]>
Commit | Line | Data |
---|---|---|
25761b6e | 1 | /* |
96bc7aec | 2 | * linux/kernel/power/snapshot.c |
25761b6e | 3 | * |
8357376d | 4 | * This file provides system snapshot/restore functionality for swsusp. |
25761b6e | 5 | * |
a2531293 | 6 | * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz> |
8357376d | 7 | * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl> |
25761b6e | 8 | * |
8357376d | 9 | * This file is released under the GPLv2. |
25761b6e RW |
10 | * |
11 | */ | |
12 | ||
f577eb30 | 13 | #include <linux/version.h> |
25761b6e RW |
14 | #include <linux/module.h> |
15 | #include <linux/mm.h> | |
16 | #include <linux/suspend.h> | |
25761b6e | 17 | #include <linux/delay.h> |
25761b6e | 18 | #include <linux/bitops.h> |
25761b6e | 19 | #include <linux/spinlock.h> |
25761b6e | 20 | #include <linux/kernel.h> |
25761b6e RW |
21 | #include <linux/pm.h> |
22 | #include <linux/device.h> | |
74dfd666 | 23 | #include <linux/init.h> |
25761b6e RW |
24 | #include <linux/bootmem.h> |
25 | #include <linux/syscalls.h> | |
26 | #include <linux/console.h> | |
27 | #include <linux/highmem.h> | |
846705de | 28 | #include <linux/list.h> |
5a0e3ad6 | 29 | #include <linux/slab.h> |
52f5684c | 30 | #include <linux/compiler.h> |
25761b6e RW |
31 | |
32 | #include <asm/uaccess.h> | |
33 | #include <asm/mmu_context.h> | |
34 | #include <asm/pgtable.h> | |
35 | #include <asm/tlbflush.h> | |
36 | #include <asm/io.h> | |
37 | ||
25761b6e RW |
38 | #include "power.h" |
39 | ||
74dfd666 RW |
40 | static int swsusp_page_is_free(struct page *); |
41 | static void swsusp_set_page_forbidden(struct page *); | |
42 | static void swsusp_unset_page_forbidden(struct page *); | |
43 | ||
ddeb6487 RW |
44 | /* |
45 | * Number of bytes to reserve for memory allocations made by device drivers | |
46 | * from their ->freeze() and ->freeze_noirq() callbacks so that they don't | |
47 | * cause image creation to fail (tunable via /sys/power/reserved_size). | |
48 | */ | |
49 | unsigned long reserved_size; | |
50 | ||
51 | void __init hibernate_reserved_size_init(void) | |
52 | { | |
53 | reserved_size = SPARE_PAGES * PAGE_SIZE; | |
54 | } | |
55 | ||
fe419535 RW |
56 | /* |
57 | * Preferred image size in bytes (tunable via /sys/power/image_size). | |
1c1be3a9 RW |
58 | * When it is set to N, swsusp will do its best to ensure the image |
59 | * size will not exceed N bytes, but if that is impossible, it will | |
60 | * try to create the smallest image possible. | |
fe419535 | 61 | */ |
ac5c24ec RW |
62 | unsigned long image_size; |
63 | ||
64 | void __init hibernate_image_size_init(void) | |
65 | { | |
1c1be3a9 | 66 | image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE; |
ac5c24ec | 67 | } |
fe419535 | 68 | |
8357376d RW |
69 | /* List of PBEs needed for restoring the pages that were allocated before |
70 | * the suspend and included in the suspend image, but have also been | |
71 | * allocated by the "resume" kernel, so their contents cannot be written | |
72 | * directly to their "original" page frames. | |
73 | */ | |
75534b50 RW |
74 | struct pbe *restore_pblist; |
75 | ||
8357376d | 76 | /* Pointer to an auxiliary buffer (1 page) */ |
940864dd | 77 | static void *buffer; |
7088a5c0 | 78 | |
f6143aa6 RW |
79 | /** |
80 | * @safe_needed - on resume, for storing the PBE list and the image, | |
81 | * we can only use memory pages that do not conflict with the pages | |
8357376d RW |
82 | * used before suspend. The unsafe pages have PageNosaveFree set |
83 | * and we count them using unsafe_pages. | |
f6143aa6 | 84 | * |
8357376d RW |
85 | * Each allocated image page is marked as PageNosave and PageNosaveFree |
86 | * so that swsusp_free() can release it. | |
f6143aa6 RW |
87 | */ |
88 | ||
0bcd888d RW |
89 | #define PG_ANY 0 |
90 | #define PG_SAFE 1 | |
91 | #define PG_UNSAFE_CLEAR 1 | |
92 | #define PG_UNSAFE_KEEP 0 | |
93 | ||
940864dd | 94 | static unsigned int allocated_unsafe_pages; |
f6143aa6 | 95 | |
8357376d | 96 | static void *get_image_page(gfp_t gfp_mask, int safe_needed) |
f6143aa6 RW |
97 | { |
98 | void *res; | |
99 | ||
100 | res = (void *)get_zeroed_page(gfp_mask); | |
101 | if (safe_needed) | |
7be98234 | 102 | while (res && swsusp_page_is_free(virt_to_page(res))) { |
f6143aa6 | 103 | /* The page is unsafe, mark it for swsusp_free() */ |
7be98234 | 104 | swsusp_set_page_forbidden(virt_to_page(res)); |
940864dd | 105 | allocated_unsafe_pages++; |
f6143aa6 RW |
106 | res = (void *)get_zeroed_page(gfp_mask); |
107 | } | |
108 | if (res) { | |
7be98234 RW |
109 | swsusp_set_page_forbidden(virt_to_page(res)); |
110 | swsusp_set_page_free(virt_to_page(res)); | |
f6143aa6 RW |
111 | } |
112 | return res; | |
113 | } | |
114 | ||
115 | unsigned long get_safe_page(gfp_t gfp_mask) | |
116 | { | |
8357376d RW |
117 | return (unsigned long)get_image_page(gfp_mask, PG_SAFE); |
118 | } | |
119 | ||
5b6d15de RW |
120 | static struct page *alloc_image_page(gfp_t gfp_mask) |
121 | { | |
8357376d RW |
122 | struct page *page; |
123 | ||
124 | page = alloc_page(gfp_mask); | |
125 | if (page) { | |
7be98234 RW |
126 | swsusp_set_page_forbidden(page); |
127 | swsusp_set_page_free(page); | |
8357376d RW |
128 | } |
129 | return page; | |
f6143aa6 RW |
130 | } |
131 | ||
132 | /** | |
133 | * free_image_page - free page represented by @addr, allocated with | |
8357376d | 134 | * get_image_page (page flags set by it must be cleared) |
f6143aa6 RW |
135 | */ |
136 | ||
137 | static inline void free_image_page(void *addr, int clear_nosave_free) | |
138 | { | |
8357376d RW |
139 | struct page *page; |
140 | ||
141 | BUG_ON(!virt_addr_valid(addr)); | |
142 | ||
143 | page = virt_to_page(addr); | |
144 | ||
7be98234 | 145 | swsusp_unset_page_forbidden(page); |
f6143aa6 | 146 | if (clear_nosave_free) |
7be98234 | 147 | swsusp_unset_page_free(page); |
8357376d RW |
148 | |
149 | __free_page(page); | |
f6143aa6 RW |
150 | } |
151 | ||
b788db79 RW |
152 | /* struct linked_page is used to build chains of pages */ |
153 | ||
154 | #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *)) | |
155 | ||
156 | struct linked_page { | |
157 | struct linked_page *next; | |
158 | char data[LINKED_PAGE_DATA_SIZE]; | |
52f5684c | 159 | } __packed; |
b788db79 RW |
160 | |
161 | static inline void | |
162 | free_list_of_pages(struct linked_page *list, int clear_page_nosave) | |
163 | { | |
164 | while (list) { | |
165 | struct linked_page *lp = list->next; | |
166 | ||
167 | free_image_page(list, clear_page_nosave); | |
168 | list = lp; | |
169 | } | |
170 | } | |
171 | ||
172 | /** | |
173 | * struct chain_allocator is used for allocating small objects out of | |
174 | * a linked list of pages called 'the chain'. | |
175 | * | |
176 | * The chain grows each time when there is no room for a new object in | |
177 | * the current page. The allocated objects cannot be freed individually. | |
178 | * It is only possible to free them all at once, by freeing the entire | |
179 | * chain. | |
180 | * | |
181 | * NOTE: The chain allocator may be inefficient if the allocated objects | |
182 | * are not much smaller than PAGE_SIZE. | |
183 | */ | |
184 | ||
185 | struct chain_allocator { | |
186 | struct linked_page *chain; /* the chain */ | |
187 | unsigned int used_space; /* total size of objects allocated out | |
188 | * of the current page | |
189 | */ | |
190 | gfp_t gfp_mask; /* mask for allocating pages */ | |
191 | int safe_needed; /* if set, only "safe" pages are allocated */ | |
192 | }; | |
193 | ||
194 | static void | |
195 | chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed) | |
196 | { | |
197 | ca->chain = NULL; | |
198 | ca->used_space = LINKED_PAGE_DATA_SIZE; | |
199 | ca->gfp_mask = gfp_mask; | |
200 | ca->safe_needed = safe_needed; | |
201 | } | |
202 | ||
203 | static void *chain_alloc(struct chain_allocator *ca, unsigned int size) | |
204 | { | |
205 | void *ret; | |
206 | ||
207 | if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) { | |
208 | struct linked_page *lp; | |
209 | ||
8357376d | 210 | lp = get_image_page(ca->gfp_mask, ca->safe_needed); |
b788db79 RW |
211 | if (!lp) |
212 | return NULL; | |
213 | ||
214 | lp->next = ca->chain; | |
215 | ca->chain = lp; | |
216 | ca->used_space = 0; | |
217 | } | |
218 | ret = ca->chain->data + ca->used_space; | |
219 | ca->used_space += size; | |
220 | return ret; | |
221 | } | |
222 | ||
b788db79 RW |
223 | /** |
224 | * Data types related to memory bitmaps. | |
225 | * | |
226 | * Memory bitmap is a structure consiting of many linked lists of | |
227 | * objects. The main list's elements are of type struct zone_bitmap | |
228 | * and each of them corresonds to one zone. For each zone bitmap | |
229 | * object there is a list of objects of type struct bm_block that | |
0d83304c | 230 | * represent each blocks of bitmap in which information is stored. |
b788db79 RW |
231 | * |
232 | * struct memory_bitmap contains a pointer to the main list of zone | |
233 | * bitmap objects, a struct bm_position used for browsing the bitmap, | |
234 | * and a pointer to the list of pages used for allocating all of the | |
235 | * zone bitmap objects and bitmap block objects. | |
236 | * | |
237 | * NOTE: It has to be possible to lay out the bitmap in memory | |
238 | * using only allocations of order 0. Additionally, the bitmap is | |
239 | * designed to work with arbitrary number of zones (this is over the | |
240 | * top for now, but let's avoid making unnecessary assumptions ;-). | |
241 | * | |
242 | * struct zone_bitmap contains a pointer to a list of bitmap block | |
243 | * objects and a pointer to the bitmap block object that has been | |
244 | * most recently used for setting bits. Additionally, it contains the | |
245 | * pfns that correspond to the start and end of the represented zone. | |
246 | * | |
247 | * struct bm_block contains a pointer to the memory page in which | |
0d83304c AM |
248 | * information is stored (in the form of a block of bitmap) |
249 | * It also contains the pfns that correspond to the start and end of | |
250 | * the represented memory area. | |
f469f02d JR |
251 | * |
252 | * The memory bitmap is organized as a radix tree to guarantee fast random | |
253 | * access to the bits. There is one radix tree for each zone (as returned | |
254 | * from create_mem_extents). | |
255 | * | |
256 | * One radix tree is represented by one struct mem_zone_bm_rtree. There are | |
257 | * two linked lists for the nodes of the tree, one for the inner nodes and | |
258 | * one for the leave nodes. The linked leave nodes are used for fast linear | |
259 | * access of the memory bitmap. | |
260 | * | |
261 | * The struct rtree_node represents one node of the radix tree. | |
b788db79 RW |
262 | */ |
263 | ||
264 | #define BM_END_OF_MAP (~0UL) | |
265 | ||
8de03073 | 266 | #define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE) |
f469f02d JR |
267 | #define BM_BLOCK_SHIFT (PAGE_SHIFT + 3) |
268 | #define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1) | |
b788db79 RW |
269 | |
270 | struct bm_block { | |
846705de | 271 | struct list_head hook; /* hook into a list of bitmap blocks */ |
b788db79 RW |
272 | unsigned long start_pfn; /* pfn represented by the first bit */ |
273 | unsigned long end_pfn; /* pfn represented by the last bit plus 1 */ | |
0d83304c | 274 | unsigned long *data; /* bitmap representing pages */ |
b788db79 RW |
275 | }; |
276 | ||
0d83304c AM |
277 | static inline unsigned long bm_block_bits(struct bm_block *bb) |
278 | { | |
279 | return bb->end_pfn - bb->start_pfn; | |
280 | } | |
281 | ||
f469f02d JR |
282 | /* |
283 | * struct rtree_node is a wrapper struct to link the nodes | |
284 | * of the rtree together for easy linear iteration over | |
285 | * bits and easy freeing | |
286 | */ | |
287 | struct rtree_node { | |
288 | struct list_head list; | |
289 | unsigned long *data; | |
290 | }; | |
291 | ||
292 | /* | |
293 | * struct mem_zone_bm_rtree represents a bitmap used for one | |
294 | * populated memory zone. | |
295 | */ | |
296 | struct mem_zone_bm_rtree { | |
297 | struct list_head list; /* Link Zones together */ | |
298 | struct list_head nodes; /* Radix Tree inner nodes */ | |
299 | struct list_head leaves; /* Radix Tree leaves */ | |
300 | unsigned long start_pfn; /* Zone start page frame */ | |
301 | unsigned long end_pfn; /* Zone end page frame + 1 */ | |
302 | struct rtree_node *rtree; /* Radix Tree Root */ | |
303 | int levels; /* Number of Radix Tree Levels */ | |
304 | unsigned int blocks; /* Number of Bitmap Blocks */ | |
305 | }; | |
306 | ||
b788db79 RW |
307 | /* strcut bm_position is used for browsing memory bitmaps */ |
308 | ||
309 | struct bm_position { | |
b788db79 | 310 | struct bm_block *block; |
b788db79 RW |
311 | int bit; |
312 | }; | |
313 | ||
314 | struct memory_bitmap { | |
f469f02d | 315 | struct list_head zones; |
846705de | 316 | struct list_head blocks; /* list of bitmap blocks */ |
b788db79 RW |
317 | struct linked_page *p_list; /* list of pages used to store zone |
318 | * bitmap objects and bitmap block | |
319 | * objects | |
320 | */ | |
321 | struct bm_position cur; /* most recently used bit position */ | |
322 | }; | |
323 | ||
324 | /* Functions that operate on memory bitmaps */ | |
325 | ||
f469f02d JR |
326 | #define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long)) |
327 | #if BITS_PER_LONG == 32 | |
328 | #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2) | |
329 | #else | |
330 | #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3) | |
331 | #endif | |
332 | #define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1) | |
333 | ||
334 | /* | |
335 | * alloc_rtree_node - Allocate a new node and add it to the radix tree. | |
336 | * | |
337 | * This function is used to allocate inner nodes as well as the | |
338 | * leave nodes of the radix tree. It also adds the node to the | |
339 | * corresponding linked list passed in by the *list parameter. | |
340 | */ | |
341 | static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed, | |
342 | struct chain_allocator *ca, | |
343 | struct list_head *list) | |
344 | { | |
345 | struct rtree_node *node; | |
346 | ||
347 | node = chain_alloc(ca, sizeof(struct rtree_node)); | |
348 | if (!node) | |
349 | return NULL; | |
350 | ||
351 | node->data = get_image_page(gfp_mask, safe_needed); | |
352 | if (!node->data) | |
353 | return NULL; | |
354 | ||
355 | list_add_tail(&node->list, list); | |
356 | ||
357 | return node; | |
358 | } | |
359 | ||
360 | /* | |
361 | * add_rtree_block - Add a new leave node to the radix tree | |
362 | * | |
363 | * The leave nodes need to be allocated in order to keep the leaves | |
364 | * linked list in order. This is guaranteed by the zone->blocks | |
365 | * counter. | |
366 | */ | |
367 | static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask, | |
368 | int safe_needed, struct chain_allocator *ca) | |
369 | { | |
370 | struct rtree_node *node, *block, **dst; | |
371 | unsigned int levels_needed, block_nr; | |
372 | int i; | |
373 | ||
374 | block_nr = zone->blocks; | |
375 | levels_needed = 0; | |
376 | ||
377 | /* How many levels do we need for this block nr? */ | |
378 | while (block_nr) { | |
379 | levels_needed += 1; | |
380 | block_nr >>= BM_RTREE_LEVEL_SHIFT; | |
381 | } | |
382 | ||
383 | /* Make sure the rtree has enough levels */ | |
384 | for (i = zone->levels; i < levels_needed; i++) { | |
385 | node = alloc_rtree_node(gfp_mask, safe_needed, ca, | |
386 | &zone->nodes); | |
387 | if (!node) | |
388 | return -ENOMEM; | |
389 | ||
390 | node->data[0] = (unsigned long)zone->rtree; | |
391 | zone->rtree = node; | |
392 | zone->levels += 1; | |
393 | } | |
394 | ||
395 | /* Allocate new block */ | |
396 | block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves); | |
397 | if (!block) | |
398 | return -ENOMEM; | |
399 | ||
400 | /* Now walk the rtree to insert the block */ | |
401 | node = zone->rtree; | |
402 | dst = &zone->rtree; | |
403 | block_nr = zone->blocks; | |
404 | for (i = zone->levels; i > 0; i--) { | |
405 | int index; | |
406 | ||
407 | if (!node) { | |
408 | node = alloc_rtree_node(gfp_mask, safe_needed, ca, | |
409 | &zone->nodes); | |
410 | if (!node) | |
411 | return -ENOMEM; | |
412 | *dst = node; | |
413 | } | |
414 | ||
415 | index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT); | |
416 | index &= BM_RTREE_LEVEL_MASK; | |
417 | dst = (struct rtree_node **)&((*dst)->data[index]); | |
418 | node = *dst; | |
419 | } | |
420 | ||
421 | zone->blocks += 1; | |
422 | *dst = block; | |
423 | ||
424 | return 0; | |
425 | } | |
426 | ||
427 | static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone, | |
428 | int clear_nosave_free); | |
429 | ||
430 | /* | |
431 | * create_zone_bm_rtree - create a radix tree for one zone | |
432 | * | |
433 | * Allocated the mem_zone_bm_rtree structure and initializes it. | |
434 | * This function also allocated and builds the radix tree for the | |
435 | * zone. | |
436 | */ | |
437 | static struct mem_zone_bm_rtree * | |
438 | create_zone_bm_rtree(gfp_t gfp_mask, int safe_needed, | |
439 | struct chain_allocator *ca, | |
440 | unsigned long start, unsigned long end) | |
441 | { | |
442 | struct mem_zone_bm_rtree *zone; | |
443 | unsigned int i, nr_blocks; | |
444 | unsigned long pages; | |
445 | ||
446 | pages = end - start; | |
447 | zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree)); | |
448 | if (!zone) | |
449 | return NULL; | |
450 | ||
451 | INIT_LIST_HEAD(&zone->nodes); | |
452 | INIT_LIST_HEAD(&zone->leaves); | |
453 | zone->start_pfn = start; | |
454 | zone->end_pfn = end; | |
455 | nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK); | |
456 | ||
457 | for (i = 0; i < nr_blocks; i++) { | |
458 | if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) { | |
459 | free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR); | |
460 | return NULL; | |
461 | } | |
462 | } | |
463 | ||
464 | return zone; | |
465 | } | |
466 | ||
467 | /* | |
468 | * free_zone_bm_rtree - Free the memory of the radix tree | |
469 | * | |
470 | * Free all node pages of the radix tree. The mem_zone_bm_rtree | |
471 | * structure itself is not freed here nor are the rtree_node | |
472 | * structs. | |
473 | */ | |
474 | static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone, | |
475 | int clear_nosave_free) | |
476 | { | |
477 | struct rtree_node *node; | |
478 | ||
479 | list_for_each_entry(node, &zone->nodes, list) | |
480 | free_image_page(node->data, clear_nosave_free); | |
481 | ||
482 | list_for_each_entry(node, &zone->leaves, list) | |
483 | free_image_page(node->data, clear_nosave_free); | |
484 | } | |
485 | ||
b788db79 RW |
486 | static void memory_bm_position_reset(struct memory_bitmap *bm) |
487 | { | |
846705de | 488 | bm->cur.block = list_entry(bm->blocks.next, struct bm_block, hook); |
0d83304c | 489 | bm->cur.bit = 0; |
b788db79 RW |
490 | } |
491 | ||
492 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free); | |
493 | ||
494 | /** | |
495 | * create_bm_block_list - create a list of block bitmap objects | |
8de03073 | 496 | * @pages - number of pages to track |
846705de RW |
497 | * @list - list to put the allocated blocks into |
498 | * @ca - chain allocator to be used for allocating memory | |
b788db79 | 499 | */ |
846705de RW |
500 | static int create_bm_block_list(unsigned long pages, |
501 | struct list_head *list, | |
502 | struct chain_allocator *ca) | |
b788db79 | 503 | { |
846705de | 504 | unsigned int nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK); |
b788db79 RW |
505 | |
506 | while (nr_blocks-- > 0) { | |
507 | struct bm_block *bb; | |
508 | ||
509 | bb = chain_alloc(ca, sizeof(struct bm_block)); | |
510 | if (!bb) | |
846705de RW |
511 | return -ENOMEM; |
512 | list_add(&bb->hook, list); | |
b788db79 | 513 | } |
846705de RW |
514 | |
515 | return 0; | |
b788db79 RW |
516 | } |
517 | ||
846705de RW |
518 | struct mem_extent { |
519 | struct list_head hook; | |
520 | unsigned long start; | |
521 | unsigned long end; | |
522 | }; | |
523 | ||
b788db79 | 524 | /** |
846705de RW |
525 | * free_mem_extents - free a list of memory extents |
526 | * @list - list of extents to empty | |
b788db79 | 527 | */ |
846705de RW |
528 | static void free_mem_extents(struct list_head *list) |
529 | { | |
530 | struct mem_extent *ext, *aux; | |
b788db79 | 531 | |
846705de RW |
532 | list_for_each_entry_safe(ext, aux, list, hook) { |
533 | list_del(&ext->hook); | |
534 | kfree(ext); | |
535 | } | |
536 | } | |
537 | ||
538 | /** | |
539 | * create_mem_extents - create a list of memory extents representing | |
540 | * contiguous ranges of PFNs | |
541 | * @list - list to put the extents into | |
542 | * @gfp_mask - mask to use for memory allocations | |
543 | */ | |
544 | static int create_mem_extents(struct list_head *list, gfp_t gfp_mask) | |
b788db79 | 545 | { |
846705de | 546 | struct zone *zone; |
b788db79 | 547 | |
846705de | 548 | INIT_LIST_HEAD(list); |
b788db79 | 549 | |
ee99c71c | 550 | for_each_populated_zone(zone) { |
846705de RW |
551 | unsigned long zone_start, zone_end; |
552 | struct mem_extent *ext, *cur, *aux; | |
553 | ||
846705de | 554 | zone_start = zone->zone_start_pfn; |
c33bc315 | 555 | zone_end = zone_end_pfn(zone); |
846705de RW |
556 | |
557 | list_for_each_entry(ext, list, hook) | |
558 | if (zone_start <= ext->end) | |
559 | break; | |
b788db79 | 560 | |
846705de RW |
561 | if (&ext->hook == list || zone_end < ext->start) { |
562 | /* New extent is necessary */ | |
563 | struct mem_extent *new_ext; | |
564 | ||
565 | new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask); | |
566 | if (!new_ext) { | |
567 | free_mem_extents(list); | |
568 | return -ENOMEM; | |
569 | } | |
570 | new_ext->start = zone_start; | |
571 | new_ext->end = zone_end; | |
572 | list_add_tail(&new_ext->hook, &ext->hook); | |
573 | continue; | |
574 | } | |
575 | ||
576 | /* Merge this zone's range of PFNs with the existing one */ | |
577 | if (zone_start < ext->start) | |
578 | ext->start = zone_start; | |
579 | if (zone_end > ext->end) | |
580 | ext->end = zone_end; | |
581 | ||
582 | /* More merging may be possible */ | |
583 | cur = ext; | |
584 | list_for_each_entry_safe_continue(cur, aux, list, hook) { | |
585 | if (zone_end < cur->start) | |
586 | break; | |
587 | if (zone_end < cur->end) | |
588 | ext->end = cur->end; | |
589 | list_del(&cur->hook); | |
590 | kfree(cur); | |
591 | } | |
b788db79 | 592 | } |
846705de RW |
593 | |
594 | return 0; | |
b788db79 RW |
595 | } |
596 | ||
597 | /** | |
598 | * memory_bm_create - allocate memory for a memory bitmap | |
599 | */ | |
b788db79 RW |
600 | static int |
601 | memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed) | |
602 | { | |
603 | struct chain_allocator ca; | |
846705de RW |
604 | struct list_head mem_extents; |
605 | struct mem_extent *ext; | |
606 | int error; | |
b788db79 RW |
607 | |
608 | chain_init(&ca, gfp_mask, safe_needed); | |
846705de | 609 | INIT_LIST_HEAD(&bm->blocks); |
f469f02d | 610 | INIT_LIST_HEAD(&bm->zones); |
b788db79 | 611 | |
846705de RW |
612 | error = create_mem_extents(&mem_extents, gfp_mask); |
613 | if (error) | |
614 | return error; | |
b788db79 | 615 | |
846705de | 616 | list_for_each_entry(ext, &mem_extents, hook) { |
f469f02d | 617 | struct mem_zone_bm_rtree *zone; |
846705de RW |
618 | struct bm_block *bb; |
619 | unsigned long pfn = ext->start; | |
620 | unsigned long pages = ext->end - ext->start; | |
b788db79 | 621 | |
846705de | 622 | bb = list_entry(bm->blocks.prev, struct bm_block, hook); |
b788db79 | 623 | |
846705de RW |
624 | error = create_bm_block_list(pages, bm->blocks.prev, &ca); |
625 | if (error) | |
626 | goto Error; | |
b788db79 | 627 | |
846705de RW |
628 | list_for_each_entry_continue(bb, &bm->blocks, hook) { |
629 | bb->data = get_image_page(gfp_mask, safe_needed); | |
630 | if (!bb->data) { | |
631 | error = -ENOMEM; | |
632 | goto Error; | |
633 | } | |
b788db79 RW |
634 | |
635 | bb->start_pfn = pfn; | |
846705de | 636 | if (pages >= BM_BITS_PER_BLOCK) { |
b788db79 | 637 | pfn += BM_BITS_PER_BLOCK; |
846705de | 638 | pages -= BM_BITS_PER_BLOCK; |
b788db79 RW |
639 | } else { |
640 | /* This is executed only once in the loop */ | |
846705de | 641 | pfn += pages; |
b788db79 RW |
642 | } |
643 | bb->end_pfn = pfn; | |
b788db79 | 644 | } |
f469f02d JR |
645 | |
646 | zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca, | |
647 | ext->start, ext->end); | |
648 | if (!zone) | |
649 | goto Error; | |
650 | list_add_tail(&zone->list, &bm->zones); | |
b788db79 | 651 | } |
846705de | 652 | |
b788db79 RW |
653 | bm->p_list = ca.chain; |
654 | memory_bm_position_reset(bm); | |
846705de RW |
655 | Exit: |
656 | free_mem_extents(&mem_extents); | |
657 | return error; | |
b788db79 | 658 | |
846705de | 659 | Error: |
b788db79 RW |
660 | bm->p_list = ca.chain; |
661 | memory_bm_free(bm, PG_UNSAFE_CLEAR); | |
846705de | 662 | goto Exit; |
b788db79 RW |
663 | } |
664 | ||
665 | /** | |
666 | * memory_bm_free - free memory occupied by the memory bitmap @bm | |
667 | */ | |
b788db79 RW |
668 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free) |
669 | { | |
f469f02d | 670 | struct mem_zone_bm_rtree *zone; |
846705de | 671 | struct bm_block *bb; |
b788db79 | 672 | |
846705de RW |
673 | list_for_each_entry(bb, &bm->blocks, hook) |
674 | if (bb->data) | |
675 | free_image_page(bb->data, clear_nosave_free); | |
b788db79 | 676 | |
f469f02d JR |
677 | list_for_each_entry(zone, &bm->zones, list) |
678 | free_zone_bm_rtree(zone, clear_nosave_free); | |
679 | ||
b788db79 | 680 | free_list_of_pages(bm->p_list, clear_nosave_free); |
846705de | 681 | |
f469f02d | 682 | INIT_LIST_HEAD(&bm->zones); |
846705de | 683 | INIT_LIST_HEAD(&bm->blocks); |
b788db79 RW |
684 | } |
685 | ||
686 | /** | |
74dfd666 | 687 | * memory_bm_find_bit - find the bit in the bitmap @bm that corresponds |
b788db79 RW |
688 | * to given pfn. The cur_zone_bm member of @bm and the cur_block member |
689 | * of @bm->cur_zone_bm are updated. | |
b788db79 | 690 | */ |
a82f7119 | 691 | static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn, |
74dfd666 | 692 | void **addr, unsigned int *bit_nr) |
b788db79 | 693 | { |
b788db79 RW |
694 | struct bm_block *bb; |
695 | ||
846705de RW |
696 | /* |
697 | * Check if the pfn corresponds to the current bitmap block and find | |
698 | * the block where it fits if this is not the case. | |
699 | */ | |
700 | bb = bm->cur.block; | |
b788db79 | 701 | if (pfn < bb->start_pfn) |
846705de RW |
702 | list_for_each_entry_continue_reverse(bb, &bm->blocks, hook) |
703 | if (pfn >= bb->start_pfn) | |
704 | break; | |
b788db79 | 705 | |
846705de RW |
706 | if (pfn >= bb->end_pfn) |
707 | list_for_each_entry_continue(bb, &bm->blocks, hook) | |
708 | if (pfn >= bb->start_pfn && pfn < bb->end_pfn) | |
709 | break; | |
74dfd666 | 710 | |
846705de RW |
711 | if (&bb->hook == &bm->blocks) |
712 | return -EFAULT; | |
713 | ||
714 | /* The block has been found */ | |
715 | bm->cur.block = bb; | |
b788db79 | 716 | pfn -= bb->start_pfn; |
846705de | 717 | bm->cur.bit = pfn + 1; |
0d83304c AM |
718 | *bit_nr = pfn; |
719 | *addr = bb->data; | |
a82f7119 | 720 | return 0; |
74dfd666 RW |
721 | } |
722 | ||
723 | static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn) | |
724 | { | |
725 | void *addr; | |
726 | unsigned int bit; | |
a82f7119 | 727 | int error; |
74dfd666 | 728 | |
a82f7119 RW |
729 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
730 | BUG_ON(error); | |
74dfd666 RW |
731 | set_bit(bit, addr); |
732 | } | |
733 | ||
a82f7119 RW |
734 | static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn) |
735 | { | |
736 | void *addr; | |
737 | unsigned int bit; | |
738 | int error; | |
739 | ||
740 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); | |
741 | if (!error) | |
742 | set_bit(bit, addr); | |
743 | return error; | |
744 | } | |
745 | ||
74dfd666 RW |
746 | static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn) |
747 | { | |
748 | void *addr; | |
749 | unsigned int bit; | |
a82f7119 | 750 | int error; |
74dfd666 | 751 | |
a82f7119 RW |
752 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
753 | BUG_ON(error); | |
74dfd666 RW |
754 | clear_bit(bit, addr); |
755 | } | |
756 | ||
757 | static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn) | |
758 | { | |
759 | void *addr; | |
760 | unsigned int bit; | |
a82f7119 | 761 | int error; |
74dfd666 | 762 | |
a82f7119 RW |
763 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
764 | BUG_ON(error); | |
74dfd666 | 765 | return test_bit(bit, addr); |
b788db79 RW |
766 | } |
767 | ||
69643279 RW |
768 | static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn) |
769 | { | |
770 | void *addr; | |
771 | unsigned int bit; | |
772 | ||
773 | return !memory_bm_find_bit(bm, pfn, &addr, &bit); | |
774 | } | |
775 | ||
b788db79 RW |
776 | /** |
777 | * memory_bm_next_pfn - find the pfn that corresponds to the next set bit | |
778 | * in the bitmap @bm. If the pfn cannot be found, BM_END_OF_MAP is | |
779 | * returned. | |
780 | * | |
781 | * It is required to run memory_bm_position_reset() before the first call to | |
782 | * this function. | |
783 | */ | |
784 | ||
785 | static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm) | |
786 | { | |
b788db79 | 787 | struct bm_block *bb; |
b788db79 RW |
788 | int bit; |
789 | ||
846705de | 790 | bb = bm->cur.block; |
b788db79 | 791 | do { |
846705de RW |
792 | bit = bm->cur.bit; |
793 | bit = find_next_bit(bb->data, bm_block_bits(bb), bit); | |
794 | if (bit < bm_block_bits(bb)) | |
795 | goto Return_pfn; | |
796 | ||
797 | bb = list_entry(bb->hook.next, struct bm_block, hook); | |
798 | bm->cur.block = bb; | |
799 | bm->cur.bit = 0; | |
800 | } while (&bb->hook != &bm->blocks); | |
801 | ||
b788db79 RW |
802 | memory_bm_position_reset(bm); |
803 | return BM_END_OF_MAP; | |
804 | ||
59a49335 | 805 | Return_pfn: |
0d83304c AM |
806 | bm->cur.bit = bit + 1; |
807 | return bb->start_pfn + bit; | |
b788db79 RW |
808 | } |
809 | ||
74dfd666 RW |
810 | /** |
811 | * This structure represents a range of page frames the contents of which | |
812 | * should not be saved during the suspend. | |
813 | */ | |
814 | ||
815 | struct nosave_region { | |
816 | struct list_head list; | |
817 | unsigned long start_pfn; | |
818 | unsigned long end_pfn; | |
819 | }; | |
820 | ||
821 | static LIST_HEAD(nosave_regions); | |
822 | ||
823 | /** | |
824 | * register_nosave_region - register a range of page frames the contents | |
825 | * of which should not be saved during the suspend (to be used in the early | |
826 | * initialization code) | |
827 | */ | |
828 | ||
829 | void __init | |
940d67f6 JB |
830 | __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn, |
831 | int use_kmalloc) | |
74dfd666 RW |
832 | { |
833 | struct nosave_region *region; | |
834 | ||
835 | if (start_pfn >= end_pfn) | |
836 | return; | |
837 | ||
838 | if (!list_empty(&nosave_regions)) { | |
839 | /* Try to extend the previous region (they should be sorted) */ | |
840 | region = list_entry(nosave_regions.prev, | |
841 | struct nosave_region, list); | |
842 | if (region->end_pfn == start_pfn) { | |
843 | region->end_pfn = end_pfn; | |
844 | goto Report; | |
845 | } | |
846 | } | |
940d67f6 JB |
847 | if (use_kmalloc) { |
848 | /* during init, this shouldn't fail */ | |
849 | region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL); | |
850 | BUG_ON(!region); | |
851 | } else | |
852 | /* This allocation cannot fail */ | |
c2f69cda | 853 | region = memblock_virt_alloc(sizeof(struct nosave_region), 0); |
74dfd666 RW |
854 | region->start_pfn = start_pfn; |
855 | region->end_pfn = end_pfn; | |
856 | list_add_tail(®ion->list, &nosave_regions); | |
857 | Report: | |
cd38ca85 BH |
858 | printk(KERN_INFO "PM: Registered nosave memory: [mem %#010llx-%#010llx]\n", |
859 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
860 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
74dfd666 RW |
861 | } |
862 | ||
863 | /* | |
864 | * Set bits in this map correspond to the page frames the contents of which | |
865 | * should not be saved during the suspend. | |
866 | */ | |
867 | static struct memory_bitmap *forbidden_pages_map; | |
868 | ||
869 | /* Set bits in this map correspond to free page frames. */ | |
870 | static struct memory_bitmap *free_pages_map; | |
871 | ||
872 | /* | |
873 | * Each page frame allocated for creating the image is marked by setting the | |
874 | * corresponding bits in forbidden_pages_map and free_pages_map simultaneously | |
875 | */ | |
876 | ||
877 | void swsusp_set_page_free(struct page *page) | |
878 | { | |
879 | if (free_pages_map) | |
880 | memory_bm_set_bit(free_pages_map, page_to_pfn(page)); | |
881 | } | |
882 | ||
883 | static int swsusp_page_is_free(struct page *page) | |
884 | { | |
885 | return free_pages_map ? | |
886 | memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0; | |
887 | } | |
888 | ||
889 | void swsusp_unset_page_free(struct page *page) | |
890 | { | |
891 | if (free_pages_map) | |
892 | memory_bm_clear_bit(free_pages_map, page_to_pfn(page)); | |
893 | } | |
894 | ||
895 | static void swsusp_set_page_forbidden(struct page *page) | |
896 | { | |
897 | if (forbidden_pages_map) | |
898 | memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page)); | |
899 | } | |
900 | ||
901 | int swsusp_page_is_forbidden(struct page *page) | |
902 | { | |
903 | return forbidden_pages_map ? | |
904 | memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0; | |
905 | } | |
906 | ||
907 | static void swsusp_unset_page_forbidden(struct page *page) | |
908 | { | |
909 | if (forbidden_pages_map) | |
910 | memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page)); | |
911 | } | |
912 | ||
913 | /** | |
914 | * mark_nosave_pages - set bits corresponding to the page frames the | |
915 | * contents of which should not be saved in a given bitmap. | |
916 | */ | |
917 | ||
918 | static void mark_nosave_pages(struct memory_bitmap *bm) | |
919 | { | |
920 | struct nosave_region *region; | |
921 | ||
922 | if (list_empty(&nosave_regions)) | |
923 | return; | |
924 | ||
925 | list_for_each_entry(region, &nosave_regions, list) { | |
926 | unsigned long pfn; | |
927 | ||
69f1d475 BH |
928 | pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n", |
929 | (unsigned long long) region->start_pfn << PAGE_SHIFT, | |
930 | ((unsigned long long) region->end_pfn << PAGE_SHIFT) | |
931 | - 1); | |
74dfd666 RW |
932 | |
933 | for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++) | |
a82f7119 RW |
934 | if (pfn_valid(pfn)) { |
935 | /* | |
936 | * It is safe to ignore the result of | |
937 | * mem_bm_set_bit_check() here, since we won't | |
938 | * touch the PFNs for which the error is | |
939 | * returned anyway. | |
940 | */ | |
941 | mem_bm_set_bit_check(bm, pfn); | |
942 | } | |
74dfd666 RW |
943 | } |
944 | } | |
945 | ||
946 | /** | |
947 | * create_basic_memory_bitmaps - create bitmaps needed for marking page | |
948 | * frames that should not be saved and free page frames. The pointers | |
949 | * forbidden_pages_map and free_pages_map are only modified if everything | |
950 | * goes well, because we don't want the bits to be used before both bitmaps | |
951 | * are set up. | |
952 | */ | |
953 | ||
954 | int create_basic_memory_bitmaps(void) | |
955 | { | |
956 | struct memory_bitmap *bm1, *bm2; | |
957 | int error = 0; | |
958 | ||
aab17289 RW |
959 | if (forbidden_pages_map && free_pages_map) |
960 | return 0; | |
961 | else | |
962 | BUG_ON(forbidden_pages_map || free_pages_map); | |
74dfd666 | 963 | |
0709db60 | 964 | bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
965 | if (!bm1) |
966 | return -ENOMEM; | |
967 | ||
0709db60 | 968 | error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
969 | if (error) |
970 | goto Free_first_object; | |
971 | ||
0709db60 | 972 | bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
973 | if (!bm2) |
974 | goto Free_first_bitmap; | |
975 | ||
0709db60 | 976 | error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
977 | if (error) |
978 | goto Free_second_object; | |
979 | ||
980 | forbidden_pages_map = bm1; | |
981 | free_pages_map = bm2; | |
982 | mark_nosave_pages(forbidden_pages_map); | |
983 | ||
23976728 | 984 | pr_debug("PM: Basic memory bitmaps created\n"); |
74dfd666 RW |
985 | |
986 | return 0; | |
987 | ||
988 | Free_second_object: | |
989 | kfree(bm2); | |
990 | Free_first_bitmap: | |
991 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
992 | Free_first_object: | |
993 | kfree(bm1); | |
994 | return -ENOMEM; | |
995 | } | |
996 | ||
997 | /** | |
998 | * free_basic_memory_bitmaps - free memory bitmaps allocated by | |
999 | * create_basic_memory_bitmaps(). The auxiliary pointers are necessary | |
1000 | * so that the bitmaps themselves are not referred to while they are being | |
1001 | * freed. | |
1002 | */ | |
1003 | ||
1004 | void free_basic_memory_bitmaps(void) | |
1005 | { | |
1006 | struct memory_bitmap *bm1, *bm2; | |
1007 | ||
6a0c7cd3 RW |
1008 | if (WARN_ON(!(forbidden_pages_map && free_pages_map))) |
1009 | return; | |
74dfd666 RW |
1010 | |
1011 | bm1 = forbidden_pages_map; | |
1012 | bm2 = free_pages_map; | |
1013 | forbidden_pages_map = NULL; | |
1014 | free_pages_map = NULL; | |
1015 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
1016 | kfree(bm1); | |
1017 | memory_bm_free(bm2, PG_UNSAFE_CLEAR); | |
1018 | kfree(bm2); | |
1019 | ||
23976728 | 1020 | pr_debug("PM: Basic memory bitmaps freed\n"); |
74dfd666 RW |
1021 | } |
1022 | ||
b788db79 RW |
1023 | /** |
1024 | * snapshot_additional_pages - estimate the number of additional pages | |
1025 | * be needed for setting up the suspend image data structures for given | |
1026 | * zone (usually the returned value is greater than the exact number) | |
1027 | */ | |
1028 | ||
1029 | unsigned int snapshot_additional_pages(struct zone *zone) | |
1030 | { | |
f469f02d | 1031 | unsigned int rtree, nodes; |
b788db79 RW |
1032 | unsigned int res; |
1033 | ||
1034 | res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK); | |
160cb5a9 NK |
1035 | res += DIV_ROUND_UP(res * sizeof(struct bm_block), |
1036 | LINKED_PAGE_DATA_SIZE); | |
f469f02d JR |
1037 | rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK); |
1038 | rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node), | |
1039 | LINKED_PAGE_DATA_SIZE); | |
1040 | while (nodes > 1) { | |
1041 | nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL); | |
1042 | rtree += nodes; | |
1043 | } | |
1044 | ||
1045 | return 2 * (res + rtree); | |
b788db79 RW |
1046 | } |
1047 | ||
8357376d RW |
1048 | #ifdef CONFIG_HIGHMEM |
1049 | /** | |
1050 | * count_free_highmem_pages - compute the total number of free highmem | |
1051 | * pages, system-wide. | |
1052 | */ | |
1053 | ||
1054 | static unsigned int count_free_highmem_pages(void) | |
1055 | { | |
1056 | struct zone *zone; | |
1057 | unsigned int cnt = 0; | |
1058 | ||
ee99c71c KM |
1059 | for_each_populated_zone(zone) |
1060 | if (is_highmem(zone)) | |
d23ad423 | 1061 | cnt += zone_page_state(zone, NR_FREE_PAGES); |
8357376d RW |
1062 | |
1063 | return cnt; | |
1064 | } | |
1065 | ||
1066 | /** | |
1067 | * saveable_highmem_page - Determine whether a highmem page should be | |
1068 | * included in the suspend image. | |
1069 | * | |
1070 | * We should save the page if it isn't Nosave or NosaveFree, or Reserved, | |
1071 | * and it isn't a part of a free chunk of pages. | |
1072 | */ | |
846705de | 1073 | static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn) |
8357376d RW |
1074 | { |
1075 | struct page *page; | |
1076 | ||
1077 | if (!pfn_valid(pfn)) | |
1078 | return NULL; | |
1079 | ||
1080 | page = pfn_to_page(pfn); | |
846705de RW |
1081 | if (page_zone(page) != zone) |
1082 | return NULL; | |
8357376d RW |
1083 | |
1084 | BUG_ON(!PageHighMem(page)); | |
1085 | ||
7be98234 RW |
1086 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) || |
1087 | PageReserved(page)) | |
8357376d RW |
1088 | return NULL; |
1089 | ||
c6968e73 SG |
1090 | if (page_is_guard(page)) |
1091 | return NULL; | |
1092 | ||
8357376d RW |
1093 | return page; |
1094 | } | |
1095 | ||
1096 | /** | |
1097 | * count_highmem_pages - compute the total number of saveable highmem | |
1098 | * pages. | |
1099 | */ | |
1100 | ||
fe419535 | 1101 | static unsigned int count_highmem_pages(void) |
8357376d RW |
1102 | { |
1103 | struct zone *zone; | |
1104 | unsigned int n = 0; | |
1105 | ||
98e73dc5 | 1106 | for_each_populated_zone(zone) { |
8357376d RW |
1107 | unsigned long pfn, max_zone_pfn; |
1108 | ||
1109 | if (!is_highmem(zone)) | |
1110 | continue; | |
1111 | ||
1112 | mark_free_pages(zone); | |
c33bc315 | 1113 | max_zone_pfn = zone_end_pfn(zone); |
8357376d | 1114 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
846705de | 1115 | if (saveable_highmem_page(zone, pfn)) |
8357376d RW |
1116 | n++; |
1117 | } | |
1118 | return n; | |
1119 | } | |
1120 | #else | |
846705de RW |
1121 | static inline void *saveable_highmem_page(struct zone *z, unsigned long p) |
1122 | { | |
1123 | return NULL; | |
1124 | } | |
8357376d RW |
1125 | #endif /* CONFIG_HIGHMEM */ |
1126 | ||
25761b6e | 1127 | /** |
8a235efa RW |
1128 | * saveable_page - Determine whether a non-highmem page should be included |
1129 | * in the suspend image. | |
25761b6e | 1130 | * |
8357376d RW |
1131 | * We should save the page if it isn't Nosave, and is not in the range |
1132 | * of pages statically defined as 'unsaveable', and it isn't a part of | |
1133 | * a free chunk of pages. | |
25761b6e | 1134 | */ |
846705de | 1135 | static struct page *saveable_page(struct zone *zone, unsigned long pfn) |
25761b6e | 1136 | { |
de491861 | 1137 | struct page *page; |
25761b6e RW |
1138 | |
1139 | if (!pfn_valid(pfn)) | |
ae83c5ee | 1140 | return NULL; |
25761b6e RW |
1141 | |
1142 | page = pfn_to_page(pfn); | |
846705de RW |
1143 | if (page_zone(page) != zone) |
1144 | return NULL; | |
ae83c5ee | 1145 | |
8357376d RW |
1146 | BUG_ON(PageHighMem(page)); |
1147 | ||
7be98234 | 1148 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page)) |
ae83c5ee | 1149 | return NULL; |
8357376d | 1150 | |
8a235efa RW |
1151 | if (PageReserved(page) |
1152 | && (!kernel_page_present(page) || pfn_is_nosave(pfn))) | |
ae83c5ee | 1153 | return NULL; |
25761b6e | 1154 | |
c6968e73 SG |
1155 | if (page_is_guard(page)) |
1156 | return NULL; | |
1157 | ||
ae83c5ee | 1158 | return page; |
25761b6e RW |
1159 | } |
1160 | ||
8357376d RW |
1161 | /** |
1162 | * count_data_pages - compute the total number of saveable non-highmem | |
1163 | * pages. | |
1164 | */ | |
1165 | ||
fe419535 | 1166 | static unsigned int count_data_pages(void) |
25761b6e RW |
1167 | { |
1168 | struct zone *zone; | |
ae83c5ee | 1169 | unsigned long pfn, max_zone_pfn; |
dc19d507 | 1170 | unsigned int n = 0; |
25761b6e | 1171 | |
98e73dc5 | 1172 | for_each_populated_zone(zone) { |
25761b6e RW |
1173 | if (is_highmem(zone)) |
1174 | continue; | |
8357376d | 1175 | |
25761b6e | 1176 | mark_free_pages(zone); |
c33bc315 | 1177 | max_zone_pfn = zone_end_pfn(zone); |
ae83c5ee | 1178 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
846705de | 1179 | if (saveable_page(zone, pfn)) |
8357376d | 1180 | n++; |
25761b6e | 1181 | } |
a0f49651 | 1182 | return n; |
25761b6e RW |
1183 | } |
1184 | ||
8357376d RW |
1185 | /* This is needed, because copy_page and memcpy are not usable for copying |
1186 | * task structs. | |
1187 | */ | |
1188 | static inline void do_copy_page(long *dst, long *src) | |
f623f0db RW |
1189 | { |
1190 | int n; | |
1191 | ||
f623f0db RW |
1192 | for (n = PAGE_SIZE / sizeof(long); n; n--) |
1193 | *dst++ = *src++; | |
1194 | } | |
1195 | ||
8a235efa RW |
1196 | |
1197 | /** | |
1198 | * safe_copy_page - check if the page we are going to copy is marked as | |
1199 | * present in the kernel page tables (this always is the case if | |
1200 | * CONFIG_DEBUG_PAGEALLOC is not set and in that case | |
1201 | * kernel_page_present() always returns 'true'). | |
1202 | */ | |
1203 | static void safe_copy_page(void *dst, struct page *s_page) | |
1204 | { | |
1205 | if (kernel_page_present(s_page)) { | |
1206 | do_copy_page(dst, page_address(s_page)); | |
1207 | } else { | |
1208 | kernel_map_pages(s_page, 1, 1); | |
1209 | do_copy_page(dst, page_address(s_page)); | |
1210 | kernel_map_pages(s_page, 1, 0); | |
1211 | } | |
1212 | } | |
1213 | ||
1214 | ||
8357376d RW |
1215 | #ifdef CONFIG_HIGHMEM |
1216 | static inline struct page * | |
1217 | page_is_saveable(struct zone *zone, unsigned long pfn) | |
1218 | { | |
1219 | return is_highmem(zone) ? | |
846705de | 1220 | saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn); |
8357376d RW |
1221 | } |
1222 | ||
8a235efa | 1223 | static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
8357376d RW |
1224 | { |
1225 | struct page *s_page, *d_page; | |
1226 | void *src, *dst; | |
1227 | ||
1228 | s_page = pfn_to_page(src_pfn); | |
1229 | d_page = pfn_to_page(dst_pfn); | |
1230 | if (PageHighMem(s_page)) { | |
0de9a1e2 CW |
1231 | src = kmap_atomic(s_page); |
1232 | dst = kmap_atomic(d_page); | |
8357376d | 1233 | do_copy_page(dst, src); |
0de9a1e2 CW |
1234 | kunmap_atomic(dst); |
1235 | kunmap_atomic(src); | |
8357376d | 1236 | } else { |
8357376d RW |
1237 | if (PageHighMem(d_page)) { |
1238 | /* Page pointed to by src may contain some kernel | |
1239 | * data modified by kmap_atomic() | |
1240 | */ | |
8a235efa | 1241 | safe_copy_page(buffer, s_page); |
0de9a1e2 | 1242 | dst = kmap_atomic(d_page); |
3ecb01df | 1243 | copy_page(dst, buffer); |
0de9a1e2 | 1244 | kunmap_atomic(dst); |
8357376d | 1245 | } else { |
8a235efa | 1246 | safe_copy_page(page_address(d_page), s_page); |
8357376d RW |
1247 | } |
1248 | } | |
1249 | } | |
1250 | #else | |
846705de | 1251 | #define page_is_saveable(zone, pfn) saveable_page(zone, pfn) |
8357376d | 1252 | |
8a235efa | 1253 | static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
8357376d | 1254 | { |
8a235efa RW |
1255 | safe_copy_page(page_address(pfn_to_page(dst_pfn)), |
1256 | pfn_to_page(src_pfn)); | |
8357376d RW |
1257 | } |
1258 | #endif /* CONFIG_HIGHMEM */ | |
1259 | ||
b788db79 RW |
1260 | static void |
1261 | copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm) | |
25761b6e RW |
1262 | { |
1263 | struct zone *zone; | |
b788db79 | 1264 | unsigned long pfn; |
25761b6e | 1265 | |
98e73dc5 | 1266 | for_each_populated_zone(zone) { |
b788db79 RW |
1267 | unsigned long max_zone_pfn; |
1268 | ||
25761b6e | 1269 | mark_free_pages(zone); |
c33bc315 | 1270 | max_zone_pfn = zone_end_pfn(zone); |
b788db79 | 1271 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
8357376d | 1272 | if (page_is_saveable(zone, pfn)) |
b788db79 | 1273 | memory_bm_set_bit(orig_bm, pfn); |
25761b6e | 1274 | } |
b788db79 RW |
1275 | memory_bm_position_reset(orig_bm); |
1276 | memory_bm_position_reset(copy_bm); | |
df7c4872 | 1277 | for(;;) { |
b788db79 | 1278 | pfn = memory_bm_next_pfn(orig_bm); |
df7c4872 FW |
1279 | if (unlikely(pfn == BM_END_OF_MAP)) |
1280 | break; | |
1281 | copy_data_page(memory_bm_next_pfn(copy_bm), pfn); | |
1282 | } | |
25761b6e RW |
1283 | } |
1284 | ||
8357376d RW |
1285 | /* Total number of image pages */ |
1286 | static unsigned int nr_copy_pages; | |
1287 | /* Number of pages needed for saving the original pfns of the image pages */ | |
1288 | static unsigned int nr_meta_pages; | |
64a473cb RW |
1289 | /* |
1290 | * Numbers of normal and highmem page frames allocated for hibernation image | |
1291 | * before suspending devices. | |
1292 | */ | |
1293 | unsigned int alloc_normal, alloc_highmem; | |
1294 | /* | |
1295 | * Memory bitmap used for marking saveable pages (during hibernation) or | |
1296 | * hibernation image pages (during restore) | |
1297 | */ | |
1298 | static struct memory_bitmap orig_bm; | |
1299 | /* | |
1300 | * Memory bitmap used during hibernation for marking allocated page frames that | |
1301 | * will contain copies of saveable pages. During restore it is initially used | |
1302 | * for marking hibernation image pages, but then the set bits from it are | |
1303 | * duplicated in @orig_bm and it is released. On highmem systems it is next | |
1304 | * used for marking "safe" highmem pages, but it has to be reinitialized for | |
1305 | * this purpose. | |
1306 | */ | |
1307 | static struct memory_bitmap copy_bm; | |
8357376d | 1308 | |
25761b6e | 1309 | /** |
940864dd | 1310 | * swsusp_free - free pages allocated for the suspend. |
cd560bb2 | 1311 | * |
940864dd RW |
1312 | * Suspend pages are alocated before the atomic copy is made, so we |
1313 | * need to release them after the resume. | |
25761b6e RW |
1314 | */ |
1315 | ||
1316 | void swsusp_free(void) | |
1317 | { | |
1318 | struct zone *zone; | |
ae83c5ee | 1319 | unsigned long pfn, max_zone_pfn; |
25761b6e | 1320 | |
98e73dc5 | 1321 | for_each_populated_zone(zone) { |
c33bc315 | 1322 | max_zone_pfn = zone_end_pfn(zone); |
ae83c5ee RW |
1323 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
1324 | if (pfn_valid(pfn)) { | |
1325 | struct page *page = pfn_to_page(pfn); | |
1326 | ||
7be98234 RW |
1327 | if (swsusp_page_is_forbidden(page) && |
1328 | swsusp_page_is_free(page)) { | |
1329 | swsusp_unset_page_forbidden(page); | |
1330 | swsusp_unset_page_free(page); | |
8357376d | 1331 | __free_page(page); |
25761b6e RW |
1332 | } |
1333 | } | |
1334 | } | |
f577eb30 RW |
1335 | nr_copy_pages = 0; |
1336 | nr_meta_pages = 0; | |
75534b50 | 1337 | restore_pblist = NULL; |
6e1819d6 | 1338 | buffer = NULL; |
64a473cb RW |
1339 | alloc_normal = 0; |
1340 | alloc_highmem = 0; | |
25761b6e RW |
1341 | } |
1342 | ||
4bb33435 RW |
1343 | /* Helper functions used for the shrinking of memory. */ |
1344 | ||
1345 | #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN) | |
1346 | ||
fe419535 | 1347 | /** |
4bb33435 RW |
1348 | * preallocate_image_pages - Allocate a number of pages for hibernation image |
1349 | * @nr_pages: Number of page frames to allocate. | |
1350 | * @mask: GFP flags to use for the allocation. | |
fe419535 | 1351 | * |
4bb33435 RW |
1352 | * Return value: Number of page frames actually allocated |
1353 | */ | |
1354 | static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask) | |
1355 | { | |
1356 | unsigned long nr_alloc = 0; | |
1357 | ||
1358 | while (nr_pages > 0) { | |
64a473cb RW |
1359 | struct page *page; |
1360 | ||
1361 | page = alloc_image_page(mask); | |
1362 | if (!page) | |
4bb33435 | 1363 | break; |
64a473cb RW |
1364 | memory_bm_set_bit(©_bm, page_to_pfn(page)); |
1365 | if (PageHighMem(page)) | |
1366 | alloc_highmem++; | |
1367 | else | |
1368 | alloc_normal++; | |
4bb33435 RW |
1369 | nr_pages--; |
1370 | nr_alloc++; | |
1371 | } | |
1372 | ||
1373 | return nr_alloc; | |
1374 | } | |
1375 | ||
6715045d RW |
1376 | static unsigned long preallocate_image_memory(unsigned long nr_pages, |
1377 | unsigned long avail_normal) | |
4bb33435 | 1378 | { |
6715045d RW |
1379 | unsigned long alloc; |
1380 | ||
1381 | if (avail_normal <= alloc_normal) | |
1382 | return 0; | |
1383 | ||
1384 | alloc = avail_normal - alloc_normal; | |
1385 | if (nr_pages < alloc) | |
1386 | alloc = nr_pages; | |
1387 | ||
1388 | return preallocate_image_pages(alloc, GFP_IMAGE); | |
4bb33435 RW |
1389 | } |
1390 | ||
1391 | #ifdef CONFIG_HIGHMEM | |
1392 | static unsigned long preallocate_image_highmem(unsigned long nr_pages) | |
1393 | { | |
1394 | return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM); | |
1395 | } | |
1396 | ||
1397 | /** | |
1398 | * __fraction - Compute (an approximation of) x * (multiplier / base) | |
fe419535 | 1399 | */ |
4bb33435 RW |
1400 | static unsigned long __fraction(u64 x, u64 multiplier, u64 base) |
1401 | { | |
1402 | x *= multiplier; | |
1403 | do_div(x, base); | |
1404 | return (unsigned long)x; | |
1405 | } | |
fe419535 | 1406 | |
4bb33435 RW |
1407 | static unsigned long preallocate_highmem_fraction(unsigned long nr_pages, |
1408 | unsigned long highmem, | |
1409 | unsigned long total) | |
fe419535 | 1410 | { |
4bb33435 RW |
1411 | unsigned long alloc = __fraction(nr_pages, highmem, total); |
1412 | ||
1413 | return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM); | |
fe419535 | 1414 | } |
4bb33435 RW |
1415 | #else /* CONFIG_HIGHMEM */ |
1416 | static inline unsigned long preallocate_image_highmem(unsigned long nr_pages) | |
1417 | { | |
1418 | return 0; | |
1419 | } | |
1420 | ||
1421 | static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages, | |
1422 | unsigned long highmem, | |
1423 | unsigned long total) | |
1424 | { | |
1425 | return 0; | |
1426 | } | |
1427 | #endif /* CONFIG_HIGHMEM */ | |
fe419535 | 1428 | |
4bb33435 | 1429 | /** |
64a473cb RW |
1430 | * free_unnecessary_pages - Release preallocated pages not needed for the image |
1431 | */ | |
1432 | static void free_unnecessary_pages(void) | |
1433 | { | |
6715045d | 1434 | unsigned long save, to_free_normal, to_free_highmem; |
64a473cb | 1435 | |
6715045d RW |
1436 | save = count_data_pages(); |
1437 | if (alloc_normal >= save) { | |
1438 | to_free_normal = alloc_normal - save; | |
1439 | save = 0; | |
1440 | } else { | |
1441 | to_free_normal = 0; | |
1442 | save -= alloc_normal; | |
1443 | } | |
1444 | save += count_highmem_pages(); | |
1445 | if (alloc_highmem >= save) { | |
1446 | to_free_highmem = alloc_highmem - save; | |
64a473cb RW |
1447 | } else { |
1448 | to_free_highmem = 0; | |
4d4cf23c RW |
1449 | save -= alloc_highmem; |
1450 | if (to_free_normal > save) | |
1451 | to_free_normal -= save; | |
1452 | else | |
1453 | to_free_normal = 0; | |
64a473cb RW |
1454 | } |
1455 | ||
1456 | memory_bm_position_reset(©_bm); | |
1457 | ||
a9c9b442 | 1458 | while (to_free_normal > 0 || to_free_highmem > 0) { |
64a473cb RW |
1459 | unsigned long pfn = memory_bm_next_pfn(©_bm); |
1460 | struct page *page = pfn_to_page(pfn); | |
1461 | ||
1462 | if (PageHighMem(page)) { | |
1463 | if (!to_free_highmem) | |
1464 | continue; | |
1465 | to_free_highmem--; | |
1466 | alloc_highmem--; | |
1467 | } else { | |
1468 | if (!to_free_normal) | |
1469 | continue; | |
1470 | to_free_normal--; | |
1471 | alloc_normal--; | |
1472 | } | |
1473 | memory_bm_clear_bit(©_bm, pfn); | |
1474 | swsusp_unset_page_forbidden(page); | |
1475 | swsusp_unset_page_free(page); | |
1476 | __free_page(page); | |
1477 | } | |
1478 | } | |
1479 | ||
ef4aede3 RW |
1480 | /** |
1481 | * minimum_image_size - Estimate the minimum acceptable size of an image | |
1482 | * @saveable: Number of saveable pages in the system. | |
1483 | * | |
1484 | * We want to avoid attempting to free too much memory too hard, so estimate the | |
1485 | * minimum acceptable size of a hibernation image to use as the lower limit for | |
1486 | * preallocating memory. | |
1487 | * | |
1488 | * We assume that the minimum image size should be proportional to | |
1489 | * | |
1490 | * [number of saveable pages] - [number of pages that can be freed in theory] | |
1491 | * | |
1492 | * where the second term is the sum of (1) reclaimable slab pages, (2) active | |
4d434820 | 1493 | * and (3) inactive anonymous pages, (4) active and (5) inactive file pages, |
ef4aede3 RW |
1494 | * minus mapped file pages. |
1495 | */ | |
1496 | static unsigned long minimum_image_size(unsigned long saveable) | |
1497 | { | |
1498 | unsigned long size; | |
1499 | ||
1500 | size = global_page_state(NR_SLAB_RECLAIMABLE) | |
1501 | + global_page_state(NR_ACTIVE_ANON) | |
1502 | + global_page_state(NR_INACTIVE_ANON) | |
1503 | + global_page_state(NR_ACTIVE_FILE) | |
1504 | + global_page_state(NR_INACTIVE_FILE) | |
1505 | - global_page_state(NR_FILE_MAPPED); | |
1506 | ||
1507 | return saveable <= size ? 0 : saveable - size; | |
1508 | } | |
1509 | ||
64a473cb RW |
1510 | /** |
1511 | * hibernate_preallocate_memory - Preallocate memory for hibernation image | |
4bb33435 RW |
1512 | * |
1513 | * To create a hibernation image it is necessary to make a copy of every page | |
1514 | * frame in use. We also need a number of page frames to be free during | |
1515 | * hibernation for allocations made while saving the image and for device | |
1516 | * drivers, in case they need to allocate memory from their hibernation | |
ddeb6487 RW |
1517 | * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough |
1518 | * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through | |
1519 | * /sys/power/reserved_size, respectively). To make this happen, we compute the | |
1520 | * total number of available page frames and allocate at least | |
4bb33435 | 1521 | * |
ddeb6487 RW |
1522 | * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2 |
1523 | * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE) | |
4bb33435 RW |
1524 | * |
1525 | * of them, which corresponds to the maximum size of a hibernation image. | |
1526 | * | |
1527 | * If image_size is set below the number following from the above formula, | |
1528 | * the preallocation of memory is continued until the total number of saveable | |
ef4aede3 RW |
1529 | * pages in the system is below the requested image size or the minimum |
1530 | * acceptable image size returned by minimum_image_size(), whichever is greater. | |
4bb33435 | 1531 | */ |
64a473cb | 1532 | int hibernate_preallocate_memory(void) |
fe419535 | 1533 | { |
fe419535 | 1534 | struct zone *zone; |
4bb33435 | 1535 | unsigned long saveable, size, max_size, count, highmem, pages = 0; |
6715045d | 1536 | unsigned long alloc, save_highmem, pages_highmem, avail_normal; |
fe419535 | 1537 | struct timeval start, stop; |
64a473cb | 1538 | int error; |
fe419535 | 1539 | |
64a473cb | 1540 | printk(KERN_INFO "PM: Preallocating image memory... "); |
fe419535 | 1541 | do_gettimeofday(&start); |
fe419535 | 1542 | |
64a473cb RW |
1543 | error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY); |
1544 | if (error) | |
1545 | goto err_out; | |
1546 | ||
1547 | error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY); | |
1548 | if (error) | |
1549 | goto err_out; | |
1550 | ||
1551 | alloc_normal = 0; | |
1552 | alloc_highmem = 0; | |
1553 | ||
4bb33435 | 1554 | /* Count the number of saveable data pages. */ |
64a473cb | 1555 | save_highmem = count_highmem_pages(); |
4bb33435 | 1556 | saveable = count_data_pages(); |
fe419535 | 1557 | |
4bb33435 RW |
1558 | /* |
1559 | * Compute the total number of page frames we can use (count) and the | |
1560 | * number of pages needed for image metadata (size). | |
1561 | */ | |
1562 | count = saveable; | |
64a473cb RW |
1563 | saveable += save_highmem; |
1564 | highmem = save_highmem; | |
4bb33435 RW |
1565 | size = 0; |
1566 | for_each_populated_zone(zone) { | |
1567 | size += snapshot_additional_pages(zone); | |
1568 | if (is_highmem(zone)) | |
1569 | highmem += zone_page_state(zone, NR_FREE_PAGES); | |
1570 | else | |
1571 | count += zone_page_state(zone, NR_FREE_PAGES); | |
1572 | } | |
6715045d | 1573 | avail_normal = count; |
4bb33435 RW |
1574 | count += highmem; |
1575 | count -= totalreserve_pages; | |
1576 | ||
85055dd8 MS |
1577 | /* Add number of pages required for page keys (s390 only). */ |
1578 | size += page_key_additional_pages(saveable); | |
1579 | ||
4bb33435 | 1580 | /* Compute the maximum number of saveable pages to leave in memory. */ |
ddeb6487 RW |
1581 | max_size = (count - (size + PAGES_FOR_IO)) / 2 |
1582 | - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE); | |
266f1a25 | 1583 | /* Compute the desired number of image pages specified by image_size. */ |
4bb33435 RW |
1584 | size = DIV_ROUND_UP(image_size, PAGE_SIZE); |
1585 | if (size > max_size) | |
1586 | size = max_size; | |
1587 | /* | |
266f1a25 RW |
1588 | * If the desired number of image pages is at least as large as the |
1589 | * current number of saveable pages in memory, allocate page frames for | |
1590 | * the image and we're done. | |
4bb33435 | 1591 | */ |
64a473cb RW |
1592 | if (size >= saveable) { |
1593 | pages = preallocate_image_highmem(save_highmem); | |
6715045d | 1594 | pages += preallocate_image_memory(saveable - pages, avail_normal); |
4bb33435 | 1595 | goto out; |
64a473cb | 1596 | } |
4bb33435 | 1597 | |
ef4aede3 RW |
1598 | /* Estimate the minimum size of the image. */ |
1599 | pages = minimum_image_size(saveable); | |
6715045d RW |
1600 | /* |
1601 | * To avoid excessive pressure on the normal zone, leave room in it to | |
1602 | * accommodate an image of the minimum size (unless it's already too | |
1603 | * small, in which case don't preallocate pages from it at all). | |
1604 | */ | |
1605 | if (avail_normal > pages) | |
1606 | avail_normal -= pages; | |
1607 | else | |
1608 | avail_normal = 0; | |
ef4aede3 RW |
1609 | if (size < pages) |
1610 | size = min_t(unsigned long, pages, max_size); | |
1611 | ||
4bb33435 RW |
1612 | /* |
1613 | * Let the memory management subsystem know that we're going to need a | |
1614 | * large number of page frames to allocate and make it free some memory. | |
1615 | * NOTE: If this is not done, performance will be hurt badly in some | |
1616 | * test cases. | |
1617 | */ | |
1618 | shrink_all_memory(saveable - size); | |
1619 | ||
1620 | /* | |
1621 | * The number of saveable pages in memory was too high, so apply some | |
1622 | * pressure to decrease it. First, make room for the largest possible | |
1623 | * image and fail if that doesn't work. Next, try to decrease the size | |
ef4aede3 RW |
1624 | * of the image as much as indicated by 'size' using allocations from |
1625 | * highmem and non-highmem zones separately. | |
4bb33435 RW |
1626 | */ |
1627 | pages_highmem = preallocate_image_highmem(highmem / 2); | |
fd432b9f AL |
1628 | alloc = count - max_size; |
1629 | if (alloc > pages_highmem) | |
1630 | alloc -= pages_highmem; | |
1631 | else | |
1632 | alloc = 0; | |
6715045d RW |
1633 | pages = preallocate_image_memory(alloc, avail_normal); |
1634 | if (pages < alloc) { | |
1635 | /* We have exhausted non-highmem pages, try highmem. */ | |
1636 | alloc -= pages; | |
1637 | pages += pages_highmem; | |
1638 | pages_highmem = preallocate_image_highmem(alloc); | |
1639 | if (pages_highmem < alloc) | |
1640 | goto err_out; | |
1641 | pages += pages_highmem; | |
1642 | /* | |
1643 | * size is the desired number of saveable pages to leave in | |
1644 | * memory, so try to preallocate (all memory - size) pages. | |
1645 | */ | |
1646 | alloc = (count - pages) - size; | |
1647 | pages += preallocate_image_highmem(alloc); | |
1648 | } else { | |
1649 | /* | |
1650 | * There are approximately max_size saveable pages at this point | |
1651 | * and we want to reduce this number down to size. | |
1652 | */ | |
1653 | alloc = max_size - size; | |
1654 | size = preallocate_highmem_fraction(alloc, highmem, count); | |
1655 | pages_highmem += size; | |
1656 | alloc -= size; | |
1657 | size = preallocate_image_memory(alloc, avail_normal); | |
1658 | pages_highmem += preallocate_image_highmem(alloc - size); | |
1659 | pages += pages_highmem + size; | |
1660 | } | |
4bb33435 | 1661 | |
64a473cb RW |
1662 | /* |
1663 | * We only need as many page frames for the image as there are saveable | |
1664 | * pages in memory, but we have allocated more. Release the excessive | |
1665 | * ones now. | |
1666 | */ | |
1667 | free_unnecessary_pages(); | |
4bb33435 RW |
1668 | |
1669 | out: | |
fe419535 | 1670 | do_gettimeofday(&stop); |
64a473cb RW |
1671 | printk(KERN_CONT "done (allocated %lu pages)\n", pages); |
1672 | swsusp_show_speed(&start, &stop, pages, "Allocated"); | |
fe419535 RW |
1673 | |
1674 | return 0; | |
64a473cb RW |
1675 | |
1676 | err_out: | |
1677 | printk(KERN_CONT "\n"); | |
1678 | swsusp_free(); | |
1679 | return -ENOMEM; | |
fe419535 RW |
1680 | } |
1681 | ||
8357376d RW |
1682 | #ifdef CONFIG_HIGHMEM |
1683 | /** | |
1684 | * count_pages_for_highmem - compute the number of non-highmem pages | |
1685 | * that will be necessary for creating copies of highmem pages. | |
1686 | */ | |
1687 | ||
1688 | static unsigned int count_pages_for_highmem(unsigned int nr_highmem) | |
1689 | { | |
64a473cb | 1690 | unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem; |
8357376d RW |
1691 | |
1692 | if (free_highmem >= nr_highmem) | |
1693 | nr_highmem = 0; | |
1694 | else | |
1695 | nr_highmem -= free_highmem; | |
1696 | ||
1697 | return nr_highmem; | |
1698 | } | |
1699 | #else | |
1700 | static unsigned int | |
1701 | count_pages_for_highmem(unsigned int nr_highmem) { return 0; } | |
1702 | #endif /* CONFIG_HIGHMEM */ | |
25761b6e RW |
1703 | |
1704 | /** | |
8357376d RW |
1705 | * enough_free_mem - Make sure we have enough free memory for the |
1706 | * snapshot image. | |
25761b6e RW |
1707 | */ |
1708 | ||
8357376d | 1709 | static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem) |
25761b6e | 1710 | { |
e5e2fa78 | 1711 | struct zone *zone; |
64a473cb | 1712 | unsigned int free = alloc_normal; |
e5e2fa78 | 1713 | |
98e73dc5 | 1714 | for_each_populated_zone(zone) |
8357376d | 1715 | if (!is_highmem(zone)) |
d23ad423 | 1716 | free += zone_page_state(zone, NR_FREE_PAGES); |
940864dd | 1717 | |
8357376d | 1718 | nr_pages += count_pages_for_highmem(nr_highmem); |
64a473cb RW |
1719 | pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n", |
1720 | nr_pages, PAGES_FOR_IO, free); | |
940864dd | 1721 | |
64a473cb | 1722 | return free > nr_pages + PAGES_FOR_IO; |
25761b6e RW |
1723 | } |
1724 | ||
8357376d RW |
1725 | #ifdef CONFIG_HIGHMEM |
1726 | /** | |
1727 | * get_highmem_buffer - if there are some highmem pages in the suspend | |
1728 | * image, we may need the buffer to copy them and/or load their data. | |
1729 | */ | |
1730 | ||
1731 | static inline int get_highmem_buffer(int safe_needed) | |
1732 | { | |
1733 | buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed); | |
1734 | return buffer ? 0 : -ENOMEM; | |
1735 | } | |
1736 | ||
1737 | /** | |
1738 | * alloc_highmem_image_pages - allocate some highmem pages for the image. | |
1739 | * Try to allocate as many pages as needed, but if the number of free | |
1740 | * highmem pages is lesser than that, allocate them all. | |
1741 | */ | |
1742 | ||
1743 | static inline unsigned int | |
64a473cb | 1744 | alloc_highmem_pages(struct memory_bitmap *bm, unsigned int nr_highmem) |
8357376d RW |
1745 | { |
1746 | unsigned int to_alloc = count_free_highmem_pages(); | |
1747 | ||
1748 | if (to_alloc > nr_highmem) | |
1749 | to_alloc = nr_highmem; | |
1750 | ||
1751 | nr_highmem -= to_alloc; | |
1752 | while (to_alloc-- > 0) { | |
1753 | struct page *page; | |
1754 | ||
1755 | page = alloc_image_page(__GFP_HIGHMEM); | |
1756 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1757 | } | |
1758 | return nr_highmem; | |
1759 | } | |
1760 | #else | |
1761 | static inline int get_highmem_buffer(int safe_needed) { return 0; } | |
1762 | ||
1763 | static inline unsigned int | |
64a473cb | 1764 | alloc_highmem_pages(struct memory_bitmap *bm, unsigned int n) { return 0; } |
8357376d RW |
1765 | #endif /* CONFIG_HIGHMEM */ |
1766 | ||
1767 | /** | |
1768 | * swsusp_alloc - allocate memory for the suspend image | |
1769 | * | |
1770 | * We first try to allocate as many highmem pages as there are | |
1771 | * saveable highmem pages in the system. If that fails, we allocate | |
1772 | * non-highmem pages for the copies of the remaining highmem ones. | |
1773 | * | |
1774 | * In this approach it is likely that the copies of highmem pages will | |
1775 | * also be located in the high memory, because of the way in which | |
1776 | * copy_data_pages() works. | |
1777 | */ | |
1778 | ||
b788db79 RW |
1779 | static int |
1780 | swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm, | |
8357376d | 1781 | unsigned int nr_pages, unsigned int nr_highmem) |
054bd4c1 | 1782 | { |
8357376d | 1783 | if (nr_highmem > 0) { |
2e725a06 | 1784 | if (get_highmem_buffer(PG_ANY)) |
64a473cb RW |
1785 | goto err_out; |
1786 | if (nr_highmem > alloc_highmem) { | |
1787 | nr_highmem -= alloc_highmem; | |
1788 | nr_pages += alloc_highmem_pages(copy_bm, nr_highmem); | |
1789 | } | |
8357376d | 1790 | } |
64a473cb RW |
1791 | if (nr_pages > alloc_normal) { |
1792 | nr_pages -= alloc_normal; | |
1793 | while (nr_pages-- > 0) { | |
1794 | struct page *page; | |
1795 | ||
1796 | page = alloc_image_page(GFP_ATOMIC | __GFP_COLD); | |
1797 | if (!page) | |
1798 | goto err_out; | |
1799 | memory_bm_set_bit(copy_bm, page_to_pfn(page)); | |
1800 | } | |
25761b6e | 1801 | } |
64a473cb | 1802 | |
b788db79 | 1803 | return 0; |
25761b6e | 1804 | |
64a473cb | 1805 | err_out: |
b788db79 | 1806 | swsusp_free(); |
2e725a06 | 1807 | return -ENOMEM; |
25761b6e RW |
1808 | } |
1809 | ||
722a9f92 | 1810 | asmlinkage __visible int swsusp_save(void) |
25761b6e | 1811 | { |
8357376d | 1812 | unsigned int nr_pages, nr_highmem; |
25761b6e | 1813 | |
07c3bb57 | 1814 | printk(KERN_INFO "PM: Creating hibernation image:\n"); |
25761b6e | 1815 | |
9f8f2172 | 1816 | drain_local_pages(NULL); |
a0f49651 | 1817 | nr_pages = count_data_pages(); |
8357376d | 1818 | nr_highmem = count_highmem_pages(); |
23976728 | 1819 | printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem); |
25761b6e | 1820 | |
8357376d | 1821 | if (!enough_free_mem(nr_pages, nr_highmem)) { |
23976728 | 1822 | printk(KERN_ERR "PM: Not enough free memory\n"); |
25761b6e RW |
1823 | return -ENOMEM; |
1824 | } | |
1825 | ||
8357376d | 1826 | if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) { |
23976728 | 1827 | printk(KERN_ERR "PM: Memory allocation failed\n"); |
a0f49651 | 1828 | return -ENOMEM; |
8357376d | 1829 | } |
25761b6e RW |
1830 | |
1831 | /* During allocating of suspend pagedir, new cold pages may appear. | |
1832 | * Kill them. | |
1833 | */ | |
9f8f2172 | 1834 | drain_local_pages(NULL); |
b788db79 | 1835 | copy_data_pages(©_bm, &orig_bm); |
25761b6e RW |
1836 | |
1837 | /* | |
1838 | * End of critical section. From now on, we can write to memory, | |
1839 | * but we should not touch disk. This specially means we must _not_ | |
1840 | * touch swap space! Except we must write out our image of course. | |
1841 | */ | |
1842 | ||
8357376d | 1843 | nr_pages += nr_highmem; |
a0f49651 | 1844 | nr_copy_pages = nr_pages; |
8357376d | 1845 | nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE); |
a0f49651 | 1846 | |
23976728 RW |
1847 | printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n", |
1848 | nr_pages); | |
8357376d | 1849 | |
25761b6e RW |
1850 | return 0; |
1851 | } | |
f577eb30 | 1852 | |
d307c4a8 RW |
1853 | #ifndef CONFIG_ARCH_HIBERNATION_HEADER |
1854 | static int init_header_complete(struct swsusp_info *info) | |
f577eb30 | 1855 | { |
d307c4a8 | 1856 | memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname)); |
f577eb30 | 1857 | info->version_code = LINUX_VERSION_CODE; |
d307c4a8 RW |
1858 | return 0; |
1859 | } | |
1860 | ||
1861 | static char *check_image_kernel(struct swsusp_info *info) | |
1862 | { | |
1863 | if (info->version_code != LINUX_VERSION_CODE) | |
1864 | return "kernel version"; | |
1865 | if (strcmp(info->uts.sysname,init_utsname()->sysname)) | |
1866 | return "system type"; | |
1867 | if (strcmp(info->uts.release,init_utsname()->release)) | |
1868 | return "kernel release"; | |
1869 | if (strcmp(info->uts.version,init_utsname()->version)) | |
1870 | return "version"; | |
1871 | if (strcmp(info->uts.machine,init_utsname()->machine)) | |
1872 | return "machine"; | |
1873 | return NULL; | |
1874 | } | |
1875 | #endif /* CONFIG_ARCH_HIBERNATION_HEADER */ | |
1876 | ||
af508b34 RW |
1877 | unsigned long snapshot_get_image_size(void) |
1878 | { | |
1879 | return nr_copy_pages + nr_meta_pages + 1; | |
1880 | } | |
1881 | ||
d307c4a8 RW |
1882 | static int init_header(struct swsusp_info *info) |
1883 | { | |
1884 | memset(info, 0, sizeof(struct swsusp_info)); | |
0ed5fd13 | 1885 | info->num_physpages = get_num_physpages(); |
f577eb30 | 1886 | info->image_pages = nr_copy_pages; |
af508b34 | 1887 | info->pages = snapshot_get_image_size(); |
6e1819d6 RW |
1888 | info->size = info->pages; |
1889 | info->size <<= PAGE_SHIFT; | |
d307c4a8 | 1890 | return init_header_complete(info); |
f577eb30 RW |
1891 | } |
1892 | ||
1893 | /** | |
940864dd RW |
1894 | * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm |
1895 | * are stored in the array @buf[] (1 page at a time) | |
f577eb30 RW |
1896 | */ |
1897 | ||
b788db79 | 1898 | static inline void |
940864dd | 1899 | pack_pfns(unsigned long *buf, struct memory_bitmap *bm) |
f577eb30 RW |
1900 | { |
1901 | int j; | |
1902 | ||
b788db79 | 1903 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
940864dd RW |
1904 | buf[j] = memory_bm_next_pfn(bm); |
1905 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
b788db79 | 1906 | break; |
85055dd8 MS |
1907 | /* Save page key for data page (s390 only). */ |
1908 | page_key_read(buf + j); | |
f577eb30 | 1909 | } |
f577eb30 RW |
1910 | } |
1911 | ||
1912 | /** | |
1913 | * snapshot_read_next - used for reading the system memory snapshot. | |
1914 | * | |
1915 | * On the first call to it @handle should point to a zeroed | |
1916 | * snapshot_handle structure. The structure gets updated and a pointer | |
1917 | * to it should be passed to this function every next time. | |
1918 | * | |
f577eb30 RW |
1919 | * On success the function returns a positive number. Then, the caller |
1920 | * is allowed to read up to the returned number of bytes from the memory | |
d3c1b24c | 1921 | * location computed by the data_of() macro. |
f577eb30 RW |
1922 | * |
1923 | * The function returns 0 to indicate the end of data stream condition, | |
1924 | * and a negative number is returned on error. In such cases the | |
1925 | * structure pointed to by @handle is not updated and should not be used | |
1926 | * any more. | |
1927 | */ | |
1928 | ||
d3c1b24c | 1929 | int snapshot_read_next(struct snapshot_handle *handle) |
f577eb30 | 1930 | { |
fb13a28b | 1931 | if (handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 1932 | return 0; |
b788db79 | 1933 | |
f577eb30 RW |
1934 | if (!buffer) { |
1935 | /* This makes the buffer be freed by swsusp_free() */ | |
8357376d | 1936 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); |
f577eb30 RW |
1937 | if (!buffer) |
1938 | return -ENOMEM; | |
1939 | } | |
d3c1b24c | 1940 | if (!handle->cur) { |
d307c4a8 RW |
1941 | int error; |
1942 | ||
1943 | error = init_header((struct swsusp_info *)buffer); | |
1944 | if (error) | |
1945 | return error; | |
f577eb30 | 1946 | handle->buffer = buffer; |
b788db79 RW |
1947 | memory_bm_position_reset(&orig_bm); |
1948 | memory_bm_position_reset(©_bm); | |
d3c1b24c | 1949 | } else if (handle->cur <= nr_meta_pages) { |
3ecb01df | 1950 | clear_page(buffer); |
d3c1b24c JS |
1951 | pack_pfns(buffer, &orig_bm); |
1952 | } else { | |
1953 | struct page *page; | |
b788db79 | 1954 | |
d3c1b24c JS |
1955 | page = pfn_to_page(memory_bm_next_pfn(©_bm)); |
1956 | if (PageHighMem(page)) { | |
1957 | /* Highmem pages are copied to the buffer, | |
1958 | * because we can't return with a kmapped | |
1959 | * highmem page (we may not be called again). | |
1960 | */ | |
1961 | void *kaddr; | |
8357376d | 1962 | |
0de9a1e2 | 1963 | kaddr = kmap_atomic(page); |
3ecb01df | 1964 | copy_page(buffer, kaddr); |
0de9a1e2 | 1965 | kunmap_atomic(kaddr); |
d3c1b24c JS |
1966 | handle->buffer = buffer; |
1967 | } else { | |
1968 | handle->buffer = page_address(page); | |
f577eb30 | 1969 | } |
f577eb30 | 1970 | } |
d3c1b24c JS |
1971 | handle->cur++; |
1972 | return PAGE_SIZE; | |
f577eb30 RW |
1973 | } |
1974 | ||
1975 | /** | |
1976 | * mark_unsafe_pages - mark the pages that cannot be used for storing | |
1977 | * the image during resume, because they conflict with the pages that | |
1978 | * had been used before suspend | |
1979 | */ | |
1980 | ||
940864dd | 1981 | static int mark_unsafe_pages(struct memory_bitmap *bm) |
f577eb30 RW |
1982 | { |
1983 | struct zone *zone; | |
ae83c5ee | 1984 | unsigned long pfn, max_zone_pfn; |
f577eb30 RW |
1985 | |
1986 | /* Clear page flags */ | |
98e73dc5 | 1987 | for_each_populated_zone(zone) { |
c33bc315 | 1988 | max_zone_pfn = zone_end_pfn(zone); |
ae83c5ee RW |
1989 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
1990 | if (pfn_valid(pfn)) | |
7be98234 | 1991 | swsusp_unset_page_free(pfn_to_page(pfn)); |
f577eb30 RW |
1992 | } |
1993 | ||
940864dd RW |
1994 | /* Mark pages that correspond to the "original" pfns as "unsafe" */ |
1995 | memory_bm_position_reset(bm); | |
1996 | do { | |
1997 | pfn = memory_bm_next_pfn(bm); | |
1998 | if (likely(pfn != BM_END_OF_MAP)) { | |
1999 | if (likely(pfn_valid(pfn))) | |
7be98234 | 2000 | swsusp_set_page_free(pfn_to_page(pfn)); |
940864dd RW |
2001 | else |
2002 | return -EFAULT; | |
2003 | } | |
2004 | } while (pfn != BM_END_OF_MAP); | |
f577eb30 | 2005 | |
940864dd | 2006 | allocated_unsafe_pages = 0; |
968808b8 | 2007 | |
f577eb30 RW |
2008 | return 0; |
2009 | } | |
2010 | ||
940864dd RW |
2011 | static void |
2012 | duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src) | |
f577eb30 | 2013 | { |
940864dd RW |
2014 | unsigned long pfn; |
2015 | ||
2016 | memory_bm_position_reset(src); | |
2017 | pfn = memory_bm_next_pfn(src); | |
2018 | while (pfn != BM_END_OF_MAP) { | |
2019 | memory_bm_set_bit(dst, pfn); | |
2020 | pfn = memory_bm_next_pfn(src); | |
f577eb30 RW |
2021 | } |
2022 | } | |
2023 | ||
d307c4a8 | 2024 | static int check_header(struct swsusp_info *info) |
f577eb30 | 2025 | { |
d307c4a8 | 2026 | char *reason; |
f577eb30 | 2027 | |
d307c4a8 | 2028 | reason = check_image_kernel(info); |
0ed5fd13 | 2029 | if (!reason && info->num_physpages != get_num_physpages()) |
f577eb30 | 2030 | reason = "memory size"; |
f577eb30 | 2031 | if (reason) { |
23976728 | 2032 | printk(KERN_ERR "PM: Image mismatch: %s\n", reason); |
f577eb30 RW |
2033 | return -EPERM; |
2034 | } | |
2035 | return 0; | |
2036 | } | |
2037 | ||
2038 | /** | |
2039 | * load header - check the image header and copy data from it | |
2040 | */ | |
2041 | ||
940864dd RW |
2042 | static int |
2043 | load_header(struct swsusp_info *info) | |
f577eb30 RW |
2044 | { |
2045 | int error; | |
f577eb30 | 2046 | |
940864dd | 2047 | restore_pblist = NULL; |
f577eb30 RW |
2048 | error = check_header(info); |
2049 | if (!error) { | |
f577eb30 RW |
2050 | nr_copy_pages = info->image_pages; |
2051 | nr_meta_pages = info->pages - info->image_pages - 1; | |
2052 | } | |
2053 | return error; | |
2054 | } | |
2055 | ||
2056 | /** | |
940864dd RW |
2057 | * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set |
2058 | * the corresponding bit in the memory bitmap @bm | |
f577eb30 | 2059 | */ |
69643279 | 2060 | static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm) |
f577eb30 RW |
2061 | { |
2062 | int j; | |
2063 | ||
940864dd RW |
2064 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
2065 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
2066 | break; | |
2067 | ||
85055dd8 MS |
2068 | /* Extract and buffer page key for data page (s390 only). */ |
2069 | page_key_memorize(buf + j); | |
2070 | ||
69643279 RW |
2071 | if (memory_bm_pfn_present(bm, buf[j])) |
2072 | memory_bm_set_bit(bm, buf[j]); | |
2073 | else | |
2074 | return -EFAULT; | |
f577eb30 | 2075 | } |
69643279 RW |
2076 | |
2077 | return 0; | |
f577eb30 RW |
2078 | } |
2079 | ||
8357376d RW |
2080 | /* List of "safe" pages that may be used to store data loaded from the suspend |
2081 | * image | |
2082 | */ | |
2083 | static struct linked_page *safe_pages_list; | |
2084 | ||
2085 | #ifdef CONFIG_HIGHMEM | |
2086 | /* struct highmem_pbe is used for creating the list of highmem pages that | |
2087 | * should be restored atomically during the resume from disk, because the page | |
2088 | * frames they have occupied before the suspend are in use. | |
2089 | */ | |
2090 | struct highmem_pbe { | |
2091 | struct page *copy_page; /* data is here now */ | |
2092 | struct page *orig_page; /* data was here before the suspend */ | |
2093 | struct highmem_pbe *next; | |
2094 | }; | |
2095 | ||
2096 | /* List of highmem PBEs needed for restoring the highmem pages that were | |
2097 | * allocated before the suspend and included in the suspend image, but have | |
2098 | * also been allocated by the "resume" kernel, so their contents cannot be | |
2099 | * written directly to their "original" page frames. | |
2100 | */ | |
2101 | static struct highmem_pbe *highmem_pblist; | |
2102 | ||
2103 | /** | |
2104 | * count_highmem_image_pages - compute the number of highmem pages in the | |
2105 | * suspend image. The bits in the memory bitmap @bm that correspond to the | |
2106 | * image pages are assumed to be set. | |
2107 | */ | |
2108 | ||
2109 | static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) | |
2110 | { | |
2111 | unsigned long pfn; | |
2112 | unsigned int cnt = 0; | |
2113 | ||
2114 | memory_bm_position_reset(bm); | |
2115 | pfn = memory_bm_next_pfn(bm); | |
2116 | while (pfn != BM_END_OF_MAP) { | |
2117 | if (PageHighMem(pfn_to_page(pfn))) | |
2118 | cnt++; | |
2119 | ||
2120 | pfn = memory_bm_next_pfn(bm); | |
2121 | } | |
2122 | return cnt; | |
2123 | } | |
2124 | ||
2125 | /** | |
2126 | * prepare_highmem_image - try to allocate as many highmem pages as | |
2127 | * there are highmem image pages (@nr_highmem_p points to the variable | |
2128 | * containing the number of highmem image pages). The pages that are | |
2129 | * "safe" (ie. will not be overwritten when the suspend image is | |
2130 | * restored) have the corresponding bits set in @bm (it must be | |
2131 | * unitialized). | |
2132 | * | |
2133 | * NOTE: This function should not be called if there are no highmem | |
2134 | * image pages. | |
2135 | */ | |
2136 | ||
2137 | static unsigned int safe_highmem_pages; | |
2138 | ||
2139 | static struct memory_bitmap *safe_highmem_bm; | |
2140 | ||
2141 | static int | |
2142 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
2143 | { | |
2144 | unsigned int to_alloc; | |
2145 | ||
2146 | if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE)) | |
2147 | return -ENOMEM; | |
2148 | ||
2149 | if (get_highmem_buffer(PG_SAFE)) | |
2150 | return -ENOMEM; | |
2151 | ||
2152 | to_alloc = count_free_highmem_pages(); | |
2153 | if (to_alloc > *nr_highmem_p) | |
2154 | to_alloc = *nr_highmem_p; | |
2155 | else | |
2156 | *nr_highmem_p = to_alloc; | |
2157 | ||
2158 | safe_highmem_pages = 0; | |
2159 | while (to_alloc-- > 0) { | |
2160 | struct page *page; | |
2161 | ||
2162 | page = alloc_page(__GFP_HIGHMEM); | |
7be98234 | 2163 | if (!swsusp_page_is_free(page)) { |
8357376d RW |
2164 | /* The page is "safe", set its bit the bitmap */ |
2165 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
2166 | safe_highmem_pages++; | |
2167 | } | |
2168 | /* Mark the page as allocated */ | |
7be98234 RW |
2169 | swsusp_set_page_forbidden(page); |
2170 | swsusp_set_page_free(page); | |
8357376d RW |
2171 | } |
2172 | memory_bm_position_reset(bm); | |
2173 | safe_highmem_bm = bm; | |
2174 | return 0; | |
2175 | } | |
2176 | ||
2177 | /** | |
2178 | * get_highmem_page_buffer - for given highmem image page find the buffer | |
2179 | * that suspend_write_next() should set for its caller to write to. | |
2180 | * | |
2181 | * If the page is to be saved to its "original" page frame or a copy of | |
2182 | * the page is to be made in the highmem, @buffer is returned. Otherwise, | |
2183 | * the copy of the page is to be made in normal memory, so the address of | |
2184 | * the copy is returned. | |
2185 | * | |
2186 | * If @buffer is returned, the caller of suspend_write_next() will write | |
2187 | * the page's contents to @buffer, so they will have to be copied to the | |
2188 | * right location on the next call to suspend_write_next() and it is done | |
2189 | * with the help of copy_last_highmem_page(). For this purpose, if | |
2190 | * @buffer is returned, @last_highmem page is set to the page to which | |
2191 | * the data will have to be copied from @buffer. | |
2192 | */ | |
2193 | ||
2194 | static struct page *last_highmem_page; | |
2195 | ||
2196 | static void * | |
2197 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
2198 | { | |
2199 | struct highmem_pbe *pbe; | |
2200 | void *kaddr; | |
2201 | ||
7be98234 | 2202 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) { |
8357376d RW |
2203 | /* We have allocated the "original" page frame and we can |
2204 | * use it directly to store the loaded page. | |
2205 | */ | |
2206 | last_highmem_page = page; | |
2207 | return buffer; | |
2208 | } | |
2209 | /* The "original" page frame has not been allocated and we have to | |
2210 | * use a "safe" page frame to store the loaded page. | |
2211 | */ | |
2212 | pbe = chain_alloc(ca, sizeof(struct highmem_pbe)); | |
2213 | if (!pbe) { | |
2214 | swsusp_free(); | |
69643279 | 2215 | return ERR_PTR(-ENOMEM); |
8357376d RW |
2216 | } |
2217 | pbe->orig_page = page; | |
2218 | if (safe_highmem_pages > 0) { | |
2219 | struct page *tmp; | |
2220 | ||
2221 | /* Copy of the page will be stored in high memory */ | |
2222 | kaddr = buffer; | |
2223 | tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm)); | |
2224 | safe_highmem_pages--; | |
2225 | last_highmem_page = tmp; | |
2226 | pbe->copy_page = tmp; | |
2227 | } else { | |
2228 | /* Copy of the page will be stored in normal memory */ | |
2229 | kaddr = safe_pages_list; | |
2230 | safe_pages_list = safe_pages_list->next; | |
2231 | pbe->copy_page = virt_to_page(kaddr); | |
2232 | } | |
2233 | pbe->next = highmem_pblist; | |
2234 | highmem_pblist = pbe; | |
2235 | return kaddr; | |
2236 | } | |
2237 | ||
2238 | /** | |
2239 | * copy_last_highmem_page - copy the contents of a highmem image from | |
2240 | * @buffer, where the caller of snapshot_write_next() has place them, | |
2241 | * to the right location represented by @last_highmem_page . | |
2242 | */ | |
2243 | ||
2244 | static void copy_last_highmem_page(void) | |
2245 | { | |
2246 | if (last_highmem_page) { | |
2247 | void *dst; | |
2248 | ||
0de9a1e2 | 2249 | dst = kmap_atomic(last_highmem_page); |
3ecb01df | 2250 | copy_page(dst, buffer); |
0de9a1e2 | 2251 | kunmap_atomic(dst); |
8357376d RW |
2252 | last_highmem_page = NULL; |
2253 | } | |
2254 | } | |
2255 | ||
2256 | static inline int last_highmem_page_copied(void) | |
2257 | { | |
2258 | return !last_highmem_page; | |
2259 | } | |
2260 | ||
2261 | static inline void free_highmem_data(void) | |
2262 | { | |
2263 | if (safe_highmem_bm) | |
2264 | memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR); | |
2265 | ||
2266 | if (buffer) | |
2267 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
2268 | } | |
2269 | #else | |
2270 | static inline int get_safe_write_buffer(void) { return 0; } | |
2271 | ||
2272 | static unsigned int | |
2273 | count_highmem_image_pages(struct memory_bitmap *bm) { return 0; } | |
2274 | ||
2275 | static inline int | |
2276 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
2277 | { | |
2278 | return 0; | |
2279 | } | |
2280 | ||
2281 | static inline void * | |
2282 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
2283 | { | |
69643279 | 2284 | return ERR_PTR(-EINVAL); |
8357376d RW |
2285 | } |
2286 | ||
2287 | static inline void copy_last_highmem_page(void) {} | |
2288 | static inline int last_highmem_page_copied(void) { return 1; } | |
2289 | static inline void free_highmem_data(void) {} | |
2290 | #endif /* CONFIG_HIGHMEM */ | |
2291 | ||
f577eb30 | 2292 | /** |
940864dd RW |
2293 | * prepare_image - use the memory bitmap @bm to mark the pages that will |
2294 | * be overwritten in the process of restoring the system memory state | |
2295 | * from the suspend image ("unsafe" pages) and allocate memory for the | |
2296 | * image. | |
968808b8 | 2297 | * |
940864dd RW |
2298 | * The idea is to allocate a new memory bitmap first and then allocate |
2299 | * as many pages as needed for the image data, but not to assign these | |
2300 | * pages to specific tasks initially. Instead, we just mark them as | |
8357376d RW |
2301 | * allocated and create a lists of "safe" pages that will be used |
2302 | * later. On systems with high memory a list of "safe" highmem pages is | |
2303 | * also created. | |
f577eb30 RW |
2304 | */ |
2305 | ||
940864dd RW |
2306 | #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe)) |
2307 | ||
940864dd RW |
2308 | static int |
2309 | prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm) | |
f577eb30 | 2310 | { |
8357376d | 2311 | unsigned int nr_pages, nr_highmem; |
940864dd RW |
2312 | struct linked_page *sp_list, *lp; |
2313 | int error; | |
f577eb30 | 2314 | |
8357376d RW |
2315 | /* If there is no highmem, the buffer will not be necessary */ |
2316 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
2317 | buffer = NULL; | |
2318 | ||
2319 | nr_highmem = count_highmem_image_pages(bm); | |
940864dd RW |
2320 | error = mark_unsafe_pages(bm); |
2321 | if (error) | |
2322 | goto Free; | |
2323 | ||
2324 | error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE); | |
2325 | if (error) | |
2326 | goto Free; | |
2327 | ||
2328 | duplicate_memory_bitmap(new_bm, bm); | |
2329 | memory_bm_free(bm, PG_UNSAFE_KEEP); | |
8357376d RW |
2330 | if (nr_highmem > 0) { |
2331 | error = prepare_highmem_image(bm, &nr_highmem); | |
2332 | if (error) | |
2333 | goto Free; | |
2334 | } | |
940864dd RW |
2335 | /* Reserve some safe pages for potential later use. |
2336 | * | |
2337 | * NOTE: This way we make sure there will be enough safe pages for the | |
2338 | * chain_alloc() in get_buffer(). It is a bit wasteful, but | |
2339 | * nr_copy_pages cannot be greater than 50% of the memory anyway. | |
2340 | */ | |
2341 | sp_list = NULL; | |
2342 | /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */ | |
8357376d | 2343 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
2344 | nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE); |
2345 | while (nr_pages > 0) { | |
8357376d | 2346 | lp = get_image_page(GFP_ATOMIC, PG_SAFE); |
940864dd | 2347 | if (!lp) { |
f577eb30 | 2348 | error = -ENOMEM; |
940864dd RW |
2349 | goto Free; |
2350 | } | |
2351 | lp->next = sp_list; | |
2352 | sp_list = lp; | |
2353 | nr_pages--; | |
f577eb30 | 2354 | } |
940864dd RW |
2355 | /* Preallocate memory for the image */ |
2356 | safe_pages_list = NULL; | |
8357376d | 2357 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
2358 | while (nr_pages > 0) { |
2359 | lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC); | |
2360 | if (!lp) { | |
2361 | error = -ENOMEM; | |
2362 | goto Free; | |
2363 | } | |
7be98234 | 2364 | if (!swsusp_page_is_free(virt_to_page(lp))) { |
940864dd RW |
2365 | /* The page is "safe", add it to the list */ |
2366 | lp->next = safe_pages_list; | |
2367 | safe_pages_list = lp; | |
968808b8 | 2368 | } |
940864dd | 2369 | /* Mark the page as allocated */ |
7be98234 RW |
2370 | swsusp_set_page_forbidden(virt_to_page(lp)); |
2371 | swsusp_set_page_free(virt_to_page(lp)); | |
940864dd | 2372 | nr_pages--; |
968808b8 | 2373 | } |
940864dd RW |
2374 | /* Free the reserved safe pages so that chain_alloc() can use them */ |
2375 | while (sp_list) { | |
2376 | lp = sp_list->next; | |
2377 | free_image_page(sp_list, PG_UNSAFE_CLEAR); | |
2378 | sp_list = lp; | |
f577eb30 | 2379 | } |
940864dd RW |
2380 | return 0; |
2381 | ||
59a49335 | 2382 | Free: |
940864dd | 2383 | swsusp_free(); |
f577eb30 RW |
2384 | return error; |
2385 | } | |
2386 | ||
940864dd RW |
2387 | /** |
2388 | * get_buffer - compute the address that snapshot_write_next() should | |
2389 | * set for its caller to write to. | |
2390 | */ | |
2391 | ||
2392 | static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca) | |
968808b8 | 2393 | { |
940864dd | 2394 | struct pbe *pbe; |
69643279 RW |
2395 | struct page *page; |
2396 | unsigned long pfn = memory_bm_next_pfn(bm); | |
968808b8 | 2397 | |
69643279 RW |
2398 | if (pfn == BM_END_OF_MAP) |
2399 | return ERR_PTR(-EFAULT); | |
2400 | ||
2401 | page = pfn_to_page(pfn); | |
8357376d RW |
2402 | if (PageHighMem(page)) |
2403 | return get_highmem_page_buffer(page, ca); | |
2404 | ||
7be98234 | 2405 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) |
940864dd RW |
2406 | /* We have allocated the "original" page frame and we can |
2407 | * use it directly to store the loaded page. | |
968808b8 | 2408 | */ |
940864dd RW |
2409 | return page_address(page); |
2410 | ||
2411 | /* The "original" page frame has not been allocated and we have to | |
2412 | * use a "safe" page frame to store the loaded page. | |
968808b8 | 2413 | */ |
940864dd RW |
2414 | pbe = chain_alloc(ca, sizeof(struct pbe)); |
2415 | if (!pbe) { | |
2416 | swsusp_free(); | |
69643279 | 2417 | return ERR_PTR(-ENOMEM); |
940864dd | 2418 | } |
8357376d RW |
2419 | pbe->orig_address = page_address(page); |
2420 | pbe->address = safe_pages_list; | |
940864dd RW |
2421 | safe_pages_list = safe_pages_list->next; |
2422 | pbe->next = restore_pblist; | |
2423 | restore_pblist = pbe; | |
8357376d | 2424 | return pbe->address; |
968808b8 RW |
2425 | } |
2426 | ||
f577eb30 RW |
2427 | /** |
2428 | * snapshot_write_next - used for writing the system memory snapshot. | |
2429 | * | |
2430 | * On the first call to it @handle should point to a zeroed | |
2431 | * snapshot_handle structure. The structure gets updated and a pointer | |
2432 | * to it should be passed to this function every next time. | |
2433 | * | |
f577eb30 RW |
2434 | * On success the function returns a positive number. Then, the caller |
2435 | * is allowed to write up to the returned number of bytes to the memory | |
d3c1b24c | 2436 | * location computed by the data_of() macro. |
f577eb30 RW |
2437 | * |
2438 | * The function returns 0 to indicate the "end of file" condition, | |
2439 | * and a negative number is returned on error. In such cases the | |
2440 | * structure pointed to by @handle is not updated and should not be used | |
2441 | * any more. | |
2442 | */ | |
2443 | ||
d3c1b24c | 2444 | int snapshot_write_next(struct snapshot_handle *handle) |
f577eb30 | 2445 | { |
940864dd | 2446 | static struct chain_allocator ca; |
f577eb30 RW |
2447 | int error = 0; |
2448 | ||
940864dd | 2449 | /* Check if we have already loaded the entire image */ |
d3c1b24c | 2450 | if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 2451 | return 0; |
940864dd | 2452 | |
d3c1b24c JS |
2453 | handle->sync_read = 1; |
2454 | ||
2455 | if (!handle->cur) { | |
8357376d RW |
2456 | if (!buffer) |
2457 | /* This makes the buffer be freed by swsusp_free() */ | |
2458 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); | |
2459 | ||
f577eb30 RW |
2460 | if (!buffer) |
2461 | return -ENOMEM; | |
8357376d | 2462 | |
f577eb30 | 2463 | handle->buffer = buffer; |
d3c1b24c JS |
2464 | } else if (handle->cur == 1) { |
2465 | error = load_header(buffer); | |
2466 | if (error) | |
2467 | return error; | |
940864dd | 2468 | |
d3c1b24c JS |
2469 | error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY); |
2470 | if (error) | |
2471 | return error; | |
2472 | ||
85055dd8 MS |
2473 | /* Allocate buffer for page keys. */ |
2474 | error = page_key_alloc(nr_copy_pages); | |
2475 | if (error) | |
2476 | return error; | |
2477 | ||
d3c1b24c JS |
2478 | } else if (handle->cur <= nr_meta_pages + 1) { |
2479 | error = unpack_orig_pfns(buffer, ©_bm); | |
2480 | if (error) | |
2481 | return error; | |
940864dd | 2482 | |
d3c1b24c JS |
2483 | if (handle->cur == nr_meta_pages + 1) { |
2484 | error = prepare_image(&orig_bm, ©_bm); | |
69643279 RW |
2485 | if (error) |
2486 | return error; | |
2487 | ||
d3c1b24c JS |
2488 | chain_init(&ca, GFP_ATOMIC, PG_SAFE); |
2489 | memory_bm_position_reset(&orig_bm); | |
2490 | restore_pblist = NULL; | |
940864dd | 2491 | handle->buffer = get_buffer(&orig_bm, &ca); |
d3c1b24c | 2492 | handle->sync_read = 0; |
69643279 RW |
2493 | if (IS_ERR(handle->buffer)) |
2494 | return PTR_ERR(handle->buffer); | |
f577eb30 | 2495 | } |
f577eb30 | 2496 | } else { |
d3c1b24c | 2497 | copy_last_highmem_page(); |
85055dd8 MS |
2498 | /* Restore page key for data page (s390 only). */ |
2499 | page_key_write(handle->buffer); | |
d3c1b24c JS |
2500 | handle->buffer = get_buffer(&orig_bm, &ca); |
2501 | if (IS_ERR(handle->buffer)) | |
2502 | return PTR_ERR(handle->buffer); | |
2503 | if (handle->buffer != buffer) | |
2504 | handle->sync_read = 0; | |
f577eb30 | 2505 | } |
d3c1b24c JS |
2506 | handle->cur++; |
2507 | return PAGE_SIZE; | |
f577eb30 RW |
2508 | } |
2509 | ||
8357376d RW |
2510 | /** |
2511 | * snapshot_write_finalize - must be called after the last call to | |
2512 | * snapshot_write_next() in case the last page in the image happens | |
2513 | * to be a highmem page and its contents should be stored in the | |
2514 | * highmem. Additionally, it releases the memory that will not be | |
2515 | * used any more. | |
2516 | */ | |
2517 | ||
2518 | void snapshot_write_finalize(struct snapshot_handle *handle) | |
2519 | { | |
2520 | copy_last_highmem_page(); | |
85055dd8 MS |
2521 | /* Restore page key for data page (s390 only). */ |
2522 | page_key_write(handle->buffer); | |
2523 | page_key_free(); | |
8357376d | 2524 | /* Free only if we have loaded the image entirely */ |
d3c1b24c | 2525 | if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) { |
8357376d RW |
2526 | memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR); |
2527 | free_highmem_data(); | |
2528 | } | |
2529 | } | |
2530 | ||
f577eb30 RW |
2531 | int snapshot_image_loaded(struct snapshot_handle *handle) |
2532 | { | |
8357376d | 2533 | return !(!nr_copy_pages || !last_highmem_page_copied() || |
940864dd RW |
2534 | handle->cur <= nr_meta_pages + nr_copy_pages); |
2535 | } | |
2536 | ||
8357376d RW |
2537 | #ifdef CONFIG_HIGHMEM |
2538 | /* Assumes that @buf is ready and points to a "safe" page */ | |
2539 | static inline void | |
2540 | swap_two_pages_data(struct page *p1, struct page *p2, void *buf) | |
940864dd | 2541 | { |
8357376d RW |
2542 | void *kaddr1, *kaddr2; |
2543 | ||
0de9a1e2 CW |
2544 | kaddr1 = kmap_atomic(p1); |
2545 | kaddr2 = kmap_atomic(p2); | |
3ecb01df JB |
2546 | copy_page(buf, kaddr1); |
2547 | copy_page(kaddr1, kaddr2); | |
2548 | copy_page(kaddr2, buf); | |
0de9a1e2 CW |
2549 | kunmap_atomic(kaddr2); |
2550 | kunmap_atomic(kaddr1); | |
8357376d RW |
2551 | } |
2552 | ||
2553 | /** | |
2554 | * restore_highmem - for each highmem page that was allocated before | |
2555 | * the suspend and included in the suspend image, and also has been | |
2556 | * allocated by the "resume" kernel swap its current (ie. "before | |
2557 | * resume") contents with the previous (ie. "before suspend") one. | |
2558 | * | |
2559 | * If the resume eventually fails, we can call this function once | |
2560 | * again and restore the "before resume" highmem state. | |
2561 | */ | |
2562 | ||
2563 | int restore_highmem(void) | |
2564 | { | |
2565 | struct highmem_pbe *pbe = highmem_pblist; | |
2566 | void *buf; | |
2567 | ||
2568 | if (!pbe) | |
2569 | return 0; | |
2570 | ||
2571 | buf = get_image_page(GFP_ATOMIC, PG_SAFE); | |
2572 | if (!buf) | |
2573 | return -ENOMEM; | |
2574 | ||
2575 | while (pbe) { | |
2576 | swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf); | |
2577 | pbe = pbe->next; | |
2578 | } | |
2579 | free_image_page(buf, PG_UNSAFE_CLEAR); | |
2580 | return 0; | |
f577eb30 | 2581 | } |
8357376d | 2582 | #endif /* CONFIG_HIGHMEM */ |