	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2398	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2399	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2400	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2401	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
2402	case PGMPOOLKIND_ROOT_PAE_PD:
2403	case PGMPOOLKIND_ROOT_PDPT:
2404	return 8;
2405
2406	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2407	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2408	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2409	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2410	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2411	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2412	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2413	case PGMPOOLKIND_ROOT_NESTED:
2414	/** @todo can we return 0? (nobody is calling this...) */
2415	AssertFailed();
2416	return 0;
2417
2418	default:
2419	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
2420	}
2421	}
2422
2423	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2424
2425	/**
2426	* Scans one shadow page table for mappings of a physical page.
2427	*
2428	* @param pVM The VM handle.
2429	* @param pPhysPage The guest page in question.
2430	* @param iShw The shadow page table.
2431	* @param cRefs The number of references made in that PT.
2432	*/
2433	static void pgmPoolTrackFlushGCPhysPTInt(PVM pVM, PCPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2434	{
2435	LogFlow(("pgmPoolTrackFlushGCPhysPT: HCPhys=%RHp iShw=%d cRefs=%d\n", pPhysPage->HCPhys, iShw, cRefs));
2436	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2437
2438	/*
2439	* Assert sanity.
2440	*/
2441	Assert(cRefs == 1);
2442	AssertFatalMsg(iShw < pPool->cCurPages && iShw != NIL_PGMPOOL_IDX, ("iShw=%d\n", iShw));
2443	PPGMPOOLPAGE pPage = &pPool->aPages[iShw];
2444
2445	/*
2446	* Then, clear the actual mappings to the page in the shadow PT.
2447	*/
2448	switch (pPage->enmKind)
2449	{
2450	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2451	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2452	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2453	{
2454	const uint32_t u32 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2455	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2456	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2457	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2458	{
2459	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX32 cRefs=%#x\n", i, pPT->a[i], cRefs));
2460	pPT->a[i].u = 0;
2461	cRefs--;
2462	if (!cRefs)
2463	return;
2464	}
2465	#if defined(DEBUG) && !defined(IN_RING0) ///@todo RTLogPrintf is missing in R0.
2466	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2467	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2468	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2469	{
2470	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2471	pPT->a[i].u = 0;
2472	}
2473	#endif
2474	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2475	break;
2476	}
2477
2478	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2479	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2480	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2481	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2482	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2483	{
2484	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2485	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2486	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2487	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2488	{
2489	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2490	pPT->a[i].u = 0;
2491	cRefs--;
2492	if (!cRefs)
2493	return;
2494	}
2495	#if defined(DEBUG) && !defined(IN_RING0) ///@todo RTLogPrintf is missing in R0.
2496	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2497	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2498	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2499	{
2500	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2501	pPT->a[i].u = 0;
2502	}
2503	#endif
2504	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2505	break;
2506	}
2507
2508	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2509	{
2510	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2511	PEPTPT pPT = (PEPTPT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2512	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2513	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2514	{
2515	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
2516	pPT->a[i].u = 0;
2517	cRefs--;
2518	if (!cRefs)
2519	return;
2520	}
2521	#if defined(DEBUG) && !defined(IN_RING0) ///@todo RTLogPrintf is missing in R0.
2522	RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
2523	for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
2524	if ((pPT->a[i].u & (EPT_PTE_PG_MASK \| X86_PTE_P)) == u64)
2525	{
2526	RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
2527	pPT->a[i].u = 0;
2528	}
2529	#endif
2530	AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
2531	break;
2532	}
2533
2534	default:
2535	AssertFatalMsgFailed(("enmKind=%d iShw=%d\n", pPage->enmKind, iShw));
2536	}
2537	}
2538
2539
2540	/**
2541	* Scans one shadow page table for mappings of a physical page.
2542	*
2543	* @param pVM The VM handle.
2544	* @param pPhysPage The guest page in question.
2545	* @param iShw The shadow page table.
2546	* @param cRefs The number of references made in that PT.
2547	*/
2548	void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
2549	{
2550	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
2551	LogFlow(("pgmPoolTrackFlushGCPhysPT: HCPhys=%RHp iShw=%d cRefs=%d\n", pPhysPage->HCPhys, iShw, cRefs));
2552	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPT, f);
2553	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, iShw, cRefs);
2554	pPhysPage->HCPhys &= MM_RAM_FLAGS_NO_REFS_MASK; /** @todo PAGE FLAGS */
2555	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPT, f);
2556	}
2557
2558
2559	/**
2560	* Flushes a list of shadow page tables mapping the same physical page.
2561	*
2562	* @param pVM The VM handle.
2563	* @param pPhysPage The guest page in question.
2564	* @param iPhysExt The physical cross reference extent list to flush.
2565	*/
2566	void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, uint16_t iPhysExt)
2567	{
2568	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2569	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTs, f);
2570	LogFlow(("pgmPoolTrackFlushGCPhysPTs: HCPhys=%RHp iPhysExt\n", pPhysPage->HCPhys, iPhysExt));
2571
2572	const uint16_t iPhysExtStart = iPhysExt;
2573	PPGMPOOLPHYSEXT pPhysExt;
2574	do
2575	{
2576	Assert(iPhysExt < pPool->cMaxPhysExts);
2577	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2578	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
2579	if (pPhysExt->aidx[i] != NIL_PGMPOOL_IDX)
2580	{
2581	pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, pPhysExt->aidx[i], 1);
2582	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
2583	}
2584
2585	/* next */
2586	iPhysExt = pPhysExt->iNext;
2587	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
2588
2589	/* insert the list into the free list and clear the ram range entry. */
2590	pPhysExt->iNext = pPool->iPhysExtFreeHead;
2591	pPool->iPhysExtFreeHead = iPhysExtStart;
2592	pPhysPage->HCPhys &= MM_RAM_FLAGS_NO_REFS_MASK; /** @todo PAGE FLAGS */
2593
2594	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTs, f);
2595	}
2596
2597	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
2598
2599	/**
2600	* Scans all shadow page tables for mappings of a physical page.
2601	*
2602	* This may be slow, but it's most likely more efficient than cleaning
2603	* out the entire page pool / cache.
2604	*
2605	* @returns VBox status code.
2606	* @retval VINF_SUCCESS if all references has been successfully cleared.
2607	* @retval VINF_PGM_GCPHYS_ALIASED if we're better off with a CR3 sync and
2608	* a page pool cleaning.
2609	*
2610	* @param pVM The VM handle.
2611	* @param pPhysPage The guest page in question.
2612	*/
2613	int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage)
2614	{
2615	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2616	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2617	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: cUsedPages=%d cPresent=%d HCPhys=%RHp\n",
2618	pPool->cUsedPages, pPool->cPresent, pPhysPage->HCPhys));
2619
2620	#if 1
2621	/*
2622	* There is a limit to what makes sense.
2623	*/
2624	if (pPool->cPresent > 1024)
2625	{
2626	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
2627	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2628	return VINF_PGM_GCPHYS_ALIASED;
2629	}
2630	#endif
2631
2632	/*
2633	* Iterate all the pages until we've encountered all that in use.
2634	* This is simple but not quite optimal solution.
2635	*/
2636	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
2637	const uint32_t u32 = u64;
2638	unsigned cLeft = pPool->cUsedPages;
2639	unsigned iPage = pPool->cCurPages;
2640	while (--iPage >= PGMPOOL_IDX_FIRST)
2641	{
2642	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
2643	if (pPage->GCPhys != NIL_RTGCPHYS)
2644	{
2645	switch (pPage->enmKind)
2646	{
2647	/*
2648	* We only care about shadow page tables.
2649	*/
2650	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2651	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2652	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2653	{
2654	unsigned cPresent = pPage->cPresent;
2655	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2656	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2657	if (pPT->a[i].n.u1Present)
2658	{
2659	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
2660	{
2661	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX32\n", iPage, i, pPT->a[i]));
2662	pPT->a[i].u = 0;
2663	}
2664	if (!--cPresent)
2665	break;
2666	}
2667	break;
2668	}
2669
2670	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2671	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2672	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2673	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2674	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2675	{
2676	unsigned cPresent = pPage->cPresent;
2677	PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
2678	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
2679	if (pPT->a[i].n.u1Present)
2680	{
2681	if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK \| X86_PTE_P)) == u64)
2682	{
2683	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
2684	pPT->a[i].u = 0;
2685	}
2686	if (!--cPresent)
2687	break;
2688	}
2689	break;
2690	}
2691	}
2692	if (!--cLeft)
2693	break;
2694	}
2695	}
2696
2697	pPhysPage->HCPhys &= MM_RAM_FLAGS_NO_REFS_MASK; /** @todo PAGE FLAGS */
2698	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
2699	return VINF_SUCCESS;
2700	}
2701
2702
2703	/**
2704	* Clears the user entry in a user table.
2705	*
2706	* This is used to remove all references to a page when flushing it.
2707	*/
2708	static void pgmPoolTrackClearPageUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PCPGMPOOLUSER pUser)
2709	{
2710	Assert(pUser->iUser != NIL_PGMPOOL_IDX);
2711	Assert(pUser->iUser < pPool->cCurPages);
2712
2713	/*
2714	* Map the user page.
2715	*/
2716	PPGMPOOLPAGE pUserPage = &pPool->aPages[pUser->iUser];
2717	union
2718	{
2719	uint64_t *pau64;
2720	uint32_t *pau32;
2721	} u;
2722	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pUserPage);
2723
2724	/* Safety precaution in case we change the paging for other modes too in the future. */
2725	Assert(PGMGetHyperCR3(pPool->CTX_SUFF(pVM)) != pPage->Core.Key);
2726
2727	#ifdef VBOX_STRICT
2728	/*
2729	* Some sanity checks.
2730	*/
2731	switch (pUserPage->enmKind)
2732	{
2733	case PGMPOOLKIND_ROOT_32BIT_PD:
2734	Assert(pUser->iUserTable < X86_PG_ENTRIES);
2735	Assert(!(u.pau32[pUser->iUserTable] & PGM_PDFLAGS_MAPPING));
2736	break;
2737	case PGMPOOLKIND_ROOT_PAE_PD:
2738	Assert(pUser->iUserTable < 2048 && pUser->iUser == PGMPOOL_IDX_PAE_PD);
2739	Assert(!(u.pau64[pUser->iUserTable] & PGM_PDFLAGS_MAPPING));
2740	break;
2741	case PGMPOOLKIND_ROOT_PDPT:
2742	Assert(pUser->iUserTable < 4);
2743	Assert(!(u.pau64[pUser->iUserTable] & PGM_PLXFLAGS_PERMANENT));
2744	break;
2745	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
2746	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2747	Assert(pUser->iUserTable < X86_PG_PAE_ENTRIES);
2748	break;
2749	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2750	Assert(pUser->iUserTable < X86_PG_PAE_ENTRIES);
2751	Assert(!(u.pau64[pUser->iUserTable] & PGM_PDFLAGS_MAPPING));
2752	break;
2753	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2754	Assert(pUser->iUserTable < X86_PG_PAE_ENTRIES);
2755	Assert(!(u.pau64[pUser->iUserTable] & PGM_PLXFLAGS_PERMANENT));
2756	break;
2757	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
2758	Assert(!(u.pau64[pUser->iUserTable] & PGM_PLXFLAGS_PERMANENT));
2759	/* GCPhys >> PAGE_SHIFT is the index here */
2760	break;
2761	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2762	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2763	Assert(pUser->iUserTable < X86_PG_PAE_ENTRIES);
2764	break;
2765
2766	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2767	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2768	Assert(pUser->iUserTable < X86_PG_PAE_ENTRIES);
2769	break;
2770
2771	case PGMPOOLKIND_ROOT_NESTED:
2772	Assert(pUser->iUserTable < X86_PG_PAE_ENTRIES);
2773	break;
2774
2775	default:
2776	AssertMsgFailed(("enmKind=%d\n", pUserPage->enmKind));
2777	break;
2778	}
2779	#endif /* VBOX_STRICT */
2780
2781	/*
2782	* Clear the entry in the user page.
2783	*/
2784	switch (pUserPage->enmKind)
2785	{
2786	/* 32-bit entries */
2787	case PGMPOOLKIND_ROOT_32BIT_PD:
2788	u.pau32[pUser->iUserTable] = 0;
2789	break;
2790
2791	/* 64-bit entries */
2792	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
2793	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2794	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2795	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2796	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
2797	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2798	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2799	case PGMPOOLKIND_ROOT_PAE_PD:
2800	case PGMPOOLKIND_ROOT_PDPT:
2801	case PGMPOOLKIND_ROOT_NESTED:
2802	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2803	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2804	u.pau64[pUser->iUserTable] = 0;
2805	break;
2806
2807	default:
2808	AssertFatalMsgFailed(("enmKind=%d iUser=%#x iUserTable=%#x\n", pUserPage->enmKind, pUser->iUser, pUser->iUserTable));
2809	}
2810	}
2811
2812
2813	/**
2814	* Clears all users of a page.
2815	*/
2816	static void pgmPoolTrackClearPageUsers(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2817	{
2818	/*
2819	* Free all the user records.
2820	*/
2821	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
2822	uint16_t i = pPage->iUserHead;
2823	while (i != NIL_PGMPOOL_USER_INDEX)
2824	{
2825	/* Clear enter in user table. */
2826	pgmPoolTrackClearPageUser(pPool, pPage, &paUsers[i]);
2827
2828	/* Free it. */
2829	const uint16_t iNext = paUsers[i].iNext;
2830	paUsers[i].iUser = NIL_PGMPOOL_IDX;
2831	paUsers[i].iNext = pPool->iUserFreeHead;
2832	pPool->iUserFreeHead = i;
2833
2834	/* Next. */
2835	i = iNext;
2836	}
2837	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
2838	}
2839
2840	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
2841
2842	/**
2843	* Allocates a new physical cross reference extent.
2844	*
2845	* @returns Pointer to the allocated extent on success. NULL if we're out of them.
2846	* @param pVM The VM handle.
2847	* @param piPhysExt Where to store the phys ext index.
2848	*/
2849	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt)
2850	{
2851	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2852	uint16_t iPhysExt = pPool->iPhysExtFreeHead;
2853	if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
2854	{
2855	STAM_COUNTER_INC(&pPool->StamTrackPhysExtAllocFailures);
2856	return NULL;
2857	}
2858	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2859	pPool->iPhysExtFreeHead = pPhysExt->iNext;
2860	pPhysExt->iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
2861	*piPhysExt = iPhysExt;
2862	return pPhysExt;
2863	}
2864
2865
2866	/**
2867	* Frees a physical cross reference extent.
2868	*
2869	* @param pVM The VM handle.
2870	* @param iPhysExt The extent to free.
2871	*/
2872	void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt)
2873	{
2874	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2875	Assert(iPhysExt < pPool->cMaxPhysExts);
2876	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2877	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
2878	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
2879	pPhysExt->iNext = pPool->iPhysExtFreeHead;
2880	pPool->iPhysExtFreeHead = iPhysExt;
2881	}
2882
2883
2884	/**
2885	* Frees a physical cross reference extent.
2886	*
2887	* @param pVM The VM handle.
2888	* @param iPhysExt The extent to free.
2889	*/
2890	void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt)
2891	{
2892	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2893
2894	const uint16_t iPhysExtStart = iPhysExt;
2895	PPGMPOOLPHYSEXT pPhysExt;
2896	do
2897	{
2898	Assert(iPhysExt < pPool->cMaxPhysExts);
2899	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
2900	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
2901	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
2902
2903	/* next */
2904	iPhysExt = pPhysExt->iNext;
2905	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
2906
2907	pPhysExt->iNext = pPool->iPhysExtFreeHead;
2908	pPool->iPhysExtFreeHead = iPhysExtStart;
2909	}
2910
2911
2912	/**
2913	* Insert a reference into a list of physical cross reference extents.
2914	*
2915	* @returns The new ram range flags (top 16-bits).
2916	*
2917	* @param pVM The VM handle.
2918	* @param iPhysExt The physical extent index of the list head.
2919	* @param iShwPT The shadow page table index.
2920	*
2921	*/
2922	static uint16_t pgmPoolTrackPhysExtInsert(PVM pVM, uint16_t iPhysExt, uint16_t iShwPT)
2923	{
2924	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2925	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
2926
2927	/* special common case. */
2928	if (paPhysExts[iPhysExt].aidx[2] == NIL_PGMPOOL_IDX)
2929	{
2930	paPhysExts[iPhysExt].aidx[2] = iShwPT;
2931	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
2932	LogFlow(("pgmPoolTrackPhysExtAddref: %d:{,,%d}\n", iPhysExt, iShwPT));
2933	return iPhysExt \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
2934	}
2935
2936	/* general treatment. */
2937	const uint16_t iPhysExtStart = iPhysExt;
2938	unsigned cMax = 15;
2939	for (;;)
2940	{
2941	Assert(iPhysExt < pPool->cMaxPhysExts);
2942	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
2943	if (paPhysExts[iPhysExt].aidx[i] == NIL_PGMPOOL_IDX)
2944	{
2945	paPhysExts[iPhysExt].aidx[i] = iShwPT;
2946	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
2947	LogFlow(("pgmPoolTrackPhysExtAddref: %d:{%d} i=%d cMax=%d\n", iPhysExt, iShwPT, i, cMax));
2948	return iPhysExtStart \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
2949	}
2950	if (!--cMax)
2951	{
2952	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
2953	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
2954	LogFlow(("pgmPoolTrackPhysExtAddref: overflow (1) iShwPT=%d\n", iShwPT));
2955	return MM_RAM_FLAGS_IDX_OVERFLOWED \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
2956	}
2957	}
2958
2959	/* add another extent to the list. */
2960	PPGMPOOLPHYSEXT pNew = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
2961	if (!pNew)
2962	{
2963	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
2964	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
2965	return MM_RAM_FLAGS_IDX_OVERFLOWED \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
2966	}
2967	pNew->iNext = iPhysExtStart;
2968	pNew->aidx[0] = iShwPT;
2969	LogFlow(("pgmPoolTrackPhysExtAddref: added new extent %d:{%d}->%d\n", iPhysExt, iShwPT, iPhysExtStart));
2970	return iPhysExt \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
2971	}
2972
2973
2974	/**
2975	* Add a reference to guest physical page where extents are in use.
2976	*
2977	* @returns The new ram range flags (top 16-bits).
2978	*
2979	* @param pVM The VM handle.
2980	* @param u16 The ram range flags (top 16-bits).
2981	* @param iShwPT The shadow page table index.
2982	*/
2983	uint16_t pgmPoolTrackPhysExtAddref(PVM pVM, uint16_t u16, uint16_t iShwPT)
2984	{
2985	if ((u16 >> (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT)) != MM_RAM_FLAGS_CREFS_PHYSEXT)
2986	{
2987	/*
2988	* Convert to extent list.
2989	*/
2990	Assert((u16 >> (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT)) == 1);
2991	uint16_t iPhysExt;
2992	PPGMPOOLPHYSEXT pPhysExt = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
2993	if (pPhysExt)
2994	{
2995	LogFlow(("pgmPoolTrackPhysExtAddref: new extent: %d:{%d, %d}\n", iPhysExt, u16 & MM_RAM_FLAGS_IDX_MASK, iShwPT));
2996	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliased);
2997	pPhysExt->aidx[0] = u16 & MM_RAM_FLAGS_IDX_MASK;
2998	pPhysExt->aidx[1] = iShwPT;
2999	u16 = iPhysExt \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
3000	}
3001	else
3002	u16 = MM_RAM_FLAGS_IDX_OVERFLOWED \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT));
3003	}
3004	else if (u16 != (MM_RAM_FLAGS_IDX_OVERFLOWED \| (MM_RAM_FLAGS_CREFS_PHYSEXT << (MM_RAM_FLAGS_CREFS_SHIFT - MM_RAM_FLAGS_IDX_SHIFT))))
3005	{
3006	/*
3007	* Insert into the extent list.
3008	*/
3009	u16 = pgmPoolTrackPhysExtInsert(pVM, u16 & MM_RAM_FLAGS_IDX_MASK, iShwPT);
3010	}
3011	else
3012	STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedLots);
3013	return u16;
3014	}
3015
3016
3017	/**
3018	* Clear references to guest physical memory.
3019	*
3020	* @param pPool The pool.
3021	* @param pPage The page.
3022	* @param pPhysPage Pointer to the aPages entry in the ram range.
3023	*/
3024	void pgmPoolTrackPhysExtDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMPAGE pPhysPage)
3025	{
3026	const unsigned cRefs = pPhysPage->HCPhys >> MM_RAM_FLAGS_CREFS_SHIFT; /** @todo PAGE FLAGS */
3027	AssertFatalMsg(cRefs == MM_RAM_FLAGS_CREFS_PHYSEXT, ("cRefs=%d HCPhys=%RHp pPage=%p:{.idx=%d}\n", cRefs, pPhysPage->HCPhys, pPage, pPage->idx));
3028
3029	uint16_t iPhysExt = (pPhysPage->HCPhys >> MM_RAM_FLAGS_IDX_SHIFT) & MM_RAM_FLAGS_IDX_MASK;
3030	if (iPhysExt != MM_RAM_FLAGS_IDX_OVERFLOWED)
3031	{
3032	uint16_t iPhysExtPrev = NIL_PGMPOOL_PHYSEXT_INDEX;
3033	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3034	do
3035	{
3036	Assert(iPhysExt < pPool->cMaxPhysExts);
3037
3038	/*
3039	* Look for the shadow page and check if it's all freed.
3040	*/
3041	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3042	{
3043	if (paPhysExts[iPhysExt].aidx[i] == pPage->idx)
3044	{
3045	paPhysExts[iPhysExt].aidx[i] = NIL_PGMPOOL_IDX;
3046
3047	for (i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
3048	if (paPhysExts[iPhysExt].aidx[i] != NIL_PGMPOOL_IDX)
3049	{
3050	LogFlow(("pgmPoolTrackPhysExtDerefGCPhys: HCPhys=%RX64 idx=%d\n", pPhysPage->HCPhys, pPage->idx));
3051	return;
3052	}
3053
3054	/* we can free the node. */
3055	PVM pVM = pPool->CTX_SUFF(pVM);
3056	const uint16_t iPhysExtNext = paPhysExts[iPhysExt].iNext;
3057	if ( iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX
3058	&& iPhysExtNext == NIL_PGMPOOL_PHYSEXT_INDEX)
3059	{
3060	/* lonely node */
3061	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3062	LogFlow(("pgmPoolTrackPhysExtDerefGCPhys: HCPhys=%RX64 idx=%d lonely\n", pPhysPage->HCPhys, pPage->idx));
3063	pPhysPage->HCPhys &= MM_RAM_FLAGS_NO_REFS_MASK; /** @todo PAGE FLAGS */
3064	}
3065	else if (iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX)
3066	{
3067	/* head */
3068	LogFlow(("pgmPoolTrackPhysExtDerefGCPhys: HCPhys=%RX64 idx=%d head\n", pPhysPage->HCPhys, pPage->idx));
3069	pPhysPage->HCPhys = (pPhysPage->HCPhys & MM_RAM_FLAGS_NO_REFS_MASK) /** @todo PAGE FLAGS */
3070	\| ((uint64_t)MM_RAM_FLAGS_CREFS_PHYSEXT << MM_RAM_FLAGS_CREFS_SHIFT)
3071	\| ((uint64_t)iPhysExtNext << MM_RAM_FLAGS_IDX_SHIFT);
3072	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3073	}
3074	else
3075	{
3076	/* in list */
3077	LogFlow(("pgmPoolTrackPhysExtDerefGCPhys: HCPhys=%RX64 idx=%d\n", pPhysPage->HCPhys, pPage->idx));
3078	paPhysExts[iPhysExtPrev].iNext = iPhysExtNext;
3079	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
3080	}
3081	iPhysExt = iPhysExtNext;
3082	return;
3083	}
3084	}
3085
3086	/* next */
3087	iPhysExtPrev = iPhysExt;
3088	iPhysExt = paPhysExts[iPhysExt].iNext;
3089	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3090
3091	AssertFatalMsgFailed(("not-found! cRefs=%d HCPhys=%RHp pPage=%p:{.idx=%d}\n", cRefs, pPhysPage->HCPhys, pPage, pPage->idx));
3092	}
3093	else /* nothing to do */
3094	LogFlow(("pgmPoolTrackPhysExtDerefGCPhys: HCPhys=%RX64\n", pPhysPage->HCPhys));
3095	}
3096
3097
3098	/**
3099	* Clear references to guest physical memory.
3100	*
3101	* This is the same as pgmPoolTracDerefGCPhys except that the guest physical address
3102	* is assumed to be correct, so the linear search can be skipped and we can assert
3103	* at an earlier point.
3104	*
3105	* @param pPool The pool.
3106	* @param pPage The page.
3107	* @param HCPhys The host physical address corresponding to the guest page.
3108	* @param GCPhys The guest physical address corresponding to HCPhys.
3109	*/
3110	static void pgmPoolTracDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhys)
3111	{
3112	/*
3113	* Walk range list.
3114	*/
3115	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3116	while (pRam)
3117	{
3118	RTGCPHYS off = GCPhys - pRam->GCPhys;
3119	if (off < pRam->cb)
3120	{
3121	/* does it match? */
3122	const unsigned iPage = off >> PAGE_SHIFT;
3123	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3124	#ifdef LOG_ENABLED
3125	RTHCPHYS HCPhysPage = PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]);
3126	Log(("pgmPoolTracDerefGCPhys %RHp vs %RHp\n", HCPhysPage, HCPhys));
3127	#endif
3128	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3129	{
3130	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3131	return;
3132	}
3133	break;
3134	}
3135	pRam = pRam->CTX_SUFF(pNext);
3136	}
3137	AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp\n", HCPhys, GCPhys));
3138	}
3139
3140
3141	/**
3142	* Clear references to guest physical memory.
3143	*
3144	* @param pPool The pool.
3145	* @param pPage The page.
3146	* @param HCPhys The host physical address corresponding to the guest page.
3147	* @param GCPhysHint The guest physical address which may corresponding to HCPhys.
3148	*/
3149	static void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint)
3150	{
3151	/*
3152	* Walk range list.
3153	*/
3154	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3155	while (pRam)
3156	{
3157	RTGCPHYS off = GCPhysHint - pRam->GCPhys;
3158	if (off < pRam->cb)
3159	{
3160	/* does it match? */
3161	const unsigned iPage = off >> PAGE_SHIFT;
3162	Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
3163	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3164	{
3165	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3166	return;
3167	}
3168	break;
3169	}
3170	pRam = pRam->CTX_SUFF(pNext);
3171	}
3172
3173	/*
3174	* Damn, the hint didn't work. We'll have to do an expensive linear search.
3175	*/
3176	STAM_COUNTER_INC(&pPool->StatTrackLinearRamSearches);
3177	pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3178	while (pRam)
3179	{
3180	unsigned iPage = pRam->cb >> PAGE_SHIFT;
3181	while (iPage-- > 0)
3182	{
3183	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
3184	{
3185	Log4(("pgmPoolTracDerefGCPhysHint: Linear HCPhys=%RHp GCPhysHint=%RGp GCPhysReal=%RGp\n",
3186	HCPhys, GCPhysHint, pRam->GCPhys + (iPage << PAGE_SHIFT)));
3187	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
3188	return;
3189	}
3190	}
3191	pRam = pRam->CTX_SUFF(pNext);
3192	}
3193
3194	AssertFatalMsgFailed(("HCPhys=%RHp GCPhysHint=%RGp\n", HCPhys, GCPhysHint));
3195	}
3196
3197
3198	/**
3199	* Clear references to guest physical memory in a 32-bit / 32-bit page table.
3200	*
3201	* @param pPool The pool.
3202	* @param pPage The page.
3203	* @param pShwPT The shadow page table (mapping of the page).
3204	* @param pGstPT The guest page table.
3205	*/
3206	DECLINLINE(void) pgmPoolTrackDerefPT32Bit32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT, PCX86PT pGstPT)
3207	{
3208	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
3209	if (pShwPT->a[i].n.u1Present)
3210	{
3211	Log4(("pgmPoolTrackDerefPT32Bit32Bit: i=%d pte=%RX32 hint=%RX32\n",
3212	i, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3213	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3214	if (!--pPage->cPresent)
3215	break;
3216	}
3217	}
3218
3219
3220	/**
3221	* Clear references to guest physical memory in a PAE / 32-bit page table.
3222	*
3223	* @param pPool The pool.
3224	* @param pPage The page.
3225	* @param pShwPT The shadow page table (mapping of the page).
3226	* @param pGstPT The guest page table (just a half one).
3227	*/
3228	DECLINLINE(void) pgmPoolTrackDerefPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PT pGstPT)
3229	{
3230	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3231	if (pShwPT->a[i].n.u1Present)
3232	{
3233	Log4(("pgmPoolTrackDerefPTPae32Bit: i=%d pte=%RX32 hint=%RX32\n",
3234	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
3235	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
3236	}
3237	}
3238
3239
3240	/**
3241	* Clear references to guest physical memory in a PAE / PAE page table.
3242	*
3243	* @param pPool The pool.
3244	* @param pPage The page.
3245	* @param pShwPT The shadow page table (mapping of the page).
3246	* @param pGstPT The guest page table.
3247	*/
3248	DECLINLINE(void) pgmPoolTrackDerefPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PTPAE pGstPT)
3249	{
3250	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
3251	if (pShwPT->a[i].n.u1Present)
3252	{
3253	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX32 hint=%RX32\n",
3254	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
3255	pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK);
3256	}
3257	}
3258
3259
3260	/**
3261	* Clear references to guest physical memory in a 32-bit / 4MB page table.
3262	*
3263	* @param pPool The pool.
3264	* @param pPage The page.
3265	* @param pShwPT The shadow page table (mapping of the page).
3266	*/
3267	DECLINLINE(void) pgmPoolTrackDerefPT32Bit4MB(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT)
3268	{
3269	RTGCPHYS GCPhys = pPage->GCPhys;
3270	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3271	if (pShwPT->a[i].n.u1Present)
3272	{
3273	Log4(("pgmPoolTrackDerefPT32Bit4MB: i=%d pte=%RX32 GCPhys=%RGp\n",
3274	i, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys));
3275	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys);
3276	}
3277	}
3278
3279
3280	/**
3281	* Clear references to guest physical memory in a PAE / 2/4MB page table.
3282	*
3283	* @param pPool The pool.
3284	* @param pPage The page.
3285	* @param pShwPT The shadow page table (mapping of the page).
3286	*/
3287	DECLINLINE(void) pgmPoolTrackDerefPTPaeBig(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT)
3288	{
3289	RTGCPHYS GCPhys = pPage->GCPhys;
3290	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3291	if (pShwPT->a[i].n.u1Present)
3292	{
3293	Log4(("pgmPoolTrackDerefPTPaeBig: i=%d pte=%RX64 hint=%RGp\n",
3294	i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys));
3295	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys);
3296	}
3297	}
3298
3299	#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
3300
3301	/**
3302	* Clear references to shadowed pages in a PAE (legacy or 64 bits) page directory.
3303	*
3304	* @param pPool The pool.
3305	* @param pPage The page.
3306	* @param pShwPD The shadow page directory (mapping of the page).
3307	*/
3308	DECLINLINE(void) pgmPoolTrackDerefPDPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPAE pShwPD)
3309	{
3310	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3311	{
3312	if (pShwPD->a[i].n.u1Present)
3313	{
3314	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PAE_PG_MASK);
3315	if (pSubPage)
3316	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3317	else
3318	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & X86_PDE_PAE_PG_MASK));
3319	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3320	}
3321	}
3322	}
3323
3324
3325	/**
3326	* Clear references to shadowed pages in a 64-bit page directory pointer table.
3327	*
3328	* @param pPool The pool.
3329	* @param pPage The page.
3330	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3331	*/
3332	DECLINLINE(void) pgmPoolTrackDerefPDPT64Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
3333	{
3334	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3335	{
3336	if (pShwPDPT->a[i].n.u1Present)
3337	{
3338	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
3339	if (pSubPage)
3340	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3341	else
3342	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
3343	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3344	}
3345	}
3346	}
3347
3348
3349	/**
3350	* Clear references to shadowed pages in a 64-bit level 4 page table.
3351	*
3352	* @param pPool The pool.
3353	* @param pPage The page.
3354	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3355	*/
3356	DECLINLINE(void) pgmPoolTrackDerefPML464Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PML4 pShwPML4)
3357	{
3358	for (unsigned i = 0; i < RT_ELEMENTS(pShwPML4->a); i++)
3359	{
3360	if (pShwPML4->a[i].n.u1Present)
3361	{
3362	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPML4->a[i].u & X86_PDPE_PG_MASK);
3363	if (pSubPage)
3364	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3365	else
3366	AssertFatalMsgFailed(("%RX64\n", pShwPML4->a[i].u & X86_PML4E_PG_MASK));
3367	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3368	}
3369	}
3370	}
3371
3372
3373	/**
3374	* Clear references to shadowed pages in an EPT page table.
3375	*
3376	* @param pPool The pool.
3377	* @param pPage The page.
3378	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
3379	*/
3380	DECLINLINE(void) pgmPoolTrackDerefPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT)
3381	{
3382	RTGCPHYS GCPhys = pPage->GCPhys;
3383	for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
3384	if (pShwPT->a[i].n.u1Present)
3385	{
3386	Log4(("pgmPoolTrackDerefPTEPT: i=%d pte=%RX64 GCPhys=%RX64\n",
3387	i, pShwPT->a[i].u & EPT_PTE_PG_MASK, pPage->GCPhys));
3388	pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & EPT_PTE_PG_MASK, GCPhys);
3389	}
3390	}
3391
3392
3393	/**
3394	* Clear references to shadowed pages in an EPT page directory.
3395	*
3396	* @param pPool The pool.
3397	* @param pPage The page.
3398	* @param pShwPD The shadow page directory (mapping of the page).
3399	*/
3400	DECLINLINE(void) pgmPoolTrackDerefPDEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPD pShwPD)
3401	{
3402	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
3403	{
3404	if (pShwPD->a[i].n.u1Present)
3405	{
3406	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & EPT_PDE_PG_MASK);
3407	if (pSubPage)
3408	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3409	else
3410	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & EPT_PDE_PG_MASK));
3411	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3412	}
3413	}
3414	}
3415
3416
3417	/**
3418	* Clear references to shadowed pages in an EPT page directory pointer table.
3419	*
3420	* @param pPool The pool.
3421	* @param pPage The page.
3422	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
3423	*/
3424	DECLINLINE(void) pgmPoolTrackDerefPDPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPDPT pShwPDPT)
3425	{
3426	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
3427	{
3428	if (pShwPDPT->a[i].n.u1Present)
3429	{
3430	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK);
3431	if (pSubPage)
3432	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
3433	else
3434	AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK));
3435	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
3436	}
3437	}
3438	}
3439
3440
3441	/**
3442	* Clears all references made by this page.
3443	*
3444	* This includes other shadow pages and GC physical addresses.
3445	*
3446	* @param pPool The pool.
3447	* @param pPage The page.
3448	*/
3449	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3450	{
3451	/*
3452	* Map the shadow page and take action according to the page kind.
3453	*/
3454	void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
3455	switch (pPage->enmKind)
3456	{
3457	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3458	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3459	{
3460	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3461	void *pvGst;
3462	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3463	pgmPoolTrackDerefPT32Bit32Bit(pPool, pPage, (PX86PT)pvShw, (PCX86PT)pvGst);
3464	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3465	break;
3466	}
3467
3468	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3469	{
3470	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3471	void *pvGst;
3472	int rc = PGM_GCPHYS_2_PTR_EX(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3473	pgmPoolTrackDerefPTPae32Bit(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PT)pvGst);
3474	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3475	break;
3476	}
3477
3478	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3479	{
3480	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3481	void *pvGst;
3482	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
3483	pgmPoolTrackDerefPTPaePae(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PTPAE)pvGst);
3484	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3485	break;
3486	}
3487
3488	case PGMPOOLKIND_32BIT_PT_FOR_PHYS: /* treat it like a 4 MB page */
3489	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3490	{
3491	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3492	pgmPoolTrackDerefPT32Bit4MB(pPool, pPage, (PX86PT)pvShw);
3493	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3494	break;
3495	}
3496
3497	case PGMPOOLKIND_PAE_PT_FOR_PHYS: /* treat it like a 2 MB page */
3498	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3499	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3500	{
3501	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
3502	pgmPoolTrackDerefPTPaeBig(pPool, pPage, (PX86PTPAE)pvShw);
3503	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
3504	break;
3505	}
3506
3507	#else /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3508	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3509	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3510	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3511	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3512	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3513	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3514	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3515	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3516	break;
3517	#endif /* !PGMPOOL_WITH_GCPHYS_TRACKING */
3518
3519	case PGMPOOLKIND_PAE_PD_FOR_32BIT_PD:
3520	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3521	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3522	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3523	pgmPoolTrackDerefPDPae(pPool, pPage, (PX86PDPAE)pvShw);
3524	break;
3525
3526	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3527	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3528	pgmPoolTrackDerefPDPT64Bit(pPool, pPage, (PX86PDPT)pvShw);
3529	break;
3530
3531	case PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4:
3532	pgmPoolTrackDerefPML464Bit(pPool, pPage, (PX86PML4)pvShw);
3533	break;
3534
3535	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3536	pgmPoolTrackDerefPTEPT(pPool, pPage, (PEPTPT)pvShw);
3537	break;
3538
3539	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3540	pgmPoolTrackDerefPDEPT(pPool, pPage, (PEPTPD)pvShw);
3541	break;
3542
3543	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3544	pgmPoolTrackDerefPDPTEPT(pPool, pPage, (PEPTPDPT)pvShw);
3545	break;
3546
3547	default:
3548	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
3549	}
3550
3551	/* paranoia, clear the shadow page. Remove this laser (i.e. let Alloc and ClearAll do it). */
3552	STAM_PROFILE_START(&pPool->StatZeroPage, z);
3553	ASMMemZeroPage(pvShw);
3554	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
3555	pPage->fZeroed = true;
3556	}
3557
3558	#endif /* PGMPOOL_WITH_USER_TRACKING */
3559
3560	/**
3561	* Flushes all the special root pages as part of a pgmPoolFlushAllInt operation.
3562	*
3563	* @param pPool The pool.
3564	*/
3565	static void pgmPoolFlushAllSpecialRoots(PPGMPOOL pPool)
3566	{
3567	/*
3568	* These special pages are all mapped into the indexes 1..PGMPOOL_IDX_FIRST.
3569	*/
3570	Assert(NIL_PGMPOOL_IDX == 0);
3571	for (unsigned i = 1; i < PGMPOOL_IDX_FIRST; i++)
3572	{
3573	/*
3574	* Get the page address.
3575	*/
3576	PPGMPOOLPAGE pPage = &pPool->aPages[i];
3577	union
3578	{
3579	uint64_t *pau64;
3580	uint32_t *pau32;
3581	} u;
3582	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
3583
3584	/*
3585	* Mark stuff not present.
3586	*/
3587	switch (pPage->enmKind)
3588	{
3589	case PGMPOOLKIND_ROOT_32BIT_PD:
3590	for (unsigned iPage = 0; iPage < X86_PG_ENTRIES; iPage++)
3591	if ((u.pau32[iPage] & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == X86_PDE_P)
3592	u.pau32[iPage] = 0;
3593	break;
3594
3595	case PGMPOOLKIND_ROOT_PAE_PD:
3596	for (unsigned iPage = 0; iPage < X86_PG_PAE_ENTRIES * X86_PG_PAE_PDPE_ENTRIES; iPage++)
3597	if ((u.pau64[iPage] & (PGM_PDFLAGS_MAPPING \| X86_PDE_P)) == X86_PDE_P)
3598	u.pau64[iPage] = 0;
3599	break;
3600
3601	case PGMPOOLKIND_ROOT_PDPT:
3602	/* Not root of shadowed pages currently, ignore it. */
3603	break;
3604
3605	case PGMPOOLKIND_ROOT_NESTED:
3606	ASMMemZero32(u.pau64, PAGE_SIZE);
3607	break;
3608	}
3609	}
3610
3611	/*
3612	* Paranoia (to be removed), flag a global CR3 sync.
3613	*/
3614	VM_FF_SET(pPool->CTX_SUFF(pVM), VM_FF_PGM_SYNC_CR3);
3615	}
3616
3617
3618	/**
3619	* Flushes the entire cache.
3620	*
3621	* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
3622	* and execute this CR3 flush.
3623	*
3624	* @param pPool The pool.
3625	*/
3626	static void pgmPoolFlushAllInt(PPGMPOOL pPool)
3627	{
3628	STAM_PROFILE_START(&pPool->StatFlushAllInt, a);
3629	LogFlow(("pgmPoolFlushAllInt:\n"));
3630
3631	/*
3632	* If there are no pages in the pool, there is nothing to do.
3633	*/
3634	if (pPool->cCurPages <= PGMPOOL_IDX_FIRST)
3635	{
3636	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
3637	return;
3638	}
3639
3640	/*
3641	* Nuke the free list and reinsert all pages into it.
3642	*/
3643	for (unsigned i = pPool->cCurPages - 1; i >= PGMPOOL_IDX_FIRST; i--)
3644	{
3645	PPGMPOOLPAGE pPage = &pPool->aPages[i];
3646
3647	#ifdef IN_RING3
3648	Assert(pPage->Core.Key == MMPage2Phys(pPool->pVMR3, pPage->pvPageR3));
3649	#endif
3650	#ifdef PGMPOOL_WITH_MONITORING
3651	if (pPage->fMonitored)
3652	pgmPoolMonitorFlush(pPool, pPage);
3653	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
3654	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
3655	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
3656	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
3657	pPage->cModifications = 0;
3658	#endif
3659	pPage->GCPhys = NIL_RTGCPHYS;
3660	pPage->enmKind = PGMPOOLKIND_FREE;
3661	Assert(pPage->idx == i);
3662	pPage->iNext = i + 1;
3663	pPage->fZeroed = false; /* This could probably be optimized, but better safe than sorry. */
3664	pPage->fSeenNonGlobal = false;
3665	pPage->fMonitored= false;
3666	pPage->fCached = false;
3667	pPage->fReusedFlushPending = false;
3668	pPage->fCR3Mix = false;
3669	#ifdef PGMPOOL_WITH_USER_TRACKING
3670	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
3671	#endif
3672	#ifdef PGMPOOL_WITH_CACHE
3673	pPage->iAgeNext = NIL_PGMPOOL_IDX;
3674	pPage->iAgePrev = NIL_PGMPOOL_IDX;
3675	#endif
3676	}
3677	pPool->aPages[pPool->cCurPages - 1].iNext = NIL_PGMPOOL_IDX;
3678	pPool->iFreeHead = PGMPOOL_IDX_FIRST;
3679	pPool->cUsedPages = 0;
3680
3681	#ifdef PGMPOOL_WITH_USER_TRACKING
3682	/*
3683	* Zap and reinitialize the user records.
3684	*/
3685	pPool->cPresent = 0;
3686	pPool->iUserFreeHead = 0;
3687	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
3688	const unsigned cMaxUsers = pPool->cMaxUsers;
3689	for (unsigned i = 0; i < cMaxUsers; i++)
3690	{
3691	paUsers[i].iNext = i + 1;
3692	paUsers[i].iUser = NIL_PGMPOOL_IDX;
3693	paUsers[i].iUserTable = 0xfffffffe;
3694	}
3695	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
3696	#endif
3697
3698	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
3699	/*
3700	* Clear all the GCPhys links and rebuild the phys ext free list.
3701	*/
3702	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
3703	pRam;
3704	pRam = pRam->CTX_SUFF(pNext))
3705	{
3706	unsigned iPage = pRam->cb >> PAGE_SHIFT;
3707	while (iPage-- > 0)
3708	pRam->aPages[iPage].HCPhys &= MM_RAM_FLAGS_NO_REFS_MASK; /** @todo PAGE FLAGS */
3709	}
3710
3711	pPool->iPhysExtFreeHead = 0;
3712	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
3713	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
3714	for (unsigned i = 0; i < cMaxPhysExts; i++)
3715	{
3716	paPhysExts[i].iNext = i + 1;
3717	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
3718	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
3719	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
3720	}
3721	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
3722	#endif
3723
3724	#ifdef PGMPOOL_WITH_MONITORING
3725	/*
3726	* Just zap the modified list.
3727	*/
3728	pPool->cModifiedPages = 0;
3729	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
3730	#endif
3731
3732	#ifdef PGMPOOL_WITH_CACHE
3733	/*
3734	* Clear the GCPhys hash and the age list.
3735	*/
3736	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
3737	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
3738	pPool->iAgeHead = NIL_PGMPOOL_IDX;
3739	pPool->iAgeTail = NIL_PGMPOOL_IDX;
3740	#endif
3741
3742	/*
3743	* Flush all the special root pages.
3744	* Reinsert active pages into the hash and ensure monitoring chains are correct.
3745	*/
3746	pgmPoolFlushAllSpecialRoots(pPool);
3747	for (unsigned i = PGMPOOL_IDX_FIRST_SPECIAL; i < PGMPOOL_IDX_FIRST; i++)
3748	{
3749	PPGMPOOLPAGE pPage = &pPool->aPages[i];
3750	pPage->iNext = NIL_PGMPOOL_IDX;
3751	#ifdef PGMPOOL_WITH_MONITORING
3752	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
3753	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
3754	pPage->cModifications = 0;
3755	/* ASSUMES that we're not sharing with any of the other special pages (safe for now). */
3756	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
3757	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
3758	if (pPage->fMonitored)
3759	{
3760	PVM pVM = pPool->CTX_SUFF(pVM);
3761	int rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
3762	pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
3763	pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
3764	pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
3765	pPool->pszAccessHandler);
3766	AssertFatalRCSuccess(rc);
3767	# ifdef PGMPOOL_WITH_CACHE
3768	pgmPoolHashInsert(pPool, pPage);
3769	# endif
3770	}
3771	#endif
3772	#ifdef PGMPOOL_WITH_USER_TRACKING
3773	Assert(pPage->iUserHead == NIL_PGMPOOL_USER_INDEX); /* for now */
3774	#endif
3775	#ifdef PGMPOOL_WITH_CACHE
3776	Assert(pPage->iAgeNext == NIL_PGMPOOL_IDX);
3777	Assert(pPage->iAgePrev == NIL_PGMPOOL_IDX);
3778	#endif
3779	}
3780
3781	STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
3782	}
3783
3784
3785	/**
3786	* Flushes a pool page.
3787	*
3788	* This moves the page to the free list after removing all user references to it.
3789	* In GC this will cause a CR3 reload if the page is traced back to an active root page.
3790	*
3791	* @returns VBox status code.
3792	* @retval VINF_SUCCESS on success.
3793	* @retval VERR_PGM_POOL_CLEARED if the deregistration of the physical handler will cause a light weight pool flush.
3794	* @param pPool The pool.
3795	* @param HCPhys The HC physical address of the shadow page.
3796	*/
3797	int pgmPoolFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3798	{
3799	int rc = VINF_SUCCESS;
3800	STAM_PROFILE_START(&pPool->StatFlushPage, f);
3801	LogFlow(("pgmPoolFlushPage: pPage=%p:{.Key=%RHp, .idx=%d, .enmKind=%d, .GCPhys=%RGp}\n",
3802	pPage, pPage->Core.Key, pPage->idx, pPage->enmKind, pPage->GCPhys));
3803
3804	/*
3805	* Quietly reject any attempts at flushing any of the special root pages.
3806	*/
3807	if (pPage->idx < PGMPOOL_IDX_FIRST)
3808	{
3809	Log(("pgmPoolFlushPage: special root page, rejected. enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
3810	return VINF_SUCCESS;
3811	}
3812
3813	/*
3814	* Quietly reject any attempts at flushing the currently active shadow CR3 mapping
3815	*/
3816	if (PGMGetHyperCR3(pPool->CTX_SUFF(pVM)) == pPage->Core.Key)
3817	{
3818	AssertMsg(pPage->enmKind == PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4,
3819	("Can't free the shadow CR3! (%RHp vs %RHp kind=%d\n", PGMGetHyperCR3(pPool->CTX_SUFF(pVM)), pPage->Core.Key, pPage->enmKind));
3820	Log(("pgmPoolFlushPage: current active shadow CR3, rejected. enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
3821	return VINF_SUCCESS;
3822	}
3823
3824	/*
3825	* Mark the page as being in need of a ASMMemZeroPage().
3826	*/
3827	pPage->fZeroed = false;
3828
3829	#ifdef PGMPOOL_WITH_USER_TRACKING
3830	/*
3831	* Clear the page.
3832	*/
3833	pgmPoolTrackClearPageUsers(pPool, pPage);
3834	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
3835	pgmPoolTrackDeref(pPool, pPage);
3836	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
3837	#endif
3838
3839	#ifdef PGMPOOL_WITH_CACHE
3840	/*
3841	* Flush it from the cache.
3842	*/
3843	pgmPoolCacheFlushPage(pPool, pPage);
3844	#endif /* PGMPOOL_WITH_CACHE */
3845
3846	#ifdef PGMPOOL_WITH_MONITORING
3847	/*
3848	* Deregistering the monitoring.
3849	*/
3850	if (pPage->fMonitored)
3851	rc = pgmPoolMonitorFlush(pPool, pPage);
3852	#endif
3853
3854	/*
3855	* Free the page.
3856	*/
3857	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
3858	pPage->iNext = pPool->iFreeHead;
3859	pPool->iFreeHead = pPage->idx;
3860	pPage->enmKind = PGMPOOLKIND_FREE;
3861	pPage->GCPhys = NIL_RTGCPHYS;
3862	pPage->fReusedFlushPending = false;
3863
3864	pPool->cUsedPages--;
3865	STAM_PROFILE_STOP(&pPool->StatFlushPage, f);
3866	return rc;
3867	}
3868
3869
3870	/**
3871	* Frees a usage of a pool page.
3872	*
3873	* The caller is responsible to updating the user table so that it no longer
3874	* references the shadow page.
3875	*
3876	* @param pPool The pool.
3877	* @param HCPhys The HC physical address of the shadow page.
3878	* @param iUser The shadow page pool index of the user table.
3879	* @param iUserTable The index into the user table (shadowed).
3880	*/
3881	void pgmPoolFreeByPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
3882	{
3883	STAM_PROFILE_START(&pPool->StatFree, a);
3884	LogFlow(("pgmPoolFreeByPage: pPage=%p:{.Key=%RHp, .idx=%d, enmKind=%d} iUser=%#x iUserTable=%#x\n",
3885	pPage, pPage->Core.Key, pPage->idx, pPage->enmKind, iUser, iUserTable));
3886	Assert(pPage->idx >= PGMPOOL_IDX_FIRST);
3887	#ifdef PGMPOOL_WITH_USER_TRACKING
3888	pgmPoolTrackFreeUser(pPool, pPage, iUser, iUserTable);
3889	#endif
3890	#ifdef PGMPOOL_WITH_CACHE
3891	if (!pPage->fCached)
3892	#endif
3893	pgmPoolFlushPage(pPool, pPage); /* ASSUMES that VERR_PGM_POOL_CLEARED can be ignored here. */
3894	STAM_PROFILE_STOP(&pPool->StatFree, a);
3895	}
3896
3897
3898	/**
3899	* Makes one or more free page free.
3900	*
3901	* @returns VBox status code.
3902	* @retval VINF_SUCCESS on success.
3903	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
3904	*
3905	* @param pPool The pool.
3906	* @param iUser The user of the page.
3907	*/
3908	static int pgmPoolMakeMoreFreePages(PPGMPOOL pPool, uint16_t iUser)
3909	{
3910	LogFlow(("pgmPoolMakeMoreFreePages: iUser=%#x\n", iUser));
3911
3912	/*
3913	* If the pool isn't full grown yet, expand it.
3914	*/
3915	if (pPool->cCurPages < pPool->cMaxPages)
3916	{
3917	STAM_PROFILE_ADV_SUSPEND(&pPool->StatAlloc, a);
3918	#ifdef IN_RING3
3919	int rc = PGMR3PoolGrow(pPool->pVMR3);
3920	#else
3921	int rc = CTXALLMID(VMM, CallHost)(pPool->CTX_SUFF(pVM), VMMCALLHOST_PGM_POOL_GROW, 0);
3922	#endif
3923	if (RT_FAILURE(rc))
3924	return rc;
3925	STAM_PROFILE_ADV_RESUME(&pPool->StatAlloc, a);
3926	if (pPool->iFreeHead != NIL_PGMPOOL_IDX)
3927	return VINF_SUCCESS;
3928	}
3929
3930	#ifdef PGMPOOL_WITH_CACHE
3931	/*
3932	* Free one cached page.
3933	*/
3934	return pgmPoolCacheFreeOne(pPool, iUser);
3935	#else
3936	/*
3937	* Flush the pool.
3938	* If we have tracking enabled, it should be possible to come up with
3939	* a cheap replacement strategy...
3940	*/
3941	/* @todo incompatible with long mode paging (cr3 root will be flushed) */
3942	Assert(!CPUMIsGuestInLongMode(pVM));
3943	pgmPoolFlushAllInt(pPool);
3944	return VERR_PGM_POOL_FLUSHED;
3945	#endif
3946	}
3947
3948
3949	/**
3950	* Allocates a page from the pool.
3951	*
3952	* This page may actually be a cached page and not in need of any processing
3953	* on the callers part.
3954	*
3955	* @returns VBox status code.
3956	* @retval VINF_SUCCESS if a NEW page was allocated.
3957	* @retval VINF_PGM_CACHED_PAGE if a CACHED page was returned.
3958	* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
3959	* @param pVM The VM handle.
3960	* @param GCPhys The GC physical address of the page we're gonna shadow.
3961	* For 4MB and 2MB PD entries, it's the first address the
3962	* shadow PT is covering.
3963	* @param enmKind The kind of mapping.
3964	* @param iUser The shadow page pool index of the user table.
3965	* @param iUserTable The index into the user table (shadowed).
3966	* @param ppPage Where to store the pointer to the page. NULL is stored here on failure.
3967	*/
3968	int pgmPoolAlloc(PVM pVM, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
3969	{
3970	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3971	STAM_PROFILE_ADV_START(&pPool->StatAlloc, a);
3972	LogFlow(("pgmPoolAlloc: GCPhys=%RGp enmKind=%d iUser=%#x iUserTable=%#x\n", GCPhys, enmKind, iUser, iUserTable));
3973	*ppPage = NULL;
3974
3975	#ifdef PGMPOOL_WITH_CACHE
3976	if (pPool->fCacheEnabled)
3977	{
3978	int rc2 = pgmPoolCacheAlloc(pPool, GCPhys, enmKind, iUser, iUserTable, ppPage);
3979	if (RT_SUCCESS(rc2))
3980	{
3981	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
3982	LogFlow(("pgmPoolAlloc: cached returns %Rrc ppPage=%p:{.Key=%RHp, .idx=%d}\n", rc2, ppPage, (ppPage)->Core.Key, (ppPage)->idx));
3983	return rc2;
3984	}
3985	}
3986	#endif
3987
3988	/*
3989	* Allocate a new one.
3990	*/
3991	int rc = VINF_SUCCESS;
3992	uint16_t iNew = pPool->iFreeHead;
3993	if (iNew == NIL_PGMPOOL_IDX)
3994	{
3995	rc = pgmPoolMakeMoreFreePages(pPool, iUser);
3996	if (RT_FAILURE(rc))
3997	{
3998	if (rc != VERR_PGM_POOL_CLEARED)
3999	{
4000	Log(("pgmPoolAlloc: returns %Rrc (Free)\n", rc));
4001	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4002	return rc;
4003	}
4004	Log(("pgmPoolMakeMoreFreePages failed with %Rrc -> return VERR_PGM_POOL_FLUSHED\n", rc));
4005	rc = VERR_PGM_POOL_FLUSHED;
4006	}
4007	iNew = pPool->iFreeHead;
4008	AssertReleaseReturn(iNew != NIL_PGMPOOL_IDX, VERR_INTERNAL_ERROR);
4009	}
4010
4011	/* unlink the free head */
4012	PPGMPOOLPAGE pPage = &pPool->aPages[iNew];
4013	pPool->iFreeHead = pPage->iNext;
4014	pPage->iNext = NIL_PGMPOOL_IDX;
4015
4016	/*
4017	* Initialize it.
4018	*/
4019	pPool->cUsedPages++; /* physical handler registration / pgmPoolTrackFlushGCPhysPTsSlow requirement. */
4020	pPage->enmKind = enmKind;
4021	pPage->GCPhys = GCPhys;
4022	pPage->fSeenNonGlobal = false; /* Set this to 'true' to disable this feature. */
4023	pPage->fMonitored = false;
4024	pPage->fCached = false;
4025	pPage->fReusedFlushPending = false;
4026	pPage->fCR3Mix = false;
4027	#ifdef PGMPOOL_WITH_MONITORING
4028	pPage->cModifications = 0;
4029	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
4030	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
4031	#endif
4032	#ifdef PGMPOOL_WITH_USER_TRACKING
4033	pPage->cPresent = 0;
4034	pPage->iFirstPresent = ~0;
4035
4036	/*
4037	* Insert into the tracking and cache. If this fails, free the page.
4038	*/
4039	int rc3 = pgmPoolTrackInsert(pPool, pPage, GCPhys, iUser, iUserTable);
4040	if (RT_FAILURE(rc3))
4041	{
4042	if (rc3 != VERR_PGM_POOL_CLEARED)
4043	{
4044	pPool->cUsedPages--;
4045	pPage->enmKind = PGMPOOLKIND_FREE;
4046	pPage->GCPhys = NIL_RTGCPHYS;
4047	pPage->iNext = pPool->iFreeHead;
4048	pPool->iFreeHead = pPage->idx;
4049	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4050	Log(("pgmPoolAlloc: returns %Rrc (Insert)\n", rc3));
4051	return rc3;
4052	}
4053	Log(("pgmPoolTrackInsert failed with %Rrc -> return VERR_PGM_POOL_FLUSHED\n", rc3));
4054	rc = VERR_PGM_POOL_FLUSHED;
4055	}
4056	#endif /* PGMPOOL_WITH_USER_TRACKING */
4057
4058	/*
4059	* Commit the allocation, clear the page and return.
4060	*/
4061	#ifdef VBOX_WITH_STATISTICS
4062	if (pPool->cUsedPages > pPool->cUsedPagesHigh)
4063	pPool->cUsedPagesHigh = pPool->cUsedPages;
4064	#endif
4065
4066	if (!pPage->fZeroed)
4067	{
4068	STAM_PROFILE_START(&pPool->StatZeroPage, z);
4069	void *pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
4070	ASMMemZeroPage(pv);
4071	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
4072	}
4073
4074	*ppPage = pPage;
4075	LogFlow(("pgmPoolAlloc: returns %Rrc *ppPage=%p:{.Key=%RHp, .idx=%d, .fCached=%RTbool, .fMonitored=%RTbool}\n",
4076	rc, pPage, pPage->Core.Key, pPage->idx, pPage->fCached, pPage->fMonitored));
4077	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
4078	return rc;
4079	}
4080
4081
4082	/**
4083	* Frees a usage of a pool page.
4084	*
4085	* @param pVM The VM handle.
4086	* @param HCPhys The HC physical address of the shadow page.
4087	* @param iUser The shadow page pool index of the user table.
4088	* @param iUserTable The index into the user table (shadowed).
4089	*/
4090	void pgmPoolFree(PVM pVM, RTHCPHYS HCPhys, uint16_t iUser, uint32_t iUserTable)
4091	{
4092	LogFlow(("pgmPoolFree: HCPhys=%RHp iUser=%#x iUserTable=%#x\n", HCPhys, iUser, iUserTable));
4093	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4094	pgmPoolFreeByPage(pPool, pgmPoolGetPage(pPool, HCPhys), iUser, iUserTable);
4095	}
4096
4097
4098	/**
4099	* Gets a in-use page in the pool by it's physical address.
4100	*
4101	* @returns Pointer to the page.
4102	* @param pVM The VM handle.
4103	* @param HCPhys The HC physical address of the shadow page.
4104	* @remark This function will NEVER return NULL. It will assert if HCPhys is invalid.
4105	*/
4106	PPGMPOOLPAGE pgmPoolGetPageByHCPhys(PVM pVM, RTHCPHYS HCPhys)
4107	{
4108	/** @todo profile this! */
4109	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4110	PPGMPOOLPAGE pPage = pgmPoolGetPage(pPool, HCPhys);
4111	Log3(("pgmPoolGetPageByHCPhys: HCPhys=%RHp -> %p:{.idx=%d .GCPhys=%RGp .enmKind=%d}\n",
4112	HCPhys, pPage, pPage->idx, pPage->GCPhys, pPage->enmKind));
4113	return pPage;
4114	}
4115
4116
4117	/**
4118	* Flushes the entire cache.
4119	*
4120	* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
4121	* and execute this CR3 flush.
4122	*
4123	* @param pPool The pool.
4124	*/
4125	void pgmPoolFlushAll(PVM pVM)
4126	{
4127	LogFlow(("pgmPoolFlushAll:\n"));
4128	pgmPoolFlushAllInt(pVM->pgm.s.CTX_SUFF(pPool));
4129	}
4130

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllPool.cpp@ 14082

Download in other formats: