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>
31 #include <asm/uaccess.h>
32 #include <asm/mmu_context.h>
33 #include <asm/pgtable.h>
34 #include <asm/tlbflush.h>
39 static int swsusp_page_is_free(struct page *);
40 static void swsusp_set_page_forbidden(struct page *);
41 static void swsusp_unset_page_forbidden(struct page *);
44 * Number of bytes to reserve for memory allocations made by device drivers
45 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
46 * cause image creation to fail (tunable via /sys/power/reserved_size).
48 unsigned long reserved_size;
50 void __init hibernate_reserved_size_init(void)
52 reserved_size = SPARE_PAGES * PAGE_SIZE;
56 * Preferred image size in bytes (tunable via /sys/power/image_size).
57 * When it is set to N, swsusp will do its best to ensure the image
58 * size will not exceed N bytes, but if that is impossible, it will
59 * try to create the smallest image possible.
61 unsigned long image_size;
63 void __init hibernate_image_size_init(void)
65 image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE;
68 /* List of PBEs needed for restoring the pages that were allocated before
69 * the suspend and included in the suspend image, but have also been
70 * allocated by the "resume" kernel, so their contents cannot be written
71 * directly to their "original" page frames.
73 struct pbe *restore_pblist;
75 /* Pointer to an auxiliary buffer (1 page) */
79 * @safe_needed - on resume, for storing the PBE list and the image,
80 * we can only use memory pages that do not conflict with the pages
81 * used before suspend. The unsafe pages have PageNosaveFree set
82 * and we count them using unsafe_pages.
84 * Each allocated image page is marked as PageNosave and PageNosaveFree
85 * so that swsusp_free() can release it.
90 #define PG_UNSAFE_CLEAR 1
91 #define PG_UNSAFE_KEEP 0
93 static unsigned int allocated_unsafe_pages;
95 static void *get_image_page(gfp_t gfp_mask, int safe_needed)
99 res = (void *)get_zeroed_page(gfp_mask);
101 while (res && swsusp_page_is_free(virt_to_page(res))) {
102 /* The page is unsafe, mark it for swsusp_free() */
103 swsusp_set_page_forbidden(virt_to_page(res));
104 allocated_unsafe_pages++;
105 res = (void *)get_zeroed_page(gfp_mask);
108 swsusp_set_page_forbidden(virt_to_page(res));
109 swsusp_set_page_free(virt_to_page(res));
114 unsigned long get_safe_page(gfp_t gfp_mask)
116 return (unsigned long)get_image_page(gfp_mask, PG_SAFE);
119 static struct page *alloc_image_page(gfp_t gfp_mask)
123 page = alloc_page(gfp_mask);
125 swsusp_set_page_forbidden(page);
126 swsusp_set_page_free(page);
132 * free_image_page - free page represented by @addr, allocated with
133 * get_image_page (page flags set by it must be cleared)
136 static inline void free_image_page(void *addr, int clear_nosave_free)
140 BUG_ON(!virt_addr_valid(addr));
142 page = virt_to_page(addr);
144 swsusp_unset_page_forbidden(page);
145 if (clear_nosave_free)
146 swsusp_unset_page_free(page);
151 /* struct linked_page is used to build chains of pages */
153 #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
156 struct linked_page *next;
157 char data[LINKED_PAGE_DATA_SIZE];
158 } __attribute__((packed));
161 free_list_of_pages(struct linked_page *list, int clear_page_nosave)
164 struct linked_page *lp = list->next;
166 free_image_page(list, clear_page_nosave);
172 * struct chain_allocator is used for allocating small objects out of
173 * a linked list of pages called 'the chain'.
175 * The chain grows each time when there is no room for a new object in
176 * the current page. The allocated objects cannot be freed individually.
177 * It is only possible to free them all at once, by freeing the entire
180 * NOTE: The chain allocator may be inefficient if the allocated objects
181 * are not much smaller than PAGE_SIZE.
184 struct chain_allocator {
185 struct linked_page *chain; /* the chain */
186 unsigned int used_space; /* total size of objects allocated out
187 * of the current page
189 gfp_t gfp_mask; /* mask for allocating pages */
190 int safe_needed; /* if set, only "safe" pages are allocated */
194 chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed)
197 ca->used_space = LINKED_PAGE_DATA_SIZE;
198 ca->gfp_mask = gfp_mask;
199 ca->safe_needed = safe_needed;
202 static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
206 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
207 struct linked_page *lp;
209 lp = get_image_page(ca->gfp_mask, ca->safe_needed);
213 lp->next = ca->chain;
217 ret = ca->chain->data + ca->used_space;
218 ca->used_space += size;
223 * Data types related to memory bitmaps.
225 * Memory bitmap is a structure consiting of many linked lists of
226 * objects. The main list's elements are of type struct zone_bitmap
227 * and each of them corresonds to one zone. For each zone bitmap
228 * object there is a list of objects of type struct bm_block that
229 * represent each blocks of bitmap in which information is stored.
231 * struct memory_bitmap contains a pointer to the main list of zone
232 * bitmap objects, a struct bm_position used for browsing the bitmap,
233 * and a pointer to the list of pages used for allocating all of the
234 * zone bitmap objects and bitmap block objects.
236 * NOTE: It has to be possible to lay out the bitmap in memory
237 * using only allocations of order 0. Additionally, the bitmap is
238 * designed to work with arbitrary number of zones (this is over the
239 * top for now, but let's avoid making unnecessary assumptions ;-).
241 * struct zone_bitmap contains a pointer to a list of bitmap block
242 * objects and a pointer to the bitmap block object that has been
243 * most recently used for setting bits. Additionally, it contains the
244 * pfns that correspond to the start and end of the represented zone.
246 * struct bm_block contains a pointer to the memory page in which
247 * information is stored (in the form of a block of bitmap)
248 * It also contains the pfns that correspond to the start and end of
249 * the represented memory area.
252 #define BM_END_OF_MAP (~0UL)
254 #define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
257 struct list_head hook; /* hook into a list of bitmap blocks */
258 unsigned long start_pfn; /* pfn represented by the first bit */
259 unsigned long end_pfn; /* pfn represented by the last bit plus 1 */
260 unsigned long *data; /* bitmap representing pages */
263 static inline unsigned long bm_block_bits(struct bm_block *bb)
265 return bb->end_pfn - bb->start_pfn;
268 /* strcut bm_position is used for browsing memory bitmaps */
271 struct bm_block *block;
275 struct memory_bitmap {
276 struct list_head blocks; /* list of bitmap blocks */
277 struct linked_page *p_list; /* list of pages used to store zone
278 * bitmap objects and bitmap block
281 struct bm_position cur; /* most recently used bit position */
284 /* Functions that operate on memory bitmaps */
286 static void memory_bm_position_reset(struct memory_bitmap *bm)
288 bm->cur.block = list_entry(bm->blocks.next, struct bm_block, hook);
292 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
295 * create_bm_block_list - create a list of block bitmap objects
296 * @pages - number of pages to track
297 * @list - list to put the allocated blocks into
298 * @ca - chain allocator to be used for allocating memory
300 static int create_bm_block_list(unsigned long pages,
301 struct list_head *list,
302 struct chain_allocator *ca)
304 unsigned int nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
306 while (nr_blocks-- > 0) {
309 bb = chain_alloc(ca, sizeof(struct bm_block));
312 list_add(&bb->hook, list);
319 struct list_head hook;
325 * free_mem_extents - free a list of memory extents
326 * @list - list of extents to empty
328 static void free_mem_extents(struct list_head *list)
330 struct mem_extent *ext, *aux;
332 list_for_each_entry_safe(ext, aux, list, hook) {
333 list_del(&ext->hook);
339 * create_mem_extents - create a list of memory extents representing
340 * contiguous ranges of PFNs
341 * @list - list to put the extents into
342 * @gfp_mask - mask to use for memory allocations
344 static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
348 INIT_LIST_HEAD(list);
350 for_each_populated_zone(zone) {
351 unsigned long zone_start, zone_end;
352 struct mem_extent *ext, *cur, *aux;
354 zone_start = zone->zone_start_pfn;
355 zone_end = zone->zone_start_pfn + zone->spanned_pages;
357 list_for_each_entry(ext, list, hook)
358 if (zone_start <= ext->end)
361 if (&ext->hook == list || zone_end < ext->start) {
362 /* New extent is necessary */
363 struct mem_extent *new_ext;
365 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
367 free_mem_extents(list);
370 new_ext->start = zone_start;
371 new_ext->end = zone_end;
372 list_add_tail(&new_ext->hook, &ext->hook);
376 /* Merge this zone's range of PFNs with the existing one */
377 if (zone_start < ext->start)
378 ext->start = zone_start;
379 if (zone_end > ext->end)
382 /* More merging may be possible */
384 list_for_each_entry_safe_continue(cur, aux, list, hook) {
385 if (zone_end < cur->start)
387 if (zone_end < cur->end)
389 list_del(&cur->hook);
398 * memory_bm_create - allocate memory for a memory bitmap
401 memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed)
403 struct chain_allocator ca;
404 struct list_head mem_extents;
405 struct mem_extent *ext;
408 chain_init(&ca, gfp_mask, safe_needed);
409 INIT_LIST_HEAD(&bm->blocks);
411 error = create_mem_extents(&mem_extents, gfp_mask);
415 list_for_each_entry(ext, &mem_extents, hook) {
417 unsigned long pfn = ext->start;
418 unsigned long pages = ext->end - ext->start;
420 bb = list_entry(bm->blocks.prev, struct bm_block, hook);
422 error = create_bm_block_list(pages, bm->blocks.prev, &ca);
426 list_for_each_entry_continue(bb, &bm->blocks, hook) {
427 bb->data = get_image_page(gfp_mask, safe_needed);
434 if (pages >= BM_BITS_PER_BLOCK) {
435 pfn += BM_BITS_PER_BLOCK;
436 pages -= BM_BITS_PER_BLOCK;
438 /* This is executed only once in the loop */
445 bm->p_list = ca.chain;
446 memory_bm_position_reset(bm);
448 free_mem_extents(&mem_extents);
452 bm->p_list = ca.chain;
453 memory_bm_free(bm, PG_UNSAFE_CLEAR);
458 * memory_bm_free - free memory occupied by the memory bitmap @bm
460 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
464 list_for_each_entry(bb, &bm->blocks, hook)
466 free_image_page(bb->data, clear_nosave_free);
468 free_list_of_pages(bm->p_list, clear_nosave_free);
470 INIT_LIST_HEAD(&bm->blocks);
474 * memory_bm_find_bit - find the bit in the bitmap @bm that corresponds
475 * to given pfn. The cur_zone_bm member of @bm and the cur_block member
476 * of @bm->cur_zone_bm are updated.
478 static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
479 void **addr, unsigned int *bit_nr)
484 * Check if the pfn corresponds to the current bitmap block and find
485 * the block where it fits if this is not the case.
488 if (pfn < bb->start_pfn)
489 list_for_each_entry_continue_reverse(bb, &bm->blocks, hook)
490 if (pfn >= bb->start_pfn)
493 if (pfn >= bb->end_pfn)
494 list_for_each_entry_continue(bb, &bm->blocks, hook)
495 if (pfn >= bb->start_pfn && pfn < bb->end_pfn)
498 if (&bb->hook == &bm->blocks)
501 /* The block has been found */
503 pfn -= bb->start_pfn;
504 bm->cur.bit = pfn + 1;
510 static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
516 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
521 static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
527 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
533 static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
539 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
541 clear_bit(bit, addr);
544 static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
550 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
552 return test_bit(bit, addr);
555 static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
560 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
564 * memory_bm_next_pfn - find the pfn that corresponds to the next set bit
565 * in the bitmap @bm. If the pfn cannot be found, BM_END_OF_MAP is
568 * It is required to run memory_bm_position_reset() before the first call to
572 static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
580 bit = find_next_bit(bb->data, bm_block_bits(bb), bit);
581 if (bit < bm_block_bits(bb))
584 bb = list_entry(bb->hook.next, struct bm_block, hook);
587 } while (&bb->hook != &bm->blocks);
589 memory_bm_position_reset(bm);
590 return BM_END_OF_MAP;
593 bm->cur.bit = bit + 1;
594 return bb->start_pfn + bit;
598 * This structure represents a range of page frames the contents of which
599 * should not be saved during the suspend.
602 struct nosave_region {
603 struct list_head list;
604 unsigned long start_pfn;
605 unsigned long end_pfn;
608 static LIST_HEAD(nosave_regions);
611 * register_nosave_region - register a range of page frames the contents
612 * of which should not be saved during the suspend (to be used in the early
613 * initialization code)
617 __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn,
620 struct nosave_region *region;
622 if (start_pfn >= end_pfn)
625 if (!list_empty(&nosave_regions)) {
626 /* Try to extend the previous region (they should be sorted) */
627 region = list_entry(nosave_regions.prev,
628 struct nosave_region, list);
629 if (region->end_pfn == start_pfn) {
630 region->end_pfn = end_pfn;
635 /* during init, this shouldn't fail */
636 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
639 /* This allocation cannot fail */
640 region = alloc_bootmem(sizeof(struct nosave_region));
641 region->start_pfn = start_pfn;
642 region->end_pfn = end_pfn;
643 list_add_tail(®ion->list, &nosave_regions);
645 printk(KERN_INFO "PM: Registered nosave memory: [mem %#010llx-%#010llx]\n",
646 (unsigned long long) start_pfn << PAGE_SHIFT,
647 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
651 * Set bits in this map correspond to the page frames the contents of which
652 * should not be saved during the suspend.
654 static struct memory_bitmap *forbidden_pages_map;
656 /* Set bits in this map correspond to free page frames. */
657 static struct memory_bitmap *free_pages_map;
660 * Each page frame allocated for creating the image is marked by setting the
661 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
664 void swsusp_set_page_free(struct page *page)
667 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
670 static int swsusp_page_is_free(struct page *page)
672 return free_pages_map ?
673 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
676 void swsusp_unset_page_free(struct page *page)
679 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
682 static void swsusp_set_page_forbidden(struct page *page)
684 if (forbidden_pages_map)
685 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
688 int swsusp_page_is_forbidden(struct page *page)
690 return forbidden_pages_map ?
691 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
694 static void swsusp_unset_page_forbidden(struct page *page)
696 if (forbidden_pages_map)
697 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
701 * mark_nosave_pages - set bits corresponding to the page frames the
702 * contents of which should not be saved in a given bitmap.
705 static void mark_nosave_pages(struct memory_bitmap *bm)
707 struct nosave_region *region;
709 if (list_empty(&nosave_regions))
712 list_for_each_entry(region, &nosave_regions, list) {
715 pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n",
716 (unsigned long long) region->start_pfn << PAGE_SHIFT,
717 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
720 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
721 if (pfn_valid(pfn)) {
723 * It is safe to ignore the result of
724 * mem_bm_set_bit_check() here, since we won't
725 * touch the PFNs for which the error is
728 mem_bm_set_bit_check(bm, pfn);
734 * create_basic_memory_bitmaps - create bitmaps needed for marking page
735 * frames that should not be saved and free page frames. The pointers
736 * forbidden_pages_map and free_pages_map are only modified if everything
737 * goes well, because we don't want the bits to be used before both bitmaps
741 int create_basic_memory_bitmaps(void)
743 struct memory_bitmap *bm1, *bm2;
746 BUG_ON(forbidden_pages_map || free_pages_map);
748 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
752 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
754 goto Free_first_object;
756 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
758 goto Free_first_bitmap;
760 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
762 goto Free_second_object;
764 forbidden_pages_map = bm1;
765 free_pages_map = bm2;
766 mark_nosave_pages(forbidden_pages_map);
768 pr_debug("PM: Basic memory bitmaps created\n");
775 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
782 * free_basic_memory_bitmaps - free memory bitmaps allocated by
783 * create_basic_memory_bitmaps(). The auxiliary pointers are necessary
784 * so that the bitmaps themselves are not referred to while they are being
788 void free_basic_memory_bitmaps(void)
790 struct memory_bitmap *bm1, *bm2;
792 BUG_ON(!(forbidden_pages_map && free_pages_map));
794 bm1 = forbidden_pages_map;
795 bm2 = free_pages_map;
796 forbidden_pages_map = NULL;
797 free_pages_map = NULL;
798 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
800 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
803 pr_debug("PM: Basic memory bitmaps freed\n");
807 * snapshot_additional_pages - estimate the number of additional pages
808 * be needed for setting up the suspend image data structures for given
809 * zone (usually the returned value is greater than the exact number)
812 unsigned int snapshot_additional_pages(struct zone *zone)
816 res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
817 res += DIV_ROUND_UP(res * sizeof(struct bm_block),
818 LINKED_PAGE_DATA_SIZE);
822 #ifdef CONFIG_HIGHMEM
824 * count_free_highmem_pages - compute the total number of free highmem
825 * pages, system-wide.
828 static unsigned int count_free_highmem_pages(void)
831 unsigned int cnt = 0;
833 for_each_populated_zone(zone)
834 if (is_highmem(zone))
835 cnt += zone_page_state(zone, NR_FREE_PAGES);
841 * saveable_highmem_page - Determine whether a highmem page should be
842 * included in the suspend image.
844 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
845 * and it isn't a part of a free chunk of pages.
847 static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
854 page = pfn_to_page(pfn);
855 if (page_zone(page) != zone)
858 BUG_ON(!PageHighMem(page));
860 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) ||
864 if (page_is_guard(page))
871 * count_highmem_pages - compute the total number of saveable highmem
875 static unsigned int count_highmem_pages(void)
880 for_each_populated_zone(zone) {
881 unsigned long pfn, max_zone_pfn;
883 if (!is_highmem(zone))
886 mark_free_pages(zone);
887 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
888 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
889 if (saveable_highmem_page(zone, pfn))
895 static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
899 #endif /* CONFIG_HIGHMEM */
902 * saveable_page - Determine whether a non-highmem page should be included
903 * in the suspend image.
905 * We should save the page if it isn't Nosave, and is not in the range
906 * of pages statically defined as 'unsaveable', and it isn't a part of
907 * a free chunk of pages.
909 static struct page *saveable_page(struct zone *zone, unsigned long pfn)
916 page = pfn_to_page(pfn);
917 if (page_zone(page) != zone)
920 BUG_ON(PageHighMem(page));
922 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
925 if (PageReserved(page)
926 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
929 if (page_is_guard(page))
936 * count_data_pages - compute the total number of saveable non-highmem
940 static unsigned int count_data_pages(void)
943 unsigned long pfn, max_zone_pfn;
946 for_each_populated_zone(zone) {
947 if (is_highmem(zone))
950 mark_free_pages(zone);
951 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
952 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
953 if (saveable_page(zone, pfn))
959 /* This is needed, because copy_page and memcpy are not usable for copying
962 static inline void do_copy_page(long *dst, long *src)
966 for (n = PAGE_SIZE / sizeof(long); n; n--)
972 * safe_copy_page - check if the page we are going to copy is marked as
973 * present in the kernel page tables (this always is the case if
974 * CONFIG_DEBUG_PAGEALLOC is not set and in that case
975 * kernel_page_present() always returns 'true').
977 static void safe_copy_page(void *dst, struct page *s_page)
979 if (kernel_page_present(s_page)) {
980 do_copy_page(dst, page_address(s_page));
982 kernel_map_pages(s_page, 1, 1);
983 do_copy_page(dst, page_address(s_page));
984 kernel_map_pages(s_page, 1, 0);
989 #ifdef CONFIG_HIGHMEM
990 static inline struct page *
991 page_is_saveable(struct zone *zone, unsigned long pfn)
993 return is_highmem(zone) ?
994 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
997 static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
999 struct page *s_page, *d_page;
1002 s_page = pfn_to_page(src_pfn);
1003 d_page = pfn_to_page(dst_pfn);
1004 if (PageHighMem(s_page)) {
1005 src = kmap_atomic(s_page);
1006 dst = kmap_atomic(d_page);
1007 do_copy_page(dst, src);
1011 if (PageHighMem(d_page)) {
1012 /* Page pointed to by src may contain some kernel
1013 * data modified by kmap_atomic()
1015 safe_copy_page(buffer, s_page);
1016 dst = kmap_atomic(d_page);
1017 copy_page(dst, buffer);
1020 safe_copy_page(page_address(d_page), s_page);
1025 #define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1027 static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1029 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1030 pfn_to_page(src_pfn));
1032 #endif /* CONFIG_HIGHMEM */
1035 copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm)
1040 for_each_populated_zone(zone) {
1041 unsigned long max_zone_pfn;
1043 mark_free_pages(zone);
1044 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1045 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1046 if (page_is_saveable(zone, pfn))
1047 memory_bm_set_bit(orig_bm, pfn);
1049 memory_bm_position_reset(orig_bm);
1050 memory_bm_position_reset(copy_bm);
1052 pfn = memory_bm_next_pfn(orig_bm);
1053 if (unlikely(pfn == BM_END_OF_MAP))
1055 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1059 /* Total number of image pages */
1060 static unsigned int nr_copy_pages;
1061 /* Number of pages needed for saving the original pfns of the image pages */
1062 static unsigned int nr_meta_pages;
1064 * Numbers of normal and highmem page frames allocated for hibernation image
1065 * before suspending devices.
1067 unsigned int alloc_normal, alloc_highmem;
1069 * Memory bitmap used for marking saveable pages (during hibernation) or
1070 * hibernation image pages (during restore)
1072 static struct memory_bitmap orig_bm;
1074 * Memory bitmap used during hibernation for marking allocated page frames that
1075 * will contain copies of saveable pages. During restore it is initially used
1076 * for marking hibernation image pages, but then the set bits from it are
1077 * duplicated in @orig_bm and it is released. On highmem systems it is next
1078 * used for marking "safe" highmem pages, but it has to be reinitialized for
1081 static struct memory_bitmap copy_bm;
1084 * swsusp_free - free pages allocated for the suspend.
1086 * Suspend pages are alocated before the atomic copy is made, so we
1087 * need to release them after the resume.
1090 void swsusp_free(void)
1093 unsigned long pfn, max_zone_pfn;
1095 for_each_populated_zone(zone) {
1096 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1097 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1098 if (pfn_valid(pfn)) {
1099 struct page *page = pfn_to_page(pfn);
1101 if (swsusp_page_is_forbidden(page) &&
1102 swsusp_page_is_free(page)) {
1103 swsusp_unset_page_forbidden(page);
1104 swsusp_unset_page_free(page);
1111 restore_pblist = NULL;
1117 /* Helper functions used for the shrinking of memory. */
1119 #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1122 * preallocate_image_pages - Allocate a number of pages for hibernation image
1123 * @nr_pages: Number of page frames to allocate.
1124 * @mask: GFP flags to use for the allocation.
1126 * Return value: Number of page frames actually allocated
1128 static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1130 unsigned long nr_alloc = 0;
1132 while (nr_pages > 0) {
1135 page = alloc_image_page(mask);
1138 memory_bm_set_bit(©_bm, page_to_pfn(page));
1139 if (PageHighMem(page))
1150 static unsigned long preallocate_image_memory(unsigned long nr_pages,
1151 unsigned long avail_normal)
1153 unsigned long alloc;
1155 if (avail_normal <= alloc_normal)
1158 alloc = avail_normal - alloc_normal;
1159 if (nr_pages < alloc)
1162 return preallocate_image_pages(alloc, GFP_IMAGE);
1165 #ifdef CONFIG_HIGHMEM
1166 static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1168 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1172 * __fraction - Compute (an approximation of) x * (multiplier / base)
1174 static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1178 return (unsigned long)x;
1181 static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1182 unsigned long highmem,
1183 unsigned long total)
1185 unsigned long alloc = __fraction(nr_pages, highmem, total);
1187 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1189 #else /* CONFIG_HIGHMEM */
1190 static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1195 static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1196 unsigned long highmem,
1197 unsigned long total)
1201 #endif /* CONFIG_HIGHMEM */
1204 * free_unnecessary_pages - Release preallocated pages not needed for the image
1206 static void free_unnecessary_pages(void)
1208 unsigned long save, to_free_normal, to_free_highmem;
1210 save = count_data_pages();
1211 if (alloc_normal >= save) {
1212 to_free_normal = alloc_normal - save;
1216 save -= alloc_normal;
1218 save += count_highmem_pages();
1219 if (alloc_highmem >= save) {
1220 to_free_highmem = alloc_highmem - save;
1222 to_free_highmem = 0;
1223 save -= alloc_highmem;
1224 if (to_free_normal > save)
1225 to_free_normal -= save;
1230 memory_bm_position_reset(©_bm);
1232 while (to_free_normal > 0 || to_free_highmem > 0) {
1233 unsigned long pfn = memory_bm_next_pfn(©_bm);
1234 struct page *page = pfn_to_page(pfn);
1236 if (PageHighMem(page)) {
1237 if (!to_free_highmem)
1242 if (!to_free_normal)
1247 memory_bm_clear_bit(©_bm, pfn);
1248 swsusp_unset_page_forbidden(page);
1249 swsusp_unset_page_free(page);
1255 * minimum_image_size - Estimate the minimum acceptable size of an image
1256 * @saveable: Number of saveable pages in the system.
1258 * We want to avoid attempting to free too much memory too hard, so estimate the
1259 * minimum acceptable size of a hibernation image to use as the lower limit for
1260 * preallocating memory.
1262 * We assume that the minimum image size should be proportional to
1264 * [number of saveable pages] - [number of pages that can be freed in theory]
1266 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1267 * and (3) inactive anonymouns pages, (4) active and (5) inactive file pages,
1268 * minus mapped file pages.
1270 static unsigned long minimum_image_size(unsigned long saveable)
1274 size = global_page_state(NR_SLAB_RECLAIMABLE)
1275 + global_page_state(NR_ACTIVE_ANON)
1276 + global_page_state(NR_INACTIVE_ANON)
1277 + global_page_state(NR_ACTIVE_FILE)
1278 + global_page_state(NR_INACTIVE_FILE)
1279 - global_page_state(NR_FILE_MAPPED);
1281 return saveable <= size ? 0 : saveable - size;
1285 * hibernate_preallocate_memory - Preallocate memory for hibernation image
1287 * To create a hibernation image it is necessary to make a copy of every page
1288 * frame in use. We also need a number of page frames to be free during
1289 * hibernation for allocations made while saving the image and for device
1290 * drivers, in case they need to allocate memory from their hibernation
1291 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1292 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1293 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1294 * total number of available page frames and allocate at least
1296 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1297 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1299 * of them, which corresponds to the maximum size of a hibernation image.
1301 * If image_size is set below the number following from the above formula,
1302 * the preallocation of memory is continued until the total number of saveable
1303 * pages in the system is below the requested image size or the minimum
1304 * acceptable image size returned by minimum_image_size(), whichever is greater.
1306 int hibernate_preallocate_memory(void)
1309 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1310 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1311 struct timeval start, stop;
1314 printk(KERN_INFO "PM: Preallocating image memory... ");
1315 do_gettimeofday(&start);
1317 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1321 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1328 /* Count the number of saveable data pages. */
1329 save_highmem = count_highmem_pages();
1330 saveable = count_data_pages();
1333 * Compute the total number of page frames we can use (count) and the
1334 * number of pages needed for image metadata (size).
1337 saveable += save_highmem;
1338 highmem = save_highmem;
1340 for_each_populated_zone(zone) {
1341 size += snapshot_additional_pages(zone);
1342 if (is_highmem(zone))
1343 highmem += zone_page_state(zone, NR_FREE_PAGES);
1345 count += zone_page_state(zone, NR_FREE_PAGES);
1347 avail_normal = count;
1349 count -= totalreserve_pages;
1351 /* Add number of pages required for page keys (s390 only). */
1352 size += page_key_additional_pages(saveable);
1354 /* Compute the maximum number of saveable pages to leave in memory. */
1355 max_size = (count - (size + PAGES_FOR_IO)) / 2
1356 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1357 /* Compute the desired number of image pages specified by image_size. */
1358 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1359 if (size > max_size)
1362 * If the desired number of image pages is at least as large as the
1363 * current number of saveable pages in memory, allocate page frames for
1364 * the image and we're done.
1366 if (size >= saveable) {
1367 pages = preallocate_image_highmem(save_highmem);
1368 pages += preallocate_image_memory(saveable - pages, avail_normal);
1372 /* Estimate the minimum size of the image. */
1373 pages = minimum_image_size(saveable);
1375 * To avoid excessive pressure on the normal zone, leave room in it to
1376 * accommodate an image of the minimum size (unless it's already too
1377 * small, in which case don't preallocate pages from it at all).
1379 if (avail_normal > pages)
1380 avail_normal -= pages;
1384 size = min_t(unsigned long, pages, max_size);
1387 * Let the memory management subsystem know that we're going to need a
1388 * large number of page frames to allocate and make it free some memory.
1389 * NOTE: If this is not done, performance will be hurt badly in some
1392 shrink_all_memory(saveable - size);
1395 * The number of saveable pages in memory was too high, so apply some
1396 * pressure to decrease it. First, make room for the largest possible
1397 * image and fail if that doesn't work. Next, try to decrease the size
1398 * of the image as much as indicated by 'size' using allocations from
1399 * highmem and non-highmem zones separately.
1401 pages_highmem = preallocate_image_highmem(highmem / 2);
1402 alloc = (count - max_size) - pages_highmem;
1403 pages = preallocate_image_memory(alloc, avail_normal);
1404 if (pages < alloc) {
1405 /* We have exhausted non-highmem pages, try highmem. */
1407 pages += pages_highmem;
1408 pages_highmem = preallocate_image_highmem(alloc);
1409 if (pages_highmem < alloc)
1411 pages += pages_highmem;
1413 * size is the desired number of saveable pages to leave in
1414 * memory, so try to preallocate (all memory - size) pages.
1416 alloc = (count - pages) - size;
1417 pages += preallocate_image_highmem(alloc);
1420 * There are approximately max_size saveable pages at this point
1421 * and we want to reduce this number down to size.
1423 alloc = max_size - size;
1424 size = preallocate_highmem_fraction(alloc, highmem, count);
1425 pages_highmem += size;
1427 size = preallocate_image_memory(alloc, avail_normal);
1428 pages_highmem += preallocate_image_highmem(alloc - size);
1429 pages += pages_highmem + size;
1433 * We only need as many page frames for the image as there are saveable
1434 * pages in memory, but we have allocated more. Release the excessive
1437 free_unnecessary_pages();
1440 do_gettimeofday(&stop);
1441 printk(KERN_CONT "done (allocated %lu pages)\n", pages);
1442 swsusp_show_speed(&start, &stop, pages, "Allocated");
1447 printk(KERN_CONT "\n");
1452 #ifdef CONFIG_HIGHMEM
1454 * count_pages_for_highmem - compute the number of non-highmem pages
1455 * that will be necessary for creating copies of highmem pages.
1458 static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1460 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1462 if (free_highmem >= nr_highmem)
1465 nr_highmem -= free_highmem;
1471 count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1472 #endif /* CONFIG_HIGHMEM */
1475 * enough_free_mem - Make sure we have enough free memory for the
1479 static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1482 unsigned int free = alloc_normal;
1484 for_each_populated_zone(zone)
1485 if (!is_highmem(zone))
1486 free += zone_page_state(zone, NR_FREE_PAGES);
1488 nr_pages += count_pages_for_highmem(nr_highmem);
1489 pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n",
1490 nr_pages, PAGES_FOR_IO, free);
1492 return free > nr_pages + PAGES_FOR_IO;
1495 #ifdef CONFIG_HIGHMEM
1497 * get_highmem_buffer - if there are some highmem pages in the suspend
1498 * image, we may need the buffer to copy them and/or load their data.
1501 static inline int get_highmem_buffer(int safe_needed)
1503 buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
1504 return buffer ? 0 : -ENOMEM;
1508 * alloc_highmem_image_pages - allocate some highmem pages for the image.
1509 * Try to allocate as many pages as needed, but if the number of free
1510 * highmem pages is lesser than that, allocate them all.
1513 static inline unsigned int
1514 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int nr_highmem)
1516 unsigned int to_alloc = count_free_highmem_pages();
1518 if (to_alloc > nr_highmem)
1519 to_alloc = nr_highmem;
1521 nr_highmem -= to_alloc;
1522 while (to_alloc-- > 0) {
1525 page = alloc_image_page(__GFP_HIGHMEM);
1526 memory_bm_set_bit(bm, page_to_pfn(page));
1531 static inline int get_highmem_buffer(int safe_needed) { return 0; }
1533 static inline unsigned int
1534 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int n) { return 0; }
1535 #endif /* CONFIG_HIGHMEM */
1538 * swsusp_alloc - allocate memory for the suspend image
1540 * We first try to allocate as many highmem pages as there are
1541 * saveable highmem pages in the system. If that fails, we allocate
1542 * non-highmem pages for the copies of the remaining highmem ones.
1544 * In this approach it is likely that the copies of highmem pages will
1545 * also be located in the high memory, because of the way in which
1546 * copy_data_pages() works.
1550 swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm,
1551 unsigned int nr_pages, unsigned int nr_highmem)
1553 if (nr_highmem > 0) {
1554 if (get_highmem_buffer(PG_ANY))
1556 if (nr_highmem > alloc_highmem) {
1557 nr_highmem -= alloc_highmem;
1558 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1561 if (nr_pages > alloc_normal) {
1562 nr_pages -= alloc_normal;
1563 while (nr_pages-- > 0) {
1566 page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
1569 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1580 asmlinkage int swsusp_save(void)
1582 unsigned int nr_pages, nr_highmem;
1584 printk(KERN_INFO "PM: Creating hibernation image:\n");
1586 drain_local_pages(NULL);
1587 nr_pages = count_data_pages();
1588 nr_highmem = count_highmem_pages();
1589 printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem);
1591 if (!enough_free_mem(nr_pages, nr_highmem)) {
1592 printk(KERN_ERR "PM: Not enough free memory\n");
1596 if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) {
1597 printk(KERN_ERR "PM: Memory allocation failed\n");
1601 /* During allocating of suspend pagedir, new cold pages may appear.
1604 drain_local_pages(NULL);
1605 copy_data_pages(©_bm, &orig_bm);
1608 * End of critical section. From now on, we can write to memory,
1609 * but we should not touch disk. This specially means we must _not_
1610 * touch swap space! Except we must write out our image of course.
1613 nr_pages += nr_highmem;
1614 nr_copy_pages = nr_pages;
1615 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1617 printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n",
1623 #ifndef CONFIG_ARCH_HIBERNATION_HEADER
1624 static int init_header_complete(struct swsusp_info *info)
1626 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
1627 info->version_code = LINUX_VERSION_CODE;
1631 static char *check_image_kernel(struct swsusp_info *info)
1633 if (info->version_code != LINUX_VERSION_CODE)
1634 return "kernel version";
1635 if (strcmp(info->uts.sysname,init_utsname()->sysname))
1636 return "system type";
1637 if (strcmp(info->uts.release,init_utsname()->release))
1638 return "kernel release";
1639 if (strcmp(info->uts.version,init_utsname()->version))
1641 if (strcmp(info->uts.machine,init_utsname()->machine))
1645 #endif /* CONFIG_ARCH_HIBERNATION_HEADER */
1647 unsigned long snapshot_get_image_size(void)
1649 return nr_copy_pages + nr_meta_pages + 1;
1652 static int init_header(struct swsusp_info *info)
1654 memset(info, 0, sizeof(struct swsusp_info));
1655 info->num_physpages = num_physpages;
1656 info->image_pages = nr_copy_pages;
1657 info->pages = snapshot_get_image_size();
1658 info->size = info->pages;
1659 info->size <<= PAGE_SHIFT;
1660 return init_header_complete(info);
1664 * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm
1665 * are stored in the array @buf[] (1 page at a time)
1669 pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
1673 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1674 buf[j] = memory_bm_next_pfn(bm);
1675 if (unlikely(buf[j] == BM_END_OF_MAP))
1677 /* Save page key for data page (s390 only). */
1678 page_key_read(buf + j);
1683 * snapshot_read_next - used for reading the system memory snapshot.
1685 * On the first call to it @handle should point to a zeroed
1686 * snapshot_handle structure. The structure gets updated and a pointer
1687 * to it should be passed to this function every next time.
1689 * On success the function returns a positive number. Then, the caller
1690 * is allowed to read up to the returned number of bytes from the memory
1691 * location computed by the data_of() macro.
1693 * The function returns 0 to indicate the end of data stream condition,
1694 * and a negative number is returned on error. In such cases the
1695 * structure pointed to by @handle is not updated and should not be used
1699 int snapshot_read_next(struct snapshot_handle *handle)
1701 if (handle->cur > nr_meta_pages + nr_copy_pages)
1705 /* This makes the buffer be freed by swsusp_free() */
1706 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
1713 error = init_header((struct swsusp_info *)buffer);
1716 handle->buffer = buffer;
1717 memory_bm_position_reset(&orig_bm);
1718 memory_bm_position_reset(©_bm);
1719 } else if (handle->cur <= nr_meta_pages) {
1721 pack_pfns(buffer, &orig_bm);
1725 page = pfn_to_page(memory_bm_next_pfn(©_bm));
1726 if (PageHighMem(page)) {
1727 /* Highmem pages are copied to the buffer,
1728 * because we can't return with a kmapped
1729 * highmem page (we may not be called again).
1733 kaddr = kmap_atomic(page);
1734 copy_page(buffer, kaddr);
1735 kunmap_atomic(kaddr);
1736 handle->buffer = buffer;
1738 handle->buffer = page_address(page);
1746 * mark_unsafe_pages - mark the pages that cannot be used for storing
1747 * the image during resume, because they conflict with the pages that
1748 * had been used before suspend
1751 static int mark_unsafe_pages(struct memory_bitmap *bm)
1754 unsigned long pfn, max_zone_pfn;
1756 /* Clear page flags */
1757 for_each_populated_zone(zone) {
1758 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1759 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1761 swsusp_unset_page_free(pfn_to_page(pfn));
1764 /* Mark pages that correspond to the "original" pfns as "unsafe" */
1765 memory_bm_position_reset(bm);
1767 pfn = memory_bm_next_pfn(bm);
1768 if (likely(pfn != BM_END_OF_MAP)) {
1769 if (likely(pfn_valid(pfn)))
1770 swsusp_set_page_free(pfn_to_page(pfn));
1774 } while (pfn != BM_END_OF_MAP);
1776 allocated_unsafe_pages = 0;
1782 duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src)
1786 memory_bm_position_reset(src);
1787 pfn = memory_bm_next_pfn(src);
1788 while (pfn != BM_END_OF_MAP) {
1789 memory_bm_set_bit(dst, pfn);
1790 pfn = memory_bm_next_pfn(src);
1794 static int check_header(struct swsusp_info *info)
1798 reason = check_image_kernel(info);
1799 if (!reason && info->num_physpages != num_physpages)
1800 reason = "memory size";
1802 printk(KERN_ERR "PM: Image mismatch: %s\n", reason);
1809 * load header - check the image header and copy data from it
1813 load_header(struct swsusp_info *info)
1817 restore_pblist = NULL;
1818 error = check_header(info);
1820 nr_copy_pages = info->image_pages;
1821 nr_meta_pages = info->pages - info->image_pages - 1;
1827 * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set
1828 * the corresponding bit in the memory bitmap @bm
1830 static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
1834 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1835 if (unlikely(buf[j] == BM_END_OF_MAP))
1838 /* Extract and buffer page key for data page (s390 only). */
1839 page_key_memorize(buf + j);
1841 if (memory_bm_pfn_present(bm, buf[j]))
1842 memory_bm_set_bit(bm, buf[j]);
1850 /* List of "safe" pages that may be used to store data loaded from the suspend
1853 static struct linked_page *safe_pages_list;
1855 #ifdef CONFIG_HIGHMEM
1856 /* struct highmem_pbe is used for creating the list of highmem pages that
1857 * should be restored atomically during the resume from disk, because the page
1858 * frames they have occupied before the suspend are in use.
1860 struct highmem_pbe {
1861 struct page *copy_page; /* data is here now */
1862 struct page *orig_page; /* data was here before the suspend */
1863 struct highmem_pbe *next;
1866 /* List of highmem PBEs needed for restoring the highmem pages that were
1867 * allocated before the suspend and included in the suspend image, but have
1868 * also been allocated by the "resume" kernel, so their contents cannot be
1869 * written directly to their "original" page frames.
1871 static struct highmem_pbe *highmem_pblist;
1874 * count_highmem_image_pages - compute the number of highmem pages in the
1875 * suspend image. The bits in the memory bitmap @bm that correspond to the
1876 * image pages are assumed to be set.
1879 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
1882 unsigned int cnt = 0;
1884 memory_bm_position_reset(bm);
1885 pfn = memory_bm_next_pfn(bm);
1886 while (pfn != BM_END_OF_MAP) {
1887 if (PageHighMem(pfn_to_page(pfn)))
1890 pfn = memory_bm_next_pfn(bm);
1896 * prepare_highmem_image - try to allocate as many highmem pages as
1897 * there are highmem image pages (@nr_highmem_p points to the variable
1898 * containing the number of highmem image pages). The pages that are
1899 * "safe" (ie. will not be overwritten when the suspend image is
1900 * restored) have the corresponding bits set in @bm (it must be
1903 * NOTE: This function should not be called if there are no highmem
1907 static unsigned int safe_highmem_pages;
1909 static struct memory_bitmap *safe_highmem_bm;
1912 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
1914 unsigned int to_alloc;
1916 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
1919 if (get_highmem_buffer(PG_SAFE))
1922 to_alloc = count_free_highmem_pages();
1923 if (to_alloc > *nr_highmem_p)
1924 to_alloc = *nr_highmem_p;
1926 *nr_highmem_p = to_alloc;
1928 safe_highmem_pages = 0;
1929 while (to_alloc-- > 0) {
1932 page = alloc_page(__GFP_HIGHMEM);
1933 if (!swsusp_page_is_free(page)) {
1934 /* The page is "safe", set its bit the bitmap */
1935 memory_bm_set_bit(bm, page_to_pfn(page));
1936 safe_highmem_pages++;
1938 /* Mark the page as allocated */
1939 swsusp_set_page_forbidden(page);
1940 swsusp_set_page_free(page);
1942 memory_bm_position_reset(bm);
1943 safe_highmem_bm = bm;
1948 * get_highmem_page_buffer - for given highmem image page find the buffer
1949 * that suspend_write_next() should set for its caller to write to.
1951 * If the page is to be saved to its "original" page frame or a copy of
1952 * the page is to be made in the highmem, @buffer is returned. Otherwise,
1953 * the copy of the page is to be made in normal memory, so the address of
1954 * the copy is returned.
1956 * If @buffer is returned, the caller of suspend_write_next() will write
1957 * the page's contents to @buffer, so they will have to be copied to the
1958 * right location on the next call to suspend_write_next() and it is done
1959 * with the help of copy_last_highmem_page(). For this purpose, if
1960 * @buffer is returned, @last_highmem page is set to the page to which
1961 * the data will have to be copied from @buffer.
1964 static struct page *last_highmem_page;
1967 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
1969 struct highmem_pbe *pbe;
1972 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
1973 /* We have allocated the "original" page frame and we can
1974 * use it directly to store the loaded page.
1976 last_highmem_page = page;
1979 /* The "original" page frame has not been allocated and we have to
1980 * use a "safe" page frame to store the loaded page.
1982 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
1985 return ERR_PTR(-ENOMEM);
1987 pbe->orig_page = page;
1988 if (safe_highmem_pages > 0) {
1991 /* Copy of the page will be stored in high memory */
1993 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
1994 safe_highmem_pages--;
1995 last_highmem_page = tmp;
1996 pbe->copy_page = tmp;
1998 /* Copy of the page will be stored in normal memory */
1999 kaddr = safe_pages_list;
2000 safe_pages_list = safe_pages_list->next;
2001 pbe->copy_page = virt_to_page(kaddr);
2003 pbe->next = highmem_pblist;
2004 highmem_pblist = pbe;
2009 * copy_last_highmem_page - copy the contents of a highmem image from
2010 * @buffer, where the caller of snapshot_write_next() has place them,
2011 * to the right location represented by @last_highmem_page .
2014 static void copy_last_highmem_page(void)
2016 if (last_highmem_page) {
2019 dst = kmap_atomic(last_highmem_page);
2020 copy_page(dst, buffer);
2022 last_highmem_page = NULL;
2026 static inline int last_highmem_page_copied(void)
2028 return !last_highmem_page;
2031 static inline void free_highmem_data(void)
2033 if (safe_highmem_bm)
2034 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2037 free_image_page(buffer, PG_UNSAFE_CLEAR);
2040 static inline int get_safe_write_buffer(void) { return 0; }
2043 count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2046 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
2051 static inline void *
2052 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
2054 return ERR_PTR(-EINVAL);
2057 static inline void copy_last_highmem_page(void) {}
2058 static inline int last_highmem_page_copied(void) { return 1; }
2059 static inline void free_highmem_data(void) {}
2060 #endif /* CONFIG_HIGHMEM */
2063 * prepare_image - use the memory bitmap @bm to mark the pages that will
2064 * be overwritten in the process of restoring the system memory state
2065 * from the suspend image ("unsafe" pages) and allocate memory for the
2068 * The idea is to allocate a new memory bitmap first and then allocate
2069 * as many pages as needed for the image data, but not to assign these
2070 * pages to specific tasks initially. Instead, we just mark them as
2071 * allocated and create a lists of "safe" pages that will be used
2072 * later. On systems with high memory a list of "safe" highmem pages is
2076 #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2079 prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2081 unsigned int nr_pages, nr_highmem;
2082 struct linked_page *sp_list, *lp;
2085 /* If there is no highmem, the buffer will not be necessary */
2086 free_image_page(buffer, PG_UNSAFE_CLEAR);
2089 nr_highmem = count_highmem_image_pages(bm);
2090 error = mark_unsafe_pages(bm);
2094 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2098 duplicate_memory_bitmap(new_bm, bm);
2099 memory_bm_free(bm, PG_UNSAFE_KEEP);
2100 if (nr_highmem > 0) {
2101 error = prepare_highmem_image(bm, &nr_highmem);
2105 /* Reserve some safe pages for potential later use.
2107 * NOTE: This way we make sure there will be enough safe pages for the
2108 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2109 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2112 /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */
2113 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2114 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2115 while (nr_pages > 0) {
2116 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2125 /* Preallocate memory for the image */
2126 safe_pages_list = NULL;
2127 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2128 while (nr_pages > 0) {
2129 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2134 if (!swsusp_page_is_free(virt_to_page(lp))) {
2135 /* The page is "safe", add it to the list */
2136 lp->next = safe_pages_list;
2137 safe_pages_list = lp;
2139 /* Mark the page as allocated */
2140 swsusp_set_page_forbidden(virt_to_page(lp));
2141 swsusp_set_page_free(virt_to_page(lp));
2144 /* Free the reserved safe pages so that chain_alloc() can use them */
2147 free_image_page(sp_list, PG_UNSAFE_CLEAR);
2158 * get_buffer - compute the address that snapshot_write_next() should
2159 * set for its caller to write to.
2162 static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2166 unsigned long pfn = memory_bm_next_pfn(bm);
2168 if (pfn == BM_END_OF_MAP)
2169 return ERR_PTR(-EFAULT);
2171 page = pfn_to_page(pfn);
2172 if (PageHighMem(page))
2173 return get_highmem_page_buffer(page, ca);
2175 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2176 /* We have allocated the "original" page frame and we can
2177 * use it directly to store the loaded page.
2179 return page_address(page);
2181 /* The "original" page frame has not been allocated and we have to
2182 * use a "safe" page frame to store the loaded page.
2184 pbe = chain_alloc(ca, sizeof(struct pbe));
2187 return ERR_PTR(-ENOMEM);
2189 pbe->orig_address = page_address(page);
2190 pbe->address = safe_pages_list;
2191 safe_pages_list = safe_pages_list->next;
2192 pbe->next = restore_pblist;
2193 restore_pblist = pbe;
2194 return pbe->address;
2198 * snapshot_write_next - used for writing the system memory snapshot.
2200 * On the first call to it @handle should point to a zeroed
2201 * snapshot_handle structure. The structure gets updated and a pointer
2202 * to it should be passed to this function every next time.
2204 * On success the function returns a positive number. Then, the caller
2205 * is allowed to write up to the returned number of bytes to the memory
2206 * location computed by the data_of() macro.
2208 * The function returns 0 to indicate the "end of file" condition,
2209 * and a negative number is returned on error. In such cases the
2210 * structure pointed to by @handle is not updated and should not be used
2214 int snapshot_write_next(struct snapshot_handle *handle)
2216 static struct chain_allocator ca;
2219 /* Check if we have already loaded the entire image */
2220 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2223 handle->sync_read = 1;
2227 /* This makes the buffer be freed by swsusp_free() */
2228 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2233 handle->buffer = buffer;
2234 } else if (handle->cur == 1) {
2235 error = load_header(buffer);
2239 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2243 /* Allocate buffer for page keys. */
2244 error = page_key_alloc(nr_copy_pages);
2248 } else if (handle->cur <= nr_meta_pages + 1) {
2249 error = unpack_orig_pfns(buffer, ©_bm);
2253 if (handle->cur == nr_meta_pages + 1) {
2254 error = prepare_image(&orig_bm, ©_bm);
2258 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2259 memory_bm_position_reset(&orig_bm);
2260 restore_pblist = NULL;
2261 handle->buffer = get_buffer(&orig_bm, &ca);
2262 handle->sync_read = 0;
2263 if (IS_ERR(handle->buffer))
2264 return PTR_ERR(handle->buffer);
2267 copy_last_highmem_page();
2268 /* Restore page key for data page (s390 only). */
2269 page_key_write(handle->buffer);
2270 handle->buffer = get_buffer(&orig_bm, &ca);
2271 if (IS_ERR(handle->buffer))
2272 return PTR_ERR(handle->buffer);
2273 if (handle->buffer != buffer)
2274 handle->sync_read = 0;
2281 * snapshot_write_finalize - must be called after the last call to
2282 * snapshot_write_next() in case the last page in the image happens
2283 * to be a highmem page and its contents should be stored in the
2284 * highmem. Additionally, it releases the memory that will not be
2288 void snapshot_write_finalize(struct snapshot_handle *handle)
2290 copy_last_highmem_page();
2291 /* Restore page key for data page (s390 only). */
2292 page_key_write(handle->buffer);
2294 /* Free only if we have loaded the image entirely */
2295 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2296 memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR);
2297 free_highmem_data();
2301 int snapshot_image_loaded(struct snapshot_handle *handle)
2303 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2304 handle->cur <= nr_meta_pages + nr_copy_pages);
2307 #ifdef CONFIG_HIGHMEM
2308 /* Assumes that @buf is ready and points to a "safe" page */
2310 swap_two_pages_data(struct page *p1, struct page *p2, void *buf)
2312 void *kaddr1, *kaddr2;
2314 kaddr1 = kmap_atomic(p1);
2315 kaddr2 = kmap_atomic(p2);
2316 copy_page(buf, kaddr1);
2317 copy_page(kaddr1, kaddr2);
2318 copy_page(kaddr2, buf);
2319 kunmap_atomic(kaddr2);
2320 kunmap_atomic(kaddr1);
2324 * restore_highmem - for each highmem page that was allocated before
2325 * the suspend and included in the suspend image, and also has been
2326 * allocated by the "resume" kernel swap its current (ie. "before
2327 * resume") contents with the previous (ie. "before suspend") one.
2329 * If the resume eventually fails, we can call this function once
2330 * again and restore the "before resume" highmem state.
2333 int restore_highmem(void)
2335 struct highmem_pbe *pbe = highmem_pblist;
2341 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2346 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2349 free_image_page(buf, PG_UNSAFE_CLEAR);
2352 #endif /* CONFIG_HIGHMEM */