PGMPool.cpp@ 44410

Last change on this file since 44410 was 44399, checked in by vboxsync, 12 years ago
DBGF,DBGC,++: PVM -> PUVM. Some refactoring and cleanup as well.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 53.6 KB

Line
1	/* $Id: PGMPool.cpp 44399 2013-01-27 21:12:53Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
4	*/
5
6	/*
7	* Copyright (C) 2006-2013 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/** @page pg_pgm_pool PGM Shadow Page Pool
19	*
20	* Motivations:
21	* -# Relationship between shadow page tables and physical guest pages. This
22	* should allow us to skip most of the global flushes now following access
23	* handler changes. The main expense is flushing shadow pages.
24	* -# Limit the pool size if necessary (default is kind of limitless).
25	* -# Allocate shadow pages from RC. We use to only do this in SyncCR3.
26	* -# Required for 64-bit guests.
27	* -# Combining the PD cache and page pool in order to simplify caching.
28	*
29	*
30	* @section sec_pgm_pool_outline Design Outline
31	*
32	* The shadow page pool tracks pages used for shadowing paging structures (i.e.
33	* page tables, page directory, page directory pointer table and page map
34	* level-4). Each page in the pool has an unique identifier. This identifier is
35	* used to link a guest physical page to a shadow PT. The identifier is a
36	* non-zero value and has a relativly low max value - say 14 bits. This makes it
37	* possible to fit it into the upper bits of the of the aHCPhys entries in the
38	* ram range.
39	*
40	* By restricting host physical memory to the first 48 bits (which is the
41	* announced physical memory range of the K8L chip (scheduled for 2008)), we
42	* can safely use the upper 16 bits for shadow page ID and reference counting.
43	*
44	* Update: The 48 bit assumption will be lifted with the new physical memory
45	* management (PGMPAGE), so we won't have any trouble when someone stuffs 2TB
46	* into a box in some years.
47	*
48	* Now, it's possible for a page to be aliased, i.e. mapped by more than one PT
49	* or PD. This is solved by creating a list of physical cross reference extents
50	* when ever this happens. Each node in the list (extent) is can contain 3 page
51	* pool indexes. The list it self is chained using indexes into the paPhysExt
52	* array.
53	*
54	*
55	* @section sec_pgm_pool_life Life Cycle of a Shadow Page
56	*
57	* -# The SyncPT function requests a page from the pool.
58	* The request includes the kind of page it is (PT/PD, PAE/legacy), the
59	* address of the page it's shadowing, and more.
60	* -# The pool responds to the request by allocating a new page.
61	* When the cache is enabled, it will first check if it's in the cache.
62	* Should the pool be exhausted, one of two things can be done:
63	* -# Flush the whole pool and current CR3.
64	* -# Use the cache to find a page which can be flushed (~age).
65	* -# The SyncPT function will sync one or more pages and insert it into the
66	* shadow PD.
67	* -# The SyncPage function may sync more pages on a later \#PFs.
68	* -# The page is freed / flushed in SyncCR3 (perhaps) and some other cases.
69	* When caching is enabled, the page isn't flush but remains in the cache.
70	*
71	*
72	* @section sec_pgm_pool_impl Monitoring
73	*
74	* We always monitor PAGE_SIZE chunks of memory. When we've got multiple shadow
75	* pages for the same PAGE_SIZE of guest memory (PAE and mixed PD/PT) the pages
76	* sharing the monitor get linked using the iMonitoredNext/Prev. The head page
77	* is the pvUser to the access handlers.
78	*
79	*
80	* @section sec_pgm_pool_impl Implementation
81	*
82	* The pool will take pages from the MM page pool. The tracking data
83	* (attributes, bitmaps and so on) are allocated from the hypervisor heap. The
84	* pool content can be accessed both by using the page id and the physical
85	* address (HC). The former is managed by means of an array, the latter by an
86	* offset based AVL tree.
87	*
88	* Flushing of a pool page means that we iterate the content (we know what kind
89	* it is) and updates the link information in the ram range.
90	*
91	* ...
92	*/
93
94
95	/*******************************************************************************
96	* Header Files *
97	*******************************************************************************/
98	#define LOG_GROUP LOG_GROUP_PGM_POOL
99	#include <VBox/vmm/pgm.h>
100	#include <VBox/vmm/mm.h>
101	#include "PGMInternal.h"
102	#include <VBox/vmm/vm.h>
103	#include <VBox/vmm/uvm.h>
104	#include "PGMInline.h"
105
106	#include <VBox/log.h>
107	#include <VBox/err.h>
108	#include <iprt/asm.h>
109	#include <iprt/string.h>
110	#include <VBox/dbg.h>
111
112
113	/*******************************************************************************
114	* Internal Functions *
115	*******************************************************************************/
116	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser);
117	#ifdef VBOX_WITH_DEBUGGER
118	static FNDBGCCMD pgmR3PoolCmdCheck;
119	#endif
120
121	#ifdef VBOX_WITH_DEBUGGER
122	/** Command descriptors. */
123	static const DBGCCMD g_aCmds[] =
124	{
125	/* pszCmd, cArgsMin, cArgsMax, paArgDesc, cArgDescs, fFlags, pfnHandler pszSyntax, ....pszDescription */
126	{ "pgmpoolcheck", 0, 0, NULL, 0, 0, pgmR3PoolCmdCheck, "", "Check the pgm pool pages." },
127	};
128	#endif
129
130	/**
131	* Initializes the pool
132	*
133	* @returns VBox status code.
134	* @param pVM Pointer to the VM.
135	*/
136	int pgmR3PoolInit(PVM pVM)
137	{
138	int rc;
139
140	AssertCompile(NIL_PGMPOOL_IDX == 0);
141	/* pPage->cLocked is an unsigned byte. */
142	AssertCompile(VMM_MAX_CPU_COUNT <= 255);
143
144	/*
145	* Query Pool config.
146	*/
147	PCFGMNODE pCfg = CFGMR3GetChild(CFGMR3GetRoot(pVM), "/PGM/Pool");
148
149	/* Default pgm pool size is 1024 pages (4MB). */
150	uint16_t cMaxPages = 1024;
151
152	/* Adjust it up relative to the RAM size, using the nested paging formula. */
153	uint64_t cbRam;
154	rc = CFGMR3QueryU64Def(CFGMR3GetRoot(pVM), "RamSize", &cbRam, 0); AssertRCReturn(rc, rc);
155	uint64_t u64MaxPages = (cbRam >> 9)
156	+ (cbRam >> 18)
157	+ (cbRam >> 27)
158	+ 32 * PAGE_SIZE;
159	u64MaxPages >>= PAGE_SHIFT;
160	if (u64MaxPages > PGMPOOL_IDX_LAST)
161	cMaxPages = PGMPOOL_IDX_LAST;
162	else
163	cMaxPages = (uint16_t)u64MaxPages;
164
165	/** @cfgm{/PGM/Pool/MaxPages, uint16_t, #pages, 16, 0x3fff, F(ram-size)}
166	* The max size of the shadow page pool in pages. The pool will grow dynamically
167	* up to this limit.
168	*/
169	rc = CFGMR3QueryU16Def(pCfg, "MaxPages", &cMaxPages, cMaxPages);
170	AssertLogRelRCReturn(rc, rc);
171	AssertLogRelMsgReturn(cMaxPages <= PGMPOOL_IDX_LAST && cMaxPages >= RT_ALIGN(PGMPOOL_IDX_FIRST, 16),
172	("cMaxPages=%u (%#x)\n", cMaxPages, cMaxPages), VERR_INVALID_PARAMETER);
173	cMaxPages = RT_ALIGN(cMaxPages, 16);
174	if (cMaxPages > PGMPOOL_IDX_LAST)
175	cMaxPages = PGMPOOL_IDX_LAST;
176	LogRel(("PGMPool: cMaxPages=%u (u64MaxPages=%llu)\n", cMaxPages, u64MaxPages));
177
178	/** todo:
179	* We need to be much more careful with our allocation strategy here.
180	* For nested paging we don't need pool user info nor extents at all, but
181	* we can't check for nested paging here (too early during init to get a
182	* confirmation it can be used). The default for large memory configs is a
183	* bit large for shadow paging, so I've restricted the extent maximum to 8k
184	* (8k * 16 = 128k of hyper heap).
185	*
186	* Also when large page support is enabled, we typically don't need so much,
187	* although that depends on the availability of 2 MB chunks on the host.
188	*/
189
190	/** @cfgm{/PGM/Pool/MaxUsers, uint16_t, #users, MaxUsers, 32K, MaxPages*2}
191	* The max number of shadow page user tracking records. Each shadow page has
192	* zero of other shadow pages (or CR3s) that references it, or uses it if you
193	* like. The structures describing these relationships are allocated from a
194	* fixed sized pool. This configuration variable defines the pool size.
195	*/
196	uint16_t cMaxUsers;
197	rc = CFGMR3QueryU16Def(pCfg, "MaxUsers", &cMaxUsers, cMaxPages * 2);
198	AssertLogRelRCReturn(rc, rc);
199	AssertLogRelMsgReturn(cMaxUsers >= cMaxPages && cMaxPages <= _32K,
200	("cMaxUsers=%u (%#x)\n", cMaxUsers, cMaxUsers), VERR_INVALID_PARAMETER);
201
202	/** @cfgm{/PGM/Pool/MaxPhysExts, uint16_t, #extents, 16, MaxPages * 2, MIN(MaxPages*2,8192)}
203	* The max number of extents for tracking aliased guest pages.
204	*/
205	uint16_t cMaxPhysExts;
206	rc = CFGMR3QueryU16Def(pCfg, "MaxPhysExts", &cMaxPhysExts,
207	RT_MIN(cMaxPages * 2, 8192 /* 8Ki max as this eat too much hyper heap */));
208	AssertLogRelRCReturn(rc, rc);
209	AssertLogRelMsgReturn(cMaxPhysExts >= 16 && cMaxPhysExts <= PGMPOOL_IDX_LAST,
210	("cMaxPhysExts=%u (%#x)\n", cMaxPhysExts, cMaxPhysExts), VERR_INVALID_PARAMETER);
211
212	/** @cfgm{/PGM/Pool/ChacheEnabled, bool, true}
213	* Enables or disabling caching of shadow pages. Caching means that we will try
214	* reuse shadow pages instead of recreating them everything SyncCR3, SyncPT or
215	* SyncPage requests one. When reusing a shadow page, we can save time
216	* reconstructing it and it's children.
217	*/
218	bool fCacheEnabled;
219	rc = CFGMR3QueryBoolDef(pCfg, "CacheEnabled", &fCacheEnabled, true);
220	AssertLogRelRCReturn(rc, rc);
221
222	LogRel(("pgmR3PoolInit: cMaxPages=%#RX16 cMaxUsers=%#RX16 cMaxPhysExts=%#RX16 fCacheEnable=%RTbool\n",
223	cMaxPages, cMaxUsers, cMaxPhysExts, fCacheEnabled));
224
225	/*
226	* Allocate the data structures.
227	*/
228	uint32_t cb = RT_OFFSETOF(PGMPOOL, aPages[cMaxPages]);
229	cb += cMaxUsers * sizeof(PGMPOOLUSER);
230	cb += cMaxPhysExts * sizeof(PGMPOOLPHYSEXT);
231	PPGMPOOL pPool;
232	rc = MMR3HyperAllocOnceNoRel(pVM, cb, 0, MM_TAG_PGM_POOL, (void **)&pPool);
233	if (RT_FAILURE(rc))
234	return rc;
235	pVM->pgm.s.pPoolR3 = pPool;
236	pVM->pgm.s.pPoolR0 = MMHyperR3ToR0(pVM, pPool);
237	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pPool);
238
239	/*
240	* Initialize it.
241	*/
242	pPool->pVMR3 = pVM;
243	pPool->pVMR0 = pVM->pVMR0;
244	pPool->pVMRC = pVM->pVMRC;
245	pPool->cMaxPages = cMaxPages;
246	pPool->cCurPages = PGMPOOL_IDX_FIRST;
247	pPool->iUserFreeHead = 0;
248	pPool->cMaxUsers = cMaxUsers;
249	PPGMPOOLUSER paUsers = (PPGMPOOLUSER)&pPool->aPages[pPool->cMaxPages];
250	pPool->paUsersR3 = paUsers;
251	pPool->paUsersR0 = MMHyperR3ToR0(pVM, paUsers);
252	pPool->paUsersRC = MMHyperR3ToRC(pVM, paUsers);
253	for (unsigned i = 0; i < cMaxUsers; i++)
254	{
255	paUsers[i].iNext = i + 1;
256	paUsers[i].iUser = NIL_PGMPOOL_IDX;
257	paUsers[i].iUserTable = 0xfffffffe;
258	}
259	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
260	pPool->iPhysExtFreeHead = 0;
261	pPool->cMaxPhysExts = cMaxPhysExts;
262	PPGMPOOLPHYSEXT paPhysExts = (PPGMPOOLPHYSEXT)&paUsers[cMaxUsers];
263	pPool->paPhysExtsR3 = paPhysExts;
264	pPool->paPhysExtsR0 = MMHyperR3ToR0(pVM, paPhysExts);
265	pPool->paPhysExtsRC = MMHyperR3ToRC(pVM, paPhysExts);
266	for (unsigned i = 0; i < cMaxPhysExts; i++)
267	{
268	paPhysExts[i].iNext = i + 1;
269	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
270	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
271	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
272	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
273	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
274	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
275	}
276	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
277	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
278	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
279	pPool->iAgeHead = NIL_PGMPOOL_IDX;
280	pPool->iAgeTail = NIL_PGMPOOL_IDX;
281	pPool->fCacheEnabled = fCacheEnabled;
282	pPool->pfnAccessHandlerR3 = pgmR3PoolAccessHandler;
283	pPool->pszAccessHandler = "Guest Paging Access Handler";
284	pPool->HCPhysTree = 0;
285
286	/* The NIL entry. */
287	Assert(NIL_PGMPOOL_IDX == 0);
288	pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
289	pPool->aPages[NIL_PGMPOOL_IDX].idx = NIL_PGMPOOL_IDX;
290
291	/* The Shadow 32-bit PD. (32 bits guest paging) */
292	pPool->aPages[PGMPOOL_IDX_PD].enmKind = PGMPOOLKIND_32BIT_PD;
293	pPool->aPages[PGMPOOL_IDX_PD].idx = PGMPOOL_IDX_PD;
294
295	/* The Shadow PDPT. */
296	pPool->aPages[PGMPOOL_IDX_PDPT].enmKind = PGMPOOLKIND_PAE_PDPT;
297	pPool->aPages[PGMPOOL_IDX_PDPT].idx = PGMPOOL_IDX_PDPT;
298
299	/* The Shadow AMD64 CR3. */
300	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].enmKind = PGMPOOLKIND_64BIT_PML4;
301	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].idx = PGMPOOL_IDX_AMD64_CR3;
302
303	/* The Nested Paging CR3. */
304	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].enmKind = PGMPOOLKIND_ROOT_NESTED;
305	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].idx = PGMPOOL_IDX_NESTED_ROOT;
306
307	/*
308	* Set common stuff.
309	*/
310	for (unsigned iPage = 0; iPage < PGMPOOL_IDX_FIRST; iPage++)
311	{
312	pPool->aPages[iPage].Core.Key = NIL_RTHCPHYS;
313	pPool->aPages[iPage].GCPhys = NIL_RTGCPHYS;
314	pPool->aPages[iPage].iNext = NIL_PGMPOOL_IDX;
315	/* pPool->aPages[iPage].cLocked = INT32_MAX; - test this out... */
316	pPool->aPages[iPage].pvPageR3 = 0;
317	pPool->aPages[iPage].iUserHead = NIL_PGMPOOL_USER_INDEX;
318	pPool->aPages[iPage].iModifiedNext = NIL_PGMPOOL_IDX;
319	pPool->aPages[iPage].iModifiedPrev = NIL_PGMPOOL_IDX;
320	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
321	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
322	pPool->aPages[iPage].iAgeNext = NIL_PGMPOOL_IDX;
323	pPool->aPages[iPage].iAgePrev = NIL_PGMPOOL_IDX;
324
325	Assert(pPool->aPages[iPage].idx == iPage);
326	Assert(pPool->aPages[iPage].GCPhys == NIL_RTGCPHYS);
327	Assert(!pPool->aPages[iPage].fSeenNonGlobal);
328	Assert(!pPool->aPages[iPage].fMonitored);
329	Assert(!pPool->aPages[iPage].fCached);
330	Assert(!pPool->aPages[iPage].fZeroed);
331	Assert(!pPool->aPages[iPage].fReusedFlushPending);
332	}
333
334	#ifdef VBOX_WITH_STATISTICS
335	/*
336	* Register statistics.
337	*/
338	STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
339	STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
340	STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
341	STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
342	STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
343	STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolClearAll.");
344	STAM_REG(pVM, &pPool->StatR3Reset, STAMTYPE_PROFILE, "/PGM/Pool/R3Reset", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolReset.");
345	STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
346	STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
347	STAM_REG(pVM, &pPool->StatForceFlushPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForce", STAMUNIT_OCCURENCES, "Counting explicit flushes by PGMPoolFlushPage().");
348	STAM_REG(pVM, &pPool->StatForceFlushDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForceDirty", STAMUNIT_OCCURENCES, "Counting explicit flushes of dirty pages by PGMPoolFlushPage().");
349	STAM_REG(pVM, &pPool->StatForceFlushReused, STAMTYPE_COUNTER, "/PGM/Pool/FlushReused", STAMUNIT_OCCURENCES, "Counting flushes for reused pages.");
350	STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spent zeroing pages. Overlaps with Alloc.");
351	STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
352	STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
353	STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackDeref.");
354	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPT.");
355	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
356	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
357	STAM_REG(pVM, &pPool->StatTrackFlushEntry, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Flush", STAMUNIT_COUNT, "Nr of flushed entries.");
358	STAM_REG(pVM, &pPool->StatTrackFlushEntryKeep, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Update", STAMUNIT_COUNT, "Nr of updated entries.");
359	STAM_REG(pVM, &pPool->StatTrackFreeUpOneUser, STAMTYPE_COUNTER, "/PGM/Pool/Track/FreeUpOneUser", STAMUNIT_TICKS_PER_CALL, "The number of times we were out of user tracking records.");
360	STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_TICKS_PER_CALL, "Profiling deref activity related tracking GC physical pages.");
361	STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
362	STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
363	STAM_REG(pVM, &pPool->StatMonitorRZ, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access handler.");
364	STAM_REG(pVM, &pPool->StatMonitorRZEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
365	STAM_REG(pVM, &pPool->StatMonitorRZFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling the pgmPoolFlushPage calls made from the RC/R0 access handler.");
366	STAM_REG(pVM, &pPool->StatMonitorRZFlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
367	STAM_REG(pVM, &pPool->StatMonitorRZFlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
368	STAM_REG(pVM, &pPool->StatMonitorRZFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
369	STAM_REG(pVM, &pPool->StatMonitorRZHandled, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/Handled", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access we've handled (except REP STOSD).");
370	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
371	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch2, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch2", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction during flushing.");
372	STAM_REG(pVM, &pPool->StatMonitorRZRepPrefix, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/RepPrefix", STAMUNIT_OCCURENCES, "The number of times we've seen rep prefixes we can't handle.");
373	STAM_REG(pVM, &pPool->StatMonitorRZRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
374	STAM_REG(pVM, &pPool->StatMonitorRZFaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
375	STAM_REG(pVM, &pPool->StatMonitorRZFaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
376	STAM_REG(pVM, &pPool->StatMonitorRZFaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
377	STAM_REG(pVM, &pPool->StatMonitorRZFaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
378	STAM_REG(pVM, &pPool->StatMonitorR3, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access handler.");
379	STAM_REG(pVM, &pPool->StatMonitorR3EmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
380	STAM_REG(pVM, &pPool->StatMonitorR3FlushPage, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling the pgmPoolFlushPage calls made from the R3 access handler.");
381	STAM_REG(pVM, &pPool->StatMonitorR3FlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
382	STAM_REG(pVM, &pPool->StatMonitorR3FlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
383	STAM_REG(pVM, &pPool->StatMonitorR3Fork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
384	STAM_REG(pVM, &pPool->StatMonitorR3Handled, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/Handled", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access we've handled (except REP STOSD).");
385	STAM_REG(pVM, &pPool->StatMonitorR3RepPrefix, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/RepPrefix", STAMUNIT_OCCURENCES, "The number of times we've seen rep prefixes we can't handle.");
386	STAM_REG(pVM, &pPool->StatMonitorR3RepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
387	STAM_REG(pVM, &pPool->StatMonitorR3FaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
388	STAM_REG(pVM, &pPool->StatMonitorR3FaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
389	STAM_REG(pVM, &pPool->StatMonitorR3FaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
390	STAM_REG(pVM, &pPool->StatMonitorR3FaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
391	STAM_REG(pVM, &pPool->StatMonitorR3Async, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Async", STAMUNIT_OCCURENCES, "Times we're called in an async thread and need to flush.");
392	STAM_REG(pVM, &pPool->cModifiedPages, STAMTYPE_U16, "/PGM/Pool/Monitor/cModifiedPages", STAMUNIT_PAGES, "The current cModifiedPages value.");
393	STAM_REG(pVM, &pPool->cModifiedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/Monitor/cModifiedPagesHigh", STAMUNIT_PAGES, "The high watermark for cModifiedPages.");
394	STAM_REG(pVM, &pPool->StatResetDirtyPages, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Resets", STAMUNIT_OCCURENCES, "Times we've called pgmPoolResetDirtyPages (and there were dirty page).");
395	STAM_REG(pVM, &pPool->StatDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Pages", STAMUNIT_OCCURENCES, "Times we've called pgmPoolAddDirtyPage.");
396	STAM_REG(pVM, &pPool->StatDirtyPageDupFlush, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/FlushDup", STAMUNIT_OCCURENCES, "Times we've had to flush duplicates for dirty page management.");
397	STAM_REG(pVM, &pPool->StatDirtyPageOverFlowFlush, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/FlushOverflow",STAMUNIT_OCCURENCES, "Times we've had to flush because of overflow.");
398	STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
399	STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
400	STAM_REG(pVM, &pPool->StatCacheKindMismatches, STAMTYPE_COUNTER, "/PGM/Pool/Cache/KindMismatches", STAMUNIT_OCCURENCES, "The number of shadow page kind mismatches. (Better be low, preferably 0!)");
401	STAM_REG(pVM, &pPool->StatCacheFreeUpOne, STAMTYPE_COUNTER, "/PGM/Pool/Cache/FreeUpOne", STAMUNIT_OCCURENCES, "The number of times the cache was asked to free up a page.");
402	STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
403	STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
404	#endif /* VBOX_WITH_STATISTICS */
405
406	#ifdef VBOX_WITH_DEBUGGER
407	/*
408	* Debugger commands.
409	*/
410	static bool s_fRegisteredCmds = false;
411	if (!s_fRegisteredCmds)
412	{
413	rc = DBGCRegisterCommands(&g_aCmds[0], RT_ELEMENTS(g_aCmds));
414	if (RT_SUCCESS(rc))
415	s_fRegisteredCmds = true;
416	}
417	#endif
418
419	return VINF_SUCCESS;
420	}
421
422
423	/**
424	* Relocate the page pool data.
425	*
426	* @param pVM Pointer to the VM.
427	*/
428	void pgmR3PoolRelocate(PVM pVM)
429	{
430	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3);
431	pVM->pgm.s.pPoolR3->pVMRC = pVM->pVMRC;
432	pVM->pgm.s.pPoolR3->paUsersRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paUsersR3);
433	pVM->pgm.s.pPoolR3->paPhysExtsRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paPhysExtsR3);
434	int rc = PDMR3LdrGetSymbolRC(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerRC);
435	AssertReleaseRC(rc);
436	/* init order hack. */
437	if (!pVM->pgm.s.pPoolR3->pfnAccessHandlerR0)
438	{
439	rc = PDMR3LdrGetSymbolR0(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerR0);
440	AssertReleaseRC(rc);
441	}
442	}
443
444
445	/**
446	* Grows the shadow page pool.
447	*
448	* I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
449	*
450	* @returns VBox status code.
451	* @param pVM Pointer to the VM.
452	*/
453	VMMR3DECL(int) PGMR3PoolGrow(PVM pVM)
454	{
455	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
456	AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_PGM_POOL_MAXED_OUT_ALREADY);
457
458	/* With 32-bit guests and no EPT, the CR3 limits the root pages to low
459	(below 4 GB) memory. */
460	/** @todo change the pool to handle ROOT page allocations specially when
461	* required. */
462	bool fCanUseHighMemory = HMIsNestedPagingActive(pVM)
463	&& HMGetShwPagingMode(pVM) == PGMMODE_EPT;
464
465	pgmLock(pVM);
466
467	/*
468	* How much to grow it by?
469	*/
470	uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
471	cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
472	LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages. fCanUseHighMemory=%RTbool\n", cPages, cPages, fCanUseHighMemory));
473
474	for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
475	{
476	PPGMPOOLPAGE pPage = &pPool->aPages[i];
477
478	if (fCanUseHighMemory)
479	pPage->pvPageR3 = MMR3PageAlloc(pVM);
480	else
481	pPage->pvPageR3 = MMR3PageAllocLow(pVM);
482	if (!pPage->pvPageR3)
483	{
484	Log(("We're out of memory!! i=%d fCanUseHighMemory=%RTbool\n", i, fCanUseHighMemory));
485	pgmUnlock(pVM);
486	return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
487	}
488	pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageR3);
489	AssertFatal(pPage->Core.Key < _4G \|\| fCanUseHighMemory);
490	pPage->GCPhys = NIL_RTGCPHYS;
491	pPage->enmKind = PGMPOOLKIND_FREE;
492	pPage->idx = pPage - &pPool->aPages[0];
493	LogFlow(("PGMR3PoolGrow: insert page #%#x - %RHp\n", pPage->idx, pPage->Core.Key));
494	pPage->iNext = pPool->iFreeHead;
495	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
496	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
497	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
498	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
499	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
500	pPage->iAgeNext = NIL_PGMPOOL_IDX;
501	pPage->iAgePrev = NIL_PGMPOOL_IDX;
502	/* commit it */
503	bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
504	pPool->iFreeHead = i;
505	pPool->cCurPages = i + 1;
506	}
507
508	pgmUnlock(pVM);
509	Assert(pPool->cCurPages <= pPool->cMaxPages);
510	return VINF_SUCCESS;
511	}
512
513
514
515	/**
516	* Worker used by pgmR3PoolAccessHandler when it's invoked by an async thread.
517	*
518	* @param pPool The pool.
519	* @param pPage The page.
520	*/
521	static DECLCALLBACK(void) pgmR3PoolFlushReusedPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
522	{
523	/* for the present this should be safe enough I think... */
524	pgmLock(pPool->pVMR3);
525	if ( pPage->fReusedFlushPending
526	&& pPage->enmKind != PGMPOOLKIND_FREE)
527	pgmPoolFlushPage(pPool, pPage);
528	pgmUnlock(pPool->pVMR3);
529	}
530
531
532	/**
533	* \#PF Handler callback for PT write accesses.
534	*
535	* The handler can not raise any faults, it's mainly for monitoring write access
536	* to certain pages.
537	*
538	* @returns VINF_SUCCESS if the handler has carried out the operation.
539	* @returns VINF_PGM_HANDLER_DO_DEFAULT if the caller should carry out the access operation.
540	* @param pVM Pointer to the VM.
541	* @param GCPhys The physical address the guest is writing to.
542	* @param pvPhys The HC mapping of that address.
543	* @param pvBuf What the guest is reading/writing.
544	* @param cbBuf How much it's reading/writing.
545	* @param enmAccessType The access type.
546	* @param pvUser User argument.
547	*/
548	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf,
549	PGMACCESSTYPE enmAccessType, void *pvUser)
550	{
551	STAM_PROFILE_START(&pVM->pgm.s.pPoolR3->StatMonitorR3, a);
552	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
553	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
554	PVMCPU pVCpu = VMMGetCpu(pVM);
555	LogFlow(("pgmR3PoolAccessHandler: GCPhys=%RGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
556	GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
557
558	NOREF(pvBuf); NOREF(enmAccessType);
559
560	/*
561	* We don't have to be very sophisticated about this since there are relativly few calls here.
562	* However, we must try our best to detect any non-cpu accesses (disk / networking).
563	*
564	* Just to make life more interesting, we'll have to deal with the async threads too.
565	* We cannot flush a page if we're in an async thread because of REM notifications.
566	*/
567	pgmLock(pVM);
568	if (PHYS_PAGE_ADDRESS(GCPhys) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
569	{
570	/* Pool page changed while we were waiting for the lock; ignore. */
571	Log(("CPU%d: pgmR3PoolAccessHandler pgm pool page for %RGp changed (to %RGp) while waiting!\n", pVCpu->idCpu, PHYS_PAGE_ADDRESS(GCPhys), PHYS_PAGE_ADDRESS(pPage->GCPhys)));
572	pgmUnlock(pVM);
573	return VINF_PGM_HANDLER_DO_DEFAULT;
574	}
575
576	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
577
578	/* @todo this code doesn't make any sense. remove the if (!pVCpu) block */
579	if (!pVCpu) /** @todo This shouldn't happen any longer, all access handlers will be called on an EMT. All ring-3 handlers, except MMIO, already own the PGM lock. @bugref{3170} */
580	{
581	Log(("pgmR3PoolAccessHandler: async thread, requesting EMT to flush the page: %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
582	pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
583	STAM_COUNTER_INC(&pPool->StatMonitorR3Async);
584	if (!pPage->fReusedFlushPending)
585	{
586	pgmUnlock(pVM);
587	int rc = VMR3ReqCallVoidNoWait(pPool->pVMR3, VMCPUID_ANY, (PFNRT)pgmR3PoolFlushReusedPage, 2, pPool, pPage);
588	AssertRCReturn(rc, rc);
589	pgmLock(pVM);
590	pPage->fReusedFlushPending = true;
591	pPage->cModifications += 0x1000;
592	}
593
594	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
595	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
596	while (cbBuf > 4)
597	{
598	cbBuf -= 4;
599	pvPhys = (uint8_t *)pvPhys + 4;
600	GCPhys += 4;
601	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
602	}
603	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
604	}
605	else if ( ( pPage->cModifications < 96 /* it's cheaper here. */
606	\|\| pgmPoolIsPageLocked(pPage)
607	)
608	&& cbBuf <= 4)
609	{
610	/* Clear the shadow entry. */
611	if (!pPage->cModifications++)
612	pgmPoolMonitorModifiedInsert(pPool, pPage);
613	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
614	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
615	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
616	}
617	else
618	{
619	pgmPoolMonitorChainFlush(pPool, pPage); /* ASSUME that VERR_PGM_POOL_CLEARED can be ignored here and that FFs will deal with it in due time. */
620	STAM_PROFILE_STOP_EX(&pPool->StatMonitorR3, &pPool->StatMonitorR3FlushPage, a);
621	}
622	pgmUnlock(pVM);
623	return VINF_PGM_HANDLER_DO_DEFAULT;
624	}
625
626
627	/**
628	* Rendezvous callback used by pgmR3PoolClearAll that clears all shadow pages
629	* and all modification counters.
630	*
631	* This is only called on one of the EMTs while the other ones are waiting for
632	* it to complete this function.
633	*
634	* @returns VINF_SUCCESS (VBox strict status code).
635	* @param pVM Pointer to the VM.
636	* @param pVCpu The VMCPU for the EMT we're being called on. Unused.
637	* @param fpvFlushRemTlb When not NULL, we'll flush the REM TLB as well.
638	* (This is the pvUser, so it has to be void *.)
639	*
640	*/
641	DECLCALLBACK(VBOXSTRICTRC) pgmR3PoolClearAllRendezvous(PVM pVM, PVMCPU pVCpu, void *fpvFlushRemTbl)
642	{
643	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
644	STAM_PROFILE_START(&pPool->StatClearAll, c);
645	NOREF(pVCpu);
646
647	pgmLock(pVM);
648	Log(("pgmR3PoolClearAllRendezvous: cUsedPages=%d fpvFlushRemTbl=%RTbool\n", pPool->cUsedPages, !!fpvFlushRemTbl));
649
650	/*
651	* Iterate all the pages until we've encountered all that are in use.
652	* This is a simple but not quite optimal solution.
653	*/
654	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
655	unsigned cLeft = pPool->cUsedPages;
656	uint32_t iPage = pPool->cCurPages;
657	while (--iPage >= PGMPOOL_IDX_FIRST)
658	{
659	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
660	if (pPage->GCPhys != NIL_RTGCPHYS)
661	{
662	switch (pPage->enmKind)
663	{
664	/*
665	* We only care about shadow page tables that reference physical memory
666	*/
667	#ifdef PGM_WITH_LARGE_PAGES
668	case PGMPOOLKIND_EPT_PD_FOR_PHYS: /* Large pages reference 2 MB of physical memory, so we must clear them. */
669	if (pPage->cPresent)
670	{
671	PX86PDPAE pShwPD = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
672	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
673	{
674	if ( pShwPD->a[i].n.u1Present
675	&& pShwPD->a[i].b.u1Size)
676	{
677	Assert(!(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING));
678	pShwPD->a[i].u = 0;
679	Assert(pPage->cPresent);
680	pPage->cPresent--;
681	}
682	}
683	if (pPage->cPresent == 0)
684	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
685	}
686	goto default_case;
687
688	case PGMPOOLKIND_PAE_PD_PHYS: /* Large pages reference 2 MB of physical memory, so we must clear them. */
689	if (pPage->cPresent)
690	{
691	PEPTPD pShwPD = (PEPTPD)PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
692	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
693	{
694	Assert((pShwPD->a[i].u & UINT64_C(0xfff0000000000f80)) == 0);
695	if ( pShwPD->a[i].n.u1Present
696	&& pShwPD->a[i].b.u1Size)
697	{
698	Assert(!(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING));
699	pShwPD->a[i].u = 0;
700	Assert(pPage->cPresent);
701	pPage->cPresent--;
702	}
703	}
704	if (pPage->cPresent == 0)
705	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
706	}
707	goto default_case;
708	#endif /* PGM_WITH_LARGE_PAGES */
709
710	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
711	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
712	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
713	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
714	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
715	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
716	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
717	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
718	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
719	{
720	if (pPage->cPresent)
721	{
722	void *pvShw = PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
723	STAM_PROFILE_START(&pPool->StatZeroPage, z);
724	#if 0
725	/* Useful check for leaking references; very expensive though. */
726	switch (pPage->enmKind)
727	{
728	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
729	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
730	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
731	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
732	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
733	{
734	bool fFoundFirst = false;
735	PPGMSHWPTPAE pPT = (PPGMSHWPTPAE)pvShw;
736	for (unsigned ptIndex = 0; ptIndex < RT_ELEMENTS(pPT->a); ptIndex++)
737	{
738	if (pPT->a[ptIndex].u)
739	{
740	if (!fFoundFirst)
741	{
742	AssertFatalMsg(pPage->iFirstPresent <= ptIndex, ("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
743	if (pPage->iFirstPresent != ptIndex)
744	Log(("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
745	fFoundFirst = true;
746	}
747	if (PGMSHWPTEPAE_IS_P(pPT->a[ptIndex]))
748	{
749	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pPT->a[ptIndex]), NIL_RTGCPHYS);
750	if (pPage->iFirstPresent == ptIndex)
751	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
752	}
753	}
754	}
755	AssertFatalMsg(pPage->cPresent == 0, ("cPresent = %d pPage = %RGv\n", pPage->cPresent, pPage->GCPhys));
756	break;
757	}
758	default:
759	break;
760	}
761	#endif
762	ASMMemZeroPage(pvShw);
763	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
764	pPage->cPresent = 0;
765	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
766	}
767	}
768	/* fall thru */
769
770	#ifdef PGM_WITH_LARGE_PAGES
771	default_case:
772	#endif
773	default:
774	Assert(!pPage->cModifications \|\| ++cModifiedPages);
775	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
776	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
777	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
778	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
779	pPage->cModifications = 0;
780	break;
781
782	}
783	if (!--cLeft)
784	break;
785	}
786	}
787
788	/* swipe the special pages too. */
789	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
790	{
791	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
792	if (pPage->GCPhys != NIL_RTGCPHYS)
793	{
794	Assert(!pPage->cModifications \|\| ++cModifiedPages);
795	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
796	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
797	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
798	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
799	pPage->cModifications = 0;
800	}
801	}
802
803	#ifndef DEBUG_michael
804	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
805	#endif
806	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
807	pPool->cModifiedPages = 0;
808
809	/*
810	* Clear all the GCPhys links and rebuild the phys ext free list.
811	*/
812	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRangesX);
813	pRam;
814	pRam = pRam->CTX_SUFF(pNext))
815	{
816	iPage = pRam->cb >> PAGE_SHIFT;
817	while (iPage-- > 0)
818	PGM_PAGE_SET_TRACKING(pVM, &pRam->aPages[iPage], 0);
819	}
820
821	pPool->iPhysExtFreeHead = 0;
822	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
823	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
824	for (unsigned i = 0; i < cMaxPhysExts; i++)
825	{
826	paPhysExts[i].iNext = i + 1;
827	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
828	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
829	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
830	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
831	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
832	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
833	}
834	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
835
836
837	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
838	/* Reset all dirty pages to reactivate the page monitoring. */
839	/* Note: we must do this after clearing all page references and shadow page tables as there might be stale references to
840	* recently removed MMIO ranges around that might otherwise end up asserting in pgmPoolTracDerefGCPhysHint
841	*/
842	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
843	{
844	PPGMPOOLPAGE pPage;
845	unsigned idxPage;
846
847	if (pPool->aDirtyPages[i].uIdx == NIL_PGMPOOL_IDX)
848	continue;
849
850	idxPage = pPool->aDirtyPages[i].uIdx;
851	AssertRelease(idxPage != NIL_PGMPOOL_IDX);
852	pPage = &pPool->aPages[idxPage];
853	Assert(pPage->idx == idxPage);
854	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
855
856	AssertMsg(pPage->fDirty, ("Page %RGp (slot=%d) not marked dirty!", pPage->GCPhys, i));
857
858	Log(("Reactivate dirty page %RGp\n", pPage->GCPhys));
859
860	/* First write protect the page again to catch all write accesses. (before checking for changes -> SMP) */
861	int rc = PGMHandlerPhysicalReset(pVM, pPage->GCPhys & PAGE_BASE_GC_MASK);
862	AssertRCSuccess(rc);
863	pPage->fDirty = false;
864
865	pPool->aDirtyPages[i].uIdx = NIL_PGMPOOL_IDX;
866	}
867
868	/* Clear all dirty pages. */
869	pPool->idxFreeDirtyPage = 0;
870	pPool->cDirtyPages = 0;
871	#endif
872
873	/* Clear the PGM_SYNC_CLEAR_PGM_POOL flag on all VCPUs to prevent redundant flushes. */
874	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
875	pVM->aCpus[idCpu].pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
876
877	/* Flush job finished. */
878	VM_FF_CLEAR(pVM, VM_FF_PGM_POOL_FLUSH_PENDING);
879	pPool->cPresent = 0;
880	pgmUnlock(pVM);
881
882	PGM_INVL_ALL_VCPU_TLBS(pVM);
883
884	if (fpvFlushRemTbl)
885	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
886	CPUMSetChangedFlags(&pVM->aCpus[idCpu], CPUM_CHANGED_GLOBAL_TLB_FLUSH);
887
888	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
889	return VINF_SUCCESS;
890	}
891
892
893	/**
894	* Clears the shadow page pool.
895	*
896	* @param pVM Pointer to the VM.
897	* @param fFlushRemTlb When set, the REM TLB is scheduled for flushing as
898	* well.
899	*/
900	void pgmR3PoolClearAll(PVM pVM, bool fFlushRemTlb)
901	{
902	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmR3PoolClearAllRendezvous, &fFlushRemTlb);
903	AssertRC(rc);
904	}
905
906
907	/**
908	* Protect all pgm pool page table entries to monitor writes
909	*
910	* @param pVM Pointer to the VM.
911	*
912	* @remarks ASSUMES the caller will flush all TLBs!!
913	*/
914	void pgmR3PoolWriteProtectPages(PVM pVM)
915	{
916	PGM_LOCK_ASSERT_OWNER(pVM);
917	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
918	unsigned cLeft = pPool->cUsedPages;
919	unsigned iPage = pPool->cCurPages;
920	while (--iPage >= PGMPOOL_IDX_FIRST)
921	{
922	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
923	if ( pPage->GCPhys != NIL_RTGCPHYS
924	&& pPage->cPresent)
925	{
926	union
927	{
928	void *pv;
929	PX86PT pPT;
930	PPGMSHWPTPAE pPTPae;
931	PEPTPT pPTEpt;
932	} uShw;
933	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
934
935	switch (pPage->enmKind)
936	{
937	/*
938	* We only care about shadow page tables.
939	*/
940	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
941	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
942	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
943	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPT->a); iShw++)
944	{
945	if (uShw.pPT->a[iShw].n.u1Present)
946	uShw.pPT->a[iShw].n.u1Write = 0;
947	}
948	break;
949
950	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
951	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
952	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
953	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
954	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
955	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTPae->a); iShw++)
956	{
957	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw]))
958	PGMSHWPTEPAE_SET_RO(uShw.pPTPae->a[iShw]);
959	}
960	break;
961
962	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
963	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTEpt->a); iShw++)
964	{
965	if (uShw.pPTEpt->a[iShw].n.u1Present)
966	uShw.pPTEpt->a[iShw].n.u1Write = 0;
967	}
968	break;
969
970	default:
971	break;
972	}
973	if (!--cLeft)
974	break;
975	}
976	}
977	}
978
979	#ifdef VBOX_WITH_DEBUGGER
980	/**
981	* @callback_method_impl{FNDBGCCMD, The '.pgmpoolcheck' command.}
982	*/
983	static DECLCALLBACK(int) pgmR3PoolCmdCheck(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PUVM pUVM, PCDBGCVAR paArgs, unsigned cArgs)
984	{
985	DBGC_CMDHLP_REQ_UVM_RET(pCmdHlp, pCmd, pUVM);
986	PVM pVM = pUVM->pVM;
987	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
988	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs == 0);
989	uint32_t cErrors = 0;
990	NOREF(paArgs);
991
992	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
993	for (unsigned i = 0; i < pPool->cCurPages; i++)
994	{
995	PPGMPOOLPAGE pPage = &pPool->aPages[i];
996	bool fFirstMsg = true;
997
998	/* Todo: cover other paging modes too. */
999	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1000	{
1001	PPGMSHWPTPAE pShwPT = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
1002	{
1003	PX86PTPAE pGstPT;
1004	PGMPAGEMAPLOCK LockPage;
1005	int rc = PGMPhysGCPhys2CCPtrReadOnly(pVM, pPage->GCPhys, (const void **)&pGstPT, &LockPage); AssertReleaseRC(rc);
1006
1007	/* Check if any PTEs are out of sync. */
1008	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
1009	{
1010	if (PGMSHWPTEPAE_IS_P(pShwPT->a[j]))
1011	{
1012	RTHCPHYS HCPhys = NIL_RTHCPHYS;
1013	rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[j].u & X86_PTE_PAE_PG_MASK, &HCPhys);
1014	if ( rc != VINF_SUCCESS
1015	\|\| PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[j]) != HCPhys)
1016	{
1017	if (fFirstMsg)
1018	{
1019	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
1020	fFirstMsg = false;
1021	}
1022	DBGCCmdHlpPrintf(pCmdHlp, "Mismatch HCPhys: rc=%Rrc idx=%d guest %RX64 shw=%RX64 vs %RHp\n", rc, j, pGstPT->a[j].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[j]), HCPhys);
1023	cErrors++;
1024	}
1025	else if ( PGMSHWPTEPAE_IS_RW(pShwPT->a[j])
1026	&& !pGstPT->a[j].n.u1Write)
1027	{
1028	if (fFirstMsg)
1029	{
1030	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
1031	fFirstMsg = false;
1032	}
1033	DBGCCmdHlpPrintf(pCmdHlp, "Mismatch r/w gst/shw: idx=%d guest %RX64 shw=%RX64 vs %RHp\n", j, pGstPT->a[j].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[j]), HCPhys);
1034	cErrors++;
1035	}
1036	}
1037	}
1038	PGMPhysReleasePageMappingLock(pVM, &LockPage);
1039	}
1040
1041	/* Make sure this page table can't be written to from any shadow mapping. */
1042	RTHCPHYS HCPhysPT = NIL_RTHCPHYS;
1043	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pPage->GCPhys, &HCPhysPT);
1044	AssertMsgRC(rc, ("PGMPhysGCPhys2HCPhys failed with rc=%d for %RGp\n", rc, pPage->GCPhys));
1045	if (rc == VINF_SUCCESS)
1046	{
1047	for (unsigned j = 0; j < pPool->cCurPages; j++)
1048	{
1049	PPGMPOOLPAGE pTempPage = &pPool->aPages[j];
1050
1051	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1052	{
1053	PPGMSHWPTPAE pShwPT2 = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pTempPage);
1054
1055	for (unsigned k = 0; k < RT_ELEMENTS(pShwPT->a); k++)
1056	{
1057	if ( PGMSHWPTEPAE_IS_P_RW(pShwPT2->a[k])
1058	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1059	&& !pPage->fDirty
1060	# endif
1061	&& PGMSHWPTEPAE_GET_HCPHYS(pShwPT2->a[k]) == HCPhysPT)
1062	{
1063	if (fFirstMsg)
1064	{
1065	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
1066	fFirstMsg = false;
1067	}
1068	DBGCCmdHlpPrintf(pCmdHlp, "Mismatch: r/w: GCPhys=%RGp idx=%d shw %RX64 %RX64\n", pTempPage->GCPhys, k, PGMSHWPTEPAE_GET_LOG(pShwPT->a[k]), PGMSHWPTEPAE_GET_LOG(pShwPT2->a[k]));
1069	cErrors++;
1070	}
1071	}
1072	}
1073	}
1074	}
1075	}
1076	}
1077	if (cErrors > 0)
1078	return DBGCCmdHlpFail(pCmdHlp, pCmd, "Found %#x errors", cErrors);
1079	return VINF_SUCCESS;
1080	}
1081	#endif /* VBOX_WITH_DEBUGGER */

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

source: vbox/trunk/src/VBox/VMM/VMMR3/PGMPool.cpp@ 44410

Download in other formats: