MM.cpp@ 2296

Last change on this file since 2296 was 1890, checked in by vboxsync, 18 years ago
Attempt to fix ALSA on Linux kernels <= 2.6.17: use mmap not memalign for allocating pages. Use madvise or mprotect to separater VM area structs inside the kernel. Most SUP* functions work on cPages now (not cBytes anymore). The free functions take a cPages parameter which is used for munmap on Linux.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 18.7 KB

Line
1	/* $Id: MM.cpp 1890 2007-04-03 16:04:19Z vboxsync $ */
2	/** @file
3	* MM - Memory Monitor(/Manager).
4	*/
5
6	/*
7	* Copyright (C) 2006 InnoTek Systemberatung GmbH
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 as published by the Free Software Foundation,
13	* in version 2 as it comes in the "COPYING" file of the VirtualBox OSE
14	* distribution. VirtualBox OSE is distributed in the hope that it will
15	* be useful, but WITHOUT ANY WARRANTY of any kind.
16	*
17	* If you received this file as part of a commercial VirtualBox
18	* distribution, then only the terms of your commercial VirtualBox
19	* license agreement apply instead of the previous paragraph.
20	*/
21
22
23	/** @page pg_mm MM - The Memory Monitor/Manager
24	*
25	* It seems like this is going to be the entity taking care of memory allocations
26	* and the locking of physical memory for a VM. MM will track these allocations and
27	* pinnings so pointer conversions, memory read and write, and correct clean up can
28	* be done.
29	*
30	* Memory types:
31	* - Hypervisor Memory Area (HMA).
32	* - Page tables.
33	* - Physical pages.
34	*
35	* The first two types are not accessible using the generic conversion functions
36	* for GC memory, there are special functions for these.
37	*
38	*
39	* A decent structure for this component need to be eveloped as we see usage. One
40	* or two rewrites is probabaly needed to get it right...
41	*
42	*
43	*
44	* @section Hypervisor Memory Area
45	*
46	* The hypervisor is give 4MB of space inside the guest, we assume that we can
47	* steal an page directory entry from the guest OS without cause trouble. In
48	* addition to these 4MB we'll be mapping memory for the graphics emulation,
49	* but that will be an independant mapping.
50	*
51	* The 4MBs are divided into two main parts:
52	* -# The static code and data
53	* -# The shortlived page mappings.
54	*
55	* The first part is used for the VM structure, the core code (VMMSwitch),
56	* GC modules, and the alloc-only-heap. The size will be determined at a
57	* later point but initially we'll say 2MB of locked memory, most of which
58	* is non contiguous physically.
59	*
60	* The second part is used for mapping pages to the hypervisor. We'll be using
61	* a simple round robin when doing these mappings. This means that no-one can
62	* assume that a mapping hangs around for very long, while the managing of the
63	* pages are very simple.
64	*
65	*
66	*
67	* @section Page Pool
68	*
69	* The MM manages a per VM page pool from which other components can allocate
70	* locked, page aligned and page granular memory objects. The pool provides
71	* facilities to convert back and forth between physical and virtual addresses
72	* (within the pool of course). Several specialized interfaces are provided
73	* for the most common alloctions and convertions to save the caller from
74	* bothersome casting and extra parameter passing.
75	*
76	*
77	*/
78
79
80
81	/*******************************************************************************
82	* Header Files *
83	*******************************************************************************/
84	#define LOG_GROUP LOG_GROUP_MM
85	#include <VBox/mm.h>
86	#include <VBox/pgm.h>
87	#include <VBox/cfgm.h>
88	#include <VBox/ssm.h>
89	#include "MMInternal.h"
90	#include <VBox/vm.h>
91	#include <VBox/err.h>
92	#include <VBox/param.h>
93
94	#include <VBox/log.h>
95	#include <iprt/alloc.h>
96	#include <iprt/assert.h>
97	#include <iprt/string.h>
98
99
100	/*******************************************************************************
101	* Internal Functions *
102	*******************************************************************************/
103	static int mmR3Term(PVM pVM, bool fKeepTheHeap);
104	static DECLCALLBACK(int) mmR3Save(PVM pVM, PSSMHANDLE pSSM);
105	static DECLCALLBACK(int) mmR3Load(PVM pVM, PSSMHANDLE pSSM, uint32_t u32Version);
106
107
108
109	/**
110	* Initializes the MM.
111	*
112	* MM is managing the virtual address space (among other things) and
113	* setup the hypvervisor memory area mapping in the VM structure and
114	* the hypvervisor alloc-only-heap. Assuming the current init order
115	* and components the hypvervisor memory area looks like this:
116	* -# VM Structure.
117	* -# Hypervisor alloc only heap (also call Hypervisor memory region).
118	* -# Core code.
119	*
120	* MM determins the virtual address of the hypvervisor memory area by
121	* checking for location at previous run. If that property isn't available
122	* it will choose a default starting location, currently 0xe0000000.
123	*
124	* @returns VBox status code.
125	* @param pVM The VM to operate on.
126	*/
127	MMR3DECL(int) MMR3Init(PVM pVM)
128	{
129	LogFlow(("MMR3Init\n"));
130
131	/*
132	* Assert alignment, sizes and order.
133	*/
134	AssertRelease(!(RT_OFFSETOF(VM, mm.s) & 31));
135	AssertRelease(sizeof(pVM->mm.s) <= sizeof(pVM->mm.padding));
136	AssertMsg(pVM->mm.s.offVM == 0, ("Already initialized!\n"));
137
138	/*
139	* Init the structure.
140	*/
141	pVM->mm.s.offVM = RT_OFFSETOF(VM, mm);
142	pVM->mm.s.offLookupHyper = NIL_OFFSET;
143
144	/*
145	* Init the heap (may already be initialized already if someone used it).
146	*/
147	if (!pVM->mm.s.pHeap)
148	{
149	int rc = mmr3HeapCreate(pVM, &pVM->mm.s.pHeap);
150	if (!VBOX_SUCCESS(rc))
151	return rc;
152	}
153
154	/*
155	* Init the page pool.
156	*/
157	int rc = mmr3PagePoolInit(pVM);
158	if (VBOX_SUCCESS(rc))
159	{
160	/*
161	* Init the hypervisor related stuff.
162	*/
163	rc = mmr3HyperInit(pVM);
164	if (VBOX_SUCCESS(rc))
165	{
166	/*
167	* Register the saved state data unit.
168	*/
169	rc = SSMR3RegisterInternal(pVM, "mm", 1, 1, sizeof(uint32_t) * 2,
170	NULL, mmR3Save, NULL,
171	NULL, mmR3Load, NULL);
172	if (VBOX_SUCCESS(rc))
173	return rc;
174	}
175
176	/* .... failure .... */
177	mmR3Term(pVM, true /* keep the heap */);
178	}
179	else
180	mmr3HeapDestroy(pVM->mm.s.pHeap);
181	return rc;
182	}
183
184
185	/**
186	* Initializes the MM parts which depends on PGM being initialized.
187	*
188	* @returns VBox status code.
189	* @param pVM The VM to operate on.
190	* @remark No cleanup necessary since MMR3Term() will be called on failure.
191	*/
192	MMR3DECL(int) MMR3InitPaging(PVM pVM)
193	{
194	LogFlow(("MMR3InitPaging:\n"));
195	bool fPreAlloc;
196	int rc = CFGMR3QueryBool(CFGMR3GetRoot(pVM), "RamPreAlloc", &fPreAlloc);
197	if (rc == VERR_CFGM_VALUE_NOT_FOUND)
198	fPreAlloc = false;
199	else
200	AssertMsgRCReturn(rc, ("Configuration error: Failed to query integer \"RamPreAlloc\", rc=%Vrc.\n", rc), rc);
201
202	uint64_t cbRam;
203	rc = CFGMR3QueryU64(CFGMR3GetRoot(pVM), "RamSize", &cbRam);
204	if (rc == VERR_CFGM_VALUE_NOT_FOUND)
205	cbRam = 0;
206	if (VBOX_SUCCESS(rc) \|\| rc == VERR_CFGM_VALUE_NOT_FOUND)
207	{
208	if (cbRam < PAGE_SIZE)
209	{
210	Log(("MM: No RAM configured\n"));
211	return VINF_SUCCESS;
212	}
213	#ifdef PGM_DYNAMIC_RAM_ALLOC
214	Log(("MM: %llu bytes of RAM%s\n", cbRam, fPreAlloc ? " (PreAlloc)" : ""));
215	pVM->mm.s.pvRamBaseHC = 0; /** @todo obsolete */
216	pVM->mm.s.cbRamBase = cbRam & PAGE_BASE_GC_MASK;
217	rc = MMR3PhysRegister(pVM, pVM->mm.s.pvRamBaseHC, 0, pVM->mm.s.cbRamBase, MM_RAM_FLAGS_DYNAMIC_ALLOC, "Main Memory");
218	if (VBOX_SUCCESS(rc))
219	{
220	/* Allocate the first chunk, as we'll map ROM ranges there. */
221	rc = PGM3PhysGrowRange(pVM, (RTGCPHYS)0);
222	if (VBOX_SUCCESS(rc))
223	{
224	/* Should we preallocate the entire guest RAM? */
225	if (fPreAlloc)
226	{
227	for (RTGCPHYS GCPhys = PGM_DYNAMIC_CHUNK_SIZE; GCPhys < cbRam; GCPhys += PGM_DYNAMIC_CHUNK_SIZE)
228	{
229	rc = PGM3PhysGrowRange(pVM, GCPhys);
230	if (VBOX_FAILURE(rc))
231	return rc;
232	}
233	}
234	return rc;
235	}
236	}
237	#else
238	unsigned cPages = cbRam >> PAGE_SHIFT;
239	Log(("MM: %llu bytes of RAM (%d pages)\n", cbRam, cPages));
240	rc = SUPPageAlloc(cPages, &pVM->mm.s.pvRamBaseHC);
241	if (VBOX_SUCCESS(rc))
242	{
243	pVM->mm.s.cbRamBase = cPages << PAGE_SHIFT;
244	rc = MMR3PhysRegister(pVM, pVM->mm.s.pvRamBaseHC, 0, pVM->mm.s.cbRamBase, 0, "Main Memory");
245	if (VBOX_SUCCESS(rc))
246	return rc;
247	SUPPageFree(pVM->mm.s.pvRamBaseHC);
248	}
249	else
250	LogRel(("MMR3InitPage: Failed to allocate %u bytes of RAM! rc=%Vrc\n", cPages << PAGE_SHIFT));
251	#endif
252	}
253	else
254	AssertMsgFailed(("Configuration error: Failed to query integer \"RamSize\", rc=%Vrc.\n", rc));
255
256	LogFlow(("MMR3InitPaging: returns %Vrc\n", rc));
257	return rc;
258	}
259
260
261	/**
262	* Terminates the MM.
263	*
264	* Termination means cleaning up and freeing all resources,
265	* the VM it self is at this point powered off or suspended.
266	*
267	* @returns VBox status code.
268	* @param pVM The VM to operate on.
269	*/
270	MMR3DECL(int) MMR3Term(PVM pVM)
271	{
272	return mmR3Term(pVM, false /* free the heap */);
273	}
274
275
276	/**
277	* Worker for MMR3Term and MMR3Init.
278	*
279	* The tricky bit here is that we must not destroy the heap if we're
280	* called from MMR3Init, otherwise we'll get into trouble when
281	* CFGMR3Term is called later in the bailout process.
282	*
283	* @returns VBox status code.
284	* @param pVM The VM to operate on.
285	* @param fKeepTheHeap Whether or not to keep the heap.
286	*/
287	static int mmR3Term(PVM pVM, bool fKeepTheHeap)
288	{
289	/*
290	* Release locked memory.
291	* (Associated record are released by the heap.)
292	*/
293	PMMLOCKEDMEM pLockedMem = pVM->mm.s.pLockedMem;
294	while (pLockedMem)
295	{
296	int rc = SUPPageUnlock(pLockedMem->pv);
297	AssertMsgRC(rc, ("SUPPageUnlock(%p) -> rc=%d\n", pLockedMem->pv, rc));
298	switch (pLockedMem->eType)
299	{
300	case MM_LOCKED_TYPE_HYPER:
301	rc = SUPPageFree(pLockedMem->pv, pLockedMem->cb >> PAGE_SHIFT);
302	AssertMsgRC(rc, ("SUPPageFree(%p) -> rc=%d\n", pLockedMem->pv, rc));
303	break;
304	case MM_LOCKED_TYPE_HYPER_NOFREE:
305	case MM_LOCKED_TYPE_HYPER_PAGES:
306	case MM_LOCKED_TYPE_PHYS:
307	/* nothing to do. */
308	break;
309	}
310	/* next */
311	pLockedMem = pLockedMem->pNext;
312	}
313
314	/*
315	* Destroy the page pool.
316	*/
317	mmr3PagePoolTerm(pVM);
318
319	/*
320	* Destroy the heap if requested.
321	*/
322	if (!fKeepTheHeap)
323	{
324	mmr3HeapDestroy(pVM->mm.s.pHeap);
325	pVM->mm.s.pHeap = NULL;
326	}
327
328	/*
329	* Zero stuff to detect after termination use of the MM interface
330	*/
331	pVM->mm.s.offLookupHyper = NIL_OFFSET;
332	pVM->mm.s.pLockedMem = NULL;
333	pVM->mm.s.pHyperHeapHC = NULL; /* freed above. */
334	pVM->mm.s.pHyperHeapGC = 0; /* freed above. */
335	pVM->mm.s.offVM = 0; /* init assertion on this */
336
337	return 0;
338	}
339
340
341	/**
342	* Execute state save operation.
343	*
344	* @returns VBox status code.
345	* @param pVM VM Handle.
346	* @param pSSM SSM operation handle.
347	*/
348	static DECLCALLBACK(int) mmR3Save(PVM pVM, PSSMHANDLE pSSM)
349	{
350	LogFlow(("mmR3Save:\n"));
351
352	/* (PGM saves the physical memory.) */
353	SSMR3PutUInt(pSSM, pVM->mm.s.cbRAMSize);
354	return SSMR3PutUInt(pSSM, pVM->mm.s.cbRamBase);
355	}
356
357
358	/**
359	* Execute state load operation.
360	*
361	* @returns VBox status code.
362	* @param pVM VM Handle.
363	* @param pSSM SSM operation handle.
364	* @param u32Version Data layout version.
365	*/
366	static DECLCALLBACK(int) mmR3Load(PVM pVM, PSSMHANDLE pSSM, uint32_t u32Version)
367	{
368	LogFlow(("mmR3Load:\n"));
369
370	/*
371	* Validate version.
372	*/
373	if (u32Version != 1)
374	{
375	Log(("mmR3Load: Invalid version u32Version=%d!\n", u32Version));
376	return VERR_SSM_UNSUPPORTED_DATA_UNIT_VERSION;
377	}
378
379	/*
380	* Check the cbRAMSize and cbRamBase values.
381	*/
382	RTUINT cb;
383	int rc = SSMR3GetUInt(pSSM, &cb);
384	if (VBOX_FAILURE(rc))
385	return rc;
386	if (cb != pVM->mm.s.cbRAMSize)
387	{
388	Log(("mmR3Load: Memory configuration has changed. cbRAMSize=%#x save %#x\n", pVM->mm.s.cbRAMSize, cb));
389	return VERR_SSM_LOAD_MEMORY_SIZE_MISMATCH;
390	}
391
392	rc = SSMR3GetUInt(pSSM, &cb);
393	if (VBOX_FAILURE(rc))
394	return rc;
395	if (cb != pVM->mm.s.cbRamBase)
396	{
397	Log(("mmR3Load: Memory configuration has changed. cbRamBase=%#x save %#x\n", pVM->mm.s.cbRamBase, cb));
398	return VERR_SSM_LOAD_MEMORY_SIZE_MISMATCH;
399	}
400
401	/* PGM restores the physical memory. */
402	return rc;
403	}
404
405
406	/**
407	* Locks physical memory which backs a virtual memory range (HC) adding
408	* the required records to the pLockedMem list.
409	*
410	* @returns VBox status code.
411	* @param pVM The VM handle.
412	* @param pv Pointer to memory range which shall be locked down.
413	* This pointer is page aligned.
414	* @param cb Size of memory range (in bytes). This size is page aligned.
415	* @param eType Memory type.
416	* @param ppLockedMem Where to store the pointer to the created locked memory record.
417	* This is optional, pass NULL if not used.
418	* @param fSilentFailure Don't raise an error when unsuccessful. Upper layer with deal with it.
419	*/
420	int mmr3LockMem(PVM pVM, void pv, size_t cb, MMLOCKEDTYPE eType, PMMLOCKEDMEM ppLockedMem, bool fSilentFailure)
421	{
422	Assert(RT_ALIGN_P(pv, PAGE_SIZE) == pv);
423	Assert(RT_ALIGN_Z(cb, PAGE_SIZE) == cb);
424
425	if (ppLockedMem)
426	*ppLockedMem = NULL;
427
428	/*
429	* Allocate locked mem structure.
430	*/
431	unsigned cPages = cb >> PAGE_SHIFT;
432	AssertReturn(cPages == (cb >> PAGE_SHIFT), VERR_OUT_OF_RANGE);
433	PMMLOCKEDMEM pLockedMem = (PMMLOCKEDMEM)MMR3HeapAlloc(pVM, MM_TAG_MM, RT_OFFSETOF(MMLOCKEDMEM, aPhysPages[cPages]));
434	if (!pLockedMem)
435	return VERR_NO_MEMORY;
436	pLockedMem->pv = pv;
437	pLockedMem->cb = cb;
438	pLockedMem->eType = eType;
439	memset(&pLockedMem->u, 0, sizeof(pLockedMem->u));
440
441	/*
442	* Lock the memory.
443	*/
444	int rc = SUPPageLock(pv, cPages, &pLockedMem->aPhysPages[0]);
445	if (VBOX_SUCCESS(rc))
446	{
447	/*
448	* Setup the reserved field.
449	*/
450	PSUPPAGE pPhysPage = &pLockedMem->aPhysPages[0];
451	for (unsigned c = cPages; c > 0; c--, pPhysPage++)
452	pPhysPage->uReserved = (RTHCUINTPTR)pLockedMem;
453
454	/*
455	* Insert into the list.
456	*
457	* ASSUME no protected needed here as only one thread in the system can possibly
458	* be doing this. No other threads will walk this list either we assume.
459	*/
460	pLockedMem->pNext = pVM->mm.s.pLockedMem;
461	pVM->mm.s.pLockedMem = pLockedMem;
462	/* Set return value. */
463	if (ppLockedMem)
464	*ppLockedMem = pLockedMem;
465	}
466	else
467	{
468	AssertMsgFailed(("SUPPageLock failed with rc=%d\n", rc));
469	MMR3HeapFree(pLockedMem);
470	if (!fSilentFailure)
471	rc = VMSetError(pVM, rc, RT_SRC_POS, N_("Failed to lock %d bytes of host memory (out of memory)"), cb);
472	}
473
474	return rc;
475	}
476
477
478	/**
479	* Maps a part of or an entire locked memory region into the guest context.
480	*
481	* @returns VBox status.
482	* God knows what happens if we fail...
483	* @param pVM VM handle.
484	* @param pLockedMem Locked memory structure.
485	* @param Addr GC Address where to start the mapping.
486	* @param iPage Page number in the locked memory region.
487	* @param cPages Number of pages to map.
488	* @param fFlags See the fFlags argument of PGR3Map().
489	*/
490	int mmr3MapLocked(PVM pVM, PMMLOCKEDMEM pLockedMem, RTGCPTR Addr, unsigned iPage, size_t cPages, unsigned fFlags)
491	{
492	/*
493	* Adjust ~0 argument
494	*/
495	if (cPages == ~(size_t)0)
496	cPages = (pLockedMem->cb >> PAGE_SHIFT) - iPage;
497	Assert(cPages != ~0U);
498	/* no incorrect arguments are accepted */
499	Assert(RT_ALIGN_GCPT(Addr, PAGE_SIZE, RTGCPTR) == Addr);
500	AssertMsg(iPage < (pLockedMem->cb >> PAGE_SHIFT), ("never even think about giving me a bad iPage(=%d)\n", iPage));
501	AssertMsg(iPage + cPages <= (pLockedMem->cb >> PAGE_SHIFT), ("never even think about giving me a bad cPages(=%d)\n", cPages));
502
503	/*
504	* Map the the pages.
505	*/
506	PSUPPAGE pPhysPage = &pLockedMem->aPhysPages[iPage];
507	while (cPages)
508	{
509	RTHCPHYS HCPhys = pPhysPage->Phys;
510	int rc = PGMMap(pVM, Addr, HCPhys, PAGE_SIZE, fFlags);
511	if (VBOX_FAILURE(rc))
512	{
513	/** @todo how the hell can we do a proper bailout here. */
514	return rc;
515	}
516
517	/* next */
518	cPages--;
519	iPage++;
520	pPhysPage++;
521	Addr += PAGE_SIZE;
522	}
523
524	return VINF_SUCCESS;
525	}
526
527
528	/**
529	* Convert HC Physical address to HC Virtual address.
530	*
531	* @returns VBox status.
532	* @param pVM VM handle.
533	* @param HCPhys The host context virtual address.
534	* @param ppv Where to store the resulting address.
535	* @thread The Emulation Thread.
536	*/
537	MMR3DECL(int) MMR3HCPhys2HCVirt(PVM pVM, RTHCPHYS HCPhys, void **ppv)
538	{
539	/*
540	* Try page tables.
541	*/
542	int rc = MMPagePhys2PageTry(pVM, HCPhys, ppv);
543	if (VBOX_SUCCESS(rc))
544	return rc;
545
546	/*
547	* Iterate the locked memory - very slow.
548	*/
549	uint32_t off = HCPhys & PAGE_OFFSET_MASK;
550	HCPhys &= X86_PTE_PAE_PG_MASK;
551	for (PMMLOCKEDMEM pCur = pVM->mm.s.pLockedMem; pCur; pCur = pCur->pNext)
552	{
553	size_t iPage = pCur->cb >> PAGE_SHIFT;
554	while (iPage-- > 0)
555	if ((pCur->aPhysPages[iPage].Phys & X86_PTE_PAE_PG_MASK) == HCPhys)
556	{
557	ppv = (char )pCur->pv + (iPage << PAGE_SHIFT) + off;
558	return VINF_SUCCESS;
559	}
560	}
561	/* give up */
562	return VERR_INVALID_POINTER;
563	}
564
565
566	/**
567	* Read memory from GC virtual address using the current guest CR3.
568	*
569	* @returns VBox status.
570	* @param pVM VM handle.
571	* @param pvDst Destination address (HC of course).
572	* @param GCPtr GC virtual address.
573	* @param cb Number of bytes to read.
574	*/
575	MMR3DECL(int) MMR3ReadGCVirt(PVM pVM, void *pvDst, RTGCPTR GCPtr, size_t cb)
576	{
577	if (GCPtr - pVM->mm.s.pvHyperAreaGC < pVM->mm.s.cbHyperArea)
578	return MMR3HyperReadGCVirt(pVM, pvDst, GCPtr, cb);
579	return PGMPhysReadGCPtr(pVM, pvDst, GCPtr, cb);
580	}
581
582
583	/**
584	* Write to memory at GC virtual address translated using the current guest CR3.
585	*
586	* @returns VBox status.
587	* @param pVM VM handle.
588	* @param GCPtrDst GC virtual address.
589	* @param pvSrc The source address (HC of course).
590	* @param cb Number of bytes to read.
591	*/
592	MMR3DECL(int) MMR3WriteGCVirt(PVM pVM, RTGCPTR GCPtrDst, const void *pvSrc, size_t cb)
593	{
594	if (GCPtrDst - pVM->mm.s.pvHyperAreaGC < pVM->mm.s.cbHyperArea)
595	return VERR_ACCESS_DENIED;
596	return PGMPhysWriteGCPtr(pVM, GCPtrDst, pvSrc, cb);
597	}
598

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

source: vbox/trunk/src/VBox/VMM/MM.cpp@ 2296

Download in other formats: