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