d3f795f01bbce83741fb717e1ee1a05e1377c561
[~shefty/rdma-dev.git] / kernel / power / snapshot.c
1 /*
2  * linux/kernel/power/snapshot.c
3  *
4  * This file provides system snapshot/restore functionality for swsusp.
5  *
6  * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
7  * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
8  *
9  * This file is released under the GPLv2.
10  *
11  */
12
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
31 #include <asm/uaccess.h>
32 #include <asm/mmu_context.h>
33 #include <asm/pgtable.h>
34 #include <asm/tlbflush.h>
35 #include <asm/io.h>
36
37 #include "power.h"
38
39 static int swsusp_page_is_free(struct page *);
40 static void swsusp_set_page_forbidden(struct page *);
41 static void swsusp_unset_page_forbidden(struct page *);
42
43 /*
44  * Preferred image size in bytes (tunable via /sys/power/image_size).
45  * When it is set to N, swsusp will do its best to ensure the image
46  * size will not exceed N bytes, but if that is impossible, it will
47  * try to create the smallest image possible.
48  */
49 unsigned long image_size = 500 * 1024 * 1024;
50
51 /* List of PBEs needed for restoring the pages that were allocated before
52  * the suspend and included in the suspend image, but have also been
53  * allocated by the "resume" kernel, so their contents cannot be written
54  * directly to their "original" page frames.
55  */
56 struct pbe *restore_pblist;
57
58 /* Pointer to an auxiliary buffer (1 page) */
59 static void *buffer;
60
61 /**
62  *      @safe_needed - on resume, for storing the PBE list and the image,
63  *      we can only use memory pages that do not conflict with the pages
64  *      used before suspend.  The unsafe pages have PageNosaveFree set
65  *      and we count them using unsafe_pages.
66  *
67  *      Each allocated image page is marked as PageNosave and PageNosaveFree
68  *      so that swsusp_free() can release it.
69  */
70
71 #define PG_ANY          0
72 #define PG_SAFE         1
73 #define PG_UNSAFE_CLEAR 1
74 #define PG_UNSAFE_KEEP  0
75
76 static unsigned int allocated_unsafe_pages;
77
78 static void *get_image_page(gfp_t gfp_mask, int safe_needed)
79 {
80         void *res;
81
82         res = (void *)get_zeroed_page(gfp_mask);
83         if (safe_needed)
84                 while (res && swsusp_page_is_free(virt_to_page(res))) {
85                         /* The page is unsafe, mark it for swsusp_free() */
86                         swsusp_set_page_forbidden(virt_to_page(res));
87                         allocated_unsafe_pages++;
88                         res = (void *)get_zeroed_page(gfp_mask);
89                 }
90         if (res) {
91                 swsusp_set_page_forbidden(virt_to_page(res));
92                 swsusp_set_page_free(virt_to_page(res));
93         }
94         return res;
95 }
96
97 unsigned long get_safe_page(gfp_t gfp_mask)
98 {
99         return (unsigned long)get_image_page(gfp_mask, PG_SAFE);
100 }
101
102 static struct page *alloc_image_page(gfp_t gfp_mask)
103 {
104         struct page *page;
105
106         page = alloc_page(gfp_mask);
107         if (page) {
108                 swsusp_set_page_forbidden(page);
109                 swsusp_set_page_free(page);
110         }
111         return page;
112 }
113
114 /**
115  *      free_image_page - free page represented by @addr, allocated with
116  *      get_image_page (page flags set by it must be cleared)
117  */
118
119 static inline void free_image_page(void *addr, int clear_nosave_free)
120 {
121         struct page *page;
122
123         BUG_ON(!virt_addr_valid(addr));
124
125         page = virt_to_page(addr);
126
127         swsusp_unset_page_forbidden(page);
128         if (clear_nosave_free)
129                 swsusp_unset_page_free(page);
130
131         __free_page(page);
132 }
133
134 /* struct linked_page is used to build chains of pages */
135
136 #define LINKED_PAGE_DATA_SIZE   (PAGE_SIZE - sizeof(void *))
137
138 struct linked_page {
139         struct linked_page *next;
140         char data[LINKED_PAGE_DATA_SIZE];
141 } __attribute__((packed));
142
143 static inline void
144 free_list_of_pages(struct linked_page *list, int clear_page_nosave)
145 {
146         while (list) {
147                 struct linked_page *lp = list->next;
148
149                 free_image_page(list, clear_page_nosave);
150                 list = lp;
151         }
152 }
153
154 /**
155   *     struct chain_allocator is used for allocating small objects out of
156   *     a linked list of pages called 'the chain'.
157   *
158   *     The chain grows each time when there is no room for a new object in
159   *     the current page.  The allocated objects cannot be freed individually.
160   *     It is only possible to free them all at once, by freeing the entire
161   *     chain.
162   *
163   *     NOTE: The chain allocator may be inefficient if the allocated objects
164   *     are not much smaller than PAGE_SIZE.
165   */
166
167 struct chain_allocator {
168         struct linked_page *chain;      /* the chain */
169         unsigned int used_space;        /* total size of objects allocated out
170                                          * of the current page
171                                          */
172         gfp_t gfp_mask;         /* mask for allocating pages */
173         int safe_needed;        /* if set, only "safe" pages are allocated */
174 };
175
176 static void
177 chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed)
178 {
179         ca->chain = NULL;
180         ca->used_space = LINKED_PAGE_DATA_SIZE;
181         ca->gfp_mask = gfp_mask;
182         ca->safe_needed = safe_needed;
183 }
184
185 static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
186 {
187         void *ret;
188
189         if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
190                 struct linked_page *lp;
191
192                 lp = get_image_page(ca->gfp_mask, ca->safe_needed);
193                 if (!lp)
194                         return NULL;
195
196                 lp->next = ca->chain;
197                 ca->chain = lp;
198                 ca->used_space = 0;
199         }
200         ret = ca->chain->data + ca->used_space;
201         ca->used_space += size;
202         return ret;
203 }
204
205 /**
206  *      Data types related to memory bitmaps.
207  *
208  *      Memory bitmap is a structure consiting of many linked lists of
209  *      objects.  The main list's elements are of type struct zone_bitmap
210  *      and each of them corresonds to one zone.  For each zone bitmap
211  *      object there is a list of objects of type struct bm_block that
212  *      represent each blocks of bitmap in which information is stored.
213  *
214  *      struct memory_bitmap contains a pointer to the main list of zone
215  *      bitmap objects, a struct bm_position used for browsing the bitmap,
216  *      and a pointer to the list of pages used for allocating all of the
217  *      zone bitmap objects and bitmap block objects.
218  *
219  *      NOTE: It has to be possible to lay out the bitmap in memory
220  *      using only allocations of order 0.  Additionally, the bitmap is
221  *      designed to work with arbitrary number of zones (this is over the
222  *      top for now, but let's avoid making unnecessary assumptions ;-).
223  *
224  *      struct zone_bitmap contains a pointer to a list of bitmap block
225  *      objects and a pointer to the bitmap block object that has been
226  *      most recently used for setting bits.  Additionally, it contains the
227  *      pfns that correspond to the start and end of the represented zone.
228  *
229  *      struct bm_block contains a pointer to the memory page in which
230  *      information is stored (in the form of a block of bitmap)
231  *      It also contains the pfns that correspond to the start and end of
232  *      the represented memory area.
233  */
234
235 #define BM_END_OF_MAP   (~0UL)
236
237 #define BM_BITS_PER_BLOCK       (PAGE_SIZE * BITS_PER_BYTE)
238
239 struct bm_block {
240         struct list_head hook;  /* hook into a list of bitmap blocks */
241         unsigned long start_pfn;        /* pfn represented by the first bit */
242         unsigned long end_pfn;  /* pfn represented by the last bit plus 1 */
243         unsigned long *data;    /* bitmap representing pages */
244 };
245
246 static inline unsigned long bm_block_bits(struct bm_block *bb)
247 {
248         return bb->end_pfn - bb->start_pfn;
249 }
250
251 /* strcut bm_position is used for browsing memory bitmaps */
252
253 struct bm_position {
254         struct bm_block *block;
255         int bit;
256 };
257
258 struct memory_bitmap {
259         struct list_head blocks;        /* list of bitmap blocks */
260         struct linked_page *p_list;     /* list of pages used to store zone
261                                          * bitmap objects and bitmap block
262                                          * objects
263                                          */
264         struct bm_position cur; /* most recently used bit position */
265 };
266
267 /* Functions that operate on memory bitmaps */
268
269 static void memory_bm_position_reset(struct memory_bitmap *bm)
270 {
271         bm->cur.block = list_entry(bm->blocks.next, struct bm_block, hook);
272         bm->cur.bit = 0;
273 }
274
275 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
276
277 /**
278  *      create_bm_block_list - create a list of block bitmap objects
279  *      @pages - number of pages to track
280  *      @list - list to put the allocated blocks into
281  *      @ca - chain allocator to be used for allocating memory
282  */
283 static int create_bm_block_list(unsigned long pages,
284                                 struct list_head *list,
285                                 struct chain_allocator *ca)
286 {
287         unsigned int nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
288
289         while (nr_blocks-- > 0) {
290                 struct bm_block *bb;
291
292                 bb = chain_alloc(ca, sizeof(struct bm_block));
293                 if (!bb)
294                         return -ENOMEM;
295                 list_add(&bb->hook, list);
296         }
297
298         return 0;
299 }
300
301 struct mem_extent {
302         struct list_head hook;
303         unsigned long start;
304         unsigned long end;
305 };
306
307 /**
308  *      free_mem_extents - free a list of memory extents
309  *      @list - list of extents to empty
310  */
311 static void free_mem_extents(struct list_head *list)
312 {
313         struct mem_extent *ext, *aux;
314
315         list_for_each_entry_safe(ext, aux, list, hook) {
316                 list_del(&ext->hook);
317                 kfree(ext);
318         }
319 }
320
321 /**
322  *      create_mem_extents - create a list of memory extents representing
323  *                           contiguous ranges of PFNs
324  *      @list - list to put the extents into
325  *      @gfp_mask - mask to use for memory allocations
326  */
327 static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
328 {
329         struct zone *zone;
330
331         INIT_LIST_HEAD(list);
332
333         for_each_populated_zone(zone) {
334                 unsigned long zone_start, zone_end;
335                 struct mem_extent *ext, *cur, *aux;
336
337                 zone_start = zone->zone_start_pfn;
338                 zone_end = zone->zone_start_pfn + zone->spanned_pages;
339
340                 list_for_each_entry(ext, list, hook)
341                         if (zone_start <= ext->end)
342                                 break;
343
344                 if (&ext->hook == list || zone_end < ext->start) {
345                         /* New extent is necessary */
346                         struct mem_extent *new_ext;
347
348                         new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
349                         if (!new_ext) {
350                                 free_mem_extents(list);
351                                 return -ENOMEM;
352                         }
353                         new_ext->start = zone_start;
354                         new_ext->end = zone_end;
355                         list_add_tail(&new_ext->hook, &ext->hook);
356                         continue;
357                 }
358
359                 /* Merge this zone's range of PFNs with the existing one */
360                 if (zone_start < ext->start)
361                         ext->start = zone_start;
362                 if (zone_end > ext->end)
363                         ext->end = zone_end;
364
365                 /* More merging may be possible */
366                 cur = ext;
367                 list_for_each_entry_safe_continue(cur, aux, list, hook) {
368                         if (zone_end < cur->start)
369                                 break;
370                         if (zone_end < cur->end)
371                                 ext->end = cur->end;
372                         list_del(&cur->hook);
373                         kfree(cur);
374                 }
375         }
376
377         return 0;
378 }
379
380 /**
381   *     memory_bm_create - allocate memory for a memory bitmap
382   */
383 static int
384 memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed)
385 {
386         struct chain_allocator ca;
387         struct list_head mem_extents;
388         struct mem_extent *ext;
389         int error;
390
391         chain_init(&ca, gfp_mask, safe_needed);
392         INIT_LIST_HEAD(&bm->blocks);
393
394         error = create_mem_extents(&mem_extents, gfp_mask);
395         if (error)
396                 return error;
397
398         list_for_each_entry(ext, &mem_extents, hook) {
399                 struct bm_block *bb;
400                 unsigned long pfn = ext->start;
401                 unsigned long pages = ext->end - ext->start;
402
403                 bb = list_entry(bm->blocks.prev, struct bm_block, hook);
404
405                 error = create_bm_block_list(pages, bm->blocks.prev, &ca);
406                 if (error)
407                         goto Error;
408
409                 list_for_each_entry_continue(bb, &bm->blocks, hook) {
410                         bb->data = get_image_page(gfp_mask, safe_needed);
411                         if (!bb->data) {
412                                 error = -ENOMEM;
413                                 goto Error;
414                         }
415
416                         bb->start_pfn = pfn;
417                         if (pages >= BM_BITS_PER_BLOCK) {
418                                 pfn += BM_BITS_PER_BLOCK;
419                                 pages -= BM_BITS_PER_BLOCK;
420                         } else {
421                                 /* This is executed only once in the loop */
422                                 pfn += pages;
423                         }
424                         bb->end_pfn = pfn;
425                 }
426         }
427
428         bm->p_list = ca.chain;
429         memory_bm_position_reset(bm);
430  Exit:
431         free_mem_extents(&mem_extents);
432         return error;
433
434  Error:
435         bm->p_list = ca.chain;
436         memory_bm_free(bm, PG_UNSAFE_CLEAR);
437         goto Exit;
438 }
439
440 /**
441   *     memory_bm_free - free memory occupied by the memory bitmap @bm
442   */
443 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
444 {
445         struct bm_block *bb;
446
447         list_for_each_entry(bb, &bm->blocks, hook)
448                 if (bb->data)
449                         free_image_page(bb->data, clear_nosave_free);
450
451         free_list_of_pages(bm->p_list, clear_nosave_free);
452
453         INIT_LIST_HEAD(&bm->blocks);
454 }
455
456 /**
457  *      memory_bm_find_bit - find the bit in the bitmap @bm that corresponds
458  *      to given pfn.  The cur_zone_bm member of @bm and the cur_block member
459  *      of @bm->cur_zone_bm are updated.
460  */
461 static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
462                                 void **addr, unsigned int *bit_nr)
463 {
464         struct bm_block *bb;
465
466         /*
467          * Check if the pfn corresponds to the current bitmap block and find
468          * the block where it fits if this is not the case.
469          */
470         bb = bm->cur.block;
471         if (pfn < bb->start_pfn)
472                 list_for_each_entry_continue_reverse(bb, &bm->blocks, hook)
473                         if (pfn >= bb->start_pfn)
474                                 break;
475
476         if (pfn >= bb->end_pfn)
477                 list_for_each_entry_continue(bb, &bm->blocks, hook)
478                         if (pfn >= bb->start_pfn && pfn < bb->end_pfn)
479                                 break;
480
481         if (&bb->hook == &bm->blocks)
482                 return -EFAULT;
483
484         /* The block has been found */
485         bm->cur.block = bb;
486         pfn -= bb->start_pfn;
487         bm->cur.bit = pfn + 1;
488         *bit_nr = pfn;
489         *addr = bb->data;
490         return 0;
491 }
492
493 static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
494 {
495         void *addr;
496         unsigned int bit;
497         int error;
498
499         error = memory_bm_find_bit(bm, pfn, &addr, &bit);
500         BUG_ON(error);
501         set_bit(bit, addr);
502 }
503
504 static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
505 {
506         void *addr;
507         unsigned int bit;
508         int error;
509
510         error = memory_bm_find_bit(bm, pfn, &addr, &bit);
511         if (!error)
512                 set_bit(bit, addr);
513         return error;
514 }
515
516 static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
517 {
518         void *addr;
519         unsigned int bit;
520         int error;
521
522         error = memory_bm_find_bit(bm, pfn, &addr, &bit);
523         BUG_ON(error);
524         clear_bit(bit, addr);
525 }
526
527 static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
528 {
529         void *addr;
530         unsigned int bit;
531         int error;
532
533         error = memory_bm_find_bit(bm, pfn, &addr, &bit);
534         BUG_ON(error);
535         return test_bit(bit, addr);
536 }
537
538 static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
539 {
540         void *addr;
541         unsigned int bit;
542
543         return !memory_bm_find_bit(bm, pfn, &addr, &bit);
544 }
545
546 /**
547  *      memory_bm_next_pfn - find the pfn that corresponds to the next set bit
548  *      in the bitmap @bm.  If the pfn cannot be found, BM_END_OF_MAP is
549  *      returned.
550  *
551  *      It is required to run memory_bm_position_reset() before the first call to
552  *      this function.
553  */
554
555 static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
556 {
557         struct bm_block *bb;
558         int bit;
559
560         bb = bm->cur.block;
561         do {
562                 bit = bm->cur.bit;
563                 bit = find_next_bit(bb->data, bm_block_bits(bb), bit);
564                 if (bit < bm_block_bits(bb))
565                         goto Return_pfn;
566
567                 bb = list_entry(bb->hook.next, struct bm_block, hook);
568                 bm->cur.block = bb;
569                 bm->cur.bit = 0;
570         } while (&bb->hook != &bm->blocks);
571
572         memory_bm_position_reset(bm);
573         return BM_END_OF_MAP;
574
575  Return_pfn:
576         bm->cur.bit = bit + 1;
577         return bb->start_pfn + bit;
578 }
579
580 /**
581  *      This structure represents a range of page frames the contents of which
582  *      should not be saved during the suspend.
583  */
584
585 struct nosave_region {
586         struct list_head list;
587         unsigned long start_pfn;
588         unsigned long end_pfn;
589 };
590
591 static LIST_HEAD(nosave_regions);
592
593 /**
594  *      register_nosave_region - register a range of page frames the contents
595  *      of which should not be saved during the suspend (to be used in the early
596  *      initialization code)
597  */
598
599 void __init
600 __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn,
601                          int use_kmalloc)
602 {
603         struct nosave_region *region;
604
605         if (start_pfn >= end_pfn)
606                 return;
607
608         if (!list_empty(&nosave_regions)) {
609                 /* Try to extend the previous region (they should be sorted) */
610                 region = list_entry(nosave_regions.prev,
611                                         struct nosave_region, list);
612                 if (region->end_pfn == start_pfn) {
613                         region->end_pfn = end_pfn;
614                         goto Report;
615                 }
616         }
617         if (use_kmalloc) {
618                 /* during init, this shouldn't fail */
619                 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
620                 BUG_ON(!region);
621         } else
622                 /* This allocation cannot fail */
623                 region = alloc_bootmem(sizeof(struct nosave_region));
624         region->start_pfn = start_pfn;
625         region->end_pfn = end_pfn;
626         list_add_tail(&region->list, &nosave_regions);
627  Report:
628         printk(KERN_INFO "PM: Registered nosave memory: %016lx - %016lx\n",
629                 start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT);
630 }
631
632 /*
633  * Set bits in this map correspond to the page frames the contents of which
634  * should not be saved during the suspend.
635  */
636 static struct memory_bitmap *forbidden_pages_map;
637
638 /* Set bits in this map correspond to free page frames. */
639 static struct memory_bitmap *free_pages_map;
640
641 /*
642  * Each page frame allocated for creating the image is marked by setting the
643  * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
644  */
645
646 void swsusp_set_page_free(struct page *page)
647 {
648         if (free_pages_map)
649                 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
650 }
651
652 static int swsusp_page_is_free(struct page *page)
653 {
654         return free_pages_map ?
655                 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
656 }
657
658 void swsusp_unset_page_free(struct page *page)
659 {
660         if (free_pages_map)
661                 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
662 }
663
664 static void swsusp_set_page_forbidden(struct page *page)
665 {
666         if (forbidden_pages_map)
667                 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
668 }
669
670 int swsusp_page_is_forbidden(struct page *page)
671 {
672         return forbidden_pages_map ?
673                 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
674 }
675
676 static void swsusp_unset_page_forbidden(struct page *page)
677 {
678         if (forbidden_pages_map)
679                 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
680 }
681
682 /**
683  *      mark_nosave_pages - set bits corresponding to the page frames the
684  *      contents of which should not be saved in a given bitmap.
685  */
686
687 static void mark_nosave_pages(struct memory_bitmap *bm)
688 {
689         struct nosave_region *region;
690
691         if (list_empty(&nosave_regions))
692                 return;
693
694         list_for_each_entry(region, &nosave_regions, list) {
695                 unsigned long pfn;
696
697                 pr_debug("PM: Marking nosave pages: %016lx - %016lx\n",
698                                 region->start_pfn << PAGE_SHIFT,
699                                 region->end_pfn << PAGE_SHIFT);
700
701                 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
702                         if (pfn_valid(pfn)) {
703                                 /*
704                                  * It is safe to ignore the result of
705                                  * mem_bm_set_bit_check() here, since we won't
706                                  * touch the PFNs for which the error is
707                                  * returned anyway.
708                                  */
709                                 mem_bm_set_bit_check(bm, pfn);
710                         }
711         }
712 }
713
714 /**
715  *      create_basic_memory_bitmaps - create bitmaps needed for marking page
716  *      frames that should not be saved and free page frames.  The pointers
717  *      forbidden_pages_map and free_pages_map are only modified if everything
718  *      goes well, because we don't want the bits to be used before both bitmaps
719  *      are set up.
720  */
721
722 int create_basic_memory_bitmaps(void)
723 {
724         struct memory_bitmap *bm1, *bm2;
725         int error = 0;
726
727         BUG_ON(forbidden_pages_map || free_pages_map);
728
729         bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
730         if (!bm1)
731                 return -ENOMEM;
732
733         error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
734         if (error)
735                 goto Free_first_object;
736
737         bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
738         if (!bm2)
739                 goto Free_first_bitmap;
740
741         error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
742         if (error)
743                 goto Free_second_object;
744
745         forbidden_pages_map = bm1;
746         free_pages_map = bm2;
747         mark_nosave_pages(forbidden_pages_map);
748
749         pr_debug("PM: Basic memory bitmaps created\n");
750
751         return 0;
752
753  Free_second_object:
754         kfree(bm2);
755  Free_first_bitmap:
756         memory_bm_free(bm1, PG_UNSAFE_CLEAR);
757  Free_first_object:
758         kfree(bm1);
759         return -ENOMEM;
760 }
761
762 /**
763  *      free_basic_memory_bitmaps - free memory bitmaps allocated by
764  *      create_basic_memory_bitmaps().  The auxiliary pointers are necessary
765  *      so that the bitmaps themselves are not referred to while they are being
766  *      freed.
767  */
768
769 void free_basic_memory_bitmaps(void)
770 {
771         struct memory_bitmap *bm1, *bm2;
772
773         BUG_ON(!(forbidden_pages_map && free_pages_map));
774
775         bm1 = forbidden_pages_map;
776         bm2 = free_pages_map;
777         forbidden_pages_map = NULL;
778         free_pages_map = NULL;
779         memory_bm_free(bm1, PG_UNSAFE_CLEAR);
780         kfree(bm1);
781         memory_bm_free(bm2, PG_UNSAFE_CLEAR);
782         kfree(bm2);
783
784         pr_debug("PM: Basic memory bitmaps freed\n");
785 }
786
787 /**
788  *      snapshot_additional_pages - estimate the number of additional pages
789  *      be needed for setting up the suspend image data structures for given
790  *      zone (usually the returned value is greater than the exact number)
791  */
792
793 unsigned int snapshot_additional_pages(struct zone *zone)
794 {
795         unsigned int res;
796
797         res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
798         res += DIV_ROUND_UP(res * sizeof(struct bm_block), PAGE_SIZE);
799         return 2 * res;
800 }
801
802 #ifdef CONFIG_HIGHMEM
803 /**
804  *      count_free_highmem_pages - compute the total number of free highmem
805  *      pages, system-wide.
806  */
807
808 static unsigned int count_free_highmem_pages(void)
809 {
810         struct zone *zone;
811         unsigned int cnt = 0;
812
813         for_each_populated_zone(zone)
814                 if (is_highmem(zone))
815                         cnt += zone_page_state(zone, NR_FREE_PAGES);
816
817         return cnt;
818 }
819
820 /**
821  *      saveable_highmem_page - Determine whether a highmem page should be
822  *      included in the suspend image.
823  *
824  *      We should save the page if it isn't Nosave or NosaveFree, or Reserved,
825  *      and it isn't a part of a free chunk of pages.
826  */
827 static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
828 {
829         struct page *page;
830
831         if (!pfn_valid(pfn))
832                 return NULL;
833
834         page = pfn_to_page(pfn);
835         if (page_zone(page) != zone)
836                 return NULL;
837
838         BUG_ON(!PageHighMem(page));
839
840         if (swsusp_page_is_forbidden(page) ||  swsusp_page_is_free(page) ||
841             PageReserved(page))
842                 return NULL;
843
844         return page;
845 }
846
847 /**
848  *      count_highmem_pages - compute the total number of saveable highmem
849  *      pages.
850  */
851
852 static unsigned int count_highmem_pages(void)
853 {
854         struct zone *zone;
855         unsigned int n = 0;
856
857         for_each_populated_zone(zone) {
858                 unsigned long pfn, max_zone_pfn;
859
860                 if (!is_highmem(zone))
861                         continue;
862
863                 mark_free_pages(zone);
864                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
865                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
866                         if (saveable_highmem_page(zone, pfn))
867                                 n++;
868         }
869         return n;
870 }
871 #else
872 static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
873 {
874         return NULL;
875 }
876 #endif /* CONFIG_HIGHMEM */
877
878 /**
879  *      saveable_page - Determine whether a non-highmem page should be included
880  *      in the suspend image.
881  *
882  *      We should save the page if it isn't Nosave, and is not in the range
883  *      of pages statically defined as 'unsaveable', and it isn't a part of
884  *      a free chunk of pages.
885  */
886 static struct page *saveable_page(struct zone *zone, unsigned long pfn)
887 {
888         struct page *page;
889
890         if (!pfn_valid(pfn))
891                 return NULL;
892
893         page = pfn_to_page(pfn);
894         if (page_zone(page) != zone)
895                 return NULL;
896
897         BUG_ON(PageHighMem(page));
898
899         if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
900                 return NULL;
901
902         if (PageReserved(page)
903             && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
904                 return NULL;
905
906         return page;
907 }
908
909 /**
910  *      count_data_pages - compute the total number of saveable non-highmem
911  *      pages.
912  */
913
914 static unsigned int count_data_pages(void)
915 {
916         struct zone *zone;
917         unsigned long pfn, max_zone_pfn;
918         unsigned int n = 0;
919
920         for_each_populated_zone(zone) {
921                 if (is_highmem(zone))
922                         continue;
923
924                 mark_free_pages(zone);
925                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
926                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
927                         if (saveable_page(zone, pfn))
928                                 n++;
929         }
930         return n;
931 }
932
933 /* This is needed, because copy_page and memcpy are not usable for copying
934  * task structs.
935  */
936 static inline void do_copy_page(long *dst, long *src)
937 {
938         int n;
939
940         for (n = PAGE_SIZE / sizeof(long); n; n--)
941                 *dst++ = *src++;
942 }
943
944
945 /**
946  *      safe_copy_page - check if the page we are going to copy is marked as
947  *              present in the kernel page tables (this always is the case if
948  *              CONFIG_DEBUG_PAGEALLOC is not set and in that case
949  *              kernel_page_present() always returns 'true').
950  */
951 static void safe_copy_page(void *dst, struct page *s_page)
952 {
953         if (kernel_page_present(s_page)) {
954                 do_copy_page(dst, page_address(s_page));
955         } else {
956                 kernel_map_pages(s_page, 1, 1);
957                 do_copy_page(dst, page_address(s_page));
958                 kernel_map_pages(s_page, 1, 0);
959         }
960 }
961
962
963 #ifdef CONFIG_HIGHMEM
964 static inline struct page *
965 page_is_saveable(struct zone *zone, unsigned long pfn)
966 {
967         return is_highmem(zone) ?
968                 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
969 }
970
971 static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
972 {
973         struct page *s_page, *d_page;
974         void *src, *dst;
975
976         s_page = pfn_to_page(src_pfn);
977         d_page = pfn_to_page(dst_pfn);
978         if (PageHighMem(s_page)) {
979                 src = kmap_atomic(s_page, KM_USER0);
980                 dst = kmap_atomic(d_page, KM_USER1);
981                 do_copy_page(dst, src);
982                 kunmap_atomic(src, KM_USER0);
983                 kunmap_atomic(dst, KM_USER1);
984         } else {
985                 if (PageHighMem(d_page)) {
986                         /* Page pointed to by src may contain some kernel
987                          * data modified by kmap_atomic()
988                          */
989                         safe_copy_page(buffer, s_page);
990                         dst = kmap_atomic(d_page, KM_USER0);
991                         memcpy(dst, buffer, PAGE_SIZE);
992                         kunmap_atomic(dst, KM_USER0);
993                 } else {
994                         safe_copy_page(page_address(d_page), s_page);
995                 }
996         }
997 }
998 #else
999 #define page_is_saveable(zone, pfn)     saveable_page(zone, pfn)
1000
1001 static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1002 {
1003         safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1004                                 pfn_to_page(src_pfn));
1005 }
1006 #endif /* CONFIG_HIGHMEM */
1007
1008 static void
1009 copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm)
1010 {
1011         struct zone *zone;
1012         unsigned long pfn;
1013
1014         for_each_populated_zone(zone) {
1015                 unsigned long max_zone_pfn;
1016
1017                 mark_free_pages(zone);
1018                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1019                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1020                         if (page_is_saveable(zone, pfn))
1021                                 memory_bm_set_bit(orig_bm, pfn);
1022         }
1023         memory_bm_position_reset(orig_bm);
1024         memory_bm_position_reset(copy_bm);
1025         for(;;) {
1026                 pfn = memory_bm_next_pfn(orig_bm);
1027                 if (unlikely(pfn == BM_END_OF_MAP))
1028                         break;
1029                 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1030         }
1031 }
1032
1033 /* Total number of image pages */
1034 static unsigned int nr_copy_pages;
1035 /* Number of pages needed for saving the original pfns of the image pages */
1036 static unsigned int nr_meta_pages;
1037 /*
1038  * Numbers of normal and highmem page frames allocated for hibernation image
1039  * before suspending devices.
1040  */
1041 unsigned int alloc_normal, alloc_highmem;
1042 /*
1043  * Memory bitmap used for marking saveable pages (during hibernation) or
1044  * hibernation image pages (during restore)
1045  */
1046 static struct memory_bitmap orig_bm;
1047 /*
1048  * Memory bitmap used during hibernation for marking allocated page frames that
1049  * will contain copies of saveable pages.  During restore it is initially used
1050  * for marking hibernation image pages, but then the set bits from it are
1051  * duplicated in @orig_bm and it is released.  On highmem systems it is next
1052  * used for marking "safe" highmem pages, but it has to be reinitialized for
1053  * this purpose.
1054  */
1055 static struct memory_bitmap copy_bm;
1056
1057 /**
1058  *      swsusp_free - free pages allocated for the suspend.
1059  *
1060  *      Suspend pages are alocated before the atomic copy is made, so we
1061  *      need to release them after the resume.
1062  */
1063
1064 void swsusp_free(void)
1065 {
1066         struct zone *zone;
1067         unsigned long pfn, max_zone_pfn;
1068
1069         for_each_populated_zone(zone) {
1070                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1071                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1072                         if (pfn_valid(pfn)) {
1073                                 struct page *page = pfn_to_page(pfn);
1074
1075                                 if (swsusp_page_is_forbidden(page) &&
1076                                     swsusp_page_is_free(page)) {
1077                                         swsusp_unset_page_forbidden(page);
1078                                         swsusp_unset_page_free(page);
1079                                         __free_page(page);
1080                                 }
1081                         }
1082         }
1083         nr_copy_pages = 0;
1084         nr_meta_pages = 0;
1085         restore_pblist = NULL;
1086         buffer = NULL;
1087         alloc_normal = 0;
1088         alloc_highmem = 0;
1089 }
1090
1091 /* Helper functions used for the shrinking of memory. */
1092
1093 #define GFP_IMAGE       (GFP_KERNEL | __GFP_NOWARN)
1094
1095 /**
1096  * preallocate_image_pages - Allocate a number of pages for hibernation image
1097  * @nr_pages: Number of page frames to allocate.
1098  * @mask: GFP flags to use for the allocation.
1099  *
1100  * Return value: Number of page frames actually allocated
1101  */
1102 static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1103 {
1104         unsigned long nr_alloc = 0;
1105
1106         while (nr_pages > 0) {
1107                 struct page *page;
1108
1109                 page = alloc_image_page(mask);
1110                 if (!page)
1111                         break;
1112                 memory_bm_set_bit(&copy_bm, page_to_pfn(page));
1113                 if (PageHighMem(page))
1114                         alloc_highmem++;
1115                 else
1116                         alloc_normal++;
1117                 nr_pages--;
1118                 nr_alloc++;
1119         }
1120
1121         return nr_alloc;
1122 }
1123
1124 static unsigned long preallocate_image_memory(unsigned long nr_pages,
1125                                               unsigned long avail_normal)
1126 {
1127         unsigned long alloc;
1128
1129         if (avail_normal <= alloc_normal)
1130                 return 0;
1131
1132         alloc = avail_normal - alloc_normal;
1133         if (nr_pages < alloc)
1134                 alloc = nr_pages;
1135
1136         return preallocate_image_pages(alloc, GFP_IMAGE);
1137 }
1138
1139 #ifdef CONFIG_HIGHMEM
1140 static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1141 {
1142         return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1143 }
1144
1145 /**
1146  *  __fraction - Compute (an approximation of) x * (multiplier / base)
1147  */
1148 static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1149 {
1150         x *= multiplier;
1151         do_div(x, base);
1152         return (unsigned long)x;
1153 }
1154
1155 static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1156                                                 unsigned long highmem,
1157                                                 unsigned long total)
1158 {
1159         unsigned long alloc = __fraction(nr_pages, highmem, total);
1160
1161         return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1162 }
1163 #else /* CONFIG_HIGHMEM */
1164 static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1165 {
1166         return 0;
1167 }
1168
1169 static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1170                                                 unsigned long highmem,
1171                                                 unsigned long total)
1172 {
1173         return 0;
1174 }
1175 #endif /* CONFIG_HIGHMEM */
1176
1177 /**
1178  * free_unnecessary_pages - Release preallocated pages not needed for the image
1179  */
1180 static void free_unnecessary_pages(void)
1181 {
1182         unsigned long save, to_free_normal, to_free_highmem;
1183
1184         save = count_data_pages();
1185         if (alloc_normal >= save) {
1186                 to_free_normal = alloc_normal - save;
1187                 save = 0;
1188         } else {
1189                 to_free_normal = 0;
1190                 save -= alloc_normal;
1191         }
1192         save += count_highmem_pages();
1193         if (alloc_highmem >= save) {
1194                 to_free_highmem = alloc_highmem - save;
1195         } else {
1196                 to_free_highmem = 0;
1197                 to_free_normal -= save - alloc_highmem;
1198         }
1199
1200         memory_bm_position_reset(&copy_bm);
1201
1202         while (to_free_normal > 0 || to_free_highmem > 0) {
1203                 unsigned long pfn = memory_bm_next_pfn(&copy_bm);
1204                 struct page *page = pfn_to_page(pfn);
1205
1206                 if (PageHighMem(page)) {
1207                         if (!to_free_highmem)
1208                                 continue;
1209                         to_free_highmem--;
1210                         alloc_highmem--;
1211                 } else {
1212                         if (!to_free_normal)
1213                                 continue;
1214                         to_free_normal--;
1215                         alloc_normal--;
1216                 }
1217                 memory_bm_clear_bit(&copy_bm, pfn);
1218                 swsusp_unset_page_forbidden(page);
1219                 swsusp_unset_page_free(page);
1220                 __free_page(page);
1221         }
1222 }
1223
1224 /**
1225  * minimum_image_size - Estimate the minimum acceptable size of an image
1226  * @saveable: Number of saveable pages in the system.
1227  *
1228  * We want to avoid attempting to free too much memory too hard, so estimate the
1229  * minimum acceptable size of a hibernation image to use as the lower limit for
1230  * preallocating memory.
1231  *
1232  * We assume that the minimum image size should be proportional to
1233  *
1234  * [number of saveable pages] - [number of pages that can be freed in theory]
1235  *
1236  * where the second term is the sum of (1) reclaimable slab pages, (2) active
1237  * and (3) inactive anonymouns pages, (4) active and (5) inactive file pages,
1238  * minus mapped file pages.
1239  */
1240 static unsigned long minimum_image_size(unsigned long saveable)
1241 {
1242         unsigned long size;
1243
1244         size = global_page_state(NR_SLAB_RECLAIMABLE)
1245                 + global_page_state(NR_ACTIVE_ANON)
1246                 + global_page_state(NR_INACTIVE_ANON)
1247                 + global_page_state(NR_ACTIVE_FILE)
1248                 + global_page_state(NR_INACTIVE_FILE)
1249                 - global_page_state(NR_FILE_MAPPED);
1250
1251         return saveable <= size ? 0 : saveable - size;
1252 }
1253
1254 /**
1255  * hibernate_preallocate_memory - Preallocate memory for hibernation image
1256  *
1257  * To create a hibernation image it is necessary to make a copy of every page
1258  * frame in use.  We also need a number of page frames to be free during
1259  * hibernation for allocations made while saving the image and for device
1260  * drivers, in case they need to allocate memory from their hibernation
1261  * callbacks (these two numbers are given by PAGES_FOR_IO and SPARE_PAGES,
1262  * respectively, both of which are rough estimates).  To make this happen, we
1263  * compute the total number of available page frames and allocate at least
1264  *
1265  * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2 + 2 * SPARE_PAGES
1266  *
1267  * of them, which corresponds to the maximum size of a hibernation image.
1268  *
1269  * If image_size is set below the number following from the above formula,
1270  * the preallocation of memory is continued until the total number of saveable
1271  * pages in the system is below the requested image size or the minimum
1272  * acceptable image size returned by minimum_image_size(), whichever is greater.
1273  */
1274 int hibernate_preallocate_memory(void)
1275 {
1276         struct zone *zone;
1277         unsigned long saveable, size, max_size, count, highmem, pages = 0;
1278         unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1279         struct timeval start, stop;
1280         int error;
1281
1282         printk(KERN_INFO "PM: Preallocating image memory... ");
1283         do_gettimeofday(&start);
1284
1285         error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1286         if (error)
1287                 goto err_out;
1288
1289         error = memory_bm_create(&copy_bm, GFP_IMAGE, PG_ANY);
1290         if (error)
1291                 goto err_out;
1292
1293         alloc_normal = 0;
1294         alloc_highmem = 0;
1295
1296         /* Count the number of saveable data pages. */
1297         save_highmem = count_highmem_pages();
1298         saveable = count_data_pages();
1299
1300         /*
1301          * Compute the total number of page frames we can use (count) and the
1302          * number of pages needed for image metadata (size).
1303          */
1304         count = saveable;
1305         saveable += save_highmem;
1306         highmem = save_highmem;
1307         size = 0;
1308         for_each_populated_zone(zone) {
1309                 size += snapshot_additional_pages(zone);
1310                 if (is_highmem(zone))
1311                         highmem += zone_page_state(zone, NR_FREE_PAGES);
1312                 else
1313                         count += zone_page_state(zone, NR_FREE_PAGES);
1314         }
1315         avail_normal = count;
1316         count += highmem;
1317         count -= totalreserve_pages;
1318
1319         /* Compute the maximum number of saveable pages to leave in memory. */
1320         max_size = (count - (size + PAGES_FOR_IO)) / 2 - 2 * SPARE_PAGES;
1321         size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1322         if (size > max_size)
1323                 size = max_size;
1324         /*
1325          * If the maximum is not less than the current number of saveable pages
1326          * in memory, allocate page frames for the image and we're done.
1327          */
1328         if (size >= saveable) {
1329                 pages = preallocate_image_highmem(save_highmem);
1330                 pages += preallocate_image_memory(saveable - pages, avail_normal);
1331                 goto out;
1332         }
1333
1334         /* Estimate the minimum size of the image. */
1335         pages = minimum_image_size(saveable);
1336         /*
1337          * To avoid excessive pressure on the normal zone, leave room in it to
1338          * accommodate an image of the minimum size (unless it's already too
1339          * small, in which case don't preallocate pages from it at all).
1340          */
1341         if (avail_normal > pages)
1342                 avail_normal -= pages;
1343         else
1344                 avail_normal = 0;
1345         if (size < pages)
1346                 size = min_t(unsigned long, pages, max_size);
1347
1348         /*
1349          * Let the memory management subsystem know that we're going to need a
1350          * large number of page frames to allocate and make it free some memory.
1351          * NOTE: If this is not done, performance will be hurt badly in some
1352          * test cases.
1353          */
1354         shrink_all_memory(saveable - size);
1355
1356         /*
1357          * The number of saveable pages in memory was too high, so apply some
1358          * pressure to decrease it.  First, make room for the largest possible
1359          * image and fail if that doesn't work.  Next, try to decrease the size
1360          * of the image as much as indicated by 'size' using allocations from
1361          * highmem and non-highmem zones separately.
1362          */
1363         pages_highmem = preallocate_image_highmem(highmem / 2);
1364         alloc = (count - max_size) - pages_highmem;
1365         pages = preallocate_image_memory(alloc, avail_normal);
1366         if (pages < alloc) {
1367                 /* We have exhausted non-highmem pages, try highmem. */
1368                 alloc -= pages;
1369                 pages += pages_highmem;
1370                 pages_highmem = preallocate_image_highmem(alloc);
1371                 if (pages_highmem < alloc)
1372                         goto err_out;
1373                 pages += pages_highmem;
1374                 /*
1375                  * size is the desired number of saveable pages to leave in
1376                  * memory, so try to preallocate (all memory - size) pages.
1377                  */
1378                 alloc = (count - pages) - size;
1379                 pages += preallocate_image_highmem(alloc);
1380         } else {
1381                 /*
1382                  * There are approximately max_size saveable pages at this point
1383                  * and we want to reduce this number down to size.
1384                  */
1385                 alloc = max_size - size;
1386                 size = preallocate_highmem_fraction(alloc, highmem, count);
1387                 pages_highmem += size;
1388                 alloc -= size;
1389                 size = preallocate_image_memory(alloc, avail_normal);
1390                 pages_highmem += preallocate_image_highmem(alloc - size);
1391                 pages += pages_highmem + size;
1392         }
1393
1394         /*
1395          * We only need as many page frames for the image as there are saveable
1396          * pages in memory, but we have allocated more.  Release the excessive
1397          * ones now.
1398          */
1399         free_unnecessary_pages();
1400
1401  out:
1402         do_gettimeofday(&stop);
1403         printk(KERN_CONT "done (allocated %lu pages)\n", pages);
1404         swsusp_show_speed(&start, &stop, pages, "Allocated");
1405
1406         return 0;
1407
1408  err_out:
1409         printk(KERN_CONT "\n");
1410         swsusp_free();
1411         return -ENOMEM;
1412 }
1413
1414 #ifdef CONFIG_HIGHMEM
1415 /**
1416   *     count_pages_for_highmem - compute the number of non-highmem pages
1417   *     that will be necessary for creating copies of highmem pages.
1418   */
1419
1420 static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1421 {
1422         unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1423
1424         if (free_highmem >= nr_highmem)
1425                 nr_highmem = 0;
1426         else
1427                 nr_highmem -= free_highmem;
1428
1429         return nr_highmem;
1430 }
1431 #else
1432 static unsigned int
1433 count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1434 #endif /* CONFIG_HIGHMEM */
1435
1436 /**
1437  *      enough_free_mem - Make sure we have enough free memory for the
1438  *      snapshot image.
1439  */
1440
1441 static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1442 {
1443         struct zone *zone;
1444         unsigned int free = alloc_normal;
1445
1446         for_each_populated_zone(zone)
1447                 if (!is_highmem(zone))
1448                         free += zone_page_state(zone, NR_FREE_PAGES);
1449
1450         nr_pages += count_pages_for_highmem(nr_highmem);
1451         pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n",
1452                 nr_pages, PAGES_FOR_IO, free);
1453
1454         return free > nr_pages + PAGES_FOR_IO;
1455 }
1456
1457 #ifdef CONFIG_HIGHMEM
1458 /**
1459  *      get_highmem_buffer - if there are some highmem pages in the suspend
1460  *      image, we may need the buffer to copy them and/or load their data.
1461  */
1462
1463 static inline int get_highmem_buffer(int safe_needed)
1464 {
1465         buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
1466         return buffer ? 0 : -ENOMEM;
1467 }
1468
1469 /**
1470  *      alloc_highmem_image_pages - allocate some highmem pages for the image.
1471  *      Try to allocate as many pages as needed, but if the number of free
1472  *      highmem pages is lesser than that, allocate them all.
1473  */
1474
1475 static inline unsigned int
1476 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int nr_highmem)
1477 {
1478         unsigned int to_alloc = count_free_highmem_pages();
1479
1480         if (to_alloc > nr_highmem)
1481                 to_alloc = nr_highmem;
1482
1483         nr_highmem -= to_alloc;
1484         while (to_alloc-- > 0) {
1485                 struct page *page;
1486
1487                 page = alloc_image_page(__GFP_HIGHMEM);
1488                 memory_bm_set_bit(bm, page_to_pfn(page));
1489         }
1490         return nr_highmem;
1491 }
1492 #else
1493 static inline int get_highmem_buffer(int safe_needed) { return 0; }
1494
1495 static inline unsigned int
1496 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int n) { return 0; }
1497 #endif /* CONFIG_HIGHMEM */
1498
1499 /**
1500  *      swsusp_alloc - allocate memory for the suspend image
1501  *
1502  *      We first try to allocate as many highmem pages as there are
1503  *      saveable highmem pages in the system.  If that fails, we allocate
1504  *      non-highmem pages for the copies of the remaining highmem ones.
1505  *
1506  *      In this approach it is likely that the copies of highmem pages will
1507  *      also be located in the high memory, because of the way in which
1508  *      copy_data_pages() works.
1509  */
1510
1511 static int
1512 swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm,
1513                 unsigned int nr_pages, unsigned int nr_highmem)
1514 {
1515         int error = 0;
1516
1517         if (nr_highmem > 0) {
1518                 error = get_highmem_buffer(PG_ANY);
1519                 if (error)
1520                         goto err_out;
1521                 if (nr_highmem > alloc_highmem) {
1522                         nr_highmem -= alloc_highmem;
1523                         nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1524                 }
1525         }
1526         if (nr_pages > alloc_normal) {
1527                 nr_pages -= alloc_normal;
1528                 while (nr_pages-- > 0) {
1529                         struct page *page;
1530
1531                         page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
1532                         if (!page)
1533                                 goto err_out;
1534                         memory_bm_set_bit(copy_bm, page_to_pfn(page));
1535                 }
1536         }
1537
1538         return 0;
1539
1540  err_out:
1541         swsusp_free();
1542         return error;
1543 }
1544
1545 asmlinkage int swsusp_save(void)
1546 {
1547         unsigned int nr_pages, nr_highmem;
1548
1549         printk(KERN_INFO "PM: Creating hibernation image:\n");
1550
1551         drain_local_pages(NULL);
1552         nr_pages = count_data_pages();
1553         nr_highmem = count_highmem_pages();
1554         printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem);
1555
1556         if (!enough_free_mem(nr_pages, nr_highmem)) {
1557                 printk(KERN_ERR "PM: Not enough free memory\n");
1558                 return -ENOMEM;
1559         }
1560
1561         if (swsusp_alloc(&orig_bm, &copy_bm, nr_pages, nr_highmem)) {
1562                 printk(KERN_ERR "PM: Memory allocation failed\n");
1563                 return -ENOMEM;
1564         }
1565
1566         /* During allocating of suspend pagedir, new cold pages may appear.
1567          * Kill them.
1568          */
1569         drain_local_pages(NULL);
1570         copy_data_pages(&copy_bm, &orig_bm);
1571
1572         /*
1573          * End of critical section. From now on, we can write to memory,
1574          * but we should not touch disk. This specially means we must _not_
1575          * touch swap space! Except we must write out our image of course.
1576          */
1577
1578         nr_pages += nr_highmem;
1579         nr_copy_pages = nr_pages;
1580         nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1581
1582         printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n",
1583                 nr_pages);
1584
1585         return 0;
1586 }
1587
1588 #ifndef CONFIG_ARCH_HIBERNATION_HEADER
1589 static int init_header_complete(struct swsusp_info *info)
1590 {
1591         memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
1592         info->version_code = LINUX_VERSION_CODE;
1593         return 0;
1594 }
1595
1596 static char *check_image_kernel(struct swsusp_info *info)
1597 {
1598         if (info->version_code != LINUX_VERSION_CODE)
1599                 return "kernel version";
1600         if (strcmp(info->uts.sysname,init_utsname()->sysname))
1601                 return "system type";
1602         if (strcmp(info->uts.release,init_utsname()->release))
1603                 return "kernel release";
1604         if (strcmp(info->uts.version,init_utsname()->version))
1605                 return "version";
1606         if (strcmp(info->uts.machine,init_utsname()->machine))
1607                 return "machine";
1608         return NULL;
1609 }
1610 #endif /* CONFIG_ARCH_HIBERNATION_HEADER */
1611
1612 unsigned long snapshot_get_image_size(void)
1613 {
1614         return nr_copy_pages + nr_meta_pages + 1;
1615 }
1616
1617 static int init_header(struct swsusp_info *info)
1618 {
1619         memset(info, 0, sizeof(struct swsusp_info));
1620         info->num_physpages = num_physpages;
1621         info->image_pages = nr_copy_pages;
1622         info->pages = snapshot_get_image_size();
1623         info->size = info->pages;
1624         info->size <<= PAGE_SHIFT;
1625         return init_header_complete(info);
1626 }
1627
1628 /**
1629  *      pack_pfns - pfns corresponding to the set bits found in the bitmap @bm
1630  *      are stored in the array @buf[] (1 page at a time)
1631  */
1632
1633 static inline void
1634 pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
1635 {
1636         int j;
1637
1638         for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1639                 buf[j] = memory_bm_next_pfn(bm);
1640                 if (unlikely(buf[j] == BM_END_OF_MAP))
1641                         break;
1642         }
1643 }
1644
1645 /**
1646  *      snapshot_read_next - used for reading the system memory snapshot.
1647  *
1648  *      On the first call to it @handle should point to a zeroed
1649  *      snapshot_handle structure.  The structure gets updated and a pointer
1650  *      to it should be passed to this function every next time.
1651  *
1652  *      On success the function returns a positive number.  Then, the caller
1653  *      is allowed to read up to the returned number of bytes from the memory
1654  *      location computed by the data_of() macro.
1655  *
1656  *      The function returns 0 to indicate the end of data stream condition,
1657  *      and a negative number is returned on error.  In such cases the
1658  *      structure pointed to by @handle is not updated and should not be used
1659  *      any more.
1660  */
1661
1662 int snapshot_read_next(struct snapshot_handle *handle)
1663 {
1664         if (handle->cur > nr_meta_pages + nr_copy_pages)
1665                 return 0;
1666
1667         if (!buffer) {
1668                 /* This makes the buffer be freed by swsusp_free() */
1669                 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
1670                 if (!buffer)
1671                         return -ENOMEM;
1672         }
1673         if (!handle->cur) {
1674                 int error;
1675
1676                 error = init_header((struct swsusp_info *)buffer);
1677                 if (error)
1678                         return error;
1679                 handle->buffer = buffer;
1680                 memory_bm_position_reset(&orig_bm);
1681                 memory_bm_position_reset(&copy_bm);
1682         } else if (handle->cur <= nr_meta_pages) {
1683                 memset(buffer, 0, PAGE_SIZE);
1684                 pack_pfns(buffer, &orig_bm);
1685         } else {
1686                 struct page *page;
1687
1688                 page = pfn_to_page(memory_bm_next_pfn(&copy_bm));
1689                 if (PageHighMem(page)) {
1690                         /* Highmem pages are copied to the buffer,
1691                          * because we can't return with a kmapped
1692                          * highmem page (we may not be called again).
1693                          */
1694                         void *kaddr;
1695
1696                         kaddr = kmap_atomic(page, KM_USER0);
1697                         memcpy(buffer, kaddr, PAGE_SIZE);
1698                         kunmap_atomic(kaddr, KM_USER0);
1699                         handle->buffer = buffer;
1700                 } else {
1701                         handle->buffer = page_address(page);
1702                 }
1703         }
1704         handle->cur++;
1705         return PAGE_SIZE;
1706 }
1707
1708 /**
1709  *      mark_unsafe_pages - mark the pages that cannot be used for storing
1710  *      the image during resume, because they conflict with the pages that
1711  *      had been used before suspend
1712  */
1713
1714 static int mark_unsafe_pages(struct memory_bitmap *bm)
1715 {
1716         struct zone *zone;
1717         unsigned long pfn, max_zone_pfn;
1718
1719         /* Clear page flags */
1720         for_each_populated_zone(zone) {
1721                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1722                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1723                         if (pfn_valid(pfn))
1724                                 swsusp_unset_page_free(pfn_to_page(pfn));
1725         }
1726
1727         /* Mark pages that correspond to the "original" pfns as "unsafe" */
1728         memory_bm_position_reset(bm);
1729         do {
1730                 pfn = memory_bm_next_pfn(bm);
1731                 if (likely(pfn != BM_END_OF_MAP)) {
1732                         if (likely(pfn_valid(pfn)))
1733                                 swsusp_set_page_free(pfn_to_page(pfn));
1734                         else
1735                                 return -EFAULT;
1736                 }
1737         } while (pfn != BM_END_OF_MAP);
1738
1739         allocated_unsafe_pages = 0;
1740
1741         return 0;
1742 }
1743
1744 static void
1745 duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src)
1746 {
1747         unsigned long pfn;
1748
1749         memory_bm_position_reset(src);
1750         pfn = memory_bm_next_pfn(src);
1751         while (pfn != BM_END_OF_MAP) {
1752                 memory_bm_set_bit(dst, pfn);
1753                 pfn = memory_bm_next_pfn(src);
1754         }
1755 }
1756
1757 static int check_header(struct swsusp_info *info)
1758 {
1759         char *reason;
1760
1761         reason = check_image_kernel(info);
1762         if (!reason && info->num_physpages != num_physpages)
1763                 reason = "memory size";
1764         if (reason) {
1765                 printk(KERN_ERR "PM: Image mismatch: %s\n", reason);
1766                 return -EPERM;
1767         }
1768         return 0;
1769 }
1770
1771 /**
1772  *      load header - check the image header and copy data from it
1773  */
1774
1775 static int
1776 load_header(struct swsusp_info *info)
1777 {
1778         int error;
1779
1780         restore_pblist = NULL;
1781         error = check_header(info);
1782         if (!error) {
1783                 nr_copy_pages = info->image_pages;
1784                 nr_meta_pages = info->pages - info->image_pages - 1;
1785         }
1786         return error;
1787 }
1788
1789 /**
1790  *      unpack_orig_pfns - for each element of @buf[] (1 page at a time) set
1791  *      the corresponding bit in the memory bitmap @bm
1792  */
1793 static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
1794 {
1795         int j;
1796
1797         for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1798                 if (unlikely(buf[j] == BM_END_OF_MAP))
1799                         break;
1800
1801                 if (memory_bm_pfn_present(bm, buf[j]))
1802                         memory_bm_set_bit(bm, buf[j]);
1803                 else
1804                         return -EFAULT;
1805         }
1806
1807         return 0;
1808 }
1809
1810 /* List of "safe" pages that may be used to store data loaded from the suspend
1811  * image
1812  */
1813 static struct linked_page *safe_pages_list;
1814
1815 #ifdef CONFIG_HIGHMEM
1816 /* struct highmem_pbe is used for creating the list of highmem pages that
1817  * should be restored atomically during the resume from disk, because the page
1818  * frames they have occupied before the suspend are in use.
1819  */
1820 struct highmem_pbe {
1821         struct page *copy_page; /* data is here now */
1822         struct page *orig_page; /* data was here before the suspend */
1823         struct highmem_pbe *next;
1824 };
1825
1826 /* List of highmem PBEs needed for restoring the highmem pages that were
1827  * allocated before the suspend and included in the suspend image, but have
1828  * also been allocated by the "resume" kernel, so their contents cannot be
1829  * written directly to their "original" page frames.
1830  */
1831 static struct highmem_pbe *highmem_pblist;
1832
1833 /**
1834  *      count_highmem_image_pages - compute the number of highmem pages in the
1835  *      suspend image.  The bits in the memory bitmap @bm that correspond to the
1836  *      image pages are assumed to be set.
1837  */
1838
1839 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
1840 {
1841         unsigned long pfn;
1842         unsigned int cnt = 0;
1843
1844         memory_bm_position_reset(bm);
1845         pfn = memory_bm_next_pfn(bm);
1846         while (pfn != BM_END_OF_MAP) {
1847                 if (PageHighMem(pfn_to_page(pfn)))
1848                         cnt++;
1849
1850                 pfn = memory_bm_next_pfn(bm);
1851         }
1852         return cnt;
1853 }
1854
1855 /**
1856  *      prepare_highmem_image - try to allocate as many highmem pages as
1857  *      there are highmem image pages (@nr_highmem_p points to the variable
1858  *      containing the number of highmem image pages).  The pages that are
1859  *      "safe" (ie. will not be overwritten when the suspend image is
1860  *      restored) have the corresponding bits set in @bm (it must be
1861  *      unitialized).
1862  *
1863  *      NOTE: This function should not be called if there are no highmem
1864  *      image pages.
1865  */
1866
1867 static unsigned int safe_highmem_pages;
1868
1869 static struct memory_bitmap *safe_highmem_bm;
1870
1871 static int
1872 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
1873 {
1874         unsigned int to_alloc;
1875
1876         if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
1877                 return -ENOMEM;
1878
1879         if (get_highmem_buffer(PG_SAFE))
1880                 return -ENOMEM;
1881
1882         to_alloc = count_free_highmem_pages();
1883         if (to_alloc > *nr_highmem_p)
1884                 to_alloc = *nr_highmem_p;
1885         else
1886                 *nr_highmem_p = to_alloc;
1887
1888         safe_highmem_pages = 0;
1889         while (to_alloc-- > 0) {
1890                 struct page *page;
1891
1892                 page = alloc_page(__GFP_HIGHMEM);
1893                 if (!swsusp_page_is_free(page)) {
1894                         /* The page is "safe", set its bit the bitmap */
1895                         memory_bm_set_bit(bm, page_to_pfn(page));
1896                         safe_highmem_pages++;
1897                 }
1898                 /* Mark the page as allocated */
1899                 swsusp_set_page_forbidden(page);
1900                 swsusp_set_page_free(page);
1901         }
1902         memory_bm_position_reset(bm);
1903         safe_highmem_bm = bm;
1904         return 0;
1905 }
1906
1907 /**
1908  *      get_highmem_page_buffer - for given highmem image page find the buffer
1909  *      that suspend_write_next() should set for its caller to write to.
1910  *
1911  *      If the page is to be saved to its "original" page frame or a copy of
1912  *      the page is to be made in the highmem, @buffer is returned.  Otherwise,
1913  *      the copy of the page is to be made in normal memory, so the address of
1914  *      the copy is returned.
1915  *
1916  *      If @buffer is returned, the caller of suspend_write_next() will write
1917  *      the page's contents to @buffer, so they will have to be copied to the
1918  *      right location on the next call to suspend_write_next() and it is done
1919  *      with the help of copy_last_highmem_page().  For this purpose, if
1920  *      @buffer is returned, @last_highmem page is set to the page to which
1921  *      the data will have to be copied from @buffer.
1922  */
1923
1924 static struct page *last_highmem_page;
1925
1926 static void *
1927 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
1928 {
1929         struct highmem_pbe *pbe;
1930         void *kaddr;
1931
1932         if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
1933                 /* We have allocated the "original" page frame and we can
1934                  * use it directly to store the loaded page.
1935                  */
1936                 last_highmem_page = page;
1937                 return buffer;
1938         }
1939         /* The "original" page frame has not been allocated and we have to
1940          * use a "safe" page frame to store the loaded page.
1941          */
1942         pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
1943         if (!pbe) {
1944                 swsusp_free();
1945                 return ERR_PTR(-ENOMEM);
1946         }
1947         pbe->orig_page = page;
1948         if (safe_highmem_pages > 0) {
1949                 struct page *tmp;
1950
1951                 /* Copy of the page will be stored in high memory */
1952                 kaddr = buffer;
1953                 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
1954                 safe_highmem_pages--;
1955                 last_highmem_page = tmp;
1956                 pbe->copy_page = tmp;
1957         } else {
1958                 /* Copy of the page will be stored in normal memory */
1959                 kaddr = safe_pages_list;
1960                 safe_pages_list = safe_pages_list->next;
1961                 pbe->copy_page = virt_to_page(kaddr);
1962         }
1963         pbe->next = highmem_pblist;
1964         highmem_pblist = pbe;
1965         return kaddr;
1966 }
1967
1968 /**
1969  *      copy_last_highmem_page - copy the contents of a highmem image from
1970  *      @buffer, where the caller of snapshot_write_next() has place them,
1971  *      to the right location represented by @last_highmem_page .
1972  */
1973
1974 static void copy_last_highmem_page(void)
1975 {
1976         if (last_highmem_page) {
1977                 void *dst;
1978
1979                 dst = kmap_atomic(last_highmem_page, KM_USER0);
1980                 memcpy(dst, buffer, PAGE_SIZE);
1981                 kunmap_atomic(dst, KM_USER0);
1982                 last_highmem_page = NULL;
1983         }
1984 }
1985
1986 static inline int last_highmem_page_copied(void)
1987 {
1988         return !last_highmem_page;
1989 }
1990
1991 static inline void free_highmem_data(void)
1992 {
1993         if (safe_highmem_bm)
1994                 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
1995
1996         if (buffer)
1997                 free_image_page(buffer, PG_UNSAFE_CLEAR);
1998 }
1999 #else
2000 static inline int get_safe_write_buffer(void) { return 0; }
2001
2002 static unsigned int
2003 count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2004
2005 static inline int
2006 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
2007 {
2008         return 0;
2009 }
2010
2011 static inline void *
2012 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
2013 {
2014         return ERR_PTR(-EINVAL);
2015 }
2016
2017 static inline void copy_last_highmem_page(void) {}
2018 static inline int last_highmem_page_copied(void) { return 1; }
2019 static inline void free_highmem_data(void) {}
2020 #endif /* CONFIG_HIGHMEM */
2021
2022 /**
2023  *      prepare_image - use the memory bitmap @bm to mark the pages that will
2024  *      be overwritten in the process of restoring the system memory state
2025  *      from the suspend image ("unsafe" pages) and allocate memory for the
2026  *      image.
2027  *
2028  *      The idea is to allocate a new memory bitmap first and then allocate
2029  *      as many pages as needed for the image data, but not to assign these
2030  *      pages to specific tasks initially.  Instead, we just mark them as
2031  *      allocated and create a lists of "safe" pages that will be used
2032  *      later.  On systems with high memory a list of "safe" highmem pages is
2033  *      also created.
2034  */
2035
2036 #define PBES_PER_LINKED_PAGE    (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2037
2038 static int
2039 prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2040 {
2041         unsigned int nr_pages, nr_highmem;
2042         struct linked_page *sp_list, *lp;
2043         int error;
2044
2045         /* If there is no highmem, the buffer will not be necessary */
2046         free_image_page(buffer, PG_UNSAFE_CLEAR);
2047         buffer = NULL;
2048
2049         nr_highmem = count_highmem_image_pages(bm);
2050         error = mark_unsafe_pages(bm);
2051         if (error)
2052                 goto Free;
2053
2054         error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2055         if (error)
2056                 goto Free;
2057
2058         duplicate_memory_bitmap(new_bm, bm);
2059         memory_bm_free(bm, PG_UNSAFE_KEEP);
2060         if (nr_highmem > 0) {
2061                 error = prepare_highmem_image(bm, &nr_highmem);
2062                 if (error)
2063                         goto Free;
2064         }
2065         /* Reserve some safe pages for potential later use.
2066          *
2067          * NOTE: This way we make sure there will be enough safe pages for the
2068          * chain_alloc() in get_buffer().  It is a bit wasteful, but
2069          * nr_copy_pages cannot be greater than 50% of the memory anyway.
2070          */
2071         sp_list = NULL;
2072         /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */
2073         nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2074         nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2075         while (nr_pages > 0) {
2076                 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2077                 if (!lp) {
2078                         error = -ENOMEM;
2079                         goto Free;
2080                 }
2081                 lp->next = sp_list;
2082                 sp_list = lp;
2083                 nr_pages--;
2084         }
2085         /* Preallocate memory for the image */
2086         safe_pages_list = NULL;
2087         nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2088         while (nr_pages > 0) {
2089                 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2090                 if (!lp) {
2091                         error = -ENOMEM;
2092                         goto Free;
2093                 }
2094                 if (!swsusp_page_is_free(virt_to_page(lp))) {
2095                         /* The page is "safe", add it to the list */
2096                         lp->next = safe_pages_list;
2097                         safe_pages_list = lp;
2098                 }
2099                 /* Mark the page as allocated */
2100                 swsusp_set_page_forbidden(virt_to_page(lp));
2101                 swsusp_set_page_free(virt_to_page(lp));
2102                 nr_pages--;
2103         }
2104         /* Free the reserved safe pages so that chain_alloc() can use them */
2105         while (sp_list) {
2106                 lp = sp_list->next;
2107                 free_image_page(sp_list, PG_UNSAFE_CLEAR);
2108                 sp_list = lp;
2109         }
2110         return 0;
2111
2112  Free:
2113         swsusp_free();
2114         return error;
2115 }
2116
2117 /**
2118  *      get_buffer - compute the address that snapshot_write_next() should
2119  *      set for its caller to write to.
2120  */
2121
2122 static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2123 {
2124         struct pbe *pbe;
2125         struct page *page;
2126         unsigned long pfn = memory_bm_next_pfn(bm);
2127
2128         if (pfn == BM_END_OF_MAP)
2129                 return ERR_PTR(-EFAULT);
2130
2131         page = pfn_to_page(pfn);
2132         if (PageHighMem(page))
2133                 return get_highmem_page_buffer(page, ca);
2134
2135         if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2136                 /* We have allocated the "original" page frame and we can
2137                  * use it directly to store the loaded page.
2138                  */
2139                 return page_address(page);
2140
2141         /* The "original" page frame has not been allocated and we have to
2142          * use a "safe" page frame to store the loaded page.
2143          */
2144         pbe = chain_alloc(ca, sizeof(struct pbe));
2145         if (!pbe) {
2146                 swsusp_free();
2147                 return ERR_PTR(-ENOMEM);
2148         }
2149         pbe->orig_address = page_address(page);
2150         pbe->address = safe_pages_list;
2151         safe_pages_list = safe_pages_list->next;
2152         pbe->next = restore_pblist;
2153         restore_pblist = pbe;
2154         return pbe->address;
2155 }
2156
2157 /**
2158  *      snapshot_write_next - used for writing the system memory snapshot.
2159  *
2160  *      On the first call to it @handle should point to a zeroed
2161  *      snapshot_handle structure.  The structure gets updated and a pointer
2162  *      to it should be passed to this function every next time.
2163  *
2164  *      On success the function returns a positive number.  Then, the caller
2165  *      is allowed to write up to the returned number of bytes to the memory
2166  *      location computed by the data_of() macro.
2167  *
2168  *      The function returns 0 to indicate the "end of file" condition,
2169  *      and a negative number is returned on error.  In such cases the
2170  *      structure pointed to by @handle is not updated and should not be used
2171  *      any more.
2172  */
2173
2174 int snapshot_write_next(struct snapshot_handle *handle)
2175 {
2176         static struct chain_allocator ca;
2177         int error = 0;
2178
2179         /* Check if we have already loaded the entire image */
2180         if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2181                 return 0;
2182
2183         handle->sync_read = 1;
2184
2185         if (!handle->cur) {
2186                 if (!buffer)
2187                         /* This makes the buffer be freed by swsusp_free() */
2188                         buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2189
2190                 if (!buffer)
2191                         return -ENOMEM;
2192
2193                 handle->buffer = buffer;
2194         } else if (handle->cur == 1) {
2195                 error = load_header(buffer);
2196                 if (error)
2197                         return error;
2198
2199                 error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
2200                 if (error)
2201                         return error;
2202
2203         } else if (handle->cur <= nr_meta_pages + 1) {
2204                 error = unpack_orig_pfns(buffer, &copy_bm);
2205                 if (error)
2206                         return error;
2207
2208                 if (handle->cur == nr_meta_pages + 1) {
2209                         error = prepare_image(&orig_bm, &copy_bm);
2210                         if (error)
2211                                 return error;
2212
2213                         chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2214                         memory_bm_position_reset(&orig_bm);
2215                         restore_pblist = NULL;
2216                         handle->buffer = get_buffer(&orig_bm, &ca);
2217                         handle->sync_read = 0;
2218                         if (IS_ERR(handle->buffer))
2219                                 return PTR_ERR(handle->buffer);
2220                 }
2221         } else {
2222                 copy_last_highmem_page();
2223                 handle->buffer = get_buffer(&orig_bm, &ca);
2224                 if (IS_ERR(handle->buffer))
2225                         return PTR_ERR(handle->buffer);
2226                 if (handle->buffer != buffer)
2227                         handle->sync_read = 0;
2228         }
2229         handle->cur++;
2230         return PAGE_SIZE;
2231 }
2232
2233 /**
2234  *      snapshot_write_finalize - must be called after the last call to
2235  *      snapshot_write_next() in case the last page in the image happens
2236  *      to be a highmem page and its contents should be stored in the
2237  *      highmem.  Additionally, it releases the memory that will not be
2238  *      used any more.
2239  */
2240
2241 void snapshot_write_finalize(struct snapshot_handle *handle)
2242 {
2243         copy_last_highmem_page();
2244         /* Free only if we have loaded the image entirely */
2245         if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2246                 memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR);
2247                 free_highmem_data();
2248         }
2249 }
2250
2251 int snapshot_image_loaded(struct snapshot_handle *handle)
2252 {
2253         return !(!nr_copy_pages || !last_highmem_page_copied() ||
2254                         handle->cur <= nr_meta_pages + nr_copy_pages);
2255 }
2256
2257 #ifdef CONFIG_HIGHMEM
2258 /* Assumes that @buf is ready and points to a "safe" page */
2259 static inline void
2260 swap_two_pages_data(struct page *p1, struct page *p2, void *buf)
2261 {
2262         void *kaddr1, *kaddr2;
2263
2264         kaddr1 = kmap_atomic(p1, KM_USER0);
2265         kaddr2 = kmap_atomic(p2, KM_USER1);
2266         memcpy(buf, kaddr1, PAGE_SIZE);
2267         memcpy(kaddr1, kaddr2, PAGE_SIZE);
2268         memcpy(kaddr2, buf, PAGE_SIZE);
2269         kunmap_atomic(kaddr1, KM_USER0);
2270         kunmap_atomic(kaddr2, KM_USER1);
2271 }
2272
2273 /**
2274  *      restore_highmem - for each highmem page that was allocated before
2275  *      the suspend and included in the suspend image, and also has been
2276  *      allocated by the "resume" kernel swap its current (ie. "before
2277  *      resume") contents with the previous (ie. "before suspend") one.
2278  *
2279  *      If the resume eventually fails, we can call this function once
2280  *      again and restore the "before resume" highmem state.
2281  */
2282
2283 int restore_highmem(void)
2284 {
2285         struct highmem_pbe *pbe = highmem_pblist;
2286         void *buf;
2287
2288         if (!pbe)
2289                 return 0;
2290
2291         buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2292         if (!buf)
2293                 return -ENOMEM;
2294
2295         while (pbe) {
2296                 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2297                 pbe = pbe->next;
2298         }
2299         free_image_page(buf, PG_UNSAFE_CLEAR);
2300         return 0;
2301 }
2302 #endif /* CONFIG_HIGHMEM */