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: %016lx - %016lx\n",
646 start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT);
650 * Set bits in this map correspond to the page frames the contents of which
651 * should not be saved during the suspend.
653 static struct memory_bitmap *forbidden_pages_map;
655 /* Set bits in this map correspond to free page frames. */
656 static struct memory_bitmap *free_pages_map;
659 * Each page frame allocated for creating the image is marked by setting the
660 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
663 void swsusp_set_page_free(struct page *page)
666 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
669 static int swsusp_page_is_free(struct page *page)
671 return free_pages_map ?
672 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
675 void swsusp_unset_page_free(struct page *page)
678 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
681 static void swsusp_set_page_forbidden(struct page *page)
683 if (forbidden_pages_map)
684 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
687 int swsusp_page_is_forbidden(struct page *page)
689 return forbidden_pages_map ?
690 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
693 static void swsusp_unset_page_forbidden(struct page *page)
695 if (forbidden_pages_map)
696 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
700 * mark_nosave_pages - set bits corresponding to the page frames the
701 * contents of which should not be saved in a given bitmap.
704 static void mark_nosave_pages(struct memory_bitmap *bm)
706 struct nosave_region *region;
708 if (list_empty(&nosave_regions))
711 list_for_each_entry(region, &nosave_regions, list) {
714 pr_debug("PM: Marking nosave pages: %016lx - %016lx\n",
715 region->start_pfn << PAGE_SHIFT,
716 region->end_pfn << PAGE_SHIFT);
718 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
719 if (pfn_valid(pfn)) {
721 * It is safe to ignore the result of
722 * mem_bm_set_bit_check() here, since we won't
723 * touch the PFNs for which the error is
726 mem_bm_set_bit_check(bm, pfn);
732 * create_basic_memory_bitmaps - create bitmaps needed for marking page
733 * frames that should not be saved and free page frames. The pointers
734 * forbidden_pages_map and free_pages_map are only modified if everything
735 * goes well, because we don't want the bits to be used before both bitmaps
739 int create_basic_memory_bitmaps(void)
741 struct memory_bitmap *bm1, *bm2;
744 BUG_ON(forbidden_pages_map || free_pages_map);
746 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
750 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
752 goto Free_first_object;
754 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
756 goto Free_first_bitmap;
758 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
760 goto Free_second_object;
762 forbidden_pages_map = bm1;
763 free_pages_map = bm2;
764 mark_nosave_pages(forbidden_pages_map);
766 pr_debug("PM: Basic memory bitmaps created\n");
773 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
780 * free_basic_memory_bitmaps - free memory bitmaps allocated by
781 * create_basic_memory_bitmaps(). The auxiliary pointers are necessary
782 * so that the bitmaps themselves are not referred to while they are being
786 void free_basic_memory_bitmaps(void)
788 struct memory_bitmap *bm1, *bm2;
790 BUG_ON(!(forbidden_pages_map && free_pages_map));
792 bm1 = forbidden_pages_map;
793 bm2 = free_pages_map;
794 forbidden_pages_map = NULL;
795 free_pages_map = NULL;
796 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
798 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
801 pr_debug("PM: Basic memory bitmaps freed\n");
805 * snapshot_additional_pages - estimate the number of additional pages
806 * be needed for setting up the suspend image data structures for given
807 * zone (usually the returned value is greater than the exact number)
810 unsigned int snapshot_additional_pages(struct zone *zone)
814 res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
815 res += DIV_ROUND_UP(res * sizeof(struct bm_block), PAGE_SIZE);
819 #ifdef CONFIG_HIGHMEM
821 * count_free_highmem_pages - compute the total number of free highmem
822 * pages, system-wide.
825 static unsigned int count_free_highmem_pages(void)
828 unsigned int cnt = 0;
830 for_each_populated_zone(zone)
831 if (is_highmem(zone))
832 cnt += zone_page_state(zone, NR_FREE_PAGES);
838 * saveable_highmem_page - Determine whether a highmem page should be
839 * included in the suspend image.
841 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
842 * and it isn't a part of a free chunk of pages.
844 static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
851 page = pfn_to_page(pfn);
852 if (page_zone(page) != zone)
855 BUG_ON(!PageHighMem(page));
857 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) ||
865 * count_highmem_pages - compute the total number of saveable highmem
869 static unsigned int count_highmem_pages(void)
874 for_each_populated_zone(zone) {
875 unsigned long pfn, max_zone_pfn;
877 if (!is_highmem(zone))
880 mark_free_pages(zone);
881 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
882 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
883 if (saveable_highmem_page(zone, pfn))
889 static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
893 #endif /* CONFIG_HIGHMEM */
896 * saveable_page - Determine whether a non-highmem page should be included
897 * in the suspend image.
899 * We should save the page if it isn't Nosave, and is not in the range
900 * of pages statically defined as 'unsaveable', and it isn't a part of
901 * a free chunk of pages.
903 static struct page *saveable_page(struct zone *zone, unsigned long pfn)
910 page = pfn_to_page(pfn);
911 if (page_zone(page) != zone)
914 BUG_ON(PageHighMem(page));
916 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
919 if (PageReserved(page)
920 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
927 * count_data_pages - compute the total number of saveable non-highmem
931 static unsigned int count_data_pages(void)
934 unsigned long pfn, max_zone_pfn;
937 for_each_populated_zone(zone) {
938 if (is_highmem(zone))
941 mark_free_pages(zone);
942 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
943 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
944 if (saveable_page(zone, pfn))
950 /* This is needed, because copy_page and memcpy are not usable for copying
953 static inline void do_copy_page(long *dst, long *src)
957 for (n = PAGE_SIZE / sizeof(long); n; n--)
963 * safe_copy_page - check if the page we are going to copy is marked as
964 * present in the kernel page tables (this always is the case if
965 * CONFIG_DEBUG_PAGEALLOC is not set and in that case
966 * kernel_page_present() always returns 'true').
968 static void safe_copy_page(void *dst, struct page *s_page)
970 if (kernel_page_present(s_page)) {
971 do_copy_page(dst, page_address(s_page));
973 kernel_map_pages(s_page, 1, 1);
974 do_copy_page(dst, page_address(s_page));
975 kernel_map_pages(s_page, 1, 0);
980 #ifdef CONFIG_HIGHMEM
981 static inline struct page *
982 page_is_saveable(struct zone *zone, unsigned long pfn)
984 return is_highmem(zone) ?
985 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
988 static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
990 struct page *s_page, *d_page;
993 s_page = pfn_to_page(src_pfn);
994 d_page = pfn_to_page(dst_pfn);
995 if (PageHighMem(s_page)) {
996 src = kmap_atomic(s_page, KM_USER0);
997 dst = kmap_atomic(d_page, KM_USER1);
998 do_copy_page(dst, src);
999 kunmap_atomic(dst, KM_USER1);
1000 kunmap_atomic(src, KM_USER0);
1002 if (PageHighMem(d_page)) {
1003 /* Page pointed to by src may contain some kernel
1004 * data modified by kmap_atomic()
1006 safe_copy_page(buffer, s_page);
1007 dst = kmap_atomic(d_page, KM_USER0);
1008 copy_page(dst, buffer);
1009 kunmap_atomic(dst, KM_USER0);
1011 safe_copy_page(page_address(d_page), s_page);
1016 #define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1018 static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1020 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1021 pfn_to_page(src_pfn));
1023 #endif /* CONFIG_HIGHMEM */
1026 copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm)
1031 for_each_populated_zone(zone) {
1032 unsigned long max_zone_pfn;
1034 mark_free_pages(zone);
1035 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1036 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1037 if (page_is_saveable(zone, pfn))
1038 memory_bm_set_bit(orig_bm, pfn);
1040 memory_bm_position_reset(orig_bm);
1041 memory_bm_position_reset(copy_bm);
1043 pfn = memory_bm_next_pfn(orig_bm);
1044 if (unlikely(pfn == BM_END_OF_MAP))
1046 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1050 /* Total number of image pages */
1051 static unsigned int nr_copy_pages;
1052 /* Number of pages needed for saving the original pfns of the image pages */
1053 static unsigned int nr_meta_pages;
1055 * Numbers of normal and highmem page frames allocated for hibernation image
1056 * before suspending devices.
1058 unsigned int alloc_normal, alloc_highmem;
1060 * Memory bitmap used for marking saveable pages (during hibernation) or
1061 * hibernation image pages (during restore)
1063 static struct memory_bitmap orig_bm;
1065 * Memory bitmap used during hibernation for marking allocated page frames that
1066 * will contain copies of saveable pages. During restore it is initially used
1067 * for marking hibernation image pages, but then the set bits from it are
1068 * duplicated in @orig_bm and it is released. On highmem systems it is next
1069 * used for marking "safe" highmem pages, but it has to be reinitialized for
1072 static struct memory_bitmap copy_bm;
1075 * swsusp_free - free pages allocated for the suspend.
1077 * Suspend pages are alocated before the atomic copy is made, so we
1078 * need to release them after the resume.
1081 void swsusp_free(void)
1084 unsigned long pfn, max_zone_pfn;
1086 for_each_populated_zone(zone) {
1087 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1088 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1089 if (pfn_valid(pfn)) {
1090 struct page *page = pfn_to_page(pfn);
1092 if (swsusp_page_is_forbidden(page) &&
1093 swsusp_page_is_free(page)) {
1094 swsusp_unset_page_forbidden(page);
1095 swsusp_unset_page_free(page);
1102 restore_pblist = NULL;
1108 /* Helper functions used for the shrinking of memory. */
1110 #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1113 * preallocate_image_pages - Allocate a number of pages for hibernation image
1114 * @nr_pages: Number of page frames to allocate.
1115 * @mask: GFP flags to use for the allocation.
1117 * Return value: Number of page frames actually allocated
1119 static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1121 unsigned long nr_alloc = 0;
1123 while (nr_pages > 0) {
1126 page = alloc_image_page(mask);
1129 memory_bm_set_bit(©_bm, page_to_pfn(page));
1130 if (PageHighMem(page))
1141 static unsigned long preallocate_image_memory(unsigned long nr_pages,
1142 unsigned long avail_normal)
1144 unsigned long alloc;
1146 if (avail_normal <= alloc_normal)
1149 alloc = avail_normal - alloc_normal;
1150 if (nr_pages < alloc)
1153 return preallocate_image_pages(alloc, GFP_IMAGE);
1156 #ifdef CONFIG_HIGHMEM
1157 static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1159 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1163 * __fraction - Compute (an approximation of) x * (multiplier / base)
1165 static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1169 return (unsigned long)x;
1172 static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1173 unsigned long highmem,
1174 unsigned long total)
1176 unsigned long alloc = __fraction(nr_pages, highmem, total);
1178 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1180 #else /* CONFIG_HIGHMEM */
1181 static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1186 static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1187 unsigned long highmem,
1188 unsigned long total)
1192 #endif /* CONFIG_HIGHMEM */
1195 * free_unnecessary_pages - Release preallocated pages not needed for the image
1197 static void free_unnecessary_pages(void)
1199 unsigned long save, to_free_normal, to_free_highmem;
1201 save = count_data_pages();
1202 if (alloc_normal >= save) {
1203 to_free_normal = alloc_normal - save;
1207 save -= alloc_normal;
1209 save += count_highmem_pages();
1210 if (alloc_highmem >= save) {
1211 to_free_highmem = alloc_highmem - save;
1213 to_free_highmem = 0;
1214 save -= alloc_highmem;
1215 if (to_free_normal > save)
1216 to_free_normal -= save;
1221 memory_bm_position_reset(©_bm);
1223 while (to_free_normal > 0 || to_free_highmem > 0) {
1224 unsigned long pfn = memory_bm_next_pfn(©_bm);
1225 struct page *page = pfn_to_page(pfn);
1227 if (PageHighMem(page)) {
1228 if (!to_free_highmem)
1233 if (!to_free_normal)
1238 memory_bm_clear_bit(©_bm, pfn);
1239 swsusp_unset_page_forbidden(page);
1240 swsusp_unset_page_free(page);
1246 * minimum_image_size - Estimate the minimum acceptable size of an image
1247 * @saveable: Number of saveable pages in the system.
1249 * We want to avoid attempting to free too much memory too hard, so estimate the
1250 * minimum acceptable size of a hibernation image to use as the lower limit for
1251 * preallocating memory.
1253 * We assume that the minimum image size should be proportional to
1255 * [number of saveable pages] - [number of pages that can be freed in theory]
1257 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1258 * and (3) inactive anonymouns pages, (4) active and (5) inactive file pages,
1259 * minus mapped file pages.
1261 static unsigned long minimum_image_size(unsigned long saveable)
1265 size = global_page_state(NR_SLAB_RECLAIMABLE)
1266 + global_page_state(NR_ACTIVE_ANON)
1267 + global_page_state(NR_INACTIVE_ANON)
1268 + global_page_state(NR_ACTIVE_FILE)
1269 + global_page_state(NR_INACTIVE_FILE)
1270 - global_page_state(NR_FILE_MAPPED);
1272 return saveable <= size ? 0 : saveable - size;
1276 * hibernate_preallocate_memory - Preallocate memory for hibernation image
1278 * To create a hibernation image it is necessary to make a copy of every page
1279 * frame in use. We also need a number of page frames to be free during
1280 * hibernation for allocations made while saving the image and for device
1281 * drivers, in case they need to allocate memory from their hibernation
1282 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1283 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1284 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1285 * total number of available page frames and allocate at least
1287 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1288 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1290 * of them, which corresponds to the maximum size of a hibernation image.
1292 * If image_size is set below the number following from the above formula,
1293 * the preallocation of memory is continued until the total number of saveable
1294 * pages in the system is below the requested image size or the minimum
1295 * acceptable image size returned by minimum_image_size(), whichever is greater.
1297 int hibernate_preallocate_memory(void)
1300 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1301 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1302 struct timeval start, stop;
1305 printk(KERN_INFO "PM: Preallocating image memory... ");
1306 do_gettimeofday(&start);
1308 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1312 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1319 /* Count the number of saveable data pages. */
1320 save_highmem = count_highmem_pages();
1321 saveable = count_data_pages();
1324 * Compute the total number of page frames we can use (count) and the
1325 * number of pages needed for image metadata (size).
1328 saveable += save_highmem;
1329 highmem = save_highmem;
1331 for_each_populated_zone(zone) {
1332 size += snapshot_additional_pages(zone);
1333 if (is_highmem(zone))
1334 highmem += zone_page_state(zone, NR_FREE_PAGES);
1336 count += zone_page_state(zone, NR_FREE_PAGES);
1338 avail_normal = count;
1340 count -= totalreserve_pages;
1342 /* Add number of pages required for page keys (s390 only). */
1343 size += page_key_additional_pages(saveable);
1345 /* Compute the maximum number of saveable pages to leave in memory. */
1346 max_size = (count - (size + PAGES_FOR_IO)) / 2
1347 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1348 /* Compute the desired number of image pages specified by image_size. */
1349 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1350 if (size > max_size)
1353 * If the desired number of image pages is at least as large as the
1354 * current number of saveable pages in memory, allocate page frames for
1355 * the image and we're done.
1357 if (size >= saveable) {
1358 pages = preallocate_image_highmem(save_highmem);
1359 pages += preallocate_image_memory(saveable - pages, avail_normal);
1363 /* Estimate the minimum size of the image. */
1364 pages = minimum_image_size(saveable);
1366 * To avoid excessive pressure on the normal zone, leave room in it to
1367 * accommodate an image of the minimum size (unless it's already too
1368 * small, in which case don't preallocate pages from it at all).
1370 if (avail_normal > pages)
1371 avail_normal -= pages;
1375 size = min_t(unsigned long, pages, max_size);
1378 * Let the memory management subsystem know that we're going to need a
1379 * large number of page frames to allocate and make it free some memory.
1380 * NOTE: If this is not done, performance will be hurt badly in some
1383 shrink_all_memory(saveable - size);
1386 * The number of saveable pages in memory was too high, so apply some
1387 * pressure to decrease it. First, make room for the largest possible
1388 * image and fail if that doesn't work. Next, try to decrease the size
1389 * of the image as much as indicated by 'size' using allocations from
1390 * highmem and non-highmem zones separately.
1392 pages_highmem = preallocate_image_highmem(highmem / 2);
1393 alloc = count - max_size;
1394 if (alloc > pages_highmem)
1395 alloc -= pages_highmem;
1398 pages = preallocate_image_memory(alloc, avail_normal);
1399 if (pages < alloc) {
1400 /* We have exhausted non-highmem pages, try highmem. */
1402 pages += pages_highmem;
1403 pages_highmem = preallocate_image_highmem(alloc);
1404 if (pages_highmem < alloc)
1406 pages += pages_highmem;
1408 * size is the desired number of saveable pages to leave in
1409 * memory, so try to preallocate (all memory - size) pages.
1411 alloc = (count - pages) - size;
1412 pages += preallocate_image_highmem(alloc);
1415 * There are approximately max_size saveable pages at this point
1416 * and we want to reduce this number down to size.
1418 alloc = max_size - size;
1419 size = preallocate_highmem_fraction(alloc, highmem, count);
1420 pages_highmem += size;
1422 size = preallocate_image_memory(alloc, avail_normal);
1423 pages_highmem += preallocate_image_highmem(alloc - size);
1424 pages += pages_highmem + size;
1428 * We only need as many page frames for the image as there are saveable
1429 * pages in memory, but we have allocated more. Release the excessive
1432 free_unnecessary_pages();
1435 do_gettimeofday(&stop);
1436 printk(KERN_CONT "done (allocated %lu pages)\n", pages);
1437 swsusp_show_speed(&start, &stop, pages, "Allocated");
1442 printk(KERN_CONT "\n");
1447 #ifdef CONFIG_HIGHMEM
1449 * count_pages_for_highmem - compute the number of non-highmem pages
1450 * that will be necessary for creating copies of highmem pages.
1453 static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1455 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1457 if (free_highmem >= nr_highmem)
1460 nr_highmem -= free_highmem;
1466 count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1467 #endif /* CONFIG_HIGHMEM */
1470 * enough_free_mem - Make sure we have enough free memory for the
1474 static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1477 unsigned int free = alloc_normal;
1479 for_each_populated_zone(zone)
1480 if (!is_highmem(zone))
1481 free += zone_page_state(zone, NR_FREE_PAGES);
1483 nr_pages += count_pages_for_highmem(nr_highmem);
1484 pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n",
1485 nr_pages, PAGES_FOR_IO, free);
1487 return free > nr_pages + PAGES_FOR_IO;
1490 #ifdef CONFIG_HIGHMEM
1492 * get_highmem_buffer - if there are some highmem pages in the suspend
1493 * image, we may need the buffer to copy them and/or load their data.
1496 static inline int get_highmem_buffer(int safe_needed)
1498 buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
1499 return buffer ? 0 : -ENOMEM;
1503 * alloc_highmem_image_pages - allocate some highmem pages for the image.
1504 * Try to allocate as many pages as needed, but if the number of free
1505 * highmem pages is lesser than that, allocate them all.
1508 static inline unsigned int
1509 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int nr_highmem)
1511 unsigned int to_alloc = count_free_highmem_pages();
1513 if (to_alloc > nr_highmem)
1514 to_alloc = nr_highmem;
1516 nr_highmem -= to_alloc;
1517 while (to_alloc-- > 0) {
1520 page = alloc_image_page(__GFP_HIGHMEM);
1521 memory_bm_set_bit(bm, page_to_pfn(page));
1526 static inline int get_highmem_buffer(int safe_needed) { return 0; }
1528 static inline unsigned int
1529 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int n) { return 0; }
1530 #endif /* CONFIG_HIGHMEM */
1533 * swsusp_alloc - allocate memory for the suspend image
1535 * We first try to allocate as many highmem pages as there are
1536 * saveable highmem pages in the system. If that fails, we allocate
1537 * non-highmem pages for the copies of the remaining highmem ones.
1539 * In this approach it is likely that the copies of highmem pages will
1540 * also be located in the high memory, because of the way in which
1541 * copy_data_pages() works.
1545 swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm,
1546 unsigned int nr_pages, unsigned int nr_highmem)
1548 if (nr_highmem > 0) {
1549 if (get_highmem_buffer(PG_ANY))
1551 if (nr_highmem > alloc_highmem) {
1552 nr_highmem -= alloc_highmem;
1553 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1556 if (nr_pages > alloc_normal) {
1557 nr_pages -= alloc_normal;
1558 while (nr_pages-- > 0) {
1561 page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
1564 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1575 asmlinkage int swsusp_save(void)
1577 unsigned int nr_pages, nr_highmem;
1579 printk(KERN_INFO "PM: Creating hibernation image:\n");
1581 drain_local_pages(NULL);
1582 nr_pages = count_data_pages();
1583 nr_highmem = count_highmem_pages();
1584 printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem);
1586 if (!enough_free_mem(nr_pages, nr_highmem)) {
1587 printk(KERN_ERR "PM: Not enough free memory\n");
1591 if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) {
1592 printk(KERN_ERR "PM: Memory allocation failed\n");
1596 /* During allocating of suspend pagedir, new cold pages may appear.
1599 drain_local_pages(NULL);
1600 copy_data_pages(©_bm, &orig_bm);
1603 * End of critical section. From now on, we can write to memory,
1604 * but we should not touch disk. This specially means we must _not_
1605 * touch swap space! Except we must write out our image of course.
1608 nr_pages += nr_highmem;
1609 nr_copy_pages = nr_pages;
1610 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1612 printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n",
1618 #ifndef CONFIG_ARCH_HIBERNATION_HEADER
1619 static int init_header_complete(struct swsusp_info *info)
1621 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
1622 info->version_code = LINUX_VERSION_CODE;
1626 static char *check_image_kernel(struct swsusp_info *info)
1628 if (info->version_code != LINUX_VERSION_CODE)
1629 return "kernel version";
1630 if (strcmp(info->uts.sysname,init_utsname()->sysname))
1631 return "system type";
1632 if (strcmp(info->uts.release,init_utsname()->release))
1633 return "kernel release";
1634 if (strcmp(info->uts.version,init_utsname()->version))
1636 if (strcmp(info->uts.machine,init_utsname()->machine))
1640 #endif /* CONFIG_ARCH_HIBERNATION_HEADER */
1642 unsigned long snapshot_get_image_size(void)
1644 return nr_copy_pages + nr_meta_pages + 1;
1647 static int init_header(struct swsusp_info *info)
1649 memset(info, 0, sizeof(struct swsusp_info));
1650 info->num_physpages = num_physpages;
1651 info->image_pages = nr_copy_pages;
1652 info->pages = snapshot_get_image_size();
1653 info->size = info->pages;
1654 info->size <<= PAGE_SHIFT;
1655 return init_header_complete(info);
1659 * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm
1660 * are stored in the array @buf[] (1 page at a time)
1664 pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
1668 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1669 buf[j] = memory_bm_next_pfn(bm);
1670 if (unlikely(buf[j] == BM_END_OF_MAP))
1672 /* Save page key for data page (s390 only). */
1673 page_key_read(buf + j);
1678 * snapshot_read_next - used for reading the system memory snapshot.
1680 * On the first call to it @handle should point to a zeroed
1681 * snapshot_handle structure. The structure gets updated and a pointer
1682 * to it should be passed to this function every next time.
1684 * On success the function returns a positive number. Then, the caller
1685 * is allowed to read up to the returned number of bytes from the memory
1686 * location computed by the data_of() macro.
1688 * The function returns 0 to indicate the end of data stream condition,
1689 * and a negative number is returned on error. In such cases the
1690 * structure pointed to by @handle is not updated and should not be used
1694 int snapshot_read_next(struct snapshot_handle *handle)
1696 if (handle->cur > nr_meta_pages + nr_copy_pages)
1700 /* This makes the buffer be freed by swsusp_free() */
1701 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
1708 error = init_header((struct swsusp_info *)buffer);
1711 handle->buffer = buffer;
1712 memory_bm_position_reset(&orig_bm);
1713 memory_bm_position_reset(©_bm);
1714 } else if (handle->cur <= nr_meta_pages) {
1716 pack_pfns(buffer, &orig_bm);
1720 page = pfn_to_page(memory_bm_next_pfn(©_bm));
1721 if (PageHighMem(page)) {
1722 /* Highmem pages are copied to the buffer,
1723 * because we can't return with a kmapped
1724 * highmem page (we may not be called again).
1728 kaddr = kmap_atomic(page, KM_USER0);
1729 copy_page(buffer, kaddr);
1730 kunmap_atomic(kaddr, KM_USER0);
1731 handle->buffer = buffer;
1733 handle->buffer = page_address(page);
1741 * mark_unsafe_pages - mark the pages that cannot be used for storing
1742 * the image during resume, because they conflict with the pages that
1743 * had been used before suspend
1746 static int mark_unsafe_pages(struct memory_bitmap *bm)
1749 unsigned long pfn, max_zone_pfn;
1751 /* Clear page flags */
1752 for_each_populated_zone(zone) {
1753 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1754 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1756 swsusp_unset_page_free(pfn_to_page(pfn));
1759 /* Mark pages that correspond to the "original" pfns as "unsafe" */
1760 memory_bm_position_reset(bm);
1762 pfn = memory_bm_next_pfn(bm);
1763 if (likely(pfn != BM_END_OF_MAP)) {
1764 if (likely(pfn_valid(pfn)))
1765 swsusp_set_page_free(pfn_to_page(pfn));
1769 } while (pfn != BM_END_OF_MAP);
1771 allocated_unsafe_pages = 0;
1777 duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src)
1781 memory_bm_position_reset(src);
1782 pfn = memory_bm_next_pfn(src);
1783 while (pfn != BM_END_OF_MAP) {
1784 memory_bm_set_bit(dst, pfn);
1785 pfn = memory_bm_next_pfn(src);
1789 static int check_header(struct swsusp_info *info)
1793 reason = check_image_kernel(info);
1794 if (!reason && info->num_physpages != num_physpages)
1795 reason = "memory size";
1797 printk(KERN_ERR "PM: Image mismatch: %s\n", reason);
1804 * load header - check the image header and copy data from it
1808 load_header(struct swsusp_info *info)
1812 restore_pblist = NULL;
1813 error = check_header(info);
1815 nr_copy_pages = info->image_pages;
1816 nr_meta_pages = info->pages - info->image_pages - 1;
1822 * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set
1823 * the corresponding bit in the memory bitmap @bm
1825 static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
1829 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1830 if (unlikely(buf[j] == BM_END_OF_MAP))
1833 /* Extract and buffer page key for data page (s390 only). */
1834 page_key_memorize(buf + j);
1836 if (memory_bm_pfn_present(bm, buf[j]))
1837 memory_bm_set_bit(bm, buf[j]);
1845 /* List of "safe" pages that may be used to store data loaded from the suspend
1848 static struct linked_page *safe_pages_list;
1850 #ifdef CONFIG_HIGHMEM
1851 /* struct highmem_pbe is used for creating the list of highmem pages that
1852 * should be restored atomically during the resume from disk, because the page
1853 * frames they have occupied before the suspend are in use.
1855 struct highmem_pbe {
1856 struct page *copy_page; /* data is here now */
1857 struct page *orig_page; /* data was here before the suspend */
1858 struct highmem_pbe *next;
1861 /* List of highmem PBEs needed for restoring the highmem pages that were
1862 * allocated before the suspend and included in the suspend image, but have
1863 * also been allocated by the "resume" kernel, so their contents cannot be
1864 * written directly to their "original" page frames.
1866 static struct highmem_pbe *highmem_pblist;
1869 * count_highmem_image_pages - compute the number of highmem pages in the
1870 * suspend image. The bits in the memory bitmap @bm that correspond to the
1871 * image pages are assumed to be set.
1874 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
1877 unsigned int cnt = 0;
1879 memory_bm_position_reset(bm);
1880 pfn = memory_bm_next_pfn(bm);
1881 while (pfn != BM_END_OF_MAP) {
1882 if (PageHighMem(pfn_to_page(pfn)))
1885 pfn = memory_bm_next_pfn(bm);
1891 * prepare_highmem_image - try to allocate as many highmem pages as
1892 * there are highmem image pages (@nr_highmem_p points to the variable
1893 * containing the number of highmem image pages). The pages that are
1894 * "safe" (ie. will not be overwritten when the suspend image is
1895 * restored) have the corresponding bits set in @bm (it must be
1898 * NOTE: This function should not be called if there are no highmem
1902 static unsigned int safe_highmem_pages;
1904 static struct memory_bitmap *safe_highmem_bm;
1907 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
1909 unsigned int to_alloc;
1911 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
1914 if (get_highmem_buffer(PG_SAFE))
1917 to_alloc = count_free_highmem_pages();
1918 if (to_alloc > *nr_highmem_p)
1919 to_alloc = *nr_highmem_p;
1921 *nr_highmem_p = to_alloc;
1923 safe_highmem_pages = 0;
1924 while (to_alloc-- > 0) {
1927 page = alloc_page(__GFP_HIGHMEM);
1928 if (!swsusp_page_is_free(page)) {
1929 /* The page is "safe", set its bit the bitmap */
1930 memory_bm_set_bit(bm, page_to_pfn(page));
1931 safe_highmem_pages++;
1933 /* Mark the page as allocated */
1934 swsusp_set_page_forbidden(page);
1935 swsusp_set_page_free(page);
1937 memory_bm_position_reset(bm);
1938 safe_highmem_bm = bm;
1943 * get_highmem_page_buffer - for given highmem image page find the buffer
1944 * that suspend_write_next() should set for its caller to write to.
1946 * If the page is to be saved to its "original" page frame or a copy of
1947 * the page is to be made in the highmem, @buffer is returned. Otherwise,
1948 * the copy of the page is to be made in normal memory, so the address of
1949 * the copy is returned.
1951 * If @buffer is returned, the caller of suspend_write_next() will write
1952 * the page's contents to @buffer, so they will have to be copied to the
1953 * right location on the next call to suspend_write_next() and it is done
1954 * with the help of copy_last_highmem_page(). For this purpose, if
1955 * @buffer is returned, @last_highmem page is set to the page to which
1956 * the data will have to be copied from @buffer.
1959 static struct page *last_highmem_page;
1962 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
1964 struct highmem_pbe *pbe;
1967 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
1968 /* We have allocated the "original" page frame and we can
1969 * use it directly to store the loaded page.
1971 last_highmem_page = page;
1974 /* The "original" page frame has not been allocated and we have to
1975 * use a "safe" page frame to store the loaded page.
1977 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
1980 return ERR_PTR(-ENOMEM);
1982 pbe->orig_page = page;
1983 if (safe_highmem_pages > 0) {
1986 /* Copy of the page will be stored in high memory */
1988 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
1989 safe_highmem_pages--;
1990 last_highmem_page = tmp;
1991 pbe->copy_page = tmp;
1993 /* Copy of the page will be stored in normal memory */
1994 kaddr = safe_pages_list;
1995 safe_pages_list = safe_pages_list->next;
1996 pbe->copy_page = virt_to_page(kaddr);
1998 pbe->next = highmem_pblist;
1999 highmem_pblist = pbe;
2004 * copy_last_highmem_page - copy the contents of a highmem image from
2005 * @buffer, where the caller of snapshot_write_next() has place them,
2006 * to the right location represented by @last_highmem_page .
2009 static void copy_last_highmem_page(void)
2011 if (last_highmem_page) {
2014 dst = kmap_atomic(last_highmem_page, KM_USER0);
2015 copy_page(dst, buffer);
2016 kunmap_atomic(dst, KM_USER0);
2017 last_highmem_page = NULL;
2021 static inline int last_highmem_page_copied(void)
2023 return !last_highmem_page;
2026 static inline void free_highmem_data(void)
2028 if (safe_highmem_bm)
2029 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2032 free_image_page(buffer, PG_UNSAFE_CLEAR);
2035 static inline int get_safe_write_buffer(void) { return 0; }
2038 count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2041 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
2046 static inline void *
2047 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
2049 return ERR_PTR(-EINVAL);
2052 static inline void copy_last_highmem_page(void) {}
2053 static inline int last_highmem_page_copied(void) { return 1; }
2054 static inline void free_highmem_data(void) {}
2055 #endif /* CONFIG_HIGHMEM */
2058 * prepare_image - use the memory bitmap @bm to mark the pages that will
2059 * be overwritten in the process of restoring the system memory state
2060 * from the suspend image ("unsafe" pages) and allocate memory for the
2063 * The idea is to allocate a new memory bitmap first and then allocate
2064 * as many pages as needed for the image data, but not to assign these
2065 * pages to specific tasks initially. Instead, we just mark them as
2066 * allocated and create a lists of "safe" pages that will be used
2067 * later. On systems with high memory a list of "safe" highmem pages is
2071 #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2074 prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2076 unsigned int nr_pages, nr_highmem;
2077 struct linked_page *sp_list, *lp;
2080 /* If there is no highmem, the buffer will not be necessary */
2081 free_image_page(buffer, PG_UNSAFE_CLEAR);
2084 nr_highmem = count_highmem_image_pages(bm);
2085 error = mark_unsafe_pages(bm);
2089 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2093 duplicate_memory_bitmap(new_bm, bm);
2094 memory_bm_free(bm, PG_UNSAFE_KEEP);
2095 if (nr_highmem > 0) {
2096 error = prepare_highmem_image(bm, &nr_highmem);
2100 /* Reserve some safe pages for potential later use.
2102 * NOTE: This way we make sure there will be enough safe pages for the
2103 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2104 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2107 /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */
2108 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2109 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2110 while (nr_pages > 0) {
2111 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2120 /* Preallocate memory for the image */
2121 safe_pages_list = NULL;
2122 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2123 while (nr_pages > 0) {
2124 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2129 if (!swsusp_page_is_free(virt_to_page(lp))) {
2130 /* The page is "safe", add it to the list */
2131 lp->next = safe_pages_list;
2132 safe_pages_list = lp;
2134 /* Mark the page as allocated */
2135 swsusp_set_page_forbidden(virt_to_page(lp));
2136 swsusp_set_page_free(virt_to_page(lp));
2139 /* Free the reserved safe pages so that chain_alloc() can use them */
2142 free_image_page(sp_list, PG_UNSAFE_CLEAR);
2153 * get_buffer - compute the address that snapshot_write_next() should
2154 * set for its caller to write to.
2157 static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2161 unsigned long pfn = memory_bm_next_pfn(bm);
2163 if (pfn == BM_END_OF_MAP)
2164 return ERR_PTR(-EFAULT);
2166 page = pfn_to_page(pfn);
2167 if (PageHighMem(page))
2168 return get_highmem_page_buffer(page, ca);
2170 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2171 /* We have allocated the "original" page frame and we can
2172 * use it directly to store the loaded page.
2174 return page_address(page);
2176 /* The "original" page frame has not been allocated and we have to
2177 * use a "safe" page frame to store the loaded page.
2179 pbe = chain_alloc(ca, sizeof(struct pbe));
2182 return ERR_PTR(-ENOMEM);
2184 pbe->orig_address = page_address(page);
2185 pbe->address = safe_pages_list;
2186 safe_pages_list = safe_pages_list->next;
2187 pbe->next = restore_pblist;
2188 restore_pblist = pbe;
2189 return pbe->address;
2193 * snapshot_write_next - used for writing the system memory snapshot.
2195 * On the first call to it @handle should point to a zeroed
2196 * snapshot_handle structure. The structure gets updated and a pointer
2197 * to it should be passed to this function every next time.
2199 * On success the function returns a positive number. Then, the caller
2200 * is allowed to write up to the returned number of bytes to the memory
2201 * location computed by the data_of() macro.
2203 * The function returns 0 to indicate the "end of file" condition,
2204 * and a negative number is returned on error. In such cases the
2205 * structure pointed to by @handle is not updated and should not be used
2209 int snapshot_write_next(struct snapshot_handle *handle)
2211 static struct chain_allocator ca;
2214 /* Check if we have already loaded the entire image */
2215 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2218 handle->sync_read = 1;
2222 /* This makes the buffer be freed by swsusp_free() */
2223 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2228 handle->buffer = buffer;
2229 } else if (handle->cur == 1) {
2230 error = load_header(buffer);
2234 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2238 /* Allocate buffer for page keys. */
2239 error = page_key_alloc(nr_copy_pages);
2243 } else if (handle->cur <= nr_meta_pages + 1) {
2244 error = unpack_orig_pfns(buffer, ©_bm);
2248 if (handle->cur == nr_meta_pages + 1) {
2249 error = prepare_image(&orig_bm, ©_bm);
2253 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2254 memory_bm_position_reset(&orig_bm);
2255 restore_pblist = NULL;
2256 handle->buffer = get_buffer(&orig_bm, &ca);
2257 handle->sync_read = 0;
2258 if (IS_ERR(handle->buffer))
2259 return PTR_ERR(handle->buffer);
2262 copy_last_highmem_page();
2263 /* Restore page key for data page (s390 only). */
2264 page_key_write(handle->buffer);
2265 handle->buffer = get_buffer(&orig_bm, &ca);
2266 if (IS_ERR(handle->buffer))
2267 return PTR_ERR(handle->buffer);
2268 if (handle->buffer != buffer)
2269 handle->sync_read = 0;
2276 * snapshot_write_finalize - must be called after the last call to
2277 * snapshot_write_next() in case the last page in the image happens
2278 * to be a highmem page and its contents should be stored in the
2279 * highmem. Additionally, it releases the memory that will not be
2283 void snapshot_write_finalize(struct snapshot_handle *handle)
2285 copy_last_highmem_page();
2286 /* Restore page key for data page (s390 only). */
2287 page_key_write(handle->buffer);
2289 /* Free only if we have loaded the image entirely */
2290 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2291 memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR);
2292 free_highmem_data();
2296 int snapshot_image_loaded(struct snapshot_handle *handle)
2298 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2299 handle->cur <= nr_meta_pages + nr_copy_pages);
2302 #ifdef CONFIG_HIGHMEM
2303 /* Assumes that @buf is ready and points to a "safe" page */
2305 swap_two_pages_data(struct page *p1, struct page *p2, void *buf)
2307 void *kaddr1, *kaddr2;
2309 kaddr1 = kmap_atomic(p1, KM_USER0);
2310 kaddr2 = kmap_atomic(p2, KM_USER1);
2311 copy_page(buf, kaddr1);
2312 copy_page(kaddr1, kaddr2);
2313 copy_page(kaddr2, buf);
2314 kunmap_atomic(kaddr2, KM_USER1);
2315 kunmap_atomic(kaddr1, KM_USER0);
2319 * restore_highmem - for each highmem page that was allocated before
2320 * the suspend and included in the suspend image, and also has been
2321 * allocated by the "resume" kernel swap its current (ie. "before
2322 * resume") contents with the previous (ie. "before suspend") one.
2324 * If the resume eventually fails, we can call this function once
2325 * again and restore the "before resume" highmem state.
2328 int restore_highmem(void)
2330 struct highmem_pbe *pbe = highmem_pblist;
2336 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2341 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2344 free_image_page(buf, PG_UNSAFE_CLEAR);
2347 #endif /* CONFIG_HIGHMEM */