diff options
Diffstat (limited to 'vm')
| -rw-r--r-- | vm/pmap.h | 15 | ||||
| -rw-r--r-- | vm/vm_fault.c | 4 | ||||
| -rw-r--r-- | vm/vm_init.c | 1 | ||||
| -rw-r--r-- | vm/vm_object.c | 3 | ||||
| -rw-r--r-- | vm/vm_page.c | 762 | ||||
| -rw-r--r-- | vm/vm_page.h | 220 | ||||
| -rw-r--r-- | vm/vm_resident.c | 546 |
7 files changed, 1137 insertions, 414 deletions
@@ -67,9 +67,6 @@ extern vm_offset_t pmap_steal_memory(vm_size_t); /* During VM initialization, report remaining unused physical pages. */ extern unsigned int pmap_free_pages(void); -/* During VM initialization, use remaining physical pages to allocate page - * frames. */ -extern void pmap_startup(vm_offset_t *, vm_offset_t *); /* Initialization, after kernel runs in virtual memory. */ extern void pmap_init(void); @@ -80,18 +77,14 @@ extern void pmap_init(void); * Otherwise, it must implement * pmap_free_pages * pmap_virtual_space - * pmap_next_page * pmap_init - * and vm/vm_resident.c implements pmap_steal_memory and pmap_startup - * using pmap_free_pages, pmap_next_page, pmap_virtual_space, - * and pmap_enter. pmap_free_pages may over-estimate the number - * of unused physical pages, and pmap_next_page may return FALSE - * to indicate that there are no more unused pages to return. + * and vm/vm_resident.c implements pmap_steal_memory using + * pmap_free_pages, pmap_virtual_space, and pmap_enter. + * + * pmap_free_pages may over-estimate the number of unused physical pages. * However, for best performance pmap_free_pages should be accurate. */ -/* During VM initialization, return the next unused physical page. */ -extern boolean_t pmap_next_page(vm_offset_t *); /* During VM initialization, report virtual space available for the kernel. */ extern void pmap_virtual_space(vm_offset_t *, vm_offset_t *); #endif /* MACHINE_PAGES */ diff --git a/vm/vm_fault.c b/vm/vm_fault.c index 46779f63..4d674174 100644 --- a/vm/vm_fault.c +++ b/vm/vm_fault.c @@ -607,7 +607,7 @@ vm_fault_return_t vm_fault_page( * won't block for pages. */ - if (m->fictitious && !vm_page_convert(m, FALSE)) { + if (m->fictitious && !vm_page_convert(&m, FALSE)) { VM_PAGE_FREE(m); vm_fault_cleanup(object, first_m); return(VM_FAULT_MEMORY_SHORTAGE); @@ -725,7 +725,7 @@ vm_fault_return_t vm_fault_page( assert(m->object == object); first_m = VM_PAGE_NULL; - if (m->fictitious && !vm_page_convert(m, !object->internal)) { + if (m->fictitious && !vm_page_convert(&m, !object->internal)) { VM_PAGE_FREE(m); vm_fault_cleanup(object, VM_PAGE_NULL); return(VM_FAULT_MEMORY_SHORTAGE); diff --git a/vm/vm_init.c b/vm/vm_init.c index 3d1081cc..23d5d46e 100644 --- a/vm/vm_init.c +++ b/vm/vm_init.c @@ -83,4 +83,5 @@ void vm_mem_init(void) { vm_object_init(); memory_object_proxy_init(); + vm_page_info_all(); } diff --git a/vm/vm_object.c b/vm/vm_object.c index 6666fcba..eda03c65 100644 --- a/vm/vm_object.c +++ b/vm/vm_object.c @@ -2891,7 +2891,8 @@ vm_object_page_map( VM_PAGE_FREE(old_page); } - vm_page_init(m, addr); + vm_page_init(m); + m->phys_addr = addr; m->private = TRUE; /* don`t free page */ m->wire_count = 1; vm_page_lock_queues(); diff --git a/vm/vm_page.c b/vm/vm_page.c new file mode 100644 index 00000000..a539ab41 --- /dev/null +++ b/vm/vm_page.c @@ -0,0 +1,762 @@ +/* + * Copyright (c) 2010-2014 Richard Braun. + * + * This program 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, either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see <http://www.gnu.org/licenses/>. + * + * + * This implementation uses the binary buddy system to manage its heap. + * Descriptions of the buddy system can be found in the following works : + * - "UNIX Internals: The New Frontiers", by Uresh Vahalia. + * - "Dynamic Storage Allocation: A Survey and Critical Review", + * by Paul R. Wilson, Mark S. Johnstone, Michael Neely, and David Boles. + * + * In addition, this allocator uses per-CPU pools of pages for order 0 + * (i.e. single page) allocations. These pools act as caches (but are named + * differently to avoid confusion with CPU caches) that reduce contention on + * multiprocessor systems. When a pool is empty and cannot provide a page, + * it is filled by transferring multiple pages from the backend buddy system. + * The symmetric case is handled likewise. + */ + +#include <string.h> +#include <kern/assert.h> +#include <kern/cpu_number.h> +#include <kern/debug.h> +#include <kern/list.h> +#include <kern/lock.h> +#include <kern/macros.h> +#include <kern/printf.h> +#include <kern/thread.h> +#include <mach/vm_param.h> +#include <machine/pmap.h> +#include <sys/types.h> +#include <vm/vm_page.h> + +#define __init +#define __initdata +#define __read_mostly + +#define thread_pin() +#define thread_unpin() + +/* + * Number of free block lists per segment. + */ +#define VM_PAGE_NR_FREE_LISTS 11 + +/* + * The size of a CPU pool is computed by dividing the number of pages in its + * containing segment by this value. + */ +#define VM_PAGE_CPU_POOL_RATIO 1024 + +/* + * Maximum number of pages in a CPU pool. + */ +#define VM_PAGE_CPU_POOL_MAX_SIZE 128 + +/* + * The transfer size of a CPU pool is computed by dividing the pool size by + * this value. + */ +#define VM_PAGE_CPU_POOL_TRANSFER_RATIO 2 + +/* + * Per-processor cache of pages. + */ +struct vm_page_cpu_pool { + simple_lock_data_t lock; + int size; + int transfer_size; + int nr_pages; + struct list pages; +} __aligned(CPU_L1_SIZE); + +/* + * Special order value for pages that aren't in a free list. Such pages are + * either allocated, or part of a free block of pages but not the head page. + */ +#define VM_PAGE_ORDER_UNLISTED ((unsigned short)-1) + +/* + * Doubly-linked list of free blocks. + */ +struct vm_page_free_list { + unsigned long size; + struct list blocks; +}; + +/* + * Segment name buffer size. + */ +#define VM_PAGE_NAME_SIZE 16 + +/* + * Segment of contiguous memory. + */ +struct vm_page_seg { + struct vm_page_cpu_pool cpu_pools[NCPUS]; + + phys_addr_t start; + phys_addr_t end; + struct vm_page *pages; + struct vm_page *pages_end; + simple_lock_data_t lock; + struct vm_page_free_list free_lists[VM_PAGE_NR_FREE_LISTS]; + unsigned long nr_free_pages; +}; + +/* + * Bootstrap information about a segment. + */ +struct vm_page_boot_seg { + phys_addr_t start; + phys_addr_t end; + phys_addr_t avail_start; + phys_addr_t avail_end; +}; + +static int vm_page_is_ready __read_mostly; + +/* + * Segment table. + * + * The system supports a maximum of 4 segments : + * - DMA: suitable for DMA + * - DMA32: suitable for DMA when devices support 32-bits addressing + * - DIRECTMAP: direct physical mapping, allows direct access from + * the kernel with a simple offset translation + * - HIGHMEM: must be mapped before it can be accessed + * + * Segments are ordered by priority, 0 being the lowest priority. Their + * relative priorities are DMA < DMA32 < DIRECTMAP < HIGHMEM. Some segments + * may actually be aliases for others, e.g. if DMA is always possible from + * the direct physical mapping, DMA and DMA32 are aliases for DIRECTMAP, + * in which case the segment table contains DIRECTMAP and HIGHMEM only. + */ +static struct vm_page_seg vm_page_segs[VM_PAGE_MAX_SEGS]; + +/* + * Bootstrap segment table. + */ +static struct vm_page_boot_seg vm_page_boot_segs[VM_PAGE_MAX_SEGS] __initdata; + +/* + * Number of loaded segments. + */ +static unsigned int vm_page_segs_size __read_mostly; + +static void __init +vm_page_init_pa(struct vm_page *page, unsigned short seg_index, phys_addr_t pa) +{ + memset(page, 0, sizeof(*page)); + vm_page_init(page); /* vm_resident members */ + page->type = VM_PT_RESERVED; + page->seg_index = seg_index; + page->order = VM_PAGE_ORDER_UNLISTED; + page->phys_addr = pa; +} + +void +vm_page_set_type(struct vm_page *page, unsigned int order, unsigned short type) +{ + unsigned int i, nr_pages; + + nr_pages = 1 << order; + + for (i = 0; i < nr_pages; i++) + page[i].type = type; +} + +static void __init +vm_page_free_list_init(struct vm_page_free_list *free_list) +{ + free_list->size = 0; + list_init(&free_list->blocks); +} + +static inline void +vm_page_free_list_insert(struct vm_page_free_list *free_list, + struct vm_page *page) +{ + assert(page->order == VM_PAGE_ORDER_UNLISTED); + + free_list->size++; + list_insert_head(&free_list->blocks, &page->node); +} + +static inline void +vm_page_free_list_remove(struct vm_page_free_list *free_list, + struct vm_page *page) +{ + assert(page->order != VM_PAGE_ORDER_UNLISTED); + + free_list->size--; + list_remove(&page->node); +} + +static struct vm_page * +vm_page_seg_alloc_from_buddy(struct vm_page_seg *seg, unsigned int order) +{ + struct vm_page_free_list *free_list = free_list; + struct vm_page *page, *buddy; + unsigned int i; + + assert(order < VM_PAGE_NR_FREE_LISTS); + + for (i = order; i < VM_PAGE_NR_FREE_LISTS; i++) { + free_list = &seg->free_lists[i]; + + if (free_list->size != 0) + break; + } + + if (i == VM_PAGE_NR_FREE_LISTS) + return NULL; + + page = list_first_entry(&free_list->blocks, struct vm_page, node); + vm_page_free_list_remove(free_list, page); + page->order = VM_PAGE_ORDER_UNLISTED; + + while (i > order) { + i--; + buddy = &page[1 << i]; + vm_page_free_list_insert(&seg->free_lists[i], buddy); + buddy->order = i; + } + + seg->nr_free_pages -= (1 << order); + return page; +} + +static void +vm_page_seg_free_to_buddy(struct vm_page_seg *seg, struct vm_page *page, + unsigned int order) +{ + struct vm_page *buddy; + phys_addr_t pa, buddy_pa; + unsigned int nr_pages; + + assert(page >= seg->pages); + assert(page < seg->pages_end); + assert(page->order == VM_PAGE_ORDER_UNLISTED); + assert(order < VM_PAGE_NR_FREE_LISTS); + + nr_pages = (1 << order); + pa = page->phys_addr; + + while (order < (VM_PAGE_NR_FREE_LISTS - 1)) { + buddy_pa = pa ^ vm_page_ptoa(1 << order); + + if ((buddy_pa < seg->start) || (buddy_pa >= seg->end)) + break; + + buddy = &seg->pages[vm_page_atop(buddy_pa - seg->start)]; + + if (buddy->order != order) + break; + + vm_page_free_list_remove(&seg->free_lists[order], buddy); + buddy->order = VM_PAGE_ORDER_UNLISTED; + order++; + pa &= -vm_page_ptoa(1 << order); + page = &seg->pages[vm_page_atop(pa - seg->start)]; + } + + vm_page_free_list_insert(&seg->free_lists[order], page); + page->order = order; + seg->nr_free_pages += nr_pages; +} + +static void __init +vm_page_cpu_pool_init(struct vm_page_cpu_pool *cpu_pool, int size) +{ + simple_lock_init(&cpu_pool->lock); + cpu_pool->size = size; + cpu_pool->transfer_size = (size + VM_PAGE_CPU_POOL_TRANSFER_RATIO - 1) + / VM_PAGE_CPU_POOL_TRANSFER_RATIO; + cpu_pool->nr_pages = 0; + list_init(&cpu_pool->pages); +} + +static inline struct vm_page_cpu_pool * +vm_page_cpu_pool_get(struct vm_page_seg *seg) +{ + return &seg->cpu_pools[cpu_number()]; +} + +static inline struct vm_page * +vm_page_cpu_pool_pop(struct vm_page_cpu_pool *cpu_pool) +{ + struct vm_page *page; + + assert(cpu_pool->nr_pages != 0); + cpu_pool->nr_pages--; + page = list_first_entry(&cpu_pool->pages, struct vm_page, node); + list_remove(&page->node); + return page; +} + +static inline void +vm_page_cpu_pool_push(struct vm_page_cpu_pool *cpu_pool, struct vm_page *page) +{ + assert(cpu_pool->nr_pages < cpu_pool->size); + cpu_pool->nr_pages++; + list_insert_head(&cpu_pool->pages, &page->node); +} + +static int +vm_page_cpu_pool_fill(struct vm_page_cpu_pool *cpu_pool, + struct vm_page_seg *seg) +{ + struct vm_page *page; + int i; + + assert(cpu_pool->nr_pages == 0); + + simple_lock(&seg->lock); + + for (i = 0; i < cpu_pool->transfer_size; i++) { + page = vm_page_seg_alloc_from_buddy(seg, 0); + + if (page == NULL) + break; + + vm_page_cpu_pool_push(cpu_pool, page); + } + + simple_unlock(&seg->lock); + + return i; +} + +static void +vm_page_cpu_pool_drain(struct vm_page_cpu_pool *cpu_pool, + struct vm_page_seg *seg) +{ + struct vm_page *page; + int i; + + assert(cpu_pool->nr_pages == cpu_pool->size); + + simple_lock(&seg->lock); + + for (i = cpu_pool->transfer_size; i > 0; i--) { + page = vm_page_cpu_pool_pop(cpu_pool); + vm_page_seg_free_to_buddy(seg, page, 0); + } + + simple_unlock(&seg->lock); +} + +static phys_addr_t __init +vm_page_seg_size(struct vm_page_seg *seg) +{ + return seg->end - seg->start; +} + +static int __init +vm_page_seg_compute_pool_size(struct vm_page_seg *seg) +{ + phys_addr_t size; + + size = vm_page_atop(vm_page_seg_size(seg)) / VM_PAGE_CPU_POOL_RATIO; + + if (size == 0) + size = 1; + else if (size > VM_PAGE_CPU_POOL_MAX_SIZE) + size = VM_PAGE_CPU_POOL_MAX_SIZE; + + return size; +} + +static void __init +vm_page_seg_init(struct vm_page_seg *seg, phys_addr_t start, phys_addr_t end, + struct vm_page *pages) +{ + phys_addr_t pa; + int pool_size; + unsigned int i; + + seg->start = start; + seg->end = end; + pool_size = vm_page_seg_compute_pool_size(seg); + + for (i = 0; i < ARRAY_SIZE(seg->cpu_pools); i++) + vm_page_cpu_pool_init(&seg->cpu_pools[i], pool_size); + + seg->pages = pages; + seg->pages_end = pages + vm_page_atop(vm_page_seg_size(seg)); + simple_lock_init(&seg->lock); + + for (i = 0; i < ARRAY_SIZE(seg->free_lists); i++) + vm_page_free_list_init(&seg->free_lists[i]); + + seg->nr_free_pages = 0; + i = seg - vm_page_segs; + + for (pa = seg->start; pa < seg->end; pa += PAGE_SIZE) + vm_page_init_pa(&pages[vm_page_atop(pa - seg->start)], i, pa); +} + +static struct vm_page * +vm_page_seg_alloc(struct vm_page_seg *seg, unsigned int order, + unsigned short type) +{ + struct vm_page_cpu_pool *cpu_pool; + struct vm_page *page; + int filled; + + assert(order < VM_PAGE_NR_FREE_LISTS); + + if (order == 0) { + thread_pin(); + cpu_pool = vm_page_cpu_pool_get(seg); + simple_lock(&cpu_pool->lock); + + if (cpu_pool->nr_pages == 0) { + filled = vm_page_cpu_pool_fill(cpu_pool, seg); + + if (!filled) { + simple_unlock(&cpu_pool->lock); + thread_unpin(); + return NULL; + } + } + + page = vm_page_cpu_pool_pop(cpu_pool); + simple_unlock(&cpu_pool->lock); + thread_unpin(); + } else { + simple_lock(&seg->lock); + page = vm_page_seg_alloc_from_buddy(seg, order); + simple_unlock(&seg->lock); + } + + assert(page->type == VM_PT_FREE); + vm_page_set_type(page, order, type); + return page; +} + +static void +vm_page_seg_free(struct vm_page_seg *seg, struct vm_page *page, + unsigned int order) +{ + struct vm_page_cpu_pool *cpu_pool; + + assert(page->type != VM_PT_FREE); + assert(order < VM_PAGE_NR_FREE_LISTS); + + vm_page_set_type(page, order, VM_PT_FREE); + + if (order == 0) { + thread_pin(); + cpu_pool = vm_page_cpu_pool_get(seg); + simple_lock(&cpu_pool->lock); + + if (cpu_pool->nr_pages == cpu_pool->size) + vm_page_cpu_pool_drain(cpu_pool, seg); + + vm_page_cpu_pool_push(cpu_pool, page); + simple_unlock(&cpu_pool->lock); + thread_unpin(); + } else { + simple_lock(&seg->lock); + vm_page_seg_free_to_buddy(seg, page, order); + simple_unlock(&seg->lock); + } +} + +void __init +vm_page_load(unsigned int seg_index, phys_addr_t start, phys_addr_t end, + phys_addr_t avail_start, phys_addr_t avail_end) +{ + struct vm_page_boot_seg *seg; + + assert(seg_index < ARRAY_SIZE(vm_page_boot_segs)); + assert(vm_page_aligned(start)); + assert(vm_page_aligned(end)); + assert(vm_page_aligned(avail_start)); + assert(vm_page_aligned(avail_end)); + assert(start < end); + assert(start <= avail_start); + assert(avail_end <= end); + assert(vm_page_segs_size < ARRAY_SIZE(vm_page_boot_segs)); + + seg = &vm_page_boot_segs[seg_index]; + seg->start = start; + seg->end = end; + seg->avail_start = avail_start; + seg->avail_end = avail_end; + vm_page_segs_size++; +} + +int +vm_page_ready(void) +{ + return vm_page_is_ready; +} + +static unsigned int +vm_page_select_alloc_seg(unsigned int selector) +{ + unsigned int seg_index; + + switch (selector) { + case VM_PAGE_SEL_DMA: + seg_index = VM_PAGE_SEG_DMA; + break; + case VM_PAGE_SEL_DMA32: + seg_index = VM_PAGE_SEG_DMA32; + break; + case VM_PAGE_SEL_DIRECTMAP: + seg_index = VM_PAGE_SEG_DIRECTMAP; + break; + case VM_PAGE_SEL_HIGHMEM: + seg_index = VM_PAGE_SEG_HIGHMEM; + break; + default: + panic("vm_page: invalid selector"); + } + + return MIN(vm_page_segs_size - 1, seg_index); +} + +static int __init +vm_page_boot_seg_loaded(const struct vm_page_boot_seg *seg) +{ + return (seg->end != 0); +} + +static void __init +vm_page_check_boot_segs(void) +{ + unsigned int i; + int expect_loaded; + + if (vm_page_segs_size == 0) + panic("vm_page: no physical memory loaded"); + + for (i = 0; i < ARRAY_SIZE(vm_page_boot_segs); i++) { + expect_loaded = (i < vm_page_segs_size); + + if (vm_page_boot_seg_loaded(&vm_page_boot_segs[i]) == expect_loaded) + continue; + + panic("vm_page: invalid boot segment table"); + } +} + +static phys_addr_t __init +vm_page_boot_seg_size(struct vm_page_boot_seg *seg) +{ + return seg->end - seg->start; +} + +static phys_addr_t __init +vm_page_boot_seg_avail_size(struct vm_page_boot_seg *seg) +{ + return seg->avail_end - seg->avail_start; +} + +unsigned long __init +vm_page_bootalloc(size_t size) +{ + struct vm_page_boot_seg *seg; + phys_addr_t pa; + unsigned int i; + + for (i = vm_page_select_alloc_seg(VM_PAGE_SEL_DIRECTMAP); + i < vm_page_segs_size; + i--) { + seg = &vm_page_boot_segs[i]; + + if (size <= vm_page_boot_seg_avail_size(seg)) { + pa = seg->avail_start; + seg->avail_start += vm_page_round(size); + return pa; + } + } + + panic("vm_page: no physical memory available"); +} + +void __init +vm_page_setup(void) +{ + struct vm_page_boot_seg *boot_seg; + struct vm_page_seg *seg; + struct vm_page *table, *page, *end; + size_t nr_pages, table_size; + unsigned long va; + unsigned int i; + phys_addr_t pa; + + vm_page_check_boot_segs(); + + /* + * Compute the page table size. + */ + nr_pages = 0; + + for (i = 0; i < vm_page_segs_size; i++) + nr_pages += vm_page_atop(vm_page_boot_seg_size(&vm_page_boot_segs[i])); + + table_size = vm_page_round(nr_pages * sizeof(struct vm_page)); + printf("vm_page: page table size: %lu entries (%luk)\n", nr_pages, + table_size >> 10); + table = (struct vm_page *)pmap_steal_memory(table_size); + va = (unsigned long)table; + + /* + * Initialize the segments, associating them to the page table. When + * the segments are initialized, all their pages are set allocated. + * Pages are then released, which populates the free lists. + */ + for (i = 0; i < vm_page_segs_size; i++) { + seg = &vm_page_segs[i]; + boot_seg = &vm_page_boot_segs[i]; + vm_page_seg_init(seg, boot_seg->start, boot_seg->end, table); + page = seg->pages + vm_page_atop(boot_seg->avail_start + - boot_seg->start); + end = seg->pages + vm_page_atop(boot_seg->avail_end + - boot_seg->start); + + while (page < end) { + page->type = VM_PT_FREE; + vm_page_seg_free_to_buddy(seg, page, 0); + page++; + + /* XXX */ + if (i <= VM_PAGE_SEG_DIRECTMAP) + vm_page_free_count++; + } + + table += vm_page_atop(vm_page_seg_size(seg)); + } + + while (va < (unsigned long)table) { + pa = pmap_extract(kernel_pmap, va); + page = vm_page_lookup_pa(pa); + assert((page != NULL) && (page->type == VM_PT_RESERVED)); + page->type = VM_PT_TABLE; + va += PAGE_SIZE; + } + + vm_page_is_ready = 1; +} + +void __init +vm_page_manage(struct vm_page *page) +{ + assert(page->seg_index < ARRAY_SIZE(vm_page_segs)); + assert(page->type == VM_PT_RESERVED); + + vm_page_set_type(page, 0, VM_PT_FREE); + vm_page_seg_free_to_buddy(&vm_page_segs[page->seg_index], page, 0); +} + +struct vm_page * +vm_page_lookup_pa(phys_addr_t pa) +{ + struct vm_page_seg *seg; + unsigned int i; + + for (i = 0; i < vm_page_segs_size; i++) { + seg = &vm_page_segs[i]; + + if ((pa >= seg->start) && (pa < seg->end)) + return &seg->pages[vm_page_atop(pa - seg->start)]; + } + + return NULL; +} + +struct vm_page * +vm_page_alloc_pa(unsigned int order, unsigned int selector, unsigned short type) +{ + struct vm_page *page; + unsigned int i; + + for (i = vm_page_select_alloc_seg(selector); i < vm_page_segs_size; i--) { + page = vm_page_seg_alloc(&vm_page_segs[i], order, type); + + if (page != NULL) + return page; + } + + if (type == VM_PT_PMAP) + panic("vm_page: unable to allocate pmap page"); + + return NULL; +} + +void +vm_page_free_pa(struct vm_page *page, unsigned int order) +{ + assert(page->seg_index < ARRAY_SIZE(vm_page_segs)); + + vm_page_seg_free(&vm_page_segs[page->seg_index], page, order); +} + +const char * +vm_page_seg_name(unsigned int seg_index) +{ + /* Don't use a switch statement since segments can be aliased */ + if (seg_index == VM_PAGE_SEG_HIGHMEM) + return "HIGHMEM"; + else if (seg_index == VM_PAGE_SEG_DIRECTMAP) + return "DIRECTMAP"; + else if (seg_index == VM_PAGE_SEG_DMA32) + return "DMA32"; + else if (seg_index == VM_PAGE_SEG_DMA) + return "DMA"; + else + panic("vm_page: invalid segment index"); +} + +void +vm_page_info_all(void) +{ + struct vm_page_seg *seg; + unsigned long pages; + unsigned int i; + + for (i = 0; i < vm_page_segs_size; i++) { + seg = &vm_page_segs[i]; + pages = (unsigned long)(seg->pages_end - seg->pages); + printf("vm_page: %s: pages: %lu (%luM), free: %lu (%luM)\n", + vm_page_seg_name(i), pages, pages >> (20 - PAGE_SHIFT), + seg->nr_free_pages, seg->nr_free_pages >> (20 - PAGE_SHIFT)); + } +} + +phys_addr_t +vm_page_mem_size(void) +{ + phys_addr_t total; + unsigned int i; + + total = 0; + + for (i = 0; i < vm_page_segs_size; i++) { + /* XXX */ + if (i > VM_PAGE_SEG_DIRECTMAP) + continue; + + total += vm_page_seg_size(&vm_page_segs[i]); + } + + return total; +} diff --git a/vm/vm_page.h b/vm/vm_page.h index e6a8c497..7607aad0 100644 --- a/vm/vm_page.h +++ b/vm/vm_page.h @@ -36,11 +36,12 @@ #include <mach/boolean.h> #include <mach/vm_prot.h> -#include <mach/vm_param.h> +#include <machine/vm_param.h> #include <vm/vm_object.h> #include <vm/vm_types.h> #include <kern/queue.h> #include <kern/lock.h> +#include <kern/log2.h> #include <kern/macros.h> #include <kern/sched_prim.h> /* definitions of wait/wakeup */ @@ -76,6 +77,22 @@ */ struct vm_page { + /* Members used in the vm_page module only */ + struct list node; + unsigned short type; + unsigned short seg_index; + unsigned short order; + + /* + * This member is used throughout the code and may only change for + * fictitious pages. + */ + phys_addr_t phys_addr; + + /* We use an empty struct as the delimiter. */ + struct {} vm_page_header; +#define VM_PAGE_HEADER_SIZE offsetof(struct vm_page, vm_page_header) + queue_chain_t pageq; /* queue info for FIFO * queue or free list (P) */ queue_chain_t listq; /* all pages in same object (O) */ @@ -110,8 +127,6 @@ struct vm_page { * without having data. (O) * [See vm_object_overwrite] */ - vm_offset_t phys_addr; /* Physical address of page, passed - * to pmap_enter (read-only) */ vm_prot_t page_lock; /* Uses prohibited by data manager (O) */ vm_prot_t unlock_request; /* Outstanding unlock request (O) */ }; @@ -140,8 +155,6 @@ struct vm_page { */ extern -vm_page_t vm_page_queue_free; /* memory free queue */ -extern vm_page_t vm_page_queue_fictitious; /* fictitious free queue */ extern queue_head_t vm_page_queue_active; /* active memory queue */ @@ -196,25 +209,21 @@ extern void vm_page_bootstrap( vm_offset_t *endp); extern void vm_page_module_init(void); -extern void vm_page_create( - vm_offset_t start, - vm_offset_t end); extern vm_page_t vm_page_lookup( vm_object_t object, vm_offset_t offset); extern vm_page_t vm_page_grab_fictitious(void); -extern void vm_page_release_fictitious(vm_page_t); -extern boolean_t vm_page_convert(vm_page_t, boolean_t); +extern boolean_t vm_page_convert(vm_page_t *, boolean_t); extern void vm_page_more_fictitious(void); extern vm_page_t vm_page_grab(boolean_t); -extern void vm_page_release(vm_page_t, boolean_t); +extern vm_page_t vm_page_grab_contig(vm_size_t, unsigned int); +extern void vm_page_free_contig(vm_page_t, vm_size_t); extern void vm_page_wait(void (*)(void)); extern vm_page_t vm_page_alloc( vm_object_t object, vm_offset_t offset); extern void vm_page_init( - vm_page_t mem, - vm_offset_t phys_addr); + vm_page_t mem); extern void vm_page_free(vm_page_t); extern void vm_page_activate(vm_page_t); extern void vm_page_deactivate(vm_page_t); @@ -312,4 +321,189 @@ extern unsigned int vm_page_info( } \ MACRO_END +/* + * Copyright (c) 2010-2014 Richard Braun. + * + * This program 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, either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see <http://www.gnu.org/licenses/>. + * + * + * Physical page management. + */ + +/* + * Address/page conversion and rounding macros (not inline functions to + * be easily usable on both virtual and physical addresses, which may not + * have the same type size). + */ +#define vm_page_atop(addr) ((addr) >> PAGE_SHIFT) +#define vm_page_ptoa(page) ((page) << PAGE_SHIFT) +#define vm_page_trunc(addr) P2ALIGN(addr, PAGE_SIZE) +#define vm_page_round(addr) P2ROUND(addr, PAGE_SIZE) +#define vm_page_aligned(addr) P2ALIGNED(addr, PAGE_SIZE) + +/* + * Segment selectors. + * + * Selector-to-segment-list translation table : + * DMA DMA + * DMA32 DMA32 DMA + * DIRECTMAP DIRECTMAP DMA32 DMA + * HIGHMEM HIGHMEM DIRECTMAP DMA32 DMA + */ +#define VM_PAGE_SEL_DMA 0 +#define VM_PAGE_SEL_DMA32 1 +#define VM_PAGE_SEL_DIRECTMAP 2 +#define VM_PAGE_SEL_HIGHMEM 3 + +/* + * Page usage types. + * + * Failing to allocate pmap pages will cause a kernel panic. + * TODO Obviously, this needs to be addressed, e.g. with a reserved pool of + * pages. + */ +#define VM_PT_FREE 0 /* Page unused */ +#define VM_PT_RESERVED 1 /* Page reserved at boot time */ +#define VM_PT_TABLE 2 /* Page is part of the page table */ +#define VM_PT_PMAP 3 /* Page stores pmap-specific data */ +#define VM_PT_KMEM 4 /* Page is part of a kmem slab */ +#define VM_PT_KERNEL 5 /* Type for generic kernel allocations */ + +static inline unsigned short +vm_page_type(const struct vm_page *page) +{ + return page->type; +} + +void vm_page_set_type(struct vm_page *page, unsigned int order, + unsigned short type); + +static inline unsigned int +vm_page_order(size_t size) +{ + return iorder2(vm_page_atop(vm_page_round(size))); +} + +static inline phys_addr_t +vm_page_to_pa(const struct vm_page *page) +{ + return page->phys_addr; +} + +#if 0 +static inline unsigned long +vm_page_direct_va(phys_addr_t pa) +{ + assert(pa < VM_PAGE_DIRECTMAP_LIMIT); + return ((unsigned long)pa + VM_MIN_DIRECTMAP_ADDRESS); +} + +static inline phys_addr_t +vm_page_direct_pa(unsigned long va) +{ + assert(va >= VM_MIN_DIRECTMAP_ADDRESS); + assert(va < VM_MAX_DIRECTMAP_ADDRESS); + return (va - VM_MIN_DIRECTMAP_ADDRESS); +} + +static inline void * +vm_page_direct_ptr(const struct vm_page *page) +{ + return (void *)vm_page_direct_va(vm_page_to_pa(page)); +} +#endif + +/* + * Load physical memory into the vm_page module at boot time. + * + * The avail_start and avail_end parameters are used to maintain a simple + * heap for bootstrap allocations. + * + * All addresses must be page-aligned. Segments can be loaded in any order. + */ +void vm_page_load(unsigned int seg_index, phys_addr_t start, phys_addr_t end, + phys_addr_t avail_start, phys_addr_t avail_end); + +/* + * Return true if the vm_page module is completely initialized, false + * otherwise, in which case only vm_page_bootalloc() can be used for + * allocations. + */ +int vm_page_ready(void); + +/* + * Early allocation function. + * + * This function is used by the vm_resident module to implement + * pmap_steal_memory. It can be used after physical segments have been loaded + * and before the vm_page module is initialized. + */ +unsigned long vm_page_bootalloc(size_t size); + +/* + * Set up the vm_page module. + * + * Architecture-specific code must have loaded segments before calling this + * function. Segments must comply with the selector-to-segment-list table, + * e.g. HIGHMEM is loaded if and only if DIRECTMAP, DMA32 and DMA are loaded, + * notwithstanding segment aliasing. + * + * Once this function returns, the vm_page module is ready, and normal + * allocation functions can be used. + */ +void vm_page_setup(void); + +/* + * Make the given page managed by the vm_page module. + * + * If additional memory can be made usable after the VM system is initialized, + * it should be reported through this function. + */ +void vm_page_manage(struct vm_page *page); + +/* + * Return the page descriptor for the given physical address. + */ +struct vm_page * vm_page_lookup_pa(phys_addr_t pa); + +/* + * Allocate a block of 2^order physical pages. + * + * The selector is used to determine the segments from which allocation can + * be attempted. + */ +struct vm_page * vm_page_alloc_pa(unsigned int order, unsigned int selector, + unsigned short type); + +/* + * Release a block of 2^order physical pages. + */ +void vm_page_free_pa(struct vm_page *page, unsigned int order); + +/* + * Return the name of the given segment. + */ +const char * vm_page_seg_name(unsigned int seg_index); + +/* + * Display internal information about the module. + */ +void vm_page_info_all(void); + +/* + * Return the total amount of physical memory. + */ +phys_addr_t vm_page_mem_size(void); + #endif /* _VM_VM_PAGE_H_ */ diff --git a/vm/vm_resident.c b/vm/vm_resident.c index c70fa734..9fd64918 100644 --- a/vm/vm_resident.c +++ b/vm/vm_resident.c @@ -72,7 +72,7 @@ /* * These variables record the values returned by vm_page_bootstrap, * for debugging purposes. The implementation of pmap_steal_memory - * and pmap_startup here also uses them internally. + * here also uses them internally. */ vm_offset_t virtual_space_start; @@ -95,21 +95,6 @@ vm_page_bucket_t *vm_page_buckets; /* Array of buckets */ unsigned int vm_page_bucket_count = 0; /* How big is array? */ unsigned int vm_page_hash_mask; /* Mask for hash function */ -/* - * Resident page structures are initialized from - * a template (see vm_page_alloc). - * - * When adding a new field to the virtual memory - * object structure, be sure to add initialization - * (see vm_page_bootstrap). - */ -struct vm_page vm_page_template; - -/* - * Resident pages that represent real memory - * are allocated from a free list. - */ -vm_page_t vm_page_queue_free; vm_page_t vm_page_queue_fictitious; decl_simple_lock_data(,vm_page_queue_free_lock) unsigned int vm_page_free_wanted; @@ -192,48 +177,15 @@ void vm_page_bootstrap( vm_offset_t *startp, vm_offset_t *endp) { - vm_page_t m; int i; /* - * Initialize the vm_page template. - */ - - m = &vm_page_template; - m->object = VM_OBJECT_NULL; /* reset later */ - m->offset = 0; /* reset later */ - m->wire_count = 0; - - m->inactive = FALSE; - m->active = FALSE; - m->laundry = FALSE; - m->free = FALSE; - m->external = FALSE; - - m->busy = TRUE; - m->wanted = FALSE; - m->tabled = FALSE; - m->fictitious = FALSE; - m->private = FALSE; - m->absent = FALSE; - m->error = FALSE; - m->dirty = FALSE; - m->precious = FALSE; - m->reference = FALSE; - - m->phys_addr = 0; /* reset later */ - - m->page_lock = VM_PROT_NONE; - m->unlock_request = VM_PROT_NONE; - - /* * Initialize the page queues. */ simple_lock_init(&vm_page_queue_free_lock); simple_lock_init(&vm_page_queue_lock); - vm_page_queue_free = VM_PAGE_NULL; vm_page_queue_fictitious = VM_PAGE_NULL; queue_init(&vm_page_queue_active); queue_init(&vm_page_queue_inactive); @@ -280,15 +232,8 @@ void vm_page_bootstrap( simple_lock_init(&bucket->lock); } - /* - * Machine-dependent code allocates the resident page table. - * It uses vm_page_init to initialize the page frames. - * The code also returns to us the virtual space available - * to the kernel. We don't trust the pmap module - * to get the alignment right. - */ + vm_page_setup(); - pmap_startup(&virtual_space_start, &virtual_space_end); virtual_space_start = round_page(virtual_space_start); virtual_space_end = trunc_page(virtual_space_end); @@ -301,8 +246,8 @@ void vm_page_bootstrap( #ifndef MACHINE_PAGES /* - * We implement pmap_steal_memory and pmap_startup with the help - * of two simpler functions, pmap_virtual_space and pmap_next_page. + * We implement pmap_steal_memory with the help + * of two simpler functions, pmap_virtual_space and vm_page_bootalloc. */ vm_offset_t pmap_steal_memory( @@ -310,11 +255,7 @@ vm_offset_t pmap_steal_memory( { vm_offset_t addr, vaddr, paddr; - /* - * We round the size to an integer multiple. - */ - - size = (size + 3) &~ 3; + size = round_page(size); /* * If this is the first call to pmap_steal_memory, @@ -347,8 +288,7 @@ vm_offset_t pmap_steal_memory( for (vaddr = round_page(addr); vaddr < addr + size; vaddr += PAGE_SIZE) { - if (!pmap_next_page(&paddr)) - panic("pmap_steal_memory"); + paddr = vm_page_bootalloc(PAGE_SIZE); /* * XXX Logically, these mappings should be wired, @@ -361,64 +301,6 @@ vm_offset_t pmap_steal_memory( return addr; } - -void pmap_startup( - vm_offset_t *startp, - vm_offset_t *endp) -{ - unsigned int i, npages, pages_initialized; - vm_page_t pages; - vm_offset_t paddr; - - /* - * We calculate how many page frames we will have - * and then allocate the page structures in one chunk. - */ - - npages = ((PAGE_SIZE * pmap_free_pages() + - (round_page(virtual_space_start) - virtual_space_start)) / - (PAGE_SIZE + sizeof *pages)); - - pages = (vm_page_t) pmap_steal_memory(npages * sizeof *pages); - - /* - * Initialize the page frames. - */ - - for (i = 0, pages_initialized = 0; i < npages; i++) { - if (!pmap_next_page(&paddr)) - break; - - vm_page_init(&pages[i], paddr); - pages_initialized++; - } - i = 0; - while (pmap_next_page(&paddr)) - i++; - if (i) - printf("%u memory page(s) left away\n", i); - - /* - * Release pages in reverse order so that physical pages - * initially get allocated in ascending addresses. This keeps - * the devices (which must address physical memory) happy if - * they require several consecutive pages. - */ - - for (i = pages_initialized; i > 0; i--) { - vm_page_release(&pages[i - 1], FALSE); - } - - /* - * We have to re-align virtual_space_start, - * because pmap_steal_memory has been using it. - */ - - virtual_space_start = round_page(virtual_space_start); - - *startp = virtual_space_start; - *endp = virtual_space_end; -} #endif /* MACHINE_PAGES */ /* @@ -434,34 +316,6 @@ void vm_page_module_init(void) } /* - * Routine: vm_page_create - * Purpose: - * After the VM system is up, machine-dependent code - * may stumble across more physical memory. For example, - * memory that it was reserving for a frame buffer. - * vm_page_create turns this memory into available pages. - */ - -void vm_page_create( - vm_offset_t start, - vm_offset_t end) -{ - vm_offset_t paddr; - vm_page_t m; - - for (paddr = round_page(start); - paddr < trunc_page(end); - paddr += PAGE_SIZE) { - m = (vm_page_t) kmem_cache_alloc(&vm_page_cache); - if (m == VM_PAGE_NULL) - panic("vm_page_create"); - - vm_page_init(m, paddr); - vm_page_release(m, FALSE); - } -} - -/* * vm_page_hash: * * Distributes the object/offset key pair among hash buckets. @@ -750,6 +604,33 @@ void vm_page_rename( vm_page_unlock_queues(); } +static void vm_page_init_template(vm_page_t m) +{ + m->object = VM_OBJECT_NULL; /* reset later */ + m->offset = 0; /* reset later */ + m->wire_count = 0; + + m->inactive = FALSE; + m->active = FALSE; + m->laundry = FALSE; + m->free = FALSE; + m->external = FALSE; + + m->busy = TRUE; + m->wanted = FALSE; + m->tabled = FALSE; + m->fictitious = FALSE; + m->private = FALSE; + m->absent = FALSE; + m->error = FALSE; + m->dirty = FALSE; + m->precious = FALSE; + m->reference = FALSE; + + m->page_lock = VM_PROT_NONE; + m->unlock_request = VM_PROT_NONE; +} + /* * vm_page_init: * @@ -758,11 +639,9 @@ void vm_page_rename( * so that it can be given to vm_page_release or vm_page_insert. */ void vm_page_init( - vm_page_t mem, - vm_offset_t phys_addr) + vm_page_t mem) { - *mem = vm_page_template; - mem->phys_addr = phys_addr; + vm_page_init_template(mem); } /* @@ -794,7 +673,7 @@ vm_page_t vm_page_grab_fictitious(void) * Release a fictitious page to the free list. */ -void vm_page_release_fictitious( +static void vm_page_release_fictitious( vm_page_t m) { simple_lock(&vm_page_queue_free_lock); @@ -826,7 +705,8 @@ void vm_page_more_fictitious(void) if (m == VM_PAGE_NULL) panic("vm_page_more_fictitious"); - vm_page_init(m, vm_page_fictitious_addr); + vm_page_init(m); + m->phys_addr = vm_page_fictitious_addr; m->fictitious = TRUE; vm_page_release_fictitious(m); } @@ -836,25 +716,46 @@ void vm_page_more_fictitious(void) * vm_page_convert: * * Attempt to convert a fictitious page into a real page. + * + * The object referenced by *MP must be locked. */ boolean_t vm_page_convert( - vm_page_t m, + struct vm_page **mp, boolean_t external) { - vm_page_t real_m; + struct vm_page *real_m, *fict_m; + vm_object_t object; + vm_offset_t offset; + + fict_m = *mp; + + assert(fict_m->fictitious); + assert(fict_m->phys_addr == vm_page_fictitious_addr); + assert(!fict_m->active); + assert(!fict_m->inactive); real_m = vm_page_grab(external); if (real_m == VM_PAGE_NULL) return FALSE; - m->phys_addr = real_m->phys_addr; - m->fictitious = FALSE; + object = fict_m->object; + offset = fict_m->offset; + vm_page_remove(fict_m); + + memcpy(&real_m->vm_page_header, + &fict_m->vm_page_header, + sizeof(*fict_m) - VM_PAGE_HEADER_SIZE); + real_m->fictitious = FALSE; - real_m->phys_addr = vm_page_fictitious_addr; - real_m->fictitious = TRUE; + vm_page_insert(real_m, object, offset); - vm_page_release_fictitious(real_m); + assert(real_m->phys_addr != vm_page_fictitious_addr); + assert(fict_m->fictitious); + assert(fict_m->phys_addr == vm_page_fictitious_addr); + + vm_page_release_fictitious(fict_m); + *mp = real_m; return TRUE; } @@ -886,15 +787,16 @@ vm_page_t vm_page_grab( return VM_PAGE_NULL; } - if (vm_page_queue_free == VM_PAGE_NULL) + mem = vm_page_alloc_pa(0, VM_PAGE_SEL_DIRECTMAP, VM_PT_KERNEL); + + if (mem == NULL) panic("vm_page_grab"); if (--vm_page_free_count < vm_page_free_count_minimum) vm_page_free_count_minimum = vm_page_free_count; if (external) vm_page_external_count++; - mem = vm_page_queue_free; - vm_page_queue_free = (vm_page_t) mem->pageq.next; + mem->free = FALSE; mem->extcounted = mem->external = external; simple_unlock(&vm_page_queue_free_lock); @@ -928,208 +830,97 @@ vm_offset_t vm_page_grab_phys_addr(void) } /* - * vm_page_grab_contiguous_pages: - * - * Take N pages off the free list, the pages should - * cover a contiguous range of physical addresses. - * [Used by device drivers to cope with DMA limitations] + * vm_page_release: * - * Returns the page descriptors in ascending order, or - * Returns KERN_RESOURCE_SHORTAGE if it could not. + * Return a page to the free list. */ -/* Biggest phys page number for the pages we handle in VM */ - -vm_size_t vm_page_big_pagenum = 0; /* Set this before call! */ - -kern_return_t -vm_page_grab_contiguous_pages( - int npages, - vm_page_t pages[], - natural_t *bits, - boolean_t external) +static void vm_page_release( + vm_page_t mem, + boolean_t external) { - int first_set; - int size, alloc_size; - kern_return_t ret; - vm_page_t mem, *prevmemp; + simple_lock(&vm_page_queue_free_lock); + if (mem->free) + panic("vm_page_release"); + mem->free = TRUE; + vm_page_free_pa(mem, 0); + vm_page_free_count++; + if (external) + vm_page_external_count--; -#ifndef NBBY -#define NBBY 8 /* size in bits of sizeof()`s unity */ -#endif + /* + * Check if we should wake up someone waiting for page. + * But don't bother waking them unless they can allocate. + * + * We wakeup only one thread, to prevent starvation. + * Because the scheduling system handles wait queues FIFO, + * if we wakeup all waiting threads, one greedy thread + * can starve multiple niceguy threads. When the threads + * all wakeup, the greedy threads runs first, grabs the page, + * and waits for another page. It will be the first to run + * when the next page is freed. + * + * However, there is a slight danger here. + * The thread we wake might not use the free page. + * Then the other threads could wait indefinitely + * while the page goes unused. To forestall this, + * the pageout daemon will keep making free pages + * as long as vm_page_free_wanted is non-zero. + */ -#define NBPEL (sizeof(natural_t)*NBBY) + if ((vm_page_free_wanted > 0) && + (vm_page_free_count >= vm_page_free_reserved)) { + vm_page_free_wanted--; + thread_wakeup_one((event_t) &vm_page_free_count); + } - size = (vm_page_big_pagenum + NBPEL - 1) - & ~(NBPEL - 1); /* in bits */ + simple_unlock(&vm_page_queue_free_lock); +} - size = size / NBBY; /* in bytes */ +/* + * vm_page_grab_contig: + * + * Remove a block of contiguous pages from the free list. + * Returns VM_PAGE_NULL if the request fails. + */ - /* - * If we are called before the VM system is fully functional - * the invoker must provide us with the work space. [one bit - * per page starting at phys 0 and up to vm_page_big_pagenum] - */ - if (bits == 0) { - alloc_size = round_page(size); - if (kmem_alloc_wired(kernel_map, - (vm_offset_t *)&bits, - alloc_size) - != KERN_SUCCESS) - return KERN_RESOURCE_SHORTAGE; - } else - alloc_size = 0; +vm_page_t vm_page_grab_contig( + vm_size_t size, + unsigned int selector) +{ + unsigned int i, order, nr_pages; + vm_page_t mem; - memset(bits, 0, size); + order = vm_page_order(size); + nr_pages = 1 << order; - /* - * A very large granularity call, its rare so that is ok - */ simple_lock(&vm_page_queue_free_lock); /* - * Do not dip into the reserved pool. + * Only let privileged threads (involved in pageout) + * dip into the reserved pool or exceed the limit + * for externally-managed pages. */ - if ((vm_page_free_count < vm_page_free_reserved) - || (vm_page_external_count >= vm_page_external_limit)) { - printf_once("no more room for vm_page_grab_contiguous_pages"); + if (((vm_page_free_count - nr_pages) <= vm_page_free_reserved) + && !current_thread()->vm_privilege) { simple_unlock(&vm_page_queue_free_lock); - return KERN_RESOURCE_SHORTAGE; - } - - /* - * First pass through, build a big bit-array of - * the pages that are free. It is not going to - * be too large anyways, in 4k we can fit info - * for 32k pages. - */ - mem = vm_page_queue_free; - while (mem) { - int word_index, bit_index; - - bit_index = (mem->phys_addr >> PAGE_SHIFT); - word_index = bit_index / NBPEL; - bit_index = bit_index - (word_index * NBPEL); - bits[word_index] |= 1 << bit_index; - - mem = (vm_page_t) mem->pageq.next; + return VM_PAGE_NULL; } - /* - * Second loop. Scan the bit array for NPAGES - * contiguous bits. That gives us, if any, - * the range of pages we will be grabbing off - * the free list. - */ - { - int bits_so_far = 0, i; - - first_set = 0; - - for (i = 0; i < size; i += sizeof(natural_t)) { - - natural_t v = bits[i / sizeof(natural_t)]; - int bitpos; - - /* - * Bitscan this one word - */ - if (v) { - /* - * keep counting them beans ? - */ - bitpos = 0; - - if (bits_so_far) { -count_ones: - while (v & 1) { - bitpos++; - /* - * got enough beans ? - */ - if (++bits_so_far == npages) - goto found_em; - v >>= 1; - } - /* if we are being lucky, roll again */ - if (bitpos == NBPEL) - continue; - } - - /* - * search for beans here - */ - bits_so_far = 0; - while ((bitpos < NBPEL) && ((v & 1) == 0)) { - bitpos++; - v >>= 1; - } - if (v & 1) { - first_set = (i * NBBY) + bitpos; - goto count_ones; - } - } - /* - * No luck - */ - bits_so_far = 0; - } - } + mem = vm_page_alloc_pa(order, selector, VM_PT_KERNEL); - /* - * We could not find enough contiguous pages. - */ - simple_unlock(&vm_page_queue_free_lock); + if (mem == NULL) + panic("vm_page_grab_contig"); - printf_once("no contiguous room for vm_page_grab_contiguous_pages"); - ret = KERN_RESOURCE_SHORTAGE; - goto out; + vm_page_free_count -= nr_pages; - /* - * Final pass. Now we know which pages we want. - * Scan the list until we find them all, grab - * pages as we go. FIRST_SET tells us where - * in the bit-array our pages start. - */ -found_em: - vm_page_free_count -= npages; if (vm_page_free_count < vm_page_free_count_minimum) vm_page_free_count_minimum = vm_page_free_count; - if (external) - vm_page_external_count += npages; - { - vm_offset_t first_phys, last_phys; - - /* cache values for compare */ - first_phys = first_set << PAGE_SHIFT; - last_phys = first_phys + (npages << PAGE_SHIFT);/* not included */ - - /* running pointers */ - mem = vm_page_queue_free; - prevmemp = &vm_page_queue_free; - - while (mem) { - - vm_offset_t addr; - - addr = mem->phys_addr; - - if ((addr >= first_phys) && - (addr < last_phys)) { - *prevmemp = (vm_page_t) mem->pageq.next; - pages[(addr - first_phys) >> PAGE_SHIFT] = mem; - mem->free = FALSE; - mem->extcounted = mem->external = external; - /* - * Got them all ? - */ - if (--npages == 0) break; - } else - prevmemp = (vm_page_t *) &mem->pageq.next; - - mem = (vm_page_t) mem->pageq.next; - } + + for (i = 0; i < nr_pages; i++) { + mem[i].free = FALSE; + mem[i].extcounted = mem[i].external = 0; } simple_unlock(&vm_page_queue_free_lock); @@ -1148,55 +939,35 @@ found_em: if ((vm_page_free_count < vm_page_free_min) || ((vm_page_free_count < vm_page_free_target) && (vm_page_inactive_count < vm_page_inactive_target))) - thread_wakeup(&vm_page_free_wanted); - - ret = KERN_SUCCESS; -out: - if (alloc_size) - kmem_free(kernel_map, (vm_offset_t) bits, alloc_size); + thread_wakeup((event_t) &vm_page_free_wanted); - return ret; + return mem; } /* - * vm_page_release: + * vm_page_free_contig: * - * Return a page to the free list. + * Return a block of contiguous pages to the free list. */ -void vm_page_release( - vm_page_t mem, - boolean_t external) +void vm_page_free_contig(vm_page_t mem, vm_size_t size) { + unsigned int i, order, nr_pages; + + order = vm_page_order(size); + nr_pages = 1 << order; + simple_lock(&vm_page_queue_free_lock); - if (mem->free) - panic("vm_page_release"); - mem->free = TRUE; - mem->pageq.next = (queue_entry_t) vm_page_queue_free; - vm_page_queue_free = mem; - vm_page_free_count++; - if (external) - vm_page_external_count--; - /* - * Check if we should wake up someone waiting for page. - * But don't bother waking them unless they can allocate. - * - * We wakeup only one thread, to prevent starvation. - * Because the scheduling system handles wait queues FIFO, - * if we wakeup all waiting threads, one greedy thread - * can starve multiple niceguy threads. When the threads - * all wakeup, the greedy threads runs first, grabs the page, - * and waits for another page. It will be the first to run - * when the next page is freed. - * - * However, there is a slight danger here. - * The thread we wake might not use the free page. - * Then the other threads could wait indefinitely - * while the page goes unused. To forestall this, - * the pageout daemon will keep making free pages - * as long as vm_page_free_wanted is non-zero. - */ + for (i = 0; i < nr_pages; i++) { + if (mem[i].free) + panic("vm_page_free_contig"); + + mem[i].free = TRUE; + } + + vm_page_free_pa(mem, order); + vm_page_free_count += nr_pages; if ((vm_page_free_wanted > 0) && (vm_page_free_count >= vm_page_free_reserved)) { @@ -1310,12 +1081,13 @@ void vm_page_free( */ if (mem->private || mem->fictitious) { - vm_page_init(mem, vm_page_fictitious_addr); + vm_page_init(mem); + mem->phys_addr = vm_page_fictitious_addr; mem->fictitious = TRUE; vm_page_release_fictitious(mem); } else { int external = mem->external && mem->extcounted; - vm_page_init(mem, mem->phys_addr); + vm_page_init(mem); vm_page_release(mem, external); } } |