2 * linux/kernel/power/snapshot.c
4 * This file provides system snapshot/restore functionality for swsusp.
6 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
7 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
9 * This file is released under the GPLv2.
13 #include <linux/version.h>
14 #include <linux/module.h>
16 #include <linux/suspend.h>
17 #include <linux/delay.h>
18 #include <linux/bitops.h>
19 #include <linux/spinlock.h>
20 #include <linux/kernel.h>
22 #include <linux/device.h>
23 #include <linux/init.h>
24 #include <linux/bootmem.h>
25 #include <linux/syscalls.h>
26 #include <linux/console.h>
27 #include <linux/highmem.h>
28 #include <linux/list.h>
29 #include <linux/slab.h>
30 #include <linux/compiler.h>
31 #include <linux/ktime.h>
33 #include <asm/uaccess.h>
34 #include <asm/mmu_context.h>
35 #include <asm/pgtable.h>
36 #include <asm/tlbflush.h>
41 static int swsusp_page_is_free(struct page *);
42 static void swsusp_set_page_forbidden(struct page *);
43 static void swsusp_unset_page_forbidden(struct page *);
46 * Number of bytes to reserve for memory allocations made by device drivers
47 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
48 * cause image creation to fail (tunable via /sys/power/reserved_size).
50 unsigned long reserved_size;
52 void __init hibernate_reserved_size_init(void)
54 reserved_size = SPARE_PAGES * PAGE_SIZE;
58 * Preferred image size in bytes (tunable via /sys/power/image_size).
59 * When it is set to N, swsusp will do its best to ensure the image
60 * size will not exceed N bytes, but if that is impossible, it will
61 * try to create the smallest image possible.
63 unsigned long image_size;
65 void __init hibernate_image_size_init(void)
67 image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE;
70 /* List of PBEs needed for restoring the pages that were allocated before
71 * the suspend and included in the suspend image, but have also been
72 * allocated by the "resume" kernel, so their contents cannot be written
73 * directly to their "original" page frames.
75 struct pbe *restore_pblist;
77 /* struct linked_page is used to build chains of pages */
79 #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
82 struct linked_page *next;
83 char data[LINKED_PAGE_DATA_SIZE];
87 * List of "safe" pages (ie. pages that were not used by the image kernel
88 * before hibernation) that may be used as temporary storage for image kernel
91 static struct linked_page *safe_pages_list;
93 /* Pointer to an auxiliary buffer (1 page) */
97 * @safe_needed - on resume, for storing the PBE list and the image,
98 * we can only use memory pages that do not conflict with the pages
99 * used before suspend. The unsafe pages have PageNosaveFree set
100 * and we count them using unsafe_pages.
102 * Each allocated image page is marked as PageNosave and PageNosaveFree
103 * so that swsusp_free() can release it.
108 #define PG_UNSAFE_CLEAR 1
109 #define PG_UNSAFE_KEEP 0
111 static unsigned int allocated_unsafe_pages;
113 static void *get_image_page(gfp_t gfp_mask, int safe_needed)
117 res = (void *)get_zeroed_page(gfp_mask);
119 while (res && swsusp_page_is_free(virt_to_page(res))) {
120 /* The page is unsafe, mark it for swsusp_free() */
121 swsusp_set_page_forbidden(virt_to_page(res));
122 allocated_unsafe_pages++;
123 res = (void *)get_zeroed_page(gfp_mask);
126 swsusp_set_page_forbidden(virt_to_page(res));
127 swsusp_set_page_free(virt_to_page(res));
132 static void *__get_safe_page(gfp_t gfp_mask)
134 if (safe_pages_list) {
135 void *ret = safe_pages_list;
137 safe_pages_list = safe_pages_list->next;
138 memset(ret, 0, PAGE_SIZE);
141 return get_image_page(gfp_mask, PG_SAFE);
144 unsigned long get_safe_page(gfp_t gfp_mask)
146 return (unsigned long)__get_safe_page(gfp_mask);
149 static struct page *alloc_image_page(gfp_t gfp_mask)
153 page = alloc_page(gfp_mask);
155 swsusp_set_page_forbidden(page);
156 swsusp_set_page_free(page);
161 static void recycle_safe_page(void *page_address)
163 struct linked_page *lp = page_address;
165 lp->next = safe_pages_list;
166 safe_pages_list = lp;
170 * free_image_page - free page represented by @addr, allocated with
171 * get_image_page (page flags set by it must be cleared)
174 static inline void free_image_page(void *addr, int clear_nosave_free)
178 BUG_ON(!virt_addr_valid(addr));
180 page = virt_to_page(addr);
182 swsusp_unset_page_forbidden(page);
183 if (clear_nosave_free)
184 swsusp_unset_page_free(page);
190 free_list_of_pages(struct linked_page *list, int clear_page_nosave)
193 struct linked_page *lp = list->next;
195 free_image_page(list, clear_page_nosave);
201 * struct chain_allocator is used for allocating small objects out of
202 * a linked list of pages called 'the chain'.
204 * The chain grows each time when there is no room for a new object in
205 * the current page. The allocated objects cannot be freed individually.
206 * It is only possible to free them all at once, by freeing the entire
209 * NOTE: The chain allocator may be inefficient if the allocated objects
210 * are not much smaller than PAGE_SIZE.
213 struct chain_allocator {
214 struct linked_page *chain; /* the chain */
215 unsigned int used_space; /* total size of objects allocated out
216 * of the current page
218 gfp_t gfp_mask; /* mask for allocating pages */
219 int safe_needed; /* if set, only "safe" pages are allocated */
223 chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed)
226 ca->used_space = LINKED_PAGE_DATA_SIZE;
227 ca->gfp_mask = gfp_mask;
228 ca->safe_needed = safe_needed;
231 static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
235 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
236 struct linked_page *lp;
238 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
239 get_image_page(ca->gfp_mask, PG_ANY);
243 lp->next = ca->chain;
247 ret = ca->chain->data + ca->used_space;
248 ca->used_space += size;
253 * Data types related to memory bitmaps.
255 * Memory bitmap is a structure consiting of many linked lists of
256 * objects. The main list's elements are of type struct zone_bitmap
257 * and each of them corresonds to one zone. For each zone bitmap
258 * object there is a list of objects of type struct bm_block that
259 * represent each blocks of bitmap in which information is stored.
261 * struct memory_bitmap contains a pointer to the main list of zone
262 * bitmap objects, a struct bm_position used for browsing the bitmap,
263 * and a pointer to the list of pages used for allocating all of the
264 * zone bitmap objects and bitmap block objects.
266 * NOTE: It has to be possible to lay out the bitmap in memory
267 * using only allocations of order 0. Additionally, the bitmap is
268 * designed to work with arbitrary number of zones (this is over the
269 * top for now, but let's avoid making unnecessary assumptions ;-).
271 * struct zone_bitmap contains a pointer to a list of bitmap block
272 * objects and a pointer to the bitmap block object that has been
273 * most recently used for setting bits. Additionally, it contains the
274 * pfns that correspond to the start and end of the represented zone.
276 * struct bm_block contains a pointer to the memory page in which
277 * information is stored (in the form of a block of bitmap)
278 * It also contains the pfns that correspond to the start and end of
279 * the represented memory area.
281 * The memory bitmap is organized as a radix tree to guarantee fast random
282 * access to the bits. There is one radix tree for each zone (as returned
283 * from create_mem_extents).
285 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
286 * two linked lists for the nodes of the tree, one for the inner nodes and
287 * one for the leave nodes. The linked leave nodes are used for fast linear
288 * access of the memory bitmap.
290 * The struct rtree_node represents one node of the radix tree.
293 #define BM_END_OF_MAP (~0UL)
295 #define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
296 #define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
297 #define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
300 * struct rtree_node is a wrapper struct to link the nodes
301 * of the rtree together for easy linear iteration over
302 * bits and easy freeing
305 struct list_head list;
310 * struct mem_zone_bm_rtree represents a bitmap used for one
311 * populated memory zone.
313 struct mem_zone_bm_rtree {
314 struct list_head list; /* Link Zones together */
315 struct list_head nodes; /* Radix Tree inner nodes */
316 struct list_head leaves; /* Radix Tree leaves */
317 unsigned long start_pfn; /* Zone start page frame */
318 unsigned long end_pfn; /* Zone end page frame + 1 */
319 struct rtree_node *rtree; /* Radix Tree Root */
320 int levels; /* Number of Radix Tree Levels */
321 unsigned int blocks; /* Number of Bitmap Blocks */
324 /* strcut bm_position is used for browsing memory bitmaps */
327 struct mem_zone_bm_rtree *zone;
328 struct rtree_node *node;
329 unsigned long node_pfn;
333 struct memory_bitmap {
334 struct list_head zones;
335 struct linked_page *p_list; /* list of pages used to store zone
336 * bitmap objects and bitmap block
339 struct bm_position cur; /* most recently used bit position */
342 /* Functions that operate on memory bitmaps */
344 #define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
345 #if BITS_PER_LONG == 32
346 #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
348 #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
350 #define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
353 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
355 * This function is used to allocate inner nodes as well as the
356 * leave nodes of the radix tree. It also adds the node to the
357 * corresponding linked list passed in by the *list parameter.
359 static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
360 struct chain_allocator *ca,
361 struct list_head *list)
363 struct rtree_node *node;
365 node = chain_alloc(ca, sizeof(struct rtree_node));
369 node->data = get_image_page(gfp_mask, safe_needed);
373 list_add_tail(&node->list, list);
379 * add_rtree_block - Add a new leave node to the radix tree
381 * The leave nodes need to be allocated in order to keep the leaves
382 * linked list in order. This is guaranteed by the zone->blocks
385 static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
386 int safe_needed, struct chain_allocator *ca)
388 struct rtree_node *node, *block, **dst;
389 unsigned int levels_needed, block_nr;
392 block_nr = zone->blocks;
395 /* How many levels do we need for this block nr? */
398 block_nr >>= BM_RTREE_LEVEL_SHIFT;
401 /* Make sure the rtree has enough levels */
402 for (i = zone->levels; i < levels_needed; i++) {
403 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
408 node->data[0] = (unsigned long)zone->rtree;
413 /* Allocate new block */
414 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
418 /* Now walk the rtree to insert the block */
421 block_nr = zone->blocks;
422 for (i = zone->levels; i > 0; i--) {
426 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
433 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
434 index &= BM_RTREE_LEVEL_MASK;
435 dst = (struct rtree_node **)&((*dst)->data[index]);
445 static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
446 int clear_nosave_free);
449 * create_zone_bm_rtree - create a radix tree for one zone
451 * Allocated the mem_zone_bm_rtree structure and initializes it.
452 * This function also allocated and builds the radix tree for the
455 static struct mem_zone_bm_rtree *
456 create_zone_bm_rtree(gfp_t gfp_mask, int safe_needed,
457 struct chain_allocator *ca,
458 unsigned long start, unsigned long end)
460 struct mem_zone_bm_rtree *zone;
461 unsigned int i, nr_blocks;
465 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
469 INIT_LIST_HEAD(&zone->nodes);
470 INIT_LIST_HEAD(&zone->leaves);
471 zone->start_pfn = start;
473 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
475 for (i = 0; i < nr_blocks; i++) {
476 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
477 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
486 * free_zone_bm_rtree - Free the memory of the radix tree
488 * Free all node pages of the radix tree. The mem_zone_bm_rtree
489 * structure itself is not freed here nor are the rtree_node
492 static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
493 int clear_nosave_free)
495 struct rtree_node *node;
497 list_for_each_entry(node, &zone->nodes, list)
498 free_image_page(node->data, clear_nosave_free);
500 list_for_each_entry(node, &zone->leaves, list)
501 free_image_page(node->data, clear_nosave_free);
504 static void memory_bm_position_reset(struct memory_bitmap *bm)
506 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
508 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
509 struct rtree_node, list);
510 bm->cur.node_pfn = 0;
511 bm->cur.node_bit = 0;
514 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
517 struct list_head hook;
523 * free_mem_extents - free a list of memory extents
524 * @list - list of extents to empty
526 static void free_mem_extents(struct list_head *list)
528 struct mem_extent *ext, *aux;
530 list_for_each_entry_safe(ext, aux, list, hook) {
531 list_del(&ext->hook);
537 * create_mem_extents - create a list of memory extents representing
538 * contiguous ranges of PFNs
539 * @list - list to put the extents into
540 * @gfp_mask - mask to use for memory allocations
542 static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
546 INIT_LIST_HEAD(list);
548 for_each_populated_zone(zone) {
549 unsigned long zone_start, zone_end;
550 struct mem_extent *ext, *cur, *aux;
552 zone_start = zone->zone_start_pfn;
553 zone_end = zone_end_pfn(zone);
555 list_for_each_entry(ext, list, hook)
556 if (zone_start <= ext->end)
559 if (&ext->hook == list || zone_end < ext->start) {
560 /* New extent is necessary */
561 struct mem_extent *new_ext;
563 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
565 free_mem_extents(list);
568 new_ext->start = zone_start;
569 new_ext->end = zone_end;
570 list_add_tail(&new_ext->hook, &ext->hook);
574 /* Merge this zone's range of PFNs with the existing one */
575 if (zone_start < ext->start)
576 ext->start = zone_start;
577 if (zone_end > ext->end)
580 /* More merging may be possible */
582 list_for_each_entry_safe_continue(cur, aux, list, hook) {
583 if (zone_end < cur->start)
585 if (zone_end < cur->end)
587 list_del(&cur->hook);
596 * memory_bm_create - allocate memory for a memory bitmap
599 memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed)
601 struct chain_allocator ca;
602 struct list_head mem_extents;
603 struct mem_extent *ext;
606 chain_init(&ca, gfp_mask, safe_needed);
607 INIT_LIST_HEAD(&bm->zones);
609 error = create_mem_extents(&mem_extents, gfp_mask);
613 list_for_each_entry(ext, &mem_extents, hook) {
614 struct mem_zone_bm_rtree *zone;
616 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
617 ext->start, ext->end);
622 list_add_tail(&zone->list, &bm->zones);
625 bm->p_list = ca.chain;
626 memory_bm_position_reset(bm);
628 free_mem_extents(&mem_extents);
632 bm->p_list = ca.chain;
633 memory_bm_free(bm, PG_UNSAFE_CLEAR);
638 * memory_bm_free - free memory occupied by the memory bitmap @bm
640 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
642 struct mem_zone_bm_rtree *zone;
644 list_for_each_entry(zone, &bm->zones, list)
645 free_zone_bm_rtree(zone, clear_nosave_free);
647 free_list_of_pages(bm->p_list, clear_nosave_free);
649 INIT_LIST_HEAD(&bm->zones);
653 * memory_bm_find_bit - Find the bit for pfn in the memory
656 * Find the bit in the bitmap @bm that corresponds to given pfn.
657 * The cur.zone, cur.block and cur.node_pfn member of @bm are
659 * It walks the radix tree to find the page which contains the bit for
660 * pfn and returns the bit position in **addr and *bit_nr.
662 static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
663 void **addr, unsigned int *bit_nr)
665 struct mem_zone_bm_rtree *curr, *zone;
666 struct rtree_node *node;
671 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
676 /* Find the right zone */
677 list_for_each_entry(curr, &bm->zones, list) {
678 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
689 * We have a zone. Now walk the radix tree to find the leave
694 if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
698 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
700 for (i = zone->levels; i > 0; i--) {
703 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
704 index &= BM_RTREE_LEVEL_MASK;
705 BUG_ON(node->data[index] == 0);
706 node = (struct rtree_node *)node->data[index];
710 /* Update last position */
713 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
715 /* Set return values */
717 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
722 static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
728 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
733 static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
739 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
746 static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
752 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
754 clear_bit(bit, addr);
757 static void memory_bm_clear_current(struct memory_bitmap *bm)
761 bit = max(bm->cur.node_bit - 1, 0);
762 clear_bit(bit, bm->cur.node->data);
765 static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
771 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
773 return test_bit(bit, addr);
776 static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
781 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
785 * rtree_next_node - Jumps to the next leave node
787 * Sets the position to the beginning of the next node in the
788 * memory bitmap. This is either the next node in the current
789 * zone's radix tree or the first node in the radix tree of the
792 * Returns true if there is a next node, false otherwise.
794 static bool rtree_next_node(struct memory_bitmap *bm)
796 bm->cur.node = list_entry(bm->cur.node->list.next,
797 struct rtree_node, list);
798 if (&bm->cur.node->list != &bm->cur.zone->leaves) {
799 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
800 bm->cur.node_bit = 0;
801 touch_softlockup_watchdog();
805 /* No more nodes, goto next zone */
806 bm->cur.zone = list_entry(bm->cur.zone->list.next,
807 struct mem_zone_bm_rtree, list);
808 if (&bm->cur.zone->list != &bm->zones) {
809 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
810 struct rtree_node, list);
811 bm->cur.node_pfn = 0;
812 bm->cur.node_bit = 0;
821 * memory_bm_rtree_next_pfn - Find the next set bit in the bitmap @bm
823 * Starting from the last returned position this function searches
824 * for the next set bit in the memory bitmap and returns its
825 * number. If no more bit is set BM_END_OF_MAP is returned.
827 * It is required to run memory_bm_position_reset() before the
828 * first call to this function.
830 static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
832 unsigned long bits, pfn, pages;
836 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
837 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
838 bit = find_next_bit(bm->cur.node->data, bits,
841 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
842 bm->cur.node_bit = bit + 1;
845 } while (rtree_next_node(bm));
847 return BM_END_OF_MAP;
851 * This structure represents a range of page frames the contents of which
852 * should not be saved during the suspend.
855 struct nosave_region {
856 struct list_head list;
857 unsigned long start_pfn;
858 unsigned long end_pfn;
861 static LIST_HEAD(nosave_regions);
863 static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
865 struct rtree_node *node;
867 list_for_each_entry(node, &zone->nodes, list)
868 recycle_safe_page(node->data);
870 list_for_each_entry(node, &zone->leaves, list)
871 recycle_safe_page(node->data);
874 static void memory_bm_recycle(struct memory_bitmap *bm)
876 struct mem_zone_bm_rtree *zone;
877 struct linked_page *p_list;
879 list_for_each_entry(zone, &bm->zones, list)
880 recycle_zone_bm_rtree(zone);
884 struct linked_page *lp = p_list;
887 recycle_safe_page(lp);
892 * register_nosave_region - register a range of page frames the contents
893 * of which should not be saved during the suspend (to be used in the early
894 * initialization code)
898 __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn,
901 struct nosave_region *region;
903 if (start_pfn >= end_pfn)
906 if (!list_empty(&nosave_regions)) {
907 /* Try to extend the previous region (they should be sorted) */
908 region = list_entry(nosave_regions.prev,
909 struct nosave_region, list);
910 if (region->end_pfn == start_pfn) {
911 region->end_pfn = end_pfn;
916 /* during init, this shouldn't fail */
917 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
920 /* This allocation cannot fail */
921 region = memblock_virt_alloc(sizeof(struct nosave_region), 0);
922 region->start_pfn = start_pfn;
923 region->end_pfn = end_pfn;
924 list_add_tail(®ion->list, &nosave_regions);
926 printk(KERN_INFO "PM: Registered nosave memory: [mem %#010llx-%#010llx]\n",
927 (unsigned long long) start_pfn << PAGE_SHIFT,
928 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
932 * Set bits in this map correspond to the page frames the contents of which
933 * should not be saved during the suspend.
935 static struct memory_bitmap *forbidden_pages_map;
937 /* Set bits in this map correspond to free page frames. */
938 static struct memory_bitmap *free_pages_map;
941 * Each page frame allocated for creating the image is marked by setting the
942 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
945 void swsusp_set_page_free(struct page *page)
948 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
951 static int swsusp_page_is_free(struct page *page)
953 return free_pages_map ?
954 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
957 void swsusp_unset_page_free(struct page *page)
960 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
963 static void swsusp_set_page_forbidden(struct page *page)
965 if (forbidden_pages_map)
966 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
969 int swsusp_page_is_forbidden(struct page *page)
971 return forbidden_pages_map ?
972 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
975 static void swsusp_unset_page_forbidden(struct page *page)
977 if (forbidden_pages_map)
978 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
982 * mark_nosave_pages - set bits corresponding to the page frames the
983 * contents of which should not be saved in a given bitmap.
986 static void mark_nosave_pages(struct memory_bitmap *bm)
988 struct nosave_region *region;
990 if (list_empty(&nosave_regions))
993 list_for_each_entry(region, &nosave_regions, list) {
996 pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n",
997 (unsigned long long) region->start_pfn << PAGE_SHIFT,
998 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1001 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1002 if (pfn_valid(pfn)) {
1004 * It is safe to ignore the result of
1005 * mem_bm_set_bit_check() here, since we won't
1006 * touch the PFNs for which the error is
1009 mem_bm_set_bit_check(bm, pfn);
1015 * create_basic_memory_bitmaps - create bitmaps needed for marking page
1016 * frames that should not be saved and free page frames. The pointers
1017 * forbidden_pages_map and free_pages_map are only modified if everything
1018 * goes well, because we don't want the bits to be used before both bitmaps
1022 int create_basic_memory_bitmaps(void)
1024 struct memory_bitmap *bm1, *bm2;
1027 if (forbidden_pages_map && free_pages_map)
1030 BUG_ON(forbidden_pages_map || free_pages_map);
1032 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1036 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1038 goto Free_first_object;
1040 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1042 goto Free_first_bitmap;
1044 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1046 goto Free_second_object;
1048 forbidden_pages_map = bm1;
1049 free_pages_map = bm2;
1050 mark_nosave_pages(forbidden_pages_map);
1052 pr_debug("PM: Basic memory bitmaps created\n");
1059 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1066 * free_basic_memory_bitmaps - free memory bitmaps allocated by
1067 * create_basic_memory_bitmaps(). The auxiliary pointers are necessary
1068 * so that the bitmaps themselves are not referred to while they are being
1072 void free_basic_memory_bitmaps(void)
1074 struct memory_bitmap *bm1, *bm2;
1076 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1079 bm1 = forbidden_pages_map;
1080 bm2 = free_pages_map;
1081 forbidden_pages_map = NULL;
1082 free_pages_map = NULL;
1083 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1085 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1088 pr_debug("PM: Basic memory bitmaps freed\n");
1092 * snapshot_additional_pages - estimate the number of additional pages
1093 * be needed for setting up the suspend image data structures for given
1094 * zone (usually the returned value is greater than the exact number)
1097 unsigned int snapshot_additional_pages(struct zone *zone)
1099 unsigned int rtree, nodes;
1101 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1102 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1103 LINKED_PAGE_DATA_SIZE);
1105 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1112 #ifdef CONFIG_HIGHMEM
1114 * count_free_highmem_pages - compute the total number of free highmem
1115 * pages, system-wide.
1118 static unsigned int count_free_highmem_pages(void)
1121 unsigned int cnt = 0;
1123 for_each_populated_zone(zone)
1124 if (is_highmem(zone))
1125 cnt += zone_page_state(zone, NR_FREE_PAGES);
1131 * saveable_highmem_page - Determine whether a highmem page should be
1132 * included in the suspend image.
1134 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1135 * and it isn't a part of a free chunk of pages.
1137 static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1141 if (!pfn_valid(pfn))
1144 page = pfn_to_page(pfn);
1145 if (page_zone(page) != zone)
1148 BUG_ON(!PageHighMem(page));
1150 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) ||
1154 if (page_is_guard(page))
1161 * count_highmem_pages - compute the total number of saveable highmem
1165 static unsigned int count_highmem_pages(void)
1170 for_each_populated_zone(zone) {
1171 unsigned long pfn, max_zone_pfn;
1173 if (!is_highmem(zone))
1176 mark_free_pages(zone);
1177 max_zone_pfn = zone_end_pfn(zone);
1178 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1179 if (saveable_highmem_page(zone, pfn))
1185 static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1189 #endif /* CONFIG_HIGHMEM */
1192 * saveable_page - Determine whether a non-highmem page should be included
1193 * in the suspend image.
1195 * We should save the page if it isn't Nosave, and is not in the range
1196 * of pages statically defined as 'unsaveable', and it isn't a part of
1197 * a free chunk of pages.
1199 static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1203 if (!pfn_valid(pfn))
1206 page = pfn_to_page(pfn);
1207 if (page_zone(page) != zone)
1210 BUG_ON(PageHighMem(page));
1212 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1215 if (PageReserved(page)
1216 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1219 if (page_is_guard(page))
1226 * count_data_pages - compute the total number of saveable non-highmem
1230 static unsigned int count_data_pages(void)
1233 unsigned long pfn, max_zone_pfn;
1236 for_each_populated_zone(zone) {
1237 if (is_highmem(zone))
1240 mark_free_pages(zone);
1241 max_zone_pfn = zone_end_pfn(zone);
1242 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1243 if (saveable_page(zone, pfn))
1249 /* This is needed, because copy_page and memcpy are not usable for copying
1252 static inline void do_copy_page(long *dst, long *src)
1256 for (n = PAGE_SIZE / sizeof(long); n; n--)
1262 * safe_copy_page - check if the page we are going to copy is marked as
1263 * present in the kernel page tables (this always is the case if
1264 * CONFIG_DEBUG_PAGEALLOC is not set and in that case
1265 * kernel_page_present() always returns 'true').
1267 static void safe_copy_page(void *dst, struct page *s_page)
1269 if (kernel_page_present(s_page)) {
1270 do_copy_page(dst, page_address(s_page));
1272 kernel_map_pages(s_page, 1, 1);
1273 do_copy_page(dst, page_address(s_page));
1274 kernel_map_pages(s_page, 1, 0);
1279 #ifdef CONFIG_HIGHMEM
1280 static inline struct page *
1281 page_is_saveable(struct zone *zone, unsigned long pfn)
1283 return is_highmem(zone) ?
1284 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1287 static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1289 struct page *s_page, *d_page;
1292 s_page = pfn_to_page(src_pfn);
1293 d_page = pfn_to_page(dst_pfn);
1294 if (PageHighMem(s_page)) {
1295 src = kmap_atomic(s_page);
1296 dst = kmap_atomic(d_page);
1297 do_copy_page(dst, src);
1301 if (PageHighMem(d_page)) {
1302 /* Page pointed to by src may contain some kernel
1303 * data modified by kmap_atomic()
1305 safe_copy_page(buffer, s_page);
1306 dst = kmap_atomic(d_page);
1307 copy_page(dst, buffer);
1310 safe_copy_page(page_address(d_page), s_page);
1315 #define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1317 static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1319 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1320 pfn_to_page(src_pfn));
1322 #endif /* CONFIG_HIGHMEM */
1325 copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm)
1330 for_each_populated_zone(zone) {
1331 unsigned long max_zone_pfn;
1333 mark_free_pages(zone);
1334 max_zone_pfn = zone_end_pfn(zone);
1335 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1336 if (page_is_saveable(zone, pfn))
1337 memory_bm_set_bit(orig_bm, pfn);
1339 memory_bm_position_reset(orig_bm);
1340 memory_bm_position_reset(copy_bm);
1342 pfn = memory_bm_next_pfn(orig_bm);
1343 if (unlikely(pfn == BM_END_OF_MAP))
1345 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1349 /* Total number of image pages */
1350 static unsigned int nr_copy_pages;
1351 /* Number of pages needed for saving the original pfns of the image pages */
1352 static unsigned int nr_meta_pages;
1354 * Numbers of normal and highmem page frames allocated for hibernation image
1355 * before suspending devices.
1357 unsigned int alloc_normal, alloc_highmem;
1359 * Memory bitmap used for marking saveable pages (during hibernation) or
1360 * hibernation image pages (during restore)
1362 static struct memory_bitmap orig_bm;
1364 * Memory bitmap used during hibernation for marking allocated page frames that
1365 * will contain copies of saveable pages. During restore it is initially used
1366 * for marking hibernation image pages, but then the set bits from it are
1367 * duplicated in @orig_bm and it is released. On highmem systems it is next
1368 * used for marking "safe" highmem pages, but it has to be reinitialized for
1371 static struct memory_bitmap copy_bm;
1374 * swsusp_free - free pages allocated for the suspend.
1376 * Suspend pages are alocated before the atomic copy is made, so we
1377 * need to release them after the resume.
1380 void swsusp_free(void)
1382 unsigned long fb_pfn, fr_pfn;
1384 if (!forbidden_pages_map || !free_pages_map)
1387 memory_bm_position_reset(forbidden_pages_map);
1388 memory_bm_position_reset(free_pages_map);
1391 fr_pfn = memory_bm_next_pfn(free_pages_map);
1392 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1395 * Find the next bit set in both bitmaps. This is guaranteed to
1396 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1399 if (fb_pfn < fr_pfn)
1400 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1401 if (fr_pfn < fb_pfn)
1402 fr_pfn = memory_bm_next_pfn(free_pages_map);
1403 } while (fb_pfn != fr_pfn);
1405 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1406 struct page *page = pfn_to_page(fr_pfn);
1408 memory_bm_clear_current(forbidden_pages_map);
1409 memory_bm_clear_current(free_pages_map);
1417 restore_pblist = NULL;
1423 /* Helper functions used for the shrinking of memory. */
1425 #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1428 * preallocate_image_pages - Allocate a number of pages for hibernation image
1429 * @nr_pages: Number of page frames to allocate.
1430 * @mask: GFP flags to use for the allocation.
1432 * Return value: Number of page frames actually allocated
1434 static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1436 unsigned long nr_alloc = 0;
1438 while (nr_pages > 0) {
1441 page = alloc_image_page(mask);
1444 memory_bm_set_bit(©_bm, page_to_pfn(page));
1445 if (PageHighMem(page))
1456 static unsigned long preallocate_image_memory(unsigned long nr_pages,
1457 unsigned long avail_normal)
1459 unsigned long alloc;
1461 if (avail_normal <= alloc_normal)
1464 alloc = avail_normal - alloc_normal;
1465 if (nr_pages < alloc)
1468 return preallocate_image_pages(alloc, GFP_IMAGE);
1471 #ifdef CONFIG_HIGHMEM
1472 static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1474 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1478 * __fraction - Compute (an approximation of) x * (multiplier / base)
1480 static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1484 return (unsigned long)x;
1487 static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1488 unsigned long highmem,
1489 unsigned long total)
1491 unsigned long alloc = __fraction(nr_pages, highmem, total);
1493 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1495 #else /* CONFIG_HIGHMEM */
1496 static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1501 static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1502 unsigned long highmem,
1503 unsigned long total)
1507 #endif /* CONFIG_HIGHMEM */
1510 * free_unnecessary_pages - Release preallocated pages not needed for the image
1512 static unsigned long free_unnecessary_pages(void)
1514 unsigned long save, to_free_normal, to_free_highmem, free;
1516 save = count_data_pages();
1517 if (alloc_normal >= save) {
1518 to_free_normal = alloc_normal - save;
1522 save -= alloc_normal;
1524 save += count_highmem_pages();
1525 if (alloc_highmem >= save) {
1526 to_free_highmem = alloc_highmem - save;
1528 to_free_highmem = 0;
1529 save -= alloc_highmem;
1530 if (to_free_normal > save)
1531 to_free_normal -= save;
1535 free = to_free_normal + to_free_highmem;
1537 memory_bm_position_reset(©_bm);
1539 while (to_free_normal > 0 || to_free_highmem > 0) {
1540 unsigned long pfn = memory_bm_next_pfn(©_bm);
1541 struct page *page = pfn_to_page(pfn);
1543 if (PageHighMem(page)) {
1544 if (!to_free_highmem)
1549 if (!to_free_normal)
1554 memory_bm_clear_bit(©_bm, pfn);
1555 swsusp_unset_page_forbidden(page);
1556 swsusp_unset_page_free(page);
1564 * minimum_image_size - Estimate the minimum acceptable size of an image
1565 * @saveable: Number of saveable pages in the system.
1567 * We want to avoid attempting to free too much memory too hard, so estimate the
1568 * minimum acceptable size of a hibernation image to use as the lower limit for
1569 * preallocating memory.
1571 * We assume that the minimum image size should be proportional to
1573 * [number of saveable pages] - [number of pages that can be freed in theory]
1575 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1576 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages,
1577 * minus mapped file pages.
1579 static unsigned long minimum_image_size(unsigned long saveable)
1583 size = global_page_state(NR_SLAB_RECLAIMABLE)
1584 + global_page_state(NR_ACTIVE_ANON)
1585 + global_page_state(NR_INACTIVE_ANON)
1586 + global_page_state(NR_ACTIVE_FILE)
1587 + global_page_state(NR_INACTIVE_FILE)
1588 - global_page_state(NR_FILE_MAPPED);
1590 return saveable <= size ? 0 : saveable - size;
1594 * hibernate_preallocate_memory - Preallocate memory for hibernation image
1596 * To create a hibernation image it is necessary to make a copy of every page
1597 * frame in use. We also need a number of page frames to be free during
1598 * hibernation for allocations made while saving the image and for device
1599 * drivers, in case they need to allocate memory from their hibernation
1600 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1601 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1602 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1603 * total number of available page frames and allocate at least
1605 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1606 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1608 * of them, which corresponds to the maximum size of a hibernation image.
1610 * If image_size is set below the number following from the above formula,
1611 * the preallocation of memory is continued until the total number of saveable
1612 * pages in the system is below the requested image size or the minimum
1613 * acceptable image size returned by minimum_image_size(), whichever is greater.
1615 int hibernate_preallocate_memory(void)
1618 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1619 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1620 ktime_t start, stop;
1623 printk(KERN_INFO "PM: Preallocating image memory... ");
1624 start = ktime_get();
1626 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1630 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1637 /* Count the number of saveable data pages. */
1638 save_highmem = count_highmem_pages();
1639 saveable = count_data_pages();
1642 * Compute the total number of page frames we can use (count) and the
1643 * number of pages needed for image metadata (size).
1646 saveable += save_highmem;
1647 highmem = save_highmem;
1649 for_each_populated_zone(zone) {
1650 size += snapshot_additional_pages(zone);
1651 if (is_highmem(zone))
1652 highmem += zone_page_state(zone, NR_FREE_PAGES);
1654 count += zone_page_state(zone, NR_FREE_PAGES);
1656 avail_normal = count;
1658 count -= totalreserve_pages;
1660 /* Add number of pages required for page keys (s390 only). */
1661 size += page_key_additional_pages(saveable);
1663 /* Compute the maximum number of saveable pages to leave in memory. */
1664 max_size = (count - (size + PAGES_FOR_IO)) / 2
1665 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1666 /* Compute the desired number of image pages specified by image_size. */
1667 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1668 if (size > max_size)
1671 * If the desired number of image pages is at least as large as the
1672 * current number of saveable pages in memory, allocate page frames for
1673 * the image and we're done.
1675 if (size >= saveable) {
1676 pages = preallocate_image_highmem(save_highmem);
1677 pages += preallocate_image_memory(saveable - pages, avail_normal);
1681 /* Estimate the minimum size of the image. */
1682 pages = minimum_image_size(saveable);
1684 * To avoid excessive pressure on the normal zone, leave room in it to
1685 * accommodate an image of the minimum size (unless it's already too
1686 * small, in which case don't preallocate pages from it at all).
1688 if (avail_normal > pages)
1689 avail_normal -= pages;
1693 size = min_t(unsigned long, pages, max_size);
1696 * Let the memory management subsystem know that we're going to need a
1697 * large number of page frames to allocate and make it free some memory.
1698 * NOTE: If this is not done, performance will be hurt badly in some
1701 shrink_all_memory(saveable - size);
1704 * The number of saveable pages in memory was too high, so apply some
1705 * pressure to decrease it. First, make room for the largest possible
1706 * image and fail if that doesn't work. Next, try to decrease the size
1707 * of the image as much as indicated by 'size' using allocations from
1708 * highmem and non-highmem zones separately.
1710 pages_highmem = preallocate_image_highmem(highmem / 2);
1711 alloc = count - max_size;
1712 if (alloc > pages_highmem)
1713 alloc -= pages_highmem;
1716 pages = preallocate_image_memory(alloc, avail_normal);
1717 if (pages < alloc) {
1718 /* We have exhausted non-highmem pages, try highmem. */
1720 pages += pages_highmem;
1721 pages_highmem = preallocate_image_highmem(alloc);
1722 if (pages_highmem < alloc)
1724 pages += pages_highmem;
1726 * size is the desired number of saveable pages to leave in
1727 * memory, so try to preallocate (all memory - size) pages.
1729 alloc = (count - pages) - size;
1730 pages += preallocate_image_highmem(alloc);
1733 * There are approximately max_size saveable pages at this point
1734 * and we want to reduce this number down to size.
1736 alloc = max_size - size;
1737 size = preallocate_highmem_fraction(alloc, highmem, count);
1738 pages_highmem += size;
1740 size = preallocate_image_memory(alloc, avail_normal);
1741 pages_highmem += preallocate_image_highmem(alloc - size);
1742 pages += pages_highmem + size;
1746 * We only need as many page frames for the image as there are saveable
1747 * pages in memory, but we have allocated more. Release the excessive
1750 pages -= free_unnecessary_pages();
1754 printk(KERN_CONT "done (allocated %lu pages)\n", pages);
1755 swsusp_show_speed(start, stop, pages, "Allocated");
1760 printk(KERN_CONT "\n");
1765 #ifdef CONFIG_HIGHMEM
1767 * count_pages_for_highmem - compute the number of non-highmem pages
1768 * that will be necessary for creating copies of highmem pages.
1771 static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1773 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1775 if (free_highmem >= nr_highmem)
1778 nr_highmem -= free_highmem;
1784 count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1785 #endif /* CONFIG_HIGHMEM */
1788 * enough_free_mem - Make sure we have enough free memory for the
1792 static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1795 unsigned int free = alloc_normal;
1797 for_each_populated_zone(zone)
1798 if (!is_highmem(zone))
1799 free += zone_page_state(zone, NR_FREE_PAGES);
1801 nr_pages += count_pages_for_highmem(nr_highmem);
1802 pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n",
1803 nr_pages, PAGES_FOR_IO, free);
1805 return free > nr_pages + PAGES_FOR_IO;
1808 #ifdef CONFIG_HIGHMEM
1810 * get_highmem_buffer - if there are some highmem pages in the suspend
1811 * image, we may need the buffer to copy them and/or load their data.
1814 static inline int get_highmem_buffer(int safe_needed)
1816 buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
1817 return buffer ? 0 : -ENOMEM;
1821 * alloc_highmem_image_pages - allocate some highmem pages for the image.
1822 * Try to allocate as many pages as needed, but if the number of free
1823 * highmem pages is lesser than that, allocate them all.
1826 static inline unsigned int
1827 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int nr_highmem)
1829 unsigned int to_alloc = count_free_highmem_pages();
1831 if (to_alloc > nr_highmem)
1832 to_alloc = nr_highmem;
1834 nr_highmem -= to_alloc;
1835 while (to_alloc-- > 0) {
1838 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1839 memory_bm_set_bit(bm, page_to_pfn(page));
1844 static inline int get_highmem_buffer(int safe_needed) { return 0; }
1846 static inline unsigned int
1847 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int n) { return 0; }
1848 #endif /* CONFIG_HIGHMEM */
1851 * swsusp_alloc - allocate memory for the suspend image
1853 * We first try to allocate as many highmem pages as there are
1854 * saveable highmem pages in the system. If that fails, we allocate
1855 * non-highmem pages for the copies of the remaining highmem ones.
1857 * In this approach it is likely that the copies of highmem pages will
1858 * also be located in the high memory, because of the way in which
1859 * copy_data_pages() works.
1863 swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm,
1864 unsigned int nr_pages, unsigned int nr_highmem)
1866 if (nr_highmem > 0) {
1867 if (get_highmem_buffer(PG_ANY))
1869 if (nr_highmem > alloc_highmem) {
1870 nr_highmem -= alloc_highmem;
1871 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1874 if (nr_pages > alloc_normal) {
1875 nr_pages -= alloc_normal;
1876 while (nr_pages-- > 0) {
1879 page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
1882 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1893 asmlinkage __visible int swsusp_save(void)
1895 unsigned int nr_pages, nr_highmem;
1897 printk(KERN_INFO "PM: Creating hibernation image:\n");
1899 drain_local_pages(NULL);
1900 nr_pages = count_data_pages();
1901 nr_highmem = count_highmem_pages();
1902 printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem);
1904 if (!enough_free_mem(nr_pages, nr_highmem)) {
1905 printk(KERN_ERR "PM: Not enough free memory\n");
1909 if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) {
1910 printk(KERN_ERR "PM: Memory allocation failed\n");
1914 /* During allocating of suspend pagedir, new cold pages may appear.
1917 drain_local_pages(NULL);
1918 copy_data_pages(©_bm, &orig_bm);
1921 * End of critical section. From now on, we can write to memory,
1922 * but we should not touch disk. This specially means we must _not_
1923 * touch swap space! Except we must write out our image of course.
1926 nr_pages += nr_highmem;
1927 nr_copy_pages = nr_pages;
1928 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1930 printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n",
1936 #ifndef CONFIG_ARCH_HIBERNATION_HEADER
1937 static int init_header_complete(struct swsusp_info *info)
1939 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
1940 info->version_code = LINUX_VERSION_CODE;
1944 static char *check_image_kernel(struct swsusp_info *info)
1946 if (info->version_code != LINUX_VERSION_CODE)
1947 return "kernel version";
1948 if (strcmp(info->uts.sysname,init_utsname()->sysname))
1949 return "system type";
1950 if (strcmp(info->uts.release,init_utsname()->release))
1951 return "kernel release";
1952 if (strcmp(info->uts.version,init_utsname()->version))
1954 if (strcmp(info->uts.machine,init_utsname()->machine))
1958 #endif /* CONFIG_ARCH_HIBERNATION_HEADER */
1960 unsigned long snapshot_get_image_size(void)
1962 return nr_copy_pages + nr_meta_pages + 1;
1965 static int init_header(struct swsusp_info *info)
1967 memset(info, 0, sizeof(struct swsusp_info));
1968 info->num_physpages = get_num_physpages();
1969 info->image_pages = nr_copy_pages;
1970 info->pages = snapshot_get_image_size();
1971 info->size = info->pages;
1972 info->size <<= PAGE_SHIFT;
1973 return init_header_complete(info);
1977 * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm
1978 * are stored in the array @buf[] (1 page at a time)
1982 pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
1986 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1987 buf[j] = memory_bm_next_pfn(bm);
1988 if (unlikely(buf[j] == BM_END_OF_MAP))
1990 /* Save page key for data page (s390 only). */
1991 page_key_read(buf + j);
1996 * snapshot_read_next - used for reading the system memory snapshot.
1998 * On the first call to it @handle should point to a zeroed
1999 * snapshot_handle structure. The structure gets updated and a pointer
2000 * to it should be passed to this function every next time.
2002 * On success the function returns a positive number. Then, the caller
2003 * is allowed to read up to the returned number of bytes from the memory
2004 * location computed by the data_of() macro.
2006 * The function returns 0 to indicate the end of data stream condition,
2007 * and a negative number is returned on error. In such cases the
2008 * structure pointed to by @handle is not updated and should not be used
2012 int snapshot_read_next(struct snapshot_handle *handle)
2014 if (handle->cur > nr_meta_pages + nr_copy_pages)
2018 /* This makes the buffer be freed by swsusp_free() */
2019 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2026 error = init_header((struct swsusp_info *)buffer);
2029 handle->buffer = buffer;
2030 memory_bm_position_reset(&orig_bm);
2031 memory_bm_position_reset(©_bm);
2032 } else if (handle->cur <= nr_meta_pages) {
2034 pack_pfns(buffer, &orig_bm);
2038 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2039 if (PageHighMem(page)) {
2040 /* Highmem pages are copied to the buffer,
2041 * because we can't return with a kmapped
2042 * highmem page (we may not be called again).
2046 kaddr = kmap_atomic(page);
2047 copy_page(buffer, kaddr);
2048 kunmap_atomic(kaddr);
2049 handle->buffer = buffer;
2051 handle->buffer = page_address(page);
2058 static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2059 struct memory_bitmap *src)
2063 memory_bm_position_reset(src);
2064 pfn = memory_bm_next_pfn(src);
2065 while (pfn != BM_END_OF_MAP) {
2066 memory_bm_set_bit(dst, pfn);
2067 pfn = memory_bm_next_pfn(src);
2072 * mark_unsafe_pages - mark the pages that cannot be used for storing
2073 * the image during resume, because they conflict with the pages that
2074 * had been used before suspend
2077 static void mark_unsafe_pages(struct memory_bitmap *bm)
2081 /* Clear the "free"/"unsafe" bit for all PFNs */
2082 memory_bm_position_reset(free_pages_map);
2083 pfn = memory_bm_next_pfn(free_pages_map);
2084 while (pfn != BM_END_OF_MAP) {
2085 memory_bm_clear_current(free_pages_map);
2086 pfn = memory_bm_next_pfn(free_pages_map);
2089 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2090 duplicate_memory_bitmap(free_pages_map, bm);
2092 allocated_unsafe_pages = 0;
2095 static int check_header(struct swsusp_info *info)
2099 reason = check_image_kernel(info);
2100 if (!reason && info->num_physpages != get_num_physpages())
2101 reason = "memory size";
2103 printk(KERN_ERR "PM: Image mismatch: %s\n", reason);
2110 * load header - check the image header and copy data from it
2114 load_header(struct swsusp_info *info)
2118 restore_pblist = NULL;
2119 error = check_header(info);
2121 nr_copy_pages = info->image_pages;
2122 nr_meta_pages = info->pages - info->image_pages - 1;
2128 * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set
2129 * the corresponding bit in the memory bitmap @bm
2131 static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2135 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2136 if (unlikely(buf[j] == BM_END_OF_MAP))
2139 /* Extract and buffer page key for data page (s390 only). */
2140 page_key_memorize(buf + j);
2142 if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2143 memory_bm_set_bit(bm, buf[j]);
2151 #ifdef CONFIG_HIGHMEM
2152 /* struct highmem_pbe is used for creating the list of highmem pages that
2153 * should be restored atomically during the resume from disk, because the page
2154 * frames they have occupied before the suspend are in use.
2156 struct highmem_pbe {
2157 struct page *copy_page; /* data is here now */
2158 struct page *orig_page; /* data was here before the suspend */
2159 struct highmem_pbe *next;
2162 /* List of highmem PBEs needed for restoring the highmem pages that were
2163 * allocated before the suspend and included in the suspend image, but have
2164 * also been allocated by the "resume" kernel, so their contents cannot be
2165 * written directly to their "original" page frames.
2167 static struct highmem_pbe *highmem_pblist;
2170 * count_highmem_image_pages - compute the number of highmem pages in the
2171 * suspend image. The bits in the memory bitmap @bm that correspond to the
2172 * image pages are assumed to be set.
2175 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2178 unsigned int cnt = 0;
2180 memory_bm_position_reset(bm);
2181 pfn = memory_bm_next_pfn(bm);
2182 while (pfn != BM_END_OF_MAP) {
2183 if (PageHighMem(pfn_to_page(pfn)))
2186 pfn = memory_bm_next_pfn(bm);
2192 * prepare_highmem_image - try to allocate as many highmem pages as
2193 * there are highmem image pages (@nr_highmem_p points to the variable
2194 * containing the number of highmem image pages). The pages that are
2195 * "safe" (ie. will not be overwritten when the suspend image is
2196 * restored) have the corresponding bits set in @bm (it must be
2199 * NOTE: This function should not be called if there are no highmem
2203 static unsigned int safe_highmem_pages;
2205 static struct memory_bitmap *safe_highmem_bm;
2208 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
2210 unsigned int to_alloc;
2212 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2215 if (get_highmem_buffer(PG_SAFE))
2218 to_alloc = count_free_highmem_pages();
2219 if (to_alloc > *nr_highmem_p)
2220 to_alloc = *nr_highmem_p;
2222 *nr_highmem_p = to_alloc;
2224 safe_highmem_pages = 0;
2225 while (to_alloc-- > 0) {
2228 page = alloc_page(__GFP_HIGHMEM);
2229 if (!swsusp_page_is_free(page)) {
2230 /* The page is "safe", set its bit the bitmap */
2231 memory_bm_set_bit(bm, page_to_pfn(page));
2232 safe_highmem_pages++;
2234 /* Mark the page as allocated */
2235 swsusp_set_page_forbidden(page);
2236 swsusp_set_page_free(page);
2238 memory_bm_position_reset(bm);
2239 safe_highmem_bm = bm;
2244 * get_highmem_page_buffer - for given highmem image page find the buffer
2245 * that suspend_write_next() should set for its caller to write to.
2247 * If the page is to be saved to its "original" page frame or a copy of
2248 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2249 * the copy of the page is to be made in normal memory, so the address of
2250 * the copy is returned.
2252 * If @buffer is returned, the caller of suspend_write_next() will write
2253 * the page's contents to @buffer, so they will have to be copied to the
2254 * right location on the next call to suspend_write_next() and it is done
2255 * with the help of copy_last_highmem_page(). For this purpose, if
2256 * @buffer is returned, @last_highmem page is set to the page to which
2257 * the data will have to be copied from @buffer.
2260 static struct page *last_highmem_page;
2263 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
2265 struct highmem_pbe *pbe;
2268 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2269 /* We have allocated the "original" page frame and we can
2270 * use it directly to store the loaded page.
2272 last_highmem_page = page;
2275 /* The "original" page frame has not been allocated and we have to
2276 * use a "safe" page frame to store the loaded page.
2278 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2281 return ERR_PTR(-ENOMEM);
2283 pbe->orig_page = page;
2284 if (safe_highmem_pages > 0) {
2287 /* Copy of the page will be stored in high memory */
2289 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2290 safe_highmem_pages--;
2291 last_highmem_page = tmp;
2292 pbe->copy_page = tmp;
2294 /* Copy of the page will be stored in normal memory */
2295 kaddr = safe_pages_list;
2296 safe_pages_list = safe_pages_list->next;
2297 pbe->copy_page = virt_to_page(kaddr);
2299 pbe->next = highmem_pblist;
2300 highmem_pblist = pbe;
2305 * copy_last_highmem_page - copy the contents of a highmem image from
2306 * @buffer, where the caller of snapshot_write_next() has place them,
2307 * to the right location represented by @last_highmem_page .
2310 static void copy_last_highmem_page(void)
2312 if (last_highmem_page) {
2315 dst = kmap_atomic(last_highmem_page);
2316 copy_page(dst, buffer);
2318 last_highmem_page = NULL;
2322 static inline int last_highmem_page_copied(void)
2324 return !last_highmem_page;
2327 static inline void free_highmem_data(void)
2329 if (safe_highmem_bm)
2330 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2333 free_image_page(buffer, PG_UNSAFE_CLEAR);
2337 count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2340 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
2345 static inline void *
2346 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
2348 return ERR_PTR(-EINVAL);
2351 static inline void copy_last_highmem_page(void) {}
2352 static inline int last_highmem_page_copied(void) { return 1; }
2353 static inline void free_highmem_data(void) {}
2354 #endif /* CONFIG_HIGHMEM */
2357 * prepare_image - use the memory bitmap @bm to mark the pages that will
2358 * be overwritten in the process of restoring the system memory state
2359 * from the suspend image ("unsafe" pages) and allocate memory for the
2362 * The idea is to allocate a new memory bitmap first and then allocate
2363 * as many pages as needed for the image data, but not to assign these
2364 * pages to specific tasks initially. Instead, we just mark them as
2365 * allocated and create a lists of "safe" pages that will be used
2366 * later. On systems with high memory a list of "safe" highmem pages is
2370 #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2373 prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2375 unsigned int nr_pages, nr_highmem;
2376 struct linked_page *lp;
2379 /* If there is no highmem, the buffer will not be necessary */
2380 free_image_page(buffer, PG_UNSAFE_CLEAR);
2383 nr_highmem = count_highmem_image_pages(bm);
2384 mark_unsafe_pages(bm);
2386 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2390 duplicate_memory_bitmap(new_bm, bm);
2391 memory_bm_free(bm, PG_UNSAFE_KEEP);
2392 if (nr_highmem > 0) {
2393 error = prepare_highmem_image(bm, &nr_highmem);
2397 /* Reserve some safe pages for potential later use.
2399 * NOTE: This way we make sure there will be enough safe pages for the
2400 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2401 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2403 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2405 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2406 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2407 while (nr_pages > 0) {
2408 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2413 lp->next = safe_pages_list;
2414 safe_pages_list = lp;
2417 /* Preallocate memory for the image */
2418 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2419 while (nr_pages > 0) {
2420 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2425 if (!swsusp_page_is_free(virt_to_page(lp))) {
2426 /* The page is "safe", add it to the list */
2427 lp->next = safe_pages_list;
2428 safe_pages_list = lp;
2430 /* Mark the page as allocated */
2431 swsusp_set_page_forbidden(virt_to_page(lp));
2432 swsusp_set_page_free(virt_to_page(lp));
2443 * get_buffer - compute the address that snapshot_write_next() should
2444 * set for its caller to write to.
2447 static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2451 unsigned long pfn = memory_bm_next_pfn(bm);
2453 if (pfn == BM_END_OF_MAP)
2454 return ERR_PTR(-EFAULT);
2456 page = pfn_to_page(pfn);
2457 if (PageHighMem(page))
2458 return get_highmem_page_buffer(page, ca);
2460 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2461 /* We have allocated the "original" page frame and we can
2462 * use it directly to store the loaded page.
2464 return page_address(page);
2466 /* The "original" page frame has not been allocated and we have to
2467 * use a "safe" page frame to store the loaded page.
2469 pbe = chain_alloc(ca, sizeof(struct pbe));
2472 return ERR_PTR(-ENOMEM);
2474 pbe->orig_address = page_address(page);
2475 pbe->address = safe_pages_list;
2476 safe_pages_list = safe_pages_list->next;
2477 pbe->next = restore_pblist;
2478 restore_pblist = pbe;
2479 return pbe->address;
2483 * snapshot_write_next - used for writing the system memory snapshot.
2485 * On the first call to it @handle should point to a zeroed
2486 * snapshot_handle structure. The structure gets updated and a pointer
2487 * to it should be passed to this function every next time.
2489 * On success the function returns a positive number. Then, the caller
2490 * is allowed to write up to the returned number of bytes to the memory
2491 * location computed by the data_of() macro.
2493 * The function returns 0 to indicate the "end of file" condition,
2494 * and a negative number is returned on error. In such cases the
2495 * structure pointed to by @handle is not updated and should not be used
2499 int snapshot_write_next(struct snapshot_handle *handle)
2501 static struct chain_allocator ca;
2504 /* Check if we have already loaded the entire image */
2505 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2508 handle->sync_read = 1;
2512 /* This makes the buffer be freed by swsusp_free() */
2513 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2518 handle->buffer = buffer;
2519 } else if (handle->cur == 1) {
2520 error = load_header(buffer);
2524 safe_pages_list = NULL;
2526 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2530 /* Allocate buffer for page keys. */
2531 error = page_key_alloc(nr_copy_pages);
2535 } else if (handle->cur <= nr_meta_pages + 1) {
2536 error = unpack_orig_pfns(buffer, ©_bm);
2540 if (handle->cur == nr_meta_pages + 1) {
2541 error = prepare_image(&orig_bm, ©_bm);
2545 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2546 memory_bm_position_reset(&orig_bm);
2547 restore_pblist = NULL;
2548 handle->buffer = get_buffer(&orig_bm, &ca);
2549 handle->sync_read = 0;
2550 if (IS_ERR(handle->buffer))
2551 return PTR_ERR(handle->buffer);
2554 copy_last_highmem_page();
2555 /* Restore page key for data page (s390 only). */
2556 page_key_write(handle->buffer);
2557 handle->buffer = get_buffer(&orig_bm, &ca);
2558 if (IS_ERR(handle->buffer))
2559 return PTR_ERR(handle->buffer);
2560 if (handle->buffer != buffer)
2561 handle->sync_read = 0;
2568 * snapshot_write_finalize - must be called after the last call to
2569 * snapshot_write_next() in case the last page in the image happens
2570 * to be a highmem page and its contents should be stored in the
2571 * highmem. Additionally, it releases the memory that will not be
2575 void snapshot_write_finalize(struct snapshot_handle *handle)
2577 copy_last_highmem_page();
2578 /* Restore page key for data page (s390 only). */
2579 page_key_write(handle->buffer);
2581 /* Do that only if we have loaded the image entirely */
2582 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2583 memory_bm_recycle(&orig_bm);
2584 free_highmem_data();
2588 int snapshot_image_loaded(struct snapshot_handle *handle)
2590 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2591 handle->cur <= nr_meta_pages + nr_copy_pages);
2594 #ifdef CONFIG_HIGHMEM
2595 /* Assumes that @buf is ready and points to a "safe" page */
2597 swap_two_pages_data(struct page *p1, struct page *p2, void *buf)
2599 void *kaddr1, *kaddr2;
2601 kaddr1 = kmap_atomic(p1);
2602 kaddr2 = kmap_atomic(p2);
2603 copy_page(buf, kaddr1);
2604 copy_page(kaddr1, kaddr2);
2605 copy_page(kaddr2, buf);
2606 kunmap_atomic(kaddr2);
2607 kunmap_atomic(kaddr1);
2611 * restore_highmem - for each highmem page that was allocated before
2612 * the suspend and included in the suspend image, and also has been
2613 * allocated by the "resume" kernel swap its current (ie. "before
2614 * resume") contents with the previous (ie. "before suspend") one.
2616 * If the resume eventually fails, we can call this function once
2617 * again and restore the "before resume" highmem state.
2620 int restore_highmem(void)
2622 struct highmem_pbe *pbe = highmem_pblist;
2628 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2633 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2636 free_image_page(buf, PG_UNSAFE_CLEAR);
2639 #endif /* CONFIG_HIGHMEM */