2 * linux/kernel/power/snapshot.c
4 * This file provides system snapshot/restore functionality for swsusp.
6 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
7 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
9 * This file is released under the GPLv2.
13 #include <linux/version.h>
14 #include <linux/module.h>
16 #include <linux/suspend.h>
17 #include <linux/delay.h>
18 #include <linux/bitops.h>
19 #include <linux/spinlock.h>
20 #include <linux/kernel.h>
22 #include <linux/device.h>
23 #include <linux/init.h>
24 #include <linux/bootmem.h>
25 #include <linux/syscalls.h>
26 #include <linux/console.h>
27 #include <linux/highmem.h>
28 #include <linux/list.h>
29 #include <linux/slab.h>
31 #include <asm/uaccess.h>
32 #include <asm/mmu_context.h>
33 #include <asm/pgtable.h>
34 #include <asm/tlbflush.h>
39 static int swsusp_page_is_free(struct page
*);
40 static void swsusp_set_page_forbidden(struct page
*);
41 static void swsusp_unset_page_forbidden(struct page
*);
44 * Number of bytes to reserve for memory allocations made by device drivers
45 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
46 * cause image creation to fail (tunable via /sys/power/reserved_size).
48 unsigned long reserved_size
;
50 void __init
hibernate_reserved_size_init(void)
52 reserved_size
= SPARE_PAGES
* PAGE_SIZE
;
56 * Preferred image size in bytes (tunable via /sys/power/image_size).
57 * When it is set to N, swsusp will do its best to ensure the image
58 * size will not exceed N bytes, but if that is impossible, it will
59 * try to create the smallest image possible.
61 unsigned long image_size
;
63 void __init
hibernate_image_size_init(void)
65 image_size
= ((totalram_pages
* 2) / 5) * PAGE_SIZE
;
68 /* List of PBEs needed for restoring the pages that were allocated before
69 * the suspend and included in the suspend image, but have also been
70 * allocated by the "resume" kernel, so their contents cannot be written
71 * directly to their "original" page frames.
73 struct pbe
*restore_pblist
;
75 /* Pointer to an auxiliary buffer (1 page) */
79 * @safe_needed - on resume, for storing the PBE list and the image,
80 * we can only use memory pages that do not conflict with the pages
81 * used before suspend. The unsafe pages have PageNosaveFree set
82 * and we count them using unsafe_pages.
84 * Each allocated image page is marked as PageNosave and PageNosaveFree
85 * so that swsusp_free() can release it.
90 #define PG_UNSAFE_CLEAR 1
91 #define PG_UNSAFE_KEEP 0
93 static unsigned int allocated_unsafe_pages
;
95 static void *get_image_page(gfp_t gfp_mask
, int safe_needed
)
99 res
= (void *)get_zeroed_page(gfp_mask
);
101 while (res
&& swsusp_page_is_free(virt_to_page(res
))) {
102 /* The page is unsafe, mark it for swsusp_free() */
103 swsusp_set_page_forbidden(virt_to_page(res
));
104 allocated_unsafe_pages
++;
105 res
= (void *)get_zeroed_page(gfp_mask
);
108 swsusp_set_page_forbidden(virt_to_page(res
));
109 swsusp_set_page_free(virt_to_page(res
));
114 unsigned long get_safe_page(gfp_t gfp_mask
)
116 return (unsigned long)get_image_page(gfp_mask
, PG_SAFE
);
119 static struct page
*alloc_image_page(gfp_t gfp_mask
)
123 page
= alloc_page(gfp_mask
);
125 swsusp_set_page_forbidden(page
);
126 swsusp_set_page_free(page
);
132 * free_image_page - free page represented by @addr, allocated with
133 * get_image_page (page flags set by it must be cleared)
136 static inline void free_image_page(void *addr
, int clear_nosave_free
)
140 BUG_ON(!virt_addr_valid(addr
));
142 page
= virt_to_page(addr
);
144 swsusp_unset_page_forbidden(page
);
145 if (clear_nosave_free
)
146 swsusp_unset_page_free(page
);
151 /* struct linked_page is used to build chains of pages */
153 #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
156 struct linked_page
*next
;
157 char data
[LINKED_PAGE_DATA_SIZE
];
158 } __attribute__((packed
));
161 free_list_of_pages(struct linked_page
*list
, int clear_page_nosave
)
164 struct linked_page
*lp
= list
->next
;
166 free_image_page(list
, clear_page_nosave
);
172 * struct chain_allocator is used for allocating small objects out of
173 * a linked list of pages called 'the chain'.
175 * The chain grows each time when there is no room for a new object in
176 * the current page. The allocated objects cannot be freed individually.
177 * It is only possible to free them all at once, by freeing the entire
180 * NOTE: The chain allocator may be inefficient if the allocated objects
181 * are not much smaller than PAGE_SIZE.
184 struct chain_allocator
{
185 struct linked_page
*chain
; /* the chain */
186 unsigned int used_space
; /* total size of objects allocated out
187 * of the current page
189 gfp_t gfp_mask
; /* mask for allocating pages */
190 int safe_needed
; /* if set, only "safe" pages are allocated */
194 chain_init(struct chain_allocator
*ca
, gfp_t gfp_mask
, int safe_needed
)
197 ca
->used_space
= LINKED_PAGE_DATA_SIZE
;
198 ca
->gfp_mask
= gfp_mask
;
199 ca
->safe_needed
= safe_needed
;
202 static void *chain_alloc(struct chain_allocator
*ca
, unsigned int size
)
206 if (LINKED_PAGE_DATA_SIZE
- ca
->used_space
< size
) {
207 struct linked_page
*lp
;
209 lp
= get_image_page(ca
->gfp_mask
, ca
->safe_needed
);
213 lp
->next
= ca
->chain
;
217 ret
= ca
->chain
->data
+ ca
->used_space
;
218 ca
->used_space
+= size
;
223 * Data types related to memory bitmaps.
225 * Memory bitmap is a structure consiting of many linked lists of
226 * objects. The main list's elements are of type struct zone_bitmap
227 * and each of them corresonds to one zone. For each zone bitmap
228 * object there is a list of objects of type struct bm_block that
229 * represent each blocks of bitmap in which information is stored.
231 * struct memory_bitmap contains a pointer to the main list of zone
232 * bitmap objects, a struct bm_position used for browsing the bitmap,
233 * and a pointer to the list of pages used for allocating all of the
234 * zone bitmap objects and bitmap block objects.
236 * NOTE: It has to be possible to lay out the bitmap in memory
237 * using only allocations of order 0. Additionally, the bitmap is
238 * designed to work with arbitrary number of zones (this is over the
239 * top for now, but let's avoid making unnecessary assumptions ;-).
241 * struct zone_bitmap contains a pointer to a list of bitmap block
242 * objects and a pointer to the bitmap block object that has been
243 * most recently used for setting bits. Additionally, it contains the
244 * pfns that correspond to the start and end of the represented zone.
246 * struct bm_block contains a pointer to the memory page in which
247 * information is stored (in the form of a block of bitmap)
248 * It also contains the pfns that correspond to the start and end of
249 * the represented memory area.
252 #define BM_END_OF_MAP (~0UL)
254 #define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
257 struct list_head hook
; /* hook into a list of bitmap blocks */
258 unsigned long start_pfn
; /* pfn represented by the first bit */
259 unsigned long end_pfn
; /* pfn represented by the last bit plus 1 */
260 unsigned long *data
; /* bitmap representing pages */
263 static inline unsigned long bm_block_bits(struct bm_block
*bb
)
265 return bb
->end_pfn
- bb
->start_pfn
;
268 /* strcut bm_position is used for browsing memory bitmaps */
271 struct bm_block
*block
;
275 struct memory_bitmap
{
276 struct list_head blocks
; /* list of bitmap blocks */
277 struct linked_page
*p_list
; /* list of pages used to store zone
278 * bitmap objects and bitmap block
281 struct bm_position cur
; /* most recently used bit position */
284 /* Functions that operate on memory bitmaps */
286 static void memory_bm_position_reset(struct memory_bitmap
*bm
)
288 bm
->cur
.block
= list_entry(bm
->blocks
.next
, struct bm_block
, hook
);
292 static void memory_bm_free(struct memory_bitmap
*bm
, int clear_nosave_free
);
295 * create_bm_block_list - create a list of block bitmap objects
296 * @pages - number of pages to track
297 * @list - list to put the allocated blocks into
298 * @ca - chain allocator to be used for allocating memory
300 static int create_bm_block_list(unsigned long pages
,
301 struct list_head
*list
,
302 struct chain_allocator
*ca
)
304 unsigned int nr_blocks
= DIV_ROUND_UP(pages
, BM_BITS_PER_BLOCK
);
306 while (nr_blocks
-- > 0) {
309 bb
= chain_alloc(ca
, sizeof(struct bm_block
));
312 list_add(&bb
->hook
, list
);
319 struct list_head hook
;
325 * free_mem_extents - free a list of memory extents
326 * @list - list of extents to empty
328 static void free_mem_extents(struct list_head
*list
)
330 struct mem_extent
*ext
, *aux
;
332 list_for_each_entry_safe(ext
, aux
, list
, hook
) {
333 list_del(&ext
->hook
);
339 * create_mem_extents - create a list of memory extents representing
340 * contiguous ranges of PFNs
341 * @list - list to put the extents into
342 * @gfp_mask - mask to use for memory allocations
344 static int create_mem_extents(struct list_head
*list
, gfp_t gfp_mask
)
348 INIT_LIST_HEAD(list
);
350 for_each_populated_zone(zone
) {
351 unsigned long zone_start
, zone_end
;
352 struct mem_extent
*ext
, *cur
, *aux
;
354 zone_start
= zone
->zone_start_pfn
;
355 zone_end
= zone_end_pfn(zone
);
357 list_for_each_entry(ext
, list
, hook
)
358 if (zone_start
<= ext
->end
)
361 if (&ext
->hook
== list
|| zone_end
< ext
->start
) {
362 /* New extent is necessary */
363 struct mem_extent
*new_ext
;
365 new_ext
= kzalloc(sizeof(struct mem_extent
), gfp_mask
);
367 free_mem_extents(list
);
370 new_ext
->start
= zone_start
;
371 new_ext
->end
= zone_end
;
372 list_add_tail(&new_ext
->hook
, &ext
->hook
);
376 /* Merge this zone's range of PFNs with the existing one */
377 if (zone_start
< ext
->start
)
378 ext
->start
= zone_start
;
379 if (zone_end
> ext
->end
)
382 /* More merging may be possible */
384 list_for_each_entry_safe_continue(cur
, aux
, list
, hook
) {
385 if (zone_end
< cur
->start
)
387 if (zone_end
< cur
->end
)
389 list_del(&cur
->hook
);
398 * memory_bm_create - allocate memory for a memory bitmap
401 memory_bm_create(struct memory_bitmap
*bm
, gfp_t gfp_mask
, int safe_needed
)
403 struct chain_allocator ca
;
404 struct list_head mem_extents
;
405 struct mem_extent
*ext
;
408 chain_init(&ca
, gfp_mask
, safe_needed
);
409 INIT_LIST_HEAD(&bm
->blocks
);
411 error
= create_mem_extents(&mem_extents
, gfp_mask
);
415 list_for_each_entry(ext
, &mem_extents
, hook
) {
417 unsigned long pfn
= ext
->start
;
418 unsigned long pages
= ext
->end
- ext
->start
;
420 bb
= list_entry(bm
->blocks
.prev
, struct bm_block
, hook
);
422 error
= create_bm_block_list(pages
, bm
->blocks
.prev
, &ca
);
426 list_for_each_entry_continue(bb
, &bm
->blocks
, hook
) {
427 bb
->data
= get_image_page(gfp_mask
, safe_needed
);
434 if (pages
>= BM_BITS_PER_BLOCK
) {
435 pfn
+= BM_BITS_PER_BLOCK
;
436 pages
-= BM_BITS_PER_BLOCK
;
438 /* This is executed only once in the loop */
445 bm
->p_list
= ca
.chain
;
446 memory_bm_position_reset(bm
);
448 free_mem_extents(&mem_extents
);
452 bm
->p_list
= ca
.chain
;
453 memory_bm_free(bm
, PG_UNSAFE_CLEAR
);
458 * memory_bm_free - free memory occupied by the memory bitmap @bm
460 static void memory_bm_free(struct memory_bitmap
*bm
, int clear_nosave_free
)
464 list_for_each_entry(bb
, &bm
->blocks
, hook
)
466 free_image_page(bb
->data
, clear_nosave_free
);
468 free_list_of_pages(bm
->p_list
, clear_nosave_free
);
470 INIT_LIST_HEAD(&bm
->blocks
);
474 * memory_bm_find_bit - find the bit in the bitmap @bm that corresponds
475 * to given pfn. The cur_zone_bm member of @bm and the cur_block member
476 * of @bm->cur_zone_bm are updated.
478 static int memory_bm_find_bit(struct memory_bitmap
*bm
, unsigned long pfn
,
479 void **addr
, unsigned int *bit_nr
)
484 * Check if the pfn corresponds to the current bitmap block and find
485 * the block where it fits if this is not the case.
488 if (pfn
< bb
->start_pfn
)
489 list_for_each_entry_continue_reverse(bb
, &bm
->blocks
, hook
)
490 if (pfn
>= bb
->start_pfn
)
493 if (pfn
>= bb
->end_pfn
)
494 list_for_each_entry_continue(bb
, &bm
->blocks
, hook
)
495 if (pfn
>= bb
->start_pfn
&& pfn
< bb
->end_pfn
)
498 if (&bb
->hook
== &bm
->blocks
)
501 /* The block has been found */
503 pfn
-= bb
->start_pfn
;
504 bm
->cur
.bit
= pfn
+ 1;
510 static void memory_bm_set_bit(struct memory_bitmap
*bm
, unsigned long pfn
)
516 error
= memory_bm_find_bit(bm
, pfn
, &addr
, &bit
);
521 static int mem_bm_set_bit_check(struct memory_bitmap
*bm
, unsigned long pfn
)
527 error
= memory_bm_find_bit(bm
, pfn
, &addr
, &bit
);
533 static void memory_bm_clear_bit(struct memory_bitmap
*bm
, unsigned long pfn
)
539 error
= memory_bm_find_bit(bm
, pfn
, &addr
, &bit
);
541 clear_bit(bit
, addr
);
544 static int memory_bm_test_bit(struct memory_bitmap
*bm
, unsigned long pfn
)
550 error
= memory_bm_find_bit(bm
, pfn
, &addr
, &bit
);
552 return test_bit(bit
, addr
);
555 static bool memory_bm_pfn_present(struct memory_bitmap
*bm
, unsigned long pfn
)
560 return !memory_bm_find_bit(bm
, pfn
, &addr
, &bit
);
564 * memory_bm_next_pfn - find the pfn that corresponds to the next set bit
565 * in the bitmap @bm. If the pfn cannot be found, BM_END_OF_MAP is
568 * It is required to run memory_bm_position_reset() before the first call to
572 static unsigned long memory_bm_next_pfn(struct memory_bitmap
*bm
)
580 bit
= find_next_bit(bb
->data
, bm_block_bits(bb
), bit
);
581 if (bit
< bm_block_bits(bb
))
584 bb
= list_entry(bb
->hook
.next
, struct bm_block
, hook
);
587 } while (&bb
->hook
!= &bm
->blocks
);
589 memory_bm_position_reset(bm
);
590 return BM_END_OF_MAP
;
593 bm
->cur
.bit
= bit
+ 1;
594 return bb
->start_pfn
+ bit
;
598 * This structure represents a range of page frames the contents of which
599 * should not be saved during the suspend.
602 struct nosave_region
{
603 struct list_head list
;
604 unsigned long start_pfn
;
605 unsigned long end_pfn
;
608 static LIST_HEAD(nosave_regions
);
611 * register_nosave_region - register a range of page frames the contents
612 * of which should not be saved during the suspend (to be used in the early
613 * initialization code)
617 __register_nosave_region(unsigned long start_pfn
, unsigned long end_pfn
,
620 struct nosave_region
*region
;
622 if (start_pfn
>= end_pfn
)
625 if (!list_empty(&nosave_regions
)) {
626 /* Try to extend the previous region (they should be sorted) */
627 region
= list_entry(nosave_regions
.prev
,
628 struct nosave_region
, list
);
629 if (region
->end_pfn
== start_pfn
) {
630 region
->end_pfn
= end_pfn
;
635 /* during init, this shouldn't fail */
636 region
= kmalloc(sizeof(struct nosave_region
), GFP_KERNEL
);
639 /* This allocation cannot fail */
640 region
= alloc_bootmem(sizeof(struct nosave_region
));
641 region
->start_pfn
= start_pfn
;
642 region
->end_pfn
= end_pfn
;
643 list_add_tail(®ion
->list
, &nosave_regions
);
645 printk(KERN_INFO
"PM: Registered nosave memory: [mem %#010llx-%#010llx]\n",
646 (unsigned long long) start_pfn
<< PAGE_SHIFT
,
647 ((unsigned long long) end_pfn
<< PAGE_SHIFT
) - 1);
651 * Set bits in this map correspond to the page frames the contents of which
652 * should not be saved during the suspend.
654 static struct memory_bitmap
*forbidden_pages_map
;
656 /* Set bits in this map correspond to free page frames. */
657 static struct memory_bitmap
*free_pages_map
;
660 * Each page frame allocated for creating the image is marked by setting the
661 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
664 void swsusp_set_page_free(struct page
*page
)
667 memory_bm_set_bit(free_pages_map
, page_to_pfn(page
));
670 static int swsusp_page_is_free(struct page
*page
)
672 return free_pages_map
?
673 memory_bm_test_bit(free_pages_map
, page_to_pfn(page
)) : 0;
676 void swsusp_unset_page_free(struct page
*page
)
679 memory_bm_clear_bit(free_pages_map
, page_to_pfn(page
));
682 static void swsusp_set_page_forbidden(struct page
*page
)
684 if (forbidden_pages_map
)
685 memory_bm_set_bit(forbidden_pages_map
, page_to_pfn(page
));
688 int swsusp_page_is_forbidden(struct page
*page
)
690 return forbidden_pages_map
?
691 memory_bm_test_bit(forbidden_pages_map
, page_to_pfn(page
)) : 0;
694 static void swsusp_unset_page_forbidden(struct page
*page
)
696 if (forbidden_pages_map
)
697 memory_bm_clear_bit(forbidden_pages_map
, page_to_pfn(page
));
701 * mark_nosave_pages - set bits corresponding to the page frames the
702 * contents of which should not be saved in a given bitmap.
705 static void mark_nosave_pages(struct memory_bitmap
*bm
)
707 struct nosave_region
*region
;
709 if (list_empty(&nosave_regions
))
712 list_for_each_entry(region
, &nosave_regions
, list
) {
715 pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n",
716 (unsigned long long) region
->start_pfn
<< PAGE_SHIFT
,
717 ((unsigned long long) region
->end_pfn
<< PAGE_SHIFT
)
720 for (pfn
= region
->start_pfn
; pfn
< region
->end_pfn
; pfn
++)
721 if (pfn_valid(pfn
)) {
723 * It is safe to ignore the result of
724 * mem_bm_set_bit_check() here, since we won't
725 * touch the PFNs for which the error is
728 mem_bm_set_bit_check(bm
, pfn
);
733 static bool is_nosave_page(unsigned long pfn
)
735 struct nosave_region
*region
;
737 list_for_each_entry(region
, &nosave_regions
, list
) {
738 if (pfn
>= region
->start_pfn
&& pfn
< region
->end_pfn
) {
739 pr_err("PM: %#010llx in e820 nosave region: "
740 "[mem %#010llx-%#010llx]\n",
741 (unsigned long long) pfn
<< PAGE_SHIFT
,
742 (unsigned long long) region
->start_pfn
<< PAGE_SHIFT
,
743 ((unsigned long long) region
->end_pfn
<< PAGE_SHIFT
)
753 * create_basic_memory_bitmaps - create bitmaps needed for marking page
754 * frames that should not be saved and free page frames. The pointers
755 * forbidden_pages_map and free_pages_map are only modified if everything
756 * goes well, because we don't want the bits to be used before both bitmaps
760 int create_basic_memory_bitmaps(void)
762 struct memory_bitmap
*bm1
, *bm2
;
765 if (forbidden_pages_map
&& free_pages_map
)
768 BUG_ON(forbidden_pages_map
|| free_pages_map
);
770 bm1
= kzalloc(sizeof(struct memory_bitmap
), GFP_KERNEL
);
774 error
= memory_bm_create(bm1
, GFP_KERNEL
, PG_ANY
);
776 goto Free_first_object
;
778 bm2
= kzalloc(sizeof(struct memory_bitmap
), GFP_KERNEL
);
780 goto Free_first_bitmap
;
782 error
= memory_bm_create(bm2
, GFP_KERNEL
, PG_ANY
);
784 goto Free_second_object
;
786 forbidden_pages_map
= bm1
;
787 free_pages_map
= bm2
;
788 mark_nosave_pages(forbidden_pages_map
);
790 pr_debug("PM: Basic memory bitmaps created\n");
797 memory_bm_free(bm1
, PG_UNSAFE_CLEAR
);
804 * free_basic_memory_bitmaps - free memory bitmaps allocated by
805 * create_basic_memory_bitmaps(). The auxiliary pointers are necessary
806 * so that the bitmaps themselves are not referred to while they are being
810 void free_basic_memory_bitmaps(void)
812 struct memory_bitmap
*bm1
, *bm2
;
814 if (WARN_ON(!(forbidden_pages_map
&& free_pages_map
)))
817 bm1
= forbidden_pages_map
;
818 bm2
= free_pages_map
;
819 forbidden_pages_map
= NULL
;
820 free_pages_map
= NULL
;
821 memory_bm_free(bm1
, PG_UNSAFE_CLEAR
);
823 memory_bm_free(bm2
, PG_UNSAFE_CLEAR
);
826 pr_debug("PM: Basic memory bitmaps freed\n");
830 * snapshot_additional_pages - estimate the number of additional pages
831 * be needed for setting up the suspend image data structures for given
832 * zone (usually the returned value is greater than the exact number)
835 unsigned int snapshot_additional_pages(struct zone
*zone
)
839 res
= DIV_ROUND_UP(zone
->spanned_pages
, BM_BITS_PER_BLOCK
);
840 res
+= DIV_ROUND_UP(res
* sizeof(struct bm_block
),
841 LINKED_PAGE_DATA_SIZE
);
845 #ifdef CONFIG_HIGHMEM
847 * count_free_highmem_pages - compute the total number of free highmem
848 * pages, system-wide.
851 static unsigned int count_free_highmem_pages(void)
854 unsigned int cnt
= 0;
856 for_each_populated_zone(zone
)
857 if (is_highmem(zone
))
858 cnt
+= zone_page_state(zone
, NR_FREE_PAGES
);
864 * saveable_highmem_page - Determine whether a highmem page should be
865 * included in the suspend image.
867 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
868 * and it isn't a part of a free chunk of pages.
870 static struct page
*saveable_highmem_page(struct zone
*zone
, unsigned long pfn
)
877 page
= pfn_to_page(pfn
);
878 if (page_zone(page
) != zone
)
881 BUG_ON(!PageHighMem(page
));
883 if (swsusp_page_is_forbidden(page
) || swsusp_page_is_free(page
) ||
887 if (page_is_guard(page
))
894 * count_highmem_pages - compute the total number of saveable highmem
898 static unsigned int count_highmem_pages(void)
903 for_each_populated_zone(zone
) {
904 unsigned long pfn
, max_zone_pfn
;
906 if (!is_highmem(zone
))
909 mark_free_pages(zone
);
910 max_zone_pfn
= zone_end_pfn(zone
);
911 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
912 if (saveable_highmem_page(zone
, pfn
))
918 static inline void *saveable_highmem_page(struct zone
*z
, unsigned long p
)
922 #endif /* CONFIG_HIGHMEM */
925 * saveable_page - Determine whether a non-highmem page should be included
926 * in the suspend image.
928 * We should save the page if it isn't Nosave, and is not in the range
929 * of pages statically defined as 'unsaveable', and it isn't a part of
930 * a free chunk of pages.
932 static struct page
*saveable_page(struct zone
*zone
, unsigned long pfn
)
939 page
= pfn_to_page(pfn
);
940 if (page_zone(page
) != zone
)
943 BUG_ON(PageHighMem(page
));
945 if (swsusp_page_is_forbidden(page
) || swsusp_page_is_free(page
))
948 if (PageReserved(page
)
949 && (!kernel_page_present(page
) || pfn_is_nosave(pfn
)))
952 if (page_is_guard(page
))
959 * count_data_pages - compute the total number of saveable non-highmem
963 static unsigned int count_data_pages(void)
966 unsigned long pfn
, max_zone_pfn
;
969 for_each_populated_zone(zone
) {
970 if (is_highmem(zone
))
973 mark_free_pages(zone
);
974 max_zone_pfn
= zone_end_pfn(zone
);
975 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
976 if (saveable_page(zone
, pfn
))
982 /* This is needed, because copy_page and memcpy are not usable for copying
985 static inline void do_copy_page(long *dst
, long *src
)
989 for (n
= PAGE_SIZE
/ sizeof(long); n
; n
--)
995 * safe_copy_page - check if the page we are going to copy is marked as
996 * present in the kernel page tables (this always is the case if
997 * CONFIG_DEBUG_PAGEALLOC is not set and in that case
998 * kernel_page_present() always returns 'true').
1000 static void safe_copy_page(void *dst
, struct page
*s_page
)
1002 if (kernel_page_present(s_page
)) {
1003 do_copy_page(dst
, page_address(s_page
));
1005 kernel_map_pages(s_page
, 1, 1);
1006 do_copy_page(dst
, page_address(s_page
));
1007 kernel_map_pages(s_page
, 1, 0);
1012 #ifdef CONFIG_HIGHMEM
1013 static inline struct page
*
1014 page_is_saveable(struct zone
*zone
, unsigned long pfn
)
1016 return is_highmem(zone
) ?
1017 saveable_highmem_page(zone
, pfn
) : saveable_page(zone
, pfn
);
1020 static void copy_data_page(unsigned long dst_pfn
, unsigned long src_pfn
)
1022 struct page
*s_page
, *d_page
;
1025 s_page
= pfn_to_page(src_pfn
);
1026 d_page
= pfn_to_page(dst_pfn
);
1027 if (PageHighMem(s_page
)) {
1028 src
= kmap_atomic(s_page
);
1029 dst
= kmap_atomic(d_page
);
1030 do_copy_page(dst
, src
);
1034 if (PageHighMem(d_page
)) {
1035 /* Page pointed to by src may contain some kernel
1036 * data modified by kmap_atomic()
1038 safe_copy_page(buffer
, s_page
);
1039 dst
= kmap_atomic(d_page
);
1040 copy_page(dst
, buffer
);
1043 safe_copy_page(page_address(d_page
), s_page
);
1048 #define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1050 static inline void copy_data_page(unsigned long dst_pfn
, unsigned long src_pfn
)
1052 safe_copy_page(page_address(pfn_to_page(dst_pfn
)),
1053 pfn_to_page(src_pfn
));
1055 #endif /* CONFIG_HIGHMEM */
1058 copy_data_pages(struct memory_bitmap
*copy_bm
, struct memory_bitmap
*orig_bm
)
1063 for_each_populated_zone(zone
) {
1064 unsigned long max_zone_pfn
;
1066 mark_free_pages(zone
);
1067 max_zone_pfn
= zone_end_pfn(zone
);
1068 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1069 if (page_is_saveable(zone
, pfn
))
1070 memory_bm_set_bit(orig_bm
, pfn
);
1072 memory_bm_position_reset(orig_bm
);
1073 memory_bm_position_reset(copy_bm
);
1075 pfn
= memory_bm_next_pfn(orig_bm
);
1076 if (unlikely(pfn
== BM_END_OF_MAP
))
1078 copy_data_page(memory_bm_next_pfn(copy_bm
), pfn
);
1082 /* Total number of image pages */
1083 static unsigned int nr_copy_pages
;
1084 /* Number of pages needed for saving the original pfns of the image pages */
1085 static unsigned int nr_meta_pages
;
1087 * Numbers of normal and highmem page frames allocated for hibernation image
1088 * before suspending devices.
1090 unsigned int alloc_normal
, alloc_highmem
;
1092 * Memory bitmap used for marking saveable pages (during hibernation) or
1093 * hibernation image pages (during restore)
1095 static struct memory_bitmap orig_bm
;
1097 * Memory bitmap used during hibernation for marking allocated page frames that
1098 * will contain copies of saveable pages. During restore it is initially used
1099 * for marking hibernation image pages, but then the set bits from it are
1100 * duplicated in @orig_bm and it is released. On highmem systems it is next
1101 * used for marking "safe" highmem pages, but it has to be reinitialized for
1104 static struct memory_bitmap copy_bm
;
1107 * swsusp_free - free pages allocated for the suspend.
1109 * Suspend pages are alocated before the atomic copy is made, so we
1110 * need to release them after the resume.
1113 void swsusp_free(void)
1116 unsigned long pfn
, max_zone_pfn
;
1118 for_each_populated_zone(zone
) {
1119 max_zone_pfn
= zone_end_pfn(zone
);
1120 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1121 if (pfn_valid(pfn
)) {
1122 struct page
*page
= pfn_to_page(pfn
);
1124 if (swsusp_page_is_forbidden(page
) &&
1125 swsusp_page_is_free(page
)) {
1126 swsusp_unset_page_forbidden(page
);
1127 swsusp_unset_page_free(page
);
1134 restore_pblist
= NULL
;
1140 /* Helper functions used for the shrinking of memory. */
1142 #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1145 * preallocate_image_pages - Allocate a number of pages for hibernation image
1146 * @nr_pages: Number of page frames to allocate.
1147 * @mask: GFP flags to use for the allocation.
1149 * Return value: Number of page frames actually allocated
1151 static unsigned long preallocate_image_pages(unsigned long nr_pages
, gfp_t mask
)
1153 unsigned long nr_alloc
= 0;
1155 while (nr_pages
> 0) {
1158 page
= alloc_image_page(mask
);
1161 memory_bm_set_bit(©_bm
, page_to_pfn(page
));
1162 if (PageHighMem(page
))
1173 static unsigned long preallocate_image_memory(unsigned long nr_pages
,
1174 unsigned long avail_normal
)
1176 unsigned long alloc
;
1178 if (avail_normal
<= alloc_normal
)
1181 alloc
= avail_normal
- alloc_normal
;
1182 if (nr_pages
< alloc
)
1185 return preallocate_image_pages(alloc
, GFP_IMAGE
);
1188 #ifdef CONFIG_HIGHMEM
1189 static unsigned long preallocate_image_highmem(unsigned long nr_pages
)
1191 return preallocate_image_pages(nr_pages
, GFP_IMAGE
| __GFP_HIGHMEM
);
1195 * __fraction - Compute (an approximation of) x * (multiplier / base)
1197 static unsigned long __fraction(u64 x
, u64 multiplier
, u64 base
)
1201 return (unsigned long)x
;
1204 static unsigned long preallocate_highmem_fraction(unsigned long nr_pages
,
1205 unsigned long highmem
,
1206 unsigned long total
)
1208 unsigned long alloc
= __fraction(nr_pages
, highmem
, total
);
1210 return preallocate_image_pages(alloc
, GFP_IMAGE
| __GFP_HIGHMEM
);
1212 #else /* CONFIG_HIGHMEM */
1213 static inline unsigned long preallocate_image_highmem(unsigned long nr_pages
)
1218 static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages
,
1219 unsigned long highmem
,
1220 unsigned long total
)
1224 #endif /* CONFIG_HIGHMEM */
1227 * free_unnecessary_pages - Release preallocated pages not needed for the image
1229 static void free_unnecessary_pages(void)
1231 unsigned long save
, to_free_normal
, to_free_highmem
;
1233 save
= count_data_pages();
1234 if (alloc_normal
>= save
) {
1235 to_free_normal
= alloc_normal
- save
;
1239 save
-= alloc_normal
;
1241 save
+= count_highmem_pages();
1242 if (alloc_highmem
>= save
) {
1243 to_free_highmem
= alloc_highmem
- save
;
1245 to_free_highmem
= 0;
1246 save
-= alloc_highmem
;
1247 if (to_free_normal
> save
)
1248 to_free_normal
-= save
;
1253 memory_bm_position_reset(©_bm
);
1255 while (to_free_normal
> 0 || to_free_highmem
> 0) {
1256 unsigned long pfn
= memory_bm_next_pfn(©_bm
);
1257 struct page
*page
= pfn_to_page(pfn
);
1259 if (PageHighMem(page
)) {
1260 if (!to_free_highmem
)
1265 if (!to_free_normal
)
1270 memory_bm_clear_bit(©_bm
, pfn
);
1271 swsusp_unset_page_forbidden(page
);
1272 swsusp_unset_page_free(page
);
1278 * minimum_image_size - Estimate the minimum acceptable size of an image
1279 * @saveable: Number of saveable pages in the system.
1281 * We want to avoid attempting to free too much memory too hard, so estimate the
1282 * minimum acceptable size of a hibernation image to use as the lower limit for
1283 * preallocating memory.
1285 * We assume that the minimum image size should be proportional to
1287 * [number of saveable pages] - [number of pages that can be freed in theory]
1289 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1290 * and (3) inactive anonymouns pages, (4) active and (5) inactive file pages,
1291 * minus mapped file pages.
1293 static unsigned long minimum_image_size(unsigned long saveable
)
1297 size
= global_page_state(NR_SLAB_RECLAIMABLE
)
1298 + global_page_state(NR_ACTIVE_ANON
)
1299 + global_page_state(NR_INACTIVE_ANON
)
1300 + global_page_state(NR_ACTIVE_FILE
)
1301 + global_page_state(NR_INACTIVE_FILE
)
1302 - global_page_state(NR_FILE_MAPPED
);
1304 return saveable
<= size
? 0 : saveable
- size
;
1308 * hibernate_preallocate_memory - Preallocate memory for hibernation image
1310 * To create a hibernation image it is necessary to make a copy of every page
1311 * frame in use. We also need a number of page frames to be free during
1312 * hibernation for allocations made while saving the image and for device
1313 * drivers, in case they need to allocate memory from their hibernation
1314 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1315 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1316 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1317 * total number of available page frames and allocate at least
1319 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1320 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1322 * of them, which corresponds to the maximum size of a hibernation image.
1324 * If image_size is set below the number following from the above formula,
1325 * the preallocation of memory is continued until the total number of saveable
1326 * pages in the system is below the requested image size or the minimum
1327 * acceptable image size returned by minimum_image_size(), whichever is greater.
1329 int hibernate_preallocate_memory(void)
1332 unsigned long saveable
, size
, max_size
, count
, highmem
, pages
= 0;
1333 unsigned long alloc
, save_highmem
, pages_highmem
, avail_normal
;
1334 struct timeval start
, stop
;
1337 printk(KERN_INFO
"PM: Preallocating image memory... ");
1338 do_gettimeofday(&start
);
1340 error
= memory_bm_create(&orig_bm
, GFP_IMAGE
, PG_ANY
);
1344 error
= memory_bm_create(©_bm
, GFP_IMAGE
, PG_ANY
);
1351 /* Count the number of saveable data pages. */
1352 save_highmem
= count_highmem_pages();
1353 saveable
= count_data_pages();
1356 * Compute the total number of page frames we can use (count) and the
1357 * number of pages needed for image metadata (size).
1360 saveable
+= save_highmem
;
1361 highmem
= save_highmem
;
1363 for_each_populated_zone(zone
) {
1364 size
+= snapshot_additional_pages(zone
);
1365 if (is_highmem(zone
))
1366 highmem
+= zone_page_state(zone
, NR_FREE_PAGES
);
1368 count
+= zone_page_state(zone
, NR_FREE_PAGES
);
1370 avail_normal
= count
;
1372 count
-= totalreserve_pages
;
1374 /* Add number of pages required for page keys (s390 only). */
1375 size
+= page_key_additional_pages(saveable
);
1377 /* Compute the maximum number of saveable pages to leave in memory. */
1378 max_size
= (count
- (size
+ PAGES_FOR_IO
)) / 2
1379 - 2 * DIV_ROUND_UP(reserved_size
, PAGE_SIZE
);
1380 /* Compute the desired number of image pages specified by image_size. */
1381 size
= DIV_ROUND_UP(image_size
, PAGE_SIZE
);
1382 if (size
> max_size
)
1385 * If the desired number of image pages is at least as large as the
1386 * current number of saveable pages in memory, allocate page frames for
1387 * the image and we're done.
1389 if (size
>= saveable
) {
1390 pages
= preallocate_image_highmem(save_highmem
);
1391 pages
+= preallocate_image_memory(saveable
- pages
, avail_normal
);
1395 /* Estimate the minimum size of the image. */
1396 pages
= minimum_image_size(saveable
);
1398 * To avoid excessive pressure on the normal zone, leave room in it to
1399 * accommodate an image of the minimum size (unless it's already too
1400 * small, in which case don't preallocate pages from it at all).
1402 if (avail_normal
> pages
)
1403 avail_normal
-= pages
;
1407 size
= min_t(unsigned long, pages
, max_size
);
1410 * Let the memory management subsystem know that we're going to need a
1411 * large number of page frames to allocate and make it free some memory.
1412 * NOTE: If this is not done, performance will be hurt badly in some
1415 shrink_all_memory(saveable
- size
);
1418 * The number of saveable pages in memory was too high, so apply some
1419 * pressure to decrease it. First, make room for the largest possible
1420 * image and fail if that doesn't work. Next, try to decrease the size
1421 * of the image as much as indicated by 'size' using allocations from
1422 * highmem and non-highmem zones separately.
1424 pages_highmem
= preallocate_image_highmem(highmem
/ 2);
1425 alloc
= count
- max_size
;
1426 if (alloc
> pages_highmem
)
1427 alloc
-= pages_highmem
;
1430 pages
= preallocate_image_memory(alloc
, avail_normal
);
1431 if (pages
< alloc
) {
1432 /* We have exhausted non-highmem pages, try highmem. */
1434 pages
+= pages_highmem
;
1435 pages_highmem
= preallocate_image_highmem(alloc
);
1436 if (pages_highmem
< alloc
)
1438 pages
+= pages_highmem
;
1440 * size is the desired number of saveable pages to leave in
1441 * memory, so try to preallocate (all memory - size) pages.
1443 alloc
= (count
- pages
) - size
;
1444 pages
+= preallocate_image_highmem(alloc
);
1447 * There are approximately max_size saveable pages at this point
1448 * and we want to reduce this number down to size.
1450 alloc
= max_size
- size
;
1451 size
= preallocate_highmem_fraction(alloc
, highmem
, count
);
1452 pages_highmem
+= size
;
1454 size
= preallocate_image_memory(alloc
, avail_normal
);
1455 pages_highmem
+= preallocate_image_highmem(alloc
- size
);
1456 pages
+= pages_highmem
+ size
;
1460 * We only need as many page frames for the image as there are saveable
1461 * pages in memory, but we have allocated more. Release the excessive
1464 free_unnecessary_pages();
1467 do_gettimeofday(&stop
);
1468 printk(KERN_CONT
"done (allocated %lu pages)\n", pages
);
1469 swsusp_show_speed(&start
, &stop
, pages
, "Allocated");
1474 printk(KERN_CONT
"\n");
1479 #ifdef CONFIG_HIGHMEM
1481 * count_pages_for_highmem - compute the number of non-highmem pages
1482 * that will be necessary for creating copies of highmem pages.
1485 static unsigned int count_pages_for_highmem(unsigned int nr_highmem
)
1487 unsigned int free_highmem
= count_free_highmem_pages() + alloc_highmem
;
1489 if (free_highmem
>= nr_highmem
)
1492 nr_highmem
-= free_highmem
;
1498 count_pages_for_highmem(unsigned int nr_highmem
) { return 0; }
1499 #endif /* CONFIG_HIGHMEM */
1502 * enough_free_mem - Make sure we have enough free memory for the
1506 static int enough_free_mem(unsigned int nr_pages
, unsigned int nr_highmem
)
1509 unsigned int free
= alloc_normal
;
1511 for_each_populated_zone(zone
)
1512 if (!is_highmem(zone
))
1513 free
+= zone_page_state(zone
, NR_FREE_PAGES
);
1515 nr_pages
+= count_pages_for_highmem(nr_highmem
);
1516 pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n",
1517 nr_pages
, PAGES_FOR_IO
, free
);
1519 return free
> nr_pages
+ PAGES_FOR_IO
;
1522 #ifdef CONFIG_HIGHMEM
1524 * get_highmem_buffer - if there are some highmem pages in the suspend
1525 * image, we may need the buffer to copy them and/or load their data.
1528 static inline int get_highmem_buffer(int safe_needed
)
1530 buffer
= get_image_page(GFP_ATOMIC
| __GFP_COLD
, safe_needed
);
1531 return buffer
? 0 : -ENOMEM
;
1535 * alloc_highmem_image_pages - allocate some highmem pages for the image.
1536 * Try to allocate as many pages as needed, but if the number of free
1537 * highmem pages is lesser than that, allocate them all.
1540 static inline unsigned int
1541 alloc_highmem_pages(struct memory_bitmap
*bm
, unsigned int nr_highmem
)
1543 unsigned int to_alloc
= count_free_highmem_pages();
1545 if (to_alloc
> nr_highmem
)
1546 to_alloc
= nr_highmem
;
1548 nr_highmem
-= to_alloc
;
1549 while (to_alloc
-- > 0) {
1552 page
= alloc_image_page(__GFP_HIGHMEM
);
1553 memory_bm_set_bit(bm
, page_to_pfn(page
));
1558 static inline int get_highmem_buffer(int safe_needed
) { return 0; }
1560 static inline unsigned int
1561 alloc_highmem_pages(struct memory_bitmap
*bm
, unsigned int n
) { return 0; }
1562 #endif /* CONFIG_HIGHMEM */
1565 * swsusp_alloc - allocate memory for the suspend image
1567 * We first try to allocate as many highmem pages as there are
1568 * saveable highmem pages in the system. If that fails, we allocate
1569 * non-highmem pages for the copies of the remaining highmem ones.
1571 * In this approach it is likely that the copies of highmem pages will
1572 * also be located in the high memory, because of the way in which
1573 * copy_data_pages() works.
1577 swsusp_alloc(struct memory_bitmap
*orig_bm
, struct memory_bitmap
*copy_bm
,
1578 unsigned int nr_pages
, unsigned int nr_highmem
)
1580 if (nr_highmem
> 0) {
1581 if (get_highmem_buffer(PG_ANY
))
1583 if (nr_highmem
> alloc_highmem
) {
1584 nr_highmem
-= alloc_highmem
;
1585 nr_pages
+= alloc_highmem_pages(copy_bm
, nr_highmem
);
1588 if (nr_pages
> alloc_normal
) {
1589 nr_pages
-= alloc_normal
;
1590 while (nr_pages
-- > 0) {
1593 page
= alloc_image_page(GFP_ATOMIC
| __GFP_COLD
);
1596 memory_bm_set_bit(copy_bm
, page_to_pfn(page
));
1607 asmlinkage
int swsusp_save(void)
1609 unsigned int nr_pages
, nr_highmem
;
1611 printk(KERN_INFO
"PM: Creating hibernation image:\n");
1613 drain_local_pages(NULL
);
1614 nr_pages
= count_data_pages();
1615 nr_highmem
= count_highmem_pages();
1616 printk(KERN_INFO
"PM: Need to copy %u pages\n", nr_pages
+ nr_highmem
);
1618 if (!enough_free_mem(nr_pages
, nr_highmem
)) {
1619 printk(KERN_ERR
"PM: Not enough free memory\n");
1623 if (swsusp_alloc(&orig_bm
, ©_bm
, nr_pages
, nr_highmem
)) {
1624 printk(KERN_ERR
"PM: Memory allocation failed\n");
1628 /* During allocating of suspend pagedir, new cold pages may appear.
1631 drain_local_pages(NULL
);
1632 copy_data_pages(©_bm
, &orig_bm
);
1635 * End of critical section. From now on, we can write to memory,
1636 * but we should not touch disk. This specially means we must _not_
1637 * touch swap space! Except we must write out our image of course.
1640 nr_pages
+= nr_highmem
;
1641 nr_copy_pages
= nr_pages
;
1642 nr_meta_pages
= DIV_ROUND_UP(nr_pages
* sizeof(long), PAGE_SIZE
);
1644 printk(KERN_INFO
"PM: Hibernation image created (%d pages copied)\n",
1650 #ifndef CONFIG_ARCH_HIBERNATION_HEADER
1651 static int init_header_complete(struct swsusp_info
*info
)
1653 memcpy(&info
->uts
, init_utsname(), sizeof(struct new_utsname
));
1654 info
->version_code
= LINUX_VERSION_CODE
;
1658 static char *check_image_kernel(struct swsusp_info
*info
)
1660 if (info
->version_code
!= LINUX_VERSION_CODE
)
1661 return "kernel version";
1662 if (strcmp(info
->uts
.sysname
,init_utsname()->sysname
))
1663 return "system type";
1664 if (strcmp(info
->uts
.release
,init_utsname()->release
))
1665 return "kernel release";
1666 if (strcmp(info
->uts
.version
,init_utsname()->version
))
1668 if (strcmp(info
->uts
.machine
,init_utsname()->machine
))
1672 #endif /* CONFIG_ARCH_HIBERNATION_HEADER */
1674 unsigned long snapshot_get_image_size(void)
1676 return nr_copy_pages
+ nr_meta_pages
+ 1;
1679 static int init_header(struct swsusp_info
*info
)
1681 memset(info
, 0, sizeof(struct swsusp_info
));
1682 info
->num_physpages
= get_num_physpages();
1683 info
->image_pages
= nr_copy_pages
;
1684 info
->pages
= snapshot_get_image_size();
1685 info
->size
= info
->pages
;
1686 info
->size
<<= PAGE_SHIFT
;
1687 return init_header_complete(info
);
1691 * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm
1692 * are stored in the array @buf[] (1 page at a time)
1696 pack_pfns(unsigned long *buf
, struct memory_bitmap
*bm
)
1700 for (j
= 0; j
< PAGE_SIZE
/ sizeof(long); j
++) {
1701 buf
[j
] = memory_bm_next_pfn(bm
);
1702 if (unlikely(buf
[j
] == BM_END_OF_MAP
))
1704 /* Save page key for data page (s390 only). */
1705 page_key_read(buf
+ j
);
1710 * snapshot_read_next - used for reading the system memory snapshot.
1712 * On the first call to it @handle should point to a zeroed
1713 * snapshot_handle structure. The structure gets updated and a pointer
1714 * to it should be passed to this function every next time.
1716 * On success the function returns a positive number. Then, the caller
1717 * is allowed to read up to the returned number of bytes from the memory
1718 * location computed by the data_of() macro.
1720 * The function returns 0 to indicate the end of data stream condition,
1721 * and a negative number is returned on error. In such cases the
1722 * structure pointed to by @handle is not updated and should not be used
1726 int snapshot_read_next(struct snapshot_handle
*handle
)
1728 if (handle
->cur
> nr_meta_pages
+ nr_copy_pages
)
1732 /* This makes the buffer be freed by swsusp_free() */
1733 buffer
= get_image_page(GFP_ATOMIC
, PG_ANY
);
1740 error
= init_header((struct swsusp_info
*)buffer
);
1743 handle
->buffer
= buffer
;
1744 memory_bm_position_reset(&orig_bm
);
1745 memory_bm_position_reset(©_bm
);
1746 } else if (handle
->cur
<= nr_meta_pages
) {
1748 pack_pfns(buffer
, &orig_bm
);
1752 page
= pfn_to_page(memory_bm_next_pfn(©_bm
));
1753 if (PageHighMem(page
)) {
1754 /* Highmem pages are copied to the buffer,
1755 * because we can't return with a kmapped
1756 * highmem page (we may not be called again).
1760 kaddr
= kmap_atomic(page
);
1761 copy_page(buffer
, kaddr
);
1762 kunmap_atomic(kaddr
);
1763 handle
->buffer
= buffer
;
1765 handle
->buffer
= page_address(page
);
1773 * mark_unsafe_pages - mark the pages that cannot be used for storing
1774 * the image during resume, because they conflict with the pages that
1775 * had been used before suspend
1778 static int mark_unsafe_pages(struct memory_bitmap
*bm
)
1781 unsigned long pfn
, max_zone_pfn
;
1783 /* Clear page flags */
1784 for_each_populated_zone(zone
) {
1785 max_zone_pfn
= zone_end_pfn(zone
);
1786 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1788 swsusp_unset_page_free(pfn_to_page(pfn
));
1791 /* Mark pages that correspond to the "original" pfns as "unsafe" */
1792 memory_bm_position_reset(bm
);
1794 pfn
= memory_bm_next_pfn(bm
);
1795 if (likely(pfn
!= BM_END_OF_MAP
)) {
1796 if (likely(pfn_valid(pfn
)) && !is_nosave_page(pfn
))
1797 swsusp_set_page_free(pfn_to_page(pfn
));
1801 } while (pfn
!= BM_END_OF_MAP
);
1803 allocated_unsafe_pages
= 0;
1809 duplicate_memory_bitmap(struct memory_bitmap
*dst
, struct memory_bitmap
*src
)
1813 memory_bm_position_reset(src
);
1814 pfn
= memory_bm_next_pfn(src
);
1815 while (pfn
!= BM_END_OF_MAP
) {
1816 memory_bm_set_bit(dst
, pfn
);
1817 pfn
= memory_bm_next_pfn(src
);
1821 static int check_header(struct swsusp_info
*info
)
1825 reason
= check_image_kernel(info
);
1826 if (!reason
&& info
->num_physpages
!= get_num_physpages())
1827 reason
= "memory size";
1829 printk(KERN_ERR
"PM: Image mismatch: %s\n", reason
);
1836 * load header - check the image header and copy data from it
1840 load_header(struct swsusp_info
*info
)
1844 restore_pblist
= NULL
;
1845 error
= check_header(info
);
1847 nr_copy_pages
= info
->image_pages
;
1848 nr_meta_pages
= info
->pages
- info
->image_pages
- 1;
1854 * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set
1855 * the corresponding bit in the memory bitmap @bm
1857 static int unpack_orig_pfns(unsigned long *buf
, struct memory_bitmap
*bm
)
1861 for (j
= 0; j
< PAGE_SIZE
/ sizeof(long); j
++) {
1862 if (unlikely(buf
[j
] == BM_END_OF_MAP
))
1865 /* Extract and buffer page key for data page (s390 only). */
1866 page_key_memorize(buf
+ j
);
1868 if (memory_bm_pfn_present(bm
, buf
[j
]))
1869 memory_bm_set_bit(bm
, buf
[j
]);
1877 /* List of "safe" pages that may be used to store data loaded from the suspend
1880 static struct linked_page
*safe_pages_list
;
1882 #ifdef CONFIG_HIGHMEM
1883 /* struct highmem_pbe is used for creating the list of highmem pages that
1884 * should be restored atomically during the resume from disk, because the page
1885 * frames they have occupied before the suspend are in use.
1887 struct highmem_pbe
{
1888 struct page
*copy_page
; /* data is here now */
1889 struct page
*orig_page
; /* data was here before the suspend */
1890 struct highmem_pbe
*next
;
1893 /* List of highmem PBEs needed for restoring the highmem pages that were
1894 * allocated before the suspend and included in the suspend image, but have
1895 * also been allocated by the "resume" kernel, so their contents cannot be
1896 * written directly to their "original" page frames.
1898 static struct highmem_pbe
*highmem_pblist
;
1901 * count_highmem_image_pages - compute the number of highmem pages in the
1902 * suspend image. The bits in the memory bitmap @bm that correspond to the
1903 * image pages are assumed to be set.
1906 static unsigned int count_highmem_image_pages(struct memory_bitmap
*bm
)
1909 unsigned int cnt
= 0;
1911 memory_bm_position_reset(bm
);
1912 pfn
= memory_bm_next_pfn(bm
);
1913 while (pfn
!= BM_END_OF_MAP
) {
1914 if (PageHighMem(pfn_to_page(pfn
)))
1917 pfn
= memory_bm_next_pfn(bm
);
1923 * prepare_highmem_image - try to allocate as many highmem pages as
1924 * there are highmem image pages (@nr_highmem_p points to the variable
1925 * containing the number of highmem image pages). The pages that are
1926 * "safe" (ie. will not be overwritten when the suspend image is
1927 * restored) have the corresponding bits set in @bm (it must be
1930 * NOTE: This function should not be called if there are no highmem
1934 static unsigned int safe_highmem_pages
;
1936 static struct memory_bitmap
*safe_highmem_bm
;
1939 prepare_highmem_image(struct memory_bitmap
*bm
, unsigned int *nr_highmem_p
)
1941 unsigned int to_alloc
;
1943 if (memory_bm_create(bm
, GFP_ATOMIC
, PG_SAFE
))
1946 if (get_highmem_buffer(PG_SAFE
))
1949 to_alloc
= count_free_highmem_pages();
1950 if (to_alloc
> *nr_highmem_p
)
1951 to_alloc
= *nr_highmem_p
;
1953 *nr_highmem_p
= to_alloc
;
1955 safe_highmem_pages
= 0;
1956 while (to_alloc
-- > 0) {
1959 page
= alloc_page(__GFP_HIGHMEM
);
1960 if (!swsusp_page_is_free(page
)) {
1961 /* The page is "safe", set its bit the bitmap */
1962 memory_bm_set_bit(bm
, page_to_pfn(page
));
1963 safe_highmem_pages
++;
1965 /* Mark the page as allocated */
1966 swsusp_set_page_forbidden(page
);
1967 swsusp_set_page_free(page
);
1969 memory_bm_position_reset(bm
);
1970 safe_highmem_bm
= bm
;
1975 * get_highmem_page_buffer - for given highmem image page find the buffer
1976 * that suspend_write_next() should set for its caller to write to.
1978 * If the page is to be saved to its "original" page frame or a copy of
1979 * the page is to be made in the highmem, @buffer is returned. Otherwise,
1980 * the copy of the page is to be made in normal memory, so the address of
1981 * the copy is returned.
1983 * If @buffer is returned, the caller of suspend_write_next() will write
1984 * the page's contents to @buffer, so they will have to be copied to the
1985 * right location on the next call to suspend_write_next() and it is done
1986 * with the help of copy_last_highmem_page(). For this purpose, if
1987 * @buffer is returned, @last_highmem page is set to the page to which
1988 * the data will have to be copied from @buffer.
1991 static struct page
*last_highmem_page
;
1994 get_highmem_page_buffer(struct page
*page
, struct chain_allocator
*ca
)
1996 struct highmem_pbe
*pbe
;
1999 if (swsusp_page_is_forbidden(page
) && swsusp_page_is_free(page
)) {
2000 /* We have allocated the "original" page frame and we can
2001 * use it directly to store the loaded page.
2003 last_highmem_page
= page
;
2006 /* The "original" page frame has not been allocated and we have to
2007 * use a "safe" page frame to store the loaded page.
2009 pbe
= chain_alloc(ca
, sizeof(struct highmem_pbe
));
2012 return ERR_PTR(-ENOMEM
);
2014 pbe
->orig_page
= page
;
2015 if (safe_highmem_pages
> 0) {
2018 /* Copy of the page will be stored in high memory */
2020 tmp
= pfn_to_page(memory_bm_next_pfn(safe_highmem_bm
));
2021 safe_highmem_pages
--;
2022 last_highmem_page
= tmp
;
2023 pbe
->copy_page
= tmp
;
2025 /* Copy of the page will be stored in normal memory */
2026 kaddr
= safe_pages_list
;
2027 safe_pages_list
= safe_pages_list
->next
;
2028 pbe
->copy_page
= virt_to_page(kaddr
);
2030 pbe
->next
= highmem_pblist
;
2031 highmem_pblist
= pbe
;
2036 * copy_last_highmem_page - copy the contents of a highmem image from
2037 * @buffer, where the caller of snapshot_write_next() has place them,
2038 * to the right location represented by @last_highmem_page .
2041 static void copy_last_highmem_page(void)
2043 if (last_highmem_page
) {
2046 dst
= kmap_atomic(last_highmem_page
);
2047 copy_page(dst
, buffer
);
2049 last_highmem_page
= NULL
;
2053 static inline int last_highmem_page_copied(void)
2055 return !last_highmem_page
;
2058 static inline void free_highmem_data(void)
2060 if (safe_highmem_bm
)
2061 memory_bm_free(safe_highmem_bm
, PG_UNSAFE_CLEAR
);
2064 free_image_page(buffer
, PG_UNSAFE_CLEAR
);
2067 static inline int get_safe_write_buffer(void) { return 0; }
2070 count_highmem_image_pages(struct memory_bitmap
*bm
) { return 0; }
2073 prepare_highmem_image(struct memory_bitmap
*bm
, unsigned int *nr_highmem_p
)
2078 static inline void *
2079 get_highmem_page_buffer(struct page
*page
, struct chain_allocator
*ca
)
2081 return ERR_PTR(-EINVAL
);
2084 static inline void copy_last_highmem_page(void) {}
2085 static inline int last_highmem_page_copied(void) { return 1; }
2086 static inline void free_highmem_data(void) {}
2087 #endif /* CONFIG_HIGHMEM */
2090 * prepare_image - use the memory bitmap @bm to mark the pages that will
2091 * be overwritten in the process of restoring the system memory state
2092 * from the suspend image ("unsafe" pages) and allocate memory for the
2095 * The idea is to allocate a new memory bitmap first and then allocate
2096 * as many pages as needed for the image data, but not to assign these
2097 * pages to specific tasks initially. Instead, we just mark them as
2098 * allocated and create a lists of "safe" pages that will be used
2099 * later. On systems with high memory a list of "safe" highmem pages is
2103 #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2106 prepare_image(struct memory_bitmap
*new_bm
, struct memory_bitmap
*bm
)
2108 unsigned int nr_pages
, nr_highmem
;
2109 struct linked_page
*sp_list
, *lp
;
2112 /* If there is no highmem, the buffer will not be necessary */
2113 free_image_page(buffer
, PG_UNSAFE_CLEAR
);
2116 nr_highmem
= count_highmem_image_pages(bm
);
2117 error
= mark_unsafe_pages(bm
);
2121 error
= memory_bm_create(new_bm
, GFP_ATOMIC
, PG_SAFE
);
2125 duplicate_memory_bitmap(new_bm
, bm
);
2126 memory_bm_free(bm
, PG_UNSAFE_KEEP
);
2127 if (nr_highmem
> 0) {
2128 error
= prepare_highmem_image(bm
, &nr_highmem
);
2132 /* Reserve some safe pages for potential later use.
2134 * NOTE: This way we make sure there will be enough safe pages for the
2135 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2136 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2139 /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */
2140 nr_pages
= nr_copy_pages
- nr_highmem
- allocated_unsafe_pages
;
2141 nr_pages
= DIV_ROUND_UP(nr_pages
, PBES_PER_LINKED_PAGE
);
2142 while (nr_pages
> 0) {
2143 lp
= get_image_page(GFP_ATOMIC
, PG_SAFE
);
2152 /* Preallocate memory for the image */
2153 safe_pages_list
= NULL
;
2154 nr_pages
= nr_copy_pages
- nr_highmem
- allocated_unsafe_pages
;
2155 while (nr_pages
> 0) {
2156 lp
= (struct linked_page
*)get_zeroed_page(GFP_ATOMIC
);
2161 if (!swsusp_page_is_free(virt_to_page(lp
))) {
2162 /* The page is "safe", add it to the list */
2163 lp
->next
= safe_pages_list
;
2164 safe_pages_list
= lp
;
2166 /* Mark the page as allocated */
2167 swsusp_set_page_forbidden(virt_to_page(lp
));
2168 swsusp_set_page_free(virt_to_page(lp
));
2171 /* Free the reserved safe pages so that chain_alloc() can use them */
2174 free_image_page(sp_list
, PG_UNSAFE_CLEAR
);
2185 * get_buffer - compute the address that snapshot_write_next() should
2186 * set for its caller to write to.
2189 static void *get_buffer(struct memory_bitmap
*bm
, struct chain_allocator
*ca
)
2193 unsigned long pfn
= memory_bm_next_pfn(bm
);
2195 if (pfn
== BM_END_OF_MAP
)
2196 return ERR_PTR(-EFAULT
);
2198 page
= pfn_to_page(pfn
);
2199 if (PageHighMem(page
))
2200 return get_highmem_page_buffer(page
, ca
);
2202 if (swsusp_page_is_forbidden(page
) && swsusp_page_is_free(page
))
2203 /* We have allocated the "original" page frame and we can
2204 * use it directly to store the loaded page.
2206 return page_address(page
);
2208 /* The "original" page frame has not been allocated and we have to
2209 * use a "safe" page frame to store the loaded page.
2211 pbe
= chain_alloc(ca
, sizeof(struct pbe
));
2214 return ERR_PTR(-ENOMEM
);
2216 pbe
->orig_address
= page_address(page
);
2217 pbe
->address
= safe_pages_list
;
2218 safe_pages_list
= safe_pages_list
->next
;
2219 pbe
->next
= restore_pblist
;
2220 restore_pblist
= pbe
;
2221 return pbe
->address
;
2225 * snapshot_write_next - used for writing the system memory snapshot.
2227 * On the first call to it @handle should point to a zeroed
2228 * snapshot_handle structure. The structure gets updated and a pointer
2229 * to it should be passed to this function every next time.
2231 * On success the function returns a positive number. Then, the caller
2232 * is allowed to write up to the returned number of bytes to the memory
2233 * location computed by the data_of() macro.
2235 * The function returns 0 to indicate the "end of file" condition,
2236 * and a negative number is returned on error. In such cases the
2237 * structure pointed to by @handle is not updated and should not be used
2241 int snapshot_write_next(struct snapshot_handle
*handle
)
2243 static struct chain_allocator ca
;
2246 /* Check if we have already loaded the entire image */
2247 if (handle
->cur
> 1 && handle
->cur
> nr_meta_pages
+ nr_copy_pages
)
2250 handle
->sync_read
= 1;
2254 /* This makes the buffer be freed by swsusp_free() */
2255 buffer
= get_image_page(GFP_ATOMIC
, PG_ANY
);
2260 handle
->buffer
= buffer
;
2261 } else if (handle
->cur
== 1) {
2262 error
= load_header(buffer
);
2266 error
= memory_bm_create(©_bm
, GFP_ATOMIC
, PG_ANY
);
2270 /* Allocate buffer for page keys. */
2271 error
= page_key_alloc(nr_copy_pages
);
2275 } else if (handle
->cur
<= nr_meta_pages
+ 1) {
2276 error
= unpack_orig_pfns(buffer
, ©_bm
);
2280 if (handle
->cur
== nr_meta_pages
+ 1) {
2281 error
= prepare_image(&orig_bm
, ©_bm
);
2285 chain_init(&ca
, GFP_ATOMIC
, PG_SAFE
);
2286 memory_bm_position_reset(&orig_bm
);
2287 restore_pblist
= NULL
;
2288 handle
->buffer
= get_buffer(&orig_bm
, &ca
);
2289 handle
->sync_read
= 0;
2290 if (IS_ERR(handle
->buffer
))
2291 return PTR_ERR(handle
->buffer
);
2294 copy_last_highmem_page();
2295 /* Restore page key for data page (s390 only). */
2296 page_key_write(handle
->buffer
);
2297 handle
->buffer
= get_buffer(&orig_bm
, &ca
);
2298 if (IS_ERR(handle
->buffer
))
2299 return PTR_ERR(handle
->buffer
);
2300 if (handle
->buffer
!= buffer
)
2301 handle
->sync_read
= 0;
2308 * snapshot_write_finalize - must be called after the last call to
2309 * snapshot_write_next() in case the last page in the image happens
2310 * to be a highmem page and its contents should be stored in the
2311 * highmem. Additionally, it releases the memory that will not be
2315 void snapshot_write_finalize(struct snapshot_handle
*handle
)
2317 copy_last_highmem_page();
2318 /* Restore page key for data page (s390 only). */
2319 page_key_write(handle
->buffer
);
2321 /* Free only if we have loaded the image entirely */
2322 if (handle
->cur
> 1 && handle
->cur
> nr_meta_pages
+ nr_copy_pages
) {
2323 memory_bm_free(&orig_bm
, PG_UNSAFE_CLEAR
);
2324 free_highmem_data();
2328 int snapshot_image_loaded(struct snapshot_handle
*handle
)
2330 return !(!nr_copy_pages
|| !last_highmem_page_copied() ||
2331 handle
->cur
<= nr_meta_pages
+ nr_copy_pages
);
2334 #ifdef CONFIG_HIGHMEM
2335 /* Assumes that @buf is ready and points to a "safe" page */
2337 swap_two_pages_data(struct page
*p1
, struct page
*p2
, void *buf
)
2339 void *kaddr1
, *kaddr2
;
2341 kaddr1
= kmap_atomic(p1
);
2342 kaddr2
= kmap_atomic(p2
);
2343 copy_page(buf
, kaddr1
);
2344 copy_page(kaddr1
, kaddr2
);
2345 copy_page(kaddr2
, buf
);
2346 kunmap_atomic(kaddr2
);
2347 kunmap_atomic(kaddr1
);
2351 * restore_highmem - for each highmem page that was allocated before
2352 * the suspend and included in the suspend image, and also has been
2353 * allocated by the "resume" kernel swap its current (ie. "before
2354 * resume") contents with the previous (ie. "before suspend") one.
2356 * If the resume eventually fails, we can call this function once
2357 * again and restore the "before resume" highmem state.
2360 int restore_highmem(void)
2362 struct highmem_pbe
*pbe
= highmem_pblist
;
2368 buf
= get_image_page(GFP_ATOMIC
, PG_SAFE
);
2373 swap_two_pages_data(pbe
->copy_page
, pbe
->orig_page
, buf
);
2376 free_image_page(buf
, PG_UNSAFE_CLEAR
);
2379 #endif /* CONFIG_HIGHMEM */