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/*
2 * linux/kernel/power/snapshot.c
3 *
4 * This file provides system snapshot/restore functionality for swsusp.
5 *
6 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
7 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
8 *
9 * This file is released under the GPLv2.
10 *
11 */
12
13#include <linux/version.h>
14#include <linux/module.h>
15#include <linux/mm.h>
16#include <linux/suspend.h>
17#include <linux/delay.h>
18#include <linux/bitops.h>
19#include <linux/spinlock.h>
20#include <linux/kernel.h>
21#include <linux/pm.h>
22#include <linux/device.h>
23#include <linux/init.h>
24#include <linux/bootmem.h>
25#include <linux/syscalls.h>
26#include <linux/console.h>
27#include <linux/highmem.h>
28#include <linux/list.h>
29#include <linux/slab.h>
30#include <linux/compiler.h>
31#include <linux/ktime.h>
32
33#include <linux/uaccess.h>
34#include <asm/mmu_context.h>
35#include <asm/pgtable.h>
36#include <asm/tlbflush.h>
37#include <asm/io.h>
38
39#include "power.h"
40
41#ifdef CONFIG_DEBUG_RODATA
42static bool hibernate_restore_protection;
43static bool hibernate_restore_protection_active;
44
45void enable_restore_image_protection(void)
46{
47 hibernate_restore_protection = true;
48}
49
50static inline void hibernate_restore_protection_begin(void)
51{
52 hibernate_restore_protection_active = hibernate_restore_protection;
53}
54
55static inline void hibernate_restore_protection_end(void)
56{
57 hibernate_restore_protection_active = false;
58}
59
60static inline void hibernate_restore_protect_page(void *page_address)
61{
62 if (hibernate_restore_protection_active)
63 set_memory_ro((unsigned long)page_address, 1);
64}
65
66static inline void hibernate_restore_unprotect_page(void *page_address)
67{
68 if (hibernate_restore_protection_active)
69 set_memory_rw((unsigned long)page_address, 1);
70}
71#else
72static inline void hibernate_restore_protection_begin(void) {}
73static inline void hibernate_restore_protection_end(void) {}
74static inline void hibernate_restore_protect_page(void *page_address) {}
75static inline void hibernate_restore_unprotect_page(void *page_address) {}
76#endif /* CONFIG_DEBUG_RODATA */
77
78static int swsusp_page_is_free(struct page *);
79static void swsusp_set_page_forbidden(struct page *);
80static void swsusp_unset_page_forbidden(struct page *);
81
82/*
83 * Number of bytes to reserve for memory allocations made by device drivers
84 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
85 * cause image creation to fail (tunable via /sys/power/reserved_size).
86 */
87unsigned long reserved_size;
88
89void __init hibernate_reserved_size_init(void)
90{
91 reserved_size = SPARE_PAGES * PAGE_SIZE;
92}
93
94/*
95 * Preferred image size in bytes (tunable via /sys/power/image_size).
96 * When it is set to N, swsusp will do its best to ensure the image
97 * size will not exceed N bytes, but if that is impossible, it will
98 * try to create the smallest image possible.
99 */
100unsigned long image_size;
101
102void __init hibernate_image_size_init(void)
103{
104 image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE;
105}
106
107/*
108 * List of PBEs needed for restoring the pages that were allocated before
109 * the suspend and included in the suspend image, but have also been
110 * allocated by the "resume" kernel, so their contents cannot be written
111 * directly to their "original" page frames.
112 */
113struct pbe *restore_pblist;
114
115/* struct linked_page is used to build chains of pages */
116
117#define LINKED_PAGE_DATA_SIZE (PAGE_SIZE - sizeof(void *))
118
119struct linked_page {
120 struct linked_page *next;
121 char data[LINKED_PAGE_DATA_SIZE];
122} __packed;
123
124/*
125 * List of "safe" pages (ie. pages that were not used by the image kernel
126 * before hibernation) that may be used as temporary storage for image kernel
127 * memory contents.
128 */
129static struct linked_page *safe_pages_list;
130
131/* Pointer to an auxiliary buffer (1 page) */
132static void *buffer;
133
134#define PG_ANY 0
135#define PG_SAFE 1
136#define PG_UNSAFE_CLEAR 1
137#define PG_UNSAFE_KEEP 0
138
139static unsigned int allocated_unsafe_pages;
140
141/**
142 * get_image_page - Allocate a page for a hibernation image.
143 * @gfp_mask: GFP mask for the allocation.
144 * @safe_needed: Get pages that were not used before hibernation (restore only)
145 *
146 * During image restoration, for storing the PBE list and the image data, we can
147 * only use memory pages that do not conflict with the pages used before
148 * hibernation. The "unsafe" pages have PageNosaveFree set and we count them
149 * using allocated_unsafe_pages.
150 *
151 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
152 * swsusp_free() can release it.
153 */
154static void *get_image_page(gfp_t gfp_mask, int safe_needed)
155{
156 void *res;
157
158 res = (void *)get_zeroed_page(gfp_mask);
159 if (safe_needed)
160 while (res && swsusp_page_is_free(virt_to_page(res))) {
161 /* The page is unsafe, mark it for swsusp_free() */
162 swsusp_set_page_forbidden(virt_to_page(res));
163 allocated_unsafe_pages++;
164 res = (void *)get_zeroed_page(gfp_mask);
165 }
166 if (res) {
167 swsusp_set_page_forbidden(virt_to_page(res));
168 swsusp_set_page_free(virt_to_page(res));
169 }
170 return res;
171}
172
173static void *__get_safe_page(gfp_t gfp_mask)
174{
175 if (safe_pages_list) {
176 void *ret = safe_pages_list;
177
178 safe_pages_list = safe_pages_list->next;
179 memset(ret, 0, PAGE_SIZE);
180 return ret;
181 }
182 return get_image_page(gfp_mask, PG_SAFE);
183}
184
185unsigned long get_safe_page(gfp_t gfp_mask)
186{
187 return (unsigned long)__get_safe_page(gfp_mask);
188}
189
190static struct page *alloc_image_page(gfp_t gfp_mask)
191{
192 struct page *page;
193
194 page = alloc_page(gfp_mask);
195 if (page) {
196 swsusp_set_page_forbidden(page);
197 swsusp_set_page_free(page);
198 }
199 return page;
200}
201
202static void recycle_safe_page(void *page_address)
203{
204 struct linked_page *lp = page_address;
205
206 lp->next = safe_pages_list;
207 safe_pages_list = lp;
208}
209
210/**
211 * free_image_page - Free a page allocated for hibernation image.
212 * @addr: Address of the page to free.
213 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
214 *
215 * The page to free should have been allocated by get_image_page() (page flags
216 * set by it are affected).
217 */
218static inline void free_image_page(void *addr, int clear_nosave_free)
219{
220 struct page *page;
221
222 BUG_ON(!virt_addr_valid(addr));
223
224 page = virt_to_page(addr);
225
226 swsusp_unset_page_forbidden(page);
227 if (clear_nosave_free)
228 swsusp_unset_page_free(page);
229
230 __free_page(page);
231}
232
233static inline void free_list_of_pages(struct linked_page *list,
234 int clear_page_nosave)
235{
236 while (list) {
237 struct linked_page *lp = list->next;
238
239 free_image_page(list, clear_page_nosave);
240 list = lp;
241 }
242}
243
244/*
245 * struct chain_allocator is used for allocating small objects out of
246 * a linked list of pages called 'the chain'.
247 *
248 * The chain grows each time when there is no room for a new object in
249 * the current page. The allocated objects cannot be freed individually.
250 * It is only possible to free them all at once, by freeing the entire
251 * chain.
252 *
253 * NOTE: The chain allocator may be inefficient if the allocated objects
254 * are not much smaller than PAGE_SIZE.
255 */
256struct chain_allocator {
257 struct linked_page *chain; /* the chain */
258 unsigned int used_space; /* total size of objects allocated out
259 of the current page */
260 gfp_t gfp_mask; /* mask for allocating pages */
261 int safe_needed; /* if set, only "safe" pages are allocated */
262};
263
264static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
265 int safe_needed)
266{
267 ca->chain = NULL;
268 ca->used_space = LINKED_PAGE_DATA_SIZE;
269 ca->gfp_mask = gfp_mask;
270 ca->safe_needed = safe_needed;
271}
272
273static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
274{
275 void *ret;
276
277 if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
278 struct linked_page *lp;
279
280 lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
281 get_image_page(ca->gfp_mask, PG_ANY);
282 if (!lp)
283 return NULL;
284
285 lp->next = ca->chain;
286 ca->chain = lp;
287 ca->used_space = 0;
288 }
289 ret = ca->chain->data + ca->used_space;
290 ca->used_space += size;
291 return ret;
292}
293
294/**
295 * Data types related to memory bitmaps.
296 *
297 * Memory bitmap is a structure consiting of many linked lists of
298 * objects. The main list's elements are of type struct zone_bitmap
299 * and each of them corresonds to one zone. For each zone bitmap
300 * object there is a list of objects of type struct bm_block that
301 * represent each blocks of bitmap in which information is stored.
302 *
303 * struct memory_bitmap contains a pointer to the main list of zone
304 * bitmap objects, a struct bm_position used for browsing the bitmap,
305 * and a pointer to the list of pages used for allocating all of the
306 * zone bitmap objects and bitmap block objects.
307 *
308 * NOTE: It has to be possible to lay out the bitmap in memory
309 * using only allocations of order 0. Additionally, the bitmap is
310 * designed to work with arbitrary number of zones (this is over the
311 * top for now, but let's avoid making unnecessary assumptions ;-).
312 *
313 * struct zone_bitmap contains a pointer to a list of bitmap block
314 * objects and a pointer to the bitmap block object that has been
315 * most recently used for setting bits. Additionally, it contains the
316 * PFNs that correspond to the start and end of the represented zone.
317 *
318 * struct bm_block contains a pointer to the memory page in which
319 * information is stored (in the form of a block of bitmap)
320 * It also contains the pfns that correspond to the start and end of
321 * the represented memory area.
322 *
323 * The memory bitmap is organized as a radix tree to guarantee fast random
324 * access to the bits. There is one radix tree for each zone (as returned
325 * from create_mem_extents).
326 *
327 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
328 * two linked lists for the nodes of the tree, one for the inner nodes and
329 * one for the leave nodes. The linked leave nodes are used for fast linear
330 * access of the memory bitmap.
331 *
332 * The struct rtree_node represents one node of the radix tree.
333 */
334
335#define BM_END_OF_MAP (~0UL)
336
337#define BM_BITS_PER_BLOCK (PAGE_SIZE * BITS_PER_BYTE)
338#define BM_BLOCK_SHIFT (PAGE_SHIFT + 3)
339#define BM_BLOCK_MASK ((1UL << BM_BLOCK_SHIFT) - 1)
340
341/*
342 * struct rtree_node is a wrapper struct to link the nodes
343 * of the rtree together for easy linear iteration over
344 * bits and easy freeing
345 */
346struct rtree_node {
347 struct list_head list;
348 unsigned long *data;
349};
350
351/*
352 * struct mem_zone_bm_rtree represents a bitmap used for one
353 * populated memory zone.
354 */
355struct mem_zone_bm_rtree {
356 struct list_head list; /* Link Zones together */
357 struct list_head nodes; /* Radix Tree inner nodes */
358 struct list_head leaves; /* Radix Tree leaves */
359 unsigned long start_pfn; /* Zone start page frame */
360 unsigned long end_pfn; /* Zone end page frame + 1 */
361 struct rtree_node *rtree; /* Radix Tree Root */
362 int levels; /* Number of Radix Tree Levels */
363 unsigned int blocks; /* Number of Bitmap Blocks */
364};
365
366/* strcut bm_position is used for browsing memory bitmaps */
367
368struct bm_position {
369 struct mem_zone_bm_rtree *zone;
370 struct rtree_node *node;
371 unsigned long node_pfn;
372 int node_bit;
373};
374
375struct memory_bitmap {
376 struct list_head zones;
377 struct linked_page *p_list; /* list of pages used to store zone
378 bitmap objects and bitmap block
379 objects */
380 struct bm_position cur; /* most recently used bit position */
381};
382
383/* Functions that operate on memory bitmaps */
384
385#define BM_ENTRIES_PER_LEVEL (PAGE_SIZE / sizeof(unsigned long))
386#if BITS_PER_LONG == 32
387#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 2)
388#else
389#define BM_RTREE_LEVEL_SHIFT (PAGE_SHIFT - 3)
390#endif
391#define BM_RTREE_LEVEL_MASK ((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
392
393/**
394 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
395 *
396 * This function is used to allocate inner nodes as well as the
397 * leave nodes of the radix tree. It also adds the node to the
398 * corresponding linked list passed in by the *list parameter.
399 */
400static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
401 struct chain_allocator *ca,
402 struct list_head *list)
403{
404 struct rtree_node *node;
405
406 node = chain_alloc(ca, sizeof(struct rtree_node));
407 if (!node)
408 return NULL;
409
410 node->data = get_image_page(gfp_mask, safe_needed);
411 if (!node->data)
412 return NULL;
413
414 list_add_tail(&node->list, list);
415
416 return node;
417}
418
419/**
420 * add_rtree_block - Add a new leave node to the radix tree.
421 *
422 * The leave nodes need to be allocated in order to keep the leaves
423 * linked list in order. This is guaranteed by the zone->blocks
424 * counter.
425 */
426static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
427 int safe_needed, struct chain_allocator *ca)
428{
429 struct rtree_node *node, *block, **dst;
430 unsigned int levels_needed, block_nr;
431 int i;
432
433 block_nr = zone->blocks;
434 levels_needed = 0;
435
436 /* How many levels do we need for this block nr? */
437 while (block_nr) {
438 levels_needed += 1;
439 block_nr >>= BM_RTREE_LEVEL_SHIFT;
440 }
441
442 /* Make sure the rtree has enough levels */
443 for (i = zone->levels; i < levels_needed; i++) {
444 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
445 &zone->nodes);
446 if (!node)
447 return -ENOMEM;
448
449 node->data[0] = (unsigned long)zone->rtree;
450 zone->rtree = node;
451 zone->levels += 1;
452 }
453
454 /* Allocate new block */
455 block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
456 if (!block)
457 return -ENOMEM;
458
459 /* Now walk the rtree to insert the block */
460 node = zone->rtree;
461 dst = &zone->rtree;
462 block_nr = zone->blocks;
463 for (i = zone->levels; i > 0; i--) {
464 int index;
465
466 if (!node) {
467 node = alloc_rtree_node(gfp_mask, safe_needed, ca,
468 &zone->nodes);
469 if (!node)
470 return -ENOMEM;
471 *dst = node;
472 }
473
474 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
475 index &= BM_RTREE_LEVEL_MASK;
476 dst = (struct rtree_node **)&((*dst)->data[index]);
477 node = *dst;
478 }
479
480 zone->blocks += 1;
481 *dst = block;
482
483 return 0;
484}
485
486static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
487 int clear_nosave_free);
488
489/**
490 * create_zone_bm_rtree - Create a radix tree for one zone.
491 *
492 * Allocated the mem_zone_bm_rtree structure and initializes it.
493 * This function also allocated and builds the radix tree for the
494 * zone.
495 */
496static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
497 int safe_needed,
498 struct chain_allocator *ca,
499 unsigned long start,
500 unsigned long end)
501{
502 struct mem_zone_bm_rtree *zone;
503 unsigned int i, nr_blocks;
504 unsigned long pages;
505
506 pages = end - start;
507 zone = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
508 if (!zone)
509 return NULL;
510
511 INIT_LIST_HEAD(&zone->nodes);
512 INIT_LIST_HEAD(&zone->leaves);
513 zone->start_pfn = start;
514 zone->end_pfn = end;
515 nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
516
517 for (i = 0; i < nr_blocks; i++) {
518 if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
519 free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
520 return NULL;
521 }
522 }
523
524 return zone;
525}
526
527/**
528 * free_zone_bm_rtree - Free the memory of the radix tree.
529 *
530 * Free all node pages of the radix tree. The mem_zone_bm_rtree
531 * structure itself is not freed here nor are the rtree_node
532 * structs.
533 */
534static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
535 int clear_nosave_free)
536{
537 struct rtree_node *node;
538
539 list_for_each_entry(node, &zone->nodes, list)
540 free_image_page(node->data, clear_nosave_free);
541
542 list_for_each_entry(node, &zone->leaves, list)
543 free_image_page(node->data, clear_nosave_free);
544}
545
546static void memory_bm_position_reset(struct memory_bitmap *bm)
547{
548 bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
549 list);
550 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
551 struct rtree_node, list);
552 bm->cur.node_pfn = 0;
553 bm->cur.node_bit = 0;
554}
555
556static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
557
558struct mem_extent {
559 struct list_head hook;
560 unsigned long start;
561 unsigned long end;
562};
563
564/**
565 * free_mem_extents - Free a list of memory extents.
566 * @list: List of extents to free.
567 */
568static void free_mem_extents(struct list_head *list)
569{
570 struct mem_extent *ext, *aux;
571
572 list_for_each_entry_safe(ext, aux, list, hook) {
573 list_del(&ext->hook);
574 kfree(ext);
575 }
576}
577
578/**
579 * create_mem_extents - Create a list of memory extents.
580 * @list: List to put the extents into.
581 * @gfp_mask: Mask to use for memory allocations.
582 *
583 * The extents represent contiguous ranges of PFNs.
584 */
585static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
586{
587 struct zone *zone;
588
589 INIT_LIST_HEAD(list);
590
591 for_each_populated_zone(zone) {
592 unsigned long zone_start, zone_end;
593 struct mem_extent *ext, *cur, *aux;
594
595 zone_start = zone->zone_start_pfn;
596 zone_end = zone_end_pfn(zone);
597
598 list_for_each_entry(ext, list, hook)
599 if (zone_start <= ext->end)
600 break;
601
602 if (&ext->hook == list || zone_end < ext->start) {
603 /* New extent is necessary */
604 struct mem_extent *new_ext;
605
606 new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
607 if (!new_ext) {
608 free_mem_extents(list);
609 return -ENOMEM;
610 }
611 new_ext->start = zone_start;
612 new_ext->end = zone_end;
613 list_add_tail(&new_ext->hook, &ext->hook);
614 continue;
615 }
616
617 /* Merge this zone's range of PFNs with the existing one */
618 if (zone_start < ext->start)
619 ext->start = zone_start;
620 if (zone_end > ext->end)
621 ext->end = zone_end;
622
623 /* More merging may be possible */
624 cur = ext;
625 list_for_each_entry_safe_continue(cur, aux, list, hook) {
626 if (zone_end < cur->start)
627 break;
628 if (zone_end < cur->end)
629 ext->end = cur->end;
630 list_del(&cur->hook);
631 kfree(cur);
632 }
633 }
634
635 return 0;
636}
637
638/**
639 * memory_bm_create - Allocate memory for a memory bitmap.
640 */
641static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
642 int safe_needed)
643{
644 struct chain_allocator ca;
645 struct list_head mem_extents;
646 struct mem_extent *ext;
647 int error;
648
649 chain_init(&ca, gfp_mask, safe_needed);
650 INIT_LIST_HEAD(&bm->zones);
651
652 error = create_mem_extents(&mem_extents, gfp_mask);
653 if (error)
654 return error;
655
656 list_for_each_entry(ext, &mem_extents, hook) {
657 struct mem_zone_bm_rtree *zone;
658
659 zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
660 ext->start, ext->end);
661 if (!zone) {
662 error = -ENOMEM;
663 goto Error;
664 }
665 list_add_tail(&zone->list, &bm->zones);
666 }
667
668 bm->p_list = ca.chain;
669 memory_bm_position_reset(bm);
670 Exit:
671 free_mem_extents(&mem_extents);
672 return error;
673
674 Error:
675 bm->p_list = ca.chain;
676 memory_bm_free(bm, PG_UNSAFE_CLEAR);
677 goto Exit;
678}
679
680/**
681 * memory_bm_free - Free memory occupied by the memory bitmap.
682 * @bm: Memory bitmap.
683 */
684static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
685{
686 struct mem_zone_bm_rtree *zone;
687
688 list_for_each_entry(zone, &bm->zones, list)
689 free_zone_bm_rtree(zone, clear_nosave_free);
690
691 free_list_of_pages(bm->p_list, clear_nosave_free);
692
693 INIT_LIST_HEAD(&bm->zones);
694}
695
696/**
697 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
698 *
699 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
700 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
701 *
702 * Walk the radix tree to find the page containing the bit that represents @pfn
703 * and return the position of the bit in @addr and @bit_nr.
704 */
705static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
706 void **addr, unsigned int *bit_nr)
707{
708 struct mem_zone_bm_rtree *curr, *zone;
709 struct rtree_node *node;
710 int i, block_nr;
711
712 zone = bm->cur.zone;
713
714 if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
715 goto zone_found;
716
717 zone = NULL;
718
719 /* Find the right zone */
720 list_for_each_entry(curr, &bm->zones, list) {
721 if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
722 zone = curr;
723 break;
724 }
725 }
726
727 if (!zone)
728 return -EFAULT;
729
730zone_found:
731 /*
732 * We have found the zone. Now walk the radix tree to find the leaf node
733 * for our PFN.
734 */
735 node = bm->cur.node;
736 if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
737 goto node_found;
738
739 node = zone->rtree;
740 block_nr = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
741
742 for (i = zone->levels; i > 0; i--) {
743 int index;
744
745 index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
746 index &= BM_RTREE_LEVEL_MASK;
747 BUG_ON(node->data[index] == 0);
748 node = (struct rtree_node *)node->data[index];
749 }
750
751node_found:
752 /* Update last position */
753 bm->cur.zone = zone;
754 bm->cur.node = node;
755 bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
756
757 /* Set return values */
758 *addr = node->data;
759 *bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
760
761 return 0;
762}
763
764static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
765{
766 void *addr;
767 unsigned int bit;
768 int error;
769
770 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
771 BUG_ON(error);
772 set_bit(bit, addr);
773}
774
775static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
776{
777 void *addr;
778 unsigned int bit;
779 int error;
780
781 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
782 if (!error)
783 set_bit(bit, addr);
784
785 return error;
786}
787
788static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
789{
790 void *addr;
791 unsigned int bit;
792 int error;
793
794 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
795 BUG_ON(error);
796 clear_bit(bit, addr);
797}
798
799static void memory_bm_clear_current(struct memory_bitmap *bm)
800{
801 int bit;
802
803 bit = max(bm->cur.node_bit - 1, 0);
804 clear_bit(bit, bm->cur.node->data);
805}
806
807static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
808{
809 void *addr;
810 unsigned int bit;
811 int error;
812
813 error = memory_bm_find_bit(bm, pfn, &addr, &bit);
814 BUG_ON(error);
815 return test_bit(bit, addr);
816}
817
818static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
819{
820 void *addr;
821 unsigned int bit;
822
823 return !memory_bm_find_bit(bm, pfn, &addr, &bit);
824}
825
826/*
827 * rtree_next_node - Jump to the next leaf node.
828 *
829 * Set the position to the beginning of the next node in the
830 * memory bitmap. This is either the next node in the current
831 * zone's radix tree or the first node in the radix tree of the
832 * next zone.
833 *
834 * Return true if there is a next node, false otherwise.
835 */
836static bool rtree_next_node(struct memory_bitmap *bm)
837{
838 if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
839 bm->cur.node = list_entry(bm->cur.node->list.next,
840 struct rtree_node, list);
841 bm->cur.node_pfn += BM_BITS_PER_BLOCK;
842 bm->cur.node_bit = 0;
843 touch_softlockup_watchdog();
844 return true;
845 }
846
847 /* No more nodes, goto next zone */
848 if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
849 bm->cur.zone = list_entry(bm->cur.zone->list.next,
850 struct mem_zone_bm_rtree, list);
851 bm->cur.node = list_entry(bm->cur.zone->leaves.next,
852 struct rtree_node, list);
853 bm->cur.node_pfn = 0;
854 bm->cur.node_bit = 0;
855 return true;
856 }
857
858 /* No more zones */
859 return false;
860}
861
862/**
863 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
864 * @bm: Memory bitmap.
865 *
866 * Starting from the last returned position this function searches for the next
867 * set bit in @bm and returns the PFN represented by it. If no more bits are
868 * set, BM_END_OF_MAP is returned.
869 *
870 * It is required to run memory_bm_position_reset() before the first call to
871 * this function for the given memory bitmap.
872 */
873static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
874{
875 unsigned long bits, pfn, pages;
876 int bit;
877
878 do {
879 pages = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
880 bits = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
881 bit = find_next_bit(bm->cur.node->data, bits,
882 bm->cur.node_bit);
883 if (bit < bits) {
884 pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
885 bm->cur.node_bit = bit + 1;
886 return pfn;
887 }
888 } while (rtree_next_node(bm));
889
890 return BM_END_OF_MAP;
891}
892
893/*
894 * This structure represents a range of page frames the contents of which
895 * should not be saved during hibernation.
896 */
897struct nosave_region {
898 struct list_head list;
899 unsigned long start_pfn;
900 unsigned long end_pfn;
901};
902
903static LIST_HEAD(nosave_regions);
904
905static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
906{
907 struct rtree_node *node;
908
909 list_for_each_entry(node, &zone->nodes, list)
910 recycle_safe_page(node->data);
911
912 list_for_each_entry(node, &zone->leaves, list)
913 recycle_safe_page(node->data);
914}
915
916static void memory_bm_recycle(struct memory_bitmap *bm)
917{
918 struct mem_zone_bm_rtree *zone;
919 struct linked_page *p_list;
920
921 list_for_each_entry(zone, &bm->zones, list)
922 recycle_zone_bm_rtree(zone);
923
924 p_list = bm->p_list;
925 while (p_list) {
926 struct linked_page *lp = p_list;
927
928 p_list = lp->next;
929 recycle_safe_page(lp);
930 }
931}
932
933/**
934 * register_nosave_region - Register a region of unsaveable memory.
935 *
936 * Register a range of page frames the contents of which should not be saved
937 * during hibernation (to be used in the early initialization code).
938 */
939void __init __register_nosave_region(unsigned long start_pfn,
940 unsigned long end_pfn, int use_kmalloc)
941{
942 struct nosave_region *region;
943
944 if (start_pfn >= end_pfn)
945 return;
946
947 if (!list_empty(&nosave_regions)) {
948 /* Try to extend the previous region (they should be sorted) */
949 region = list_entry(nosave_regions.prev,
950 struct nosave_region, list);
951 if (region->end_pfn == start_pfn) {
952 region->end_pfn = end_pfn;
953 goto Report;
954 }
955 }
956 if (use_kmalloc) {
957 /* During init, this shouldn't fail */
958 region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
959 BUG_ON(!region);
960 } else {
961 /* This allocation cannot fail */
962 region = memblock_virt_alloc(sizeof(struct nosave_region), 0);
963 }
964 region->start_pfn = start_pfn;
965 region->end_pfn = end_pfn;
966 list_add_tail(®ion->list, &nosave_regions);
967 Report:
968 printk(KERN_INFO "PM: Registered nosave memory: [mem %#010llx-%#010llx]\n",
969 (unsigned long long) start_pfn << PAGE_SHIFT,
970 ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
971}
972
973/*
974 * Set bits in this map correspond to the page frames the contents of which
975 * should not be saved during the suspend.
976 */
977static struct memory_bitmap *forbidden_pages_map;
978
979/* Set bits in this map correspond to free page frames. */
980static struct memory_bitmap *free_pages_map;
981
982/*
983 * Each page frame allocated for creating the image is marked by setting the
984 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
985 */
986
987void swsusp_set_page_free(struct page *page)
988{
989 if (free_pages_map)
990 memory_bm_set_bit(free_pages_map, page_to_pfn(page));
991}
992
993static int swsusp_page_is_free(struct page *page)
994{
995 return free_pages_map ?
996 memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
997}
998
999void swsusp_unset_page_free(struct page *page)
1000{
1001 if (free_pages_map)
1002 memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1003}
1004
1005static void swsusp_set_page_forbidden(struct page *page)
1006{
1007 if (forbidden_pages_map)
1008 memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1009}
1010
1011int swsusp_page_is_forbidden(struct page *page)
1012{
1013 return forbidden_pages_map ?
1014 memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1015}
1016
1017static void swsusp_unset_page_forbidden(struct page *page)
1018{
1019 if (forbidden_pages_map)
1020 memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1021}
1022
1023/**
1024 * mark_nosave_pages - Mark pages that should not be saved.
1025 * @bm: Memory bitmap.
1026 *
1027 * Set the bits in @bm that correspond to the page frames the contents of which
1028 * should not be saved.
1029 */
1030static void mark_nosave_pages(struct memory_bitmap *bm)
1031{
1032 struct nosave_region *region;
1033
1034 if (list_empty(&nosave_regions))
1035 return;
1036
1037 list_for_each_entry(region, &nosave_regions, list) {
1038 unsigned long pfn;
1039
1040 pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n",
1041 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1042 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1043 - 1);
1044
1045 for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1046 if (pfn_valid(pfn)) {
1047 /*
1048 * It is safe to ignore the result of
1049 * mem_bm_set_bit_check() here, since we won't
1050 * touch the PFNs for which the error is
1051 * returned anyway.
1052 */
1053 mem_bm_set_bit_check(bm, pfn);
1054 }
1055 }
1056}
1057
1058/**
1059 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1060 *
1061 * Create bitmaps needed for marking page frames that should not be saved and
1062 * free page frames. The forbidden_pages_map and free_pages_map pointers are
1063 * only modified if everything goes well, because we don't want the bits to be
1064 * touched before both bitmaps are set up.
1065 */
1066int create_basic_memory_bitmaps(void)
1067{
1068 struct memory_bitmap *bm1, *bm2;
1069 int error = 0;
1070
1071 if (forbidden_pages_map && free_pages_map)
1072 return 0;
1073 else
1074 BUG_ON(forbidden_pages_map || free_pages_map);
1075
1076 bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1077 if (!bm1)
1078 return -ENOMEM;
1079
1080 error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1081 if (error)
1082 goto Free_first_object;
1083
1084 bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1085 if (!bm2)
1086 goto Free_first_bitmap;
1087
1088 error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1089 if (error)
1090 goto Free_second_object;
1091
1092 forbidden_pages_map = bm1;
1093 free_pages_map = bm2;
1094 mark_nosave_pages(forbidden_pages_map);
1095
1096 pr_debug("PM: Basic memory bitmaps created\n");
1097
1098 return 0;
1099
1100 Free_second_object:
1101 kfree(bm2);
1102 Free_first_bitmap:
1103 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1104 Free_first_object:
1105 kfree(bm1);
1106 return -ENOMEM;
1107}
1108
1109/**
1110 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1111 *
1112 * Free memory bitmaps allocated by create_basic_memory_bitmaps(). The
1113 * auxiliary pointers are necessary so that the bitmaps themselves are not
1114 * referred to while they are being freed.
1115 */
1116void free_basic_memory_bitmaps(void)
1117{
1118 struct memory_bitmap *bm1, *bm2;
1119
1120 if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1121 return;
1122
1123 bm1 = forbidden_pages_map;
1124 bm2 = free_pages_map;
1125 forbidden_pages_map = NULL;
1126 free_pages_map = NULL;
1127 memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1128 kfree(bm1);
1129 memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1130 kfree(bm2);
1131
1132 pr_debug("PM: Basic memory bitmaps freed\n");
1133}
1134
1135void clear_free_pages(void)
1136{
1137#ifdef CONFIG_PAGE_POISONING_ZERO
1138 struct memory_bitmap *bm = free_pages_map;
1139 unsigned long pfn;
1140
1141 if (WARN_ON(!(free_pages_map)))
1142 return;
1143
1144 memory_bm_position_reset(bm);
1145 pfn = memory_bm_next_pfn(bm);
1146 while (pfn != BM_END_OF_MAP) {
1147 if (pfn_valid(pfn))
1148 clear_highpage(pfn_to_page(pfn));
1149
1150 pfn = memory_bm_next_pfn(bm);
1151 }
1152 memory_bm_position_reset(bm);
1153 pr_info("PM: free pages cleared after restore\n");
1154#endif /* PAGE_POISONING_ZERO */
1155}
1156
1157/**
1158 * snapshot_additional_pages - Estimate the number of extra pages needed.
1159 * @zone: Memory zone to carry out the computation for.
1160 *
1161 * Estimate the number of additional pages needed for setting up a hibernation
1162 * image data structures for @zone (usually, the returned value is greater than
1163 * the exact number).
1164 */
1165unsigned int snapshot_additional_pages(struct zone *zone)
1166{
1167 unsigned int rtree, nodes;
1168
1169 rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1170 rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1171 LINKED_PAGE_DATA_SIZE);
1172 while (nodes > 1) {
1173 nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1174 rtree += nodes;
1175 }
1176
1177 return 2 * rtree;
1178}
1179
1180#ifdef CONFIG_HIGHMEM
1181/**
1182 * count_free_highmem_pages - Compute the total number of free highmem pages.
1183 *
1184 * The returned number is system-wide.
1185 */
1186static unsigned int count_free_highmem_pages(void)
1187{
1188 struct zone *zone;
1189 unsigned int cnt = 0;
1190
1191 for_each_populated_zone(zone)
1192 if (is_highmem(zone))
1193 cnt += zone_page_state(zone, NR_FREE_PAGES);
1194
1195 return cnt;
1196}
1197
1198/**
1199 * saveable_highmem_page - Check if a highmem page is saveable.
1200 *
1201 * Determine whether a highmem page should be included in a hibernation image.
1202 *
1203 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1204 * and it isn't part of a free chunk of pages.
1205 */
1206static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1207{
1208 struct page *page;
1209
1210 if (!pfn_valid(pfn))
1211 return NULL;
1212
1213 page = pfn_to_page(pfn);
1214 if (page_zone(page) != zone)
1215 return NULL;
1216
1217 BUG_ON(!PageHighMem(page));
1218
1219 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page) ||
1220 PageReserved(page))
1221 return NULL;
1222
1223 if (page_is_guard(page))
1224 return NULL;
1225
1226 return page;
1227}
1228
1229/**
1230 * count_highmem_pages - Compute the total number of saveable highmem pages.
1231 */
1232static unsigned int count_highmem_pages(void)
1233{
1234 struct zone *zone;
1235 unsigned int n = 0;
1236
1237 for_each_populated_zone(zone) {
1238 unsigned long pfn, max_zone_pfn;
1239
1240 if (!is_highmem(zone))
1241 continue;
1242
1243 mark_free_pages(zone);
1244 max_zone_pfn = zone_end_pfn(zone);
1245 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1246 if (saveable_highmem_page(zone, pfn))
1247 n++;
1248 }
1249 return n;
1250}
1251#else
1252static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1253{
1254 return NULL;
1255}
1256#endif /* CONFIG_HIGHMEM */
1257
1258/**
1259 * saveable_page - Check if the given page is saveable.
1260 *
1261 * Determine whether a non-highmem page should be included in a hibernation
1262 * image.
1263 *
1264 * We should save the page if it isn't Nosave, and is not in the range
1265 * of pages statically defined as 'unsaveable', and it isn't part of
1266 * a free chunk of pages.
1267 */
1268static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1269{
1270 struct page *page;
1271
1272 if (!pfn_valid(pfn))
1273 return NULL;
1274
1275 page = pfn_to_page(pfn);
1276 if (page_zone(page) != zone)
1277 return NULL;
1278
1279 BUG_ON(PageHighMem(page));
1280
1281 if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1282 return NULL;
1283
1284 if (PageReserved(page)
1285 && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1286 return NULL;
1287
1288 if (page_is_guard(page))
1289 return NULL;
1290
1291 return page;
1292}
1293
1294/**
1295 * count_data_pages - Compute the total number of saveable non-highmem pages.
1296 */
1297static unsigned int count_data_pages(void)
1298{
1299 struct zone *zone;
1300 unsigned long pfn, max_zone_pfn;
1301 unsigned int n = 0;
1302
1303 for_each_populated_zone(zone) {
1304 if (is_highmem(zone))
1305 continue;
1306
1307 mark_free_pages(zone);
1308 max_zone_pfn = zone_end_pfn(zone);
1309 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1310 if (saveable_page(zone, pfn))
1311 n++;
1312 }
1313 return n;
1314}
1315
1316/*
1317 * This is needed, because copy_page and memcpy are not usable for copying
1318 * task structs.
1319 */
1320static inline void do_copy_page(long *dst, long *src)
1321{
1322 int n;
1323
1324 for (n = PAGE_SIZE / sizeof(long); n; n--)
1325 *dst++ = *src++;
1326}
1327
1328/**
1329 * safe_copy_page - Copy a page in a safe way.
1330 *
1331 * Check if the page we are going to copy is marked as present in the kernel
1332 * page tables (this always is the case if CONFIG_DEBUG_PAGEALLOC is not set
1333 * and in that case kernel_page_present() always returns 'true').
1334 */
1335static void safe_copy_page(void *dst, struct page *s_page)
1336{
1337 if (kernel_page_present(s_page)) {
1338 do_copy_page(dst, page_address(s_page));
1339 } else {
1340 kernel_map_pages(s_page, 1, 1);
1341 do_copy_page(dst, page_address(s_page));
1342 kernel_map_pages(s_page, 1, 0);
1343 }
1344}
1345
1346#ifdef CONFIG_HIGHMEM
1347static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1348{
1349 return is_highmem(zone) ?
1350 saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1351}
1352
1353static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1354{
1355 struct page *s_page, *d_page;
1356 void *src, *dst;
1357
1358 s_page = pfn_to_page(src_pfn);
1359 d_page = pfn_to_page(dst_pfn);
1360 if (PageHighMem(s_page)) {
1361 src = kmap_atomic(s_page);
1362 dst = kmap_atomic(d_page);
1363 do_copy_page(dst, src);
1364 kunmap_atomic(dst);
1365 kunmap_atomic(src);
1366 } else {
1367 if (PageHighMem(d_page)) {
1368 /*
1369 * The page pointed to by src may contain some kernel
1370 * data modified by kmap_atomic()
1371 */
1372 safe_copy_page(buffer, s_page);
1373 dst = kmap_atomic(d_page);
1374 copy_page(dst, buffer);
1375 kunmap_atomic(dst);
1376 } else {
1377 safe_copy_page(page_address(d_page), s_page);
1378 }
1379 }
1380}
1381#else
1382#define page_is_saveable(zone, pfn) saveable_page(zone, pfn)
1383
1384static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1385{
1386 safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1387 pfn_to_page(src_pfn));
1388}
1389#endif /* CONFIG_HIGHMEM */
1390
1391static void copy_data_pages(struct memory_bitmap *copy_bm,
1392 struct memory_bitmap *orig_bm)
1393{
1394 struct zone *zone;
1395 unsigned long pfn;
1396
1397 for_each_populated_zone(zone) {
1398 unsigned long max_zone_pfn;
1399
1400 mark_free_pages(zone);
1401 max_zone_pfn = zone_end_pfn(zone);
1402 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1403 if (page_is_saveable(zone, pfn))
1404 memory_bm_set_bit(orig_bm, pfn);
1405 }
1406 memory_bm_position_reset(orig_bm);
1407 memory_bm_position_reset(copy_bm);
1408 for(;;) {
1409 pfn = memory_bm_next_pfn(orig_bm);
1410 if (unlikely(pfn == BM_END_OF_MAP))
1411 break;
1412 copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1413 }
1414}
1415
1416/* Total number of image pages */
1417static unsigned int nr_copy_pages;
1418/* Number of pages needed for saving the original pfns of the image pages */
1419static unsigned int nr_meta_pages;
1420/*
1421 * Numbers of normal and highmem page frames allocated for hibernation image
1422 * before suspending devices.
1423 */
1424unsigned int alloc_normal, alloc_highmem;
1425/*
1426 * Memory bitmap used for marking saveable pages (during hibernation) or
1427 * hibernation image pages (during restore)
1428 */
1429static struct memory_bitmap orig_bm;
1430/*
1431 * Memory bitmap used during hibernation for marking allocated page frames that
1432 * will contain copies of saveable pages. During restore it is initially used
1433 * for marking hibernation image pages, but then the set bits from it are
1434 * duplicated in @orig_bm and it is released. On highmem systems it is next
1435 * used for marking "safe" highmem pages, but it has to be reinitialized for
1436 * this purpose.
1437 */
1438static struct memory_bitmap copy_bm;
1439
1440/**
1441 * swsusp_free - Free pages allocated for hibernation image.
1442 *
1443 * Image pages are alocated before snapshot creation, so they need to be
1444 * released after resume.
1445 */
1446void swsusp_free(void)
1447{
1448 unsigned long fb_pfn, fr_pfn;
1449
1450 if (!forbidden_pages_map || !free_pages_map)
1451 goto out;
1452
1453 memory_bm_position_reset(forbidden_pages_map);
1454 memory_bm_position_reset(free_pages_map);
1455
1456loop:
1457 fr_pfn = memory_bm_next_pfn(free_pages_map);
1458 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1459
1460 /*
1461 * Find the next bit set in both bitmaps. This is guaranteed to
1462 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1463 */
1464 do {
1465 if (fb_pfn < fr_pfn)
1466 fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1467 if (fr_pfn < fb_pfn)
1468 fr_pfn = memory_bm_next_pfn(free_pages_map);
1469 } while (fb_pfn != fr_pfn);
1470
1471 if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1472 struct page *page = pfn_to_page(fr_pfn);
1473
1474 memory_bm_clear_current(forbidden_pages_map);
1475 memory_bm_clear_current(free_pages_map);
1476 hibernate_restore_unprotect_page(page_address(page));
1477 __free_page(page);
1478 goto loop;
1479 }
1480
1481out:
1482 nr_copy_pages = 0;
1483 nr_meta_pages = 0;
1484 restore_pblist = NULL;
1485 buffer = NULL;
1486 alloc_normal = 0;
1487 alloc_highmem = 0;
1488 hibernate_restore_protection_end();
1489}
1490
1491/* Helper functions used for the shrinking of memory. */
1492
1493#define GFP_IMAGE (GFP_KERNEL | __GFP_NOWARN)
1494
1495/**
1496 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1497 * @nr_pages: Number of page frames to allocate.
1498 * @mask: GFP flags to use for the allocation.
1499 *
1500 * Return value: Number of page frames actually allocated
1501 */
1502static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1503{
1504 unsigned long nr_alloc = 0;
1505
1506 while (nr_pages > 0) {
1507 struct page *page;
1508
1509 page = alloc_image_page(mask);
1510 if (!page)
1511 break;
1512 memory_bm_set_bit(©_bm, page_to_pfn(page));
1513 if (PageHighMem(page))
1514 alloc_highmem++;
1515 else
1516 alloc_normal++;
1517 nr_pages--;
1518 nr_alloc++;
1519 }
1520
1521 return nr_alloc;
1522}
1523
1524static unsigned long preallocate_image_memory(unsigned long nr_pages,
1525 unsigned long avail_normal)
1526{
1527 unsigned long alloc;
1528
1529 if (avail_normal <= alloc_normal)
1530 return 0;
1531
1532 alloc = avail_normal - alloc_normal;
1533 if (nr_pages < alloc)
1534 alloc = nr_pages;
1535
1536 return preallocate_image_pages(alloc, GFP_IMAGE);
1537}
1538
1539#ifdef CONFIG_HIGHMEM
1540static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1541{
1542 return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1543}
1544
1545/**
1546 * __fraction - Compute (an approximation of) x * (multiplier / base).
1547 */
1548static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1549{
1550 x *= multiplier;
1551 do_div(x, base);
1552 return (unsigned long)x;
1553}
1554
1555static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1556 unsigned long highmem,
1557 unsigned long total)
1558{
1559 unsigned long alloc = __fraction(nr_pages, highmem, total);
1560
1561 return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1562}
1563#else /* CONFIG_HIGHMEM */
1564static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1565{
1566 return 0;
1567}
1568
1569static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1570 unsigned long highmem,
1571 unsigned long total)
1572{
1573 return 0;
1574}
1575#endif /* CONFIG_HIGHMEM */
1576
1577/**
1578 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1579 */
1580static unsigned long free_unnecessary_pages(void)
1581{
1582 unsigned long save, to_free_normal, to_free_highmem, free;
1583
1584 save = count_data_pages();
1585 if (alloc_normal >= save) {
1586 to_free_normal = alloc_normal - save;
1587 save = 0;
1588 } else {
1589 to_free_normal = 0;
1590 save -= alloc_normal;
1591 }
1592 save += count_highmem_pages();
1593 if (alloc_highmem >= save) {
1594 to_free_highmem = alloc_highmem - save;
1595 } else {
1596 to_free_highmem = 0;
1597 save -= alloc_highmem;
1598 if (to_free_normal > save)
1599 to_free_normal -= save;
1600 else
1601 to_free_normal = 0;
1602 }
1603 free = to_free_normal + to_free_highmem;
1604
1605 memory_bm_position_reset(©_bm);
1606
1607 while (to_free_normal > 0 || to_free_highmem > 0) {
1608 unsigned long pfn = memory_bm_next_pfn(©_bm);
1609 struct page *page = pfn_to_page(pfn);
1610
1611 if (PageHighMem(page)) {
1612 if (!to_free_highmem)
1613 continue;
1614 to_free_highmem--;
1615 alloc_highmem--;
1616 } else {
1617 if (!to_free_normal)
1618 continue;
1619 to_free_normal--;
1620 alloc_normal--;
1621 }
1622 memory_bm_clear_bit(©_bm, pfn);
1623 swsusp_unset_page_forbidden(page);
1624 swsusp_unset_page_free(page);
1625 __free_page(page);
1626 }
1627
1628 return free;
1629}
1630
1631/**
1632 * minimum_image_size - Estimate the minimum acceptable size of an image.
1633 * @saveable: Number of saveable pages in the system.
1634 *
1635 * We want to avoid attempting to free too much memory too hard, so estimate the
1636 * minimum acceptable size of a hibernation image to use as the lower limit for
1637 * preallocating memory.
1638 *
1639 * We assume that the minimum image size should be proportional to
1640 *
1641 * [number of saveable pages] - [number of pages that can be freed in theory]
1642 *
1643 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1644 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages,
1645 * minus mapped file pages.
1646 */
1647static unsigned long minimum_image_size(unsigned long saveable)
1648{
1649 unsigned long size;
1650
1651 size = global_page_state(NR_SLAB_RECLAIMABLE)
1652 + global_node_page_state(NR_ACTIVE_ANON)
1653 + global_node_page_state(NR_INACTIVE_ANON)
1654 + global_node_page_state(NR_ACTIVE_FILE)
1655 + global_node_page_state(NR_INACTIVE_FILE)
1656 - global_node_page_state(NR_FILE_MAPPED);
1657
1658 return saveable <= size ? 0 : saveable - size;
1659}
1660
1661/**
1662 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1663 *
1664 * To create a hibernation image it is necessary to make a copy of every page
1665 * frame in use. We also need a number of page frames to be free during
1666 * hibernation for allocations made while saving the image and for device
1667 * drivers, in case they need to allocate memory from their hibernation
1668 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1669 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1670 * /sys/power/reserved_size, respectively). To make this happen, we compute the
1671 * total number of available page frames and allocate at least
1672 *
1673 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1674 * + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1675 *
1676 * of them, which corresponds to the maximum size of a hibernation image.
1677 *
1678 * If image_size is set below the number following from the above formula,
1679 * the preallocation of memory is continued until the total number of saveable
1680 * pages in the system is below the requested image size or the minimum
1681 * acceptable image size returned by minimum_image_size(), whichever is greater.
1682 */
1683int hibernate_preallocate_memory(void)
1684{
1685 struct zone *zone;
1686 unsigned long saveable, size, max_size, count, highmem, pages = 0;
1687 unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1688 ktime_t start, stop;
1689 int error;
1690
1691 printk(KERN_INFO "PM: Preallocating image memory... ");
1692 start = ktime_get();
1693
1694 error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1695 if (error)
1696 goto err_out;
1697
1698 error = memory_bm_create(©_bm, GFP_IMAGE, PG_ANY);
1699 if (error)
1700 goto err_out;
1701
1702 alloc_normal = 0;
1703 alloc_highmem = 0;
1704
1705 /* Count the number of saveable data pages. */
1706 save_highmem = count_highmem_pages();
1707 saveable = count_data_pages();
1708
1709 /*
1710 * Compute the total number of page frames we can use (count) and the
1711 * number of pages needed for image metadata (size).
1712 */
1713 count = saveable;
1714 saveable += save_highmem;
1715 highmem = save_highmem;
1716 size = 0;
1717 for_each_populated_zone(zone) {
1718 size += snapshot_additional_pages(zone);
1719 if (is_highmem(zone))
1720 highmem += zone_page_state(zone, NR_FREE_PAGES);
1721 else
1722 count += zone_page_state(zone, NR_FREE_PAGES);
1723 }
1724 avail_normal = count;
1725 count += highmem;
1726 count -= totalreserve_pages;
1727
1728 /* Add number of pages required for page keys (s390 only). */
1729 size += page_key_additional_pages(saveable);
1730
1731 /* Compute the maximum number of saveable pages to leave in memory. */
1732 max_size = (count - (size + PAGES_FOR_IO)) / 2
1733 - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1734 /* Compute the desired number of image pages specified by image_size. */
1735 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1736 if (size > max_size)
1737 size = max_size;
1738 /*
1739 * If the desired number of image pages is at least as large as the
1740 * current number of saveable pages in memory, allocate page frames for
1741 * the image and we're done.
1742 */
1743 if (size >= saveable) {
1744 pages = preallocate_image_highmem(save_highmem);
1745 pages += preallocate_image_memory(saveable - pages, avail_normal);
1746 goto out;
1747 }
1748
1749 /* Estimate the minimum size of the image. */
1750 pages = minimum_image_size(saveable);
1751 /*
1752 * To avoid excessive pressure on the normal zone, leave room in it to
1753 * accommodate an image of the minimum size (unless it's already too
1754 * small, in which case don't preallocate pages from it at all).
1755 */
1756 if (avail_normal > pages)
1757 avail_normal -= pages;
1758 else
1759 avail_normal = 0;
1760 if (size < pages)
1761 size = min_t(unsigned long, pages, max_size);
1762
1763 /*
1764 * Let the memory management subsystem know that we're going to need a
1765 * large number of page frames to allocate and make it free some memory.
1766 * NOTE: If this is not done, performance will be hurt badly in some
1767 * test cases.
1768 */
1769 shrink_all_memory(saveable - size);
1770
1771 /*
1772 * The number of saveable pages in memory was too high, so apply some
1773 * pressure to decrease it. First, make room for the largest possible
1774 * image and fail if that doesn't work. Next, try to decrease the size
1775 * of the image as much as indicated by 'size' using allocations from
1776 * highmem and non-highmem zones separately.
1777 */
1778 pages_highmem = preallocate_image_highmem(highmem / 2);
1779 alloc = count - max_size;
1780 if (alloc > pages_highmem)
1781 alloc -= pages_highmem;
1782 else
1783 alloc = 0;
1784 pages = preallocate_image_memory(alloc, avail_normal);
1785 if (pages < alloc) {
1786 /* We have exhausted non-highmem pages, try highmem. */
1787 alloc -= pages;
1788 pages += pages_highmem;
1789 pages_highmem = preallocate_image_highmem(alloc);
1790 if (pages_highmem < alloc)
1791 goto err_out;
1792 pages += pages_highmem;
1793 /*
1794 * size is the desired number of saveable pages to leave in
1795 * memory, so try to preallocate (all memory - size) pages.
1796 */
1797 alloc = (count - pages) - size;
1798 pages += preallocate_image_highmem(alloc);
1799 } else {
1800 /*
1801 * There are approximately max_size saveable pages at this point
1802 * and we want to reduce this number down to size.
1803 */
1804 alloc = max_size - size;
1805 size = preallocate_highmem_fraction(alloc, highmem, count);
1806 pages_highmem += size;
1807 alloc -= size;
1808 size = preallocate_image_memory(alloc, avail_normal);
1809 pages_highmem += preallocate_image_highmem(alloc - size);
1810 pages += pages_highmem + size;
1811 }
1812
1813 /*
1814 * We only need as many page frames for the image as there are saveable
1815 * pages in memory, but we have allocated more. Release the excessive
1816 * ones now.
1817 */
1818 pages -= free_unnecessary_pages();
1819
1820 out:
1821 stop = ktime_get();
1822 printk(KERN_CONT "done (allocated %lu pages)\n", pages);
1823 swsusp_show_speed(start, stop, pages, "Allocated");
1824
1825 return 0;
1826
1827 err_out:
1828 printk(KERN_CONT "\n");
1829 swsusp_free();
1830 return -ENOMEM;
1831}
1832
1833#ifdef CONFIG_HIGHMEM
1834/**
1835 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1836 *
1837 * Compute the number of non-highmem pages that will be necessary for creating
1838 * copies of highmem pages.
1839 */
1840static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1841{
1842 unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1843
1844 if (free_highmem >= nr_highmem)
1845 nr_highmem = 0;
1846 else
1847 nr_highmem -= free_highmem;
1848
1849 return nr_highmem;
1850}
1851#else
1852static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1853#endif /* CONFIG_HIGHMEM */
1854
1855/**
1856 * enough_free_mem - Check if there is enough free memory for the image.
1857 */
1858static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1859{
1860 struct zone *zone;
1861 unsigned int free = alloc_normal;
1862
1863 for_each_populated_zone(zone)
1864 if (!is_highmem(zone))
1865 free += zone_page_state(zone, NR_FREE_PAGES);
1866
1867 nr_pages += count_pages_for_highmem(nr_highmem);
1868 pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n",
1869 nr_pages, PAGES_FOR_IO, free);
1870
1871 return free > nr_pages + PAGES_FOR_IO;
1872}
1873
1874#ifdef CONFIG_HIGHMEM
1875/**
1876 * get_highmem_buffer - Allocate a buffer for highmem pages.
1877 *
1878 * If there are some highmem pages in the hibernation image, we may need a
1879 * buffer to copy them and/or load their data.
1880 */
1881static inline int get_highmem_buffer(int safe_needed)
1882{
1883 buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
1884 return buffer ? 0 : -ENOMEM;
1885}
1886
1887/**
1888 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1889 *
1890 * Try to allocate as many pages as needed, but if the number of free highmem
1891 * pages is less than that, allocate them all.
1892 */
1893static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1894 unsigned int nr_highmem)
1895{
1896 unsigned int to_alloc = count_free_highmem_pages();
1897
1898 if (to_alloc > nr_highmem)
1899 to_alloc = nr_highmem;
1900
1901 nr_highmem -= to_alloc;
1902 while (to_alloc-- > 0) {
1903 struct page *page;
1904
1905 page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1906 memory_bm_set_bit(bm, page_to_pfn(page));
1907 }
1908 return nr_highmem;
1909}
1910#else
1911static inline int get_highmem_buffer(int safe_needed) { return 0; }
1912
1913static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1914 unsigned int n) { return 0; }
1915#endif /* CONFIG_HIGHMEM */
1916
1917/**
1918 * swsusp_alloc - Allocate memory for hibernation image.
1919 *
1920 * We first try to allocate as many highmem pages as there are
1921 * saveable highmem pages in the system. If that fails, we allocate
1922 * non-highmem pages for the copies of the remaining highmem ones.
1923 *
1924 * In this approach it is likely that the copies of highmem pages will
1925 * also be located in the high memory, because of the way in which
1926 * copy_data_pages() works.
1927 */
1928static int swsusp_alloc(struct memory_bitmap *orig_bm,
1929 struct memory_bitmap *copy_bm,
1930 unsigned int nr_pages, unsigned int nr_highmem)
1931{
1932 if (nr_highmem > 0) {
1933 if (get_highmem_buffer(PG_ANY))
1934 goto err_out;
1935 if (nr_highmem > alloc_highmem) {
1936 nr_highmem -= alloc_highmem;
1937 nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1938 }
1939 }
1940 if (nr_pages > alloc_normal) {
1941 nr_pages -= alloc_normal;
1942 while (nr_pages-- > 0) {
1943 struct page *page;
1944
1945 page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
1946 if (!page)
1947 goto err_out;
1948 memory_bm_set_bit(copy_bm, page_to_pfn(page));
1949 }
1950 }
1951
1952 return 0;
1953
1954 err_out:
1955 swsusp_free();
1956 return -ENOMEM;
1957}
1958
1959asmlinkage __visible int swsusp_save(void)
1960{
1961 unsigned int nr_pages, nr_highmem;
1962
1963 printk(KERN_INFO "PM: Creating hibernation image:\n");
1964
1965 drain_local_pages(NULL);
1966 nr_pages = count_data_pages();
1967 nr_highmem = count_highmem_pages();
1968 printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem);
1969
1970 if (!enough_free_mem(nr_pages, nr_highmem)) {
1971 printk(KERN_ERR "PM: Not enough free memory\n");
1972 return -ENOMEM;
1973 }
1974
1975 if (swsusp_alloc(&orig_bm, ©_bm, nr_pages, nr_highmem)) {
1976 printk(KERN_ERR "PM: Memory allocation failed\n");
1977 return -ENOMEM;
1978 }
1979
1980 /*
1981 * During allocating of suspend pagedir, new cold pages may appear.
1982 * Kill them.
1983 */
1984 drain_local_pages(NULL);
1985 copy_data_pages(©_bm, &orig_bm);
1986
1987 /*
1988 * End of critical section. From now on, we can write to memory,
1989 * but we should not touch disk. This specially means we must _not_
1990 * touch swap space! Except we must write out our image of course.
1991 */
1992
1993 nr_pages += nr_highmem;
1994 nr_copy_pages = nr_pages;
1995 nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1996
1997 printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n",
1998 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 printk(KERN_ERR "PM: 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 */