PGMPool.cpp@ 64115

Last change on this file since 64115 was 63560, checked in by vboxsync, 8 years ago
scm: cleaning up todos
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 47.7 KB

Line
1	/* $Id: PGMPool.cpp 63560 2016-08-16 14:01:20Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
4	*/
5
6	/*
7	* Copyright (C) 2006-2016 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_monitoring 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	#ifdef VBOX_WITH_DEBUGGER
117	static FNDBGCCMD pgmR3PoolCmdCheck;
118	#endif
119
120	#ifdef VBOX_WITH_DEBUGGER
121	/** Command descriptors. */
122	static const DBGCCMD g_aCmds[] =
123	{
124	/* pszCmd, cArgsMin, cArgsMax, paArgDesc, cArgDescs, fFlags, pfnHandler pszSyntax, ....pszDescription */
125	{ "pgmpoolcheck", 0, 0, NULL, 0, 0, pgmR3PoolCmdCheck, "", "Check the pgm pool pages." },
126	};
127	#endif
128
129	/**
130	* Initializes the pool
131	*
132	* @returns VBox status code.
133	* @param pVM The cross context VM structure.
134	*/
135	int pgmR3PoolInit(PVM pVM)
136	{
137	int rc;
138
139	AssertCompile(NIL_PGMPOOL_IDX == 0);
140	/* pPage->cLocked is an unsigned byte. */
141	AssertCompile(VMM_MAX_CPU_COUNT <= 255);
142
143	/*
144	* Query Pool config.
145	*/
146	PCFGMNODE pCfg = CFGMR3GetChild(CFGMR3GetRoot(pVM), "/PGM/Pool");
147
148	/* Default pgm pool size is 1024 pages (4MB). */
149	uint16_t cMaxPages = 1024;
150
151	/* Adjust it up relative to the RAM size, using the nested paging formula. */
152	uint64_t cbRam;
153	rc = CFGMR3QueryU64Def(CFGMR3GetRoot(pVM), "RamSize", &cbRam, 0); AssertRCReturn(rc, rc);
154	uint64_t u64MaxPages = (cbRam >> 9)
155	+ (cbRam >> 18)
156	+ (cbRam >> 27)
157	+ 32 * PAGE_SIZE;
158	u64MaxPages >>= PAGE_SHIFT;
159	if (u64MaxPages > PGMPOOL_IDX_LAST)
160	cMaxPages = PGMPOOL_IDX_LAST;
161	else
162	cMaxPages = (uint16_t)u64MaxPages;
163
164	/** @cfgm{/PGM/Pool/MaxPages, uint16_t, \#pages, 16, 0x3fff, F(ram-size)}
165	* The max size of the shadow page pool in pages. The pool will grow dynamically
166	* up to this limit.
167	*/
168	rc = CFGMR3QueryU16Def(pCfg, "MaxPages", &cMaxPages, cMaxPages);
169	AssertLogRelRCReturn(rc, rc);
170	AssertLogRelMsgReturn(cMaxPages <= PGMPOOL_IDX_LAST && cMaxPages >= RT_ALIGN(PGMPOOL_IDX_FIRST, 16),
171	("cMaxPages=%u (%#x)\n", cMaxPages, cMaxPages), VERR_INVALID_PARAMETER);
172	cMaxPages = RT_ALIGN(cMaxPages, 16);
173	if (cMaxPages > PGMPOOL_IDX_LAST)
174	cMaxPages = PGMPOOL_IDX_LAST;
175	LogRel(("PGM: PGMPool: cMaxPages=%u (u64MaxPages=%llu)\n", cMaxPages, u64MaxPages));
176
177	/** @todo
178	* We need to be much more careful with our allocation strategy here.
179	* For nested paging we don't need pool user info nor extents at all, but
180	* we can't check for nested paging here (too early during init to get a
181	* confirmation it can be used). The default for large memory configs is a
182	* bit large for shadow paging, so I've restricted the extent maximum to 8k
183	* (8k * 16 = 128k of hyper heap).
184	*
185	* Also when large page support is enabled, we typically don't need so much,
186	* although that depends on the availability of 2 MB chunks on the host.
187	*/
188
189	/** @cfgm{/PGM/Pool/MaxUsers, uint16_t, \#users, MaxUsers, 32K, MaxPages*2}
190	* The max number of shadow page user tracking records. Each shadow page has
191	* zero of other shadow pages (or CR3s) that references it, or uses it if you
192	* like. The structures describing these relationships are allocated from a
193	* fixed sized pool. This configuration variable defines the pool size.
194	*/
195	uint16_t cMaxUsers;
196	rc = CFGMR3QueryU16Def(pCfg, "MaxUsers", &cMaxUsers, cMaxPages * 2);
197	AssertLogRelRCReturn(rc, rc);
198	AssertLogRelMsgReturn(cMaxUsers >= cMaxPages && cMaxPages <= _32K,
199	("cMaxUsers=%u (%#x)\n", cMaxUsers, cMaxUsers), VERR_INVALID_PARAMETER);
200
201	/** @cfgm{/PGM/Pool/MaxPhysExts, uint16_t, \#extents, 16, MaxPages * 2, MIN(MaxPages*2\,8192)}
202	* The max number of extents for tracking aliased guest pages.
203	*/
204	uint16_t cMaxPhysExts;
205	rc = CFGMR3QueryU16Def(pCfg, "MaxPhysExts", &cMaxPhysExts,
206	RT_MIN(cMaxPages * 2, 8192 /* 8Ki max as this eat too much hyper heap */));
207	AssertLogRelRCReturn(rc, rc);
208	AssertLogRelMsgReturn(cMaxPhysExts >= 16 && cMaxPhysExts <= PGMPOOL_IDX_LAST,
209	("cMaxPhysExts=%u (%#x)\n", cMaxPhysExts, cMaxPhysExts), VERR_INVALID_PARAMETER);
210
211	/** @cfgm{/PGM/Pool/ChacheEnabled, bool, true}
212	* Enables or disabling caching of shadow pages. Caching means that we will try
213	* reuse shadow pages instead of recreating them everything SyncCR3, SyncPT or
214	* SyncPage requests one. When reusing a shadow page, we can save time
215	* reconstructing it and it's children.
216	*/
217	bool fCacheEnabled;
218	rc = CFGMR3QueryBoolDef(pCfg, "CacheEnabled", &fCacheEnabled, true);
219	AssertLogRelRCReturn(rc, rc);
220
221	LogRel(("PGM: pgmR3PoolInit: cMaxPages=%#RX16 cMaxUsers=%#RX16 cMaxPhysExts=%#RX16 fCacheEnable=%RTbool\n",
222	cMaxPages, cMaxUsers, cMaxPhysExts, fCacheEnabled));
223
224	/*
225	* Allocate the data structures.
226	*/
227	uint32_t cb = RT_OFFSETOF(PGMPOOL, aPages[cMaxPages]);
228	cb += cMaxUsers * sizeof(PGMPOOLUSER);
229	cb += cMaxPhysExts * sizeof(PGMPOOLPHYSEXT);
230	PPGMPOOL pPool;
231	rc = MMR3HyperAllocOnceNoRel(pVM, cb, 0, MM_TAG_PGM_POOL, (void **)&pPool);
232	if (RT_FAILURE(rc))
233	return rc;
234	pVM->pgm.s.pPoolR3 = pPool;
235	pVM->pgm.s.pPoolR0 = MMHyperR3ToR0(pVM, pPool);
236	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pPool);
237
238	/*
239	* Initialize it.
240	*/
241	pPool->pVMR3 = pVM;
242	pPool->pVMR0 = pVM->pVMR0;
243	pPool->pVMRC = pVM->pVMRC;
244	pPool->cMaxPages = cMaxPages;
245	pPool->cCurPages = PGMPOOL_IDX_FIRST;
246	pPool->iUserFreeHead = 0;
247	pPool->cMaxUsers = cMaxUsers;
248	PPGMPOOLUSER paUsers = (PPGMPOOLUSER)&pPool->aPages[pPool->cMaxPages];
249	pPool->paUsersR3 = paUsers;
250	pPool->paUsersR0 = MMHyperR3ToR0(pVM, paUsers);
251	pPool->paUsersRC = MMHyperR3ToRC(pVM, paUsers);
252	for (unsigned i = 0; i < cMaxUsers; i++)
253	{
254	paUsers[i].iNext = i + 1;
255	paUsers[i].iUser = NIL_PGMPOOL_IDX;
256	paUsers[i].iUserTable = 0xfffffffe;
257	}
258	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
259	pPool->iPhysExtFreeHead = 0;
260	pPool->cMaxPhysExts = cMaxPhysExts;
261	PPGMPOOLPHYSEXT paPhysExts = (PPGMPOOLPHYSEXT)&paUsers[cMaxUsers];
262	pPool->paPhysExtsR3 = paPhysExts;
263	pPool->paPhysExtsR0 = MMHyperR3ToR0(pVM, paPhysExts);
264	pPool->paPhysExtsRC = MMHyperR3ToRC(pVM, paPhysExts);
265	for (unsigned i = 0; i < cMaxPhysExts; i++)
266	{
267	paPhysExts[i].iNext = i + 1;
268	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
269	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
270	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
271	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
272	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
273	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
274	}
275	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
276	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
277	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
278	pPool->iAgeHead = NIL_PGMPOOL_IDX;
279	pPool->iAgeTail = NIL_PGMPOOL_IDX;
280	pPool->fCacheEnabled = fCacheEnabled;
281
282	pPool->hAccessHandlerType = NIL_PGMPHYSHANDLERTYPE;
283	rc = PGMR3HandlerPhysicalTypeRegister(pVM, PGMPHYSHANDLERKIND_WRITE,
284	pgmPoolAccessHandler,
285	NULL, "pgmPoolAccessHandler", "pgmPoolAccessPfHandler",
286	NULL, "pgmPoolAccessHandler", "pgmPoolAccessPfHandler",
287	"Guest Paging Access Handler",
288	&pPool->hAccessHandlerType);
289	AssertLogRelRCReturn(rc, rc);
290
291	pPool->HCPhysTree = 0;
292
293	/*
294	* The NIL entry.
295	*/
296	Assert(NIL_PGMPOOL_IDX == 0);
297	pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
298	pPool->aPages[NIL_PGMPOOL_IDX].idx = NIL_PGMPOOL_IDX;
299	pPool->aPages[NIL_PGMPOOL_IDX].Core.Key = NIL_RTHCPHYS;
300	pPool->aPages[NIL_PGMPOOL_IDX].GCPhys = NIL_RTGCPHYS;
301	pPool->aPages[NIL_PGMPOOL_IDX].iNext = NIL_PGMPOOL_IDX;
302	/* pPool->aPages[NIL_PGMPOOL_IDX].cLocked = INT32_MAX; - test this out... */
303	pPool->aPages[NIL_PGMPOOL_IDX].pvPageR3 = 0;
304	pPool->aPages[NIL_PGMPOOL_IDX].iUserHead = NIL_PGMPOOL_USER_INDEX;
305	pPool->aPages[NIL_PGMPOOL_IDX].iModifiedNext = NIL_PGMPOOL_IDX;
306	pPool->aPages[NIL_PGMPOOL_IDX].iModifiedPrev = NIL_PGMPOOL_IDX;
307	pPool->aPages[NIL_PGMPOOL_IDX].iMonitoredNext = NIL_PGMPOOL_IDX;
308	pPool->aPages[NIL_PGMPOOL_IDX].iMonitoredPrev = NIL_PGMPOOL_IDX;
309	pPool->aPages[NIL_PGMPOOL_IDX].iAgeNext = NIL_PGMPOOL_IDX;
310	pPool->aPages[NIL_PGMPOOL_IDX].iAgePrev = NIL_PGMPOOL_IDX;
311
312	Assert(pPool->aPages[NIL_PGMPOOL_IDX].idx == NIL_PGMPOOL_IDX);
313	Assert(pPool->aPages[NIL_PGMPOOL_IDX].GCPhys == NIL_RTGCPHYS);
314	Assert(!pPool->aPages[NIL_PGMPOOL_IDX].fSeenNonGlobal);
315	Assert(!pPool->aPages[NIL_PGMPOOL_IDX].fMonitored);
316	Assert(!pPool->aPages[NIL_PGMPOOL_IDX].fCached);
317	Assert(!pPool->aPages[NIL_PGMPOOL_IDX].fZeroed);
318	Assert(!pPool->aPages[NIL_PGMPOOL_IDX].fReusedFlushPending);
319
320	#ifdef VBOX_WITH_STATISTICS
321	/*
322	* Register statistics.
323	*/
324	STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
325	STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
326	STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
327	STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
328	STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
329	STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolClearAll.");
330	STAM_REG(pVM, &pPool->StatR3Reset, STAMTYPE_PROFILE, "/PGM/Pool/R3Reset", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolReset.");
331	STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
332	STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
333	STAM_REG(pVM, &pPool->StatForceFlushPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForce", STAMUNIT_OCCURENCES, "Counting explicit flushes by PGMPoolFlushPage().");
334	STAM_REG(pVM, &pPool->StatForceFlushDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForceDirty", STAMUNIT_OCCURENCES, "Counting explicit flushes of dirty pages by PGMPoolFlushPage().");
335	STAM_REG(pVM, &pPool->StatForceFlushReused, STAMTYPE_COUNTER, "/PGM/Pool/FlushReused", STAMUNIT_OCCURENCES, "Counting flushes for reused pages.");
336	STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spent zeroing pages. Overlaps with Alloc.");
337	STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
338	STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
339	STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackDeref.");
340	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPT.");
341	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
342	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
343	STAM_REG(pVM, &pPool->StatTrackFlushEntry, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Flush", STAMUNIT_COUNT, "Nr of flushed entries.");
344	STAM_REG(pVM, &pPool->StatTrackFlushEntryKeep, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Update", STAMUNIT_COUNT, "Nr of updated entries.");
345	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.");
346	STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_TICKS_PER_CALL, "Profiling deref activity related tracking GC physical pages.");
347	STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
348	STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
349	STAM_REG(pVM, &pPool->StatMonitorRZ, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access handler.");
350	STAM_REG(pVM, &pPool->StatMonitorRZEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
351	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.");
352	STAM_REG(pVM, &pPool->StatMonitorRZFlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
353	STAM_REG(pVM, &pPool->StatMonitorRZFlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
354	STAM_REG(pVM, &pPool->StatMonitorRZFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
355	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).");
356	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
357	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.");
358	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.");
359	STAM_REG(pVM, &pPool->StatMonitorRZRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
360	STAM_REG(pVM, &pPool->StatMonitorRZFaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
361	STAM_REG(pVM, &pPool->StatMonitorRZFaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
362	STAM_REG(pVM, &pPool->StatMonitorRZFaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
363	STAM_REG(pVM, &pPool->StatMonitorRZFaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
364	STAM_REG(pVM, &pPool->StatMonitorR3, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access handler.");
365	STAM_REG(pVM, &pPool->StatMonitorR3EmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
366	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.");
367	STAM_REG(pVM, &pPool->StatMonitorR3FlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
368	STAM_REG(pVM, &pPool->StatMonitorR3FlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
369	STAM_REG(pVM, &pPool->StatMonitorR3Fork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
370	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).");
371	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.");
372	STAM_REG(pVM, &pPool->StatMonitorR3RepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
373	STAM_REG(pVM, &pPool->StatMonitorR3FaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
374	STAM_REG(pVM, &pPool->StatMonitorR3FaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
375	STAM_REG(pVM, &pPool->StatMonitorR3FaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
376	STAM_REG(pVM, &pPool->StatMonitorR3FaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
377	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.");
378	STAM_REG(pVM, &pPool->cModifiedPages, STAMTYPE_U16, "/PGM/Pool/Monitor/cModifiedPages", STAMUNIT_PAGES, "The current cModifiedPages value.");
379	STAM_REG(pVM, &pPool->cModifiedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/Monitor/cModifiedPagesHigh", STAMUNIT_PAGES, "The high watermark for cModifiedPages.");
380	STAM_REG(pVM, &pPool->StatResetDirtyPages, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Resets", STAMUNIT_OCCURENCES, "Times we've called pgmPoolResetDirtyPages (and there were dirty page).");
381	STAM_REG(pVM, &pPool->StatDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Pages", STAMUNIT_OCCURENCES, "Times we've called pgmPoolAddDirtyPage.");
382	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.");
383	STAM_REG(pVM, &pPool->StatDirtyPageOverFlowFlush, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/FlushOverflow",STAMUNIT_OCCURENCES, "Times we've had to flush because of overflow.");
384	STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
385	STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
386	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!)");
387	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.");
388	STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
389	STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
390	#endif /* VBOX_WITH_STATISTICS */
391
392	#ifdef VBOX_WITH_DEBUGGER
393	/*
394	* Debugger commands.
395	*/
396	static bool s_fRegisteredCmds = false;
397	if (!s_fRegisteredCmds)
398	{
399	rc = DBGCRegisterCommands(&g_aCmds[0], RT_ELEMENTS(g_aCmds));
400	if (RT_SUCCESS(rc))
401	s_fRegisteredCmds = true;
402	}
403	#endif
404
405	return VINF_SUCCESS;
406	}
407
408
409	/**
410	* Relocate the page pool data.
411	*
412	* @param pVM The cross context VM structure.
413	*/
414	void pgmR3PoolRelocate(PVM pVM)
415	{
416	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3);
417	pVM->pgm.s.pPoolR3->pVMRC = pVM->pVMRC;
418	pVM->pgm.s.pPoolR3->paUsersRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paUsersR3);
419	pVM->pgm.s.pPoolR3->paPhysExtsRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paPhysExtsR3);
420	}
421
422
423	/**
424	* Grows the shadow page pool.
425	*
426	* I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
427	*
428	* @returns VBox status code.
429	* @param pVM The cross context VM structure.
430	*/
431	VMMR3DECL(int) PGMR3PoolGrow(PVM pVM)
432	{
433	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
434	AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_PGM_POOL_MAXED_OUT_ALREADY);
435
436	/* With 32-bit guests and no EPT, the CR3 limits the root pages to low
437	(below 4 GB) memory. */
438	/** @todo change the pool to handle ROOT page allocations specially when
439	* required. */
440	bool fCanUseHighMemory = HMIsNestedPagingActive(pVM)
441	&& HMGetShwPagingMode(pVM) == PGMMODE_EPT;
442
443	pgmLock(pVM);
444
445	/*
446	* How much to grow it by?
447	*/
448	uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
449	cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
450	LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages. fCanUseHighMemory=%RTbool\n", cPages, cPages, fCanUseHighMemory));
451
452	for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
453	{
454	PPGMPOOLPAGE pPage = &pPool->aPages[i];
455
456	if (fCanUseHighMemory)
457	pPage->pvPageR3 = MMR3PageAlloc(pVM);
458	else
459	pPage->pvPageR3 = MMR3PageAllocLow(pVM);
460	if (!pPage->pvPageR3)
461	{
462	Log(("We're out of memory!! i=%d fCanUseHighMemory=%RTbool\n", i, fCanUseHighMemory));
463	pgmUnlock(pVM);
464	return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
465	}
466	pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageR3);
467	AssertFatal(pPage->Core.Key < _4G \|\| fCanUseHighMemory);
468	pPage->GCPhys = NIL_RTGCPHYS;
469	pPage->enmKind = PGMPOOLKIND_FREE;
470	pPage->idx = pPage - &pPool->aPages[0];
471	LogFlow(("PGMR3PoolGrow: insert page #%#x - %RHp\n", pPage->idx, pPage->Core.Key));
472	pPage->iNext = pPool->iFreeHead;
473	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
474	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
475	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
476	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
477	pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
478	pPage->iAgeNext = NIL_PGMPOOL_IDX;
479	pPage->iAgePrev = NIL_PGMPOOL_IDX;
480	/* commit it */
481	bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
482	pPool->iFreeHead = i;
483	pPool->cCurPages = i + 1;
484	}
485
486	pgmUnlock(pVM);
487	Assert(pPool->cCurPages <= pPool->cMaxPages);
488	return VINF_SUCCESS;
489	}
490
491
492	/**
493	* Rendezvous callback used by pgmR3PoolClearAll that clears all shadow pages
494	* and all modification counters.
495	*
496	* This is only called on one of the EMTs while the other ones are waiting for
497	* it to complete this function.
498	*
499	* @returns VINF_SUCCESS (VBox strict status code).
500	* @param pVM The cross context VM structure.
501	* @param pVCpu The cross context virtual CPU structure of the calling EMT. Unused.
502	* @param fpvFlushRemTlb When not NULL, we'll flush the REM TLB as well.
503	* (This is the pvUser, so it has to be void *.)
504	*
505	*/
506	DECLCALLBACK(VBOXSTRICTRC) pgmR3PoolClearAllRendezvous(PVM pVM, PVMCPU pVCpu, void *fpvFlushRemTlb)
507	{
508	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
509	STAM_PROFILE_START(&pPool->StatClearAll, c);
510	NOREF(pVCpu);
511
512	pgmLock(pVM);
513	Log(("pgmR3PoolClearAllRendezvous: cUsedPages=%d fpvFlushRemTlb=%RTbool\n", pPool->cUsedPages, !!fpvFlushRemTlb));
514
515	/*
516	* Iterate all the pages until we've encountered all that are in use.
517	* This is a simple but not quite optimal solution.
518	*/
519	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
520	unsigned cLeft = pPool->cUsedPages;
521	uint32_t iPage = pPool->cCurPages;
522	while (--iPage >= PGMPOOL_IDX_FIRST)
523	{
524	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
525	if (pPage->GCPhys != NIL_RTGCPHYS)
526	{
527	switch (pPage->enmKind)
528	{
529	/*
530	* We only care about shadow page tables that reference physical memory
531	*/
532	#ifdef PGM_WITH_LARGE_PAGES
533	case PGMPOOLKIND_EPT_PD_FOR_PHYS: /* Large pages reference 2 MB of physical memory, so we must clear them. */
534	if (pPage->cPresent)
535	{
536	PX86PDPAE pShwPD = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
537	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
538	{
539	if ( pShwPD->a[i].n.u1Present
540	&& pShwPD->a[i].b.u1Size)
541	{
542	Assert(!(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING));
543	pShwPD->a[i].u = 0;
544	Assert(pPage->cPresent);
545	pPage->cPresent--;
546	}
547	}
548	if (pPage->cPresent == 0)
549	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
550	}
551	goto default_case;
552
553	case PGMPOOLKIND_PAE_PD_PHYS: /* Large pages reference 2 MB of physical memory, so we must clear them. */
554	if (pPage->cPresent)
555	{
556	PEPTPD pShwPD = (PEPTPD)PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
557	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
558	{
559	Assert((pShwPD->a[i].u & UINT64_C(0xfff0000000000f80)) == 0);
560	if ( pShwPD->a[i].n.u1Present
561	&& pShwPD->a[i].b.u1Size)
562	{
563	Assert(!(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING));
564	pShwPD->a[i].u = 0;
565	Assert(pPage->cPresent);
566	pPage->cPresent--;
567	}
568	}
569	if (pPage->cPresent == 0)
570	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
571	}
572	goto default_case;
573	#endif /* PGM_WITH_LARGE_PAGES */
574
575	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
576	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
577	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
578	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
579	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
580	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
581	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
582	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
583	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
584	{
585	if (pPage->cPresent)
586	{
587	void *pvShw = PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
588	STAM_PROFILE_START(&pPool->StatZeroPage, z);
589	#if 0
590	/* Useful check for leaking references; very expensive though. */
591	switch (pPage->enmKind)
592	{
593	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
594	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
595	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
596	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
597	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
598	{
599	bool fFoundFirst = false;
600	PPGMSHWPTPAE pPT = (PPGMSHWPTPAE)pvShw;
601	for (unsigned ptIndex = 0; ptIndex < RT_ELEMENTS(pPT->a); ptIndex++)
602	{
603	if (pPT->a[ptIndex].u)
604	{
605	if (!fFoundFirst)
606	{
607	AssertFatalMsg(pPage->iFirstPresent <= ptIndex, ("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
608	if (pPage->iFirstPresent != ptIndex)
609	Log(("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
610	fFoundFirst = true;
611	}
612	if (PGMSHWPTEPAE_IS_P(pPT->a[ptIndex]))
613	{
614	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pPT->a[ptIndex]), NIL_RTGCPHYS);
615	if (pPage->iFirstPresent == ptIndex)
616	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
617	}
618	}
619	}
620	AssertFatalMsg(pPage->cPresent == 0, ("cPresent = %d pPage = %RGv\n", pPage->cPresent, pPage->GCPhys));
621	break;
622	}
623	default:
624	break;
625	}
626	#endif
627	ASMMemZeroPage(pvShw);
628	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
629	pPage->cPresent = 0;
630	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
631	}
632	}
633	/* fall thru */
634
635	#ifdef PGM_WITH_LARGE_PAGES
636	default_case:
637	#endif
638	default:
639	Assert(!pPage->cModifications \|\| ++cModifiedPages);
640	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
641	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
642	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
643	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
644	pPage->cModifications = 0;
645	break;
646
647	}
648	if (!--cLeft)
649	break;
650	}
651	}
652
653	#ifndef DEBUG_michael
654	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
655	#endif
656	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
657	pPool->cModifiedPages = 0;
658
659	/*
660	* Clear all the GCPhys links and rebuild the phys ext free list.
661	*/
662	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRangesX);
663	pRam;
664	pRam = pRam->CTX_SUFF(pNext))
665	{
666	iPage = pRam->cb >> PAGE_SHIFT;
667	while (iPage-- > 0)
668	PGM_PAGE_SET_TRACKING(pVM, &pRam->aPages[iPage], 0);
669	}
670
671	pPool->iPhysExtFreeHead = 0;
672	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
673	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
674	for (unsigned i = 0; i < cMaxPhysExts; i++)
675	{
676	paPhysExts[i].iNext = i + 1;
677	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
678	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
679	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
680	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
681	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
682	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
683	}
684	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
685
686
687	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
688	/* Reset all dirty pages to reactivate the page monitoring. */
689	/* Note: we must do this after clearing all page references and shadow page tables as there might be stale references to
690	* recently removed MMIO ranges around that might otherwise end up asserting in pgmPoolTracDerefGCPhysHint
691	*/
692	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
693	{
694	PPGMPOOLPAGE pPage;
695	unsigned idxPage;
696
697	if (pPool->aDirtyPages[i].uIdx == NIL_PGMPOOL_IDX)
698	continue;
699
700	idxPage = pPool->aDirtyPages[i].uIdx;
701	AssertRelease(idxPage != NIL_PGMPOOL_IDX);
702	pPage = &pPool->aPages[idxPage];
703	Assert(pPage->idx == idxPage);
704	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
705
706	AssertMsg(pPage->fDirty, ("Page %RGp (slot=%d) not marked dirty!", pPage->GCPhys, i));
707
708	Log(("Reactivate dirty page %RGp\n", pPage->GCPhys));
709
710	/* First write protect the page again to catch all write accesses. (before checking for changes -> SMP) */
711	int rc = PGMHandlerPhysicalReset(pVM, pPage->GCPhys & PAGE_BASE_GC_MASK);
712	AssertRCSuccess(rc);
713	pPage->fDirty = false;
714
715	pPool->aDirtyPages[i].uIdx = NIL_PGMPOOL_IDX;
716	}
717
718	/* Clear all dirty pages. */
719	pPool->idxFreeDirtyPage = 0;
720	pPool->cDirtyPages = 0;
721	#endif
722
723	/* Clear the PGM_SYNC_CLEAR_PGM_POOL flag on all VCPUs to prevent redundant flushes. */
724	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
725	pVM->aCpus[idCpu].pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
726
727	/* Flush job finished. */
728	VM_FF_CLEAR(pVM, VM_FF_PGM_POOL_FLUSH_PENDING);
729	pPool->cPresent = 0;
730	pgmUnlock(pVM);
731
732	PGM_INVL_ALL_VCPU_TLBS(pVM);
733
734	if (fpvFlushRemTlb)
735	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
736	CPUMSetChangedFlags(&pVM->aCpus[idCpu], CPUM_CHANGED_GLOBAL_TLB_FLUSH);
737
738	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
739	return VINF_SUCCESS;
740	}
741
742
743	/**
744	* Clears the shadow page pool.
745	*
746	* @param pVM The cross context VM structure.
747	* @param fFlushRemTlb When set, the REM TLB is scheduled for flushing as
748	* well.
749	*/
750	void pgmR3PoolClearAll(PVM pVM, bool fFlushRemTlb)
751	{
752	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmR3PoolClearAllRendezvous, &fFlushRemTlb);
753	AssertRC(rc);
754	}
755
756
757	/**
758	* Protect all pgm pool page table entries to monitor writes
759	*
760	* @param pVM The cross context VM structure.
761	*
762	* @remarks ASSUMES the caller will flush all TLBs!!
763	*/
764	void pgmR3PoolWriteProtectPages(PVM pVM)
765	{
766	PGM_LOCK_ASSERT_OWNER(pVM);
767	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
768	unsigned cLeft = pPool->cUsedPages;
769	unsigned iPage = pPool->cCurPages;
770	while (--iPage >= PGMPOOL_IDX_FIRST)
771	{
772	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
773	if ( pPage->GCPhys != NIL_RTGCPHYS
774	&& pPage->cPresent)
775	{
776	union
777	{
778	void *pv;
779	PX86PT pPT;
780	PPGMSHWPTPAE pPTPae;
781	PEPTPT pPTEpt;
782	} uShw;
783	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
784
785	switch (pPage->enmKind)
786	{
787	/*
788	* We only care about shadow page tables.
789	*/
790	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
791	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
792	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
793	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPT->a); iShw++)
794	{
795	if (uShw.pPT->a[iShw].n.u1Present)
796	uShw.pPT->a[iShw].n.u1Write = 0;
797	}
798	break;
799
800	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
801	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
802	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
803	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
804	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
805	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTPae->a); iShw++)
806	{
807	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw]))
808	PGMSHWPTEPAE_SET_RO(uShw.pPTPae->a[iShw]);
809	}
810	break;
811
812	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
813	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTEpt->a); iShw++)
814	{
815	if (uShw.pPTEpt->a[iShw].n.u1Present)
816	uShw.pPTEpt->a[iShw].n.u1Write = 0;
817	}
818	break;
819
820	default:
821	break;
822	}
823	if (!--cLeft)
824	break;
825	}
826	}
827	}
828
829	#ifdef VBOX_WITH_DEBUGGER
830	/**
831	* @callback_method_impl{FNDBGCCMD, The '.pgmpoolcheck' command.}
832	*/
833	static DECLCALLBACK(int) pgmR3PoolCmdCheck(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PUVM pUVM, PCDBGCVAR paArgs, unsigned cArgs)
834	{
835	DBGC_CMDHLP_REQ_UVM_RET(pCmdHlp, pCmd, pUVM);
836	PVM pVM = pUVM->pVM;
837	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
838	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs == 0);
839	uint32_t cErrors = 0;
840	NOREF(paArgs);
841
842	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
843	for (unsigned i = 0; i < pPool->cCurPages; i++)
844	{
845	PPGMPOOLPAGE pPage = &pPool->aPages[i];
846	bool fFirstMsg = true;
847
848	/** @todo cover other paging modes too. */
849	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
850	{
851	PPGMSHWPTPAE pShwPT = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
852	{
853	PX86PTPAE pGstPT;
854	PGMPAGEMAPLOCK LockPage;
855	int rc = PGMPhysGCPhys2CCPtrReadOnly(pVM, pPage->GCPhys, (const void **)&pGstPT, &LockPage); AssertReleaseRC(rc);
856
857	/* Check if any PTEs are out of sync. */
858	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
859	{
860	if (PGMSHWPTEPAE_IS_P(pShwPT->a[j]))
861	{
862	RTHCPHYS HCPhys = NIL_RTHCPHYS;
863	rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[j].u & X86_PTE_PAE_PG_MASK, &HCPhys);
864	if ( rc != VINF_SUCCESS
865	\|\| PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[j]) != HCPhys)
866	{
867	if (fFirstMsg)
868	{
869	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
870	fFirstMsg = false;
871	}
872	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);
873	cErrors++;
874	}
875	else if ( PGMSHWPTEPAE_IS_RW(pShwPT->a[j])
876	&& !pGstPT->a[j].n.u1Write)
877	{
878	if (fFirstMsg)
879	{
880	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
881	fFirstMsg = false;
882	}
883	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);
884	cErrors++;
885	}
886	}
887	}
888	PGMPhysReleasePageMappingLock(pVM, &LockPage);
889	}
890
891	/* Make sure this page table can't be written to from any shadow mapping. */
892	RTHCPHYS HCPhysPT = NIL_RTHCPHYS;
893	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pPage->GCPhys, &HCPhysPT);
894	AssertMsgRC(rc, ("PGMPhysGCPhys2HCPhys failed with rc=%d for %RGp\n", rc, pPage->GCPhys));
895	if (rc == VINF_SUCCESS)
896	{
897	for (unsigned j = 0; j < pPool->cCurPages; j++)
898	{
899	PPGMPOOLPAGE pTempPage = &pPool->aPages[j];
900
901	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
902	{
903	PPGMSHWPTPAE pShwPT2 = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pTempPage);
904
905	for (unsigned k = 0; k < RT_ELEMENTS(pShwPT->a); k++)
906	{
907	if ( PGMSHWPTEPAE_IS_P_RW(pShwPT2->a[k])
908	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
909	&& !pPage->fDirty
910	# endif
911	&& PGMSHWPTEPAE_GET_HCPHYS(pShwPT2->a[k]) == HCPhysPT)
912	{
913	if (fFirstMsg)
914	{
915	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
916	fFirstMsg = false;
917	}
918	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]));
919	cErrors++;
920	}
921	}
922	}
923	}
924	}
925	}
926	}
927	if (cErrors > 0)
928	return DBGCCmdHlpFail(pCmdHlp, pCmd, "Found %#x errors", cErrors);
929	return VINF_SUCCESS;
930	}
931	#endif /* VBOX_WITH_DEBUGGER */

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

source: vbox/trunk/src/VBox/VMM/VMMR3/PGMPool.cpp@ 64115

Download in other formats: