PGMPool.cpp@ 45530

Last change on this file since 45530 was 45103, checked in by vboxsync, 12 years ago
PGMPool: Eliminated fictive page (except NIL). #6367
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 52.3 KB

Line
1	/* $Id: PGMPool.cpp 45103 2013-03-20 11:13:27Z 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	/*
287	* The NIL entry.
288	*/
289	Assert(NIL_PGMPOOL_IDX == 0);
290	pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
291	pPool->aPages[NIL_PGMPOOL_IDX].idx = NIL_PGMPOOL_IDX;
292	pPool->aPages[NIL_PGMPOOL_IDX].Core.Key = NIL_RTHCPHYS;
293	pPool->aPages[NIL_PGMPOOL_IDX].GCPhys = NIL_RTGCPHYS;
294	pPool->aPages[NIL_PGMPOOL_IDX].iNext = NIL_PGMPOOL_IDX;
295	/* pPool->aPages[NIL_PGMPOOL_IDX].cLocked = INT32_MAX; - test this out... */
296	pPool->aPages[NIL_PGMPOOL_IDX].pvPageR3 = 0;
297	pPool->aPages[NIL_PGMPOOL_IDX].iUserHead = NIL_PGMPOOL_USER_INDEX;
298	pPool->aPages[NIL_PGMPOOL_IDX].iModifiedNext = NIL_PGMPOOL_IDX;
299	pPool->aPages[NIL_PGMPOOL_IDX].iModifiedPrev = NIL_PGMPOOL_IDX;
300	pPool->aPages[NIL_PGMPOOL_IDX].iMonitoredNext = NIL_PGMPOOL_IDX;
301	pPool->aPages[NIL_PGMPOOL_IDX].iMonitoredNext = NIL_PGMPOOL_IDX;
302	pPool->aPages[NIL_PGMPOOL_IDX].iAgeNext = NIL_PGMPOOL_IDX;
303	pPool->aPages[NIL_PGMPOOL_IDX].iAgePrev = NIL_PGMPOOL_IDX;
304
305	Assert(pPool->aPages[NIL_PGMPOOL_IDX].idx == NIL_PGMPOOL_IDX);
306	Assert(pPool->aPages[NIL_PGMPOOL_IDX].GCPhys == NIL_RTGCPHYS);
307	Assert(!pPool->aPages[NIL_PGMPOOL_IDX].fSeenNonGlobal);
308	Assert(!pPool->aPages[NIL_PGMPOOL_IDX].fMonitored);
309	Assert(!pPool->aPages[NIL_PGMPOOL_IDX].fCached);
310	Assert(!pPool->aPages[NIL_PGMPOOL_IDX].fZeroed);
311	Assert(!pPool->aPages[NIL_PGMPOOL_IDX].fReusedFlushPending);
312
313	#ifdef VBOX_WITH_STATISTICS
314	/*
315	* Register statistics.
316	*/
317	STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
318	STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
319	STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
320	STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
321	STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
322	STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolClearAll.");
323	STAM_REG(pVM, &pPool->StatR3Reset, STAMTYPE_PROFILE, "/PGM/Pool/R3Reset", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolReset.");
324	STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
325	STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
326	STAM_REG(pVM, &pPool->StatForceFlushPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForce", STAMUNIT_OCCURENCES, "Counting explicit flushes by PGMPoolFlushPage().");
327	STAM_REG(pVM, &pPool->StatForceFlushDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForceDirty", STAMUNIT_OCCURENCES, "Counting explicit flushes of dirty pages by PGMPoolFlushPage().");
328	STAM_REG(pVM, &pPool->StatForceFlushReused, STAMTYPE_COUNTER, "/PGM/Pool/FlushReused", STAMUNIT_OCCURENCES, "Counting flushes for reused pages.");
329	STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spent zeroing pages. Overlaps with Alloc.");
330	STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
331	STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
332	STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackDeref.");
333	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPT.");
334	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
335	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
336	STAM_REG(pVM, &pPool->StatTrackFlushEntry, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Flush", STAMUNIT_COUNT, "Nr of flushed entries.");
337	STAM_REG(pVM, &pPool->StatTrackFlushEntryKeep, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Update", STAMUNIT_COUNT, "Nr of updated entries.");
338	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.");
339	STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_TICKS_PER_CALL, "Profiling deref activity related tracking GC physical pages.");
340	STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
341	STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
342	STAM_REG(pVM, &pPool->StatMonitorRZ, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access handler.");
343	STAM_REG(pVM, &pPool->StatMonitorRZEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
344	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.");
345	STAM_REG(pVM, &pPool->StatMonitorRZFlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
346	STAM_REG(pVM, &pPool->StatMonitorRZFlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
347	STAM_REG(pVM, &pPool->StatMonitorRZFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
348	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).");
349	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
350	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.");
351	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.");
352	STAM_REG(pVM, &pPool->StatMonitorRZRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
353	STAM_REG(pVM, &pPool->StatMonitorRZFaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
354	STAM_REG(pVM, &pPool->StatMonitorRZFaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
355	STAM_REG(pVM, &pPool->StatMonitorRZFaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
356	STAM_REG(pVM, &pPool->StatMonitorRZFaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
357	STAM_REG(pVM, &pPool->StatMonitorR3, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access handler.");
358	STAM_REG(pVM, &pPool->StatMonitorR3EmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
359	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.");
360	STAM_REG(pVM, &pPool->StatMonitorR3FlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
361	STAM_REG(pVM, &pPool->StatMonitorR3FlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
362	STAM_REG(pVM, &pPool->StatMonitorR3Fork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
363	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).");
364	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.");
365	STAM_REG(pVM, &pPool->StatMonitorR3RepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
366	STAM_REG(pVM, &pPool->StatMonitorR3FaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
367	STAM_REG(pVM, &pPool->StatMonitorR3FaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
368	STAM_REG(pVM, &pPool->StatMonitorR3FaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
369	STAM_REG(pVM, &pPool->StatMonitorR3FaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
370	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.");
371	STAM_REG(pVM, &pPool->cModifiedPages, STAMTYPE_U16, "/PGM/Pool/Monitor/cModifiedPages", STAMUNIT_PAGES, "The current cModifiedPages value.");
372	STAM_REG(pVM, &pPool->cModifiedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/Monitor/cModifiedPagesHigh", STAMUNIT_PAGES, "The high watermark for cModifiedPages.");
373	STAM_REG(pVM, &pPool->StatResetDirtyPages, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Resets", STAMUNIT_OCCURENCES, "Times we've called pgmPoolResetDirtyPages (and there were dirty page).");
374	STAM_REG(pVM, &pPool->StatDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Pages", STAMUNIT_OCCURENCES, "Times we've called pgmPoolAddDirtyPage.");
375	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.");
376	STAM_REG(pVM, &pPool->StatDirtyPageOverFlowFlush, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/FlushOverflow",STAMUNIT_OCCURENCES, "Times we've had to flush because of overflow.");
377	STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
378	STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
379	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!)");
380	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.");
381	STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
382	STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
383	#endif /* VBOX_WITH_STATISTICS */
384
385	#ifdef VBOX_WITH_DEBUGGER
386	/*
387	* Debugger commands.
388	*/
389	static bool s_fRegisteredCmds = false;
390	if (!s_fRegisteredCmds)
391	{
392	rc = DBGCRegisterCommands(&g_aCmds[0], RT_ELEMENTS(g_aCmds));
393	if (RT_SUCCESS(rc))
394	s_fRegisteredCmds = true;
395	}
396	#endif
397
398	return VINF_SUCCESS;
399	}
400
401
402	/**
403	* Relocate the page pool data.
404	*
405	* @param pVM Pointer to the VM.
406	*/
407	void pgmR3PoolRelocate(PVM pVM)
408	{
409	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3);
410	pVM->pgm.s.pPoolR3->pVMRC = pVM->pVMRC;
411	pVM->pgm.s.pPoolR3->paUsersRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paUsersR3);
412	pVM->pgm.s.pPoolR3->paPhysExtsRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paPhysExtsR3);
413	int rc = PDMR3LdrGetSymbolRC(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerRC);
414	AssertReleaseRC(rc);
415	/* init order hack. */
416	if (!pVM->pgm.s.pPoolR3->pfnAccessHandlerR0)
417	{
418	rc = PDMR3LdrGetSymbolR0(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerR0);
419	AssertReleaseRC(rc);
420	}
421	}
422
423
424	/**
425	* Grows the shadow page pool.
426	*
427	* I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
428	*
429	* @returns VBox status code.
430	* @param pVM Pointer to the VM.
431	*/
432	VMMR3DECL(int) PGMR3PoolGrow(PVM pVM)
433	{
434	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
435	AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_PGM_POOL_MAXED_OUT_ALREADY);
436
437	/* With 32-bit guests and no EPT, the CR3 limits the root pages to low
438	(below 4 GB) memory. */
439	/** @todo change the pool to handle ROOT page allocations specially when
440	* required. */
441	bool fCanUseHighMemory = HMIsNestedPagingActive(pVM)
442	&& HMGetShwPagingMode(pVM) == PGMMODE_EPT;
443
444	pgmLock(pVM);
445
446	/*
447	* How much to grow it by?
448	*/
449	uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
450	cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
451	LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages. fCanUseHighMemory=%RTbool\n", cPages, cPages, fCanUseHighMemory));
452
453	for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
454	{
455	PPGMPOOLPAGE pPage = &pPool->aPages[i];
456
457	if (fCanUseHighMemory)
458	pPage->pvPageR3 = MMR3PageAlloc(pVM);
459	else
460	pPage->pvPageR3 = MMR3PageAllocLow(pVM);
461	if (!pPage->pvPageR3)
462	{
463	Log(("We're out of memory!! i=%d fCanUseHighMemory=%RTbool\n", i, fCanUseHighMemory));
464	pgmUnlock(pVM);
465	return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
466	}
467	pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageR3);
468	AssertFatal(pPage->Core.Key < _4G \|\| fCanUseHighMemory);
469	pPage->GCPhys = NIL_RTGCPHYS;
470	pPage->enmKind = PGMPOOLKIND_FREE;
471	pPage->idx = pPage - &pPool->aPages[0];
472	LogFlow(("PGMR3PoolGrow: insert page #%#x - %RHp\n", pPage->idx, pPage->Core.Key));
473	pPage->iNext = pPool->iFreeHead;
474	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
475	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
476	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
477	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
478	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
479	pPage->iAgeNext = NIL_PGMPOOL_IDX;
480	pPage->iAgePrev = NIL_PGMPOOL_IDX;
481	/* commit it */
482	bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
483	pPool->iFreeHead = i;
484	pPool->cCurPages = i + 1;
485	}
486
487	pgmUnlock(pVM);
488	Assert(pPool->cCurPages <= pPool->cMaxPages);
489	return VINF_SUCCESS;
490	}
491
492
493
494	/**
495	* Worker used by pgmR3PoolAccessHandler when it's invoked by an async thread.
496	*
497	* @param pPool The pool.
498	* @param pPage The page.
499	*/
500	static DECLCALLBACK(void) pgmR3PoolFlushReusedPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
501	{
502	/* for the present this should be safe enough I think... */
503	pgmLock(pPool->pVMR3);
504	if ( pPage->fReusedFlushPending
505	&& pPage->enmKind != PGMPOOLKIND_FREE)
506	pgmPoolFlushPage(pPool, pPage);
507	pgmUnlock(pPool->pVMR3);
508	}
509
510
511	/**
512	* \#PF Handler callback for PT write accesses.
513	*
514	* The handler can not raise any faults, it's mainly for monitoring write access
515	* to certain pages.
516	*
517	* @returns VINF_SUCCESS if the handler has carried out the operation.
518	* @returns VINF_PGM_HANDLER_DO_DEFAULT if the caller should carry out the access operation.
519	* @param pVM Pointer to the VM.
520	* @param GCPhys The physical address the guest is writing to.
521	* @param pvPhys The HC mapping of that address.
522	* @param pvBuf What the guest is reading/writing.
523	* @param cbBuf How much it's reading/writing.
524	* @param enmAccessType The access type.
525	* @param pvUser User argument.
526	*/
527	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf,
528	PGMACCESSTYPE enmAccessType, void *pvUser)
529	{
530	STAM_PROFILE_START(&pVM->pgm.s.pPoolR3->StatMonitorR3, a);
531	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
532	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
533	PVMCPU pVCpu = VMMGetCpu(pVM);
534	LogFlow(("pgmR3PoolAccessHandler: GCPhys=%RGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
535	GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
536
537	NOREF(pvBuf); NOREF(enmAccessType);
538
539	/*
540	* We don't have to be very sophisticated about this since there are relativly few calls here.
541	* However, we must try our best to detect any non-cpu accesses (disk / networking).
542	*
543	* Just to make life more interesting, we'll have to deal with the async threads too.
544	* We cannot flush a page if we're in an async thread because of REM notifications.
545	*/
546	pgmLock(pVM);
547	if (PHYS_PAGE_ADDRESS(GCPhys) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
548	{
549	/* Pool page changed while we were waiting for the lock; ignore. */
550	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)));
551	pgmUnlock(pVM);
552	return VINF_PGM_HANDLER_DO_DEFAULT;
553	}
554
555	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
556
557	/* @todo this code doesn't make any sense. remove the if (!pVCpu) block */
558	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} */
559	{
560	Log(("pgmR3PoolAccessHandler: async thread, requesting EMT to flush the page: %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
561	pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
562	STAM_COUNTER_INC(&pPool->StatMonitorR3Async);
563	if (!pPage->fReusedFlushPending)
564	{
565	pgmUnlock(pVM);
566	int rc = VMR3ReqCallVoidNoWait(pPool->pVMR3, VMCPUID_ANY, (PFNRT)pgmR3PoolFlushReusedPage, 2, pPool, pPage);
567	AssertRCReturn(rc, rc);
568	pgmLock(pVM);
569	pPage->fReusedFlushPending = true;
570	pPage->cModifications += 0x1000;
571	}
572
573	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
574	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
575	while (cbBuf > 4)
576	{
577	cbBuf -= 4;
578	pvPhys = (uint8_t *)pvPhys + 4;
579	GCPhys += 4;
580	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
581	}
582	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
583	}
584	else if ( ( pPage->cModifications < 96 /* it's cheaper here. */
585	\|\| pgmPoolIsPageLocked(pPage)
586	)
587	&& cbBuf <= 4)
588	{
589	/* Clear the shadow entry. */
590	if (!pPage->cModifications++)
591	pgmPoolMonitorModifiedInsert(pPool, pPage);
592	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
593	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
594	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
595	}
596	else
597	{
598	pgmPoolMonitorChainFlush(pPool, pPage); /* ASSUME that VERR_PGM_POOL_CLEARED can be ignored here and that FFs will deal with it in due time. */
599	STAM_PROFILE_STOP_EX(&pPool->StatMonitorR3, &pPool->StatMonitorR3FlushPage, a);
600	}
601	pgmUnlock(pVM);
602	return VINF_PGM_HANDLER_DO_DEFAULT;
603	}
604
605
606	/**
607	* Rendezvous callback used by pgmR3PoolClearAll that clears all shadow pages
608	* and all modification counters.
609	*
610	* This is only called on one of the EMTs while the other ones are waiting for
611	* it to complete this function.
612	*
613	* @returns VINF_SUCCESS (VBox strict status code).
614	* @param pVM Pointer to the VM.
615	* @param pVCpu The VMCPU for the EMT we're being called on. Unused.
616	* @param fpvFlushRemTlb When not NULL, we'll flush the REM TLB as well.
617	* (This is the pvUser, so it has to be void *.)
618	*
619	*/
620	DECLCALLBACK(VBOXSTRICTRC) pgmR3PoolClearAllRendezvous(PVM pVM, PVMCPU pVCpu, void *fpvFlushRemTbl)
621	{
622	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
623	STAM_PROFILE_START(&pPool->StatClearAll, c);
624	NOREF(pVCpu);
625
626	pgmLock(pVM);
627	Log(("pgmR3PoolClearAllRendezvous: cUsedPages=%d fpvFlushRemTbl=%RTbool\n", pPool->cUsedPages, !!fpvFlushRemTbl));
628
629	/*
630	* Iterate all the pages until we've encountered all that are in use.
631	* This is a simple but not quite optimal solution.
632	*/
633	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
634	unsigned cLeft = pPool->cUsedPages;
635	uint32_t iPage = pPool->cCurPages;
636	while (--iPage >= PGMPOOL_IDX_FIRST)
637	{
638	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
639	if (pPage->GCPhys != NIL_RTGCPHYS)
640	{
641	switch (pPage->enmKind)
642	{
643	/*
644	* We only care about shadow page tables that reference physical memory
645	*/
646	#ifdef PGM_WITH_LARGE_PAGES
647	case PGMPOOLKIND_EPT_PD_FOR_PHYS: /* Large pages reference 2 MB of physical memory, so we must clear them. */
648	if (pPage->cPresent)
649	{
650	PX86PDPAE pShwPD = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
651	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
652	{
653	if ( pShwPD->a[i].n.u1Present
654	&& pShwPD->a[i].b.u1Size)
655	{
656	Assert(!(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING));
657	pShwPD->a[i].u = 0;
658	Assert(pPage->cPresent);
659	pPage->cPresent--;
660	}
661	}
662	if (pPage->cPresent == 0)
663	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
664	}
665	goto default_case;
666
667	case PGMPOOLKIND_PAE_PD_PHYS: /* Large pages reference 2 MB of physical memory, so we must clear them. */
668	if (pPage->cPresent)
669	{
670	PEPTPD pShwPD = (PEPTPD)PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
671	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
672	{
673	Assert((pShwPD->a[i].u & UINT64_C(0xfff0000000000f80)) == 0);
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	#endif /* PGM_WITH_LARGE_PAGES */
688
689	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
690	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
691	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
692	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
693	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
694	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
695	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
696	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
697	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
698	{
699	if (pPage->cPresent)
700	{
701	void *pvShw = PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
702	STAM_PROFILE_START(&pPool->StatZeroPage, z);
703	#if 0
704	/* Useful check for leaking references; very expensive though. */
705	switch (pPage->enmKind)
706	{
707	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
708	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
709	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
710	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
711	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
712	{
713	bool fFoundFirst = false;
714	PPGMSHWPTPAE pPT = (PPGMSHWPTPAE)pvShw;
715	for (unsigned ptIndex = 0; ptIndex < RT_ELEMENTS(pPT->a); ptIndex++)
716	{
717	if (pPT->a[ptIndex].u)
718	{
719	if (!fFoundFirst)
720	{
721	AssertFatalMsg(pPage->iFirstPresent <= ptIndex, ("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
722	if (pPage->iFirstPresent != ptIndex)
723	Log(("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
724	fFoundFirst = true;
725	}
726	if (PGMSHWPTEPAE_IS_P(pPT->a[ptIndex]))
727	{
728	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pPT->a[ptIndex]), NIL_RTGCPHYS);
729	if (pPage->iFirstPresent == ptIndex)
730	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
731	}
732	}
733	}
734	AssertFatalMsg(pPage->cPresent == 0, ("cPresent = %d pPage = %RGv\n", pPage->cPresent, pPage->GCPhys));
735	break;
736	}
737	default:
738	break;
739	}
740	#endif
741	ASMMemZeroPage(pvShw);
742	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
743	pPage->cPresent = 0;
744	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
745	}
746	}
747	/* fall thru */
748
749	#ifdef PGM_WITH_LARGE_PAGES
750	default_case:
751	#endif
752	default:
753	Assert(!pPage->cModifications \|\| ++cModifiedPages);
754	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
755	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
756	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
757	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
758	pPage->cModifications = 0;
759	break;
760
761	}
762	if (!--cLeft)
763	break;
764	}
765	}
766
767	#ifndef DEBUG_michael
768	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
769	#endif
770	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
771	pPool->cModifiedPages = 0;
772
773	/*
774	* Clear all the GCPhys links and rebuild the phys ext free list.
775	*/
776	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRangesX);
777	pRam;
778	pRam = pRam->CTX_SUFF(pNext))
779	{
780	iPage = pRam->cb >> PAGE_SHIFT;
781	while (iPage-- > 0)
782	PGM_PAGE_SET_TRACKING(pVM, &pRam->aPages[iPage], 0);
783	}
784
785	pPool->iPhysExtFreeHead = 0;
786	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
787	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
788	for (unsigned i = 0; i < cMaxPhysExts; i++)
789	{
790	paPhysExts[i].iNext = i + 1;
791	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
792	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
793	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
794	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
795	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
796	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
797	}
798	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
799
800
801	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
802	/* Reset all dirty pages to reactivate the page monitoring. */
803	/* Note: we must do this after clearing all page references and shadow page tables as there might be stale references to
804	* recently removed MMIO ranges around that might otherwise end up asserting in pgmPoolTracDerefGCPhysHint
805	*/
806	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
807	{
808	PPGMPOOLPAGE pPage;
809	unsigned idxPage;
810
811	if (pPool->aDirtyPages[i].uIdx == NIL_PGMPOOL_IDX)
812	continue;
813
814	idxPage = pPool->aDirtyPages[i].uIdx;
815	AssertRelease(idxPage != NIL_PGMPOOL_IDX);
816	pPage = &pPool->aPages[idxPage];
817	Assert(pPage->idx == idxPage);
818	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
819
820	AssertMsg(pPage->fDirty, ("Page %RGp (slot=%d) not marked dirty!", pPage->GCPhys, i));
821
822	Log(("Reactivate dirty page %RGp\n", pPage->GCPhys));
823
824	/* First write protect the page again to catch all write accesses. (before checking for changes -> SMP) */
825	int rc = PGMHandlerPhysicalReset(pVM, pPage->GCPhys & PAGE_BASE_GC_MASK);
826	AssertRCSuccess(rc);
827	pPage->fDirty = false;
828
829	pPool->aDirtyPages[i].uIdx = NIL_PGMPOOL_IDX;
830	}
831
832	/* Clear all dirty pages. */
833	pPool->idxFreeDirtyPage = 0;
834	pPool->cDirtyPages = 0;
835	#endif
836
837	/* Clear the PGM_SYNC_CLEAR_PGM_POOL flag on all VCPUs to prevent redundant flushes. */
838	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
839	pVM->aCpus[idCpu].pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
840
841	/* Flush job finished. */
842	VM_FF_CLEAR(pVM, VM_FF_PGM_POOL_FLUSH_PENDING);
843	pPool->cPresent = 0;
844	pgmUnlock(pVM);
845
846	PGM_INVL_ALL_VCPU_TLBS(pVM);
847
848	if (fpvFlushRemTbl)
849	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
850	CPUMSetChangedFlags(&pVM->aCpus[idCpu], CPUM_CHANGED_GLOBAL_TLB_FLUSH);
851
852	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
853	return VINF_SUCCESS;
854	}
855
856
857	/**
858	* Clears the shadow page pool.
859	*
860	* @param pVM Pointer to the VM.
861	* @param fFlushRemTlb When set, the REM TLB is scheduled for flushing as
862	* well.
863	*/
864	void pgmR3PoolClearAll(PVM pVM, bool fFlushRemTlb)
865	{
866	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmR3PoolClearAllRendezvous, &fFlushRemTlb);
867	AssertRC(rc);
868	}
869
870
871	/**
872	* Protect all pgm pool page table entries to monitor writes
873	*
874	* @param pVM Pointer to the VM.
875	*
876	* @remarks ASSUMES the caller will flush all TLBs!!
877	*/
878	void pgmR3PoolWriteProtectPages(PVM pVM)
879	{
880	PGM_LOCK_ASSERT_OWNER(pVM);
881	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
882	unsigned cLeft = pPool->cUsedPages;
883	unsigned iPage = pPool->cCurPages;
884	while (--iPage >= PGMPOOL_IDX_FIRST)
885	{
886	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
887	if ( pPage->GCPhys != NIL_RTGCPHYS
888	&& pPage->cPresent)
889	{
890	union
891	{
892	void *pv;
893	PX86PT pPT;
894	PPGMSHWPTPAE pPTPae;
895	PEPTPT pPTEpt;
896	} uShw;
897	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
898
899	switch (pPage->enmKind)
900	{
901	/*
902	* We only care about shadow page tables.
903	*/
904	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
905	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
906	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
907	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPT->a); iShw++)
908	{
909	if (uShw.pPT->a[iShw].n.u1Present)
910	uShw.pPT->a[iShw].n.u1Write = 0;
911	}
912	break;
913
914	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
915	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
916	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
917	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
918	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
919	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTPae->a); iShw++)
920	{
921	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw]))
922	PGMSHWPTEPAE_SET_RO(uShw.pPTPae->a[iShw]);
923	}
924	break;
925
926	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
927	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTEpt->a); iShw++)
928	{
929	if (uShw.pPTEpt->a[iShw].n.u1Present)
930	uShw.pPTEpt->a[iShw].n.u1Write = 0;
931	}
932	break;
933
934	default:
935	break;
936	}
937	if (!--cLeft)
938	break;
939	}
940	}
941	}
942
943	#ifdef VBOX_WITH_DEBUGGER
944	/**
945	* @callback_method_impl{FNDBGCCMD, The '.pgmpoolcheck' command.}
946	*/
947	static DECLCALLBACK(int) pgmR3PoolCmdCheck(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PUVM pUVM, PCDBGCVAR paArgs, unsigned cArgs)
948	{
949	DBGC_CMDHLP_REQ_UVM_RET(pCmdHlp, pCmd, pUVM);
950	PVM pVM = pUVM->pVM;
951	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
952	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs == 0);
953	uint32_t cErrors = 0;
954	NOREF(paArgs);
955
956	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
957	for (unsigned i = 0; i < pPool->cCurPages; i++)
958	{
959	PPGMPOOLPAGE pPage = &pPool->aPages[i];
960	bool fFirstMsg = true;
961
962	/* Todo: cover other paging modes too. */
963	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
964	{
965	PPGMSHWPTPAE pShwPT = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
966	{
967	PX86PTPAE pGstPT;
968	PGMPAGEMAPLOCK LockPage;
969	int rc = PGMPhysGCPhys2CCPtrReadOnly(pVM, pPage->GCPhys, (const void **)&pGstPT, &LockPage); AssertReleaseRC(rc);
970
971	/* Check if any PTEs are out of sync. */
972	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
973	{
974	if (PGMSHWPTEPAE_IS_P(pShwPT->a[j]))
975	{
976	RTHCPHYS HCPhys = NIL_RTHCPHYS;
977	rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[j].u & X86_PTE_PAE_PG_MASK, &HCPhys);
978	if ( rc != VINF_SUCCESS
979	\|\| PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[j]) != HCPhys)
980	{
981	if (fFirstMsg)
982	{
983	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
984	fFirstMsg = false;
985	}
986	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);
987	cErrors++;
988	}
989	else if ( PGMSHWPTEPAE_IS_RW(pShwPT->a[j])
990	&& !pGstPT->a[j].n.u1Write)
991	{
992	if (fFirstMsg)
993	{
994	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
995	fFirstMsg = false;
996	}
997	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);
998	cErrors++;
999	}
1000	}
1001	}
1002	PGMPhysReleasePageMappingLock(pVM, &LockPage);
1003	}
1004
1005	/* Make sure this page table can't be written to from any shadow mapping. */
1006	RTHCPHYS HCPhysPT = NIL_RTHCPHYS;
1007	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pPage->GCPhys, &HCPhysPT);
1008	AssertMsgRC(rc, ("PGMPhysGCPhys2HCPhys failed with rc=%d for %RGp\n", rc, pPage->GCPhys));
1009	if (rc == VINF_SUCCESS)
1010	{
1011	for (unsigned j = 0; j < pPool->cCurPages; j++)
1012	{
1013	PPGMPOOLPAGE pTempPage = &pPool->aPages[j];
1014
1015	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1016	{
1017	PPGMSHWPTPAE pShwPT2 = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pTempPage);
1018
1019	for (unsigned k = 0; k < RT_ELEMENTS(pShwPT->a); k++)
1020	{
1021	if ( PGMSHWPTEPAE_IS_P_RW(pShwPT2->a[k])
1022	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1023	&& !pPage->fDirty
1024	# endif
1025	&& PGMSHWPTEPAE_GET_HCPHYS(pShwPT2->a[k]) == HCPhysPT)
1026	{
1027	if (fFirstMsg)
1028	{
1029	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
1030	fFirstMsg = false;
1031	}
1032	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]));
1033	cErrors++;
1034	}
1035	}
1036	}
1037	}
1038	}
1039	}
1040	}
1041	if (cErrors > 0)
1042	return DBGCCmdHlpFail(pCmdHlp, pCmd, "Found %#x errors", cErrors);
1043	return VINF_SUCCESS;
1044	}
1045	#endif /* VBOX_WITH_DEBUGGER */

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

source: vbox/trunk/src/VBox/VMM/VMMR3/PGMPool.cpp@ 45530

Download in other formats: