powerpc/powernv: Report size of OPAL memcons log
[linux/fpc-iii.git] / kernel / power / snapshot.c
blob2d8e2b227db84598f1523fa8fdadca5ccd7e9f44
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@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>
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>
29 #include <linux/slab.h>
30 #include <linux/compiler.h>
31 #include <linux/ktime.h>
33 #include <linux/uaccess.h>
34 #include <asm/mmu_context.h>
35 #include <asm/pgtable.h>
36 #include <asm/tlbflush.h>
37 #include <asm/io.h>
39 #include "power.h"
41 #ifdef CONFIG_DEBUG_RODATA
42 static bool hibernate_restore_protection;
43 static bool hibernate_restore_protection_active;
45 void enable_restore_image_protection(void)
47 hibernate_restore_protection = true;
50 static inline void hibernate_restore_protection_begin(void)
52 hibernate_restore_protection_active = hibernate_restore_protection;
55 static inline void hibernate_restore_protection_end(void)
57 hibernate_restore_protection_active = false;
60 static inline void hibernate_restore_protect_page(void *page_address)
62 if (hibernate_restore_protection_active)
63 set_memory_ro((unsigned long)page_address, 1);
66 static inline void hibernate_restore_unprotect_page(void *page_address)
68 if (hibernate_restore_protection_active)
69 set_memory_rw((unsigned long)page_address, 1);
71 #else
72 static inline void hibernate_restore_protection_begin(void) {}
73 static inline void hibernate_restore_protection_end(void) {}
74 static inline void hibernate_restore_protect_page(void *page_address) {}
75 static inline void hibernate_restore_unprotect_page(void *page_address) {}
76 #endif /* CONFIG_DEBUG_RODATA */
78 static int swsusp_page_is_free(struct page *);
79 static void swsusp_set_page_forbidden(struct page *);
80 static void swsusp_unset_page_forbidden(struct page *);
83 * Number of bytes to reserve for memory allocations made by device drivers
84 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
85 * cause image creation to fail (tunable via /sys/power/reserved_size).
87 unsigned long reserved_size;
89 void __init hibernate_reserved_size_init(void)
91 reserved_size = SPARE_PAGES * PAGE_SIZE;
95 * Preferred image size in bytes (tunable via /sys/power/image_size).
96 * When it is set to N, swsusp will do its best to ensure the image
97 * size will not exceed N bytes, but if that is impossible, it will
98 * try to create the smallest image possible.
100 unsigned long image_size;
102 void __init hibernate_image_size_init(void)
104 image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE;
108 * List of PBEs needed for restoring the pages that were allocated before
109 * the suspend and included in the suspend image, but have also been
110 * allocated by the "resume" kernel, so their contents cannot be written
111 * directly to their "original" page frames.
113 struct pbe *restore_pblist;
115 /* struct linked_page is used to build chains of pages */
117 #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
119 struct linked_page {
120 struct linked_page *next;
121 char data[LINKED_PAGE_DATA_SIZE];
122 } __packed;
125 * List of "safe" pages (ie. pages that were not used by the image kernel
126 * before hibernation) that may be used as temporary storage for image kernel
127 * memory contents.
129 static struct linked_page *safe_pages_list;
131 /* Pointer to an auxiliary buffer (1 page) */
132 static void *buffer;
134 #define PG_ANY 0
135 #define PG_SAFE 1
136 #define PG_UNSAFE_CLEAR 1
137 #define PG_UNSAFE_KEEP 0
139 static unsigned int allocated_unsafe_pages;
142 * get_image_page - Allocate a page for a hibernation image.
143 * @gfp_mask: GFP mask for the allocation.
144 * @safe_needed: Get pages that were not used before hibernation (restore only)
146 * During image restoration, for storing the PBE list and the image data, we can
147 * only use memory pages that do not conflict with the pages used before
148 * hibernation. The "unsafe" pages have PageNosaveFree set and we count them
149 * using allocated_unsafe_pages.
151 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
152 * swsusp_free() can release it.
154 static void *get_image_page(gfp_t gfp_mask, int safe_needed)
156 void *res;
158 res = (void *)get_zeroed_page(gfp_mask);
159 if (safe_needed)
160 while (res && swsusp_page_is_free(virt_to_page(res))) {
161 /* The page is unsafe, mark it for swsusp_free() */
162 swsusp_set_page_forbidden(virt_to_page(res));
163 allocated_unsafe_pages++;
164 res = (void *)get_zeroed_page(gfp_mask);
166 if (res) {
167 swsusp_set_page_forbidden(virt_to_page(res));
168 swsusp_set_page_free(virt_to_page(res));
170 return res;
173 static void *__get_safe_page(gfp_t gfp_mask)
175 if (safe_pages_list) {
176 void *ret = safe_pages_list;
178 safe_pages_list = safe_pages_list->next;
179 memset(ret, 0, PAGE_SIZE);
180 return ret;
182 return get_image_page(gfp_mask, PG_SAFE);
185 unsigned long get_safe_page(gfp_t gfp_mask)
187 return (unsigned long)__get_safe_page(gfp_mask);
190 static struct page *alloc_image_page(gfp_t gfp_mask)
192 struct page *page;
194 page = alloc_page(gfp_mask);
195 if (page) {
196 swsusp_set_page_forbidden(page);
197 swsusp_set_page_free(page);
199 return page;
202 static void recycle_safe_page(void *page_address)
204 struct linked_page *lp = page_address;
206 lp->next = safe_pages_list;
207 safe_pages_list = lp;
211 * free_image_page - Free a page allocated for hibernation image.
212 * @addr: Address of the page to free.
213 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
215 * The page to free should have been allocated by get_image_page() (page flags
216 * set by it are affected).
218 static inline void free_image_page(void *addr, int clear_nosave_free)
220 struct page *page;
222 BUG_ON(!virt_addr_valid(addr));
224 page = virt_to_page(addr);
226 swsusp_unset_page_forbidden(page);
227 if (clear_nosave_free)
228 swsusp_unset_page_free(page);
230 __free_page(page);
233 static inline void free_list_of_pages(struct linked_page *list,
234 int clear_page_nosave)
236 while (list) {
237 struct linked_page *lp = list->next;
239 free_image_page(list, clear_page_nosave);
240 list = lp;
245 * struct chain_allocator is used for allocating small objects out of
246 * a linked list of pages called 'the chain'.
248 * The chain grows each time when there is no room for a new object in
249 * the current page. The allocated objects cannot be freed individually.
250 * It is only possible to free them all at once, by freeing the entire
251 * chain.
253 * NOTE: The chain allocator may be inefficient if the allocated objects
254 * are not much smaller than PAGE_SIZE.
256 struct chain_allocator {
257 struct linked_page *chain; /* the chain */
258 unsigned int used_space; /* total size of objects allocated out
259 of the current page */
260 gfp_t gfp_mask; /* mask for allocating pages */
261 int safe_needed; /* if set, only "safe" pages are allocated */
264 static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
265 int safe_needed)
267 ca->chain = NULL;
268 ca->used_space = LINKED_PAGE_DATA_SIZE;
269 ca->gfp_mask = gfp_mask;
270 ca->safe_needed = safe_needed;
273 static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
275 void *ret;
277 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
278 struct linked_page *lp;
280 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
281 get_image_page(ca->gfp_mask, PG_ANY);
282 if (!lp)
283 return NULL;
285 lp->next = ca->chain;
286 ca->chain = lp;
287 ca->used_space = 0;
289 ret = ca->chain->data + ca->used_space;
290 ca->used_space += size;
291 return ret;
295 * Data types related to memory bitmaps.
297 * Memory bitmap is a structure consiting of many linked lists of
298 * objects. The main list's elements are of type struct zone_bitmap
299 * and each of them corresonds to one zone. For each zone bitmap
300 * object there is a list of objects of type struct bm_block that
301 * represent each blocks of bitmap in which information is stored.
303 * struct memory_bitmap contains a pointer to the main list of zone
304 * bitmap objects, a struct bm_position used for browsing the bitmap,
305 * and a pointer to the list of pages used for allocating all of the
306 * zone bitmap objects and bitmap block objects.
308 * NOTE: It has to be possible to lay out the bitmap in memory
309 * using only allocations of order 0. Additionally, the bitmap is
310 * designed to work with arbitrary number of zones (this is over the
311 * top for now, but let's avoid making unnecessary assumptions ;-).
313 * struct zone_bitmap contains a pointer to a list of bitmap block
314 * objects and a pointer to the bitmap block object that has been
315 * most recently used for setting bits. Additionally, it contains the
316 * PFNs that correspond to the start and end of the represented zone.
318 * struct bm_block contains a pointer to the memory page in which
319 * information is stored (in the form of a block of bitmap)
320 * It also contains the pfns that correspond to the start and end of
321 * the represented memory area.
323 * The memory bitmap is organized as a radix tree to guarantee fast random
324 * access to the bits. There is one radix tree for each zone (as returned
325 * from create_mem_extents).
327 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
328 * two linked lists for the nodes of the tree, one for the inner nodes and
329 * one for the leave nodes. The linked leave nodes are used for fast linear
330 * access of the memory bitmap.
332 * The struct rtree_node represents one node of the radix tree.
335 #define BM_END_OF_MAP (~0UL)
337 #define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
338 #define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
339 #define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
342 * struct rtree_node is a wrapper struct to link the nodes
343 * of the rtree together for easy linear iteration over
344 * bits and easy freeing
346 struct rtree_node {
347 struct list_head list;
348 unsigned long *data;
352 * struct mem_zone_bm_rtree represents a bitmap used for one
353 * populated memory zone.
355 struct mem_zone_bm_rtree {
356 struct list_head list; /* Link Zones together */
357 struct list_head nodes; /* Radix Tree inner nodes */
358 struct list_head leaves; /* Radix Tree leaves */
359 unsigned long start_pfn; /* Zone start page frame */
360 unsigned long end_pfn; /* Zone end page frame + 1 */
361 struct rtree_node *rtree; /* Radix Tree Root */
362 int levels; /* Number of Radix Tree Levels */
363 unsigned int blocks; /* Number of Bitmap Blocks */
366 /* strcut bm_position is used for browsing memory bitmaps */
368 struct bm_position {
369 struct mem_zone_bm_rtree *zone;
370 struct rtree_node *node;
371 unsigned long node_pfn;
372 int node_bit;
375 struct memory_bitmap {
376 struct list_head zones;
377 struct linked_page *p_list; /* list of pages used to store zone
378 bitmap objects and bitmap block
379 objects */
380 struct bm_position cur; /* most recently used bit position */
383 /* Functions that operate on memory bitmaps */
385 #define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
386 #if BITS_PER_LONG == 32
387 #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
388 #else
389 #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
390 #endif
391 #define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
394 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
396 * This function is used to allocate inner nodes as well as the
397 * leave nodes of the radix tree. It also adds the node to the
398 * corresponding linked list passed in by the *list parameter.
400 static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
401 struct chain_allocator *ca,
402 struct list_head *list)
404 struct rtree_node *node;
406 node = chain_alloc(ca, sizeof(struct rtree_node));
407 if (!node)
408 return NULL;
410 node->data = get_image_page(gfp_mask, safe_needed);
411 if (!node->data)
412 return NULL;
414 list_add_tail(&node->list, list);
416 return node;
420 * add_rtree_block - Add a new leave node to the radix tree.
422 * The leave nodes need to be allocated in order to keep the leaves
423 * linked list in order. This is guaranteed by the zone->blocks
424 * counter.
426 static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
427 int safe_needed, struct chain_allocator *ca)
429 struct rtree_node *node, *block, **dst;
430 unsigned int levels_needed, block_nr;
431 int i;
433 block_nr = zone->blocks;
434 levels_needed = 0;
436 /* How many levels do we need for this block nr? */
437 while (block_nr) {
438 levels_needed += 1;
439 block_nr >>= BM_RTREE_LEVEL_SHIFT;
442 /* Make sure the rtree has enough levels */
443 for (i = zone->levels; i < levels_needed; i++) {
444 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
445 &zone->nodes);
446 if (!node)
447 return -ENOMEM;
449 node->data[0] = (unsigned long)zone->rtree;
450 zone->rtree = node;
451 zone->levels += 1;
454 /* Allocate new block */
455 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
456 if (!block)
457 return -ENOMEM;
459 /* Now walk the rtree to insert the block */
460 node = zone->rtree;
461 dst = &zone->rtree;
462 block_nr = zone->blocks;
463 for (i = zone->levels; i > 0; i--) {
464 int index;
466 if (!node) {
467 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
468 &zone->nodes);
469 if (!node)
470 return -ENOMEM;
471 *dst = node;
474 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
475 index &= BM_RTREE_LEVEL_MASK;
476 dst = (struct rtree_node **)&((*dst)->data[index]);
477 node = *dst;
480 zone->blocks += 1;
481 *dst = block;
483 return 0;
486 static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
487 int clear_nosave_free);
490 * create_zone_bm_rtree - Create a radix tree for one zone.
492 * Allocated the mem_zone_bm_rtree structure and initializes it.
493 * This function also allocated and builds the radix tree for the
494 * zone.
496 static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
497 int safe_needed,
498 struct chain_allocator *ca,
499 unsigned long start,
500 unsigned long end)
502 struct mem_zone_bm_rtree *zone;
503 unsigned int i, nr_blocks;
504 unsigned long pages;
506 pages = end - start;
507 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
508 if (!zone)
509 return NULL;
511 INIT_LIST_HEAD(&zone->nodes);
512 INIT_LIST_HEAD(&zone->leaves);
513 zone->start_pfn = start;
514 zone->end_pfn = end;
515 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
517 for (i = 0; i < nr_blocks; i++) {
518 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
519 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
520 return NULL;
524 return zone;
528 * free_zone_bm_rtree - Free the memory of the radix tree.
530 * Free all node pages of the radix tree. The mem_zone_bm_rtree
531 * structure itself is not freed here nor are the rtree_node
532 * structs.
534 static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
535 int clear_nosave_free)
537 struct rtree_node *node;
539 list_for_each_entry(node, &zone->nodes, list)
540 free_image_page(node->data, clear_nosave_free);
542 list_for_each_entry(node, &zone->leaves, list)
543 free_image_page(node->data, clear_nosave_free);
546 static void memory_bm_position_reset(struct memory_bitmap *bm)
548 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
549 list);
550 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
551 struct rtree_node, list);
552 bm->cur.node_pfn = 0;
553 bm->cur.node_bit = 0;
556 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
558 struct mem_extent {
559 struct list_head hook;
560 unsigned long start;
561 unsigned long end;
565 * free_mem_extents - Free a list of memory extents.
566 * @list: List of extents to free.
568 static void free_mem_extents(struct list_head *list)
570 struct mem_extent *ext, *aux;
572 list_for_each_entry_safe(ext, aux, list, hook) {
573 list_del(&ext->hook);
574 kfree(ext);
579 * create_mem_extents - Create a list of memory extents.
580 * @list: List to put the extents into.
581 * @gfp_mask: Mask to use for memory allocations.
583 * The extents represent contiguous ranges of PFNs.
585 static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
587 struct zone *zone;
589 INIT_LIST_HEAD(list);
591 for_each_populated_zone(zone) {
592 unsigned long zone_start, zone_end;
593 struct mem_extent *ext, *cur, *aux;
595 zone_start = zone->zone_start_pfn;
596 zone_end = zone_end_pfn(zone);
598 list_for_each_entry(ext, list, hook)
599 if (zone_start <= ext->end)
600 break;
602 if (&ext->hook == list || zone_end < ext->start) {
603 /* New extent is necessary */
604 struct mem_extent *new_ext;
606 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
607 if (!new_ext) {
608 free_mem_extents(list);
609 return -ENOMEM;
611 new_ext->start = zone_start;
612 new_ext->end = zone_end;
613 list_add_tail(&new_ext->hook, &ext->hook);
614 continue;
617 /* Merge this zone's range of PFNs with the existing one */
618 if (zone_start < ext->start)
619 ext->start = zone_start;
620 if (zone_end > ext->end)
621 ext->end = zone_end;
623 /* More merging may be possible */
624 cur = ext;
625 list_for_each_entry_safe_continue(cur, aux, list, hook) {
626 if (zone_end < cur->start)
627 break;
628 if (zone_end < cur->end)
629 ext->end = cur->end;
630 list_del(&cur->hook);
631 kfree(cur);
635 return 0;
639 * memory_bm_create - Allocate memory for a memory bitmap.
641 static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
642 int safe_needed)
644 struct chain_allocator ca;
645 struct list_head mem_extents;
646 struct mem_extent *ext;
647 int error;
649 chain_init(&ca, gfp_mask, safe_needed);
650 INIT_LIST_HEAD(&bm->zones);
652 error = create_mem_extents(&mem_extents, gfp_mask);
653 if (error)
654 return error;
656 list_for_each_entry(ext, &mem_extents, hook) {
657 struct mem_zone_bm_rtree *zone;
659 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
660 ext->start, ext->end);
661 if (!zone) {
662 error = -ENOMEM;
663 goto Error;
665 list_add_tail(&zone->list, &bm->zones);
668 bm->p_list = ca.chain;
669 memory_bm_position_reset(bm);
670 Exit:
671 free_mem_extents(&mem_extents);
672 return error;
674 Error:
675 bm->p_list = ca.chain;
676 memory_bm_free(bm, PG_UNSAFE_CLEAR);
677 goto Exit;
681 * memory_bm_free - Free memory occupied by the memory bitmap.
682 * @bm: Memory bitmap.
684 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
686 struct mem_zone_bm_rtree *zone;
688 list_for_each_entry(zone, &bm->zones, list)
689 free_zone_bm_rtree(zone, clear_nosave_free);
691 free_list_of_pages(bm->p_list, clear_nosave_free);
693 INIT_LIST_HEAD(&bm->zones);
697 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
699 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
700 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
702 * Walk the radix tree to find the page containing the bit that represents @pfn
703 * and return the position of the bit in @addr and @bit_nr.
705 static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
706 void **addr, unsigned int *bit_nr)
708 struct mem_zone_bm_rtree *curr, *zone;
709 struct rtree_node *node;
710 int i, block_nr;
712 zone = bm->cur.zone;
714 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
715 goto zone_found;
717 zone = NULL;
719 /* Find the right zone */
720 list_for_each_entry(curr, &bm->zones, list) {
721 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
722 zone = curr;
723 break;
727 if (!zone)
728 return -EFAULT;
730 zone_found:
732 * We have found the zone. Now walk the radix tree to find the leaf node
733 * for our PFN.
735 node = bm->cur.node;
736 if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
737 goto node_found;
739 node = zone->rtree;
740 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
742 for (i = zone->levels; i > 0; i--) {
743 int index;
745 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
746 index &= BM_RTREE_LEVEL_MASK;
747 BUG_ON(node->data[index] == 0);
748 node = (struct rtree_node *)node->data[index];
751 node_found:
752 /* Update last position */
753 bm->cur.zone = zone;
754 bm->cur.node = node;
755 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
757 /* Set return values */
758 *addr = node->data;
759 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
761 return 0;
764 static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
766 void *addr;
767 unsigned int bit;
768 int error;
770 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
771 BUG_ON(error);
772 set_bit(bit, addr);
775 static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
777 void *addr;
778 unsigned int bit;
779 int error;
781 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
782 if (!error)
783 set_bit(bit, addr);
785 return error;
788 static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
790 void *addr;
791 unsigned int bit;
792 int error;
794 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
795 BUG_ON(error);
796 clear_bit(bit, addr);
799 static void memory_bm_clear_current(struct memory_bitmap *bm)
801 int bit;
803 bit = max(bm->cur.node_bit - 1, 0);
804 clear_bit(bit, bm->cur.node->data);
807 static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
809 void *addr;
810 unsigned int bit;
811 int error;
813 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
814 BUG_ON(error);
815 return test_bit(bit, addr);
818 static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
820 void *addr;
821 unsigned int bit;
823 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
827 * rtree_next_node - Jump to the next leaf node.
829 * Set the position to the beginning of the next node in the
830 * memory bitmap. This is either the next node in the current
831 * zone's radix tree or the first node in the radix tree of the
832 * next zone.
834 * Return true if there is a next node, false otherwise.
836 static bool rtree_next_node(struct memory_bitmap *bm)
838 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
839 bm->cur.node = list_entry(bm->cur.node->list.next,
840 struct rtree_node, list);
841 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
842 bm->cur.node_bit = 0;
843 touch_softlockup_watchdog();
844 return true;
847 /* No more nodes, goto next zone */
848 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
849 bm->cur.zone = list_entry(bm->cur.zone->list.next,
850 struct mem_zone_bm_rtree, list);
851 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
852 struct rtree_node, list);
853 bm->cur.node_pfn = 0;
854 bm->cur.node_bit = 0;
855 return true;
858 /* No more zones */
859 return false;
863 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
864 * @bm: Memory bitmap.
866 * Starting from the last returned position this function searches for the next
867 * set bit in @bm and returns the PFN represented by it. If no more bits are
868 * set, BM_END_OF_MAP is returned.
870 * It is required to run memory_bm_position_reset() before the first call to
871 * this function for the given memory bitmap.
873 static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
875 unsigned long bits, pfn, pages;
876 int bit;
878 do {
879 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
880 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
881 bit = find_next_bit(bm->cur.node->data, bits,
882 bm->cur.node_bit);
883 if (bit < bits) {
884 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
885 bm->cur.node_bit = bit + 1;
886 return pfn;
888 } while (rtree_next_node(bm));
890 return BM_END_OF_MAP;
894 * This structure represents a range of page frames the contents of which
895 * should not be saved during hibernation.
897 struct nosave_region {
898 struct list_head list;
899 unsigned long start_pfn;
900 unsigned long end_pfn;
903 static LIST_HEAD(nosave_regions);
905 static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
907 struct rtree_node *node;
909 list_for_each_entry(node, &zone->nodes, list)
910 recycle_safe_page(node->data);
912 list_for_each_entry(node, &zone->leaves, list)
913 recycle_safe_page(node->data);
916 static void memory_bm_recycle(struct memory_bitmap *bm)
918 struct mem_zone_bm_rtree *zone;
919 struct linked_page *p_list;
921 list_for_each_entry(zone, &bm->zones, list)
922 recycle_zone_bm_rtree(zone);
924 p_list = bm->p_list;
925 while (p_list) {
926 struct linked_page *lp = p_list;
928 p_list = lp->next;
929 recycle_safe_page(lp);
934 * register_nosave_region - Register a region of unsaveable memory.
936 * Register a range of page frames the contents of which should not be saved
937 * during hibernation (to be used in the early initialization code).
939 void __init __register_nosave_region(unsigned long start_pfn,
940 unsigned long end_pfn, int use_kmalloc)
942 struct nosave_region *region;
944 if (start_pfn >= end_pfn)
945 return;
947 if (!list_empty(&nosave_regions)) {
948 /* Try to extend the previous region (they should be sorted) */
949 region = list_entry(nosave_regions.prev,
950 struct nosave_region, list);
951 if (region->end_pfn == start_pfn) {
952 region->end_pfn = end_pfn;
953 goto Report;
956 if (use_kmalloc) {
957 /* During init, this shouldn't fail */
958 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
959 BUG_ON(!region);
960 } else {
961 /* This allocation cannot fail */
962 region = memblock_virt_alloc(sizeof(struct nosave_region), 0);
964 region->start_pfn = start_pfn;
965 region->end_pfn = end_pfn;
966 list_add_tail(&region->list, &nosave_regions);
967 Report:
968 printk(KERN_INFO "PM: Registered nosave memory: [mem %#010llx-%#010llx]\n",
969 (unsigned long long) start_pfn << PAGE_SHIFT,
970 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
974 * Set bits in this map correspond to the page frames the contents of which
975 * should not be saved during the suspend.
977 static struct memory_bitmap *forbidden_pages_map;
979 /* Set bits in this map correspond to free page frames. */
980 static struct memory_bitmap *free_pages_map;
983 * Each page frame allocated for creating the image is marked by setting the
984 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
987 void swsusp_set_page_free(struct page *page)
989 if (free_pages_map)
990 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
993 static int swsusp_page_is_free(struct page *page)
995 return free_pages_map ?
996 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
999 void swsusp_unset_page_free(struct page *page)
1001 if (free_pages_map)
1002 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1005 static void swsusp_set_page_forbidden(struct page *page)
1007 if (forbidden_pages_map)
1008 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1011 int swsusp_page_is_forbidden(struct page *page)
1013 return forbidden_pages_map ?
1014 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1017 static void swsusp_unset_page_forbidden(struct page *page)
1019 if (forbidden_pages_map)
1020 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1024 * mark_nosave_pages - Mark pages that should not be saved.
1025 * @bm: Memory bitmap.
1027 * Set the bits in @bm that correspond to the page frames the contents of which
1028 * should not be saved.
1030 static void mark_nosave_pages(struct memory_bitmap *bm)
1032 struct nosave_region *region;
1034 if (list_empty(&nosave_regions))
1035 return;
1037 list_for_each_entry(region, &nosave_regions, list) {
1038 unsigned long pfn;
1040 pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n",
1041 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1042 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1043 - 1);
1045 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1046 if (pfn_valid(pfn)) {
1048 * It is safe to ignore the result of
1049 * mem_bm_set_bit_check() here, since we won't
1050 * touch the PFNs for which the error is
1051 * returned anyway.
1053 mem_bm_set_bit_check(bm, pfn);
1059 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1061 * Create bitmaps needed for marking page frames that should not be saved and
1062 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1063 * only modified if everything goes well, because we don't want the bits to be
1064 * touched before both bitmaps are set up.
1066 int create_basic_memory_bitmaps(void)
1068 struct memory_bitmap *bm1, *bm2;
1069 int error = 0;
1071 if (forbidden_pages_map && free_pages_map)
1072 return 0;
1073 else
1074 BUG_ON(forbidden_pages_map || free_pages_map);
1076 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1077 if (!bm1)
1078 return -ENOMEM;
1080 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1081 if (error)
1082 goto Free_first_object;
1084 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1085 if (!bm2)
1086 goto Free_first_bitmap;
1088 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1089 if (error)
1090 goto Free_second_object;
1092 forbidden_pages_map = bm1;
1093 free_pages_map = bm2;
1094 mark_nosave_pages(forbidden_pages_map);
1096 pr_debug("PM: Basic memory bitmaps created\n");
1098 return 0;
1100 Free_second_object:
1101 kfree(bm2);
1102 Free_first_bitmap:
1103 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1104 Free_first_object:
1105 kfree(bm1);
1106 return -ENOMEM;
1110 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1112 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1113 * auxiliary pointers are necessary so that the bitmaps themselves are not
1114 * referred to while they are being freed.
1116 void free_basic_memory_bitmaps(void)
1118 struct memory_bitmap *bm1, *bm2;
1120 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1121 return;
1123 bm1 = forbidden_pages_map;
1124 bm2 = free_pages_map;
1125 forbidden_pages_map = NULL;
1126 free_pages_map = NULL;
1127 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1128 kfree(bm1);
1129 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1130 kfree(bm2);
1132 pr_debug("PM: Basic memory bitmaps freed\n");
1135 void clear_free_pages(void)
1137 #ifdef CONFIG_PAGE_POISONING_ZERO
1138 struct memory_bitmap *bm = free_pages_map;
1139 unsigned long pfn;
1141 if (WARN_ON(!(free_pages_map)))
1142 return;
1144 memory_bm_position_reset(bm);
1145 pfn = memory_bm_next_pfn(bm);
1146 while (pfn != BM_END_OF_MAP) {
1147 if (pfn_valid(pfn))
1148 clear_highpage(pfn_to_page(pfn));
1150 pfn = memory_bm_next_pfn(bm);
1152 memory_bm_position_reset(bm);
1153 pr_info("PM: free pages cleared after restore\n");
1154 #endif /* PAGE_POISONING_ZERO */
1158 * snapshot_additional_pages - Estimate the number of extra pages needed.
1159 * @zone: Memory zone to carry out the computation for.
1161 * Estimate the number of additional pages needed for setting up a hibernation
1162 * image data structures for @zone (usually, the returned value is greater than
1163 * the exact number).
1165 unsigned int snapshot_additional_pages(struct zone *zone)
1167 unsigned int rtree, nodes;
1169 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1170 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1171 LINKED_PAGE_DATA_SIZE);
1172 while (nodes > 1) {
1173 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1174 rtree += nodes;
1177 return 2 * rtree;
1180 #ifdef CONFIG_HIGHMEM
1182 * count_free_highmem_pages - Compute the total number of free highmem pages.
1184 * The returned number is system-wide.
1186 static unsigned int count_free_highmem_pages(void)
1188 struct zone *zone;
1189 unsigned int cnt = 0;
1191 for_each_populated_zone(zone)
1192 if (is_highmem(zone))
1193 cnt += zone_page_state(zone, NR_FREE_PAGES);
1195 return cnt;
1199 * saveable_highmem_page - Check if a highmem page is saveable.
1201 * Determine whether a highmem page should be included in a hibernation image.
1203 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1204 * and it isn't part of a free chunk of pages.
1206 static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1208 struct page *page;
1210 if (!pfn_valid(pfn))
1211 return NULL;
1213 page = pfn_to_page(pfn);
1214 if (page_zone(page) != zone)
1215 return NULL;
1217 BUG_ON(!PageHighMem(page));
1219 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) ||
1220 PageReserved(page))
1221 return NULL;
1223 if (page_is_guard(page))
1224 return NULL;
1226 return page;
1230 * count_highmem_pages - Compute the total number of saveable highmem pages.
1232 static unsigned int count_highmem_pages(void)
1234 struct zone *zone;
1235 unsigned int n = 0;
1237 for_each_populated_zone(zone) {
1238 unsigned long pfn, max_zone_pfn;
1240 if (!is_highmem(zone))
1241 continue;
1243 mark_free_pages(zone);
1244 max_zone_pfn = zone_end_pfn(zone);
1245 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1246 if (saveable_highmem_page(zone, pfn))
1247 n++;
1249 return n;
1251 #else
1252 static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1254 return NULL;
1256 #endif /* CONFIG_HIGHMEM */
1259 * saveable_page - Check if the given page is saveable.
1261 * Determine whether a non-highmem page should be included in a hibernation
1262 * image.
1264 * We should save the page if it isn't Nosave, and is not in the range
1265 * of pages statically defined as 'unsaveable', and it isn't part of
1266 * a free chunk of pages.
1268 static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1270 struct page *page;
1272 if (!pfn_valid(pfn))
1273 return NULL;
1275 page = pfn_to_page(pfn);
1276 if (page_zone(page) != zone)
1277 return NULL;
1279 BUG_ON(PageHighMem(page));
1281 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1282 return NULL;
1284 if (PageReserved(page)
1285 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1286 return NULL;
1288 if (page_is_guard(page))
1289 return NULL;
1291 return page;
1295 * count_data_pages - Compute the total number of saveable non-highmem pages.
1297 static unsigned int count_data_pages(void)
1299 struct zone *zone;
1300 unsigned long pfn, max_zone_pfn;
1301 unsigned int n = 0;
1303 for_each_populated_zone(zone) {
1304 if (is_highmem(zone))
1305 continue;
1307 mark_free_pages(zone);
1308 max_zone_pfn = zone_end_pfn(zone);
1309 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1310 if (saveable_page(zone, pfn))
1311 n++;
1313 return n;
1317 * This is needed, because copy_page and memcpy are not usable for copying
1318 * task structs.
1320 static inline void do_copy_page(long *dst, long *src)
1322 int n;
1324 for (n = PAGE_SIZE / sizeof(long); n; n--)
1325 *dst++ = *src++;
1329 * safe_copy_page - Copy a page in a safe way.
1331 * Check if the page we are going to copy is marked as present in the kernel
1332 * page tables (this always is the case if CONFIG_DEBUG_PAGEALLOC is not set
1333 * and in that case kernel_page_present() always returns 'true').
1335 static void safe_copy_page(void *dst, struct page *s_page)
1337 if (kernel_page_present(s_page)) {
1338 do_copy_page(dst, page_address(s_page));
1339 } else {
1340 kernel_map_pages(s_page, 1, 1);
1341 do_copy_page(dst, page_address(s_page));
1342 kernel_map_pages(s_page, 1, 0);
1346 #ifdef CONFIG_HIGHMEM
1347 static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1349 return is_highmem(zone) ?
1350 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1353 static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1355 struct page *s_page, *d_page;
1356 void *src, *dst;
1358 s_page = pfn_to_page(src_pfn);
1359 d_page = pfn_to_page(dst_pfn);
1360 if (PageHighMem(s_page)) {
1361 src = kmap_atomic(s_page);
1362 dst = kmap_atomic(d_page);
1363 do_copy_page(dst, src);
1364 kunmap_atomic(dst);
1365 kunmap_atomic(src);
1366 } else {
1367 if (PageHighMem(d_page)) {
1369 * The page pointed to by src may contain some kernel
1370 * data modified by kmap_atomic()
1372 safe_copy_page(buffer, s_page);
1373 dst = kmap_atomic(d_page);
1374 copy_page(dst, buffer);
1375 kunmap_atomic(dst);
1376 } else {
1377 safe_copy_page(page_address(d_page), s_page);
1381 #else
1382 #define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1384 static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1386 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1387 pfn_to_page(src_pfn));
1389 #endif /* CONFIG_HIGHMEM */
1391 static void copy_data_pages(struct memory_bitmap *copy_bm,
1392 struct memory_bitmap *orig_bm)
1394 struct zone *zone;
1395 unsigned long pfn;
1397 for_each_populated_zone(zone) {
1398 unsigned long max_zone_pfn;
1400 mark_free_pages(zone);
1401 max_zone_pfn = zone_end_pfn(zone);
1402 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1403 if (page_is_saveable(zone, pfn))
1404 memory_bm_set_bit(orig_bm, pfn);
1406 memory_bm_position_reset(orig_bm);
1407 memory_bm_position_reset(copy_bm);
1408 for(;;) {
1409 pfn = memory_bm_next_pfn(orig_bm);
1410 if (unlikely(pfn == BM_END_OF_MAP))
1411 break;
1412 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1416 /* Total number of image pages */
1417 static unsigned int nr_copy_pages;
1418 /* Number of pages needed for saving the original pfns of the image pages */
1419 static unsigned int nr_meta_pages;
1421 * Numbers of normal and highmem page frames allocated for hibernation image
1422 * before suspending devices.
1424 unsigned int alloc_normal, alloc_highmem;
1426 * Memory bitmap used for marking saveable pages (during hibernation) or
1427 * hibernation image pages (during restore)
1429 static struct memory_bitmap orig_bm;
1431 * Memory bitmap used during hibernation for marking allocated page frames that
1432 * will contain copies of saveable pages. During restore it is initially used
1433 * for marking hibernation image pages, but then the set bits from it are
1434 * duplicated in @orig_bm and it is released. On highmem systems it is next
1435 * used for marking "safe" highmem pages, but it has to be reinitialized for
1436 * this purpose.
1438 static struct memory_bitmap copy_bm;
1441 * swsusp_free - Free pages allocated for hibernation image.
1443 * Image pages are alocated before snapshot creation, so they need to be
1444 * released after resume.
1446 void swsusp_free(void)
1448 unsigned long fb_pfn, fr_pfn;
1450 if (!forbidden_pages_map || !free_pages_map)
1451 goto out;
1453 memory_bm_position_reset(forbidden_pages_map);
1454 memory_bm_position_reset(free_pages_map);
1456 loop:
1457 fr_pfn = memory_bm_next_pfn(free_pages_map);
1458 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1461 * Find the next bit set in both bitmaps. This is guaranteed to
1462 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1464 do {
1465 if (fb_pfn < fr_pfn)
1466 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1467 if (fr_pfn < fb_pfn)
1468 fr_pfn = memory_bm_next_pfn(free_pages_map);
1469 } while (fb_pfn != fr_pfn);
1471 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1472 struct page *page = pfn_to_page(fr_pfn);
1474 memory_bm_clear_current(forbidden_pages_map);
1475 memory_bm_clear_current(free_pages_map);
1476 hibernate_restore_unprotect_page(page_address(page));
1477 __free_page(page);
1478 goto loop;
1481 out:
1482 nr_copy_pages = 0;
1483 nr_meta_pages = 0;
1484 restore_pblist = NULL;
1485 buffer = NULL;
1486 alloc_normal = 0;
1487 alloc_highmem = 0;
1488 hibernate_restore_protection_end();
1491 /* Helper functions used for the shrinking of memory. */
1493 #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1496 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1497 * @nr_pages: Number of page frames to allocate.
1498 * @mask: GFP flags to use for the allocation.
1500 * Return value: Number of page frames actually allocated
1502 static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1504 unsigned long nr_alloc = 0;
1506 while (nr_pages > 0) {
1507 struct page *page;
1509 page = alloc_image_page(mask);
1510 if (!page)
1511 break;
1512 memory_bm_set_bit(&copy_bm, page_to_pfn(page));
1513 if (PageHighMem(page))
1514 alloc_highmem++;
1515 else
1516 alloc_normal++;
1517 nr_pages--;
1518 nr_alloc++;
1521 return nr_alloc;
1524 static unsigned long preallocate_image_memory(unsigned long nr_pages,
1525 unsigned long avail_normal)
1527 unsigned long alloc;
1529 if (avail_normal <= alloc_normal)
1530 return 0;
1532 alloc = avail_normal - alloc_normal;
1533 if (nr_pages < alloc)
1534 alloc = nr_pages;
1536 return preallocate_image_pages(alloc, GFP_IMAGE);
1539 #ifdef CONFIG_HIGHMEM
1540 static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1542 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1546 * __fraction - Compute (an approximation of) x * (multiplier / base).
1548 static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1550 x *= multiplier;
1551 do_div(x, base);
1552 return (unsigned long)x;
1555 static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1556 unsigned long highmem,
1557 unsigned long total)
1559 unsigned long alloc = __fraction(nr_pages, highmem, total);
1561 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1563 #else /* CONFIG_HIGHMEM */
1564 static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1566 return 0;
1569 static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1570 unsigned long highmem,
1571 unsigned long total)
1573 return 0;
1575 #endif /* CONFIG_HIGHMEM */
1578 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1580 static unsigned long free_unnecessary_pages(void)
1582 unsigned long save, to_free_normal, to_free_highmem, free;
1584 save = count_data_pages();
1585 if (alloc_normal >= save) {
1586 to_free_normal = alloc_normal - save;
1587 save = 0;
1588 } else {
1589 to_free_normal = 0;
1590 save -= alloc_normal;
1592 save += count_highmem_pages();
1593 if (alloc_highmem >= save) {
1594 to_free_highmem = alloc_highmem - save;
1595 } else {
1596 to_free_highmem = 0;
1597 save -= alloc_highmem;
1598 if (to_free_normal > save)
1599 to_free_normal -= save;
1600 else
1601 to_free_normal = 0;
1603 free = to_free_normal + to_free_highmem;
1605 memory_bm_position_reset(&copy_bm);
1607 while (to_free_normal > 0 || to_free_highmem > 0) {
1608 unsigned long pfn = memory_bm_next_pfn(&copy_bm);
1609 struct page *page = pfn_to_page(pfn);
1611 if (PageHighMem(page)) {
1612 if (!to_free_highmem)
1613 continue;
1614 to_free_highmem--;
1615 alloc_highmem--;
1616 } else {
1617 if (!to_free_normal)
1618 continue;
1619 to_free_normal--;
1620 alloc_normal--;
1622 memory_bm_clear_bit(&copy_bm, pfn);
1623 swsusp_unset_page_forbidden(page);
1624 swsusp_unset_page_free(page);
1625 __free_page(page);
1628 return free;
1632 * minimum_image_size - Estimate the minimum acceptable size of an image.
1633 * @saveable: Number of saveable pages in the system.
1635 * We want to avoid attempting to free too much memory too hard, so estimate the
1636 * minimum acceptable size of a hibernation image to use as the lower limit for
1637 * preallocating memory.
1639 * We assume that the minimum image size should be proportional to
1641 * [number of saveable pages] - [number of pages that can be freed in theory]
1643 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1644 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages,
1645 * minus mapped file pages.
1647 static unsigned long minimum_image_size(unsigned long saveable)
1649 unsigned long size;
1651 size = global_page_state(NR_SLAB_RECLAIMABLE)
1652 + global_node_page_state(NR_ACTIVE_ANON)
1653 + global_node_page_state(NR_INACTIVE_ANON)
1654 + global_node_page_state(NR_ACTIVE_FILE)
1655 + global_node_page_state(NR_INACTIVE_FILE)
1656 - global_node_page_state(NR_FILE_MAPPED);
1658 return saveable <= size ? 0 : saveable - size;
1662 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1664 * To create a hibernation image it is necessary to make a copy of every page
1665 * frame in use. We also need a number of page frames to be free during
1666 * hibernation for allocations made while saving the image and for device
1667 * drivers, in case they need to allocate memory from their hibernation
1668 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1669 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1670 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1671 * total number of available page frames and allocate at least
1673 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1674 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1676 * of them, which corresponds to the maximum size of a hibernation image.
1678 * If image_size is set below the number following from the above formula,
1679 * the preallocation of memory is continued until the total number of saveable
1680 * pages in the system is below the requested image size or the minimum
1681 * acceptable image size returned by minimum_image_size(), whichever is greater.
1683 int hibernate_preallocate_memory(void)
1685 struct zone *zone;
1686 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1687 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1688 ktime_t start, stop;
1689 int error;
1691 printk(KERN_INFO "PM: Preallocating image memory... ");
1692 start = ktime_get();
1694 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1695 if (error)
1696 goto err_out;
1698 error = memory_bm_create(&copy_bm, GFP_IMAGE, PG_ANY);
1699 if (error)
1700 goto err_out;
1702 alloc_normal = 0;
1703 alloc_highmem = 0;
1705 /* Count the number of saveable data pages. */
1706 save_highmem = count_highmem_pages();
1707 saveable = count_data_pages();
1710 * Compute the total number of page frames we can use (count) and the
1711 * number of pages needed for image metadata (size).
1713 count = saveable;
1714 saveable += save_highmem;
1715 highmem = save_highmem;
1716 size = 0;
1717 for_each_populated_zone(zone) {
1718 size += snapshot_additional_pages(zone);
1719 if (is_highmem(zone))
1720 highmem += zone_page_state(zone, NR_FREE_PAGES);
1721 else
1722 count += zone_page_state(zone, NR_FREE_PAGES);
1724 avail_normal = count;
1725 count += highmem;
1726 count -= totalreserve_pages;
1728 /* Add number of pages required for page keys (s390 only). */
1729 size += page_key_additional_pages(saveable);
1731 /* Compute the maximum number of saveable pages to leave in memory. */
1732 max_size = (count - (size + PAGES_FOR_IO)) / 2
1733 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1734 /* Compute the desired number of image pages specified by image_size. */
1735 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1736 if (size > max_size)
1737 size = max_size;
1739 * If the desired number of image pages is at least as large as the
1740 * current number of saveable pages in memory, allocate page frames for
1741 * the image and we're done.
1743 if (size >= saveable) {
1744 pages = preallocate_image_highmem(save_highmem);
1745 pages += preallocate_image_memory(saveable - pages, avail_normal);
1746 goto out;
1749 /* Estimate the minimum size of the image. */
1750 pages = minimum_image_size(saveable);
1752 * To avoid excessive pressure on the normal zone, leave room in it to
1753 * accommodate an image of the minimum size (unless it's already too
1754 * small, in which case don't preallocate pages from it at all).
1756 if (avail_normal > pages)
1757 avail_normal -= pages;
1758 else
1759 avail_normal = 0;
1760 if (size < pages)
1761 size = min_t(unsigned long, pages, max_size);
1764 * Let the memory management subsystem know that we're going to need a
1765 * large number of page frames to allocate and make it free some memory.
1766 * NOTE: If this is not done, performance will be hurt badly in some
1767 * test cases.
1769 shrink_all_memory(saveable - size);
1772 * The number of saveable pages in memory was too high, so apply some
1773 * pressure to decrease it. First, make room for the largest possible
1774 * image and fail if that doesn't work. Next, try to decrease the size
1775 * of the image as much as indicated by 'size' using allocations from
1776 * highmem and non-highmem zones separately.
1778 pages_highmem = preallocate_image_highmem(highmem / 2);
1779 alloc = count - max_size;
1780 if (alloc > pages_highmem)
1781 alloc -= pages_highmem;
1782 else
1783 alloc = 0;
1784 pages = preallocate_image_memory(alloc, avail_normal);
1785 if (pages < alloc) {
1786 /* We have exhausted non-highmem pages, try highmem. */
1787 alloc -= pages;
1788 pages += pages_highmem;
1789 pages_highmem = preallocate_image_highmem(alloc);
1790 if (pages_highmem < alloc)
1791 goto err_out;
1792 pages += pages_highmem;
1794 * size is the desired number of saveable pages to leave in
1795 * memory, so try to preallocate (all memory - size) pages.
1797 alloc = (count - pages) - size;
1798 pages += preallocate_image_highmem(alloc);
1799 } else {
1801 * There are approximately max_size saveable pages at this point
1802 * and we want to reduce this number down to size.
1804 alloc = max_size - size;
1805 size = preallocate_highmem_fraction(alloc, highmem, count);
1806 pages_highmem += size;
1807 alloc -= size;
1808 size = preallocate_image_memory(alloc, avail_normal);
1809 pages_highmem += preallocate_image_highmem(alloc - size);
1810 pages += pages_highmem + size;
1814 * We only need as many page frames for the image as there are saveable
1815 * pages in memory, but we have allocated more. Release the excessive
1816 * ones now.
1818 pages -= free_unnecessary_pages();
1820 out:
1821 stop = ktime_get();
1822 printk(KERN_CONT "done (allocated %lu pages)\n", pages);
1823 swsusp_show_speed(start, stop, pages, "Allocated");
1825 return 0;
1827 err_out:
1828 printk(KERN_CONT "\n");
1829 swsusp_free();
1830 return -ENOMEM;
1833 #ifdef CONFIG_HIGHMEM
1835 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1837 * Compute the number of non-highmem pages that will be necessary for creating
1838 * copies of highmem pages.
1840 static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1842 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1844 if (free_highmem >= nr_highmem)
1845 nr_highmem = 0;
1846 else
1847 nr_highmem -= free_highmem;
1849 return nr_highmem;
1851 #else
1852 static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1853 #endif /* CONFIG_HIGHMEM */
1856 * enough_free_mem - Check if there is enough free memory for the image.
1858 static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1860 struct zone *zone;
1861 unsigned int free = alloc_normal;
1863 for_each_populated_zone(zone)
1864 if (!is_highmem(zone))
1865 free += zone_page_state(zone, NR_FREE_PAGES);
1867 nr_pages += count_pages_for_highmem(nr_highmem);
1868 pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n",
1869 nr_pages, PAGES_FOR_IO, free);
1871 return free > nr_pages + PAGES_FOR_IO;
1874 #ifdef CONFIG_HIGHMEM
1876 * get_highmem_buffer - Allocate a buffer for highmem pages.
1878 * If there are some highmem pages in the hibernation image, we may need a
1879 * buffer to copy them and/or load their data.
1881 static inline int get_highmem_buffer(int safe_needed)
1883 buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
1884 return buffer ? 0 : -ENOMEM;
1888 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1890 * Try to allocate as many pages as needed, but if the number of free highmem
1891 * pages is less than that, allocate them all.
1893 static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1894 unsigned int nr_highmem)
1896 unsigned int to_alloc = count_free_highmem_pages();
1898 if (to_alloc > nr_highmem)
1899 to_alloc = nr_highmem;
1901 nr_highmem -= to_alloc;
1902 while (to_alloc-- > 0) {
1903 struct page *page;
1905 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1906 memory_bm_set_bit(bm, page_to_pfn(page));
1908 return nr_highmem;
1910 #else
1911 static inline int get_highmem_buffer(int safe_needed) { return 0; }
1913 static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1914 unsigned int n) { return 0; }
1915 #endif /* CONFIG_HIGHMEM */
1918 * swsusp_alloc - Allocate memory for hibernation image.
1920 * We first try to allocate as many highmem pages as there are
1921 * saveable highmem pages in the system. If that fails, we allocate
1922 * non-highmem pages for the copies of the remaining highmem ones.
1924 * In this approach it is likely that the copies of highmem pages will
1925 * also be located in the high memory, because of the way in which
1926 * copy_data_pages() works.
1928 static int swsusp_alloc(struct memory_bitmap *orig_bm,
1929 struct memory_bitmap *copy_bm,
1930 unsigned int nr_pages, unsigned int nr_highmem)
1932 if (nr_highmem > 0) {
1933 if (get_highmem_buffer(PG_ANY))
1934 goto err_out;
1935 if (nr_highmem > alloc_highmem) {
1936 nr_highmem -= alloc_highmem;
1937 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1940 if (nr_pages > alloc_normal) {
1941 nr_pages -= alloc_normal;
1942 while (nr_pages-- > 0) {
1943 struct page *page;
1945 page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
1946 if (!page)
1947 goto err_out;
1948 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1952 return 0;
1954 err_out:
1955 swsusp_free();
1956 return -ENOMEM;
1959 asmlinkage __visible int swsusp_save(void)
1961 unsigned int nr_pages, nr_highmem;
1963 printk(KERN_INFO "PM: Creating hibernation image:\n");
1965 drain_local_pages(NULL);
1966 nr_pages = count_data_pages();
1967 nr_highmem = count_highmem_pages();
1968 printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem);
1970 if (!enough_free_mem(nr_pages, nr_highmem)) {
1971 printk(KERN_ERR "PM: Not enough free memory\n");
1972 return -ENOMEM;
1975 if (swsusp_alloc(&orig_bm, &copy_bm, nr_pages, nr_highmem)) {
1976 printk(KERN_ERR "PM: Memory allocation failed\n");
1977 return -ENOMEM;
1981 * During allocating of suspend pagedir, new cold pages may appear.
1982 * Kill them.
1984 drain_local_pages(NULL);
1985 copy_data_pages(&copy_bm, &orig_bm);
1988 * End of critical section. From now on, we can write to memory,
1989 * but we should not touch disk. This specially means we must _not_
1990 * touch swap space! Except we must write out our image of course.
1993 nr_pages += nr_highmem;
1994 nr_copy_pages = nr_pages;
1995 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1997 printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n",
1998 nr_pages);
2000 return 0;
2003 #ifndef CONFIG_ARCH_HIBERNATION_HEADER
2004 static int init_header_complete(struct swsusp_info *info)
2006 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2007 info->version_code = LINUX_VERSION_CODE;
2008 return 0;
2011 static char *check_image_kernel(struct swsusp_info *info)
2013 if (info->version_code != LINUX_VERSION_CODE)
2014 return "kernel version";
2015 if (strcmp(info->uts.sysname,init_utsname()->sysname))
2016 return "system type";
2017 if (strcmp(info->uts.release,init_utsname()->release))
2018 return "kernel release";
2019 if (strcmp(info->uts.version,init_utsname()->version))
2020 return "version";
2021 if (strcmp(info->uts.machine,init_utsname()->machine))
2022 return "machine";
2023 return NULL;
2025 #endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2027 unsigned long snapshot_get_image_size(void)
2029 return nr_copy_pages + nr_meta_pages + 1;
2032 static int init_header(struct swsusp_info *info)
2034 memset(info, 0, sizeof(struct swsusp_info));
2035 info->num_physpages = get_num_physpages();
2036 info->image_pages = nr_copy_pages;
2037 info->pages = snapshot_get_image_size();
2038 info->size = info->pages;
2039 info->size <<= PAGE_SHIFT;
2040 return init_header_complete(info);
2044 * pack_pfns - Prepare PFNs for saving.
2045 * @bm: Memory bitmap.
2046 * @buf: Memory buffer to store the PFNs in.
2048 * PFNs corresponding to set bits in @bm are stored in the area of memory
2049 * pointed to by @buf (1 page at a time).
2051 static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2053 int j;
2055 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2056 buf[j] = memory_bm_next_pfn(bm);
2057 if (unlikely(buf[j] == BM_END_OF_MAP))
2058 break;
2059 /* Save page key for data page (s390 only). */
2060 page_key_read(buf + j);
2065 * snapshot_read_next - Get the address to read the next image page from.
2066 * @handle: Snapshot handle to be used for the reading.
2068 * On the first call, @handle should point to a zeroed snapshot_handle
2069 * structure. The structure gets populated then and a pointer to it should be
2070 * passed to this function every next time.
2072 * On success, the function returns a positive number. Then, the caller
2073 * is allowed to read up to the returned number of bytes from the memory
2074 * location computed by the data_of() macro.
2076 * The function returns 0 to indicate the end of the data stream condition,
2077 * and negative numbers are returned on errors. If that happens, the structure
2078 * pointed to by @handle is not updated and should not be used any more.
2080 int snapshot_read_next(struct snapshot_handle *handle)
2082 if (handle->cur > nr_meta_pages + nr_copy_pages)
2083 return 0;
2085 if (!buffer) {
2086 /* This makes the buffer be freed by swsusp_free() */
2087 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2088 if (!buffer)
2089 return -ENOMEM;
2091 if (!handle->cur) {
2092 int error;
2094 error = init_header((struct swsusp_info *)buffer);
2095 if (error)
2096 return error;
2097 handle->buffer = buffer;
2098 memory_bm_position_reset(&orig_bm);
2099 memory_bm_position_reset(&copy_bm);
2100 } else if (handle->cur <= nr_meta_pages) {
2101 clear_page(buffer);
2102 pack_pfns(buffer, &orig_bm);
2103 } else {
2104 struct page *page;
2106 page = pfn_to_page(memory_bm_next_pfn(&copy_bm));
2107 if (PageHighMem(page)) {
2109 * Highmem pages are copied to the buffer,
2110 * because we can't return with a kmapped
2111 * highmem page (we may not be called again).
2113 void *kaddr;
2115 kaddr = kmap_atomic(page);
2116 copy_page(buffer, kaddr);
2117 kunmap_atomic(kaddr);
2118 handle->buffer = buffer;
2119 } else {
2120 handle->buffer = page_address(page);
2123 handle->cur++;
2124 return PAGE_SIZE;
2127 static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2128 struct memory_bitmap *src)
2130 unsigned long pfn;
2132 memory_bm_position_reset(src);
2133 pfn = memory_bm_next_pfn(src);
2134 while (pfn != BM_END_OF_MAP) {
2135 memory_bm_set_bit(dst, pfn);
2136 pfn = memory_bm_next_pfn(src);
2141 * mark_unsafe_pages - Mark pages that were used before hibernation.
2143 * Mark the pages that cannot be used for storing the image during restoration,
2144 * because they conflict with the pages that had been used before hibernation.
2146 static void mark_unsafe_pages(struct memory_bitmap *bm)
2148 unsigned long pfn;
2150 /* Clear the "free"/"unsafe" bit for all PFNs */
2151 memory_bm_position_reset(free_pages_map);
2152 pfn = memory_bm_next_pfn(free_pages_map);
2153 while (pfn != BM_END_OF_MAP) {
2154 memory_bm_clear_current(free_pages_map);
2155 pfn = memory_bm_next_pfn(free_pages_map);
2158 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2159 duplicate_memory_bitmap(free_pages_map, bm);
2161 allocated_unsafe_pages = 0;
2164 static int check_header(struct swsusp_info *info)
2166 char *reason;
2168 reason = check_image_kernel(info);
2169 if (!reason && info->num_physpages != get_num_physpages())
2170 reason = "memory size";
2171 if (reason) {
2172 printk(KERN_ERR "PM: Image mismatch: %s\n", reason);
2173 return -EPERM;
2175 return 0;
2179 * load header - Check the image header and copy the data from it.
2181 static int load_header(struct swsusp_info *info)
2183 int error;
2185 restore_pblist = NULL;
2186 error = check_header(info);
2187 if (!error) {
2188 nr_copy_pages = info->image_pages;
2189 nr_meta_pages = info->pages - info->image_pages - 1;
2191 return error;
2195 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2196 * @bm: Memory bitmap.
2197 * @buf: Area of memory containing the PFNs.
2199 * For each element of the array pointed to by @buf (1 page at a time), set the
2200 * corresponding bit in @bm.
2202 static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2204 int j;
2206 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2207 if (unlikely(buf[j] == BM_END_OF_MAP))
2208 break;
2210 /* Extract and buffer page key for data page (s390 only). */
2211 page_key_memorize(buf + j);
2213 if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2214 memory_bm_set_bit(bm, buf[j]);
2215 else
2216 return -EFAULT;
2219 return 0;
2222 #ifdef CONFIG_HIGHMEM
2224 * struct highmem_pbe is used for creating the list of highmem pages that
2225 * should be restored atomically during the resume from disk, because the page
2226 * frames they have occupied before the suspend are in use.
2228 struct highmem_pbe {
2229 struct page *copy_page; /* data is here now */
2230 struct page *orig_page; /* data was here before the suspend */
2231 struct highmem_pbe *next;
2235 * List of highmem PBEs needed for restoring the highmem pages that were
2236 * allocated before the suspend and included in the suspend image, but have
2237 * also been allocated by the "resume" kernel, so their contents cannot be
2238 * written directly to their "original" page frames.
2240 static struct highmem_pbe *highmem_pblist;
2243 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2244 * @bm: Memory bitmap.
2246 * The bits in @bm that correspond to image pages are assumed to be set.
2248 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2250 unsigned long pfn;
2251 unsigned int cnt = 0;
2253 memory_bm_position_reset(bm);
2254 pfn = memory_bm_next_pfn(bm);
2255 while (pfn != BM_END_OF_MAP) {
2256 if (PageHighMem(pfn_to_page(pfn)))
2257 cnt++;
2259 pfn = memory_bm_next_pfn(bm);
2261 return cnt;
2264 static unsigned int safe_highmem_pages;
2266 static struct memory_bitmap *safe_highmem_bm;
2269 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2270 * @bm: Pointer to an uninitialized memory bitmap structure.
2271 * @nr_highmem_p: Pointer to the number of highmem image pages.
2273 * Try to allocate as many highmem pages as there are highmem image pages
2274 * (@nr_highmem_p points to the variable containing the number of highmem image
2275 * pages). The pages that are "safe" (ie. will not be overwritten when the
2276 * hibernation image is restored entirely) have the corresponding bits set in
2277 * @bm (it must be unitialized).
2279 * NOTE: This function should not be called if there are no highmem image pages.
2281 static int prepare_highmem_image(struct memory_bitmap *bm,
2282 unsigned int *nr_highmem_p)
2284 unsigned int to_alloc;
2286 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2287 return -ENOMEM;
2289 if (get_highmem_buffer(PG_SAFE))
2290 return -ENOMEM;
2292 to_alloc = count_free_highmem_pages();
2293 if (to_alloc > *nr_highmem_p)
2294 to_alloc = *nr_highmem_p;
2295 else
2296 *nr_highmem_p = to_alloc;
2298 safe_highmem_pages = 0;
2299 while (to_alloc-- > 0) {
2300 struct page *page;
2302 page = alloc_page(__GFP_HIGHMEM);
2303 if (!swsusp_page_is_free(page)) {
2304 /* The page is "safe", set its bit the bitmap */
2305 memory_bm_set_bit(bm, page_to_pfn(page));
2306 safe_highmem_pages++;
2308 /* Mark the page as allocated */
2309 swsusp_set_page_forbidden(page);
2310 swsusp_set_page_free(page);
2312 memory_bm_position_reset(bm);
2313 safe_highmem_bm = bm;
2314 return 0;
2317 static struct page *last_highmem_page;
2320 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2322 * For a given highmem image page get a buffer that suspend_write_next() should
2323 * return to its caller to write to.
2325 * If the page is to be saved to its "original" page frame or a copy of
2326 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2327 * the copy of the page is to be made in normal memory, so the address of
2328 * the copy is returned.
2330 * If @buffer is returned, the caller of suspend_write_next() will write
2331 * the page's contents to @buffer, so they will have to be copied to the
2332 * right location on the next call to suspend_write_next() and it is done
2333 * with the help of copy_last_highmem_page(). For this purpose, if
2334 * @buffer is returned, @last_highmem_page is set to the page to which
2335 * the data will have to be copied from @buffer.
2337 static void *get_highmem_page_buffer(struct page *page,
2338 struct chain_allocator *ca)
2340 struct highmem_pbe *pbe;
2341 void *kaddr;
2343 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2345 * We have allocated the "original" page frame and we can
2346 * use it directly to store the loaded page.
2348 last_highmem_page = page;
2349 return buffer;
2352 * The "original" page frame has not been allocated and we have to
2353 * use a "safe" page frame to store the loaded page.
2355 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2356 if (!pbe) {
2357 swsusp_free();
2358 return ERR_PTR(-ENOMEM);
2360 pbe->orig_page = page;
2361 if (safe_highmem_pages > 0) {
2362 struct page *tmp;
2364 /* Copy of the page will be stored in high memory */
2365 kaddr = buffer;
2366 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2367 safe_highmem_pages--;
2368 last_highmem_page = tmp;
2369 pbe->copy_page = tmp;
2370 } else {
2371 /* Copy of the page will be stored in normal memory */
2372 kaddr = safe_pages_list;
2373 safe_pages_list = safe_pages_list->next;
2374 pbe->copy_page = virt_to_page(kaddr);
2376 pbe->next = highmem_pblist;
2377 highmem_pblist = pbe;
2378 return kaddr;
2382 * copy_last_highmem_page - Copy most the most recent highmem image page.
2384 * Copy the contents of a highmem image from @buffer, where the caller of
2385 * snapshot_write_next() has stored them, to the right location represented by
2386 * @last_highmem_page .
2388 static void copy_last_highmem_page(void)
2390 if (last_highmem_page) {
2391 void *dst;
2393 dst = kmap_atomic(last_highmem_page);
2394 copy_page(dst, buffer);
2395 kunmap_atomic(dst);
2396 last_highmem_page = NULL;
2400 static inline int last_highmem_page_copied(void)
2402 return !last_highmem_page;
2405 static inline void free_highmem_data(void)
2407 if (safe_highmem_bm)
2408 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2410 if (buffer)
2411 free_image_page(buffer, PG_UNSAFE_CLEAR);
2413 #else
2414 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2416 static inline int prepare_highmem_image(struct memory_bitmap *bm,
2417 unsigned int *nr_highmem_p) { return 0; }
2419 static inline void *get_highmem_page_buffer(struct page *page,
2420 struct chain_allocator *ca)
2422 return ERR_PTR(-EINVAL);
2425 static inline void copy_last_highmem_page(void) {}
2426 static inline int last_highmem_page_copied(void) { return 1; }
2427 static inline void free_highmem_data(void) {}
2428 #endif /* CONFIG_HIGHMEM */
2430 #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2433 * prepare_image - Make room for loading hibernation image.
2434 * @new_bm: Unitialized memory bitmap structure.
2435 * @bm: Memory bitmap with unsafe pages marked.
2437 * Use @bm to mark the pages that will be overwritten in the process of
2438 * restoring the system memory state from the suspend image ("unsafe" pages)
2439 * and allocate memory for the image.
2441 * The idea is to allocate a new memory bitmap first and then allocate
2442 * as many pages as needed for image data, but without specifying what those
2443 * pages will be used for just yet. Instead, we mark them all as allocated and
2444 * create a lists of "safe" pages to be used later. On systems with high
2445 * memory a list of "safe" highmem pages is created too.
2447 static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2449 unsigned int nr_pages, nr_highmem;
2450 struct linked_page *lp;
2451 int error;
2453 /* If there is no highmem, the buffer will not be necessary */
2454 free_image_page(buffer, PG_UNSAFE_CLEAR);
2455 buffer = NULL;
2457 nr_highmem = count_highmem_image_pages(bm);
2458 mark_unsafe_pages(bm);
2460 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2461 if (error)
2462 goto Free;
2464 duplicate_memory_bitmap(new_bm, bm);
2465 memory_bm_free(bm, PG_UNSAFE_KEEP);
2466 if (nr_highmem > 0) {
2467 error = prepare_highmem_image(bm, &nr_highmem);
2468 if (error)
2469 goto Free;
2472 * Reserve some safe pages for potential later use.
2474 * NOTE: This way we make sure there will be enough safe pages for the
2475 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2476 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2478 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2480 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2481 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2482 while (nr_pages > 0) {
2483 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2484 if (!lp) {
2485 error = -ENOMEM;
2486 goto Free;
2488 lp->next = safe_pages_list;
2489 safe_pages_list = lp;
2490 nr_pages--;
2492 /* Preallocate memory for the image */
2493 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2494 while (nr_pages > 0) {
2495 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2496 if (!lp) {
2497 error = -ENOMEM;
2498 goto Free;
2500 if (!swsusp_page_is_free(virt_to_page(lp))) {
2501 /* The page is "safe", add it to the list */
2502 lp->next = safe_pages_list;
2503 safe_pages_list = lp;
2505 /* Mark the page as allocated */
2506 swsusp_set_page_forbidden(virt_to_page(lp));
2507 swsusp_set_page_free(virt_to_page(lp));
2508 nr_pages--;
2510 return 0;
2512 Free:
2513 swsusp_free();
2514 return error;
2518 * get_buffer - Get the address to store the next image data page.
2520 * Get the address that snapshot_write_next() should return to its caller to
2521 * write to.
2523 static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2525 struct pbe *pbe;
2526 struct page *page;
2527 unsigned long pfn = memory_bm_next_pfn(bm);
2529 if (pfn == BM_END_OF_MAP)
2530 return ERR_PTR(-EFAULT);
2532 page = pfn_to_page(pfn);
2533 if (PageHighMem(page))
2534 return get_highmem_page_buffer(page, ca);
2536 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2538 * We have allocated the "original" page frame and we can
2539 * use it directly to store the loaded page.
2541 return page_address(page);
2544 * The "original" page frame has not been allocated and we have to
2545 * use a "safe" page frame to store the loaded page.
2547 pbe = chain_alloc(ca, sizeof(struct pbe));
2548 if (!pbe) {
2549 swsusp_free();
2550 return ERR_PTR(-ENOMEM);
2552 pbe->orig_address = page_address(page);
2553 pbe->address = safe_pages_list;
2554 safe_pages_list = safe_pages_list->next;
2555 pbe->next = restore_pblist;
2556 restore_pblist = pbe;
2557 return pbe->address;
2561 * snapshot_write_next - Get the address to store the next image page.
2562 * @handle: Snapshot handle structure to guide the writing.
2564 * On the first call, @handle should point to a zeroed snapshot_handle
2565 * structure. The structure gets populated then and a pointer to it should be
2566 * passed to this function every next time.
2568 * On success, the function returns a positive number. Then, the caller
2569 * is allowed to write up to the returned number of bytes to the memory
2570 * location computed by the data_of() macro.
2572 * The function returns 0 to indicate the "end of file" condition. Negative
2573 * numbers are returned on errors, in which cases the structure pointed to by
2574 * @handle is not updated and should not be used any more.
2576 int snapshot_write_next(struct snapshot_handle *handle)
2578 static struct chain_allocator ca;
2579 int error = 0;
2581 /* Check if we have already loaded the entire image */
2582 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2583 return 0;
2585 handle->sync_read = 1;
2587 if (!handle->cur) {
2588 if (!buffer)
2589 /* This makes the buffer be freed by swsusp_free() */
2590 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2592 if (!buffer)
2593 return -ENOMEM;
2595 handle->buffer = buffer;
2596 } else if (handle->cur == 1) {
2597 error = load_header(buffer);
2598 if (error)
2599 return error;
2601 safe_pages_list = NULL;
2603 error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
2604 if (error)
2605 return error;
2607 /* Allocate buffer for page keys. */
2608 error = page_key_alloc(nr_copy_pages);
2609 if (error)
2610 return error;
2612 hibernate_restore_protection_begin();
2613 } else if (handle->cur <= nr_meta_pages + 1) {
2614 error = unpack_orig_pfns(buffer, &copy_bm);
2615 if (error)
2616 return error;
2618 if (handle->cur == nr_meta_pages + 1) {
2619 error = prepare_image(&orig_bm, &copy_bm);
2620 if (error)
2621 return error;
2623 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2624 memory_bm_position_reset(&orig_bm);
2625 restore_pblist = NULL;
2626 handle->buffer = get_buffer(&orig_bm, &ca);
2627 handle->sync_read = 0;
2628 if (IS_ERR(handle->buffer))
2629 return PTR_ERR(handle->buffer);
2631 } else {
2632 copy_last_highmem_page();
2633 /* Restore page key for data page (s390 only). */
2634 page_key_write(handle->buffer);
2635 hibernate_restore_protect_page(handle->buffer);
2636 handle->buffer = get_buffer(&orig_bm, &ca);
2637 if (IS_ERR(handle->buffer))
2638 return PTR_ERR(handle->buffer);
2639 if (handle->buffer != buffer)
2640 handle->sync_read = 0;
2642 handle->cur++;
2643 return PAGE_SIZE;
2647 * snapshot_write_finalize - Complete the loading of a hibernation image.
2649 * Must be called after the last call to snapshot_write_next() in case the last
2650 * page in the image happens to be a highmem page and its contents should be
2651 * stored in highmem. Additionally, it recycles bitmap memory that's not
2652 * necessary any more.
2654 void snapshot_write_finalize(struct snapshot_handle *handle)
2656 copy_last_highmem_page();
2657 /* Restore page key for data page (s390 only). */
2658 page_key_write(handle->buffer);
2659 page_key_free();
2660 hibernate_restore_protect_page(handle->buffer);
2661 /* Do that only if we have loaded the image entirely */
2662 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2663 memory_bm_recycle(&orig_bm);
2664 free_highmem_data();
2668 int snapshot_image_loaded(struct snapshot_handle *handle)
2670 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2671 handle->cur <= nr_meta_pages + nr_copy_pages);
2674 #ifdef CONFIG_HIGHMEM
2675 /* Assumes that @buf is ready and points to a "safe" page */
2676 static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2677 void *buf)
2679 void *kaddr1, *kaddr2;
2681 kaddr1 = kmap_atomic(p1);
2682 kaddr2 = kmap_atomic(p2);
2683 copy_page(buf, kaddr1);
2684 copy_page(kaddr1, kaddr2);
2685 copy_page(kaddr2, buf);
2686 kunmap_atomic(kaddr2);
2687 kunmap_atomic(kaddr1);
2691 * restore_highmem - Put highmem image pages into their original locations.
2693 * For each highmem page that was in use before hibernation and is included in
2694 * the image, and also has been allocated by the "restore" kernel, swap its
2695 * current contents with the previous (ie. "before hibernation") ones.
2697 * If the restore eventually fails, we can call this function once again and
2698 * restore the highmem state as seen by the restore kernel.
2700 int restore_highmem(void)
2702 struct highmem_pbe *pbe = highmem_pblist;
2703 void *buf;
2705 if (!pbe)
2706 return 0;
2708 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2709 if (!buf)
2710 return -ENOMEM;
2712 while (pbe) {
2713 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2714 pbe = pbe->next;
2716 free_image_page(buf, PG_UNSAFE_CLEAR);
2717 return 0;
2719 #endif /* CONFIG_HIGHMEM */