]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/kernel/power/snapshot.c | |
3 | * | |
4 | * This file provides system snapshot/restore functionality for swsusp. | |
5 | * | |
6 | * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz> | |
7 | * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl> | |
8 | * | |
9 | * This file is released under the GPLv2. | |
10 | * | |
11 | */ | |
12 | ||
13 | #include <linux/version.h> | |
14 | #include <linux/module.h> | |
15 | #include <linux/mm.h> | |
16 | #include <linux/suspend.h> | |
17 | #include <linux/delay.h> | |
18 | #include <linux/bitops.h> | |
19 | #include <linux/spinlock.h> | |
20 | #include <linux/kernel.h> | |
21 | #include <linux/pm.h> | |
22 | #include <linux/device.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/bootmem.h> | |
25 | #include <linux/nmi.h> | |
26 | #include <linux/syscalls.h> | |
27 | #include <linux/console.h> | |
28 | #include <linux/highmem.h> | |
29 | #include <linux/list.h> | |
30 | #include <linux/slab.h> | |
31 | #include <linux/compiler.h> | |
32 | #include <linux/ktime.h> | |
33 | #include <linux/set_memory.h> | |
34 | ||
35 | #include <linux/uaccess.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <asm/pgtable.h> | |
38 | #include <asm/tlbflush.h> | |
39 | #include <asm/io.h> | |
40 | ||
41 | #include "power.h" | |
42 | ||
43 | #if defined(CONFIG_STRICT_KERNEL_RWX) && defined(CONFIG_ARCH_HAS_SET_MEMORY) | |
44 | static bool hibernate_restore_protection; | |
45 | static bool hibernate_restore_protection_active; | |
46 | ||
47 | void enable_restore_image_protection(void) | |
48 | { | |
49 | hibernate_restore_protection = true; | |
50 | } | |
51 | ||
52 | static inline void hibernate_restore_protection_begin(void) | |
53 | { | |
54 | hibernate_restore_protection_active = hibernate_restore_protection; | |
55 | } | |
56 | ||
57 | static inline void hibernate_restore_protection_end(void) | |
58 | { | |
59 | hibernate_restore_protection_active = false; | |
60 | } | |
61 | ||
62 | static inline void hibernate_restore_protect_page(void *page_address) | |
63 | { | |
64 | if (hibernate_restore_protection_active) | |
65 | set_memory_ro((unsigned long)page_address, 1); | |
66 | } | |
67 | ||
68 | static inline void hibernate_restore_unprotect_page(void *page_address) | |
69 | { | |
70 | if (hibernate_restore_protection_active) | |
71 | set_memory_rw((unsigned long)page_address, 1); | |
72 | } | |
73 | #else | |
74 | static inline void hibernate_restore_protection_begin(void) {} | |
75 | static inline void hibernate_restore_protection_end(void) {} | |
76 | static inline void hibernate_restore_protect_page(void *page_address) {} | |
77 | static inline void hibernate_restore_unprotect_page(void *page_address) {} | |
78 | #endif /* CONFIG_STRICT_KERNEL_RWX && CONFIG_ARCH_HAS_SET_MEMORY */ | |
79 | ||
80 | static int swsusp_page_is_free(struct page *); | |
81 | static void swsusp_set_page_forbidden(struct page *); | |
82 | static void swsusp_unset_page_forbidden(struct page *); | |
83 | ||
84 | /* | |
85 | * Number of bytes to reserve for memory allocations made by device drivers | |
86 | * from their ->freeze() and ->freeze_noirq() callbacks so that they don't | |
87 | * cause image creation to fail (tunable via /sys/power/reserved_size). | |
88 | */ | |
89 | unsigned long reserved_size; | |
90 | ||
91 | void __init hibernate_reserved_size_init(void) | |
92 | { | |
93 | reserved_size = SPARE_PAGES * PAGE_SIZE; | |
94 | } | |
95 | ||
96 | /* | |
97 | * Preferred image size in bytes (tunable via /sys/power/image_size). | |
98 | * When it is set to N, swsusp will do its best to ensure the image | |
99 | * size will not exceed N bytes, but if that is impossible, it will | |
100 | * try to create the smallest image possible. | |
101 | */ | |
102 | unsigned long image_size; | |
103 | ||
104 | void __init hibernate_image_size_init(void) | |
105 | { | |
106 | image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE; | |
107 | } | |
108 | ||
109 | /* | |
110 | * List of PBEs needed for restoring the pages that were allocated before | |
111 | * the suspend and included in the suspend image, but have also been | |
112 | * allocated by the "resume" kernel, so their contents cannot be written | |
113 | * directly to their "original" page frames. | |
114 | */ | |
115 | struct pbe *restore_pblist; | |
116 | ||
117 | /* struct linked_page is used to build chains of pages */ | |
118 | ||
119 | #define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *)) | |
120 | ||
121 | struct linked_page { | |
122 | struct linked_page *next; | |
123 | char data[LINKED_PAGE_DATA_SIZE]; | |
124 | } __packed; | |
125 | ||
126 | /* | |
127 | * List of "safe" pages (ie. pages that were not used by the image kernel | |
128 | * before hibernation) that may be used as temporary storage for image kernel | |
129 | * memory contents. | |
130 | */ | |
131 | static struct linked_page *safe_pages_list; | |
132 | ||
133 | /* Pointer to an auxiliary buffer (1 page) */ | |
134 | static void *buffer; | |
135 | ||
136 | #define PG_ANY 0 | |
137 | #define PG_SAFE 1 | |
138 | #define PG_UNSAFE_CLEAR 1 | |
139 | #define PG_UNSAFE_KEEP 0 | |
140 | ||
141 | static unsigned int allocated_unsafe_pages; | |
142 | ||
143 | /** | |
144 | * get_image_page - Allocate a page for a hibernation image. | |
145 | * @gfp_mask: GFP mask for the allocation. | |
146 | * @safe_needed: Get pages that were not used before hibernation (restore only) | |
147 | * | |
148 | * During image restoration, for storing the PBE list and the image data, we can | |
149 | * only use memory pages that do not conflict with the pages used before | |
150 | * hibernation. The "unsafe" pages have PageNosaveFree set and we count them | |
151 | * using allocated_unsafe_pages. | |
152 | * | |
153 | * Each allocated image page is marked as PageNosave and PageNosaveFree so that | |
154 | * swsusp_free() can release it. | |
155 | */ | |
156 | static void *get_image_page(gfp_t gfp_mask, int safe_needed) | |
157 | { | |
158 | void *res; | |
159 | ||
160 | res = (void *)get_zeroed_page(gfp_mask); | |
161 | if (safe_needed) | |
162 | while (res && swsusp_page_is_free(virt_to_page(res))) { | |
163 | /* The page is unsafe, mark it for swsusp_free() */ | |
164 | swsusp_set_page_forbidden(virt_to_page(res)); | |
165 | allocated_unsafe_pages++; | |
166 | res = (void *)get_zeroed_page(gfp_mask); | |
167 | } | |
168 | if (res) { | |
169 | swsusp_set_page_forbidden(virt_to_page(res)); | |
170 | swsusp_set_page_free(virt_to_page(res)); | |
171 | } | |
172 | return res; | |
173 | } | |
174 | ||
175 | static void *__get_safe_page(gfp_t gfp_mask) | |
176 | { | |
177 | if (safe_pages_list) { | |
178 | void *ret = safe_pages_list; | |
179 | ||
180 | safe_pages_list = safe_pages_list->next; | |
181 | memset(ret, 0, PAGE_SIZE); | |
182 | return ret; | |
183 | } | |
184 | return get_image_page(gfp_mask, PG_SAFE); | |
185 | } | |
186 | ||
187 | unsigned long get_safe_page(gfp_t gfp_mask) | |
188 | { | |
189 | return (unsigned long)__get_safe_page(gfp_mask); | |
190 | } | |
191 | ||
192 | static struct page *alloc_image_page(gfp_t gfp_mask) | |
193 | { | |
194 | struct page *page; | |
195 | ||
196 | page = alloc_page(gfp_mask); | |
197 | if (page) { | |
198 | swsusp_set_page_forbidden(page); | |
199 | swsusp_set_page_free(page); | |
200 | } | |
201 | return page; | |
202 | } | |
203 | ||
204 | static void recycle_safe_page(void *page_address) | |
205 | { | |
206 | struct linked_page *lp = page_address; | |
207 | ||
208 | lp->next = safe_pages_list; | |
209 | safe_pages_list = lp; | |
210 | } | |
211 | ||
212 | /** | |
213 | * free_image_page - Free a page allocated for hibernation image. | |
214 | * @addr: Address of the page to free. | |
215 | * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page. | |
216 | * | |
217 | * The page to free should have been allocated by get_image_page() (page flags | |
218 | * set by it are affected). | |
219 | */ | |
220 | static inline void free_image_page(void *addr, int clear_nosave_free) | |
221 | { | |
222 | struct page *page; | |
223 | ||
224 | BUG_ON(!virt_addr_valid(addr)); | |
225 | ||
226 | page = virt_to_page(addr); | |
227 | ||
228 | swsusp_unset_page_forbidden(page); | |
229 | if (clear_nosave_free) | |
230 | swsusp_unset_page_free(page); | |
231 | ||
232 | __free_page(page); | |
233 | } | |
234 | ||
235 | static inline void free_list_of_pages(struct linked_page *list, | |
236 | int clear_page_nosave) | |
237 | { | |
238 | while (list) { | |
239 | struct linked_page *lp = list->next; | |
240 | ||
241 | free_image_page(list, clear_page_nosave); | |
242 | list = lp; | |
243 | } | |
244 | } | |
245 | ||
246 | /* | |
247 | * struct chain_allocator is used for allocating small objects out of | |
248 | * a linked list of pages called 'the chain'. | |
249 | * | |
250 | * The chain grows each time when there is no room for a new object in | |
251 | * the current page. The allocated objects cannot be freed individually. | |
252 | * It is only possible to free them all at once, by freeing the entire | |
253 | * chain. | |
254 | * | |
255 | * NOTE: The chain allocator may be inefficient if the allocated objects | |
256 | * are not much smaller than PAGE_SIZE. | |
257 | */ | |
258 | struct chain_allocator { | |
259 | struct linked_page *chain; /* the chain */ | |
260 | unsigned int used_space; /* total size of objects allocated out | |
261 | of the current page */ | |
262 | gfp_t gfp_mask; /* mask for allocating pages */ | |
263 | int safe_needed; /* if set, only "safe" pages are allocated */ | |
264 | }; | |
265 | ||
266 | static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask, | |
267 | int safe_needed) | |
268 | { | |
269 | ca->chain = NULL; | |
270 | ca->used_space = LINKED_PAGE_DATA_SIZE; | |
271 | ca->gfp_mask = gfp_mask; | |
272 | ca->safe_needed = safe_needed; | |
273 | } | |
274 | ||
275 | static void *chain_alloc(struct chain_allocator *ca, unsigned int size) | |
276 | { | |
277 | void *ret; | |
278 | ||
279 | if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) { | |
280 | struct linked_page *lp; | |
281 | ||
282 | lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) : | |
283 | get_image_page(ca->gfp_mask, PG_ANY); | |
284 | if (!lp) | |
285 | return NULL; | |
286 | ||
287 | lp->next = ca->chain; | |
288 | ca->chain = lp; | |
289 | ca->used_space = 0; | |
290 | } | |
291 | ret = ca->chain->data + ca->used_space; | |
292 | ca->used_space += size; | |
293 | return ret; | |
294 | } | |
295 | ||
296 | /** | |
297 | * Data types related to memory bitmaps. | |
298 | * | |
299 | * Memory bitmap is a structure consiting of many linked lists of | |
300 | * objects. The main list's elements are of type struct zone_bitmap | |
301 | * and each of them corresonds to one zone. For each zone bitmap | |
302 | * object there is a list of objects of type struct bm_block that | |
303 | * represent each blocks of bitmap in which information is stored. | |
304 | * | |
305 | * struct memory_bitmap contains a pointer to the main list of zone | |
306 | * bitmap objects, a struct bm_position used for browsing the bitmap, | |
307 | * and a pointer to the list of pages used for allocating all of the | |
308 | * zone bitmap objects and bitmap block objects. | |
309 | * | |
310 | * NOTE: It has to be possible to lay out the bitmap in memory | |
311 | * using only allocations of order 0. Additionally, the bitmap is | |
312 | * designed to work with arbitrary number of zones (this is over the | |
313 | * top for now, but let's avoid making unnecessary assumptions ;-). | |
314 | * | |
315 | * struct zone_bitmap contains a pointer to a list of bitmap block | |
316 | * objects and a pointer to the bitmap block object that has been | |
317 | * most recently used for setting bits. Additionally, it contains the | |
318 | * PFNs that correspond to the start and end of the represented zone. | |
319 | * | |
320 | * struct bm_block contains a pointer to the memory page in which | |
321 | * information is stored (in the form of a block of bitmap) | |
322 | * It also contains the pfns that correspond to the start and end of | |
323 | * the represented memory area. | |
324 | * | |
325 | * The memory bitmap is organized as a radix tree to guarantee fast random | |
326 | * access to the bits. There is one radix tree for each zone (as returned | |
327 | * from create_mem_extents). | |
328 | * | |
329 | * One radix tree is represented by one struct mem_zone_bm_rtree. There are | |
330 | * two linked lists for the nodes of the tree, one for the inner nodes and | |
331 | * one for the leave nodes. The linked leave nodes are used for fast linear | |
332 | * access of the memory bitmap. | |
333 | * | |
334 | * The struct rtree_node represents one node of the radix tree. | |
335 | */ | |
336 | ||
337 | #define BM_END_OF_MAP (~0UL) | |
338 | ||
339 | #define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE) | |
340 | #define BM_BLOCK_SHIFT (PAGE_SHIFT + 3) | |
341 | #define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1) | |
342 | ||
343 | /* | |
344 | * struct rtree_node is a wrapper struct to link the nodes | |
345 | * of the rtree together for easy linear iteration over | |
346 | * bits and easy freeing | |
347 | */ | |
348 | struct rtree_node { | |
349 | struct list_head list; | |
350 | unsigned long *data; | |
351 | }; | |
352 | ||
353 | /* | |
354 | * struct mem_zone_bm_rtree represents a bitmap used for one | |
355 | * populated memory zone. | |
356 | */ | |
357 | struct mem_zone_bm_rtree { | |
358 | struct list_head list; /* Link Zones together */ | |
359 | struct list_head nodes; /* Radix Tree inner nodes */ | |
360 | struct list_head leaves; /* Radix Tree leaves */ | |
361 | unsigned long start_pfn; /* Zone start page frame */ | |
362 | unsigned long end_pfn; /* Zone end page frame + 1 */ | |
363 | struct rtree_node *rtree; /* Radix Tree Root */ | |
364 | int levels; /* Number of Radix Tree Levels */ | |
365 | unsigned int blocks; /* Number of Bitmap Blocks */ | |
366 | }; | |
367 | ||
368 | /* strcut bm_position is used for browsing memory bitmaps */ | |
369 | ||
370 | struct bm_position { | |
371 | struct mem_zone_bm_rtree *zone; | |
372 | struct rtree_node *node; | |
373 | unsigned long node_pfn; | |
374 | int node_bit; | |
375 | }; | |
376 | ||
377 | struct memory_bitmap { | |
378 | struct list_head zones; | |
379 | struct linked_page *p_list; /* list of pages used to store zone | |
380 | bitmap objects and bitmap block | |
381 | objects */ | |
382 | struct bm_position cur; /* most recently used bit position */ | |
383 | }; | |
384 | ||
385 | /* Functions that operate on memory bitmaps */ | |
386 | ||
387 | #define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long)) | |
388 | #if BITS_PER_LONG == 32 | |
389 | #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2) | |
390 | #else | |
391 | #define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3) | |
392 | #endif | |
393 | #define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1) | |
394 | ||
395 | /** | |
396 | * alloc_rtree_node - Allocate a new node and add it to the radix tree. | |
397 | * | |
398 | * This function is used to allocate inner nodes as well as the | |
399 | * leave nodes of the radix tree. It also adds the node to the | |
400 | * corresponding linked list passed in by the *list parameter. | |
401 | */ | |
402 | static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed, | |
403 | struct chain_allocator *ca, | |
404 | struct list_head *list) | |
405 | { | |
406 | struct rtree_node *node; | |
407 | ||
408 | node = chain_alloc(ca, sizeof(struct rtree_node)); | |
409 | if (!node) | |
410 | return NULL; | |
411 | ||
412 | node->data = get_image_page(gfp_mask, safe_needed); | |
413 | if (!node->data) | |
414 | return NULL; | |
415 | ||
416 | list_add_tail(&node->list, list); | |
417 | ||
418 | return node; | |
419 | } | |
420 | ||
421 | /** | |
422 | * add_rtree_block - Add a new leave node to the radix tree. | |
423 | * | |
424 | * The leave nodes need to be allocated in order to keep the leaves | |
425 | * linked list in order. This is guaranteed by the zone->blocks | |
426 | * counter. | |
427 | */ | |
428 | static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask, | |
429 | int safe_needed, struct chain_allocator *ca) | |
430 | { | |
431 | struct rtree_node *node, *block, **dst; | |
432 | unsigned int levels_needed, block_nr; | |
433 | int i; | |
434 | ||
435 | block_nr = zone->blocks; | |
436 | levels_needed = 0; | |
437 | ||
438 | /* How many levels do we need for this block nr? */ | |
439 | while (block_nr) { | |
440 | levels_needed += 1; | |
441 | block_nr >>= BM_RTREE_LEVEL_SHIFT; | |
442 | } | |
443 | ||
444 | /* Make sure the rtree has enough levels */ | |
445 | for (i = zone->levels; i < levels_needed; i++) { | |
446 | node = alloc_rtree_node(gfp_mask, safe_needed, ca, | |
447 | &zone->nodes); | |
448 | if (!node) | |
449 | return -ENOMEM; | |
450 | ||
451 | node->data[0] = (unsigned long)zone->rtree; | |
452 | zone->rtree = node; | |
453 | zone->levels += 1; | |
454 | } | |
455 | ||
456 | /* Allocate new block */ | |
457 | block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves); | |
458 | if (!block) | |
459 | return -ENOMEM; | |
460 | ||
461 | /* Now walk the rtree to insert the block */ | |
462 | node = zone->rtree; | |
463 | dst = &zone->rtree; | |
464 | block_nr = zone->blocks; | |
465 | for (i = zone->levels; i > 0; i--) { | |
466 | int index; | |
467 | ||
468 | if (!node) { | |
469 | node = alloc_rtree_node(gfp_mask, safe_needed, ca, | |
470 | &zone->nodes); | |
471 | if (!node) | |
472 | return -ENOMEM; | |
473 | *dst = node; | |
474 | } | |
475 | ||
476 | index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT); | |
477 | index &= BM_RTREE_LEVEL_MASK; | |
478 | dst = (struct rtree_node **)&((*dst)->data[index]); | |
479 | node = *dst; | |
480 | } | |
481 | ||
482 | zone->blocks += 1; | |
483 | *dst = block; | |
484 | ||
485 | return 0; | |
486 | } | |
487 | ||
488 | static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone, | |
489 | int clear_nosave_free); | |
490 | ||
491 | /** | |
492 | * create_zone_bm_rtree - Create a radix tree for one zone. | |
493 | * | |
494 | * Allocated the mem_zone_bm_rtree structure and initializes it. | |
495 | * This function also allocated and builds the radix tree for the | |
496 | * zone. | |
497 | */ | |
498 | static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask, | |
499 | int safe_needed, | |
500 | struct chain_allocator *ca, | |
501 | unsigned long start, | |
502 | unsigned long end) | |
503 | { | |
504 | struct mem_zone_bm_rtree *zone; | |
505 | unsigned int i, nr_blocks; | |
506 | unsigned long pages; | |
507 | ||
508 | pages = end - start; | |
509 | zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree)); | |
510 | if (!zone) | |
511 | return NULL; | |
512 | ||
513 | INIT_LIST_HEAD(&zone->nodes); | |
514 | INIT_LIST_HEAD(&zone->leaves); | |
515 | zone->start_pfn = start; | |
516 | zone->end_pfn = end; | |
517 | nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK); | |
518 | ||
519 | for (i = 0; i < nr_blocks; i++) { | |
520 | if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) { | |
521 | free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR); | |
522 | return NULL; | |
523 | } | |
524 | } | |
525 | ||
526 | return zone; | |
527 | } | |
528 | ||
529 | /** | |
530 | * free_zone_bm_rtree - Free the memory of the radix tree. | |
531 | * | |
532 | * Free all node pages of the radix tree. The mem_zone_bm_rtree | |
533 | * structure itself is not freed here nor are the rtree_node | |
534 | * structs. | |
535 | */ | |
536 | static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone, | |
537 | int clear_nosave_free) | |
538 | { | |
539 | struct rtree_node *node; | |
540 | ||
541 | list_for_each_entry(node, &zone->nodes, list) | |
542 | free_image_page(node->data, clear_nosave_free); | |
543 | ||
544 | list_for_each_entry(node, &zone->leaves, list) | |
545 | free_image_page(node->data, clear_nosave_free); | |
546 | } | |
547 | ||
548 | static void memory_bm_position_reset(struct memory_bitmap *bm) | |
549 | { | |
550 | bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree, | |
551 | list); | |
552 | bm->cur.node = list_entry(bm->cur.zone->leaves.next, | |
553 | struct rtree_node, list); | |
554 | bm->cur.node_pfn = 0; | |
555 | bm->cur.node_bit = 0; | |
556 | } | |
557 | ||
558 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free); | |
559 | ||
560 | struct mem_extent { | |
561 | struct list_head hook; | |
562 | unsigned long start; | |
563 | unsigned long end; | |
564 | }; | |
565 | ||
566 | /** | |
567 | * free_mem_extents - Free a list of memory extents. | |
568 | * @list: List of extents to free. | |
569 | */ | |
570 | static void free_mem_extents(struct list_head *list) | |
571 | { | |
572 | struct mem_extent *ext, *aux; | |
573 | ||
574 | list_for_each_entry_safe(ext, aux, list, hook) { | |
575 | list_del(&ext->hook); | |
576 | kfree(ext); | |
577 | } | |
578 | } | |
579 | ||
580 | /** | |
581 | * create_mem_extents - Create a list of memory extents. | |
582 | * @list: List to put the extents into. | |
583 | * @gfp_mask: Mask to use for memory allocations. | |
584 | * | |
585 | * The extents represent contiguous ranges of PFNs. | |
586 | */ | |
587 | static int create_mem_extents(struct list_head *list, gfp_t gfp_mask) | |
588 | { | |
589 | struct zone *zone; | |
590 | ||
591 | INIT_LIST_HEAD(list); | |
592 | ||
593 | for_each_populated_zone(zone) { | |
594 | unsigned long zone_start, zone_end; | |
595 | struct mem_extent *ext, *cur, *aux; | |
596 | ||
597 | zone_start = zone->zone_start_pfn; | |
598 | zone_end = zone_end_pfn(zone); | |
599 | ||
600 | list_for_each_entry(ext, list, hook) | |
601 | if (zone_start <= ext->end) | |
602 | break; | |
603 | ||
604 | if (&ext->hook == list || zone_end < ext->start) { | |
605 | /* New extent is necessary */ | |
606 | struct mem_extent *new_ext; | |
607 | ||
608 | new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask); | |
609 | if (!new_ext) { | |
610 | free_mem_extents(list); | |
611 | return -ENOMEM; | |
612 | } | |
613 | new_ext->start = zone_start; | |
614 | new_ext->end = zone_end; | |
615 | list_add_tail(&new_ext->hook, &ext->hook); | |
616 | continue; | |
617 | } | |
618 | ||
619 | /* Merge this zone's range of PFNs with the existing one */ | |
620 | if (zone_start < ext->start) | |
621 | ext->start = zone_start; | |
622 | if (zone_end > ext->end) | |
623 | ext->end = zone_end; | |
624 | ||
625 | /* More merging may be possible */ | |
626 | cur = ext; | |
627 | list_for_each_entry_safe_continue(cur, aux, list, hook) { | |
628 | if (zone_end < cur->start) | |
629 | break; | |
630 | if (zone_end < cur->end) | |
631 | ext->end = cur->end; | |
632 | list_del(&cur->hook); | |
633 | kfree(cur); | |
634 | } | |
635 | } | |
636 | ||
637 | return 0; | |
638 | } | |
639 | ||
640 | /** | |
641 | * memory_bm_create - Allocate memory for a memory bitmap. | |
642 | */ | |
643 | static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, | |
644 | int safe_needed) | |
645 | { | |
646 | struct chain_allocator ca; | |
647 | struct list_head mem_extents; | |
648 | struct mem_extent *ext; | |
649 | int error; | |
650 | ||
651 | chain_init(&ca, gfp_mask, safe_needed); | |
652 | INIT_LIST_HEAD(&bm->zones); | |
653 | ||
654 | error = create_mem_extents(&mem_extents, gfp_mask); | |
655 | if (error) | |
656 | return error; | |
657 | ||
658 | list_for_each_entry(ext, &mem_extents, hook) { | |
659 | struct mem_zone_bm_rtree *zone; | |
660 | ||
661 | zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca, | |
662 | ext->start, ext->end); | |
663 | if (!zone) { | |
664 | error = -ENOMEM; | |
665 | goto Error; | |
666 | } | |
667 | list_add_tail(&zone->list, &bm->zones); | |
668 | } | |
669 | ||
670 | bm->p_list = ca.chain; | |
671 | memory_bm_position_reset(bm); | |
672 | Exit: | |
673 | free_mem_extents(&mem_extents); | |
674 | return error; | |
675 | ||
676 | Error: | |
677 | bm->p_list = ca.chain; | |
678 | memory_bm_free(bm, PG_UNSAFE_CLEAR); | |
679 | goto Exit; | |
680 | } | |
681 | ||
682 | /** | |
683 | * memory_bm_free - Free memory occupied by the memory bitmap. | |
684 | * @bm: Memory bitmap. | |
685 | */ | |
686 | static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free) | |
687 | { | |
688 | struct mem_zone_bm_rtree *zone; | |
689 | ||
690 | list_for_each_entry(zone, &bm->zones, list) | |
691 | free_zone_bm_rtree(zone, clear_nosave_free); | |
692 | ||
693 | free_list_of_pages(bm->p_list, clear_nosave_free); | |
694 | ||
695 | INIT_LIST_HEAD(&bm->zones); | |
696 | } | |
697 | ||
698 | /** | |
699 | * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap. | |
700 | * | |
701 | * Find the bit in memory bitmap @bm that corresponds to the given PFN. | |
702 | * The cur.zone, cur.block and cur.node_pfn members of @bm are updated. | |
703 | * | |
704 | * Walk the radix tree to find the page containing the bit that represents @pfn | |
705 | * and return the position of the bit in @addr and @bit_nr. | |
706 | */ | |
707 | static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn, | |
708 | void **addr, unsigned int *bit_nr) | |
709 | { | |
710 | struct mem_zone_bm_rtree *curr, *zone; | |
711 | struct rtree_node *node; | |
712 | int i, block_nr; | |
713 | ||
714 | zone = bm->cur.zone; | |
715 | ||
716 | if (pfn >= zone->start_pfn && pfn < zone->end_pfn) | |
717 | goto zone_found; | |
718 | ||
719 | zone = NULL; | |
720 | ||
721 | /* Find the right zone */ | |
722 | list_for_each_entry(curr, &bm->zones, list) { | |
723 | if (pfn >= curr->start_pfn && pfn < curr->end_pfn) { | |
724 | zone = curr; | |
725 | break; | |
726 | } | |
727 | } | |
728 | ||
729 | if (!zone) | |
730 | return -EFAULT; | |
731 | ||
732 | zone_found: | |
733 | /* | |
734 | * We have found the zone. Now walk the radix tree to find the leaf node | |
735 | * for our PFN. | |
736 | */ | |
737 | node = bm->cur.node; | |
738 | if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn) | |
739 | goto node_found; | |
740 | ||
741 | node = zone->rtree; | |
742 | block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT; | |
743 | ||
744 | for (i = zone->levels; i > 0; i--) { | |
745 | int index; | |
746 | ||
747 | index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT); | |
748 | index &= BM_RTREE_LEVEL_MASK; | |
749 | BUG_ON(node->data[index] == 0); | |
750 | node = (struct rtree_node *)node->data[index]; | |
751 | } | |
752 | ||
753 | node_found: | |
754 | /* Update last position */ | |
755 | bm->cur.zone = zone; | |
756 | bm->cur.node = node; | |
757 | bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK; | |
758 | ||
759 | /* Set return values */ | |
760 | *addr = node->data; | |
761 | *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK; | |
762 | ||
763 | return 0; | |
764 | } | |
765 | ||
766 | static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn) | |
767 | { | |
768 | void *addr; | |
769 | unsigned int bit; | |
770 | int error; | |
771 | ||
772 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); | |
773 | BUG_ON(error); | |
774 | set_bit(bit, addr); | |
775 | } | |
776 | ||
777 | static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn) | |
778 | { | |
779 | void *addr; | |
780 | unsigned int bit; | |
781 | int error; | |
782 | ||
783 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); | |
784 | if (!error) | |
785 | set_bit(bit, addr); | |
786 | ||
787 | return error; | |
788 | } | |
789 | ||
790 | static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn) | |
791 | { | |
792 | void *addr; | |
793 | unsigned int bit; | |
794 | int error; | |
795 | ||
796 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); | |
797 | BUG_ON(error); | |
798 | clear_bit(bit, addr); | |
799 | } | |
800 | ||
801 | static void memory_bm_clear_current(struct memory_bitmap *bm) | |
802 | { | |
803 | int bit; | |
804 | ||
805 | bit = max(bm->cur.node_bit - 1, 0); | |
806 | clear_bit(bit, bm->cur.node->data); | |
807 | } | |
808 | ||
809 | static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn) | |
810 | { | |
811 | void *addr; | |
812 | unsigned int bit; | |
813 | int error; | |
814 | ||
815 | error = memory_bm_find_bit(bm, pfn, &addr, &bit); | |
816 | BUG_ON(error); | |
817 | return test_bit(bit, addr); | |
818 | } | |
819 | ||
820 | static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn) | |
821 | { | |
822 | void *addr; | |
823 | unsigned int bit; | |
824 | ||
825 | return !memory_bm_find_bit(bm, pfn, &addr, &bit); | |
826 | } | |
827 | ||
828 | /* | |
829 | * rtree_next_node - Jump to the next leaf node. | |
830 | * | |
831 | * Set the position to the beginning of the next node in the | |
832 | * memory bitmap. This is either the next node in the current | |
833 | * zone's radix tree or the first node in the radix tree of the | |
834 | * next zone. | |
835 | * | |
836 | * Return true if there is a next node, false otherwise. | |
837 | */ | |
838 | static bool rtree_next_node(struct memory_bitmap *bm) | |
839 | { | |
840 | if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) { | |
841 | bm->cur.node = list_entry(bm->cur.node->list.next, | |
842 | struct rtree_node, list); | |
843 | bm->cur.node_pfn += BM_BITS_PER_BLOCK; | |
844 | bm->cur.node_bit = 0; | |
845 | touch_softlockup_watchdog(); | |
846 | return true; | |
847 | } | |
848 | ||
849 | /* No more nodes, goto next zone */ | |
850 | if (!list_is_last(&bm->cur.zone->list, &bm->zones)) { | |
851 | bm->cur.zone = list_entry(bm->cur.zone->list.next, | |
852 | struct mem_zone_bm_rtree, list); | |
853 | bm->cur.node = list_entry(bm->cur.zone->leaves.next, | |
854 | struct rtree_node, list); | |
855 | bm->cur.node_pfn = 0; | |
856 | bm->cur.node_bit = 0; | |
857 | return true; | |
858 | } | |
859 | ||
860 | /* No more zones */ | |
861 | return false; | |
862 | } | |
863 | ||
864 | /** | |
865 | * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap. | |
866 | * @bm: Memory bitmap. | |
867 | * | |
868 | * Starting from the last returned position this function searches for the next | |
869 | * set bit in @bm and returns the PFN represented by it. If no more bits are | |
870 | * set, BM_END_OF_MAP is returned. | |
871 | * | |
872 | * It is required to run memory_bm_position_reset() before the first call to | |
873 | * this function for the given memory bitmap. | |
874 | */ | |
875 | static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm) | |
876 | { | |
877 | unsigned long bits, pfn, pages; | |
878 | int bit; | |
879 | ||
880 | do { | |
881 | pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn; | |
882 | bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK); | |
883 | bit = find_next_bit(bm->cur.node->data, bits, | |
884 | bm->cur.node_bit); | |
885 | if (bit < bits) { | |
886 | pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit; | |
887 | bm->cur.node_bit = bit + 1; | |
888 | return pfn; | |
889 | } | |
890 | } while (rtree_next_node(bm)); | |
891 | ||
892 | return BM_END_OF_MAP; | |
893 | } | |
894 | ||
895 | /* | |
896 | * This structure represents a range of page frames the contents of which | |
897 | * should not be saved during hibernation. | |
898 | */ | |
899 | struct nosave_region { | |
900 | struct list_head list; | |
901 | unsigned long start_pfn; | |
902 | unsigned long end_pfn; | |
903 | }; | |
904 | ||
905 | static LIST_HEAD(nosave_regions); | |
906 | ||
907 | static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone) | |
908 | { | |
909 | struct rtree_node *node; | |
910 | ||
911 | list_for_each_entry(node, &zone->nodes, list) | |
912 | recycle_safe_page(node->data); | |
913 | ||
914 | list_for_each_entry(node, &zone->leaves, list) | |
915 | recycle_safe_page(node->data); | |
916 | } | |
917 | ||
918 | static void memory_bm_recycle(struct memory_bitmap *bm) | |
919 | { | |
920 | struct mem_zone_bm_rtree *zone; | |
921 | struct linked_page *p_list; | |
922 | ||
923 | list_for_each_entry(zone, &bm->zones, list) | |
924 | recycle_zone_bm_rtree(zone); | |
925 | ||
926 | p_list = bm->p_list; | |
927 | while (p_list) { | |
928 | struct linked_page *lp = p_list; | |
929 | ||
930 | p_list = lp->next; | |
931 | recycle_safe_page(lp); | |
932 | } | |
933 | } | |
934 | ||
935 | /** | |
936 | * register_nosave_region - Register a region of unsaveable memory. | |
937 | * | |
938 | * Register a range of page frames the contents of which should not be saved | |
939 | * during hibernation (to be used in the early initialization code). | |
940 | */ | |
941 | void __init __register_nosave_region(unsigned long start_pfn, | |
942 | unsigned long end_pfn, int use_kmalloc) | |
943 | { | |
944 | struct nosave_region *region; | |
945 | ||
946 | if (start_pfn >= end_pfn) | |
947 | return; | |
948 | ||
949 | if (!list_empty(&nosave_regions)) { | |
950 | /* Try to extend the previous region (they should be sorted) */ | |
951 | region = list_entry(nosave_regions.prev, | |
952 | struct nosave_region, list); | |
953 | if (region->end_pfn == start_pfn) { | |
954 | region->end_pfn = end_pfn; | |
955 | goto Report; | |
956 | } | |
957 | } | |
958 | if (use_kmalloc) { | |
959 | /* During init, this shouldn't fail */ | |
960 | region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL); | |
961 | BUG_ON(!region); | |
962 | } else { | |
963 | /* This allocation cannot fail */ | |
964 | region = memblock_virt_alloc(sizeof(struct nosave_region), 0); | |
965 | } | |
966 | region->start_pfn = start_pfn; | |
967 | region->end_pfn = end_pfn; | |
968 | list_add_tail(®ion->list, &nosave_regions); | |
969 | Report: | |
970 | printk(KERN_INFO "PM: Registered nosave memory: [mem %#010llx-%#010llx]\n", | |
971 | (unsigned long long) start_pfn << PAGE_SHIFT, | |
972 | ((unsigned long long) end_pfn << PAGE_SHIFT) - 1); | |
973 | } | |
974 | ||
975 | /* | |
976 | * Set bits in this map correspond to the page frames the contents of which | |
977 | * should not be saved during the suspend. | |
978 | */ | |
979 | static struct memory_bitmap *forbidden_pages_map; | |
980 | ||
981 | /* Set bits in this map correspond to free page frames. */ | |
982 | static struct memory_bitmap *free_pages_map; | |
983 | ||
984 | /* | |
985 | * Each page frame allocated for creating the image is marked by setting the | |
986 | * corresponding bits in forbidden_pages_map and free_pages_map simultaneously | |
987 | */ | |
988 | ||
989 | void swsusp_set_page_free(struct page *page) | |
990 | { | |
991 | if (free_pages_map) | |
992 | memory_bm_set_bit(free_pages_map, page_to_pfn(page)); | |
993 | } | |
994 | ||
995 | static int swsusp_page_is_free(struct page *page) | |
996 | { | |
997 | return free_pages_map ? | |
998 | memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0; | |
999 | } | |
1000 | ||
1001 | void swsusp_unset_page_free(struct page *page) | |
1002 | { | |
1003 | if (free_pages_map) | |
1004 | memory_bm_clear_bit(free_pages_map, page_to_pfn(page)); | |
1005 | } | |
1006 | ||
1007 | static void swsusp_set_page_forbidden(struct page *page) | |
1008 | { | |
1009 | if (forbidden_pages_map) | |
1010 | memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page)); | |
1011 | } | |
1012 | ||
1013 | int swsusp_page_is_forbidden(struct page *page) | |
1014 | { | |
1015 | return forbidden_pages_map ? | |
1016 | memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0; | |
1017 | } | |
1018 | ||
1019 | static void swsusp_unset_page_forbidden(struct page *page) | |
1020 | { | |
1021 | if (forbidden_pages_map) | |
1022 | memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page)); | |
1023 | } | |
1024 | ||
1025 | /** | |
1026 | * mark_nosave_pages - Mark pages that should not be saved. | |
1027 | * @bm: Memory bitmap. | |
1028 | * | |
1029 | * Set the bits in @bm that correspond to the page frames the contents of which | |
1030 | * should not be saved. | |
1031 | */ | |
1032 | static void mark_nosave_pages(struct memory_bitmap *bm) | |
1033 | { | |
1034 | struct nosave_region *region; | |
1035 | ||
1036 | if (list_empty(&nosave_regions)) | |
1037 | return; | |
1038 | ||
1039 | list_for_each_entry(region, &nosave_regions, list) { | |
1040 | unsigned long pfn; | |
1041 | ||
1042 | pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n", | |
1043 | (unsigned long long) region->start_pfn << PAGE_SHIFT, | |
1044 | ((unsigned long long) region->end_pfn << PAGE_SHIFT) | |
1045 | - 1); | |
1046 | ||
1047 | for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++) | |
1048 | if (pfn_valid(pfn)) { | |
1049 | /* | |
1050 | * It is safe to ignore the result of | |
1051 | * mem_bm_set_bit_check() here, since we won't | |
1052 | * touch the PFNs for which the error is | |
1053 | * returned anyway. | |
1054 | */ | |
1055 | mem_bm_set_bit_check(bm, pfn); | |
1056 | } | |
1057 | } | |
1058 | } | |
1059 | ||
1060 | /** | |
1061 | * create_basic_memory_bitmaps - Create bitmaps to hold basic page information. | |
1062 | * | |
1063 | * Create bitmaps needed for marking page frames that should not be saved and | |
1064 | * free page frames. The forbidden_pages_map and free_pages_map pointers are | |
1065 | * only modified if everything goes well, because we don't want the bits to be | |
1066 | * touched before both bitmaps are set up. | |
1067 | */ | |
1068 | int create_basic_memory_bitmaps(void) | |
1069 | { | |
1070 | struct memory_bitmap *bm1, *bm2; | |
1071 | int error = 0; | |
1072 | ||
1073 | if (forbidden_pages_map && free_pages_map) | |
1074 | return 0; | |
1075 | else | |
1076 | BUG_ON(forbidden_pages_map || free_pages_map); | |
1077 | ||
1078 | bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); | |
1079 | if (!bm1) | |
1080 | return -ENOMEM; | |
1081 | ||
1082 | error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY); | |
1083 | if (error) | |
1084 | goto Free_first_object; | |
1085 | ||
1086 | bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL); | |
1087 | if (!bm2) | |
1088 | goto Free_first_bitmap; | |
1089 | ||
1090 | error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY); | |
1091 | if (error) | |
1092 | goto Free_second_object; | |
1093 | ||
1094 | forbidden_pages_map = bm1; | |
1095 | free_pages_map = bm2; | |
1096 | mark_nosave_pages(forbidden_pages_map); | |
1097 | ||
1098 | pr_debug("PM: Basic memory bitmaps created\n"); | |
1099 | ||
1100 | return 0; | |
1101 | ||
1102 | Free_second_object: | |
1103 | kfree(bm2); | |
1104 | Free_first_bitmap: | |
1105 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
1106 | Free_first_object: | |
1107 | kfree(bm1); | |
1108 | return -ENOMEM; | |
1109 | } | |
1110 | ||
1111 | /** | |
1112 | * free_basic_memory_bitmaps - Free memory bitmaps holding basic information. | |
1113 | * | |
1114 | * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The | |
1115 | * auxiliary pointers are necessary so that the bitmaps themselves are not | |
1116 | * referred to while they are being freed. | |
1117 | */ | |
1118 | void free_basic_memory_bitmaps(void) | |
1119 | { | |
1120 | struct memory_bitmap *bm1, *bm2; | |
1121 | ||
1122 | if (WARN_ON(!(forbidden_pages_map && free_pages_map))) | |
1123 | return; | |
1124 | ||
1125 | bm1 = forbidden_pages_map; | |
1126 | bm2 = free_pages_map; | |
1127 | forbidden_pages_map = NULL; | |
1128 | free_pages_map = NULL; | |
1129 | memory_bm_free(bm1, PG_UNSAFE_CLEAR); | |
1130 | kfree(bm1); | |
1131 | memory_bm_free(bm2, PG_UNSAFE_CLEAR); | |
1132 | kfree(bm2); | |
1133 | ||
1134 | pr_debug("PM: Basic memory bitmaps freed\n"); | |
1135 | } | |
1136 | ||
1137 | void clear_free_pages(void) | |
1138 | { | |
1139 | #ifdef CONFIG_PAGE_POISONING_ZERO | |
1140 | struct memory_bitmap *bm = free_pages_map; | |
1141 | unsigned long pfn; | |
1142 | ||
1143 | if (WARN_ON(!(free_pages_map))) | |
1144 | return; | |
1145 | ||
1146 | memory_bm_position_reset(bm); | |
1147 | pfn = memory_bm_next_pfn(bm); | |
1148 | while (pfn != BM_END_OF_MAP) { | |
1149 | if (pfn_valid(pfn)) | |
1150 | clear_highpage(pfn_to_page(pfn)); | |
1151 | ||
1152 | pfn = memory_bm_next_pfn(bm); | |
1153 | } | |
1154 | memory_bm_position_reset(bm); | |
1155 | pr_info("PM: free pages cleared after restore\n"); | |
1156 | #endif /* PAGE_POISONING_ZERO */ | |
1157 | } | |
1158 | ||
1159 | /** | |
1160 | * snapshot_additional_pages - Estimate the number of extra pages needed. | |
1161 | * @zone: Memory zone to carry out the computation for. | |
1162 | * | |
1163 | * Estimate the number of additional pages needed for setting up a hibernation | |
1164 | * image data structures for @zone (usually, the returned value is greater than | |
1165 | * the exact number). | |
1166 | */ | |
1167 | unsigned int snapshot_additional_pages(struct zone *zone) | |
1168 | { | |
1169 | unsigned int rtree, nodes; | |
1170 | ||
1171 | rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK); | |
1172 | rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node), | |
1173 | LINKED_PAGE_DATA_SIZE); | |
1174 | while (nodes > 1) { | |
1175 | nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL); | |
1176 | rtree += nodes; | |
1177 | } | |
1178 | ||
1179 | return 2 * rtree; | |
1180 | } | |
1181 | ||
1182 | #ifdef CONFIG_HIGHMEM | |
1183 | /** | |
1184 | * count_free_highmem_pages - Compute the total number of free highmem pages. | |
1185 | * | |
1186 | * The returned number is system-wide. | |
1187 | */ | |
1188 | static unsigned int count_free_highmem_pages(void) | |
1189 | { | |
1190 | struct zone *zone; | |
1191 | unsigned int cnt = 0; | |
1192 | ||
1193 | for_each_populated_zone(zone) | |
1194 | if (is_highmem(zone)) | |
1195 | cnt += zone_page_state(zone, NR_FREE_PAGES); | |
1196 | ||
1197 | return cnt; | |
1198 | } | |
1199 | ||
1200 | /** | |
1201 | * saveable_highmem_page - Check if a highmem page is saveable. | |
1202 | * | |
1203 | * Determine whether a highmem page should be included in a hibernation image. | |
1204 | * | |
1205 | * We should save the page if it isn't Nosave or NosaveFree, or Reserved, | |
1206 | * and it isn't part of a free chunk of pages. | |
1207 | */ | |
1208 | static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn) | |
1209 | { | |
1210 | struct page *page; | |
1211 | ||
1212 | if (!pfn_valid(pfn)) | |
1213 | return NULL; | |
1214 | ||
1215 | page = pfn_to_page(pfn); | |
1216 | if (page_zone(page) != zone) | |
1217 | return NULL; | |
1218 | ||
1219 | BUG_ON(!PageHighMem(page)); | |
1220 | ||
1221 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) || | |
1222 | PageReserved(page)) | |
1223 | return NULL; | |
1224 | ||
1225 | if (page_is_guard(page)) | |
1226 | return NULL; | |
1227 | ||
1228 | return page; | |
1229 | } | |
1230 | ||
1231 | /** | |
1232 | * count_highmem_pages - Compute the total number of saveable highmem pages. | |
1233 | */ | |
1234 | static unsigned int count_highmem_pages(void) | |
1235 | { | |
1236 | struct zone *zone; | |
1237 | unsigned int n = 0; | |
1238 | ||
1239 | for_each_populated_zone(zone) { | |
1240 | unsigned long pfn, max_zone_pfn; | |
1241 | ||
1242 | if (!is_highmem(zone)) | |
1243 | continue; | |
1244 | ||
1245 | mark_free_pages(zone); | |
1246 | max_zone_pfn = zone_end_pfn(zone); | |
1247 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1248 | if (saveable_highmem_page(zone, pfn)) | |
1249 | n++; | |
1250 | } | |
1251 | return n; | |
1252 | } | |
1253 | #else | |
1254 | static inline void *saveable_highmem_page(struct zone *z, unsigned long p) | |
1255 | { | |
1256 | return NULL; | |
1257 | } | |
1258 | #endif /* CONFIG_HIGHMEM */ | |
1259 | ||
1260 | /** | |
1261 | * saveable_page - Check if the given page is saveable. | |
1262 | * | |
1263 | * Determine whether a non-highmem page should be included in a hibernation | |
1264 | * image. | |
1265 | * | |
1266 | * We should save the page if it isn't Nosave, and is not in the range | |
1267 | * of pages statically defined as 'unsaveable', and it isn't part of | |
1268 | * a free chunk of pages. | |
1269 | */ | |
1270 | static struct page *saveable_page(struct zone *zone, unsigned long pfn) | |
1271 | { | |
1272 | struct page *page; | |
1273 | ||
1274 | if (!pfn_valid(pfn)) | |
1275 | return NULL; | |
1276 | ||
1277 | page = pfn_to_page(pfn); | |
1278 | if (page_zone(page) != zone) | |
1279 | return NULL; | |
1280 | ||
1281 | BUG_ON(PageHighMem(page)); | |
1282 | ||
1283 | if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page)) | |
1284 | return NULL; | |
1285 | ||
1286 | if (PageReserved(page) | |
1287 | && (!kernel_page_present(page) || pfn_is_nosave(pfn))) | |
1288 | return NULL; | |
1289 | ||
1290 | if (page_is_guard(page)) | |
1291 | return NULL; | |
1292 | ||
1293 | return page; | |
1294 | } | |
1295 | ||
1296 | /** | |
1297 | * count_data_pages - Compute the total number of saveable non-highmem pages. | |
1298 | */ | |
1299 | static unsigned int count_data_pages(void) | |
1300 | { | |
1301 | struct zone *zone; | |
1302 | unsigned long pfn, max_zone_pfn; | |
1303 | unsigned int n = 0; | |
1304 | ||
1305 | for_each_populated_zone(zone) { | |
1306 | if (is_highmem(zone)) | |
1307 | continue; | |
1308 | ||
1309 | mark_free_pages(zone); | |
1310 | max_zone_pfn = zone_end_pfn(zone); | |
1311 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1312 | if (saveable_page(zone, pfn)) | |
1313 | n++; | |
1314 | } | |
1315 | return n; | |
1316 | } | |
1317 | ||
1318 | /* | |
1319 | * This is needed, because copy_page and memcpy are not usable for copying | |
1320 | * task structs. | |
1321 | */ | |
1322 | static inline void do_copy_page(long *dst, long *src) | |
1323 | { | |
1324 | int n; | |
1325 | ||
1326 | for (n = PAGE_SIZE / sizeof(long); n; n--) | |
1327 | *dst++ = *src++; | |
1328 | } | |
1329 | ||
1330 | /** | |
1331 | * safe_copy_page - Copy a page in a safe way. | |
1332 | * | |
1333 | * Check if the page we are going to copy is marked as present in the kernel | |
1334 | * page tables (this always is the case if CONFIG_DEBUG_PAGEALLOC is not set | |
1335 | * and in that case kernel_page_present() always returns 'true'). | |
1336 | */ | |
1337 | static void safe_copy_page(void *dst, struct page *s_page) | |
1338 | { | |
1339 | if (kernel_page_present(s_page)) { | |
1340 | do_copy_page(dst, page_address(s_page)); | |
1341 | } else { | |
1342 | kernel_map_pages(s_page, 1, 1); | |
1343 | do_copy_page(dst, page_address(s_page)); | |
1344 | kernel_map_pages(s_page, 1, 0); | |
1345 | } | |
1346 | } | |
1347 | ||
1348 | #ifdef CONFIG_HIGHMEM | |
1349 | static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn) | |
1350 | { | |
1351 | return is_highmem(zone) ? | |
1352 | saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn); | |
1353 | } | |
1354 | ||
1355 | static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) | |
1356 | { | |
1357 | struct page *s_page, *d_page; | |
1358 | void *src, *dst; | |
1359 | ||
1360 | s_page = pfn_to_page(src_pfn); | |
1361 | d_page = pfn_to_page(dst_pfn); | |
1362 | if (PageHighMem(s_page)) { | |
1363 | src = kmap_atomic(s_page); | |
1364 | dst = kmap_atomic(d_page); | |
1365 | do_copy_page(dst, src); | |
1366 | kunmap_atomic(dst); | |
1367 | kunmap_atomic(src); | |
1368 | } else { | |
1369 | if (PageHighMem(d_page)) { | |
1370 | /* | |
1371 | * The page pointed to by src may contain some kernel | |
1372 | * data modified by kmap_atomic() | |
1373 | */ | |
1374 | safe_copy_page(buffer, s_page); | |
1375 | dst = kmap_atomic(d_page); | |
1376 | copy_page(dst, buffer); | |
1377 | kunmap_atomic(dst); | |
1378 | } else { | |
1379 | safe_copy_page(page_address(d_page), s_page); | |
1380 | } | |
1381 | } | |
1382 | } | |
1383 | #else | |
1384 | #define page_is_saveable(zone, pfn) saveable_page(zone, pfn) | |
1385 | ||
1386 | static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn) | |
1387 | { | |
1388 | safe_copy_page(page_address(pfn_to_page(dst_pfn)), | |
1389 | pfn_to_page(src_pfn)); | |
1390 | } | |
1391 | #endif /* CONFIG_HIGHMEM */ | |
1392 | ||
1393 | static void copy_data_pages(struct memory_bitmap *copy_bm, | |
1394 | struct memory_bitmap *orig_bm) | |
1395 | { | |
1396 | struct zone *zone; | |
1397 | unsigned long pfn; | |
1398 | ||
1399 | for_each_populated_zone(zone) { | |
1400 | unsigned long max_zone_pfn; | |
1401 | ||
1402 | mark_free_pages(zone); | |
1403 | max_zone_pfn = zone_end_pfn(zone); | |
1404 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) | |
1405 | if (page_is_saveable(zone, pfn)) | |
1406 | memory_bm_set_bit(orig_bm, pfn); | |
1407 | } | |
1408 | memory_bm_position_reset(orig_bm); | |
1409 | memory_bm_position_reset(copy_bm); | |
1410 | for(;;) { | |
1411 | pfn = memory_bm_next_pfn(orig_bm); | |
1412 | if (unlikely(pfn == BM_END_OF_MAP)) | |
1413 | break; | |
1414 | copy_data_page(memory_bm_next_pfn(copy_bm), pfn); | |
1415 | } | |
1416 | } | |
1417 | ||
1418 | /* Total number of image pages */ | |
1419 | static unsigned int nr_copy_pages; | |
1420 | /* Number of pages needed for saving the original pfns of the image pages */ | |
1421 | static unsigned int nr_meta_pages; | |
1422 | /* | |
1423 | * Numbers of normal and highmem page frames allocated for hibernation image | |
1424 | * before suspending devices. | |
1425 | */ | |
1426 | static unsigned int alloc_normal, alloc_highmem; | |
1427 | /* | |
1428 | * Memory bitmap used for marking saveable pages (during hibernation) or | |
1429 | * hibernation image pages (during restore) | |
1430 | */ | |
1431 | static struct memory_bitmap orig_bm; | |
1432 | /* | |
1433 | * Memory bitmap used during hibernation for marking allocated page frames that | |
1434 | * will contain copies of saveable pages. During restore it is initially used | |
1435 | * for marking hibernation image pages, but then the set bits from it are | |
1436 | * duplicated in @orig_bm and it is released. On highmem systems it is next | |
1437 | * used for marking "safe" highmem pages, but it has to be reinitialized for | |
1438 | * this purpose. | |
1439 | */ | |
1440 | static struct memory_bitmap copy_bm; | |
1441 | ||
1442 | /** | |
1443 | * swsusp_free - Free pages allocated for hibernation image. | |
1444 | * | |
1445 | * Image pages are alocated before snapshot creation, so they need to be | |
1446 | * released after resume. | |
1447 | */ | |
1448 | void swsusp_free(void) | |
1449 | { | |
1450 | unsigned long fb_pfn, fr_pfn; | |
1451 | ||
1452 | if (!forbidden_pages_map || !free_pages_map) | |
1453 | goto out; | |
1454 | ||
1455 | memory_bm_position_reset(forbidden_pages_map); | |
1456 | memory_bm_position_reset(free_pages_map); | |
1457 | ||
1458 | loop: | |
1459 | fr_pfn = memory_bm_next_pfn(free_pages_map); | |
1460 | fb_pfn = memory_bm_next_pfn(forbidden_pages_map); | |
1461 | ||
1462 | /* | |
1463 | * Find the next bit set in both bitmaps. This is guaranteed to | |
1464 | * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP. | |
1465 | */ | |
1466 | do { | |
1467 | if (fb_pfn < fr_pfn) | |
1468 | fb_pfn = memory_bm_next_pfn(forbidden_pages_map); | |
1469 | if (fr_pfn < fb_pfn) | |
1470 | fr_pfn = memory_bm_next_pfn(free_pages_map); | |
1471 | } while (fb_pfn != fr_pfn); | |
1472 | ||
1473 | if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) { | |
1474 | struct page *page = pfn_to_page(fr_pfn); | |
1475 | ||
1476 | memory_bm_clear_current(forbidden_pages_map); | |
1477 | memory_bm_clear_current(free_pages_map); | |
1478 | hibernate_restore_unprotect_page(page_address(page)); | |
1479 | __free_page(page); | |
1480 | goto loop; | |
1481 | } | |
1482 | ||
1483 | out: | |
1484 | nr_copy_pages = 0; | |
1485 | nr_meta_pages = 0; | |
1486 | restore_pblist = NULL; | |
1487 | buffer = NULL; | |
1488 | alloc_normal = 0; | |
1489 | alloc_highmem = 0; | |
1490 | hibernate_restore_protection_end(); | |
1491 | } | |
1492 | ||
1493 | /* Helper functions used for the shrinking of memory. */ | |
1494 | ||
1495 | #define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN) | |
1496 | ||
1497 | /** | |
1498 | * preallocate_image_pages - Allocate a number of pages for hibernation image. | |
1499 | * @nr_pages: Number of page frames to allocate. | |
1500 | * @mask: GFP flags to use for the allocation. | |
1501 | * | |
1502 | * Return value: Number of page frames actually allocated | |
1503 | */ | |
1504 | static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask) | |
1505 | { | |
1506 | unsigned long nr_alloc = 0; | |
1507 | ||
1508 | while (nr_pages > 0) { | |
1509 | struct page *page; | |
1510 | ||
1511 | page = alloc_image_page(mask); | |
1512 | if (!page) | |
1513 | break; | |
1514 | memory_bm_set_bit(©_bm, page_to_pfn(page)); | |
1515 | if (PageHighMem(page)) | |
1516 | alloc_highmem++; | |
1517 | else | |
1518 | alloc_normal++; | |
1519 | nr_pages--; | |
1520 | nr_alloc++; | |
1521 | } | |
1522 | ||
1523 | return nr_alloc; | |
1524 | } | |
1525 | ||
1526 | static unsigned long preallocate_image_memory(unsigned long nr_pages, | |
1527 | unsigned long avail_normal) | |
1528 | { | |
1529 | unsigned long alloc; | |
1530 | ||
1531 | if (avail_normal <= alloc_normal) | |
1532 | return 0; | |
1533 | ||
1534 | alloc = avail_normal - alloc_normal; | |
1535 | if (nr_pages < alloc) | |
1536 | alloc = nr_pages; | |
1537 | ||
1538 | return preallocate_image_pages(alloc, GFP_IMAGE); | |
1539 | } | |
1540 | ||
1541 | #ifdef CONFIG_HIGHMEM | |
1542 | static unsigned long preallocate_image_highmem(unsigned long nr_pages) | |
1543 | { | |
1544 | return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM); | |
1545 | } | |
1546 | ||
1547 | /** | |
1548 | * __fraction - Compute (an approximation of) x * (multiplier / base). | |
1549 | */ | |
1550 | static unsigned long __fraction(u64 x, u64 multiplier, u64 base) | |
1551 | { | |
1552 | x *= multiplier; | |
1553 | do_div(x, base); | |
1554 | return (unsigned long)x; | |
1555 | } | |
1556 | ||
1557 | static unsigned long preallocate_highmem_fraction(unsigned long nr_pages, | |
1558 | unsigned long highmem, | |
1559 | unsigned long total) | |
1560 | { | |
1561 | unsigned long alloc = __fraction(nr_pages, highmem, total); | |
1562 | ||
1563 | return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM); | |
1564 | } | |
1565 | #else /* CONFIG_HIGHMEM */ | |
1566 | static inline unsigned long preallocate_image_highmem(unsigned long nr_pages) | |
1567 | { | |
1568 | return 0; | |
1569 | } | |
1570 | ||
1571 | static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages, | |
1572 | unsigned long highmem, | |
1573 | unsigned long total) | |
1574 | { | |
1575 | return 0; | |
1576 | } | |
1577 | #endif /* CONFIG_HIGHMEM */ | |
1578 | ||
1579 | /** | |
1580 | * free_unnecessary_pages - Release preallocated pages not needed for the image. | |
1581 | */ | |
1582 | static unsigned long free_unnecessary_pages(void) | |
1583 | { | |
1584 | unsigned long save, to_free_normal, to_free_highmem, free; | |
1585 | ||
1586 | save = count_data_pages(); | |
1587 | if (alloc_normal >= save) { | |
1588 | to_free_normal = alloc_normal - save; | |
1589 | save = 0; | |
1590 | } else { | |
1591 | to_free_normal = 0; | |
1592 | save -= alloc_normal; | |
1593 | } | |
1594 | save += count_highmem_pages(); | |
1595 | if (alloc_highmem >= save) { | |
1596 | to_free_highmem = alloc_highmem - save; | |
1597 | } else { | |
1598 | to_free_highmem = 0; | |
1599 | save -= alloc_highmem; | |
1600 | if (to_free_normal > save) | |
1601 | to_free_normal -= save; | |
1602 | else | |
1603 | to_free_normal = 0; | |
1604 | } | |
1605 | free = to_free_normal + to_free_highmem; | |
1606 | ||
1607 | memory_bm_position_reset(©_bm); | |
1608 | ||
1609 | while (to_free_normal > 0 || to_free_highmem > 0) { | |
1610 | unsigned long pfn = memory_bm_next_pfn(©_bm); | |
1611 | struct page *page = pfn_to_page(pfn); | |
1612 | ||
1613 | if (PageHighMem(page)) { | |
1614 | if (!to_free_highmem) | |
1615 | continue; | |
1616 | to_free_highmem--; | |
1617 | alloc_highmem--; | |
1618 | } else { | |
1619 | if (!to_free_normal) | |
1620 | continue; | |
1621 | to_free_normal--; | |
1622 | alloc_normal--; | |
1623 | } | |
1624 | memory_bm_clear_bit(©_bm, pfn); | |
1625 | swsusp_unset_page_forbidden(page); | |
1626 | swsusp_unset_page_free(page); | |
1627 | __free_page(page); | |
1628 | } | |
1629 | ||
1630 | return free; | |
1631 | } | |
1632 | ||
1633 | /** | |
1634 | * minimum_image_size - Estimate the minimum acceptable size of an image. | |
1635 | * @saveable: Number of saveable pages in the system. | |
1636 | * | |
1637 | * We want to avoid attempting to free too much memory too hard, so estimate the | |
1638 | * minimum acceptable size of a hibernation image to use as the lower limit for | |
1639 | * preallocating memory. | |
1640 | * | |
1641 | * We assume that the minimum image size should be proportional to | |
1642 | * | |
1643 | * [number of saveable pages] - [number of pages that can be freed in theory] | |
1644 | * | |
1645 | * where the second term is the sum of (1) reclaimable slab pages, (2) active | |
1646 | * and (3) inactive anonymous pages, (4) active and (5) inactive file pages, | |
1647 | * minus mapped file pages. | |
1648 | */ | |
1649 | static unsigned long minimum_image_size(unsigned long saveable) | |
1650 | { | |
1651 | unsigned long size; | |
1652 | ||
1653 | size = global_node_page_state(NR_SLAB_RECLAIMABLE) | |
1654 | + global_node_page_state(NR_ACTIVE_ANON) | |
1655 | + global_node_page_state(NR_INACTIVE_ANON) | |
1656 | + global_node_page_state(NR_ACTIVE_FILE) | |
1657 | + global_node_page_state(NR_INACTIVE_FILE) | |
1658 | - global_node_page_state(NR_FILE_MAPPED); | |
1659 | ||
1660 | return saveable <= size ? 0 : saveable - size; | |
1661 | } | |
1662 | ||
1663 | /** | |
1664 | * hibernate_preallocate_memory - Preallocate memory for hibernation image. | |
1665 | * | |
1666 | * To create a hibernation image it is necessary to make a copy of every page | |
1667 | * frame in use. We also need a number of page frames to be free during | |
1668 | * hibernation for allocations made while saving the image and for device | |
1669 | * drivers, in case they need to allocate memory from their hibernation | |
1670 | * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough | |
1671 | * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through | |
1672 | * /sys/power/reserved_size, respectively). To make this happen, we compute the | |
1673 | * total number of available page frames and allocate at least | |
1674 | * | |
1675 | * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2 | |
1676 | * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE) | |
1677 | * | |
1678 | * of them, which corresponds to the maximum size of a hibernation image. | |
1679 | * | |
1680 | * If image_size is set below the number following from the above formula, | |
1681 | * the preallocation of memory is continued until the total number of saveable | |
1682 | * pages in the system is below the requested image size or the minimum | |
1683 | * acceptable image size returned by minimum_image_size(), whichever is greater. | |
1684 | */ | |
1685 | int hibernate_preallocate_memory(void) | |
1686 | { | |
1687 | struct zone *zone; | |
1688 | unsigned long saveable, size, max_size, count, highmem, pages = 0; | |
1689 | unsigned long alloc, save_highmem, pages_highmem, avail_normal; | |
1690 | ktime_t start, stop; | |
1691 | int error; | |
1692 | ||
1693 | printk(KERN_INFO "PM: Preallocating image memory... "); | |
1694 | start = ktime_get(); | |
1695 | ||
1696 | error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY); | |
1697 | if (error) | |
1698 | goto err_out; | |
1699 | ||
1700 | error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY); | |
1701 | if (error) | |
1702 | goto err_out; | |
1703 | ||
1704 | alloc_normal = 0; | |
1705 | alloc_highmem = 0; | |
1706 | ||
1707 | /* Count the number of saveable data pages. */ | |
1708 | save_highmem = count_highmem_pages(); | |
1709 | saveable = count_data_pages(); | |
1710 | ||
1711 | /* | |
1712 | * Compute the total number of page frames we can use (count) and the | |
1713 | * number of pages needed for image metadata (size). | |
1714 | */ | |
1715 | count = saveable; | |
1716 | saveable += save_highmem; | |
1717 | highmem = save_highmem; | |
1718 | size = 0; | |
1719 | for_each_populated_zone(zone) { | |
1720 | size += snapshot_additional_pages(zone); | |
1721 | if (is_highmem(zone)) | |
1722 | highmem += zone_page_state(zone, NR_FREE_PAGES); | |
1723 | else | |
1724 | count += zone_page_state(zone, NR_FREE_PAGES); | |
1725 | } | |
1726 | avail_normal = count; | |
1727 | count += highmem; | |
1728 | count -= totalreserve_pages; | |
1729 | ||
1730 | /* Add number of pages required for page keys (s390 only). */ | |
1731 | size += page_key_additional_pages(saveable); | |
1732 | ||
1733 | /* Compute the maximum number of saveable pages to leave in memory. */ | |
1734 | max_size = (count - (size + PAGES_FOR_IO)) / 2 | |
1735 | - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE); | |
1736 | /* Compute the desired number of image pages specified by image_size. */ | |
1737 | size = DIV_ROUND_UP(image_size, PAGE_SIZE); | |
1738 | if (size > max_size) | |
1739 | size = max_size; | |
1740 | /* | |
1741 | * If the desired number of image pages is at least as large as the | |
1742 | * current number of saveable pages in memory, allocate page frames for | |
1743 | * the image and we're done. | |
1744 | */ | |
1745 | if (size >= saveable) { | |
1746 | pages = preallocate_image_highmem(save_highmem); | |
1747 | pages += preallocate_image_memory(saveable - pages, avail_normal); | |
1748 | goto out; | |
1749 | } | |
1750 | ||
1751 | /* Estimate the minimum size of the image. */ | |
1752 | pages = minimum_image_size(saveable); | |
1753 | /* | |
1754 | * To avoid excessive pressure on the normal zone, leave room in it to | |
1755 | * accommodate an image of the minimum size (unless it's already too | |
1756 | * small, in which case don't preallocate pages from it at all). | |
1757 | */ | |
1758 | if (avail_normal > pages) | |
1759 | avail_normal -= pages; | |
1760 | else | |
1761 | avail_normal = 0; | |
1762 | if (size < pages) | |
1763 | size = min_t(unsigned long, pages, max_size); | |
1764 | ||
1765 | /* | |
1766 | * Let the memory management subsystem know that we're going to need a | |
1767 | * large number of page frames to allocate and make it free some memory. | |
1768 | * NOTE: If this is not done, performance will be hurt badly in some | |
1769 | * test cases. | |
1770 | */ | |
1771 | shrink_all_memory(saveable - size); | |
1772 | ||
1773 | /* | |
1774 | * The number of saveable pages in memory was too high, so apply some | |
1775 | * pressure to decrease it. First, make room for the largest possible | |
1776 | * image and fail if that doesn't work. Next, try to decrease the size | |
1777 | * of the image as much as indicated by 'size' using allocations from | |
1778 | * highmem and non-highmem zones separately. | |
1779 | */ | |
1780 | pages_highmem = preallocate_image_highmem(highmem / 2); | |
1781 | alloc = count - max_size; | |
1782 | if (alloc > pages_highmem) | |
1783 | alloc -= pages_highmem; | |
1784 | else | |
1785 | alloc = 0; | |
1786 | pages = preallocate_image_memory(alloc, avail_normal); | |
1787 | if (pages < alloc) { | |
1788 | /* We have exhausted non-highmem pages, try highmem. */ | |
1789 | alloc -= pages; | |
1790 | pages += pages_highmem; | |
1791 | pages_highmem = preallocate_image_highmem(alloc); | |
1792 | if (pages_highmem < alloc) | |
1793 | goto err_out; | |
1794 | pages += pages_highmem; | |
1795 | /* | |
1796 | * size is the desired number of saveable pages to leave in | |
1797 | * memory, so try to preallocate (all memory - size) pages. | |
1798 | */ | |
1799 | alloc = (count - pages) - size; | |
1800 | pages += preallocate_image_highmem(alloc); | |
1801 | } else { | |
1802 | /* | |
1803 | * There are approximately max_size saveable pages at this point | |
1804 | * and we want to reduce this number down to size. | |
1805 | */ | |
1806 | alloc = max_size - size; | |
1807 | size = preallocate_highmem_fraction(alloc, highmem, count); | |
1808 | pages_highmem += size; | |
1809 | alloc -= size; | |
1810 | size = preallocate_image_memory(alloc, avail_normal); | |
1811 | pages_highmem += preallocate_image_highmem(alloc - size); | |
1812 | pages += pages_highmem + size; | |
1813 | } | |
1814 | ||
1815 | /* | |
1816 | * We only need as many page frames for the image as there are saveable | |
1817 | * pages in memory, but we have allocated more. Release the excessive | |
1818 | * ones now. | |
1819 | */ | |
1820 | pages -= free_unnecessary_pages(); | |
1821 | ||
1822 | out: | |
1823 | stop = ktime_get(); | |
1824 | printk(KERN_CONT "done (allocated %lu pages)\n", pages); | |
1825 | swsusp_show_speed(start, stop, pages, "Allocated"); | |
1826 | ||
1827 | return 0; | |
1828 | ||
1829 | err_out: | |
1830 | printk(KERN_CONT "\n"); | |
1831 | swsusp_free(); | |
1832 | return -ENOMEM; | |
1833 | } | |
1834 | ||
1835 | #ifdef CONFIG_HIGHMEM | |
1836 | /** | |
1837 | * count_pages_for_highmem - Count non-highmem pages needed for copying highmem. | |
1838 | * | |
1839 | * Compute the number of non-highmem pages that will be necessary for creating | |
1840 | * copies of highmem pages. | |
1841 | */ | |
1842 | static unsigned int count_pages_for_highmem(unsigned int nr_highmem) | |
1843 | { | |
1844 | unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem; | |
1845 | ||
1846 | if (free_highmem >= nr_highmem) | |
1847 | nr_highmem = 0; | |
1848 | else | |
1849 | nr_highmem -= free_highmem; | |
1850 | ||
1851 | return nr_highmem; | |
1852 | } | |
1853 | #else | |
1854 | static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; } | |
1855 | #endif /* CONFIG_HIGHMEM */ | |
1856 | ||
1857 | /** | |
1858 | * enough_free_mem - Check if there is enough free memory for the image. | |
1859 | */ | |
1860 | static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem) | |
1861 | { | |
1862 | struct zone *zone; | |
1863 | unsigned int free = alloc_normal; | |
1864 | ||
1865 | for_each_populated_zone(zone) | |
1866 | if (!is_highmem(zone)) | |
1867 | free += zone_page_state(zone, NR_FREE_PAGES); | |
1868 | ||
1869 | nr_pages += count_pages_for_highmem(nr_highmem); | |
1870 | pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n", | |
1871 | nr_pages, PAGES_FOR_IO, free); | |
1872 | ||
1873 | return free > nr_pages + PAGES_FOR_IO; | |
1874 | } | |
1875 | ||
1876 | #ifdef CONFIG_HIGHMEM | |
1877 | /** | |
1878 | * get_highmem_buffer - Allocate a buffer for highmem pages. | |
1879 | * | |
1880 | * If there are some highmem pages in the hibernation image, we may need a | |
1881 | * buffer to copy them and/or load their data. | |
1882 | */ | |
1883 | static inline int get_highmem_buffer(int safe_needed) | |
1884 | { | |
1885 | buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed); | |
1886 | return buffer ? 0 : -ENOMEM; | |
1887 | } | |
1888 | ||
1889 | /** | |
1890 | * alloc_highmem_image_pages - Allocate some highmem pages for the image. | |
1891 | * | |
1892 | * Try to allocate as many pages as needed, but if the number of free highmem | |
1893 | * pages is less than that, allocate them all. | |
1894 | */ | |
1895 | static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm, | |
1896 | unsigned int nr_highmem) | |
1897 | { | |
1898 | unsigned int to_alloc = count_free_highmem_pages(); | |
1899 | ||
1900 | if (to_alloc > nr_highmem) | |
1901 | to_alloc = nr_highmem; | |
1902 | ||
1903 | nr_highmem -= to_alloc; | |
1904 | while (to_alloc-- > 0) { | |
1905 | struct page *page; | |
1906 | ||
1907 | page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM); | |
1908 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
1909 | } | |
1910 | return nr_highmem; | |
1911 | } | |
1912 | #else | |
1913 | static inline int get_highmem_buffer(int safe_needed) { return 0; } | |
1914 | ||
1915 | static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm, | |
1916 | unsigned int n) { return 0; } | |
1917 | #endif /* CONFIG_HIGHMEM */ | |
1918 | ||
1919 | /** | |
1920 | * swsusp_alloc - Allocate memory for hibernation image. | |
1921 | * | |
1922 | * We first try to allocate as many highmem pages as there are | |
1923 | * saveable highmem pages in the system. If that fails, we allocate | |
1924 | * non-highmem pages for the copies of the remaining highmem ones. | |
1925 | * | |
1926 | * In this approach it is likely that the copies of highmem pages will | |
1927 | * also be located in the high memory, because of the way in which | |
1928 | * copy_data_pages() works. | |
1929 | */ | |
1930 | static int swsusp_alloc(struct memory_bitmap *copy_bm, | |
1931 | unsigned int nr_pages, unsigned int nr_highmem) | |
1932 | { | |
1933 | if (nr_highmem > 0) { | |
1934 | if (get_highmem_buffer(PG_ANY)) | |
1935 | goto err_out; | |
1936 | if (nr_highmem > alloc_highmem) { | |
1937 | nr_highmem -= alloc_highmem; | |
1938 | nr_pages += alloc_highmem_pages(copy_bm, nr_highmem); | |
1939 | } | |
1940 | } | |
1941 | if (nr_pages > alloc_normal) { | |
1942 | nr_pages -= alloc_normal; | |
1943 | while (nr_pages-- > 0) { | |
1944 | struct page *page; | |
1945 | ||
1946 | page = alloc_image_page(GFP_ATOMIC | __GFP_COLD); | |
1947 | if (!page) | |
1948 | goto err_out; | |
1949 | memory_bm_set_bit(copy_bm, page_to_pfn(page)); | |
1950 | } | |
1951 | } | |
1952 | ||
1953 | return 0; | |
1954 | ||
1955 | err_out: | |
1956 | swsusp_free(); | |
1957 | return -ENOMEM; | |
1958 | } | |
1959 | ||
1960 | asmlinkage __visible int swsusp_save(void) | |
1961 | { | |
1962 | unsigned int nr_pages, nr_highmem; | |
1963 | ||
1964 | printk(KERN_INFO "PM: Creating hibernation image:\n"); | |
1965 | ||
1966 | drain_local_pages(NULL); | |
1967 | nr_pages = count_data_pages(); | |
1968 | nr_highmem = count_highmem_pages(); | |
1969 | printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem); | |
1970 | ||
1971 | if (!enough_free_mem(nr_pages, nr_highmem)) { | |
1972 | printk(KERN_ERR "PM: Not enough free memory\n"); | |
1973 | return -ENOMEM; | |
1974 | } | |
1975 | ||
1976 | if (swsusp_alloc(©_bm, nr_pages, nr_highmem)) { | |
1977 | printk(KERN_ERR "PM: Memory allocation failed\n"); | |
1978 | return -ENOMEM; | |
1979 | } | |
1980 | ||
1981 | /* | |
1982 | * During allocating of suspend pagedir, new cold pages may appear. | |
1983 | * Kill them. | |
1984 | */ | |
1985 | drain_local_pages(NULL); | |
1986 | copy_data_pages(©_bm, &orig_bm); | |
1987 | ||
1988 | /* | |
1989 | * End of critical section. From now on, we can write to memory, | |
1990 | * but we should not touch disk. This specially means we must _not_ | |
1991 | * touch swap space! Except we must write out our image of course. | |
1992 | */ | |
1993 | ||
1994 | nr_pages += nr_highmem; | |
1995 | nr_copy_pages = nr_pages; | |
1996 | nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE); | |
1997 | ||
1998 | printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n", | |
1999 | nr_pages); | |
2000 | ||
2001 | return 0; | |
2002 | } | |
2003 | ||
2004 | #ifndef CONFIG_ARCH_HIBERNATION_HEADER | |
2005 | static int init_header_complete(struct swsusp_info *info) | |
2006 | { | |
2007 | memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname)); | |
2008 | info->version_code = LINUX_VERSION_CODE; | |
2009 | return 0; | |
2010 | } | |
2011 | ||
2012 | static char *check_image_kernel(struct swsusp_info *info) | |
2013 | { | |
2014 | if (info->version_code != LINUX_VERSION_CODE) | |
2015 | return "kernel version"; | |
2016 | if (strcmp(info->uts.sysname,init_utsname()->sysname)) | |
2017 | return "system type"; | |
2018 | if (strcmp(info->uts.release,init_utsname()->release)) | |
2019 | return "kernel release"; | |
2020 | if (strcmp(info->uts.version,init_utsname()->version)) | |
2021 | return "version"; | |
2022 | if (strcmp(info->uts.machine,init_utsname()->machine)) | |
2023 | return "machine"; | |
2024 | return NULL; | |
2025 | } | |
2026 | #endif /* CONFIG_ARCH_HIBERNATION_HEADER */ | |
2027 | ||
2028 | unsigned long snapshot_get_image_size(void) | |
2029 | { | |
2030 | return nr_copy_pages + nr_meta_pages + 1; | |
2031 | } | |
2032 | ||
2033 | static int init_header(struct swsusp_info *info) | |
2034 | { | |
2035 | memset(info, 0, sizeof(struct swsusp_info)); | |
2036 | info->num_physpages = get_num_physpages(); | |
2037 | info->image_pages = nr_copy_pages; | |
2038 | info->pages = snapshot_get_image_size(); | |
2039 | info->size = info->pages; | |
2040 | info->size <<= PAGE_SHIFT; | |
2041 | return init_header_complete(info); | |
2042 | } | |
2043 | ||
2044 | /** | |
2045 | * pack_pfns - Prepare PFNs for saving. | |
2046 | * @bm: Memory bitmap. | |
2047 | * @buf: Memory buffer to store the PFNs in. | |
2048 | * | |
2049 | * PFNs corresponding to set bits in @bm are stored in the area of memory | |
2050 | * pointed to by @buf (1 page at a time). | |
2051 | */ | |
2052 | static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm) | |
2053 | { | |
2054 | int j; | |
2055 | ||
2056 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { | |
2057 | buf[j] = memory_bm_next_pfn(bm); | |
2058 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
2059 | break; | |
2060 | /* Save page key for data page (s390 only). */ | |
2061 | page_key_read(buf + j); | |
2062 | } | |
2063 | } | |
2064 | ||
2065 | /** | |
2066 | * snapshot_read_next - Get the address to read the next image page from. | |
2067 | * @handle: Snapshot handle to be used for the reading. | |
2068 | * | |
2069 | * On the first call, @handle should point to a zeroed snapshot_handle | |
2070 | * structure. The structure gets populated then and a pointer to it should be | |
2071 | * passed to this function every next time. | |
2072 | * | |
2073 | * On success, the function returns a positive number. Then, the caller | |
2074 | * is allowed to read up to the returned number of bytes from the memory | |
2075 | * location computed by the data_of() macro. | |
2076 | * | |
2077 | * The function returns 0 to indicate the end of the data stream condition, | |
2078 | * and negative numbers are returned on errors. If that happens, the structure | |
2079 | * pointed to by @handle is not updated and should not be used any more. | |
2080 | */ | |
2081 | int snapshot_read_next(struct snapshot_handle *handle) | |
2082 | { | |
2083 | if (handle->cur > nr_meta_pages + nr_copy_pages) | |
2084 | return 0; | |
2085 | ||
2086 | if (!buffer) { | |
2087 | /* This makes the buffer be freed by swsusp_free() */ | |
2088 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); | |
2089 | if (!buffer) | |
2090 | return -ENOMEM; | |
2091 | } | |
2092 | if (!handle->cur) { | |
2093 | int error; | |
2094 | ||
2095 | error = init_header((struct swsusp_info *)buffer); | |
2096 | if (error) | |
2097 | return error; | |
2098 | handle->buffer = buffer; | |
2099 | memory_bm_position_reset(&orig_bm); | |
2100 | memory_bm_position_reset(©_bm); | |
2101 | } else if (handle->cur <= nr_meta_pages) { | |
2102 | clear_page(buffer); | |
2103 | pack_pfns(buffer, &orig_bm); | |
2104 | } else { | |
2105 | struct page *page; | |
2106 | ||
2107 | page = pfn_to_page(memory_bm_next_pfn(©_bm)); | |
2108 | if (PageHighMem(page)) { | |
2109 | /* | |
2110 | * Highmem pages are copied to the buffer, | |
2111 | * because we can't return with a kmapped | |
2112 | * highmem page (we may not be called again). | |
2113 | */ | |
2114 | void *kaddr; | |
2115 | ||
2116 | kaddr = kmap_atomic(page); | |
2117 | copy_page(buffer, kaddr); | |
2118 | kunmap_atomic(kaddr); | |
2119 | handle->buffer = buffer; | |
2120 | } else { | |
2121 | handle->buffer = page_address(page); | |
2122 | } | |
2123 | } | |
2124 | handle->cur++; | |
2125 | return PAGE_SIZE; | |
2126 | } | |
2127 | ||
2128 | static void duplicate_memory_bitmap(struct memory_bitmap *dst, | |
2129 | struct memory_bitmap *src) | |
2130 | { | |
2131 | unsigned long pfn; | |
2132 | ||
2133 | memory_bm_position_reset(src); | |
2134 | pfn = memory_bm_next_pfn(src); | |
2135 | while (pfn != BM_END_OF_MAP) { | |
2136 | memory_bm_set_bit(dst, pfn); | |
2137 | pfn = memory_bm_next_pfn(src); | |
2138 | } | |
2139 | } | |
2140 | ||
2141 | /** | |
2142 | * mark_unsafe_pages - Mark pages that were used before hibernation. | |
2143 | * | |
2144 | * Mark the pages that cannot be used for storing the image during restoration, | |
2145 | * because they conflict with the pages that had been used before hibernation. | |
2146 | */ | |
2147 | static void mark_unsafe_pages(struct memory_bitmap *bm) | |
2148 | { | |
2149 | unsigned long pfn; | |
2150 | ||
2151 | /* Clear the "free"/"unsafe" bit for all PFNs */ | |
2152 | memory_bm_position_reset(free_pages_map); | |
2153 | pfn = memory_bm_next_pfn(free_pages_map); | |
2154 | while (pfn != BM_END_OF_MAP) { | |
2155 | memory_bm_clear_current(free_pages_map); | |
2156 | pfn = memory_bm_next_pfn(free_pages_map); | |
2157 | } | |
2158 | ||
2159 | /* Mark pages that correspond to the "original" PFNs as "unsafe" */ | |
2160 | duplicate_memory_bitmap(free_pages_map, bm); | |
2161 | ||
2162 | allocated_unsafe_pages = 0; | |
2163 | } | |
2164 | ||
2165 | static int check_header(struct swsusp_info *info) | |
2166 | { | |
2167 | char *reason; | |
2168 | ||
2169 | reason = check_image_kernel(info); | |
2170 | if (!reason && info->num_physpages != get_num_physpages()) | |
2171 | reason = "memory size"; | |
2172 | if (reason) { | |
2173 | printk(KERN_ERR "PM: Image mismatch: %s\n", reason); | |
2174 | return -EPERM; | |
2175 | } | |
2176 | return 0; | |
2177 | } | |
2178 | ||
2179 | /** | |
2180 | * load header - Check the image header and copy the data from it. | |
2181 | */ | |
2182 | static int load_header(struct swsusp_info *info) | |
2183 | { | |
2184 | int error; | |
2185 | ||
2186 | restore_pblist = NULL; | |
2187 | error = check_header(info); | |
2188 | if (!error) { | |
2189 | nr_copy_pages = info->image_pages; | |
2190 | nr_meta_pages = info->pages - info->image_pages - 1; | |
2191 | } | |
2192 | return error; | |
2193 | } | |
2194 | ||
2195 | /** | |
2196 | * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap. | |
2197 | * @bm: Memory bitmap. | |
2198 | * @buf: Area of memory containing the PFNs. | |
2199 | * | |
2200 | * For each element of the array pointed to by @buf (1 page at a time), set the | |
2201 | * corresponding bit in @bm. | |
2202 | */ | |
2203 | static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm) | |
2204 | { | |
2205 | int j; | |
2206 | ||
2207 | for (j = 0; j < PAGE_SIZE / sizeof(long); j++) { | |
2208 | if (unlikely(buf[j] == BM_END_OF_MAP)) | |
2209 | break; | |
2210 | ||
2211 | /* Extract and buffer page key for data page (s390 only). */ | |
2212 | page_key_memorize(buf + j); | |
2213 | ||
2214 | if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j])) | |
2215 | memory_bm_set_bit(bm, buf[j]); | |
2216 | else | |
2217 | return -EFAULT; | |
2218 | } | |
2219 | ||
2220 | return 0; | |
2221 | } | |
2222 | ||
2223 | #ifdef CONFIG_HIGHMEM | |
2224 | /* | |
2225 | * struct highmem_pbe is used for creating the list of highmem pages that | |
2226 | * should be restored atomically during the resume from disk, because the page | |
2227 | * frames they have occupied before the suspend are in use. | |
2228 | */ | |
2229 | struct highmem_pbe { | |
2230 | struct page *copy_page; /* data is here now */ | |
2231 | struct page *orig_page; /* data was here before the suspend */ | |
2232 | struct highmem_pbe *next; | |
2233 | }; | |
2234 | ||
2235 | /* | |
2236 | * List of highmem PBEs needed for restoring the highmem pages that were | |
2237 | * allocated before the suspend and included in the suspend image, but have | |
2238 | * also been allocated by the "resume" kernel, so their contents cannot be | |
2239 | * written directly to their "original" page frames. | |
2240 | */ | |
2241 | static struct highmem_pbe *highmem_pblist; | |
2242 | ||
2243 | /** | |
2244 | * count_highmem_image_pages - Compute the number of highmem pages in the image. | |
2245 | * @bm: Memory bitmap. | |
2246 | * | |
2247 | * The bits in @bm that correspond to image pages are assumed to be set. | |
2248 | */ | |
2249 | static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) | |
2250 | { | |
2251 | unsigned long pfn; | |
2252 | unsigned int cnt = 0; | |
2253 | ||
2254 | memory_bm_position_reset(bm); | |
2255 | pfn = memory_bm_next_pfn(bm); | |
2256 | while (pfn != BM_END_OF_MAP) { | |
2257 | if (PageHighMem(pfn_to_page(pfn))) | |
2258 | cnt++; | |
2259 | ||
2260 | pfn = memory_bm_next_pfn(bm); | |
2261 | } | |
2262 | return cnt; | |
2263 | } | |
2264 | ||
2265 | static unsigned int safe_highmem_pages; | |
2266 | ||
2267 | static struct memory_bitmap *safe_highmem_bm; | |
2268 | ||
2269 | /** | |
2270 | * prepare_highmem_image - Allocate memory for loading highmem data from image. | |
2271 | * @bm: Pointer to an uninitialized memory bitmap structure. | |
2272 | * @nr_highmem_p: Pointer to the number of highmem image pages. | |
2273 | * | |
2274 | * Try to allocate as many highmem pages as there are highmem image pages | |
2275 | * (@nr_highmem_p points to the variable containing the number of highmem image | |
2276 | * pages). The pages that are "safe" (ie. will not be overwritten when the | |
2277 | * hibernation image is restored entirely) have the corresponding bits set in | |
2278 | * @bm (it must be unitialized). | |
2279 | * | |
2280 | * NOTE: This function should not be called if there are no highmem image pages. | |
2281 | */ | |
2282 | static int prepare_highmem_image(struct memory_bitmap *bm, | |
2283 | unsigned int *nr_highmem_p) | |
2284 | { | |
2285 | unsigned int to_alloc; | |
2286 | ||
2287 | if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE)) | |
2288 | return -ENOMEM; | |
2289 | ||
2290 | if (get_highmem_buffer(PG_SAFE)) | |
2291 | return -ENOMEM; | |
2292 | ||
2293 | to_alloc = count_free_highmem_pages(); | |
2294 | if (to_alloc > *nr_highmem_p) | |
2295 | to_alloc = *nr_highmem_p; | |
2296 | else | |
2297 | *nr_highmem_p = to_alloc; | |
2298 | ||
2299 | safe_highmem_pages = 0; | |
2300 | while (to_alloc-- > 0) { | |
2301 | struct page *page; | |
2302 | ||
2303 | page = alloc_page(__GFP_HIGHMEM); | |
2304 | if (!swsusp_page_is_free(page)) { | |
2305 | /* The page is "safe", set its bit the bitmap */ | |
2306 | memory_bm_set_bit(bm, page_to_pfn(page)); | |
2307 | safe_highmem_pages++; | |
2308 | } | |
2309 | /* Mark the page as allocated */ | |
2310 | swsusp_set_page_forbidden(page); | |
2311 | swsusp_set_page_free(page); | |
2312 | } | |
2313 | memory_bm_position_reset(bm); | |
2314 | safe_highmem_bm = bm; | |
2315 | return 0; | |
2316 | } | |
2317 | ||
2318 | static struct page *last_highmem_page; | |
2319 | ||
2320 | /** | |
2321 | * get_highmem_page_buffer - Prepare a buffer to store a highmem image page. | |
2322 | * | |
2323 | * For a given highmem image page get a buffer that suspend_write_next() should | |
2324 | * return to its caller to write to. | |
2325 | * | |
2326 | * If the page is to be saved to its "original" page frame or a copy of | |
2327 | * the page is to be made in the highmem, @buffer is returned. Otherwise, | |
2328 | * the copy of the page is to be made in normal memory, so the address of | |
2329 | * the copy is returned. | |
2330 | * | |
2331 | * If @buffer is returned, the caller of suspend_write_next() will write | |
2332 | * the page's contents to @buffer, so they will have to be copied to the | |
2333 | * right location on the next call to suspend_write_next() and it is done | |
2334 | * with the help of copy_last_highmem_page(). For this purpose, if | |
2335 | * @buffer is returned, @last_highmem_page is set to the page to which | |
2336 | * the data will have to be copied from @buffer. | |
2337 | */ | |
2338 | static void *get_highmem_page_buffer(struct page *page, | |
2339 | struct chain_allocator *ca) | |
2340 | { | |
2341 | struct highmem_pbe *pbe; | |
2342 | void *kaddr; | |
2343 | ||
2344 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) { | |
2345 | /* | |
2346 | * We have allocated the "original" page frame and we can | |
2347 | * use it directly to store the loaded page. | |
2348 | */ | |
2349 | last_highmem_page = page; | |
2350 | return buffer; | |
2351 | } | |
2352 | /* | |
2353 | * The "original" page frame has not been allocated and we have to | |
2354 | * use a "safe" page frame to store the loaded page. | |
2355 | */ | |
2356 | pbe = chain_alloc(ca, sizeof(struct highmem_pbe)); | |
2357 | if (!pbe) { | |
2358 | swsusp_free(); | |
2359 | return ERR_PTR(-ENOMEM); | |
2360 | } | |
2361 | pbe->orig_page = page; | |
2362 | if (safe_highmem_pages > 0) { | |
2363 | struct page *tmp; | |
2364 | ||
2365 | /* Copy of the page will be stored in high memory */ | |
2366 | kaddr = buffer; | |
2367 | tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm)); | |
2368 | safe_highmem_pages--; | |
2369 | last_highmem_page = tmp; | |
2370 | pbe->copy_page = tmp; | |
2371 | } else { | |
2372 | /* Copy of the page will be stored in normal memory */ | |
2373 | kaddr = safe_pages_list; | |
2374 | safe_pages_list = safe_pages_list->next; | |
2375 | pbe->copy_page = virt_to_page(kaddr); | |
2376 | } | |
2377 | pbe->next = highmem_pblist; | |
2378 | highmem_pblist = pbe; | |
2379 | return kaddr; | |
2380 | } | |
2381 | ||
2382 | /** | |
2383 | * copy_last_highmem_page - Copy most the most recent highmem image page. | |
2384 | * | |
2385 | * Copy the contents of a highmem image from @buffer, where the caller of | |
2386 | * snapshot_write_next() has stored them, to the right location represented by | |
2387 | * @last_highmem_page . | |
2388 | */ | |
2389 | static void copy_last_highmem_page(void) | |
2390 | { | |
2391 | if (last_highmem_page) { | |
2392 | void *dst; | |
2393 | ||
2394 | dst = kmap_atomic(last_highmem_page); | |
2395 | copy_page(dst, buffer); | |
2396 | kunmap_atomic(dst); | |
2397 | last_highmem_page = NULL; | |
2398 | } | |
2399 | } | |
2400 | ||
2401 | static inline int last_highmem_page_copied(void) | |
2402 | { | |
2403 | return !last_highmem_page; | |
2404 | } | |
2405 | ||
2406 | static inline void free_highmem_data(void) | |
2407 | { | |
2408 | if (safe_highmem_bm) | |
2409 | memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR); | |
2410 | ||
2411 | if (buffer) | |
2412 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
2413 | } | |
2414 | #else | |
2415 | static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; } | |
2416 | ||
2417 | static inline int prepare_highmem_image(struct memory_bitmap *bm, | |
2418 | unsigned int *nr_highmem_p) { return 0; } | |
2419 | ||
2420 | static inline void *get_highmem_page_buffer(struct page *page, | |
2421 | struct chain_allocator *ca) | |
2422 | { | |
2423 | return ERR_PTR(-EINVAL); | |
2424 | } | |
2425 | ||
2426 | static inline void copy_last_highmem_page(void) {} | |
2427 | static inline int last_highmem_page_copied(void) { return 1; } | |
2428 | static inline void free_highmem_data(void) {} | |
2429 | #endif /* CONFIG_HIGHMEM */ | |
2430 | ||
2431 | #define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe)) | |
2432 | ||
2433 | /** | |
2434 | * prepare_image - Make room for loading hibernation image. | |
2435 | * @new_bm: Unitialized memory bitmap structure. | |
2436 | * @bm: Memory bitmap with unsafe pages marked. | |
2437 | * | |
2438 | * Use @bm to mark the pages that will be overwritten in the process of | |
2439 | * restoring the system memory state from the suspend image ("unsafe" pages) | |
2440 | * and allocate memory for the image. | |
2441 | * | |
2442 | * The idea is to allocate a new memory bitmap first and then allocate | |
2443 | * as many pages as needed for image data, but without specifying what those | |
2444 | * pages will be used for just yet. Instead, we mark them all as allocated and | |
2445 | * create a lists of "safe" pages to be used later. On systems with high | |
2446 | * memory a list of "safe" highmem pages is created too. | |
2447 | */ | |
2448 | static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm) | |
2449 | { | |
2450 | unsigned int nr_pages, nr_highmem; | |
2451 | struct linked_page *lp; | |
2452 | int error; | |
2453 | ||
2454 | /* If there is no highmem, the buffer will not be necessary */ | |
2455 | free_image_page(buffer, PG_UNSAFE_CLEAR); | |
2456 | buffer = NULL; | |
2457 | ||
2458 | nr_highmem = count_highmem_image_pages(bm); | |
2459 | mark_unsafe_pages(bm); | |
2460 | ||
2461 | error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE); | |
2462 | if (error) | |
2463 | goto Free; | |
2464 | ||
2465 | duplicate_memory_bitmap(new_bm, bm); | |
2466 | memory_bm_free(bm, PG_UNSAFE_KEEP); | |
2467 | if (nr_highmem > 0) { | |
2468 | error = prepare_highmem_image(bm, &nr_highmem); | |
2469 | if (error) | |
2470 | goto Free; | |
2471 | } | |
2472 | /* | |
2473 | * Reserve some safe pages for potential later use. | |
2474 | * | |
2475 | * NOTE: This way we make sure there will be enough safe pages for the | |
2476 | * chain_alloc() in get_buffer(). It is a bit wasteful, but | |
2477 | * nr_copy_pages cannot be greater than 50% of the memory anyway. | |
2478 | * | |
2479 | * nr_copy_pages cannot be less than allocated_unsafe_pages too. | |
2480 | */ | |
2481 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; | |
2482 | nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE); | |
2483 | while (nr_pages > 0) { | |
2484 | lp = get_image_page(GFP_ATOMIC, PG_SAFE); | |
2485 | if (!lp) { | |
2486 | error = -ENOMEM; | |
2487 | goto Free; | |
2488 | } | |
2489 | lp->next = safe_pages_list; | |
2490 | safe_pages_list = lp; | |
2491 | nr_pages--; | |
2492 | } | |
2493 | /* Preallocate memory for the image */ | |
2494 | nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages; | |
2495 | while (nr_pages > 0) { | |
2496 | lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC); | |
2497 | if (!lp) { | |
2498 | error = -ENOMEM; | |
2499 | goto Free; | |
2500 | } | |
2501 | if (!swsusp_page_is_free(virt_to_page(lp))) { | |
2502 | /* The page is "safe", add it to the list */ | |
2503 | lp->next = safe_pages_list; | |
2504 | safe_pages_list = lp; | |
2505 | } | |
2506 | /* Mark the page as allocated */ | |
2507 | swsusp_set_page_forbidden(virt_to_page(lp)); | |
2508 | swsusp_set_page_free(virt_to_page(lp)); | |
2509 | nr_pages--; | |
2510 | } | |
2511 | return 0; | |
2512 | ||
2513 | Free: | |
2514 | swsusp_free(); | |
2515 | return error; | |
2516 | } | |
2517 | ||
2518 | /** | |
2519 | * get_buffer - Get the address to store the next image data page. | |
2520 | * | |
2521 | * Get the address that snapshot_write_next() should return to its caller to | |
2522 | * write to. | |
2523 | */ | |
2524 | static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca) | |
2525 | { | |
2526 | struct pbe *pbe; | |
2527 | struct page *page; | |
2528 | unsigned long pfn = memory_bm_next_pfn(bm); | |
2529 | ||
2530 | if (pfn == BM_END_OF_MAP) | |
2531 | return ERR_PTR(-EFAULT); | |
2532 | ||
2533 | page = pfn_to_page(pfn); | |
2534 | if (PageHighMem(page)) | |
2535 | return get_highmem_page_buffer(page, ca); | |
2536 | ||
2537 | if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) | |
2538 | /* | |
2539 | * We have allocated the "original" page frame and we can | |
2540 | * use it directly to store the loaded page. | |
2541 | */ | |
2542 | return page_address(page); | |
2543 | ||
2544 | /* | |
2545 | * The "original" page frame has not been allocated and we have to | |
2546 | * use a "safe" page frame to store the loaded page. | |
2547 | */ | |
2548 | pbe = chain_alloc(ca, sizeof(struct pbe)); | |
2549 | if (!pbe) { | |
2550 | swsusp_free(); | |
2551 | return ERR_PTR(-ENOMEM); | |
2552 | } | |
2553 | pbe->orig_address = page_address(page); | |
2554 | pbe->address = safe_pages_list; | |
2555 | safe_pages_list = safe_pages_list->next; | |
2556 | pbe->next = restore_pblist; | |
2557 | restore_pblist = pbe; | |
2558 | return pbe->address; | |
2559 | } | |
2560 | ||
2561 | /** | |
2562 | * snapshot_write_next - Get the address to store the next image page. | |
2563 | * @handle: Snapshot handle structure to guide the writing. | |
2564 | * | |
2565 | * On the first call, @handle should point to a zeroed snapshot_handle | |
2566 | * structure. The structure gets populated then and a pointer to it should be | |
2567 | * passed to this function every next time. | |
2568 | * | |
2569 | * On success, the function returns a positive number. Then, the caller | |
2570 | * is allowed to write up to the returned number of bytes to the memory | |
2571 | * location computed by the data_of() macro. | |
2572 | * | |
2573 | * The function returns 0 to indicate the "end of file" condition. Negative | |
2574 | * numbers are returned on errors, in which cases the structure pointed to by | |
2575 | * @handle is not updated and should not be used any more. | |
2576 | */ | |
2577 | int snapshot_write_next(struct snapshot_handle *handle) | |
2578 | { | |
2579 | static struct chain_allocator ca; | |
2580 | int error = 0; | |
2581 | ||
2582 | /* Check if we have already loaded the entire image */ | |
2583 | if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) | |
2584 | return 0; | |
2585 | ||
2586 | handle->sync_read = 1; | |
2587 | ||
2588 | if (!handle->cur) { | |
2589 | if (!buffer) | |
2590 | /* This makes the buffer be freed by swsusp_free() */ | |
2591 | buffer = get_image_page(GFP_ATOMIC, PG_ANY); | |
2592 | ||
2593 | if (!buffer) | |
2594 | return -ENOMEM; | |
2595 | ||
2596 | handle->buffer = buffer; | |
2597 | } else if (handle->cur == 1) { | |
2598 | error = load_header(buffer); | |
2599 | if (error) | |
2600 | return error; | |
2601 | ||
2602 | safe_pages_list = NULL; | |
2603 | ||
2604 | error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY); | |
2605 | if (error) | |
2606 | return error; | |
2607 | ||
2608 | /* Allocate buffer for page keys. */ | |
2609 | error = page_key_alloc(nr_copy_pages); | |
2610 | if (error) | |
2611 | return error; | |
2612 | ||
2613 | hibernate_restore_protection_begin(); | |
2614 | } else if (handle->cur <= nr_meta_pages + 1) { | |
2615 | error = unpack_orig_pfns(buffer, ©_bm); | |
2616 | if (error) | |
2617 | return error; | |
2618 | ||
2619 | if (handle->cur == nr_meta_pages + 1) { | |
2620 | error = prepare_image(&orig_bm, ©_bm); | |
2621 | if (error) | |
2622 | return error; | |
2623 | ||
2624 | chain_init(&ca, GFP_ATOMIC, PG_SAFE); | |
2625 | memory_bm_position_reset(&orig_bm); | |
2626 | restore_pblist = NULL; | |
2627 | handle->buffer = get_buffer(&orig_bm, &ca); | |
2628 | handle->sync_read = 0; | |
2629 | if (IS_ERR(handle->buffer)) | |
2630 | return PTR_ERR(handle->buffer); | |
2631 | } | |
2632 | } else { | |
2633 | copy_last_highmem_page(); | |
2634 | /* Restore page key for data page (s390 only). */ | |
2635 | page_key_write(handle->buffer); | |
2636 | hibernate_restore_protect_page(handle->buffer); | |
2637 | handle->buffer = get_buffer(&orig_bm, &ca); | |
2638 | if (IS_ERR(handle->buffer)) | |
2639 | return PTR_ERR(handle->buffer); | |
2640 | if (handle->buffer != buffer) | |
2641 | handle->sync_read = 0; | |
2642 | } | |
2643 | handle->cur++; | |
2644 | return PAGE_SIZE; | |
2645 | } | |
2646 | ||
2647 | /** | |
2648 | * snapshot_write_finalize - Complete the loading of a hibernation image. | |
2649 | * | |
2650 | * Must be called after the last call to snapshot_write_next() in case the last | |
2651 | * page in the image happens to be a highmem page and its contents should be | |
2652 | * stored in highmem. Additionally, it recycles bitmap memory that's not | |
2653 | * necessary any more. | |
2654 | */ | |
2655 | void snapshot_write_finalize(struct snapshot_handle *handle) | |
2656 | { | |
2657 | copy_last_highmem_page(); | |
2658 | /* Restore page key for data page (s390 only). */ | |
2659 | page_key_write(handle->buffer); | |
2660 | page_key_free(); | |
2661 | hibernate_restore_protect_page(handle->buffer); | |
2662 | /* Do that only if we have loaded the image entirely */ | |
2663 | if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) { | |
2664 | memory_bm_recycle(&orig_bm); | |
2665 | free_highmem_data(); | |
2666 | } | |
2667 | } | |
2668 | ||
2669 | int snapshot_image_loaded(struct snapshot_handle *handle) | |
2670 | { | |
2671 | return !(!nr_copy_pages || !last_highmem_page_copied() || | |
2672 | handle->cur <= nr_meta_pages + nr_copy_pages); | |
2673 | } | |
2674 | ||
2675 | #ifdef CONFIG_HIGHMEM | |
2676 | /* Assumes that @buf is ready and points to a "safe" page */ | |
2677 | static inline void swap_two_pages_data(struct page *p1, struct page *p2, | |
2678 | void *buf) | |
2679 | { | |
2680 | void *kaddr1, *kaddr2; | |
2681 | ||
2682 | kaddr1 = kmap_atomic(p1); | |
2683 | kaddr2 = kmap_atomic(p2); | |
2684 | copy_page(buf, kaddr1); | |
2685 | copy_page(kaddr1, kaddr2); | |
2686 | copy_page(kaddr2, buf); | |
2687 | kunmap_atomic(kaddr2); | |
2688 | kunmap_atomic(kaddr1); | |
2689 | } | |
2690 | ||
2691 | /** | |
2692 | * restore_highmem - Put highmem image pages into their original locations. | |
2693 | * | |
2694 | * For each highmem page that was in use before hibernation and is included in | |
2695 | * the image, and also has been allocated by the "restore" kernel, swap its | |
2696 | * current contents with the previous (ie. "before hibernation") ones. | |
2697 | * | |
2698 | * If the restore eventually fails, we can call this function once again and | |
2699 | * restore the highmem state as seen by the restore kernel. | |
2700 | */ | |
2701 | int restore_highmem(void) | |
2702 | { | |
2703 | struct highmem_pbe *pbe = highmem_pblist; | |
2704 | void *buf; | |
2705 | ||
2706 | if (!pbe) | |
2707 | return 0; | |
2708 | ||
2709 | buf = get_image_page(GFP_ATOMIC, PG_SAFE); | |
2710 | if (!buf) | |
2711 | return -ENOMEM; | |
2712 | ||
2713 | while (pbe) { | |
2714 | swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf); | |
2715 | pbe = pbe->next; | |
2716 | } | |
2717 | free_image_page(buf, PG_UNSAFE_CLEAR); | |
2718 | return 0; | |
2719 | } | |
2720 | #endif /* CONFIG_HIGHMEM */ |