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: " 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/pgtable.h>
38#include <asm/tlbflush.h>
39#include <asm/io.h>
40
41#include "power.h"
42
43#if defined(CONFIG_STRICT_KERNEL_RWX) && defined(CONFIG_ARCH_HAS_SET_MEMORY)
44static bool hibernate_restore_protection;
45static bool hibernate_restore_protection_active;
46
47void enable_restore_image_protection(void)
48{
49 hibernate_restore_protection = true;
50}
51
52static inline void hibernate_restore_protection_begin(void)
53{
54 hibernate_restore_protection_active = hibernate_restore_protection;
55}
56
57static inline void hibernate_restore_protection_end(void)
58{
59 hibernate_restore_protection_active = false;
60}
61
62static inline void hibernate_restore_protect_page(void *page_address)
63{
64 if (hibernate_restore_protection_active)
65 set_memory_ro((unsigned long)page_address, 1);
66}
67
68static inline void hibernate_restore_unprotect_page(void *page_address)
69{
70 if (hibernate_restore_protection_active)
71 set_memory_rw((unsigned long)page_address, 1);
72}
73#else
74static inline void hibernate_restore_protection_begin(void) {}
75static inline void hibernate_restore_protection_end(void) {}
76static inline void hibernate_restore_protect_page(void *page_address) {}
77static inline void hibernate_restore_unprotect_page(void *page_address) {}
78#endif /* CONFIG_STRICT_KERNEL_RWX && CONFIG_ARCH_HAS_SET_MEMORY */
79
80static int swsusp_page_is_free(struct page *);
81static void swsusp_set_page_forbidden(struct page *);
82static void swsusp_unset_page_forbidden(struct page *);
83
84/*
85 * Number of bytes to reserve for memory allocations made by device drivers
86 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
87 * cause image creation to fail (tunable via /sys/power/reserved_size).
88 */
89unsigned long reserved_size;
90
91void __init hibernate_reserved_size_init(void)
92{
93 reserved_size = SPARE_PAGES * PAGE_SIZE;
94}
95
96/*
97 * Preferred image size in bytes (tunable via /sys/power/image_size).
98 * When it is set to N, swsusp will do its best to ensure the image
99 * size will not exceed N bytes, but if that is impossible, it will
100 * try to create the smallest image possible.
101 */
102unsigned long image_size;
103
104void __init hibernate_image_size_init(void)
105{
106 image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
107}
108
109/*
110 * List of PBEs needed for restoring the pages that were allocated before
111 * the suspend and included in the suspend image, but have also been
112 * allocated by the "resume" kernel, so their contents cannot be written
113 * directly to their "original" page frames.
114 */
115struct pbe *restore_pblist;
116
117/* struct linked_page is used to build chains of pages */
118
119#define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
120
121struct linked_page {
122 struct linked_page *next;
123 char data[LINKED_PAGE_DATA_SIZE];
124} __packed;
125
126/*
127 * List of "safe" pages (ie. pages that were not used by the image kernel
128 * before hibernation) that may be used as temporary storage for image kernel
129 * memory contents.
130 */
131static struct linked_page *safe_pages_list;
132
133/* Pointer to an auxiliary buffer (1 page) */
134static void *buffer;
135
136#define PG_ANY 0
137#define PG_SAFE 1
138#define PG_UNSAFE_CLEAR 1
139#define PG_UNSAFE_KEEP 0
140
141static unsigned int allocated_unsafe_pages;
142
143/**
144 * get_image_page - Allocate a page for a hibernation image.
145 * @gfp_mask: GFP mask for the allocation.
146 * @safe_needed: Get pages that were not used before hibernation (restore only)
147 *
148 * During image restoration, for storing the PBE list and the image data, we can
149 * only use memory pages that do not conflict with the pages used before
150 * hibernation. The "unsafe" pages have PageNosaveFree set and we count them
151 * using allocated_unsafe_pages.
152 *
153 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
154 * swsusp_free() can release it.
155 */
156static void *get_image_page(gfp_t gfp_mask, int safe_needed)
157{
158 void *res;
159
160 res = (void *)get_zeroed_page(gfp_mask);
161 if (safe_needed)
162 while (res && swsusp_page_is_free(virt_to_page(res))) {
163 /* The page is unsafe, mark it for swsusp_free() */
164 swsusp_set_page_forbidden(virt_to_page(res));
165 allocated_unsafe_pages++;
166 res = (void *)get_zeroed_page(gfp_mask);
167 }
168 if (res) {
169 swsusp_set_page_forbidden(virt_to_page(res));
170 swsusp_set_page_free(virt_to_page(res));
171 }
172 return res;
173}
174
175static void *__get_safe_page(gfp_t gfp_mask)
176{
177 if (safe_pages_list) {
178 void *ret = safe_pages_list;
179
180 safe_pages_list = safe_pages_list->next;
181 memset(ret, 0, PAGE_SIZE);
182 return ret;
183 }
184 return get_image_page(gfp_mask, PG_SAFE);
185}
186
187unsigned long get_safe_page(gfp_t gfp_mask)
188{
189 return (unsigned long)__get_safe_page(gfp_mask);
190}
191
192static struct page *alloc_image_page(gfp_t gfp_mask)
193{
194 struct page *page;
195
196 page = alloc_page(gfp_mask);
197 if (page) {
198 swsusp_set_page_forbidden(page);
199 swsusp_set_page_free(page);
200 }
201 return page;
202}
203
204static void recycle_safe_page(void *page_address)
205{
206 struct linked_page *lp = page_address;
207
208 lp->next = safe_pages_list;
209 safe_pages_list = lp;
210}
211
212/**
213 * free_image_page - Free a page allocated for hibernation image.
214 * @addr: Address of the page to free.
215 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
216 *
217 * The page to free should have been allocated by get_image_page() (page flags
218 * set by it are affected).
219 */
220static inline void free_image_page(void *addr, int clear_nosave_free)
221{
222 struct page *page;
223
224 BUG_ON(!virt_addr_valid(addr));
225
226 page = virt_to_page(addr);
227
228 swsusp_unset_page_forbidden(page);
229 if (clear_nosave_free)
230 swsusp_unset_page_free(page);
231
232 __free_page(page);
233}
234
235static inline void free_list_of_pages(struct linked_page *list,
236 int clear_page_nosave)
237{
238 while (list) {
239 struct linked_page *lp = list->next;
240
241 free_image_page(list, clear_page_nosave);
242 list = lp;
243 }
244}
245
246/*
247 * struct chain_allocator is used for allocating small objects out of
248 * a linked list of pages called 'the chain'.
249 *
250 * The chain grows each time when there is no room for a new object in
251 * the current page. The allocated objects cannot be freed individually.
252 * It is only possible to free them all at once, by freeing the entire
253 * chain.
254 *
255 * NOTE: The chain allocator may be inefficient if the allocated objects
256 * are not much smaller than PAGE_SIZE.
257 */
258struct chain_allocator {
259 struct linked_page *chain; /* the chain */
260 unsigned int used_space; /* total size of objects allocated out
261 of the current page */
262 gfp_t gfp_mask; /* mask for allocating pages */
263 int safe_needed; /* if set, only "safe" pages are allocated */
264};
265
266static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
267 int safe_needed)
268{
269 ca->chain = NULL;
270 ca->used_space = LINKED_PAGE_DATA_SIZE;
271 ca->gfp_mask = gfp_mask;
272 ca->safe_needed = safe_needed;
273}
274
275static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
276{
277 void *ret;
278
279 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
280 struct linked_page *lp;
281
282 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
283 get_image_page(ca->gfp_mask, PG_ANY);
284 if (!lp)
285 return NULL;
286
287 lp->next = ca->chain;
288 ca->chain = lp;
289 ca->used_space = 0;
290 }
291 ret = ca->chain->data + ca->used_space;
292 ca->used_space += size;
293 return ret;
294}
295
296/**
297 * Data types related to memory bitmaps.
298 *
299 * Memory bitmap is a structure consiting of many linked lists of
300 * objects. The main list's elements are of type struct zone_bitmap
301 * and each of them corresonds to one zone. For each zone bitmap
302 * object there is a list of objects of type struct bm_block that
303 * represent each blocks of bitmap in which information is stored.
304 *
305 * struct memory_bitmap contains a pointer to the main list of zone
306 * bitmap objects, a struct bm_position used for browsing the bitmap,
307 * and a pointer to the list of pages used for allocating all of the
308 * zone bitmap objects and bitmap block objects.
309 *
310 * NOTE: It has to be possible to lay out the bitmap in memory
311 * using only allocations of order 0. Additionally, the bitmap is
312 * designed to work with arbitrary number of zones (this is over the
313 * top for now, but let's avoid making unnecessary assumptions ;-).
314 *
315 * struct zone_bitmap contains a pointer to a list of bitmap block
316 * objects and a pointer to the bitmap block object that has been
317 * most recently used for setting bits. Additionally, it contains the
318 * PFNs that correspond to the start and end of the represented zone.
319 *
320 * struct bm_block contains a pointer to the memory page in which
321 * information is stored (in the form of a block of bitmap)
322 * It also contains the pfns that correspond to the start and end of
323 * the represented memory area.
324 *
325 * The memory bitmap is organized as a radix tree to guarantee fast random
326 * access to the bits. There is one radix tree for each zone (as returned
327 * from create_mem_extents).
328 *
329 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
330 * two linked lists for the nodes of the tree, one for the inner nodes and
331 * one for the leave nodes. The linked leave nodes are used for fast linear
332 * access of the memory bitmap.
333 *
334 * The struct rtree_node represents one node of the radix tree.
335 */
336
337#define BM_END_OF_MAP (~0UL)
338
339#define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
340#define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
341#define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
342
343/*
344 * struct rtree_node is a wrapper struct to link the nodes
345 * of the rtree together for easy linear iteration over
346 * bits and easy freeing
347 */
348struct rtree_node {
349 struct list_head list;
350 unsigned long *data;
351};
352
353/*
354 * struct mem_zone_bm_rtree represents a bitmap used for one
355 * populated memory zone.
356 */
357struct mem_zone_bm_rtree {
358 struct list_head list; /* Link Zones together */
359 struct list_head nodes; /* Radix Tree inner nodes */
360 struct list_head leaves; /* Radix Tree leaves */
361 unsigned long start_pfn; /* Zone start page frame */
362 unsigned long end_pfn; /* Zone end page frame + 1 */
363 struct rtree_node *rtree; /* Radix Tree Root */
364 int levels; /* Number of Radix Tree Levels */
365 unsigned int blocks; /* Number of Bitmap Blocks */
366};
367
368/* strcut bm_position is used for browsing memory bitmaps */
369
370struct bm_position {
371 struct mem_zone_bm_rtree *zone;
372 struct rtree_node *node;
373 unsigned long node_pfn;
374 int node_bit;
375};
376
377struct memory_bitmap {
378 struct list_head zones;
379 struct linked_page *p_list; /* list of pages used to store zone
380 bitmap objects and bitmap block
381 objects */
382 struct bm_position cur; /* most recently used bit position */
383};
384
385/* Functions that operate on memory bitmaps */
386
387#define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
388#if BITS_PER_LONG == 32
389#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
390#else
391#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
392#endif
393#define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
394
395/**
396 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
397 *
398 * This function is used to allocate inner nodes as well as the
399 * leave nodes of the radix tree. It also adds the node to the
400 * corresponding linked list passed in by the *list parameter.
401 */
402static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
403 struct chain_allocator *ca,
404 struct list_head *list)
405{
406 struct rtree_node *node;
407
408 node = chain_alloc(ca, sizeof(struct rtree_node));
409 if (!node)
410 return NULL;
411
412 node->data = get_image_page(gfp_mask, safe_needed);
413 if (!node->data)
414 return NULL;
415
416 list_add_tail(&node->list, list);
417
418 return node;
419}
420
421/**
422 * add_rtree_block - Add a new leave node to the radix tree.
423 *
424 * The leave nodes need to be allocated in order to keep the leaves
425 * linked list in order. This is guaranteed by the zone->blocks
426 * counter.
427 */
428static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
429 int safe_needed, struct chain_allocator *ca)
430{
431 struct rtree_node *node, *block, **dst;
432 unsigned int levels_needed, block_nr;
433 int i;
434
435 block_nr = zone->blocks;
436 levels_needed = 0;
437
438 /* How many levels do we need for this block nr? */
439 while (block_nr) {
440 levels_needed += 1;
441 block_nr >>= BM_RTREE_LEVEL_SHIFT;
442 }
443
444 /* Make sure the rtree has enough levels */
445 for (i = zone->levels; i < levels_needed; i++) {
446 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
447 &zone->nodes);
448 if (!node)
449 return -ENOMEM;
450
451 node->data[0] = (unsigned long)zone->rtree;
452 zone->rtree = node;
453 zone->levels += 1;
454 }
455
456 /* Allocate new block */
457 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
458 if (!block)
459 return -ENOMEM;
460
461 /* Now walk the rtree to insert the block */
462 node = zone->rtree;
463 dst = &zone->rtree;
464 block_nr = zone->blocks;
465 for (i = zone->levels; i > 0; i--) {
466 int index;
467
468 if (!node) {
469 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
470 &zone->nodes);
471 if (!node)
472 return -ENOMEM;
473 *dst = node;
474 }
475
476 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
477 index &= BM_RTREE_LEVEL_MASK;
478 dst = (struct rtree_node **)&((*dst)->data[index]);
479 node = *dst;
480 }
481
482 zone->blocks += 1;
483 *dst = block;
484
485 return 0;
486}
487
488static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
489 int clear_nosave_free);
490
491/**
492 * create_zone_bm_rtree - Create a radix tree for one zone.
493 *
494 * Allocated the mem_zone_bm_rtree structure and initializes it.
495 * This function also allocated and builds the radix tree for the
496 * zone.
497 */
498static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
499 int safe_needed,
500 struct chain_allocator *ca,
501 unsigned long start,
502 unsigned long end)
503{
504 struct mem_zone_bm_rtree *zone;
505 unsigned int i, nr_blocks;
506 unsigned long pages;
507
508 pages = end - start;
509 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
510 if (!zone)
511 return NULL;
512
513 INIT_LIST_HEAD(&zone->nodes);
514 INIT_LIST_HEAD(&zone->leaves);
515 zone->start_pfn = start;
516 zone->end_pfn = end;
517 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
518
519 for (i = 0; i < nr_blocks; i++) {
520 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
521 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
522 return NULL;
523 }
524 }
525
526 return zone;
527}
528
529/**
530 * free_zone_bm_rtree - Free the memory of the radix tree.
531 *
532 * Free all node pages of the radix tree. The mem_zone_bm_rtree
533 * structure itself is not freed here nor are the rtree_node
534 * structs.
535 */
536static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
537 int clear_nosave_free)
538{
539 struct rtree_node *node;
540
541 list_for_each_entry(node, &zone->nodes, list)
542 free_image_page(node->data, clear_nosave_free);
543
544 list_for_each_entry(node, &zone->leaves, list)
545 free_image_page(node->data, clear_nosave_free);
546}
547
548static void memory_bm_position_reset(struct memory_bitmap *bm)
549{
550 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
551 list);
552 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
553 struct rtree_node, list);
554 bm->cur.node_pfn = 0;
555 bm->cur.node_bit = 0;
556}
557
558static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
559
560struct mem_extent {
561 struct list_head hook;
562 unsigned long start;
563 unsigned long end;
564};
565
566/**
567 * free_mem_extents - Free a list of memory extents.
568 * @list: List of extents to free.
569 */
570static void free_mem_extents(struct list_head *list)
571{
572 struct mem_extent *ext, *aux;
573
574 list_for_each_entry_safe(ext, aux, list, hook) {
575 list_del(&ext->hook);
576 kfree(ext);
577 }
578}
579
580/**
581 * create_mem_extents - Create a list of memory extents.
582 * @list: List to put the extents into.
583 * @gfp_mask: Mask to use for memory allocations.
584 *
585 * The extents represent contiguous ranges of PFNs.
586 */
587static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
588{
589 struct zone *zone;
590
591 INIT_LIST_HEAD(list);
592
593 for_each_populated_zone(zone) {
594 unsigned long zone_start, zone_end;
595 struct mem_extent *ext, *cur, *aux;
596
597 zone_start = zone->zone_start_pfn;
598 zone_end = zone_end_pfn(zone);
599
600 list_for_each_entry(ext, list, hook)
601 if (zone_start <= ext->end)
602 break;
603
604 if (&ext->hook == list || zone_end < ext->start) {
605 /* New extent is necessary */
606 struct mem_extent *new_ext;
607
608 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
609 if (!new_ext) {
610 free_mem_extents(list);
611 return -ENOMEM;
612 }
613 new_ext->start = zone_start;
614 new_ext->end = zone_end;
615 list_add_tail(&new_ext->hook, &ext->hook);
616 continue;
617 }
618
619 /* Merge this zone's range of PFNs with the existing one */
620 if (zone_start < ext->start)
621 ext->start = zone_start;
622 if (zone_end > ext->end)
623 ext->end = zone_end;
624
625 /* More merging may be possible */
626 cur = ext;
627 list_for_each_entry_safe_continue(cur, aux, list, hook) {
628 if (zone_end < cur->start)
629 break;
630 if (zone_end < cur->end)
631 ext->end = cur->end;
632 list_del(&cur->hook);
633 kfree(cur);
634 }
635 }
636
637 return 0;
638}
639
640/**
641 * memory_bm_create - Allocate memory for a memory bitmap.
642 */
643static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
644 int safe_needed)
645{
646 struct chain_allocator ca;
647 struct list_head mem_extents;
648 struct mem_extent *ext;
649 int error;
650
651 chain_init(&ca, gfp_mask, safe_needed);
652 INIT_LIST_HEAD(&bm->zones);
653
654 error = create_mem_extents(&mem_extents, gfp_mask);
655 if (error)
656 return error;
657
658 list_for_each_entry(ext, &mem_extents, hook) {
659 struct mem_zone_bm_rtree *zone;
660
661 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
662 ext->start, ext->end);
663 if (!zone) {
664 error = -ENOMEM;
665 goto Error;
666 }
667 list_add_tail(&zone->list, &bm->zones);
668 }
669
670 bm->p_list = ca.chain;
671 memory_bm_position_reset(bm);
672 Exit:
673 free_mem_extents(&mem_extents);
674 return error;
675
676 Error:
677 bm->p_list = ca.chain;
678 memory_bm_free(bm, PG_UNSAFE_CLEAR);
679 goto Exit;
680}
681
682/**
683 * memory_bm_free - Free memory occupied by the memory bitmap.
684 * @bm: Memory bitmap.
685 */
686static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
687{
688 struct mem_zone_bm_rtree *zone;
689
690 list_for_each_entry(zone, &bm->zones, list)
691 free_zone_bm_rtree(zone, clear_nosave_free);
692
693 free_list_of_pages(bm->p_list, clear_nosave_free);
694
695 INIT_LIST_HEAD(&bm->zones);
696}
697
698/**
699 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
700 *
701 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
702 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
703 *
704 * Walk the radix tree to find the page containing the bit that represents @pfn
705 * and return the position of the bit in @addr and @bit_nr.
706 */
707static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
708 void **addr, unsigned int *bit_nr)
709{
710 struct mem_zone_bm_rtree *curr, *zone;
711 struct rtree_node *node;
712 int i, block_nr;
713
714 zone = bm->cur.zone;
715
716 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
717 goto zone_found;
718
719 zone = NULL;
720
721 /* Find the right zone */
722 list_for_each_entry(curr, &bm->zones, list) {
723 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
724 zone = curr;
725 break;
726 }
727 }
728
729 if (!zone)
730 return -EFAULT;
731
732zone_found:
733 /*
734 * We have found the zone. Now walk the radix tree to find the leaf node
735 * for our PFN.
736 */
737 node = bm->cur.node;
738 if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
739 goto node_found;
740
741 node = zone->rtree;
742 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
743
744 for (i = zone->levels; i > 0; i--) {
745 int index;
746
747 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
748 index &= BM_RTREE_LEVEL_MASK;
749 BUG_ON(node->data[index] == 0);
750 node = (struct rtree_node *)node->data[index];
751 }
752
753node_found:
754 /* Update last position */
755 bm->cur.zone = zone;
756 bm->cur.node = node;
757 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
758
759 /* Set return values */
760 *addr = node->data;
761 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
762
763 return 0;
764}
765
766static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
767{
768 void *addr;
769 unsigned int bit;
770 int error;
771
772 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
773 BUG_ON(error);
774 set_bit(bit, addr);
775}
776
777static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
778{
779 void *addr;
780 unsigned int bit;
781 int error;
782
783 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
784 if (!error)
785 set_bit(bit, addr);
786
787 return error;
788}
789
790static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
791{
792 void *addr;
793 unsigned int bit;
794 int error;
795
796 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
797 BUG_ON(error);
798 clear_bit(bit, addr);
799}
800
801static void memory_bm_clear_current(struct memory_bitmap *bm)
802{
803 int bit;
804
805 bit = max(bm->cur.node_bit - 1, 0);
806 clear_bit(bit, bm->cur.node->data);
807}
808
809static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
810{
811 void *addr;
812 unsigned int bit;
813 int error;
814
815 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
816 BUG_ON(error);
817 return test_bit(bit, addr);
818}
819
820static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
821{
822 void *addr;
823 unsigned int bit;
824
825 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
826}
827
828/*
829 * rtree_next_node - Jump to the next leaf node.
830 *
831 * Set the position to the beginning of the next node in the
832 * memory bitmap. This is either the next node in the current
833 * zone's radix tree or the first node in the radix tree of the
834 * next zone.
835 *
836 * Return true if there is a next node, false otherwise.
837 */
838static bool rtree_next_node(struct memory_bitmap *bm)
839{
840 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
841 bm->cur.node = list_entry(bm->cur.node->list.next,
842 struct rtree_node, list);
843 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
844 bm->cur.node_bit = 0;
845 touch_softlockup_watchdog();
846 return true;
847 }
848
849 /* No more nodes, goto next zone */
850 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
851 bm->cur.zone = list_entry(bm->cur.zone->list.next,
852 struct mem_zone_bm_rtree, list);
853 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
854 struct rtree_node, list);
855 bm->cur.node_pfn = 0;
856 bm->cur.node_bit = 0;
857 return true;
858 }
859
860 /* No more zones */
861 return false;
862}
863
864/**
865 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
866 * @bm: Memory bitmap.
867 *
868 * Starting from the last returned position this function searches for the next
869 * set bit in @bm and returns the PFN represented by it. If no more bits are
870 * set, BM_END_OF_MAP is returned.
871 *
872 * It is required to run memory_bm_position_reset() before the first call to
873 * this function for the given memory bitmap.
874 */
875static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
876{
877 unsigned long bits, pfn, pages;
878 int bit;
879
880 do {
881 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
882 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
883 bit = find_next_bit(bm->cur.node->data, bits,
884 bm->cur.node_bit);
885 if (bit < bits) {
886 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
887 bm->cur.node_bit = bit + 1;
888 return pfn;
889 }
890 } while (rtree_next_node(bm));
891
892 return BM_END_OF_MAP;
893}
894
895/*
896 * This structure represents a range of page frames the contents of which
897 * should not be saved during hibernation.
898 */
899struct nosave_region {
900 struct list_head list;
901 unsigned long start_pfn;
902 unsigned long end_pfn;
903};
904
905static LIST_HEAD(nosave_regions);
906
907static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
908{
909 struct rtree_node *node;
910
911 list_for_each_entry(node, &zone->nodes, list)
912 recycle_safe_page(node->data);
913
914 list_for_each_entry(node, &zone->leaves, list)
915 recycle_safe_page(node->data);
916}
917
918static void memory_bm_recycle(struct memory_bitmap *bm)
919{
920 struct mem_zone_bm_rtree *zone;
921 struct linked_page *p_list;
922
923 list_for_each_entry(zone, &bm->zones, list)
924 recycle_zone_bm_rtree(zone);
925
926 p_list = bm->p_list;
927 while (p_list) {
928 struct linked_page *lp = p_list;
929
930 p_list = lp->next;
931 recycle_safe_page(lp);
932 }
933}
934
935/**
936 * register_nosave_region - Register a region of unsaveable memory.
937 *
938 * Register a range of page frames the contents of which should not be saved
939 * during hibernation (to be used in the early initialization code).
940 */
941void __init __register_nosave_region(unsigned long start_pfn,
942 unsigned long end_pfn, int use_kmalloc)
943{
944 struct nosave_region *region;
945
946 if (start_pfn >= end_pfn)
947 return;
948
949 if (!list_empty(&nosave_regions)) {
950 /* Try to extend the previous region (they should be sorted) */
951 region = list_entry(nosave_regions.prev,
952 struct nosave_region, list);
953 if (region->end_pfn == start_pfn) {
954 region->end_pfn = end_pfn;
955 goto Report;
956 }
957 }
958 if (use_kmalloc) {
959 /* During init, this shouldn't fail */
960 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
961 BUG_ON(!region);
962 } else {
963 /* This allocation cannot fail */
964 region = memblock_alloc(sizeof(struct nosave_region),
965 SMP_CACHE_BYTES);
966 if (!region)
967 panic("%s: Failed to allocate %zu bytes\n", __func__,
968 sizeof(struct nosave_region));
969 }
970 region->start_pfn = start_pfn;
971 region->end_pfn = end_pfn;
972 list_add_tail(®ion->list, &nosave_regions);
973 Report:
974 pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
975 (unsigned long long) start_pfn << PAGE_SHIFT,
976 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
977}
978
979/*
980 * Set bits in this map correspond to the page frames the contents of which
981 * should not be saved during the suspend.
982 */
983static struct memory_bitmap *forbidden_pages_map;
984
985/* Set bits in this map correspond to free page frames. */
986static struct memory_bitmap *free_pages_map;
987
988/*
989 * Each page frame allocated for creating the image is marked by setting the
990 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
991 */
992
993void swsusp_set_page_free(struct page *page)
994{
995 if (free_pages_map)
996 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
997}
998
999static int swsusp_page_is_free(struct page *page)
1000{
1001 return free_pages_map ?
1002 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1003}
1004
1005void swsusp_unset_page_free(struct page *page)
1006{
1007 if (free_pages_map)
1008 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1009}
1010
1011static void swsusp_set_page_forbidden(struct page *page)
1012{
1013 if (forbidden_pages_map)
1014 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1015}
1016
1017int swsusp_page_is_forbidden(struct page *page)
1018{
1019 return forbidden_pages_map ?
1020 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1021}
1022
1023static void swsusp_unset_page_forbidden(struct page *page)
1024{
1025 if (forbidden_pages_map)
1026 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1027}
1028
1029/**
1030 * mark_nosave_pages - Mark pages that should not be saved.
1031 * @bm: Memory bitmap.
1032 *
1033 * Set the bits in @bm that correspond to the page frames the contents of which
1034 * should not be saved.
1035 */
1036static void mark_nosave_pages(struct memory_bitmap *bm)
1037{
1038 struct nosave_region *region;
1039
1040 if (list_empty(&nosave_regions))
1041 return;
1042
1043 list_for_each_entry(region, &nosave_regions, list) {
1044 unsigned long pfn;
1045
1046 pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1047 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1048 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1049 - 1);
1050
1051 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1052 if (pfn_valid(pfn)) {
1053 /*
1054 * It is safe to ignore the result of
1055 * mem_bm_set_bit_check() here, since we won't
1056 * touch the PFNs for which the error is
1057 * returned anyway.
1058 */
1059 mem_bm_set_bit_check(bm, pfn);
1060 }
1061 }
1062}
1063
1064/**
1065 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1066 *
1067 * Create bitmaps needed for marking page frames that should not be saved and
1068 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1069 * only modified if everything goes well, because we don't want the bits to be
1070 * touched before both bitmaps are set up.
1071 */
1072int create_basic_memory_bitmaps(void)
1073{
1074 struct memory_bitmap *bm1, *bm2;
1075 int error = 0;
1076
1077 if (forbidden_pages_map && free_pages_map)
1078 return 0;
1079 else
1080 BUG_ON(forbidden_pages_map || free_pages_map);
1081
1082 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1083 if (!bm1)
1084 return -ENOMEM;
1085
1086 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1087 if (error)
1088 goto Free_first_object;
1089
1090 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1091 if (!bm2)
1092 goto Free_first_bitmap;
1093
1094 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1095 if (error)
1096 goto Free_second_object;
1097
1098 forbidden_pages_map = bm1;
1099 free_pages_map = bm2;
1100 mark_nosave_pages(forbidden_pages_map);
1101
1102 pr_debug("Basic memory bitmaps created\n");
1103
1104 return 0;
1105
1106 Free_second_object:
1107 kfree(bm2);
1108 Free_first_bitmap:
1109 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1110 Free_first_object:
1111 kfree(bm1);
1112 return -ENOMEM;
1113}
1114
1115/**
1116 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1117 *
1118 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1119 * auxiliary pointers are necessary so that the bitmaps themselves are not
1120 * referred to while they are being freed.
1121 */
1122void free_basic_memory_bitmaps(void)
1123{
1124 struct memory_bitmap *bm1, *bm2;
1125
1126 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1127 return;
1128
1129 bm1 = forbidden_pages_map;
1130 bm2 = free_pages_map;
1131 forbidden_pages_map = NULL;
1132 free_pages_map = NULL;
1133 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1134 kfree(bm1);
1135 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1136 kfree(bm2);
1137
1138 pr_debug("Basic memory bitmaps freed\n");
1139}
1140
1141void clear_free_pages(void)
1142{
1143#ifdef CONFIG_PAGE_POISONING_ZERO
1144 struct memory_bitmap *bm = free_pages_map;
1145 unsigned long pfn;
1146
1147 if (WARN_ON(!(free_pages_map)))
1148 return;
1149
1150 memory_bm_position_reset(bm);
1151 pfn = memory_bm_next_pfn(bm);
1152 while (pfn != BM_END_OF_MAP) {
1153 if (pfn_valid(pfn))
1154 clear_highpage(pfn_to_page(pfn));
1155
1156 pfn = memory_bm_next_pfn(bm);
1157 }
1158 memory_bm_position_reset(bm);
1159 pr_info("free pages cleared after restore\n");
1160#endif /* PAGE_POISONING_ZERO */
1161}
1162
1163/**
1164 * snapshot_additional_pages - Estimate the number of extra pages needed.
1165 * @zone: Memory zone to carry out the computation for.
1166 *
1167 * Estimate the number of additional pages needed for setting up a hibernation
1168 * image data structures for @zone (usually, the returned value is greater than
1169 * the exact number).
1170 */
1171unsigned int snapshot_additional_pages(struct zone *zone)
1172{
1173 unsigned int rtree, nodes;
1174
1175 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1176 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1177 LINKED_PAGE_DATA_SIZE);
1178 while (nodes > 1) {
1179 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1180 rtree += nodes;
1181 }
1182
1183 return 2 * rtree;
1184}
1185
1186#ifdef CONFIG_HIGHMEM
1187/**
1188 * count_free_highmem_pages - Compute the total number of free highmem pages.
1189 *
1190 * The returned number is system-wide.
1191 */
1192static unsigned int count_free_highmem_pages(void)
1193{
1194 struct zone *zone;
1195 unsigned int cnt = 0;
1196
1197 for_each_populated_zone(zone)
1198 if (is_highmem(zone))
1199 cnt += zone_page_state(zone, NR_FREE_PAGES);
1200
1201 return cnt;
1202}
1203
1204/**
1205 * saveable_highmem_page - Check if a highmem page is saveable.
1206 *
1207 * Determine whether a highmem page should be included in a hibernation image.
1208 *
1209 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1210 * and it isn't part of a free chunk of pages.
1211 */
1212static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1213{
1214 struct page *page;
1215
1216 if (!pfn_valid(pfn))
1217 return NULL;
1218
1219 page = pfn_to_online_page(pfn);
1220 if (!page || page_zone(page) != zone)
1221 return NULL;
1222
1223 BUG_ON(!PageHighMem(page));
1224
1225 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1226 return NULL;
1227
1228 if (PageReserved(page) || PageOffline(page))
1229 return NULL;
1230
1231 if (page_is_guard(page))
1232 return NULL;
1233
1234 return page;
1235}
1236
1237/**
1238 * count_highmem_pages - Compute the total number of saveable highmem pages.
1239 */
1240static unsigned int count_highmem_pages(void)
1241{
1242 struct zone *zone;
1243 unsigned int n = 0;
1244
1245 for_each_populated_zone(zone) {
1246 unsigned long pfn, max_zone_pfn;
1247
1248 if (!is_highmem(zone))
1249 continue;
1250
1251 mark_free_pages(zone);
1252 max_zone_pfn = zone_end_pfn(zone);
1253 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1254 if (saveable_highmem_page(zone, pfn))
1255 n++;
1256 }
1257 return n;
1258}
1259#else
1260static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1261{
1262 return NULL;
1263}
1264#endif /* CONFIG_HIGHMEM */
1265
1266/**
1267 * saveable_page - Check if the given page is saveable.
1268 *
1269 * Determine whether a non-highmem page should be included in a hibernation
1270 * image.
1271 *
1272 * We should save the page if it isn't Nosave, and is not in the range
1273 * of pages statically defined as 'unsaveable', and it isn't part of
1274 * a free chunk of pages.
1275 */
1276static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1277{
1278 struct page *page;
1279
1280 if (!pfn_valid(pfn))
1281 return NULL;
1282
1283 page = pfn_to_online_page(pfn);
1284 if (!page || page_zone(page) != zone)
1285 return NULL;
1286
1287 BUG_ON(PageHighMem(page));
1288
1289 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1290 return NULL;
1291
1292 if (PageOffline(page))
1293 return NULL;
1294
1295 if (PageReserved(page)
1296 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1297 return NULL;
1298
1299 if (page_is_guard(page))
1300 return NULL;
1301
1302 return page;
1303}
1304
1305/**
1306 * count_data_pages - Compute the total number of saveable non-highmem pages.
1307 */
1308static unsigned int count_data_pages(void)
1309{
1310 struct zone *zone;
1311 unsigned long pfn, max_zone_pfn;
1312 unsigned int n = 0;
1313
1314 for_each_populated_zone(zone) {
1315 if (is_highmem(zone))
1316 continue;
1317
1318 mark_free_pages(zone);
1319 max_zone_pfn = zone_end_pfn(zone);
1320 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1321 if (saveable_page(zone, pfn))
1322 n++;
1323 }
1324 return n;
1325}
1326
1327/*
1328 * This is needed, because copy_page and memcpy are not usable for copying
1329 * task structs.
1330 */
1331static inline void do_copy_page(long *dst, long *src)
1332{
1333 int n;
1334
1335 for (n = PAGE_SIZE / sizeof(long); n; n--)
1336 *dst++ = *src++;
1337}
1338
1339/**
1340 * safe_copy_page - Copy a page in a safe way.
1341 *
1342 * Check if the page we are going to copy is marked as present in the kernel
1343 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1344 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1345 * always returns 'true'.
1346 */
1347static void safe_copy_page(void *dst, struct page *s_page)
1348{
1349 if (kernel_page_present(s_page)) {
1350 do_copy_page(dst, page_address(s_page));
1351 } else {
1352 kernel_map_pages(s_page, 1, 1);
1353 do_copy_page(dst, page_address(s_page));
1354 kernel_map_pages(s_page, 1, 0);
1355 }
1356}
1357
1358#ifdef CONFIG_HIGHMEM
1359static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1360{
1361 return is_highmem(zone) ?
1362 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1363}
1364
1365static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1366{
1367 struct page *s_page, *d_page;
1368 void *src, *dst;
1369
1370 s_page = pfn_to_page(src_pfn);
1371 d_page = pfn_to_page(dst_pfn);
1372 if (PageHighMem(s_page)) {
1373 src = kmap_atomic(s_page);
1374 dst = kmap_atomic(d_page);
1375 do_copy_page(dst, src);
1376 kunmap_atomic(dst);
1377 kunmap_atomic(src);
1378 } else {
1379 if (PageHighMem(d_page)) {
1380 /*
1381 * The page pointed to by src may contain some kernel
1382 * data modified by kmap_atomic()
1383 */
1384 safe_copy_page(buffer, s_page);
1385 dst = kmap_atomic(d_page);
1386 copy_page(dst, buffer);
1387 kunmap_atomic(dst);
1388 } else {
1389 safe_copy_page(page_address(d_page), s_page);
1390 }
1391 }
1392}
1393#else
1394#define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1395
1396static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1397{
1398 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1399 pfn_to_page(src_pfn));
1400}
1401#endif /* CONFIG_HIGHMEM */
1402
1403static void copy_data_pages(struct memory_bitmap *copy_bm,
1404 struct memory_bitmap *orig_bm)
1405{
1406 struct zone *zone;
1407 unsigned long pfn;
1408
1409 for_each_populated_zone(zone) {
1410 unsigned long max_zone_pfn;
1411
1412 mark_free_pages(zone);
1413 max_zone_pfn = zone_end_pfn(zone);
1414 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1415 if (page_is_saveable(zone, pfn))
1416 memory_bm_set_bit(orig_bm, pfn);
1417 }
1418 memory_bm_position_reset(orig_bm);
1419 memory_bm_position_reset(copy_bm);
1420 for(;;) {
1421 pfn = memory_bm_next_pfn(orig_bm);
1422 if (unlikely(pfn == BM_END_OF_MAP))
1423 break;
1424 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1425 }
1426}
1427
1428/* Total number of image pages */
1429static unsigned int nr_copy_pages;
1430/* Number of pages needed for saving the original pfns of the image pages */
1431static unsigned int nr_meta_pages;
1432/*
1433 * Numbers of normal and highmem page frames allocated for hibernation image
1434 * before suspending devices.
1435 */
1436static unsigned int alloc_normal, alloc_highmem;
1437/*
1438 * Memory bitmap used for marking saveable pages (during hibernation) or
1439 * hibernation image pages (during restore)
1440 */
1441static struct memory_bitmap orig_bm;
1442/*
1443 * Memory bitmap used during hibernation for marking allocated page frames that
1444 * will contain copies of saveable pages. During restore it is initially used
1445 * for marking hibernation image pages, but then the set bits from it are
1446 * duplicated in @orig_bm and it is released. On highmem systems it is next
1447 * used for marking "safe" highmem pages, but it has to be reinitialized for
1448 * this purpose.
1449 */
1450static struct memory_bitmap copy_bm;
1451
1452/**
1453 * swsusp_free - Free pages allocated for hibernation image.
1454 *
1455 * Image pages are alocated before snapshot creation, so they need to be
1456 * released after resume.
1457 */
1458void swsusp_free(void)
1459{
1460 unsigned long fb_pfn, fr_pfn;
1461
1462 if (!forbidden_pages_map || !free_pages_map)
1463 goto out;
1464
1465 memory_bm_position_reset(forbidden_pages_map);
1466 memory_bm_position_reset(free_pages_map);
1467
1468loop:
1469 fr_pfn = memory_bm_next_pfn(free_pages_map);
1470 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1471
1472 /*
1473 * Find the next bit set in both bitmaps. This is guaranteed to
1474 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1475 */
1476 do {
1477 if (fb_pfn < fr_pfn)
1478 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1479 if (fr_pfn < fb_pfn)
1480 fr_pfn = memory_bm_next_pfn(free_pages_map);
1481 } while (fb_pfn != fr_pfn);
1482
1483 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1484 struct page *page = pfn_to_page(fr_pfn);
1485
1486 memory_bm_clear_current(forbidden_pages_map);
1487 memory_bm_clear_current(free_pages_map);
1488 hibernate_restore_unprotect_page(page_address(page));
1489 __free_page(page);
1490 goto loop;
1491 }
1492
1493out:
1494 nr_copy_pages = 0;
1495 nr_meta_pages = 0;
1496 restore_pblist = NULL;
1497 buffer = NULL;
1498 alloc_normal = 0;
1499 alloc_highmem = 0;
1500 hibernate_restore_protection_end();
1501}
1502
1503/* Helper functions used for the shrinking of memory. */
1504
1505#define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1506
1507/**
1508 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1509 * @nr_pages: Number of page frames to allocate.
1510 * @mask: GFP flags to use for the allocation.
1511 *
1512 * Return value: Number of page frames actually allocated
1513 */
1514static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1515{
1516 unsigned long nr_alloc = 0;
1517
1518 while (nr_pages > 0) {
1519 struct page *page;
1520
1521 page = alloc_image_page(mask);
1522 if (!page)
1523 break;
1524 memory_bm_set_bit(©_bm, page_to_pfn(page));
1525 if (PageHighMem(page))
1526 alloc_highmem++;
1527 else
1528 alloc_normal++;
1529 nr_pages--;
1530 nr_alloc++;
1531 }
1532
1533 return nr_alloc;
1534}
1535
1536static unsigned long preallocate_image_memory(unsigned long nr_pages,
1537 unsigned long avail_normal)
1538{
1539 unsigned long alloc;
1540
1541 if (avail_normal <= alloc_normal)
1542 return 0;
1543
1544 alloc = avail_normal - alloc_normal;
1545 if (nr_pages < alloc)
1546 alloc = nr_pages;
1547
1548 return preallocate_image_pages(alloc, GFP_IMAGE);
1549}
1550
1551#ifdef CONFIG_HIGHMEM
1552static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1553{
1554 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1555}
1556
1557/**
1558 * __fraction - Compute (an approximation of) x * (multiplier / base).
1559 */
1560static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1561{
1562 x *= multiplier;
1563 do_div(x, base);
1564 return (unsigned long)x;
1565}
1566
1567static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1568 unsigned long highmem,
1569 unsigned long total)
1570{
1571 unsigned long alloc = __fraction(nr_pages, highmem, total);
1572
1573 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1574}
1575#else /* CONFIG_HIGHMEM */
1576static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1577{
1578 return 0;
1579}
1580
1581static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1582 unsigned long highmem,
1583 unsigned long total)
1584{
1585 return 0;
1586}
1587#endif /* CONFIG_HIGHMEM */
1588
1589/**
1590 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1591 */
1592static unsigned long free_unnecessary_pages(void)
1593{
1594 unsigned long save, to_free_normal, to_free_highmem, free;
1595
1596 save = count_data_pages();
1597 if (alloc_normal >= save) {
1598 to_free_normal = alloc_normal - save;
1599 save = 0;
1600 } else {
1601 to_free_normal = 0;
1602 save -= alloc_normal;
1603 }
1604 save += count_highmem_pages();
1605 if (alloc_highmem >= save) {
1606 to_free_highmem = alloc_highmem - save;
1607 } else {
1608 to_free_highmem = 0;
1609 save -= alloc_highmem;
1610 if (to_free_normal > save)
1611 to_free_normal -= save;
1612 else
1613 to_free_normal = 0;
1614 }
1615 free = to_free_normal + to_free_highmem;
1616
1617 memory_bm_position_reset(©_bm);
1618
1619 while (to_free_normal > 0 || to_free_highmem > 0) {
1620 unsigned long pfn = memory_bm_next_pfn(©_bm);
1621 struct page *page = pfn_to_page(pfn);
1622
1623 if (PageHighMem(page)) {
1624 if (!to_free_highmem)
1625 continue;
1626 to_free_highmem--;
1627 alloc_highmem--;
1628 } else {
1629 if (!to_free_normal)
1630 continue;
1631 to_free_normal--;
1632 alloc_normal--;
1633 }
1634 memory_bm_clear_bit(©_bm, pfn);
1635 swsusp_unset_page_forbidden(page);
1636 swsusp_unset_page_free(page);
1637 __free_page(page);
1638 }
1639
1640 return free;
1641}
1642
1643/**
1644 * minimum_image_size - Estimate the minimum acceptable size of an image.
1645 * @saveable: Number of saveable pages in the system.
1646 *
1647 * We want to avoid attempting to free too much memory too hard, so estimate the
1648 * minimum acceptable size of a hibernation image to use as the lower limit for
1649 * preallocating memory.
1650 *
1651 * We assume that the minimum image size should be proportional to
1652 *
1653 * [number of saveable pages] - [number of pages that can be freed in theory]
1654 *
1655 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1656 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1657 */
1658static unsigned long minimum_image_size(unsigned long saveable)
1659{
1660 unsigned long size;
1661
1662 size = global_node_page_state(NR_SLAB_RECLAIMABLE)
1663 + global_node_page_state(NR_ACTIVE_ANON)
1664 + global_node_page_state(NR_INACTIVE_ANON)
1665 + global_node_page_state(NR_ACTIVE_FILE)
1666 + global_node_page_state(NR_INACTIVE_FILE);
1667
1668 return saveable <= size ? 0 : saveable - size;
1669}
1670
1671/**
1672 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1673 *
1674 * To create a hibernation image it is necessary to make a copy of every page
1675 * frame in use. We also need a number of page frames to be free during
1676 * hibernation for allocations made while saving the image and for device
1677 * drivers, in case they need to allocate memory from their hibernation
1678 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1679 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1680 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1681 * total number of available page frames and allocate at least
1682 *
1683 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1684 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1685 *
1686 * of them, which corresponds to the maximum size of a hibernation image.
1687 *
1688 * If image_size is set below the number following from the above formula,
1689 * the preallocation of memory is continued until the total number of saveable
1690 * pages in the system is below the requested image size or the minimum
1691 * acceptable image size returned by minimum_image_size(), whichever is greater.
1692 */
1693int hibernate_preallocate_memory(void)
1694{
1695 struct zone *zone;
1696 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1697 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1698 ktime_t start, stop;
1699 int error;
1700
1701 pr_info("Preallocating image memory... ");
1702 start = ktime_get();
1703
1704 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1705 if (error)
1706 goto err_out;
1707
1708 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1709 if (error)
1710 goto err_out;
1711
1712 alloc_normal = 0;
1713 alloc_highmem = 0;
1714
1715 /* Count the number of saveable data pages. */
1716 save_highmem = count_highmem_pages();
1717 saveable = count_data_pages();
1718
1719 /*
1720 * Compute the total number of page frames we can use (count) and the
1721 * number of pages needed for image metadata (size).
1722 */
1723 count = saveable;
1724 saveable += save_highmem;
1725 highmem = save_highmem;
1726 size = 0;
1727 for_each_populated_zone(zone) {
1728 size += snapshot_additional_pages(zone);
1729 if (is_highmem(zone))
1730 highmem += zone_page_state(zone, NR_FREE_PAGES);
1731 else
1732 count += zone_page_state(zone, NR_FREE_PAGES);
1733 }
1734 avail_normal = count;
1735 count += highmem;
1736 count -= totalreserve_pages;
1737
1738 /* Add number of pages required for page keys (s390 only). */
1739 size += page_key_additional_pages(saveable);
1740
1741 /* Compute the maximum number of saveable pages to leave in memory. */
1742 max_size = (count - (size + PAGES_FOR_IO)) / 2
1743 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1744 /* Compute the desired number of image pages specified by image_size. */
1745 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1746 if (size > max_size)
1747 size = max_size;
1748 /*
1749 * If the desired number of image pages is at least as large as the
1750 * current number of saveable pages in memory, allocate page frames for
1751 * the image and we're done.
1752 */
1753 if (size >= saveable) {
1754 pages = preallocate_image_highmem(save_highmem);
1755 pages += preallocate_image_memory(saveable - pages, avail_normal);
1756 goto out;
1757 }
1758
1759 /* Estimate the minimum size of the image. */
1760 pages = minimum_image_size(saveable);
1761 /*
1762 * To avoid excessive pressure on the normal zone, leave room in it to
1763 * accommodate an image of the minimum size (unless it's already too
1764 * small, in which case don't preallocate pages from it at all).
1765 */
1766 if (avail_normal > pages)
1767 avail_normal -= pages;
1768 else
1769 avail_normal = 0;
1770 if (size < pages)
1771 size = min_t(unsigned long, pages, max_size);
1772
1773 /*
1774 * Let the memory management subsystem know that we're going to need a
1775 * large number of page frames to allocate and make it free some memory.
1776 * NOTE: If this is not done, performance will be hurt badly in some
1777 * test cases.
1778 */
1779 shrink_all_memory(saveable - size);
1780
1781 /*
1782 * The number of saveable pages in memory was too high, so apply some
1783 * pressure to decrease it. First, make room for the largest possible
1784 * image and fail if that doesn't work. Next, try to decrease the size
1785 * of the image as much as indicated by 'size' using allocations from
1786 * highmem and non-highmem zones separately.
1787 */
1788 pages_highmem = preallocate_image_highmem(highmem / 2);
1789 alloc = count - max_size;
1790 if (alloc > pages_highmem)
1791 alloc -= pages_highmem;
1792 else
1793 alloc = 0;
1794 pages = preallocate_image_memory(alloc, avail_normal);
1795 if (pages < alloc) {
1796 /* We have exhausted non-highmem pages, try highmem. */
1797 alloc -= pages;
1798 pages += pages_highmem;
1799 pages_highmem = preallocate_image_highmem(alloc);
1800 if (pages_highmem < alloc)
1801 goto err_out;
1802 pages += pages_highmem;
1803 /*
1804 * size is the desired number of saveable pages to leave in
1805 * memory, so try to preallocate (all memory - size) pages.
1806 */
1807 alloc = (count - pages) - size;
1808 pages += preallocate_image_highmem(alloc);
1809 } else {
1810 /*
1811 * There are approximately max_size saveable pages at this point
1812 * and we want to reduce this number down to size.
1813 */
1814 alloc = max_size - size;
1815 size = preallocate_highmem_fraction(alloc, highmem, count);
1816 pages_highmem += size;
1817 alloc -= size;
1818 size = preallocate_image_memory(alloc, avail_normal);
1819 pages_highmem += preallocate_image_highmem(alloc - size);
1820 pages += pages_highmem + size;
1821 }
1822
1823 /*
1824 * We only need as many page frames for the image as there are saveable
1825 * pages in memory, but we have allocated more. Release the excessive
1826 * ones now.
1827 */
1828 pages -= free_unnecessary_pages();
1829
1830 out:
1831 stop = ktime_get();
1832 pr_cont("done (allocated %lu pages)\n", pages);
1833 swsusp_show_speed(start, stop, pages, "Allocated");
1834
1835 return 0;
1836
1837 err_out:
1838 pr_cont("\n");
1839 swsusp_free();
1840 return -ENOMEM;
1841}
1842
1843#ifdef CONFIG_HIGHMEM
1844/**
1845 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1846 *
1847 * Compute the number of non-highmem pages that will be necessary for creating
1848 * copies of highmem pages.
1849 */
1850static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1851{
1852 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1853
1854 if (free_highmem >= nr_highmem)
1855 nr_highmem = 0;
1856 else
1857 nr_highmem -= free_highmem;
1858
1859 return nr_highmem;
1860}
1861#else
1862static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1863#endif /* CONFIG_HIGHMEM */
1864
1865/**
1866 * enough_free_mem - Check if there is enough free memory for the image.
1867 */
1868static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1869{
1870 struct zone *zone;
1871 unsigned int free = alloc_normal;
1872
1873 for_each_populated_zone(zone)
1874 if (!is_highmem(zone))
1875 free += zone_page_state(zone, NR_FREE_PAGES);
1876
1877 nr_pages += count_pages_for_highmem(nr_highmem);
1878 pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
1879 nr_pages, PAGES_FOR_IO, free);
1880
1881 return free > nr_pages + PAGES_FOR_IO;
1882}
1883
1884#ifdef CONFIG_HIGHMEM
1885/**
1886 * get_highmem_buffer - Allocate a buffer for highmem pages.
1887 *
1888 * If there are some highmem pages in the hibernation image, we may need a
1889 * buffer to copy them and/or load their data.
1890 */
1891static inline int get_highmem_buffer(int safe_needed)
1892{
1893 buffer = get_image_page(GFP_ATOMIC, safe_needed);
1894 return buffer ? 0 : -ENOMEM;
1895}
1896
1897/**
1898 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1899 *
1900 * Try to allocate as many pages as needed, but if the number of free highmem
1901 * pages is less than that, allocate them all.
1902 */
1903static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1904 unsigned int nr_highmem)
1905{
1906 unsigned int to_alloc = count_free_highmem_pages();
1907
1908 if (to_alloc > nr_highmem)
1909 to_alloc = nr_highmem;
1910
1911 nr_highmem -= to_alloc;
1912 while (to_alloc-- > 0) {
1913 struct page *page;
1914
1915 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1916 memory_bm_set_bit(bm, page_to_pfn(page));
1917 }
1918 return nr_highmem;
1919}
1920#else
1921static inline int get_highmem_buffer(int safe_needed) { return 0; }
1922
1923static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1924 unsigned int n) { return 0; }
1925#endif /* CONFIG_HIGHMEM */
1926
1927/**
1928 * swsusp_alloc - Allocate memory for hibernation image.
1929 *
1930 * We first try to allocate as many highmem pages as there are
1931 * saveable highmem pages in the system. If that fails, we allocate
1932 * non-highmem pages for the copies of the remaining highmem ones.
1933 *
1934 * In this approach it is likely that the copies of highmem pages will
1935 * also be located in the high memory, because of the way in which
1936 * copy_data_pages() works.
1937 */
1938static int swsusp_alloc(struct memory_bitmap *copy_bm,
1939 unsigned int nr_pages, unsigned int nr_highmem)
1940{
1941 if (nr_highmem > 0) {
1942 if (get_highmem_buffer(PG_ANY))
1943 goto err_out;
1944 if (nr_highmem > alloc_highmem) {
1945 nr_highmem -= alloc_highmem;
1946 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1947 }
1948 }
1949 if (nr_pages > alloc_normal) {
1950 nr_pages -= alloc_normal;
1951 while (nr_pages-- > 0) {
1952 struct page *page;
1953
1954 page = alloc_image_page(GFP_ATOMIC);
1955 if (!page)
1956 goto err_out;
1957 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1958 }
1959 }
1960
1961 return 0;
1962
1963 err_out:
1964 swsusp_free();
1965 return -ENOMEM;
1966}
1967
1968asmlinkage __visible int swsusp_save(void)
1969{
1970 unsigned int nr_pages, nr_highmem;
1971
1972 pr_info("Creating hibernation image:\n");
1973
1974 drain_local_pages(NULL);
1975 nr_pages = count_data_pages();
1976 nr_highmem = count_highmem_pages();
1977 pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
1978
1979 if (!enough_free_mem(nr_pages, nr_highmem)) {
1980 pr_err("Not enough free memory\n");
1981 return -ENOMEM;
1982 }
1983
1984 if (swsusp_alloc(©_bm, nr_pages, nr_highmem)) {
1985 pr_err("Memory allocation failed\n");
1986 return -ENOMEM;
1987 }
1988
1989 /*
1990 * During allocating of suspend pagedir, new cold pages may appear.
1991 * Kill them.
1992 */
1993 drain_local_pages(NULL);
1994 copy_data_pages(©_bm, &orig_bm);
1995
1996 /*
1997 * End of critical section. From now on, we can write to memory,
1998 * but we should not touch disk. This specially means we must _not_
1999 * touch swap space! Except we must write out our image of course.
2000 */
2001
2002 nr_pages += nr_highmem;
2003 nr_copy_pages = nr_pages;
2004 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2005
2006 pr_info("Hibernation image created (%d pages copied)\n", nr_pages);
2007
2008 return 0;
2009}
2010
2011#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2012static int init_header_complete(struct swsusp_info *info)
2013{
2014 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2015 info->version_code = LINUX_VERSION_CODE;
2016 return 0;
2017}
2018
2019static char *check_image_kernel(struct swsusp_info *info)
2020{
2021 if (info->version_code != LINUX_VERSION_CODE)
2022 return "kernel version";
2023 if (strcmp(info->uts.sysname,init_utsname()->sysname))
2024 return "system type";
2025 if (strcmp(info->uts.release,init_utsname()->release))
2026 return "kernel release";
2027 if (strcmp(info->uts.version,init_utsname()->version))
2028 return "version";
2029 if (strcmp(info->uts.machine,init_utsname()->machine))
2030 return "machine";
2031 return NULL;
2032}
2033#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2034
2035unsigned long snapshot_get_image_size(void)
2036{
2037 return nr_copy_pages + nr_meta_pages + 1;
2038}
2039
2040static int init_header(struct swsusp_info *info)
2041{
2042 memset(info, 0, sizeof(struct swsusp_info));
2043 info->num_physpages = get_num_physpages();
2044 info->image_pages = nr_copy_pages;
2045 info->pages = snapshot_get_image_size();
2046 info->size = info->pages;
2047 info->size <<= PAGE_SHIFT;
2048 return init_header_complete(info);
2049}
2050
2051/**
2052 * pack_pfns - Prepare PFNs for saving.
2053 * @bm: Memory bitmap.
2054 * @buf: Memory buffer to store the PFNs in.
2055 *
2056 * PFNs corresponding to set bits in @bm are stored in the area of memory
2057 * pointed to by @buf (1 page at a time).
2058 */
2059static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2060{
2061 int j;
2062
2063 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2064 buf[j] = memory_bm_next_pfn(bm);
2065 if (unlikely(buf[j] == BM_END_OF_MAP))
2066 break;
2067 /* Save page key for data page (s390 only). */
2068 page_key_read(buf + j);
2069 }
2070}
2071
2072/**
2073 * snapshot_read_next - Get the address to read the next image page from.
2074 * @handle: Snapshot handle to be used for the reading.
2075 *
2076 * On the first call, @handle should point to a zeroed snapshot_handle
2077 * structure. The structure gets populated then and a pointer to it should be
2078 * passed to this function every next time.
2079 *
2080 * On success, the function returns a positive number. Then, the caller
2081 * is allowed to read up to the returned number of bytes from the memory
2082 * location computed by the data_of() macro.
2083 *
2084 * The function returns 0 to indicate the end of the data stream condition,
2085 * and negative numbers are returned on errors. If that happens, the structure
2086 * pointed to by @handle is not updated and should not be used any more.
2087 */
2088int snapshot_read_next(struct snapshot_handle *handle)
2089{
2090 if (handle->cur > nr_meta_pages + nr_copy_pages)
2091 return 0;
2092
2093 if (!buffer) {
2094 /* This makes the buffer be freed by swsusp_free() */
2095 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2096 if (!buffer)
2097 return -ENOMEM;
2098 }
2099 if (!handle->cur) {
2100 int error;
2101
2102 error = init_header((struct swsusp_info *)buffer);
2103 if (error)
2104 return error;
2105 handle->buffer = buffer;
2106 memory_bm_position_reset(&orig_bm);
2107 memory_bm_position_reset(©_bm);
2108 } else if (handle->cur <= nr_meta_pages) {
2109 clear_page(buffer);
2110 pack_pfns(buffer, &orig_bm);
2111 } else {
2112 struct page *page;
2113
2114 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2115 if (PageHighMem(page)) {
2116 /*
2117 * Highmem pages are copied to the buffer,
2118 * because we can't return with a kmapped
2119 * highmem page (we may not be called again).
2120 */
2121 void *kaddr;
2122
2123 kaddr = kmap_atomic(page);
2124 copy_page(buffer, kaddr);
2125 kunmap_atomic(kaddr);
2126 handle->buffer = buffer;
2127 } else {
2128 handle->buffer = page_address(page);
2129 }
2130 }
2131 handle->cur++;
2132 return PAGE_SIZE;
2133}
2134
2135static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2136 struct memory_bitmap *src)
2137{
2138 unsigned long pfn;
2139
2140 memory_bm_position_reset(src);
2141 pfn = memory_bm_next_pfn(src);
2142 while (pfn != BM_END_OF_MAP) {
2143 memory_bm_set_bit(dst, pfn);
2144 pfn = memory_bm_next_pfn(src);
2145 }
2146}
2147
2148/**
2149 * mark_unsafe_pages - Mark pages that were used before hibernation.
2150 *
2151 * Mark the pages that cannot be used for storing the image during restoration,
2152 * because they conflict with the pages that had been used before hibernation.
2153 */
2154static void mark_unsafe_pages(struct memory_bitmap *bm)
2155{
2156 unsigned long pfn;
2157
2158 /* Clear the "free"/"unsafe" bit for all PFNs */
2159 memory_bm_position_reset(free_pages_map);
2160 pfn = memory_bm_next_pfn(free_pages_map);
2161 while (pfn != BM_END_OF_MAP) {
2162 memory_bm_clear_current(free_pages_map);
2163 pfn = memory_bm_next_pfn(free_pages_map);
2164 }
2165
2166 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2167 duplicate_memory_bitmap(free_pages_map, bm);
2168
2169 allocated_unsafe_pages = 0;
2170}
2171
2172static int check_header(struct swsusp_info *info)
2173{
2174 char *reason;
2175
2176 reason = check_image_kernel(info);
2177 if (!reason && info->num_physpages != get_num_physpages())
2178 reason = "memory size";
2179 if (reason) {
2180 pr_err("Image mismatch: %s\n", reason);
2181 return -EPERM;
2182 }
2183 return 0;
2184}
2185
2186/**
2187 * load header - Check the image header and copy the data from it.
2188 */
2189static int load_header(struct swsusp_info *info)
2190{
2191 int error;
2192
2193 restore_pblist = NULL;
2194 error = check_header(info);
2195 if (!error) {
2196 nr_copy_pages = info->image_pages;
2197 nr_meta_pages = info->pages - info->image_pages - 1;
2198 }
2199 return error;
2200}
2201
2202/**
2203 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2204 * @bm: Memory bitmap.
2205 * @buf: Area of memory containing the PFNs.
2206 *
2207 * For each element of the array pointed to by @buf (1 page at a time), set the
2208 * corresponding bit in @bm.
2209 */
2210static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2211{
2212 int j;
2213
2214 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2215 if (unlikely(buf[j] == BM_END_OF_MAP))
2216 break;
2217
2218 /* Extract and buffer page key for data page (s390 only). */
2219 page_key_memorize(buf + j);
2220
2221 if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2222 memory_bm_set_bit(bm, buf[j]);
2223 else
2224 return -EFAULT;
2225 }
2226
2227 return 0;
2228}
2229
2230#ifdef CONFIG_HIGHMEM
2231/*
2232 * struct highmem_pbe is used for creating the list of highmem pages that
2233 * should be restored atomically during the resume from disk, because the page
2234 * frames they have occupied before the suspend are in use.
2235 */
2236struct highmem_pbe {
2237 struct page *copy_page; /* data is here now */
2238 struct page *orig_page; /* data was here before the suspend */
2239 struct highmem_pbe *next;
2240};
2241
2242/*
2243 * List of highmem PBEs needed for restoring the highmem pages that were
2244 * allocated before the suspend and included in the suspend image, but have
2245 * also been allocated by the "resume" kernel, so their contents cannot be
2246 * written directly to their "original" page frames.
2247 */
2248static struct highmem_pbe *highmem_pblist;
2249
2250/**
2251 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2252 * @bm: Memory bitmap.
2253 *
2254 * The bits in @bm that correspond to image pages are assumed to be set.
2255 */
2256static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2257{
2258 unsigned long pfn;
2259 unsigned int cnt = 0;
2260
2261 memory_bm_position_reset(bm);
2262 pfn = memory_bm_next_pfn(bm);
2263 while (pfn != BM_END_OF_MAP) {
2264 if (PageHighMem(pfn_to_page(pfn)))
2265 cnt++;
2266
2267 pfn = memory_bm_next_pfn(bm);
2268 }
2269 return cnt;
2270}
2271
2272static unsigned int safe_highmem_pages;
2273
2274static struct memory_bitmap *safe_highmem_bm;
2275
2276/**
2277 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2278 * @bm: Pointer to an uninitialized memory bitmap structure.
2279 * @nr_highmem_p: Pointer to the number of highmem image pages.
2280 *
2281 * Try to allocate as many highmem pages as there are highmem image pages
2282 * (@nr_highmem_p points to the variable containing the number of highmem image
2283 * pages). The pages that are "safe" (ie. will not be overwritten when the
2284 * hibernation image is restored entirely) have the corresponding bits set in
2285 * @bm (it must be unitialized).
2286 *
2287 * NOTE: This function should not be called if there are no highmem image pages.
2288 */
2289static int prepare_highmem_image(struct memory_bitmap *bm,
2290 unsigned int *nr_highmem_p)
2291{
2292 unsigned int to_alloc;
2293
2294 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2295 return -ENOMEM;
2296
2297 if (get_highmem_buffer(PG_SAFE))
2298 return -ENOMEM;
2299
2300 to_alloc = count_free_highmem_pages();
2301 if (to_alloc > *nr_highmem_p)
2302 to_alloc = *nr_highmem_p;
2303 else
2304 *nr_highmem_p = to_alloc;
2305
2306 safe_highmem_pages = 0;
2307 while (to_alloc-- > 0) {
2308 struct page *page;
2309
2310 page = alloc_page(__GFP_HIGHMEM);
2311 if (!swsusp_page_is_free(page)) {
2312 /* The page is "safe", set its bit the bitmap */
2313 memory_bm_set_bit(bm, page_to_pfn(page));
2314 safe_highmem_pages++;
2315 }
2316 /* Mark the page as allocated */
2317 swsusp_set_page_forbidden(page);
2318 swsusp_set_page_free(page);
2319 }
2320 memory_bm_position_reset(bm);
2321 safe_highmem_bm = bm;
2322 return 0;
2323}
2324
2325static struct page *last_highmem_page;
2326
2327/**
2328 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2329 *
2330 * For a given highmem image page get a buffer that suspend_write_next() should
2331 * return to its caller to write to.
2332 *
2333 * If the page is to be saved to its "original" page frame or a copy of
2334 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2335 * the copy of the page is to be made in normal memory, so the address of
2336 * the copy is returned.
2337 *
2338 * If @buffer is returned, the caller of suspend_write_next() will write
2339 * the page's contents to @buffer, so they will have to be copied to the
2340 * right location on the next call to suspend_write_next() and it is done
2341 * with the help of copy_last_highmem_page(). For this purpose, if
2342 * @buffer is returned, @last_highmem_page is set to the page to which
2343 * the data will have to be copied from @buffer.
2344 */
2345static void *get_highmem_page_buffer(struct page *page,
2346 struct chain_allocator *ca)
2347{
2348 struct highmem_pbe *pbe;
2349 void *kaddr;
2350
2351 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2352 /*
2353 * We have allocated the "original" page frame and we can
2354 * use it directly to store the loaded page.
2355 */
2356 last_highmem_page = page;
2357 return buffer;
2358 }
2359 /*
2360 * The "original" page frame has not been allocated and we have to
2361 * use a "safe" page frame to store the loaded page.
2362 */
2363 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2364 if (!pbe) {
2365 swsusp_free();
2366 return ERR_PTR(-ENOMEM);
2367 }
2368 pbe->orig_page = page;
2369 if (safe_highmem_pages > 0) {
2370 struct page *tmp;
2371
2372 /* Copy of the page will be stored in high memory */
2373 kaddr = buffer;
2374 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2375 safe_highmem_pages--;
2376 last_highmem_page = tmp;
2377 pbe->copy_page = tmp;
2378 } else {
2379 /* Copy of the page will be stored in normal memory */
2380 kaddr = safe_pages_list;
2381 safe_pages_list = safe_pages_list->next;
2382 pbe->copy_page = virt_to_page(kaddr);
2383 }
2384 pbe->next = highmem_pblist;
2385 highmem_pblist = pbe;
2386 return kaddr;
2387}
2388
2389/**
2390 * copy_last_highmem_page - Copy most the most recent highmem image page.
2391 *
2392 * Copy the contents of a highmem image from @buffer, where the caller of
2393 * snapshot_write_next() has stored them, to the right location represented by
2394 * @last_highmem_page .
2395 */
2396static void copy_last_highmem_page(void)
2397{
2398 if (last_highmem_page) {
2399 void *dst;
2400
2401 dst = kmap_atomic(last_highmem_page);
2402 copy_page(dst, buffer);
2403 kunmap_atomic(dst);
2404 last_highmem_page = NULL;
2405 }
2406}
2407
2408static inline int last_highmem_page_copied(void)
2409{
2410 return !last_highmem_page;
2411}
2412
2413static inline void free_highmem_data(void)
2414{
2415 if (safe_highmem_bm)
2416 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2417
2418 if (buffer)
2419 free_image_page(buffer, PG_UNSAFE_CLEAR);
2420}
2421#else
2422static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2423
2424static inline int prepare_highmem_image(struct memory_bitmap *bm,
2425 unsigned int *nr_highmem_p) { return 0; }
2426
2427static inline void *get_highmem_page_buffer(struct page *page,
2428 struct chain_allocator *ca)
2429{
2430 return ERR_PTR(-EINVAL);
2431}
2432
2433static inline void copy_last_highmem_page(void) {}
2434static inline int last_highmem_page_copied(void) { return 1; }
2435static inline void free_highmem_data(void) {}
2436#endif /* CONFIG_HIGHMEM */
2437
2438#define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2439
2440/**
2441 * prepare_image - Make room for loading hibernation image.
2442 * @new_bm: Unitialized memory bitmap structure.
2443 * @bm: Memory bitmap with unsafe pages marked.
2444 *
2445 * Use @bm to mark the pages that will be overwritten in the process of
2446 * restoring the system memory state from the suspend image ("unsafe" pages)
2447 * and allocate memory for the image.
2448 *
2449 * The idea is to allocate a new memory bitmap first and then allocate
2450 * as many pages as needed for image data, but without specifying what those
2451 * pages will be used for just yet. Instead, we mark them all as allocated and
2452 * create a lists of "safe" pages to be used later. On systems with high
2453 * memory a list of "safe" highmem pages is created too.
2454 */
2455static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2456{
2457 unsigned int nr_pages, nr_highmem;
2458 struct linked_page *lp;
2459 int error;
2460
2461 /* If there is no highmem, the buffer will not be necessary */
2462 free_image_page(buffer, PG_UNSAFE_CLEAR);
2463 buffer = NULL;
2464
2465 nr_highmem = count_highmem_image_pages(bm);
2466 mark_unsafe_pages(bm);
2467
2468 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2469 if (error)
2470 goto Free;
2471
2472 duplicate_memory_bitmap(new_bm, bm);
2473 memory_bm_free(bm, PG_UNSAFE_KEEP);
2474 if (nr_highmem > 0) {
2475 error = prepare_highmem_image(bm, &nr_highmem);
2476 if (error)
2477 goto Free;
2478 }
2479 /*
2480 * Reserve some safe pages for potential later use.
2481 *
2482 * NOTE: This way we make sure there will be enough safe pages for the
2483 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2484 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2485 *
2486 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2487 */
2488 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2489 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2490 while (nr_pages > 0) {
2491 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2492 if (!lp) {
2493 error = -ENOMEM;
2494 goto Free;
2495 }
2496 lp->next = safe_pages_list;
2497 safe_pages_list = lp;
2498 nr_pages--;
2499 }
2500 /* Preallocate memory for the image */
2501 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2502 while (nr_pages > 0) {
2503 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2504 if (!lp) {
2505 error = -ENOMEM;
2506 goto Free;
2507 }
2508 if (!swsusp_page_is_free(virt_to_page(lp))) {
2509 /* The page is "safe", add it to the list */
2510 lp->next = safe_pages_list;
2511 safe_pages_list = lp;
2512 }
2513 /* Mark the page as allocated */
2514 swsusp_set_page_forbidden(virt_to_page(lp));
2515 swsusp_set_page_free(virt_to_page(lp));
2516 nr_pages--;
2517 }
2518 return 0;
2519
2520 Free:
2521 swsusp_free();
2522 return error;
2523}
2524
2525/**
2526 * get_buffer - Get the address to store the next image data page.
2527 *
2528 * Get the address that snapshot_write_next() should return to its caller to
2529 * write to.
2530 */
2531static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2532{
2533 struct pbe *pbe;
2534 struct page *page;
2535 unsigned long pfn = memory_bm_next_pfn(bm);
2536
2537 if (pfn == BM_END_OF_MAP)
2538 return ERR_PTR(-EFAULT);
2539
2540 page = pfn_to_page(pfn);
2541 if (PageHighMem(page))
2542 return get_highmem_page_buffer(page, ca);
2543
2544 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2545 /*
2546 * We have allocated the "original" page frame and we can
2547 * use it directly to store the loaded page.
2548 */
2549 return page_address(page);
2550
2551 /*
2552 * The "original" page frame has not been allocated and we have to
2553 * use a "safe" page frame to store the loaded page.
2554 */
2555 pbe = chain_alloc(ca, sizeof(struct pbe));
2556 if (!pbe) {
2557 swsusp_free();
2558 return ERR_PTR(-ENOMEM);
2559 }
2560 pbe->orig_address = page_address(page);
2561 pbe->address = safe_pages_list;
2562 safe_pages_list = safe_pages_list->next;
2563 pbe->next = restore_pblist;
2564 restore_pblist = pbe;
2565 return pbe->address;
2566}
2567
2568/**
2569 * snapshot_write_next - Get the address to store the next image page.
2570 * @handle: Snapshot handle structure to guide the writing.
2571 *
2572 * On the first call, @handle should point to a zeroed snapshot_handle
2573 * structure. The structure gets populated then and a pointer to it should be
2574 * passed to this function every next time.
2575 *
2576 * On success, the function returns a positive number. Then, the caller
2577 * is allowed to write up to the returned number of bytes to the memory
2578 * location computed by the data_of() macro.
2579 *
2580 * The function returns 0 to indicate the "end of file" condition. Negative
2581 * numbers are returned on errors, in which cases the structure pointed to by
2582 * @handle is not updated and should not be used any more.
2583 */
2584int snapshot_write_next(struct snapshot_handle *handle)
2585{
2586 static struct chain_allocator ca;
2587 int error = 0;
2588
2589 /* Check if we have already loaded the entire image */
2590 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2591 return 0;
2592
2593 handle->sync_read = 1;
2594
2595 if (!handle->cur) {
2596 if (!buffer)
2597 /* This makes the buffer be freed by swsusp_free() */
2598 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2599
2600 if (!buffer)
2601 return -ENOMEM;
2602
2603 handle->buffer = buffer;
2604 } else if (handle->cur == 1) {
2605 error = load_header(buffer);
2606 if (error)
2607 return error;
2608
2609 safe_pages_list = NULL;
2610
2611 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2612 if (error)
2613 return error;
2614
2615 /* Allocate buffer for page keys. */
2616 error = page_key_alloc(nr_copy_pages);
2617 if (error)
2618 return error;
2619
2620 hibernate_restore_protection_begin();
2621 } else if (handle->cur <= nr_meta_pages + 1) {
2622 error = unpack_orig_pfns(buffer, ©_bm);
2623 if (error)
2624 return error;
2625
2626 if (handle->cur == nr_meta_pages + 1) {
2627 error = prepare_image(&orig_bm, ©_bm);
2628 if (error)
2629 return error;
2630
2631 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2632 memory_bm_position_reset(&orig_bm);
2633 restore_pblist = NULL;
2634 handle->buffer = get_buffer(&orig_bm, &ca);
2635 handle->sync_read = 0;
2636 if (IS_ERR(handle->buffer))
2637 return PTR_ERR(handle->buffer);
2638 }
2639 } else {
2640 copy_last_highmem_page();
2641 /* Restore page key for data page (s390 only). */
2642 page_key_write(handle->buffer);
2643 hibernate_restore_protect_page(handle->buffer);
2644 handle->buffer = get_buffer(&orig_bm, &ca);
2645 if (IS_ERR(handle->buffer))
2646 return PTR_ERR(handle->buffer);
2647 if (handle->buffer != buffer)
2648 handle->sync_read = 0;
2649 }
2650 handle->cur++;
2651 return PAGE_SIZE;
2652}
2653
2654/**
2655 * snapshot_write_finalize - Complete the loading of a hibernation image.
2656 *
2657 * Must be called after the last call to snapshot_write_next() in case the last
2658 * page in the image happens to be a highmem page and its contents should be
2659 * stored in highmem. Additionally, it recycles bitmap memory that's not
2660 * necessary any more.
2661 */
2662void snapshot_write_finalize(struct snapshot_handle *handle)
2663{
2664 copy_last_highmem_page();
2665 /* Restore page key for data page (s390 only). */
2666 page_key_write(handle->buffer);
2667 page_key_free();
2668 hibernate_restore_protect_page(handle->buffer);
2669 /* Do that only if we have loaded the image entirely */
2670 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2671 memory_bm_recycle(&orig_bm);
2672 free_highmem_data();
2673 }
2674}
2675
2676int snapshot_image_loaded(struct snapshot_handle *handle)
2677{
2678 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2679 handle->cur <= nr_meta_pages + nr_copy_pages);
2680}
2681
2682#ifdef CONFIG_HIGHMEM
2683/* Assumes that @buf is ready and points to a "safe" page */
2684static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2685 void *buf)
2686{
2687 void *kaddr1, *kaddr2;
2688
2689 kaddr1 = kmap_atomic(p1);
2690 kaddr2 = kmap_atomic(p2);
2691 copy_page(buf, kaddr1);
2692 copy_page(kaddr1, kaddr2);
2693 copy_page(kaddr2, buf);
2694 kunmap_atomic(kaddr2);
2695 kunmap_atomic(kaddr1);
2696}
2697
2698/**
2699 * restore_highmem - Put highmem image pages into their original locations.
2700 *
2701 * For each highmem page that was in use before hibernation and is included in
2702 * the image, and also has been allocated by the "restore" kernel, swap its
2703 * current contents with the previous (ie. "before hibernation") ones.
2704 *
2705 * If the restore eventually fails, we can call this function once again and
2706 * restore the highmem state as seen by the restore kernel.
2707 */
2708int restore_highmem(void)
2709{
2710 struct highmem_pbe *pbe = highmem_pblist;
2711 void *buf;
2712
2713 if (!pbe)
2714 return 0;
2715
2716 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2717 if (!buf)
2718 return -ENOMEM;
2719
2720 while (pbe) {
2721 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2722 pbe = pbe->next;
2723 }
2724 free_image_page(buf, PG_UNSAFE_CLEAR);
2725 return 0;
2726}
2727#endif /* CONFIG_HIGHMEM */
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/power/snapshot.c
4 *
5 * This file provides system snapshot/restore functionality for swsusp.
6 *
7 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
8 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
9 */
10
11#define pr_fmt(fmt) "PM: hibernation: " fmt
12
13#include <linux/version.h>
14#include <linux/module.h>
15#include <linux/mm.h>
16#include <linux/suspend.h>
17#include <linux/delay.h>
18#include <linux/bitops.h>
19#include <linux/spinlock.h>
20#include <linux/kernel.h>
21#include <linux/pm.h>
22#include <linux/device.h>
23#include <linux/init.h>
24#include <linux/memblock.h>
25#include <linux/nmi.h>
26#include <linux/syscalls.h>
27#include <linux/console.h>
28#include <linux/highmem.h>
29#include <linux/list.h>
30#include <linux/slab.h>
31#include <linux/compiler.h>
32#include <linux/ktime.h>
33#include <linux/set_memory.h>
34
35#include <linux/uaccess.h>
36#include <asm/mmu_context.h>
37#include <asm/tlbflush.h>
38#include <asm/io.h>
39
40#include "power.h"
41
42#if defined(CONFIG_STRICT_KERNEL_RWX) && defined(CONFIG_ARCH_HAS_SET_MEMORY)
43static bool hibernate_restore_protection;
44static bool hibernate_restore_protection_active;
45
46void enable_restore_image_protection(void)
47{
48 hibernate_restore_protection = true;
49}
50
51static inline void hibernate_restore_protection_begin(void)
52{
53 hibernate_restore_protection_active = hibernate_restore_protection;
54}
55
56static inline void hibernate_restore_protection_end(void)
57{
58 hibernate_restore_protection_active = false;
59}
60
61static inline void hibernate_restore_protect_page(void *page_address)
62{
63 if (hibernate_restore_protection_active)
64 set_memory_ro((unsigned long)page_address, 1);
65}
66
67static inline void hibernate_restore_unprotect_page(void *page_address)
68{
69 if (hibernate_restore_protection_active)
70 set_memory_rw((unsigned long)page_address, 1);
71}
72#else
73static inline void hibernate_restore_protection_begin(void) {}
74static inline void hibernate_restore_protection_end(void) {}
75static inline void hibernate_restore_protect_page(void *page_address) {}
76static inline void hibernate_restore_unprotect_page(void *page_address) {}
77#endif /* CONFIG_STRICT_KERNEL_RWX && CONFIG_ARCH_HAS_SET_MEMORY */
78
79
80/*
81 * The calls to set_direct_map_*() should not fail because remapping a page
82 * here means that we only update protection bits in an existing PTE.
83 * It is still worth to have a warning here if something changes and this
84 * will no longer be the case.
85 */
86static inline void hibernate_map_page(struct page *page)
87{
88 if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
89 int ret = set_direct_map_default_noflush(page);
90
91 if (ret)
92 pr_warn_once("Failed to remap page\n");
93 } else {
94 debug_pagealloc_map_pages(page, 1);
95 }
96}
97
98static inline void hibernate_unmap_page(struct page *page)
99{
100 if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
101 unsigned long addr = (unsigned long)page_address(page);
102 int ret = set_direct_map_invalid_noflush(page);
103
104 if (ret)
105 pr_warn_once("Failed to remap page\n");
106
107 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
108 } else {
109 debug_pagealloc_unmap_pages(page, 1);
110 }
111}
112
113static int swsusp_page_is_free(struct page *);
114static void swsusp_set_page_forbidden(struct page *);
115static void swsusp_unset_page_forbidden(struct page *);
116
117/*
118 * Number of bytes to reserve for memory allocations made by device drivers
119 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
120 * cause image creation to fail (tunable via /sys/power/reserved_size).
121 */
122unsigned long reserved_size;
123
124void __init hibernate_reserved_size_init(void)
125{
126 reserved_size = SPARE_PAGES * PAGE_SIZE;
127}
128
129/*
130 * Preferred image size in bytes (tunable via /sys/power/image_size).
131 * When it is set to N, swsusp will do its best to ensure the image
132 * size will not exceed N bytes, but if that is impossible, it will
133 * try to create the smallest image possible.
134 */
135unsigned long image_size;
136
137void __init hibernate_image_size_init(void)
138{
139 image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
140}
141
142/*
143 * List of PBEs needed for restoring the pages that were allocated before
144 * the suspend and included in the suspend image, but have also been
145 * allocated by the "resume" kernel, so their contents cannot be written
146 * directly to their "original" page frames.
147 */
148struct pbe *restore_pblist;
149
150/* struct linked_page is used to build chains of pages */
151
152#define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
153
154struct linked_page {
155 struct linked_page *next;
156 char data[LINKED_PAGE_DATA_SIZE];
157} __packed;
158
159/*
160 * List of "safe" pages (ie. pages that were not used by the image kernel
161 * before hibernation) that may be used as temporary storage for image kernel
162 * memory contents.
163 */
164static struct linked_page *safe_pages_list;
165
166/* Pointer to an auxiliary buffer (1 page) */
167static void *buffer;
168
169#define PG_ANY 0
170#define PG_SAFE 1
171#define PG_UNSAFE_CLEAR 1
172#define PG_UNSAFE_KEEP 0
173
174static unsigned int allocated_unsafe_pages;
175
176/**
177 * get_image_page - Allocate a page for a hibernation image.
178 * @gfp_mask: GFP mask for the allocation.
179 * @safe_needed: Get pages that were not used before hibernation (restore only)
180 *
181 * During image restoration, for storing the PBE list and the image data, we can
182 * only use memory pages that do not conflict with the pages used before
183 * hibernation. The "unsafe" pages have PageNosaveFree set and we count them
184 * using allocated_unsafe_pages.
185 *
186 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
187 * swsusp_free() can release it.
188 */
189static void *get_image_page(gfp_t gfp_mask, int safe_needed)
190{
191 void *res;
192
193 res = (void *)get_zeroed_page(gfp_mask);
194 if (safe_needed)
195 while (res && swsusp_page_is_free(virt_to_page(res))) {
196 /* The page is unsafe, mark it for swsusp_free() */
197 swsusp_set_page_forbidden(virt_to_page(res));
198 allocated_unsafe_pages++;
199 res = (void *)get_zeroed_page(gfp_mask);
200 }
201 if (res) {
202 swsusp_set_page_forbidden(virt_to_page(res));
203 swsusp_set_page_free(virt_to_page(res));
204 }
205 return res;
206}
207
208static void *__get_safe_page(gfp_t gfp_mask)
209{
210 if (safe_pages_list) {
211 void *ret = safe_pages_list;
212
213 safe_pages_list = safe_pages_list->next;
214 memset(ret, 0, PAGE_SIZE);
215 return ret;
216 }
217 return get_image_page(gfp_mask, PG_SAFE);
218}
219
220unsigned long get_safe_page(gfp_t gfp_mask)
221{
222 return (unsigned long)__get_safe_page(gfp_mask);
223}
224
225static struct page *alloc_image_page(gfp_t gfp_mask)
226{
227 struct page *page;
228
229 page = alloc_page(gfp_mask);
230 if (page) {
231 swsusp_set_page_forbidden(page);
232 swsusp_set_page_free(page);
233 }
234 return page;
235}
236
237static void recycle_safe_page(void *page_address)
238{
239 struct linked_page *lp = page_address;
240
241 lp->next = safe_pages_list;
242 safe_pages_list = lp;
243}
244
245/**
246 * free_image_page - Free a page allocated for hibernation image.
247 * @addr: Address of the page to free.
248 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
249 *
250 * The page to free should have been allocated by get_image_page() (page flags
251 * set by it are affected).
252 */
253static inline void free_image_page(void *addr, int clear_nosave_free)
254{
255 struct page *page;
256
257 BUG_ON(!virt_addr_valid(addr));
258
259 page = virt_to_page(addr);
260
261 swsusp_unset_page_forbidden(page);
262 if (clear_nosave_free)
263 swsusp_unset_page_free(page);
264
265 __free_page(page);
266}
267
268static inline void free_list_of_pages(struct linked_page *list,
269 int clear_page_nosave)
270{
271 while (list) {
272 struct linked_page *lp = list->next;
273
274 free_image_page(list, clear_page_nosave);
275 list = lp;
276 }
277}
278
279/*
280 * struct chain_allocator is used for allocating small objects out of
281 * a linked list of pages called 'the chain'.
282 *
283 * The chain grows each time when there is no room for a new object in
284 * the current page. The allocated objects cannot be freed individually.
285 * It is only possible to free them all at once, by freeing the entire
286 * chain.
287 *
288 * NOTE: The chain allocator may be inefficient if the allocated objects
289 * are not much smaller than PAGE_SIZE.
290 */
291struct chain_allocator {
292 struct linked_page *chain; /* the chain */
293 unsigned int used_space; /* total size of objects allocated out
294 of the current page */
295 gfp_t gfp_mask; /* mask for allocating pages */
296 int safe_needed; /* if set, only "safe" pages are allocated */
297};
298
299static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
300 int safe_needed)
301{
302 ca->chain = NULL;
303 ca->used_space = LINKED_PAGE_DATA_SIZE;
304 ca->gfp_mask = gfp_mask;
305 ca->safe_needed = safe_needed;
306}
307
308static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
309{
310 void *ret;
311
312 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
313 struct linked_page *lp;
314
315 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
316 get_image_page(ca->gfp_mask, PG_ANY);
317 if (!lp)
318 return NULL;
319
320 lp->next = ca->chain;
321 ca->chain = lp;
322 ca->used_space = 0;
323 }
324 ret = ca->chain->data + ca->used_space;
325 ca->used_space += size;
326 return ret;
327}
328
329/*
330 * Data types related to memory bitmaps.
331 *
332 * Memory bitmap is a structure consisting of many linked lists of
333 * objects. The main list's elements are of type struct zone_bitmap
334 * and each of them corresponds to one zone. For each zone bitmap
335 * object there is a list of objects of type struct bm_block that
336 * represent each blocks of bitmap in which information is stored.
337 *
338 * struct memory_bitmap contains a pointer to the main list of zone
339 * bitmap objects, a struct bm_position used for browsing the bitmap,
340 * and a pointer to the list of pages used for allocating all of the
341 * zone bitmap objects and bitmap block objects.
342 *
343 * NOTE: It has to be possible to lay out the bitmap in memory
344 * using only allocations of order 0. Additionally, the bitmap is
345 * designed to work with arbitrary number of zones (this is over the
346 * top for now, but let's avoid making unnecessary assumptions ;-).
347 *
348 * struct zone_bitmap contains a pointer to a list of bitmap block
349 * objects and a pointer to the bitmap block object that has been
350 * most recently used for setting bits. Additionally, it contains the
351 * PFNs that correspond to the start and end of the represented zone.
352 *
353 * struct bm_block contains a pointer to the memory page in which
354 * information is stored (in the form of a block of bitmap)
355 * It also contains the pfns that correspond to the start and end of
356 * the represented memory area.
357 *
358 * The memory bitmap is organized as a radix tree to guarantee fast random
359 * access to the bits. There is one radix tree for each zone (as returned
360 * from create_mem_extents).
361 *
362 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
363 * two linked lists for the nodes of the tree, one for the inner nodes and
364 * one for the leave nodes. The linked leave nodes are used for fast linear
365 * access of the memory bitmap.
366 *
367 * The struct rtree_node represents one node of the radix tree.
368 */
369
370#define BM_END_OF_MAP (~0UL)
371
372#define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
373#define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
374#define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
375
376/*
377 * struct rtree_node is a wrapper struct to link the nodes
378 * of the rtree together for easy linear iteration over
379 * bits and easy freeing
380 */
381struct rtree_node {
382 struct list_head list;
383 unsigned long *data;
384};
385
386/*
387 * struct mem_zone_bm_rtree represents a bitmap used for one
388 * populated memory zone.
389 */
390struct mem_zone_bm_rtree {
391 struct list_head list; /* Link Zones together */
392 struct list_head nodes; /* Radix Tree inner nodes */
393 struct list_head leaves; /* Radix Tree leaves */
394 unsigned long start_pfn; /* Zone start page frame */
395 unsigned long end_pfn; /* Zone end page frame + 1 */
396 struct rtree_node *rtree; /* Radix Tree Root */
397 int levels; /* Number of Radix Tree Levels */
398 unsigned int blocks; /* Number of Bitmap Blocks */
399};
400
401/* struct bm_position is used for browsing memory bitmaps */
402
403struct bm_position {
404 struct mem_zone_bm_rtree *zone;
405 struct rtree_node *node;
406 unsigned long node_pfn;
407 unsigned long cur_pfn;
408 int node_bit;
409};
410
411struct memory_bitmap {
412 struct list_head zones;
413 struct linked_page *p_list; /* list of pages used to store zone
414 bitmap objects and bitmap block
415 objects */
416 struct bm_position cur; /* most recently used bit position */
417};
418
419/* Functions that operate on memory bitmaps */
420
421#define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
422#if BITS_PER_LONG == 32
423#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
424#else
425#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
426#endif
427#define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
428
429/**
430 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
431 * @gfp_mask: GFP mask for the allocation.
432 * @safe_needed: Get pages not used before hibernation (restore only)
433 * @ca: Pointer to a linked list of pages ("a chain") to allocate from
434 * @list: Radix Tree node to add.
435 *
436 * This function is used to allocate inner nodes as well as the
437 * leave nodes of the radix tree. It also adds the node to the
438 * corresponding linked list passed in by the *list parameter.
439 */
440static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
441 struct chain_allocator *ca,
442 struct list_head *list)
443{
444 struct rtree_node *node;
445
446 node = chain_alloc(ca, sizeof(struct rtree_node));
447 if (!node)
448 return NULL;
449
450 node->data = get_image_page(gfp_mask, safe_needed);
451 if (!node->data)
452 return NULL;
453
454 list_add_tail(&node->list, list);
455
456 return node;
457}
458
459/**
460 * add_rtree_block - Add a new leave node to the radix tree.
461 *
462 * The leave nodes need to be allocated in order to keep the leaves
463 * linked list in order. This is guaranteed by the zone->blocks
464 * counter.
465 */
466static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
467 int safe_needed, struct chain_allocator *ca)
468{
469 struct rtree_node *node, *block, **dst;
470 unsigned int levels_needed, block_nr;
471 int i;
472
473 block_nr = zone->blocks;
474 levels_needed = 0;
475
476 /* How many levels do we need for this block nr? */
477 while (block_nr) {
478 levels_needed += 1;
479 block_nr >>= BM_RTREE_LEVEL_SHIFT;
480 }
481
482 /* Make sure the rtree has enough levels */
483 for (i = zone->levels; i < levels_needed; i++) {
484 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
485 &zone->nodes);
486 if (!node)
487 return -ENOMEM;
488
489 node->data[0] = (unsigned long)zone->rtree;
490 zone->rtree = node;
491 zone->levels += 1;
492 }
493
494 /* Allocate new block */
495 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
496 if (!block)
497 return -ENOMEM;
498
499 /* Now walk the rtree to insert the block */
500 node = zone->rtree;
501 dst = &zone->rtree;
502 block_nr = zone->blocks;
503 for (i = zone->levels; i > 0; i--) {
504 int index;
505
506 if (!node) {
507 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
508 &zone->nodes);
509 if (!node)
510 return -ENOMEM;
511 *dst = node;
512 }
513
514 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
515 index &= BM_RTREE_LEVEL_MASK;
516 dst = (struct rtree_node **)&((*dst)->data[index]);
517 node = *dst;
518 }
519
520 zone->blocks += 1;
521 *dst = block;
522
523 return 0;
524}
525
526static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
527 int clear_nosave_free);
528
529/**
530 * create_zone_bm_rtree - Create a radix tree for one zone.
531 *
532 * Allocated the mem_zone_bm_rtree structure and initializes it.
533 * This function also allocated and builds the radix tree for the
534 * zone.
535 */
536static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
537 int safe_needed,
538 struct chain_allocator *ca,
539 unsigned long start,
540 unsigned long end)
541{
542 struct mem_zone_bm_rtree *zone;
543 unsigned int i, nr_blocks;
544 unsigned long pages;
545
546 pages = end - start;
547 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
548 if (!zone)
549 return NULL;
550
551 INIT_LIST_HEAD(&zone->nodes);
552 INIT_LIST_HEAD(&zone->leaves);
553 zone->start_pfn = start;
554 zone->end_pfn = end;
555 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
556
557 for (i = 0; i < nr_blocks; i++) {
558 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
559 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
560 return NULL;
561 }
562 }
563
564 return zone;
565}
566
567/**
568 * free_zone_bm_rtree - Free the memory of the radix tree.
569 *
570 * Free all node pages of the radix tree. The mem_zone_bm_rtree
571 * structure itself is not freed here nor are the rtree_node
572 * structs.
573 */
574static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
575 int clear_nosave_free)
576{
577 struct rtree_node *node;
578
579 list_for_each_entry(node, &zone->nodes, list)
580 free_image_page(node->data, clear_nosave_free);
581
582 list_for_each_entry(node, &zone->leaves, list)
583 free_image_page(node->data, clear_nosave_free);
584}
585
586static void memory_bm_position_reset(struct memory_bitmap *bm)
587{
588 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
589 list);
590 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
591 struct rtree_node, list);
592 bm->cur.node_pfn = 0;
593 bm->cur.cur_pfn = BM_END_OF_MAP;
594 bm->cur.node_bit = 0;
595}
596
597static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
598
599struct mem_extent {
600 struct list_head hook;
601 unsigned long start;
602 unsigned long end;
603};
604
605/**
606 * free_mem_extents - Free a list of memory extents.
607 * @list: List of extents to free.
608 */
609static void free_mem_extents(struct list_head *list)
610{
611 struct mem_extent *ext, *aux;
612
613 list_for_each_entry_safe(ext, aux, list, hook) {
614 list_del(&ext->hook);
615 kfree(ext);
616 }
617}
618
619/**
620 * create_mem_extents - Create a list of memory extents.
621 * @list: List to put the extents into.
622 * @gfp_mask: Mask to use for memory allocations.
623 *
624 * The extents represent contiguous ranges of PFNs.
625 */
626static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
627{
628 struct zone *zone;
629
630 INIT_LIST_HEAD(list);
631
632 for_each_populated_zone(zone) {
633 unsigned long zone_start, zone_end;
634 struct mem_extent *ext, *cur, *aux;
635
636 zone_start = zone->zone_start_pfn;
637 zone_end = zone_end_pfn(zone);
638
639 list_for_each_entry(ext, list, hook)
640 if (zone_start <= ext->end)
641 break;
642
643 if (&ext->hook == list || zone_end < ext->start) {
644 /* New extent is necessary */
645 struct mem_extent *new_ext;
646
647 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
648 if (!new_ext) {
649 free_mem_extents(list);
650 return -ENOMEM;
651 }
652 new_ext->start = zone_start;
653 new_ext->end = zone_end;
654 list_add_tail(&new_ext->hook, &ext->hook);
655 continue;
656 }
657
658 /* Merge this zone's range of PFNs with the existing one */
659 if (zone_start < ext->start)
660 ext->start = zone_start;
661 if (zone_end > ext->end)
662 ext->end = zone_end;
663
664 /* More merging may be possible */
665 cur = ext;
666 list_for_each_entry_safe_continue(cur, aux, list, hook) {
667 if (zone_end < cur->start)
668 break;
669 if (zone_end < cur->end)
670 ext->end = cur->end;
671 list_del(&cur->hook);
672 kfree(cur);
673 }
674 }
675
676 return 0;
677}
678
679/**
680 * memory_bm_create - Allocate memory for a memory bitmap.
681 */
682static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
683 int safe_needed)
684{
685 struct chain_allocator ca;
686 struct list_head mem_extents;
687 struct mem_extent *ext;
688 int error;
689
690 chain_init(&ca, gfp_mask, safe_needed);
691 INIT_LIST_HEAD(&bm->zones);
692
693 error = create_mem_extents(&mem_extents, gfp_mask);
694 if (error)
695 return error;
696
697 list_for_each_entry(ext, &mem_extents, hook) {
698 struct mem_zone_bm_rtree *zone;
699
700 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
701 ext->start, ext->end);
702 if (!zone) {
703 error = -ENOMEM;
704 goto Error;
705 }
706 list_add_tail(&zone->list, &bm->zones);
707 }
708
709 bm->p_list = ca.chain;
710 memory_bm_position_reset(bm);
711 Exit:
712 free_mem_extents(&mem_extents);
713 return error;
714
715 Error:
716 bm->p_list = ca.chain;
717 memory_bm_free(bm, PG_UNSAFE_CLEAR);
718 goto Exit;
719}
720
721/**
722 * memory_bm_free - Free memory occupied by the memory bitmap.
723 * @bm: Memory bitmap.
724 */
725static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
726{
727 struct mem_zone_bm_rtree *zone;
728
729 list_for_each_entry(zone, &bm->zones, list)
730 free_zone_bm_rtree(zone, clear_nosave_free);
731
732 free_list_of_pages(bm->p_list, clear_nosave_free);
733
734 INIT_LIST_HEAD(&bm->zones);
735}
736
737/**
738 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
739 *
740 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
741 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
742 *
743 * Walk the radix tree to find the page containing the bit that represents @pfn
744 * and return the position of the bit in @addr and @bit_nr.
745 */
746static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
747 void **addr, unsigned int *bit_nr)
748{
749 struct mem_zone_bm_rtree *curr, *zone;
750 struct rtree_node *node;
751 int i, block_nr;
752
753 zone = bm->cur.zone;
754
755 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
756 goto zone_found;
757
758 zone = NULL;
759
760 /* Find the right zone */
761 list_for_each_entry(curr, &bm->zones, list) {
762 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
763 zone = curr;
764 break;
765 }
766 }
767
768 if (!zone)
769 return -EFAULT;
770
771zone_found:
772 /*
773 * We have found the zone. Now walk the radix tree to find the leaf node
774 * for our PFN.
775 */
776
777 /*
778 * If the zone we wish to scan is the current zone and the
779 * pfn falls into the current node then we do not need to walk
780 * the tree.
781 */
782 node = bm->cur.node;
783 if (zone == bm->cur.zone &&
784 ((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
785 goto node_found;
786
787 node = zone->rtree;
788 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
789
790 for (i = zone->levels; i > 0; i--) {
791 int index;
792
793 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
794 index &= BM_RTREE_LEVEL_MASK;
795 BUG_ON(node->data[index] == 0);
796 node = (struct rtree_node *)node->data[index];
797 }
798
799node_found:
800 /* Update last position */
801 bm->cur.zone = zone;
802 bm->cur.node = node;
803 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
804 bm->cur.cur_pfn = pfn;
805
806 /* Set return values */
807 *addr = node->data;
808 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
809
810 return 0;
811}
812
813static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
814{
815 void *addr;
816 unsigned int bit;
817 int error;
818
819 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
820 BUG_ON(error);
821 set_bit(bit, addr);
822}
823
824static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
825{
826 void *addr;
827 unsigned int bit;
828 int error;
829
830 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
831 if (!error)
832 set_bit(bit, addr);
833
834 return error;
835}
836
837static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
838{
839 void *addr;
840 unsigned int bit;
841 int error;
842
843 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
844 BUG_ON(error);
845 clear_bit(bit, addr);
846}
847
848static void memory_bm_clear_current(struct memory_bitmap *bm)
849{
850 int bit;
851
852 bit = max(bm->cur.node_bit - 1, 0);
853 clear_bit(bit, bm->cur.node->data);
854}
855
856static unsigned long memory_bm_get_current(struct memory_bitmap *bm)
857{
858 return bm->cur.cur_pfn;
859}
860
861static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
862{
863 void *addr;
864 unsigned int bit;
865 int error;
866
867 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
868 BUG_ON(error);
869 return test_bit(bit, addr);
870}
871
872static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
873{
874 void *addr;
875 unsigned int bit;
876
877 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
878}
879
880/*
881 * rtree_next_node - Jump to the next leaf node.
882 *
883 * Set the position to the beginning of the next node in the
884 * memory bitmap. This is either the next node in the current
885 * zone's radix tree or the first node in the radix tree of the
886 * next zone.
887 *
888 * Return true if there is a next node, false otherwise.
889 */
890static bool rtree_next_node(struct memory_bitmap *bm)
891{
892 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
893 bm->cur.node = list_entry(bm->cur.node->list.next,
894 struct rtree_node, list);
895 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
896 bm->cur.node_bit = 0;
897 touch_softlockup_watchdog();
898 return true;
899 }
900
901 /* No more nodes, goto next zone */
902 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
903 bm->cur.zone = list_entry(bm->cur.zone->list.next,
904 struct mem_zone_bm_rtree, list);
905 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
906 struct rtree_node, list);
907 bm->cur.node_pfn = 0;
908 bm->cur.node_bit = 0;
909 return true;
910 }
911
912 /* No more zones */
913 return false;
914}
915
916/**
917 * memory_bm_next_pfn - Find the next set bit in a memory bitmap.
918 * @bm: Memory bitmap.
919 *
920 * Starting from the last returned position this function searches for the next
921 * set bit in @bm and returns the PFN represented by it. If no more bits are
922 * set, BM_END_OF_MAP is returned.
923 *
924 * It is required to run memory_bm_position_reset() before the first call to
925 * this function for the given memory bitmap.
926 */
927static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
928{
929 unsigned long bits, pfn, pages;
930 int bit;
931
932 do {
933 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
934 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
935 bit = find_next_bit(bm->cur.node->data, bits,
936 bm->cur.node_bit);
937 if (bit < bits) {
938 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
939 bm->cur.node_bit = bit + 1;
940 bm->cur.cur_pfn = pfn;
941 return pfn;
942 }
943 } while (rtree_next_node(bm));
944
945 bm->cur.cur_pfn = BM_END_OF_MAP;
946 return BM_END_OF_MAP;
947}
948
949/*
950 * This structure represents a range of page frames the contents of which
951 * should not be saved during hibernation.
952 */
953struct nosave_region {
954 struct list_head list;
955 unsigned long start_pfn;
956 unsigned long end_pfn;
957};
958
959static LIST_HEAD(nosave_regions);
960
961static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
962{
963 struct rtree_node *node;
964
965 list_for_each_entry(node, &zone->nodes, list)
966 recycle_safe_page(node->data);
967
968 list_for_each_entry(node, &zone->leaves, list)
969 recycle_safe_page(node->data);
970}
971
972static void memory_bm_recycle(struct memory_bitmap *bm)
973{
974 struct mem_zone_bm_rtree *zone;
975 struct linked_page *p_list;
976
977 list_for_each_entry(zone, &bm->zones, list)
978 recycle_zone_bm_rtree(zone);
979
980 p_list = bm->p_list;
981 while (p_list) {
982 struct linked_page *lp = p_list;
983
984 p_list = lp->next;
985 recycle_safe_page(lp);
986 }
987}
988
989/**
990 * register_nosave_region - Register a region of unsaveable memory.
991 *
992 * Register a range of page frames the contents of which should not be saved
993 * during hibernation (to be used in the early initialization code).
994 */
995void __init register_nosave_region(unsigned long start_pfn, unsigned long end_pfn)
996{
997 struct nosave_region *region;
998
999 if (start_pfn >= end_pfn)
1000 return;
1001
1002 if (!list_empty(&nosave_regions)) {
1003 /* Try to extend the previous region (they should be sorted) */
1004 region = list_entry(nosave_regions.prev,
1005 struct nosave_region, list);
1006 if (region->end_pfn == start_pfn) {
1007 region->end_pfn = end_pfn;
1008 goto Report;
1009 }
1010 }
1011 /* This allocation cannot fail */
1012 region = memblock_alloc(sizeof(struct nosave_region),
1013 SMP_CACHE_BYTES);
1014 if (!region)
1015 panic("%s: Failed to allocate %zu bytes\n", __func__,
1016 sizeof(struct nosave_region));
1017 region->start_pfn = start_pfn;
1018 region->end_pfn = end_pfn;
1019 list_add_tail(®ion->list, &nosave_regions);
1020 Report:
1021 pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
1022 (unsigned long long) start_pfn << PAGE_SHIFT,
1023 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1024}
1025
1026/*
1027 * Set bits in this map correspond to the page frames the contents of which
1028 * should not be saved during the suspend.
1029 */
1030static struct memory_bitmap *forbidden_pages_map;
1031
1032/* Set bits in this map correspond to free page frames. */
1033static struct memory_bitmap *free_pages_map;
1034
1035/*
1036 * Each page frame allocated for creating the image is marked by setting the
1037 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
1038 */
1039
1040void swsusp_set_page_free(struct page *page)
1041{
1042 if (free_pages_map)
1043 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
1044}
1045
1046static int swsusp_page_is_free(struct page *page)
1047{
1048 return free_pages_map ?
1049 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1050}
1051
1052void swsusp_unset_page_free(struct page *page)
1053{
1054 if (free_pages_map)
1055 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1056}
1057
1058static void swsusp_set_page_forbidden(struct page *page)
1059{
1060 if (forbidden_pages_map)
1061 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1062}
1063
1064int swsusp_page_is_forbidden(struct page *page)
1065{
1066 return forbidden_pages_map ?
1067 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1068}
1069
1070static void swsusp_unset_page_forbidden(struct page *page)
1071{
1072 if (forbidden_pages_map)
1073 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1074}
1075
1076/**
1077 * mark_nosave_pages - Mark pages that should not be saved.
1078 * @bm: Memory bitmap.
1079 *
1080 * Set the bits in @bm that correspond to the page frames the contents of which
1081 * should not be saved.
1082 */
1083static void mark_nosave_pages(struct memory_bitmap *bm)
1084{
1085 struct nosave_region *region;
1086
1087 if (list_empty(&nosave_regions))
1088 return;
1089
1090 list_for_each_entry(region, &nosave_regions, list) {
1091 unsigned long pfn;
1092
1093 pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1094 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1095 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1096 - 1);
1097
1098 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1099 if (pfn_valid(pfn)) {
1100 /*
1101 * It is safe to ignore the result of
1102 * mem_bm_set_bit_check() here, since we won't
1103 * touch the PFNs for which the error is
1104 * returned anyway.
1105 */
1106 mem_bm_set_bit_check(bm, pfn);
1107 }
1108 }
1109}
1110
1111/**
1112 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1113 *
1114 * Create bitmaps needed for marking page frames that should not be saved and
1115 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1116 * only modified if everything goes well, because we don't want the bits to be
1117 * touched before both bitmaps are set up.
1118 */
1119int create_basic_memory_bitmaps(void)
1120{
1121 struct memory_bitmap *bm1, *bm2;
1122 int error;
1123
1124 if (forbidden_pages_map && free_pages_map)
1125 return 0;
1126 else
1127 BUG_ON(forbidden_pages_map || free_pages_map);
1128
1129 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1130 if (!bm1)
1131 return -ENOMEM;
1132
1133 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1134 if (error)
1135 goto Free_first_object;
1136
1137 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1138 if (!bm2)
1139 goto Free_first_bitmap;
1140
1141 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1142 if (error)
1143 goto Free_second_object;
1144
1145 forbidden_pages_map = bm1;
1146 free_pages_map = bm2;
1147 mark_nosave_pages(forbidden_pages_map);
1148
1149 pr_debug("Basic memory bitmaps created\n");
1150
1151 return 0;
1152
1153 Free_second_object:
1154 kfree(bm2);
1155 Free_first_bitmap:
1156 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1157 Free_first_object:
1158 kfree(bm1);
1159 return -ENOMEM;
1160}
1161
1162/**
1163 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1164 *
1165 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1166 * auxiliary pointers are necessary so that the bitmaps themselves are not
1167 * referred to while they are being freed.
1168 */
1169void free_basic_memory_bitmaps(void)
1170{
1171 struct memory_bitmap *bm1, *bm2;
1172
1173 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1174 return;
1175
1176 bm1 = forbidden_pages_map;
1177 bm2 = free_pages_map;
1178 forbidden_pages_map = NULL;
1179 free_pages_map = NULL;
1180 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1181 kfree(bm1);
1182 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1183 kfree(bm2);
1184
1185 pr_debug("Basic memory bitmaps freed\n");
1186}
1187
1188static void clear_or_poison_free_page(struct page *page)
1189{
1190 if (page_poisoning_enabled_static())
1191 __kernel_poison_pages(page, 1);
1192 else if (want_init_on_free())
1193 clear_highpage(page);
1194}
1195
1196void clear_or_poison_free_pages(void)
1197{
1198 struct memory_bitmap *bm = free_pages_map;
1199 unsigned long pfn;
1200
1201 if (WARN_ON(!(free_pages_map)))
1202 return;
1203
1204 if (page_poisoning_enabled() || want_init_on_free()) {
1205 memory_bm_position_reset(bm);
1206 pfn = memory_bm_next_pfn(bm);
1207 while (pfn != BM_END_OF_MAP) {
1208 if (pfn_valid(pfn))
1209 clear_or_poison_free_page(pfn_to_page(pfn));
1210
1211 pfn = memory_bm_next_pfn(bm);
1212 }
1213 memory_bm_position_reset(bm);
1214 pr_info("free pages cleared after restore\n");
1215 }
1216}
1217
1218/**
1219 * snapshot_additional_pages - Estimate the number of extra pages needed.
1220 * @zone: Memory zone to carry out the computation for.
1221 *
1222 * Estimate the number of additional pages needed for setting up a hibernation
1223 * image data structures for @zone (usually, the returned value is greater than
1224 * the exact number).
1225 */
1226unsigned int snapshot_additional_pages(struct zone *zone)
1227{
1228 unsigned int rtree, nodes;
1229
1230 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1231 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1232 LINKED_PAGE_DATA_SIZE);
1233 while (nodes > 1) {
1234 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1235 rtree += nodes;
1236 }
1237
1238 return 2 * rtree;
1239}
1240
1241/*
1242 * Touch the watchdog for every WD_PAGE_COUNT pages.
1243 */
1244#define WD_PAGE_COUNT (128*1024)
1245
1246static void mark_free_pages(struct zone *zone)
1247{
1248 unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
1249 unsigned long flags;
1250 unsigned int order, t;
1251 struct page *page;
1252
1253 if (zone_is_empty(zone))
1254 return;
1255
1256 spin_lock_irqsave(&zone->lock, flags);
1257
1258 max_zone_pfn = zone_end_pfn(zone);
1259 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1260 if (pfn_valid(pfn)) {
1261 page = pfn_to_page(pfn);
1262
1263 if (!--page_count) {
1264 touch_nmi_watchdog();
1265 page_count = WD_PAGE_COUNT;
1266 }
1267
1268 if (page_zone(page) != zone)
1269 continue;
1270
1271 if (!swsusp_page_is_forbidden(page))
1272 swsusp_unset_page_free(page);
1273 }
1274
1275 for_each_migratetype_order(order, t) {
1276 list_for_each_entry(page,
1277 &zone->free_area[order].free_list[t], buddy_list) {
1278 unsigned long i;
1279
1280 pfn = page_to_pfn(page);
1281 for (i = 0; i < (1UL << order); i++) {
1282 if (!--page_count) {
1283 touch_nmi_watchdog();
1284 page_count = WD_PAGE_COUNT;
1285 }
1286 swsusp_set_page_free(pfn_to_page(pfn + i));
1287 }
1288 }
1289 }
1290 spin_unlock_irqrestore(&zone->lock, flags);
1291}
1292
1293#ifdef CONFIG_HIGHMEM
1294/**
1295 * count_free_highmem_pages - Compute the total number of free highmem pages.
1296 *
1297 * The returned number is system-wide.
1298 */
1299static unsigned int count_free_highmem_pages(void)
1300{
1301 struct zone *zone;
1302 unsigned int cnt = 0;
1303
1304 for_each_populated_zone(zone)
1305 if (is_highmem(zone))
1306 cnt += zone_page_state(zone, NR_FREE_PAGES);
1307
1308 return cnt;
1309}
1310
1311/**
1312 * saveable_highmem_page - Check if a highmem page is saveable.
1313 *
1314 * Determine whether a highmem page should be included in a hibernation image.
1315 *
1316 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1317 * and it isn't part of a free chunk of pages.
1318 */
1319static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1320{
1321 struct page *page;
1322
1323 if (!pfn_valid(pfn))
1324 return NULL;
1325
1326 page = pfn_to_online_page(pfn);
1327 if (!page || page_zone(page) != zone)
1328 return NULL;
1329
1330 BUG_ON(!PageHighMem(page));
1331
1332 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1333 return NULL;
1334
1335 if (PageReserved(page) || PageOffline(page))
1336 return NULL;
1337
1338 if (page_is_guard(page))
1339 return NULL;
1340
1341 return page;
1342}
1343
1344/**
1345 * count_highmem_pages - Compute the total number of saveable highmem pages.
1346 */
1347static unsigned int count_highmem_pages(void)
1348{
1349 struct zone *zone;
1350 unsigned int n = 0;
1351
1352 for_each_populated_zone(zone) {
1353 unsigned long pfn, max_zone_pfn;
1354
1355 if (!is_highmem(zone))
1356 continue;
1357
1358 mark_free_pages(zone);
1359 max_zone_pfn = zone_end_pfn(zone);
1360 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1361 if (saveable_highmem_page(zone, pfn))
1362 n++;
1363 }
1364 return n;
1365}
1366#else
1367static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1368{
1369 return NULL;
1370}
1371#endif /* CONFIG_HIGHMEM */
1372
1373/**
1374 * saveable_page - Check if the given page is saveable.
1375 *
1376 * Determine whether a non-highmem page should be included in a hibernation
1377 * image.
1378 *
1379 * We should save the page if it isn't Nosave, and is not in the range
1380 * of pages statically defined as 'unsaveable', and it isn't part of
1381 * a free chunk of pages.
1382 */
1383static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1384{
1385 struct page *page;
1386
1387 if (!pfn_valid(pfn))
1388 return NULL;
1389
1390 page = pfn_to_online_page(pfn);
1391 if (!page || page_zone(page) != zone)
1392 return NULL;
1393
1394 BUG_ON(PageHighMem(page));
1395
1396 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1397 return NULL;
1398
1399 if (PageOffline(page))
1400 return NULL;
1401
1402 if (PageReserved(page)
1403 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1404 return NULL;
1405
1406 if (page_is_guard(page))
1407 return NULL;
1408
1409 return page;
1410}
1411
1412/**
1413 * count_data_pages - Compute the total number of saveable non-highmem pages.
1414 */
1415static unsigned int count_data_pages(void)
1416{
1417 struct zone *zone;
1418 unsigned long pfn, max_zone_pfn;
1419 unsigned int n = 0;
1420
1421 for_each_populated_zone(zone) {
1422 if (is_highmem(zone))
1423 continue;
1424
1425 mark_free_pages(zone);
1426 max_zone_pfn = zone_end_pfn(zone);
1427 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1428 if (saveable_page(zone, pfn))
1429 n++;
1430 }
1431 return n;
1432}
1433
1434/*
1435 * This is needed, because copy_page and memcpy are not usable for copying
1436 * task structs. Returns true if the page was filled with only zeros,
1437 * otherwise false.
1438 */
1439static inline bool do_copy_page(long *dst, long *src)
1440{
1441 long z = 0;
1442 int n;
1443
1444 for (n = PAGE_SIZE / sizeof(long); n; n--) {
1445 z |= *src;
1446 *dst++ = *src++;
1447 }
1448 return !z;
1449}
1450
1451/**
1452 * safe_copy_page - Copy a page in a safe way.
1453 *
1454 * Check if the page we are going to copy is marked as present in the kernel
1455 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1456 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1457 * always returns 'true'. Returns true if the page was entirely composed of
1458 * zeros, otherwise it will return false.
1459 */
1460static bool safe_copy_page(void *dst, struct page *s_page)
1461{
1462 bool zeros_only;
1463
1464 if (kernel_page_present(s_page)) {
1465 zeros_only = do_copy_page(dst, page_address(s_page));
1466 } else {
1467 hibernate_map_page(s_page);
1468 zeros_only = do_copy_page(dst, page_address(s_page));
1469 hibernate_unmap_page(s_page);
1470 }
1471 return zeros_only;
1472}
1473
1474#ifdef CONFIG_HIGHMEM
1475static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1476{
1477 return is_highmem(zone) ?
1478 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1479}
1480
1481static bool copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1482{
1483 struct page *s_page, *d_page;
1484 void *src, *dst;
1485 bool zeros_only;
1486
1487 s_page = pfn_to_page(src_pfn);
1488 d_page = pfn_to_page(dst_pfn);
1489 if (PageHighMem(s_page)) {
1490 src = kmap_local_page(s_page);
1491 dst = kmap_local_page(d_page);
1492 zeros_only = do_copy_page(dst, src);
1493 kunmap_local(dst);
1494 kunmap_local(src);
1495 } else {
1496 if (PageHighMem(d_page)) {
1497 /*
1498 * The page pointed to by src may contain some kernel
1499 * data modified by kmap_atomic()
1500 */
1501 zeros_only = safe_copy_page(buffer, s_page);
1502 dst = kmap_local_page(d_page);
1503 copy_page(dst, buffer);
1504 kunmap_local(dst);
1505 } else {
1506 zeros_only = safe_copy_page(page_address(d_page), s_page);
1507 }
1508 }
1509 return zeros_only;
1510}
1511#else
1512#define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1513
1514static inline int copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1515{
1516 return safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1517 pfn_to_page(src_pfn));
1518}
1519#endif /* CONFIG_HIGHMEM */
1520
1521/*
1522 * Copy data pages will copy all pages into pages pulled from the copy_bm.
1523 * If a page was entirely filled with zeros it will be marked in the zero_bm.
1524 *
1525 * Returns the number of pages copied.
1526 */
1527static unsigned long copy_data_pages(struct memory_bitmap *copy_bm,
1528 struct memory_bitmap *orig_bm,
1529 struct memory_bitmap *zero_bm)
1530{
1531 unsigned long copied_pages = 0;
1532 struct zone *zone;
1533 unsigned long pfn, copy_pfn;
1534
1535 for_each_populated_zone(zone) {
1536 unsigned long max_zone_pfn;
1537
1538 mark_free_pages(zone);
1539 max_zone_pfn = zone_end_pfn(zone);
1540 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1541 if (page_is_saveable(zone, pfn))
1542 memory_bm_set_bit(orig_bm, pfn);
1543 }
1544 memory_bm_position_reset(orig_bm);
1545 memory_bm_position_reset(copy_bm);
1546 copy_pfn = memory_bm_next_pfn(copy_bm);
1547 for(;;) {
1548 pfn = memory_bm_next_pfn(orig_bm);
1549 if (unlikely(pfn == BM_END_OF_MAP))
1550 break;
1551 if (copy_data_page(copy_pfn, pfn)) {
1552 memory_bm_set_bit(zero_bm, pfn);
1553 /* Use this copy_pfn for a page that is not full of zeros */
1554 continue;
1555 }
1556 copied_pages++;
1557 copy_pfn = memory_bm_next_pfn(copy_bm);
1558 }
1559 return copied_pages;
1560}
1561
1562/* Total number of image pages */
1563static unsigned int nr_copy_pages;
1564/* Number of pages needed for saving the original pfns of the image pages */
1565static unsigned int nr_meta_pages;
1566/* Number of zero pages */
1567static unsigned int nr_zero_pages;
1568
1569/*
1570 * Numbers of normal and highmem page frames allocated for hibernation image
1571 * before suspending devices.
1572 */
1573static unsigned int alloc_normal, alloc_highmem;
1574/*
1575 * Memory bitmap used for marking saveable pages (during hibernation) or
1576 * hibernation image pages (during restore)
1577 */
1578static struct memory_bitmap orig_bm;
1579/*
1580 * Memory bitmap used during hibernation for marking allocated page frames that
1581 * will contain copies of saveable pages. During restore it is initially used
1582 * for marking hibernation image pages, but then the set bits from it are
1583 * duplicated in @orig_bm and it is released. On highmem systems it is next
1584 * used for marking "safe" highmem pages, but it has to be reinitialized for
1585 * this purpose.
1586 */
1587static struct memory_bitmap copy_bm;
1588
1589/* Memory bitmap which tracks which saveable pages were zero filled. */
1590static struct memory_bitmap zero_bm;
1591
1592/**
1593 * swsusp_free - Free pages allocated for hibernation image.
1594 *
1595 * Image pages are allocated before snapshot creation, so they need to be
1596 * released after resume.
1597 */
1598void swsusp_free(void)
1599{
1600 unsigned long fb_pfn, fr_pfn;
1601
1602 if (!forbidden_pages_map || !free_pages_map)
1603 goto out;
1604
1605 memory_bm_position_reset(forbidden_pages_map);
1606 memory_bm_position_reset(free_pages_map);
1607
1608loop:
1609 fr_pfn = memory_bm_next_pfn(free_pages_map);
1610 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1611
1612 /*
1613 * Find the next bit set in both bitmaps. This is guaranteed to
1614 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1615 */
1616 do {
1617 if (fb_pfn < fr_pfn)
1618 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1619 if (fr_pfn < fb_pfn)
1620 fr_pfn = memory_bm_next_pfn(free_pages_map);
1621 } while (fb_pfn != fr_pfn);
1622
1623 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1624 struct page *page = pfn_to_page(fr_pfn);
1625
1626 memory_bm_clear_current(forbidden_pages_map);
1627 memory_bm_clear_current(free_pages_map);
1628 hibernate_restore_unprotect_page(page_address(page));
1629 __free_page(page);
1630 goto loop;
1631 }
1632
1633out:
1634 nr_copy_pages = 0;
1635 nr_meta_pages = 0;
1636 nr_zero_pages = 0;
1637 restore_pblist = NULL;
1638 buffer = NULL;
1639 alloc_normal = 0;
1640 alloc_highmem = 0;
1641 hibernate_restore_protection_end();
1642}
1643
1644/* Helper functions used for the shrinking of memory. */
1645
1646#define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1647
1648/**
1649 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1650 * @nr_pages: Number of page frames to allocate.
1651 * @mask: GFP flags to use for the allocation.
1652 *
1653 * Return value: Number of page frames actually allocated
1654 */
1655static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1656{
1657 unsigned long nr_alloc = 0;
1658
1659 while (nr_pages > 0) {
1660 struct page *page;
1661
1662 page = alloc_image_page(mask);
1663 if (!page)
1664 break;
1665 memory_bm_set_bit(©_bm, page_to_pfn(page));
1666 if (PageHighMem(page))
1667 alloc_highmem++;
1668 else
1669 alloc_normal++;
1670 nr_pages--;
1671 nr_alloc++;
1672 }
1673
1674 return nr_alloc;
1675}
1676
1677static unsigned long preallocate_image_memory(unsigned long nr_pages,
1678 unsigned long avail_normal)
1679{
1680 unsigned long alloc;
1681
1682 if (avail_normal <= alloc_normal)
1683 return 0;
1684
1685 alloc = avail_normal - alloc_normal;
1686 if (nr_pages < alloc)
1687 alloc = nr_pages;
1688
1689 return preallocate_image_pages(alloc, GFP_IMAGE);
1690}
1691
1692#ifdef CONFIG_HIGHMEM
1693static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1694{
1695 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1696}
1697
1698/**
1699 * __fraction - Compute (an approximation of) x * (multiplier / base).
1700 */
1701static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1702{
1703 return div64_u64(x * multiplier, base);
1704}
1705
1706static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1707 unsigned long highmem,
1708 unsigned long total)
1709{
1710 unsigned long alloc = __fraction(nr_pages, highmem, total);
1711
1712 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1713}
1714#else /* CONFIG_HIGHMEM */
1715static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1716{
1717 return 0;
1718}
1719
1720static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1721 unsigned long highmem,
1722 unsigned long total)
1723{
1724 return 0;
1725}
1726#endif /* CONFIG_HIGHMEM */
1727
1728/**
1729 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1730 */
1731static unsigned long free_unnecessary_pages(void)
1732{
1733 unsigned long save, to_free_normal, to_free_highmem, free;
1734
1735 save = count_data_pages();
1736 if (alloc_normal >= save) {
1737 to_free_normal = alloc_normal - save;
1738 save = 0;
1739 } else {
1740 to_free_normal = 0;
1741 save -= alloc_normal;
1742 }
1743 save += count_highmem_pages();
1744 if (alloc_highmem >= save) {
1745 to_free_highmem = alloc_highmem - save;
1746 } else {
1747 to_free_highmem = 0;
1748 save -= alloc_highmem;
1749 if (to_free_normal > save)
1750 to_free_normal -= save;
1751 else
1752 to_free_normal = 0;
1753 }
1754 free = to_free_normal + to_free_highmem;
1755
1756 memory_bm_position_reset(©_bm);
1757
1758 while (to_free_normal > 0 || to_free_highmem > 0) {
1759 unsigned long pfn = memory_bm_next_pfn(©_bm);
1760 struct page *page = pfn_to_page(pfn);
1761
1762 if (PageHighMem(page)) {
1763 if (!to_free_highmem)
1764 continue;
1765 to_free_highmem--;
1766 alloc_highmem--;
1767 } else {
1768 if (!to_free_normal)
1769 continue;
1770 to_free_normal--;
1771 alloc_normal--;
1772 }
1773 memory_bm_clear_bit(©_bm, pfn);
1774 swsusp_unset_page_forbidden(page);
1775 swsusp_unset_page_free(page);
1776 __free_page(page);
1777 }
1778
1779 return free;
1780}
1781
1782/**
1783 * minimum_image_size - Estimate the minimum acceptable size of an image.
1784 * @saveable: Number of saveable pages in the system.
1785 *
1786 * We want to avoid attempting to free too much memory too hard, so estimate the
1787 * minimum acceptable size of a hibernation image to use as the lower limit for
1788 * preallocating memory.
1789 *
1790 * We assume that the minimum image size should be proportional to
1791 *
1792 * [number of saveable pages] - [number of pages that can be freed in theory]
1793 *
1794 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1795 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1796 */
1797static unsigned long minimum_image_size(unsigned long saveable)
1798{
1799 unsigned long size;
1800
1801 size = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B)
1802 + global_node_page_state(NR_ACTIVE_ANON)
1803 + global_node_page_state(NR_INACTIVE_ANON)
1804 + global_node_page_state(NR_ACTIVE_FILE)
1805 + global_node_page_state(NR_INACTIVE_FILE);
1806
1807 return saveable <= size ? 0 : saveable - size;
1808}
1809
1810/**
1811 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1812 *
1813 * To create a hibernation image it is necessary to make a copy of every page
1814 * frame in use. We also need a number of page frames to be free during
1815 * hibernation for allocations made while saving the image and for device
1816 * drivers, in case they need to allocate memory from their hibernation
1817 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1818 * estimate) and reserved_size divided by PAGE_SIZE (which is tunable through
1819 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1820 * total number of available page frames and allocate at least
1821 *
1822 * ([page frames total] - PAGES_FOR_IO - [metadata pages]) / 2
1823 * - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1824 *
1825 * of them, which corresponds to the maximum size of a hibernation image.
1826 *
1827 * If image_size is set below the number following from the above formula,
1828 * the preallocation of memory is continued until the total number of saveable
1829 * pages in the system is below the requested image size or the minimum
1830 * acceptable image size returned by minimum_image_size(), whichever is greater.
1831 */
1832int hibernate_preallocate_memory(void)
1833{
1834 struct zone *zone;
1835 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1836 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1837 ktime_t start, stop;
1838 int error;
1839
1840 pr_info("Preallocating image memory\n");
1841 start = ktime_get();
1842
1843 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1844 if (error) {
1845 pr_err("Cannot allocate original bitmap\n");
1846 goto err_out;
1847 }
1848
1849 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1850 if (error) {
1851 pr_err("Cannot allocate copy bitmap\n");
1852 goto err_out;
1853 }
1854
1855 error = memory_bm_create(&zero_bm, GFP_IMAGE, PG_ANY);
1856 if (error) {
1857 pr_err("Cannot allocate zero bitmap\n");
1858 goto err_out;
1859 }
1860
1861 alloc_normal = 0;
1862 alloc_highmem = 0;
1863 nr_zero_pages = 0;
1864
1865 /* Count the number of saveable data pages. */
1866 save_highmem = count_highmem_pages();
1867 saveable = count_data_pages();
1868
1869 /*
1870 * Compute the total number of page frames we can use (count) and the
1871 * number of pages needed for image metadata (size).
1872 */
1873 count = saveable;
1874 saveable += save_highmem;
1875 highmem = save_highmem;
1876 size = 0;
1877 for_each_populated_zone(zone) {
1878 size += snapshot_additional_pages(zone);
1879 if (is_highmem(zone))
1880 highmem += zone_page_state(zone, NR_FREE_PAGES);
1881 else
1882 count += zone_page_state(zone, NR_FREE_PAGES);
1883 }
1884 avail_normal = count;
1885 count += highmem;
1886 count -= totalreserve_pages;
1887
1888 /* Compute the maximum number of saveable pages to leave in memory. */
1889 max_size = (count - (size + PAGES_FOR_IO)) / 2
1890 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1891 /* Compute the desired number of image pages specified by image_size. */
1892 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1893 if (size > max_size)
1894 size = max_size;
1895 /*
1896 * If the desired number of image pages is at least as large as the
1897 * current number of saveable pages in memory, allocate page frames for
1898 * the image and we're done.
1899 */
1900 if (size >= saveable) {
1901 pages = preallocate_image_highmem(save_highmem);
1902 pages += preallocate_image_memory(saveable - pages, avail_normal);
1903 goto out;
1904 }
1905
1906 /* Estimate the minimum size of the image. */
1907 pages = minimum_image_size(saveable);
1908 /*
1909 * To avoid excessive pressure on the normal zone, leave room in it to
1910 * accommodate an image of the minimum size (unless it's already too
1911 * small, in which case don't preallocate pages from it at all).
1912 */
1913 if (avail_normal > pages)
1914 avail_normal -= pages;
1915 else
1916 avail_normal = 0;
1917 if (size < pages)
1918 size = min_t(unsigned long, pages, max_size);
1919
1920 /*
1921 * Let the memory management subsystem know that we're going to need a
1922 * large number of page frames to allocate and make it free some memory.
1923 * NOTE: If this is not done, performance will be hurt badly in some
1924 * test cases.
1925 */
1926 shrink_all_memory(saveable - size);
1927
1928 /*
1929 * The number of saveable pages in memory was too high, so apply some
1930 * pressure to decrease it. First, make room for the largest possible
1931 * image and fail if that doesn't work. Next, try to decrease the size
1932 * of the image as much as indicated by 'size' using allocations from
1933 * highmem and non-highmem zones separately.
1934 */
1935 pages_highmem = preallocate_image_highmem(highmem / 2);
1936 alloc = count - max_size;
1937 if (alloc > pages_highmem)
1938 alloc -= pages_highmem;
1939 else
1940 alloc = 0;
1941 pages = preallocate_image_memory(alloc, avail_normal);
1942 if (pages < alloc) {
1943 /* We have exhausted non-highmem pages, try highmem. */
1944 alloc -= pages;
1945 pages += pages_highmem;
1946 pages_highmem = preallocate_image_highmem(alloc);
1947 if (pages_highmem < alloc) {
1948 pr_err("Image allocation is %lu pages short\n",
1949 alloc - pages_highmem);
1950 goto err_out;
1951 }
1952 pages += pages_highmem;
1953 /*
1954 * size is the desired number of saveable pages to leave in
1955 * memory, so try to preallocate (all memory - size) pages.
1956 */
1957 alloc = (count - pages) - size;
1958 pages += preallocate_image_highmem(alloc);
1959 } else {
1960 /*
1961 * There are approximately max_size saveable pages at this point
1962 * and we want to reduce this number down to size.
1963 */
1964 alloc = max_size - size;
1965 size = preallocate_highmem_fraction(alloc, highmem, count);
1966 pages_highmem += size;
1967 alloc -= size;
1968 size = preallocate_image_memory(alloc, avail_normal);
1969 pages_highmem += preallocate_image_highmem(alloc - size);
1970 pages += pages_highmem + size;
1971 }
1972
1973 /*
1974 * We only need as many page frames for the image as there are saveable
1975 * pages in memory, but we have allocated more. Release the excessive
1976 * ones now.
1977 */
1978 pages -= free_unnecessary_pages();
1979
1980 out:
1981 stop = ktime_get();
1982 pr_info("Allocated %lu pages for snapshot\n", pages);
1983 swsusp_show_speed(start, stop, pages, "Allocated");
1984
1985 return 0;
1986
1987 err_out:
1988 swsusp_free();
1989 return -ENOMEM;
1990}
1991
1992#ifdef CONFIG_HIGHMEM
1993/**
1994 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1995 *
1996 * Compute the number of non-highmem pages that will be necessary for creating
1997 * copies of highmem pages.
1998 */
1999static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
2000{
2001 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
2002
2003 if (free_highmem >= nr_highmem)
2004 nr_highmem = 0;
2005 else
2006 nr_highmem -= free_highmem;
2007
2008 return nr_highmem;
2009}
2010#else
2011static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
2012#endif /* CONFIG_HIGHMEM */
2013
2014/**
2015 * enough_free_mem - Check if there is enough free memory for the image.
2016 */
2017static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
2018{
2019 struct zone *zone;
2020 unsigned int free = alloc_normal;
2021
2022 for_each_populated_zone(zone)
2023 if (!is_highmem(zone))
2024 free += zone_page_state(zone, NR_FREE_PAGES);
2025
2026 nr_pages += count_pages_for_highmem(nr_highmem);
2027 pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
2028 nr_pages, PAGES_FOR_IO, free);
2029
2030 return free > nr_pages + PAGES_FOR_IO;
2031}
2032
2033#ifdef CONFIG_HIGHMEM
2034/**
2035 * get_highmem_buffer - Allocate a buffer for highmem pages.
2036 *
2037 * If there are some highmem pages in the hibernation image, we may need a
2038 * buffer to copy them and/or load their data.
2039 */
2040static inline int get_highmem_buffer(int safe_needed)
2041{
2042 buffer = get_image_page(GFP_ATOMIC, safe_needed);
2043 return buffer ? 0 : -ENOMEM;
2044}
2045
2046/**
2047 * alloc_highmem_pages - Allocate some highmem pages for the image.
2048 *
2049 * Try to allocate as many pages as needed, but if the number of free highmem
2050 * pages is less than that, allocate them all.
2051 */
2052static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
2053 unsigned int nr_highmem)
2054{
2055 unsigned int to_alloc = count_free_highmem_pages();
2056
2057 if (to_alloc > nr_highmem)
2058 to_alloc = nr_highmem;
2059
2060 nr_highmem -= to_alloc;
2061 while (to_alloc-- > 0) {
2062 struct page *page;
2063
2064 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
2065 memory_bm_set_bit(bm, page_to_pfn(page));
2066 }
2067 return nr_highmem;
2068}
2069#else
2070static inline int get_highmem_buffer(int safe_needed) { return 0; }
2071
2072static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
2073 unsigned int n) { return 0; }
2074#endif /* CONFIG_HIGHMEM */
2075
2076/**
2077 * swsusp_alloc - Allocate memory for hibernation image.
2078 *
2079 * We first try to allocate as many highmem pages as there are
2080 * saveable highmem pages in the system. If that fails, we allocate
2081 * non-highmem pages for the copies of the remaining highmem ones.
2082 *
2083 * In this approach it is likely that the copies of highmem pages will
2084 * also be located in the high memory, because of the way in which
2085 * copy_data_pages() works.
2086 */
2087static int swsusp_alloc(struct memory_bitmap *copy_bm,
2088 unsigned int nr_pages, unsigned int nr_highmem)
2089{
2090 if (nr_highmem > 0) {
2091 if (get_highmem_buffer(PG_ANY))
2092 goto err_out;
2093 if (nr_highmem > alloc_highmem) {
2094 nr_highmem -= alloc_highmem;
2095 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
2096 }
2097 }
2098 if (nr_pages > alloc_normal) {
2099 nr_pages -= alloc_normal;
2100 while (nr_pages-- > 0) {
2101 struct page *page;
2102
2103 page = alloc_image_page(GFP_ATOMIC);
2104 if (!page)
2105 goto err_out;
2106 memory_bm_set_bit(copy_bm, page_to_pfn(page));
2107 }
2108 }
2109
2110 return 0;
2111
2112 err_out:
2113 swsusp_free();
2114 return -ENOMEM;
2115}
2116
2117asmlinkage __visible int swsusp_save(void)
2118{
2119 unsigned int nr_pages, nr_highmem;
2120
2121 pr_info("Creating image:\n");
2122
2123 drain_local_pages(NULL);
2124 nr_pages = count_data_pages();
2125 nr_highmem = count_highmem_pages();
2126 pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
2127
2128 if (!enough_free_mem(nr_pages, nr_highmem)) {
2129 pr_err("Not enough free memory\n");
2130 return -ENOMEM;
2131 }
2132
2133 if (swsusp_alloc(©_bm, nr_pages, nr_highmem)) {
2134 pr_err("Memory allocation failed\n");
2135 return -ENOMEM;
2136 }
2137
2138 /*
2139 * During allocating of suspend pagedir, new cold pages may appear.
2140 * Kill them.
2141 */
2142 drain_local_pages(NULL);
2143 nr_copy_pages = copy_data_pages(©_bm, &orig_bm, &zero_bm);
2144
2145 /*
2146 * End of critical section. From now on, we can write to memory,
2147 * but we should not touch disk. This specially means we must _not_
2148 * touch swap space! Except we must write out our image of course.
2149 */
2150 nr_pages += nr_highmem;
2151 /* We don't actually copy the zero pages */
2152 nr_zero_pages = nr_pages - nr_copy_pages;
2153 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2154
2155 pr_info("Image created (%d pages copied, %d zero pages)\n", nr_copy_pages, nr_zero_pages);
2156
2157 return 0;
2158}
2159
2160#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2161static int init_header_complete(struct swsusp_info *info)
2162{
2163 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2164 info->version_code = LINUX_VERSION_CODE;
2165 return 0;
2166}
2167
2168static const char *check_image_kernel(struct swsusp_info *info)
2169{
2170 if (info->version_code != LINUX_VERSION_CODE)
2171 return "kernel version";
2172 if (strcmp(info->uts.sysname,init_utsname()->sysname))
2173 return "system type";
2174 if (strcmp(info->uts.release,init_utsname()->release))
2175 return "kernel release";
2176 if (strcmp(info->uts.version,init_utsname()->version))
2177 return "version";
2178 if (strcmp(info->uts.machine,init_utsname()->machine))
2179 return "machine";
2180 return NULL;
2181}
2182#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2183
2184unsigned long snapshot_get_image_size(void)
2185{
2186 return nr_copy_pages + nr_meta_pages + 1;
2187}
2188
2189static int init_header(struct swsusp_info *info)
2190{
2191 memset(info, 0, sizeof(struct swsusp_info));
2192 info->num_physpages = get_num_physpages();
2193 info->image_pages = nr_copy_pages;
2194 info->pages = snapshot_get_image_size();
2195 info->size = info->pages;
2196 info->size <<= PAGE_SHIFT;
2197 return init_header_complete(info);
2198}
2199
2200#define ENCODED_PFN_ZERO_FLAG ((unsigned long)1 << (BITS_PER_LONG - 1))
2201#define ENCODED_PFN_MASK (~ENCODED_PFN_ZERO_FLAG)
2202
2203/**
2204 * pack_pfns - Prepare PFNs for saving.
2205 * @bm: Memory bitmap.
2206 * @buf: Memory buffer to store the PFNs in.
2207 * @zero_bm: Memory bitmap containing PFNs of zero pages.
2208 *
2209 * PFNs corresponding to set bits in @bm are stored in the area of memory
2210 * pointed to by @buf (1 page at a time). Pages which were filled with only
2211 * zeros will have the highest bit set in the packed format to distinguish
2212 * them from PFNs which will be contained in the image file.
2213 */
2214static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm,
2215 struct memory_bitmap *zero_bm)
2216{
2217 int j;
2218
2219 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2220 buf[j] = memory_bm_next_pfn(bm);
2221 if (unlikely(buf[j] == BM_END_OF_MAP))
2222 break;
2223 if (memory_bm_test_bit(zero_bm, buf[j]))
2224 buf[j] |= ENCODED_PFN_ZERO_FLAG;
2225 }
2226}
2227
2228/**
2229 * snapshot_read_next - Get the address to read the next image page from.
2230 * @handle: Snapshot handle to be used for the reading.
2231 *
2232 * On the first call, @handle should point to a zeroed snapshot_handle
2233 * structure. The structure gets populated then and a pointer to it should be
2234 * passed to this function every next time.
2235 *
2236 * On success, the function returns a positive number. Then, the caller
2237 * is allowed to read up to the returned number of bytes from the memory
2238 * location computed by the data_of() macro.
2239 *
2240 * The function returns 0 to indicate the end of the data stream condition,
2241 * and negative numbers are returned on errors. If that happens, the structure
2242 * pointed to by @handle is not updated and should not be used any more.
2243 */
2244int snapshot_read_next(struct snapshot_handle *handle)
2245{
2246 if (handle->cur > nr_meta_pages + nr_copy_pages)
2247 return 0;
2248
2249 if (!buffer) {
2250 /* This makes the buffer be freed by swsusp_free() */
2251 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2252 if (!buffer)
2253 return -ENOMEM;
2254 }
2255 if (!handle->cur) {
2256 int error;
2257
2258 error = init_header((struct swsusp_info *)buffer);
2259 if (error)
2260 return error;
2261 handle->buffer = buffer;
2262 memory_bm_position_reset(&orig_bm);
2263 memory_bm_position_reset(©_bm);
2264 } else if (handle->cur <= nr_meta_pages) {
2265 clear_page(buffer);
2266 pack_pfns(buffer, &orig_bm, &zero_bm);
2267 } else {
2268 struct page *page;
2269
2270 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2271 if (PageHighMem(page)) {
2272 /*
2273 * Highmem pages are copied to the buffer,
2274 * because we can't return with a kmapped
2275 * highmem page (we may not be called again).
2276 */
2277 void *kaddr;
2278
2279 kaddr = kmap_atomic(page);
2280 copy_page(buffer, kaddr);
2281 kunmap_atomic(kaddr);
2282 handle->buffer = buffer;
2283 } else {
2284 handle->buffer = page_address(page);
2285 }
2286 }
2287 handle->cur++;
2288 return PAGE_SIZE;
2289}
2290
2291static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2292 struct memory_bitmap *src)
2293{
2294 unsigned long pfn;
2295
2296 memory_bm_position_reset(src);
2297 pfn = memory_bm_next_pfn(src);
2298 while (pfn != BM_END_OF_MAP) {
2299 memory_bm_set_bit(dst, pfn);
2300 pfn = memory_bm_next_pfn(src);
2301 }
2302}
2303
2304/**
2305 * mark_unsafe_pages - Mark pages that were used before hibernation.
2306 *
2307 * Mark the pages that cannot be used for storing the image during restoration,
2308 * because they conflict with the pages that had been used before hibernation.
2309 */
2310static void mark_unsafe_pages(struct memory_bitmap *bm)
2311{
2312 unsigned long pfn;
2313
2314 /* Clear the "free"/"unsafe" bit for all PFNs */
2315 memory_bm_position_reset(free_pages_map);
2316 pfn = memory_bm_next_pfn(free_pages_map);
2317 while (pfn != BM_END_OF_MAP) {
2318 memory_bm_clear_current(free_pages_map);
2319 pfn = memory_bm_next_pfn(free_pages_map);
2320 }
2321
2322 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2323 duplicate_memory_bitmap(free_pages_map, bm);
2324
2325 allocated_unsafe_pages = 0;
2326}
2327
2328static int check_header(struct swsusp_info *info)
2329{
2330 const char *reason;
2331
2332 reason = check_image_kernel(info);
2333 if (!reason && info->num_physpages != get_num_physpages())
2334 reason = "memory size";
2335 if (reason) {
2336 pr_err("Image mismatch: %s\n", reason);
2337 return -EPERM;
2338 }
2339 return 0;
2340}
2341
2342/**
2343 * load_header - Check the image header and copy the data from it.
2344 */
2345static int load_header(struct swsusp_info *info)
2346{
2347 int error;
2348
2349 restore_pblist = NULL;
2350 error = check_header(info);
2351 if (!error) {
2352 nr_copy_pages = info->image_pages;
2353 nr_meta_pages = info->pages - info->image_pages - 1;
2354 }
2355 return error;
2356}
2357
2358/**
2359 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2360 * @bm: Memory bitmap.
2361 * @buf: Area of memory containing the PFNs.
2362 * @zero_bm: Memory bitmap with the zero PFNs marked.
2363 *
2364 * For each element of the array pointed to by @buf (1 page at a time), set the
2365 * corresponding bit in @bm. If the page was originally populated with only
2366 * zeros then a corresponding bit will also be set in @zero_bm.
2367 */
2368static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm,
2369 struct memory_bitmap *zero_bm)
2370{
2371 unsigned long decoded_pfn;
2372 bool zero;
2373 int j;
2374
2375 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2376 if (unlikely(buf[j] == BM_END_OF_MAP))
2377 break;
2378
2379 zero = !!(buf[j] & ENCODED_PFN_ZERO_FLAG);
2380 decoded_pfn = buf[j] & ENCODED_PFN_MASK;
2381 if (pfn_valid(decoded_pfn) && memory_bm_pfn_present(bm, decoded_pfn)) {
2382 memory_bm_set_bit(bm, decoded_pfn);
2383 if (zero) {
2384 memory_bm_set_bit(zero_bm, decoded_pfn);
2385 nr_zero_pages++;
2386 }
2387 } else {
2388 if (!pfn_valid(decoded_pfn))
2389 pr_err(FW_BUG "Memory map mismatch at 0x%llx after hibernation\n",
2390 (unsigned long long)PFN_PHYS(decoded_pfn));
2391 return -EFAULT;
2392 }
2393 }
2394
2395 return 0;
2396}
2397
2398#ifdef CONFIG_HIGHMEM
2399/*
2400 * struct highmem_pbe is used for creating the list of highmem pages that
2401 * should be restored atomically during the resume from disk, because the page
2402 * frames they have occupied before the suspend are in use.
2403 */
2404struct highmem_pbe {
2405 struct page *copy_page; /* data is here now */
2406 struct page *orig_page; /* data was here before the suspend */
2407 struct highmem_pbe *next;
2408};
2409
2410/*
2411 * List of highmem PBEs needed for restoring the highmem pages that were
2412 * allocated before the suspend and included in the suspend image, but have
2413 * also been allocated by the "resume" kernel, so their contents cannot be
2414 * written directly to their "original" page frames.
2415 */
2416static struct highmem_pbe *highmem_pblist;
2417
2418/**
2419 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2420 * @bm: Memory bitmap.
2421 *
2422 * The bits in @bm that correspond to image pages are assumed to be set.
2423 */
2424static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2425{
2426 unsigned long pfn;
2427 unsigned int cnt = 0;
2428
2429 memory_bm_position_reset(bm);
2430 pfn = memory_bm_next_pfn(bm);
2431 while (pfn != BM_END_OF_MAP) {
2432 if (PageHighMem(pfn_to_page(pfn)))
2433 cnt++;
2434
2435 pfn = memory_bm_next_pfn(bm);
2436 }
2437 return cnt;
2438}
2439
2440static unsigned int safe_highmem_pages;
2441
2442static struct memory_bitmap *safe_highmem_bm;
2443
2444/**
2445 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2446 * @bm: Pointer to an uninitialized memory bitmap structure.
2447 * @nr_highmem_p: Pointer to the number of highmem image pages.
2448 *
2449 * Try to allocate as many highmem pages as there are highmem image pages
2450 * (@nr_highmem_p points to the variable containing the number of highmem image
2451 * pages). The pages that are "safe" (ie. will not be overwritten when the
2452 * hibernation image is restored entirely) have the corresponding bits set in
2453 * @bm (it must be uninitialized).
2454 *
2455 * NOTE: This function should not be called if there are no highmem image pages.
2456 */
2457static int prepare_highmem_image(struct memory_bitmap *bm,
2458 unsigned int *nr_highmem_p)
2459{
2460 unsigned int to_alloc;
2461
2462 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2463 return -ENOMEM;
2464
2465 if (get_highmem_buffer(PG_SAFE))
2466 return -ENOMEM;
2467
2468 to_alloc = count_free_highmem_pages();
2469 if (to_alloc > *nr_highmem_p)
2470 to_alloc = *nr_highmem_p;
2471 else
2472 *nr_highmem_p = to_alloc;
2473
2474 safe_highmem_pages = 0;
2475 while (to_alloc-- > 0) {
2476 struct page *page;
2477
2478 page = alloc_page(__GFP_HIGHMEM);
2479 if (!swsusp_page_is_free(page)) {
2480 /* The page is "safe", set its bit the bitmap */
2481 memory_bm_set_bit(bm, page_to_pfn(page));
2482 safe_highmem_pages++;
2483 }
2484 /* Mark the page as allocated */
2485 swsusp_set_page_forbidden(page);
2486 swsusp_set_page_free(page);
2487 }
2488 memory_bm_position_reset(bm);
2489 safe_highmem_bm = bm;
2490 return 0;
2491}
2492
2493static struct page *last_highmem_page;
2494
2495/**
2496 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2497 *
2498 * For a given highmem image page get a buffer that suspend_write_next() should
2499 * return to its caller to write to.
2500 *
2501 * If the page is to be saved to its "original" page frame or a copy of
2502 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2503 * the copy of the page is to be made in normal memory, so the address of
2504 * the copy is returned.
2505 *
2506 * If @buffer is returned, the caller of suspend_write_next() will write
2507 * the page's contents to @buffer, so they will have to be copied to the
2508 * right location on the next call to suspend_write_next() and it is done
2509 * with the help of copy_last_highmem_page(). For this purpose, if
2510 * @buffer is returned, @last_highmem_page is set to the page to which
2511 * the data will have to be copied from @buffer.
2512 */
2513static void *get_highmem_page_buffer(struct page *page,
2514 struct chain_allocator *ca)
2515{
2516 struct highmem_pbe *pbe;
2517 void *kaddr;
2518
2519 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2520 /*
2521 * We have allocated the "original" page frame and we can
2522 * use it directly to store the loaded page.
2523 */
2524 last_highmem_page = page;
2525 return buffer;
2526 }
2527 /*
2528 * The "original" page frame has not been allocated and we have to
2529 * use a "safe" page frame to store the loaded page.
2530 */
2531 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2532 if (!pbe) {
2533 swsusp_free();
2534 return ERR_PTR(-ENOMEM);
2535 }
2536 pbe->orig_page = page;
2537 if (safe_highmem_pages > 0) {
2538 struct page *tmp;
2539
2540 /* Copy of the page will be stored in high memory */
2541 kaddr = buffer;
2542 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2543 safe_highmem_pages--;
2544 last_highmem_page = tmp;
2545 pbe->copy_page = tmp;
2546 } else {
2547 /* Copy of the page will be stored in normal memory */
2548 kaddr = __get_safe_page(ca->gfp_mask);
2549 if (!kaddr)
2550 return ERR_PTR(-ENOMEM);
2551 pbe->copy_page = virt_to_page(kaddr);
2552 }
2553 pbe->next = highmem_pblist;
2554 highmem_pblist = pbe;
2555 return kaddr;
2556}
2557
2558/**
2559 * copy_last_highmem_page - Copy most the most recent highmem image page.
2560 *
2561 * Copy the contents of a highmem image from @buffer, where the caller of
2562 * snapshot_write_next() has stored them, to the right location represented by
2563 * @last_highmem_page .
2564 */
2565static void copy_last_highmem_page(void)
2566{
2567 if (last_highmem_page) {
2568 void *dst;
2569
2570 dst = kmap_atomic(last_highmem_page);
2571 copy_page(dst, buffer);
2572 kunmap_atomic(dst);
2573 last_highmem_page = NULL;
2574 }
2575}
2576
2577static inline int last_highmem_page_copied(void)
2578{
2579 return !last_highmem_page;
2580}
2581
2582static inline void free_highmem_data(void)
2583{
2584 if (safe_highmem_bm)
2585 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2586
2587 if (buffer)
2588 free_image_page(buffer, PG_UNSAFE_CLEAR);
2589}
2590#else
2591static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2592
2593static inline int prepare_highmem_image(struct memory_bitmap *bm,
2594 unsigned int *nr_highmem_p) { return 0; }
2595
2596static inline void *get_highmem_page_buffer(struct page *page,
2597 struct chain_allocator *ca)
2598{
2599 return ERR_PTR(-EINVAL);
2600}
2601
2602static inline void copy_last_highmem_page(void) {}
2603static inline int last_highmem_page_copied(void) { return 1; }
2604static inline void free_highmem_data(void) {}
2605#endif /* CONFIG_HIGHMEM */
2606
2607#define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2608
2609/**
2610 * prepare_image - Make room for loading hibernation image.
2611 * @new_bm: Uninitialized memory bitmap structure.
2612 * @bm: Memory bitmap with unsafe pages marked.
2613 * @zero_bm: Memory bitmap containing the zero pages.
2614 *
2615 * Use @bm to mark the pages that will be overwritten in the process of
2616 * restoring the system memory state from the suspend image ("unsafe" pages)
2617 * and allocate memory for the image.
2618 *
2619 * The idea is to allocate a new memory bitmap first and then allocate
2620 * as many pages as needed for image data, but without specifying what those
2621 * pages will be used for just yet. Instead, we mark them all as allocated and
2622 * create a lists of "safe" pages to be used later. On systems with high
2623 * memory a list of "safe" highmem pages is created too.
2624 *
2625 * Because it was not known which pages were unsafe when @zero_bm was created,
2626 * make a copy of it and recreate it within safe pages.
2627 */
2628static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm,
2629 struct memory_bitmap *zero_bm)
2630{
2631 unsigned int nr_pages, nr_highmem;
2632 struct memory_bitmap tmp;
2633 struct linked_page *lp;
2634 int error;
2635
2636 /* If there is no highmem, the buffer will not be necessary */
2637 free_image_page(buffer, PG_UNSAFE_CLEAR);
2638 buffer = NULL;
2639
2640 nr_highmem = count_highmem_image_pages(bm);
2641 mark_unsafe_pages(bm);
2642
2643 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2644 if (error)
2645 goto Free;
2646
2647 duplicate_memory_bitmap(new_bm, bm);
2648 memory_bm_free(bm, PG_UNSAFE_KEEP);
2649
2650 /* Make a copy of zero_bm so it can be created in safe pages */
2651 error = memory_bm_create(&tmp, GFP_ATOMIC, PG_SAFE);
2652 if (error)
2653 goto Free;
2654
2655 duplicate_memory_bitmap(&tmp, zero_bm);
2656 memory_bm_free(zero_bm, PG_UNSAFE_KEEP);
2657
2658 /* Recreate zero_bm in safe pages */
2659 error = memory_bm_create(zero_bm, GFP_ATOMIC, PG_SAFE);
2660 if (error)
2661 goto Free;
2662
2663 duplicate_memory_bitmap(zero_bm, &tmp);
2664 memory_bm_free(&tmp, PG_UNSAFE_CLEAR);
2665 /* At this point zero_bm is in safe pages and it can be used for restoring. */
2666
2667 if (nr_highmem > 0) {
2668 error = prepare_highmem_image(bm, &nr_highmem);
2669 if (error)
2670 goto Free;
2671 }
2672 /*
2673 * Reserve some safe pages for potential later use.
2674 *
2675 * NOTE: This way we make sure there will be enough safe pages for the
2676 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2677 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2678 *
2679 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2680 */
2681 nr_pages = (nr_zero_pages + nr_copy_pages) - nr_highmem - allocated_unsafe_pages;
2682 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2683 while (nr_pages > 0) {
2684 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2685 if (!lp) {
2686 error = -ENOMEM;
2687 goto Free;
2688 }
2689 lp->next = safe_pages_list;
2690 safe_pages_list = lp;
2691 nr_pages--;
2692 }
2693 /* Preallocate memory for the image */
2694 nr_pages = (nr_zero_pages + nr_copy_pages) - nr_highmem - allocated_unsafe_pages;
2695 while (nr_pages > 0) {
2696 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2697 if (!lp) {
2698 error = -ENOMEM;
2699 goto Free;
2700 }
2701 if (!swsusp_page_is_free(virt_to_page(lp))) {
2702 /* The page is "safe", add it to the list */
2703 lp->next = safe_pages_list;
2704 safe_pages_list = lp;
2705 }
2706 /* Mark the page as allocated */
2707 swsusp_set_page_forbidden(virt_to_page(lp));
2708 swsusp_set_page_free(virt_to_page(lp));
2709 nr_pages--;
2710 }
2711 return 0;
2712
2713 Free:
2714 swsusp_free();
2715 return error;
2716}
2717
2718/**
2719 * get_buffer - Get the address to store the next image data page.
2720 *
2721 * Get the address that snapshot_write_next() should return to its caller to
2722 * write to.
2723 */
2724static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2725{
2726 struct pbe *pbe;
2727 struct page *page;
2728 unsigned long pfn = memory_bm_next_pfn(bm);
2729
2730 if (pfn == BM_END_OF_MAP)
2731 return ERR_PTR(-EFAULT);
2732
2733 page = pfn_to_page(pfn);
2734 if (PageHighMem(page))
2735 return get_highmem_page_buffer(page, ca);
2736
2737 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2738 /*
2739 * We have allocated the "original" page frame and we can
2740 * use it directly to store the loaded page.
2741 */
2742 return page_address(page);
2743
2744 /*
2745 * The "original" page frame has not been allocated and we have to
2746 * use a "safe" page frame to store the loaded page.
2747 */
2748 pbe = chain_alloc(ca, sizeof(struct pbe));
2749 if (!pbe) {
2750 swsusp_free();
2751 return ERR_PTR(-ENOMEM);
2752 }
2753 pbe->orig_address = page_address(page);
2754 pbe->address = __get_safe_page(ca->gfp_mask);
2755 if (!pbe->address)
2756 return ERR_PTR(-ENOMEM);
2757 pbe->next = restore_pblist;
2758 restore_pblist = pbe;
2759 return pbe->address;
2760}
2761
2762/**
2763 * snapshot_write_next - Get the address to store the next image page.
2764 * @handle: Snapshot handle structure to guide the writing.
2765 *
2766 * On the first call, @handle should point to a zeroed snapshot_handle
2767 * structure. The structure gets populated then and a pointer to it should be
2768 * passed to this function every next time.
2769 *
2770 * On success, the function returns a positive number. Then, the caller
2771 * is allowed to write up to the returned number of bytes to the memory
2772 * location computed by the data_of() macro.
2773 *
2774 * The function returns 0 to indicate the "end of file" condition. Negative
2775 * numbers are returned on errors, in which cases the structure pointed to by
2776 * @handle is not updated and should not be used any more.
2777 */
2778int snapshot_write_next(struct snapshot_handle *handle)
2779{
2780 static struct chain_allocator ca;
2781 int error;
2782
2783next:
2784 /* Check if we have already loaded the entire image */
2785 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages + nr_zero_pages)
2786 return 0;
2787
2788 if (!handle->cur) {
2789 if (!buffer)
2790 /* This makes the buffer be freed by swsusp_free() */
2791 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2792
2793 if (!buffer)
2794 return -ENOMEM;
2795
2796 handle->buffer = buffer;
2797 } else if (handle->cur == 1) {
2798 error = load_header(buffer);
2799 if (error)
2800 return error;
2801
2802 safe_pages_list = NULL;
2803
2804 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2805 if (error)
2806 return error;
2807
2808 error = memory_bm_create(&zero_bm, GFP_ATOMIC, PG_ANY);
2809 if (error)
2810 return error;
2811
2812 nr_zero_pages = 0;
2813
2814 hibernate_restore_protection_begin();
2815 } else if (handle->cur <= nr_meta_pages + 1) {
2816 error = unpack_orig_pfns(buffer, ©_bm, &zero_bm);
2817 if (error)
2818 return error;
2819
2820 if (handle->cur == nr_meta_pages + 1) {
2821 error = prepare_image(&orig_bm, ©_bm, &zero_bm);
2822 if (error)
2823 return error;
2824
2825 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2826 memory_bm_position_reset(&orig_bm);
2827 memory_bm_position_reset(&zero_bm);
2828 restore_pblist = NULL;
2829 handle->buffer = get_buffer(&orig_bm, &ca);
2830 if (IS_ERR(handle->buffer))
2831 return PTR_ERR(handle->buffer);
2832 }
2833 } else {
2834 copy_last_highmem_page();
2835 hibernate_restore_protect_page(handle->buffer);
2836 handle->buffer = get_buffer(&orig_bm, &ca);
2837 if (IS_ERR(handle->buffer))
2838 return PTR_ERR(handle->buffer);
2839 }
2840 handle->sync_read = (handle->buffer == buffer);
2841 handle->cur++;
2842
2843 /* Zero pages were not included in the image, memset it and move on. */
2844 if (handle->cur > nr_meta_pages + 1 &&
2845 memory_bm_test_bit(&zero_bm, memory_bm_get_current(&orig_bm))) {
2846 memset(handle->buffer, 0, PAGE_SIZE);
2847 goto next;
2848 }
2849
2850 return PAGE_SIZE;
2851}
2852
2853/**
2854 * snapshot_write_finalize - Complete the loading of a hibernation image.
2855 *
2856 * Must be called after the last call to snapshot_write_next() in case the last
2857 * page in the image happens to be a highmem page and its contents should be
2858 * stored in highmem. Additionally, it recycles bitmap memory that's not
2859 * necessary any more.
2860 */
2861void snapshot_write_finalize(struct snapshot_handle *handle)
2862{
2863 copy_last_highmem_page();
2864 hibernate_restore_protect_page(handle->buffer);
2865 /* Do that only if we have loaded the image entirely */
2866 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages + nr_zero_pages) {
2867 memory_bm_recycle(&orig_bm);
2868 free_highmem_data();
2869 }
2870}
2871
2872int snapshot_image_loaded(struct snapshot_handle *handle)
2873{
2874 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2875 handle->cur <= nr_meta_pages + nr_copy_pages + nr_zero_pages);
2876}
2877
2878#ifdef CONFIG_HIGHMEM
2879/* Assumes that @buf is ready and points to a "safe" page */
2880static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2881 void *buf)
2882{
2883 void *kaddr1, *kaddr2;
2884
2885 kaddr1 = kmap_atomic(p1);
2886 kaddr2 = kmap_atomic(p2);
2887 copy_page(buf, kaddr1);
2888 copy_page(kaddr1, kaddr2);
2889 copy_page(kaddr2, buf);
2890 kunmap_atomic(kaddr2);
2891 kunmap_atomic(kaddr1);
2892}
2893
2894/**
2895 * restore_highmem - Put highmem image pages into their original locations.
2896 *
2897 * For each highmem page that was in use before hibernation and is included in
2898 * the image, and also has been allocated by the "restore" kernel, swap its
2899 * current contents with the previous (ie. "before hibernation") ones.
2900 *
2901 * If the restore eventually fails, we can call this function once again and
2902 * restore the highmem state as seen by the restore kernel.
2903 */
2904int restore_highmem(void)
2905{
2906 struct highmem_pbe *pbe = highmem_pblist;
2907 void *buf;
2908
2909 if (!pbe)
2910 return 0;
2911
2912 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2913 if (!buf)
2914 return -ENOMEM;
2915
2916 while (pbe) {
2917 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2918 pbe = pbe->next;
2919 }
2920 free_image_page(buf, PG_UNSAFE_CLEAR);
2921 return 0;
2922}
2923#endif /* CONFIG_HIGHMEM */