]>
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 | ||
8357376d RW |
42 | /* List of PBEs needed for restoring the pages that were allocated before |
43 | * the suspend and included in the suspend image, but have also been | |
44 | * allocated by the "resume" kernel, so their contents cannot be written | |
45 | * directly to their "original" page frames. | |
46 | */ | |
75534b50 RW |
47 | struct pbe *restore_pblist; |
48 | ||
8357376d | 49 | /* Pointer to an auxiliary buffer (1 page) */ |
940864dd | 50 | static void *buffer; |
7088a5c0 | 51 | |
f6143aa6 RW |
52 | /** |
53 | * @safe_needed - on resume, for storing the PBE list and the image, | |
54 | * we can only use memory pages that do not conflict with the pages | |
8357376d RW |
55 | * used before suspend. The unsafe pages have PageNosaveFree set |
56 | * and we count them using unsafe_pages. | |
f6143aa6 | 57 | * |
8357376d RW |
58 | * Each allocated image page is marked as PageNosave and PageNosaveFree |
59 | * so that swsusp_free() can release it. | |
f6143aa6 RW |
60 | */ |
61 | ||
0bcd888d RW |
62 | #define PG_ANY 0 |
63 | #define PG_SAFE 1 | |
64 | #define PG_UNSAFE_CLEAR 1 | |
65 | #define PG_UNSAFE_KEEP 0 | |
66 | ||
940864dd | 67 | static unsigned int allocated_unsafe_pages; |
f6143aa6 | 68 | |
8357376d | 69 | static void *get_image_page(gfp_t gfp_mask, int safe_needed) |
f6143aa6 RW |
70 | { |
71 | void *res; | |
72 | ||
73 | res = (void *)get_zeroed_page(gfp_mask); | |
74 | if (safe_needed) | |
7be98234 | 75 | while (res && swsusp_page_is_free(virt_to_page(res))) { |
f6143aa6 | 76 | /* The page is unsafe, mark it for swsusp_free() */ |
7be98234 | 77 | swsusp_set_page_forbidden(virt_to_page(res)); |
940864dd | 78 | allocated_unsafe_pages++; |
f6143aa6 RW |
79 | res = (void *)get_zeroed_page(gfp_mask); |
80 | } | |
81 | if (res) { | |
7be98234 RW |
82 | swsusp_set_page_forbidden(virt_to_page(res)); |
83 | swsusp_set_page_free(virt_to_page(res)); | |
f6143aa6 RW |
84 | } |
85 | return res; | |
86 | } | |
87 | ||
88 | unsigned long get_safe_page(gfp_t gfp_mask) | |
89 | { | |
8357376d RW |
90 | return (unsigned long)get_image_page(gfp_mask, PG_SAFE); |
91 | } | |
92 | ||
5b6d15de RW |
93 | static struct page *alloc_image_page(gfp_t gfp_mask) |
94 | { | |
8357376d RW |
95 | struct page *page; |
96 | ||
97 | page = alloc_page(gfp_mask); | |
98 | if (page) { | |
7be98234 RW |
99 | swsusp_set_page_forbidden(page); |
100 | swsusp_set_page_free(page); | |
8357376d RW |
101 | } |
102 | return page; | |
f6143aa6 RW |
103 | } |
104 | ||
105 | /** | |
106 | * free_image_page - free page represented by @addr, allocated with | |
8357376d | 107 | * get_image_page (page flags set by it must be cleared) |
f6143aa6 RW |
108 | */ |
109 | ||
110 | static inline void free_image_page(void *addr, int clear_nosave_free) | |
111 | { | |
8357376d RW |
112 | struct page *page; |
113 | ||
114 | BUG_ON(!virt_addr_valid(addr)); | |
115 | ||
116 | page = virt_to_page(addr); | |
117 | ||
7be98234 | 118 | swsusp_unset_page_forbidden(page); |
f6143aa6 | 119 | if (clear_nosave_free) |
7be98234 | 120 | swsusp_unset_page_free(page); |
8357376d RW |
121 | |
122 | __free_page(page); | |
f6143aa6 RW |
123 | } |
124 | ||
b788db79 RW |
125 | /* struct linked_page is used to build chains of pages */ |
126 | ||
127 | #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *)) | |
128 | ||
129 | struct linked_page { | |
130 | struct linked_page *next; | |
131 | char data[LINKED_PAGE_DATA_SIZE]; | |
132 | } __attribute__((packed)); | |
133 | ||
134 | static inline void | |
135 | free_list_of_pages(struct linked_page *list, int clear_page_nosave) | |
136 | { | |
137 | while (list) { | |
138 | struct linked_page *lp = list->next; | |
139 | ||
140 | free_image_page(list, clear_page_nosave); | |
141 | list = lp; | |
142 | } | |
143 | } | |
144 | ||
145 | /** | |
146 | * struct chain_allocator is used for allocating small objects out of | |
147 | * a linked list of pages called 'the chain'. | |
148 | * | |
149 | * The chain grows each time when there is no room for a new object in | |
150 | * the current page. The allocated objects cannot be freed individually. | |
151 | * It is only possible to free them all at once, by freeing the entire | |
152 | * chain. | |
153 | * | |
154 | * NOTE: The chain allocator may be inefficient if the allocated objects | |
155 | * are not much smaller than PAGE_SIZE. | |
156 | */ | |
157 | ||
158 | struct chain_allocator { | |
159 | struct linked_page *chain; /* the chain */ | |
160 | unsigned int used_space; /* total size of objects allocated out | |
161 | * of the current page | |
162 | */ | |
163 | gfp_t gfp_mask; /* mask for allocating pages */ | |
164 | int safe_needed; /* if set, only "safe" pages are allocated */ | |
165 | }; | |
166 | ||
167 | static void | |
168 | chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed) | |
169 | { | |
170 | ca->chain = NULL; | |
171 | ca->used_space = LINKED_PAGE_DATA_SIZE; | |
172 | ca->gfp_mask = gfp_mask; | |
173 | ca->safe_needed = safe_needed; | |
174 | } | |
175 | ||
176 | static void *chain_alloc(struct chain_allocator *ca, unsigned int size) | |
177 | { | |
178 | void *ret; | |
179 | ||
180 | if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) { | |
181 | struct linked_page *lp; | |
182 | ||
8357376d | 183 | lp = get_image_page(ca->gfp_mask, ca->safe_needed); |
b788db79 RW |
184 | if (!lp) |
185 | return NULL; | |
186 | ||
187 | lp->next = ca->chain; | |
188 | ca->chain = lp; | |
189 | ca->used_space = 0; | |
190 | } | |
191 | ret = ca->chain->data + ca->used_space; | |
192 | ca->used_space += size; | |
193 | return ret; | |
194 | } | |
195 | ||
b788db79 RW |
196 | /** |
197 | * Data types related to memory bitmaps. | |
198 | * | |
199 | * Memory bitmap is a structure consiting of many linked lists of | |
200 | * objects. The main list's elements are of type struct zone_bitmap | |
201 | * and each of them corresonds to one zone. For each zone bitmap | |
202 | * object there is a list of objects of type struct bm_block that | |
0d83304c | 203 | * represent each blocks of bitmap in which information is stored. |
b788db79 RW |
204 | * |
205 | * struct memory_bitmap contains a pointer to the main list of zone | |
206 | * bitmap objects, a struct bm_position used for browsing the bitmap, | |
207 | * and a pointer to the list of pages used for allocating all of the | |
208 | * zone bitmap objects and bitmap block objects. | |
209 | * | |
210 | * NOTE: It has to be possible to lay out the bitmap in memory | |
211 | * using only allocations of order 0. Additionally, the bitmap is | |
212 | * designed to work with arbitrary number of zones (this is over the | |
213 | * top for now, but let's avoid making unnecessary assumptions ;-). | |
214 | * | |
215 | * struct zone_bitmap contains a pointer to a list of bitmap block | |
216 | * objects and a pointer to the bitmap block object that has been | |
217 | * most recently used for setting bits. Additionally, it contains the | |
218 | * pfns that correspond to the start and end of the represented zone. | |
219 | * | |
220 | * struct bm_block contains a pointer to the memory page in which | |
0d83304c AM |
221 | * information is stored (in the form of a block of bitmap) |
222 | * It also contains the pfns that correspond to the start and end of | |
223 | * the represented memory area. | |
b788db79 RW |
224 | */ |
225 | ||
226 | #define BM_END_OF_MAP (~0UL) | |
227 | ||
b788db79 RW |
228 | #define BM_BITS_PER_BLOCK (PAGE_SIZE << 3) |
229 | ||
230 | struct bm_block { | |
846705de | 231 | struct list_head hook; /* hook into a list of bitmap blocks */ |
b788db79 RW |
232 | unsigned long start_pfn; /* pfn represented by the first bit */ |
233 | unsigned long end_pfn; /* pfn represented by the last bit plus 1 */ | |
0d83304c | 234 | unsigned long *data; /* bitmap representing pages */ |
b788db79 RW |
235 | }; |
236 | ||
0d83304c AM |
237 | static inline unsigned long bm_block_bits(struct bm_block *bb) |
238 | { | |
239 | return bb->end_pfn - bb->start_pfn; | |
240 | } | |
241 | ||
b788db79 RW |
242 | /* strcut bm_position is used for browsing memory bitmaps */ |
243 | ||
244 | struct bm_position { | |
b788db79 | 245 | struct bm_block *block; |
b788db79 RW |
246 | int bit; |
247 | }; | |
248 | ||
249 | struct memory_bitmap { | |
846705de | 250 | struct list_head blocks; /* list of bitmap blocks */ |
b788db79 RW |
251 | struct linked_page *p_list; /* list of pages used to store zone |
252 | * bitmap objects and bitmap block | |
253 | * objects | |
254 | */ | |
255 | struct bm_position cur; /* most recently used bit position */ | |
256 | }; | |
257 | ||
258 | /* Functions that operate on memory bitmaps */ | |
259 | ||
b788db79 RW |
260 | static void memory_bm_position_reset(struct memory_bitmap *bm) |
261 | { | |
846705de | 262 | bm->cur.block = list_entry(bm->blocks.next, struct bm_block, hook); |
0d83304c | 263 | bm->cur.bit = 0; |
b788db79 RW |
264 | } |
265 | ||
266 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free); | |
267 | ||
268 | /** | |
269 | * create_bm_block_list - create a list of block bitmap objects | |
846705de RW |
270 | * @nr_blocks - number of blocks to allocate |
271 | * @list - list to put the allocated blocks into | |
272 | * @ca - chain allocator to be used for allocating memory | |
b788db79 | 273 | */ |
846705de RW |
274 | static int create_bm_block_list(unsigned long pages, |
275 | struct list_head *list, | |
276 | struct chain_allocator *ca) | |
b788db79 | 277 | { |
846705de | 278 | unsigned int nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK); |
b788db79 RW |
279 | |
280 | while (nr_blocks-- > 0) { | |
281 | struct bm_block *bb; | |
282 | ||
283 | bb = chain_alloc(ca, sizeof(struct bm_block)); | |
284 | if (!bb) | |
846705de RW |
285 | return -ENOMEM; |
286 | list_add(&bb->hook, list); | |
b788db79 | 287 | } |
846705de RW |
288 | |
289 | return 0; | |
b788db79 RW |
290 | } |
291 | ||
846705de RW |
292 | struct mem_extent { |
293 | struct list_head hook; | |
294 | unsigned long start; | |
295 | unsigned long end; | |
296 | }; | |
297 | ||
b788db79 | 298 | /** |
846705de RW |
299 | * free_mem_extents - free a list of memory extents |
300 | * @list - list of extents to empty | |
b788db79 | 301 | */ |
846705de RW |
302 | static void free_mem_extents(struct list_head *list) |
303 | { | |
304 | struct mem_extent *ext, *aux; | |
b788db79 | 305 | |
846705de RW |
306 | list_for_each_entry_safe(ext, aux, list, hook) { |
307 | list_del(&ext->hook); | |
308 | kfree(ext); | |
309 | } | |
310 | } | |
311 | ||
312 | /** | |
313 | * create_mem_extents - create a list of memory extents representing | |
314 | * contiguous ranges of PFNs | |
315 | * @list - list to put the extents into | |
316 | * @gfp_mask - mask to use for memory allocations | |
317 | */ | |
318 | static int create_mem_extents(struct list_head *list, gfp_t gfp_mask) | |
b788db79 | 319 | { |
846705de | 320 | struct zone *zone; |
b788db79 | 321 | |
846705de | 322 | INIT_LIST_HEAD(list); |
b788db79 | 323 | |
846705de RW |
324 | for_each_zone(zone) { |
325 | unsigned long zone_start, zone_end; | |
326 | struct mem_extent *ext, *cur, *aux; | |
327 | ||
328 | if (!populated_zone(zone)) | |
329 | continue; | |
330 | ||
331 | zone_start = zone->zone_start_pfn; | |
332 | zone_end = zone->zone_start_pfn + zone->spanned_pages; | |
333 | ||
334 | list_for_each_entry(ext, list, hook) | |
335 | if (zone_start <= ext->end) | |
336 | break; | |
b788db79 | 337 | |
846705de RW |
338 | if (&ext->hook == list || zone_end < ext->start) { |
339 | /* New extent is necessary */ | |
340 | struct mem_extent *new_ext; | |
341 | ||
342 | new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask); | |
343 | if (!new_ext) { | |
344 | free_mem_extents(list); | |
345 | return -ENOMEM; | |
346 | } | |
347 | new_ext->start = zone_start; | |
348 | new_ext->end = zone_end; | |
349 | list_add_tail(&new_ext->hook, &ext->hook); | |
350 | continue; | |
351 | } | |
352 | ||
353 | /* Merge this zone's range of PFNs with the existing one */ | |
354 | if (zone_start < ext->start) | |
355 | ext->start = zone_start; | |
356 | if (zone_end > ext->end) | |
357 | ext->end = zone_end; | |
358 | ||
359 | /* More merging may be possible */ | |
360 | cur = ext; | |
361 | list_for_each_entry_safe_continue(cur, aux, list, hook) { | |
362 | if (zone_end < cur->start) | |
363 | break; | |
364 | if (zone_end < cur->end) | |
365 | ext->end = cur->end; | |
366 | list_del(&cur->hook); | |
367 | kfree(cur); | |
368 | } | |
b788db79 | 369 | } |
846705de RW |
370 | |
371 | return 0; | |
b788db79 RW |
372 | } |
373 | ||
374 | /** | |
375 | * memory_bm_create - allocate memory for a memory bitmap | |
376 | */ | |
b788db79 RW |
377 | static int |
378 | memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed) | |
379 | { | |
380 | struct chain_allocator ca; | |
846705de RW |
381 | struct list_head mem_extents; |
382 | struct mem_extent *ext; | |
383 | int error; | |
b788db79 RW |
384 | |
385 | chain_init(&ca, gfp_mask, safe_needed); | |
846705de | 386 | INIT_LIST_HEAD(&bm->blocks); |
b788db79 | 387 | |
846705de RW |
388 | error = create_mem_extents(&mem_extents, gfp_mask); |
389 | if (error) | |
390 | return error; | |
b788db79 | 391 | |
846705de RW |
392 | list_for_each_entry(ext, &mem_extents, hook) { |
393 | struct bm_block *bb; | |
394 | unsigned long pfn = ext->start; | |
395 | unsigned long pages = ext->end - ext->start; | |
b788db79 | 396 | |
846705de | 397 | bb = list_entry(bm->blocks.prev, struct bm_block, hook); |
b788db79 | 398 | |
846705de RW |
399 | error = create_bm_block_list(pages, bm->blocks.prev, &ca); |
400 | if (error) | |
401 | goto Error; | |
b788db79 | 402 | |
846705de RW |
403 | list_for_each_entry_continue(bb, &bm->blocks, hook) { |
404 | bb->data = get_image_page(gfp_mask, safe_needed); | |
405 | if (!bb->data) { | |
406 | error = -ENOMEM; | |
407 | goto Error; | |
408 | } | |
b788db79 RW |
409 | |
410 | bb->start_pfn = pfn; | |
846705de | 411 | if (pages >= BM_BITS_PER_BLOCK) { |
b788db79 | 412 | pfn += BM_BITS_PER_BLOCK; |
846705de | 413 | pages -= BM_BITS_PER_BLOCK; |
b788db79 RW |
414 | } else { |
415 | /* This is executed only once in the loop */ | |
846705de | 416 | pfn += pages; |
b788db79 RW |
417 | } |
418 | bb->end_pfn = pfn; | |
b788db79 | 419 | } |
b788db79 | 420 | } |
846705de | 421 | |
b788db79 RW |
422 | bm->p_list = ca.chain; |
423 | memory_bm_position_reset(bm); | |
846705de RW |
424 | Exit: |
425 | free_mem_extents(&mem_extents); | |
426 | return error; | |
b788db79 | 427 | |
846705de | 428 | Error: |
b788db79 RW |
429 | bm->p_list = ca.chain; |
430 | memory_bm_free(bm, PG_UNSAFE_CLEAR); | |
846705de | 431 | goto Exit; |
b788db79 RW |
432 | } |
433 | ||
434 | /** | |
435 | * memory_bm_free - free memory occupied by the memory bitmap @bm | |
436 | */ | |
b788db79 RW |
437 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free) |
438 | { | |
846705de | 439 | struct bm_block *bb; |
b788db79 | 440 | |
846705de RW |
441 | list_for_each_entry(bb, &bm->blocks, hook) |
442 | if (bb->data) | |
443 | free_image_page(bb->data, clear_nosave_free); | |
b788db79 | 444 | |
b788db79 | 445 | free_list_of_pages(bm->p_list, clear_nosave_free); |
846705de RW |
446 | |
447 | INIT_LIST_HEAD(&bm->blocks); | |
b788db79 RW |
448 | } |
449 | ||
450 | /** | |
74dfd666 | 451 | * memory_bm_find_bit - find the bit in the bitmap @bm that corresponds |
b788db79 RW |
452 | * to given pfn. The cur_zone_bm member of @bm and the cur_block member |
453 | * of @bm->cur_zone_bm are updated. | |
b788db79 | 454 | */ |
a82f7119 | 455 | static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn, |
74dfd666 | 456 | void **addr, unsigned int *bit_nr) |
b788db79 | 457 | { |
b788db79 RW |
458 | struct bm_block *bb; |
459 | ||
846705de RW |
460 | /* |
461 | * Check if the pfn corresponds to the current bitmap block and find | |
462 | * the block where it fits if this is not the case. | |
463 | */ | |
464 | bb = bm->cur.block; | |
b788db79 | 465 | if (pfn < bb->start_pfn) |
846705de RW |
466 | list_for_each_entry_continue_reverse(bb, &bm->blocks, hook) |
467 | if (pfn >= bb->start_pfn) | |
468 | break; | |
b788db79 | 469 | |
846705de RW |
470 | if (pfn >= bb->end_pfn) |
471 | list_for_each_entry_continue(bb, &bm->blocks, hook) | |
472 | if (pfn >= bb->start_pfn && pfn < bb->end_pfn) | |
473 | break; | |
74dfd666 | 474 | |
846705de RW |
475 | if (&bb->hook == &bm->blocks) |
476 | return -EFAULT; | |
477 | ||
478 | /* The block has been found */ | |
479 | bm->cur.block = bb; | |
b788db79 | 480 | pfn -= bb->start_pfn; |
846705de | 481 | bm->cur.bit = pfn + 1; |
0d83304c AM |
482 | *bit_nr = pfn; |
483 | *addr = bb->data; | |
a82f7119 | 484 | return 0; |
74dfd666 RW |
485 | } |
486 | ||
487 | static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn) | |
488 | { | |
489 | void *addr; | |
490 | unsigned int bit; | |
a82f7119 | 491 | int error; |
74dfd666 | 492 | |
a82f7119 RW |
493 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
494 | BUG_ON(error); | |
74dfd666 RW |
495 | set_bit(bit, addr); |
496 | } | |
497 | ||
a82f7119 RW |
498 | static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn) |
499 | { | |
500 | void *addr; | |
501 | unsigned int bit; | |
502 | int error; | |
503 | ||
504 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); | |
505 | if (!error) | |
506 | set_bit(bit, addr); | |
507 | return error; | |
508 | } | |
509 | ||
74dfd666 RW |
510 | static void memory_bm_clear_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 | clear_bit(bit, addr); |
519 | } | |
520 | ||
521 | static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn) | |
522 | { | |
523 | void *addr; | |
524 | unsigned int bit; | |
a82f7119 | 525 | int error; |
74dfd666 | 526 | |
a82f7119 RW |
527 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
528 | BUG_ON(error); | |
74dfd666 | 529 | return test_bit(bit, addr); |
b788db79 RW |
530 | } |
531 | ||
69643279 RW |
532 | static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn) |
533 | { | |
534 | void *addr; | |
535 | unsigned int bit; | |
536 | ||
537 | return !memory_bm_find_bit(bm, pfn, &addr, &bit); | |
538 | } | |
539 | ||
b788db79 RW |
540 | /** |
541 | * memory_bm_next_pfn - find the pfn that corresponds to the next set bit | |
542 | * in the bitmap @bm. If the pfn cannot be found, BM_END_OF_MAP is | |
543 | * returned. | |
544 | * | |
545 | * It is required to run memory_bm_position_reset() before the first call to | |
546 | * this function. | |
547 | */ | |
548 | ||
549 | static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm) | |
550 | { | |
b788db79 | 551 | struct bm_block *bb; |
b788db79 RW |
552 | int bit; |
553 | ||
846705de | 554 | bb = bm->cur.block; |
b788db79 | 555 | do { |
846705de RW |
556 | bit = bm->cur.bit; |
557 | bit = find_next_bit(bb->data, bm_block_bits(bb), bit); | |
558 | if (bit < bm_block_bits(bb)) | |
559 | goto Return_pfn; | |
560 | ||
561 | bb = list_entry(bb->hook.next, struct bm_block, hook); | |
562 | bm->cur.block = bb; | |
563 | bm->cur.bit = 0; | |
564 | } while (&bb->hook != &bm->blocks); | |
565 | ||
b788db79 RW |
566 | memory_bm_position_reset(bm); |
567 | return BM_END_OF_MAP; | |
568 | ||
59a49335 | 569 | Return_pfn: |
0d83304c AM |
570 | bm->cur.bit = bit + 1; |
571 | return bb->start_pfn + bit; | |
b788db79 RW |
572 | } |
573 | ||
74dfd666 RW |
574 | /** |
575 | * This structure represents a range of page frames the contents of which | |
576 | * should not be saved during the suspend. | |
577 | */ | |
578 | ||
579 | struct nosave_region { | |
580 | struct list_head list; | |
581 | unsigned long start_pfn; | |
582 | unsigned long end_pfn; | |
583 | }; | |
584 | ||
585 | static LIST_HEAD(nosave_regions); | |
586 | ||
587 | /** | |
588 | * register_nosave_region - register a range of page frames the contents | |
589 | * of which should not be saved during the suspend (to be used in the early | |
590 | * initialization code) | |
591 | */ | |
592 | ||
593 | void __init | |
940d67f6 JB |
594 | __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn, |
595 | int use_kmalloc) | |
74dfd666 RW |
596 | { |
597 | struct nosave_region *region; | |
598 | ||
599 | if (start_pfn >= end_pfn) | |
600 | return; | |
601 | ||
602 | if (!list_empty(&nosave_regions)) { | |
603 | /* Try to extend the previous region (they should be sorted) */ | |
604 | region = list_entry(nosave_regions.prev, | |
605 | struct nosave_region, list); | |
606 | if (region->end_pfn == start_pfn) { | |
607 | region->end_pfn = end_pfn; | |
608 | goto Report; | |
609 | } | |
610 | } | |
940d67f6 JB |
611 | if (use_kmalloc) { |
612 | /* during init, this shouldn't fail */ | |
613 | region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL); | |
614 | BUG_ON(!region); | |
615 | } else | |
616 | /* This allocation cannot fail */ | |
617 | region = alloc_bootmem_low(sizeof(struct nosave_region)); | |
74dfd666 RW |
618 | region->start_pfn = start_pfn; |
619 | region->end_pfn = end_pfn; | |
620 | list_add_tail(®ion->list, &nosave_regions); | |
621 | Report: | |
23976728 | 622 | printk(KERN_INFO "PM: Registered nosave memory: %016lx - %016lx\n", |
74dfd666 RW |
623 | start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT); |
624 | } | |
625 | ||
626 | /* | |
627 | * Set bits in this map correspond to the page frames the contents of which | |
628 | * should not be saved during the suspend. | |
629 | */ | |
630 | static struct memory_bitmap *forbidden_pages_map; | |
631 | ||
632 | /* Set bits in this map correspond to free page frames. */ | |
633 | static struct memory_bitmap *free_pages_map; | |
634 | ||
635 | /* | |
636 | * Each page frame allocated for creating the image is marked by setting the | |
637 | * corresponding bits in forbidden_pages_map and free_pages_map simultaneously | |
638 | */ | |
639 | ||
640 | void swsusp_set_page_free(struct page *page) | |
641 | { | |
642 | if (free_pages_map) | |
643 | memory_bm_set_bit(free_pages_map, page_to_pfn(page)); | |
644 | } | |
645 | ||
646 | static int swsusp_page_is_free(struct page *page) | |
647 | { | |
648 | return free_pages_map ? | |
649 | memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0; | |
650 | } | |
651 | ||
652 | void swsusp_unset_page_free(struct page *page) | |
653 | { | |
654 | if (free_pages_map) | |
655 | memory_bm_clear_bit(free_pages_map, page_to_pfn(page)); | |
656 | } | |
657 | ||
658 | static void swsusp_set_page_forbidden(struct page *page) | |
659 | { | |
660 | if (forbidden_pages_map) | |
661 | memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page)); | |
662 | } | |
663 | ||
664 | int swsusp_page_is_forbidden(struct page *page) | |
665 | { | |
666 | return forbidden_pages_map ? | |
667 | memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0; | |
668 | } | |
669 | ||
670 | static void swsusp_unset_page_forbidden(struct page *page) | |
671 | { | |
672 | if (forbidden_pages_map) | |
673 | memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page)); | |
674 | } | |
675 | ||
676 | /** | |
677 | * mark_nosave_pages - set bits corresponding to the page frames the | |
678 | * contents of which should not be saved in a given bitmap. | |
679 | */ | |
680 | ||
681 | static void mark_nosave_pages(struct memory_bitmap *bm) | |
682 | { | |
683 | struct nosave_region *region; | |
684 | ||
685 | if (list_empty(&nosave_regions)) | |
686 | return; | |
687 | ||
688 | list_for_each_entry(region, &nosave_regions, list) { | |
689 | unsigned long pfn; | |
690 | ||
23976728 | 691 | pr_debug("PM: Marking nosave pages: %016lx - %016lx\n", |
74dfd666 RW |
692 | region->start_pfn << PAGE_SHIFT, |
693 | region->end_pfn << PAGE_SHIFT); | |
694 | ||
695 | for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++) | |
a82f7119 RW |
696 | if (pfn_valid(pfn)) { |
697 | /* | |
698 | * It is safe to ignore the result of | |
699 | * mem_bm_set_bit_check() here, since we won't | |
700 | * touch the PFNs for which the error is | |
701 | * returned anyway. | |
702 | */ | |
703 | mem_bm_set_bit_check(bm, pfn); | |
704 | } | |
74dfd666 RW |
705 | } |
706 | } | |
707 | ||
708 | /** | |
709 | * create_basic_memory_bitmaps - create bitmaps needed for marking page | |
710 | * frames that should not be saved and free page frames. The pointers | |
711 | * forbidden_pages_map and free_pages_map are only modified if everything | |
712 | * goes well, because we don't want the bits to be used before both bitmaps | |
713 | * are set up. | |
714 | */ | |
715 | ||
716 | int create_basic_memory_bitmaps(void) | |
717 | { | |
718 | struct memory_bitmap *bm1, *bm2; | |
719 | int error = 0; | |
720 | ||
721 | BUG_ON(forbidden_pages_map || free_pages_map); | |
722 | ||
0709db60 | 723 | bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
724 | if (!bm1) |
725 | return -ENOMEM; | |
726 | ||
0709db60 | 727 | error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
728 | if (error) |
729 | goto Free_first_object; | |
730 | ||
0709db60 | 731 | bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
732 | if (!bm2) |
733 | goto Free_first_bitmap; | |
734 | ||
0709db60 | 735 | error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
736 | if (error) |
737 | goto Free_second_object; | |
738 | ||
739 | forbidden_pages_map = bm1; | |
740 | free_pages_map = bm2; | |
741 | mark_nosave_pages(forbidden_pages_map); | |
742 | ||
23976728 | 743 | pr_debug("PM: Basic memory bitmaps created\n"); |
74dfd666 RW |
744 | |
745 | return 0; | |
746 | ||
747 | Free_second_object: | |
748 | kfree(bm2); | |
749 | Free_first_bitmap: | |
750 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
751 | Free_first_object: | |
752 | kfree(bm1); | |
753 | return -ENOMEM; | |
754 | } | |
755 | ||
756 | /** | |
757 | * free_basic_memory_bitmaps - free memory bitmaps allocated by | |
758 | * create_basic_memory_bitmaps(). The auxiliary pointers are necessary | |
759 | * so that the bitmaps themselves are not referred to while they are being | |
760 | * freed. | |
761 | */ | |
762 | ||
763 | void free_basic_memory_bitmaps(void) | |
764 | { | |
765 | struct memory_bitmap *bm1, *bm2; | |
766 | ||
767 | BUG_ON(!(forbidden_pages_map && free_pages_map)); | |
768 | ||
769 | bm1 = forbidden_pages_map; | |
770 | bm2 = free_pages_map; | |
771 | forbidden_pages_map = NULL; | |
772 | free_pages_map = NULL; | |
773 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
774 | kfree(bm1); | |
775 | memory_bm_free(bm2, PG_UNSAFE_CLEAR); | |
776 | kfree(bm2); | |
777 | ||
23976728 | 778 | pr_debug("PM: Basic memory bitmaps freed\n"); |
74dfd666 RW |
779 | } |
780 | ||
b788db79 RW |
781 | /** |
782 | * snapshot_additional_pages - estimate the number of additional pages | |
783 | * be needed for setting up the suspend image data structures for given | |
784 | * zone (usually the returned value is greater than the exact number) | |
785 | */ | |
786 | ||
787 | unsigned int snapshot_additional_pages(struct zone *zone) | |
788 | { | |
789 | unsigned int res; | |
790 | ||
791 | res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK); | |
792 | res += DIV_ROUND_UP(res * sizeof(struct bm_block), PAGE_SIZE); | |
8357376d | 793 | return 2 * res; |
b788db79 RW |
794 | } |
795 | ||
8357376d RW |
796 | #ifdef CONFIG_HIGHMEM |
797 | /** | |
798 | * count_free_highmem_pages - compute the total number of free highmem | |
799 | * pages, system-wide. | |
800 | */ | |
801 | ||
802 | static unsigned int count_free_highmem_pages(void) | |
803 | { | |
804 | struct zone *zone; | |
805 | unsigned int cnt = 0; | |
806 | ||
807 | for_each_zone(zone) | |
808 | if (populated_zone(zone) && is_highmem(zone)) | |
d23ad423 | 809 | cnt += zone_page_state(zone, NR_FREE_PAGES); |
8357376d RW |
810 | |
811 | return cnt; | |
812 | } | |
813 | ||
814 | /** | |
815 | * saveable_highmem_page - Determine whether a highmem page should be | |
816 | * included in the suspend image. | |
817 | * | |
818 | * We should save the page if it isn't Nosave or NosaveFree, or Reserved, | |
819 | * and it isn't a part of a free chunk of pages. | |
820 | */ | |
846705de | 821 | static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn) |
8357376d RW |
822 | { |
823 | struct page *page; | |
824 | ||
825 | if (!pfn_valid(pfn)) | |
826 | return NULL; | |
827 | ||
828 | page = pfn_to_page(pfn); | |
846705de RW |
829 | if (page_zone(page) != zone) |
830 | return NULL; | |
8357376d RW |
831 | |
832 | BUG_ON(!PageHighMem(page)); | |
833 | ||
7be98234 RW |
834 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) || |
835 | PageReserved(page)) | |
8357376d RW |
836 | return NULL; |
837 | ||
838 | return page; | |
839 | } | |
840 | ||
841 | /** | |
842 | * count_highmem_pages - compute the total number of saveable highmem | |
843 | * pages. | |
844 | */ | |
845 | ||
846 | unsigned int count_highmem_pages(void) | |
847 | { | |
848 | struct zone *zone; | |
849 | unsigned int n = 0; | |
850 | ||
851 | for_each_zone(zone) { | |
852 | unsigned long pfn, max_zone_pfn; | |
853 | ||
854 | if (!is_highmem(zone)) | |
855 | continue; | |
856 | ||
857 | mark_free_pages(zone); | |
858 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
859 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
846705de | 860 | if (saveable_highmem_page(zone, pfn)) |
8357376d RW |
861 | n++; |
862 | } | |
863 | return n; | |
864 | } | |
865 | #else | |
846705de RW |
866 | static inline void *saveable_highmem_page(struct zone *z, unsigned long p) |
867 | { | |
868 | return NULL; | |
869 | } | |
8357376d RW |
870 | #endif /* CONFIG_HIGHMEM */ |
871 | ||
25761b6e | 872 | /** |
8a235efa RW |
873 | * saveable_page - Determine whether a non-highmem page should be included |
874 | * in the suspend image. | |
25761b6e | 875 | * |
8357376d RW |
876 | * We should save the page if it isn't Nosave, and is not in the range |
877 | * of pages statically defined as 'unsaveable', and it isn't a part of | |
878 | * a free chunk of pages. | |
25761b6e | 879 | */ |
846705de | 880 | static struct page *saveable_page(struct zone *zone, unsigned long pfn) |
25761b6e | 881 | { |
de491861 | 882 | struct page *page; |
25761b6e RW |
883 | |
884 | if (!pfn_valid(pfn)) | |
ae83c5ee | 885 | return NULL; |
25761b6e RW |
886 | |
887 | page = pfn_to_page(pfn); | |
846705de RW |
888 | if (page_zone(page) != zone) |
889 | return NULL; | |
ae83c5ee | 890 | |
8357376d RW |
891 | BUG_ON(PageHighMem(page)); |
892 | ||
7be98234 | 893 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page)) |
ae83c5ee | 894 | return NULL; |
8357376d | 895 | |
8a235efa RW |
896 | if (PageReserved(page) |
897 | && (!kernel_page_present(page) || pfn_is_nosave(pfn))) | |
ae83c5ee | 898 | return NULL; |
25761b6e | 899 | |
ae83c5ee | 900 | return page; |
25761b6e RW |
901 | } |
902 | ||
8357376d RW |
903 | /** |
904 | * count_data_pages - compute the total number of saveable non-highmem | |
905 | * pages. | |
906 | */ | |
907 | ||
72a97e08 | 908 | unsigned int count_data_pages(void) |
25761b6e RW |
909 | { |
910 | struct zone *zone; | |
ae83c5ee | 911 | unsigned long pfn, max_zone_pfn; |
dc19d507 | 912 | unsigned int n = 0; |
25761b6e | 913 | |
8357376d | 914 | for_each_zone(zone) { |
25761b6e RW |
915 | if (is_highmem(zone)) |
916 | continue; | |
8357376d | 917 | |
25761b6e | 918 | mark_free_pages(zone); |
ae83c5ee RW |
919 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
920 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
846705de | 921 | if (saveable_page(zone, pfn)) |
8357376d | 922 | n++; |
25761b6e | 923 | } |
a0f49651 | 924 | return n; |
25761b6e RW |
925 | } |
926 | ||
8357376d RW |
927 | /* This is needed, because copy_page and memcpy are not usable for copying |
928 | * task structs. | |
929 | */ | |
930 | static inline void do_copy_page(long *dst, long *src) | |
f623f0db RW |
931 | { |
932 | int n; | |
933 | ||
f623f0db RW |
934 | for (n = PAGE_SIZE / sizeof(long); n; n--) |
935 | *dst++ = *src++; | |
936 | } | |
937 | ||
8a235efa RW |
938 | |
939 | /** | |
940 | * safe_copy_page - check if the page we are going to copy is marked as | |
941 | * present in the kernel page tables (this always is the case if | |
942 | * CONFIG_DEBUG_PAGEALLOC is not set and in that case | |
943 | * kernel_page_present() always returns 'true'). | |
944 | */ | |
945 | static void safe_copy_page(void *dst, struct page *s_page) | |
946 | { | |
947 | if (kernel_page_present(s_page)) { | |
948 | do_copy_page(dst, page_address(s_page)); | |
949 | } else { | |
950 | kernel_map_pages(s_page, 1, 1); | |
951 | do_copy_page(dst, page_address(s_page)); | |
952 | kernel_map_pages(s_page, 1, 0); | |
953 | } | |
954 | } | |
955 | ||
956 | ||
8357376d RW |
957 | #ifdef CONFIG_HIGHMEM |
958 | static inline struct page * | |
959 | page_is_saveable(struct zone *zone, unsigned long pfn) | |
960 | { | |
961 | return is_highmem(zone) ? | |
846705de | 962 | saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn); |
8357376d RW |
963 | } |
964 | ||
8a235efa | 965 | static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
8357376d RW |
966 | { |
967 | struct page *s_page, *d_page; | |
968 | void *src, *dst; | |
969 | ||
970 | s_page = pfn_to_page(src_pfn); | |
971 | d_page = pfn_to_page(dst_pfn); | |
972 | if (PageHighMem(s_page)) { | |
973 | src = kmap_atomic(s_page, KM_USER0); | |
974 | dst = kmap_atomic(d_page, KM_USER1); | |
975 | do_copy_page(dst, src); | |
976 | kunmap_atomic(src, KM_USER0); | |
977 | kunmap_atomic(dst, KM_USER1); | |
978 | } else { | |
8357376d RW |
979 | if (PageHighMem(d_page)) { |
980 | /* Page pointed to by src may contain some kernel | |
981 | * data modified by kmap_atomic() | |
982 | */ | |
8a235efa | 983 | safe_copy_page(buffer, s_page); |
8357376d RW |
984 | dst = kmap_atomic(pfn_to_page(dst_pfn), KM_USER0); |
985 | memcpy(dst, buffer, PAGE_SIZE); | |
986 | kunmap_atomic(dst, KM_USER0); | |
987 | } else { | |
8a235efa | 988 | safe_copy_page(page_address(d_page), s_page); |
8357376d RW |
989 | } |
990 | } | |
991 | } | |
992 | #else | |
846705de | 993 | #define page_is_saveable(zone, pfn) saveable_page(zone, pfn) |
8357376d | 994 | |
8a235efa | 995 | static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
8357376d | 996 | { |
8a235efa RW |
997 | safe_copy_page(page_address(pfn_to_page(dst_pfn)), |
998 | pfn_to_page(src_pfn)); | |
8357376d RW |
999 | } |
1000 | #endif /* CONFIG_HIGHMEM */ | |
1001 | ||
b788db79 RW |
1002 | static void |
1003 | copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm) | |
25761b6e RW |
1004 | { |
1005 | struct zone *zone; | |
b788db79 | 1006 | unsigned long pfn; |
25761b6e | 1007 | |
8357376d | 1008 | for_each_zone(zone) { |
b788db79 RW |
1009 | unsigned long max_zone_pfn; |
1010 | ||
25761b6e | 1011 | mark_free_pages(zone); |
ae83c5ee | 1012 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
b788db79 | 1013 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
8357376d | 1014 | if (page_is_saveable(zone, pfn)) |
b788db79 | 1015 | memory_bm_set_bit(orig_bm, pfn); |
25761b6e | 1016 | } |
b788db79 RW |
1017 | memory_bm_position_reset(orig_bm); |
1018 | memory_bm_position_reset(copy_bm); | |
df7c4872 | 1019 | for(;;) { |
b788db79 | 1020 | pfn = memory_bm_next_pfn(orig_bm); |
df7c4872 FW |
1021 | if (unlikely(pfn == BM_END_OF_MAP)) |
1022 | break; | |
1023 | copy_data_page(memory_bm_next_pfn(copy_bm), pfn); | |
1024 | } | |
25761b6e RW |
1025 | } |
1026 | ||
8357376d RW |
1027 | /* Total number of image pages */ |
1028 | static unsigned int nr_copy_pages; | |
1029 | /* Number of pages needed for saving the original pfns of the image pages */ | |
1030 | static unsigned int nr_meta_pages; | |
1031 | ||
25761b6e | 1032 | /** |
940864dd | 1033 | * swsusp_free - free pages allocated for the suspend. |
cd560bb2 | 1034 | * |
940864dd RW |
1035 | * Suspend pages are alocated before the atomic copy is made, so we |
1036 | * need to release them after the resume. | |
25761b6e RW |
1037 | */ |
1038 | ||
1039 | void swsusp_free(void) | |
1040 | { | |
1041 | struct zone *zone; | |
ae83c5ee | 1042 | unsigned long pfn, max_zone_pfn; |
25761b6e RW |
1043 | |
1044 | for_each_zone(zone) { | |
ae83c5ee RW |
1045 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
1046 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1047 | if (pfn_valid(pfn)) { | |
1048 | struct page *page = pfn_to_page(pfn); | |
1049 | ||
7be98234 RW |
1050 | if (swsusp_page_is_forbidden(page) && |
1051 | swsusp_page_is_free(page)) { | |
1052 | swsusp_unset_page_forbidden(page); | |
1053 | swsusp_unset_page_free(page); | |
8357376d | 1054 | __free_page(page); |
25761b6e RW |
1055 | } |
1056 | } | |
1057 | } | |
f577eb30 RW |
1058 | nr_copy_pages = 0; |
1059 | nr_meta_pages = 0; | |
75534b50 | 1060 | restore_pblist = NULL; |
6e1819d6 | 1061 | buffer = NULL; |
25761b6e RW |
1062 | } |
1063 | ||
8357376d RW |
1064 | #ifdef CONFIG_HIGHMEM |
1065 | /** | |
1066 | * count_pages_for_highmem - compute the number of non-highmem pages | |
1067 | * that will be necessary for creating copies of highmem pages. | |
1068 | */ | |
1069 | ||
1070 | static unsigned int count_pages_for_highmem(unsigned int nr_highmem) | |
1071 | { | |
1072 | unsigned int free_highmem = count_free_highmem_pages(); | |
1073 | ||
1074 | if (free_highmem >= nr_highmem) | |
1075 | nr_highmem = 0; | |
1076 | else | |
1077 | nr_highmem -= free_highmem; | |
1078 | ||
1079 | return nr_highmem; | |
1080 | } | |
1081 | #else | |
1082 | static unsigned int | |
1083 | count_pages_for_highmem(unsigned int nr_highmem) { return 0; } | |
1084 | #endif /* CONFIG_HIGHMEM */ | |
25761b6e RW |
1085 | |
1086 | /** | |
8357376d RW |
1087 | * enough_free_mem - Make sure we have enough free memory for the |
1088 | * snapshot image. | |
25761b6e RW |
1089 | */ |
1090 | ||
8357376d | 1091 | static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem) |
25761b6e | 1092 | { |
e5e2fa78 | 1093 | struct zone *zone; |
940864dd | 1094 | unsigned int free = 0, meta = 0; |
e5e2fa78 | 1095 | |
8357376d RW |
1096 | for_each_zone(zone) { |
1097 | meta += snapshot_additional_pages(zone); | |
1098 | if (!is_highmem(zone)) | |
d23ad423 | 1099 | free += zone_page_state(zone, NR_FREE_PAGES); |
8357376d | 1100 | } |
940864dd | 1101 | |
8357376d | 1102 | nr_pages += count_pages_for_highmem(nr_highmem); |
23976728 | 1103 | pr_debug("PM: Normal pages needed: %u + %u + %u, available pages: %u\n", |
940864dd RW |
1104 | nr_pages, PAGES_FOR_IO, meta, free); |
1105 | ||
1106 | return free > nr_pages + PAGES_FOR_IO + meta; | |
25761b6e RW |
1107 | } |
1108 | ||
8357376d RW |
1109 | #ifdef CONFIG_HIGHMEM |
1110 | /** | |
1111 | * get_highmem_buffer - if there are some highmem pages in the suspend | |
1112 | * image, we may need the buffer to copy them and/or load their data. | |
1113 | */ | |
1114 | ||
1115 | static inline int get_highmem_buffer(int safe_needed) | |
1116 | { | |
1117 | buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed); | |
1118 | return buffer ? 0 : -ENOMEM; | |
1119 | } | |
1120 | ||
1121 | /** | |
1122 | * alloc_highmem_image_pages - allocate some highmem pages for the image. | |
1123 | * Try to allocate as many pages as needed, but if the number of free | |
1124 | * highmem pages is lesser than that, allocate them all. | |
1125 | */ | |
1126 | ||
1127 | static inline unsigned int | |
1128 | alloc_highmem_image_pages(struct memory_bitmap *bm, unsigned int nr_highmem) | |
1129 | { | |
1130 | unsigned int to_alloc = count_free_highmem_pages(); | |
1131 | ||
1132 | if (to_alloc > nr_highmem) | |
1133 | to_alloc = nr_highmem; | |
1134 | ||
1135 | nr_highmem -= to_alloc; | |
1136 | while (to_alloc-- > 0) { | |
1137 | struct page *page; | |
1138 | ||
1139 | page = alloc_image_page(__GFP_HIGHMEM); | |
1140 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1141 | } | |
1142 | return nr_highmem; | |
1143 | } | |
1144 | #else | |
1145 | static inline int get_highmem_buffer(int safe_needed) { return 0; } | |
1146 | ||
1147 | static inline unsigned int | |
1148 | alloc_highmem_image_pages(struct memory_bitmap *bm, unsigned int n) { return 0; } | |
1149 | #endif /* CONFIG_HIGHMEM */ | |
1150 | ||
1151 | /** | |
1152 | * swsusp_alloc - allocate memory for the suspend image | |
1153 | * | |
1154 | * We first try to allocate as many highmem pages as there are | |
1155 | * saveable highmem pages in the system. If that fails, we allocate | |
1156 | * non-highmem pages for the copies of the remaining highmem ones. | |
1157 | * | |
1158 | * In this approach it is likely that the copies of highmem pages will | |
1159 | * also be located in the high memory, because of the way in which | |
1160 | * copy_data_pages() works. | |
1161 | */ | |
1162 | ||
b788db79 RW |
1163 | static int |
1164 | swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm, | |
8357376d | 1165 | unsigned int nr_pages, unsigned int nr_highmem) |
054bd4c1 | 1166 | { |
b788db79 | 1167 | int error; |
054bd4c1 | 1168 | |
b788db79 RW |
1169 | error = memory_bm_create(orig_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY); |
1170 | if (error) | |
1171 | goto Free; | |
25761b6e | 1172 | |
b788db79 RW |
1173 | error = memory_bm_create(copy_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY); |
1174 | if (error) | |
1175 | goto Free; | |
25761b6e | 1176 | |
8357376d RW |
1177 | if (nr_highmem > 0) { |
1178 | error = get_highmem_buffer(PG_ANY); | |
1179 | if (error) | |
1180 | goto Free; | |
1181 | ||
1182 | nr_pages += alloc_highmem_image_pages(copy_bm, nr_highmem); | |
1183 | } | |
b788db79 | 1184 | while (nr_pages-- > 0) { |
8357376d RW |
1185 | struct page *page = alloc_image_page(GFP_ATOMIC | __GFP_COLD); |
1186 | ||
b788db79 RW |
1187 | if (!page) |
1188 | goto Free; | |
25761b6e | 1189 | |
b788db79 | 1190 | memory_bm_set_bit(copy_bm, page_to_pfn(page)); |
25761b6e | 1191 | } |
b788db79 | 1192 | return 0; |
25761b6e | 1193 | |
59a49335 | 1194 | Free: |
b788db79 RW |
1195 | swsusp_free(); |
1196 | return -ENOMEM; | |
25761b6e RW |
1197 | } |
1198 | ||
8357376d RW |
1199 | /* Memory bitmap used for marking saveable pages (during suspend) or the |
1200 | * suspend image pages (during resume) | |
1201 | */ | |
b788db79 | 1202 | static struct memory_bitmap orig_bm; |
8357376d RW |
1203 | /* Memory bitmap used on suspend for marking allocated pages that will contain |
1204 | * the copies of saveable pages. During resume it is initially used for | |
1205 | * marking the suspend image pages, but then its set bits are duplicated in | |
1206 | * @orig_bm and it is released. Next, on systems with high memory, it may be | |
1207 | * used for marking "safe" highmem pages, but it has to be reinitialized for | |
1208 | * this purpose. | |
b788db79 RW |
1209 | */ |
1210 | static struct memory_bitmap copy_bm; | |
1211 | ||
2e32a43e | 1212 | asmlinkage int swsusp_save(void) |
25761b6e | 1213 | { |
8357376d | 1214 | unsigned int nr_pages, nr_highmem; |
25761b6e | 1215 | |
23976728 | 1216 | printk(KERN_INFO "PM: Creating hibernation image: \n"); |
25761b6e | 1217 | |
9f8f2172 | 1218 | drain_local_pages(NULL); |
a0f49651 | 1219 | nr_pages = count_data_pages(); |
8357376d | 1220 | nr_highmem = count_highmem_pages(); |
23976728 | 1221 | printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem); |
25761b6e | 1222 | |
8357376d | 1223 | if (!enough_free_mem(nr_pages, nr_highmem)) { |
23976728 | 1224 | printk(KERN_ERR "PM: Not enough free memory\n"); |
25761b6e RW |
1225 | return -ENOMEM; |
1226 | } | |
1227 | ||
8357376d | 1228 | if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) { |
23976728 | 1229 | printk(KERN_ERR "PM: Memory allocation failed\n"); |
a0f49651 | 1230 | return -ENOMEM; |
8357376d | 1231 | } |
25761b6e RW |
1232 | |
1233 | /* During allocating of suspend pagedir, new cold pages may appear. | |
1234 | * Kill them. | |
1235 | */ | |
9f8f2172 | 1236 | drain_local_pages(NULL); |
b788db79 | 1237 | copy_data_pages(©_bm, &orig_bm); |
25761b6e RW |
1238 | |
1239 | /* | |
1240 | * End of critical section. From now on, we can write to memory, | |
1241 | * but we should not touch disk. This specially means we must _not_ | |
1242 | * touch swap space! Except we must write out our image of course. | |
1243 | */ | |
1244 | ||
8357376d | 1245 | nr_pages += nr_highmem; |
a0f49651 | 1246 | nr_copy_pages = nr_pages; |
8357376d | 1247 | nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE); |
a0f49651 | 1248 | |
23976728 RW |
1249 | printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n", |
1250 | nr_pages); | |
8357376d | 1251 | |
25761b6e RW |
1252 | return 0; |
1253 | } | |
f577eb30 | 1254 | |
d307c4a8 RW |
1255 | #ifndef CONFIG_ARCH_HIBERNATION_HEADER |
1256 | static int init_header_complete(struct swsusp_info *info) | |
f577eb30 | 1257 | { |
d307c4a8 | 1258 | memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname)); |
f577eb30 | 1259 | info->version_code = LINUX_VERSION_CODE; |
d307c4a8 RW |
1260 | return 0; |
1261 | } | |
1262 | ||
1263 | static char *check_image_kernel(struct swsusp_info *info) | |
1264 | { | |
1265 | if (info->version_code != LINUX_VERSION_CODE) | |
1266 | return "kernel version"; | |
1267 | if (strcmp(info->uts.sysname,init_utsname()->sysname)) | |
1268 | return "system type"; | |
1269 | if (strcmp(info->uts.release,init_utsname()->release)) | |
1270 | return "kernel release"; | |
1271 | if (strcmp(info->uts.version,init_utsname()->version)) | |
1272 | return "version"; | |
1273 | if (strcmp(info->uts.machine,init_utsname()->machine)) | |
1274 | return "machine"; | |
1275 | return NULL; | |
1276 | } | |
1277 | #endif /* CONFIG_ARCH_HIBERNATION_HEADER */ | |
1278 | ||
af508b34 RW |
1279 | unsigned long snapshot_get_image_size(void) |
1280 | { | |
1281 | return nr_copy_pages + nr_meta_pages + 1; | |
1282 | } | |
1283 | ||
d307c4a8 RW |
1284 | static int init_header(struct swsusp_info *info) |
1285 | { | |
1286 | memset(info, 0, sizeof(struct swsusp_info)); | |
f577eb30 | 1287 | info->num_physpages = num_physpages; |
f577eb30 | 1288 | info->image_pages = nr_copy_pages; |
af508b34 | 1289 | info->pages = snapshot_get_image_size(); |
6e1819d6 RW |
1290 | info->size = info->pages; |
1291 | info->size <<= PAGE_SHIFT; | |
d307c4a8 | 1292 | return init_header_complete(info); |
f577eb30 RW |
1293 | } |
1294 | ||
1295 | /** | |
940864dd RW |
1296 | * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm |
1297 | * are stored in the array @buf[] (1 page at a time) | |
f577eb30 RW |
1298 | */ |
1299 | ||
b788db79 | 1300 | static inline void |
940864dd | 1301 | pack_pfns(unsigned long *buf, struct memory_bitmap *bm) |
f577eb30 RW |
1302 | { |
1303 | int j; | |
1304 | ||
b788db79 | 1305 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
940864dd RW |
1306 | buf[j] = memory_bm_next_pfn(bm); |
1307 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
b788db79 | 1308 | break; |
f577eb30 | 1309 | } |
f577eb30 RW |
1310 | } |
1311 | ||
1312 | /** | |
1313 | * snapshot_read_next - used for reading the system memory snapshot. | |
1314 | * | |
1315 | * On the first call to it @handle should point to a zeroed | |
1316 | * snapshot_handle structure. The structure gets updated and a pointer | |
1317 | * to it should be passed to this function every next time. | |
1318 | * | |
1319 | * The @count parameter should contain the number of bytes the caller | |
1320 | * wants to read from the snapshot. It must not be zero. | |
1321 | * | |
1322 | * On success the function returns a positive number. Then, the caller | |
1323 | * is allowed to read up to the returned number of bytes from the memory | |
1324 | * location computed by the data_of() macro. The number returned | |
1325 | * may be smaller than @count, but this only happens if the read would | |
1326 | * cross a page boundary otherwise. | |
1327 | * | |
1328 | * The function returns 0 to indicate the end of data stream condition, | |
1329 | * and a negative number is returned on error. In such cases the | |
1330 | * structure pointed to by @handle is not updated and should not be used | |
1331 | * any more. | |
1332 | */ | |
1333 | ||
1334 | int snapshot_read_next(struct snapshot_handle *handle, size_t count) | |
1335 | { | |
fb13a28b | 1336 | if (handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 1337 | return 0; |
b788db79 | 1338 | |
f577eb30 RW |
1339 | if (!buffer) { |
1340 | /* This makes the buffer be freed by swsusp_free() */ | |
8357376d | 1341 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); |
f577eb30 RW |
1342 | if (!buffer) |
1343 | return -ENOMEM; | |
1344 | } | |
1345 | if (!handle->offset) { | |
d307c4a8 RW |
1346 | int error; |
1347 | ||
1348 | error = init_header((struct swsusp_info *)buffer); | |
1349 | if (error) | |
1350 | return error; | |
f577eb30 | 1351 | handle->buffer = buffer; |
b788db79 RW |
1352 | memory_bm_position_reset(&orig_bm); |
1353 | memory_bm_position_reset(©_bm); | |
f577eb30 | 1354 | } |
fb13a28b RW |
1355 | if (handle->prev < handle->cur) { |
1356 | if (handle->cur <= nr_meta_pages) { | |
b788db79 | 1357 | memset(buffer, 0, PAGE_SIZE); |
940864dd | 1358 | pack_pfns(buffer, &orig_bm); |
f577eb30 | 1359 | } else { |
8357376d | 1360 | struct page *page; |
b788db79 | 1361 | |
8357376d RW |
1362 | page = pfn_to_page(memory_bm_next_pfn(©_bm)); |
1363 | if (PageHighMem(page)) { | |
1364 | /* Highmem pages are copied to the buffer, | |
1365 | * because we can't return with a kmapped | |
1366 | * highmem page (we may not be called again). | |
1367 | */ | |
1368 | void *kaddr; | |
1369 | ||
1370 | kaddr = kmap_atomic(page, KM_USER0); | |
1371 | memcpy(buffer, kaddr, PAGE_SIZE); | |
1372 | kunmap_atomic(kaddr, KM_USER0); | |
1373 | handle->buffer = buffer; | |
1374 | } else { | |
1375 | handle->buffer = page_address(page); | |
1376 | } | |
f577eb30 | 1377 | } |
fb13a28b | 1378 | handle->prev = handle->cur; |
f577eb30 | 1379 | } |
fb13a28b RW |
1380 | handle->buf_offset = handle->cur_offset; |
1381 | if (handle->cur_offset + count >= PAGE_SIZE) { | |
1382 | count = PAGE_SIZE - handle->cur_offset; | |
1383 | handle->cur_offset = 0; | |
1384 | handle->cur++; | |
f577eb30 | 1385 | } else { |
fb13a28b | 1386 | handle->cur_offset += count; |
f577eb30 RW |
1387 | } |
1388 | handle->offset += count; | |
1389 | return count; | |
1390 | } | |
1391 | ||
1392 | /** | |
1393 | * mark_unsafe_pages - mark the pages that cannot be used for storing | |
1394 | * the image during resume, because they conflict with the pages that | |
1395 | * had been used before suspend | |
1396 | */ | |
1397 | ||
940864dd | 1398 | static int mark_unsafe_pages(struct memory_bitmap *bm) |
f577eb30 RW |
1399 | { |
1400 | struct zone *zone; | |
ae83c5ee | 1401 | unsigned long pfn, max_zone_pfn; |
f577eb30 RW |
1402 | |
1403 | /* Clear page flags */ | |
8357376d | 1404 | for_each_zone(zone) { |
ae83c5ee RW |
1405 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
1406 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1407 | if (pfn_valid(pfn)) | |
7be98234 | 1408 | swsusp_unset_page_free(pfn_to_page(pfn)); |
f577eb30 RW |
1409 | } |
1410 | ||
940864dd RW |
1411 | /* Mark pages that correspond to the "original" pfns as "unsafe" */ |
1412 | memory_bm_position_reset(bm); | |
1413 | do { | |
1414 | pfn = memory_bm_next_pfn(bm); | |
1415 | if (likely(pfn != BM_END_OF_MAP)) { | |
1416 | if (likely(pfn_valid(pfn))) | |
7be98234 | 1417 | swsusp_set_page_free(pfn_to_page(pfn)); |
940864dd RW |
1418 | else |
1419 | return -EFAULT; | |
1420 | } | |
1421 | } while (pfn != BM_END_OF_MAP); | |
f577eb30 | 1422 | |
940864dd | 1423 | allocated_unsafe_pages = 0; |
968808b8 | 1424 | |
f577eb30 RW |
1425 | return 0; |
1426 | } | |
1427 | ||
940864dd RW |
1428 | static void |
1429 | duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src) | |
f577eb30 | 1430 | { |
940864dd RW |
1431 | unsigned long pfn; |
1432 | ||
1433 | memory_bm_position_reset(src); | |
1434 | pfn = memory_bm_next_pfn(src); | |
1435 | while (pfn != BM_END_OF_MAP) { | |
1436 | memory_bm_set_bit(dst, pfn); | |
1437 | pfn = memory_bm_next_pfn(src); | |
f577eb30 RW |
1438 | } |
1439 | } | |
1440 | ||
d307c4a8 | 1441 | static int check_header(struct swsusp_info *info) |
f577eb30 | 1442 | { |
d307c4a8 | 1443 | char *reason; |
f577eb30 | 1444 | |
d307c4a8 RW |
1445 | reason = check_image_kernel(info); |
1446 | if (!reason && info->num_physpages != num_physpages) | |
f577eb30 | 1447 | reason = "memory size"; |
f577eb30 | 1448 | if (reason) { |
23976728 | 1449 | printk(KERN_ERR "PM: Image mismatch: %s\n", reason); |
f577eb30 RW |
1450 | return -EPERM; |
1451 | } | |
1452 | return 0; | |
1453 | } | |
1454 | ||
1455 | /** | |
1456 | * load header - check the image header and copy data from it | |
1457 | */ | |
1458 | ||
940864dd RW |
1459 | static int |
1460 | load_header(struct swsusp_info *info) | |
f577eb30 RW |
1461 | { |
1462 | int error; | |
f577eb30 | 1463 | |
940864dd | 1464 | restore_pblist = NULL; |
f577eb30 RW |
1465 | error = check_header(info); |
1466 | if (!error) { | |
f577eb30 RW |
1467 | nr_copy_pages = info->image_pages; |
1468 | nr_meta_pages = info->pages - info->image_pages - 1; | |
1469 | } | |
1470 | return error; | |
1471 | } | |
1472 | ||
1473 | /** | |
940864dd RW |
1474 | * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set |
1475 | * the corresponding bit in the memory bitmap @bm | |
f577eb30 | 1476 | */ |
69643279 | 1477 | static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm) |
f577eb30 RW |
1478 | { |
1479 | int j; | |
1480 | ||
940864dd RW |
1481 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
1482 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
1483 | break; | |
1484 | ||
69643279 RW |
1485 | if (memory_bm_pfn_present(bm, buf[j])) |
1486 | memory_bm_set_bit(bm, buf[j]); | |
1487 | else | |
1488 | return -EFAULT; | |
f577eb30 | 1489 | } |
69643279 RW |
1490 | |
1491 | return 0; | |
f577eb30 RW |
1492 | } |
1493 | ||
8357376d RW |
1494 | /* List of "safe" pages that may be used to store data loaded from the suspend |
1495 | * image | |
1496 | */ | |
1497 | static struct linked_page *safe_pages_list; | |
1498 | ||
1499 | #ifdef CONFIG_HIGHMEM | |
1500 | /* struct highmem_pbe is used for creating the list of highmem pages that | |
1501 | * should be restored atomically during the resume from disk, because the page | |
1502 | * frames they have occupied before the suspend are in use. | |
1503 | */ | |
1504 | struct highmem_pbe { | |
1505 | struct page *copy_page; /* data is here now */ | |
1506 | struct page *orig_page; /* data was here before the suspend */ | |
1507 | struct highmem_pbe *next; | |
1508 | }; | |
1509 | ||
1510 | /* List of highmem PBEs needed for restoring the highmem pages that were | |
1511 | * allocated before the suspend and included in the suspend image, but have | |
1512 | * also been allocated by the "resume" kernel, so their contents cannot be | |
1513 | * written directly to their "original" page frames. | |
1514 | */ | |
1515 | static struct highmem_pbe *highmem_pblist; | |
1516 | ||
1517 | /** | |
1518 | * count_highmem_image_pages - compute the number of highmem pages in the | |
1519 | * suspend image. The bits in the memory bitmap @bm that correspond to the | |
1520 | * image pages are assumed to be set. | |
1521 | */ | |
1522 | ||
1523 | static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) | |
1524 | { | |
1525 | unsigned long pfn; | |
1526 | unsigned int cnt = 0; | |
1527 | ||
1528 | memory_bm_position_reset(bm); | |
1529 | pfn = memory_bm_next_pfn(bm); | |
1530 | while (pfn != BM_END_OF_MAP) { | |
1531 | if (PageHighMem(pfn_to_page(pfn))) | |
1532 | cnt++; | |
1533 | ||
1534 | pfn = memory_bm_next_pfn(bm); | |
1535 | } | |
1536 | return cnt; | |
1537 | } | |
1538 | ||
1539 | /** | |
1540 | * prepare_highmem_image - try to allocate as many highmem pages as | |
1541 | * there are highmem image pages (@nr_highmem_p points to the variable | |
1542 | * containing the number of highmem image pages). The pages that are | |
1543 | * "safe" (ie. will not be overwritten when the suspend image is | |
1544 | * restored) have the corresponding bits set in @bm (it must be | |
1545 | * unitialized). | |
1546 | * | |
1547 | * NOTE: This function should not be called if there are no highmem | |
1548 | * image pages. | |
1549 | */ | |
1550 | ||
1551 | static unsigned int safe_highmem_pages; | |
1552 | ||
1553 | static struct memory_bitmap *safe_highmem_bm; | |
1554 | ||
1555 | static int | |
1556 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
1557 | { | |
1558 | unsigned int to_alloc; | |
1559 | ||
1560 | if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE)) | |
1561 | return -ENOMEM; | |
1562 | ||
1563 | if (get_highmem_buffer(PG_SAFE)) | |
1564 | return -ENOMEM; | |
1565 | ||
1566 | to_alloc = count_free_highmem_pages(); | |
1567 | if (to_alloc > *nr_highmem_p) | |
1568 | to_alloc = *nr_highmem_p; | |
1569 | else | |
1570 | *nr_highmem_p = to_alloc; | |
1571 | ||
1572 | safe_highmem_pages = 0; | |
1573 | while (to_alloc-- > 0) { | |
1574 | struct page *page; | |
1575 | ||
1576 | page = alloc_page(__GFP_HIGHMEM); | |
7be98234 | 1577 | if (!swsusp_page_is_free(page)) { |
8357376d RW |
1578 | /* The page is "safe", set its bit the bitmap */ |
1579 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1580 | safe_highmem_pages++; | |
1581 | } | |
1582 | /* Mark the page as allocated */ | |
7be98234 RW |
1583 | swsusp_set_page_forbidden(page); |
1584 | swsusp_set_page_free(page); | |
8357376d RW |
1585 | } |
1586 | memory_bm_position_reset(bm); | |
1587 | safe_highmem_bm = bm; | |
1588 | return 0; | |
1589 | } | |
1590 | ||
1591 | /** | |
1592 | * get_highmem_page_buffer - for given highmem image page find the buffer | |
1593 | * that suspend_write_next() should set for its caller to write to. | |
1594 | * | |
1595 | * If the page is to be saved to its "original" page frame or a copy of | |
1596 | * the page is to be made in the highmem, @buffer is returned. Otherwise, | |
1597 | * the copy of the page is to be made in normal memory, so the address of | |
1598 | * the copy is returned. | |
1599 | * | |
1600 | * If @buffer is returned, the caller of suspend_write_next() will write | |
1601 | * the page's contents to @buffer, so they will have to be copied to the | |
1602 | * right location on the next call to suspend_write_next() and it is done | |
1603 | * with the help of copy_last_highmem_page(). For this purpose, if | |
1604 | * @buffer is returned, @last_highmem page is set to the page to which | |
1605 | * the data will have to be copied from @buffer. | |
1606 | */ | |
1607 | ||
1608 | static struct page *last_highmem_page; | |
1609 | ||
1610 | static void * | |
1611 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
1612 | { | |
1613 | struct highmem_pbe *pbe; | |
1614 | void *kaddr; | |
1615 | ||
7be98234 | 1616 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) { |
8357376d RW |
1617 | /* We have allocated the "original" page frame and we can |
1618 | * use it directly to store the loaded page. | |
1619 | */ | |
1620 | last_highmem_page = page; | |
1621 | return buffer; | |
1622 | } | |
1623 | /* The "original" page frame has not been allocated and we have to | |
1624 | * use a "safe" page frame to store the loaded page. | |
1625 | */ | |
1626 | pbe = chain_alloc(ca, sizeof(struct highmem_pbe)); | |
1627 | if (!pbe) { | |
1628 | swsusp_free(); | |
69643279 | 1629 | return ERR_PTR(-ENOMEM); |
8357376d RW |
1630 | } |
1631 | pbe->orig_page = page; | |
1632 | if (safe_highmem_pages > 0) { | |
1633 | struct page *tmp; | |
1634 | ||
1635 | /* Copy of the page will be stored in high memory */ | |
1636 | kaddr = buffer; | |
1637 | tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm)); | |
1638 | safe_highmem_pages--; | |
1639 | last_highmem_page = tmp; | |
1640 | pbe->copy_page = tmp; | |
1641 | } else { | |
1642 | /* Copy of the page will be stored in normal memory */ | |
1643 | kaddr = safe_pages_list; | |
1644 | safe_pages_list = safe_pages_list->next; | |
1645 | pbe->copy_page = virt_to_page(kaddr); | |
1646 | } | |
1647 | pbe->next = highmem_pblist; | |
1648 | highmem_pblist = pbe; | |
1649 | return kaddr; | |
1650 | } | |
1651 | ||
1652 | /** | |
1653 | * copy_last_highmem_page - copy the contents of a highmem image from | |
1654 | * @buffer, where the caller of snapshot_write_next() has place them, | |
1655 | * to the right location represented by @last_highmem_page . | |
1656 | */ | |
1657 | ||
1658 | static void copy_last_highmem_page(void) | |
1659 | { | |
1660 | if (last_highmem_page) { | |
1661 | void *dst; | |
1662 | ||
1663 | dst = kmap_atomic(last_highmem_page, KM_USER0); | |
1664 | memcpy(dst, buffer, PAGE_SIZE); | |
1665 | kunmap_atomic(dst, KM_USER0); | |
1666 | last_highmem_page = NULL; | |
1667 | } | |
1668 | } | |
1669 | ||
1670 | static inline int last_highmem_page_copied(void) | |
1671 | { | |
1672 | return !last_highmem_page; | |
1673 | } | |
1674 | ||
1675 | static inline void free_highmem_data(void) | |
1676 | { | |
1677 | if (safe_highmem_bm) | |
1678 | memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR); | |
1679 | ||
1680 | if (buffer) | |
1681 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
1682 | } | |
1683 | #else | |
1684 | static inline int get_safe_write_buffer(void) { return 0; } | |
1685 | ||
1686 | static unsigned int | |
1687 | count_highmem_image_pages(struct memory_bitmap *bm) { return 0; } | |
1688 | ||
1689 | static inline int | |
1690 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
1691 | { | |
1692 | return 0; | |
1693 | } | |
1694 | ||
1695 | static inline void * | |
1696 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
1697 | { | |
69643279 | 1698 | return ERR_PTR(-EINVAL); |
8357376d RW |
1699 | } |
1700 | ||
1701 | static inline void copy_last_highmem_page(void) {} | |
1702 | static inline int last_highmem_page_copied(void) { return 1; } | |
1703 | static inline void free_highmem_data(void) {} | |
1704 | #endif /* CONFIG_HIGHMEM */ | |
1705 | ||
f577eb30 | 1706 | /** |
940864dd RW |
1707 | * prepare_image - use the memory bitmap @bm to mark the pages that will |
1708 | * be overwritten in the process of restoring the system memory state | |
1709 | * from the suspend image ("unsafe" pages) and allocate memory for the | |
1710 | * image. | |
968808b8 | 1711 | * |
940864dd RW |
1712 | * The idea is to allocate a new memory bitmap first and then allocate |
1713 | * as many pages as needed for the image data, but not to assign these | |
1714 | * pages to specific tasks initially. Instead, we just mark them as | |
8357376d RW |
1715 | * allocated and create a lists of "safe" pages that will be used |
1716 | * later. On systems with high memory a list of "safe" highmem pages is | |
1717 | * also created. | |
f577eb30 RW |
1718 | */ |
1719 | ||
940864dd RW |
1720 | #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe)) |
1721 | ||
940864dd RW |
1722 | static int |
1723 | prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm) | |
f577eb30 | 1724 | { |
8357376d | 1725 | unsigned int nr_pages, nr_highmem; |
940864dd RW |
1726 | struct linked_page *sp_list, *lp; |
1727 | int error; | |
f577eb30 | 1728 | |
8357376d RW |
1729 | /* If there is no highmem, the buffer will not be necessary */ |
1730 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
1731 | buffer = NULL; | |
1732 | ||
1733 | nr_highmem = count_highmem_image_pages(bm); | |
940864dd RW |
1734 | error = mark_unsafe_pages(bm); |
1735 | if (error) | |
1736 | goto Free; | |
1737 | ||
1738 | error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE); | |
1739 | if (error) | |
1740 | goto Free; | |
1741 | ||
1742 | duplicate_memory_bitmap(new_bm, bm); | |
1743 | memory_bm_free(bm, PG_UNSAFE_KEEP); | |
8357376d RW |
1744 | if (nr_highmem > 0) { |
1745 | error = prepare_highmem_image(bm, &nr_highmem); | |
1746 | if (error) | |
1747 | goto Free; | |
1748 | } | |
940864dd RW |
1749 | /* Reserve some safe pages for potential later use. |
1750 | * | |
1751 | * NOTE: This way we make sure there will be enough safe pages for the | |
1752 | * chain_alloc() in get_buffer(). It is a bit wasteful, but | |
1753 | * nr_copy_pages cannot be greater than 50% of the memory anyway. | |
1754 | */ | |
1755 | sp_list = NULL; | |
1756 | /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */ | |
8357376d | 1757 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
1758 | nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE); |
1759 | while (nr_pages > 0) { | |
8357376d | 1760 | lp = get_image_page(GFP_ATOMIC, PG_SAFE); |
940864dd | 1761 | if (!lp) { |
f577eb30 | 1762 | error = -ENOMEM; |
940864dd RW |
1763 | goto Free; |
1764 | } | |
1765 | lp->next = sp_list; | |
1766 | sp_list = lp; | |
1767 | nr_pages--; | |
f577eb30 | 1768 | } |
940864dd RW |
1769 | /* Preallocate memory for the image */ |
1770 | safe_pages_list = NULL; | |
8357376d | 1771 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
1772 | while (nr_pages > 0) { |
1773 | lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC); | |
1774 | if (!lp) { | |
1775 | error = -ENOMEM; | |
1776 | goto Free; | |
1777 | } | |
7be98234 | 1778 | if (!swsusp_page_is_free(virt_to_page(lp))) { |
940864dd RW |
1779 | /* The page is "safe", add it to the list */ |
1780 | lp->next = safe_pages_list; | |
1781 | safe_pages_list = lp; | |
968808b8 | 1782 | } |
940864dd | 1783 | /* Mark the page as allocated */ |
7be98234 RW |
1784 | swsusp_set_page_forbidden(virt_to_page(lp)); |
1785 | swsusp_set_page_free(virt_to_page(lp)); | |
940864dd | 1786 | nr_pages--; |
968808b8 | 1787 | } |
940864dd RW |
1788 | /* Free the reserved safe pages so that chain_alloc() can use them */ |
1789 | while (sp_list) { | |
1790 | lp = sp_list->next; | |
1791 | free_image_page(sp_list, PG_UNSAFE_CLEAR); | |
1792 | sp_list = lp; | |
f577eb30 | 1793 | } |
940864dd RW |
1794 | return 0; |
1795 | ||
59a49335 | 1796 | Free: |
940864dd | 1797 | swsusp_free(); |
f577eb30 RW |
1798 | return error; |
1799 | } | |
1800 | ||
940864dd RW |
1801 | /** |
1802 | * get_buffer - compute the address that snapshot_write_next() should | |
1803 | * set for its caller to write to. | |
1804 | */ | |
1805 | ||
1806 | static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca) | |
968808b8 | 1807 | { |
940864dd | 1808 | struct pbe *pbe; |
69643279 RW |
1809 | struct page *page; |
1810 | unsigned long pfn = memory_bm_next_pfn(bm); | |
968808b8 | 1811 | |
69643279 RW |
1812 | if (pfn == BM_END_OF_MAP) |
1813 | return ERR_PTR(-EFAULT); | |
1814 | ||
1815 | page = pfn_to_page(pfn); | |
8357376d RW |
1816 | if (PageHighMem(page)) |
1817 | return get_highmem_page_buffer(page, ca); | |
1818 | ||
7be98234 | 1819 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) |
940864dd RW |
1820 | /* We have allocated the "original" page frame and we can |
1821 | * use it directly to store the loaded page. | |
968808b8 | 1822 | */ |
940864dd RW |
1823 | return page_address(page); |
1824 | ||
1825 | /* The "original" page frame has not been allocated and we have to | |
1826 | * use a "safe" page frame to store the loaded page. | |
968808b8 | 1827 | */ |
940864dd RW |
1828 | pbe = chain_alloc(ca, sizeof(struct pbe)); |
1829 | if (!pbe) { | |
1830 | swsusp_free(); | |
69643279 | 1831 | return ERR_PTR(-ENOMEM); |
940864dd | 1832 | } |
8357376d RW |
1833 | pbe->orig_address = page_address(page); |
1834 | pbe->address = safe_pages_list; | |
940864dd RW |
1835 | safe_pages_list = safe_pages_list->next; |
1836 | pbe->next = restore_pblist; | |
1837 | restore_pblist = pbe; | |
8357376d | 1838 | return pbe->address; |
968808b8 RW |
1839 | } |
1840 | ||
f577eb30 RW |
1841 | /** |
1842 | * snapshot_write_next - used for writing the system memory snapshot. | |
1843 | * | |
1844 | * On the first call to it @handle should point to a zeroed | |
1845 | * snapshot_handle structure. The structure gets updated and a pointer | |
1846 | * to it should be passed to this function every next time. | |
1847 | * | |
1848 | * The @count parameter should contain the number of bytes the caller | |
1849 | * wants to write to the image. It must not be zero. | |
1850 | * | |
1851 | * On success the function returns a positive number. Then, the caller | |
1852 | * is allowed to write up to the returned number of bytes to the memory | |
1853 | * location computed by the data_of() macro. The number returned | |
1854 | * may be smaller than @count, but this only happens if the write would | |
1855 | * cross a page boundary otherwise. | |
1856 | * | |
1857 | * The function returns 0 to indicate the "end of file" condition, | |
1858 | * and a negative number is returned on error. In such cases the | |
1859 | * structure pointed to by @handle is not updated and should not be used | |
1860 | * any more. | |
1861 | */ | |
1862 | ||
1863 | int snapshot_write_next(struct snapshot_handle *handle, size_t count) | |
1864 | { | |
940864dd | 1865 | static struct chain_allocator ca; |
f577eb30 RW |
1866 | int error = 0; |
1867 | ||
940864dd | 1868 | /* Check if we have already loaded the entire image */ |
fb13a28b | 1869 | if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 1870 | return 0; |
940864dd | 1871 | |
8357376d RW |
1872 | if (handle->offset == 0) { |
1873 | if (!buffer) | |
1874 | /* This makes the buffer be freed by swsusp_free() */ | |
1875 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); | |
1876 | ||
f577eb30 RW |
1877 | if (!buffer) |
1878 | return -ENOMEM; | |
8357376d | 1879 | |
f577eb30 | 1880 | handle->buffer = buffer; |
8357376d | 1881 | } |
546e0d27 | 1882 | handle->sync_read = 1; |
fb13a28b | 1883 | if (handle->prev < handle->cur) { |
940864dd RW |
1884 | if (handle->prev == 0) { |
1885 | error = load_header(buffer); | |
1886 | if (error) | |
1887 | return error; | |
1888 | ||
1889 | error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY); | |
f577eb30 RW |
1890 | if (error) |
1891 | return error; | |
940864dd | 1892 | |
f577eb30 | 1893 | } else if (handle->prev <= nr_meta_pages) { |
69643279 RW |
1894 | error = unpack_orig_pfns(buffer, ©_bm); |
1895 | if (error) | |
1896 | return error; | |
1897 | ||
940864dd RW |
1898 | if (handle->prev == nr_meta_pages) { |
1899 | error = prepare_image(&orig_bm, ©_bm); | |
f577eb30 RW |
1900 | if (error) |
1901 | return error; | |
940864dd RW |
1902 | |
1903 | chain_init(&ca, GFP_ATOMIC, PG_SAFE); | |
1904 | memory_bm_position_reset(&orig_bm); | |
1905 | restore_pblist = NULL; | |
1906 | handle->buffer = get_buffer(&orig_bm, &ca); | |
546e0d27 | 1907 | handle->sync_read = 0; |
69643279 RW |
1908 | if (IS_ERR(handle->buffer)) |
1909 | return PTR_ERR(handle->buffer); | |
f577eb30 RW |
1910 | } |
1911 | } else { | |
8357376d | 1912 | copy_last_highmem_page(); |
940864dd | 1913 | handle->buffer = get_buffer(&orig_bm, &ca); |
69643279 RW |
1914 | if (IS_ERR(handle->buffer)) |
1915 | return PTR_ERR(handle->buffer); | |
8357376d RW |
1916 | if (handle->buffer != buffer) |
1917 | handle->sync_read = 0; | |
f577eb30 | 1918 | } |
fb13a28b | 1919 | handle->prev = handle->cur; |
f577eb30 | 1920 | } |
fb13a28b RW |
1921 | handle->buf_offset = handle->cur_offset; |
1922 | if (handle->cur_offset + count >= PAGE_SIZE) { | |
1923 | count = PAGE_SIZE - handle->cur_offset; | |
1924 | handle->cur_offset = 0; | |
1925 | handle->cur++; | |
f577eb30 | 1926 | } else { |
fb13a28b | 1927 | handle->cur_offset += count; |
f577eb30 RW |
1928 | } |
1929 | handle->offset += count; | |
1930 | return count; | |
1931 | } | |
1932 | ||
8357376d RW |
1933 | /** |
1934 | * snapshot_write_finalize - must be called after the last call to | |
1935 | * snapshot_write_next() in case the last page in the image happens | |
1936 | * to be a highmem page and its contents should be stored in the | |
1937 | * highmem. Additionally, it releases the memory that will not be | |
1938 | * used any more. | |
1939 | */ | |
1940 | ||
1941 | void snapshot_write_finalize(struct snapshot_handle *handle) | |
1942 | { | |
1943 | copy_last_highmem_page(); | |
1944 | /* Free only if we have loaded the image entirely */ | |
1945 | if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages) { | |
1946 | memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR); | |
1947 | free_highmem_data(); | |
1948 | } | |
1949 | } | |
1950 | ||
f577eb30 RW |
1951 | int snapshot_image_loaded(struct snapshot_handle *handle) |
1952 | { | |
8357376d | 1953 | return !(!nr_copy_pages || !last_highmem_page_copied() || |
940864dd RW |
1954 | handle->cur <= nr_meta_pages + nr_copy_pages); |
1955 | } | |
1956 | ||
8357376d RW |
1957 | #ifdef CONFIG_HIGHMEM |
1958 | /* Assumes that @buf is ready and points to a "safe" page */ | |
1959 | static inline void | |
1960 | swap_two_pages_data(struct page *p1, struct page *p2, void *buf) | |
940864dd | 1961 | { |
8357376d RW |
1962 | void *kaddr1, *kaddr2; |
1963 | ||
1964 | kaddr1 = kmap_atomic(p1, KM_USER0); | |
1965 | kaddr2 = kmap_atomic(p2, KM_USER1); | |
1966 | memcpy(buf, kaddr1, PAGE_SIZE); | |
1967 | memcpy(kaddr1, kaddr2, PAGE_SIZE); | |
1968 | memcpy(kaddr2, buf, PAGE_SIZE); | |
1969 | kunmap_atomic(kaddr1, KM_USER0); | |
1970 | kunmap_atomic(kaddr2, KM_USER1); | |
1971 | } | |
1972 | ||
1973 | /** | |
1974 | * restore_highmem - for each highmem page that was allocated before | |
1975 | * the suspend and included in the suspend image, and also has been | |
1976 | * allocated by the "resume" kernel swap its current (ie. "before | |
1977 | * resume") contents with the previous (ie. "before suspend") one. | |
1978 | * | |
1979 | * If the resume eventually fails, we can call this function once | |
1980 | * again and restore the "before resume" highmem state. | |
1981 | */ | |
1982 | ||
1983 | int restore_highmem(void) | |
1984 | { | |
1985 | struct highmem_pbe *pbe = highmem_pblist; | |
1986 | void *buf; | |
1987 | ||
1988 | if (!pbe) | |
1989 | return 0; | |
1990 | ||
1991 | buf = get_image_page(GFP_ATOMIC, PG_SAFE); | |
1992 | if (!buf) | |
1993 | return -ENOMEM; | |
1994 | ||
1995 | while (pbe) { | |
1996 | swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf); | |
1997 | pbe = pbe->next; | |
1998 | } | |
1999 | free_image_page(buf, PG_UNSAFE_CLEAR); | |
2000 | return 0; | |
f577eb30 | 2001 | } |
8357376d | 2002 | #endif /* CONFIG_HIGHMEM */ |