Pull platform-drivers into test branch
[linux-2.6] / 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@suse.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/smp_lock.h>
18 #include <linux/delay.h>
19 #include <linux/bitops.h>
20 #include <linux/spinlock.h>
21 #include <linux/kernel.h>
22 #include <linux/pm.h>
23 #include <linux/device.h>
24 #include <linux/bootmem.h>
25 #include <linux/syscalls.h>
26 #include <linux/console.h>
27 #include <linux/highmem.h>
28
29 #include <asm/uaccess.h>
30 #include <asm/mmu_context.h>
31 #include <asm/pgtable.h>
32 #include <asm/tlbflush.h>
33 #include <asm/io.h>
34
35 #include "power.h"
36
37 /* List of PBEs needed for restoring the pages that were allocated before
38  * the suspend and included in the suspend image, but have also been
39  * allocated by the "resume" kernel, so their contents cannot be written
40  * directly to their "original" page frames.
41  */
42 struct pbe *restore_pblist;
43
44 /* Pointer to an auxiliary buffer (1 page) */
45 static void *buffer;
46
47 /**
48  *      @safe_needed - on resume, for storing the PBE list and the image,
49  *      we can only use memory pages that do not conflict with the pages
50  *      used before suspend.  The unsafe pages have PageNosaveFree set
51  *      and we count them using unsafe_pages.
52  *
53  *      Each allocated image page is marked as PageNosave and PageNosaveFree
54  *      so that swsusp_free() can release it.
55  */
56
57 #define PG_ANY          0
58 #define PG_SAFE         1
59 #define PG_UNSAFE_CLEAR 1
60 #define PG_UNSAFE_KEEP  0
61
62 static unsigned int allocated_unsafe_pages;
63
64 static void *get_image_page(gfp_t gfp_mask, int safe_needed)
65 {
66         void *res;
67
68         res = (void *)get_zeroed_page(gfp_mask);
69         if (safe_needed)
70                 while (res && PageNosaveFree(virt_to_page(res))) {
71                         /* The page is unsafe, mark it for swsusp_free() */
72                         SetPageNosave(virt_to_page(res));
73                         allocated_unsafe_pages++;
74                         res = (void *)get_zeroed_page(gfp_mask);
75                 }
76         if (res) {
77                 SetPageNosave(virt_to_page(res));
78                 SetPageNosaveFree(virt_to_page(res));
79         }
80         return res;
81 }
82
83 unsigned long get_safe_page(gfp_t gfp_mask)
84 {
85         return (unsigned long)get_image_page(gfp_mask, PG_SAFE);
86 }
87
88 static struct page *alloc_image_page(gfp_t gfp_mask)
89 {
90         struct page *page;
91
92         page = alloc_page(gfp_mask);
93         if (page) {
94                 SetPageNosave(page);
95                 SetPageNosaveFree(page);
96         }
97         return page;
98 }
99
100 /**
101  *      free_image_page - free page represented by @addr, allocated with
102  *      get_image_page (page flags set by it must be cleared)
103  */
104
105 static inline void free_image_page(void *addr, int clear_nosave_free)
106 {
107         struct page *page;
108
109         BUG_ON(!virt_addr_valid(addr));
110
111         page = virt_to_page(addr);
112
113         ClearPageNosave(page);
114         if (clear_nosave_free)
115                 ClearPageNosaveFree(page);
116
117         __free_page(page);
118 }
119
120 /* struct linked_page is used to build chains of pages */
121
122 #define LINKED_PAGE_DATA_SIZE   (PAGE_SIZE - sizeof(void *))
123
124 struct linked_page {
125         struct linked_page *next;
126         char data[LINKED_PAGE_DATA_SIZE];
127 } __attribute__((packed));
128
129 static inline void
130 free_list_of_pages(struct linked_page *list, int clear_page_nosave)
131 {
132         while (list) {
133                 struct linked_page *lp = list->next;
134
135                 free_image_page(list, clear_page_nosave);
136                 list = lp;
137         }
138 }
139
140 /**
141   *     struct chain_allocator is used for allocating small objects out of
142   *     a linked list of pages called 'the chain'.
143   *
144   *     The chain grows each time when there is no room for a new object in
145   *     the current page.  The allocated objects cannot be freed individually.
146   *     It is only possible to free them all at once, by freeing the entire
147   *     chain.
148   *
149   *     NOTE: The chain allocator may be inefficient if the allocated objects
150   *     are not much smaller than PAGE_SIZE.
151   */
152
153 struct chain_allocator {
154         struct linked_page *chain;      /* the chain */
155         unsigned int used_space;        /* total size of objects allocated out
156                                          * of the current page
157                                          */
158         gfp_t gfp_mask;         /* mask for allocating pages */
159         int safe_needed;        /* if set, only "safe" pages are allocated */
160 };
161
162 static void
163 chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed)
164 {
165         ca->chain = NULL;
166         ca->used_space = LINKED_PAGE_DATA_SIZE;
167         ca->gfp_mask = gfp_mask;
168         ca->safe_needed = safe_needed;
169 }
170
171 static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
172 {
173         void *ret;
174
175         if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
176                 struct linked_page *lp;
177
178                 lp = get_image_page(ca->gfp_mask, ca->safe_needed);
179                 if (!lp)
180                         return NULL;
181
182                 lp->next = ca->chain;
183                 ca->chain = lp;
184                 ca->used_space = 0;
185         }
186         ret = ca->chain->data + ca->used_space;
187         ca->used_space += size;
188         return ret;
189 }
190
191 static void chain_free(struct chain_allocator *ca, int clear_page_nosave)
192 {
193         free_list_of_pages(ca->chain, clear_page_nosave);
194         memset(ca, 0, sizeof(struct chain_allocator));
195 }
196
197 /**
198  *      Data types related to memory bitmaps.
199  *
200  *      Memory bitmap is a structure consiting of many linked lists of
201  *      objects.  The main list's elements are of type struct zone_bitmap
202  *      and each of them corresonds to one zone.  For each zone bitmap
203  *      object there is a list of objects of type struct bm_block that
204  *      represent each blocks of bit chunks in which information is
205  *      stored.
206  *
207  *      struct memory_bitmap contains a pointer to the main list of zone
208  *      bitmap objects, a struct bm_position used for browsing the bitmap,
209  *      and a pointer to the list of pages used for allocating all of the
210  *      zone bitmap objects and bitmap block objects.
211  *
212  *      NOTE: It has to be possible to lay out the bitmap in memory
213  *      using only allocations of order 0.  Additionally, the bitmap is
214  *      designed to work with arbitrary number of zones (this is over the
215  *      top for now, but let's avoid making unnecessary assumptions ;-).
216  *
217  *      struct zone_bitmap contains a pointer to a list of bitmap block
218  *      objects and a pointer to the bitmap block object that has been
219  *      most recently used for setting bits.  Additionally, it contains the
220  *      pfns that correspond to the start and end of the represented zone.
221  *
222  *      struct bm_block contains a pointer to the memory page in which
223  *      information is stored (in the form of a block of bit chunks
224  *      of type unsigned long each).  It also contains the pfns that
225  *      correspond to the start and end of the represented memory area and
226  *      the number of bit chunks in the block.
227  *
228  *      NOTE: Memory bitmaps are used for two types of operations only:
229  *      "set a bit" and "find the next bit set".  Moreover, the searching
230  *      is always carried out after all of the "set a bit" operations
231  *      on given bitmap.
232  */
233
234 #define BM_END_OF_MAP   (~0UL)
235
236 #define BM_CHUNKS_PER_BLOCK     (PAGE_SIZE / sizeof(long))
237 #define BM_BITS_PER_CHUNK       (sizeof(long) << 3)
238 #define BM_BITS_PER_BLOCK       (PAGE_SIZE << 3)
239
240 struct bm_block {
241         struct bm_block *next;          /* next element of the list */
242         unsigned long start_pfn;        /* pfn represented by the first bit */
243         unsigned long end_pfn;  /* pfn represented by the last bit plus 1 */
244         unsigned int size;      /* number of bit chunks */
245         unsigned long *data;    /* chunks of bits representing pages */
246 };
247
248 struct zone_bitmap {
249         struct zone_bitmap *next;       /* next element of the list */
250         unsigned long start_pfn;        /* minimal pfn in this zone */
251         unsigned long end_pfn;          /* maximal pfn in this zone plus 1 */
252         struct bm_block *bm_blocks;     /* list of bitmap blocks */
253         struct bm_block *cur_block;     /* recently used bitmap block */
254 };
255
256 /* strcut bm_position is used for browsing memory bitmaps */
257
258 struct bm_position {
259         struct zone_bitmap *zone_bm;
260         struct bm_block *block;
261         int chunk;
262         int bit;
263 };
264
265 struct memory_bitmap {
266         struct zone_bitmap *zone_bm_list;       /* list of zone bitmaps */
267         struct linked_page *p_list;     /* list of pages used to store zone
268                                          * bitmap objects and bitmap block
269                                          * objects
270                                          */
271         struct bm_position cur; /* most recently used bit position */
272 };
273
274 /* Functions that operate on memory bitmaps */
275
276 static inline void memory_bm_reset_chunk(struct memory_bitmap *bm)
277 {
278         bm->cur.chunk = 0;
279         bm->cur.bit = -1;
280 }
281
282 static void memory_bm_position_reset(struct memory_bitmap *bm)
283 {
284         struct zone_bitmap *zone_bm;
285
286         zone_bm = bm->zone_bm_list;
287         bm->cur.zone_bm = zone_bm;
288         bm->cur.block = zone_bm->bm_blocks;
289         memory_bm_reset_chunk(bm);
290 }
291
292 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
293
294 /**
295  *      create_bm_block_list - create a list of block bitmap objects
296  */
297
298 static inline struct bm_block *
299 create_bm_block_list(unsigned int nr_blocks, struct chain_allocator *ca)
300 {
301         struct bm_block *bblist = NULL;
302
303         while (nr_blocks-- > 0) {
304                 struct bm_block *bb;
305
306                 bb = chain_alloc(ca, sizeof(struct bm_block));
307                 if (!bb)
308                         return NULL;
309
310                 bb->next = bblist;
311                 bblist = bb;
312         }
313         return bblist;
314 }
315
316 /**
317  *      create_zone_bm_list - create a list of zone bitmap objects
318  */
319
320 static inline struct zone_bitmap *
321 create_zone_bm_list(unsigned int nr_zones, struct chain_allocator *ca)
322 {
323         struct zone_bitmap *zbmlist = NULL;
324
325         while (nr_zones-- > 0) {
326                 struct zone_bitmap *zbm;
327
328                 zbm = chain_alloc(ca, sizeof(struct zone_bitmap));
329                 if (!zbm)
330                         return NULL;
331
332                 zbm->next = zbmlist;
333                 zbmlist = zbm;
334         }
335         return zbmlist;
336 }
337
338 /**
339   *     memory_bm_create - allocate memory for a memory bitmap
340   */
341
342 static int
343 memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed)
344 {
345         struct chain_allocator ca;
346         struct zone *zone;
347         struct zone_bitmap *zone_bm;
348         struct bm_block *bb;
349         unsigned int nr;
350
351         chain_init(&ca, gfp_mask, safe_needed);
352
353         /* Compute the number of zones */
354         nr = 0;
355         for_each_zone(zone)
356                 if (populated_zone(zone))
357                         nr++;
358
359         /* Allocate the list of zones bitmap objects */
360         zone_bm = create_zone_bm_list(nr, &ca);
361         bm->zone_bm_list = zone_bm;
362         if (!zone_bm) {
363                 chain_free(&ca, PG_UNSAFE_CLEAR);
364                 return -ENOMEM;
365         }
366
367         /* Initialize the zone bitmap objects */
368         for_each_zone(zone) {
369                 unsigned long pfn;
370
371                 if (!populated_zone(zone))
372                         continue;
373
374                 zone_bm->start_pfn = zone->zone_start_pfn;
375                 zone_bm->end_pfn = zone->zone_start_pfn + zone->spanned_pages;
376                 /* Allocate the list of bitmap block objects */
377                 nr = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
378                 bb = create_bm_block_list(nr, &ca);
379                 zone_bm->bm_blocks = bb;
380                 zone_bm->cur_block = bb;
381                 if (!bb)
382                         goto Free;
383
384                 nr = zone->spanned_pages;
385                 pfn = zone->zone_start_pfn;
386                 /* Initialize the bitmap block objects */
387                 while (bb) {
388                         unsigned long *ptr;
389
390                         ptr = get_image_page(gfp_mask, safe_needed);
391                         bb->data = ptr;
392                         if (!ptr)
393                                 goto Free;
394
395                         bb->start_pfn = pfn;
396                         if (nr >= BM_BITS_PER_BLOCK) {
397                                 pfn += BM_BITS_PER_BLOCK;
398                                 bb->size = BM_CHUNKS_PER_BLOCK;
399                                 nr -= BM_BITS_PER_BLOCK;
400                         } else {
401                                 /* This is executed only once in the loop */
402                                 pfn += nr;
403                                 bb->size = DIV_ROUND_UP(nr, BM_BITS_PER_CHUNK);
404                         }
405                         bb->end_pfn = pfn;
406                         bb = bb->next;
407                 }
408                 zone_bm = zone_bm->next;
409         }
410         bm->p_list = ca.chain;
411         memory_bm_position_reset(bm);
412         return 0;
413
414  Free:
415         bm->p_list = ca.chain;
416         memory_bm_free(bm, PG_UNSAFE_CLEAR);
417         return -ENOMEM;
418 }
419
420 /**
421   *     memory_bm_free - free memory occupied by the memory bitmap @bm
422   */
423
424 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
425 {
426         struct zone_bitmap *zone_bm;
427
428         /* Free the list of bit blocks for each zone_bitmap object */
429         zone_bm = bm->zone_bm_list;
430         while (zone_bm) {
431                 struct bm_block *bb;
432
433                 bb = zone_bm->bm_blocks;
434                 while (bb) {
435                         if (bb->data)
436                                 free_image_page(bb->data, clear_nosave_free);
437                         bb = bb->next;
438                 }
439                 zone_bm = zone_bm->next;
440         }
441         free_list_of_pages(bm->p_list, clear_nosave_free);
442         bm->zone_bm_list = NULL;
443 }
444
445 /**
446  *      memory_bm_set_bit - set the bit in the bitmap @bm that corresponds
447  *      to given pfn.  The cur_zone_bm member of @bm and the cur_block member
448  *      of @bm->cur_zone_bm are updated.
449  *
450  *      If the bit cannot be set, the function returns -EINVAL .
451  */
452
453 static int
454 memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
455 {
456         struct zone_bitmap *zone_bm;
457         struct bm_block *bb;
458
459         /* Check if the pfn is from the current zone */
460         zone_bm = bm->cur.zone_bm;
461         if (pfn < zone_bm->start_pfn || pfn >= zone_bm->end_pfn) {
462                 zone_bm = bm->zone_bm_list;
463                 /* We don't assume that the zones are sorted by pfns */
464                 while (pfn < zone_bm->start_pfn || pfn >= zone_bm->end_pfn) {
465                         zone_bm = zone_bm->next;
466                         if (unlikely(!zone_bm))
467                                 return -EINVAL;
468                 }
469                 bm->cur.zone_bm = zone_bm;
470         }
471         /* Check if the pfn corresponds to the current bitmap block */
472         bb = zone_bm->cur_block;
473         if (pfn < bb->start_pfn)
474                 bb = zone_bm->bm_blocks;
475
476         while (pfn >= bb->end_pfn) {
477                 bb = bb->next;
478                 if (unlikely(!bb))
479                         return -EINVAL;
480         }
481         zone_bm->cur_block = bb;
482         pfn -= bb->start_pfn;
483         set_bit(pfn % BM_BITS_PER_CHUNK, bb->data + pfn / BM_BITS_PER_CHUNK);
484         return 0;
485 }
486
487 /* Two auxiliary functions for memory_bm_next_pfn */
488
489 /* Find the first set bit in the given chunk, if there is one */
490
491 static inline int next_bit_in_chunk(int bit, unsigned long *chunk_p)
492 {
493         bit++;
494         while (bit < BM_BITS_PER_CHUNK) {
495                 if (test_bit(bit, chunk_p))
496                         return bit;
497
498                 bit++;
499         }
500         return -1;
501 }
502
503 /* Find a chunk containing some bits set in given block of bits */
504
505 static inline int next_chunk_in_block(int n, struct bm_block *bb)
506 {
507         n++;
508         while (n < bb->size) {
509                 if (bb->data[n])
510                         return n;
511
512                 n++;
513         }
514         return -1;
515 }
516
517 /**
518  *      memory_bm_next_pfn - find the pfn that corresponds to the next set bit
519  *      in the bitmap @bm.  If the pfn cannot be found, BM_END_OF_MAP is
520  *      returned.
521  *
522  *      It is required to run memory_bm_position_reset() before the first call to
523  *      this function.
524  */
525
526 static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
527 {
528         struct zone_bitmap *zone_bm;
529         struct bm_block *bb;
530         int chunk;
531         int bit;
532
533         do {
534                 bb = bm->cur.block;
535                 do {
536                         chunk = bm->cur.chunk;
537                         bit = bm->cur.bit;
538                         do {
539                                 bit = next_bit_in_chunk(bit, bb->data + chunk);
540                                 if (bit >= 0)
541                                         goto Return_pfn;
542
543                                 chunk = next_chunk_in_block(chunk, bb);
544                                 bit = -1;
545                         } while (chunk >= 0);
546                         bb = bb->next;
547                         bm->cur.block = bb;
548                         memory_bm_reset_chunk(bm);
549                 } while (bb);
550                 zone_bm = bm->cur.zone_bm->next;
551                 if (zone_bm) {
552                         bm->cur.zone_bm = zone_bm;
553                         bm->cur.block = zone_bm->bm_blocks;
554                         memory_bm_reset_chunk(bm);
555                 }
556         } while (zone_bm);
557         memory_bm_position_reset(bm);
558         return BM_END_OF_MAP;
559
560  Return_pfn:
561         bm->cur.chunk = chunk;
562         bm->cur.bit = bit;
563         return bb->start_pfn + chunk * BM_BITS_PER_CHUNK + bit;
564 }
565
566 /**
567  *      snapshot_additional_pages - estimate the number of additional pages
568  *      be needed for setting up the suspend image data structures for given
569  *      zone (usually the returned value is greater than the exact number)
570  */
571
572 unsigned int snapshot_additional_pages(struct zone *zone)
573 {
574         unsigned int res;
575
576         res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
577         res += DIV_ROUND_UP(res * sizeof(struct bm_block), PAGE_SIZE);
578         return 2 * res;
579 }
580
581 #ifdef CONFIG_HIGHMEM
582 /**
583  *      count_free_highmem_pages - compute the total number of free highmem
584  *      pages, system-wide.
585  */
586
587 static unsigned int count_free_highmem_pages(void)
588 {
589         struct zone *zone;
590         unsigned int cnt = 0;
591
592         for_each_zone(zone)
593                 if (populated_zone(zone) && is_highmem(zone))
594                         cnt += zone->free_pages;
595
596         return cnt;
597 }
598
599 /**
600  *      saveable_highmem_page - Determine whether a highmem page should be
601  *      included in the suspend image.
602  *
603  *      We should save the page if it isn't Nosave or NosaveFree, or Reserved,
604  *      and it isn't a part of a free chunk of pages.
605  */
606
607 static struct page *saveable_highmem_page(unsigned long pfn)
608 {
609         struct page *page;
610
611         if (!pfn_valid(pfn))
612                 return NULL;
613
614         page = pfn_to_page(pfn);
615
616         BUG_ON(!PageHighMem(page));
617
618         if (PageNosave(page) || PageReserved(page) || PageNosaveFree(page))
619                 return NULL;
620
621         return page;
622 }
623
624 /**
625  *      count_highmem_pages - compute the total number of saveable highmem
626  *      pages.
627  */
628
629 unsigned int count_highmem_pages(void)
630 {
631         struct zone *zone;
632         unsigned int n = 0;
633
634         for_each_zone(zone) {
635                 unsigned long pfn, max_zone_pfn;
636
637                 if (!is_highmem(zone))
638                         continue;
639
640                 mark_free_pages(zone);
641                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
642                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
643                         if (saveable_highmem_page(pfn))
644                                 n++;
645         }
646         return n;
647 }
648 #else
649 static inline void *saveable_highmem_page(unsigned long pfn) { return NULL; }
650 static inline unsigned int count_highmem_pages(void) { return 0; }
651 #endif /* CONFIG_HIGHMEM */
652
653 /**
654  *      pfn_is_nosave - check if given pfn is in the 'nosave' section
655  */
656
657 static inline int pfn_is_nosave(unsigned long pfn)
658 {
659         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
660         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
661         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
662 }
663
664 /**
665  *      saveable - Determine whether a non-highmem page should be included in
666  *      the suspend image.
667  *
668  *      We should save the page if it isn't Nosave, and is not in the range
669  *      of pages statically defined as 'unsaveable', and it isn't a part of
670  *      a free chunk of pages.
671  */
672
673 static struct page *saveable_page(unsigned long pfn)
674 {
675         struct page *page;
676
677         if (!pfn_valid(pfn))
678                 return NULL;
679
680         page = pfn_to_page(pfn);
681
682         BUG_ON(PageHighMem(page));
683
684         if (PageNosave(page) || PageNosaveFree(page))
685                 return NULL;
686
687         if (PageReserved(page) && pfn_is_nosave(pfn))
688                 return NULL;
689
690         return page;
691 }
692
693 /**
694  *      count_data_pages - compute the total number of saveable non-highmem
695  *      pages.
696  */
697
698 unsigned int count_data_pages(void)
699 {
700         struct zone *zone;
701         unsigned long pfn, max_zone_pfn;
702         unsigned int n = 0;
703
704         for_each_zone(zone) {
705                 if (is_highmem(zone))
706                         continue;
707
708                 mark_free_pages(zone);
709                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
710                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
711                         if(saveable_page(pfn))
712                                 n++;
713         }
714         return n;
715 }
716
717 /* This is needed, because copy_page and memcpy are not usable for copying
718  * task structs.
719  */
720 static inline void do_copy_page(long *dst, long *src)
721 {
722         int n;
723
724         for (n = PAGE_SIZE / sizeof(long); n; n--)
725                 *dst++ = *src++;
726 }
727
728 #ifdef CONFIG_HIGHMEM
729 static inline struct page *
730 page_is_saveable(struct zone *zone, unsigned long pfn)
731 {
732         return is_highmem(zone) ?
733                         saveable_highmem_page(pfn) : saveable_page(pfn);
734 }
735
736 static inline void
737 copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
738 {
739         struct page *s_page, *d_page;
740         void *src, *dst;
741
742         s_page = pfn_to_page(src_pfn);
743         d_page = pfn_to_page(dst_pfn);
744         if (PageHighMem(s_page)) {
745                 src = kmap_atomic(s_page, KM_USER0);
746                 dst = kmap_atomic(d_page, KM_USER1);
747                 do_copy_page(dst, src);
748                 kunmap_atomic(src, KM_USER0);
749                 kunmap_atomic(dst, KM_USER1);
750         } else {
751                 src = page_address(s_page);
752                 if (PageHighMem(d_page)) {
753                         /* Page pointed to by src may contain some kernel
754                          * data modified by kmap_atomic()
755                          */
756                         do_copy_page(buffer, src);
757                         dst = kmap_atomic(pfn_to_page(dst_pfn), KM_USER0);
758                         memcpy(dst, buffer, PAGE_SIZE);
759                         kunmap_atomic(dst, KM_USER0);
760                 } else {
761                         dst = page_address(d_page);
762                         do_copy_page(dst, src);
763                 }
764         }
765 }
766 #else
767 #define page_is_saveable(zone, pfn)     saveable_page(pfn)
768
769 static inline void
770 copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
771 {
772         do_copy_page(page_address(pfn_to_page(dst_pfn)),
773                         page_address(pfn_to_page(src_pfn)));
774 }
775 #endif /* CONFIG_HIGHMEM */
776
777 static void
778 copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm)
779 {
780         struct zone *zone;
781         unsigned long pfn;
782
783         for_each_zone(zone) {
784                 unsigned long max_zone_pfn;
785
786                 mark_free_pages(zone);
787                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
788                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
789                         if (page_is_saveable(zone, pfn))
790                                 memory_bm_set_bit(orig_bm, pfn);
791         }
792         memory_bm_position_reset(orig_bm);
793         memory_bm_position_reset(copy_bm);
794         do {
795                 pfn = memory_bm_next_pfn(orig_bm);
796                 if (likely(pfn != BM_END_OF_MAP))
797                         copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
798         } while (pfn != BM_END_OF_MAP);
799 }
800
801 /* Total number of image pages */
802 static unsigned int nr_copy_pages;
803 /* Number of pages needed for saving the original pfns of the image pages */
804 static unsigned int nr_meta_pages;
805
806 /**
807  *      swsusp_free - free pages allocated for the suspend.
808  *
809  *      Suspend pages are alocated before the atomic copy is made, so we
810  *      need to release them after the resume.
811  */
812
813 void swsusp_free(void)
814 {
815         struct zone *zone;
816         unsigned long pfn, max_zone_pfn;
817
818         for_each_zone(zone) {
819                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
820                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
821                         if (pfn_valid(pfn)) {
822                                 struct page *page = pfn_to_page(pfn);
823
824                                 if (PageNosave(page) && PageNosaveFree(page)) {
825                                         ClearPageNosave(page);
826                                         ClearPageNosaveFree(page);
827                                         __free_page(page);
828                                 }
829                         }
830         }
831         nr_copy_pages = 0;
832         nr_meta_pages = 0;
833         restore_pblist = NULL;
834         buffer = NULL;
835 }
836
837 #ifdef CONFIG_HIGHMEM
838 /**
839   *     count_pages_for_highmem - compute the number of non-highmem pages
840   *     that will be necessary for creating copies of highmem pages.
841   */
842
843 static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
844 {
845         unsigned int free_highmem = count_free_highmem_pages();
846
847         if (free_highmem >= nr_highmem)
848                 nr_highmem = 0;
849         else
850                 nr_highmem -= free_highmem;
851
852         return nr_highmem;
853 }
854 #else
855 static unsigned int
856 count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
857 #endif /* CONFIG_HIGHMEM */
858
859 /**
860  *      enough_free_mem - Make sure we have enough free memory for the
861  *      snapshot image.
862  */
863
864 static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
865 {
866         struct zone *zone;
867         unsigned int free = 0, meta = 0;
868
869         for_each_zone(zone) {
870                 meta += snapshot_additional_pages(zone);
871                 if (!is_highmem(zone))
872                         free += zone->free_pages;
873         }
874
875         nr_pages += count_pages_for_highmem(nr_highmem);
876         pr_debug("swsusp: Normal pages needed: %u + %u + %u, available pages: %u\n",
877                 nr_pages, PAGES_FOR_IO, meta, free);
878
879         return free > nr_pages + PAGES_FOR_IO + meta;
880 }
881
882 #ifdef CONFIG_HIGHMEM
883 /**
884  *      get_highmem_buffer - if there are some highmem pages in the suspend
885  *      image, we may need the buffer to copy them and/or load their data.
886  */
887
888 static inline int get_highmem_buffer(int safe_needed)
889 {
890         buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
891         return buffer ? 0 : -ENOMEM;
892 }
893
894 /**
895  *      alloc_highmem_image_pages - allocate some highmem pages for the image.
896  *      Try to allocate as many pages as needed, but if the number of free
897  *      highmem pages is lesser than that, allocate them all.
898  */
899
900 static inline unsigned int
901 alloc_highmem_image_pages(struct memory_bitmap *bm, unsigned int nr_highmem)
902 {
903         unsigned int to_alloc = count_free_highmem_pages();
904
905         if (to_alloc > nr_highmem)
906                 to_alloc = nr_highmem;
907
908         nr_highmem -= to_alloc;
909         while (to_alloc-- > 0) {
910                 struct page *page;
911
912                 page = alloc_image_page(__GFP_HIGHMEM);
913                 memory_bm_set_bit(bm, page_to_pfn(page));
914         }
915         return nr_highmem;
916 }
917 #else
918 static inline int get_highmem_buffer(int safe_needed) { return 0; }
919
920 static inline unsigned int
921 alloc_highmem_image_pages(struct memory_bitmap *bm, unsigned int n) { return 0; }
922 #endif /* CONFIG_HIGHMEM */
923
924 /**
925  *      swsusp_alloc - allocate memory for the suspend image
926  *
927  *      We first try to allocate as many highmem pages as there are
928  *      saveable highmem pages in the system.  If that fails, we allocate
929  *      non-highmem pages for the copies of the remaining highmem ones.
930  *
931  *      In this approach it is likely that the copies of highmem pages will
932  *      also be located in the high memory, because of the way in which
933  *      copy_data_pages() works.
934  */
935
936 static int
937 swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm,
938                 unsigned int nr_pages, unsigned int nr_highmem)
939 {
940         int error;
941
942         error = memory_bm_create(orig_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY);
943         if (error)
944                 goto Free;
945
946         error = memory_bm_create(copy_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY);
947         if (error)
948                 goto Free;
949
950         if (nr_highmem > 0) {
951                 error = get_highmem_buffer(PG_ANY);
952                 if (error)
953                         goto Free;
954
955                 nr_pages += alloc_highmem_image_pages(copy_bm, nr_highmem);
956         }
957         while (nr_pages-- > 0) {
958                 struct page *page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
959
960                 if (!page)
961                         goto Free;
962
963                 memory_bm_set_bit(copy_bm, page_to_pfn(page));
964         }
965         return 0;
966
967  Free:
968         swsusp_free();
969         return -ENOMEM;
970 }
971
972 /* Memory bitmap used for marking saveable pages (during suspend) or the
973  * suspend image pages (during resume)
974  */
975 static struct memory_bitmap orig_bm;
976 /* Memory bitmap used on suspend for marking allocated pages that will contain
977  * the copies of saveable pages.  During resume it is initially used for
978  * marking the suspend image pages, but then its set bits are duplicated in
979  * @orig_bm and it is released.  Next, on systems with high memory, it may be
980  * used for marking "safe" highmem pages, but it has to be reinitialized for
981  * this purpose.
982  */
983 static struct memory_bitmap copy_bm;
984
985 asmlinkage int swsusp_save(void)
986 {
987         unsigned int nr_pages, nr_highmem;
988
989         printk("swsusp: critical section: \n");
990
991         drain_local_pages();
992         nr_pages = count_data_pages();
993         nr_highmem = count_highmem_pages();
994         printk("swsusp: Need to copy %u pages\n", nr_pages + nr_highmem);
995
996         if (!enough_free_mem(nr_pages, nr_highmem)) {
997                 printk(KERN_ERR "swsusp: Not enough free memory\n");
998                 return -ENOMEM;
999         }
1000
1001         if (swsusp_alloc(&orig_bm, &copy_bm, nr_pages, nr_highmem)) {
1002                 printk(KERN_ERR "swsusp: Memory allocation failed\n");
1003                 return -ENOMEM;
1004         }
1005
1006         /* During allocating of suspend pagedir, new cold pages may appear.
1007          * Kill them.
1008          */
1009         drain_local_pages();
1010         copy_data_pages(&copy_bm, &orig_bm);
1011
1012         /*
1013          * End of critical section. From now on, we can write to memory,
1014          * but we should not touch disk. This specially means we must _not_
1015          * touch swap space! Except we must write out our image of course.
1016          */
1017
1018         nr_pages += nr_highmem;
1019         nr_copy_pages = nr_pages;
1020         nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1021
1022         printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
1023
1024         return 0;
1025 }
1026
1027 static void init_header(struct swsusp_info *info)
1028 {
1029         memset(info, 0, sizeof(struct swsusp_info));
1030         info->version_code = LINUX_VERSION_CODE;
1031         info->num_physpages = num_physpages;
1032         memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
1033         info->cpus = num_online_cpus();
1034         info->image_pages = nr_copy_pages;
1035         info->pages = nr_copy_pages + nr_meta_pages + 1;
1036         info->size = info->pages;
1037         info->size <<= PAGE_SHIFT;
1038 }
1039
1040 /**
1041  *      pack_pfns - pfns corresponding to the set bits found in the bitmap @bm
1042  *      are stored in the array @buf[] (1 page at a time)
1043  */
1044
1045 static inline void
1046 pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
1047 {
1048         int j;
1049
1050         for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1051                 buf[j] = memory_bm_next_pfn(bm);
1052                 if (unlikely(buf[j] == BM_END_OF_MAP))
1053                         break;
1054         }
1055 }
1056
1057 /**
1058  *      snapshot_read_next - used for reading the system memory snapshot.
1059  *
1060  *      On the first call to it @handle should point to a zeroed
1061  *      snapshot_handle structure.  The structure gets updated and a pointer
1062  *      to it should be passed to this function every next time.
1063  *
1064  *      The @count parameter should contain the number of bytes the caller
1065  *      wants to read from the snapshot.  It must not be zero.
1066  *
1067  *      On success the function returns a positive number.  Then, the caller
1068  *      is allowed to read up to the returned number of bytes from the memory
1069  *      location computed by the data_of() macro.  The number returned
1070  *      may be smaller than @count, but this only happens if the read would
1071  *      cross a page boundary otherwise.
1072  *
1073  *      The function returns 0 to indicate the end of data stream condition,
1074  *      and a negative number is returned on error.  In such cases the
1075  *      structure pointed to by @handle is not updated and should not be used
1076  *      any more.
1077  */
1078
1079 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
1080 {
1081         if (handle->cur > nr_meta_pages + nr_copy_pages)
1082                 return 0;
1083
1084         if (!buffer) {
1085                 /* This makes the buffer be freed by swsusp_free() */
1086                 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
1087                 if (!buffer)
1088                         return -ENOMEM;
1089         }
1090         if (!handle->offset) {
1091                 init_header((struct swsusp_info *)buffer);
1092                 handle->buffer = buffer;
1093                 memory_bm_position_reset(&orig_bm);
1094                 memory_bm_position_reset(&copy_bm);
1095         }
1096         if (handle->prev < handle->cur) {
1097                 if (handle->cur <= nr_meta_pages) {
1098                         memset(buffer, 0, PAGE_SIZE);
1099                         pack_pfns(buffer, &orig_bm);
1100                 } else {
1101                         struct page *page;
1102
1103                         page = pfn_to_page(memory_bm_next_pfn(&copy_bm));
1104                         if (PageHighMem(page)) {
1105                                 /* Highmem pages are copied to the buffer,
1106                                  * because we can't return with a kmapped
1107                                  * highmem page (we may not be called again).
1108                                  */
1109                                 void *kaddr;
1110
1111                                 kaddr = kmap_atomic(page, KM_USER0);
1112                                 memcpy(buffer, kaddr, PAGE_SIZE);
1113                                 kunmap_atomic(kaddr, KM_USER0);
1114                                 handle->buffer = buffer;
1115                         } else {
1116                                 handle->buffer = page_address(page);
1117                         }
1118                 }
1119                 handle->prev = handle->cur;
1120         }
1121         handle->buf_offset = handle->cur_offset;
1122         if (handle->cur_offset + count >= PAGE_SIZE) {
1123                 count = PAGE_SIZE - handle->cur_offset;
1124                 handle->cur_offset = 0;
1125                 handle->cur++;
1126         } else {
1127                 handle->cur_offset += count;
1128         }
1129         handle->offset += count;
1130         return count;
1131 }
1132
1133 /**
1134  *      mark_unsafe_pages - mark the pages that cannot be used for storing
1135  *      the image during resume, because they conflict with the pages that
1136  *      had been used before suspend
1137  */
1138
1139 static int mark_unsafe_pages(struct memory_bitmap *bm)
1140 {
1141         struct zone *zone;
1142         unsigned long pfn, max_zone_pfn;
1143
1144         /* Clear page flags */
1145         for_each_zone(zone) {
1146                 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1147                 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1148                         if (pfn_valid(pfn))
1149                                 ClearPageNosaveFree(pfn_to_page(pfn));
1150         }
1151
1152         /* Mark pages that correspond to the "original" pfns as "unsafe" */
1153         memory_bm_position_reset(bm);
1154         do {
1155                 pfn = memory_bm_next_pfn(bm);
1156                 if (likely(pfn != BM_END_OF_MAP)) {
1157                         if (likely(pfn_valid(pfn)))
1158                                 SetPageNosaveFree(pfn_to_page(pfn));
1159                         else
1160                                 return -EFAULT;
1161                 }
1162         } while (pfn != BM_END_OF_MAP);
1163
1164         allocated_unsafe_pages = 0;
1165
1166         return 0;
1167 }
1168
1169 static void
1170 duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src)
1171 {
1172         unsigned long pfn;
1173
1174         memory_bm_position_reset(src);
1175         pfn = memory_bm_next_pfn(src);
1176         while (pfn != BM_END_OF_MAP) {
1177                 memory_bm_set_bit(dst, pfn);
1178                 pfn = memory_bm_next_pfn(src);
1179         }
1180 }
1181
1182 static inline int check_header(struct swsusp_info *info)
1183 {
1184         char *reason = NULL;
1185
1186         if (info->version_code != LINUX_VERSION_CODE)
1187                 reason = "kernel version";
1188         if (info->num_physpages != num_physpages)
1189                 reason = "memory size";
1190         if (strcmp(info->uts.sysname,init_utsname()->sysname))
1191                 reason = "system type";
1192         if (strcmp(info->uts.release,init_utsname()->release))
1193                 reason = "kernel release";
1194         if (strcmp(info->uts.version,init_utsname()->version))
1195                 reason = "version";
1196         if (strcmp(info->uts.machine,init_utsname()->machine))
1197                 reason = "machine";
1198         if (reason) {
1199                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
1200                 return -EPERM;
1201         }
1202         return 0;
1203 }
1204
1205 /**
1206  *      load header - check the image header and copy data from it
1207  */
1208
1209 static int
1210 load_header(struct swsusp_info *info)
1211 {
1212         int error;
1213
1214         restore_pblist = NULL;
1215         error = check_header(info);
1216         if (!error) {
1217                 nr_copy_pages = info->image_pages;
1218                 nr_meta_pages = info->pages - info->image_pages - 1;
1219         }
1220         return error;
1221 }
1222
1223 /**
1224  *      unpack_orig_pfns - for each element of @buf[] (1 page at a time) set
1225  *      the corresponding bit in the memory bitmap @bm
1226  */
1227
1228 static inline void
1229 unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
1230 {
1231         int j;
1232
1233         for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1234                 if (unlikely(buf[j] == BM_END_OF_MAP))
1235                         break;
1236
1237                 memory_bm_set_bit(bm, buf[j]);
1238         }
1239 }
1240
1241 /* List of "safe" pages that may be used to store data loaded from the suspend
1242  * image
1243  */
1244 static struct linked_page *safe_pages_list;
1245
1246 #ifdef CONFIG_HIGHMEM
1247 /* struct highmem_pbe is used for creating the list of highmem pages that
1248  * should be restored atomically during the resume from disk, because the page
1249  * frames they have occupied before the suspend are in use.
1250  */
1251 struct highmem_pbe {
1252         struct page *copy_page; /* data is here now */
1253         struct page *orig_page; /* data was here before the suspend */
1254         struct highmem_pbe *next;
1255 };
1256
1257 /* List of highmem PBEs needed for restoring the highmem pages that were
1258  * allocated before the suspend and included in the suspend image, but have
1259  * also been allocated by the "resume" kernel, so their contents cannot be
1260  * written directly to their "original" page frames.
1261  */
1262 static struct highmem_pbe *highmem_pblist;
1263
1264 /**
1265  *      count_highmem_image_pages - compute the number of highmem pages in the
1266  *      suspend image.  The bits in the memory bitmap @bm that correspond to the
1267  *      image pages are assumed to be set.
1268  */
1269
1270 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
1271 {
1272         unsigned long pfn;
1273         unsigned int cnt = 0;
1274
1275         memory_bm_position_reset(bm);
1276         pfn = memory_bm_next_pfn(bm);
1277         while (pfn != BM_END_OF_MAP) {
1278                 if (PageHighMem(pfn_to_page(pfn)))
1279                         cnt++;
1280
1281                 pfn = memory_bm_next_pfn(bm);
1282         }
1283         return cnt;
1284 }
1285
1286 /**
1287  *      prepare_highmem_image - try to allocate as many highmem pages as
1288  *      there are highmem image pages (@nr_highmem_p points to the variable
1289  *      containing the number of highmem image pages).  The pages that are
1290  *      "safe" (ie. will not be overwritten when the suspend image is
1291  *      restored) have the corresponding bits set in @bm (it must be
1292  *      unitialized).
1293  *
1294  *      NOTE: This function should not be called if there are no highmem
1295  *      image pages.
1296  */
1297
1298 static unsigned int safe_highmem_pages;
1299
1300 static struct memory_bitmap *safe_highmem_bm;
1301
1302 static int
1303 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
1304 {
1305         unsigned int to_alloc;
1306
1307         if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
1308                 return -ENOMEM;
1309
1310         if (get_highmem_buffer(PG_SAFE))
1311                 return -ENOMEM;
1312
1313         to_alloc = count_free_highmem_pages();
1314         if (to_alloc > *nr_highmem_p)
1315                 to_alloc = *nr_highmem_p;
1316         else
1317                 *nr_highmem_p = to_alloc;
1318
1319         safe_highmem_pages = 0;
1320         while (to_alloc-- > 0) {
1321                 struct page *page;
1322
1323                 page = alloc_page(__GFP_HIGHMEM);
1324                 if (!PageNosaveFree(page)) {
1325                         /* The page is "safe", set its bit the bitmap */
1326                         memory_bm_set_bit(bm, page_to_pfn(page));
1327                         safe_highmem_pages++;
1328                 }
1329                 /* Mark the page as allocated */
1330                 SetPageNosave(page);
1331                 SetPageNosaveFree(page);
1332         }
1333         memory_bm_position_reset(bm);
1334         safe_highmem_bm = bm;
1335         return 0;
1336 }
1337
1338 /**
1339  *      get_highmem_page_buffer - for given highmem image page find the buffer
1340  *      that suspend_write_next() should set for its caller to write to.
1341  *
1342  *      If the page is to be saved to its "original" page frame or a copy of
1343  *      the page is to be made in the highmem, @buffer is returned.  Otherwise,
1344  *      the copy of the page is to be made in normal memory, so the address of
1345  *      the copy is returned.
1346  *
1347  *      If @buffer is returned, the caller of suspend_write_next() will write
1348  *      the page's contents to @buffer, so they will have to be copied to the
1349  *      right location on the next call to suspend_write_next() and it is done
1350  *      with the help of copy_last_highmem_page().  For this purpose, if
1351  *      @buffer is returned, @last_highmem page is set to the page to which
1352  *      the data will have to be copied from @buffer.
1353  */
1354
1355 static struct page *last_highmem_page;
1356
1357 static void *
1358 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
1359 {
1360         struct highmem_pbe *pbe;
1361         void *kaddr;
1362
1363         if (PageNosave(page) && PageNosaveFree(page)) {
1364                 /* We have allocated the "original" page frame and we can
1365                  * use it directly to store the loaded page.
1366                  */
1367                 last_highmem_page = page;
1368                 return buffer;
1369         }
1370         /* The "original" page frame has not been allocated and we have to
1371          * use a "safe" page frame to store the loaded page.
1372          */
1373         pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
1374         if (!pbe) {
1375                 swsusp_free();
1376                 return NULL;
1377         }
1378         pbe->orig_page = page;
1379         if (safe_highmem_pages > 0) {
1380                 struct page *tmp;
1381
1382                 /* Copy of the page will be stored in high memory */
1383                 kaddr = buffer;
1384                 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
1385                 safe_highmem_pages--;
1386                 last_highmem_page = tmp;
1387                 pbe->copy_page = tmp;
1388         } else {
1389                 /* Copy of the page will be stored in normal memory */
1390                 kaddr = safe_pages_list;
1391                 safe_pages_list = safe_pages_list->next;
1392                 pbe->copy_page = virt_to_page(kaddr);
1393         }
1394         pbe->next = highmem_pblist;
1395         highmem_pblist = pbe;
1396         return kaddr;
1397 }
1398
1399 /**
1400  *      copy_last_highmem_page - copy the contents of a highmem image from
1401  *      @buffer, where the caller of snapshot_write_next() has place them,
1402  *      to the right location represented by @last_highmem_page .
1403  */
1404
1405 static void copy_last_highmem_page(void)
1406 {
1407         if (last_highmem_page) {
1408                 void *dst;
1409
1410                 dst = kmap_atomic(last_highmem_page, KM_USER0);
1411                 memcpy(dst, buffer, PAGE_SIZE);
1412                 kunmap_atomic(dst, KM_USER0);
1413                 last_highmem_page = NULL;
1414         }
1415 }
1416
1417 static inline int last_highmem_page_copied(void)
1418 {
1419         return !last_highmem_page;
1420 }
1421
1422 static inline void free_highmem_data(void)
1423 {
1424         if (safe_highmem_bm)
1425                 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
1426
1427         if (buffer)
1428                 free_image_page(buffer, PG_UNSAFE_CLEAR);
1429 }
1430 #else
1431 static inline int get_safe_write_buffer(void) { return 0; }
1432
1433 static unsigned int
1434 count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
1435
1436 static inline int
1437 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
1438 {
1439         return 0;
1440 }
1441
1442 static inline void *
1443 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
1444 {
1445         return NULL;
1446 }
1447
1448 static inline void copy_last_highmem_page(void) {}
1449 static inline int last_highmem_page_copied(void) { return 1; }
1450 static inline void free_highmem_data(void) {}
1451 #endif /* CONFIG_HIGHMEM */
1452
1453 /**
1454  *      prepare_image - use the memory bitmap @bm to mark the pages that will
1455  *      be overwritten in the process of restoring the system memory state
1456  *      from the suspend image ("unsafe" pages) and allocate memory for the
1457  *      image.
1458  *
1459  *      The idea is to allocate a new memory bitmap first and then allocate
1460  *      as many pages as needed for the image data, but not to assign these
1461  *      pages to specific tasks initially.  Instead, we just mark them as
1462  *      allocated and create a lists of "safe" pages that will be used
1463  *      later.  On systems with high memory a list of "safe" highmem pages is
1464  *      also created.
1465  */
1466
1467 #define PBES_PER_LINKED_PAGE    (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
1468
1469 static int
1470 prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
1471 {
1472         unsigned int nr_pages, nr_highmem;
1473         struct linked_page *sp_list, *lp;
1474         int error;
1475
1476         /* If there is no highmem, the buffer will not be necessary */
1477         free_image_page(buffer, PG_UNSAFE_CLEAR);
1478         buffer = NULL;
1479
1480         nr_highmem = count_highmem_image_pages(bm);
1481         error = mark_unsafe_pages(bm);
1482         if (error)
1483                 goto Free;
1484
1485         error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
1486         if (error)
1487                 goto Free;
1488
1489         duplicate_memory_bitmap(new_bm, bm);
1490         memory_bm_free(bm, PG_UNSAFE_KEEP);
1491         if (nr_highmem > 0) {
1492                 error = prepare_highmem_image(bm, &nr_highmem);
1493                 if (error)
1494                         goto Free;
1495         }
1496         /* Reserve some safe pages for potential later use.
1497          *
1498          * NOTE: This way we make sure there will be enough safe pages for the
1499          * chain_alloc() in get_buffer().  It is a bit wasteful, but
1500          * nr_copy_pages cannot be greater than 50% of the memory anyway.
1501          */
1502         sp_list = NULL;
1503         /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */
1504         nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
1505         nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
1506         while (nr_pages > 0) {
1507                 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
1508                 if (!lp) {
1509                         error = -ENOMEM;
1510                         goto Free;
1511                 }
1512                 lp->next = sp_list;
1513                 sp_list = lp;
1514                 nr_pages--;
1515         }
1516         /* Preallocate memory for the image */
1517         safe_pages_list = NULL;
1518         nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
1519         while (nr_pages > 0) {
1520                 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
1521                 if (!lp) {
1522                         error = -ENOMEM;
1523                         goto Free;
1524                 }
1525                 if (!PageNosaveFree(virt_to_page(lp))) {
1526                         /* The page is "safe", add it to the list */
1527                         lp->next = safe_pages_list;
1528                         safe_pages_list = lp;
1529                 }
1530                 /* Mark the page as allocated */
1531                 SetPageNosave(virt_to_page(lp));
1532                 SetPageNosaveFree(virt_to_page(lp));
1533                 nr_pages--;
1534         }
1535         /* Free the reserved safe pages so that chain_alloc() can use them */
1536         while (sp_list) {
1537                 lp = sp_list->next;
1538                 free_image_page(sp_list, PG_UNSAFE_CLEAR);
1539                 sp_list = lp;
1540         }
1541         return 0;
1542
1543  Free:
1544         swsusp_free();
1545         return error;
1546 }
1547
1548 /**
1549  *      get_buffer - compute the address that snapshot_write_next() should
1550  *      set for its caller to write to.
1551  */
1552
1553 static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
1554 {
1555         struct pbe *pbe;
1556         struct page *page = pfn_to_page(memory_bm_next_pfn(bm));
1557
1558         if (PageHighMem(page))
1559                 return get_highmem_page_buffer(page, ca);
1560
1561         if (PageNosave(page) && PageNosaveFree(page))
1562                 /* We have allocated the "original" page frame and we can
1563                  * use it directly to store the loaded page.
1564                  */
1565                 return page_address(page);
1566
1567         /* The "original" page frame has not been allocated and we have to
1568          * use a "safe" page frame to store the loaded page.
1569          */
1570         pbe = chain_alloc(ca, sizeof(struct pbe));
1571         if (!pbe) {
1572                 swsusp_free();
1573                 return NULL;
1574         }
1575         pbe->orig_address = page_address(page);
1576         pbe->address = safe_pages_list;
1577         safe_pages_list = safe_pages_list->next;
1578         pbe->next = restore_pblist;
1579         restore_pblist = pbe;
1580         return pbe->address;
1581 }
1582
1583 /**
1584  *      snapshot_write_next - used for writing the system memory snapshot.
1585  *
1586  *      On the first call to it @handle should point to a zeroed
1587  *      snapshot_handle structure.  The structure gets updated and a pointer
1588  *      to it should be passed to this function every next time.
1589  *
1590  *      The @count parameter should contain the number of bytes the caller
1591  *      wants to write to the image.  It must not be zero.
1592  *
1593  *      On success the function returns a positive number.  Then, the caller
1594  *      is allowed to write up to the returned number of bytes to the memory
1595  *      location computed by the data_of() macro.  The number returned
1596  *      may be smaller than @count, but this only happens if the write would
1597  *      cross a page boundary otherwise.
1598  *
1599  *      The function returns 0 to indicate the "end of file" condition,
1600  *      and a negative number is returned on error.  In such cases the
1601  *      structure pointed to by @handle is not updated and should not be used
1602  *      any more.
1603  */
1604
1605 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
1606 {
1607         static struct chain_allocator ca;
1608         int error = 0;
1609
1610         /* Check if we have already loaded the entire image */
1611         if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages)
1612                 return 0;
1613
1614         if (handle->offset == 0) {
1615                 if (!buffer)
1616                         /* This makes the buffer be freed by swsusp_free() */
1617                         buffer = get_image_page(GFP_ATOMIC, PG_ANY);
1618
1619                 if (!buffer)
1620                         return -ENOMEM;
1621
1622                 handle->buffer = buffer;
1623         }
1624         handle->sync_read = 1;
1625         if (handle->prev < handle->cur) {
1626                 if (handle->prev == 0) {
1627                         error = load_header(buffer);
1628                         if (error)
1629                                 return error;
1630
1631                         error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
1632                         if (error)
1633                                 return error;
1634
1635                 } else if (handle->prev <= nr_meta_pages) {
1636                         unpack_orig_pfns(buffer, &copy_bm);
1637                         if (handle->prev == nr_meta_pages) {
1638                                 error = prepare_image(&orig_bm, &copy_bm);
1639                                 if (error)
1640                                         return error;
1641
1642                                 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
1643                                 memory_bm_position_reset(&orig_bm);
1644                                 restore_pblist = NULL;
1645                                 handle->buffer = get_buffer(&orig_bm, &ca);
1646                                 handle->sync_read = 0;
1647                                 if (!handle->buffer)
1648                                         return -ENOMEM;
1649                         }
1650                 } else {
1651                         copy_last_highmem_page();
1652                         handle->buffer = get_buffer(&orig_bm, &ca);
1653                         if (handle->buffer != buffer)
1654                                 handle->sync_read = 0;
1655                 }
1656                 handle->prev = handle->cur;
1657         }
1658         handle->buf_offset = handle->cur_offset;
1659         if (handle->cur_offset + count >= PAGE_SIZE) {
1660                 count = PAGE_SIZE - handle->cur_offset;
1661                 handle->cur_offset = 0;
1662                 handle->cur++;
1663         } else {
1664                 handle->cur_offset += count;
1665         }
1666         handle->offset += count;
1667         return count;
1668 }
1669
1670 /**
1671  *      snapshot_write_finalize - must be called after the last call to
1672  *      snapshot_write_next() in case the last page in the image happens
1673  *      to be a highmem page and its contents should be stored in the
1674  *      highmem.  Additionally, it releases the memory that will not be
1675  *      used any more.
1676  */
1677
1678 void snapshot_write_finalize(struct snapshot_handle *handle)
1679 {
1680         copy_last_highmem_page();
1681         /* Free only if we have loaded the image entirely */
1682         if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages) {
1683                 memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR);
1684                 free_highmem_data();
1685         }
1686 }
1687
1688 int snapshot_image_loaded(struct snapshot_handle *handle)
1689 {
1690         return !(!nr_copy_pages || !last_highmem_page_copied() ||
1691                         handle->cur <= nr_meta_pages + nr_copy_pages);
1692 }
1693
1694 #ifdef CONFIG_HIGHMEM
1695 /* Assumes that @buf is ready and points to a "safe" page */
1696 static inline void
1697 swap_two_pages_data(struct page *p1, struct page *p2, void *buf)
1698 {
1699         void *kaddr1, *kaddr2;
1700
1701         kaddr1 = kmap_atomic(p1, KM_USER0);
1702         kaddr2 = kmap_atomic(p2, KM_USER1);
1703         memcpy(buf, kaddr1, PAGE_SIZE);
1704         memcpy(kaddr1, kaddr2, PAGE_SIZE);
1705         memcpy(kaddr2, buf, PAGE_SIZE);
1706         kunmap_atomic(kaddr1, KM_USER0);
1707         kunmap_atomic(kaddr2, KM_USER1);
1708 }
1709
1710 /**
1711  *      restore_highmem - for each highmem page that was allocated before
1712  *      the suspend and included in the suspend image, and also has been
1713  *      allocated by the "resume" kernel swap its current (ie. "before
1714  *      resume") contents with the previous (ie. "before suspend") one.
1715  *
1716  *      If the resume eventually fails, we can call this function once
1717  *      again and restore the "before resume" highmem state.
1718  */
1719
1720 int restore_highmem(void)
1721 {
1722         struct highmem_pbe *pbe = highmem_pblist;
1723         void *buf;
1724
1725         if (!pbe)
1726                 return 0;
1727
1728         buf = get_image_page(GFP_ATOMIC, PG_SAFE);
1729         if (!buf)
1730                 return -ENOMEM;
1731
1732         while (pbe) {
1733                 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
1734                 pbe = pbe->next;
1735         }
1736         free_image_page(buf, PG_UNSAFE_CLEAR);
1737         return 0;
1738 }
1739 #endif /* CONFIG_HIGHMEM */