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>
32 #include <asm/uaccess.h>
33 #include <asm/mmu_context.h>
34 #include <asm/pgtable.h>
35 #include <asm/tlbflush.h>
40 static int swsusp_page_is_free(struct page *);
41 static void swsusp_set_page_forbidden(struct page *);
42 static void swsusp_unset_page_forbidden(struct page *);
45 * Number of bytes to reserve for memory allocations made by device drivers
46 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
47 * cause image creation to fail (tunable via /sys/power/reserved_size).
49 unsigned long reserved_size;
51 void __init hibernate_reserved_size_init(void)
53 reserved_size = SPARE_PAGES * PAGE_SIZE;
57 * Preferred image size in bytes (tunable via /sys/power/image_size).
58 * When it is set to N, swsusp will do its best to ensure the image
59 * size will not exceed N bytes, but if that is impossible, it will
60 * try to create the smallest image possible.
62 unsigned long image_size;
64 void __init hibernate_image_size_init(void)
66 image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE;
69 /* List of PBEs needed for restoring the pages that were allocated before
70 * the suspend and included in the suspend image, but have also been
71 * allocated by the "resume" kernel, so their contents cannot be written
72 * directly to their "original" page frames.
74 struct pbe *restore_pblist;
76 /* Pointer to an auxiliary buffer (1 page) */
80 * @safe_needed - on resume, for storing the PBE list and the image,
81 * we can only use memory pages that do not conflict with the pages
82 * used before suspend. The unsafe pages have PageNosaveFree set
83 * and we count them using unsafe_pages.
85 * Each allocated image page is marked as PageNosave and PageNosaveFree
86 * so that swsusp_free() can release it.
91 #define PG_UNSAFE_CLEAR 1
92 #define PG_UNSAFE_KEEP 0
94 static unsigned int allocated_unsafe_pages;
96 static void *get_image_page(gfp_t gfp_mask, int safe_needed)
100 res = (void *)get_zeroed_page(gfp_mask);
102 while (res && swsusp_page_is_free(virt_to_page(res))) {
103 /* The page is unsafe, mark it for swsusp_free() */
104 swsusp_set_page_forbidden(virt_to_page(res));
105 allocated_unsafe_pages++;
106 res = (void *)get_zeroed_page(gfp_mask);
109 swsusp_set_page_forbidden(virt_to_page(res));
110 swsusp_set_page_free(virt_to_page(res));
115 unsigned long get_safe_page(gfp_t gfp_mask)
117 return (unsigned long)get_image_page(gfp_mask, PG_SAFE);
120 static struct page *alloc_image_page(gfp_t gfp_mask)
124 page = alloc_page(gfp_mask);
126 swsusp_set_page_forbidden(page);
127 swsusp_set_page_free(page);
133 * free_image_page - free page represented by @addr, allocated with
134 * get_image_page (page flags set by it must be cleared)
137 static inline void free_image_page(void *addr, int clear_nosave_free)
141 BUG_ON(!virt_addr_valid(addr));
143 page = virt_to_page(addr);
145 swsusp_unset_page_forbidden(page);
146 if (clear_nosave_free)
147 swsusp_unset_page_free(page);
152 /* struct linked_page is used to build chains of pages */
154 #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
157 struct linked_page *next;
158 char data[LINKED_PAGE_DATA_SIZE];
162 free_list_of_pages(struct linked_page *list, int clear_page_nosave)
165 struct linked_page *lp = list->next;
167 free_image_page(list, clear_page_nosave);
173 * struct chain_allocator is used for allocating small objects out of
174 * a linked list of pages called 'the chain'.
176 * The chain grows each time when there is no room for a new object in
177 * the current page. The allocated objects cannot be freed individually.
178 * It is only possible to free them all at once, by freeing the entire
181 * NOTE: The chain allocator may be inefficient if the allocated objects
182 * are not much smaller than PAGE_SIZE.
185 struct chain_allocator {
186 struct linked_page *chain; /* the chain */
187 unsigned int used_space; /* total size of objects allocated out
188 * of the current page
190 gfp_t gfp_mask; /* mask for allocating pages */
191 int safe_needed; /* if set, only "safe" pages are allocated */
195 chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed)
198 ca->used_space = LINKED_PAGE_DATA_SIZE;
199 ca->gfp_mask = gfp_mask;
200 ca->safe_needed = safe_needed;
203 static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
207 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
208 struct linked_page *lp;
210 lp = get_image_page(ca->gfp_mask, ca->safe_needed);
214 lp->next = ca->chain;
218 ret = ca->chain->data + ca->used_space;
219 ca->used_space += size;
224 * Data types related to memory bitmaps.
226 * Memory bitmap is a structure consiting of many linked lists of
227 * objects. The main list's elements are of type struct zone_bitmap
228 * and each of them corresonds to one zone. For each zone bitmap
229 * object there is a list of objects of type struct bm_block that
230 * represent each blocks of bitmap in which information is stored.
232 * struct memory_bitmap contains a pointer to the main list of zone
233 * bitmap objects, a struct bm_position used for browsing the bitmap,
234 * and a pointer to the list of pages used for allocating all of the
235 * zone bitmap objects and bitmap block objects.
237 * NOTE: It has to be possible to lay out the bitmap in memory
238 * using only allocations of order 0. Additionally, the bitmap is
239 * designed to work with arbitrary number of zones (this is over the
240 * top for now, but let's avoid making unnecessary assumptions ;-).
242 * struct zone_bitmap contains a pointer to a list of bitmap block
243 * objects and a pointer to the bitmap block object that has been
244 * most recently used for setting bits. Additionally, it contains the
245 * pfns that correspond to the start and end of the represented zone.
247 * struct bm_block contains a pointer to the memory page in which
248 * information is stored (in the form of a block of bitmap)
249 * It also contains the pfns that correspond to the start and end of
250 * the represented memory area.
252 * The memory bitmap is organized as a radix tree to guarantee fast random
253 * access to the bits. There is one radix tree for each zone (as returned
254 * from create_mem_extents).
256 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
257 * two linked lists for the nodes of the tree, one for the inner nodes and
258 * one for the leave nodes. The linked leave nodes are used for fast linear
259 * access of the memory bitmap.
261 * The struct rtree_node represents one node of the radix tree.
264 #define BM_END_OF_MAP (~0UL)
266 #define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
267 #define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
268 #define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
271 * struct rtree_node is a wrapper struct to link the nodes
272 * of the rtree together for easy linear iteration over
273 * bits and easy freeing
276 struct list_head list;
281 * struct mem_zone_bm_rtree represents a bitmap used for one
282 * populated memory zone.
284 struct mem_zone_bm_rtree {
285 struct list_head list; /* Link Zones together */
286 struct list_head nodes; /* Radix Tree inner nodes */
287 struct list_head leaves; /* Radix Tree leaves */
288 unsigned long start_pfn; /* Zone start page frame */
289 unsigned long end_pfn; /* Zone end page frame + 1 */
290 struct rtree_node *rtree; /* Radix Tree Root */
291 int levels; /* Number of Radix Tree Levels */
292 unsigned int blocks; /* Number of Bitmap Blocks */
295 /* strcut bm_position is used for browsing memory bitmaps */
298 struct mem_zone_bm_rtree *zone;
299 struct rtree_node *node;
300 unsigned long node_pfn;
304 struct memory_bitmap {
305 struct list_head zones;
306 struct linked_page *p_list; /* list of pages used to store zone
307 * bitmap objects and bitmap block
310 struct bm_position cur; /* most recently used bit position */
313 /* Functions that operate on memory bitmaps */
315 #define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
316 #if BITS_PER_LONG == 32
317 #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
319 #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
321 #define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
324 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
326 * This function is used to allocate inner nodes as well as the
327 * leave nodes of the radix tree. It also adds the node to the
328 * corresponding linked list passed in by the *list parameter.
330 static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
331 struct chain_allocator *ca,
332 struct list_head *list)
334 struct rtree_node *node;
336 node = chain_alloc(ca, sizeof(struct rtree_node));
340 node->data = get_image_page(gfp_mask, safe_needed);
344 list_add_tail(&node->list, list);
350 * add_rtree_block - Add a new leave node to the radix tree
352 * The leave nodes need to be allocated in order to keep the leaves
353 * linked list in order. This is guaranteed by the zone->blocks
356 static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
357 int safe_needed, struct chain_allocator *ca)
359 struct rtree_node *node, *block, **dst;
360 unsigned int levels_needed, block_nr;
363 block_nr = zone->blocks;
366 /* How many levels do we need for this block nr? */
369 block_nr >>= BM_RTREE_LEVEL_SHIFT;
372 /* Make sure the rtree has enough levels */
373 for (i = zone->levels; i < levels_needed; i++) {
374 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
379 node->data[0] = (unsigned long)zone->rtree;
384 /* Allocate new block */
385 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
389 /* Now walk the rtree to insert the block */
392 block_nr = zone->blocks;
393 for (i = zone->levels; i > 0; i--) {
397 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
404 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
405 index &= BM_RTREE_LEVEL_MASK;
406 dst = (struct rtree_node **)&((*dst)->data[index]);
416 static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
417 int clear_nosave_free);
420 * create_zone_bm_rtree - create a radix tree for one zone
422 * Allocated the mem_zone_bm_rtree structure and initializes it.
423 * This function also allocated and builds the radix tree for the
426 static struct mem_zone_bm_rtree *
427 create_zone_bm_rtree(gfp_t gfp_mask, int safe_needed,
428 struct chain_allocator *ca,
429 unsigned long start, unsigned long end)
431 struct mem_zone_bm_rtree *zone;
432 unsigned int i, nr_blocks;
436 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
440 INIT_LIST_HEAD(&zone->nodes);
441 INIT_LIST_HEAD(&zone->leaves);
442 zone->start_pfn = start;
444 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
446 for (i = 0; i < nr_blocks; i++) {
447 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
448 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
457 * free_zone_bm_rtree - Free the memory of the radix tree
459 * Free all node pages of the radix tree. The mem_zone_bm_rtree
460 * structure itself is not freed here nor are the rtree_node
463 static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
464 int clear_nosave_free)
466 struct rtree_node *node;
468 list_for_each_entry(node, &zone->nodes, list)
469 free_image_page(node->data, clear_nosave_free);
471 list_for_each_entry(node, &zone->leaves, list)
472 free_image_page(node->data, clear_nosave_free);
475 static void memory_bm_position_reset(struct memory_bitmap *bm)
477 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
479 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
480 struct rtree_node, list);
481 bm->cur.node_pfn = 0;
482 bm->cur.node_bit = 0;
485 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
488 struct list_head hook;
494 * free_mem_extents - free a list of memory extents
495 * @list - list of extents to empty
497 static void free_mem_extents(struct list_head *list)
499 struct mem_extent *ext, *aux;
501 list_for_each_entry_safe(ext, aux, list, hook) {
502 list_del(&ext->hook);
508 * create_mem_extents - create a list of memory extents representing
509 * contiguous ranges of PFNs
510 * @list - list to put the extents into
511 * @gfp_mask - mask to use for memory allocations
513 static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
517 INIT_LIST_HEAD(list);
519 for_each_populated_zone(zone) {
520 unsigned long zone_start, zone_end;
521 struct mem_extent *ext, *cur, *aux;
523 zone_start = zone->zone_start_pfn;
524 zone_end = zone_end_pfn(zone);
526 list_for_each_entry(ext, list, hook)
527 if (zone_start <= ext->end)
530 if (&ext->hook == list || zone_end < ext->start) {
531 /* New extent is necessary */
532 struct mem_extent *new_ext;
534 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
536 free_mem_extents(list);
539 new_ext->start = zone_start;
540 new_ext->end = zone_end;
541 list_add_tail(&new_ext->hook, &ext->hook);
545 /* Merge this zone's range of PFNs with the existing one */
546 if (zone_start < ext->start)
547 ext->start = zone_start;
548 if (zone_end > ext->end)
551 /* More merging may be possible */
553 list_for_each_entry_safe_continue(cur, aux, list, hook) {
554 if (zone_end < cur->start)
556 if (zone_end < cur->end)
558 list_del(&cur->hook);
567 * memory_bm_create - allocate memory for a memory bitmap
570 memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed)
572 struct chain_allocator ca;
573 struct list_head mem_extents;
574 struct mem_extent *ext;
577 chain_init(&ca, gfp_mask, safe_needed);
578 INIT_LIST_HEAD(&bm->zones);
580 error = create_mem_extents(&mem_extents, gfp_mask);
584 list_for_each_entry(ext, &mem_extents, hook) {
585 struct mem_zone_bm_rtree *zone;
587 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
588 ext->start, ext->end);
593 list_add_tail(&zone->list, &bm->zones);
596 bm->p_list = ca.chain;
597 memory_bm_position_reset(bm);
599 free_mem_extents(&mem_extents);
603 bm->p_list = ca.chain;
604 memory_bm_free(bm, PG_UNSAFE_CLEAR);
609 * memory_bm_free - free memory occupied by the memory bitmap @bm
611 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
613 struct mem_zone_bm_rtree *zone;
615 list_for_each_entry(zone, &bm->zones, list)
616 free_zone_bm_rtree(zone, clear_nosave_free);
618 free_list_of_pages(bm->p_list, clear_nosave_free);
620 INIT_LIST_HEAD(&bm->zones);
624 * memory_bm_find_bit - Find the bit for pfn in the memory
627 * Find the bit in the bitmap @bm that corresponds to given pfn.
628 * The cur.zone, cur.block and cur.node_pfn member of @bm are
630 * It walks the radix tree to find the page which contains the bit for
631 * pfn and returns the bit position in **addr and *bit_nr.
633 static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
634 void **addr, unsigned int *bit_nr)
636 struct mem_zone_bm_rtree *curr, *zone;
637 struct rtree_node *node;
642 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
647 /* Find the right zone */
648 list_for_each_entry(curr, &bm->zones, list) {
649 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
660 * We have a zone. Now walk the radix tree to find the leave
665 if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
669 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
671 for (i = zone->levels; i > 0; i--) {
674 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
675 index &= BM_RTREE_LEVEL_MASK;
676 BUG_ON(node->data[index] == 0);
677 node = (struct rtree_node *)node->data[index];
681 /* Update last position */
684 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
686 /* Set return values */
688 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
693 static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
699 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
704 static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
710 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
717 static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
723 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
725 clear_bit(bit, addr);
728 static void memory_bm_clear_current(struct memory_bitmap *bm)
732 bit = max(bm->cur.node_bit - 1, 0);
733 clear_bit(bit, bm->cur.node->data);
736 static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
742 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
744 return test_bit(bit, addr);
747 static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
752 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
756 * rtree_next_node - Jumps to the next leave node
758 * Sets the position to the beginning of the next node in the
759 * memory bitmap. This is either the next node in the current
760 * zone's radix tree or the first node in the radix tree of the
763 * Returns true if there is a next node, false otherwise.
765 static bool rtree_next_node(struct memory_bitmap *bm)
767 bm->cur.node = list_entry(bm->cur.node->list.next,
768 struct rtree_node, list);
769 if (&bm->cur.node->list != &bm->cur.zone->leaves) {
770 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
771 bm->cur.node_bit = 0;
772 touch_softlockup_watchdog();
776 /* No more nodes, goto next zone */
777 bm->cur.zone = list_entry(bm->cur.zone->list.next,
778 struct mem_zone_bm_rtree, list);
779 if (&bm->cur.zone->list != &bm->zones) {
780 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
781 struct rtree_node, list);
782 bm->cur.node_pfn = 0;
783 bm->cur.node_bit = 0;
792 * memory_bm_rtree_next_pfn - Find the next set bit in the bitmap @bm
794 * Starting from the last returned position this function searches
795 * for the next set bit in the memory bitmap and returns its
796 * number. If no more bit is set BM_END_OF_MAP is returned.
798 * It is required to run memory_bm_position_reset() before the
799 * first call to this function.
801 static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
803 unsigned long bits, pfn, pages;
807 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
808 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
809 bit = find_next_bit(bm->cur.node->data, bits,
812 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
813 bm->cur.node_bit = bit + 1;
816 } while (rtree_next_node(bm));
818 return BM_END_OF_MAP;
822 * This structure represents a range of page frames the contents of which
823 * should not be saved during the suspend.
826 struct nosave_region {
827 struct list_head list;
828 unsigned long start_pfn;
829 unsigned long end_pfn;
832 static LIST_HEAD(nosave_regions);
835 * register_nosave_region - register a range of page frames the contents
836 * of which should not be saved during the suspend (to be used in the early
837 * initialization code)
841 __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn,
844 struct nosave_region *region;
846 if (start_pfn >= end_pfn)
849 if (!list_empty(&nosave_regions)) {
850 /* Try to extend the previous region (they should be sorted) */
851 region = list_entry(nosave_regions.prev,
852 struct nosave_region, list);
853 if (region->end_pfn == start_pfn) {
854 region->end_pfn = end_pfn;
859 /* during init, this shouldn't fail */
860 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
863 /* This allocation cannot fail */
864 region = memblock_virt_alloc(sizeof(struct nosave_region), 0);
865 region->start_pfn = start_pfn;
866 region->end_pfn = end_pfn;
867 list_add_tail(®ion->list, &nosave_regions);
869 printk(KERN_INFO "PM: Registered nosave memory: [mem %#010llx-%#010llx]\n",
870 (unsigned long long) start_pfn << PAGE_SHIFT,
871 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
875 * Set bits in this map correspond to the page frames the contents of which
876 * should not be saved during the suspend.
878 static struct memory_bitmap *forbidden_pages_map;
880 /* Set bits in this map correspond to free page frames. */
881 static struct memory_bitmap *free_pages_map;
884 * Each page frame allocated for creating the image is marked by setting the
885 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
888 void swsusp_set_page_free(struct page *page)
891 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
894 static int swsusp_page_is_free(struct page *page)
896 return free_pages_map ?
897 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
900 void swsusp_unset_page_free(struct page *page)
903 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
906 static void swsusp_set_page_forbidden(struct page *page)
908 if (forbidden_pages_map)
909 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
912 int swsusp_page_is_forbidden(struct page *page)
914 return forbidden_pages_map ?
915 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
918 static void swsusp_unset_page_forbidden(struct page *page)
920 if (forbidden_pages_map)
921 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
925 * mark_nosave_pages - set bits corresponding to the page frames the
926 * contents of which should not be saved in a given bitmap.
929 static void mark_nosave_pages(struct memory_bitmap *bm)
931 struct nosave_region *region;
933 if (list_empty(&nosave_regions))
936 list_for_each_entry(region, &nosave_regions, list) {
939 pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n",
940 (unsigned long long) region->start_pfn << PAGE_SHIFT,
941 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
944 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
945 if (pfn_valid(pfn)) {
947 * It is safe to ignore the result of
948 * mem_bm_set_bit_check() here, since we won't
949 * touch the PFNs for which the error is
952 mem_bm_set_bit_check(bm, pfn);
957 static bool is_nosave_page(unsigned long pfn)
959 struct nosave_region *region;
961 list_for_each_entry(region, &nosave_regions, list) {
962 if (pfn >= region->start_pfn && pfn < region->end_pfn) {
963 pr_err("PM: %#010llx in e820 nosave region: "
964 "[mem %#010llx-%#010llx]\n",
965 (unsigned long long) pfn << PAGE_SHIFT,
966 (unsigned long long) region->start_pfn << PAGE_SHIFT,
967 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
977 * create_basic_memory_bitmaps - create bitmaps needed for marking page
978 * frames that should not be saved and free page frames. The pointers
979 * forbidden_pages_map and free_pages_map are only modified if everything
980 * goes well, because we don't want the bits to be used before both bitmaps
984 int create_basic_memory_bitmaps(void)
986 struct memory_bitmap *bm1, *bm2;
989 if (forbidden_pages_map && free_pages_map)
992 BUG_ON(forbidden_pages_map || free_pages_map);
994 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
998 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1000 goto Free_first_object;
1002 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1004 goto Free_first_bitmap;
1006 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1008 goto Free_second_object;
1010 forbidden_pages_map = bm1;
1011 free_pages_map = bm2;
1012 mark_nosave_pages(forbidden_pages_map);
1014 pr_debug("PM: Basic memory bitmaps created\n");
1021 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1028 * free_basic_memory_bitmaps - free memory bitmaps allocated by
1029 * create_basic_memory_bitmaps(). The auxiliary pointers are necessary
1030 * so that the bitmaps themselves are not referred to while they are being
1034 void free_basic_memory_bitmaps(void)
1036 struct memory_bitmap *bm1, *bm2;
1038 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1041 bm1 = forbidden_pages_map;
1042 bm2 = free_pages_map;
1043 forbidden_pages_map = NULL;
1044 free_pages_map = NULL;
1045 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1047 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1050 pr_debug("PM: Basic memory bitmaps freed\n");
1054 * snapshot_additional_pages - estimate the number of additional pages
1055 * be needed for setting up the suspend image data structures for given
1056 * zone (usually the returned value is greater than the exact number)
1059 unsigned int snapshot_additional_pages(struct zone *zone)
1061 unsigned int rtree, nodes;
1063 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1064 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1065 LINKED_PAGE_DATA_SIZE);
1067 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1074 #ifdef CONFIG_HIGHMEM
1076 * count_free_highmem_pages - compute the total number of free highmem
1077 * pages, system-wide.
1080 static unsigned int count_free_highmem_pages(void)
1083 unsigned int cnt = 0;
1085 for_each_populated_zone(zone)
1086 if (is_highmem(zone))
1087 cnt += zone_page_state(zone, NR_FREE_PAGES);
1093 * saveable_highmem_page - Determine whether a highmem page should be
1094 * included in the suspend image.
1096 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1097 * and it isn't a part of a free chunk of pages.
1099 static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1103 if (!pfn_valid(pfn))
1106 page = pfn_to_page(pfn);
1107 if (page_zone(page) != zone)
1110 BUG_ON(!PageHighMem(page));
1112 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) ||
1116 if (page_is_guard(page))
1123 * count_highmem_pages - compute the total number of saveable highmem
1127 static unsigned int count_highmem_pages(void)
1132 for_each_populated_zone(zone) {
1133 unsigned long pfn, max_zone_pfn;
1135 if (!is_highmem(zone))
1138 mark_free_pages(zone);
1139 max_zone_pfn = zone_end_pfn(zone);
1140 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1141 if (saveable_highmem_page(zone, pfn))
1147 static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1151 #endif /* CONFIG_HIGHMEM */
1154 * saveable_page - Determine whether a non-highmem page should be included
1155 * in the suspend image.
1157 * We should save the page if it isn't Nosave, and is not in the range
1158 * of pages statically defined as 'unsaveable', and it isn't a part of
1159 * a free chunk of pages.
1161 static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1165 if (!pfn_valid(pfn))
1168 page = pfn_to_page(pfn);
1169 if (page_zone(page) != zone)
1172 BUG_ON(PageHighMem(page));
1174 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1177 if (PageReserved(page)
1178 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1181 if (page_is_guard(page))
1188 * count_data_pages - compute the total number of saveable non-highmem
1192 static unsigned int count_data_pages(void)
1195 unsigned long pfn, max_zone_pfn;
1198 for_each_populated_zone(zone) {
1199 if (is_highmem(zone))
1202 mark_free_pages(zone);
1203 max_zone_pfn = zone_end_pfn(zone);
1204 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1205 if (saveable_page(zone, pfn))
1211 /* This is needed, because copy_page and memcpy are not usable for copying
1214 static inline void do_copy_page(long *dst, long *src)
1218 for (n = PAGE_SIZE / sizeof(long); n; n--)
1224 * safe_copy_page - check if the page we are going to copy is marked as
1225 * present in the kernel page tables (this always is the case if
1226 * CONFIG_DEBUG_PAGEALLOC is not set and in that case
1227 * kernel_page_present() always returns 'true').
1229 static void safe_copy_page(void *dst, struct page *s_page)
1231 if (kernel_page_present(s_page)) {
1232 do_copy_page(dst, page_address(s_page));
1234 kernel_map_pages(s_page, 1, 1);
1235 do_copy_page(dst, page_address(s_page));
1236 kernel_map_pages(s_page, 1, 0);
1241 #ifdef CONFIG_HIGHMEM
1242 static inline struct page *
1243 page_is_saveable(struct zone *zone, unsigned long pfn)
1245 return is_highmem(zone) ?
1246 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1249 static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1251 struct page *s_page, *d_page;
1254 s_page = pfn_to_page(src_pfn);
1255 d_page = pfn_to_page(dst_pfn);
1256 if (PageHighMem(s_page)) {
1257 src = kmap_atomic(s_page);
1258 dst = kmap_atomic(d_page);
1259 do_copy_page(dst, src);
1263 if (PageHighMem(d_page)) {
1264 /* Page pointed to by src may contain some kernel
1265 * data modified by kmap_atomic()
1267 safe_copy_page(buffer, s_page);
1268 dst = kmap_atomic(d_page);
1269 copy_page(dst, buffer);
1272 safe_copy_page(page_address(d_page), s_page);
1277 #define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1279 static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1281 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1282 pfn_to_page(src_pfn));
1284 #endif /* CONFIG_HIGHMEM */
1287 copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm)
1292 for_each_populated_zone(zone) {
1293 unsigned long max_zone_pfn;
1295 mark_free_pages(zone);
1296 max_zone_pfn = zone_end_pfn(zone);
1297 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1298 if (page_is_saveable(zone, pfn))
1299 memory_bm_set_bit(orig_bm, pfn);
1301 memory_bm_position_reset(orig_bm);
1302 memory_bm_position_reset(copy_bm);
1304 pfn = memory_bm_next_pfn(orig_bm);
1305 if (unlikely(pfn == BM_END_OF_MAP))
1307 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1311 /* Total number of image pages */
1312 static unsigned int nr_copy_pages;
1313 /* Number of pages needed for saving the original pfns of the image pages */
1314 static unsigned int nr_meta_pages;
1316 * Numbers of normal and highmem page frames allocated for hibernation image
1317 * before suspending devices.
1319 unsigned int alloc_normal, alloc_highmem;
1321 * Memory bitmap used for marking saveable pages (during hibernation) or
1322 * hibernation image pages (during restore)
1324 static struct memory_bitmap orig_bm;
1326 * Memory bitmap used during hibernation for marking allocated page frames that
1327 * will contain copies of saveable pages. During restore it is initially used
1328 * for marking hibernation image pages, but then the set bits from it are
1329 * duplicated in @orig_bm and it is released. On highmem systems it is next
1330 * used for marking "safe" highmem pages, but it has to be reinitialized for
1333 static struct memory_bitmap copy_bm;
1336 * swsusp_free - free pages allocated for the suspend.
1338 * Suspend pages are alocated before the atomic copy is made, so we
1339 * need to release them after the resume.
1342 void swsusp_free(void)
1344 unsigned long fb_pfn, fr_pfn;
1346 if (!forbidden_pages_map || !free_pages_map)
1349 memory_bm_position_reset(forbidden_pages_map);
1350 memory_bm_position_reset(free_pages_map);
1353 fr_pfn = memory_bm_next_pfn(free_pages_map);
1354 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1357 * Find the next bit set in both bitmaps. This is guaranteed to
1358 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1361 if (fb_pfn < fr_pfn)
1362 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1363 if (fr_pfn < fb_pfn)
1364 fr_pfn = memory_bm_next_pfn(free_pages_map);
1365 } while (fb_pfn != fr_pfn);
1367 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1368 struct page *page = pfn_to_page(fr_pfn);
1370 memory_bm_clear_current(forbidden_pages_map);
1371 memory_bm_clear_current(free_pages_map);
1379 restore_pblist = NULL;
1385 /* Helper functions used for the shrinking of memory. */
1387 #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1390 * preallocate_image_pages - Allocate a number of pages for hibernation image
1391 * @nr_pages: Number of page frames to allocate.
1392 * @mask: GFP flags to use for the allocation.
1394 * Return value: Number of page frames actually allocated
1396 static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1398 unsigned long nr_alloc = 0;
1400 while (nr_pages > 0) {
1403 page = alloc_image_page(mask);
1406 memory_bm_set_bit(©_bm, page_to_pfn(page));
1407 if (PageHighMem(page))
1418 static unsigned long preallocate_image_memory(unsigned long nr_pages,
1419 unsigned long avail_normal)
1421 unsigned long alloc;
1423 if (avail_normal <= alloc_normal)
1426 alloc = avail_normal - alloc_normal;
1427 if (nr_pages < alloc)
1430 return preallocate_image_pages(alloc, GFP_IMAGE);
1433 #ifdef CONFIG_HIGHMEM
1434 static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1436 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1440 * __fraction - Compute (an approximation of) x * (multiplier / base)
1442 static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1446 return (unsigned long)x;
1449 static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1450 unsigned long highmem,
1451 unsigned long total)
1453 unsigned long alloc = __fraction(nr_pages, highmem, total);
1455 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1457 #else /* CONFIG_HIGHMEM */
1458 static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1463 static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1464 unsigned long highmem,
1465 unsigned long total)
1469 #endif /* CONFIG_HIGHMEM */
1472 * free_unnecessary_pages - Release preallocated pages not needed for the image
1474 static void free_unnecessary_pages(void)
1476 unsigned long save, to_free_normal, to_free_highmem;
1478 save = count_data_pages();
1479 if (alloc_normal >= save) {
1480 to_free_normal = alloc_normal - save;
1484 save -= alloc_normal;
1486 save += count_highmem_pages();
1487 if (alloc_highmem >= save) {
1488 to_free_highmem = alloc_highmem - save;
1490 to_free_highmem = 0;
1491 save -= alloc_highmem;
1492 if (to_free_normal > save)
1493 to_free_normal -= save;
1498 memory_bm_position_reset(©_bm);
1500 while (to_free_normal > 0 || to_free_highmem > 0) {
1501 unsigned long pfn = memory_bm_next_pfn(©_bm);
1502 struct page *page = pfn_to_page(pfn);
1504 if (PageHighMem(page)) {
1505 if (!to_free_highmem)
1510 if (!to_free_normal)
1515 memory_bm_clear_bit(©_bm, pfn);
1516 swsusp_unset_page_forbidden(page);
1517 swsusp_unset_page_free(page);
1523 * minimum_image_size - Estimate the minimum acceptable size of an image
1524 * @saveable: Number of saveable pages in the system.
1526 * We want to avoid attempting to free too much memory too hard, so estimate the
1527 * minimum acceptable size of a hibernation image to use as the lower limit for
1528 * preallocating memory.
1530 * We assume that the minimum image size should be proportional to
1532 * [number of saveable pages] - [number of pages that can be freed in theory]
1534 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1535 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages,
1536 * minus mapped file pages.
1538 static unsigned long minimum_image_size(unsigned long saveable)
1542 size = global_page_state(NR_SLAB_RECLAIMABLE)
1543 + global_page_state(NR_ACTIVE_ANON)
1544 + global_page_state(NR_INACTIVE_ANON)
1545 + global_page_state(NR_ACTIVE_FILE)
1546 + global_page_state(NR_INACTIVE_FILE)
1547 - global_page_state(NR_FILE_MAPPED);
1549 return saveable <= size ? 0 : saveable - size;
1553 * hibernate_preallocate_memory - Preallocate memory for hibernation image
1555 * To create a hibernation image it is necessary to make a copy of every page
1556 * frame in use. We also need a number of page frames to be free during
1557 * hibernation for allocations made while saving the image and for device
1558 * drivers, in case they need to allocate memory from their hibernation
1559 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1560 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1561 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1562 * total number of available page frames and allocate at least
1564 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1565 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1567 * of them, which corresponds to the maximum size of a hibernation image.
1569 * If image_size is set below the number following from the above formula,
1570 * the preallocation of memory is continued until the total number of saveable
1571 * pages in the system is below the requested image size or the minimum
1572 * acceptable image size returned by minimum_image_size(), whichever is greater.
1574 int hibernate_preallocate_memory(void)
1577 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1578 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1579 struct timeval start, stop;
1582 printk(KERN_INFO "PM: Preallocating image memory... ");
1583 do_gettimeofday(&start);
1585 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1589 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1596 /* Count the number of saveable data pages. */
1597 save_highmem = count_highmem_pages();
1598 saveable = count_data_pages();
1601 * Compute the total number of page frames we can use (count) and the
1602 * number of pages needed for image metadata (size).
1605 saveable += save_highmem;
1606 highmem = save_highmem;
1608 for_each_populated_zone(zone) {
1609 size += snapshot_additional_pages(zone);
1610 if (is_highmem(zone))
1611 highmem += zone_page_state(zone, NR_FREE_PAGES);
1613 count += zone_page_state(zone, NR_FREE_PAGES);
1615 avail_normal = count;
1617 count -= totalreserve_pages;
1619 /* Add number of pages required for page keys (s390 only). */
1620 size += page_key_additional_pages(saveable);
1622 /* Compute the maximum number of saveable pages to leave in memory. */
1623 max_size = (count - (size + PAGES_FOR_IO)) / 2
1624 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1625 /* Compute the desired number of image pages specified by image_size. */
1626 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1627 if (size > max_size)
1630 * If the desired number of image pages is at least as large as the
1631 * current number of saveable pages in memory, allocate page frames for
1632 * the image and we're done.
1634 if (size >= saveable) {
1635 pages = preallocate_image_highmem(save_highmem);
1636 pages += preallocate_image_memory(saveable - pages, avail_normal);
1640 /* Estimate the minimum size of the image. */
1641 pages = minimum_image_size(saveable);
1643 * To avoid excessive pressure on the normal zone, leave room in it to
1644 * accommodate an image of the minimum size (unless it's already too
1645 * small, in which case don't preallocate pages from it at all).
1647 if (avail_normal > pages)
1648 avail_normal -= pages;
1652 size = min_t(unsigned long, pages, max_size);
1655 * Let the memory management subsystem know that we're going to need a
1656 * large number of page frames to allocate and make it free some memory.
1657 * NOTE: If this is not done, performance will be hurt badly in some
1660 shrink_all_memory(saveable - size);
1663 * The number of saveable pages in memory was too high, so apply some
1664 * pressure to decrease it. First, make room for the largest possible
1665 * image and fail if that doesn't work. Next, try to decrease the size
1666 * of the image as much as indicated by 'size' using allocations from
1667 * highmem and non-highmem zones separately.
1669 pages_highmem = preallocate_image_highmem(highmem / 2);
1670 alloc = count - max_size;
1671 if (alloc > pages_highmem)
1672 alloc -= pages_highmem;
1675 pages = preallocate_image_memory(alloc, avail_normal);
1676 if (pages < alloc) {
1677 /* We have exhausted non-highmem pages, try highmem. */
1679 pages += pages_highmem;
1680 pages_highmem = preallocate_image_highmem(alloc);
1681 if (pages_highmem < alloc)
1683 pages += pages_highmem;
1685 * size is the desired number of saveable pages to leave in
1686 * memory, so try to preallocate (all memory - size) pages.
1688 alloc = (count - pages) - size;
1689 pages += preallocate_image_highmem(alloc);
1692 * There are approximately max_size saveable pages at this point
1693 * and we want to reduce this number down to size.
1695 alloc = max_size - size;
1696 size = preallocate_highmem_fraction(alloc, highmem, count);
1697 pages_highmem += size;
1699 size = preallocate_image_memory(alloc, avail_normal);
1700 pages_highmem += preallocate_image_highmem(alloc - size);
1701 pages += pages_highmem + size;
1705 * We only need as many page frames for the image as there are saveable
1706 * pages in memory, but we have allocated more. Release the excessive
1709 free_unnecessary_pages();
1712 do_gettimeofday(&stop);
1713 printk(KERN_CONT "done (allocated %lu pages)\n", pages);
1714 swsusp_show_speed(&start, &stop, pages, "Allocated");
1719 printk(KERN_CONT "\n");
1724 #ifdef CONFIG_HIGHMEM
1726 * count_pages_for_highmem - compute the number of non-highmem pages
1727 * that will be necessary for creating copies of highmem pages.
1730 static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1732 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1734 if (free_highmem >= nr_highmem)
1737 nr_highmem -= free_highmem;
1743 count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1744 #endif /* CONFIG_HIGHMEM */
1747 * enough_free_mem - Make sure we have enough free memory for the
1751 static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1754 unsigned int free = alloc_normal;
1756 for_each_populated_zone(zone)
1757 if (!is_highmem(zone))
1758 free += zone_page_state(zone, NR_FREE_PAGES);
1760 nr_pages += count_pages_for_highmem(nr_highmem);
1761 pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n",
1762 nr_pages, PAGES_FOR_IO, free);
1764 return free > nr_pages + PAGES_FOR_IO;
1767 #ifdef CONFIG_HIGHMEM
1769 * get_highmem_buffer - if there are some highmem pages in the suspend
1770 * image, we may need the buffer to copy them and/or load their data.
1773 static inline int get_highmem_buffer(int safe_needed)
1775 buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
1776 return buffer ? 0 : -ENOMEM;
1780 * alloc_highmem_image_pages - allocate some highmem pages for the image.
1781 * Try to allocate as many pages as needed, but if the number of free
1782 * highmem pages is lesser than that, allocate them all.
1785 static inline unsigned int
1786 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int nr_highmem)
1788 unsigned int to_alloc = count_free_highmem_pages();
1790 if (to_alloc > nr_highmem)
1791 to_alloc = nr_highmem;
1793 nr_highmem -= to_alloc;
1794 while (to_alloc-- > 0) {
1797 page = alloc_image_page(__GFP_HIGHMEM);
1798 memory_bm_set_bit(bm, page_to_pfn(page));
1803 static inline int get_highmem_buffer(int safe_needed) { return 0; }
1805 static inline unsigned int
1806 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int n) { return 0; }
1807 #endif /* CONFIG_HIGHMEM */
1810 * swsusp_alloc - allocate memory for the suspend image
1812 * We first try to allocate as many highmem pages as there are
1813 * saveable highmem pages in the system. If that fails, we allocate
1814 * non-highmem pages for the copies of the remaining highmem ones.
1816 * In this approach it is likely that the copies of highmem pages will
1817 * also be located in the high memory, because of the way in which
1818 * copy_data_pages() works.
1822 swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm,
1823 unsigned int nr_pages, unsigned int nr_highmem)
1825 if (nr_highmem > 0) {
1826 if (get_highmem_buffer(PG_ANY))
1828 if (nr_highmem > alloc_highmem) {
1829 nr_highmem -= alloc_highmem;
1830 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1833 if (nr_pages > alloc_normal) {
1834 nr_pages -= alloc_normal;
1835 while (nr_pages-- > 0) {
1838 page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
1841 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1852 asmlinkage __visible int swsusp_save(void)
1854 unsigned int nr_pages, nr_highmem;
1856 printk(KERN_INFO "PM: Creating hibernation image:\n");
1858 drain_local_pages(NULL);
1859 nr_pages = count_data_pages();
1860 nr_highmem = count_highmem_pages();
1861 printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem);
1863 if (!enough_free_mem(nr_pages, nr_highmem)) {
1864 printk(KERN_ERR "PM: Not enough free memory\n");
1868 if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) {
1869 printk(KERN_ERR "PM: Memory allocation failed\n");
1873 /* During allocating of suspend pagedir, new cold pages may appear.
1876 drain_local_pages(NULL);
1877 copy_data_pages(©_bm, &orig_bm);
1880 * End of critical section. From now on, we can write to memory,
1881 * but we should not touch disk. This specially means we must _not_
1882 * touch swap space! Except we must write out our image of course.
1885 nr_pages += nr_highmem;
1886 nr_copy_pages = nr_pages;
1887 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1889 printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n",
1895 #ifndef CONFIG_ARCH_HIBERNATION_HEADER
1896 static int init_header_complete(struct swsusp_info *info)
1898 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
1899 info->version_code = LINUX_VERSION_CODE;
1903 static char *check_image_kernel(struct swsusp_info *info)
1905 if (info->version_code != LINUX_VERSION_CODE)
1906 return "kernel version";
1907 if (strcmp(info->uts.sysname,init_utsname()->sysname))
1908 return "system type";
1909 if (strcmp(info->uts.release,init_utsname()->release))
1910 return "kernel release";
1911 if (strcmp(info->uts.version,init_utsname()->version))
1913 if (strcmp(info->uts.machine,init_utsname()->machine))
1917 #endif /* CONFIG_ARCH_HIBERNATION_HEADER */
1919 unsigned long snapshot_get_image_size(void)
1921 return nr_copy_pages + nr_meta_pages + 1;
1924 static int init_header(struct swsusp_info *info)
1926 memset(info, 0, sizeof(struct swsusp_info));
1927 info->num_physpages = get_num_physpages();
1928 info->image_pages = nr_copy_pages;
1929 info->pages = snapshot_get_image_size();
1930 info->size = info->pages;
1931 info->size <<= PAGE_SHIFT;
1932 return init_header_complete(info);
1936 * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm
1937 * are stored in the array @buf[] (1 page at a time)
1941 pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
1945 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1946 buf[j] = memory_bm_next_pfn(bm);
1947 if (unlikely(buf[j] == BM_END_OF_MAP))
1949 /* Save page key for data page (s390 only). */
1950 page_key_read(buf + j);
1955 * snapshot_read_next - used for reading the system memory snapshot.
1957 * On the first call to it @handle should point to a zeroed
1958 * snapshot_handle structure. The structure gets updated and a pointer
1959 * to it should be passed to this function every next time.
1961 * On success the function returns a positive number. Then, the caller
1962 * is allowed to read up to the returned number of bytes from the memory
1963 * location computed by the data_of() macro.
1965 * The function returns 0 to indicate the end of data stream condition,
1966 * and a negative number is returned on error. In such cases the
1967 * structure pointed to by @handle is not updated and should not be used
1971 int snapshot_read_next(struct snapshot_handle *handle)
1973 if (handle->cur > nr_meta_pages + nr_copy_pages)
1977 /* This makes the buffer be freed by swsusp_free() */
1978 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
1985 error = init_header((struct swsusp_info *)buffer);
1988 handle->buffer = buffer;
1989 memory_bm_position_reset(&orig_bm);
1990 memory_bm_position_reset(©_bm);
1991 } else if (handle->cur <= nr_meta_pages) {
1993 pack_pfns(buffer, &orig_bm);
1997 page = pfn_to_page(memory_bm_next_pfn(©_bm));
1998 if (PageHighMem(page)) {
1999 /* Highmem pages are copied to the buffer,
2000 * because we can't return with a kmapped
2001 * highmem page (we may not be called again).
2005 kaddr = kmap_atomic(page);
2006 copy_page(buffer, kaddr);
2007 kunmap_atomic(kaddr);
2008 handle->buffer = buffer;
2010 handle->buffer = page_address(page);
2018 * mark_unsafe_pages - mark the pages that cannot be used for storing
2019 * the image during resume, because they conflict with the pages that
2020 * had been used before suspend
2023 static int mark_unsafe_pages(struct memory_bitmap *bm)
2026 unsigned long pfn, max_zone_pfn;
2028 /* Clear page flags */
2029 for_each_populated_zone(zone) {
2030 max_zone_pfn = zone_end_pfn(zone);
2031 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2033 swsusp_unset_page_free(pfn_to_page(pfn));
2036 /* Mark pages that correspond to the "original" pfns as "unsafe" */
2037 memory_bm_position_reset(bm);
2039 pfn = memory_bm_next_pfn(bm);
2040 if (likely(pfn != BM_END_OF_MAP)) {
2041 if (likely(pfn_valid(pfn)) && !is_nosave_page(pfn))
2042 swsusp_set_page_free(pfn_to_page(pfn));
2046 } while (pfn != BM_END_OF_MAP);
2048 allocated_unsafe_pages = 0;
2054 duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src)
2058 memory_bm_position_reset(src);
2059 pfn = memory_bm_next_pfn(src);
2060 while (pfn != BM_END_OF_MAP) {
2061 memory_bm_set_bit(dst, pfn);
2062 pfn = memory_bm_next_pfn(src);
2066 static int check_header(struct swsusp_info *info)
2070 reason = check_image_kernel(info);
2071 if (!reason && info->num_physpages != get_num_physpages())
2072 reason = "memory size";
2074 printk(KERN_ERR "PM: Image mismatch: %s\n", reason);
2081 * load header - check the image header and copy data from it
2085 load_header(struct swsusp_info *info)
2089 restore_pblist = NULL;
2090 error = check_header(info);
2092 nr_copy_pages = info->image_pages;
2093 nr_meta_pages = info->pages - info->image_pages - 1;
2099 * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set
2100 * the corresponding bit in the memory bitmap @bm
2102 static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2106 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2107 if (unlikely(buf[j] == BM_END_OF_MAP))
2110 /* Extract and buffer page key for data page (s390 only). */
2111 page_key_memorize(buf + j);
2113 if (memory_bm_pfn_present(bm, buf[j]))
2114 memory_bm_set_bit(bm, buf[j]);
2122 /* List of "safe" pages that may be used to store data loaded from the suspend
2125 static struct linked_page *safe_pages_list;
2127 #ifdef CONFIG_HIGHMEM
2128 /* struct highmem_pbe is used for creating the list of highmem pages that
2129 * should be restored atomically during the resume from disk, because the page
2130 * frames they have occupied before the suspend are in use.
2132 struct highmem_pbe {
2133 struct page *copy_page; /* data is here now */
2134 struct page *orig_page; /* data was here before the suspend */
2135 struct highmem_pbe *next;
2138 /* List of highmem PBEs needed for restoring the highmem pages that were
2139 * allocated before the suspend and included in the suspend image, but have
2140 * also been allocated by the "resume" kernel, so their contents cannot be
2141 * written directly to their "original" page frames.
2143 static struct highmem_pbe *highmem_pblist;
2146 * count_highmem_image_pages - compute the number of highmem pages in the
2147 * suspend image. The bits in the memory bitmap @bm that correspond to the
2148 * image pages are assumed to be set.
2151 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2154 unsigned int cnt = 0;
2156 memory_bm_position_reset(bm);
2157 pfn = memory_bm_next_pfn(bm);
2158 while (pfn != BM_END_OF_MAP) {
2159 if (PageHighMem(pfn_to_page(pfn)))
2162 pfn = memory_bm_next_pfn(bm);
2168 * prepare_highmem_image - try to allocate as many highmem pages as
2169 * there are highmem image pages (@nr_highmem_p points to the variable
2170 * containing the number of highmem image pages). The pages that are
2171 * "safe" (ie. will not be overwritten when the suspend image is
2172 * restored) have the corresponding bits set in @bm (it must be
2175 * NOTE: This function should not be called if there are no highmem
2179 static unsigned int safe_highmem_pages;
2181 static struct memory_bitmap *safe_highmem_bm;
2184 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
2186 unsigned int to_alloc;
2188 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2191 if (get_highmem_buffer(PG_SAFE))
2194 to_alloc = count_free_highmem_pages();
2195 if (to_alloc > *nr_highmem_p)
2196 to_alloc = *nr_highmem_p;
2198 *nr_highmem_p = to_alloc;
2200 safe_highmem_pages = 0;
2201 while (to_alloc-- > 0) {
2204 page = alloc_page(__GFP_HIGHMEM);
2205 if (!swsusp_page_is_free(page)) {
2206 /* The page is "safe", set its bit the bitmap */
2207 memory_bm_set_bit(bm, page_to_pfn(page));
2208 safe_highmem_pages++;
2210 /* Mark the page as allocated */
2211 swsusp_set_page_forbidden(page);
2212 swsusp_set_page_free(page);
2214 memory_bm_position_reset(bm);
2215 safe_highmem_bm = bm;
2220 * get_highmem_page_buffer - for given highmem image page find the buffer
2221 * that suspend_write_next() should set for its caller to write to.
2223 * If the page is to be saved to its "original" page frame or a copy of
2224 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2225 * the copy of the page is to be made in normal memory, so the address of
2226 * the copy is returned.
2228 * If @buffer is returned, the caller of suspend_write_next() will write
2229 * the page's contents to @buffer, so they will have to be copied to the
2230 * right location on the next call to suspend_write_next() and it is done
2231 * with the help of copy_last_highmem_page(). For this purpose, if
2232 * @buffer is returned, @last_highmem page is set to the page to which
2233 * the data will have to be copied from @buffer.
2236 static struct page *last_highmem_page;
2239 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
2241 struct highmem_pbe *pbe;
2244 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2245 /* We have allocated the "original" page frame and we can
2246 * use it directly to store the loaded page.
2248 last_highmem_page = page;
2251 /* The "original" page frame has not been allocated and we have to
2252 * use a "safe" page frame to store the loaded page.
2254 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2257 return ERR_PTR(-ENOMEM);
2259 pbe->orig_page = page;
2260 if (safe_highmem_pages > 0) {
2263 /* Copy of the page will be stored in high memory */
2265 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2266 safe_highmem_pages--;
2267 last_highmem_page = tmp;
2268 pbe->copy_page = tmp;
2270 /* Copy of the page will be stored in normal memory */
2271 kaddr = safe_pages_list;
2272 safe_pages_list = safe_pages_list->next;
2273 pbe->copy_page = virt_to_page(kaddr);
2275 pbe->next = highmem_pblist;
2276 highmem_pblist = pbe;
2281 * copy_last_highmem_page - copy the contents of a highmem image from
2282 * @buffer, where the caller of snapshot_write_next() has place them,
2283 * to the right location represented by @last_highmem_page .
2286 static void copy_last_highmem_page(void)
2288 if (last_highmem_page) {
2291 dst = kmap_atomic(last_highmem_page);
2292 copy_page(dst, buffer);
2294 last_highmem_page = NULL;
2298 static inline int last_highmem_page_copied(void)
2300 return !last_highmem_page;
2303 static inline void free_highmem_data(void)
2305 if (safe_highmem_bm)
2306 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2309 free_image_page(buffer, PG_UNSAFE_CLEAR);
2312 static inline int get_safe_write_buffer(void) { return 0; }
2315 count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2318 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
2323 static inline void *
2324 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
2326 return ERR_PTR(-EINVAL);
2329 static inline void copy_last_highmem_page(void) {}
2330 static inline int last_highmem_page_copied(void) { return 1; }
2331 static inline void free_highmem_data(void) {}
2332 #endif /* CONFIG_HIGHMEM */
2335 * prepare_image - use the memory bitmap @bm to mark the pages that will
2336 * be overwritten in the process of restoring the system memory state
2337 * from the suspend image ("unsafe" pages) and allocate memory for the
2340 * The idea is to allocate a new memory bitmap first and then allocate
2341 * as many pages as needed for the image data, but not to assign these
2342 * pages to specific tasks initially. Instead, we just mark them as
2343 * allocated and create a lists of "safe" pages that will be used
2344 * later. On systems with high memory a list of "safe" highmem pages is
2348 #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2351 prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2353 unsigned int nr_pages, nr_highmem;
2354 struct linked_page *sp_list, *lp;
2357 /* If there is no highmem, the buffer will not be necessary */
2358 free_image_page(buffer, PG_UNSAFE_CLEAR);
2361 nr_highmem = count_highmem_image_pages(bm);
2362 error = mark_unsafe_pages(bm);
2366 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2370 duplicate_memory_bitmap(new_bm, bm);
2371 memory_bm_free(bm, PG_UNSAFE_KEEP);
2372 if (nr_highmem > 0) {
2373 error = prepare_highmem_image(bm, &nr_highmem);
2377 /* Reserve some safe pages for potential later use.
2379 * NOTE: This way we make sure there will be enough safe pages for the
2380 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2381 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2384 /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */
2385 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2386 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2387 while (nr_pages > 0) {
2388 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2397 /* Preallocate memory for the image */
2398 safe_pages_list = NULL;
2399 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2400 while (nr_pages > 0) {
2401 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2406 if (!swsusp_page_is_free(virt_to_page(lp))) {
2407 /* The page is "safe", add it to the list */
2408 lp->next = safe_pages_list;
2409 safe_pages_list = lp;
2411 /* Mark the page as allocated */
2412 swsusp_set_page_forbidden(virt_to_page(lp));
2413 swsusp_set_page_free(virt_to_page(lp));
2416 /* Free the reserved safe pages so that chain_alloc() can use them */
2419 free_image_page(sp_list, PG_UNSAFE_CLEAR);
2430 * get_buffer - compute the address that snapshot_write_next() should
2431 * set for its caller to write to.
2434 static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2438 unsigned long pfn = memory_bm_next_pfn(bm);
2440 if (pfn == BM_END_OF_MAP)
2441 return ERR_PTR(-EFAULT);
2443 page = pfn_to_page(pfn);
2444 if (PageHighMem(page))
2445 return get_highmem_page_buffer(page, ca);
2447 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2448 /* We have allocated the "original" page frame and we can
2449 * use it directly to store the loaded page.
2451 return page_address(page);
2453 /* The "original" page frame has not been allocated and we have to
2454 * use a "safe" page frame to store the loaded page.
2456 pbe = chain_alloc(ca, sizeof(struct pbe));
2459 return ERR_PTR(-ENOMEM);
2461 pbe->orig_address = page_address(page);
2462 pbe->address = safe_pages_list;
2463 safe_pages_list = safe_pages_list->next;
2464 pbe->next = restore_pblist;
2465 restore_pblist = pbe;
2466 return pbe->address;
2470 * snapshot_write_next - used for writing the system memory snapshot.
2472 * On the first call to it @handle should point to a zeroed
2473 * snapshot_handle structure. The structure gets updated and a pointer
2474 * to it should be passed to this function every next time.
2476 * On success the function returns a positive number. Then, the caller
2477 * is allowed to write up to the returned number of bytes to the memory
2478 * location computed by the data_of() macro.
2480 * The function returns 0 to indicate the "end of file" condition,
2481 * and a negative number is returned on error. In such cases the
2482 * structure pointed to by @handle is not updated and should not be used
2486 int snapshot_write_next(struct snapshot_handle *handle)
2488 static struct chain_allocator ca;
2491 /* Check if we have already loaded the entire image */
2492 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2495 handle->sync_read = 1;
2499 /* This makes the buffer be freed by swsusp_free() */
2500 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2505 handle->buffer = buffer;
2506 } else if (handle->cur == 1) {
2507 error = load_header(buffer);
2511 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2515 /* Allocate buffer for page keys. */
2516 error = page_key_alloc(nr_copy_pages);
2520 } else if (handle->cur <= nr_meta_pages + 1) {
2521 error = unpack_orig_pfns(buffer, ©_bm);
2525 if (handle->cur == nr_meta_pages + 1) {
2526 error = prepare_image(&orig_bm, ©_bm);
2530 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2531 memory_bm_position_reset(&orig_bm);
2532 restore_pblist = NULL;
2533 handle->buffer = get_buffer(&orig_bm, &ca);
2534 handle->sync_read = 0;
2535 if (IS_ERR(handle->buffer))
2536 return PTR_ERR(handle->buffer);
2539 copy_last_highmem_page();
2540 /* Restore page key for data page (s390 only). */
2541 page_key_write(handle->buffer);
2542 handle->buffer = get_buffer(&orig_bm, &ca);
2543 if (IS_ERR(handle->buffer))
2544 return PTR_ERR(handle->buffer);
2545 if (handle->buffer != buffer)
2546 handle->sync_read = 0;
2553 * snapshot_write_finalize - must be called after the last call to
2554 * snapshot_write_next() in case the last page in the image happens
2555 * to be a highmem page and its contents should be stored in the
2556 * highmem. Additionally, it releases the memory that will not be
2560 void snapshot_write_finalize(struct snapshot_handle *handle)
2562 copy_last_highmem_page();
2563 /* Restore page key for data page (s390 only). */
2564 page_key_write(handle->buffer);
2566 /* Free only if we have loaded the image entirely */
2567 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2568 memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR);
2569 free_highmem_data();
2573 int snapshot_image_loaded(struct snapshot_handle *handle)
2575 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2576 handle->cur <= nr_meta_pages + nr_copy_pages);
2579 #ifdef CONFIG_HIGHMEM
2580 /* Assumes that @buf is ready and points to a "safe" page */
2582 swap_two_pages_data(struct page *p1, struct page *p2, void *buf)
2584 void *kaddr1, *kaddr2;
2586 kaddr1 = kmap_atomic(p1);
2587 kaddr2 = kmap_atomic(p2);
2588 copy_page(buf, kaddr1);
2589 copy_page(kaddr1, kaddr2);
2590 copy_page(kaddr2, buf);
2591 kunmap_atomic(kaddr2);
2592 kunmap_atomic(kaddr1);
2596 * restore_highmem - for each highmem page that was allocated before
2597 * the suspend and included in the suspend image, and also has been
2598 * allocated by the "resume" kernel swap its current (ie. "before
2599 * resume") contents with the previous (ie. "before suspend") one.
2601 * If the resume eventually fails, we can call this function once
2602 * again and restore the "before resume" highmem state.
2605 int restore_highmem(void)
2607 struct highmem_pbe *pbe = highmem_pblist;
2613 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2618 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2621 free_image_page(buf, PG_UNSAFE_CLEAR);
2624 #endif /* CONFIG_HIGHMEM */