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