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// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/power/snapshot.c
4 *
5 * This file provides system snapshot/restore functionality for swsusp.
6 *
7 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
8 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
9 */
10
11#define pr_fmt(fmt) "PM: hibernation: " fmt
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/memblock.h>
25#include <linux/nmi.h>
26#include <linux/syscalls.h>
27#include <linux/console.h>
28#include <linux/highmem.h>
29#include <linux/list.h>
30#include <linux/slab.h>
31#include <linux/compiler.h>
32#include <linux/ktime.h>
33#include <linux/set_memory.h>
34
35#include <linux/uaccess.h>
36#include <asm/mmu_context.h>
37#include <asm/tlbflush.h>
38#include <asm/io.h>
39
40#include "power.h"
41
42#if defined(CONFIG_STRICT_KERNEL_RWX) && defined(CONFIG_ARCH_HAS_SET_MEMORY)
43static bool hibernate_restore_protection;
44static bool hibernate_restore_protection_active;
45
46void enable_restore_image_protection(void)
47{
48 hibernate_restore_protection = true;
49}
50
51static inline void hibernate_restore_protection_begin(void)
52{
53 hibernate_restore_protection_active = hibernate_restore_protection;
54}
55
56static inline void hibernate_restore_protection_end(void)
57{
58 hibernate_restore_protection_active = false;
59}
60
61static inline void hibernate_restore_protect_page(void *page_address)
62{
63 if (hibernate_restore_protection_active)
64 set_memory_ro((unsigned long)page_address, 1);
65}
66
67static inline void hibernate_restore_unprotect_page(void *page_address)
68{
69 if (hibernate_restore_protection_active)
70 set_memory_rw((unsigned long)page_address, 1);
71}
72#else
73static inline void hibernate_restore_protection_begin(void) {}
74static inline void hibernate_restore_protection_end(void) {}
75static inline void hibernate_restore_protect_page(void *page_address) {}
76static inline void hibernate_restore_unprotect_page(void *page_address) {}
77#endif /* CONFIG_STRICT_KERNEL_RWX && CONFIG_ARCH_HAS_SET_MEMORY */
78
79
80/*
81 * The calls to set_direct_map_*() should not fail because remapping a page
82 * here means that we only update protection bits in an existing PTE.
83 * It is still worth to have a warning here if something changes and this
84 * will no longer be the case.
85 */
86static inline void hibernate_map_page(struct page *page)
87{
88 if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
89 int ret = set_direct_map_default_noflush(page);
90
91 if (ret)
92 pr_warn_once("Failed to remap page\n");
93 } else {
94 debug_pagealloc_map_pages(page, 1);
95 }
96}
97
98static inline void hibernate_unmap_page(struct page *page)
99{
100 if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
101 unsigned long addr = (unsigned long)page_address(page);
102 int ret = set_direct_map_invalid_noflush(page);
103
104 if (ret)
105 pr_warn_once("Failed to remap page\n");
106
107 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
108 } else {
109 debug_pagealloc_unmap_pages(page, 1);
110 }
111}
112
113static int swsusp_page_is_free(struct page *);
114static void swsusp_set_page_forbidden(struct page *);
115static void swsusp_unset_page_forbidden(struct page *);
116
117/*
118 * Number of bytes to reserve for memory allocations made by device drivers
119 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
120 * cause image creation to fail (tunable via /sys/power/reserved_size).
121 */
122unsigned long reserved_size;
123
124void __init hibernate_reserved_size_init(void)
125{
126 reserved_size = SPARE_PAGES * PAGE_SIZE;
127}
128
129/*
130 * Preferred image size in bytes (tunable via /sys/power/image_size).
131 * When it is set to N, swsusp will do its best to ensure the image
132 * size will not exceed N bytes, but if that is impossible, it will
133 * try to create the smallest image possible.
134 */
135unsigned long image_size;
136
137void __init hibernate_image_size_init(void)
138{
139 image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
140}
141
142/*
143 * List of PBEs needed for restoring the pages that were allocated before
144 * the suspend and included in the suspend image, but have also been
145 * allocated by the "resume" kernel, so their contents cannot be written
146 * directly to their "original" page frames.
147 */
148struct pbe *restore_pblist;
149
150/* struct linked_page is used to build chains of pages */
151
152#define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
153
154struct linked_page {
155 struct linked_page *next;
156 char data[LINKED_PAGE_DATA_SIZE];
157} __packed;
158
159/*
160 * List of "safe" pages (ie. pages that were not used by the image kernel
161 * before hibernation) that may be used as temporary storage for image kernel
162 * memory contents.
163 */
164static struct linked_page *safe_pages_list;
165
166/* Pointer to an auxiliary buffer (1 page) */
167static void *buffer;
168
169#define PG_ANY 0
170#define PG_SAFE 1
171#define PG_UNSAFE_CLEAR 1
172#define PG_UNSAFE_KEEP 0
173
174static unsigned int allocated_unsafe_pages;
175
176/**
177 * get_image_page - Allocate a page for a hibernation image.
178 * @gfp_mask: GFP mask for the allocation.
179 * @safe_needed: Get pages that were not used before hibernation (restore only)
180 *
181 * During image restoration, for storing the PBE list and the image data, we can
182 * only use memory pages that do not conflict with the pages used before
183 * hibernation. The "unsafe" pages have PageNosaveFree set and we count them
184 * using allocated_unsafe_pages.
185 *
186 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
187 * swsusp_free() can release it.
188 */
189static void *get_image_page(gfp_t gfp_mask, int safe_needed)
190{
191 void *res;
192
193 res = (void *)get_zeroed_page(gfp_mask);
194 if (safe_needed)
195 while (res && swsusp_page_is_free(virt_to_page(res))) {
196 /* The page is unsafe, mark it for swsusp_free() */
197 swsusp_set_page_forbidden(virt_to_page(res));
198 allocated_unsafe_pages++;
199 res = (void *)get_zeroed_page(gfp_mask);
200 }
201 if (res) {
202 swsusp_set_page_forbidden(virt_to_page(res));
203 swsusp_set_page_free(virt_to_page(res));
204 }
205 return res;
206}
207
208static void *__get_safe_page(gfp_t gfp_mask)
209{
210 if (safe_pages_list) {
211 void *ret = safe_pages_list;
212
213 safe_pages_list = safe_pages_list->next;
214 memset(ret, 0, PAGE_SIZE);
215 return ret;
216 }
217 return get_image_page(gfp_mask, PG_SAFE);
218}
219
220unsigned long get_safe_page(gfp_t gfp_mask)
221{
222 return (unsigned long)__get_safe_page(gfp_mask);
223}
224
225static struct page *alloc_image_page(gfp_t gfp_mask)
226{
227 struct page *page;
228
229 page = alloc_page(gfp_mask);
230 if (page) {
231 swsusp_set_page_forbidden(page);
232 swsusp_set_page_free(page);
233 }
234 return page;
235}
236
237static void recycle_safe_page(void *page_address)
238{
239 struct linked_page *lp = page_address;
240
241 lp->next = safe_pages_list;
242 safe_pages_list = lp;
243}
244
245/**
246 * free_image_page - Free a page allocated for hibernation image.
247 * @addr: Address of the page to free.
248 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
249 *
250 * The page to free should have been allocated by get_image_page() (page flags
251 * set by it are affected).
252 */
253static inline void free_image_page(void *addr, int clear_nosave_free)
254{
255 struct page *page;
256
257 BUG_ON(!virt_addr_valid(addr));
258
259 page = virt_to_page(addr);
260
261 swsusp_unset_page_forbidden(page);
262 if (clear_nosave_free)
263 swsusp_unset_page_free(page);
264
265 __free_page(page);
266}
267
268static inline void free_list_of_pages(struct linked_page *list,
269 int clear_page_nosave)
270{
271 while (list) {
272 struct linked_page *lp = list->next;
273
274 free_image_page(list, clear_page_nosave);
275 list = lp;
276 }
277}
278
279/*
280 * struct chain_allocator is used for allocating small objects out of
281 * a linked list of pages called 'the chain'.
282 *
283 * The chain grows each time when there is no room for a new object in
284 * the current page. The allocated objects cannot be freed individually.
285 * It is only possible to free them all at once, by freeing the entire
286 * chain.
287 *
288 * NOTE: The chain allocator may be inefficient if the allocated objects
289 * are not much smaller than PAGE_SIZE.
290 */
291struct chain_allocator {
292 struct linked_page *chain; /* the chain */
293 unsigned int used_space; /* total size of objects allocated out
294 of the current page */
295 gfp_t gfp_mask; /* mask for allocating pages */
296 int safe_needed; /* if set, only "safe" pages are allocated */
297};
298
299static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
300 int safe_needed)
301{
302 ca->chain = NULL;
303 ca->used_space = LINKED_PAGE_DATA_SIZE;
304 ca->gfp_mask = gfp_mask;
305 ca->safe_needed = safe_needed;
306}
307
308static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
309{
310 void *ret;
311
312 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
313 struct linked_page *lp;
314
315 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
316 get_image_page(ca->gfp_mask, PG_ANY);
317 if (!lp)
318 return NULL;
319
320 lp->next = ca->chain;
321 ca->chain = lp;
322 ca->used_space = 0;
323 }
324 ret = ca->chain->data + ca->used_space;
325 ca->used_space += size;
326 return ret;
327}
328
329/**
330 * Data types related to memory bitmaps.
331 *
332 * Memory bitmap is a structure consisting of many linked lists of
333 * objects. The main list's elements are of type struct zone_bitmap
334 * and each of them corresponds to one zone. For each zone bitmap
335 * object there is a list of objects of type struct bm_block that
336 * represent each blocks of bitmap in which information is stored.
337 *
338 * struct memory_bitmap contains a pointer to the main list of zone
339 * bitmap objects, a struct bm_position used for browsing the bitmap,
340 * and a pointer to the list of pages used for allocating all of the
341 * zone bitmap objects and bitmap block objects.
342 *
343 * NOTE: It has to be possible to lay out the bitmap in memory
344 * using only allocations of order 0. Additionally, the bitmap is
345 * designed to work with arbitrary number of zones (this is over the
346 * top for now, but let's avoid making unnecessary assumptions ;-).
347 *
348 * struct zone_bitmap contains a pointer to a list of bitmap block
349 * objects and a pointer to the bitmap block object that has been
350 * most recently used for setting bits. Additionally, it contains the
351 * PFNs that correspond to the start and end of the represented zone.
352 *
353 * struct bm_block contains a pointer to the memory page in which
354 * information is stored (in the form of a block of bitmap)
355 * It also contains the pfns that correspond to the start and end of
356 * the represented memory area.
357 *
358 * The memory bitmap is organized as a radix tree to guarantee fast random
359 * access to the bits. There is one radix tree for each zone (as returned
360 * from create_mem_extents).
361 *
362 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
363 * two linked lists for the nodes of the tree, one for the inner nodes and
364 * one for the leave nodes. The linked leave nodes are used for fast linear
365 * access of the memory bitmap.
366 *
367 * The struct rtree_node represents one node of the radix tree.
368 */
369
370#define BM_END_OF_MAP (~0UL)
371
372#define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
373#define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
374#define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
375
376/*
377 * struct rtree_node is a wrapper struct to link the nodes
378 * of the rtree together for easy linear iteration over
379 * bits and easy freeing
380 */
381struct rtree_node {
382 struct list_head list;
383 unsigned long *data;
384};
385
386/*
387 * struct mem_zone_bm_rtree represents a bitmap used for one
388 * populated memory zone.
389 */
390struct mem_zone_bm_rtree {
391 struct list_head list; /* Link Zones together */
392 struct list_head nodes; /* Radix Tree inner nodes */
393 struct list_head leaves; /* Radix Tree leaves */
394 unsigned long start_pfn; /* Zone start page frame */
395 unsigned long end_pfn; /* Zone end page frame + 1 */
396 struct rtree_node *rtree; /* Radix Tree Root */
397 int levels; /* Number of Radix Tree Levels */
398 unsigned int blocks; /* Number of Bitmap Blocks */
399};
400
401/* strcut bm_position is used for browsing memory bitmaps */
402
403struct bm_position {
404 struct mem_zone_bm_rtree *zone;
405 struct rtree_node *node;
406 unsigned long node_pfn;
407 int node_bit;
408};
409
410struct memory_bitmap {
411 struct list_head zones;
412 struct linked_page *p_list; /* list of pages used to store zone
413 bitmap objects and bitmap block
414 objects */
415 struct bm_position cur; /* most recently used bit position */
416};
417
418/* Functions that operate on memory bitmaps */
419
420#define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
421#if BITS_PER_LONG == 32
422#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
423#else
424#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
425#endif
426#define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
427
428/**
429 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
430 *
431 * This function is used to allocate inner nodes as well as the
432 * leave nodes of the radix tree. It also adds the node to the
433 * corresponding linked list passed in by the *list parameter.
434 */
435static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
436 struct chain_allocator *ca,
437 struct list_head *list)
438{
439 struct rtree_node *node;
440
441 node = chain_alloc(ca, sizeof(struct rtree_node));
442 if (!node)
443 return NULL;
444
445 node->data = get_image_page(gfp_mask, safe_needed);
446 if (!node->data)
447 return NULL;
448
449 list_add_tail(&node->list, list);
450
451 return node;
452}
453
454/**
455 * add_rtree_block - Add a new leave node to the radix tree.
456 *
457 * The leave nodes need to be allocated in order to keep the leaves
458 * linked list in order. This is guaranteed by the zone->blocks
459 * counter.
460 */
461static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
462 int safe_needed, struct chain_allocator *ca)
463{
464 struct rtree_node *node, *block, **dst;
465 unsigned int levels_needed, block_nr;
466 int i;
467
468 block_nr = zone->blocks;
469 levels_needed = 0;
470
471 /* How many levels do we need for this block nr? */
472 while (block_nr) {
473 levels_needed += 1;
474 block_nr >>= BM_RTREE_LEVEL_SHIFT;
475 }
476
477 /* Make sure the rtree has enough levels */
478 for (i = zone->levels; i < levels_needed; i++) {
479 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
480 &zone->nodes);
481 if (!node)
482 return -ENOMEM;
483
484 node->data[0] = (unsigned long)zone->rtree;
485 zone->rtree = node;
486 zone->levels += 1;
487 }
488
489 /* Allocate new block */
490 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
491 if (!block)
492 return -ENOMEM;
493
494 /* Now walk the rtree to insert the block */
495 node = zone->rtree;
496 dst = &zone->rtree;
497 block_nr = zone->blocks;
498 for (i = zone->levels; i > 0; i--) {
499 int index;
500
501 if (!node) {
502 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
503 &zone->nodes);
504 if (!node)
505 return -ENOMEM;
506 *dst = node;
507 }
508
509 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
510 index &= BM_RTREE_LEVEL_MASK;
511 dst = (struct rtree_node **)&((*dst)->data[index]);
512 node = *dst;
513 }
514
515 zone->blocks += 1;
516 *dst = block;
517
518 return 0;
519}
520
521static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
522 int clear_nosave_free);
523
524/**
525 * create_zone_bm_rtree - Create a radix tree for one zone.
526 *
527 * Allocated the mem_zone_bm_rtree structure and initializes it.
528 * This function also allocated and builds the radix tree for the
529 * zone.
530 */
531static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
532 int safe_needed,
533 struct chain_allocator *ca,
534 unsigned long start,
535 unsigned long end)
536{
537 struct mem_zone_bm_rtree *zone;
538 unsigned int i, nr_blocks;
539 unsigned long pages;
540
541 pages = end - start;
542 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
543 if (!zone)
544 return NULL;
545
546 INIT_LIST_HEAD(&zone->nodes);
547 INIT_LIST_HEAD(&zone->leaves);
548 zone->start_pfn = start;
549 zone->end_pfn = end;
550 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
551
552 for (i = 0; i < nr_blocks; i++) {
553 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
554 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
555 return NULL;
556 }
557 }
558
559 return zone;
560}
561
562/**
563 * free_zone_bm_rtree - Free the memory of the radix tree.
564 *
565 * Free all node pages of the radix tree. The mem_zone_bm_rtree
566 * structure itself is not freed here nor are the rtree_node
567 * structs.
568 */
569static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
570 int clear_nosave_free)
571{
572 struct rtree_node *node;
573
574 list_for_each_entry(node, &zone->nodes, list)
575 free_image_page(node->data, clear_nosave_free);
576
577 list_for_each_entry(node, &zone->leaves, list)
578 free_image_page(node->data, clear_nosave_free);
579}
580
581static void memory_bm_position_reset(struct memory_bitmap *bm)
582{
583 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
584 list);
585 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
586 struct rtree_node, list);
587 bm->cur.node_pfn = 0;
588 bm->cur.node_bit = 0;
589}
590
591static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
592
593struct mem_extent {
594 struct list_head hook;
595 unsigned long start;
596 unsigned long end;
597};
598
599/**
600 * free_mem_extents - Free a list of memory extents.
601 * @list: List of extents to free.
602 */
603static void free_mem_extents(struct list_head *list)
604{
605 struct mem_extent *ext, *aux;
606
607 list_for_each_entry_safe(ext, aux, list, hook) {
608 list_del(&ext->hook);
609 kfree(ext);
610 }
611}
612
613/**
614 * create_mem_extents - Create a list of memory extents.
615 * @list: List to put the extents into.
616 * @gfp_mask: Mask to use for memory allocations.
617 *
618 * The extents represent contiguous ranges of PFNs.
619 */
620static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
621{
622 struct zone *zone;
623
624 INIT_LIST_HEAD(list);
625
626 for_each_populated_zone(zone) {
627 unsigned long zone_start, zone_end;
628 struct mem_extent *ext, *cur, *aux;
629
630 zone_start = zone->zone_start_pfn;
631 zone_end = zone_end_pfn(zone);
632
633 list_for_each_entry(ext, list, hook)
634 if (zone_start <= ext->end)
635 break;
636
637 if (&ext->hook == list || zone_end < ext->start) {
638 /* New extent is necessary */
639 struct mem_extent *new_ext;
640
641 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
642 if (!new_ext) {
643 free_mem_extents(list);
644 return -ENOMEM;
645 }
646 new_ext->start = zone_start;
647 new_ext->end = zone_end;
648 list_add_tail(&new_ext->hook, &ext->hook);
649 continue;
650 }
651
652 /* Merge this zone's range of PFNs with the existing one */
653 if (zone_start < ext->start)
654 ext->start = zone_start;
655 if (zone_end > ext->end)
656 ext->end = zone_end;
657
658 /* More merging may be possible */
659 cur = ext;
660 list_for_each_entry_safe_continue(cur, aux, list, hook) {
661 if (zone_end < cur->start)
662 break;
663 if (zone_end < cur->end)
664 ext->end = cur->end;
665 list_del(&cur->hook);
666 kfree(cur);
667 }
668 }
669
670 return 0;
671}
672
673/**
674 * memory_bm_create - Allocate memory for a memory bitmap.
675 */
676static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
677 int safe_needed)
678{
679 struct chain_allocator ca;
680 struct list_head mem_extents;
681 struct mem_extent *ext;
682 int error;
683
684 chain_init(&ca, gfp_mask, safe_needed);
685 INIT_LIST_HEAD(&bm->zones);
686
687 error = create_mem_extents(&mem_extents, gfp_mask);
688 if (error)
689 return error;
690
691 list_for_each_entry(ext, &mem_extents, hook) {
692 struct mem_zone_bm_rtree *zone;
693
694 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
695 ext->start, ext->end);
696 if (!zone) {
697 error = -ENOMEM;
698 goto Error;
699 }
700 list_add_tail(&zone->list, &bm->zones);
701 }
702
703 bm->p_list = ca.chain;
704 memory_bm_position_reset(bm);
705 Exit:
706 free_mem_extents(&mem_extents);
707 return error;
708
709 Error:
710 bm->p_list = ca.chain;
711 memory_bm_free(bm, PG_UNSAFE_CLEAR);
712 goto Exit;
713}
714
715/**
716 * memory_bm_free - Free memory occupied by the memory bitmap.
717 * @bm: Memory bitmap.
718 */
719static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
720{
721 struct mem_zone_bm_rtree *zone;
722
723 list_for_each_entry(zone, &bm->zones, list)
724 free_zone_bm_rtree(zone, clear_nosave_free);
725
726 free_list_of_pages(bm->p_list, clear_nosave_free);
727
728 INIT_LIST_HEAD(&bm->zones);
729}
730
731/**
732 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
733 *
734 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
735 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
736 *
737 * Walk the radix tree to find the page containing the bit that represents @pfn
738 * and return the position of the bit in @addr and @bit_nr.
739 */
740static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
741 void **addr, unsigned int *bit_nr)
742{
743 struct mem_zone_bm_rtree *curr, *zone;
744 struct rtree_node *node;
745 int i, block_nr;
746
747 zone = bm->cur.zone;
748
749 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
750 goto zone_found;
751
752 zone = NULL;
753
754 /* Find the right zone */
755 list_for_each_entry(curr, &bm->zones, list) {
756 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
757 zone = curr;
758 break;
759 }
760 }
761
762 if (!zone)
763 return -EFAULT;
764
765zone_found:
766 /*
767 * We have found the zone. Now walk the radix tree to find the leaf node
768 * for our PFN.
769 */
770
771 /*
772 * If the zone we wish to scan is the current zone and the
773 * pfn falls into the current node then we do not need to walk
774 * the tree.
775 */
776 node = bm->cur.node;
777 if (zone == bm->cur.zone &&
778 ((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
779 goto node_found;
780
781 node = zone->rtree;
782 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
783
784 for (i = zone->levels; i > 0; i--) {
785 int index;
786
787 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
788 index &= BM_RTREE_LEVEL_MASK;
789 BUG_ON(node->data[index] == 0);
790 node = (struct rtree_node *)node->data[index];
791 }
792
793node_found:
794 /* Update last position */
795 bm->cur.zone = zone;
796 bm->cur.node = node;
797 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
798
799 /* Set return values */
800 *addr = node->data;
801 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
802
803 return 0;
804}
805
806static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
807{
808 void *addr;
809 unsigned int bit;
810 int error;
811
812 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
813 BUG_ON(error);
814 set_bit(bit, addr);
815}
816
817static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
818{
819 void *addr;
820 unsigned int bit;
821 int error;
822
823 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
824 if (!error)
825 set_bit(bit, addr);
826
827 return error;
828}
829
830static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
831{
832 void *addr;
833 unsigned int bit;
834 int error;
835
836 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
837 BUG_ON(error);
838 clear_bit(bit, addr);
839}
840
841static void memory_bm_clear_current(struct memory_bitmap *bm)
842{
843 int bit;
844
845 bit = max(bm->cur.node_bit - 1, 0);
846 clear_bit(bit, bm->cur.node->data);
847}
848
849static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
850{
851 void *addr;
852 unsigned int bit;
853 int error;
854
855 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
856 BUG_ON(error);
857 return test_bit(bit, addr);
858}
859
860static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
861{
862 void *addr;
863 unsigned int bit;
864
865 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
866}
867
868/*
869 * rtree_next_node - Jump to the next leaf node.
870 *
871 * Set the position to the beginning of the next node in the
872 * memory bitmap. This is either the next node in the current
873 * zone's radix tree or the first node in the radix tree of the
874 * next zone.
875 *
876 * Return true if there is a next node, false otherwise.
877 */
878static bool rtree_next_node(struct memory_bitmap *bm)
879{
880 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
881 bm->cur.node = list_entry(bm->cur.node->list.next,
882 struct rtree_node, list);
883 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
884 bm->cur.node_bit = 0;
885 touch_softlockup_watchdog();
886 return true;
887 }
888
889 /* No more nodes, goto next zone */
890 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
891 bm->cur.zone = list_entry(bm->cur.zone->list.next,
892 struct mem_zone_bm_rtree, list);
893 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
894 struct rtree_node, list);
895 bm->cur.node_pfn = 0;
896 bm->cur.node_bit = 0;
897 return true;
898 }
899
900 /* No more zones */
901 return false;
902}
903
904/**
905 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
906 * @bm: Memory bitmap.
907 *
908 * Starting from the last returned position this function searches for the next
909 * set bit in @bm and returns the PFN represented by it. If no more bits are
910 * set, BM_END_OF_MAP is returned.
911 *
912 * It is required to run memory_bm_position_reset() before the first call to
913 * this function for the given memory bitmap.
914 */
915static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
916{
917 unsigned long bits, pfn, pages;
918 int bit;
919
920 do {
921 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
922 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
923 bit = find_next_bit(bm->cur.node->data, bits,
924 bm->cur.node_bit);
925 if (bit < bits) {
926 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
927 bm->cur.node_bit = bit + 1;
928 return pfn;
929 }
930 } while (rtree_next_node(bm));
931
932 return BM_END_OF_MAP;
933}
934
935/*
936 * This structure represents a range of page frames the contents of which
937 * should not be saved during hibernation.
938 */
939struct nosave_region {
940 struct list_head list;
941 unsigned long start_pfn;
942 unsigned long end_pfn;
943};
944
945static LIST_HEAD(nosave_regions);
946
947static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
948{
949 struct rtree_node *node;
950
951 list_for_each_entry(node, &zone->nodes, list)
952 recycle_safe_page(node->data);
953
954 list_for_each_entry(node, &zone->leaves, list)
955 recycle_safe_page(node->data);
956}
957
958static void memory_bm_recycle(struct memory_bitmap *bm)
959{
960 struct mem_zone_bm_rtree *zone;
961 struct linked_page *p_list;
962
963 list_for_each_entry(zone, &bm->zones, list)
964 recycle_zone_bm_rtree(zone);
965
966 p_list = bm->p_list;
967 while (p_list) {
968 struct linked_page *lp = p_list;
969
970 p_list = lp->next;
971 recycle_safe_page(lp);
972 }
973}
974
975/**
976 * register_nosave_region - Register a region of unsaveable memory.
977 *
978 * Register a range of page frames the contents of which should not be saved
979 * during hibernation (to be used in the early initialization code).
980 */
981void __init __register_nosave_region(unsigned long start_pfn,
982 unsigned long end_pfn, int use_kmalloc)
983{
984 struct nosave_region *region;
985
986 if (start_pfn >= end_pfn)
987 return;
988
989 if (!list_empty(&nosave_regions)) {
990 /* Try to extend the previous region (they should be sorted) */
991 region = list_entry(nosave_regions.prev,
992 struct nosave_region, list);
993 if (region->end_pfn == start_pfn) {
994 region->end_pfn = end_pfn;
995 goto Report;
996 }
997 }
998 if (use_kmalloc) {
999 /* During init, this shouldn't fail */
1000 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
1001 BUG_ON(!region);
1002 } else {
1003 /* This allocation cannot fail */
1004 region = memblock_alloc(sizeof(struct nosave_region),
1005 SMP_CACHE_BYTES);
1006 if (!region)
1007 panic("%s: Failed to allocate %zu bytes\n", __func__,
1008 sizeof(struct nosave_region));
1009 }
1010 region->start_pfn = start_pfn;
1011 region->end_pfn = end_pfn;
1012 list_add_tail(®ion->list, &nosave_regions);
1013 Report:
1014 pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
1015 (unsigned long long) start_pfn << PAGE_SHIFT,
1016 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1017}
1018
1019/*
1020 * Set bits in this map correspond to the page frames the contents of which
1021 * should not be saved during the suspend.
1022 */
1023static struct memory_bitmap *forbidden_pages_map;
1024
1025/* Set bits in this map correspond to free page frames. */
1026static struct memory_bitmap *free_pages_map;
1027
1028/*
1029 * Each page frame allocated for creating the image is marked by setting the
1030 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
1031 */
1032
1033void swsusp_set_page_free(struct page *page)
1034{
1035 if (free_pages_map)
1036 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
1037}
1038
1039static int swsusp_page_is_free(struct page *page)
1040{
1041 return free_pages_map ?
1042 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1043}
1044
1045void swsusp_unset_page_free(struct page *page)
1046{
1047 if (free_pages_map)
1048 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1049}
1050
1051static void swsusp_set_page_forbidden(struct page *page)
1052{
1053 if (forbidden_pages_map)
1054 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1055}
1056
1057int swsusp_page_is_forbidden(struct page *page)
1058{
1059 return forbidden_pages_map ?
1060 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1061}
1062
1063static void swsusp_unset_page_forbidden(struct page *page)
1064{
1065 if (forbidden_pages_map)
1066 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1067}
1068
1069/**
1070 * mark_nosave_pages - Mark pages that should not be saved.
1071 * @bm: Memory bitmap.
1072 *
1073 * Set the bits in @bm that correspond to the page frames the contents of which
1074 * should not be saved.
1075 */
1076static void mark_nosave_pages(struct memory_bitmap *bm)
1077{
1078 struct nosave_region *region;
1079
1080 if (list_empty(&nosave_regions))
1081 return;
1082
1083 list_for_each_entry(region, &nosave_regions, list) {
1084 unsigned long pfn;
1085
1086 pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1087 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1088 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1089 - 1);
1090
1091 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1092 if (pfn_valid(pfn)) {
1093 /*
1094 * It is safe to ignore the result of
1095 * mem_bm_set_bit_check() here, since we won't
1096 * touch the PFNs for which the error is
1097 * returned anyway.
1098 */
1099 mem_bm_set_bit_check(bm, pfn);
1100 }
1101 }
1102}
1103
1104/**
1105 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1106 *
1107 * Create bitmaps needed for marking page frames that should not be saved and
1108 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1109 * only modified if everything goes well, because we don't want the bits to be
1110 * touched before both bitmaps are set up.
1111 */
1112int create_basic_memory_bitmaps(void)
1113{
1114 struct memory_bitmap *bm1, *bm2;
1115 int error = 0;
1116
1117 if (forbidden_pages_map && free_pages_map)
1118 return 0;
1119 else
1120 BUG_ON(forbidden_pages_map || free_pages_map);
1121
1122 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1123 if (!bm1)
1124 return -ENOMEM;
1125
1126 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1127 if (error)
1128 goto Free_first_object;
1129
1130 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1131 if (!bm2)
1132 goto Free_first_bitmap;
1133
1134 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1135 if (error)
1136 goto Free_second_object;
1137
1138 forbidden_pages_map = bm1;
1139 free_pages_map = bm2;
1140 mark_nosave_pages(forbidden_pages_map);
1141
1142 pr_debug("Basic memory bitmaps created\n");
1143
1144 return 0;
1145
1146 Free_second_object:
1147 kfree(bm2);
1148 Free_first_bitmap:
1149 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1150 Free_first_object:
1151 kfree(bm1);
1152 return -ENOMEM;
1153}
1154
1155/**
1156 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1157 *
1158 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1159 * auxiliary pointers are necessary so that the bitmaps themselves are not
1160 * referred to while they are being freed.
1161 */
1162void free_basic_memory_bitmaps(void)
1163{
1164 struct memory_bitmap *bm1, *bm2;
1165
1166 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1167 return;
1168
1169 bm1 = forbidden_pages_map;
1170 bm2 = free_pages_map;
1171 forbidden_pages_map = NULL;
1172 free_pages_map = NULL;
1173 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1174 kfree(bm1);
1175 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1176 kfree(bm2);
1177
1178 pr_debug("Basic memory bitmaps freed\n");
1179}
1180
1181static void clear_or_poison_free_page(struct page *page)
1182{
1183 if (page_poisoning_enabled_static())
1184 __kernel_poison_pages(page, 1);
1185 else if (want_init_on_free())
1186 clear_highpage(page);
1187}
1188
1189void clear_or_poison_free_pages(void)
1190{
1191 struct memory_bitmap *bm = free_pages_map;
1192 unsigned long pfn;
1193
1194 if (WARN_ON(!(free_pages_map)))
1195 return;
1196
1197 if (page_poisoning_enabled() || want_init_on_free()) {
1198 memory_bm_position_reset(bm);
1199 pfn = memory_bm_next_pfn(bm);
1200 while (pfn != BM_END_OF_MAP) {
1201 if (pfn_valid(pfn))
1202 clear_or_poison_free_page(pfn_to_page(pfn));
1203
1204 pfn = memory_bm_next_pfn(bm);
1205 }
1206 memory_bm_position_reset(bm);
1207 pr_info("free pages cleared after restore\n");
1208 }
1209}
1210
1211/**
1212 * snapshot_additional_pages - Estimate the number of extra pages needed.
1213 * @zone: Memory zone to carry out the computation for.
1214 *
1215 * Estimate the number of additional pages needed for setting up a hibernation
1216 * image data structures for @zone (usually, the returned value is greater than
1217 * the exact number).
1218 */
1219unsigned int snapshot_additional_pages(struct zone *zone)
1220{
1221 unsigned int rtree, nodes;
1222
1223 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1224 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1225 LINKED_PAGE_DATA_SIZE);
1226 while (nodes > 1) {
1227 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1228 rtree += nodes;
1229 }
1230
1231 return 2 * rtree;
1232}
1233
1234#ifdef CONFIG_HIGHMEM
1235/**
1236 * count_free_highmem_pages - Compute the total number of free highmem pages.
1237 *
1238 * The returned number is system-wide.
1239 */
1240static unsigned int count_free_highmem_pages(void)
1241{
1242 struct zone *zone;
1243 unsigned int cnt = 0;
1244
1245 for_each_populated_zone(zone)
1246 if (is_highmem(zone))
1247 cnt += zone_page_state(zone, NR_FREE_PAGES);
1248
1249 return cnt;
1250}
1251
1252/**
1253 * saveable_highmem_page - Check if a highmem page is saveable.
1254 *
1255 * Determine whether a highmem page should be included in a hibernation image.
1256 *
1257 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1258 * and it isn't part of a free chunk of pages.
1259 */
1260static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1261{
1262 struct page *page;
1263
1264 if (!pfn_valid(pfn))
1265 return NULL;
1266
1267 page = pfn_to_online_page(pfn);
1268 if (!page || page_zone(page) != zone)
1269 return NULL;
1270
1271 BUG_ON(!PageHighMem(page));
1272
1273 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1274 return NULL;
1275
1276 if (PageReserved(page) || PageOffline(page))
1277 return NULL;
1278
1279 if (page_is_guard(page))
1280 return NULL;
1281
1282 return page;
1283}
1284
1285/**
1286 * count_highmem_pages - Compute the total number of saveable highmem pages.
1287 */
1288static unsigned int count_highmem_pages(void)
1289{
1290 struct zone *zone;
1291 unsigned int n = 0;
1292
1293 for_each_populated_zone(zone) {
1294 unsigned long pfn, max_zone_pfn;
1295
1296 if (!is_highmem(zone))
1297 continue;
1298
1299 mark_free_pages(zone);
1300 max_zone_pfn = zone_end_pfn(zone);
1301 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1302 if (saveable_highmem_page(zone, pfn))
1303 n++;
1304 }
1305 return n;
1306}
1307#else
1308static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1309{
1310 return NULL;
1311}
1312#endif /* CONFIG_HIGHMEM */
1313
1314/**
1315 * saveable_page - Check if the given page is saveable.
1316 *
1317 * Determine whether a non-highmem page should be included in a hibernation
1318 * image.
1319 *
1320 * We should save the page if it isn't Nosave, and is not in the range
1321 * of pages statically defined as 'unsaveable', and it isn't part of
1322 * a free chunk of pages.
1323 */
1324static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1325{
1326 struct page *page;
1327
1328 if (!pfn_valid(pfn))
1329 return NULL;
1330
1331 page = pfn_to_online_page(pfn);
1332 if (!page || page_zone(page) != zone)
1333 return NULL;
1334
1335 BUG_ON(PageHighMem(page));
1336
1337 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1338 return NULL;
1339
1340 if (PageOffline(page))
1341 return NULL;
1342
1343 if (PageReserved(page)
1344 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1345 return NULL;
1346
1347 if (page_is_guard(page))
1348 return NULL;
1349
1350 return page;
1351}
1352
1353/**
1354 * count_data_pages - Compute the total number of saveable non-highmem pages.
1355 */
1356static unsigned int count_data_pages(void)
1357{
1358 struct zone *zone;
1359 unsigned long pfn, max_zone_pfn;
1360 unsigned int n = 0;
1361
1362 for_each_populated_zone(zone) {
1363 if (is_highmem(zone))
1364 continue;
1365
1366 mark_free_pages(zone);
1367 max_zone_pfn = zone_end_pfn(zone);
1368 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1369 if (saveable_page(zone, pfn))
1370 n++;
1371 }
1372 return n;
1373}
1374
1375/*
1376 * This is needed, because copy_page and memcpy are not usable for copying
1377 * task structs.
1378 */
1379static inline void do_copy_page(long *dst, long *src)
1380{
1381 int n;
1382
1383 for (n = PAGE_SIZE / sizeof(long); n; n--)
1384 *dst++ = *src++;
1385}
1386
1387/**
1388 * safe_copy_page - Copy a page in a safe way.
1389 *
1390 * Check if the page we are going to copy is marked as present in the kernel
1391 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1392 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1393 * always returns 'true'.
1394 */
1395static void safe_copy_page(void *dst, struct page *s_page)
1396{
1397 if (kernel_page_present(s_page)) {
1398 do_copy_page(dst, page_address(s_page));
1399 } else {
1400 hibernate_map_page(s_page);
1401 do_copy_page(dst, page_address(s_page));
1402 hibernate_unmap_page(s_page);
1403 }
1404}
1405
1406#ifdef CONFIG_HIGHMEM
1407static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1408{
1409 return is_highmem(zone) ?
1410 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1411}
1412
1413static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1414{
1415 struct page *s_page, *d_page;
1416 void *src, *dst;
1417
1418 s_page = pfn_to_page(src_pfn);
1419 d_page = pfn_to_page(dst_pfn);
1420 if (PageHighMem(s_page)) {
1421 src = kmap_atomic(s_page);
1422 dst = kmap_atomic(d_page);
1423 do_copy_page(dst, src);
1424 kunmap_atomic(dst);
1425 kunmap_atomic(src);
1426 } else {
1427 if (PageHighMem(d_page)) {
1428 /*
1429 * The page pointed to by src may contain some kernel
1430 * data modified by kmap_atomic()
1431 */
1432 safe_copy_page(buffer, s_page);
1433 dst = kmap_atomic(d_page);
1434 copy_page(dst, buffer);
1435 kunmap_atomic(dst);
1436 } else {
1437 safe_copy_page(page_address(d_page), s_page);
1438 }
1439 }
1440}
1441#else
1442#define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1443
1444static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1445{
1446 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1447 pfn_to_page(src_pfn));
1448}
1449#endif /* CONFIG_HIGHMEM */
1450
1451static void copy_data_pages(struct memory_bitmap *copy_bm,
1452 struct memory_bitmap *orig_bm)
1453{
1454 struct zone *zone;
1455 unsigned long pfn;
1456
1457 for_each_populated_zone(zone) {
1458 unsigned long max_zone_pfn;
1459
1460 mark_free_pages(zone);
1461 max_zone_pfn = zone_end_pfn(zone);
1462 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1463 if (page_is_saveable(zone, pfn))
1464 memory_bm_set_bit(orig_bm, pfn);
1465 }
1466 memory_bm_position_reset(orig_bm);
1467 memory_bm_position_reset(copy_bm);
1468 for(;;) {
1469 pfn = memory_bm_next_pfn(orig_bm);
1470 if (unlikely(pfn == BM_END_OF_MAP))
1471 break;
1472 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1473 }
1474}
1475
1476/* Total number of image pages */
1477static unsigned int nr_copy_pages;
1478/* Number of pages needed for saving the original pfns of the image pages */
1479static unsigned int nr_meta_pages;
1480/*
1481 * Numbers of normal and highmem page frames allocated for hibernation image
1482 * before suspending devices.
1483 */
1484static unsigned int alloc_normal, alloc_highmem;
1485/*
1486 * Memory bitmap used for marking saveable pages (during hibernation) or
1487 * hibernation image pages (during restore)
1488 */
1489static struct memory_bitmap orig_bm;
1490/*
1491 * Memory bitmap used during hibernation for marking allocated page frames that
1492 * will contain copies of saveable pages. During restore it is initially used
1493 * for marking hibernation image pages, but then the set bits from it are
1494 * duplicated in @orig_bm and it is released. On highmem systems it is next
1495 * used for marking "safe" highmem pages, but it has to be reinitialized for
1496 * this purpose.
1497 */
1498static struct memory_bitmap copy_bm;
1499
1500/**
1501 * swsusp_free - Free pages allocated for hibernation image.
1502 *
1503 * Image pages are allocated before snapshot creation, so they need to be
1504 * released after resume.
1505 */
1506void swsusp_free(void)
1507{
1508 unsigned long fb_pfn, fr_pfn;
1509
1510 if (!forbidden_pages_map || !free_pages_map)
1511 goto out;
1512
1513 memory_bm_position_reset(forbidden_pages_map);
1514 memory_bm_position_reset(free_pages_map);
1515
1516loop:
1517 fr_pfn = memory_bm_next_pfn(free_pages_map);
1518 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1519
1520 /*
1521 * Find the next bit set in both bitmaps. This is guaranteed to
1522 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1523 */
1524 do {
1525 if (fb_pfn < fr_pfn)
1526 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1527 if (fr_pfn < fb_pfn)
1528 fr_pfn = memory_bm_next_pfn(free_pages_map);
1529 } while (fb_pfn != fr_pfn);
1530
1531 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1532 struct page *page = pfn_to_page(fr_pfn);
1533
1534 memory_bm_clear_current(forbidden_pages_map);
1535 memory_bm_clear_current(free_pages_map);
1536 hibernate_restore_unprotect_page(page_address(page));
1537 __free_page(page);
1538 goto loop;
1539 }
1540
1541out:
1542 nr_copy_pages = 0;
1543 nr_meta_pages = 0;
1544 restore_pblist = NULL;
1545 buffer = NULL;
1546 alloc_normal = 0;
1547 alloc_highmem = 0;
1548 hibernate_restore_protection_end();
1549}
1550
1551/* Helper functions used for the shrinking of memory. */
1552
1553#define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1554
1555/**
1556 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1557 * @nr_pages: Number of page frames to allocate.
1558 * @mask: GFP flags to use for the allocation.
1559 *
1560 * Return value: Number of page frames actually allocated
1561 */
1562static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1563{
1564 unsigned long nr_alloc = 0;
1565
1566 while (nr_pages > 0) {
1567 struct page *page;
1568
1569 page = alloc_image_page(mask);
1570 if (!page)
1571 break;
1572 memory_bm_set_bit(©_bm, page_to_pfn(page));
1573 if (PageHighMem(page))
1574 alloc_highmem++;
1575 else
1576 alloc_normal++;
1577 nr_pages--;
1578 nr_alloc++;
1579 }
1580
1581 return nr_alloc;
1582}
1583
1584static unsigned long preallocate_image_memory(unsigned long nr_pages,
1585 unsigned long avail_normal)
1586{
1587 unsigned long alloc;
1588
1589 if (avail_normal <= alloc_normal)
1590 return 0;
1591
1592 alloc = avail_normal - alloc_normal;
1593 if (nr_pages < alloc)
1594 alloc = nr_pages;
1595
1596 return preallocate_image_pages(alloc, GFP_IMAGE);
1597}
1598
1599#ifdef CONFIG_HIGHMEM
1600static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1601{
1602 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1603}
1604
1605/**
1606 * __fraction - Compute (an approximation of) x * (multiplier / base).
1607 */
1608static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1609{
1610 return div64_u64(x * multiplier, base);
1611}
1612
1613static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1614 unsigned long highmem,
1615 unsigned long total)
1616{
1617 unsigned long alloc = __fraction(nr_pages, highmem, total);
1618
1619 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1620}
1621#else /* CONFIG_HIGHMEM */
1622static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1623{
1624 return 0;
1625}
1626
1627static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1628 unsigned long highmem,
1629 unsigned long total)
1630{
1631 return 0;
1632}
1633#endif /* CONFIG_HIGHMEM */
1634
1635/**
1636 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1637 */
1638static unsigned long free_unnecessary_pages(void)
1639{
1640 unsigned long save, to_free_normal, to_free_highmem, free;
1641
1642 save = count_data_pages();
1643 if (alloc_normal >= save) {
1644 to_free_normal = alloc_normal - save;
1645 save = 0;
1646 } else {
1647 to_free_normal = 0;
1648 save -= alloc_normal;
1649 }
1650 save += count_highmem_pages();
1651 if (alloc_highmem >= save) {
1652 to_free_highmem = alloc_highmem - save;
1653 } else {
1654 to_free_highmem = 0;
1655 save -= alloc_highmem;
1656 if (to_free_normal > save)
1657 to_free_normal -= save;
1658 else
1659 to_free_normal = 0;
1660 }
1661 free = to_free_normal + to_free_highmem;
1662
1663 memory_bm_position_reset(©_bm);
1664
1665 while (to_free_normal > 0 || to_free_highmem > 0) {
1666 unsigned long pfn = memory_bm_next_pfn(©_bm);
1667 struct page *page = pfn_to_page(pfn);
1668
1669 if (PageHighMem(page)) {
1670 if (!to_free_highmem)
1671 continue;
1672 to_free_highmem--;
1673 alloc_highmem--;
1674 } else {
1675 if (!to_free_normal)
1676 continue;
1677 to_free_normal--;
1678 alloc_normal--;
1679 }
1680 memory_bm_clear_bit(©_bm, pfn);
1681 swsusp_unset_page_forbidden(page);
1682 swsusp_unset_page_free(page);
1683 __free_page(page);
1684 }
1685
1686 return free;
1687}
1688
1689/**
1690 * minimum_image_size - Estimate the minimum acceptable size of an image.
1691 * @saveable: Number of saveable pages in the system.
1692 *
1693 * We want to avoid attempting to free too much memory too hard, so estimate the
1694 * minimum acceptable size of a hibernation image to use as the lower limit for
1695 * preallocating memory.
1696 *
1697 * We assume that the minimum image size should be proportional to
1698 *
1699 * [number of saveable pages] - [number of pages that can be freed in theory]
1700 *
1701 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1702 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1703 */
1704static unsigned long minimum_image_size(unsigned long saveable)
1705{
1706 unsigned long size;
1707
1708 size = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B)
1709 + global_node_page_state(NR_ACTIVE_ANON)
1710 + global_node_page_state(NR_INACTIVE_ANON)
1711 + global_node_page_state(NR_ACTIVE_FILE)
1712 + global_node_page_state(NR_INACTIVE_FILE);
1713
1714 return saveable <= size ? 0 : saveable - size;
1715}
1716
1717/**
1718 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1719 *
1720 * To create a hibernation image it is necessary to make a copy of every page
1721 * frame in use. We also need a number of page frames to be free during
1722 * hibernation for allocations made while saving the image and for device
1723 * drivers, in case they need to allocate memory from their hibernation
1724 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1725 * estimate) and reserved_size divided by PAGE_SIZE (which is tunable through
1726 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1727 * total number of available page frames and allocate at least
1728 *
1729 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1730 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1731 *
1732 * of them, which corresponds to the maximum size of a hibernation image.
1733 *
1734 * If image_size is set below the number following from the above formula,
1735 * the preallocation of memory is continued until the total number of saveable
1736 * pages in the system is below the requested image size or the minimum
1737 * acceptable image size returned by minimum_image_size(), whichever is greater.
1738 */
1739int hibernate_preallocate_memory(void)
1740{
1741 struct zone *zone;
1742 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1743 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1744 ktime_t start, stop;
1745 int error;
1746
1747 pr_info("Preallocating image memory\n");
1748 start = ktime_get();
1749
1750 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1751 if (error) {
1752 pr_err("Cannot allocate original bitmap\n");
1753 goto err_out;
1754 }
1755
1756 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1757 if (error) {
1758 pr_err("Cannot allocate copy bitmap\n");
1759 goto err_out;
1760 }
1761
1762 alloc_normal = 0;
1763 alloc_highmem = 0;
1764
1765 /* Count the number of saveable data pages. */
1766 save_highmem = count_highmem_pages();
1767 saveable = count_data_pages();
1768
1769 /*
1770 * Compute the total number of page frames we can use (count) and the
1771 * number of pages needed for image metadata (size).
1772 */
1773 count = saveable;
1774 saveable += save_highmem;
1775 highmem = save_highmem;
1776 size = 0;
1777 for_each_populated_zone(zone) {
1778 size += snapshot_additional_pages(zone);
1779 if (is_highmem(zone))
1780 highmem += zone_page_state(zone, NR_FREE_PAGES);
1781 else
1782 count += zone_page_state(zone, NR_FREE_PAGES);
1783 }
1784 avail_normal = count;
1785 count += highmem;
1786 count -= totalreserve_pages;
1787
1788 /* Compute the maximum number of saveable pages to leave in memory. */
1789 max_size = (count - (size + PAGES_FOR_IO)) / 2
1790 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1791 /* Compute the desired number of image pages specified by image_size. */
1792 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1793 if (size > max_size)
1794 size = max_size;
1795 /*
1796 * If the desired number of image pages is at least as large as the
1797 * current number of saveable pages in memory, allocate page frames for
1798 * the image and we're done.
1799 */
1800 if (size >= saveable) {
1801 pages = preallocate_image_highmem(save_highmem);
1802 pages += preallocate_image_memory(saveable - pages, avail_normal);
1803 goto out;
1804 }
1805
1806 /* Estimate the minimum size of the image. */
1807 pages = minimum_image_size(saveable);
1808 /*
1809 * To avoid excessive pressure on the normal zone, leave room in it to
1810 * accommodate an image of the minimum size (unless it's already too
1811 * small, in which case don't preallocate pages from it at all).
1812 */
1813 if (avail_normal > pages)
1814 avail_normal -= pages;
1815 else
1816 avail_normal = 0;
1817 if (size < pages)
1818 size = min_t(unsigned long, pages, max_size);
1819
1820 /*
1821 * Let the memory management subsystem know that we're going to need a
1822 * large number of page frames to allocate and make it free some memory.
1823 * NOTE: If this is not done, performance will be hurt badly in some
1824 * test cases.
1825 */
1826 shrink_all_memory(saveable - size);
1827
1828 /*
1829 * The number of saveable pages in memory was too high, so apply some
1830 * pressure to decrease it. First, make room for the largest possible
1831 * image and fail if that doesn't work. Next, try to decrease the size
1832 * of the image as much as indicated by 'size' using allocations from
1833 * highmem and non-highmem zones separately.
1834 */
1835 pages_highmem = preallocate_image_highmem(highmem / 2);
1836 alloc = count - max_size;
1837 if (alloc > pages_highmem)
1838 alloc -= pages_highmem;
1839 else
1840 alloc = 0;
1841 pages = preallocate_image_memory(alloc, avail_normal);
1842 if (pages < alloc) {
1843 /* We have exhausted non-highmem pages, try highmem. */
1844 alloc -= pages;
1845 pages += pages_highmem;
1846 pages_highmem = preallocate_image_highmem(alloc);
1847 if (pages_highmem < alloc) {
1848 pr_err("Image allocation is %lu pages short\n",
1849 alloc - pages_highmem);
1850 goto err_out;
1851 }
1852 pages += pages_highmem;
1853 /*
1854 * size is the desired number of saveable pages to leave in
1855 * memory, so try to preallocate (all memory - size) pages.
1856 */
1857 alloc = (count - pages) - size;
1858 pages += preallocate_image_highmem(alloc);
1859 } else {
1860 /*
1861 * There are approximately max_size saveable pages at this point
1862 * and we want to reduce this number down to size.
1863 */
1864 alloc = max_size - size;
1865 size = preallocate_highmem_fraction(alloc, highmem, count);
1866 pages_highmem += size;
1867 alloc -= size;
1868 size = preallocate_image_memory(alloc, avail_normal);
1869 pages_highmem += preallocate_image_highmem(alloc - size);
1870 pages += pages_highmem + size;
1871 }
1872
1873 /*
1874 * We only need as many page frames for the image as there are saveable
1875 * pages in memory, but we have allocated more. Release the excessive
1876 * ones now.
1877 */
1878 pages -= free_unnecessary_pages();
1879
1880 out:
1881 stop = ktime_get();
1882 pr_info("Allocated %lu pages for snapshot\n", pages);
1883 swsusp_show_speed(start, stop, pages, "Allocated");
1884
1885 return 0;
1886
1887 err_out:
1888 swsusp_free();
1889 return -ENOMEM;
1890}
1891
1892#ifdef CONFIG_HIGHMEM
1893/**
1894 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1895 *
1896 * Compute the number of non-highmem pages that will be necessary for creating
1897 * copies of highmem pages.
1898 */
1899static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1900{
1901 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1902
1903 if (free_highmem >= nr_highmem)
1904 nr_highmem = 0;
1905 else
1906 nr_highmem -= free_highmem;
1907
1908 return nr_highmem;
1909}
1910#else
1911static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1912#endif /* CONFIG_HIGHMEM */
1913
1914/**
1915 * enough_free_mem - Check if there is enough free memory for the image.
1916 */
1917static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1918{
1919 struct zone *zone;
1920 unsigned int free = alloc_normal;
1921
1922 for_each_populated_zone(zone)
1923 if (!is_highmem(zone))
1924 free += zone_page_state(zone, NR_FREE_PAGES);
1925
1926 nr_pages += count_pages_for_highmem(nr_highmem);
1927 pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
1928 nr_pages, PAGES_FOR_IO, free);
1929
1930 return free > nr_pages + PAGES_FOR_IO;
1931}
1932
1933#ifdef CONFIG_HIGHMEM
1934/**
1935 * get_highmem_buffer - Allocate a buffer for highmem pages.
1936 *
1937 * If there are some highmem pages in the hibernation image, we may need a
1938 * buffer to copy them and/or load their data.
1939 */
1940static inline int get_highmem_buffer(int safe_needed)
1941{
1942 buffer = get_image_page(GFP_ATOMIC, safe_needed);
1943 return buffer ? 0 : -ENOMEM;
1944}
1945
1946/**
1947 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1948 *
1949 * Try to allocate as many pages as needed, but if the number of free highmem
1950 * pages is less than that, allocate them all.
1951 */
1952static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1953 unsigned int nr_highmem)
1954{
1955 unsigned int to_alloc = count_free_highmem_pages();
1956
1957 if (to_alloc > nr_highmem)
1958 to_alloc = nr_highmem;
1959
1960 nr_highmem -= to_alloc;
1961 while (to_alloc-- > 0) {
1962 struct page *page;
1963
1964 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1965 memory_bm_set_bit(bm, page_to_pfn(page));
1966 }
1967 return nr_highmem;
1968}
1969#else
1970static inline int get_highmem_buffer(int safe_needed) { return 0; }
1971
1972static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1973 unsigned int n) { return 0; }
1974#endif /* CONFIG_HIGHMEM */
1975
1976/**
1977 * swsusp_alloc - Allocate memory for hibernation image.
1978 *
1979 * We first try to allocate as many highmem pages as there are
1980 * saveable highmem pages in the system. If that fails, we allocate
1981 * non-highmem pages for the copies of the remaining highmem ones.
1982 *
1983 * In this approach it is likely that the copies of highmem pages will
1984 * also be located in the high memory, because of the way in which
1985 * copy_data_pages() works.
1986 */
1987static int swsusp_alloc(struct memory_bitmap *copy_bm,
1988 unsigned int nr_pages, unsigned int nr_highmem)
1989{
1990 if (nr_highmem > 0) {
1991 if (get_highmem_buffer(PG_ANY))
1992 goto err_out;
1993 if (nr_highmem > alloc_highmem) {
1994 nr_highmem -= alloc_highmem;
1995 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1996 }
1997 }
1998 if (nr_pages > alloc_normal) {
1999 nr_pages -= alloc_normal;
2000 while (nr_pages-- > 0) {
2001 struct page *page;
2002
2003 page = alloc_image_page(GFP_ATOMIC);
2004 if (!page)
2005 goto err_out;
2006 memory_bm_set_bit(copy_bm, page_to_pfn(page));
2007 }
2008 }
2009
2010 return 0;
2011
2012 err_out:
2013 swsusp_free();
2014 return -ENOMEM;
2015}
2016
2017asmlinkage __visible int swsusp_save(void)
2018{
2019 unsigned int nr_pages, nr_highmem;
2020
2021 pr_info("Creating image:\n");
2022
2023 drain_local_pages(NULL);
2024 nr_pages = count_data_pages();
2025 nr_highmem = count_highmem_pages();
2026 pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
2027
2028 if (!enough_free_mem(nr_pages, nr_highmem)) {
2029 pr_err("Not enough free memory\n");
2030 return -ENOMEM;
2031 }
2032
2033 if (swsusp_alloc(©_bm, nr_pages, nr_highmem)) {
2034 pr_err("Memory allocation failed\n");
2035 return -ENOMEM;
2036 }
2037
2038 /*
2039 * During allocating of suspend pagedir, new cold pages may appear.
2040 * Kill them.
2041 */
2042 drain_local_pages(NULL);
2043 copy_data_pages(©_bm, &orig_bm);
2044
2045 /*
2046 * End of critical section. From now on, we can write to memory,
2047 * but we should not touch disk. This specially means we must _not_
2048 * touch swap space! Except we must write out our image of course.
2049 */
2050
2051 nr_pages += nr_highmem;
2052 nr_copy_pages = nr_pages;
2053 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2054
2055 pr_info("Image created (%d pages copied)\n", nr_pages);
2056
2057 return 0;
2058}
2059
2060#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2061static int init_header_complete(struct swsusp_info *info)
2062{
2063 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2064 info->version_code = LINUX_VERSION_CODE;
2065 return 0;
2066}
2067
2068static const char *check_image_kernel(struct swsusp_info *info)
2069{
2070 if (info->version_code != LINUX_VERSION_CODE)
2071 return "kernel version";
2072 if (strcmp(info->uts.sysname,init_utsname()->sysname))
2073 return "system type";
2074 if (strcmp(info->uts.release,init_utsname()->release))
2075 return "kernel release";
2076 if (strcmp(info->uts.version,init_utsname()->version))
2077 return "version";
2078 if (strcmp(info->uts.machine,init_utsname()->machine))
2079 return "machine";
2080 return NULL;
2081}
2082#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2083
2084unsigned long snapshot_get_image_size(void)
2085{
2086 return nr_copy_pages + nr_meta_pages + 1;
2087}
2088
2089static int init_header(struct swsusp_info *info)
2090{
2091 memset(info, 0, sizeof(struct swsusp_info));
2092 info->num_physpages = get_num_physpages();
2093 info->image_pages = nr_copy_pages;
2094 info->pages = snapshot_get_image_size();
2095 info->size = info->pages;
2096 info->size <<= PAGE_SHIFT;
2097 return init_header_complete(info);
2098}
2099
2100/**
2101 * pack_pfns - Prepare PFNs for saving.
2102 * @bm: Memory bitmap.
2103 * @buf: Memory buffer to store the PFNs in.
2104 *
2105 * PFNs corresponding to set bits in @bm are stored in the area of memory
2106 * pointed to by @buf (1 page at a time).
2107 */
2108static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2109{
2110 int j;
2111
2112 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2113 buf[j] = memory_bm_next_pfn(bm);
2114 if (unlikely(buf[j] == BM_END_OF_MAP))
2115 break;
2116 }
2117}
2118
2119/**
2120 * snapshot_read_next - Get the address to read the next image page from.
2121 * @handle: Snapshot handle to be used for the reading.
2122 *
2123 * On the first call, @handle should point to a zeroed snapshot_handle
2124 * structure. The structure gets populated then and a pointer to it should be
2125 * passed to this function every next time.
2126 *
2127 * On success, the function returns a positive number. Then, the caller
2128 * is allowed to read up to the returned number of bytes from the memory
2129 * location computed by the data_of() macro.
2130 *
2131 * The function returns 0 to indicate the end of the data stream condition,
2132 * and negative numbers are returned on errors. If that happens, the structure
2133 * pointed to by @handle is not updated and should not be used any more.
2134 */
2135int snapshot_read_next(struct snapshot_handle *handle)
2136{
2137 if (handle->cur > nr_meta_pages + nr_copy_pages)
2138 return 0;
2139
2140 if (!buffer) {
2141 /* This makes the buffer be freed by swsusp_free() */
2142 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2143 if (!buffer)
2144 return -ENOMEM;
2145 }
2146 if (!handle->cur) {
2147 int error;
2148
2149 error = init_header((struct swsusp_info *)buffer);
2150 if (error)
2151 return error;
2152 handle->buffer = buffer;
2153 memory_bm_position_reset(&orig_bm);
2154 memory_bm_position_reset(©_bm);
2155 } else if (handle->cur <= nr_meta_pages) {
2156 clear_page(buffer);
2157 pack_pfns(buffer, &orig_bm);
2158 } else {
2159 struct page *page;
2160
2161 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2162 if (PageHighMem(page)) {
2163 /*
2164 * Highmem pages are copied to the buffer,
2165 * because we can't return with a kmapped
2166 * highmem page (we may not be called again).
2167 */
2168 void *kaddr;
2169
2170 kaddr = kmap_atomic(page);
2171 copy_page(buffer, kaddr);
2172 kunmap_atomic(kaddr);
2173 handle->buffer = buffer;
2174 } else {
2175 handle->buffer = page_address(page);
2176 }
2177 }
2178 handle->cur++;
2179 return PAGE_SIZE;
2180}
2181
2182static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2183 struct memory_bitmap *src)
2184{
2185 unsigned long pfn;
2186
2187 memory_bm_position_reset(src);
2188 pfn = memory_bm_next_pfn(src);
2189 while (pfn != BM_END_OF_MAP) {
2190 memory_bm_set_bit(dst, pfn);
2191 pfn = memory_bm_next_pfn(src);
2192 }
2193}
2194
2195/**
2196 * mark_unsafe_pages - Mark pages that were used before hibernation.
2197 *
2198 * Mark the pages that cannot be used for storing the image during restoration,
2199 * because they conflict with the pages that had been used before hibernation.
2200 */
2201static void mark_unsafe_pages(struct memory_bitmap *bm)
2202{
2203 unsigned long pfn;
2204
2205 /* Clear the "free"/"unsafe" bit for all PFNs */
2206 memory_bm_position_reset(free_pages_map);
2207 pfn = memory_bm_next_pfn(free_pages_map);
2208 while (pfn != BM_END_OF_MAP) {
2209 memory_bm_clear_current(free_pages_map);
2210 pfn = memory_bm_next_pfn(free_pages_map);
2211 }
2212
2213 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2214 duplicate_memory_bitmap(free_pages_map, bm);
2215
2216 allocated_unsafe_pages = 0;
2217}
2218
2219static int check_header(struct swsusp_info *info)
2220{
2221 const char *reason;
2222
2223 reason = check_image_kernel(info);
2224 if (!reason && info->num_physpages != get_num_physpages())
2225 reason = "memory size";
2226 if (reason) {
2227 pr_err("Image mismatch: %s\n", reason);
2228 return -EPERM;
2229 }
2230 return 0;
2231}
2232
2233/**
2234 * load header - Check the image header and copy the data from it.
2235 */
2236static int load_header(struct swsusp_info *info)
2237{
2238 int error;
2239
2240 restore_pblist = NULL;
2241 error = check_header(info);
2242 if (!error) {
2243 nr_copy_pages = info->image_pages;
2244 nr_meta_pages = info->pages - info->image_pages - 1;
2245 }
2246 return error;
2247}
2248
2249/**
2250 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2251 * @bm: Memory bitmap.
2252 * @buf: Area of memory containing the PFNs.
2253 *
2254 * For each element of the array pointed to by @buf (1 page at a time), set the
2255 * corresponding bit in @bm.
2256 */
2257static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2258{
2259 int j;
2260
2261 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2262 if (unlikely(buf[j] == BM_END_OF_MAP))
2263 break;
2264
2265 if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2266 memory_bm_set_bit(bm, buf[j]);
2267 else
2268 return -EFAULT;
2269 }
2270
2271 return 0;
2272}
2273
2274#ifdef CONFIG_HIGHMEM
2275/*
2276 * struct highmem_pbe is used for creating the list of highmem pages that
2277 * should be restored atomically during the resume from disk, because the page
2278 * frames they have occupied before the suspend are in use.
2279 */
2280struct highmem_pbe {
2281 struct page *copy_page; /* data is here now */
2282 struct page *orig_page; /* data was here before the suspend */
2283 struct highmem_pbe *next;
2284};
2285
2286/*
2287 * List of highmem PBEs needed for restoring the highmem pages that were
2288 * allocated before the suspend and included in the suspend image, but have
2289 * also been allocated by the "resume" kernel, so their contents cannot be
2290 * written directly to their "original" page frames.
2291 */
2292static struct highmem_pbe *highmem_pblist;
2293
2294/**
2295 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2296 * @bm: Memory bitmap.
2297 *
2298 * The bits in @bm that correspond to image pages are assumed to be set.
2299 */
2300static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2301{
2302 unsigned long pfn;
2303 unsigned int cnt = 0;
2304
2305 memory_bm_position_reset(bm);
2306 pfn = memory_bm_next_pfn(bm);
2307 while (pfn != BM_END_OF_MAP) {
2308 if (PageHighMem(pfn_to_page(pfn)))
2309 cnt++;
2310
2311 pfn = memory_bm_next_pfn(bm);
2312 }
2313 return cnt;
2314}
2315
2316static unsigned int safe_highmem_pages;
2317
2318static struct memory_bitmap *safe_highmem_bm;
2319
2320/**
2321 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2322 * @bm: Pointer to an uninitialized memory bitmap structure.
2323 * @nr_highmem_p: Pointer to the number of highmem image pages.
2324 *
2325 * Try to allocate as many highmem pages as there are highmem image pages
2326 * (@nr_highmem_p points to the variable containing the number of highmem image
2327 * pages). The pages that are "safe" (ie. will not be overwritten when the
2328 * hibernation image is restored entirely) have the corresponding bits set in
2329 * @bm (it must be uninitialized).
2330 *
2331 * NOTE: This function should not be called if there are no highmem image pages.
2332 */
2333static int prepare_highmem_image(struct memory_bitmap *bm,
2334 unsigned int *nr_highmem_p)
2335{
2336 unsigned int to_alloc;
2337
2338 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2339 return -ENOMEM;
2340
2341 if (get_highmem_buffer(PG_SAFE))
2342 return -ENOMEM;
2343
2344 to_alloc = count_free_highmem_pages();
2345 if (to_alloc > *nr_highmem_p)
2346 to_alloc = *nr_highmem_p;
2347 else
2348 *nr_highmem_p = to_alloc;
2349
2350 safe_highmem_pages = 0;
2351 while (to_alloc-- > 0) {
2352 struct page *page;
2353
2354 page = alloc_page(__GFP_HIGHMEM);
2355 if (!swsusp_page_is_free(page)) {
2356 /* The page is "safe", set its bit the bitmap */
2357 memory_bm_set_bit(bm, page_to_pfn(page));
2358 safe_highmem_pages++;
2359 }
2360 /* Mark the page as allocated */
2361 swsusp_set_page_forbidden(page);
2362 swsusp_set_page_free(page);
2363 }
2364 memory_bm_position_reset(bm);
2365 safe_highmem_bm = bm;
2366 return 0;
2367}
2368
2369static struct page *last_highmem_page;
2370
2371/**
2372 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2373 *
2374 * For a given highmem image page get a buffer that suspend_write_next() should
2375 * return to its caller to write to.
2376 *
2377 * If the page is to be saved to its "original" page frame or a copy of
2378 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2379 * the copy of the page is to be made in normal memory, so the address of
2380 * the copy is returned.
2381 *
2382 * If @buffer is returned, the caller of suspend_write_next() will write
2383 * the page's contents to @buffer, so they will have to be copied to the
2384 * right location on the next call to suspend_write_next() and it is done
2385 * with the help of copy_last_highmem_page(). For this purpose, if
2386 * @buffer is returned, @last_highmem_page is set to the page to which
2387 * the data will have to be copied from @buffer.
2388 */
2389static void *get_highmem_page_buffer(struct page *page,
2390 struct chain_allocator *ca)
2391{
2392 struct highmem_pbe *pbe;
2393 void *kaddr;
2394
2395 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2396 /*
2397 * We have allocated the "original" page frame and we can
2398 * use it directly to store the loaded page.
2399 */
2400 last_highmem_page = page;
2401 return buffer;
2402 }
2403 /*
2404 * The "original" page frame has not been allocated and we have to
2405 * use a "safe" page frame to store the loaded page.
2406 */
2407 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2408 if (!pbe) {
2409 swsusp_free();
2410 return ERR_PTR(-ENOMEM);
2411 }
2412 pbe->orig_page = page;
2413 if (safe_highmem_pages > 0) {
2414 struct page *tmp;
2415
2416 /* Copy of the page will be stored in high memory */
2417 kaddr = buffer;
2418 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2419 safe_highmem_pages--;
2420 last_highmem_page = tmp;
2421 pbe->copy_page = tmp;
2422 } else {
2423 /* Copy of the page will be stored in normal memory */
2424 kaddr = safe_pages_list;
2425 safe_pages_list = safe_pages_list->next;
2426 pbe->copy_page = virt_to_page(kaddr);
2427 }
2428 pbe->next = highmem_pblist;
2429 highmem_pblist = pbe;
2430 return kaddr;
2431}
2432
2433/**
2434 * copy_last_highmem_page - Copy most the most recent highmem image page.
2435 *
2436 * Copy the contents of a highmem image from @buffer, where the caller of
2437 * snapshot_write_next() has stored them, to the right location represented by
2438 * @last_highmem_page .
2439 */
2440static void copy_last_highmem_page(void)
2441{
2442 if (last_highmem_page) {
2443 void *dst;
2444
2445 dst = kmap_atomic(last_highmem_page);
2446 copy_page(dst, buffer);
2447 kunmap_atomic(dst);
2448 last_highmem_page = NULL;
2449 }
2450}
2451
2452static inline int last_highmem_page_copied(void)
2453{
2454 return !last_highmem_page;
2455}
2456
2457static inline void free_highmem_data(void)
2458{
2459 if (safe_highmem_bm)
2460 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2461
2462 if (buffer)
2463 free_image_page(buffer, PG_UNSAFE_CLEAR);
2464}
2465#else
2466static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2467
2468static inline int prepare_highmem_image(struct memory_bitmap *bm,
2469 unsigned int *nr_highmem_p) { return 0; }
2470
2471static inline void *get_highmem_page_buffer(struct page *page,
2472 struct chain_allocator *ca)
2473{
2474 return ERR_PTR(-EINVAL);
2475}
2476
2477static inline void copy_last_highmem_page(void) {}
2478static inline int last_highmem_page_copied(void) { return 1; }
2479static inline void free_highmem_data(void) {}
2480#endif /* CONFIG_HIGHMEM */
2481
2482#define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2483
2484/**
2485 * prepare_image - Make room for loading hibernation image.
2486 * @new_bm: Uninitialized memory bitmap structure.
2487 * @bm: Memory bitmap with unsafe pages marked.
2488 *
2489 * Use @bm to mark the pages that will be overwritten in the process of
2490 * restoring the system memory state from the suspend image ("unsafe" pages)
2491 * and allocate memory for the image.
2492 *
2493 * The idea is to allocate a new memory bitmap first and then allocate
2494 * as many pages as needed for image data, but without specifying what those
2495 * pages will be used for just yet. Instead, we mark them all as allocated and
2496 * create a lists of "safe" pages to be used later. On systems with high
2497 * memory a list of "safe" highmem pages is created too.
2498 */
2499static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2500{
2501 unsigned int nr_pages, nr_highmem;
2502 struct linked_page *lp;
2503 int error;
2504
2505 /* If there is no highmem, the buffer will not be necessary */
2506 free_image_page(buffer, PG_UNSAFE_CLEAR);
2507 buffer = NULL;
2508
2509 nr_highmem = count_highmem_image_pages(bm);
2510 mark_unsafe_pages(bm);
2511
2512 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2513 if (error)
2514 goto Free;
2515
2516 duplicate_memory_bitmap(new_bm, bm);
2517 memory_bm_free(bm, PG_UNSAFE_KEEP);
2518 if (nr_highmem > 0) {
2519 error = prepare_highmem_image(bm, &nr_highmem);
2520 if (error)
2521 goto Free;
2522 }
2523 /*
2524 * Reserve some safe pages for potential later use.
2525 *
2526 * NOTE: This way we make sure there will be enough safe pages for the
2527 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2528 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2529 *
2530 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2531 */
2532 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2533 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2534 while (nr_pages > 0) {
2535 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2536 if (!lp) {
2537 error = -ENOMEM;
2538 goto Free;
2539 }
2540 lp->next = safe_pages_list;
2541 safe_pages_list = lp;
2542 nr_pages--;
2543 }
2544 /* Preallocate memory for the image */
2545 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2546 while (nr_pages > 0) {
2547 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2548 if (!lp) {
2549 error = -ENOMEM;
2550 goto Free;
2551 }
2552 if (!swsusp_page_is_free(virt_to_page(lp))) {
2553 /* The page is "safe", add it to the list */
2554 lp->next = safe_pages_list;
2555 safe_pages_list = lp;
2556 }
2557 /* Mark the page as allocated */
2558 swsusp_set_page_forbidden(virt_to_page(lp));
2559 swsusp_set_page_free(virt_to_page(lp));
2560 nr_pages--;
2561 }
2562 return 0;
2563
2564 Free:
2565 swsusp_free();
2566 return error;
2567}
2568
2569/**
2570 * get_buffer - Get the address to store the next image data page.
2571 *
2572 * Get the address that snapshot_write_next() should return to its caller to
2573 * write to.
2574 */
2575static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2576{
2577 struct pbe *pbe;
2578 struct page *page;
2579 unsigned long pfn = memory_bm_next_pfn(bm);
2580
2581 if (pfn == BM_END_OF_MAP)
2582 return ERR_PTR(-EFAULT);
2583
2584 page = pfn_to_page(pfn);
2585 if (PageHighMem(page))
2586 return get_highmem_page_buffer(page, ca);
2587
2588 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2589 /*
2590 * We have allocated the "original" page frame and we can
2591 * use it directly to store the loaded page.
2592 */
2593 return page_address(page);
2594
2595 /*
2596 * The "original" page frame has not been allocated and we have to
2597 * use a "safe" page frame to store the loaded page.
2598 */
2599 pbe = chain_alloc(ca, sizeof(struct pbe));
2600 if (!pbe) {
2601 swsusp_free();
2602 return ERR_PTR(-ENOMEM);
2603 }
2604 pbe->orig_address = page_address(page);
2605 pbe->address = safe_pages_list;
2606 safe_pages_list = safe_pages_list->next;
2607 pbe->next = restore_pblist;
2608 restore_pblist = pbe;
2609 return pbe->address;
2610}
2611
2612/**
2613 * snapshot_write_next - Get the address to store the next image page.
2614 * @handle: Snapshot handle structure to guide the writing.
2615 *
2616 * On the first call, @handle should point to a zeroed snapshot_handle
2617 * structure. The structure gets populated then and a pointer to it should be
2618 * passed to this function every next time.
2619 *
2620 * On success, the function returns a positive number. Then, the caller
2621 * is allowed to write up to the returned number of bytes to the memory
2622 * location computed by the data_of() macro.
2623 *
2624 * The function returns 0 to indicate the "end of file" condition. Negative
2625 * numbers are returned on errors, in which cases the structure pointed to by
2626 * @handle is not updated and should not be used any more.
2627 */
2628int snapshot_write_next(struct snapshot_handle *handle)
2629{
2630 static struct chain_allocator ca;
2631 int error = 0;
2632
2633 /* Check if we have already loaded the entire image */
2634 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2635 return 0;
2636
2637 handle->sync_read = 1;
2638
2639 if (!handle->cur) {
2640 if (!buffer)
2641 /* This makes the buffer be freed by swsusp_free() */
2642 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2643
2644 if (!buffer)
2645 return -ENOMEM;
2646
2647 handle->buffer = buffer;
2648 } else if (handle->cur == 1) {
2649 error = load_header(buffer);
2650 if (error)
2651 return error;
2652
2653 safe_pages_list = NULL;
2654
2655 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2656 if (error)
2657 return error;
2658
2659 hibernate_restore_protection_begin();
2660 } else if (handle->cur <= nr_meta_pages + 1) {
2661 error = unpack_orig_pfns(buffer, ©_bm);
2662 if (error)
2663 return error;
2664
2665 if (handle->cur == nr_meta_pages + 1) {
2666 error = prepare_image(&orig_bm, ©_bm);
2667 if (error)
2668 return error;
2669
2670 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2671 memory_bm_position_reset(&orig_bm);
2672 restore_pblist = NULL;
2673 handle->buffer = get_buffer(&orig_bm, &ca);
2674 handle->sync_read = 0;
2675 if (IS_ERR(handle->buffer))
2676 return PTR_ERR(handle->buffer);
2677 }
2678 } else {
2679 copy_last_highmem_page();
2680 hibernate_restore_protect_page(handle->buffer);
2681 handle->buffer = get_buffer(&orig_bm, &ca);
2682 if (IS_ERR(handle->buffer))
2683 return PTR_ERR(handle->buffer);
2684 if (handle->buffer != buffer)
2685 handle->sync_read = 0;
2686 }
2687 handle->cur++;
2688 return PAGE_SIZE;
2689}
2690
2691/**
2692 * snapshot_write_finalize - Complete the loading of a hibernation image.
2693 *
2694 * Must be called after the last call to snapshot_write_next() in case the last
2695 * page in the image happens to be a highmem page and its contents should be
2696 * stored in highmem. Additionally, it recycles bitmap memory that's not
2697 * necessary any more.
2698 */
2699void snapshot_write_finalize(struct snapshot_handle *handle)
2700{
2701 copy_last_highmem_page();
2702 hibernate_restore_protect_page(handle->buffer);
2703 /* Do that only if we have loaded the image entirely */
2704 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2705 memory_bm_recycle(&orig_bm);
2706 free_highmem_data();
2707 }
2708}
2709
2710int snapshot_image_loaded(struct snapshot_handle *handle)
2711{
2712 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2713 handle->cur <= nr_meta_pages + nr_copy_pages);
2714}
2715
2716#ifdef CONFIG_HIGHMEM
2717/* Assumes that @buf is ready and points to a "safe" page */
2718static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2719 void *buf)
2720{
2721 void *kaddr1, *kaddr2;
2722
2723 kaddr1 = kmap_atomic(p1);
2724 kaddr2 = kmap_atomic(p2);
2725 copy_page(buf, kaddr1);
2726 copy_page(kaddr1, kaddr2);
2727 copy_page(kaddr2, buf);
2728 kunmap_atomic(kaddr2);
2729 kunmap_atomic(kaddr1);
2730}
2731
2732/**
2733 * restore_highmem - Put highmem image pages into their original locations.
2734 *
2735 * For each highmem page that was in use before hibernation and is included in
2736 * the image, and also has been allocated by the "restore" kernel, swap its
2737 * current contents with the previous (ie. "before hibernation") ones.
2738 *
2739 * If the restore eventually fails, we can call this function once again and
2740 * restore the highmem state as seen by the restore kernel.
2741 */
2742int restore_highmem(void)
2743{
2744 struct highmem_pbe *pbe = highmem_pblist;
2745 void *buf;
2746
2747 if (!pbe)
2748 return 0;
2749
2750 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2751 if (!buf)
2752 return -ENOMEM;
2753
2754 while (pbe) {
2755 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2756 pbe = pbe->next;
2757 }
2758 free_image_page(buf, PG_UNSAFE_CLEAR);
2759 return 0;
2760}
2761#endif /* CONFIG_HIGHMEM */