1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/kernel/power/snapshot.c
5 * This file provides system snapshot/restore functionality for swsusp.
7 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
8 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
11 #define pr_fmt(fmt) "PM: hibernation: " fmt
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/memblock.h>
25 #include <linux/nmi.h>
26 #include <linux/syscalls.h>
27 #include <linux/console.h>
28 #include <linux/highmem.h>
29 #include <linux/list.h>
30 #include <linux/slab.h>
31 #include <linux/compiler.h>
32 #include <linux/ktime.h>
33 #include <linux/set_memory.h>
35 #include <linux/uaccess.h>
36 #include <asm/mmu_context.h>
37 #include <asm/tlbflush.h>
42 #if defined(CONFIG_STRICT_KERNEL_RWX) && defined(CONFIG_ARCH_HAS_SET_MEMORY)
43 static bool hibernate_restore_protection;
44 static bool hibernate_restore_protection_active;
46 void enable_restore_image_protection(void)
48 hibernate_restore_protection = true;
51 static inline void hibernate_restore_protection_begin(void)
53 hibernate_restore_protection_active = hibernate_restore_protection;
56 static inline void hibernate_restore_protection_end(void)
58 hibernate_restore_protection_active = false;
61 static inline void hibernate_restore_protect_page(void *page_address)
63 if (hibernate_restore_protection_active)
64 set_memory_ro((unsigned long)page_address, 1);
67 static inline void hibernate_restore_unprotect_page(void *page_address)
69 if (hibernate_restore_protection_active)
70 set_memory_rw((unsigned long)page_address, 1);
73 static inline void hibernate_restore_protection_begin(void) {}
74 static inline void hibernate_restore_protection_end(void) {}
75 static inline void hibernate_restore_protect_page(void *page_address) {}
76 static inline void hibernate_restore_unprotect_page(void *page_address) {}
77 #endif /* CONFIG_STRICT_KERNEL_RWX && CONFIG_ARCH_HAS_SET_MEMORY */
81 * The calls to set_direct_map_*() should not fail because remapping a page
82 * here means that we only update protection bits in an existing PTE.
83 * It is still worth to have a warning here if something changes and this
84 * will no longer be the case.
86 static inline void hibernate_map_page(struct page *page)
88 if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
89 int ret = set_direct_map_default_noflush(page);
92 pr_warn_once("Failed to remap page\n");
94 debug_pagealloc_map_pages(page, 1);
98 static inline void hibernate_unmap_page(struct page *page)
100 if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
101 unsigned long addr = (unsigned long)page_address(page);
102 int ret = set_direct_map_invalid_noflush(page);
105 pr_warn_once("Failed to remap page\n");
107 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
109 debug_pagealloc_unmap_pages(page, 1);
113 static int swsusp_page_is_free(struct page *);
114 static void swsusp_set_page_forbidden(struct page *);
115 static void swsusp_unset_page_forbidden(struct page *);
118 * Number of bytes to reserve for memory allocations made by device drivers
119 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
120 * cause image creation to fail (tunable via /sys/power/reserved_size).
122 unsigned long reserved_size;
124 void __init hibernate_reserved_size_init(void)
126 reserved_size = SPARE_PAGES * PAGE_SIZE;
130 * Preferred image size in bytes (tunable via /sys/power/image_size).
131 * When it is set to N, swsusp will do its best to ensure the image
132 * size will not exceed N bytes, but if that is impossible, it will
133 * try to create the smallest image possible.
135 unsigned long image_size;
137 void __init hibernate_image_size_init(void)
139 image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
143 * List of PBEs needed for restoring the pages that were allocated before
144 * the suspend and included in the suspend image, but have also been
145 * allocated by the "resume" kernel, so their contents cannot be written
146 * directly to their "original" page frames.
148 struct pbe *restore_pblist;
150 /* struct linked_page is used to build chains of pages */
152 #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
155 struct linked_page *next;
156 char data[LINKED_PAGE_DATA_SIZE];
160 * List of "safe" pages (ie. pages that were not used by the image kernel
161 * before hibernation) that may be used as temporary storage for image kernel
164 static struct linked_page *safe_pages_list;
166 /* Pointer to an auxiliary buffer (1 page) */
171 #define PG_UNSAFE_CLEAR 1
172 #define PG_UNSAFE_KEEP 0
174 static unsigned int allocated_unsafe_pages;
177 * get_image_page - Allocate a page for a hibernation image.
178 * @gfp_mask: GFP mask for the allocation.
179 * @safe_needed: Get pages that were not used before hibernation (restore only)
181 * During image restoration, for storing the PBE list and the image data, we can
182 * only use memory pages that do not conflict with the pages used before
183 * hibernation. The "unsafe" pages have PageNosaveFree set and we count them
184 * using allocated_unsafe_pages.
186 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
187 * swsusp_free() can release it.
189 static void *get_image_page(gfp_t gfp_mask, int safe_needed)
193 res = (void *)get_zeroed_page(gfp_mask);
195 while (res && swsusp_page_is_free(virt_to_page(res))) {
196 /* The page is unsafe, mark it for swsusp_free() */
197 swsusp_set_page_forbidden(virt_to_page(res));
198 allocated_unsafe_pages++;
199 res = (void *)get_zeroed_page(gfp_mask);
202 swsusp_set_page_forbidden(virt_to_page(res));
203 swsusp_set_page_free(virt_to_page(res));
208 static void *__get_safe_page(gfp_t gfp_mask)
210 if (safe_pages_list) {
211 void *ret = safe_pages_list;
213 safe_pages_list = safe_pages_list->next;
214 memset(ret, 0, PAGE_SIZE);
217 return get_image_page(gfp_mask, PG_SAFE);
220 unsigned long get_safe_page(gfp_t gfp_mask)
222 return (unsigned long)__get_safe_page(gfp_mask);
225 static struct page *alloc_image_page(gfp_t gfp_mask)
229 page = alloc_page(gfp_mask);
231 swsusp_set_page_forbidden(page);
232 swsusp_set_page_free(page);
237 static void recycle_safe_page(void *page_address)
239 struct linked_page *lp = page_address;
241 lp->next = safe_pages_list;
242 safe_pages_list = lp;
246 * free_image_page - Free a page allocated for hibernation image.
247 * @addr: Address of the page to free.
248 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
250 * The page to free should have been allocated by get_image_page() (page flags
251 * set by it are affected).
253 static inline void free_image_page(void *addr, int clear_nosave_free)
257 BUG_ON(!virt_addr_valid(addr));
259 page = virt_to_page(addr);
261 swsusp_unset_page_forbidden(page);
262 if (clear_nosave_free)
263 swsusp_unset_page_free(page);
268 static inline void free_list_of_pages(struct linked_page *list,
269 int clear_page_nosave)
272 struct linked_page *lp = list->next;
274 free_image_page(list, clear_page_nosave);
280 * struct chain_allocator is used for allocating small objects out of
281 * a linked list of pages called 'the chain'.
283 * The chain grows each time when there is no room for a new object in
284 * the current page. The allocated objects cannot be freed individually.
285 * It is only possible to free them all at once, by freeing the entire
288 * NOTE: The chain allocator may be inefficient if the allocated objects
289 * are not much smaller than PAGE_SIZE.
291 struct chain_allocator {
292 struct linked_page *chain; /* the chain */
293 unsigned int used_space; /* total size of objects allocated out
294 of the current page */
295 gfp_t gfp_mask; /* mask for allocating pages */
296 int safe_needed; /* if set, only "safe" pages are allocated */
299 static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
303 ca->used_space = LINKED_PAGE_DATA_SIZE;
304 ca->gfp_mask = gfp_mask;
305 ca->safe_needed = safe_needed;
308 static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
312 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
313 struct linked_page *lp;
315 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
316 get_image_page(ca->gfp_mask, PG_ANY);
320 lp->next = ca->chain;
324 ret = ca->chain->data + ca->used_space;
325 ca->used_space += size;
330 * Data types related to memory bitmaps.
332 * Memory bitmap is a structure consiting of many linked lists of
333 * objects. The main list's elements are of type struct zone_bitmap
334 * and each of them corresonds to one zone. For each zone bitmap
335 * object there is a list of objects of type struct bm_block that
336 * represent each blocks of bitmap in which information is stored.
338 * struct memory_bitmap contains a pointer to the main list of zone
339 * bitmap objects, a struct bm_position used for browsing the bitmap,
340 * and a pointer to the list of pages used for allocating all of the
341 * zone bitmap objects and bitmap block objects.
343 * NOTE: It has to be possible to lay out the bitmap in memory
344 * using only allocations of order 0. Additionally, the bitmap is
345 * designed to work with arbitrary number of zones (this is over the
346 * top for now, but let's avoid making unnecessary assumptions ;-).
348 * struct zone_bitmap contains a pointer to a list of bitmap block
349 * objects and a pointer to the bitmap block object that has been
350 * most recently used for setting bits. Additionally, it contains the
351 * PFNs that correspond to the start and end of the represented zone.
353 * struct bm_block contains a pointer to the memory page in which
354 * information is stored (in the form of a block of bitmap)
355 * It also contains the pfns that correspond to the start and end of
356 * the represented memory area.
358 * The memory bitmap is organized as a radix tree to guarantee fast random
359 * access to the bits. There is one radix tree for each zone (as returned
360 * from create_mem_extents).
362 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
363 * two linked lists for the nodes of the tree, one for the inner nodes and
364 * one for the leave nodes. The linked leave nodes are used for fast linear
365 * access of the memory bitmap.
367 * The struct rtree_node represents one node of the radix tree.
370 #define BM_END_OF_MAP (~0UL)
372 #define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
373 #define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
374 #define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
377 * struct rtree_node is a wrapper struct to link the nodes
378 * of the rtree together for easy linear iteration over
379 * bits and easy freeing
382 struct list_head list;
387 * struct mem_zone_bm_rtree represents a bitmap used for one
388 * populated memory zone.
390 struct mem_zone_bm_rtree {
391 struct list_head list; /* Link Zones together */
392 struct list_head nodes; /* Radix Tree inner nodes */
393 struct list_head leaves; /* Radix Tree leaves */
394 unsigned long start_pfn; /* Zone start page frame */
395 unsigned long end_pfn; /* Zone end page frame + 1 */
396 struct rtree_node *rtree; /* Radix Tree Root */
397 int levels; /* Number of Radix Tree Levels */
398 unsigned int blocks; /* Number of Bitmap Blocks */
401 /* strcut bm_position is used for browsing memory bitmaps */
404 struct mem_zone_bm_rtree *zone;
405 struct rtree_node *node;
406 unsigned long node_pfn;
410 struct memory_bitmap {
411 struct list_head zones;
412 struct linked_page *p_list; /* list of pages used to store zone
413 bitmap objects and bitmap block
415 struct bm_position cur; /* most recently used bit position */
418 /* Functions that operate on memory bitmaps */
420 #define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
421 #if BITS_PER_LONG == 32
422 #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
424 #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
426 #define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
429 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
431 * This function is used to allocate inner nodes as well as the
432 * leave nodes of the radix tree. It also adds the node to the
433 * corresponding linked list passed in by the *list parameter.
435 static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
436 struct chain_allocator *ca,
437 struct list_head *list)
439 struct rtree_node *node;
441 node = chain_alloc(ca, sizeof(struct rtree_node));
445 node->data = get_image_page(gfp_mask, safe_needed);
449 list_add_tail(&node->list, list);
455 * add_rtree_block - Add a new leave node to the radix tree.
457 * The leave nodes need to be allocated in order to keep the leaves
458 * linked list in order. This is guaranteed by the zone->blocks
461 static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
462 int safe_needed, struct chain_allocator *ca)
464 struct rtree_node *node, *block, **dst;
465 unsigned int levels_needed, block_nr;
468 block_nr = zone->blocks;
471 /* How many levels do we need for this block nr? */
474 block_nr >>= BM_RTREE_LEVEL_SHIFT;
477 /* Make sure the rtree has enough levels */
478 for (i = zone->levels; i < levels_needed; i++) {
479 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
484 node->data[0] = (unsigned long)zone->rtree;
489 /* Allocate new block */
490 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
494 /* Now walk the rtree to insert the block */
497 block_nr = zone->blocks;
498 for (i = zone->levels; i > 0; i--) {
502 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
509 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
510 index &= BM_RTREE_LEVEL_MASK;
511 dst = (struct rtree_node **)&((*dst)->data[index]);
521 static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
522 int clear_nosave_free);
525 * create_zone_bm_rtree - Create a radix tree for one zone.
527 * Allocated the mem_zone_bm_rtree structure and initializes it.
528 * This function also allocated and builds the radix tree for the
531 static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
533 struct chain_allocator *ca,
537 struct mem_zone_bm_rtree *zone;
538 unsigned int i, nr_blocks;
542 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
546 INIT_LIST_HEAD(&zone->nodes);
547 INIT_LIST_HEAD(&zone->leaves);
548 zone->start_pfn = start;
550 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
552 for (i = 0; i < nr_blocks; i++) {
553 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
554 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
563 * free_zone_bm_rtree - Free the memory of the radix tree.
565 * Free all node pages of the radix tree. The mem_zone_bm_rtree
566 * structure itself is not freed here nor are the rtree_node
569 static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
570 int clear_nosave_free)
572 struct rtree_node *node;
574 list_for_each_entry(node, &zone->nodes, list)
575 free_image_page(node->data, clear_nosave_free);
577 list_for_each_entry(node, &zone->leaves, list)
578 free_image_page(node->data, clear_nosave_free);
581 static void memory_bm_position_reset(struct memory_bitmap *bm)
583 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
585 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
586 struct rtree_node, list);
587 bm->cur.node_pfn = 0;
588 bm->cur.node_bit = 0;
591 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
594 struct list_head hook;
600 * free_mem_extents - Free a list of memory extents.
601 * @list: List of extents to free.
603 static void free_mem_extents(struct list_head *list)
605 struct mem_extent *ext, *aux;
607 list_for_each_entry_safe(ext, aux, list, hook) {
608 list_del(&ext->hook);
614 * create_mem_extents - Create a list of memory extents.
615 * @list: List to put the extents into.
616 * @gfp_mask: Mask to use for memory allocations.
618 * The extents represent contiguous ranges of PFNs.
620 static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
624 INIT_LIST_HEAD(list);
626 for_each_populated_zone(zone) {
627 unsigned long zone_start, zone_end;
628 struct mem_extent *ext, *cur, *aux;
630 zone_start = zone->zone_start_pfn;
631 zone_end = zone_end_pfn(zone);
633 list_for_each_entry(ext, list, hook)
634 if (zone_start <= ext->end)
637 if (&ext->hook == list || zone_end < ext->start) {
638 /* New extent is necessary */
639 struct mem_extent *new_ext;
641 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
643 free_mem_extents(list);
646 new_ext->start = zone_start;
647 new_ext->end = zone_end;
648 list_add_tail(&new_ext->hook, &ext->hook);
652 /* Merge this zone's range of PFNs with the existing one */
653 if (zone_start < ext->start)
654 ext->start = zone_start;
655 if (zone_end > ext->end)
658 /* More merging may be possible */
660 list_for_each_entry_safe_continue(cur, aux, list, hook) {
661 if (zone_end < cur->start)
663 if (zone_end < cur->end)
665 list_del(&cur->hook);
674 * memory_bm_create - Allocate memory for a memory bitmap.
676 static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
679 struct chain_allocator ca;
680 struct list_head mem_extents;
681 struct mem_extent *ext;
684 chain_init(&ca, gfp_mask, safe_needed);
685 INIT_LIST_HEAD(&bm->zones);
687 error = create_mem_extents(&mem_extents, gfp_mask);
691 list_for_each_entry(ext, &mem_extents, hook) {
692 struct mem_zone_bm_rtree *zone;
694 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
695 ext->start, ext->end);
700 list_add_tail(&zone->list, &bm->zones);
703 bm->p_list = ca.chain;
704 memory_bm_position_reset(bm);
706 free_mem_extents(&mem_extents);
710 bm->p_list = ca.chain;
711 memory_bm_free(bm, PG_UNSAFE_CLEAR);
716 * memory_bm_free - Free memory occupied by the memory bitmap.
717 * @bm: Memory bitmap.
719 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
721 struct mem_zone_bm_rtree *zone;
723 list_for_each_entry(zone, &bm->zones, list)
724 free_zone_bm_rtree(zone, clear_nosave_free);
726 free_list_of_pages(bm->p_list, clear_nosave_free);
728 INIT_LIST_HEAD(&bm->zones);
732 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
734 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
735 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
737 * Walk the radix tree to find the page containing the bit that represents @pfn
738 * and return the position of the bit in @addr and @bit_nr.
740 static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
741 void **addr, unsigned int *bit_nr)
743 struct mem_zone_bm_rtree *curr, *zone;
744 struct rtree_node *node;
749 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
754 /* Find the right zone */
755 list_for_each_entry(curr, &bm->zones, list) {
756 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
767 * We have found the zone. Now walk the radix tree to find the leaf node
772 * If the zone we wish to scan is the current zone and the
773 * pfn falls into the current node then we do not need to walk
777 if (zone == bm->cur.zone &&
778 ((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
782 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
784 for (i = zone->levels; i > 0; i--) {
787 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
788 index &= BM_RTREE_LEVEL_MASK;
789 BUG_ON(node->data[index] == 0);
790 node = (struct rtree_node *)node->data[index];
794 /* Update last position */
797 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
799 /* Set return values */
801 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
806 static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
812 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
817 static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
823 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
830 static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
836 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
838 clear_bit(bit, addr);
841 static void memory_bm_clear_current(struct memory_bitmap *bm)
845 bit = max(bm->cur.node_bit - 1, 0);
846 clear_bit(bit, bm->cur.node->data);
849 static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
855 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
857 return test_bit(bit, addr);
860 static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
865 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
869 * rtree_next_node - Jump to the next leaf node.
871 * Set the position to the beginning of the next node in the
872 * memory bitmap. This is either the next node in the current
873 * zone's radix tree or the first node in the radix tree of the
876 * Return true if there is a next node, false otherwise.
878 static bool rtree_next_node(struct memory_bitmap *bm)
880 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
881 bm->cur.node = list_entry(bm->cur.node->list.next,
882 struct rtree_node, list);
883 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
884 bm->cur.node_bit = 0;
885 touch_softlockup_watchdog();
889 /* No more nodes, goto next zone */
890 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
891 bm->cur.zone = list_entry(bm->cur.zone->list.next,
892 struct mem_zone_bm_rtree, list);
893 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
894 struct rtree_node, list);
895 bm->cur.node_pfn = 0;
896 bm->cur.node_bit = 0;
905 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
906 * @bm: Memory bitmap.
908 * Starting from the last returned position this function searches for the next
909 * set bit in @bm and returns the PFN represented by it. If no more bits are
910 * set, BM_END_OF_MAP is returned.
912 * It is required to run memory_bm_position_reset() before the first call to
913 * this function for the given memory bitmap.
915 static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
917 unsigned long bits, pfn, pages;
921 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
922 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
923 bit = find_next_bit(bm->cur.node->data, bits,
926 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
927 bm->cur.node_bit = bit + 1;
930 } while (rtree_next_node(bm));
932 return BM_END_OF_MAP;
936 * This structure represents a range of page frames the contents of which
937 * should not be saved during hibernation.
939 struct nosave_region {
940 struct list_head list;
941 unsigned long start_pfn;
942 unsigned long end_pfn;
945 static LIST_HEAD(nosave_regions);
947 static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
949 struct rtree_node *node;
951 list_for_each_entry(node, &zone->nodes, list)
952 recycle_safe_page(node->data);
954 list_for_each_entry(node, &zone->leaves, list)
955 recycle_safe_page(node->data);
958 static void memory_bm_recycle(struct memory_bitmap *bm)
960 struct mem_zone_bm_rtree *zone;
961 struct linked_page *p_list;
963 list_for_each_entry(zone, &bm->zones, list)
964 recycle_zone_bm_rtree(zone);
968 struct linked_page *lp = p_list;
971 recycle_safe_page(lp);
976 * register_nosave_region - Register a region of unsaveable memory.
978 * Register a range of page frames the contents of which should not be saved
979 * during hibernation (to be used in the early initialization code).
981 void __init __register_nosave_region(unsigned long start_pfn,
982 unsigned long end_pfn, int use_kmalloc)
984 struct nosave_region *region;
986 if (start_pfn >= end_pfn)
989 if (!list_empty(&nosave_regions)) {
990 /* Try to extend the previous region (they should be sorted) */
991 region = list_entry(nosave_regions.prev,
992 struct nosave_region, list);
993 if (region->end_pfn == start_pfn) {
994 region->end_pfn = end_pfn;
999 /* During init, this shouldn't fail */
1000 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
1003 /* This allocation cannot fail */
1004 region = memblock_alloc(sizeof(struct nosave_region),
1007 panic("%s: Failed to allocate %zu bytes\n", __func__,
1008 sizeof(struct nosave_region));
1010 region->start_pfn = start_pfn;
1011 region->end_pfn = end_pfn;
1012 list_add_tail(®ion->list, &nosave_regions);
1014 pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
1015 (unsigned long long) start_pfn << PAGE_SHIFT,
1016 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1020 * Set bits in this map correspond to the page frames the contents of which
1021 * should not be saved during the suspend.
1023 static struct memory_bitmap *forbidden_pages_map;
1025 /* Set bits in this map correspond to free page frames. */
1026 static struct memory_bitmap *free_pages_map;
1029 * Each page frame allocated for creating the image is marked by setting the
1030 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
1033 void swsusp_set_page_free(struct page *page)
1036 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
1039 static int swsusp_page_is_free(struct page *page)
1041 return free_pages_map ?
1042 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1045 void swsusp_unset_page_free(struct page *page)
1048 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1051 static void swsusp_set_page_forbidden(struct page *page)
1053 if (forbidden_pages_map)
1054 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1057 int swsusp_page_is_forbidden(struct page *page)
1059 return forbidden_pages_map ?
1060 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1063 static void swsusp_unset_page_forbidden(struct page *page)
1065 if (forbidden_pages_map)
1066 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1070 * mark_nosave_pages - Mark pages that should not be saved.
1071 * @bm: Memory bitmap.
1073 * Set the bits in @bm that correspond to the page frames the contents of which
1074 * should not be saved.
1076 static void mark_nosave_pages(struct memory_bitmap *bm)
1078 struct nosave_region *region;
1080 if (list_empty(&nosave_regions))
1083 list_for_each_entry(region, &nosave_regions, list) {
1086 pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1087 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1088 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1091 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1092 if (pfn_valid(pfn)) {
1094 * It is safe to ignore the result of
1095 * mem_bm_set_bit_check() here, since we won't
1096 * touch the PFNs for which the error is
1099 mem_bm_set_bit_check(bm, pfn);
1105 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1107 * Create bitmaps needed for marking page frames that should not be saved and
1108 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1109 * only modified if everything goes well, because we don't want the bits to be
1110 * touched before both bitmaps are set up.
1112 int create_basic_memory_bitmaps(void)
1114 struct memory_bitmap *bm1, *bm2;
1117 if (forbidden_pages_map && free_pages_map)
1120 BUG_ON(forbidden_pages_map || free_pages_map);
1122 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1126 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1128 goto Free_first_object;
1130 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1132 goto Free_first_bitmap;
1134 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1136 goto Free_second_object;
1138 forbidden_pages_map = bm1;
1139 free_pages_map = bm2;
1140 mark_nosave_pages(forbidden_pages_map);
1142 pr_debug("Basic memory bitmaps created\n");
1149 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1156 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1158 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1159 * auxiliary pointers are necessary so that the bitmaps themselves are not
1160 * referred to while they are being freed.
1162 void free_basic_memory_bitmaps(void)
1164 struct memory_bitmap *bm1, *bm2;
1166 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1169 bm1 = forbidden_pages_map;
1170 bm2 = free_pages_map;
1171 forbidden_pages_map = NULL;
1172 free_pages_map = NULL;
1173 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1175 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1178 pr_debug("Basic memory bitmaps freed\n");
1181 void clear_free_pages(void)
1183 struct memory_bitmap *bm = free_pages_map;
1186 if (WARN_ON(!(free_pages_map)))
1189 if (IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) || want_init_on_free()) {
1190 memory_bm_position_reset(bm);
1191 pfn = memory_bm_next_pfn(bm);
1192 while (pfn != BM_END_OF_MAP) {
1194 clear_highpage(pfn_to_page(pfn));
1196 pfn = memory_bm_next_pfn(bm);
1198 memory_bm_position_reset(bm);
1199 pr_info("free pages cleared after restore\n");
1204 * snapshot_additional_pages - Estimate the number of extra pages needed.
1205 * @zone: Memory zone to carry out the computation for.
1207 * Estimate the number of additional pages needed for setting up a hibernation
1208 * image data structures for @zone (usually, the returned value is greater than
1209 * the exact number).
1211 unsigned int snapshot_additional_pages(struct zone *zone)
1213 unsigned int rtree, nodes;
1215 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1216 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1217 LINKED_PAGE_DATA_SIZE);
1219 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1226 #ifdef CONFIG_HIGHMEM
1228 * count_free_highmem_pages - Compute the total number of free highmem pages.
1230 * The returned number is system-wide.
1232 static unsigned int count_free_highmem_pages(void)
1235 unsigned int cnt = 0;
1237 for_each_populated_zone(zone)
1238 if (is_highmem(zone))
1239 cnt += zone_page_state(zone, NR_FREE_PAGES);
1245 * saveable_highmem_page - Check if a highmem page is saveable.
1247 * Determine whether a highmem page should be included in a hibernation image.
1249 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1250 * and it isn't part of a free chunk of pages.
1252 static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1256 if (!pfn_valid(pfn))
1259 page = pfn_to_online_page(pfn);
1260 if (!page || page_zone(page) != zone)
1263 BUG_ON(!PageHighMem(page));
1265 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1268 if (PageReserved(page) || PageOffline(page))
1271 if (page_is_guard(page))
1278 * count_highmem_pages - Compute the total number of saveable highmem pages.
1280 static unsigned int count_highmem_pages(void)
1285 for_each_populated_zone(zone) {
1286 unsigned long pfn, max_zone_pfn;
1288 if (!is_highmem(zone))
1291 mark_free_pages(zone);
1292 max_zone_pfn = zone_end_pfn(zone);
1293 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1294 if (saveable_highmem_page(zone, pfn))
1300 static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1304 #endif /* CONFIG_HIGHMEM */
1307 * saveable_page - Check if the given page is saveable.
1309 * Determine whether a non-highmem page should be included in a hibernation
1312 * We should save the page if it isn't Nosave, and is not in the range
1313 * of pages statically defined as 'unsaveable', and it isn't part of
1314 * a free chunk of pages.
1316 static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1320 if (!pfn_valid(pfn))
1323 page = pfn_to_online_page(pfn);
1324 if (!page || page_zone(page) != zone)
1327 BUG_ON(PageHighMem(page));
1329 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1332 if (PageOffline(page))
1335 if (PageReserved(page)
1336 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1339 if (page_is_guard(page))
1346 * count_data_pages - Compute the total number of saveable non-highmem pages.
1348 static unsigned int count_data_pages(void)
1351 unsigned long pfn, max_zone_pfn;
1354 for_each_populated_zone(zone) {
1355 if (is_highmem(zone))
1358 mark_free_pages(zone);
1359 max_zone_pfn = zone_end_pfn(zone);
1360 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1361 if (saveable_page(zone, pfn))
1368 * This is needed, because copy_page and memcpy are not usable for copying
1371 static inline void do_copy_page(long *dst, long *src)
1375 for (n = PAGE_SIZE / sizeof(long); n; n--)
1380 * safe_copy_page - Copy a page in a safe way.
1382 * Check if the page we are going to copy is marked as present in the kernel
1383 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1384 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1385 * always returns 'true'.
1387 static void safe_copy_page(void *dst, struct page *s_page)
1389 if (kernel_page_present(s_page)) {
1390 do_copy_page(dst, page_address(s_page));
1392 hibernate_map_page(s_page);
1393 do_copy_page(dst, page_address(s_page));
1394 hibernate_unmap_page(s_page);
1398 #ifdef CONFIG_HIGHMEM
1399 static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1401 return is_highmem(zone) ?
1402 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1405 static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1407 struct page *s_page, *d_page;
1410 s_page = pfn_to_page(src_pfn);
1411 d_page = pfn_to_page(dst_pfn);
1412 if (PageHighMem(s_page)) {
1413 src = kmap_atomic(s_page);
1414 dst = kmap_atomic(d_page);
1415 do_copy_page(dst, src);
1419 if (PageHighMem(d_page)) {
1421 * The page pointed to by src may contain some kernel
1422 * data modified by kmap_atomic()
1424 safe_copy_page(buffer, s_page);
1425 dst = kmap_atomic(d_page);
1426 copy_page(dst, buffer);
1429 safe_copy_page(page_address(d_page), s_page);
1434 #define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1436 static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1438 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1439 pfn_to_page(src_pfn));
1441 #endif /* CONFIG_HIGHMEM */
1443 static void copy_data_pages(struct memory_bitmap *copy_bm,
1444 struct memory_bitmap *orig_bm)
1449 for_each_populated_zone(zone) {
1450 unsigned long max_zone_pfn;
1452 mark_free_pages(zone);
1453 max_zone_pfn = zone_end_pfn(zone);
1454 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1455 if (page_is_saveable(zone, pfn))
1456 memory_bm_set_bit(orig_bm, pfn);
1458 memory_bm_position_reset(orig_bm);
1459 memory_bm_position_reset(copy_bm);
1461 pfn = memory_bm_next_pfn(orig_bm);
1462 if (unlikely(pfn == BM_END_OF_MAP))
1464 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1468 /* Total number of image pages */
1469 static unsigned int nr_copy_pages;
1470 /* Number of pages needed for saving the original pfns of the image pages */
1471 static unsigned int nr_meta_pages;
1473 * Numbers of normal and highmem page frames allocated for hibernation image
1474 * before suspending devices.
1476 static unsigned int alloc_normal, alloc_highmem;
1478 * Memory bitmap used for marking saveable pages (during hibernation) or
1479 * hibernation image pages (during restore)
1481 static struct memory_bitmap orig_bm;
1483 * Memory bitmap used during hibernation for marking allocated page frames that
1484 * will contain copies of saveable pages. During restore it is initially used
1485 * for marking hibernation image pages, but then the set bits from it are
1486 * duplicated in @orig_bm and it is released. On highmem systems it is next
1487 * used for marking "safe" highmem pages, but it has to be reinitialized for
1490 static struct memory_bitmap copy_bm;
1493 * swsusp_free - Free pages allocated for hibernation image.
1495 * Image pages are alocated before snapshot creation, so they need to be
1496 * released after resume.
1498 void swsusp_free(void)
1500 unsigned long fb_pfn, fr_pfn;
1502 if (!forbidden_pages_map || !free_pages_map)
1505 memory_bm_position_reset(forbidden_pages_map);
1506 memory_bm_position_reset(free_pages_map);
1509 fr_pfn = memory_bm_next_pfn(free_pages_map);
1510 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1513 * Find the next bit set in both bitmaps. This is guaranteed to
1514 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1517 if (fb_pfn < fr_pfn)
1518 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1519 if (fr_pfn < fb_pfn)
1520 fr_pfn = memory_bm_next_pfn(free_pages_map);
1521 } while (fb_pfn != fr_pfn);
1523 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1524 struct page *page = pfn_to_page(fr_pfn);
1526 memory_bm_clear_current(forbidden_pages_map);
1527 memory_bm_clear_current(free_pages_map);
1528 hibernate_restore_unprotect_page(page_address(page));
1536 restore_pblist = NULL;
1540 hibernate_restore_protection_end();
1543 /* Helper functions used for the shrinking of memory. */
1545 #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1548 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1549 * @nr_pages: Number of page frames to allocate.
1550 * @mask: GFP flags to use for the allocation.
1552 * Return value: Number of page frames actually allocated
1554 static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1556 unsigned long nr_alloc = 0;
1558 while (nr_pages > 0) {
1561 page = alloc_image_page(mask);
1564 memory_bm_set_bit(©_bm, page_to_pfn(page));
1565 if (PageHighMem(page))
1576 static unsigned long preallocate_image_memory(unsigned long nr_pages,
1577 unsigned long avail_normal)
1579 unsigned long alloc;
1581 if (avail_normal <= alloc_normal)
1584 alloc = avail_normal - alloc_normal;
1585 if (nr_pages < alloc)
1588 return preallocate_image_pages(alloc, GFP_IMAGE);
1591 #ifdef CONFIG_HIGHMEM
1592 static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1594 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1598 * __fraction - Compute (an approximation of) x * (multiplier / base).
1600 static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1602 return div64_u64(x * multiplier, base);
1605 static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1606 unsigned long highmem,
1607 unsigned long total)
1609 unsigned long alloc = __fraction(nr_pages, highmem, total);
1611 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1613 #else /* CONFIG_HIGHMEM */
1614 static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1619 static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1620 unsigned long highmem,
1621 unsigned long total)
1625 #endif /* CONFIG_HIGHMEM */
1628 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1630 static unsigned long free_unnecessary_pages(void)
1632 unsigned long save, to_free_normal, to_free_highmem, free;
1634 save = count_data_pages();
1635 if (alloc_normal >= save) {
1636 to_free_normal = alloc_normal - save;
1640 save -= alloc_normal;
1642 save += count_highmem_pages();
1643 if (alloc_highmem >= save) {
1644 to_free_highmem = alloc_highmem - save;
1646 to_free_highmem = 0;
1647 save -= alloc_highmem;
1648 if (to_free_normal > save)
1649 to_free_normal -= save;
1653 free = to_free_normal + to_free_highmem;
1655 memory_bm_position_reset(©_bm);
1657 while (to_free_normal > 0 || to_free_highmem > 0) {
1658 unsigned long pfn = memory_bm_next_pfn(©_bm);
1659 struct page *page = pfn_to_page(pfn);
1661 if (PageHighMem(page)) {
1662 if (!to_free_highmem)
1667 if (!to_free_normal)
1672 memory_bm_clear_bit(©_bm, pfn);
1673 swsusp_unset_page_forbidden(page);
1674 swsusp_unset_page_free(page);
1682 * minimum_image_size - Estimate the minimum acceptable size of an image.
1683 * @saveable: Number of saveable pages in the system.
1685 * We want to avoid attempting to free too much memory too hard, so estimate the
1686 * minimum acceptable size of a hibernation image to use as the lower limit for
1687 * preallocating memory.
1689 * We assume that the minimum image size should be proportional to
1691 * [number of saveable pages] - [number of pages that can be freed in theory]
1693 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1694 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1696 static unsigned long minimum_image_size(unsigned long saveable)
1700 size = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B)
1701 + global_node_page_state(NR_ACTIVE_ANON)
1702 + global_node_page_state(NR_INACTIVE_ANON)
1703 + global_node_page_state(NR_ACTIVE_FILE)
1704 + global_node_page_state(NR_INACTIVE_FILE);
1706 return saveable <= size ? 0 : saveable - size;
1710 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1712 * To create a hibernation image it is necessary to make a copy of every page
1713 * frame in use. We also need a number of page frames to be free during
1714 * hibernation for allocations made while saving the image and for device
1715 * drivers, in case they need to allocate memory from their hibernation
1716 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1717 * estimate) and reserved_size divided by PAGE_SIZE (which is tunable through
1718 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1719 * total number of available page frames and allocate at least
1721 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1722 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1724 * of them, which corresponds to the maximum size of a hibernation image.
1726 * If image_size is set below the number following from the above formula,
1727 * the preallocation of memory is continued until the total number of saveable
1728 * pages in the system is below the requested image size or the minimum
1729 * acceptable image size returned by minimum_image_size(), whichever is greater.
1731 int hibernate_preallocate_memory(void)
1734 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1735 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1736 ktime_t start, stop;
1739 pr_info("Preallocating image memory\n");
1740 start = ktime_get();
1742 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1744 pr_err("Cannot allocate original bitmap\n");
1748 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1750 pr_err("Cannot allocate copy bitmap\n");
1757 /* Count the number of saveable data pages. */
1758 save_highmem = count_highmem_pages();
1759 saveable = count_data_pages();
1762 * Compute the total number of page frames we can use (count) and the
1763 * number of pages needed for image metadata (size).
1766 saveable += save_highmem;
1767 highmem = save_highmem;
1769 for_each_populated_zone(zone) {
1770 size += snapshot_additional_pages(zone);
1771 if (is_highmem(zone))
1772 highmem += zone_page_state(zone, NR_FREE_PAGES);
1774 count += zone_page_state(zone, NR_FREE_PAGES);
1776 avail_normal = count;
1778 count -= totalreserve_pages;
1780 /* Compute the maximum number of saveable pages to leave in memory. */
1781 max_size = (count - (size + PAGES_FOR_IO)) / 2
1782 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1783 /* Compute the desired number of image pages specified by image_size. */
1784 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1785 if (size > max_size)
1788 * If the desired number of image pages is at least as large as the
1789 * current number of saveable pages in memory, allocate page frames for
1790 * the image and we're done.
1792 if (size >= saveable) {
1793 pages = preallocate_image_highmem(save_highmem);
1794 pages += preallocate_image_memory(saveable - pages, avail_normal);
1798 /* Estimate the minimum size of the image. */
1799 pages = minimum_image_size(saveable);
1801 * To avoid excessive pressure on the normal zone, leave room in it to
1802 * accommodate an image of the minimum size (unless it's already too
1803 * small, in which case don't preallocate pages from it at all).
1805 if (avail_normal > pages)
1806 avail_normal -= pages;
1810 size = min_t(unsigned long, pages, max_size);
1813 * Let the memory management subsystem know that we're going to need a
1814 * large number of page frames to allocate and make it free some memory.
1815 * NOTE: If this is not done, performance will be hurt badly in some
1818 shrink_all_memory(saveable - size);
1821 * The number of saveable pages in memory was too high, so apply some
1822 * pressure to decrease it. First, make room for the largest possible
1823 * image and fail if that doesn't work. Next, try to decrease the size
1824 * of the image as much as indicated by 'size' using allocations from
1825 * highmem and non-highmem zones separately.
1827 pages_highmem = preallocate_image_highmem(highmem / 2);
1828 alloc = count - max_size;
1829 if (alloc > pages_highmem)
1830 alloc -= pages_highmem;
1833 pages = preallocate_image_memory(alloc, avail_normal);
1834 if (pages < alloc) {
1835 /* We have exhausted non-highmem pages, try highmem. */
1837 pages += pages_highmem;
1838 pages_highmem = preallocate_image_highmem(alloc);
1839 if (pages_highmem < alloc) {
1840 pr_err("Image allocation is %lu pages short\n",
1841 alloc - pages_highmem);
1844 pages += pages_highmem;
1846 * size is the desired number of saveable pages to leave in
1847 * memory, so try to preallocate (all memory - size) pages.
1849 alloc = (count - pages) - size;
1850 pages += preallocate_image_highmem(alloc);
1853 * There are approximately max_size saveable pages at this point
1854 * and we want to reduce this number down to size.
1856 alloc = max_size - size;
1857 size = preallocate_highmem_fraction(alloc, highmem, count);
1858 pages_highmem += size;
1860 size = preallocate_image_memory(alloc, avail_normal);
1861 pages_highmem += preallocate_image_highmem(alloc - size);
1862 pages += pages_highmem + size;
1866 * We only need as many page frames for the image as there are saveable
1867 * pages in memory, but we have allocated more. Release the excessive
1870 pages -= free_unnecessary_pages();
1874 pr_info("Allocated %lu pages for snapshot\n", pages);
1875 swsusp_show_speed(start, stop, pages, "Allocated");
1884 #ifdef CONFIG_HIGHMEM
1886 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1888 * Compute the number of non-highmem pages that will be necessary for creating
1889 * copies of highmem pages.
1891 static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1893 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1895 if (free_highmem >= nr_highmem)
1898 nr_highmem -= free_highmem;
1903 static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1904 #endif /* CONFIG_HIGHMEM */
1907 * enough_free_mem - Check if there is enough free memory for the image.
1909 static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1912 unsigned int free = alloc_normal;
1914 for_each_populated_zone(zone)
1915 if (!is_highmem(zone))
1916 free += zone_page_state(zone, NR_FREE_PAGES);
1918 nr_pages += count_pages_for_highmem(nr_highmem);
1919 pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
1920 nr_pages, PAGES_FOR_IO, free);
1922 return free > nr_pages + PAGES_FOR_IO;
1925 #ifdef CONFIG_HIGHMEM
1927 * get_highmem_buffer - Allocate a buffer for highmem pages.
1929 * If there are some highmem pages in the hibernation image, we may need a
1930 * buffer to copy them and/or load their data.
1932 static inline int get_highmem_buffer(int safe_needed)
1934 buffer = get_image_page(GFP_ATOMIC, safe_needed);
1935 return buffer ? 0 : -ENOMEM;
1939 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1941 * Try to allocate as many pages as needed, but if the number of free highmem
1942 * pages is less than that, allocate them all.
1944 static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1945 unsigned int nr_highmem)
1947 unsigned int to_alloc = count_free_highmem_pages();
1949 if (to_alloc > nr_highmem)
1950 to_alloc = nr_highmem;
1952 nr_highmem -= to_alloc;
1953 while (to_alloc-- > 0) {
1956 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1957 memory_bm_set_bit(bm, page_to_pfn(page));
1962 static inline int get_highmem_buffer(int safe_needed) { return 0; }
1964 static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1965 unsigned int n) { return 0; }
1966 #endif /* CONFIG_HIGHMEM */
1969 * swsusp_alloc - Allocate memory for hibernation image.
1971 * We first try to allocate as many highmem pages as there are
1972 * saveable highmem pages in the system. If that fails, we allocate
1973 * non-highmem pages for the copies of the remaining highmem ones.
1975 * In this approach it is likely that the copies of highmem pages will
1976 * also be located in the high memory, because of the way in which
1977 * copy_data_pages() works.
1979 static int swsusp_alloc(struct memory_bitmap *copy_bm,
1980 unsigned int nr_pages, unsigned int nr_highmem)
1982 if (nr_highmem > 0) {
1983 if (get_highmem_buffer(PG_ANY))
1985 if (nr_highmem > alloc_highmem) {
1986 nr_highmem -= alloc_highmem;
1987 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1990 if (nr_pages > alloc_normal) {
1991 nr_pages -= alloc_normal;
1992 while (nr_pages-- > 0) {
1995 page = alloc_image_page(GFP_ATOMIC);
1998 memory_bm_set_bit(copy_bm, page_to_pfn(page));
2009 asmlinkage __visible int swsusp_save(void)
2011 unsigned int nr_pages, nr_highmem;
2013 pr_info("Creating image:\n");
2015 drain_local_pages(NULL);
2016 nr_pages = count_data_pages();
2017 nr_highmem = count_highmem_pages();
2018 pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
2020 if (!enough_free_mem(nr_pages, nr_highmem)) {
2021 pr_err("Not enough free memory\n");
2025 if (swsusp_alloc(©_bm, nr_pages, nr_highmem)) {
2026 pr_err("Memory allocation failed\n");
2031 * During allocating of suspend pagedir, new cold pages may appear.
2034 drain_local_pages(NULL);
2035 copy_data_pages(©_bm, &orig_bm);
2038 * End of critical section. From now on, we can write to memory,
2039 * but we should not touch disk. This specially means we must _not_
2040 * touch swap space! Except we must write out our image of course.
2043 nr_pages += nr_highmem;
2044 nr_copy_pages = nr_pages;
2045 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2047 pr_info("Image created (%d pages copied)\n", nr_pages);
2052 #ifndef CONFIG_ARCH_HIBERNATION_HEADER
2053 static int init_header_complete(struct swsusp_info *info)
2055 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2056 info->version_code = LINUX_VERSION_CODE;
2060 static const char *check_image_kernel(struct swsusp_info *info)
2062 if (info->version_code != LINUX_VERSION_CODE)
2063 return "kernel version";
2064 if (strcmp(info->uts.sysname,init_utsname()->sysname))
2065 return "system type";
2066 if (strcmp(info->uts.release,init_utsname()->release))
2067 return "kernel release";
2068 if (strcmp(info->uts.version,init_utsname()->version))
2070 if (strcmp(info->uts.machine,init_utsname()->machine))
2074 #endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2076 unsigned long snapshot_get_image_size(void)
2078 return nr_copy_pages + nr_meta_pages + 1;
2081 static int init_header(struct swsusp_info *info)
2083 memset(info, 0, sizeof(struct swsusp_info));
2084 info->num_physpages = get_num_physpages();
2085 info->image_pages = nr_copy_pages;
2086 info->pages = snapshot_get_image_size();
2087 info->size = info->pages;
2088 info->size <<= PAGE_SHIFT;
2089 return init_header_complete(info);
2093 * pack_pfns - Prepare PFNs for saving.
2094 * @bm: Memory bitmap.
2095 * @buf: Memory buffer to store the PFNs in.
2097 * PFNs corresponding to set bits in @bm are stored in the area of memory
2098 * pointed to by @buf (1 page at a time).
2100 static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2104 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2105 buf[j] = memory_bm_next_pfn(bm);
2106 if (unlikely(buf[j] == BM_END_OF_MAP))
2112 * snapshot_read_next - Get the address to read the next image page from.
2113 * @handle: Snapshot handle to be used for the reading.
2115 * On the first call, @handle should point to a zeroed snapshot_handle
2116 * structure. The structure gets populated then and a pointer to it should be
2117 * passed to this function every next time.
2119 * On success, the function returns a positive number. Then, the caller
2120 * is allowed to read up to the returned number of bytes from the memory
2121 * location computed by the data_of() macro.
2123 * The function returns 0 to indicate the end of the data stream condition,
2124 * and negative numbers are returned on errors. If that happens, the structure
2125 * pointed to by @handle is not updated and should not be used any more.
2127 int snapshot_read_next(struct snapshot_handle *handle)
2129 if (handle->cur > nr_meta_pages + nr_copy_pages)
2133 /* This makes the buffer be freed by swsusp_free() */
2134 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2141 error = init_header((struct swsusp_info *)buffer);
2144 handle->buffer = buffer;
2145 memory_bm_position_reset(&orig_bm);
2146 memory_bm_position_reset(©_bm);
2147 } else if (handle->cur <= nr_meta_pages) {
2149 pack_pfns(buffer, &orig_bm);
2153 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2154 if (PageHighMem(page)) {
2156 * Highmem pages are copied to the buffer,
2157 * because we can't return with a kmapped
2158 * highmem page (we may not be called again).
2162 kaddr = kmap_atomic(page);
2163 copy_page(buffer, kaddr);
2164 kunmap_atomic(kaddr);
2165 handle->buffer = buffer;
2167 handle->buffer = page_address(page);
2174 static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2175 struct memory_bitmap *src)
2179 memory_bm_position_reset(src);
2180 pfn = memory_bm_next_pfn(src);
2181 while (pfn != BM_END_OF_MAP) {
2182 memory_bm_set_bit(dst, pfn);
2183 pfn = memory_bm_next_pfn(src);
2188 * mark_unsafe_pages - Mark pages that were used before hibernation.
2190 * Mark the pages that cannot be used for storing the image during restoration,
2191 * because they conflict with the pages that had been used before hibernation.
2193 static void mark_unsafe_pages(struct memory_bitmap *bm)
2197 /* Clear the "free"/"unsafe" bit for all PFNs */
2198 memory_bm_position_reset(free_pages_map);
2199 pfn = memory_bm_next_pfn(free_pages_map);
2200 while (pfn != BM_END_OF_MAP) {
2201 memory_bm_clear_current(free_pages_map);
2202 pfn = memory_bm_next_pfn(free_pages_map);
2205 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2206 duplicate_memory_bitmap(free_pages_map, bm);
2208 allocated_unsafe_pages = 0;
2211 static int check_header(struct swsusp_info *info)
2215 reason = check_image_kernel(info);
2216 if (!reason && info->num_physpages != get_num_physpages())
2217 reason = "memory size";
2219 pr_err("Image mismatch: %s\n", reason);
2226 * load header - Check the image header and copy the data from it.
2228 static int load_header(struct swsusp_info *info)
2232 restore_pblist = NULL;
2233 error = check_header(info);
2235 nr_copy_pages = info->image_pages;
2236 nr_meta_pages = info->pages - info->image_pages - 1;
2242 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2243 * @bm: Memory bitmap.
2244 * @buf: Area of memory containing the PFNs.
2246 * For each element of the array pointed to by @buf (1 page at a time), set the
2247 * corresponding bit in @bm.
2249 static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2253 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2254 if (unlikely(buf[j] == BM_END_OF_MAP))
2257 if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2258 memory_bm_set_bit(bm, buf[j]);
2266 #ifdef CONFIG_HIGHMEM
2268 * struct highmem_pbe is used for creating the list of highmem pages that
2269 * should be restored atomically during the resume from disk, because the page
2270 * frames they have occupied before the suspend are in use.
2272 struct highmem_pbe {
2273 struct page *copy_page; /* data is here now */
2274 struct page *orig_page; /* data was here before the suspend */
2275 struct highmem_pbe *next;
2279 * List of highmem PBEs needed for restoring the highmem pages that were
2280 * allocated before the suspend and included in the suspend image, but have
2281 * also been allocated by the "resume" kernel, so their contents cannot be
2282 * written directly to their "original" page frames.
2284 static struct highmem_pbe *highmem_pblist;
2287 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2288 * @bm: Memory bitmap.
2290 * The bits in @bm that correspond to image pages are assumed to be set.
2292 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2295 unsigned int cnt = 0;
2297 memory_bm_position_reset(bm);
2298 pfn = memory_bm_next_pfn(bm);
2299 while (pfn != BM_END_OF_MAP) {
2300 if (PageHighMem(pfn_to_page(pfn)))
2303 pfn = memory_bm_next_pfn(bm);
2308 static unsigned int safe_highmem_pages;
2310 static struct memory_bitmap *safe_highmem_bm;
2313 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2314 * @bm: Pointer to an uninitialized memory bitmap structure.
2315 * @nr_highmem_p: Pointer to the number of highmem image pages.
2317 * Try to allocate as many highmem pages as there are highmem image pages
2318 * (@nr_highmem_p points to the variable containing the number of highmem image
2319 * pages). The pages that are "safe" (ie. will not be overwritten when the
2320 * hibernation image is restored entirely) have the corresponding bits set in
2321 * @bm (it must be unitialized).
2323 * NOTE: This function should not be called if there are no highmem image pages.
2325 static int prepare_highmem_image(struct memory_bitmap *bm,
2326 unsigned int *nr_highmem_p)
2328 unsigned int to_alloc;
2330 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2333 if (get_highmem_buffer(PG_SAFE))
2336 to_alloc = count_free_highmem_pages();
2337 if (to_alloc > *nr_highmem_p)
2338 to_alloc = *nr_highmem_p;
2340 *nr_highmem_p = to_alloc;
2342 safe_highmem_pages = 0;
2343 while (to_alloc-- > 0) {
2346 page = alloc_page(__GFP_HIGHMEM);
2347 if (!swsusp_page_is_free(page)) {
2348 /* The page is "safe", set its bit the bitmap */
2349 memory_bm_set_bit(bm, page_to_pfn(page));
2350 safe_highmem_pages++;
2352 /* Mark the page as allocated */
2353 swsusp_set_page_forbidden(page);
2354 swsusp_set_page_free(page);
2356 memory_bm_position_reset(bm);
2357 safe_highmem_bm = bm;
2361 static struct page *last_highmem_page;
2364 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2366 * For a given highmem image page get a buffer that suspend_write_next() should
2367 * return to its caller to write to.
2369 * If the page is to be saved to its "original" page frame or a copy of
2370 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2371 * the copy of the page is to be made in normal memory, so the address of
2372 * the copy is returned.
2374 * If @buffer is returned, the caller of suspend_write_next() will write
2375 * the page's contents to @buffer, so they will have to be copied to the
2376 * right location on the next call to suspend_write_next() and it is done
2377 * with the help of copy_last_highmem_page(). For this purpose, if
2378 * @buffer is returned, @last_highmem_page is set to the page to which
2379 * the data will have to be copied from @buffer.
2381 static void *get_highmem_page_buffer(struct page *page,
2382 struct chain_allocator *ca)
2384 struct highmem_pbe *pbe;
2387 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2389 * We have allocated the "original" page frame and we can
2390 * use it directly to store the loaded page.
2392 last_highmem_page = page;
2396 * The "original" page frame has not been allocated and we have to
2397 * use a "safe" page frame to store the loaded page.
2399 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2402 return ERR_PTR(-ENOMEM);
2404 pbe->orig_page = page;
2405 if (safe_highmem_pages > 0) {
2408 /* Copy of the page will be stored in high memory */
2410 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2411 safe_highmem_pages--;
2412 last_highmem_page = tmp;
2413 pbe->copy_page = tmp;
2415 /* Copy of the page will be stored in normal memory */
2416 kaddr = safe_pages_list;
2417 safe_pages_list = safe_pages_list->next;
2418 pbe->copy_page = virt_to_page(kaddr);
2420 pbe->next = highmem_pblist;
2421 highmem_pblist = pbe;
2426 * copy_last_highmem_page - Copy most the most recent highmem image page.
2428 * Copy the contents of a highmem image from @buffer, where the caller of
2429 * snapshot_write_next() has stored them, to the right location represented by
2430 * @last_highmem_page .
2432 static void copy_last_highmem_page(void)
2434 if (last_highmem_page) {
2437 dst = kmap_atomic(last_highmem_page);
2438 copy_page(dst, buffer);
2440 last_highmem_page = NULL;
2444 static inline int last_highmem_page_copied(void)
2446 return !last_highmem_page;
2449 static inline void free_highmem_data(void)
2451 if (safe_highmem_bm)
2452 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2455 free_image_page(buffer, PG_UNSAFE_CLEAR);
2458 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2460 static inline int prepare_highmem_image(struct memory_bitmap *bm,
2461 unsigned int *nr_highmem_p) { return 0; }
2463 static inline void *get_highmem_page_buffer(struct page *page,
2464 struct chain_allocator *ca)
2466 return ERR_PTR(-EINVAL);
2469 static inline void copy_last_highmem_page(void) {}
2470 static inline int last_highmem_page_copied(void) { return 1; }
2471 static inline void free_highmem_data(void) {}
2472 #endif /* CONFIG_HIGHMEM */
2474 #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2477 * prepare_image - Make room for loading hibernation image.
2478 * @new_bm: Unitialized memory bitmap structure.
2479 * @bm: Memory bitmap with unsafe pages marked.
2481 * Use @bm to mark the pages that will be overwritten in the process of
2482 * restoring the system memory state from the suspend image ("unsafe" pages)
2483 * and allocate memory for the image.
2485 * The idea is to allocate a new memory bitmap first and then allocate
2486 * as many pages as needed for image data, but without specifying what those
2487 * pages will be used for just yet. Instead, we mark them all as allocated and
2488 * create a lists of "safe" pages to be used later. On systems with high
2489 * memory a list of "safe" highmem pages is created too.
2491 static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2493 unsigned int nr_pages, nr_highmem;
2494 struct linked_page *lp;
2497 /* If there is no highmem, the buffer will not be necessary */
2498 free_image_page(buffer, PG_UNSAFE_CLEAR);
2501 nr_highmem = count_highmem_image_pages(bm);
2502 mark_unsafe_pages(bm);
2504 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2508 duplicate_memory_bitmap(new_bm, bm);
2509 memory_bm_free(bm, PG_UNSAFE_KEEP);
2510 if (nr_highmem > 0) {
2511 error = prepare_highmem_image(bm, &nr_highmem);
2516 * Reserve some safe pages for potential later use.
2518 * NOTE: This way we make sure there will be enough safe pages for the
2519 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2520 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2522 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2524 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2525 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2526 while (nr_pages > 0) {
2527 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2532 lp->next = safe_pages_list;
2533 safe_pages_list = lp;
2536 /* Preallocate memory for the image */
2537 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2538 while (nr_pages > 0) {
2539 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2544 if (!swsusp_page_is_free(virt_to_page(lp))) {
2545 /* The page is "safe", add it to the list */
2546 lp->next = safe_pages_list;
2547 safe_pages_list = lp;
2549 /* Mark the page as allocated */
2550 swsusp_set_page_forbidden(virt_to_page(lp));
2551 swsusp_set_page_free(virt_to_page(lp));
2562 * get_buffer - Get the address to store the next image data page.
2564 * Get the address that snapshot_write_next() should return to its caller to
2567 static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2571 unsigned long pfn = memory_bm_next_pfn(bm);
2573 if (pfn == BM_END_OF_MAP)
2574 return ERR_PTR(-EFAULT);
2576 page = pfn_to_page(pfn);
2577 if (PageHighMem(page))
2578 return get_highmem_page_buffer(page, ca);
2580 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2582 * We have allocated the "original" page frame and we can
2583 * use it directly to store the loaded page.
2585 return page_address(page);
2588 * The "original" page frame has not been allocated and we have to
2589 * use a "safe" page frame to store the loaded page.
2591 pbe = chain_alloc(ca, sizeof(struct pbe));
2594 return ERR_PTR(-ENOMEM);
2596 pbe->orig_address = page_address(page);
2597 pbe->address = safe_pages_list;
2598 safe_pages_list = safe_pages_list->next;
2599 pbe->next = restore_pblist;
2600 restore_pblist = pbe;
2601 return pbe->address;
2605 * snapshot_write_next - Get the address to store the next image page.
2606 * @handle: Snapshot handle structure to guide the writing.
2608 * On the first call, @handle should point to a zeroed snapshot_handle
2609 * structure. The structure gets populated then and a pointer to it should be
2610 * passed to this function every next time.
2612 * On success, the function returns a positive number. Then, the caller
2613 * is allowed to write up to the returned number of bytes to the memory
2614 * location computed by the data_of() macro.
2616 * The function returns 0 to indicate the "end of file" condition. Negative
2617 * numbers are returned on errors, in which cases the structure pointed to by
2618 * @handle is not updated and should not be used any more.
2620 int snapshot_write_next(struct snapshot_handle *handle)
2622 static struct chain_allocator ca;
2625 /* Check if we have already loaded the entire image */
2626 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2629 handle->sync_read = 1;
2633 /* This makes the buffer be freed by swsusp_free() */
2634 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2639 handle->buffer = buffer;
2640 } else if (handle->cur == 1) {
2641 error = load_header(buffer);
2645 safe_pages_list = NULL;
2647 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2651 hibernate_restore_protection_begin();
2652 } else if (handle->cur <= nr_meta_pages + 1) {
2653 error = unpack_orig_pfns(buffer, ©_bm);
2657 if (handle->cur == nr_meta_pages + 1) {
2658 error = prepare_image(&orig_bm, ©_bm);
2662 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2663 memory_bm_position_reset(&orig_bm);
2664 restore_pblist = NULL;
2665 handle->buffer = get_buffer(&orig_bm, &ca);
2666 handle->sync_read = 0;
2667 if (IS_ERR(handle->buffer))
2668 return PTR_ERR(handle->buffer);
2671 copy_last_highmem_page();
2672 hibernate_restore_protect_page(handle->buffer);
2673 handle->buffer = get_buffer(&orig_bm, &ca);
2674 if (IS_ERR(handle->buffer))
2675 return PTR_ERR(handle->buffer);
2676 if (handle->buffer != buffer)
2677 handle->sync_read = 0;
2684 * snapshot_write_finalize - Complete the loading of a hibernation image.
2686 * Must be called after the last call to snapshot_write_next() in case the last
2687 * page in the image happens to be a highmem page and its contents should be
2688 * stored in highmem. Additionally, it recycles bitmap memory that's not
2689 * necessary any more.
2691 void snapshot_write_finalize(struct snapshot_handle *handle)
2693 copy_last_highmem_page();
2694 hibernate_restore_protect_page(handle->buffer);
2695 /* Do that only if we have loaded the image entirely */
2696 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2697 memory_bm_recycle(&orig_bm);
2698 free_highmem_data();
2702 int snapshot_image_loaded(struct snapshot_handle *handle)
2704 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2705 handle->cur <= nr_meta_pages + nr_copy_pages);
2708 #ifdef CONFIG_HIGHMEM
2709 /* Assumes that @buf is ready and points to a "safe" page */
2710 static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2713 void *kaddr1, *kaddr2;
2715 kaddr1 = kmap_atomic(p1);
2716 kaddr2 = kmap_atomic(p2);
2717 copy_page(buf, kaddr1);
2718 copy_page(kaddr1, kaddr2);
2719 copy_page(kaddr2, buf);
2720 kunmap_atomic(kaddr2);
2721 kunmap_atomic(kaddr1);
2725 * restore_highmem - Put highmem image pages into their original locations.
2727 * For each highmem page that was in use before hibernation and is included in
2728 * the image, and also has been allocated by the "restore" kernel, swap its
2729 * current contents with the previous (ie. "before hibernation") ones.
2731 * If the restore eventually fails, we can call this function once again and
2732 * restore the highmem state as seen by the restore kernel.
2734 int restore_highmem(void)
2736 struct highmem_pbe *pbe = highmem_pblist;
2742 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2747 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2750 free_image_page(buf, PG_UNSAFE_CLEAR);
2753 #endif /* CONFIG_HIGHMEM */