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> | |
846705de | 28 | #include <linux/list.h> |
25761b6e RW |
29 | |
30 | #include <asm/uaccess.h> | |
31 | #include <asm/mmu_context.h> | |
32 | #include <asm/pgtable.h> | |
33 | #include <asm/tlbflush.h> | |
34 | #include <asm/io.h> | |
35 | ||
25761b6e RW |
36 | #include "power.h" |
37 | ||
74dfd666 RW |
38 | static int swsusp_page_is_free(struct page *); |
39 | static void swsusp_set_page_forbidden(struct page *); | |
40 | static void swsusp_unset_page_forbidden(struct page *); | |
41 | ||
fe419535 RW |
42 | /* |
43 | * Preferred image size in bytes (tunable via /sys/power/image_size). | |
44 | * When it is set to N, swsusp will do its best to ensure the image | |
45 | * size will not exceed N bytes, but if that is impossible, it will | |
46 | * try to create the smallest image possible. | |
47 | */ | |
48 | unsigned long image_size = 500 * 1024 * 1024; | |
49 | ||
8357376d RW |
50 | /* List of PBEs needed for restoring the pages that were allocated before |
51 | * the suspend and included in the suspend image, but have also been | |
52 | * allocated by the "resume" kernel, so their contents cannot be written | |
53 | * directly to their "original" page frames. | |
54 | */ | |
75534b50 RW |
55 | struct pbe *restore_pblist; |
56 | ||
8357376d | 57 | /* Pointer to an auxiliary buffer (1 page) */ |
940864dd | 58 | static void *buffer; |
7088a5c0 | 59 | |
f6143aa6 RW |
60 | /** |
61 | * @safe_needed - on resume, for storing the PBE list and the image, | |
62 | * we can only use memory pages that do not conflict with the pages | |
8357376d RW |
63 | * used before suspend. The unsafe pages have PageNosaveFree set |
64 | * and we count them using unsafe_pages. | |
f6143aa6 | 65 | * |
8357376d RW |
66 | * Each allocated image page is marked as PageNosave and PageNosaveFree |
67 | * so that swsusp_free() can release it. | |
f6143aa6 RW |
68 | */ |
69 | ||
0bcd888d RW |
70 | #define PG_ANY 0 |
71 | #define PG_SAFE 1 | |
72 | #define PG_UNSAFE_CLEAR 1 | |
73 | #define PG_UNSAFE_KEEP 0 | |
74 | ||
940864dd | 75 | static unsigned int allocated_unsafe_pages; |
f6143aa6 | 76 | |
8357376d | 77 | static void *get_image_page(gfp_t gfp_mask, int safe_needed) |
f6143aa6 RW |
78 | { |
79 | void *res; | |
80 | ||
81 | res = (void *)get_zeroed_page(gfp_mask); | |
82 | if (safe_needed) | |
7be98234 | 83 | while (res && swsusp_page_is_free(virt_to_page(res))) { |
f6143aa6 | 84 | /* The page is unsafe, mark it for swsusp_free() */ |
7be98234 | 85 | swsusp_set_page_forbidden(virt_to_page(res)); |
940864dd | 86 | allocated_unsafe_pages++; |
f6143aa6 RW |
87 | res = (void *)get_zeroed_page(gfp_mask); |
88 | } | |
89 | if (res) { | |
7be98234 RW |
90 | swsusp_set_page_forbidden(virt_to_page(res)); |
91 | swsusp_set_page_free(virt_to_page(res)); | |
f6143aa6 RW |
92 | } |
93 | return res; | |
94 | } | |
95 | ||
96 | unsigned long get_safe_page(gfp_t gfp_mask) | |
97 | { | |
8357376d RW |
98 | return (unsigned long)get_image_page(gfp_mask, PG_SAFE); |
99 | } | |
100 | ||
5b6d15de RW |
101 | static struct page *alloc_image_page(gfp_t gfp_mask) |
102 | { | |
8357376d RW |
103 | struct page *page; |
104 | ||
105 | page = alloc_page(gfp_mask); | |
106 | if (page) { | |
7be98234 RW |
107 | swsusp_set_page_forbidden(page); |
108 | swsusp_set_page_free(page); | |
8357376d RW |
109 | } |
110 | return page; | |
f6143aa6 RW |
111 | } |
112 | ||
113 | /** | |
114 | * free_image_page - free page represented by @addr, allocated with | |
8357376d | 115 | * get_image_page (page flags set by it must be cleared) |
f6143aa6 RW |
116 | */ |
117 | ||
118 | static inline void free_image_page(void *addr, int clear_nosave_free) | |
119 | { | |
8357376d RW |
120 | struct page *page; |
121 | ||
122 | BUG_ON(!virt_addr_valid(addr)); | |
123 | ||
124 | page = virt_to_page(addr); | |
125 | ||
7be98234 | 126 | swsusp_unset_page_forbidden(page); |
f6143aa6 | 127 | if (clear_nosave_free) |
7be98234 | 128 | swsusp_unset_page_free(page); |
8357376d RW |
129 | |
130 | __free_page(page); | |
f6143aa6 RW |
131 | } |
132 | ||
b788db79 RW |
133 | /* struct linked_page is used to build chains of pages */ |
134 | ||
135 | #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *)) | |
136 | ||
137 | struct linked_page { | |
138 | struct linked_page *next; | |
139 | char data[LINKED_PAGE_DATA_SIZE]; | |
140 | } __attribute__((packed)); | |
141 | ||
142 | static inline void | |
143 | free_list_of_pages(struct linked_page *list, int clear_page_nosave) | |
144 | { | |
145 | while (list) { | |
146 | struct linked_page *lp = list->next; | |
147 | ||
148 | free_image_page(list, clear_page_nosave); | |
149 | list = lp; | |
150 | } | |
151 | } | |
152 | ||
153 | /** | |
154 | * struct chain_allocator is used for allocating small objects out of | |
155 | * a linked list of pages called 'the chain'. | |
156 | * | |
157 | * The chain grows each time when there is no room for a new object in | |
158 | * the current page. The allocated objects cannot be freed individually. | |
159 | * It is only possible to free them all at once, by freeing the entire | |
160 | * chain. | |
161 | * | |
162 | * NOTE: The chain allocator may be inefficient if the allocated objects | |
163 | * are not much smaller than PAGE_SIZE. | |
164 | */ | |
165 | ||
166 | struct chain_allocator { | |
167 | struct linked_page *chain; /* the chain */ | |
168 | unsigned int used_space; /* total size of objects allocated out | |
169 | * of the current page | |
170 | */ | |
171 | gfp_t gfp_mask; /* mask for allocating pages */ | |
172 | int safe_needed; /* if set, only "safe" pages are allocated */ | |
173 | }; | |
174 | ||
175 | static void | |
176 | chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed) | |
177 | { | |
178 | ca->chain = NULL; | |
179 | ca->used_space = LINKED_PAGE_DATA_SIZE; | |
180 | ca->gfp_mask = gfp_mask; | |
181 | ca->safe_needed = safe_needed; | |
182 | } | |
183 | ||
184 | static void *chain_alloc(struct chain_allocator *ca, unsigned int size) | |
185 | { | |
186 | void *ret; | |
187 | ||
188 | if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) { | |
189 | struct linked_page *lp; | |
190 | ||
8357376d | 191 | lp = get_image_page(ca->gfp_mask, ca->safe_needed); |
b788db79 RW |
192 | if (!lp) |
193 | return NULL; | |
194 | ||
195 | lp->next = ca->chain; | |
196 | ca->chain = lp; | |
197 | ca->used_space = 0; | |
198 | } | |
199 | ret = ca->chain->data + ca->used_space; | |
200 | ca->used_space += size; | |
201 | return ret; | |
202 | } | |
203 | ||
b788db79 RW |
204 | /** |
205 | * Data types related to memory bitmaps. | |
206 | * | |
207 | * Memory bitmap is a structure consiting of many linked lists of | |
208 | * objects. The main list's elements are of type struct zone_bitmap | |
209 | * and each of them corresonds to one zone. For each zone bitmap | |
210 | * object there is a list of objects of type struct bm_block that | |
0d83304c | 211 | * represent each blocks of bitmap in which information is stored. |
b788db79 RW |
212 | * |
213 | * struct memory_bitmap contains a pointer to the main list of zone | |
214 | * bitmap objects, a struct bm_position used for browsing the bitmap, | |
215 | * and a pointer to the list of pages used for allocating all of the | |
216 | * zone bitmap objects and bitmap block objects. | |
217 | * | |
218 | * NOTE: It has to be possible to lay out the bitmap in memory | |
219 | * using only allocations of order 0. Additionally, the bitmap is | |
220 | * designed to work with arbitrary number of zones (this is over the | |
221 | * top for now, but let's avoid making unnecessary assumptions ;-). | |
222 | * | |
223 | * struct zone_bitmap contains a pointer to a list of bitmap block | |
224 | * objects and a pointer to the bitmap block object that has been | |
225 | * most recently used for setting bits. Additionally, it contains the | |
226 | * pfns that correspond to the start and end of the represented zone. | |
227 | * | |
228 | * struct bm_block contains a pointer to the memory page in which | |
0d83304c AM |
229 | * information is stored (in the form of a block of bitmap) |
230 | * It also contains the pfns that correspond to the start and end of | |
231 | * the represented memory area. | |
b788db79 RW |
232 | */ |
233 | ||
234 | #define BM_END_OF_MAP (~0UL) | |
235 | ||
b788db79 RW |
236 | #define BM_BITS_PER_BLOCK (PAGE_SIZE << 3) |
237 | ||
238 | struct bm_block { | |
846705de | 239 | struct list_head hook; /* hook into a list of bitmap blocks */ |
b788db79 RW |
240 | unsigned long start_pfn; /* pfn represented by the first bit */ |
241 | unsigned long end_pfn; /* pfn represented by the last bit plus 1 */ | |
0d83304c | 242 | unsigned long *data; /* bitmap representing pages */ |
b788db79 RW |
243 | }; |
244 | ||
0d83304c AM |
245 | static inline unsigned long bm_block_bits(struct bm_block *bb) |
246 | { | |
247 | return bb->end_pfn - bb->start_pfn; | |
248 | } | |
249 | ||
b788db79 RW |
250 | /* strcut bm_position is used for browsing memory bitmaps */ |
251 | ||
252 | struct bm_position { | |
b788db79 | 253 | struct bm_block *block; |
b788db79 RW |
254 | int bit; |
255 | }; | |
256 | ||
257 | struct memory_bitmap { | |
846705de | 258 | struct list_head blocks; /* list of bitmap blocks */ |
b788db79 RW |
259 | struct linked_page *p_list; /* list of pages used to store zone |
260 | * bitmap objects and bitmap block | |
261 | * objects | |
262 | */ | |
263 | struct bm_position cur; /* most recently used bit position */ | |
264 | }; | |
265 | ||
266 | /* Functions that operate on memory bitmaps */ | |
267 | ||
b788db79 RW |
268 | static void memory_bm_position_reset(struct memory_bitmap *bm) |
269 | { | |
846705de | 270 | bm->cur.block = list_entry(bm->blocks.next, struct bm_block, hook); |
0d83304c | 271 | bm->cur.bit = 0; |
b788db79 RW |
272 | } |
273 | ||
274 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free); | |
275 | ||
276 | /** | |
277 | * create_bm_block_list - create a list of block bitmap objects | |
846705de RW |
278 | * @nr_blocks - number of blocks to allocate |
279 | * @list - list to put the allocated blocks into | |
280 | * @ca - chain allocator to be used for allocating memory | |
b788db79 | 281 | */ |
846705de RW |
282 | static int create_bm_block_list(unsigned long pages, |
283 | struct list_head *list, | |
284 | struct chain_allocator *ca) | |
b788db79 | 285 | { |
846705de | 286 | unsigned int nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK); |
b788db79 RW |
287 | |
288 | while (nr_blocks-- > 0) { | |
289 | struct bm_block *bb; | |
290 | ||
291 | bb = chain_alloc(ca, sizeof(struct bm_block)); | |
292 | if (!bb) | |
846705de RW |
293 | return -ENOMEM; |
294 | list_add(&bb->hook, list); | |
b788db79 | 295 | } |
846705de RW |
296 | |
297 | return 0; | |
b788db79 RW |
298 | } |
299 | ||
846705de RW |
300 | struct mem_extent { |
301 | struct list_head hook; | |
302 | unsigned long start; | |
303 | unsigned long end; | |
304 | }; | |
305 | ||
b788db79 | 306 | /** |
846705de RW |
307 | * free_mem_extents - free a list of memory extents |
308 | * @list - list of extents to empty | |
b788db79 | 309 | */ |
846705de RW |
310 | static void free_mem_extents(struct list_head *list) |
311 | { | |
312 | struct mem_extent *ext, *aux; | |
b788db79 | 313 | |
846705de RW |
314 | list_for_each_entry_safe(ext, aux, list, hook) { |
315 | list_del(&ext->hook); | |
316 | kfree(ext); | |
317 | } | |
318 | } | |
319 | ||
320 | /** | |
321 | * create_mem_extents - create a list of memory extents representing | |
322 | * contiguous ranges of PFNs | |
323 | * @list - list to put the extents into | |
324 | * @gfp_mask - mask to use for memory allocations | |
325 | */ | |
326 | static int create_mem_extents(struct list_head *list, gfp_t gfp_mask) | |
b788db79 | 327 | { |
846705de | 328 | struct zone *zone; |
b788db79 | 329 | |
846705de | 330 | INIT_LIST_HEAD(list); |
b788db79 | 331 | |
ee99c71c | 332 | for_each_populated_zone(zone) { |
846705de RW |
333 | unsigned long zone_start, zone_end; |
334 | struct mem_extent *ext, *cur, *aux; | |
335 | ||
846705de RW |
336 | zone_start = zone->zone_start_pfn; |
337 | zone_end = zone->zone_start_pfn + zone->spanned_pages; | |
338 | ||
339 | list_for_each_entry(ext, list, hook) | |
340 | if (zone_start <= ext->end) | |
341 | break; | |
b788db79 | 342 | |
846705de RW |
343 | if (&ext->hook == list || zone_end < ext->start) { |
344 | /* New extent is necessary */ | |
345 | struct mem_extent *new_ext; | |
346 | ||
347 | new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask); | |
348 | if (!new_ext) { | |
349 | free_mem_extents(list); | |
350 | return -ENOMEM; | |
351 | } | |
352 | new_ext->start = zone_start; | |
353 | new_ext->end = zone_end; | |
354 | list_add_tail(&new_ext->hook, &ext->hook); | |
355 | continue; | |
356 | } | |
357 | ||
358 | /* Merge this zone's range of PFNs with the existing one */ | |
359 | if (zone_start < ext->start) | |
360 | ext->start = zone_start; | |
361 | if (zone_end > ext->end) | |
362 | ext->end = zone_end; | |
363 | ||
364 | /* More merging may be possible */ | |
365 | cur = ext; | |
366 | list_for_each_entry_safe_continue(cur, aux, list, hook) { | |
367 | if (zone_end < cur->start) | |
368 | break; | |
369 | if (zone_end < cur->end) | |
370 | ext->end = cur->end; | |
371 | list_del(&cur->hook); | |
372 | kfree(cur); | |
373 | } | |
b788db79 | 374 | } |
846705de RW |
375 | |
376 | return 0; | |
b788db79 RW |
377 | } |
378 | ||
379 | /** | |
380 | * memory_bm_create - allocate memory for a memory bitmap | |
381 | */ | |
b788db79 RW |
382 | static int |
383 | memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed) | |
384 | { | |
385 | struct chain_allocator ca; | |
846705de RW |
386 | struct list_head mem_extents; |
387 | struct mem_extent *ext; | |
388 | int error; | |
b788db79 RW |
389 | |
390 | chain_init(&ca, gfp_mask, safe_needed); | |
846705de | 391 | INIT_LIST_HEAD(&bm->blocks); |
b788db79 | 392 | |
846705de RW |
393 | error = create_mem_extents(&mem_extents, gfp_mask); |
394 | if (error) | |
395 | return error; | |
b788db79 | 396 | |
846705de RW |
397 | list_for_each_entry(ext, &mem_extents, hook) { |
398 | struct bm_block *bb; | |
399 | unsigned long pfn = ext->start; | |
400 | unsigned long pages = ext->end - ext->start; | |
b788db79 | 401 | |
846705de | 402 | bb = list_entry(bm->blocks.prev, struct bm_block, hook); |
b788db79 | 403 | |
846705de RW |
404 | error = create_bm_block_list(pages, bm->blocks.prev, &ca); |
405 | if (error) | |
406 | goto Error; | |
b788db79 | 407 | |
846705de RW |
408 | list_for_each_entry_continue(bb, &bm->blocks, hook) { |
409 | bb->data = get_image_page(gfp_mask, safe_needed); | |
410 | if (!bb->data) { | |
411 | error = -ENOMEM; | |
412 | goto Error; | |
413 | } | |
b788db79 RW |
414 | |
415 | bb->start_pfn = pfn; | |
846705de | 416 | if (pages >= BM_BITS_PER_BLOCK) { |
b788db79 | 417 | pfn += BM_BITS_PER_BLOCK; |
846705de | 418 | pages -= BM_BITS_PER_BLOCK; |
b788db79 RW |
419 | } else { |
420 | /* This is executed only once in the loop */ | |
846705de | 421 | pfn += pages; |
b788db79 RW |
422 | } |
423 | bb->end_pfn = pfn; | |
b788db79 | 424 | } |
b788db79 | 425 | } |
846705de | 426 | |
b788db79 RW |
427 | bm->p_list = ca.chain; |
428 | memory_bm_position_reset(bm); | |
846705de RW |
429 | Exit: |
430 | free_mem_extents(&mem_extents); | |
431 | return error; | |
b788db79 | 432 | |
846705de | 433 | Error: |
b788db79 RW |
434 | bm->p_list = ca.chain; |
435 | memory_bm_free(bm, PG_UNSAFE_CLEAR); | |
846705de | 436 | goto Exit; |
b788db79 RW |
437 | } |
438 | ||
439 | /** | |
440 | * memory_bm_free - free memory occupied by the memory bitmap @bm | |
441 | */ | |
b788db79 RW |
442 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free) |
443 | { | |
846705de | 444 | struct bm_block *bb; |
b788db79 | 445 | |
846705de RW |
446 | list_for_each_entry(bb, &bm->blocks, hook) |
447 | if (bb->data) | |
448 | free_image_page(bb->data, clear_nosave_free); | |
b788db79 | 449 | |
b788db79 | 450 | free_list_of_pages(bm->p_list, clear_nosave_free); |
846705de RW |
451 | |
452 | INIT_LIST_HEAD(&bm->blocks); | |
b788db79 RW |
453 | } |
454 | ||
455 | /** | |
74dfd666 | 456 | * memory_bm_find_bit - find the bit in the bitmap @bm that corresponds |
b788db79 RW |
457 | * to given pfn. The cur_zone_bm member of @bm and the cur_block member |
458 | * of @bm->cur_zone_bm are updated. | |
b788db79 | 459 | */ |
a82f7119 | 460 | static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn, |
74dfd666 | 461 | void **addr, unsigned int *bit_nr) |
b788db79 | 462 | { |
b788db79 RW |
463 | struct bm_block *bb; |
464 | ||
846705de RW |
465 | /* |
466 | * Check if the pfn corresponds to the current bitmap block and find | |
467 | * the block where it fits if this is not the case. | |
468 | */ | |
469 | bb = bm->cur.block; | |
b788db79 | 470 | if (pfn < bb->start_pfn) |
846705de RW |
471 | list_for_each_entry_continue_reverse(bb, &bm->blocks, hook) |
472 | if (pfn >= bb->start_pfn) | |
473 | break; | |
b788db79 | 474 | |
846705de RW |
475 | if (pfn >= bb->end_pfn) |
476 | list_for_each_entry_continue(bb, &bm->blocks, hook) | |
477 | if (pfn >= bb->start_pfn && pfn < bb->end_pfn) | |
478 | break; | |
74dfd666 | 479 | |
846705de RW |
480 | if (&bb->hook == &bm->blocks) |
481 | return -EFAULT; | |
482 | ||
483 | /* The block has been found */ | |
484 | bm->cur.block = bb; | |
b788db79 | 485 | pfn -= bb->start_pfn; |
846705de | 486 | bm->cur.bit = pfn + 1; |
0d83304c AM |
487 | *bit_nr = pfn; |
488 | *addr = bb->data; | |
a82f7119 | 489 | return 0; |
74dfd666 RW |
490 | } |
491 | ||
492 | static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn) | |
493 | { | |
494 | void *addr; | |
495 | unsigned int bit; | |
a82f7119 | 496 | int error; |
74dfd666 | 497 | |
a82f7119 RW |
498 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
499 | BUG_ON(error); | |
74dfd666 RW |
500 | set_bit(bit, addr); |
501 | } | |
502 | ||
a82f7119 RW |
503 | static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn) |
504 | { | |
505 | void *addr; | |
506 | unsigned int bit; | |
507 | int error; | |
508 | ||
509 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); | |
510 | if (!error) | |
511 | set_bit(bit, addr); | |
512 | return error; | |
513 | } | |
514 | ||
74dfd666 RW |
515 | static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn) |
516 | { | |
517 | void *addr; | |
518 | unsigned int bit; | |
a82f7119 | 519 | int error; |
74dfd666 | 520 | |
a82f7119 RW |
521 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
522 | BUG_ON(error); | |
74dfd666 RW |
523 | clear_bit(bit, addr); |
524 | } | |
525 | ||
526 | static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn) | |
527 | { | |
528 | void *addr; | |
529 | unsigned int bit; | |
a82f7119 | 530 | int error; |
74dfd666 | 531 | |
a82f7119 RW |
532 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
533 | BUG_ON(error); | |
74dfd666 | 534 | return test_bit(bit, addr); |
b788db79 RW |
535 | } |
536 | ||
69643279 RW |
537 | static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn) |
538 | { | |
539 | void *addr; | |
540 | unsigned int bit; | |
541 | ||
542 | return !memory_bm_find_bit(bm, pfn, &addr, &bit); | |
543 | } | |
544 | ||
b788db79 RW |
545 | /** |
546 | * memory_bm_next_pfn - find the pfn that corresponds to the next set bit | |
547 | * in the bitmap @bm. If the pfn cannot be found, BM_END_OF_MAP is | |
548 | * returned. | |
549 | * | |
550 | * It is required to run memory_bm_position_reset() before the first call to | |
551 | * this function. | |
552 | */ | |
553 | ||
554 | static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm) | |
555 | { | |
b788db79 | 556 | struct bm_block *bb; |
b788db79 RW |
557 | int bit; |
558 | ||
846705de | 559 | bb = bm->cur.block; |
b788db79 | 560 | do { |
846705de RW |
561 | bit = bm->cur.bit; |
562 | bit = find_next_bit(bb->data, bm_block_bits(bb), bit); | |
563 | if (bit < bm_block_bits(bb)) | |
564 | goto Return_pfn; | |
565 | ||
566 | bb = list_entry(bb->hook.next, struct bm_block, hook); | |
567 | bm->cur.block = bb; | |
568 | bm->cur.bit = 0; | |
569 | } while (&bb->hook != &bm->blocks); | |
570 | ||
b788db79 RW |
571 | memory_bm_position_reset(bm); |
572 | return BM_END_OF_MAP; | |
573 | ||
59a49335 | 574 | Return_pfn: |
0d83304c AM |
575 | bm->cur.bit = bit + 1; |
576 | return bb->start_pfn + bit; | |
b788db79 RW |
577 | } |
578 | ||
74dfd666 RW |
579 | /** |
580 | * This structure represents a range of page frames the contents of which | |
581 | * should not be saved during the suspend. | |
582 | */ | |
583 | ||
584 | struct nosave_region { | |
585 | struct list_head list; | |
586 | unsigned long start_pfn; | |
587 | unsigned long end_pfn; | |
588 | }; | |
589 | ||
590 | static LIST_HEAD(nosave_regions); | |
591 | ||
592 | /** | |
593 | * register_nosave_region - register a range of page frames the contents | |
594 | * of which should not be saved during the suspend (to be used in the early | |
595 | * initialization code) | |
596 | */ | |
597 | ||
598 | void __init | |
940d67f6 JB |
599 | __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn, |
600 | int use_kmalloc) | |
74dfd666 RW |
601 | { |
602 | struct nosave_region *region; | |
603 | ||
604 | if (start_pfn >= end_pfn) | |
605 | return; | |
606 | ||
607 | if (!list_empty(&nosave_regions)) { | |
608 | /* Try to extend the previous region (they should be sorted) */ | |
609 | region = list_entry(nosave_regions.prev, | |
610 | struct nosave_region, list); | |
611 | if (region->end_pfn == start_pfn) { | |
612 | region->end_pfn = end_pfn; | |
613 | goto Report; | |
614 | } | |
615 | } | |
940d67f6 JB |
616 | if (use_kmalloc) { |
617 | /* during init, this shouldn't fail */ | |
618 | region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL); | |
619 | BUG_ON(!region); | |
620 | } else | |
621 | /* This allocation cannot fail */ | |
622 | region = alloc_bootmem_low(sizeof(struct nosave_region)); | |
74dfd666 RW |
623 | region->start_pfn = start_pfn; |
624 | region->end_pfn = end_pfn; | |
625 | list_add_tail(®ion->list, &nosave_regions); | |
626 | Report: | |
23976728 | 627 | printk(KERN_INFO "PM: Registered nosave memory: %016lx - %016lx\n", |
74dfd666 RW |
628 | start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT); |
629 | } | |
630 | ||
631 | /* | |
632 | * Set bits in this map correspond to the page frames the contents of which | |
633 | * should not be saved during the suspend. | |
634 | */ | |
635 | static struct memory_bitmap *forbidden_pages_map; | |
636 | ||
637 | /* Set bits in this map correspond to free page frames. */ | |
638 | static struct memory_bitmap *free_pages_map; | |
639 | ||
640 | /* | |
641 | * Each page frame allocated for creating the image is marked by setting the | |
642 | * corresponding bits in forbidden_pages_map and free_pages_map simultaneously | |
643 | */ | |
644 | ||
645 | void swsusp_set_page_free(struct page *page) | |
646 | { | |
647 | if (free_pages_map) | |
648 | memory_bm_set_bit(free_pages_map, page_to_pfn(page)); | |
649 | } | |
650 | ||
651 | static int swsusp_page_is_free(struct page *page) | |
652 | { | |
653 | return free_pages_map ? | |
654 | memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0; | |
655 | } | |
656 | ||
657 | void swsusp_unset_page_free(struct page *page) | |
658 | { | |
659 | if (free_pages_map) | |
660 | memory_bm_clear_bit(free_pages_map, page_to_pfn(page)); | |
661 | } | |
662 | ||
663 | static void swsusp_set_page_forbidden(struct page *page) | |
664 | { | |
665 | if (forbidden_pages_map) | |
666 | memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page)); | |
667 | } | |
668 | ||
669 | int swsusp_page_is_forbidden(struct page *page) | |
670 | { | |
671 | return forbidden_pages_map ? | |
672 | memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0; | |
673 | } | |
674 | ||
675 | static void swsusp_unset_page_forbidden(struct page *page) | |
676 | { | |
677 | if (forbidden_pages_map) | |
678 | memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page)); | |
679 | } | |
680 | ||
681 | /** | |
682 | * mark_nosave_pages - set bits corresponding to the page frames the | |
683 | * contents of which should not be saved in a given bitmap. | |
684 | */ | |
685 | ||
686 | static void mark_nosave_pages(struct memory_bitmap *bm) | |
687 | { | |
688 | struct nosave_region *region; | |
689 | ||
690 | if (list_empty(&nosave_regions)) | |
691 | return; | |
692 | ||
693 | list_for_each_entry(region, &nosave_regions, list) { | |
694 | unsigned long pfn; | |
695 | ||
23976728 | 696 | pr_debug("PM: Marking nosave pages: %016lx - %016lx\n", |
74dfd666 RW |
697 | region->start_pfn << PAGE_SHIFT, |
698 | region->end_pfn << PAGE_SHIFT); | |
699 | ||
700 | for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++) | |
a82f7119 RW |
701 | if (pfn_valid(pfn)) { |
702 | /* | |
703 | * It is safe to ignore the result of | |
704 | * mem_bm_set_bit_check() here, since we won't | |
705 | * touch the PFNs for which the error is | |
706 | * returned anyway. | |
707 | */ | |
708 | mem_bm_set_bit_check(bm, pfn); | |
709 | } | |
74dfd666 RW |
710 | } |
711 | } | |
712 | ||
713 | /** | |
714 | * create_basic_memory_bitmaps - create bitmaps needed for marking page | |
715 | * frames that should not be saved and free page frames. The pointers | |
716 | * forbidden_pages_map and free_pages_map are only modified if everything | |
717 | * goes well, because we don't want the bits to be used before both bitmaps | |
718 | * are set up. | |
719 | */ | |
720 | ||
721 | int create_basic_memory_bitmaps(void) | |
722 | { | |
723 | struct memory_bitmap *bm1, *bm2; | |
724 | int error = 0; | |
725 | ||
726 | BUG_ON(forbidden_pages_map || free_pages_map); | |
727 | ||
0709db60 | 728 | bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
729 | if (!bm1) |
730 | return -ENOMEM; | |
731 | ||
0709db60 | 732 | error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
733 | if (error) |
734 | goto Free_first_object; | |
735 | ||
0709db60 | 736 | bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
737 | if (!bm2) |
738 | goto Free_first_bitmap; | |
739 | ||
0709db60 | 740 | error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
741 | if (error) |
742 | goto Free_second_object; | |
743 | ||
744 | forbidden_pages_map = bm1; | |
745 | free_pages_map = bm2; | |
746 | mark_nosave_pages(forbidden_pages_map); | |
747 | ||
23976728 | 748 | pr_debug("PM: Basic memory bitmaps created\n"); |
74dfd666 RW |
749 | |
750 | return 0; | |
751 | ||
752 | Free_second_object: | |
753 | kfree(bm2); | |
754 | Free_first_bitmap: | |
755 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
756 | Free_first_object: | |
757 | kfree(bm1); | |
758 | return -ENOMEM; | |
759 | } | |
760 | ||
761 | /** | |
762 | * free_basic_memory_bitmaps - free memory bitmaps allocated by | |
763 | * create_basic_memory_bitmaps(). The auxiliary pointers are necessary | |
764 | * so that the bitmaps themselves are not referred to while they are being | |
765 | * freed. | |
766 | */ | |
767 | ||
768 | void free_basic_memory_bitmaps(void) | |
769 | { | |
770 | struct memory_bitmap *bm1, *bm2; | |
771 | ||
772 | BUG_ON(!(forbidden_pages_map && free_pages_map)); | |
773 | ||
774 | bm1 = forbidden_pages_map; | |
775 | bm2 = free_pages_map; | |
776 | forbidden_pages_map = NULL; | |
777 | free_pages_map = NULL; | |
778 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
779 | kfree(bm1); | |
780 | memory_bm_free(bm2, PG_UNSAFE_CLEAR); | |
781 | kfree(bm2); | |
782 | ||
23976728 | 783 | pr_debug("PM: Basic memory bitmaps freed\n"); |
74dfd666 RW |
784 | } |
785 | ||
b788db79 RW |
786 | /** |
787 | * snapshot_additional_pages - estimate the number of additional pages | |
788 | * be needed for setting up the suspend image data structures for given | |
789 | * zone (usually the returned value is greater than the exact number) | |
790 | */ | |
791 | ||
792 | unsigned int snapshot_additional_pages(struct zone *zone) | |
793 | { | |
794 | unsigned int res; | |
795 | ||
796 | res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK); | |
797 | res += DIV_ROUND_UP(res * sizeof(struct bm_block), PAGE_SIZE); | |
8357376d | 798 | return 2 * res; |
b788db79 RW |
799 | } |
800 | ||
8357376d RW |
801 | #ifdef CONFIG_HIGHMEM |
802 | /** | |
803 | * count_free_highmem_pages - compute the total number of free highmem | |
804 | * pages, system-wide. | |
805 | */ | |
806 | ||
807 | static unsigned int count_free_highmem_pages(void) | |
808 | { | |
809 | struct zone *zone; | |
810 | unsigned int cnt = 0; | |
811 | ||
ee99c71c KM |
812 | for_each_populated_zone(zone) |
813 | if (is_highmem(zone)) | |
d23ad423 | 814 | cnt += zone_page_state(zone, NR_FREE_PAGES); |
8357376d RW |
815 | |
816 | return cnt; | |
817 | } | |
818 | ||
819 | /** | |
820 | * saveable_highmem_page - Determine whether a highmem page should be | |
821 | * included in the suspend image. | |
822 | * | |
823 | * We should save the page if it isn't Nosave or NosaveFree, or Reserved, | |
824 | * and it isn't a part of a free chunk of pages. | |
825 | */ | |
846705de | 826 | static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn) |
8357376d RW |
827 | { |
828 | struct page *page; | |
829 | ||
830 | if (!pfn_valid(pfn)) | |
831 | return NULL; | |
832 | ||
833 | page = pfn_to_page(pfn); | |
846705de RW |
834 | if (page_zone(page) != zone) |
835 | return NULL; | |
8357376d RW |
836 | |
837 | BUG_ON(!PageHighMem(page)); | |
838 | ||
7be98234 RW |
839 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) || |
840 | PageReserved(page)) | |
8357376d RW |
841 | return NULL; |
842 | ||
843 | return page; | |
844 | } | |
845 | ||
846 | /** | |
847 | * count_highmem_pages - compute the total number of saveable highmem | |
848 | * pages. | |
849 | */ | |
850 | ||
fe419535 | 851 | static unsigned int count_highmem_pages(void) |
8357376d RW |
852 | { |
853 | struct zone *zone; | |
854 | unsigned int n = 0; | |
855 | ||
856 | for_each_zone(zone) { | |
857 | unsigned long pfn, max_zone_pfn; | |
858 | ||
859 | if (!is_highmem(zone)) | |
860 | continue; | |
861 | ||
862 | mark_free_pages(zone); | |
863 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
864 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
846705de | 865 | if (saveable_highmem_page(zone, pfn)) |
8357376d RW |
866 | n++; |
867 | } | |
868 | return n; | |
869 | } | |
870 | #else | |
846705de RW |
871 | static inline void *saveable_highmem_page(struct zone *z, unsigned long p) |
872 | { | |
873 | return NULL; | |
874 | } | |
8357376d RW |
875 | #endif /* CONFIG_HIGHMEM */ |
876 | ||
25761b6e | 877 | /** |
8a235efa RW |
878 | * saveable_page - Determine whether a non-highmem page should be included |
879 | * in the suspend image. | |
25761b6e | 880 | * |
8357376d RW |
881 | * We should save the page if it isn't Nosave, and is not in the range |
882 | * of pages statically defined as 'unsaveable', and it isn't a part of | |
883 | * a free chunk of pages. | |
25761b6e | 884 | */ |
846705de | 885 | static struct page *saveable_page(struct zone *zone, unsigned long pfn) |
25761b6e | 886 | { |
de491861 | 887 | struct page *page; |
25761b6e RW |
888 | |
889 | if (!pfn_valid(pfn)) | |
ae83c5ee | 890 | return NULL; |
25761b6e RW |
891 | |
892 | page = pfn_to_page(pfn); | |
846705de RW |
893 | if (page_zone(page) != zone) |
894 | return NULL; | |
ae83c5ee | 895 | |
8357376d RW |
896 | BUG_ON(PageHighMem(page)); |
897 | ||
7be98234 | 898 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page)) |
ae83c5ee | 899 | return NULL; |
8357376d | 900 | |
8a235efa RW |
901 | if (PageReserved(page) |
902 | && (!kernel_page_present(page) || pfn_is_nosave(pfn))) | |
ae83c5ee | 903 | return NULL; |
25761b6e | 904 | |
ae83c5ee | 905 | return page; |
25761b6e RW |
906 | } |
907 | ||
8357376d RW |
908 | /** |
909 | * count_data_pages - compute the total number of saveable non-highmem | |
910 | * pages. | |
911 | */ | |
912 | ||
fe419535 | 913 | static unsigned int count_data_pages(void) |
25761b6e RW |
914 | { |
915 | struct zone *zone; | |
ae83c5ee | 916 | unsigned long pfn, max_zone_pfn; |
dc19d507 | 917 | unsigned int n = 0; |
25761b6e | 918 | |
8357376d | 919 | for_each_zone(zone) { |
25761b6e RW |
920 | if (is_highmem(zone)) |
921 | continue; | |
8357376d | 922 | |
25761b6e | 923 | mark_free_pages(zone); |
ae83c5ee RW |
924 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
925 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
846705de | 926 | if (saveable_page(zone, pfn)) |
8357376d | 927 | n++; |
25761b6e | 928 | } |
a0f49651 | 929 | return n; |
25761b6e RW |
930 | } |
931 | ||
8357376d RW |
932 | /* This is needed, because copy_page and memcpy are not usable for copying |
933 | * task structs. | |
934 | */ | |
935 | static inline void do_copy_page(long *dst, long *src) | |
f623f0db RW |
936 | { |
937 | int n; | |
938 | ||
f623f0db RW |
939 | for (n = PAGE_SIZE / sizeof(long); n; n--) |
940 | *dst++ = *src++; | |
941 | } | |
942 | ||
8a235efa RW |
943 | |
944 | /** | |
945 | * safe_copy_page - check if the page we are going to copy is marked as | |
946 | * present in the kernel page tables (this always is the case if | |
947 | * CONFIG_DEBUG_PAGEALLOC is not set and in that case | |
948 | * kernel_page_present() always returns 'true'). | |
949 | */ | |
950 | static void safe_copy_page(void *dst, struct page *s_page) | |
951 | { | |
952 | if (kernel_page_present(s_page)) { | |
953 | do_copy_page(dst, page_address(s_page)); | |
954 | } else { | |
955 | kernel_map_pages(s_page, 1, 1); | |
956 | do_copy_page(dst, page_address(s_page)); | |
957 | kernel_map_pages(s_page, 1, 0); | |
958 | } | |
959 | } | |
960 | ||
961 | ||
8357376d RW |
962 | #ifdef CONFIG_HIGHMEM |
963 | static inline struct page * | |
964 | page_is_saveable(struct zone *zone, unsigned long pfn) | |
965 | { | |
966 | return is_highmem(zone) ? | |
846705de | 967 | saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn); |
8357376d RW |
968 | } |
969 | ||
8a235efa | 970 | static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
8357376d RW |
971 | { |
972 | struct page *s_page, *d_page; | |
973 | void *src, *dst; | |
974 | ||
975 | s_page = pfn_to_page(src_pfn); | |
976 | d_page = pfn_to_page(dst_pfn); | |
977 | if (PageHighMem(s_page)) { | |
978 | src = kmap_atomic(s_page, KM_USER0); | |
979 | dst = kmap_atomic(d_page, KM_USER1); | |
980 | do_copy_page(dst, src); | |
981 | kunmap_atomic(src, KM_USER0); | |
982 | kunmap_atomic(dst, KM_USER1); | |
983 | } else { | |
8357376d RW |
984 | if (PageHighMem(d_page)) { |
985 | /* Page pointed to by src may contain some kernel | |
986 | * data modified by kmap_atomic() | |
987 | */ | |
8a235efa | 988 | safe_copy_page(buffer, s_page); |
baa5835d | 989 | dst = kmap_atomic(d_page, KM_USER0); |
8357376d RW |
990 | memcpy(dst, buffer, PAGE_SIZE); |
991 | kunmap_atomic(dst, KM_USER0); | |
992 | } else { | |
8a235efa | 993 | safe_copy_page(page_address(d_page), s_page); |
8357376d RW |
994 | } |
995 | } | |
996 | } | |
997 | #else | |
846705de | 998 | #define page_is_saveable(zone, pfn) saveable_page(zone, pfn) |
8357376d | 999 | |
8a235efa | 1000 | static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
8357376d | 1001 | { |
8a235efa RW |
1002 | safe_copy_page(page_address(pfn_to_page(dst_pfn)), |
1003 | pfn_to_page(src_pfn)); | |
8357376d RW |
1004 | } |
1005 | #endif /* CONFIG_HIGHMEM */ | |
1006 | ||
b788db79 RW |
1007 | static void |
1008 | copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm) | |
25761b6e RW |
1009 | { |
1010 | struct zone *zone; | |
b788db79 | 1011 | unsigned long pfn; |
25761b6e | 1012 | |
8357376d | 1013 | for_each_zone(zone) { |
b788db79 RW |
1014 | unsigned long max_zone_pfn; |
1015 | ||
25761b6e | 1016 | mark_free_pages(zone); |
ae83c5ee | 1017 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
b788db79 | 1018 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
8357376d | 1019 | if (page_is_saveable(zone, pfn)) |
b788db79 | 1020 | memory_bm_set_bit(orig_bm, pfn); |
25761b6e | 1021 | } |
b788db79 RW |
1022 | memory_bm_position_reset(orig_bm); |
1023 | memory_bm_position_reset(copy_bm); | |
df7c4872 | 1024 | for(;;) { |
b788db79 | 1025 | pfn = memory_bm_next_pfn(orig_bm); |
df7c4872 FW |
1026 | if (unlikely(pfn == BM_END_OF_MAP)) |
1027 | break; | |
1028 | copy_data_page(memory_bm_next_pfn(copy_bm), pfn); | |
1029 | } | |
25761b6e RW |
1030 | } |
1031 | ||
8357376d RW |
1032 | /* Total number of image pages */ |
1033 | static unsigned int nr_copy_pages; | |
1034 | /* Number of pages needed for saving the original pfns of the image pages */ | |
1035 | static unsigned int nr_meta_pages; | |
1036 | ||
25761b6e | 1037 | /** |
940864dd | 1038 | * swsusp_free - free pages allocated for the suspend. |
cd560bb2 | 1039 | * |
940864dd RW |
1040 | * Suspend pages are alocated before the atomic copy is made, so we |
1041 | * need to release them after the resume. | |
25761b6e RW |
1042 | */ |
1043 | ||
1044 | void swsusp_free(void) | |
1045 | { | |
1046 | struct zone *zone; | |
ae83c5ee | 1047 | unsigned long pfn, max_zone_pfn; |
25761b6e RW |
1048 | |
1049 | for_each_zone(zone) { | |
ae83c5ee RW |
1050 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
1051 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1052 | if (pfn_valid(pfn)) { | |
1053 | struct page *page = pfn_to_page(pfn); | |
1054 | ||
7be98234 RW |
1055 | if (swsusp_page_is_forbidden(page) && |
1056 | swsusp_page_is_free(page)) { | |
1057 | swsusp_unset_page_forbidden(page); | |
1058 | swsusp_unset_page_free(page); | |
8357376d | 1059 | __free_page(page); |
25761b6e RW |
1060 | } |
1061 | } | |
1062 | } | |
f577eb30 RW |
1063 | nr_copy_pages = 0; |
1064 | nr_meta_pages = 0; | |
75534b50 | 1065 | restore_pblist = NULL; |
6e1819d6 | 1066 | buffer = NULL; |
25761b6e RW |
1067 | } |
1068 | ||
fe419535 RW |
1069 | /** |
1070 | * swsusp_shrink_memory - Try to free as much memory as needed | |
1071 | * | |
1072 | * ... but do not OOM-kill anyone | |
1073 | * | |
1074 | * Notice: all userland should be stopped before it is called, or | |
1075 | * livelock is possible. | |
1076 | */ | |
1077 | ||
1078 | #define SHRINK_BITE 10000 | |
1079 | static inline unsigned long __shrink_memory(long tmp) | |
1080 | { | |
1081 | if (tmp > SHRINK_BITE) | |
1082 | tmp = SHRINK_BITE; | |
1083 | return shrink_all_memory(tmp); | |
1084 | } | |
1085 | ||
1086 | int swsusp_shrink_memory(void) | |
1087 | { | |
1088 | long tmp; | |
1089 | struct zone *zone; | |
1090 | unsigned long pages = 0; | |
1091 | unsigned int i = 0; | |
1092 | char *p = "-\\|/"; | |
1093 | struct timeval start, stop; | |
1094 | ||
1095 | printk(KERN_INFO "PM: Shrinking memory... "); | |
1096 | do_gettimeofday(&start); | |
1097 | do { | |
1098 | long size, highmem_size; | |
1099 | ||
1100 | highmem_size = count_highmem_pages(); | |
1101 | size = count_data_pages() + PAGES_FOR_IO + SPARE_PAGES; | |
1102 | tmp = size; | |
1103 | size += highmem_size; | |
1104 | for_each_populated_zone(zone) { | |
1105 | tmp += snapshot_additional_pages(zone); | |
1106 | if (is_highmem(zone)) { | |
1107 | highmem_size -= | |
1108 | zone_page_state(zone, NR_FREE_PAGES); | |
1109 | } else { | |
1110 | tmp -= zone_page_state(zone, NR_FREE_PAGES); | |
1111 | tmp += zone->lowmem_reserve[ZONE_NORMAL]; | |
1112 | } | |
1113 | } | |
1114 | ||
1115 | if (highmem_size < 0) | |
1116 | highmem_size = 0; | |
1117 | ||
1118 | tmp += highmem_size; | |
1119 | if (tmp > 0) { | |
1120 | tmp = __shrink_memory(tmp); | |
1121 | if (!tmp) | |
1122 | return -ENOMEM; | |
1123 | pages += tmp; | |
1124 | } else if (size > image_size / PAGE_SIZE) { | |
1125 | tmp = __shrink_memory(size - (image_size / PAGE_SIZE)); | |
1126 | pages += tmp; | |
1127 | } | |
1128 | printk("\b%c", p[i++%4]); | |
1129 | } while (tmp > 0); | |
1130 | do_gettimeofday(&stop); | |
1131 | printk("\bdone (%lu pages freed)\n", pages); | |
1132 | swsusp_show_speed(&start, &stop, pages, "Freed"); | |
1133 | ||
1134 | return 0; | |
1135 | } | |
1136 | ||
8357376d RW |
1137 | #ifdef CONFIG_HIGHMEM |
1138 | /** | |
1139 | * count_pages_for_highmem - compute the number of non-highmem pages | |
1140 | * that will be necessary for creating copies of highmem pages. | |
1141 | */ | |
1142 | ||
1143 | static unsigned int count_pages_for_highmem(unsigned int nr_highmem) | |
1144 | { | |
1145 | unsigned int free_highmem = count_free_highmem_pages(); | |
1146 | ||
1147 | if (free_highmem >= nr_highmem) | |
1148 | nr_highmem = 0; | |
1149 | else | |
1150 | nr_highmem -= free_highmem; | |
1151 | ||
1152 | return nr_highmem; | |
1153 | } | |
1154 | #else | |
1155 | static unsigned int | |
1156 | count_pages_for_highmem(unsigned int nr_highmem) { return 0; } | |
1157 | #endif /* CONFIG_HIGHMEM */ | |
25761b6e RW |
1158 | |
1159 | /** | |
8357376d RW |
1160 | * enough_free_mem - Make sure we have enough free memory for the |
1161 | * snapshot image. | |
25761b6e RW |
1162 | */ |
1163 | ||
8357376d | 1164 | static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem) |
25761b6e | 1165 | { |
e5e2fa78 | 1166 | struct zone *zone; |
940864dd | 1167 | unsigned int free = 0, meta = 0; |
e5e2fa78 | 1168 | |
8357376d RW |
1169 | for_each_zone(zone) { |
1170 | meta += snapshot_additional_pages(zone); | |
1171 | if (!is_highmem(zone)) | |
d23ad423 | 1172 | free += zone_page_state(zone, NR_FREE_PAGES); |
8357376d | 1173 | } |
940864dd | 1174 | |
8357376d | 1175 | nr_pages += count_pages_for_highmem(nr_highmem); |
23976728 | 1176 | pr_debug("PM: Normal pages needed: %u + %u + %u, available pages: %u\n", |
940864dd RW |
1177 | nr_pages, PAGES_FOR_IO, meta, free); |
1178 | ||
1179 | return free > nr_pages + PAGES_FOR_IO + meta; | |
25761b6e RW |
1180 | } |
1181 | ||
8357376d RW |
1182 | #ifdef CONFIG_HIGHMEM |
1183 | /** | |
1184 | * get_highmem_buffer - if there are some highmem pages in the suspend | |
1185 | * image, we may need the buffer to copy them and/or load their data. | |
1186 | */ | |
1187 | ||
1188 | static inline int get_highmem_buffer(int safe_needed) | |
1189 | { | |
1190 | buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed); | |
1191 | return buffer ? 0 : -ENOMEM; | |
1192 | } | |
1193 | ||
1194 | /** | |
1195 | * alloc_highmem_image_pages - allocate some highmem pages for the image. | |
1196 | * Try to allocate as many pages as needed, but if the number of free | |
1197 | * highmem pages is lesser than that, allocate them all. | |
1198 | */ | |
1199 | ||
1200 | static inline unsigned int | |
1201 | alloc_highmem_image_pages(struct memory_bitmap *bm, unsigned int nr_highmem) | |
1202 | { | |
1203 | unsigned int to_alloc = count_free_highmem_pages(); | |
1204 | ||
1205 | if (to_alloc > nr_highmem) | |
1206 | to_alloc = nr_highmem; | |
1207 | ||
1208 | nr_highmem -= to_alloc; | |
1209 | while (to_alloc-- > 0) { | |
1210 | struct page *page; | |
1211 | ||
1212 | page = alloc_image_page(__GFP_HIGHMEM); | |
1213 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1214 | } | |
1215 | return nr_highmem; | |
1216 | } | |
1217 | #else | |
1218 | static inline int get_highmem_buffer(int safe_needed) { return 0; } | |
1219 | ||
1220 | static inline unsigned int | |
1221 | alloc_highmem_image_pages(struct memory_bitmap *bm, unsigned int n) { return 0; } | |
1222 | #endif /* CONFIG_HIGHMEM */ | |
1223 | ||
1224 | /** | |
1225 | * swsusp_alloc - allocate memory for the suspend image | |
1226 | * | |
1227 | * We first try to allocate as many highmem pages as there are | |
1228 | * saveable highmem pages in the system. If that fails, we allocate | |
1229 | * non-highmem pages for the copies of the remaining highmem ones. | |
1230 | * | |
1231 | * In this approach it is likely that the copies of highmem pages will | |
1232 | * also be located in the high memory, because of the way in which | |
1233 | * copy_data_pages() works. | |
1234 | */ | |
1235 | ||
b788db79 RW |
1236 | static int |
1237 | swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm, | |
8357376d | 1238 | unsigned int nr_pages, unsigned int nr_highmem) |
054bd4c1 | 1239 | { |
b788db79 | 1240 | int error; |
054bd4c1 | 1241 | |
b788db79 RW |
1242 | error = memory_bm_create(orig_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY); |
1243 | if (error) | |
1244 | goto Free; | |
25761b6e | 1245 | |
b788db79 RW |
1246 | error = memory_bm_create(copy_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY); |
1247 | if (error) | |
1248 | goto Free; | |
25761b6e | 1249 | |
8357376d RW |
1250 | if (nr_highmem > 0) { |
1251 | error = get_highmem_buffer(PG_ANY); | |
1252 | if (error) | |
1253 | goto Free; | |
1254 | ||
1255 | nr_pages += alloc_highmem_image_pages(copy_bm, nr_highmem); | |
1256 | } | |
b788db79 | 1257 | while (nr_pages-- > 0) { |
8357376d RW |
1258 | struct page *page = alloc_image_page(GFP_ATOMIC | __GFP_COLD); |
1259 | ||
b788db79 RW |
1260 | if (!page) |
1261 | goto Free; | |
25761b6e | 1262 | |
b788db79 | 1263 | memory_bm_set_bit(copy_bm, page_to_pfn(page)); |
25761b6e | 1264 | } |
b788db79 | 1265 | return 0; |
25761b6e | 1266 | |
59a49335 | 1267 | Free: |
b788db79 RW |
1268 | swsusp_free(); |
1269 | return -ENOMEM; | |
25761b6e RW |
1270 | } |
1271 | ||
8357376d RW |
1272 | /* Memory bitmap used for marking saveable pages (during suspend) or the |
1273 | * suspend image pages (during resume) | |
1274 | */ | |
b788db79 | 1275 | static struct memory_bitmap orig_bm; |
8357376d RW |
1276 | /* Memory bitmap used on suspend for marking allocated pages that will contain |
1277 | * the copies of saveable pages. During resume it is initially used for | |
1278 | * marking the suspend image pages, but then its set bits are duplicated in | |
1279 | * @orig_bm and it is released. Next, on systems with high memory, it may be | |
1280 | * used for marking "safe" highmem pages, but it has to be reinitialized for | |
1281 | * this purpose. | |
b788db79 RW |
1282 | */ |
1283 | static struct memory_bitmap copy_bm; | |
1284 | ||
2e32a43e | 1285 | asmlinkage int swsusp_save(void) |
25761b6e | 1286 | { |
8357376d | 1287 | unsigned int nr_pages, nr_highmem; |
25761b6e | 1288 | |
23976728 | 1289 | printk(KERN_INFO "PM: Creating hibernation image: \n"); |
25761b6e | 1290 | |
9f8f2172 | 1291 | drain_local_pages(NULL); |
a0f49651 | 1292 | nr_pages = count_data_pages(); |
8357376d | 1293 | nr_highmem = count_highmem_pages(); |
23976728 | 1294 | printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem); |
25761b6e | 1295 | |
8357376d | 1296 | if (!enough_free_mem(nr_pages, nr_highmem)) { |
23976728 | 1297 | printk(KERN_ERR "PM: Not enough free memory\n"); |
25761b6e RW |
1298 | return -ENOMEM; |
1299 | } | |
1300 | ||
8357376d | 1301 | if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) { |
23976728 | 1302 | printk(KERN_ERR "PM: Memory allocation failed\n"); |
a0f49651 | 1303 | return -ENOMEM; |
8357376d | 1304 | } |
25761b6e RW |
1305 | |
1306 | /* During allocating of suspend pagedir, new cold pages may appear. | |
1307 | * Kill them. | |
1308 | */ | |
9f8f2172 | 1309 | drain_local_pages(NULL); |
b788db79 | 1310 | copy_data_pages(©_bm, &orig_bm); |
25761b6e RW |
1311 | |
1312 | /* | |
1313 | * End of critical section. From now on, we can write to memory, | |
1314 | * but we should not touch disk. This specially means we must _not_ | |
1315 | * touch swap space! Except we must write out our image of course. | |
1316 | */ | |
1317 | ||
8357376d | 1318 | nr_pages += nr_highmem; |
a0f49651 | 1319 | nr_copy_pages = nr_pages; |
8357376d | 1320 | nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE); |
a0f49651 | 1321 | |
23976728 RW |
1322 | printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n", |
1323 | nr_pages); | |
8357376d | 1324 | |
25761b6e RW |
1325 | return 0; |
1326 | } | |
f577eb30 | 1327 | |
d307c4a8 RW |
1328 | #ifndef CONFIG_ARCH_HIBERNATION_HEADER |
1329 | static int init_header_complete(struct swsusp_info *info) | |
f577eb30 | 1330 | { |
d307c4a8 | 1331 | memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname)); |
f577eb30 | 1332 | info->version_code = LINUX_VERSION_CODE; |
d307c4a8 RW |
1333 | return 0; |
1334 | } | |
1335 | ||
1336 | static char *check_image_kernel(struct swsusp_info *info) | |
1337 | { | |
1338 | if (info->version_code != LINUX_VERSION_CODE) | |
1339 | return "kernel version"; | |
1340 | if (strcmp(info->uts.sysname,init_utsname()->sysname)) | |
1341 | return "system type"; | |
1342 | if (strcmp(info->uts.release,init_utsname()->release)) | |
1343 | return "kernel release"; | |
1344 | if (strcmp(info->uts.version,init_utsname()->version)) | |
1345 | return "version"; | |
1346 | if (strcmp(info->uts.machine,init_utsname()->machine)) | |
1347 | return "machine"; | |
1348 | return NULL; | |
1349 | } | |
1350 | #endif /* CONFIG_ARCH_HIBERNATION_HEADER */ | |
1351 | ||
af508b34 RW |
1352 | unsigned long snapshot_get_image_size(void) |
1353 | { | |
1354 | return nr_copy_pages + nr_meta_pages + 1; | |
1355 | } | |
1356 | ||
d307c4a8 RW |
1357 | static int init_header(struct swsusp_info *info) |
1358 | { | |
1359 | memset(info, 0, sizeof(struct swsusp_info)); | |
f577eb30 | 1360 | info->num_physpages = num_physpages; |
f577eb30 | 1361 | info->image_pages = nr_copy_pages; |
af508b34 | 1362 | info->pages = snapshot_get_image_size(); |
6e1819d6 RW |
1363 | info->size = info->pages; |
1364 | info->size <<= PAGE_SHIFT; | |
d307c4a8 | 1365 | return init_header_complete(info); |
f577eb30 RW |
1366 | } |
1367 | ||
1368 | /** | |
940864dd RW |
1369 | * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm |
1370 | * are stored in the array @buf[] (1 page at a time) | |
f577eb30 RW |
1371 | */ |
1372 | ||
b788db79 | 1373 | static inline void |
940864dd | 1374 | pack_pfns(unsigned long *buf, struct memory_bitmap *bm) |
f577eb30 RW |
1375 | { |
1376 | int j; | |
1377 | ||
b788db79 | 1378 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
940864dd RW |
1379 | buf[j] = memory_bm_next_pfn(bm); |
1380 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
b788db79 | 1381 | break; |
f577eb30 | 1382 | } |
f577eb30 RW |
1383 | } |
1384 | ||
1385 | /** | |
1386 | * snapshot_read_next - used for reading the system memory snapshot. | |
1387 | * | |
1388 | * On the first call to it @handle should point to a zeroed | |
1389 | * snapshot_handle structure. The structure gets updated and a pointer | |
1390 | * to it should be passed to this function every next time. | |
1391 | * | |
1392 | * The @count parameter should contain the number of bytes the caller | |
1393 | * wants to read from the snapshot. It must not be zero. | |
1394 | * | |
1395 | * On success the function returns a positive number. Then, the caller | |
1396 | * is allowed to read up to the returned number of bytes from the memory | |
1397 | * location computed by the data_of() macro. The number returned | |
1398 | * may be smaller than @count, but this only happens if the read would | |
1399 | * cross a page boundary otherwise. | |
1400 | * | |
1401 | * The function returns 0 to indicate the end of data stream condition, | |
1402 | * and a negative number is returned on error. In such cases the | |
1403 | * structure pointed to by @handle is not updated and should not be used | |
1404 | * any more. | |
1405 | */ | |
1406 | ||
1407 | int snapshot_read_next(struct snapshot_handle *handle, size_t count) | |
1408 | { | |
fb13a28b | 1409 | if (handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 1410 | return 0; |
b788db79 | 1411 | |
f577eb30 RW |
1412 | if (!buffer) { |
1413 | /* This makes the buffer be freed by swsusp_free() */ | |
8357376d | 1414 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); |
f577eb30 RW |
1415 | if (!buffer) |
1416 | return -ENOMEM; | |
1417 | } | |
1418 | if (!handle->offset) { | |
d307c4a8 RW |
1419 | int error; |
1420 | ||
1421 | error = init_header((struct swsusp_info *)buffer); | |
1422 | if (error) | |
1423 | return error; | |
f577eb30 | 1424 | handle->buffer = buffer; |
b788db79 RW |
1425 | memory_bm_position_reset(&orig_bm); |
1426 | memory_bm_position_reset(©_bm); | |
f577eb30 | 1427 | } |
fb13a28b RW |
1428 | if (handle->prev < handle->cur) { |
1429 | if (handle->cur <= nr_meta_pages) { | |
b788db79 | 1430 | memset(buffer, 0, PAGE_SIZE); |
940864dd | 1431 | pack_pfns(buffer, &orig_bm); |
f577eb30 | 1432 | } else { |
8357376d | 1433 | struct page *page; |
b788db79 | 1434 | |
8357376d RW |
1435 | page = pfn_to_page(memory_bm_next_pfn(©_bm)); |
1436 | if (PageHighMem(page)) { | |
1437 | /* Highmem pages are copied to the buffer, | |
1438 | * because we can't return with a kmapped | |
1439 | * highmem page (we may not be called again). | |
1440 | */ | |
1441 | void *kaddr; | |
1442 | ||
1443 | kaddr = kmap_atomic(page, KM_USER0); | |
1444 | memcpy(buffer, kaddr, PAGE_SIZE); | |
1445 | kunmap_atomic(kaddr, KM_USER0); | |
1446 | handle->buffer = buffer; | |
1447 | } else { | |
1448 | handle->buffer = page_address(page); | |
1449 | } | |
f577eb30 | 1450 | } |
fb13a28b | 1451 | handle->prev = handle->cur; |
f577eb30 | 1452 | } |
fb13a28b RW |
1453 | handle->buf_offset = handle->cur_offset; |
1454 | if (handle->cur_offset + count >= PAGE_SIZE) { | |
1455 | count = PAGE_SIZE - handle->cur_offset; | |
1456 | handle->cur_offset = 0; | |
1457 | handle->cur++; | |
f577eb30 | 1458 | } else { |
fb13a28b | 1459 | handle->cur_offset += count; |
f577eb30 RW |
1460 | } |
1461 | handle->offset += count; | |
1462 | return count; | |
1463 | } | |
1464 | ||
1465 | /** | |
1466 | * mark_unsafe_pages - mark the pages that cannot be used for storing | |
1467 | * the image during resume, because they conflict with the pages that | |
1468 | * had been used before suspend | |
1469 | */ | |
1470 | ||
940864dd | 1471 | static int mark_unsafe_pages(struct memory_bitmap *bm) |
f577eb30 RW |
1472 | { |
1473 | struct zone *zone; | |
ae83c5ee | 1474 | unsigned long pfn, max_zone_pfn; |
f577eb30 RW |
1475 | |
1476 | /* Clear page flags */ | |
8357376d | 1477 | for_each_zone(zone) { |
ae83c5ee RW |
1478 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
1479 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1480 | if (pfn_valid(pfn)) | |
7be98234 | 1481 | swsusp_unset_page_free(pfn_to_page(pfn)); |
f577eb30 RW |
1482 | } |
1483 | ||
940864dd RW |
1484 | /* Mark pages that correspond to the "original" pfns as "unsafe" */ |
1485 | memory_bm_position_reset(bm); | |
1486 | do { | |
1487 | pfn = memory_bm_next_pfn(bm); | |
1488 | if (likely(pfn != BM_END_OF_MAP)) { | |
1489 | if (likely(pfn_valid(pfn))) | |
7be98234 | 1490 | swsusp_set_page_free(pfn_to_page(pfn)); |
940864dd RW |
1491 | else |
1492 | return -EFAULT; | |
1493 | } | |
1494 | } while (pfn != BM_END_OF_MAP); | |
f577eb30 | 1495 | |
940864dd | 1496 | allocated_unsafe_pages = 0; |
968808b8 | 1497 | |
f577eb30 RW |
1498 | return 0; |
1499 | } | |
1500 | ||
940864dd RW |
1501 | static void |
1502 | duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src) | |
f577eb30 | 1503 | { |
940864dd RW |
1504 | unsigned long pfn; |
1505 | ||
1506 | memory_bm_position_reset(src); | |
1507 | pfn = memory_bm_next_pfn(src); | |
1508 | while (pfn != BM_END_OF_MAP) { | |
1509 | memory_bm_set_bit(dst, pfn); | |
1510 | pfn = memory_bm_next_pfn(src); | |
f577eb30 RW |
1511 | } |
1512 | } | |
1513 | ||
d307c4a8 | 1514 | static int check_header(struct swsusp_info *info) |
f577eb30 | 1515 | { |
d307c4a8 | 1516 | char *reason; |
f577eb30 | 1517 | |
d307c4a8 RW |
1518 | reason = check_image_kernel(info); |
1519 | if (!reason && info->num_physpages != num_physpages) | |
f577eb30 | 1520 | reason = "memory size"; |
f577eb30 | 1521 | if (reason) { |
23976728 | 1522 | printk(KERN_ERR "PM: Image mismatch: %s\n", reason); |
f577eb30 RW |
1523 | return -EPERM; |
1524 | } | |
1525 | return 0; | |
1526 | } | |
1527 | ||
1528 | /** | |
1529 | * load header - check the image header and copy data from it | |
1530 | */ | |
1531 | ||
940864dd RW |
1532 | static int |
1533 | load_header(struct swsusp_info *info) | |
f577eb30 RW |
1534 | { |
1535 | int error; | |
f577eb30 | 1536 | |
940864dd | 1537 | restore_pblist = NULL; |
f577eb30 RW |
1538 | error = check_header(info); |
1539 | if (!error) { | |
f577eb30 RW |
1540 | nr_copy_pages = info->image_pages; |
1541 | nr_meta_pages = info->pages - info->image_pages - 1; | |
1542 | } | |
1543 | return error; | |
1544 | } | |
1545 | ||
1546 | /** | |
940864dd RW |
1547 | * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set |
1548 | * the corresponding bit in the memory bitmap @bm | |
f577eb30 | 1549 | */ |
69643279 | 1550 | static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm) |
f577eb30 RW |
1551 | { |
1552 | int j; | |
1553 | ||
940864dd RW |
1554 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
1555 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
1556 | break; | |
1557 | ||
69643279 RW |
1558 | if (memory_bm_pfn_present(bm, buf[j])) |
1559 | memory_bm_set_bit(bm, buf[j]); | |
1560 | else | |
1561 | return -EFAULT; | |
f577eb30 | 1562 | } |
69643279 RW |
1563 | |
1564 | return 0; | |
f577eb30 RW |
1565 | } |
1566 | ||
8357376d RW |
1567 | /* List of "safe" pages that may be used to store data loaded from the suspend |
1568 | * image | |
1569 | */ | |
1570 | static struct linked_page *safe_pages_list; | |
1571 | ||
1572 | #ifdef CONFIG_HIGHMEM | |
1573 | /* struct highmem_pbe is used for creating the list of highmem pages that | |
1574 | * should be restored atomically during the resume from disk, because the page | |
1575 | * frames they have occupied before the suspend are in use. | |
1576 | */ | |
1577 | struct highmem_pbe { | |
1578 | struct page *copy_page; /* data is here now */ | |
1579 | struct page *orig_page; /* data was here before the suspend */ | |
1580 | struct highmem_pbe *next; | |
1581 | }; | |
1582 | ||
1583 | /* List of highmem PBEs needed for restoring the highmem pages that were | |
1584 | * allocated before the suspend and included in the suspend image, but have | |
1585 | * also been allocated by the "resume" kernel, so their contents cannot be | |
1586 | * written directly to their "original" page frames. | |
1587 | */ | |
1588 | static struct highmem_pbe *highmem_pblist; | |
1589 | ||
1590 | /** | |
1591 | * count_highmem_image_pages - compute the number of highmem pages in the | |
1592 | * suspend image. The bits in the memory bitmap @bm that correspond to the | |
1593 | * image pages are assumed to be set. | |
1594 | */ | |
1595 | ||
1596 | static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) | |
1597 | { | |
1598 | unsigned long pfn; | |
1599 | unsigned int cnt = 0; | |
1600 | ||
1601 | memory_bm_position_reset(bm); | |
1602 | pfn = memory_bm_next_pfn(bm); | |
1603 | while (pfn != BM_END_OF_MAP) { | |
1604 | if (PageHighMem(pfn_to_page(pfn))) | |
1605 | cnt++; | |
1606 | ||
1607 | pfn = memory_bm_next_pfn(bm); | |
1608 | } | |
1609 | return cnt; | |
1610 | } | |
1611 | ||
1612 | /** | |
1613 | * prepare_highmem_image - try to allocate as many highmem pages as | |
1614 | * there are highmem image pages (@nr_highmem_p points to the variable | |
1615 | * containing the number of highmem image pages). The pages that are | |
1616 | * "safe" (ie. will not be overwritten when the suspend image is | |
1617 | * restored) have the corresponding bits set in @bm (it must be | |
1618 | * unitialized). | |
1619 | * | |
1620 | * NOTE: This function should not be called if there are no highmem | |
1621 | * image pages. | |
1622 | */ | |
1623 | ||
1624 | static unsigned int safe_highmem_pages; | |
1625 | ||
1626 | static struct memory_bitmap *safe_highmem_bm; | |
1627 | ||
1628 | static int | |
1629 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
1630 | { | |
1631 | unsigned int to_alloc; | |
1632 | ||
1633 | if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE)) | |
1634 | return -ENOMEM; | |
1635 | ||
1636 | if (get_highmem_buffer(PG_SAFE)) | |
1637 | return -ENOMEM; | |
1638 | ||
1639 | to_alloc = count_free_highmem_pages(); | |
1640 | if (to_alloc > *nr_highmem_p) | |
1641 | to_alloc = *nr_highmem_p; | |
1642 | else | |
1643 | *nr_highmem_p = to_alloc; | |
1644 | ||
1645 | safe_highmem_pages = 0; | |
1646 | while (to_alloc-- > 0) { | |
1647 | struct page *page; | |
1648 | ||
1649 | page = alloc_page(__GFP_HIGHMEM); | |
7be98234 | 1650 | if (!swsusp_page_is_free(page)) { |
8357376d RW |
1651 | /* The page is "safe", set its bit the bitmap */ |
1652 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1653 | safe_highmem_pages++; | |
1654 | } | |
1655 | /* Mark the page as allocated */ | |
7be98234 RW |
1656 | swsusp_set_page_forbidden(page); |
1657 | swsusp_set_page_free(page); | |
8357376d RW |
1658 | } |
1659 | memory_bm_position_reset(bm); | |
1660 | safe_highmem_bm = bm; | |
1661 | return 0; | |
1662 | } | |
1663 | ||
1664 | /** | |
1665 | * get_highmem_page_buffer - for given highmem image page find the buffer | |
1666 | * that suspend_write_next() should set for its caller to write to. | |
1667 | * | |
1668 | * If the page is to be saved to its "original" page frame or a copy of | |
1669 | * the page is to be made in the highmem, @buffer is returned. Otherwise, | |
1670 | * the copy of the page is to be made in normal memory, so the address of | |
1671 | * the copy is returned. | |
1672 | * | |
1673 | * If @buffer is returned, the caller of suspend_write_next() will write | |
1674 | * the page's contents to @buffer, so they will have to be copied to the | |
1675 | * right location on the next call to suspend_write_next() and it is done | |
1676 | * with the help of copy_last_highmem_page(). For this purpose, if | |
1677 | * @buffer is returned, @last_highmem page is set to the page to which | |
1678 | * the data will have to be copied from @buffer. | |
1679 | */ | |
1680 | ||
1681 | static struct page *last_highmem_page; | |
1682 | ||
1683 | static void * | |
1684 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
1685 | { | |
1686 | struct highmem_pbe *pbe; | |
1687 | void *kaddr; | |
1688 | ||
7be98234 | 1689 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) { |
8357376d RW |
1690 | /* We have allocated the "original" page frame and we can |
1691 | * use it directly to store the loaded page. | |
1692 | */ | |
1693 | last_highmem_page = page; | |
1694 | return buffer; | |
1695 | } | |
1696 | /* The "original" page frame has not been allocated and we have to | |
1697 | * use a "safe" page frame to store the loaded page. | |
1698 | */ | |
1699 | pbe = chain_alloc(ca, sizeof(struct highmem_pbe)); | |
1700 | if (!pbe) { | |
1701 | swsusp_free(); | |
69643279 | 1702 | return ERR_PTR(-ENOMEM); |
8357376d RW |
1703 | } |
1704 | pbe->orig_page = page; | |
1705 | if (safe_highmem_pages > 0) { | |
1706 | struct page *tmp; | |
1707 | ||
1708 | /* Copy of the page will be stored in high memory */ | |
1709 | kaddr = buffer; | |
1710 | tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm)); | |
1711 | safe_highmem_pages--; | |
1712 | last_highmem_page = tmp; | |
1713 | pbe->copy_page = tmp; | |
1714 | } else { | |
1715 | /* Copy of the page will be stored in normal memory */ | |
1716 | kaddr = safe_pages_list; | |
1717 | safe_pages_list = safe_pages_list->next; | |
1718 | pbe->copy_page = virt_to_page(kaddr); | |
1719 | } | |
1720 | pbe->next = highmem_pblist; | |
1721 | highmem_pblist = pbe; | |
1722 | return kaddr; | |
1723 | } | |
1724 | ||
1725 | /** | |
1726 | * copy_last_highmem_page - copy the contents of a highmem image from | |
1727 | * @buffer, where the caller of snapshot_write_next() has place them, | |
1728 | * to the right location represented by @last_highmem_page . | |
1729 | */ | |
1730 | ||
1731 | static void copy_last_highmem_page(void) | |
1732 | { | |
1733 | if (last_highmem_page) { | |
1734 | void *dst; | |
1735 | ||
1736 | dst = kmap_atomic(last_highmem_page, KM_USER0); | |
1737 | memcpy(dst, buffer, PAGE_SIZE); | |
1738 | kunmap_atomic(dst, KM_USER0); | |
1739 | last_highmem_page = NULL; | |
1740 | } | |
1741 | } | |
1742 | ||
1743 | static inline int last_highmem_page_copied(void) | |
1744 | { | |
1745 | return !last_highmem_page; | |
1746 | } | |
1747 | ||
1748 | static inline void free_highmem_data(void) | |
1749 | { | |
1750 | if (safe_highmem_bm) | |
1751 | memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR); | |
1752 | ||
1753 | if (buffer) | |
1754 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
1755 | } | |
1756 | #else | |
1757 | static inline int get_safe_write_buffer(void) { return 0; } | |
1758 | ||
1759 | static unsigned int | |
1760 | count_highmem_image_pages(struct memory_bitmap *bm) { return 0; } | |
1761 | ||
1762 | static inline int | |
1763 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
1764 | { | |
1765 | return 0; | |
1766 | } | |
1767 | ||
1768 | static inline void * | |
1769 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
1770 | { | |
69643279 | 1771 | return ERR_PTR(-EINVAL); |
8357376d RW |
1772 | } |
1773 | ||
1774 | static inline void copy_last_highmem_page(void) {} | |
1775 | static inline int last_highmem_page_copied(void) { return 1; } | |
1776 | static inline void free_highmem_data(void) {} | |
1777 | #endif /* CONFIG_HIGHMEM */ | |
1778 | ||
f577eb30 | 1779 | /** |
940864dd RW |
1780 | * prepare_image - use the memory bitmap @bm to mark the pages that will |
1781 | * be overwritten in the process of restoring the system memory state | |
1782 | * from the suspend image ("unsafe" pages) and allocate memory for the | |
1783 | * image. | |
968808b8 | 1784 | * |
940864dd RW |
1785 | * The idea is to allocate a new memory bitmap first and then allocate |
1786 | * as many pages as needed for the image data, but not to assign these | |
1787 | * pages to specific tasks initially. Instead, we just mark them as | |
8357376d RW |
1788 | * allocated and create a lists of "safe" pages that will be used |
1789 | * later. On systems with high memory a list of "safe" highmem pages is | |
1790 | * also created. | |
f577eb30 RW |
1791 | */ |
1792 | ||
940864dd RW |
1793 | #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe)) |
1794 | ||
940864dd RW |
1795 | static int |
1796 | prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm) | |
f577eb30 | 1797 | { |
8357376d | 1798 | unsigned int nr_pages, nr_highmem; |
940864dd RW |
1799 | struct linked_page *sp_list, *lp; |
1800 | int error; | |
f577eb30 | 1801 | |
8357376d RW |
1802 | /* If there is no highmem, the buffer will not be necessary */ |
1803 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
1804 | buffer = NULL; | |
1805 | ||
1806 | nr_highmem = count_highmem_image_pages(bm); | |
940864dd RW |
1807 | error = mark_unsafe_pages(bm); |
1808 | if (error) | |
1809 | goto Free; | |
1810 | ||
1811 | error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE); | |
1812 | if (error) | |
1813 | goto Free; | |
1814 | ||
1815 | duplicate_memory_bitmap(new_bm, bm); | |
1816 | memory_bm_free(bm, PG_UNSAFE_KEEP); | |
8357376d RW |
1817 | if (nr_highmem > 0) { |
1818 | error = prepare_highmem_image(bm, &nr_highmem); | |
1819 | if (error) | |
1820 | goto Free; | |
1821 | } | |
940864dd RW |
1822 | /* Reserve some safe pages for potential later use. |
1823 | * | |
1824 | * NOTE: This way we make sure there will be enough safe pages for the | |
1825 | * chain_alloc() in get_buffer(). It is a bit wasteful, but | |
1826 | * nr_copy_pages cannot be greater than 50% of the memory anyway. | |
1827 | */ | |
1828 | sp_list = NULL; | |
1829 | /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */ | |
8357376d | 1830 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
1831 | nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE); |
1832 | while (nr_pages > 0) { | |
8357376d | 1833 | lp = get_image_page(GFP_ATOMIC, PG_SAFE); |
940864dd | 1834 | if (!lp) { |
f577eb30 | 1835 | error = -ENOMEM; |
940864dd RW |
1836 | goto Free; |
1837 | } | |
1838 | lp->next = sp_list; | |
1839 | sp_list = lp; | |
1840 | nr_pages--; | |
f577eb30 | 1841 | } |
940864dd RW |
1842 | /* Preallocate memory for the image */ |
1843 | safe_pages_list = NULL; | |
8357376d | 1844 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
1845 | while (nr_pages > 0) { |
1846 | lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC); | |
1847 | if (!lp) { | |
1848 | error = -ENOMEM; | |
1849 | goto Free; | |
1850 | } | |
7be98234 | 1851 | if (!swsusp_page_is_free(virt_to_page(lp))) { |
940864dd RW |
1852 | /* The page is "safe", add it to the list */ |
1853 | lp->next = safe_pages_list; | |
1854 | safe_pages_list = lp; | |
968808b8 | 1855 | } |
940864dd | 1856 | /* Mark the page as allocated */ |
7be98234 RW |
1857 | swsusp_set_page_forbidden(virt_to_page(lp)); |
1858 | swsusp_set_page_free(virt_to_page(lp)); | |
940864dd | 1859 | nr_pages--; |
968808b8 | 1860 | } |
940864dd RW |
1861 | /* Free the reserved safe pages so that chain_alloc() can use them */ |
1862 | while (sp_list) { | |
1863 | lp = sp_list->next; | |
1864 | free_image_page(sp_list, PG_UNSAFE_CLEAR); | |
1865 | sp_list = lp; | |
f577eb30 | 1866 | } |
940864dd RW |
1867 | return 0; |
1868 | ||
59a49335 | 1869 | Free: |
940864dd | 1870 | swsusp_free(); |
f577eb30 RW |
1871 | return error; |
1872 | } | |
1873 | ||
940864dd RW |
1874 | /** |
1875 | * get_buffer - compute the address that snapshot_write_next() should | |
1876 | * set for its caller to write to. | |
1877 | */ | |
1878 | ||
1879 | static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca) | |
968808b8 | 1880 | { |
940864dd | 1881 | struct pbe *pbe; |
69643279 RW |
1882 | struct page *page; |
1883 | unsigned long pfn = memory_bm_next_pfn(bm); | |
968808b8 | 1884 | |
69643279 RW |
1885 | if (pfn == BM_END_OF_MAP) |
1886 | return ERR_PTR(-EFAULT); | |
1887 | ||
1888 | page = pfn_to_page(pfn); | |
8357376d RW |
1889 | if (PageHighMem(page)) |
1890 | return get_highmem_page_buffer(page, ca); | |
1891 | ||
7be98234 | 1892 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) |
940864dd RW |
1893 | /* We have allocated the "original" page frame and we can |
1894 | * use it directly to store the loaded page. | |
968808b8 | 1895 | */ |
940864dd RW |
1896 | return page_address(page); |
1897 | ||
1898 | /* The "original" page frame has not been allocated and we have to | |
1899 | * use a "safe" page frame to store the loaded page. | |
968808b8 | 1900 | */ |
940864dd RW |
1901 | pbe = chain_alloc(ca, sizeof(struct pbe)); |
1902 | if (!pbe) { | |
1903 | swsusp_free(); | |
69643279 | 1904 | return ERR_PTR(-ENOMEM); |
940864dd | 1905 | } |
8357376d RW |
1906 | pbe->orig_address = page_address(page); |
1907 | pbe->address = safe_pages_list; | |
940864dd RW |
1908 | safe_pages_list = safe_pages_list->next; |
1909 | pbe->next = restore_pblist; | |
1910 | restore_pblist = pbe; | |
8357376d | 1911 | return pbe->address; |
968808b8 RW |
1912 | } |
1913 | ||
f577eb30 RW |
1914 | /** |
1915 | * snapshot_write_next - used for writing the system memory snapshot. | |
1916 | * | |
1917 | * On the first call to it @handle should point to a zeroed | |
1918 | * snapshot_handle structure. The structure gets updated and a pointer | |
1919 | * to it should be passed to this function every next time. | |
1920 | * | |
1921 | * The @count parameter should contain the number of bytes the caller | |
1922 | * wants to write to the image. It must not be zero. | |
1923 | * | |
1924 | * On success the function returns a positive number. Then, the caller | |
1925 | * is allowed to write up to the returned number of bytes to the memory | |
1926 | * location computed by the data_of() macro. The number returned | |
1927 | * may be smaller than @count, but this only happens if the write would | |
1928 | * cross a page boundary otherwise. | |
1929 | * | |
1930 | * The function returns 0 to indicate the "end of file" condition, | |
1931 | * and a negative number is returned on error. In such cases the | |
1932 | * structure pointed to by @handle is not updated and should not be used | |
1933 | * any more. | |
1934 | */ | |
1935 | ||
1936 | int snapshot_write_next(struct snapshot_handle *handle, size_t count) | |
1937 | { | |
940864dd | 1938 | static struct chain_allocator ca; |
f577eb30 RW |
1939 | int error = 0; |
1940 | ||
940864dd | 1941 | /* Check if we have already loaded the entire image */ |
fb13a28b | 1942 | if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 1943 | return 0; |
940864dd | 1944 | |
8357376d RW |
1945 | if (handle->offset == 0) { |
1946 | if (!buffer) | |
1947 | /* This makes the buffer be freed by swsusp_free() */ | |
1948 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); | |
1949 | ||
f577eb30 RW |
1950 | if (!buffer) |
1951 | return -ENOMEM; | |
8357376d | 1952 | |
f577eb30 | 1953 | handle->buffer = buffer; |
8357376d | 1954 | } |
546e0d27 | 1955 | handle->sync_read = 1; |
fb13a28b | 1956 | if (handle->prev < handle->cur) { |
940864dd RW |
1957 | if (handle->prev == 0) { |
1958 | error = load_header(buffer); | |
1959 | if (error) | |
1960 | return error; | |
1961 | ||
1962 | error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY); | |
f577eb30 RW |
1963 | if (error) |
1964 | return error; | |
940864dd | 1965 | |
f577eb30 | 1966 | } else if (handle->prev <= nr_meta_pages) { |
69643279 RW |
1967 | error = unpack_orig_pfns(buffer, ©_bm); |
1968 | if (error) | |
1969 | return error; | |
1970 | ||
940864dd RW |
1971 | if (handle->prev == nr_meta_pages) { |
1972 | error = prepare_image(&orig_bm, ©_bm); | |
f577eb30 RW |
1973 | if (error) |
1974 | return error; | |
940864dd RW |
1975 | |
1976 | chain_init(&ca, GFP_ATOMIC, PG_SAFE); | |
1977 | memory_bm_position_reset(&orig_bm); | |
1978 | restore_pblist = NULL; | |
1979 | handle->buffer = get_buffer(&orig_bm, &ca); | |
546e0d27 | 1980 | handle->sync_read = 0; |
69643279 RW |
1981 | if (IS_ERR(handle->buffer)) |
1982 | return PTR_ERR(handle->buffer); | |
f577eb30 RW |
1983 | } |
1984 | } else { | |
8357376d | 1985 | copy_last_highmem_page(); |
940864dd | 1986 | handle->buffer = get_buffer(&orig_bm, &ca); |
69643279 RW |
1987 | if (IS_ERR(handle->buffer)) |
1988 | return PTR_ERR(handle->buffer); | |
8357376d RW |
1989 | if (handle->buffer != buffer) |
1990 | handle->sync_read = 0; | |
f577eb30 | 1991 | } |
fb13a28b | 1992 | handle->prev = handle->cur; |
f577eb30 | 1993 | } |
fb13a28b RW |
1994 | handle->buf_offset = handle->cur_offset; |
1995 | if (handle->cur_offset + count >= PAGE_SIZE) { | |
1996 | count = PAGE_SIZE - handle->cur_offset; | |
1997 | handle->cur_offset = 0; | |
1998 | handle->cur++; | |
f577eb30 | 1999 | } else { |
fb13a28b | 2000 | handle->cur_offset += count; |
f577eb30 RW |
2001 | } |
2002 | handle->offset += count; | |
2003 | return count; | |
2004 | } | |
2005 | ||
8357376d RW |
2006 | /** |
2007 | * snapshot_write_finalize - must be called after the last call to | |
2008 | * snapshot_write_next() in case the last page in the image happens | |
2009 | * to be a highmem page and its contents should be stored in the | |
2010 | * highmem. Additionally, it releases the memory that will not be | |
2011 | * used any more. | |
2012 | */ | |
2013 | ||
2014 | void snapshot_write_finalize(struct snapshot_handle *handle) | |
2015 | { | |
2016 | copy_last_highmem_page(); | |
2017 | /* Free only if we have loaded the image entirely */ | |
2018 | if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages) { | |
2019 | memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR); | |
2020 | free_highmem_data(); | |
2021 | } | |
2022 | } | |
2023 | ||
f577eb30 RW |
2024 | int snapshot_image_loaded(struct snapshot_handle *handle) |
2025 | { | |
8357376d | 2026 | return !(!nr_copy_pages || !last_highmem_page_copied() || |
940864dd RW |
2027 | handle->cur <= nr_meta_pages + nr_copy_pages); |
2028 | } | |
2029 | ||
8357376d RW |
2030 | #ifdef CONFIG_HIGHMEM |
2031 | /* Assumes that @buf is ready and points to a "safe" page */ | |
2032 | static inline void | |
2033 | swap_two_pages_data(struct page *p1, struct page *p2, void *buf) | |
940864dd | 2034 | { |
8357376d RW |
2035 | void *kaddr1, *kaddr2; |
2036 | ||
2037 | kaddr1 = kmap_atomic(p1, KM_USER0); | |
2038 | kaddr2 = kmap_atomic(p2, KM_USER1); | |
2039 | memcpy(buf, kaddr1, PAGE_SIZE); | |
2040 | memcpy(kaddr1, kaddr2, PAGE_SIZE); | |
2041 | memcpy(kaddr2, buf, PAGE_SIZE); | |
2042 | kunmap_atomic(kaddr1, KM_USER0); | |
2043 | kunmap_atomic(kaddr2, KM_USER1); | |
2044 | } | |
2045 | ||
2046 | /** | |
2047 | * restore_highmem - for each highmem page that was allocated before | |
2048 | * the suspend and included in the suspend image, and also has been | |
2049 | * allocated by the "resume" kernel swap its current (ie. "before | |
2050 | * resume") contents with the previous (ie. "before suspend") one. | |
2051 | * | |
2052 | * If the resume eventually fails, we can call this function once | |
2053 | * again and restore the "before resume" highmem state. | |
2054 | */ | |
2055 | ||
2056 | int restore_highmem(void) | |
2057 | { | |
2058 | struct highmem_pbe *pbe = highmem_pblist; | |
2059 | void *buf; | |
2060 | ||
2061 | if (!pbe) | |
2062 | return 0; | |
2063 | ||
2064 | buf = get_image_page(GFP_ATOMIC, PG_SAFE); | |
2065 | if (!buf) | |
2066 | return -ENOMEM; | |
2067 | ||
2068 | while (pbe) { | |
2069 | swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf); | |
2070 | pbe = pbe->next; | |
2071 | } | |
2072 | free_image_page(buf, PG_UNSAFE_CLEAR); | |
2073 | return 0; | |
f577eb30 | 2074 | } |
8357376d | 2075 | #endif /* CONFIG_HIGHMEM */ |