2 * linux/kernel/power/snapshot.c
4 * This file provides system snapshot/restore functionality for swsusp.
6 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
7 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
9 * This file is released under the GPLv2.
13 #include <linux/version.h>
14 #include <linux/module.h>
16 #include <linux/suspend.h>
17 #include <linux/delay.h>
18 #include <linux/bitops.h>
19 #include <linux/spinlock.h>
20 #include <linux/kernel.h>
22 #include <linux/device.h>
23 #include <linux/init.h>
24 #include <linux/bootmem.h>
25 #include <linux/syscalls.h>
26 #include <linux/console.h>
27 #include <linux/highmem.h>
28 #include <linux/list.h>
29 #include <linux/slab.h>
30 #include <linux/compiler.h>
31 #include <linux/ktime.h>
33 #include <asm/uaccess.h>
34 #include <asm/mmu_context.h>
35 #include <asm/pgtable.h>
36 #include <asm/tlbflush.h>
41 static int swsusp_page_is_free(struct page *);
42 static void swsusp_set_page_forbidden(struct page *);
43 static void swsusp_unset_page_forbidden(struct page *);
46 * Number of bytes to reserve for memory allocations made by device drivers
47 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
48 * cause image creation to fail (tunable via /sys/power/reserved_size).
50 unsigned long reserved_size;
52 void __init hibernate_reserved_size_init(void)
54 reserved_size = SPARE_PAGES * PAGE_SIZE;
58 * Preferred image size in bytes (tunable via /sys/power/image_size).
59 * When it is set to N, swsusp will do its best to ensure the image
60 * size will not exceed N bytes, but if that is impossible, it will
61 * try to create the smallest image possible.
63 unsigned long image_size;
65 void __init hibernate_image_size_init(void)
67 image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE;
70 /* List of PBEs needed for restoring the pages that were allocated before
71 * the suspend and included in the suspend image, but have also been
72 * allocated by the "resume" kernel, so their contents cannot be written
73 * directly to their "original" page frames.
75 struct pbe *restore_pblist;
77 /* struct linked_page is used to build chains of pages */
79 #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
82 struct linked_page *next;
83 char data[LINKED_PAGE_DATA_SIZE];
87 * List of "safe" pages (ie. pages that were not used by the image kernel
88 * before hibernation) that may be used as temporary storage for image kernel
91 static struct linked_page *safe_pages_list;
93 /* Pointer to an auxiliary buffer (1 page) */
97 * @safe_needed - on resume, for storing the PBE list and the image,
98 * we can only use memory pages that do not conflict with the pages
99 * used before suspend. The unsafe pages have PageNosaveFree set
100 * and we count them using unsafe_pages.
102 * Each allocated image page is marked as PageNosave and PageNosaveFree
103 * so that swsusp_free() can release it.
108 #define PG_UNSAFE_CLEAR 1
109 #define PG_UNSAFE_KEEP 0
111 static unsigned int allocated_unsafe_pages;
113 static void *get_image_page(gfp_t gfp_mask, int safe_needed)
117 res = (void *)get_zeroed_page(gfp_mask);
119 while (res && swsusp_page_is_free(virt_to_page(res))) {
120 /* The page is unsafe, mark it for swsusp_free() */
121 swsusp_set_page_forbidden(virt_to_page(res));
122 allocated_unsafe_pages++;
123 res = (void *)get_zeroed_page(gfp_mask);
126 swsusp_set_page_forbidden(virt_to_page(res));
127 swsusp_set_page_free(virt_to_page(res));
132 static void *__get_safe_page(gfp_t gfp_mask)
134 if (safe_pages_list) {
135 void *ret = safe_pages_list;
137 safe_pages_list = safe_pages_list->next;
138 memset(ret, 0, PAGE_SIZE);
141 return get_image_page(gfp_mask, PG_SAFE);
144 unsigned long get_safe_page(gfp_t gfp_mask)
146 return (unsigned long)__get_safe_page(gfp_mask);
149 static struct page *alloc_image_page(gfp_t gfp_mask)
153 page = alloc_page(gfp_mask);
155 swsusp_set_page_forbidden(page);
156 swsusp_set_page_free(page);
161 static void recycle_safe_page(void *page_address)
163 struct linked_page *lp = page_address;
165 lp->next = safe_pages_list;
166 safe_pages_list = lp;
170 * free_image_page - free page represented by @addr, allocated with
171 * get_image_page (page flags set by it must be cleared)
174 static inline void free_image_page(void *addr, int clear_nosave_free)
178 BUG_ON(!virt_addr_valid(addr));
180 page = virt_to_page(addr);
182 swsusp_unset_page_forbidden(page);
183 if (clear_nosave_free)
184 swsusp_unset_page_free(page);
189 static inline void free_list_of_pages(struct linked_page *list,
190 int clear_page_nosave)
193 struct linked_page *lp = list->next;
195 free_image_page(list, clear_page_nosave);
201 * struct chain_allocator is used for allocating small objects out of
202 * a linked list of pages called 'the chain'.
204 * The chain grows each time when there is no room for a new object in
205 * the current page. The allocated objects cannot be freed individually.
206 * It is only possible to free them all at once, by freeing the entire
209 * NOTE: The chain allocator may be inefficient if the allocated objects
210 * are not much smaller than PAGE_SIZE.
213 struct chain_allocator {
214 struct linked_page *chain; /* the chain */
215 unsigned int used_space; /* total size of objects allocated out
216 * of the current page
218 gfp_t gfp_mask; /* mask for allocating pages */
219 int safe_needed; /* if set, only "safe" pages are allocated */
222 static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
226 ca->used_space = LINKED_PAGE_DATA_SIZE;
227 ca->gfp_mask = gfp_mask;
228 ca->safe_needed = safe_needed;
231 static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
235 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
236 struct linked_page *lp;
238 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
239 get_image_page(ca->gfp_mask, PG_ANY);
243 lp->next = ca->chain;
247 ret = ca->chain->data + ca->used_space;
248 ca->used_space += size;
253 * Data types related to memory bitmaps.
255 * Memory bitmap is a structure consiting of many linked lists of
256 * objects. The main list's elements are of type struct zone_bitmap
257 * and each of them corresonds to one zone. For each zone bitmap
258 * object there is a list of objects of type struct bm_block that
259 * represent each blocks of bitmap in which information is stored.
261 * struct memory_bitmap contains a pointer to the main list of zone
262 * bitmap objects, a struct bm_position used for browsing the bitmap,
263 * and a pointer to the list of pages used for allocating all of the
264 * zone bitmap objects and bitmap block objects.
266 * NOTE: It has to be possible to lay out the bitmap in memory
267 * using only allocations of order 0. Additionally, the bitmap is
268 * designed to work with arbitrary number of zones (this is over the
269 * top for now, but let's avoid making unnecessary assumptions ;-).
271 * struct zone_bitmap contains a pointer to a list of bitmap block
272 * objects and a pointer to the bitmap block object that has been
273 * most recently used for setting bits. Additionally, it contains the
274 * pfns that correspond to the start and end of the represented zone.
276 * struct bm_block contains a pointer to the memory page in which
277 * information is stored (in the form of a block of bitmap)
278 * It also contains the pfns that correspond to the start and end of
279 * the represented memory area.
281 * The memory bitmap is organized as a radix tree to guarantee fast random
282 * access to the bits. There is one radix tree for each zone (as returned
283 * from create_mem_extents).
285 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
286 * two linked lists for the nodes of the tree, one for the inner nodes and
287 * one for the leave nodes. The linked leave nodes are used for fast linear
288 * access of the memory bitmap.
290 * The struct rtree_node represents one node of the radix tree.
293 #define BM_END_OF_MAP (~0UL)
295 #define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
296 #define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
297 #define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
300 * struct rtree_node is a wrapper struct to link the nodes
301 * of the rtree together for easy linear iteration over
302 * bits and easy freeing
305 struct list_head list;
310 * struct mem_zone_bm_rtree represents a bitmap used for one
311 * populated memory zone.
313 struct mem_zone_bm_rtree {
314 struct list_head list; /* Link Zones together */
315 struct list_head nodes; /* Radix Tree inner nodes */
316 struct list_head leaves; /* Radix Tree leaves */
317 unsigned long start_pfn; /* Zone start page frame */
318 unsigned long end_pfn; /* Zone end page frame + 1 */
319 struct rtree_node *rtree; /* Radix Tree Root */
320 int levels; /* Number of Radix Tree Levels */
321 unsigned int blocks; /* Number of Bitmap Blocks */
324 /* strcut bm_position is used for browsing memory bitmaps */
327 struct mem_zone_bm_rtree *zone;
328 struct rtree_node *node;
329 unsigned long node_pfn;
333 struct memory_bitmap {
334 struct list_head zones;
335 struct linked_page *p_list; /* list of pages used to store zone
336 * bitmap objects and bitmap block
339 struct bm_position cur; /* most recently used bit position */
342 /* Functions that operate on memory bitmaps */
344 #define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
345 #if BITS_PER_LONG == 32
346 #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
348 #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
350 #define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
353 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
355 * This function is used to allocate inner nodes as well as the
356 * leave nodes of the radix tree. It also adds the node to the
357 * corresponding linked list passed in by the *list parameter.
359 static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
360 struct chain_allocator *ca,
361 struct list_head *list)
363 struct rtree_node *node;
365 node = chain_alloc(ca, sizeof(struct rtree_node));
369 node->data = get_image_page(gfp_mask, safe_needed);
373 list_add_tail(&node->list, list);
379 * add_rtree_block - Add a new leave node to the radix tree
381 * The leave nodes need to be allocated in order to keep the leaves
382 * linked list in order. This is guaranteed by the zone->blocks
385 static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
386 int safe_needed, struct chain_allocator *ca)
388 struct rtree_node *node, *block, **dst;
389 unsigned int levels_needed, block_nr;
392 block_nr = zone->blocks;
395 /* How many levels do we need for this block nr? */
398 block_nr >>= BM_RTREE_LEVEL_SHIFT;
401 /* Make sure the rtree has enough levels */
402 for (i = zone->levels; i < levels_needed; i++) {
403 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
408 node->data[0] = (unsigned long)zone->rtree;
413 /* Allocate new block */
414 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
418 /* Now walk the rtree to insert the block */
421 block_nr = zone->blocks;
422 for (i = zone->levels; i > 0; i--) {
426 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
433 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
434 index &= BM_RTREE_LEVEL_MASK;
435 dst = (struct rtree_node **)&((*dst)->data[index]);
445 static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
446 int clear_nosave_free);
449 * create_zone_bm_rtree - create a radix tree for one zone
451 * Allocated the mem_zone_bm_rtree structure and initializes it.
452 * This function also allocated and builds the radix tree for the
455 static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
457 struct chain_allocator *ca,
461 struct mem_zone_bm_rtree *zone;
462 unsigned int i, nr_blocks;
466 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
470 INIT_LIST_HEAD(&zone->nodes);
471 INIT_LIST_HEAD(&zone->leaves);
472 zone->start_pfn = start;
474 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
476 for (i = 0; i < nr_blocks; i++) {
477 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
478 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
487 * free_zone_bm_rtree - Free the memory of the radix tree
489 * Free all node pages of the radix tree. The mem_zone_bm_rtree
490 * structure itself is not freed here nor are the rtree_node
493 static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
494 int clear_nosave_free)
496 struct rtree_node *node;
498 list_for_each_entry(node, &zone->nodes, list)
499 free_image_page(node->data, clear_nosave_free);
501 list_for_each_entry(node, &zone->leaves, list)
502 free_image_page(node->data, clear_nosave_free);
505 static void memory_bm_position_reset(struct memory_bitmap *bm)
507 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
509 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
510 struct rtree_node, list);
511 bm->cur.node_pfn = 0;
512 bm->cur.node_bit = 0;
515 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
518 struct list_head hook;
524 * free_mem_extents - free a list of memory extents
525 * @list - list of extents to empty
527 static void free_mem_extents(struct list_head *list)
529 struct mem_extent *ext, *aux;
531 list_for_each_entry_safe(ext, aux, list, hook) {
532 list_del(&ext->hook);
538 * create_mem_extents - create a list of memory extents representing
539 * contiguous ranges of PFNs
540 * @list - list to put the extents into
541 * @gfp_mask - mask to use for memory allocations
543 static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
547 INIT_LIST_HEAD(list);
549 for_each_populated_zone(zone) {
550 unsigned long zone_start, zone_end;
551 struct mem_extent *ext, *cur, *aux;
553 zone_start = zone->zone_start_pfn;
554 zone_end = zone_end_pfn(zone);
556 list_for_each_entry(ext, list, hook)
557 if (zone_start <= ext->end)
560 if (&ext->hook == list || zone_end < ext->start) {
561 /* New extent is necessary */
562 struct mem_extent *new_ext;
564 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
566 free_mem_extents(list);
569 new_ext->start = zone_start;
570 new_ext->end = zone_end;
571 list_add_tail(&new_ext->hook, &ext->hook);
575 /* Merge this zone's range of PFNs with the existing one */
576 if (zone_start < ext->start)
577 ext->start = zone_start;
578 if (zone_end > ext->end)
581 /* More merging may be possible */
583 list_for_each_entry_safe_continue(cur, aux, list, hook) {
584 if (zone_end < cur->start)
586 if (zone_end < cur->end)
588 list_del(&cur->hook);
597 * memory_bm_create - allocate memory for a memory bitmap
599 static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
602 struct chain_allocator ca;
603 struct list_head mem_extents;
604 struct mem_extent *ext;
607 chain_init(&ca, gfp_mask, safe_needed);
608 INIT_LIST_HEAD(&bm->zones);
610 error = create_mem_extents(&mem_extents, gfp_mask);
614 list_for_each_entry(ext, &mem_extents, hook) {
615 struct mem_zone_bm_rtree *zone;
617 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
618 ext->start, ext->end);
623 list_add_tail(&zone->list, &bm->zones);
626 bm->p_list = ca.chain;
627 memory_bm_position_reset(bm);
629 free_mem_extents(&mem_extents);
633 bm->p_list = ca.chain;
634 memory_bm_free(bm, PG_UNSAFE_CLEAR);
639 * memory_bm_free - free memory occupied by the memory bitmap @bm
641 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
643 struct mem_zone_bm_rtree *zone;
645 list_for_each_entry(zone, &bm->zones, list)
646 free_zone_bm_rtree(zone, clear_nosave_free);
648 free_list_of_pages(bm->p_list, clear_nosave_free);
650 INIT_LIST_HEAD(&bm->zones);
654 * memory_bm_find_bit - Find the bit for pfn in the memory
657 * Find the bit in the bitmap @bm that corresponds to given pfn.
658 * The cur.zone, cur.block and cur.node_pfn member of @bm are
660 * It walks the radix tree to find the page which contains the bit for
661 * pfn and returns the bit position in **addr and *bit_nr.
663 static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
664 void **addr, unsigned int *bit_nr)
666 struct mem_zone_bm_rtree *curr, *zone;
667 struct rtree_node *node;
672 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
677 /* Find the right zone */
678 list_for_each_entry(curr, &bm->zones, list) {
679 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
690 * We have a zone. Now walk the radix tree to find the leave
695 if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
699 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
701 for (i = zone->levels; i > 0; i--) {
704 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
705 index &= BM_RTREE_LEVEL_MASK;
706 BUG_ON(node->data[index] == 0);
707 node = (struct rtree_node *)node->data[index];
711 /* Update last position */
714 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
716 /* Set return values */
718 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
723 static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
729 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
734 static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
740 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
747 static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
753 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
755 clear_bit(bit, addr);
758 static void memory_bm_clear_current(struct memory_bitmap *bm)
762 bit = max(bm->cur.node_bit - 1, 0);
763 clear_bit(bit, bm->cur.node->data);
766 static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
772 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
774 return test_bit(bit, addr);
777 static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
782 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
786 * rtree_next_node - Jumps to the next leave node
788 * Sets the position to the beginning of the next node in the
789 * memory bitmap. This is either the next node in the current
790 * zone's radix tree or the first node in the radix tree of the
793 * Returns true if there is a next node, false otherwise.
795 static bool rtree_next_node(struct memory_bitmap *bm)
797 bm->cur.node = list_entry(bm->cur.node->list.next,
798 struct rtree_node, list);
799 if (&bm->cur.node->list != &bm->cur.zone->leaves) {
800 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
801 bm->cur.node_bit = 0;
802 touch_softlockup_watchdog();
806 /* No more nodes, goto next zone */
807 bm->cur.zone = list_entry(bm->cur.zone->list.next,
808 struct mem_zone_bm_rtree, list);
809 if (&bm->cur.zone->list != &bm->zones) {
810 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
811 struct rtree_node, list);
812 bm->cur.node_pfn = 0;
813 bm->cur.node_bit = 0;
822 * memory_bm_rtree_next_pfn - Find the next set bit in the bitmap @bm
824 * Starting from the last returned position this function searches
825 * for the next set bit in the memory bitmap and returns its
826 * number. If no more bit is set BM_END_OF_MAP is returned.
828 * It is required to run memory_bm_position_reset() before the
829 * first call to this function.
831 static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
833 unsigned long bits, pfn, pages;
837 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
838 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
839 bit = find_next_bit(bm->cur.node->data, bits,
842 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
843 bm->cur.node_bit = bit + 1;
846 } while (rtree_next_node(bm));
848 return BM_END_OF_MAP;
852 * This structure represents a range of page frames the contents of which
853 * should not be saved during the suspend.
856 struct nosave_region {
857 struct list_head list;
858 unsigned long start_pfn;
859 unsigned long end_pfn;
862 static LIST_HEAD(nosave_regions);
864 static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
866 struct rtree_node *node;
868 list_for_each_entry(node, &zone->nodes, list)
869 recycle_safe_page(node->data);
871 list_for_each_entry(node, &zone->leaves, list)
872 recycle_safe_page(node->data);
875 static void memory_bm_recycle(struct memory_bitmap *bm)
877 struct mem_zone_bm_rtree *zone;
878 struct linked_page *p_list;
880 list_for_each_entry(zone, &bm->zones, list)
881 recycle_zone_bm_rtree(zone);
885 struct linked_page *lp = p_list;
888 recycle_safe_page(lp);
893 * register_nosave_region - register a range of page frames the contents
894 * of which should not be saved during the suspend (to be used in the early
895 * initialization code)
898 void __init __register_nosave_region(unsigned long start_pfn,
899 unsigned long end_pfn, int use_kmalloc)
901 struct nosave_region *region;
903 if (start_pfn >= end_pfn)
906 if (!list_empty(&nosave_regions)) {
907 /* Try to extend the previous region (they should be sorted) */
908 region = list_entry(nosave_regions.prev,
909 struct nosave_region, list);
910 if (region->end_pfn == start_pfn) {
911 region->end_pfn = end_pfn;
916 /* during init, this shouldn't fail */
917 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
920 /* This allocation cannot fail */
921 region = memblock_virt_alloc(sizeof(struct nosave_region), 0);
922 region->start_pfn = start_pfn;
923 region->end_pfn = end_pfn;
924 list_add_tail(®ion->list, &nosave_regions);
926 printk(KERN_INFO "PM: Registered nosave memory: [mem %#010llx-%#010llx]\n",
927 (unsigned long long) start_pfn << PAGE_SHIFT,
928 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
932 * Set bits in this map correspond to the page frames the contents of which
933 * should not be saved during the suspend.
935 static struct memory_bitmap *forbidden_pages_map;
937 /* Set bits in this map correspond to free page frames. */
938 static struct memory_bitmap *free_pages_map;
941 * Each page frame allocated for creating the image is marked by setting the
942 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
945 void swsusp_set_page_free(struct page *page)
948 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
951 static int swsusp_page_is_free(struct page *page)
953 return free_pages_map ?
954 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
957 void swsusp_unset_page_free(struct page *page)
960 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
963 static void swsusp_set_page_forbidden(struct page *page)
965 if (forbidden_pages_map)
966 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
969 int swsusp_page_is_forbidden(struct page *page)
971 return forbidden_pages_map ?
972 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
975 static void swsusp_unset_page_forbidden(struct page *page)
977 if (forbidden_pages_map)
978 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
982 * mark_nosave_pages - set bits corresponding to the page frames the
983 * contents of which should not be saved in a given bitmap.
986 static void mark_nosave_pages(struct memory_bitmap *bm)
988 struct nosave_region *region;
990 if (list_empty(&nosave_regions))
993 list_for_each_entry(region, &nosave_regions, list) {
996 pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n",
997 (unsigned long long) region->start_pfn << PAGE_SHIFT,
998 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1001 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1002 if (pfn_valid(pfn)) {
1004 * It is safe to ignore the result of
1005 * mem_bm_set_bit_check() here, since we won't
1006 * touch the PFNs for which the error is
1009 mem_bm_set_bit_check(bm, pfn);
1015 * create_basic_memory_bitmaps - create bitmaps needed for marking page
1016 * frames that should not be saved and free page frames. The pointers
1017 * forbidden_pages_map and free_pages_map are only modified if everything
1018 * goes well, because we don't want the bits to be used before both bitmaps
1022 int create_basic_memory_bitmaps(void)
1024 struct memory_bitmap *bm1, *bm2;
1027 if (forbidden_pages_map && free_pages_map)
1030 BUG_ON(forbidden_pages_map || free_pages_map);
1032 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1036 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1038 goto Free_first_object;
1040 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1042 goto Free_first_bitmap;
1044 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1046 goto Free_second_object;
1048 forbidden_pages_map = bm1;
1049 free_pages_map = bm2;
1050 mark_nosave_pages(forbidden_pages_map);
1052 pr_debug("PM: Basic memory bitmaps created\n");
1059 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1066 * free_basic_memory_bitmaps - free memory bitmaps allocated by
1067 * create_basic_memory_bitmaps(). The auxiliary pointers are necessary
1068 * so that the bitmaps themselves are not referred to while they are being
1072 void free_basic_memory_bitmaps(void)
1074 struct memory_bitmap *bm1, *bm2;
1076 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1079 bm1 = forbidden_pages_map;
1080 bm2 = free_pages_map;
1081 forbidden_pages_map = NULL;
1082 free_pages_map = NULL;
1083 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1085 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1088 pr_debug("PM: Basic memory bitmaps freed\n");
1092 * snapshot_additional_pages - estimate the number of additional pages
1093 * be needed for setting up the suspend image data structures for given
1094 * zone (usually the returned value is greater than the exact number)
1097 unsigned int snapshot_additional_pages(struct zone *zone)
1099 unsigned int rtree, nodes;
1101 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1102 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1103 LINKED_PAGE_DATA_SIZE);
1105 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1112 #ifdef CONFIG_HIGHMEM
1114 * count_free_highmem_pages - compute the total number of free highmem
1115 * pages, system-wide.
1118 static unsigned int count_free_highmem_pages(void)
1121 unsigned int cnt = 0;
1123 for_each_populated_zone(zone)
1124 if (is_highmem(zone))
1125 cnt += zone_page_state(zone, NR_FREE_PAGES);
1131 * saveable_highmem_page - Determine whether a highmem page should be
1132 * included in the suspend image.
1134 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1135 * and it isn't a part of a free chunk of pages.
1137 static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1141 if (!pfn_valid(pfn))
1144 page = pfn_to_page(pfn);
1145 if (page_zone(page) != zone)
1148 BUG_ON(!PageHighMem(page));
1150 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) ||
1154 if (page_is_guard(page))
1161 * count_highmem_pages - compute the total number of saveable highmem
1165 static unsigned int count_highmem_pages(void)
1170 for_each_populated_zone(zone) {
1171 unsigned long pfn, max_zone_pfn;
1173 if (!is_highmem(zone))
1176 mark_free_pages(zone);
1177 max_zone_pfn = zone_end_pfn(zone);
1178 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1179 if (saveable_highmem_page(zone, pfn))
1185 static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1189 #endif /* CONFIG_HIGHMEM */
1192 * saveable_page - Determine whether a non-highmem page should be included
1193 * in the suspend image.
1195 * We should save the page if it isn't Nosave, and is not in the range
1196 * of pages statically defined as 'unsaveable', and it isn't a part of
1197 * a free chunk of pages.
1199 static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1203 if (!pfn_valid(pfn))
1206 page = pfn_to_page(pfn);
1207 if (page_zone(page) != zone)
1210 BUG_ON(PageHighMem(page));
1212 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1215 if (PageReserved(page)
1216 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1219 if (page_is_guard(page))
1226 * count_data_pages - compute the total number of saveable non-highmem
1230 static unsigned int count_data_pages(void)
1233 unsigned long pfn, max_zone_pfn;
1236 for_each_populated_zone(zone) {
1237 if (is_highmem(zone))
1240 mark_free_pages(zone);
1241 max_zone_pfn = zone_end_pfn(zone);
1242 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1243 if (saveable_page(zone, pfn))
1249 /* This is needed, because copy_page and memcpy are not usable for copying
1252 static inline void do_copy_page(long *dst, long *src)
1256 for (n = PAGE_SIZE / sizeof(long); n; n--)
1262 * safe_copy_page - check if the page we are going to copy is marked as
1263 * present in the kernel page tables (this always is the case if
1264 * CONFIG_DEBUG_PAGEALLOC is not set and in that case
1265 * kernel_page_present() always returns 'true').
1267 static void safe_copy_page(void *dst, struct page *s_page)
1269 if (kernel_page_present(s_page)) {
1270 do_copy_page(dst, page_address(s_page));
1272 kernel_map_pages(s_page, 1, 1);
1273 do_copy_page(dst, page_address(s_page));
1274 kernel_map_pages(s_page, 1, 0);
1279 #ifdef CONFIG_HIGHMEM
1280 static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1282 return is_highmem(zone) ?
1283 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1286 static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1288 struct page *s_page, *d_page;
1291 s_page = pfn_to_page(src_pfn);
1292 d_page = pfn_to_page(dst_pfn);
1293 if (PageHighMem(s_page)) {
1294 src = kmap_atomic(s_page);
1295 dst = kmap_atomic(d_page);
1296 do_copy_page(dst, src);
1300 if (PageHighMem(d_page)) {
1301 /* Page pointed to by src may contain some kernel
1302 * data modified by kmap_atomic()
1304 safe_copy_page(buffer, s_page);
1305 dst = kmap_atomic(d_page);
1306 copy_page(dst, buffer);
1309 safe_copy_page(page_address(d_page), s_page);
1314 #define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1316 static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1318 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1319 pfn_to_page(src_pfn));
1321 #endif /* CONFIG_HIGHMEM */
1323 static void copy_data_pages(struct memory_bitmap *copy_bm,
1324 struct memory_bitmap *orig_bm)
1329 for_each_populated_zone(zone) {
1330 unsigned long max_zone_pfn;
1332 mark_free_pages(zone);
1333 max_zone_pfn = zone_end_pfn(zone);
1334 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1335 if (page_is_saveable(zone, pfn))
1336 memory_bm_set_bit(orig_bm, pfn);
1338 memory_bm_position_reset(orig_bm);
1339 memory_bm_position_reset(copy_bm);
1341 pfn = memory_bm_next_pfn(orig_bm);
1342 if (unlikely(pfn == BM_END_OF_MAP))
1344 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1348 /* Total number of image pages */
1349 static unsigned int nr_copy_pages;
1350 /* Number of pages needed for saving the original pfns of the image pages */
1351 static unsigned int nr_meta_pages;
1353 * Numbers of normal and highmem page frames allocated for hibernation image
1354 * before suspending devices.
1356 unsigned int alloc_normal, alloc_highmem;
1358 * Memory bitmap used for marking saveable pages (during hibernation) or
1359 * hibernation image pages (during restore)
1361 static struct memory_bitmap orig_bm;
1363 * Memory bitmap used during hibernation for marking allocated page frames that
1364 * will contain copies of saveable pages. During restore it is initially used
1365 * for marking hibernation image pages, but then the set bits from it are
1366 * duplicated in @orig_bm and it is released. On highmem systems it is next
1367 * used for marking "safe" highmem pages, but it has to be reinitialized for
1370 static struct memory_bitmap copy_bm;
1373 * swsusp_free - free pages allocated for the suspend.
1375 * Suspend pages are alocated before the atomic copy is made, so we
1376 * need to release them after the resume.
1379 void swsusp_free(void)
1381 unsigned long fb_pfn, fr_pfn;
1383 if (!forbidden_pages_map || !free_pages_map)
1386 memory_bm_position_reset(forbidden_pages_map);
1387 memory_bm_position_reset(free_pages_map);
1390 fr_pfn = memory_bm_next_pfn(free_pages_map);
1391 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1394 * Find the next bit set in both bitmaps. This is guaranteed to
1395 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1398 if (fb_pfn < fr_pfn)
1399 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1400 if (fr_pfn < fb_pfn)
1401 fr_pfn = memory_bm_next_pfn(free_pages_map);
1402 } while (fb_pfn != fr_pfn);
1404 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1405 struct page *page = pfn_to_page(fr_pfn);
1407 memory_bm_clear_current(forbidden_pages_map);
1408 memory_bm_clear_current(free_pages_map);
1416 restore_pblist = NULL;
1422 /* Helper functions used for the shrinking of memory. */
1424 #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1427 * preallocate_image_pages - Allocate a number of pages for hibernation image
1428 * @nr_pages: Number of page frames to allocate.
1429 * @mask: GFP flags to use for the allocation.
1431 * Return value: Number of page frames actually allocated
1433 static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1435 unsigned long nr_alloc = 0;
1437 while (nr_pages > 0) {
1440 page = alloc_image_page(mask);
1443 memory_bm_set_bit(©_bm, page_to_pfn(page));
1444 if (PageHighMem(page))
1455 static unsigned long preallocate_image_memory(unsigned long nr_pages,
1456 unsigned long avail_normal)
1458 unsigned long alloc;
1460 if (avail_normal <= alloc_normal)
1463 alloc = avail_normal - alloc_normal;
1464 if (nr_pages < alloc)
1467 return preallocate_image_pages(alloc, GFP_IMAGE);
1470 #ifdef CONFIG_HIGHMEM
1471 static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1473 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1477 * __fraction - Compute (an approximation of) x * (multiplier / base)
1479 static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1483 return (unsigned long)x;
1486 static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1487 unsigned long highmem,
1488 unsigned long total)
1490 unsigned long alloc = __fraction(nr_pages, highmem, total);
1492 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1494 #else /* CONFIG_HIGHMEM */
1495 static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1500 static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1501 unsigned long highmem,
1502 unsigned long total)
1506 #endif /* CONFIG_HIGHMEM */
1509 * free_unnecessary_pages - Release preallocated pages not needed for the image
1511 static unsigned long free_unnecessary_pages(void)
1513 unsigned long save, to_free_normal, to_free_highmem, free;
1515 save = count_data_pages();
1516 if (alloc_normal >= save) {
1517 to_free_normal = alloc_normal - save;
1521 save -= alloc_normal;
1523 save += count_highmem_pages();
1524 if (alloc_highmem >= save) {
1525 to_free_highmem = alloc_highmem - save;
1527 to_free_highmem = 0;
1528 save -= alloc_highmem;
1529 if (to_free_normal > save)
1530 to_free_normal -= save;
1534 free = to_free_normal + to_free_highmem;
1536 memory_bm_position_reset(©_bm);
1538 while (to_free_normal > 0 || to_free_highmem > 0) {
1539 unsigned long pfn = memory_bm_next_pfn(©_bm);
1540 struct page *page = pfn_to_page(pfn);
1542 if (PageHighMem(page)) {
1543 if (!to_free_highmem)
1548 if (!to_free_normal)
1553 memory_bm_clear_bit(©_bm, pfn);
1554 swsusp_unset_page_forbidden(page);
1555 swsusp_unset_page_free(page);
1563 * minimum_image_size - Estimate the minimum acceptable size of an image
1564 * @saveable: Number of saveable pages in the system.
1566 * We want to avoid attempting to free too much memory too hard, so estimate the
1567 * minimum acceptable size of a hibernation image to use as the lower limit for
1568 * preallocating memory.
1570 * We assume that the minimum image size should be proportional to
1572 * [number of saveable pages] - [number of pages that can be freed in theory]
1574 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1575 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages,
1576 * minus mapped file pages.
1578 static unsigned long minimum_image_size(unsigned long saveable)
1582 size = global_page_state(NR_SLAB_RECLAIMABLE)
1583 + global_page_state(NR_ACTIVE_ANON)
1584 + global_page_state(NR_INACTIVE_ANON)
1585 + global_page_state(NR_ACTIVE_FILE)
1586 + global_page_state(NR_INACTIVE_FILE)
1587 - global_page_state(NR_FILE_MAPPED);
1589 return saveable <= size ? 0 : saveable - size;
1593 * hibernate_preallocate_memory - Preallocate memory for hibernation image
1595 * To create a hibernation image it is necessary to make a copy of every page
1596 * frame in use. We also need a number of page frames to be free during
1597 * hibernation for allocations made while saving the image and for device
1598 * drivers, in case they need to allocate memory from their hibernation
1599 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1600 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1601 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1602 * total number of available page frames and allocate at least
1604 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1605 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1607 * of them, which corresponds to the maximum size of a hibernation image.
1609 * If image_size is set below the number following from the above formula,
1610 * the preallocation of memory is continued until the total number of saveable
1611 * pages in the system is below the requested image size or the minimum
1612 * acceptable image size returned by minimum_image_size(), whichever is greater.
1614 int hibernate_preallocate_memory(void)
1617 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1618 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1619 ktime_t start, stop;
1622 printk(KERN_INFO "PM: Preallocating image memory... ");
1623 start = ktime_get();
1625 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1629 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1636 /* Count the number of saveable data pages. */
1637 save_highmem = count_highmem_pages();
1638 saveable = count_data_pages();
1641 * Compute the total number of page frames we can use (count) and the
1642 * number of pages needed for image metadata (size).
1645 saveable += save_highmem;
1646 highmem = save_highmem;
1648 for_each_populated_zone(zone) {
1649 size += snapshot_additional_pages(zone);
1650 if (is_highmem(zone))
1651 highmem += zone_page_state(zone, NR_FREE_PAGES);
1653 count += zone_page_state(zone, NR_FREE_PAGES);
1655 avail_normal = count;
1657 count -= totalreserve_pages;
1659 /* Add number of pages required for page keys (s390 only). */
1660 size += page_key_additional_pages(saveable);
1662 /* Compute the maximum number of saveable pages to leave in memory. */
1663 max_size = (count - (size + PAGES_FOR_IO)) / 2
1664 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1665 /* Compute the desired number of image pages specified by image_size. */
1666 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1667 if (size > max_size)
1670 * If the desired number of image pages is at least as large as the
1671 * current number of saveable pages in memory, allocate page frames for
1672 * the image and we're done.
1674 if (size >= saveable) {
1675 pages = preallocate_image_highmem(save_highmem);
1676 pages += preallocate_image_memory(saveable - pages, avail_normal);
1680 /* Estimate the minimum size of the image. */
1681 pages = minimum_image_size(saveable);
1683 * To avoid excessive pressure on the normal zone, leave room in it to
1684 * accommodate an image of the minimum size (unless it's already too
1685 * small, in which case don't preallocate pages from it at all).
1687 if (avail_normal > pages)
1688 avail_normal -= pages;
1692 size = min_t(unsigned long, pages, max_size);
1695 * Let the memory management subsystem know that we're going to need a
1696 * large number of page frames to allocate and make it free some memory.
1697 * NOTE: If this is not done, performance will be hurt badly in some
1700 shrink_all_memory(saveable - size);
1703 * The number of saveable pages in memory was too high, so apply some
1704 * pressure to decrease it. First, make room for the largest possible
1705 * image and fail if that doesn't work. Next, try to decrease the size
1706 * of the image as much as indicated by 'size' using allocations from
1707 * highmem and non-highmem zones separately.
1709 pages_highmem = preallocate_image_highmem(highmem / 2);
1710 alloc = count - max_size;
1711 if (alloc > pages_highmem)
1712 alloc -= pages_highmem;
1715 pages = preallocate_image_memory(alloc, avail_normal);
1716 if (pages < alloc) {
1717 /* We have exhausted non-highmem pages, try highmem. */
1719 pages += pages_highmem;
1720 pages_highmem = preallocate_image_highmem(alloc);
1721 if (pages_highmem < alloc)
1723 pages += pages_highmem;
1725 * size is the desired number of saveable pages to leave in
1726 * memory, so try to preallocate (all memory - size) pages.
1728 alloc = (count - pages) - size;
1729 pages += preallocate_image_highmem(alloc);
1732 * There are approximately max_size saveable pages at this point
1733 * and we want to reduce this number down to size.
1735 alloc = max_size - size;
1736 size = preallocate_highmem_fraction(alloc, highmem, count);
1737 pages_highmem += size;
1739 size = preallocate_image_memory(alloc, avail_normal);
1740 pages_highmem += preallocate_image_highmem(alloc - size);
1741 pages += pages_highmem + size;
1745 * We only need as many page frames for the image as there are saveable
1746 * pages in memory, but we have allocated more. Release the excessive
1749 pages -= free_unnecessary_pages();
1753 printk(KERN_CONT "done (allocated %lu pages)\n", pages);
1754 swsusp_show_speed(start, stop, pages, "Allocated");
1759 printk(KERN_CONT "\n");
1764 #ifdef CONFIG_HIGHMEM
1766 * count_pages_for_highmem - compute the number of non-highmem pages
1767 * that will be necessary for creating copies of highmem pages.
1770 static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1772 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1774 if (free_highmem >= nr_highmem)
1777 nr_highmem -= free_highmem;
1782 static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1783 #endif /* CONFIG_HIGHMEM */
1786 * enough_free_mem - Make sure we have enough free memory for the
1790 static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1793 unsigned int free = alloc_normal;
1795 for_each_populated_zone(zone)
1796 if (!is_highmem(zone))
1797 free += zone_page_state(zone, NR_FREE_PAGES);
1799 nr_pages += count_pages_for_highmem(nr_highmem);
1800 pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n",
1801 nr_pages, PAGES_FOR_IO, free);
1803 return free > nr_pages + PAGES_FOR_IO;
1806 #ifdef CONFIG_HIGHMEM
1808 * get_highmem_buffer - if there are some highmem pages in the suspend
1809 * image, we may need the buffer to copy them and/or load their data.
1812 static inline int get_highmem_buffer(int safe_needed)
1814 buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
1815 return buffer ? 0 : -ENOMEM;
1819 * alloc_highmem_image_pages - allocate some highmem pages for the image.
1820 * Try to allocate as many pages as needed, but if the number of free
1821 * highmem pages is lesser than that, allocate them all.
1824 static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1825 unsigned int nr_highmem)
1827 unsigned int to_alloc = count_free_highmem_pages();
1829 if (to_alloc > nr_highmem)
1830 to_alloc = nr_highmem;
1832 nr_highmem -= to_alloc;
1833 while (to_alloc-- > 0) {
1836 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1837 memory_bm_set_bit(bm, page_to_pfn(page));
1842 static inline int get_highmem_buffer(int safe_needed) { return 0; }
1844 static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1845 unsigned int n) { return 0; }
1846 #endif /* CONFIG_HIGHMEM */
1849 * swsusp_alloc - allocate memory for the suspend image
1851 * We first try to allocate as many highmem pages as there are
1852 * saveable highmem pages in the system. If that fails, we allocate
1853 * non-highmem pages for the copies of the remaining highmem ones.
1855 * In this approach it is likely that the copies of highmem pages will
1856 * also be located in the high memory, because of the way in which
1857 * copy_data_pages() works.
1860 static int swsusp_alloc(struct memory_bitmap *orig_bm,
1861 struct memory_bitmap *copy_bm,
1862 unsigned int nr_pages, unsigned int nr_highmem)
1864 if (nr_highmem > 0) {
1865 if (get_highmem_buffer(PG_ANY))
1867 if (nr_highmem > alloc_highmem) {
1868 nr_highmem -= alloc_highmem;
1869 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1872 if (nr_pages > alloc_normal) {
1873 nr_pages -= alloc_normal;
1874 while (nr_pages-- > 0) {
1877 page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
1880 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1891 asmlinkage __visible int swsusp_save(void)
1893 unsigned int nr_pages, nr_highmem;
1895 printk(KERN_INFO "PM: Creating hibernation image:\n");
1897 drain_local_pages(NULL);
1898 nr_pages = count_data_pages();
1899 nr_highmem = count_highmem_pages();
1900 printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem);
1902 if (!enough_free_mem(nr_pages, nr_highmem)) {
1903 printk(KERN_ERR "PM: Not enough free memory\n");
1907 if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) {
1908 printk(KERN_ERR "PM: Memory allocation failed\n");
1912 /* During allocating of suspend pagedir, new cold pages may appear.
1915 drain_local_pages(NULL);
1916 copy_data_pages(©_bm, &orig_bm);
1919 * End of critical section. From now on, we can write to memory,
1920 * but we should not touch disk. This specially means we must _not_
1921 * touch swap space! Except we must write out our image of course.
1924 nr_pages += nr_highmem;
1925 nr_copy_pages = nr_pages;
1926 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1928 printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n",
1934 #ifndef CONFIG_ARCH_HIBERNATION_HEADER
1935 static int init_header_complete(struct swsusp_info *info)
1937 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
1938 info->version_code = LINUX_VERSION_CODE;
1942 static char *check_image_kernel(struct swsusp_info *info)
1944 if (info->version_code != LINUX_VERSION_CODE)
1945 return "kernel version";
1946 if (strcmp(info->uts.sysname,init_utsname()->sysname))
1947 return "system type";
1948 if (strcmp(info->uts.release,init_utsname()->release))
1949 return "kernel release";
1950 if (strcmp(info->uts.version,init_utsname()->version))
1952 if (strcmp(info->uts.machine,init_utsname()->machine))
1956 #endif /* CONFIG_ARCH_HIBERNATION_HEADER */
1958 unsigned long snapshot_get_image_size(void)
1960 return nr_copy_pages + nr_meta_pages + 1;
1963 static int init_header(struct swsusp_info *info)
1965 memset(info, 0, sizeof(struct swsusp_info));
1966 info->num_physpages = get_num_physpages();
1967 info->image_pages = nr_copy_pages;
1968 info->pages = snapshot_get_image_size();
1969 info->size = info->pages;
1970 info->size <<= PAGE_SHIFT;
1971 return init_header_complete(info);
1975 * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm
1976 * are stored in the array @buf[] (1 page at a time)
1979 static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
1983 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1984 buf[j] = memory_bm_next_pfn(bm);
1985 if (unlikely(buf[j] == BM_END_OF_MAP))
1987 /* Save page key for data page (s390 only). */
1988 page_key_read(buf + j);
1993 * snapshot_read_next - used for reading the system memory snapshot.
1995 * On the first call to it @handle should point to a zeroed
1996 * snapshot_handle structure. The structure gets updated and a pointer
1997 * to it should be passed to this function every next time.
1999 * On success the function returns a positive number. Then, the caller
2000 * is allowed to read up to the returned number of bytes from the memory
2001 * location computed by the data_of() macro.
2003 * The function returns 0 to indicate the end of data stream condition,
2004 * and a negative number is returned on error. In such cases the
2005 * structure pointed to by @handle is not updated and should not be used
2009 int snapshot_read_next(struct snapshot_handle *handle)
2011 if (handle->cur > nr_meta_pages + nr_copy_pages)
2015 /* This makes the buffer be freed by swsusp_free() */
2016 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2023 error = init_header((struct swsusp_info *)buffer);
2026 handle->buffer = buffer;
2027 memory_bm_position_reset(&orig_bm);
2028 memory_bm_position_reset(©_bm);
2029 } else if (handle->cur <= nr_meta_pages) {
2031 pack_pfns(buffer, &orig_bm);
2035 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2036 if (PageHighMem(page)) {
2037 /* Highmem pages are copied to the buffer,
2038 * because we can't return with a kmapped
2039 * highmem page (we may not be called again).
2043 kaddr = kmap_atomic(page);
2044 copy_page(buffer, kaddr);
2045 kunmap_atomic(kaddr);
2046 handle->buffer = buffer;
2048 handle->buffer = page_address(page);
2055 static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2056 struct memory_bitmap *src)
2060 memory_bm_position_reset(src);
2061 pfn = memory_bm_next_pfn(src);
2062 while (pfn != BM_END_OF_MAP) {
2063 memory_bm_set_bit(dst, pfn);
2064 pfn = memory_bm_next_pfn(src);
2069 * mark_unsafe_pages - mark the pages that cannot be used for storing
2070 * the image during resume, because they conflict with the pages that
2071 * had been used before suspend
2074 static void mark_unsafe_pages(struct memory_bitmap *bm)
2078 /* Clear the "free"/"unsafe" bit for all PFNs */
2079 memory_bm_position_reset(free_pages_map);
2080 pfn = memory_bm_next_pfn(free_pages_map);
2081 while (pfn != BM_END_OF_MAP) {
2082 memory_bm_clear_current(free_pages_map);
2083 pfn = memory_bm_next_pfn(free_pages_map);
2086 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2087 duplicate_memory_bitmap(free_pages_map, bm);
2089 allocated_unsafe_pages = 0;
2092 static int check_header(struct swsusp_info *info)
2096 reason = check_image_kernel(info);
2097 if (!reason && info->num_physpages != get_num_physpages())
2098 reason = "memory size";
2100 printk(KERN_ERR "PM: Image mismatch: %s\n", reason);
2107 * load header - check the image header and copy data from it
2110 static int load_header(struct swsusp_info *info)
2114 restore_pblist = NULL;
2115 error = check_header(info);
2117 nr_copy_pages = info->image_pages;
2118 nr_meta_pages = info->pages - info->image_pages - 1;
2124 * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set
2125 * the corresponding bit in the memory bitmap @bm
2127 static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2131 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2132 if (unlikely(buf[j] == BM_END_OF_MAP))
2135 /* Extract and buffer page key for data page (s390 only). */
2136 page_key_memorize(buf + j);
2138 if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2139 memory_bm_set_bit(bm, buf[j]);
2147 #ifdef CONFIG_HIGHMEM
2148 /* struct highmem_pbe is used for creating the list of highmem pages that
2149 * should be restored atomically during the resume from disk, because the page
2150 * frames they have occupied before the suspend are in use.
2152 struct highmem_pbe {
2153 struct page *copy_page; /* data is here now */
2154 struct page *orig_page; /* data was here before the suspend */
2155 struct highmem_pbe *next;
2158 /* List of highmem PBEs needed for restoring the highmem pages that were
2159 * allocated before the suspend and included in the suspend image, but have
2160 * also been allocated by the "resume" kernel, so their contents cannot be
2161 * written directly to their "original" page frames.
2163 static struct highmem_pbe *highmem_pblist;
2166 * count_highmem_image_pages - compute the number of highmem pages in the
2167 * suspend image. The bits in the memory bitmap @bm that correspond to the
2168 * image pages are assumed to be set.
2171 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2174 unsigned int cnt = 0;
2176 memory_bm_position_reset(bm);
2177 pfn = memory_bm_next_pfn(bm);
2178 while (pfn != BM_END_OF_MAP) {
2179 if (PageHighMem(pfn_to_page(pfn)))
2182 pfn = memory_bm_next_pfn(bm);
2188 * prepare_highmem_image - try to allocate as many highmem pages as
2189 * there are highmem image pages (@nr_highmem_p points to the variable
2190 * containing the number of highmem image pages). The pages that are
2191 * "safe" (ie. will not be overwritten when the suspend image is
2192 * restored) have the corresponding bits set in @bm (it must be
2195 * NOTE: This function should not be called if there are no highmem
2199 static unsigned int safe_highmem_pages;
2201 static struct memory_bitmap *safe_highmem_bm;
2203 static int prepare_highmem_image(struct memory_bitmap *bm,
2204 unsigned int *nr_highmem_p)
2206 unsigned int to_alloc;
2208 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2211 if (get_highmem_buffer(PG_SAFE))
2214 to_alloc = count_free_highmem_pages();
2215 if (to_alloc > *nr_highmem_p)
2216 to_alloc = *nr_highmem_p;
2218 *nr_highmem_p = to_alloc;
2220 safe_highmem_pages = 0;
2221 while (to_alloc-- > 0) {
2224 page = alloc_page(__GFP_HIGHMEM);
2225 if (!swsusp_page_is_free(page)) {
2226 /* The page is "safe", set its bit the bitmap */
2227 memory_bm_set_bit(bm, page_to_pfn(page));
2228 safe_highmem_pages++;
2230 /* Mark the page as allocated */
2231 swsusp_set_page_forbidden(page);
2232 swsusp_set_page_free(page);
2234 memory_bm_position_reset(bm);
2235 safe_highmem_bm = bm;
2240 * get_highmem_page_buffer - for given highmem image page find the buffer
2241 * that suspend_write_next() should set for its caller to write to.
2243 * If the page is to be saved to its "original" page frame or a copy of
2244 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2245 * the copy of the page is to be made in normal memory, so the address of
2246 * the copy is returned.
2248 * If @buffer is returned, the caller of suspend_write_next() will write
2249 * the page's contents to @buffer, so they will have to be copied to the
2250 * right location on the next call to suspend_write_next() and it is done
2251 * with the help of copy_last_highmem_page(). For this purpose, if
2252 * @buffer is returned, @last_highmem page is set to the page to which
2253 * the data will have to be copied from @buffer.
2256 static struct page *last_highmem_page;
2258 static void *get_highmem_page_buffer(struct page *page,
2259 struct chain_allocator *ca)
2261 struct highmem_pbe *pbe;
2264 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2265 /* We have allocated the "original" page frame and we can
2266 * use it directly to store the loaded page.
2268 last_highmem_page = page;
2271 /* The "original" page frame has not been allocated and we have to
2272 * use a "safe" page frame to store the loaded page.
2274 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2277 return ERR_PTR(-ENOMEM);
2279 pbe->orig_page = page;
2280 if (safe_highmem_pages > 0) {
2283 /* Copy of the page will be stored in high memory */
2285 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2286 safe_highmem_pages--;
2287 last_highmem_page = tmp;
2288 pbe->copy_page = tmp;
2290 /* Copy of the page will be stored in normal memory */
2291 kaddr = safe_pages_list;
2292 safe_pages_list = safe_pages_list->next;
2293 pbe->copy_page = virt_to_page(kaddr);
2295 pbe->next = highmem_pblist;
2296 highmem_pblist = pbe;
2301 * copy_last_highmem_page - copy the contents of a highmem image from
2302 * @buffer, where the caller of snapshot_write_next() has place them,
2303 * to the right location represented by @last_highmem_page .
2306 static void copy_last_highmem_page(void)
2308 if (last_highmem_page) {
2311 dst = kmap_atomic(last_highmem_page);
2312 copy_page(dst, buffer);
2314 last_highmem_page = NULL;
2318 static inline int last_highmem_page_copied(void)
2320 return !last_highmem_page;
2323 static inline void free_highmem_data(void)
2325 if (safe_highmem_bm)
2326 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2329 free_image_page(buffer, PG_UNSAFE_CLEAR);
2332 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2334 static inline int prepare_highmem_image(struct memory_bitmap *bm,
2335 unsigned int *nr_highmem_p) { return 0; }
2337 static inline void *get_highmem_page_buffer(struct page *page,
2338 struct chain_allocator *ca)
2340 return ERR_PTR(-EINVAL);
2343 static inline void copy_last_highmem_page(void) {}
2344 static inline int last_highmem_page_copied(void) { return 1; }
2345 static inline void free_highmem_data(void) {}
2346 #endif /* CONFIG_HIGHMEM */
2349 * prepare_image - use the memory bitmap @bm to mark the pages that will
2350 * be overwritten in the process of restoring the system memory state
2351 * from the suspend image ("unsafe" pages) and allocate memory for the
2354 * The idea is to allocate a new memory bitmap first and then allocate
2355 * as many pages as needed for the image data, but not to assign these
2356 * pages to specific tasks initially. Instead, we just mark them as
2357 * allocated and create a lists of "safe" pages that will be used
2358 * later. On systems with high memory a list of "safe" highmem pages is
2362 #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2364 static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2366 unsigned int nr_pages, nr_highmem;
2367 struct linked_page *lp;
2370 /* If there is no highmem, the buffer will not be necessary */
2371 free_image_page(buffer, PG_UNSAFE_CLEAR);
2374 nr_highmem = count_highmem_image_pages(bm);
2375 mark_unsafe_pages(bm);
2377 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2381 duplicate_memory_bitmap(new_bm, bm);
2382 memory_bm_free(bm, PG_UNSAFE_KEEP);
2383 if (nr_highmem > 0) {
2384 error = prepare_highmem_image(bm, &nr_highmem);
2388 /* Reserve some safe pages for potential later use.
2390 * NOTE: This way we make sure there will be enough safe pages for the
2391 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2392 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2394 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2396 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2397 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2398 while (nr_pages > 0) {
2399 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2404 lp->next = safe_pages_list;
2405 safe_pages_list = lp;
2408 /* Preallocate memory for the image */
2409 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2410 while (nr_pages > 0) {
2411 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2416 if (!swsusp_page_is_free(virt_to_page(lp))) {
2417 /* The page is "safe", add it to the list */
2418 lp->next = safe_pages_list;
2419 safe_pages_list = lp;
2421 /* Mark the page as allocated */
2422 swsusp_set_page_forbidden(virt_to_page(lp));
2423 swsusp_set_page_free(virt_to_page(lp));
2434 * get_buffer - compute the address that snapshot_write_next() should
2435 * set for its caller to write to.
2438 static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2442 unsigned long pfn = memory_bm_next_pfn(bm);
2444 if (pfn == BM_END_OF_MAP)
2445 return ERR_PTR(-EFAULT);
2447 page = pfn_to_page(pfn);
2448 if (PageHighMem(page))
2449 return get_highmem_page_buffer(page, ca);
2451 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2452 /* We have allocated the "original" page frame and we can
2453 * use it directly to store the loaded page.
2455 return page_address(page);
2457 /* The "original" page frame has not been allocated and we have to
2458 * use a "safe" page frame to store the loaded page.
2460 pbe = chain_alloc(ca, sizeof(struct pbe));
2463 return ERR_PTR(-ENOMEM);
2465 pbe->orig_address = page_address(page);
2466 pbe->address = safe_pages_list;
2467 safe_pages_list = safe_pages_list->next;
2468 pbe->next = restore_pblist;
2469 restore_pblist = pbe;
2470 return pbe->address;
2474 * snapshot_write_next - used for writing the system memory snapshot.
2476 * On the first call to it @handle should point to a zeroed
2477 * snapshot_handle structure. The structure gets updated and a pointer
2478 * to it should be passed to this function every next time.
2480 * On success the function returns a positive number. Then, the caller
2481 * is allowed to write up to the returned number of bytes to the memory
2482 * location computed by the data_of() macro.
2484 * The function returns 0 to indicate the "end of file" condition,
2485 * and a negative number is returned on error. In such cases the
2486 * structure pointed to by @handle is not updated and should not be used
2490 int snapshot_write_next(struct snapshot_handle *handle)
2492 static struct chain_allocator ca;
2495 /* Check if we have already loaded the entire image */
2496 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2499 handle->sync_read = 1;
2503 /* This makes the buffer be freed by swsusp_free() */
2504 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2509 handle->buffer = buffer;
2510 } else if (handle->cur == 1) {
2511 error = load_header(buffer);
2515 safe_pages_list = NULL;
2517 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2521 /* Allocate buffer for page keys. */
2522 error = page_key_alloc(nr_copy_pages);
2526 } else if (handle->cur <= nr_meta_pages + 1) {
2527 error = unpack_orig_pfns(buffer, ©_bm);
2531 if (handle->cur == nr_meta_pages + 1) {
2532 error = prepare_image(&orig_bm, ©_bm);
2536 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2537 memory_bm_position_reset(&orig_bm);
2538 restore_pblist = NULL;
2539 handle->buffer = get_buffer(&orig_bm, &ca);
2540 handle->sync_read = 0;
2541 if (IS_ERR(handle->buffer))
2542 return PTR_ERR(handle->buffer);
2545 copy_last_highmem_page();
2546 /* Restore page key for data page (s390 only). */
2547 page_key_write(handle->buffer);
2548 handle->buffer = get_buffer(&orig_bm, &ca);
2549 if (IS_ERR(handle->buffer))
2550 return PTR_ERR(handle->buffer);
2551 if (handle->buffer != buffer)
2552 handle->sync_read = 0;
2559 * snapshot_write_finalize - must be called after the last call to
2560 * snapshot_write_next() in case the last page in the image happens
2561 * to be a highmem page and its contents should be stored in the
2562 * highmem. Additionally, it releases the memory that will not be
2566 void snapshot_write_finalize(struct snapshot_handle *handle)
2568 copy_last_highmem_page();
2569 /* Restore page key for data page (s390 only). */
2570 page_key_write(handle->buffer);
2572 /* Do that only if we have loaded the image entirely */
2573 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2574 memory_bm_recycle(&orig_bm);
2575 free_highmem_data();
2579 int snapshot_image_loaded(struct snapshot_handle *handle)
2581 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2582 handle->cur <= nr_meta_pages + nr_copy_pages);
2585 #ifdef CONFIG_HIGHMEM
2586 /* Assumes that @buf is ready and points to a "safe" page */
2587 static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2590 void *kaddr1, *kaddr2;
2592 kaddr1 = kmap_atomic(p1);
2593 kaddr2 = kmap_atomic(p2);
2594 copy_page(buf, kaddr1);
2595 copy_page(kaddr1, kaddr2);
2596 copy_page(kaddr2, buf);
2597 kunmap_atomic(kaddr2);
2598 kunmap_atomic(kaddr1);
2602 * restore_highmem - for each highmem page that was allocated before
2603 * the suspend and included in the suspend image, and also has been
2604 * allocated by the "resume" kernel swap its current (ie. "before
2605 * resume") contents with the previous (ie. "before suspend") one.
2607 * If the resume eventually fails, we can call this function once
2608 * again and restore the "before resume" highmem state.
2611 int restore_highmem(void)
2613 struct highmem_pbe *pbe = highmem_pblist;
2619 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2624 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2627 free_image_page(buf, PG_UNSAFE_CLEAR);
2630 #endif /* CONFIG_HIGHMEM */