aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/power/snapshot.c
diff options
context:
space:
mode:
Diffstat (limited to 'kernel/power/snapshot.c')
-rw-r--r--kernel/power/snapshot.c1155
1 files changed, 801 insertions, 354 deletions
diff --git a/kernel/power/snapshot.c b/kernel/power/snapshot.c
index 75d4886e648e..1b84313cbab5 100644
--- a/kernel/power/snapshot.c
+++ b/kernel/power/snapshot.c
@@ -34,10 +34,12 @@
34 34
35#include "power.h" 35#include "power.h"
36 36
37struct pbe *pagedir_nosave; 37/* List of PBEs used for creating and restoring the suspend image */
38struct pbe *restore_pblist;
39
38static unsigned int nr_copy_pages; 40static unsigned int nr_copy_pages;
39static unsigned int nr_meta_pages; 41static unsigned int nr_meta_pages;
40static unsigned long *buffer; 42static void *buffer;
41 43
42#ifdef CONFIG_HIGHMEM 44#ifdef CONFIG_HIGHMEM
43unsigned int count_highmem_pages(void) 45unsigned int count_highmem_pages(void)
@@ -156,240 +158,637 @@ static inline int save_highmem(void) {return 0;}
156static inline int restore_highmem(void) {return 0;} 158static inline int restore_highmem(void) {return 0;}
157#endif 159#endif
158 160
159static int pfn_is_nosave(unsigned long pfn) 161/**
162 * @safe_needed - on resume, for storing the PBE list and the image,
163 * we can only use memory pages that do not conflict with the pages
164 * used before suspend.
165 *
166 * The unsafe pages are marked with the PG_nosave_free flag
167 * and we count them using unsafe_pages
168 */
169
170#define PG_ANY 0
171#define PG_SAFE 1
172#define PG_UNSAFE_CLEAR 1
173#define PG_UNSAFE_KEEP 0
174
175static unsigned int allocated_unsafe_pages;
176
177static void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
160{ 178{
161 unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT; 179 void *res;
162 unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT; 180
163 return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn); 181 res = (void *)get_zeroed_page(gfp_mask);
182 if (safe_needed)
183 while (res && PageNosaveFree(virt_to_page(res))) {
184 /* The page is unsafe, mark it for swsusp_free() */
185 SetPageNosave(virt_to_page(res));
186 allocated_unsafe_pages++;
187 res = (void *)get_zeroed_page(gfp_mask);
188 }
189 if (res) {
190 SetPageNosave(virt_to_page(res));
191 SetPageNosaveFree(virt_to_page(res));
192 }
193 return res;
194}
195
196unsigned long get_safe_page(gfp_t gfp_mask)
197{
198 return (unsigned long)alloc_image_page(gfp_mask, PG_SAFE);
164} 199}
165 200
166/** 201/**
167 * saveable - Determine whether a page should be cloned or not. 202 * free_image_page - free page represented by @addr, allocated with
168 * @pfn: The page 203 * alloc_image_page (page flags set by it must be cleared)
169 *
170 * We save a page if it's Reserved, and not in the range of pages
171 * statically defined as 'unsaveable', or if it isn't reserved, and
172 * isn't part of a free chunk of pages.
173 */ 204 */
174 205
175static int saveable(struct zone *zone, unsigned long *zone_pfn) 206static inline void free_image_page(void *addr, int clear_nosave_free)
176{ 207{
177 unsigned long pfn = *zone_pfn + zone->zone_start_pfn; 208 ClearPageNosave(virt_to_page(addr));
178 struct page *page; 209 if (clear_nosave_free)
210 ClearPageNosaveFree(virt_to_page(addr));
211 free_page((unsigned long)addr);
212}
179 213
180 if (!pfn_valid(pfn)) 214/* struct linked_page is used to build chains of pages */
181 return 0;
182 215
183 page = pfn_to_page(pfn); 216#define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
184 BUG_ON(PageReserved(page) && PageNosave(page));
185 if (PageNosave(page))
186 return 0;
187 if (PageReserved(page) && pfn_is_nosave(pfn))
188 return 0;
189 if (PageNosaveFree(page))
190 return 0;
191 217
192 return 1; 218struct linked_page {
193} 219 struct linked_page *next;
220 char data[LINKED_PAGE_DATA_SIZE];
221} __attribute__((packed));
194 222
195unsigned int count_data_pages(void) 223static inline void
224free_list_of_pages(struct linked_page *list, int clear_page_nosave)
196{ 225{
197 struct zone *zone; 226 while (list) {
198 unsigned long zone_pfn; 227 struct linked_page *lp = list->next;
199 unsigned int n = 0;
200 228
201 for_each_zone (zone) { 229 free_image_page(list, clear_page_nosave);
202 if (is_highmem(zone)) 230 list = lp;
203 continue;
204 mark_free_pages(zone);
205 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
206 n += saveable(zone, &zone_pfn);
207 } 231 }
208 return n;
209} 232}
210 233
211static void copy_data_pages(struct pbe *pblist) 234/**
235 * struct chain_allocator is used for allocating small objects out of
236 * a linked list of pages called 'the chain'.
237 *
238 * The chain grows each time when there is no room for a new object in
239 * the current page. The allocated objects cannot be freed individually.
240 * It is only possible to free them all at once, by freeing the entire
241 * chain.
242 *
243 * NOTE: The chain allocator may be inefficient if the allocated objects
244 * are not much smaller than PAGE_SIZE.
245 */
246
247struct chain_allocator {
248 struct linked_page *chain; /* the chain */
249 unsigned int used_space; /* total size of objects allocated out
250 * of the current page
251 */
252 gfp_t gfp_mask; /* mask for allocating pages */
253 int safe_needed; /* if set, only "safe" pages are allocated */
254};
255
256static void
257chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed)
212{ 258{
213 struct zone *zone; 259 ca->chain = NULL;
214 unsigned long zone_pfn; 260 ca->used_space = LINKED_PAGE_DATA_SIZE;
215 struct pbe *pbe, *p; 261 ca->gfp_mask = gfp_mask;
262 ca->safe_needed = safe_needed;
263}
216 264
217 pbe = pblist; 265static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
218 for_each_zone (zone) { 266{
219 if (is_highmem(zone)) 267 void *ret;
220 continue; 268
221 mark_free_pages(zone); 269 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
222 /* This is necessary for swsusp_free() */ 270 struct linked_page *lp;
223 for_each_pb_page (p, pblist) 271
224 SetPageNosaveFree(virt_to_page(p)); 272 lp = alloc_image_page(ca->gfp_mask, ca->safe_needed);
225 for_each_pbe (p, pblist) 273 if (!lp)
226 SetPageNosaveFree(virt_to_page(p->address)); 274 return NULL;
227 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) { 275
228 if (saveable(zone, &zone_pfn)) { 276 lp->next = ca->chain;
229 struct page *page; 277 ca->chain = lp;
230 long *src, *dst; 278 ca->used_space = 0;
231 int n;
232
233 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
234 BUG_ON(!pbe);
235 pbe->orig_address = (unsigned long)page_address(page);
236 /* copy_page and memcpy are not usable for copying task structs. */
237 dst = (long *)pbe->address;
238 src = (long *)pbe->orig_address;
239 for (n = PAGE_SIZE / sizeof(long); n; n--)
240 *dst++ = *src++;
241 pbe = pbe->next;
242 }
243 }
244 } 279 }
245 BUG_ON(pbe); 280 ret = ca->chain->data + ca->used_space;
281 ca->used_space += size;
282 return ret;
246} 283}
247 284
285static void chain_free(struct chain_allocator *ca, int clear_page_nosave)
286{
287 free_list_of_pages(ca->chain, clear_page_nosave);
288 memset(ca, 0, sizeof(struct chain_allocator));
289}
248 290
249/** 291/**
250 * free_pagedir - free pages allocated with alloc_pagedir() 292 * Data types related to memory bitmaps.
293 *
294 * Memory bitmap is a structure consiting of many linked lists of
295 * objects. The main list's elements are of type struct zone_bitmap
296 * and each of them corresonds to one zone. For each zone bitmap
297 * object there is a list of objects of type struct bm_block that
298 * represent each blocks of bit chunks in which information is
299 * stored.
300 *
301 * struct memory_bitmap contains a pointer to the main list of zone
302 * bitmap objects, a struct bm_position used for browsing the bitmap,
303 * and a pointer to the list of pages used for allocating all of the
304 * zone bitmap objects and bitmap block objects.
305 *
306 * NOTE: It has to be possible to lay out the bitmap in memory
307 * using only allocations of order 0. Additionally, the bitmap is
308 * designed to work with arbitrary number of zones (this is over the
309 * top for now, but let's avoid making unnecessary assumptions ;-).
310 *
311 * struct zone_bitmap contains a pointer to a list of bitmap block
312 * objects and a pointer to the bitmap block object that has been
313 * most recently used for setting bits. Additionally, it contains the
314 * pfns that correspond to the start and end of the represented zone.
315 *
316 * struct bm_block contains a pointer to the memory page in which
317 * information is stored (in the form of a block of bit chunks
318 * of type unsigned long each). It also contains the pfns that
319 * correspond to the start and end of the represented memory area and
320 * the number of bit chunks in the block.
321 *
322 * NOTE: Memory bitmaps are used for two types of operations only:
323 * "set a bit" and "find the next bit set". Moreover, the searching
324 * is always carried out after all of the "set a bit" operations
325 * on given bitmap.
251 */ 326 */
252 327
253static void free_pagedir(struct pbe *pblist, int clear_nosave_free) 328#define BM_END_OF_MAP (~0UL)
329
330#define BM_CHUNKS_PER_BLOCK (PAGE_SIZE / sizeof(long))
331#define BM_BITS_PER_CHUNK (sizeof(long) << 3)
332#define BM_BITS_PER_BLOCK (PAGE_SIZE << 3)
333
334struct bm_block {
335 struct bm_block *next; /* next element of the list */
336 unsigned long start_pfn; /* pfn represented by the first bit */
337 unsigned long end_pfn; /* pfn represented by the last bit plus 1 */
338 unsigned int size; /* number of bit chunks */
339 unsigned long *data; /* chunks of bits representing pages */
340};
341
342struct zone_bitmap {
343 struct zone_bitmap *next; /* next element of the list */
344 unsigned long start_pfn; /* minimal pfn in this zone */
345 unsigned long end_pfn; /* maximal pfn in this zone plus 1 */
346 struct bm_block *bm_blocks; /* list of bitmap blocks */
347 struct bm_block *cur_block; /* recently used bitmap block */
348};
349
350/* strcut bm_position is used for browsing memory bitmaps */
351
352struct bm_position {
353 struct zone_bitmap *zone_bm;
354 struct bm_block *block;
355 int chunk;
356 int bit;
357};
358
359struct memory_bitmap {
360 struct zone_bitmap *zone_bm_list; /* list of zone bitmaps */
361 struct linked_page *p_list; /* list of pages used to store zone
362 * bitmap objects and bitmap block
363 * objects
364 */
365 struct bm_position cur; /* most recently used bit position */
366};
367
368/* Functions that operate on memory bitmaps */
369
370static inline void memory_bm_reset_chunk(struct memory_bitmap *bm)
254{ 371{
255 struct pbe *pbe; 372 bm->cur.chunk = 0;
373 bm->cur.bit = -1;
374}
256 375
257 while (pblist) { 376static void memory_bm_position_reset(struct memory_bitmap *bm)
258 pbe = (pblist + PB_PAGE_SKIP)->next; 377{
259 ClearPageNosave(virt_to_page(pblist)); 378 struct zone_bitmap *zone_bm;
260 if (clear_nosave_free) 379
261 ClearPageNosaveFree(virt_to_page(pblist)); 380 zone_bm = bm->zone_bm_list;
262 free_page((unsigned long)pblist); 381 bm->cur.zone_bm = zone_bm;
263 pblist = pbe; 382 bm->cur.block = zone_bm->bm_blocks;
264 } 383 memory_bm_reset_chunk(bm);
265} 384}
266 385
386static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
387
267/** 388/**
268 * fill_pb_page - Create a list of PBEs on a given memory page 389 * create_bm_block_list - create a list of block bitmap objects
269 */ 390 */
270 391
271static inline void fill_pb_page(struct pbe *pbpage) 392static inline struct bm_block *
393create_bm_block_list(unsigned int nr_blocks, struct chain_allocator *ca)
272{ 394{
273 struct pbe *p; 395 struct bm_block *bblist = NULL;
396
397 while (nr_blocks-- > 0) {
398 struct bm_block *bb;
274 399
275 p = pbpage; 400 bb = chain_alloc(ca, sizeof(struct bm_block));
276 pbpage += PB_PAGE_SKIP; 401 if (!bb)
277 do 402 return NULL;
278 p->next = p + 1; 403
279 while (++p < pbpage); 404 bb->next = bblist;
405 bblist = bb;
406 }
407 return bblist;
280} 408}
281 409
282/** 410/**
283 * create_pbe_list - Create a list of PBEs on top of a given chain 411 * create_zone_bm_list - create a list of zone bitmap objects
284 * of memory pages allocated with alloc_pagedir()
285 */ 412 */
286 413
287static inline void create_pbe_list(struct pbe *pblist, unsigned int nr_pages) 414static inline struct zone_bitmap *
415create_zone_bm_list(unsigned int nr_zones, struct chain_allocator *ca)
288{ 416{
289 struct pbe *pbpage, *p; 417 struct zone_bitmap *zbmlist = NULL;
290 unsigned int num = PBES_PER_PAGE;
291 418
292 for_each_pb_page (pbpage, pblist) { 419 while (nr_zones-- > 0) {
293 if (num >= nr_pages) 420 struct zone_bitmap *zbm;
294 break; 421
422 zbm = chain_alloc(ca, sizeof(struct zone_bitmap));
423 if (!zbm)
424 return NULL;
425
426 zbm->next = zbmlist;
427 zbmlist = zbm;
428 }
429 return zbmlist;
430}
431
432/**
433 * memory_bm_create - allocate memory for a memory bitmap
434 */
435
436static int
437memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed)
438{
439 struct chain_allocator ca;
440 struct zone *zone;
441 struct zone_bitmap *zone_bm;
442 struct bm_block *bb;
443 unsigned int nr;
444
445 chain_init(&ca, gfp_mask, safe_needed);
295 446
296 fill_pb_page(pbpage); 447 /* Compute the number of zones */
297 num += PBES_PER_PAGE; 448 nr = 0;
449 for_each_zone (zone)
450 if (populated_zone(zone) && !is_highmem(zone))
451 nr++;
452
453 /* Allocate the list of zones bitmap objects */
454 zone_bm = create_zone_bm_list(nr, &ca);
455 bm->zone_bm_list = zone_bm;
456 if (!zone_bm) {
457 chain_free(&ca, PG_UNSAFE_CLEAR);
458 return -ENOMEM;
298 } 459 }
299 if (pbpage) { 460
300 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++) 461 /* Initialize the zone bitmap objects */
301 p->next = p + 1; 462 for_each_zone (zone) {
302 p->next = NULL; 463 unsigned long pfn;
464
465 if (!populated_zone(zone) || is_highmem(zone))
466 continue;
467
468 zone_bm->start_pfn = zone->zone_start_pfn;
469 zone_bm->end_pfn = zone->zone_start_pfn + zone->spanned_pages;
470 /* Allocate the list of bitmap block objects */
471 nr = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
472 bb = create_bm_block_list(nr, &ca);
473 zone_bm->bm_blocks = bb;
474 zone_bm->cur_block = bb;
475 if (!bb)
476 goto Free;
477
478 nr = zone->spanned_pages;
479 pfn = zone->zone_start_pfn;
480 /* Initialize the bitmap block objects */
481 while (bb) {
482 unsigned long *ptr;
483
484 ptr = alloc_image_page(gfp_mask, safe_needed);
485 bb->data = ptr;
486 if (!ptr)
487 goto Free;
488
489 bb->start_pfn = pfn;
490 if (nr >= BM_BITS_PER_BLOCK) {
491 pfn += BM_BITS_PER_BLOCK;
492 bb->size = BM_CHUNKS_PER_BLOCK;
493 nr -= BM_BITS_PER_BLOCK;
494 } else {
495 /* This is executed only once in the loop */
496 pfn += nr;
497 bb->size = DIV_ROUND_UP(nr, BM_BITS_PER_CHUNK);
498 }
499 bb->end_pfn = pfn;
500 bb = bb->next;
501 }
502 zone_bm = zone_bm->next;
303 } 503 }
504 bm->p_list = ca.chain;
505 memory_bm_position_reset(bm);
506 return 0;
507
508Free:
509 bm->p_list = ca.chain;
510 memory_bm_free(bm, PG_UNSAFE_CLEAR);
511 return -ENOMEM;
304} 512}
305 513
306static unsigned int unsafe_pages; 514/**
515 * memory_bm_free - free memory occupied by the memory bitmap @bm
516 */
517
518static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
519{
520 struct zone_bitmap *zone_bm;
521
522 /* Free the list of bit blocks for each zone_bitmap object */
523 zone_bm = bm->zone_bm_list;
524 while (zone_bm) {
525 struct bm_block *bb;
526
527 bb = zone_bm->bm_blocks;
528 while (bb) {
529 if (bb->data)
530 free_image_page(bb->data, clear_nosave_free);
531 bb = bb->next;
532 }
533 zone_bm = zone_bm->next;
534 }
535 free_list_of_pages(bm->p_list, clear_nosave_free);
536 bm->zone_bm_list = NULL;
537}
307 538
308/** 539/**
309 * @safe_needed - on resume, for storing the PBE list and the image, 540 * memory_bm_set_bit - set the bit in the bitmap @bm that corresponds
310 * we can only use memory pages that do not conflict with the pages 541 * to given pfn. The cur_zone_bm member of @bm and the cur_block member
311 * used before suspend. 542 * of @bm->cur_zone_bm are updated.
312 * 543 *
313 * The unsafe pages are marked with the PG_nosave_free flag 544 * If the bit cannot be set, the function returns -EINVAL .
314 * and we count them using unsafe_pages
315 */ 545 */
316 546
317static inline void *alloc_image_page(gfp_t gfp_mask, int safe_needed) 547static int
548memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
318{ 549{
319 void *res; 550 struct zone_bitmap *zone_bm;
320 551 struct bm_block *bb;
321 res = (void *)get_zeroed_page(gfp_mask); 552
322 if (safe_needed) 553 /* Check if the pfn is from the current zone */
323 while (res && PageNosaveFree(virt_to_page(res))) { 554 zone_bm = bm->cur.zone_bm;
324 /* The page is unsafe, mark it for swsusp_free() */ 555 if (pfn < zone_bm->start_pfn || pfn >= zone_bm->end_pfn) {
325 SetPageNosave(virt_to_page(res)); 556 zone_bm = bm->zone_bm_list;
326 unsafe_pages++; 557 /* We don't assume that the zones are sorted by pfns */
327 res = (void *)get_zeroed_page(gfp_mask); 558 while (pfn < zone_bm->start_pfn || pfn >= zone_bm->end_pfn) {
559 zone_bm = zone_bm->next;
560 if (unlikely(!zone_bm))
561 return -EINVAL;
328 } 562 }
329 if (res) { 563 bm->cur.zone_bm = zone_bm;
330 SetPageNosave(virt_to_page(res));
331 SetPageNosaveFree(virt_to_page(res));
332 } 564 }
333 return res; 565 /* Check if the pfn corresponds to the current bitmap block */
566 bb = zone_bm->cur_block;
567 if (pfn < bb->start_pfn)
568 bb = zone_bm->bm_blocks;
569
570 while (pfn >= bb->end_pfn) {
571 bb = bb->next;
572 if (unlikely(!bb))
573 return -EINVAL;
574 }
575 zone_bm->cur_block = bb;
576 pfn -= bb->start_pfn;
577 set_bit(pfn % BM_BITS_PER_CHUNK, bb->data + pfn / BM_BITS_PER_CHUNK);
578 return 0;
334} 579}
335 580
336unsigned long get_safe_page(gfp_t gfp_mask) 581/* Two auxiliary functions for memory_bm_next_pfn */
582
583/* Find the first set bit in the given chunk, if there is one */
584
585static inline int next_bit_in_chunk(int bit, unsigned long *chunk_p)
337{ 586{
338 return (unsigned long)alloc_image_page(gfp_mask, 1); 587 bit++;
588 while (bit < BM_BITS_PER_CHUNK) {
589 if (test_bit(bit, chunk_p))
590 return bit;
591
592 bit++;
593 }
594 return -1;
595}
596
597/* Find a chunk containing some bits set in given block of bits */
598
599static inline int next_chunk_in_block(int n, struct bm_block *bb)
600{
601 n++;
602 while (n < bb->size) {
603 if (bb->data[n])
604 return n;
605
606 n++;
607 }
608 return -1;
339} 609}
340 610
341/** 611/**
342 * alloc_pagedir - Allocate the page directory. 612 * memory_bm_next_pfn - find the pfn that corresponds to the next set bit
343 * 613 * in the bitmap @bm. If the pfn cannot be found, BM_END_OF_MAP is
344 * First, determine exactly how many pages we need and 614 * returned.
345 * allocate them.
346 * 615 *
347 * We arrange the pages in a chain: each page is an array of PBES_PER_PAGE 616 * It is required to run memory_bm_position_reset() before the first call to
348 * struct pbe elements (pbes) and the last element in the page points 617 * this function.
349 * to the next page. 618 */
619
620static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
621{
622 struct zone_bitmap *zone_bm;
623 struct bm_block *bb;
624 int chunk;
625 int bit;
626
627 do {
628 bb = bm->cur.block;
629 do {
630 chunk = bm->cur.chunk;
631 bit = bm->cur.bit;
632 do {
633 bit = next_bit_in_chunk(bit, bb->data + chunk);
634 if (bit >= 0)
635 goto Return_pfn;
636
637 chunk = next_chunk_in_block(chunk, bb);
638 bit = -1;
639 } while (chunk >= 0);
640 bb = bb->next;
641 bm->cur.block = bb;
642 memory_bm_reset_chunk(bm);
643 } while (bb);
644 zone_bm = bm->cur.zone_bm->next;
645 if (zone_bm) {
646 bm->cur.zone_bm = zone_bm;
647 bm->cur.block = zone_bm->bm_blocks;
648 memory_bm_reset_chunk(bm);
649 }
650 } while (zone_bm);
651 memory_bm_position_reset(bm);
652 return BM_END_OF_MAP;
653
654Return_pfn:
655 bm->cur.chunk = chunk;
656 bm->cur.bit = bit;
657 return bb->start_pfn + chunk * BM_BITS_PER_CHUNK + bit;
658}
659
660/**
661 * snapshot_additional_pages - estimate the number of additional pages
662 * be needed for setting up the suspend image data structures for given
663 * zone (usually the returned value is greater than the exact number)
664 */
665
666unsigned int snapshot_additional_pages(struct zone *zone)
667{
668 unsigned int res;
669
670 res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
671 res += DIV_ROUND_UP(res * sizeof(struct bm_block), PAGE_SIZE);
672 return res;
673}
674
675/**
676 * pfn_is_nosave - check if given pfn is in the 'nosave' section
677 */
678
679static inline int pfn_is_nosave(unsigned long pfn)
680{
681 unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
682 unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
683 return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
684}
685
686/**
687 * saveable - Determine whether a page should be cloned or not.
688 * @pfn: The page
350 * 689 *
351 * On each page we set up a list of struct_pbe elements. 690 * We save a page if it isn't Nosave, and is not in the range of pages
691 * statically defined as 'unsaveable', and it
692 * isn't a part of a free chunk of pages.
352 */ 693 */
353 694
354static struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask, 695static struct page *saveable_page(unsigned long pfn)
355 int safe_needed)
356{ 696{
357 unsigned int num; 697 struct page *page;
358 struct pbe *pblist, *pbe; 698
699 if (!pfn_valid(pfn))
700 return NULL;
359 701
360 if (!nr_pages) 702 page = pfn_to_page(pfn);
703
704 if (PageNosave(page))
705 return NULL;
706 if (PageReserved(page) && pfn_is_nosave(pfn))
361 return NULL; 707 return NULL;
708 if (PageNosaveFree(page))
709 return NULL;
710
711 return page;
712}
713
714unsigned int count_data_pages(void)
715{
716 struct zone *zone;
717 unsigned long pfn, max_zone_pfn;
718 unsigned int n = 0;
362 719
363 pblist = alloc_image_page(gfp_mask, safe_needed); 720 for_each_zone (zone) {
364 /* FIXME: rewrite this ugly loop */ 721 if (is_highmem(zone))
365 for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages; 722 continue;
366 pbe = pbe->next, num += PBES_PER_PAGE) { 723 mark_free_pages(zone);
367 pbe += PB_PAGE_SKIP; 724 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
368 pbe->next = alloc_image_page(gfp_mask, safe_needed); 725 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
726 n += !!saveable_page(pfn);
369 } 727 }
370 if (!pbe) { /* get_zeroed_page() failed */ 728 return n;
371 free_pagedir(pblist, 1); 729}
372 pblist = NULL; 730
373 } else 731static inline void copy_data_page(long *dst, long *src)
374 create_pbe_list(pblist, nr_pages); 732{
375 return pblist; 733 int n;
734
735 /* copy_page and memcpy are not usable for copying task structs. */
736 for (n = PAGE_SIZE / sizeof(long); n; n--)
737 *dst++ = *src++;
738}
739
740static void
741copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm)
742{
743 struct zone *zone;
744 unsigned long pfn;
745
746 for_each_zone (zone) {
747 unsigned long max_zone_pfn;
748
749 if (is_highmem(zone))
750 continue;
751
752 mark_free_pages(zone);
753 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
754 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
755 if (saveable_page(pfn))
756 memory_bm_set_bit(orig_bm, pfn);
757 }
758 memory_bm_position_reset(orig_bm);
759 memory_bm_position_reset(copy_bm);
760 do {
761 pfn = memory_bm_next_pfn(orig_bm);
762 if (likely(pfn != BM_END_OF_MAP)) {
763 struct page *page;
764 void *src;
765
766 page = pfn_to_page(pfn);
767 src = page_address(page);
768 page = pfn_to_page(memory_bm_next_pfn(copy_bm));
769 copy_data_page(page_address(page), src);
770 }
771 } while (pfn != BM_END_OF_MAP);
376} 772}
377 773
378/** 774/**
379 * Free pages we allocated for suspend. Suspend pages are alocated 775 * swsusp_free - free pages allocated for the suspend.
380 * before atomic copy, so we need to free them after resume. 776 *
777 * Suspend pages are alocated before the atomic copy is made, so we
778 * need to release them after the resume.
381 */ 779 */
382 780
383void swsusp_free(void) 781void swsusp_free(void)
384{ 782{
385 struct zone *zone; 783 struct zone *zone;
386 unsigned long zone_pfn; 784 unsigned long pfn, max_zone_pfn;
387 785
388 for_each_zone(zone) { 786 for_each_zone(zone) {
389 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) 787 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
390 if (pfn_valid(zone_pfn + zone->zone_start_pfn)) { 788 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
391 struct page *page; 789 if (pfn_valid(pfn)) {
392 page = pfn_to_page(zone_pfn + zone->zone_start_pfn); 790 struct page *page = pfn_to_page(pfn);
791
393 if (PageNosave(page) && PageNosaveFree(page)) { 792 if (PageNosave(page) && PageNosaveFree(page)) {
394 ClearPageNosave(page); 793 ClearPageNosave(page);
395 ClearPageNosaveFree(page); 794 ClearPageNosaveFree(page);
@@ -399,7 +798,7 @@ void swsusp_free(void)
399 } 798 }
400 nr_copy_pages = 0; 799 nr_copy_pages = 0;
401 nr_meta_pages = 0; 800 nr_meta_pages = 0;
402 pagedir_nosave = NULL; 801 restore_pblist = NULL;
403 buffer = NULL; 802 buffer = NULL;
404} 803}
405 804
@@ -414,46 +813,57 @@ void swsusp_free(void)
414static int enough_free_mem(unsigned int nr_pages) 813static int enough_free_mem(unsigned int nr_pages)
415{ 814{
416 struct zone *zone; 815 struct zone *zone;
417 unsigned int n = 0; 816 unsigned int free = 0, meta = 0;
418 817
419 for_each_zone (zone) 818 for_each_zone (zone)
420 if (!is_highmem(zone)) 819 if (!is_highmem(zone)) {
421 n += zone->free_pages; 820 free += zone->free_pages;
422 pr_debug("swsusp: available memory: %u pages\n", n); 821 meta += snapshot_additional_pages(zone);
423 return n > (nr_pages + PAGES_FOR_IO + 822 }
424 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
425}
426 823
427static int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed) 824 pr_debug("swsusp: pages needed: %u + %u + %u, available pages: %u\n",
428{ 825 nr_pages, PAGES_FOR_IO, meta, free);
429 struct pbe *p;
430 826
431 for_each_pbe (p, pblist) { 827 return free > nr_pages + PAGES_FOR_IO + meta;
432 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
433 if (!p->address)
434 return -ENOMEM;
435 }
436 return 0;
437} 828}
438 829
439static struct pbe *swsusp_alloc(unsigned int nr_pages) 830static int
831swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm,
832 unsigned int nr_pages)
440{ 833{
441 struct pbe *pblist; 834 int error;
442 835
443 if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) { 836 error = memory_bm_create(orig_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY);
444 printk(KERN_ERR "suspend: Allocating pagedir failed.\n"); 837 if (error)
445 return NULL; 838 goto Free;
446 }
447 839
448 if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) { 840 error = memory_bm_create(copy_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY);
449 printk(KERN_ERR "suspend: Allocating image pages failed.\n"); 841 if (error)
450 swsusp_free(); 842 goto Free;
451 return NULL; 843
844 while (nr_pages-- > 0) {
845 struct page *page = alloc_page(GFP_ATOMIC | __GFP_COLD);
846 if (!page)
847 goto Free;
848
849 SetPageNosave(page);
850 SetPageNosaveFree(page);
851 memory_bm_set_bit(copy_bm, page_to_pfn(page));
452 } 852 }
853 return 0;
453 854
454 return pblist; 855Free:
856 swsusp_free();
857 return -ENOMEM;
455} 858}
456 859
860/* Memory bitmap used for marking saveable pages */
861static struct memory_bitmap orig_bm;
862/* Memory bitmap used for marking allocated pages that will contain the copies
863 * of saveable pages
864 */
865static struct memory_bitmap copy_bm;
866
457asmlinkage int swsusp_save(void) 867asmlinkage int swsusp_save(void)
458{ 868{
459 unsigned int nr_pages; 869 unsigned int nr_pages;
@@ -464,25 +874,19 @@ asmlinkage int swsusp_save(void)
464 nr_pages = count_data_pages(); 874 nr_pages = count_data_pages();
465 printk("swsusp: Need to copy %u pages\n", nr_pages); 875 printk("swsusp: Need to copy %u pages\n", nr_pages);
466 876
467 pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
468 nr_pages,
469 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
470 PAGES_FOR_IO, nr_free_pages());
471
472 if (!enough_free_mem(nr_pages)) { 877 if (!enough_free_mem(nr_pages)) {
473 printk(KERN_ERR "swsusp: Not enough free memory\n"); 878 printk(KERN_ERR "swsusp: Not enough free memory\n");
474 return -ENOMEM; 879 return -ENOMEM;
475 } 880 }
476 881
477 pagedir_nosave = swsusp_alloc(nr_pages); 882 if (swsusp_alloc(&orig_bm, &copy_bm, nr_pages))
478 if (!pagedir_nosave)
479 return -ENOMEM; 883 return -ENOMEM;
480 884
481 /* During allocating of suspend pagedir, new cold pages may appear. 885 /* During allocating of suspend pagedir, new cold pages may appear.
482 * Kill them. 886 * Kill them.
483 */ 887 */
484 drain_local_pages(); 888 drain_local_pages();
485 copy_data_pages(pagedir_nosave); 889 copy_data_pages(&copy_bm, &orig_bm);
486 890
487 /* 891 /*
488 * End of critical section. From now on, we can write to memory, 892 * End of critical section. From now on, we can write to memory,
@@ -511,22 +915,20 @@ static void init_header(struct swsusp_info *info)
511} 915}
512 916
513/** 917/**
514 * pack_orig_addresses - the .orig_address fields of the PBEs from the 918 * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm
515 * list starting at @pbe are stored in the array @buf[] (1 page) 919 * are stored in the array @buf[] (1 page at a time)
516 */ 920 */
517 921
518static inline struct pbe *pack_orig_addresses(unsigned long *buf, struct pbe *pbe) 922static inline void
923pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
519{ 924{
520 int j; 925 int j;
521 926
522 for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) { 927 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
523 buf[j] = pbe->orig_address; 928 buf[j] = memory_bm_next_pfn(bm);
524 pbe = pbe->next; 929 if (unlikely(buf[j] == BM_END_OF_MAP))
930 break;
525 } 931 }
526 if (!pbe)
527 for (; j < PAGE_SIZE / sizeof(long); j++)
528 buf[j] = 0;
529 return pbe;
530} 932}
531 933
532/** 934/**
@@ -553,37 +955,39 @@ static inline struct pbe *pack_orig_addresses(unsigned long *buf, struct pbe *pb
553 955
554int snapshot_read_next(struct snapshot_handle *handle, size_t count) 956int snapshot_read_next(struct snapshot_handle *handle, size_t count)
555{ 957{
556 if (handle->page > nr_meta_pages + nr_copy_pages) 958 if (handle->cur > nr_meta_pages + nr_copy_pages)
557 return 0; 959 return 0;
960
558 if (!buffer) { 961 if (!buffer) {
559 /* This makes the buffer be freed by swsusp_free() */ 962 /* This makes the buffer be freed by swsusp_free() */
560 buffer = alloc_image_page(GFP_ATOMIC, 0); 963 buffer = alloc_image_page(GFP_ATOMIC, PG_ANY);
561 if (!buffer) 964 if (!buffer)
562 return -ENOMEM; 965 return -ENOMEM;
563 } 966 }
564 if (!handle->offset) { 967 if (!handle->offset) {
565 init_header((struct swsusp_info *)buffer); 968 init_header((struct swsusp_info *)buffer);
566 handle->buffer = buffer; 969 handle->buffer = buffer;
567 handle->pbe = pagedir_nosave; 970 memory_bm_position_reset(&orig_bm);
971 memory_bm_position_reset(&copy_bm);
568 } 972 }
569 if (handle->prev < handle->page) { 973 if (handle->prev < handle->cur) {
570 if (handle->page <= nr_meta_pages) { 974 if (handle->cur <= nr_meta_pages) {
571 handle->pbe = pack_orig_addresses(buffer, handle->pbe); 975 memset(buffer, 0, PAGE_SIZE);
572 if (!handle->pbe) 976 pack_pfns(buffer, &orig_bm);
573 handle->pbe = pagedir_nosave;
574 } else { 977 } else {
575 handle->buffer = (void *)handle->pbe->address; 978 unsigned long pfn = memory_bm_next_pfn(&copy_bm);
576 handle->pbe = handle->pbe->next; 979
980 handle->buffer = page_address(pfn_to_page(pfn));
577 } 981 }
578 handle->prev = handle->page; 982 handle->prev = handle->cur;
579 } 983 }
580 handle->buf_offset = handle->page_offset; 984 handle->buf_offset = handle->cur_offset;
581 if (handle->page_offset + count >= PAGE_SIZE) { 985 if (handle->cur_offset + count >= PAGE_SIZE) {
582 count = PAGE_SIZE - handle->page_offset; 986 count = PAGE_SIZE - handle->cur_offset;
583 handle->page_offset = 0; 987 handle->cur_offset = 0;
584 handle->page++; 988 handle->cur++;
585 } else { 989 } else {
586 handle->page_offset += count; 990 handle->cur_offset += count;
587 } 991 }
588 handle->offset += count; 992 handle->offset += count;
589 return count; 993 return count;
@@ -595,47 +999,50 @@ int snapshot_read_next(struct snapshot_handle *handle, size_t count)
595 * had been used before suspend 999 * had been used before suspend
596 */ 1000 */
597 1001
598static int mark_unsafe_pages(struct pbe *pblist) 1002static int mark_unsafe_pages(struct memory_bitmap *bm)
599{ 1003{
600 struct zone *zone; 1004 struct zone *zone;
601 unsigned long zone_pfn; 1005 unsigned long pfn, max_zone_pfn;
602 struct pbe *p;
603
604 if (!pblist) /* a sanity check */
605 return -EINVAL;
606 1006
607 /* Clear page flags */ 1007 /* Clear page flags */
608 for_each_zone (zone) { 1008 for_each_zone (zone) {
609 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) 1009 max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
610 if (pfn_valid(zone_pfn + zone->zone_start_pfn)) 1010 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
611 ClearPageNosaveFree(pfn_to_page(zone_pfn + 1011 if (pfn_valid(pfn))
612 zone->zone_start_pfn)); 1012 ClearPageNosaveFree(pfn_to_page(pfn));
613 } 1013 }
614 1014
615 /* Mark orig addresses */ 1015 /* Mark pages that correspond to the "original" pfns as "unsafe" */
616 for_each_pbe (p, pblist) { 1016 memory_bm_position_reset(bm);
617 if (virt_addr_valid(p->orig_address)) 1017 do {
618 SetPageNosaveFree(virt_to_page(p->orig_address)); 1018 pfn = memory_bm_next_pfn(bm);
619 else 1019 if (likely(pfn != BM_END_OF_MAP)) {
620 return -EFAULT; 1020 if (likely(pfn_valid(pfn)))
621 } 1021 SetPageNosaveFree(pfn_to_page(pfn));
1022 else
1023 return -EFAULT;
1024 }
1025 } while (pfn != BM_END_OF_MAP);
622 1026
623 unsafe_pages = 0; 1027 allocated_unsafe_pages = 0;
624 1028
625 return 0; 1029 return 0;
626} 1030}
627 1031
628static void copy_page_backup_list(struct pbe *dst, struct pbe *src) 1032static void
1033duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src)
629{ 1034{
630 /* We assume both lists contain the same number of elements */ 1035 unsigned long pfn;
631 while (src) { 1036
632 dst->orig_address = src->orig_address; 1037 memory_bm_position_reset(src);
633 dst = dst->next; 1038 pfn = memory_bm_next_pfn(src);
634 src = src->next; 1039 while (pfn != BM_END_OF_MAP) {
1040 memory_bm_set_bit(dst, pfn);
1041 pfn = memory_bm_next_pfn(src);
635 } 1042 }
636} 1043}
637 1044
638static int check_header(struct swsusp_info *info) 1045static inline int check_header(struct swsusp_info *info)
639{ 1046{
640 char *reason = NULL; 1047 char *reason = NULL;
641 1048
@@ -662,19 +1069,14 @@ static int check_header(struct swsusp_info *info)
662 * load header - check the image header and copy data from it 1069 * load header - check the image header and copy data from it
663 */ 1070 */
664 1071
665static int load_header(struct snapshot_handle *handle, 1072static int
666 struct swsusp_info *info) 1073load_header(struct swsusp_info *info)
667{ 1074{
668 int error; 1075 int error;
669 struct pbe *pblist;
670 1076
1077 restore_pblist = NULL;
671 error = check_header(info); 1078 error = check_header(info);
672 if (!error) { 1079 if (!error) {
673 pblist = alloc_pagedir(info->image_pages, GFP_ATOMIC, 0);
674 if (!pblist)
675 return -ENOMEM;
676 pagedir_nosave = pblist;
677 handle->pbe = pblist;
678 nr_copy_pages = info->image_pages; 1080 nr_copy_pages = info->image_pages;
679 nr_meta_pages = info->pages - info->image_pages - 1; 1081 nr_meta_pages = info->pages - info->image_pages - 1;
680 } 1082 }
@@ -682,113 +1084,137 @@ static int load_header(struct snapshot_handle *handle,
682} 1084}
683 1085
684/** 1086/**
685 * unpack_orig_addresses - copy the elements of @buf[] (1 page) to 1087 * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set
686 * the PBEs in the list starting at @pbe 1088 * the corresponding bit in the memory bitmap @bm
687 */ 1089 */
688 1090
689static inline struct pbe *unpack_orig_addresses(unsigned long *buf, 1091static inline void
690 struct pbe *pbe) 1092unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
691{ 1093{
692 int j; 1094 int j;
693 1095
694 for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) { 1096 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
695 pbe->orig_address = buf[j]; 1097 if (unlikely(buf[j] == BM_END_OF_MAP))
696 pbe = pbe->next; 1098 break;
1099
1100 memory_bm_set_bit(bm, buf[j]);
697 } 1101 }
698 return pbe;
699} 1102}
700 1103
701/** 1104/**
702 * prepare_image - use metadata contained in the PBE list 1105 * prepare_image - use the memory bitmap @bm to mark the pages that will
703 * pointed to by pagedir_nosave to mark the pages that will 1106 * be overwritten in the process of restoring the system memory state
704 * be overwritten in the process of restoring the system 1107 * from the suspend image ("unsafe" pages) and allocate memory for the
705 * memory state from the image ("unsafe" pages) and allocate 1108 * image.
706 * memory for the image
707 * 1109 *
708 * The idea is to allocate the PBE list first and then 1110 * The idea is to allocate a new memory bitmap first and then allocate
709 * allocate as many pages as it's needed for the image data, 1111 * as many pages as needed for the image data, but not to assign these
710 * but not to assign these pages to the PBEs initially. 1112 * pages to specific tasks initially. Instead, we just mark them as
711 * Instead, we just mark them as allocated and create a list 1113 * allocated and create a list of "safe" pages that will be used later.
712 * of "safe" which will be used later
713 */ 1114 */
714 1115
715struct safe_page { 1116#define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
716 struct safe_page *next;
717 char padding[PAGE_SIZE - sizeof(void *)];
718};
719 1117
720static struct safe_page *safe_pages; 1118static struct linked_page *safe_pages_list;
721 1119
722static int prepare_image(struct snapshot_handle *handle) 1120static int
1121prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
723{ 1122{
724 int error = 0; 1123 unsigned int nr_pages;
725 unsigned int nr_pages = nr_copy_pages; 1124 struct linked_page *sp_list, *lp;
726 struct pbe *p, *pblist = NULL; 1125 int error;
727 1126
728 p = pagedir_nosave; 1127 error = mark_unsafe_pages(bm);
729 error = mark_unsafe_pages(p); 1128 if (error)
730 if (!error) { 1129 goto Free;
731 pblist = alloc_pagedir(nr_pages, GFP_ATOMIC, 1); 1130
732 if (pblist) 1131 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
733 copy_page_backup_list(pblist, p); 1132 if (error)
734 free_pagedir(p, 0); 1133 goto Free;
735 if (!pblist) 1134
1135 duplicate_memory_bitmap(new_bm, bm);
1136 memory_bm_free(bm, PG_UNSAFE_KEEP);
1137 /* Reserve some safe pages for potential later use.
1138 *
1139 * NOTE: This way we make sure there will be enough safe pages for the
1140 * chain_alloc() in get_buffer(). It is a bit wasteful, but
1141 * nr_copy_pages cannot be greater than 50% of the memory anyway.
1142 */
1143 sp_list = NULL;
1144 /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */
1145 nr_pages = nr_copy_pages - allocated_unsafe_pages;
1146 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
1147 while (nr_pages > 0) {
1148 lp = alloc_image_page(GFP_ATOMIC, PG_SAFE);
1149 if (!lp) {
736 error = -ENOMEM; 1150 error = -ENOMEM;
1151 goto Free;
1152 }
1153 lp->next = sp_list;
1154 sp_list = lp;
1155 nr_pages--;
737 } 1156 }
738 safe_pages = NULL; 1157 /* Preallocate memory for the image */
739 if (!error && nr_pages > unsafe_pages) { 1158 safe_pages_list = NULL;
740 nr_pages -= unsafe_pages; 1159 nr_pages = nr_copy_pages - allocated_unsafe_pages;
741 while (nr_pages--) { 1160 while (nr_pages > 0) {
742 struct safe_page *ptr; 1161 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
743 1162 if (!lp) {
744 ptr = (struct safe_page *)get_zeroed_page(GFP_ATOMIC); 1163 error = -ENOMEM;
745 if (!ptr) { 1164 goto Free;
746 error = -ENOMEM; 1165 }
747 break; 1166 if (!PageNosaveFree(virt_to_page(lp))) {
748 } 1167 /* The page is "safe", add it to the list */
749 if (!PageNosaveFree(virt_to_page(ptr))) { 1168 lp->next = safe_pages_list;
750 /* The page is "safe", add it to the list */ 1169 safe_pages_list = lp;
751 ptr->next = safe_pages;
752 safe_pages = ptr;
753 }
754 /* Mark the page as allocated */
755 SetPageNosave(virt_to_page(ptr));
756 SetPageNosaveFree(virt_to_page(ptr));
757 } 1170 }
1171 /* Mark the page as allocated */
1172 SetPageNosave(virt_to_page(lp));
1173 SetPageNosaveFree(virt_to_page(lp));
1174 nr_pages--;
758 } 1175 }
759 if (!error) { 1176 /* Free the reserved safe pages so that chain_alloc() can use them */
760 pagedir_nosave = pblist; 1177 while (sp_list) {
761 } else { 1178 lp = sp_list->next;
762 handle->pbe = NULL; 1179 free_image_page(sp_list, PG_UNSAFE_CLEAR);
763 swsusp_free(); 1180 sp_list = lp;
764 } 1181 }
1182 return 0;
1183
1184Free:
1185 swsusp_free();
765 return error; 1186 return error;
766} 1187}
767 1188
768static void *get_buffer(struct snapshot_handle *handle) 1189/**
1190 * get_buffer - compute the address that snapshot_write_next() should
1191 * set for its caller to write to.
1192 */
1193
1194static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
769{ 1195{
770 struct pbe *pbe = handle->pbe, *last = handle->last_pbe; 1196 struct pbe *pbe;
771 struct page *page = virt_to_page(pbe->orig_address); 1197 struct page *page = pfn_to_page(memory_bm_next_pfn(bm));
772 1198
773 if (PageNosave(page) && PageNosaveFree(page)) { 1199 if (PageNosave(page) && PageNosaveFree(page))
774 /* 1200 /* We have allocated the "original" page frame and we can
775 * We have allocated the "original" page frame and we can 1201 * use it directly to store the loaded page.
776 * use it directly to store the read page
777 */ 1202 */
778 pbe->address = 0; 1203 return page_address(page);
779 if (last && last->next) 1204
780 last->next = NULL; 1205 /* The "original" page frame has not been allocated and we have to
781 return (void *)pbe->orig_address; 1206 * use a "safe" page frame to store the loaded page.
782 }
783 /*
784 * The "original" page frame has not been allocated and we have to
785 * use a "safe" page frame to store the read page
786 */ 1207 */
787 pbe->address = (unsigned long)safe_pages; 1208 pbe = chain_alloc(ca, sizeof(struct pbe));
788 safe_pages = safe_pages->next; 1209 if (!pbe) {
789 if (last) 1210 swsusp_free();
790 last->next = pbe; 1211 return NULL;
791 handle->last_pbe = pbe; 1212 }
1213 pbe->orig_address = (unsigned long)page_address(page);
1214 pbe->address = (unsigned long)safe_pages_list;
1215 safe_pages_list = safe_pages_list->next;
1216 pbe->next = restore_pblist;
1217 restore_pblist = pbe;
792 return (void *)pbe->address; 1218 return (void *)pbe->address;
793} 1219}
794 1220
@@ -816,46 +1242,60 @@ static void *get_buffer(struct snapshot_handle *handle)
816 1242
817int snapshot_write_next(struct snapshot_handle *handle, size_t count) 1243int snapshot_write_next(struct snapshot_handle *handle, size_t count)
818{ 1244{
1245 static struct chain_allocator ca;
819 int error = 0; 1246 int error = 0;
820 1247
821 if (handle->prev && handle->page > nr_meta_pages + nr_copy_pages) 1248 /* Check if we have already loaded the entire image */
1249 if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages)
822 return 0; 1250 return 0;
1251
823 if (!buffer) { 1252 if (!buffer) {
824 /* This makes the buffer be freed by swsusp_free() */ 1253 /* This makes the buffer be freed by swsusp_free() */
825 buffer = alloc_image_page(GFP_ATOMIC, 0); 1254 buffer = alloc_image_page(GFP_ATOMIC, PG_ANY);
826 if (!buffer) 1255 if (!buffer)
827 return -ENOMEM; 1256 return -ENOMEM;
828 } 1257 }
829 if (!handle->offset) 1258 if (!handle->offset)
830 handle->buffer = buffer; 1259 handle->buffer = buffer;
831 if (handle->prev < handle->page) { 1260 handle->sync_read = 1;
832 if (!handle->prev) { 1261 if (handle->prev < handle->cur) {
833 error = load_header(handle, (struct swsusp_info *)buffer); 1262 if (handle->prev == 0) {
1263 error = load_header(buffer);
834 if (error) 1264 if (error)
835 return error; 1265 return error;
1266
1267 error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
1268 if (error)
1269 return error;
1270
836 } else if (handle->prev <= nr_meta_pages) { 1271 } else if (handle->prev <= nr_meta_pages) {
837 handle->pbe = unpack_orig_addresses(buffer, handle->pbe); 1272 unpack_orig_pfns(buffer, &copy_bm);
838 if (!handle->pbe) { 1273 if (handle->prev == nr_meta_pages) {
839 error = prepare_image(handle); 1274 error = prepare_image(&orig_bm, &copy_bm);
840 if (error) 1275 if (error)
841 return error; 1276 return error;
842 handle->pbe = pagedir_nosave; 1277
843 handle->last_pbe = NULL; 1278 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
844 handle->buffer = get_buffer(handle); 1279 memory_bm_position_reset(&orig_bm);
1280 restore_pblist = NULL;
1281 handle->buffer = get_buffer(&orig_bm, &ca);
1282 handle->sync_read = 0;
1283 if (!handle->buffer)
1284 return -ENOMEM;
845 } 1285 }
846 } else { 1286 } else {
847 handle->pbe = handle->pbe->next; 1287 handle->buffer = get_buffer(&orig_bm, &ca);
848 handle->buffer = get_buffer(handle); 1288 handle->sync_read = 0;
849 } 1289 }
850 handle->prev = handle->page; 1290 handle->prev = handle->cur;
851 } 1291 }
852 handle->buf_offset = handle->page_offset; 1292 handle->buf_offset = handle->cur_offset;
853 if (handle->page_offset + count >= PAGE_SIZE) { 1293 if (handle->cur_offset + count >= PAGE_SIZE) {
854 count = PAGE_SIZE - handle->page_offset; 1294 count = PAGE_SIZE - handle->cur_offset;
855 handle->page_offset = 0; 1295 handle->cur_offset = 0;
856 handle->page++; 1296 handle->cur++;
857 } else { 1297 } else {
858 handle->page_offset += count; 1298 handle->cur_offset += count;
859 } 1299 }
860 handle->offset += count; 1300 handle->offset += count;
861 return count; 1301 return count;
@@ -863,6 +1303,13 @@ int snapshot_write_next(struct snapshot_handle *handle, size_t count)
863 1303
864int snapshot_image_loaded(struct snapshot_handle *handle) 1304int snapshot_image_loaded(struct snapshot_handle *handle)
865{ 1305{
866 return !(!handle->pbe || handle->pbe->next || !nr_copy_pages || 1306 return !(!nr_copy_pages ||
867 handle->page <= nr_meta_pages + nr_copy_pages); 1307 handle->cur <= nr_meta_pages + nr_copy_pages);
1308}
1309
1310void snapshot_free_unused_memory(struct snapshot_handle *handle)
1311{
1312 /* Free only if we have loaded the image entirely */
1313 if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages)
1314 memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR);
868} 1315}