Hibernation: Arbitrary boot kernel support - generic code
[deliverable/linux.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@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/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
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 static int swsusp_page_is_free(struct page *);
38 static void swsusp_set_page_forbidden(struct page *);
39 static void swsusp_unset_page_forbidden(struct page *);
40
41 /* List of PBEs needed for restoring the pages that were allocated before
42 * the suspend and included in the suspend image, but have also been
43 * allocated by the "resume" kernel, so their contents cannot be written
44 * directly to their "original" page frames.
45 */
46 struct pbe *restore_pblist;
47
48 /* Pointer to an auxiliary buffer (1 page) */
49 static void *buffer;
50
51 /**
52 * @safe_needed - on resume, for storing the PBE list and the image,
53 * we can only use memory pages that do not conflict with the pages
54 * used before suspend. The unsafe pages have PageNosaveFree set
55 * and we count them using unsafe_pages.
56 *
57 * Each allocated image page is marked as PageNosave and PageNosaveFree
58 * so that swsusp_free() can release it.
59 */
60
61 #define PG_ANY 0
62 #define PG_SAFE 1
63 #define PG_UNSAFE_CLEAR 1
64 #define PG_UNSAFE_KEEP 0
65
66 static unsigned int allocated_unsafe_pages;
67
68 static void *get_image_page(gfp_t gfp_mask, int safe_needed)
69 {
70 void *res;
71
72 res = (void *)get_zeroed_page(gfp_mask);
73 if (safe_needed)
74 while (res && swsusp_page_is_free(virt_to_page(res))) {
75 /* The page is unsafe, mark it for swsusp_free() */
76 swsusp_set_page_forbidden(virt_to_page(res));
77 allocated_unsafe_pages++;
78 res = (void *)get_zeroed_page(gfp_mask);
79 }
80 if (res) {
81 swsusp_set_page_forbidden(virt_to_page(res));
82 swsusp_set_page_free(virt_to_page(res));
83 }
84 return res;
85 }
86
87 unsigned long get_safe_page(gfp_t gfp_mask)
88 {
89 return (unsigned long)get_image_page(gfp_mask, PG_SAFE);
90 }
91
92 static struct page *alloc_image_page(gfp_t gfp_mask)
93 {
94 struct page *page;
95
96 page = alloc_page(gfp_mask);
97 if (page) {
98 swsusp_set_page_forbidden(page);
99 swsusp_set_page_free(page);
100 }
101 return page;
102 }
103
104 /**
105 * free_image_page - free page represented by @addr, allocated with
106 * get_image_page (page flags set by it must be cleared)
107 */
108
109 static inline void free_image_page(void *addr, int clear_nosave_free)
110 {
111 struct page *page;
112
113 BUG_ON(!virt_addr_valid(addr));
114
115 page = virt_to_page(addr);
116
117 swsusp_unset_page_forbidden(page);
118 if (clear_nosave_free)
119 swsusp_unset_page_free(page);
120
121 __free_page(page);
122 }
123
124 /* struct linked_page is used to build chains of pages */
125
126 #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
127
128 struct linked_page {
129 struct linked_page *next;
130 char data[LINKED_PAGE_DATA_SIZE];
131 } __attribute__((packed));
132
133 static inline void
134 free_list_of_pages(struct linked_page *list, int clear_page_nosave)
135 {
136 while (list) {
137 struct linked_page *lp = list->next;
138
139 free_image_page(list, clear_page_nosave);
140 list = lp;
141 }
142 }
143
144 /**
145 * struct chain_allocator is used for allocating small objects out of
146 * a linked list of pages called 'the chain'.
147 *
148 * The chain grows each time when there is no room for a new object in
149 * the current page. The allocated objects cannot be freed individually.
150 * It is only possible to free them all at once, by freeing the entire
151 * chain.
152 *
153 * NOTE: The chain allocator may be inefficient if the allocated objects
154 * are not much smaller than PAGE_SIZE.
155 */
156
157 struct chain_allocator {
158 struct linked_page *chain; /* the chain */
159 unsigned int used_space; /* total size of objects allocated out
160 * of the current page
161 */
162 gfp_t gfp_mask; /* mask for allocating pages */
163 int safe_needed; /* if set, only "safe" pages are allocated */
164 };
165
166 static void
167 chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed)
168 {
169 ca->chain = NULL;
170 ca->used_space = LINKED_PAGE_DATA_SIZE;
171 ca->gfp_mask = gfp_mask;
172 ca->safe_needed = safe_needed;
173 }
174
175 static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
176 {
177 void *ret;
178
179 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
180 struct linked_page *lp;
181
182 lp = get_image_page(ca->gfp_mask, ca->safe_needed);
183 if (!lp)
184 return NULL;
185
186 lp->next = ca->chain;
187 ca->chain = lp;
188 ca->used_space = 0;
189 }
190 ret = ca->chain->data + ca->used_space;
191 ca->used_space += size;
192 return ret;
193 }
194
195 static void chain_free(struct chain_allocator *ca, int clear_page_nosave)
196 {
197 free_list_of_pages(ca->chain, clear_page_nosave);
198 memset(ca, 0, sizeof(struct chain_allocator));
199 }
200
201 /**
202 * Data types related to memory bitmaps.
203 *
204 * Memory bitmap is a structure consiting of many linked lists of
205 * objects. The main list's elements are of type struct zone_bitmap
206 * and each of them corresonds to one zone. For each zone bitmap
207 * object there is a list of objects of type struct bm_block that
208 * represent each blocks of bit chunks in which information is
209 * stored.
210 *
211 * struct memory_bitmap contains a pointer to the main list of zone
212 * bitmap objects, a struct bm_position used for browsing the bitmap,
213 * and a pointer to the list of pages used for allocating all of the
214 * zone bitmap objects and bitmap block objects.
215 *
216 * NOTE: It has to be possible to lay out the bitmap in memory
217 * using only allocations of order 0. Additionally, the bitmap is
218 * designed to work with arbitrary number of zones (this is over the
219 * top for now, but let's avoid making unnecessary assumptions ;-).
220 *
221 * struct zone_bitmap contains a pointer to a list of bitmap block
222 * objects and a pointer to the bitmap block object that has been
223 * most recently used for setting bits. Additionally, it contains the
224 * pfns that correspond to the start and end of the represented zone.
225 *
226 * struct bm_block contains a pointer to the memory page in which
227 * information is stored (in the form of a block of bit chunks
228 * of type unsigned long each). It also contains the pfns that
229 * correspond to the start and end of the represented memory area and
230 * the number of bit chunks in the block.
231 */
232
233 #define BM_END_OF_MAP (~0UL)
234
235 #define BM_CHUNKS_PER_BLOCK (PAGE_SIZE / sizeof(long))
236 #define BM_BITS_PER_CHUNK (sizeof(long) << 3)
237 #define BM_BITS_PER_BLOCK (PAGE_SIZE << 3)
238
239 struct bm_block {
240 struct bm_block *next; /* next element of the list */
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 int size; /* number of bit chunks */
244 unsigned long *data; /* chunks of bits representing pages */
245 };
246
247 struct zone_bitmap {
248 struct zone_bitmap *next; /* next element of the list */
249 unsigned long start_pfn; /* minimal pfn in this zone */
250 unsigned long end_pfn; /* maximal pfn in this zone plus 1 */
251 struct bm_block *bm_blocks; /* list of bitmap blocks */
252 struct bm_block *cur_block; /* recently used bitmap block */
253 };
254
255 /* strcut bm_position is used for browsing memory bitmaps */
256
257 struct bm_position {
258 struct zone_bitmap *zone_bm;
259 struct bm_block *block;
260 int chunk;
261 int bit;
262 };
263
264 struct memory_bitmap {
265 struct zone_bitmap *zone_bm_list; /* list of zone bitmaps */
266 struct linked_page *p_list; /* list of pages used to store zone
267 * bitmap objects and bitmap block
268 * objects
269 */
270 struct bm_position cur; /* most recently used bit position */
271 };
272
273 /* Functions that operate on memory bitmaps */
274
275 static inline void memory_bm_reset_chunk(struct memory_bitmap *bm)
276 {
277 bm->cur.chunk = 0;
278 bm->cur.bit = -1;
279 }
280
281 static void memory_bm_position_reset(struct memory_bitmap *bm)
282 {
283 struct zone_bitmap *zone_bm;
284
285 zone_bm = bm->zone_bm_list;
286 bm->cur.zone_bm = zone_bm;
287 bm->cur.block = zone_bm->bm_blocks;
288 memory_bm_reset_chunk(bm);
289 }
290
291 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
292
293 /**
294 * create_bm_block_list - create a list of block bitmap objects
295 */
296
297 static inline struct bm_block *
298 create_bm_block_list(unsigned int nr_blocks, struct chain_allocator *ca)
299 {
300 struct bm_block *bblist = NULL;
301
302 while (nr_blocks-- > 0) {
303 struct bm_block *bb;
304
305 bb = chain_alloc(ca, sizeof(struct bm_block));
306 if (!bb)
307 return NULL;
308
309 bb->next = bblist;
310 bblist = bb;
311 }
312 return bblist;
313 }
314
315 /**
316 * create_zone_bm_list - create a list of zone bitmap objects
317 */
318
319 static inline struct zone_bitmap *
320 create_zone_bm_list(unsigned int nr_zones, struct chain_allocator *ca)
321 {
322 struct zone_bitmap *zbmlist = NULL;
323
324 while (nr_zones-- > 0) {
325 struct zone_bitmap *zbm;
326
327 zbm = chain_alloc(ca, sizeof(struct zone_bitmap));
328 if (!zbm)
329 return NULL;
330
331 zbm->next = zbmlist;
332 zbmlist = zbm;
333 }
334 return zbmlist;
335 }
336
337 /**
338 * memory_bm_create - allocate memory for a memory bitmap
339 */
340
341 static int
342 memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed)
343 {
344 struct chain_allocator ca;
345 struct zone *zone;
346 struct zone_bitmap *zone_bm;
347 struct bm_block *bb;
348 unsigned int nr;
349
350 chain_init(&ca, gfp_mask, safe_needed);
351
352 /* Compute the number of zones */
353 nr = 0;
354 for_each_zone(zone)
355 if (populated_zone(zone))
356 nr++;
357
358 /* Allocate the list of zones bitmap objects */
359 zone_bm = create_zone_bm_list(nr, &ca);
360 bm->zone_bm_list = zone_bm;
361 if (!zone_bm) {
362 chain_free(&ca, PG_UNSAFE_CLEAR);
363 return -ENOMEM;
364 }
365
366 /* Initialize the zone bitmap objects */
367 for_each_zone(zone) {
368 unsigned long pfn;
369
370 if (!populated_zone(zone))
371 continue;
372
373 zone_bm->start_pfn = zone->zone_start_pfn;
374 zone_bm->end_pfn = zone->zone_start_pfn + zone->spanned_pages;
375 /* Allocate the list of bitmap block objects */
376 nr = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
377 bb = create_bm_block_list(nr, &ca);
378 zone_bm->bm_blocks = bb;
379 zone_bm->cur_block = bb;
380 if (!bb)
381 goto Free;
382
383 nr = zone->spanned_pages;
384 pfn = zone->zone_start_pfn;
385 /* Initialize the bitmap block objects */
386 while (bb) {
387 unsigned long *ptr;
388
389 ptr = get_image_page(gfp_mask, safe_needed);
390 bb->data = ptr;
391 if (!ptr)
392 goto Free;
393
394 bb->start_pfn = pfn;
395 if (nr >= BM_BITS_PER_BLOCK) {
396 pfn += BM_BITS_PER_BLOCK;
397 bb->size = BM_CHUNKS_PER_BLOCK;
398 nr -= BM_BITS_PER_BLOCK;
399 } else {
400 /* This is executed only once in the loop */
401 pfn += nr;
402 bb->size = DIV_ROUND_UP(nr, BM_BITS_PER_CHUNK);
403 }
404 bb->end_pfn = pfn;
405 bb = bb->next;
406 }
407 zone_bm = zone_bm->next;
408 }
409 bm->p_list = ca.chain;
410 memory_bm_position_reset(bm);
411 return 0;
412
413 Free:
414 bm->p_list = ca.chain;
415 memory_bm_free(bm, PG_UNSAFE_CLEAR);
416 return -ENOMEM;
417 }
418
419 /**
420 * memory_bm_free - free memory occupied by the memory bitmap @bm
421 */
422
423 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
424 {
425 struct zone_bitmap *zone_bm;
426
427 /* Free the list of bit blocks for each zone_bitmap object */
428 zone_bm = bm->zone_bm_list;
429 while (zone_bm) {
430 struct bm_block *bb;
431
432 bb = zone_bm->bm_blocks;
433 while (bb) {
434 if (bb->data)
435 free_image_page(bb->data, clear_nosave_free);
436 bb = bb->next;
437 }
438 zone_bm = zone_bm->next;
439 }
440 free_list_of_pages(bm->p_list, clear_nosave_free);
441 bm->zone_bm_list = NULL;
442 }
443
444 /**
445 * memory_bm_find_bit - find the bit in the bitmap @bm that corresponds
446 * to given pfn. The cur_zone_bm member of @bm and the cur_block member
447 * of @bm->cur_zone_bm are updated.
448 */
449
450 static void memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
451 void **addr, unsigned int *bit_nr)
452 {
453 struct zone_bitmap *zone_bm;
454 struct bm_block *bb;
455
456 /* Check if the pfn is from the current zone */
457 zone_bm = bm->cur.zone_bm;
458 if (pfn < zone_bm->start_pfn || pfn >= zone_bm->end_pfn) {
459 zone_bm = bm->zone_bm_list;
460 /* We don't assume that the zones are sorted by pfns */
461 while (pfn < zone_bm->start_pfn || pfn >= zone_bm->end_pfn) {
462 zone_bm = zone_bm->next;
463
464 BUG_ON(!zone_bm);
465 }
466 bm->cur.zone_bm = zone_bm;
467 }
468 /* Check if the pfn corresponds to the current bitmap block */
469 bb = zone_bm->cur_block;
470 if (pfn < bb->start_pfn)
471 bb = zone_bm->bm_blocks;
472
473 while (pfn >= bb->end_pfn) {
474 bb = bb->next;
475
476 BUG_ON(!bb);
477 }
478 zone_bm->cur_block = bb;
479 pfn -= bb->start_pfn;
480 *bit_nr = pfn % BM_BITS_PER_CHUNK;
481 *addr = bb->data + pfn / BM_BITS_PER_CHUNK;
482 }
483
484 static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
485 {
486 void *addr;
487 unsigned int bit;
488
489 memory_bm_find_bit(bm, pfn, &addr, &bit);
490 set_bit(bit, addr);
491 }
492
493 static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
494 {
495 void *addr;
496 unsigned int bit;
497
498 memory_bm_find_bit(bm, pfn, &addr, &bit);
499 clear_bit(bit, addr);
500 }
501
502 static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
503 {
504 void *addr;
505 unsigned int bit;
506
507 memory_bm_find_bit(bm, pfn, &addr, &bit);
508 return test_bit(bit, addr);
509 }
510
511 /* Two auxiliary functions for memory_bm_next_pfn */
512
513 /* Find the first set bit in the given chunk, if there is one */
514
515 static inline int next_bit_in_chunk(int bit, unsigned long *chunk_p)
516 {
517 bit++;
518 while (bit < BM_BITS_PER_CHUNK) {
519 if (test_bit(bit, chunk_p))
520 return bit;
521
522 bit++;
523 }
524 return -1;
525 }
526
527 /* Find a chunk containing some bits set in given block of bits */
528
529 static inline int next_chunk_in_block(int n, struct bm_block *bb)
530 {
531 n++;
532 while (n < bb->size) {
533 if (bb->data[n])
534 return n;
535
536 n++;
537 }
538 return -1;
539 }
540
541 /**
542 * memory_bm_next_pfn - find the pfn that corresponds to the next set bit
543 * in the bitmap @bm. If the pfn cannot be found, BM_END_OF_MAP is
544 * returned.
545 *
546 * It is required to run memory_bm_position_reset() before the first call to
547 * this function.
548 */
549
550 static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
551 {
552 struct zone_bitmap *zone_bm;
553 struct bm_block *bb;
554 int chunk;
555 int bit;
556
557 do {
558 bb = bm->cur.block;
559 do {
560 chunk = bm->cur.chunk;
561 bit = bm->cur.bit;
562 do {
563 bit = next_bit_in_chunk(bit, bb->data + chunk);
564 if (bit >= 0)
565 goto Return_pfn;
566
567 chunk = next_chunk_in_block(chunk, bb);
568 bit = -1;
569 } while (chunk >= 0);
570 bb = bb->next;
571 bm->cur.block = bb;
572 memory_bm_reset_chunk(bm);
573 } while (bb);
574 zone_bm = bm->cur.zone_bm->next;
575 if (zone_bm) {
576 bm->cur.zone_bm = zone_bm;
577 bm->cur.block = zone_bm->bm_blocks;
578 memory_bm_reset_chunk(bm);
579 }
580 } while (zone_bm);
581 memory_bm_position_reset(bm);
582 return BM_END_OF_MAP;
583
584 Return_pfn:
585 bm->cur.chunk = chunk;
586 bm->cur.bit = bit;
587 return bb->start_pfn + chunk * BM_BITS_PER_CHUNK + bit;
588 }
589
590 /**
591 * This structure represents a range of page frames the contents of which
592 * should not be saved during the suspend.
593 */
594
595 struct nosave_region {
596 struct list_head list;
597 unsigned long start_pfn;
598 unsigned long end_pfn;
599 };
600
601 static LIST_HEAD(nosave_regions);
602
603 /**
604 * register_nosave_region - register a range of page frames the contents
605 * of which should not be saved during the suspend (to be used in the early
606 * initialization code)
607 */
608
609 void __init
610 __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn,
611 int use_kmalloc)
612 {
613 struct nosave_region *region;
614
615 if (start_pfn >= end_pfn)
616 return;
617
618 if (!list_empty(&nosave_regions)) {
619 /* Try to extend the previous region (they should be sorted) */
620 region = list_entry(nosave_regions.prev,
621 struct nosave_region, list);
622 if (region->end_pfn == start_pfn) {
623 region->end_pfn = end_pfn;
624 goto Report;
625 }
626 }
627 if (use_kmalloc) {
628 /* during init, this shouldn't fail */
629 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
630 BUG_ON(!region);
631 } else
632 /* This allocation cannot fail */
633 region = alloc_bootmem_low(sizeof(struct nosave_region));
634 region->start_pfn = start_pfn;
635 region->end_pfn = end_pfn;
636 list_add_tail(&region->list, &nosave_regions);
637 Report:
638 printk("swsusp: Registered nosave memory region: %016lx - %016lx\n",
639 start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT);
640 }
641
642 /*
643 * Set bits in this map correspond to the page frames the contents of which
644 * should not be saved during the suspend.
645 */
646 static struct memory_bitmap *forbidden_pages_map;
647
648 /* Set bits in this map correspond to free page frames. */
649 static struct memory_bitmap *free_pages_map;
650
651 /*
652 * Each page frame allocated for creating the image is marked by setting the
653 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
654 */
655
656 void swsusp_set_page_free(struct page *page)
657 {
658 if (free_pages_map)
659 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
660 }
661
662 static int swsusp_page_is_free(struct page *page)
663 {
664 return free_pages_map ?
665 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
666 }
667
668 void swsusp_unset_page_free(struct page *page)
669 {
670 if (free_pages_map)
671 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
672 }
673
674 static void swsusp_set_page_forbidden(struct page *page)
675 {
676 if (forbidden_pages_map)
677 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
678 }
679
680 int swsusp_page_is_forbidden(struct page *page)
681 {
682 return forbidden_pages_map ?
683 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
684 }
685
686 static void swsusp_unset_page_forbidden(struct page *page)
687 {
688 if (forbidden_pages_map)
689 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
690 }
691
692 /**
693 * mark_nosave_pages - set bits corresponding to the page frames the
694 * contents of which should not be saved in a given bitmap.
695 */
696
697 static void mark_nosave_pages(struct memory_bitmap *bm)
698 {
699 struct nosave_region *region;
700
701 if (list_empty(&nosave_regions))
702 return;
703
704 list_for_each_entry(region, &nosave_regions, list) {
705 unsigned long pfn;
706
707 printk("swsusp: Marking nosave pages: %016lx - %016lx\n",
708 region->start_pfn << PAGE_SHIFT,
709 region->end_pfn << PAGE_SHIFT);
710
711 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
712 if (pfn_valid(pfn))
713 memory_bm_set_bit(bm, pfn);
714 }
715 }
716
717 /**
718 * create_basic_memory_bitmaps - create bitmaps needed for marking page
719 * frames that should not be saved and free page frames. The pointers
720 * forbidden_pages_map and free_pages_map are only modified if everything
721 * goes well, because we don't want the bits to be used before both bitmaps
722 * are set up.
723 */
724
725 int create_basic_memory_bitmaps(void)
726 {
727 struct memory_bitmap *bm1, *bm2;
728 int error = 0;
729
730 BUG_ON(forbidden_pages_map || free_pages_map);
731
732 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
733 if (!bm1)
734 return -ENOMEM;
735
736 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
737 if (error)
738 goto Free_first_object;
739
740 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
741 if (!bm2)
742 goto Free_first_bitmap;
743
744 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
745 if (error)
746 goto Free_second_object;
747
748 forbidden_pages_map = bm1;
749 free_pages_map = bm2;
750 mark_nosave_pages(forbidden_pages_map);
751
752 printk("swsusp: Basic memory bitmaps created\n");
753
754 return 0;
755
756 Free_second_object:
757 kfree(bm2);
758 Free_first_bitmap:
759 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
760 Free_first_object:
761 kfree(bm1);
762 return -ENOMEM;
763 }
764
765 /**
766 * free_basic_memory_bitmaps - free memory bitmaps allocated by
767 * create_basic_memory_bitmaps(). The auxiliary pointers are necessary
768 * so that the bitmaps themselves are not referred to while they are being
769 * freed.
770 */
771
772 void free_basic_memory_bitmaps(void)
773 {
774 struct memory_bitmap *bm1, *bm2;
775
776 BUG_ON(!(forbidden_pages_map && free_pages_map));
777
778 bm1 = forbidden_pages_map;
779 bm2 = free_pages_map;
780 forbidden_pages_map = NULL;
781 free_pages_map = NULL;
782 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
783 kfree(bm1);
784 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
785 kfree(bm2);
786
787 printk("swsusp: Basic memory bitmaps freed\n");
788 }
789
790 /**
791 * snapshot_additional_pages - estimate the number of additional pages
792 * be needed for setting up the suspend image data structures for given
793 * zone (usually the returned value is greater than the exact number)
794 */
795
796 unsigned int snapshot_additional_pages(struct zone *zone)
797 {
798 unsigned int res;
799
800 res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
801 res += DIV_ROUND_UP(res * sizeof(struct bm_block), PAGE_SIZE);
802 return 2 * res;
803 }
804
805 #ifdef CONFIG_HIGHMEM
806 /**
807 * count_free_highmem_pages - compute the total number of free highmem
808 * pages, system-wide.
809 */
810
811 static unsigned int count_free_highmem_pages(void)
812 {
813 struct zone *zone;
814 unsigned int cnt = 0;
815
816 for_each_zone(zone)
817 if (populated_zone(zone) && is_highmem(zone))
818 cnt += zone_page_state(zone, NR_FREE_PAGES);
819
820 return cnt;
821 }
822
823 /**
824 * saveable_highmem_page - Determine whether a highmem page should be
825 * included in the suspend image.
826 *
827 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
828 * and it isn't a part of a free chunk of pages.
829 */
830
831 static struct page *saveable_highmem_page(unsigned long pfn)
832 {
833 struct page *page;
834
835 if (!pfn_valid(pfn))
836 return NULL;
837
838 page = pfn_to_page(pfn);
839
840 BUG_ON(!PageHighMem(page));
841
842 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) ||
843 PageReserved(page))
844 return NULL;
845
846 return page;
847 }
848
849 /**
850 * count_highmem_pages - compute the total number of saveable highmem
851 * pages.
852 */
853
854 unsigned int count_highmem_pages(void)
855 {
856 struct zone *zone;
857 unsigned int n = 0;
858
859 for_each_zone(zone) {
860 unsigned long pfn, max_zone_pfn;
861
862 if (!is_highmem(zone))
863 continue;
864
865 mark_free_pages(zone);
866 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
867 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
868 if (saveable_highmem_page(pfn))
869 n++;
870 }
871 return n;
872 }
873 #else
874 static inline void *saveable_highmem_page(unsigned long pfn) { return NULL; }
875 static inline unsigned int count_highmem_pages(void) { return 0; }
876 #endif /* CONFIG_HIGHMEM */
877
878 /**
879 * saveable - Determine whether a non-highmem page should be included in
880 * 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
887 static struct page *saveable_page(unsigned long pfn)
888 {
889 struct page *page;
890
891 if (!pfn_valid(pfn))
892 return NULL;
893
894 page = pfn_to_page(pfn);
895
896 BUG_ON(PageHighMem(page));
897
898 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
899 return NULL;
900
901 if (PageReserved(page) && pfn_is_nosave(pfn))
902 return NULL;
903
904 return page;
905 }
906
907 /**
908 * count_data_pages - compute the total number of saveable non-highmem
909 * pages.
910 */
911
912 unsigned int count_data_pages(void)
913 {
914 struct zone *zone;
915 unsigned long pfn, max_zone_pfn;
916 unsigned int n = 0;
917
918 for_each_zone(zone) {
919 if (is_highmem(zone))
920 continue;
921
922 mark_free_pages(zone);
923 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
924 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
925 if(saveable_page(pfn))
926 n++;
927 }
928 return n;
929 }
930
931 /* This is needed, because copy_page and memcpy are not usable for copying
932 * task structs.
933 */
934 static inline void do_copy_page(long *dst, long *src)
935 {
936 int n;
937
938 for (n = PAGE_SIZE / sizeof(long); n; n--)
939 *dst++ = *src++;
940 }
941
942 #ifdef CONFIG_HIGHMEM
943 static inline struct page *
944 page_is_saveable(struct zone *zone, unsigned long pfn)
945 {
946 return is_highmem(zone) ?
947 saveable_highmem_page(pfn) : saveable_page(pfn);
948 }
949
950 static inline void
951 copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
952 {
953 struct page *s_page, *d_page;
954 void *src, *dst;
955
956 s_page = pfn_to_page(src_pfn);
957 d_page = pfn_to_page(dst_pfn);
958 if (PageHighMem(s_page)) {
959 src = kmap_atomic(s_page, KM_USER0);
960 dst = kmap_atomic(d_page, KM_USER1);
961 do_copy_page(dst, src);
962 kunmap_atomic(src, KM_USER0);
963 kunmap_atomic(dst, KM_USER1);
964 } else {
965 src = page_address(s_page);
966 if (PageHighMem(d_page)) {
967 /* Page pointed to by src may contain some kernel
968 * data modified by kmap_atomic()
969 */
970 do_copy_page(buffer, src);
971 dst = kmap_atomic(pfn_to_page(dst_pfn), KM_USER0);
972 memcpy(dst, buffer, PAGE_SIZE);
973 kunmap_atomic(dst, KM_USER0);
974 } else {
975 dst = page_address(d_page);
976 do_copy_page(dst, src);
977 }
978 }
979 }
980 #else
981 #define page_is_saveable(zone, pfn) saveable_page(pfn)
982
983 static inline void
984 copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
985 {
986 do_copy_page(page_address(pfn_to_page(dst_pfn)),
987 page_address(pfn_to_page(src_pfn)));
988 }
989 #endif /* CONFIG_HIGHMEM */
990
991 static void
992 copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm)
993 {
994 struct zone *zone;
995 unsigned long pfn;
996
997 for_each_zone(zone) {
998 unsigned long max_zone_pfn;
999
1000 mark_free_pages(zone);
1001 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1002 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1003 if (page_is_saveable(zone, pfn))
1004 memory_bm_set_bit(orig_bm, pfn);
1005 }
1006 memory_bm_position_reset(orig_bm);
1007 memory_bm_position_reset(copy_bm);
1008 do {
1009 pfn = memory_bm_next_pfn(orig_bm);
1010 if (likely(pfn != BM_END_OF_MAP))
1011 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1012 } while (pfn != BM_END_OF_MAP);
1013 }
1014
1015 /* Total number of image pages */
1016 static unsigned int nr_copy_pages;
1017 /* Number of pages needed for saving the original pfns of the image pages */
1018 static unsigned int nr_meta_pages;
1019
1020 /**
1021 * swsusp_free - free pages allocated for the suspend.
1022 *
1023 * Suspend pages are alocated before the atomic copy is made, so we
1024 * need to release them after the resume.
1025 */
1026
1027 void swsusp_free(void)
1028 {
1029 struct zone *zone;
1030 unsigned long pfn, max_zone_pfn;
1031
1032 for_each_zone(zone) {
1033 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1034 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1035 if (pfn_valid(pfn)) {
1036 struct page *page = pfn_to_page(pfn);
1037
1038 if (swsusp_page_is_forbidden(page) &&
1039 swsusp_page_is_free(page)) {
1040 swsusp_unset_page_forbidden(page);
1041 swsusp_unset_page_free(page);
1042 __free_page(page);
1043 }
1044 }
1045 }
1046 nr_copy_pages = 0;
1047 nr_meta_pages = 0;
1048 restore_pblist = NULL;
1049 buffer = NULL;
1050 }
1051
1052 #ifdef CONFIG_HIGHMEM
1053 /**
1054 * count_pages_for_highmem - compute the number of non-highmem pages
1055 * that will be necessary for creating copies of highmem pages.
1056 */
1057
1058 static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1059 {
1060 unsigned int free_highmem = count_free_highmem_pages();
1061
1062 if (free_highmem >= nr_highmem)
1063 nr_highmem = 0;
1064 else
1065 nr_highmem -= free_highmem;
1066
1067 return nr_highmem;
1068 }
1069 #else
1070 static unsigned int
1071 count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1072 #endif /* CONFIG_HIGHMEM */
1073
1074 /**
1075 * enough_free_mem - Make sure we have enough free memory for the
1076 * snapshot image.
1077 */
1078
1079 static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1080 {
1081 struct zone *zone;
1082 unsigned int free = 0, meta = 0;
1083
1084 for_each_zone(zone) {
1085 meta += snapshot_additional_pages(zone);
1086 if (!is_highmem(zone))
1087 free += zone_page_state(zone, NR_FREE_PAGES);
1088 }
1089
1090 nr_pages += count_pages_for_highmem(nr_highmem);
1091 pr_debug("swsusp: Normal pages needed: %u + %u + %u, available pages: %u\n",
1092 nr_pages, PAGES_FOR_IO, meta, free);
1093
1094 return free > nr_pages + PAGES_FOR_IO + meta;
1095 }
1096
1097 #ifdef CONFIG_HIGHMEM
1098 /**
1099 * get_highmem_buffer - if there are some highmem pages in the suspend
1100 * image, we may need the buffer to copy them and/or load their data.
1101 */
1102
1103 static inline int get_highmem_buffer(int safe_needed)
1104 {
1105 buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
1106 return buffer ? 0 : -ENOMEM;
1107 }
1108
1109 /**
1110 * alloc_highmem_image_pages - allocate some highmem pages for the image.
1111 * Try to allocate as many pages as needed, but if the number of free
1112 * highmem pages is lesser than that, allocate them all.
1113 */
1114
1115 static inline unsigned int
1116 alloc_highmem_image_pages(struct memory_bitmap *bm, unsigned int nr_highmem)
1117 {
1118 unsigned int to_alloc = count_free_highmem_pages();
1119
1120 if (to_alloc > nr_highmem)
1121 to_alloc = nr_highmem;
1122
1123 nr_highmem -= to_alloc;
1124 while (to_alloc-- > 0) {
1125 struct page *page;
1126
1127 page = alloc_image_page(__GFP_HIGHMEM);
1128 memory_bm_set_bit(bm, page_to_pfn(page));
1129 }
1130 return nr_highmem;
1131 }
1132 #else
1133 static inline int get_highmem_buffer(int safe_needed) { return 0; }
1134
1135 static inline unsigned int
1136 alloc_highmem_image_pages(struct memory_bitmap *bm, unsigned int n) { return 0; }
1137 #endif /* CONFIG_HIGHMEM */
1138
1139 /**
1140 * swsusp_alloc - allocate memory for the suspend image
1141 *
1142 * We first try to allocate as many highmem pages as there are
1143 * saveable highmem pages in the system. If that fails, we allocate
1144 * non-highmem pages for the copies of the remaining highmem ones.
1145 *
1146 * In this approach it is likely that the copies of highmem pages will
1147 * also be located in the high memory, because of the way in which
1148 * copy_data_pages() works.
1149 */
1150
1151 static int
1152 swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm,
1153 unsigned int nr_pages, unsigned int nr_highmem)
1154 {
1155 int error;
1156
1157 error = memory_bm_create(orig_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY);
1158 if (error)
1159 goto Free;
1160
1161 error = memory_bm_create(copy_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY);
1162 if (error)
1163 goto Free;
1164
1165 if (nr_highmem > 0) {
1166 error = get_highmem_buffer(PG_ANY);
1167 if (error)
1168 goto Free;
1169
1170 nr_pages += alloc_highmem_image_pages(copy_bm, nr_highmem);
1171 }
1172 while (nr_pages-- > 0) {
1173 struct page *page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
1174
1175 if (!page)
1176 goto Free;
1177
1178 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1179 }
1180 return 0;
1181
1182 Free:
1183 swsusp_free();
1184 return -ENOMEM;
1185 }
1186
1187 /* Memory bitmap used for marking saveable pages (during suspend) or the
1188 * suspend image pages (during resume)
1189 */
1190 static struct memory_bitmap orig_bm;
1191 /* Memory bitmap used on suspend for marking allocated pages that will contain
1192 * the copies of saveable pages. During resume it is initially used for
1193 * marking the suspend image pages, but then its set bits are duplicated in
1194 * @orig_bm and it is released. Next, on systems with high memory, it may be
1195 * used for marking "safe" highmem pages, but it has to be reinitialized for
1196 * this purpose.
1197 */
1198 static struct memory_bitmap copy_bm;
1199
1200 asmlinkage int swsusp_save(void)
1201 {
1202 unsigned int nr_pages, nr_highmem;
1203
1204 printk("swsusp: critical section: \n");
1205
1206 drain_local_pages();
1207 nr_pages = count_data_pages();
1208 nr_highmem = count_highmem_pages();
1209 printk("swsusp: Need to copy %u pages\n", nr_pages + nr_highmem);
1210
1211 if (!enough_free_mem(nr_pages, nr_highmem)) {
1212 printk(KERN_ERR "swsusp: Not enough free memory\n");
1213 return -ENOMEM;
1214 }
1215
1216 if (swsusp_alloc(&orig_bm, &copy_bm, nr_pages, nr_highmem)) {
1217 printk(KERN_ERR "swsusp: Memory allocation failed\n");
1218 return -ENOMEM;
1219 }
1220
1221 /* During allocating of suspend pagedir, new cold pages may appear.
1222 * Kill them.
1223 */
1224 drain_local_pages();
1225 copy_data_pages(&copy_bm, &orig_bm);
1226
1227 /*
1228 * End of critical section. From now on, we can write to memory,
1229 * but we should not touch disk. This specially means we must _not_
1230 * touch swap space! Except we must write out our image of course.
1231 */
1232
1233 nr_pages += nr_highmem;
1234 nr_copy_pages = nr_pages;
1235 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1236
1237 printk("swsusp: critical section: done (%d pages copied)\n", nr_pages);
1238
1239 return 0;
1240 }
1241
1242 #ifndef CONFIG_ARCH_HIBERNATION_HEADER
1243 static int init_header_complete(struct swsusp_info *info)
1244 {
1245 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
1246 info->version_code = LINUX_VERSION_CODE;
1247 return 0;
1248 }
1249
1250 static char *check_image_kernel(struct swsusp_info *info)
1251 {
1252 if (info->version_code != LINUX_VERSION_CODE)
1253 return "kernel version";
1254 if (strcmp(info->uts.sysname,init_utsname()->sysname))
1255 return "system type";
1256 if (strcmp(info->uts.release,init_utsname()->release))
1257 return "kernel release";
1258 if (strcmp(info->uts.version,init_utsname()->version))
1259 return "version";
1260 if (strcmp(info->uts.machine,init_utsname()->machine))
1261 return "machine";
1262 return NULL;
1263 }
1264 #endif /* CONFIG_ARCH_HIBERNATION_HEADER */
1265
1266 static int init_header(struct swsusp_info *info)
1267 {
1268 memset(info, 0, sizeof(struct swsusp_info));
1269 info->num_physpages = num_physpages;
1270 info->image_pages = nr_copy_pages;
1271 info->pages = nr_copy_pages + nr_meta_pages + 1;
1272 info->size = info->pages;
1273 info->size <<= PAGE_SHIFT;
1274 return init_header_complete(info);
1275 }
1276
1277 /**
1278 * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm
1279 * are stored in the array @buf[] (1 page at a time)
1280 */
1281
1282 static inline void
1283 pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
1284 {
1285 int j;
1286
1287 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1288 buf[j] = memory_bm_next_pfn(bm);
1289 if (unlikely(buf[j] == BM_END_OF_MAP))
1290 break;
1291 }
1292 }
1293
1294 /**
1295 * snapshot_read_next - used for reading the system memory snapshot.
1296 *
1297 * On the first call to it @handle should point to a zeroed
1298 * snapshot_handle structure. The structure gets updated and a pointer
1299 * to it should be passed to this function every next time.
1300 *
1301 * The @count parameter should contain the number of bytes the caller
1302 * wants to read from the snapshot. It must not be zero.
1303 *
1304 * On success the function returns a positive number. Then, the caller
1305 * is allowed to read up to the returned number of bytes from the memory
1306 * location computed by the data_of() macro. The number returned
1307 * may be smaller than @count, but this only happens if the read would
1308 * cross a page boundary otherwise.
1309 *
1310 * The function returns 0 to indicate the end of data stream condition,
1311 * and a negative number is returned on error. In such cases the
1312 * structure pointed to by @handle is not updated and should not be used
1313 * any more.
1314 */
1315
1316 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
1317 {
1318 if (handle->cur > nr_meta_pages + nr_copy_pages)
1319 return 0;
1320
1321 if (!buffer) {
1322 /* This makes the buffer be freed by swsusp_free() */
1323 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
1324 if (!buffer)
1325 return -ENOMEM;
1326 }
1327 if (!handle->offset) {
1328 int error;
1329
1330 error = init_header((struct swsusp_info *)buffer);
1331 if (error)
1332 return error;
1333 handle->buffer = buffer;
1334 memory_bm_position_reset(&orig_bm);
1335 memory_bm_position_reset(&copy_bm);
1336 }
1337 if (handle->prev < handle->cur) {
1338 if (handle->cur <= nr_meta_pages) {
1339 memset(buffer, 0, PAGE_SIZE);
1340 pack_pfns(buffer, &orig_bm);
1341 } else {
1342 struct page *page;
1343
1344 page = pfn_to_page(memory_bm_next_pfn(&copy_bm));
1345 if (PageHighMem(page)) {
1346 /* Highmem pages are copied to the buffer,
1347 * because we can't return with a kmapped
1348 * highmem page (we may not be called again).
1349 */
1350 void *kaddr;
1351
1352 kaddr = kmap_atomic(page, KM_USER0);
1353 memcpy(buffer, kaddr, PAGE_SIZE);
1354 kunmap_atomic(kaddr, KM_USER0);
1355 handle->buffer = buffer;
1356 } else {
1357 handle->buffer = page_address(page);
1358 }
1359 }
1360 handle->prev = handle->cur;
1361 }
1362 handle->buf_offset = handle->cur_offset;
1363 if (handle->cur_offset + count >= PAGE_SIZE) {
1364 count = PAGE_SIZE - handle->cur_offset;
1365 handle->cur_offset = 0;
1366 handle->cur++;
1367 } else {
1368 handle->cur_offset += count;
1369 }
1370 handle->offset += count;
1371 return count;
1372 }
1373
1374 /**
1375 * mark_unsafe_pages - mark the pages that cannot be used for storing
1376 * the image during resume, because they conflict with the pages that
1377 * had been used before suspend
1378 */
1379
1380 static int mark_unsafe_pages(struct memory_bitmap *bm)
1381 {
1382 struct zone *zone;
1383 unsigned long pfn, max_zone_pfn;
1384
1385 /* Clear page flags */
1386 for_each_zone(zone) {
1387 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1388 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1389 if (pfn_valid(pfn))
1390 swsusp_unset_page_free(pfn_to_page(pfn));
1391 }
1392
1393 /* Mark pages that correspond to the "original" pfns as "unsafe" */
1394 memory_bm_position_reset(bm);
1395 do {
1396 pfn = memory_bm_next_pfn(bm);
1397 if (likely(pfn != BM_END_OF_MAP)) {
1398 if (likely(pfn_valid(pfn)))
1399 swsusp_set_page_free(pfn_to_page(pfn));
1400 else
1401 return -EFAULT;
1402 }
1403 } while (pfn != BM_END_OF_MAP);
1404
1405 allocated_unsafe_pages = 0;
1406
1407 return 0;
1408 }
1409
1410 static void
1411 duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src)
1412 {
1413 unsigned long pfn;
1414
1415 memory_bm_position_reset(src);
1416 pfn = memory_bm_next_pfn(src);
1417 while (pfn != BM_END_OF_MAP) {
1418 memory_bm_set_bit(dst, pfn);
1419 pfn = memory_bm_next_pfn(src);
1420 }
1421 }
1422
1423 static int check_header(struct swsusp_info *info)
1424 {
1425 char *reason;
1426
1427 reason = check_image_kernel(info);
1428 if (!reason && info->num_physpages != num_physpages)
1429 reason = "memory size";
1430 if (reason) {
1431 printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
1432 return -EPERM;
1433 }
1434 return 0;
1435 }
1436
1437 /**
1438 * load header - check the image header and copy data from it
1439 */
1440
1441 static int
1442 load_header(struct swsusp_info *info)
1443 {
1444 int error;
1445
1446 restore_pblist = NULL;
1447 error = check_header(info);
1448 if (!error) {
1449 nr_copy_pages = info->image_pages;
1450 nr_meta_pages = info->pages - info->image_pages - 1;
1451 }
1452 return error;
1453 }
1454
1455 /**
1456 * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set
1457 * the corresponding bit in the memory bitmap @bm
1458 */
1459
1460 static inline void
1461 unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
1462 {
1463 int j;
1464
1465 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1466 if (unlikely(buf[j] == BM_END_OF_MAP))
1467 break;
1468
1469 memory_bm_set_bit(bm, buf[j]);
1470 }
1471 }
1472
1473 /* List of "safe" pages that may be used to store data loaded from the suspend
1474 * image
1475 */
1476 static struct linked_page *safe_pages_list;
1477
1478 #ifdef CONFIG_HIGHMEM
1479 /* struct highmem_pbe is used for creating the list of highmem pages that
1480 * should be restored atomically during the resume from disk, because the page
1481 * frames they have occupied before the suspend are in use.
1482 */
1483 struct highmem_pbe {
1484 struct page *copy_page; /* data is here now */
1485 struct page *orig_page; /* data was here before the suspend */
1486 struct highmem_pbe *next;
1487 };
1488
1489 /* List of highmem PBEs needed for restoring the highmem pages that were
1490 * allocated before the suspend and included in the suspend image, but have
1491 * also been allocated by the "resume" kernel, so their contents cannot be
1492 * written directly to their "original" page frames.
1493 */
1494 static struct highmem_pbe *highmem_pblist;
1495
1496 /**
1497 * count_highmem_image_pages - compute the number of highmem pages in the
1498 * suspend image. The bits in the memory bitmap @bm that correspond to the
1499 * image pages are assumed to be set.
1500 */
1501
1502 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
1503 {
1504 unsigned long pfn;
1505 unsigned int cnt = 0;
1506
1507 memory_bm_position_reset(bm);
1508 pfn = memory_bm_next_pfn(bm);
1509 while (pfn != BM_END_OF_MAP) {
1510 if (PageHighMem(pfn_to_page(pfn)))
1511 cnt++;
1512
1513 pfn = memory_bm_next_pfn(bm);
1514 }
1515 return cnt;
1516 }
1517
1518 /**
1519 * prepare_highmem_image - try to allocate as many highmem pages as
1520 * there are highmem image pages (@nr_highmem_p points to the variable
1521 * containing the number of highmem image pages). The pages that are
1522 * "safe" (ie. will not be overwritten when the suspend image is
1523 * restored) have the corresponding bits set in @bm (it must be
1524 * unitialized).
1525 *
1526 * NOTE: This function should not be called if there are no highmem
1527 * image pages.
1528 */
1529
1530 static unsigned int safe_highmem_pages;
1531
1532 static struct memory_bitmap *safe_highmem_bm;
1533
1534 static int
1535 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
1536 {
1537 unsigned int to_alloc;
1538
1539 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
1540 return -ENOMEM;
1541
1542 if (get_highmem_buffer(PG_SAFE))
1543 return -ENOMEM;
1544
1545 to_alloc = count_free_highmem_pages();
1546 if (to_alloc > *nr_highmem_p)
1547 to_alloc = *nr_highmem_p;
1548 else
1549 *nr_highmem_p = to_alloc;
1550
1551 safe_highmem_pages = 0;
1552 while (to_alloc-- > 0) {
1553 struct page *page;
1554
1555 page = alloc_page(__GFP_HIGHMEM);
1556 if (!swsusp_page_is_free(page)) {
1557 /* The page is "safe", set its bit the bitmap */
1558 memory_bm_set_bit(bm, page_to_pfn(page));
1559 safe_highmem_pages++;
1560 }
1561 /* Mark the page as allocated */
1562 swsusp_set_page_forbidden(page);
1563 swsusp_set_page_free(page);
1564 }
1565 memory_bm_position_reset(bm);
1566 safe_highmem_bm = bm;
1567 return 0;
1568 }
1569
1570 /**
1571 * get_highmem_page_buffer - for given highmem image page find the buffer
1572 * that suspend_write_next() should set for its caller to write to.
1573 *
1574 * If the page is to be saved to its "original" page frame or a copy of
1575 * the page is to be made in the highmem, @buffer is returned. Otherwise,
1576 * the copy of the page is to be made in normal memory, so the address of
1577 * the copy is returned.
1578 *
1579 * If @buffer is returned, the caller of suspend_write_next() will write
1580 * the page's contents to @buffer, so they will have to be copied to the
1581 * right location on the next call to suspend_write_next() and it is done
1582 * with the help of copy_last_highmem_page(). For this purpose, if
1583 * @buffer is returned, @last_highmem page is set to the page to which
1584 * the data will have to be copied from @buffer.
1585 */
1586
1587 static struct page *last_highmem_page;
1588
1589 static void *
1590 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
1591 {
1592 struct highmem_pbe *pbe;
1593 void *kaddr;
1594
1595 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
1596 /* We have allocated the "original" page frame and we can
1597 * use it directly to store the loaded page.
1598 */
1599 last_highmem_page = page;
1600 return buffer;
1601 }
1602 /* The "original" page frame has not been allocated and we have to
1603 * use a "safe" page frame to store the loaded page.
1604 */
1605 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
1606 if (!pbe) {
1607 swsusp_free();
1608 return NULL;
1609 }
1610 pbe->orig_page = page;
1611 if (safe_highmem_pages > 0) {
1612 struct page *tmp;
1613
1614 /* Copy of the page will be stored in high memory */
1615 kaddr = buffer;
1616 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
1617 safe_highmem_pages--;
1618 last_highmem_page = tmp;
1619 pbe->copy_page = tmp;
1620 } else {
1621 /* Copy of the page will be stored in normal memory */
1622 kaddr = safe_pages_list;
1623 safe_pages_list = safe_pages_list->next;
1624 pbe->copy_page = virt_to_page(kaddr);
1625 }
1626 pbe->next = highmem_pblist;
1627 highmem_pblist = pbe;
1628 return kaddr;
1629 }
1630
1631 /**
1632 * copy_last_highmem_page - copy the contents of a highmem image from
1633 * @buffer, where the caller of snapshot_write_next() has place them,
1634 * to the right location represented by @last_highmem_page .
1635 */
1636
1637 static void copy_last_highmem_page(void)
1638 {
1639 if (last_highmem_page) {
1640 void *dst;
1641
1642 dst = kmap_atomic(last_highmem_page, KM_USER0);
1643 memcpy(dst, buffer, PAGE_SIZE);
1644 kunmap_atomic(dst, KM_USER0);
1645 last_highmem_page = NULL;
1646 }
1647 }
1648
1649 static inline int last_highmem_page_copied(void)
1650 {
1651 return !last_highmem_page;
1652 }
1653
1654 static inline void free_highmem_data(void)
1655 {
1656 if (safe_highmem_bm)
1657 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
1658
1659 if (buffer)
1660 free_image_page(buffer, PG_UNSAFE_CLEAR);
1661 }
1662 #else
1663 static inline int get_safe_write_buffer(void) { return 0; }
1664
1665 static unsigned int
1666 count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
1667
1668 static inline int
1669 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
1670 {
1671 return 0;
1672 }
1673
1674 static inline void *
1675 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
1676 {
1677 return NULL;
1678 }
1679
1680 static inline void copy_last_highmem_page(void) {}
1681 static inline int last_highmem_page_copied(void) { return 1; }
1682 static inline void free_highmem_data(void) {}
1683 #endif /* CONFIG_HIGHMEM */
1684
1685 /**
1686 * prepare_image - use the memory bitmap @bm to mark the pages that will
1687 * be overwritten in the process of restoring the system memory state
1688 * from the suspend image ("unsafe" pages) and allocate memory for the
1689 * image.
1690 *
1691 * The idea is to allocate a new memory bitmap first and then allocate
1692 * as many pages as needed for the image data, but not to assign these
1693 * pages to specific tasks initially. Instead, we just mark them as
1694 * allocated and create a lists of "safe" pages that will be used
1695 * later. On systems with high memory a list of "safe" highmem pages is
1696 * also created.
1697 */
1698
1699 #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
1700
1701 static int
1702 prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
1703 {
1704 unsigned int nr_pages, nr_highmem;
1705 struct linked_page *sp_list, *lp;
1706 int error;
1707
1708 /* If there is no highmem, the buffer will not be necessary */
1709 free_image_page(buffer, PG_UNSAFE_CLEAR);
1710 buffer = NULL;
1711
1712 nr_highmem = count_highmem_image_pages(bm);
1713 error = mark_unsafe_pages(bm);
1714 if (error)
1715 goto Free;
1716
1717 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
1718 if (error)
1719 goto Free;
1720
1721 duplicate_memory_bitmap(new_bm, bm);
1722 memory_bm_free(bm, PG_UNSAFE_KEEP);
1723 if (nr_highmem > 0) {
1724 error = prepare_highmem_image(bm, &nr_highmem);
1725 if (error)
1726 goto Free;
1727 }
1728 /* Reserve some safe pages for potential later use.
1729 *
1730 * NOTE: This way we make sure there will be enough safe pages for the
1731 * chain_alloc() in get_buffer(). It is a bit wasteful, but
1732 * nr_copy_pages cannot be greater than 50% of the memory anyway.
1733 */
1734 sp_list = NULL;
1735 /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */
1736 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
1737 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
1738 while (nr_pages > 0) {
1739 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
1740 if (!lp) {
1741 error = -ENOMEM;
1742 goto Free;
1743 }
1744 lp->next = sp_list;
1745 sp_list = lp;
1746 nr_pages--;
1747 }
1748 /* Preallocate memory for the image */
1749 safe_pages_list = NULL;
1750 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
1751 while (nr_pages > 0) {
1752 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
1753 if (!lp) {
1754 error = -ENOMEM;
1755 goto Free;
1756 }
1757 if (!swsusp_page_is_free(virt_to_page(lp))) {
1758 /* The page is "safe", add it to the list */
1759 lp->next = safe_pages_list;
1760 safe_pages_list = lp;
1761 }
1762 /* Mark the page as allocated */
1763 swsusp_set_page_forbidden(virt_to_page(lp));
1764 swsusp_set_page_free(virt_to_page(lp));
1765 nr_pages--;
1766 }
1767 /* Free the reserved safe pages so that chain_alloc() can use them */
1768 while (sp_list) {
1769 lp = sp_list->next;
1770 free_image_page(sp_list, PG_UNSAFE_CLEAR);
1771 sp_list = lp;
1772 }
1773 return 0;
1774
1775 Free:
1776 swsusp_free();
1777 return error;
1778 }
1779
1780 /**
1781 * get_buffer - compute the address that snapshot_write_next() should
1782 * set for its caller to write to.
1783 */
1784
1785 static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
1786 {
1787 struct pbe *pbe;
1788 struct page *page = pfn_to_page(memory_bm_next_pfn(bm));
1789
1790 if (PageHighMem(page))
1791 return get_highmem_page_buffer(page, ca);
1792
1793 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
1794 /* We have allocated the "original" page frame and we can
1795 * use it directly to store the loaded page.
1796 */
1797 return page_address(page);
1798
1799 /* The "original" page frame has not been allocated and we have to
1800 * use a "safe" page frame to store the loaded page.
1801 */
1802 pbe = chain_alloc(ca, sizeof(struct pbe));
1803 if (!pbe) {
1804 swsusp_free();
1805 return NULL;
1806 }
1807 pbe->orig_address = page_address(page);
1808 pbe->address = safe_pages_list;
1809 safe_pages_list = safe_pages_list->next;
1810 pbe->next = restore_pblist;
1811 restore_pblist = pbe;
1812 return pbe->address;
1813 }
1814
1815 /**
1816 * snapshot_write_next - used for writing the system memory snapshot.
1817 *
1818 * On the first call to it @handle should point to a zeroed
1819 * snapshot_handle structure. The structure gets updated and a pointer
1820 * to it should be passed to this function every next time.
1821 *
1822 * The @count parameter should contain the number of bytes the caller
1823 * wants to write to the image. It must not be zero.
1824 *
1825 * On success the function returns a positive number. Then, the caller
1826 * is allowed to write up to the returned number of bytes to the memory
1827 * location computed by the data_of() macro. The number returned
1828 * may be smaller than @count, but this only happens if the write would
1829 * cross a page boundary otherwise.
1830 *
1831 * The function returns 0 to indicate the "end of file" condition,
1832 * and a negative number is returned on error. In such cases the
1833 * structure pointed to by @handle is not updated and should not be used
1834 * any more.
1835 */
1836
1837 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
1838 {
1839 static struct chain_allocator ca;
1840 int error = 0;
1841
1842 /* Check if we have already loaded the entire image */
1843 if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages)
1844 return 0;
1845
1846 if (handle->offset == 0) {
1847 if (!buffer)
1848 /* This makes the buffer be freed by swsusp_free() */
1849 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
1850
1851 if (!buffer)
1852 return -ENOMEM;
1853
1854 handle->buffer = buffer;
1855 }
1856 handle->sync_read = 1;
1857 if (handle->prev < handle->cur) {
1858 if (handle->prev == 0) {
1859 error = load_header(buffer);
1860 if (error)
1861 return error;
1862
1863 error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
1864 if (error)
1865 return error;
1866
1867 } else if (handle->prev <= nr_meta_pages) {
1868 unpack_orig_pfns(buffer, &copy_bm);
1869 if (handle->prev == nr_meta_pages) {
1870 error = prepare_image(&orig_bm, &copy_bm);
1871 if (error)
1872 return error;
1873
1874 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
1875 memory_bm_position_reset(&orig_bm);
1876 restore_pblist = NULL;
1877 handle->buffer = get_buffer(&orig_bm, &ca);
1878 handle->sync_read = 0;
1879 if (!handle->buffer)
1880 return -ENOMEM;
1881 }
1882 } else {
1883 copy_last_highmem_page();
1884 handle->buffer = get_buffer(&orig_bm, &ca);
1885 if (handle->buffer != buffer)
1886 handle->sync_read = 0;
1887 }
1888 handle->prev = handle->cur;
1889 }
1890 handle->buf_offset = handle->cur_offset;
1891 if (handle->cur_offset + count >= PAGE_SIZE) {
1892 count = PAGE_SIZE - handle->cur_offset;
1893 handle->cur_offset = 0;
1894 handle->cur++;
1895 } else {
1896 handle->cur_offset += count;
1897 }
1898 handle->offset += count;
1899 return count;
1900 }
1901
1902 /**
1903 * snapshot_write_finalize - must be called after the last call to
1904 * snapshot_write_next() in case the last page in the image happens
1905 * to be a highmem page and its contents should be stored in the
1906 * highmem. Additionally, it releases the memory that will not be
1907 * used any more.
1908 */
1909
1910 void snapshot_write_finalize(struct snapshot_handle *handle)
1911 {
1912 copy_last_highmem_page();
1913 /* Free only if we have loaded the image entirely */
1914 if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages) {
1915 memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR);
1916 free_highmem_data();
1917 }
1918 }
1919
1920 int snapshot_image_loaded(struct snapshot_handle *handle)
1921 {
1922 return !(!nr_copy_pages || !last_highmem_page_copied() ||
1923 handle->cur <= nr_meta_pages + nr_copy_pages);
1924 }
1925
1926 #ifdef CONFIG_HIGHMEM
1927 /* Assumes that @buf is ready and points to a "safe" page */
1928 static inline void
1929 swap_two_pages_data(struct page *p1, struct page *p2, void *buf)
1930 {
1931 void *kaddr1, *kaddr2;
1932
1933 kaddr1 = kmap_atomic(p1, KM_USER0);
1934 kaddr2 = kmap_atomic(p2, KM_USER1);
1935 memcpy(buf, kaddr1, PAGE_SIZE);
1936 memcpy(kaddr1, kaddr2, PAGE_SIZE);
1937 memcpy(kaddr2, buf, PAGE_SIZE);
1938 kunmap_atomic(kaddr1, KM_USER0);
1939 kunmap_atomic(kaddr2, KM_USER1);
1940 }
1941
1942 /**
1943 * restore_highmem - for each highmem page that was allocated before
1944 * the suspend and included in the suspend image, and also has been
1945 * allocated by the "resume" kernel swap its current (ie. "before
1946 * resume") contents with the previous (ie. "before suspend") one.
1947 *
1948 * If the resume eventually fails, we can call this function once
1949 * again and restore the "before resume" highmem state.
1950 */
1951
1952 int restore_highmem(void)
1953 {
1954 struct highmem_pbe *pbe = highmem_pblist;
1955 void *buf;
1956
1957 if (!pbe)
1958 return 0;
1959
1960 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
1961 if (!buf)
1962 return -ENOMEM;
1963
1964 while (pbe) {
1965 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
1966 pbe = pbe->next;
1967 }
1968 free_image_page(buf, PG_UNSAFE_CLEAR);
1969 return 0;
1970 }
1971 #endif /* CONFIG_HIGHMEM */
This page took 0.087999 seconds and 5 git commands to generate.