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/* struct 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 unsigned long cur_pfn;
408 int node_bit;
409};
410
411struct memory_bitmap {
412 struct list_head zones;
413 struct linked_page *p_list; /* list of pages used to store zone
414 bitmap objects and bitmap block
415 objects */
416 struct bm_position cur; /* most recently used bit position */
417};
418
419/* Functions that operate on memory bitmaps */
420
421#define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
422#if BITS_PER_LONG == 32
423#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
424#else
425#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
426#endif
427#define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
428
429/**
430 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
431 * @gfp_mask: GFP mask for the allocation.
432 * @safe_needed: Get pages not used before hibernation (restore only)
433 * @ca: Pointer to a linked list of pages ("a chain") to allocate from
434 * @list: Radix Tree node to add.
435 *
436 * This function is used to allocate inner nodes as well as the
437 * leave nodes of the radix tree. It also adds the node to the
438 * corresponding linked list passed in by the *list parameter.
439 */
440static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
441 struct chain_allocator *ca,
442 struct list_head *list)
443{
444 struct rtree_node *node;
445
446 node = chain_alloc(ca, sizeof(struct rtree_node));
447 if (!node)
448 return NULL;
449
450 node->data = get_image_page(gfp_mask, safe_needed);
451 if (!node->data)
452 return NULL;
453
454 list_add_tail(&node->list, list);
455
456 return node;
457}
458
459/**
460 * add_rtree_block - Add a new leave node to the radix tree.
461 *
462 * The leave nodes need to be allocated in order to keep the leaves
463 * linked list in order. This is guaranteed by the zone->blocks
464 * counter.
465 */
466static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
467 int safe_needed, struct chain_allocator *ca)
468{
469 struct rtree_node *node, *block, **dst;
470 unsigned int levels_needed, block_nr;
471 int i;
472
473 block_nr = zone->blocks;
474 levels_needed = 0;
475
476 /* How many levels do we need for this block nr? */
477 while (block_nr) {
478 levels_needed += 1;
479 block_nr >>= BM_RTREE_LEVEL_SHIFT;
480 }
481
482 /* Make sure the rtree has enough levels */
483 for (i = zone->levels; i < levels_needed; i++) {
484 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
485 &zone->nodes);
486 if (!node)
487 return -ENOMEM;
488
489 node->data[0] = (unsigned long)zone->rtree;
490 zone->rtree = node;
491 zone->levels += 1;
492 }
493
494 /* Allocate new block */
495 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
496 if (!block)
497 return -ENOMEM;
498
499 /* Now walk the rtree to insert the block */
500 node = zone->rtree;
501 dst = &zone->rtree;
502 block_nr = zone->blocks;
503 for (i = zone->levels; i > 0; i--) {
504 int index;
505
506 if (!node) {
507 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
508 &zone->nodes);
509 if (!node)
510 return -ENOMEM;
511 *dst = node;
512 }
513
514 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
515 index &= BM_RTREE_LEVEL_MASK;
516 dst = (struct rtree_node **)&((*dst)->data[index]);
517 node = *dst;
518 }
519
520 zone->blocks += 1;
521 *dst = block;
522
523 return 0;
524}
525
526static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
527 int clear_nosave_free);
528
529/**
530 * create_zone_bm_rtree - Create a radix tree for one zone.
531 *
532 * Allocated the mem_zone_bm_rtree structure and initializes it.
533 * This function also allocated and builds the radix tree for the
534 * zone.
535 */
536static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
537 int safe_needed,
538 struct chain_allocator *ca,
539 unsigned long start,
540 unsigned long end)
541{
542 struct mem_zone_bm_rtree *zone;
543 unsigned int i, nr_blocks;
544 unsigned long pages;
545
546 pages = end - start;
547 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
548 if (!zone)
549 return NULL;
550
551 INIT_LIST_HEAD(&zone->nodes);
552 INIT_LIST_HEAD(&zone->leaves);
553 zone->start_pfn = start;
554 zone->end_pfn = end;
555 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
556
557 for (i = 0; i < nr_blocks; i++) {
558 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
559 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
560 return NULL;
561 }
562 }
563
564 return zone;
565}
566
567/**
568 * free_zone_bm_rtree - Free the memory of the radix tree.
569 *
570 * Free all node pages of the radix tree. The mem_zone_bm_rtree
571 * structure itself is not freed here nor are the rtree_node
572 * structs.
573 */
574static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
575 int clear_nosave_free)
576{
577 struct rtree_node *node;
578
579 list_for_each_entry(node, &zone->nodes, list)
580 free_image_page(node->data, clear_nosave_free);
581
582 list_for_each_entry(node, &zone->leaves, list)
583 free_image_page(node->data, clear_nosave_free);
584}
585
586static void memory_bm_position_reset(struct memory_bitmap *bm)
587{
588 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
589 list);
590 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
591 struct rtree_node, list);
592 bm->cur.node_pfn = 0;
593 bm->cur.cur_pfn = BM_END_OF_MAP;
594 bm->cur.node_bit = 0;
595}
596
597static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
598
599struct mem_extent {
600 struct list_head hook;
601 unsigned long start;
602 unsigned long end;
603};
604
605/**
606 * free_mem_extents - Free a list of memory extents.
607 * @list: List of extents to free.
608 */
609static void free_mem_extents(struct list_head *list)
610{
611 struct mem_extent *ext, *aux;
612
613 list_for_each_entry_safe(ext, aux, list, hook) {
614 list_del(&ext->hook);
615 kfree(ext);
616 }
617}
618
619/**
620 * create_mem_extents - Create a list of memory extents.
621 * @list: List to put the extents into.
622 * @gfp_mask: Mask to use for memory allocations.
623 *
624 * The extents represent contiguous ranges of PFNs.
625 */
626static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
627{
628 struct zone *zone;
629
630 INIT_LIST_HEAD(list);
631
632 for_each_populated_zone(zone) {
633 unsigned long zone_start, zone_end;
634 struct mem_extent *ext, *cur, *aux;
635
636 zone_start = zone->zone_start_pfn;
637 zone_end = zone_end_pfn(zone);
638
639 list_for_each_entry(ext, list, hook)
640 if (zone_start <= ext->end)
641 break;
642
643 if (&ext->hook == list || zone_end < ext->start) {
644 /* New extent is necessary */
645 struct mem_extent *new_ext;
646
647 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
648 if (!new_ext) {
649 free_mem_extents(list);
650 return -ENOMEM;
651 }
652 new_ext->start = zone_start;
653 new_ext->end = zone_end;
654 list_add_tail(&new_ext->hook, &ext->hook);
655 continue;
656 }
657
658 /* Merge this zone's range of PFNs with the existing one */
659 if (zone_start < ext->start)
660 ext->start = zone_start;
661 if (zone_end > ext->end)
662 ext->end = zone_end;
663
664 /* More merging may be possible */
665 cur = ext;
666 list_for_each_entry_safe_continue(cur, aux, list, hook) {
667 if (zone_end < cur->start)
668 break;
669 if (zone_end < cur->end)
670 ext->end = cur->end;
671 list_del(&cur->hook);
672 kfree(cur);
673 }
674 }
675
676 return 0;
677}
678
679/**
680 * memory_bm_create - Allocate memory for a memory bitmap.
681 */
682static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
683 int safe_needed)
684{
685 struct chain_allocator ca;
686 struct list_head mem_extents;
687 struct mem_extent *ext;
688 int error;
689
690 chain_init(&ca, gfp_mask, safe_needed);
691 INIT_LIST_HEAD(&bm->zones);
692
693 error = create_mem_extents(&mem_extents, gfp_mask);
694 if (error)
695 return error;
696
697 list_for_each_entry(ext, &mem_extents, hook) {
698 struct mem_zone_bm_rtree *zone;
699
700 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
701 ext->start, ext->end);
702 if (!zone) {
703 error = -ENOMEM;
704 goto Error;
705 }
706 list_add_tail(&zone->list, &bm->zones);
707 }
708
709 bm->p_list = ca.chain;
710 memory_bm_position_reset(bm);
711 Exit:
712 free_mem_extents(&mem_extents);
713 return error;
714
715 Error:
716 bm->p_list = ca.chain;
717 memory_bm_free(bm, PG_UNSAFE_CLEAR);
718 goto Exit;
719}
720
721/**
722 * memory_bm_free - Free memory occupied by the memory bitmap.
723 * @bm: Memory bitmap.
724 */
725static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
726{
727 struct mem_zone_bm_rtree *zone;
728
729 list_for_each_entry(zone, &bm->zones, list)
730 free_zone_bm_rtree(zone, clear_nosave_free);
731
732 free_list_of_pages(bm->p_list, clear_nosave_free);
733
734 INIT_LIST_HEAD(&bm->zones);
735}
736
737/**
738 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
739 *
740 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
741 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
742 *
743 * Walk the radix tree to find the page containing the bit that represents @pfn
744 * and return the position of the bit in @addr and @bit_nr.
745 */
746static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
747 void **addr, unsigned int *bit_nr)
748{
749 struct mem_zone_bm_rtree *curr, *zone;
750 struct rtree_node *node;
751 int i, block_nr;
752
753 zone = bm->cur.zone;
754
755 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
756 goto zone_found;
757
758 zone = NULL;
759
760 /* Find the right zone */
761 list_for_each_entry(curr, &bm->zones, list) {
762 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
763 zone = curr;
764 break;
765 }
766 }
767
768 if (!zone)
769 return -EFAULT;
770
771zone_found:
772 /*
773 * We have found the zone. Now walk the radix tree to find the leaf node
774 * for our PFN.
775 */
776
777 /*
778 * If the zone we wish to scan is the current zone and the
779 * pfn falls into the current node then we do not need to walk
780 * the tree.
781 */
782 node = bm->cur.node;
783 if (zone == bm->cur.zone &&
784 ((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
785 goto node_found;
786
787 node = zone->rtree;
788 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
789
790 for (i = zone->levels; i > 0; i--) {
791 int index;
792
793 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
794 index &= BM_RTREE_LEVEL_MASK;
795 BUG_ON(node->data[index] == 0);
796 node = (struct rtree_node *)node->data[index];
797 }
798
799node_found:
800 /* Update last position */
801 bm->cur.zone = zone;
802 bm->cur.node = node;
803 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
804 bm->cur.cur_pfn = pfn;
805
806 /* Set return values */
807 *addr = node->data;
808 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
809
810 return 0;
811}
812
813static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
814{
815 void *addr;
816 unsigned int bit;
817 int error;
818
819 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
820 BUG_ON(error);
821 set_bit(bit, addr);
822}
823
824static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
825{
826 void *addr;
827 unsigned int bit;
828 int error;
829
830 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
831 if (!error)
832 set_bit(bit, addr);
833
834 return error;
835}
836
837static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
838{
839 void *addr;
840 unsigned int bit;
841 int error;
842
843 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
844 BUG_ON(error);
845 clear_bit(bit, addr);
846}
847
848static void memory_bm_clear_current(struct memory_bitmap *bm)
849{
850 int bit;
851
852 bit = max(bm->cur.node_bit - 1, 0);
853 clear_bit(bit, bm->cur.node->data);
854}
855
856static unsigned long memory_bm_get_current(struct memory_bitmap *bm)
857{
858 return bm->cur.cur_pfn;
859}
860
861static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
862{
863 void *addr;
864 unsigned int bit;
865 int error;
866
867 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
868 BUG_ON(error);
869 return test_bit(bit, addr);
870}
871
872static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
873{
874 void *addr;
875 unsigned int bit;
876
877 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
878}
879
880/*
881 * rtree_next_node - Jump to the next leaf node.
882 *
883 * Set the position to the beginning of the next node in the
884 * memory bitmap. This is either the next node in the current
885 * zone's radix tree or the first node in the radix tree of the
886 * next zone.
887 *
888 * Return true if there is a next node, false otherwise.
889 */
890static bool rtree_next_node(struct memory_bitmap *bm)
891{
892 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
893 bm->cur.node = list_entry(bm->cur.node->list.next,
894 struct rtree_node, list);
895 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
896 bm->cur.node_bit = 0;
897 touch_softlockup_watchdog();
898 return true;
899 }
900
901 /* No more nodes, goto next zone */
902 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
903 bm->cur.zone = list_entry(bm->cur.zone->list.next,
904 struct mem_zone_bm_rtree, list);
905 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
906 struct rtree_node, list);
907 bm->cur.node_pfn = 0;
908 bm->cur.node_bit = 0;
909 return true;
910 }
911
912 /* No more zones */
913 return false;
914}
915
916/**
917 * memory_bm_next_pfn - Find the next set bit in a memory bitmap.
918 * @bm: Memory bitmap.
919 *
920 * Starting from the last returned position this function searches for the next
921 * set bit in @bm and returns the PFN represented by it. If no more bits are
922 * set, BM_END_OF_MAP is returned.
923 *
924 * It is required to run memory_bm_position_reset() before the first call to
925 * this function for the given memory bitmap.
926 */
927static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
928{
929 unsigned long bits, pfn, pages;
930 int bit;
931
932 do {
933 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
934 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
935 bit = find_next_bit(bm->cur.node->data, bits,
936 bm->cur.node_bit);
937 if (bit < bits) {
938 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
939 bm->cur.node_bit = bit + 1;
940 bm->cur.cur_pfn = pfn;
941 return pfn;
942 }
943 } while (rtree_next_node(bm));
944
945 bm->cur.cur_pfn = BM_END_OF_MAP;
946 return BM_END_OF_MAP;
947}
948
949/*
950 * This structure represents a range of page frames the contents of which
951 * should not be saved during hibernation.
952 */
953struct nosave_region {
954 struct list_head list;
955 unsigned long start_pfn;
956 unsigned long end_pfn;
957};
958
959static LIST_HEAD(nosave_regions);
960
961static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
962{
963 struct rtree_node *node;
964
965 list_for_each_entry(node, &zone->nodes, list)
966 recycle_safe_page(node->data);
967
968 list_for_each_entry(node, &zone->leaves, list)
969 recycle_safe_page(node->data);
970}
971
972static void memory_bm_recycle(struct memory_bitmap *bm)
973{
974 struct mem_zone_bm_rtree *zone;
975 struct linked_page *p_list;
976
977 list_for_each_entry(zone, &bm->zones, list)
978 recycle_zone_bm_rtree(zone);
979
980 p_list = bm->p_list;
981 while (p_list) {
982 struct linked_page *lp = p_list;
983
984 p_list = lp->next;
985 recycle_safe_page(lp);
986 }
987}
988
989/**
990 * register_nosave_region - Register a region of unsaveable memory.
991 *
992 * Register a range of page frames the contents of which should not be saved
993 * during hibernation (to be used in the early initialization code).
994 */
995void __init register_nosave_region(unsigned long start_pfn, unsigned long end_pfn)
996{
997 struct nosave_region *region;
998
999 if (start_pfn >= end_pfn)
1000 return;
1001
1002 if (!list_empty(&nosave_regions)) {
1003 /* Try to extend the previous region (they should be sorted) */
1004 region = list_entry(nosave_regions.prev,
1005 struct nosave_region, list);
1006 if (region->end_pfn == start_pfn) {
1007 region->end_pfn = end_pfn;
1008 goto Report;
1009 }
1010 }
1011 /* This allocation cannot fail */
1012 region = memblock_alloc(sizeof(struct nosave_region),
1013 SMP_CACHE_BYTES);
1014 if (!region)
1015 panic("%s: Failed to allocate %zu bytes\n", __func__,
1016 sizeof(struct nosave_region));
1017 region->start_pfn = start_pfn;
1018 region->end_pfn = end_pfn;
1019 list_add_tail(®ion->list, &nosave_regions);
1020 Report:
1021 pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
1022 (unsigned long long) start_pfn << PAGE_SHIFT,
1023 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1024}
1025
1026/*
1027 * Set bits in this map correspond to the page frames the contents of which
1028 * should not be saved during the suspend.
1029 */
1030static struct memory_bitmap *forbidden_pages_map;
1031
1032/* Set bits in this map correspond to free page frames. */
1033static struct memory_bitmap *free_pages_map;
1034
1035/*
1036 * Each page frame allocated for creating the image is marked by setting the
1037 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
1038 */
1039
1040void swsusp_set_page_free(struct page *page)
1041{
1042 if (free_pages_map)
1043 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
1044}
1045
1046static int swsusp_page_is_free(struct page *page)
1047{
1048 return free_pages_map ?
1049 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1050}
1051
1052void swsusp_unset_page_free(struct page *page)
1053{
1054 if (free_pages_map)
1055 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1056}
1057
1058static void swsusp_set_page_forbidden(struct page *page)
1059{
1060 if (forbidden_pages_map)
1061 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1062}
1063
1064int swsusp_page_is_forbidden(struct page *page)
1065{
1066 return forbidden_pages_map ?
1067 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1068}
1069
1070static void swsusp_unset_page_forbidden(struct page *page)
1071{
1072 if (forbidden_pages_map)
1073 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1074}
1075
1076/**
1077 * mark_nosave_pages - Mark pages that should not be saved.
1078 * @bm: Memory bitmap.
1079 *
1080 * Set the bits in @bm that correspond to the page frames the contents of which
1081 * should not be saved.
1082 */
1083static void mark_nosave_pages(struct memory_bitmap *bm)
1084{
1085 struct nosave_region *region;
1086
1087 if (list_empty(&nosave_regions))
1088 return;
1089
1090 list_for_each_entry(region, &nosave_regions, list) {
1091 unsigned long pfn;
1092
1093 pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1094 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1095 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1096 - 1);
1097
1098 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1099 if (pfn_valid(pfn)) {
1100 /*
1101 * It is safe to ignore the result of
1102 * mem_bm_set_bit_check() here, since we won't
1103 * touch the PFNs for which the error is
1104 * returned anyway.
1105 */
1106 mem_bm_set_bit_check(bm, pfn);
1107 }
1108 }
1109}
1110
1111/**
1112 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1113 *
1114 * Create bitmaps needed for marking page frames that should not be saved and
1115 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1116 * only modified if everything goes well, because we don't want the bits to be
1117 * touched before both bitmaps are set up.
1118 */
1119int create_basic_memory_bitmaps(void)
1120{
1121 struct memory_bitmap *bm1, *bm2;
1122 int error;
1123
1124 if (forbidden_pages_map && free_pages_map)
1125 return 0;
1126 else
1127 BUG_ON(forbidden_pages_map || free_pages_map);
1128
1129 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1130 if (!bm1)
1131 return -ENOMEM;
1132
1133 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1134 if (error)
1135 goto Free_first_object;
1136
1137 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1138 if (!bm2)
1139 goto Free_first_bitmap;
1140
1141 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1142 if (error)
1143 goto Free_second_object;
1144
1145 forbidden_pages_map = bm1;
1146 free_pages_map = bm2;
1147 mark_nosave_pages(forbidden_pages_map);
1148
1149 pr_debug("Basic memory bitmaps created\n");
1150
1151 return 0;
1152
1153 Free_second_object:
1154 kfree(bm2);
1155 Free_first_bitmap:
1156 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1157 Free_first_object:
1158 kfree(bm1);
1159 return -ENOMEM;
1160}
1161
1162/**
1163 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1164 *
1165 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1166 * auxiliary pointers are necessary so that the bitmaps themselves are not
1167 * referred to while they are being freed.
1168 */
1169void free_basic_memory_bitmaps(void)
1170{
1171 struct memory_bitmap *bm1, *bm2;
1172
1173 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1174 return;
1175
1176 bm1 = forbidden_pages_map;
1177 bm2 = free_pages_map;
1178 forbidden_pages_map = NULL;
1179 free_pages_map = NULL;
1180 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1181 kfree(bm1);
1182 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1183 kfree(bm2);
1184
1185 pr_debug("Basic memory bitmaps freed\n");
1186}
1187
1188static void clear_or_poison_free_page(struct page *page)
1189{
1190 if (page_poisoning_enabled_static())
1191 __kernel_poison_pages(page, 1);
1192 else if (want_init_on_free())
1193 clear_highpage(page);
1194}
1195
1196void clear_or_poison_free_pages(void)
1197{
1198 struct memory_bitmap *bm = free_pages_map;
1199 unsigned long pfn;
1200
1201 if (WARN_ON(!(free_pages_map)))
1202 return;
1203
1204 if (page_poisoning_enabled() || want_init_on_free()) {
1205 memory_bm_position_reset(bm);
1206 pfn = memory_bm_next_pfn(bm);
1207 while (pfn != BM_END_OF_MAP) {
1208 if (pfn_valid(pfn))
1209 clear_or_poison_free_page(pfn_to_page(pfn));
1210
1211 pfn = memory_bm_next_pfn(bm);
1212 }
1213 memory_bm_position_reset(bm);
1214 pr_info("free pages cleared after restore\n");
1215 }
1216}
1217
1218/**
1219 * snapshot_additional_pages - Estimate the number of extra pages needed.
1220 * @zone: Memory zone to carry out the computation for.
1221 *
1222 * Estimate the number of additional pages needed for setting up a hibernation
1223 * image data structures for @zone (usually, the returned value is greater than
1224 * the exact number).
1225 */
1226unsigned int snapshot_additional_pages(struct zone *zone)
1227{
1228 unsigned int rtree, nodes;
1229
1230 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1231 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1232 LINKED_PAGE_DATA_SIZE);
1233 while (nodes > 1) {
1234 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1235 rtree += nodes;
1236 }
1237
1238 return 2 * rtree;
1239}
1240
1241/*
1242 * Touch the watchdog for every WD_PAGE_COUNT pages.
1243 */
1244#define WD_PAGE_COUNT (128*1024)
1245
1246static void mark_free_pages(struct zone *zone)
1247{
1248 unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
1249 unsigned long flags;
1250 unsigned int order, t;
1251 struct page *page;
1252
1253 if (zone_is_empty(zone))
1254 return;
1255
1256 spin_lock_irqsave(&zone->lock, flags);
1257
1258 max_zone_pfn = zone_end_pfn(zone);
1259 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1260 if (pfn_valid(pfn)) {
1261 page = pfn_to_page(pfn);
1262
1263 if (!--page_count) {
1264 touch_nmi_watchdog();
1265 page_count = WD_PAGE_COUNT;
1266 }
1267
1268 if (page_zone(page) != zone)
1269 continue;
1270
1271 if (!swsusp_page_is_forbidden(page))
1272 swsusp_unset_page_free(page);
1273 }
1274
1275 for_each_migratetype_order(order, t) {
1276 list_for_each_entry(page,
1277 &zone->free_area[order].free_list[t], buddy_list) {
1278 unsigned long i;
1279
1280 pfn = page_to_pfn(page);
1281 for (i = 0; i < (1UL << order); i++) {
1282 if (!--page_count) {
1283 touch_nmi_watchdog();
1284 page_count = WD_PAGE_COUNT;
1285 }
1286 swsusp_set_page_free(pfn_to_page(pfn + i));
1287 }
1288 }
1289 }
1290 spin_unlock_irqrestore(&zone->lock, flags);
1291}
1292
1293#ifdef CONFIG_HIGHMEM
1294/**
1295 * count_free_highmem_pages - Compute the total number of free highmem pages.
1296 *
1297 * The returned number is system-wide.
1298 */
1299static unsigned int count_free_highmem_pages(void)
1300{
1301 struct zone *zone;
1302 unsigned int cnt = 0;
1303
1304 for_each_populated_zone(zone)
1305 if (is_highmem(zone))
1306 cnt += zone_page_state(zone, NR_FREE_PAGES);
1307
1308 return cnt;
1309}
1310
1311/**
1312 * saveable_highmem_page - Check if a highmem page is saveable.
1313 *
1314 * Determine whether a highmem page should be included in a hibernation image.
1315 *
1316 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1317 * and it isn't part of a free chunk of pages.
1318 */
1319static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1320{
1321 struct page *page;
1322
1323 if (!pfn_valid(pfn))
1324 return NULL;
1325
1326 page = pfn_to_online_page(pfn);
1327 if (!page || page_zone(page) != zone)
1328 return NULL;
1329
1330 BUG_ON(!PageHighMem(page));
1331
1332 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1333 return NULL;
1334
1335 if (PageReserved(page) || PageOffline(page))
1336 return NULL;
1337
1338 if (page_is_guard(page))
1339 return NULL;
1340
1341 return page;
1342}
1343
1344/**
1345 * count_highmem_pages - Compute the total number of saveable highmem pages.
1346 */
1347static unsigned int count_highmem_pages(void)
1348{
1349 struct zone *zone;
1350 unsigned int n = 0;
1351
1352 for_each_populated_zone(zone) {
1353 unsigned long pfn, max_zone_pfn;
1354
1355 if (!is_highmem(zone))
1356 continue;
1357
1358 mark_free_pages(zone);
1359 max_zone_pfn = zone_end_pfn(zone);
1360 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1361 if (saveable_highmem_page(zone, pfn))
1362 n++;
1363 }
1364 return n;
1365}
1366#else
1367static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1368{
1369 return NULL;
1370}
1371#endif /* CONFIG_HIGHMEM */
1372
1373/**
1374 * saveable_page - Check if the given page is saveable.
1375 *
1376 * Determine whether a non-highmem page should be included in a hibernation
1377 * image.
1378 *
1379 * We should save the page if it isn't Nosave, and is not in the range
1380 * of pages statically defined as 'unsaveable', and it isn't part of
1381 * a free chunk of pages.
1382 */
1383static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1384{
1385 struct page *page;
1386
1387 if (!pfn_valid(pfn))
1388 return NULL;
1389
1390 page = pfn_to_online_page(pfn);
1391 if (!page || page_zone(page) != zone)
1392 return NULL;
1393
1394 BUG_ON(PageHighMem(page));
1395
1396 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1397 return NULL;
1398
1399 if (PageOffline(page))
1400 return NULL;
1401
1402 if (PageReserved(page)
1403 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1404 return NULL;
1405
1406 if (page_is_guard(page))
1407 return NULL;
1408
1409 return page;
1410}
1411
1412/**
1413 * count_data_pages - Compute the total number of saveable non-highmem pages.
1414 */
1415static unsigned int count_data_pages(void)
1416{
1417 struct zone *zone;
1418 unsigned long pfn, max_zone_pfn;
1419 unsigned int n = 0;
1420
1421 for_each_populated_zone(zone) {
1422 if (is_highmem(zone))
1423 continue;
1424
1425 mark_free_pages(zone);
1426 max_zone_pfn = zone_end_pfn(zone);
1427 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1428 if (saveable_page(zone, pfn))
1429 n++;
1430 }
1431 return n;
1432}
1433
1434/*
1435 * This is needed, because copy_page and memcpy are not usable for copying
1436 * task structs. Returns true if the page was filled with only zeros,
1437 * otherwise false.
1438 */
1439static inline bool do_copy_page(long *dst, long *src)
1440{
1441 long z = 0;
1442 int n;
1443
1444 for (n = PAGE_SIZE / sizeof(long); n; n--) {
1445 z |= *src;
1446 *dst++ = *src++;
1447 }
1448 return !z;
1449}
1450
1451/**
1452 * safe_copy_page - Copy a page in a safe way.
1453 *
1454 * Check if the page we are going to copy is marked as present in the kernel
1455 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1456 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1457 * always returns 'true'. Returns true if the page was entirely composed of
1458 * zeros, otherwise it will return false.
1459 */
1460static bool safe_copy_page(void *dst, struct page *s_page)
1461{
1462 bool zeros_only;
1463
1464 if (kernel_page_present(s_page)) {
1465 zeros_only = do_copy_page(dst, page_address(s_page));
1466 } else {
1467 hibernate_map_page(s_page);
1468 zeros_only = do_copy_page(dst, page_address(s_page));
1469 hibernate_unmap_page(s_page);
1470 }
1471 return zeros_only;
1472}
1473
1474#ifdef CONFIG_HIGHMEM
1475static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1476{
1477 return is_highmem(zone) ?
1478 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1479}
1480
1481static bool copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1482{
1483 struct page *s_page, *d_page;
1484 void *src, *dst;
1485 bool zeros_only;
1486
1487 s_page = pfn_to_page(src_pfn);
1488 d_page = pfn_to_page(dst_pfn);
1489 if (PageHighMem(s_page)) {
1490 src = kmap_local_page(s_page);
1491 dst = kmap_local_page(d_page);
1492 zeros_only = do_copy_page(dst, src);
1493 kunmap_local(dst);
1494 kunmap_local(src);
1495 } else {
1496 if (PageHighMem(d_page)) {
1497 /*
1498 * The page pointed to by src may contain some kernel
1499 * data modified by kmap_atomic()
1500 */
1501 zeros_only = safe_copy_page(buffer, s_page);
1502 dst = kmap_local_page(d_page);
1503 copy_page(dst, buffer);
1504 kunmap_local(dst);
1505 } else {
1506 zeros_only = safe_copy_page(page_address(d_page), s_page);
1507 }
1508 }
1509 return zeros_only;
1510}
1511#else
1512#define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1513
1514static inline int copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1515{
1516 return safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1517 pfn_to_page(src_pfn));
1518}
1519#endif /* CONFIG_HIGHMEM */
1520
1521/*
1522 * Copy data pages will copy all pages into pages pulled from the copy_bm.
1523 * If a page was entirely filled with zeros it will be marked in the zero_bm.
1524 *
1525 * Returns the number of pages copied.
1526 */
1527static unsigned long copy_data_pages(struct memory_bitmap *copy_bm,
1528 struct memory_bitmap *orig_bm,
1529 struct memory_bitmap *zero_bm)
1530{
1531 unsigned long copied_pages = 0;
1532 struct zone *zone;
1533 unsigned long pfn, copy_pfn;
1534
1535 for_each_populated_zone(zone) {
1536 unsigned long max_zone_pfn;
1537
1538 mark_free_pages(zone);
1539 max_zone_pfn = zone_end_pfn(zone);
1540 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1541 if (page_is_saveable(zone, pfn))
1542 memory_bm_set_bit(orig_bm, pfn);
1543 }
1544 memory_bm_position_reset(orig_bm);
1545 memory_bm_position_reset(copy_bm);
1546 copy_pfn = memory_bm_next_pfn(copy_bm);
1547 for(;;) {
1548 pfn = memory_bm_next_pfn(orig_bm);
1549 if (unlikely(pfn == BM_END_OF_MAP))
1550 break;
1551 if (copy_data_page(copy_pfn, pfn)) {
1552 memory_bm_set_bit(zero_bm, pfn);
1553 /* Use this copy_pfn for a page that is not full of zeros */
1554 continue;
1555 }
1556 copied_pages++;
1557 copy_pfn = memory_bm_next_pfn(copy_bm);
1558 }
1559 return copied_pages;
1560}
1561
1562/* Total number of image pages */
1563static unsigned int nr_copy_pages;
1564/* Number of pages needed for saving the original pfns of the image pages */
1565static unsigned int nr_meta_pages;
1566/* Number of zero pages */
1567static unsigned int nr_zero_pages;
1568
1569/*
1570 * Numbers of normal and highmem page frames allocated for hibernation image
1571 * before suspending devices.
1572 */
1573static unsigned int alloc_normal, alloc_highmem;
1574/*
1575 * Memory bitmap used for marking saveable pages (during hibernation) or
1576 * hibernation image pages (during restore)
1577 */
1578static struct memory_bitmap orig_bm;
1579/*
1580 * Memory bitmap used during hibernation for marking allocated page frames that
1581 * will contain copies of saveable pages. During restore it is initially used
1582 * for marking hibernation image pages, but then the set bits from it are
1583 * duplicated in @orig_bm and it is released. On highmem systems it is next
1584 * used for marking "safe" highmem pages, but it has to be reinitialized for
1585 * this purpose.
1586 */
1587static struct memory_bitmap copy_bm;
1588
1589/* Memory bitmap which tracks which saveable pages were zero filled. */
1590static struct memory_bitmap zero_bm;
1591
1592/**
1593 * swsusp_free - Free pages allocated for hibernation image.
1594 *
1595 * Image pages are allocated before snapshot creation, so they need to be
1596 * released after resume.
1597 */
1598void swsusp_free(void)
1599{
1600 unsigned long fb_pfn, fr_pfn;
1601
1602 if (!forbidden_pages_map || !free_pages_map)
1603 goto out;
1604
1605 memory_bm_position_reset(forbidden_pages_map);
1606 memory_bm_position_reset(free_pages_map);
1607
1608loop:
1609 fr_pfn = memory_bm_next_pfn(free_pages_map);
1610 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1611
1612 /*
1613 * Find the next bit set in both bitmaps. This is guaranteed to
1614 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1615 */
1616 do {
1617 if (fb_pfn < fr_pfn)
1618 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1619 if (fr_pfn < fb_pfn)
1620 fr_pfn = memory_bm_next_pfn(free_pages_map);
1621 } while (fb_pfn != fr_pfn);
1622
1623 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1624 struct page *page = pfn_to_page(fr_pfn);
1625
1626 memory_bm_clear_current(forbidden_pages_map);
1627 memory_bm_clear_current(free_pages_map);
1628 hibernate_restore_unprotect_page(page_address(page));
1629 __free_page(page);
1630 goto loop;
1631 }
1632
1633out:
1634 nr_copy_pages = 0;
1635 nr_meta_pages = 0;
1636 nr_zero_pages = 0;
1637 restore_pblist = NULL;
1638 buffer = NULL;
1639 alloc_normal = 0;
1640 alloc_highmem = 0;
1641 hibernate_restore_protection_end();
1642}
1643
1644/* Helper functions used for the shrinking of memory. */
1645
1646#define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1647
1648/**
1649 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1650 * @nr_pages: Number of page frames to allocate.
1651 * @mask: GFP flags to use for the allocation.
1652 *
1653 * Return value: Number of page frames actually allocated
1654 */
1655static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1656{
1657 unsigned long nr_alloc = 0;
1658
1659 while (nr_pages > 0) {
1660 struct page *page;
1661
1662 page = alloc_image_page(mask);
1663 if (!page)
1664 break;
1665 memory_bm_set_bit(©_bm, page_to_pfn(page));
1666 if (PageHighMem(page))
1667 alloc_highmem++;
1668 else
1669 alloc_normal++;
1670 nr_pages--;
1671 nr_alloc++;
1672 }
1673
1674 return nr_alloc;
1675}
1676
1677static unsigned long preallocate_image_memory(unsigned long nr_pages,
1678 unsigned long avail_normal)
1679{
1680 unsigned long alloc;
1681
1682 if (avail_normal <= alloc_normal)
1683 return 0;
1684
1685 alloc = avail_normal - alloc_normal;
1686 if (nr_pages < alloc)
1687 alloc = nr_pages;
1688
1689 return preallocate_image_pages(alloc, GFP_IMAGE);
1690}
1691
1692#ifdef CONFIG_HIGHMEM
1693static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1694{
1695 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1696}
1697
1698/**
1699 * __fraction - Compute (an approximation of) x * (multiplier / base).
1700 */
1701static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1702{
1703 return div64_u64(x * multiplier, base);
1704}
1705
1706static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1707 unsigned long highmem,
1708 unsigned long total)
1709{
1710 unsigned long alloc = __fraction(nr_pages, highmem, total);
1711
1712 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1713}
1714#else /* CONFIG_HIGHMEM */
1715static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1716{
1717 return 0;
1718}
1719
1720static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1721 unsigned long highmem,
1722 unsigned long total)
1723{
1724 return 0;
1725}
1726#endif /* CONFIG_HIGHMEM */
1727
1728/**
1729 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1730 */
1731static unsigned long free_unnecessary_pages(void)
1732{
1733 unsigned long save, to_free_normal, to_free_highmem, free;
1734
1735 save = count_data_pages();
1736 if (alloc_normal >= save) {
1737 to_free_normal = alloc_normal - save;
1738 save = 0;
1739 } else {
1740 to_free_normal = 0;
1741 save -= alloc_normal;
1742 }
1743 save += count_highmem_pages();
1744 if (alloc_highmem >= save) {
1745 to_free_highmem = alloc_highmem - save;
1746 } else {
1747 to_free_highmem = 0;
1748 save -= alloc_highmem;
1749 if (to_free_normal > save)
1750 to_free_normal -= save;
1751 else
1752 to_free_normal = 0;
1753 }
1754 free = to_free_normal + to_free_highmem;
1755
1756 memory_bm_position_reset(©_bm);
1757
1758 while (to_free_normal > 0 || to_free_highmem > 0) {
1759 unsigned long pfn = memory_bm_next_pfn(©_bm);
1760 struct page *page = pfn_to_page(pfn);
1761
1762 if (PageHighMem(page)) {
1763 if (!to_free_highmem)
1764 continue;
1765 to_free_highmem--;
1766 alloc_highmem--;
1767 } else {
1768 if (!to_free_normal)
1769 continue;
1770 to_free_normal--;
1771 alloc_normal--;
1772 }
1773 memory_bm_clear_bit(©_bm, pfn);
1774 swsusp_unset_page_forbidden(page);
1775 swsusp_unset_page_free(page);
1776 __free_page(page);
1777 }
1778
1779 return free;
1780}
1781
1782/**
1783 * minimum_image_size - Estimate the minimum acceptable size of an image.
1784 * @saveable: Number of saveable pages in the system.
1785 *
1786 * We want to avoid attempting to free too much memory too hard, so estimate the
1787 * minimum acceptable size of a hibernation image to use as the lower limit for
1788 * preallocating memory.
1789 *
1790 * We assume that the minimum image size should be proportional to
1791 *
1792 * [number of saveable pages] - [number of pages that can be freed in theory]
1793 *
1794 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1795 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1796 */
1797static unsigned long minimum_image_size(unsigned long saveable)
1798{
1799 unsigned long size;
1800
1801 size = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B)
1802 + global_node_page_state(NR_ACTIVE_ANON)
1803 + global_node_page_state(NR_INACTIVE_ANON)
1804 + global_node_page_state(NR_ACTIVE_FILE)
1805 + global_node_page_state(NR_INACTIVE_FILE);
1806
1807 return saveable <= size ? 0 : saveable - size;
1808}
1809
1810/**
1811 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1812 *
1813 * To create a hibernation image it is necessary to make a copy of every page
1814 * frame in use. We also need a number of page frames to be free during
1815 * hibernation for allocations made while saving the image and for device
1816 * drivers, in case they need to allocate memory from their hibernation
1817 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1818 * estimate) and reserved_size divided by PAGE_SIZE (which is tunable through
1819 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1820 * total number of available page frames and allocate at least
1821 *
1822 * ([page frames total] - PAGES_FOR_IO - [metadata pages]) / 2
1823 * - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1824 *
1825 * of them, which corresponds to the maximum size of a hibernation image.
1826 *
1827 * If image_size is set below the number following from the above formula,
1828 * the preallocation of memory is continued until the total number of saveable
1829 * pages in the system is below the requested image size or the minimum
1830 * acceptable image size returned by minimum_image_size(), whichever is greater.
1831 */
1832int hibernate_preallocate_memory(void)
1833{
1834 struct zone *zone;
1835 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1836 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1837 ktime_t start, stop;
1838 int error;
1839
1840 pr_info("Preallocating image memory\n");
1841 start = ktime_get();
1842
1843 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1844 if (error) {
1845 pr_err("Cannot allocate original bitmap\n");
1846 goto err_out;
1847 }
1848
1849 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1850 if (error) {
1851 pr_err("Cannot allocate copy bitmap\n");
1852 goto err_out;
1853 }
1854
1855 error = memory_bm_create(&zero_bm, GFP_IMAGE, PG_ANY);
1856 if (error) {
1857 pr_err("Cannot allocate zero bitmap\n");
1858 goto err_out;
1859 }
1860
1861 alloc_normal = 0;
1862 alloc_highmem = 0;
1863 nr_zero_pages = 0;
1864
1865 /* Count the number of saveable data pages. */
1866 save_highmem = count_highmem_pages();
1867 saveable = count_data_pages();
1868
1869 /*
1870 * Compute the total number of page frames we can use (count) and the
1871 * number of pages needed for image metadata (size).
1872 */
1873 count = saveable;
1874 saveable += save_highmem;
1875 highmem = save_highmem;
1876 size = 0;
1877 for_each_populated_zone(zone) {
1878 size += snapshot_additional_pages(zone);
1879 if (is_highmem(zone))
1880 highmem += zone_page_state(zone, NR_FREE_PAGES);
1881 else
1882 count += zone_page_state(zone, NR_FREE_PAGES);
1883 }
1884 avail_normal = count;
1885 count += highmem;
1886 count -= totalreserve_pages;
1887
1888 /* Compute the maximum number of saveable pages to leave in memory. */
1889 max_size = (count - (size + PAGES_FOR_IO)) / 2
1890 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1891 /* Compute the desired number of image pages specified by image_size. */
1892 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1893 if (size > max_size)
1894 size = max_size;
1895 /*
1896 * If the desired number of image pages is at least as large as the
1897 * current number of saveable pages in memory, allocate page frames for
1898 * the image and we're done.
1899 */
1900 if (size >= saveable) {
1901 pages = preallocate_image_highmem(save_highmem);
1902 pages += preallocate_image_memory(saveable - pages, avail_normal);
1903 goto out;
1904 }
1905
1906 /* Estimate the minimum size of the image. */
1907 pages = minimum_image_size(saveable);
1908 /*
1909 * To avoid excessive pressure on the normal zone, leave room in it to
1910 * accommodate an image of the minimum size (unless it's already too
1911 * small, in which case don't preallocate pages from it at all).
1912 */
1913 if (avail_normal > pages)
1914 avail_normal -= pages;
1915 else
1916 avail_normal = 0;
1917 if (size < pages)
1918 size = min_t(unsigned long, pages, max_size);
1919
1920 /*
1921 * Let the memory management subsystem know that we're going to need a
1922 * large number of page frames to allocate and make it free some memory.
1923 * NOTE: If this is not done, performance will be hurt badly in some
1924 * test cases.
1925 */
1926 shrink_all_memory(saveable - size);
1927
1928 /*
1929 * The number of saveable pages in memory was too high, so apply some
1930 * pressure to decrease it. First, make room for the largest possible
1931 * image and fail if that doesn't work. Next, try to decrease the size
1932 * of the image as much as indicated by 'size' using allocations from
1933 * highmem and non-highmem zones separately.
1934 */
1935 pages_highmem = preallocate_image_highmem(highmem / 2);
1936 alloc = count - max_size;
1937 if (alloc > pages_highmem)
1938 alloc -= pages_highmem;
1939 else
1940 alloc = 0;
1941 pages = preallocate_image_memory(alloc, avail_normal);
1942 if (pages < alloc) {
1943 /* We have exhausted non-highmem pages, try highmem. */
1944 alloc -= pages;
1945 pages += pages_highmem;
1946 pages_highmem = preallocate_image_highmem(alloc);
1947 if (pages_highmem < alloc) {
1948 pr_err("Image allocation is %lu pages short\n",
1949 alloc - pages_highmem);
1950 goto err_out;
1951 }
1952 pages += pages_highmem;
1953 /*
1954 * size is the desired number of saveable pages to leave in
1955 * memory, so try to preallocate (all memory - size) pages.
1956 */
1957 alloc = (count - pages) - size;
1958 pages += preallocate_image_highmem(alloc);
1959 } else {
1960 /*
1961 * There are approximately max_size saveable pages at this point
1962 * and we want to reduce this number down to size.
1963 */
1964 alloc = max_size - size;
1965 size = preallocate_highmem_fraction(alloc, highmem, count);
1966 pages_highmem += size;
1967 alloc -= size;
1968 size = preallocate_image_memory(alloc, avail_normal);
1969 pages_highmem += preallocate_image_highmem(alloc - size);
1970 pages += pages_highmem + size;
1971 }
1972
1973 /*
1974 * We only need as many page frames for the image as there are saveable
1975 * pages in memory, but we have allocated more. Release the excessive
1976 * ones now.
1977 */
1978 pages -= free_unnecessary_pages();
1979
1980 out:
1981 stop = ktime_get();
1982 pr_info("Allocated %lu pages for snapshot\n", pages);
1983 swsusp_show_speed(start, stop, pages, "Allocated");
1984
1985 return 0;
1986
1987 err_out:
1988 swsusp_free();
1989 return -ENOMEM;
1990}
1991
1992#ifdef CONFIG_HIGHMEM
1993/**
1994 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1995 *
1996 * Compute the number of non-highmem pages that will be necessary for creating
1997 * copies of highmem pages.
1998 */
1999static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
2000{
2001 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
2002
2003 if (free_highmem >= nr_highmem)
2004 nr_highmem = 0;
2005 else
2006 nr_highmem -= free_highmem;
2007
2008 return nr_highmem;
2009}
2010#else
2011static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
2012#endif /* CONFIG_HIGHMEM */
2013
2014/**
2015 * enough_free_mem - Check if there is enough free memory for the image.
2016 */
2017static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
2018{
2019 struct zone *zone;
2020 unsigned int free = alloc_normal;
2021
2022 for_each_populated_zone(zone)
2023 if (!is_highmem(zone))
2024 free += zone_page_state(zone, NR_FREE_PAGES);
2025
2026 nr_pages += count_pages_for_highmem(nr_highmem);
2027 pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
2028 nr_pages, PAGES_FOR_IO, free);
2029
2030 return free > nr_pages + PAGES_FOR_IO;
2031}
2032
2033#ifdef CONFIG_HIGHMEM
2034/**
2035 * get_highmem_buffer - Allocate a buffer for highmem pages.
2036 *
2037 * If there are some highmem pages in the hibernation image, we may need a
2038 * buffer to copy them and/or load their data.
2039 */
2040static inline int get_highmem_buffer(int safe_needed)
2041{
2042 buffer = get_image_page(GFP_ATOMIC, safe_needed);
2043 return buffer ? 0 : -ENOMEM;
2044}
2045
2046/**
2047 * alloc_highmem_pages - Allocate some highmem pages for the image.
2048 *
2049 * Try to allocate as many pages as needed, but if the number of free highmem
2050 * pages is less than that, allocate them all.
2051 */
2052static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
2053 unsigned int nr_highmem)
2054{
2055 unsigned int to_alloc = count_free_highmem_pages();
2056
2057 if (to_alloc > nr_highmem)
2058 to_alloc = nr_highmem;
2059
2060 nr_highmem -= to_alloc;
2061 while (to_alloc-- > 0) {
2062 struct page *page;
2063
2064 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
2065 memory_bm_set_bit(bm, page_to_pfn(page));
2066 }
2067 return nr_highmem;
2068}
2069#else
2070static inline int get_highmem_buffer(int safe_needed) { return 0; }
2071
2072static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
2073 unsigned int n) { return 0; }
2074#endif /* CONFIG_HIGHMEM */
2075
2076/**
2077 * swsusp_alloc - Allocate memory for hibernation image.
2078 *
2079 * We first try to allocate as many highmem pages as there are
2080 * saveable highmem pages in the system. If that fails, we allocate
2081 * non-highmem pages for the copies of the remaining highmem ones.
2082 *
2083 * In this approach it is likely that the copies of highmem pages will
2084 * also be located in the high memory, because of the way in which
2085 * copy_data_pages() works.
2086 */
2087static int swsusp_alloc(struct memory_bitmap *copy_bm,
2088 unsigned int nr_pages, unsigned int nr_highmem)
2089{
2090 if (nr_highmem > 0) {
2091 if (get_highmem_buffer(PG_ANY))
2092 goto err_out;
2093 if (nr_highmem > alloc_highmem) {
2094 nr_highmem -= alloc_highmem;
2095 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
2096 }
2097 }
2098 if (nr_pages > alloc_normal) {
2099 nr_pages -= alloc_normal;
2100 while (nr_pages-- > 0) {
2101 struct page *page;
2102
2103 page = alloc_image_page(GFP_ATOMIC);
2104 if (!page)
2105 goto err_out;
2106 memory_bm_set_bit(copy_bm, page_to_pfn(page));
2107 }
2108 }
2109
2110 return 0;
2111
2112 err_out:
2113 swsusp_free();
2114 return -ENOMEM;
2115}
2116
2117asmlinkage __visible int swsusp_save(void)
2118{
2119 unsigned int nr_pages, nr_highmem;
2120
2121 pr_info("Creating image:\n");
2122
2123 drain_local_pages(NULL);
2124 nr_pages = count_data_pages();
2125 nr_highmem = count_highmem_pages();
2126 pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
2127
2128 if (!enough_free_mem(nr_pages, nr_highmem)) {
2129 pr_err("Not enough free memory\n");
2130 return -ENOMEM;
2131 }
2132
2133 if (swsusp_alloc(©_bm, nr_pages, nr_highmem)) {
2134 pr_err("Memory allocation failed\n");
2135 return -ENOMEM;
2136 }
2137
2138 /*
2139 * During allocating of suspend pagedir, new cold pages may appear.
2140 * Kill them.
2141 */
2142 drain_local_pages(NULL);
2143 nr_copy_pages = copy_data_pages(©_bm, &orig_bm, &zero_bm);
2144
2145 /*
2146 * End of critical section. From now on, we can write to memory,
2147 * but we should not touch disk. This specially means we must _not_
2148 * touch swap space! Except we must write out our image of course.
2149 */
2150 nr_pages += nr_highmem;
2151 /* We don't actually copy the zero pages */
2152 nr_zero_pages = nr_pages - nr_copy_pages;
2153 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2154
2155 pr_info("Image created (%d pages copied, %d zero pages)\n", nr_copy_pages, nr_zero_pages);
2156
2157 return 0;
2158}
2159
2160#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2161static int init_header_complete(struct swsusp_info *info)
2162{
2163 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2164 info->version_code = LINUX_VERSION_CODE;
2165 return 0;
2166}
2167
2168static const char *check_image_kernel(struct swsusp_info *info)
2169{
2170 if (info->version_code != LINUX_VERSION_CODE)
2171 return "kernel version";
2172 if (strcmp(info->uts.sysname,init_utsname()->sysname))
2173 return "system type";
2174 if (strcmp(info->uts.release,init_utsname()->release))
2175 return "kernel release";
2176 if (strcmp(info->uts.version,init_utsname()->version))
2177 return "version";
2178 if (strcmp(info->uts.machine,init_utsname()->machine))
2179 return "machine";
2180 return NULL;
2181}
2182#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2183
2184unsigned long snapshot_get_image_size(void)
2185{
2186 return nr_copy_pages + nr_meta_pages + 1;
2187}
2188
2189static int init_header(struct swsusp_info *info)
2190{
2191 memset(info, 0, sizeof(struct swsusp_info));
2192 info->num_physpages = get_num_physpages();
2193 info->image_pages = nr_copy_pages;
2194 info->pages = snapshot_get_image_size();
2195 info->size = info->pages;
2196 info->size <<= PAGE_SHIFT;
2197 return init_header_complete(info);
2198}
2199
2200#define ENCODED_PFN_ZERO_FLAG ((unsigned long)1 << (BITS_PER_LONG - 1))
2201#define ENCODED_PFN_MASK (~ENCODED_PFN_ZERO_FLAG)
2202
2203/**
2204 * pack_pfns - Prepare PFNs for saving.
2205 * @bm: Memory bitmap.
2206 * @buf: Memory buffer to store the PFNs in.
2207 * @zero_bm: Memory bitmap containing PFNs of zero pages.
2208 *
2209 * PFNs corresponding to set bits in @bm are stored in the area of memory
2210 * pointed to by @buf (1 page at a time). Pages which were filled with only
2211 * zeros will have the highest bit set in the packed format to distinguish
2212 * them from PFNs which will be contained in the image file.
2213 */
2214static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm,
2215 struct memory_bitmap *zero_bm)
2216{
2217 int j;
2218
2219 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2220 buf[j] = memory_bm_next_pfn(bm);
2221 if (unlikely(buf[j] == BM_END_OF_MAP))
2222 break;
2223 if (memory_bm_test_bit(zero_bm, buf[j]))
2224 buf[j] |= ENCODED_PFN_ZERO_FLAG;
2225 }
2226}
2227
2228/**
2229 * snapshot_read_next - Get the address to read the next image page from.
2230 * @handle: Snapshot handle to be used for the reading.
2231 *
2232 * On the first call, @handle should point to a zeroed snapshot_handle
2233 * structure. The structure gets populated then and a pointer to it should be
2234 * passed to this function every next time.
2235 *
2236 * On success, the function returns a positive number. Then, the caller
2237 * is allowed to read up to the returned number of bytes from the memory
2238 * location computed by the data_of() macro.
2239 *
2240 * The function returns 0 to indicate the end of the data stream condition,
2241 * and negative numbers are returned on errors. If that happens, the structure
2242 * pointed to by @handle is not updated and should not be used any more.
2243 */
2244int snapshot_read_next(struct snapshot_handle *handle)
2245{
2246 if (handle->cur > nr_meta_pages + nr_copy_pages)
2247 return 0;
2248
2249 if (!buffer) {
2250 /* This makes the buffer be freed by swsusp_free() */
2251 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2252 if (!buffer)
2253 return -ENOMEM;
2254 }
2255 if (!handle->cur) {
2256 int error;
2257
2258 error = init_header((struct swsusp_info *)buffer);
2259 if (error)
2260 return error;
2261 handle->buffer = buffer;
2262 memory_bm_position_reset(&orig_bm);
2263 memory_bm_position_reset(©_bm);
2264 } else if (handle->cur <= nr_meta_pages) {
2265 clear_page(buffer);
2266 pack_pfns(buffer, &orig_bm, &zero_bm);
2267 } else {
2268 struct page *page;
2269
2270 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2271 if (PageHighMem(page)) {
2272 /*
2273 * Highmem pages are copied to the buffer,
2274 * because we can't return with a kmapped
2275 * highmem page (we may not be called again).
2276 */
2277 void *kaddr;
2278
2279 kaddr = kmap_atomic(page);
2280 copy_page(buffer, kaddr);
2281 kunmap_atomic(kaddr);
2282 handle->buffer = buffer;
2283 } else {
2284 handle->buffer = page_address(page);
2285 }
2286 }
2287 handle->cur++;
2288 return PAGE_SIZE;
2289}
2290
2291static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2292 struct memory_bitmap *src)
2293{
2294 unsigned long pfn;
2295
2296 memory_bm_position_reset(src);
2297 pfn = memory_bm_next_pfn(src);
2298 while (pfn != BM_END_OF_MAP) {
2299 memory_bm_set_bit(dst, pfn);
2300 pfn = memory_bm_next_pfn(src);
2301 }
2302}
2303
2304/**
2305 * mark_unsafe_pages - Mark pages that were used before hibernation.
2306 *
2307 * Mark the pages that cannot be used for storing the image during restoration,
2308 * because they conflict with the pages that had been used before hibernation.
2309 */
2310static void mark_unsafe_pages(struct memory_bitmap *bm)
2311{
2312 unsigned long pfn;
2313
2314 /* Clear the "free"/"unsafe" bit for all PFNs */
2315 memory_bm_position_reset(free_pages_map);
2316 pfn = memory_bm_next_pfn(free_pages_map);
2317 while (pfn != BM_END_OF_MAP) {
2318 memory_bm_clear_current(free_pages_map);
2319 pfn = memory_bm_next_pfn(free_pages_map);
2320 }
2321
2322 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2323 duplicate_memory_bitmap(free_pages_map, bm);
2324
2325 allocated_unsafe_pages = 0;
2326}
2327
2328static int check_header(struct swsusp_info *info)
2329{
2330 const char *reason;
2331
2332 reason = check_image_kernel(info);
2333 if (!reason && info->num_physpages != get_num_physpages())
2334 reason = "memory size";
2335 if (reason) {
2336 pr_err("Image mismatch: %s\n", reason);
2337 return -EPERM;
2338 }
2339 return 0;
2340}
2341
2342/**
2343 * load_header - Check the image header and copy the data from it.
2344 */
2345static int load_header(struct swsusp_info *info)
2346{
2347 int error;
2348
2349 restore_pblist = NULL;
2350 error = check_header(info);
2351 if (!error) {
2352 nr_copy_pages = info->image_pages;
2353 nr_meta_pages = info->pages - info->image_pages - 1;
2354 }
2355 return error;
2356}
2357
2358/**
2359 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2360 * @bm: Memory bitmap.
2361 * @buf: Area of memory containing the PFNs.
2362 * @zero_bm: Memory bitmap with the zero PFNs marked.
2363 *
2364 * For each element of the array pointed to by @buf (1 page at a time), set the
2365 * corresponding bit in @bm. If the page was originally populated with only
2366 * zeros then a corresponding bit will also be set in @zero_bm.
2367 */
2368static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm,
2369 struct memory_bitmap *zero_bm)
2370{
2371 unsigned long decoded_pfn;
2372 bool zero;
2373 int j;
2374
2375 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2376 if (unlikely(buf[j] == BM_END_OF_MAP))
2377 break;
2378
2379 zero = !!(buf[j] & ENCODED_PFN_ZERO_FLAG);
2380 decoded_pfn = buf[j] & ENCODED_PFN_MASK;
2381 if (pfn_valid(decoded_pfn) && memory_bm_pfn_present(bm, decoded_pfn)) {
2382 memory_bm_set_bit(bm, decoded_pfn);
2383 if (zero) {
2384 memory_bm_set_bit(zero_bm, decoded_pfn);
2385 nr_zero_pages++;
2386 }
2387 } else {
2388 if (!pfn_valid(decoded_pfn))
2389 pr_err(FW_BUG "Memory map mismatch at 0x%llx after hibernation\n",
2390 (unsigned long long)PFN_PHYS(decoded_pfn));
2391 return -EFAULT;
2392 }
2393 }
2394
2395 return 0;
2396}
2397
2398#ifdef CONFIG_HIGHMEM
2399/*
2400 * struct highmem_pbe is used for creating the list of highmem pages that
2401 * should be restored atomically during the resume from disk, because the page
2402 * frames they have occupied before the suspend are in use.
2403 */
2404struct highmem_pbe {
2405 struct page *copy_page; /* data is here now */
2406 struct page *orig_page; /* data was here before the suspend */
2407 struct highmem_pbe *next;
2408};
2409
2410/*
2411 * List of highmem PBEs needed for restoring the highmem pages that were
2412 * allocated before the suspend and included in the suspend image, but have
2413 * also been allocated by the "resume" kernel, so their contents cannot be
2414 * written directly to their "original" page frames.
2415 */
2416static struct highmem_pbe *highmem_pblist;
2417
2418/**
2419 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2420 * @bm: Memory bitmap.
2421 *
2422 * The bits in @bm that correspond to image pages are assumed to be set.
2423 */
2424static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2425{
2426 unsigned long pfn;
2427 unsigned int cnt = 0;
2428
2429 memory_bm_position_reset(bm);
2430 pfn = memory_bm_next_pfn(bm);
2431 while (pfn != BM_END_OF_MAP) {
2432 if (PageHighMem(pfn_to_page(pfn)))
2433 cnt++;
2434
2435 pfn = memory_bm_next_pfn(bm);
2436 }
2437 return cnt;
2438}
2439
2440static unsigned int safe_highmem_pages;
2441
2442static struct memory_bitmap *safe_highmem_bm;
2443
2444/**
2445 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2446 * @bm: Pointer to an uninitialized memory bitmap structure.
2447 * @nr_highmem_p: Pointer to the number of highmem image pages.
2448 *
2449 * Try to allocate as many highmem pages as there are highmem image pages
2450 * (@nr_highmem_p points to the variable containing the number of highmem image
2451 * pages). The pages that are "safe" (ie. will not be overwritten when the
2452 * hibernation image is restored entirely) have the corresponding bits set in
2453 * @bm (it must be uninitialized).
2454 *
2455 * NOTE: This function should not be called if there are no highmem image pages.
2456 */
2457static int prepare_highmem_image(struct memory_bitmap *bm,
2458 unsigned int *nr_highmem_p)
2459{
2460 unsigned int to_alloc;
2461
2462 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2463 return -ENOMEM;
2464
2465 if (get_highmem_buffer(PG_SAFE))
2466 return -ENOMEM;
2467
2468 to_alloc = count_free_highmem_pages();
2469 if (to_alloc > *nr_highmem_p)
2470 to_alloc = *nr_highmem_p;
2471 else
2472 *nr_highmem_p = to_alloc;
2473
2474 safe_highmem_pages = 0;
2475 while (to_alloc-- > 0) {
2476 struct page *page;
2477
2478 page = alloc_page(__GFP_HIGHMEM);
2479 if (!swsusp_page_is_free(page)) {
2480 /* The page is "safe", set its bit the bitmap */
2481 memory_bm_set_bit(bm, page_to_pfn(page));
2482 safe_highmem_pages++;
2483 }
2484 /* Mark the page as allocated */
2485 swsusp_set_page_forbidden(page);
2486 swsusp_set_page_free(page);
2487 }
2488 memory_bm_position_reset(bm);
2489 safe_highmem_bm = bm;
2490 return 0;
2491}
2492
2493static struct page *last_highmem_page;
2494
2495/**
2496 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2497 *
2498 * For a given highmem image page get a buffer that suspend_write_next() should
2499 * return to its caller to write to.
2500 *
2501 * If the page is to be saved to its "original" page frame or a copy of
2502 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2503 * the copy of the page is to be made in normal memory, so the address of
2504 * the copy is returned.
2505 *
2506 * If @buffer is returned, the caller of suspend_write_next() will write
2507 * the page's contents to @buffer, so they will have to be copied to the
2508 * right location on the next call to suspend_write_next() and it is done
2509 * with the help of copy_last_highmem_page(). For this purpose, if
2510 * @buffer is returned, @last_highmem_page is set to the page to which
2511 * the data will have to be copied from @buffer.
2512 */
2513static void *get_highmem_page_buffer(struct page *page,
2514 struct chain_allocator *ca)
2515{
2516 struct highmem_pbe *pbe;
2517 void *kaddr;
2518
2519 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2520 /*
2521 * We have allocated the "original" page frame and we can
2522 * use it directly to store the loaded page.
2523 */
2524 last_highmem_page = page;
2525 return buffer;
2526 }
2527 /*
2528 * The "original" page frame has not been allocated and we have to
2529 * use a "safe" page frame to store the loaded page.
2530 */
2531 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2532 if (!pbe) {
2533 swsusp_free();
2534 return ERR_PTR(-ENOMEM);
2535 }
2536 pbe->orig_page = page;
2537 if (safe_highmem_pages > 0) {
2538 struct page *tmp;
2539
2540 /* Copy of the page will be stored in high memory */
2541 kaddr = buffer;
2542 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2543 safe_highmem_pages--;
2544 last_highmem_page = tmp;
2545 pbe->copy_page = tmp;
2546 } else {
2547 /* Copy of the page will be stored in normal memory */
2548 kaddr = __get_safe_page(ca->gfp_mask);
2549 if (!kaddr)
2550 return ERR_PTR(-ENOMEM);
2551 pbe->copy_page = virt_to_page(kaddr);
2552 }
2553 pbe->next = highmem_pblist;
2554 highmem_pblist = pbe;
2555 return kaddr;
2556}
2557
2558/**
2559 * copy_last_highmem_page - Copy most the most recent highmem image page.
2560 *
2561 * Copy the contents of a highmem image from @buffer, where the caller of
2562 * snapshot_write_next() has stored them, to the right location represented by
2563 * @last_highmem_page .
2564 */
2565static void copy_last_highmem_page(void)
2566{
2567 if (last_highmem_page) {
2568 void *dst;
2569
2570 dst = kmap_atomic(last_highmem_page);
2571 copy_page(dst, buffer);
2572 kunmap_atomic(dst);
2573 last_highmem_page = NULL;
2574 }
2575}
2576
2577static inline int last_highmem_page_copied(void)
2578{
2579 return !last_highmem_page;
2580}
2581
2582static inline void free_highmem_data(void)
2583{
2584 if (safe_highmem_bm)
2585 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2586
2587 if (buffer)
2588 free_image_page(buffer, PG_UNSAFE_CLEAR);
2589}
2590#else
2591static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2592
2593static inline int prepare_highmem_image(struct memory_bitmap *bm,
2594 unsigned int *nr_highmem_p) { return 0; }
2595
2596static inline void *get_highmem_page_buffer(struct page *page,
2597 struct chain_allocator *ca)
2598{
2599 return ERR_PTR(-EINVAL);
2600}
2601
2602static inline void copy_last_highmem_page(void) {}
2603static inline int last_highmem_page_copied(void) { return 1; }
2604static inline void free_highmem_data(void) {}
2605#endif /* CONFIG_HIGHMEM */
2606
2607#define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2608
2609/**
2610 * prepare_image - Make room for loading hibernation image.
2611 * @new_bm: Uninitialized memory bitmap structure.
2612 * @bm: Memory bitmap with unsafe pages marked.
2613 * @zero_bm: Memory bitmap containing the zero pages.
2614 *
2615 * Use @bm to mark the pages that will be overwritten in the process of
2616 * restoring the system memory state from the suspend image ("unsafe" pages)
2617 * and allocate memory for the image.
2618 *
2619 * The idea is to allocate a new memory bitmap first and then allocate
2620 * as many pages as needed for image data, but without specifying what those
2621 * pages will be used for just yet. Instead, we mark them all as allocated and
2622 * create a lists of "safe" pages to be used later. On systems with high
2623 * memory a list of "safe" highmem pages is created too.
2624 *
2625 * Because it was not known which pages were unsafe when @zero_bm was created,
2626 * make a copy of it and recreate it within safe pages.
2627 */
2628static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm,
2629 struct memory_bitmap *zero_bm)
2630{
2631 unsigned int nr_pages, nr_highmem;
2632 struct memory_bitmap tmp;
2633 struct linked_page *lp;
2634 int error;
2635
2636 /* If there is no highmem, the buffer will not be necessary */
2637 free_image_page(buffer, PG_UNSAFE_CLEAR);
2638 buffer = NULL;
2639
2640 nr_highmem = count_highmem_image_pages(bm);
2641 mark_unsafe_pages(bm);
2642
2643 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2644 if (error)
2645 goto Free;
2646
2647 duplicate_memory_bitmap(new_bm, bm);
2648 memory_bm_free(bm, PG_UNSAFE_KEEP);
2649
2650 /* Make a copy of zero_bm so it can be created in safe pages */
2651 error = memory_bm_create(&tmp, GFP_ATOMIC, PG_SAFE);
2652 if (error)
2653 goto Free;
2654
2655 duplicate_memory_bitmap(&tmp, zero_bm);
2656 memory_bm_free(zero_bm, PG_UNSAFE_KEEP);
2657
2658 /* Recreate zero_bm in safe pages */
2659 error = memory_bm_create(zero_bm, GFP_ATOMIC, PG_SAFE);
2660 if (error)
2661 goto Free;
2662
2663 duplicate_memory_bitmap(zero_bm, &tmp);
2664 memory_bm_free(&tmp, PG_UNSAFE_CLEAR);
2665 /* At this point zero_bm is in safe pages and it can be used for restoring. */
2666
2667 if (nr_highmem > 0) {
2668 error = prepare_highmem_image(bm, &nr_highmem);
2669 if (error)
2670 goto Free;
2671 }
2672 /*
2673 * Reserve some safe pages for potential later use.
2674 *
2675 * NOTE: This way we make sure there will be enough safe pages for the
2676 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2677 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2678 *
2679 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2680 */
2681 nr_pages = (nr_zero_pages + nr_copy_pages) - nr_highmem - allocated_unsafe_pages;
2682 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2683 while (nr_pages > 0) {
2684 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2685 if (!lp) {
2686 error = -ENOMEM;
2687 goto Free;
2688 }
2689 lp->next = safe_pages_list;
2690 safe_pages_list = lp;
2691 nr_pages--;
2692 }
2693 /* Preallocate memory for the image */
2694 nr_pages = (nr_zero_pages + nr_copy_pages) - nr_highmem - allocated_unsafe_pages;
2695 while (nr_pages > 0) {
2696 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2697 if (!lp) {
2698 error = -ENOMEM;
2699 goto Free;
2700 }
2701 if (!swsusp_page_is_free(virt_to_page(lp))) {
2702 /* The page is "safe", add it to the list */
2703 lp->next = safe_pages_list;
2704 safe_pages_list = lp;
2705 }
2706 /* Mark the page as allocated */
2707 swsusp_set_page_forbidden(virt_to_page(lp));
2708 swsusp_set_page_free(virt_to_page(lp));
2709 nr_pages--;
2710 }
2711 return 0;
2712
2713 Free:
2714 swsusp_free();
2715 return error;
2716}
2717
2718/**
2719 * get_buffer - Get the address to store the next image data page.
2720 *
2721 * Get the address that snapshot_write_next() should return to its caller to
2722 * write to.
2723 */
2724static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2725{
2726 struct pbe *pbe;
2727 struct page *page;
2728 unsigned long pfn = memory_bm_next_pfn(bm);
2729
2730 if (pfn == BM_END_OF_MAP)
2731 return ERR_PTR(-EFAULT);
2732
2733 page = pfn_to_page(pfn);
2734 if (PageHighMem(page))
2735 return get_highmem_page_buffer(page, ca);
2736
2737 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2738 /*
2739 * We have allocated the "original" page frame and we can
2740 * use it directly to store the loaded page.
2741 */
2742 return page_address(page);
2743
2744 /*
2745 * The "original" page frame has not been allocated and we have to
2746 * use a "safe" page frame to store the loaded page.
2747 */
2748 pbe = chain_alloc(ca, sizeof(struct pbe));
2749 if (!pbe) {
2750 swsusp_free();
2751 return ERR_PTR(-ENOMEM);
2752 }
2753 pbe->orig_address = page_address(page);
2754 pbe->address = __get_safe_page(ca->gfp_mask);
2755 if (!pbe->address)
2756 return ERR_PTR(-ENOMEM);
2757 pbe->next = restore_pblist;
2758 restore_pblist = pbe;
2759 return pbe->address;
2760}
2761
2762/**
2763 * snapshot_write_next - Get the address to store the next image page.
2764 * @handle: Snapshot handle structure to guide the writing.
2765 *
2766 * On the first call, @handle should point to a zeroed snapshot_handle
2767 * structure. The structure gets populated then and a pointer to it should be
2768 * passed to this function every next time.
2769 *
2770 * On success, the function returns a positive number. Then, the caller
2771 * is allowed to write up to the returned number of bytes to the memory
2772 * location computed by the data_of() macro.
2773 *
2774 * The function returns 0 to indicate the "end of file" condition. Negative
2775 * numbers are returned on errors, in which cases the structure pointed to by
2776 * @handle is not updated and should not be used any more.
2777 */
2778int snapshot_write_next(struct snapshot_handle *handle)
2779{
2780 static struct chain_allocator ca;
2781 int error;
2782
2783next:
2784 /* Check if we have already loaded the entire image */
2785 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages + nr_zero_pages)
2786 return 0;
2787
2788 if (!handle->cur) {
2789 if (!buffer)
2790 /* This makes the buffer be freed by swsusp_free() */
2791 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2792
2793 if (!buffer)
2794 return -ENOMEM;
2795
2796 handle->buffer = buffer;
2797 } else if (handle->cur == 1) {
2798 error = load_header(buffer);
2799 if (error)
2800 return error;
2801
2802 safe_pages_list = NULL;
2803
2804 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2805 if (error)
2806 return error;
2807
2808 error = memory_bm_create(&zero_bm, GFP_ATOMIC, PG_ANY);
2809 if (error)
2810 return error;
2811
2812 nr_zero_pages = 0;
2813
2814 hibernate_restore_protection_begin();
2815 } else if (handle->cur <= nr_meta_pages + 1) {
2816 error = unpack_orig_pfns(buffer, ©_bm, &zero_bm);
2817 if (error)
2818 return error;
2819
2820 if (handle->cur == nr_meta_pages + 1) {
2821 error = prepare_image(&orig_bm, ©_bm, &zero_bm);
2822 if (error)
2823 return error;
2824
2825 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2826 memory_bm_position_reset(&orig_bm);
2827 memory_bm_position_reset(&zero_bm);
2828 restore_pblist = NULL;
2829 handle->buffer = get_buffer(&orig_bm, &ca);
2830 if (IS_ERR(handle->buffer))
2831 return PTR_ERR(handle->buffer);
2832 }
2833 } else {
2834 copy_last_highmem_page();
2835 hibernate_restore_protect_page(handle->buffer);
2836 handle->buffer = get_buffer(&orig_bm, &ca);
2837 if (IS_ERR(handle->buffer))
2838 return PTR_ERR(handle->buffer);
2839 }
2840 handle->sync_read = (handle->buffer == buffer);
2841 handle->cur++;
2842
2843 /* Zero pages were not included in the image, memset it and move on. */
2844 if (handle->cur > nr_meta_pages + 1 &&
2845 memory_bm_test_bit(&zero_bm, memory_bm_get_current(&orig_bm))) {
2846 memset(handle->buffer, 0, PAGE_SIZE);
2847 goto next;
2848 }
2849
2850 return PAGE_SIZE;
2851}
2852
2853/**
2854 * snapshot_write_finalize - Complete the loading of a hibernation image.
2855 *
2856 * Must be called after the last call to snapshot_write_next() in case the last
2857 * page in the image happens to be a highmem page and its contents should be
2858 * stored in highmem. Additionally, it recycles bitmap memory that's not
2859 * necessary any more.
2860 */
2861void snapshot_write_finalize(struct snapshot_handle *handle)
2862{
2863 copy_last_highmem_page();
2864 hibernate_restore_protect_page(handle->buffer);
2865 /* Do that only if we have loaded the image entirely */
2866 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages + nr_zero_pages) {
2867 memory_bm_recycle(&orig_bm);
2868 free_highmem_data();
2869 }
2870}
2871
2872int snapshot_image_loaded(struct snapshot_handle *handle)
2873{
2874 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2875 handle->cur <= nr_meta_pages + nr_copy_pages + nr_zero_pages);
2876}
2877
2878#ifdef CONFIG_HIGHMEM
2879/* Assumes that @buf is ready and points to a "safe" page */
2880static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2881 void *buf)
2882{
2883 void *kaddr1, *kaddr2;
2884
2885 kaddr1 = kmap_atomic(p1);
2886 kaddr2 = kmap_atomic(p2);
2887 copy_page(buf, kaddr1);
2888 copy_page(kaddr1, kaddr2);
2889 copy_page(kaddr2, buf);
2890 kunmap_atomic(kaddr2);
2891 kunmap_atomic(kaddr1);
2892}
2893
2894/**
2895 * restore_highmem - Put highmem image pages into their original locations.
2896 *
2897 * For each highmem page that was in use before hibernation and is included in
2898 * the image, and also has been allocated by the "restore" kernel, swap its
2899 * current contents with the previous (ie. "before hibernation") ones.
2900 *
2901 * If the restore eventually fails, we can call this function once again and
2902 * restore the highmem state as seen by the restore kernel.
2903 */
2904int restore_highmem(void)
2905{
2906 struct highmem_pbe *pbe = highmem_pblist;
2907 void *buf;
2908
2909 if (!pbe)
2910 return 0;
2911
2912 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2913 if (!buf)
2914 return -ENOMEM;
2915
2916 while (pbe) {
2917 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2918 pbe = pbe->next;
2919 }
2920 free_image_page(buf, PG_UNSAFE_CLEAR);
2921 return 0;
2922}
2923#endif /* CONFIG_HIGHMEM */