sysfs: sysfs_sd_setattr set iattrs unconditionally
[linux-2.6/linux-2.6-openrd.git] / kernel / power / snapshot.c
blob36cb168e4330936a9120691fd19e80ddb8d2db9d
1 /*
2 * linux/kernel/power/snapshot.c
4 * This file provides system snapshot/restore functionality for swsusp.
6 * Copyright (C) 1998-2005 Pavel Machek <pavel@suse.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>
15 #include <linux/mm.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>
21 #include <linux/pm.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>
30 #include <asm/uaccess.h>
31 #include <asm/mmu_context.h>
32 #include <asm/pgtable.h>
33 #include <asm/tlbflush.h>
34 #include <asm/io.h>
36 #include "power.h"
38 static int swsusp_page_is_free(struct page *);
39 static void swsusp_set_page_forbidden(struct page *);
40 static void swsusp_unset_page_forbidden(struct page *);
43 * Preferred image size in bytes (tunable via /sys/power/image_size).
44 * When it is set to N, swsusp will do its best to ensure the image
45 * size will not exceed N bytes, but if that is impossible, it will
46 * try to create the smallest image possible.
48 unsigned long image_size = 500 * 1024 * 1024;
50 /* List of PBEs needed for restoring the pages that were allocated before
51 * the suspend and included in the suspend image, but have also been
52 * allocated by the "resume" kernel, so their contents cannot be written
53 * directly to their "original" page frames.
55 struct pbe *restore_pblist;
57 /* Pointer to an auxiliary buffer (1 page) */
58 static void *buffer;
60 /**
61 * @safe_needed - on resume, for storing the PBE list and the image,
62 * we can only use memory pages that do not conflict with the pages
63 * used before suspend. The unsafe pages have PageNosaveFree set
64 * and we count them using unsafe_pages.
66 * Each allocated image page is marked as PageNosave and PageNosaveFree
67 * so that swsusp_free() can release it.
70 #define PG_ANY 0
71 #define PG_SAFE 1
72 #define PG_UNSAFE_CLEAR 1
73 #define PG_UNSAFE_KEEP 0
75 static unsigned int allocated_unsafe_pages;
77 static void *get_image_page(gfp_t gfp_mask, int safe_needed)
79 void *res;
81 res = (void *)get_zeroed_page(gfp_mask);
82 if (safe_needed)
83 while (res && swsusp_page_is_free(virt_to_page(res))) {
84 /* The page is unsafe, mark it for swsusp_free() */
85 swsusp_set_page_forbidden(virt_to_page(res));
86 allocated_unsafe_pages++;
87 res = (void *)get_zeroed_page(gfp_mask);
89 if (res) {
90 swsusp_set_page_forbidden(virt_to_page(res));
91 swsusp_set_page_free(virt_to_page(res));
93 return res;
96 unsigned long get_safe_page(gfp_t gfp_mask)
98 return (unsigned long)get_image_page(gfp_mask, PG_SAFE);
101 static struct page *alloc_image_page(gfp_t gfp_mask)
103 struct page *page;
105 page = alloc_page(gfp_mask);
106 if (page) {
107 swsusp_set_page_forbidden(page);
108 swsusp_set_page_free(page);
110 return page;
114 * free_image_page - free page represented by @addr, allocated with
115 * get_image_page (page flags set by it must be cleared)
118 static inline void free_image_page(void *addr, int clear_nosave_free)
120 struct page *page;
122 BUG_ON(!virt_addr_valid(addr));
124 page = virt_to_page(addr);
126 swsusp_unset_page_forbidden(page);
127 if (clear_nosave_free)
128 swsusp_unset_page_free(page);
130 __free_page(page);
133 /* struct linked_page is used to build chains of pages */
135 #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
137 struct linked_page {
138 struct linked_page *next;
139 char data[LINKED_PAGE_DATA_SIZE];
140 } __attribute__((packed));
142 static inline void
143 free_list_of_pages(struct linked_page *list, int clear_page_nosave)
145 while (list) {
146 struct linked_page *lp = list->next;
148 free_image_page(list, clear_page_nosave);
149 list = lp;
154 * struct chain_allocator is used for allocating small objects out of
155 * a linked list of pages called 'the chain'.
157 * The chain grows each time when there is no room for a new object in
158 * the current page. The allocated objects cannot be freed individually.
159 * It is only possible to free them all at once, by freeing the entire
160 * chain.
162 * NOTE: The chain allocator may be inefficient if the allocated objects
163 * are not much smaller than PAGE_SIZE.
166 struct chain_allocator {
167 struct linked_page *chain; /* the chain */
168 unsigned int used_space; /* total size of objects allocated out
169 * of the current page
171 gfp_t gfp_mask; /* mask for allocating pages */
172 int safe_needed; /* if set, only "safe" pages are allocated */
175 static void
176 chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed)
178 ca->chain = NULL;
179 ca->used_space = LINKED_PAGE_DATA_SIZE;
180 ca->gfp_mask = gfp_mask;
181 ca->safe_needed = safe_needed;
184 static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
186 void *ret;
188 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
189 struct linked_page *lp;
191 lp = get_image_page(ca->gfp_mask, ca->safe_needed);
192 if (!lp)
193 return NULL;
195 lp->next = ca->chain;
196 ca->chain = lp;
197 ca->used_space = 0;
199 ret = ca->chain->data + ca->used_space;
200 ca->used_space += size;
201 return ret;
205 * Data types related to memory bitmaps.
207 * Memory bitmap is a structure consiting of many linked lists of
208 * objects. The main list's elements are of type struct zone_bitmap
209 * and each of them corresonds to one zone. For each zone bitmap
210 * object there is a list of objects of type struct bm_block that
211 * represent each blocks of bitmap in which information is stored.
213 * struct memory_bitmap contains a pointer to the main list of zone
214 * bitmap objects, a struct bm_position used for browsing the bitmap,
215 * and a pointer to the list of pages used for allocating all of the
216 * zone bitmap objects and bitmap block objects.
218 * NOTE: It has to be possible to lay out the bitmap in memory
219 * using only allocations of order 0. Additionally, the bitmap is
220 * designed to work with arbitrary number of zones (this is over the
221 * top for now, but let's avoid making unnecessary assumptions ;-).
223 * struct zone_bitmap contains a pointer to a list of bitmap block
224 * objects and a pointer to the bitmap block object that has been
225 * most recently used for setting bits. Additionally, it contains the
226 * pfns that correspond to the start and end of the represented zone.
228 * struct bm_block contains a pointer to the memory page in which
229 * information is stored (in the form of a block of bitmap)
230 * It also contains the pfns that correspond to the start and end of
231 * the represented memory area.
234 #define BM_END_OF_MAP (~0UL)
236 #define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
238 struct bm_block {
239 struct list_head hook; /* hook into a list of bitmap blocks */
240 unsigned long start_pfn; /* pfn represented by the first bit */
241 unsigned long end_pfn; /* pfn represented by the last bit plus 1 */
242 unsigned long *data; /* bitmap representing pages */
245 static inline unsigned long bm_block_bits(struct bm_block *bb)
247 return bb->end_pfn - bb->start_pfn;
250 /* strcut bm_position is used for browsing memory bitmaps */
252 struct bm_position {
253 struct bm_block *block;
254 int bit;
257 struct memory_bitmap {
258 struct list_head blocks; /* list of bitmap blocks */
259 struct linked_page *p_list; /* list of pages used to store zone
260 * bitmap objects and bitmap block
261 * objects
263 struct bm_position cur; /* most recently used bit position */
266 /* Functions that operate on memory bitmaps */
268 static void memory_bm_position_reset(struct memory_bitmap *bm)
270 bm->cur.block = list_entry(bm->blocks.next, struct bm_block, hook);
271 bm->cur.bit = 0;
274 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
277 * create_bm_block_list - create a list of block bitmap objects
278 * @pages - number of pages to track
279 * @list - list to put the allocated blocks into
280 * @ca - chain allocator to be used for allocating memory
282 static int create_bm_block_list(unsigned long pages,
283 struct list_head *list,
284 struct chain_allocator *ca)
286 unsigned int nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
288 while (nr_blocks-- > 0) {
289 struct bm_block *bb;
291 bb = chain_alloc(ca, sizeof(struct bm_block));
292 if (!bb)
293 return -ENOMEM;
294 list_add(&bb->hook, list);
297 return 0;
300 struct mem_extent {
301 struct list_head hook;
302 unsigned long start;
303 unsigned long end;
307 * free_mem_extents - free a list of memory extents
308 * @list - list of extents to empty
310 static void free_mem_extents(struct list_head *list)
312 struct mem_extent *ext, *aux;
314 list_for_each_entry_safe(ext, aux, list, hook) {
315 list_del(&ext->hook);
316 kfree(ext);
321 * create_mem_extents - create a list of memory extents representing
322 * contiguous ranges of PFNs
323 * @list - list to put the extents into
324 * @gfp_mask - mask to use for memory allocations
326 static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
328 struct zone *zone;
330 INIT_LIST_HEAD(list);
332 for_each_populated_zone(zone) {
333 unsigned long zone_start, zone_end;
334 struct mem_extent *ext, *cur, *aux;
336 zone_start = zone->zone_start_pfn;
337 zone_end = zone->zone_start_pfn + zone->spanned_pages;
339 list_for_each_entry(ext, list, hook)
340 if (zone_start <= ext->end)
341 break;
343 if (&ext->hook == list || zone_end < ext->start) {
344 /* New extent is necessary */
345 struct mem_extent *new_ext;
347 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
348 if (!new_ext) {
349 free_mem_extents(list);
350 return -ENOMEM;
352 new_ext->start = zone_start;
353 new_ext->end = zone_end;
354 list_add_tail(&new_ext->hook, &ext->hook);
355 continue;
358 /* Merge this zone's range of PFNs with the existing one */
359 if (zone_start < ext->start)
360 ext->start = zone_start;
361 if (zone_end > ext->end)
362 ext->end = zone_end;
364 /* More merging may be possible */
365 cur = ext;
366 list_for_each_entry_safe_continue(cur, aux, list, hook) {
367 if (zone_end < cur->start)
368 break;
369 if (zone_end < cur->end)
370 ext->end = cur->end;
371 list_del(&cur->hook);
372 kfree(cur);
376 return 0;
380 * memory_bm_create - allocate memory for a memory bitmap
382 static int
383 memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed)
385 struct chain_allocator ca;
386 struct list_head mem_extents;
387 struct mem_extent *ext;
388 int error;
390 chain_init(&ca, gfp_mask, safe_needed);
391 INIT_LIST_HEAD(&bm->blocks);
393 error = create_mem_extents(&mem_extents, gfp_mask);
394 if (error)
395 return error;
397 list_for_each_entry(ext, &mem_extents, hook) {
398 struct bm_block *bb;
399 unsigned long pfn = ext->start;
400 unsigned long pages = ext->end - ext->start;
402 bb = list_entry(bm->blocks.prev, struct bm_block, hook);
404 error = create_bm_block_list(pages, bm->blocks.prev, &ca);
405 if (error)
406 goto Error;
408 list_for_each_entry_continue(bb, &bm->blocks, hook) {
409 bb->data = get_image_page(gfp_mask, safe_needed);
410 if (!bb->data) {
411 error = -ENOMEM;
412 goto Error;
415 bb->start_pfn = pfn;
416 if (pages >= BM_BITS_PER_BLOCK) {
417 pfn += BM_BITS_PER_BLOCK;
418 pages -= BM_BITS_PER_BLOCK;
419 } else {
420 /* This is executed only once in the loop */
421 pfn += pages;
423 bb->end_pfn = pfn;
427 bm->p_list = ca.chain;
428 memory_bm_position_reset(bm);
429 Exit:
430 free_mem_extents(&mem_extents);
431 return error;
433 Error:
434 bm->p_list = ca.chain;
435 memory_bm_free(bm, PG_UNSAFE_CLEAR);
436 goto Exit;
440 * memory_bm_free - free memory occupied by the memory bitmap @bm
442 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
444 struct bm_block *bb;
446 list_for_each_entry(bb, &bm->blocks, hook)
447 if (bb->data)
448 free_image_page(bb->data, clear_nosave_free);
450 free_list_of_pages(bm->p_list, clear_nosave_free);
452 INIT_LIST_HEAD(&bm->blocks);
456 * memory_bm_find_bit - find the bit in the bitmap @bm that corresponds
457 * to given pfn. The cur_zone_bm member of @bm and the cur_block member
458 * of @bm->cur_zone_bm are updated.
460 static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
461 void **addr, unsigned int *bit_nr)
463 struct bm_block *bb;
466 * Check if the pfn corresponds to the current bitmap block and find
467 * the block where it fits if this is not the case.
469 bb = bm->cur.block;
470 if (pfn < bb->start_pfn)
471 list_for_each_entry_continue_reverse(bb, &bm->blocks, hook)
472 if (pfn >= bb->start_pfn)
473 break;
475 if (pfn >= bb->end_pfn)
476 list_for_each_entry_continue(bb, &bm->blocks, hook)
477 if (pfn >= bb->start_pfn && pfn < bb->end_pfn)
478 break;
480 if (&bb->hook == &bm->blocks)
481 return -EFAULT;
483 /* The block has been found */
484 bm->cur.block = bb;
485 pfn -= bb->start_pfn;
486 bm->cur.bit = pfn + 1;
487 *bit_nr = pfn;
488 *addr = bb->data;
489 return 0;
492 static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
494 void *addr;
495 unsigned int bit;
496 int error;
498 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
499 BUG_ON(error);
500 set_bit(bit, addr);
503 static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
505 void *addr;
506 unsigned int bit;
507 int error;
509 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
510 if (!error)
511 set_bit(bit, addr);
512 return error;
515 static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
517 void *addr;
518 unsigned int bit;
519 int error;
521 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
522 BUG_ON(error);
523 clear_bit(bit, addr);
526 static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
528 void *addr;
529 unsigned int bit;
530 int error;
532 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
533 BUG_ON(error);
534 return test_bit(bit, addr);
537 static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
539 void *addr;
540 unsigned int bit;
542 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
546 * memory_bm_next_pfn - find the pfn that corresponds to the next set bit
547 * in the bitmap @bm. If the pfn cannot be found, BM_END_OF_MAP is
548 * returned.
550 * It is required to run memory_bm_position_reset() before the first call to
551 * this function.
554 static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
556 struct bm_block *bb;
557 int bit;
559 bb = bm->cur.block;
560 do {
561 bit = bm->cur.bit;
562 bit = find_next_bit(bb->data, bm_block_bits(bb), bit);
563 if (bit < bm_block_bits(bb))
564 goto Return_pfn;
566 bb = list_entry(bb->hook.next, struct bm_block, hook);
567 bm->cur.block = bb;
568 bm->cur.bit = 0;
569 } while (&bb->hook != &bm->blocks);
571 memory_bm_position_reset(bm);
572 return BM_END_OF_MAP;
574 Return_pfn:
575 bm->cur.bit = bit + 1;
576 return bb->start_pfn + bit;
580 * This structure represents a range of page frames the contents of which
581 * should not be saved during the suspend.
584 struct nosave_region {
585 struct list_head list;
586 unsigned long start_pfn;
587 unsigned long end_pfn;
590 static LIST_HEAD(nosave_regions);
593 * register_nosave_region - register a range of page frames the contents
594 * of which should not be saved during the suspend (to be used in the early
595 * initialization code)
598 void __init
599 __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn,
600 int use_kmalloc)
602 struct nosave_region *region;
604 if (start_pfn >= end_pfn)
605 return;
607 if (!list_empty(&nosave_regions)) {
608 /* Try to extend the previous region (they should be sorted) */
609 region = list_entry(nosave_regions.prev,
610 struct nosave_region, list);
611 if (region->end_pfn == start_pfn) {
612 region->end_pfn = end_pfn;
613 goto Report;
616 if (use_kmalloc) {
617 /* during init, this shouldn't fail */
618 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
619 BUG_ON(!region);
620 } else
621 /* This allocation cannot fail */
622 region = alloc_bootmem(sizeof(struct nosave_region));
623 region->start_pfn = start_pfn;
624 region->end_pfn = end_pfn;
625 list_add_tail(&region->list, &nosave_regions);
626 Report:
627 printk(KERN_INFO "PM: Registered nosave memory: %016lx - %016lx\n",
628 start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT);
632 * Set bits in this map correspond to the page frames the contents of which
633 * should not be saved during the suspend.
635 static struct memory_bitmap *forbidden_pages_map;
637 /* Set bits in this map correspond to free page frames. */
638 static struct memory_bitmap *free_pages_map;
641 * Each page frame allocated for creating the image is marked by setting the
642 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
645 void swsusp_set_page_free(struct page *page)
647 if (free_pages_map)
648 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
651 static int swsusp_page_is_free(struct page *page)
653 return free_pages_map ?
654 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
657 void swsusp_unset_page_free(struct page *page)
659 if (free_pages_map)
660 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
663 static void swsusp_set_page_forbidden(struct page *page)
665 if (forbidden_pages_map)
666 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
669 int swsusp_page_is_forbidden(struct page *page)
671 return forbidden_pages_map ?
672 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
675 static void swsusp_unset_page_forbidden(struct page *page)
677 if (forbidden_pages_map)
678 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
682 * mark_nosave_pages - set bits corresponding to the page frames the
683 * contents of which should not be saved in a given bitmap.
686 static void mark_nosave_pages(struct memory_bitmap *bm)
688 struct nosave_region *region;
690 if (list_empty(&nosave_regions))
691 return;
693 list_for_each_entry(region, &nosave_regions, list) {
694 unsigned long pfn;
696 pr_debug("PM: Marking nosave pages: %016lx - %016lx\n",
697 region->start_pfn << PAGE_SHIFT,
698 region->end_pfn << PAGE_SHIFT);
700 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
701 if (pfn_valid(pfn)) {
703 * It is safe to ignore the result of
704 * mem_bm_set_bit_check() here, since we won't
705 * touch the PFNs for which the error is
706 * returned anyway.
708 mem_bm_set_bit_check(bm, pfn);
714 * create_basic_memory_bitmaps - create bitmaps needed for marking page
715 * frames that should not be saved and free page frames. The pointers
716 * forbidden_pages_map and free_pages_map are only modified if everything
717 * goes well, because we don't want the bits to be used before both bitmaps
718 * are set up.
721 int create_basic_memory_bitmaps(void)
723 struct memory_bitmap *bm1, *bm2;
724 int error = 0;
726 BUG_ON(forbidden_pages_map || free_pages_map);
728 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
729 if (!bm1)
730 return -ENOMEM;
732 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
733 if (error)
734 goto Free_first_object;
736 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
737 if (!bm2)
738 goto Free_first_bitmap;
740 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
741 if (error)
742 goto Free_second_object;
744 forbidden_pages_map = bm1;
745 free_pages_map = bm2;
746 mark_nosave_pages(forbidden_pages_map);
748 pr_debug("PM: Basic memory bitmaps created\n");
750 return 0;
752 Free_second_object:
753 kfree(bm2);
754 Free_first_bitmap:
755 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
756 Free_first_object:
757 kfree(bm1);
758 return -ENOMEM;
762 * free_basic_memory_bitmaps - free memory bitmaps allocated by
763 * create_basic_memory_bitmaps(). The auxiliary pointers are necessary
764 * so that the bitmaps themselves are not referred to while they are being
765 * freed.
768 void free_basic_memory_bitmaps(void)
770 struct memory_bitmap *bm1, *bm2;
772 BUG_ON(!(forbidden_pages_map && free_pages_map));
774 bm1 = forbidden_pages_map;
775 bm2 = free_pages_map;
776 forbidden_pages_map = NULL;
777 free_pages_map = NULL;
778 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
779 kfree(bm1);
780 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
781 kfree(bm2);
783 pr_debug("PM: Basic memory bitmaps freed\n");
787 * snapshot_additional_pages - estimate the number of additional pages
788 * be needed for setting up the suspend image data structures for given
789 * zone (usually the returned value is greater than the exact number)
792 unsigned int snapshot_additional_pages(struct zone *zone)
794 unsigned int res;
796 res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
797 res += DIV_ROUND_UP(res * sizeof(struct bm_block), PAGE_SIZE);
798 return 2 * res;
801 #ifdef CONFIG_HIGHMEM
803 * count_free_highmem_pages - compute the total number of free highmem
804 * pages, system-wide.
807 static unsigned int count_free_highmem_pages(void)
809 struct zone *zone;
810 unsigned int cnt = 0;
812 for_each_populated_zone(zone)
813 if (is_highmem(zone))
814 cnt += zone_page_state(zone, NR_FREE_PAGES);
816 return cnt;
820 * saveable_highmem_page - Determine whether a highmem page should be
821 * included in the suspend image.
823 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
824 * and it isn't a part of a free chunk of pages.
826 static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
828 struct page *page;
830 if (!pfn_valid(pfn))
831 return NULL;
833 page = pfn_to_page(pfn);
834 if (page_zone(page) != zone)
835 return NULL;
837 BUG_ON(!PageHighMem(page));
839 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) ||
840 PageReserved(page))
841 return NULL;
843 return page;
847 * count_highmem_pages - compute the total number of saveable highmem
848 * pages.
851 static unsigned int count_highmem_pages(void)
853 struct zone *zone;
854 unsigned int n = 0;
856 for_each_populated_zone(zone) {
857 unsigned long pfn, max_zone_pfn;
859 if (!is_highmem(zone))
860 continue;
862 mark_free_pages(zone);
863 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
864 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
865 if (saveable_highmem_page(zone, pfn))
866 n++;
868 return n;
870 #else
871 static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
873 return NULL;
875 #endif /* CONFIG_HIGHMEM */
878 * saveable_page - Determine whether a non-highmem page should be included
879 * in the suspend image.
881 * We should save the page if it isn't Nosave, and is not in the range
882 * of pages statically defined as 'unsaveable', and it isn't a part of
883 * a free chunk of pages.
885 static struct page *saveable_page(struct zone *zone, unsigned long pfn)
887 struct page *page;
889 if (!pfn_valid(pfn))
890 return NULL;
892 page = pfn_to_page(pfn);
893 if (page_zone(page) != zone)
894 return NULL;
896 BUG_ON(PageHighMem(page));
898 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
899 return NULL;
901 if (PageReserved(page)
902 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
903 return NULL;
905 return page;
909 * count_data_pages - compute the total number of saveable non-highmem
910 * pages.
913 static unsigned int count_data_pages(void)
915 struct zone *zone;
916 unsigned long pfn, max_zone_pfn;
917 unsigned int n = 0;
919 for_each_populated_zone(zone) {
920 if (is_highmem(zone))
921 continue;
923 mark_free_pages(zone);
924 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
925 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
926 if (saveable_page(zone, pfn))
927 n++;
929 return n;
932 /* This is needed, because copy_page and memcpy are not usable for copying
933 * task structs.
935 static inline void do_copy_page(long *dst, long *src)
937 int n;
939 for (n = PAGE_SIZE / sizeof(long); n; n--)
940 *dst++ = *src++;
945 * safe_copy_page - check if the page we are going to copy is marked as
946 * present in the kernel page tables (this always is the case if
947 * CONFIG_DEBUG_PAGEALLOC is not set and in that case
948 * kernel_page_present() always returns 'true').
950 static void safe_copy_page(void *dst, struct page *s_page)
952 if (kernel_page_present(s_page)) {
953 do_copy_page(dst, page_address(s_page));
954 } else {
955 kernel_map_pages(s_page, 1, 1);
956 do_copy_page(dst, page_address(s_page));
957 kernel_map_pages(s_page, 1, 0);
962 #ifdef CONFIG_HIGHMEM
963 static inline struct page *
964 page_is_saveable(struct zone *zone, unsigned long pfn)
966 return is_highmem(zone) ?
967 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
970 static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
972 struct page *s_page, *d_page;
973 void *src, *dst;
975 s_page = pfn_to_page(src_pfn);
976 d_page = pfn_to_page(dst_pfn);
977 if (PageHighMem(s_page)) {
978 src = kmap_atomic(s_page, KM_USER0);
979 dst = kmap_atomic(d_page, KM_USER1);
980 do_copy_page(dst, src);
981 kunmap_atomic(src, KM_USER0);
982 kunmap_atomic(dst, KM_USER1);
983 } else {
984 if (PageHighMem(d_page)) {
985 /* Page pointed to by src may contain some kernel
986 * data modified by kmap_atomic()
988 safe_copy_page(buffer, s_page);
989 dst = kmap_atomic(d_page, KM_USER0);
990 memcpy(dst, buffer, PAGE_SIZE);
991 kunmap_atomic(dst, KM_USER0);
992 } else {
993 safe_copy_page(page_address(d_page), s_page);
997 #else
998 #define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1000 static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1002 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1003 pfn_to_page(src_pfn));
1005 #endif /* CONFIG_HIGHMEM */
1007 static void
1008 copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm)
1010 struct zone *zone;
1011 unsigned long pfn;
1013 for_each_populated_zone(zone) {
1014 unsigned long max_zone_pfn;
1016 mark_free_pages(zone);
1017 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1018 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1019 if (page_is_saveable(zone, pfn))
1020 memory_bm_set_bit(orig_bm, pfn);
1022 memory_bm_position_reset(orig_bm);
1023 memory_bm_position_reset(copy_bm);
1024 for(;;) {
1025 pfn = memory_bm_next_pfn(orig_bm);
1026 if (unlikely(pfn == BM_END_OF_MAP))
1027 break;
1028 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1032 /* Total number of image pages */
1033 static unsigned int nr_copy_pages;
1034 /* Number of pages needed for saving the original pfns of the image pages */
1035 static unsigned int nr_meta_pages;
1037 * Numbers of normal and highmem page frames allocated for hibernation image
1038 * before suspending devices.
1040 unsigned int alloc_normal, alloc_highmem;
1042 * Memory bitmap used for marking saveable pages (during hibernation) or
1043 * hibernation image pages (during restore)
1045 static struct memory_bitmap orig_bm;
1047 * Memory bitmap used during hibernation for marking allocated page frames that
1048 * will contain copies of saveable pages. During restore it is initially used
1049 * for marking hibernation image pages, but then the set bits from it are
1050 * duplicated in @orig_bm and it is released. On highmem systems it is next
1051 * used for marking "safe" highmem pages, but it has to be reinitialized for
1052 * this purpose.
1054 static struct memory_bitmap copy_bm;
1057 * swsusp_free - free pages allocated for the suspend.
1059 * Suspend pages are alocated before the atomic copy is made, so we
1060 * need to release them after the resume.
1063 void swsusp_free(void)
1065 struct zone *zone;
1066 unsigned long pfn, max_zone_pfn;
1068 for_each_populated_zone(zone) {
1069 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1070 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1071 if (pfn_valid(pfn)) {
1072 struct page *page = pfn_to_page(pfn);
1074 if (swsusp_page_is_forbidden(page) &&
1075 swsusp_page_is_free(page)) {
1076 swsusp_unset_page_forbidden(page);
1077 swsusp_unset_page_free(page);
1078 __free_page(page);
1082 nr_copy_pages = 0;
1083 nr_meta_pages = 0;
1084 restore_pblist = NULL;
1085 buffer = NULL;
1086 alloc_normal = 0;
1087 alloc_highmem = 0;
1090 /* Helper functions used for the shrinking of memory. */
1092 #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1095 * preallocate_image_pages - Allocate a number of pages for hibernation image
1096 * @nr_pages: Number of page frames to allocate.
1097 * @mask: GFP flags to use for the allocation.
1099 * Return value: Number of page frames actually allocated
1101 static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1103 unsigned long nr_alloc = 0;
1105 while (nr_pages > 0) {
1106 struct page *page;
1108 page = alloc_image_page(mask);
1109 if (!page)
1110 break;
1111 memory_bm_set_bit(&copy_bm, page_to_pfn(page));
1112 if (PageHighMem(page))
1113 alloc_highmem++;
1114 else
1115 alloc_normal++;
1116 nr_pages--;
1117 nr_alloc++;
1120 return nr_alloc;
1123 static unsigned long preallocate_image_memory(unsigned long nr_pages)
1125 return preallocate_image_pages(nr_pages, GFP_IMAGE);
1128 #ifdef CONFIG_HIGHMEM
1129 static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1131 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1135 * __fraction - Compute (an approximation of) x * (multiplier / base)
1137 static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1139 x *= multiplier;
1140 do_div(x, base);
1141 return (unsigned long)x;
1144 static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1145 unsigned long highmem,
1146 unsigned long total)
1148 unsigned long alloc = __fraction(nr_pages, highmem, total);
1150 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1152 #else /* CONFIG_HIGHMEM */
1153 static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1155 return 0;
1158 static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1159 unsigned long highmem,
1160 unsigned long total)
1162 return 0;
1164 #endif /* CONFIG_HIGHMEM */
1167 * free_unnecessary_pages - Release preallocated pages not needed for the image
1169 static void free_unnecessary_pages(void)
1171 unsigned long save_highmem, to_free_normal, to_free_highmem;
1173 to_free_normal = alloc_normal - count_data_pages();
1174 save_highmem = count_highmem_pages();
1175 if (alloc_highmem > save_highmem) {
1176 to_free_highmem = alloc_highmem - save_highmem;
1177 } else {
1178 to_free_highmem = 0;
1179 to_free_normal -= save_highmem - alloc_highmem;
1182 memory_bm_position_reset(&copy_bm);
1184 while (to_free_normal > 0 && to_free_highmem > 0) {
1185 unsigned long pfn = memory_bm_next_pfn(&copy_bm);
1186 struct page *page = pfn_to_page(pfn);
1188 if (PageHighMem(page)) {
1189 if (!to_free_highmem)
1190 continue;
1191 to_free_highmem--;
1192 alloc_highmem--;
1193 } else {
1194 if (!to_free_normal)
1195 continue;
1196 to_free_normal--;
1197 alloc_normal--;
1199 memory_bm_clear_bit(&copy_bm, pfn);
1200 swsusp_unset_page_forbidden(page);
1201 swsusp_unset_page_free(page);
1202 __free_page(page);
1207 * minimum_image_size - Estimate the minimum acceptable size of an image
1208 * @saveable: Number of saveable pages in the system.
1210 * We want to avoid attempting to free too much memory too hard, so estimate the
1211 * minimum acceptable size of a hibernation image to use as the lower limit for
1212 * preallocating memory.
1214 * We assume that the minimum image size should be proportional to
1216 * [number of saveable pages] - [number of pages that can be freed in theory]
1218 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1219 * and (3) inactive anonymouns pages, (4) active and (5) inactive file pages,
1220 * minus mapped file pages.
1222 static unsigned long minimum_image_size(unsigned long saveable)
1224 unsigned long size;
1226 size = global_page_state(NR_SLAB_RECLAIMABLE)
1227 + global_page_state(NR_ACTIVE_ANON)
1228 + global_page_state(NR_INACTIVE_ANON)
1229 + global_page_state(NR_ACTIVE_FILE)
1230 + global_page_state(NR_INACTIVE_FILE)
1231 - global_page_state(NR_FILE_MAPPED);
1233 return saveable <= size ? 0 : saveable - size;
1237 * hibernate_preallocate_memory - Preallocate memory for hibernation image
1239 * To create a hibernation image it is necessary to make a copy of every page
1240 * frame in use. We also need a number of page frames to be free during
1241 * hibernation for allocations made while saving the image and for device
1242 * drivers, in case they need to allocate memory from their hibernation
1243 * callbacks (these two numbers are given by PAGES_FOR_IO and SPARE_PAGES,
1244 * respectively, both of which are rough estimates). To make this happen, we
1245 * compute the total number of available page frames and allocate at least
1247 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2 + 2 * SPARE_PAGES
1249 * of them, which corresponds to the maximum size of a hibernation image.
1251 * If image_size is set below the number following from the above formula,
1252 * the preallocation of memory is continued until the total number of saveable
1253 * pages in the system is below the requested image size or the minimum
1254 * acceptable image size returned by minimum_image_size(), whichever is greater.
1256 int hibernate_preallocate_memory(void)
1258 struct zone *zone;
1259 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1260 unsigned long alloc, save_highmem, pages_highmem;
1261 struct timeval start, stop;
1262 int error;
1264 printk(KERN_INFO "PM: Preallocating image memory... ");
1265 do_gettimeofday(&start);
1267 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1268 if (error)
1269 goto err_out;
1271 error = memory_bm_create(&copy_bm, GFP_IMAGE, PG_ANY);
1272 if (error)
1273 goto err_out;
1275 alloc_normal = 0;
1276 alloc_highmem = 0;
1278 /* Count the number of saveable data pages. */
1279 save_highmem = count_highmem_pages();
1280 saveable = count_data_pages();
1283 * Compute the total number of page frames we can use (count) and the
1284 * number of pages needed for image metadata (size).
1286 count = saveable;
1287 saveable += save_highmem;
1288 highmem = save_highmem;
1289 size = 0;
1290 for_each_populated_zone(zone) {
1291 size += snapshot_additional_pages(zone);
1292 if (is_highmem(zone))
1293 highmem += zone_page_state(zone, NR_FREE_PAGES);
1294 else
1295 count += zone_page_state(zone, NR_FREE_PAGES);
1297 count += highmem;
1298 count -= totalreserve_pages;
1300 /* Compute the maximum number of saveable pages to leave in memory. */
1301 max_size = (count - (size + PAGES_FOR_IO)) / 2 - 2 * SPARE_PAGES;
1302 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1303 if (size > max_size)
1304 size = max_size;
1306 * If the maximum is not less than the current number of saveable pages
1307 * in memory, allocate page frames for the image and we're done.
1309 if (size >= saveable) {
1310 pages = preallocate_image_highmem(save_highmem);
1311 pages += preallocate_image_memory(saveable - pages);
1312 goto out;
1315 /* Estimate the minimum size of the image. */
1316 pages = minimum_image_size(saveable);
1317 if (size < pages)
1318 size = min_t(unsigned long, pages, max_size);
1321 * Let the memory management subsystem know that we're going to need a
1322 * large number of page frames to allocate and make it free some memory.
1323 * NOTE: If this is not done, performance will be hurt badly in some
1324 * test cases.
1326 shrink_all_memory(saveable - size);
1329 * The number of saveable pages in memory was too high, so apply some
1330 * pressure to decrease it. First, make room for the largest possible
1331 * image and fail if that doesn't work. Next, try to decrease the size
1332 * of the image as much as indicated by 'size' using allocations from
1333 * highmem and non-highmem zones separately.
1335 pages_highmem = preallocate_image_highmem(highmem / 2);
1336 alloc = (count - max_size) - pages_highmem;
1337 pages = preallocate_image_memory(alloc);
1338 if (pages < alloc)
1339 goto err_out;
1340 size = max_size - size;
1341 alloc = size;
1342 size = preallocate_highmem_fraction(size, highmem, count);
1343 pages_highmem += size;
1344 alloc -= size;
1345 pages += preallocate_image_memory(alloc);
1346 pages += pages_highmem;
1349 * We only need as many page frames for the image as there are saveable
1350 * pages in memory, but we have allocated more. Release the excessive
1351 * ones now.
1353 free_unnecessary_pages();
1355 out:
1356 do_gettimeofday(&stop);
1357 printk(KERN_CONT "done (allocated %lu pages)\n", pages);
1358 swsusp_show_speed(&start, &stop, pages, "Allocated");
1360 return 0;
1362 err_out:
1363 printk(KERN_CONT "\n");
1364 swsusp_free();
1365 return -ENOMEM;
1368 #ifdef CONFIG_HIGHMEM
1370 * count_pages_for_highmem - compute the number of non-highmem pages
1371 * that will be necessary for creating copies of highmem pages.
1374 static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1376 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1378 if (free_highmem >= nr_highmem)
1379 nr_highmem = 0;
1380 else
1381 nr_highmem -= free_highmem;
1383 return nr_highmem;
1385 #else
1386 static unsigned int
1387 count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1388 #endif /* CONFIG_HIGHMEM */
1391 * enough_free_mem - Make sure we have enough free memory for the
1392 * snapshot image.
1395 static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1397 struct zone *zone;
1398 unsigned int free = alloc_normal;
1400 for_each_populated_zone(zone)
1401 if (!is_highmem(zone))
1402 free += zone_page_state(zone, NR_FREE_PAGES);
1404 nr_pages += count_pages_for_highmem(nr_highmem);
1405 pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n",
1406 nr_pages, PAGES_FOR_IO, free);
1408 return free > nr_pages + PAGES_FOR_IO;
1411 #ifdef CONFIG_HIGHMEM
1413 * get_highmem_buffer - if there are some highmem pages in the suspend
1414 * image, we may need the buffer to copy them and/or load their data.
1417 static inline int get_highmem_buffer(int safe_needed)
1419 buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
1420 return buffer ? 0 : -ENOMEM;
1424 * alloc_highmem_image_pages - allocate some highmem pages for the image.
1425 * Try to allocate as many pages as needed, but if the number of free
1426 * highmem pages is lesser than that, allocate them all.
1429 static inline unsigned int
1430 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int nr_highmem)
1432 unsigned int to_alloc = count_free_highmem_pages();
1434 if (to_alloc > nr_highmem)
1435 to_alloc = nr_highmem;
1437 nr_highmem -= to_alloc;
1438 while (to_alloc-- > 0) {
1439 struct page *page;
1441 page = alloc_image_page(__GFP_HIGHMEM);
1442 memory_bm_set_bit(bm, page_to_pfn(page));
1444 return nr_highmem;
1446 #else
1447 static inline int get_highmem_buffer(int safe_needed) { return 0; }
1449 static inline unsigned int
1450 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int n) { return 0; }
1451 #endif /* CONFIG_HIGHMEM */
1454 * swsusp_alloc - allocate memory for the suspend image
1456 * We first try to allocate as many highmem pages as there are
1457 * saveable highmem pages in the system. If that fails, we allocate
1458 * non-highmem pages for the copies of the remaining highmem ones.
1460 * In this approach it is likely that the copies of highmem pages will
1461 * also be located in the high memory, because of the way in which
1462 * copy_data_pages() works.
1465 static int
1466 swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm,
1467 unsigned int nr_pages, unsigned int nr_highmem)
1469 int error = 0;
1471 if (nr_highmem > 0) {
1472 error = get_highmem_buffer(PG_ANY);
1473 if (error)
1474 goto err_out;
1475 if (nr_highmem > alloc_highmem) {
1476 nr_highmem -= alloc_highmem;
1477 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1480 if (nr_pages > alloc_normal) {
1481 nr_pages -= alloc_normal;
1482 while (nr_pages-- > 0) {
1483 struct page *page;
1485 page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
1486 if (!page)
1487 goto err_out;
1488 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1492 return 0;
1494 err_out:
1495 swsusp_free();
1496 return error;
1499 asmlinkage int swsusp_save(void)
1501 unsigned int nr_pages, nr_highmem;
1503 printk(KERN_INFO "PM: Creating hibernation image: \n");
1505 drain_local_pages(NULL);
1506 nr_pages = count_data_pages();
1507 nr_highmem = count_highmem_pages();
1508 printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem);
1510 if (!enough_free_mem(nr_pages, nr_highmem)) {
1511 printk(KERN_ERR "PM: Not enough free memory\n");
1512 return -ENOMEM;
1515 if (swsusp_alloc(&orig_bm, &copy_bm, nr_pages, nr_highmem)) {
1516 printk(KERN_ERR "PM: Memory allocation failed\n");
1517 return -ENOMEM;
1520 /* During allocating of suspend pagedir, new cold pages may appear.
1521 * Kill them.
1523 drain_local_pages(NULL);
1524 copy_data_pages(&copy_bm, &orig_bm);
1527 * End of critical section. From now on, we can write to memory,
1528 * but we should not touch disk. This specially means we must _not_
1529 * touch swap space! Except we must write out our image of course.
1532 nr_pages += nr_highmem;
1533 nr_copy_pages = nr_pages;
1534 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1536 printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n",
1537 nr_pages);
1539 return 0;
1542 #ifndef CONFIG_ARCH_HIBERNATION_HEADER
1543 static int init_header_complete(struct swsusp_info *info)
1545 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
1546 info->version_code = LINUX_VERSION_CODE;
1547 return 0;
1550 static char *check_image_kernel(struct swsusp_info *info)
1552 if (info->version_code != LINUX_VERSION_CODE)
1553 return "kernel version";
1554 if (strcmp(info->uts.sysname,init_utsname()->sysname))
1555 return "system type";
1556 if (strcmp(info->uts.release,init_utsname()->release))
1557 return "kernel release";
1558 if (strcmp(info->uts.version,init_utsname()->version))
1559 return "version";
1560 if (strcmp(info->uts.machine,init_utsname()->machine))
1561 return "machine";
1562 return NULL;
1564 #endif /* CONFIG_ARCH_HIBERNATION_HEADER */
1566 unsigned long snapshot_get_image_size(void)
1568 return nr_copy_pages + nr_meta_pages + 1;
1571 static int init_header(struct swsusp_info *info)
1573 memset(info, 0, sizeof(struct swsusp_info));
1574 info->num_physpages = num_physpages;
1575 info->image_pages = nr_copy_pages;
1576 info->pages = snapshot_get_image_size();
1577 info->size = info->pages;
1578 info->size <<= PAGE_SHIFT;
1579 return init_header_complete(info);
1583 * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm
1584 * are stored in the array @buf[] (1 page at a time)
1587 static inline void
1588 pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
1590 int j;
1592 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1593 buf[j] = memory_bm_next_pfn(bm);
1594 if (unlikely(buf[j] == BM_END_OF_MAP))
1595 break;
1600 * snapshot_read_next - used for reading the system memory snapshot.
1602 * On the first call to it @handle should point to a zeroed
1603 * snapshot_handle structure. The structure gets updated and a pointer
1604 * to it should be passed to this function every next time.
1606 * The @count parameter should contain the number of bytes the caller
1607 * wants to read from the snapshot. It must not be zero.
1609 * On success the function returns a positive number. Then, the caller
1610 * is allowed to read up to the returned number of bytes from the memory
1611 * location computed by the data_of() macro. The number returned
1612 * may be smaller than @count, but this only happens if the read would
1613 * cross a page boundary otherwise.
1615 * The function returns 0 to indicate the end of data stream condition,
1616 * and a negative number is returned on error. In such cases the
1617 * structure pointed to by @handle is not updated and should not be used
1618 * any more.
1621 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
1623 if (handle->cur > nr_meta_pages + nr_copy_pages)
1624 return 0;
1626 if (!buffer) {
1627 /* This makes the buffer be freed by swsusp_free() */
1628 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
1629 if (!buffer)
1630 return -ENOMEM;
1632 if (!handle->offset) {
1633 int error;
1635 error = init_header((struct swsusp_info *)buffer);
1636 if (error)
1637 return error;
1638 handle->buffer = buffer;
1639 memory_bm_position_reset(&orig_bm);
1640 memory_bm_position_reset(&copy_bm);
1642 if (handle->prev < handle->cur) {
1643 if (handle->cur <= nr_meta_pages) {
1644 memset(buffer, 0, PAGE_SIZE);
1645 pack_pfns(buffer, &orig_bm);
1646 } else {
1647 struct page *page;
1649 page = pfn_to_page(memory_bm_next_pfn(&copy_bm));
1650 if (PageHighMem(page)) {
1651 /* Highmem pages are copied to the buffer,
1652 * because we can't return with a kmapped
1653 * highmem page (we may not be called again).
1655 void *kaddr;
1657 kaddr = kmap_atomic(page, KM_USER0);
1658 memcpy(buffer, kaddr, PAGE_SIZE);
1659 kunmap_atomic(kaddr, KM_USER0);
1660 handle->buffer = buffer;
1661 } else {
1662 handle->buffer = page_address(page);
1665 handle->prev = handle->cur;
1667 handle->buf_offset = handle->cur_offset;
1668 if (handle->cur_offset + count >= PAGE_SIZE) {
1669 count = PAGE_SIZE - handle->cur_offset;
1670 handle->cur_offset = 0;
1671 handle->cur++;
1672 } else {
1673 handle->cur_offset += count;
1675 handle->offset += count;
1676 return count;
1680 * mark_unsafe_pages - mark the pages that cannot be used for storing
1681 * the image during resume, because they conflict with the pages that
1682 * had been used before suspend
1685 static int mark_unsafe_pages(struct memory_bitmap *bm)
1687 struct zone *zone;
1688 unsigned long pfn, max_zone_pfn;
1690 /* Clear page flags */
1691 for_each_populated_zone(zone) {
1692 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1693 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1694 if (pfn_valid(pfn))
1695 swsusp_unset_page_free(pfn_to_page(pfn));
1698 /* Mark pages that correspond to the "original" pfns as "unsafe" */
1699 memory_bm_position_reset(bm);
1700 do {
1701 pfn = memory_bm_next_pfn(bm);
1702 if (likely(pfn != BM_END_OF_MAP)) {
1703 if (likely(pfn_valid(pfn)))
1704 swsusp_set_page_free(pfn_to_page(pfn));
1705 else
1706 return -EFAULT;
1708 } while (pfn != BM_END_OF_MAP);
1710 allocated_unsafe_pages = 0;
1712 return 0;
1715 static void
1716 duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src)
1718 unsigned long pfn;
1720 memory_bm_position_reset(src);
1721 pfn = memory_bm_next_pfn(src);
1722 while (pfn != BM_END_OF_MAP) {
1723 memory_bm_set_bit(dst, pfn);
1724 pfn = memory_bm_next_pfn(src);
1728 static int check_header(struct swsusp_info *info)
1730 char *reason;
1732 reason = check_image_kernel(info);
1733 if (!reason && info->num_physpages != num_physpages)
1734 reason = "memory size";
1735 if (reason) {
1736 printk(KERN_ERR "PM: Image mismatch: %s\n", reason);
1737 return -EPERM;
1739 return 0;
1743 * load header - check the image header and copy data from it
1746 static int
1747 load_header(struct swsusp_info *info)
1749 int error;
1751 restore_pblist = NULL;
1752 error = check_header(info);
1753 if (!error) {
1754 nr_copy_pages = info->image_pages;
1755 nr_meta_pages = info->pages - info->image_pages - 1;
1757 return error;
1761 * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set
1762 * the corresponding bit in the memory bitmap @bm
1764 static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
1766 int j;
1768 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1769 if (unlikely(buf[j] == BM_END_OF_MAP))
1770 break;
1772 if (memory_bm_pfn_present(bm, buf[j]))
1773 memory_bm_set_bit(bm, buf[j]);
1774 else
1775 return -EFAULT;
1778 return 0;
1781 /* List of "safe" pages that may be used to store data loaded from the suspend
1782 * image
1784 static struct linked_page *safe_pages_list;
1786 #ifdef CONFIG_HIGHMEM
1787 /* struct highmem_pbe is used for creating the list of highmem pages that
1788 * should be restored atomically during the resume from disk, because the page
1789 * frames they have occupied before the suspend are in use.
1791 struct highmem_pbe {
1792 struct page *copy_page; /* data is here now */
1793 struct page *orig_page; /* data was here before the suspend */
1794 struct highmem_pbe *next;
1797 /* List of highmem PBEs needed for restoring the highmem pages that were
1798 * allocated before the suspend and included in the suspend image, but have
1799 * also been allocated by the "resume" kernel, so their contents cannot be
1800 * written directly to their "original" page frames.
1802 static struct highmem_pbe *highmem_pblist;
1805 * count_highmem_image_pages - compute the number of highmem pages in the
1806 * suspend image. The bits in the memory bitmap @bm that correspond to the
1807 * image pages are assumed to be set.
1810 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
1812 unsigned long pfn;
1813 unsigned int cnt = 0;
1815 memory_bm_position_reset(bm);
1816 pfn = memory_bm_next_pfn(bm);
1817 while (pfn != BM_END_OF_MAP) {
1818 if (PageHighMem(pfn_to_page(pfn)))
1819 cnt++;
1821 pfn = memory_bm_next_pfn(bm);
1823 return cnt;
1827 * prepare_highmem_image - try to allocate as many highmem pages as
1828 * there are highmem image pages (@nr_highmem_p points to the variable
1829 * containing the number of highmem image pages). The pages that are
1830 * "safe" (ie. will not be overwritten when the suspend image is
1831 * restored) have the corresponding bits set in @bm (it must be
1832 * unitialized).
1834 * NOTE: This function should not be called if there are no highmem
1835 * image pages.
1838 static unsigned int safe_highmem_pages;
1840 static struct memory_bitmap *safe_highmem_bm;
1842 static int
1843 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
1845 unsigned int to_alloc;
1847 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
1848 return -ENOMEM;
1850 if (get_highmem_buffer(PG_SAFE))
1851 return -ENOMEM;
1853 to_alloc = count_free_highmem_pages();
1854 if (to_alloc > *nr_highmem_p)
1855 to_alloc = *nr_highmem_p;
1856 else
1857 *nr_highmem_p = to_alloc;
1859 safe_highmem_pages = 0;
1860 while (to_alloc-- > 0) {
1861 struct page *page;
1863 page = alloc_page(__GFP_HIGHMEM);
1864 if (!swsusp_page_is_free(page)) {
1865 /* The page is "safe", set its bit the bitmap */
1866 memory_bm_set_bit(bm, page_to_pfn(page));
1867 safe_highmem_pages++;
1869 /* Mark the page as allocated */
1870 swsusp_set_page_forbidden(page);
1871 swsusp_set_page_free(page);
1873 memory_bm_position_reset(bm);
1874 safe_highmem_bm = bm;
1875 return 0;
1879 * get_highmem_page_buffer - for given highmem image page find the buffer
1880 * that suspend_write_next() should set for its caller to write to.
1882 * If the page is to be saved to its "original" page frame or a copy of
1883 * the page is to be made in the highmem, @buffer is returned. Otherwise,
1884 * the copy of the page is to be made in normal memory, so the address of
1885 * the copy is returned.
1887 * If @buffer is returned, the caller of suspend_write_next() will write
1888 * the page's contents to @buffer, so they will have to be copied to the
1889 * right location on the next call to suspend_write_next() and it is done
1890 * with the help of copy_last_highmem_page(). For this purpose, if
1891 * @buffer is returned, @last_highmem page is set to the page to which
1892 * the data will have to be copied from @buffer.
1895 static struct page *last_highmem_page;
1897 static void *
1898 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
1900 struct highmem_pbe *pbe;
1901 void *kaddr;
1903 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
1904 /* We have allocated the "original" page frame and we can
1905 * use it directly to store the loaded page.
1907 last_highmem_page = page;
1908 return buffer;
1910 /* The "original" page frame has not been allocated and we have to
1911 * use a "safe" page frame to store the loaded page.
1913 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
1914 if (!pbe) {
1915 swsusp_free();
1916 return ERR_PTR(-ENOMEM);
1918 pbe->orig_page = page;
1919 if (safe_highmem_pages > 0) {
1920 struct page *tmp;
1922 /* Copy of the page will be stored in high memory */
1923 kaddr = buffer;
1924 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
1925 safe_highmem_pages--;
1926 last_highmem_page = tmp;
1927 pbe->copy_page = tmp;
1928 } else {
1929 /* Copy of the page will be stored in normal memory */
1930 kaddr = safe_pages_list;
1931 safe_pages_list = safe_pages_list->next;
1932 pbe->copy_page = virt_to_page(kaddr);
1934 pbe->next = highmem_pblist;
1935 highmem_pblist = pbe;
1936 return kaddr;
1940 * copy_last_highmem_page - copy the contents of a highmem image from
1941 * @buffer, where the caller of snapshot_write_next() has place them,
1942 * to the right location represented by @last_highmem_page .
1945 static void copy_last_highmem_page(void)
1947 if (last_highmem_page) {
1948 void *dst;
1950 dst = kmap_atomic(last_highmem_page, KM_USER0);
1951 memcpy(dst, buffer, PAGE_SIZE);
1952 kunmap_atomic(dst, KM_USER0);
1953 last_highmem_page = NULL;
1957 static inline int last_highmem_page_copied(void)
1959 return !last_highmem_page;
1962 static inline void free_highmem_data(void)
1964 if (safe_highmem_bm)
1965 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
1967 if (buffer)
1968 free_image_page(buffer, PG_UNSAFE_CLEAR);
1970 #else
1971 static inline int get_safe_write_buffer(void) { return 0; }
1973 static unsigned int
1974 count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
1976 static inline int
1977 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
1979 return 0;
1982 static inline void *
1983 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
1985 return ERR_PTR(-EINVAL);
1988 static inline void copy_last_highmem_page(void) {}
1989 static inline int last_highmem_page_copied(void) { return 1; }
1990 static inline void free_highmem_data(void) {}
1991 #endif /* CONFIG_HIGHMEM */
1994 * prepare_image - use the memory bitmap @bm to mark the pages that will
1995 * be overwritten in the process of restoring the system memory state
1996 * from the suspend image ("unsafe" pages) and allocate memory for the
1997 * image.
1999 * The idea is to allocate a new memory bitmap first and then allocate
2000 * as many pages as needed for the image data, but not to assign these
2001 * pages to specific tasks initially. Instead, we just mark them as
2002 * allocated and create a lists of "safe" pages that will be used
2003 * later. On systems with high memory a list of "safe" highmem pages is
2004 * also created.
2007 #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2009 static int
2010 prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2012 unsigned int nr_pages, nr_highmem;
2013 struct linked_page *sp_list, *lp;
2014 int error;
2016 /* If there is no highmem, the buffer will not be necessary */
2017 free_image_page(buffer, PG_UNSAFE_CLEAR);
2018 buffer = NULL;
2020 nr_highmem = count_highmem_image_pages(bm);
2021 error = mark_unsafe_pages(bm);
2022 if (error)
2023 goto Free;
2025 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2026 if (error)
2027 goto Free;
2029 duplicate_memory_bitmap(new_bm, bm);
2030 memory_bm_free(bm, PG_UNSAFE_KEEP);
2031 if (nr_highmem > 0) {
2032 error = prepare_highmem_image(bm, &nr_highmem);
2033 if (error)
2034 goto Free;
2036 /* Reserve some safe pages for potential later use.
2038 * NOTE: This way we make sure there will be enough safe pages for the
2039 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2040 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2042 sp_list = NULL;
2043 /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */
2044 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2045 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2046 while (nr_pages > 0) {
2047 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2048 if (!lp) {
2049 error = -ENOMEM;
2050 goto Free;
2052 lp->next = sp_list;
2053 sp_list = lp;
2054 nr_pages--;
2056 /* Preallocate memory for the image */
2057 safe_pages_list = NULL;
2058 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2059 while (nr_pages > 0) {
2060 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2061 if (!lp) {
2062 error = -ENOMEM;
2063 goto Free;
2065 if (!swsusp_page_is_free(virt_to_page(lp))) {
2066 /* The page is "safe", add it to the list */
2067 lp->next = safe_pages_list;
2068 safe_pages_list = lp;
2070 /* Mark the page as allocated */
2071 swsusp_set_page_forbidden(virt_to_page(lp));
2072 swsusp_set_page_free(virt_to_page(lp));
2073 nr_pages--;
2075 /* Free the reserved safe pages so that chain_alloc() can use them */
2076 while (sp_list) {
2077 lp = sp_list->next;
2078 free_image_page(sp_list, PG_UNSAFE_CLEAR);
2079 sp_list = lp;
2081 return 0;
2083 Free:
2084 swsusp_free();
2085 return error;
2089 * get_buffer - compute the address that snapshot_write_next() should
2090 * set for its caller to write to.
2093 static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2095 struct pbe *pbe;
2096 struct page *page;
2097 unsigned long pfn = memory_bm_next_pfn(bm);
2099 if (pfn == BM_END_OF_MAP)
2100 return ERR_PTR(-EFAULT);
2102 page = pfn_to_page(pfn);
2103 if (PageHighMem(page))
2104 return get_highmem_page_buffer(page, ca);
2106 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2107 /* We have allocated the "original" page frame and we can
2108 * use it directly to store the loaded page.
2110 return page_address(page);
2112 /* The "original" page frame has not been allocated and we have to
2113 * use a "safe" page frame to store the loaded page.
2115 pbe = chain_alloc(ca, sizeof(struct pbe));
2116 if (!pbe) {
2117 swsusp_free();
2118 return ERR_PTR(-ENOMEM);
2120 pbe->orig_address = page_address(page);
2121 pbe->address = safe_pages_list;
2122 safe_pages_list = safe_pages_list->next;
2123 pbe->next = restore_pblist;
2124 restore_pblist = pbe;
2125 return pbe->address;
2129 * snapshot_write_next - used for writing the system memory snapshot.
2131 * On the first call to it @handle should point to a zeroed
2132 * snapshot_handle structure. The structure gets updated and a pointer
2133 * to it should be passed to this function every next time.
2135 * The @count parameter should contain the number of bytes the caller
2136 * wants to write to the image. It must not be zero.
2138 * On success the function returns a positive number. Then, the caller
2139 * is allowed to write up to the returned number of bytes to the memory
2140 * location computed by the data_of() macro. The number returned
2141 * may be smaller than @count, but this only happens if the write would
2142 * cross a page boundary otherwise.
2144 * The function returns 0 to indicate the "end of file" condition,
2145 * and a negative number is returned on error. In such cases the
2146 * structure pointed to by @handle is not updated and should not be used
2147 * any more.
2150 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
2152 static struct chain_allocator ca;
2153 int error = 0;
2155 /* Check if we have already loaded the entire image */
2156 if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages)
2157 return 0;
2159 if (handle->offset == 0) {
2160 if (!buffer)
2161 /* This makes the buffer be freed by swsusp_free() */
2162 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2164 if (!buffer)
2165 return -ENOMEM;
2167 handle->buffer = buffer;
2169 handle->sync_read = 1;
2170 if (handle->prev < handle->cur) {
2171 if (handle->prev == 0) {
2172 error = load_header(buffer);
2173 if (error)
2174 return error;
2176 error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
2177 if (error)
2178 return error;
2180 } else if (handle->prev <= nr_meta_pages) {
2181 error = unpack_orig_pfns(buffer, &copy_bm);
2182 if (error)
2183 return error;
2185 if (handle->prev == nr_meta_pages) {
2186 error = prepare_image(&orig_bm, &copy_bm);
2187 if (error)
2188 return error;
2190 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2191 memory_bm_position_reset(&orig_bm);
2192 restore_pblist = NULL;
2193 handle->buffer = get_buffer(&orig_bm, &ca);
2194 handle->sync_read = 0;
2195 if (IS_ERR(handle->buffer))
2196 return PTR_ERR(handle->buffer);
2198 } else {
2199 copy_last_highmem_page();
2200 handle->buffer = get_buffer(&orig_bm, &ca);
2201 if (IS_ERR(handle->buffer))
2202 return PTR_ERR(handle->buffer);
2203 if (handle->buffer != buffer)
2204 handle->sync_read = 0;
2206 handle->prev = handle->cur;
2208 handle->buf_offset = handle->cur_offset;
2209 if (handle->cur_offset + count >= PAGE_SIZE) {
2210 count = PAGE_SIZE - handle->cur_offset;
2211 handle->cur_offset = 0;
2212 handle->cur++;
2213 } else {
2214 handle->cur_offset += count;
2216 handle->offset += count;
2217 return count;
2221 * snapshot_write_finalize - must be called after the last call to
2222 * snapshot_write_next() in case the last page in the image happens
2223 * to be a highmem page and its contents should be stored in the
2224 * highmem. Additionally, it releases the memory that will not be
2225 * used any more.
2228 void snapshot_write_finalize(struct snapshot_handle *handle)
2230 copy_last_highmem_page();
2231 /* Free only if we have loaded the image entirely */
2232 if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages) {
2233 memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR);
2234 free_highmem_data();
2238 int snapshot_image_loaded(struct snapshot_handle *handle)
2240 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2241 handle->cur <= nr_meta_pages + nr_copy_pages);
2244 #ifdef CONFIG_HIGHMEM
2245 /* Assumes that @buf is ready and points to a "safe" page */
2246 static inline void
2247 swap_two_pages_data(struct page *p1, struct page *p2, void *buf)
2249 void *kaddr1, *kaddr2;
2251 kaddr1 = kmap_atomic(p1, KM_USER0);
2252 kaddr2 = kmap_atomic(p2, KM_USER1);
2253 memcpy(buf, kaddr1, PAGE_SIZE);
2254 memcpy(kaddr1, kaddr2, PAGE_SIZE);
2255 memcpy(kaddr2, buf, PAGE_SIZE);
2256 kunmap_atomic(kaddr1, KM_USER0);
2257 kunmap_atomic(kaddr2, KM_USER1);
2261 * restore_highmem - for each highmem page that was allocated before
2262 * the suspend and included in the suspend image, and also has been
2263 * allocated by the "resume" kernel swap its current (ie. "before
2264 * resume") contents with the previous (ie. "before suspend") one.
2266 * If the resume eventually fails, we can call this function once
2267 * again and restore the "before resume" highmem state.
2270 int restore_highmem(void)
2272 struct highmem_pbe *pbe = highmem_pblist;
2273 void *buf;
2275 if (!pbe)
2276 return 0;
2278 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2279 if (!buf)
2280 return -ENOMEM;
2282 while (pbe) {
2283 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2284 pbe = pbe->next;
2286 free_image_page(buf, PG_UNSAFE_CLEAR);
2287 return 0;
2289 #endif /* CONFIG_HIGHMEM */