Loading...
1/*
2 * linux/kernel/power/snapshot.c
3 *
4 * This file provides system snapshot/restore functionality for swsusp.
5 *
6 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
7 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
8 *
9 * This file is released under the GPLv2.
10 *
11 */
12
13#define pr_fmt(fmt) "PM: " fmt
14
15#include <linux/version.h>
16#include <linux/module.h>
17#include <linux/mm.h>
18#include <linux/suspend.h>
19#include <linux/delay.h>
20#include <linux/bitops.h>
21#include <linux/spinlock.h>
22#include <linux/kernel.h>
23#include <linux/pm.h>
24#include <linux/device.h>
25#include <linux/init.h>
26#include <linux/bootmem.h>
27#include <linux/nmi.h>
28#include <linux/syscalls.h>
29#include <linux/console.h>
30#include <linux/highmem.h>
31#include <linux/list.h>
32#include <linux/slab.h>
33#include <linux/compiler.h>
34#include <linux/ktime.h>
35#include <linux/set_memory.h>
36
37#include <linux/uaccess.h>
38#include <asm/mmu_context.h>
39#include <asm/pgtable.h>
40#include <asm/tlbflush.h>
41#include <asm/io.h>
42
43#include "power.h"
44
45#if defined(CONFIG_STRICT_KERNEL_RWX) && defined(CONFIG_ARCH_HAS_SET_MEMORY)
46static bool hibernate_restore_protection;
47static bool hibernate_restore_protection_active;
48
49void enable_restore_image_protection(void)
50{
51 hibernate_restore_protection = true;
52}
53
54static inline void hibernate_restore_protection_begin(void)
55{
56 hibernate_restore_protection_active = hibernate_restore_protection;
57}
58
59static inline void hibernate_restore_protection_end(void)
60{
61 hibernate_restore_protection_active = false;
62}
63
64static inline void hibernate_restore_protect_page(void *page_address)
65{
66 if (hibernate_restore_protection_active)
67 set_memory_ro((unsigned long)page_address, 1);
68}
69
70static inline void hibernate_restore_unprotect_page(void *page_address)
71{
72 if (hibernate_restore_protection_active)
73 set_memory_rw((unsigned long)page_address, 1);
74}
75#else
76static inline void hibernate_restore_protection_begin(void) {}
77static inline void hibernate_restore_protection_end(void) {}
78static inline void hibernate_restore_protect_page(void *page_address) {}
79static inline void hibernate_restore_unprotect_page(void *page_address) {}
80#endif /* CONFIG_STRICT_KERNEL_RWX && CONFIG_ARCH_HAS_SET_MEMORY */
81
82static int swsusp_page_is_free(struct page *);
83static void swsusp_set_page_forbidden(struct page *);
84static void swsusp_unset_page_forbidden(struct page *);
85
86/*
87 * Number of bytes to reserve for memory allocations made by device drivers
88 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
89 * cause image creation to fail (tunable via /sys/power/reserved_size).
90 */
91unsigned long reserved_size;
92
93void __init hibernate_reserved_size_init(void)
94{
95 reserved_size = SPARE_PAGES * PAGE_SIZE;
96}
97
98/*
99 * Preferred image size in bytes (tunable via /sys/power/image_size).
100 * When it is set to N, swsusp will do its best to ensure the image
101 * size will not exceed N bytes, but if that is impossible, it will
102 * try to create the smallest image possible.
103 */
104unsigned long image_size;
105
106void __init hibernate_image_size_init(void)
107{
108 image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE;
109}
110
111/*
112 * List of PBEs needed for restoring the pages that were allocated before
113 * the suspend and included in the suspend image, but have also been
114 * allocated by the "resume" kernel, so their contents cannot be written
115 * directly to their "original" page frames.
116 */
117struct pbe *restore_pblist;
118
119/* struct linked_page is used to build chains of pages */
120
121#define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
122
123struct linked_page {
124 struct linked_page *next;
125 char data[LINKED_PAGE_DATA_SIZE];
126} __packed;
127
128/*
129 * List of "safe" pages (ie. pages that were not used by the image kernel
130 * before hibernation) that may be used as temporary storage for image kernel
131 * memory contents.
132 */
133static struct linked_page *safe_pages_list;
134
135/* Pointer to an auxiliary buffer (1 page) */
136static void *buffer;
137
138#define PG_ANY 0
139#define PG_SAFE 1
140#define PG_UNSAFE_CLEAR 1
141#define PG_UNSAFE_KEEP 0
142
143static unsigned int allocated_unsafe_pages;
144
145/**
146 * get_image_page - Allocate a page for a hibernation image.
147 * @gfp_mask: GFP mask for the allocation.
148 * @safe_needed: Get pages that were not used before hibernation (restore only)
149 *
150 * During image restoration, for storing the PBE list and the image data, we can
151 * only use memory pages that do not conflict with the pages used before
152 * hibernation. The "unsafe" pages have PageNosaveFree set and we count them
153 * using allocated_unsafe_pages.
154 *
155 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
156 * swsusp_free() can release it.
157 */
158static void *get_image_page(gfp_t gfp_mask, int safe_needed)
159{
160 void *res;
161
162 res = (void *)get_zeroed_page(gfp_mask);
163 if (safe_needed)
164 while (res && swsusp_page_is_free(virt_to_page(res))) {
165 /* The page is unsafe, mark it for swsusp_free() */
166 swsusp_set_page_forbidden(virt_to_page(res));
167 allocated_unsafe_pages++;
168 res = (void *)get_zeroed_page(gfp_mask);
169 }
170 if (res) {
171 swsusp_set_page_forbidden(virt_to_page(res));
172 swsusp_set_page_free(virt_to_page(res));
173 }
174 return res;
175}
176
177static void *__get_safe_page(gfp_t gfp_mask)
178{
179 if (safe_pages_list) {
180 void *ret = safe_pages_list;
181
182 safe_pages_list = safe_pages_list->next;
183 memset(ret, 0, PAGE_SIZE);
184 return ret;
185 }
186 return get_image_page(gfp_mask, PG_SAFE);
187}
188
189unsigned long get_safe_page(gfp_t gfp_mask)
190{
191 return (unsigned long)__get_safe_page(gfp_mask);
192}
193
194static struct page *alloc_image_page(gfp_t gfp_mask)
195{
196 struct page *page;
197
198 page = alloc_page(gfp_mask);
199 if (page) {
200 swsusp_set_page_forbidden(page);
201 swsusp_set_page_free(page);
202 }
203 return page;
204}
205
206static void recycle_safe_page(void *page_address)
207{
208 struct linked_page *lp = page_address;
209
210 lp->next = safe_pages_list;
211 safe_pages_list = lp;
212}
213
214/**
215 * free_image_page - Free a page allocated for hibernation image.
216 * @addr: Address of the page to free.
217 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
218 *
219 * The page to free should have been allocated by get_image_page() (page flags
220 * set by it are affected).
221 */
222static inline void free_image_page(void *addr, int clear_nosave_free)
223{
224 struct page *page;
225
226 BUG_ON(!virt_addr_valid(addr));
227
228 page = virt_to_page(addr);
229
230 swsusp_unset_page_forbidden(page);
231 if (clear_nosave_free)
232 swsusp_unset_page_free(page);
233
234 __free_page(page);
235}
236
237static inline void free_list_of_pages(struct linked_page *list,
238 int clear_page_nosave)
239{
240 while (list) {
241 struct linked_page *lp = list->next;
242
243 free_image_page(list, clear_page_nosave);
244 list = lp;
245 }
246}
247
248/*
249 * struct chain_allocator is used for allocating small objects out of
250 * a linked list of pages called 'the chain'.
251 *
252 * The chain grows each time when there is no room for a new object in
253 * the current page. The allocated objects cannot be freed individually.
254 * It is only possible to free them all at once, by freeing the entire
255 * chain.
256 *
257 * NOTE: The chain allocator may be inefficient if the allocated objects
258 * are not much smaller than PAGE_SIZE.
259 */
260struct chain_allocator {
261 struct linked_page *chain; /* the chain */
262 unsigned int used_space; /* total size of objects allocated out
263 of the current page */
264 gfp_t gfp_mask; /* mask for allocating pages */
265 int safe_needed; /* if set, only "safe" pages are allocated */
266};
267
268static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
269 int safe_needed)
270{
271 ca->chain = NULL;
272 ca->used_space = LINKED_PAGE_DATA_SIZE;
273 ca->gfp_mask = gfp_mask;
274 ca->safe_needed = safe_needed;
275}
276
277static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
278{
279 void *ret;
280
281 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
282 struct linked_page *lp;
283
284 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
285 get_image_page(ca->gfp_mask, PG_ANY);
286 if (!lp)
287 return NULL;
288
289 lp->next = ca->chain;
290 ca->chain = lp;
291 ca->used_space = 0;
292 }
293 ret = ca->chain->data + ca->used_space;
294 ca->used_space += size;
295 return ret;
296}
297
298/**
299 * Data types related to memory bitmaps.
300 *
301 * Memory bitmap is a structure consiting of many linked lists of
302 * objects. The main list's elements are of type struct zone_bitmap
303 * and each of them corresonds to one zone. For each zone bitmap
304 * object there is a list of objects of type struct bm_block that
305 * represent each blocks of bitmap in which information is stored.
306 *
307 * struct memory_bitmap contains a pointer to the main list of zone
308 * bitmap objects, a struct bm_position used for browsing the bitmap,
309 * and a pointer to the list of pages used for allocating all of the
310 * zone bitmap objects and bitmap block objects.
311 *
312 * NOTE: It has to be possible to lay out the bitmap in memory
313 * using only allocations of order 0. Additionally, the bitmap is
314 * designed to work with arbitrary number of zones (this is over the
315 * top for now, but let's avoid making unnecessary assumptions ;-).
316 *
317 * struct zone_bitmap contains a pointer to a list of bitmap block
318 * objects and a pointer to the bitmap block object that has been
319 * most recently used for setting bits. Additionally, it contains the
320 * PFNs that correspond to the start and end of the represented zone.
321 *
322 * struct bm_block contains a pointer to the memory page in which
323 * information is stored (in the form of a block of bitmap)
324 * It also contains the pfns that correspond to the start and end of
325 * the represented memory area.
326 *
327 * The memory bitmap is organized as a radix tree to guarantee fast random
328 * access to the bits. There is one radix tree for each zone (as returned
329 * from create_mem_extents).
330 *
331 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
332 * two linked lists for the nodes of the tree, one for the inner nodes and
333 * one for the leave nodes. The linked leave nodes are used for fast linear
334 * access of the memory bitmap.
335 *
336 * The struct rtree_node represents one node of the radix tree.
337 */
338
339#define BM_END_OF_MAP (~0UL)
340
341#define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
342#define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
343#define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
344
345/*
346 * struct rtree_node is a wrapper struct to link the nodes
347 * of the rtree together for easy linear iteration over
348 * bits and easy freeing
349 */
350struct rtree_node {
351 struct list_head list;
352 unsigned long *data;
353};
354
355/*
356 * struct mem_zone_bm_rtree represents a bitmap used for one
357 * populated memory zone.
358 */
359struct mem_zone_bm_rtree {
360 struct list_head list; /* Link Zones together */
361 struct list_head nodes; /* Radix Tree inner nodes */
362 struct list_head leaves; /* Radix Tree leaves */
363 unsigned long start_pfn; /* Zone start page frame */
364 unsigned long end_pfn; /* Zone end page frame + 1 */
365 struct rtree_node *rtree; /* Radix Tree Root */
366 int levels; /* Number of Radix Tree Levels */
367 unsigned int blocks; /* Number of Bitmap Blocks */
368};
369
370/* strcut bm_position is used for browsing memory bitmaps */
371
372struct bm_position {
373 struct mem_zone_bm_rtree *zone;
374 struct rtree_node *node;
375 unsigned long node_pfn;
376 int node_bit;
377};
378
379struct memory_bitmap {
380 struct list_head zones;
381 struct linked_page *p_list; /* list of pages used to store zone
382 bitmap objects and bitmap block
383 objects */
384 struct bm_position cur; /* most recently used bit position */
385};
386
387/* Functions that operate on memory bitmaps */
388
389#define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
390#if BITS_PER_LONG == 32
391#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
392#else
393#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
394#endif
395#define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
396
397/**
398 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
399 *
400 * This function is used to allocate inner nodes as well as the
401 * leave nodes of the radix tree. It also adds the node to the
402 * corresponding linked list passed in by the *list parameter.
403 */
404static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
405 struct chain_allocator *ca,
406 struct list_head *list)
407{
408 struct rtree_node *node;
409
410 node = chain_alloc(ca, sizeof(struct rtree_node));
411 if (!node)
412 return NULL;
413
414 node->data = get_image_page(gfp_mask, safe_needed);
415 if (!node->data)
416 return NULL;
417
418 list_add_tail(&node->list, list);
419
420 return node;
421}
422
423/**
424 * add_rtree_block - Add a new leave node to the radix tree.
425 *
426 * The leave nodes need to be allocated in order to keep the leaves
427 * linked list in order. This is guaranteed by the zone->blocks
428 * counter.
429 */
430static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
431 int safe_needed, struct chain_allocator *ca)
432{
433 struct rtree_node *node, *block, **dst;
434 unsigned int levels_needed, block_nr;
435 int i;
436
437 block_nr = zone->blocks;
438 levels_needed = 0;
439
440 /* How many levels do we need for this block nr? */
441 while (block_nr) {
442 levels_needed += 1;
443 block_nr >>= BM_RTREE_LEVEL_SHIFT;
444 }
445
446 /* Make sure the rtree has enough levels */
447 for (i = zone->levels; i < levels_needed; i++) {
448 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
449 &zone->nodes);
450 if (!node)
451 return -ENOMEM;
452
453 node->data[0] = (unsigned long)zone->rtree;
454 zone->rtree = node;
455 zone->levels += 1;
456 }
457
458 /* Allocate new block */
459 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
460 if (!block)
461 return -ENOMEM;
462
463 /* Now walk the rtree to insert the block */
464 node = zone->rtree;
465 dst = &zone->rtree;
466 block_nr = zone->blocks;
467 for (i = zone->levels; i > 0; i--) {
468 int index;
469
470 if (!node) {
471 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
472 &zone->nodes);
473 if (!node)
474 return -ENOMEM;
475 *dst = node;
476 }
477
478 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
479 index &= BM_RTREE_LEVEL_MASK;
480 dst = (struct rtree_node **)&((*dst)->data[index]);
481 node = *dst;
482 }
483
484 zone->blocks += 1;
485 *dst = block;
486
487 return 0;
488}
489
490static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
491 int clear_nosave_free);
492
493/**
494 * create_zone_bm_rtree - Create a radix tree for one zone.
495 *
496 * Allocated the mem_zone_bm_rtree structure and initializes it.
497 * This function also allocated and builds the radix tree for the
498 * zone.
499 */
500static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
501 int safe_needed,
502 struct chain_allocator *ca,
503 unsigned long start,
504 unsigned long end)
505{
506 struct mem_zone_bm_rtree *zone;
507 unsigned int i, nr_blocks;
508 unsigned long pages;
509
510 pages = end - start;
511 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
512 if (!zone)
513 return NULL;
514
515 INIT_LIST_HEAD(&zone->nodes);
516 INIT_LIST_HEAD(&zone->leaves);
517 zone->start_pfn = start;
518 zone->end_pfn = end;
519 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
520
521 for (i = 0; i < nr_blocks; i++) {
522 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
523 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
524 return NULL;
525 }
526 }
527
528 return zone;
529}
530
531/**
532 * free_zone_bm_rtree - Free the memory of the radix tree.
533 *
534 * Free all node pages of the radix tree. The mem_zone_bm_rtree
535 * structure itself is not freed here nor are the rtree_node
536 * structs.
537 */
538static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
539 int clear_nosave_free)
540{
541 struct rtree_node *node;
542
543 list_for_each_entry(node, &zone->nodes, list)
544 free_image_page(node->data, clear_nosave_free);
545
546 list_for_each_entry(node, &zone->leaves, list)
547 free_image_page(node->data, clear_nosave_free);
548}
549
550static void memory_bm_position_reset(struct memory_bitmap *bm)
551{
552 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
553 list);
554 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
555 struct rtree_node, list);
556 bm->cur.node_pfn = 0;
557 bm->cur.node_bit = 0;
558}
559
560static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
561
562struct mem_extent {
563 struct list_head hook;
564 unsigned long start;
565 unsigned long end;
566};
567
568/**
569 * free_mem_extents - Free a list of memory extents.
570 * @list: List of extents to free.
571 */
572static void free_mem_extents(struct list_head *list)
573{
574 struct mem_extent *ext, *aux;
575
576 list_for_each_entry_safe(ext, aux, list, hook) {
577 list_del(&ext->hook);
578 kfree(ext);
579 }
580}
581
582/**
583 * create_mem_extents - Create a list of memory extents.
584 * @list: List to put the extents into.
585 * @gfp_mask: Mask to use for memory allocations.
586 *
587 * The extents represent contiguous ranges of PFNs.
588 */
589static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
590{
591 struct zone *zone;
592
593 INIT_LIST_HEAD(list);
594
595 for_each_populated_zone(zone) {
596 unsigned long zone_start, zone_end;
597 struct mem_extent *ext, *cur, *aux;
598
599 zone_start = zone->zone_start_pfn;
600 zone_end = zone_end_pfn(zone);
601
602 list_for_each_entry(ext, list, hook)
603 if (zone_start <= ext->end)
604 break;
605
606 if (&ext->hook == list || zone_end < ext->start) {
607 /* New extent is necessary */
608 struct mem_extent *new_ext;
609
610 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
611 if (!new_ext) {
612 free_mem_extents(list);
613 return -ENOMEM;
614 }
615 new_ext->start = zone_start;
616 new_ext->end = zone_end;
617 list_add_tail(&new_ext->hook, &ext->hook);
618 continue;
619 }
620
621 /* Merge this zone's range of PFNs with the existing one */
622 if (zone_start < ext->start)
623 ext->start = zone_start;
624 if (zone_end > ext->end)
625 ext->end = zone_end;
626
627 /* More merging may be possible */
628 cur = ext;
629 list_for_each_entry_safe_continue(cur, aux, list, hook) {
630 if (zone_end < cur->start)
631 break;
632 if (zone_end < cur->end)
633 ext->end = cur->end;
634 list_del(&cur->hook);
635 kfree(cur);
636 }
637 }
638
639 return 0;
640}
641
642/**
643 * memory_bm_create - Allocate memory for a memory bitmap.
644 */
645static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
646 int safe_needed)
647{
648 struct chain_allocator ca;
649 struct list_head mem_extents;
650 struct mem_extent *ext;
651 int error;
652
653 chain_init(&ca, gfp_mask, safe_needed);
654 INIT_LIST_HEAD(&bm->zones);
655
656 error = create_mem_extents(&mem_extents, gfp_mask);
657 if (error)
658 return error;
659
660 list_for_each_entry(ext, &mem_extents, hook) {
661 struct mem_zone_bm_rtree *zone;
662
663 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
664 ext->start, ext->end);
665 if (!zone) {
666 error = -ENOMEM;
667 goto Error;
668 }
669 list_add_tail(&zone->list, &bm->zones);
670 }
671
672 bm->p_list = ca.chain;
673 memory_bm_position_reset(bm);
674 Exit:
675 free_mem_extents(&mem_extents);
676 return error;
677
678 Error:
679 bm->p_list = ca.chain;
680 memory_bm_free(bm, PG_UNSAFE_CLEAR);
681 goto Exit;
682}
683
684/**
685 * memory_bm_free - Free memory occupied by the memory bitmap.
686 * @bm: Memory bitmap.
687 */
688static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
689{
690 struct mem_zone_bm_rtree *zone;
691
692 list_for_each_entry(zone, &bm->zones, list)
693 free_zone_bm_rtree(zone, clear_nosave_free);
694
695 free_list_of_pages(bm->p_list, clear_nosave_free);
696
697 INIT_LIST_HEAD(&bm->zones);
698}
699
700/**
701 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
702 *
703 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
704 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
705 *
706 * Walk the radix tree to find the page containing the bit that represents @pfn
707 * and return the position of the bit in @addr and @bit_nr.
708 */
709static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
710 void **addr, unsigned int *bit_nr)
711{
712 struct mem_zone_bm_rtree *curr, *zone;
713 struct rtree_node *node;
714 int i, block_nr;
715
716 zone = bm->cur.zone;
717
718 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
719 goto zone_found;
720
721 zone = NULL;
722
723 /* Find the right zone */
724 list_for_each_entry(curr, &bm->zones, list) {
725 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
726 zone = curr;
727 break;
728 }
729 }
730
731 if (!zone)
732 return -EFAULT;
733
734zone_found:
735 /*
736 * We have found the zone. Now walk the radix tree to find the leaf node
737 * for our PFN.
738 */
739 node = bm->cur.node;
740 if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
741 goto node_found;
742
743 node = zone->rtree;
744 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
745
746 for (i = zone->levels; i > 0; i--) {
747 int index;
748
749 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
750 index &= BM_RTREE_LEVEL_MASK;
751 BUG_ON(node->data[index] == 0);
752 node = (struct rtree_node *)node->data[index];
753 }
754
755node_found:
756 /* Update last position */
757 bm->cur.zone = zone;
758 bm->cur.node = node;
759 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
760
761 /* Set return values */
762 *addr = node->data;
763 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
764
765 return 0;
766}
767
768static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
769{
770 void *addr;
771 unsigned int bit;
772 int error;
773
774 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
775 BUG_ON(error);
776 set_bit(bit, addr);
777}
778
779static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
780{
781 void *addr;
782 unsigned int bit;
783 int error;
784
785 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
786 if (!error)
787 set_bit(bit, addr);
788
789 return error;
790}
791
792static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
793{
794 void *addr;
795 unsigned int bit;
796 int error;
797
798 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
799 BUG_ON(error);
800 clear_bit(bit, addr);
801}
802
803static void memory_bm_clear_current(struct memory_bitmap *bm)
804{
805 int bit;
806
807 bit = max(bm->cur.node_bit - 1, 0);
808 clear_bit(bit, bm->cur.node->data);
809}
810
811static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
812{
813 void *addr;
814 unsigned int bit;
815 int error;
816
817 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
818 BUG_ON(error);
819 return test_bit(bit, addr);
820}
821
822static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
823{
824 void *addr;
825 unsigned int bit;
826
827 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
828}
829
830/*
831 * rtree_next_node - Jump to the next leaf node.
832 *
833 * Set the position to the beginning of the next node in the
834 * memory bitmap. This is either the next node in the current
835 * zone's radix tree or the first node in the radix tree of the
836 * next zone.
837 *
838 * Return true if there is a next node, false otherwise.
839 */
840static bool rtree_next_node(struct memory_bitmap *bm)
841{
842 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
843 bm->cur.node = list_entry(bm->cur.node->list.next,
844 struct rtree_node, list);
845 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
846 bm->cur.node_bit = 0;
847 touch_softlockup_watchdog();
848 return true;
849 }
850
851 /* No more nodes, goto next zone */
852 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
853 bm->cur.zone = list_entry(bm->cur.zone->list.next,
854 struct mem_zone_bm_rtree, list);
855 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
856 struct rtree_node, list);
857 bm->cur.node_pfn = 0;
858 bm->cur.node_bit = 0;
859 return true;
860 }
861
862 /* No more zones */
863 return false;
864}
865
866/**
867 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
868 * @bm: Memory bitmap.
869 *
870 * Starting from the last returned position this function searches for the next
871 * set bit in @bm and returns the PFN represented by it. If no more bits are
872 * set, BM_END_OF_MAP is returned.
873 *
874 * It is required to run memory_bm_position_reset() before the first call to
875 * this function for the given memory bitmap.
876 */
877static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
878{
879 unsigned long bits, pfn, pages;
880 int bit;
881
882 do {
883 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
884 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
885 bit = find_next_bit(bm->cur.node->data, bits,
886 bm->cur.node_bit);
887 if (bit < bits) {
888 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
889 bm->cur.node_bit = bit + 1;
890 return pfn;
891 }
892 } while (rtree_next_node(bm));
893
894 return BM_END_OF_MAP;
895}
896
897/*
898 * This structure represents a range of page frames the contents of which
899 * should not be saved during hibernation.
900 */
901struct nosave_region {
902 struct list_head list;
903 unsigned long start_pfn;
904 unsigned long end_pfn;
905};
906
907static LIST_HEAD(nosave_regions);
908
909static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
910{
911 struct rtree_node *node;
912
913 list_for_each_entry(node, &zone->nodes, list)
914 recycle_safe_page(node->data);
915
916 list_for_each_entry(node, &zone->leaves, list)
917 recycle_safe_page(node->data);
918}
919
920static void memory_bm_recycle(struct memory_bitmap *bm)
921{
922 struct mem_zone_bm_rtree *zone;
923 struct linked_page *p_list;
924
925 list_for_each_entry(zone, &bm->zones, list)
926 recycle_zone_bm_rtree(zone);
927
928 p_list = bm->p_list;
929 while (p_list) {
930 struct linked_page *lp = p_list;
931
932 p_list = lp->next;
933 recycle_safe_page(lp);
934 }
935}
936
937/**
938 * register_nosave_region - Register a region of unsaveable memory.
939 *
940 * Register a range of page frames the contents of which should not be saved
941 * during hibernation (to be used in the early initialization code).
942 */
943void __init __register_nosave_region(unsigned long start_pfn,
944 unsigned long end_pfn, int use_kmalloc)
945{
946 struct nosave_region *region;
947
948 if (start_pfn >= end_pfn)
949 return;
950
951 if (!list_empty(&nosave_regions)) {
952 /* Try to extend the previous region (they should be sorted) */
953 region = list_entry(nosave_regions.prev,
954 struct nosave_region, list);
955 if (region->end_pfn == start_pfn) {
956 region->end_pfn = end_pfn;
957 goto Report;
958 }
959 }
960 if (use_kmalloc) {
961 /* During init, this shouldn't fail */
962 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
963 BUG_ON(!region);
964 } else {
965 /* This allocation cannot fail */
966 region = memblock_virt_alloc(sizeof(struct nosave_region), 0);
967 }
968 region->start_pfn = start_pfn;
969 region->end_pfn = end_pfn;
970 list_add_tail(®ion->list, &nosave_regions);
971 Report:
972 pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
973 (unsigned long long) start_pfn << PAGE_SHIFT,
974 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
975}
976
977/*
978 * Set bits in this map correspond to the page frames the contents of which
979 * should not be saved during the suspend.
980 */
981static struct memory_bitmap *forbidden_pages_map;
982
983/* Set bits in this map correspond to free page frames. */
984static struct memory_bitmap *free_pages_map;
985
986/*
987 * Each page frame allocated for creating the image is marked by setting the
988 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
989 */
990
991void swsusp_set_page_free(struct page *page)
992{
993 if (free_pages_map)
994 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
995}
996
997static int swsusp_page_is_free(struct page *page)
998{
999 return free_pages_map ?
1000 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1001}
1002
1003void swsusp_unset_page_free(struct page *page)
1004{
1005 if (free_pages_map)
1006 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1007}
1008
1009static void swsusp_set_page_forbidden(struct page *page)
1010{
1011 if (forbidden_pages_map)
1012 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1013}
1014
1015int swsusp_page_is_forbidden(struct page *page)
1016{
1017 return forbidden_pages_map ?
1018 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1019}
1020
1021static void swsusp_unset_page_forbidden(struct page *page)
1022{
1023 if (forbidden_pages_map)
1024 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1025}
1026
1027/**
1028 * mark_nosave_pages - Mark pages that should not be saved.
1029 * @bm: Memory bitmap.
1030 *
1031 * Set the bits in @bm that correspond to the page frames the contents of which
1032 * should not be saved.
1033 */
1034static void mark_nosave_pages(struct memory_bitmap *bm)
1035{
1036 struct nosave_region *region;
1037
1038 if (list_empty(&nosave_regions))
1039 return;
1040
1041 list_for_each_entry(region, &nosave_regions, list) {
1042 unsigned long pfn;
1043
1044 pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1045 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1046 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1047 - 1);
1048
1049 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1050 if (pfn_valid(pfn)) {
1051 /*
1052 * It is safe to ignore the result of
1053 * mem_bm_set_bit_check() here, since we won't
1054 * touch the PFNs for which the error is
1055 * returned anyway.
1056 */
1057 mem_bm_set_bit_check(bm, pfn);
1058 }
1059 }
1060}
1061
1062/**
1063 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1064 *
1065 * Create bitmaps needed for marking page frames that should not be saved and
1066 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1067 * only modified if everything goes well, because we don't want the bits to be
1068 * touched before both bitmaps are set up.
1069 */
1070int create_basic_memory_bitmaps(void)
1071{
1072 struct memory_bitmap *bm1, *bm2;
1073 int error = 0;
1074
1075 if (forbidden_pages_map && free_pages_map)
1076 return 0;
1077 else
1078 BUG_ON(forbidden_pages_map || free_pages_map);
1079
1080 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1081 if (!bm1)
1082 return -ENOMEM;
1083
1084 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1085 if (error)
1086 goto Free_first_object;
1087
1088 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1089 if (!bm2)
1090 goto Free_first_bitmap;
1091
1092 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1093 if (error)
1094 goto Free_second_object;
1095
1096 forbidden_pages_map = bm1;
1097 free_pages_map = bm2;
1098 mark_nosave_pages(forbidden_pages_map);
1099
1100 pr_debug("Basic memory bitmaps created\n");
1101
1102 return 0;
1103
1104 Free_second_object:
1105 kfree(bm2);
1106 Free_first_bitmap:
1107 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1108 Free_first_object:
1109 kfree(bm1);
1110 return -ENOMEM;
1111}
1112
1113/**
1114 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1115 *
1116 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1117 * auxiliary pointers are necessary so that the bitmaps themselves are not
1118 * referred to while they are being freed.
1119 */
1120void free_basic_memory_bitmaps(void)
1121{
1122 struct memory_bitmap *bm1, *bm2;
1123
1124 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1125 return;
1126
1127 bm1 = forbidden_pages_map;
1128 bm2 = free_pages_map;
1129 forbidden_pages_map = NULL;
1130 free_pages_map = NULL;
1131 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1132 kfree(bm1);
1133 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1134 kfree(bm2);
1135
1136 pr_debug("Basic memory bitmaps freed\n");
1137}
1138
1139void clear_free_pages(void)
1140{
1141#ifdef CONFIG_PAGE_POISONING_ZERO
1142 struct memory_bitmap *bm = free_pages_map;
1143 unsigned long pfn;
1144
1145 if (WARN_ON(!(free_pages_map)))
1146 return;
1147
1148 memory_bm_position_reset(bm);
1149 pfn = memory_bm_next_pfn(bm);
1150 while (pfn != BM_END_OF_MAP) {
1151 if (pfn_valid(pfn))
1152 clear_highpage(pfn_to_page(pfn));
1153
1154 pfn = memory_bm_next_pfn(bm);
1155 }
1156 memory_bm_position_reset(bm);
1157 pr_info("free pages cleared after restore\n");
1158#endif /* PAGE_POISONING_ZERO */
1159}
1160
1161/**
1162 * snapshot_additional_pages - Estimate the number of extra pages needed.
1163 * @zone: Memory zone to carry out the computation for.
1164 *
1165 * Estimate the number of additional pages needed for setting up a hibernation
1166 * image data structures for @zone (usually, the returned value is greater than
1167 * the exact number).
1168 */
1169unsigned int snapshot_additional_pages(struct zone *zone)
1170{
1171 unsigned int rtree, nodes;
1172
1173 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1174 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1175 LINKED_PAGE_DATA_SIZE);
1176 while (nodes > 1) {
1177 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1178 rtree += nodes;
1179 }
1180
1181 return 2 * rtree;
1182}
1183
1184#ifdef CONFIG_HIGHMEM
1185/**
1186 * count_free_highmem_pages - Compute the total number of free highmem pages.
1187 *
1188 * The returned number is system-wide.
1189 */
1190static unsigned int count_free_highmem_pages(void)
1191{
1192 struct zone *zone;
1193 unsigned int cnt = 0;
1194
1195 for_each_populated_zone(zone)
1196 if (is_highmem(zone))
1197 cnt += zone_page_state(zone, NR_FREE_PAGES);
1198
1199 return cnt;
1200}
1201
1202/**
1203 * saveable_highmem_page - Check if a highmem page is saveable.
1204 *
1205 * Determine whether a highmem page should be included in a hibernation image.
1206 *
1207 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1208 * and it isn't part of a free chunk of pages.
1209 */
1210static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1211{
1212 struct page *page;
1213
1214 if (!pfn_valid(pfn))
1215 return NULL;
1216
1217 page = pfn_to_page(pfn);
1218 if (page_zone(page) != zone)
1219 return NULL;
1220
1221 BUG_ON(!PageHighMem(page));
1222
1223 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) ||
1224 PageReserved(page))
1225 return NULL;
1226
1227 if (page_is_guard(page))
1228 return NULL;
1229
1230 return page;
1231}
1232
1233/**
1234 * count_highmem_pages - Compute the total number of saveable highmem pages.
1235 */
1236static unsigned int count_highmem_pages(void)
1237{
1238 struct zone *zone;
1239 unsigned int n = 0;
1240
1241 for_each_populated_zone(zone) {
1242 unsigned long pfn, max_zone_pfn;
1243
1244 if (!is_highmem(zone))
1245 continue;
1246
1247 mark_free_pages(zone);
1248 max_zone_pfn = zone_end_pfn(zone);
1249 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1250 if (saveable_highmem_page(zone, pfn))
1251 n++;
1252 }
1253 return n;
1254}
1255#else
1256static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1257{
1258 return NULL;
1259}
1260#endif /* CONFIG_HIGHMEM */
1261
1262/**
1263 * saveable_page - Check if the given page is saveable.
1264 *
1265 * Determine whether a non-highmem page should be included in a hibernation
1266 * image.
1267 *
1268 * We should save the page if it isn't Nosave, and is not in the range
1269 * of pages statically defined as 'unsaveable', and it isn't part of
1270 * a free chunk of pages.
1271 */
1272static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1273{
1274 struct page *page;
1275
1276 if (!pfn_valid(pfn))
1277 return NULL;
1278
1279 page = pfn_to_page(pfn);
1280 if (page_zone(page) != zone)
1281 return NULL;
1282
1283 BUG_ON(PageHighMem(page));
1284
1285 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1286 return NULL;
1287
1288 if (PageReserved(page)
1289 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1290 return NULL;
1291
1292 if (page_is_guard(page))
1293 return NULL;
1294
1295 return page;
1296}
1297
1298/**
1299 * count_data_pages - Compute the total number of saveable non-highmem pages.
1300 */
1301static unsigned int count_data_pages(void)
1302{
1303 struct zone *zone;
1304 unsigned long pfn, max_zone_pfn;
1305 unsigned int n = 0;
1306
1307 for_each_populated_zone(zone) {
1308 if (is_highmem(zone))
1309 continue;
1310
1311 mark_free_pages(zone);
1312 max_zone_pfn = zone_end_pfn(zone);
1313 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1314 if (saveable_page(zone, pfn))
1315 n++;
1316 }
1317 return n;
1318}
1319
1320/*
1321 * This is needed, because copy_page and memcpy are not usable for copying
1322 * task structs.
1323 */
1324static inline void do_copy_page(long *dst, long *src)
1325{
1326 int n;
1327
1328 for (n = PAGE_SIZE / sizeof(long); n; n--)
1329 *dst++ = *src++;
1330}
1331
1332/**
1333 * safe_copy_page - Copy a page in a safe way.
1334 *
1335 * Check if the page we are going to copy is marked as present in the kernel
1336 * page tables (this always is the case if CONFIG_DEBUG_PAGEALLOC is not set
1337 * and in that case kernel_page_present() always returns 'true').
1338 */
1339static void safe_copy_page(void *dst, struct page *s_page)
1340{
1341 if (kernel_page_present(s_page)) {
1342 do_copy_page(dst, page_address(s_page));
1343 } else {
1344 kernel_map_pages(s_page, 1, 1);
1345 do_copy_page(dst, page_address(s_page));
1346 kernel_map_pages(s_page, 1, 0);
1347 }
1348}
1349
1350#ifdef CONFIG_HIGHMEM
1351static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1352{
1353 return is_highmem(zone) ?
1354 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1355}
1356
1357static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1358{
1359 struct page *s_page, *d_page;
1360 void *src, *dst;
1361
1362 s_page = pfn_to_page(src_pfn);
1363 d_page = pfn_to_page(dst_pfn);
1364 if (PageHighMem(s_page)) {
1365 src = kmap_atomic(s_page);
1366 dst = kmap_atomic(d_page);
1367 do_copy_page(dst, src);
1368 kunmap_atomic(dst);
1369 kunmap_atomic(src);
1370 } else {
1371 if (PageHighMem(d_page)) {
1372 /*
1373 * The page pointed to by src may contain some kernel
1374 * data modified by kmap_atomic()
1375 */
1376 safe_copy_page(buffer, s_page);
1377 dst = kmap_atomic(d_page);
1378 copy_page(dst, buffer);
1379 kunmap_atomic(dst);
1380 } else {
1381 safe_copy_page(page_address(d_page), s_page);
1382 }
1383 }
1384}
1385#else
1386#define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1387
1388static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1389{
1390 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1391 pfn_to_page(src_pfn));
1392}
1393#endif /* CONFIG_HIGHMEM */
1394
1395static void copy_data_pages(struct memory_bitmap *copy_bm,
1396 struct memory_bitmap *orig_bm)
1397{
1398 struct zone *zone;
1399 unsigned long pfn;
1400
1401 for_each_populated_zone(zone) {
1402 unsigned long max_zone_pfn;
1403
1404 mark_free_pages(zone);
1405 max_zone_pfn = zone_end_pfn(zone);
1406 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1407 if (page_is_saveable(zone, pfn))
1408 memory_bm_set_bit(orig_bm, pfn);
1409 }
1410 memory_bm_position_reset(orig_bm);
1411 memory_bm_position_reset(copy_bm);
1412 for(;;) {
1413 pfn = memory_bm_next_pfn(orig_bm);
1414 if (unlikely(pfn == BM_END_OF_MAP))
1415 break;
1416 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1417 }
1418}
1419
1420/* Total number of image pages */
1421static unsigned int nr_copy_pages;
1422/* Number of pages needed for saving the original pfns of the image pages */
1423static unsigned int nr_meta_pages;
1424/*
1425 * Numbers of normal and highmem page frames allocated for hibernation image
1426 * before suspending devices.
1427 */
1428static unsigned int alloc_normal, alloc_highmem;
1429/*
1430 * Memory bitmap used for marking saveable pages (during hibernation) or
1431 * hibernation image pages (during restore)
1432 */
1433static struct memory_bitmap orig_bm;
1434/*
1435 * Memory bitmap used during hibernation for marking allocated page frames that
1436 * will contain copies of saveable pages. During restore it is initially used
1437 * for marking hibernation image pages, but then the set bits from it are
1438 * duplicated in @orig_bm and it is released. On highmem systems it is next
1439 * used for marking "safe" highmem pages, but it has to be reinitialized for
1440 * this purpose.
1441 */
1442static struct memory_bitmap copy_bm;
1443
1444/**
1445 * swsusp_free - Free pages allocated for hibernation image.
1446 *
1447 * Image pages are alocated before snapshot creation, so they need to be
1448 * released after resume.
1449 */
1450void swsusp_free(void)
1451{
1452 unsigned long fb_pfn, fr_pfn;
1453
1454 if (!forbidden_pages_map || !free_pages_map)
1455 goto out;
1456
1457 memory_bm_position_reset(forbidden_pages_map);
1458 memory_bm_position_reset(free_pages_map);
1459
1460loop:
1461 fr_pfn = memory_bm_next_pfn(free_pages_map);
1462 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1463
1464 /*
1465 * Find the next bit set in both bitmaps. This is guaranteed to
1466 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1467 */
1468 do {
1469 if (fb_pfn < fr_pfn)
1470 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1471 if (fr_pfn < fb_pfn)
1472 fr_pfn = memory_bm_next_pfn(free_pages_map);
1473 } while (fb_pfn != fr_pfn);
1474
1475 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1476 struct page *page = pfn_to_page(fr_pfn);
1477
1478 memory_bm_clear_current(forbidden_pages_map);
1479 memory_bm_clear_current(free_pages_map);
1480 hibernate_restore_unprotect_page(page_address(page));
1481 __free_page(page);
1482 goto loop;
1483 }
1484
1485out:
1486 nr_copy_pages = 0;
1487 nr_meta_pages = 0;
1488 restore_pblist = NULL;
1489 buffer = NULL;
1490 alloc_normal = 0;
1491 alloc_highmem = 0;
1492 hibernate_restore_protection_end();
1493}
1494
1495/* Helper functions used for the shrinking of memory. */
1496
1497#define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1498
1499/**
1500 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1501 * @nr_pages: Number of page frames to allocate.
1502 * @mask: GFP flags to use for the allocation.
1503 *
1504 * Return value: Number of page frames actually allocated
1505 */
1506static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1507{
1508 unsigned long nr_alloc = 0;
1509
1510 while (nr_pages > 0) {
1511 struct page *page;
1512
1513 page = alloc_image_page(mask);
1514 if (!page)
1515 break;
1516 memory_bm_set_bit(©_bm, page_to_pfn(page));
1517 if (PageHighMem(page))
1518 alloc_highmem++;
1519 else
1520 alloc_normal++;
1521 nr_pages--;
1522 nr_alloc++;
1523 }
1524
1525 return nr_alloc;
1526}
1527
1528static unsigned long preallocate_image_memory(unsigned long nr_pages,
1529 unsigned long avail_normal)
1530{
1531 unsigned long alloc;
1532
1533 if (avail_normal <= alloc_normal)
1534 return 0;
1535
1536 alloc = avail_normal - alloc_normal;
1537 if (nr_pages < alloc)
1538 alloc = nr_pages;
1539
1540 return preallocate_image_pages(alloc, GFP_IMAGE);
1541}
1542
1543#ifdef CONFIG_HIGHMEM
1544static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1545{
1546 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1547}
1548
1549/**
1550 * __fraction - Compute (an approximation of) x * (multiplier / base).
1551 */
1552static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1553{
1554 x *= multiplier;
1555 do_div(x, base);
1556 return (unsigned long)x;
1557}
1558
1559static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1560 unsigned long highmem,
1561 unsigned long total)
1562{
1563 unsigned long alloc = __fraction(nr_pages, highmem, total);
1564
1565 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1566}
1567#else /* CONFIG_HIGHMEM */
1568static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1569{
1570 return 0;
1571}
1572
1573static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1574 unsigned long highmem,
1575 unsigned long total)
1576{
1577 return 0;
1578}
1579#endif /* CONFIG_HIGHMEM */
1580
1581/**
1582 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1583 */
1584static unsigned long free_unnecessary_pages(void)
1585{
1586 unsigned long save, to_free_normal, to_free_highmem, free;
1587
1588 save = count_data_pages();
1589 if (alloc_normal >= save) {
1590 to_free_normal = alloc_normal - save;
1591 save = 0;
1592 } else {
1593 to_free_normal = 0;
1594 save -= alloc_normal;
1595 }
1596 save += count_highmem_pages();
1597 if (alloc_highmem >= save) {
1598 to_free_highmem = alloc_highmem - save;
1599 } else {
1600 to_free_highmem = 0;
1601 save -= alloc_highmem;
1602 if (to_free_normal > save)
1603 to_free_normal -= save;
1604 else
1605 to_free_normal = 0;
1606 }
1607 free = to_free_normal + to_free_highmem;
1608
1609 memory_bm_position_reset(©_bm);
1610
1611 while (to_free_normal > 0 || to_free_highmem > 0) {
1612 unsigned long pfn = memory_bm_next_pfn(©_bm);
1613 struct page *page = pfn_to_page(pfn);
1614
1615 if (PageHighMem(page)) {
1616 if (!to_free_highmem)
1617 continue;
1618 to_free_highmem--;
1619 alloc_highmem--;
1620 } else {
1621 if (!to_free_normal)
1622 continue;
1623 to_free_normal--;
1624 alloc_normal--;
1625 }
1626 memory_bm_clear_bit(©_bm, pfn);
1627 swsusp_unset_page_forbidden(page);
1628 swsusp_unset_page_free(page);
1629 __free_page(page);
1630 }
1631
1632 return free;
1633}
1634
1635/**
1636 * minimum_image_size - Estimate the minimum acceptable size of an image.
1637 * @saveable: Number of saveable pages in the system.
1638 *
1639 * We want to avoid attempting to free too much memory too hard, so estimate the
1640 * minimum acceptable size of a hibernation image to use as the lower limit for
1641 * preallocating memory.
1642 *
1643 * We assume that the minimum image size should be proportional to
1644 *
1645 * [number of saveable pages] - [number of pages that can be freed in theory]
1646 *
1647 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1648 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1649 */
1650static unsigned long minimum_image_size(unsigned long saveable)
1651{
1652 unsigned long size;
1653
1654 size = global_node_page_state(NR_SLAB_RECLAIMABLE)
1655 + global_node_page_state(NR_ACTIVE_ANON)
1656 + global_node_page_state(NR_INACTIVE_ANON)
1657 + global_node_page_state(NR_ACTIVE_FILE)
1658 + global_node_page_state(NR_INACTIVE_FILE);
1659
1660 return saveable <= size ? 0 : saveable - size;
1661}
1662
1663/**
1664 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1665 *
1666 * To create a hibernation image it is necessary to make a copy of every page
1667 * frame in use. We also need a number of page frames to be free during
1668 * hibernation for allocations made while saving the image and for device
1669 * drivers, in case they need to allocate memory from their hibernation
1670 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1671 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1672 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1673 * total number of available page frames and allocate at least
1674 *
1675 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1676 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1677 *
1678 * of them, which corresponds to the maximum size of a hibernation image.
1679 *
1680 * If image_size is set below the number following from the above formula,
1681 * the preallocation of memory is continued until the total number of saveable
1682 * pages in the system is below the requested image size or the minimum
1683 * acceptable image size returned by minimum_image_size(), whichever is greater.
1684 */
1685int hibernate_preallocate_memory(void)
1686{
1687 struct zone *zone;
1688 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1689 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1690 ktime_t start, stop;
1691 int error;
1692
1693 pr_info("Preallocating image memory... ");
1694 start = ktime_get();
1695
1696 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1697 if (error)
1698 goto err_out;
1699
1700 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1701 if (error)
1702 goto err_out;
1703
1704 alloc_normal = 0;
1705 alloc_highmem = 0;
1706
1707 /* Count the number of saveable data pages. */
1708 save_highmem = count_highmem_pages();
1709 saveable = count_data_pages();
1710
1711 /*
1712 * Compute the total number of page frames we can use (count) and the
1713 * number of pages needed for image metadata (size).
1714 */
1715 count = saveable;
1716 saveable += save_highmem;
1717 highmem = save_highmem;
1718 size = 0;
1719 for_each_populated_zone(zone) {
1720 size += snapshot_additional_pages(zone);
1721 if (is_highmem(zone))
1722 highmem += zone_page_state(zone, NR_FREE_PAGES);
1723 else
1724 count += zone_page_state(zone, NR_FREE_PAGES);
1725 }
1726 avail_normal = count;
1727 count += highmem;
1728 count -= totalreserve_pages;
1729
1730 /* Add number of pages required for page keys (s390 only). */
1731 size += page_key_additional_pages(saveable);
1732
1733 /* Compute the maximum number of saveable pages to leave in memory. */
1734 max_size = (count - (size + PAGES_FOR_IO)) / 2
1735 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1736 /* Compute the desired number of image pages specified by image_size. */
1737 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1738 if (size > max_size)
1739 size = max_size;
1740 /*
1741 * If the desired number of image pages is at least as large as the
1742 * current number of saveable pages in memory, allocate page frames for
1743 * the image and we're done.
1744 */
1745 if (size >= saveable) {
1746 pages = preallocate_image_highmem(save_highmem);
1747 pages += preallocate_image_memory(saveable - pages, avail_normal);
1748 goto out;
1749 }
1750
1751 /* Estimate the minimum size of the image. */
1752 pages = minimum_image_size(saveable);
1753 /*
1754 * To avoid excessive pressure on the normal zone, leave room in it to
1755 * accommodate an image of the minimum size (unless it's already too
1756 * small, in which case don't preallocate pages from it at all).
1757 */
1758 if (avail_normal > pages)
1759 avail_normal -= pages;
1760 else
1761 avail_normal = 0;
1762 if (size < pages)
1763 size = min_t(unsigned long, pages, max_size);
1764
1765 /*
1766 * Let the memory management subsystem know that we're going to need a
1767 * large number of page frames to allocate and make it free some memory.
1768 * NOTE: If this is not done, performance will be hurt badly in some
1769 * test cases.
1770 */
1771 shrink_all_memory(saveable - size);
1772
1773 /*
1774 * The number of saveable pages in memory was too high, so apply some
1775 * pressure to decrease it. First, make room for the largest possible
1776 * image and fail if that doesn't work. Next, try to decrease the size
1777 * of the image as much as indicated by 'size' using allocations from
1778 * highmem and non-highmem zones separately.
1779 */
1780 pages_highmem = preallocate_image_highmem(highmem / 2);
1781 alloc = count - max_size;
1782 if (alloc > pages_highmem)
1783 alloc -= pages_highmem;
1784 else
1785 alloc = 0;
1786 pages = preallocate_image_memory(alloc, avail_normal);
1787 if (pages < alloc) {
1788 /* We have exhausted non-highmem pages, try highmem. */
1789 alloc -= pages;
1790 pages += pages_highmem;
1791 pages_highmem = preallocate_image_highmem(alloc);
1792 if (pages_highmem < alloc)
1793 goto err_out;
1794 pages += pages_highmem;
1795 /*
1796 * size is the desired number of saveable pages to leave in
1797 * memory, so try to preallocate (all memory - size) pages.
1798 */
1799 alloc = (count - pages) - size;
1800 pages += preallocate_image_highmem(alloc);
1801 } else {
1802 /*
1803 * There are approximately max_size saveable pages at this point
1804 * and we want to reduce this number down to size.
1805 */
1806 alloc = max_size - size;
1807 size = preallocate_highmem_fraction(alloc, highmem, count);
1808 pages_highmem += size;
1809 alloc -= size;
1810 size = preallocate_image_memory(alloc, avail_normal);
1811 pages_highmem += preallocate_image_highmem(alloc - size);
1812 pages += pages_highmem + size;
1813 }
1814
1815 /*
1816 * We only need as many page frames for the image as there are saveable
1817 * pages in memory, but we have allocated more. Release the excessive
1818 * ones now.
1819 */
1820 pages -= free_unnecessary_pages();
1821
1822 out:
1823 stop = ktime_get();
1824 pr_cont("done (allocated %lu pages)\n", pages);
1825 swsusp_show_speed(start, stop, pages, "Allocated");
1826
1827 return 0;
1828
1829 err_out:
1830 pr_cont("\n");
1831 swsusp_free();
1832 return -ENOMEM;
1833}
1834
1835#ifdef CONFIG_HIGHMEM
1836/**
1837 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1838 *
1839 * Compute the number of non-highmem pages that will be necessary for creating
1840 * copies of highmem pages.
1841 */
1842static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1843{
1844 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1845
1846 if (free_highmem >= nr_highmem)
1847 nr_highmem = 0;
1848 else
1849 nr_highmem -= free_highmem;
1850
1851 return nr_highmem;
1852}
1853#else
1854static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1855#endif /* CONFIG_HIGHMEM */
1856
1857/**
1858 * enough_free_mem - Check if there is enough free memory for the image.
1859 */
1860static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1861{
1862 struct zone *zone;
1863 unsigned int free = alloc_normal;
1864
1865 for_each_populated_zone(zone)
1866 if (!is_highmem(zone))
1867 free += zone_page_state(zone, NR_FREE_PAGES);
1868
1869 nr_pages += count_pages_for_highmem(nr_highmem);
1870 pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
1871 nr_pages, PAGES_FOR_IO, free);
1872
1873 return free > nr_pages + PAGES_FOR_IO;
1874}
1875
1876#ifdef CONFIG_HIGHMEM
1877/**
1878 * get_highmem_buffer - Allocate a buffer for highmem pages.
1879 *
1880 * If there are some highmem pages in the hibernation image, we may need a
1881 * buffer to copy them and/or load their data.
1882 */
1883static inline int get_highmem_buffer(int safe_needed)
1884{
1885 buffer = get_image_page(GFP_ATOMIC, safe_needed);
1886 return buffer ? 0 : -ENOMEM;
1887}
1888
1889/**
1890 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1891 *
1892 * Try to allocate as many pages as needed, but if the number of free highmem
1893 * pages is less than that, allocate them all.
1894 */
1895static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1896 unsigned int nr_highmem)
1897{
1898 unsigned int to_alloc = count_free_highmem_pages();
1899
1900 if (to_alloc > nr_highmem)
1901 to_alloc = nr_highmem;
1902
1903 nr_highmem -= to_alloc;
1904 while (to_alloc-- > 0) {
1905 struct page *page;
1906
1907 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1908 memory_bm_set_bit(bm, page_to_pfn(page));
1909 }
1910 return nr_highmem;
1911}
1912#else
1913static inline int get_highmem_buffer(int safe_needed) { return 0; }
1914
1915static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1916 unsigned int n) { return 0; }
1917#endif /* CONFIG_HIGHMEM */
1918
1919/**
1920 * swsusp_alloc - Allocate memory for hibernation image.
1921 *
1922 * We first try to allocate as many highmem pages as there are
1923 * saveable highmem pages in the system. If that fails, we allocate
1924 * non-highmem pages for the copies of the remaining highmem ones.
1925 *
1926 * In this approach it is likely that the copies of highmem pages will
1927 * also be located in the high memory, because of the way in which
1928 * copy_data_pages() works.
1929 */
1930static int swsusp_alloc(struct memory_bitmap *copy_bm,
1931 unsigned int nr_pages, unsigned int nr_highmem)
1932{
1933 if (nr_highmem > 0) {
1934 if (get_highmem_buffer(PG_ANY))
1935 goto err_out;
1936 if (nr_highmem > alloc_highmem) {
1937 nr_highmem -= alloc_highmem;
1938 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1939 }
1940 }
1941 if (nr_pages > alloc_normal) {
1942 nr_pages -= alloc_normal;
1943 while (nr_pages-- > 0) {
1944 struct page *page;
1945
1946 page = alloc_image_page(GFP_ATOMIC);
1947 if (!page)
1948 goto err_out;
1949 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1950 }
1951 }
1952
1953 return 0;
1954
1955 err_out:
1956 swsusp_free();
1957 return -ENOMEM;
1958}
1959
1960asmlinkage __visible int swsusp_save(void)
1961{
1962 unsigned int nr_pages, nr_highmem;
1963
1964 pr_info("Creating hibernation image:\n");
1965
1966 drain_local_pages(NULL);
1967 nr_pages = count_data_pages();
1968 nr_highmem = count_highmem_pages();
1969 pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
1970
1971 if (!enough_free_mem(nr_pages, nr_highmem)) {
1972 pr_err("Not enough free memory\n");
1973 return -ENOMEM;
1974 }
1975
1976 if (swsusp_alloc(©_bm, nr_pages, nr_highmem)) {
1977 pr_err("Memory allocation failed\n");
1978 return -ENOMEM;
1979 }
1980
1981 /*
1982 * During allocating of suspend pagedir, new cold pages may appear.
1983 * Kill them.
1984 */
1985 drain_local_pages(NULL);
1986 copy_data_pages(©_bm, &orig_bm);
1987
1988 /*
1989 * End of critical section. From now on, we can write to memory,
1990 * but we should not touch disk. This specially means we must _not_
1991 * touch swap space! Except we must write out our image of course.
1992 */
1993
1994 nr_pages += nr_highmem;
1995 nr_copy_pages = nr_pages;
1996 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1997
1998 pr_info("Hibernation image created (%d pages copied)\n", nr_pages);
1999
2000 return 0;
2001}
2002
2003#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2004static int init_header_complete(struct swsusp_info *info)
2005{
2006 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2007 info->version_code = LINUX_VERSION_CODE;
2008 return 0;
2009}
2010
2011static char *check_image_kernel(struct swsusp_info *info)
2012{
2013 if (info->version_code != LINUX_VERSION_CODE)
2014 return "kernel version";
2015 if (strcmp(info->uts.sysname,init_utsname()->sysname))
2016 return "system type";
2017 if (strcmp(info->uts.release,init_utsname()->release))
2018 return "kernel release";
2019 if (strcmp(info->uts.version,init_utsname()->version))
2020 return "version";
2021 if (strcmp(info->uts.machine,init_utsname()->machine))
2022 return "machine";
2023 return NULL;
2024}
2025#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2026
2027unsigned long snapshot_get_image_size(void)
2028{
2029 return nr_copy_pages + nr_meta_pages + 1;
2030}
2031
2032static int init_header(struct swsusp_info *info)
2033{
2034 memset(info, 0, sizeof(struct swsusp_info));
2035 info->num_physpages = get_num_physpages();
2036 info->image_pages = nr_copy_pages;
2037 info->pages = snapshot_get_image_size();
2038 info->size = info->pages;
2039 info->size <<= PAGE_SHIFT;
2040 return init_header_complete(info);
2041}
2042
2043/**
2044 * pack_pfns - Prepare PFNs for saving.
2045 * @bm: Memory bitmap.
2046 * @buf: Memory buffer to store the PFNs in.
2047 *
2048 * PFNs corresponding to set bits in @bm are stored in the area of memory
2049 * pointed to by @buf (1 page at a time).
2050 */
2051static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2052{
2053 int j;
2054
2055 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2056 buf[j] = memory_bm_next_pfn(bm);
2057 if (unlikely(buf[j] == BM_END_OF_MAP))
2058 break;
2059 /* Save page key for data page (s390 only). */
2060 page_key_read(buf + j);
2061 }
2062}
2063
2064/**
2065 * snapshot_read_next - Get the address to read the next image page from.
2066 * @handle: Snapshot handle to be used for the reading.
2067 *
2068 * On the first call, @handle should point to a zeroed snapshot_handle
2069 * structure. The structure gets populated then and a pointer to it should be
2070 * passed to this function every next time.
2071 *
2072 * On success, the function returns a positive number. Then, the caller
2073 * is allowed to read up to the returned number of bytes from the memory
2074 * location computed by the data_of() macro.
2075 *
2076 * The function returns 0 to indicate the end of the data stream condition,
2077 * and negative numbers are returned on errors. If that happens, the structure
2078 * pointed to by @handle is not updated and should not be used any more.
2079 */
2080int snapshot_read_next(struct snapshot_handle *handle)
2081{
2082 if (handle->cur > nr_meta_pages + nr_copy_pages)
2083 return 0;
2084
2085 if (!buffer) {
2086 /* This makes the buffer be freed by swsusp_free() */
2087 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2088 if (!buffer)
2089 return -ENOMEM;
2090 }
2091 if (!handle->cur) {
2092 int error;
2093
2094 error = init_header((struct swsusp_info *)buffer);
2095 if (error)
2096 return error;
2097 handle->buffer = buffer;
2098 memory_bm_position_reset(&orig_bm);
2099 memory_bm_position_reset(©_bm);
2100 } else if (handle->cur <= nr_meta_pages) {
2101 clear_page(buffer);
2102 pack_pfns(buffer, &orig_bm);
2103 } else {
2104 struct page *page;
2105
2106 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2107 if (PageHighMem(page)) {
2108 /*
2109 * Highmem pages are copied to the buffer,
2110 * because we can't return with a kmapped
2111 * highmem page (we may not be called again).
2112 */
2113 void *kaddr;
2114
2115 kaddr = kmap_atomic(page);
2116 copy_page(buffer, kaddr);
2117 kunmap_atomic(kaddr);
2118 handle->buffer = buffer;
2119 } else {
2120 handle->buffer = page_address(page);
2121 }
2122 }
2123 handle->cur++;
2124 return PAGE_SIZE;
2125}
2126
2127static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2128 struct memory_bitmap *src)
2129{
2130 unsigned long pfn;
2131
2132 memory_bm_position_reset(src);
2133 pfn = memory_bm_next_pfn(src);
2134 while (pfn != BM_END_OF_MAP) {
2135 memory_bm_set_bit(dst, pfn);
2136 pfn = memory_bm_next_pfn(src);
2137 }
2138}
2139
2140/**
2141 * mark_unsafe_pages - Mark pages that were used before hibernation.
2142 *
2143 * Mark the pages that cannot be used for storing the image during restoration,
2144 * because they conflict with the pages that had been used before hibernation.
2145 */
2146static void mark_unsafe_pages(struct memory_bitmap *bm)
2147{
2148 unsigned long pfn;
2149
2150 /* Clear the "free"/"unsafe" bit for all PFNs */
2151 memory_bm_position_reset(free_pages_map);
2152 pfn = memory_bm_next_pfn(free_pages_map);
2153 while (pfn != BM_END_OF_MAP) {
2154 memory_bm_clear_current(free_pages_map);
2155 pfn = memory_bm_next_pfn(free_pages_map);
2156 }
2157
2158 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2159 duplicate_memory_bitmap(free_pages_map, bm);
2160
2161 allocated_unsafe_pages = 0;
2162}
2163
2164static int check_header(struct swsusp_info *info)
2165{
2166 char *reason;
2167
2168 reason = check_image_kernel(info);
2169 if (!reason && info->num_physpages != get_num_physpages())
2170 reason = "memory size";
2171 if (reason) {
2172 pr_err("Image mismatch: %s\n", reason);
2173 return -EPERM;
2174 }
2175 return 0;
2176}
2177
2178/**
2179 * load header - Check the image header and copy the data from it.
2180 */
2181static int load_header(struct swsusp_info *info)
2182{
2183 int error;
2184
2185 restore_pblist = NULL;
2186 error = check_header(info);
2187 if (!error) {
2188 nr_copy_pages = info->image_pages;
2189 nr_meta_pages = info->pages - info->image_pages - 1;
2190 }
2191 return error;
2192}
2193
2194/**
2195 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2196 * @bm: Memory bitmap.
2197 * @buf: Area of memory containing the PFNs.
2198 *
2199 * For each element of the array pointed to by @buf (1 page at a time), set the
2200 * corresponding bit in @bm.
2201 */
2202static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2203{
2204 int j;
2205
2206 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2207 if (unlikely(buf[j] == BM_END_OF_MAP))
2208 break;
2209
2210 /* Extract and buffer page key for data page (s390 only). */
2211 page_key_memorize(buf + j);
2212
2213 if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2214 memory_bm_set_bit(bm, buf[j]);
2215 else
2216 return -EFAULT;
2217 }
2218
2219 return 0;
2220}
2221
2222#ifdef CONFIG_HIGHMEM
2223/*
2224 * struct highmem_pbe is used for creating the list of highmem pages that
2225 * should be restored atomically during the resume from disk, because the page
2226 * frames they have occupied before the suspend are in use.
2227 */
2228struct highmem_pbe {
2229 struct page *copy_page; /* data is here now */
2230 struct page *orig_page; /* data was here before the suspend */
2231 struct highmem_pbe *next;
2232};
2233
2234/*
2235 * List of highmem PBEs needed for restoring the highmem pages that were
2236 * allocated before the suspend and included in the suspend image, but have
2237 * also been allocated by the "resume" kernel, so their contents cannot be
2238 * written directly to their "original" page frames.
2239 */
2240static struct highmem_pbe *highmem_pblist;
2241
2242/**
2243 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2244 * @bm: Memory bitmap.
2245 *
2246 * The bits in @bm that correspond to image pages are assumed to be set.
2247 */
2248static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2249{
2250 unsigned long pfn;
2251 unsigned int cnt = 0;
2252
2253 memory_bm_position_reset(bm);
2254 pfn = memory_bm_next_pfn(bm);
2255 while (pfn != BM_END_OF_MAP) {
2256 if (PageHighMem(pfn_to_page(pfn)))
2257 cnt++;
2258
2259 pfn = memory_bm_next_pfn(bm);
2260 }
2261 return cnt;
2262}
2263
2264static unsigned int safe_highmem_pages;
2265
2266static struct memory_bitmap *safe_highmem_bm;
2267
2268/**
2269 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2270 * @bm: Pointer to an uninitialized memory bitmap structure.
2271 * @nr_highmem_p: Pointer to the number of highmem image pages.
2272 *
2273 * Try to allocate as many highmem pages as there are highmem image pages
2274 * (@nr_highmem_p points to the variable containing the number of highmem image
2275 * pages). The pages that are "safe" (ie. will not be overwritten when the
2276 * hibernation image is restored entirely) have the corresponding bits set in
2277 * @bm (it must be unitialized).
2278 *
2279 * NOTE: This function should not be called if there are no highmem image pages.
2280 */
2281static int prepare_highmem_image(struct memory_bitmap *bm,
2282 unsigned int *nr_highmem_p)
2283{
2284 unsigned int to_alloc;
2285
2286 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2287 return -ENOMEM;
2288
2289 if (get_highmem_buffer(PG_SAFE))
2290 return -ENOMEM;
2291
2292 to_alloc = count_free_highmem_pages();
2293 if (to_alloc > *nr_highmem_p)
2294 to_alloc = *nr_highmem_p;
2295 else
2296 *nr_highmem_p = to_alloc;
2297
2298 safe_highmem_pages = 0;
2299 while (to_alloc-- > 0) {
2300 struct page *page;
2301
2302 page = alloc_page(__GFP_HIGHMEM);
2303 if (!swsusp_page_is_free(page)) {
2304 /* The page is "safe", set its bit the bitmap */
2305 memory_bm_set_bit(bm, page_to_pfn(page));
2306 safe_highmem_pages++;
2307 }
2308 /* Mark the page as allocated */
2309 swsusp_set_page_forbidden(page);
2310 swsusp_set_page_free(page);
2311 }
2312 memory_bm_position_reset(bm);
2313 safe_highmem_bm = bm;
2314 return 0;
2315}
2316
2317static struct page *last_highmem_page;
2318
2319/**
2320 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2321 *
2322 * For a given highmem image page get a buffer that suspend_write_next() should
2323 * return to its caller to write to.
2324 *
2325 * If the page is to be saved to its "original" page frame or a copy of
2326 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2327 * the copy of the page is to be made in normal memory, so the address of
2328 * the copy is returned.
2329 *
2330 * If @buffer is returned, the caller of suspend_write_next() will write
2331 * the page's contents to @buffer, so they will have to be copied to the
2332 * right location on the next call to suspend_write_next() and it is done
2333 * with the help of copy_last_highmem_page(). For this purpose, if
2334 * @buffer is returned, @last_highmem_page is set to the page to which
2335 * the data will have to be copied from @buffer.
2336 */
2337static void *get_highmem_page_buffer(struct page *page,
2338 struct chain_allocator *ca)
2339{
2340 struct highmem_pbe *pbe;
2341 void *kaddr;
2342
2343 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2344 /*
2345 * We have allocated the "original" page frame and we can
2346 * use it directly to store the loaded page.
2347 */
2348 last_highmem_page = page;
2349 return buffer;
2350 }
2351 /*
2352 * The "original" page frame has not been allocated and we have to
2353 * use a "safe" page frame to store the loaded page.
2354 */
2355 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2356 if (!pbe) {
2357 swsusp_free();
2358 return ERR_PTR(-ENOMEM);
2359 }
2360 pbe->orig_page = page;
2361 if (safe_highmem_pages > 0) {
2362 struct page *tmp;
2363
2364 /* Copy of the page will be stored in high memory */
2365 kaddr = buffer;
2366 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2367 safe_highmem_pages--;
2368 last_highmem_page = tmp;
2369 pbe->copy_page = tmp;
2370 } else {
2371 /* Copy of the page will be stored in normal memory */
2372 kaddr = safe_pages_list;
2373 safe_pages_list = safe_pages_list->next;
2374 pbe->copy_page = virt_to_page(kaddr);
2375 }
2376 pbe->next = highmem_pblist;
2377 highmem_pblist = pbe;
2378 return kaddr;
2379}
2380
2381/**
2382 * copy_last_highmem_page - Copy most the most recent highmem image page.
2383 *
2384 * Copy the contents of a highmem image from @buffer, where the caller of
2385 * snapshot_write_next() has stored them, to the right location represented by
2386 * @last_highmem_page .
2387 */
2388static void copy_last_highmem_page(void)
2389{
2390 if (last_highmem_page) {
2391 void *dst;
2392
2393 dst = kmap_atomic(last_highmem_page);
2394 copy_page(dst, buffer);
2395 kunmap_atomic(dst);
2396 last_highmem_page = NULL;
2397 }
2398}
2399
2400static inline int last_highmem_page_copied(void)
2401{
2402 return !last_highmem_page;
2403}
2404
2405static inline void free_highmem_data(void)
2406{
2407 if (safe_highmem_bm)
2408 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2409
2410 if (buffer)
2411 free_image_page(buffer, PG_UNSAFE_CLEAR);
2412}
2413#else
2414static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2415
2416static inline int prepare_highmem_image(struct memory_bitmap *bm,
2417 unsigned int *nr_highmem_p) { return 0; }
2418
2419static inline void *get_highmem_page_buffer(struct page *page,
2420 struct chain_allocator *ca)
2421{
2422 return ERR_PTR(-EINVAL);
2423}
2424
2425static inline void copy_last_highmem_page(void) {}
2426static inline int last_highmem_page_copied(void) { return 1; }
2427static inline void free_highmem_data(void) {}
2428#endif /* CONFIG_HIGHMEM */
2429
2430#define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2431
2432/**
2433 * prepare_image - Make room for loading hibernation image.
2434 * @new_bm: Unitialized memory bitmap structure.
2435 * @bm: Memory bitmap with unsafe pages marked.
2436 *
2437 * Use @bm to mark the pages that will be overwritten in the process of
2438 * restoring the system memory state from the suspend image ("unsafe" pages)
2439 * and allocate memory for the image.
2440 *
2441 * The idea is to allocate a new memory bitmap first and then allocate
2442 * as many pages as needed for image data, but without specifying what those
2443 * pages will be used for just yet. Instead, we mark them all as allocated and
2444 * create a lists of "safe" pages to be used later. On systems with high
2445 * memory a list of "safe" highmem pages is created too.
2446 */
2447static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2448{
2449 unsigned int nr_pages, nr_highmem;
2450 struct linked_page *lp;
2451 int error;
2452
2453 /* If there is no highmem, the buffer will not be necessary */
2454 free_image_page(buffer, PG_UNSAFE_CLEAR);
2455 buffer = NULL;
2456
2457 nr_highmem = count_highmem_image_pages(bm);
2458 mark_unsafe_pages(bm);
2459
2460 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2461 if (error)
2462 goto Free;
2463
2464 duplicate_memory_bitmap(new_bm, bm);
2465 memory_bm_free(bm, PG_UNSAFE_KEEP);
2466 if (nr_highmem > 0) {
2467 error = prepare_highmem_image(bm, &nr_highmem);
2468 if (error)
2469 goto Free;
2470 }
2471 /*
2472 * Reserve some safe pages for potential later use.
2473 *
2474 * NOTE: This way we make sure there will be enough safe pages for the
2475 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2476 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2477 *
2478 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2479 */
2480 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2481 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2482 while (nr_pages > 0) {
2483 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2484 if (!lp) {
2485 error = -ENOMEM;
2486 goto Free;
2487 }
2488 lp->next = safe_pages_list;
2489 safe_pages_list = lp;
2490 nr_pages--;
2491 }
2492 /* Preallocate memory for the image */
2493 nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2494 while (nr_pages > 0) {
2495 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2496 if (!lp) {
2497 error = -ENOMEM;
2498 goto Free;
2499 }
2500 if (!swsusp_page_is_free(virt_to_page(lp))) {
2501 /* The page is "safe", add it to the list */
2502 lp->next = safe_pages_list;
2503 safe_pages_list = lp;
2504 }
2505 /* Mark the page as allocated */
2506 swsusp_set_page_forbidden(virt_to_page(lp));
2507 swsusp_set_page_free(virt_to_page(lp));
2508 nr_pages--;
2509 }
2510 return 0;
2511
2512 Free:
2513 swsusp_free();
2514 return error;
2515}
2516
2517/**
2518 * get_buffer - Get the address to store the next image data page.
2519 *
2520 * Get the address that snapshot_write_next() should return to its caller to
2521 * write to.
2522 */
2523static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2524{
2525 struct pbe *pbe;
2526 struct page *page;
2527 unsigned long pfn = memory_bm_next_pfn(bm);
2528
2529 if (pfn == BM_END_OF_MAP)
2530 return ERR_PTR(-EFAULT);
2531
2532 page = pfn_to_page(pfn);
2533 if (PageHighMem(page))
2534 return get_highmem_page_buffer(page, ca);
2535
2536 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2537 /*
2538 * We have allocated the "original" page frame and we can
2539 * use it directly to store the loaded page.
2540 */
2541 return page_address(page);
2542
2543 /*
2544 * The "original" page frame has not been allocated and we have to
2545 * use a "safe" page frame to store the loaded page.
2546 */
2547 pbe = chain_alloc(ca, sizeof(struct pbe));
2548 if (!pbe) {
2549 swsusp_free();
2550 return ERR_PTR(-ENOMEM);
2551 }
2552 pbe->orig_address = page_address(page);
2553 pbe->address = safe_pages_list;
2554 safe_pages_list = safe_pages_list->next;
2555 pbe->next = restore_pblist;
2556 restore_pblist = pbe;
2557 return pbe->address;
2558}
2559
2560/**
2561 * snapshot_write_next - Get the address to store the next image page.
2562 * @handle: Snapshot handle structure to guide the writing.
2563 *
2564 * On the first call, @handle should point to a zeroed snapshot_handle
2565 * structure. The structure gets populated then and a pointer to it should be
2566 * passed to this function every next time.
2567 *
2568 * On success, the function returns a positive number. Then, the caller
2569 * is allowed to write up to the returned number of bytes to the memory
2570 * location computed by the data_of() macro.
2571 *
2572 * The function returns 0 to indicate the "end of file" condition. Negative
2573 * numbers are returned on errors, in which cases the structure pointed to by
2574 * @handle is not updated and should not be used any more.
2575 */
2576int snapshot_write_next(struct snapshot_handle *handle)
2577{
2578 static struct chain_allocator ca;
2579 int error = 0;
2580
2581 /* Check if we have already loaded the entire image */
2582 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2583 return 0;
2584
2585 handle->sync_read = 1;
2586
2587 if (!handle->cur) {
2588 if (!buffer)
2589 /* This makes the buffer be freed by swsusp_free() */
2590 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2591
2592 if (!buffer)
2593 return -ENOMEM;
2594
2595 handle->buffer = buffer;
2596 } else if (handle->cur == 1) {
2597 error = load_header(buffer);
2598 if (error)
2599 return error;
2600
2601 safe_pages_list = NULL;
2602
2603 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2604 if (error)
2605 return error;
2606
2607 /* Allocate buffer for page keys. */
2608 error = page_key_alloc(nr_copy_pages);
2609 if (error)
2610 return error;
2611
2612 hibernate_restore_protection_begin();
2613 } else if (handle->cur <= nr_meta_pages + 1) {
2614 error = unpack_orig_pfns(buffer, ©_bm);
2615 if (error)
2616 return error;
2617
2618 if (handle->cur == nr_meta_pages + 1) {
2619 error = prepare_image(&orig_bm, ©_bm);
2620 if (error)
2621 return error;
2622
2623 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2624 memory_bm_position_reset(&orig_bm);
2625 restore_pblist = NULL;
2626 handle->buffer = get_buffer(&orig_bm, &ca);
2627 handle->sync_read = 0;
2628 if (IS_ERR(handle->buffer))
2629 return PTR_ERR(handle->buffer);
2630 }
2631 } else {
2632 copy_last_highmem_page();
2633 /* Restore page key for data page (s390 only). */
2634 page_key_write(handle->buffer);
2635 hibernate_restore_protect_page(handle->buffer);
2636 handle->buffer = get_buffer(&orig_bm, &ca);
2637 if (IS_ERR(handle->buffer))
2638 return PTR_ERR(handle->buffer);
2639 if (handle->buffer != buffer)
2640 handle->sync_read = 0;
2641 }
2642 handle->cur++;
2643 return PAGE_SIZE;
2644}
2645
2646/**
2647 * snapshot_write_finalize - Complete the loading of a hibernation image.
2648 *
2649 * Must be called after the last call to snapshot_write_next() in case the last
2650 * page in the image happens to be a highmem page and its contents should be
2651 * stored in highmem. Additionally, it recycles bitmap memory that's not
2652 * necessary any more.
2653 */
2654void snapshot_write_finalize(struct snapshot_handle *handle)
2655{
2656 copy_last_highmem_page();
2657 /* Restore page key for data page (s390 only). */
2658 page_key_write(handle->buffer);
2659 page_key_free();
2660 hibernate_restore_protect_page(handle->buffer);
2661 /* Do that only if we have loaded the image entirely */
2662 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2663 memory_bm_recycle(&orig_bm);
2664 free_highmem_data();
2665 }
2666}
2667
2668int snapshot_image_loaded(struct snapshot_handle *handle)
2669{
2670 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2671 handle->cur <= nr_meta_pages + nr_copy_pages);
2672}
2673
2674#ifdef CONFIG_HIGHMEM
2675/* Assumes that @buf is ready and points to a "safe" page */
2676static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2677 void *buf)
2678{
2679 void *kaddr1, *kaddr2;
2680
2681 kaddr1 = kmap_atomic(p1);
2682 kaddr2 = kmap_atomic(p2);
2683 copy_page(buf, kaddr1);
2684 copy_page(kaddr1, kaddr2);
2685 copy_page(kaddr2, buf);
2686 kunmap_atomic(kaddr2);
2687 kunmap_atomic(kaddr1);
2688}
2689
2690/**
2691 * restore_highmem - Put highmem image pages into their original locations.
2692 *
2693 * For each highmem page that was in use before hibernation and is included in
2694 * the image, and also has been allocated by the "restore" kernel, swap its
2695 * current contents with the previous (ie. "before hibernation") ones.
2696 *
2697 * If the restore eventually fails, we can call this function once again and
2698 * restore the highmem state as seen by the restore kernel.
2699 */
2700int restore_highmem(void)
2701{
2702 struct highmem_pbe *pbe = highmem_pblist;
2703 void *buf;
2704
2705 if (!pbe)
2706 return 0;
2707
2708 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2709 if (!buf)
2710 return -ENOMEM;
2711
2712 while (pbe) {
2713 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2714 pbe = pbe->next;
2715 }
2716 free_image_page(buf, PG_UNSAFE_CLEAR);
2717 return 0;
2718}
2719#endif /* CONFIG_HIGHMEM */
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/power/snapshot.c
4 *
5 * This file provides system snapshot/restore functionality for swsusp.
6 *
7 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
8 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
9 */
10
11#define pr_fmt(fmt) "PM: hibernation: " fmt
12
13#include <linux/version.h>
14#include <linux/module.h>
15#include <linux/mm.h>
16#include <linux/suspend.h>
17#include <linux/delay.h>
18#include <linux/bitops.h>
19#include <linux/spinlock.h>
20#include <linux/kernel.h>
21#include <linux/pm.h>
22#include <linux/device.h>
23#include <linux/init.h>
24#include <linux/memblock.h>
25#include <linux/nmi.h>
26#include <linux/syscalls.h>
27#include <linux/console.h>
28#include <linux/highmem.h>
29#include <linux/list.h>
30#include <linux/slab.h>
31#include <linux/compiler.h>
32#include <linux/ktime.h>
33#include <linux/set_memory.h>
34
35#include <linux/uaccess.h>
36#include <asm/mmu_context.h>
37#include <asm/tlbflush.h>
38#include <asm/io.h>
39
40#include "power.h"
41
42#if defined(CONFIG_STRICT_KERNEL_RWX) && defined(CONFIG_ARCH_HAS_SET_MEMORY)
43static bool hibernate_restore_protection;
44static bool hibernate_restore_protection_active;
45
46void enable_restore_image_protection(void)
47{
48 hibernate_restore_protection = true;
49}
50
51static inline void hibernate_restore_protection_begin(void)
52{
53 hibernate_restore_protection_active = hibernate_restore_protection;
54}
55
56static inline void hibernate_restore_protection_end(void)
57{
58 hibernate_restore_protection_active = false;
59}
60
61static inline int __must_check hibernate_restore_protect_page(void *page_address)
62{
63 if (hibernate_restore_protection_active)
64 return set_memory_ro((unsigned long)page_address, 1);
65 return 0;
66}
67
68static inline int hibernate_restore_unprotect_page(void *page_address)
69{
70 if (hibernate_restore_protection_active)
71 return set_memory_rw((unsigned long)page_address, 1);
72 return 0;
73}
74#else
75static inline void hibernate_restore_protection_begin(void) {}
76static inline void hibernate_restore_protection_end(void) {}
77static inline int __must_check hibernate_restore_protect_page(void *page_address) {return 0; }
78static inline int hibernate_restore_unprotect_page(void *page_address) {return 0; }
79#endif /* CONFIG_STRICT_KERNEL_RWX && CONFIG_ARCH_HAS_SET_MEMORY */
80
81
82/*
83 * The calls to set_direct_map_*() should not fail because remapping a page
84 * here means that we only update protection bits in an existing PTE.
85 * It is still worth to have a warning here if something changes and this
86 * will no longer be the case.
87 */
88static inline void hibernate_map_page(struct page *page)
89{
90 if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
91 int ret = set_direct_map_default_noflush(page);
92
93 if (ret)
94 pr_warn_once("Failed to remap page\n");
95 } else {
96 debug_pagealloc_map_pages(page, 1);
97 }
98}
99
100static inline void hibernate_unmap_page(struct page *page)
101{
102 if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
103 unsigned long addr = (unsigned long)page_address(page);
104 int ret = set_direct_map_invalid_noflush(page);
105
106 if (ret)
107 pr_warn_once("Failed to remap page\n");
108
109 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
110 } else {
111 debug_pagealloc_unmap_pages(page, 1);
112 }
113}
114
115static int swsusp_page_is_free(struct page *);
116static void swsusp_set_page_forbidden(struct page *);
117static void swsusp_unset_page_forbidden(struct page *);
118
119/*
120 * Number of bytes to reserve for memory allocations made by device drivers
121 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
122 * cause image creation to fail (tunable via /sys/power/reserved_size).
123 */
124unsigned long reserved_size;
125
126void __init hibernate_reserved_size_init(void)
127{
128 reserved_size = SPARE_PAGES * PAGE_SIZE;
129}
130
131/*
132 * Preferred image size in bytes (tunable via /sys/power/image_size).
133 * When it is set to N, swsusp will do its best to ensure the image
134 * size will not exceed N bytes, but if that is impossible, it will
135 * try to create the smallest image possible.
136 */
137unsigned long image_size;
138
139void __init hibernate_image_size_init(void)
140{
141 image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
142}
143
144/*
145 * List of PBEs needed for restoring the pages that were allocated before
146 * the suspend and included in the suspend image, but have also been
147 * allocated by the "resume" kernel, so their contents cannot be written
148 * directly to their "original" page frames.
149 */
150struct pbe *restore_pblist;
151
152/* struct linked_page is used to build chains of pages */
153
154#define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
155
156struct linked_page {
157 struct linked_page *next;
158 char data[LINKED_PAGE_DATA_SIZE];
159} __packed;
160
161/*
162 * List of "safe" pages (ie. pages that were not used by the image kernel
163 * before hibernation) that may be used as temporary storage for image kernel
164 * memory contents.
165 */
166static struct linked_page *safe_pages_list;
167
168/* Pointer to an auxiliary buffer (1 page) */
169static void *buffer;
170
171#define PG_ANY 0
172#define PG_SAFE 1
173#define PG_UNSAFE_CLEAR 1
174#define PG_UNSAFE_KEEP 0
175
176static unsigned int allocated_unsafe_pages;
177
178/**
179 * get_image_page - Allocate a page for a hibernation image.
180 * @gfp_mask: GFP mask for the allocation.
181 * @safe_needed: Get pages that were not used before hibernation (restore only)
182 *
183 * During image restoration, for storing the PBE list and the image data, we can
184 * only use memory pages that do not conflict with the pages used before
185 * hibernation. The "unsafe" pages have PageNosaveFree set and we count them
186 * using allocated_unsafe_pages.
187 *
188 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
189 * swsusp_free() can release it.
190 */
191static void *get_image_page(gfp_t gfp_mask, int safe_needed)
192{
193 void *res;
194
195 res = (void *)get_zeroed_page(gfp_mask);
196 if (safe_needed)
197 while (res && swsusp_page_is_free(virt_to_page(res))) {
198 /* The page is unsafe, mark it for swsusp_free() */
199 swsusp_set_page_forbidden(virt_to_page(res));
200 allocated_unsafe_pages++;
201 res = (void *)get_zeroed_page(gfp_mask);
202 }
203 if (res) {
204 swsusp_set_page_forbidden(virt_to_page(res));
205 swsusp_set_page_free(virt_to_page(res));
206 }
207 return res;
208}
209
210static void *__get_safe_page(gfp_t gfp_mask)
211{
212 if (safe_pages_list) {
213 void *ret = safe_pages_list;
214
215 safe_pages_list = safe_pages_list->next;
216 memset(ret, 0, PAGE_SIZE);
217 return ret;
218 }
219 return get_image_page(gfp_mask, PG_SAFE);
220}
221
222unsigned long get_safe_page(gfp_t gfp_mask)
223{
224 return (unsigned long)__get_safe_page(gfp_mask);
225}
226
227static struct page *alloc_image_page(gfp_t gfp_mask)
228{
229 struct page *page;
230
231 page = alloc_page(gfp_mask);
232 if (page) {
233 swsusp_set_page_forbidden(page);
234 swsusp_set_page_free(page);
235 }
236 return page;
237}
238
239static void recycle_safe_page(void *page_address)
240{
241 struct linked_page *lp = page_address;
242
243 lp->next = safe_pages_list;
244 safe_pages_list = lp;
245}
246
247/**
248 * free_image_page - Free a page allocated for hibernation image.
249 * @addr: Address of the page to free.
250 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
251 *
252 * The page to free should have been allocated by get_image_page() (page flags
253 * set by it are affected).
254 */
255static inline void free_image_page(void *addr, int clear_nosave_free)
256{
257 struct page *page;
258
259 BUG_ON(!virt_addr_valid(addr));
260
261 page = virt_to_page(addr);
262
263 swsusp_unset_page_forbidden(page);
264 if (clear_nosave_free)
265 swsusp_unset_page_free(page);
266
267 __free_page(page);
268}
269
270static inline void free_list_of_pages(struct linked_page *list,
271 int clear_page_nosave)
272{
273 while (list) {
274 struct linked_page *lp = list->next;
275
276 free_image_page(list, clear_page_nosave);
277 list = lp;
278 }
279}
280
281/*
282 * struct chain_allocator is used for allocating small objects out of
283 * a linked list of pages called 'the chain'.
284 *
285 * The chain grows each time when there is no room for a new object in
286 * the current page. The allocated objects cannot be freed individually.
287 * It is only possible to free them all at once, by freeing the entire
288 * chain.
289 *
290 * NOTE: The chain allocator may be inefficient if the allocated objects
291 * are not much smaller than PAGE_SIZE.
292 */
293struct chain_allocator {
294 struct linked_page *chain; /* the chain */
295 unsigned int used_space; /* total size of objects allocated out
296 of the current page */
297 gfp_t gfp_mask; /* mask for allocating pages */
298 int safe_needed; /* if set, only "safe" pages are allocated */
299};
300
301static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
302 int safe_needed)
303{
304 ca->chain = NULL;
305 ca->used_space = LINKED_PAGE_DATA_SIZE;
306 ca->gfp_mask = gfp_mask;
307 ca->safe_needed = safe_needed;
308}
309
310static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
311{
312 void *ret;
313
314 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
315 struct linked_page *lp;
316
317 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
318 get_image_page(ca->gfp_mask, PG_ANY);
319 if (!lp)
320 return NULL;
321
322 lp->next = ca->chain;
323 ca->chain = lp;
324 ca->used_space = 0;
325 }
326 ret = ca->chain->data + ca->used_space;
327 ca->used_space += size;
328 return ret;
329}
330
331/*
332 * Data types related to memory bitmaps.
333 *
334 * Memory bitmap is a structure consisting of many linked lists of
335 * objects. The main list's elements are of type struct zone_bitmap
336 * and each of them corresponds to one zone. For each zone bitmap
337 * object there is a list of objects of type struct bm_block that
338 * represent each blocks of bitmap in which information is stored.
339 *
340 * struct memory_bitmap contains a pointer to the main list of zone
341 * bitmap objects, a struct bm_position used for browsing the bitmap,
342 * and a pointer to the list of pages used for allocating all of the
343 * zone bitmap objects and bitmap block objects.
344 *
345 * NOTE: It has to be possible to lay out the bitmap in memory
346 * using only allocations of order 0. Additionally, the bitmap is
347 * designed to work with arbitrary number of zones (this is over the
348 * top for now, but let's avoid making unnecessary assumptions ;-).
349 *
350 * struct zone_bitmap contains a pointer to a list of bitmap block
351 * objects and a pointer to the bitmap block object that has been
352 * most recently used for setting bits. Additionally, it contains the
353 * PFNs that correspond to the start and end of the represented zone.
354 *
355 * struct bm_block contains a pointer to the memory page in which
356 * information is stored (in the form of a block of bitmap)
357 * It also contains the pfns that correspond to the start and end of
358 * the represented memory area.
359 *
360 * The memory bitmap is organized as a radix tree to guarantee fast random
361 * access to the bits. There is one radix tree for each zone (as returned
362 * from create_mem_extents).
363 *
364 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
365 * two linked lists for the nodes of the tree, one for the inner nodes and
366 * one for the leave nodes. The linked leave nodes are used for fast linear
367 * access of the memory bitmap.
368 *
369 * The struct rtree_node represents one node of the radix tree.
370 */
371
372#define BM_END_OF_MAP (~0UL)
373
374#define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
375#define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
376#define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
377
378/*
379 * struct rtree_node is a wrapper struct to link the nodes
380 * of the rtree together for easy linear iteration over
381 * bits and easy freeing
382 */
383struct rtree_node {
384 struct list_head list;
385 unsigned long *data;
386};
387
388/*
389 * struct mem_zone_bm_rtree represents a bitmap used for one
390 * populated memory zone.
391 */
392struct mem_zone_bm_rtree {
393 struct list_head list; /* Link Zones together */
394 struct list_head nodes; /* Radix Tree inner nodes */
395 struct list_head leaves; /* Radix Tree leaves */
396 unsigned long start_pfn; /* Zone start page frame */
397 unsigned long end_pfn; /* Zone end page frame + 1 */
398 struct rtree_node *rtree; /* Radix Tree Root */
399 int levels; /* Number of Radix Tree Levels */
400 unsigned int blocks; /* Number of Bitmap Blocks */
401};
402
403/* struct bm_position is used for browsing memory bitmaps */
404
405struct bm_position {
406 struct mem_zone_bm_rtree *zone;
407 struct rtree_node *node;
408 unsigned long node_pfn;
409 unsigned long cur_pfn;
410 int node_bit;
411};
412
413struct memory_bitmap {
414 struct list_head zones;
415 struct linked_page *p_list; /* list of pages used to store zone
416 bitmap objects and bitmap block
417 objects */
418 struct bm_position cur; /* most recently used bit position */
419};
420
421/* Functions that operate on memory bitmaps */
422
423#define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
424#if BITS_PER_LONG == 32
425#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
426#else
427#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
428#endif
429#define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
430
431/**
432 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
433 * @gfp_mask: GFP mask for the allocation.
434 * @safe_needed: Get pages not used before hibernation (restore only)
435 * @ca: Pointer to a linked list of pages ("a chain") to allocate from
436 * @list: Radix Tree node to add.
437 *
438 * This function is used to allocate inner nodes as well as the
439 * leave nodes of the radix tree. It also adds the node to the
440 * corresponding linked list passed in by the *list parameter.
441 */
442static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
443 struct chain_allocator *ca,
444 struct list_head *list)
445{
446 struct rtree_node *node;
447
448 node = chain_alloc(ca, sizeof(struct rtree_node));
449 if (!node)
450 return NULL;
451
452 node->data = get_image_page(gfp_mask, safe_needed);
453 if (!node->data)
454 return NULL;
455
456 list_add_tail(&node->list, list);
457
458 return node;
459}
460
461/**
462 * add_rtree_block - Add a new leave node to the radix tree.
463 *
464 * The leave nodes need to be allocated in order to keep the leaves
465 * linked list in order. This is guaranteed by the zone->blocks
466 * counter.
467 */
468static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
469 int safe_needed, struct chain_allocator *ca)
470{
471 struct rtree_node *node, *block, **dst;
472 unsigned int levels_needed, block_nr;
473 int i;
474
475 block_nr = zone->blocks;
476 levels_needed = 0;
477
478 /* How many levels do we need for this block nr? */
479 while (block_nr) {
480 levels_needed += 1;
481 block_nr >>= BM_RTREE_LEVEL_SHIFT;
482 }
483
484 /* Make sure the rtree has enough levels */
485 for (i = zone->levels; i < levels_needed; i++) {
486 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
487 &zone->nodes);
488 if (!node)
489 return -ENOMEM;
490
491 node->data[0] = (unsigned long)zone->rtree;
492 zone->rtree = node;
493 zone->levels += 1;
494 }
495
496 /* Allocate new block */
497 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
498 if (!block)
499 return -ENOMEM;
500
501 /* Now walk the rtree to insert the block */
502 node = zone->rtree;
503 dst = &zone->rtree;
504 block_nr = zone->blocks;
505 for (i = zone->levels; i > 0; i--) {
506 int index;
507
508 if (!node) {
509 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
510 &zone->nodes);
511 if (!node)
512 return -ENOMEM;
513 *dst = node;
514 }
515
516 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
517 index &= BM_RTREE_LEVEL_MASK;
518 dst = (struct rtree_node **)&((*dst)->data[index]);
519 node = *dst;
520 }
521
522 zone->blocks += 1;
523 *dst = block;
524
525 return 0;
526}
527
528static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
529 int clear_nosave_free);
530
531/**
532 * create_zone_bm_rtree - Create a radix tree for one zone.
533 *
534 * Allocated the mem_zone_bm_rtree structure and initializes it.
535 * This function also allocated and builds the radix tree for the
536 * zone.
537 */
538static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
539 int safe_needed,
540 struct chain_allocator *ca,
541 unsigned long start,
542 unsigned long end)
543{
544 struct mem_zone_bm_rtree *zone;
545 unsigned int i, nr_blocks;
546 unsigned long pages;
547
548 pages = end - start;
549 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
550 if (!zone)
551 return NULL;
552
553 INIT_LIST_HEAD(&zone->nodes);
554 INIT_LIST_HEAD(&zone->leaves);
555 zone->start_pfn = start;
556 zone->end_pfn = end;
557 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
558
559 for (i = 0; i < nr_blocks; i++) {
560 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
561 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
562 return NULL;
563 }
564 }
565
566 return zone;
567}
568
569/**
570 * free_zone_bm_rtree - Free the memory of the radix tree.
571 *
572 * Free all node pages of the radix tree. The mem_zone_bm_rtree
573 * structure itself is not freed here nor are the rtree_node
574 * structs.
575 */
576static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
577 int clear_nosave_free)
578{
579 struct rtree_node *node;
580
581 list_for_each_entry(node, &zone->nodes, list)
582 free_image_page(node->data, clear_nosave_free);
583
584 list_for_each_entry(node, &zone->leaves, list)
585 free_image_page(node->data, clear_nosave_free);
586}
587
588static void memory_bm_position_reset(struct memory_bitmap *bm)
589{
590 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
591 list);
592 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
593 struct rtree_node, list);
594 bm->cur.node_pfn = 0;
595 bm->cur.cur_pfn = BM_END_OF_MAP;
596 bm->cur.node_bit = 0;
597}
598
599static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
600
601struct mem_extent {
602 struct list_head hook;
603 unsigned long start;
604 unsigned long end;
605};
606
607/**
608 * free_mem_extents - Free a list of memory extents.
609 * @list: List of extents to free.
610 */
611static void free_mem_extents(struct list_head *list)
612{
613 struct mem_extent *ext, *aux;
614
615 list_for_each_entry_safe(ext, aux, list, hook) {
616 list_del(&ext->hook);
617 kfree(ext);
618 }
619}
620
621/**
622 * create_mem_extents - Create a list of memory extents.
623 * @list: List to put the extents into.
624 * @gfp_mask: Mask to use for memory allocations.
625 *
626 * The extents represent contiguous ranges of PFNs.
627 */
628static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
629{
630 struct zone *zone;
631
632 INIT_LIST_HEAD(list);
633
634 for_each_populated_zone(zone) {
635 unsigned long zone_start, zone_end;
636 struct mem_extent *ext, *cur, *aux;
637
638 zone_start = zone->zone_start_pfn;
639 zone_end = zone_end_pfn(zone);
640
641 list_for_each_entry(ext, list, hook)
642 if (zone_start <= ext->end)
643 break;
644
645 if (&ext->hook == list || zone_end < ext->start) {
646 /* New extent is necessary */
647 struct mem_extent *new_ext;
648
649 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
650 if (!new_ext) {
651 free_mem_extents(list);
652 return -ENOMEM;
653 }
654 new_ext->start = zone_start;
655 new_ext->end = zone_end;
656 list_add_tail(&new_ext->hook, &ext->hook);
657 continue;
658 }
659
660 /* Merge this zone's range of PFNs with the existing one */
661 if (zone_start < ext->start)
662 ext->start = zone_start;
663 if (zone_end > ext->end)
664 ext->end = zone_end;
665
666 /* More merging may be possible */
667 cur = ext;
668 list_for_each_entry_safe_continue(cur, aux, list, hook) {
669 if (zone_end < cur->start)
670 break;
671 if (zone_end < cur->end)
672 ext->end = cur->end;
673 list_del(&cur->hook);
674 kfree(cur);
675 }
676 }
677
678 return 0;
679}
680
681/**
682 * memory_bm_create - Allocate memory for a memory bitmap.
683 */
684static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
685 int safe_needed)
686{
687 struct chain_allocator ca;
688 struct list_head mem_extents;
689 struct mem_extent *ext;
690 int error;
691
692 chain_init(&ca, gfp_mask, safe_needed);
693 INIT_LIST_HEAD(&bm->zones);
694
695 error = create_mem_extents(&mem_extents, gfp_mask);
696 if (error)
697 return error;
698
699 list_for_each_entry(ext, &mem_extents, hook) {
700 struct mem_zone_bm_rtree *zone;
701
702 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
703 ext->start, ext->end);
704 if (!zone) {
705 error = -ENOMEM;
706 goto Error;
707 }
708 list_add_tail(&zone->list, &bm->zones);
709 }
710
711 bm->p_list = ca.chain;
712 memory_bm_position_reset(bm);
713 Exit:
714 free_mem_extents(&mem_extents);
715 return error;
716
717 Error:
718 bm->p_list = ca.chain;
719 memory_bm_free(bm, PG_UNSAFE_CLEAR);
720 goto Exit;
721}
722
723/**
724 * memory_bm_free - Free memory occupied by the memory bitmap.
725 * @bm: Memory bitmap.
726 */
727static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
728{
729 struct mem_zone_bm_rtree *zone;
730
731 list_for_each_entry(zone, &bm->zones, list)
732 free_zone_bm_rtree(zone, clear_nosave_free);
733
734 free_list_of_pages(bm->p_list, clear_nosave_free);
735
736 INIT_LIST_HEAD(&bm->zones);
737}
738
739/**
740 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
741 *
742 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
743 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
744 *
745 * Walk the radix tree to find the page containing the bit that represents @pfn
746 * and return the position of the bit in @addr and @bit_nr.
747 */
748static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
749 void **addr, unsigned int *bit_nr)
750{
751 struct mem_zone_bm_rtree *curr, *zone;
752 struct rtree_node *node;
753 int i, block_nr;
754
755 zone = bm->cur.zone;
756
757 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
758 goto zone_found;
759
760 zone = NULL;
761
762 /* Find the right zone */
763 list_for_each_entry(curr, &bm->zones, list) {
764 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
765 zone = curr;
766 break;
767 }
768 }
769
770 if (!zone)
771 return -EFAULT;
772
773zone_found:
774 /*
775 * We have found the zone. Now walk the radix tree to find the leaf node
776 * for our PFN.
777 */
778
779 /*
780 * If the zone we wish to scan is the current zone and the
781 * pfn falls into the current node then we do not need to walk
782 * the tree.
783 */
784 node = bm->cur.node;
785 if (zone == bm->cur.zone &&
786 ((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
787 goto node_found;
788
789 node = zone->rtree;
790 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
791
792 for (i = zone->levels; i > 0; i--) {
793 int index;
794
795 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
796 index &= BM_RTREE_LEVEL_MASK;
797 BUG_ON(node->data[index] == 0);
798 node = (struct rtree_node *)node->data[index];
799 }
800
801node_found:
802 /* Update last position */
803 bm->cur.zone = zone;
804 bm->cur.node = node;
805 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
806 bm->cur.cur_pfn = pfn;
807
808 /* Set return values */
809 *addr = node->data;
810 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
811
812 return 0;
813}
814
815static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
816{
817 void *addr;
818 unsigned int bit;
819 int error;
820
821 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
822 BUG_ON(error);
823 set_bit(bit, addr);
824}
825
826static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
827{
828 void *addr;
829 unsigned int bit;
830 int error;
831
832 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
833 if (!error)
834 set_bit(bit, addr);
835
836 return error;
837}
838
839static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
840{
841 void *addr;
842 unsigned int bit;
843 int error;
844
845 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
846 BUG_ON(error);
847 clear_bit(bit, addr);
848}
849
850static void memory_bm_clear_current(struct memory_bitmap *bm)
851{
852 int bit;
853
854 bit = max(bm->cur.node_bit - 1, 0);
855 clear_bit(bit, bm->cur.node->data);
856}
857
858static unsigned long memory_bm_get_current(struct memory_bitmap *bm)
859{
860 return bm->cur.cur_pfn;
861}
862
863static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
864{
865 void *addr;
866 unsigned int bit;
867 int error;
868
869 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
870 BUG_ON(error);
871 return test_bit(bit, addr);
872}
873
874static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
875{
876 void *addr;
877 unsigned int bit;
878
879 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
880}
881
882/*
883 * rtree_next_node - Jump to the next leaf node.
884 *
885 * Set the position to the beginning of the next node in the
886 * memory bitmap. This is either the next node in the current
887 * zone's radix tree or the first node in the radix tree of the
888 * next zone.
889 *
890 * Return true if there is a next node, false otherwise.
891 */
892static bool rtree_next_node(struct memory_bitmap *bm)
893{
894 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
895 bm->cur.node = list_entry(bm->cur.node->list.next,
896 struct rtree_node, list);
897 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
898 bm->cur.node_bit = 0;
899 touch_softlockup_watchdog();
900 return true;
901 }
902
903 /* No more nodes, goto next zone */
904 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
905 bm->cur.zone = list_entry(bm->cur.zone->list.next,
906 struct mem_zone_bm_rtree, list);
907 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
908 struct rtree_node, list);
909 bm->cur.node_pfn = 0;
910 bm->cur.node_bit = 0;
911 return true;
912 }
913
914 /* No more zones */
915 return false;
916}
917
918/**
919 * memory_bm_next_pfn - Find the next set bit in a memory bitmap.
920 * @bm: Memory bitmap.
921 *
922 * Starting from the last returned position this function searches for the next
923 * set bit in @bm and returns the PFN represented by it. If no more bits are
924 * set, BM_END_OF_MAP is returned.
925 *
926 * It is required to run memory_bm_position_reset() before the first call to
927 * this function for the given memory bitmap.
928 */
929static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
930{
931 unsigned long bits, pfn, pages;
932 int bit;
933
934 do {
935 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
936 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
937 bit = find_next_bit(bm->cur.node->data, bits,
938 bm->cur.node_bit);
939 if (bit < bits) {
940 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
941 bm->cur.node_bit = bit + 1;
942 bm->cur.cur_pfn = pfn;
943 return pfn;
944 }
945 } while (rtree_next_node(bm));
946
947 bm->cur.cur_pfn = BM_END_OF_MAP;
948 return BM_END_OF_MAP;
949}
950
951/*
952 * This structure represents a range of page frames the contents of which
953 * should not be saved during hibernation.
954 */
955struct nosave_region {
956 struct list_head list;
957 unsigned long start_pfn;
958 unsigned long end_pfn;
959};
960
961static LIST_HEAD(nosave_regions);
962
963static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
964{
965 struct rtree_node *node;
966
967 list_for_each_entry(node, &zone->nodes, list)
968 recycle_safe_page(node->data);
969
970 list_for_each_entry(node, &zone->leaves, list)
971 recycle_safe_page(node->data);
972}
973
974static void memory_bm_recycle(struct memory_bitmap *bm)
975{
976 struct mem_zone_bm_rtree *zone;
977 struct linked_page *p_list;
978
979 list_for_each_entry(zone, &bm->zones, list)
980 recycle_zone_bm_rtree(zone);
981
982 p_list = bm->p_list;
983 while (p_list) {
984 struct linked_page *lp = p_list;
985
986 p_list = lp->next;
987 recycle_safe_page(lp);
988 }
989}
990
991/**
992 * register_nosave_region - Register a region of unsaveable memory.
993 *
994 * Register a range of page frames the contents of which should not be saved
995 * during hibernation (to be used in the early initialization code).
996 */
997void __init register_nosave_region(unsigned long start_pfn, unsigned long end_pfn)
998{
999 struct nosave_region *region;
1000
1001 if (start_pfn >= end_pfn)
1002 return;
1003
1004 if (!list_empty(&nosave_regions)) {
1005 /* Try to extend the previous region (they should be sorted) */
1006 region = list_entry(nosave_regions.prev,
1007 struct nosave_region, list);
1008 if (region->end_pfn == start_pfn) {
1009 region->end_pfn = end_pfn;
1010 goto Report;
1011 }
1012 }
1013 /* This allocation cannot fail */
1014 region = memblock_alloc(sizeof(struct nosave_region),
1015 SMP_CACHE_BYTES);
1016 if (!region)
1017 panic("%s: Failed to allocate %zu bytes\n", __func__,
1018 sizeof(struct nosave_region));
1019 region->start_pfn = start_pfn;
1020 region->end_pfn = end_pfn;
1021 list_add_tail(®ion->list, &nosave_regions);
1022 Report:
1023 pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
1024 (unsigned long long) start_pfn << PAGE_SHIFT,
1025 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1026}
1027
1028/*
1029 * Set bits in this map correspond to the page frames the contents of which
1030 * should not be saved during the suspend.
1031 */
1032static struct memory_bitmap *forbidden_pages_map;
1033
1034/* Set bits in this map correspond to free page frames. */
1035static struct memory_bitmap *free_pages_map;
1036
1037/*
1038 * Each page frame allocated for creating the image is marked by setting the
1039 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
1040 */
1041
1042void swsusp_set_page_free(struct page *page)
1043{
1044 if (free_pages_map)
1045 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
1046}
1047
1048static int swsusp_page_is_free(struct page *page)
1049{
1050 return free_pages_map ?
1051 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1052}
1053
1054void swsusp_unset_page_free(struct page *page)
1055{
1056 if (free_pages_map)
1057 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1058}
1059
1060static void swsusp_set_page_forbidden(struct page *page)
1061{
1062 if (forbidden_pages_map)
1063 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1064}
1065
1066int swsusp_page_is_forbidden(struct page *page)
1067{
1068 return forbidden_pages_map ?
1069 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1070}
1071
1072static void swsusp_unset_page_forbidden(struct page *page)
1073{
1074 if (forbidden_pages_map)
1075 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1076}
1077
1078/**
1079 * mark_nosave_pages - Mark pages that should not be saved.
1080 * @bm: Memory bitmap.
1081 *
1082 * Set the bits in @bm that correspond to the page frames the contents of which
1083 * should not be saved.
1084 */
1085static void mark_nosave_pages(struct memory_bitmap *bm)
1086{
1087 struct nosave_region *region;
1088
1089 if (list_empty(&nosave_regions))
1090 return;
1091
1092 list_for_each_entry(region, &nosave_regions, list) {
1093 unsigned long pfn;
1094
1095 pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1096 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1097 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1098 - 1);
1099
1100 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1101 if (pfn_valid(pfn)) {
1102 /*
1103 * It is safe to ignore the result of
1104 * mem_bm_set_bit_check() here, since we won't
1105 * touch the PFNs for which the error is
1106 * returned anyway.
1107 */
1108 mem_bm_set_bit_check(bm, pfn);
1109 }
1110 }
1111}
1112
1113/**
1114 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1115 *
1116 * Create bitmaps needed for marking page frames that should not be saved and
1117 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1118 * only modified if everything goes well, because we don't want the bits to be
1119 * touched before both bitmaps are set up.
1120 */
1121int create_basic_memory_bitmaps(void)
1122{
1123 struct memory_bitmap *bm1, *bm2;
1124 int error;
1125
1126 if (forbidden_pages_map && free_pages_map)
1127 return 0;
1128 else
1129 BUG_ON(forbidden_pages_map || free_pages_map);
1130
1131 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1132 if (!bm1)
1133 return -ENOMEM;
1134
1135 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1136 if (error)
1137 goto Free_first_object;
1138
1139 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1140 if (!bm2)
1141 goto Free_first_bitmap;
1142
1143 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1144 if (error)
1145 goto Free_second_object;
1146
1147 forbidden_pages_map = bm1;
1148 free_pages_map = bm2;
1149 mark_nosave_pages(forbidden_pages_map);
1150
1151 pr_debug("Basic memory bitmaps created\n");
1152
1153 return 0;
1154
1155 Free_second_object:
1156 kfree(bm2);
1157 Free_first_bitmap:
1158 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1159 Free_first_object:
1160 kfree(bm1);
1161 return -ENOMEM;
1162}
1163
1164/**
1165 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1166 *
1167 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1168 * auxiliary pointers are necessary so that the bitmaps themselves are not
1169 * referred to while they are being freed.
1170 */
1171void free_basic_memory_bitmaps(void)
1172{
1173 struct memory_bitmap *bm1, *bm2;
1174
1175 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1176 return;
1177
1178 bm1 = forbidden_pages_map;
1179 bm2 = free_pages_map;
1180 forbidden_pages_map = NULL;
1181 free_pages_map = NULL;
1182 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1183 kfree(bm1);
1184 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1185 kfree(bm2);
1186
1187 pr_debug("Basic memory bitmaps freed\n");
1188}
1189
1190static void clear_or_poison_free_page(struct page *page)
1191{
1192 if (page_poisoning_enabled_static())
1193 __kernel_poison_pages(page, 1);
1194 else if (want_init_on_free())
1195 clear_highpage(page);
1196}
1197
1198void clear_or_poison_free_pages(void)
1199{
1200 struct memory_bitmap *bm = free_pages_map;
1201 unsigned long pfn;
1202
1203 if (WARN_ON(!(free_pages_map)))
1204 return;
1205
1206 if (page_poisoning_enabled() || want_init_on_free()) {
1207 memory_bm_position_reset(bm);
1208 pfn = memory_bm_next_pfn(bm);
1209 while (pfn != BM_END_OF_MAP) {
1210 if (pfn_valid(pfn))
1211 clear_or_poison_free_page(pfn_to_page(pfn));
1212
1213 pfn = memory_bm_next_pfn(bm);
1214 }
1215 memory_bm_position_reset(bm);
1216 pr_info("free pages cleared after restore\n");
1217 }
1218}
1219
1220/**
1221 * snapshot_additional_pages - Estimate the number of extra pages needed.
1222 * @zone: Memory zone to carry out the computation for.
1223 *
1224 * Estimate the number of additional pages needed for setting up a hibernation
1225 * image data structures for @zone (usually, the returned value is greater than
1226 * the exact number).
1227 */
1228unsigned int snapshot_additional_pages(struct zone *zone)
1229{
1230 unsigned int rtree, nodes;
1231
1232 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1233 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1234 LINKED_PAGE_DATA_SIZE);
1235 while (nodes > 1) {
1236 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1237 rtree += nodes;
1238 }
1239
1240 return 2 * rtree;
1241}
1242
1243/*
1244 * Touch the watchdog for every WD_PAGE_COUNT pages.
1245 */
1246#define WD_PAGE_COUNT (128*1024)
1247
1248static void mark_free_pages(struct zone *zone)
1249{
1250 unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
1251 unsigned long flags;
1252 unsigned int order, t;
1253 struct page *page;
1254
1255 if (zone_is_empty(zone))
1256 return;
1257
1258 spin_lock_irqsave(&zone->lock, flags);
1259
1260 max_zone_pfn = zone_end_pfn(zone);
1261 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1262 if (pfn_valid(pfn)) {
1263 page = pfn_to_page(pfn);
1264
1265 if (!--page_count) {
1266 touch_nmi_watchdog();
1267 page_count = WD_PAGE_COUNT;
1268 }
1269
1270 if (page_zone(page) != zone)
1271 continue;
1272
1273 if (!swsusp_page_is_forbidden(page))
1274 swsusp_unset_page_free(page);
1275 }
1276
1277 for_each_migratetype_order(order, t) {
1278 list_for_each_entry(page,
1279 &zone->free_area[order].free_list[t], buddy_list) {
1280 unsigned long i;
1281
1282 pfn = page_to_pfn(page);
1283 for (i = 0; i < (1UL << order); i++) {
1284 if (!--page_count) {
1285 touch_nmi_watchdog();
1286 page_count = WD_PAGE_COUNT;
1287 }
1288 swsusp_set_page_free(pfn_to_page(pfn + i));
1289 }
1290 }
1291 }
1292 spin_unlock_irqrestore(&zone->lock, flags);
1293}
1294
1295#ifdef CONFIG_HIGHMEM
1296/**
1297 * count_free_highmem_pages - Compute the total number of free highmem pages.
1298 *
1299 * The returned number is system-wide.
1300 */
1301static unsigned int count_free_highmem_pages(void)
1302{
1303 struct zone *zone;
1304 unsigned int cnt = 0;
1305
1306 for_each_populated_zone(zone)
1307 if (is_highmem(zone))
1308 cnt += zone_page_state(zone, NR_FREE_PAGES);
1309
1310 return cnt;
1311}
1312
1313/**
1314 * saveable_highmem_page - Check if a highmem page is saveable.
1315 *
1316 * Determine whether a highmem page should be included in a hibernation image.
1317 *
1318 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1319 * and it isn't part of a free chunk of pages.
1320 */
1321static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1322{
1323 struct page *page;
1324
1325 if (!pfn_valid(pfn))
1326 return NULL;
1327
1328 page = pfn_to_online_page(pfn);
1329 if (!page || page_zone(page) != zone)
1330 return NULL;
1331
1332 BUG_ON(!PageHighMem(page));
1333
1334 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1335 return NULL;
1336
1337 if (PageReserved(page) || PageOffline(page))
1338 return NULL;
1339
1340 if (page_is_guard(page))
1341 return NULL;
1342
1343 return page;
1344}
1345
1346/**
1347 * count_highmem_pages - Compute the total number of saveable highmem pages.
1348 */
1349static unsigned int count_highmem_pages(void)
1350{
1351 struct zone *zone;
1352 unsigned int n = 0;
1353
1354 for_each_populated_zone(zone) {
1355 unsigned long pfn, max_zone_pfn;
1356
1357 if (!is_highmem(zone))
1358 continue;
1359
1360 mark_free_pages(zone);
1361 max_zone_pfn = zone_end_pfn(zone);
1362 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1363 if (saveable_highmem_page(zone, pfn))
1364 n++;
1365 }
1366 return n;
1367}
1368#else
1369static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1370{
1371 return NULL;
1372}
1373#endif /* CONFIG_HIGHMEM */
1374
1375/**
1376 * saveable_page - Check if the given page is saveable.
1377 *
1378 * Determine whether a non-highmem page should be included in a hibernation
1379 * image.
1380 *
1381 * We should save the page if it isn't Nosave, and is not in the range
1382 * of pages statically defined as 'unsaveable', and it isn't part of
1383 * a free chunk of pages.
1384 */
1385static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1386{
1387 struct page *page;
1388
1389 if (!pfn_valid(pfn))
1390 return NULL;
1391
1392 page = pfn_to_online_page(pfn);
1393 if (!page || page_zone(page) != zone)
1394 return NULL;
1395
1396 BUG_ON(PageHighMem(page));
1397
1398 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1399 return NULL;
1400
1401 if (PageOffline(page))
1402 return NULL;
1403
1404 if (PageReserved(page)
1405 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1406 return NULL;
1407
1408 if (page_is_guard(page))
1409 return NULL;
1410
1411 return page;
1412}
1413
1414/**
1415 * count_data_pages - Compute the total number of saveable non-highmem pages.
1416 */
1417static unsigned int count_data_pages(void)
1418{
1419 struct zone *zone;
1420 unsigned long pfn, max_zone_pfn;
1421 unsigned int n = 0;
1422
1423 for_each_populated_zone(zone) {
1424 if (is_highmem(zone))
1425 continue;
1426
1427 mark_free_pages(zone);
1428 max_zone_pfn = zone_end_pfn(zone);
1429 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1430 if (saveable_page(zone, pfn))
1431 n++;
1432 }
1433 return n;
1434}
1435
1436/*
1437 * This is needed, because copy_page and memcpy are not usable for copying
1438 * task structs. Returns true if the page was filled with only zeros,
1439 * otherwise false.
1440 */
1441static inline bool do_copy_page(long *dst, long *src)
1442{
1443 long z = 0;
1444 int n;
1445
1446 for (n = PAGE_SIZE / sizeof(long); n; n--) {
1447 z |= *src;
1448 *dst++ = *src++;
1449 }
1450 return !z;
1451}
1452
1453/**
1454 * safe_copy_page - Copy a page in a safe way.
1455 *
1456 * Check if the page we are going to copy is marked as present in the kernel
1457 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1458 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1459 * always returns 'true'. Returns true if the page was entirely composed of
1460 * zeros, otherwise it will return false.
1461 */
1462static bool safe_copy_page(void *dst, struct page *s_page)
1463{
1464 bool zeros_only;
1465
1466 if (kernel_page_present(s_page)) {
1467 zeros_only = do_copy_page(dst, page_address(s_page));
1468 } else {
1469 hibernate_map_page(s_page);
1470 zeros_only = do_copy_page(dst, page_address(s_page));
1471 hibernate_unmap_page(s_page);
1472 }
1473 return zeros_only;
1474}
1475
1476#ifdef CONFIG_HIGHMEM
1477static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1478{
1479 return is_highmem(zone) ?
1480 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1481}
1482
1483static bool copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1484{
1485 struct page *s_page, *d_page;
1486 void *src, *dst;
1487 bool zeros_only;
1488
1489 s_page = pfn_to_page(src_pfn);
1490 d_page = pfn_to_page(dst_pfn);
1491 if (PageHighMem(s_page)) {
1492 src = kmap_local_page(s_page);
1493 dst = kmap_local_page(d_page);
1494 zeros_only = do_copy_page(dst, src);
1495 kunmap_local(dst);
1496 kunmap_local(src);
1497 } else {
1498 if (PageHighMem(d_page)) {
1499 /*
1500 * The page pointed to by src may contain some kernel
1501 * data modified by kmap_atomic()
1502 */
1503 zeros_only = safe_copy_page(buffer, s_page);
1504 dst = kmap_local_page(d_page);
1505 copy_page(dst, buffer);
1506 kunmap_local(dst);
1507 } else {
1508 zeros_only = safe_copy_page(page_address(d_page), s_page);
1509 }
1510 }
1511 return zeros_only;
1512}
1513#else
1514#define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1515
1516static inline int copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1517{
1518 return safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1519 pfn_to_page(src_pfn));
1520}
1521#endif /* CONFIG_HIGHMEM */
1522
1523/*
1524 * Copy data pages will copy all pages into pages pulled from the copy_bm.
1525 * If a page was entirely filled with zeros it will be marked in the zero_bm.
1526 *
1527 * Returns the number of pages copied.
1528 */
1529static unsigned long copy_data_pages(struct memory_bitmap *copy_bm,
1530 struct memory_bitmap *orig_bm,
1531 struct memory_bitmap *zero_bm)
1532{
1533 unsigned long copied_pages = 0;
1534 struct zone *zone;
1535 unsigned long pfn, copy_pfn;
1536
1537 for_each_populated_zone(zone) {
1538 unsigned long max_zone_pfn;
1539
1540 mark_free_pages(zone);
1541 max_zone_pfn = zone_end_pfn(zone);
1542 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1543 if (page_is_saveable(zone, pfn))
1544 memory_bm_set_bit(orig_bm, pfn);
1545 }
1546 memory_bm_position_reset(orig_bm);
1547 memory_bm_position_reset(copy_bm);
1548 copy_pfn = memory_bm_next_pfn(copy_bm);
1549 for(;;) {
1550 pfn = memory_bm_next_pfn(orig_bm);
1551 if (unlikely(pfn == BM_END_OF_MAP))
1552 break;
1553 if (copy_data_page(copy_pfn, pfn)) {
1554 memory_bm_set_bit(zero_bm, pfn);
1555 /* Use this copy_pfn for a page that is not full of zeros */
1556 continue;
1557 }
1558 copied_pages++;
1559 copy_pfn = memory_bm_next_pfn(copy_bm);
1560 }
1561 return copied_pages;
1562}
1563
1564/* Total number of image pages */
1565static unsigned int nr_copy_pages;
1566/* Number of pages needed for saving the original pfns of the image pages */
1567static unsigned int nr_meta_pages;
1568/* Number of zero pages */
1569static unsigned int nr_zero_pages;
1570
1571/*
1572 * Numbers of normal and highmem page frames allocated for hibernation image
1573 * before suspending devices.
1574 */
1575static unsigned int alloc_normal, alloc_highmem;
1576/*
1577 * Memory bitmap used for marking saveable pages (during hibernation) or
1578 * hibernation image pages (during restore)
1579 */
1580static struct memory_bitmap orig_bm;
1581/*
1582 * Memory bitmap used during hibernation for marking allocated page frames that
1583 * will contain copies of saveable pages. During restore it is initially used
1584 * for marking hibernation image pages, but then the set bits from it are
1585 * duplicated in @orig_bm and it is released. On highmem systems it is next
1586 * used for marking "safe" highmem pages, but it has to be reinitialized for
1587 * this purpose.
1588 */
1589static struct memory_bitmap copy_bm;
1590
1591/* Memory bitmap which tracks which saveable pages were zero filled. */
1592static struct memory_bitmap zero_bm;
1593
1594/**
1595 * swsusp_free - Free pages allocated for hibernation image.
1596 *
1597 * Image pages are allocated before snapshot creation, so they need to be
1598 * released after resume.
1599 */
1600void swsusp_free(void)
1601{
1602 unsigned long fb_pfn, fr_pfn;
1603
1604 if (!forbidden_pages_map || !free_pages_map)
1605 goto out;
1606
1607 memory_bm_position_reset(forbidden_pages_map);
1608 memory_bm_position_reset(free_pages_map);
1609
1610loop:
1611 fr_pfn = memory_bm_next_pfn(free_pages_map);
1612 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1613
1614 /*
1615 * Find the next bit set in both bitmaps. This is guaranteed to
1616 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1617 */
1618 do {
1619 if (fb_pfn < fr_pfn)
1620 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1621 if (fr_pfn < fb_pfn)
1622 fr_pfn = memory_bm_next_pfn(free_pages_map);
1623 } while (fb_pfn != fr_pfn);
1624
1625 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1626 struct page *page = pfn_to_page(fr_pfn);
1627
1628 memory_bm_clear_current(forbidden_pages_map);
1629 memory_bm_clear_current(free_pages_map);
1630 hibernate_restore_unprotect_page(page_address(page));
1631 __free_page(page);
1632 goto loop;
1633 }
1634
1635out:
1636 nr_copy_pages = 0;
1637 nr_meta_pages = 0;
1638 nr_zero_pages = 0;
1639 restore_pblist = NULL;
1640 buffer = NULL;
1641 alloc_normal = 0;
1642 alloc_highmem = 0;
1643 hibernate_restore_protection_end();
1644}
1645
1646/* Helper functions used for the shrinking of memory. */
1647
1648#define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1649
1650/**
1651 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1652 * @nr_pages: Number of page frames to allocate.
1653 * @mask: GFP flags to use for the allocation.
1654 *
1655 * Return value: Number of page frames actually allocated
1656 */
1657static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1658{
1659 unsigned long nr_alloc = 0;
1660
1661 while (nr_pages > 0) {
1662 struct page *page;
1663
1664 page = alloc_image_page(mask);
1665 if (!page)
1666 break;
1667 memory_bm_set_bit(©_bm, page_to_pfn(page));
1668 if (PageHighMem(page))
1669 alloc_highmem++;
1670 else
1671 alloc_normal++;
1672 nr_pages--;
1673 nr_alloc++;
1674 }
1675
1676 return nr_alloc;
1677}
1678
1679static unsigned long preallocate_image_memory(unsigned long nr_pages,
1680 unsigned long avail_normal)
1681{
1682 unsigned long alloc;
1683
1684 if (avail_normal <= alloc_normal)
1685 return 0;
1686
1687 alloc = avail_normal - alloc_normal;
1688 if (nr_pages < alloc)
1689 alloc = nr_pages;
1690
1691 return preallocate_image_pages(alloc, GFP_IMAGE);
1692}
1693
1694#ifdef CONFIG_HIGHMEM
1695static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1696{
1697 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1698}
1699
1700/**
1701 * __fraction - Compute (an approximation of) x * (multiplier / base).
1702 */
1703static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1704{
1705 return div64_u64(x * multiplier, base);
1706}
1707
1708static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1709 unsigned long highmem,
1710 unsigned long total)
1711{
1712 unsigned long alloc = __fraction(nr_pages, highmem, total);
1713
1714 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1715}
1716#else /* CONFIG_HIGHMEM */
1717static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1718{
1719 return 0;
1720}
1721
1722static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1723 unsigned long highmem,
1724 unsigned long total)
1725{
1726 return 0;
1727}
1728#endif /* CONFIG_HIGHMEM */
1729
1730/**
1731 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1732 */
1733static unsigned long free_unnecessary_pages(void)
1734{
1735 unsigned long save, to_free_normal, to_free_highmem, free;
1736
1737 save = count_data_pages();
1738 if (alloc_normal >= save) {
1739 to_free_normal = alloc_normal - save;
1740 save = 0;
1741 } else {
1742 to_free_normal = 0;
1743 save -= alloc_normal;
1744 }
1745 save += count_highmem_pages();
1746 if (alloc_highmem >= save) {
1747 to_free_highmem = alloc_highmem - save;
1748 } else {
1749 to_free_highmem = 0;
1750 save -= alloc_highmem;
1751 if (to_free_normal > save)
1752 to_free_normal -= save;
1753 else
1754 to_free_normal = 0;
1755 }
1756 free = to_free_normal + to_free_highmem;
1757
1758 memory_bm_position_reset(©_bm);
1759
1760 while (to_free_normal > 0 || to_free_highmem > 0) {
1761 unsigned long pfn = memory_bm_next_pfn(©_bm);
1762 struct page *page = pfn_to_page(pfn);
1763
1764 if (PageHighMem(page)) {
1765 if (!to_free_highmem)
1766 continue;
1767 to_free_highmem--;
1768 alloc_highmem--;
1769 } else {
1770 if (!to_free_normal)
1771 continue;
1772 to_free_normal--;
1773 alloc_normal--;
1774 }
1775 memory_bm_clear_bit(©_bm, pfn);
1776 swsusp_unset_page_forbidden(page);
1777 swsusp_unset_page_free(page);
1778 __free_page(page);
1779 }
1780
1781 return free;
1782}
1783
1784/**
1785 * minimum_image_size - Estimate the minimum acceptable size of an image.
1786 * @saveable: Number of saveable pages in the system.
1787 *
1788 * We want to avoid attempting to free too much memory too hard, so estimate the
1789 * minimum acceptable size of a hibernation image to use as the lower limit for
1790 * preallocating memory.
1791 *
1792 * We assume that the minimum image size should be proportional to
1793 *
1794 * [number of saveable pages] - [number of pages that can be freed in theory]
1795 *
1796 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1797 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1798 */
1799static unsigned long minimum_image_size(unsigned long saveable)
1800{
1801 unsigned long size;
1802
1803 size = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B)
1804 + global_node_page_state(NR_ACTIVE_ANON)
1805 + global_node_page_state(NR_INACTIVE_ANON)
1806 + global_node_page_state(NR_ACTIVE_FILE)
1807 + global_node_page_state(NR_INACTIVE_FILE);
1808
1809 return saveable <= size ? 0 : saveable - size;
1810}
1811
1812/**
1813 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1814 *
1815 * To create a hibernation image it is necessary to make a copy of every page
1816 * frame in use. We also need a number of page frames to be free during
1817 * hibernation for allocations made while saving the image and for device
1818 * drivers, in case they need to allocate memory from their hibernation
1819 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1820 * estimate) and reserved_size divided by PAGE_SIZE (which is tunable through
1821 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1822 * total number of available page frames and allocate at least
1823 *
1824 * ([page frames total] - PAGES_FOR_IO - [metadata pages]) / 2
1825 * - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1826 *
1827 * of them, which corresponds to the maximum size of a hibernation image.
1828 *
1829 * If image_size is set below the number following from the above formula,
1830 * the preallocation of memory is continued until the total number of saveable
1831 * pages in the system is below the requested image size or the minimum
1832 * acceptable image size returned by minimum_image_size(), whichever is greater.
1833 */
1834int hibernate_preallocate_memory(void)
1835{
1836 struct zone *zone;
1837 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1838 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1839 ktime_t start, stop;
1840 int error;
1841
1842 pr_info("Preallocating image memory\n");
1843 start = ktime_get();
1844
1845 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1846 if (error) {
1847 pr_err("Cannot allocate original bitmap\n");
1848 goto err_out;
1849 }
1850
1851 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1852 if (error) {
1853 pr_err("Cannot allocate copy bitmap\n");
1854 goto err_out;
1855 }
1856
1857 error = memory_bm_create(&zero_bm, GFP_IMAGE, PG_ANY);
1858 if (error) {
1859 pr_err("Cannot allocate zero bitmap\n");
1860 goto err_out;
1861 }
1862
1863 alloc_normal = 0;
1864 alloc_highmem = 0;
1865 nr_zero_pages = 0;
1866
1867 /* Count the number of saveable data pages. */
1868 save_highmem = count_highmem_pages();
1869 saveable = count_data_pages();
1870
1871 /*
1872 * Compute the total number of page frames we can use (count) and the
1873 * number of pages needed for image metadata (size).
1874 */
1875 count = saveable;
1876 saveable += save_highmem;
1877 highmem = save_highmem;
1878 size = 0;
1879 for_each_populated_zone(zone) {
1880 size += snapshot_additional_pages(zone);
1881 if (is_highmem(zone))
1882 highmem += zone_page_state(zone, NR_FREE_PAGES);
1883 else
1884 count += zone_page_state(zone, NR_FREE_PAGES);
1885 }
1886 avail_normal = count;
1887 count += highmem;
1888 count -= totalreserve_pages;
1889
1890 /* Compute the maximum number of saveable pages to leave in memory. */
1891 max_size = (count - (size + PAGES_FOR_IO)) / 2
1892 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1893 /* Compute the desired number of image pages specified by image_size. */
1894 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1895 if (size > max_size)
1896 size = max_size;
1897 /*
1898 * If the desired number of image pages is at least as large as the
1899 * current number of saveable pages in memory, allocate page frames for
1900 * the image and we're done.
1901 */
1902 if (size >= saveable) {
1903 pages = preallocate_image_highmem(save_highmem);
1904 pages += preallocate_image_memory(saveable - pages, avail_normal);
1905 goto out;
1906 }
1907
1908 /* Estimate the minimum size of the image. */
1909 pages = minimum_image_size(saveable);
1910 /*
1911 * To avoid excessive pressure on the normal zone, leave room in it to
1912 * accommodate an image of the minimum size (unless it's already too
1913 * small, in which case don't preallocate pages from it at all).
1914 */
1915 if (avail_normal > pages)
1916 avail_normal -= pages;
1917 else
1918 avail_normal = 0;
1919 if (size < pages)
1920 size = min_t(unsigned long, pages, max_size);
1921
1922 /*
1923 * Let the memory management subsystem know that we're going to need a
1924 * large number of page frames to allocate and make it free some memory.
1925 * NOTE: If this is not done, performance will be hurt badly in some
1926 * test cases.
1927 */
1928 shrink_all_memory(saveable - size);
1929
1930 /*
1931 * The number of saveable pages in memory was too high, so apply some
1932 * pressure to decrease it. First, make room for the largest possible
1933 * image and fail if that doesn't work. Next, try to decrease the size
1934 * of the image as much as indicated by 'size' using allocations from
1935 * highmem and non-highmem zones separately.
1936 */
1937 pages_highmem = preallocate_image_highmem(highmem / 2);
1938 alloc = count - max_size;
1939 if (alloc > pages_highmem)
1940 alloc -= pages_highmem;
1941 else
1942 alloc = 0;
1943 pages = preallocate_image_memory(alloc, avail_normal);
1944 if (pages < alloc) {
1945 /* We have exhausted non-highmem pages, try highmem. */
1946 alloc -= pages;
1947 pages += pages_highmem;
1948 pages_highmem = preallocate_image_highmem(alloc);
1949 if (pages_highmem < alloc) {
1950 pr_err("Image allocation is %lu pages short\n",
1951 alloc - pages_highmem);
1952 goto err_out;
1953 }
1954 pages += pages_highmem;
1955 /*
1956 * size is the desired number of saveable pages to leave in
1957 * memory, so try to preallocate (all memory - size) pages.
1958 */
1959 alloc = (count - pages) - size;
1960 pages += preallocate_image_highmem(alloc);
1961 } else {
1962 /*
1963 * There are approximately max_size saveable pages at this point
1964 * and we want to reduce this number down to size.
1965 */
1966 alloc = max_size - size;
1967 size = preallocate_highmem_fraction(alloc, highmem, count);
1968 pages_highmem += size;
1969 alloc -= size;
1970 size = preallocate_image_memory(alloc, avail_normal);
1971 pages_highmem += preallocate_image_highmem(alloc - size);
1972 pages += pages_highmem + size;
1973 }
1974
1975 /*
1976 * We only need as many page frames for the image as there are saveable
1977 * pages in memory, but we have allocated more. Release the excessive
1978 * ones now.
1979 */
1980 pages -= free_unnecessary_pages();
1981
1982 out:
1983 stop = ktime_get();
1984 pr_info("Allocated %lu pages for snapshot\n", pages);
1985 swsusp_show_speed(start, stop, pages, "Allocated");
1986
1987 return 0;
1988
1989 err_out:
1990 swsusp_free();
1991 return -ENOMEM;
1992}
1993
1994#ifdef CONFIG_HIGHMEM
1995/**
1996 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1997 *
1998 * Compute the number of non-highmem pages that will be necessary for creating
1999 * copies of highmem pages.
2000 */
2001static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
2002{
2003 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
2004
2005 if (free_highmem >= nr_highmem)
2006 nr_highmem = 0;
2007 else
2008 nr_highmem -= free_highmem;
2009
2010 return nr_highmem;
2011}
2012#else
2013static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
2014#endif /* CONFIG_HIGHMEM */
2015
2016/**
2017 * enough_free_mem - Check if there is enough free memory for the image.
2018 */
2019static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
2020{
2021 struct zone *zone;
2022 unsigned int free = alloc_normal;
2023
2024 for_each_populated_zone(zone)
2025 if (!is_highmem(zone))
2026 free += zone_page_state(zone, NR_FREE_PAGES);
2027
2028 nr_pages += count_pages_for_highmem(nr_highmem);
2029 pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
2030 nr_pages, PAGES_FOR_IO, free);
2031
2032 return free > nr_pages + PAGES_FOR_IO;
2033}
2034
2035#ifdef CONFIG_HIGHMEM
2036/**
2037 * get_highmem_buffer - Allocate a buffer for highmem pages.
2038 *
2039 * If there are some highmem pages in the hibernation image, we may need a
2040 * buffer to copy them and/or load their data.
2041 */
2042static inline int get_highmem_buffer(int safe_needed)
2043{
2044 buffer = get_image_page(GFP_ATOMIC, safe_needed);
2045 return buffer ? 0 : -ENOMEM;
2046}
2047
2048/**
2049 * alloc_highmem_pages - Allocate some highmem pages for the image.
2050 *
2051 * Try to allocate as many pages as needed, but if the number of free highmem
2052 * pages is less than that, allocate them all.
2053 */
2054static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
2055 unsigned int nr_highmem)
2056{
2057 unsigned int to_alloc = count_free_highmem_pages();
2058
2059 if (to_alloc > nr_highmem)
2060 to_alloc = nr_highmem;
2061
2062 nr_highmem -= to_alloc;
2063 while (to_alloc-- > 0) {
2064 struct page *page;
2065
2066 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
2067 memory_bm_set_bit(bm, page_to_pfn(page));
2068 }
2069 return nr_highmem;
2070}
2071#else
2072static inline int get_highmem_buffer(int safe_needed) { return 0; }
2073
2074static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
2075 unsigned int n) { return 0; }
2076#endif /* CONFIG_HIGHMEM */
2077
2078/**
2079 * swsusp_alloc - Allocate memory for hibernation image.
2080 *
2081 * We first try to allocate as many highmem pages as there are
2082 * saveable highmem pages in the system. If that fails, we allocate
2083 * non-highmem pages for the copies of the remaining highmem ones.
2084 *
2085 * In this approach it is likely that the copies of highmem pages will
2086 * also be located in the high memory, because of the way in which
2087 * copy_data_pages() works.
2088 */
2089static int swsusp_alloc(struct memory_bitmap *copy_bm,
2090 unsigned int nr_pages, unsigned int nr_highmem)
2091{
2092 if (nr_highmem > 0) {
2093 if (get_highmem_buffer(PG_ANY))
2094 goto err_out;
2095 if (nr_highmem > alloc_highmem) {
2096 nr_highmem -= alloc_highmem;
2097 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
2098 }
2099 }
2100 if (nr_pages > alloc_normal) {
2101 nr_pages -= alloc_normal;
2102 while (nr_pages-- > 0) {
2103 struct page *page;
2104
2105 page = alloc_image_page(GFP_ATOMIC);
2106 if (!page)
2107 goto err_out;
2108 memory_bm_set_bit(copy_bm, page_to_pfn(page));
2109 }
2110 }
2111
2112 return 0;
2113
2114 err_out:
2115 swsusp_free();
2116 return -ENOMEM;
2117}
2118
2119asmlinkage __visible int swsusp_save(void)
2120{
2121 unsigned int nr_pages, nr_highmem;
2122
2123 pr_info("Creating image:\n");
2124
2125 drain_local_pages(NULL);
2126 nr_pages = count_data_pages();
2127 nr_highmem = count_highmem_pages();
2128 pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
2129
2130 if (!enough_free_mem(nr_pages, nr_highmem)) {
2131 pr_err("Not enough free memory\n");
2132 return -ENOMEM;
2133 }
2134
2135 if (swsusp_alloc(©_bm, nr_pages, nr_highmem)) {
2136 pr_err("Memory allocation failed\n");
2137 return -ENOMEM;
2138 }
2139
2140 /*
2141 * During allocating of suspend pagedir, new cold pages may appear.
2142 * Kill them.
2143 */
2144 drain_local_pages(NULL);
2145 nr_copy_pages = copy_data_pages(©_bm, &orig_bm, &zero_bm);
2146
2147 /*
2148 * End of critical section. From now on, we can write to memory,
2149 * but we should not touch disk. This specially means we must _not_
2150 * touch swap space! Except we must write out our image of course.
2151 */
2152 nr_pages += nr_highmem;
2153 /* We don't actually copy the zero pages */
2154 nr_zero_pages = nr_pages - nr_copy_pages;
2155 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2156
2157 pr_info("Image created (%d pages copied, %d zero pages)\n", nr_copy_pages, nr_zero_pages);
2158
2159 return 0;
2160}
2161
2162#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2163static int init_header_complete(struct swsusp_info *info)
2164{
2165 memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2166 info->version_code = LINUX_VERSION_CODE;
2167 return 0;
2168}
2169
2170static const char *check_image_kernel(struct swsusp_info *info)
2171{
2172 if (info->version_code != LINUX_VERSION_CODE)
2173 return "kernel version";
2174 if (strcmp(info->uts.sysname,init_utsname()->sysname))
2175 return "system type";
2176 if (strcmp(info->uts.release,init_utsname()->release))
2177 return "kernel release";
2178 if (strcmp(info->uts.version,init_utsname()->version))
2179 return "version";
2180 if (strcmp(info->uts.machine,init_utsname()->machine))
2181 return "machine";
2182 return NULL;
2183}
2184#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2185
2186unsigned long snapshot_get_image_size(void)
2187{
2188 return nr_copy_pages + nr_meta_pages + 1;
2189}
2190
2191static int init_header(struct swsusp_info *info)
2192{
2193 memset(info, 0, sizeof(struct swsusp_info));
2194 info->num_physpages = get_num_physpages();
2195 info->image_pages = nr_copy_pages;
2196 info->pages = snapshot_get_image_size();
2197 info->size = info->pages;
2198 info->size <<= PAGE_SHIFT;
2199 return init_header_complete(info);
2200}
2201
2202#define ENCODED_PFN_ZERO_FLAG ((unsigned long)1 << (BITS_PER_LONG - 1))
2203#define ENCODED_PFN_MASK (~ENCODED_PFN_ZERO_FLAG)
2204
2205/**
2206 * pack_pfns - Prepare PFNs for saving.
2207 * @bm: Memory bitmap.
2208 * @buf: Memory buffer to store the PFNs in.
2209 * @zero_bm: Memory bitmap containing PFNs of zero pages.
2210 *
2211 * PFNs corresponding to set bits in @bm are stored in the area of memory
2212 * pointed to by @buf (1 page at a time). Pages which were filled with only
2213 * zeros will have the highest bit set in the packed format to distinguish
2214 * them from PFNs which will be contained in the image file.
2215 */
2216static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm,
2217 struct memory_bitmap *zero_bm)
2218{
2219 int j;
2220
2221 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2222 buf[j] = memory_bm_next_pfn(bm);
2223 if (unlikely(buf[j] == BM_END_OF_MAP))
2224 break;
2225 if (memory_bm_test_bit(zero_bm, buf[j]))
2226 buf[j] |= ENCODED_PFN_ZERO_FLAG;
2227 }
2228}
2229
2230/**
2231 * snapshot_read_next - Get the address to read the next image page from.
2232 * @handle: Snapshot handle to be used for the reading.
2233 *
2234 * On the first call, @handle should point to a zeroed snapshot_handle
2235 * structure. The structure gets populated then and a pointer to it should be
2236 * passed to this function every next time.
2237 *
2238 * On success, the function returns a positive number. Then, the caller
2239 * is allowed to read up to the returned number of bytes from the memory
2240 * location computed by the data_of() macro.
2241 *
2242 * The function returns 0 to indicate the end of the data stream condition,
2243 * and negative numbers are returned on errors. If that happens, the structure
2244 * pointed to by @handle is not updated and should not be used any more.
2245 */
2246int snapshot_read_next(struct snapshot_handle *handle)
2247{
2248 if (handle->cur > nr_meta_pages + nr_copy_pages)
2249 return 0;
2250
2251 if (!buffer) {
2252 /* This makes the buffer be freed by swsusp_free() */
2253 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2254 if (!buffer)
2255 return -ENOMEM;
2256 }
2257 if (!handle->cur) {
2258 int error;
2259
2260 error = init_header((struct swsusp_info *)buffer);
2261 if (error)
2262 return error;
2263 handle->buffer = buffer;
2264 memory_bm_position_reset(&orig_bm);
2265 memory_bm_position_reset(©_bm);
2266 } else if (handle->cur <= nr_meta_pages) {
2267 clear_page(buffer);
2268 pack_pfns(buffer, &orig_bm, &zero_bm);
2269 } else {
2270 struct page *page;
2271
2272 page = pfn_to_page(memory_bm_next_pfn(©_bm));
2273 if (PageHighMem(page)) {
2274 /*
2275 * Highmem pages are copied to the buffer,
2276 * because we can't return with a kmapped
2277 * highmem page (we may not be called again).
2278 */
2279 void *kaddr;
2280
2281 kaddr = kmap_atomic(page);
2282 copy_page(buffer, kaddr);
2283 kunmap_atomic(kaddr);
2284 handle->buffer = buffer;
2285 } else {
2286 handle->buffer = page_address(page);
2287 }
2288 }
2289 handle->cur++;
2290 return PAGE_SIZE;
2291}
2292
2293static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2294 struct memory_bitmap *src)
2295{
2296 unsigned long pfn;
2297
2298 memory_bm_position_reset(src);
2299 pfn = memory_bm_next_pfn(src);
2300 while (pfn != BM_END_OF_MAP) {
2301 memory_bm_set_bit(dst, pfn);
2302 pfn = memory_bm_next_pfn(src);
2303 }
2304}
2305
2306/**
2307 * mark_unsafe_pages - Mark pages that were used before hibernation.
2308 *
2309 * Mark the pages that cannot be used for storing the image during restoration,
2310 * because they conflict with the pages that had been used before hibernation.
2311 */
2312static void mark_unsafe_pages(struct memory_bitmap *bm)
2313{
2314 unsigned long pfn;
2315
2316 /* Clear the "free"/"unsafe" bit for all PFNs */
2317 memory_bm_position_reset(free_pages_map);
2318 pfn = memory_bm_next_pfn(free_pages_map);
2319 while (pfn != BM_END_OF_MAP) {
2320 memory_bm_clear_current(free_pages_map);
2321 pfn = memory_bm_next_pfn(free_pages_map);
2322 }
2323
2324 /* Mark pages that correspond to the "original" PFNs as "unsafe" */
2325 duplicate_memory_bitmap(free_pages_map, bm);
2326
2327 allocated_unsafe_pages = 0;
2328}
2329
2330static int check_header(struct swsusp_info *info)
2331{
2332 const char *reason;
2333
2334 reason = check_image_kernel(info);
2335 if (!reason && info->num_physpages != get_num_physpages())
2336 reason = "memory size";
2337 if (reason) {
2338 pr_err("Image mismatch: %s\n", reason);
2339 return -EPERM;
2340 }
2341 return 0;
2342}
2343
2344/**
2345 * load_header - Check the image header and copy the data from it.
2346 */
2347static int load_header(struct swsusp_info *info)
2348{
2349 int error;
2350
2351 restore_pblist = NULL;
2352 error = check_header(info);
2353 if (!error) {
2354 nr_copy_pages = info->image_pages;
2355 nr_meta_pages = info->pages - info->image_pages - 1;
2356 }
2357 return error;
2358}
2359
2360/**
2361 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2362 * @bm: Memory bitmap.
2363 * @buf: Area of memory containing the PFNs.
2364 * @zero_bm: Memory bitmap with the zero PFNs marked.
2365 *
2366 * For each element of the array pointed to by @buf (1 page at a time), set the
2367 * corresponding bit in @bm. If the page was originally populated with only
2368 * zeros then a corresponding bit will also be set in @zero_bm.
2369 */
2370static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm,
2371 struct memory_bitmap *zero_bm)
2372{
2373 unsigned long decoded_pfn;
2374 bool zero;
2375 int j;
2376
2377 for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2378 if (unlikely(buf[j] == BM_END_OF_MAP))
2379 break;
2380
2381 zero = !!(buf[j] & ENCODED_PFN_ZERO_FLAG);
2382 decoded_pfn = buf[j] & ENCODED_PFN_MASK;
2383 if (pfn_valid(decoded_pfn) && memory_bm_pfn_present(bm, decoded_pfn)) {
2384 memory_bm_set_bit(bm, decoded_pfn);
2385 if (zero) {
2386 memory_bm_set_bit(zero_bm, decoded_pfn);
2387 nr_zero_pages++;
2388 }
2389 } else {
2390 if (!pfn_valid(decoded_pfn))
2391 pr_err(FW_BUG "Memory map mismatch at 0x%llx after hibernation\n",
2392 (unsigned long long)PFN_PHYS(decoded_pfn));
2393 return -EFAULT;
2394 }
2395 }
2396
2397 return 0;
2398}
2399
2400#ifdef CONFIG_HIGHMEM
2401/*
2402 * struct highmem_pbe is used for creating the list of highmem pages that
2403 * should be restored atomically during the resume from disk, because the page
2404 * frames they have occupied before the suspend are in use.
2405 */
2406struct highmem_pbe {
2407 struct page *copy_page; /* data is here now */
2408 struct page *orig_page; /* data was here before the suspend */
2409 struct highmem_pbe *next;
2410};
2411
2412/*
2413 * List of highmem PBEs needed for restoring the highmem pages that were
2414 * allocated before the suspend and included in the suspend image, but have
2415 * also been allocated by the "resume" kernel, so their contents cannot be
2416 * written directly to their "original" page frames.
2417 */
2418static struct highmem_pbe *highmem_pblist;
2419
2420/**
2421 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2422 * @bm: Memory bitmap.
2423 *
2424 * The bits in @bm that correspond to image pages are assumed to be set.
2425 */
2426static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2427{
2428 unsigned long pfn;
2429 unsigned int cnt = 0;
2430
2431 memory_bm_position_reset(bm);
2432 pfn = memory_bm_next_pfn(bm);
2433 while (pfn != BM_END_OF_MAP) {
2434 if (PageHighMem(pfn_to_page(pfn)))
2435 cnt++;
2436
2437 pfn = memory_bm_next_pfn(bm);
2438 }
2439 return cnt;
2440}
2441
2442static unsigned int safe_highmem_pages;
2443
2444static struct memory_bitmap *safe_highmem_bm;
2445
2446/**
2447 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2448 * @bm: Pointer to an uninitialized memory bitmap structure.
2449 * @nr_highmem_p: Pointer to the number of highmem image pages.
2450 *
2451 * Try to allocate as many highmem pages as there are highmem image pages
2452 * (@nr_highmem_p points to the variable containing the number of highmem image
2453 * pages). The pages that are "safe" (ie. will not be overwritten when the
2454 * hibernation image is restored entirely) have the corresponding bits set in
2455 * @bm (it must be uninitialized).
2456 *
2457 * NOTE: This function should not be called if there are no highmem image pages.
2458 */
2459static int prepare_highmem_image(struct memory_bitmap *bm,
2460 unsigned int *nr_highmem_p)
2461{
2462 unsigned int to_alloc;
2463
2464 if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2465 return -ENOMEM;
2466
2467 if (get_highmem_buffer(PG_SAFE))
2468 return -ENOMEM;
2469
2470 to_alloc = count_free_highmem_pages();
2471 if (to_alloc > *nr_highmem_p)
2472 to_alloc = *nr_highmem_p;
2473 else
2474 *nr_highmem_p = to_alloc;
2475
2476 safe_highmem_pages = 0;
2477 while (to_alloc-- > 0) {
2478 struct page *page;
2479
2480 page = alloc_page(__GFP_HIGHMEM);
2481 if (!swsusp_page_is_free(page)) {
2482 /* The page is "safe", set its bit the bitmap */
2483 memory_bm_set_bit(bm, page_to_pfn(page));
2484 safe_highmem_pages++;
2485 }
2486 /* Mark the page as allocated */
2487 swsusp_set_page_forbidden(page);
2488 swsusp_set_page_free(page);
2489 }
2490 memory_bm_position_reset(bm);
2491 safe_highmem_bm = bm;
2492 return 0;
2493}
2494
2495static struct page *last_highmem_page;
2496
2497/**
2498 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2499 *
2500 * For a given highmem image page get a buffer that suspend_write_next() should
2501 * return to its caller to write to.
2502 *
2503 * If the page is to be saved to its "original" page frame or a copy of
2504 * the page is to be made in the highmem, @buffer is returned. Otherwise,
2505 * the copy of the page is to be made in normal memory, so the address of
2506 * the copy is returned.
2507 *
2508 * If @buffer is returned, the caller of suspend_write_next() will write
2509 * the page's contents to @buffer, so they will have to be copied to the
2510 * right location on the next call to suspend_write_next() and it is done
2511 * with the help of copy_last_highmem_page(). For this purpose, if
2512 * @buffer is returned, @last_highmem_page is set to the page to which
2513 * the data will have to be copied from @buffer.
2514 */
2515static void *get_highmem_page_buffer(struct page *page,
2516 struct chain_allocator *ca)
2517{
2518 struct highmem_pbe *pbe;
2519 void *kaddr;
2520
2521 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2522 /*
2523 * We have allocated the "original" page frame and we can
2524 * use it directly to store the loaded page.
2525 */
2526 last_highmem_page = page;
2527 return buffer;
2528 }
2529 /*
2530 * The "original" page frame has not been allocated and we have to
2531 * use a "safe" page frame to store the loaded page.
2532 */
2533 pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2534 if (!pbe) {
2535 swsusp_free();
2536 return ERR_PTR(-ENOMEM);
2537 }
2538 pbe->orig_page = page;
2539 if (safe_highmem_pages > 0) {
2540 struct page *tmp;
2541
2542 /* Copy of the page will be stored in high memory */
2543 kaddr = buffer;
2544 tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2545 safe_highmem_pages--;
2546 last_highmem_page = tmp;
2547 pbe->copy_page = tmp;
2548 } else {
2549 /* Copy of the page will be stored in normal memory */
2550 kaddr = __get_safe_page(ca->gfp_mask);
2551 if (!kaddr)
2552 return ERR_PTR(-ENOMEM);
2553 pbe->copy_page = virt_to_page(kaddr);
2554 }
2555 pbe->next = highmem_pblist;
2556 highmem_pblist = pbe;
2557 return kaddr;
2558}
2559
2560/**
2561 * copy_last_highmem_page - Copy most the most recent highmem image page.
2562 *
2563 * Copy the contents of a highmem image from @buffer, where the caller of
2564 * snapshot_write_next() has stored them, to the right location represented by
2565 * @last_highmem_page .
2566 */
2567static void copy_last_highmem_page(void)
2568{
2569 if (last_highmem_page) {
2570 void *dst;
2571
2572 dst = kmap_atomic(last_highmem_page);
2573 copy_page(dst, buffer);
2574 kunmap_atomic(dst);
2575 last_highmem_page = NULL;
2576 }
2577}
2578
2579static inline int last_highmem_page_copied(void)
2580{
2581 return !last_highmem_page;
2582}
2583
2584static inline void free_highmem_data(void)
2585{
2586 if (safe_highmem_bm)
2587 memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2588
2589 if (buffer)
2590 free_image_page(buffer, PG_UNSAFE_CLEAR);
2591}
2592#else
2593static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2594
2595static inline int prepare_highmem_image(struct memory_bitmap *bm,
2596 unsigned int *nr_highmem_p) { return 0; }
2597
2598static inline void *get_highmem_page_buffer(struct page *page,
2599 struct chain_allocator *ca)
2600{
2601 return ERR_PTR(-EINVAL);
2602}
2603
2604static inline void copy_last_highmem_page(void) {}
2605static inline int last_highmem_page_copied(void) { return 1; }
2606static inline void free_highmem_data(void) {}
2607#endif /* CONFIG_HIGHMEM */
2608
2609#define PBES_PER_LINKED_PAGE (LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2610
2611/**
2612 * prepare_image - Make room for loading hibernation image.
2613 * @new_bm: Uninitialized memory bitmap structure.
2614 * @bm: Memory bitmap with unsafe pages marked.
2615 * @zero_bm: Memory bitmap containing the zero pages.
2616 *
2617 * Use @bm to mark the pages that will be overwritten in the process of
2618 * restoring the system memory state from the suspend image ("unsafe" pages)
2619 * and allocate memory for the image.
2620 *
2621 * The idea is to allocate a new memory bitmap first and then allocate
2622 * as many pages as needed for image data, but without specifying what those
2623 * pages will be used for just yet. Instead, we mark them all as allocated and
2624 * create a lists of "safe" pages to be used later. On systems with high
2625 * memory a list of "safe" highmem pages is created too.
2626 *
2627 * Because it was not known which pages were unsafe when @zero_bm was created,
2628 * make a copy of it and recreate it within safe pages.
2629 */
2630static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm,
2631 struct memory_bitmap *zero_bm)
2632{
2633 unsigned int nr_pages, nr_highmem;
2634 struct memory_bitmap tmp;
2635 struct linked_page *lp;
2636 int error;
2637
2638 /* If there is no highmem, the buffer will not be necessary */
2639 free_image_page(buffer, PG_UNSAFE_CLEAR);
2640 buffer = NULL;
2641
2642 nr_highmem = count_highmem_image_pages(bm);
2643 mark_unsafe_pages(bm);
2644
2645 error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2646 if (error)
2647 goto Free;
2648
2649 duplicate_memory_bitmap(new_bm, bm);
2650 memory_bm_free(bm, PG_UNSAFE_KEEP);
2651
2652 /* Make a copy of zero_bm so it can be created in safe pages */
2653 error = memory_bm_create(&tmp, GFP_ATOMIC, PG_SAFE);
2654 if (error)
2655 goto Free;
2656
2657 duplicate_memory_bitmap(&tmp, zero_bm);
2658 memory_bm_free(zero_bm, PG_UNSAFE_KEEP);
2659
2660 /* Recreate zero_bm in safe pages */
2661 error = memory_bm_create(zero_bm, GFP_ATOMIC, PG_SAFE);
2662 if (error)
2663 goto Free;
2664
2665 duplicate_memory_bitmap(zero_bm, &tmp);
2666 memory_bm_free(&tmp, PG_UNSAFE_CLEAR);
2667 /* At this point zero_bm is in safe pages and it can be used for restoring. */
2668
2669 if (nr_highmem > 0) {
2670 error = prepare_highmem_image(bm, &nr_highmem);
2671 if (error)
2672 goto Free;
2673 }
2674 /*
2675 * Reserve some safe pages for potential later use.
2676 *
2677 * NOTE: This way we make sure there will be enough safe pages for the
2678 * chain_alloc() in get_buffer(). It is a bit wasteful, but
2679 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2680 *
2681 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2682 */
2683 nr_pages = (nr_zero_pages + nr_copy_pages) - nr_highmem - allocated_unsafe_pages;
2684 nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2685 while (nr_pages > 0) {
2686 lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2687 if (!lp) {
2688 error = -ENOMEM;
2689 goto Free;
2690 }
2691 lp->next = safe_pages_list;
2692 safe_pages_list = lp;
2693 nr_pages--;
2694 }
2695 /* Preallocate memory for the image */
2696 nr_pages = (nr_zero_pages + nr_copy_pages) - nr_highmem - allocated_unsafe_pages;
2697 while (nr_pages > 0) {
2698 lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2699 if (!lp) {
2700 error = -ENOMEM;
2701 goto Free;
2702 }
2703 if (!swsusp_page_is_free(virt_to_page(lp))) {
2704 /* The page is "safe", add it to the list */
2705 lp->next = safe_pages_list;
2706 safe_pages_list = lp;
2707 }
2708 /* Mark the page as allocated */
2709 swsusp_set_page_forbidden(virt_to_page(lp));
2710 swsusp_set_page_free(virt_to_page(lp));
2711 nr_pages--;
2712 }
2713 return 0;
2714
2715 Free:
2716 swsusp_free();
2717 return error;
2718}
2719
2720/**
2721 * get_buffer - Get the address to store the next image data page.
2722 *
2723 * Get the address that snapshot_write_next() should return to its caller to
2724 * write to.
2725 */
2726static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2727{
2728 struct pbe *pbe;
2729 struct page *page;
2730 unsigned long pfn = memory_bm_next_pfn(bm);
2731
2732 if (pfn == BM_END_OF_MAP)
2733 return ERR_PTR(-EFAULT);
2734
2735 page = pfn_to_page(pfn);
2736 if (PageHighMem(page))
2737 return get_highmem_page_buffer(page, ca);
2738
2739 if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2740 /*
2741 * We have allocated the "original" page frame and we can
2742 * use it directly to store the loaded page.
2743 */
2744 return page_address(page);
2745
2746 /*
2747 * The "original" page frame has not been allocated and we have to
2748 * use a "safe" page frame to store the loaded page.
2749 */
2750 pbe = chain_alloc(ca, sizeof(struct pbe));
2751 if (!pbe) {
2752 swsusp_free();
2753 return ERR_PTR(-ENOMEM);
2754 }
2755 pbe->orig_address = page_address(page);
2756 pbe->address = __get_safe_page(ca->gfp_mask);
2757 if (!pbe->address)
2758 return ERR_PTR(-ENOMEM);
2759 pbe->next = restore_pblist;
2760 restore_pblist = pbe;
2761 return pbe->address;
2762}
2763
2764/**
2765 * snapshot_write_next - Get the address to store the next image page.
2766 * @handle: Snapshot handle structure to guide the writing.
2767 *
2768 * On the first call, @handle should point to a zeroed snapshot_handle
2769 * structure. The structure gets populated then and a pointer to it should be
2770 * passed to this function every next time.
2771 *
2772 * On success, the function returns a positive number. Then, the caller
2773 * is allowed to write up to the returned number of bytes to the memory
2774 * location computed by the data_of() macro.
2775 *
2776 * The function returns 0 to indicate the "end of file" condition. Negative
2777 * numbers are returned on errors, in which cases the structure pointed to by
2778 * @handle is not updated and should not be used any more.
2779 */
2780int snapshot_write_next(struct snapshot_handle *handle)
2781{
2782 static struct chain_allocator ca;
2783 int error;
2784
2785next:
2786 /* Check if we have already loaded the entire image */
2787 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages + nr_zero_pages)
2788 return 0;
2789
2790 if (!handle->cur) {
2791 if (!buffer)
2792 /* This makes the buffer be freed by swsusp_free() */
2793 buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2794
2795 if (!buffer)
2796 return -ENOMEM;
2797
2798 handle->buffer = buffer;
2799 } else if (handle->cur == 1) {
2800 error = load_header(buffer);
2801 if (error)
2802 return error;
2803
2804 safe_pages_list = NULL;
2805
2806 error = memory_bm_create(©_bm, GFP_ATOMIC, PG_ANY);
2807 if (error)
2808 return error;
2809
2810 error = memory_bm_create(&zero_bm, GFP_ATOMIC, PG_ANY);
2811 if (error)
2812 return error;
2813
2814 nr_zero_pages = 0;
2815
2816 hibernate_restore_protection_begin();
2817 } else if (handle->cur <= nr_meta_pages + 1) {
2818 error = unpack_orig_pfns(buffer, ©_bm, &zero_bm);
2819 if (error)
2820 return error;
2821
2822 if (handle->cur == nr_meta_pages + 1) {
2823 error = prepare_image(&orig_bm, ©_bm, &zero_bm);
2824 if (error)
2825 return error;
2826
2827 chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2828 memory_bm_position_reset(&orig_bm);
2829 memory_bm_position_reset(&zero_bm);
2830 restore_pblist = NULL;
2831 handle->buffer = get_buffer(&orig_bm, &ca);
2832 if (IS_ERR(handle->buffer))
2833 return PTR_ERR(handle->buffer);
2834 }
2835 } else {
2836 copy_last_highmem_page();
2837 error = hibernate_restore_protect_page(handle->buffer);
2838 if (error)
2839 return error;
2840 handle->buffer = get_buffer(&orig_bm, &ca);
2841 if (IS_ERR(handle->buffer))
2842 return PTR_ERR(handle->buffer);
2843 }
2844 handle->sync_read = (handle->buffer == buffer);
2845 handle->cur++;
2846
2847 /* Zero pages were not included in the image, memset it and move on. */
2848 if (handle->cur > nr_meta_pages + 1 &&
2849 memory_bm_test_bit(&zero_bm, memory_bm_get_current(&orig_bm))) {
2850 memset(handle->buffer, 0, PAGE_SIZE);
2851 goto next;
2852 }
2853
2854 return PAGE_SIZE;
2855}
2856
2857/**
2858 * snapshot_write_finalize - Complete the loading of a hibernation image.
2859 *
2860 * Must be called after the last call to snapshot_write_next() in case the last
2861 * page in the image happens to be a highmem page and its contents should be
2862 * stored in highmem. Additionally, it recycles bitmap memory that's not
2863 * necessary any more.
2864 */
2865int snapshot_write_finalize(struct snapshot_handle *handle)
2866{
2867 int error;
2868
2869 copy_last_highmem_page();
2870 error = hibernate_restore_protect_page(handle->buffer);
2871 /* Do that only if we have loaded the image entirely */
2872 if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages + nr_zero_pages) {
2873 memory_bm_recycle(&orig_bm);
2874 free_highmem_data();
2875 }
2876 return error;
2877}
2878
2879int snapshot_image_loaded(struct snapshot_handle *handle)
2880{
2881 return !(!nr_copy_pages || !last_highmem_page_copied() ||
2882 handle->cur <= nr_meta_pages + nr_copy_pages + nr_zero_pages);
2883}
2884
2885#ifdef CONFIG_HIGHMEM
2886/* Assumes that @buf is ready and points to a "safe" page */
2887static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2888 void *buf)
2889{
2890 void *kaddr1, *kaddr2;
2891
2892 kaddr1 = kmap_atomic(p1);
2893 kaddr2 = kmap_atomic(p2);
2894 copy_page(buf, kaddr1);
2895 copy_page(kaddr1, kaddr2);
2896 copy_page(kaddr2, buf);
2897 kunmap_atomic(kaddr2);
2898 kunmap_atomic(kaddr1);
2899}
2900
2901/**
2902 * restore_highmem - Put highmem image pages into their original locations.
2903 *
2904 * For each highmem page that was in use before hibernation and is included in
2905 * the image, and also has been allocated by the "restore" kernel, swap its
2906 * current contents with the previous (ie. "before hibernation") ones.
2907 *
2908 * If the restore eventually fails, we can call this function once again and
2909 * restore the highmem state as seen by the restore kernel.
2910 */
2911int restore_highmem(void)
2912{
2913 struct highmem_pbe *pbe = highmem_pblist;
2914 void *buf;
2915
2916 if (!pbe)
2917 return 0;
2918
2919 buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2920 if (!buf)
2921 return -ENOMEM;
2922
2923 while (pbe) {
2924 swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2925 pbe = pbe->next;
2926 }
2927 free_image_page(buf, PG_UNSAFE_CLEAR);
2928 return 0;
2929}
2930#endif /* CONFIG_HIGHMEM */