Commit | Line | Data |
---|---|---|
25761b6e | 1 | /* |
96bc7aec | 2 | * linux/kernel/power/snapshot.c |
25761b6e | 3 | * |
8357376d | 4 | * This file provides system snapshot/restore functionality for swsusp. |
25761b6e RW |
5 | * |
6 | * Copyright (C) 1998-2005 Pavel Machek <pavel@suse.cz> | |
8357376d | 7 | * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl> |
25761b6e | 8 | * |
8357376d | 9 | * This file is released under the GPLv2. |
25761b6e RW |
10 | * |
11 | */ | |
12 | ||
f577eb30 | 13 | #include <linux/version.h> |
25761b6e RW |
14 | #include <linux/module.h> |
15 | #include <linux/mm.h> | |
16 | #include <linux/suspend.h> | |
25761b6e | 17 | #include <linux/delay.h> |
25761b6e | 18 | #include <linux/bitops.h> |
25761b6e | 19 | #include <linux/spinlock.h> |
25761b6e | 20 | #include <linux/kernel.h> |
25761b6e RW |
21 | #include <linux/pm.h> |
22 | #include <linux/device.h> | |
74dfd666 | 23 | #include <linux/init.h> |
25761b6e RW |
24 | #include <linux/bootmem.h> |
25 | #include <linux/syscalls.h> | |
26 | #include <linux/console.h> | |
27 | #include <linux/highmem.h> | |
25761b6e RW |
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 | ||
25761b6e RW |
35 | #include "power.h" |
36 | ||
74dfd666 RW |
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 | ||
8357376d RW |
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 | */ | |
75534b50 RW |
46 | struct pbe *restore_pblist; |
47 | ||
8357376d | 48 | /* Pointer to an auxiliary buffer (1 page) */ |
940864dd | 49 | static void *buffer; |
7088a5c0 | 50 | |
f6143aa6 RW |
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 | |
8357376d RW |
54 | * used before suspend. The unsafe pages have PageNosaveFree set |
55 | * and we count them using unsafe_pages. | |
f6143aa6 | 56 | * |
8357376d RW |
57 | * Each allocated image page is marked as PageNosave and PageNosaveFree |
58 | * so that swsusp_free() can release it. | |
f6143aa6 RW |
59 | */ |
60 | ||
0bcd888d RW |
61 | #define PG_ANY 0 |
62 | #define PG_SAFE 1 | |
63 | #define PG_UNSAFE_CLEAR 1 | |
64 | #define PG_UNSAFE_KEEP 0 | |
65 | ||
940864dd | 66 | static unsigned int allocated_unsafe_pages; |
f6143aa6 | 67 | |
8357376d | 68 | static void *get_image_page(gfp_t gfp_mask, int safe_needed) |
f6143aa6 RW |
69 | { |
70 | void *res; | |
71 | ||
72 | res = (void *)get_zeroed_page(gfp_mask); | |
73 | if (safe_needed) | |
7be98234 | 74 | while (res && swsusp_page_is_free(virt_to_page(res))) { |
f6143aa6 | 75 | /* The page is unsafe, mark it for swsusp_free() */ |
7be98234 | 76 | swsusp_set_page_forbidden(virt_to_page(res)); |
940864dd | 77 | allocated_unsafe_pages++; |
f6143aa6 RW |
78 | res = (void *)get_zeroed_page(gfp_mask); |
79 | } | |
80 | if (res) { | |
7be98234 RW |
81 | swsusp_set_page_forbidden(virt_to_page(res)); |
82 | swsusp_set_page_free(virt_to_page(res)); | |
f6143aa6 RW |
83 | } |
84 | return res; | |
85 | } | |
86 | ||
87 | unsigned long get_safe_page(gfp_t gfp_mask) | |
88 | { | |
8357376d RW |
89 | return (unsigned long)get_image_page(gfp_mask, PG_SAFE); |
90 | } | |
91 | ||
5b6d15de RW |
92 | static struct page *alloc_image_page(gfp_t gfp_mask) |
93 | { | |
8357376d RW |
94 | struct page *page; |
95 | ||
96 | page = alloc_page(gfp_mask); | |
97 | if (page) { | |
7be98234 RW |
98 | swsusp_set_page_forbidden(page); |
99 | swsusp_set_page_free(page); | |
8357376d RW |
100 | } |
101 | return page; | |
f6143aa6 RW |
102 | } |
103 | ||
104 | /** | |
105 | * free_image_page - free page represented by @addr, allocated with | |
8357376d | 106 | * get_image_page (page flags set by it must be cleared) |
f6143aa6 RW |
107 | */ |
108 | ||
109 | static inline void free_image_page(void *addr, int clear_nosave_free) | |
110 | { | |
8357376d RW |
111 | struct page *page; |
112 | ||
113 | BUG_ON(!virt_addr_valid(addr)); | |
114 | ||
115 | page = virt_to_page(addr); | |
116 | ||
7be98234 | 117 | swsusp_unset_page_forbidden(page); |
f6143aa6 | 118 | if (clear_nosave_free) |
7be98234 | 119 | swsusp_unset_page_free(page); |
8357376d RW |
120 | |
121 | __free_page(page); | |
f6143aa6 RW |
122 | } |
123 | ||
b788db79 RW |
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 | ||
8357376d | 182 | lp = get_image_page(ca->gfp_mask, ca->safe_needed); |
b788db79 RW |
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. | |
b788db79 RW |
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; | |
8357376d RW |
354 | for_each_zone(zone) |
355 | if (populated_zone(zone)) | |
b788db79 RW |
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 */ | |
8357376d | 367 | for_each_zone(zone) { |
b788db79 RW |
368 | unsigned long pfn; |
369 | ||
8357376d | 370 | if (!populated_zone(zone)) |
b788db79 RW |
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 | ||
8357376d | 389 | ptr = get_image_page(gfp_mask, safe_needed); |
b788db79 RW |
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 | ||
59a49335 | 413 | Free: |
b788db79 RW |
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 | /** | |
74dfd666 | 445 | * memory_bm_find_bit - find the bit in the bitmap @bm that corresponds |
b788db79 RW |
446 | * to given pfn. The cur_zone_bm member of @bm and the cur_block member |
447 | * of @bm->cur_zone_bm are updated. | |
b788db79 RW |
448 | */ |
449 | ||
74dfd666 RW |
450 | static void memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn, |
451 | void **addr, unsigned int *bit_nr) | |
b788db79 RW |
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; | |
74dfd666 RW |
463 | |
464 | BUG_ON(!zone_bm); | |
b788db79 RW |
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; | |
74dfd666 RW |
475 | |
476 | BUG_ON(!bb); | |
b788db79 RW |
477 | } |
478 | zone_bm->cur_block = bb; | |
479 | pfn -= bb->start_pfn; | |
74dfd666 RW |
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); | |
b788db79 RW |
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 | ||
59a49335 | 584 | Return_pfn: |
b788db79 RW |
585 | bm->cur.chunk = chunk; |
586 | bm->cur.bit = bit; | |
587 | return bb->start_pfn + chunk * BM_BITS_PER_CHUNK + bit; | |
588 | } | |
589 | ||
74dfd666 RW |
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 | { | |
612 | struct nosave_region *region; | |
613 | ||
614 | if (start_pfn >= end_pfn) | |
615 | return; | |
616 | ||
617 | if (!list_empty(&nosave_regions)) { | |
618 | /* Try to extend the previous region (they should be sorted) */ | |
619 | region = list_entry(nosave_regions.prev, | |
620 | struct nosave_region, list); | |
621 | if (region->end_pfn == start_pfn) { | |
622 | region->end_pfn = end_pfn; | |
623 | goto Report; | |
624 | } | |
625 | } | |
626 | /* This allocation cannot fail */ | |
627 | region = alloc_bootmem_low(sizeof(struct nosave_region)); | |
628 | region->start_pfn = start_pfn; | |
629 | region->end_pfn = end_pfn; | |
630 | list_add_tail(®ion->list, &nosave_regions); | |
631 | Report: | |
632 | printk("swsusp: Registered nosave memory region: %016lx - %016lx\n", | |
633 | start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT); | |
634 | } | |
635 | ||
636 | /* | |
637 | * Set bits in this map correspond to the page frames the contents of which | |
638 | * should not be saved during the suspend. | |
639 | */ | |
640 | static struct memory_bitmap *forbidden_pages_map; | |
641 | ||
642 | /* Set bits in this map correspond to free page frames. */ | |
643 | static struct memory_bitmap *free_pages_map; | |
644 | ||
645 | /* | |
646 | * Each page frame allocated for creating the image is marked by setting the | |
647 | * corresponding bits in forbidden_pages_map and free_pages_map simultaneously | |
648 | */ | |
649 | ||
650 | void swsusp_set_page_free(struct page *page) | |
651 | { | |
652 | if (free_pages_map) | |
653 | memory_bm_set_bit(free_pages_map, page_to_pfn(page)); | |
654 | } | |
655 | ||
656 | static int swsusp_page_is_free(struct page *page) | |
657 | { | |
658 | return free_pages_map ? | |
659 | memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0; | |
660 | } | |
661 | ||
662 | void swsusp_unset_page_free(struct page *page) | |
663 | { | |
664 | if (free_pages_map) | |
665 | memory_bm_clear_bit(free_pages_map, page_to_pfn(page)); | |
666 | } | |
667 | ||
668 | static void swsusp_set_page_forbidden(struct page *page) | |
669 | { | |
670 | if (forbidden_pages_map) | |
671 | memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page)); | |
672 | } | |
673 | ||
674 | int swsusp_page_is_forbidden(struct page *page) | |
675 | { | |
676 | return forbidden_pages_map ? | |
677 | memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0; | |
678 | } | |
679 | ||
680 | static void swsusp_unset_page_forbidden(struct page *page) | |
681 | { | |
682 | if (forbidden_pages_map) | |
683 | memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page)); | |
684 | } | |
685 | ||
686 | /** | |
687 | * mark_nosave_pages - set bits corresponding to the page frames the | |
688 | * contents of which should not be saved in a given bitmap. | |
689 | */ | |
690 | ||
691 | static void mark_nosave_pages(struct memory_bitmap *bm) | |
692 | { | |
693 | struct nosave_region *region; | |
694 | ||
695 | if (list_empty(&nosave_regions)) | |
696 | return; | |
697 | ||
698 | list_for_each_entry(region, &nosave_regions, list) { | |
699 | unsigned long pfn; | |
700 | ||
701 | printk("swsusp: Marking nosave pages: %016lx - %016lx\n", | |
702 | region->start_pfn << PAGE_SHIFT, | |
703 | region->end_pfn << PAGE_SHIFT); | |
704 | ||
705 | for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++) | |
706 | memory_bm_set_bit(bm, pfn); | |
707 | } | |
708 | } | |
709 | ||
710 | /** | |
711 | * create_basic_memory_bitmaps - create bitmaps needed for marking page | |
712 | * frames that should not be saved and free page frames. The pointers | |
713 | * forbidden_pages_map and free_pages_map are only modified if everything | |
714 | * goes well, because we don't want the bits to be used before both bitmaps | |
715 | * are set up. | |
716 | */ | |
717 | ||
718 | int create_basic_memory_bitmaps(void) | |
719 | { | |
720 | struct memory_bitmap *bm1, *bm2; | |
721 | int error = 0; | |
722 | ||
723 | BUG_ON(forbidden_pages_map || free_pages_map); | |
724 | ||
0709db60 | 725 | bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
726 | if (!bm1) |
727 | return -ENOMEM; | |
728 | ||
0709db60 | 729 | error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
730 | if (error) |
731 | goto Free_first_object; | |
732 | ||
0709db60 | 733 | bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
734 | if (!bm2) |
735 | goto Free_first_bitmap; | |
736 | ||
0709db60 | 737 | error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
738 | if (error) |
739 | goto Free_second_object; | |
740 | ||
741 | forbidden_pages_map = bm1; | |
742 | free_pages_map = bm2; | |
743 | mark_nosave_pages(forbidden_pages_map); | |
744 | ||
745 | printk("swsusp: Basic memory bitmaps created\n"); | |
746 | ||
747 | return 0; | |
748 | ||
749 | Free_second_object: | |
750 | kfree(bm2); | |
751 | Free_first_bitmap: | |
752 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
753 | Free_first_object: | |
754 | kfree(bm1); | |
755 | return -ENOMEM; | |
756 | } | |
757 | ||
758 | /** | |
759 | * free_basic_memory_bitmaps - free memory bitmaps allocated by | |
760 | * create_basic_memory_bitmaps(). The auxiliary pointers are necessary | |
761 | * so that the bitmaps themselves are not referred to while they are being | |
762 | * freed. | |
763 | */ | |
764 | ||
765 | void free_basic_memory_bitmaps(void) | |
766 | { | |
767 | struct memory_bitmap *bm1, *bm2; | |
768 | ||
769 | BUG_ON(!(forbidden_pages_map && free_pages_map)); | |
770 | ||
771 | bm1 = forbidden_pages_map; | |
772 | bm2 = free_pages_map; | |
773 | forbidden_pages_map = NULL; | |
774 | free_pages_map = NULL; | |
775 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
776 | kfree(bm1); | |
777 | memory_bm_free(bm2, PG_UNSAFE_CLEAR); | |
778 | kfree(bm2); | |
779 | ||
780 | printk("swsusp: Basic memory bitmaps freed\n"); | |
781 | } | |
782 | ||
b788db79 RW |
783 | /** |
784 | * snapshot_additional_pages - estimate the number of additional pages | |
785 | * be needed for setting up the suspend image data structures for given | |
786 | * zone (usually the returned value is greater than the exact number) | |
787 | */ | |
788 | ||
789 | unsigned int snapshot_additional_pages(struct zone *zone) | |
790 | { | |
791 | unsigned int res; | |
792 | ||
793 | res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK); | |
794 | res += DIV_ROUND_UP(res * sizeof(struct bm_block), PAGE_SIZE); | |
8357376d | 795 | return 2 * res; |
b788db79 RW |
796 | } |
797 | ||
8357376d RW |
798 | #ifdef CONFIG_HIGHMEM |
799 | /** | |
800 | * count_free_highmem_pages - compute the total number of free highmem | |
801 | * pages, system-wide. | |
802 | */ | |
803 | ||
804 | static unsigned int count_free_highmem_pages(void) | |
805 | { | |
806 | struct zone *zone; | |
807 | unsigned int cnt = 0; | |
808 | ||
809 | for_each_zone(zone) | |
810 | if (populated_zone(zone) && is_highmem(zone)) | |
d23ad423 | 811 | cnt += zone_page_state(zone, NR_FREE_PAGES); |
8357376d RW |
812 | |
813 | return cnt; | |
814 | } | |
815 | ||
816 | /** | |
817 | * saveable_highmem_page - Determine whether a highmem page should be | |
818 | * included in the suspend image. | |
819 | * | |
820 | * We should save the page if it isn't Nosave or NosaveFree, or Reserved, | |
821 | * and it isn't a part of a free chunk of pages. | |
822 | */ | |
823 | ||
824 | static struct page *saveable_highmem_page(unsigned long pfn) | |
825 | { | |
826 | struct page *page; | |
827 | ||
828 | if (!pfn_valid(pfn)) | |
829 | return NULL; | |
830 | ||
831 | page = pfn_to_page(pfn); | |
832 | ||
833 | BUG_ON(!PageHighMem(page)); | |
834 | ||
7be98234 RW |
835 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) || |
836 | PageReserved(page)) | |
8357376d RW |
837 | return NULL; |
838 | ||
839 | return page; | |
840 | } | |
841 | ||
842 | /** | |
843 | * count_highmem_pages - compute the total number of saveable highmem | |
844 | * pages. | |
845 | */ | |
846 | ||
847 | unsigned int count_highmem_pages(void) | |
848 | { | |
849 | struct zone *zone; | |
850 | unsigned int n = 0; | |
851 | ||
852 | for_each_zone(zone) { | |
853 | unsigned long pfn, max_zone_pfn; | |
854 | ||
855 | if (!is_highmem(zone)) | |
856 | continue; | |
857 | ||
858 | mark_free_pages(zone); | |
859 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
860 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
861 | if (saveable_highmem_page(pfn)) | |
862 | n++; | |
863 | } | |
864 | return n; | |
865 | } | |
866 | #else | |
867 | static inline void *saveable_highmem_page(unsigned long pfn) { return NULL; } | |
868 | static inline unsigned int count_highmem_pages(void) { return 0; } | |
869 | #endif /* CONFIG_HIGHMEM */ | |
870 | ||
25761b6e | 871 | /** |
8357376d RW |
872 | * saveable - Determine whether a non-highmem page should be included in |
873 | * the suspend image. | |
25761b6e | 874 | * |
8357376d RW |
875 | * We should save the page if it isn't Nosave, and is not in the range |
876 | * of pages statically defined as 'unsaveable', and it isn't a part of | |
877 | * a free chunk of pages. | |
25761b6e RW |
878 | */ |
879 | ||
ae83c5ee | 880 | static struct page *saveable_page(unsigned long pfn) |
25761b6e | 881 | { |
de491861 | 882 | struct page *page; |
25761b6e RW |
883 | |
884 | if (!pfn_valid(pfn)) | |
ae83c5ee | 885 | return NULL; |
25761b6e RW |
886 | |
887 | page = pfn_to_page(pfn); | |
ae83c5ee | 888 | |
8357376d RW |
889 | BUG_ON(PageHighMem(page)); |
890 | ||
7be98234 | 891 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page)) |
ae83c5ee | 892 | return NULL; |
8357376d | 893 | |
72a97e08 | 894 | if (PageReserved(page) && pfn_is_nosave(pfn)) |
ae83c5ee | 895 | return NULL; |
25761b6e | 896 | |
ae83c5ee | 897 | return page; |
25761b6e RW |
898 | } |
899 | ||
8357376d RW |
900 | /** |
901 | * count_data_pages - compute the total number of saveable non-highmem | |
902 | * pages. | |
903 | */ | |
904 | ||
72a97e08 | 905 | unsigned int count_data_pages(void) |
25761b6e RW |
906 | { |
907 | struct zone *zone; | |
ae83c5ee | 908 | unsigned long pfn, max_zone_pfn; |
dc19d507 | 909 | unsigned int n = 0; |
25761b6e | 910 | |
8357376d | 911 | for_each_zone(zone) { |
25761b6e RW |
912 | if (is_highmem(zone)) |
913 | continue; | |
8357376d | 914 | |
25761b6e | 915 | mark_free_pages(zone); |
ae83c5ee RW |
916 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
917 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
8357376d RW |
918 | if(saveable_page(pfn)) |
919 | n++; | |
25761b6e | 920 | } |
a0f49651 | 921 | return n; |
25761b6e RW |
922 | } |
923 | ||
8357376d RW |
924 | /* This is needed, because copy_page and memcpy are not usable for copying |
925 | * task structs. | |
926 | */ | |
927 | static inline void do_copy_page(long *dst, long *src) | |
f623f0db RW |
928 | { |
929 | int n; | |
930 | ||
f623f0db RW |
931 | for (n = PAGE_SIZE / sizeof(long); n; n--) |
932 | *dst++ = *src++; | |
933 | } | |
934 | ||
8357376d RW |
935 | #ifdef CONFIG_HIGHMEM |
936 | static inline struct page * | |
937 | page_is_saveable(struct zone *zone, unsigned long pfn) | |
938 | { | |
939 | return is_highmem(zone) ? | |
940 | saveable_highmem_page(pfn) : saveable_page(pfn); | |
941 | } | |
942 | ||
943 | static inline void | |
944 | copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) | |
945 | { | |
946 | struct page *s_page, *d_page; | |
947 | void *src, *dst; | |
948 | ||
949 | s_page = pfn_to_page(src_pfn); | |
950 | d_page = pfn_to_page(dst_pfn); | |
951 | if (PageHighMem(s_page)) { | |
952 | src = kmap_atomic(s_page, KM_USER0); | |
953 | dst = kmap_atomic(d_page, KM_USER1); | |
954 | do_copy_page(dst, src); | |
955 | kunmap_atomic(src, KM_USER0); | |
956 | kunmap_atomic(dst, KM_USER1); | |
957 | } else { | |
958 | src = page_address(s_page); | |
959 | if (PageHighMem(d_page)) { | |
960 | /* Page pointed to by src may contain some kernel | |
961 | * data modified by kmap_atomic() | |
962 | */ | |
963 | do_copy_page(buffer, src); | |
964 | dst = kmap_atomic(pfn_to_page(dst_pfn), KM_USER0); | |
965 | memcpy(dst, buffer, PAGE_SIZE); | |
966 | kunmap_atomic(dst, KM_USER0); | |
967 | } else { | |
968 | dst = page_address(d_page); | |
969 | do_copy_page(dst, src); | |
970 | } | |
971 | } | |
972 | } | |
973 | #else | |
974 | #define page_is_saveable(zone, pfn) saveable_page(pfn) | |
975 | ||
976 | static inline void | |
977 | copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) | |
978 | { | |
979 | do_copy_page(page_address(pfn_to_page(dst_pfn)), | |
980 | page_address(pfn_to_page(src_pfn))); | |
981 | } | |
982 | #endif /* CONFIG_HIGHMEM */ | |
983 | ||
b788db79 RW |
984 | static void |
985 | copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm) | |
25761b6e RW |
986 | { |
987 | struct zone *zone; | |
b788db79 | 988 | unsigned long pfn; |
25761b6e | 989 | |
8357376d | 990 | for_each_zone(zone) { |
b788db79 RW |
991 | unsigned long max_zone_pfn; |
992 | ||
25761b6e | 993 | mark_free_pages(zone); |
ae83c5ee | 994 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
b788db79 | 995 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
8357376d | 996 | if (page_is_saveable(zone, pfn)) |
b788db79 | 997 | memory_bm_set_bit(orig_bm, pfn); |
25761b6e | 998 | } |
b788db79 RW |
999 | memory_bm_position_reset(orig_bm); |
1000 | memory_bm_position_reset(copy_bm); | |
1001 | do { | |
1002 | pfn = memory_bm_next_pfn(orig_bm); | |
8357376d RW |
1003 | if (likely(pfn != BM_END_OF_MAP)) |
1004 | copy_data_page(memory_bm_next_pfn(copy_bm), pfn); | |
b788db79 | 1005 | } while (pfn != BM_END_OF_MAP); |
25761b6e RW |
1006 | } |
1007 | ||
8357376d RW |
1008 | /* Total number of image pages */ |
1009 | static unsigned int nr_copy_pages; | |
1010 | /* Number of pages needed for saving the original pfns of the image pages */ | |
1011 | static unsigned int nr_meta_pages; | |
1012 | ||
25761b6e | 1013 | /** |
940864dd | 1014 | * swsusp_free - free pages allocated for the suspend. |
cd560bb2 | 1015 | * |
940864dd RW |
1016 | * Suspend pages are alocated before the atomic copy is made, so we |
1017 | * need to release them after the resume. | |
25761b6e RW |
1018 | */ |
1019 | ||
1020 | void swsusp_free(void) | |
1021 | { | |
1022 | struct zone *zone; | |
ae83c5ee | 1023 | unsigned long pfn, max_zone_pfn; |
25761b6e RW |
1024 | |
1025 | for_each_zone(zone) { | |
ae83c5ee RW |
1026 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
1027 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1028 | if (pfn_valid(pfn)) { | |
1029 | struct page *page = pfn_to_page(pfn); | |
1030 | ||
7be98234 RW |
1031 | if (swsusp_page_is_forbidden(page) && |
1032 | swsusp_page_is_free(page)) { | |
1033 | swsusp_unset_page_forbidden(page); | |
1034 | swsusp_unset_page_free(page); | |
8357376d | 1035 | __free_page(page); |
25761b6e RW |
1036 | } |
1037 | } | |
1038 | } | |
f577eb30 RW |
1039 | nr_copy_pages = 0; |
1040 | nr_meta_pages = 0; | |
75534b50 | 1041 | restore_pblist = NULL; |
6e1819d6 | 1042 | buffer = NULL; |
25761b6e RW |
1043 | } |
1044 | ||
8357376d RW |
1045 | #ifdef CONFIG_HIGHMEM |
1046 | /** | |
1047 | * count_pages_for_highmem - compute the number of non-highmem pages | |
1048 | * that will be necessary for creating copies of highmem pages. | |
1049 | */ | |
1050 | ||
1051 | static unsigned int count_pages_for_highmem(unsigned int nr_highmem) | |
1052 | { | |
1053 | unsigned int free_highmem = count_free_highmem_pages(); | |
1054 | ||
1055 | if (free_highmem >= nr_highmem) | |
1056 | nr_highmem = 0; | |
1057 | else | |
1058 | nr_highmem -= free_highmem; | |
1059 | ||
1060 | return nr_highmem; | |
1061 | } | |
1062 | #else | |
1063 | static unsigned int | |
1064 | count_pages_for_highmem(unsigned int nr_highmem) { return 0; } | |
1065 | #endif /* CONFIG_HIGHMEM */ | |
25761b6e RW |
1066 | |
1067 | /** | |
8357376d RW |
1068 | * enough_free_mem - Make sure we have enough free memory for the |
1069 | * snapshot image. | |
25761b6e RW |
1070 | */ |
1071 | ||
8357376d | 1072 | static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem) |
25761b6e | 1073 | { |
e5e2fa78 | 1074 | struct zone *zone; |
940864dd | 1075 | unsigned int free = 0, meta = 0; |
e5e2fa78 | 1076 | |
8357376d RW |
1077 | for_each_zone(zone) { |
1078 | meta += snapshot_additional_pages(zone); | |
1079 | if (!is_highmem(zone)) | |
d23ad423 | 1080 | free += zone_page_state(zone, NR_FREE_PAGES); |
8357376d | 1081 | } |
940864dd | 1082 | |
8357376d RW |
1083 | nr_pages += count_pages_for_highmem(nr_highmem); |
1084 | pr_debug("swsusp: Normal pages needed: %u + %u + %u, available pages: %u\n", | |
940864dd RW |
1085 | nr_pages, PAGES_FOR_IO, meta, free); |
1086 | ||
1087 | return free > nr_pages + PAGES_FOR_IO + meta; | |
25761b6e RW |
1088 | } |
1089 | ||
8357376d RW |
1090 | #ifdef CONFIG_HIGHMEM |
1091 | /** | |
1092 | * get_highmem_buffer - if there are some highmem pages in the suspend | |
1093 | * image, we may need the buffer to copy them and/or load their data. | |
1094 | */ | |
1095 | ||
1096 | static inline int get_highmem_buffer(int safe_needed) | |
1097 | { | |
1098 | buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed); | |
1099 | return buffer ? 0 : -ENOMEM; | |
1100 | } | |
1101 | ||
1102 | /** | |
1103 | * alloc_highmem_image_pages - allocate some highmem pages for the image. | |
1104 | * Try to allocate as many pages as needed, but if the number of free | |
1105 | * highmem pages is lesser than that, allocate them all. | |
1106 | */ | |
1107 | ||
1108 | static inline unsigned int | |
1109 | alloc_highmem_image_pages(struct memory_bitmap *bm, unsigned int nr_highmem) | |
1110 | { | |
1111 | unsigned int to_alloc = count_free_highmem_pages(); | |
1112 | ||
1113 | if (to_alloc > nr_highmem) | |
1114 | to_alloc = nr_highmem; | |
1115 | ||
1116 | nr_highmem -= to_alloc; | |
1117 | while (to_alloc-- > 0) { | |
1118 | struct page *page; | |
1119 | ||
1120 | page = alloc_image_page(__GFP_HIGHMEM); | |
1121 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1122 | } | |
1123 | return nr_highmem; | |
1124 | } | |
1125 | #else | |
1126 | static inline int get_highmem_buffer(int safe_needed) { return 0; } | |
1127 | ||
1128 | static inline unsigned int | |
1129 | alloc_highmem_image_pages(struct memory_bitmap *bm, unsigned int n) { return 0; } | |
1130 | #endif /* CONFIG_HIGHMEM */ | |
1131 | ||
1132 | /** | |
1133 | * swsusp_alloc - allocate memory for the suspend image | |
1134 | * | |
1135 | * We first try to allocate as many highmem pages as there are | |
1136 | * saveable highmem pages in the system. If that fails, we allocate | |
1137 | * non-highmem pages for the copies of the remaining highmem ones. | |
1138 | * | |
1139 | * In this approach it is likely that the copies of highmem pages will | |
1140 | * also be located in the high memory, because of the way in which | |
1141 | * copy_data_pages() works. | |
1142 | */ | |
1143 | ||
b788db79 RW |
1144 | static int |
1145 | swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm, | |
8357376d | 1146 | unsigned int nr_pages, unsigned int nr_highmem) |
054bd4c1 | 1147 | { |
b788db79 | 1148 | int error; |
054bd4c1 | 1149 | |
b788db79 RW |
1150 | error = memory_bm_create(orig_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY); |
1151 | if (error) | |
1152 | goto Free; | |
25761b6e | 1153 | |
b788db79 RW |
1154 | error = memory_bm_create(copy_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY); |
1155 | if (error) | |
1156 | goto Free; | |
25761b6e | 1157 | |
8357376d RW |
1158 | if (nr_highmem > 0) { |
1159 | error = get_highmem_buffer(PG_ANY); | |
1160 | if (error) | |
1161 | goto Free; | |
1162 | ||
1163 | nr_pages += alloc_highmem_image_pages(copy_bm, nr_highmem); | |
1164 | } | |
b788db79 | 1165 | while (nr_pages-- > 0) { |
8357376d RW |
1166 | struct page *page = alloc_image_page(GFP_ATOMIC | __GFP_COLD); |
1167 | ||
b788db79 RW |
1168 | if (!page) |
1169 | goto Free; | |
25761b6e | 1170 | |
b788db79 | 1171 | memory_bm_set_bit(copy_bm, page_to_pfn(page)); |
25761b6e | 1172 | } |
b788db79 | 1173 | return 0; |
25761b6e | 1174 | |
59a49335 | 1175 | Free: |
b788db79 RW |
1176 | swsusp_free(); |
1177 | return -ENOMEM; | |
25761b6e RW |
1178 | } |
1179 | ||
8357376d RW |
1180 | /* Memory bitmap used for marking saveable pages (during suspend) or the |
1181 | * suspend image pages (during resume) | |
1182 | */ | |
b788db79 | 1183 | static struct memory_bitmap orig_bm; |
8357376d RW |
1184 | /* Memory bitmap used on suspend for marking allocated pages that will contain |
1185 | * the copies of saveable pages. During resume it is initially used for | |
1186 | * marking the suspend image pages, but then its set bits are duplicated in | |
1187 | * @orig_bm and it is released. Next, on systems with high memory, it may be | |
1188 | * used for marking "safe" highmem pages, but it has to be reinitialized for | |
1189 | * this purpose. | |
b788db79 RW |
1190 | */ |
1191 | static struct memory_bitmap copy_bm; | |
1192 | ||
2e32a43e | 1193 | asmlinkage int swsusp_save(void) |
25761b6e | 1194 | { |
8357376d | 1195 | unsigned int nr_pages, nr_highmem; |
25761b6e | 1196 | |
8357376d | 1197 | printk("swsusp: critical section: \n"); |
25761b6e RW |
1198 | |
1199 | drain_local_pages(); | |
a0f49651 | 1200 | nr_pages = count_data_pages(); |
8357376d RW |
1201 | nr_highmem = count_highmem_pages(); |
1202 | printk("swsusp: Need to copy %u pages\n", nr_pages + nr_highmem); | |
25761b6e | 1203 | |
8357376d | 1204 | if (!enough_free_mem(nr_pages, nr_highmem)) { |
25761b6e RW |
1205 | printk(KERN_ERR "swsusp: Not enough free memory\n"); |
1206 | return -ENOMEM; | |
1207 | } | |
1208 | ||
8357376d RW |
1209 | if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) { |
1210 | printk(KERN_ERR "swsusp: Memory allocation failed\n"); | |
a0f49651 | 1211 | return -ENOMEM; |
8357376d | 1212 | } |
25761b6e RW |
1213 | |
1214 | /* During allocating of suspend pagedir, new cold pages may appear. | |
1215 | * Kill them. | |
1216 | */ | |
1217 | drain_local_pages(); | |
b788db79 | 1218 | copy_data_pages(©_bm, &orig_bm); |
25761b6e RW |
1219 | |
1220 | /* | |
1221 | * End of critical section. From now on, we can write to memory, | |
1222 | * but we should not touch disk. This specially means we must _not_ | |
1223 | * touch swap space! Except we must write out our image of course. | |
1224 | */ | |
1225 | ||
8357376d | 1226 | nr_pages += nr_highmem; |
a0f49651 | 1227 | nr_copy_pages = nr_pages; |
8357376d | 1228 | nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE); |
a0f49651 RW |
1229 | |
1230 | printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages); | |
8357376d | 1231 | |
25761b6e RW |
1232 | return 0; |
1233 | } | |
f577eb30 RW |
1234 | |
1235 | static void init_header(struct swsusp_info *info) | |
1236 | { | |
1237 | memset(info, 0, sizeof(struct swsusp_info)); | |
1238 | info->version_code = LINUX_VERSION_CODE; | |
1239 | info->num_physpages = num_physpages; | |
96b644bd | 1240 | memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname)); |
f577eb30 RW |
1241 | info->cpus = num_online_cpus(); |
1242 | info->image_pages = nr_copy_pages; | |
1243 | info->pages = nr_copy_pages + nr_meta_pages + 1; | |
6e1819d6 RW |
1244 | info->size = info->pages; |
1245 | info->size <<= PAGE_SHIFT; | |
f577eb30 RW |
1246 | } |
1247 | ||
1248 | /** | |
940864dd RW |
1249 | * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm |
1250 | * are stored in the array @buf[] (1 page at a time) | |
f577eb30 RW |
1251 | */ |
1252 | ||
b788db79 | 1253 | static inline void |
940864dd | 1254 | pack_pfns(unsigned long *buf, struct memory_bitmap *bm) |
f577eb30 RW |
1255 | { |
1256 | int j; | |
1257 | ||
b788db79 | 1258 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
940864dd RW |
1259 | buf[j] = memory_bm_next_pfn(bm); |
1260 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
b788db79 | 1261 | break; |
f577eb30 | 1262 | } |
f577eb30 RW |
1263 | } |
1264 | ||
1265 | /** | |
1266 | * snapshot_read_next - used for reading the system memory snapshot. | |
1267 | * | |
1268 | * On the first call to it @handle should point to a zeroed | |
1269 | * snapshot_handle structure. The structure gets updated and a pointer | |
1270 | * to it should be passed to this function every next time. | |
1271 | * | |
1272 | * The @count parameter should contain the number of bytes the caller | |
1273 | * wants to read from the snapshot. It must not be zero. | |
1274 | * | |
1275 | * On success the function returns a positive number. Then, the caller | |
1276 | * is allowed to read up to the returned number of bytes from the memory | |
1277 | * location computed by the data_of() macro. The number returned | |
1278 | * may be smaller than @count, but this only happens if the read would | |
1279 | * cross a page boundary otherwise. | |
1280 | * | |
1281 | * The function returns 0 to indicate the end of data stream condition, | |
1282 | * and a negative number is returned on error. In such cases the | |
1283 | * structure pointed to by @handle is not updated and should not be used | |
1284 | * any more. | |
1285 | */ | |
1286 | ||
1287 | int snapshot_read_next(struct snapshot_handle *handle, size_t count) | |
1288 | { | |
fb13a28b | 1289 | if (handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 1290 | return 0; |
b788db79 | 1291 | |
f577eb30 RW |
1292 | if (!buffer) { |
1293 | /* This makes the buffer be freed by swsusp_free() */ | |
8357376d | 1294 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); |
f577eb30 RW |
1295 | if (!buffer) |
1296 | return -ENOMEM; | |
1297 | } | |
1298 | if (!handle->offset) { | |
1299 | init_header((struct swsusp_info *)buffer); | |
1300 | handle->buffer = buffer; | |
b788db79 RW |
1301 | memory_bm_position_reset(&orig_bm); |
1302 | memory_bm_position_reset(©_bm); | |
f577eb30 | 1303 | } |
fb13a28b RW |
1304 | if (handle->prev < handle->cur) { |
1305 | if (handle->cur <= nr_meta_pages) { | |
b788db79 | 1306 | memset(buffer, 0, PAGE_SIZE); |
940864dd | 1307 | pack_pfns(buffer, &orig_bm); |
f577eb30 | 1308 | } else { |
8357376d | 1309 | struct page *page; |
b788db79 | 1310 | |
8357376d RW |
1311 | page = pfn_to_page(memory_bm_next_pfn(©_bm)); |
1312 | if (PageHighMem(page)) { | |
1313 | /* Highmem pages are copied to the buffer, | |
1314 | * because we can't return with a kmapped | |
1315 | * highmem page (we may not be called again). | |
1316 | */ | |
1317 | void *kaddr; | |
1318 | ||
1319 | kaddr = kmap_atomic(page, KM_USER0); | |
1320 | memcpy(buffer, kaddr, PAGE_SIZE); | |
1321 | kunmap_atomic(kaddr, KM_USER0); | |
1322 | handle->buffer = buffer; | |
1323 | } else { | |
1324 | handle->buffer = page_address(page); | |
1325 | } | |
f577eb30 | 1326 | } |
fb13a28b | 1327 | handle->prev = handle->cur; |
f577eb30 | 1328 | } |
fb13a28b RW |
1329 | handle->buf_offset = handle->cur_offset; |
1330 | if (handle->cur_offset + count >= PAGE_SIZE) { | |
1331 | count = PAGE_SIZE - handle->cur_offset; | |
1332 | handle->cur_offset = 0; | |
1333 | handle->cur++; | |
f577eb30 | 1334 | } else { |
fb13a28b | 1335 | handle->cur_offset += count; |
f577eb30 RW |
1336 | } |
1337 | handle->offset += count; | |
1338 | return count; | |
1339 | } | |
1340 | ||
1341 | /** | |
1342 | * mark_unsafe_pages - mark the pages that cannot be used for storing | |
1343 | * the image during resume, because they conflict with the pages that | |
1344 | * had been used before suspend | |
1345 | */ | |
1346 | ||
940864dd | 1347 | static int mark_unsafe_pages(struct memory_bitmap *bm) |
f577eb30 RW |
1348 | { |
1349 | struct zone *zone; | |
ae83c5ee | 1350 | unsigned long pfn, max_zone_pfn; |
f577eb30 RW |
1351 | |
1352 | /* Clear page flags */ | |
8357376d | 1353 | for_each_zone(zone) { |
ae83c5ee RW |
1354 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
1355 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1356 | if (pfn_valid(pfn)) | |
7be98234 | 1357 | swsusp_unset_page_free(pfn_to_page(pfn)); |
f577eb30 RW |
1358 | } |
1359 | ||
940864dd RW |
1360 | /* Mark pages that correspond to the "original" pfns as "unsafe" */ |
1361 | memory_bm_position_reset(bm); | |
1362 | do { | |
1363 | pfn = memory_bm_next_pfn(bm); | |
1364 | if (likely(pfn != BM_END_OF_MAP)) { | |
1365 | if (likely(pfn_valid(pfn))) | |
7be98234 | 1366 | swsusp_set_page_free(pfn_to_page(pfn)); |
940864dd RW |
1367 | else |
1368 | return -EFAULT; | |
1369 | } | |
1370 | } while (pfn != BM_END_OF_MAP); | |
f577eb30 | 1371 | |
940864dd | 1372 | allocated_unsafe_pages = 0; |
968808b8 | 1373 | |
f577eb30 RW |
1374 | return 0; |
1375 | } | |
1376 | ||
940864dd RW |
1377 | static void |
1378 | duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src) | |
f577eb30 | 1379 | { |
940864dd RW |
1380 | unsigned long pfn; |
1381 | ||
1382 | memory_bm_position_reset(src); | |
1383 | pfn = memory_bm_next_pfn(src); | |
1384 | while (pfn != BM_END_OF_MAP) { | |
1385 | memory_bm_set_bit(dst, pfn); | |
1386 | pfn = memory_bm_next_pfn(src); | |
f577eb30 RW |
1387 | } |
1388 | } | |
1389 | ||
940864dd | 1390 | static inline int check_header(struct swsusp_info *info) |
f577eb30 RW |
1391 | { |
1392 | char *reason = NULL; | |
1393 | ||
1394 | if (info->version_code != LINUX_VERSION_CODE) | |
1395 | reason = "kernel version"; | |
1396 | if (info->num_physpages != num_physpages) | |
1397 | reason = "memory size"; | |
96b644bd | 1398 | if (strcmp(info->uts.sysname,init_utsname()->sysname)) |
f577eb30 | 1399 | reason = "system type"; |
96b644bd | 1400 | if (strcmp(info->uts.release,init_utsname()->release)) |
f577eb30 | 1401 | reason = "kernel release"; |
96b644bd | 1402 | if (strcmp(info->uts.version,init_utsname()->version)) |
f577eb30 | 1403 | reason = "version"; |
96b644bd | 1404 | if (strcmp(info->uts.machine,init_utsname()->machine)) |
f577eb30 RW |
1405 | reason = "machine"; |
1406 | if (reason) { | |
1407 | printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason); | |
1408 | return -EPERM; | |
1409 | } | |
1410 | return 0; | |
1411 | } | |
1412 | ||
1413 | /** | |
1414 | * load header - check the image header and copy data from it | |
1415 | */ | |
1416 | ||
940864dd RW |
1417 | static int |
1418 | load_header(struct swsusp_info *info) | |
f577eb30 RW |
1419 | { |
1420 | int error; | |
f577eb30 | 1421 | |
940864dd | 1422 | restore_pblist = NULL; |
f577eb30 RW |
1423 | error = check_header(info); |
1424 | if (!error) { | |
f577eb30 RW |
1425 | nr_copy_pages = info->image_pages; |
1426 | nr_meta_pages = info->pages - info->image_pages - 1; | |
1427 | } | |
1428 | return error; | |
1429 | } | |
1430 | ||
1431 | /** | |
940864dd RW |
1432 | * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set |
1433 | * the corresponding bit in the memory bitmap @bm | |
f577eb30 RW |
1434 | */ |
1435 | ||
940864dd RW |
1436 | static inline void |
1437 | unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm) | |
f577eb30 RW |
1438 | { |
1439 | int j; | |
1440 | ||
940864dd RW |
1441 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
1442 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
1443 | break; | |
1444 | ||
1445 | memory_bm_set_bit(bm, buf[j]); | |
f577eb30 | 1446 | } |
f577eb30 RW |
1447 | } |
1448 | ||
8357376d RW |
1449 | /* List of "safe" pages that may be used to store data loaded from the suspend |
1450 | * image | |
1451 | */ | |
1452 | static struct linked_page *safe_pages_list; | |
1453 | ||
1454 | #ifdef CONFIG_HIGHMEM | |
1455 | /* struct highmem_pbe is used for creating the list of highmem pages that | |
1456 | * should be restored atomically during the resume from disk, because the page | |
1457 | * frames they have occupied before the suspend are in use. | |
1458 | */ | |
1459 | struct highmem_pbe { | |
1460 | struct page *copy_page; /* data is here now */ | |
1461 | struct page *orig_page; /* data was here before the suspend */ | |
1462 | struct highmem_pbe *next; | |
1463 | }; | |
1464 | ||
1465 | /* List of highmem PBEs needed for restoring the highmem pages that were | |
1466 | * allocated before the suspend and included in the suspend image, but have | |
1467 | * also been allocated by the "resume" kernel, so their contents cannot be | |
1468 | * written directly to their "original" page frames. | |
1469 | */ | |
1470 | static struct highmem_pbe *highmem_pblist; | |
1471 | ||
1472 | /** | |
1473 | * count_highmem_image_pages - compute the number of highmem pages in the | |
1474 | * suspend image. The bits in the memory bitmap @bm that correspond to the | |
1475 | * image pages are assumed to be set. | |
1476 | */ | |
1477 | ||
1478 | static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) | |
1479 | { | |
1480 | unsigned long pfn; | |
1481 | unsigned int cnt = 0; | |
1482 | ||
1483 | memory_bm_position_reset(bm); | |
1484 | pfn = memory_bm_next_pfn(bm); | |
1485 | while (pfn != BM_END_OF_MAP) { | |
1486 | if (PageHighMem(pfn_to_page(pfn))) | |
1487 | cnt++; | |
1488 | ||
1489 | pfn = memory_bm_next_pfn(bm); | |
1490 | } | |
1491 | return cnt; | |
1492 | } | |
1493 | ||
1494 | /** | |
1495 | * prepare_highmem_image - try to allocate as many highmem pages as | |
1496 | * there are highmem image pages (@nr_highmem_p points to the variable | |
1497 | * containing the number of highmem image pages). The pages that are | |
1498 | * "safe" (ie. will not be overwritten when the suspend image is | |
1499 | * restored) have the corresponding bits set in @bm (it must be | |
1500 | * unitialized). | |
1501 | * | |
1502 | * NOTE: This function should not be called if there are no highmem | |
1503 | * image pages. | |
1504 | */ | |
1505 | ||
1506 | static unsigned int safe_highmem_pages; | |
1507 | ||
1508 | static struct memory_bitmap *safe_highmem_bm; | |
1509 | ||
1510 | static int | |
1511 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
1512 | { | |
1513 | unsigned int to_alloc; | |
1514 | ||
1515 | if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE)) | |
1516 | return -ENOMEM; | |
1517 | ||
1518 | if (get_highmem_buffer(PG_SAFE)) | |
1519 | return -ENOMEM; | |
1520 | ||
1521 | to_alloc = count_free_highmem_pages(); | |
1522 | if (to_alloc > *nr_highmem_p) | |
1523 | to_alloc = *nr_highmem_p; | |
1524 | else | |
1525 | *nr_highmem_p = to_alloc; | |
1526 | ||
1527 | safe_highmem_pages = 0; | |
1528 | while (to_alloc-- > 0) { | |
1529 | struct page *page; | |
1530 | ||
1531 | page = alloc_page(__GFP_HIGHMEM); | |
7be98234 | 1532 | if (!swsusp_page_is_free(page)) { |
8357376d RW |
1533 | /* The page is "safe", set its bit the bitmap */ |
1534 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1535 | safe_highmem_pages++; | |
1536 | } | |
1537 | /* Mark the page as allocated */ | |
7be98234 RW |
1538 | swsusp_set_page_forbidden(page); |
1539 | swsusp_set_page_free(page); | |
8357376d RW |
1540 | } |
1541 | memory_bm_position_reset(bm); | |
1542 | safe_highmem_bm = bm; | |
1543 | return 0; | |
1544 | } | |
1545 | ||
1546 | /** | |
1547 | * get_highmem_page_buffer - for given highmem image page find the buffer | |
1548 | * that suspend_write_next() should set for its caller to write to. | |
1549 | * | |
1550 | * If the page is to be saved to its "original" page frame or a copy of | |
1551 | * the page is to be made in the highmem, @buffer is returned. Otherwise, | |
1552 | * the copy of the page is to be made in normal memory, so the address of | |
1553 | * the copy is returned. | |
1554 | * | |
1555 | * If @buffer is returned, the caller of suspend_write_next() will write | |
1556 | * the page's contents to @buffer, so they will have to be copied to the | |
1557 | * right location on the next call to suspend_write_next() and it is done | |
1558 | * with the help of copy_last_highmem_page(). For this purpose, if | |
1559 | * @buffer is returned, @last_highmem page is set to the page to which | |
1560 | * the data will have to be copied from @buffer. | |
1561 | */ | |
1562 | ||
1563 | static struct page *last_highmem_page; | |
1564 | ||
1565 | static void * | |
1566 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
1567 | { | |
1568 | struct highmem_pbe *pbe; | |
1569 | void *kaddr; | |
1570 | ||
7be98234 | 1571 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) { |
8357376d RW |
1572 | /* We have allocated the "original" page frame and we can |
1573 | * use it directly to store the loaded page. | |
1574 | */ | |
1575 | last_highmem_page = page; | |
1576 | return buffer; | |
1577 | } | |
1578 | /* The "original" page frame has not been allocated and we have to | |
1579 | * use a "safe" page frame to store the loaded page. | |
1580 | */ | |
1581 | pbe = chain_alloc(ca, sizeof(struct highmem_pbe)); | |
1582 | if (!pbe) { | |
1583 | swsusp_free(); | |
1584 | return NULL; | |
1585 | } | |
1586 | pbe->orig_page = page; | |
1587 | if (safe_highmem_pages > 0) { | |
1588 | struct page *tmp; | |
1589 | ||
1590 | /* Copy of the page will be stored in high memory */ | |
1591 | kaddr = buffer; | |
1592 | tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm)); | |
1593 | safe_highmem_pages--; | |
1594 | last_highmem_page = tmp; | |
1595 | pbe->copy_page = tmp; | |
1596 | } else { | |
1597 | /* Copy of the page will be stored in normal memory */ | |
1598 | kaddr = safe_pages_list; | |
1599 | safe_pages_list = safe_pages_list->next; | |
1600 | pbe->copy_page = virt_to_page(kaddr); | |
1601 | } | |
1602 | pbe->next = highmem_pblist; | |
1603 | highmem_pblist = pbe; | |
1604 | return kaddr; | |
1605 | } | |
1606 | ||
1607 | /** | |
1608 | * copy_last_highmem_page - copy the contents of a highmem image from | |
1609 | * @buffer, where the caller of snapshot_write_next() has place them, | |
1610 | * to the right location represented by @last_highmem_page . | |
1611 | */ | |
1612 | ||
1613 | static void copy_last_highmem_page(void) | |
1614 | { | |
1615 | if (last_highmem_page) { | |
1616 | void *dst; | |
1617 | ||
1618 | dst = kmap_atomic(last_highmem_page, KM_USER0); | |
1619 | memcpy(dst, buffer, PAGE_SIZE); | |
1620 | kunmap_atomic(dst, KM_USER0); | |
1621 | last_highmem_page = NULL; | |
1622 | } | |
1623 | } | |
1624 | ||
1625 | static inline int last_highmem_page_copied(void) | |
1626 | { | |
1627 | return !last_highmem_page; | |
1628 | } | |
1629 | ||
1630 | static inline void free_highmem_data(void) | |
1631 | { | |
1632 | if (safe_highmem_bm) | |
1633 | memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR); | |
1634 | ||
1635 | if (buffer) | |
1636 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
1637 | } | |
1638 | #else | |
1639 | static inline int get_safe_write_buffer(void) { return 0; } | |
1640 | ||
1641 | static unsigned int | |
1642 | count_highmem_image_pages(struct memory_bitmap *bm) { return 0; } | |
1643 | ||
1644 | static inline int | |
1645 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
1646 | { | |
1647 | return 0; | |
1648 | } | |
1649 | ||
1650 | static inline void * | |
1651 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
1652 | { | |
1653 | return NULL; | |
1654 | } | |
1655 | ||
1656 | static inline void copy_last_highmem_page(void) {} | |
1657 | static inline int last_highmem_page_copied(void) { return 1; } | |
1658 | static inline void free_highmem_data(void) {} | |
1659 | #endif /* CONFIG_HIGHMEM */ | |
1660 | ||
f577eb30 | 1661 | /** |
940864dd RW |
1662 | * prepare_image - use the memory bitmap @bm to mark the pages that will |
1663 | * be overwritten in the process of restoring the system memory state | |
1664 | * from the suspend image ("unsafe" pages) and allocate memory for the | |
1665 | * image. | |
968808b8 | 1666 | * |
940864dd RW |
1667 | * The idea is to allocate a new memory bitmap first and then allocate |
1668 | * as many pages as needed for the image data, but not to assign these | |
1669 | * pages to specific tasks initially. Instead, we just mark them as | |
8357376d RW |
1670 | * allocated and create a lists of "safe" pages that will be used |
1671 | * later. On systems with high memory a list of "safe" highmem pages is | |
1672 | * also created. | |
f577eb30 RW |
1673 | */ |
1674 | ||
940864dd RW |
1675 | #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe)) |
1676 | ||
940864dd RW |
1677 | static int |
1678 | prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm) | |
f577eb30 | 1679 | { |
8357376d | 1680 | unsigned int nr_pages, nr_highmem; |
940864dd RW |
1681 | struct linked_page *sp_list, *lp; |
1682 | int error; | |
f577eb30 | 1683 | |
8357376d RW |
1684 | /* If there is no highmem, the buffer will not be necessary */ |
1685 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
1686 | buffer = NULL; | |
1687 | ||
1688 | nr_highmem = count_highmem_image_pages(bm); | |
940864dd RW |
1689 | error = mark_unsafe_pages(bm); |
1690 | if (error) | |
1691 | goto Free; | |
1692 | ||
1693 | error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE); | |
1694 | if (error) | |
1695 | goto Free; | |
1696 | ||
1697 | duplicate_memory_bitmap(new_bm, bm); | |
1698 | memory_bm_free(bm, PG_UNSAFE_KEEP); | |
8357376d RW |
1699 | if (nr_highmem > 0) { |
1700 | error = prepare_highmem_image(bm, &nr_highmem); | |
1701 | if (error) | |
1702 | goto Free; | |
1703 | } | |
940864dd RW |
1704 | /* Reserve some safe pages for potential later use. |
1705 | * | |
1706 | * NOTE: This way we make sure there will be enough safe pages for the | |
1707 | * chain_alloc() in get_buffer(). It is a bit wasteful, but | |
1708 | * nr_copy_pages cannot be greater than 50% of the memory anyway. | |
1709 | */ | |
1710 | sp_list = NULL; | |
1711 | /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */ | |
8357376d | 1712 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
1713 | nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE); |
1714 | while (nr_pages > 0) { | |
8357376d | 1715 | lp = get_image_page(GFP_ATOMIC, PG_SAFE); |
940864dd | 1716 | if (!lp) { |
f577eb30 | 1717 | error = -ENOMEM; |
940864dd RW |
1718 | goto Free; |
1719 | } | |
1720 | lp->next = sp_list; | |
1721 | sp_list = lp; | |
1722 | nr_pages--; | |
f577eb30 | 1723 | } |
940864dd RW |
1724 | /* Preallocate memory for the image */ |
1725 | safe_pages_list = NULL; | |
8357376d | 1726 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
1727 | while (nr_pages > 0) { |
1728 | lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC); | |
1729 | if (!lp) { | |
1730 | error = -ENOMEM; | |
1731 | goto Free; | |
1732 | } | |
7be98234 | 1733 | if (!swsusp_page_is_free(virt_to_page(lp))) { |
940864dd RW |
1734 | /* The page is "safe", add it to the list */ |
1735 | lp->next = safe_pages_list; | |
1736 | safe_pages_list = lp; | |
968808b8 | 1737 | } |
940864dd | 1738 | /* Mark the page as allocated */ |
7be98234 RW |
1739 | swsusp_set_page_forbidden(virt_to_page(lp)); |
1740 | swsusp_set_page_free(virt_to_page(lp)); | |
940864dd | 1741 | nr_pages--; |
968808b8 | 1742 | } |
940864dd RW |
1743 | /* Free the reserved safe pages so that chain_alloc() can use them */ |
1744 | while (sp_list) { | |
1745 | lp = sp_list->next; | |
1746 | free_image_page(sp_list, PG_UNSAFE_CLEAR); | |
1747 | sp_list = lp; | |
f577eb30 | 1748 | } |
940864dd RW |
1749 | return 0; |
1750 | ||
59a49335 | 1751 | Free: |
940864dd | 1752 | swsusp_free(); |
f577eb30 RW |
1753 | return error; |
1754 | } | |
1755 | ||
940864dd RW |
1756 | /** |
1757 | * get_buffer - compute the address that snapshot_write_next() should | |
1758 | * set for its caller to write to. | |
1759 | */ | |
1760 | ||
1761 | static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca) | |
968808b8 | 1762 | { |
940864dd RW |
1763 | struct pbe *pbe; |
1764 | struct page *page = pfn_to_page(memory_bm_next_pfn(bm)); | |
968808b8 | 1765 | |
8357376d RW |
1766 | if (PageHighMem(page)) |
1767 | return get_highmem_page_buffer(page, ca); | |
1768 | ||
7be98234 | 1769 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) |
940864dd RW |
1770 | /* We have allocated the "original" page frame and we can |
1771 | * use it directly to store the loaded page. | |
968808b8 | 1772 | */ |
940864dd RW |
1773 | return page_address(page); |
1774 | ||
1775 | /* The "original" page frame has not been allocated and we have to | |
1776 | * use a "safe" page frame to store the loaded page. | |
968808b8 | 1777 | */ |
940864dd RW |
1778 | pbe = chain_alloc(ca, sizeof(struct pbe)); |
1779 | if (!pbe) { | |
1780 | swsusp_free(); | |
1781 | return NULL; | |
1782 | } | |
8357376d RW |
1783 | pbe->orig_address = page_address(page); |
1784 | pbe->address = safe_pages_list; | |
940864dd RW |
1785 | safe_pages_list = safe_pages_list->next; |
1786 | pbe->next = restore_pblist; | |
1787 | restore_pblist = pbe; | |
8357376d | 1788 | return pbe->address; |
968808b8 RW |
1789 | } |
1790 | ||
f577eb30 RW |
1791 | /** |
1792 | * snapshot_write_next - used for writing the system memory snapshot. | |
1793 | * | |
1794 | * On the first call to it @handle should point to a zeroed | |
1795 | * snapshot_handle structure. The structure gets updated and a pointer | |
1796 | * to it should be passed to this function every next time. | |
1797 | * | |
1798 | * The @count parameter should contain the number of bytes the caller | |
1799 | * wants to write to the image. It must not be zero. | |
1800 | * | |
1801 | * On success the function returns a positive number. Then, the caller | |
1802 | * is allowed to write up to the returned number of bytes to the memory | |
1803 | * location computed by the data_of() macro. The number returned | |
1804 | * may be smaller than @count, but this only happens if the write would | |
1805 | * cross a page boundary otherwise. | |
1806 | * | |
1807 | * The function returns 0 to indicate the "end of file" condition, | |
1808 | * and a negative number is returned on error. In such cases the | |
1809 | * structure pointed to by @handle is not updated and should not be used | |
1810 | * any more. | |
1811 | */ | |
1812 | ||
1813 | int snapshot_write_next(struct snapshot_handle *handle, size_t count) | |
1814 | { | |
940864dd | 1815 | static struct chain_allocator ca; |
f577eb30 RW |
1816 | int error = 0; |
1817 | ||
940864dd | 1818 | /* Check if we have already loaded the entire image */ |
fb13a28b | 1819 | if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 1820 | return 0; |
940864dd | 1821 | |
8357376d RW |
1822 | if (handle->offset == 0) { |
1823 | if (!buffer) | |
1824 | /* This makes the buffer be freed by swsusp_free() */ | |
1825 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); | |
1826 | ||
f577eb30 RW |
1827 | if (!buffer) |
1828 | return -ENOMEM; | |
8357376d | 1829 | |
f577eb30 | 1830 | handle->buffer = buffer; |
8357376d | 1831 | } |
546e0d27 | 1832 | handle->sync_read = 1; |
fb13a28b | 1833 | if (handle->prev < handle->cur) { |
940864dd RW |
1834 | if (handle->prev == 0) { |
1835 | error = load_header(buffer); | |
1836 | if (error) | |
1837 | return error; | |
1838 | ||
1839 | error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY); | |
f577eb30 RW |
1840 | if (error) |
1841 | return error; | |
940864dd | 1842 | |
f577eb30 | 1843 | } else if (handle->prev <= nr_meta_pages) { |
940864dd RW |
1844 | unpack_orig_pfns(buffer, ©_bm); |
1845 | if (handle->prev == nr_meta_pages) { | |
1846 | error = prepare_image(&orig_bm, ©_bm); | |
f577eb30 RW |
1847 | if (error) |
1848 | return error; | |
940864dd RW |
1849 | |
1850 | chain_init(&ca, GFP_ATOMIC, PG_SAFE); | |
1851 | memory_bm_position_reset(&orig_bm); | |
1852 | restore_pblist = NULL; | |
1853 | handle->buffer = get_buffer(&orig_bm, &ca); | |
546e0d27 | 1854 | handle->sync_read = 0; |
940864dd RW |
1855 | if (!handle->buffer) |
1856 | return -ENOMEM; | |
f577eb30 RW |
1857 | } |
1858 | } else { | |
8357376d | 1859 | copy_last_highmem_page(); |
940864dd | 1860 | handle->buffer = get_buffer(&orig_bm, &ca); |
8357376d RW |
1861 | if (handle->buffer != buffer) |
1862 | handle->sync_read = 0; | |
f577eb30 | 1863 | } |
fb13a28b | 1864 | handle->prev = handle->cur; |
f577eb30 | 1865 | } |
fb13a28b RW |
1866 | handle->buf_offset = handle->cur_offset; |
1867 | if (handle->cur_offset + count >= PAGE_SIZE) { | |
1868 | count = PAGE_SIZE - handle->cur_offset; | |
1869 | handle->cur_offset = 0; | |
1870 | handle->cur++; | |
f577eb30 | 1871 | } else { |
fb13a28b | 1872 | handle->cur_offset += count; |
f577eb30 RW |
1873 | } |
1874 | handle->offset += count; | |
1875 | return count; | |
1876 | } | |
1877 | ||
8357376d RW |
1878 | /** |
1879 | * snapshot_write_finalize - must be called after the last call to | |
1880 | * snapshot_write_next() in case the last page in the image happens | |
1881 | * to be a highmem page and its contents should be stored in the | |
1882 | * highmem. Additionally, it releases the memory that will not be | |
1883 | * used any more. | |
1884 | */ | |
1885 | ||
1886 | void snapshot_write_finalize(struct snapshot_handle *handle) | |
1887 | { | |
1888 | copy_last_highmem_page(); | |
1889 | /* Free only if we have loaded the image entirely */ | |
1890 | if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages) { | |
1891 | memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR); | |
1892 | free_highmem_data(); | |
1893 | } | |
1894 | } | |
1895 | ||
f577eb30 RW |
1896 | int snapshot_image_loaded(struct snapshot_handle *handle) |
1897 | { | |
8357376d | 1898 | return !(!nr_copy_pages || !last_highmem_page_copied() || |
940864dd RW |
1899 | handle->cur <= nr_meta_pages + nr_copy_pages); |
1900 | } | |
1901 | ||
8357376d RW |
1902 | #ifdef CONFIG_HIGHMEM |
1903 | /* Assumes that @buf is ready and points to a "safe" page */ | |
1904 | static inline void | |
1905 | swap_two_pages_data(struct page *p1, struct page *p2, void *buf) | |
940864dd | 1906 | { |
8357376d RW |
1907 | void *kaddr1, *kaddr2; |
1908 | ||
1909 | kaddr1 = kmap_atomic(p1, KM_USER0); | |
1910 | kaddr2 = kmap_atomic(p2, KM_USER1); | |
1911 | memcpy(buf, kaddr1, PAGE_SIZE); | |
1912 | memcpy(kaddr1, kaddr2, PAGE_SIZE); | |
1913 | memcpy(kaddr2, buf, PAGE_SIZE); | |
1914 | kunmap_atomic(kaddr1, KM_USER0); | |
1915 | kunmap_atomic(kaddr2, KM_USER1); | |
1916 | } | |
1917 | ||
1918 | /** | |
1919 | * restore_highmem - for each highmem page that was allocated before | |
1920 | * the suspend and included in the suspend image, and also has been | |
1921 | * allocated by the "resume" kernel swap its current (ie. "before | |
1922 | * resume") contents with the previous (ie. "before suspend") one. | |
1923 | * | |
1924 | * If the resume eventually fails, we can call this function once | |
1925 | * again and restore the "before resume" highmem state. | |
1926 | */ | |
1927 | ||
1928 | int restore_highmem(void) | |
1929 | { | |
1930 | struct highmem_pbe *pbe = highmem_pblist; | |
1931 | void *buf; | |
1932 | ||
1933 | if (!pbe) | |
1934 | return 0; | |
1935 | ||
1936 | buf = get_image_page(GFP_ATOMIC, PG_SAFE); | |
1937 | if (!buf) | |
1938 | return -ENOMEM; | |
1939 | ||
1940 | while (pbe) { | |
1941 | swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf); | |
1942 | pbe = pbe->next; | |
1943 | } | |
1944 | free_image_page(buf, PG_UNSAFE_CLEAR); | |
1945 | return 0; | |
f577eb30 | 1946 | } |
8357376d | 1947 | #endif /* CONFIG_HIGHMEM */ |