Commit | Line | Data |
---|---|---|
25761b6e | 1 | /* |
96bc7aec | 2 | * linux/kernel/power/snapshot.c |
25761b6e | 3 | * |
8357376d | 4 | * This file provides system snapshot/restore functionality for swsusp. |
25761b6e RW |
5 | * |
6 | * Copyright (C) 1998-2005 Pavel Machek <pavel@suse.cz> | |
8357376d | 7 | * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl> |
25761b6e | 8 | * |
8357376d | 9 | * This file is released under the GPLv2. |
25761b6e RW |
10 | * |
11 | */ | |
12 | ||
f577eb30 | 13 | #include <linux/version.h> |
25761b6e RW |
14 | #include <linux/module.h> |
15 | #include <linux/mm.h> | |
16 | #include <linux/suspend.h> | |
25761b6e | 17 | #include <linux/delay.h> |
25761b6e | 18 | #include <linux/bitops.h> |
25761b6e | 19 | #include <linux/spinlock.h> |
25761b6e | 20 | #include <linux/kernel.h> |
25761b6e RW |
21 | #include <linux/pm.h> |
22 | #include <linux/device.h> | |
74dfd666 | 23 | #include <linux/init.h> |
25761b6e RW |
24 | #include <linux/bootmem.h> |
25 | #include <linux/syscalls.h> | |
26 | #include <linux/console.h> | |
27 | #include <linux/highmem.h> | |
25761b6e RW |
28 | |
29 | #include <asm/uaccess.h> | |
30 | #include <asm/mmu_context.h> | |
31 | #include <asm/pgtable.h> | |
32 | #include <asm/tlbflush.h> | |
33 | #include <asm/io.h> | |
34 | ||
25761b6e RW |
35 | #include "power.h" |
36 | ||
74dfd666 RW |
37 | static int swsusp_page_is_free(struct page *); |
38 | static void swsusp_set_page_forbidden(struct page *); | |
39 | static void swsusp_unset_page_forbidden(struct page *); | |
40 | ||
8357376d RW |
41 | /* List of PBEs needed for restoring the pages that were allocated before |
42 | * the suspend and included in the suspend image, but have also been | |
43 | * allocated by the "resume" kernel, so their contents cannot be written | |
44 | * directly to their "original" page frames. | |
45 | */ | |
75534b50 RW |
46 | struct pbe *restore_pblist; |
47 | ||
8357376d | 48 | /* Pointer to an auxiliary buffer (1 page) */ |
940864dd | 49 | static void *buffer; |
7088a5c0 | 50 | |
f6143aa6 RW |
51 | /** |
52 | * @safe_needed - on resume, for storing the PBE list and the image, | |
53 | * we can only use memory pages that do not conflict with the pages | |
8357376d RW |
54 | * used before suspend. The unsafe pages have PageNosaveFree set |
55 | * and we count them using unsafe_pages. | |
f6143aa6 | 56 | * |
8357376d RW |
57 | * Each allocated image page is marked as PageNosave and PageNosaveFree |
58 | * so that swsusp_free() can release it. | |
f6143aa6 RW |
59 | */ |
60 | ||
0bcd888d RW |
61 | #define PG_ANY 0 |
62 | #define PG_SAFE 1 | |
63 | #define PG_UNSAFE_CLEAR 1 | |
64 | #define PG_UNSAFE_KEEP 0 | |
65 | ||
940864dd | 66 | static unsigned int allocated_unsafe_pages; |
f6143aa6 | 67 | |
8357376d | 68 | static void *get_image_page(gfp_t gfp_mask, int safe_needed) |
f6143aa6 RW |
69 | { |
70 | void *res; | |
71 | ||
72 | res = (void *)get_zeroed_page(gfp_mask); | |
73 | if (safe_needed) | |
7be98234 | 74 | while (res && swsusp_page_is_free(virt_to_page(res))) { |
f6143aa6 | 75 | /* The page is unsafe, mark it for swsusp_free() */ |
7be98234 | 76 | swsusp_set_page_forbidden(virt_to_page(res)); |
940864dd | 77 | allocated_unsafe_pages++; |
f6143aa6 RW |
78 | res = (void *)get_zeroed_page(gfp_mask); |
79 | } | |
80 | if (res) { | |
7be98234 RW |
81 | swsusp_set_page_forbidden(virt_to_page(res)); |
82 | swsusp_set_page_free(virt_to_page(res)); | |
f6143aa6 RW |
83 | } |
84 | return res; | |
85 | } | |
86 | ||
87 | unsigned long get_safe_page(gfp_t gfp_mask) | |
88 | { | |
8357376d RW |
89 | return (unsigned long)get_image_page(gfp_mask, PG_SAFE); |
90 | } | |
91 | ||
5b6d15de RW |
92 | static struct page *alloc_image_page(gfp_t gfp_mask) |
93 | { | |
8357376d RW |
94 | struct page *page; |
95 | ||
96 | page = alloc_page(gfp_mask); | |
97 | if (page) { | |
7be98234 RW |
98 | swsusp_set_page_forbidden(page); |
99 | swsusp_set_page_free(page); | |
8357376d RW |
100 | } |
101 | return page; | |
f6143aa6 RW |
102 | } |
103 | ||
104 | /** | |
105 | * free_image_page - free page represented by @addr, allocated with | |
8357376d | 106 | * get_image_page (page flags set by it must be cleared) |
f6143aa6 RW |
107 | */ |
108 | ||
109 | static inline void free_image_page(void *addr, int clear_nosave_free) | |
110 | { | |
8357376d RW |
111 | struct page *page; |
112 | ||
113 | BUG_ON(!virt_addr_valid(addr)); | |
114 | ||
115 | page = virt_to_page(addr); | |
116 | ||
7be98234 | 117 | swsusp_unset_page_forbidden(page); |
f6143aa6 | 118 | if (clear_nosave_free) |
7be98234 | 119 | swsusp_unset_page_free(page); |
8357376d RW |
120 | |
121 | __free_page(page); | |
f6143aa6 RW |
122 | } |
123 | ||
b788db79 RW |
124 | /* struct linked_page is used to build chains of pages */ |
125 | ||
126 | #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *)) | |
127 | ||
128 | struct linked_page { | |
129 | struct linked_page *next; | |
130 | char data[LINKED_PAGE_DATA_SIZE]; | |
131 | } __attribute__((packed)); | |
132 | ||
133 | static inline void | |
134 | free_list_of_pages(struct linked_page *list, int clear_page_nosave) | |
135 | { | |
136 | while (list) { | |
137 | struct linked_page *lp = list->next; | |
138 | ||
139 | free_image_page(list, clear_page_nosave); | |
140 | list = lp; | |
141 | } | |
142 | } | |
143 | ||
144 | /** | |
145 | * struct chain_allocator is used for allocating small objects out of | |
146 | * a linked list of pages called 'the chain'. | |
147 | * | |
148 | * The chain grows each time when there is no room for a new object in | |
149 | * the current page. The allocated objects cannot be freed individually. | |
150 | * It is only possible to free them all at once, by freeing the entire | |
151 | * chain. | |
152 | * | |
153 | * NOTE: The chain allocator may be inefficient if the allocated objects | |
154 | * are not much smaller than PAGE_SIZE. | |
155 | */ | |
156 | ||
157 | struct chain_allocator { | |
158 | struct linked_page *chain; /* the chain */ | |
159 | unsigned int used_space; /* total size of objects allocated out | |
160 | * of the current page | |
161 | */ | |
162 | gfp_t gfp_mask; /* mask for allocating pages */ | |
163 | int safe_needed; /* if set, only "safe" pages are allocated */ | |
164 | }; | |
165 | ||
166 | static void | |
167 | chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed) | |
168 | { | |
169 | ca->chain = NULL; | |
170 | ca->used_space = LINKED_PAGE_DATA_SIZE; | |
171 | ca->gfp_mask = gfp_mask; | |
172 | ca->safe_needed = safe_needed; | |
173 | } | |
174 | ||
175 | static void *chain_alloc(struct chain_allocator *ca, unsigned int size) | |
176 | { | |
177 | void *ret; | |
178 | ||
179 | if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) { | |
180 | struct linked_page *lp; | |
181 | ||
8357376d | 182 | lp = get_image_page(ca->gfp_mask, ca->safe_needed); |
b788db79 RW |
183 | if (!lp) |
184 | return NULL; | |
185 | ||
186 | lp->next = ca->chain; | |
187 | ca->chain = lp; | |
188 | ca->used_space = 0; | |
189 | } | |
190 | ret = ca->chain->data + ca->used_space; | |
191 | ca->used_space += size; | |
192 | return ret; | |
193 | } | |
194 | ||
195 | static void chain_free(struct chain_allocator *ca, int clear_page_nosave) | |
196 | { | |
197 | free_list_of_pages(ca->chain, clear_page_nosave); | |
198 | memset(ca, 0, sizeof(struct chain_allocator)); | |
199 | } | |
200 | ||
201 | /** | |
202 | * Data types related to memory bitmaps. | |
203 | * | |
204 | * Memory bitmap is a structure consiting of many linked lists of | |
205 | * objects. The main list's elements are of type struct zone_bitmap | |
206 | * and each of them corresonds to one zone. For each zone bitmap | |
207 | * object there is a list of objects of type struct bm_block that | |
208 | * represent each blocks of bit chunks in which information is | |
209 | * stored. | |
210 | * | |
211 | * struct memory_bitmap contains a pointer to the main list of zone | |
212 | * bitmap objects, a struct bm_position used for browsing the bitmap, | |
213 | * and a pointer to the list of pages used for allocating all of the | |
214 | * zone bitmap objects and bitmap block objects. | |
215 | * | |
216 | * NOTE: It has to be possible to lay out the bitmap in memory | |
217 | * using only allocations of order 0. Additionally, the bitmap is | |
218 | * designed to work with arbitrary number of zones (this is over the | |
219 | * top for now, but let's avoid making unnecessary assumptions ;-). | |
220 | * | |
221 | * struct zone_bitmap contains a pointer to a list of bitmap block | |
222 | * objects and a pointer to the bitmap block object that has been | |
223 | * most recently used for setting bits. Additionally, it contains the | |
224 | * pfns that correspond to the start and end of the represented zone. | |
225 | * | |
226 | * struct bm_block contains a pointer to the memory page in which | |
227 | * information is stored (in the form of a block of bit chunks | |
228 | * of type unsigned long each). It also contains the pfns that | |
229 | * correspond to the start and end of the represented memory area and | |
230 | * the number of bit chunks in the block. | |
b788db79 RW |
231 | */ |
232 | ||
233 | #define BM_END_OF_MAP (~0UL) | |
234 | ||
235 | #define BM_CHUNKS_PER_BLOCK (PAGE_SIZE / sizeof(long)) | |
236 | #define BM_BITS_PER_CHUNK (sizeof(long) << 3) | |
237 | #define BM_BITS_PER_BLOCK (PAGE_SIZE << 3) | |
238 | ||
239 | struct bm_block { | |
240 | struct bm_block *next; /* next element of the list */ | |
241 | unsigned long start_pfn; /* pfn represented by the first bit */ | |
242 | unsigned long end_pfn; /* pfn represented by the last bit plus 1 */ | |
243 | unsigned int size; /* number of bit chunks */ | |
244 | unsigned long *data; /* chunks of bits representing pages */ | |
245 | }; | |
246 | ||
247 | struct zone_bitmap { | |
248 | struct zone_bitmap *next; /* next element of the list */ | |
249 | unsigned long start_pfn; /* minimal pfn in this zone */ | |
250 | unsigned long end_pfn; /* maximal pfn in this zone plus 1 */ | |
251 | struct bm_block *bm_blocks; /* list of bitmap blocks */ | |
252 | struct bm_block *cur_block; /* recently used bitmap block */ | |
253 | }; | |
254 | ||
255 | /* strcut bm_position is used for browsing memory bitmaps */ | |
256 | ||
257 | struct bm_position { | |
258 | struct zone_bitmap *zone_bm; | |
259 | struct bm_block *block; | |
260 | int chunk; | |
261 | int bit; | |
262 | }; | |
263 | ||
264 | struct memory_bitmap { | |
265 | struct zone_bitmap *zone_bm_list; /* list of zone bitmaps */ | |
266 | struct linked_page *p_list; /* list of pages used to store zone | |
267 | * bitmap objects and bitmap block | |
268 | * objects | |
269 | */ | |
270 | struct bm_position cur; /* most recently used bit position */ | |
271 | }; | |
272 | ||
273 | /* Functions that operate on memory bitmaps */ | |
274 | ||
275 | static inline void memory_bm_reset_chunk(struct memory_bitmap *bm) | |
276 | { | |
277 | bm->cur.chunk = 0; | |
278 | bm->cur.bit = -1; | |
279 | } | |
280 | ||
281 | static void memory_bm_position_reset(struct memory_bitmap *bm) | |
282 | { | |
283 | struct zone_bitmap *zone_bm; | |
284 | ||
285 | zone_bm = bm->zone_bm_list; | |
286 | bm->cur.zone_bm = zone_bm; | |
287 | bm->cur.block = zone_bm->bm_blocks; | |
288 | memory_bm_reset_chunk(bm); | |
289 | } | |
290 | ||
291 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free); | |
292 | ||
293 | /** | |
294 | * create_bm_block_list - create a list of block bitmap objects | |
295 | */ | |
296 | ||
297 | static inline struct bm_block * | |
298 | create_bm_block_list(unsigned int nr_blocks, struct chain_allocator *ca) | |
299 | { | |
300 | struct bm_block *bblist = NULL; | |
301 | ||
302 | while (nr_blocks-- > 0) { | |
303 | struct bm_block *bb; | |
304 | ||
305 | bb = chain_alloc(ca, sizeof(struct bm_block)); | |
306 | if (!bb) | |
307 | return NULL; | |
308 | ||
309 | bb->next = bblist; | |
310 | bblist = bb; | |
311 | } | |
312 | return bblist; | |
313 | } | |
314 | ||
315 | /** | |
316 | * create_zone_bm_list - create a list of zone bitmap objects | |
317 | */ | |
318 | ||
319 | static inline struct zone_bitmap * | |
320 | create_zone_bm_list(unsigned int nr_zones, struct chain_allocator *ca) | |
321 | { | |
322 | struct zone_bitmap *zbmlist = NULL; | |
323 | ||
324 | while (nr_zones-- > 0) { | |
325 | struct zone_bitmap *zbm; | |
326 | ||
327 | zbm = chain_alloc(ca, sizeof(struct zone_bitmap)); | |
328 | if (!zbm) | |
329 | return NULL; | |
330 | ||
331 | zbm->next = zbmlist; | |
332 | zbmlist = zbm; | |
333 | } | |
334 | return zbmlist; | |
335 | } | |
336 | ||
337 | /** | |
338 | * memory_bm_create - allocate memory for a memory bitmap | |
339 | */ | |
340 | ||
341 | static int | |
342 | memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed) | |
343 | { | |
344 | struct chain_allocator ca; | |
345 | struct zone *zone; | |
346 | struct zone_bitmap *zone_bm; | |
347 | struct bm_block *bb; | |
348 | unsigned int nr; | |
349 | ||
350 | chain_init(&ca, gfp_mask, safe_needed); | |
351 | ||
352 | /* Compute the number of zones */ | |
353 | nr = 0; | |
8357376d RW |
354 | for_each_zone(zone) |
355 | if (populated_zone(zone)) | |
b788db79 RW |
356 | nr++; |
357 | ||
358 | /* Allocate the list of zones bitmap objects */ | |
359 | zone_bm = create_zone_bm_list(nr, &ca); | |
360 | bm->zone_bm_list = zone_bm; | |
361 | if (!zone_bm) { | |
362 | chain_free(&ca, PG_UNSAFE_CLEAR); | |
363 | return -ENOMEM; | |
364 | } | |
365 | ||
366 | /* Initialize the zone bitmap objects */ | |
8357376d | 367 | for_each_zone(zone) { |
b788db79 RW |
368 | unsigned long pfn; |
369 | ||
8357376d | 370 | if (!populated_zone(zone)) |
b788db79 RW |
371 | continue; |
372 | ||
373 | zone_bm->start_pfn = zone->zone_start_pfn; | |
374 | zone_bm->end_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
375 | /* Allocate the list of bitmap block objects */ | |
376 | nr = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK); | |
377 | bb = create_bm_block_list(nr, &ca); | |
378 | zone_bm->bm_blocks = bb; | |
379 | zone_bm->cur_block = bb; | |
380 | if (!bb) | |
381 | goto Free; | |
382 | ||
383 | nr = zone->spanned_pages; | |
384 | pfn = zone->zone_start_pfn; | |
385 | /* Initialize the bitmap block objects */ | |
386 | while (bb) { | |
387 | unsigned long *ptr; | |
388 | ||
8357376d | 389 | ptr = get_image_page(gfp_mask, safe_needed); |
b788db79 RW |
390 | bb->data = ptr; |
391 | if (!ptr) | |
392 | goto Free; | |
393 | ||
394 | bb->start_pfn = pfn; | |
395 | if (nr >= BM_BITS_PER_BLOCK) { | |
396 | pfn += BM_BITS_PER_BLOCK; | |
397 | bb->size = BM_CHUNKS_PER_BLOCK; | |
398 | nr -= BM_BITS_PER_BLOCK; | |
399 | } else { | |
400 | /* This is executed only once in the loop */ | |
401 | pfn += nr; | |
402 | bb->size = DIV_ROUND_UP(nr, BM_BITS_PER_CHUNK); | |
403 | } | |
404 | bb->end_pfn = pfn; | |
405 | bb = bb->next; | |
406 | } | |
407 | zone_bm = zone_bm->next; | |
408 | } | |
409 | bm->p_list = ca.chain; | |
410 | memory_bm_position_reset(bm); | |
411 | return 0; | |
412 | ||
59a49335 | 413 | Free: |
b788db79 RW |
414 | bm->p_list = ca.chain; |
415 | memory_bm_free(bm, PG_UNSAFE_CLEAR); | |
416 | return -ENOMEM; | |
417 | } | |
418 | ||
419 | /** | |
420 | * memory_bm_free - free memory occupied by the memory bitmap @bm | |
421 | */ | |
422 | ||
423 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free) | |
424 | { | |
425 | struct zone_bitmap *zone_bm; | |
426 | ||
427 | /* Free the list of bit blocks for each zone_bitmap object */ | |
428 | zone_bm = bm->zone_bm_list; | |
429 | while (zone_bm) { | |
430 | struct bm_block *bb; | |
431 | ||
432 | bb = zone_bm->bm_blocks; | |
433 | while (bb) { | |
434 | if (bb->data) | |
435 | free_image_page(bb->data, clear_nosave_free); | |
436 | bb = bb->next; | |
437 | } | |
438 | zone_bm = zone_bm->next; | |
439 | } | |
440 | free_list_of_pages(bm->p_list, clear_nosave_free); | |
441 | bm->zone_bm_list = NULL; | |
442 | } | |
443 | ||
444 | /** | |
74dfd666 | 445 | * memory_bm_find_bit - find the bit in the bitmap @bm that corresponds |
b788db79 RW |
446 | * to given pfn. The cur_zone_bm member of @bm and the cur_block member |
447 | * of @bm->cur_zone_bm are updated. | |
b788db79 RW |
448 | */ |
449 | ||
a82f7119 | 450 | static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn, |
74dfd666 | 451 | void **addr, unsigned int *bit_nr) |
b788db79 RW |
452 | { |
453 | struct zone_bitmap *zone_bm; | |
454 | struct bm_block *bb; | |
455 | ||
456 | /* Check if the pfn is from the current zone */ | |
457 | zone_bm = bm->cur.zone_bm; | |
458 | if (pfn < zone_bm->start_pfn || pfn >= zone_bm->end_pfn) { | |
459 | zone_bm = bm->zone_bm_list; | |
460 | /* We don't assume that the zones are sorted by pfns */ | |
461 | while (pfn < zone_bm->start_pfn || pfn >= zone_bm->end_pfn) { | |
462 | zone_bm = zone_bm->next; | |
74dfd666 | 463 | |
a82f7119 RW |
464 | if (!zone_bm) |
465 | return -EFAULT; | |
b788db79 RW |
466 | } |
467 | bm->cur.zone_bm = zone_bm; | |
468 | } | |
469 | /* Check if the pfn corresponds to the current bitmap block */ | |
470 | bb = zone_bm->cur_block; | |
471 | if (pfn < bb->start_pfn) | |
472 | bb = zone_bm->bm_blocks; | |
473 | ||
474 | while (pfn >= bb->end_pfn) { | |
475 | bb = bb->next; | |
74dfd666 RW |
476 | |
477 | BUG_ON(!bb); | |
b788db79 RW |
478 | } |
479 | zone_bm->cur_block = bb; | |
480 | pfn -= bb->start_pfn; | |
74dfd666 RW |
481 | *bit_nr = pfn % BM_BITS_PER_CHUNK; |
482 | *addr = bb->data + pfn / BM_BITS_PER_CHUNK; | |
a82f7119 | 483 | return 0; |
74dfd666 RW |
484 | } |
485 | ||
486 | static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn) | |
487 | { | |
488 | void *addr; | |
489 | unsigned int bit; | |
a82f7119 | 490 | int error; |
74dfd666 | 491 | |
a82f7119 RW |
492 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
493 | BUG_ON(error); | |
74dfd666 RW |
494 | set_bit(bit, addr); |
495 | } | |
496 | ||
a82f7119 RW |
497 | static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn) |
498 | { | |
499 | void *addr; | |
500 | unsigned int bit; | |
501 | int error; | |
502 | ||
503 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); | |
504 | if (!error) | |
505 | set_bit(bit, addr); | |
506 | return error; | |
507 | } | |
508 | ||
74dfd666 RW |
509 | static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn) |
510 | { | |
511 | void *addr; | |
512 | unsigned int bit; | |
a82f7119 | 513 | int error; |
74dfd666 | 514 | |
a82f7119 RW |
515 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
516 | BUG_ON(error); | |
74dfd666 RW |
517 | clear_bit(bit, addr); |
518 | } | |
519 | ||
520 | static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn) | |
521 | { | |
522 | void *addr; | |
523 | unsigned int bit; | |
a82f7119 | 524 | int error; |
74dfd666 | 525 | |
a82f7119 RW |
526 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
527 | BUG_ON(error); | |
74dfd666 | 528 | return test_bit(bit, addr); |
b788db79 RW |
529 | } |
530 | ||
531 | /* Two auxiliary functions for memory_bm_next_pfn */ | |
532 | ||
533 | /* Find the first set bit in the given chunk, if there is one */ | |
534 | ||
535 | static inline int next_bit_in_chunk(int bit, unsigned long *chunk_p) | |
536 | { | |
537 | bit++; | |
538 | while (bit < BM_BITS_PER_CHUNK) { | |
539 | if (test_bit(bit, chunk_p)) | |
540 | return bit; | |
541 | ||
542 | bit++; | |
543 | } | |
544 | return -1; | |
545 | } | |
546 | ||
547 | /* Find a chunk containing some bits set in given block of bits */ | |
548 | ||
549 | static inline int next_chunk_in_block(int n, struct bm_block *bb) | |
550 | { | |
551 | n++; | |
552 | while (n < bb->size) { | |
553 | if (bb->data[n]) | |
554 | return n; | |
555 | ||
556 | n++; | |
557 | } | |
558 | return -1; | |
559 | } | |
560 | ||
561 | /** | |
562 | * memory_bm_next_pfn - find the pfn that corresponds to the next set bit | |
563 | * in the bitmap @bm. If the pfn cannot be found, BM_END_OF_MAP is | |
564 | * returned. | |
565 | * | |
566 | * It is required to run memory_bm_position_reset() before the first call to | |
567 | * this function. | |
568 | */ | |
569 | ||
570 | static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm) | |
571 | { | |
572 | struct zone_bitmap *zone_bm; | |
573 | struct bm_block *bb; | |
574 | int chunk; | |
575 | int bit; | |
576 | ||
577 | do { | |
578 | bb = bm->cur.block; | |
579 | do { | |
580 | chunk = bm->cur.chunk; | |
581 | bit = bm->cur.bit; | |
582 | do { | |
583 | bit = next_bit_in_chunk(bit, bb->data + chunk); | |
584 | if (bit >= 0) | |
585 | goto Return_pfn; | |
586 | ||
587 | chunk = next_chunk_in_block(chunk, bb); | |
588 | bit = -1; | |
589 | } while (chunk >= 0); | |
590 | bb = bb->next; | |
591 | bm->cur.block = bb; | |
592 | memory_bm_reset_chunk(bm); | |
593 | } while (bb); | |
594 | zone_bm = bm->cur.zone_bm->next; | |
595 | if (zone_bm) { | |
596 | bm->cur.zone_bm = zone_bm; | |
597 | bm->cur.block = zone_bm->bm_blocks; | |
598 | memory_bm_reset_chunk(bm); | |
599 | } | |
600 | } while (zone_bm); | |
601 | memory_bm_position_reset(bm); | |
602 | return BM_END_OF_MAP; | |
603 | ||
59a49335 | 604 | Return_pfn: |
b788db79 RW |
605 | bm->cur.chunk = chunk; |
606 | bm->cur.bit = bit; | |
607 | return bb->start_pfn + chunk * BM_BITS_PER_CHUNK + bit; | |
608 | } | |
609 | ||
74dfd666 RW |
610 | /** |
611 | * This structure represents a range of page frames the contents of which | |
612 | * should not be saved during the suspend. | |
613 | */ | |
614 | ||
615 | struct nosave_region { | |
616 | struct list_head list; | |
617 | unsigned long start_pfn; | |
618 | unsigned long end_pfn; | |
619 | }; | |
620 | ||
621 | static LIST_HEAD(nosave_regions); | |
622 | ||
623 | /** | |
624 | * register_nosave_region - register a range of page frames the contents | |
625 | * of which should not be saved during the suspend (to be used in the early | |
626 | * initialization code) | |
627 | */ | |
628 | ||
629 | void __init | |
940d67f6 JB |
630 | __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn, |
631 | int use_kmalloc) | |
74dfd666 RW |
632 | { |
633 | struct nosave_region *region; | |
634 | ||
635 | if (start_pfn >= end_pfn) | |
636 | return; | |
637 | ||
638 | if (!list_empty(&nosave_regions)) { | |
639 | /* Try to extend the previous region (they should be sorted) */ | |
640 | region = list_entry(nosave_regions.prev, | |
641 | struct nosave_region, list); | |
642 | if (region->end_pfn == start_pfn) { | |
643 | region->end_pfn = end_pfn; | |
644 | goto Report; | |
645 | } | |
646 | } | |
940d67f6 JB |
647 | if (use_kmalloc) { |
648 | /* during init, this shouldn't fail */ | |
649 | region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL); | |
650 | BUG_ON(!region); | |
651 | } else | |
652 | /* This allocation cannot fail */ | |
653 | region = alloc_bootmem_low(sizeof(struct nosave_region)); | |
74dfd666 RW |
654 | region->start_pfn = start_pfn; |
655 | region->end_pfn = end_pfn; | |
656 | list_add_tail(®ion->list, &nosave_regions); | |
657 | Report: | |
23976728 | 658 | printk(KERN_INFO "PM: Registered nosave memory: %016lx - %016lx\n", |
74dfd666 RW |
659 | start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT); |
660 | } | |
661 | ||
662 | /* | |
663 | * Set bits in this map correspond to the page frames the contents of which | |
664 | * should not be saved during the suspend. | |
665 | */ | |
666 | static struct memory_bitmap *forbidden_pages_map; | |
667 | ||
668 | /* Set bits in this map correspond to free page frames. */ | |
669 | static struct memory_bitmap *free_pages_map; | |
670 | ||
671 | /* | |
672 | * Each page frame allocated for creating the image is marked by setting the | |
673 | * corresponding bits in forbidden_pages_map and free_pages_map simultaneously | |
674 | */ | |
675 | ||
676 | void swsusp_set_page_free(struct page *page) | |
677 | { | |
678 | if (free_pages_map) | |
679 | memory_bm_set_bit(free_pages_map, page_to_pfn(page)); | |
680 | } | |
681 | ||
682 | static int swsusp_page_is_free(struct page *page) | |
683 | { | |
684 | return free_pages_map ? | |
685 | memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0; | |
686 | } | |
687 | ||
688 | void swsusp_unset_page_free(struct page *page) | |
689 | { | |
690 | if (free_pages_map) | |
691 | memory_bm_clear_bit(free_pages_map, page_to_pfn(page)); | |
692 | } | |
693 | ||
694 | static void swsusp_set_page_forbidden(struct page *page) | |
695 | { | |
696 | if (forbidden_pages_map) | |
697 | memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page)); | |
698 | } | |
699 | ||
700 | int swsusp_page_is_forbidden(struct page *page) | |
701 | { | |
702 | return forbidden_pages_map ? | |
703 | memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0; | |
704 | } | |
705 | ||
706 | static void swsusp_unset_page_forbidden(struct page *page) | |
707 | { | |
708 | if (forbidden_pages_map) | |
709 | memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page)); | |
710 | } | |
711 | ||
712 | /** | |
713 | * mark_nosave_pages - set bits corresponding to the page frames the | |
714 | * contents of which should not be saved in a given bitmap. | |
715 | */ | |
716 | ||
717 | static void mark_nosave_pages(struct memory_bitmap *bm) | |
718 | { | |
719 | struct nosave_region *region; | |
720 | ||
721 | if (list_empty(&nosave_regions)) | |
722 | return; | |
723 | ||
724 | list_for_each_entry(region, &nosave_regions, list) { | |
725 | unsigned long pfn; | |
726 | ||
23976728 | 727 | pr_debug("PM: Marking nosave pages: %016lx - %016lx\n", |
74dfd666 RW |
728 | region->start_pfn << PAGE_SHIFT, |
729 | region->end_pfn << PAGE_SHIFT); | |
730 | ||
731 | for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++) | |
a82f7119 RW |
732 | if (pfn_valid(pfn)) { |
733 | /* | |
734 | * It is safe to ignore the result of | |
735 | * mem_bm_set_bit_check() here, since we won't | |
736 | * touch the PFNs for which the error is | |
737 | * returned anyway. | |
738 | */ | |
739 | mem_bm_set_bit_check(bm, pfn); | |
740 | } | |
74dfd666 RW |
741 | } |
742 | } | |
743 | ||
744 | /** | |
745 | * create_basic_memory_bitmaps - create bitmaps needed for marking page | |
746 | * frames that should not be saved and free page frames. The pointers | |
747 | * forbidden_pages_map and free_pages_map are only modified if everything | |
748 | * goes well, because we don't want the bits to be used before both bitmaps | |
749 | * are set up. | |
750 | */ | |
751 | ||
752 | int create_basic_memory_bitmaps(void) | |
753 | { | |
754 | struct memory_bitmap *bm1, *bm2; | |
755 | int error = 0; | |
756 | ||
757 | BUG_ON(forbidden_pages_map || free_pages_map); | |
758 | ||
0709db60 | 759 | bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
760 | if (!bm1) |
761 | return -ENOMEM; | |
762 | ||
0709db60 | 763 | error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
764 | if (error) |
765 | goto Free_first_object; | |
766 | ||
0709db60 | 767 | bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
768 | if (!bm2) |
769 | goto Free_first_bitmap; | |
770 | ||
0709db60 | 771 | error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
772 | if (error) |
773 | goto Free_second_object; | |
774 | ||
775 | forbidden_pages_map = bm1; | |
776 | free_pages_map = bm2; | |
777 | mark_nosave_pages(forbidden_pages_map); | |
778 | ||
23976728 | 779 | pr_debug("PM: Basic memory bitmaps created\n"); |
74dfd666 RW |
780 | |
781 | return 0; | |
782 | ||
783 | Free_second_object: | |
784 | kfree(bm2); | |
785 | Free_first_bitmap: | |
786 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
787 | Free_first_object: | |
788 | kfree(bm1); | |
789 | return -ENOMEM; | |
790 | } | |
791 | ||
792 | /** | |
793 | * free_basic_memory_bitmaps - free memory bitmaps allocated by | |
794 | * create_basic_memory_bitmaps(). The auxiliary pointers are necessary | |
795 | * so that the bitmaps themselves are not referred to while they are being | |
796 | * freed. | |
797 | */ | |
798 | ||
799 | void free_basic_memory_bitmaps(void) | |
800 | { | |
801 | struct memory_bitmap *bm1, *bm2; | |
802 | ||
803 | BUG_ON(!(forbidden_pages_map && free_pages_map)); | |
804 | ||
805 | bm1 = forbidden_pages_map; | |
806 | bm2 = free_pages_map; | |
807 | forbidden_pages_map = NULL; | |
808 | free_pages_map = NULL; | |
809 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
810 | kfree(bm1); | |
811 | memory_bm_free(bm2, PG_UNSAFE_CLEAR); | |
812 | kfree(bm2); | |
813 | ||
23976728 | 814 | pr_debug("PM: Basic memory bitmaps freed\n"); |
74dfd666 RW |
815 | } |
816 | ||
b788db79 RW |
817 | /** |
818 | * snapshot_additional_pages - estimate the number of additional pages | |
819 | * be needed for setting up the suspend image data structures for given | |
820 | * zone (usually the returned value is greater than the exact number) | |
821 | */ | |
822 | ||
823 | unsigned int snapshot_additional_pages(struct zone *zone) | |
824 | { | |
825 | unsigned int res; | |
826 | ||
827 | res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK); | |
828 | res += DIV_ROUND_UP(res * sizeof(struct bm_block), PAGE_SIZE); | |
8357376d | 829 | return 2 * res; |
b788db79 RW |
830 | } |
831 | ||
8357376d RW |
832 | #ifdef CONFIG_HIGHMEM |
833 | /** | |
834 | * count_free_highmem_pages - compute the total number of free highmem | |
835 | * pages, system-wide. | |
836 | */ | |
837 | ||
838 | static unsigned int count_free_highmem_pages(void) | |
839 | { | |
840 | struct zone *zone; | |
841 | unsigned int cnt = 0; | |
842 | ||
843 | for_each_zone(zone) | |
844 | if (populated_zone(zone) && is_highmem(zone)) | |
d23ad423 | 845 | cnt += zone_page_state(zone, NR_FREE_PAGES); |
8357376d RW |
846 | |
847 | return cnt; | |
848 | } | |
849 | ||
850 | /** | |
851 | * saveable_highmem_page - Determine whether a highmem page should be | |
852 | * included in the suspend image. | |
853 | * | |
854 | * We should save the page if it isn't Nosave or NosaveFree, or Reserved, | |
855 | * and it isn't a part of a free chunk of pages. | |
856 | */ | |
857 | ||
858 | static struct page *saveable_highmem_page(unsigned long pfn) | |
859 | { | |
860 | struct page *page; | |
861 | ||
862 | if (!pfn_valid(pfn)) | |
863 | return NULL; | |
864 | ||
865 | page = pfn_to_page(pfn); | |
866 | ||
867 | BUG_ON(!PageHighMem(page)); | |
868 | ||
7be98234 RW |
869 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) || |
870 | PageReserved(page)) | |
8357376d RW |
871 | return NULL; |
872 | ||
873 | return page; | |
874 | } | |
875 | ||
876 | /** | |
877 | * count_highmem_pages - compute the total number of saveable highmem | |
878 | * pages. | |
879 | */ | |
880 | ||
881 | unsigned int count_highmem_pages(void) | |
882 | { | |
883 | struct zone *zone; | |
884 | unsigned int n = 0; | |
885 | ||
886 | for_each_zone(zone) { | |
887 | unsigned long pfn, max_zone_pfn; | |
888 | ||
889 | if (!is_highmem(zone)) | |
890 | continue; | |
891 | ||
892 | mark_free_pages(zone); | |
893 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; | |
894 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
895 | if (saveable_highmem_page(pfn)) | |
896 | n++; | |
897 | } | |
898 | return n; | |
899 | } | |
900 | #else | |
901 | static inline void *saveable_highmem_page(unsigned long pfn) { return NULL; } | |
8357376d RW |
902 | #endif /* CONFIG_HIGHMEM */ |
903 | ||
25761b6e | 904 | /** |
8a235efa RW |
905 | * saveable_page - Determine whether a non-highmem page should be included |
906 | * in the suspend image. | |
25761b6e | 907 | * |
8357376d RW |
908 | * We should save the page if it isn't Nosave, and is not in the range |
909 | * of pages statically defined as 'unsaveable', and it isn't a part of | |
910 | * a free chunk of pages. | |
25761b6e RW |
911 | */ |
912 | ||
ae83c5ee | 913 | static struct page *saveable_page(unsigned long pfn) |
25761b6e | 914 | { |
de491861 | 915 | struct page *page; |
25761b6e RW |
916 | |
917 | if (!pfn_valid(pfn)) | |
ae83c5ee | 918 | return NULL; |
25761b6e RW |
919 | |
920 | page = pfn_to_page(pfn); | |
ae83c5ee | 921 | |
8357376d RW |
922 | BUG_ON(PageHighMem(page)); |
923 | ||
7be98234 | 924 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page)) |
ae83c5ee | 925 | return NULL; |
8357376d | 926 | |
8a235efa RW |
927 | if (PageReserved(page) |
928 | && (!kernel_page_present(page) || pfn_is_nosave(pfn))) | |
ae83c5ee | 929 | return NULL; |
25761b6e | 930 | |
ae83c5ee | 931 | return page; |
25761b6e RW |
932 | } |
933 | ||
8357376d RW |
934 | /** |
935 | * count_data_pages - compute the total number of saveable non-highmem | |
936 | * pages. | |
937 | */ | |
938 | ||
72a97e08 | 939 | unsigned int count_data_pages(void) |
25761b6e RW |
940 | { |
941 | struct zone *zone; | |
ae83c5ee | 942 | unsigned long pfn, max_zone_pfn; |
dc19d507 | 943 | unsigned int n = 0; |
25761b6e | 944 | |
8357376d | 945 | for_each_zone(zone) { |
25761b6e RW |
946 | if (is_highmem(zone)) |
947 | continue; | |
8357376d | 948 | |
25761b6e | 949 | mark_free_pages(zone); |
ae83c5ee RW |
950 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
951 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
8357376d RW |
952 | if(saveable_page(pfn)) |
953 | n++; | |
25761b6e | 954 | } |
a0f49651 | 955 | return n; |
25761b6e RW |
956 | } |
957 | ||
8357376d RW |
958 | /* This is needed, because copy_page and memcpy are not usable for copying |
959 | * task structs. | |
960 | */ | |
961 | static inline void do_copy_page(long *dst, long *src) | |
f623f0db RW |
962 | { |
963 | int n; | |
964 | ||
f623f0db RW |
965 | for (n = PAGE_SIZE / sizeof(long); n; n--) |
966 | *dst++ = *src++; | |
967 | } | |
968 | ||
8a235efa RW |
969 | |
970 | /** | |
971 | * safe_copy_page - check if the page we are going to copy is marked as | |
972 | * present in the kernel page tables (this always is the case if | |
973 | * CONFIG_DEBUG_PAGEALLOC is not set and in that case | |
974 | * kernel_page_present() always returns 'true'). | |
975 | */ | |
976 | static void safe_copy_page(void *dst, struct page *s_page) | |
977 | { | |
978 | if (kernel_page_present(s_page)) { | |
979 | do_copy_page(dst, page_address(s_page)); | |
980 | } else { | |
981 | kernel_map_pages(s_page, 1, 1); | |
982 | do_copy_page(dst, page_address(s_page)); | |
983 | kernel_map_pages(s_page, 1, 0); | |
984 | } | |
985 | } | |
986 | ||
987 | ||
8357376d RW |
988 | #ifdef CONFIG_HIGHMEM |
989 | static inline struct page * | |
990 | page_is_saveable(struct zone *zone, unsigned long pfn) | |
991 | { | |
992 | return is_highmem(zone) ? | |
993 | saveable_highmem_page(pfn) : saveable_page(pfn); | |
994 | } | |
995 | ||
8a235efa | 996 | static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
8357376d RW |
997 | { |
998 | struct page *s_page, *d_page; | |
999 | void *src, *dst; | |
1000 | ||
1001 | s_page = pfn_to_page(src_pfn); | |
1002 | d_page = pfn_to_page(dst_pfn); | |
1003 | if (PageHighMem(s_page)) { | |
1004 | src = kmap_atomic(s_page, KM_USER0); | |
1005 | dst = kmap_atomic(d_page, KM_USER1); | |
1006 | do_copy_page(dst, src); | |
1007 | kunmap_atomic(src, KM_USER0); | |
1008 | kunmap_atomic(dst, KM_USER1); | |
1009 | } else { | |
8357376d RW |
1010 | if (PageHighMem(d_page)) { |
1011 | /* Page pointed to by src may contain some kernel | |
1012 | * data modified by kmap_atomic() | |
1013 | */ | |
8a235efa | 1014 | safe_copy_page(buffer, s_page); |
8357376d RW |
1015 | dst = kmap_atomic(pfn_to_page(dst_pfn), KM_USER0); |
1016 | memcpy(dst, buffer, PAGE_SIZE); | |
1017 | kunmap_atomic(dst, KM_USER0); | |
1018 | } else { | |
8a235efa | 1019 | safe_copy_page(page_address(d_page), s_page); |
8357376d RW |
1020 | } |
1021 | } | |
1022 | } | |
1023 | #else | |
1024 | #define page_is_saveable(zone, pfn) saveable_page(pfn) | |
1025 | ||
8a235efa | 1026 | static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
8357376d | 1027 | { |
8a235efa RW |
1028 | safe_copy_page(page_address(pfn_to_page(dst_pfn)), |
1029 | pfn_to_page(src_pfn)); | |
8357376d RW |
1030 | } |
1031 | #endif /* CONFIG_HIGHMEM */ | |
1032 | ||
b788db79 RW |
1033 | static void |
1034 | copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm) | |
25761b6e RW |
1035 | { |
1036 | struct zone *zone; | |
b788db79 | 1037 | unsigned long pfn; |
25761b6e | 1038 | |
8357376d | 1039 | for_each_zone(zone) { |
b788db79 RW |
1040 | unsigned long max_zone_pfn; |
1041 | ||
25761b6e | 1042 | mark_free_pages(zone); |
ae83c5ee | 1043 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
b788db79 | 1044 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
8357376d | 1045 | if (page_is_saveable(zone, pfn)) |
b788db79 | 1046 | memory_bm_set_bit(orig_bm, pfn); |
25761b6e | 1047 | } |
b788db79 RW |
1048 | memory_bm_position_reset(orig_bm); |
1049 | memory_bm_position_reset(copy_bm); | |
df7c4872 | 1050 | for(;;) { |
b788db79 | 1051 | pfn = memory_bm_next_pfn(orig_bm); |
df7c4872 FW |
1052 | if (unlikely(pfn == BM_END_OF_MAP)) |
1053 | break; | |
1054 | copy_data_page(memory_bm_next_pfn(copy_bm), pfn); | |
1055 | } | |
25761b6e RW |
1056 | } |
1057 | ||
8357376d RW |
1058 | /* Total number of image pages */ |
1059 | static unsigned int nr_copy_pages; | |
1060 | /* Number of pages needed for saving the original pfns of the image pages */ | |
1061 | static unsigned int nr_meta_pages; | |
1062 | ||
25761b6e | 1063 | /** |
940864dd | 1064 | * swsusp_free - free pages allocated for the suspend. |
cd560bb2 | 1065 | * |
940864dd RW |
1066 | * Suspend pages are alocated before the atomic copy is made, so we |
1067 | * need to release them after the resume. | |
25761b6e RW |
1068 | */ |
1069 | ||
1070 | void swsusp_free(void) | |
1071 | { | |
1072 | struct zone *zone; | |
ae83c5ee | 1073 | unsigned long pfn, max_zone_pfn; |
25761b6e RW |
1074 | |
1075 | for_each_zone(zone) { | |
ae83c5ee RW |
1076 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
1077 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1078 | if (pfn_valid(pfn)) { | |
1079 | struct page *page = pfn_to_page(pfn); | |
1080 | ||
7be98234 RW |
1081 | if (swsusp_page_is_forbidden(page) && |
1082 | swsusp_page_is_free(page)) { | |
1083 | swsusp_unset_page_forbidden(page); | |
1084 | swsusp_unset_page_free(page); | |
8357376d | 1085 | __free_page(page); |
25761b6e RW |
1086 | } |
1087 | } | |
1088 | } | |
f577eb30 RW |
1089 | nr_copy_pages = 0; |
1090 | nr_meta_pages = 0; | |
75534b50 | 1091 | restore_pblist = NULL; |
6e1819d6 | 1092 | buffer = NULL; |
25761b6e RW |
1093 | } |
1094 | ||
8357376d RW |
1095 | #ifdef CONFIG_HIGHMEM |
1096 | /** | |
1097 | * count_pages_for_highmem - compute the number of non-highmem pages | |
1098 | * that will be necessary for creating copies of highmem pages. | |
1099 | */ | |
1100 | ||
1101 | static unsigned int count_pages_for_highmem(unsigned int nr_highmem) | |
1102 | { | |
1103 | unsigned int free_highmem = count_free_highmem_pages(); | |
1104 | ||
1105 | if (free_highmem >= nr_highmem) | |
1106 | nr_highmem = 0; | |
1107 | else | |
1108 | nr_highmem -= free_highmem; | |
1109 | ||
1110 | return nr_highmem; | |
1111 | } | |
1112 | #else | |
1113 | static unsigned int | |
1114 | count_pages_for_highmem(unsigned int nr_highmem) { return 0; } | |
1115 | #endif /* CONFIG_HIGHMEM */ | |
25761b6e RW |
1116 | |
1117 | /** | |
8357376d RW |
1118 | * enough_free_mem - Make sure we have enough free memory for the |
1119 | * snapshot image. | |
25761b6e RW |
1120 | */ |
1121 | ||
8357376d | 1122 | static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem) |
25761b6e | 1123 | { |
e5e2fa78 | 1124 | struct zone *zone; |
940864dd | 1125 | unsigned int free = 0, meta = 0; |
e5e2fa78 | 1126 | |
8357376d RW |
1127 | for_each_zone(zone) { |
1128 | meta += snapshot_additional_pages(zone); | |
1129 | if (!is_highmem(zone)) | |
d23ad423 | 1130 | free += zone_page_state(zone, NR_FREE_PAGES); |
8357376d | 1131 | } |
940864dd | 1132 | |
8357376d | 1133 | nr_pages += count_pages_for_highmem(nr_highmem); |
23976728 | 1134 | pr_debug("PM: Normal pages needed: %u + %u + %u, available pages: %u\n", |
940864dd RW |
1135 | nr_pages, PAGES_FOR_IO, meta, free); |
1136 | ||
1137 | return free > nr_pages + PAGES_FOR_IO + meta; | |
25761b6e RW |
1138 | } |
1139 | ||
8357376d RW |
1140 | #ifdef CONFIG_HIGHMEM |
1141 | /** | |
1142 | * get_highmem_buffer - if there are some highmem pages in the suspend | |
1143 | * image, we may need the buffer to copy them and/or load their data. | |
1144 | */ | |
1145 | ||
1146 | static inline int get_highmem_buffer(int safe_needed) | |
1147 | { | |
1148 | buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed); | |
1149 | return buffer ? 0 : -ENOMEM; | |
1150 | } | |
1151 | ||
1152 | /** | |
1153 | * alloc_highmem_image_pages - allocate some highmem pages for the image. | |
1154 | * Try to allocate as many pages as needed, but if the number of free | |
1155 | * highmem pages is lesser than that, allocate them all. | |
1156 | */ | |
1157 | ||
1158 | static inline unsigned int | |
1159 | alloc_highmem_image_pages(struct memory_bitmap *bm, unsigned int nr_highmem) | |
1160 | { | |
1161 | unsigned int to_alloc = count_free_highmem_pages(); | |
1162 | ||
1163 | if (to_alloc > nr_highmem) | |
1164 | to_alloc = nr_highmem; | |
1165 | ||
1166 | nr_highmem -= to_alloc; | |
1167 | while (to_alloc-- > 0) { | |
1168 | struct page *page; | |
1169 | ||
1170 | page = alloc_image_page(__GFP_HIGHMEM); | |
1171 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1172 | } | |
1173 | return nr_highmem; | |
1174 | } | |
1175 | #else | |
1176 | static inline int get_highmem_buffer(int safe_needed) { return 0; } | |
1177 | ||
1178 | static inline unsigned int | |
1179 | alloc_highmem_image_pages(struct memory_bitmap *bm, unsigned int n) { return 0; } | |
1180 | #endif /* CONFIG_HIGHMEM */ | |
1181 | ||
1182 | /** | |
1183 | * swsusp_alloc - allocate memory for the suspend image | |
1184 | * | |
1185 | * We first try to allocate as many highmem pages as there are | |
1186 | * saveable highmem pages in the system. If that fails, we allocate | |
1187 | * non-highmem pages for the copies of the remaining highmem ones. | |
1188 | * | |
1189 | * In this approach it is likely that the copies of highmem pages will | |
1190 | * also be located in the high memory, because of the way in which | |
1191 | * copy_data_pages() works. | |
1192 | */ | |
1193 | ||
b788db79 RW |
1194 | static int |
1195 | swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm, | |
8357376d | 1196 | unsigned int nr_pages, unsigned int nr_highmem) |
054bd4c1 | 1197 | { |
b788db79 | 1198 | int error; |
054bd4c1 | 1199 | |
b788db79 RW |
1200 | error = memory_bm_create(orig_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY); |
1201 | if (error) | |
1202 | goto Free; | |
25761b6e | 1203 | |
b788db79 RW |
1204 | error = memory_bm_create(copy_bm, GFP_ATOMIC | __GFP_COLD, PG_ANY); |
1205 | if (error) | |
1206 | goto Free; | |
25761b6e | 1207 | |
8357376d RW |
1208 | if (nr_highmem > 0) { |
1209 | error = get_highmem_buffer(PG_ANY); | |
1210 | if (error) | |
1211 | goto Free; | |
1212 | ||
1213 | nr_pages += alloc_highmem_image_pages(copy_bm, nr_highmem); | |
1214 | } | |
b788db79 | 1215 | while (nr_pages-- > 0) { |
8357376d RW |
1216 | struct page *page = alloc_image_page(GFP_ATOMIC | __GFP_COLD); |
1217 | ||
b788db79 RW |
1218 | if (!page) |
1219 | goto Free; | |
25761b6e | 1220 | |
b788db79 | 1221 | memory_bm_set_bit(copy_bm, page_to_pfn(page)); |
25761b6e | 1222 | } |
b788db79 | 1223 | return 0; |
25761b6e | 1224 | |
59a49335 | 1225 | Free: |
b788db79 RW |
1226 | swsusp_free(); |
1227 | return -ENOMEM; | |
25761b6e RW |
1228 | } |
1229 | ||
8357376d RW |
1230 | /* Memory bitmap used for marking saveable pages (during suspend) or the |
1231 | * suspend image pages (during resume) | |
1232 | */ | |
b788db79 | 1233 | static struct memory_bitmap orig_bm; |
8357376d RW |
1234 | /* Memory bitmap used on suspend for marking allocated pages that will contain |
1235 | * the copies of saveable pages. During resume it is initially used for | |
1236 | * marking the suspend image pages, but then its set bits are duplicated in | |
1237 | * @orig_bm and it is released. Next, on systems with high memory, it may be | |
1238 | * used for marking "safe" highmem pages, but it has to be reinitialized for | |
1239 | * this purpose. | |
b788db79 RW |
1240 | */ |
1241 | static struct memory_bitmap copy_bm; | |
1242 | ||
2e32a43e | 1243 | asmlinkage int swsusp_save(void) |
25761b6e | 1244 | { |
8357376d | 1245 | unsigned int nr_pages, nr_highmem; |
25761b6e | 1246 | |
23976728 | 1247 | printk(KERN_INFO "PM: Creating hibernation image: \n"); |
25761b6e | 1248 | |
9f8f2172 | 1249 | drain_local_pages(NULL); |
a0f49651 | 1250 | nr_pages = count_data_pages(); |
8357376d | 1251 | nr_highmem = count_highmem_pages(); |
23976728 | 1252 | printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem); |
25761b6e | 1253 | |
8357376d | 1254 | if (!enough_free_mem(nr_pages, nr_highmem)) { |
23976728 | 1255 | printk(KERN_ERR "PM: Not enough free memory\n"); |
25761b6e RW |
1256 | return -ENOMEM; |
1257 | } | |
1258 | ||
8357376d | 1259 | if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) { |
23976728 | 1260 | printk(KERN_ERR "PM: Memory allocation failed\n"); |
a0f49651 | 1261 | return -ENOMEM; |
8357376d | 1262 | } |
25761b6e RW |
1263 | |
1264 | /* During allocating of suspend pagedir, new cold pages may appear. | |
1265 | * Kill them. | |
1266 | */ | |
9f8f2172 | 1267 | drain_local_pages(NULL); |
b788db79 | 1268 | copy_data_pages(©_bm, &orig_bm); |
25761b6e RW |
1269 | |
1270 | /* | |
1271 | * End of critical section. From now on, we can write to memory, | |
1272 | * but we should not touch disk. This specially means we must _not_ | |
1273 | * touch swap space! Except we must write out our image of course. | |
1274 | */ | |
1275 | ||
8357376d | 1276 | nr_pages += nr_highmem; |
a0f49651 | 1277 | nr_copy_pages = nr_pages; |
8357376d | 1278 | nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE); |
a0f49651 | 1279 | |
23976728 RW |
1280 | printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n", |
1281 | nr_pages); | |
8357376d | 1282 | |
25761b6e RW |
1283 | return 0; |
1284 | } | |
f577eb30 | 1285 | |
d307c4a8 RW |
1286 | #ifndef CONFIG_ARCH_HIBERNATION_HEADER |
1287 | static int init_header_complete(struct swsusp_info *info) | |
f577eb30 | 1288 | { |
d307c4a8 | 1289 | memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname)); |
f577eb30 | 1290 | info->version_code = LINUX_VERSION_CODE; |
d307c4a8 RW |
1291 | return 0; |
1292 | } | |
1293 | ||
1294 | static char *check_image_kernel(struct swsusp_info *info) | |
1295 | { | |
1296 | if (info->version_code != LINUX_VERSION_CODE) | |
1297 | return "kernel version"; | |
1298 | if (strcmp(info->uts.sysname,init_utsname()->sysname)) | |
1299 | return "system type"; | |
1300 | if (strcmp(info->uts.release,init_utsname()->release)) | |
1301 | return "kernel release"; | |
1302 | if (strcmp(info->uts.version,init_utsname()->version)) | |
1303 | return "version"; | |
1304 | if (strcmp(info->uts.machine,init_utsname()->machine)) | |
1305 | return "machine"; | |
1306 | return NULL; | |
1307 | } | |
1308 | #endif /* CONFIG_ARCH_HIBERNATION_HEADER */ | |
1309 | ||
af508b34 RW |
1310 | unsigned long snapshot_get_image_size(void) |
1311 | { | |
1312 | return nr_copy_pages + nr_meta_pages + 1; | |
1313 | } | |
1314 | ||
d307c4a8 RW |
1315 | static int init_header(struct swsusp_info *info) |
1316 | { | |
1317 | memset(info, 0, sizeof(struct swsusp_info)); | |
f577eb30 | 1318 | info->num_physpages = num_physpages; |
f577eb30 | 1319 | info->image_pages = nr_copy_pages; |
af508b34 | 1320 | info->pages = snapshot_get_image_size(); |
6e1819d6 RW |
1321 | info->size = info->pages; |
1322 | info->size <<= PAGE_SHIFT; | |
d307c4a8 | 1323 | return init_header_complete(info); |
f577eb30 RW |
1324 | } |
1325 | ||
1326 | /** | |
940864dd RW |
1327 | * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm |
1328 | * are stored in the array @buf[] (1 page at a time) | |
f577eb30 RW |
1329 | */ |
1330 | ||
b788db79 | 1331 | static inline void |
940864dd | 1332 | pack_pfns(unsigned long *buf, struct memory_bitmap *bm) |
f577eb30 RW |
1333 | { |
1334 | int j; | |
1335 | ||
b788db79 | 1336 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
940864dd RW |
1337 | buf[j] = memory_bm_next_pfn(bm); |
1338 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
b788db79 | 1339 | break; |
f577eb30 | 1340 | } |
f577eb30 RW |
1341 | } |
1342 | ||
1343 | /** | |
1344 | * snapshot_read_next - used for reading the system memory snapshot. | |
1345 | * | |
1346 | * On the first call to it @handle should point to a zeroed | |
1347 | * snapshot_handle structure. The structure gets updated and a pointer | |
1348 | * to it should be passed to this function every next time. | |
1349 | * | |
1350 | * The @count parameter should contain the number of bytes the caller | |
1351 | * wants to read from the snapshot. It must not be zero. | |
1352 | * | |
1353 | * On success the function returns a positive number. Then, the caller | |
1354 | * is allowed to read up to the returned number of bytes from the memory | |
1355 | * location computed by the data_of() macro. The number returned | |
1356 | * may be smaller than @count, but this only happens if the read would | |
1357 | * cross a page boundary otherwise. | |
1358 | * | |
1359 | * The function returns 0 to indicate the end of data stream condition, | |
1360 | * and a negative number is returned on error. In such cases the | |
1361 | * structure pointed to by @handle is not updated and should not be used | |
1362 | * any more. | |
1363 | */ | |
1364 | ||
1365 | int snapshot_read_next(struct snapshot_handle *handle, size_t count) | |
1366 | { | |
fb13a28b | 1367 | if (handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 1368 | return 0; |
b788db79 | 1369 | |
f577eb30 RW |
1370 | if (!buffer) { |
1371 | /* This makes the buffer be freed by swsusp_free() */ | |
8357376d | 1372 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); |
f577eb30 RW |
1373 | if (!buffer) |
1374 | return -ENOMEM; | |
1375 | } | |
1376 | if (!handle->offset) { | |
d307c4a8 RW |
1377 | int error; |
1378 | ||
1379 | error = init_header((struct swsusp_info *)buffer); | |
1380 | if (error) | |
1381 | return error; | |
f577eb30 | 1382 | handle->buffer = buffer; |
b788db79 RW |
1383 | memory_bm_position_reset(&orig_bm); |
1384 | memory_bm_position_reset(©_bm); | |
f577eb30 | 1385 | } |
fb13a28b RW |
1386 | if (handle->prev < handle->cur) { |
1387 | if (handle->cur <= nr_meta_pages) { | |
b788db79 | 1388 | memset(buffer, 0, PAGE_SIZE); |
940864dd | 1389 | pack_pfns(buffer, &orig_bm); |
f577eb30 | 1390 | } else { |
8357376d | 1391 | struct page *page; |
b788db79 | 1392 | |
8357376d RW |
1393 | page = pfn_to_page(memory_bm_next_pfn(©_bm)); |
1394 | if (PageHighMem(page)) { | |
1395 | /* Highmem pages are copied to the buffer, | |
1396 | * because we can't return with a kmapped | |
1397 | * highmem page (we may not be called again). | |
1398 | */ | |
1399 | void *kaddr; | |
1400 | ||
1401 | kaddr = kmap_atomic(page, KM_USER0); | |
1402 | memcpy(buffer, kaddr, PAGE_SIZE); | |
1403 | kunmap_atomic(kaddr, KM_USER0); | |
1404 | handle->buffer = buffer; | |
1405 | } else { | |
1406 | handle->buffer = page_address(page); | |
1407 | } | |
f577eb30 | 1408 | } |
fb13a28b | 1409 | handle->prev = handle->cur; |
f577eb30 | 1410 | } |
fb13a28b RW |
1411 | handle->buf_offset = handle->cur_offset; |
1412 | if (handle->cur_offset + count >= PAGE_SIZE) { | |
1413 | count = PAGE_SIZE - handle->cur_offset; | |
1414 | handle->cur_offset = 0; | |
1415 | handle->cur++; | |
f577eb30 | 1416 | } else { |
fb13a28b | 1417 | handle->cur_offset += count; |
f577eb30 RW |
1418 | } |
1419 | handle->offset += count; | |
1420 | return count; | |
1421 | } | |
1422 | ||
1423 | /** | |
1424 | * mark_unsafe_pages - mark the pages that cannot be used for storing | |
1425 | * the image during resume, because they conflict with the pages that | |
1426 | * had been used before suspend | |
1427 | */ | |
1428 | ||
940864dd | 1429 | static int mark_unsafe_pages(struct memory_bitmap *bm) |
f577eb30 RW |
1430 | { |
1431 | struct zone *zone; | |
ae83c5ee | 1432 | unsigned long pfn, max_zone_pfn; |
f577eb30 RW |
1433 | |
1434 | /* Clear page flags */ | |
8357376d | 1435 | for_each_zone(zone) { |
ae83c5ee RW |
1436 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
1437 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1438 | if (pfn_valid(pfn)) | |
7be98234 | 1439 | swsusp_unset_page_free(pfn_to_page(pfn)); |
f577eb30 RW |
1440 | } |
1441 | ||
940864dd RW |
1442 | /* Mark pages that correspond to the "original" pfns as "unsafe" */ |
1443 | memory_bm_position_reset(bm); | |
1444 | do { | |
1445 | pfn = memory_bm_next_pfn(bm); | |
1446 | if (likely(pfn != BM_END_OF_MAP)) { | |
1447 | if (likely(pfn_valid(pfn))) | |
7be98234 | 1448 | swsusp_set_page_free(pfn_to_page(pfn)); |
940864dd RW |
1449 | else |
1450 | return -EFAULT; | |
1451 | } | |
1452 | } while (pfn != BM_END_OF_MAP); | |
f577eb30 | 1453 | |
940864dd | 1454 | allocated_unsafe_pages = 0; |
968808b8 | 1455 | |
f577eb30 RW |
1456 | return 0; |
1457 | } | |
1458 | ||
940864dd RW |
1459 | static void |
1460 | duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src) | |
f577eb30 | 1461 | { |
940864dd RW |
1462 | unsigned long pfn; |
1463 | ||
1464 | memory_bm_position_reset(src); | |
1465 | pfn = memory_bm_next_pfn(src); | |
1466 | while (pfn != BM_END_OF_MAP) { | |
1467 | memory_bm_set_bit(dst, pfn); | |
1468 | pfn = memory_bm_next_pfn(src); | |
f577eb30 RW |
1469 | } |
1470 | } | |
1471 | ||
d307c4a8 | 1472 | static int check_header(struct swsusp_info *info) |
f577eb30 | 1473 | { |
d307c4a8 | 1474 | char *reason; |
f577eb30 | 1475 | |
d307c4a8 RW |
1476 | reason = check_image_kernel(info); |
1477 | if (!reason && info->num_physpages != num_physpages) | |
f577eb30 | 1478 | reason = "memory size"; |
f577eb30 | 1479 | if (reason) { |
23976728 | 1480 | printk(KERN_ERR "PM: Image mismatch: %s\n", reason); |
f577eb30 RW |
1481 | return -EPERM; |
1482 | } | |
1483 | return 0; | |
1484 | } | |
1485 | ||
1486 | /** | |
1487 | * load header - check the image header and copy data from it | |
1488 | */ | |
1489 | ||
940864dd RW |
1490 | static int |
1491 | load_header(struct swsusp_info *info) | |
f577eb30 RW |
1492 | { |
1493 | int error; | |
f577eb30 | 1494 | |
940864dd | 1495 | restore_pblist = NULL; |
f577eb30 RW |
1496 | error = check_header(info); |
1497 | if (!error) { | |
f577eb30 RW |
1498 | nr_copy_pages = info->image_pages; |
1499 | nr_meta_pages = info->pages - info->image_pages - 1; | |
1500 | } | |
1501 | return error; | |
1502 | } | |
1503 | ||
1504 | /** | |
940864dd RW |
1505 | * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set |
1506 | * the corresponding bit in the memory bitmap @bm | |
f577eb30 RW |
1507 | */ |
1508 | ||
940864dd RW |
1509 | static inline void |
1510 | unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm) | |
f577eb30 RW |
1511 | { |
1512 | int j; | |
1513 | ||
940864dd RW |
1514 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
1515 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
1516 | break; | |
1517 | ||
1518 | memory_bm_set_bit(bm, buf[j]); | |
f577eb30 | 1519 | } |
f577eb30 RW |
1520 | } |
1521 | ||
8357376d RW |
1522 | /* List of "safe" pages that may be used to store data loaded from the suspend |
1523 | * image | |
1524 | */ | |
1525 | static struct linked_page *safe_pages_list; | |
1526 | ||
1527 | #ifdef CONFIG_HIGHMEM | |
1528 | /* struct highmem_pbe is used for creating the list of highmem pages that | |
1529 | * should be restored atomically during the resume from disk, because the page | |
1530 | * frames they have occupied before the suspend are in use. | |
1531 | */ | |
1532 | struct highmem_pbe { | |
1533 | struct page *copy_page; /* data is here now */ | |
1534 | struct page *orig_page; /* data was here before the suspend */ | |
1535 | struct highmem_pbe *next; | |
1536 | }; | |
1537 | ||
1538 | /* List of highmem PBEs needed for restoring the highmem pages that were | |
1539 | * allocated before the suspend and included in the suspend image, but have | |
1540 | * also been allocated by the "resume" kernel, so their contents cannot be | |
1541 | * written directly to their "original" page frames. | |
1542 | */ | |
1543 | static struct highmem_pbe *highmem_pblist; | |
1544 | ||
1545 | /** | |
1546 | * count_highmem_image_pages - compute the number of highmem pages in the | |
1547 | * suspend image. The bits in the memory bitmap @bm that correspond to the | |
1548 | * image pages are assumed to be set. | |
1549 | */ | |
1550 | ||
1551 | static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) | |
1552 | { | |
1553 | unsigned long pfn; | |
1554 | unsigned int cnt = 0; | |
1555 | ||
1556 | memory_bm_position_reset(bm); | |
1557 | pfn = memory_bm_next_pfn(bm); | |
1558 | while (pfn != BM_END_OF_MAP) { | |
1559 | if (PageHighMem(pfn_to_page(pfn))) | |
1560 | cnt++; | |
1561 | ||
1562 | pfn = memory_bm_next_pfn(bm); | |
1563 | } | |
1564 | return cnt; | |
1565 | } | |
1566 | ||
1567 | /** | |
1568 | * prepare_highmem_image - try to allocate as many highmem pages as | |
1569 | * there are highmem image pages (@nr_highmem_p points to the variable | |
1570 | * containing the number of highmem image pages). The pages that are | |
1571 | * "safe" (ie. will not be overwritten when the suspend image is | |
1572 | * restored) have the corresponding bits set in @bm (it must be | |
1573 | * unitialized). | |
1574 | * | |
1575 | * NOTE: This function should not be called if there are no highmem | |
1576 | * image pages. | |
1577 | */ | |
1578 | ||
1579 | static unsigned int safe_highmem_pages; | |
1580 | ||
1581 | static struct memory_bitmap *safe_highmem_bm; | |
1582 | ||
1583 | static int | |
1584 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
1585 | { | |
1586 | unsigned int to_alloc; | |
1587 | ||
1588 | if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE)) | |
1589 | return -ENOMEM; | |
1590 | ||
1591 | if (get_highmem_buffer(PG_SAFE)) | |
1592 | return -ENOMEM; | |
1593 | ||
1594 | to_alloc = count_free_highmem_pages(); | |
1595 | if (to_alloc > *nr_highmem_p) | |
1596 | to_alloc = *nr_highmem_p; | |
1597 | else | |
1598 | *nr_highmem_p = to_alloc; | |
1599 | ||
1600 | safe_highmem_pages = 0; | |
1601 | while (to_alloc-- > 0) { | |
1602 | struct page *page; | |
1603 | ||
1604 | page = alloc_page(__GFP_HIGHMEM); | |
7be98234 | 1605 | if (!swsusp_page_is_free(page)) { |
8357376d RW |
1606 | /* The page is "safe", set its bit the bitmap */ |
1607 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1608 | safe_highmem_pages++; | |
1609 | } | |
1610 | /* Mark the page as allocated */ | |
7be98234 RW |
1611 | swsusp_set_page_forbidden(page); |
1612 | swsusp_set_page_free(page); | |
8357376d RW |
1613 | } |
1614 | memory_bm_position_reset(bm); | |
1615 | safe_highmem_bm = bm; | |
1616 | return 0; | |
1617 | } | |
1618 | ||
1619 | /** | |
1620 | * get_highmem_page_buffer - for given highmem image page find the buffer | |
1621 | * that suspend_write_next() should set for its caller to write to. | |
1622 | * | |
1623 | * If the page is to be saved to its "original" page frame or a copy of | |
1624 | * the page is to be made in the highmem, @buffer is returned. Otherwise, | |
1625 | * the copy of the page is to be made in normal memory, so the address of | |
1626 | * the copy is returned. | |
1627 | * | |
1628 | * If @buffer is returned, the caller of suspend_write_next() will write | |
1629 | * the page's contents to @buffer, so they will have to be copied to the | |
1630 | * right location on the next call to suspend_write_next() and it is done | |
1631 | * with the help of copy_last_highmem_page(). For this purpose, if | |
1632 | * @buffer is returned, @last_highmem page is set to the page to which | |
1633 | * the data will have to be copied from @buffer. | |
1634 | */ | |
1635 | ||
1636 | static struct page *last_highmem_page; | |
1637 | ||
1638 | static void * | |
1639 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
1640 | { | |
1641 | struct highmem_pbe *pbe; | |
1642 | void *kaddr; | |
1643 | ||
7be98234 | 1644 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) { |
8357376d RW |
1645 | /* We have allocated the "original" page frame and we can |
1646 | * use it directly to store the loaded page. | |
1647 | */ | |
1648 | last_highmem_page = page; | |
1649 | return buffer; | |
1650 | } | |
1651 | /* The "original" page frame has not been allocated and we have to | |
1652 | * use a "safe" page frame to store the loaded page. | |
1653 | */ | |
1654 | pbe = chain_alloc(ca, sizeof(struct highmem_pbe)); | |
1655 | if (!pbe) { | |
1656 | swsusp_free(); | |
1657 | return NULL; | |
1658 | } | |
1659 | pbe->orig_page = page; | |
1660 | if (safe_highmem_pages > 0) { | |
1661 | struct page *tmp; | |
1662 | ||
1663 | /* Copy of the page will be stored in high memory */ | |
1664 | kaddr = buffer; | |
1665 | tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm)); | |
1666 | safe_highmem_pages--; | |
1667 | last_highmem_page = tmp; | |
1668 | pbe->copy_page = tmp; | |
1669 | } else { | |
1670 | /* Copy of the page will be stored in normal memory */ | |
1671 | kaddr = safe_pages_list; | |
1672 | safe_pages_list = safe_pages_list->next; | |
1673 | pbe->copy_page = virt_to_page(kaddr); | |
1674 | } | |
1675 | pbe->next = highmem_pblist; | |
1676 | highmem_pblist = pbe; | |
1677 | return kaddr; | |
1678 | } | |
1679 | ||
1680 | /** | |
1681 | * copy_last_highmem_page - copy the contents of a highmem image from | |
1682 | * @buffer, where the caller of snapshot_write_next() has place them, | |
1683 | * to the right location represented by @last_highmem_page . | |
1684 | */ | |
1685 | ||
1686 | static void copy_last_highmem_page(void) | |
1687 | { | |
1688 | if (last_highmem_page) { | |
1689 | void *dst; | |
1690 | ||
1691 | dst = kmap_atomic(last_highmem_page, KM_USER0); | |
1692 | memcpy(dst, buffer, PAGE_SIZE); | |
1693 | kunmap_atomic(dst, KM_USER0); | |
1694 | last_highmem_page = NULL; | |
1695 | } | |
1696 | } | |
1697 | ||
1698 | static inline int last_highmem_page_copied(void) | |
1699 | { | |
1700 | return !last_highmem_page; | |
1701 | } | |
1702 | ||
1703 | static inline void free_highmem_data(void) | |
1704 | { | |
1705 | if (safe_highmem_bm) | |
1706 | memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR); | |
1707 | ||
1708 | if (buffer) | |
1709 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
1710 | } | |
1711 | #else | |
1712 | static inline int get_safe_write_buffer(void) { return 0; } | |
1713 | ||
1714 | static unsigned int | |
1715 | count_highmem_image_pages(struct memory_bitmap *bm) { return 0; } | |
1716 | ||
1717 | static inline int | |
1718 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
1719 | { | |
1720 | return 0; | |
1721 | } | |
1722 | ||
1723 | static inline void * | |
1724 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
1725 | { | |
1726 | return NULL; | |
1727 | } | |
1728 | ||
1729 | static inline void copy_last_highmem_page(void) {} | |
1730 | static inline int last_highmem_page_copied(void) { return 1; } | |
1731 | static inline void free_highmem_data(void) {} | |
1732 | #endif /* CONFIG_HIGHMEM */ | |
1733 | ||
f577eb30 | 1734 | /** |
940864dd RW |
1735 | * prepare_image - use the memory bitmap @bm to mark the pages that will |
1736 | * be overwritten in the process of restoring the system memory state | |
1737 | * from the suspend image ("unsafe" pages) and allocate memory for the | |
1738 | * image. | |
968808b8 | 1739 | * |
940864dd RW |
1740 | * The idea is to allocate a new memory bitmap first and then allocate |
1741 | * as many pages as needed for the image data, but not to assign these | |
1742 | * pages to specific tasks initially. Instead, we just mark them as | |
8357376d RW |
1743 | * allocated and create a lists of "safe" pages that will be used |
1744 | * later. On systems with high memory a list of "safe" highmem pages is | |
1745 | * also created. | |
f577eb30 RW |
1746 | */ |
1747 | ||
940864dd RW |
1748 | #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe)) |
1749 | ||
940864dd RW |
1750 | static int |
1751 | prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm) | |
f577eb30 | 1752 | { |
8357376d | 1753 | unsigned int nr_pages, nr_highmem; |
940864dd RW |
1754 | struct linked_page *sp_list, *lp; |
1755 | int error; | |
f577eb30 | 1756 | |
8357376d RW |
1757 | /* If there is no highmem, the buffer will not be necessary */ |
1758 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
1759 | buffer = NULL; | |
1760 | ||
1761 | nr_highmem = count_highmem_image_pages(bm); | |
940864dd RW |
1762 | error = mark_unsafe_pages(bm); |
1763 | if (error) | |
1764 | goto Free; | |
1765 | ||
1766 | error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE); | |
1767 | if (error) | |
1768 | goto Free; | |
1769 | ||
1770 | duplicate_memory_bitmap(new_bm, bm); | |
1771 | memory_bm_free(bm, PG_UNSAFE_KEEP); | |
8357376d RW |
1772 | if (nr_highmem > 0) { |
1773 | error = prepare_highmem_image(bm, &nr_highmem); | |
1774 | if (error) | |
1775 | goto Free; | |
1776 | } | |
940864dd RW |
1777 | /* Reserve some safe pages for potential later use. |
1778 | * | |
1779 | * NOTE: This way we make sure there will be enough safe pages for the | |
1780 | * chain_alloc() in get_buffer(). It is a bit wasteful, but | |
1781 | * nr_copy_pages cannot be greater than 50% of the memory anyway. | |
1782 | */ | |
1783 | sp_list = NULL; | |
1784 | /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */ | |
8357376d | 1785 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
1786 | nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE); |
1787 | while (nr_pages > 0) { | |
8357376d | 1788 | lp = get_image_page(GFP_ATOMIC, PG_SAFE); |
940864dd | 1789 | if (!lp) { |
f577eb30 | 1790 | error = -ENOMEM; |
940864dd RW |
1791 | goto Free; |
1792 | } | |
1793 | lp->next = sp_list; | |
1794 | sp_list = lp; | |
1795 | nr_pages--; | |
f577eb30 | 1796 | } |
940864dd RW |
1797 | /* Preallocate memory for the image */ |
1798 | safe_pages_list = NULL; | |
8357376d | 1799 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
1800 | while (nr_pages > 0) { |
1801 | lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC); | |
1802 | if (!lp) { | |
1803 | error = -ENOMEM; | |
1804 | goto Free; | |
1805 | } | |
7be98234 | 1806 | if (!swsusp_page_is_free(virt_to_page(lp))) { |
940864dd RW |
1807 | /* The page is "safe", add it to the list */ |
1808 | lp->next = safe_pages_list; | |
1809 | safe_pages_list = lp; | |
968808b8 | 1810 | } |
940864dd | 1811 | /* Mark the page as allocated */ |
7be98234 RW |
1812 | swsusp_set_page_forbidden(virt_to_page(lp)); |
1813 | swsusp_set_page_free(virt_to_page(lp)); | |
940864dd | 1814 | nr_pages--; |
968808b8 | 1815 | } |
940864dd RW |
1816 | /* Free the reserved safe pages so that chain_alloc() can use them */ |
1817 | while (sp_list) { | |
1818 | lp = sp_list->next; | |
1819 | free_image_page(sp_list, PG_UNSAFE_CLEAR); | |
1820 | sp_list = lp; | |
f577eb30 | 1821 | } |
940864dd RW |
1822 | return 0; |
1823 | ||
59a49335 | 1824 | Free: |
940864dd | 1825 | swsusp_free(); |
f577eb30 RW |
1826 | return error; |
1827 | } | |
1828 | ||
940864dd RW |
1829 | /** |
1830 | * get_buffer - compute the address that snapshot_write_next() should | |
1831 | * set for its caller to write to. | |
1832 | */ | |
1833 | ||
1834 | static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca) | |
968808b8 | 1835 | { |
940864dd RW |
1836 | struct pbe *pbe; |
1837 | struct page *page = pfn_to_page(memory_bm_next_pfn(bm)); | |
968808b8 | 1838 | |
8357376d RW |
1839 | if (PageHighMem(page)) |
1840 | return get_highmem_page_buffer(page, ca); | |
1841 | ||
7be98234 | 1842 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) |
940864dd RW |
1843 | /* We have allocated the "original" page frame and we can |
1844 | * use it directly to store the loaded page. | |
968808b8 | 1845 | */ |
940864dd RW |
1846 | return page_address(page); |
1847 | ||
1848 | /* The "original" page frame has not been allocated and we have to | |
1849 | * use a "safe" page frame to store the loaded page. | |
968808b8 | 1850 | */ |
940864dd RW |
1851 | pbe = chain_alloc(ca, sizeof(struct pbe)); |
1852 | if (!pbe) { | |
1853 | swsusp_free(); | |
1854 | return NULL; | |
1855 | } | |
8357376d RW |
1856 | pbe->orig_address = page_address(page); |
1857 | pbe->address = safe_pages_list; | |
940864dd RW |
1858 | safe_pages_list = safe_pages_list->next; |
1859 | pbe->next = restore_pblist; | |
1860 | restore_pblist = pbe; | |
8357376d | 1861 | return pbe->address; |
968808b8 RW |
1862 | } |
1863 | ||
f577eb30 RW |
1864 | /** |
1865 | * snapshot_write_next - used for writing the system memory snapshot. | |
1866 | * | |
1867 | * On the first call to it @handle should point to a zeroed | |
1868 | * snapshot_handle structure. The structure gets updated and a pointer | |
1869 | * to it should be passed to this function every next time. | |
1870 | * | |
1871 | * The @count parameter should contain the number of bytes the caller | |
1872 | * wants to write to the image. It must not be zero. | |
1873 | * | |
1874 | * On success the function returns a positive number. Then, the caller | |
1875 | * is allowed to write up to the returned number of bytes to the memory | |
1876 | * location computed by the data_of() macro. The number returned | |
1877 | * may be smaller than @count, but this only happens if the write would | |
1878 | * cross a page boundary otherwise. | |
1879 | * | |
1880 | * The function returns 0 to indicate the "end of file" condition, | |
1881 | * and a negative number is returned on error. In such cases the | |
1882 | * structure pointed to by @handle is not updated and should not be used | |
1883 | * any more. | |
1884 | */ | |
1885 | ||
1886 | int snapshot_write_next(struct snapshot_handle *handle, size_t count) | |
1887 | { | |
940864dd | 1888 | static struct chain_allocator ca; |
f577eb30 RW |
1889 | int error = 0; |
1890 | ||
940864dd | 1891 | /* Check if we have already loaded the entire image */ |
fb13a28b | 1892 | if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 1893 | return 0; |
940864dd | 1894 | |
8357376d RW |
1895 | if (handle->offset == 0) { |
1896 | if (!buffer) | |
1897 | /* This makes the buffer be freed by swsusp_free() */ | |
1898 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); | |
1899 | ||
f577eb30 RW |
1900 | if (!buffer) |
1901 | return -ENOMEM; | |
8357376d | 1902 | |
f577eb30 | 1903 | handle->buffer = buffer; |
8357376d | 1904 | } |
546e0d27 | 1905 | handle->sync_read = 1; |
fb13a28b | 1906 | if (handle->prev < handle->cur) { |
940864dd RW |
1907 | if (handle->prev == 0) { |
1908 | error = load_header(buffer); | |
1909 | if (error) | |
1910 | return error; | |
1911 | ||
1912 | error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY); | |
f577eb30 RW |
1913 | if (error) |
1914 | return error; | |
940864dd | 1915 | |
f577eb30 | 1916 | } else if (handle->prev <= nr_meta_pages) { |
940864dd RW |
1917 | unpack_orig_pfns(buffer, ©_bm); |
1918 | if (handle->prev == nr_meta_pages) { | |
1919 | error = prepare_image(&orig_bm, ©_bm); | |
f577eb30 RW |
1920 | if (error) |
1921 | return error; | |
940864dd RW |
1922 | |
1923 | chain_init(&ca, GFP_ATOMIC, PG_SAFE); | |
1924 | memory_bm_position_reset(&orig_bm); | |
1925 | restore_pblist = NULL; | |
1926 | handle->buffer = get_buffer(&orig_bm, &ca); | |
546e0d27 | 1927 | handle->sync_read = 0; |
940864dd RW |
1928 | if (!handle->buffer) |
1929 | return -ENOMEM; | |
f577eb30 RW |
1930 | } |
1931 | } else { | |
8357376d | 1932 | copy_last_highmem_page(); |
940864dd | 1933 | handle->buffer = get_buffer(&orig_bm, &ca); |
8357376d RW |
1934 | if (handle->buffer != buffer) |
1935 | handle->sync_read = 0; | |
f577eb30 | 1936 | } |
fb13a28b | 1937 | handle->prev = handle->cur; |
f577eb30 | 1938 | } |
fb13a28b RW |
1939 | handle->buf_offset = handle->cur_offset; |
1940 | if (handle->cur_offset + count >= PAGE_SIZE) { | |
1941 | count = PAGE_SIZE - handle->cur_offset; | |
1942 | handle->cur_offset = 0; | |
1943 | handle->cur++; | |
f577eb30 | 1944 | } else { |
fb13a28b | 1945 | handle->cur_offset += count; |
f577eb30 RW |
1946 | } |
1947 | handle->offset += count; | |
1948 | return count; | |
1949 | } | |
1950 | ||
8357376d RW |
1951 | /** |
1952 | * snapshot_write_finalize - must be called after the last call to | |
1953 | * snapshot_write_next() in case the last page in the image happens | |
1954 | * to be a highmem page and its contents should be stored in the | |
1955 | * highmem. Additionally, it releases the memory that will not be | |
1956 | * used any more. | |
1957 | */ | |
1958 | ||
1959 | void snapshot_write_finalize(struct snapshot_handle *handle) | |
1960 | { | |
1961 | copy_last_highmem_page(); | |
1962 | /* Free only if we have loaded the image entirely */ | |
1963 | if (handle->prev && handle->cur > nr_meta_pages + nr_copy_pages) { | |
1964 | memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR); | |
1965 | free_highmem_data(); | |
1966 | } | |
1967 | } | |
1968 | ||
f577eb30 RW |
1969 | int snapshot_image_loaded(struct snapshot_handle *handle) |
1970 | { | |
8357376d | 1971 | return !(!nr_copy_pages || !last_highmem_page_copied() || |
940864dd RW |
1972 | handle->cur <= nr_meta_pages + nr_copy_pages); |
1973 | } | |
1974 | ||
8357376d RW |
1975 | #ifdef CONFIG_HIGHMEM |
1976 | /* Assumes that @buf is ready and points to a "safe" page */ | |
1977 | static inline void | |
1978 | swap_two_pages_data(struct page *p1, struct page *p2, void *buf) | |
940864dd | 1979 | { |
8357376d RW |
1980 | void *kaddr1, *kaddr2; |
1981 | ||
1982 | kaddr1 = kmap_atomic(p1, KM_USER0); | |
1983 | kaddr2 = kmap_atomic(p2, KM_USER1); | |
1984 | memcpy(buf, kaddr1, PAGE_SIZE); | |
1985 | memcpy(kaddr1, kaddr2, PAGE_SIZE); | |
1986 | memcpy(kaddr2, buf, PAGE_SIZE); | |
1987 | kunmap_atomic(kaddr1, KM_USER0); | |
1988 | kunmap_atomic(kaddr2, KM_USER1); | |
1989 | } | |
1990 | ||
1991 | /** | |
1992 | * restore_highmem - for each highmem page that was allocated before | |
1993 | * the suspend and included in the suspend image, and also has been | |
1994 | * allocated by the "resume" kernel swap its current (ie. "before | |
1995 | * resume") contents with the previous (ie. "before suspend") one. | |
1996 | * | |
1997 | * If the resume eventually fails, we can call this function once | |
1998 | * again and restore the "before resume" highmem state. | |
1999 | */ | |
2000 | ||
2001 | int restore_highmem(void) | |
2002 | { | |
2003 | struct highmem_pbe *pbe = highmem_pblist; | |
2004 | void *buf; | |
2005 | ||
2006 | if (!pbe) | |
2007 | return 0; | |
2008 | ||
2009 | buf = get_image_page(GFP_ATOMIC, PG_SAFE); | |
2010 | if (!buf) | |
2011 | return -ENOMEM; | |
2012 | ||
2013 | while (pbe) { | |
2014 | swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf); | |
2015 | pbe = pbe->next; | |
2016 | } | |
2017 | free_image_page(buf, PG_UNSAFE_CLEAR); | |
2018 | return 0; | |
f577eb30 | 2019 | } |
8357376d | 2020 | #endif /* CONFIG_HIGHMEM */ |