X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/8f6c56a50524aa785f7e596d52dddfb331e18961..743345f9a4b36f7e2f9ba37691e70c50baecb56e:/osfmk/vm/vm_kern.c diff --git a/osfmk/vm/vm_kern.c b/osfmk/vm/vm_kern.c index d5ed1f659..8d37f4cf0 100644 --- a/osfmk/vm/vm_kern.c +++ b/osfmk/vm/vm_kern.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2005 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2007 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -66,7 +66,6 @@ #include #include #include -#include #include #include #include @@ -77,6 +76,10 @@ #include #include + +#include +#include + /* * Variables exported by this module. */ @@ -84,20 +87,15 @@ vm_map_t kernel_map; vm_map_t kernel_pageable_map; +extern boolean_t vm_kernel_ready; + /* * Forward declarations for internal functions. */ extern kern_return_t kmem_alloc_pages( - register vm_object_t object, - register vm_object_offset_t offset, - register vm_object_size_t size); - -extern void kmem_remap_pages( - register vm_object_t object, - register vm_object_offset_t offset, - register vm_offset_t start, - register vm_offset_t end, - vm_prot_t protection); + vm_object_t object, + vm_object_offset_t offset, + vm_object_size_t size); kern_return_t kmem_alloc_contig( @@ -105,7 +103,10 @@ kmem_alloc_contig( vm_offset_t *addrp, vm_size_t size, vm_offset_t mask, - int flags) + ppnum_t max_pnum, + ppnum_t pnum_mask, + int flags, + vm_tag_t tag) { vm_object_t object; vm_object_offset_t offset; @@ -116,17 +117,19 @@ kmem_alloc_contig( vm_page_t m, pages; kern_return_t kr; - if (map == VM_MAP_NULL || (flags && (flags ^ KMA_KOBJECT))) + if (map == VM_MAP_NULL || (flags & ~(KMA_KOBJECT | KMA_LOMEM | KMA_NOPAGEWAIT))) return KERN_INVALID_ARGUMENT; + + map_size = vm_map_round_page(size, + VM_MAP_PAGE_MASK(map)); + map_mask = (vm_map_offset_t)mask; - if (size == 0) { + /* Check for zero allocation size (either directly or via overflow) */ + if (map_size == 0) { *addrp = 0; return KERN_INVALID_ARGUMENT; } - map_size = vm_map_round_page(size); - map_mask = (vm_map_offset_t)mask; - /* * Allocate a new object (if necessary) and the reference we * will be donating to the map entry. We must do this before @@ -139,25 +142,34 @@ kmem_alloc_contig( object = vm_object_allocate(map_size); } - kr = vm_map_find_space(map, &map_addr, map_size, map_mask, &entry); + kr = vm_map_find_space(map, &map_addr, map_size, map_mask, 0, &entry); if (KERN_SUCCESS != kr) { vm_object_deallocate(object); return kr; } - entry->object.vm_object = object; - entry->offset = offset = (object == kernel_object) ? - map_addr - VM_MIN_KERNEL_ADDRESS : 0; + if (object == kernel_object) { + offset = map_addr; + } else { + offset = 0; + } + VME_OBJECT_SET(entry, object); + VME_OFFSET_SET(entry, offset); + VME_ALIAS_SET(entry, tag); /* Take an extra object ref in case the map entry gets deleted */ vm_object_reference(object); vm_map_unlock(map); - kr = cpm_allocate(CAST_DOWN(vm_size_t, map_size), &pages, FALSE); + kr = cpm_allocate(CAST_DOWN(vm_size_t, map_size), &pages, max_pnum, pnum_mask, FALSE, flags); if (kr != KERN_SUCCESS) { - vm_map_remove(map, vm_map_trunc_page(map_addr), - vm_map_round_page(map_addr + map_size), 0); + vm_map_remove(map, + vm_map_trunc_page(map_addr, + VM_MAP_PAGE_MASK(map)), + vm_map_round_page(map_addr + map_size, + VM_MAP_PAGE_MASK(map)), + 0); vm_object_deallocate(object); *addrp = 0; return kr; @@ -167,21 +179,32 @@ kmem_alloc_contig( for (i = 0; i < map_size; i += PAGE_SIZE) { m = pages; pages = NEXT_PAGE(m); + *(NEXT_PAGE_PTR(m)) = VM_PAGE_NULL; m->busy = FALSE; vm_page_insert(m, object, offset + i); } vm_object_unlock(object); - if ((kr = vm_map_wire(map, vm_map_trunc_page(map_addr), - vm_map_round_page(map_addr + map_size), VM_PROT_DEFAULT, FALSE)) - != KERN_SUCCESS) { + kr = vm_map_wire(map, + vm_map_trunc_page(map_addr, + VM_MAP_PAGE_MASK(map)), + vm_map_round_page(map_addr + map_size, + VM_MAP_PAGE_MASK(map)), + VM_PROT_DEFAULT | VM_PROT_MEMORY_TAG_MAKE(tag), + FALSE); + + if (kr != KERN_SUCCESS) { if (object == kernel_object) { vm_object_lock(object); vm_object_page_remove(object, offset, offset + map_size); vm_object_unlock(object); } - vm_map_remove(map, vm_map_trunc_page(map_addr), - vm_map_round_page(map_addr + map_size), 0); + vm_map_remove(map, + vm_map_trunc_page(map_addr, + VM_MAP_PAGE_MASK(map)), + vm_map_round_page(map_addr + map_size, + VM_MAP_PAGE_MASK(map)), + 0); vm_object_deallocate(object); return kr; } @@ -190,7 +213,8 @@ kmem_alloc_contig( if (object == kernel_object) vm_map_simplify(map, map_addr); - *addrp = map_addr; + *addrp = (vm_offset_t) map_addr; + assert((vm_map_offset_t) *addrp == map_addr); return KERN_SUCCESS; } @@ -205,32 +229,156 @@ kmem_alloc_contig( * KMA_HERE *addrp is base address, else "anywhere" * KMA_NOPAGEWAIT don't wait for pages if unavailable * KMA_KOBJECT use kernel_object + * KMA_LOMEM support for 32 bit devices in a 64 bit world + * if set and a lomemory pool is available + * grab pages from it... this also implies + * KMA_NOPAGEWAIT */ kern_return_t kernel_memory_allocate( - register vm_map_t map, - register vm_offset_t *addrp, - register vm_size_t size, - register vm_offset_t mask, - int flags) + vm_map_t map, + vm_offset_t *addrp, + vm_size_t size, + vm_offset_t mask, + int flags, + vm_tag_t tag) { vm_object_t object; vm_object_offset_t offset; - vm_map_entry_t entry; - vm_map_offset_t map_addr; + vm_object_offset_t pg_offset; + vm_map_entry_t entry = NULL; + vm_map_offset_t map_addr, fill_start; vm_map_offset_t map_mask; - vm_map_size_t map_size; - vm_map_size_t i; - kern_return_t kr; + vm_map_size_t map_size, fill_size; + kern_return_t kr, pe_result; + vm_page_t mem; + vm_page_t guard_page_list = NULL; + vm_page_t wired_page_list = NULL; + int guard_page_count = 0; + int wired_page_count = 0; + int i; + int vm_alloc_flags; + vm_prot_t kma_prot; + + if (! vm_kernel_ready) { + panic("kernel_memory_allocate: VM is not ready"); + } + + map_size = vm_map_round_page(size, + VM_MAP_PAGE_MASK(map)); + map_mask = (vm_map_offset_t) mask; + + vm_alloc_flags = VM_MAKE_TAG(tag); - if (size == 0) { + /* Check for zero allocation size (either directly or via overflow) */ + if (map_size == 0) { *addrp = 0; return KERN_INVALID_ARGUMENT; } - map_size = vm_map_round_page(size); - map_mask = (vm_map_offset_t) mask; + /* + * limit the size of a single extent of wired memory + * to try and limit the damage to the system if + * too many pages get wired down + * limit raised to 2GB with 128GB max physical limit, + * but scaled by installed memory above this + */ + if ( !(flags & KMA_VAONLY) && map_size > MAX(1ULL<<31, sane_size/64)) { + return KERN_RESOURCE_SHORTAGE; + } + + /* + * Guard pages: + * + * Guard pages are implemented as ficticious pages. By placing guard pages + * on either end of a stack, they can help detect cases where a thread walks + * off either end of its stack. They are allocated and set up here and attempts + * to access those pages are trapped in vm_fault_page(). + * + * The map_size we were passed may include extra space for + * guard pages. If those were requested, then back it out of fill_size + * since vm_map_find_space() takes just the actual size not including + * guard pages. Similarly, fill_start indicates where the actual pages + * will begin in the range. + */ + + fill_start = 0; + fill_size = map_size; + + if (flags & KMA_GUARD_FIRST) { + vm_alloc_flags |= VM_FLAGS_GUARD_BEFORE; + fill_start += PAGE_SIZE_64; + fill_size -= PAGE_SIZE_64; + if (map_size < fill_start + fill_size) { + /* no space for a guard page */ + *addrp = 0; + return KERN_INVALID_ARGUMENT; + } + guard_page_count++; + } + if (flags & KMA_GUARD_LAST) { + vm_alloc_flags |= VM_FLAGS_GUARD_AFTER; + fill_size -= PAGE_SIZE_64; + if (map_size <= fill_start + fill_size) { + /* no space for a guard page */ + *addrp = 0; + return KERN_INVALID_ARGUMENT; + } + guard_page_count++; + } + wired_page_count = (int) (fill_size / PAGE_SIZE_64); + assert(wired_page_count * PAGE_SIZE_64 == fill_size); + + for (i = 0; i < guard_page_count; i++) { + for (;;) { + mem = vm_page_grab_guard(); + + if (mem != VM_PAGE_NULL) + break; + if (flags & KMA_NOPAGEWAIT) { + kr = KERN_RESOURCE_SHORTAGE; + goto out; + } + vm_page_more_fictitious(); + } + mem->snext = guard_page_list; + guard_page_list = mem; + } + + if (! (flags & KMA_VAONLY)) { + for (i = 0; i < wired_page_count; i++) { + uint64_t unavailable; + + for (;;) { + if (flags & KMA_LOMEM) + mem = vm_page_grablo(); + else + mem = vm_page_grab(); + + if (mem != VM_PAGE_NULL) + break; + + if (flags & KMA_NOPAGEWAIT) { + kr = KERN_RESOURCE_SHORTAGE; + goto out; + } + if ((flags & KMA_LOMEM) && (vm_lopage_needed == TRUE)) { + kr = KERN_RESOURCE_SHORTAGE; + goto out; + } + unavailable = (vm_page_wire_count + vm_page_free_target) * PAGE_SIZE; + + if (unavailable > max_mem || map_size > (max_mem - unavailable)) { + kr = KERN_RESOURCE_SHORTAGE; + goto out; + } + VM_PAGE_WAIT(); + } + mem->snext = wired_page_list; + wired_page_list = mem; + } + } /* * Allocate a new object (if necessary). We must do this before @@ -239,66 +387,400 @@ kernel_memory_allocate( if ((flags & KMA_KOBJECT) != 0) { object = kernel_object; vm_object_reference(object); + } else if ((flags & KMA_COMPRESSOR) != 0) { + object = compressor_object; + vm_object_reference(object); } else { object = vm_object_allocate(map_size); } - kr = vm_map_find_space(map, &map_addr, map_size, map_mask, &entry); + if (flags & KMA_ATOMIC) + vm_alloc_flags |= VM_FLAGS_ATOMIC_ENTRY; + + kr = vm_map_find_space(map, &map_addr, + fill_size, map_mask, + vm_alloc_flags, &entry); if (KERN_SUCCESS != kr) { vm_object_deallocate(object); - return kr; + goto out; } - entry->object.vm_object = object; - entry->offset = offset = (object == kernel_object) ? - map_addr - VM_MIN_KERNEL_ADDRESS : 0; + if (object == kernel_object || object == compressor_object) { + offset = map_addr; + } else { + offset = 0; + } + VME_OBJECT_SET(entry, object); + VME_OFFSET_SET(entry, offset); + + if (object != compressor_object) + entry->wired_count++; - vm_object_reference(object); - vm_map_unlock(map); + if (flags & KMA_PERMANENT) + entry->permanent = TRUE; + + if (object != kernel_object && object != compressor_object) + vm_object_reference(object); vm_object_lock(object); - for (i = 0; i < map_size; i += PAGE_SIZE) { - vm_page_t mem; + vm_map_unlock(map); - while (VM_PAGE_NULL == - (mem = vm_page_alloc(object, offset + i))) { - if (flags & KMA_NOPAGEWAIT) { - if (object == kernel_object) - vm_object_page_remove(object, offset, offset + i); - vm_object_unlock(object); - vm_map_remove(map, map_addr, map_addr + map_size, 0); - vm_object_deallocate(object); - return KERN_RESOURCE_SHORTAGE; - } + pg_offset = 0; + + if (fill_start) { + if (guard_page_list == NULL) + panic("kernel_memory_allocate: guard_page_list == NULL"); + + mem = guard_page_list; + guard_page_list = mem->snext; + mem->snext = NULL; + + vm_page_insert(mem, object, offset + pg_offset); + + mem->busy = FALSE; + pg_offset += PAGE_SIZE_64; + } + + kma_prot = VM_PROT_READ | VM_PROT_WRITE; + + if (flags & KMA_VAONLY) { + pg_offset = fill_start + fill_size; + } else { + for (pg_offset = fill_start; pg_offset < fill_start + fill_size; pg_offset += PAGE_SIZE_64) { + if (wired_page_list == NULL) + panic("kernel_memory_allocate: wired_page_list == NULL"); + + mem = wired_page_list; + wired_page_list = mem->snext; + mem->snext = NULL; + + assert(mem->wire_count == 0); + assert(mem->vm_page_q_state == VM_PAGE_NOT_ON_Q); + + mem->vm_page_q_state = VM_PAGE_IS_WIRED; + mem->wire_count++; + if (__improbable(mem->wire_count == 0)) { + panic("kernel_memory_allocate(%p): wire_count overflow", + mem); + } + + vm_page_insert_wired(mem, object, offset + pg_offset, tag); + + mem->busy = FALSE; + mem->pmapped = TRUE; + mem->wpmapped = TRUE; + + PMAP_ENTER_OPTIONS(kernel_pmap, map_addr + pg_offset, mem, + kma_prot, VM_PROT_NONE, ((flags & KMA_KSTACK) ? VM_MEM_STACK : 0), TRUE, + PMAP_OPTIONS_NOWAIT, pe_result); + + if (pe_result == KERN_RESOURCE_SHORTAGE) { vm_object_unlock(object); - VM_PAGE_WAIT(); + + PMAP_ENTER(kernel_pmap, map_addr + pg_offset, mem, + kma_prot, VM_PROT_NONE, ((flags & KMA_KSTACK) ? VM_MEM_STACK : 0), TRUE); + vm_object_lock(object); } + if (flags & KMA_NOENCRYPT) { + bzero(CAST_DOWN(void *, (map_addr + pg_offset)), PAGE_SIZE); + + pmap_set_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)); + } + } + } + if ((fill_start + fill_size) < map_size) { + if (guard_page_list == NULL) + panic("kernel_memory_allocate: guard_page_list == NULL"); + + mem = guard_page_list; + guard_page_list = mem->snext; + mem->snext = NULL; + + vm_page_insert(mem, object, offset + pg_offset); + mem->busy = FALSE; } - vm_object_unlock(object); + if (guard_page_list || wired_page_list) + panic("kernel_memory_allocate: non empty list\n"); - if ((kr = vm_map_wire(map, map_addr, map_addr + map_size, VM_PROT_DEFAULT, FALSE)) - != KERN_SUCCESS) { - if (object == kernel_object) { - vm_object_lock(object); - vm_object_page_remove(object, offset, offset + map_size); - vm_object_unlock(object); - } - vm_map_remove(map, map_addr, map_addr + map_size, 0); - vm_object_deallocate(object); - return (kr); + if (! (flags & KMA_VAONLY)) { + vm_page_lockspin_queues(); + vm_page_wire_count += wired_page_count; + vm_page_unlock_queues(); } - /* now that the page is wired, we no longer have to fear coalesce */ - vm_object_deallocate(object); - if (object == kernel_object) + + vm_object_unlock(object); + + /* + * now that the pages are wired, we no longer have to fear coalesce + */ + if (object == kernel_object || object == compressor_object) vm_map_simplify(map, map_addr); + else + vm_object_deallocate(object); /* * Return the memory, not zeroed. */ *addrp = CAST_DOWN(vm_offset_t, map_addr); return KERN_SUCCESS; + +out: + if (guard_page_list) + vm_page_free_list(guard_page_list, FALSE); + + if (wired_page_list) + vm_page_free_list(wired_page_list, FALSE); + + return kr; +} + +kern_return_t +kernel_memory_populate( + vm_map_t map, + vm_offset_t addr, + vm_size_t size, + int flags, + vm_tag_t tag) +{ + vm_object_t object; + vm_object_offset_t offset, pg_offset; + kern_return_t kr, pe_result; + vm_page_t mem; + vm_page_t page_list = NULL; + int page_count = 0; + int i; + + page_count = (int) (size / PAGE_SIZE_64); + + assert((flags & (KMA_COMPRESSOR|KMA_KOBJECT)) != (KMA_COMPRESSOR|KMA_KOBJECT)); + + if (flags & KMA_COMPRESSOR) { + + pg_offset = page_count * PAGE_SIZE_64; + + do { + for (;;) { + mem = vm_page_grab(); + + if (mem != VM_PAGE_NULL) + break; + + VM_PAGE_WAIT(); + } + mem->snext = page_list; + page_list = mem; + + pg_offset -= PAGE_SIZE_64; + + kr = pmap_enter_options(kernel_pmap, + addr + pg_offset, VM_PAGE_GET_PHYS_PAGE(mem), + VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE, 0, TRUE, + PMAP_OPTIONS_INTERNAL, NULL); + assert(kr == KERN_SUCCESS); + + } while (pg_offset); + + offset = addr; + object = compressor_object; + + vm_object_lock(object); + + for (pg_offset = 0; + pg_offset < size; + pg_offset += PAGE_SIZE_64) { + + mem = page_list; + page_list = mem->snext; + mem->snext = NULL; + + vm_page_insert(mem, object, offset + pg_offset); + assert(mem->busy); + + mem->busy = FALSE; + mem->pmapped = TRUE; + mem->wpmapped = TRUE; + mem->vm_page_q_state = VM_PAGE_USED_BY_COMPRESSOR; + } + vm_object_unlock(object); + + return KERN_SUCCESS; + } + + for (i = 0; i < page_count; i++) { + for (;;) { + if (flags & KMA_LOMEM) + mem = vm_page_grablo(); + else + mem = vm_page_grab(); + + if (mem != VM_PAGE_NULL) + break; + + if (flags & KMA_NOPAGEWAIT) { + kr = KERN_RESOURCE_SHORTAGE; + goto out; + } + if ((flags & KMA_LOMEM) && + (vm_lopage_needed == TRUE)) { + kr = KERN_RESOURCE_SHORTAGE; + goto out; + } + VM_PAGE_WAIT(); + } + mem->snext = page_list; + page_list = mem; + } + if (flags & KMA_KOBJECT) { + offset = addr; + object = kernel_object; + + vm_object_lock(object); + } else { + /* + * If it's not the kernel object, we need to: + * lock map; + * lookup entry; + * lock object; + * take reference on object; + * unlock map; + */ + panic("kernel_memory_populate(%p,0x%llx,0x%llx,0x%x): " + "!KMA_KOBJECT", + map, (uint64_t) addr, (uint64_t) size, flags); + } + + for (pg_offset = 0; + pg_offset < size; + pg_offset += PAGE_SIZE_64) { + + if (page_list == NULL) + panic("kernel_memory_populate: page_list == NULL"); + + mem = page_list; + page_list = mem->snext; + mem->snext = NULL; + + assert(mem->vm_page_q_state == VM_PAGE_NOT_ON_Q); + mem->vm_page_q_state = VM_PAGE_IS_WIRED; + mem->wire_count++; + if (__improbable(mem->wire_count == 0)) { + panic("kernel_memory_populate(%p): wire_count overflow", + mem); + } + + vm_page_insert_wired(mem, object, offset + pg_offset, tag); + + mem->busy = FALSE; + mem->pmapped = TRUE; + mem->wpmapped = TRUE; + + PMAP_ENTER_OPTIONS(kernel_pmap, addr + pg_offset, mem, + VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE, + ((flags & KMA_KSTACK) ? VM_MEM_STACK : 0), TRUE, + PMAP_OPTIONS_NOWAIT, pe_result); + + if (pe_result == KERN_RESOURCE_SHORTAGE) { + + vm_object_unlock(object); + + PMAP_ENTER(kernel_pmap, addr + pg_offset, mem, + VM_PROT_READ | VM_PROT_WRITE, VM_PROT_NONE, + ((flags & KMA_KSTACK) ? VM_MEM_STACK : 0), TRUE); + + vm_object_lock(object); + } + if (flags & KMA_NOENCRYPT) { + bzero(CAST_DOWN(void *, (addr + pg_offset)), PAGE_SIZE); + pmap_set_noencrypt(VM_PAGE_GET_PHYS_PAGE(mem)); + } + } + vm_page_lock_queues(); + vm_page_wire_count += page_count; + vm_page_unlock_queues(); + + vm_object_unlock(object); + + return KERN_SUCCESS; + +out: + if (page_list) + vm_page_free_list(page_list, FALSE); + + return kr; +} + + +void +kernel_memory_depopulate( + vm_map_t map, + vm_offset_t addr, + vm_size_t size, + int flags) +{ + vm_object_t object; + vm_object_offset_t offset, pg_offset; + vm_page_t mem; + vm_page_t local_freeq = NULL; + + assert((flags & (KMA_COMPRESSOR|KMA_KOBJECT)) != (KMA_COMPRESSOR|KMA_KOBJECT)); + + if (flags & KMA_COMPRESSOR) { + offset = addr; + object = compressor_object; + + vm_object_lock(object); + } else if (flags & KMA_KOBJECT) { + offset = addr; + object = kernel_object; + + vm_object_lock(object); + } else { + offset = 0; + object = NULL; + /* + * If it's not the kernel object, we need to: + * lock map; + * lookup entry; + * lock object; + * unlock map; + */ + panic("kernel_memory_depopulate(%p,0x%llx,0x%llx,0x%x): " + "!KMA_KOBJECT", + map, (uint64_t) addr, (uint64_t) size, flags); + } + pmap_protect(kernel_map->pmap, offset, offset + size, VM_PROT_NONE); + + for (pg_offset = 0; + pg_offset < size; + pg_offset += PAGE_SIZE_64) { + + mem = vm_page_lookup(object, offset + pg_offset); + + assert(mem); + + if (mem->vm_page_q_state != VM_PAGE_USED_BY_COMPRESSOR) + pmap_disconnect(VM_PAGE_GET_PHYS_PAGE(mem)); + + mem->busy = TRUE; + + assert(mem->tabled); + vm_page_remove(mem, TRUE); + assert(mem->busy); + + assert(mem->pageq.next == 0 && mem->pageq.prev == 0); + assert((mem->vm_page_q_state == VM_PAGE_USED_BY_COMPRESSOR) || + (mem->vm_page_q_state == VM_PAGE_NOT_ON_Q)); + + mem->vm_page_q_state = VM_PAGE_NOT_ON_Q; + mem->snext = local_freeq; + local_freeq = mem; + } + vm_object_unlock(object); + + if (local_freeq) + vm_page_free_list(local_freeq, TRUE); } /* @@ -309,12 +791,36 @@ kernel_memory_allocate( */ kern_return_t -kmem_alloc( +kmem_alloc_external( vm_map_t map, vm_offset_t *addrp, vm_size_t size) { - return kernel_memory_allocate(map, addrp, size, 0, 0); + return (kmem_alloc(map, addrp, size, vm_tag_bt())); +} + + +kern_return_t +kmem_alloc( + vm_map_t map, + vm_offset_t *addrp, + vm_size_t size, + vm_tag_t tag) +{ + return kmem_alloc_flags(map, addrp, size, tag, 0); +} + +kern_return_t +kmem_alloc_flags( + vm_map_t map, + vm_offset_t *addrp, + vm_size_t size, + vm_tag_t tag, + int flags) +{ + kern_return_t kr = kernel_memory_allocate(map, addrp, size, 0, flags, tag); + TRACE_MACHLEAKS(KMEM_ALLOC_CODE, KMEM_ALLOC_CODE_2, size, *addrp); + return kr; } /* @@ -333,7 +839,8 @@ kmem_realloc( vm_offset_t oldaddr, vm_size_t oldsize, vm_offset_t *newaddrp, - vm_size_t newsize) + vm_size_t newsize, + vm_tag_t tag) { vm_object_t object; vm_object_offset_t offset; @@ -347,10 +854,13 @@ kmem_realloc( vm_page_t mem; kern_return_t kr; - oldmapmin = vm_map_trunc_page(oldaddr); - oldmapmax = vm_map_round_page(oldaddr + oldsize); + oldmapmin = vm_map_trunc_page(oldaddr, + VM_MAP_PAGE_MASK(map)); + oldmapmax = vm_map_round_page(oldaddr + oldsize, + VM_MAP_PAGE_MASK(map)); oldmapsize = oldmapmax - oldmapmin; - newmapsize = vm_map_round_page(newsize); + newmapsize = vm_map_round_page(newsize, + VM_MAP_PAGE_MASK(map)); /* @@ -361,7 +871,7 @@ kmem_realloc( if (!vm_map_lookup_entry(map, oldmapmin, &oldentry)) panic("kmem_realloc"); - object = oldentry->object.vm_object; + object = VME_OBJECT(oldentry); /* * Increase the size of the object and @@ -374,9 +884,9 @@ kmem_realloc( /* attempt is made to realloc a kmem_alloc'd area */ vm_object_lock(object); vm_map_unlock(map); - if (object->size != oldmapsize) + if (object->vo_size != oldmapsize) panic("kmem_realloc"); - object->size = newmapsize; + object->vo_size = newmapsize; vm_object_unlock(object); /* allocate the new pages while expanded portion of the */ @@ -389,25 +899,24 @@ kmem_realloc( */ kr = vm_map_find_space(map, &newmapaddr, newmapsize, - (vm_map_offset_t) 0, &newentry); + (vm_map_offset_t) 0, 0, &newentry); if (kr != KERN_SUCCESS) { vm_object_lock(object); for(offset = oldmapsize; offset < newmapsize; offset += PAGE_SIZE) { if ((mem = vm_page_lookup(object, offset)) != VM_PAGE_NULL) { - vm_page_lock_queues(); - vm_page_free(mem); - vm_page_unlock_queues(); + VM_PAGE_FREE(mem); } } - object->size = oldmapsize; + object->vo_size = oldmapsize; vm_object_unlock(object); vm_object_deallocate(object); return kr; } - newentry->object.vm_object = object; - newentry->offset = 0; - assert (newentry->wired_count == 0); + VME_OBJECT_SET(newentry, object); + VME_OFFSET_SET(newentry, 0); + VME_ALIAS_SET(newentry, tag); + assert(newentry->wired_count == 0); /* add an extra reference in case we have someone doing an */ @@ -415,18 +924,17 @@ kmem_realloc( vm_object_reference(object); vm_map_unlock(map); - kr = vm_map_wire(map, newmapaddr, newmapaddr + newmapsize, VM_PROT_DEFAULT, FALSE); + kr = vm_map_wire(map, newmapaddr, newmapaddr + newmapsize, + VM_PROT_DEFAULT | VM_PROT_MEMORY_TAG_MAKE(tag), FALSE); if (KERN_SUCCESS != kr) { vm_map_remove(map, newmapaddr, newmapaddr + newmapsize, 0); vm_object_lock(object); for(offset = oldsize; offset < newmapsize; offset += PAGE_SIZE) { if ((mem = vm_page_lookup(object, offset)) != VM_PAGE_NULL) { - vm_page_lock_queues(); - vm_page_free(mem); - vm_page_unlock_queues(); + VM_PAGE_FREE(mem); } } - object->size = oldmapsize; + object->vo_size = oldmapsize; vm_object_unlock(object); vm_object_deallocate(object); return (kr); @@ -438,7 +946,7 @@ kmem_realloc( } /* - * kmem_alloc_wired: + * kmem_alloc_kobject: * * Allocate wired-down memory in the kernel's address map * or a submap. The memory is not zero-filled. @@ -449,18 +957,28 @@ kmem_realloc( */ kern_return_t -kmem_alloc_wired( +kmem_alloc_kobject_external( vm_map_t map, vm_offset_t *addrp, vm_size_t size) { - return kernel_memory_allocate(map, addrp, size, 0, KMA_KOBJECT); + return (kmem_alloc_kobject(map, addrp, size, vm_tag_bt())); +} + +kern_return_t +kmem_alloc_kobject( + vm_map_t map, + vm_offset_t *addrp, + vm_size_t size, + vm_tag_t tag) +{ + return kernel_memory_allocate(map, addrp, size, 0, KMA_KOBJECT, tag); } /* * kmem_alloc_aligned: * - * Like kmem_alloc_wired, except that the memory is aligned. + * Like kmem_alloc_kobject, except that the memory is aligned. * The size should be a power-of-2. */ @@ -468,11 +986,12 @@ kern_return_t kmem_alloc_aligned( vm_map_t map, vm_offset_t *addrp, - vm_size_t size) + vm_size_t size, + vm_tag_t tag) { if ((size & (size - 1)) != 0) panic("kmem_alloc_aligned: size not aligned"); - return kernel_memory_allocate(map, addrp, size, size - 1, KMA_KOBJECT); + return kernel_memory_allocate(map, addrp, size, size - 1, KMA_KOBJECT, tag); } /* @@ -482,24 +1001,36 @@ kmem_alloc_aligned( */ kern_return_t -kmem_alloc_pageable( +kmem_alloc_pageable_external( vm_map_t map, vm_offset_t *addrp, vm_size_t size) +{ + return (kmem_alloc_pageable(map, addrp, size, vm_tag_bt())); +} + +kern_return_t +kmem_alloc_pageable( + vm_map_t map, + vm_offset_t *addrp, + vm_size_t size, + vm_tag_t tag) { vm_map_offset_t map_addr; vm_map_size_t map_size; kern_return_t kr; #ifndef normal - map_addr = (vm_map_min(map)) + 0x1000; + map_addr = (vm_map_min(map)) + PAGE_SIZE; #else map_addr = vm_map_min(map); #endif - map_size = vm_map_round_page(size); + map_size = vm_map_round_page(size, + VM_MAP_PAGE_MASK(map)); kr = vm_map_enter(map, &map_addr, map_size, - (vm_map_offset_t) 0, VM_FLAGS_ANYWHERE, + (vm_map_offset_t) 0, + VM_FLAGS_ANYWHERE | VM_MAKE_TAG(tag), VM_OBJECT_NULL, (vm_object_offset_t) 0, FALSE, VM_PROT_DEFAULT, VM_PROT_ALL, VM_INHERIT_DEFAULT); @@ -514,7 +1045,7 @@ kmem_alloc_pageable( * kmem_free: * * Release a region of kernel virtual memory allocated - * with kmem_alloc, kmem_alloc_wired, or kmem_alloc_pageable, + * with kmem_alloc, kmem_alloc_kobject, or kmem_alloc_pageable, * and return the physical pages associated with that region. */ @@ -526,9 +1057,23 @@ kmem_free( { kern_return_t kr; - kr = vm_map_remove(map, vm_map_trunc_page(addr), - vm_map_round_page(addr + size), - VM_MAP_REMOVE_KUNWIRE); + assert(addr >= VM_MIN_KERNEL_AND_KEXT_ADDRESS); + + TRACE_MACHLEAKS(KMEM_FREE_CODE, KMEM_FREE_CODE_2, size, addr); + + if(size == 0) { +#if MACH_ASSERT + printf("kmem_free called with size==0 for map: %p with addr: 0x%llx\n",map,(uint64_t)addr); +#endif + return; + } + + kr = vm_map_remove(map, + vm_map_trunc_page(addr, + VM_MAP_PAGE_MASK(map)), + vm_map_round_page(addr + size, + VM_MAP_PAGE_MASK(map)), + VM_MAP_REMOVE_KUNWIRE); if (kr != KERN_SUCCESS) panic("kmem_free"); } @@ -539,16 +1084,16 @@ kmem_free( kern_return_t kmem_alloc_pages( - register vm_object_t object, - register vm_object_offset_t offset, - register vm_object_size_t size) + vm_object_t object, + vm_object_offset_t offset, + vm_object_size_t size) { vm_object_size_t alloc_size; alloc_size = vm_object_round_page(size); vm_object_lock(object); while (alloc_size) { - register vm_page_t mem; + vm_page_t mem; /* @@ -569,72 +1114,6 @@ kmem_alloc_pages( return KERN_SUCCESS; } -/* - * Remap wired pages in an object into a new region. - * The object is assumed to be mapped into the kernel map or - * a submap. - */ -void -kmem_remap_pages( - register vm_object_t object, - register vm_object_offset_t offset, - register vm_offset_t start, - register vm_offset_t end, - vm_prot_t protection) -{ - - vm_map_offset_t map_start; - vm_map_offset_t map_end; - - /* - * Mark the pmap region as not pageable. - */ - map_start = vm_map_trunc_page(start); - map_end = vm_map_round_page(end); - - pmap_pageable(kernel_pmap, map_start, map_end, FALSE); - - while (map_start < map_end) { - register vm_page_t mem; - - vm_object_lock(object); - - /* - * Find a page - */ - if ((mem = vm_page_lookup(object, offset)) == VM_PAGE_NULL) - panic("kmem_remap_pages"); - - /* - * Wire it down (again) - */ - vm_page_lock_queues(); - vm_page_wire(mem); - vm_page_unlock_queues(); - vm_object_unlock(object); - - /* - * ENCRYPTED SWAP: - * The page is supposed to be wired now, so it - * shouldn't be encrypted at this point. It can - * safely be entered in the page table. - */ - ASSERT_PAGE_DECRYPTED(mem); - - /* - * Enter it in the kernel pmap. The page isn't busy, - * but this shouldn't be a problem because it is wired. - */ - PMAP_ENTER(kernel_pmap, map_start, mem, protection, - ((unsigned int)(mem->object->wimg_bits)) - & VM_WIMG_MASK, - TRUE); - - map_start += PAGE_SIZE; - offset += PAGE_SIZE; - } -} - /* * kmem_suballoc: * @@ -664,7 +1143,8 @@ kmem_suballoc( vm_map_size_t map_size; kern_return_t kr; - map_size = vm_map_round_page(size); + map_size = vm_map_round_page(size, + VM_MAP_PAGE_MASK(parent)); /* * Need reference on submap object because it is internal @@ -673,8 +1153,10 @@ kmem_suballoc( */ vm_object_reference(vm_submap_object); - map_addr = (flags & VM_FLAGS_ANYWHERE) ? - vm_map_min(parent) : vm_map_trunc_page(*addr); + map_addr = ((flags & VM_FLAGS_ANYWHERE) + ? vm_map_min(parent) + : vm_map_trunc_page(*addr, + VM_MAP_PAGE_MASK(parent))); kr = vm_map_enter(parent, &map_addr, map_size, (vm_map_offset_t) 0, flags, @@ -689,6 +1171,8 @@ kmem_suballoc( map = vm_map_create(vm_map_pmap(parent), map_addr, map_addr + map_size, pageable); if (map == VM_MAP_NULL) panic("kmem_suballoc: vm_map_create failed"); /* "can't happen" */ + /* inherit the parent map's page size */ + vm_map_set_page_shift(map, VM_MAP_PAGE_SHIFT(parent)); kr = vm_map_submap(parent, map_addr, map_addr + map_size, map, map_addr, FALSE); if (kr != KERN_SUCCESS) { @@ -719,39 +1203,52 @@ kmem_init( vm_map_offset_t map_start; vm_map_offset_t map_end; - map_start = vm_map_trunc_page(start); - map_end = vm_map_round_page(end); - - kernel_map = vm_map_create(pmap_kernel(),VM_MIN_KERNEL_ADDRESS, - map_end, FALSE); + map_start = vm_map_trunc_page(start, + VM_MAP_PAGE_MASK(kernel_map)); + map_end = vm_map_round_page(end, + VM_MAP_PAGE_MASK(kernel_map)); + kernel_map = vm_map_create(pmap_kernel(),VM_MIN_KERNEL_AND_KEXT_ADDRESS, + map_end, FALSE); /* * Reserve virtual memory allocated up to this time. */ - - if (start != VM_MIN_KERNEL_ADDRESS) { + if (start != VM_MIN_KERNEL_AND_KEXT_ADDRESS) { vm_map_offset_t map_addr; + kern_return_t kr; + + map_addr = VM_MIN_KERNEL_AND_KEXT_ADDRESS; + kr = vm_map_enter(kernel_map, + &map_addr, + (vm_map_size_t)(map_start - VM_MIN_KERNEL_AND_KEXT_ADDRESS), + (vm_map_offset_t) 0, + VM_FLAGS_FIXED | VM_FLAGS_NO_PMAP_CHECK, + VM_OBJECT_NULL, + (vm_object_offset_t) 0, FALSE, + VM_PROT_NONE, VM_PROT_NONE, + VM_INHERIT_DEFAULT); + + if (kr != KERN_SUCCESS) { + panic("kmem_init(0x%llx,0x%llx): vm_map_enter(0x%llx,0x%llx) error 0x%x\n", + (uint64_t) start, (uint64_t) end, + (uint64_t) VM_MIN_KERNEL_AND_KEXT_ADDRESS, + (uint64_t) (map_start - VM_MIN_KERNEL_AND_KEXT_ADDRESS), + kr); + } + } - map_addr = VM_MIN_KERNEL_ADDRESS; - (void) vm_map_enter(kernel_map, - &map_addr, - (vm_map_size_t)(map_start - VM_MIN_KERNEL_ADDRESS), - (vm_map_offset_t) 0, - VM_FLAGS_ANYWHERE | VM_FLAGS_NO_PMAP_CHECK, - VM_OBJECT_NULL, - (vm_object_offset_t) 0, FALSE, - VM_PROT_DEFAULT, VM_PROT_ALL, - VM_INHERIT_DEFAULT); - } - - /* - * Account for kernel memory (text, data, bss, vm shenanigans). - * This may include inaccessible "holes" as determined by what - * the machine-dependent init code includes in max_mem. - */ - vm_page_wire_count = (atop_64(max_mem) - (vm_page_free_count - + vm_page_active_count - + vm_page_inactive_count)); + /* + * Set the default global user wire limit which limits the amount of + * memory that can be locked via mlock(). We set this to the total + * amount of memory that are potentially usable by a user app (max_mem) + * minus a certain amount. This can be overridden via a sysctl. + */ + vm_global_no_user_wire_amount = MIN(max_mem*20/100, + VM_NOT_USER_WIREABLE); + vm_global_user_wire_limit = max_mem - vm_global_no_user_wire_amount; + + /* the default per user limit is the same as the global limit */ + vm_user_wire_limit = vm_global_user_wire_limit; } @@ -851,7 +1348,7 @@ vm_conflict_check( while(vm_map_lookup_entry(map, off, &entry)) { local_len = len; - if (entry->object.vm_object == VM_OBJECT_NULL) { + if (VME_OBJECT(entry) == VM_OBJECT_NULL) { vm_map_unlock(map); return KERN_SUCCESS; } @@ -859,16 +1356,16 @@ vm_conflict_check( vm_map_t old_map; old_map = map; - vm_map_lock(entry->object.sub_map); - map = entry->object.sub_map; - off = entry->offset + (off - entry->vme_start); + vm_map_lock(VME_SUBMAP(entry)); + map = VME_SUBMAP(entry); + off = VME_OFFSET(entry) + (off - entry->vme_start); vm_map_unlock(old_map); continue; } - obj = entry->object.vm_object; - obj_off = (off - entry->vme_start) + entry->offset; + obj = VME_OBJECT(entry); + obj_off = (off - entry->vme_start) + VME_OFFSET(entry); while(obj->shadow) { - obj_off += obj->shadow_offset; + obj_off += obj->vo_shadow_offset; obj = obj->shadow; } if((obj->pager_created) && (obj->pager == pager)) { @@ -938,3 +1435,55 @@ vm_conflict_check( vm_map_unlock(map); return kr; } + +/* + * + * The following two functions are to be used when exposing kernel + * addresses to userspace via any of the various debug or info + * facilities that exist. These are basically the same as VM_KERNEL_ADDRPERM() + * and VM_KERNEL_UNSLIDE_OR_PERM() except they use a different random seed and + * are exported to KEXTs. + * + * NOTE: USE THE MACRO VERSIONS OF THESE FUNCTIONS (in vm_param.h) FROM WITHIN THE KERNEL + */ + +/* + * vm_kernel_addrperm_external: + * + * Used when exposing an address to userspace which is in the kernel's + * "heap". These addresses are not loaded from anywhere and are resultingly + * unslid. We apply a permutation value to obscure the address. + */ +void +vm_kernel_addrperm_external( + vm_offset_t addr, + vm_offset_t *perm_addr) +{ + if (addr == 0) { + *perm_addr = 0; + return; + } + + *perm_addr = (addr + vm_kernel_addrperm_ext); + return; +} + +/* + * vm_kernel_unslide_or_perm_external: + * + * Use this macro when exposing an address to userspace that could come from + * either kernel text/data *or* the heap. + */ +void +vm_kernel_unslide_or_perm_external( + vm_offset_t addr, + vm_offset_t *up_addr) +{ + if (VM_KERNEL_IS_SLID(addr)) { + *up_addr = addr - vm_kernel_slide; + return; + } + + vm_kernel_addrperm_external(addr, up_addr); + return; +}