/* $Id: MM.cpp 247 2007-01-23 17:10:04Z vboxsync $ */ /** @file * MM - Memory Monitor(/Manager). */ /* * Copyright (C) 2006 InnoTek Systemberatung GmbH * * This file is part of VirtualBox Open Source Edition (OSE), as * available from http://www.virtualbox.org. This file is free software; * you can redistribute it and/or modify it under the terms of the GNU * General Public License as published by the Free Software Foundation, * in version 2 as it comes in the "COPYING" file of the VirtualBox OSE * distribution. VirtualBox OSE is distributed in the hope that it will * be useful, but WITHOUT ANY WARRANTY of any kind. * * If you received this file as part of a commercial VirtualBox * distribution, then only the terms of your commercial VirtualBox * license agreement apply instead of the previous paragraph. */ /** @page pg_mm MM - The Memory Monitor/Manager * * It seems like this is going to be the entity taking care of memory allocations * and the locking of physical memory for a VM. MM will track these allocations and * pinnings so pointer conversions, memory read and write, and correct clean up can * be done. * * Memory types: * - Hypervisor Memory Area (HMA). * - Page tables. * - Physical pages. * * The first two types are not accessible using the generic conversion functions * for GC memory, there are special functions for these. * * * A decent structure for this component need to be eveloped as we see usage. One * or two rewrites is probabaly needed to get it right... * * * * @section Hypervisor Memory Area * * The hypervisor is give 4MB of space inside the guest, we assume that we can * steal an page directory entry from the guest OS without cause trouble. In * addition to these 4MB we'll be mapping memory for the graphics emulation, * but that will be an independant mapping. * * The 4MBs are divided into two main parts: * -# The static code and data * -# The shortlived page mappings. * * The first part is used for the VM structure, the core code (VMMSwitch), * GC modules, and the alloc-only-heap. The size will be determined at a * later point but initially we'll say 2MB of locked memory, most of which * is non contiguous physically. * * The second part is used for mapping pages to the hypervisor. We'll be using * a simple round robin when doing these mappings. This means that no-one can * assume that a mapping hangs around for very long, while the managing of the * pages are very simple. * * * * @section Page Pool * * The MM manages a per VM page pool from which other components can allocate * locked, page aligned and page granular memory objects. The pool provides * facilities to convert back and forth between physical and virtual addresses * (within the pool of course). Several specialized interfaces are provided * for the most common alloctions and convertions to save the caller from * bothersome casting and extra parameter passing. * * */ /******************************************************************************* * Header Files * *******************************************************************************/ #define LOG_GROUP LOG_GROUP_MM #include #include #include #include #include "MMInternal.h" #include #include #include #include #include #include #include /******************************************************************************* * Internal Functions * *******************************************************************************/ static int mmR3Term(PVM pVM, bool fKeepTheHeap); static DECLCALLBACK(int) mmR3Save(PVM pVM, PSSMHANDLE pSSM); static DECLCALLBACK(int) mmR3Load(PVM pVM, PSSMHANDLE pSSM, uint32_t u32Version); /** * Initializes the MM. * * MM is managing the virtual address space (among other things) and * setup the hypvervisor memory area mapping in the VM structure and * the hypvervisor alloc-only-heap. Assuming the current init order * and components the hypvervisor memory area looks like this: * -# VM Structure. * -# Hypervisor alloc only heap (also call Hypervisor memory region). * -# Core code. * * MM determins the virtual address of the hypvervisor memory area by * checking for location at previous run. If that property isn't available * it will choose a default starting location, currently 0xe0000000. * * @returns VBox status code. * @param pVM The VM to operate on. */ MMR3DECL(int) MMR3Init(PVM pVM) { LogFlow(("MMR3Init\n")); /* * Assert alignment, sizes and order. */ AssertRelease(!(RT_OFFSETOF(VM, mm.s) & 31)); AssertRelease(sizeof(pVM->mm.s) <= sizeof(pVM->mm.padding)); AssertMsg(pVM->mm.s.offVM == 0, ("Already initialized!\n")); /* * Init the structure. */ pVM->mm.s.offVM = RT_OFFSETOF(VM, mm); pVM->mm.s.offLookupHyper = NIL_OFFSET; /* * Init the heap (may already be initialized already if someone used it). */ if (!pVM->mm.s.pHeap) { int rc = mmr3HeapCreate(pVM, &pVM->mm.s.pHeap); if (!VBOX_SUCCESS(rc)) return rc; } /* * Init the page pool. */ int rc = mmr3PagePoolInit(pVM); if (VBOX_SUCCESS(rc)) { /* * Init the hypervisor related stuff. */ rc = mmr3HyperInit(pVM); if (VBOX_SUCCESS(rc)) { /* * Register the saved state data unit. */ rc = SSMR3RegisterInternal(pVM, "mm", 1, 1, sizeof(uint32_t) * 2, NULL, mmR3Save, NULL, NULL, mmR3Load, NULL); if (VBOX_SUCCESS(rc)) return rc; } /* .... failure .... */ mmR3Term(pVM, true /* keep the heap */); } else mmr3HeapDestroy(pVM->mm.s.pHeap); return rc; } /** * Initializes the MM parts which depends on PGM being initialized. * * @returns VBox status code. * @param pVM The VM to operate on. * @remark No cleanup necessary since MMR3Term() will be called on failure. */ MMR3DECL(int) MMR3InitPaging(PVM pVM) { LogFlow(("MMR3InitPaging:\n")); bool fPreAlloc; int rc = CFGMR3QueryBool(CFGMR3GetRoot(pVM), "RamPreAlloc", &fPreAlloc); if (rc == VERR_CFGM_VALUE_NOT_FOUND) fPreAlloc = false; else AssertMsgRCReturn(rc, ("Configuration error: Failed to query integer \"RamPreAlloc\", rc=%Vrc.\n", rc), rc); uint64_t cbRam; rc = CFGMR3QueryU64(CFGMR3GetRoot(pVM), "RamSize", &cbRam); if (rc == VERR_CFGM_VALUE_NOT_FOUND) cbRam = 0; if (VBOX_SUCCESS(rc) || rc == VERR_CFGM_VALUE_NOT_FOUND) { if (cbRam < PAGE_SIZE) { Log(("MM: No RAM configured\n")); return VINF_SUCCESS; } #ifdef PGM_DYNAMIC_RAM_ALLOC Log(("MM: %llu bytes of RAM%s\n", cbRam, fPreAlloc ? " (PreAlloc)" : "")); pVM->mm.s.pvRamBaseHC = 0; /** @todo obsolete */ pVM->mm.s.cbRamBase = cbRam & PAGE_BASE_GC_MASK; rc = MMR3PhysRegister(pVM, pVM->mm.s.pvRamBaseHC, 0, pVM->mm.s.cbRamBase, MM_RAM_FLAGS_DYNAMIC_ALLOC, "Main Memory"); if (VBOX_SUCCESS(rc)) { /* Allocate the first chunk, as we'll map ROM ranges there. */ rc = PGM3PhysGrowRange(pVM, (RTGCPHYS)0); if (VBOX_SUCCESS(rc)) { /* Should we preallocate the entire guest RAM? */ if (fPreAlloc) { for (RTGCPHYS GCPhys = PGM_DYNAMIC_CHUNK_SIZE; GCPhys < cbRam; GCPhys += PGM_DYNAMIC_CHUNK_SIZE) { rc = PGM3PhysGrowRange(pVM, GCPhys); if (VBOX_FAILURE(rc)) return rc; } } return rc; } } #else unsigned cPages = cbRam >> PAGE_SHIFT; Log(("MM: %llu bytes of RAM (%d pages)\n", cbRam, cPages)); rc = SUPPageAlloc(cPages, &pVM->mm.s.pvRamBaseHC); if (VBOX_SUCCESS(rc)) { pVM->mm.s.cbRamBase = cPages << PAGE_SHIFT; rc = MMR3PhysRegister(pVM, pVM->mm.s.pvRamBaseHC, 0, pVM->mm.s.cbRamBase, 0, "Main Memory"); if (VBOX_SUCCESS(rc)) return rc; SUPPageFree(pVM->mm.s.pvRamBaseHC); } else LogRel(("MMR3InitPage: Failed to allocate %u bytes of RAM! rc=%Vrc\n", cPages << PAGE_SHIFT)); #endif } else AssertMsgFailed(("Configuration error: Failed to query integer \"RamSize\", rc=%Vrc.\n", rc)); LogFlow(("MMR3InitPaging: returns %Vrc\n", rc)); return rc; } /** * Terminates the MM. * * Termination means cleaning up and freeing all resources, * the VM it self is at this point powered off or suspended. * * @returns VBox status code. * @param pVM The VM to operate on. */ MMR3DECL(int) MMR3Term(PVM pVM) { return mmR3Term(pVM, false /* free the heap */); } /** * Worker for MMR3Term and MMR3Init. * * The tricky bit here is that we must not destroy the heap if we're * called from MMR3Init, otherwise we'll get into trouble when * CFGMR3Term is called later in the bailout process. * * @returns VBox status code. * @param pVM The VM to operate on. * @param fKeepTheHeap Whether or not to keep the heap. */ static int mmR3Term(PVM pVM, bool fKeepTheHeap) { /* * Release locked memory. * (Associated record are released by the heap.) */ PMMLOCKEDMEM pLockedMem = pVM->mm.s.pLockedMem; while (pLockedMem) { int rc = SUPPageUnlock(pLockedMem->pv); AssertMsgRC(rc, ("SUPPageUnlock(%p) -> rc=%d\n", pLockedMem->pv, rc)); switch (pLockedMem->eType) { case MM_LOCKED_TYPE_PHYS: case MM_LOCKED_TYPE_HYPER_NOFREE: break; case MM_LOCKED_TYPE_HYPER: rc = SUPPageFree(pLockedMem->pv); AssertMsgRC(rc, ("SUPPageFree(%p) -> rc=%d\n", pLockedMem->pv, rc)); break; } /* next */ pLockedMem = pLockedMem->pNext; } /* * Destroy the page pool. */ mmr3PagePoolTerm(pVM); /* * Destroy the heap if requested. */ if (!fKeepTheHeap) { mmr3HeapDestroy(pVM->mm.s.pHeap); pVM->mm.s.pHeap = NULL; } /* * Zero stuff to detect after termination use of the MM interface */ pVM->mm.s.offLookupHyper = NIL_OFFSET; pVM->mm.s.pLockedMem = NULL; pVM->mm.s.pHyperHeapHC = NULL; /* freed above. */ pVM->mm.s.pHyperHeapGC = 0; /* freed above. */ pVM->mm.s.offVM = 0; /* init assertion on this */ return 0; } /** * Execute state save operation. * * @returns VBox status code. * @param pVM VM Handle. * @param pSSM SSM operation handle. */ static DECLCALLBACK(int) mmR3Save(PVM pVM, PSSMHANDLE pSSM) { LogFlow(("mmR3Save:\n")); /* (PGM saves the physical memory.) */ SSMR3PutUInt(pSSM, pVM->mm.s.cbRAMSize); return SSMR3PutUInt(pSSM, pVM->mm.s.cbRamBase); } /** * Execute state load operation. * * @returns VBox status code. * @param pVM VM Handle. * @param pSSM SSM operation handle. * @param u32Version Data layout version. */ static DECLCALLBACK(int) mmR3Load(PVM pVM, PSSMHANDLE pSSM, uint32_t u32Version) { LogFlow(("mmR3Load:\n")); /* * Validate version. */ if (u32Version != 1) { Log(("mmR3Load: Invalid version u32Version=%d!\n", u32Version)); return VERR_SSM_UNSUPPORTED_DATA_UNIT_VERSION; } /* * Check the cbRAMSize and cbRamBase values. */ RTUINT cb; int rc = SSMR3GetUInt(pSSM, &cb); if (VBOX_FAILURE(rc)) return rc; if (cb != pVM->mm.s.cbRAMSize) { Log(("mmR3Load: Memory configuration has changed. cbRAMSize=%#x save %#x\n", pVM->mm.s.cbRAMSize, cb)); return VERR_SSM_LOAD_MEMORY_SIZE_MISMATCH; } rc = SSMR3GetUInt(pSSM, &cb); if (VBOX_FAILURE(rc)) return rc; if (cb != pVM->mm.s.cbRamBase) { Log(("mmR3Load: Memory configuration has changed. cbRamBase=%#x save %#x\n", pVM->mm.s.cbRamBase, cb)); return VERR_SSM_LOAD_MEMORY_SIZE_MISMATCH; } /* PGM restores the physical memory. */ return rc; } /** * Locks physical memory which backs a virtual memory range (HC) adding * the required records to the pLockedMem list. * * @returns VBox status code. * @param pVM The VM handle. * @param pv Pointer to memory range which shall be locked down. * This pointer is page aligned. * @param cb Size of memory range (in bytes). This size is page aligned. * @param eType Memory type. * @param ppLockedMem Where to store the pointer to the created locked memory record. * This is optional, pass NULL if not used. * @param fSilentFailure Don't raise an error when unsuccessful. Upper layer with deal with it. */ int mmr3LockMem(PVM pVM, void *pv, size_t cb, MMLOCKEDTYPE eType, PMMLOCKEDMEM *ppLockedMem, bool fSilentFailure) { Assert(RT_ALIGN_P(pv, PAGE_SIZE) == pv); Assert(RT_ALIGN_Z(cb, PAGE_SIZE) == cb); if (ppLockedMem) *ppLockedMem = NULL; /* * Allocate locked mem structure. */ unsigned cPages = cb >> PAGE_SHIFT; AssertReturn(cPages == (cb >> PAGE_SHIFT), VERR_OUT_OF_RANGE); PMMLOCKEDMEM pLockedMem = (PMMLOCKEDMEM)MMR3HeapAlloc(pVM, MM_TAG_MM, RT_OFFSETOF(MMLOCKEDMEM, aPhysPages[cPages])); if (!pLockedMem) return VERR_NO_MEMORY; pLockedMem->pv = pv; pLockedMem->cb = cb; pLockedMem->eType = eType; memset(&pLockedMem->u, 0, sizeof(pLockedMem->u)); /* * Lock the memory. */ int rc = SUPPageLock(pv, cb, &pLockedMem->aPhysPages[0]); if (VBOX_SUCCESS(rc)) { /* * Setup the reserved field. */ PSUPPAGE pPhysPage = &pLockedMem->aPhysPages[0]; for (unsigned c = cPages; c > 0; c--, pPhysPage++) pPhysPage->uReserved = (RTHCUINTPTR)pLockedMem; /* * Insert into the list. * * ASSUME no protected needed here as only one thread in the system can possibly * be doing this. No other threads will walk this list either we assume. */ pLockedMem->pNext = pVM->mm.s.pLockedMem; pVM->mm.s.pLockedMem = pLockedMem; /* Set return value. */ if (ppLockedMem) *ppLockedMem = pLockedMem; } else { AssertMsgFailed(("SUPPageLock failed with rc=%d\n", rc)); MMR3HeapFree(pLockedMem); if (!fSilentFailure) rc = VMSetError(pVM, rc, RT_SRC_POS, N_("Failed to lock %d bytes of host memory (out of memory)"), cb); } return rc; } /** * Maps a part of or an entire locked memory region into the guest context. * * @returns VBox status. * God knows what happens if we fail... * @param pVM VM handle. * @param pLockedMem Locked memory structure. * @param Addr GC Address where to start the mapping. * @param iPage Page number in the locked memory region. * @param cPages Number of pages to map. * @param fFlags See the fFlags argument of PGR3Map(). */ int mmr3MapLocked(PVM pVM, PMMLOCKEDMEM pLockedMem, RTGCPTR Addr, unsigned iPage, size_t cPages, unsigned fFlags) { /* * Adjust ~0 argument */ if (cPages == ~(size_t)0) cPages = (pLockedMem->cb >> PAGE_SHIFT) - iPage; Assert(cPages != ~0U); /* no incorrect arguments are accepted */ Assert(RT_ALIGN_GCPT(Addr, PAGE_SIZE, RTGCPTR) == Addr); AssertMsg(iPage < (pLockedMem->cb >> PAGE_SHIFT), ("never even think about giving me a bad iPage(=%d)\n", iPage)); AssertMsg(iPage + cPages <= (pLockedMem->cb >> PAGE_SHIFT), ("never even think about giving me a bad cPages(=%d)\n", cPages)); /* * Map the the pages. */ PSUPPAGE pPhysPage = &pLockedMem->aPhysPages[iPage]; while (cPages) { RTHCPHYS HCPhys = pPhysPage->Phys; int rc = PGMMap(pVM, Addr, HCPhys, PAGE_SIZE, fFlags); if (VBOX_FAILURE(rc)) { /** @todo how the hell can we do a proper bailout here. */ return rc; } /* next */ cPages--; iPage++; pPhysPage++; Addr += PAGE_SIZE; } return VINF_SUCCESS; } /** * Convert HC Physical address to HC Virtual address. * * @returns VBox status. * @param pVM VM handle. * @param HCPhys The host context virtual address. * @param ppv Where to store the resulting address. * @thread The Emulation Thread. */ MMR3DECL(int) MMR3HCPhys2HCVirt(PVM pVM, RTHCPHYS HCPhys, void **ppv) { /* * Try page tables. */ int rc = MMPagePhys2PageTry(pVM, HCPhys, ppv); if (VBOX_SUCCESS(rc)) return rc; /* * The VM structure? */ uint32_t off = (uint32_t)(HCPhys - pVM->HCPhysVM); if (off < RT_ALIGN_32(sizeof(*pVM), PAGE_SIZE)) { *ppv = (char *)pVM + off; return VINF_SUCCESS; } /* * Iterate the locked memory - very slow. */ off = HCPhys & PAGE_OFFSET_MASK; HCPhys &= X86_PTE_PAE_PG_MASK; for (PMMLOCKEDMEM pCur = pVM->mm.s.pLockedMem; pCur; pCur = pCur->pNext) { size_t iPage = pCur->cb >> PAGE_SHIFT; while (iPage-- > 0) if ((pCur->aPhysPages[iPage].Phys & X86_PTE_PAE_PG_MASK) == HCPhys) { *ppv = (char *)pCur->pv + (iPage << PAGE_SHIFT) + off; return VINF_SUCCESS; } } /* give up */ return VERR_INVALID_POINTER; } /** * Read memory from GC virtual address using the current guest CR3. * * @returns VBox status. * @param pVM VM handle. * @param pvDst Destination address (HC of course). * @param GCPtr GC virtual address. * @param cb Number of bytes to read. */ MMR3DECL(int) MMR3ReadGCVirt(PVM pVM, void *pvDst, RTGCPTR GCPtr, size_t cb) { if (GCPtr - pVM->mm.s.pvHyperAreaGC < pVM->mm.s.cbHyperArea) return MMR3HyperReadGCVirt(pVM, pvDst, GCPtr, cb); return PGMPhysReadGCPtr(pVM, pvDst, GCPtr, cb); } /** * Write to memory at GC virtual address translated using the current guest CR3. * * @returns VBox status. * @param pVM VM handle. * @param GCPtrDst GC virtual address. * @param pvSrc The source address (HC of course). * @param cb Number of bytes to read. */ MMR3DECL(int) MMR3WriteGCVirt(PVM pVM, RTGCPTR GCPtrDst, const void *pvSrc, size_t cb) { if (GCPtrDst - pVM->mm.s.pvHyperAreaGC < pVM->mm.s.cbHyperArea) return VERR_ACCESS_DENIED; return PGMPhysWriteGCPtr(pVM, GCPtrDst, pvSrc, cb); }