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 | 5 | * |
a2531293 | 6 | * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.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> |
5a0e3ad6 | 29 | #include <linux/slab.h> |
25761b6e RW |
30 | |
31 | #include <asm/uaccess.h> | |
32 | #include <asm/mmu_context.h> | |
33 | #include <asm/pgtable.h> | |
34 | #include <asm/tlbflush.h> | |
35 | #include <asm/io.h> | |
36 | ||
25761b6e RW |
37 | #include "power.h" |
38 | ||
74dfd666 RW |
39 | static int swsusp_page_is_free(struct page *); |
40 | static void swsusp_set_page_forbidden(struct page *); | |
41 | static void swsusp_unset_page_forbidden(struct page *); | |
42 | ||
ddeb6487 RW |
43 | /* |
44 | * Number of bytes to reserve for memory allocations made by device drivers | |
45 | * from their ->freeze() and ->freeze_noirq() callbacks so that they don't | |
46 | * cause image creation to fail (tunable via /sys/power/reserved_size). | |
47 | */ | |
48 | unsigned long reserved_size; | |
49 | ||
50 | void __init hibernate_reserved_size_init(void) | |
51 | { | |
52 | reserved_size = SPARE_PAGES * PAGE_SIZE; | |
53 | } | |
54 | ||
fe419535 RW |
55 | /* |
56 | * Preferred image size in bytes (tunable via /sys/power/image_size). | |
1c1be3a9 RW |
57 | * When it is set to N, swsusp will do its best to ensure the image |
58 | * size will not exceed N bytes, but if that is impossible, it will | |
59 | * try to create the smallest image possible. | |
fe419535 | 60 | */ |
ac5c24ec RW |
61 | unsigned long image_size; |
62 | ||
63 | void __init hibernate_image_size_init(void) | |
64 | { | |
1c1be3a9 | 65 | image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE; |
ac5c24ec | 66 | } |
fe419535 | 67 | |
8357376d RW |
68 | /* List of PBEs needed for restoring the pages that were allocated before |
69 | * the suspend and included in the suspend image, but have also been | |
70 | * allocated by the "resume" kernel, so their contents cannot be written | |
71 | * directly to their "original" page frames. | |
72 | */ | |
75534b50 RW |
73 | struct pbe *restore_pblist; |
74 | ||
8357376d | 75 | /* Pointer to an auxiliary buffer (1 page) */ |
940864dd | 76 | static void *buffer; |
7088a5c0 | 77 | |
f6143aa6 RW |
78 | /** |
79 | * @safe_needed - on resume, for storing the PBE list and the image, | |
80 | * we can only use memory pages that do not conflict with the pages | |
8357376d RW |
81 | * used before suspend. The unsafe pages have PageNosaveFree set |
82 | * and we count them using unsafe_pages. | |
f6143aa6 | 83 | * |
8357376d RW |
84 | * Each allocated image page is marked as PageNosave and PageNosaveFree |
85 | * so that swsusp_free() can release it. | |
f6143aa6 RW |
86 | */ |
87 | ||
0bcd888d RW |
88 | #define PG_ANY 0 |
89 | #define PG_SAFE 1 | |
90 | #define PG_UNSAFE_CLEAR 1 | |
91 | #define PG_UNSAFE_KEEP 0 | |
92 | ||
940864dd | 93 | static unsigned int allocated_unsafe_pages; |
f6143aa6 | 94 | |
8357376d | 95 | static void *get_image_page(gfp_t gfp_mask, int safe_needed) |
f6143aa6 RW |
96 | { |
97 | void *res; | |
98 | ||
99 | res = (void *)get_zeroed_page(gfp_mask); | |
100 | if (safe_needed) | |
7be98234 | 101 | while (res && swsusp_page_is_free(virt_to_page(res))) { |
f6143aa6 | 102 | /* The page is unsafe, mark it for swsusp_free() */ |
7be98234 | 103 | swsusp_set_page_forbidden(virt_to_page(res)); |
940864dd | 104 | allocated_unsafe_pages++; |
f6143aa6 RW |
105 | res = (void *)get_zeroed_page(gfp_mask); |
106 | } | |
107 | if (res) { | |
7be98234 RW |
108 | swsusp_set_page_forbidden(virt_to_page(res)); |
109 | swsusp_set_page_free(virt_to_page(res)); | |
f6143aa6 RW |
110 | } |
111 | return res; | |
112 | } | |
113 | ||
114 | unsigned long get_safe_page(gfp_t gfp_mask) | |
115 | { | |
8357376d RW |
116 | return (unsigned long)get_image_page(gfp_mask, PG_SAFE); |
117 | } | |
118 | ||
5b6d15de RW |
119 | static struct page *alloc_image_page(gfp_t gfp_mask) |
120 | { | |
8357376d RW |
121 | struct page *page; |
122 | ||
123 | page = alloc_page(gfp_mask); | |
124 | if (page) { | |
7be98234 RW |
125 | swsusp_set_page_forbidden(page); |
126 | swsusp_set_page_free(page); | |
8357376d RW |
127 | } |
128 | return page; | |
f6143aa6 RW |
129 | } |
130 | ||
131 | /** | |
132 | * free_image_page - free page represented by @addr, allocated with | |
8357376d | 133 | * get_image_page (page flags set by it must be cleared) |
f6143aa6 RW |
134 | */ |
135 | ||
136 | static inline void free_image_page(void *addr, int clear_nosave_free) | |
137 | { | |
8357376d RW |
138 | struct page *page; |
139 | ||
140 | BUG_ON(!virt_addr_valid(addr)); | |
141 | ||
142 | page = virt_to_page(addr); | |
143 | ||
7be98234 | 144 | swsusp_unset_page_forbidden(page); |
f6143aa6 | 145 | if (clear_nosave_free) |
7be98234 | 146 | swsusp_unset_page_free(page); |
8357376d RW |
147 | |
148 | __free_page(page); | |
f6143aa6 RW |
149 | } |
150 | ||
b788db79 RW |
151 | /* struct linked_page is used to build chains of pages */ |
152 | ||
153 | #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *)) | |
154 | ||
155 | struct linked_page { | |
156 | struct linked_page *next; | |
157 | char data[LINKED_PAGE_DATA_SIZE]; | |
158 | } __attribute__((packed)); | |
159 | ||
160 | static inline void | |
161 | free_list_of_pages(struct linked_page *list, int clear_page_nosave) | |
162 | { | |
163 | while (list) { | |
164 | struct linked_page *lp = list->next; | |
165 | ||
166 | free_image_page(list, clear_page_nosave); | |
167 | list = lp; | |
168 | } | |
169 | } | |
170 | ||
171 | /** | |
172 | * struct chain_allocator is used for allocating small objects out of | |
173 | * a linked list of pages called 'the chain'. | |
174 | * | |
175 | * The chain grows each time when there is no room for a new object in | |
176 | * the current page. The allocated objects cannot be freed individually. | |
177 | * It is only possible to free them all at once, by freeing the entire | |
178 | * chain. | |
179 | * | |
180 | * NOTE: The chain allocator may be inefficient if the allocated objects | |
181 | * are not much smaller than PAGE_SIZE. | |
182 | */ | |
183 | ||
184 | struct chain_allocator { | |
185 | struct linked_page *chain; /* the chain */ | |
186 | unsigned int used_space; /* total size of objects allocated out | |
187 | * of the current page | |
188 | */ | |
189 | gfp_t gfp_mask; /* mask for allocating pages */ | |
190 | int safe_needed; /* if set, only "safe" pages are allocated */ | |
191 | }; | |
192 | ||
193 | static void | |
194 | chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed) | |
195 | { | |
196 | ca->chain = NULL; | |
197 | ca->used_space = LINKED_PAGE_DATA_SIZE; | |
198 | ca->gfp_mask = gfp_mask; | |
199 | ca->safe_needed = safe_needed; | |
200 | } | |
201 | ||
202 | static void *chain_alloc(struct chain_allocator *ca, unsigned int size) | |
203 | { | |
204 | void *ret; | |
205 | ||
206 | if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) { | |
207 | struct linked_page *lp; | |
208 | ||
8357376d | 209 | lp = get_image_page(ca->gfp_mask, ca->safe_needed); |
b788db79 RW |
210 | if (!lp) |
211 | return NULL; | |
212 | ||
213 | lp->next = ca->chain; | |
214 | ca->chain = lp; | |
215 | ca->used_space = 0; | |
216 | } | |
217 | ret = ca->chain->data + ca->used_space; | |
218 | ca->used_space += size; | |
219 | return ret; | |
220 | } | |
221 | ||
b788db79 RW |
222 | /** |
223 | * Data types related to memory bitmaps. | |
224 | * | |
225 | * Memory bitmap is a structure consiting of many linked lists of | |
226 | * objects. The main list's elements are of type struct zone_bitmap | |
227 | * and each of them corresonds to one zone. For each zone bitmap | |
228 | * object there is a list of objects of type struct bm_block that | |
0d83304c | 229 | * represent each blocks of bitmap in which information is stored. |
b788db79 RW |
230 | * |
231 | * struct memory_bitmap contains a pointer to the main list of zone | |
232 | * bitmap objects, a struct bm_position used for browsing the bitmap, | |
233 | * and a pointer to the list of pages used for allocating all of the | |
234 | * zone bitmap objects and bitmap block objects. | |
235 | * | |
236 | * NOTE: It has to be possible to lay out the bitmap in memory | |
237 | * using only allocations of order 0. Additionally, the bitmap is | |
238 | * designed to work with arbitrary number of zones (this is over the | |
239 | * top for now, but let's avoid making unnecessary assumptions ;-). | |
240 | * | |
241 | * struct zone_bitmap contains a pointer to a list of bitmap block | |
242 | * objects and a pointer to the bitmap block object that has been | |
243 | * most recently used for setting bits. Additionally, it contains the | |
244 | * pfns that correspond to the start and end of the represented zone. | |
245 | * | |
246 | * struct bm_block contains a pointer to the memory page in which | |
0d83304c AM |
247 | * information is stored (in the form of a block of bitmap) |
248 | * It also contains the pfns that correspond to the start and end of | |
249 | * the represented memory area. | |
b788db79 RW |
250 | */ |
251 | ||
252 | #define BM_END_OF_MAP (~0UL) | |
253 | ||
8de03073 | 254 | #define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE) |
b788db79 RW |
255 | |
256 | struct bm_block { | |
846705de | 257 | struct list_head hook; /* hook into a list of bitmap blocks */ |
b788db79 RW |
258 | unsigned long start_pfn; /* pfn represented by the first bit */ |
259 | unsigned long end_pfn; /* pfn represented by the last bit plus 1 */ | |
0d83304c | 260 | unsigned long *data; /* bitmap representing pages */ |
b788db79 RW |
261 | }; |
262 | ||
0d83304c AM |
263 | static inline unsigned long bm_block_bits(struct bm_block *bb) |
264 | { | |
265 | return bb->end_pfn - bb->start_pfn; | |
266 | } | |
267 | ||
b788db79 RW |
268 | /* strcut bm_position is used for browsing memory bitmaps */ |
269 | ||
270 | struct bm_position { | |
b788db79 | 271 | struct bm_block *block; |
b788db79 RW |
272 | int bit; |
273 | }; | |
274 | ||
275 | struct memory_bitmap { | |
846705de | 276 | struct list_head blocks; /* list of bitmap blocks */ |
b788db79 RW |
277 | struct linked_page *p_list; /* list of pages used to store zone |
278 | * bitmap objects and bitmap block | |
279 | * objects | |
280 | */ | |
281 | struct bm_position cur; /* most recently used bit position */ | |
282 | }; | |
283 | ||
284 | /* Functions that operate on memory bitmaps */ | |
285 | ||
b788db79 RW |
286 | static void memory_bm_position_reset(struct memory_bitmap *bm) |
287 | { | |
846705de | 288 | bm->cur.block = list_entry(bm->blocks.next, struct bm_block, hook); |
0d83304c | 289 | bm->cur.bit = 0; |
b788db79 RW |
290 | } |
291 | ||
292 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free); | |
293 | ||
294 | /** | |
295 | * create_bm_block_list - create a list of block bitmap objects | |
8de03073 | 296 | * @pages - number of pages to track |
846705de RW |
297 | * @list - list to put the allocated blocks into |
298 | * @ca - chain allocator to be used for allocating memory | |
b788db79 | 299 | */ |
846705de RW |
300 | static int create_bm_block_list(unsigned long pages, |
301 | struct list_head *list, | |
302 | struct chain_allocator *ca) | |
b788db79 | 303 | { |
846705de | 304 | unsigned int nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK); |
b788db79 RW |
305 | |
306 | while (nr_blocks-- > 0) { | |
307 | struct bm_block *bb; | |
308 | ||
309 | bb = chain_alloc(ca, sizeof(struct bm_block)); | |
310 | if (!bb) | |
846705de RW |
311 | return -ENOMEM; |
312 | list_add(&bb->hook, list); | |
b788db79 | 313 | } |
846705de RW |
314 | |
315 | return 0; | |
b788db79 RW |
316 | } |
317 | ||
846705de RW |
318 | struct mem_extent { |
319 | struct list_head hook; | |
320 | unsigned long start; | |
321 | unsigned long end; | |
322 | }; | |
323 | ||
b788db79 | 324 | /** |
846705de RW |
325 | * free_mem_extents - free a list of memory extents |
326 | * @list - list of extents to empty | |
b788db79 | 327 | */ |
846705de RW |
328 | static void free_mem_extents(struct list_head *list) |
329 | { | |
330 | struct mem_extent *ext, *aux; | |
b788db79 | 331 | |
846705de RW |
332 | list_for_each_entry_safe(ext, aux, list, hook) { |
333 | list_del(&ext->hook); | |
334 | kfree(ext); | |
335 | } | |
336 | } | |
337 | ||
338 | /** | |
339 | * create_mem_extents - create a list of memory extents representing | |
340 | * contiguous ranges of PFNs | |
341 | * @list - list to put the extents into | |
342 | * @gfp_mask - mask to use for memory allocations | |
343 | */ | |
344 | static int create_mem_extents(struct list_head *list, gfp_t gfp_mask) | |
b788db79 | 345 | { |
846705de | 346 | struct zone *zone; |
b788db79 | 347 | |
846705de | 348 | INIT_LIST_HEAD(list); |
b788db79 | 349 | |
ee99c71c | 350 | for_each_populated_zone(zone) { |
846705de RW |
351 | unsigned long zone_start, zone_end; |
352 | struct mem_extent *ext, *cur, *aux; | |
353 | ||
846705de RW |
354 | zone_start = zone->zone_start_pfn; |
355 | zone_end = zone->zone_start_pfn + zone->spanned_pages; | |
356 | ||
357 | list_for_each_entry(ext, list, hook) | |
358 | if (zone_start <= ext->end) | |
359 | break; | |
b788db79 | 360 | |
846705de RW |
361 | if (&ext->hook == list || zone_end < ext->start) { |
362 | /* New extent is necessary */ | |
363 | struct mem_extent *new_ext; | |
364 | ||
365 | new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask); | |
366 | if (!new_ext) { | |
367 | free_mem_extents(list); | |
368 | return -ENOMEM; | |
369 | } | |
370 | new_ext->start = zone_start; | |
371 | new_ext->end = zone_end; | |
372 | list_add_tail(&new_ext->hook, &ext->hook); | |
373 | continue; | |
374 | } | |
375 | ||
376 | /* Merge this zone's range of PFNs with the existing one */ | |
377 | if (zone_start < ext->start) | |
378 | ext->start = zone_start; | |
379 | if (zone_end > ext->end) | |
380 | ext->end = zone_end; | |
381 | ||
382 | /* More merging may be possible */ | |
383 | cur = ext; | |
384 | list_for_each_entry_safe_continue(cur, aux, list, hook) { | |
385 | if (zone_end < cur->start) | |
386 | break; | |
387 | if (zone_end < cur->end) | |
388 | ext->end = cur->end; | |
389 | list_del(&cur->hook); | |
390 | kfree(cur); | |
391 | } | |
b788db79 | 392 | } |
846705de RW |
393 | |
394 | return 0; | |
b788db79 RW |
395 | } |
396 | ||
397 | /** | |
398 | * memory_bm_create - allocate memory for a memory bitmap | |
399 | */ | |
b788db79 RW |
400 | static int |
401 | memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed) | |
402 | { | |
403 | struct chain_allocator ca; | |
846705de RW |
404 | struct list_head mem_extents; |
405 | struct mem_extent *ext; | |
406 | int error; | |
b788db79 RW |
407 | |
408 | chain_init(&ca, gfp_mask, safe_needed); | |
846705de | 409 | INIT_LIST_HEAD(&bm->blocks); |
b788db79 | 410 | |
846705de RW |
411 | error = create_mem_extents(&mem_extents, gfp_mask); |
412 | if (error) | |
413 | return error; | |
b788db79 | 414 | |
846705de RW |
415 | list_for_each_entry(ext, &mem_extents, hook) { |
416 | struct bm_block *bb; | |
417 | unsigned long pfn = ext->start; | |
418 | unsigned long pages = ext->end - ext->start; | |
b788db79 | 419 | |
846705de | 420 | bb = list_entry(bm->blocks.prev, struct bm_block, hook); |
b788db79 | 421 | |
846705de RW |
422 | error = create_bm_block_list(pages, bm->blocks.prev, &ca); |
423 | if (error) | |
424 | goto Error; | |
b788db79 | 425 | |
846705de RW |
426 | list_for_each_entry_continue(bb, &bm->blocks, hook) { |
427 | bb->data = get_image_page(gfp_mask, safe_needed); | |
428 | if (!bb->data) { | |
429 | error = -ENOMEM; | |
430 | goto Error; | |
431 | } | |
b788db79 RW |
432 | |
433 | bb->start_pfn = pfn; | |
846705de | 434 | if (pages >= BM_BITS_PER_BLOCK) { |
b788db79 | 435 | pfn += BM_BITS_PER_BLOCK; |
846705de | 436 | pages -= BM_BITS_PER_BLOCK; |
b788db79 RW |
437 | } else { |
438 | /* This is executed only once in the loop */ | |
846705de | 439 | pfn += pages; |
b788db79 RW |
440 | } |
441 | bb->end_pfn = pfn; | |
b788db79 | 442 | } |
b788db79 | 443 | } |
846705de | 444 | |
b788db79 RW |
445 | bm->p_list = ca.chain; |
446 | memory_bm_position_reset(bm); | |
846705de RW |
447 | Exit: |
448 | free_mem_extents(&mem_extents); | |
449 | return error; | |
b788db79 | 450 | |
846705de | 451 | Error: |
b788db79 RW |
452 | bm->p_list = ca.chain; |
453 | memory_bm_free(bm, PG_UNSAFE_CLEAR); | |
846705de | 454 | goto Exit; |
b788db79 RW |
455 | } |
456 | ||
457 | /** | |
458 | * memory_bm_free - free memory occupied by the memory bitmap @bm | |
459 | */ | |
b788db79 RW |
460 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free) |
461 | { | |
846705de | 462 | struct bm_block *bb; |
b788db79 | 463 | |
846705de RW |
464 | list_for_each_entry(bb, &bm->blocks, hook) |
465 | if (bb->data) | |
466 | free_image_page(bb->data, clear_nosave_free); | |
b788db79 | 467 | |
b788db79 | 468 | free_list_of_pages(bm->p_list, clear_nosave_free); |
846705de RW |
469 | |
470 | INIT_LIST_HEAD(&bm->blocks); | |
b788db79 RW |
471 | } |
472 | ||
473 | /** | |
74dfd666 | 474 | * memory_bm_find_bit - find the bit in the bitmap @bm that corresponds |
b788db79 RW |
475 | * to given pfn. The cur_zone_bm member of @bm and the cur_block member |
476 | * of @bm->cur_zone_bm are updated. | |
b788db79 | 477 | */ |
a82f7119 | 478 | static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn, |
74dfd666 | 479 | void **addr, unsigned int *bit_nr) |
b788db79 | 480 | { |
b788db79 RW |
481 | struct bm_block *bb; |
482 | ||
846705de RW |
483 | /* |
484 | * Check if the pfn corresponds to the current bitmap block and find | |
485 | * the block where it fits if this is not the case. | |
486 | */ | |
487 | bb = bm->cur.block; | |
b788db79 | 488 | if (pfn < bb->start_pfn) |
846705de RW |
489 | list_for_each_entry_continue_reverse(bb, &bm->blocks, hook) |
490 | if (pfn >= bb->start_pfn) | |
491 | break; | |
b788db79 | 492 | |
846705de RW |
493 | if (pfn >= bb->end_pfn) |
494 | list_for_each_entry_continue(bb, &bm->blocks, hook) | |
495 | if (pfn >= bb->start_pfn && pfn < bb->end_pfn) | |
496 | break; | |
74dfd666 | 497 | |
846705de RW |
498 | if (&bb->hook == &bm->blocks) |
499 | return -EFAULT; | |
500 | ||
501 | /* The block has been found */ | |
502 | bm->cur.block = bb; | |
b788db79 | 503 | pfn -= bb->start_pfn; |
846705de | 504 | bm->cur.bit = pfn + 1; |
0d83304c AM |
505 | *bit_nr = pfn; |
506 | *addr = bb->data; | |
a82f7119 | 507 | return 0; |
74dfd666 RW |
508 | } |
509 | ||
510 | static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn) | |
511 | { | |
512 | void *addr; | |
513 | unsigned int bit; | |
a82f7119 | 514 | int error; |
74dfd666 | 515 | |
a82f7119 RW |
516 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
517 | BUG_ON(error); | |
74dfd666 RW |
518 | set_bit(bit, addr); |
519 | } | |
520 | ||
a82f7119 RW |
521 | static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn) |
522 | { | |
523 | void *addr; | |
524 | unsigned int bit; | |
525 | int error; | |
526 | ||
527 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); | |
528 | if (!error) | |
529 | set_bit(bit, addr); | |
530 | return error; | |
531 | } | |
532 | ||
74dfd666 RW |
533 | static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn) |
534 | { | |
535 | void *addr; | |
536 | unsigned int bit; | |
a82f7119 | 537 | int error; |
74dfd666 | 538 | |
a82f7119 RW |
539 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
540 | BUG_ON(error); | |
74dfd666 RW |
541 | clear_bit(bit, addr); |
542 | } | |
543 | ||
544 | static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn) | |
545 | { | |
546 | void *addr; | |
547 | unsigned int bit; | |
a82f7119 | 548 | int error; |
74dfd666 | 549 | |
a82f7119 RW |
550 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
551 | BUG_ON(error); | |
74dfd666 | 552 | return test_bit(bit, addr); |
b788db79 RW |
553 | } |
554 | ||
69643279 RW |
555 | static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn) |
556 | { | |
557 | void *addr; | |
558 | unsigned int bit; | |
559 | ||
560 | return !memory_bm_find_bit(bm, pfn, &addr, &bit); | |
561 | } | |
562 | ||
b788db79 RW |
563 | /** |
564 | * memory_bm_next_pfn - find the pfn that corresponds to the next set bit | |
565 | * in the bitmap @bm. If the pfn cannot be found, BM_END_OF_MAP is | |
566 | * returned. | |
567 | * | |
568 | * It is required to run memory_bm_position_reset() before the first call to | |
569 | * this function. | |
570 | */ | |
571 | ||
572 | static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm) | |
573 | { | |
b788db79 | 574 | struct bm_block *bb; |
b788db79 RW |
575 | int bit; |
576 | ||
846705de | 577 | bb = bm->cur.block; |
b788db79 | 578 | do { |
846705de RW |
579 | bit = bm->cur.bit; |
580 | bit = find_next_bit(bb->data, bm_block_bits(bb), bit); | |
581 | if (bit < bm_block_bits(bb)) | |
582 | goto Return_pfn; | |
583 | ||
584 | bb = list_entry(bb->hook.next, struct bm_block, hook); | |
585 | bm->cur.block = bb; | |
586 | bm->cur.bit = 0; | |
587 | } while (&bb->hook != &bm->blocks); | |
588 | ||
b788db79 RW |
589 | memory_bm_position_reset(bm); |
590 | return BM_END_OF_MAP; | |
591 | ||
59a49335 | 592 | Return_pfn: |
0d83304c AM |
593 | bm->cur.bit = bit + 1; |
594 | return bb->start_pfn + bit; | |
b788db79 RW |
595 | } |
596 | ||
74dfd666 RW |
597 | /** |
598 | * This structure represents a range of page frames the contents of which | |
599 | * should not be saved during the suspend. | |
600 | */ | |
601 | ||
602 | struct nosave_region { | |
603 | struct list_head list; | |
604 | unsigned long start_pfn; | |
605 | unsigned long end_pfn; | |
606 | }; | |
607 | ||
608 | static LIST_HEAD(nosave_regions); | |
609 | ||
610 | /** | |
611 | * register_nosave_region - register a range of page frames the contents | |
612 | * of which should not be saved during the suspend (to be used in the early | |
613 | * initialization code) | |
614 | */ | |
615 | ||
616 | void __init | |
940d67f6 JB |
617 | __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn, |
618 | int use_kmalloc) | |
74dfd666 RW |
619 | { |
620 | struct nosave_region *region; | |
621 | ||
622 | if (start_pfn >= end_pfn) | |
623 | return; | |
624 | ||
625 | if (!list_empty(&nosave_regions)) { | |
626 | /* Try to extend the previous region (they should be sorted) */ | |
627 | region = list_entry(nosave_regions.prev, | |
628 | struct nosave_region, list); | |
629 | if (region->end_pfn == start_pfn) { | |
630 | region->end_pfn = end_pfn; | |
631 | goto Report; | |
632 | } | |
633 | } | |
940d67f6 JB |
634 | if (use_kmalloc) { |
635 | /* during init, this shouldn't fail */ | |
636 | region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL); | |
637 | BUG_ON(!region); | |
638 | } else | |
639 | /* This allocation cannot fail */ | |
3c1596ef | 640 | region = alloc_bootmem(sizeof(struct nosave_region)); |
74dfd666 RW |
641 | region->start_pfn = start_pfn; |
642 | region->end_pfn = end_pfn; | |
643 | list_add_tail(®ion->list, &nosave_regions); | |
644 | Report: | |
cd38ca85 BH |
645 | printk(KERN_INFO "PM: Registered nosave memory: [mem %#010llx-%#010llx]\n", |
646 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
647 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
74dfd666 RW |
648 | } |
649 | ||
650 | /* | |
651 | * Set bits in this map correspond to the page frames the contents of which | |
652 | * should not be saved during the suspend. | |
653 | */ | |
654 | static struct memory_bitmap *forbidden_pages_map; | |
655 | ||
656 | /* Set bits in this map correspond to free page frames. */ | |
657 | static struct memory_bitmap *free_pages_map; | |
658 | ||
659 | /* | |
660 | * Each page frame allocated for creating the image is marked by setting the | |
661 | * corresponding bits in forbidden_pages_map and free_pages_map simultaneously | |
662 | */ | |
663 | ||
664 | void swsusp_set_page_free(struct page *page) | |
665 | { | |
666 | if (free_pages_map) | |
667 | memory_bm_set_bit(free_pages_map, page_to_pfn(page)); | |
668 | } | |
669 | ||
670 | static int swsusp_page_is_free(struct page *page) | |
671 | { | |
672 | return free_pages_map ? | |
673 | memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0; | |
674 | } | |
675 | ||
676 | void swsusp_unset_page_free(struct page *page) | |
677 | { | |
678 | if (free_pages_map) | |
679 | memory_bm_clear_bit(free_pages_map, page_to_pfn(page)); | |
680 | } | |
681 | ||
682 | static void swsusp_set_page_forbidden(struct page *page) | |
683 | { | |
684 | if (forbidden_pages_map) | |
685 | memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page)); | |
686 | } | |
687 | ||
688 | int swsusp_page_is_forbidden(struct page *page) | |
689 | { | |
690 | return forbidden_pages_map ? | |
691 | memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0; | |
692 | } | |
693 | ||
694 | static void swsusp_unset_page_forbidden(struct page *page) | |
695 | { | |
696 | if (forbidden_pages_map) | |
697 | memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page)); | |
698 | } | |
699 | ||
700 | /** | |
701 | * mark_nosave_pages - set bits corresponding to the page frames the | |
702 | * contents of which should not be saved in a given bitmap. | |
703 | */ | |
704 | ||
705 | static void mark_nosave_pages(struct memory_bitmap *bm) | |
706 | { | |
707 | struct nosave_region *region; | |
708 | ||
709 | if (list_empty(&nosave_regions)) | |
710 | return; | |
711 | ||
712 | list_for_each_entry(region, &nosave_regions, list) { | |
713 | unsigned long pfn; | |
714 | ||
69f1d475 BH |
715 | pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n", |
716 | (unsigned long long) region->start_pfn << PAGE_SHIFT, | |
717 | ((unsigned long long) region->end_pfn << PAGE_SHIFT) | |
718 | - 1); | |
74dfd666 RW |
719 | |
720 | for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++) | |
a82f7119 RW |
721 | if (pfn_valid(pfn)) { |
722 | /* | |
723 | * It is safe to ignore the result of | |
724 | * mem_bm_set_bit_check() here, since we won't | |
725 | * touch the PFNs for which the error is | |
726 | * returned anyway. | |
727 | */ | |
728 | mem_bm_set_bit_check(bm, pfn); | |
729 | } | |
74dfd666 RW |
730 | } |
731 | } | |
732 | ||
733 | /** | |
734 | * create_basic_memory_bitmaps - create bitmaps needed for marking page | |
735 | * frames that should not be saved and free page frames. The pointers | |
736 | * forbidden_pages_map and free_pages_map are only modified if everything | |
737 | * goes well, because we don't want the bits to be used before both bitmaps | |
738 | * are set up. | |
739 | */ | |
740 | ||
741 | int create_basic_memory_bitmaps(void) | |
742 | { | |
743 | struct memory_bitmap *bm1, *bm2; | |
744 | int error = 0; | |
745 | ||
746 | BUG_ON(forbidden_pages_map || free_pages_map); | |
747 | ||
0709db60 | 748 | bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
749 | if (!bm1) |
750 | return -ENOMEM; | |
751 | ||
0709db60 | 752 | error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
753 | if (error) |
754 | goto Free_first_object; | |
755 | ||
0709db60 | 756 | bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
757 | if (!bm2) |
758 | goto Free_first_bitmap; | |
759 | ||
0709db60 | 760 | error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
761 | if (error) |
762 | goto Free_second_object; | |
763 | ||
764 | forbidden_pages_map = bm1; | |
765 | free_pages_map = bm2; | |
766 | mark_nosave_pages(forbidden_pages_map); | |
767 | ||
23976728 | 768 | pr_debug("PM: Basic memory bitmaps created\n"); |
74dfd666 RW |
769 | |
770 | return 0; | |
771 | ||
772 | Free_second_object: | |
773 | kfree(bm2); | |
774 | Free_first_bitmap: | |
775 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
776 | Free_first_object: | |
777 | kfree(bm1); | |
778 | return -ENOMEM; | |
779 | } | |
780 | ||
781 | /** | |
782 | * free_basic_memory_bitmaps - free memory bitmaps allocated by | |
783 | * create_basic_memory_bitmaps(). The auxiliary pointers are necessary | |
784 | * so that the bitmaps themselves are not referred to while they are being | |
785 | * freed. | |
786 | */ | |
787 | ||
788 | void free_basic_memory_bitmaps(void) | |
789 | { | |
790 | struct memory_bitmap *bm1, *bm2; | |
791 | ||
792 | BUG_ON(!(forbidden_pages_map && free_pages_map)); | |
793 | ||
794 | bm1 = forbidden_pages_map; | |
795 | bm2 = free_pages_map; | |
796 | forbidden_pages_map = NULL; | |
797 | free_pages_map = NULL; | |
798 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
799 | kfree(bm1); | |
800 | memory_bm_free(bm2, PG_UNSAFE_CLEAR); | |
801 | kfree(bm2); | |
802 | ||
23976728 | 803 | pr_debug("PM: Basic memory bitmaps freed\n"); |
74dfd666 RW |
804 | } |
805 | ||
b788db79 RW |
806 | /** |
807 | * snapshot_additional_pages - estimate the number of additional pages | |
808 | * be needed for setting up the suspend image data structures for given | |
809 | * zone (usually the returned value is greater than the exact number) | |
810 | */ | |
811 | ||
812 | unsigned int snapshot_additional_pages(struct zone *zone) | |
813 | { | |
814 | unsigned int res; | |
815 | ||
816 | res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK); | |
160cb5a9 NK |
817 | res += DIV_ROUND_UP(res * sizeof(struct bm_block), |
818 | LINKED_PAGE_DATA_SIZE); | |
8357376d | 819 | return 2 * res; |
b788db79 RW |
820 | } |
821 | ||
8357376d RW |
822 | #ifdef CONFIG_HIGHMEM |
823 | /** | |
824 | * count_free_highmem_pages - compute the total number of free highmem | |
825 | * pages, system-wide. | |
826 | */ | |
827 | ||
828 | static unsigned int count_free_highmem_pages(void) | |
829 | { | |
830 | struct zone *zone; | |
831 | unsigned int cnt = 0; | |
832 | ||
ee99c71c KM |
833 | for_each_populated_zone(zone) |
834 | if (is_highmem(zone)) | |
d23ad423 | 835 | cnt += zone_page_state(zone, NR_FREE_PAGES); |
8357376d RW |
836 | |
837 | return cnt; | |
838 | } | |
839 | ||
840 | /** | |
841 | * saveable_highmem_page - Determine whether a highmem page should be | |
842 | * included in the suspend image. | |
843 | * | |
844 | * We should save the page if it isn't Nosave or NosaveFree, or Reserved, | |
845 | * and it isn't a part of a free chunk of pages. | |
846 | */ | |
846705de | 847 | static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn) |
8357376d RW |
848 | { |
849 | struct page *page; | |
850 | ||
851 | if (!pfn_valid(pfn)) | |
852 | return NULL; | |
853 | ||
854 | page = pfn_to_page(pfn); | |
846705de RW |
855 | if (page_zone(page) != zone) |
856 | return NULL; | |
8357376d RW |
857 | |
858 | BUG_ON(!PageHighMem(page)); | |
859 | ||
7be98234 RW |
860 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) || |
861 | PageReserved(page)) | |
8357376d RW |
862 | return NULL; |
863 | ||
c6968e73 SG |
864 | if (page_is_guard(page)) |
865 | return NULL; | |
866 | ||
8357376d RW |
867 | return page; |
868 | } | |
869 | ||
870 | /** | |
871 | * count_highmem_pages - compute the total number of saveable highmem | |
872 | * pages. | |
873 | */ | |
874 | ||
fe419535 | 875 | static unsigned int count_highmem_pages(void) |
8357376d RW |
876 | { |
877 | struct zone *zone; | |
878 | unsigned int n = 0; | |
879 | ||
98e73dc5 | 880 | for_each_populated_zone(zone) { |
8357376d RW |
881 | unsigned long pfn, max_zone_pfn; |
882 | ||
883 | if (!is_highmem(zone)) | |
884 | continue; | |
885 | ||
886 | mark_free_pages(zone); | |
887 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
888 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
846705de | 889 | if (saveable_highmem_page(zone, pfn)) |
8357376d RW |
890 | n++; |
891 | } | |
892 | return n; | |
893 | } | |
894 | #else | |
846705de RW |
895 | static inline void *saveable_highmem_page(struct zone *z, unsigned long p) |
896 | { | |
897 | return NULL; | |
898 | } | |
8357376d RW |
899 | #endif /* CONFIG_HIGHMEM */ |
900 | ||
25761b6e | 901 | /** |
8a235efa RW |
902 | * saveable_page - Determine whether a non-highmem page should be included |
903 | * in the suspend image. | |
25761b6e | 904 | * |
8357376d RW |
905 | * We should save the page if it isn't Nosave, and is not in the range |
906 | * of pages statically defined as 'unsaveable', and it isn't a part of | |
907 | * a free chunk of pages. | |
25761b6e | 908 | */ |
846705de | 909 | static struct page *saveable_page(struct zone *zone, unsigned long pfn) |
25761b6e | 910 | { |
de491861 | 911 | struct page *page; |
25761b6e RW |
912 | |
913 | if (!pfn_valid(pfn)) | |
ae83c5ee | 914 | return NULL; |
25761b6e RW |
915 | |
916 | page = pfn_to_page(pfn); | |
846705de RW |
917 | if (page_zone(page) != zone) |
918 | return NULL; | |
ae83c5ee | 919 | |
8357376d RW |
920 | BUG_ON(PageHighMem(page)); |
921 | ||
7be98234 | 922 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page)) |
ae83c5ee | 923 | return NULL; |
8357376d | 924 | |
8a235efa RW |
925 | if (PageReserved(page) |
926 | && (!kernel_page_present(page) || pfn_is_nosave(pfn))) | |
ae83c5ee | 927 | return NULL; |
25761b6e | 928 | |
c6968e73 SG |
929 | if (page_is_guard(page)) |
930 | return NULL; | |
931 | ||
ae83c5ee | 932 | return page; |
25761b6e RW |
933 | } |
934 | ||
8357376d RW |
935 | /** |
936 | * count_data_pages - compute the total number of saveable non-highmem | |
937 | * pages. | |
938 | */ | |
939 | ||
fe419535 | 940 | static unsigned int count_data_pages(void) |
25761b6e RW |
941 | { |
942 | struct zone *zone; | |
ae83c5ee | 943 | unsigned long pfn, max_zone_pfn; |
dc19d507 | 944 | unsigned int n = 0; |
25761b6e | 945 | |
98e73dc5 | 946 | for_each_populated_zone(zone) { |
25761b6e RW |
947 | if (is_highmem(zone)) |
948 | continue; | |
8357376d | 949 | |
25761b6e | 950 | mark_free_pages(zone); |
ae83c5ee RW |
951 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
952 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
846705de | 953 | if (saveable_page(zone, pfn)) |
8357376d | 954 | n++; |
25761b6e | 955 | } |
a0f49651 | 956 | return n; |
25761b6e RW |
957 | } |
958 | ||
8357376d RW |
959 | /* This is needed, because copy_page and memcpy are not usable for copying |
960 | * task structs. | |
961 | */ | |
962 | static inline void do_copy_page(long *dst, long *src) | |
f623f0db RW |
963 | { |
964 | int n; | |
965 | ||
f623f0db RW |
966 | for (n = PAGE_SIZE / sizeof(long); n; n--) |
967 | *dst++ = *src++; | |
968 | } | |
969 | ||
8a235efa RW |
970 | |
971 | /** | |
972 | * safe_copy_page - check if the page we are going to copy is marked as | |
973 | * present in the kernel page tables (this always is the case if | |
974 | * CONFIG_DEBUG_PAGEALLOC is not set and in that case | |
975 | * kernel_page_present() always returns 'true'). | |
976 | */ | |
977 | static void safe_copy_page(void *dst, struct page *s_page) | |
978 | { | |
979 | if (kernel_page_present(s_page)) { | |
980 | do_copy_page(dst, page_address(s_page)); | |
981 | } else { | |
982 | kernel_map_pages(s_page, 1, 1); | |
983 | do_copy_page(dst, page_address(s_page)); | |
984 | kernel_map_pages(s_page, 1, 0); | |
985 | } | |
986 | } | |
987 | ||
988 | ||
8357376d RW |
989 | #ifdef CONFIG_HIGHMEM |
990 | static inline struct page * | |
991 | page_is_saveable(struct zone *zone, unsigned long pfn) | |
992 | { | |
993 | return is_highmem(zone) ? | |
846705de | 994 | saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn); |
8357376d RW |
995 | } |
996 | ||
8a235efa | 997 | static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
8357376d RW |
998 | { |
999 | struct page *s_page, *d_page; | |
1000 | void *src, *dst; | |
1001 | ||
1002 | s_page = pfn_to_page(src_pfn); | |
1003 | d_page = pfn_to_page(dst_pfn); | |
1004 | if (PageHighMem(s_page)) { | |
0de9a1e2 CW |
1005 | src = kmap_atomic(s_page); |
1006 | dst = kmap_atomic(d_page); | |
8357376d | 1007 | do_copy_page(dst, src); |
0de9a1e2 CW |
1008 | kunmap_atomic(dst); |
1009 | kunmap_atomic(src); | |
8357376d | 1010 | } else { |
8357376d RW |
1011 | if (PageHighMem(d_page)) { |
1012 | /* Page pointed to by src may contain some kernel | |
1013 | * data modified by kmap_atomic() | |
1014 | */ | |
8a235efa | 1015 | safe_copy_page(buffer, s_page); |
0de9a1e2 | 1016 | dst = kmap_atomic(d_page); |
3ecb01df | 1017 | copy_page(dst, buffer); |
0de9a1e2 | 1018 | kunmap_atomic(dst); |
8357376d | 1019 | } else { |
8a235efa | 1020 | safe_copy_page(page_address(d_page), s_page); |
8357376d RW |
1021 | } |
1022 | } | |
1023 | } | |
1024 | #else | |
846705de | 1025 | #define page_is_saveable(zone, pfn) saveable_page(zone, pfn) |
8357376d | 1026 | |
8a235efa | 1027 | static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
8357376d | 1028 | { |
8a235efa RW |
1029 | safe_copy_page(page_address(pfn_to_page(dst_pfn)), |
1030 | pfn_to_page(src_pfn)); | |
8357376d RW |
1031 | } |
1032 | #endif /* CONFIG_HIGHMEM */ | |
1033 | ||
b788db79 RW |
1034 | static void |
1035 | copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm) | |
25761b6e RW |
1036 | { |
1037 | struct zone *zone; | |
b788db79 | 1038 | unsigned long pfn; |
25761b6e | 1039 | |
98e73dc5 | 1040 | for_each_populated_zone(zone) { |
b788db79 RW |
1041 | unsigned long max_zone_pfn; |
1042 | ||
25761b6e | 1043 | mark_free_pages(zone); |
ae83c5ee | 1044 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
b788db79 | 1045 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
8357376d | 1046 | if (page_is_saveable(zone, pfn)) |
b788db79 | 1047 | memory_bm_set_bit(orig_bm, pfn); |
25761b6e | 1048 | } |
b788db79 RW |
1049 | memory_bm_position_reset(orig_bm); |
1050 | memory_bm_position_reset(copy_bm); | |
df7c4872 | 1051 | for(;;) { |
b788db79 | 1052 | pfn = memory_bm_next_pfn(orig_bm); |
df7c4872 FW |
1053 | if (unlikely(pfn == BM_END_OF_MAP)) |
1054 | break; | |
1055 | copy_data_page(memory_bm_next_pfn(copy_bm), pfn); | |
1056 | } | |
25761b6e RW |
1057 | } |
1058 | ||
8357376d RW |
1059 | /* Total number of image pages */ |
1060 | static unsigned int nr_copy_pages; | |
1061 | /* Number of pages needed for saving the original pfns of the image pages */ | |
1062 | static unsigned int nr_meta_pages; | |
64a473cb RW |
1063 | /* |
1064 | * Numbers of normal and highmem page frames allocated for hibernation image | |
1065 | * before suspending devices. | |
1066 | */ | |
1067 | unsigned int alloc_normal, alloc_highmem; | |
1068 | /* | |
1069 | * Memory bitmap used for marking saveable pages (during hibernation) or | |
1070 | * hibernation image pages (during restore) | |
1071 | */ | |
1072 | static struct memory_bitmap orig_bm; | |
1073 | /* | |
1074 | * Memory bitmap used during hibernation for marking allocated page frames that | |
1075 | * will contain copies of saveable pages. During restore it is initially used | |
1076 | * for marking hibernation image pages, but then the set bits from it are | |
1077 | * duplicated in @orig_bm and it is released. On highmem systems it is next | |
1078 | * used for marking "safe" highmem pages, but it has to be reinitialized for | |
1079 | * this purpose. | |
1080 | */ | |
1081 | static struct memory_bitmap copy_bm; | |
8357376d | 1082 | |
25761b6e | 1083 | /** |
940864dd | 1084 | * swsusp_free - free pages allocated for the suspend. |
cd560bb2 | 1085 | * |
940864dd RW |
1086 | * Suspend pages are alocated before the atomic copy is made, so we |
1087 | * need to release them after the resume. | |
25761b6e RW |
1088 | */ |
1089 | ||
1090 | void swsusp_free(void) | |
1091 | { | |
1092 | struct zone *zone; | |
ae83c5ee | 1093 | unsigned long pfn, max_zone_pfn; |
25761b6e | 1094 | |
98e73dc5 | 1095 | for_each_populated_zone(zone) { |
ae83c5ee RW |
1096 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
1097 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1098 | if (pfn_valid(pfn)) { | |
1099 | struct page *page = pfn_to_page(pfn); | |
1100 | ||
7be98234 RW |
1101 | if (swsusp_page_is_forbidden(page) && |
1102 | swsusp_page_is_free(page)) { | |
1103 | swsusp_unset_page_forbidden(page); | |
1104 | swsusp_unset_page_free(page); | |
8357376d | 1105 | __free_page(page); |
25761b6e RW |
1106 | } |
1107 | } | |
1108 | } | |
f577eb30 RW |
1109 | nr_copy_pages = 0; |
1110 | nr_meta_pages = 0; | |
75534b50 | 1111 | restore_pblist = NULL; |
6e1819d6 | 1112 | buffer = NULL; |
64a473cb RW |
1113 | alloc_normal = 0; |
1114 | alloc_highmem = 0; | |
25761b6e RW |
1115 | } |
1116 | ||
4bb33435 RW |
1117 | /* Helper functions used for the shrinking of memory. */ |
1118 | ||
1119 | #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN) | |
1120 | ||
fe419535 | 1121 | /** |
4bb33435 RW |
1122 | * preallocate_image_pages - Allocate a number of pages for hibernation image |
1123 | * @nr_pages: Number of page frames to allocate. | |
1124 | * @mask: GFP flags to use for the allocation. | |
fe419535 | 1125 | * |
4bb33435 RW |
1126 | * Return value: Number of page frames actually allocated |
1127 | */ | |
1128 | static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask) | |
1129 | { | |
1130 | unsigned long nr_alloc = 0; | |
1131 | ||
1132 | while (nr_pages > 0) { | |
64a473cb RW |
1133 | struct page *page; |
1134 | ||
1135 | page = alloc_image_page(mask); | |
1136 | if (!page) | |
4bb33435 | 1137 | break; |
64a473cb RW |
1138 | memory_bm_set_bit(©_bm, page_to_pfn(page)); |
1139 | if (PageHighMem(page)) | |
1140 | alloc_highmem++; | |
1141 | else | |
1142 | alloc_normal++; | |
4bb33435 RW |
1143 | nr_pages--; |
1144 | nr_alloc++; | |
1145 | } | |
1146 | ||
1147 | return nr_alloc; | |
1148 | } | |
1149 | ||
6715045d RW |
1150 | static unsigned long preallocate_image_memory(unsigned long nr_pages, |
1151 | unsigned long avail_normal) | |
4bb33435 | 1152 | { |
6715045d RW |
1153 | unsigned long alloc; |
1154 | ||
1155 | if (avail_normal <= alloc_normal) | |
1156 | return 0; | |
1157 | ||
1158 | alloc = avail_normal - alloc_normal; | |
1159 | if (nr_pages < alloc) | |
1160 | alloc = nr_pages; | |
1161 | ||
1162 | return preallocate_image_pages(alloc, GFP_IMAGE); | |
4bb33435 RW |
1163 | } |
1164 | ||
1165 | #ifdef CONFIG_HIGHMEM | |
1166 | static unsigned long preallocate_image_highmem(unsigned long nr_pages) | |
1167 | { | |
1168 | return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM); | |
1169 | } | |
1170 | ||
1171 | /** | |
1172 | * __fraction - Compute (an approximation of) x * (multiplier / base) | |
fe419535 | 1173 | */ |
4bb33435 RW |
1174 | static unsigned long __fraction(u64 x, u64 multiplier, u64 base) |
1175 | { | |
1176 | x *= multiplier; | |
1177 | do_div(x, base); | |
1178 | return (unsigned long)x; | |
1179 | } | |
fe419535 | 1180 | |
4bb33435 RW |
1181 | static unsigned long preallocate_highmem_fraction(unsigned long nr_pages, |
1182 | unsigned long highmem, | |
1183 | unsigned long total) | |
fe419535 | 1184 | { |
4bb33435 RW |
1185 | unsigned long alloc = __fraction(nr_pages, highmem, total); |
1186 | ||
1187 | return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM); | |
fe419535 | 1188 | } |
4bb33435 RW |
1189 | #else /* CONFIG_HIGHMEM */ |
1190 | static inline unsigned long preallocate_image_highmem(unsigned long nr_pages) | |
1191 | { | |
1192 | return 0; | |
1193 | } | |
1194 | ||
1195 | static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages, | |
1196 | unsigned long highmem, | |
1197 | unsigned long total) | |
1198 | { | |
1199 | return 0; | |
1200 | } | |
1201 | #endif /* CONFIG_HIGHMEM */ | |
fe419535 | 1202 | |
4bb33435 | 1203 | /** |
64a473cb RW |
1204 | * free_unnecessary_pages - Release preallocated pages not needed for the image |
1205 | */ | |
1206 | static void free_unnecessary_pages(void) | |
1207 | { | |
6715045d | 1208 | unsigned long save, to_free_normal, to_free_highmem; |
64a473cb | 1209 | |
6715045d RW |
1210 | save = count_data_pages(); |
1211 | if (alloc_normal >= save) { | |
1212 | to_free_normal = alloc_normal - save; | |
1213 | save = 0; | |
1214 | } else { | |
1215 | to_free_normal = 0; | |
1216 | save -= alloc_normal; | |
1217 | } | |
1218 | save += count_highmem_pages(); | |
1219 | if (alloc_highmem >= save) { | |
1220 | to_free_highmem = alloc_highmem - save; | |
64a473cb RW |
1221 | } else { |
1222 | to_free_highmem = 0; | |
4d4cf23c RW |
1223 | save -= alloc_highmem; |
1224 | if (to_free_normal > save) | |
1225 | to_free_normal -= save; | |
1226 | else | |
1227 | to_free_normal = 0; | |
64a473cb RW |
1228 | } |
1229 | ||
1230 | memory_bm_position_reset(©_bm); | |
1231 | ||
a9c9b442 | 1232 | while (to_free_normal > 0 || to_free_highmem > 0) { |
64a473cb RW |
1233 | unsigned long pfn = memory_bm_next_pfn(©_bm); |
1234 | struct page *page = pfn_to_page(pfn); | |
1235 | ||
1236 | if (PageHighMem(page)) { | |
1237 | if (!to_free_highmem) | |
1238 | continue; | |
1239 | to_free_highmem--; | |
1240 | alloc_highmem--; | |
1241 | } else { | |
1242 | if (!to_free_normal) | |
1243 | continue; | |
1244 | to_free_normal--; | |
1245 | alloc_normal--; | |
1246 | } | |
1247 | memory_bm_clear_bit(©_bm, pfn); | |
1248 | swsusp_unset_page_forbidden(page); | |
1249 | swsusp_unset_page_free(page); | |
1250 | __free_page(page); | |
1251 | } | |
1252 | } | |
1253 | ||
ef4aede3 RW |
1254 | /** |
1255 | * minimum_image_size - Estimate the minimum acceptable size of an image | |
1256 | * @saveable: Number of saveable pages in the system. | |
1257 | * | |
1258 | * We want to avoid attempting to free too much memory too hard, so estimate the | |
1259 | * minimum acceptable size of a hibernation image to use as the lower limit for | |
1260 | * preallocating memory. | |
1261 | * | |
1262 | * We assume that the minimum image size should be proportional to | |
1263 | * | |
1264 | * [number of saveable pages] - [number of pages that can be freed in theory] | |
1265 | * | |
1266 | * where the second term is the sum of (1) reclaimable slab pages, (2) active | |
1267 | * and (3) inactive anonymouns pages, (4) active and (5) inactive file pages, | |
1268 | * minus mapped file pages. | |
1269 | */ | |
1270 | static unsigned long minimum_image_size(unsigned long saveable) | |
1271 | { | |
1272 | unsigned long size; | |
1273 | ||
1274 | size = global_page_state(NR_SLAB_RECLAIMABLE) | |
1275 | + global_page_state(NR_ACTIVE_ANON) | |
1276 | + global_page_state(NR_INACTIVE_ANON) | |
1277 | + global_page_state(NR_ACTIVE_FILE) | |
1278 | + global_page_state(NR_INACTIVE_FILE) | |
1279 | - global_page_state(NR_FILE_MAPPED); | |
1280 | ||
1281 | return saveable <= size ? 0 : saveable - size; | |
1282 | } | |
1283 | ||
64a473cb RW |
1284 | /** |
1285 | * hibernate_preallocate_memory - Preallocate memory for hibernation image | |
4bb33435 RW |
1286 | * |
1287 | * To create a hibernation image it is necessary to make a copy of every page | |
1288 | * frame in use. We also need a number of page frames to be free during | |
1289 | * hibernation for allocations made while saving the image and for device | |
1290 | * drivers, in case they need to allocate memory from their hibernation | |
ddeb6487 RW |
1291 | * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough |
1292 | * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through | |
1293 | * /sys/power/reserved_size, respectively). To make this happen, we compute the | |
1294 | * total number of available page frames and allocate at least | |
4bb33435 | 1295 | * |
ddeb6487 RW |
1296 | * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2 |
1297 | * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE) | |
4bb33435 RW |
1298 | * |
1299 | * of them, which corresponds to the maximum size of a hibernation image. | |
1300 | * | |
1301 | * If image_size is set below the number following from the above formula, | |
1302 | * the preallocation of memory is continued until the total number of saveable | |
ef4aede3 RW |
1303 | * pages in the system is below the requested image size or the minimum |
1304 | * acceptable image size returned by minimum_image_size(), whichever is greater. | |
4bb33435 | 1305 | */ |
64a473cb | 1306 | int hibernate_preallocate_memory(void) |
fe419535 | 1307 | { |
fe419535 | 1308 | struct zone *zone; |
4bb33435 | 1309 | unsigned long saveable, size, max_size, count, highmem, pages = 0; |
6715045d | 1310 | unsigned long alloc, save_highmem, pages_highmem, avail_normal; |
fe419535 | 1311 | struct timeval start, stop; |
64a473cb | 1312 | int error; |
fe419535 | 1313 | |
64a473cb | 1314 | printk(KERN_INFO "PM: Preallocating image memory... "); |
fe419535 | 1315 | do_gettimeofday(&start); |
fe419535 | 1316 | |
64a473cb RW |
1317 | error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY); |
1318 | if (error) | |
1319 | goto err_out; | |
1320 | ||
1321 | error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY); | |
1322 | if (error) | |
1323 | goto err_out; | |
1324 | ||
1325 | alloc_normal = 0; | |
1326 | alloc_highmem = 0; | |
1327 | ||
4bb33435 | 1328 | /* Count the number of saveable data pages. */ |
64a473cb | 1329 | save_highmem = count_highmem_pages(); |
4bb33435 | 1330 | saveable = count_data_pages(); |
fe419535 | 1331 | |
4bb33435 RW |
1332 | /* |
1333 | * Compute the total number of page frames we can use (count) and the | |
1334 | * number of pages needed for image metadata (size). | |
1335 | */ | |
1336 | count = saveable; | |
64a473cb RW |
1337 | saveable += save_highmem; |
1338 | highmem = save_highmem; | |
4bb33435 RW |
1339 | size = 0; |
1340 | for_each_populated_zone(zone) { | |
1341 | size += snapshot_additional_pages(zone); | |
1342 | if (is_highmem(zone)) | |
1343 | highmem += zone_page_state(zone, NR_FREE_PAGES); | |
1344 | else | |
1345 | count += zone_page_state(zone, NR_FREE_PAGES); | |
1346 | } | |
6715045d | 1347 | avail_normal = count; |
4bb33435 RW |
1348 | count += highmem; |
1349 | count -= totalreserve_pages; | |
1350 | ||
85055dd8 MS |
1351 | /* Add number of pages required for page keys (s390 only). */ |
1352 | size += page_key_additional_pages(saveable); | |
1353 | ||
4bb33435 | 1354 | /* Compute the maximum number of saveable pages to leave in memory. */ |
ddeb6487 RW |
1355 | max_size = (count - (size + PAGES_FOR_IO)) / 2 |
1356 | - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE); | |
266f1a25 | 1357 | /* Compute the desired number of image pages specified by image_size. */ |
4bb33435 RW |
1358 | size = DIV_ROUND_UP(image_size, PAGE_SIZE); |
1359 | if (size > max_size) | |
1360 | size = max_size; | |
1361 | /* | |
266f1a25 RW |
1362 | * If the desired number of image pages is at least as large as the |
1363 | * current number of saveable pages in memory, allocate page frames for | |
1364 | * the image and we're done. | |
4bb33435 | 1365 | */ |
64a473cb RW |
1366 | if (size >= saveable) { |
1367 | pages = preallocate_image_highmem(save_highmem); | |
6715045d | 1368 | pages += preallocate_image_memory(saveable - pages, avail_normal); |
4bb33435 | 1369 | goto out; |
64a473cb | 1370 | } |
4bb33435 | 1371 | |
ef4aede3 RW |
1372 | /* Estimate the minimum size of the image. */ |
1373 | pages = minimum_image_size(saveable); | |
6715045d RW |
1374 | /* |
1375 | * To avoid excessive pressure on the normal zone, leave room in it to | |
1376 | * accommodate an image of the minimum size (unless it's already too | |
1377 | * small, in which case don't preallocate pages from it at all). | |
1378 | */ | |
1379 | if (avail_normal > pages) | |
1380 | avail_normal -= pages; | |
1381 | else | |
1382 | avail_normal = 0; | |
ef4aede3 RW |
1383 | if (size < pages) |
1384 | size = min_t(unsigned long, pages, max_size); | |
1385 | ||
4bb33435 RW |
1386 | /* |
1387 | * Let the memory management subsystem know that we're going to need a | |
1388 | * large number of page frames to allocate and make it free some memory. | |
1389 | * NOTE: If this is not done, performance will be hurt badly in some | |
1390 | * test cases. | |
1391 | */ | |
1392 | shrink_all_memory(saveable - size); | |
1393 | ||
1394 | /* | |
1395 | * The number of saveable pages in memory was too high, so apply some | |
1396 | * pressure to decrease it. First, make room for the largest possible | |
1397 | * image and fail if that doesn't work. Next, try to decrease the size | |
ef4aede3 RW |
1398 | * of the image as much as indicated by 'size' using allocations from |
1399 | * highmem and non-highmem zones separately. | |
4bb33435 RW |
1400 | */ |
1401 | pages_highmem = preallocate_image_highmem(highmem / 2); | |
1402 | alloc = (count - max_size) - pages_highmem; | |
6715045d RW |
1403 | pages = preallocate_image_memory(alloc, avail_normal); |
1404 | if (pages < alloc) { | |
1405 | /* We have exhausted non-highmem pages, try highmem. */ | |
1406 | alloc -= pages; | |
1407 | pages += pages_highmem; | |
1408 | pages_highmem = preallocate_image_highmem(alloc); | |
1409 | if (pages_highmem < alloc) | |
1410 | goto err_out; | |
1411 | pages += pages_highmem; | |
1412 | /* | |
1413 | * size is the desired number of saveable pages to leave in | |
1414 | * memory, so try to preallocate (all memory - size) pages. | |
1415 | */ | |
1416 | alloc = (count - pages) - size; | |
1417 | pages += preallocate_image_highmem(alloc); | |
1418 | } else { | |
1419 | /* | |
1420 | * There are approximately max_size saveable pages at this point | |
1421 | * and we want to reduce this number down to size. | |
1422 | */ | |
1423 | alloc = max_size - size; | |
1424 | size = preallocate_highmem_fraction(alloc, highmem, count); | |
1425 | pages_highmem += size; | |
1426 | alloc -= size; | |
1427 | size = preallocate_image_memory(alloc, avail_normal); | |
1428 | pages_highmem += preallocate_image_highmem(alloc - size); | |
1429 | pages += pages_highmem + size; | |
1430 | } | |
4bb33435 | 1431 | |
64a473cb RW |
1432 | /* |
1433 | * We only need as many page frames for the image as there are saveable | |
1434 | * pages in memory, but we have allocated more. Release the excessive | |
1435 | * ones now. | |
1436 | */ | |
1437 | free_unnecessary_pages(); | |
4bb33435 RW |
1438 | |
1439 | out: | |
fe419535 | 1440 | do_gettimeofday(&stop); |
64a473cb RW |
1441 | printk(KERN_CONT "done (allocated %lu pages)\n", pages); |
1442 | swsusp_show_speed(&start, &stop, pages, "Allocated"); | |
fe419535 RW |
1443 | |
1444 | return 0; | |
64a473cb RW |
1445 | |
1446 | err_out: | |
1447 | printk(KERN_CONT "\n"); | |
1448 | swsusp_free(); | |
1449 | return -ENOMEM; | |
fe419535 RW |
1450 | } |
1451 | ||
8357376d RW |
1452 | #ifdef CONFIG_HIGHMEM |
1453 | /** | |
1454 | * count_pages_for_highmem - compute the number of non-highmem pages | |
1455 | * that will be necessary for creating copies of highmem pages. | |
1456 | */ | |
1457 | ||
1458 | static unsigned int count_pages_for_highmem(unsigned int nr_highmem) | |
1459 | { | |
64a473cb | 1460 | unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem; |
8357376d RW |
1461 | |
1462 | if (free_highmem >= nr_highmem) | |
1463 | nr_highmem = 0; | |
1464 | else | |
1465 | nr_highmem -= free_highmem; | |
1466 | ||
1467 | return nr_highmem; | |
1468 | } | |
1469 | #else | |
1470 | static unsigned int | |
1471 | count_pages_for_highmem(unsigned int nr_highmem) { return 0; } | |
1472 | #endif /* CONFIG_HIGHMEM */ | |
25761b6e RW |
1473 | |
1474 | /** | |
8357376d RW |
1475 | * enough_free_mem - Make sure we have enough free memory for the |
1476 | * snapshot image. | |
25761b6e RW |
1477 | */ |
1478 | ||
8357376d | 1479 | static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem) |
25761b6e | 1480 | { |
e5e2fa78 | 1481 | struct zone *zone; |
64a473cb | 1482 | unsigned int free = alloc_normal; |
e5e2fa78 | 1483 | |
98e73dc5 | 1484 | for_each_populated_zone(zone) |
8357376d | 1485 | if (!is_highmem(zone)) |
d23ad423 | 1486 | free += zone_page_state(zone, NR_FREE_PAGES); |
940864dd | 1487 | |
8357376d | 1488 | nr_pages += count_pages_for_highmem(nr_highmem); |
64a473cb RW |
1489 | pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n", |
1490 | nr_pages, PAGES_FOR_IO, free); | |
940864dd | 1491 | |
64a473cb | 1492 | return free > nr_pages + PAGES_FOR_IO; |
25761b6e RW |
1493 | } |
1494 | ||
8357376d RW |
1495 | #ifdef CONFIG_HIGHMEM |
1496 | /** | |
1497 | * get_highmem_buffer - if there are some highmem pages in the suspend | |
1498 | * image, we may need the buffer to copy them and/or load their data. | |
1499 | */ | |
1500 | ||
1501 | static inline int get_highmem_buffer(int safe_needed) | |
1502 | { | |
1503 | buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed); | |
1504 | return buffer ? 0 : -ENOMEM; | |
1505 | } | |
1506 | ||
1507 | /** | |
1508 | * alloc_highmem_image_pages - allocate some highmem pages for the image. | |
1509 | * Try to allocate as many pages as needed, but if the number of free | |
1510 | * highmem pages is lesser than that, allocate them all. | |
1511 | */ | |
1512 | ||
1513 | static inline unsigned int | |
64a473cb | 1514 | alloc_highmem_pages(struct memory_bitmap *bm, unsigned int nr_highmem) |
8357376d RW |
1515 | { |
1516 | unsigned int to_alloc = count_free_highmem_pages(); | |
1517 | ||
1518 | if (to_alloc > nr_highmem) | |
1519 | to_alloc = nr_highmem; | |
1520 | ||
1521 | nr_highmem -= to_alloc; | |
1522 | while (to_alloc-- > 0) { | |
1523 | struct page *page; | |
1524 | ||
1525 | page = alloc_image_page(__GFP_HIGHMEM); | |
1526 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1527 | } | |
1528 | return nr_highmem; | |
1529 | } | |
1530 | #else | |
1531 | static inline int get_highmem_buffer(int safe_needed) { return 0; } | |
1532 | ||
1533 | static inline unsigned int | |
64a473cb | 1534 | alloc_highmem_pages(struct memory_bitmap *bm, unsigned int n) { return 0; } |
8357376d RW |
1535 | #endif /* CONFIG_HIGHMEM */ |
1536 | ||
1537 | /** | |
1538 | * swsusp_alloc - allocate memory for the suspend image | |
1539 | * | |
1540 | * We first try to allocate as many highmem pages as there are | |
1541 | * saveable highmem pages in the system. If that fails, we allocate | |
1542 | * non-highmem pages for the copies of the remaining highmem ones. | |
1543 | * | |
1544 | * In this approach it is likely that the copies of highmem pages will | |
1545 | * also be located in the high memory, because of the way in which | |
1546 | * copy_data_pages() works. | |
1547 | */ | |
1548 | ||
b788db79 RW |
1549 | static int |
1550 | swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm, | |
8357376d | 1551 | unsigned int nr_pages, unsigned int nr_highmem) |
054bd4c1 | 1552 | { |
8357376d | 1553 | if (nr_highmem > 0) { |
2e725a06 | 1554 | if (get_highmem_buffer(PG_ANY)) |
64a473cb RW |
1555 | goto err_out; |
1556 | if (nr_highmem > alloc_highmem) { | |
1557 | nr_highmem -= alloc_highmem; | |
1558 | nr_pages += alloc_highmem_pages(copy_bm, nr_highmem); | |
1559 | } | |
8357376d | 1560 | } |
64a473cb RW |
1561 | if (nr_pages > alloc_normal) { |
1562 | nr_pages -= alloc_normal; | |
1563 | while (nr_pages-- > 0) { | |
1564 | struct page *page; | |
1565 | ||
1566 | page = alloc_image_page(GFP_ATOMIC | __GFP_COLD); | |
1567 | if (!page) | |
1568 | goto err_out; | |
1569 | memory_bm_set_bit(copy_bm, page_to_pfn(page)); | |
1570 | } | |
25761b6e | 1571 | } |
64a473cb | 1572 | |
b788db79 | 1573 | return 0; |
25761b6e | 1574 | |
64a473cb | 1575 | err_out: |
b788db79 | 1576 | swsusp_free(); |
2e725a06 | 1577 | return -ENOMEM; |
25761b6e RW |
1578 | } |
1579 | ||
2e32a43e | 1580 | asmlinkage int swsusp_save(void) |
25761b6e | 1581 | { |
8357376d | 1582 | unsigned int nr_pages, nr_highmem; |
25761b6e | 1583 | |
07c3bb57 | 1584 | printk(KERN_INFO "PM: Creating hibernation image:\n"); |
25761b6e | 1585 | |
9f8f2172 | 1586 | drain_local_pages(NULL); |
a0f49651 | 1587 | nr_pages = count_data_pages(); |
8357376d | 1588 | nr_highmem = count_highmem_pages(); |
23976728 | 1589 | printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem); |
25761b6e | 1590 | |
8357376d | 1591 | if (!enough_free_mem(nr_pages, nr_highmem)) { |
23976728 | 1592 | printk(KERN_ERR "PM: Not enough free memory\n"); |
25761b6e RW |
1593 | return -ENOMEM; |
1594 | } | |
1595 | ||
8357376d | 1596 | if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) { |
23976728 | 1597 | printk(KERN_ERR "PM: Memory allocation failed\n"); |
a0f49651 | 1598 | return -ENOMEM; |
8357376d | 1599 | } |
25761b6e RW |
1600 | |
1601 | /* During allocating of suspend pagedir, new cold pages may appear. | |
1602 | * Kill them. | |
1603 | */ | |
9f8f2172 | 1604 | drain_local_pages(NULL); |
b788db79 | 1605 | copy_data_pages(©_bm, &orig_bm); |
25761b6e RW |
1606 | |
1607 | /* | |
1608 | * End of critical section. From now on, we can write to memory, | |
1609 | * but we should not touch disk. This specially means we must _not_ | |
1610 | * touch swap space! Except we must write out our image of course. | |
1611 | */ | |
1612 | ||
8357376d | 1613 | nr_pages += nr_highmem; |
a0f49651 | 1614 | nr_copy_pages = nr_pages; |
8357376d | 1615 | nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE); |
a0f49651 | 1616 | |
23976728 RW |
1617 | printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n", |
1618 | nr_pages); | |
8357376d | 1619 | |
25761b6e RW |
1620 | return 0; |
1621 | } | |
f577eb30 | 1622 | |
d307c4a8 RW |
1623 | #ifndef CONFIG_ARCH_HIBERNATION_HEADER |
1624 | static int init_header_complete(struct swsusp_info *info) | |
f577eb30 | 1625 | { |
d307c4a8 | 1626 | memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname)); |
f577eb30 | 1627 | info->version_code = LINUX_VERSION_CODE; |
d307c4a8 RW |
1628 | return 0; |
1629 | } | |
1630 | ||
1631 | static char *check_image_kernel(struct swsusp_info *info) | |
1632 | { | |
1633 | if (info->version_code != LINUX_VERSION_CODE) | |
1634 | return "kernel version"; | |
1635 | if (strcmp(info->uts.sysname,init_utsname()->sysname)) | |
1636 | return "system type"; | |
1637 | if (strcmp(info->uts.release,init_utsname()->release)) | |
1638 | return "kernel release"; | |
1639 | if (strcmp(info->uts.version,init_utsname()->version)) | |
1640 | return "version"; | |
1641 | if (strcmp(info->uts.machine,init_utsname()->machine)) | |
1642 | return "machine"; | |
1643 | return NULL; | |
1644 | } | |
1645 | #endif /* CONFIG_ARCH_HIBERNATION_HEADER */ | |
1646 | ||
af508b34 RW |
1647 | unsigned long snapshot_get_image_size(void) |
1648 | { | |
1649 | return nr_copy_pages + nr_meta_pages + 1; | |
1650 | } | |
1651 | ||
d307c4a8 RW |
1652 | static int init_header(struct swsusp_info *info) |
1653 | { | |
1654 | memset(info, 0, sizeof(struct swsusp_info)); | |
0ed5fd13 | 1655 | info->num_physpages = get_num_physpages(); |
f577eb30 | 1656 | info->image_pages = nr_copy_pages; |
af508b34 | 1657 | info->pages = snapshot_get_image_size(); |
6e1819d6 RW |
1658 | info->size = info->pages; |
1659 | info->size <<= PAGE_SHIFT; | |
d307c4a8 | 1660 | return init_header_complete(info); |
f577eb30 RW |
1661 | } |
1662 | ||
1663 | /** | |
940864dd RW |
1664 | * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm |
1665 | * are stored in the array @buf[] (1 page at a time) | |
f577eb30 RW |
1666 | */ |
1667 | ||
b788db79 | 1668 | static inline void |
940864dd | 1669 | pack_pfns(unsigned long *buf, struct memory_bitmap *bm) |
f577eb30 RW |
1670 | { |
1671 | int j; | |
1672 | ||
b788db79 | 1673 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
940864dd RW |
1674 | buf[j] = memory_bm_next_pfn(bm); |
1675 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
b788db79 | 1676 | break; |
85055dd8 MS |
1677 | /* Save page key for data page (s390 only). */ |
1678 | page_key_read(buf + j); | |
f577eb30 | 1679 | } |
f577eb30 RW |
1680 | } |
1681 | ||
1682 | /** | |
1683 | * snapshot_read_next - used for reading the system memory snapshot. | |
1684 | * | |
1685 | * On the first call to it @handle should point to a zeroed | |
1686 | * snapshot_handle structure. The structure gets updated and a pointer | |
1687 | * to it should be passed to this function every next time. | |
1688 | * | |
f577eb30 RW |
1689 | * On success the function returns a positive number. Then, the caller |
1690 | * is allowed to read up to the returned number of bytes from the memory | |
d3c1b24c | 1691 | * location computed by the data_of() macro. |
f577eb30 RW |
1692 | * |
1693 | * The function returns 0 to indicate the end of data stream condition, | |
1694 | * and a negative number is returned on error. In such cases the | |
1695 | * structure pointed to by @handle is not updated and should not be used | |
1696 | * any more. | |
1697 | */ | |
1698 | ||
d3c1b24c | 1699 | int snapshot_read_next(struct snapshot_handle *handle) |
f577eb30 | 1700 | { |
fb13a28b | 1701 | if (handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 1702 | return 0; |
b788db79 | 1703 | |
f577eb30 RW |
1704 | if (!buffer) { |
1705 | /* This makes the buffer be freed by swsusp_free() */ | |
8357376d | 1706 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); |
f577eb30 RW |
1707 | if (!buffer) |
1708 | return -ENOMEM; | |
1709 | } | |
d3c1b24c | 1710 | if (!handle->cur) { |
d307c4a8 RW |
1711 | int error; |
1712 | ||
1713 | error = init_header((struct swsusp_info *)buffer); | |
1714 | if (error) | |
1715 | return error; | |
f577eb30 | 1716 | handle->buffer = buffer; |
b788db79 RW |
1717 | memory_bm_position_reset(&orig_bm); |
1718 | memory_bm_position_reset(©_bm); | |
d3c1b24c | 1719 | } else if (handle->cur <= nr_meta_pages) { |
3ecb01df | 1720 | clear_page(buffer); |
d3c1b24c JS |
1721 | pack_pfns(buffer, &orig_bm); |
1722 | } else { | |
1723 | struct page *page; | |
b788db79 | 1724 | |
d3c1b24c JS |
1725 | page = pfn_to_page(memory_bm_next_pfn(©_bm)); |
1726 | if (PageHighMem(page)) { | |
1727 | /* Highmem pages are copied to the buffer, | |
1728 | * because we can't return with a kmapped | |
1729 | * highmem page (we may not be called again). | |
1730 | */ | |
1731 | void *kaddr; | |
8357376d | 1732 | |
0de9a1e2 | 1733 | kaddr = kmap_atomic(page); |
3ecb01df | 1734 | copy_page(buffer, kaddr); |
0de9a1e2 | 1735 | kunmap_atomic(kaddr); |
d3c1b24c JS |
1736 | handle->buffer = buffer; |
1737 | } else { | |
1738 | handle->buffer = page_address(page); | |
f577eb30 | 1739 | } |
f577eb30 | 1740 | } |
d3c1b24c JS |
1741 | handle->cur++; |
1742 | return PAGE_SIZE; | |
f577eb30 RW |
1743 | } |
1744 | ||
1745 | /** | |
1746 | * mark_unsafe_pages - mark the pages that cannot be used for storing | |
1747 | * the image during resume, because they conflict with the pages that | |
1748 | * had been used before suspend | |
1749 | */ | |
1750 | ||
940864dd | 1751 | static int mark_unsafe_pages(struct memory_bitmap *bm) |
f577eb30 RW |
1752 | { |
1753 | struct zone *zone; | |
ae83c5ee | 1754 | unsigned long pfn, max_zone_pfn; |
f577eb30 RW |
1755 | |
1756 | /* Clear page flags */ | |
98e73dc5 | 1757 | for_each_populated_zone(zone) { |
ae83c5ee RW |
1758 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
1759 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1760 | if (pfn_valid(pfn)) | |
7be98234 | 1761 | swsusp_unset_page_free(pfn_to_page(pfn)); |
f577eb30 RW |
1762 | } |
1763 | ||
940864dd RW |
1764 | /* Mark pages that correspond to the "original" pfns as "unsafe" */ |
1765 | memory_bm_position_reset(bm); | |
1766 | do { | |
1767 | pfn = memory_bm_next_pfn(bm); | |
1768 | if (likely(pfn != BM_END_OF_MAP)) { | |
1769 | if (likely(pfn_valid(pfn))) | |
7be98234 | 1770 | swsusp_set_page_free(pfn_to_page(pfn)); |
940864dd RW |
1771 | else |
1772 | return -EFAULT; | |
1773 | } | |
1774 | } while (pfn != BM_END_OF_MAP); | |
f577eb30 | 1775 | |
940864dd | 1776 | allocated_unsafe_pages = 0; |
968808b8 | 1777 | |
f577eb30 RW |
1778 | return 0; |
1779 | } | |
1780 | ||
940864dd RW |
1781 | static void |
1782 | duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src) | |
f577eb30 | 1783 | { |
940864dd RW |
1784 | unsigned long pfn; |
1785 | ||
1786 | memory_bm_position_reset(src); | |
1787 | pfn = memory_bm_next_pfn(src); | |
1788 | while (pfn != BM_END_OF_MAP) { | |
1789 | memory_bm_set_bit(dst, pfn); | |
1790 | pfn = memory_bm_next_pfn(src); | |
f577eb30 RW |
1791 | } |
1792 | } | |
1793 | ||
d307c4a8 | 1794 | static int check_header(struct swsusp_info *info) |
f577eb30 | 1795 | { |
d307c4a8 | 1796 | char *reason; |
f577eb30 | 1797 | |
d307c4a8 | 1798 | reason = check_image_kernel(info); |
0ed5fd13 | 1799 | if (!reason && info->num_physpages != get_num_physpages()) |
f577eb30 | 1800 | reason = "memory size"; |
f577eb30 | 1801 | if (reason) { |
23976728 | 1802 | printk(KERN_ERR "PM: Image mismatch: %s\n", reason); |
f577eb30 RW |
1803 | return -EPERM; |
1804 | } | |
1805 | return 0; | |
1806 | } | |
1807 | ||
1808 | /** | |
1809 | * load header - check the image header and copy data from it | |
1810 | */ | |
1811 | ||
940864dd RW |
1812 | static int |
1813 | load_header(struct swsusp_info *info) | |
f577eb30 RW |
1814 | { |
1815 | int error; | |
f577eb30 | 1816 | |
940864dd | 1817 | restore_pblist = NULL; |
f577eb30 RW |
1818 | error = check_header(info); |
1819 | if (!error) { | |
f577eb30 RW |
1820 | nr_copy_pages = info->image_pages; |
1821 | nr_meta_pages = info->pages - info->image_pages - 1; | |
1822 | } | |
1823 | return error; | |
1824 | } | |
1825 | ||
1826 | /** | |
940864dd RW |
1827 | * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set |
1828 | * the corresponding bit in the memory bitmap @bm | |
f577eb30 | 1829 | */ |
69643279 | 1830 | static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm) |
f577eb30 RW |
1831 | { |
1832 | int j; | |
1833 | ||
940864dd RW |
1834 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
1835 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
1836 | break; | |
1837 | ||
85055dd8 MS |
1838 | /* Extract and buffer page key for data page (s390 only). */ |
1839 | page_key_memorize(buf + j); | |
1840 | ||
69643279 RW |
1841 | if (memory_bm_pfn_present(bm, buf[j])) |
1842 | memory_bm_set_bit(bm, buf[j]); | |
1843 | else | |
1844 | return -EFAULT; | |
f577eb30 | 1845 | } |
69643279 RW |
1846 | |
1847 | return 0; | |
f577eb30 RW |
1848 | } |
1849 | ||
8357376d RW |
1850 | /* List of "safe" pages that may be used to store data loaded from the suspend |
1851 | * image | |
1852 | */ | |
1853 | static struct linked_page *safe_pages_list; | |
1854 | ||
1855 | #ifdef CONFIG_HIGHMEM | |
1856 | /* struct highmem_pbe is used for creating the list of highmem pages that | |
1857 | * should be restored atomically during the resume from disk, because the page | |
1858 | * frames they have occupied before the suspend are in use. | |
1859 | */ | |
1860 | struct highmem_pbe { | |
1861 | struct page *copy_page; /* data is here now */ | |
1862 | struct page *orig_page; /* data was here before the suspend */ | |
1863 | struct highmem_pbe *next; | |
1864 | }; | |
1865 | ||
1866 | /* List of highmem PBEs needed for restoring the highmem pages that were | |
1867 | * allocated before the suspend and included in the suspend image, but have | |
1868 | * also been allocated by the "resume" kernel, so their contents cannot be | |
1869 | * written directly to their "original" page frames. | |
1870 | */ | |
1871 | static struct highmem_pbe *highmem_pblist; | |
1872 | ||
1873 | /** | |
1874 | * count_highmem_image_pages - compute the number of highmem pages in the | |
1875 | * suspend image. The bits in the memory bitmap @bm that correspond to the | |
1876 | * image pages are assumed to be set. | |
1877 | */ | |
1878 | ||
1879 | static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) | |
1880 | { | |
1881 | unsigned long pfn; | |
1882 | unsigned int cnt = 0; | |
1883 | ||
1884 | memory_bm_position_reset(bm); | |
1885 | pfn = memory_bm_next_pfn(bm); | |
1886 | while (pfn != BM_END_OF_MAP) { | |
1887 | if (PageHighMem(pfn_to_page(pfn))) | |
1888 | cnt++; | |
1889 | ||
1890 | pfn = memory_bm_next_pfn(bm); | |
1891 | } | |
1892 | return cnt; | |
1893 | } | |
1894 | ||
1895 | /** | |
1896 | * prepare_highmem_image - try to allocate as many highmem pages as | |
1897 | * there are highmem image pages (@nr_highmem_p points to the variable | |
1898 | * containing the number of highmem image pages). The pages that are | |
1899 | * "safe" (ie. will not be overwritten when the suspend image is | |
1900 | * restored) have the corresponding bits set in @bm (it must be | |
1901 | * unitialized). | |
1902 | * | |
1903 | * NOTE: This function should not be called if there are no highmem | |
1904 | * image pages. | |
1905 | */ | |
1906 | ||
1907 | static unsigned int safe_highmem_pages; | |
1908 | ||
1909 | static struct memory_bitmap *safe_highmem_bm; | |
1910 | ||
1911 | static int | |
1912 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
1913 | { | |
1914 | unsigned int to_alloc; | |
1915 | ||
1916 | if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE)) | |
1917 | return -ENOMEM; | |
1918 | ||
1919 | if (get_highmem_buffer(PG_SAFE)) | |
1920 | return -ENOMEM; | |
1921 | ||
1922 | to_alloc = count_free_highmem_pages(); | |
1923 | if (to_alloc > *nr_highmem_p) | |
1924 | to_alloc = *nr_highmem_p; | |
1925 | else | |
1926 | *nr_highmem_p = to_alloc; | |
1927 | ||
1928 | safe_highmem_pages = 0; | |
1929 | while (to_alloc-- > 0) { | |
1930 | struct page *page; | |
1931 | ||
1932 | page = alloc_page(__GFP_HIGHMEM); | |
7be98234 | 1933 | if (!swsusp_page_is_free(page)) { |
8357376d RW |
1934 | /* The page is "safe", set its bit the bitmap */ |
1935 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1936 | safe_highmem_pages++; | |
1937 | } | |
1938 | /* Mark the page as allocated */ | |
7be98234 RW |
1939 | swsusp_set_page_forbidden(page); |
1940 | swsusp_set_page_free(page); | |
8357376d RW |
1941 | } |
1942 | memory_bm_position_reset(bm); | |
1943 | safe_highmem_bm = bm; | |
1944 | return 0; | |
1945 | } | |
1946 | ||
1947 | /** | |
1948 | * get_highmem_page_buffer - for given highmem image page find the buffer | |
1949 | * that suspend_write_next() should set for its caller to write to. | |
1950 | * | |
1951 | * If the page is to be saved to its "original" page frame or a copy of | |
1952 | * the page is to be made in the highmem, @buffer is returned. Otherwise, | |
1953 | * the copy of the page is to be made in normal memory, so the address of | |
1954 | * the copy is returned. | |
1955 | * | |
1956 | * If @buffer is returned, the caller of suspend_write_next() will write | |
1957 | * the page's contents to @buffer, so they will have to be copied to the | |
1958 | * right location on the next call to suspend_write_next() and it is done | |
1959 | * with the help of copy_last_highmem_page(). For this purpose, if | |
1960 | * @buffer is returned, @last_highmem page is set to the page to which | |
1961 | * the data will have to be copied from @buffer. | |
1962 | */ | |
1963 | ||
1964 | static struct page *last_highmem_page; | |
1965 | ||
1966 | static void * | |
1967 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
1968 | { | |
1969 | struct highmem_pbe *pbe; | |
1970 | void *kaddr; | |
1971 | ||
7be98234 | 1972 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) { |
8357376d RW |
1973 | /* We have allocated the "original" page frame and we can |
1974 | * use it directly to store the loaded page. | |
1975 | */ | |
1976 | last_highmem_page = page; | |
1977 | return buffer; | |
1978 | } | |
1979 | /* The "original" page frame has not been allocated and we have to | |
1980 | * use a "safe" page frame to store the loaded page. | |
1981 | */ | |
1982 | pbe = chain_alloc(ca, sizeof(struct highmem_pbe)); | |
1983 | if (!pbe) { | |
1984 | swsusp_free(); | |
69643279 | 1985 | return ERR_PTR(-ENOMEM); |
8357376d RW |
1986 | } |
1987 | pbe->orig_page = page; | |
1988 | if (safe_highmem_pages > 0) { | |
1989 | struct page *tmp; | |
1990 | ||
1991 | /* Copy of the page will be stored in high memory */ | |
1992 | kaddr = buffer; | |
1993 | tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm)); | |
1994 | safe_highmem_pages--; | |
1995 | last_highmem_page = tmp; | |
1996 | pbe->copy_page = tmp; | |
1997 | } else { | |
1998 | /* Copy of the page will be stored in normal memory */ | |
1999 | kaddr = safe_pages_list; | |
2000 | safe_pages_list = safe_pages_list->next; | |
2001 | pbe->copy_page = virt_to_page(kaddr); | |
2002 | } | |
2003 | pbe->next = highmem_pblist; | |
2004 | highmem_pblist = pbe; | |
2005 | return kaddr; | |
2006 | } | |
2007 | ||
2008 | /** | |
2009 | * copy_last_highmem_page - copy the contents of a highmem image from | |
2010 | * @buffer, where the caller of snapshot_write_next() has place them, | |
2011 | * to the right location represented by @last_highmem_page . | |
2012 | */ | |
2013 | ||
2014 | static void copy_last_highmem_page(void) | |
2015 | { | |
2016 | if (last_highmem_page) { | |
2017 | void *dst; | |
2018 | ||
0de9a1e2 | 2019 | dst = kmap_atomic(last_highmem_page); |
3ecb01df | 2020 | copy_page(dst, buffer); |
0de9a1e2 | 2021 | kunmap_atomic(dst); |
8357376d RW |
2022 | last_highmem_page = NULL; |
2023 | } | |
2024 | } | |
2025 | ||
2026 | static inline int last_highmem_page_copied(void) | |
2027 | { | |
2028 | return !last_highmem_page; | |
2029 | } | |
2030 | ||
2031 | static inline void free_highmem_data(void) | |
2032 | { | |
2033 | if (safe_highmem_bm) | |
2034 | memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR); | |
2035 | ||
2036 | if (buffer) | |
2037 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
2038 | } | |
2039 | #else | |
2040 | static inline int get_safe_write_buffer(void) { return 0; } | |
2041 | ||
2042 | static unsigned int | |
2043 | count_highmem_image_pages(struct memory_bitmap *bm) { return 0; } | |
2044 | ||
2045 | static inline int | |
2046 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
2047 | { | |
2048 | return 0; | |
2049 | } | |
2050 | ||
2051 | static inline void * | |
2052 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
2053 | { | |
69643279 | 2054 | return ERR_PTR(-EINVAL); |
8357376d RW |
2055 | } |
2056 | ||
2057 | static inline void copy_last_highmem_page(void) {} | |
2058 | static inline int last_highmem_page_copied(void) { return 1; } | |
2059 | static inline void free_highmem_data(void) {} | |
2060 | #endif /* CONFIG_HIGHMEM */ | |
2061 | ||
f577eb30 | 2062 | /** |
940864dd RW |
2063 | * prepare_image - use the memory bitmap @bm to mark the pages that will |
2064 | * be overwritten in the process of restoring the system memory state | |
2065 | * from the suspend image ("unsafe" pages) and allocate memory for the | |
2066 | * image. | |
968808b8 | 2067 | * |
940864dd RW |
2068 | * The idea is to allocate a new memory bitmap first and then allocate |
2069 | * as many pages as needed for the image data, but not to assign these | |
2070 | * pages to specific tasks initially. Instead, we just mark them as | |
8357376d RW |
2071 | * allocated and create a lists of "safe" pages that will be used |
2072 | * later. On systems with high memory a list of "safe" highmem pages is | |
2073 | * also created. | |
f577eb30 RW |
2074 | */ |
2075 | ||
940864dd RW |
2076 | #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe)) |
2077 | ||
940864dd RW |
2078 | static int |
2079 | prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm) | |
f577eb30 | 2080 | { |
8357376d | 2081 | unsigned int nr_pages, nr_highmem; |
940864dd RW |
2082 | struct linked_page *sp_list, *lp; |
2083 | int error; | |
f577eb30 | 2084 | |
8357376d RW |
2085 | /* If there is no highmem, the buffer will not be necessary */ |
2086 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
2087 | buffer = NULL; | |
2088 | ||
2089 | nr_highmem = count_highmem_image_pages(bm); | |
940864dd RW |
2090 | error = mark_unsafe_pages(bm); |
2091 | if (error) | |
2092 | goto Free; | |
2093 | ||
2094 | error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE); | |
2095 | if (error) | |
2096 | goto Free; | |
2097 | ||
2098 | duplicate_memory_bitmap(new_bm, bm); | |
2099 | memory_bm_free(bm, PG_UNSAFE_KEEP); | |
8357376d RW |
2100 | if (nr_highmem > 0) { |
2101 | error = prepare_highmem_image(bm, &nr_highmem); | |
2102 | if (error) | |
2103 | goto Free; | |
2104 | } | |
940864dd RW |
2105 | /* Reserve some safe pages for potential later use. |
2106 | * | |
2107 | * NOTE: This way we make sure there will be enough safe pages for the | |
2108 | * chain_alloc() in get_buffer(). It is a bit wasteful, but | |
2109 | * nr_copy_pages cannot be greater than 50% of the memory anyway. | |
2110 | */ | |
2111 | sp_list = NULL; | |
2112 | /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */ | |
8357376d | 2113 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
2114 | nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE); |
2115 | while (nr_pages > 0) { | |
8357376d | 2116 | lp = get_image_page(GFP_ATOMIC, PG_SAFE); |
940864dd | 2117 | if (!lp) { |
f577eb30 | 2118 | error = -ENOMEM; |
940864dd RW |
2119 | goto Free; |
2120 | } | |
2121 | lp->next = sp_list; | |
2122 | sp_list = lp; | |
2123 | nr_pages--; | |
f577eb30 | 2124 | } |
940864dd RW |
2125 | /* Preallocate memory for the image */ |
2126 | safe_pages_list = NULL; | |
8357376d | 2127 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
2128 | while (nr_pages > 0) { |
2129 | lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC); | |
2130 | if (!lp) { | |
2131 | error = -ENOMEM; | |
2132 | goto Free; | |
2133 | } | |
7be98234 | 2134 | if (!swsusp_page_is_free(virt_to_page(lp))) { |
940864dd RW |
2135 | /* The page is "safe", add it to the list */ |
2136 | lp->next = safe_pages_list; | |
2137 | safe_pages_list = lp; | |
968808b8 | 2138 | } |
940864dd | 2139 | /* Mark the page as allocated */ |
7be98234 RW |
2140 | swsusp_set_page_forbidden(virt_to_page(lp)); |
2141 | swsusp_set_page_free(virt_to_page(lp)); | |
940864dd | 2142 | nr_pages--; |
968808b8 | 2143 | } |
940864dd RW |
2144 | /* Free the reserved safe pages so that chain_alloc() can use them */ |
2145 | while (sp_list) { | |
2146 | lp = sp_list->next; | |
2147 | free_image_page(sp_list, PG_UNSAFE_CLEAR); | |
2148 | sp_list = lp; | |
f577eb30 | 2149 | } |
940864dd RW |
2150 | return 0; |
2151 | ||
59a49335 | 2152 | Free: |
940864dd | 2153 | swsusp_free(); |
f577eb30 RW |
2154 | return error; |
2155 | } | |
2156 | ||
940864dd RW |
2157 | /** |
2158 | * get_buffer - compute the address that snapshot_write_next() should | |
2159 | * set for its caller to write to. | |
2160 | */ | |
2161 | ||
2162 | static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca) | |
968808b8 | 2163 | { |
940864dd | 2164 | struct pbe *pbe; |
69643279 RW |
2165 | struct page *page; |
2166 | unsigned long pfn = memory_bm_next_pfn(bm); | |
968808b8 | 2167 | |
69643279 RW |
2168 | if (pfn == BM_END_OF_MAP) |
2169 | return ERR_PTR(-EFAULT); | |
2170 | ||
2171 | page = pfn_to_page(pfn); | |
8357376d RW |
2172 | if (PageHighMem(page)) |
2173 | return get_highmem_page_buffer(page, ca); | |
2174 | ||
7be98234 | 2175 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) |
940864dd RW |
2176 | /* We have allocated the "original" page frame and we can |
2177 | * use it directly to store the loaded page. | |
968808b8 | 2178 | */ |
940864dd RW |
2179 | return page_address(page); |
2180 | ||
2181 | /* The "original" page frame has not been allocated and we have to | |
2182 | * use a "safe" page frame to store the loaded page. | |
968808b8 | 2183 | */ |
940864dd RW |
2184 | pbe = chain_alloc(ca, sizeof(struct pbe)); |
2185 | if (!pbe) { | |
2186 | swsusp_free(); | |
69643279 | 2187 | return ERR_PTR(-ENOMEM); |
940864dd | 2188 | } |
8357376d RW |
2189 | pbe->orig_address = page_address(page); |
2190 | pbe->address = safe_pages_list; | |
940864dd RW |
2191 | safe_pages_list = safe_pages_list->next; |
2192 | pbe->next = restore_pblist; | |
2193 | restore_pblist = pbe; | |
8357376d | 2194 | return pbe->address; |
968808b8 RW |
2195 | } |
2196 | ||
f577eb30 RW |
2197 | /** |
2198 | * snapshot_write_next - used for writing the system memory snapshot. | |
2199 | * | |
2200 | * On the first call to it @handle should point to a zeroed | |
2201 | * snapshot_handle structure. The structure gets updated and a pointer | |
2202 | * to it should be passed to this function every next time. | |
2203 | * | |
f577eb30 RW |
2204 | * On success the function returns a positive number. Then, the caller |
2205 | * is allowed to write up to the returned number of bytes to the memory | |
d3c1b24c | 2206 | * location computed by the data_of() macro. |
f577eb30 RW |
2207 | * |
2208 | * The function returns 0 to indicate the "end of file" condition, | |
2209 | * and a negative number is returned on error. In such cases the | |
2210 | * structure pointed to by @handle is not updated and should not be used | |
2211 | * any more. | |
2212 | */ | |
2213 | ||
d3c1b24c | 2214 | int snapshot_write_next(struct snapshot_handle *handle) |
f577eb30 | 2215 | { |
940864dd | 2216 | static struct chain_allocator ca; |
f577eb30 RW |
2217 | int error = 0; |
2218 | ||
940864dd | 2219 | /* Check if we have already loaded the entire image */ |
d3c1b24c | 2220 | if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 2221 | return 0; |
940864dd | 2222 | |
d3c1b24c JS |
2223 | handle->sync_read = 1; |
2224 | ||
2225 | if (!handle->cur) { | |
8357376d RW |
2226 | if (!buffer) |
2227 | /* This makes the buffer be freed by swsusp_free() */ | |
2228 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); | |
2229 | ||
f577eb30 RW |
2230 | if (!buffer) |
2231 | return -ENOMEM; | |
8357376d | 2232 | |
f577eb30 | 2233 | handle->buffer = buffer; |
d3c1b24c JS |
2234 | } else if (handle->cur == 1) { |
2235 | error = load_header(buffer); | |
2236 | if (error) | |
2237 | return error; | |
940864dd | 2238 | |
d3c1b24c JS |
2239 | error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY); |
2240 | if (error) | |
2241 | return error; | |
2242 | ||
85055dd8 MS |
2243 | /* Allocate buffer for page keys. */ |
2244 | error = page_key_alloc(nr_copy_pages); | |
2245 | if (error) | |
2246 | return error; | |
2247 | ||
d3c1b24c JS |
2248 | } else if (handle->cur <= nr_meta_pages + 1) { |
2249 | error = unpack_orig_pfns(buffer, ©_bm); | |
2250 | if (error) | |
2251 | return error; | |
940864dd | 2252 | |
d3c1b24c JS |
2253 | if (handle->cur == nr_meta_pages + 1) { |
2254 | error = prepare_image(&orig_bm, ©_bm); | |
69643279 RW |
2255 | if (error) |
2256 | return error; | |
2257 | ||
d3c1b24c JS |
2258 | chain_init(&ca, GFP_ATOMIC, PG_SAFE); |
2259 | memory_bm_position_reset(&orig_bm); | |
2260 | restore_pblist = NULL; | |
940864dd | 2261 | handle->buffer = get_buffer(&orig_bm, &ca); |
d3c1b24c | 2262 | handle->sync_read = 0; |
69643279 RW |
2263 | if (IS_ERR(handle->buffer)) |
2264 | return PTR_ERR(handle->buffer); | |
f577eb30 | 2265 | } |
f577eb30 | 2266 | } else { |
d3c1b24c | 2267 | copy_last_highmem_page(); |
85055dd8 MS |
2268 | /* Restore page key for data page (s390 only). */ |
2269 | page_key_write(handle->buffer); | |
d3c1b24c JS |
2270 | handle->buffer = get_buffer(&orig_bm, &ca); |
2271 | if (IS_ERR(handle->buffer)) | |
2272 | return PTR_ERR(handle->buffer); | |
2273 | if (handle->buffer != buffer) | |
2274 | handle->sync_read = 0; | |
f577eb30 | 2275 | } |
d3c1b24c JS |
2276 | handle->cur++; |
2277 | return PAGE_SIZE; | |
f577eb30 RW |
2278 | } |
2279 | ||
8357376d RW |
2280 | /** |
2281 | * snapshot_write_finalize - must be called after the last call to | |
2282 | * snapshot_write_next() in case the last page in the image happens | |
2283 | * to be a highmem page and its contents should be stored in the | |
2284 | * highmem. Additionally, it releases the memory that will not be | |
2285 | * used any more. | |
2286 | */ | |
2287 | ||
2288 | void snapshot_write_finalize(struct snapshot_handle *handle) | |
2289 | { | |
2290 | copy_last_highmem_page(); | |
85055dd8 MS |
2291 | /* Restore page key for data page (s390 only). */ |
2292 | page_key_write(handle->buffer); | |
2293 | page_key_free(); | |
8357376d | 2294 | /* Free only if we have loaded the image entirely */ |
d3c1b24c | 2295 | if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) { |
8357376d RW |
2296 | memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR); |
2297 | free_highmem_data(); | |
2298 | } | |
2299 | } | |
2300 | ||
f577eb30 RW |
2301 | int snapshot_image_loaded(struct snapshot_handle *handle) |
2302 | { | |
8357376d | 2303 | return !(!nr_copy_pages || !last_highmem_page_copied() || |
940864dd RW |
2304 | handle->cur <= nr_meta_pages + nr_copy_pages); |
2305 | } | |
2306 | ||
8357376d RW |
2307 | #ifdef CONFIG_HIGHMEM |
2308 | /* Assumes that @buf is ready and points to a "safe" page */ | |
2309 | static inline void | |
2310 | swap_two_pages_data(struct page *p1, struct page *p2, void *buf) | |
940864dd | 2311 | { |
8357376d RW |
2312 | void *kaddr1, *kaddr2; |
2313 | ||
0de9a1e2 CW |
2314 | kaddr1 = kmap_atomic(p1); |
2315 | kaddr2 = kmap_atomic(p2); | |
3ecb01df JB |
2316 | copy_page(buf, kaddr1); |
2317 | copy_page(kaddr1, kaddr2); | |
2318 | copy_page(kaddr2, buf); | |
0de9a1e2 CW |
2319 | kunmap_atomic(kaddr2); |
2320 | kunmap_atomic(kaddr1); | |
8357376d RW |
2321 | } |
2322 | ||
2323 | /** | |
2324 | * restore_highmem - for each highmem page that was allocated before | |
2325 | * the suspend and included in the suspend image, and also has been | |
2326 | * allocated by the "resume" kernel swap its current (ie. "before | |
2327 | * resume") contents with the previous (ie. "before suspend") one. | |
2328 | * | |
2329 | * If the resume eventually fails, we can call this function once | |
2330 | * again and restore the "before resume" highmem state. | |
2331 | */ | |
2332 | ||
2333 | int restore_highmem(void) | |
2334 | { | |
2335 | struct highmem_pbe *pbe = highmem_pblist; | |
2336 | void *buf; | |
2337 | ||
2338 | if (!pbe) | |
2339 | return 0; | |
2340 | ||
2341 | buf = get_image_page(GFP_ATOMIC, PG_SAFE); | |
2342 | if (!buf) | |
2343 | return -ENOMEM; | |
2344 | ||
2345 | while (pbe) { | |
2346 | swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf); | |
2347 | pbe = pbe->next; | |
2348 | } | |
2349 | free_image_page(buf, PG_UNSAFE_CLEAR); | |
2350 | return 0; | |
f577eb30 | 2351 | } |
8357376d | 2352 | #endif /* CONFIG_HIGHMEM */ |