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#define pr_fmt(fmt) "PM: " fmt
14
15#include <linux/version.h>
16#include <linux/module.h>
17#include <linux/mm.h>
18#include <linux/suspend.h>
19#include <linux/delay.h>
20#include <linux/bitops.h>
21#include <linux/spinlock.h>
22#include <linux/kernel.h>
23#include <linux/pm.h>
24#include <linux/device.h>
25#include <linux/init.h>
26#include <linux/bootmem.h>
27#include <linux/nmi.h>
28#include <linux/syscalls.h>
29#include <linux/console.h>
30#include <linux/highmem.h>
31#include <linux/list.h>
32#include <linux/slab.h>
33#include <linux/compiler.h>
34#include <linux/ktime.h>
35#include <linux/set_memory.h>
36
37#include <linux/uaccess.h>
38#include <asm/mmu_context.h>
39#include <asm/pgtable.h>
40#include <asm/tlbflush.h>
41#include <asm/io.h>
42
43#include "power.h"
44
45#if defined(CONFIG_STRICT_KERNEL_RWX) && defined(CONFIG_ARCH_HAS_SET_MEMORY)
46static bool hibernate_restore_protection;
47static bool hibernate_restore_protection_active;
48
49void enable_restore_image_protection(void)
50{
51 hibernate_restore_protection = true;
52}
53
54static inline void hibernate_restore_protection_begin(void)
55{
56 hibernate_restore_protection_active = hibernate_restore_protection;
57}
58
59static inline void hibernate_restore_protection_end(void)
60{
61 hibernate_restore_protection_active = false;
62}
63
64static inline void hibernate_restore_protect_page(void *page_address)
65{
66 if (hibernate_restore_protection_active)
67 set_memory_ro((unsigned long)page_address, 1);
68}
69
70static inline void hibernate_restore_unprotect_page(void *page_address)
71{
72 if (hibernate_restore_protection_active)
73 set_memory_rw((unsigned long)page_address, 1);
74}
75#else
76static inline void hibernate_restore_protection_begin(void) {}
77static inline void hibernate_restore_protection_end(void) {}
78static inline void hibernate_restore_protect_page(void *page_address) {}
79static inline void hibernate_restore_unprotect_page(void *page_address) {}
80#endif /* CONFIG_STRICT_KERNEL_RWX && CONFIG_ARCH_HAS_SET_MEMORY */
81
82static int swsusp_page_is_free(struct page *);
83static void swsusp_set_page_forbidden(struct page *);
84static void swsusp_unset_page_forbidden(struct page *);
85
86/*
87 * Number of bytes to reserve for memory allocations made by device drivers
88 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
89 * cause image creation to fail (tunable via /sys/power/reserved_size).
90 */
91unsigned long reserved_size;
92
93void __init hibernate_reserved_size_init(void)
94{
95 reserved_size = SPARE_PAGES * PAGE_SIZE;
96}
97
98/*
99 * Preferred image size in bytes (tunable via /sys/power/image_size).
100 * When it is set to N, swsusp will do its best to ensure the image
101 * size will not exceed N bytes, but if that is impossible, it will
102 * try to create the smallest image possible.
103 */
104unsigned long image_size;
105
106void __init hibernate_image_size_init(void)
107{
108 image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE;
109}
110
111/*
112 * List of PBEs needed for restoring the pages that were allocated before
113 * the suspend and included in the suspend image, but have also been
114 * allocated by the "resume" kernel, so their contents cannot be written
115 * directly to their "original" page frames.
116 */
117struct pbe *restore_pblist;
118
119/* struct linked_page is used to build chains of pages */
120
121#define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
122
123struct linked_page {
124 struct linked_page *next;
125 char data[LINKED_PAGE_DATA_SIZE];
126} __packed;
127
128/*
129 * List of "safe" pages (ie. pages that were not used by the image kernel
130 * before hibernation) that may be used as temporary storage for image kernel
131 * memory contents.
132 */
133static struct linked_page *safe_pages_list;
134
135/* Pointer to an auxiliary buffer (1 page) */
136static void *buffer;
137
138#define PG_ANY 0
139#define PG_SAFE 1
140#define PG_UNSAFE_CLEAR 1
141#define PG_UNSAFE_KEEP 0
142
143static unsigned int allocated_unsafe_pages;
144
145/**
146 * get_image_page - Allocate a page for a hibernation image.
147 * @gfp_mask: GFP mask for the allocation.
148 * @safe_needed: Get pages that were not used before hibernation (restore only)
149 *
150 * During image restoration, for storing the PBE list and the image data, we can
151 * only use memory pages that do not conflict with the pages used before
152 * hibernation. The "unsafe" pages have PageNosaveFree set and we count them
153 * using allocated_unsafe_pages.
154 *
155 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
156 * swsusp_free() can release it.
157 */
158static void *get_image_page(gfp_t gfp_mask, int safe_needed)
159{
160 void *res;
161
162 res = (void *)get_zeroed_page(gfp_mask);
163 if (safe_needed)
164 while (res && swsusp_page_is_free(virt_to_page(res))) {
165 /* The page is unsafe, mark it for swsusp_free() */
166 swsusp_set_page_forbidden(virt_to_page(res));
167 allocated_unsafe_pages++;
168 res = (void *)get_zeroed_page(gfp_mask);
169 }
170 if (res) {
171 swsusp_set_page_forbidden(virt_to_page(res));
172 swsusp_set_page_free(virt_to_page(res));
173 }
174 return res;
175}
176
177static void *__get_safe_page(gfp_t gfp_mask)
178{
179 if (safe_pages_list) {
180 void *ret = safe_pages_list;
181
182 safe_pages_list = safe_pages_list->next;
183 memset(ret, 0, PAGE_SIZE);
184 return ret;
185 }
186 return get_image_page(gfp_mask, PG_SAFE);
187}
188
189unsigned long get_safe_page(gfp_t gfp_mask)
190{
191 return (unsigned long)__get_safe_page(gfp_mask);
192}
193
194static struct page *alloc_image_page(gfp_t gfp_mask)
195{
196 struct page *page;
197
198 page = alloc_page(gfp_mask);
199 if (page) {
200 swsusp_set_page_forbidden(page);
201 swsusp_set_page_free(page);
202 }
203 return page;
204}
205
206static void recycle_safe_page(void *page_address)
207{
208 struct linked_page *lp = page_address;
209
210 lp->next = safe_pages_list;
211 safe_pages_list = lp;
212}
213
214/**
215 * free_image_page - Free a page allocated for hibernation image.
216 * @addr: Address of the page to free.
217 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
218 *
219 * The page to free should have been allocated by get_image_page() (page flags
220 * set by it are affected).
221 */
222static inline void free_image_page(void *addr, int clear_nosave_free)
223{
224 struct page *page;
225
226 BUG_ON(!virt_addr_valid(addr));
227
228 page = virt_to_page(addr);
229
230 swsusp_unset_page_forbidden(page);
231 if (clear_nosave_free)
232 swsusp_unset_page_free(page);
233
234 __free_page(page);
235}
236
237static inline void free_list_of_pages(struct linked_page *list,
238 int clear_page_nosave)
239{
240 while (list) {
241 struct linked_page *lp = list->next;
242
243 free_image_page(list, clear_page_nosave);
244 list = lp;
245 }
246}
247
248/*
249 * struct chain_allocator is used for allocating small objects out of
250 * a linked list of pages called 'the chain'.
251 *
252 * The chain grows each time when there is no room for a new object in
253 * the current page. The allocated objects cannot be freed individually.
254 * It is only possible to free them all at once, by freeing the entire
255 * chain.
256 *
257 * NOTE: The chain allocator may be inefficient if the allocated objects
258 * are not much smaller than PAGE_SIZE.
259 */
260struct chain_allocator {
261 struct linked_page *chain; /* the chain */
262 unsigned int used_space; /* total size of objects allocated out
263 of the current page */
264 gfp_t gfp_mask; /* mask for allocating pages */
265 int safe_needed; /* if set, only "safe" pages are allocated */
266};
267
268static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
269 int safe_needed)
270{
271 ca->chain = NULL;
272 ca->used_space = LINKED_PAGE_DATA_SIZE;
273 ca->gfp_mask = gfp_mask;
274 ca->safe_needed = safe_needed;
275}
276
277static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
278{
279 void *ret;
280
281 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
282 struct linked_page *lp;
283
284 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
285 get_image_page(ca->gfp_mask, PG_ANY);
286 if (!lp)
287 return NULL;
288
289 lp->next = ca->chain;
290 ca->chain = lp;
291 ca->used_space = 0;
292 }
293 ret = ca->chain->data + ca->used_space;
294 ca->used_space += size;
295 return ret;
296}
297
298/**
299 * Data types related to memory bitmaps.
300 *
301 * Memory bitmap is a structure consiting of many linked lists of
302 * objects. The main list's elements are of type struct zone_bitmap
303 * and each of them corresonds to one zone. For each zone bitmap
304 * object there is a list of objects of type struct bm_block that
305 * represent each blocks of bitmap in which information is stored.
306 *
307 * struct memory_bitmap contains a pointer to the main list of zone
308 * bitmap objects, a struct bm_position used for browsing the bitmap,
309 * and a pointer to the list of pages used for allocating all of the
310 * zone bitmap objects and bitmap block objects.
311 *
312 * NOTE: It has to be possible to lay out the bitmap in memory
313 * using only allocations of order 0. Additionally, the bitmap is
314 * designed to work with arbitrary number of zones (this is over the
315 * top for now, but let's avoid making unnecessary assumptions ;-).
316 *
317 * struct zone_bitmap contains a pointer to a list of bitmap block
318 * objects and a pointer to the bitmap block object that has been
319 * most recently used for setting bits. Additionally, it contains the
320 * PFNs that correspond to the start and end of the represented zone.
321 *
322 * struct bm_block contains a pointer to the memory page in which
323 * information is stored (in the form of a block of bitmap)
324 * It also contains the pfns that correspond to the start and end of
325 * the represented memory area.
326 *
327 * The memory bitmap is organized as a radix tree to guarantee fast random
328 * access to the bits. There is one radix tree for each zone (as returned
329 * from create_mem_extents).
330 *
331 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
332 * two linked lists for the nodes of the tree, one for the inner nodes and
333 * one for the leave nodes. The linked leave nodes are used for fast linear
334 * access of the memory bitmap.
335 *
336 * The struct rtree_node represents one node of the radix tree.
337 */
338
339#define BM_END_OF_MAP (~0UL)
340
341#define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
342#define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
343#define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
344
345/*
346 * struct rtree_node is a wrapper struct to link the nodes
347 * of the rtree together for easy linear iteration over
348 * bits and easy freeing
349 */
350struct rtree_node {
351 struct list_head list;
352 unsigned long *data;
353};
354
355/*
356 * struct mem_zone_bm_rtree represents a bitmap used for one
357 * populated memory zone.
358 */
359struct mem_zone_bm_rtree {
360 struct list_head list; /* Link Zones together */
361 struct list_head nodes; /* Radix Tree inner nodes */
362 struct list_head leaves; /* Radix Tree leaves */
363 unsigned long start_pfn; /* Zone start page frame */
364 unsigned long end_pfn; /* Zone end page frame + 1 */
365 struct rtree_node *rtree; /* Radix Tree Root */
366 int levels; /* Number of Radix Tree Levels */
367 unsigned int blocks; /* Number of Bitmap Blocks */
368};
369
370/* strcut bm_position is used for browsing memory bitmaps */
371
372struct bm_position {
373 struct mem_zone_bm_rtree *zone;
374 struct rtree_node *node;
375 unsigned long node_pfn;
376 int node_bit;
377};
378
379struct memory_bitmap {
380 struct list_head zones;
381 struct linked_page *p_list; /* list of pages used to store zone
382 bitmap objects and bitmap block
383 objects */
384 struct bm_position cur; /* most recently used bit position */
385};
386
387/* Functions that operate on memory bitmaps */
388
389#define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
390#if BITS_PER_LONG == 32
391#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
392#else
393#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
394#endif
395#define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
396
397/**
398 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
399 *
400 * This function is used to allocate inner nodes as well as the
401 * leave nodes of the radix tree. It also adds the node to the
402 * corresponding linked list passed in by the *list parameter.
403 */
404static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
405 struct chain_allocator *ca,
406 struct list_head *list)
407{
408 struct rtree_node *node;
409
410 node = chain_alloc(ca, sizeof(struct rtree_node));
411 if (!node)
412 return NULL;
413
414 node->data = get_image_page(gfp_mask, safe_needed);
415 if (!node->data)
416 return NULL;
417
418 list_add_tail(&node->list, list);
419
420 return node;
421}
422
423/**
424 * add_rtree_block - Add a new leave node to the radix tree.
425 *
426 * The leave nodes need to be allocated in order to keep the leaves
427 * linked list in order. This is guaranteed by the zone->blocks
428 * counter.
429 */
430static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
431 int safe_needed, struct chain_allocator *ca)
432{
433 struct rtree_node *node, *block, **dst;
434 unsigned int levels_needed, block_nr;
435 int i;
436
437 block_nr = zone->blocks;
438 levels_needed = 0;
439
440 /* How many levels do we need for this block nr? */
441 while (block_nr) {
442 levels_needed += 1;
443 block_nr >>= BM_RTREE_LEVEL_SHIFT;
444 }
445
446 /* Make sure the rtree has enough levels */
447 for (i = zone->levels; i < levels_needed; i++) {
448 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
449 &zone->nodes);
450 if (!node)
451 return -ENOMEM;
452
453 node->data[0] = (unsigned long)zone->rtree;
454 zone->rtree = node;
455 zone->levels += 1;
456 }
457
458 /* Allocate new block */
459 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
460 if (!block)
461 return -ENOMEM;
462
463 /* Now walk the rtree to insert the block */
464 node = zone->rtree;
465 dst = &zone->rtree;
466 block_nr = zone->blocks;
467 for (i = zone->levels; i > 0; i--) {
468 int index;
469
470 if (!node) {
471 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
472 &zone->nodes);
473 if (!node)
474 return -ENOMEM;
475 *dst = node;
476 }
477
478 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
479 index &= BM_RTREE_LEVEL_MASK;
480 dst = (struct rtree_node **)&((*dst)->data[index]);
481 node = *dst;
482 }
483
484 zone->blocks += 1;
485 *dst = block;
486
487 return 0;
488}
489
490static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
491 int clear_nosave_free);
492
493/**
494 * create_zone_bm_rtree - Create a radix tree for one zone.
495 *
496 * Allocated the mem_zone_bm_rtree structure and initializes it.
497 * This function also allocated and builds the radix tree for the
498 * zone.
499 */
500static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
501 int safe_needed,
502 struct chain_allocator *ca,
503 unsigned long start,
504 unsigned long end)
505{
506 struct mem_zone_bm_rtree *zone;
507 unsigned int i, nr_blocks;
508 unsigned long pages;
509
510 pages = end - start;
511 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
512 if (!zone)
513 return NULL;
514
515 INIT_LIST_HEAD(&zone->nodes);
516 INIT_LIST_HEAD(&zone->leaves);
517 zone->start_pfn = start;
518 zone->end_pfn = end;
519 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
520
521 for (i = 0; i < nr_blocks; i++) {
522 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
523 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
524 return NULL;
525 }
526 }
527
528 return zone;
529}
530
531/**
532 * free_zone_bm_rtree - Free the memory of the radix tree.
533 *
534 * Free all node pages of the radix tree. The mem_zone_bm_rtree
535 * structure itself is not freed here nor are the rtree_node
536 * structs.
537 */
538static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
539 int clear_nosave_free)
540{
541 struct rtree_node *node;
542
543 list_for_each_entry(node, &zone->nodes, list)
544 free_image_page(node->data, clear_nosave_free);
545
546 list_for_each_entry(node, &zone->leaves, list)
547 free_image_page(node->data, clear_nosave_free);
548}
549
550static void memory_bm_position_reset(struct memory_bitmap *bm)
551{
552 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
553 list);
554 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
555 struct rtree_node, list);
556 bm->cur.node_pfn = 0;
557 bm->cur.node_bit = 0;
558}
559
560static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
561
562struct mem_extent {
563 struct list_head hook;
564 unsigned long start;
565 unsigned long end;
566};
567
568/**
569 * free_mem_extents - Free a list of memory extents.
570 * @list: List of extents to free.
571 */
572static void free_mem_extents(struct list_head *list)
573{
574 struct mem_extent *ext, *aux;
575
576 list_for_each_entry_safe(ext, aux, list, hook) {
577 list_del(&ext->hook);
578 kfree(ext);
579 }
580}
581
582/**
583 * create_mem_extents - Create a list of memory extents.
584 * @list: List to put the extents into.
585 * @gfp_mask: Mask to use for memory allocations.
586 *
587 * The extents represent contiguous ranges of PFNs.
588 */
589static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
590{
591 struct zone *zone;
592
593 INIT_LIST_HEAD(list);
594
595 for_each_populated_zone(zone) {
596 unsigned long zone_start, zone_end;
597 struct mem_extent *ext, *cur, *aux;
598
599 zone_start = zone->zone_start_pfn;
600 zone_end = zone_end_pfn(zone);
601
602 list_for_each_entry(ext, list, hook)
603 if (zone_start <= ext->end)
604 break;
605
606 if (&ext->hook == list || zone_end < ext->start) {
607 /* New extent is necessary */
608 struct mem_extent *new_ext;
609
610 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
611 if (!new_ext) {
612 free_mem_extents(list);
613 return -ENOMEM;
614 }
615 new_ext->start = zone_start;
616 new_ext->end = zone_end;
617 list_add_tail(&new_ext->hook, &ext->hook);
618 continue;
619 }
620
621 /* Merge this zone's range of PFNs with the existing one */
622 if (zone_start < ext->start)
623 ext->start = zone_start;
624 if (zone_end > ext->end)
625 ext->end = zone_end;
626
627 /* More merging may be possible */
628 cur = ext;
629 list_for_each_entry_safe_continue(cur, aux, list, hook) {
630 if (zone_end < cur->start)
631 break;
632 if (zone_end < cur->end)
633 ext->end = cur->end;
634 list_del(&cur->hook);
635 kfree(cur);
636 }
637 }
638
639 return 0;
640}
641
642/**
643 * memory_bm_create - Allocate memory for a memory bitmap.
644 */
645static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
646 int safe_needed)
647{
648 struct chain_allocator ca;
649 struct list_head mem_extents;
650 struct mem_extent *ext;
651 int error;
652
653 chain_init(&ca, gfp_mask, safe_needed);
654 INIT_LIST_HEAD(&bm->zones);
655
656 error = create_mem_extents(&mem_extents, gfp_mask);
657 if (error)
658 return error;
659
660 list_for_each_entry(ext, &mem_extents, hook) {
661 struct mem_zone_bm_rtree *zone;
662
663 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
664 ext->start, ext->end);
665 if (!zone) {
666 error = -ENOMEM;
667 goto Error;
668 }
669 list_add_tail(&zone->list, &bm->zones);
670 }
671
672 bm->p_list = ca.chain;
673 memory_bm_position_reset(bm);
674 Exit:
675 free_mem_extents(&mem_extents);
676 return error;
677
678 Error:
679 bm->p_list = ca.chain;
680 memory_bm_free(bm, PG_UNSAFE_CLEAR);
681 goto Exit;
682}
683
684/**
685 * memory_bm_free - Free memory occupied by the memory bitmap.
686 * @bm: Memory bitmap.
687 */
688static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
689{
690 struct mem_zone_bm_rtree *zone;
691
692 list_for_each_entry(zone, &bm->zones, list)
693 free_zone_bm_rtree(zone, clear_nosave_free);
694
695 free_list_of_pages(bm->p_list, clear_nosave_free);
696
697 INIT_LIST_HEAD(&bm->zones);
698}
699
700/**
701 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
702 *
703 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
704 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
705 *
706 * Walk the radix tree to find the page containing the bit that represents @pfn
707 * and return the position of the bit in @addr and @bit_nr.
708 */
709static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
710 void **addr, unsigned int *bit_nr)
711{
712 struct mem_zone_bm_rtree *curr, *zone;
713 struct rtree_node *node;
714 int i, block_nr;
715
716 zone = bm->cur.zone;
717
718 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
719 goto zone_found;
720
721 zone = NULL;
722
723 /* Find the right zone */
724 list_for_each_entry(curr, &bm->zones, list) {
725 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
726 zone = curr;
727 break;
728 }
729 }
730
731 if (!zone)
732 return -EFAULT;
733
734zone_found:
735 /*
736 * We have found the zone. Now walk the radix tree to find the leaf node
737 * for our PFN.
738 */
739 node = bm->cur.node;
740 if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
741 goto node_found;
742
743 node = zone->rtree;
744 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
745
746 for (i = zone->levels; i > 0; i--) {
747 int index;
748
749 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
750 index &= BM_RTREE_LEVEL_MASK;
751 BUG_ON(node->data[index] == 0);
752 node = (struct rtree_node *)node->data[index];
753 }
754
755node_found:
756 /* Update last position */
757 bm->cur.zone = zone;
758 bm->cur.node = node;
759 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
760
761 /* Set return values */
762 *addr = node->data;
763 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
764
765 return 0;
766}
767
768static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
769{
770 void *addr;
771 unsigned int bit;
772 int error;
773
774 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
775 BUG_ON(error);
776 set_bit(bit, addr);
777}
778
779static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
780{
781 void *addr;
782 unsigned int bit;
783 int error;
784
785 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
786 if (!error)
787 set_bit(bit, addr);
788
789 return error;
790}
791
792static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
793{
794 void *addr;
795 unsigned int bit;
796 int error;
797
798 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
799 BUG_ON(error);
800 clear_bit(bit, addr);
801}
802
803static void memory_bm_clear_current(struct memory_bitmap *bm)
804{
805 int bit;
806
807 bit = max(bm->cur.node_bit - 1, 0);
808 clear_bit(bit, bm->cur.node->data);
809}
810
811static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
812{
813 void *addr;
814 unsigned int bit;
815 int error;
816
817 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
818 BUG_ON(error);
819 return test_bit(bit, addr);
820}
821
822static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
823{
824 void *addr;
825 unsigned int bit;
826
827 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
828}
829
830/*
831 * rtree_next_node - Jump to the next leaf node.
832 *
833 * Set the position to the beginning of the next node in the
834 * memory bitmap. This is either the next node in the current
835 * zone's radix tree or the first node in the radix tree of the
836 * next zone.
837 *
838 * Return true if there is a next node, false otherwise.
839 */
840static bool rtree_next_node(struct memory_bitmap *bm)
841{
842 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
843 bm->cur.node = list_entry(bm->cur.node->list.next,
844 struct rtree_node, list);
845 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
846 bm->cur.node_bit = 0;
847 touch_softlockup_watchdog();
848 return true;
849 }
850
851 /* No more nodes, goto next zone */
852 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
853 bm->cur.zone = list_entry(bm->cur.zone->list.next,
854 struct mem_zone_bm_rtree, list);
855 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
856 struct rtree_node, list);
857 bm->cur.node_pfn = 0;
858 bm->cur.node_bit = 0;
859 return true;
860 }
861
862 /* No more zones */
863 return false;
864}
865
866/**
867 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
868 * @bm: Memory bitmap.
869 *
870 * Starting from the last returned position this function searches for the next
871 * set bit in @bm and returns the PFN represented by it. If no more bits are
872 * set, BM_END_OF_MAP is returned.
873 *
874 * It is required to run memory_bm_position_reset() before the first call to
875 * this function for the given memory bitmap.
876 */
877static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
878{
879 unsigned long bits, pfn, pages;
880 int bit;
881
882 do {
883 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
884 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
885 bit = find_next_bit(bm->cur.node->data, bits,
886 bm->cur.node_bit);
887 if (bit < bits) {
888 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
889 bm->cur.node_bit = bit + 1;
890 return pfn;
891 }
892 } while (rtree_next_node(bm));
893
894 return BM_END_OF_MAP;
895}
896
897/*
898 * This structure represents a range of page frames the contents of which
899 * should not be saved during hibernation.
900 */
901struct nosave_region {
902 struct list_head list;
903 unsigned long start_pfn;
904 unsigned long end_pfn;
905};
906
907static LIST_HEAD(nosave_regions);
908
909static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
910{
911 struct rtree_node *node;
912
913 list_for_each_entry(node, &zone->nodes, list)
914 recycle_safe_page(node->data);
915
916 list_for_each_entry(node, &zone->leaves, list)
917 recycle_safe_page(node->data);
918}
919
920static void memory_bm_recycle(struct memory_bitmap *bm)
921{
922 struct mem_zone_bm_rtree *zone;
923 struct linked_page *p_list;
924
925 list_for_each_entry(zone, &bm->zones, list)
926 recycle_zone_bm_rtree(zone);
927
928 p_list = bm->p_list;
929 while (p_list) {
930 struct linked_page *lp = p_list;
931
932 p_list = lp->next;
933 recycle_safe_page(lp);
934 }
935}
936
937/**
938 * register_nosave_region - Register a region of unsaveable memory.
939 *
940 * Register a range of page frames the contents of which should not be saved
941 * during hibernation (to be used in the early initialization code).
942 */
943void __init __register_nosave_region(unsigned long start_pfn,
944 unsigned long end_pfn, int use_kmalloc)
945{
946 struct nosave_region *region;
947
948 if (start_pfn >= end_pfn)
949 return;
950
951 if (!list_empty(&nosave_regions)) {
952 /* Try to extend the previous region (they should be sorted) */
953 region = list_entry(nosave_regions.prev,
954 struct nosave_region, list);
955 if (region->end_pfn == start_pfn) {
956 region->end_pfn = end_pfn;
957 goto Report;
958 }
959 }
960 if (use_kmalloc) {
961 /* During init, this shouldn't fail */
962 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
963 BUG_ON(!region);
964 } else {
965 /* This allocation cannot fail */
966 region = memblock_virt_alloc(sizeof(struct nosave_region), 0);
967 }
968 region->start_pfn = start_pfn;
969 region->end_pfn = end_pfn;
970 list_add_tail(®ion->list, &nosave_regions);
971 Report:
972 pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
973 (unsigned long long) start_pfn << PAGE_SHIFT,
974 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
975}
976
977/*
978 * Set bits in this map correspond to the page frames the contents of which
979 * should not be saved during the suspend.
980 */
981static struct memory_bitmap *forbidden_pages_map;
982
983/* Set bits in this map correspond to free page frames. */
984static struct memory_bitmap *free_pages_map;
985
986/*
987 * Each page frame allocated for creating the image is marked by setting the
988 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
989 */
990
991void swsusp_set_page_free(struct page *page)
992{
993 if (free_pages_map)
994 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
995}
996
997static int swsusp_page_is_free(struct page *page)
998{
999 return free_pages_map ?
1000 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1001}
1002
1003void swsusp_unset_page_free(struct page *page)
1004{
1005 if (free_pages_map)
1006 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1007}
1008
1009static void swsusp_set_page_forbidden(struct page *page)
1010{
1011 if (forbidden_pages_map)
1012 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1013}
1014
1015int swsusp_page_is_forbidden(struct page *page)
1016{
1017 return forbidden_pages_map ?
1018 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1019}
1020
1021static void swsusp_unset_page_forbidden(struct page *page)
1022{
1023 if (forbidden_pages_map)
1024 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1025}
1026
1027/**
1028 * mark_nosave_pages - Mark pages that should not be saved.
1029 * @bm: Memory bitmap.
1030 *
1031 * Set the bits in @bm that correspond to the page frames the contents of which
1032 * should not be saved.
1033 */
1034static void mark_nosave_pages(struct memory_bitmap *bm)
1035{
1036 struct nosave_region *region;
1037
1038 if (list_empty(&nosave_regions))
1039 return;
1040
1041 list_for_each_entry(region, &nosave_regions, list) {
1042 unsigned long pfn;
1043
1044 pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1045 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1046 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1047 - 1);
1048
1049 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1050 if (pfn_valid(pfn)) {
1051 /*
1052 * It is safe to ignore the result of
1053 * mem_bm_set_bit_check() here, since we won't
1054 * touch the PFNs for which the error is
1055 * returned anyway.
1056 */
1057 mem_bm_set_bit_check(bm, pfn);
1058 }
1059 }
1060}
1061
1062/**
1063 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1064 *
1065 * Create bitmaps needed for marking page frames that should not be saved and
1066 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1067 * only modified if everything goes well, because we don't want the bits to be
1068 * touched before both bitmaps are set up.
1069 */
1070int create_basic_memory_bitmaps(void)
1071{
1072 struct memory_bitmap *bm1, *bm2;
1073 int error = 0;
1074
1075 if (forbidden_pages_map && free_pages_map)
1076 return 0;
1077 else
1078 BUG_ON(forbidden_pages_map || free_pages_map);
1079
1080 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1081 if (!bm1)
1082 return -ENOMEM;
1083
1084 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1085 if (error)
1086 goto Free_first_object;
1087
1088 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1089 if (!bm2)
1090 goto Free_first_bitmap;
1091
1092 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1093 if (error)
1094 goto Free_second_object;
1095
1096 forbidden_pages_map = bm1;
1097 free_pages_map = bm2;
1098 mark_nosave_pages(forbidden_pages_map);
1099
1100 pr_debug("Basic memory bitmaps created\n");
1101
1102 return 0;
1103
1104 Free_second_object:
1105 kfree(bm2);
1106 Free_first_bitmap:
1107 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1108 Free_first_object:
1109 kfree(bm1);
1110 return -ENOMEM;
1111}
1112
1113/**
1114 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1115 *
1116 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1117 * auxiliary pointers are necessary so that the bitmaps themselves are not
1118 * referred to while they are being freed.
1119 */
1120void free_basic_memory_bitmaps(void)
1121{
1122 struct memory_bitmap *bm1, *bm2;
1123
1124 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1125 return;
1126
1127 bm1 = forbidden_pages_map;
1128 bm2 = free_pages_map;
1129 forbidden_pages_map = NULL;
1130 free_pages_map = NULL;
1131 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1132 kfree(bm1);
1133 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1134 kfree(bm2);
1135
1136 pr_debug("Basic memory bitmaps freed\n");
1137}
1138
1139void clear_free_pages(void)
1140{
1141#ifdef CONFIG_PAGE_POISONING_ZERO
1142 struct memory_bitmap *bm = free_pages_map;
1143 unsigned long pfn;
1144
1145 if (WARN_ON(!(free_pages_map)))
1146 return;
1147
1148 memory_bm_position_reset(bm);
1149 pfn = memory_bm_next_pfn(bm);
1150 while (pfn != BM_END_OF_MAP) {
1151 if (pfn_valid(pfn))
1152 clear_highpage(pfn_to_page(pfn));
1153
1154 pfn = memory_bm_next_pfn(bm);
1155 }
1156 memory_bm_position_reset(bm);
1157 pr_info("free pages cleared after restore\n");
1158#endif /* PAGE_POISONING_ZERO */
1159}
1160
1161/**
1162 * snapshot_additional_pages - Estimate the number of extra pages needed.
1163 * @zone: Memory zone to carry out the computation for.
1164 *
1165 * Estimate the number of additional pages needed for setting up a hibernation
1166 * image data structures for @zone (usually, the returned value is greater than
1167 * the exact number).
1168 */
1169unsigned int snapshot_additional_pages(struct zone *zone)
1170{
1171 unsigned int rtree, nodes;
1172
1173 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1174 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1175 LINKED_PAGE_DATA_SIZE);
1176 while (nodes > 1) {
1177 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1178 rtree += nodes;
1179 }
1180
1181 return 2 * rtree;
1182}
1183
1184#ifdef CONFIG_HIGHMEM
1185/**
1186 * count_free_highmem_pages - Compute the total number of free highmem pages.
1187 *
1188 * The returned number is system-wide.
1189 */
1190static unsigned int count_free_highmem_pages(void)
1191{
1192 struct zone *zone;
1193 unsigned int cnt = 0;
1194
1195 for_each_populated_zone(zone)
1196 if (is_highmem(zone))
1197 cnt += zone_page_state(zone, NR_FREE_PAGES);
1198
1199 return cnt;
1200}
1201
1202/**
1203 * saveable_highmem_page - Check if a highmem page is saveable.
1204 *
1205 * Determine whether a highmem page should be included in a hibernation image.
1206 *
1207 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1208 * and it isn't part of a free chunk of pages.
1209 */
1210static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1211{
1212 struct page *page;
1213
1214 if (!pfn_valid(pfn))
1215 return NULL;
1216
1217 page = pfn_to_page(pfn);
1218 if (page_zone(page) != zone)
1219 return NULL;
1220
1221 BUG_ON(!PageHighMem(page));
1222
1223 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) ||
1224 PageReserved(page))
1225 return NULL;
1226
1227 if (page_is_guard(page))
1228 return NULL;
1229
1230 return page;
1231}
1232
1233/**
1234 * count_highmem_pages - Compute the total number of saveable highmem pages.
1235 */
1236static unsigned int count_highmem_pages(void)
1237{
1238 struct zone *zone;
1239 unsigned int n = 0;
1240
1241 for_each_populated_zone(zone) {
1242 unsigned long pfn, max_zone_pfn;
1243
1244 if (!is_highmem(zone))
1245 continue;
1246
1247 mark_free_pages(zone);
1248 max_zone_pfn = zone_end_pfn(zone);
1249 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1250 if (saveable_highmem_page(zone, pfn))
1251 n++;
1252 }
1253 return n;
1254}
1255#else
1256static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1257{
1258 return NULL;
1259}
1260#endif /* CONFIG_HIGHMEM */
1261
1262/**
1263 * saveable_page - Check if the given page is saveable.
1264 *
1265 * Determine whether a non-highmem page should be included in a hibernation
1266 * image.
1267 *
1268 * We should save the page if it isn't Nosave, and is not in the range
1269 * of pages statically defined as 'unsaveable', and it isn't part of
1270 * a free chunk of pages.
1271 */
1272static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1273{
1274 struct page *page;
1275
1276 if (!pfn_valid(pfn))
1277 return NULL;
1278
1279 page = pfn_to_page(pfn);
1280 if (page_zone(page) != zone)
1281 return NULL;
1282
1283 BUG_ON(PageHighMem(page));
1284
1285 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1286 return NULL;
1287
1288 if (PageReserved(page)
1289 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1290 return NULL;
1291
1292 if (page_is_guard(page))
1293 return NULL;
1294
1295 return page;
1296}
1297
1298/**
1299 * count_data_pages - Compute the total number of saveable non-highmem pages.
1300 */
1301static unsigned int count_data_pages(void)
1302{
1303 struct zone *zone;
1304 unsigned long pfn, max_zone_pfn;
1305 unsigned int n = 0;
1306
1307 for_each_populated_zone(zone) {
1308 if (is_highmem(zone))
1309 continue;
1310
1311 mark_free_pages(zone);
1312 max_zone_pfn = zone_end_pfn(zone);
1313 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1314 if (saveable_page(zone, pfn))
1315 n++;
1316 }
1317 return n;
1318}
1319
1320/*
1321 * This is needed, because copy_page and memcpy are not usable for copying
1322 * task structs.
1323 */
1324static inline void do_copy_page(long *dst, long *src)
1325{
1326 int n;
1327
1328 for (n = PAGE_SIZE / sizeof(long); n; n--)
1329 *dst++ = *src++;
1330}
1331
1332/**
1333 * safe_copy_page - Copy a page in a safe way.
1334 *
1335 * Check if the page we are going to copy is marked as present in the kernel
1336 * page tables (this always is the case if CONFIG_DEBUG_PAGEALLOC is not set
1337 * and in that case kernel_page_present() always returns 'true').
1338 */
1339static void safe_copy_page(void *dst, struct page *s_page)
1340{
1341 if (kernel_page_present(s_page)) {
1342 do_copy_page(dst, page_address(s_page));
1343 } else {
1344 kernel_map_pages(s_page, 1, 1);
1345 do_copy_page(dst, page_address(s_page));
1346 kernel_map_pages(s_page, 1, 0);
1347 }
1348}
1349
1350#ifdef CONFIG_HIGHMEM
1351static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1352{
1353 return is_highmem(zone) ?
1354 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1355}
1356
1357static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1358{
1359 struct page *s_page, *d_page;
1360 void *src, *dst;
1361
1362 s_page = pfn_to_page(src_pfn);
1363 d_page = pfn_to_page(dst_pfn);
1364 if (PageHighMem(s_page)) {
1365 src = kmap_atomic(s_page);
1366 dst = kmap_atomic(d_page);
1367 do_copy_page(dst, src);
1368 kunmap_atomic(dst);
1369 kunmap_atomic(src);
1370 } else {
1371 if (PageHighMem(d_page)) {
1372 /*
1373 * The page pointed to by src may contain some kernel
1374 * data modified by kmap_atomic()
1375 */
1376 safe_copy_page(buffer, s_page);
1377 dst = kmap_atomic(d_page);
1378 copy_page(dst, buffer);
1379 kunmap_atomic(dst);
1380 } else {
1381 safe_copy_page(page_address(d_page), s_page);
1382 }
1383 }
1384}
1385#else
1386#define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1387
1388static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1389{
1390 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1391 pfn_to_page(src_pfn));
1392}
1393#endif /* CONFIG_HIGHMEM */
1394
1395static void copy_data_pages(struct memory_bitmap *copy_bm,
1396 struct memory_bitmap *orig_bm)
1397{
1398 struct zone *zone;
1399 unsigned long pfn;
1400
1401 for_each_populated_zone(zone) {
1402 unsigned long max_zone_pfn;
1403
1404 mark_free_pages(zone);
1405 max_zone_pfn = zone_end_pfn(zone);
1406 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1407 if (page_is_saveable(zone, pfn))
1408 memory_bm_set_bit(orig_bm, pfn);
1409 }
1410 memory_bm_position_reset(orig_bm);
1411 memory_bm_position_reset(copy_bm);
1412 for(;;) {
1413 pfn = memory_bm_next_pfn(orig_bm);
1414 if (unlikely(pfn == BM_END_OF_MAP))
1415 break;
1416 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1417 }
1418}
1419
1420/* Total number of image pages */
1421static unsigned int nr_copy_pages;
1422/* Number of pages needed for saving the original pfns of the image pages */
1423static unsigned int nr_meta_pages;
1424/*
1425 * Numbers of normal and highmem page frames allocated for hibernation image
1426 * before suspending devices.
1427 */
1428static unsigned int alloc_normal, alloc_highmem;
1429/*
1430 * Memory bitmap used for marking saveable pages (during hibernation) or
1431 * hibernation image pages (during restore)
1432 */
1433static struct memory_bitmap orig_bm;
1434/*
1435 * Memory bitmap used during hibernation for marking allocated page frames that
1436 * will contain copies of saveable pages. During restore it is initially used
1437 * for marking hibernation image pages, but then the set bits from it are
1438 * duplicated in @orig_bm and it is released. On highmem systems it is next
1439 * used for marking "safe" highmem pages, but it has to be reinitialized for
1440 * this purpose.
1441 */
1442static struct memory_bitmap copy_bm;
1443
1444/**
1445 * swsusp_free - Free pages allocated for hibernation image.
1446 *
1447 * Image pages are alocated before snapshot creation, so they need to be
1448 * released after resume.
1449 */
1450void swsusp_free(void)
1451{
1452 unsigned long fb_pfn, fr_pfn;
1453
1454 if (!forbidden_pages_map || !free_pages_map)
1455 goto out;
1456
1457 memory_bm_position_reset(forbidden_pages_map);
1458 memory_bm_position_reset(free_pages_map);
1459
1460loop:
1461 fr_pfn = memory_bm_next_pfn(free_pages_map);
1462 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1463
1464 /*
1465 * Find the next bit set in both bitmaps. This is guaranteed to
1466 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1467 */
1468 do {
1469 if (fb_pfn < fr_pfn)
1470 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1471 if (fr_pfn < fb_pfn)
1472 fr_pfn = memory_bm_next_pfn(free_pages_map);
1473 } while (fb_pfn != fr_pfn);
1474
1475 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1476 struct page *page = pfn_to_page(fr_pfn);
1477
1478 memory_bm_clear_current(forbidden_pages_map);
1479 memory_bm_clear_current(free_pages_map);
1480 hibernate_restore_unprotect_page(page_address(page));
1481 __free_page(page);
1482 goto loop;
1483 }
1484
1485out:
1486 nr_copy_pages = 0;
1487 nr_meta_pages = 0;
1488 restore_pblist = NULL;
1489 buffer = NULL;
1490 alloc_normal = 0;
1491 alloc_highmem = 0;
1492 hibernate_restore_protection_end();
1493}
1494
1495/* Helper functions used for the shrinking of memory. */
1496
1497#define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1498
1499/**
1500 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1501 * @nr_pages: Number of page frames to allocate.
1502 * @mask: GFP flags to use for the allocation.
1503 *
1504 * Return value: Number of page frames actually allocated
1505 */
1506static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1507{
1508 unsigned long nr_alloc = 0;
1509
1510 while (nr_pages > 0) {
1511 struct page *page;
1512
1513 page = alloc_image_page(mask);
1514 if (!page)
1515 break;
1516 memory_bm_set_bit(©_bm, page_to_pfn(page));
1517 if (PageHighMem(page))
1518 alloc_highmem++;
1519 else
1520 alloc_normal++;
1521 nr_pages--;
1522 nr_alloc++;
1523 }
1524
1525 return nr_alloc;
1526}
1527
1528static unsigned long preallocate_image_memory(unsigned long nr_pages,
1529 unsigned long avail_normal)
1530{
1531 unsigned long alloc;
1532
1533 if (avail_normal <= alloc_normal)
1534 return 0;
1535
1536 alloc = avail_normal - alloc_normal;
1537 if (nr_pages < alloc)
1538 alloc = nr_pages;
1539
1540 return preallocate_image_pages(alloc, GFP_IMAGE);
1541}
1542
1543#ifdef CONFIG_HIGHMEM
1544static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1545{
1546 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1547}
1548
1549/**
1550 * __fraction - Compute (an approximation of) x * (multiplier / base).
1551 */
1552static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1553{
1554 x *= multiplier;
1555 do_div(x, base);
1556 return (unsigned long)x;
1557}
1558
1559static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1560 unsigned long highmem,
1561 unsigned long total)
1562{
1563 unsigned long alloc = __fraction(nr_pages, highmem, total);
1564
1565 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1566}
1567#else /* CONFIG_HIGHMEM */
1568static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1569{
1570 return 0;
1571}
1572
1573static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1574 unsigned long highmem,
1575 unsigned long total)
1576{
1577 return 0;
1578}
1579#endif /* CONFIG_HIGHMEM */
1580
1581/**
1582 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1583 */
1584static unsigned long free_unnecessary_pages(void)
1585{
1586 unsigned long save, to_free_normal, to_free_highmem, free;
1587
1588 save = count_data_pages();
1589 if (alloc_normal >= save) {
1590 to_free_normal = alloc_normal - save;
1591 save = 0;
1592 } else {
1593 to_free_normal = 0;
1594 save -= alloc_normal;
1595 }
1596 save += count_highmem_pages();
1597 if (alloc_highmem >= save) {
1598 to_free_highmem = alloc_highmem - save;
1599 } else {
1600 to_free_highmem = 0;
1601 save -= alloc_highmem;
1602 if (to_free_normal > save)
1603 to_free_normal -= save;
1604 else
1605 to_free_normal = 0;
1606 }
1607 free = to_free_normal + to_free_highmem;
1608
1609 memory_bm_position_reset(©_bm);
1610
1611 while (to_free_normal > 0 || to_free_highmem > 0) {
1612 unsigned long pfn = memory_bm_next_pfn(©_bm);
1613 struct page *page = pfn_to_page(pfn);
1614
1615 if (PageHighMem(page)) {
1616 if (!to_free_highmem)
1617 continue;
1618 to_free_highmem--;
1619 alloc_highmem--;
1620 } else {
1621 if (!to_free_normal)
1622 continue;
1623 to_free_normal--;
1624 alloc_normal--;
1625 }
1626 memory_bm_clear_bit(©_bm, pfn);
1627 swsusp_unset_page_forbidden(page);
1628 swsusp_unset_page_free(page);
1629 __free_page(page);
1630 }
1631
1632 return free;
1633}
1634
1635/**
1636 * minimum_image_size - Estimate the minimum acceptable size of an image.
1637 * @saveable: Number of saveable pages in the system.
1638 *
1639 * We want to avoid attempting to free too much memory too hard, so estimate the
1640 * minimum acceptable size of a hibernation image to use as the lower limit for
1641 * preallocating memory.
1642 *
1643 * We assume that the minimum image size should be proportional to
1644 *
1645 * [number of saveable pages] - [number of pages that can be freed in theory]
1646 *
1647 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1648 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1649 */
1650static unsigned long minimum_image_size(unsigned long saveable)
1651{
1652 unsigned long size;
1653
1654 size = global_node_page_state(NR_SLAB_RECLAIMABLE)
1655 + global_node_page_state(NR_ACTIVE_ANON)
1656 + global_node_page_state(NR_INACTIVE_ANON)
1657 + global_node_page_state(NR_ACTIVE_FILE)
1658 + global_node_page_state(NR_INACTIVE_FILE);
1659
1660 return saveable <= size ? 0 : saveable - size;
1661}
1662
1663/**
1664 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1665 *
1666 * To create a hibernation image it is necessary to make a copy of every page
1667 * frame in use. We also need a number of page frames to be free during
1668 * hibernation for allocations made while saving the image and for device
1669 * drivers, in case they need to allocate memory from their hibernation
1670 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1671 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1672 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1673 * total number of available page frames and allocate at least
1674 *
1675 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1676 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1677 *
1678 * of them, which corresponds to the maximum size of a hibernation image.
1679 *
1680 * If image_size is set below the number following from the above formula,
1681 * the preallocation of memory is continued until the total number of saveable
1682 * pages in the system is below the requested image size or the minimum
1683 * acceptable image size returned by minimum_image_size(), whichever is greater.
1684 */
1685int hibernate_preallocate_memory(void)
1686{
1687 struct zone *zone;
1688 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1689 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1690 ktime_t start, stop;
1691 int error;
1692
1693 pr_info("Preallocating image memory... ");
1694 start = ktime_get();
1695
1696 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1697 if (error)
1698 goto err_out;
1699
1700 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1701 if (error)
1702 goto err_out;
1703
1704 alloc_normal = 0;
1705 alloc_highmem = 0;
1706
1707 /* Count the number of saveable data pages. */
1708 save_highmem = count_highmem_pages();
1709 saveable = count_data_pages();
1710
1711 /*
1712 * Compute the total number of page frames we can use (count) and the
1713 * number of pages needed for image metadata (size).
1714 */
1715 count = saveable;
1716 saveable += save_highmem;
1717 highmem = save_highmem;
1718 size = 0;
1719 for_each_populated_zone(zone) {
1720 size += snapshot_additional_pages(zone);
1721 if (is_highmem(zone))
1722 highmem += zone_page_state(zone, NR_FREE_PAGES);
1723 else
1724 count += zone_page_state(zone, NR_FREE_PAGES);
1725 }
1726 avail_normal = count;
1727 count += highmem;
1728 count -= totalreserve_pages;
1729
1730 /* Add number of pages required for page keys (s390 only). */
1731 size += page_key_additional_pages(saveable);
1732
1733 /* Compute the maximum number of saveable pages to leave in memory. */
1734 max_size = (count - (size + PAGES_FOR_IO)) / 2
1735 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1736 /* Compute the desired number of image pages specified by image_size. */
1737 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1738 if (size > max_size)
1739 size = max_size;
1740 /*
1741 * If the desired number of image pages is at least as large as the
1742 * current number of saveable pages in memory, allocate page frames for
1743 * the image and we're done.
1744 */
1745 if (size >= saveable) {
1746 pages = preallocate_image_highmem(save_highmem);
1747 pages += preallocate_image_memory(saveable - pages, avail_normal);
1748 goto out;
1749 }
1750
1751 /* Estimate the minimum size of the image. */
1752 pages = minimum_image_size(saveable);
1753 /*
1754 * To avoid excessive pressure on the normal zone, leave room in it to
1755 * accommodate an image of the minimum size (unless it's already too
1756 * small, in which case don't preallocate pages from it at all).
1757 */
1758 if (avail_normal > pages)
1759 avail_normal -= pages;
1760 else
1761 avail_normal = 0;
1762 if (size < pages)
1763 size = min_t(unsigned long, pages, max_size);
1764
1765 /*
1766 * Let the memory management subsystem know that we're going to need a
1767 * large number of page frames to allocate and make it free some memory.
1768 * NOTE: If this is not done, performance will be hurt badly in some
1769 * test cases.
1770 */
1771 shrink_all_memory(saveable - size);
1772
1773 /*
1774 * The number of saveable pages in memory was too high, so apply some
1775 * pressure to decrease it. First, make room for the largest possible
1776 * image and fail if that doesn't work. Next, try to decrease the size
1777 * of the image as much as indicated by 'size' using allocations from
1778 * highmem and non-highmem zones separately.
1779 */
1780 pages_highmem = preallocate_image_highmem(highmem / 2);
1781 alloc = count - max_size;
1782 if (alloc > pages_highmem)
1783 alloc -= pages_highmem;
1784 else
1785 alloc = 0;
1786 pages = preallocate_image_memory(alloc, avail_normal);
1787 if (pages < alloc) {
1788 /* We have exhausted non-highmem pages, try highmem. */
1789 alloc -= pages;
1790 pages += pages_highmem;
1791 pages_highmem = preallocate_image_highmem(alloc);
1792 if (pages_highmem < alloc)
1793 goto err_out;
1794 pages += pages_highmem;
1795 /*
1796 * size is the desired number of saveable pages to leave in
1797 * memory, so try to preallocate (all memory - size) pages.
1798 */
1799 alloc = (count - pages) - size;
1800 pages += preallocate_image_highmem(alloc);
1801 } else {
1802 /*
1803 * There are approximately max_size saveable pages at this point
1804 * and we want to reduce this number down to size.
1805 */
1806 alloc = max_size - size;
1807 size = preallocate_highmem_fraction(alloc, highmem, count);
1808 pages_highmem += size;
1809 alloc -= size;
1810 size = preallocate_image_memory(alloc, avail_normal);
1811 pages_highmem += preallocate_image_highmem(alloc - size);
1812 pages += pages_highmem + size;
1813 }
1814
1815 /*
1816 * We only need as many page frames for the image as there are saveable
1817 * pages in memory, but we have allocated more. Release the excessive
1818 * ones now.
1819 */
1820 pages -= free_unnecessary_pages();
1821
1822 out:
1823 stop = ktime_get();
1824 pr_cont("done (allocated %lu pages)\n", pages);
1825 swsusp_show_speed(start, stop, pages, "Allocated");
1826
1827 return 0;
1828
1829 err_out:
1830 pr_cont("\n");
1831 swsusp_free();
1832 return -ENOMEM;
1833}
1834
1835#ifdef CONFIG_HIGHMEM
1836/**
1837 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1838 *
1839 * Compute the number of non-highmem pages that will be necessary for creating
1840 * copies of highmem pages.
1841 */
1842static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1843{
1844 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1845
1846 if (free_highmem >= nr_highmem)
1847 nr_highmem = 0;
1848 else
1849 nr_highmem -= free_highmem;
1850
1851 return nr_highmem;
1852}
1853#else
1854static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1855#endif /* CONFIG_HIGHMEM */
1856
1857/**
1858 * enough_free_mem - Check if there is enough free memory for the image.
1859 */
1860static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1861{
1862 struct zone *zone;
1863 unsigned int free = alloc_normal;
1864
1865 for_each_populated_zone(zone)
1866 if (!is_highmem(zone))
1867 free += zone_page_state(zone, NR_FREE_PAGES);
1868
1869 nr_pages += count_pages_for_highmem(nr_highmem);
1870 pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
1871 nr_pages, PAGES_FOR_IO, free);
1872
1873 return free > nr_pages + PAGES_FOR_IO;
1874}
1875
1876#ifdef CONFIG_HIGHMEM
1877/**
1878 * get_highmem_buffer - Allocate a buffer for highmem pages.
1879 *
1880 * If there are some highmem pages in the hibernation image, we may need a
1881 * buffer to copy them and/or load their data.
1882 */
1883static inline int get_highmem_buffer(int safe_needed)
1884{
1885 buffer = get_image_page(GFP_ATOMIC, safe_needed);
1886 return buffer ? 0 : -ENOMEM;
1887}
1888
1889/**
1890 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1891 *
1892 * Try to allocate as many pages as needed, but if the number of free highmem
1893 * pages is less than that, allocate them all.
1894 */
1895static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1896 unsigned int nr_highmem)
1897{
1898 unsigned int to_alloc = count_free_highmem_pages();
1899
1900 if (to_alloc > nr_highmem)
1901 to_alloc = nr_highmem;
1902
1903 nr_highmem -= to_alloc;
1904 while (to_alloc-- > 0) {
1905 struct page *page;
1906
1907 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1908 memory_bm_set_bit(bm, page_to_pfn(page));
1909 }
1910 return nr_highmem;
1911}
1912#else
1913static inline int get_highmem_buffer(int safe_needed) { return 0; }
1914
1915static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1916 unsigned int n) { return 0; }
1917#endif /* CONFIG_HIGHMEM */
1918
1919/**
1920 * swsusp_alloc - Allocate memory for hibernation image.
1921 *
1922 * We first try to allocate as many highmem pages as there are
1923 * saveable highmem pages in the system. If that fails, we allocate
1924 * non-highmem pages for the copies of the remaining highmem ones.
1925 *
1926 * In this approach it is likely that the copies of highmem pages will
1927 * also be located in the high memory, because of the way in which
1928 * copy_data_pages() works.
1929 */
1930static int swsusp_alloc(struct memory_bitmap *copy_bm,
1931 unsigned int nr_pages, unsigned int nr_highmem)
1932{
1933 if (nr_highmem > 0) {
1934 if (get_highmem_buffer(PG_ANY))
1935 goto err_out;
1936 if (nr_highmem > alloc_highmem) {
1937 nr_highmem -= alloc_highmem;
1938 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1939 }
1940 }
1941 if (nr_pages > alloc_normal) {
1942 nr_pages -= alloc_normal;
1943 while (nr_pages-- > 0) {
1944 struct page *page;
1945
1946 page = alloc_image_page(GFP_ATOMIC);
1947 if (!page)
1948 goto err_out;
1949 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1950 }
1951 }
1952
1953 return 0;
1954
1955 err_out:
1956 swsusp_free();
1957 return -ENOMEM;
1958}
1959
1960asmlinkage __visible int swsusp_save(void)
1961{
1962 unsigned int nr_pages, nr_highmem;
1963
1964 pr_info("Creating hibernation image:\n");
1965
1966 drain_local_pages(NULL);
1967 nr_pages = count_data_pages();
1968 nr_highmem = count_highmem_pages();
1969 pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
1970
1971 if (!enough_free_mem(nr_pages, nr_highmem)) {
1972 pr_err("Not enough free memory\n");
1973 return -ENOMEM;
1974 }
1975
1976 if (swsusp_alloc(©_bm, nr_pages, nr_highmem)) {
1977 pr_err("Memory allocation failed\n");
1978 return -ENOMEM;
1979 }
1980
1981 /*
1982 * During allocating of suspend pagedir, new cold pages may appear.
1983 * Kill them.
1984 */
1985 drain_local_pages(NULL);
1986 copy_data_pages(©_bm, &orig_bm);
1987
1988 /*
1989 * End of critical section. From now on, we can write to memory,
1990 * but we should not touch disk. This specially means we must _not_
1991 * touch swap space! Except we must write out our image of course.
1992 */
1993
1994 nr_pages += nr_highmem;
1995 nr_copy_pages = nr_pages;
1996 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1997
1998 pr_info("Hibernation image created (%d pages copied)\n", nr_pages);
1999
2000 return 0;
2001}
2002
2003#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2004static int init_header_complete(struct swsusp_info *info)
2005{
2006 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2007 info->version_code = LINUX_VERSION_CODE;
2008 return 0;
2009}
2010
2011static char *check_image_kernel(struct swsusp_info *info)
2012{
2013 if (info->version_code != LINUX_VERSION_CODE)
2014 return "kernel version";
2015 if (strcmp(info->uts.sysname,init_utsname()->sysname))
2016 return "system type";
2017 if (strcmp(info->uts.release,init_utsname()->release))
2018 return "kernel release";
2019 if (strcmp(info->uts.version,init_utsname()->version))
2020 return "version";
2021 if (strcmp(info->uts.machine,init_utsname()->machine))
2022 return "machine";
2023 return NULL;
2024}
2025#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2026
2027unsigned long snapshot_get_image_size(void)
2028{
2029 return nr_copy_pages + nr_meta_pages + 1;
2030}
2031
2032static int init_header(struct swsusp_info *info)
2033{
2034 memset(info, 0, sizeof(struct swsusp_info));
2035 info->num_physpages = get_num_physpages();
2036 info->image_pages = nr_copy_pages;
2037 info->pages = snapshot_get_image_size();
2038 info->size = info->pages;
2039 info->size <<= PAGE_SHIFT;
2040 return init_header_complete(info);
2041}
2042
2043/**
2044 * pack_pfns - Prepare PFNs for saving.
2045 * @bm: Memory bitmap.
2046 * @buf: Memory buffer to store the PFNs in.
2047 *
2048 * PFNs corresponding to set bits in @bm are stored in the area of memory
2049 * pointed to by @buf (1 page at a time).
2050 */
2051static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2052{
2053 int j;
2054
2055 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2056 buf[j] = memory_bm_next_pfn(bm);
2057 if (unlikely(buf[j] == BM_END_OF_MAP))
2058 break;
2059 /* Save page key for data page (s390 only). */
2060 page_key_read(buf + j);
2061 }
2062}
2063
2064/**
2065 * snapshot_read_next - Get the address to read the next image page from.
2066 * @handle: Snapshot handle to be used for the reading.
2067 *
2068 * On the first call, @handle should point to a zeroed snapshot_handle
2069 * structure. The structure gets populated then and a pointer to it should be
2070 * passed to this function every next time.
2071 *
2072 * On success, the function returns a positive number. Then, the caller
2073 * is allowed to read up to the returned number of bytes from the memory
2074 * location computed by the data_of() macro.
2075 *
2076 * The function returns 0 to indicate the end of the data stream condition,
2077 * and negative numbers are returned on errors. If that happens, the structure
2078 * pointed to by @handle is not updated and should not be used any more.
2079 */
2080int snapshot_read_next(struct snapshot_handle *handle)
2081{
2082 if (handle->cur > nr_meta_pages + nr_copy_pages)
2083 return 0;
2084
2085 if (!buffer) {
2086 /* This makes the buffer be freed by swsusp_free() */
2087 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2088 if (!buffer)
2089 return -ENOMEM;
2090 }
2091 if (!handle->cur) {
2092 int error;
2093
2094 error = init_header((struct swsusp_info *)buffer);
2095 if (error)
2096 return error;
2097 handle->buffer = buffer;
2098 memory_bm_position_reset(&orig_bm);
2099 memory_bm_position_reset(©_bm);
2100 } else if (handle->cur <= nr_meta_pages) {
2101 clear_page(buffer);
2102 pack_pfns(buffer, &orig_bm);
2103 } else {
2104 struct page *page;
2105
2106 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2107 if (PageHighMem(page)) {
2108 /*
2109 * Highmem pages are copied to the buffer,
2110 * because we can't return with a kmapped
2111 * highmem page (we may not be called again).
2112 */
2113 void *kaddr;
2114
2115 kaddr = kmap_atomic(page);
2116 copy_page(buffer, kaddr);
2117 kunmap_atomic(kaddr);
2118 handle->buffer = buffer;
2119 } else {
2120 handle->buffer = page_address(page);
2121 }
2122 }
2123 handle->cur++;
2124 return PAGE_SIZE;
2125}
2126
2127static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2128 struct memory_bitmap *src)
2129{
2130 unsigned long pfn;
2131
2132 memory_bm_position_reset(src);
2133 pfn = memory_bm_next_pfn(src);
2134 while (pfn != BM_END_OF_MAP) {
2135 memory_bm_set_bit(dst, pfn);
2136 pfn = memory_bm_next_pfn(src);
2137 }
2138}
2139
2140/**
2141 * mark_unsafe_pages - Mark pages that were used before hibernation.
2142 *
2143 * Mark the pages that cannot be used for storing the image during restoration,
2144 * because they conflict with the pages that had been used before hibernation.
2145 */
2146static void mark_unsafe_pages(struct memory_bitmap *bm)
2147{
2148 unsigned long pfn;
2149
2150 /* Clear the "free"/"unsafe" bit for all PFNs */
2151 memory_bm_position_reset(free_pages_map);
2152 pfn = memory_bm_next_pfn(free_pages_map);
2153 while (pfn != BM_END_OF_MAP) {
2154 memory_bm_clear_current(free_pages_map);
2155 pfn = memory_bm_next_pfn(free_pages_map);
2156 }
2157
2158 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2159 duplicate_memory_bitmap(free_pages_map, bm);
2160
2161 allocated_unsafe_pages = 0;
2162}
2163
2164static int check_header(struct swsusp_info *info)
2165{
2166 char *reason;
2167
2168 reason = check_image_kernel(info);
2169 if (!reason && info->num_physpages != get_num_physpages())
2170 reason = "memory size";
2171 if (reason) {
2172 pr_err("Image mismatch: %s\n", reason);
2173 return -EPERM;
2174 }
2175 return 0;
2176}
2177
2178/**
2179 * load header - Check the image header and copy the data from it.
2180 */
2181static int load_header(struct swsusp_info *info)
2182{
2183 int error;
2184
2185 restore_pblist = NULL;
2186 error = check_header(info);
2187 if (!error) {
2188 nr_copy_pages = info->image_pages;
2189 nr_meta_pages = info->pages - info->image_pages - 1;
2190 }
2191 return error;
2192}
2193
2194/**
2195 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2196 * @bm: Memory bitmap.
2197 * @buf: Area of memory containing the PFNs.
2198 *
2199 * For each element of the array pointed to by @buf (1 page at a time), set the
2200 * corresponding bit in @bm.
2201 */
2202static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2203{
2204 int j;
2205
2206 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2207 if (unlikely(buf[j] == BM_END_OF_MAP))
2208 break;
2209
2210 /* Extract and buffer page key for data page (s390 only). */
2211 page_key_memorize(buf + j);
2212
2213 if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2214 memory_bm_set_bit(bm, buf[j]);
2215 else
2216 return -EFAULT;
2217 }
2218
2219 return 0;
2220}
2221
2222#ifdef CONFIG_HIGHMEM
2223/*
2224 * struct highmem_pbe is used for creating the list of highmem pages that
2225 * should be restored atomically during the resume from disk, because the page
2226 * frames they have occupied before the suspend are in use.
2227 */
2228struct highmem_pbe {
2229 struct page *copy_page; /* data is here now */
2230 struct page *orig_page; /* data was here before the suspend */
2231 struct highmem_pbe *next;
2232};
2233
2234/*
2235 * List of highmem PBEs needed for restoring the highmem pages that were
2236 * allocated before the suspend and included in the suspend image, but have
2237 * also been allocated by the "resume" kernel, so their contents cannot be
2238 * written directly to their "original" page frames.
2239 */
2240static struct highmem_pbe *highmem_pblist;
2241
2242/**
2243 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2244 * @bm: Memory bitmap.
2245 *
2246 * The bits in @bm that correspond to image pages are assumed to be set.
2247 */
2248static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2249{
2250 unsigned long pfn;
2251 unsigned int cnt = 0;
2252
2253 memory_bm_position_reset(bm);
2254 pfn = memory_bm_next_pfn(bm);
2255 while (pfn != BM_END_OF_MAP) {
2256 if (PageHighMem(pfn_to_page(pfn)))
2257 cnt++;
2258
2259 pfn = memory_bm_next_pfn(bm);
2260 }
2261 return cnt;
2262}
2263
2264static unsigned int safe_highmem_pages;
2265
2266static struct memory_bitmap *safe_highmem_bm;
2267
2268/**
2269 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2270 * @bm: Pointer to an uninitialized memory bitmap structure.
2271 * @nr_highmem_p: Pointer to the number of highmem image pages.
2272 *
2273 * Try to allocate as many highmem pages as there are highmem image pages
2274 * (@nr_highmem_p points to the variable containing the number of highmem image
2275 * pages). The pages that are "safe" (ie. will not be overwritten when the
2276 * hibernation image is restored entirely) have the corresponding bits set in
2277 * @bm (it must be unitialized).
2278 *
2279 * NOTE: This function should not be called if there are no highmem image pages.
2280 */
2281static int prepare_highmem_image(struct memory_bitmap *bm,
2282 unsigned int *nr_highmem_p)
2283{
2284 unsigned int to_alloc;
2285
2286 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2287 return -ENOMEM;
2288
2289 if (get_highmem_buffer(PG_SAFE))
2290 return -ENOMEM;
2291
2292 to_alloc = count_free_highmem_pages();
2293 if (to_alloc > *nr_highmem_p)
2294 to_alloc = *nr_highmem_p;
2295 else
2296 *nr_highmem_p = to_alloc;
2297
2298 safe_highmem_pages = 0;
2299 while (to_alloc-- > 0) {
2300 struct page *page;
2301
2302 page = alloc_page(__GFP_HIGHMEM);
2303 if (!swsusp_page_is_free(page)) {
2304 /* The page is "safe", set its bit the bitmap */
2305 memory_bm_set_bit(bm, page_to_pfn(page));
2306 safe_highmem_pages++;
2307 }
2308 /* Mark the page as allocated */
2309 swsusp_set_page_forbidden(page);
2310 swsusp_set_page_free(page);
2311 }
2312 memory_bm_position_reset(bm);
2313 safe_highmem_bm = bm;
2314 return 0;
2315}
2316
2317static struct page *last_highmem_page;
2318
2319/**
2320 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2321 *
2322 * For a given highmem image page get a buffer that suspend_write_next() should
2323 * return to its caller to write to.
2324 *
2325 * If the page is to be saved to its "original" page frame or a copy of
2326 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2327 * the copy of the page is to be made in normal memory, so the address of
2328 * the copy is returned.
2329 *
2330 * If @buffer is returned, the caller of suspend_write_next() will write
2331 * the page's contents to @buffer, so they will have to be copied to the
2332 * right location on the next call to suspend_write_next() and it is done
2333 * with the help of copy_last_highmem_page(). For this purpose, if
2334 * @buffer is returned, @last_highmem_page is set to the page to which
2335 * the data will have to be copied from @buffer.
2336 */
2337static void *get_highmem_page_buffer(struct page *page,
2338 struct chain_allocator *ca)
2339{
2340 struct highmem_pbe *pbe;
2341 void *kaddr;
2342
2343 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2344 /*
2345 * We have allocated the "original" page frame and we can
2346 * use it directly to store the loaded page.
2347 */
2348 last_highmem_page = page;
2349 return buffer;
2350 }
2351 /*
2352 * The "original" page frame has not been allocated and we have to
2353 * use a "safe" page frame to store the loaded page.
2354 */
2355 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2356 if (!pbe) {
2357 swsusp_free();
2358 return ERR_PTR(-ENOMEM);
2359 }
2360 pbe->orig_page = page;
2361 if (safe_highmem_pages > 0) {
2362 struct page *tmp;
2363
2364 /* Copy of the page will be stored in high memory */
2365 kaddr = buffer;
2366 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2367 safe_highmem_pages--;
2368 last_highmem_page = tmp;
2369 pbe->copy_page = tmp;
2370 } else {
2371 /* Copy of the page will be stored in normal memory */
2372 kaddr = safe_pages_list;
2373 safe_pages_list = safe_pages_list->next;
2374 pbe->copy_page = virt_to_page(kaddr);
2375 }
2376 pbe->next = highmem_pblist;
2377 highmem_pblist = pbe;
2378 return kaddr;
2379}
2380
2381/**
2382 * copy_last_highmem_page - Copy most the most recent highmem image page.
2383 *
2384 * Copy the contents of a highmem image from @buffer, where the caller of
2385 * snapshot_write_next() has stored them, to the right location represented by
2386 * @last_highmem_page .
2387 */
2388static void copy_last_highmem_page(void)
2389{
2390 if (last_highmem_page) {
2391 void *dst;
2392
2393 dst = kmap_atomic(last_highmem_page);
2394 copy_page(dst, buffer);
2395 kunmap_atomic(dst);
2396 last_highmem_page = NULL;
2397 }
2398}
2399
2400static inline int last_highmem_page_copied(void)
2401{
2402 return !last_highmem_page;
2403}
2404
2405static inline void free_highmem_data(void)
2406{
2407 if (safe_highmem_bm)
2408 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2409
2410 if (buffer)
2411 free_image_page(buffer, PG_UNSAFE_CLEAR);
2412}
2413#else
2414static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2415
2416static inline int prepare_highmem_image(struct memory_bitmap *bm,
2417 unsigned int *nr_highmem_p) { return 0; }
2418
2419static inline void *get_highmem_page_buffer(struct page *page,
2420 struct chain_allocator *ca)
2421{
2422 return ERR_PTR(-EINVAL);
2423}
2424
2425static inline void copy_last_highmem_page(void) {}
2426static inline int last_highmem_page_copied(void) { return 1; }
2427static inline void free_highmem_data(void) {}
2428#endif /* CONFIG_HIGHMEM */
2429
2430#define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2431
2432/**
2433 * prepare_image - Make room for loading hibernation image.
2434 * @new_bm: Unitialized memory bitmap structure.
2435 * @bm: Memory bitmap with unsafe pages marked.
2436 *
2437 * Use @bm to mark the pages that will be overwritten in the process of
2438 * restoring the system memory state from the suspend image ("unsafe" pages)
2439 * and allocate memory for the image.
2440 *
2441 * The idea is to allocate a new memory bitmap first and then allocate
2442 * as many pages as needed for image data, but without specifying what those
2443 * pages will be used for just yet. Instead, we mark them all as allocated and
2444 * create a lists of "safe" pages to be used later. On systems with high
2445 * memory a list of "safe" highmem pages is created too.
2446 */
2447static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2448{
2449 unsigned int nr_pages, nr_highmem;
2450 struct linked_page *lp;
2451 int error;
2452
2453 /* If there is no highmem, the buffer will not be necessary */
2454 free_image_page(buffer, PG_UNSAFE_CLEAR);
2455 buffer = NULL;
2456
2457 nr_highmem = count_highmem_image_pages(bm);
2458 mark_unsafe_pages(bm);
2459
2460 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2461 if (error)
2462 goto Free;
2463
2464 duplicate_memory_bitmap(new_bm, bm);
2465 memory_bm_free(bm, PG_UNSAFE_KEEP);
2466 if (nr_highmem > 0) {
2467 error = prepare_highmem_image(bm, &nr_highmem);
2468 if (error)
2469 goto Free;
2470 }
2471 /*
2472 * Reserve some safe pages for potential later use.
2473 *
2474 * NOTE: This way we make sure there will be enough safe pages for the
2475 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2476 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2477 *
2478 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2479 */
2480 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2481 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2482 while (nr_pages > 0) {
2483 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2484 if (!lp) {
2485 error = -ENOMEM;
2486 goto Free;
2487 }
2488 lp->next = safe_pages_list;
2489 safe_pages_list = lp;
2490 nr_pages--;
2491 }
2492 /* Preallocate memory for the image */
2493 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2494 while (nr_pages > 0) {
2495 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2496 if (!lp) {
2497 error = -ENOMEM;
2498 goto Free;
2499 }
2500 if (!swsusp_page_is_free(virt_to_page(lp))) {
2501 /* The page is "safe", add it to the list */
2502 lp->next = safe_pages_list;
2503 safe_pages_list = lp;
2504 }
2505 /* Mark the page as allocated */
2506 swsusp_set_page_forbidden(virt_to_page(lp));
2507 swsusp_set_page_free(virt_to_page(lp));
2508 nr_pages--;
2509 }
2510 return 0;
2511
2512 Free:
2513 swsusp_free();
2514 return error;
2515}
2516
2517/**
2518 * get_buffer - Get the address to store the next image data page.
2519 *
2520 * Get the address that snapshot_write_next() should return to its caller to
2521 * write to.
2522 */
2523static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2524{
2525 struct pbe *pbe;
2526 struct page *page;
2527 unsigned long pfn = memory_bm_next_pfn(bm);
2528
2529 if (pfn == BM_END_OF_MAP)
2530 return ERR_PTR(-EFAULT);
2531
2532 page = pfn_to_page(pfn);
2533 if (PageHighMem(page))
2534 return get_highmem_page_buffer(page, ca);
2535
2536 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2537 /*
2538 * We have allocated the "original" page frame and we can
2539 * use it directly to store the loaded page.
2540 */
2541 return page_address(page);
2542
2543 /*
2544 * The "original" page frame has not been allocated and we have to
2545 * use a "safe" page frame to store the loaded page.
2546 */
2547 pbe = chain_alloc(ca, sizeof(struct pbe));
2548 if (!pbe) {
2549 swsusp_free();
2550 return ERR_PTR(-ENOMEM);
2551 }
2552 pbe->orig_address = page_address(page);
2553 pbe->address = safe_pages_list;
2554 safe_pages_list = safe_pages_list->next;
2555 pbe->next = restore_pblist;
2556 restore_pblist = pbe;
2557 return pbe->address;
2558}
2559
2560/**
2561 * snapshot_write_next - Get the address to store the next image page.
2562 * @handle: Snapshot handle structure to guide the writing.
2563 *
2564 * On the first call, @handle should point to a zeroed snapshot_handle
2565 * structure. The structure gets populated then and a pointer to it should be
2566 * passed to this function every next time.
2567 *
2568 * On success, the function returns a positive number. Then, the caller
2569 * is allowed to write up to the returned number of bytes to the memory
2570 * location computed by the data_of() macro.
2571 *
2572 * The function returns 0 to indicate the "end of file" condition. Negative
2573 * numbers are returned on errors, in which cases the structure pointed to by
2574 * @handle is not updated and should not be used any more.
2575 */
2576int snapshot_write_next(struct snapshot_handle *handle)
2577{
2578 static struct chain_allocator ca;
2579 int error = 0;
2580
2581 /* Check if we have already loaded the entire image */
2582 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2583 return 0;
2584
2585 handle->sync_read = 1;
2586
2587 if (!handle->cur) {
2588 if (!buffer)
2589 /* This makes the buffer be freed by swsusp_free() */
2590 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2591
2592 if (!buffer)
2593 return -ENOMEM;
2594
2595 handle->buffer = buffer;
2596 } else if (handle->cur == 1) {
2597 error = load_header(buffer);
2598 if (error)
2599 return error;
2600
2601 safe_pages_list = NULL;
2602
2603 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2604 if (error)
2605 return error;
2606
2607 /* Allocate buffer for page keys. */
2608 error = page_key_alloc(nr_copy_pages);
2609 if (error)
2610 return error;
2611
2612 hibernate_restore_protection_begin();
2613 } else if (handle->cur <= nr_meta_pages + 1) {
2614 error = unpack_orig_pfns(buffer, ©_bm);
2615 if (error)
2616 return error;
2617
2618 if (handle->cur == nr_meta_pages + 1) {
2619 error = prepare_image(&orig_bm, ©_bm);
2620 if (error)
2621 return error;
2622
2623 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2624 memory_bm_position_reset(&orig_bm);
2625 restore_pblist = NULL;
2626 handle->buffer = get_buffer(&orig_bm, &ca);
2627 handle->sync_read = 0;
2628 if (IS_ERR(handle->buffer))
2629 return PTR_ERR(handle->buffer);
2630 }
2631 } else {
2632 copy_last_highmem_page();
2633 /* Restore page key for data page (s390 only). */
2634 page_key_write(handle->buffer);
2635 hibernate_restore_protect_page(handle->buffer);
2636 handle->buffer = get_buffer(&orig_bm, &ca);
2637 if (IS_ERR(handle->buffer))
2638 return PTR_ERR(handle->buffer);
2639 if (handle->buffer != buffer)
2640 handle->sync_read = 0;
2641 }
2642 handle->cur++;
2643 return PAGE_SIZE;
2644}
2645
2646/**
2647 * snapshot_write_finalize - Complete the loading of a hibernation image.
2648 *
2649 * Must be called after the last call to snapshot_write_next() in case the last
2650 * page in the image happens to be a highmem page and its contents should be
2651 * stored in highmem. Additionally, it recycles bitmap memory that's not
2652 * necessary any more.
2653 */
2654void snapshot_write_finalize(struct snapshot_handle *handle)
2655{
2656 copy_last_highmem_page();
2657 /* Restore page key for data page (s390 only). */
2658 page_key_write(handle->buffer);
2659 page_key_free();
2660 hibernate_restore_protect_page(handle->buffer);
2661 /* Do that only if we have loaded the image entirely */
2662 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2663 memory_bm_recycle(&orig_bm);
2664 free_highmem_data();
2665 }
2666}
2667
2668int snapshot_image_loaded(struct snapshot_handle *handle)
2669{
2670 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2671 handle->cur <= nr_meta_pages + nr_copy_pages);
2672}
2673
2674#ifdef CONFIG_HIGHMEM
2675/* Assumes that @buf is ready and points to a "safe" page */
2676static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2677 void *buf)
2678{
2679 void *kaddr1, *kaddr2;
2680
2681 kaddr1 = kmap_atomic(p1);
2682 kaddr2 = kmap_atomic(p2);
2683 copy_page(buf, kaddr1);
2684 copy_page(kaddr1, kaddr2);
2685 copy_page(kaddr2, buf);
2686 kunmap_atomic(kaddr2);
2687 kunmap_atomic(kaddr1);
2688}
2689
2690/**
2691 * restore_highmem - Put highmem image pages into their original locations.
2692 *
2693 * For each highmem page that was in use before hibernation and is included in
2694 * the image, and also has been allocated by the "restore" kernel, swap its
2695 * current contents with the previous (ie. "before hibernation") ones.
2696 *
2697 * If the restore eventually fails, we can call this function once again and
2698 * restore the highmem state as seen by the restore kernel.
2699 */
2700int restore_highmem(void)
2701{
2702 struct highmem_pbe *pbe = highmem_pblist;
2703 void *buf;
2704
2705 if (!pbe)
2706 return 0;
2707
2708 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2709 if (!buf)
2710 return -ENOMEM;
2711
2712 while (pbe) {
2713 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2714 pbe = pbe->next;
2715 }
2716 free_image_page(buf, PG_UNSAFE_CLEAR);
2717 return 0;
2718}
2719#endif /* CONFIG_HIGHMEM */
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/power/snapshot.c
4 *
5 * This file provides system snapshot/restore functionality for swsusp.
6 *
7 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
8 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
9 */
10
11#define pr_fmt(fmt) "PM: hibernation: " fmt
12
13#include <linux/version.h>
14#include <linux/module.h>
15#include <linux/mm.h>
16#include <linux/suspend.h>
17#include <linux/delay.h>
18#include <linux/bitops.h>
19#include <linux/spinlock.h>
20#include <linux/kernel.h>
21#include <linux/pm.h>
22#include <linux/device.h>
23#include <linux/init.h>
24#include <linux/memblock.h>
25#include <linux/nmi.h>
26#include <linux/syscalls.h>
27#include <linux/console.h>
28#include <linux/highmem.h>
29#include <linux/list.h>
30#include <linux/slab.h>
31#include <linux/compiler.h>
32#include <linux/ktime.h>
33#include <linux/set_memory.h>
34
35#include <linux/uaccess.h>
36#include <asm/mmu_context.h>
37#include <asm/tlbflush.h>
38#include <asm/io.h>
39
40#include "power.h"
41
42#if defined(CONFIG_STRICT_KERNEL_RWX) && defined(CONFIG_ARCH_HAS_SET_MEMORY)
43static bool hibernate_restore_protection;
44static bool hibernate_restore_protection_active;
45
46void enable_restore_image_protection(void)
47{
48 hibernate_restore_protection = true;
49}
50
51static inline void hibernate_restore_protection_begin(void)
52{
53 hibernate_restore_protection_active = hibernate_restore_protection;
54}
55
56static inline void hibernate_restore_protection_end(void)
57{
58 hibernate_restore_protection_active = false;
59}
60
61static inline void hibernate_restore_protect_page(void *page_address)
62{
63 if (hibernate_restore_protection_active)
64 set_memory_ro((unsigned long)page_address, 1);
65}
66
67static inline void hibernate_restore_unprotect_page(void *page_address)
68{
69 if (hibernate_restore_protection_active)
70 set_memory_rw((unsigned long)page_address, 1);
71}
72#else
73static inline void hibernate_restore_protection_begin(void) {}
74static inline void hibernate_restore_protection_end(void) {}
75static inline void hibernate_restore_protect_page(void *page_address) {}
76static inline void hibernate_restore_unprotect_page(void *page_address) {}
77#endif /* CONFIG_STRICT_KERNEL_RWX && CONFIG_ARCH_HAS_SET_MEMORY */
78
79
80/*
81 * The calls to set_direct_map_*() should not fail because remapping a page
82 * here means that we only update protection bits in an existing PTE.
83 * It is still worth to have a warning here if something changes and this
84 * will no longer be the case.
85 */
86static inline void hibernate_map_page(struct page *page)
87{
88 if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
89 int ret = set_direct_map_default_noflush(page);
90
91 if (ret)
92 pr_warn_once("Failed to remap page\n");
93 } else {
94 debug_pagealloc_map_pages(page, 1);
95 }
96}
97
98static inline void hibernate_unmap_page(struct page *page)
99{
100 if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
101 unsigned long addr = (unsigned long)page_address(page);
102 int ret = set_direct_map_invalid_noflush(page);
103
104 if (ret)
105 pr_warn_once("Failed to remap page\n");
106
107 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
108 } else {
109 debug_pagealloc_unmap_pages(page, 1);
110 }
111}
112
113static int swsusp_page_is_free(struct page *);
114static void swsusp_set_page_forbidden(struct page *);
115static void swsusp_unset_page_forbidden(struct page *);
116
117/*
118 * Number of bytes to reserve for memory allocations made by device drivers
119 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
120 * cause image creation to fail (tunable via /sys/power/reserved_size).
121 */
122unsigned long reserved_size;
123
124void __init hibernate_reserved_size_init(void)
125{
126 reserved_size = SPARE_PAGES * PAGE_SIZE;
127}
128
129/*
130 * Preferred image size in bytes (tunable via /sys/power/image_size).
131 * When it is set to N, swsusp will do its best to ensure the image
132 * size will not exceed N bytes, but if that is impossible, it will
133 * try to create the smallest image possible.
134 */
135unsigned long image_size;
136
137void __init hibernate_image_size_init(void)
138{
139 image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
140}
141
142/*
143 * List of PBEs needed for restoring the pages that were allocated before
144 * the suspend and included in the suspend image, but have also been
145 * allocated by the "resume" kernel, so their contents cannot be written
146 * directly to their "original" page frames.
147 */
148struct pbe *restore_pblist;
149
150/* struct linked_page is used to build chains of pages */
151
152#define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
153
154struct linked_page {
155 struct linked_page *next;
156 char data[LINKED_PAGE_DATA_SIZE];
157} __packed;
158
159/*
160 * List of "safe" pages (ie. pages that were not used by the image kernel
161 * before hibernation) that may be used as temporary storage for image kernel
162 * memory contents.
163 */
164static struct linked_page *safe_pages_list;
165
166/* Pointer to an auxiliary buffer (1 page) */
167static void *buffer;
168
169#define PG_ANY 0
170#define PG_SAFE 1
171#define PG_UNSAFE_CLEAR 1
172#define PG_UNSAFE_KEEP 0
173
174static unsigned int allocated_unsafe_pages;
175
176/**
177 * get_image_page - Allocate a page for a hibernation image.
178 * @gfp_mask: GFP mask for the allocation.
179 * @safe_needed: Get pages that were not used before hibernation (restore only)
180 *
181 * During image restoration, for storing the PBE list and the image data, we can
182 * only use memory pages that do not conflict with the pages used before
183 * hibernation. The "unsafe" pages have PageNosaveFree set and we count them
184 * using allocated_unsafe_pages.
185 *
186 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
187 * swsusp_free() can release it.
188 */
189static void *get_image_page(gfp_t gfp_mask, int safe_needed)
190{
191 void *res;
192
193 res = (void *)get_zeroed_page(gfp_mask);
194 if (safe_needed)
195 while (res && swsusp_page_is_free(virt_to_page(res))) {
196 /* The page is unsafe, mark it for swsusp_free() */
197 swsusp_set_page_forbidden(virt_to_page(res));
198 allocated_unsafe_pages++;
199 res = (void *)get_zeroed_page(gfp_mask);
200 }
201 if (res) {
202 swsusp_set_page_forbidden(virt_to_page(res));
203 swsusp_set_page_free(virt_to_page(res));
204 }
205 return res;
206}
207
208static void *__get_safe_page(gfp_t gfp_mask)
209{
210 if (safe_pages_list) {
211 void *ret = safe_pages_list;
212
213 safe_pages_list = safe_pages_list->next;
214 memset(ret, 0, PAGE_SIZE);
215 return ret;
216 }
217 return get_image_page(gfp_mask, PG_SAFE);
218}
219
220unsigned long get_safe_page(gfp_t gfp_mask)
221{
222 return (unsigned long)__get_safe_page(gfp_mask);
223}
224
225static struct page *alloc_image_page(gfp_t gfp_mask)
226{
227 struct page *page;
228
229 page = alloc_page(gfp_mask);
230 if (page) {
231 swsusp_set_page_forbidden(page);
232 swsusp_set_page_free(page);
233 }
234 return page;
235}
236
237static void recycle_safe_page(void *page_address)
238{
239 struct linked_page *lp = page_address;
240
241 lp->next = safe_pages_list;
242 safe_pages_list = lp;
243}
244
245/**
246 * free_image_page - Free a page allocated for hibernation image.
247 * @addr: Address of the page to free.
248 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
249 *
250 * The page to free should have been allocated by get_image_page() (page flags
251 * set by it are affected).
252 */
253static inline void free_image_page(void *addr, int clear_nosave_free)
254{
255 struct page *page;
256
257 BUG_ON(!virt_addr_valid(addr));
258
259 page = virt_to_page(addr);
260
261 swsusp_unset_page_forbidden(page);
262 if (clear_nosave_free)
263 swsusp_unset_page_free(page);
264
265 __free_page(page);
266}
267
268static inline void free_list_of_pages(struct linked_page *list,
269 int clear_page_nosave)
270{
271 while (list) {
272 struct linked_page *lp = list->next;
273
274 free_image_page(list, clear_page_nosave);
275 list = lp;
276 }
277}
278
279/*
280 * struct chain_allocator is used for allocating small objects out of
281 * a linked list of pages called 'the chain'.
282 *
283 * The chain grows each time when there is no room for a new object in
284 * the current page. The allocated objects cannot be freed individually.
285 * It is only possible to free them all at once, by freeing the entire
286 * chain.
287 *
288 * NOTE: The chain allocator may be inefficient if the allocated objects
289 * are not much smaller than PAGE_SIZE.
290 */
291struct chain_allocator {
292 struct linked_page *chain; /* the chain */
293 unsigned int used_space; /* total size of objects allocated out
294 of the current page */
295 gfp_t gfp_mask; /* mask for allocating pages */
296 int safe_needed; /* if set, only "safe" pages are allocated */
297};
298
299static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
300 int safe_needed)
301{
302 ca->chain = NULL;
303 ca->used_space = LINKED_PAGE_DATA_SIZE;
304 ca->gfp_mask = gfp_mask;
305 ca->safe_needed = safe_needed;
306}
307
308static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
309{
310 void *ret;
311
312 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
313 struct linked_page *lp;
314
315 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
316 get_image_page(ca->gfp_mask, PG_ANY);
317 if (!lp)
318 return NULL;
319
320 lp->next = ca->chain;
321 ca->chain = lp;
322 ca->used_space = 0;
323 }
324 ret = ca->chain->data + ca->used_space;
325 ca->used_space += size;
326 return ret;
327}
328
329/**
330 * Data types related to memory bitmaps.
331 *
332 * Memory bitmap is a structure consisting of many linked lists of
333 * objects. The main list's elements are of type struct zone_bitmap
334 * and each of them corresponds to one zone. For each zone bitmap
335 * object there is a list of objects of type struct bm_block that
336 * represent each blocks of bitmap in which information is stored.
337 *
338 * struct memory_bitmap contains a pointer to the main list of zone
339 * bitmap objects, a struct bm_position used for browsing the bitmap,
340 * and a pointer to the list of pages used for allocating all of the
341 * zone bitmap objects and bitmap block objects.
342 *
343 * NOTE: It has to be possible to lay out the bitmap in memory
344 * using only allocations of order 0. Additionally, the bitmap is
345 * designed to work with arbitrary number of zones (this is over the
346 * top for now, but let's avoid making unnecessary assumptions ;-).
347 *
348 * struct zone_bitmap contains a pointer to a list of bitmap block
349 * objects and a pointer to the bitmap block object that has been
350 * most recently used for setting bits. Additionally, it contains the
351 * PFNs that correspond to the start and end of the represented zone.
352 *
353 * struct bm_block contains a pointer to the memory page in which
354 * information is stored (in the form of a block of bitmap)
355 * It also contains the pfns that correspond to the start and end of
356 * the represented memory area.
357 *
358 * The memory bitmap is organized as a radix tree to guarantee fast random
359 * access to the bits. There is one radix tree for each zone (as returned
360 * from create_mem_extents).
361 *
362 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
363 * two linked lists for the nodes of the tree, one for the inner nodes and
364 * one for the leave nodes. The linked leave nodes are used for fast linear
365 * access of the memory bitmap.
366 *
367 * The struct rtree_node represents one node of the radix tree.
368 */
369
370#define BM_END_OF_MAP (~0UL)
371
372#define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
373#define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
374#define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
375
376/*
377 * struct rtree_node is a wrapper struct to link the nodes
378 * of the rtree together for easy linear iteration over
379 * bits and easy freeing
380 */
381struct rtree_node {
382 struct list_head list;
383 unsigned long *data;
384};
385
386/*
387 * struct mem_zone_bm_rtree represents a bitmap used for one
388 * populated memory zone.
389 */
390struct mem_zone_bm_rtree {
391 struct list_head list; /* Link Zones together */
392 struct list_head nodes; /* Radix Tree inner nodes */
393 struct list_head leaves; /* Radix Tree leaves */
394 unsigned long start_pfn; /* Zone start page frame */
395 unsigned long end_pfn; /* Zone end page frame + 1 */
396 struct rtree_node *rtree; /* Radix Tree Root */
397 int levels; /* Number of Radix Tree Levels */
398 unsigned int blocks; /* Number of Bitmap Blocks */
399};
400
401/* strcut bm_position is used for browsing memory bitmaps */
402
403struct bm_position {
404 struct mem_zone_bm_rtree *zone;
405 struct rtree_node *node;
406 unsigned long node_pfn;
407 int node_bit;
408};
409
410struct memory_bitmap {
411 struct list_head zones;
412 struct linked_page *p_list; /* list of pages used to store zone
413 bitmap objects and bitmap block
414 objects */
415 struct bm_position cur; /* most recently used bit position */
416};
417
418/* Functions that operate on memory bitmaps */
419
420#define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
421#if BITS_PER_LONG == 32
422#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
423#else
424#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
425#endif
426#define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
427
428/**
429 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
430 *
431 * This function is used to allocate inner nodes as well as the
432 * leave nodes of the radix tree. It also adds the node to the
433 * corresponding linked list passed in by the *list parameter.
434 */
435static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
436 struct chain_allocator *ca,
437 struct list_head *list)
438{
439 struct rtree_node *node;
440
441 node = chain_alloc(ca, sizeof(struct rtree_node));
442 if (!node)
443 return NULL;
444
445 node->data = get_image_page(gfp_mask, safe_needed);
446 if (!node->data)
447 return NULL;
448
449 list_add_tail(&node->list, list);
450
451 return node;
452}
453
454/**
455 * add_rtree_block - Add a new leave node to the radix tree.
456 *
457 * The leave nodes need to be allocated in order to keep the leaves
458 * linked list in order. This is guaranteed by the zone->blocks
459 * counter.
460 */
461static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
462 int safe_needed, struct chain_allocator *ca)
463{
464 struct rtree_node *node, *block, **dst;
465 unsigned int levels_needed, block_nr;
466 int i;
467
468 block_nr = zone->blocks;
469 levels_needed = 0;
470
471 /* How many levels do we need for this block nr? */
472 while (block_nr) {
473 levels_needed += 1;
474 block_nr >>= BM_RTREE_LEVEL_SHIFT;
475 }
476
477 /* Make sure the rtree has enough levels */
478 for (i = zone->levels; i < levels_needed; i++) {
479 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
480 &zone->nodes);
481 if (!node)
482 return -ENOMEM;
483
484 node->data[0] = (unsigned long)zone->rtree;
485 zone->rtree = node;
486 zone->levels += 1;
487 }
488
489 /* Allocate new block */
490 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
491 if (!block)
492 return -ENOMEM;
493
494 /* Now walk the rtree to insert the block */
495 node = zone->rtree;
496 dst = &zone->rtree;
497 block_nr = zone->blocks;
498 for (i = zone->levels; i > 0; i--) {
499 int index;
500
501 if (!node) {
502 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
503 &zone->nodes);
504 if (!node)
505 return -ENOMEM;
506 *dst = node;
507 }
508
509 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
510 index &= BM_RTREE_LEVEL_MASK;
511 dst = (struct rtree_node **)&((*dst)->data[index]);
512 node = *dst;
513 }
514
515 zone->blocks += 1;
516 *dst = block;
517
518 return 0;
519}
520
521static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
522 int clear_nosave_free);
523
524/**
525 * create_zone_bm_rtree - Create a radix tree for one zone.
526 *
527 * Allocated the mem_zone_bm_rtree structure and initializes it.
528 * This function also allocated and builds the radix tree for the
529 * zone.
530 */
531static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
532 int safe_needed,
533 struct chain_allocator *ca,
534 unsigned long start,
535 unsigned long end)
536{
537 struct mem_zone_bm_rtree *zone;
538 unsigned int i, nr_blocks;
539 unsigned long pages;
540
541 pages = end - start;
542 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
543 if (!zone)
544 return NULL;
545
546 INIT_LIST_HEAD(&zone->nodes);
547 INIT_LIST_HEAD(&zone->leaves);
548 zone->start_pfn = start;
549 zone->end_pfn = end;
550 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
551
552 for (i = 0; i < nr_blocks; i++) {
553 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
554 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
555 return NULL;
556 }
557 }
558
559 return zone;
560}
561
562/**
563 * free_zone_bm_rtree - Free the memory of the radix tree.
564 *
565 * Free all node pages of the radix tree. The mem_zone_bm_rtree
566 * structure itself is not freed here nor are the rtree_node
567 * structs.
568 */
569static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
570 int clear_nosave_free)
571{
572 struct rtree_node *node;
573
574 list_for_each_entry(node, &zone->nodes, list)
575 free_image_page(node->data, clear_nosave_free);
576
577 list_for_each_entry(node, &zone->leaves, list)
578 free_image_page(node->data, clear_nosave_free);
579}
580
581static void memory_bm_position_reset(struct memory_bitmap *bm)
582{
583 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
584 list);
585 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
586 struct rtree_node, list);
587 bm->cur.node_pfn = 0;
588 bm->cur.node_bit = 0;
589}
590
591static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
592
593struct mem_extent {
594 struct list_head hook;
595 unsigned long start;
596 unsigned long end;
597};
598
599/**
600 * free_mem_extents - Free a list of memory extents.
601 * @list: List of extents to free.
602 */
603static void free_mem_extents(struct list_head *list)
604{
605 struct mem_extent *ext, *aux;
606
607 list_for_each_entry_safe(ext, aux, list, hook) {
608 list_del(&ext->hook);
609 kfree(ext);
610 }
611}
612
613/**
614 * create_mem_extents - Create a list of memory extents.
615 * @list: List to put the extents into.
616 * @gfp_mask: Mask to use for memory allocations.
617 *
618 * The extents represent contiguous ranges of PFNs.
619 */
620static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
621{
622 struct zone *zone;
623
624 INIT_LIST_HEAD(list);
625
626 for_each_populated_zone(zone) {
627 unsigned long zone_start, zone_end;
628 struct mem_extent *ext, *cur, *aux;
629
630 zone_start = zone->zone_start_pfn;
631 zone_end = zone_end_pfn(zone);
632
633 list_for_each_entry(ext, list, hook)
634 if (zone_start <= ext->end)
635 break;
636
637 if (&ext->hook == list || zone_end < ext->start) {
638 /* New extent is necessary */
639 struct mem_extent *new_ext;
640
641 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
642 if (!new_ext) {
643 free_mem_extents(list);
644 return -ENOMEM;
645 }
646 new_ext->start = zone_start;
647 new_ext->end = zone_end;
648 list_add_tail(&new_ext->hook, &ext->hook);
649 continue;
650 }
651
652 /* Merge this zone's range of PFNs with the existing one */
653 if (zone_start < ext->start)
654 ext->start = zone_start;
655 if (zone_end > ext->end)
656 ext->end = zone_end;
657
658 /* More merging may be possible */
659 cur = ext;
660 list_for_each_entry_safe_continue(cur, aux, list, hook) {
661 if (zone_end < cur->start)
662 break;
663 if (zone_end < cur->end)
664 ext->end = cur->end;
665 list_del(&cur->hook);
666 kfree(cur);
667 }
668 }
669
670 return 0;
671}
672
673/**
674 * memory_bm_create - Allocate memory for a memory bitmap.
675 */
676static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
677 int safe_needed)
678{
679 struct chain_allocator ca;
680 struct list_head mem_extents;
681 struct mem_extent *ext;
682 int error;
683
684 chain_init(&ca, gfp_mask, safe_needed);
685 INIT_LIST_HEAD(&bm->zones);
686
687 error = create_mem_extents(&mem_extents, gfp_mask);
688 if (error)
689 return error;
690
691 list_for_each_entry(ext, &mem_extents, hook) {
692 struct mem_zone_bm_rtree *zone;
693
694 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
695 ext->start, ext->end);
696 if (!zone) {
697 error = -ENOMEM;
698 goto Error;
699 }
700 list_add_tail(&zone->list, &bm->zones);
701 }
702
703 bm->p_list = ca.chain;
704 memory_bm_position_reset(bm);
705 Exit:
706 free_mem_extents(&mem_extents);
707 return error;
708
709 Error:
710 bm->p_list = ca.chain;
711 memory_bm_free(bm, PG_UNSAFE_CLEAR);
712 goto Exit;
713}
714
715/**
716 * memory_bm_free - Free memory occupied by the memory bitmap.
717 * @bm: Memory bitmap.
718 */
719static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
720{
721 struct mem_zone_bm_rtree *zone;
722
723 list_for_each_entry(zone, &bm->zones, list)
724 free_zone_bm_rtree(zone, clear_nosave_free);
725
726 free_list_of_pages(bm->p_list, clear_nosave_free);
727
728 INIT_LIST_HEAD(&bm->zones);
729}
730
731/**
732 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
733 *
734 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
735 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
736 *
737 * Walk the radix tree to find the page containing the bit that represents @pfn
738 * and return the position of the bit in @addr and @bit_nr.
739 */
740static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
741 void **addr, unsigned int *bit_nr)
742{
743 struct mem_zone_bm_rtree *curr, *zone;
744 struct rtree_node *node;
745 int i, block_nr;
746
747 zone = bm->cur.zone;
748
749 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
750 goto zone_found;
751
752 zone = NULL;
753
754 /* Find the right zone */
755 list_for_each_entry(curr, &bm->zones, list) {
756 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
757 zone = curr;
758 break;
759 }
760 }
761
762 if (!zone)
763 return -EFAULT;
764
765zone_found:
766 /*
767 * We have found the zone. Now walk the radix tree to find the leaf node
768 * for our PFN.
769 */
770
771 /*
772 * If the zone we wish to scan is the current zone and the
773 * pfn falls into the current node then we do not need to walk
774 * the tree.
775 */
776 node = bm->cur.node;
777 if (zone == bm->cur.zone &&
778 ((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
779 goto node_found;
780
781 node = zone->rtree;
782 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
783
784 for (i = zone->levels; i > 0; i--) {
785 int index;
786
787 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
788 index &= BM_RTREE_LEVEL_MASK;
789 BUG_ON(node->data[index] == 0);
790 node = (struct rtree_node *)node->data[index];
791 }
792
793node_found:
794 /* Update last position */
795 bm->cur.zone = zone;
796 bm->cur.node = node;
797 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
798
799 /* Set return values */
800 *addr = node->data;
801 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
802
803 return 0;
804}
805
806static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
807{
808 void *addr;
809 unsigned int bit;
810 int error;
811
812 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
813 BUG_ON(error);
814 set_bit(bit, addr);
815}
816
817static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
818{
819 void *addr;
820 unsigned int bit;
821 int error;
822
823 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
824 if (!error)
825 set_bit(bit, addr);
826
827 return error;
828}
829
830static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
831{
832 void *addr;
833 unsigned int bit;
834 int error;
835
836 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
837 BUG_ON(error);
838 clear_bit(bit, addr);
839}
840
841static void memory_bm_clear_current(struct memory_bitmap *bm)
842{
843 int bit;
844
845 bit = max(bm->cur.node_bit - 1, 0);
846 clear_bit(bit, bm->cur.node->data);
847}
848
849static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
850{
851 void *addr;
852 unsigned int bit;
853 int error;
854
855 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
856 BUG_ON(error);
857 return test_bit(bit, addr);
858}
859
860static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
861{
862 void *addr;
863 unsigned int bit;
864
865 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
866}
867
868/*
869 * rtree_next_node - Jump to the next leaf node.
870 *
871 * Set the position to the beginning of the next node in the
872 * memory bitmap. This is either the next node in the current
873 * zone's radix tree or the first node in the radix tree of the
874 * next zone.
875 *
876 * Return true if there is a next node, false otherwise.
877 */
878static bool rtree_next_node(struct memory_bitmap *bm)
879{
880 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
881 bm->cur.node = list_entry(bm->cur.node->list.next,
882 struct rtree_node, list);
883 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
884 bm->cur.node_bit = 0;
885 touch_softlockup_watchdog();
886 return true;
887 }
888
889 /* No more nodes, goto next zone */
890 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
891 bm->cur.zone = list_entry(bm->cur.zone->list.next,
892 struct mem_zone_bm_rtree, list);
893 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
894 struct rtree_node, list);
895 bm->cur.node_pfn = 0;
896 bm->cur.node_bit = 0;
897 return true;
898 }
899
900 /* No more zones */
901 return false;
902}
903
904/**
905 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
906 * @bm: Memory bitmap.
907 *
908 * Starting from the last returned position this function searches for the next
909 * set bit in @bm and returns the PFN represented by it. If no more bits are
910 * set, BM_END_OF_MAP is returned.
911 *
912 * It is required to run memory_bm_position_reset() before the first call to
913 * this function for the given memory bitmap.
914 */
915static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
916{
917 unsigned long bits, pfn, pages;
918 int bit;
919
920 do {
921 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
922 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
923 bit = find_next_bit(bm->cur.node->data, bits,
924 bm->cur.node_bit);
925 if (bit < bits) {
926 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
927 bm->cur.node_bit = bit + 1;
928 return pfn;
929 }
930 } while (rtree_next_node(bm));
931
932 return BM_END_OF_MAP;
933}
934
935/*
936 * This structure represents a range of page frames the contents of which
937 * should not be saved during hibernation.
938 */
939struct nosave_region {
940 struct list_head list;
941 unsigned long start_pfn;
942 unsigned long end_pfn;
943};
944
945static LIST_HEAD(nosave_regions);
946
947static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
948{
949 struct rtree_node *node;
950
951 list_for_each_entry(node, &zone->nodes, list)
952 recycle_safe_page(node->data);
953
954 list_for_each_entry(node, &zone->leaves, list)
955 recycle_safe_page(node->data);
956}
957
958static void memory_bm_recycle(struct memory_bitmap *bm)
959{
960 struct mem_zone_bm_rtree *zone;
961 struct linked_page *p_list;
962
963 list_for_each_entry(zone, &bm->zones, list)
964 recycle_zone_bm_rtree(zone);
965
966 p_list = bm->p_list;
967 while (p_list) {
968 struct linked_page *lp = p_list;
969
970 p_list = lp->next;
971 recycle_safe_page(lp);
972 }
973}
974
975/**
976 * register_nosave_region - Register a region of unsaveable memory.
977 *
978 * Register a range of page frames the contents of which should not be saved
979 * during hibernation (to be used in the early initialization code).
980 */
981void __init __register_nosave_region(unsigned long start_pfn,
982 unsigned long end_pfn, int use_kmalloc)
983{
984 struct nosave_region *region;
985
986 if (start_pfn >= end_pfn)
987 return;
988
989 if (!list_empty(&nosave_regions)) {
990 /* Try to extend the previous region (they should be sorted) */
991 region = list_entry(nosave_regions.prev,
992 struct nosave_region, list);
993 if (region->end_pfn == start_pfn) {
994 region->end_pfn = end_pfn;
995 goto Report;
996 }
997 }
998 if (use_kmalloc) {
999 /* During init, this shouldn't fail */
1000 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
1001 BUG_ON(!region);
1002 } else {
1003 /* This allocation cannot fail */
1004 region = memblock_alloc(sizeof(struct nosave_region),
1005 SMP_CACHE_BYTES);
1006 if (!region)
1007 panic("%s: Failed to allocate %zu bytes\n", __func__,
1008 sizeof(struct nosave_region));
1009 }
1010 region->start_pfn = start_pfn;
1011 region->end_pfn = end_pfn;
1012 list_add_tail(®ion->list, &nosave_regions);
1013 Report:
1014 pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
1015 (unsigned long long) start_pfn << PAGE_SHIFT,
1016 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1017}
1018
1019/*
1020 * Set bits in this map correspond to the page frames the contents of which
1021 * should not be saved during the suspend.
1022 */
1023static struct memory_bitmap *forbidden_pages_map;
1024
1025/* Set bits in this map correspond to free page frames. */
1026static struct memory_bitmap *free_pages_map;
1027
1028/*
1029 * Each page frame allocated for creating the image is marked by setting the
1030 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
1031 */
1032
1033void swsusp_set_page_free(struct page *page)
1034{
1035 if (free_pages_map)
1036 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
1037}
1038
1039static int swsusp_page_is_free(struct page *page)
1040{
1041 return free_pages_map ?
1042 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1043}
1044
1045void swsusp_unset_page_free(struct page *page)
1046{
1047 if (free_pages_map)
1048 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1049}
1050
1051static void swsusp_set_page_forbidden(struct page *page)
1052{
1053 if (forbidden_pages_map)
1054 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1055}
1056
1057int swsusp_page_is_forbidden(struct page *page)
1058{
1059 return forbidden_pages_map ?
1060 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1061}
1062
1063static void swsusp_unset_page_forbidden(struct page *page)
1064{
1065 if (forbidden_pages_map)
1066 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1067}
1068
1069/**
1070 * mark_nosave_pages - Mark pages that should not be saved.
1071 * @bm: Memory bitmap.
1072 *
1073 * Set the bits in @bm that correspond to the page frames the contents of which
1074 * should not be saved.
1075 */
1076static void mark_nosave_pages(struct memory_bitmap *bm)
1077{
1078 struct nosave_region *region;
1079
1080 if (list_empty(&nosave_regions))
1081 return;
1082
1083 list_for_each_entry(region, &nosave_regions, list) {
1084 unsigned long pfn;
1085
1086 pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1087 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1088 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1089 - 1);
1090
1091 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1092 if (pfn_valid(pfn)) {
1093 /*
1094 * It is safe to ignore the result of
1095 * mem_bm_set_bit_check() here, since we won't
1096 * touch the PFNs for which the error is
1097 * returned anyway.
1098 */
1099 mem_bm_set_bit_check(bm, pfn);
1100 }
1101 }
1102}
1103
1104/**
1105 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1106 *
1107 * Create bitmaps needed for marking page frames that should not be saved and
1108 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1109 * only modified if everything goes well, because we don't want the bits to be
1110 * touched before both bitmaps are set up.
1111 */
1112int create_basic_memory_bitmaps(void)
1113{
1114 struct memory_bitmap *bm1, *bm2;
1115 int error = 0;
1116
1117 if (forbidden_pages_map && free_pages_map)
1118 return 0;
1119 else
1120 BUG_ON(forbidden_pages_map || free_pages_map);
1121
1122 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1123 if (!bm1)
1124 return -ENOMEM;
1125
1126 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1127 if (error)
1128 goto Free_first_object;
1129
1130 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1131 if (!bm2)
1132 goto Free_first_bitmap;
1133
1134 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1135 if (error)
1136 goto Free_second_object;
1137
1138 forbidden_pages_map = bm1;
1139 free_pages_map = bm2;
1140 mark_nosave_pages(forbidden_pages_map);
1141
1142 pr_debug("Basic memory bitmaps created\n");
1143
1144 return 0;
1145
1146 Free_second_object:
1147 kfree(bm2);
1148 Free_first_bitmap:
1149 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1150 Free_first_object:
1151 kfree(bm1);
1152 return -ENOMEM;
1153}
1154
1155/**
1156 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1157 *
1158 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1159 * auxiliary pointers are necessary so that the bitmaps themselves are not
1160 * referred to while they are being freed.
1161 */
1162void free_basic_memory_bitmaps(void)
1163{
1164 struct memory_bitmap *bm1, *bm2;
1165
1166 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1167 return;
1168
1169 bm1 = forbidden_pages_map;
1170 bm2 = free_pages_map;
1171 forbidden_pages_map = NULL;
1172 free_pages_map = NULL;
1173 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1174 kfree(bm1);
1175 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1176 kfree(bm2);
1177
1178 pr_debug("Basic memory bitmaps freed\n");
1179}
1180
1181static void clear_or_poison_free_page(struct page *page)
1182{
1183 if (page_poisoning_enabled_static())
1184 __kernel_poison_pages(page, 1);
1185 else if (want_init_on_free())
1186 clear_highpage(page);
1187}
1188
1189void clear_or_poison_free_pages(void)
1190{
1191 struct memory_bitmap *bm = free_pages_map;
1192 unsigned long pfn;
1193
1194 if (WARN_ON(!(free_pages_map)))
1195 return;
1196
1197 if (page_poisoning_enabled() || want_init_on_free()) {
1198 memory_bm_position_reset(bm);
1199 pfn = memory_bm_next_pfn(bm);
1200 while (pfn != BM_END_OF_MAP) {
1201 if (pfn_valid(pfn))
1202 clear_or_poison_free_page(pfn_to_page(pfn));
1203
1204 pfn = memory_bm_next_pfn(bm);
1205 }
1206 memory_bm_position_reset(bm);
1207 pr_info("free pages cleared after restore\n");
1208 }
1209}
1210
1211/**
1212 * snapshot_additional_pages - Estimate the number of extra pages needed.
1213 * @zone: Memory zone to carry out the computation for.
1214 *
1215 * Estimate the number of additional pages needed for setting up a hibernation
1216 * image data structures for @zone (usually, the returned value is greater than
1217 * the exact number).
1218 */
1219unsigned int snapshot_additional_pages(struct zone *zone)
1220{
1221 unsigned int rtree, nodes;
1222
1223 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1224 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1225 LINKED_PAGE_DATA_SIZE);
1226 while (nodes > 1) {
1227 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1228 rtree += nodes;
1229 }
1230
1231 return 2 * rtree;
1232}
1233
1234#ifdef CONFIG_HIGHMEM
1235/**
1236 * count_free_highmem_pages - Compute the total number of free highmem pages.
1237 *
1238 * The returned number is system-wide.
1239 */
1240static unsigned int count_free_highmem_pages(void)
1241{
1242 struct zone *zone;
1243 unsigned int cnt = 0;
1244
1245 for_each_populated_zone(zone)
1246 if (is_highmem(zone))
1247 cnt += zone_page_state(zone, NR_FREE_PAGES);
1248
1249 return cnt;
1250}
1251
1252/**
1253 * saveable_highmem_page - Check if a highmem page is saveable.
1254 *
1255 * Determine whether a highmem page should be included in a hibernation image.
1256 *
1257 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1258 * and it isn't part of a free chunk of pages.
1259 */
1260static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1261{
1262 struct page *page;
1263
1264 if (!pfn_valid(pfn))
1265 return NULL;
1266
1267 page = pfn_to_online_page(pfn);
1268 if (!page || page_zone(page) != zone)
1269 return NULL;
1270
1271 BUG_ON(!PageHighMem(page));
1272
1273 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1274 return NULL;
1275
1276 if (PageReserved(page) || PageOffline(page))
1277 return NULL;
1278
1279 if (page_is_guard(page))
1280 return NULL;
1281
1282 return page;
1283}
1284
1285/**
1286 * count_highmem_pages - Compute the total number of saveable highmem pages.
1287 */
1288static unsigned int count_highmem_pages(void)
1289{
1290 struct zone *zone;
1291 unsigned int n = 0;
1292
1293 for_each_populated_zone(zone) {
1294 unsigned long pfn, max_zone_pfn;
1295
1296 if (!is_highmem(zone))
1297 continue;
1298
1299 mark_free_pages(zone);
1300 max_zone_pfn = zone_end_pfn(zone);
1301 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1302 if (saveable_highmem_page(zone, pfn))
1303 n++;
1304 }
1305 return n;
1306}
1307#else
1308static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1309{
1310 return NULL;
1311}
1312#endif /* CONFIG_HIGHMEM */
1313
1314/**
1315 * saveable_page - Check if the given page is saveable.
1316 *
1317 * Determine whether a non-highmem page should be included in a hibernation
1318 * image.
1319 *
1320 * We should save the page if it isn't Nosave, and is not in the range
1321 * of pages statically defined as 'unsaveable', and it isn't part of
1322 * a free chunk of pages.
1323 */
1324static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1325{
1326 struct page *page;
1327
1328 if (!pfn_valid(pfn))
1329 return NULL;
1330
1331 page = pfn_to_online_page(pfn);
1332 if (!page || page_zone(page) != zone)
1333 return NULL;
1334
1335 BUG_ON(PageHighMem(page));
1336
1337 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1338 return NULL;
1339
1340 if (PageOffline(page))
1341 return NULL;
1342
1343 if (PageReserved(page)
1344 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1345 return NULL;
1346
1347 if (page_is_guard(page))
1348 return NULL;
1349
1350 return page;
1351}
1352
1353/**
1354 * count_data_pages - Compute the total number of saveable non-highmem pages.
1355 */
1356static unsigned int count_data_pages(void)
1357{
1358 struct zone *zone;
1359 unsigned long pfn, max_zone_pfn;
1360 unsigned int n = 0;
1361
1362 for_each_populated_zone(zone) {
1363 if (is_highmem(zone))
1364 continue;
1365
1366 mark_free_pages(zone);
1367 max_zone_pfn = zone_end_pfn(zone);
1368 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1369 if (saveable_page(zone, pfn))
1370 n++;
1371 }
1372 return n;
1373}
1374
1375/*
1376 * This is needed, because copy_page and memcpy are not usable for copying
1377 * task structs.
1378 */
1379static inline void do_copy_page(long *dst, long *src)
1380{
1381 int n;
1382
1383 for (n = PAGE_SIZE / sizeof(long); n; n--)
1384 *dst++ = *src++;
1385}
1386
1387/**
1388 * safe_copy_page - Copy a page in a safe way.
1389 *
1390 * Check if the page we are going to copy is marked as present in the kernel
1391 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1392 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1393 * always returns 'true'.
1394 */
1395static void safe_copy_page(void *dst, struct page *s_page)
1396{
1397 if (kernel_page_present(s_page)) {
1398 do_copy_page(dst, page_address(s_page));
1399 } else {
1400 hibernate_map_page(s_page);
1401 do_copy_page(dst, page_address(s_page));
1402 hibernate_unmap_page(s_page);
1403 }
1404}
1405
1406#ifdef CONFIG_HIGHMEM
1407static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1408{
1409 return is_highmem(zone) ?
1410 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1411}
1412
1413static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1414{
1415 struct page *s_page, *d_page;
1416 void *src, *dst;
1417
1418 s_page = pfn_to_page(src_pfn);
1419 d_page = pfn_to_page(dst_pfn);
1420 if (PageHighMem(s_page)) {
1421 src = kmap_atomic(s_page);
1422 dst = kmap_atomic(d_page);
1423 do_copy_page(dst, src);
1424 kunmap_atomic(dst);
1425 kunmap_atomic(src);
1426 } else {
1427 if (PageHighMem(d_page)) {
1428 /*
1429 * The page pointed to by src may contain some kernel
1430 * data modified by kmap_atomic()
1431 */
1432 safe_copy_page(buffer, s_page);
1433 dst = kmap_atomic(d_page);
1434 copy_page(dst, buffer);
1435 kunmap_atomic(dst);
1436 } else {
1437 safe_copy_page(page_address(d_page), s_page);
1438 }
1439 }
1440}
1441#else
1442#define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1443
1444static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1445{
1446 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1447 pfn_to_page(src_pfn));
1448}
1449#endif /* CONFIG_HIGHMEM */
1450
1451static void copy_data_pages(struct memory_bitmap *copy_bm,
1452 struct memory_bitmap *orig_bm)
1453{
1454 struct zone *zone;
1455 unsigned long pfn;
1456
1457 for_each_populated_zone(zone) {
1458 unsigned long max_zone_pfn;
1459
1460 mark_free_pages(zone);
1461 max_zone_pfn = zone_end_pfn(zone);
1462 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1463 if (page_is_saveable(zone, pfn))
1464 memory_bm_set_bit(orig_bm, pfn);
1465 }
1466 memory_bm_position_reset(orig_bm);
1467 memory_bm_position_reset(copy_bm);
1468 for(;;) {
1469 pfn = memory_bm_next_pfn(orig_bm);
1470 if (unlikely(pfn == BM_END_OF_MAP))
1471 break;
1472 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1473 }
1474}
1475
1476/* Total number of image pages */
1477static unsigned int nr_copy_pages;
1478/* Number of pages needed for saving the original pfns of the image pages */
1479static unsigned int nr_meta_pages;
1480/*
1481 * Numbers of normal and highmem page frames allocated for hibernation image
1482 * before suspending devices.
1483 */
1484static unsigned int alloc_normal, alloc_highmem;
1485/*
1486 * Memory bitmap used for marking saveable pages (during hibernation) or
1487 * hibernation image pages (during restore)
1488 */
1489static struct memory_bitmap orig_bm;
1490/*
1491 * Memory bitmap used during hibernation for marking allocated page frames that
1492 * will contain copies of saveable pages. During restore it is initially used
1493 * for marking hibernation image pages, but then the set bits from it are
1494 * duplicated in @orig_bm and it is released. On highmem systems it is next
1495 * used for marking "safe" highmem pages, but it has to be reinitialized for
1496 * this purpose.
1497 */
1498static struct memory_bitmap copy_bm;
1499
1500/**
1501 * swsusp_free - Free pages allocated for hibernation image.
1502 *
1503 * Image pages are allocated before snapshot creation, so they need to be
1504 * released after resume.
1505 */
1506void swsusp_free(void)
1507{
1508 unsigned long fb_pfn, fr_pfn;
1509
1510 if (!forbidden_pages_map || !free_pages_map)
1511 goto out;
1512
1513 memory_bm_position_reset(forbidden_pages_map);
1514 memory_bm_position_reset(free_pages_map);
1515
1516loop:
1517 fr_pfn = memory_bm_next_pfn(free_pages_map);
1518 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1519
1520 /*
1521 * Find the next bit set in both bitmaps. This is guaranteed to
1522 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1523 */
1524 do {
1525 if (fb_pfn < fr_pfn)
1526 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1527 if (fr_pfn < fb_pfn)
1528 fr_pfn = memory_bm_next_pfn(free_pages_map);
1529 } while (fb_pfn != fr_pfn);
1530
1531 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1532 struct page *page = pfn_to_page(fr_pfn);
1533
1534 memory_bm_clear_current(forbidden_pages_map);
1535 memory_bm_clear_current(free_pages_map);
1536 hibernate_restore_unprotect_page(page_address(page));
1537 __free_page(page);
1538 goto loop;
1539 }
1540
1541out:
1542 nr_copy_pages = 0;
1543 nr_meta_pages = 0;
1544 restore_pblist = NULL;
1545 buffer = NULL;
1546 alloc_normal = 0;
1547 alloc_highmem = 0;
1548 hibernate_restore_protection_end();
1549}
1550
1551/* Helper functions used for the shrinking of memory. */
1552
1553#define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1554
1555/**
1556 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1557 * @nr_pages: Number of page frames to allocate.
1558 * @mask: GFP flags to use for the allocation.
1559 *
1560 * Return value: Number of page frames actually allocated
1561 */
1562static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1563{
1564 unsigned long nr_alloc = 0;
1565
1566 while (nr_pages > 0) {
1567 struct page *page;
1568
1569 page = alloc_image_page(mask);
1570 if (!page)
1571 break;
1572 memory_bm_set_bit(©_bm, page_to_pfn(page));
1573 if (PageHighMem(page))
1574 alloc_highmem++;
1575 else
1576 alloc_normal++;
1577 nr_pages--;
1578 nr_alloc++;
1579 }
1580
1581 return nr_alloc;
1582}
1583
1584static unsigned long preallocate_image_memory(unsigned long nr_pages,
1585 unsigned long avail_normal)
1586{
1587 unsigned long alloc;
1588
1589 if (avail_normal <= alloc_normal)
1590 return 0;
1591
1592 alloc = avail_normal - alloc_normal;
1593 if (nr_pages < alloc)
1594 alloc = nr_pages;
1595
1596 return preallocate_image_pages(alloc, GFP_IMAGE);
1597}
1598
1599#ifdef CONFIG_HIGHMEM
1600static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1601{
1602 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1603}
1604
1605/**
1606 * __fraction - Compute (an approximation of) x * (multiplier / base).
1607 */
1608static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1609{
1610 return div64_u64(x * multiplier, base);
1611}
1612
1613static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1614 unsigned long highmem,
1615 unsigned long total)
1616{
1617 unsigned long alloc = __fraction(nr_pages, highmem, total);
1618
1619 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1620}
1621#else /* CONFIG_HIGHMEM */
1622static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1623{
1624 return 0;
1625}
1626
1627static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1628 unsigned long highmem,
1629 unsigned long total)
1630{
1631 return 0;
1632}
1633#endif /* CONFIG_HIGHMEM */
1634
1635/**
1636 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1637 */
1638static unsigned long free_unnecessary_pages(void)
1639{
1640 unsigned long save, to_free_normal, to_free_highmem, free;
1641
1642 save = count_data_pages();
1643 if (alloc_normal >= save) {
1644 to_free_normal = alloc_normal - save;
1645 save = 0;
1646 } else {
1647 to_free_normal = 0;
1648 save -= alloc_normal;
1649 }
1650 save += count_highmem_pages();
1651 if (alloc_highmem >= save) {
1652 to_free_highmem = alloc_highmem - save;
1653 } else {
1654 to_free_highmem = 0;
1655 save -= alloc_highmem;
1656 if (to_free_normal > save)
1657 to_free_normal -= save;
1658 else
1659 to_free_normal = 0;
1660 }
1661 free = to_free_normal + to_free_highmem;
1662
1663 memory_bm_position_reset(©_bm);
1664
1665 while (to_free_normal > 0 || to_free_highmem > 0) {
1666 unsigned long pfn = memory_bm_next_pfn(©_bm);
1667 struct page *page = pfn_to_page(pfn);
1668
1669 if (PageHighMem(page)) {
1670 if (!to_free_highmem)
1671 continue;
1672 to_free_highmem--;
1673 alloc_highmem--;
1674 } else {
1675 if (!to_free_normal)
1676 continue;
1677 to_free_normal--;
1678 alloc_normal--;
1679 }
1680 memory_bm_clear_bit(©_bm, pfn);
1681 swsusp_unset_page_forbidden(page);
1682 swsusp_unset_page_free(page);
1683 __free_page(page);
1684 }
1685
1686 return free;
1687}
1688
1689/**
1690 * minimum_image_size - Estimate the minimum acceptable size of an image.
1691 * @saveable: Number of saveable pages in the system.
1692 *
1693 * We want to avoid attempting to free too much memory too hard, so estimate the
1694 * minimum acceptable size of a hibernation image to use as the lower limit for
1695 * preallocating memory.
1696 *
1697 * We assume that the minimum image size should be proportional to
1698 *
1699 * [number of saveable pages] - [number of pages that can be freed in theory]
1700 *
1701 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1702 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1703 */
1704static unsigned long minimum_image_size(unsigned long saveable)
1705{
1706 unsigned long size;
1707
1708 size = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B)
1709 + global_node_page_state(NR_ACTIVE_ANON)
1710 + global_node_page_state(NR_INACTIVE_ANON)
1711 + global_node_page_state(NR_ACTIVE_FILE)
1712 + global_node_page_state(NR_INACTIVE_FILE);
1713
1714 return saveable <= size ? 0 : saveable - size;
1715}
1716
1717/**
1718 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1719 *
1720 * To create a hibernation image it is necessary to make a copy of every page
1721 * frame in use. We also need a number of page frames to be free during
1722 * hibernation for allocations made while saving the image and for device
1723 * drivers, in case they need to allocate memory from their hibernation
1724 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1725 * estimate) and reserved_size divided by PAGE_SIZE (which is tunable through
1726 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1727 * total number of available page frames and allocate at least
1728 *
1729 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1730 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1731 *
1732 * of them, which corresponds to the maximum size of a hibernation image.
1733 *
1734 * If image_size is set below the number following from the above formula,
1735 * the preallocation of memory is continued until the total number of saveable
1736 * pages in the system is below the requested image size or the minimum
1737 * acceptable image size returned by minimum_image_size(), whichever is greater.
1738 */
1739int hibernate_preallocate_memory(void)
1740{
1741 struct zone *zone;
1742 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1743 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1744 ktime_t start, stop;
1745 int error;
1746
1747 pr_info("Preallocating image memory\n");
1748 start = ktime_get();
1749
1750 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1751 if (error) {
1752 pr_err("Cannot allocate original bitmap\n");
1753 goto err_out;
1754 }
1755
1756 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1757 if (error) {
1758 pr_err("Cannot allocate copy bitmap\n");
1759 goto err_out;
1760 }
1761
1762 alloc_normal = 0;
1763 alloc_highmem = 0;
1764
1765 /* Count the number of saveable data pages. */
1766 save_highmem = count_highmem_pages();
1767 saveable = count_data_pages();
1768
1769 /*
1770 * Compute the total number of page frames we can use (count) and the
1771 * number of pages needed for image metadata (size).
1772 */
1773 count = saveable;
1774 saveable += save_highmem;
1775 highmem = save_highmem;
1776 size = 0;
1777 for_each_populated_zone(zone) {
1778 size += snapshot_additional_pages(zone);
1779 if (is_highmem(zone))
1780 highmem += zone_page_state(zone, NR_FREE_PAGES);
1781 else
1782 count += zone_page_state(zone, NR_FREE_PAGES);
1783 }
1784 avail_normal = count;
1785 count += highmem;
1786 count -= totalreserve_pages;
1787
1788 /* Compute the maximum number of saveable pages to leave in memory. */
1789 max_size = (count - (size + PAGES_FOR_IO)) / 2
1790 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1791 /* Compute the desired number of image pages specified by image_size. */
1792 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1793 if (size > max_size)
1794 size = max_size;
1795 /*
1796 * If the desired number of image pages is at least as large as the
1797 * current number of saveable pages in memory, allocate page frames for
1798 * the image and we're done.
1799 */
1800 if (size >= saveable) {
1801 pages = preallocate_image_highmem(save_highmem);
1802 pages += preallocate_image_memory(saveable - pages, avail_normal);
1803 goto out;
1804 }
1805
1806 /* Estimate the minimum size of the image. */
1807 pages = minimum_image_size(saveable);
1808 /*
1809 * To avoid excessive pressure on the normal zone, leave room in it to
1810 * accommodate an image of the minimum size (unless it's already too
1811 * small, in which case don't preallocate pages from it at all).
1812 */
1813 if (avail_normal > pages)
1814 avail_normal -= pages;
1815 else
1816 avail_normal = 0;
1817 if (size < pages)
1818 size = min_t(unsigned long, pages, max_size);
1819
1820 /*
1821 * Let the memory management subsystem know that we're going to need a
1822 * large number of page frames to allocate and make it free some memory.
1823 * NOTE: If this is not done, performance will be hurt badly in some
1824 * test cases.
1825 */
1826 shrink_all_memory(saveable - size);
1827
1828 /*
1829 * The number of saveable pages in memory was too high, so apply some
1830 * pressure to decrease it. First, make room for the largest possible
1831 * image and fail if that doesn't work. Next, try to decrease the size
1832 * of the image as much as indicated by 'size' using allocations from
1833 * highmem and non-highmem zones separately.
1834 */
1835 pages_highmem = preallocate_image_highmem(highmem / 2);
1836 alloc = count - max_size;
1837 if (alloc > pages_highmem)
1838 alloc -= pages_highmem;
1839 else
1840 alloc = 0;
1841 pages = preallocate_image_memory(alloc, avail_normal);
1842 if (pages < alloc) {
1843 /* We have exhausted non-highmem pages, try highmem. */
1844 alloc -= pages;
1845 pages += pages_highmem;
1846 pages_highmem = preallocate_image_highmem(alloc);
1847 if (pages_highmem < alloc) {
1848 pr_err("Image allocation is %lu pages short\n",
1849 alloc - pages_highmem);
1850 goto err_out;
1851 }
1852 pages += pages_highmem;
1853 /*
1854 * size is the desired number of saveable pages to leave in
1855 * memory, so try to preallocate (all memory - size) pages.
1856 */
1857 alloc = (count - pages) - size;
1858 pages += preallocate_image_highmem(alloc);
1859 } else {
1860 /*
1861 * There are approximately max_size saveable pages at this point
1862 * and we want to reduce this number down to size.
1863 */
1864 alloc = max_size - size;
1865 size = preallocate_highmem_fraction(alloc, highmem, count);
1866 pages_highmem += size;
1867 alloc -= size;
1868 size = preallocate_image_memory(alloc, avail_normal);
1869 pages_highmem += preallocate_image_highmem(alloc - size);
1870 pages += pages_highmem + size;
1871 }
1872
1873 /*
1874 * We only need as many page frames for the image as there are saveable
1875 * pages in memory, but we have allocated more. Release the excessive
1876 * ones now.
1877 */
1878 pages -= free_unnecessary_pages();
1879
1880 out:
1881 stop = ktime_get();
1882 pr_info("Allocated %lu pages for snapshot\n", pages);
1883 swsusp_show_speed(start, stop, pages, "Allocated");
1884
1885 return 0;
1886
1887 err_out:
1888 swsusp_free();
1889 return -ENOMEM;
1890}
1891
1892#ifdef CONFIG_HIGHMEM
1893/**
1894 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1895 *
1896 * Compute the number of non-highmem pages that will be necessary for creating
1897 * copies of highmem pages.
1898 */
1899static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1900{
1901 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1902
1903 if (free_highmem >= nr_highmem)
1904 nr_highmem = 0;
1905 else
1906 nr_highmem -= free_highmem;
1907
1908 return nr_highmem;
1909}
1910#else
1911static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1912#endif /* CONFIG_HIGHMEM */
1913
1914/**
1915 * enough_free_mem - Check if there is enough free memory for the image.
1916 */
1917static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1918{
1919 struct zone *zone;
1920 unsigned int free = alloc_normal;
1921
1922 for_each_populated_zone(zone)
1923 if (!is_highmem(zone))
1924 free += zone_page_state(zone, NR_FREE_PAGES);
1925
1926 nr_pages += count_pages_for_highmem(nr_highmem);
1927 pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
1928 nr_pages, PAGES_FOR_IO, free);
1929
1930 return free > nr_pages + PAGES_FOR_IO;
1931}
1932
1933#ifdef CONFIG_HIGHMEM
1934/**
1935 * get_highmem_buffer - Allocate a buffer for highmem pages.
1936 *
1937 * If there are some highmem pages in the hibernation image, we may need a
1938 * buffer to copy them and/or load their data.
1939 */
1940static inline int get_highmem_buffer(int safe_needed)
1941{
1942 buffer = get_image_page(GFP_ATOMIC, safe_needed);
1943 return buffer ? 0 : -ENOMEM;
1944}
1945
1946/**
1947 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1948 *
1949 * Try to allocate as many pages as needed, but if the number of free highmem
1950 * pages is less than that, allocate them all.
1951 */
1952static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1953 unsigned int nr_highmem)
1954{
1955 unsigned int to_alloc = count_free_highmem_pages();
1956
1957 if (to_alloc > nr_highmem)
1958 to_alloc = nr_highmem;
1959
1960 nr_highmem -= to_alloc;
1961 while (to_alloc-- > 0) {
1962 struct page *page;
1963
1964 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1965 memory_bm_set_bit(bm, page_to_pfn(page));
1966 }
1967 return nr_highmem;
1968}
1969#else
1970static inline int get_highmem_buffer(int safe_needed) { return 0; }
1971
1972static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1973 unsigned int n) { return 0; }
1974#endif /* CONFIG_HIGHMEM */
1975
1976/**
1977 * swsusp_alloc - Allocate memory for hibernation image.
1978 *
1979 * We first try to allocate as many highmem pages as there are
1980 * saveable highmem pages in the system. If that fails, we allocate
1981 * non-highmem pages for the copies of the remaining highmem ones.
1982 *
1983 * In this approach it is likely that the copies of highmem pages will
1984 * also be located in the high memory, because of the way in which
1985 * copy_data_pages() works.
1986 */
1987static int swsusp_alloc(struct memory_bitmap *copy_bm,
1988 unsigned int nr_pages, unsigned int nr_highmem)
1989{
1990 if (nr_highmem > 0) {
1991 if (get_highmem_buffer(PG_ANY))
1992 goto err_out;
1993 if (nr_highmem > alloc_highmem) {
1994 nr_highmem -= alloc_highmem;
1995 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1996 }
1997 }
1998 if (nr_pages > alloc_normal) {
1999 nr_pages -= alloc_normal;
2000 while (nr_pages-- > 0) {
2001 struct page *page;
2002
2003 page = alloc_image_page(GFP_ATOMIC);
2004 if (!page)
2005 goto err_out;
2006 memory_bm_set_bit(copy_bm, page_to_pfn(page));
2007 }
2008 }
2009
2010 return 0;
2011
2012 err_out:
2013 swsusp_free();
2014 return -ENOMEM;
2015}
2016
2017asmlinkage __visible int swsusp_save(void)
2018{
2019 unsigned int nr_pages, nr_highmem;
2020
2021 pr_info("Creating image:\n");
2022
2023 drain_local_pages(NULL);
2024 nr_pages = count_data_pages();
2025 nr_highmem = count_highmem_pages();
2026 pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
2027
2028 if (!enough_free_mem(nr_pages, nr_highmem)) {
2029 pr_err("Not enough free memory\n");
2030 return -ENOMEM;
2031 }
2032
2033 if (swsusp_alloc(©_bm, nr_pages, nr_highmem)) {
2034 pr_err("Memory allocation failed\n");
2035 return -ENOMEM;
2036 }
2037
2038 /*
2039 * During allocating of suspend pagedir, new cold pages may appear.
2040 * Kill them.
2041 */
2042 drain_local_pages(NULL);
2043 copy_data_pages(©_bm, &orig_bm);
2044
2045 /*
2046 * End of critical section. From now on, we can write to memory,
2047 * but we should not touch disk. This specially means we must _not_
2048 * touch swap space! Except we must write out our image of course.
2049 */
2050
2051 nr_pages += nr_highmem;
2052 nr_copy_pages = nr_pages;
2053 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2054
2055 pr_info("Image created (%d pages copied)\n", nr_pages);
2056
2057 return 0;
2058}
2059
2060#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2061static int init_header_complete(struct swsusp_info *info)
2062{
2063 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2064 info->version_code = LINUX_VERSION_CODE;
2065 return 0;
2066}
2067
2068static const char *check_image_kernel(struct swsusp_info *info)
2069{
2070 if (info->version_code != LINUX_VERSION_CODE)
2071 return "kernel version";
2072 if (strcmp(info->uts.sysname,init_utsname()->sysname))
2073 return "system type";
2074 if (strcmp(info->uts.release,init_utsname()->release))
2075 return "kernel release";
2076 if (strcmp(info->uts.version,init_utsname()->version))
2077 return "version";
2078 if (strcmp(info->uts.machine,init_utsname()->machine))
2079 return "machine";
2080 return NULL;
2081}
2082#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2083
2084unsigned long snapshot_get_image_size(void)
2085{
2086 return nr_copy_pages + nr_meta_pages + 1;
2087}
2088
2089static int init_header(struct swsusp_info *info)
2090{
2091 memset(info, 0, sizeof(struct swsusp_info));
2092 info->num_physpages = get_num_physpages();
2093 info->image_pages = nr_copy_pages;
2094 info->pages = snapshot_get_image_size();
2095 info->size = info->pages;
2096 info->size <<= PAGE_SHIFT;
2097 return init_header_complete(info);
2098}
2099
2100/**
2101 * pack_pfns - Prepare PFNs for saving.
2102 * @bm: Memory bitmap.
2103 * @buf: Memory buffer to store the PFNs in.
2104 *
2105 * PFNs corresponding to set bits in @bm are stored in the area of memory
2106 * pointed to by @buf (1 page at a time).
2107 */
2108static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2109{
2110 int j;
2111
2112 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2113 buf[j] = memory_bm_next_pfn(bm);
2114 if (unlikely(buf[j] == BM_END_OF_MAP))
2115 break;
2116 }
2117}
2118
2119/**
2120 * snapshot_read_next - Get the address to read the next image page from.
2121 * @handle: Snapshot handle to be used for the reading.
2122 *
2123 * On the first call, @handle should point to a zeroed snapshot_handle
2124 * structure. The structure gets populated then and a pointer to it should be
2125 * passed to this function every next time.
2126 *
2127 * On success, the function returns a positive number. Then, the caller
2128 * is allowed to read up to the returned number of bytes from the memory
2129 * location computed by the data_of() macro.
2130 *
2131 * The function returns 0 to indicate the end of the data stream condition,
2132 * and negative numbers are returned on errors. If that happens, the structure
2133 * pointed to by @handle is not updated and should not be used any more.
2134 */
2135int snapshot_read_next(struct snapshot_handle *handle)
2136{
2137 if (handle->cur > nr_meta_pages + nr_copy_pages)
2138 return 0;
2139
2140 if (!buffer) {
2141 /* This makes the buffer be freed by swsusp_free() */
2142 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2143 if (!buffer)
2144 return -ENOMEM;
2145 }
2146 if (!handle->cur) {
2147 int error;
2148
2149 error = init_header((struct swsusp_info *)buffer);
2150 if (error)
2151 return error;
2152 handle->buffer = buffer;
2153 memory_bm_position_reset(&orig_bm);
2154 memory_bm_position_reset(©_bm);
2155 } else if (handle->cur <= nr_meta_pages) {
2156 clear_page(buffer);
2157 pack_pfns(buffer, &orig_bm);
2158 } else {
2159 struct page *page;
2160
2161 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2162 if (PageHighMem(page)) {
2163 /*
2164 * Highmem pages are copied to the buffer,
2165 * because we can't return with a kmapped
2166 * highmem page (we may not be called again).
2167 */
2168 void *kaddr;
2169
2170 kaddr = kmap_atomic(page);
2171 copy_page(buffer, kaddr);
2172 kunmap_atomic(kaddr);
2173 handle->buffer = buffer;
2174 } else {
2175 handle->buffer = page_address(page);
2176 }
2177 }
2178 handle->cur++;
2179 return PAGE_SIZE;
2180}
2181
2182static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2183 struct memory_bitmap *src)
2184{
2185 unsigned long pfn;
2186
2187 memory_bm_position_reset(src);
2188 pfn = memory_bm_next_pfn(src);
2189 while (pfn != BM_END_OF_MAP) {
2190 memory_bm_set_bit(dst, pfn);
2191 pfn = memory_bm_next_pfn(src);
2192 }
2193}
2194
2195/**
2196 * mark_unsafe_pages - Mark pages that were used before hibernation.
2197 *
2198 * Mark the pages that cannot be used for storing the image during restoration,
2199 * because they conflict with the pages that had been used before hibernation.
2200 */
2201static void mark_unsafe_pages(struct memory_bitmap *bm)
2202{
2203 unsigned long pfn;
2204
2205 /* Clear the "free"/"unsafe" bit for all PFNs */
2206 memory_bm_position_reset(free_pages_map);
2207 pfn = memory_bm_next_pfn(free_pages_map);
2208 while (pfn != BM_END_OF_MAP) {
2209 memory_bm_clear_current(free_pages_map);
2210 pfn = memory_bm_next_pfn(free_pages_map);
2211 }
2212
2213 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2214 duplicate_memory_bitmap(free_pages_map, bm);
2215
2216 allocated_unsafe_pages = 0;
2217}
2218
2219static int check_header(struct swsusp_info *info)
2220{
2221 const char *reason;
2222
2223 reason = check_image_kernel(info);
2224 if (!reason && info->num_physpages != get_num_physpages())
2225 reason = "memory size";
2226 if (reason) {
2227 pr_err("Image mismatch: %s\n", reason);
2228 return -EPERM;
2229 }
2230 return 0;
2231}
2232
2233/**
2234 * load header - Check the image header and copy the data from it.
2235 */
2236static int load_header(struct swsusp_info *info)
2237{
2238 int error;
2239
2240 restore_pblist = NULL;
2241 error = check_header(info);
2242 if (!error) {
2243 nr_copy_pages = info->image_pages;
2244 nr_meta_pages = info->pages - info->image_pages - 1;
2245 }
2246 return error;
2247}
2248
2249/**
2250 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2251 * @bm: Memory bitmap.
2252 * @buf: Area of memory containing the PFNs.
2253 *
2254 * For each element of the array pointed to by @buf (1 page at a time), set the
2255 * corresponding bit in @bm.
2256 */
2257static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2258{
2259 int j;
2260
2261 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2262 if (unlikely(buf[j] == BM_END_OF_MAP))
2263 break;
2264
2265 if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2266 memory_bm_set_bit(bm, buf[j]);
2267 else
2268 return -EFAULT;
2269 }
2270
2271 return 0;
2272}
2273
2274#ifdef CONFIG_HIGHMEM
2275/*
2276 * struct highmem_pbe is used for creating the list of highmem pages that
2277 * should be restored atomically during the resume from disk, because the page
2278 * frames they have occupied before the suspend are in use.
2279 */
2280struct highmem_pbe {
2281 struct page *copy_page; /* data is here now */
2282 struct page *orig_page; /* data was here before the suspend */
2283 struct highmem_pbe *next;
2284};
2285
2286/*
2287 * List of highmem PBEs needed for restoring the highmem pages that were
2288 * allocated before the suspend and included in the suspend image, but have
2289 * also been allocated by the "resume" kernel, so their contents cannot be
2290 * written directly to their "original" page frames.
2291 */
2292static struct highmem_pbe *highmem_pblist;
2293
2294/**
2295 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2296 * @bm: Memory bitmap.
2297 *
2298 * The bits in @bm that correspond to image pages are assumed to be set.
2299 */
2300static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2301{
2302 unsigned long pfn;
2303 unsigned int cnt = 0;
2304
2305 memory_bm_position_reset(bm);
2306 pfn = memory_bm_next_pfn(bm);
2307 while (pfn != BM_END_OF_MAP) {
2308 if (PageHighMem(pfn_to_page(pfn)))
2309 cnt++;
2310
2311 pfn = memory_bm_next_pfn(bm);
2312 }
2313 return cnt;
2314}
2315
2316static unsigned int safe_highmem_pages;
2317
2318static struct memory_bitmap *safe_highmem_bm;
2319
2320/**
2321 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2322 * @bm: Pointer to an uninitialized memory bitmap structure.
2323 * @nr_highmem_p: Pointer to the number of highmem image pages.
2324 *
2325 * Try to allocate as many highmem pages as there are highmem image pages
2326 * (@nr_highmem_p points to the variable containing the number of highmem image
2327 * pages). The pages that are "safe" (ie. will not be overwritten when the
2328 * hibernation image is restored entirely) have the corresponding bits set in
2329 * @bm (it must be uninitialized).
2330 *
2331 * NOTE: This function should not be called if there are no highmem image pages.
2332 */
2333static int prepare_highmem_image(struct memory_bitmap *bm,
2334 unsigned int *nr_highmem_p)
2335{
2336 unsigned int to_alloc;
2337
2338 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2339 return -ENOMEM;
2340
2341 if (get_highmem_buffer(PG_SAFE))
2342 return -ENOMEM;
2343
2344 to_alloc = count_free_highmem_pages();
2345 if (to_alloc > *nr_highmem_p)
2346 to_alloc = *nr_highmem_p;
2347 else
2348 *nr_highmem_p = to_alloc;
2349
2350 safe_highmem_pages = 0;
2351 while (to_alloc-- > 0) {
2352 struct page *page;
2353
2354 page = alloc_page(__GFP_HIGHMEM);
2355 if (!swsusp_page_is_free(page)) {
2356 /* The page is "safe", set its bit the bitmap */
2357 memory_bm_set_bit(bm, page_to_pfn(page));
2358 safe_highmem_pages++;
2359 }
2360 /* Mark the page as allocated */
2361 swsusp_set_page_forbidden(page);
2362 swsusp_set_page_free(page);
2363 }
2364 memory_bm_position_reset(bm);
2365 safe_highmem_bm = bm;
2366 return 0;
2367}
2368
2369static struct page *last_highmem_page;
2370
2371/**
2372 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2373 *
2374 * For a given highmem image page get a buffer that suspend_write_next() should
2375 * return to its caller to write to.
2376 *
2377 * If the page is to be saved to its "original" page frame or a copy of
2378 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2379 * the copy of the page is to be made in normal memory, so the address of
2380 * the copy is returned.
2381 *
2382 * If @buffer is returned, the caller of suspend_write_next() will write
2383 * the page's contents to @buffer, so they will have to be copied to the
2384 * right location on the next call to suspend_write_next() and it is done
2385 * with the help of copy_last_highmem_page(). For this purpose, if
2386 * @buffer is returned, @last_highmem_page is set to the page to which
2387 * the data will have to be copied from @buffer.
2388 */
2389static void *get_highmem_page_buffer(struct page *page,
2390 struct chain_allocator *ca)
2391{
2392 struct highmem_pbe *pbe;
2393 void *kaddr;
2394
2395 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2396 /*
2397 * We have allocated the "original" page frame and we can
2398 * use it directly to store the loaded page.
2399 */
2400 last_highmem_page = page;
2401 return buffer;
2402 }
2403 /*
2404 * The "original" page frame has not been allocated and we have to
2405 * use a "safe" page frame to store the loaded page.
2406 */
2407 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2408 if (!pbe) {
2409 swsusp_free();
2410 return ERR_PTR(-ENOMEM);
2411 }
2412 pbe->orig_page = page;
2413 if (safe_highmem_pages > 0) {
2414 struct page *tmp;
2415
2416 /* Copy of the page will be stored in high memory */
2417 kaddr = buffer;
2418 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2419 safe_highmem_pages--;
2420 last_highmem_page = tmp;
2421 pbe->copy_page = tmp;
2422 } else {
2423 /* Copy of the page will be stored in normal memory */
2424 kaddr = safe_pages_list;
2425 safe_pages_list = safe_pages_list->next;
2426 pbe->copy_page = virt_to_page(kaddr);
2427 }
2428 pbe->next = highmem_pblist;
2429 highmem_pblist = pbe;
2430 return kaddr;
2431}
2432
2433/**
2434 * copy_last_highmem_page - Copy most the most recent highmem image page.
2435 *
2436 * Copy the contents of a highmem image from @buffer, where the caller of
2437 * snapshot_write_next() has stored them, to the right location represented by
2438 * @last_highmem_page .
2439 */
2440static void copy_last_highmem_page(void)
2441{
2442 if (last_highmem_page) {
2443 void *dst;
2444
2445 dst = kmap_atomic(last_highmem_page);
2446 copy_page(dst, buffer);
2447 kunmap_atomic(dst);
2448 last_highmem_page = NULL;
2449 }
2450}
2451
2452static inline int last_highmem_page_copied(void)
2453{
2454 return !last_highmem_page;
2455}
2456
2457static inline void free_highmem_data(void)
2458{
2459 if (safe_highmem_bm)
2460 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2461
2462 if (buffer)
2463 free_image_page(buffer, PG_UNSAFE_CLEAR);
2464}
2465#else
2466static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2467
2468static inline int prepare_highmem_image(struct memory_bitmap *bm,
2469 unsigned int *nr_highmem_p) { return 0; }
2470
2471static inline void *get_highmem_page_buffer(struct page *page,
2472 struct chain_allocator *ca)
2473{
2474 return ERR_PTR(-EINVAL);
2475}
2476
2477static inline void copy_last_highmem_page(void) {}
2478static inline int last_highmem_page_copied(void) { return 1; }
2479static inline void free_highmem_data(void) {}
2480#endif /* CONFIG_HIGHMEM */
2481
2482#define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2483
2484/**
2485 * prepare_image - Make room for loading hibernation image.
2486 * @new_bm: Uninitialized memory bitmap structure.
2487 * @bm: Memory bitmap with unsafe pages marked.
2488 *
2489 * Use @bm to mark the pages that will be overwritten in the process of
2490 * restoring the system memory state from the suspend image ("unsafe" pages)
2491 * and allocate memory for the image.
2492 *
2493 * The idea is to allocate a new memory bitmap first and then allocate
2494 * as many pages as needed for image data, but without specifying what those
2495 * pages will be used for just yet. Instead, we mark them all as allocated and
2496 * create a lists of "safe" pages to be used later. On systems with high
2497 * memory a list of "safe" highmem pages is created too.
2498 */
2499static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2500{
2501 unsigned int nr_pages, nr_highmem;
2502 struct linked_page *lp;
2503 int error;
2504
2505 /* If there is no highmem, the buffer will not be necessary */
2506 free_image_page(buffer, PG_UNSAFE_CLEAR);
2507 buffer = NULL;
2508
2509 nr_highmem = count_highmem_image_pages(bm);
2510 mark_unsafe_pages(bm);
2511
2512 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2513 if (error)
2514 goto Free;
2515
2516 duplicate_memory_bitmap(new_bm, bm);
2517 memory_bm_free(bm, PG_UNSAFE_KEEP);
2518 if (nr_highmem > 0) {
2519 error = prepare_highmem_image(bm, &nr_highmem);
2520 if (error)
2521 goto Free;
2522 }
2523 /*
2524 * Reserve some safe pages for potential later use.
2525 *
2526 * NOTE: This way we make sure there will be enough safe pages for the
2527 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2528 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2529 *
2530 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2531 */
2532 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2533 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2534 while (nr_pages > 0) {
2535 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2536 if (!lp) {
2537 error = -ENOMEM;
2538 goto Free;
2539 }
2540 lp->next = safe_pages_list;
2541 safe_pages_list = lp;
2542 nr_pages--;
2543 }
2544 /* Preallocate memory for the image */
2545 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2546 while (nr_pages > 0) {
2547 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2548 if (!lp) {
2549 error = -ENOMEM;
2550 goto Free;
2551 }
2552 if (!swsusp_page_is_free(virt_to_page(lp))) {
2553 /* The page is "safe", add it to the list */
2554 lp->next = safe_pages_list;
2555 safe_pages_list = lp;
2556 }
2557 /* Mark the page as allocated */
2558 swsusp_set_page_forbidden(virt_to_page(lp));
2559 swsusp_set_page_free(virt_to_page(lp));
2560 nr_pages--;
2561 }
2562 return 0;
2563
2564 Free:
2565 swsusp_free();
2566 return error;
2567}
2568
2569/**
2570 * get_buffer - Get the address to store the next image data page.
2571 *
2572 * Get the address that snapshot_write_next() should return to its caller to
2573 * write to.
2574 */
2575static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2576{
2577 struct pbe *pbe;
2578 struct page *page;
2579 unsigned long pfn = memory_bm_next_pfn(bm);
2580
2581 if (pfn == BM_END_OF_MAP)
2582 return ERR_PTR(-EFAULT);
2583
2584 page = pfn_to_page(pfn);
2585 if (PageHighMem(page))
2586 return get_highmem_page_buffer(page, ca);
2587
2588 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2589 /*
2590 * We have allocated the "original" page frame and we can
2591 * use it directly to store the loaded page.
2592 */
2593 return page_address(page);
2594
2595 /*
2596 * The "original" page frame has not been allocated and we have to
2597 * use a "safe" page frame to store the loaded page.
2598 */
2599 pbe = chain_alloc(ca, sizeof(struct pbe));
2600 if (!pbe) {
2601 swsusp_free();
2602 return ERR_PTR(-ENOMEM);
2603 }
2604 pbe->orig_address = page_address(page);
2605 pbe->address = safe_pages_list;
2606 safe_pages_list = safe_pages_list->next;
2607 pbe->next = restore_pblist;
2608 restore_pblist = pbe;
2609 return pbe->address;
2610}
2611
2612/**
2613 * snapshot_write_next - Get the address to store the next image page.
2614 * @handle: Snapshot handle structure to guide the writing.
2615 *
2616 * On the first call, @handle should point to a zeroed snapshot_handle
2617 * structure. The structure gets populated then and a pointer to it should be
2618 * passed to this function every next time.
2619 *
2620 * On success, the function returns a positive number. Then, the caller
2621 * is allowed to write up to the returned number of bytes to the memory
2622 * location computed by the data_of() macro.
2623 *
2624 * The function returns 0 to indicate the "end of file" condition. Negative
2625 * numbers are returned on errors, in which cases the structure pointed to by
2626 * @handle is not updated and should not be used any more.
2627 */
2628int snapshot_write_next(struct snapshot_handle *handle)
2629{
2630 static struct chain_allocator ca;
2631 int error = 0;
2632
2633 /* Check if we have already loaded the entire image */
2634 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2635 return 0;
2636
2637 handle->sync_read = 1;
2638
2639 if (!handle->cur) {
2640 if (!buffer)
2641 /* This makes the buffer be freed by swsusp_free() */
2642 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2643
2644 if (!buffer)
2645 return -ENOMEM;
2646
2647 handle->buffer = buffer;
2648 } else if (handle->cur == 1) {
2649 error = load_header(buffer);
2650 if (error)
2651 return error;
2652
2653 safe_pages_list = NULL;
2654
2655 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2656 if (error)
2657 return error;
2658
2659 hibernate_restore_protection_begin();
2660 } else if (handle->cur <= nr_meta_pages + 1) {
2661 error = unpack_orig_pfns(buffer, ©_bm);
2662 if (error)
2663 return error;
2664
2665 if (handle->cur == nr_meta_pages + 1) {
2666 error = prepare_image(&orig_bm, ©_bm);
2667 if (error)
2668 return error;
2669
2670 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2671 memory_bm_position_reset(&orig_bm);
2672 restore_pblist = NULL;
2673 handle->buffer = get_buffer(&orig_bm, &ca);
2674 handle->sync_read = 0;
2675 if (IS_ERR(handle->buffer))
2676 return PTR_ERR(handle->buffer);
2677 }
2678 } else {
2679 copy_last_highmem_page();
2680 hibernate_restore_protect_page(handle->buffer);
2681 handle->buffer = get_buffer(&orig_bm, &ca);
2682 if (IS_ERR(handle->buffer))
2683 return PTR_ERR(handle->buffer);
2684 if (handle->buffer != buffer)
2685 handle->sync_read = 0;
2686 }
2687 handle->cur++;
2688 return PAGE_SIZE;
2689}
2690
2691/**
2692 * snapshot_write_finalize - Complete the loading of a hibernation image.
2693 *
2694 * Must be called after the last call to snapshot_write_next() in case the last
2695 * page in the image happens to be a highmem page and its contents should be
2696 * stored in highmem. Additionally, it recycles bitmap memory that's not
2697 * necessary any more.
2698 */
2699void snapshot_write_finalize(struct snapshot_handle *handle)
2700{
2701 copy_last_highmem_page();
2702 hibernate_restore_protect_page(handle->buffer);
2703 /* Do that only if we have loaded the image entirely */
2704 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2705 memory_bm_recycle(&orig_bm);
2706 free_highmem_data();
2707 }
2708}
2709
2710int snapshot_image_loaded(struct snapshot_handle *handle)
2711{
2712 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2713 handle->cur <= nr_meta_pages + nr_copy_pages);
2714}
2715
2716#ifdef CONFIG_HIGHMEM
2717/* Assumes that @buf is ready and points to a "safe" page */
2718static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2719 void *buf)
2720{
2721 void *kaddr1, *kaddr2;
2722
2723 kaddr1 = kmap_atomic(p1);
2724 kaddr2 = kmap_atomic(p2);
2725 copy_page(buf, kaddr1);
2726 copy_page(kaddr1, kaddr2);
2727 copy_page(kaddr2, buf);
2728 kunmap_atomic(kaddr2);
2729 kunmap_atomic(kaddr1);
2730}
2731
2732/**
2733 * restore_highmem - Put highmem image pages into their original locations.
2734 *
2735 * For each highmem page that was in use before hibernation and is included in
2736 * the image, and also has been allocated by the "restore" kernel, swap its
2737 * current contents with the previous (ie. "before hibernation") ones.
2738 *
2739 * If the restore eventually fails, we can call this function once again and
2740 * restore the highmem state as seen by the restore kernel.
2741 */
2742int restore_highmem(void)
2743{
2744 struct highmem_pbe *pbe = highmem_pblist;
2745 void *buf;
2746
2747 if (!pbe)
2748 return 0;
2749
2750 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2751 if (!buf)
2752 return -ENOMEM;
2753
2754 while (pbe) {
2755 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2756 pbe = pbe->next;
2757 }
2758 free_image_page(buf, PG_UNSAFE_CLEAR);
2759 return 0;
2760}
2761#endif /* CONFIG_HIGHMEM */