Loading...
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
79static int swsusp_page_is_free(struct page *);
80static void swsusp_set_page_forbidden(struct page *);
81static void swsusp_unset_page_forbidden(struct page *);
82
83/*
84 * Number of bytes to reserve for memory allocations made by device drivers
85 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
86 * cause image creation to fail (tunable via /sys/power/reserved_size).
87 */
88unsigned long reserved_size;
89
90void __init hibernate_reserved_size_init(void)
91{
92 reserved_size = SPARE_PAGES * PAGE_SIZE;
93}
94
95/*
96 * Preferred image size in bytes (tunable via /sys/power/image_size).
97 * When it is set to N, swsusp will do its best to ensure the image
98 * size will not exceed N bytes, but if that is impossible, it will
99 * try to create the smallest image possible.
100 */
101unsigned long image_size;
102
103void __init hibernate_image_size_init(void)
104{
105 image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
106}
107
108/*
109 * List of PBEs needed for restoring the pages that were allocated before
110 * the suspend and included in the suspend image, but have also been
111 * allocated by the "resume" kernel, so their contents cannot be written
112 * directly to their "original" page frames.
113 */
114struct pbe *restore_pblist;
115
116/* struct linked_page is used to build chains of pages */
117
118#define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
119
120struct linked_page {
121 struct linked_page *next;
122 char data[LINKED_PAGE_DATA_SIZE];
123} __packed;
124
125/*
126 * List of "safe" pages (ie. pages that were not used by the image kernel
127 * before hibernation) that may be used as temporary storage for image kernel
128 * memory contents.
129 */
130static struct linked_page *safe_pages_list;
131
132/* Pointer to an auxiliary buffer (1 page) */
133static void *buffer;
134
135#define PG_ANY 0
136#define PG_SAFE 1
137#define PG_UNSAFE_CLEAR 1
138#define PG_UNSAFE_KEEP 0
139
140static unsigned int allocated_unsafe_pages;
141
142/**
143 * get_image_page - Allocate a page for a hibernation image.
144 * @gfp_mask: GFP mask for the allocation.
145 * @safe_needed: Get pages that were not used before hibernation (restore only)
146 *
147 * During image restoration, for storing the PBE list and the image data, we can
148 * only use memory pages that do not conflict with the pages used before
149 * hibernation. The "unsafe" pages have PageNosaveFree set and we count them
150 * using allocated_unsafe_pages.
151 *
152 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
153 * swsusp_free() can release it.
154 */
155static void *get_image_page(gfp_t gfp_mask, int safe_needed)
156{
157 void *res;
158
159 res = (void *)get_zeroed_page(gfp_mask);
160 if (safe_needed)
161 while (res && swsusp_page_is_free(virt_to_page(res))) {
162 /* The page is unsafe, mark it for swsusp_free() */
163 swsusp_set_page_forbidden(virt_to_page(res));
164 allocated_unsafe_pages++;
165 res = (void *)get_zeroed_page(gfp_mask);
166 }
167 if (res) {
168 swsusp_set_page_forbidden(virt_to_page(res));
169 swsusp_set_page_free(virt_to_page(res));
170 }
171 return res;
172}
173
174static void *__get_safe_page(gfp_t gfp_mask)
175{
176 if (safe_pages_list) {
177 void *ret = safe_pages_list;
178
179 safe_pages_list = safe_pages_list->next;
180 memset(ret, 0, PAGE_SIZE);
181 return ret;
182 }
183 return get_image_page(gfp_mask, PG_SAFE);
184}
185
186unsigned long get_safe_page(gfp_t gfp_mask)
187{
188 return (unsigned long)__get_safe_page(gfp_mask);
189}
190
191static struct page *alloc_image_page(gfp_t gfp_mask)
192{
193 struct page *page;
194
195 page = alloc_page(gfp_mask);
196 if (page) {
197 swsusp_set_page_forbidden(page);
198 swsusp_set_page_free(page);
199 }
200 return page;
201}
202
203static void recycle_safe_page(void *page_address)
204{
205 struct linked_page *lp = page_address;
206
207 lp->next = safe_pages_list;
208 safe_pages_list = lp;
209}
210
211/**
212 * free_image_page - Free a page allocated for hibernation image.
213 * @addr: Address of the page to free.
214 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
215 *
216 * The page to free should have been allocated by get_image_page() (page flags
217 * set by it are affected).
218 */
219static inline void free_image_page(void *addr, int clear_nosave_free)
220{
221 struct page *page;
222
223 BUG_ON(!virt_addr_valid(addr));
224
225 page = virt_to_page(addr);
226
227 swsusp_unset_page_forbidden(page);
228 if (clear_nosave_free)
229 swsusp_unset_page_free(page);
230
231 __free_page(page);
232}
233
234static inline void free_list_of_pages(struct linked_page *list,
235 int clear_page_nosave)
236{
237 while (list) {
238 struct linked_page *lp = list->next;
239
240 free_image_page(list, clear_page_nosave);
241 list = lp;
242 }
243}
244
245/*
246 * struct chain_allocator is used for allocating small objects out of
247 * a linked list of pages called 'the chain'.
248 *
249 * The chain grows each time when there is no room for a new object in
250 * the current page. The allocated objects cannot be freed individually.
251 * It is only possible to free them all at once, by freeing the entire
252 * chain.
253 *
254 * NOTE: The chain allocator may be inefficient if the allocated objects
255 * are not much smaller than PAGE_SIZE.
256 */
257struct chain_allocator {
258 struct linked_page *chain; /* the chain */
259 unsigned int used_space; /* total size of objects allocated out
260 of the current page */
261 gfp_t gfp_mask; /* mask for allocating pages */
262 int safe_needed; /* if set, only "safe" pages are allocated */
263};
264
265static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
266 int safe_needed)
267{
268 ca->chain = NULL;
269 ca->used_space = LINKED_PAGE_DATA_SIZE;
270 ca->gfp_mask = gfp_mask;
271 ca->safe_needed = safe_needed;
272}
273
274static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
275{
276 void *ret;
277
278 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
279 struct linked_page *lp;
280
281 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
282 get_image_page(ca->gfp_mask, PG_ANY);
283 if (!lp)
284 return NULL;
285
286 lp->next = ca->chain;
287 ca->chain = lp;
288 ca->used_space = 0;
289 }
290 ret = ca->chain->data + ca->used_space;
291 ca->used_space += size;
292 return ret;
293}
294
295/**
296 * Data types related to memory bitmaps.
297 *
298 * Memory bitmap is a structure consiting of many linked lists of
299 * objects. The main list's elements are of type struct zone_bitmap
300 * and each of them corresonds to one zone. For each zone bitmap
301 * object there is a list of objects of type struct bm_block that
302 * represent each blocks of bitmap in which information is stored.
303 *
304 * struct memory_bitmap contains a pointer to the main list of zone
305 * bitmap objects, a struct bm_position used for browsing the bitmap,
306 * and a pointer to the list of pages used for allocating all of the
307 * zone bitmap objects and bitmap block objects.
308 *
309 * NOTE: It has to be possible to lay out the bitmap in memory
310 * using only allocations of order 0. Additionally, the bitmap is
311 * designed to work with arbitrary number of zones (this is over the
312 * top for now, but let's avoid making unnecessary assumptions ;-).
313 *
314 * struct zone_bitmap contains a pointer to a list of bitmap block
315 * objects and a pointer to the bitmap block object that has been
316 * most recently used for setting bits. Additionally, it contains the
317 * PFNs that correspond to the start and end of the represented zone.
318 *
319 * struct bm_block contains a pointer to the memory page in which
320 * information is stored (in the form of a block of bitmap)
321 * It also contains the pfns that correspond to the start and end of
322 * the represented memory area.
323 *
324 * The memory bitmap is organized as a radix tree to guarantee fast random
325 * access to the bits. There is one radix tree for each zone (as returned
326 * from create_mem_extents).
327 *
328 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
329 * two linked lists for the nodes of the tree, one for the inner nodes and
330 * one for the leave nodes. The linked leave nodes are used for fast linear
331 * access of the memory bitmap.
332 *
333 * The struct rtree_node represents one node of the radix tree.
334 */
335
336#define BM_END_OF_MAP (~0UL)
337
338#define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
339#define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
340#define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
341
342/*
343 * struct rtree_node is a wrapper struct to link the nodes
344 * of the rtree together for easy linear iteration over
345 * bits and easy freeing
346 */
347struct rtree_node {
348 struct list_head list;
349 unsigned long *data;
350};
351
352/*
353 * struct mem_zone_bm_rtree represents a bitmap used for one
354 * populated memory zone.
355 */
356struct mem_zone_bm_rtree {
357 struct list_head list; /* Link Zones together */
358 struct list_head nodes; /* Radix Tree inner nodes */
359 struct list_head leaves; /* Radix Tree leaves */
360 unsigned long start_pfn; /* Zone start page frame */
361 unsigned long end_pfn; /* Zone end page frame + 1 */
362 struct rtree_node *rtree; /* Radix Tree Root */
363 int levels; /* Number of Radix Tree Levels */
364 unsigned int blocks; /* Number of Bitmap Blocks */
365};
366
367/* strcut bm_position is used for browsing memory bitmaps */
368
369struct bm_position {
370 struct mem_zone_bm_rtree *zone;
371 struct rtree_node *node;
372 unsigned long node_pfn;
373 int node_bit;
374};
375
376struct memory_bitmap {
377 struct list_head zones;
378 struct linked_page *p_list; /* list of pages used to store zone
379 bitmap objects and bitmap block
380 objects */
381 struct bm_position cur; /* most recently used bit position */
382};
383
384/* Functions that operate on memory bitmaps */
385
386#define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
387#if BITS_PER_LONG == 32
388#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
389#else
390#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
391#endif
392#define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
393
394/**
395 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
396 *
397 * This function is used to allocate inner nodes as well as the
398 * leave nodes of the radix tree. It also adds the node to the
399 * corresponding linked list passed in by the *list parameter.
400 */
401static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
402 struct chain_allocator *ca,
403 struct list_head *list)
404{
405 struct rtree_node *node;
406
407 node = chain_alloc(ca, sizeof(struct rtree_node));
408 if (!node)
409 return NULL;
410
411 node->data = get_image_page(gfp_mask, safe_needed);
412 if (!node->data)
413 return NULL;
414
415 list_add_tail(&node->list, list);
416
417 return node;
418}
419
420/**
421 * add_rtree_block - Add a new leave node to the radix tree.
422 *
423 * The leave nodes need to be allocated in order to keep the leaves
424 * linked list in order. This is guaranteed by the zone->blocks
425 * counter.
426 */
427static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
428 int safe_needed, struct chain_allocator *ca)
429{
430 struct rtree_node *node, *block, **dst;
431 unsigned int levels_needed, block_nr;
432 int i;
433
434 block_nr = zone->blocks;
435 levels_needed = 0;
436
437 /* How many levels do we need for this block nr? */
438 while (block_nr) {
439 levels_needed += 1;
440 block_nr >>= BM_RTREE_LEVEL_SHIFT;
441 }
442
443 /* Make sure the rtree has enough levels */
444 for (i = zone->levels; i < levels_needed; i++) {
445 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
446 &zone->nodes);
447 if (!node)
448 return -ENOMEM;
449
450 node->data[0] = (unsigned long)zone->rtree;
451 zone->rtree = node;
452 zone->levels += 1;
453 }
454
455 /* Allocate new block */
456 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
457 if (!block)
458 return -ENOMEM;
459
460 /* Now walk the rtree to insert the block */
461 node = zone->rtree;
462 dst = &zone->rtree;
463 block_nr = zone->blocks;
464 for (i = zone->levels; i > 0; i--) {
465 int index;
466
467 if (!node) {
468 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
469 &zone->nodes);
470 if (!node)
471 return -ENOMEM;
472 *dst = node;
473 }
474
475 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
476 index &= BM_RTREE_LEVEL_MASK;
477 dst = (struct rtree_node **)&((*dst)->data[index]);
478 node = *dst;
479 }
480
481 zone->blocks += 1;
482 *dst = block;
483
484 return 0;
485}
486
487static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
488 int clear_nosave_free);
489
490/**
491 * create_zone_bm_rtree - Create a radix tree for one zone.
492 *
493 * Allocated the mem_zone_bm_rtree structure and initializes it.
494 * This function also allocated and builds the radix tree for the
495 * zone.
496 */
497static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
498 int safe_needed,
499 struct chain_allocator *ca,
500 unsigned long start,
501 unsigned long end)
502{
503 struct mem_zone_bm_rtree *zone;
504 unsigned int i, nr_blocks;
505 unsigned long pages;
506
507 pages = end - start;
508 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
509 if (!zone)
510 return NULL;
511
512 INIT_LIST_HEAD(&zone->nodes);
513 INIT_LIST_HEAD(&zone->leaves);
514 zone->start_pfn = start;
515 zone->end_pfn = end;
516 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
517
518 for (i = 0; i < nr_blocks; i++) {
519 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
520 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
521 return NULL;
522 }
523 }
524
525 return zone;
526}
527
528/**
529 * free_zone_bm_rtree - Free the memory of the radix tree.
530 *
531 * Free all node pages of the radix tree. The mem_zone_bm_rtree
532 * structure itself is not freed here nor are the rtree_node
533 * structs.
534 */
535static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
536 int clear_nosave_free)
537{
538 struct rtree_node *node;
539
540 list_for_each_entry(node, &zone->nodes, list)
541 free_image_page(node->data, clear_nosave_free);
542
543 list_for_each_entry(node, &zone->leaves, list)
544 free_image_page(node->data, clear_nosave_free);
545}
546
547static void memory_bm_position_reset(struct memory_bitmap *bm)
548{
549 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
550 list);
551 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
552 struct rtree_node, list);
553 bm->cur.node_pfn = 0;
554 bm->cur.node_bit = 0;
555}
556
557static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
558
559struct mem_extent {
560 struct list_head hook;
561 unsigned long start;
562 unsigned long end;
563};
564
565/**
566 * free_mem_extents - Free a list of memory extents.
567 * @list: List of extents to free.
568 */
569static void free_mem_extents(struct list_head *list)
570{
571 struct mem_extent *ext, *aux;
572
573 list_for_each_entry_safe(ext, aux, list, hook) {
574 list_del(&ext->hook);
575 kfree(ext);
576 }
577}
578
579/**
580 * create_mem_extents - Create a list of memory extents.
581 * @list: List to put the extents into.
582 * @gfp_mask: Mask to use for memory allocations.
583 *
584 * The extents represent contiguous ranges of PFNs.
585 */
586static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
587{
588 struct zone *zone;
589
590 INIT_LIST_HEAD(list);
591
592 for_each_populated_zone(zone) {
593 unsigned long zone_start, zone_end;
594 struct mem_extent *ext, *cur, *aux;
595
596 zone_start = zone->zone_start_pfn;
597 zone_end = zone_end_pfn(zone);
598
599 list_for_each_entry(ext, list, hook)
600 if (zone_start <= ext->end)
601 break;
602
603 if (&ext->hook == list || zone_end < ext->start) {
604 /* New extent is necessary */
605 struct mem_extent *new_ext;
606
607 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
608 if (!new_ext) {
609 free_mem_extents(list);
610 return -ENOMEM;
611 }
612 new_ext->start = zone_start;
613 new_ext->end = zone_end;
614 list_add_tail(&new_ext->hook, &ext->hook);
615 continue;
616 }
617
618 /* Merge this zone's range of PFNs with the existing one */
619 if (zone_start < ext->start)
620 ext->start = zone_start;
621 if (zone_end > ext->end)
622 ext->end = zone_end;
623
624 /* More merging may be possible */
625 cur = ext;
626 list_for_each_entry_safe_continue(cur, aux, list, hook) {
627 if (zone_end < cur->start)
628 break;
629 if (zone_end < cur->end)
630 ext->end = cur->end;
631 list_del(&cur->hook);
632 kfree(cur);
633 }
634 }
635
636 return 0;
637}
638
639/**
640 * memory_bm_create - Allocate memory for a memory bitmap.
641 */
642static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
643 int safe_needed)
644{
645 struct chain_allocator ca;
646 struct list_head mem_extents;
647 struct mem_extent *ext;
648 int error;
649
650 chain_init(&ca, gfp_mask, safe_needed);
651 INIT_LIST_HEAD(&bm->zones);
652
653 error = create_mem_extents(&mem_extents, gfp_mask);
654 if (error)
655 return error;
656
657 list_for_each_entry(ext, &mem_extents, hook) {
658 struct mem_zone_bm_rtree *zone;
659
660 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
661 ext->start, ext->end);
662 if (!zone) {
663 error = -ENOMEM;
664 goto Error;
665 }
666 list_add_tail(&zone->list, &bm->zones);
667 }
668
669 bm->p_list = ca.chain;
670 memory_bm_position_reset(bm);
671 Exit:
672 free_mem_extents(&mem_extents);
673 return error;
674
675 Error:
676 bm->p_list = ca.chain;
677 memory_bm_free(bm, PG_UNSAFE_CLEAR);
678 goto Exit;
679}
680
681/**
682 * memory_bm_free - Free memory occupied by the memory bitmap.
683 * @bm: Memory bitmap.
684 */
685static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
686{
687 struct mem_zone_bm_rtree *zone;
688
689 list_for_each_entry(zone, &bm->zones, list)
690 free_zone_bm_rtree(zone, clear_nosave_free);
691
692 free_list_of_pages(bm->p_list, clear_nosave_free);
693
694 INIT_LIST_HEAD(&bm->zones);
695}
696
697/**
698 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
699 *
700 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
701 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
702 *
703 * Walk the radix tree to find the page containing the bit that represents @pfn
704 * and return the position of the bit in @addr and @bit_nr.
705 */
706static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
707 void **addr, unsigned int *bit_nr)
708{
709 struct mem_zone_bm_rtree *curr, *zone;
710 struct rtree_node *node;
711 int i, block_nr;
712
713 zone = bm->cur.zone;
714
715 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
716 goto zone_found;
717
718 zone = NULL;
719
720 /* Find the right zone */
721 list_for_each_entry(curr, &bm->zones, list) {
722 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
723 zone = curr;
724 break;
725 }
726 }
727
728 if (!zone)
729 return -EFAULT;
730
731zone_found:
732 /*
733 * We have found the zone. Now walk the radix tree to find the leaf node
734 * for our PFN.
735 */
736
737 /*
738 * If the zone we wish to scan is the the current zone and the
739 * pfn falls into the current node then we do not need to walk
740 * the tree.
741 */
742 node = bm->cur.node;
743 if (zone == bm->cur.zone &&
744 ((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
745 goto node_found;
746
747 node = zone->rtree;
748 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
749
750 for (i = zone->levels; i > 0; i--) {
751 int index;
752
753 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
754 index &= BM_RTREE_LEVEL_MASK;
755 BUG_ON(node->data[index] == 0);
756 node = (struct rtree_node *)node->data[index];
757 }
758
759node_found:
760 /* Update last position */
761 bm->cur.zone = zone;
762 bm->cur.node = node;
763 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
764
765 /* Set return values */
766 *addr = node->data;
767 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
768
769 return 0;
770}
771
772static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
773{
774 void *addr;
775 unsigned int bit;
776 int error;
777
778 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
779 BUG_ON(error);
780 set_bit(bit, addr);
781}
782
783static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
784{
785 void *addr;
786 unsigned int bit;
787 int error;
788
789 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
790 if (!error)
791 set_bit(bit, addr);
792
793 return error;
794}
795
796static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
797{
798 void *addr;
799 unsigned int bit;
800 int error;
801
802 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
803 BUG_ON(error);
804 clear_bit(bit, addr);
805}
806
807static void memory_bm_clear_current(struct memory_bitmap *bm)
808{
809 int bit;
810
811 bit = max(bm->cur.node_bit - 1, 0);
812 clear_bit(bit, bm->cur.node->data);
813}
814
815static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
816{
817 void *addr;
818 unsigned int bit;
819 int error;
820
821 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
822 BUG_ON(error);
823 return test_bit(bit, addr);
824}
825
826static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
827{
828 void *addr;
829 unsigned int bit;
830
831 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
832}
833
834/*
835 * rtree_next_node - Jump to the next leaf node.
836 *
837 * Set the position to the beginning of the next node in the
838 * memory bitmap. This is either the next node in the current
839 * zone's radix tree or the first node in the radix tree of the
840 * next zone.
841 *
842 * Return true if there is a next node, false otherwise.
843 */
844static bool rtree_next_node(struct memory_bitmap *bm)
845{
846 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
847 bm->cur.node = list_entry(bm->cur.node->list.next,
848 struct rtree_node, list);
849 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
850 bm->cur.node_bit = 0;
851 touch_softlockup_watchdog();
852 return true;
853 }
854
855 /* No more nodes, goto next zone */
856 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
857 bm->cur.zone = list_entry(bm->cur.zone->list.next,
858 struct mem_zone_bm_rtree, list);
859 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
860 struct rtree_node, list);
861 bm->cur.node_pfn = 0;
862 bm->cur.node_bit = 0;
863 return true;
864 }
865
866 /* No more zones */
867 return false;
868}
869
870/**
871 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
872 * @bm: Memory bitmap.
873 *
874 * Starting from the last returned position this function searches for the next
875 * set bit in @bm and returns the PFN represented by it. If no more bits are
876 * set, BM_END_OF_MAP is returned.
877 *
878 * It is required to run memory_bm_position_reset() before the first call to
879 * this function for the given memory bitmap.
880 */
881static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
882{
883 unsigned long bits, pfn, pages;
884 int bit;
885
886 do {
887 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
888 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
889 bit = find_next_bit(bm->cur.node->data, bits,
890 bm->cur.node_bit);
891 if (bit < bits) {
892 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
893 bm->cur.node_bit = bit + 1;
894 return pfn;
895 }
896 } while (rtree_next_node(bm));
897
898 return BM_END_OF_MAP;
899}
900
901/*
902 * This structure represents a range of page frames the contents of which
903 * should not be saved during hibernation.
904 */
905struct nosave_region {
906 struct list_head list;
907 unsigned long start_pfn;
908 unsigned long end_pfn;
909};
910
911static LIST_HEAD(nosave_regions);
912
913static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
914{
915 struct rtree_node *node;
916
917 list_for_each_entry(node, &zone->nodes, list)
918 recycle_safe_page(node->data);
919
920 list_for_each_entry(node, &zone->leaves, list)
921 recycle_safe_page(node->data);
922}
923
924static void memory_bm_recycle(struct memory_bitmap *bm)
925{
926 struct mem_zone_bm_rtree *zone;
927 struct linked_page *p_list;
928
929 list_for_each_entry(zone, &bm->zones, list)
930 recycle_zone_bm_rtree(zone);
931
932 p_list = bm->p_list;
933 while (p_list) {
934 struct linked_page *lp = p_list;
935
936 p_list = lp->next;
937 recycle_safe_page(lp);
938 }
939}
940
941/**
942 * register_nosave_region - Register a region of unsaveable memory.
943 *
944 * Register a range of page frames the contents of which should not be saved
945 * during hibernation (to be used in the early initialization code).
946 */
947void __init __register_nosave_region(unsigned long start_pfn,
948 unsigned long end_pfn, int use_kmalloc)
949{
950 struct nosave_region *region;
951
952 if (start_pfn >= end_pfn)
953 return;
954
955 if (!list_empty(&nosave_regions)) {
956 /* Try to extend the previous region (they should be sorted) */
957 region = list_entry(nosave_regions.prev,
958 struct nosave_region, list);
959 if (region->end_pfn == start_pfn) {
960 region->end_pfn = end_pfn;
961 goto Report;
962 }
963 }
964 if (use_kmalloc) {
965 /* During init, this shouldn't fail */
966 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
967 BUG_ON(!region);
968 } else {
969 /* This allocation cannot fail */
970 region = memblock_alloc(sizeof(struct nosave_region),
971 SMP_CACHE_BYTES);
972 if (!region)
973 panic("%s: Failed to allocate %zu bytes\n", __func__,
974 sizeof(struct nosave_region));
975 }
976 region->start_pfn = start_pfn;
977 region->end_pfn = end_pfn;
978 list_add_tail(®ion->list, &nosave_regions);
979 Report:
980 pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
981 (unsigned long long) start_pfn << PAGE_SHIFT,
982 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
983}
984
985/*
986 * Set bits in this map correspond to the page frames the contents of which
987 * should not be saved during the suspend.
988 */
989static struct memory_bitmap *forbidden_pages_map;
990
991/* Set bits in this map correspond to free page frames. */
992static struct memory_bitmap *free_pages_map;
993
994/*
995 * Each page frame allocated for creating the image is marked by setting the
996 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
997 */
998
999void swsusp_set_page_free(struct page *page)
1000{
1001 if (free_pages_map)
1002 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
1003}
1004
1005static int swsusp_page_is_free(struct page *page)
1006{
1007 return free_pages_map ?
1008 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1009}
1010
1011void swsusp_unset_page_free(struct page *page)
1012{
1013 if (free_pages_map)
1014 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1015}
1016
1017static void swsusp_set_page_forbidden(struct page *page)
1018{
1019 if (forbidden_pages_map)
1020 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1021}
1022
1023int swsusp_page_is_forbidden(struct page *page)
1024{
1025 return forbidden_pages_map ?
1026 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1027}
1028
1029static void swsusp_unset_page_forbidden(struct page *page)
1030{
1031 if (forbidden_pages_map)
1032 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1033}
1034
1035/**
1036 * mark_nosave_pages - Mark pages that should not be saved.
1037 * @bm: Memory bitmap.
1038 *
1039 * Set the bits in @bm that correspond to the page frames the contents of which
1040 * should not be saved.
1041 */
1042static void mark_nosave_pages(struct memory_bitmap *bm)
1043{
1044 struct nosave_region *region;
1045
1046 if (list_empty(&nosave_regions))
1047 return;
1048
1049 list_for_each_entry(region, &nosave_regions, list) {
1050 unsigned long pfn;
1051
1052 pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1053 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1054 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1055 - 1);
1056
1057 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1058 if (pfn_valid(pfn)) {
1059 /*
1060 * It is safe to ignore the result of
1061 * mem_bm_set_bit_check() here, since we won't
1062 * touch the PFNs for which the error is
1063 * returned anyway.
1064 */
1065 mem_bm_set_bit_check(bm, pfn);
1066 }
1067 }
1068}
1069
1070/**
1071 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1072 *
1073 * Create bitmaps needed for marking page frames that should not be saved and
1074 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1075 * only modified if everything goes well, because we don't want the bits to be
1076 * touched before both bitmaps are set up.
1077 */
1078int create_basic_memory_bitmaps(void)
1079{
1080 struct memory_bitmap *bm1, *bm2;
1081 int error = 0;
1082
1083 if (forbidden_pages_map && free_pages_map)
1084 return 0;
1085 else
1086 BUG_ON(forbidden_pages_map || free_pages_map);
1087
1088 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1089 if (!bm1)
1090 return -ENOMEM;
1091
1092 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1093 if (error)
1094 goto Free_first_object;
1095
1096 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1097 if (!bm2)
1098 goto Free_first_bitmap;
1099
1100 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1101 if (error)
1102 goto Free_second_object;
1103
1104 forbidden_pages_map = bm1;
1105 free_pages_map = bm2;
1106 mark_nosave_pages(forbidden_pages_map);
1107
1108 pr_debug("Basic memory bitmaps created\n");
1109
1110 return 0;
1111
1112 Free_second_object:
1113 kfree(bm2);
1114 Free_first_bitmap:
1115 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1116 Free_first_object:
1117 kfree(bm1);
1118 return -ENOMEM;
1119}
1120
1121/**
1122 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1123 *
1124 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1125 * auxiliary pointers are necessary so that the bitmaps themselves are not
1126 * referred to while they are being freed.
1127 */
1128void free_basic_memory_bitmaps(void)
1129{
1130 struct memory_bitmap *bm1, *bm2;
1131
1132 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1133 return;
1134
1135 bm1 = forbidden_pages_map;
1136 bm2 = free_pages_map;
1137 forbidden_pages_map = NULL;
1138 free_pages_map = NULL;
1139 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1140 kfree(bm1);
1141 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1142 kfree(bm2);
1143
1144 pr_debug("Basic memory bitmaps freed\n");
1145}
1146
1147void clear_free_pages(void)
1148{
1149 struct memory_bitmap *bm = free_pages_map;
1150 unsigned long pfn;
1151
1152 if (WARN_ON(!(free_pages_map)))
1153 return;
1154
1155 if (IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) || want_init_on_free()) {
1156 memory_bm_position_reset(bm);
1157 pfn = memory_bm_next_pfn(bm);
1158 while (pfn != BM_END_OF_MAP) {
1159 if (pfn_valid(pfn))
1160 clear_highpage(pfn_to_page(pfn));
1161
1162 pfn = memory_bm_next_pfn(bm);
1163 }
1164 memory_bm_position_reset(bm);
1165 pr_info("free pages cleared after restore\n");
1166 }
1167}
1168
1169/**
1170 * snapshot_additional_pages - Estimate the number of extra pages needed.
1171 * @zone: Memory zone to carry out the computation for.
1172 *
1173 * Estimate the number of additional pages needed for setting up a hibernation
1174 * image data structures for @zone (usually, the returned value is greater than
1175 * the exact number).
1176 */
1177unsigned int snapshot_additional_pages(struct zone *zone)
1178{
1179 unsigned int rtree, nodes;
1180
1181 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1182 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1183 LINKED_PAGE_DATA_SIZE);
1184 while (nodes > 1) {
1185 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1186 rtree += nodes;
1187 }
1188
1189 return 2 * rtree;
1190}
1191
1192#ifdef CONFIG_HIGHMEM
1193/**
1194 * count_free_highmem_pages - Compute the total number of free highmem pages.
1195 *
1196 * The returned number is system-wide.
1197 */
1198static unsigned int count_free_highmem_pages(void)
1199{
1200 struct zone *zone;
1201 unsigned int cnt = 0;
1202
1203 for_each_populated_zone(zone)
1204 if (is_highmem(zone))
1205 cnt += zone_page_state(zone, NR_FREE_PAGES);
1206
1207 return cnt;
1208}
1209
1210/**
1211 * saveable_highmem_page - Check if a highmem page is saveable.
1212 *
1213 * Determine whether a highmem page should be included in a hibernation image.
1214 *
1215 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1216 * and it isn't part of a free chunk of pages.
1217 */
1218static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1219{
1220 struct page *page;
1221
1222 if (!pfn_valid(pfn))
1223 return NULL;
1224
1225 page = pfn_to_online_page(pfn);
1226 if (!page || page_zone(page) != zone)
1227 return NULL;
1228
1229 BUG_ON(!PageHighMem(page));
1230
1231 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1232 return NULL;
1233
1234 if (PageReserved(page) || PageOffline(page))
1235 return NULL;
1236
1237 if (page_is_guard(page))
1238 return NULL;
1239
1240 return page;
1241}
1242
1243/**
1244 * count_highmem_pages - Compute the total number of saveable highmem pages.
1245 */
1246static unsigned int count_highmem_pages(void)
1247{
1248 struct zone *zone;
1249 unsigned int n = 0;
1250
1251 for_each_populated_zone(zone) {
1252 unsigned long pfn, max_zone_pfn;
1253
1254 if (!is_highmem(zone))
1255 continue;
1256
1257 mark_free_pages(zone);
1258 max_zone_pfn = zone_end_pfn(zone);
1259 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1260 if (saveable_highmem_page(zone, pfn))
1261 n++;
1262 }
1263 return n;
1264}
1265#else
1266static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1267{
1268 return NULL;
1269}
1270#endif /* CONFIG_HIGHMEM */
1271
1272/**
1273 * saveable_page - Check if the given page is saveable.
1274 *
1275 * Determine whether a non-highmem page should be included in a hibernation
1276 * image.
1277 *
1278 * We should save the page if it isn't Nosave, and is not in the range
1279 * of pages statically defined as 'unsaveable', and it isn't part of
1280 * a free chunk of pages.
1281 */
1282static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1283{
1284 struct page *page;
1285
1286 if (!pfn_valid(pfn))
1287 return NULL;
1288
1289 page = pfn_to_online_page(pfn);
1290 if (!page || page_zone(page) != zone)
1291 return NULL;
1292
1293 BUG_ON(PageHighMem(page));
1294
1295 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1296 return NULL;
1297
1298 if (PageOffline(page))
1299 return NULL;
1300
1301 if (PageReserved(page)
1302 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1303 return NULL;
1304
1305 if (page_is_guard(page))
1306 return NULL;
1307
1308 return page;
1309}
1310
1311/**
1312 * count_data_pages - Compute the total number of saveable non-highmem pages.
1313 */
1314static unsigned int count_data_pages(void)
1315{
1316 struct zone *zone;
1317 unsigned long pfn, max_zone_pfn;
1318 unsigned int n = 0;
1319
1320 for_each_populated_zone(zone) {
1321 if (is_highmem(zone))
1322 continue;
1323
1324 mark_free_pages(zone);
1325 max_zone_pfn = zone_end_pfn(zone);
1326 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1327 if (saveable_page(zone, pfn))
1328 n++;
1329 }
1330 return n;
1331}
1332
1333/*
1334 * This is needed, because copy_page and memcpy are not usable for copying
1335 * task structs.
1336 */
1337static inline void do_copy_page(long *dst, long *src)
1338{
1339 int n;
1340
1341 for (n = PAGE_SIZE / sizeof(long); n; n--)
1342 *dst++ = *src++;
1343}
1344
1345/**
1346 * safe_copy_page - Copy a page in a safe way.
1347 *
1348 * Check if the page we are going to copy is marked as present in the kernel
1349 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1350 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1351 * always returns 'true'.
1352 */
1353static void safe_copy_page(void *dst, struct page *s_page)
1354{
1355 if (kernel_page_present(s_page)) {
1356 do_copy_page(dst, page_address(s_page));
1357 } else {
1358 kernel_map_pages(s_page, 1, 1);
1359 do_copy_page(dst, page_address(s_page));
1360 kernel_map_pages(s_page, 1, 0);
1361 }
1362}
1363
1364#ifdef CONFIG_HIGHMEM
1365static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1366{
1367 return is_highmem(zone) ?
1368 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1369}
1370
1371static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1372{
1373 struct page *s_page, *d_page;
1374 void *src, *dst;
1375
1376 s_page = pfn_to_page(src_pfn);
1377 d_page = pfn_to_page(dst_pfn);
1378 if (PageHighMem(s_page)) {
1379 src = kmap_atomic(s_page);
1380 dst = kmap_atomic(d_page);
1381 do_copy_page(dst, src);
1382 kunmap_atomic(dst);
1383 kunmap_atomic(src);
1384 } else {
1385 if (PageHighMem(d_page)) {
1386 /*
1387 * The page pointed to by src may contain some kernel
1388 * data modified by kmap_atomic()
1389 */
1390 safe_copy_page(buffer, s_page);
1391 dst = kmap_atomic(d_page);
1392 copy_page(dst, buffer);
1393 kunmap_atomic(dst);
1394 } else {
1395 safe_copy_page(page_address(d_page), s_page);
1396 }
1397 }
1398}
1399#else
1400#define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1401
1402static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1403{
1404 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1405 pfn_to_page(src_pfn));
1406}
1407#endif /* CONFIG_HIGHMEM */
1408
1409static void copy_data_pages(struct memory_bitmap *copy_bm,
1410 struct memory_bitmap *orig_bm)
1411{
1412 struct zone *zone;
1413 unsigned long pfn;
1414
1415 for_each_populated_zone(zone) {
1416 unsigned long max_zone_pfn;
1417
1418 mark_free_pages(zone);
1419 max_zone_pfn = zone_end_pfn(zone);
1420 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1421 if (page_is_saveable(zone, pfn))
1422 memory_bm_set_bit(orig_bm, pfn);
1423 }
1424 memory_bm_position_reset(orig_bm);
1425 memory_bm_position_reset(copy_bm);
1426 for(;;) {
1427 pfn = memory_bm_next_pfn(orig_bm);
1428 if (unlikely(pfn == BM_END_OF_MAP))
1429 break;
1430 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1431 }
1432}
1433
1434/* Total number of image pages */
1435static unsigned int nr_copy_pages;
1436/* Number of pages needed for saving the original pfns of the image pages */
1437static unsigned int nr_meta_pages;
1438/*
1439 * Numbers of normal and highmem page frames allocated for hibernation image
1440 * before suspending devices.
1441 */
1442static unsigned int alloc_normal, alloc_highmem;
1443/*
1444 * Memory bitmap used for marking saveable pages (during hibernation) or
1445 * hibernation image pages (during restore)
1446 */
1447static struct memory_bitmap orig_bm;
1448/*
1449 * Memory bitmap used during hibernation for marking allocated page frames that
1450 * will contain copies of saveable pages. During restore it is initially used
1451 * for marking hibernation image pages, but then the set bits from it are
1452 * duplicated in @orig_bm and it is released. On highmem systems it is next
1453 * used for marking "safe" highmem pages, but it has to be reinitialized for
1454 * this purpose.
1455 */
1456static struct memory_bitmap copy_bm;
1457
1458/**
1459 * swsusp_free - Free pages allocated for hibernation image.
1460 *
1461 * Image pages are alocated before snapshot creation, so they need to be
1462 * released after resume.
1463 */
1464void swsusp_free(void)
1465{
1466 unsigned long fb_pfn, fr_pfn;
1467
1468 if (!forbidden_pages_map || !free_pages_map)
1469 goto out;
1470
1471 memory_bm_position_reset(forbidden_pages_map);
1472 memory_bm_position_reset(free_pages_map);
1473
1474loop:
1475 fr_pfn = memory_bm_next_pfn(free_pages_map);
1476 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1477
1478 /*
1479 * Find the next bit set in both bitmaps. This is guaranteed to
1480 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1481 */
1482 do {
1483 if (fb_pfn < fr_pfn)
1484 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1485 if (fr_pfn < fb_pfn)
1486 fr_pfn = memory_bm_next_pfn(free_pages_map);
1487 } while (fb_pfn != fr_pfn);
1488
1489 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1490 struct page *page = pfn_to_page(fr_pfn);
1491
1492 memory_bm_clear_current(forbidden_pages_map);
1493 memory_bm_clear_current(free_pages_map);
1494 hibernate_restore_unprotect_page(page_address(page));
1495 __free_page(page);
1496 goto loop;
1497 }
1498
1499out:
1500 nr_copy_pages = 0;
1501 nr_meta_pages = 0;
1502 restore_pblist = NULL;
1503 buffer = NULL;
1504 alloc_normal = 0;
1505 alloc_highmem = 0;
1506 hibernate_restore_protection_end();
1507}
1508
1509/* Helper functions used for the shrinking of memory. */
1510
1511#define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1512
1513/**
1514 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1515 * @nr_pages: Number of page frames to allocate.
1516 * @mask: GFP flags to use for the allocation.
1517 *
1518 * Return value: Number of page frames actually allocated
1519 */
1520static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1521{
1522 unsigned long nr_alloc = 0;
1523
1524 while (nr_pages > 0) {
1525 struct page *page;
1526
1527 page = alloc_image_page(mask);
1528 if (!page)
1529 break;
1530 memory_bm_set_bit(©_bm, page_to_pfn(page));
1531 if (PageHighMem(page))
1532 alloc_highmem++;
1533 else
1534 alloc_normal++;
1535 nr_pages--;
1536 nr_alloc++;
1537 }
1538
1539 return nr_alloc;
1540}
1541
1542static unsigned long preallocate_image_memory(unsigned long nr_pages,
1543 unsigned long avail_normal)
1544{
1545 unsigned long alloc;
1546
1547 if (avail_normal <= alloc_normal)
1548 return 0;
1549
1550 alloc = avail_normal - alloc_normal;
1551 if (nr_pages < alloc)
1552 alloc = nr_pages;
1553
1554 return preallocate_image_pages(alloc, GFP_IMAGE);
1555}
1556
1557#ifdef CONFIG_HIGHMEM
1558static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1559{
1560 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1561}
1562
1563/**
1564 * __fraction - Compute (an approximation of) x * (multiplier / base).
1565 */
1566static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1567{
1568 return div64_u64(x * multiplier, base);
1569}
1570
1571static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1572 unsigned long highmem,
1573 unsigned long total)
1574{
1575 unsigned long alloc = __fraction(nr_pages, highmem, total);
1576
1577 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1578}
1579#else /* CONFIG_HIGHMEM */
1580static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1581{
1582 return 0;
1583}
1584
1585static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1586 unsigned long highmem,
1587 unsigned long total)
1588{
1589 return 0;
1590}
1591#endif /* CONFIG_HIGHMEM */
1592
1593/**
1594 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1595 */
1596static unsigned long free_unnecessary_pages(void)
1597{
1598 unsigned long save, to_free_normal, to_free_highmem, free;
1599
1600 save = count_data_pages();
1601 if (alloc_normal >= save) {
1602 to_free_normal = alloc_normal - save;
1603 save = 0;
1604 } else {
1605 to_free_normal = 0;
1606 save -= alloc_normal;
1607 }
1608 save += count_highmem_pages();
1609 if (alloc_highmem >= save) {
1610 to_free_highmem = alloc_highmem - save;
1611 } else {
1612 to_free_highmem = 0;
1613 save -= alloc_highmem;
1614 if (to_free_normal > save)
1615 to_free_normal -= save;
1616 else
1617 to_free_normal = 0;
1618 }
1619 free = to_free_normal + to_free_highmem;
1620
1621 memory_bm_position_reset(©_bm);
1622
1623 while (to_free_normal > 0 || to_free_highmem > 0) {
1624 unsigned long pfn = memory_bm_next_pfn(©_bm);
1625 struct page *page = pfn_to_page(pfn);
1626
1627 if (PageHighMem(page)) {
1628 if (!to_free_highmem)
1629 continue;
1630 to_free_highmem--;
1631 alloc_highmem--;
1632 } else {
1633 if (!to_free_normal)
1634 continue;
1635 to_free_normal--;
1636 alloc_normal--;
1637 }
1638 memory_bm_clear_bit(©_bm, pfn);
1639 swsusp_unset_page_forbidden(page);
1640 swsusp_unset_page_free(page);
1641 __free_page(page);
1642 }
1643
1644 return free;
1645}
1646
1647/**
1648 * minimum_image_size - Estimate the minimum acceptable size of an image.
1649 * @saveable: Number of saveable pages in the system.
1650 *
1651 * We want to avoid attempting to free too much memory too hard, so estimate the
1652 * minimum acceptable size of a hibernation image to use as the lower limit for
1653 * preallocating memory.
1654 *
1655 * We assume that the minimum image size should be proportional to
1656 *
1657 * [number of saveable pages] - [number of pages that can be freed in theory]
1658 *
1659 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1660 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1661 */
1662static unsigned long minimum_image_size(unsigned long saveable)
1663{
1664 unsigned long size;
1665
1666 size = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B)
1667 + global_node_page_state(NR_ACTIVE_ANON)
1668 + global_node_page_state(NR_INACTIVE_ANON)
1669 + global_node_page_state(NR_ACTIVE_FILE)
1670 + global_node_page_state(NR_INACTIVE_FILE);
1671
1672 return saveable <= size ? 0 : saveable - size;
1673}
1674
1675/**
1676 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1677 *
1678 * To create a hibernation image it is necessary to make a copy of every page
1679 * frame in use. We also need a number of page frames to be free during
1680 * hibernation for allocations made while saving the image and for device
1681 * drivers, in case they need to allocate memory from their hibernation
1682 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1683 * estimate) and reserved_size divided by PAGE_SIZE (which is tunable through
1684 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1685 * total number of available page frames and allocate at least
1686 *
1687 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1688 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1689 *
1690 * of them, which corresponds to the maximum size of a hibernation image.
1691 *
1692 * If image_size is set below the number following from the above formula,
1693 * the preallocation of memory is continued until the total number of saveable
1694 * pages in the system is below the requested image size or the minimum
1695 * acceptable image size returned by minimum_image_size(), whichever is greater.
1696 */
1697int hibernate_preallocate_memory(void)
1698{
1699 struct zone *zone;
1700 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1701 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1702 ktime_t start, stop;
1703 int error;
1704
1705 pr_info("Preallocating image memory\n");
1706 start = ktime_get();
1707
1708 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1709 if (error) {
1710 pr_err("Cannot allocate original bitmap\n");
1711 goto err_out;
1712 }
1713
1714 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1715 if (error) {
1716 pr_err("Cannot allocate copy bitmap\n");
1717 goto err_out;
1718 }
1719
1720 alloc_normal = 0;
1721 alloc_highmem = 0;
1722
1723 /* Count the number of saveable data pages. */
1724 save_highmem = count_highmem_pages();
1725 saveable = count_data_pages();
1726
1727 /*
1728 * Compute the total number of page frames we can use (count) and the
1729 * number of pages needed for image metadata (size).
1730 */
1731 count = saveable;
1732 saveable += save_highmem;
1733 highmem = save_highmem;
1734 size = 0;
1735 for_each_populated_zone(zone) {
1736 size += snapshot_additional_pages(zone);
1737 if (is_highmem(zone))
1738 highmem += zone_page_state(zone, NR_FREE_PAGES);
1739 else
1740 count += zone_page_state(zone, NR_FREE_PAGES);
1741 }
1742 avail_normal = count;
1743 count += highmem;
1744 count -= totalreserve_pages;
1745
1746 /* Compute the maximum number of saveable pages to leave in memory. */
1747 max_size = (count - (size + PAGES_FOR_IO)) / 2
1748 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1749 /* Compute the desired number of image pages specified by image_size. */
1750 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1751 if (size > max_size)
1752 size = max_size;
1753 /*
1754 * If the desired number of image pages is at least as large as the
1755 * current number of saveable pages in memory, allocate page frames for
1756 * the image and we're done.
1757 */
1758 if (size >= saveable) {
1759 pages = preallocate_image_highmem(save_highmem);
1760 pages += preallocate_image_memory(saveable - pages, avail_normal);
1761 goto out;
1762 }
1763
1764 /* Estimate the minimum size of the image. */
1765 pages = minimum_image_size(saveable);
1766 /*
1767 * To avoid excessive pressure on the normal zone, leave room in it to
1768 * accommodate an image of the minimum size (unless it's already too
1769 * small, in which case don't preallocate pages from it at all).
1770 */
1771 if (avail_normal > pages)
1772 avail_normal -= pages;
1773 else
1774 avail_normal = 0;
1775 if (size < pages)
1776 size = min_t(unsigned long, pages, max_size);
1777
1778 /*
1779 * Let the memory management subsystem know that we're going to need a
1780 * large number of page frames to allocate and make it free some memory.
1781 * NOTE: If this is not done, performance will be hurt badly in some
1782 * test cases.
1783 */
1784 shrink_all_memory(saveable - size);
1785
1786 /*
1787 * The number of saveable pages in memory was too high, so apply some
1788 * pressure to decrease it. First, make room for the largest possible
1789 * image and fail if that doesn't work. Next, try to decrease the size
1790 * of the image as much as indicated by 'size' using allocations from
1791 * highmem and non-highmem zones separately.
1792 */
1793 pages_highmem = preallocate_image_highmem(highmem / 2);
1794 alloc = count - max_size;
1795 if (alloc > pages_highmem)
1796 alloc -= pages_highmem;
1797 else
1798 alloc = 0;
1799 pages = preallocate_image_memory(alloc, avail_normal);
1800 if (pages < alloc) {
1801 /* We have exhausted non-highmem pages, try highmem. */
1802 alloc -= pages;
1803 pages += pages_highmem;
1804 pages_highmem = preallocate_image_highmem(alloc);
1805 if (pages_highmem < alloc) {
1806 pr_err("Image allocation is %lu pages short\n",
1807 alloc - pages_highmem);
1808 goto err_out;
1809 }
1810 pages += pages_highmem;
1811 /*
1812 * size is the desired number of saveable pages to leave in
1813 * memory, so try to preallocate (all memory - size) pages.
1814 */
1815 alloc = (count - pages) - size;
1816 pages += preallocate_image_highmem(alloc);
1817 } else {
1818 /*
1819 * There are approximately max_size saveable pages at this point
1820 * and we want to reduce this number down to size.
1821 */
1822 alloc = max_size - size;
1823 size = preallocate_highmem_fraction(alloc, highmem, count);
1824 pages_highmem += size;
1825 alloc -= size;
1826 size = preallocate_image_memory(alloc, avail_normal);
1827 pages_highmem += preallocate_image_highmem(alloc - size);
1828 pages += pages_highmem + size;
1829 }
1830
1831 /*
1832 * We only need as many page frames for the image as there are saveable
1833 * pages in memory, but we have allocated more. Release the excessive
1834 * ones now.
1835 */
1836 pages -= free_unnecessary_pages();
1837
1838 out:
1839 stop = ktime_get();
1840 pr_info("Allocated %lu pages for snapshot\n", pages);
1841 swsusp_show_speed(start, stop, pages, "Allocated");
1842
1843 return 0;
1844
1845 err_out:
1846 swsusp_free();
1847 return -ENOMEM;
1848}
1849
1850#ifdef CONFIG_HIGHMEM
1851/**
1852 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1853 *
1854 * Compute the number of non-highmem pages that will be necessary for creating
1855 * copies of highmem pages.
1856 */
1857static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1858{
1859 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1860
1861 if (free_highmem >= nr_highmem)
1862 nr_highmem = 0;
1863 else
1864 nr_highmem -= free_highmem;
1865
1866 return nr_highmem;
1867}
1868#else
1869static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1870#endif /* CONFIG_HIGHMEM */
1871
1872/**
1873 * enough_free_mem - Check if there is enough free memory for the image.
1874 */
1875static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1876{
1877 struct zone *zone;
1878 unsigned int free = alloc_normal;
1879
1880 for_each_populated_zone(zone)
1881 if (!is_highmem(zone))
1882 free += zone_page_state(zone, NR_FREE_PAGES);
1883
1884 nr_pages += count_pages_for_highmem(nr_highmem);
1885 pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
1886 nr_pages, PAGES_FOR_IO, free);
1887
1888 return free > nr_pages + PAGES_FOR_IO;
1889}
1890
1891#ifdef CONFIG_HIGHMEM
1892/**
1893 * get_highmem_buffer - Allocate a buffer for highmem pages.
1894 *
1895 * If there are some highmem pages in the hibernation image, we may need a
1896 * buffer to copy them and/or load their data.
1897 */
1898static inline int get_highmem_buffer(int safe_needed)
1899{
1900 buffer = get_image_page(GFP_ATOMIC, safe_needed);
1901 return buffer ? 0 : -ENOMEM;
1902}
1903
1904/**
1905 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1906 *
1907 * Try to allocate as many pages as needed, but if the number of free highmem
1908 * pages is less than that, allocate them all.
1909 */
1910static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1911 unsigned int nr_highmem)
1912{
1913 unsigned int to_alloc = count_free_highmem_pages();
1914
1915 if (to_alloc > nr_highmem)
1916 to_alloc = nr_highmem;
1917
1918 nr_highmem -= to_alloc;
1919 while (to_alloc-- > 0) {
1920 struct page *page;
1921
1922 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1923 memory_bm_set_bit(bm, page_to_pfn(page));
1924 }
1925 return nr_highmem;
1926}
1927#else
1928static inline int get_highmem_buffer(int safe_needed) { return 0; }
1929
1930static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1931 unsigned int n) { return 0; }
1932#endif /* CONFIG_HIGHMEM */
1933
1934/**
1935 * swsusp_alloc - Allocate memory for hibernation image.
1936 *
1937 * We first try to allocate as many highmem pages as there are
1938 * saveable highmem pages in the system. If that fails, we allocate
1939 * non-highmem pages for the copies of the remaining highmem ones.
1940 *
1941 * In this approach it is likely that the copies of highmem pages will
1942 * also be located in the high memory, because of the way in which
1943 * copy_data_pages() works.
1944 */
1945static int swsusp_alloc(struct memory_bitmap *copy_bm,
1946 unsigned int nr_pages, unsigned int nr_highmem)
1947{
1948 if (nr_highmem > 0) {
1949 if (get_highmem_buffer(PG_ANY))
1950 goto err_out;
1951 if (nr_highmem > alloc_highmem) {
1952 nr_highmem -= alloc_highmem;
1953 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1954 }
1955 }
1956 if (nr_pages > alloc_normal) {
1957 nr_pages -= alloc_normal;
1958 while (nr_pages-- > 0) {
1959 struct page *page;
1960
1961 page = alloc_image_page(GFP_ATOMIC);
1962 if (!page)
1963 goto err_out;
1964 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1965 }
1966 }
1967
1968 return 0;
1969
1970 err_out:
1971 swsusp_free();
1972 return -ENOMEM;
1973}
1974
1975asmlinkage __visible int swsusp_save(void)
1976{
1977 unsigned int nr_pages, nr_highmem;
1978
1979 pr_info("Creating image:\n");
1980
1981 drain_local_pages(NULL);
1982 nr_pages = count_data_pages();
1983 nr_highmem = count_highmem_pages();
1984 pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
1985
1986 if (!enough_free_mem(nr_pages, nr_highmem)) {
1987 pr_err("Not enough free memory\n");
1988 return -ENOMEM;
1989 }
1990
1991 if (swsusp_alloc(©_bm, nr_pages, nr_highmem)) {
1992 pr_err("Memory allocation failed\n");
1993 return -ENOMEM;
1994 }
1995
1996 /*
1997 * During allocating of suspend pagedir, new cold pages may appear.
1998 * Kill them.
1999 */
2000 drain_local_pages(NULL);
2001 copy_data_pages(©_bm, &orig_bm);
2002
2003 /*
2004 * End of critical section. From now on, we can write to memory,
2005 * but we should not touch disk. This specially means we must _not_
2006 * touch swap space! Except we must write out our image of course.
2007 */
2008
2009 nr_pages += nr_highmem;
2010 nr_copy_pages = nr_pages;
2011 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2012
2013 pr_info("Image created (%d pages copied)\n", nr_pages);
2014
2015 return 0;
2016}
2017
2018#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2019static int init_header_complete(struct swsusp_info *info)
2020{
2021 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2022 info->version_code = LINUX_VERSION_CODE;
2023 return 0;
2024}
2025
2026static const char *check_image_kernel(struct swsusp_info *info)
2027{
2028 if (info->version_code != LINUX_VERSION_CODE)
2029 return "kernel version";
2030 if (strcmp(info->uts.sysname,init_utsname()->sysname))
2031 return "system type";
2032 if (strcmp(info->uts.release,init_utsname()->release))
2033 return "kernel release";
2034 if (strcmp(info->uts.version,init_utsname()->version))
2035 return "version";
2036 if (strcmp(info->uts.machine,init_utsname()->machine))
2037 return "machine";
2038 return NULL;
2039}
2040#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2041
2042unsigned long snapshot_get_image_size(void)
2043{
2044 return nr_copy_pages + nr_meta_pages + 1;
2045}
2046
2047static int init_header(struct swsusp_info *info)
2048{
2049 memset(info, 0, sizeof(struct swsusp_info));
2050 info->num_physpages = get_num_physpages();
2051 info->image_pages = nr_copy_pages;
2052 info->pages = snapshot_get_image_size();
2053 info->size = info->pages;
2054 info->size <<= PAGE_SHIFT;
2055 return init_header_complete(info);
2056}
2057
2058/**
2059 * pack_pfns - Prepare PFNs for saving.
2060 * @bm: Memory bitmap.
2061 * @buf: Memory buffer to store the PFNs in.
2062 *
2063 * PFNs corresponding to set bits in @bm are stored in the area of memory
2064 * pointed to by @buf (1 page at a time).
2065 */
2066static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2067{
2068 int j;
2069
2070 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2071 buf[j] = memory_bm_next_pfn(bm);
2072 if (unlikely(buf[j] == BM_END_OF_MAP))
2073 break;
2074 }
2075}
2076
2077/**
2078 * snapshot_read_next - Get the address to read the next image page from.
2079 * @handle: Snapshot handle to be used for the reading.
2080 *
2081 * On the first call, @handle should point to a zeroed snapshot_handle
2082 * structure. The structure gets populated then and a pointer to it should be
2083 * passed to this function every next time.
2084 *
2085 * On success, the function returns a positive number. Then, the caller
2086 * is allowed to read up to the returned number of bytes from the memory
2087 * location computed by the data_of() macro.
2088 *
2089 * The function returns 0 to indicate the end of the data stream condition,
2090 * and negative numbers are returned on errors. If that happens, the structure
2091 * pointed to by @handle is not updated and should not be used any more.
2092 */
2093int snapshot_read_next(struct snapshot_handle *handle)
2094{
2095 if (handle->cur > nr_meta_pages + nr_copy_pages)
2096 return 0;
2097
2098 if (!buffer) {
2099 /* This makes the buffer be freed by swsusp_free() */
2100 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2101 if (!buffer)
2102 return -ENOMEM;
2103 }
2104 if (!handle->cur) {
2105 int error;
2106
2107 error = init_header((struct swsusp_info *)buffer);
2108 if (error)
2109 return error;
2110 handle->buffer = buffer;
2111 memory_bm_position_reset(&orig_bm);
2112 memory_bm_position_reset(©_bm);
2113 } else if (handle->cur <= nr_meta_pages) {
2114 clear_page(buffer);
2115 pack_pfns(buffer, &orig_bm);
2116 } else {
2117 struct page *page;
2118
2119 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2120 if (PageHighMem(page)) {
2121 /*
2122 * Highmem pages are copied to the buffer,
2123 * because we can't return with a kmapped
2124 * highmem page (we may not be called again).
2125 */
2126 void *kaddr;
2127
2128 kaddr = kmap_atomic(page);
2129 copy_page(buffer, kaddr);
2130 kunmap_atomic(kaddr);
2131 handle->buffer = buffer;
2132 } else {
2133 handle->buffer = page_address(page);
2134 }
2135 }
2136 handle->cur++;
2137 return PAGE_SIZE;
2138}
2139
2140static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2141 struct memory_bitmap *src)
2142{
2143 unsigned long pfn;
2144
2145 memory_bm_position_reset(src);
2146 pfn = memory_bm_next_pfn(src);
2147 while (pfn != BM_END_OF_MAP) {
2148 memory_bm_set_bit(dst, pfn);
2149 pfn = memory_bm_next_pfn(src);
2150 }
2151}
2152
2153/**
2154 * mark_unsafe_pages - Mark pages that were used before hibernation.
2155 *
2156 * Mark the pages that cannot be used for storing the image during restoration,
2157 * because they conflict with the pages that had been used before hibernation.
2158 */
2159static void mark_unsafe_pages(struct memory_bitmap *bm)
2160{
2161 unsigned long pfn;
2162
2163 /* Clear the "free"/"unsafe" bit for all PFNs */
2164 memory_bm_position_reset(free_pages_map);
2165 pfn = memory_bm_next_pfn(free_pages_map);
2166 while (pfn != BM_END_OF_MAP) {
2167 memory_bm_clear_current(free_pages_map);
2168 pfn = memory_bm_next_pfn(free_pages_map);
2169 }
2170
2171 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2172 duplicate_memory_bitmap(free_pages_map, bm);
2173
2174 allocated_unsafe_pages = 0;
2175}
2176
2177static int check_header(struct swsusp_info *info)
2178{
2179 const char *reason;
2180
2181 reason = check_image_kernel(info);
2182 if (!reason && info->num_physpages != get_num_physpages())
2183 reason = "memory size";
2184 if (reason) {
2185 pr_err("Image mismatch: %s\n", reason);
2186 return -EPERM;
2187 }
2188 return 0;
2189}
2190
2191/**
2192 * load header - Check the image header and copy the data from it.
2193 */
2194static int load_header(struct swsusp_info *info)
2195{
2196 int error;
2197
2198 restore_pblist = NULL;
2199 error = check_header(info);
2200 if (!error) {
2201 nr_copy_pages = info->image_pages;
2202 nr_meta_pages = info->pages - info->image_pages - 1;
2203 }
2204 return error;
2205}
2206
2207/**
2208 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2209 * @bm: Memory bitmap.
2210 * @buf: Area of memory containing the PFNs.
2211 *
2212 * For each element of the array pointed to by @buf (1 page at a time), set the
2213 * corresponding bit in @bm.
2214 */
2215static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2216{
2217 int j;
2218
2219 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2220 if (unlikely(buf[j] == BM_END_OF_MAP))
2221 break;
2222
2223 if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2224 memory_bm_set_bit(bm, buf[j]);
2225 else
2226 return -EFAULT;
2227 }
2228
2229 return 0;
2230}
2231
2232#ifdef CONFIG_HIGHMEM
2233/*
2234 * struct highmem_pbe is used for creating the list of highmem pages that
2235 * should be restored atomically during the resume from disk, because the page
2236 * frames they have occupied before the suspend are in use.
2237 */
2238struct highmem_pbe {
2239 struct page *copy_page; /* data is here now */
2240 struct page *orig_page; /* data was here before the suspend */
2241 struct highmem_pbe *next;
2242};
2243
2244/*
2245 * List of highmem PBEs needed for restoring the highmem pages that were
2246 * allocated before the suspend and included in the suspend image, but have
2247 * also been allocated by the "resume" kernel, so their contents cannot be
2248 * written directly to their "original" page frames.
2249 */
2250static struct highmem_pbe *highmem_pblist;
2251
2252/**
2253 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2254 * @bm: Memory bitmap.
2255 *
2256 * The bits in @bm that correspond to image pages are assumed to be set.
2257 */
2258static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2259{
2260 unsigned long pfn;
2261 unsigned int cnt = 0;
2262
2263 memory_bm_position_reset(bm);
2264 pfn = memory_bm_next_pfn(bm);
2265 while (pfn != BM_END_OF_MAP) {
2266 if (PageHighMem(pfn_to_page(pfn)))
2267 cnt++;
2268
2269 pfn = memory_bm_next_pfn(bm);
2270 }
2271 return cnt;
2272}
2273
2274static unsigned int safe_highmem_pages;
2275
2276static struct memory_bitmap *safe_highmem_bm;
2277
2278/**
2279 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2280 * @bm: Pointer to an uninitialized memory bitmap structure.
2281 * @nr_highmem_p: Pointer to the number of highmem image pages.
2282 *
2283 * Try to allocate as many highmem pages as there are highmem image pages
2284 * (@nr_highmem_p points to the variable containing the number of highmem image
2285 * pages). The pages that are "safe" (ie. will not be overwritten when the
2286 * hibernation image is restored entirely) have the corresponding bits set in
2287 * @bm (it must be unitialized).
2288 *
2289 * NOTE: This function should not be called if there are no highmem image pages.
2290 */
2291static int prepare_highmem_image(struct memory_bitmap *bm,
2292 unsigned int *nr_highmem_p)
2293{
2294 unsigned int to_alloc;
2295
2296 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2297 return -ENOMEM;
2298
2299 if (get_highmem_buffer(PG_SAFE))
2300 return -ENOMEM;
2301
2302 to_alloc = count_free_highmem_pages();
2303 if (to_alloc > *nr_highmem_p)
2304 to_alloc = *nr_highmem_p;
2305 else
2306 *nr_highmem_p = to_alloc;
2307
2308 safe_highmem_pages = 0;
2309 while (to_alloc-- > 0) {
2310 struct page *page;
2311
2312 page = alloc_page(__GFP_HIGHMEM);
2313 if (!swsusp_page_is_free(page)) {
2314 /* The page is "safe", set its bit the bitmap */
2315 memory_bm_set_bit(bm, page_to_pfn(page));
2316 safe_highmem_pages++;
2317 }
2318 /* Mark the page as allocated */
2319 swsusp_set_page_forbidden(page);
2320 swsusp_set_page_free(page);
2321 }
2322 memory_bm_position_reset(bm);
2323 safe_highmem_bm = bm;
2324 return 0;
2325}
2326
2327static struct page *last_highmem_page;
2328
2329/**
2330 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2331 *
2332 * For a given highmem image page get a buffer that suspend_write_next() should
2333 * return to its caller to write to.
2334 *
2335 * If the page is to be saved to its "original" page frame or a copy of
2336 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2337 * the copy of the page is to be made in normal memory, so the address of
2338 * the copy is returned.
2339 *
2340 * If @buffer is returned, the caller of suspend_write_next() will write
2341 * the page's contents to @buffer, so they will have to be copied to the
2342 * right location on the next call to suspend_write_next() and it is done
2343 * with the help of copy_last_highmem_page(). For this purpose, if
2344 * @buffer is returned, @last_highmem_page is set to the page to which
2345 * the data will have to be copied from @buffer.
2346 */
2347static void *get_highmem_page_buffer(struct page *page,
2348 struct chain_allocator *ca)
2349{
2350 struct highmem_pbe *pbe;
2351 void *kaddr;
2352
2353 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2354 /*
2355 * We have allocated the "original" page frame and we can
2356 * use it directly to store the loaded page.
2357 */
2358 last_highmem_page = page;
2359 return buffer;
2360 }
2361 /*
2362 * The "original" page frame has not been allocated and we have to
2363 * use a "safe" page frame to store the loaded page.
2364 */
2365 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2366 if (!pbe) {
2367 swsusp_free();
2368 return ERR_PTR(-ENOMEM);
2369 }
2370 pbe->orig_page = page;
2371 if (safe_highmem_pages > 0) {
2372 struct page *tmp;
2373
2374 /* Copy of the page will be stored in high memory */
2375 kaddr = buffer;
2376 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2377 safe_highmem_pages--;
2378 last_highmem_page = tmp;
2379 pbe->copy_page = tmp;
2380 } else {
2381 /* Copy of the page will be stored in normal memory */
2382 kaddr = safe_pages_list;
2383 safe_pages_list = safe_pages_list->next;
2384 pbe->copy_page = virt_to_page(kaddr);
2385 }
2386 pbe->next = highmem_pblist;
2387 highmem_pblist = pbe;
2388 return kaddr;
2389}
2390
2391/**
2392 * copy_last_highmem_page - Copy most the most recent highmem image page.
2393 *
2394 * Copy the contents of a highmem image from @buffer, where the caller of
2395 * snapshot_write_next() has stored them, to the right location represented by
2396 * @last_highmem_page .
2397 */
2398static void copy_last_highmem_page(void)
2399{
2400 if (last_highmem_page) {
2401 void *dst;
2402
2403 dst = kmap_atomic(last_highmem_page);
2404 copy_page(dst, buffer);
2405 kunmap_atomic(dst);
2406 last_highmem_page = NULL;
2407 }
2408}
2409
2410static inline int last_highmem_page_copied(void)
2411{
2412 return !last_highmem_page;
2413}
2414
2415static inline void free_highmem_data(void)
2416{
2417 if (safe_highmem_bm)
2418 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2419
2420 if (buffer)
2421 free_image_page(buffer, PG_UNSAFE_CLEAR);
2422}
2423#else
2424static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2425
2426static inline int prepare_highmem_image(struct memory_bitmap *bm,
2427 unsigned int *nr_highmem_p) { return 0; }
2428
2429static inline void *get_highmem_page_buffer(struct page *page,
2430 struct chain_allocator *ca)
2431{
2432 return ERR_PTR(-EINVAL);
2433}
2434
2435static inline void copy_last_highmem_page(void) {}
2436static inline int last_highmem_page_copied(void) { return 1; }
2437static inline void free_highmem_data(void) {}
2438#endif /* CONFIG_HIGHMEM */
2439
2440#define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2441
2442/**
2443 * prepare_image - Make room for loading hibernation image.
2444 * @new_bm: Unitialized memory bitmap structure.
2445 * @bm: Memory bitmap with unsafe pages marked.
2446 *
2447 * Use @bm to mark the pages that will be overwritten in the process of
2448 * restoring the system memory state from the suspend image ("unsafe" pages)
2449 * and allocate memory for the image.
2450 *
2451 * The idea is to allocate a new memory bitmap first and then allocate
2452 * as many pages as needed for image data, but without specifying what those
2453 * pages will be used for just yet. Instead, we mark them all as allocated and
2454 * create a lists of "safe" pages to be used later. On systems with high
2455 * memory a list of "safe" highmem pages is created too.
2456 */
2457static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2458{
2459 unsigned int nr_pages, nr_highmem;
2460 struct linked_page *lp;
2461 int error;
2462
2463 /* If there is no highmem, the buffer will not be necessary */
2464 free_image_page(buffer, PG_UNSAFE_CLEAR);
2465 buffer = NULL;
2466
2467 nr_highmem = count_highmem_image_pages(bm);
2468 mark_unsafe_pages(bm);
2469
2470 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2471 if (error)
2472 goto Free;
2473
2474 duplicate_memory_bitmap(new_bm, bm);
2475 memory_bm_free(bm, PG_UNSAFE_KEEP);
2476 if (nr_highmem > 0) {
2477 error = prepare_highmem_image(bm, &nr_highmem);
2478 if (error)
2479 goto Free;
2480 }
2481 /*
2482 * Reserve some safe pages for potential later use.
2483 *
2484 * NOTE: This way we make sure there will be enough safe pages for the
2485 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2486 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2487 *
2488 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2489 */
2490 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2491 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2492 while (nr_pages > 0) {
2493 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2494 if (!lp) {
2495 error = -ENOMEM;
2496 goto Free;
2497 }
2498 lp->next = safe_pages_list;
2499 safe_pages_list = lp;
2500 nr_pages--;
2501 }
2502 /* Preallocate memory for the image */
2503 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2504 while (nr_pages > 0) {
2505 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2506 if (!lp) {
2507 error = -ENOMEM;
2508 goto Free;
2509 }
2510 if (!swsusp_page_is_free(virt_to_page(lp))) {
2511 /* The page is "safe", add it to the list */
2512 lp->next = safe_pages_list;
2513 safe_pages_list = lp;
2514 }
2515 /* Mark the page as allocated */
2516 swsusp_set_page_forbidden(virt_to_page(lp));
2517 swsusp_set_page_free(virt_to_page(lp));
2518 nr_pages--;
2519 }
2520 return 0;
2521
2522 Free:
2523 swsusp_free();
2524 return error;
2525}
2526
2527/**
2528 * get_buffer - Get the address to store the next image data page.
2529 *
2530 * Get the address that snapshot_write_next() should return to its caller to
2531 * write to.
2532 */
2533static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2534{
2535 struct pbe *pbe;
2536 struct page *page;
2537 unsigned long pfn = memory_bm_next_pfn(bm);
2538
2539 if (pfn == BM_END_OF_MAP)
2540 return ERR_PTR(-EFAULT);
2541
2542 page = pfn_to_page(pfn);
2543 if (PageHighMem(page))
2544 return get_highmem_page_buffer(page, ca);
2545
2546 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2547 /*
2548 * We have allocated the "original" page frame and we can
2549 * use it directly to store the loaded page.
2550 */
2551 return page_address(page);
2552
2553 /*
2554 * The "original" page frame has not been allocated and we have to
2555 * use a "safe" page frame to store the loaded page.
2556 */
2557 pbe = chain_alloc(ca, sizeof(struct pbe));
2558 if (!pbe) {
2559 swsusp_free();
2560 return ERR_PTR(-ENOMEM);
2561 }
2562 pbe->orig_address = page_address(page);
2563 pbe->address = safe_pages_list;
2564 safe_pages_list = safe_pages_list->next;
2565 pbe->next = restore_pblist;
2566 restore_pblist = pbe;
2567 return pbe->address;
2568}
2569
2570/**
2571 * snapshot_write_next - Get the address to store the next image page.
2572 * @handle: Snapshot handle structure to guide the writing.
2573 *
2574 * On the first call, @handle should point to a zeroed snapshot_handle
2575 * structure. The structure gets populated then and a pointer to it should be
2576 * passed to this function every next time.
2577 *
2578 * On success, the function returns a positive number. Then, the caller
2579 * is allowed to write up to the returned number of bytes to the memory
2580 * location computed by the data_of() macro.
2581 *
2582 * The function returns 0 to indicate the "end of file" condition. Negative
2583 * numbers are returned on errors, in which cases the structure pointed to by
2584 * @handle is not updated and should not be used any more.
2585 */
2586int snapshot_write_next(struct snapshot_handle *handle)
2587{
2588 static struct chain_allocator ca;
2589 int error = 0;
2590
2591 /* Check if we have already loaded the entire image */
2592 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2593 return 0;
2594
2595 handle->sync_read = 1;
2596
2597 if (!handle->cur) {
2598 if (!buffer)
2599 /* This makes the buffer be freed by swsusp_free() */
2600 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2601
2602 if (!buffer)
2603 return -ENOMEM;
2604
2605 handle->buffer = buffer;
2606 } else if (handle->cur == 1) {
2607 error = load_header(buffer);
2608 if (error)
2609 return error;
2610
2611 safe_pages_list = NULL;
2612
2613 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2614 if (error)
2615 return error;
2616
2617 hibernate_restore_protection_begin();
2618 } else if (handle->cur <= nr_meta_pages + 1) {
2619 error = unpack_orig_pfns(buffer, ©_bm);
2620 if (error)
2621 return error;
2622
2623 if (handle->cur == nr_meta_pages + 1) {
2624 error = prepare_image(&orig_bm, ©_bm);
2625 if (error)
2626 return error;
2627
2628 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2629 memory_bm_position_reset(&orig_bm);
2630 restore_pblist = NULL;
2631 handle->buffer = get_buffer(&orig_bm, &ca);
2632 handle->sync_read = 0;
2633 if (IS_ERR(handle->buffer))
2634 return PTR_ERR(handle->buffer);
2635 }
2636 } else {
2637 copy_last_highmem_page();
2638 hibernate_restore_protect_page(handle->buffer);
2639 handle->buffer = get_buffer(&orig_bm, &ca);
2640 if (IS_ERR(handle->buffer))
2641 return PTR_ERR(handle->buffer);
2642 if (handle->buffer != buffer)
2643 handle->sync_read = 0;
2644 }
2645 handle->cur++;
2646 return PAGE_SIZE;
2647}
2648
2649/**
2650 * snapshot_write_finalize - Complete the loading of a hibernation image.
2651 *
2652 * Must be called after the last call to snapshot_write_next() in case the last
2653 * page in the image happens to be a highmem page and its contents should be
2654 * stored in highmem. Additionally, it recycles bitmap memory that's not
2655 * necessary any more.
2656 */
2657void snapshot_write_finalize(struct snapshot_handle *handle)
2658{
2659 copy_last_highmem_page();
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 int __must_check hibernate_restore_protect_page(void *page_address)
62{
63 if (hibernate_restore_protection_active)
64 return set_memory_ro((unsigned long)page_address, 1);
65 return 0;
66}
67
68static inline int hibernate_restore_unprotect_page(void *page_address)
69{
70 if (hibernate_restore_protection_active)
71 return set_memory_rw((unsigned long)page_address, 1);
72 return 0;
73}
74#else
75static inline void hibernate_restore_protection_begin(void) {}
76static inline void hibernate_restore_protection_end(void) {}
77static inline int __must_check hibernate_restore_protect_page(void *page_address) {return 0; }
78static inline int hibernate_restore_unprotect_page(void *page_address) {return 0; }
79#endif /* CONFIG_STRICT_KERNEL_RWX && CONFIG_ARCH_HAS_SET_MEMORY */
80
81
82/*
83 * The calls to set_direct_map_*() should not fail because remapping a page
84 * here means that we only update protection bits in an existing PTE.
85 * It is still worth to have a warning here if something changes and this
86 * will no longer be the case.
87 */
88static inline void hibernate_map_page(struct page *page)
89{
90 if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
91 int ret = set_direct_map_default_noflush(page);
92
93 if (ret)
94 pr_warn_once("Failed to remap page\n");
95 } else {
96 debug_pagealloc_map_pages(page, 1);
97 }
98}
99
100static inline void hibernate_unmap_page(struct page *page)
101{
102 if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
103 unsigned long addr = (unsigned long)page_address(page);
104 int ret = set_direct_map_invalid_noflush(page);
105
106 if (ret)
107 pr_warn_once("Failed to remap page\n");
108
109 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
110 } else {
111 debug_pagealloc_unmap_pages(page, 1);
112 }
113}
114
115static int swsusp_page_is_free(struct page *);
116static void swsusp_set_page_forbidden(struct page *);
117static void swsusp_unset_page_forbidden(struct page *);
118
119/*
120 * Number of bytes to reserve for memory allocations made by device drivers
121 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
122 * cause image creation to fail (tunable via /sys/power/reserved_size).
123 */
124unsigned long reserved_size;
125
126void __init hibernate_reserved_size_init(void)
127{
128 reserved_size = SPARE_PAGES * PAGE_SIZE;
129}
130
131/*
132 * Preferred image size in bytes (tunable via /sys/power/image_size).
133 * When it is set to N, swsusp will do its best to ensure the image
134 * size will not exceed N bytes, but if that is impossible, it will
135 * try to create the smallest image possible.
136 */
137unsigned long image_size;
138
139void __init hibernate_image_size_init(void)
140{
141 image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
142}
143
144/*
145 * List of PBEs needed for restoring the pages that were allocated before
146 * the suspend and included in the suspend image, but have also been
147 * allocated by the "resume" kernel, so their contents cannot be written
148 * directly to their "original" page frames.
149 */
150struct pbe *restore_pblist;
151
152/* struct linked_page is used to build chains of pages */
153
154#define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
155
156struct linked_page {
157 struct linked_page *next;
158 char data[LINKED_PAGE_DATA_SIZE];
159} __packed;
160
161/*
162 * List of "safe" pages (ie. pages that were not used by the image kernel
163 * before hibernation) that may be used as temporary storage for image kernel
164 * memory contents.
165 */
166static struct linked_page *safe_pages_list;
167
168/* Pointer to an auxiliary buffer (1 page) */
169static void *buffer;
170
171#define PG_ANY 0
172#define PG_SAFE 1
173#define PG_UNSAFE_CLEAR 1
174#define PG_UNSAFE_KEEP 0
175
176static unsigned int allocated_unsafe_pages;
177
178/**
179 * get_image_page - Allocate a page for a hibernation image.
180 * @gfp_mask: GFP mask for the allocation.
181 * @safe_needed: Get pages that were not used before hibernation (restore only)
182 *
183 * During image restoration, for storing the PBE list and the image data, we can
184 * only use memory pages that do not conflict with the pages used before
185 * hibernation. The "unsafe" pages have PageNosaveFree set and we count them
186 * using allocated_unsafe_pages.
187 *
188 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
189 * swsusp_free() can release it.
190 */
191static void *get_image_page(gfp_t gfp_mask, int safe_needed)
192{
193 void *res;
194
195 res = (void *)get_zeroed_page(gfp_mask);
196 if (safe_needed)
197 while (res && swsusp_page_is_free(virt_to_page(res))) {
198 /* The page is unsafe, mark it for swsusp_free() */
199 swsusp_set_page_forbidden(virt_to_page(res));
200 allocated_unsafe_pages++;
201 res = (void *)get_zeroed_page(gfp_mask);
202 }
203 if (res) {
204 swsusp_set_page_forbidden(virt_to_page(res));
205 swsusp_set_page_free(virt_to_page(res));
206 }
207 return res;
208}
209
210static void *__get_safe_page(gfp_t gfp_mask)
211{
212 if (safe_pages_list) {
213 void *ret = safe_pages_list;
214
215 safe_pages_list = safe_pages_list->next;
216 memset(ret, 0, PAGE_SIZE);
217 return ret;
218 }
219 return get_image_page(gfp_mask, PG_SAFE);
220}
221
222unsigned long get_safe_page(gfp_t gfp_mask)
223{
224 return (unsigned long)__get_safe_page(gfp_mask);
225}
226
227static struct page *alloc_image_page(gfp_t gfp_mask)
228{
229 struct page *page;
230
231 page = alloc_page(gfp_mask);
232 if (page) {
233 swsusp_set_page_forbidden(page);
234 swsusp_set_page_free(page);
235 }
236 return page;
237}
238
239static void recycle_safe_page(void *page_address)
240{
241 struct linked_page *lp = page_address;
242
243 lp->next = safe_pages_list;
244 safe_pages_list = lp;
245}
246
247/**
248 * free_image_page - Free a page allocated for hibernation image.
249 * @addr: Address of the page to free.
250 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
251 *
252 * The page to free should have been allocated by get_image_page() (page flags
253 * set by it are affected).
254 */
255static inline void free_image_page(void *addr, int clear_nosave_free)
256{
257 struct page *page;
258
259 BUG_ON(!virt_addr_valid(addr));
260
261 page = virt_to_page(addr);
262
263 swsusp_unset_page_forbidden(page);
264 if (clear_nosave_free)
265 swsusp_unset_page_free(page);
266
267 __free_page(page);
268}
269
270static inline void free_list_of_pages(struct linked_page *list,
271 int clear_page_nosave)
272{
273 while (list) {
274 struct linked_page *lp = list->next;
275
276 free_image_page(list, clear_page_nosave);
277 list = lp;
278 }
279}
280
281/*
282 * struct chain_allocator is used for allocating small objects out of
283 * a linked list of pages called 'the chain'.
284 *
285 * The chain grows each time when there is no room for a new object in
286 * the current page. The allocated objects cannot be freed individually.
287 * It is only possible to free them all at once, by freeing the entire
288 * chain.
289 *
290 * NOTE: The chain allocator may be inefficient if the allocated objects
291 * are not much smaller than PAGE_SIZE.
292 */
293struct chain_allocator {
294 struct linked_page *chain; /* the chain */
295 unsigned int used_space; /* total size of objects allocated out
296 of the current page */
297 gfp_t gfp_mask; /* mask for allocating pages */
298 int safe_needed; /* if set, only "safe" pages are allocated */
299};
300
301static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
302 int safe_needed)
303{
304 ca->chain = NULL;
305 ca->used_space = LINKED_PAGE_DATA_SIZE;
306 ca->gfp_mask = gfp_mask;
307 ca->safe_needed = safe_needed;
308}
309
310static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
311{
312 void *ret;
313
314 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
315 struct linked_page *lp;
316
317 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
318 get_image_page(ca->gfp_mask, PG_ANY);
319 if (!lp)
320 return NULL;
321
322 lp->next = ca->chain;
323 ca->chain = lp;
324 ca->used_space = 0;
325 }
326 ret = ca->chain->data + ca->used_space;
327 ca->used_space += size;
328 return ret;
329}
330
331/*
332 * Data types related to memory bitmaps.
333 *
334 * Memory bitmap is a structure consisting of many linked lists of
335 * objects. The main list's elements are of type struct zone_bitmap
336 * and each of them corresponds to one zone. For each zone bitmap
337 * object there is a list of objects of type struct bm_block that
338 * represent each blocks of bitmap in which information is stored.
339 *
340 * struct memory_bitmap contains a pointer to the main list of zone
341 * bitmap objects, a struct bm_position used for browsing the bitmap,
342 * and a pointer to the list of pages used for allocating all of the
343 * zone bitmap objects and bitmap block objects.
344 *
345 * NOTE: It has to be possible to lay out the bitmap in memory
346 * using only allocations of order 0. Additionally, the bitmap is
347 * designed to work with arbitrary number of zones (this is over the
348 * top for now, but let's avoid making unnecessary assumptions ;-).
349 *
350 * struct zone_bitmap contains a pointer to a list of bitmap block
351 * objects and a pointer to the bitmap block object that has been
352 * most recently used for setting bits. Additionally, it contains the
353 * PFNs that correspond to the start and end of the represented zone.
354 *
355 * struct bm_block contains a pointer to the memory page in which
356 * information is stored (in the form of a block of bitmap)
357 * It also contains the pfns that correspond to the start and end of
358 * the represented memory area.
359 *
360 * The memory bitmap is organized as a radix tree to guarantee fast random
361 * access to the bits. There is one radix tree for each zone (as returned
362 * from create_mem_extents).
363 *
364 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
365 * two linked lists for the nodes of the tree, one for the inner nodes and
366 * one for the leave nodes. The linked leave nodes are used for fast linear
367 * access of the memory bitmap.
368 *
369 * The struct rtree_node represents one node of the radix tree.
370 */
371
372#define BM_END_OF_MAP (~0UL)
373
374#define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
375#define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
376#define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
377
378/*
379 * struct rtree_node is a wrapper struct to link the nodes
380 * of the rtree together for easy linear iteration over
381 * bits and easy freeing
382 */
383struct rtree_node {
384 struct list_head list;
385 unsigned long *data;
386};
387
388/*
389 * struct mem_zone_bm_rtree represents a bitmap used for one
390 * populated memory zone.
391 */
392struct mem_zone_bm_rtree {
393 struct list_head list; /* Link Zones together */
394 struct list_head nodes; /* Radix Tree inner nodes */
395 struct list_head leaves; /* Radix Tree leaves */
396 unsigned long start_pfn; /* Zone start page frame */
397 unsigned long end_pfn; /* Zone end page frame + 1 */
398 struct rtree_node *rtree; /* Radix Tree Root */
399 int levels; /* Number of Radix Tree Levels */
400 unsigned int blocks; /* Number of Bitmap Blocks */
401};
402
403/* struct bm_position is used for browsing memory bitmaps */
404
405struct bm_position {
406 struct mem_zone_bm_rtree *zone;
407 struct rtree_node *node;
408 unsigned long node_pfn;
409 unsigned long cur_pfn;
410 int node_bit;
411};
412
413struct memory_bitmap {
414 struct list_head zones;
415 struct linked_page *p_list; /* list of pages used to store zone
416 bitmap objects and bitmap block
417 objects */
418 struct bm_position cur; /* most recently used bit position */
419};
420
421/* Functions that operate on memory bitmaps */
422
423#define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
424#if BITS_PER_LONG == 32
425#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
426#else
427#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
428#endif
429#define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
430
431/**
432 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
433 * @gfp_mask: GFP mask for the allocation.
434 * @safe_needed: Get pages not used before hibernation (restore only)
435 * @ca: Pointer to a linked list of pages ("a chain") to allocate from
436 * @list: Radix Tree node to add.
437 *
438 * This function is used to allocate inner nodes as well as the
439 * leave nodes of the radix tree. It also adds the node to the
440 * corresponding linked list passed in by the *list parameter.
441 */
442static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
443 struct chain_allocator *ca,
444 struct list_head *list)
445{
446 struct rtree_node *node;
447
448 node = chain_alloc(ca, sizeof(struct rtree_node));
449 if (!node)
450 return NULL;
451
452 node->data = get_image_page(gfp_mask, safe_needed);
453 if (!node->data)
454 return NULL;
455
456 list_add_tail(&node->list, list);
457
458 return node;
459}
460
461/**
462 * add_rtree_block - Add a new leave node to the radix tree.
463 *
464 * The leave nodes need to be allocated in order to keep the leaves
465 * linked list in order. This is guaranteed by the zone->blocks
466 * counter.
467 */
468static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
469 int safe_needed, struct chain_allocator *ca)
470{
471 struct rtree_node *node, *block, **dst;
472 unsigned int levels_needed, block_nr;
473 int i;
474
475 block_nr = zone->blocks;
476 levels_needed = 0;
477
478 /* How many levels do we need for this block nr? */
479 while (block_nr) {
480 levels_needed += 1;
481 block_nr >>= BM_RTREE_LEVEL_SHIFT;
482 }
483
484 /* Make sure the rtree has enough levels */
485 for (i = zone->levels; i < levels_needed; i++) {
486 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
487 &zone->nodes);
488 if (!node)
489 return -ENOMEM;
490
491 node->data[0] = (unsigned long)zone->rtree;
492 zone->rtree = node;
493 zone->levels += 1;
494 }
495
496 /* Allocate new block */
497 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
498 if (!block)
499 return -ENOMEM;
500
501 /* Now walk the rtree to insert the block */
502 node = zone->rtree;
503 dst = &zone->rtree;
504 block_nr = zone->blocks;
505 for (i = zone->levels; i > 0; i--) {
506 int index;
507
508 if (!node) {
509 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
510 &zone->nodes);
511 if (!node)
512 return -ENOMEM;
513 *dst = node;
514 }
515
516 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
517 index &= BM_RTREE_LEVEL_MASK;
518 dst = (struct rtree_node **)&((*dst)->data[index]);
519 node = *dst;
520 }
521
522 zone->blocks += 1;
523 *dst = block;
524
525 return 0;
526}
527
528static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
529 int clear_nosave_free);
530
531/**
532 * create_zone_bm_rtree - Create a radix tree for one zone.
533 *
534 * Allocated the mem_zone_bm_rtree structure and initializes it.
535 * This function also allocated and builds the radix tree for the
536 * zone.
537 */
538static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
539 int safe_needed,
540 struct chain_allocator *ca,
541 unsigned long start,
542 unsigned long end)
543{
544 struct mem_zone_bm_rtree *zone;
545 unsigned int i, nr_blocks;
546 unsigned long pages;
547
548 pages = end - start;
549 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
550 if (!zone)
551 return NULL;
552
553 INIT_LIST_HEAD(&zone->nodes);
554 INIT_LIST_HEAD(&zone->leaves);
555 zone->start_pfn = start;
556 zone->end_pfn = end;
557 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
558
559 for (i = 0; i < nr_blocks; i++) {
560 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
561 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
562 return NULL;
563 }
564 }
565
566 return zone;
567}
568
569/**
570 * free_zone_bm_rtree - Free the memory of the radix tree.
571 *
572 * Free all node pages of the radix tree. The mem_zone_bm_rtree
573 * structure itself is not freed here nor are the rtree_node
574 * structs.
575 */
576static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
577 int clear_nosave_free)
578{
579 struct rtree_node *node;
580
581 list_for_each_entry(node, &zone->nodes, list)
582 free_image_page(node->data, clear_nosave_free);
583
584 list_for_each_entry(node, &zone->leaves, list)
585 free_image_page(node->data, clear_nosave_free);
586}
587
588static void memory_bm_position_reset(struct memory_bitmap *bm)
589{
590 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
591 list);
592 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
593 struct rtree_node, list);
594 bm->cur.node_pfn = 0;
595 bm->cur.cur_pfn = BM_END_OF_MAP;
596 bm->cur.node_bit = 0;
597}
598
599static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
600
601struct mem_extent {
602 struct list_head hook;
603 unsigned long start;
604 unsigned long end;
605};
606
607/**
608 * free_mem_extents - Free a list of memory extents.
609 * @list: List of extents to free.
610 */
611static void free_mem_extents(struct list_head *list)
612{
613 struct mem_extent *ext, *aux;
614
615 list_for_each_entry_safe(ext, aux, list, hook) {
616 list_del(&ext->hook);
617 kfree(ext);
618 }
619}
620
621/**
622 * create_mem_extents - Create a list of memory extents.
623 * @list: List to put the extents into.
624 * @gfp_mask: Mask to use for memory allocations.
625 *
626 * The extents represent contiguous ranges of PFNs.
627 */
628static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
629{
630 struct zone *zone;
631
632 INIT_LIST_HEAD(list);
633
634 for_each_populated_zone(zone) {
635 unsigned long zone_start, zone_end;
636 struct mem_extent *ext, *cur, *aux;
637
638 zone_start = zone->zone_start_pfn;
639 zone_end = zone_end_pfn(zone);
640
641 list_for_each_entry(ext, list, hook)
642 if (zone_start <= ext->end)
643 break;
644
645 if (&ext->hook == list || zone_end < ext->start) {
646 /* New extent is necessary */
647 struct mem_extent *new_ext;
648
649 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
650 if (!new_ext) {
651 free_mem_extents(list);
652 return -ENOMEM;
653 }
654 new_ext->start = zone_start;
655 new_ext->end = zone_end;
656 list_add_tail(&new_ext->hook, &ext->hook);
657 continue;
658 }
659
660 /* Merge this zone's range of PFNs with the existing one */
661 if (zone_start < ext->start)
662 ext->start = zone_start;
663 if (zone_end > ext->end)
664 ext->end = zone_end;
665
666 /* More merging may be possible */
667 cur = ext;
668 list_for_each_entry_safe_continue(cur, aux, list, hook) {
669 if (zone_end < cur->start)
670 break;
671 if (zone_end < cur->end)
672 ext->end = cur->end;
673 list_del(&cur->hook);
674 kfree(cur);
675 }
676 }
677
678 return 0;
679}
680
681/**
682 * memory_bm_create - Allocate memory for a memory bitmap.
683 */
684static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
685 int safe_needed)
686{
687 struct chain_allocator ca;
688 struct list_head mem_extents;
689 struct mem_extent *ext;
690 int error;
691
692 chain_init(&ca, gfp_mask, safe_needed);
693 INIT_LIST_HEAD(&bm->zones);
694
695 error = create_mem_extents(&mem_extents, gfp_mask);
696 if (error)
697 return error;
698
699 list_for_each_entry(ext, &mem_extents, hook) {
700 struct mem_zone_bm_rtree *zone;
701
702 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
703 ext->start, ext->end);
704 if (!zone) {
705 error = -ENOMEM;
706 goto Error;
707 }
708 list_add_tail(&zone->list, &bm->zones);
709 }
710
711 bm->p_list = ca.chain;
712 memory_bm_position_reset(bm);
713 Exit:
714 free_mem_extents(&mem_extents);
715 return error;
716
717 Error:
718 bm->p_list = ca.chain;
719 memory_bm_free(bm, PG_UNSAFE_CLEAR);
720 goto Exit;
721}
722
723/**
724 * memory_bm_free - Free memory occupied by the memory bitmap.
725 * @bm: Memory bitmap.
726 */
727static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
728{
729 struct mem_zone_bm_rtree *zone;
730
731 list_for_each_entry(zone, &bm->zones, list)
732 free_zone_bm_rtree(zone, clear_nosave_free);
733
734 free_list_of_pages(bm->p_list, clear_nosave_free);
735
736 INIT_LIST_HEAD(&bm->zones);
737}
738
739/**
740 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
741 *
742 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
743 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
744 *
745 * Walk the radix tree to find the page containing the bit that represents @pfn
746 * and return the position of the bit in @addr and @bit_nr.
747 */
748static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
749 void **addr, unsigned int *bit_nr)
750{
751 struct mem_zone_bm_rtree *curr, *zone;
752 struct rtree_node *node;
753 int i, block_nr;
754
755 zone = bm->cur.zone;
756
757 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
758 goto zone_found;
759
760 zone = NULL;
761
762 /* Find the right zone */
763 list_for_each_entry(curr, &bm->zones, list) {
764 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
765 zone = curr;
766 break;
767 }
768 }
769
770 if (!zone)
771 return -EFAULT;
772
773zone_found:
774 /*
775 * We have found the zone. Now walk the radix tree to find the leaf node
776 * for our PFN.
777 */
778
779 /*
780 * If the zone we wish to scan is the current zone and the
781 * pfn falls into the current node then we do not need to walk
782 * the tree.
783 */
784 node = bm->cur.node;
785 if (zone == bm->cur.zone &&
786 ((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
787 goto node_found;
788
789 node = zone->rtree;
790 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
791
792 for (i = zone->levels; i > 0; i--) {
793 int index;
794
795 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
796 index &= BM_RTREE_LEVEL_MASK;
797 BUG_ON(node->data[index] == 0);
798 node = (struct rtree_node *)node->data[index];
799 }
800
801node_found:
802 /* Update last position */
803 bm->cur.zone = zone;
804 bm->cur.node = node;
805 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
806 bm->cur.cur_pfn = pfn;
807
808 /* Set return values */
809 *addr = node->data;
810 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
811
812 return 0;
813}
814
815static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
816{
817 void *addr;
818 unsigned int bit;
819 int error;
820
821 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
822 BUG_ON(error);
823 set_bit(bit, addr);
824}
825
826static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
827{
828 void *addr;
829 unsigned int bit;
830 int error;
831
832 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
833 if (!error)
834 set_bit(bit, addr);
835
836 return error;
837}
838
839static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
840{
841 void *addr;
842 unsigned int bit;
843 int error;
844
845 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
846 BUG_ON(error);
847 clear_bit(bit, addr);
848}
849
850static void memory_bm_clear_current(struct memory_bitmap *bm)
851{
852 int bit;
853
854 bit = max(bm->cur.node_bit - 1, 0);
855 clear_bit(bit, bm->cur.node->data);
856}
857
858static unsigned long memory_bm_get_current(struct memory_bitmap *bm)
859{
860 return bm->cur.cur_pfn;
861}
862
863static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
864{
865 void *addr;
866 unsigned int bit;
867 int error;
868
869 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
870 BUG_ON(error);
871 return test_bit(bit, addr);
872}
873
874static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
875{
876 void *addr;
877 unsigned int bit;
878
879 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
880}
881
882/*
883 * rtree_next_node - Jump to the next leaf node.
884 *
885 * Set the position to the beginning of the next node in the
886 * memory bitmap. This is either the next node in the current
887 * zone's radix tree or the first node in the radix tree of the
888 * next zone.
889 *
890 * Return true if there is a next node, false otherwise.
891 */
892static bool rtree_next_node(struct memory_bitmap *bm)
893{
894 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
895 bm->cur.node = list_entry(bm->cur.node->list.next,
896 struct rtree_node, list);
897 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
898 bm->cur.node_bit = 0;
899 touch_softlockup_watchdog();
900 return true;
901 }
902
903 /* No more nodes, goto next zone */
904 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
905 bm->cur.zone = list_entry(bm->cur.zone->list.next,
906 struct mem_zone_bm_rtree, list);
907 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
908 struct rtree_node, list);
909 bm->cur.node_pfn = 0;
910 bm->cur.node_bit = 0;
911 return true;
912 }
913
914 /* No more zones */
915 return false;
916}
917
918/**
919 * memory_bm_next_pfn - Find the next set bit in a memory bitmap.
920 * @bm: Memory bitmap.
921 *
922 * Starting from the last returned position this function searches for the next
923 * set bit in @bm and returns the PFN represented by it. If no more bits are
924 * set, BM_END_OF_MAP is returned.
925 *
926 * It is required to run memory_bm_position_reset() before the first call to
927 * this function for the given memory bitmap.
928 */
929static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
930{
931 unsigned long bits, pfn, pages;
932 int bit;
933
934 do {
935 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
936 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
937 bit = find_next_bit(bm->cur.node->data, bits,
938 bm->cur.node_bit);
939 if (bit < bits) {
940 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
941 bm->cur.node_bit = bit + 1;
942 bm->cur.cur_pfn = pfn;
943 return pfn;
944 }
945 } while (rtree_next_node(bm));
946
947 bm->cur.cur_pfn = BM_END_OF_MAP;
948 return BM_END_OF_MAP;
949}
950
951/*
952 * This structure represents a range of page frames the contents of which
953 * should not be saved during hibernation.
954 */
955struct nosave_region {
956 struct list_head list;
957 unsigned long start_pfn;
958 unsigned long end_pfn;
959};
960
961static LIST_HEAD(nosave_regions);
962
963static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
964{
965 struct rtree_node *node;
966
967 list_for_each_entry(node, &zone->nodes, list)
968 recycle_safe_page(node->data);
969
970 list_for_each_entry(node, &zone->leaves, list)
971 recycle_safe_page(node->data);
972}
973
974static void memory_bm_recycle(struct memory_bitmap *bm)
975{
976 struct mem_zone_bm_rtree *zone;
977 struct linked_page *p_list;
978
979 list_for_each_entry(zone, &bm->zones, list)
980 recycle_zone_bm_rtree(zone);
981
982 p_list = bm->p_list;
983 while (p_list) {
984 struct linked_page *lp = p_list;
985
986 p_list = lp->next;
987 recycle_safe_page(lp);
988 }
989}
990
991/**
992 * register_nosave_region - Register a region of unsaveable memory.
993 *
994 * Register a range of page frames the contents of which should not be saved
995 * during hibernation (to be used in the early initialization code).
996 */
997void __init register_nosave_region(unsigned long start_pfn, unsigned long end_pfn)
998{
999 struct nosave_region *region;
1000
1001 if (start_pfn >= end_pfn)
1002 return;
1003
1004 if (!list_empty(&nosave_regions)) {
1005 /* Try to extend the previous region (they should be sorted) */
1006 region = list_entry(nosave_regions.prev,
1007 struct nosave_region, list);
1008 if (region->end_pfn == start_pfn) {
1009 region->end_pfn = end_pfn;
1010 goto Report;
1011 }
1012 }
1013 /* This allocation cannot fail */
1014 region = memblock_alloc(sizeof(struct nosave_region),
1015 SMP_CACHE_BYTES);
1016 if (!region)
1017 panic("%s: Failed to allocate %zu bytes\n", __func__,
1018 sizeof(struct nosave_region));
1019 region->start_pfn = start_pfn;
1020 region->end_pfn = end_pfn;
1021 list_add_tail(®ion->list, &nosave_regions);
1022 Report:
1023 pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
1024 (unsigned long long) start_pfn << PAGE_SHIFT,
1025 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1026}
1027
1028/*
1029 * Set bits in this map correspond to the page frames the contents of which
1030 * should not be saved during the suspend.
1031 */
1032static struct memory_bitmap *forbidden_pages_map;
1033
1034/* Set bits in this map correspond to free page frames. */
1035static struct memory_bitmap *free_pages_map;
1036
1037/*
1038 * Each page frame allocated for creating the image is marked by setting the
1039 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
1040 */
1041
1042void swsusp_set_page_free(struct page *page)
1043{
1044 if (free_pages_map)
1045 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
1046}
1047
1048static int swsusp_page_is_free(struct page *page)
1049{
1050 return free_pages_map ?
1051 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1052}
1053
1054void swsusp_unset_page_free(struct page *page)
1055{
1056 if (free_pages_map)
1057 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1058}
1059
1060static void swsusp_set_page_forbidden(struct page *page)
1061{
1062 if (forbidden_pages_map)
1063 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1064}
1065
1066int swsusp_page_is_forbidden(struct page *page)
1067{
1068 return forbidden_pages_map ?
1069 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1070}
1071
1072static void swsusp_unset_page_forbidden(struct page *page)
1073{
1074 if (forbidden_pages_map)
1075 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1076}
1077
1078/**
1079 * mark_nosave_pages - Mark pages that should not be saved.
1080 * @bm: Memory bitmap.
1081 *
1082 * Set the bits in @bm that correspond to the page frames the contents of which
1083 * should not be saved.
1084 */
1085static void mark_nosave_pages(struct memory_bitmap *bm)
1086{
1087 struct nosave_region *region;
1088
1089 if (list_empty(&nosave_regions))
1090 return;
1091
1092 list_for_each_entry(region, &nosave_regions, list) {
1093 unsigned long pfn;
1094
1095 pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1096 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1097 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1098 - 1);
1099
1100 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1101 if (pfn_valid(pfn)) {
1102 /*
1103 * It is safe to ignore the result of
1104 * mem_bm_set_bit_check() here, since we won't
1105 * touch the PFNs for which the error is
1106 * returned anyway.
1107 */
1108 mem_bm_set_bit_check(bm, pfn);
1109 }
1110 }
1111}
1112
1113/**
1114 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1115 *
1116 * Create bitmaps needed for marking page frames that should not be saved and
1117 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1118 * only modified if everything goes well, because we don't want the bits to be
1119 * touched before both bitmaps are set up.
1120 */
1121int create_basic_memory_bitmaps(void)
1122{
1123 struct memory_bitmap *bm1, *bm2;
1124 int error;
1125
1126 if (forbidden_pages_map && free_pages_map)
1127 return 0;
1128 else
1129 BUG_ON(forbidden_pages_map || free_pages_map);
1130
1131 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1132 if (!bm1)
1133 return -ENOMEM;
1134
1135 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1136 if (error)
1137 goto Free_first_object;
1138
1139 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1140 if (!bm2)
1141 goto Free_first_bitmap;
1142
1143 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1144 if (error)
1145 goto Free_second_object;
1146
1147 forbidden_pages_map = bm1;
1148 free_pages_map = bm2;
1149 mark_nosave_pages(forbidden_pages_map);
1150
1151 pr_debug("Basic memory bitmaps created\n");
1152
1153 return 0;
1154
1155 Free_second_object:
1156 kfree(bm2);
1157 Free_first_bitmap:
1158 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1159 Free_first_object:
1160 kfree(bm1);
1161 return -ENOMEM;
1162}
1163
1164/**
1165 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1166 *
1167 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1168 * auxiliary pointers are necessary so that the bitmaps themselves are not
1169 * referred to while they are being freed.
1170 */
1171void free_basic_memory_bitmaps(void)
1172{
1173 struct memory_bitmap *bm1, *bm2;
1174
1175 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1176 return;
1177
1178 bm1 = forbidden_pages_map;
1179 bm2 = free_pages_map;
1180 forbidden_pages_map = NULL;
1181 free_pages_map = NULL;
1182 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1183 kfree(bm1);
1184 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1185 kfree(bm2);
1186
1187 pr_debug("Basic memory bitmaps freed\n");
1188}
1189
1190static void clear_or_poison_free_page(struct page *page)
1191{
1192 if (page_poisoning_enabled_static())
1193 __kernel_poison_pages(page, 1);
1194 else if (want_init_on_free())
1195 clear_highpage(page);
1196}
1197
1198void clear_or_poison_free_pages(void)
1199{
1200 struct memory_bitmap *bm = free_pages_map;
1201 unsigned long pfn;
1202
1203 if (WARN_ON(!(free_pages_map)))
1204 return;
1205
1206 if (page_poisoning_enabled() || want_init_on_free()) {
1207 memory_bm_position_reset(bm);
1208 pfn = memory_bm_next_pfn(bm);
1209 while (pfn != BM_END_OF_MAP) {
1210 if (pfn_valid(pfn))
1211 clear_or_poison_free_page(pfn_to_page(pfn));
1212
1213 pfn = memory_bm_next_pfn(bm);
1214 }
1215 memory_bm_position_reset(bm);
1216 pr_info("free pages cleared after restore\n");
1217 }
1218}
1219
1220/**
1221 * snapshot_additional_pages - Estimate the number of extra pages needed.
1222 * @zone: Memory zone to carry out the computation for.
1223 *
1224 * Estimate the number of additional pages needed for setting up a hibernation
1225 * image data structures for @zone (usually, the returned value is greater than
1226 * the exact number).
1227 */
1228unsigned int snapshot_additional_pages(struct zone *zone)
1229{
1230 unsigned int rtree, nodes;
1231
1232 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1233 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1234 LINKED_PAGE_DATA_SIZE);
1235 while (nodes > 1) {
1236 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1237 rtree += nodes;
1238 }
1239
1240 return 2 * rtree;
1241}
1242
1243/*
1244 * Touch the watchdog for every WD_PAGE_COUNT pages.
1245 */
1246#define WD_PAGE_COUNT (128*1024)
1247
1248static void mark_free_pages(struct zone *zone)
1249{
1250 unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
1251 unsigned long flags;
1252 unsigned int order, t;
1253 struct page *page;
1254
1255 if (zone_is_empty(zone))
1256 return;
1257
1258 spin_lock_irqsave(&zone->lock, flags);
1259
1260 max_zone_pfn = zone_end_pfn(zone);
1261 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1262 if (pfn_valid(pfn)) {
1263 page = pfn_to_page(pfn);
1264
1265 if (!--page_count) {
1266 touch_nmi_watchdog();
1267 page_count = WD_PAGE_COUNT;
1268 }
1269
1270 if (page_zone(page) != zone)
1271 continue;
1272
1273 if (!swsusp_page_is_forbidden(page))
1274 swsusp_unset_page_free(page);
1275 }
1276
1277 for_each_migratetype_order(order, t) {
1278 list_for_each_entry(page,
1279 &zone->free_area[order].free_list[t], buddy_list) {
1280 unsigned long i;
1281
1282 pfn = page_to_pfn(page);
1283 for (i = 0; i < (1UL << order); i++) {
1284 if (!--page_count) {
1285 touch_nmi_watchdog();
1286 page_count = WD_PAGE_COUNT;
1287 }
1288 swsusp_set_page_free(pfn_to_page(pfn + i));
1289 }
1290 }
1291 }
1292 spin_unlock_irqrestore(&zone->lock, flags);
1293}
1294
1295#ifdef CONFIG_HIGHMEM
1296/**
1297 * count_free_highmem_pages - Compute the total number of free highmem pages.
1298 *
1299 * The returned number is system-wide.
1300 */
1301static unsigned int count_free_highmem_pages(void)
1302{
1303 struct zone *zone;
1304 unsigned int cnt = 0;
1305
1306 for_each_populated_zone(zone)
1307 if (is_highmem(zone))
1308 cnt += zone_page_state(zone, NR_FREE_PAGES);
1309
1310 return cnt;
1311}
1312
1313/**
1314 * saveable_highmem_page - Check if a highmem page is saveable.
1315 *
1316 * Determine whether a highmem page should be included in a hibernation image.
1317 *
1318 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1319 * and it isn't part of a free chunk of pages.
1320 */
1321static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1322{
1323 struct page *page;
1324
1325 if (!pfn_valid(pfn))
1326 return NULL;
1327
1328 page = pfn_to_online_page(pfn);
1329 if (!page || page_zone(page) != zone)
1330 return NULL;
1331
1332 BUG_ON(!PageHighMem(page));
1333
1334 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1335 return NULL;
1336
1337 if (PageReserved(page) || PageOffline(page))
1338 return NULL;
1339
1340 if (page_is_guard(page))
1341 return NULL;
1342
1343 return page;
1344}
1345
1346/**
1347 * count_highmem_pages - Compute the total number of saveable highmem pages.
1348 */
1349static unsigned int count_highmem_pages(void)
1350{
1351 struct zone *zone;
1352 unsigned int n = 0;
1353
1354 for_each_populated_zone(zone) {
1355 unsigned long pfn, max_zone_pfn;
1356
1357 if (!is_highmem(zone))
1358 continue;
1359
1360 mark_free_pages(zone);
1361 max_zone_pfn = zone_end_pfn(zone);
1362 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1363 if (saveable_highmem_page(zone, pfn))
1364 n++;
1365 }
1366 return n;
1367}
1368#endif /* CONFIG_HIGHMEM */
1369
1370/**
1371 * saveable_page - Check if the given page is saveable.
1372 *
1373 * Determine whether a non-highmem page should be included in a hibernation
1374 * image.
1375 *
1376 * We should save the page if it isn't Nosave, and is not in the range
1377 * of pages statically defined as 'unsaveable', and it isn't part of
1378 * a free chunk of pages.
1379 */
1380static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1381{
1382 struct page *page;
1383
1384 if (!pfn_valid(pfn))
1385 return NULL;
1386
1387 page = pfn_to_online_page(pfn);
1388 if (!page || page_zone(page) != zone)
1389 return NULL;
1390
1391 BUG_ON(PageHighMem(page));
1392
1393 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1394 return NULL;
1395
1396 if (PageOffline(page))
1397 return NULL;
1398
1399 if (PageReserved(page)
1400 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1401 return NULL;
1402
1403 if (page_is_guard(page))
1404 return NULL;
1405
1406 return page;
1407}
1408
1409/**
1410 * count_data_pages - Compute the total number of saveable non-highmem pages.
1411 */
1412static unsigned int count_data_pages(void)
1413{
1414 struct zone *zone;
1415 unsigned long pfn, max_zone_pfn;
1416 unsigned int n = 0;
1417
1418 for_each_populated_zone(zone) {
1419 if (is_highmem(zone))
1420 continue;
1421
1422 mark_free_pages(zone);
1423 max_zone_pfn = zone_end_pfn(zone);
1424 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1425 if (saveable_page(zone, pfn))
1426 n++;
1427 }
1428 return n;
1429}
1430
1431/*
1432 * This is needed, because copy_page and memcpy are not usable for copying
1433 * task structs. Returns true if the page was filled with only zeros,
1434 * otherwise false.
1435 */
1436static inline bool do_copy_page(long *dst, long *src)
1437{
1438 long z = 0;
1439 int n;
1440
1441 for (n = PAGE_SIZE / sizeof(long); n; n--) {
1442 z |= *src;
1443 *dst++ = *src++;
1444 }
1445 return !z;
1446}
1447
1448/**
1449 * safe_copy_page - Copy a page in a safe way.
1450 *
1451 * Check if the page we are going to copy is marked as present in the kernel
1452 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1453 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1454 * always returns 'true'. Returns true if the page was entirely composed of
1455 * zeros, otherwise it will return false.
1456 */
1457static bool safe_copy_page(void *dst, struct page *s_page)
1458{
1459 bool zeros_only;
1460
1461 if (kernel_page_present(s_page)) {
1462 zeros_only = do_copy_page(dst, page_address(s_page));
1463 } else {
1464 hibernate_map_page(s_page);
1465 zeros_only = do_copy_page(dst, page_address(s_page));
1466 hibernate_unmap_page(s_page);
1467 }
1468 return zeros_only;
1469}
1470
1471#ifdef CONFIG_HIGHMEM
1472static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1473{
1474 return is_highmem(zone) ?
1475 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1476}
1477
1478static bool copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1479{
1480 struct page *s_page, *d_page;
1481 void *src, *dst;
1482 bool zeros_only;
1483
1484 s_page = pfn_to_page(src_pfn);
1485 d_page = pfn_to_page(dst_pfn);
1486 if (PageHighMem(s_page)) {
1487 src = kmap_local_page(s_page);
1488 dst = kmap_local_page(d_page);
1489 zeros_only = do_copy_page(dst, src);
1490 kunmap_local(dst);
1491 kunmap_local(src);
1492 } else {
1493 if (PageHighMem(d_page)) {
1494 /*
1495 * The page pointed to by src may contain some kernel
1496 * data modified by kmap_atomic()
1497 */
1498 zeros_only = safe_copy_page(buffer, s_page);
1499 dst = kmap_local_page(d_page);
1500 copy_page(dst, buffer);
1501 kunmap_local(dst);
1502 } else {
1503 zeros_only = safe_copy_page(page_address(d_page), s_page);
1504 }
1505 }
1506 return zeros_only;
1507}
1508#else
1509#define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1510
1511static inline int copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1512{
1513 return safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1514 pfn_to_page(src_pfn));
1515}
1516#endif /* CONFIG_HIGHMEM */
1517
1518/*
1519 * Copy data pages will copy all pages into pages pulled from the copy_bm.
1520 * If a page was entirely filled with zeros it will be marked in the zero_bm.
1521 *
1522 * Returns the number of pages copied.
1523 */
1524static unsigned long copy_data_pages(struct memory_bitmap *copy_bm,
1525 struct memory_bitmap *orig_bm,
1526 struct memory_bitmap *zero_bm)
1527{
1528 unsigned long copied_pages = 0;
1529 struct zone *zone;
1530 unsigned long pfn, copy_pfn;
1531
1532 for_each_populated_zone(zone) {
1533 unsigned long max_zone_pfn;
1534
1535 mark_free_pages(zone);
1536 max_zone_pfn = zone_end_pfn(zone);
1537 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1538 if (page_is_saveable(zone, pfn))
1539 memory_bm_set_bit(orig_bm, pfn);
1540 }
1541 memory_bm_position_reset(orig_bm);
1542 memory_bm_position_reset(copy_bm);
1543 copy_pfn = memory_bm_next_pfn(copy_bm);
1544 for(;;) {
1545 pfn = memory_bm_next_pfn(orig_bm);
1546 if (unlikely(pfn == BM_END_OF_MAP))
1547 break;
1548 if (copy_data_page(copy_pfn, pfn)) {
1549 memory_bm_set_bit(zero_bm, pfn);
1550 /* Use this copy_pfn for a page that is not full of zeros */
1551 continue;
1552 }
1553 copied_pages++;
1554 copy_pfn = memory_bm_next_pfn(copy_bm);
1555 }
1556 return copied_pages;
1557}
1558
1559/* Total number of image pages */
1560static unsigned int nr_copy_pages;
1561/* Number of pages needed for saving the original pfns of the image pages */
1562static unsigned int nr_meta_pages;
1563/* Number of zero pages */
1564static unsigned int nr_zero_pages;
1565
1566/*
1567 * Numbers of normal and highmem page frames allocated for hibernation image
1568 * before suspending devices.
1569 */
1570static unsigned int alloc_normal, alloc_highmem;
1571/*
1572 * Memory bitmap used for marking saveable pages (during hibernation) or
1573 * hibernation image pages (during restore)
1574 */
1575static struct memory_bitmap orig_bm;
1576/*
1577 * Memory bitmap used during hibernation for marking allocated page frames that
1578 * will contain copies of saveable pages. During restore it is initially used
1579 * for marking hibernation image pages, but then the set bits from it are
1580 * duplicated in @orig_bm and it is released. On highmem systems it is next
1581 * used for marking "safe" highmem pages, but it has to be reinitialized for
1582 * this purpose.
1583 */
1584static struct memory_bitmap copy_bm;
1585
1586/* Memory bitmap which tracks which saveable pages were zero filled. */
1587static struct memory_bitmap zero_bm;
1588
1589/**
1590 * swsusp_free - Free pages allocated for hibernation image.
1591 *
1592 * Image pages are allocated before snapshot creation, so they need to be
1593 * released after resume.
1594 */
1595void swsusp_free(void)
1596{
1597 unsigned long fb_pfn, fr_pfn;
1598
1599 if (!forbidden_pages_map || !free_pages_map)
1600 goto out;
1601
1602 memory_bm_position_reset(forbidden_pages_map);
1603 memory_bm_position_reset(free_pages_map);
1604
1605loop:
1606 fr_pfn = memory_bm_next_pfn(free_pages_map);
1607 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1608
1609 /*
1610 * Find the next bit set in both bitmaps. This is guaranteed to
1611 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1612 */
1613 do {
1614 if (fb_pfn < fr_pfn)
1615 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1616 if (fr_pfn < fb_pfn)
1617 fr_pfn = memory_bm_next_pfn(free_pages_map);
1618 } while (fb_pfn != fr_pfn);
1619
1620 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1621 struct page *page = pfn_to_page(fr_pfn);
1622
1623 memory_bm_clear_current(forbidden_pages_map);
1624 memory_bm_clear_current(free_pages_map);
1625 hibernate_restore_unprotect_page(page_address(page));
1626 __free_page(page);
1627 goto loop;
1628 }
1629
1630out:
1631 nr_copy_pages = 0;
1632 nr_meta_pages = 0;
1633 nr_zero_pages = 0;
1634 restore_pblist = NULL;
1635 buffer = NULL;
1636 alloc_normal = 0;
1637 alloc_highmem = 0;
1638 hibernate_restore_protection_end();
1639}
1640
1641/* Helper functions used for the shrinking of memory. */
1642
1643#define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1644
1645/**
1646 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1647 * @nr_pages: Number of page frames to allocate.
1648 * @mask: GFP flags to use for the allocation.
1649 *
1650 * Return value: Number of page frames actually allocated
1651 */
1652static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1653{
1654 unsigned long nr_alloc = 0;
1655
1656 while (nr_pages > 0) {
1657 struct page *page;
1658
1659 page = alloc_image_page(mask);
1660 if (!page)
1661 break;
1662 memory_bm_set_bit(©_bm, page_to_pfn(page));
1663 if (PageHighMem(page))
1664 alloc_highmem++;
1665 else
1666 alloc_normal++;
1667 nr_pages--;
1668 nr_alloc++;
1669 }
1670
1671 return nr_alloc;
1672}
1673
1674static unsigned long preallocate_image_memory(unsigned long nr_pages,
1675 unsigned long avail_normal)
1676{
1677 unsigned long alloc;
1678
1679 if (avail_normal <= alloc_normal)
1680 return 0;
1681
1682 alloc = avail_normal - alloc_normal;
1683 if (nr_pages < alloc)
1684 alloc = nr_pages;
1685
1686 return preallocate_image_pages(alloc, GFP_IMAGE);
1687}
1688
1689#ifdef CONFIG_HIGHMEM
1690static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1691{
1692 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1693}
1694
1695/**
1696 * __fraction - Compute (an approximation of) x * (multiplier / base).
1697 */
1698static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1699{
1700 return div64_u64(x * multiplier, base);
1701}
1702
1703static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1704 unsigned long highmem,
1705 unsigned long total)
1706{
1707 unsigned long alloc = __fraction(nr_pages, highmem, total);
1708
1709 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1710}
1711#else /* CONFIG_HIGHMEM */
1712static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1713{
1714 return 0;
1715}
1716
1717static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1718 unsigned long highmem,
1719 unsigned long total)
1720{
1721 return 0;
1722}
1723#endif /* CONFIG_HIGHMEM */
1724
1725/**
1726 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1727 */
1728static unsigned long free_unnecessary_pages(void)
1729{
1730 unsigned long save, to_free_normal, to_free_highmem, free;
1731
1732 save = count_data_pages();
1733 if (alloc_normal >= save) {
1734 to_free_normal = alloc_normal - save;
1735 save = 0;
1736 } else {
1737 to_free_normal = 0;
1738 save -= alloc_normal;
1739 }
1740 save += count_highmem_pages();
1741 if (alloc_highmem >= save) {
1742 to_free_highmem = alloc_highmem - save;
1743 } else {
1744 to_free_highmem = 0;
1745 save -= alloc_highmem;
1746 if (to_free_normal > save)
1747 to_free_normal -= save;
1748 else
1749 to_free_normal = 0;
1750 }
1751 free = to_free_normal + to_free_highmem;
1752
1753 memory_bm_position_reset(©_bm);
1754
1755 while (to_free_normal > 0 || to_free_highmem > 0) {
1756 unsigned long pfn = memory_bm_next_pfn(©_bm);
1757 struct page *page = pfn_to_page(pfn);
1758
1759 if (PageHighMem(page)) {
1760 if (!to_free_highmem)
1761 continue;
1762 to_free_highmem--;
1763 alloc_highmem--;
1764 } else {
1765 if (!to_free_normal)
1766 continue;
1767 to_free_normal--;
1768 alloc_normal--;
1769 }
1770 memory_bm_clear_bit(©_bm, pfn);
1771 swsusp_unset_page_forbidden(page);
1772 swsusp_unset_page_free(page);
1773 __free_page(page);
1774 }
1775
1776 return free;
1777}
1778
1779/**
1780 * minimum_image_size - Estimate the minimum acceptable size of an image.
1781 * @saveable: Number of saveable pages in the system.
1782 *
1783 * We want to avoid attempting to free too much memory too hard, so estimate the
1784 * minimum acceptable size of a hibernation image to use as the lower limit for
1785 * preallocating memory.
1786 *
1787 * We assume that the minimum image size should be proportional to
1788 *
1789 * [number of saveable pages] - [number of pages that can be freed in theory]
1790 *
1791 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1792 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1793 */
1794static unsigned long minimum_image_size(unsigned long saveable)
1795{
1796 unsigned long size;
1797
1798 size = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B)
1799 + global_node_page_state(NR_ACTIVE_ANON)
1800 + global_node_page_state(NR_INACTIVE_ANON)
1801 + global_node_page_state(NR_ACTIVE_FILE)
1802 + global_node_page_state(NR_INACTIVE_FILE);
1803
1804 return saveable <= size ? 0 : saveable - size;
1805}
1806
1807/**
1808 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1809 *
1810 * To create a hibernation image it is necessary to make a copy of every page
1811 * frame in use. We also need a number of page frames to be free during
1812 * hibernation for allocations made while saving the image and for device
1813 * drivers, in case they need to allocate memory from their hibernation
1814 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1815 * estimate) and reserved_size divided by PAGE_SIZE (which is tunable through
1816 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1817 * total number of available page frames and allocate at least
1818 *
1819 * ([page frames total] - PAGES_FOR_IO - [metadata pages]) / 2
1820 * - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1821 *
1822 * of them, which corresponds to the maximum size of a hibernation image.
1823 *
1824 * If image_size is set below the number following from the above formula,
1825 * the preallocation of memory is continued until the total number of saveable
1826 * pages in the system is below the requested image size or the minimum
1827 * acceptable image size returned by minimum_image_size(), whichever is greater.
1828 */
1829int hibernate_preallocate_memory(void)
1830{
1831 struct zone *zone;
1832 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1833 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1834 ktime_t start, stop;
1835 int error;
1836
1837 pr_info("Preallocating image memory\n");
1838 start = ktime_get();
1839
1840 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1841 if (error) {
1842 pr_err("Cannot allocate original bitmap\n");
1843 goto err_out;
1844 }
1845
1846 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1847 if (error) {
1848 pr_err("Cannot allocate copy bitmap\n");
1849 goto err_out;
1850 }
1851
1852 error = memory_bm_create(&zero_bm, GFP_IMAGE, PG_ANY);
1853 if (error) {
1854 pr_err("Cannot allocate zero bitmap\n");
1855 goto err_out;
1856 }
1857
1858 alloc_normal = 0;
1859 alloc_highmem = 0;
1860 nr_zero_pages = 0;
1861
1862 /* Count the number of saveable data pages. */
1863 save_highmem = count_highmem_pages();
1864 saveable = count_data_pages();
1865
1866 /*
1867 * Compute the total number of page frames we can use (count) and the
1868 * number of pages needed for image metadata (size).
1869 */
1870 count = saveable;
1871 saveable += save_highmem;
1872 highmem = save_highmem;
1873 size = 0;
1874 for_each_populated_zone(zone) {
1875 size += snapshot_additional_pages(zone);
1876 if (is_highmem(zone))
1877 highmem += zone_page_state(zone, NR_FREE_PAGES);
1878 else
1879 count += zone_page_state(zone, NR_FREE_PAGES);
1880 }
1881 avail_normal = count;
1882 count += highmem;
1883 count -= totalreserve_pages;
1884
1885 /* Compute the maximum number of saveable pages to leave in memory. */
1886 max_size = (count - (size + PAGES_FOR_IO)) / 2
1887 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1888 /* Compute the desired number of image pages specified by image_size. */
1889 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1890 if (size > max_size)
1891 size = max_size;
1892 /*
1893 * If the desired number of image pages is at least as large as the
1894 * current number of saveable pages in memory, allocate page frames for
1895 * the image and we're done.
1896 */
1897 if (size >= saveable) {
1898 pages = preallocate_image_highmem(save_highmem);
1899 pages += preallocate_image_memory(saveable - pages, avail_normal);
1900 goto out;
1901 }
1902
1903 /* Estimate the minimum size of the image. */
1904 pages = minimum_image_size(saveable);
1905 /*
1906 * To avoid excessive pressure on the normal zone, leave room in it to
1907 * accommodate an image of the minimum size (unless it's already too
1908 * small, in which case don't preallocate pages from it at all).
1909 */
1910 if (avail_normal > pages)
1911 avail_normal -= pages;
1912 else
1913 avail_normal = 0;
1914 if (size < pages)
1915 size = min_t(unsigned long, pages, max_size);
1916
1917 /*
1918 * Let the memory management subsystem know that we're going to need a
1919 * large number of page frames to allocate and make it free some memory.
1920 * NOTE: If this is not done, performance will be hurt badly in some
1921 * test cases.
1922 */
1923 shrink_all_memory(saveable - size);
1924
1925 /*
1926 * The number of saveable pages in memory was too high, so apply some
1927 * pressure to decrease it. First, make room for the largest possible
1928 * image and fail if that doesn't work. Next, try to decrease the size
1929 * of the image as much as indicated by 'size' using allocations from
1930 * highmem and non-highmem zones separately.
1931 */
1932 pages_highmem = preallocate_image_highmem(highmem / 2);
1933 alloc = count - max_size;
1934 if (alloc > pages_highmem)
1935 alloc -= pages_highmem;
1936 else
1937 alloc = 0;
1938 pages = preallocate_image_memory(alloc, avail_normal);
1939 if (pages < alloc) {
1940 /* We have exhausted non-highmem pages, try highmem. */
1941 alloc -= pages;
1942 pages += pages_highmem;
1943 pages_highmem = preallocate_image_highmem(alloc);
1944 if (pages_highmem < alloc) {
1945 pr_err("Image allocation is %lu pages short\n",
1946 alloc - pages_highmem);
1947 goto err_out;
1948 }
1949 pages += pages_highmem;
1950 /*
1951 * size is the desired number of saveable pages to leave in
1952 * memory, so try to preallocate (all memory - size) pages.
1953 */
1954 alloc = (count - pages) - size;
1955 pages += preallocate_image_highmem(alloc);
1956 } else {
1957 /*
1958 * There are approximately max_size saveable pages at this point
1959 * and we want to reduce this number down to size.
1960 */
1961 alloc = max_size - size;
1962 size = preallocate_highmem_fraction(alloc, highmem, count);
1963 pages_highmem += size;
1964 alloc -= size;
1965 size = preallocate_image_memory(alloc, avail_normal);
1966 pages_highmem += preallocate_image_highmem(alloc - size);
1967 pages += pages_highmem + size;
1968 }
1969
1970 /*
1971 * We only need as many page frames for the image as there are saveable
1972 * pages in memory, but we have allocated more. Release the excessive
1973 * ones now.
1974 */
1975 pages -= free_unnecessary_pages();
1976
1977 out:
1978 stop = ktime_get();
1979 pr_info("Allocated %lu pages for snapshot\n", pages);
1980 swsusp_show_speed(start, stop, pages, "Allocated");
1981
1982 return 0;
1983
1984 err_out:
1985 swsusp_free();
1986 return -ENOMEM;
1987}
1988
1989#ifdef CONFIG_HIGHMEM
1990/**
1991 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1992 *
1993 * Compute the number of non-highmem pages that will be necessary for creating
1994 * copies of highmem pages.
1995 */
1996static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1997{
1998 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1999
2000 if (free_highmem >= nr_highmem)
2001 nr_highmem = 0;
2002 else
2003 nr_highmem -= free_highmem;
2004
2005 return nr_highmem;
2006}
2007#else
2008static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
2009#endif /* CONFIG_HIGHMEM */
2010
2011/**
2012 * enough_free_mem - Check if there is enough free memory for the image.
2013 */
2014static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
2015{
2016 struct zone *zone;
2017 unsigned int free = alloc_normal;
2018
2019 for_each_populated_zone(zone)
2020 if (!is_highmem(zone))
2021 free += zone_page_state(zone, NR_FREE_PAGES);
2022
2023 nr_pages += count_pages_for_highmem(nr_highmem);
2024 pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
2025 nr_pages, PAGES_FOR_IO, free);
2026
2027 return free > nr_pages + PAGES_FOR_IO;
2028}
2029
2030#ifdef CONFIG_HIGHMEM
2031/**
2032 * get_highmem_buffer - Allocate a buffer for highmem pages.
2033 *
2034 * If there are some highmem pages in the hibernation image, we may need a
2035 * buffer to copy them and/or load their data.
2036 */
2037static inline int get_highmem_buffer(int safe_needed)
2038{
2039 buffer = get_image_page(GFP_ATOMIC, safe_needed);
2040 return buffer ? 0 : -ENOMEM;
2041}
2042
2043/**
2044 * alloc_highmem_pages - Allocate some highmem pages for the image.
2045 *
2046 * Try to allocate as many pages as needed, but if the number of free highmem
2047 * pages is less than that, allocate them all.
2048 */
2049static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
2050 unsigned int nr_highmem)
2051{
2052 unsigned int to_alloc = count_free_highmem_pages();
2053
2054 if (to_alloc > nr_highmem)
2055 to_alloc = nr_highmem;
2056
2057 nr_highmem -= to_alloc;
2058 while (to_alloc-- > 0) {
2059 struct page *page;
2060
2061 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
2062 memory_bm_set_bit(bm, page_to_pfn(page));
2063 }
2064 return nr_highmem;
2065}
2066#else
2067static inline int get_highmem_buffer(int safe_needed) { return 0; }
2068
2069static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
2070 unsigned int n) { return 0; }
2071#endif /* CONFIG_HIGHMEM */
2072
2073/**
2074 * swsusp_alloc - Allocate memory for hibernation image.
2075 *
2076 * We first try to allocate as many highmem pages as there are
2077 * saveable highmem pages in the system. If that fails, we allocate
2078 * non-highmem pages for the copies of the remaining highmem ones.
2079 *
2080 * In this approach it is likely that the copies of highmem pages will
2081 * also be located in the high memory, because of the way in which
2082 * copy_data_pages() works.
2083 */
2084static int swsusp_alloc(struct memory_bitmap *copy_bm,
2085 unsigned int nr_pages, unsigned int nr_highmem)
2086{
2087 if (nr_highmem > 0) {
2088 if (get_highmem_buffer(PG_ANY))
2089 goto err_out;
2090 if (nr_highmem > alloc_highmem) {
2091 nr_highmem -= alloc_highmem;
2092 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
2093 }
2094 }
2095 if (nr_pages > alloc_normal) {
2096 nr_pages -= alloc_normal;
2097 while (nr_pages-- > 0) {
2098 struct page *page;
2099
2100 page = alloc_image_page(GFP_ATOMIC);
2101 if (!page)
2102 goto err_out;
2103 memory_bm_set_bit(copy_bm, page_to_pfn(page));
2104 }
2105 }
2106
2107 return 0;
2108
2109 err_out:
2110 swsusp_free();
2111 return -ENOMEM;
2112}
2113
2114asmlinkage __visible int swsusp_save(void)
2115{
2116 unsigned int nr_pages, nr_highmem;
2117
2118 pr_info("Creating image:\n");
2119
2120 drain_local_pages(NULL);
2121 nr_pages = count_data_pages();
2122 nr_highmem = count_highmem_pages();
2123 pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
2124
2125 if (!enough_free_mem(nr_pages, nr_highmem)) {
2126 pr_err("Not enough free memory\n");
2127 return -ENOMEM;
2128 }
2129
2130 if (swsusp_alloc(©_bm, nr_pages, nr_highmem)) {
2131 pr_err("Memory allocation failed\n");
2132 return -ENOMEM;
2133 }
2134
2135 /*
2136 * During allocating of suspend pagedir, new cold pages may appear.
2137 * Kill them.
2138 */
2139 drain_local_pages(NULL);
2140 nr_copy_pages = copy_data_pages(©_bm, &orig_bm, &zero_bm);
2141
2142 /*
2143 * End of critical section. From now on, we can write to memory,
2144 * but we should not touch disk. This specially means we must _not_
2145 * touch swap space! Except we must write out our image of course.
2146 */
2147 nr_pages += nr_highmem;
2148 /* We don't actually copy the zero pages */
2149 nr_zero_pages = nr_pages - nr_copy_pages;
2150 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2151
2152 pr_info("Image created (%d pages copied, %d zero pages)\n", nr_copy_pages, nr_zero_pages);
2153
2154 return 0;
2155}
2156
2157#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2158static int init_header_complete(struct swsusp_info *info)
2159{
2160 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2161 info->version_code = LINUX_VERSION_CODE;
2162 return 0;
2163}
2164
2165static const char *check_image_kernel(struct swsusp_info *info)
2166{
2167 if (info->version_code != LINUX_VERSION_CODE)
2168 return "kernel version";
2169 if (strcmp(info->uts.sysname,init_utsname()->sysname))
2170 return "system type";
2171 if (strcmp(info->uts.release,init_utsname()->release))
2172 return "kernel release";
2173 if (strcmp(info->uts.version,init_utsname()->version))
2174 return "version";
2175 if (strcmp(info->uts.machine,init_utsname()->machine))
2176 return "machine";
2177 return NULL;
2178}
2179#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2180
2181unsigned long snapshot_get_image_size(void)
2182{
2183 return nr_copy_pages + nr_meta_pages + 1;
2184}
2185
2186static int init_header(struct swsusp_info *info)
2187{
2188 memset(info, 0, sizeof(struct swsusp_info));
2189 info->num_physpages = get_num_physpages();
2190 info->image_pages = nr_copy_pages;
2191 info->pages = snapshot_get_image_size();
2192 info->size = info->pages;
2193 info->size <<= PAGE_SHIFT;
2194 return init_header_complete(info);
2195}
2196
2197#define ENCODED_PFN_ZERO_FLAG ((unsigned long)1 << (BITS_PER_LONG - 1))
2198#define ENCODED_PFN_MASK (~ENCODED_PFN_ZERO_FLAG)
2199
2200/**
2201 * pack_pfns - Prepare PFNs for saving.
2202 * @bm: Memory bitmap.
2203 * @buf: Memory buffer to store the PFNs in.
2204 * @zero_bm: Memory bitmap containing PFNs of zero pages.
2205 *
2206 * PFNs corresponding to set bits in @bm are stored in the area of memory
2207 * pointed to by @buf (1 page at a time). Pages which were filled with only
2208 * zeros will have the highest bit set in the packed format to distinguish
2209 * them from PFNs which will be contained in the image file.
2210 */
2211static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm,
2212 struct memory_bitmap *zero_bm)
2213{
2214 int j;
2215
2216 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2217 buf[j] = memory_bm_next_pfn(bm);
2218 if (unlikely(buf[j] == BM_END_OF_MAP))
2219 break;
2220 if (memory_bm_test_bit(zero_bm, buf[j]))
2221 buf[j] |= ENCODED_PFN_ZERO_FLAG;
2222 }
2223}
2224
2225/**
2226 * snapshot_read_next - Get the address to read the next image page from.
2227 * @handle: Snapshot handle to be used for the reading.
2228 *
2229 * On the first call, @handle should point to a zeroed snapshot_handle
2230 * structure. The structure gets populated then and a pointer to it should be
2231 * passed to this function every next time.
2232 *
2233 * On success, the function returns a positive number. Then, the caller
2234 * is allowed to read up to the returned number of bytes from the memory
2235 * location computed by the data_of() macro.
2236 *
2237 * The function returns 0 to indicate the end of the data stream condition,
2238 * and negative numbers are returned on errors. If that happens, the structure
2239 * pointed to by @handle is not updated and should not be used any more.
2240 */
2241int snapshot_read_next(struct snapshot_handle *handle)
2242{
2243 if (handle->cur > nr_meta_pages + nr_copy_pages)
2244 return 0;
2245
2246 if (!buffer) {
2247 /* This makes the buffer be freed by swsusp_free() */
2248 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2249 if (!buffer)
2250 return -ENOMEM;
2251 }
2252 if (!handle->cur) {
2253 int error;
2254
2255 error = init_header((struct swsusp_info *)buffer);
2256 if (error)
2257 return error;
2258 handle->buffer = buffer;
2259 memory_bm_position_reset(&orig_bm);
2260 memory_bm_position_reset(©_bm);
2261 } else if (handle->cur <= nr_meta_pages) {
2262 clear_page(buffer);
2263 pack_pfns(buffer, &orig_bm, &zero_bm);
2264 } else {
2265 struct page *page;
2266
2267 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2268 if (PageHighMem(page)) {
2269 /*
2270 * Highmem pages are copied to the buffer,
2271 * because we can't return with a kmapped
2272 * highmem page (we may not be called again).
2273 */
2274 void *kaddr;
2275
2276 kaddr = kmap_atomic(page);
2277 copy_page(buffer, kaddr);
2278 kunmap_atomic(kaddr);
2279 handle->buffer = buffer;
2280 } else {
2281 handle->buffer = page_address(page);
2282 }
2283 }
2284 handle->cur++;
2285 return PAGE_SIZE;
2286}
2287
2288static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2289 struct memory_bitmap *src)
2290{
2291 unsigned long pfn;
2292
2293 memory_bm_position_reset(src);
2294 pfn = memory_bm_next_pfn(src);
2295 while (pfn != BM_END_OF_MAP) {
2296 memory_bm_set_bit(dst, pfn);
2297 pfn = memory_bm_next_pfn(src);
2298 }
2299}
2300
2301/**
2302 * mark_unsafe_pages - Mark pages that were used before hibernation.
2303 *
2304 * Mark the pages that cannot be used for storing the image during restoration,
2305 * because they conflict with the pages that had been used before hibernation.
2306 */
2307static void mark_unsafe_pages(struct memory_bitmap *bm)
2308{
2309 unsigned long pfn;
2310
2311 /* Clear the "free"/"unsafe" bit for all PFNs */
2312 memory_bm_position_reset(free_pages_map);
2313 pfn = memory_bm_next_pfn(free_pages_map);
2314 while (pfn != BM_END_OF_MAP) {
2315 memory_bm_clear_current(free_pages_map);
2316 pfn = memory_bm_next_pfn(free_pages_map);
2317 }
2318
2319 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2320 duplicate_memory_bitmap(free_pages_map, bm);
2321
2322 allocated_unsafe_pages = 0;
2323}
2324
2325static int check_header(struct swsusp_info *info)
2326{
2327 const char *reason;
2328
2329 reason = check_image_kernel(info);
2330 if (!reason && info->num_physpages != get_num_physpages())
2331 reason = "memory size";
2332 if (reason) {
2333 pr_err("Image mismatch: %s\n", reason);
2334 return -EPERM;
2335 }
2336 return 0;
2337}
2338
2339/**
2340 * load_header - Check the image header and copy the data from it.
2341 */
2342static int load_header(struct swsusp_info *info)
2343{
2344 int error;
2345
2346 restore_pblist = NULL;
2347 error = check_header(info);
2348 if (!error) {
2349 nr_copy_pages = info->image_pages;
2350 nr_meta_pages = info->pages - info->image_pages - 1;
2351 }
2352 return error;
2353}
2354
2355/**
2356 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2357 * @bm: Memory bitmap.
2358 * @buf: Area of memory containing the PFNs.
2359 * @zero_bm: Memory bitmap with the zero PFNs marked.
2360 *
2361 * For each element of the array pointed to by @buf (1 page at a time), set the
2362 * corresponding bit in @bm. If the page was originally populated with only
2363 * zeros then a corresponding bit will also be set in @zero_bm.
2364 */
2365static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm,
2366 struct memory_bitmap *zero_bm)
2367{
2368 unsigned long decoded_pfn;
2369 bool zero;
2370 int j;
2371
2372 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2373 if (unlikely(buf[j] == BM_END_OF_MAP))
2374 break;
2375
2376 zero = !!(buf[j] & ENCODED_PFN_ZERO_FLAG);
2377 decoded_pfn = buf[j] & ENCODED_PFN_MASK;
2378 if (pfn_valid(decoded_pfn) && memory_bm_pfn_present(bm, decoded_pfn)) {
2379 memory_bm_set_bit(bm, decoded_pfn);
2380 if (zero) {
2381 memory_bm_set_bit(zero_bm, decoded_pfn);
2382 nr_zero_pages++;
2383 }
2384 } else {
2385 if (!pfn_valid(decoded_pfn))
2386 pr_err(FW_BUG "Memory map mismatch at 0x%llx after hibernation\n",
2387 (unsigned long long)PFN_PHYS(decoded_pfn));
2388 return -EFAULT;
2389 }
2390 }
2391
2392 return 0;
2393}
2394
2395#ifdef CONFIG_HIGHMEM
2396/*
2397 * struct highmem_pbe is used for creating the list of highmem pages that
2398 * should be restored atomically during the resume from disk, because the page
2399 * frames they have occupied before the suspend are in use.
2400 */
2401struct highmem_pbe {
2402 struct page *copy_page; /* data is here now */
2403 struct page *orig_page; /* data was here before the suspend */
2404 struct highmem_pbe *next;
2405};
2406
2407/*
2408 * List of highmem PBEs needed for restoring the highmem pages that were
2409 * allocated before the suspend and included in the suspend image, but have
2410 * also been allocated by the "resume" kernel, so their contents cannot be
2411 * written directly to their "original" page frames.
2412 */
2413static struct highmem_pbe *highmem_pblist;
2414
2415/**
2416 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2417 * @bm: Memory bitmap.
2418 *
2419 * The bits in @bm that correspond to image pages are assumed to be set.
2420 */
2421static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2422{
2423 unsigned long pfn;
2424 unsigned int cnt = 0;
2425
2426 memory_bm_position_reset(bm);
2427 pfn = memory_bm_next_pfn(bm);
2428 while (pfn != BM_END_OF_MAP) {
2429 if (PageHighMem(pfn_to_page(pfn)))
2430 cnt++;
2431
2432 pfn = memory_bm_next_pfn(bm);
2433 }
2434 return cnt;
2435}
2436
2437static unsigned int safe_highmem_pages;
2438
2439static struct memory_bitmap *safe_highmem_bm;
2440
2441/**
2442 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2443 * @bm: Pointer to an uninitialized memory bitmap structure.
2444 * @nr_highmem_p: Pointer to the number of highmem image pages.
2445 *
2446 * Try to allocate as many highmem pages as there are highmem image pages
2447 * (@nr_highmem_p points to the variable containing the number of highmem image
2448 * pages). The pages that are "safe" (ie. will not be overwritten when the
2449 * hibernation image is restored entirely) have the corresponding bits set in
2450 * @bm (it must be uninitialized).
2451 *
2452 * NOTE: This function should not be called if there are no highmem image pages.
2453 */
2454static int prepare_highmem_image(struct memory_bitmap *bm,
2455 unsigned int *nr_highmem_p)
2456{
2457 unsigned int to_alloc;
2458
2459 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2460 return -ENOMEM;
2461
2462 if (get_highmem_buffer(PG_SAFE))
2463 return -ENOMEM;
2464
2465 to_alloc = count_free_highmem_pages();
2466 if (to_alloc > *nr_highmem_p)
2467 to_alloc = *nr_highmem_p;
2468 else
2469 *nr_highmem_p = to_alloc;
2470
2471 safe_highmem_pages = 0;
2472 while (to_alloc-- > 0) {
2473 struct page *page;
2474
2475 page = alloc_page(__GFP_HIGHMEM);
2476 if (!swsusp_page_is_free(page)) {
2477 /* The page is "safe", set its bit the bitmap */
2478 memory_bm_set_bit(bm, page_to_pfn(page));
2479 safe_highmem_pages++;
2480 }
2481 /* Mark the page as allocated */
2482 swsusp_set_page_forbidden(page);
2483 swsusp_set_page_free(page);
2484 }
2485 memory_bm_position_reset(bm);
2486 safe_highmem_bm = bm;
2487 return 0;
2488}
2489
2490static struct page *last_highmem_page;
2491
2492/**
2493 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2494 *
2495 * For a given highmem image page get a buffer that suspend_write_next() should
2496 * return to its caller to write to.
2497 *
2498 * If the page is to be saved to its "original" page frame or a copy of
2499 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2500 * the copy of the page is to be made in normal memory, so the address of
2501 * the copy is returned.
2502 *
2503 * If @buffer is returned, the caller of suspend_write_next() will write
2504 * the page's contents to @buffer, so they will have to be copied to the
2505 * right location on the next call to suspend_write_next() and it is done
2506 * with the help of copy_last_highmem_page(). For this purpose, if
2507 * @buffer is returned, @last_highmem_page is set to the page to which
2508 * the data will have to be copied from @buffer.
2509 */
2510static void *get_highmem_page_buffer(struct page *page,
2511 struct chain_allocator *ca)
2512{
2513 struct highmem_pbe *pbe;
2514 void *kaddr;
2515
2516 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2517 /*
2518 * We have allocated the "original" page frame and we can
2519 * use it directly to store the loaded page.
2520 */
2521 last_highmem_page = page;
2522 return buffer;
2523 }
2524 /*
2525 * The "original" page frame has not been allocated and we have to
2526 * use a "safe" page frame to store the loaded page.
2527 */
2528 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2529 if (!pbe) {
2530 swsusp_free();
2531 return ERR_PTR(-ENOMEM);
2532 }
2533 pbe->orig_page = page;
2534 if (safe_highmem_pages > 0) {
2535 struct page *tmp;
2536
2537 /* Copy of the page will be stored in high memory */
2538 kaddr = buffer;
2539 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2540 safe_highmem_pages--;
2541 last_highmem_page = tmp;
2542 pbe->copy_page = tmp;
2543 } else {
2544 /* Copy of the page will be stored in normal memory */
2545 kaddr = __get_safe_page(ca->gfp_mask);
2546 if (!kaddr)
2547 return ERR_PTR(-ENOMEM);
2548 pbe->copy_page = virt_to_page(kaddr);
2549 }
2550 pbe->next = highmem_pblist;
2551 highmem_pblist = pbe;
2552 return kaddr;
2553}
2554
2555/**
2556 * copy_last_highmem_page - Copy most the most recent highmem image page.
2557 *
2558 * Copy the contents of a highmem image from @buffer, where the caller of
2559 * snapshot_write_next() has stored them, to the right location represented by
2560 * @last_highmem_page .
2561 */
2562static void copy_last_highmem_page(void)
2563{
2564 if (last_highmem_page) {
2565 void *dst;
2566
2567 dst = kmap_atomic(last_highmem_page);
2568 copy_page(dst, buffer);
2569 kunmap_atomic(dst);
2570 last_highmem_page = NULL;
2571 }
2572}
2573
2574static inline int last_highmem_page_copied(void)
2575{
2576 return !last_highmem_page;
2577}
2578
2579static inline void free_highmem_data(void)
2580{
2581 if (safe_highmem_bm)
2582 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2583
2584 if (buffer)
2585 free_image_page(buffer, PG_UNSAFE_CLEAR);
2586}
2587#else
2588static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2589
2590static inline int prepare_highmem_image(struct memory_bitmap *bm,
2591 unsigned int *nr_highmem_p) { return 0; }
2592
2593static inline void *get_highmem_page_buffer(struct page *page,
2594 struct chain_allocator *ca)
2595{
2596 return ERR_PTR(-EINVAL);
2597}
2598
2599static inline void copy_last_highmem_page(void) {}
2600static inline int last_highmem_page_copied(void) { return 1; }
2601static inline void free_highmem_data(void) {}
2602#endif /* CONFIG_HIGHMEM */
2603
2604#define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2605
2606/**
2607 * prepare_image - Make room for loading hibernation image.
2608 * @new_bm: Uninitialized memory bitmap structure.
2609 * @bm: Memory bitmap with unsafe pages marked.
2610 * @zero_bm: Memory bitmap containing the zero pages.
2611 *
2612 * Use @bm to mark the pages that will be overwritten in the process of
2613 * restoring the system memory state from the suspend image ("unsafe" pages)
2614 * and allocate memory for the image.
2615 *
2616 * The idea is to allocate a new memory bitmap first and then allocate
2617 * as many pages as needed for image data, but without specifying what those
2618 * pages will be used for just yet. Instead, we mark them all as allocated and
2619 * create a lists of "safe" pages to be used later. On systems with high
2620 * memory a list of "safe" highmem pages is created too.
2621 *
2622 * Because it was not known which pages were unsafe when @zero_bm was created,
2623 * make a copy of it and recreate it within safe pages.
2624 */
2625static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm,
2626 struct memory_bitmap *zero_bm)
2627{
2628 unsigned int nr_pages, nr_highmem;
2629 struct memory_bitmap tmp;
2630 struct linked_page *lp;
2631 int error;
2632
2633 /* If there is no highmem, the buffer will not be necessary */
2634 free_image_page(buffer, PG_UNSAFE_CLEAR);
2635 buffer = NULL;
2636
2637 nr_highmem = count_highmem_image_pages(bm);
2638 mark_unsafe_pages(bm);
2639
2640 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2641 if (error)
2642 goto Free;
2643
2644 duplicate_memory_bitmap(new_bm, bm);
2645 memory_bm_free(bm, PG_UNSAFE_KEEP);
2646
2647 /* Make a copy of zero_bm so it can be created in safe pages */
2648 error = memory_bm_create(&tmp, GFP_ATOMIC, PG_SAFE);
2649 if (error)
2650 goto Free;
2651
2652 duplicate_memory_bitmap(&tmp, zero_bm);
2653 memory_bm_free(zero_bm, PG_UNSAFE_KEEP);
2654
2655 /* Recreate zero_bm in safe pages */
2656 error = memory_bm_create(zero_bm, GFP_ATOMIC, PG_SAFE);
2657 if (error)
2658 goto Free;
2659
2660 duplicate_memory_bitmap(zero_bm, &tmp);
2661 memory_bm_free(&tmp, PG_UNSAFE_CLEAR);
2662 /* At this point zero_bm is in safe pages and it can be used for restoring. */
2663
2664 if (nr_highmem > 0) {
2665 error = prepare_highmem_image(bm, &nr_highmem);
2666 if (error)
2667 goto Free;
2668 }
2669 /*
2670 * Reserve some safe pages for potential later use.
2671 *
2672 * NOTE: This way we make sure there will be enough safe pages for the
2673 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2674 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2675 *
2676 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2677 */
2678 nr_pages = (nr_zero_pages + nr_copy_pages) - nr_highmem - allocated_unsafe_pages;
2679 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2680 while (nr_pages > 0) {
2681 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2682 if (!lp) {
2683 error = -ENOMEM;
2684 goto Free;
2685 }
2686 lp->next = safe_pages_list;
2687 safe_pages_list = lp;
2688 nr_pages--;
2689 }
2690 /* Preallocate memory for the image */
2691 nr_pages = (nr_zero_pages + nr_copy_pages) - nr_highmem - allocated_unsafe_pages;
2692 while (nr_pages > 0) {
2693 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2694 if (!lp) {
2695 error = -ENOMEM;
2696 goto Free;
2697 }
2698 if (!swsusp_page_is_free(virt_to_page(lp))) {
2699 /* The page is "safe", add it to the list */
2700 lp->next = safe_pages_list;
2701 safe_pages_list = lp;
2702 }
2703 /* Mark the page as allocated */
2704 swsusp_set_page_forbidden(virt_to_page(lp));
2705 swsusp_set_page_free(virt_to_page(lp));
2706 nr_pages--;
2707 }
2708 return 0;
2709
2710 Free:
2711 swsusp_free();
2712 return error;
2713}
2714
2715/**
2716 * get_buffer - Get the address to store the next image data page.
2717 *
2718 * Get the address that snapshot_write_next() should return to its caller to
2719 * write to.
2720 */
2721static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2722{
2723 struct pbe *pbe;
2724 struct page *page;
2725 unsigned long pfn = memory_bm_next_pfn(bm);
2726
2727 if (pfn == BM_END_OF_MAP)
2728 return ERR_PTR(-EFAULT);
2729
2730 page = pfn_to_page(pfn);
2731 if (PageHighMem(page))
2732 return get_highmem_page_buffer(page, ca);
2733
2734 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2735 /*
2736 * We have allocated the "original" page frame and we can
2737 * use it directly to store the loaded page.
2738 */
2739 return page_address(page);
2740
2741 /*
2742 * The "original" page frame has not been allocated and we have to
2743 * use a "safe" page frame to store the loaded page.
2744 */
2745 pbe = chain_alloc(ca, sizeof(struct pbe));
2746 if (!pbe) {
2747 swsusp_free();
2748 return ERR_PTR(-ENOMEM);
2749 }
2750 pbe->orig_address = page_address(page);
2751 pbe->address = __get_safe_page(ca->gfp_mask);
2752 if (!pbe->address)
2753 return ERR_PTR(-ENOMEM);
2754 pbe->next = restore_pblist;
2755 restore_pblist = pbe;
2756 return pbe->address;
2757}
2758
2759/**
2760 * snapshot_write_next - Get the address to store the next image page.
2761 * @handle: Snapshot handle structure to guide the writing.
2762 *
2763 * On the first call, @handle should point to a zeroed snapshot_handle
2764 * structure. The structure gets populated then and a pointer to it should be
2765 * passed to this function every next time.
2766 *
2767 * On success, the function returns a positive number. Then, the caller
2768 * is allowed to write up to the returned number of bytes to the memory
2769 * location computed by the data_of() macro.
2770 *
2771 * The function returns 0 to indicate the "end of file" condition. Negative
2772 * numbers are returned on errors, in which cases the structure pointed to by
2773 * @handle is not updated and should not be used any more.
2774 */
2775int snapshot_write_next(struct snapshot_handle *handle)
2776{
2777 static struct chain_allocator ca;
2778 int error;
2779
2780next:
2781 /* Check if we have already loaded the entire image */
2782 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages + nr_zero_pages)
2783 return 0;
2784
2785 if (!handle->cur) {
2786 if (!buffer)
2787 /* This makes the buffer be freed by swsusp_free() */
2788 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2789
2790 if (!buffer)
2791 return -ENOMEM;
2792
2793 handle->buffer = buffer;
2794 } else if (handle->cur == 1) {
2795 error = load_header(buffer);
2796 if (error)
2797 return error;
2798
2799 safe_pages_list = NULL;
2800
2801 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2802 if (error)
2803 return error;
2804
2805 error = memory_bm_create(&zero_bm, GFP_ATOMIC, PG_ANY);
2806 if (error)
2807 return error;
2808
2809 nr_zero_pages = 0;
2810
2811 hibernate_restore_protection_begin();
2812 } else if (handle->cur <= nr_meta_pages + 1) {
2813 error = unpack_orig_pfns(buffer, ©_bm, &zero_bm);
2814 if (error)
2815 return error;
2816
2817 if (handle->cur == nr_meta_pages + 1) {
2818 error = prepare_image(&orig_bm, ©_bm, &zero_bm);
2819 if (error)
2820 return error;
2821
2822 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2823 memory_bm_position_reset(&orig_bm);
2824 memory_bm_position_reset(&zero_bm);
2825 restore_pblist = NULL;
2826 handle->buffer = get_buffer(&orig_bm, &ca);
2827 if (IS_ERR(handle->buffer))
2828 return PTR_ERR(handle->buffer);
2829 }
2830 } else {
2831 copy_last_highmem_page();
2832 error = hibernate_restore_protect_page(handle->buffer);
2833 if (error)
2834 return error;
2835 handle->buffer = get_buffer(&orig_bm, &ca);
2836 if (IS_ERR(handle->buffer))
2837 return PTR_ERR(handle->buffer);
2838 }
2839 handle->sync_read = (handle->buffer == buffer);
2840 handle->cur++;
2841
2842 /* Zero pages were not included in the image, memset it and move on. */
2843 if (handle->cur > nr_meta_pages + 1 &&
2844 memory_bm_test_bit(&zero_bm, memory_bm_get_current(&orig_bm))) {
2845 memset(handle->buffer, 0, PAGE_SIZE);
2846 goto next;
2847 }
2848
2849 return PAGE_SIZE;
2850}
2851
2852/**
2853 * snapshot_write_finalize - Complete the loading of a hibernation image.
2854 *
2855 * Must be called after the last call to snapshot_write_next() in case the last
2856 * page in the image happens to be a highmem page and its contents should be
2857 * stored in highmem. Additionally, it recycles bitmap memory that's not
2858 * necessary any more.
2859 */
2860int snapshot_write_finalize(struct snapshot_handle *handle)
2861{
2862 int error;
2863
2864 copy_last_highmem_page();
2865 error = hibernate_restore_protect_page(handle->buffer);
2866 /* Do that only if we have loaded the image entirely */
2867 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages + nr_zero_pages) {
2868 memory_bm_recycle(&orig_bm);
2869 free_highmem_data();
2870 }
2871 return error;
2872}
2873
2874int snapshot_image_loaded(struct snapshot_handle *handle)
2875{
2876 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2877 handle->cur <= nr_meta_pages + nr_copy_pages + nr_zero_pages);
2878}
2879
2880#ifdef CONFIG_HIGHMEM
2881/* Assumes that @buf is ready and points to a "safe" page */
2882static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2883 void *buf)
2884{
2885 void *kaddr1, *kaddr2;
2886
2887 kaddr1 = kmap_atomic(p1);
2888 kaddr2 = kmap_atomic(p2);
2889 copy_page(buf, kaddr1);
2890 copy_page(kaddr1, kaddr2);
2891 copy_page(kaddr2, buf);
2892 kunmap_atomic(kaddr2);
2893 kunmap_atomic(kaddr1);
2894}
2895
2896/**
2897 * restore_highmem - Put highmem image pages into their original locations.
2898 *
2899 * For each highmem page that was in use before hibernation and is included in
2900 * the image, and also has been allocated by the "restore" kernel, swap its
2901 * current contents with the previous (ie. "before hibernation") ones.
2902 *
2903 * If the restore eventually fails, we can call this function once again and
2904 * restore the highmem state as seen by the restore kernel.
2905 */
2906int restore_highmem(void)
2907{
2908 struct highmem_pbe *pbe = highmem_pblist;
2909 void *buf;
2910
2911 if (!pbe)
2912 return 0;
2913
2914 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2915 if (!buf)
2916 return -ENOMEM;
2917
2918 while (pbe) {
2919 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2920 pbe = pbe->next;
2921 }
2922 free_image_page(buf, PG_UNSAFE_CLEAR);
2923 return 0;
2924}
2925#endif /* CONFIG_HIGHMEM */