]>
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 | 269 | |
f469f02d JR |
270 | /* |
271 | * struct rtree_node is a wrapper struct to link the nodes | |
272 | * of the rtree together for easy linear iteration over | |
273 | * bits and easy freeing | |
274 | */ | |
275 | struct rtree_node { | |
276 | struct list_head list; | |
277 | unsigned long *data; | |
278 | }; | |
279 | ||
280 | /* | |
281 | * struct mem_zone_bm_rtree represents a bitmap used for one | |
282 | * populated memory zone. | |
283 | */ | |
284 | struct mem_zone_bm_rtree { | |
285 | struct list_head list; /* Link Zones together */ | |
286 | struct list_head nodes; /* Radix Tree inner nodes */ | |
287 | struct list_head leaves; /* Radix Tree leaves */ | |
288 | unsigned long start_pfn; /* Zone start page frame */ | |
289 | unsigned long end_pfn; /* Zone end page frame + 1 */ | |
290 | struct rtree_node *rtree; /* Radix Tree Root */ | |
291 | int levels; /* Number of Radix Tree Levels */ | |
292 | unsigned int blocks; /* Number of Bitmap Blocks */ | |
293 | }; | |
294 | ||
b788db79 RW |
295 | /* strcut bm_position is used for browsing memory bitmaps */ |
296 | ||
297 | struct bm_position { | |
3a20cb17 JR |
298 | struct mem_zone_bm_rtree *zone; |
299 | struct rtree_node *node; | |
300 | unsigned long node_pfn; | |
301 | int node_bit; | |
b788db79 RW |
302 | }; |
303 | ||
304 | struct memory_bitmap { | |
f469f02d | 305 | struct list_head zones; |
b788db79 RW |
306 | struct linked_page *p_list; /* list of pages used to store zone |
307 | * bitmap objects and bitmap block | |
308 | * objects | |
309 | */ | |
310 | struct bm_position cur; /* most recently used bit position */ | |
311 | }; | |
312 | ||
313 | /* Functions that operate on memory bitmaps */ | |
314 | ||
f469f02d JR |
315 | #define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long)) |
316 | #if BITS_PER_LONG == 32 | |
317 | #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2) | |
318 | #else | |
319 | #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3) | |
320 | #endif | |
321 | #define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1) | |
322 | ||
323 | /* | |
324 | * alloc_rtree_node - Allocate a new node and add it to the radix tree. | |
325 | * | |
326 | * This function is used to allocate inner nodes as well as the | |
327 | * leave nodes of the radix tree. It also adds the node to the | |
328 | * corresponding linked list passed in by the *list parameter. | |
329 | */ | |
330 | static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed, | |
331 | struct chain_allocator *ca, | |
332 | struct list_head *list) | |
333 | { | |
334 | struct rtree_node *node; | |
335 | ||
336 | node = chain_alloc(ca, sizeof(struct rtree_node)); | |
337 | if (!node) | |
338 | return NULL; | |
339 | ||
340 | node->data = get_image_page(gfp_mask, safe_needed); | |
341 | if (!node->data) | |
342 | return NULL; | |
343 | ||
344 | list_add_tail(&node->list, list); | |
345 | ||
346 | return node; | |
347 | } | |
348 | ||
349 | /* | |
350 | * add_rtree_block - Add a new leave node to the radix tree | |
351 | * | |
352 | * The leave nodes need to be allocated in order to keep the leaves | |
353 | * linked list in order. This is guaranteed by the zone->blocks | |
354 | * counter. | |
355 | */ | |
356 | static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask, | |
357 | int safe_needed, struct chain_allocator *ca) | |
358 | { | |
359 | struct rtree_node *node, *block, **dst; | |
360 | unsigned int levels_needed, block_nr; | |
361 | int i; | |
362 | ||
363 | block_nr = zone->blocks; | |
364 | levels_needed = 0; | |
365 | ||
366 | /* How many levels do we need for this block nr? */ | |
367 | while (block_nr) { | |
368 | levels_needed += 1; | |
369 | block_nr >>= BM_RTREE_LEVEL_SHIFT; | |
370 | } | |
371 | ||
372 | /* Make sure the rtree has enough levels */ | |
373 | for (i = zone->levels; i < levels_needed; i++) { | |
374 | node = alloc_rtree_node(gfp_mask, safe_needed, ca, | |
375 | &zone->nodes); | |
376 | if (!node) | |
377 | return -ENOMEM; | |
378 | ||
379 | node->data[0] = (unsigned long)zone->rtree; | |
380 | zone->rtree = node; | |
381 | zone->levels += 1; | |
382 | } | |
383 | ||
384 | /* Allocate new block */ | |
385 | block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves); | |
386 | if (!block) | |
387 | return -ENOMEM; | |
388 | ||
389 | /* Now walk the rtree to insert the block */ | |
390 | node = zone->rtree; | |
391 | dst = &zone->rtree; | |
392 | block_nr = zone->blocks; | |
393 | for (i = zone->levels; i > 0; i--) { | |
394 | int index; | |
395 | ||
396 | if (!node) { | |
397 | node = alloc_rtree_node(gfp_mask, safe_needed, ca, | |
398 | &zone->nodes); | |
399 | if (!node) | |
400 | return -ENOMEM; | |
401 | *dst = node; | |
402 | } | |
403 | ||
404 | index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT); | |
405 | index &= BM_RTREE_LEVEL_MASK; | |
406 | dst = (struct rtree_node **)&((*dst)->data[index]); | |
407 | node = *dst; | |
408 | } | |
409 | ||
410 | zone->blocks += 1; | |
411 | *dst = block; | |
412 | ||
413 | return 0; | |
414 | } | |
415 | ||
416 | static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone, | |
417 | int clear_nosave_free); | |
418 | ||
419 | /* | |
420 | * create_zone_bm_rtree - create a radix tree for one zone | |
421 | * | |
422 | * Allocated the mem_zone_bm_rtree structure and initializes it. | |
423 | * This function also allocated and builds the radix tree for the | |
424 | * zone. | |
425 | */ | |
426 | static struct mem_zone_bm_rtree * | |
427 | create_zone_bm_rtree(gfp_t gfp_mask, int safe_needed, | |
428 | struct chain_allocator *ca, | |
429 | unsigned long start, unsigned long end) | |
430 | { | |
431 | struct mem_zone_bm_rtree *zone; | |
432 | unsigned int i, nr_blocks; | |
433 | unsigned long pages; | |
434 | ||
435 | pages = end - start; | |
436 | zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree)); | |
437 | if (!zone) | |
438 | return NULL; | |
439 | ||
440 | INIT_LIST_HEAD(&zone->nodes); | |
441 | INIT_LIST_HEAD(&zone->leaves); | |
442 | zone->start_pfn = start; | |
443 | zone->end_pfn = end; | |
444 | nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK); | |
445 | ||
446 | for (i = 0; i < nr_blocks; i++) { | |
447 | if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) { | |
448 | free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR); | |
449 | return NULL; | |
450 | } | |
451 | } | |
452 | ||
453 | return zone; | |
454 | } | |
455 | ||
456 | /* | |
457 | * free_zone_bm_rtree - Free the memory of the radix tree | |
458 | * | |
459 | * Free all node pages of the radix tree. The mem_zone_bm_rtree | |
460 | * structure itself is not freed here nor are the rtree_node | |
461 | * structs. | |
462 | */ | |
463 | static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone, | |
464 | int clear_nosave_free) | |
465 | { | |
466 | struct rtree_node *node; | |
467 | ||
468 | list_for_each_entry(node, &zone->nodes, list) | |
469 | free_image_page(node->data, clear_nosave_free); | |
470 | ||
471 | list_for_each_entry(node, &zone->leaves, list) | |
472 | free_image_page(node->data, clear_nosave_free); | |
473 | } | |
474 | ||
b788db79 RW |
475 | static void memory_bm_position_reset(struct memory_bitmap *bm) |
476 | { | |
3a20cb17 JR |
477 | bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree, |
478 | list); | |
479 | bm->cur.node = list_entry(bm->cur.zone->leaves.next, | |
480 | struct rtree_node, list); | |
481 | bm->cur.node_pfn = 0; | |
482 | bm->cur.node_bit = 0; | |
b788db79 RW |
483 | } |
484 | ||
485 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free); | |
486 | ||
846705de RW |
487 | struct mem_extent { |
488 | struct list_head hook; | |
489 | unsigned long start; | |
490 | unsigned long end; | |
491 | }; | |
492 | ||
b788db79 | 493 | /** |
846705de RW |
494 | * free_mem_extents - free a list of memory extents |
495 | * @list - list of extents to empty | |
b788db79 | 496 | */ |
846705de RW |
497 | static void free_mem_extents(struct list_head *list) |
498 | { | |
499 | struct mem_extent *ext, *aux; | |
b788db79 | 500 | |
846705de RW |
501 | list_for_each_entry_safe(ext, aux, list, hook) { |
502 | list_del(&ext->hook); | |
503 | kfree(ext); | |
504 | } | |
505 | } | |
506 | ||
507 | /** | |
508 | * create_mem_extents - create a list of memory extents representing | |
509 | * contiguous ranges of PFNs | |
510 | * @list - list to put the extents into | |
511 | * @gfp_mask - mask to use for memory allocations | |
512 | */ | |
513 | static int create_mem_extents(struct list_head *list, gfp_t gfp_mask) | |
b788db79 | 514 | { |
846705de | 515 | struct zone *zone; |
b788db79 | 516 | |
846705de | 517 | INIT_LIST_HEAD(list); |
b788db79 | 518 | |
ee99c71c | 519 | for_each_populated_zone(zone) { |
846705de RW |
520 | unsigned long zone_start, zone_end; |
521 | struct mem_extent *ext, *cur, *aux; | |
522 | ||
846705de | 523 | zone_start = zone->zone_start_pfn; |
c33bc315 | 524 | zone_end = zone_end_pfn(zone); |
846705de RW |
525 | |
526 | list_for_each_entry(ext, list, hook) | |
527 | if (zone_start <= ext->end) | |
528 | break; | |
b788db79 | 529 | |
846705de RW |
530 | if (&ext->hook == list || zone_end < ext->start) { |
531 | /* New extent is necessary */ | |
532 | struct mem_extent *new_ext; | |
533 | ||
534 | new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask); | |
535 | if (!new_ext) { | |
536 | free_mem_extents(list); | |
537 | return -ENOMEM; | |
538 | } | |
539 | new_ext->start = zone_start; | |
540 | new_ext->end = zone_end; | |
541 | list_add_tail(&new_ext->hook, &ext->hook); | |
542 | continue; | |
543 | } | |
544 | ||
545 | /* Merge this zone's range of PFNs with the existing one */ | |
546 | if (zone_start < ext->start) | |
547 | ext->start = zone_start; | |
548 | if (zone_end > ext->end) | |
549 | ext->end = zone_end; | |
550 | ||
551 | /* More merging may be possible */ | |
552 | cur = ext; | |
553 | list_for_each_entry_safe_continue(cur, aux, list, hook) { | |
554 | if (zone_end < cur->start) | |
555 | break; | |
556 | if (zone_end < cur->end) | |
557 | ext->end = cur->end; | |
558 | list_del(&cur->hook); | |
559 | kfree(cur); | |
560 | } | |
b788db79 | 561 | } |
846705de RW |
562 | |
563 | return 0; | |
b788db79 RW |
564 | } |
565 | ||
566 | /** | |
567 | * memory_bm_create - allocate memory for a memory bitmap | |
568 | */ | |
b788db79 RW |
569 | static int |
570 | memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed) | |
571 | { | |
572 | struct chain_allocator ca; | |
846705de RW |
573 | struct list_head mem_extents; |
574 | struct mem_extent *ext; | |
575 | int error; | |
b788db79 RW |
576 | |
577 | chain_init(&ca, gfp_mask, safe_needed); | |
f469f02d | 578 | INIT_LIST_HEAD(&bm->zones); |
b788db79 | 579 | |
846705de RW |
580 | error = create_mem_extents(&mem_extents, gfp_mask); |
581 | if (error) | |
582 | return error; | |
b788db79 | 583 | |
846705de | 584 | list_for_each_entry(ext, &mem_extents, hook) { |
f469f02d | 585 | struct mem_zone_bm_rtree *zone; |
f469f02d JR |
586 | |
587 | zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca, | |
588 | ext->start, ext->end); | |
9047eb62 JR |
589 | if (!zone) { |
590 | error = -ENOMEM; | |
f469f02d | 591 | goto Error; |
9047eb62 | 592 | } |
f469f02d | 593 | list_add_tail(&zone->list, &bm->zones); |
b788db79 | 594 | } |
846705de | 595 | |
b788db79 RW |
596 | bm->p_list = ca.chain; |
597 | memory_bm_position_reset(bm); | |
846705de RW |
598 | Exit: |
599 | free_mem_extents(&mem_extents); | |
600 | return error; | |
b788db79 | 601 | |
846705de | 602 | Error: |
b788db79 RW |
603 | bm->p_list = ca.chain; |
604 | memory_bm_free(bm, PG_UNSAFE_CLEAR); | |
846705de | 605 | goto Exit; |
b788db79 RW |
606 | } |
607 | ||
608 | /** | |
609 | * memory_bm_free - free memory occupied by the memory bitmap @bm | |
610 | */ | |
b788db79 RW |
611 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free) |
612 | { | |
f469f02d | 613 | struct mem_zone_bm_rtree *zone; |
b788db79 | 614 | |
f469f02d JR |
615 | list_for_each_entry(zone, &bm->zones, list) |
616 | free_zone_bm_rtree(zone, clear_nosave_free); | |
617 | ||
b788db79 | 618 | free_list_of_pages(bm->p_list, clear_nosave_free); |
846705de | 619 | |
f469f02d | 620 | INIT_LIST_HEAD(&bm->zones); |
b788db79 RW |
621 | } |
622 | ||
623 | /** | |
9047eb62 JR |
624 | * memory_bm_find_bit - Find the bit for pfn in the memory |
625 | * bitmap | |
07a33823 | 626 | * |
9047eb62 JR |
627 | * Find the bit in the bitmap @bm that corresponds to given pfn. |
628 | * The cur.zone, cur.block and cur.node_pfn member of @bm are | |
629 | * updated. | |
630 | * It walks the radix tree to find the page which contains the bit for | |
07a33823 JR |
631 | * pfn and returns the bit position in **addr and *bit_nr. |
632 | */ | |
9047eb62 JR |
633 | static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn, |
634 | void **addr, unsigned int *bit_nr) | |
07a33823 JR |
635 | { |
636 | struct mem_zone_bm_rtree *curr, *zone; | |
637 | struct rtree_node *node; | |
638 | int i, block_nr; | |
639 | ||
3a20cb17 JR |
640 | zone = bm->cur.zone; |
641 | ||
642 | if (pfn >= zone->start_pfn && pfn < zone->end_pfn) | |
643 | goto zone_found; | |
644 | ||
07a33823 JR |
645 | zone = NULL; |
646 | ||
647 | /* Find the right zone */ | |
648 | list_for_each_entry(curr, &bm->zones, list) { | |
649 | if (pfn >= curr->start_pfn && pfn < curr->end_pfn) { | |
650 | zone = curr; | |
651 | break; | |
652 | } | |
653 | } | |
654 | ||
655 | if (!zone) | |
656 | return -EFAULT; | |
657 | ||
3a20cb17 | 658 | zone_found: |
07a33823 JR |
659 | /* |
660 | * We have a zone. Now walk the radix tree to find the leave | |
661 | * node for our pfn. | |
662 | */ | |
3a20cb17 JR |
663 | |
664 | node = bm->cur.node; | |
665 | if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn) | |
666 | goto node_found; | |
667 | ||
07a33823 JR |
668 | node = zone->rtree; |
669 | block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT; | |
670 | ||
671 | for (i = zone->levels; i > 0; i--) { | |
672 | int index; | |
673 | ||
674 | index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT); | |
675 | index &= BM_RTREE_LEVEL_MASK; | |
676 | BUG_ON(node->data[index] == 0); | |
677 | node = (struct rtree_node *)node->data[index]; | |
678 | } | |
679 | ||
3a20cb17 JR |
680 | node_found: |
681 | /* Update last position */ | |
682 | bm->cur.zone = zone; | |
683 | bm->cur.node = node; | |
684 | bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK; | |
685 | ||
07a33823 JR |
686 | /* Set return values */ |
687 | *addr = node->data; | |
688 | *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK; | |
689 | ||
690 | return 0; | |
691 | } | |
692 | ||
74dfd666 RW |
693 | static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn) |
694 | { | |
695 | void *addr; | |
696 | unsigned int bit; | |
a82f7119 | 697 | int error; |
74dfd666 | 698 | |
a82f7119 RW |
699 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
700 | BUG_ON(error); | |
74dfd666 RW |
701 | set_bit(bit, addr); |
702 | } | |
703 | ||
a82f7119 RW |
704 | static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn) |
705 | { | |
706 | void *addr; | |
707 | unsigned int bit; | |
708 | int error; | |
709 | ||
710 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); | |
07a33823 JR |
711 | if (!error) |
712 | set_bit(bit, addr); | |
713 | ||
a82f7119 RW |
714 | return error; |
715 | } | |
716 | ||
74dfd666 RW |
717 | static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn) |
718 | { | |
719 | void *addr; | |
720 | unsigned int bit; | |
a82f7119 | 721 | int error; |
74dfd666 | 722 | |
a82f7119 RW |
723 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
724 | BUG_ON(error); | |
74dfd666 RW |
725 | clear_bit(bit, addr); |
726 | } | |
727 | ||
6efde38f JR |
728 | static void memory_bm_clear_current(struct memory_bitmap *bm) |
729 | { | |
730 | int bit; | |
731 | ||
732 | bit = max(bm->cur.node_bit - 1, 0); | |
733 | clear_bit(bit, bm->cur.node->data); | |
6efde38f JR |
734 | } |
735 | ||
74dfd666 RW |
736 | static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn) |
737 | { | |
738 | void *addr; | |
739 | unsigned int bit; | |
9047eb62 | 740 | int error; |
74dfd666 | 741 | |
a82f7119 RW |
742 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); |
743 | BUG_ON(error); | |
9047eb62 | 744 | return test_bit(bit, addr); |
b788db79 RW |
745 | } |
746 | ||
69643279 RW |
747 | static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn) |
748 | { | |
749 | void *addr; | |
750 | unsigned int bit; | |
07a33823 | 751 | |
9047eb62 | 752 | return !memory_bm_find_bit(bm, pfn, &addr, &bit); |
b788db79 RW |
753 | } |
754 | ||
3a20cb17 JR |
755 | /* |
756 | * rtree_next_node - Jumps to the next leave node | |
757 | * | |
758 | * Sets the position to the beginning of the next node in the | |
759 | * memory bitmap. This is either the next node in the current | |
760 | * zone's radix tree or the first node in the radix tree of the | |
761 | * next zone. | |
762 | * | |
763 | * Returns true if there is a next node, false otherwise. | |
764 | */ | |
765 | static bool rtree_next_node(struct memory_bitmap *bm) | |
766 | { | |
767 | bm->cur.node = list_entry(bm->cur.node->list.next, | |
768 | struct rtree_node, list); | |
769 | if (&bm->cur.node->list != &bm->cur.zone->leaves) { | |
770 | bm->cur.node_pfn += BM_BITS_PER_BLOCK; | |
771 | bm->cur.node_bit = 0; | |
772 | return true; | |
773 | } | |
774 | ||
775 | /* No more nodes, goto next zone */ | |
776 | bm->cur.zone = list_entry(bm->cur.zone->list.next, | |
777 | struct mem_zone_bm_rtree, list); | |
778 | if (&bm->cur.zone->list != &bm->zones) { | |
779 | bm->cur.node = list_entry(bm->cur.zone->leaves.next, | |
780 | struct rtree_node, list); | |
781 | bm->cur.node_pfn = 0; | |
782 | bm->cur.node_bit = 0; | |
783 | return true; | |
784 | } | |
785 | ||
786 | /* No more zones */ | |
787 | return false; | |
788 | } | |
789 | ||
9047eb62 JR |
790 | /** |
791 | * memory_bm_rtree_next_pfn - Find the next set bit in the bitmap @bm | |
3a20cb17 JR |
792 | * |
793 | * Starting from the last returned position this function searches | |
794 | * for the next set bit in the memory bitmap and returns its | |
795 | * number. If no more bit is set BM_END_OF_MAP is returned. | |
9047eb62 JR |
796 | * |
797 | * It is required to run memory_bm_position_reset() before the | |
798 | * first call to this function. | |
3a20cb17 | 799 | */ |
9047eb62 | 800 | static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm) |
3a20cb17 JR |
801 | { |
802 | unsigned long bits, pfn, pages; | |
803 | int bit; | |
804 | ||
805 | do { | |
806 | pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn; | |
807 | bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK); | |
808 | bit = find_next_bit(bm->cur.node->data, bits, | |
809 | bm->cur.node_bit); | |
810 | if (bit < bits) { | |
811 | pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit; | |
812 | bm->cur.node_bit = bit + 1; | |
813 | return pfn; | |
814 | } | |
815 | } while (rtree_next_node(bm)); | |
816 | ||
817 | return BM_END_OF_MAP; | |
818 | } | |
819 | ||
74dfd666 RW |
820 | /** |
821 | * This structure represents a range of page frames the contents of which | |
822 | * should not be saved during the suspend. | |
823 | */ | |
824 | ||
825 | struct nosave_region { | |
826 | struct list_head list; | |
827 | unsigned long start_pfn; | |
828 | unsigned long end_pfn; | |
829 | }; | |
830 | ||
831 | static LIST_HEAD(nosave_regions); | |
832 | ||
833 | /** | |
834 | * register_nosave_region - register a range of page frames the contents | |
835 | * of which should not be saved during the suspend (to be used in the early | |
836 | * initialization code) | |
837 | */ | |
838 | ||
839 | void __init | |
940d67f6 JB |
840 | __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn, |
841 | int use_kmalloc) | |
74dfd666 RW |
842 | { |
843 | struct nosave_region *region; | |
844 | ||
845 | if (start_pfn >= end_pfn) | |
846 | return; | |
847 | ||
848 | if (!list_empty(&nosave_regions)) { | |
849 | /* Try to extend the previous region (they should be sorted) */ | |
850 | region = list_entry(nosave_regions.prev, | |
851 | struct nosave_region, list); | |
852 | if (region->end_pfn == start_pfn) { | |
853 | region->end_pfn = end_pfn; | |
854 | goto Report; | |
855 | } | |
856 | } | |
940d67f6 JB |
857 | if (use_kmalloc) { |
858 | /* during init, this shouldn't fail */ | |
859 | region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL); | |
860 | BUG_ON(!region); | |
861 | } else | |
862 | /* This allocation cannot fail */ | |
c2f69cda | 863 | region = memblock_virt_alloc(sizeof(struct nosave_region), 0); |
74dfd666 RW |
864 | region->start_pfn = start_pfn; |
865 | region->end_pfn = end_pfn; | |
866 | list_add_tail(®ion->list, &nosave_regions); | |
867 | Report: | |
cd38ca85 BH |
868 | printk(KERN_INFO "PM: Registered nosave memory: [mem %#010llx-%#010llx]\n", |
869 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
870 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
74dfd666 RW |
871 | } |
872 | ||
873 | /* | |
874 | * Set bits in this map correspond to the page frames the contents of which | |
875 | * should not be saved during the suspend. | |
876 | */ | |
877 | static struct memory_bitmap *forbidden_pages_map; | |
878 | ||
879 | /* Set bits in this map correspond to free page frames. */ | |
880 | static struct memory_bitmap *free_pages_map; | |
881 | ||
882 | /* | |
883 | * Each page frame allocated for creating the image is marked by setting the | |
884 | * corresponding bits in forbidden_pages_map and free_pages_map simultaneously | |
885 | */ | |
886 | ||
887 | void swsusp_set_page_free(struct page *page) | |
888 | { | |
889 | if (free_pages_map) | |
890 | memory_bm_set_bit(free_pages_map, page_to_pfn(page)); | |
891 | } | |
892 | ||
893 | static int swsusp_page_is_free(struct page *page) | |
894 | { | |
895 | return free_pages_map ? | |
896 | memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0; | |
897 | } | |
898 | ||
899 | void swsusp_unset_page_free(struct page *page) | |
900 | { | |
901 | if (free_pages_map) | |
902 | memory_bm_clear_bit(free_pages_map, page_to_pfn(page)); | |
903 | } | |
904 | ||
905 | static void swsusp_set_page_forbidden(struct page *page) | |
906 | { | |
907 | if (forbidden_pages_map) | |
908 | memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page)); | |
909 | } | |
910 | ||
911 | int swsusp_page_is_forbidden(struct page *page) | |
912 | { | |
913 | return forbidden_pages_map ? | |
914 | memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0; | |
915 | } | |
916 | ||
917 | static void swsusp_unset_page_forbidden(struct page *page) | |
918 | { | |
919 | if (forbidden_pages_map) | |
920 | memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page)); | |
921 | } | |
922 | ||
923 | /** | |
924 | * mark_nosave_pages - set bits corresponding to the page frames the | |
925 | * contents of which should not be saved in a given bitmap. | |
926 | */ | |
927 | ||
928 | static void mark_nosave_pages(struct memory_bitmap *bm) | |
929 | { | |
930 | struct nosave_region *region; | |
931 | ||
932 | if (list_empty(&nosave_regions)) | |
933 | return; | |
934 | ||
935 | list_for_each_entry(region, &nosave_regions, list) { | |
936 | unsigned long pfn; | |
937 | ||
69f1d475 BH |
938 | pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n", |
939 | (unsigned long long) region->start_pfn << PAGE_SHIFT, | |
940 | ((unsigned long long) region->end_pfn << PAGE_SHIFT) | |
941 | - 1); | |
74dfd666 RW |
942 | |
943 | for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++) | |
a82f7119 RW |
944 | if (pfn_valid(pfn)) { |
945 | /* | |
946 | * It is safe to ignore the result of | |
947 | * mem_bm_set_bit_check() here, since we won't | |
948 | * touch the PFNs for which the error is | |
949 | * returned anyway. | |
950 | */ | |
951 | mem_bm_set_bit_check(bm, pfn); | |
952 | } | |
74dfd666 RW |
953 | } |
954 | } | |
955 | ||
956 | /** | |
957 | * create_basic_memory_bitmaps - create bitmaps needed for marking page | |
958 | * frames that should not be saved and free page frames. The pointers | |
959 | * forbidden_pages_map and free_pages_map are only modified if everything | |
960 | * goes well, because we don't want the bits to be used before both bitmaps | |
961 | * are set up. | |
962 | */ | |
963 | ||
964 | int create_basic_memory_bitmaps(void) | |
965 | { | |
966 | struct memory_bitmap *bm1, *bm2; | |
967 | int error = 0; | |
968 | ||
aab17289 RW |
969 | if (forbidden_pages_map && free_pages_map) |
970 | return 0; | |
971 | else | |
972 | BUG_ON(forbidden_pages_map || free_pages_map); | |
74dfd666 | 973 | |
0709db60 | 974 | bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
975 | if (!bm1) |
976 | return -ENOMEM; | |
977 | ||
0709db60 | 978 | error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
979 | if (error) |
980 | goto Free_first_object; | |
981 | ||
0709db60 | 982 | bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); |
74dfd666 RW |
983 | if (!bm2) |
984 | goto Free_first_bitmap; | |
985 | ||
0709db60 | 986 | error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY); |
74dfd666 RW |
987 | if (error) |
988 | goto Free_second_object; | |
989 | ||
990 | forbidden_pages_map = bm1; | |
991 | free_pages_map = bm2; | |
992 | mark_nosave_pages(forbidden_pages_map); | |
993 | ||
23976728 | 994 | pr_debug("PM: Basic memory bitmaps created\n"); |
74dfd666 RW |
995 | |
996 | return 0; | |
997 | ||
998 | Free_second_object: | |
999 | kfree(bm2); | |
1000 | Free_first_bitmap: | |
1001 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
1002 | Free_first_object: | |
1003 | kfree(bm1); | |
1004 | return -ENOMEM; | |
1005 | } | |
1006 | ||
1007 | /** | |
1008 | * free_basic_memory_bitmaps - free memory bitmaps allocated by | |
1009 | * create_basic_memory_bitmaps(). The auxiliary pointers are necessary | |
1010 | * so that the bitmaps themselves are not referred to while they are being | |
1011 | * freed. | |
1012 | */ | |
1013 | ||
1014 | void free_basic_memory_bitmaps(void) | |
1015 | { | |
1016 | struct memory_bitmap *bm1, *bm2; | |
1017 | ||
6a0c7cd3 RW |
1018 | if (WARN_ON(!(forbidden_pages_map && free_pages_map))) |
1019 | return; | |
74dfd666 RW |
1020 | |
1021 | bm1 = forbidden_pages_map; | |
1022 | bm2 = free_pages_map; | |
1023 | forbidden_pages_map = NULL; | |
1024 | free_pages_map = NULL; | |
1025 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
1026 | kfree(bm1); | |
1027 | memory_bm_free(bm2, PG_UNSAFE_CLEAR); | |
1028 | kfree(bm2); | |
1029 | ||
23976728 | 1030 | pr_debug("PM: Basic memory bitmaps freed\n"); |
74dfd666 RW |
1031 | } |
1032 | ||
b788db79 RW |
1033 | /** |
1034 | * snapshot_additional_pages - estimate the number of additional pages | |
1035 | * be needed for setting up the suspend image data structures for given | |
1036 | * zone (usually the returned value is greater than the exact number) | |
1037 | */ | |
1038 | ||
1039 | unsigned int snapshot_additional_pages(struct zone *zone) | |
1040 | { | |
f469f02d | 1041 | unsigned int rtree, nodes; |
b788db79 | 1042 | |
f469f02d JR |
1043 | rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK); |
1044 | rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node), | |
1045 | LINKED_PAGE_DATA_SIZE); | |
1046 | while (nodes > 1) { | |
1047 | nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL); | |
1048 | rtree += nodes; | |
1049 | } | |
1050 | ||
9047eb62 | 1051 | return 2 * rtree; |
b788db79 RW |
1052 | } |
1053 | ||
8357376d RW |
1054 | #ifdef CONFIG_HIGHMEM |
1055 | /** | |
1056 | * count_free_highmem_pages - compute the total number of free highmem | |
1057 | * pages, system-wide. | |
1058 | */ | |
1059 | ||
1060 | static unsigned int count_free_highmem_pages(void) | |
1061 | { | |
1062 | struct zone *zone; | |
1063 | unsigned int cnt = 0; | |
1064 | ||
ee99c71c KM |
1065 | for_each_populated_zone(zone) |
1066 | if (is_highmem(zone)) | |
d23ad423 | 1067 | cnt += zone_page_state(zone, NR_FREE_PAGES); |
8357376d RW |
1068 | |
1069 | return cnt; | |
1070 | } | |
1071 | ||
1072 | /** | |
1073 | * saveable_highmem_page - Determine whether a highmem page should be | |
1074 | * included in the suspend image. | |
1075 | * | |
1076 | * We should save the page if it isn't Nosave or NosaveFree, or Reserved, | |
1077 | * and it isn't a part of a free chunk of pages. | |
1078 | */ | |
846705de | 1079 | static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn) |
8357376d RW |
1080 | { |
1081 | struct page *page; | |
1082 | ||
1083 | if (!pfn_valid(pfn)) | |
1084 | return NULL; | |
1085 | ||
1086 | page = pfn_to_page(pfn); | |
846705de RW |
1087 | if (page_zone(page) != zone) |
1088 | return NULL; | |
8357376d RW |
1089 | |
1090 | BUG_ON(!PageHighMem(page)); | |
1091 | ||
7be98234 RW |
1092 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) || |
1093 | PageReserved(page)) | |
8357376d RW |
1094 | return NULL; |
1095 | ||
c6968e73 SG |
1096 | if (page_is_guard(page)) |
1097 | return NULL; | |
1098 | ||
8357376d RW |
1099 | return page; |
1100 | } | |
1101 | ||
1102 | /** | |
1103 | * count_highmem_pages - compute the total number of saveable highmem | |
1104 | * pages. | |
1105 | */ | |
1106 | ||
fe419535 | 1107 | static unsigned int count_highmem_pages(void) |
8357376d RW |
1108 | { |
1109 | struct zone *zone; | |
1110 | unsigned int n = 0; | |
1111 | ||
98e73dc5 | 1112 | for_each_populated_zone(zone) { |
8357376d RW |
1113 | unsigned long pfn, max_zone_pfn; |
1114 | ||
1115 | if (!is_highmem(zone)) | |
1116 | continue; | |
1117 | ||
1118 | mark_free_pages(zone); | |
c33bc315 | 1119 | max_zone_pfn = zone_end_pfn(zone); |
8357376d | 1120 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
846705de | 1121 | if (saveable_highmem_page(zone, pfn)) |
8357376d RW |
1122 | n++; |
1123 | } | |
1124 | return n; | |
1125 | } | |
1126 | #else | |
846705de RW |
1127 | static inline void *saveable_highmem_page(struct zone *z, unsigned long p) |
1128 | { | |
1129 | return NULL; | |
1130 | } | |
8357376d RW |
1131 | #endif /* CONFIG_HIGHMEM */ |
1132 | ||
25761b6e | 1133 | /** |
8a235efa RW |
1134 | * saveable_page - Determine whether a non-highmem page should be included |
1135 | * in the suspend image. | |
25761b6e | 1136 | * |
8357376d RW |
1137 | * We should save the page if it isn't Nosave, and is not in the range |
1138 | * of pages statically defined as 'unsaveable', and it isn't a part of | |
1139 | * a free chunk of pages. | |
25761b6e | 1140 | */ |
846705de | 1141 | static struct page *saveable_page(struct zone *zone, unsigned long pfn) |
25761b6e | 1142 | { |
de491861 | 1143 | struct page *page; |
25761b6e RW |
1144 | |
1145 | if (!pfn_valid(pfn)) | |
ae83c5ee | 1146 | return NULL; |
25761b6e RW |
1147 | |
1148 | page = pfn_to_page(pfn); | |
846705de RW |
1149 | if (page_zone(page) != zone) |
1150 | return NULL; | |
ae83c5ee | 1151 | |
8357376d RW |
1152 | BUG_ON(PageHighMem(page)); |
1153 | ||
7be98234 | 1154 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page)) |
ae83c5ee | 1155 | return NULL; |
8357376d | 1156 | |
8a235efa RW |
1157 | if (PageReserved(page) |
1158 | && (!kernel_page_present(page) || pfn_is_nosave(pfn))) | |
ae83c5ee | 1159 | return NULL; |
25761b6e | 1160 | |
c6968e73 SG |
1161 | if (page_is_guard(page)) |
1162 | return NULL; | |
1163 | ||
ae83c5ee | 1164 | return page; |
25761b6e RW |
1165 | } |
1166 | ||
8357376d RW |
1167 | /** |
1168 | * count_data_pages - compute the total number of saveable non-highmem | |
1169 | * pages. | |
1170 | */ | |
1171 | ||
fe419535 | 1172 | static unsigned int count_data_pages(void) |
25761b6e RW |
1173 | { |
1174 | struct zone *zone; | |
ae83c5ee | 1175 | unsigned long pfn, max_zone_pfn; |
dc19d507 | 1176 | unsigned int n = 0; |
25761b6e | 1177 | |
98e73dc5 | 1178 | for_each_populated_zone(zone) { |
25761b6e RW |
1179 | if (is_highmem(zone)) |
1180 | continue; | |
8357376d | 1181 | |
25761b6e | 1182 | mark_free_pages(zone); |
c33bc315 | 1183 | max_zone_pfn = zone_end_pfn(zone); |
ae83c5ee | 1184 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
846705de | 1185 | if (saveable_page(zone, pfn)) |
8357376d | 1186 | n++; |
25761b6e | 1187 | } |
a0f49651 | 1188 | return n; |
25761b6e RW |
1189 | } |
1190 | ||
8357376d RW |
1191 | /* This is needed, because copy_page and memcpy are not usable for copying |
1192 | * task structs. | |
1193 | */ | |
1194 | static inline void do_copy_page(long *dst, long *src) | |
f623f0db RW |
1195 | { |
1196 | int n; | |
1197 | ||
f623f0db RW |
1198 | for (n = PAGE_SIZE / sizeof(long); n; n--) |
1199 | *dst++ = *src++; | |
1200 | } | |
1201 | ||
8a235efa RW |
1202 | |
1203 | /** | |
1204 | * safe_copy_page - check if the page we are going to copy is marked as | |
1205 | * present in the kernel page tables (this always is the case if | |
1206 | * CONFIG_DEBUG_PAGEALLOC is not set and in that case | |
1207 | * kernel_page_present() always returns 'true'). | |
1208 | */ | |
1209 | static void safe_copy_page(void *dst, struct page *s_page) | |
1210 | { | |
1211 | if (kernel_page_present(s_page)) { | |
1212 | do_copy_page(dst, page_address(s_page)); | |
1213 | } else { | |
1214 | kernel_map_pages(s_page, 1, 1); | |
1215 | do_copy_page(dst, page_address(s_page)); | |
1216 | kernel_map_pages(s_page, 1, 0); | |
1217 | } | |
1218 | } | |
1219 | ||
1220 | ||
8357376d RW |
1221 | #ifdef CONFIG_HIGHMEM |
1222 | static inline struct page * | |
1223 | page_is_saveable(struct zone *zone, unsigned long pfn) | |
1224 | { | |
1225 | return is_highmem(zone) ? | |
846705de | 1226 | saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn); |
8357376d RW |
1227 | } |
1228 | ||
8a235efa | 1229 | static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
8357376d RW |
1230 | { |
1231 | struct page *s_page, *d_page; | |
1232 | void *src, *dst; | |
1233 | ||
1234 | s_page = pfn_to_page(src_pfn); | |
1235 | d_page = pfn_to_page(dst_pfn); | |
1236 | if (PageHighMem(s_page)) { | |
0de9a1e2 CW |
1237 | src = kmap_atomic(s_page); |
1238 | dst = kmap_atomic(d_page); | |
8357376d | 1239 | do_copy_page(dst, src); |
0de9a1e2 CW |
1240 | kunmap_atomic(dst); |
1241 | kunmap_atomic(src); | |
8357376d | 1242 | } else { |
8357376d RW |
1243 | if (PageHighMem(d_page)) { |
1244 | /* Page pointed to by src may contain some kernel | |
1245 | * data modified by kmap_atomic() | |
1246 | */ | |
8a235efa | 1247 | safe_copy_page(buffer, s_page); |
0de9a1e2 | 1248 | dst = kmap_atomic(d_page); |
3ecb01df | 1249 | copy_page(dst, buffer); |
0de9a1e2 | 1250 | kunmap_atomic(dst); |
8357376d | 1251 | } else { |
8a235efa | 1252 | safe_copy_page(page_address(d_page), s_page); |
8357376d RW |
1253 | } |
1254 | } | |
1255 | } | |
1256 | #else | |
846705de | 1257 | #define page_is_saveable(zone, pfn) saveable_page(zone, pfn) |
8357376d | 1258 | |
8a235efa | 1259 | static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) |
8357376d | 1260 | { |
8a235efa RW |
1261 | safe_copy_page(page_address(pfn_to_page(dst_pfn)), |
1262 | pfn_to_page(src_pfn)); | |
8357376d RW |
1263 | } |
1264 | #endif /* CONFIG_HIGHMEM */ | |
1265 | ||
b788db79 RW |
1266 | static void |
1267 | copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm) | |
25761b6e RW |
1268 | { |
1269 | struct zone *zone; | |
b788db79 | 1270 | unsigned long pfn; |
25761b6e | 1271 | |
98e73dc5 | 1272 | for_each_populated_zone(zone) { |
b788db79 RW |
1273 | unsigned long max_zone_pfn; |
1274 | ||
25761b6e | 1275 | mark_free_pages(zone); |
c33bc315 | 1276 | max_zone_pfn = zone_end_pfn(zone); |
b788db79 | 1277 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
8357376d | 1278 | if (page_is_saveable(zone, pfn)) |
b788db79 | 1279 | memory_bm_set_bit(orig_bm, pfn); |
25761b6e | 1280 | } |
b788db79 RW |
1281 | memory_bm_position_reset(orig_bm); |
1282 | memory_bm_position_reset(copy_bm); | |
df7c4872 | 1283 | for(;;) { |
b788db79 | 1284 | pfn = memory_bm_next_pfn(orig_bm); |
df7c4872 FW |
1285 | if (unlikely(pfn == BM_END_OF_MAP)) |
1286 | break; | |
1287 | copy_data_page(memory_bm_next_pfn(copy_bm), pfn); | |
1288 | } | |
25761b6e RW |
1289 | } |
1290 | ||
8357376d RW |
1291 | /* Total number of image pages */ |
1292 | static unsigned int nr_copy_pages; | |
1293 | /* Number of pages needed for saving the original pfns of the image pages */ | |
1294 | static unsigned int nr_meta_pages; | |
64a473cb RW |
1295 | /* |
1296 | * Numbers of normal and highmem page frames allocated for hibernation image | |
1297 | * before suspending devices. | |
1298 | */ | |
1299 | unsigned int alloc_normal, alloc_highmem; | |
1300 | /* | |
1301 | * Memory bitmap used for marking saveable pages (during hibernation) or | |
1302 | * hibernation image pages (during restore) | |
1303 | */ | |
1304 | static struct memory_bitmap orig_bm; | |
1305 | /* | |
1306 | * Memory bitmap used during hibernation for marking allocated page frames that | |
1307 | * will contain copies of saveable pages. During restore it is initially used | |
1308 | * for marking hibernation image pages, but then the set bits from it are | |
1309 | * duplicated in @orig_bm and it is released. On highmem systems it is next | |
1310 | * used for marking "safe" highmem pages, but it has to be reinitialized for | |
1311 | * this purpose. | |
1312 | */ | |
1313 | static struct memory_bitmap copy_bm; | |
8357376d | 1314 | |
25761b6e | 1315 | /** |
940864dd | 1316 | * swsusp_free - free pages allocated for the suspend. |
cd560bb2 | 1317 | * |
940864dd RW |
1318 | * Suspend pages are alocated before the atomic copy is made, so we |
1319 | * need to release them after the resume. | |
25761b6e RW |
1320 | */ |
1321 | ||
1322 | void swsusp_free(void) | |
1323 | { | |
6efde38f | 1324 | unsigned long fb_pfn, fr_pfn; |
25761b6e | 1325 | |
6efde38f JR |
1326 | memory_bm_position_reset(forbidden_pages_map); |
1327 | memory_bm_position_reset(free_pages_map); | |
1328 | ||
1329 | loop: | |
1330 | fr_pfn = memory_bm_next_pfn(free_pages_map); | |
1331 | fb_pfn = memory_bm_next_pfn(forbidden_pages_map); | |
1332 | ||
1333 | /* | |
1334 | * Find the next bit set in both bitmaps. This is guaranteed to | |
1335 | * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP. | |
1336 | */ | |
1337 | do { | |
1338 | if (fb_pfn < fr_pfn) | |
1339 | fb_pfn = memory_bm_next_pfn(forbidden_pages_map); | |
1340 | if (fr_pfn < fb_pfn) | |
1341 | fr_pfn = memory_bm_next_pfn(free_pages_map); | |
1342 | } while (fb_pfn != fr_pfn); | |
1343 | ||
1344 | if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) { | |
1345 | struct page *page = pfn_to_page(fr_pfn); | |
1346 | ||
1347 | memory_bm_clear_current(forbidden_pages_map); | |
1348 | memory_bm_clear_current(free_pages_map); | |
1349 | __free_page(page); | |
1350 | goto loop; | |
25761b6e | 1351 | } |
6efde38f | 1352 | |
f577eb30 RW |
1353 | nr_copy_pages = 0; |
1354 | nr_meta_pages = 0; | |
75534b50 | 1355 | restore_pblist = NULL; |
6e1819d6 | 1356 | buffer = NULL; |
64a473cb RW |
1357 | alloc_normal = 0; |
1358 | alloc_highmem = 0; | |
25761b6e RW |
1359 | } |
1360 | ||
4bb33435 RW |
1361 | /* Helper functions used for the shrinking of memory. */ |
1362 | ||
1363 | #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN) | |
1364 | ||
fe419535 | 1365 | /** |
4bb33435 RW |
1366 | * preallocate_image_pages - Allocate a number of pages for hibernation image |
1367 | * @nr_pages: Number of page frames to allocate. | |
1368 | * @mask: GFP flags to use for the allocation. | |
fe419535 | 1369 | * |
4bb33435 RW |
1370 | * Return value: Number of page frames actually allocated |
1371 | */ | |
1372 | static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask) | |
1373 | { | |
1374 | unsigned long nr_alloc = 0; | |
1375 | ||
1376 | while (nr_pages > 0) { | |
64a473cb RW |
1377 | struct page *page; |
1378 | ||
1379 | page = alloc_image_page(mask); | |
1380 | if (!page) | |
4bb33435 | 1381 | break; |
64a473cb RW |
1382 | memory_bm_set_bit(©_bm, page_to_pfn(page)); |
1383 | if (PageHighMem(page)) | |
1384 | alloc_highmem++; | |
1385 | else | |
1386 | alloc_normal++; | |
4bb33435 RW |
1387 | nr_pages--; |
1388 | nr_alloc++; | |
1389 | } | |
1390 | ||
1391 | return nr_alloc; | |
1392 | } | |
1393 | ||
6715045d RW |
1394 | static unsigned long preallocate_image_memory(unsigned long nr_pages, |
1395 | unsigned long avail_normal) | |
4bb33435 | 1396 | { |
6715045d RW |
1397 | unsigned long alloc; |
1398 | ||
1399 | if (avail_normal <= alloc_normal) | |
1400 | return 0; | |
1401 | ||
1402 | alloc = avail_normal - alloc_normal; | |
1403 | if (nr_pages < alloc) | |
1404 | alloc = nr_pages; | |
1405 | ||
1406 | return preallocate_image_pages(alloc, GFP_IMAGE); | |
4bb33435 RW |
1407 | } |
1408 | ||
1409 | #ifdef CONFIG_HIGHMEM | |
1410 | static unsigned long preallocate_image_highmem(unsigned long nr_pages) | |
1411 | { | |
1412 | return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM); | |
1413 | } | |
1414 | ||
1415 | /** | |
1416 | * __fraction - Compute (an approximation of) x * (multiplier / base) | |
fe419535 | 1417 | */ |
4bb33435 RW |
1418 | static unsigned long __fraction(u64 x, u64 multiplier, u64 base) |
1419 | { | |
1420 | x *= multiplier; | |
1421 | do_div(x, base); | |
1422 | return (unsigned long)x; | |
1423 | } | |
fe419535 | 1424 | |
4bb33435 RW |
1425 | static unsigned long preallocate_highmem_fraction(unsigned long nr_pages, |
1426 | unsigned long highmem, | |
1427 | unsigned long total) | |
fe419535 | 1428 | { |
4bb33435 RW |
1429 | unsigned long alloc = __fraction(nr_pages, highmem, total); |
1430 | ||
1431 | return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM); | |
fe419535 | 1432 | } |
4bb33435 RW |
1433 | #else /* CONFIG_HIGHMEM */ |
1434 | static inline unsigned long preallocate_image_highmem(unsigned long nr_pages) | |
1435 | { | |
1436 | return 0; | |
1437 | } | |
1438 | ||
1439 | static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages, | |
1440 | unsigned long highmem, | |
1441 | unsigned long total) | |
1442 | { | |
1443 | return 0; | |
1444 | } | |
1445 | #endif /* CONFIG_HIGHMEM */ | |
fe419535 | 1446 | |
4bb33435 | 1447 | /** |
64a473cb RW |
1448 | * free_unnecessary_pages - Release preallocated pages not needed for the image |
1449 | */ | |
1450 | static void free_unnecessary_pages(void) | |
1451 | { | |
6715045d | 1452 | unsigned long save, to_free_normal, to_free_highmem; |
64a473cb | 1453 | |
6715045d RW |
1454 | save = count_data_pages(); |
1455 | if (alloc_normal >= save) { | |
1456 | to_free_normal = alloc_normal - save; | |
1457 | save = 0; | |
1458 | } else { | |
1459 | to_free_normal = 0; | |
1460 | save -= alloc_normal; | |
1461 | } | |
1462 | save += count_highmem_pages(); | |
1463 | if (alloc_highmem >= save) { | |
1464 | to_free_highmem = alloc_highmem - save; | |
64a473cb RW |
1465 | } else { |
1466 | to_free_highmem = 0; | |
4d4cf23c RW |
1467 | save -= alloc_highmem; |
1468 | if (to_free_normal > save) | |
1469 | to_free_normal -= save; | |
1470 | else | |
1471 | to_free_normal = 0; | |
64a473cb RW |
1472 | } |
1473 | ||
1474 | memory_bm_position_reset(©_bm); | |
1475 | ||
a9c9b442 | 1476 | while (to_free_normal > 0 || to_free_highmem > 0) { |
64a473cb RW |
1477 | unsigned long pfn = memory_bm_next_pfn(©_bm); |
1478 | struct page *page = pfn_to_page(pfn); | |
1479 | ||
1480 | if (PageHighMem(page)) { | |
1481 | if (!to_free_highmem) | |
1482 | continue; | |
1483 | to_free_highmem--; | |
1484 | alloc_highmem--; | |
1485 | } else { | |
1486 | if (!to_free_normal) | |
1487 | continue; | |
1488 | to_free_normal--; | |
1489 | alloc_normal--; | |
1490 | } | |
1491 | memory_bm_clear_bit(©_bm, pfn); | |
1492 | swsusp_unset_page_forbidden(page); | |
1493 | swsusp_unset_page_free(page); | |
1494 | __free_page(page); | |
1495 | } | |
1496 | } | |
1497 | ||
ef4aede3 RW |
1498 | /** |
1499 | * minimum_image_size - Estimate the minimum acceptable size of an image | |
1500 | * @saveable: Number of saveable pages in the system. | |
1501 | * | |
1502 | * We want to avoid attempting to free too much memory too hard, so estimate the | |
1503 | * minimum acceptable size of a hibernation image to use as the lower limit for | |
1504 | * preallocating memory. | |
1505 | * | |
1506 | * We assume that the minimum image size should be proportional to | |
1507 | * | |
1508 | * [number of saveable pages] - [number of pages that can be freed in theory] | |
1509 | * | |
1510 | * where the second term is the sum of (1) reclaimable slab pages, (2) active | |
4d434820 | 1511 | * and (3) inactive anonymous pages, (4) active and (5) inactive file pages, |
ef4aede3 RW |
1512 | * minus mapped file pages. |
1513 | */ | |
1514 | static unsigned long minimum_image_size(unsigned long saveable) | |
1515 | { | |
1516 | unsigned long size; | |
1517 | ||
1518 | size = global_page_state(NR_SLAB_RECLAIMABLE) | |
1519 | + global_page_state(NR_ACTIVE_ANON) | |
1520 | + global_page_state(NR_INACTIVE_ANON) | |
1521 | + global_page_state(NR_ACTIVE_FILE) | |
1522 | + global_page_state(NR_INACTIVE_FILE) | |
1523 | - global_page_state(NR_FILE_MAPPED); | |
1524 | ||
1525 | return saveable <= size ? 0 : saveable - size; | |
1526 | } | |
1527 | ||
64a473cb RW |
1528 | /** |
1529 | * hibernate_preallocate_memory - Preallocate memory for hibernation image | |
4bb33435 RW |
1530 | * |
1531 | * To create a hibernation image it is necessary to make a copy of every page | |
1532 | * frame in use. We also need a number of page frames to be free during | |
1533 | * hibernation for allocations made while saving the image and for device | |
1534 | * drivers, in case they need to allocate memory from their hibernation | |
ddeb6487 RW |
1535 | * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough |
1536 | * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through | |
1537 | * /sys/power/reserved_size, respectively). To make this happen, we compute the | |
1538 | * total number of available page frames and allocate at least | |
4bb33435 | 1539 | * |
ddeb6487 RW |
1540 | * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2 |
1541 | * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE) | |
4bb33435 RW |
1542 | * |
1543 | * of them, which corresponds to the maximum size of a hibernation image. | |
1544 | * | |
1545 | * If image_size is set below the number following from the above formula, | |
1546 | * the preallocation of memory is continued until the total number of saveable | |
ef4aede3 RW |
1547 | * pages in the system is below the requested image size or the minimum |
1548 | * acceptable image size returned by minimum_image_size(), whichever is greater. | |
4bb33435 | 1549 | */ |
64a473cb | 1550 | int hibernate_preallocate_memory(void) |
fe419535 | 1551 | { |
fe419535 | 1552 | struct zone *zone; |
4bb33435 | 1553 | unsigned long saveable, size, max_size, count, highmem, pages = 0; |
6715045d | 1554 | unsigned long alloc, save_highmem, pages_highmem, avail_normal; |
fe419535 | 1555 | struct timeval start, stop; |
64a473cb | 1556 | int error; |
fe419535 | 1557 | |
64a473cb | 1558 | printk(KERN_INFO "PM: Preallocating image memory... "); |
fe419535 | 1559 | do_gettimeofday(&start); |
fe419535 | 1560 | |
64a473cb RW |
1561 | error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY); |
1562 | if (error) | |
1563 | goto err_out; | |
1564 | ||
1565 | error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY); | |
1566 | if (error) | |
1567 | goto err_out; | |
1568 | ||
1569 | alloc_normal = 0; | |
1570 | alloc_highmem = 0; | |
1571 | ||
4bb33435 | 1572 | /* Count the number of saveable data pages. */ |
64a473cb | 1573 | save_highmem = count_highmem_pages(); |
4bb33435 | 1574 | saveable = count_data_pages(); |
fe419535 | 1575 | |
4bb33435 RW |
1576 | /* |
1577 | * Compute the total number of page frames we can use (count) and the | |
1578 | * number of pages needed for image metadata (size). | |
1579 | */ | |
1580 | count = saveable; | |
64a473cb RW |
1581 | saveable += save_highmem; |
1582 | highmem = save_highmem; | |
4bb33435 RW |
1583 | size = 0; |
1584 | for_each_populated_zone(zone) { | |
1585 | size += snapshot_additional_pages(zone); | |
1586 | if (is_highmem(zone)) | |
1587 | highmem += zone_page_state(zone, NR_FREE_PAGES); | |
1588 | else | |
1589 | count += zone_page_state(zone, NR_FREE_PAGES); | |
1590 | } | |
6715045d | 1591 | avail_normal = count; |
4bb33435 RW |
1592 | count += highmem; |
1593 | count -= totalreserve_pages; | |
1594 | ||
85055dd8 MS |
1595 | /* Add number of pages required for page keys (s390 only). */ |
1596 | size += page_key_additional_pages(saveable); | |
1597 | ||
4bb33435 | 1598 | /* Compute the maximum number of saveable pages to leave in memory. */ |
ddeb6487 RW |
1599 | max_size = (count - (size + PAGES_FOR_IO)) / 2 |
1600 | - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE); | |
266f1a25 | 1601 | /* Compute the desired number of image pages specified by image_size. */ |
4bb33435 RW |
1602 | size = DIV_ROUND_UP(image_size, PAGE_SIZE); |
1603 | if (size > max_size) | |
1604 | size = max_size; | |
1605 | /* | |
266f1a25 RW |
1606 | * If the desired number of image pages is at least as large as the |
1607 | * current number of saveable pages in memory, allocate page frames for | |
1608 | * the image and we're done. | |
4bb33435 | 1609 | */ |
64a473cb RW |
1610 | if (size >= saveable) { |
1611 | pages = preallocate_image_highmem(save_highmem); | |
6715045d | 1612 | pages += preallocate_image_memory(saveable - pages, avail_normal); |
4bb33435 | 1613 | goto out; |
64a473cb | 1614 | } |
4bb33435 | 1615 | |
ef4aede3 RW |
1616 | /* Estimate the minimum size of the image. */ |
1617 | pages = minimum_image_size(saveable); | |
6715045d RW |
1618 | /* |
1619 | * To avoid excessive pressure on the normal zone, leave room in it to | |
1620 | * accommodate an image of the minimum size (unless it's already too | |
1621 | * small, in which case don't preallocate pages from it at all). | |
1622 | */ | |
1623 | if (avail_normal > pages) | |
1624 | avail_normal -= pages; | |
1625 | else | |
1626 | avail_normal = 0; | |
ef4aede3 RW |
1627 | if (size < pages) |
1628 | size = min_t(unsigned long, pages, max_size); | |
1629 | ||
4bb33435 RW |
1630 | /* |
1631 | * Let the memory management subsystem know that we're going to need a | |
1632 | * large number of page frames to allocate and make it free some memory. | |
1633 | * NOTE: If this is not done, performance will be hurt badly in some | |
1634 | * test cases. | |
1635 | */ | |
1636 | shrink_all_memory(saveable - size); | |
1637 | ||
1638 | /* | |
1639 | * The number of saveable pages in memory was too high, so apply some | |
1640 | * pressure to decrease it. First, make room for the largest possible | |
1641 | * image and fail if that doesn't work. Next, try to decrease the size | |
ef4aede3 RW |
1642 | * of the image as much as indicated by 'size' using allocations from |
1643 | * highmem and non-highmem zones separately. | |
4bb33435 RW |
1644 | */ |
1645 | pages_highmem = preallocate_image_highmem(highmem / 2); | |
fd432b9f AL |
1646 | alloc = count - max_size; |
1647 | if (alloc > pages_highmem) | |
1648 | alloc -= pages_highmem; | |
1649 | else | |
1650 | alloc = 0; | |
6715045d RW |
1651 | pages = preallocate_image_memory(alloc, avail_normal); |
1652 | if (pages < alloc) { | |
1653 | /* We have exhausted non-highmem pages, try highmem. */ | |
1654 | alloc -= pages; | |
1655 | pages += pages_highmem; | |
1656 | pages_highmem = preallocate_image_highmem(alloc); | |
1657 | if (pages_highmem < alloc) | |
1658 | goto err_out; | |
1659 | pages += pages_highmem; | |
1660 | /* | |
1661 | * size is the desired number of saveable pages to leave in | |
1662 | * memory, so try to preallocate (all memory - size) pages. | |
1663 | */ | |
1664 | alloc = (count - pages) - size; | |
1665 | pages += preallocate_image_highmem(alloc); | |
1666 | } else { | |
1667 | /* | |
1668 | * There are approximately max_size saveable pages at this point | |
1669 | * and we want to reduce this number down to size. | |
1670 | */ | |
1671 | alloc = max_size - size; | |
1672 | size = preallocate_highmem_fraction(alloc, highmem, count); | |
1673 | pages_highmem += size; | |
1674 | alloc -= size; | |
1675 | size = preallocate_image_memory(alloc, avail_normal); | |
1676 | pages_highmem += preallocate_image_highmem(alloc - size); | |
1677 | pages += pages_highmem + size; | |
1678 | } | |
4bb33435 | 1679 | |
64a473cb RW |
1680 | /* |
1681 | * We only need as many page frames for the image as there are saveable | |
1682 | * pages in memory, but we have allocated more. Release the excessive | |
1683 | * ones now. | |
1684 | */ | |
1685 | free_unnecessary_pages(); | |
4bb33435 RW |
1686 | |
1687 | out: | |
fe419535 | 1688 | do_gettimeofday(&stop); |
64a473cb RW |
1689 | printk(KERN_CONT "done (allocated %lu pages)\n", pages); |
1690 | swsusp_show_speed(&start, &stop, pages, "Allocated"); | |
fe419535 RW |
1691 | |
1692 | return 0; | |
64a473cb RW |
1693 | |
1694 | err_out: | |
1695 | printk(KERN_CONT "\n"); | |
1696 | swsusp_free(); | |
1697 | return -ENOMEM; | |
fe419535 RW |
1698 | } |
1699 | ||
8357376d RW |
1700 | #ifdef CONFIG_HIGHMEM |
1701 | /** | |
1702 | * count_pages_for_highmem - compute the number of non-highmem pages | |
1703 | * that will be necessary for creating copies of highmem pages. | |
1704 | */ | |
1705 | ||
1706 | static unsigned int count_pages_for_highmem(unsigned int nr_highmem) | |
1707 | { | |
64a473cb | 1708 | unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem; |
8357376d RW |
1709 | |
1710 | if (free_highmem >= nr_highmem) | |
1711 | nr_highmem = 0; | |
1712 | else | |
1713 | nr_highmem -= free_highmem; | |
1714 | ||
1715 | return nr_highmem; | |
1716 | } | |
1717 | #else | |
1718 | static unsigned int | |
1719 | count_pages_for_highmem(unsigned int nr_highmem) { return 0; } | |
1720 | #endif /* CONFIG_HIGHMEM */ | |
25761b6e RW |
1721 | |
1722 | /** | |
8357376d RW |
1723 | * enough_free_mem - Make sure we have enough free memory for the |
1724 | * snapshot image. | |
25761b6e RW |
1725 | */ |
1726 | ||
8357376d | 1727 | static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem) |
25761b6e | 1728 | { |
e5e2fa78 | 1729 | struct zone *zone; |
64a473cb | 1730 | unsigned int free = alloc_normal; |
e5e2fa78 | 1731 | |
98e73dc5 | 1732 | for_each_populated_zone(zone) |
8357376d | 1733 | if (!is_highmem(zone)) |
d23ad423 | 1734 | free += zone_page_state(zone, NR_FREE_PAGES); |
940864dd | 1735 | |
8357376d | 1736 | nr_pages += count_pages_for_highmem(nr_highmem); |
64a473cb RW |
1737 | pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n", |
1738 | nr_pages, PAGES_FOR_IO, free); | |
940864dd | 1739 | |
64a473cb | 1740 | return free > nr_pages + PAGES_FOR_IO; |
25761b6e RW |
1741 | } |
1742 | ||
8357376d RW |
1743 | #ifdef CONFIG_HIGHMEM |
1744 | /** | |
1745 | * get_highmem_buffer - if there are some highmem pages in the suspend | |
1746 | * image, we may need the buffer to copy them and/or load their data. | |
1747 | */ | |
1748 | ||
1749 | static inline int get_highmem_buffer(int safe_needed) | |
1750 | { | |
1751 | buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed); | |
1752 | return buffer ? 0 : -ENOMEM; | |
1753 | } | |
1754 | ||
1755 | /** | |
1756 | * alloc_highmem_image_pages - allocate some highmem pages for the image. | |
1757 | * Try to allocate as many pages as needed, but if the number of free | |
1758 | * highmem pages is lesser than that, allocate them all. | |
1759 | */ | |
1760 | ||
1761 | static inline unsigned int | |
64a473cb | 1762 | alloc_highmem_pages(struct memory_bitmap *bm, unsigned int nr_highmem) |
8357376d RW |
1763 | { |
1764 | unsigned int to_alloc = count_free_highmem_pages(); | |
1765 | ||
1766 | if (to_alloc > nr_highmem) | |
1767 | to_alloc = nr_highmem; | |
1768 | ||
1769 | nr_highmem -= to_alloc; | |
1770 | while (to_alloc-- > 0) { | |
1771 | struct page *page; | |
1772 | ||
1773 | page = alloc_image_page(__GFP_HIGHMEM); | |
1774 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1775 | } | |
1776 | return nr_highmem; | |
1777 | } | |
1778 | #else | |
1779 | static inline int get_highmem_buffer(int safe_needed) { return 0; } | |
1780 | ||
1781 | static inline unsigned int | |
64a473cb | 1782 | alloc_highmem_pages(struct memory_bitmap *bm, unsigned int n) { return 0; } |
8357376d RW |
1783 | #endif /* CONFIG_HIGHMEM */ |
1784 | ||
1785 | /** | |
1786 | * swsusp_alloc - allocate memory for the suspend image | |
1787 | * | |
1788 | * We first try to allocate as many highmem pages as there are | |
1789 | * saveable highmem pages in the system. If that fails, we allocate | |
1790 | * non-highmem pages for the copies of the remaining highmem ones. | |
1791 | * | |
1792 | * In this approach it is likely that the copies of highmem pages will | |
1793 | * also be located in the high memory, because of the way in which | |
1794 | * copy_data_pages() works. | |
1795 | */ | |
1796 | ||
b788db79 RW |
1797 | static int |
1798 | swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm, | |
8357376d | 1799 | unsigned int nr_pages, unsigned int nr_highmem) |
054bd4c1 | 1800 | { |
8357376d | 1801 | if (nr_highmem > 0) { |
2e725a06 | 1802 | if (get_highmem_buffer(PG_ANY)) |
64a473cb RW |
1803 | goto err_out; |
1804 | if (nr_highmem > alloc_highmem) { | |
1805 | nr_highmem -= alloc_highmem; | |
1806 | nr_pages += alloc_highmem_pages(copy_bm, nr_highmem); | |
1807 | } | |
8357376d | 1808 | } |
64a473cb RW |
1809 | if (nr_pages > alloc_normal) { |
1810 | nr_pages -= alloc_normal; | |
1811 | while (nr_pages-- > 0) { | |
1812 | struct page *page; | |
1813 | ||
1814 | page = alloc_image_page(GFP_ATOMIC | __GFP_COLD); | |
1815 | if (!page) | |
1816 | goto err_out; | |
1817 | memory_bm_set_bit(copy_bm, page_to_pfn(page)); | |
1818 | } | |
25761b6e | 1819 | } |
64a473cb | 1820 | |
b788db79 | 1821 | return 0; |
25761b6e | 1822 | |
64a473cb | 1823 | err_out: |
b788db79 | 1824 | swsusp_free(); |
2e725a06 | 1825 | return -ENOMEM; |
25761b6e RW |
1826 | } |
1827 | ||
722a9f92 | 1828 | asmlinkage __visible int swsusp_save(void) |
25761b6e | 1829 | { |
8357376d | 1830 | unsigned int nr_pages, nr_highmem; |
25761b6e | 1831 | |
07c3bb57 | 1832 | printk(KERN_INFO "PM: Creating hibernation image:\n"); |
25761b6e | 1833 | |
9f8f2172 | 1834 | drain_local_pages(NULL); |
a0f49651 | 1835 | nr_pages = count_data_pages(); |
8357376d | 1836 | nr_highmem = count_highmem_pages(); |
23976728 | 1837 | printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem); |
25761b6e | 1838 | |
8357376d | 1839 | if (!enough_free_mem(nr_pages, nr_highmem)) { |
23976728 | 1840 | printk(KERN_ERR "PM: Not enough free memory\n"); |
25761b6e RW |
1841 | return -ENOMEM; |
1842 | } | |
1843 | ||
8357376d | 1844 | if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) { |
23976728 | 1845 | printk(KERN_ERR "PM: Memory allocation failed\n"); |
a0f49651 | 1846 | return -ENOMEM; |
8357376d | 1847 | } |
25761b6e RW |
1848 | |
1849 | /* During allocating of suspend pagedir, new cold pages may appear. | |
1850 | * Kill them. | |
1851 | */ | |
9f8f2172 | 1852 | drain_local_pages(NULL); |
b788db79 | 1853 | copy_data_pages(©_bm, &orig_bm); |
25761b6e RW |
1854 | |
1855 | /* | |
1856 | * End of critical section. From now on, we can write to memory, | |
1857 | * but we should not touch disk. This specially means we must _not_ | |
1858 | * touch swap space! Except we must write out our image of course. | |
1859 | */ | |
1860 | ||
8357376d | 1861 | nr_pages += nr_highmem; |
a0f49651 | 1862 | nr_copy_pages = nr_pages; |
8357376d | 1863 | nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE); |
a0f49651 | 1864 | |
23976728 RW |
1865 | printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n", |
1866 | nr_pages); | |
8357376d | 1867 | |
25761b6e RW |
1868 | return 0; |
1869 | } | |
f577eb30 | 1870 | |
d307c4a8 RW |
1871 | #ifndef CONFIG_ARCH_HIBERNATION_HEADER |
1872 | static int init_header_complete(struct swsusp_info *info) | |
f577eb30 | 1873 | { |
d307c4a8 | 1874 | memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname)); |
f577eb30 | 1875 | info->version_code = LINUX_VERSION_CODE; |
d307c4a8 RW |
1876 | return 0; |
1877 | } | |
1878 | ||
1879 | static char *check_image_kernel(struct swsusp_info *info) | |
1880 | { | |
1881 | if (info->version_code != LINUX_VERSION_CODE) | |
1882 | return "kernel version"; | |
1883 | if (strcmp(info->uts.sysname,init_utsname()->sysname)) | |
1884 | return "system type"; | |
1885 | if (strcmp(info->uts.release,init_utsname()->release)) | |
1886 | return "kernel release"; | |
1887 | if (strcmp(info->uts.version,init_utsname()->version)) | |
1888 | return "version"; | |
1889 | if (strcmp(info->uts.machine,init_utsname()->machine)) | |
1890 | return "machine"; | |
1891 | return NULL; | |
1892 | } | |
1893 | #endif /* CONFIG_ARCH_HIBERNATION_HEADER */ | |
1894 | ||
af508b34 RW |
1895 | unsigned long snapshot_get_image_size(void) |
1896 | { | |
1897 | return nr_copy_pages + nr_meta_pages + 1; | |
1898 | } | |
1899 | ||
d307c4a8 RW |
1900 | static int init_header(struct swsusp_info *info) |
1901 | { | |
1902 | memset(info, 0, sizeof(struct swsusp_info)); | |
0ed5fd13 | 1903 | info->num_physpages = get_num_physpages(); |
f577eb30 | 1904 | info->image_pages = nr_copy_pages; |
af508b34 | 1905 | info->pages = snapshot_get_image_size(); |
6e1819d6 RW |
1906 | info->size = info->pages; |
1907 | info->size <<= PAGE_SHIFT; | |
d307c4a8 | 1908 | return init_header_complete(info); |
f577eb30 RW |
1909 | } |
1910 | ||
1911 | /** | |
940864dd RW |
1912 | * pack_pfns - pfns corresponding to the set bits found in the bitmap @bm |
1913 | * are stored in the array @buf[] (1 page at a time) | |
f577eb30 RW |
1914 | */ |
1915 | ||
b788db79 | 1916 | static inline void |
940864dd | 1917 | pack_pfns(unsigned long *buf, struct memory_bitmap *bm) |
f577eb30 RW |
1918 | { |
1919 | int j; | |
1920 | ||
b788db79 | 1921 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
940864dd RW |
1922 | buf[j] = memory_bm_next_pfn(bm); |
1923 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
b788db79 | 1924 | break; |
85055dd8 MS |
1925 | /* Save page key for data page (s390 only). */ |
1926 | page_key_read(buf + j); | |
f577eb30 | 1927 | } |
f577eb30 RW |
1928 | } |
1929 | ||
1930 | /** | |
1931 | * snapshot_read_next - used for reading the system memory snapshot. | |
1932 | * | |
1933 | * On the first call to it @handle should point to a zeroed | |
1934 | * snapshot_handle structure. The structure gets updated and a pointer | |
1935 | * to it should be passed to this function every next time. | |
1936 | * | |
f577eb30 RW |
1937 | * On success the function returns a positive number. Then, the caller |
1938 | * is allowed to read up to the returned number of bytes from the memory | |
d3c1b24c | 1939 | * location computed by the data_of() macro. |
f577eb30 RW |
1940 | * |
1941 | * The function returns 0 to indicate the end of data stream condition, | |
1942 | * and a negative number is returned on error. In such cases the | |
1943 | * structure pointed to by @handle is not updated and should not be used | |
1944 | * any more. | |
1945 | */ | |
1946 | ||
d3c1b24c | 1947 | int snapshot_read_next(struct snapshot_handle *handle) |
f577eb30 | 1948 | { |
fb13a28b | 1949 | if (handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 1950 | return 0; |
b788db79 | 1951 | |
f577eb30 RW |
1952 | if (!buffer) { |
1953 | /* This makes the buffer be freed by swsusp_free() */ | |
8357376d | 1954 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); |
f577eb30 RW |
1955 | if (!buffer) |
1956 | return -ENOMEM; | |
1957 | } | |
d3c1b24c | 1958 | if (!handle->cur) { |
d307c4a8 RW |
1959 | int error; |
1960 | ||
1961 | error = init_header((struct swsusp_info *)buffer); | |
1962 | if (error) | |
1963 | return error; | |
f577eb30 | 1964 | handle->buffer = buffer; |
b788db79 RW |
1965 | memory_bm_position_reset(&orig_bm); |
1966 | memory_bm_position_reset(©_bm); | |
d3c1b24c | 1967 | } else if (handle->cur <= nr_meta_pages) { |
3ecb01df | 1968 | clear_page(buffer); |
d3c1b24c JS |
1969 | pack_pfns(buffer, &orig_bm); |
1970 | } else { | |
1971 | struct page *page; | |
b788db79 | 1972 | |
d3c1b24c JS |
1973 | page = pfn_to_page(memory_bm_next_pfn(©_bm)); |
1974 | if (PageHighMem(page)) { | |
1975 | /* Highmem pages are copied to the buffer, | |
1976 | * because we can't return with a kmapped | |
1977 | * highmem page (we may not be called again). | |
1978 | */ | |
1979 | void *kaddr; | |
8357376d | 1980 | |
0de9a1e2 | 1981 | kaddr = kmap_atomic(page); |
3ecb01df | 1982 | copy_page(buffer, kaddr); |
0de9a1e2 | 1983 | kunmap_atomic(kaddr); |
d3c1b24c JS |
1984 | handle->buffer = buffer; |
1985 | } else { | |
1986 | handle->buffer = page_address(page); | |
f577eb30 | 1987 | } |
f577eb30 | 1988 | } |
d3c1b24c JS |
1989 | handle->cur++; |
1990 | return PAGE_SIZE; | |
f577eb30 RW |
1991 | } |
1992 | ||
1993 | /** | |
1994 | * mark_unsafe_pages - mark the pages that cannot be used for storing | |
1995 | * the image during resume, because they conflict with the pages that | |
1996 | * had been used before suspend | |
1997 | */ | |
1998 | ||
940864dd | 1999 | static int mark_unsafe_pages(struct memory_bitmap *bm) |
f577eb30 RW |
2000 | { |
2001 | struct zone *zone; | |
ae83c5ee | 2002 | unsigned long pfn, max_zone_pfn; |
f577eb30 RW |
2003 | |
2004 | /* Clear page flags */ | |
98e73dc5 | 2005 | for_each_populated_zone(zone) { |
c33bc315 | 2006 | max_zone_pfn = zone_end_pfn(zone); |
ae83c5ee RW |
2007 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
2008 | if (pfn_valid(pfn)) | |
7be98234 | 2009 | swsusp_unset_page_free(pfn_to_page(pfn)); |
f577eb30 RW |
2010 | } |
2011 | ||
940864dd RW |
2012 | /* Mark pages that correspond to the "original" pfns as "unsafe" */ |
2013 | memory_bm_position_reset(bm); | |
2014 | do { | |
2015 | pfn = memory_bm_next_pfn(bm); | |
2016 | if (likely(pfn != BM_END_OF_MAP)) { | |
2017 | if (likely(pfn_valid(pfn))) | |
7be98234 | 2018 | swsusp_set_page_free(pfn_to_page(pfn)); |
940864dd RW |
2019 | else |
2020 | return -EFAULT; | |
2021 | } | |
2022 | } while (pfn != BM_END_OF_MAP); | |
f577eb30 | 2023 | |
940864dd | 2024 | allocated_unsafe_pages = 0; |
968808b8 | 2025 | |
f577eb30 RW |
2026 | return 0; |
2027 | } | |
2028 | ||
940864dd RW |
2029 | static void |
2030 | duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src) | |
f577eb30 | 2031 | { |
940864dd RW |
2032 | unsigned long pfn; |
2033 | ||
2034 | memory_bm_position_reset(src); | |
2035 | pfn = memory_bm_next_pfn(src); | |
2036 | while (pfn != BM_END_OF_MAP) { | |
2037 | memory_bm_set_bit(dst, pfn); | |
2038 | pfn = memory_bm_next_pfn(src); | |
f577eb30 RW |
2039 | } |
2040 | } | |
2041 | ||
d307c4a8 | 2042 | static int check_header(struct swsusp_info *info) |
f577eb30 | 2043 | { |
d307c4a8 | 2044 | char *reason; |
f577eb30 | 2045 | |
d307c4a8 | 2046 | reason = check_image_kernel(info); |
0ed5fd13 | 2047 | if (!reason && info->num_physpages != get_num_physpages()) |
f577eb30 | 2048 | reason = "memory size"; |
f577eb30 | 2049 | if (reason) { |
23976728 | 2050 | printk(KERN_ERR "PM: Image mismatch: %s\n", reason); |
f577eb30 RW |
2051 | return -EPERM; |
2052 | } | |
2053 | return 0; | |
2054 | } | |
2055 | ||
2056 | /** | |
2057 | * load header - check the image header and copy data from it | |
2058 | */ | |
2059 | ||
940864dd RW |
2060 | static int |
2061 | load_header(struct swsusp_info *info) | |
f577eb30 RW |
2062 | { |
2063 | int error; | |
f577eb30 | 2064 | |
940864dd | 2065 | restore_pblist = NULL; |
f577eb30 RW |
2066 | error = check_header(info); |
2067 | if (!error) { | |
f577eb30 RW |
2068 | nr_copy_pages = info->image_pages; |
2069 | nr_meta_pages = info->pages - info->image_pages - 1; | |
2070 | } | |
2071 | return error; | |
2072 | } | |
2073 | ||
2074 | /** | |
940864dd RW |
2075 | * unpack_orig_pfns - for each element of @buf[] (1 page at a time) set |
2076 | * the corresponding bit in the memory bitmap @bm | |
f577eb30 | 2077 | */ |
69643279 | 2078 | static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm) |
f577eb30 RW |
2079 | { |
2080 | int j; | |
2081 | ||
940864dd RW |
2082 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { |
2083 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
2084 | break; | |
2085 | ||
85055dd8 MS |
2086 | /* Extract and buffer page key for data page (s390 only). */ |
2087 | page_key_memorize(buf + j); | |
2088 | ||
69643279 RW |
2089 | if (memory_bm_pfn_present(bm, buf[j])) |
2090 | memory_bm_set_bit(bm, buf[j]); | |
2091 | else | |
2092 | return -EFAULT; | |
f577eb30 | 2093 | } |
69643279 RW |
2094 | |
2095 | return 0; | |
f577eb30 RW |
2096 | } |
2097 | ||
8357376d RW |
2098 | /* List of "safe" pages that may be used to store data loaded from the suspend |
2099 | * image | |
2100 | */ | |
2101 | static struct linked_page *safe_pages_list; | |
2102 | ||
2103 | #ifdef CONFIG_HIGHMEM | |
2104 | /* struct highmem_pbe is used for creating the list of highmem pages that | |
2105 | * should be restored atomically during the resume from disk, because the page | |
2106 | * frames they have occupied before the suspend are in use. | |
2107 | */ | |
2108 | struct highmem_pbe { | |
2109 | struct page *copy_page; /* data is here now */ | |
2110 | struct page *orig_page; /* data was here before the suspend */ | |
2111 | struct highmem_pbe *next; | |
2112 | }; | |
2113 | ||
2114 | /* List of highmem PBEs needed for restoring the highmem pages that were | |
2115 | * allocated before the suspend and included in the suspend image, but have | |
2116 | * also been allocated by the "resume" kernel, so their contents cannot be | |
2117 | * written directly to their "original" page frames. | |
2118 | */ | |
2119 | static struct highmem_pbe *highmem_pblist; | |
2120 | ||
2121 | /** | |
2122 | * count_highmem_image_pages - compute the number of highmem pages in the | |
2123 | * suspend image. The bits in the memory bitmap @bm that correspond to the | |
2124 | * image pages are assumed to be set. | |
2125 | */ | |
2126 | ||
2127 | static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) | |
2128 | { | |
2129 | unsigned long pfn; | |
2130 | unsigned int cnt = 0; | |
2131 | ||
2132 | memory_bm_position_reset(bm); | |
2133 | pfn = memory_bm_next_pfn(bm); | |
2134 | while (pfn != BM_END_OF_MAP) { | |
2135 | if (PageHighMem(pfn_to_page(pfn))) | |
2136 | cnt++; | |
2137 | ||
2138 | pfn = memory_bm_next_pfn(bm); | |
2139 | } | |
2140 | return cnt; | |
2141 | } | |
2142 | ||
2143 | /** | |
2144 | * prepare_highmem_image - try to allocate as many highmem pages as | |
2145 | * there are highmem image pages (@nr_highmem_p points to the variable | |
2146 | * containing the number of highmem image pages). The pages that are | |
2147 | * "safe" (ie. will not be overwritten when the suspend image is | |
2148 | * restored) have the corresponding bits set in @bm (it must be | |
2149 | * unitialized). | |
2150 | * | |
2151 | * NOTE: This function should not be called if there are no highmem | |
2152 | * image pages. | |
2153 | */ | |
2154 | ||
2155 | static unsigned int safe_highmem_pages; | |
2156 | ||
2157 | static struct memory_bitmap *safe_highmem_bm; | |
2158 | ||
2159 | static int | |
2160 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
2161 | { | |
2162 | unsigned int to_alloc; | |
2163 | ||
2164 | if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE)) | |
2165 | return -ENOMEM; | |
2166 | ||
2167 | if (get_highmem_buffer(PG_SAFE)) | |
2168 | return -ENOMEM; | |
2169 | ||
2170 | to_alloc = count_free_highmem_pages(); | |
2171 | if (to_alloc > *nr_highmem_p) | |
2172 | to_alloc = *nr_highmem_p; | |
2173 | else | |
2174 | *nr_highmem_p = to_alloc; | |
2175 | ||
2176 | safe_highmem_pages = 0; | |
2177 | while (to_alloc-- > 0) { | |
2178 | struct page *page; | |
2179 | ||
2180 | page = alloc_page(__GFP_HIGHMEM); | |
7be98234 | 2181 | if (!swsusp_page_is_free(page)) { |
8357376d RW |
2182 | /* The page is "safe", set its bit the bitmap */ |
2183 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
2184 | safe_highmem_pages++; | |
2185 | } | |
2186 | /* Mark the page as allocated */ | |
7be98234 RW |
2187 | swsusp_set_page_forbidden(page); |
2188 | swsusp_set_page_free(page); | |
8357376d RW |
2189 | } |
2190 | memory_bm_position_reset(bm); | |
2191 | safe_highmem_bm = bm; | |
2192 | return 0; | |
2193 | } | |
2194 | ||
2195 | /** | |
2196 | * get_highmem_page_buffer - for given highmem image page find the buffer | |
2197 | * that suspend_write_next() should set for its caller to write to. | |
2198 | * | |
2199 | * If the page is to be saved to its "original" page frame or a copy of | |
2200 | * the page is to be made in the highmem, @buffer is returned. Otherwise, | |
2201 | * the copy of the page is to be made in normal memory, so the address of | |
2202 | * the copy is returned. | |
2203 | * | |
2204 | * If @buffer is returned, the caller of suspend_write_next() will write | |
2205 | * the page's contents to @buffer, so they will have to be copied to the | |
2206 | * right location on the next call to suspend_write_next() and it is done | |
2207 | * with the help of copy_last_highmem_page(). For this purpose, if | |
2208 | * @buffer is returned, @last_highmem page is set to the page to which | |
2209 | * the data will have to be copied from @buffer. | |
2210 | */ | |
2211 | ||
2212 | static struct page *last_highmem_page; | |
2213 | ||
2214 | static void * | |
2215 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
2216 | { | |
2217 | struct highmem_pbe *pbe; | |
2218 | void *kaddr; | |
2219 | ||
7be98234 | 2220 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) { |
8357376d RW |
2221 | /* We have allocated the "original" page frame and we can |
2222 | * use it directly to store the loaded page. | |
2223 | */ | |
2224 | last_highmem_page = page; | |
2225 | return buffer; | |
2226 | } | |
2227 | /* The "original" page frame has not been allocated and we have to | |
2228 | * use a "safe" page frame to store the loaded page. | |
2229 | */ | |
2230 | pbe = chain_alloc(ca, sizeof(struct highmem_pbe)); | |
2231 | if (!pbe) { | |
2232 | swsusp_free(); | |
69643279 | 2233 | return ERR_PTR(-ENOMEM); |
8357376d RW |
2234 | } |
2235 | pbe->orig_page = page; | |
2236 | if (safe_highmem_pages > 0) { | |
2237 | struct page *tmp; | |
2238 | ||
2239 | /* Copy of the page will be stored in high memory */ | |
2240 | kaddr = buffer; | |
2241 | tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm)); | |
2242 | safe_highmem_pages--; | |
2243 | last_highmem_page = tmp; | |
2244 | pbe->copy_page = tmp; | |
2245 | } else { | |
2246 | /* Copy of the page will be stored in normal memory */ | |
2247 | kaddr = safe_pages_list; | |
2248 | safe_pages_list = safe_pages_list->next; | |
2249 | pbe->copy_page = virt_to_page(kaddr); | |
2250 | } | |
2251 | pbe->next = highmem_pblist; | |
2252 | highmem_pblist = pbe; | |
2253 | return kaddr; | |
2254 | } | |
2255 | ||
2256 | /** | |
2257 | * copy_last_highmem_page - copy the contents of a highmem image from | |
2258 | * @buffer, where the caller of snapshot_write_next() has place them, | |
2259 | * to the right location represented by @last_highmem_page . | |
2260 | */ | |
2261 | ||
2262 | static void copy_last_highmem_page(void) | |
2263 | { | |
2264 | if (last_highmem_page) { | |
2265 | void *dst; | |
2266 | ||
0de9a1e2 | 2267 | dst = kmap_atomic(last_highmem_page); |
3ecb01df | 2268 | copy_page(dst, buffer); |
0de9a1e2 | 2269 | kunmap_atomic(dst); |
8357376d RW |
2270 | last_highmem_page = NULL; |
2271 | } | |
2272 | } | |
2273 | ||
2274 | static inline int last_highmem_page_copied(void) | |
2275 | { | |
2276 | return !last_highmem_page; | |
2277 | } | |
2278 | ||
2279 | static inline void free_highmem_data(void) | |
2280 | { | |
2281 | if (safe_highmem_bm) | |
2282 | memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR); | |
2283 | ||
2284 | if (buffer) | |
2285 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
2286 | } | |
2287 | #else | |
2288 | static inline int get_safe_write_buffer(void) { return 0; } | |
2289 | ||
2290 | static unsigned int | |
2291 | count_highmem_image_pages(struct memory_bitmap *bm) { return 0; } | |
2292 | ||
2293 | static inline int | |
2294 | prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p) | |
2295 | { | |
2296 | return 0; | |
2297 | } | |
2298 | ||
2299 | static inline void * | |
2300 | get_highmem_page_buffer(struct page *page, struct chain_allocator *ca) | |
2301 | { | |
69643279 | 2302 | return ERR_PTR(-EINVAL); |
8357376d RW |
2303 | } |
2304 | ||
2305 | static inline void copy_last_highmem_page(void) {} | |
2306 | static inline int last_highmem_page_copied(void) { return 1; } | |
2307 | static inline void free_highmem_data(void) {} | |
2308 | #endif /* CONFIG_HIGHMEM */ | |
2309 | ||
f577eb30 | 2310 | /** |
940864dd RW |
2311 | * prepare_image - use the memory bitmap @bm to mark the pages that will |
2312 | * be overwritten in the process of restoring the system memory state | |
2313 | * from the suspend image ("unsafe" pages) and allocate memory for the | |
2314 | * image. | |
968808b8 | 2315 | * |
940864dd RW |
2316 | * The idea is to allocate a new memory bitmap first and then allocate |
2317 | * as many pages as needed for the image data, but not to assign these | |
2318 | * pages to specific tasks initially. Instead, we just mark them as | |
8357376d RW |
2319 | * allocated and create a lists of "safe" pages that will be used |
2320 | * later. On systems with high memory a list of "safe" highmem pages is | |
2321 | * also created. | |
f577eb30 RW |
2322 | */ |
2323 | ||
940864dd RW |
2324 | #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe)) |
2325 | ||
940864dd RW |
2326 | static int |
2327 | prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm) | |
f577eb30 | 2328 | { |
8357376d | 2329 | unsigned int nr_pages, nr_highmem; |
940864dd RW |
2330 | struct linked_page *sp_list, *lp; |
2331 | int error; | |
f577eb30 | 2332 | |
8357376d RW |
2333 | /* If there is no highmem, the buffer will not be necessary */ |
2334 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
2335 | buffer = NULL; | |
2336 | ||
2337 | nr_highmem = count_highmem_image_pages(bm); | |
940864dd RW |
2338 | error = mark_unsafe_pages(bm); |
2339 | if (error) | |
2340 | goto Free; | |
2341 | ||
2342 | error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE); | |
2343 | if (error) | |
2344 | goto Free; | |
2345 | ||
2346 | duplicate_memory_bitmap(new_bm, bm); | |
2347 | memory_bm_free(bm, PG_UNSAFE_KEEP); | |
8357376d RW |
2348 | if (nr_highmem > 0) { |
2349 | error = prepare_highmem_image(bm, &nr_highmem); | |
2350 | if (error) | |
2351 | goto Free; | |
2352 | } | |
940864dd RW |
2353 | /* Reserve some safe pages for potential later use. |
2354 | * | |
2355 | * NOTE: This way we make sure there will be enough safe pages for the | |
2356 | * chain_alloc() in get_buffer(). It is a bit wasteful, but | |
2357 | * nr_copy_pages cannot be greater than 50% of the memory anyway. | |
2358 | */ | |
2359 | sp_list = NULL; | |
2360 | /* nr_copy_pages cannot be lesser than allocated_unsafe_pages */ | |
8357376d | 2361 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
2362 | nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE); |
2363 | while (nr_pages > 0) { | |
8357376d | 2364 | lp = get_image_page(GFP_ATOMIC, PG_SAFE); |
940864dd | 2365 | if (!lp) { |
f577eb30 | 2366 | error = -ENOMEM; |
940864dd RW |
2367 | goto Free; |
2368 | } | |
2369 | lp->next = sp_list; | |
2370 | sp_list = lp; | |
2371 | nr_pages--; | |
f577eb30 | 2372 | } |
940864dd RW |
2373 | /* Preallocate memory for the image */ |
2374 | safe_pages_list = NULL; | |
8357376d | 2375 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; |
940864dd RW |
2376 | while (nr_pages > 0) { |
2377 | lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC); | |
2378 | if (!lp) { | |
2379 | error = -ENOMEM; | |
2380 | goto Free; | |
2381 | } | |
7be98234 | 2382 | if (!swsusp_page_is_free(virt_to_page(lp))) { |
940864dd RW |
2383 | /* The page is "safe", add it to the list */ |
2384 | lp->next = safe_pages_list; | |
2385 | safe_pages_list = lp; | |
968808b8 | 2386 | } |
940864dd | 2387 | /* Mark the page as allocated */ |
7be98234 RW |
2388 | swsusp_set_page_forbidden(virt_to_page(lp)); |
2389 | swsusp_set_page_free(virt_to_page(lp)); | |
940864dd | 2390 | nr_pages--; |
968808b8 | 2391 | } |
940864dd RW |
2392 | /* Free the reserved safe pages so that chain_alloc() can use them */ |
2393 | while (sp_list) { | |
2394 | lp = sp_list->next; | |
2395 | free_image_page(sp_list, PG_UNSAFE_CLEAR); | |
2396 | sp_list = lp; | |
f577eb30 | 2397 | } |
940864dd RW |
2398 | return 0; |
2399 | ||
59a49335 | 2400 | Free: |
940864dd | 2401 | swsusp_free(); |
f577eb30 RW |
2402 | return error; |
2403 | } | |
2404 | ||
940864dd RW |
2405 | /** |
2406 | * get_buffer - compute the address that snapshot_write_next() should | |
2407 | * set for its caller to write to. | |
2408 | */ | |
2409 | ||
2410 | static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca) | |
968808b8 | 2411 | { |
940864dd | 2412 | struct pbe *pbe; |
69643279 RW |
2413 | struct page *page; |
2414 | unsigned long pfn = memory_bm_next_pfn(bm); | |
968808b8 | 2415 | |
69643279 RW |
2416 | if (pfn == BM_END_OF_MAP) |
2417 | return ERR_PTR(-EFAULT); | |
2418 | ||
2419 | page = pfn_to_page(pfn); | |
8357376d RW |
2420 | if (PageHighMem(page)) |
2421 | return get_highmem_page_buffer(page, ca); | |
2422 | ||
7be98234 | 2423 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) |
940864dd RW |
2424 | /* We have allocated the "original" page frame and we can |
2425 | * use it directly to store the loaded page. | |
968808b8 | 2426 | */ |
940864dd RW |
2427 | return page_address(page); |
2428 | ||
2429 | /* The "original" page frame has not been allocated and we have to | |
2430 | * use a "safe" page frame to store the loaded page. | |
968808b8 | 2431 | */ |
940864dd RW |
2432 | pbe = chain_alloc(ca, sizeof(struct pbe)); |
2433 | if (!pbe) { | |
2434 | swsusp_free(); | |
69643279 | 2435 | return ERR_PTR(-ENOMEM); |
940864dd | 2436 | } |
8357376d RW |
2437 | pbe->orig_address = page_address(page); |
2438 | pbe->address = safe_pages_list; | |
940864dd RW |
2439 | safe_pages_list = safe_pages_list->next; |
2440 | pbe->next = restore_pblist; | |
2441 | restore_pblist = pbe; | |
8357376d | 2442 | return pbe->address; |
968808b8 RW |
2443 | } |
2444 | ||
f577eb30 RW |
2445 | /** |
2446 | * snapshot_write_next - used for writing the system memory snapshot. | |
2447 | * | |
2448 | * On the first call to it @handle should point to a zeroed | |
2449 | * snapshot_handle structure. The structure gets updated and a pointer | |
2450 | * to it should be passed to this function every next time. | |
2451 | * | |
f577eb30 RW |
2452 | * On success the function returns a positive number. Then, the caller |
2453 | * is allowed to write up to the returned number of bytes to the memory | |
d3c1b24c | 2454 | * location computed by the data_of() macro. |
f577eb30 RW |
2455 | * |
2456 | * The function returns 0 to indicate the "end of file" condition, | |
2457 | * and a negative number is returned on error. In such cases the | |
2458 | * structure pointed to by @handle is not updated and should not be used | |
2459 | * any more. | |
2460 | */ | |
2461 | ||
d3c1b24c | 2462 | int snapshot_write_next(struct snapshot_handle *handle) |
f577eb30 | 2463 | { |
940864dd | 2464 | static struct chain_allocator ca; |
f577eb30 RW |
2465 | int error = 0; |
2466 | ||
940864dd | 2467 | /* Check if we have already loaded the entire image */ |
d3c1b24c | 2468 | if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) |
f577eb30 | 2469 | return 0; |
940864dd | 2470 | |
d3c1b24c JS |
2471 | handle->sync_read = 1; |
2472 | ||
2473 | if (!handle->cur) { | |
8357376d RW |
2474 | if (!buffer) |
2475 | /* This makes the buffer be freed by swsusp_free() */ | |
2476 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); | |
2477 | ||
f577eb30 RW |
2478 | if (!buffer) |
2479 | return -ENOMEM; | |
8357376d | 2480 | |
f577eb30 | 2481 | handle->buffer = buffer; |
d3c1b24c JS |
2482 | } else if (handle->cur == 1) { |
2483 | error = load_header(buffer); | |
2484 | if (error) | |
2485 | return error; | |
940864dd | 2486 | |
d3c1b24c JS |
2487 | error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY); |
2488 | if (error) | |
2489 | return error; | |
2490 | ||
85055dd8 MS |
2491 | /* Allocate buffer for page keys. */ |
2492 | error = page_key_alloc(nr_copy_pages); | |
2493 | if (error) | |
2494 | return error; | |
2495 | ||
d3c1b24c JS |
2496 | } else if (handle->cur <= nr_meta_pages + 1) { |
2497 | error = unpack_orig_pfns(buffer, ©_bm); | |
2498 | if (error) | |
2499 | return error; | |
940864dd | 2500 | |
d3c1b24c JS |
2501 | if (handle->cur == nr_meta_pages + 1) { |
2502 | error = prepare_image(&orig_bm, ©_bm); | |
69643279 RW |
2503 | if (error) |
2504 | return error; | |
2505 | ||
d3c1b24c JS |
2506 | chain_init(&ca, GFP_ATOMIC, PG_SAFE); |
2507 | memory_bm_position_reset(&orig_bm); | |
2508 | restore_pblist = NULL; | |
940864dd | 2509 | handle->buffer = get_buffer(&orig_bm, &ca); |
d3c1b24c | 2510 | handle->sync_read = 0; |
69643279 RW |
2511 | if (IS_ERR(handle->buffer)) |
2512 | return PTR_ERR(handle->buffer); | |
f577eb30 | 2513 | } |
f577eb30 | 2514 | } else { |
d3c1b24c | 2515 | copy_last_highmem_page(); |
85055dd8 MS |
2516 | /* Restore page key for data page (s390 only). */ |
2517 | page_key_write(handle->buffer); | |
d3c1b24c JS |
2518 | handle->buffer = get_buffer(&orig_bm, &ca); |
2519 | if (IS_ERR(handle->buffer)) | |
2520 | return PTR_ERR(handle->buffer); | |
2521 | if (handle->buffer != buffer) | |
2522 | handle->sync_read = 0; | |
f577eb30 | 2523 | } |
d3c1b24c JS |
2524 | handle->cur++; |
2525 | return PAGE_SIZE; | |
f577eb30 RW |
2526 | } |
2527 | ||
8357376d RW |
2528 | /** |
2529 | * snapshot_write_finalize - must be called after the last call to | |
2530 | * snapshot_write_next() in case the last page in the image happens | |
2531 | * to be a highmem page and its contents should be stored in the | |
2532 | * highmem. Additionally, it releases the memory that will not be | |
2533 | * used any more. | |
2534 | */ | |
2535 | ||
2536 | void snapshot_write_finalize(struct snapshot_handle *handle) | |
2537 | { | |
2538 | copy_last_highmem_page(); | |
85055dd8 MS |
2539 | /* Restore page key for data page (s390 only). */ |
2540 | page_key_write(handle->buffer); | |
2541 | page_key_free(); | |
8357376d | 2542 | /* Free only if we have loaded the image entirely */ |
d3c1b24c | 2543 | if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) { |
8357376d RW |
2544 | memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR); |
2545 | free_highmem_data(); | |
2546 | } | |
2547 | } | |
2548 | ||
f577eb30 RW |
2549 | int snapshot_image_loaded(struct snapshot_handle *handle) |
2550 | { | |
8357376d | 2551 | return !(!nr_copy_pages || !last_highmem_page_copied() || |
940864dd RW |
2552 | handle->cur <= nr_meta_pages + nr_copy_pages); |
2553 | } | |
2554 | ||
8357376d RW |
2555 | #ifdef CONFIG_HIGHMEM |
2556 | /* Assumes that @buf is ready and points to a "safe" page */ | |
2557 | static inline void | |
2558 | swap_two_pages_data(struct page *p1, struct page *p2, void *buf) | |
940864dd | 2559 | { |
8357376d RW |
2560 | void *kaddr1, *kaddr2; |
2561 | ||
0de9a1e2 CW |
2562 | kaddr1 = kmap_atomic(p1); |
2563 | kaddr2 = kmap_atomic(p2); | |
3ecb01df JB |
2564 | copy_page(buf, kaddr1); |
2565 | copy_page(kaddr1, kaddr2); | |
2566 | copy_page(kaddr2, buf); | |
0de9a1e2 CW |
2567 | kunmap_atomic(kaddr2); |
2568 | kunmap_atomic(kaddr1); | |
8357376d RW |
2569 | } |
2570 | ||
2571 | /** | |
2572 | * restore_highmem - for each highmem page that was allocated before | |
2573 | * the suspend and included in the suspend image, and also has been | |
2574 | * allocated by the "resume" kernel swap its current (ie. "before | |
2575 | * resume") contents with the previous (ie. "before suspend") one. | |
2576 | * | |
2577 | * If the resume eventually fails, we can call this function once | |
2578 | * again and restore the "before resume" highmem state. | |
2579 | */ | |
2580 | ||
2581 | int restore_highmem(void) | |
2582 | { | |
2583 | struct highmem_pbe *pbe = highmem_pblist; | |
2584 | void *buf; | |
2585 | ||
2586 | if (!pbe) | |
2587 | return 0; | |
2588 | ||
2589 | buf = get_image_page(GFP_ATOMIC, PG_SAFE); | |
2590 | if (!buf) | |
2591 | return -ENOMEM; | |
2592 | ||
2593 | while (pbe) { | |
2594 | swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf); | |
2595 | pbe = pbe->next; | |
2596 | } | |
2597 | free_image_page(buf, PG_UNSAFE_CLEAR); | |
2598 | return 0; | |
f577eb30 | 2599 | } |
8357376d | 2600 | #endif /* CONFIG_HIGHMEM */ |