PGMAllPool.cpp@ 99208

Last change on this file since 99208 was 99208, checked in by vboxsync, 20 months ago
Disassembler,VMM,Runtime: Get rid of deprecated DISCPUSTATE types (preparation for architecture specific separation in order to support ARMv8), bugref:10394
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 224.6 KB

Line
1	/* $Id: PGMAllPool.cpp 99208 2023-03-29 14:13:56Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
4	*/
5
6	/*
7	* Copyright (C) 2006-2023 Oracle and/or its affiliates.
8	*
9	* This file is part of VirtualBox base platform packages, as
10	* available from https://www.virtualbox.org.
11	*
12	* This program is free software; you can redistribute it and/or
13	* modify it under the terms of the GNU General Public License
14	* as published by the Free Software Foundation, in version 3 of the
15	* License.
16	*
17	* This program is distributed in the hope that it will be useful, but
18	* WITHOUT ANY WARRANTY; without even the implied warranty of
19	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20	* General Public License for more details.
21	*
22	* You should have received a copy of the GNU General Public License
23	* along with this program; if not, see <https://www.gnu.org/licenses>.
24	*
25	* SPDX-License-Identifier: GPL-3.0-only
26	*/
27
28
29	/*********************************************************************************************************************************
30	* Header Files *
31	*********************************************************************************************************************************/
32	#define LOG_GROUP LOG_GROUP_PGM_POOL
33	#define VBOX_WITHOUT_PAGING_BIT_FIELDS /* 64-bit bitfields are just asking for trouble. See @bugref{9841} and others. */
34	#include <VBox/vmm/pgm.h>
35	#include <VBox/vmm/mm.h>
36	#include <VBox/vmm/em.h>
37	#include <VBox/vmm/cpum.h>
38	#include "PGMInternal.h"
39	#include <VBox/vmm/vmcc.h>
40	#include "PGMInline.h"
41	#include <VBox/disopcode.h>
42	#include <VBox/vmm/hm_vmx.h>
43
44	#include <VBox/log.h>
45	#include <VBox/err.h>
46	#include <iprt/asm.h>
47	#include <iprt/string.h>
48
49
50	/*********************************************************************************************************************************
51	* Internal Functions *
52	*********************************************************************************************************************************/
53	RT_C_DECLS_BEGIN
54	#if 0 /* unused */
55	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind);
56	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind);
57	#endif /* unused */
58	static void pgmPoolTrackClearPageUsers(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
59	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
60	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable);
61	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
62	#if defined(LOG_ENABLED) \|\| defined(VBOX_STRICT)
63	static const char *pgmPoolPoolKindToStr(uint8_t enmKind);
64	#endif
65	#if 0 /defined(VBOX_STRICT) && defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT)/
66	static void pgmPoolTrackCheckPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PTPAE pGstPT);
67	#endif
68
69	int pgmPoolTrackFlushGCPhysPTsSlow(PVMCC pVM, PPGMPAGE pPhysPage);
70	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVMCC pVM, uint16_t *piPhysExt);
71	void pgmPoolTrackPhysExtFree(PVMCC pVM, uint16_t iPhysExt);
72	void pgmPoolTrackPhysExtFreeList(PVMCC pVM, uint16_t iPhysExt);
73
74	RT_C_DECLS_END
75
76
77	#if 0 /* unused */
78	/**
79	* Checks if the specified page pool kind is for a 4MB or 2MB guest page.
80	*
81	* @returns true if it's the shadow of a 4MB or 2MB guest page, otherwise false.
82	* @param enmKind The page kind.
83	*/
84	DECLINLINE(bool) pgmPoolIsBigPage(PGMPOOLKIND enmKind)
85	{
86	switch (enmKind)
87	{
88	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
89	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
90	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
91	return true;
92	default:
93	return false;
94	}
95	}
96	#endif /* unused */
97
98
99	/**
100	* Flushes a chain of pages sharing the same access monitor.
101	*
102	* @param pPool The pool.
103	* @param pPage A page in the chain.
104	*/
105	void pgmPoolMonitorChainFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
106	{
107	LogFlow(("pgmPoolMonitorChainFlush: Flush page %RGp type=%d\n", pPage->GCPhys, pPage->enmKind));
108
109	/*
110	* Find the list head.
111	*/
112	uint16_t idx = pPage->idx;
113	if (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
114	{
115	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
116	{
117	idx = pPage->iMonitoredPrev;
118	Assert(idx != pPage->idx);
119	pPage = &pPool->aPages[idx];
120	}
121	}
122
123	/*
124	* Iterate the list flushing each shadow page.
125	*/
126	for (;;)
127	{
128	idx = pPage->iMonitoredNext;
129	Assert(idx != pPage->idx);
130	if (pPage->idx >= PGMPOOL_IDX_FIRST)
131	{
132	int rc2 = pgmPoolFlushPage(pPool, pPage);
133	AssertRC(rc2);
134	}
135	/* next */
136	if (idx == NIL_PGMPOOL_IDX)
137	break;
138	pPage = &pPool->aPages[idx];
139	}
140	}
141
142
143	/**
144	* Wrapper for getting the current context pointer to the entry being modified.
145	*
146	* @returns VBox status code suitable for scheduling.
147	* @param pVM The cross context VM structure.
148	* @param pvDst Destination address
149	* @param pvSrc Pointer to the mapping of @a GCPhysSrc or NULL depending
150	* on the context (e.g. \#PF in R0 & RC).
151	* @param GCPhysSrc The source guest physical address.
152	* @param cb Size of data to read
153	*/
154	DECLINLINE(int) pgmPoolPhysSimpleReadGCPhys(PVMCC pVM, void pvDst, void const pvSrc, RTGCPHYS GCPhysSrc, size_t cb)
155	{
156	#if defined(IN_RING3)
157	NOREF(pVM); NOREF(GCPhysSrc);
158	memcpy(pvDst, (RTHCPTR)((uintptr_t)pvSrc & ~(RTHCUINTPTR)(cb - 1)), cb);
159	return VINF_SUCCESS;
160	#else
161	/** @todo in RC we could attempt to use the virtual address, although this can cause many faults (PAE Windows XP guest). */
162	NOREF(pvSrc);
163	return PGMPhysSimpleReadGCPhys(pVM, pvDst, GCPhysSrc & ~(RTGCPHYS)(cb - 1), cb);
164	#endif
165	}
166
167
168	/**
169	* Process shadow entries before they are changed by the guest.
170	*
171	* For PT entries we will clear them. For PD entries, we'll simply check
172	* for mapping conflicts and set the SyncCR3 FF if found.
173	*
174	* @param pVCpu The cross context virtual CPU structure.
175	* @param pPool The pool.
176	* @param pPage The head page.
177	* @param GCPhysFault The guest physical fault address.
178	* @param pvAddress Pointer to the mapping of @a GCPhysFault or NULL
179	* depending on the context (e.g. \#PF in R0 & RC).
180	* @param cbWrite Write size; might be zero if the caller knows we're not crossing entry boundaries
181	*/
182	static void pgmPoolMonitorChainChanging(PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhysFault,
183	void const *pvAddress, unsigned cbWrite)
184	{
185	AssertMsg(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX, ("%u (idx=%u)\n", pPage->iMonitoredPrev, pPage->idx));
186	const unsigned off = GCPhysFault & GUEST_PAGE_OFFSET_MASK;
187	PVMCC pVM = pPool->CTX_SUFF(pVM);
188	NOREF(pVCpu);
189
190	LogFlow(("pgmPoolMonitorChainChanging: %RGv phys=%RGp cbWrite=%d\n",
191	(RTGCPTR)(CTXTYPE(RTGCPTR, uintptr_t, RTGCPTR))(uintptr_t)pvAddress, GCPhysFault, cbWrite));
192
193	if (PGMPOOL_PAGE_IS_NESTED(pPage))
194	Log7Func(("%RGv phys=%RGp cbWrite=%d\n", (RTGCPTR)(CTXTYPE(RTGCPTR, uintptr_t, RTGCPTR))(uintptr_t)pvAddress, GCPhysFault, cbWrite));
195
196	for (;;)
197	{
198	union
199	{
200	void *pv;
201	PX86PT pPT;
202	PPGMSHWPTPAE pPTPae;
203	PX86PD pPD;
204	PX86PDPAE pPDPae;
205	PX86PDPT pPDPT;
206	PX86PML4 pPML4;
207	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
208	PEPTPDPT pEptPdpt;
209	PEPTPD pEptPd;
210	PEPTPT pEptPt;
211	#endif
212	} uShw;
213
214	LogFlow(("pgmPoolMonitorChainChanging: page idx=%d phys=%RGp (next=%d) kind=%s write=%#x\n",
215	pPage->idx, pPage->GCPhys, pPage->iMonitoredNext, pgmPoolPoolKindToStr(pPage->enmKind), cbWrite));
216
217	uShw.pv = NULL;
218	switch (pPage->enmKind)
219	{
220	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
221	{
222	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPT));
223	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
224	const unsigned iShw = off / sizeof(X86PTE);
225	LogFlow(("PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT iShw=%x\n", iShw));
226	X86PGUINT const uPde = uShw.pPT->a[iShw].u;
227	if (uPde & X86_PTE_P)
228	{
229	X86PTE GstPte;
230	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
231	AssertRC(rc);
232	Log4(("pgmPoolMonitorChainChanging 32_32: deref %016RX64 GCPhys %08RX32\n", uPde & X86_PTE_PG_MASK, GstPte.u & X86_PTE_PG_MASK));
233	pgmPoolTracDerefGCPhysHint(pPool, pPage, uPde & X86_PTE_PG_MASK, GstPte.u & X86_PTE_PG_MASK, iShw);
234	ASMAtomicWriteU32(&uShw.pPT->a[iShw].u, 0);
235	}
236	break;
237	}
238
239	/* page/2 sized */
240	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
241	{
242	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPT));
243	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
244	if (!((off ^ pPage->GCPhys) & (PAGE_SIZE / 2)))
245	{
246	const unsigned iShw = (off / sizeof(X86PTE)) & (X86_PG_PAE_ENTRIES - 1);
247	LogFlow(("PGMPOOLKIND_PAE_PT_FOR_32BIT_PT iShw=%x\n", iShw));
248	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw]))
249	{
250	X86PTE GstPte;
251	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
252	AssertRC(rc);
253
254	Log4(("pgmPoolMonitorChainChanging pae_32: deref %016RX64 GCPhys %08RX32\n", uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PG_MASK));
255	pgmPoolTracDerefGCPhysHint(pPool, pPage,
256	PGMSHWPTEPAE_GET_HCPHYS(uShw.pPTPae->a[iShw]),
257	GstPte.u & X86_PTE_PG_MASK,
258	iShw);
259	PGMSHWPTEPAE_ATOMIC_SET(uShw.pPTPae->a[iShw], 0);
260	}
261	}
262	break;
263	}
264
265	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
266	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
267	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
268	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
269	{
270	unsigned iGst = off / sizeof(X86PDE);
271	unsigned iShwPdpt = iGst / 256;
272	unsigned iShw = (iGst % 256) * 2;
273	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
274
275	LogFlow(("pgmPoolMonitorChainChanging PAE for 32 bits: iGst=%x iShw=%x idx = %d page idx=%d\n", iGst, iShw, iShwPdpt, pPage->enmKind - PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD));
276	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPD));
277	if (iShwPdpt == pPage->enmKind - (unsigned)PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD)
278	{
279	for (unsigned i = 0; i < 2; i++)
280	{
281	X86PGPAEUINT const uPde = uShw.pPDPae->a[iShw + i].u;
282	if (uPde & X86_PDE_P)
283	{
284	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw + i, uPde));
285	pgmPoolFree(pVM, uPde & X86_PDE_PAE_PG_MASK, pPage->idx, iShw + i);
286	ASMAtomicWriteU64(&uShw.pPDPae->a[iShw + i].u, 0);
287	}
288
289	/* paranoia / a bit assumptive. */
290	if ( (off & 3)
291	&& (off & 3) + cbWrite > 4)
292	{
293	const unsigned iShw2 = iShw + 2 + i;
294	if (iShw2 < RT_ELEMENTS(uShw.pPDPae->a))
295	{
296	X86PGPAEUINT const uPde2 = uShw.pPDPae->a[iShw2].u;
297	if (uPde2 & X86_PDE_P)
298	{
299	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uPde2));
300	pgmPoolFree(pVM, uPde2 & X86_PDE_PAE_PG_MASK, pPage->idx, iShw2);
301	ASMAtomicWriteU64(&uShw.pPDPae->a[iShw2].u, 0);
302	}
303	}
304	}
305	}
306	}
307	break;
308	}
309
310	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
311	{
312	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
313	const unsigned iShw = off / sizeof(X86PTEPAE);
314	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPT));
315	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw]))
316	{
317	X86PTEPAE GstPte;
318	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
319	AssertRC(rc);
320
321	Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", PGMSHWPTEPAE_GET_HCPHYS(uShw.pPTPae->a[iShw]), GstPte.u & X86_PTE_PAE_PG_MASK));
322	pgmPoolTracDerefGCPhysHint(pPool, pPage,
323	PGMSHWPTEPAE_GET_HCPHYS(uShw.pPTPae->a[iShw]),
324	GstPte.u & X86_PTE_PAE_PG_MASK,
325	iShw);
326	PGMSHWPTEPAE_ATOMIC_SET(uShw.pPTPae->a[iShw], 0);
327	}
328
329	/* paranoia / a bit assumptive. */
330	if ( (off & 7)
331	&& (off & 7) + cbWrite > sizeof(X86PTEPAE))
332	{
333	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTEPAE);
334	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPTPae->a));
335
336	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw2]))
337	{
338	X86PTEPAE GstPte;
339	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte,
340	pvAddress ? (uint8_t const *)pvAddress + sizeof(GstPte) : NULL,
341	GCPhysFault + sizeof(GstPte), sizeof(GstPte));
342	AssertRC(rc);
343	Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", PGMSHWPTEPAE_GET_HCPHYS(uShw.pPTPae->a[iShw2]), GstPte.u & X86_PTE_PAE_PG_MASK));
344	pgmPoolTracDerefGCPhysHint(pPool, pPage,
345	PGMSHWPTEPAE_GET_HCPHYS(uShw.pPTPae->a[iShw2]),
346	GstPte.u & X86_PTE_PAE_PG_MASK,
347	iShw2);
348	PGMSHWPTEPAE_ATOMIC_SET(uShw.pPTPae->a[iShw2], 0);
349	}
350	}
351	break;
352	}
353
354	case PGMPOOLKIND_32BIT_PD:
355	{
356	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
357	const unsigned iShw = off / sizeof(X86PTE); // ASSUMING 32-bit guest paging!
358
359	LogFlow(("pgmPoolMonitorChainChanging: PGMPOOLKIND_32BIT_PD %x\n", iShw));
360	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPD));
361	X86PGUINT const uPde = uShw.pPD->a[iShw].u;
362	if (uPde & X86_PDE_P)
363	{
364	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uPde));
365	pgmPoolFree(pVM, uPde & X86_PDE_PG_MASK, pPage->idx, iShw);
366	ASMAtomicWriteU32(&uShw.pPD->a[iShw].u, 0);
367	}
368
369	/* paranoia / a bit assumptive. */
370	if ( (off & 3)
371	&& (off & 3) + cbWrite > sizeof(X86PTE))
372	{
373	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTE);
374	if ( iShw2 != iShw
375	&& iShw2 < RT_ELEMENTS(uShw.pPD->a))
376	{
377	X86PGUINT const uPde2 = uShw.pPD->a[iShw2].u;
378	if (uPde2 & X86_PDE_P)
379	{
380	LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uPde2));
381	pgmPoolFree(pVM, uPde2 & X86_PDE_PG_MASK, pPage->idx, iShw2);
382	ASMAtomicWriteU32(&uShw.pPD->a[iShw2].u, 0);
383	}
384	}
385	}
386	#if 0 /* useful when running PGMAssertCR3(), a bit too troublesome for general use (TLBs). - not working any longer... */
387	if ( uShw.pPD->a[iShw].n.u1Present
388	&& !VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
389	{
390	LogFlow(("pgmPoolMonitorChainChanging: iShw=%#x: %RX32 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
391	pgmPoolFree(pVM, uShw.pPD->a[iShw].u & X86_PDE_PG_MASK, pPage->idx, iShw);
392	ASMAtomicWriteU32(&uShw.pPD->a[iShw].u, 0);
393	}
394	#endif
395	break;
396	}
397
398	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
399	{
400	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
401	const unsigned iShw = off / sizeof(X86PDEPAE);
402	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPD));
403
404	/*
405	* Causes trouble when the guest uses a PDE to refer to the whole page table level
406	* structure. (Invalidate here; faults later on when it tries to change the page
407	* table entries -> recheck; probably only applies to the RC case.)
408	*/
409	X86PGPAEUINT const uPde = uShw.pPDPae->a[iShw].u;
410	if (uPde & X86_PDE_P)
411	{
412	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uPde));
413	pgmPoolFree(pVM, uPde & X86_PDE_PAE_PG_MASK, pPage->idx, iShw);
414	ASMAtomicWriteU64(&uShw.pPDPae->a[iShw].u, 0);
415	}
416
417	/* paranoia / a bit assumptive. */
418	if ( (off & 7)
419	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
420	{
421	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
422	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
423
424	X86PGPAEUINT const uPde2 = uShw.pPDPae->a[iShw2].u;
425	if (uPde2 & X86_PDE_P)
426	{
427	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uPde2));
428	pgmPoolFree(pVM, uPde2 & X86_PDE_PAE_PG_MASK, pPage->idx, iShw2);
429	ASMAtomicWriteU64(&uShw.pPDPae->a[iShw2].u, 0);
430	}
431	}
432	break;
433	}
434
435	case PGMPOOLKIND_PAE_PDPT:
436	{
437	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPDPT));
438	/*
439	* Hopefully this doesn't happen very often:
440	* - touching unused parts of the page
441	* - messing with the bits of pd pointers without changing the physical address
442	*/
443	/* PDPT roots are not page aligned; 32 byte only! */
444	const unsigned offPdpt = GCPhysFault - pPage->GCPhys;
445
446	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
447	const unsigned iShw = offPdpt / sizeof(X86PDPE);
448	if (iShw < X86_PG_PAE_PDPE_ENTRIES) /* don't use RT_ELEMENTS(uShw.pPDPT->a), because that's for long mode only */
449	{
450	X86PGPAEUINT const uPdpe = uShw.pPDPT->a[iShw].u;
451	if (uPdpe & X86_PDPE_P)
452	{
453	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
454	pgmPoolFree(pVM, uPdpe & X86_PDPE_PG_MASK, pPage->idx, iShw);
455	ASMAtomicWriteU64(&uShw.pPDPT->a[iShw].u, 0);
456	}
457
458	/* paranoia / a bit assumptive. */
459	if ( (offPdpt & 7)
460	&& (offPdpt & 7) + cbWrite > sizeof(X86PDPE))
461	{
462	const unsigned iShw2 = (offPdpt + cbWrite - 1) / sizeof(X86PDPE);
463	if ( iShw2 != iShw
464	&& iShw2 < X86_PG_PAE_PDPE_ENTRIES)
465	{
466	X86PGPAEUINT const uPdpe2 = uShw.pPDPT->a[iShw2].u;
467	if (uPdpe2 & X86_PDPE_P)
468	{
469	LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
470	pgmPoolFree(pVM, uPdpe2 & X86_PDPE_PG_MASK, pPage->idx, iShw2);
471	ASMAtomicWriteU64(&uShw.pPDPT->a[iShw2].u, 0);
472	}
473	}
474	}
475	}
476	break;
477	}
478
479	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
480	{
481	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPD));
482	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
483	const unsigned iShw = off / sizeof(X86PDEPAE);
484	X86PGPAEUINT const uPde = uShw.pPDPae->a[iShw].u;
485	if (uPde & X86_PDE_P)
486	{
487	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uPde));
488	pgmPoolFree(pVM, uPde & X86_PDE_PAE_PG_MASK, pPage->idx, iShw);
489	ASMAtomicWriteU64(&uShw.pPDPae->a[iShw].u, 0);
490	}
491
492	/* paranoia / a bit assumptive. */
493	if ( (off & 7)
494	&& (off & 7) + cbWrite > sizeof(X86PDEPAE))
495	{
496	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
497	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
498	X86PGPAEUINT const uPde2 = uShw.pPDPae->a[iShw2].u;
499	if (uPde2 & X86_PDE_P)
500	{
501	LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uPde2));
502	pgmPoolFree(pVM, uPde2 & X86_PDE_PAE_PG_MASK, pPage->idx, iShw2);
503	ASMAtomicWriteU64(&uShw.pPDPae->a[iShw2].u, 0);
504	}
505	}
506	break;
507	}
508
509	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
510	{
511	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPDPT));
512	/*
513	* Hopefully this doesn't happen very often:
514	* - messing with the bits of pd pointers without changing the physical address
515	*/
516	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
517	const unsigned iShw = off / sizeof(X86PDPE);
518	X86PGPAEUINT const uPdpe = uShw.pPDPT->a[iShw].u;
519	if (uPdpe & X86_PDPE_P)
520	{
521	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uPdpe));
522	pgmPoolFree(pVM, uPdpe & X86_PDPE_PG_MASK, pPage->idx, iShw);
523	ASMAtomicWriteU64(&uShw.pPDPT->a[iShw].u, 0);
524	}
525	/* paranoia / a bit assumptive. */
526	if ( (off & 7)
527	&& (off & 7) + cbWrite > sizeof(X86PDPE))
528	{
529	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDPE);
530	X86PGPAEUINT const uPdpe2 = uShw.pPDPT->a[iShw2].u;
531	if (uPdpe2 & X86_PDPE_P)
532	{
533	LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uPdpe2));
534	pgmPoolFree(pVM, uPdpe2 & X86_PDPE_PG_MASK, pPage->idx, iShw2);
535	ASMAtomicWriteU64(&uShw.pPDPT->a[iShw2].u, 0);
536	}
537	}
538	break;
539	}
540
541	case PGMPOOLKIND_64BIT_PML4:
542	{
543	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPML4));
544	/*
545	* Hopefully this doesn't happen very often:
546	* - messing with the bits of pd pointers without changing the physical address
547	*/
548	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
549	const unsigned iShw = off / sizeof(X86PDPE);
550	X86PGPAEUINT const uPml4e = uShw.pPML4->a[iShw].u;
551	if (uPml4e & X86_PML4E_P)
552	{
553	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw=%#x: %RX64 -> freeing it!\n", iShw, uPml4e));
554	pgmPoolFree(pVM, uPml4e & X86_PML4E_PG_MASK, pPage->idx, iShw);
555	ASMAtomicWriteU64(&uShw.pPML4->a[iShw].u, 0);
556	}
557	/* paranoia / a bit assumptive. */
558	if ( (off & 7)
559	&& (off & 7) + cbWrite > sizeof(X86PDPE))
560	{
561	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PML4E);
562	X86PGPAEUINT const uPml4e2 = uShw.pPML4->a[iShw2].u;
563	if (uPml4e2 & X86_PML4E_P)
564	{
565	LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uPml4e2));
566	pgmPoolFree(pVM, uPml4e2 & X86_PML4E_PG_MASK, pPage->idx, iShw2);
567	ASMAtomicWriteU64(&uShw.pPML4->a[iShw2].u, 0);
568	}
569	}
570	break;
571	}
572
573	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
574	case PGMPOOLKIND_EPT_PML4_FOR_EPT_PML4:
575	{
576	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
577	const unsigned iShw = off / sizeof(EPTPML4E);
578	X86PGPAEUINT const uPml4e = uShw.pPML4->a[iShw].u;
579	if (uPml4e & EPT_PRESENT_MASK)
580	{
581	Log7Func(("PML4 iShw=%#x: %RX64 (%RGp) -> freeing it!\n", iShw, uPml4e, pPage->GCPhys));
582	pgmPoolFree(pVM, uPml4e & X86_PML4E_PG_MASK, pPage->idx, iShw);
583	ASMAtomicWriteU64(&uShw.pPML4->a[iShw].u, 0);
584	}
585
586	/* paranoia / a bit assumptive. */
587	if ( (off & 7)
588	&& (off & 7) + cbWrite > sizeof(X86PML4E))
589	{
590	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PML4E);
591	X86PGPAEUINT const uPml4e2 = uShw.pPML4->a[iShw2].u;
592	if (uPml4e2 & EPT_PRESENT_MASK)
593	{
594	Log7Func(("PML4 iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uPml4e2));
595	pgmPoolFree(pVM, uPml4e2 & X86_PML4E_PG_MASK, pPage->idx, iShw2);
596	ASMAtomicWriteU64(&uShw.pPML4->a[iShw2].u, 0);
597	}
598	}
599	break;
600	}
601
602	case PGMPOOLKIND_EPT_PDPT_FOR_EPT_PDPT:
603	{
604	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
605	const unsigned iShw = off / sizeof(EPTPDPTE);
606	X86PGPAEUINT const uPdpte = uShw.pEptPdpt->a[iShw].u;
607	if (uPdpte & EPT_PRESENT_MASK)
608	{
609	Log7Func(("EPT PDPT iShw=%#x: %RX64 (%RGp) -> freeing it!\n", iShw, uPdpte, pPage->GCPhys));
610	pgmPoolFree(pVM, uPdpte & EPT_PDPTE_PG_MASK, pPage->idx, iShw);
611	ASMAtomicWriteU64(&uShw.pEptPdpt->a[iShw].u, 0);
612	}
613
614	/* paranoia / a bit assumptive. */
615	if ( (off & 7)
616	&& (off & 7) + cbWrite > sizeof(EPTPDPTE))
617	{
618	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(EPTPDPTE);
619	X86PGPAEUINT const uPdpte2 = uShw.pEptPdpt->a[iShw2].u;
620	if (uPdpte2 & EPT_PRESENT_MASK)
621	{
622	Log7Func(("EPT PDPT iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uPdpte2));
623	pgmPoolFree(pVM, uPdpte2 & EPT_PDPTE_PG_MASK, pPage->idx, iShw2);
624	ASMAtomicWriteU64(&uShw.pEptPdpt->a[iShw2].u, 0);
625	}
626	}
627	break;
628	}
629
630	case PGMPOOLKIND_EPT_PD_FOR_EPT_PD:
631	{
632	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
633	const unsigned iShw = off / sizeof(EPTPDE);
634	X86PGPAEUINT const uPde = uShw.pEptPd->a[iShw].u;
635	if (uPde & EPT_PRESENT_MASK)
636	{
637	Log7Func(("EPT PD iShw=%#x: %RX64 (%RGp) -> freeing it!\n", iShw, uPde, pPage->GCPhys));
638	pgmPoolFree(pVM, uPde & EPT_PDE_PG_MASK, pPage->idx, iShw);
639	ASMAtomicWriteU64(&uShw.pEptPd->a[iShw].u, 0);
640	}
641
642	/* paranoia / a bit assumptive. */
643	if ( (off & 7)
644	&& (off & 7) + cbWrite > sizeof(EPTPDE))
645	{
646	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(EPTPDE);
647	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pEptPd->a));
648	X86PGPAEUINT const uPde2 = uShw.pEptPd->a[iShw2].u;
649	if (uPde2 & EPT_PRESENT_MASK)
650	{
651	Log7Func(("EPT PD (2): iShw2=%#x: %RX64 (%RGp) -> freeing it!\n", iShw2, uPde2, pPage->GCPhys));
652	pgmPoolFree(pVM, uPde2 & EPT_PDE_PG_MASK, pPage->idx, iShw2);
653	ASMAtomicWriteU64(&uShw.pEptPd->a[iShw2].u, 0);
654	}
655	}
656	break;
657	}
658
659	case PGMPOOLKIND_EPT_PT_FOR_EPT_PT:
660	{
661	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
662	const unsigned iShw = off / sizeof(EPTPTE);
663	X86PGPAEUINT const uPte = uShw.pEptPt->a[iShw].u;
664	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,FaultPT));
665	if (uPte & EPT_PRESENT_MASK)
666	{
667	EPTPTE GstPte;
668	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
669	AssertRC(rc);
670
671	Log7Func(("EPT PT: iShw=%#x %RX64 (%RGp)\n", iShw, uPte, pPage->GCPhys));
672	pgmPoolTracDerefGCPhysHint(pPool, pPage,
673	uShw.pEptPt->a[iShw].u & EPT_PTE_PG_MASK,
674	GstPte.u & EPT_PTE_PG_MASK,
675	iShw);
676	ASMAtomicWriteU64(&uShw.pEptPt->a[iShw].u, 0);
677	}
678
679	/* paranoia / a bit assumptive. */
680	if ( (off & 7)
681	&& (off & 7) + cbWrite > sizeof(EPTPTE))
682	{
683	const unsigned iShw2 = (off + cbWrite - 1) / sizeof(EPTPTE);
684	AssertBreak(iShw2 < RT_ELEMENTS(uShw.pEptPt->a));
685	X86PGPAEUINT const uPte2 = uShw.pEptPt->a[iShw2].u;
686	if (uPte2 & EPT_PRESENT_MASK)
687	{
688	EPTPTE GstPte;
689	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte,
690	pvAddress ? (uint8_t const *)pvAddress + sizeof(GstPte) : NULL,
691	GCPhysFault + sizeof(GstPte), sizeof(GstPte));
692	AssertRC(rc);
693	Log7Func(("EPT PT (2): iShw=%#x %RX64 (%RGp)\n", iShw2, uPte2, pPage->GCPhys));
694	pgmPoolTracDerefGCPhysHint(pPool, pPage,
695	uShw.pEptPt->a[iShw2].u & EPT_PTE_PG_MASK,
696	GstPte.u & EPT_PTE_PG_MASK,
697	iShw2);
698	ASMAtomicWriteU64(&uShw.pEptPt->a[iShw2].u, 0);
699	}
700	}
701	break;
702	}
703	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX_EPT */
704
705	default:
706	AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
707	}
708	PGM_DYNMAP_UNUSED_HINT_VM(pVM, uShw.pv);
709
710	/* next */
711	if (pPage->iMonitoredNext == NIL_PGMPOOL_IDX)
712	return;
713	pPage = &pPool->aPages[pPage->iMonitoredNext];
714	}
715	}
716
717	#ifndef IN_RING3
718
719	/**
720	* Checks if a access could be a fork operation in progress.
721	*
722	* Meaning, that the guest is setting up the parent process for Copy-On-Write.
723	*
724	* @returns true if it's likely that we're forking, otherwise false.
725	* @param pPool The pool.
726	* @param pDis The disassembled instruction.
727	* @param offFault The access offset.
728	*/
729	DECLINLINE(bool) pgmRZPoolMonitorIsForking(PPGMPOOL pPool, PDISSTATE pDis, unsigned offFault)
730	{
731	/*
732	* i386 linux is using btr to clear X86_PTE_RW.
733	* The functions involved are (2.6.16 source inspection):
734	* clear_bit
735	* ptep_set_wrprotect
736	* copy_one_pte
737	* copy_pte_range
738	* copy_pmd_range
739	* copy_pud_range
740	* copy_page_range
741	* dup_mmap
742	* dup_mm
743	* copy_mm
744	* copy_process
745	* do_fork
746	*/
747	if ( pDis->pCurInstr->uOpcode == OP_BTR
748	&& !(offFault & 4)
749	/** @todo Validate that the bit index is X86_PTE_RW. */
750	)
751	{
752	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitorPf,Fork)); RT_NOREF_PV(pPool);
753	return true;
754	}
755	return false;
756	}
757
758
759	/**
760	* Determine whether the page is likely to have been reused.
761	*
762	* @returns true if we consider the page as being reused for a different purpose.
763	* @returns false if we consider it to still be a paging page.
764	* @param pVM The cross context VM structure.
765	* @param pVCpu The cross context virtual CPU structure.
766	* @param pCtx Pointer to the register context for the CPU.
767	* @param pDis The disassembly info for the faulting instruction.
768	* @param pvFault The fault address.
769	* @param pPage The pool page being accessed.
770	*
771	* @remark The REP prefix check is left to the caller because of STOSD/W.
772	*/
773	DECLINLINE(bool) pgmRZPoolMonitorIsReused(PVMCC pVM, PVMCPUCC pVCpu, PCPUMCTX pCtx, PDISSTATE pDis, RTGCPTR pvFault,
774	PPGMPOOLPAGE pPage)
775	{
776	/* Locked (CR3, PDPTR*4) should not be reusable. Considering them as
777	such may cause loops booting tst-ubuntu-15_10-64-efi, ++. */
778	if (pPage->cLocked)
779	{
780	Log2(("pgmRZPoolMonitorIsReused: %RGv (%p) can't have been resued, because it's locked!\n", pvFault, pPage));
781	return false;
782	}
783
784	/** @todo could make this general, faulting close to rsp should be a safe reuse heuristic. */
785	if ( HMHasPendingIrq(pVM)
786	&& pCtx->rsp - pvFault < 32)
787	{
788	/* Fault caused by stack writes while trying to inject an interrupt event. */
789	Log(("pgmRZPoolMonitorIsReused: reused %RGv for interrupt stack (rsp=%RGv).\n", pvFault, pCtx->rsp));
790	return true;
791	}
792
793	LogFlow(("Reused instr %RGv %d at %RGv param1.fUse=%llx param1.reg=%d\n", pCtx->rip, pDis->pCurInstr->uOpcode, pvFault, pDis->Param1.fUse, pDis->Param1.Base.idxGenReg));
794
795	/* Non-supervisor mode write means it's used for something else. */
796	if (CPUMGetGuestCPL(pVCpu) == 3)
797	return true;
798
799	switch (pDis->pCurInstr->uOpcode)
800	{
801	/* call implies the actual push of the return address faulted */
802	case OP_CALL:
803	Log4(("pgmRZPoolMonitorIsReused: CALL\n"));
804	return true;
805	case OP_PUSH:
806	Log4(("pgmRZPoolMonitorIsReused: PUSH\n"));
807	return true;
808	case OP_PUSHF:
809	Log4(("pgmRZPoolMonitorIsReused: PUSHF\n"));
810	return true;
811	case OP_PUSHA:
812	Log4(("pgmRZPoolMonitorIsReused: PUSHA\n"));
813	return true;
814	case OP_FXSAVE:
815	Log4(("pgmRZPoolMonitorIsReused: FXSAVE\n"));
816	return true;
817	case OP_MOVNTI: /* solaris - block_zero_no_xmm */
818	Log4(("pgmRZPoolMonitorIsReused: MOVNTI\n"));
819	return true;
820	case OP_MOVNTDQ: /* solaris - hwblkclr & hwblkpagecopy */
821	Log4(("pgmRZPoolMonitorIsReused: MOVNTDQ\n"));
822	return true;
823	case OP_MOVSWD:
824	case OP_STOSWD:
825	if ( pDis->fPrefix == (DISPREFIX_REP\|DISPREFIX_REX)
826	&& pCtx->rcx >= 0x40
827	)
828	{
829	Assert(pDis->uCpuMode == DISCPUMODE_64BIT);
830
831	Log(("pgmRZPoolMonitorIsReused: OP_STOSQ\n"));
832	return true;
833	}
834	break;
835
836	default:
837	/*
838	* Anything having ESP on the left side means stack writes.
839	*/
840	if ( ( (pDis->Param1.fUse & DISUSE_REG_GEN32)
841	\|\| (pDis->Param1.fUse & DISUSE_REG_GEN64))
842	&& (pDis->Param1.Base.idxGenReg == DISGREG_ESP))
843	{
844	Log4(("pgmRZPoolMonitorIsReused: ESP\n"));
845	return true;
846	}
847	break;
848	}
849
850	/*
851	* Page table updates are very very unlikely to be crossing page boundraries,
852	* and we don't want to deal with that in pgmPoolMonitorChainChanging and such.
853	*/
854	uint32_t const cbWrite = DISGetParamSize(pDis, &pDis->Param1);
855	if ( (((uintptr_t)pvFault + cbWrite) >> X86_PAGE_SHIFT) != ((uintptr_t)pvFault >> X86_PAGE_SHIFT) )
856	{
857	Log4(("pgmRZPoolMonitorIsReused: cross page write\n"));
858	return true;
859	}
860
861	/*
862	* Nobody does an unaligned 8 byte write to a page table, right.
863	*/
864	if (cbWrite >= 8 && ((uintptr_t)pvFault & 7) != 0)
865	{
866	Log4(("pgmRZPoolMonitorIsReused: Unaligned 8+ byte write\n"));
867	return true;
868	}
869
870	return false;
871	}
872
873
874	/**
875	* Flushes the page being accessed.
876	*
877	* @returns VBox status code suitable for scheduling.
878	* @param pVM The cross context VM structure.
879	* @param pVCpu The cross context virtual CPU structure.
880	* @param pPool The pool.
881	* @param pPage The pool page (head).
882	* @param pDis The disassembly of the write instruction.
883	* @param pCtx Pointer to the register context for the CPU.
884	* @param GCPhysFault The fault address as guest physical address.
885	* @todo VBOXSTRICTRC
886	*/
887	static int pgmRZPoolAccessPfHandlerFlush(PVMCC pVM, PVMCPUCC pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISSTATE pDis,
888	PCPUMCTX pCtx, RTGCPHYS GCPhysFault)
889	{
890	NOREF(pVM); NOREF(GCPhysFault);
891
892	/*
893	* First, do the flushing.
894	*/
895	pgmPoolMonitorChainFlush(pPool, pPage);
896
897	/*
898	* Emulate the instruction (xp/w2k problem, requires pc/cr2/sp detection).
899	* Must do this in raw mode (!); XP boot will fail otherwise.
900	*/
901	int rc = VINF_SUCCESS;
902	VBOXSTRICTRC rc2 = EMInterpretInstructionDisasState(pVCpu, pDis, pCtx->rip);
903	if (rc2 == VINF_SUCCESS)
904	{ /* do nothing */ }
905	else if (rc2 == VINF_EM_RESCHEDULE)
906	{
907	rc = VBOXSTRICTRC_VAL(rc2);
908	# ifndef IN_RING3
909	VMCPU_FF_SET(pVCpu, VMCPU_FF_TO_R3);
910	# endif
911	}
912	else if (rc2 == VERR_EM_INTERPRETER)
913	{
914	rc = VINF_EM_RAW_EMULATE_INSTR;
915	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitorPf,EmulateInstr));
916	}
917	else if (RT_FAILURE_NP(rc2))
918	rc = VBOXSTRICTRC_VAL(rc2);
919	else
920	AssertMsgFailed(("%Rrc\n", VBOXSTRICTRC_VAL(rc2))); /* ASSUMES no complicated stuff here. */
921
922	LogFlow(("pgmRZPoolAccessPfHandlerFlush: returns %Rrc (flushed)\n", rc));
923	return rc;
924	}
925
926
927	/**
928	* Handles the STOSD write accesses.
929	*
930	* @returns VBox status code suitable for scheduling.
931	* @param pVM The cross context VM structure.
932	* @param pPool The pool.
933	* @param pPage The pool page (head).
934	* @param pDis The disassembly of the write instruction.
935	* @param pCtx Pointer to the register context for the CPU.
936	* @param GCPhysFault The fault address as guest physical address.
937	* @param pvFault The fault address.
938	*/
939	DECLINLINE(int) pgmRZPoolAccessPfHandlerSTOSD(PVMCC pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISSTATE pDis,
940	PCPUMCTX pCtx, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
941	{
942	unsigned uIncrement = pDis->Param1.cb;
943	NOREF(pVM);
944
945	Assert(pDis->uCpuMode == DISCPUMODE_32BIT \|\| pDis->uCpuMode == DISCPUMODE_64BIT);
946	Assert(pCtx->rcx <= 0x20);
947
948	# ifdef VBOX_STRICT
949	if (pDis->uOpMode == DISCPUMODE_32BIT)
950	Assert(uIncrement == 4);
951	else
952	Assert(uIncrement == 8);
953	# endif
954
955	Log3(("pgmRZPoolAccessPfHandlerSTOSD\n"));
956
957	/*
958	* Increment the modification counter and insert it into the list
959	* of modified pages the first time.
960	*/
961	if (!pPage->cModifications++)
962	pgmPoolMonitorModifiedInsert(pPool, pPage);
963
964	/*
965	* Execute REP STOSD.
966	*
967	* This ASSUMES that we're not invoked by Trap0e on in a out-of-sync
968	* write situation, meaning that it's safe to write here.
969	*/
970	PVMCPUCC pVCpu = VMMGetCpu(pPool->CTX_SUFF(pVM));
971	RTGCUINTPTR pu32 = (RTGCUINTPTR)pvFault;
972	while (pCtx->rcx)
973	{
974	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, NULL, uIncrement);
975	PGMPhysSimpleWriteGCPhys(pVM, GCPhysFault, &pCtx->rax, uIncrement);
976	pu32 += uIncrement;
977	GCPhysFault += uIncrement;
978	pCtx->rdi += uIncrement;
979	pCtx->rcx--;
980	}
981	pCtx->rip += pDis->cbInstr;
982
983	LogFlow(("pgmRZPoolAccessPfHandlerSTOSD: returns\n"));
984	return VINF_SUCCESS;
985	}
986
987
988	/**
989	* Handles the simple write accesses.
990	*
991	* @returns VBox status code suitable for scheduling.
992	* @param pVM The cross context VM structure.
993	* @param pVCpu The cross context virtual CPU structure.
994	* @param pPool The pool.
995	* @param pPage The pool page (head).
996	* @param pDis The disassembly of the write instruction.
997	* @param pCtx Pointer to the register context for the CPU.
998	* @param GCPhysFault The fault address as guest physical address.
999	* @param pfReused Reused state (in/out)
1000	*/
1001	DECLINLINE(int) pgmRZPoolAccessPfHandlerSimple(PVMCC pVM, PVMCPUCC pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISSTATE pDis,
1002	PCPUMCTX pCtx, RTGCPHYS GCPhysFault, bool *pfReused)
1003	{
1004	Log3(("pgmRZPoolAccessPfHandlerSimple\n"));
1005	NOREF(pVM);
1006	NOREF(pfReused); /* initialized by caller */
1007
1008	/*
1009	* Increment the modification counter and insert it into the list
1010	* of modified pages the first time.
1011	*/
1012	if (!pPage->cModifications++)
1013	pgmPoolMonitorModifiedInsert(pPool, pPage);
1014
1015	/*
1016	* Clear all the pages.
1017	*/
1018	uint32_t cbWrite = DISGetParamSize(pDis, &pDis->Param1);
1019	if (cbWrite <= 8)
1020	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, NULL, cbWrite);
1021	else if (cbWrite <= 16)
1022	{
1023	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, NULL, 8);
1024	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault + 8, NULL, cbWrite - 8);
1025	}
1026	else
1027	{
1028	Assert(cbWrite <= 32);
1029	for (uint32_t off = 0; off < cbWrite; off += 8)
1030	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault + off, NULL, RT_MIN(8, cbWrite - off));
1031	}
1032
1033	/*
1034	* Interpret the instruction.
1035	*/
1036	VBOXSTRICTRC rc = EMInterpretInstructionDisasState(pVCpu, pDis, pCtx->rip);
1037	if (RT_SUCCESS(rc))
1038	AssertMsg(rc == VINF_SUCCESS, ("%Rrc\n", VBOXSTRICTRC_VAL(rc))); /* ASSUMES no complicated stuff here. */
1039	else if (rc == VERR_EM_INTERPRETER)
1040	{
1041	LogFlow(("pgmRZPoolAccessPfHandlerSimple: Interpretation failed for %04x:%RGv - opcode=%d\n",
1042	pCtx->cs.Sel, (RTGCPTR)pCtx->rip, pDis->pCurInstr->uOpcode));
1043	rc = VINF_EM_RAW_EMULATE_INSTR;
1044	STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitorPf,EmulateInstr));
1045	}
1046
1047	# if 0 /* experimental code */
1048	if (rc == VINF_SUCCESS)
1049	{
1050	switch (pPage->enmKind)
1051	{
1052	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
1053	{
1054	X86PTEPAE GstPte;
1055	int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvFault, GCPhysFault, sizeof(GstPte));
1056	AssertRC(rc);
1057
1058	/* Check the new value written by the guest. If present and with a bogus physical address, then
1059	* it's fairly safe to assume the guest is reusing the PT.
1060	*/
1061	if (GstPte.n.u1Present)
1062	{
1063	RTHCPHYS HCPhys = -1;
1064	int rc = PGMPhysGCPhys2HCPhys(pVM, GstPte.u & X86_PTE_PAE_PG_MASK, &HCPhys);
1065	if (rc != VINF_SUCCESS)
1066	{
1067	*pfReused = true;
1068	STAM_COUNTER_INC(&pPool->StatForceFlushReused);
1069	}
1070	}
1071	break;
1072	}
1073	}
1074	}
1075	# endif
1076
1077	LogFlow(("pgmRZPoolAccessPfHandlerSimple: returns %Rrc\n", VBOXSTRICTRC_VAL(rc)));
1078	return VBOXSTRICTRC_VAL(rc);
1079	}
1080
1081
1082	/**
1083	* @callback_method_impl{FNPGMRZPHYSPFHANDLER,
1084	* \#PF access handler callback for page table pages.}
1085	*
1086	* @remarks The @a uUser argument is the index of the PGMPOOLPAGE.
1087	*/
1088	DECLCALLBACK(VBOXSTRICTRC) pgmRZPoolAccessPfHandler(PVMCC pVM, PVMCPUCC pVCpu, RTGCUINT uErrorCode, PCPUMCTX pCtx,
1089	RTGCPTR pvFault, RTGCPHYS GCPhysFault, uint64_t uUser)
1090	{
1091	STAM_PROFILE_START(&pVM->pgm.s.CTX_SUFF(pPool)->StatMonitorRZ, a);
1092	PPGMPOOL const pPool = pVM->pgm.s.CTX_SUFF(pPool);
1093	AssertReturn(uUser < pPool->cCurPages, VERR_PGM_POOL_IPE);
1094	PPGMPOOLPAGE const pPage = &pPool->aPages[uUser];
1095	unsigned cMaxModifications;
1096	bool fForcedFlush = false;
1097	RT_NOREF_PV(uErrorCode);
1098
1099	# ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
1100	AssertMsg(pVCpu->pgm.s.enmGuestSlatMode == PGMSLAT_DIRECT,
1101	("pvFault=%RGv pPage=%p:{.idx=%d} GCPhysFault=%RGp\n", pvFault, pPage, pPage->idx, GCPhysFault));
1102	# endif
1103	LogFlow(("pgmRZPoolAccessPfHandler: pvFault=%RGv pPage=%p:{.idx=%d} GCPhysFault=%RGp\n", pvFault, pPage, pPage->idx, GCPhysFault));
1104
1105	PGM_LOCK_VOID(pVM);
1106	if (PHYS_PAGE_ADDRESS(GCPhysFault) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
1107	{
1108	/* Pool page changed while we were waiting for the lock; ignore. */
1109	Log(("CPU%d: pgmRZPoolAccessPfHandler pgm pool page for %RGp changed (to %RGp) while waiting!\n", pVCpu->idCpu, PHYS_PAGE_ADDRESS(GCPhysFault), PHYS_PAGE_ADDRESS(pPage->GCPhys)));
1110	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->StatMonitorPfRZ, &pPool->StatMonitorPfRZHandled, a);
1111	PGM_UNLOCK(pVM);
1112	return VINF_SUCCESS;
1113	}
1114	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1115	if (pPage->fDirty)
1116	{
1117	# ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
1118	Assert(!PGMPOOL_PAGE_IS_NESTED(pPage));
1119	# endif
1120	Assert(VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TLB_FLUSH));
1121	PGM_UNLOCK(pVM);
1122	return VINF_SUCCESS; /* SMP guest case where we were blocking on the pgm lock while the same page was being marked dirty. */
1123	}
1124	# endif
1125
1126	# if 0 /* test code defined(VBOX_STRICT) && defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT) */
1127	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1128	{
1129	void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
1130	void *pvGst;
1131	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
1132	pgmPoolTrackCheckPTPaePae(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PTPAE)pvGst);
1133	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvGst);
1134	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvShw);
1135	}
1136	# endif
1137
1138	# ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
1139	if (PGMPOOL_PAGE_IS_NESTED(pPage))
1140	{
1141	Assert(!CPUMIsGuestInVmxNonRootMode(CPUMQueryGuestCtxPtr(pVCpu)));
1142	Log7Func(("Flushing pvFault=%RGv GCPhysFault=%RGp\n", pvFault, GCPhysFault));
1143	pgmPoolMonitorChainFlush(pPool, pPage);
1144	PGM_UNLOCK(pVM);
1145	return VINF_SUCCESS;
1146	}
1147	# endif
1148
1149	/*
1150	* Disassemble the faulting instruction.
1151	*/
1152	PDISSTATE pDis = &pVCpu->pgm.s.Dis;
1153	int rc = EMInterpretDisasCurrent(pVCpu, pDis, NULL);
1154	if (RT_UNLIKELY(rc != VINF_SUCCESS))
1155	{
1156	AssertMsg(rc == VERR_PAGE_NOT_PRESENT \|\| rc == VERR_PAGE_TABLE_NOT_PRESENT, ("Unexpected rc %d\n", rc));
1157	PGM_UNLOCK(pVM);
1158	return rc;
1159	}
1160
1161	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
1162
1163	/*
1164	* We should ALWAYS have the list head as user parameter. This
1165	* is because we use that page to record the changes.
1166	*/
1167	Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1168
1169	# ifdef IN_RING0
1170	/* Maximum nr of modifications depends on the page type. */
1171	if ( pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT
1172	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_32BIT_PT)
1173	cMaxModifications = 4;
1174	else
1175	cMaxModifications = 24;
1176	# else
1177	cMaxModifications = 48;
1178	# endif
1179
1180	/*
1181	* Incremental page table updates should weigh more than random ones.
1182	* (Only applies when started from offset 0)
1183	*/
1184	pVCpu->pgm.s.cPoolAccessHandler++;
1185	if ( pPage->GCPtrLastAccessHandlerRip >= pCtx->rip - 0x40 /* observed loops in Windows 7 x64 */
1186	&& pPage->GCPtrLastAccessHandlerRip < pCtx->rip + 0x40
1187	&& pvFault == (pPage->GCPtrLastAccessHandlerFault + pDis->Param1.cb)
1188	&& pVCpu->pgm.s.cPoolAccessHandler == pPage->cLastAccessHandler + 1)
1189	{
1190	Log(("Possible page reuse cMods=%d -> %d (locked=%d type=%s)\n", pPage->cModifications, pPage->cModifications * 2, pgmPoolIsPageLocked(pPage), pgmPoolPoolKindToStr(pPage->enmKind)));
1191	Assert(pPage->cModifications < 32000);
1192	pPage->cModifications = pPage->cModifications * 2;
1193	pPage->GCPtrLastAccessHandlerFault = pvFault;
1194	pPage->cLastAccessHandler = pVCpu->pgm.s.cPoolAccessHandler;
1195	if (pPage->cModifications >= cMaxModifications)
1196	{
1197	STAM_COUNTER_INC(&pPool->StatMonitorPfRZFlushReinit);
1198	fForcedFlush = true;
1199	}
1200	}
1201
1202	if (pPage->cModifications >= cMaxModifications)
1203	Log(("Mod overflow %RGv cMods=%d (locked=%d type=%s)\n", pvFault, pPage->cModifications, pgmPoolIsPageLocked(pPage), pgmPoolPoolKindToStr(pPage->enmKind)));
1204
1205	/*
1206	* Check if it's worth dealing with.
1207	*/
1208	bool fReused = false;
1209	bool fNotReusedNotForking = false;
1210	if ( ( pPage->cModifications < cMaxModifications /** @todo \#define / /* @todo need to check that it's not mapping EIP. / /* @todo adjust this! */
1211	\|\| pgmPoolIsPageLocked(pPage)
1212	)
1213	&& !(fReused = pgmRZPoolMonitorIsReused(pVM, pVCpu, pCtx, pDis, pvFault, pPage))
1214	&& !pgmRZPoolMonitorIsForking(pPool, pDis, GCPhysFault & PAGE_OFFSET_MASK))
1215	{
1216	/*
1217	* Simple instructions, no REP prefix.
1218	*/
1219	if (!(pDis->fPrefix & (DISPREFIX_REP \| DISPREFIX_REPNE)))
1220	{
1221	rc = pgmRZPoolAccessPfHandlerSimple(pVM, pVCpu, pPool, pPage, pDis, pCtx, GCPhysFault, &fReused);
1222	if (fReused)
1223	goto flushPage;
1224
1225	/* A mov instruction to change the first page table entry will be remembered so we can detect
1226	* full page table changes early on. This will reduce the amount of unnecessary traps we'll take.
1227	*/
1228	if ( rc == VINF_SUCCESS
1229	&& !pPage->cLocked /* only applies to unlocked pages as we can't free locked ones (e.g. cr3 root). */
1230	&& pDis->pCurInstr->uOpcode == OP_MOV
1231	&& (pvFault & PAGE_OFFSET_MASK) == 0)
1232	{
1233	pPage->GCPtrLastAccessHandlerFault = pvFault;
1234	pPage->cLastAccessHandler = pVCpu->pgm.s.cPoolAccessHandler;
1235	pPage->GCPtrLastAccessHandlerRip = pCtx->rip;
1236	/* Make sure we don't kick out a page too quickly. */
1237	if (pPage->cModifications > 8)
1238	pPage->cModifications = 2;
1239	}
1240	else if (pPage->GCPtrLastAccessHandlerFault == pvFault)
1241	{
1242	/* ignore the 2nd write to this page table entry. */
1243	pPage->cLastAccessHandler = pVCpu->pgm.s.cPoolAccessHandler;
1244	}
1245	else
1246	{
1247	pPage->GCPtrLastAccessHandlerFault = NIL_RTGCPTR;
1248	pPage->GCPtrLastAccessHandlerRip = 0;
1249	}
1250
1251	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->StatMonitorPfRZ, &pPool->StatMonitorPfRZHandled, a);
1252	PGM_UNLOCK(pVM);
1253	return rc;
1254	}
1255
1256	/*
1257	* Windows is frequently doing small memset() operations (netio test 4k+).
1258	* We have to deal with these or we'll kill the cache and performance.
1259	*/
1260	if ( pDis->pCurInstr->uOpcode == OP_STOSWD
1261	&& !pCtx->eflags.Bits.u1DF
1262	&& pDis->uOpMode == pDis->uCpuMode
1263	&& pDis->uAddrMode == pDis->uCpuMode)
1264	{
1265	bool fValidStosd = false;
1266
1267	if ( pDis->uCpuMode == DISCPUMODE_32BIT
1268	&& pDis->fPrefix == DISPREFIX_REP
1269	&& pCtx->ecx <= 0x20
1270	&& pCtx->ecx * 4 <= GUEST_PAGE_SIZE - ((uintptr_t)pvFault & GUEST_PAGE_OFFSET_MASK)
1271	&& !((uintptr_t)pvFault & 3)
1272	&& (pCtx->eax == 0 \|\| pCtx->eax == 0x80) /* the two values observed. */
1273	)
1274	{
1275	fValidStosd = true;
1276	pCtx->rcx &= 0xffffffff; /* paranoia */
1277	}
1278	else
1279	if ( pDis->uCpuMode == DISCPUMODE_64BIT
1280	&& pDis->fPrefix == (DISPREFIX_REP \| DISPREFIX_REX)
1281	&& pCtx->rcx <= 0x20
1282	&& pCtx->rcx * 8 <= GUEST_PAGE_SIZE - ((uintptr_t)pvFault & GUEST_PAGE_OFFSET_MASK)
1283	&& !((uintptr_t)pvFault & 7)
1284	&& (pCtx->rax == 0 \|\| pCtx->rax == 0x80) /* the two values observed. */
1285	)
1286	{
1287	fValidStosd = true;
1288	}
1289
1290	if (fValidStosd)
1291	{
1292	rc = pgmRZPoolAccessPfHandlerSTOSD(pVM, pPool, pPage, pDis, pCtx, GCPhysFault, pvFault);
1293	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->StatMonitorPfRZ, &pPool->StatMonitorPfRZRepStosd, a);
1294	PGM_UNLOCK(pVM);
1295	return rc;
1296	}
1297	}
1298
1299	/* REP prefix, don't bother. */
1300	STAM_COUNTER_INC(&pPool->StatMonitorPfRZRepPrefix);
1301	Log4(("pgmRZPoolAccessPfHandler: eax=%#x ecx=%#x edi=%#x esi=%#x rip=%RGv opcode=%d prefix=%#x\n",
1302	pCtx->eax, pCtx->ecx, pCtx->edi, pCtx->esi, (RTGCPTR)pCtx->rip, pDis->pCurInstr->uOpcode, pDis->fPrefix));
1303	fNotReusedNotForking = true;
1304	}
1305
1306	# if defined(PGMPOOL_WITH_OPTIMIZED_DIRTY_PT) && defined(IN_RING0)
1307	/* E.g. Windows 7 x64 initializes page tables and touches some pages in the table during the process. This
1308	* leads to pgm pool trashing and an excessive amount of write faults due to page monitoring.
1309	*/
1310	if ( pPage->cModifications >= cMaxModifications
1311	&& !fForcedFlush
1312	&& (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT \|\| pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_32BIT_PT)
1313	&& ( fNotReusedNotForking
1314	\|\| ( !pgmRZPoolMonitorIsReused(pVM, pVCpu, pCtx, pDis, pvFault, pPage)
1315	&& !pgmRZPoolMonitorIsForking(pPool, pDis, GCPhysFault & PAGE_OFFSET_MASK))
1316	)
1317	)
1318	{
1319	Assert(!pgmPoolIsPageLocked(pPage));
1320	Assert(pPage->fDirty == false);
1321
1322	/* Flush any monitored duplicates as we will disable write protection. */
1323	if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
1324	\|\| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
1325	{
1326	PPGMPOOLPAGE pPageHead = pPage;
1327
1328	/* Find the monitor head. */
1329	while (pPageHead->iMonitoredPrev != NIL_PGMPOOL_IDX)
1330	pPageHead = &pPool->aPages[pPageHead->iMonitoredPrev];
1331
1332	while (pPageHead)
1333	{
1334	unsigned idxNext = pPageHead->iMonitoredNext;
1335
1336	if (pPageHead != pPage)
1337	{
1338	STAM_COUNTER_INC(&pPool->StatDirtyPageDupFlush);
1339	Log(("Flush duplicate page idx=%d GCPhys=%RGp type=%s\n", pPageHead->idx, pPageHead->GCPhys, pgmPoolPoolKindToStr(pPageHead->enmKind)));
1340	int rc2 = pgmPoolFlushPage(pPool, pPageHead);
1341	AssertRC(rc2);
1342	}
1343
1344	if (idxNext == NIL_PGMPOOL_IDX)
1345	break;
1346
1347	pPageHead = &pPool->aPages[idxNext];
1348	}
1349	}
1350
1351	/* The flushing above might fail for locked pages, so double check. */
1352	if ( pPage->iMonitoredNext == NIL_PGMPOOL_IDX
1353	&& pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
1354	{
1355	pgmPoolAddDirtyPage(pVM, pPool, pPage);
1356
1357	/* Temporarily allow write access to the page table again. */
1358	rc = PGMHandlerPhysicalPageTempOff(pVM,
1359	pPage->GCPhys & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK,
1360	pPage->GCPhys & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK);
1361	if (rc == VINF_SUCCESS)
1362	{
1363	rc = PGMShwMakePageWritable(pVCpu, pvFault, PGM_MK_PG_IS_WRITE_FAULT);
1364	AssertMsg(rc == VINF_SUCCESS
1365	/* In the SMP case the page table might be removed while we wait for the PGM lock in the trap handler. */
1366	\|\| rc == VERR_PAGE_TABLE_NOT_PRESENT
1367	\|\| rc == VERR_PAGE_NOT_PRESENT,
1368	("PGMShwModifyPage -> GCPtr=%RGv rc=%d\n", pvFault, rc));
1369	# ifdef VBOX_STRICT
1370	pPage->GCPtrDirtyFault = pvFault;
1371	# endif
1372
1373	STAM_PROFILE_STOP(&pVM->pgm.s.CTX_SUFF(pPool)->StatMonitorPfRZ, a);
1374	PGM_UNLOCK(pVM);
1375	return rc;
1376	}
1377	}
1378	}
1379	# endif /* PGMPOOL_WITH_OPTIMIZED_DIRTY_PT && IN_RING0 */
1380
1381	STAM_COUNTER_INC(&pPool->StatMonitorPfRZFlushModOverflow);
1382	flushPage:
1383	/*
1384	* Not worth it, so flush it.
1385	*
1386	* If we considered it to be reused, don't go back to ring-3
1387	* to emulate failed instructions since we usually cannot
1388	* interpret then. This may be a bit risky, in which case
1389	* the reuse detection must be fixed.
1390	*/
1391	rc = pgmRZPoolAccessPfHandlerFlush(pVM, pVCpu, pPool, pPage, pDis, pCtx, GCPhysFault);
1392	if ( rc == VINF_EM_RAW_EMULATE_INSTR
1393	&& fReused)
1394	{
1395	Assert(!PGMPOOL_PAGE_IS_NESTED(pPage)); /* temporary, remove later. */
1396	/* Make sure that the current instruction still has shadow page backing, otherwise we'll end up in a loop. */
1397	if (PGMShwGetPage(pVCpu, pCtx->rip, NULL, NULL) == VINF_SUCCESS)
1398	rc = VINF_SUCCESS; /* safe to restart the instruction. */
1399	}
1400	STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->StatMonitorPfRZ, &pPool->StatMonitorPfRZFlushPage, a);
1401	PGM_UNLOCK(pVM);
1402	return rc;
1403	}
1404
1405	#endif /* !IN_RING3 */
1406
1407	/**
1408	* @callback_method_impl{FNPGMPHYSHANDLER,
1409	* Access handler for shadowed page table pages.}
1410	*
1411	* @remarks Only uses the VINF_PGM_HANDLER_DO_DEFAULT status.
1412	* @note The @a uUser argument is the index of the PGMPOOLPAGE.
1413	*/
1414	DECLCALLBACK(VBOXSTRICTRC)
1415	pgmPoolAccessHandler(PVMCC pVM, PVMCPUCC pVCpu, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf,
1416	PGMACCESSTYPE enmAccessType, PGMACCESSORIGIN enmOrigin, uint64_t uUser)
1417	{
1418	PPGMPOOL const pPool = pVM->pgm.s.CTX_SUFF(pPool);
1419	STAM_PROFILE_START(&pPool->CTX_SUFF_Z(StatMonitor), a);
1420	AssertReturn(uUser < pPool->cCurPages, VERR_PGM_POOL_IPE);
1421	PPGMPOOLPAGE const pPage = &pPool->aPages[uUser];
1422	LogFlow(("PGM_ALL_CB_DECL: GCPhys=%RGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
1423	GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
1424
1425	NOREF(pvPhys); NOREF(pvBuf); NOREF(enmAccessType);
1426
1427	PGM_LOCK_VOID(pVM);
1428
1429	#ifdef VBOX_WITH_STATISTICS
1430	/*
1431	* Collect stats on the access.
1432	*/
1433	AssertCompile(RT_ELEMENTS(pPool->CTX_MID_Z(aStatMonitor,Sizes)) == 19);
1434	if (cbBuf <= 16 && cbBuf > 0)
1435	STAM_COUNTER_INC(&pPool->CTX_MID_Z(aStatMonitor,Sizes)[cbBuf - 1]);
1436	else if (cbBuf >= 17 && cbBuf < 32)
1437	STAM_COUNTER_INC(&pPool->CTX_MID_Z(aStatMonitor,Sizes)[16]);
1438	else if (cbBuf >= 32 && cbBuf < 64)
1439	STAM_COUNTER_INC(&pPool->CTX_MID_Z(aStatMonitor,Sizes)[17]);
1440	else if (cbBuf >= 64)
1441	STAM_COUNTER_INC(&pPool->CTX_MID_Z(aStatMonitor,Sizes)[18]);
1442
1443	uint8_t cbAlign;
1444	switch (pPage->enmKind)
1445	{
1446	default:
1447	cbAlign = 7;
1448	break;
1449	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
1450	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
1451	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
1452	case PGMPOOLKIND_32BIT_PD:
1453	case PGMPOOLKIND_32BIT_PD_PHYS:
1454	cbAlign = 3;
1455	break;
1456	}
1457	AssertCompile(RT_ELEMENTS(pPool->CTX_MID_Z(aStatMonitor,Misaligned)) == 7);
1458	if ((uint8_t)GCPhys & cbAlign)
1459	STAM_COUNTER_INC(&pPool->CTX_MID_Z(aStatMonitor,Misaligned)[((uint8_t)GCPhys & cbAlign) - 1]);
1460	#endif
1461
1462	/*
1463	* Make sure the pool page wasn't modified by a different CPU.
1464	*/
1465	if (PHYS_PAGE_ADDRESS(GCPhys) == PHYS_PAGE_ADDRESS(pPage->GCPhys))
1466	{
1467	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
1468
1469	/* The max modification count before flushing depends on the context and page type. */
1470	#ifdef IN_RING3
1471	uint16_t const cMaxModifications = 96; /* it's cheaper here, right? */
1472	#else
1473	uint16_t cMaxModifications;
1474	if ( pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT
1475	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_32BIT_PT)
1476	cMaxModifications = 4;
1477	else
1478	cMaxModifications = 24;
1479	#endif
1480
1481	/*
1482	* We don't have to be very sophisticated about this since there are relativly few calls here.
1483	* However, we must try our best to detect any non-cpu accesses (disk / networking).
1484	*/
1485	if ( ( pPage->cModifications < cMaxModifications
1486	\|\| pgmPoolIsPageLocked(pPage) )
1487	&& enmOrigin != PGMACCESSORIGIN_DEVICE
1488	&& cbBuf <= 16)
1489	{
1490	/* Clear the shadow entry. */
1491	if (!pPage->cModifications++)
1492	pgmPoolMonitorModifiedInsert(pPool, pPage);
1493
1494	if (cbBuf <= 8)
1495	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvBuf, (uint32_t)cbBuf);
1496	else
1497	{
1498	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvBuf, 8);
1499	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys + 8, (uint8_t *)pvBuf + 8, (uint32_t)cbBuf - 8);
1500	}
1501	}
1502	else
1503	pgmPoolMonitorChainFlush(pPool, pPage);
1504
1505	STAM_PROFILE_STOP_EX(&pPool->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,FlushPage), a);
1506	}
1507	else
1508	Log(("CPU%d: PGM_ALL_CB_DECL pgm pool page for %RGp changed (to %RGp) while waiting!\n", pVCpu->idCpu, PHYS_PAGE_ADDRESS(GCPhys), PHYS_PAGE_ADDRESS(pPage->GCPhys)));
1509	PGM_UNLOCK(pVM);
1510	return VINF_PGM_HANDLER_DO_DEFAULT;
1511	}
1512
1513
1514	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1515
1516	# if defined(VBOX_STRICT) && !defined(IN_RING3)
1517
1518	/**
1519	* Check references to guest physical memory in a PAE / PAE page table.
1520	*
1521	* @param pPool The pool.
1522	* @param pPage The page.
1523	* @param pShwPT The shadow page table (mapping of the page).
1524	* @param pGstPT The guest page table.
1525	*/
1526	static void pgmPoolTrackCheckPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PTPAE pGstPT)
1527	{
1528	unsigned cErrors = 0;
1529	int LastRc = -1; /* initialized to shut up gcc */
1530	unsigned LastPTE = ~0U; /* initialized to shut up gcc */
1531	RTHCPHYS LastHCPhys = NIL_RTHCPHYS; /* initialized to shut up gcc */
1532	PVMCC pVM = pPool->CTX_SUFF(pVM);
1533
1534	# ifdef VBOX_STRICT
1535	for (unsigned i = 0; i < RT_MIN(RT_ELEMENTS(pShwPT->a), pPage->iFirstPresent); i++)
1536	AssertMsg(!PGMSHWPTEPAE_IS_P(pShwPT->a[i]), ("Unexpected PTE: idx=%d %RX64 (first=%d)\n", i, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), pPage->iFirstPresent));
1537	# endif
1538	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
1539	{
1540	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
1541	{
1542	RTHCPHYS HCPhys = NIL_RTHCPHYS;
1543	int rc = PGMPhysGCPhys2HCPhys(pVM, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK, &HCPhys);
1544	if ( rc != VINF_SUCCESS
1545	\|\| PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]) != HCPhys)
1546	{
1547	Log(("rc=%d idx=%d guest %RX64 shw=%RX64 vs %RHp\n", rc, i, pGstPT->a[i].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), HCPhys));
1548	LastPTE = i;
1549	LastRc = rc;
1550	LastHCPhys = HCPhys;
1551	cErrors++;
1552
1553	RTHCPHYS HCPhysPT = NIL_RTHCPHYS;
1554	rc = PGMPhysGCPhys2HCPhys(pVM, pPage->GCPhys, &HCPhysPT);
1555	AssertRC(rc);
1556
1557	for (unsigned iPage = 0; iPage < pPool->cCurPages; iPage++)
1558	{
1559	PPGMPOOLPAGE pTempPage = &pPool->aPages[iPage];
1560
1561	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1562	{
1563	PPGMSHWPTPAE pShwPT2 = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pVM, pTempPage);
1564
1565	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
1566	{
1567	if ( PGMSHWPTEPAE_IS_P_RW(pShwPT2->a[j])
1568	&& PGMSHWPTEPAE_GET_HCPHYS(pShwPT2->a[j]) == HCPhysPT)
1569	{
1570	Log(("GCPhys=%RGp idx=%d %RX64 vs %RX64\n", pTempPage->GCPhys, j, PGMSHWPTEPAE_GET_LOG(pShwPT->a[j]), PGMSHWPTEPAE_GET_LOG(pShwPT2->a[j])));
1571	}
1572	}
1573
1574	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pShwPT2);
1575	}
1576	}
1577	}
1578	}
1579	}
1580	AssertMsg(!cErrors, ("cErrors=%d: last rc=%d idx=%d guest %RX64 shw=%RX64 vs %RHp\n", cErrors, LastRc, LastPTE, pGstPT->a[LastPTE].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[LastPTE]), LastHCPhys));
1581	}
1582
1583
1584	/**
1585	* Check references to guest physical memory in a PAE / 32-bit page table.
1586	*
1587	* @param pPool The pool.
1588	* @param pPage The page.
1589	* @param pShwPT The shadow page table (mapping of the page).
1590	* @param pGstPT The guest page table.
1591	*/
1592	static void pgmPoolTrackCheckPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PT pGstPT)
1593	{
1594	unsigned cErrors = 0;
1595	int LastRc = -1; /* initialized to shut up gcc */
1596	unsigned LastPTE = ~0U; /* initialized to shut up gcc */
1597	RTHCPHYS LastHCPhys = NIL_RTHCPHYS; /* initialized to shut up gcc */
1598	PVMCC pVM = pPool->CTX_SUFF(pVM);
1599
1600	# ifdef VBOX_STRICT
1601	for (unsigned i = 0; i < RT_MIN(RT_ELEMENTS(pShwPT->a), pPage->iFirstPresent); i++)
1602	AssertMsg(!PGMSHWPTEPAE_IS_P(pShwPT->a[i]), ("Unexpected PTE: idx=%d %RX64 (first=%d)\n", i, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), pPage->iFirstPresent));
1603	# endif
1604	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
1605	{
1606	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
1607	{
1608	RTHCPHYS HCPhys = NIL_RTHCPHYS;
1609	int rc = PGMPhysGCPhys2HCPhys(pVM, pGstPT->a[i].u & X86_PTE_PG_MASK, &HCPhys);
1610	if ( rc != VINF_SUCCESS
1611	\|\| PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]) != HCPhys)
1612	{
1613	Log(("rc=%d idx=%d guest %x shw=%RX64 vs %RHp\n", rc, i, pGstPT->a[i].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), HCPhys));
1614	LastPTE = i;
1615	LastRc = rc;
1616	LastHCPhys = HCPhys;
1617	cErrors++;
1618
1619	RTHCPHYS HCPhysPT = NIL_RTHCPHYS;
1620	rc = PGMPhysGCPhys2HCPhys(pVM, pPage->GCPhys, &HCPhysPT);
1621	AssertRC(rc);
1622
1623	for (unsigned iPage = 0; iPage < pPool->cCurPages; iPage++)
1624	{
1625	PPGMPOOLPAGE pTempPage = &pPool->aPages[iPage];
1626
1627	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_32BIT_PT)
1628	{
1629	PPGMSHWPTPAE pShwPT2 = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pVM, pTempPage);
1630
1631	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
1632	{
1633	if ( PGMSHWPTEPAE_IS_P_RW(pShwPT2->a[j])
1634	&& PGMSHWPTEPAE_GET_HCPHYS(pShwPT2->a[j]) == HCPhysPT)
1635	{
1636	Log(("GCPhys=%RGp idx=%d %RX64 vs %RX64\n", pTempPage->GCPhys, j, PGMSHWPTEPAE_GET_LOG(pShwPT->a[j]), PGMSHWPTEPAE_GET_LOG(pShwPT2->a[j])));
1637	}
1638	}
1639
1640	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pShwPT2);
1641	}
1642	}
1643	}
1644	}
1645	}
1646	AssertMsg(!cErrors, ("cErrors=%d: last rc=%d idx=%d guest %x shw=%RX64 vs %RHp\n", cErrors, LastRc, LastPTE, pGstPT->a[LastPTE].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[LastPTE]), LastHCPhys));
1647	}
1648
1649	# endif /* VBOX_STRICT && !IN_RING3 */
1650
1651	/**
1652	* Clear references to guest physical memory in a PAE / PAE page table.
1653	*
1654	* @returns nr of changed PTEs
1655	* @param pPool The pool.
1656	* @param pPage The page.
1657	* @param pShwPT The shadow page table (mapping of the page).
1658	* @param pGstPT The guest page table.
1659	* @param pOldGstPT The old cached guest page table.
1660	* @param fAllowRemoval Bail out as soon as we encounter an invalid PTE
1661	* @param pfFlush Flush reused page table (out)
1662	*/
1663	DECLINLINE(unsigned) pgmPoolTrackFlushPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PTPAE pGstPT,
1664	PCX86PTPAE pOldGstPT, bool fAllowRemoval, bool *pfFlush)
1665	{
1666	unsigned cChanged = 0;
1667
1668	# ifdef VBOX_STRICT
1669	for (unsigned i = 0; i < RT_MIN(RT_ELEMENTS(pShwPT->a), pPage->iFirstPresent); i++)
1670	AssertMsg(!PGMSHWPTEPAE_IS_P(pShwPT->a[i]), ("Unexpected PTE: idx=%d %RX64 (first=%d)\n", i, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), pPage->iFirstPresent));
1671	# endif
1672	*pfFlush = false;
1673
1674	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
1675	{
1676	/* Check the new value written by the guest. If present and with a bogus physical address, then
1677	* it's fairly safe to assume the guest is reusing the PT.
1678	*/
1679	if ( fAllowRemoval
1680	&& (pGstPT->a[i].u & X86_PTE_P))
1681	{
1682	if (!PGMPhysIsGCPhysValid(pPool->CTX_SUFF(pVM), pGstPT->a[i].u & X86_PTE_PAE_PG_MASK))
1683	{
1684	*pfFlush = true;
1685	return ++cChanged;
1686	}
1687	}
1688	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
1689	{
1690	/* If the old cached PTE is identical, then there's no need to flush the shadow copy. */
1691	if ((pGstPT->a[i].u & X86_PTE_PAE_PG_MASK) == (pOldGstPT->a[i].u & X86_PTE_PAE_PG_MASK))
1692	{
1693	# ifdef VBOX_STRICT
1694	RTHCPHYS HCPhys = NIL_RTGCPHYS;
1695	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[i].u & X86_PTE_PAE_PG_MASK, &HCPhys);
1696	AssertMsg(rc == VINF_SUCCESS && PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]) == HCPhys, ("rc=%d guest %RX64 old %RX64 shw=%RX64 vs %RHp\n", rc, pGstPT->a[i].u, pOldGstPT->a[i].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), HCPhys));
1697	# endif
1698	uint64_t uHostAttr = PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & (X86_PTE_P \| X86_PTE_US \| X86_PTE_A \| X86_PTE_D \| X86_PTE_G \| X86_PTE_PAE_NX);
1699	bool fHostRW = !!(PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & X86_PTE_RW);
1700	uint64_t uGuestAttr = pGstPT->a[i].u & (X86_PTE_P \| X86_PTE_US \| X86_PTE_A \| X86_PTE_D \| X86_PTE_G \| X86_PTE_PAE_NX);
1701	bool fGuestRW = !!(pGstPT->a[i].u & X86_PTE_RW);
1702
1703	if ( uHostAttr == uGuestAttr
1704	&& fHostRW <= fGuestRW)
1705	continue;
1706	}
1707	cChanged++;
1708	/* Something was changed, so flush it. */
1709	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX64 hint=%RX64\n",
1710	i, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pOldGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
1711	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pOldGstPT->a[i].u & X86_PTE_PAE_PG_MASK, i);
1712	PGMSHWPTEPAE_ATOMIC_SET(pShwPT->a[i], 0);
1713	}
1714	}
1715	return cChanged;
1716	}
1717
1718
1719	/**
1720	* Clear references to guest physical memory in a PAE / PAE page table.
1721	*
1722	* @returns nr of changed PTEs
1723	* @param pPool The pool.
1724	* @param pPage The page.
1725	* @param pShwPT The shadow page table (mapping of the page).
1726	* @param pGstPT The guest page table.
1727	* @param pOldGstPT The old cached guest page table.
1728	* @param fAllowRemoval Bail out as soon as we encounter an invalid PTE
1729	* @param pfFlush Flush reused page table (out)
1730	*/
1731	DECLINLINE(unsigned) pgmPoolTrackFlushPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PT pGstPT,
1732	PCX86PT pOldGstPT, bool fAllowRemoval, bool *pfFlush)
1733	{
1734	unsigned cChanged = 0;
1735
1736	# ifdef VBOX_STRICT
1737	for (unsigned i = 0; i < RT_MIN(RT_ELEMENTS(pShwPT->a), pPage->iFirstPresent); i++)
1738	AssertMsg(!PGMSHWPTEPAE_IS_P(pShwPT->a[i]), ("Unexpected PTE: idx=%d %RX64 (first=%d)\n", i, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), pPage->iFirstPresent));
1739	# endif
1740	*pfFlush = false;
1741
1742	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
1743	{
1744	/* Check the new value written by the guest. If present and with a bogus physical address, then
1745	* it's fairly safe to assume the guest is reusing the PT. */
1746	if (fAllowRemoval)
1747	{
1748	X86PGUINT const uPte = pGstPT->a[i].u;
1749	if ( (uPte & X86_PTE_P)
1750	&& !PGMPhysIsGCPhysValid(pPool->CTX_SUFF(pVM), uPte & X86_PTE_PG_MASK))
1751	{
1752	*pfFlush = true;
1753	return ++cChanged;
1754	}
1755	}
1756	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
1757	{
1758	/* If the old cached PTE is identical, then there's no need to flush the shadow copy. */
1759	if ((pGstPT->a[i].u & X86_PTE_PG_MASK) == (pOldGstPT->a[i].u & X86_PTE_PG_MASK))
1760	{
1761	# ifdef VBOX_STRICT
1762	RTHCPHYS HCPhys = NIL_RTGCPHYS;
1763	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[i].u & X86_PTE_PG_MASK, &HCPhys);
1764	AssertMsg(rc == VINF_SUCCESS && PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]) == HCPhys, ("rc=%d guest %x old %x shw=%RX64 vs %RHp\n", rc, pGstPT->a[i].u, pOldGstPT->a[i].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[i]), HCPhys));
1765	# endif
1766	uint64_t uHostAttr = PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & (X86_PTE_P \| X86_PTE_US \| X86_PTE_A \| X86_PTE_D \| X86_PTE_G);
1767	bool fHostRW = !!(PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & X86_PTE_RW);
1768	uint64_t uGuestAttr = pGstPT->a[i].u & (X86_PTE_P \| X86_PTE_US \| X86_PTE_A \| X86_PTE_D \| X86_PTE_G);
1769	bool fGuestRW = !!(pGstPT->a[i].u & X86_PTE_RW);
1770
1771	if ( uHostAttr == uGuestAttr
1772	&& fHostRW <= fGuestRW)
1773	continue;
1774	}
1775	cChanged++;
1776	/* Something was changed, so flush it. */
1777	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX64 hint=%x\n",
1778	i, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pOldGstPT->a[i].u & X86_PTE_PG_MASK));
1779	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pOldGstPT->a[i].u & X86_PTE_PG_MASK, i);
1780	PGMSHWPTEPAE_ATOMIC_SET(pShwPT->a[i], 0);
1781	}
1782	}
1783	return cChanged;
1784	}
1785
1786
1787	/**
1788	* Flush a dirty page
1789	*
1790	* @param pVM The cross context VM structure.
1791	* @param pPool The pool.
1792	* @param idxSlot Dirty array slot index
1793	* @param fAllowRemoval Allow a reused page table to be removed
1794	*/
1795	static void pgmPoolFlushDirtyPage(PVMCC pVM, PPGMPOOL pPool, unsigned idxSlot, bool fAllowRemoval = false)
1796	{
1797	AssertCompile(RT_ELEMENTS(pPool->aidxDirtyPages) == RT_ELEMENTS(pPool->aDirtyPages));
1798
1799	Assert(idxSlot < RT_ELEMENTS(pPool->aDirtyPages));
1800	unsigned idxPage = pPool->aidxDirtyPages[idxSlot];
1801	if (idxPage == NIL_PGMPOOL_IDX)
1802	return;
1803
1804	PPGMPOOLPAGE pPage = &pPool->aPages[idxPage];
1805	Assert(pPage->idx == idxPage);
1806	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1807
1808	AssertMsg(pPage->fDirty, ("Page %RGp (slot=%d) not marked dirty!", pPage->GCPhys, idxSlot));
1809	Log(("Flush dirty page %RGp cMods=%d\n", pPage->GCPhys, pPage->cModifications));
1810
1811	/* First write protect the page again to catch all write accesses. (before checking for changes -> SMP) */
1812	int rc = PGMHandlerPhysicalReset(pVM, pPage->GCPhys & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK);
1813	Assert(rc == VINF_SUCCESS);
1814	pPage->fDirty = false;
1815
1816	# ifdef VBOX_STRICT
1817	uint64_t fFlags = 0;
1818	RTHCPHYS HCPhys;
1819	rc = PGMShwGetPage(VMMGetCpu(pVM), pPage->GCPtrDirtyFault, &fFlags, &HCPhys);
1820	AssertMsg( ( rc == VINF_SUCCESS
1821	&& (!(fFlags & X86_PTE_RW) \|\| HCPhys != pPage->Core.Key))
1822	/* In the SMP case the page table might be removed while we wait for the PGM lock in the trap handler. */
1823	\|\| rc == VERR_PAGE_TABLE_NOT_PRESENT
1824	\|\| rc == VERR_PAGE_NOT_PRESENT,
1825	("PGMShwGetPage -> GCPtr=%RGv rc=%d flags=%RX64\n", pPage->GCPtrDirtyFault, rc, fFlags));
1826	# endif
1827
1828	/* Flush those PTEs that have changed. */
1829	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
1830	void *pvShw = PGMPOOL_PAGE_2_PTR(pVM, pPage);
1831	void *pvGst;
1832	rc = PGM_GCPHYS_2_PTR_EX(pVM, pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
1833	bool fFlush;
1834	unsigned cChanges;
1835
1836	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1837	cChanges = pgmPoolTrackFlushPTPaePae(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PTPAE)pvGst,
1838	(PCX86PTPAE)&pPool->aDirtyPages[idxSlot].aPage[0], fAllowRemoval, &fFlush);
1839	else
1840	{
1841	Assert(!PGMPOOL_PAGE_IS_NESTED(pPage)); /* temporary, remove later. */
1842	cChanges = pgmPoolTrackFlushPTPae32Bit(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PT)pvGst,
1843	(PCX86PT)&pPool->aDirtyPages[idxSlot].aPage[0], fAllowRemoval, &fFlush);
1844	}
1845
1846	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvGst);
1847	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvShw);
1848	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
1849	/* Note: we might want to consider keeping the dirty page active in case there were many changes. */
1850
1851	/* This page is likely to be modified again, so reduce the nr of modifications just a bit here. */
1852	Assert(pPage->cModifications);
1853	if (cChanges < 4)
1854	pPage->cModifications = 1; /* must use > 0 here */
1855	else
1856	pPage->cModifications = RT_MAX(1, pPage->cModifications / 2);
1857
1858	STAM_COUNTER_INC(&pPool->StatResetDirtyPages);
1859	if (pPool->cDirtyPages == RT_ELEMENTS(pPool->aDirtyPages))
1860	pPool->idxFreeDirtyPage = idxSlot;
1861
1862	pPool->cDirtyPages--;
1863	pPool->aidxDirtyPages[idxSlot] = NIL_PGMPOOL_IDX;
1864	Assert(pPool->cDirtyPages <= RT_ELEMENTS(pPool->aDirtyPages));
1865	if (fFlush)
1866	{
1867	Assert(fAllowRemoval);
1868	Log(("Flush reused page table!\n"));
1869	pgmPoolFlushPage(pPool, pPage);
1870	STAM_COUNTER_INC(&pPool->StatForceFlushReused);
1871	}
1872	else
1873	Log(("Removed dirty page %RGp cMods=%d cChanges=%d\n", pPage->GCPhys, pPage->cModifications, cChanges));
1874	}
1875
1876
1877	# ifndef IN_RING3
1878	/**
1879	* Add a new dirty page
1880	*
1881	* @param pVM The cross context VM structure.
1882	* @param pPool The pool.
1883	* @param pPage The page.
1884	*/
1885	void pgmPoolAddDirtyPage(PVMCC pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage)
1886	{
1887	PGM_LOCK_ASSERT_OWNER(pVM);
1888	AssertCompile(RT_ELEMENTS(pPool->aDirtyPages) == 8 \|\| RT_ELEMENTS(pPool->aDirtyPages) == 16);
1889	Assert(!pPage->fDirty);
1890	Assert(!PGMPOOL_PAGE_IS_NESTED(pPage));
1891
1892	unsigned idxFree = pPool->idxFreeDirtyPage;
1893	Assert(idxFree < RT_ELEMENTS(pPool->aDirtyPages));
1894	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
1895
1896	if (pPool->cDirtyPages >= RT_ELEMENTS(pPool->aDirtyPages))
1897	{
1898	STAM_COUNTER_INC(&pPool->StatDirtyPageOverFlowFlush);
1899	pgmPoolFlushDirtyPage(pVM, pPool, idxFree, true /* allow removal of reused page tables*/);
1900	}
1901	Assert(pPool->cDirtyPages < RT_ELEMENTS(pPool->aDirtyPages));
1902	AssertMsg(pPool->aidxDirtyPages[idxFree] == NIL_PGMPOOL_IDX, ("idxFree=%d cDirtyPages=%d\n", idxFree, pPool->cDirtyPages));
1903
1904	Log(("Add dirty page %RGp (slot=%d)\n", pPage->GCPhys, idxFree));
1905
1906	/*
1907	* Make a copy of the guest page table as we require valid GCPhys addresses
1908	* when removing references to physical pages.
1909	* (The HCPhys linear lookup is extremely expensive!)
1910	*/
1911	void *pvGst;
1912	int rc = PGM_GCPHYS_2_PTR_EX(pVM, pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
1913	memcpy(&pPool->aDirtyPages[idxFree].aPage[0], pvGst,
1914	pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT ? PAGE_SIZE : PAGE_SIZE / 2);
1915	# ifdef VBOX_STRICT
1916	void *pvShw = PGMPOOL_PAGE_2_PTR(pVM, pPage);
1917	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1918	pgmPoolTrackCheckPTPaePae(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PTPAE)pvGst);
1919	else
1920	pgmPoolTrackCheckPTPae32Bit(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PT)pvGst);
1921	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvShw);
1922	# endif
1923	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvGst);
1924
1925	STAM_COUNTER_INC(&pPool->StatDirtyPage);
1926	pPage->fDirty = true;
1927	pPage->idxDirtyEntry = (uint8_t)idxFree; Assert(pPage->idxDirtyEntry == idxFree);
1928	pPool->aidxDirtyPages[idxFree] = pPage->idx;
1929	pPool->cDirtyPages++;
1930
1931	pPool->idxFreeDirtyPage = (pPool->idxFreeDirtyPage + 1) & (RT_ELEMENTS(pPool->aDirtyPages) - 1);
1932	if ( pPool->cDirtyPages < RT_ELEMENTS(pPool->aDirtyPages)
1933	&& pPool->aidxDirtyPages[pPool->idxFreeDirtyPage] != NIL_PGMPOOL_IDX)
1934	{
1935	unsigned i;
1936	for (i = 1; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
1937	{
1938	idxFree = (pPool->idxFreeDirtyPage + i) & (RT_ELEMENTS(pPool->aDirtyPages) - 1);
1939	if (pPool->aidxDirtyPages[idxFree] == NIL_PGMPOOL_IDX)
1940	{
1941	pPool->idxFreeDirtyPage = idxFree;
1942	break;
1943	}
1944	}
1945	Assert(i != RT_ELEMENTS(pPool->aDirtyPages));
1946	}
1947
1948	Assert(pPool->cDirtyPages == RT_ELEMENTS(pPool->aDirtyPages) \|\| pPool->aidxDirtyPages[pPool->idxFreeDirtyPage] == NIL_PGMPOOL_IDX);
1949
1950	/*
1951	* Clear all references to this shadow table. See @bugref{7298}.
1952	*/
1953	pgmPoolTrackClearPageUsers(pPool, pPage);
1954	}
1955	# endif /* !IN_RING3 */
1956
1957
1958	/**
1959	* Check if the specified page is dirty (not write monitored)
1960	*
1961	* @return dirty or not
1962	* @param pVM The cross context VM structure.
1963	* @param GCPhys Guest physical address
1964	*/
1965	bool pgmPoolIsDirtyPageSlow(PVMCC pVM, RTGCPHYS GCPhys)
1966	{
1967	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1968	PGM_LOCK_ASSERT_OWNER(pVM);
1969	if (!pPool->cDirtyPages)
1970	return false;
1971
1972	GCPhys = GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK;
1973
1974	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
1975	{
1976	unsigned idxPage = pPool->aidxDirtyPages[i];
1977	if (idxPage != NIL_PGMPOOL_IDX)
1978	{
1979	PPGMPOOLPAGE pPage = &pPool->aPages[idxPage];
1980	if (pPage->GCPhys == GCPhys)
1981	return true;
1982	}
1983	}
1984	return false;
1985	}
1986
1987
1988	/**
1989	* Reset all dirty pages by reinstating page monitoring.
1990	*
1991	* @param pVM The cross context VM structure.
1992	*/
1993	void pgmPoolResetDirtyPages(PVMCC pVM)
1994	{
1995	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1996	PGM_LOCK_ASSERT_OWNER(pVM);
1997	Assert(pPool->cDirtyPages <= RT_ELEMENTS(pPool->aDirtyPages));
1998
1999	if (!pPool->cDirtyPages)
2000	return;
2001
2002	Log(("pgmPoolResetDirtyPages\n"));
2003	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
2004	pgmPoolFlushDirtyPage(pVM, pPool, i, true /* allow removal of reused page tables*/);
2005
2006	pPool->idxFreeDirtyPage = 0;
2007	if ( pPool->cDirtyPages != RT_ELEMENTS(pPool->aDirtyPages)
2008	&& pPool->aidxDirtyPages[pPool->idxFreeDirtyPage] != NIL_PGMPOOL_IDX)
2009	{
2010	unsigned i;
2011	for (i = 1; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
2012	{
2013	if (pPool->aidxDirtyPages[i] == NIL_PGMPOOL_IDX)
2014	{
2015	pPool->idxFreeDirtyPage = i;
2016	break;
2017	}
2018	}
2019	AssertMsg(i != RT_ELEMENTS(pPool->aDirtyPages), ("cDirtyPages %d", pPool->cDirtyPages));
2020	}
2021
2022	Assert(pPool->aidxDirtyPages[pPool->idxFreeDirtyPage] == NIL_PGMPOOL_IDX \|\| pPool->cDirtyPages == RT_ELEMENTS(pPool->aDirtyPages));
2023	return;
2024	}
2025
2026
2027	/**
2028	* Invalidate the PT entry for the specified page
2029	*
2030	* @param pVM The cross context VM structure.
2031	* @param GCPtrPage Guest page to invalidate
2032	*/
2033	void pgmPoolResetDirtyPage(PVMCC pVM, RTGCPTR GCPtrPage)
2034	{
2035	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2036	PGM_LOCK_ASSERT_OWNER(pVM);
2037	Assert(pPool->cDirtyPages <= RT_ELEMENTS(pPool->aDirtyPages));
2038
2039	if (!pPool->cDirtyPages)
2040	return;
2041
2042	Log(("pgmPoolResetDirtyPage %RGv\n", GCPtrPage)); RT_NOREF_PV(GCPtrPage);
2043	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
2044	{
2045	/** @todo What was intended here??? This looks incomplete... */
2046	}
2047	}
2048
2049
2050	/**
2051	* Reset all dirty pages by reinstating page monitoring.
2052	*
2053	* @param pVM The cross context VM structure.
2054	* @param GCPhysPT Physical address of the page table
2055	*/
2056	void pgmPoolInvalidateDirtyPage(PVMCC pVM, RTGCPHYS GCPhysPT)
2057	{
2058	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2059	PGM_LOCK_ASSERT_OWNER(pVM);
2060	Assert(pPool->cDirtyPages <= RT_ELEMENTS(pPool->aDirtyPages));
2061	unsigned idxDirtyPage = RT_ELEMENTS(pPool->aDirtyPages);
2062
2063	if (!pPool->cDirtyPages)
2064	return;
2065
2066	GCPhysPT = GCPhysPT & ~(RTGCPHYS)PAGE_OFFSET_MASK;
2067
2068	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
2069	{
2070	unsigned idxPage = pPool->aidxDirtyPages[i];
2071	if (idxPage != NIL_PGMPOOL_IDX)
2072	{
2073	PPGMPOOLPAGE pPage = &pPool->aPages[idxPage];
2074	if (pPage->GCPhys == GCPhysPT)
2075	{
2076	idxDirtyPage = i;
2077	break;
2078	}
2079	}
2080	}
2081
2082	if (idxDirtyPage != RT_ELEMENTS(pPool->aDirtyPages))
2083	{
2084	pgmPoolFlushDirtyPage(pVM, pPool, idxDirtyPage, true /* allow removal of reused page tables*/);
2085	if ( pPool->cDirtyPages != RT_ELEMENTS(pPool->aDirtyPages)
2086	&& pPool->aidxDirtyPages[pPool->idxFreeDirtyPage] != NIL_PGMPOOL_IDX)
2087	{
2088	unsigned i;
2089	for (i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
2090	{
2091	if (pPool->aidxDirtyPages[i] == NIL_PGMPOOL_IDX)
2092	{
2093	pPool->idxFreeDirtyPage = i;
2094	break;
2095	}
2096	}
2097	AssertMsg(i != RT_ELEMENTS(pPool->aDirtyPages), ("cDirtyPages %d", pPool->cDirtyPages));
2098	}
2099	}
2100	}
2101
2102	#endif /* PGMPOOL_WITH_OPTIMIZED_DIRTY_PT */
2103
2104	/**
2105	* Inserts a page into the GCPhys hash table.
2106	*
2107	* @param pPool The pool.
2108	* @param pPage The page.
2109	*/
2110	DECLINLINE(void) pgmPoolHashInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2111	{
2112	Log3(("pgmPoolHashInsert: %RGp\n", pPage->GCPhys));
2113	Assert(pPage->GCPhys != NIL_RTGCPHYS); Assert(pPage->iNext == NIL_PGMPOOL_IDX);
2114	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
2115	pPage->iNext = pPool->aiHash[iHash];
2116	pPool->aiHash[iHash] = pPage->idx;
2117	}
2118
2119
2120	/**
2121	* Removes a page from the GCPhys hash table.
2122	*
2123	* @param pPool The pool.
2124	* @param pPage The page.
2125	*/
2126	DECLINLINE(void) pgmPoolHashRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2127	{
2128	Log3(("pgmPoolHashRemove: %RGp\n", pPage->GCPhys));
2129	uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
2130	if (pPool->aiHash[iHash] == pPage->idx)
2131	pPool->aiHash[iHash] = pPage->iNext;
2132	else
2133	{
2134	uint16_t iPrev = pPool->aiHash[iHash];
2135	for (;;)
2136	{
2137	const int16_t i = pPool->aPages[iPrev].iNext;
2138	if (i == pPage->idx)
2139	{
2140	pPool->aPages[iPrev].iNext = pPage->iNext;
2141	break;
2142	}
2143	if (i == NIL_PGMPOOL_IDX)
2144	{
2145	AssertReleaseMsgFailed(("GCPhys=%RGp idx=%d\n", pPage->GCPhys, pPage->idx));
2146	break;
2147	}
2148	iPrev = i;
2149	}
2150	}
2151	pPage->iNext = NIL_PGMPOOL_IDX;
2152	}
2153
2154
2155	/**
2156	* Frees up one cache page.
2157	*
2158	* @returns VBox status code.
2159	* @retval VINF_SUCCESS on success.
2160	* @param pPool The pool.
2161	* @param iUser The user index.
2162	*/
2163	static int pgmPoolCacheFreeOne(PPGMPOOL pPool, uint16_t iUser)
2164	{
2165	const PVMCC pVM = pPool->CTX_SUFF(pVM);
2166	Assert(pPool->iAgeHead != pPool->iAgeTail); /* We shouldn't be here if there < 2 cached entries! */
2167	STAM_COUNTER_INC(&pPool->StatCacheFreeUpOne);
2168
2169	/*
2170	* Select one page from the tail of the age list.
2171	*/
2172	PPGMPOOLPAGE pPage;
2173	for (unsigned iLoop = 0; ; iLoop++)
2174	{
2175	uint16_t iToFree = pPool->iAgeTail;
2176	if (iToFree == iUser && iUser != NIL_PGMPOOL_IDX)
2177	iToFree = pPool->aPages[iToFree].iAgePrev;
2178	/* This is the alternative to the SyncCR3 pgmPoolCacheUsed calls.
2179	if (pPool->aPages[iToFree].iUserHead != NIL_PGMPOOL_USER_INDEX)
2180	{
2181	uint16_t i = pPool->aPages[iToFree].iAgePrev;
2182	for (unsigned j = 0; j < 10 && i != NIL_PGMPOOL_USER_INDEX; j++, i = pPool->aPages[i].iAgePrev)
2183	{
2184	if (pPool->aPages[iToFree].iUserHead == NIL_PGMPOOL_USER_INDEX)
2185	continue;
2186	iToFree = i;
2187	break;
2188	}
2189	}
2190	*/
2191	Assert(iToFree != iUser);
2192	AssertReleaseMsg(iToFree != NIL_PGMPOOL_IDX,
2193	("iToFree=%#x (iAgeTail=%#x) iUser=%#x iLoop=%u - pPool=%p LB %#zx\n",
2194	iToFree, pPool->iAgeTail, iUser, iLoop, pPool,
2195	RT_UOFFSETOF_DYN(PGMPOOL, aPages[pPool->cMaxPages])
2196	+ pPool->cMaxUsers * sizeof(PGMPOOLUSER)
2197	+ pPool->cMaxPhysExts * sizeof(PGMPOOLPHYSEXT) ));
2198
2199	pPage = &pPool->aPages[iToFree];
2200
2201	/*
2202	* Reject any attempts at flushing the currently active shadow CR3 mapping.
2203	* Call pgmPoolCacheUsed to move the page to the head of the age list.
2204	*/
2205	if ( !pgmPoolIsPageLocked(pPage)
2206	&& pPage->idx >= PGMPOOL_IDX_FIRST /* paranoia (#6349) */)
2207	break;
2208	LogFlow(("pgmPoolCacheFreeOne: refuse CR3 mapping\n"));
2209	pgmPoolCacheUsed(pPool, pPage);
2210	AssertLogRelReturn(iLoop < 8192, VERR_PGM_POOL_TOO_MANY_LOOPS);
2211	}
2212
2213	/*
2214	* Found a usable page, flush it and return.
2215	*/
2216	int rc = pgmPoolFlushPage(pPool, pPage);
2217	/* This flush was initiated by us and not the guest, so explicitly flush the TLB. */
2218	/** @todo find out why this is necessary; pgmPoolFlushPage should trigger a flush if one is really needed. */
2219	if (rc == VINF_SUCCESS)
2220	PGM_INVL_ALL_VCPU_TLBS(pVM);
2221	return rc;
2222	}
2223
2224
2225	/**
2226	* Checks if a kind mismatch is really a page being reused
2227	* or if it's just normal remappings.
2228	*
2229	* @returns true if reused and the cached page (enmKind1) should be flushed
2230	* @returns false if not reused.
2231	* @param enmKind1 The kind of the cached page.
2232	* @param enmKind2 The kind of the requested page.
2233	*/
2234	static bool pgmPoolCacheReusedByKind(PGMPOOLKIND enmKind1, PGMPOOLKIND enmKind2)
2235	{
2236	switch (enmKind1)
2237	{
2238	/*
2239	* Never reuse them. There is no remapping in non-paging mode.
2240	*/
2241	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2242	case PGMPOOLKIND_32BIT_PD_PHYS:
2243	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2244	case PGMPOOLKIND_PAE_PD_PHYS:
2245	case PGMPOOLKIND_PAE_PDPT_PHYS:
2246	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2247	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2248	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2249	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2250	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2251	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT: /* never reuse them for other types */
2252	return false;
2253
2254	/*
2255	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
2256	*/
2257	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2258	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2259	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2260	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2261	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2262	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2263	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2264	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2265	case PGMPOOLKIND_32BIT_PD:
2266	case PGMPOOLKIND_PAE_PDPT:
2267	Assert(!PGMPOOL_PAGE_IS_KIND_NESTED(enmKind2));
2268	switch (enmKind2)
2269	{
2270	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2271	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2272	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2273	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2274	case PGMPOOLKIND_64BIT_PML4:
2275	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2276	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2277	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2278	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2279	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2280	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2281	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2282	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2283	return true;
2284	default:
2285	return false;
2286	}
2287
2288	/*
2289	* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
2290	*/
2291	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2292	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2293	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2294	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2295	case PGMPOOLKIND_64BIT_PML4:
2296	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2297	Assert(!PGMPOOL_PAGE_IS_KIND_NESTED(enmKind2));
2298	switch (enmKind2)
2299	{
2300	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2301	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2302	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2303	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2304	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2305	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2306	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2307	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2308	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2309	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2310	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2311	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2312	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2313	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2314	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2315	return true;
2316	default:
2317	return false;
2318	}
2319
2320	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
2321	case PGMPOOLKIND_EPT_PT_FOR_EPT_PT:
2322	case PGMPOOLKIND_EPT_PT_FOR_EPT_2MB:
2323	case PGMPOOLKIND_EPT_PD_FOR_EPT_PD:
2324	case PGMPOOLKIND_EPT_PDPT_FOR_EPT_PDPT:
2325	return PGMPOOL_PAGE_IS_KIND_NESTED(enmKind2);
2326
2327	case PGMPOOLKIND_EPT_PML4_FOR_EPT_PML4:
2328	return false;
2329	#endif
2330
2331	/*
2332	* These cannot be flushed, and it's common to reuse the PDs as PTs.
2333	*/
2334	case PGMPOOLKIND_ROOT_NESTED:
2335	return false;
2336
2337	default:
2338	AssertFatalMsgFailed(("enmKind1=%d\n", enmKind1));
2339	}
2340	}
2341
2342
2343	/**
2344	* Attempts to satisfy a pgmPoolAlloc request from the cache.
2345	*
2346	* @returns VBox status code.
2347	* @retval VINF_PGM_CACHED_PAGE on success.
2348	* @retval VERR_FILE_NOT_FOUND if not found.
2349	* @param pPool The pool.
2350	* @param GCPhys The GC physical address of the page we're gonna shadow.
2351	* @param enmKind The kind of mapping.
2352	* @param enmAccess Access type for the mapping (only relevant for big pages)
2353	* @param fA20Enabled Whether the CPU has the A20 gate enabled.
2354	* @param iUser The shadow page pool index of the user table. This is
2355	* NIL_PGMPOOL_IDX for root pages.
2356	* @param iUserTable The index into the user table (shadowed). Ignored if
2357	* root page
2358	* @param ppPage Where to store the pointer to the page.
2359	*/
2360	static int pgmPoolCacheAlloc(PPGMPOOL pPool, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, PGMPOOLACCESS enmAccess, bool fA20Enabled,
2361	uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
2362	{
2363	/*
2364	* Look up the GCPhys in the hash.
2365	*/
2366	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
2367	Log3(("pgmPoolCacheAlloc: %RGp kind %s iUser=%d iUserTable=%x SLOT=%d\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable, i));
2368	if (i != NIL_PGMPOOL_IDX)
2369	{
2370	do
2371	{
2372	PPGMPOOLPAGE pPage = &pPool->aPages[i];
2373	Log4(("pgmPoolCacheAlloc: slot %d found page %RGp\n", i, pPage->GCPhys));
2374	if (pPage->GCPhys == GCPhys)
2375	{
2376	if ( (PGMPOOLKIND)pPage->enmKind == enmKind
2377	&& (PGMPOOLACCESS)pPage->enmAccess == enmAccess
2378	&& pPage->fA20Enabled == fA20Enabled)
2379	{
2380	/* Put it at the start of the use list to make sure pgmPoolTrackAddUser
2381	* doesn't flush it in case there are no more free use records.
2382	*/
2383	pgmPoolCacheUsed(pPool, pPage);
2384
2385	int rc = VINF_SUCCESS;
2386	if (iUser != NIL_PGMPOOL_IDX)
2387	rc = pgmPoolTrackAddUser(pPool, pPage, iUser, iUserTable);
2388	if (RT_SUCCESS(rc))
2389	{
2390	Assert((PGMPOOLKIND)pPage->enmKind == enmKind);
2391	*ppPage = pPage;
2392	if (pPage->cModifications)
2393	pPage->cModifications = 1; /* reset counter (can't use 0, or else it will be reinserted in the modified list) */
2394	STAM_COUNTER_INC(&pPool->StatCacheHits);
2395	return VINF_PGM_CACHED_PAGE;
2396	}
2397	return rc;
2398	}
2399
2400	if ((PGMPOOLKIND)pPage->enmKind != enmKind)
2401	{
2402	/*
2403	* The kind is different. In some cases we should now flush the page
2404	* as it has been reused, but in most cases this is normal remapping
2405	* of PDs as PT or big pages using the GCPhys field in a slightly
2406	* different way than the other kinds.
2407	*/
2408	if (pgmPoolCacheReusedByKind((PGMPOOLKIND)pPage->enmKind, enmKind))
2409	{
2410	STAM_COUNTER_INC(&pPool->StatCacheKindMismatches);
2411	pgmPoolFlushPage(pPool, pPage);
2412	break;
2413	}
2414	}
2415	}
2416
2417	/* next */
2418	i = pPage->iNext;
2419	} while (i != NIL_PGMPOOL_IDX);
2420	}
2421
2422	Log3(("pgmPoolCacheAlloc: Missed GCPhys=%RGp enmKind=%s\n", GCPhys, pgmPoolPoolKindToStr(enmKind)));
2423	STAM_COUNTER_INC(&pPool->StatCacheMisses);
2424	return VERR_FILE_NOT_FOUND;
2425	}
2426
2427
2428	/**
2429	* Inserts a page into the cache.
2430	*
2431	* @param pPool The pool.
2432	* @param pPage The cached page.
2433	* @param fCanBeCached Set if the page is fit for caching from the caller's point of view.
2434	*/
2435	static void pgmPoolCacheInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fCanBeCached)
2436	{
2437	/*
2438	* Insert into the GCPhys hash if the page is fit for that.
2439	*/
2440	Assert(!pPage->fCached);
2441	if (fCanBeCached)
2442	{
2443	pPage->fCached = true;
2444	pgmPoolHashInsert(pPool, pPage);
2445	Log3(("pgmPoolCacheInsert: Caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
2446	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
2447	STAM_COUNTER_INC(&pPool->StatCacheCacheable);
2448	}
2449	else
2450	{
2451	Log3(("pgmPoolCacheInsert: Not caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
2452	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
2453	STAM_COUNTER_INC(&pPool->StatCacheUncacheable);
2454	}
2455
2456	/*
2457	* Insert at the head of the age list.
2458	*/
2459	pPage->iAgePrev = NIL_PGMPOOL_IDX;
2460	pPage->iAgeNext = pPool->iAgeHead;
2461	if (pPool->iAgeHead != NIL_PGMPOOL_IDX)
2462	pPool->aPages[pPool->iAgeHead].iAgePrev = pPage->idx;
2463	else
2464	pPool->iAgeTail = pPage->idx;
2465	pPool->iAgeHead = pPage->idx;
2466	}
2467
2468
2469	/**
2470	* Flushes a cached page.
2471	*
2472	* @param pPool The pool.
2473	* @param pPage The cached page.
2474	*/
2475	static void pgmPoolCacheFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2476	{
2477	Log3(("pgmPoolCacheFlushPage: %RGp\n", pPage->GCPhys));
2478
2479	/*
2480	* Remove the page from the hash.
2481	*/
2482	if (pPage->fCached)
2483	{
2484	pPage->fCached = false;
2485	pgmPoolHashRemove(pPool, pPage);
2486	}
2487	else
2488	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
2489
2490	/*
2491	* Remove it from the age list.
2492	*/
2493	if (pPage->iAgeNext != NIL_PGMPOOL_IDX)
2494	pPool->aPages[pPage->iAgeNext].iAgePrev = pPage->iAgePrev;
2495	else
2496	pPool->iAgeTail = pPage->iAgePrev;
2497	if (pPage->iAgePrev != NIL_PGMPOOL_IDX)
2498	pPool->aPages[pPage->iAgePrev].iAgeNext = pPage->iAgeNext;
2499	else
2500	pPool->iAgeHead = pPage->iAgeNext;
2501	pPage->iAgeNext = NIL_PGMPOOL_IDX;
2502	pPage->iAgePrev = NIL_PGMPOOL_IDX;
2503	}
2504
2505
2506	/**
2507	* Looks for pages sharing the monitor.
2508	*
2509	* @returns Pointer to the head page.
2510	* @returns NULL if not found.
2511	* @param pPool The Pool
2512	* @param pNewPage The page which is going to be monitored.
2513	*/
2514	static PPGMPOOLPAGE pgmPoolMonitorGetPageByGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pNewPage)
2515	{
2516	/*
2517	* Look up the GCPhys in the hash.
2518	*/
2519	RTGCPHYS GCPhys = pNewPage->GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK;
2520	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
2521	if (i == NIL_PGMPOOL_IDX)
2522	return NULL;
2523	do
2524	{
2525	PPGMPOOLPAGE pPage = &pPool->aPages[i];
2526	if ( pPage->GCPhys - GCPhys < PAGE_SIZE
2527	&& pPage != pNewPage)
2528	{
2529	switch (pPage->enmKind)
2530	{
2531	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2532	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2533	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2534	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2535	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2536	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2537	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2538	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2539	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2540	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2541	case PGMPOOLKIND_64BIT_PML4:
2542	case PGMPOOLKIND_32BIT_PD:
2543	case PGMPOOLKIND_PAE_PDPT:
2544	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
2545	case PGMPOOLKIND_EPT_PT_FOR_EPT_PT:
2546	case PGMPOOLKIND_EPT_PD_FOR_EPT_PD:
2547	case PGMPOOLKIND_EPT_PDPT_FOR_EPT_PDPT:
2548	#endif
2549	{
2550	/* find the head */
2551	while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
2552	{
2553	Assert(pPage->iMonitoredPrev != pPage->idx);
2554	pPage = &pPool->aPages[pPage->iMonitoredPrev];
2555	}
2556	return pPage;
2557	}
2558
2559	/* ignore, no monitoring. */
2560	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2561	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2562	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2563	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2564	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2565	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2566	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2567	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2568	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2569	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2570	case PGMPOOLKIND_ROOT_NESTED:
2571	case PGMPOOLKIND_PAE_PD_PHYS:
2572	case PGMPOOLKIND_PAE_PDPT_PHYS:
2573	case PGMPOOLKIND_32BIT_PD_PHYS:
2574	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2575	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
2576	case PGMPOOLKIND_EPT_PT_FOR_EPT_2MB:
2577	case PGMPOOLKIND_EPT_PML4_FOR_EPT_PML4:
2578	#endif
2579	break;
2580	default:
2581	AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
2582	}
2583	}
2584
2585	/* next */
2586	i = pPage->iNext;
2587	} while (i != NIL_PGMPOOL_IDX);
2588	return NULL;
2589	}
2590
2591
2592	/**
2593	* Enabled write monitoring of a guest page.
2594	*
2595	* @returns VBox status code.
2596	* @retval VINF_SUCCESS on success.
2597	* @param pPool The pool.
2598	* @param pPage The cached page.
2599	*/
2600	static int pgmPoolMonitorInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2601	{
2602	LogFlow(("pgmPoolMonitorInsert %RGp\n", pPage->GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK));
2603
2604	/*
2605	* Filter out the relevant kinds.
2606	*/
2607	switch (pPage->enmKind)
2608	{
2609	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2610	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2611	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2612	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2613	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2614	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2615	case PGMPOOLKIND_64BIT_PML4:
2616	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2617	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2618	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2619	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2620	case PGMPOOLKIND_32BIT_PD:
2621	case PGMPOOLKIND_PAE_PDPT:
2622	break;
2623
2624	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2625	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2626	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2627	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2628	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2629	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2630	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2631	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2632	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2633	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2634	case PGMPOOLKIND_ROOT_NESTED:
2635	/* Nothing to monitor here. */
2636	return VINF_SUCCESS;
2637
2638	case PGMPOOLKIND_32BIT_PD_PHYS:
2639	case PGMPOOLKIND_PAE_PDPT_PHYS:
2640	case PGMPOOLKIND_PAE_PD_PHYS:
2641	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
2642	/* Nothing to monitor here. */
2643	return VINF_SUCCESS;
2644
2645	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
2646	case PGMPOOLKIND_EPT_PT_FOR_EPT_PT:
2647	case PGMPOOLKIND_EPT_PD_FOR_EPT_PD:
2648	case PGMPOOLKIND_EPT_PDPT_FOR_EPT_PDPT:
2649	break;
2650
2651	case PGMPOOLKIND_EPT_PT_FOR_EPT_2MB:
2652	case PGMPOOLKIND_EPT_PML4_FOR_EPT_PML4:
2653	/* Nothing to monitor here. */
2654	return VINF_SUCCESS;
2655	#endif
2656
2657	default:
2658	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
2659	}
2660
2661	/*
2662	* Install handler.
2663	*/
2664	int rc;
2665	PPGMPOOLPAGE pPageHead = pgmPoolMonitorGetPageByGCPhys(pPool, pPage);
2666	if (pPageHead)
2667	{
2668	Assert(pPageHead != pPage); Assert(pPageHead->iMonitoredNext != pPage->idx);
2669	Assert(pPageHead->iMonitoredPrev != pPage->idx);
2670
2671	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
2672	if (pPageHead->fDirty)
2673	pgmPoolFlushDirtyPage(pPool->CTX_SUFF(pVM), pPool, pPageHead->idxDirtyEntry, false /* do not remove */);
2674	#endif
2675
2676	pPage->iMonitoredPrev = pPageHead->idx;
2677	pPage->iMonitoredNext = pPageHead->iMonitoredNext;
2678	if (pPageHead->iMonitoredNext != NIL_PGMPOOL_IDX)
2679	pPool->aPages[pPageHead->iMonitoredNext].iMonitoredPrev = pPage->idx;
2680	pPageHead->iMonitoredNext = pPage->idx;
2681	rc = VINF_SUCCESS;
2682	if (PGMPOOL_PAGE_IS_NESTED(pPage))
2683	Log7Func(("Adding to monitoring list GCPhysPage=%RGp\n", pPage->GCPhys));
2684	}
2685	else
2686	{
2687	if (PGMPOOL_PAGE_IS_NESTED(pPage))
2688	Log7Func(("Started monitoring GCPhysPage=%RGp HCPhys=%RHp enmKind=%s\n", pPage->GCPhys, pPage->Core.Key, pgmPoolPoolKindToStr(pPage->enmKind)));
2689
2690	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX); Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
2691	PVMCC pVM = pPool->CTX_SUFF(pVM);
2692	const RTGCPHYS GCPhysPage = pPage->GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK;
2693	rc = PGMHandlerPhysicalRegister(pVM, GCPhysPage, GCPhysPage + PAGE_OFFSET_MASK, pPool->hAccessHandlerType,
2694	pPage - &pPool->aPages[0], NIL_RTR3PTR /pszDesc/);
2695	/** @todo we should probably deal with out-of-memory conditions here, but for now increasing
2696	* the heap size should suffice. */
2697	AssertFatalMsgRC(rc, ("PGMHandlerPhysicalRegisterEx %RGp failed with %Rrc\n", GCPhysPage, rc));
2698	PVMCPU pVCpu = VMMGetCpu(pVM);
2699	AssertFatalMsg(!(pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3), ("fSyncFlags=%x syncff=%d\n", pVCpu->pgm.s.fSyncFlags, VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3)));
2700	}
2701	pPage->fMonitored = true;
2702	return rc;
2703	}
2704
2705
2706	/**
2707	* Disables write monitoring of a guest page.
2708	*
2709	* @returns VBox status code.
2710	* @retval VINF_SUCCESS on success.
2711	* @param pPool The pool.
2712	* @param pPage The cached page.
2713	*/
2714	static int pgmPoolMonitorFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2715	{
2716	/*
2717	* Filter out the relevant kinds.
2718	*/
2719	switch (pPage->enmKind)
2720	{
2721	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
2722	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
2723	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
2724	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
2725	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
2726	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
2727	case PGMPOOLKIND_64BIT_PML4:
2728	case PGMPOOLKIND_32BIT_PD:
2729	case PGMPOOLKIND_PAE_PDPT:
2730	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
2731	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
2732	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
2733	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
2734	break;
2735
2736	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
2737	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
2738	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
2739	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
2740	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
2741	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
2742	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
2743	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
2744	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
2745	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
2746	case PGMPOOLKIND_ROOT_NESTED:
2747	case PGMPOOLKIND_PAE_PD_PHYS:
2748	case PGMPOOLKIND_PAE_PDPT_PHYS:
2749	case PGMPOOLKIND_32BIT_PD_PHYS:
2750	/* Nothing to monitor here. */
2751	Assert(!pPage->fMonitored);
2752	return VINF_SUCCESS;
2753
2754	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
2755	case PGMPOOLKIND_EPT_PT_FOR_EPT_PT:
2756	case PGMPOOLKIND_EPT_PD_FOR_EPT_PD:
2757	case PGMPOOLKIND_EPT_PDPT_FOR_EPT_PDPT:
2758	break;
2759
2760	case PGMPOOLKIND_EPT_PT_FOR_EPT_2MB:
2761	case PGMPOOLKIND_EPT_PML4_FOR_EPT_PML4:
2762	/* Nothing to monitor here. */
2763	Assert(!pPage->fMonitored);
2764	return VINF_SUCCESS;
2765	#endif
2766
2767	default:
2768	AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
2769	}
2770	Assert(pPage->fMonitored);
2771
2772	/*
2773	* Remove the page from the monitored list or uninstall it if last.
2774	*/
2775	const PVMCC pVM = pPool->CTX_SUFF(pVM);
2776	int rc;
2777	if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
2778	\|\| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
2779	{
2780	if (pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
2781	{
2782	PPGMPOOLPAGE pNewHead = &pPool->aPages[pPage->iMonitoredNext];
2783	pNewHead->iMonitoredPrev = NIL_PGMPOOL_IDX;
2784	rc = PGMHandlerPhysicalChangeUserArg(pVM, pPage->GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK, pPage->iMonitoredNext);
2785
2786	AssertFatalRCSuccess(rc);
2787	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
2788	}
2789	else
2790	{
2791	pPool->aPages[pPage->iMonitoredPrev].iMonitoredNext = pPage->iMonitoredNext;
2792	if (pPage->iMonitoredNext != NIL_PGMPOOL_IDX)
2793	{
2794	pPool->aPages[pPage->iMonitoredNext].iMonitoredPrev = pPage->iMonitoredPrev;
2795	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
2796	}
2797	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
2798	rc = VINF_SUCCESS;
2799	}
2800	}
2801	else
2802	{
2803	rc = PGMHandlerPhysicalDeregister(pVM, pPage->GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK);
2804	AssertFatalRC(rc);
2805	PVMCPU pVCpu = VMMGetCpu(pVM);
2806	AssertFatalMsg(!(pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL) \|\| VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3),
2807	("%#x %#x\n", pVCpu->pgm.s.fSyncFlags, pVM->fGlobalForcedActions));
2808	}
2809	pPage->fMonitored = false;
2810
2811	/*
2812	* Remove it from the list of modified pages (if in it).
2813	*/
2814	pgmPoolMonitorModifiedRemove(pPool, pPage);
2815
2816	if (PGMPOOL_PAGE_IS_NESTED(pPage))
2817	Log7Func(("Stopped monitoring %RGp\n", pPage->GCPhys));
2818
2819	return rc;
2820	}
2821
2822
2823	/**
2824	* Inserts the page into the list of modified pages.
2825	*
2826	* @param pPool The pool.
2827	* @param pPage The page.
2828	*/
2829	void pgmPoolMonitorModifiedInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2830	{
2831	Log3(("pgmPoolMonitorModifiedInsert: idx=%d\n", pPage->idx));
2832	AssertMsg( pPage->iModifiedNext == NIL_PGMPOOL_IDX
2833	&& pPage->iModifiedPrev == NIL_PGMPOOL_IDX
2834	&& pPool->iModifiedHead != pPage->idx,
2835	("Next=%d Prev=%d idx=%d cModifications=%d Head=%d cModifiedPages=%d\n",
2836	pPage->iModifiedNext, pPage->iModifiedPrev, pPage->idx, pPage->cModifications,
2837	pPool->iModifiedHead, pPool->cModifiedPages));
2838
2839	pPage->iModifiedNext = pPool->iModifiedHead;
2840	if (pPool->iModifiedHead != NIL_PGMPOOL_IDX)
2841	pPool->aPages[pPool->iModifiedHead].iModifiedPrev = pPage->idx;
2842	pPool->iModifiedHead = pPage->idx;
2843	pPool->cModifiedPages++;
2844	#ifdef VBOX_WITH_STATISTICS
2845	if (pPool->cModifiedPages > pPool->cModifiedPagesHigh)
2846	pPool->cModifiedPagesHigh = pPool->cModifiedPages;
2847	#endif
2848	}
2849
2850
2851	/**
2852	* Removes the page from the list of modified pages and resets the
2853	* modification counter.
2854	*
2855	* @param pPool The pool.
2856	* @param pPage The page which is believed to be in the list of modified pages.
2857	*/
2858	static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
2859	{
2860	Log3(("pgmPoolMonitorModifiedRemove: idx=%d cModifications=%d\n", pPage->idx, pPage->cModifications));
2861	if (pPool->iModifiedHead == pPage->idx)
2862	{
2863	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
2864	pPool->iModifiedHead = pPage->iModifiedNext;
2865	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
2866	{
2867	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = NIL_PGMPOOL_IDX;
2868	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2869	}
2870	pPool->cModifiedPages--;
2871	}
2872	else if (pPage->iModifiedPrev != NIL_PGMPOOL_IDX)
2873	{
2874	pPool->aPages[pPage->iModifiedPrev].iModifiedNext = pPage->iModifiedNext;
2875	if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
2876	{
2877	pPool->aPages[pPage->iModifiedNext].iModifiedPrev = pPage->iModifiedPrev;
2878	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2879	}
2880	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2881	pPool->cModifiedPages--;
2882	}
2883	else
2884	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
2885	pPage->cModifications = 0;
2886	}
2887
2888
2889	/**
2890	* Zaps the list of modified pages, resetting their modification counters in the process.
2891	*
2892	* @param pVM The cross context VM structure.
2893	*/
2894	static void pgmPoolMonitorModifiedClearAll(PVMCC pVM)
2895	{
2896	PGM_LOCK_VOID(pVM);
2897	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2898	LogFlow(("pgmPoolMonitorModifiedClearAll: cModifiedPages=%d\n", pPool->cModifiedPages));
2899
2900	unsigned cPages = 0; NOREF(cPages);
2901
2902	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
2903	pgmPoolResetDirtyPages(pVM);
2904	#endif
2905
2906	uint16_t idx = pPool->iModifiedHead;
2907	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
2908	while (idx != NIL_PGMPOOL_IDX)
2909	{
2910	PPGMPOOLPAGE pPage = &pPool->aPages[idx];
2911	idx = pPage->iModifiedNext;
2912	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
2913	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
2914	pPage->cModifications = 0;
2915	Assert(++cPages);
2916	}
2917	AssertMsg(cPages == pPool->cModifiedPages, ("%d != %d\n", cPages, pPool->cModifiedPages));
2918	pPool->cModifiedPages = 0;
2919	PGM_UNLOCK(pVM);
2920	}
2921
2922
2923	/**
2924	* Handle SyncCR3 pool tasks
2925	*
2926	* @returns VBox status code.
2927	* @retval VINF_SUCCESS if successfully added.
2928	* @retval VINF_PGM_SYNC_CR3 is it needs to be deferred to ring 3 (GC only)
2929	* @param pVCpu The cross context virtual CPU structure.
2930	* @remark Should only be used when monitoring is available, thus placed in
2931	* the PGMPOOL_WITH_MONITORING \#ifdef.
2932	*/
2933	int pgmPoolSyncCR3(PVMCPUCC pVCpu)
2934	{
2935	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2936	LogFlow(("pgmPoolSyncCR3 fSyncFlags=%x\n", pVCpu->pgm.s.fSyncFlags));
2937
2938	/*
2939	* When monitoring shadowed pages, we reset the modification counters on CR3 sync.
2940	* Occasionally we will have to clear all the shadow page tables because we wanted
2941	* to monitor a page which was mapped by too many shadowed page tables. This operation
2942	* sometimes referred to as a 'lightweight flush'.
2943	*/
2944	# ifdef IN_RING3 /* Don't flush in ring-0 or raw mode, it's taking too long. */
2945	if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
2946	pgmR3PoolClearAll(pVM, false /fFlushRemTlb/);
2947	# else /* !IN_RING3 */
2948	if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
2949	{
2950	Log(("SyncCR3: PGM_SYNC_CLEAR_PGM_POOL is set -> VINF_PGM_SYNC_CR3\n"));
2951	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
2952
2953	/* Make sure all other VCPUs return to ring 3. */
2954	if (pVM->cCpus > 1)
2955	{
2956	VM_FF_SET(pVM, VM_FF_PGM_POOL_FLUSH_PENDING);
2957	PGM_INVL_ALL_VCPU_TLBS(pVM);
2958	}
2959	return VINF_PGM_SYNC_CR3;
2960	}
2961	# endif /* !IN_RING3 */
2962	else
2963	{
2964	pgmPoolMonitorModifiedClearAll(pVM);
2965
2966	/* pgmPoolMonitorModifiedClearAll can cause a pgm pool flush (dirty page clearing), so make sure we handle this! */
2967	if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
2968	{
2969	Log(("pgmPoolMonitorModifiedClearAll caused a pgm flush -> call pgmPoolSyncCR3 again!\n"));
2970	return pgmPoolSyncCR3(pVCpu);
2971	}
2972	}
2973	return VINF_SUCCESS;
2974	}
2975
2976
2977	/**
2978	* Frees up at least one user entry.
2979	*
2980	* @returns VBox status code.
2981	* @retval VINF_SUCCESS if successfully added.
2982	*
2983	* @param pPool The pool.
2984	* @param iUser The user index.
2985	*/
2986	static int pgmPoolTrackFreeOneUser(PPGMPOOL pPool, uint16_t iUser)
2987	{
2988	STAM_COUNTER_INC(&pPool->StatTrackFreeUpOneUser);
2989	/*
2990	* Just free cached pages in a braindead fashion.
2991	*/
2992	/** @todo walk the age list backwards and free the first with usage. */
2993	int rc = VINF_SUCCESS;
2994	do
2995	{
2996	int rc2 = pgmPoolCacheFreeOne(pPool, iUser);
2997	if (RT_FAILURE(rc2) && rc == VINF_SUCCESS)
2998	rc = rc2;
2999	} while (pPool->iUserFreeHead == NIL_PGMPOOL_USER_INDEX);
3000	return rc;
3001	}
3002
3003
3004	/**
3005	* Inserts a page into the cache.
3006	*
3007	* This will create user node for the page, insert it into the GCPhys
3008	* hash, and insert it into the age list.
3009	*
3010	* @returns VBox status code.
3011	* @retval VINF_SUCCESS if successfully added.
3012	*
3013	* @param pPool The pool.
3014	* @param pPage The cached page.
3015	* @param GCPhys The GC physical address of the page we're gonna shadow.
3016	* @param iUser The user index.
3017	* @param iUserTable The user table index.
3018	*/
3019	DECLINLINE(int) pgmPoolTrackInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhys, uint16_t iUser, uint32_t iUserTable)
3020	{
3021	int rc = VINF_SUCCESS;
3022	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
3023
3024	LogFlow(("pgmPoolTrackInsert GCPhys=%RGp iUser=%d iUserTable=%x\n", GCPhys, iUser, iUserTable)); RT_NOREF_PV(GCPhys);
3025
3026	if (iUser != NIL_PGMPOOL_IDX)
3027	{
3028	#ifdef VBOX_STRICT
3029	/*
3030	* Check that the entry doesn't already exists.
3031	*/
3032	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
3033	{
3034	uint16_t i = pPage->iUserHead;
3035	do
3036	{
3037	Assert(i < pPool->cMaxUsers);
3038	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
3039	i = paUsers[i].iNext;
3040	} while (i != NIL_PGMPOOL_USER_INDEX);
3041	}
3042	#endif
3043
3044	/*
3045	* Find free a user node.
3046	*/
3047	uint16_t i = pPool->iUserFreeHead;
3048	if (i == NIL_PGMPOOL_USER_INDEX)
3049	{
3050	rc = pgmPoolTrackFreeOneUser(pPool, iUser);
3051	if (RT_FAILURE(rc))
3052	return rc;
3053	i = pPool->iUserFreeHead;
3054	}
3055
3056	/*
3057	* Unlink the user node from the free list,
3058	* initialize and insert it into the user list.
3059	*/
3060	pPool->iUserFreeHead = paUsers[i].iNext;
3061	paUsers[i].iNext = NIL_PGMPOOL_USER_INDEX;
3062	paUsers[i].iUser = iUser;
3063	paUsers[i].iUserTable = iUserTable;
3064	pPage->iUserHead = i;
3065	}
3066	else
3067	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
3068
3069
3070	/*
3071	* Insert into cache and enable monitoring of the guest page if enabled.
3072	*
3073	* Until we implement caching of all levels, including the CR3 one, we'll
3074	* have to make sure we don't try monitor & cache any recursive reuse of
3075	* a monitored CR3 page. Because all windows versions are doing this we'll
3076	* have to be able to do combined access monitoring, CR3 + PT and
3077	* PD + PT (guest PAE).
3078	*
3079	* Update:
3080	* We're now cooperating with the CR3 monitor if an uncachable page is found.
3081	*/
3082	const bool fCanBeMonitored = true;
3083	pgmPoolCacheInsert(pPool, pPage, fCanBeMonitored); /* This can be expanded. */
3084	if (fCanBeMonitored)
3085	{
3086	rc = pgmPoolMonitorInsert(pPool, pPage);
3087	AssertRC(rc);
3088	}
3089	return rc;
3090	}
3091
3092
3093	/**
3094	* Adds a user reference to a page.
3095	*
3096	* This will move the page to the head of the
3097	*
3098	* @returns VBox status code.
3099	* @retval VINF_SUCCESS if successfully added.
3100	*
3101	* @param pPool The pool.
3102	* @param pPage The cached page.
3103	* @param iUser The user index.
3104	* @param iUserTable The user table.
3105	*/
3106	static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
3107	{
3108	Log3(("pgmPoolTrackAddUser: GCPhys=%RGp iUser=%x iUserTable=%x\n", pPage->GCPhys, iUser, iUserTable));
3109	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
3110	Assert(iUser != NIL_PGMPOOL_IDX);
3111
3112	# ifdef VBOX_STRICT
3113	/*
3114	* Check that the entry doesn't already exists. We only allow multiple
3115	* users of top-level paging structures (SHW_POOL_ROOT_IDX).
3116	*/
3117	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
3118	{
3119	uint16_t i = pPage->iUserHead;
3120	do
3121	{
3122	Assert(i < pPool->cMaxUsers);
3123	/** @todo this assertion looks odd... Shouldn't it be && here? */
3124	AssertMsg(paUsers[i].iUser != iUser \|\| paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
3125	i = paUsers[i].iNext;
3126	} while (i != NIL_PGMPOOL_USER_INDEX);
3127	}
3128	# endif
3129
3130	/*
3131	* Allocate a user node.
3132	*/
3133	uint16_t i = pPool->iUserFreeHead;
3134	if (i == NIL_PGMPOOL_USER_INDEX)
3135	{
3136	int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
3137	if (RT_FAILURE(rc))
3138	return rc;
3139	i = pPool->iUserFreeHead;
3140	}
3141	pPool->iUserFreeHead = paUsers[i].iNext;
3142
3143	/*
3144	* Initialize the user node and insert it.
3145	*/
3146	paUsers[i].iNext = pPage->iUserHead;
3147	paUsers[i].iUser = iUser;
3148	paUsers[i].iUserTable = iUserTable;
3149	pPage->iUserHead = i;
3150
3151	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
3152	if (pPage->fDirty)
3153	pgmPoolFlushDirtyPage(pPool->CTX_SUFF(pVM), pPool, pPage->idxDirtyEntry, false /* do not remove */);
3154	# endif
3155
3156	/*
3157	* Tell the cache to update its replacement stats for this page.
3158	*/
3159	pgmPoolCacheUsed(pPool, pPage);
3160	return VINF_SUCCESS;
3161	}
3162
3163
3164	/**
3165	* Frees a user record associated with a page.
3166	*
3167	* This does not clear the entry in the user table, it simply replaces the
3168	* user record to the chain of free records.
3169	*
3170	* @param pPool The pool.
3171	* @param pPage The shadow page.
3172	* @param iUser The shadow page pool index of the user table.
3173	* @param iUserTable The index into the user table (shadowed).
3174	*
3175	* @remarks Don't call this for root pages.
3176	*/
3177	static void pgmPoolTrackFreeUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
3178	{
3179	Log3(("pgmPoolTrackFreeUser %RGp %x %x\n", pPage->GCPhys, iUser, iUserTable));
3180	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
3181	Assert(iUser != NIL_PGMPOOL_IDX);
3182
3183	/*
3184	* Unlink and free the specified user entry.
3185	*/
3186
3187	/* Special: For PAE and 32-bit paging, there is usually no more than one user. */
3188	uint16_t i = pPage->iUserHead;
3189	if ( i != NIL_PGMPOOL_USER_INDEX
3190	&& paUsers[i].iUser == iUser
3191	&& paUsers[i].iUserTable == iUserTable)
3192	{
3193	pPage->iUserHead = paUsers[i].iNext;
3194
3195	paUsers[i].iUser = NIL_PGMPOOL_IDX;
3196	paUsers[i].iNext = pPool->iUserFreeHead;
3197	pPool->iUserFreeHead = i;
3198	return;
3199	}
3200
3201	/* General: Linear search. */
3202	uint16_t iPrev = NIL_PGMPOOL_USER_INDEX;
3203	while (i != NIL_PGMPOOL_USER_INDEX)
3204	{
3205	if ( paUsers[i].iUser == iUser
3206	&& paUsers[i].iUserTable == iUserTable)
3207	{
3208	if (iPrev != NIL_PGMPOOL_USER_INDEX)
3209	paUsers[iPrev].iNext = paUsers[i].iNext;
3210	else
3211	pPage->iUserHead = paUsers[i].iNext;
3212
3213	paUsers[i].iUser = NIL_PGMPOOL_IDX;
3214	paUsers[i].iNext = pPool->iUserFreeHead;
3215	pPool->iUserFreeHead = i;
3216	return;
3217	}
3218	iPrev = i;
3219	i = paUsers[i].iNext;
3220	}
3221
3222	/* Fatal: didn't find it */
3223	AssertFatalMsgFailed(("Didn't find the user entry! iUser=%d iUserTable=%#x GCPhys=%RGp\n",
3224	iUser, iUserTable, pPage->GCPhys));
3225	}
3226
3227
3228	#if 0 /* unused */
3229	/**
3230	* Gets the entry size of a shadow table.
3231	*
3232	* @param enmKind The kind of page.
3233	*
3234	* @returns The size of the entry in bytes. That is, 4 or 8.
3235	* @returns If the kind is not for a table, an assertion is raised and 0 is
3236	* returned.
3237	*/
3238	DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind)
3239	{
3240	switch (enmKind)
3241	{
3242	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3243	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3244	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3245	case PGMPOOLKIND_32BIT_PD:
3246	case PGMPOOLKIND_32BIT_PD_PHYS:
3247	return 4;
3248
3249	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3250	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3251	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3252	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3253	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3254	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3255	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3256	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3257	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3258	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3259	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3260	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3261	case PGMPOOLKIND_64BIT_PML4:
3262	case PGMPOOLKIND_PAE_PDPT:
3263	case PGMPOOLKIND_ROOT_NESTED:
3264	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3265	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3266	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3267	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3268	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3269	case PGMPOOLKIND_PAE_PD_PHYS:
3270	case PGMPOOLKIND_PAE_PDPT_PHYS:
3271	return 8;
3272
3273	default:
3274	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
3275	}
3276	}
3277	#endif /* unused */
3278
3279	#if 0 /* unused */
3280	/**
3281	* Gets the entry size of a guest table.
3282	*
3283	* @param enmKind The kind of page.
3284	*
3285	* @returns The size of the entry in bytes. That is, 0, 4 or 8.
3286	* @returns If the kind is not for a table, an assertion is raised and 0 is
3287	* returned.
3288	*/
3289	DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind)
3290	{
3291	switch (enmKind)
3292	{
3293	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3294	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3295	case PGMPOOLKIND_32BIT_PD:
3296	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3297	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3298	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3299	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3300	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3301	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3302	return 4;
3303
3304	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3305	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3306	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3307	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3308	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3309	case PGMPOOLKIND_64BIT_PML4:
3310	case PGMPOOLKIND_PAE_PDPT:
3311	return 8;
3312
3313	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3314	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3315	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
3316	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
3317	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
3318	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3319	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3320	case PGMPOOLKIND_ROOT_NESTED:
3321	case PGMPOOLKIND_PAE_PD_PHYS:
3322	case PGMPOOLKIND_PAE_PDPT_PHYS:
3323	case PGMPOOLKIND_32BIT_PD_PHYS:
3324	/** @todo can we return 0? (nobody is calling this...) */
3325	AssertFailed();
3326	return 0;
3327
3328	default:
3329	AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
3330	}
3331	}
3332	#endif /* unused */
3333
3334
3335	/**
3336	* Checks one shadow page table entry for a mapping of a physical page.
3337	*
3338	* @returns true / false indicating removal of all relevant PTEs
3339	*
3340	* @param pVM The cross context VM structure.
3341	* @param pPhysPage The guest page in question.
3342	* @param fFlushPTEs Flush PTEs or allow them to be updated (e.g. in case of an RW bit change)
3343	* @param iShw The shadow page table.
3344	* @param iPte Page table entry or NIL_PGMPOOL_PHYSEXT_IDX_PTE if unknown
3345	*/
3346	static bool pgmPoolTrackFlushGCPhysPTInt(PVM pVM, PCPGMPAGE pPhysPage, bool fFlushPTEs, uint16_t iShw, uint16_t iPte)
3347	{
3348	LogFlow(("pgmPoolTrackFlushGCPhysPTInt: pPhysPage=%RHp iShw=%d iPte=%d\n", PGM_PAGE_GET_HCPHYS(pPhysPage), iShw, iPte));
3349	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3350	bool fRet = false;
3351
3352	/*
3353	* Assert sanity.
3354	*/
3355	Assert(iPte != NIL_PGMPOOL_PHYSEXT_IDX_PTE);
3356	AssertFatalMsg(iShw < pPool->cCurPages && iShw != NIL_PGMPOOL_IDX, ("iShw=%d\n", iShw));
3357	PPGMPOOLPAGE pPage = &pPool->aPages[iShw];
3358
3359	/*
3360	* Then, clear the actual mappings to the page in the shadow PT.
3361	*/
3362	switch (pPage->enmKind)
3363	{
3364	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3365	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3366	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3367	{
3368	const uint32_t u32 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
3369	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3370	uint32_t u32AndMask = 0;
3371	uint32_t u32OrMask = 0;
3372
3373	if (!fFlushPTEs)
3374	{
3375	/* Note! Disregarding the PGMPHYSHANDLER_F_NOT_IN_HM bit here. Should be harmless. */
3376	switch (PGM_PAGE_GET_HNDL_PHYS_STATE(pPhysPage))
3377	{
3378	case PGM_PAGE_HNDL_PHYS_STATE_NONE: /* No handler installed. */
3379	case PGM_PAGE_HNDL_PHYS_STATE_DISABLED: /* Monitoring is temporarily disabled. */
3380	u32OrMask = X86_PTE_RW;
3381	u32AndMask = UINT32_MAX;
3382	fRet = true;
3383	STAM_COUNTER_INC(&pPool->StatTrackFlushEntryKeep);
3384	break;
3385
3386	case PGM_PAGE_HNDL_PHYS_STATE_WRITE: /* Write access is monitored. */
3387	u32OrMask = 0;
3388	u32AndMask = ~X86_PTE_RW;
3389	fRet = true;
3390	STAM_COUNTER_INC(&pPool->StatTrackFlushEntryKeep);
3391	break;
3392	default:
3393	/* We will end up here when called with an "ALL" access handler. */
3394	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3395	break;
3396	}
3397	}
3398	else
3399	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3400
3401	/* Update the counter if we're removing references. */
3402	if (!u32AndMask)
3403	{
3404	Assert(pPage->cPresent);
3405	Assert(pPool->cPresent);
3406	pPage->cPresent--;
3407	pPool->cPresent--;
3408	}
3409
3410	if ((pPT->a[iPte].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
3411	{
3412	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX32\n", iPte, pPT->a[iPte]));
3413	X86PTE Pte;
3414	Pte.u = (pPT->a[iPte].u & u32AndMask) \| u32OrMask;
3415	if (Pte.u & PGM_PTFLAGS_TRACK_DIRTY)
3416	Pte.u &= ~(X86PGUINT)X86_PTE_RW; /* need to disallow writes when dirty bit tracking is still active. */
3417
3418	ASMAtomicWriteU32(&pPT->a[iPte].u, Pte.u);
3419	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);
3420	return fRet;
3421	}
3422	#ifdef LOG_ENABLED
3423	Log(("iFirstPresent=%d cPresent=%d\n", pPage->iFirstPresent, pPage->cPresent));
3424	for (unsigned i = 0, cFound = 0; i < RT_ELEMENTS(pPT->a); i++)
3425	if ((pPT->a[i].u & (X86_PTE_PG_MASK \| X86_PTE_P)) == u32)
3426	{
3427	Log(("i=%d cFound=%d\n", i, ++cFound));
3428	}
3429	#endif
3430	AssertFatalMsgFailed(("iFirstPresent=%d cPresent=%d u32=%RX32 poolkind=%x\n", pPage->iFirstPresent, pPage->cPresent, u32, pPage->enmKind));
3431	/PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);/
3432	break;
3433	}
3434
3435	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3436	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3437	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3438	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3439	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3440	case PGMPOOLKIND_EPT_PT_FOR_PHYS: /* physical mask the same as PAE; RW bit as well; be careful! */
3441	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
3442	case PGMPOOLKIND_EPT_PT_FOR_EPT_PT:
3443	# ifdef PGM_WITH_LARGE_PAGES
3444	case PGMPOOLKIND_EPT_PT_FOR_EPT_2MB:
3445	# endif
3446	#endif
3447	{
3448	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PTE_P;
3449	PPGMSHWPTPAE pPT = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3450	uint64_t u64OrMask = 0;
3451	uint64_t u64AndMask = 0;
3452
3453	if (!fFlushPTEs)
3454	{
3455	/* Note! Disregarding the PGMPHYSHANDLER_F_NOT_IN_HM bit here. Should be harmless. */
3456	switch (PGM_PAGE_GET_HNDL_PHYS_STATE(pPhysPage))
3457	{
3458	case PGM_PAGE_HNDL_PHYS_STATE_NONE: /* No handler installed. */
3459	case PGM_PAGE_HNDL_PHYS_STATE_DISABLED: /* Monitoring is temporarily disabled. */
3460	u64OrMask = X86_PTE_RW;
3461	u64AndMask = UINT64_MAX;
3462	fRet = true;
3463	STAM_COUNTER_INC(&pPool->StatTrackFlushEntryKeep);
3464	break;
3465
3466	case PGM_PAGE_HNDL_PHYS_STATE_WRITE: /* Write access is monitored. */
3467	u64OrMask = 0;
3468	u64AndMask = ~(uint64_t)X86_PTE_RW;
3469	fRet = true;
3470	STAM_COUNTER_INC(&pPool->StatTrackFlushEntryKeep);
3471	break;
3472
3473	default:
3474	/* We will end up here when called with an "ALL" access handler. */
3475	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3476	break;
3477	}
3478	}
3479	else
3480	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3481
3482	/* Update the counter if we're removing references. */
3483	if (!u64AndMask)
3484	{
3485	Assert(pPage->cPresent);
3486	Assert(pPool->cPresent);
3487	pPage->cPresent--;
3488	pPool->cPresent--;
3489	}
3490
3491	if ((PGMSHWPTEPAE_GET_U(pPT->a[iPte]) & (X86_PTE_PAE_PG_MASK \| X86_PTE_P \| X86_PTE_PAE_MBZ_MASK_NX)) == u64)
3492	{
3493	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64\n", iPte, PGMSHWPTEPAE_GET_LOG(pPT->a[iPte])));
3494	X86PTEPAE Pte;
3495	Pte.u = (PGMSHWPTEPAE_GET_U(pPT->a[iPte]) & u64AndMask) \| u64OrMask;
3496	if (Pte.u & PGM_PTFLAGS_TRACK_DIRTY)
3497	Pte.u &= ~(X86PGPAEUINT)X86_PTE_RW; /* need to disallow writes when dirty bit tracking is still active. */
3498
3499	PGMSHWPTEPAE_ATOMIC_SET(pPT->a[iPte], Pte.u);
3500	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);
3501	return fRet;
3502	}
3503	#ifdef LOG_ENABLED
3504	Log(("iFirstPresent=%d cPresent=%d\n", pPage->iFirstPresent, pPage->cPresent));
3505	Log(("Found %RX64 expected %RX64\n", PGMSHWPTEPAE_GET_U(pPT->a[iPte]) & (X86_PTE_PAE_PG_MASK \| X86_PTE_P \| X86_PTE_PAE_MBZ_MASK_NX), u64));
3506	for (unsigned i = 0, cFound = 0; i < RT_ELEMENTS(pPT->a); i++)
3507	if ((PGMSHWPTEPAE_GET_U(pPT->a[i]) & (X86_PTE_PAE_PG_MASK \| X86_PTE_P \| X86_PTE_PAE_MBZ_MASK_NX)) == u64)
3508	Log(("i=%d cFound=%d\n", i, ++cFound));
3509	#endif
3510	AssertFatalMsgFailed(("iFirstPresent=%d cPresent=%d u64=%RX64 poolkind=%x iPte=%d PT=%RX64\n", pPage->iFirstPresent, pPage->cPresent, u64, pPage->enmKind, iPte, PGMSHWPTEPAE_GET_LOG(pPT->a[iPte])));
3511	/PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);/
3512	break;
3513	}
3514
3515	#ifdef PGM_WITH_LARGE_PAGES
3516	/* Large page case only. */
3517	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
3518	{
3519	Assert(pVM->pgm.s.fNestedPaging);
3520
3521	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PDE4M_P \| X86_PDE4M_PS;
3522	PEPTPD pPD = (PEPTPD)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3523
3524	if ((pPD->a[iPte].u & (EPT_PDE2M_PG_MASK \| X86_PDE4M_P \| X86_PDE4M_PS)) == u64)
3525	{
3526	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pde=%RX64\n", iPte, pPD->a[iPte]));
3527	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3528	pPD->a[iPte].u = 0;
3529	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPD);
3530
3531	/* Update the counter as we're removing references. */
3532	Assert(pPage->cPresent);
3533	Assert(pPool->cPresent);
3534	pPage->cPresent--;
3535	pPool->cPresent--;
3536
3537	return fRet;
3538	}
3539	# ifdef LOG_ENABLED
3540	Log(("iFirstPresent=%d cPresent=%d\n", pPage->iFirstPresent, pPage->cPresent));
3541	for (unsigned i = 0, cFound = 0; i < RT_ELEMENTS(pPD->a); i++)
3542	if ((pPD->a[i].u & (EPT_PDE2M_PG_MASK \| X86_PDE4M_P \| X86_PDE4M_PS)) == u64)
3543	Log(("i=%d cFound=%d\n", i, ++cFound));
3544	# endif
3545	AssertFatalMsgFailed(("iFirstPresent=%d cPresent=%d\n", pPage->iFirstPresent, pPage->cPresent));
3546	/PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPD);/
3547	break;
3548	}
3549
3550	/* AMD-V nested paging / /* @todo merge with EPT as we only check the parts that are identical. */
3551	case PGMPOOLKIND_PAE_PD_PHYS:
3552	{
3553	Assert(pVM->pgm.s.fNestedPaging);
3554
3555	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) \| X86_PDE4M_P \| X86_PDE4M_PS;
3556	PX86PDPAE pPD = (PX86PDPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3557
3558	if ((pPD->a[iPte].u & (X86_PDE2M_PAE_PG_MASK \| X86_PDE4M_P \| X86_PDE4M_PS)) == u64)
3559	{
3560	Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pde=%RX64\n", iPte, pPD->a[iPte]));
3561	STAM_COUNTER_INC(&pPool->StatTrackFlushEntry);
3562	pPD->a[iPte].u = 0;
3563	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPD);
3564
3565	/* Update the counter as we're removing references. */
3566	Assert(pPage->cPresent);
3567	Assert(pPool->cPresent);
3568	pPage->cPresent--;
3569	pPool->cPresent--;
3570	return fRet;
3571	}
3572	# ifdef LOG_ENABLED
3573	Log(("iFirstPresent=%d cPresent=%d\n", pPage->iFirstPresent, pPage->cPresent));
3574	for (unsigned i = 0, cFound = 0; i < RT_ELEMENTS(pPD->a); i++)
3575	if ((pPD->a[i].u & (X86_PDE2M_PAE_PG_MASK \| X86_PDE4M_P \| X86_PDE4M_PS)) == u64)
3576	Log(("i=%d cFound=%d\n", i, ++cFound));
3577	# endif
3578	AssertFatalMsgFailed(("iFirstPresent=%d cPresent=%d\n", pPage->iFirstPresent, pPage->cPresent));
3579	/PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPD);/
3580	break;
3581	}
3582	#endif /* PGM_WITH_LARGE_PAGES */
3583
3584	default:
3585	AssertFatalMsgFailed(("enmKind=%d iShw=%d\n", pPage->enmKind, iShw));
3586	}
3587
3588	/* not reached. */
3589	#ifndef _MSC_VER
3590	return fRet;
3591	#endif
3592	}
3593
3594
3595	/**
3596	* Scans one shadow page table for mappings of a physical page.
3597	*
3598	* @param pVM The cross context VM structure.
3599	* @param pPhysPage The guest page in question.
3600	* @param fFlushPTEs Flush PTEs or allow them to be updated (e.g. in case of an RW bit change)
3601	* @param iShw The shadow page table.
3602	*/
3603	static void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, bool fFlushPTEs, uint16_t iShw)
3604	{
3605	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
3606
3607	/* We should only come here with when there's only one reference to this physical page. */
3608	Assert(PGMPOOL_TD_GET_CREFS(PGM_PAGE_GET_TRACKING(pPhysPage)) == 1);
3609
3610	Log2(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%RHp iShw=%d\n", PGM_PAGE_GET_HCPHYS(pPhysPage), iShw));
3611	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPT, f);
3612	bool fKeptPTEs = pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, fFlushPTEs, iShw, PGM_PAGE_GET_PTE_INDEX(pPhysPage));
3613	if (!fKeptPTEs)
3614	PGM_PAGE_SET_TRACKING(pVM, pPhysPage, 0);
3615	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPT, f);
3616	}
3617
3618
3619	/**
3620	* Flushes a list of shadow page tables mapping the same physical page.
3621	*
3622	* @param pVM The cross context VM structure.
3623	* @param pPhysPage The guest page in question.
3624	* @param fFlushPTEs Flush PTEs or allow them to be updated (e.g. in case of an RW bit change)
3625	* @param iPhysExt The physical cross reference extent list to flush.
3626	*/
3627	static void pgmPoolTrackFlushGCPhysPTs(PVMCC pVM, PPGMPAGE pPhysPage, bool fFlushPTEs, uint16_t iPhysExt)
3628	{
3629	PGM_LOCK_ASSERT_OWNER(pVM);
3630	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3631	bool fKeepList = false;
3632
3633	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTs, f);
3634	Log2(("pgmPoolTrackFlushGCPhysPTs: pPhysPage=%RHp iPhysExt=%u\n", PGM_PAGE_GET_HCPHYS(pPhysPage), iPhysExt));
3635
3636	const uint16_t iPhysExtStart = iPhysExt;
3637	PPGMPOOLPHYSEXT pPhysExt;
3638	do
3639	{
3640	Assert(iPhysExt < pPool->cMaxPhysExts);
3641	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
3642	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
3643	{
3644	if (pPhysExt->aidx[i] != NIL_PGMPOOL_IDX)
3645	{
3646	bool fKeptPTEs = pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, fFlushPTEs, pPhysExt->aidx[i], pPhysExt->apte[i]);
3647	if (!fKeptPTEs)
3648	{
3649	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
3650	pPhysExt->apte[i] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
3651	}
3652	else
3653	fKeepList = true;
3654	}
3655	}
3656	/* next */
3657	iPhysExt = pPhysExt->iNext;
3658	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
3659
3660	if (!fKeepList)
3661	{
3662	/* insert the list into the free list and clear the ram range entry. */
3663	pPhysExt->iNext = pPool->iPhysExtFreeHead;
3664	pPool->iPhysExtFreeHead = iPhysExtStart;
3665	/* Invalidate the tracking data. */
3666	PGM_PAGE_SET_TRACKING(pVM, pPhysPage, 0);
3667	}
3668
3669	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTs, f);
3670	}
3671
3672
3673	/**
3674	* Flushes all shadow page table mappings of the given guest page.
3675	*
3676	* This is typically called when the host page backing the guest one has been
3677	* replaced or when the page protection was changed due to a guest access
3678	* caught by the monitoring.
3679	*
3680	* @returns VBox status code.
3681	* @retval VINF_SUCCESS if all references has been successfully cleared.
3682	* @retval VINF_PGM_SYNC_CR3 if we're better off with a CR3 sync and a page
3683	* pool cleaning. FF and sync flags are set.
3684	*
3685	* @param pVM The cross context VM structure.
3686	* @param GCPhysPage GC physical address of the page in question
3687	* @param pPhysPage The guest page in question.
3688	* @param fFlushPTEs Flush PTEs or allow them to be updated (e.g. in case of an RW bit change)
3689	* @param pfFlushTLBs This is set to @a true if the shadow TLBs should be
3690	* flushed, it is NOT touched if this isn't necessary.
3691	* The caller MUST initialized this to @a false.
3692	*/
3693	int pgmPoolTrackUpdateGCPhys(PVMCC pVM, RTGCPHYS GCPhysPage, PPGMPAGE pPhysPage, bool fFlushPTEs, bool *pfFlushTLBs)
3694	{
3695	PVMCPUCC pVCpu = VMMGetCpu(pVM);
3696	PGM_LOCK_VOID(pVM);
3697	int rc = VINF_SUCCESS;
3698
3699	#ifdef PGM_WITH_LARGE_PAGES
3700	/* Is this page part of a large page? */
3701	if (PGM_PAGE_GET_PDE_TYPE(pPhysPage) == PGM_PAGE_PDE_TYPE_PDE)
3702	{
3703	RTGCPHYS GCPhysBase = GCPhysPage & X86_PDE2M_PAE_PG_MASK;
3704	GCPhysPage &= X86_PDE_PAE_PG_MASK;
3705
3706	/* Fetch the large page base. */
3707	PPGMPAGE pLargePage;
3708	if (GCPhysBase != GCPhysPage)
3709	{
3710	pLargePage = pgmPhysGetPage(pVM, GCPhysBase);
3711	AssertFatal(pLargePage);
3712	}
3713	else
3714	pLargePage = pPhysPage;
3715
3716	Log(("pgmPoolTrackUpdateGCPhys: update large page PDE for %RGp (%RGp)\n", GCPhysBase, GCPhysPage));
3717
3718	if (PGM_PAGE_GET_PDE_TYPE(pLargePage) == PGM_PAGE_PDE_TYPE_PDE)
3719	{
3720	/* Mark the large page as disabled as we need to break it up to change a single page in the 2 MB range. */
3721	PGM_PAGE_SET_PDE_TYPE(pVM, pLargePage, PGM_PAGE_PDE_TYPE_PDE_DISABLED);
3722	pVM->pgm.s.cLargePagesDisabled++;
3723
3724	/* Update the base as that only that one has a reference and there's only one PDE to clear. */
3725	rc = pgmPoolTrackUpdateGCPhys(pVM, GCPhysBase, pLargePage, fFlushPTEs, pfFlushTLBs);
3726
3727	*pfFlushTLBs = true;
3728	PGM_UNLOCK(pVM);
3729	return rc;
3730	}
3731	}
3732	#else
3733	NOREF(GCPhysPage);
3734	#endif /* PGM_WITH_LARGE_PAGES */
3735
3736	const uint16_t u16 = PGM_PAGE_GET_TRACKING(pPhysPage);
3737	if (u16)
3738	{
3739	/*
3740	* The zero page is currently screwing up the tracking and we'll
3741	* have to flush the whole shebang. Unless VBOX_WITH_NEW_LAZY_PAGE_ALLOC
3742	* is defined, zero pages won't normally be mapped. Some kind of solution
3743	* will be needed for this problem of course, but it will have to wait...
3744	*/
3745	if ( PGM_PAGE_IS_ZERO(pPhysPage)
3746	\|\| PGM_PAGE_IS_BALLOONED(pPhysPage))
3747	rc = VINF_PGM_GCPHYS_ALIASED;
3748	else
3749	{
3750	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
3751	{
3752	Assert(PGMPOOL_TD_GET_CREFS(u16) == 1);
3753	pgmPoolTrackFlushGCPhysPT(pVM,
3754	pPhysPage,
3755	fFlushPTEs,
3756	PGMPOOL_TD_GET_IDX(u16));
3757	}
3758	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
3759	pgmPoolTrackFlushGCPhysPTs(pVM, pPhysPage, fFlushPTEs, PGMPOOL_TD_GET_IDX(u16));
3760	else
3761	rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
3762	*pfFlushTLBs = true;
3763	}
3764	}
3765
3766	if (rc == VINF_PGM_GCPHYS_ALIASED)
3767	{
3768	pVCpu->pgm.s.fSyncFlags \|= PGM_SYNC_CLEAR_PGM_POOL;
3769	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
3770	rc = VINF_PGM_SYNC_CR3;
3771	}
3772	PGM_UNLOCK(pVM);
3773	return rc;
3774	}
3775
3776
3777	/**
3778	* Scans all shadow page tables for mappings of a physical page.
3779	*
3780	* This may be slow, but it's most likely more efficient than cleaning
3781	* out the entire page pool / cache.
3782	*
3783	* @returns VBox status code.
3784	* @retval VINF_SUCCESS if all references has been successfully cleared.
3785	* @retval VINF_PGM_GCPHYS_ALIASED if we're better off with a CR3 sync and
3786	* a page pool cleaning.
3787	*
3788	* @param pVM The cross context VM structure.
3789	* @param pPhysPage The guest page in question.
3790	*/
3791	int pgmPoolTrackFlushGCPhysPTsSlow(PVMCC pVM, PPGMPAGE pPhysPage)
3792	{
3793	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
3794	STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTsSlow, s);
3795	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: cUsedPages=%d cPresent=%d pPhysPage=%R[pgmpage]\n",
3796	pPool->cUsedPages, pPool->cPresent, pPhysPage));
3797
3798	/*
3799	* There is a limit to what makes sense.
3800	*/
3801	if ( pPool->cPresent > 1024
3802	&& pVM->cCpus == 1)
3803	{
3804	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
3805	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
3806	return VINF_PGM_GCPHYS_ALIASED;
3807	}
3808
3809	/*
3810	* Iterate all the pages until we've encountered all that in use.
3811	* This is simple but not quite optimal solution.
3812	*/
3813	const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage);
3814	unsigned cLeft = pPool->cUsedPages;
3815	unsigned iPage = pPool->cCurPages;
3816	while (--iPage >= PGMPOOL_IDX_FIRST)
3817	{
3818	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
3819	if ( pPage->GCPhys != NIL_RTGCPHYS
3820	&& pPage->cPresent)
3821	{
3822	Assert(!PGMPOOL_PAGE_IS_NESTED(pPage)); /* see if it hits */
3823	switch (pPage->enmKind)
3824	{
3825	/*
3826	* We only care about shadow page tables.
3827	*/
3828	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
3829	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3830	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
3831	{
3832	const uint32_t u32 = (uint32_t)u64;
3833	unsigned cPresent = pPage->cPresent;
3834	PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3835	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
3836	{
3837	const X86PGUINT uPte = pPT->a[i].u;
3838	if (uPte & X86_PTE_P)
3839	{
3840	if ((uPte & X86_PTE_PG_MASK) == u32)
3841	{
3842	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX32\n", iPage, i, pPT->a[i]));
3843	ASMAtomicWriteU32(&pPT->a[i].u, 0);
3844
3845	/* Update the counter as we're removing references. */
3846	Assert(pPage->cPresent);
3847	Assert(pPool->cPresent);
3848	pPage->cPresent--;
3849	pPool->cPresent--;
3850	}
3851	if (!--cPresent)
3852	break;
3853	}
3854	}
3855	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);
3856	break;
3857	}
3858
3859	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
3860	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
3861	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
3862	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3863	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
3864	{
3865	unsigned cPresent = pPage->cPresent;
3866	PPGMSHWPTPAE pPT = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3867	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
3868	if (PGMSHWPTEPAE_IS_P(pPT->a[i]))
3869	{
3870	if ((PGMSHWPTEPAE_GET_U(pPT->a[i]) & X86_PTE_PAE_PG_MASK) == u64)
3871	{
3872	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
3873	PGMSHWPTEPAE_ATOMIC_SET(pPT->a[i], 0); /// @todo why not atomic?
3874
3875	/* Update the counter as we're removing references. */
3876	Assert(pPage->cPresent);
3877	Assert(pPool->cPresent);
3878	pPage->cPresent--;
3879	pPool->cPresent--;
3880	}
3881	if (!--cPresent)
3882	break;
3883	}
3884	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);
3885	break;
3886	}
3887
3888	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
3889	{
3890	unsigned cPresent = pPage->cPresent;
3891	PEPTPT pPT = (PEPTPT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
3892	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
3893	{
3894	X86PGPAEUINT const uPte = pPT->a[i].u;
3895	if (uPte & EPT_E_READ)
3896	{
3897	if ((uPte & EPT_PTE_PG_MASK) == u64)
3898	{
3899	//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
3900	ASMAtomicWriteU64(&pPT->a[i].u, 0);
3901
3902	/* Update the counter as we're removing references. */
3903	Assert(pPage->cPresent);
3904	Assert(pPool->cPresent);
3905	pPage->cPresent--;
3906	pPool->cPresent--;
3907	}
3908	if (!--cPresent)
3909	break;
3910	}
3911	}
3912	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pPT);
3913	break;
3914	}
3915	}
3916
3917	if (!--cLeft)
3918	break;
3919	}
3920	}
3921
3922	PGM_PAGE_SET_TRACKING(pVM, pPhysPage, 0);
3923	STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
3924
3925	/*
3926	* There is a limit to what makes sense. The above search is very expensive, so force a pgm pool flush.
3927	*/
3928	if (pPool->cPresent > 1024)
3929	{
3930	LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
3931	return VINF_PGM_GCPHYS_ALIASED;
3932	}
3933
3934	return VINF_SUCCESS;
3935	}
3936
3937
3938	/**
3939	* Clears the user entry in a user table.
3940	*
3941	* This is used to remove all references to a page when flushing it.
3942	*/
3943	static void pgmPoolTrackClearPageUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PCPGMPOOLUSER pUser)
3944	{
3945	Assert(pUser->iUser != NIL_PGMPOOL_IDX);
3946	Assert(pUser->iUser < pPool->cCurPages);
3947	uint32_t iUserTable = pUser->iUserTable;
3948
3949	/*
3950	* Map the user page. Ignore references made by fictitious pages.
3951	*/
3952	PPGMPOOLPAGE pUserPage = &pPool->aPages[pUser->iUser];
3953	LogFlow(("pgmPoolTrackClearPageUser: clear %x in %s (%RGp) (flushing %s)\n", iUserTable, pgmPoolPoolKindToStr(pUserPage->enmKind), pUserPage->Core.Key, pgmPoolPoolKindToStr(pPage->enmKind)));
3954	union
3955	{
3956	uint64_t *pau64;
3957	uint32_t *pau32;
3958	} u;
3959	if (pUserPage->idx < PGMPOOL_IDX_FIRST)
3960	{
3961	Assert(!pUserPage->pvPageR3);
3962	return;
3963	}
3964	u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pUserPage);
3965
3966
3967	/* Safety precaution in case we change the paging for other modes too in the future. */
3968	Assert(!pgmPoolIsPageLocked(pPage)); RT_NOREF_PV(pPage);
3969
3970	#ifdef VBOX_STRICT
3971	/*
3972	* Some sanity checks.
3973	*/
3974	switch (pUserPage->enmKind)
3975	{
3976	case PGMPOOLKIND_32BIT_PD:
3977	case PGMPOOLKIND_32BIT_PD_PHYS:
3978	Assert(iUserTable < X86_PG_ENTRIES);
3979	break;
3980	case PGMPOOLKIND_PAE_PDPT:
3981	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
3982	case PGMPOOLKIND_PAE_PDPT_PHYS:
3983	Assert(iUserTable < 4);
3984	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
3985	break;
3986	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
3987	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
3988	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
3989	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
3990	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
3991	case PGMPOOLKIND_PAE_PD_PHYS:
3992	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3993	break;
3994	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
3995	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3996	break;
3997	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
3998	Assert(iUserTable < X86_PG_PAE_ENTRIES);
3999	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
4000	break;
4001	case PGMPOOLKIND_64BIT_PML4:
4002	Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
4003	/* GCPhys >> PAGE_SHIFT is the index here */
4004	break;
4005	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
4006	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
4007	Assert(iUserTable < X86_PG_PAE_ENTRIES);
4008	break;
4009
4010	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
4011	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
4012	Assert(iUserTable < X86_PG_PAE_ENTRIES);
4013	break;
4014
4015	case PGMPOOLKIND_ROOT_NESTED:
4016	Assert(iUserTable < X86_PG_PAE_ENTRIES);
4017	break;
4018
4019	# ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
4020	case PGMPOOLKIND_EPT_PT_FOR_EPT_PT:
4021	case PGMPOOLKIND_EPT_PT_FOR_EPT_2MB:
4022	case PGMPOOLKIND_EPT_PD_FOR_EPT_PD:
4023	case PGMPOOLKIND_EPT_PDPT_FOR_EPT_PDPT:
4024	case PGMPOOLKIND_EPT_PML4_FOR_EPT_PML4:
4025	Assert(iUserTable < EPT_PG_ENTRIES);
4026	break;
4027	# endif
4028
4029	default:
4030	AssertMsgFailed(("enmKind=%d GCPhys=%RGp\n", pUserPage->enmKind, pPage->GCPhys));
4031	break;
4032	}
4033	#endif /* VBOX_STRICT */
4034
4035	/*
4036	* Clear the entry in the user page.
4037	*/
4038	switch (pUserPage->enmKind)
4039	{
4040	/* 32-bit entries */
4041	case PGMPOOLKIND_32BIT_PD:
4042	case PGMPOOLKIND_32BIT_PD_PHYS:
4043	ASMAtomicWriteU32(&u.pau32[iUserTable], 0);
4044	break;
4045
4046	/* 64-bit entries */
4047	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
4048	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
4049	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
4050	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
4051	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
4052	case PGMPOOLKIND_PAE_PD_PHYS:
4053	case PGMPOOLKIND_PAE_PDPT_PHYS:
4054	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
4055	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
4056	case PGMPOOLKIND_64BIT_PML4:
4057	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
4058	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
4059	case PGMPOOLKIND_PAE_PDPT:
4060	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
4061	case PGMPOOLKIND_ROOT_NESTED:
4062	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
4063	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
4064	# ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
4065	case PGMPOOLKIND_EPT_PT_FOR_EPT_PT:
4066	case PGMPOOLKIND_EPT_PT_FOR_EPT_2MB:
4067	case PGMPOOLKIND_EPT_PD_FOR_EPT_PD:
4068	case PGMPOOLKIND_EPT_PDPT_FOR_EPT_PDPT:
4069	case PGMPOOLKIND_EPT_PML4_FOR_EPT_PML4:
4070	#endif
4071	ASMAtomicWriteU64(&u.pau64[iUserTable], 0);
4072	break;
4073
4074	default:
4075	AssertFatalMsgFailed(("enmKind=%d iUser=%d iUserTable=%#x\n", pUserPage->enmKind, pUser->iUser, pUser->iUserTable));
4076	}
4077	PGM_DYNMAP_UNUSED_HINT_VM(pPool->CTX_SUFF(pVM), u.pau64);
4078	}
4079
4080
4081	/**
4082	* Clears all users of a page.
4083	*/
4084	static void pgmPoolTrackClearPageUsers(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
4085	{
4086	/*
4087	* Free all the user records.
4088	*/
4089	LogFlow(("pgmPoolTrackClearPageUsers %RGp\n", pPage->GCPhys));
4090
4091	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
4092	uint16_t i = pPage->iUserHead;
4093	while (i != NIL_PGMPOOL_USER_INDEX)
4094	{
4095	/* Clear enter in user table. */
4096	pgmPoolTrackClearPageUser(pPool, pPage, &paUsers[i]);
4097
4098	/* Free it. */
4099	const uint16_t iNext = paUsers[i].iNext;
4100	paUsers[i].iUser = NIL_PGMPOOL_IDX;
4101	paUsers[i].iNext = pPool->iUserFreeHead;
4102	pPool->iUserFreeHead = i;
4103
4104	/* Next. */
4105	i = iNext;
4106	}
4107	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
4108	}
4109
4110
4111	/**
4112	* Allocates a new physical cross reference extent.
4113	*
4114	* @returns Pointer to the allocated extent on success. NULL if we're out of them.
4115	* @param pVM The cross context VM structure.
4116	* @param piPhysExt Where to store the phys ext index.
4117	*/
4118	PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVMCC pVM, uint16_t *piPhysExt)
4119	{
4120	PGM_LOCK_ASSERT_OWNER(pVM);
4121	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4122	uint16_t iPhysExt = pPool->iPhysExtFreeHead;
4123	if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
4124	{
4125	STAM_COUNTER_INC(&pPool->StamTrackPhysExtAllocFailures);
4126	return NULL;
4127	}
4128	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
4129	pPool->iPhysExtFreeHead = pPhysExt->iNext;
4130	pPhysExt->iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
4131	*piPhysExt = iPhysExt;
4132	return pPhysExt;
4133	}
4134
4135
4136	/**
4137	* Frees a physical cross reference extent.
4138	*
4139	* @param pVM The cross context VM structure.
4140	* @param iPhysExt The extent to free.
4141	*/
4142	void pgmPoolTrackPhysExtFree(PVMCC pVM, uint16_t iPhysExt)
4143	{
4144	PGM_LOCK_ASSERT_OWNER(pVM);
4145	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4146	Assert(iPhysExt < pPool->cMaxPhysExts);
4147	PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
4148	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
4149	{
4150	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
4151	pPhysExt->apte[i] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
4152	}
4153	pPhysExt->iNext = pPool->iPhysExtFreeHead;
4154	pPool->iPhysExtFreeHead = iPhysExt;
4155	}
4156
4157
4158	/**
4159	* Frees a physical cross reference extent.
4160	*
4161	* @param pVM The cross context VM structure.
4162	* @param iPhysExt The extent to free.
4163	*/
4164	void pgmPoolTrackPhysExtFreeList(PVMCC pVM, uint16_t iPhysExt)
4165	{
4166	PGM_LOCK_ASSERT_OWNER(pVM);
4167	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4168
4169	const uint16_t iPhysExtStart = iPhysExt;
4170	PPGMPOOLPHYSEXT pPhysExt;
4171	do
4172	{
4173	Assert(iPhysExt < pPool->cMaxPhysExts);
4174	pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
4175	for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
4176	{
4177	pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
4178	pPhysExt->apte[i] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
4179	}
4180
4181	/* next */
4182	iPhysExt = pPhysExt->iNext;
4183	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
4184
4185	pPhysExt->iNext = pPool->iPhysExtFreeHead;
4186	pPool->iPhysExtFreeHead = iPhysExtStart;
4187	}
4188
4189
4190	/**
4191	* Insert a reference into a list of physical cross reference extents.
4192	*
4193	* @returns The new tracking data for PGMPAGE.
4194	*
4195	* @param pVM The cross context VM structure.
4196	* @param iPhysExt The physical extent index of the list head.
4197	* @param iShwPT The shadow page table index.
4198	* @param iPte Page table entry
4199	*
4200	*/
4201	static uint16_t pgmPoolTrackPhysExtInsert(PVMCC pVM, uint16_t iPhysExt, uint16_t iShwPT, uint16_t iPte)
4202	{
4203	PGM_LOCK_ASSERT_OWNER(pVM);
4204	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
4205	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
4206
4207	/*
4208	* Special common cases.
4209	*/
4210	if (paPhysExts[iPhysExt].aidx[1] == NIL_PGMPOOL_IDX)
4211	{
4212	paPhysExts[iPhysExt].aidx[1] = iShwPT;
4213	paPhysExts[iPhysExt].apte[1] = iPte;
4214	STAM_COUNTER_INC(&pVM->pgm.s.Stats.StatTrackAliasedMany);
4215	LogFlow(("pgmPoolTrackPhysExtInsert: %d:{,%d pte %d,}\n", iPhysExt, iShwPT, iPte));
4216	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
4217	}
4218	if (paPhysExts[iPhysExt].aidx[2] == NIL_PGMPOOL_IDX)
4219	{
4220	paPhysExts[iPhysExt].aidx[2] = iShwPT;
4221	paPhysExts[iPhysExt].apte[2] = iPte;
4222	STAM_COUNTER_INC(&pVM->pgm.s.Stats.StatTrackAliasedMany);
4223	LogFlow(("pgmPoolTrackPhysExtInsert: %d:{,,%d pte %d}\n", iPhysExt, iShwPT, iPte));
4224	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
4225	}
4226	AssertCompile(RT_ELEMENTS(paPhysExts[iPhysExt].aidx) == 3);
4227
4228	/*
4229	* General treatment.
4230	*/
4231	const uint16_t iPhysExtStart = iPhysExt;
4232	unsigned cMax = 15;
4233	for (;;)
4234	{
4235	Assert(iPhysExt < pPool->cMaxPhysExts);
4236	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
4237	if (paPhysExts[iPhysExt].aidx[i] == NIL_PGMPOOL_IDX)
4238	{
4239	paPhysExts[iPhysExt].aidx[i] = iShwPT;
4240	paPhysExts[iPhysExt].apte[i] = iPte;
4241	STAM_COUNTER_INC(&pVM->pgm.s.Stats.StatTrackAliasedMany);
4242	LogFlow(("pgmPoolTrackPhysExtInsert: %d:{%d pte %d} i=%d cMax=%d\n", iPhysExt, iShwPT, iPte, i, cMax));
4243	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtStart);
4244	}
4245	if (!--cMax)
4246	{
4247	STAM_COUNTER_INC(&pVM->pgm.s.Stats.StatTrackOverflows);
4248	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
4249	LogFlow(("pgmPoolTrackPhysExtInsert: overflow (1) iShwPT=%d\n", iShwPT));
4250	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
4251	}
4252
4253	/* advance */
4254	iPhysExt = paPhysExts[iPhysExt].iNext;
4255	if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
4256	break;
4257	}
4258
4259	/*
4260	* Add another extent to the list.
4261	*/
4262	PPGMPOOLPHYSEXT pNew = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
4263	if (!pNew)
4264	{
4265	STAM_COUNTER_INC(&pVM->pgm.s.Stats.StatTrackNoExtentsLeft);
4266	pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
4267	LogFlow(("pgmPoolTrackPhysExtInsert: pgmPoolTrackPhysExtAlloc failed iShwPT=%d\n", iShwPT));
4268	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
4269	}
4270	pNew->iNext = iPhysExtStart;
4271	pNew->aidx[0] = iShwPT;
4272	pNew->apte[0] = iPte;
4273	LogFlow(("pgmPoolTrackPhysExtInsert: added new extent %d:{%d pte %d}->%d\n", iPhysExt, iShwPT, iPte, iPhysExtStart));
4274	return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
4275	}
4276
4277
4278	/**
4279	* Add a reference to guest physical page where extents are in use.
4280	*
4281	* @returns The new tracking data for PGMPAGE.
4282	*
4283	* @param pVM The cross context VM structure.
4284	* @param pPhysPage Pointer to the aPages entry in the ram range.
4285	* @param u16 The ram range flags (top 16-bits).
4286	* @param iShwPT The shadow page table index.
4287	* @param iPte Page table entry
4288	*/
4289	uint16_t pgmPoolTrackPhysExtAddref(PVMCC pVM, PPGMPAGE pPhysPage, uint16_t u16, uint16_t iShwPT, uint16_t iPte)
4290	{
4291	PGM_LOCK_VOID(pVM);
4292	if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
4293	{
4294	/*
4295	* Convert to extent list.
4296	*/
4297	Assert(PGMPOOL_TD_GET_CREFS(u16) == 1);
4298	uint16_t iPhysExt;
4299	PPGMPOOLPHYSEXT pPhysExt = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
4300	if (pPhysExt)
4301	{
4302	LogFlow(("pgmPoolTrackPhysExtAddref: new extent: %d:{%d, %d}\n", iPhysExt, PGMPOOL_TD_GET_IDX(u16), iShwPT));
4303	STAM_COUNTER_INC(&pVM->pgm.s.Stats.StatTrackAliased);
4304	pPhysExt->aidx[0] = PGMPOOL_TD_GET_IDX(u16);
4305	pPhysExt->apte[0] = PGM_PAGE_GET_PTE_INDEX(pPhysPage);
4306	pPhysExt->aidx[1] = iShwPT;
4307	pPhysExt->apte[1] = iPte;
4308	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
4309	}
4310	else
4311	u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
4312	}
4313	else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
4314	{
4315	/*
4316	* Insert into the extent list.
4317	*/
4318	u16 = pgmPoolTrackPhysExtInsert(pVM, PGMPOOL_TD_GET_IDX(u16), iShwPT, iPte);
4319	}
4320	else
4321	STAM_COUNTER_INC(&pVM->pgm.s.Stats.StatTrackAliasedLots);
4322	PGM_UNLOCK(pVM);
4323	return u16;
4324	}
4325
4326
4327	/**
4328	* Clear references to guest physical memory.
4329	*
4330	* @param pPool The pool.
4331	* @param pPage The page.
4332	* @param pPhysPage Pointer to the aPages entry in the ram range.
4333	* @param iPte Shadow PTE index
4334	*/
4335	void pgmPoolTrackPhysExtDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMPAGE pPhysPage, uint16_t iPte)
4336	{
4337	PVMCC pVM = pPool->CTX_SUFF(pVM);
4338	const unsigned cRefs = PGM_PAGE_GET_TD_CREFS(pPhysPage);
4339	AssertFatalMsg(cRefs == PGMPOOL_TD_CREFS_PHYSEXT, ("cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
4340
4341	uint16_t iPhysExt = PGM_PAGE_GET_TD_IDX(pPhysPage);
4342	if (iPhysExt != PGMPOOL_TD_IDX_OVERFLOWED)
4343	{
4344	PGM_LOCK_VOID(pVM);
4345
4346	uint16_t iPhysExtPrev = NIL_PGMPOOL_PHYSEXT_INDEX;
4347	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
4348	do
4349	{
4350	Assert(iPhysExt < pPool->cMaxPhysExts);
4351
4352	/*
4353	* Look for the shadow page and check if it's all freed.
4354	*/
4355	for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
4356	{
4357	if ( paPhysExts[iPhysExt].aidx[i] == pPage->idx
4358	&& paPhysExts[iPhysExt].apte[i] == iPte)
4359	{
4360	paPhysExts[iPhysExt].aidx[i] = NIL_PGMPOOL_IDX;
4361	paPhysExts[iPhysExt].apte[i] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
4362
4363	for (i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
4364	if (paPhysExts[iPhysExt].aidx[i] != NIL_PGMPOOL_IDX)
4365	{
4366	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
4367	PGM_UNLOCK(pVM);
4368	return;
4369	}
4370
4371	/* we can free the node. */
4372	const uint16_t iPhysExtNext = paPhysExts[iPhysExt].iNext;
4373	if ( iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX
4374	&& iPhysExtNext == NIL_PGMPOOL_PHYSEXT_INDEX)
4375	{
4376	/* lonely node */
4377	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
4378	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d lonely\n", pPhysPage, pPage->idx));
4379	PGM_PAGE_SET_TRACKING(pVM, pPhysPage, 0);
4380	}
4381	else if (iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX)
4382	{
4383	/* head */
4384	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d head\n", pPhysPage, pPage->idx));
4385	PGM_PAGE_SET_TRACKING(pVM, pPhysPage, PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtNext));
4386	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
4387	}
4388	else
4389	{
4390	/* in list */
4391	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d in list\n", pPhysPage, pPage->idx));
4392	paPhysExts[iPhysExtPrev].iNext = iPhysExtNext;
4393	pgmPoolTrackPhysExtFree(pVM, iPhysExt);
4394	}
4395	iPhysExt = iPhysExtNext;
4396	PGM_UNLOCK(pVM);
4397	return;
4398	}
4399	}
4400
4401	/* next */
4402	iPhysExtPrev = iPhysExt;
4403	iPhysExt = paPhysExts[iPhysExt].iNext;
4404	} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
4405
4406	PGM_UNLOCK(pVM);
4407	AssertFatalMsgFailed(("not-found! cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
4408	}
4409	else /* nothing to do */
4410	Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage]\n", pPhysPage));
4411	}
4412
4413	/**
4414	* Clear references to guest physical memory.
4415	*
4416	* This is the same as pgmPoolTracDerefGCPhysHint except that the guest
4417	* physical address is assumed to be correct, so the linear search can be
4418	* skipped and we can assert at an earlier point.
4419	*
4420	* @param pPool The pool.
4421	* @param pPage The page.
4422	* @param HCPhys The host physical address corresponding to the guest page.
4423	* @param GCPhys The guest physical address corresponding to HCPhys.
4424	* @param iPte Shadow PTE index
4425	*/
4426	static void pgmPoolTracDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhys, uint16_t iPte)
4427	{
4428	/*
4429	* Lookup the page and check if it checks out before derefing it.
4430	*/
4431	PVMCC pVM = pPool->CTX_SUFF(pVM);
4432	PPGMPAGE pPhysPage = pgmPhysGetPage(pVM, GCPhys);
4433	if (pPhysPage)
4434	{
4435	Assert(PGM_PAGE_GET_HCPHYS(pPhysPage));
4436	#ifdef LOG_ENABLED
4437	RTHCPHYS HCPhysPage = PGM_PAGE_GET_HCPHYS(pPhysPage);
4438	Log2(("pgmPoolTracDerefGCPhys %RHp vs %RHp\n", HCPhysPage, HCPhys));
4439	#endif
4440	if (PGM_PAGE_GET_HCPHYS(pPhysPage) == HCPhys)
4441	{
4442	Assert(pPage->cPresent);
4443	Assert(pPool->cPresent);
4444	pPage->cPresent--;
4445	pPool->cPresent--;
4446	pgmTrackDerefGCPhys(pPool, pPage, pPhysPage, iPte);
4447	return;
4448	}
4449
4450	AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp; found page has HCPhys=%RHp iPte=%u fIsNested=%RTbool\n",
4451	HCPhys, GCPhys, PGM_PAGE_GET_HCPHYS(pPhysPage), iPte, PGMPOOL_PAGE_IS_NESTED(pPage)));
4452	}
4453	AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp\n", HCPhys, GCPhys));
4454	}
4455
4456
4457	/**
4458	* Clear references to guest physical memory.
4459	*
4460	* @param pPool The pool.
4461	* @param pPage The page.
4462	* @param HCPhys The host physical address corresponding to the guest page.
4463	* @param GCPhysHint The guest physical address which may corresponding to HCPhys.
4464	* @param iPte Shadow pte index
4465	*/
4466	void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint, uint16_t iPte)
4467	{
4468	Log4(("pgmPoolTracDerefGCPhysHint %RHp %RGp\n", HCPhys, GCPhysHint));
4469
4470	/*
4471	* Try the hint first.
4472	*/
4473	RTHCPHYS HCPhysHinted;
4474	PVMCC pVM = pPool->CTX_SUFF(pVM);
4475	PPGMPAGE pPhysPage = pgmPhysGetPage(pVM, GCPhysHint);
4476	if (pPhysPage)
4477	{
4478	HCPhysHinted = PGM_PAGE_GET_HCPHYS(pPhysPage);
4479	Assert(HCPhysHinted);
4480	if (HCPhysHinted == HCPhys)
4481	{
4482	Assert(pPage->cPresent);
4483	Assert(pPool->cPresent);
4484	pPage->cPresent--;
4485	pPool->cPresent--;
4486	pgmTrackDerefGCPhys(pPool, pPage, pPhysPage, iPte);
4487	return;
4488	}
4489	}
4490	else
4491	HCPhysHinted = UINT64_C(0xdeadbeefdeadbeef);
4492
4493	/*
4494	* Damn, the hint didn't work. We'll have to do an expensive linear search.
4495	*/
4496	STAM_COUNTER_INC(&pPool->StatTrackLinearRamSearches);
4497	PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRangesX);
4498	while (pRam)
4499	{
4500	unsigned iPage = pRam->cb >> PAGE_SHIFT;
4501	while (iPage-- > 0)
4502	{
4503	if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
4504	{
4505	Log4(("pgmPoolTracDerefGCPhysHint: Linear HCPhys=%RHp GCPhysHint=%RGp GCPhysReal=%RGp\n",
4506	HCPhys, GCPhysHint, pRam->GCPhys + (iPage << PAGE_SHIFT)));
4507	Assert(pPage->cPresent);
4508	Assert(pPool->cPresent);
4509	pPage->cPresent--;
4510	pPool->cPresent--;
4511	pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage], iPte);
4512	return;
4513	}
4514	}
4515	pRam = pRam->CTX_SUFF(pNext);
4516	}
4517
4518	AssertFatalMsgFailed(("HCPhys=%RHp GCPhysHint=%RGp (Hinted page has HCPhys = %RHp)\n", HCPhys, GCPhysHint, HCPhysHinted));
4519	}
4520
4521
4522	/**
4523	* Clear references to guest physical memory in a 32-bit / 32-bit page table.
4524	*
4525	* @param pPool The pool.
4526	* @param pPage The page.
4527	* @param pShwPT The shadow page table (mapping of the page).
4528	* @param pGstPT The guest page table.
4529	*/
4530	DECLINLINE(void) pgmPoolTrackDerefPT32Bit32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT, PCX86PT pGstPT)
4531	{
4532	RTGCPHYS32 const fPgMask = pPage->fA20Enabled ? X86_PTE_PG_MASK : X86_PTE_PG_MASK & ~RT_BIT_32(20);
4533	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
4534	{
4535	const X86PGUINT uPte = pShwPT->a[i].u;
4536	Assert(!(uPte & RT_BIT_32(10)));
4537	if (uPte & X86_PTE_P)
4538	{
4539	Log4(("pgmPoolTrackDerefPT32Bit32Bit: i=%d pte=%RX32 hint=%RX32\n",
4540	i, uPte & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
4541	pgmPoolTracDerefGCPhysHint(pPool, pPage, uPte & X86_PTE_PG_MASK, pGstPT->a[i].u & fPgMask, i);
4542	if (!pPage->cPresent)
4543	break;
4544	}
4545	}
4546	}
4547
4548
4549	/**
4550	* Clear references to guest physical memory in a PAE / 32-bit page table.
4551	*
4552	* @param pPool The pool.
4553	* @param pPage The page.
4554	* @param pShwPT The shadow page table (mapping of the page).
4555	* @param pGstPT The guest page table (just a half one).
4556	*/
4557	DECLINLINE(void) pgmPoolTrackDerefPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PT pGstPT)
4558	{
4559	RTGCPHYS32 const fPgMask = pPage->fA20Enabled ? X86_PTE_PG_MASK : X86_PTE_PG_MASK & ~RT_BIT_32(20);
4560	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
4561	{
4562	Assert( (PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & UINT64_C(0x7ff0000000000400)) == 0
4563	\|\| (PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & UINT64_C(0x7ff0000000000400)) == UINT64_C(0x7ff0000000000000));
4564	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
4565	{
4566	Log4(("pgmPoolTrackDerefPTPae32Bit: i=%d pte=%RX64 hint=%RX32\n",
4567	i, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pGstPT->a[i].u & X86_PTE_PG_MASK));
4568	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pGstPT->a[i].u & fPgMask, i);
4569	if (!pPage->cPresent)
4570	break;
4571	}
4572	}
4573	}
4574
4575
4576	/**
4577	* Clear references to guest physical memory in a PAE / PAE page table.
4578	*
4579	* @param pPool The pool.
4580	* @param pPage The page.
4581	* @param pShwPT The shadow page table (mapping of the page).
4582	* @param pGstPT The guest page table.
4583	*/
4584	DECLINLINE(void) pgmPoolTrackDerefPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT, PCX86PTPAE pGstPT)
4585	{
4586	RTGCPHYS const fPgMask = pPage->fA20Enabled ? X86_PTE_PAE_PG_MASK : X86_PTE_PAE_PG_MASK & ~RT_BIT_64(20);
4587	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
4588	{
4589	Assert( (PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & UINT64_C(0x7ff0000000000400)) == 0
4590	\|\| (PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & UINT64_C(0x7ff0000000000400)) == UINT64_C(0x7ff0000000000000));
4591	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
4592	{
4593	Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX32 hint=%RX32\n",
4594	i, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
4595	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), pGstPT->a[i].u & fPgMask, i);
4596	if (!pPage->cPresent)
4597	break;
4598	}
4599	}
4600	}
4601
4602
4603	/**
4604	* Clear references to guest physical memory in a 32-bit / 4MB page table.
4605	*
4606	* @param pPool The pool.
4607	* @param pPage The page.
4608	* @param pShwPT The shadow page table (mapping of the page).
4609	*/
4610	DECLINLINE(void) pgmPoolTrackDerefPT32Bit4MB(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT)
4611	{
4612	RTGCPHYS const GCPhysA20Mask = pPage->fA20Enabled ? UINT64_MAX : ~RT_BIT_64(20);
4613	RTGCPHYS GCPhys = pPage->GCPhys + PAGE_SIZE * pPage->iFirstPresent;
4614	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
4615	{
4616	const X86PGUINT uPte = pShwPT->a[i].u;
4617	Assert(!(uPte & RT_BIT_32(10)));
4618	if (uPte & X86_PTE_P)
4619	{
4620	Log4(("pgmPoolTrackDerefPT32Bit4MB: i=%d pte=%RX32 GCPhys=%RGp\n",
4621	i, uPte & X86_PTE_PG_MASK, GCPhys));
4622	pgmPoolTracDerefGCPhys(pPool, pPage, uPte & X86_PTE_PG_MASK, GCPhys & GCPhysA20Mask, i);
4623	if (!pPage->cPresent)
4624	break;
4625	}
4626	}
4627	}
4628
4629
4630	/**
4631	* Clear references to guest physical memory in a PAE / 2/4MB page table.
4632	*
4633	* @param pPool The pool.
4634	* @param pPage The page.
4635	* @param pShwPT The shadow page table (mapping of the page).
4636	*/
4637	DECLINLINE(void) pgmPoolTrackDerefPTPaeBig(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMSHWPTPAE pShwPT)
4638	{
4639	RTGCPHYS const GCPhysA20Mask = pPage->fA20Enabled ? UINT64_MAX : ~RT_BIT_64(20);
4640	RTGCPHYS GCPhys = pPage->GCPhys + PAGE_SIZE * pPage->iFirstPresent;
4641	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
4642	{
4643	Assert( (PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & UINT64_C(0x7ff0000000000400)) == 0
4644	\|\| (PGMSHWPTEPAE_GET_U(pShwPT->a[i]) & UINT64_C(0x7ff0000000000400)) == UINT64_C(0x7ff0000000000000));
4645	if (PGMSHWPTEPAE_IS_P(pShwPT->a[i]))
4646	{
4647	Log4(("pgmPoolTrackDerefPTPaeBig: i=%d pte=%RX64 hint=%RGp\n",
4648	i, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), GCPhys));
4649	pgmPoolTracDerefGCPhys(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[i]), GCPhys & GCPhysA20Mask, i);
4650	if (!pPage->cPresent)
4651	break;
4652	}
4653	}
4654	}
4655
4656
4657	/**
4658	* Clear references to shadowed pages in an EPT page table.
4659	*
4660	* @param pPool The pool.
4661	* @param pPage The page.
4662	* @param pShwPT The shadow page directory pointer table (mapping of the
4663	* page).
4664	*/
4665	DECLINLINE(void) pgmPoolTrackDerefPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT)
4666	{
4667	RTGCPHYS const GCPhysA20Mask = pPage->fA20Enabled ? UINT64_MAX : ~RT_BIT_64(20);
4668	RTGCPHYS GCPhys = pPage->GCPhys + PAGE_SIZE * pPage->iFirstPresent;
4669	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
4670	{
4671	X86PGPAEUINT const uPte = pShwPT->a[i].u;
4672	Assert((uPte & UINT64_C(0xfff0000000000f80)) == 0);
4673	if (uPte & EPT_E_READ)
4674	{
4675	Log4(("pgmPoolTrackDerefPTEPT: i=%d pte=%RX64 GCPhys=%RX64\n",
4676	i, uPte & EPT_PTE_PG_MASK, pPage->GCPhys));
4677	pgmPoolTracDerefGCPhys(pPool, pPage, uPte & EPT_PTE_PG_MASK, GCPhys & GCPhysA20Mask, i);
4678	if (!pPage->cPresent)
4679	break;
4680	}
4681	}
4682	}
4683
4684	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
4685
4686	/**
4687	* Clears references to shadowed pages in a SLAT EPT page table.
4688	*
4689	* @param pPool The pool.
4690	* @param pPage The page.
4691	* @param pShwPT The shadow page table (mapping of the page).
4692	* @param pGstPT The guest page table.
4693	*/
4694	DECLINLINE(void) pgmPoolTrackDerefNestedPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT, PCEPTPT pGstPT)
4695	{
4696	Assert(PGMPOOL_PAGE_IS_NESTED(pPage));
4697	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
4698	{
4699	X86PGPAEUINT const uShwPte = pShwPT->a[i].u;
4700	Assert((uShwPte & UINT64_C(0xfff0000000000f80)) == 0); /* Access, Dirty, UserX (not supported) and ignored bits 7, 11. */
4701	if (uShwPte & EPT_PRESENT_MASK)
4702	{
4703	Log7Func(("Shw=%RX64 GstPte=%RX64\n", uShwPte, pGstPT->a[i].u));
4704	pgmPoolTracDerefGCPhys(pPool, pPage, uShwPte & EPT_PTE_PG_MASK, pGstPT->a[i].u & EPT_PTE_PG_MASK, i);
4705	if (!pPage->cPresent)
4706	break;
4707	}
4708	}
4709	}
4710
4711
4712	/**
4713	* Clear references to guest physical memory in a SLAT 2MB EPT page table.
4714	*
4715	* @param pPool The pool.
4716	* @param pPage The page.
4717	* @param pShwPT The shadow page table (mapping of the page).
4718	*/
4719	DECLINLINE(void) pgmPoolTrackDerefNestedPTEPT2MB(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT)
4720	{
4721	Assert(pPage->fA20Enabled);
4722	RTGCPHYS GCPhys = pPage->GCPhys + PAGE_SIZE * pPage->iFirstPresent;
4723	for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
4724	{
4725	X86PGPAEUINT const uShwPte = pShwPT->a[i].u;
4726	Assert((uShwPte & UINT64_C(0xfff0000000000f80)) == 0); /* Access, Dirty, UserX (not supported) and ignored bits 7, 11. */
4727	if (uShwPte & EPT_PRESENT_MASK)
4728	{
4729	Log7Func(("Shw=%RX64 GstPte=%RX64\n", uShwPte, GCPhys));
4730	pgmPoolTracDerefGCPhys(pPool, pPage, uShwPte & EPT_PTE_PG_MASK, GCPhys, i);
4731	if (!pPage->cPresent)
4732	break;
4733	}
4734	}
4735	}
4736
4737
4738	/**
4739	* Clear references to shadowed pages in a SLAT EPT page directory.
4740	*
4741	* @param pPool The pool.
4742	* @param pPage The page.
4743	* @param pShwPD The shadow page directory (mapping of the page).
4744	* @param pGstPD The guest page directory.
4745	*/
4746	DECLINLINE(void) pgmPoolTrackDerefNestedPDEpt(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPD pShwPD, PCEPTPD pGstPD)
4747	{
4748	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
4749	{
4750	X86PGPAEUINT const uPde = pShwPD->a[i].u;
4751	#ifdef PGM_WITH_LARGE_PAGES
4752	AssertMsg((uPde & UINT64_C(0xfff0000000000f00)) == 0, ("uPde=%RX64\n", uPde));
4753	#else
4754	AssertMsg((uPde & UINT64_C(0xfff0000000000f80)) == 0, ("uPde=%RX64\n", uPde));
4755	#endif
4756	if (uPde & EPT_PRESENT_MASK)
4757	{
4758	#ifdef PGM_WITH_LARGE_PAGES
4759	if (uPde & EPT_E_LEAF)
4760	{
4761	Log4(("pgmPoolTrackDerefPDEPT: i=%d pde=%RX64 GCPhys=%RX64\n", i, uPde & EPT_PDE2M_PG_MASK, pPage->GCPhys));
4762	pgmPoolTracDerefGCPhys(pPool, pPage, uPde & EPT_PDE2M_PG_MASK, pGstPD->a[i].u & EPT_PDE2M_PG_MASK, i);
4763	}
4764	else
4765	#endif
4766	{
4767	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, uPde & EPT_PDE_PG_MASK);
4768	if (pSubPage)
4769	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4770	else
4771	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & EPT_PDE_PG_MASK));
4772	}
4773	}
4774	}
4775	}
4776
4777	#endif /* VBOX_WITH_NESTED_HWVIRT_VMX_EPT */
4778
4779
4780	/**
4781	* Clear references to shadowed pages in a 32 bits page directory.
4782	*
4783	* @param pPool The pool.
4784	* @param pPage The page.
4785	* @param pShwPD The shadow page directory (mapping of the page).
4786	*/
4787	DECLINLINE(void) pgmPoolTrackDerefPD(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PD pShwPD)
4788	{
4789	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
4790	{
4791	X86PGUINT const uPde = pShwPD->a[i].u;
4792	if (uPde & X86_PDE_P)
4793	{
4794	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PG_MASK);
4795	if (pSubPage)
4796	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4797	else
4798	AssertFatalMsgFailed(("%x\n", pShwPD->a[i].u & X86_PDE_PG_MASK));
4799	}
4800	}
4801	}
4802
4803
4804	/**
4805	* Clear references to shadowed pages in a PAE (legacy or 64 bits) page directory.
4806	*
4807	* @param pPool The pool.
4808	* @param pPage The page.
4809	* @param pShwPD The shadow page directory (mapping of the page).
4810	*/
4811	DECLINLINE(void) pgmPoolTrackDerefPDPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPAE pShwPD)
4812	{
4813	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
4814	{
4815	X86PGPAEUINT const uPde = pShwPD->a[i].u;
4816	if (uPde & X86_PDE_P)
4817	{
4818	#ifdef PGM_WITH_LARGE_PAGES
4819	if (uPde & X86_PDE_PS)
4820	{
4821	Log4(("pgmPoolTrackDerefPDPae: i=%d pde=%RX64 GCPhys=%RX64\n",
4822	i, uPde & X86_PDE2M_PAE_PG_MASK, pPage->GCPhys));
4823	pgmPoolTracDerefGCPhys(pPool, pPage, uPde & X86_PDE2M_PAE_PG_MASK,
4824	pPage->GCPhys + i * 2 * _1M /* pPage->GCPhys = base address of the memory described by the PD */,
4825	i);
4826	}
4827	else
4828	#endif
4829	{
4830	Assert((uPde & (X86_PDE_PAE_MBZ_MASK_NX \| UINT64_C(0x7ff0000000000000))) == 0);
4831	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, uPde & X86_PDE_PAE_PG_MASK);
4832	if (pSubPage)
4833	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4834	else
4835	AssertFatalMsgFailed(("%RX64\n", uPde & X86_PDE_PAE_PG_MASK));
4836	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4837	}
4838	}
4839	}
4840	}
4841
4842
4843	/**
4844	* Clear references to shadowed pages in a PAE page directory pointer table.
4845	*
4846	* @param pPool The pool.
4847	* @param pPage The page.
4848	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
4849	*/
4850	DECLINLINE(void) pgmPoolTrackDerefPDPTPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
4851	{
4852	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
4853	{
4854	X86PGPAEUINT const uPdpe = pShwPDPT->a[i].u;
4855	Assert((uPdpe & (X86_PDPE_PAE_MBZ_MASK \| UINT64_C(0x7ff0000000000200))) == 0);
4856	if (uPdpe & X86_PDPE_P)
4857	{
4858	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, uPdpe & X86_PDPE_PG_MASK);
4859	if (pSubPage)
4860	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4861	else
4862	AssertFatalMsgFailed(("%RX64\n", uPdpe & X86_PDPE_PG_MASK));
4863	}
4864	}
4865	}
4866
4867
4868	/**
4869	* Clear references to shadowed pages in a 64-bit page directory pointer table.
4870	*
4871	* @param pPool The pool.
4872	* @param pPage The page.
4873	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
4874	*/
4875	DECLINLINE(void) pgmPoolTrackDerefPDPT64Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
4876	{
4877	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
4878	{
4879	X86PGPAEUINT const uPdpe = pShwPDPT->a[i].u;
4880	Assert((uPdpe & (X86_PDPE_LM_MBZ_MASK_NX \| UINT64_C(0x7ff0000000000200))) == 0);
4881	if (uPdpe & X86_PDPE_P)
4882	{
4883	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, uPdpe & X86_PDPE_PG_MASK);
4884	if (pSubPage)
4885	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4886	else
4887	AssertFatalMsgFailed(("%RX64\n", uPdpe & X86_PDPE_PG_MASK));
4888	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4889	}
4890	}
4891	}
4892
4893
4894	/**
4895	* Clear references to shadowed pages in a 64-bit level 4 page table.
4896	*
4897	* @param pPool The pool.
4898	* @param pPage The page.
4899	* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
4900	*/
4901	DECLINLINE(void) pgmPoolTrackDerefPML464Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PML4 pShwPML4)
4902	{
4903	for (unsigned i = 0; i < RT_ELEMENTS(pShwPML4->a); i++)
4904	{
4905	X86PGPAEUINT const uPml4e = pShwPML4->a[i].u;
4906	Assert((uPml4e & (X86_PML4E_MBZ_MASK_NX \| UINT64_C(0x7ff0000000000200))) == 0);
4907	if (uPml4e & X86_PML4E_P)
4908	{
4909	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, uPml4e & X86_PDPE_PG_MASK);
4910	if (pSubPage)
4911	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4912	else
4913	AssertFatalMsgFailed(("%RX64\n", uPml4e & X86_PML4E_PG_MASK));
4914	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4915	}
4916	}
4917	}
4918
4919
4920	/**
4921	* Clear references to shadowed pages in an EPT page directory.
4922	*
4923	* @param pPool The pool.
4924	* @param pPage The page.
4925	* @param pShwPD The shadow page directory (mapping of the page).
4926	*/
4927	DECLINLINE(void) pgmPoolTrackDerefPDEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPD pShwPD)
4928	{
4929	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
4930	{
4931	X86PGPAEUINT const uPde = pShwPD->a[i].u;
4932	#ifdef PGM_WITH_LARGE_PAGES
4933	AssertMsg((uPde & UINT64_C(0xfff0000000000f00)) == 0, ("uPde=%RX64\n", uPde));
4934	#else
4935	AssertMsg((uPde & UINT64_C(0xfff0000000000f80)) == 0, ("uPde=%RX64\n", uPde));
4936	#endif
4937	if (uPde & EPT_E_READ)
4938	{
4939	#ifdef PGM_WITH_LARGE_PAGES
4940	if (uPde & EPT_E_LEAF)
4941	{
4942	Log4(("pgmPoolTrackDerefPDEPT: i=%d pde=%RX64 GCPhys=%RX64\n",
4943	i, uPde & EPT_PDE2M_PG_MASK, pPage->GCPhys));
4944	pgmPoolTracDerefGCPhys(pPool, pPage, uPde & EPT_PDE2M_PG_MASK,
4945	pPage->GCPhys + i * 2 * _1M /* pPage->GCPhys = base address of the memory described by the PD */,
4946	i);
4947	}
4948	else
4949	#endif
4950	{
4951	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, uPde & EPT_PDE_PG_MASK);
4952	if (pSubPage)
4953	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4954	else
4955	AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & EPT_PDE_PG_MASK));
4956	}
4957	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4958	}
4959	}
4960	}
4961
4962
4963	/**
4964	* Clear references to shadowed pages in an EPT page directory pointer table.
4965	*
4966	* @param pPool The pool.
4967	* @param pPage The page.
4968	* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
4969	*/
4970	DECLINLINE(void) pgmPoolTrackDerefPDPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPDPT pShwPDPT)
4971	{
4972	for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
4973	{
4974	X86PGPAEUINT const uPdpe = pShwPDPT->a[i].u;
4975	Assert((uPdpe & UINT64_C(0xfff0000000000f80)) == 0);
4976	if (uPdpe & EPT_E_READ)
4977	{
4978	PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, uPdpe & EPT_PDPTE_PG_MASK);
4979	if (pSubPage)
4980	pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
4981	else
4982	AssertFatalMsgFailed(("%RX64\n", uPdpe & EPT_PDPTE_PG_MASK));
4983	/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
4984	}
4985	}
4986	}
4987
4988
4989	/**
4990	* Clears all references made by this page.
4991	*
4992	* This includes other shadow pages and GC physical addresses.
4993	*
4994	* @param pPool The pool.
4995	* @param pPage The page.
4996	*/
4997	static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
4998	{
4999	/*
5000	* Map the shadow page and take action according to the page kind.
5001	*/
5002	PVMCC pVM = pPool->CTX_SUFF(pVM);
5003	void *pvShw = PGMPOOL_PAGE_2_PTR(pVM, pPage);
5004	switch (pPage->enmKind)
5005	{
5006	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
5007	{
5008	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
5009	void *pvGst;
5010	int rc = PGM_GCPHYS_2_PTR(pVM, pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
5011	pgmPoolTrackDerefPT32Bit32Bit(pPool, pPage, (PX86PT)pvShw, (PCX86PT)pvGst);
5012	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvGst);
5013	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
5014	break;
5015	}
5016
5017	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
5018	{
5019	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
5020	void *pvGst;
5021	int rc = PGM_GCPHYS_2_PTR_EX(pVM, pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
5022	pgmPoolTrackDerefPTPae32Bit(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PT)pvGst);
5023	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvGst);
5024	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
5025	break;
5026	}
5027
5028	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
5029	{
5030	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
5031	void *pvGst;
5032	int rc = PGM_GCPHYS_2_PTR(pVM, pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
5033	pgmPoolTrackDerefPTPaePae(pPool, pPage, (PPGMSHWPTPAE)pvShw, (PCX86PTPAE)pvGst);
5034	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvGst);
5035	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
5036	break;
5037	}
5038
5039	case PGMPOOLKIND_32BIT_PT_FOR_PHYS: /* treat it like a 4 MB page */
5040	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
5041	{
5042	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
5043	pgmPoolTrackDerefPT32Bit4MB(pPool, pPage, (PX86PT)pvShw);
5044	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
5045	break;
5046	}
5047
5048	case PGMPOOLKIND_PAE_PT_FOR_PHYS: /* treat it like a 2 MB page */
5049	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
5050	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
5051	{
5052	STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
5053	pgmPoolTrackDerefPTPaeBig(pPool, pPage, (PPGMSHWPTPAE)pvShw);
5054	STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
5055	break;
5056	}
5057
5058	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
5059	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
5060	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
5061	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
5062	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
5063	case PGMPOOLKIND_PAE_PD_PHYS:
5064	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
5065	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
5066	pgmPoolTrackDerefPDPae(pPool, pPage, (PX86PDPAE)pvShw);
5067	break;
5068
5069	case PGMPOOLKIND_32BIT_PD_PHYS:
5070	case PGMPOOLKIND_32BIT_PD:
5071	pgmPoolTrackDerefPD(pPool, pPage, (PX86PD)pvShw);
5072	break;
5073
5074	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
5075	case PGMPOOLKIND_PAE_PDPT:
5076	case PGMPOOLKIND_PAE_PDPT_PHYS:
5077	pgmPoolTrackDerefPDPTPae(pPool, pPage, (PX86PDPT)pvShw);
5078	break;
5079
5080	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
5081	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
5082	pgmPoolTrackDerefPDPT64Bit(pPool, pPage, (PX86PDPT)pvShw);
5083	break;
5084
5085	case PGMPOOLKIND_64BIT_PML4:
5086	pgmPoolTrackDerefPML464Bit(pPool, pPage, (PX86PML4)pvShw);
5087	break;
5088
5089	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
5090	pgmPoolTrackDerefPTEPT(pPool, pPage, (PEPTPT)pvShw);
5091	break;
5092
5093	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
5094	pgmPoolTrackDerefPDEPT(pPool, pPage, (PEPTPD)pvShw);
5095	break;
5096
5097	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
5098	pgmPoolTrackDerefPDPTEPT(pPool, pPage, (PEPTPDPT)pvShw);
5099	break;
5100
5101	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX_EPT
5102	case PGMPOOLKIND_EPT_PT_FOR_EPT_PT:
5103	{
5104	void *pvGst;
5105	int const rc = PGM_GCPHYS_2_PTR(pVM, pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
5106	pgmPoolTrackDerefNestedPTEPT(pPool, pPage, (PEPTPT)pvShw, (PCEPTPT)pvGst);
5107	break;
5108	}
5109
5110	case PGMPOOLKIND_EPT_PT_FOR_EPT_2MB:
5111	pgmPoolTrackDerefNestedPTEPT2MB(pPool, pPage, (PEPTPT)pvShw);
5112	break;
5113
5114	case PGMPOOLKIND_EPT_PD_FOR_EPT_PD:
5115	{
5116	void *pvGst;
5117	int const rc = PGM_GCPHYS_2_PTR(pVM, pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
5118	pgmPoolTrackDerefNestedPDEpt(pPool, pPage, (PEPTPD)pvShw, (PCEPTPD)pvGst);
5119	break;
5120	}
5121
5122	case PGMPOOLKIND_EPT_PDPT_FOR_EPT_PDPT:
5123	pgmPoolTrackDerefPDPTEPT(pPool, pPage, (PEPTPDPT)pvShw);
5124	break;
5125	#endif
5126
5127	default:
5128	AssertFatalMsgFailed(("enmKind=%d GCPhys=%RGp\n", pPage->enmKind, pPage->GCPhys));
5129	}
5130
5131	/* paranoia, clear the shadow page. Remove this laser (i.e. let Alloc and ClearAll do it). */
5132	STAM_PROFILE_START(&pPool->StatZeroPage, z);
5133	ASMMemZeroPage(pvShw);
5134	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
5135	pPage->fZeroed = true;
5136	Assert(!pPage->cPresent);
5137	PGM_DYNMAP_UNUSED_HINT_VM(pVM, pvShw);
5138	}
5139
5140
5141	/**
5142	* Flushes a pool page.
5143	*
5144	* This moves the page to the free list after removing all user references to it.
5145	*
5146	* @returns VBox status code.
5147	* @retval VINF_SUCCESS on success.
5148	* @param pPool The pool.
5149	* @param pPage The shadow page.
5150	* @param fFlush Flush the TLBS when required (should only be false in very specific use cases!!)
5151	*/
5152	int pgmPoolFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fFlush)
5153	{
5154	PVMCC pVM = pPool->CTX_SUFF(pVM);
5155	bool fFlushRequired = false;
5156
5157	int rc = VINF_SUCCESS;
5158	STAM_PROFILE_START(&pPool->StatFlushPage, f);
5159	LogFlow(("pgmPoolFlushPage: pPage=%p:{.Key=%RHp, .idx=%d, .enmKind=%s, .GCPhys=%RGp}\n",
5160	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
5161
5162	if (PGMPOOL_PAGE_IS_NESTED(pPage))
5163	Log7Func(("pPage=%p:{.Key=%RHp, .idx=%d, .enmKind=%s, .GCPhys=%RGp}\n",
5164	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
5165
5166	/*
5167	* Reject any attempts at flushing any of the special root pages (shall
5168	* not happen).
5169	*/
5170	AssertMsgReturn(pPage->idx >= PGMPOOL_IDX_FIRST,
5171	("pgmPoolFlushPage: special root page, rejected. enmKind=%s idx=%d\n",
5172	pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx),
5173	VINF_SUCCESS);
5174
5175	PGM_LOCK_VOID(pVM);
5176
5177	/*
5178	* Quietly reject any attempts at flushing the currently active shadow CR3 mapping
5179	*/
5180	if (pgmPoolIsPageLocked(pPage))
5181	{
5182	#if !defined(VBOX_VMM_TARGET_ARMV8)
5183	AssertMsg( pPage->enmKind == PGMPOOLKIND_64BIT_PML4
5184	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT
5185	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PDPT_FOR_32BIT
5186	\|\| pPage->enmKind == PGMPOOLKIND_32BIT_PD
5187	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD
5188	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD
5189	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD
5190	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD
5191	\|\| pPage->enmKind == PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD
5192	\|\| pPage->enmKind == PGMPOOLKIND_ROOT_NESTED,
5193	("Can't free the shadow CR3! (%RHp vs %RHp kind=%d\n", PGMGetHyperCR3(VMMGetCpu(pVM)), pPage->Core.Key, pPage->enmKind));
5194	#endif
5195	Log(("pgmPoolFlushPage: current active shadow CR3, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
5196	PGM_UNLOCK(pVM);
5197	return VINF_SUCCESS;
5198	}
5199
5200	/*
5201	* Mark the page as being in need of an ASMMemZeroPage().
5202	*/
5203	pPage->fZeroed = false;
5204
5205	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
5206	if (pPage->fDirty)
5207	pgmPoolFlushDirtyPage(pVM, pPool, pPage->idxDirtyEntry, false /* do not remove */);
5208	#endif
5209
5210	/* If there are any users of this table, then we must issue a tlb flush on all VCPUs. */
5211	if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
5212	fFlushRequired = true;
5213
5214	/*
5215	* Clear the page.
5216	*/
5217	pgmPoolTrackClearPageUsers(pPool, pPage);
5218	STAM_PROFILE_START(&pPool->StatTrackDeref,a);
5219	pgmPoolTrackDeref(pPool, pPage);
5220	STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
5221
5222	/*
5223	* Flush it from the cache.
5224	*/
5225	pgmPoolCacheFlushPage(pPool, pPage);
5226
5227	/*
5228	* Deregistering the monitoring.
5229	*/
5230	if (pPage->fMonitored)
5231	rc = pgmPoolMonitorFlush(pPool, pPage);
5232
5233	/*
5234	* Free the page.
5235	*/
5236	Assert(pPage->iNext == NIL_PGMPOOL_IDX);
5237	pPage->iNext = pPool->iFreeHead;
5238	pPool->iFreeHead = pPage->idx;
5239	pPage->enmKind = PGMPOOLKIND_FREE;
5240	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
5241	pPage->GCPhys = NIL_RTGCPHYS;
5242	pPage->fReusedFlushPending = false;
5243
5244	pPool->cUsedPages--;
5245
5246	/* Flush the TLBs of all VCPUs if required. */
5247	if ( fFlushRequired
5248	&& fFlush)
5249	{
5250	PGM_INVL_ALL_VCPU_TLBS(pVM);
5251	}
5252
5253	PGM_UNLOCK(pVM);
5254	STAM_PROFILE_STOP(&pPool->StatFlushPage, f);
5255	return rc;
5256	}
5257
5258
5259	/**
5260	* Frees a usage of a pool page.
5261	*
5262	* The caller is responsible to updating the user table so that it no longer
5263	* references the shadow page.
5264	*
5265	* @param pPool The pool.
5266	* @param pPage The shadow page.
5267	* @param iUser The shadow page pool index of the user table.
5268	* NIL_PGMPOOL_IDX for root pages.
5269	* @param iUserTable The index into the user table (shadowed). Ignored if
5270	* root page.
5271	*/
5272	void pgmPoolFreeByPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
5273	{
5274	PVMCC pVM = pPool->CTX_SUFF(pVM);
5275
5276	STAM_PROFILE_START(&pPool->StatFree, a);
5277	LogFlow(("pgmPoolFreeByPage: pPage=%p:{.Key=%RHp, .idx=%d, enmKind=%s} iUser=%d iUserTable=%#x\n",
5278	pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), iUser, iUserTable));
5279	AssertReturnVoid(pPage->idx >= PGMPOOL_IDX_FIRST); /* paranoia (#6349) */
5280
5281	PGM_LOCK_VOID(pVM);
5282	if (iUser != NIL_PGMPOOL_IDX)
5283	pgmPoolTrackFreeUser(pPool, pPage, iUser, iUserTable);
5284	if (!pPage->fCached)
5285	pgmPoolFlushPage(pPool, pPage);
5286	PGM_UNLOCK(pVM);
5287	STAM_PROFILE_STOP(&pPool->StatFree, a);
5288	}
5289
5290
5291	/**
5292	* Makes one or more free page free.
5293	*
5294	* @returns VBox status code.
5295	* @retval VINF_SUCCESS on success.
5296	*
5297	* @param pPool The pool.
5298	* @param enmKind Page table kind
5299	* @param iUser The user of the page.
5300	*/
5301	static int pgmPoolMakeMoreFreePages(PPGMPOOL pPool, PGMPOOLKIND enmKind, uint16_t iUser)
5302	{
5303	PVMCC pVM = pPool->CTX_SUFF(pVM);
5304	LogFlow(("pgmPoolMakeMoreFreePages: enmKind=%d iUser=%d\n", enmKind, iUser));
5305	NOREF(enmKind);
5306
5307	/*
5308	* If the pool isn't full grown yet, expand it.
5309	*/
5310	if (pPool->cCurPages < pPool->cMaxPages)
5311	{
5312	STAM_PROFILE_ADV_SUSPEND(&pPool->StatAlloc, a);
5313	#ifdef IN_RING3
5314	int rc = PGMR3PoolGrow(pVM, VMMGetCpu(pVM));
5315	#else
5316	int rc = PGMR0PoolGrow(pVM, VMMGetCpuId(pVM));
5317	#endif
5318	if (RT_FAILURE(rc))
5319	return rc;
5320	STAM_PROFILE_ADV_RESUME(&pPool->StatAlloc, a);
5321	if (pPool->iFreeHead != NIL_PGMPOOL_IDX)
5322	return VINF_SUCCESS;
5323	}
5324
5325	/*
5326	* Free one cached page.
5327	*/
5328	return pgmPoolCacheFreeOne(pPool, iUser);
5329	}
5330
5331
5332	/**
5333	* Allocates a page from the pool.
5334	*
5335	* This page may actually be a cached page and not in need of any processing
5336	* on the callers part.
5337	*
5338	* @returns VBox status code.
5339	* @retval VINF_SUCCESS if a NEW page was allocated.
5340	* @retval VINF_PGM_CACHED_PAGE if a CACHED page was returned.
5341	*
5342	* @param pVM The cross context VM structure.
5343	* @param GCPhys The GC physical address of the page we're gonna shadow.
5344	* For 4MB and 2MB PD entries, it's the first address the
5345	* shadow PT is covering.
5346	* @param enmKind The kind of mapping.
5347	* @param enmAccess Access type for the mapping (only relevant for big pages)
5348	* @param fA20Enabled Whether the A20 gate is enabled or not.
5349	* @param iUser The shadow page pool index of the user table. Root
5350	* pages should pass NIL_PGMPOOL_IDX.
5351	* @param iUserTable The index into the user table (shadowed). Ignored for
5352	* root pages (iUser == NIL_PGMPOOL_IDX).
5353	* @param fLockPage Lock the page
5354	* @param ppPage Where to store the pointer to the page. NULL is stored here on failure.
5355	*/
5356	int pgmPoolAlloc(PVMCC pVM, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, PGMPOOLACCESS enmAccess, bool fA20Enabled,
5357	uint16_t iUser, uint32_t iUserTable, bool fLockPage, PPPGMPOOLPAGE ppPage)
5358	{
5359	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
5360	STAM_PROFILE_ADV_START(&pPool->StatAlloc, a);
5361	LogFlow(("pgmPoolAlloc: GCPhys=%RGp enmKind=%s iUser=%d iUserTable=%#x\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable));
5362	*ppPage = NULL;
5363	/** @todo CSAM/PGMPrefetchPage messes up here during CSAMR3CheckGates
5364	* (TRPMR3SyncIDT) because of FF priority. Try fix that?
5365	* Assert(!(pVM->pgm.s.fGlobalSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)); */
5366
5367	#if defined(VBOX_STRICT) && defined(VBOX_WITH_NESTED_HWVIRT_VMX_EPT)
5368	PVMCPUCC pVCpu = VMMGetCpu(pVM);
5369	Assert(pVCpu->pgm.s.enmGuestSlatMode == PGMSLAT_DIRECT \|\| PGMPOOL_PAGE_IS_KIND_NESTED(enmKind));
5370	#endif
5371
5372	PGM_LOCK_VOID(pVM);
5373
5374	if (pPool->fCacheEnabled)
5375	{
5376	int rc2 = pgmPoolCacheAlloc(pPool, GCPhys, enmKind, enmAccess, fA20Enabled, iUser, iUserTable, ppPage);
5377	if (RT_SUCCESS(rc2))
5378	{
5379	if (fLockPage)
5380	pgmPoolLockPage(pPool, *ppPage);
5381	PGM_UNLOCK(pVM);
5382	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
5383	LogFlow(("pgmPoolAlloc: cached returns %Rrc ppPage=%p:{.Key=%RHp, .idx=%d}\n", rc2, ppPage, (ppPage)->Core.Key, (ppPage)->idx));
5384	return rc2;
5385	}
5386	}
5387
5388	/*
5389	* Allocate a new one.
5390	*/
5391	int rc = VINF_SUCCESS;
5392	uint16_t iNew = pPool->iFreeHead;
5393	if (iNew == NIL_PGMPOOL_IDX)
5394	{
5395	rc = pgmPoolMakeMoreFreePages(pPool, enmKind, iUser);
5396	if (RT_FAILURE(rc))
5397	{
5398	PGM_UNLOCK(pVM);
5399	Log(("pgmPoolAlloc: returns %Rrc (Free)\n", rc));
5400	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
5401	return rc;
5402	}
5403	iNew = pPool->iFreeHead;
5404	AssertReleaseMsgReturn(iNew != NIL_PGMPOOL_IDX, ("iNew=%#x\n", iNew), VERR_PGM_POOL_IPE);
5405	}
5406
5407	/* unlink the free head */
5408	PPGMPOOLPAGE pPage = &pPool->aPages[iNew];
5409	pPool->iFreeHead = pPage->iNext;
5410	pPage->iNext = NIL_PGMPOOL_IDX;
5411
5412	/*
5413	* Initialize it.
5414	*/
5415	pPool->cUsedPages++; /* physical handler registration / pgmPoolTrackFlushGCPhysPTsSlow requirement. */
5416	pPage->enmKind = enmKind;
5417	pPage->enmAccess = enmAccess;
5418	pPage->GCPhys = GCPhys;
5419	pPage->fA20Enabled = fA20Enabled;
5420	pPage->fSeenNonGlobal = false; /* Set this to 'true' to disable this feature. */
5421	pPage->fMonitored = false;
5422	pPage->fCached = false;
5423	pPage->fDirty = false;
5424	pPage->fReusedFlushPending = false;
5425	pPage->cModifications = 0;
5426	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
5427	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
5428	pPage->cPresent = 0;
5429	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
5430	pPage->idxDirtyEntry = 0;
5431	pPage->GCPtrLastAccessHandlerFault = NIL_RTGCPTR;
5432	pPage->GCPtrLastAccessHandlerRip = NIL_RTGCPTR;
5433	pPage->cLastAccessHandler = 0;
5434	pPage->cLocked = 0;
5435	# ifdef VBOX_STRICT
5436	pPage->GCPtrDirtyFault = NIL_RTGCPTR;
5437	# endif
5438
5439	/*
5440	* Insert into the tracking and cache. If this fails, free the page.
5441	*/
5442	int rc3 = pgmPoolTrackInsert(pPool, pPage, GCPhys, iUser, iUserTable);
5443	if (RT_FAILURE(rc3))
5444	{
5445	pPool->cUsedPages--;
5446	pPage->enmKind = PGMPOOLKIND_FREE;
5447	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
5448	pPage->GCPhys = NIL_RTGCPHYS;
5449	pPage->iNext = pPool->iFreeHead;
5450	pPool->iFreeHead = pPage->idx;
5451	PGM_UNLOCK(pVM);
5452	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
5453	Log(("pgmPoolAlloc: returns %Rrc (Insert)\n", rc3));
5454	return rc3;
5455	}
5456
5457	/*
5458	* Commit the allocation, clear the page and return.
5459	*/
5460	#ifdef VBOX_WITH_STATISTICS
5461	if (pPool->cUsedPages > pPool->cUsedPagesHigh)
5462	pPool->cUsedPagesHigh = pPool->cUsedPages;
5463	#endif
5464
5465	if (!pPage->fZeroed)
5466	{
5467	STAM_PROFILE_START(&pPool->StatZeroPage, z);
5468	void *pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
5469	ASMMemZeroPage(pv);
5470	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
5471	}
5472
5473	*ppPage = pPage;
5474	if (fLockPage)
5475	pgmPoolLockPage(pPool, pPage);
5476	PGM_UNLOCK(pVM);
5477	LogFlow(("pgmPoolAlloc: returns %Rrc *ppPage=%p:{.Key=%RHp, .idx=%d, .fCached=%RTbool, .fMonitored=%RTbool}\n",
5478	rc, pPage, pPage->Core.Key, pPage->idx, pPage->fCached, pPage->fMonitored));
5479	STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
5480	return rc;
5481	}
5482
5483
5484	/**
5485	* Frees a usage of a pool page.
5486	*
5487	* @param pVM The cross context VM structure.
5488	* @param HCPhys The HC physical address of the shadow page.
5489	* @param iUser The shadow page pool index of the user table.
5490	* NIL_PGMPOOL_IDX if root page.
5491	* @param iUserTable The index into the user table (shadowed). Ignored if
5492	* root page.
5493	*/
5494	void pgmPoolFree(PVM pVM, RTHCPHYS HCPhys, uint16_t iUser, uint32_t iUserTable)
5495	{
5496	LogFlow(("pgmPoolFree: HCPhys=%RHp iUser=%d iUserTable=%#x\n", HCPhys, iUser, iUserTable));
5497	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
5498	pgmPoolFreeByPage(pPool, pgmPoolGetPage(pPool, HCPhys), iUser, iUserTable);
5499	}
5500
5501
5502	/**
5503	* Internal worker for finding a 'in-use' shadow page give by it's physical address.
5504	*
5505	* @returns Pointer to the shadow page structure.
5506	* @param pPool The pool.
5507	* @param HCPhys The HC physical address of the shadow page.
5508	*/
5509	PPGMPOOLPAGE pgmPoolGetPage(PPGMPOOL pPool, RTHCPHYS HCPhys)
5510	{
5511	PGM_LOCK_ASSERT_OWNER(pPool->CTX_SUFF(pVM));
5512
5513	/*
5514	* Look up the page.
5515	*/
5516	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, HCPhys & X86_PTE_PAE_PG_MASK);
5517
5518	AssertFatalMsg(pPage && pPage->enmKind != PGMPOOLKIND_FREE, ("HCPhys=%RHp pPage=%p idx=%d\n", HCPhys, pPage, (pPage) ? pPage->idx : 0));
5519	return pPage;
5520	}
5521
5522
5523	/**
5524	* Internal worker for finding a page for debugging purposes, no assertions.
5525	*
5526	* @returns Pointer to the shadow page structure. NULL on if not found.
5527	* @param pPool The pool.
5528	* @param HCPhys The HC physical address of the shadow page.
5529	*/
5530	PPGMPOOLPAGE pgmPoolQueryPageForDbg(PPGMPOOL pPool, RTHCPHYS HCPhys)
5531	{
5532	PGM_LOCK_ASSERT_OWNER(pPool->CTX_SUFF(pVM));
5533	return (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, HCPhys & X86_PTE_PAE_PG_MASK);
5534	}
5535
5536
5537	/**
5538	* Internal worker for PGM_HCPHYS_2_PTR.
5539	*
5540	* @returns VBox status code.
5541	* @param pVM The cross context VM structure.
5542	* @param HCPhys The HC physical address of the shadow page.
5543	* @param ppv Where to return the address.
5544	*/
5545	int pgmPoolHCPhys2Ptr(PVM pVM, RTHCPHYS HCPhys, void **ppv)
5546	{
5547	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pVM->pgm.s.CTX_SUFF(pPool)->HCPhysTree, HCPhys & X86_PTE_PAE_PG_MASK);
5548	AssertMsgReturn(pPage && pPage->enmKind != PGMPOOLKIND_FREE,
5549	("HCPhys=%RHp pPage=%p idx=%d\n", HCPhys, pPage, (pPage) ? pPage->idx : 0),
5550	VERR_PGM_POOL_GET_PAGE_FAILED);
5551	ppv = (uint8_t )pPage->CTX_SUFF(pvPage) + (HCPhys & PAGE_OFFSET_MASK);
5552	return VINF_SUCCESS;
5553	}
5554
5555	#ifdef IN_RING3 /* currently only used in ring 3; save some space in the R0 & GC modules (left it here as we might need it elsewhere later on) */
5556
5557	/**
5558	* Flush the specified page if present
5559	*
5560	* @param pVM The cross context VM structure.
5561	* @param GCPhys Guest physical address of the page to flush
5562	*/
5563	void pgmPoolFlushPageByGCPhys(PVM pVM, RTGCPHYS GCPhys)
5564	{
5565	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
5566
5567	VM_ASSERT_EMT(pVM);
5568
5569	/*
5570	* Look up the GCPhys in the hash.
5571	*/
5572	GCPhys = GCPhys & ~(RTGCPHYS)PAGE_OFFSET_MASK;
5573	unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
5574	if (i == NIL_PGMPOOL_IDX)
5575	return;
5576
5577	do
5578	{
5579	PPGMPOOLPAGE pPage = &pPool->aPages[i];
5580	if (pPage->GCPhys - GCPhys < PAGE_SIZE)
5581	{
5582	Assert(!PGMPOOL_PAGE_IS_NESTED(pPage)); /* Temporary to see if it hits. Remove later. */
5583	switch (pPage->enmKind)
5584	{
5585	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
5586	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
5587	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
5588	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
5589	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
5590	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
5591	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
5592	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
5593	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
5594	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
5595	case PGMPOOLKIND_64BIT_PML4:
5596	case PGMPOOLKIND_32BIT_PD:
5597	case PGMPOOLKIND_PAE_PDPT:
5598	{
5599	Log(("PGMPoolFlushPage: found pgm pool pages for %RGp\n", GCPhys));
5600	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
5601	if (pPage->fDirty)
5602	STAM_COUNTER_INC(&pPool->StatForceFlushDirtyPage);
5603	else
5604	# endif
5605	STAM_COUNTER_INC(&pPool->StatForceFlushPage);
5606	Assert(!pgmPoolIsPageLocked(pPage));
5607	pgmPoolMonitorChainFlush(pPool, pPage);
5608	return;
5609	}
5610
5611	/* ignore, no monitoring. */
5612	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
5613	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
5614	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
5615	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
5616	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
5617	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
5618	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
5619	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
5620	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
5621	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
5622	case PGMPOOLKIND_ROOT_NESTED:
5623	case PGMPOOLKIND_PAE_PD_PHYS:
5624	case PGMPOOLKIND_PAE_PDPT_PHYS:
5625	case PGMPOOLKIND_32BIT_PD_PHYS:
5626	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
5627	break;
5628
5629	default:
5630	AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
5631	}
5632	}
5633
5634	/* next */
5635	i = pPage->iNext;
5636	} while (i != NIL_PGMPOOL_IDX);
5637	return;
5638	}
5639
5640
5641	/**
5642	* Reset CPU on hot plugging.
5643	*
5644	* @param pVM The cross context VM structure.
5645	* @param pVCpu The cross context virtual CPU structure.
5646	*/
5647	void pgmR3PoolResetUnpluggedCpu(PVM pVM, PVMCPU pVCpu)
5648	{
5649	pgmR3ExitShadowModeBeforePoolFlush(pVCpu);
5650
5651	pgmR3ReEnterShadowModeAfterPoolFlush(pVM, pVCpu);
5652	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
5653	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
5654	}
5655
5656
5657	/**
5658	* Flushes the entire cache.
5659	*
5660	* It will assert a global CR3 flush (FF) and assumes the caller is aware of
5661	* this and execute this CR3 flush.
5662	*
5663	* @param pVM The cross context VM structure.
5664	*/
5665	void pgmR3PoolReset(PVM pVM)
5666	{
5667	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
5668
5669	PGM_LOCK_ASSERT_OWNER(pVM);
5670	STAM_PROFILE_START(&pPool->StatR3Reset, a);
5671	LogFlow(("pgmR3PoolReset:\n"));
5672
5673	/*
5674	* If there are no pages in the pool, there is nothing to do.
5675	*/
5676	if (pPool->cCurPages <= PGMPOOL_IDX_FIRST)
5677	{
5678	STAM_PROFILE_STOP(&pPool->StatR3Reset, a);
5679	return;
5680	}
5681
5682	/*
5683	* Exit the shadow mode since we're going to clear everything,
5684	* including the root page.
5685	*/
5686	VMCC_FOR_EACH_VMCPU(pVM)
5687	pgmR3ExitShadowModeBeforePoolFlush(pVCpu);
5688	VMCC_FOR_EACH_VMCPU_END(pVM);
5689
5690
5691	/*
5692	* Nuke the free list and reinsert all pages into it.
5693	*/
5694	for (unsigned i = pPool->cCurPages - 1; i >= PGMPOOL_IDX_FIRST; i--)
5695	{
5696	PPGMPOOLPAGE pPage = &pPool->aPages[i];
5697
5698	if (pPage->fMonitored)
5699	pgmPoolMonitorFlush(pPool, pPage);
5700	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
5701	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
5702	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
5703	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
5704	pPage->GCPhys = NIL_RTGCPHYS;
5705	pPage->enmKind = PGMPOOLKIND_FREE;
5706	pPage->enmAccess = PGMPOOLACCESS_DONTCARE;
5707	Assert(pPage->idx == i);
5708	pPage->iNext = i + 1;
5709	pPage->fA20Enabled = true;
5710	pPage->fZeroed = false; /* This could probably be optimized, but better safe than sorry. */
5711	pPage->fSeenNonGlobal = false;
5712	pPage->fMonitored = false;
5713	pPage->fDirty = false;
5714	pPage->fCached = false;
5715	pPage->fReusedFlushPending = false;
5716	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
5717	pPage->cPresent = 0;
5718	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
5719	pPage->cModifications = 0;
5720	pPage->iAgeNext = NIL_PGMPOOL_IDX;
5721	pPage->iAgePrev = NIL_PGMPOOL_IDX;
5722	pPage->idxDirtyEntry = 0;
5723	pPage->GCPtrLastAccessHandlerRip = NIL_RTGCPTR;
5724	pPage->GCPtrLastAccessHandlerFault = NIL_RTGCPTR;
5725	pPage->cLastAccessHandler = 0;
5726	pPage->cLocked = 0;
5727	# ifdef VBOX_STRICT
5728	pPage->GCPtrDirtyFault = NIL_RTGCPTR;
5729	# endif
5730	}
5731	pPool->aPages[pPool->cCurPages - 1].iNext = NIL_PGMPOOL_IDX;
5732	pPool->iFreeHead = PGMPOOL_IDX_FIRST;
5733	pPool->cUsedPages = 0;
5734
5735	/*
5736	* Zap and reinitialize the user records.
5737	*/
5738	pPool->cPresent = 0;
5739	pPool->iUserFreeHead = 0;
5740	PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
5741	const unsigned cMaxUsers = pPool->cMaxUsers;
5742	for (unsigned i = 0; i < cMaxUsers; i++)
5743	{
5744	paUsers[i].iNext = i + 1;
5745	paUsers[i].iUser = NIL_PGMPOOL_IDX;
5746	paUsers[i].iUserTable = 0xfffffffe;
5747	}
5748	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
5749
5750	/*
5751	* Clear all the GCPhys links and rebuild the phys ext free list.
5752	*/
5753	for (PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRangesX);
5754	pRam;
5755	pRam = pRam->CTX_SUFF(pNext))
5756	{
5757	unsigned iPage = pRam->cb >> PAGE_SHIFT;
5758	while (iPage-- > 0)
5759	PGM_PAGE_SET_TRACKING(pVM, &pRam->aPages[iPage], 0);
5760	}
5761
5762	pPool->iPhysExtFreeHead = 0;
5763	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
5764	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
5765	for (unsigned i = 0; i < cMaxPhysExts; i++)
5766	{
5767	paPhysExts[i].iNext = i + 1;
5768	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
5769	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
5770	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
5771	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
5772	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
5773	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
5774	}
5775	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
5776
5777	/*
5778	* Just zap the modified list.
5779	*/
5780	pPool->cModifiedPages = 0;
5781	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
5782
5783	/*
5784	* Clear the GCPhys hash and the age list.
5785	*/
5786	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
5787	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
5788	pPool->iAgeHead = NIL_PGMPOOL_IDX;
5789	pPool->iAgeTail = NIL_PGMPOOL_IDX;
5790
5791	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
5792	/* Clear all dirty pages. */
5793	pPool->idxFreeDirtyPage = 0;
5794	pPool->cDirtyPages = 0;
5795	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aidxDirtyPages); i++)
5796	pPool->aidxDirtyPages[i] = NIL_PGMPOOL_IDX;
5797	# endif
5798
5799	/*
5800	* Reinsert active pages into the hash and ensure monitoring chains are correct.
5801	*/
5802	VMCC_FOR_EACH_VMCPU(pVM)
5803	{
5804	/*
5805	* Re-enter the shadowing mode and assert Sync CR3 FF.
5806	*/
5807	pgmR3ReEnterShadowModeAfterPoolFlush(pVM, pVCpu);
5808	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
5809	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
5810	}
5811	VMCC_FOR_EACH_VMCPU_END(pVM);
5812
5813	STAM_PROFILE_STOP(&pPool->StatR3Reset, a);
5814	}
5815
5816	#endif /* IN_RING3 */
5817
5818	#if defined(LOG_ENABLED) \|\| defined(VBOX_STRICT)
5819	/**
5820	* Stringifies a PGMPOOLKIND value.
5821	*/
5822	static const char *pgmPoolPoolKindToStr(uint8_t enmKind)
5823	{
5824	switch ((PGMPOOLKIND)enmKind)
5825	{
5826	case PGMPOOLKIND_INVALID:
5827	return "PGMPOOLKIND_INVALID";
5828	case PGMPOOLKIND_FREE:
5829	return "PGMPOOLKIND_FREE";
5830	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
5831	return "PGMPOOLKIND_32BIT_PT_FOR_PHYS";
5832	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
5833	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT";
5834	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
5835	return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB";
5836	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
5837	return "PGMPOOLKIND_PAE_PT_FOR_PHYS";
5838	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
5839	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_PT";
5840	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
5841	return "PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB";
5842	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
5843	return "PGMPOOLKIND_PAE_PT_FOR_PAE_PT";
5844	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
5845	return "PGMPOOLKIND_PAE_PT_FOR_PAE_2MB";
5846	case PGMPOOLKIND_32BIT_PD:
5847	return "PGMPOOLKIND_32BIT_PD";
5848	case PGMPOOLKIND_32BIT_PD_PHYS:
5849	return "PGMPOOLKIND_32BIT_PD_PHYS";
5850	case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
5851	return "PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD";
5852	case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
5853	return "PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD";
5854	case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
5855	return "PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD";
5856	case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
5857	return "PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD";
5858	case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
5859	return "PGMPOOLKIND_PAE_PD_FOR_PAE_PD";
5860	case PGMPOOLKIND_PAE_PD_PHYS:
5861	return "PGMPOOLKIND_PAE_PD_PHYS";
5862	case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
5863	return "PGMPOOLKIND_PAE_PDPT_FOR_32BIT";
5864	case PGMPOOLKIND_PAE_PDPT:
5865	return "PGMPOOLKIND_PAE_PDPT";
5866	case PGMPOOLKIND_PAE_PDPT_PHYS:
5867	return "PGMPOOLKIND_PAE_PDPT_PHYS";
5868	case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
5869	return "PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT";
5870	case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
5871	return "PGMPOOLKIND_64BIT_PDPT_FOR_PHYS";
5872	case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
5873	return "PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD";
5874	case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
5875	return "PGMPOOLKIND_64BIT_PD_FOR_PHYS";
5876	case PGMPOOLKIND_64BIT_PML4:
5877	return "PGMPOOLKIND_64BIT_PML4";
5878	case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
5879	return "PGMPOOLKIND_EPT_PDPT_FOR_PHYS";
5880	case PGMPOOLKIND_EPT_PD_FOR_PHYS:
5881	return "PGMPOOLKIND_EPT_PD_FOR_PHYS";
5882	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
5883	return "PGMPOOLKIND_EPT_PT_FOR_PHYS";
5884	case PGMPOOLKIND_ROOT_NESTED:
5885	return "PGMPOOLKIND_ROOT_NESTED";
5886	case PGMPOOLKIND_EPT_PT_FOR_EPT_PT:
5887	return "PGMPOOLKIND_EPT_PT_FOR_EPT_PT";
5888	case PGMPOOLKIND_EPT_PT_FOR_EPT_2MB:
5889	return "PGMPOOLKIND_EPT_PT_FOR_EPT_2MB";
5890	case PGMPOOLKIND_EPT_PD_FOR_EPT_PD:
5891	return "PGMPOOLKIND_EPT_PD_FOR_EPT_PD";
5892	case PGMPOOLKIND_EPT_PDPT_FOR_EPT_PDPT:
5893	return "PGMPOOLKIND_EPT_PDPT_FOR_EPT_PDPT";
5894	case PGMPOOLKIND_EPT_PML4_FOR_EPT_PML4:
5895	return "PGMPOOLKIND_EPT_PML4_FOR_EPT_PML4";
5896	}
5897	return "Unknown kind!";
5898	}
5899	#endif /* LOG_ENABLED \|\| VBOX_STRICT */
5900

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

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAllPool.cpp@ 99208

Download in other formats: