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 
31 #include <asm/uaccess.h>
32 #include <asm/mmu_context.h>
33 #include <asm/pgtable.h>
34 #include <asm/tlbflush.h>
35 #include <asm/io.h>
36 
37 #include "power.h"
38 
39 static int swsusp_page_is_free(struct page *);
40 static void swsusp_set_page_forbidden(struct page *);
41 static void swsusp_unset_page_forbidden(struct page *);
42 
43 /*
44  * Number of bytes to reserve for memory allocations made by device drivers
45  * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
46  * cause image creation to fail (tunable via /sys/power/reserved_size).
47  */
48 unsigned long reserved_size;
49 
hibernate_reserved_size_init(void)50 void __init hibernate_reserved_size_init(void)
51 {
52 	reserved_size = SPARE_PAGES * PAGE_SIZE;
53 }
54 
55 /*
56  * Preferred image size in bytes (tunable via /sys/power/image_size).
57  * When it is set to N, swsusp will do its best to ensure the image
58  * size will not exceed N bytes, but if that is impossible, it will
59  * try to create the smallest image possible.
60  */
61 unsigned long image_size;
62 
hibernate_image_size_init(void)63 void __init hibernate_image_size_init(void)
64 {
65 	image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE;
66 }
67 
68 /* List of PBEs needed for restoring the pages that were allocated before
69  * the suspend and included in the suspend image, but have also been
70  * allocated by the "resume" kernel, so their contents cannot be written
71  * directly to their "original" page frames.
72  */
73 struct pbe *restore_pblist;
74 
75 /* Pointer to an auxiliary buffer (1 page) */
76 static void *buffer;
77 
78 /**
79  *	@safe_needed - on resume, for storing the PBE list and the image,
80  *	we can only use memory pages that do not conflict with the pages
81  *	used before suspend.  The unsafe pages have PageNosaveFree set
82  *	and we count them using unsafe_pages.
83  *
84  *	Each allocated image page is marked as PageNosave and PageNosaveFree
85  *	so that swsusp_free() can release it.
86  */
87 
88 #define PG_ANY		0
89 #define PG_SAFE		1
90 #define PG_UNSAFE_CLEAR	1
91 #define PG_UNSAFE_KEEP	0
92 
93 static unsigned int allocated_unsafe_pages;
94 
get_image_page(gfp_t gfp_mask,int safe_needed)95 static void *get_image_page(gfp_t gfp_mask, int safe_needed)
96 {
97 	void *res;
98 
99 	res = (void *)get_zeroed_page(gfp_mask);
100 	if (safe_needed)
101 		while (res && swsusp_page_is_free(virt_to_page(res))) {
102 			/* The page is unsafe, mark it for swsusp_free() */
103 			swsusp_set_page_forbidden(virt_to_page(res));
104 			allocated_unsafe_pages++;
105 			res = (void *)get_zeroed_page(gfp_mask);
106 		}
107 	if (res) {
108 		swsusp_set_page_forbidden(virt_to_page(res));
109 		swsusp_set_page_free(virt_to_page(res));
110 	}
111 	return res;
112 }
113 
get_safe_page(gfp_t gfp_mask)114 unsigned long get_safe_page(gfp_t gfp_mask)
115 {
116 	return (unsigned long)get_image_page(gfp_mask, PG_SAFE);
117 }
118 
alloc_image_page(gfp_t gfp_mask)119 static struct page *alloc_image_page(gfp_t gfp_mask)
120 {
121 	struct page *page;
122 
123 	page = alloc_page(gfp_mask);
124 	if (page) {
125 		swsusp_set_page_forbidden(page);
126 		swsusp_set_page_free(page);
127 	}
128 	return page;
129 }
130 
131 /**
132  *	free_image_page - free page represented by @addr, allocated with
133  *	get_image_page (page flags set by it must be cleared)
134  */
135 
free_image_page(void * addr,int clear_nosave_free)136 static inline void free_image_page(void *addr, int clear_nosave_free)
137 {
138 	struct page *page;
139 
140 	BUG_ON(!virt_addr_valid(addr));
141 
142 	page = virt_to_page(addr);
143 
144 	swsusp_unset_page_forbidden(page);
145 	if (clear_nosave_free)
146 		swsusp_unset_page_free(page);
147 
148 	__free_page(page);
149 }
150 
151 /* struct linked_page is used to build chains of pages */
152 
153 #define LINKED_PAGE_DATA_SIZE	(PAGE_SIZE - sizeof(void *))
154 
155 struct linked_page {
156 	struct linked_page *next;
157 	char data[LINKED_PAGE_DATA_SIZE];
158 } __attribute__((packed));
159 
160 static inline void
free_list_of_pages(struct linked_page * list,int clear_page_nosave)161 free_list_of_pages(struct linked_page *list, int clear_page_nosave)
162 {
163 	while (list) {
164 		struct linked_page *lp = list->next;
165 
166 		free_image_page(list, clear_page_nosave);
167 		list = lp;
168 	}
169 }
170 
171 /**
172   *	struct chain_allocator is used for allocating small objects out of
173   *	a linked list of pages called 'the chain'.
174   *
175   *	The chain grows each time when there is no room for a new object in
176   *	the current page.  The allocated objects cannot be freed individually.
177   *	It is only possible to free them all at once, by freeing the entire
178   *	chain.
179   *
180   *	NOTE: The chain allocator may be inefficient if the allocated objects
181   *	are not much smaller than PAGE_SIZE.
182   */
183 
184 struct chain_allocator {
185 	struct linked_page *chain;	/* the chain */
186 	unsigned int used_space;	/* total size of objects allocated out
187 					 * of the current page
188 					 */
189 	gfp_t gfp_mask;		/* mask for allocating pages */
190 	int safe_needed;	/* if set, only "safe" pages are allocated */
191 };
192 
193 static void
chain_init(struct chain_allocator * ca,gfp_t gfp_mask,int safe_needed)194 chain_init(struct chain_allocator *ca, gfp_t gfp_mask, int safe_needed)
195 {
196 	ca->chain = NULL;
197 	ca->used_space = LINKED_PAGE_DATA_SIZE;
198 	ca->gfp_mask = gfp_mask;
199 	ca->safe_needed = safe_needed;
200 }
201 
chain_alloc(struct chain_allocator * ca,unsigned int size)202 static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
203 {
204 	void *ret;
205 
206 	if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
207 		struct linked_page *lp;
208 
209 		lp = get_image_page(ca->gfp_mask, ca->safe_needed);
210 		if (!lp)
211 			return NULL;
212 
213 		lp->next = ca->chain;
214 		ca->chain = lp;
215 		ca->used_space = 0;
216 	}
217 	ret = ca->chain->data + ca->used_space;
218 	ca->used_space += size;
219 	return ret;
220 }
221 
222 /**
223  *	Data types related to memory bitmaps.
224  *
225  *	Memory bitmap is a structure consiting of many linked lists of
226  *	objects.  The main list's elements are of type struct zone_bitmap
227  *	and each of them corresonds to one zone.  For each zone bitmap
228  *	object there is a list of objects of type struct bm_block that
229  *	represent each blocks of bitmap in which information is stored.
230  *
231  *	struct memory_bitmap contains a pointer to the main list of zone
232  *	bitmap objects, a struct bm_position used for browsing the bitmap,
233  *	and a pointer to the list of pages used for allocating all of the
234  *	zone bitmap objects and bitmap block objects.
235  *
236  *	NOTE: It has to be possible to lay out the bitmap in memory
237  *	using only allocations of order 0.  Additionally, the bitmap is
238  *	designed to work with arbitrary number of zones (this is over the
239  *	top for now, but let's avoid making unnecessary assumptions ;-).
240  *
241  *	struct zone_bitmap contains a pointer to a list of bitmap block
242  *	objects and a pointer to the bitmap block object that has been
243  *	most recently used for setting bits.  Additionally, it contains the
244  *	pfns that correspond to the start and end of the represented zone.
245  *
246  *	struct bm_block contains a pointer to the memory page in which
247  *	information is stored (in the form of a block of bitmap)
248  *	It also contains the pfns that correspond to the start and end of
249  *	the represented memory area.
250  */
251 
252 #define BM_END_OF_MAP	(~0UL)
253 
254 #define BM_BITS_PER_BLOCK	(PAGE_SIZE * BITS_PER_BYTE)
255 
256 struct bm_block {
257 	struct list_head hook;	/* hook into a list of bitmap blocks */
258 	unsigned long start_pfn;	/* pfn represented by the first bit */
259 	unsigned long end_pfn;	/* pfn represented by the last bit plus 1 */
260 	unsigned long *data;	/* bitmap representing pages */
261 };
262 
bm_block_bits(struct bm_block * bb)263 static inline unsigned long bm_block_bits(struct bm_block *bb)
264 {
265 	return bb->end_pfn - bb->start_pfn;
266 }
267 
268 /* strcut bm_position is used for browsing memory bitmaps */
269 
270 struct bm_position {
271 	struct bm_block *block;
272 	int bit;
273 };
274 
275 struct memory_bitmap {
276 	struct list_head blocks;	/* list of bitmap blocks */
277 	struct linked_page *p_list;	/* list of pages used to store zone
278 					 * bitmap objects and bitmap block
279 					 * objects
280 					 */
281 	struct bm_position cur;	/* most recently used bit position */
282 };
283 
284 /* Functions that operate on memory bitmaps */
285 
memory_bm_position_reset(struct memory_bitmap * bm)286 static void memory_bm_position_reset(struct memory_bitmap *bm)
287 {
288 	bm->cur.block = list_entry(bm->blocks.next, struct bm_block, hook);
289 	bm->cur.bit = 0;
290 }
291 
292 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
293 
294 /**
295  *	create_bm_block_list - create a list of block bitmap objects
296  *	@pages - number of pages to track
297  *	@list - list to put the allocated blocks into
298  *	@ca - chain allocator to be used for allocating memory
299  */
create_bm_block_list(unsigned long pages,struct list_head * list,struct chain_allocator * ca)300 static int create_bm_block_list(unsigned long pages,
301 				struct list_head *list,
302 				struct chain_allocator *ca)
303 {
304 	unsigned int nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
305 
306 	while (nr_blocks-- > 0) {
307 		struct bm_block *bb;
308 
309 		bb = chain_alloc(ca, sizeof(struct bm_block));
310 		if (!bb)
311 			return -ENOMEM;
312 		list_add(&bb->hook, list);
313 	}
314 
315 	return 0;
316 }
317 
318 struct mem_extent {
319 	struct list_head hook;
320 	unsigned long start;
321 	unsigned long end;
322 };
323 
324 /**
325  *	free_mem_extents - free a list of memory extents
326  *	@list - list of extents to empty
327  */
free_mem_extents(struct list_head * list)328 static void free_mem_extents(struct list_head *list)
329 {
330 	struct mem_extent *ext, *aux;
331 
332 	list_for_each_entry_safe(ext, aux, list, hook) {
333 		list_del(&ext->hook);
334 		kfree(ext);
335 	}
336 }
337 
338 /**
339  *	create_mem_extents - create a list of memory extents representing
340  *	                     contiguous ranges of PFNs
341  *	@list - list to put the extents into
342  *	@gfp_mask - mask to use for memory allocations
343  */
create_mem_extents(struct list_head * list,gfp_t gfp_mask)344 static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
345 {
346 	struct zone *zone;
347 
348 	INIT_LIST_HEAD(list);
349 
350 	for_each_populated_zone(zone) {
351 		unsigned long zone_start, zone_end;
352 		struct mem_extent *ext, *cur, *aux;
353 
354 		zone_start = zone->zone_start_pfn;
355 		zone_end = zone->zone_start_pfn + zone->spanned_pages;
356 
357 		list_for_each_entry(ext, list, hook)
358 			if (zone_start <= ext->end)
359 				break;
360 
361 		if (&ext->hook == list || zone_end < ext->start) {
362 			/* New extent is necessary */
363 			struct mem_extent *new_ext;
364 
365 			new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
366 			if (!new_ext) {
367 				free_mem_extents(list);
368 				return -ENOMEM;
369 			}
370 			new_ext->start = zone_start;
371 			new_ext->end = zone_end;
372 			list_add_tail(&new_ext->hook, &ext->hook);
373 			continue;
374 		}
375 
376 		/* Merge this zone's range of PFNs with the existing one */
377 		if (zone_start < ext->start)
378 			ext->start = zone_start;
379 		if (zone_end > ext->end)
380 			ext->end = zone_end;
381 
382 		/* More merging may be possible */
383 		cur = ext;
384 		list_for_each_entry_safe_continue(cur, aux, list, hook) {
385 			if (zone_end < cur->start)
386 				break;
387 			if (zone_end < cur->end)
388 				ext->end = cur->end;
389 			list_del(&cur->hook);
390 			kfree(cur);
391 		}
392 	}
393 
394 	return 0;
395 }
396 
397 /**
398   *	memory_bm_create - allocate memory for a memory bitmap
399   */
400 static int
memory_bm_create(struct memory_bitmap * bm,gfp_t gfp_mask,int safe_needed)401 memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask, int safe_needed)
402 {
403 	struct chain_allocator ca;
404 	struct list_head mem_extents;
405 	struct mem_extent *ext;
406 	int error;
407 
408 	chain_init(&ca, gfp_mask, safe_needed);
409 	INIT_LIST_HEAD(&bm->blocks);
410 
411 	error = create_mem_extents(&mem_extents, gfp_mask);
412 	if (error)
413 		return error;
414 
415 	list_for_each_entry(ext, &mem_extents, hook) {
416 		struct bm_block *bb;
417 		unsigned long pfn = ext->start;
418 		unsigned long pages = ext->end - ext->start;
419 
420 		bb = list_entry(bm->blocks.prev, struct bm_block, hook);
421 
422 		error = create_bm_block_list(pages, bm->blocks.prev, &ca);
423 		if (error)
424 			goto Error;
425 
426 		list_for_each_entry_continue(bb, &bm->blocks, hook) {
427 			bb->data = get_image_page(gfp_mask, safe_needed);
428 			if (!bb->data) {
429 				error = -ENOMEM;
430 				goto Error;
431 			}
432 
433 			bb->start_pfn = pfn;
434 			if (pages >= BM_BITS_PER_BLOCK) {
435 				pfn += BM_BITS_PER_BLOCK;
436 				pages -= BM_BITS_PER_BLOCK;
437 			} else {
438 				/* This is executed only once in the loop */
439 				pfn += pages;
440 			}
441 			bb->end_pfn = pfn;
442 		}
443 	}
444 
445 	bm->p_list = ca.chain;
446 	memory_bm_position_reset(bm);
447  Exit:
448 	free_mem_extents(&mem_extents);
449 	return error;
450 
451  Error:
452 	bm->p_list = ca.chain;
453 	memory_bm_free(bm, PG_UNSAFE_CLEAR);
454 	goto Exit;
455 }
456 
457 /**
458   *	memory_bm_free - free memory occupied by the memory bitmap @bm
459   */
memory_bm_free(struct memory_bitmap * bm,int clear_nosave_free)460 static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
461 {
462 	struct bm_block *bb;
463 
464 	list_for_each_entry(bb, &bm->blocks, hook)
465 		if (bb->data)
466 			free_image_page(bb->data, clear_nosave_free);
467 
468 	free_list_of_pages(bm->p_list, clear_nosave_free);
469 
470 	INIT_LIST_HEAD(&bm->blocks);
471 }
472 
473 /**
474  *	memory_bm_find_bit - find the bit in the bitmap @bm that corresponds
475  *	to given pfn.  The cur_zone_bm member of @bm and the cur_block member
476  *	of @bm->cur_zone_bm are updated.
477  */
memory_bm_find_bit(struct memory_bitmap * bm,unsigned long pfn,void ** addr,unsigned int * bit_nr)478 static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
479 				void **addr, unsigned int *bit_nr)
480 {
481 	struct bm_block *bb;
482 
483 	/*
484 	 * Check if the pfn corresponds to the current bitmap block and find
485 	 * the block where it fits if this is not the case.
486 	 */
487 	bb = bm->cur.block;
488 	if (pfn < bb->start_pfn)
489 		list_for_each_entry_continue_reverse(bb, &bm->blocks, hook)
490 			if (pfn >= bb->start_pfn)
491 				break;
492 
493 	if (pfn >= bb->end_pfn)
494 		list_for_each_entry_continue(bb, &bm->blocks, hook)
495 			if (pfn >= bb->start_pfn && pfn < bb->end_pfn)
496 				break;
497 
498 	if (&bb->hook == &bm->blocks)
499 		return -EFAULT;
500 
501 	/* The block has been found */
502 	bm->cur.block = bb;
503 	pfn -= bb->start_pfn;
504 	bm->cur.bit = pfn + 1;
505 	*bit_nr = pfn;
506 	*addr = bb->data;
507 	return 0;
508 }
509 
memory_bm_set_bit(struct memory_bitmap * bm,unsigned long pfn)510 static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
511 {
512 	void *addr;
513 	unsigned int bit;
514 	int error;
515 
516 	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
517 	BUG_ON(error);
518 	set_bit(bit, addr);
519 }
520 
mem_bm_set_bit_check(struct memory_bitmap * bm,unsigned long pfn)521 static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
522 {
523 	void *addr;
524 	unsigned int bit;
525 	int error;
526 
527 	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
528 	if (!error)
529 		set_bit(bit, addr);
530 	return error;
531 }
532 
memory_bm_clear_bit(struct memory_bitmap * bm,unsigned long pfn)533 static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
534 {
535 	void *addr;
536 	unsigned int bit;
537 	int error;
538 
539 	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
540 	BUG_ON(error);
541 	clear_bit(bit, addr);
542 }
543 
memory_bm_test_bit(struct memory_bitmap * bm,unsigned long pfn)544 static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
545 {
546 	void *addr;
547 	unsigned int bit;
548 	int error;
549 
550 	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
551 	BUG_ON(error);
552 	return test_bit(bit, addr);
553 }
554 
memory_bm_pfn_present(struct memory_bitmap * bm,unsigned long pfn)555 static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
556 {
557 	void *addr;
558 	unsigned int bit;
559 
560 	return !memory_bm_find_bit(bm, pfn, &addr, &bit);
561 }
562 
563 /**
564  *	memory_bm_next_pfn - find the pfn that corresponds to the next set bit
565  *	in the bitmap @bm.  If the pfn cannot be found, BM_END_OF_MAP is
566  *	returned.
567  *
568  *	It is required to run memory_bm_position_reset() before the first call to
569  *	this function.
570  */
571 
memory_bm_next_pfn(struct memory_bitmap * bm)572 static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
573 {
574 	struct bm_block *bb;
575 	int bit;
576 
577 	bb = bm->cur.block;
578 	do {
579 		bit = bm->cur.bit;
580 		bit = find_next_bit(bb->data, bm_block_bits(bb), bit);
581 		if (bit < bm_block_bits(bb))
582 			goto Return_pfn;
583 
584 		bb = list_entry(bb->hook.next, struct bm_block, hook);
585 		bm->cur.block = bb;
586 		bm->cur.bit = 0;
587 	} while (&bb->hook != &bm->blocks);
588 
589 	memory_bm_position_reset(bm);
590 	return BM_END_OF_MAP;
591 
592  Return_pfn:
593 	bm->cur.bit = bit + 1;
594 	return bb->start_pfn + bit;
595 }
596 
597 /**
598  *	This structure represents a range of page frames the contents of which
599  *	should not be saved during the suspend.
600  */
601 
602 struct nosave_region {
603 	struct list_head list;
604 	unsigned long start_pfn;
605 	unsigned long end_pfn;
606 };
607 
608 static LIST_HEAD(nosave_regions);
609 
610 /**
611  *	register_nosave_region - register a range of page frames the contents
612  *	of which should not be saved during the suspend (to be used in the early
613  *	initialization code)
614  */
615 
616 void __init
__register_nosave_region(unsigned long start_pfn,unsigned long end_pfn,int use_kmalloc)617 __register_nosave_region(unsigned long start_pfn, unsigned long end_pfn,
618 			 int use_kmalloc)
619 {
620 	struct nosave_region *region;
621 
622 	if (start_pfn >= end_pfn)
623 		return;
624 
625 	if (!list_empty(&nosave_regions)) {
626 		/* Try to extend the previous region (they should be sorted) */
627 		region = list_entry(nosave_regions.prev,
628 					struct nosave_region, list);
629 		if (region->end_pfn == start_pfn) {
630 			region->end_pfn = end_pfn;
631 			goto Report;
632 		}
633 	}
634 	if (use_kmalloc) {
635 		/* during init, this shouldn't fail */
636 		region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
637 		BUG_ON(!region);
638 	} else
639 		/* This allocation cannot fail */
640 		region = alloc_bootmem(sizeof(struct nosave_region));
641 	region->start_pfn = start_pfn;
642 	region->end_pfn = end_pfn;
643 	list_add_tail(&region->list, &nosave_regions);
644  Report:
645 	printk(KERN_INFO "PM: Registered nosave memory: %016lx - %016lx\n",
646 		start_pfn << PAGE_SHIFT, end_pfn << PAGE_SHIFT);
647 }
648 
649 /*
650  * Set bits in this map correspond to the page frames the contents of which
651  * should not be saved during the suspend.
652  */
653 static struct memory_bitmap *forbidden_pages_map;
654 
655 /* Set bits in this map correspond to free page frames. */
656 static struct memory_bitmap *free_pages_map;
657 
658 /*
659  * Each page frame allocated for creating the image is marked by setting the
660  * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
661  */
662 
swsusp_set_page_free(struct page * page)663 void swsusp_set_page_free(struct page *page)
664 {
665 	if (free_pages_map)
666 		memory_bm_set_bit(free_pages_map, page_to_pfn(page));
667 }
668 
swsusp_page_is_free(struct page * page)669 static int swsusp_page_is_free(struct page *page)
670 {
671 	return free_pages_map ?
672 		memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
673 }
674 
swsusp_unset_page_free(struct page * page)675 void swsusp_unset_page_free(struct page *page)
676 {
677 	if (free_pages_map)
678 		memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
679 }
680 
swsusp_set_page_forbidden(struct page * page)681 static void swsusp_set_page_forbidden(struct page *page)
682 {
683 	if (forbidden_pages_map)
684 		memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
685 }
686 
swsusp_page_is_forbidden(struct page * page)687 int swsusp_page_is_forbidden(struct page *page)
688 {
689 	return forbidden_pages_map ?
690 		memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
691 }
692 
swsusp_unset_page_forbidden(struct page * page)693 static void swsusp_unset_page_forbidden(struct page *page)
694 {
695 	if (forbidden_pages_map)
696 		memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
697 }
698 
699 /**
700  *	mark_nosave_pages - set bits corresponding to the page frames the
701  *	contents of which should not be saved in a given bitmap.
702  */
703 
mark_nosave_pages(struct memory_bitmap * bm)704 static void mark_nosave_pages(struct memory_bitmap *bm)
705 {
706 	struct nosave_region *region;
707 
708 	if (list_empty(&nosave_regions))
709 		return;
710 
711 	list_for_each_entry(region, &nosave_regions, list) {
712 		unsigned long pfn;
713 
714 		pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n",
715 			 (unsigned long long) region->start_pfn << PAGE_SHIFT,
716 			 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
717 				- 1);
718 
719 		for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
720 			if (pfn_valid(pfn)) {
721 				/*
722 				 * It is safe to ignore the result of
723 				 * mem_bm_set_bit_check() here, since we won't
724 				 * touch the PFNs for which the error is
725 				 * returned anyway.
726 				 */
727 				mem_bm_set_bit_check(bm, pfn);
728 			}
729 	}
730 }
731 
732 /**
733  *	create_basic_memory_bitmaps - create bitmaps needed for marking page
734  *	frames that should not be saved and free page frames.  The pointers
735  *	forbidden_pages_map and free_pages_map are only modified if everything
736  *	goes well, because we don't want the bits to be used before both bitmaps
737  *	are set up.
738  */
739 
create_basic_memory_bitmaps(void)740 int create_basic_memory_bitmaps(void)
741 {
742 	struct memory_bitmap *bm1, *bm2;
743 	int error = 0;
744 
745 	BUG_ON(forbidden_pages_map || free_pages_map);
746 
747 	bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
748 	if (!bm1)
749 		return -ENOMEM;
750 
751 	error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
752 	if (error)
753 		goto Free_first_object;
754 
755 	bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
756 	if (!bm2)
757 		goto Free_first_bitmap;
758 
759 	error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
760 	if (error)
761 		goto Free_second_object;
762 
763 	forbidden_pages_map = bm1;
764 	free_pages_map = bm2;
765 	mark_nosave_pages(forbidden_pages_map);
766 
767 	pr_debug("PM: Basic memory bitmaps created\n");
768 
769 	return 0;
770 
771  Free_second_object:
772 	kfree(bm2);
773  Free_first_bitmap:
774  	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
775  Free_first_object:
776 	kfree(bm1);
777 	return -ENOMEM;
778 }
779 
780 /**
781  *	free_basic_memory_bitmaps - free memory bitmaps allocated by
782  *	create_basic_memory_bitmaps().  The auxiliary pointers are necessary
783  *	so that the bitmaps themselves are not referred to while they are being
784  *	freed.
785  */
786 
free_basic_memory_bitmaps(void)787 void free_basic_memory_bitmaps(void)
788 {
789 	struct memory_bitmap *bm1, *bm2;
790 
791 	BUG_ON(!(forbidden_pages_map && free_pages_map));
792 
793 	bm1 = forbidden_pages_map;
794 	bm2 = free_pages_map;
795 	forbidden_pages_map = NULL;
796 	free_pages_map = NULL;
797 	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
798 	kfree(bm1);
799 	memory_bm_free(bm2, PG_UNSAFE_CLEAR);
800 	kfree(bm2);
801 
802 	pr_debug("PM: Basic memory bitmaps freed\n");
803 }
804 
805 /**
806  *	snapshot_additional_pages - estimate the number of additional pages
807  *	be needed for setting up the suspend image data structures for given
808  *	zone (usually the returned value is greater than the exact number)
809  */
810 
snapshot_additional_pages(struct zone * zone)811 unsigned int snapshot_additional_pages(struct zone *zone)
812 {
813 	unsigned int res;
814 
815 	res = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
816 	res += DIV_ROUND_UP(res * sizeof(struct bm_block),
817 			    LINKED_PAGE_DATA_SIZE);
818 	return 2 * res;
819 }
820 
821 #ifdef CONFIG_HIGHMEM
822 /**
823  *	count_free_highmem_pages - compute the total number of free highmem
824  *	pages, system-wide.
825  */
826 
count_free_highmem_pages(void)827 static unsigned int count_free_highmem_pages(void)
828 {
829 	struct zone *zone;
830 	unsigned int cnt = 0;
831 
832 	for_each_populated_zone(zone)
833 		if (is_highmem(zone))
834 			cnt += zone_page_state(zone, NR_FREE_PAGES);
835 
836 	return cnt;
837 }
838 
839 /**
840  *	saveable_highmem_page - Determine whether a highmem page should be
841  *	included in the suspend image.
842  *
843  *	We should save the page if it isn't Nosave or NosaveFree, or Reserved,
844  *	and it isn't a part of a free chunk of pages.
845  */
saveable_highmem_page(struct zone * zone,unsigned long pfn)846 static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
847 {
848 	struct page *page;
849 
850 	if (!pfn_valid(pfn))
851 		return NULL;
852 
853 	page = pfn_to_page(pfn);
854 	if (page_zone(page) != zone)
855 		return NULL;
856 
857 	BUG_ON(!PageHighMem(page));
858 
859 	if (swsusp_page_is_forbidden(page) ||  swsusp_page_is_free(page) ||
860 	    PageReserved(page))
861 		return NULL;
862 
863 	if (page_is_guard(page))
864 		return NULL;
865 
866 	return page;
867 }
868 
869 /**
870  *	count_highmem_pages - compute the total number of saveable highmem
871  *	pages.
872  */
873 
count_highmem_pages(void)874 static unsigned int count_highmem_pages(void)
875 {
876 	struct zone *zone;
877 	unsigned int n = 0;
878 
879 	for_each_populated_zone(zone) {
880 		unsigned long pfn, max_zone_pfn;
881 
882 		if (!is_highmem(zone))
883 			continue;
884 
885 		mark_free_pages(zone);
886 		max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
887 		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
888 			if (saveable_highmem_page(zone, pfn))
889 				n++;
890 	}
891 	return n;
892 }
893 #else
saveable_highmem_page(struct zone * z,unsigned long p)894 static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
895 {
896 	return NULL;
897 }
898 #endif /* CONFIG_HIGHMEM */
899 
900 /**
901  *	saveable_page - Determine whether a non-highmem page should be included
902  *	in the suspend image.
903  *
904  *	We should save the page if it isn't Nosave, and is not in the range
905  *	of pages statically defined as 'unsaveable', and it isn't a part of
906  *	a free chunk of pages.
907  */
saveable_page(struct zone * zone,unsigned long pfn)908 static struct page *saveable_page(struct zone *zone, unsigned long pfn)
909 {
910 	struct page *page;
911 
912 	if (!pfn_valid(pfn))
913 		return NULL;
914 
915 	page = pfn_to_page(pfn);
916 	if (page_zone(page) != zone)
917 		return NULL;
918 
919 	BUG_ON(PageHighMem(page));
920 
921 	if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
922 		return NULL;
923 
924 	if (PageReserved(page)
925 	    && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
926 		return NULL;
927 
928 	if (page_is_guard(page))
929 		return NULL;
930 
931 	return page;
932 }
933 
934 /**
935  *	count_data_pages - compute the total number of saveable non-highmem
936  *	pages.
937  */
938 
count_data_pages(void)939 static unsigned int count_data_pages(void)
940 {
941 	struct zone *zone;
942 	unsigned long pfn, max_zone_pfn;
943 	unsigned int n = 0;
944 
945 	for_each_populated_zone(zone) {
946 		if (is_highmem(zone))
947 			continue;
948 
949 		mark_free_pages(zone);
950 		max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
951 		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
952 			if (saveable_page(zone, pfn))
953 				n++;
954 	}
955 	return n;
956 }
957 
958 /* This is needed, because copy_page and memcpy are not usable for copying
959  * task structs.
960  */
do_copy_page(long * dst,long * src)961 static inline void do_copy_page(long *dst, long *src)
962 {
963 	int n;
964 
965 	for (n = PAGE_SIZE / sizeof(long); n; n--)
966 		*dst++ = *src++;
967 }
968 
969 
970 /**
971  *	safe_copy_page - check if the page we are going to copy is marked as
972  *		present in the kernel page tables (this always is the case if
973  *		CONFIG_DEBUG_PAGEALLOC is not set and in that case
974  *		kernel_page_present() always returns 'true').
975  */
safe_copy_page(void * dst,struct page * s_page)976 static void safe_copy_page(void *dst, struct page *s_page)
977 {
978 	if (kernel_page_present(s_page)) {
979 		do_copy_page(dst, page_address(s_page));
980 	} else {
981 		kernel_map_pages(s_page, 1, 1);
982 		do_copy_page(dst, page_address(s_page));
983 		kernel_map_pages(s_page, 1, 0);
984 	}
985 }
986 
987 
988 #ifdef CONFIG_HIGHMEM
989 static inline struct page *
page_is_saveable(struct zone * zone,unsigned long pfn)990 page_is_saveable(struct zone *zone, unsigned long pfn)
991 {
992 	return is_highmem(zone) ?
993 		saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
994 }
995 
copy_data_page(unsigned long dst_pfn,unsigned long src_pfn)996 static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
997 {
998 	struct page *s_page, *d_page;
999 	void *src, *dst;
1000 
1001 	s_page = pfn_to_page(src_pfn);
1002 	d_page = pfn_to_page(dst_pfn);
1003 	if (PageHighMem(s_page)) {
1004 		src = kmap_atomic(s_page);
1005 		dst = kmap_atomic(d_page);
1006 		do_copy_page(dst, src);
1007 		kunmap_atomic(dst);
1008 		kunmap_atomic(src);
1009 	} else {
1010 		if (PageHighMem(d_page)) {
1011 			/* Page pointed to by src may contain some kernel
1012 			 * data modified by kmap_atomic()
1013 			 */
1014 			safe_copy_page(buffer, s_page);
1015 			dst = kmap_atomic(d_page);
1016 			copy_page(dst, buffer);
1017 			kunmap_atomic(dst);
1018 		} else {
1019 			safe_copy_page(page_address(d_page), s_page);
1020 		}
1021 	}
1022 }
1023 #else
1024 #define page_is_saveable(zone, pfn)	saveable_page(zone, pfn)
1025 
copy_data_page(unsigned long dst_pfn,unsigned long src_pfn)1026 static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1027 {
1028 	safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1029 				pfn_to_page(src_pfn));
1030 }
1031 #endif /* CONFIG_HIGHMEM */
1032 
1033 static void
copy_data_pages(struct memory_bitmap * copy_bm,struct memory_bitmap * orig_bm)1034 copy_data_pages(struct memory_bitmap *copy_bm, struct memory_bitmap *orig_bm)
1035 {
1036 	struct zone *zone;
1037 	unsigned long pfn;
1038 
1039 	for_each_populated_zone(zone) {
1040 		unsigned long max_zone_pfn;
1041 
1042 		mark_free_pages(zone);
1043 		max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1044 		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1045 			if (page_is_saveable(zone, pfn))
1046 				memory_bm_set_bit(orig_bm, pfn);
1047 	}
1048 	memory_bm_position_reset(orig_bm);
1049 	memory_bm_position_reset(copy_bm);
1050 	for(;;) {
1051 		pfn = memory_bm_next_pfn(orig_bm);
1052 		if (unlikely(pfn == BM_END_OF_MAP))
1053 			break;
1054 		copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1055 	}
1056 }
1057 
1058 /* Total number of image pages */
1059 static unsigned int nr_copy_pages;
1060 /* Number of pages needed for saving the original pfns of the image pages */
1061 static unsigned int nr_meta_pages;
1062 /*
1063  * Numbers of normal and highmem page frames allocated for hibernation image
1064  * before suspending devices.
1065  */
1066 unsigned int alloc_normal, alloc_highmem;
1067 /*
1068  * Memory bitmap used for marking saveable pages (during hibernation) or
1069  * hibernation image pages (during restore)
1070  */
1071 static struct memory_bitmap orig_bm;
1072 /*
1073  * Memory bitmap used during hibernation for marking allocated page frames that
1074  * will contain copies of saveable pages.  During restore it is initially used
1075  * for marking hibernation image pages, but then the set bits from it are
1076  * duplicated in @orig_bm and it is released.  On highmem systems it is next
1077  * used for marking "safe" highmem pages, but it has to be reinitialized for
1078  * this purpose.
1079  */
1080 static struct memory_bitmap copy_bm;
1081 
1082 /**
1083  *	swsusp_free - free pages allocated for the suspend.
1084  *
1085  *	Suspend pages are alocated before the atomic copy is made, so we
1086  *	need to release them after the resume.
1087  */
1088 
swsusp_free(void)1089 void swsusp_free(void)
1090 {
1091 	struct zone *zone;
1092 	unsigned long pfn, max_zone_pfn;
1093 
1094 	for_each_populated_zone(zone) {
1095 		max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1096 		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1097 			if (pfn_valid(pfn)) {
1098 				struct page *page = pfn_to_page(pfn);
1099 
1100 				if (swsusp_page_is_forbidden(page) &&
1101 				    swsusp_page_is_free(page)) {
1102 					swsusp_unset_page_forbidden(page);
1103 					swsusp_unset_page_free(page);
1104 					__free_page(page);
1105 				}
1106 			}
1107 	}
1108 	nr_copy_pages = 0;
1109 	nr_meta_pages = 0;
1110 	restore_pblist = NULL;
1111 	buffer = NULL;
1112 	alloc_normal = 0;
1113 	alloc_highmem = 0;
1114 }
1115 
1116 /* Helper functions used for the shrinking of memory. */
1117 
1118 #define GFP_IMAGE	(GFP_KERNEL | __GFP_NOWARN)
1119 
1120 /**
1121  * preallocate_image_pages - Allocate a number of pages for hibernation image
1122  * @nr_pages: Number of page frames to allocate.
1123  * @mask: GFP flags to use for the allocation.
1124  *
1125  * Return value: Number of page frames actually allocated
1126  */
preallocate_image_pages(unsigned long nr_pages,gfp_t mask)1127 static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1128 {
1129 	unsigned long nr_alloc = 0;
1130 
1131 	while (nr_pages > 0) {
1132 		struct page *page;
1133 
1134 		page = alloc_image_page(mask);
1135 		if (!page)
1136 			break;
1137 		memory_bm_set_bit(&copy_bm, page_to_pfn(page));
1138 		if (PageHighMem(page))
1139 			alloc_highmem++;
1140 		else
1141 			alloc_normal++;
1142 		nr_pages--;
1143 		nr_alloc++;
1144 	}
1145 
1146 	return nr_alloc;
1147 }
1148 
preallocate_image_memory(unsigned long nr_pages,unsigned long avail_normal)1149 static unsigned long preallocate_image_memory(unsigned long nr_pages,
1150 					      unsigned long avail_normal)
1151 {
1152 	unsigned long alloc;
1153 
1154 	if (avail_normal <= alloc_normal)
1155 		return 0;
1156 
1157 	alloc = avail_normal - alloc_normal;
1158 	if (nr_pages < alloc)
1159 		alloc = nr_pages;
1160 
1161 	return preallocate_image_pages(alloc, GFP_IMAGE);
1162 }
1163 
1164 #ifdef CONFIG_HIGHMEM
preallocate_image_highmem(unsigned long nr_pages)1165 static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1166 {
1167 	return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1168 }
1169 
1170 /**
1171  *  __fraction - Compute (an approximation of) x * (multiplier / base)
1172  */
__fraction(u64 x,u64 multiplier,u64 base)1173 static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1174 {
1175 	x *= multiplier;
1176 	do_div(x, base);
1177 	return (unsigned long)x;
1178 }
1179 
preallocate_highmem_fraction(unsigned long nr_pages,unsigned long highmem,unsigned long total)1180 static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1181 						unsigned long highmem,
1182 						unsigned long total)
1183 {
1184 	unsigned long alloc = __fraction(nr_pages, highmem, total);
1185 
1186 	return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1187 }
1188 #else /* CONFIG_HIGHMEM */
preallocate_image_highmem(unsigned long nr_pages)1189 static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1190 {
1191 	return 0;
1192 }
1193 
preallocate_highmem_fraction(unsigned long nr_pages,unsigned long highmem,unsigned long total)1194 static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1195 						unsigned long highmem,
1196 						unsigned long total)
1197 {
1198 	return 0;
1199 }
1200 #endif /* CONFIG_HIGHMEM */
1201 
1202 /**
1203  * free_unnecessary_pages - Release preallocated pages not needed for the image
1204  */
free_unnecessary_pages(void)1205 static void free_unnecessary_pages(void)
1206 {
1207 	unsigned long save, to_free_normal, to_free_highmem;
1208 
1209 	save = count_data_pages();
1210 	if (alloc_normal >= save) {
1211 		to_free_normal = alloc_normal - save;
1212 		save = 0;
1213 	} else {
1214 		to_free_normal = 0;
1215 		save -= alloc_normal;
1216 	}
1217 	save += count_highmem_pages();
1218 	if (alloc_highmem >= save) {
1219 		to_free_highmem = alloc_highmem - save;
1220 	} else {
1221 		to_free_highmem = 0;
1222 		save -= alloc_highmem;
1223 		if (to_free_normal > save)
1224 			to_free_normal -= save;
1225 		else
1226 			to_free_normal = 0;
1227 	}
1228 
1229 	memory_bm_position_reset(&copy_bm);
1230 
1231 	while (to_free_normal > 0 || to_free_highmem > 0) {
1232 		unsigned long pfn = memory_bm_next_pfn(&copy_bm);
1233 		struct page *page = pfn_to_page(pfn);
1234 
1235 		if (PageHighMem(page)) {
1236 			if (!to_free_highmem)
1237 				continue;
1238 			to_free_highmem--;
1239 			alloc_highmem--;
1240 		} else {
1241 			if (!to_free_normal)
1242 				continue;
1243 			to_free_normal--;
1244 			alloc_normal--;
1245 		}
1246 		memory_bm_clear_bit(&copy_bm, pfn);
1247 		swsusp_unset_page_forbidden(page);
1248 		swsusp_unset_page_free(page);
1249 		__free_page(page);
1250 	}
1251 }
1252 
1253 /**
1254  * minimum_image_size - Estimate the minimum acceptable size of an image
1255  * @saveable: Number of saveable pages in the system.
1256  *
1257  * We want to avoid attempting to free too much memory too hard, so estimate the
1258  * minimum acceptable size of a hibernation image to use as the lower limit for
1259  * preallocating memory.
1260  *
1261  * We assume that the minimum image size should be proportional to
1262  *
1263  * [number of saveable pages] - [number of pages that can be freed in theory]
1264  *
1265  * where the second term is the sum of (1) reclaimable slab pages, (2) active
1266  * and (3) inactive anonymouns pages, (4) active and (5) inactive file pages,
1267  * minus mapped file pages.
1268  */
minimum_image_size(unsigned long saveable)1269 static unsigned long minimum_image_size(unsigned long saveable)
1270 {
1271 	unsigned long size;
1272 
1273 	size = global_page_state(NR_SLAB_RECLAIMABLE)
1274 		+ global_page_state(NR_ACTIVE_ANON)
1275 		+ global_page_state(NR_INACTIVE_ANON)
1276 		+ global_page_state(NR_ACTIVE_FILE)
1277 		+ global_page_state(NR_INACTIVE_FILE)
1278 		- global_page_state(NR_FILE_MAPPED);
1279 
1280 	return saveable <= size ? 0 : saveable - size;
1281 }
1282 
1283 /**
1284  * hibernate_preallocate_memory - Preallocate memory for hibernation image
1285  *
1286  * To create a hibernation image it is necessary to make a copy of every page
1287  * frame in use.  We also need a number of page frames to be free during
1288  * hibernation for allocations made while saving the image and for device
1289  * drivers, in case they need to allocate memory from their hibernation
1290  * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1291  * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1292  * /sys/power/reserved_size, respectively).  To make this happen, we compute the
1293  * total number of available page frames and allocate at least
1294  *
1295  * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1296  *  + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1297  *
1298  * of them, which corresponds to the maximum size of a hibernation image.
1299  *
1300  * If image_size is set below the number following from the above formula,
1301  * the preallocation of memory is continued until the total number of saveable
1302  * pages in the system is below the requested image size or the minimum
1303  * acceptable image size returned by minimum_image_size(), whichever is greater.
1304  */
hibernate_preallocate_memory(void)1305 int hibernate_preallocate_memory(void)
1306 {
1307 	struct zone *zone;
1308 	unsigned long saveable, size, max_size, count, highmem, pages = 0;
1309 	unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1310 	struct timeval start, stop;
1311 	int error;
1312 
1313 	printk(KERN_INFO "PM: Preallocating image memory... ");
1314 	do_gettimeofday(&start);
1315 
1316 	error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1317 	if (error)
1318 		goto err_out;
1319 
1320 	error = memory_bm_create(&copy_bm, GFP_IMAGE, PG_ANY);
1321 	if (error)
1322 		goto err_out;
1323 
1324 	alloc_normal = 0;
1325 	alloc_highmem = 0;
1326 
1327 	/* Count the number of saveable data pages. */
1328 	save_highmem = count_highmem_pages();
1329 	saveable = count_data_pages();
1330 
1331 	/*
1332 	 * Compute the total number of page frames we can use (count) and the
1333 	 * number of pages needed for image metadata (size).
1334 	 */
1335 	count = saveable;
1336 	saveable += save_highmem;
1337 	highmem = save_highmem;
1338 	size = 0;
1339 	for_each_populated_zone(zone) {
1340 		size += snapshot_additional_pages(zone);
1341 		if (is_highmem(zone))
1342 			highmem += zone_page_state(zone, NR_FREE_PAGES);
1343 		else
1344 			count += zone_page_state(zone, NR_FREE_PAGES);
1345 	}
1346 	avail_normal = count;
1347 	count += highmem;
1348 	count -= totalreserve_pages;
1349 
1350 	/* Add number of pages required for page keys (s390 only). */
1351 	size += page_key_additional_pages(saveable);
1352 
1353 	/* Compute the maximum number of saveable pages to leave in memory. */
1354 	max_size = (count - (size + PAGES_FOR_IO)) / 2
1355 			- 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1356 	/* Compute the desired number of image pages specified by image_size. */
1357 	size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1358 	if (size > max_size)
1359 		size = max_size;
1360 	/*
1361 	 * If the desired number of image pages is at least as large as the
1362 	 * current number of saveable pages in memory, allocate page frames for
1363 	 * the image and we're done.
1364 	 */
1365 	if (size >= saveable) {
1366 		pages = preallocate_image_highmem(save_highmem);
1367 		pages += preallocate_image_memory(saveable - pages, avail_normal);
1368 		goto out;
1369 	}
1370 
1371 	/* Estimate the minimum size of the image. */
1372 	pages = minimum_image_size(saveable);
1373 	/*
1374 	 * To avoid excessive pressure on the normal zone, leave room in it to
1375 	 * accommodate an image of the minimum size (unless it's already too
1376 	 * small, in which case don't preallocate pages from it at all).
1377 	 */
1378 	if (avail_normal > pages)
1379 		avail_normal -= pages;
1380 	else
1381 		avail_normal = 0;
1382 	if (size < pages)
1383 		size = min_t(unsigned long, pages, max_size);
1384 
1385 	/*
1386 	 * Let the memory management subsystem know that we're going to need a
1387 	 * large number of page frames to allocate and make it free some memory.
1388 	 * NOTE: If this is not done, performance will be hurt badly in some
1389 	 * test cases.
1390 	 */
1391 	shrink_all_memory(saveable - size);
1392 
1393 	/*
1394 	 * The number of saveable pages in memory was too high, so apply some
1395 	 * pressure to decrease it.  First, make room for the largest possible
1396 	 * image and fail if that doesn't work.  Next, try to decrease the size
1397 	 * of the image as much as indicated by 'size' using allocations from
1398 	 * highmem and non-highmem zones separately.
1399 	 */
1400 	pages_highmem = preallocate_image_highmem(highmem / 2);
1401 	alloc = count - max_size;
1402 	if (alloc > pages_highmem)
1403 		alloc -= pages_highmem;
1404 	else
1405 		alloc = 0;
1406 	pages = preallocate_image_memory(alloc, avail_normal);
1407 	if (pages < alloc) {
1408 		/* We have exhausted non-highmem pages, try highmem. */
1409 		alloc -= pages;
1410 		pages += pages_highmem;
1411 		pages_highmem = preallocate_image_highmem(alloc);
1412 		if (pages_highmem < alloc)
1413 			goto err_out;
1414 		pages += pages_highmem;
1415 		/*
1416 		 * size is the desired number of saveable pages to leave in
1417 		 * memory, so try to preallocate (all memory - size) pages.
1418 		 */
1419 		alloc = (count - pages) - size;
1420 		pages += preallocate_image_highmem(alloc);
1421 	} else {
1422 		/*
1423 		 * There are approximately max_size saveable pages at this point
1424 		 * and we want to reduce this number down to size.
1425 		 */
1426 		alloc = max_size - size;
1427 		size = preallocate_highmem_fraction(alloc, highmem, count);
1428 		pages_highmem += size;
1429 		alloc -= size;
1430 		size = preallocate_image_memory(alloc, avail_normal);
1431 		pages_highmem += preallocate_image_highmem(alloc - size);
1432 		pages += pages_highmem + size;
1433 	}
1434 
1435 	/*
1436 	 * We only need as many page frames for the image as there are saveable
1437 	 * pages in memory, but we have allocated more.  Release the excessive
1438 	 * ones now.
1439 	 */
1440 	free_unnecessary_pages();
1441 
1442  out:
1443 	do_gettimeofday(&stop);
1444 	printk(KERN_CONT "done (allocated %lu pages)\n", pages);
1445 	swsusp_show_speed(&start, &stop, pages, "Allocated");
1446 
1447 	return 0;
1448 
1449  err_out:
1450 	printk(KERN_CONT "\n");
1451 	swsusp_free();
1452 	return -ENOMEM;
1453 }
1454 
1455 #ifdef CONFIG_HIGHMEM
1456 /**
1457   *	count_pages_for_highmem - compute the number of non-highmem pages
1458   *	that will be necessary for creating copies of highmem pages.
1459   */
1460 
count_pages_for_highmem(unsigned int nr_highmem)1461 static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1462 {
1463 	unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1464 
1465 	if (free_highmem >= nr_highmem)
1466 		nr_highmem = 0;
1467 	else
1468 		nr_highmem -= free_highmem;
1469 
1470 	return nr_highmem;
1471 }
1472 #else
1473 static unsigned int
count_pages_for_highmem(unsigned int nr_highmem)1474 count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1475 #endif /* CONFIG_HIGHMEM */
1476 
1477 /**
1478  *	enough_free_mem - Make sure we have enough free memory for the
1479  *	snapshot image.
1480  */
1481 
enough_free_mem(unsigned int nr_pages,unsigned int nr_highmem)1482 static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1483 {
1484 	struct zone *zone;
1485 	unsigned int free = alloc_normal;
1486 
1487 	for_each_populated_zone(zone)
1488 		if (!is_highmem(zone))
1489 			free += zone_page_state(zone, NR_FREE_PAGES);
1490 
1491 	nr_pages += count_pages_for_highmem(nr_highmem);
1492 	pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n",
1493 		nr_pages, PAGES_FOR_IO, free);
1494 
1495 	return free > nr_pages + PAGES_FOR_IO;
1496 }
1497 
1498 #ifdef CONFIG_HIGHMEM
1499 /**
1500  *	get_highmem_buffer - if there are some highmem pages in the suspend
1501  *	image, we may need the buffer to copy them and/or load their data.
1502  */
1503 
get_highmem_buffer(int safe_needed)1504 static inline int get_highmem_buffer(int safe_needed)
1505 {
1506 	buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
1507 	return buffer ? 0 : -ENOMEM;
1508 }
1509 
1510 /**
1511  *	alloc_highmem_image_pages - allocate some highmem pages for the image.
1512  *	Try to allocate as many pages as needed, but if the number of free
1513  *	highmem pages is lesser than that, allocate them all.
1514  */
1515 
1516 static inline unsigned int
alloc_highmem_pages(struct memory_bitmap * bm,unsigned int nr_highmem)1517 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int nr_highmem)
1518 {
1519 	unsigned int to_alloc = count_free_highmem_pages();
1520 
1521 	if (to_alloc > nr_highmem)
1522 		to_alloc = nr_highmem;
1523 
1524 	nr_highmem -= to_alloc;
1525 	while (to_alloc-- > 0) {
1526 		struct page *page;
1527 
1528 		page = alloc_image_page(__GFP_HIGHMEM);
1529 		memory_bm_set_bit(bm, page_to_pfn(page));
1530 	}
1531 	return nr_highmem;
1532 }
1533 #else
get_highmem_buffer(int safe_needed)1534 static inline int get_highmem_buffer(int safe_needed) { return 0; }
1535 
1536 static inline unsigned int
alloc_highmem_pages(struct memory_bitmap * bm,unsigned int n)1537 alloc_highmem_pages(struct memory_bitmap *bm, unsigned int n) { return 0; }
1538 #endif /* CONFIG_HIGHMEM */
1539 
1540 /**
1541  *	swsusp_alloc - allocate memory for the suspend image
1542  *
1543  *	We first try to allocate as many highmem pages as there are
1544  *	saveable highmem pages in the system.  If that fails, we allocate
1545  *	non-highmem pages for the copies of the remaining highmem ones.
1546  *
1547  *	In this approach it is likely that the copies of highmem pages will
1548  *	also be located in the high memory, because of the way in which
1549  *	copy_data_pages() works.
1550  */
1551 
1552 static int
swsusp_alloc(struct memory_bitmap * orig_bm,struct memory_bitmap * copy_bm,unsigned int nr_pages,unsigned int nr_highmem)1553 swsusp_alloc(struct memory_bitmap *orig_bm, struct memory_bitmap *copy_bm,
1554 		unsigned int nr_pages, unsigned int nr_highmem)
1555 {
1556 	if (nr_highmem > 0) {
1557 		if (get_highmem_buffer(PG_ANY))
1558 			goto err_out;
1559 		if (nr_highmem > alloc_highmem) {
1560 			nr_highmem -= alloc_highmem;
1561 			nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1562 		}
1563 	}
1564 	if (nr_pages > alloc_normal) {
1565 		nr_pages -= alloc_normal;
1566 		while (nr_pages-- > 0) {
1567 			struct page *page;
1568 
1569 			page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
1570 			if (!page)
1571 				goto err_out;
1572 			memory_bm_set_bit(copy_bm, page_to_pfn(page));
1573 		}
1574 	}
1575 
1576 	return 0;
1577 
1578  err_out:
1579 	swsusp_free();
1580 	return -ENOMEM;
1581 }
1582 
swsusp_save(void)1583 asmlinkage int swsusp_save(void)
1584 {
1585 	unsigned int nr_pages, nr_highmem;
1586 
1587 	printk(KERN_INFO "PM: Creating hibernation image:\n");
1588 
1589 	drain_local_pages(NULL);
1590 	nr_pages = count_data_pages();
1591 	nr_highmem = count_highmem_pages();
1592 	printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem);
1593 
1594 	if (!enough_free_mem(nr_pages, nr_highmem)) {
1595 		printk(KERN_ERR "PM: Not enough free memory\n");
1596 		return -ENOMEM;
1597 	}
1598 
1599 	if (swsusp_alloc(&orig_bm, &copy_bm, nr_pages, nr_highmem)) {
1600 		printk(KERN_ERR "PM: Memory allocation failed\n");
1601 		return -ENOMEM;
1602 	}
1603 
1604 	/* During allocating of suspend pagedir, new cold pages may appear.
1605 	 * Kill them.
1606 	 */
1607 	drain_local_pages(NULL);
1608 	copy_data_pages(&copy_bm, &orig_bm);
1609 
1610 	/*
1611 	 * End of critical section. From now on, we can write to memory,
1612 	 * but we should not touch disk. This specially means we must _not_
1613 	 * touch swap space! Except we must write out our image of course.
1614 	 */
1615 
1616 	nr_pages += nr_highmem;
1617 	nr_copy_pages = nr_pages;
1618 	nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1619 
1620 	printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n",
1621 		nr_pages);
1622 
1623 	return 0;
1624 }
1625 
1626 #ifndef CONFIG_ARCH_HIBERNATION_HEADER
init_header_complete(struct swsusp_info * info)1627 static int init_header_complete(struct swsusp_info *info)
1628 {
1629 	memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
1630 	info->version_code = LINUX_VERSION_CODE;
1631 	return 0;
1632 }
1633 
check_image_kernel(struct swsusp_info * info)1634 static char *check_image_kernel(struct swsusp_info *info)
1635 {
1636 	if (info->version_code != LINUX_VERSION_CODE)
1637 		return "kernel version";
1638 	if (strcmp(info->uts.sysname,init_utsname()->sysname))
1639 		return "system type";
1640 	if (strcmp(info->uts.release,init_utsname()->release))
1641 		return "kernel release";
1642 	if (strcmp(info->uts.version,init_utsname()->version))
1643 		return "version";
1644 	if (strcmp(info->uts.machine,init_utsname()->machine))
1645 		return "machine";
1646 	return NULL;
1647 }
1648 #endif /* CONFIG_ARCH_HIBERNATION_HEADER */
1649 
snapshot_get_image_size(void)1650 unsigned long snapshot_get_image_size(void)
1651 {
1652 	return nr_copy_pages + nr_meta_pages + 1;
1653 }
1654 
init_header(struct swsusp_info * info)1655 static int init_header(struct swsusp_info *info)
1656 {
1657 	memset(info, 0, sizeof(struct swsusp_info));
1658 	info->num_physpages = num_physpages;
1659 	info->image_pages = nr_copy_pages;
1660 	info->pages = snapshot_get_image_size();
1661 	info->size = info->pages;
1662 	info->size <<= PAGE_SHIFT;
1663 	return init_header_complete(info);
1664 }
1665 
1666 /**
1667  *	pack_pfns - pfns corresponding to the set bits found in the bitmap @bm
1668  *	are stored in the array @buf[] (1 page at a time)
1669  */
1670 
1671 static inline void
pack_pfns(unsigned long * buf,struct memory_bitmap * bm)1672 pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
1673 {
1674 	int j;
1675 
1676 	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1677 		buf[j] = memory_bm_next_pfn(bm);
1678 		if (unlikely(buf[j] == BM_END_OF_MAP))
1679 			break;
1680 		/* Save page key for data page (s390 only). */
1681 		page_key_read(buf + j);
1682 	}
1683 }
1684 
1685 /**
1686  *	snapshot_read_next - used for reading the system memory snapshot.
1687  *
1688  *	On the first call to it @handle should point to a zeroed
1689  *	snapshot_handle structure.  The structure gets updated and a pointer
1690  *	to it should be passed to this function every next time.
1691  *
1692  *	On success the function returns a positive number.  Then, the caller
1693  *	is allowed to read up to the returned number of bytes from the memory
1694  *	location computed by the data_of() macro.
1695  *
1696  *	The function returns 0 to indicate the end of data stream condition,
1697  *	and a negative number is returned on error.  In such cases the
1698  *	structure pointed to by @handle is not updated and should not be used
1699  *	any more.
1700  */
1701 
snapshot_read_next(struct snapshot_handle * handle)1702 int snapshot_read_next(struct snapshot_handle *handle)
1703 {
1704 	if (handle->cur > nr_meta_pages + nr_copy_pages)
1705 		return 0;
1706 
1707 	if (!buffer) {
1708 		/* This makes the buffer be freed by swsusp_free() */
1709 		buffer = get_image_page(GFP_ATOMIC, PG_ANY);
1710 		if (!buffer)
1711 			return -ENOMEM;
1712 	}
1713 	if (!handle->cur) {
1714 		int error;
1715 
1716 		error = init_header((struct swsusp_info *)buffer);
1717 		if (error)
1718 			return error;
1719 		handle->buffer = buffer;
1720 		memory_bm_position_reset(&orig_bm);
1721 		memory_bm_position_reset(&copy_bm);
1722 	} else if (handle->cur <= nr_meta_pages) {
1723 		clear_page(buffer);
1724 		pack_pfns(buffer, &orig_bm);
1725 	} else {
1726 		struct page *page;
1727 
1728 		page = pfn_to_page(memory_bm_next_pfn(&copy_bm));
1729 		if (PageHighMem(page)) {
1730 			/* Highmem pages are copied to the buffer,
1731 			 * because we can't return with a kmapped
1732 			 * highmem page (we may not be called again).
1733 			 */
1734 			void *kaddr;
1735 
1736 			kaddr = kmap_atomic(page);
1737 			copy_page(buffer, kaddr);
1738 			kunmap_atomic(kaddr);
1739 			handle->buffer = buffer;
1740 		} else {
1741 			handle->buffer = page_address(page);
1742 		}
1743 	}
1744 	handle->cur++;
1745 	return PAGE_SIZE;
1746 }
1747 
1748 /**
1749  *	mark_unsafe_pages - mark the pages that cannot be used for storing
1750  *	the image during resume, because they conflict with the pages that
1751  *	had been used before suspend
1752  */
1753 
mark_unsafe_pages(struct memory_bitmap * bm)1754 static int mark_unsafe_pages(struct memory_bitmap *bm)
1755 {
1756 	struct zone *zone;
1757 	unsigned long pfn, max_zone_pfn;
1758 
1759 	/* Clear page flags */
1760 	for_each_populated_zone(zone) {
1761 		max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
1762 		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1763 			if (pfn_valid(pfn))
1764 				swsusp_unset_page_free(pfn_to_page(pfn));
1765 	}
1766 
1767 	/* Mark pages that correspond to the "original" pfns as "unsafe" */
1768 	memory_bm_position_reset(bm);
1769 	do {
1770 		pfn = memory_bm_next_pfn(bm);
1771 		if (likely(pfn != BM_END_OF_MAP)) {
1772 			if (likely(pfn_valid(pfn)))
1773 				swsusp_set_page_free(pfn_to_page(pfn));
1774 			else
1775 				return -EFAULT;
1776 		}
1777 	} while (pfn != BM_END_OF_MAP);
1778 
1779 	allocated_unsafe_pages = 0;
1780 
1781 	return 0;
1782 }
1783 
1784 static void
duplicate_memory_bitmap(struct memory_bitmap * dst,struct memory_bitmap * src)1785 duplicate_memory_bitmap(struct memory_bitmap *dst, struct memory_bitmap *src)
1786 {
1787 	unsigned long pfn;
1788 
1789 	memory_bm_position_reset(src);
1790 	pfn = memory_bm_next_pfn(src);
1791 	while (pfn != BM_END_OF_MAP) {
1792 		memory_bm_set_bit(dst, pfn);
1793 		pfn = memory_bm_next_pfn(src);
1794 	}
1795 }
1796 
check_header(struct swsusp_info * info)1797 static int check_header(struct swsusp_info *info)
1798 {
1799 	char *reason;
1800 
1801 	reason = check_image_kernel(info);
1802 	if (!reason && info->num_physpages != num_physpages)
1803 		reason = "memory size";
1804 	if (reason) {
1805 		printk(KERN_ERR "PM: Image mismatch: %s\n", reason);
1806 		return -EPERM;
1807 	}
1808 	return 0;
1809 }
1810 
1811 /**
1812  *	load header - check the image header and copy data from it
1813  */
1814 
1815 static int
load_header(struct swsusp_info * info)1816 load_header(struct swsusp_info *info)
1817 {
1818 	int error;
1819 
1820 	restore_pblist = NULL;
1821 	error = check_header(info);
1822 	if (!error) {
1823 		nr_copy_pages = info->image_pages;
1824 		nr_meta_pages = info->pages - info->image_pages - 1;
1825 	}
1826 	return error;
1827 }
1828 
1829 /**
1830  *	unpack_orig_pfns - for each element of @buf[] (1 page at a time) set
1831  *	the corresponding bit in the memory bitmap @bm
1832  */
unpack_orig_pfns(unsigned long * buf,struct memory_bitmap * bm)1833 static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
1834 {
1835 	int j;
1836 
1837 	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
1838 		if (unlikely(buf[j] == BM_END_OF_MAP))
1839 			break;
1840 
1841 		/* Extract and buffer page key for data page (s390 only). */
1842 		page_key_memorize(buf + j);
1843 
1844 		if (memory_bm_pfn_present(bm, buf[j]))
1845 			memory_bm_set_bit(bm, buf[j]);
1846 		else
1847 			return -EFAULT;
1848 	}
1849 
1850 	return 0;
1851 }
1852 
1853 /* List of "safe" pages that may be used to store data loaded from the suspend
1854  * image
1855  */
1856 static struct linked_page *safe_pages_list;
1857 
1858 #ifdef CONFIG_HIGHMEM
1859 /* struct highmem_pbe is used for creating the list of highmem pages that
1860  * should be restored atomically during the resume from disk, because the page
1861  * frames they have occupied before the suspend are in use.
1862  */
1863 struct highmem_pbe {
1864 	struct page *copy_page;	/* data is here now */
1865 	struct page *orig_page;	/* data was here before the suspend */
1866 	struct highmem_pbe *next;
1867 };
1868 
1869 /* List of highmem PBEs needed for restoring the highmem pages that were
1870  * allocated before the suspend and included in the suspend image, but have
1871  * also been allocated by the "resume" kernel, so their contents cannot be
1872  * written directly to their "original" page frames.
1873  */
1874 static struct highmem_pbe *highmem_pblist;
1875 
1876 /**
1877  *	count_highmem_image_pages - compute the number of highmem pages in the
1878  *	suspend image.  The bits in the memory bitmap @bm that correspond to the
1879  *	image pages are assumed to be set.
1880  */
1881 
count_highmem_image_pages(struct memory_bitmap * bm)1882 static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
1883 {
1884 	unsigned long pfn;
1885 	unsigned int cnt = 0;
1886 
1887 	memory_bm_position_reset(bm);
1888 	pfn = memory_bm_next_pfn(bm);
1889 	while (pfn != BM_END_OF_MAP) {
1890 		if (PageHighMem(pfn_to_page(pfn)))
1891 			cnt++;
1892 
1893 		pfn = memory_bm_next_pfn(bm);
1894 	}
1895 	return cnt;
1896 }
1897 
1898 /**
1899  *	prepare_highmem_image - try to allocate as many highmem pages as
1900  *	there are highmem image pages (@nr_highmem_p points to the variable
1901  *	containing the number of highmem image pages).  The pages that are
1902  *	"safe" (ie. will not be overwritten when the suspend image is
1903  *	restored) have the corresponding bits set in @bm (it must be
1904  *	unitialized).
1905  *
1906  *	NOTE: This function should not be called if there are no highmem
1907  *	image pages.
1908  */
1909 
1910 static unsigned int safe_highmem_pages;
1911 
1912 static struct memory_bitmap *safe_highmem_bm;
1913 
1914 static int
prepare_highmem_image(struct memory_bitmap * bm,unsigned int * nr_highmem_p)1915 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
1916 {
1917 	unsigned int to_alloc;
1918 
1919 	if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
1920 		return -ENOMEM;
1921 
1922 	if (get_highmem_buffer(PG_SAFE))
1923 		return -ENOMEM;
1924 
1925 	to_alloc = count_free_highmem_pages();
1926 	if (to_alloc > *nr_highmem_p)
1927 		to_alloc = *nr_highmem_p;
1928 	else
1929 		*nr_highmem_p = to_alloc;
1930 
1931 	safe_highmem_pages = 0;
1932 	while (to_alloc-- > 0) {
1933 		struct page *page;
1934 
1935 		page = alloc_page(__GFP_HIGHMEM);
1936 		if (!swsusp_page_is_free(page)) {
1937 			/* The page is "safe", set its bit the bitmap */
1938 			memory_bm_set_bit(bm, page_to_pfn(page));
1939 			safe_highmem_pages++;
1940 		}
1941 		/* Mark the page as allocated */
1942 		swsusp_set_page_forbidden(page);
1943 		swsusp_set_page_free(page);
1944 	}
1945 	memory_bm_position_reset(bm);
1946 	safe_highmem_bm = bm;
1947 	return 0;
1948 }
1949 
1950 /**
1951  *	get_highmem_page_buffer - for given highmem image page find the buffer
1952  *	that suspend_write_next() should set for its caller to write to.
1953  *
1954  *	If the page is to be saved to its "original" page frame or a copy of
1955  *	the page is to be made in the highmem, @buffer is returned.  Otherwise,
1956  *	the copy of the page is to be made in normal memory, so the address of
1957  *	the copy is returned.
1958  *
1959  *	If @buffer is returned, the caller of suspend_write_next() will write
1960  *	the page's contents to @buffer, so they will have to be copied to the
1961  *	right location on the next call to suspend_write_next() and it is done
1962  *	with the help of copy_last_highmem_page().  For this purpose, if
1963  *	@buffer is returned, @last_highmem page is set to the page to which
1964  *	the data will have to be copied from @buffer.
1965  */
1966 
1967 static struct page *last_highmem_page;
1968 
1969 static void *
get_highmem_page_buffer(struct page * page,struct chain_allocator * ca)1970 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
1971 {
1972 	struct highmem_pbe *pbe;
1973 	void *kaddr;
1974 
1975 	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
1976 		/* We have allocated the "original" page frame and we can
1977 		 * use it directly to store the loaded page.
1978 		 */
1979 		last_highmem_page = page;
1980 		return buffer;
1981 	}
1982 	/* The "original" page frame has not been allocated and we have to
1983 	 * use a "safe" page frame to store the loaded page.
1984 	 */
1985 	pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
1986 	if (!pbe) {
1987 		swsusp_free();
1988 		return ERR_PTR(-ENOMEM);
1989 	}
1990 	pbe->orig_page = page;
1991 	if (safe_highmem_pages > 0) {
1992 		struct page *tmp;
1993 
1994 		/* Copy of the page will be stored in high memory */
1995 		kaddr = buffer;
1996 		tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
1997 		safe_highmem_pages--;
1998 		last_highmem_page = tmp;
1999 		pbe->copy_page = tmp;
2000 	} else {
2001 		/* Copy of the page will be stored in normal memory */
2002 		kaddr = safe_pages_list;
2003 		safe_pages_list = safe_pages_list->next;
2004 		pbe->copy_page = virt_to_page(kaddr);
2005 	}
2006 	pbe->next = highmem_pblist;
2007 	highmem_pblist = pbe;
2008 	return kaddr;
2009 }
2010 
2011 /**
2012  *	copy_last_highmem_page - copy the contents of a highmem image from
2013  *	@buffer, where the caller of snapshot_write_next() has place them,
2014  *	to the right location represented by @last_highmem_page .
2015  */
2016 
copy_last_highmem_page(void)2017 static void copy_last_highmem_page(void)
2018 {
2019 	if (last_highmem_page) {
2020 		void *dst;
2021 
2022 		dst = kmap_atomic(last_highmem_page);
2023 		copy_page(dst, buffer);
2024 		kunmap_atomic(dst);
2025 		last_highmem_page = NULL;
2026 	}
2027 }
2028 
last_highmem_page_copied(void)2029 static inline int last_highmem_page_copied(void)
2030 {
2031 	return !last_highmem_page;
2032 }
2033 
free_highmem_data(void)2034 static inline void free_highmem_data(void)
2035 {
2036 	if (safe_highmem_bm)
2037 		memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2038 
2039 	if (buffer)
2040 		free_image_page(buffer, PG_UNSAFE_CLEAR);
2041 }
2042 #else
get_safe_write_buffer(void)2043 static inline int get_safe_write_buffer(void) { return 0; }
2044 
2045 static unsigned int
count_highmem_image_pages(struct memory_bitmap * bm)2046 count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2047 
2048 static inline int
prepare_highmem_image(struct memory_bitmap * bm,unsigned int * nr_highmem_p)2049 prepare_highmem_image(struct memory_bitmap *bm, unsigned int *nr_highmem_p)
2050 {
2051 	return 0;
2052 }
2053 
2054 static inline void *
get_highmem_page_buffer(struct page * page,struct chain_allocator * ca)2055 get_highmem_page_buffer(struct page *page, struct chain_allocator *ca)
2056 {
2057 	return ERR_PTR(-EINVAL);
2058 }
2059 
copy_last_highmem_page(void)2060 static inline void copy_last_highmem_page(void) {}
last_highmem_page_copied(void)2061 static inline int last_highmem_page_copied(void) { return 1; }
free_highmem_data(void)2062 static inline void free_highmem_data(void) {}
2063 #endif /* CONFIG_HIGHMEM */
2064 
2065 /**
2066  *	prepare_image - use the memory bitmap @bm to mark the pages that will
2067  *	be overwritten in the process of restoring the system memory state
2068  *	from the suspend image ("unsafe" pages) and allocate memory for the
2069  *	image.
2070  *
2071  *	The idea is to allocate a new memory bitmap first and then allocate
2072  *	as many pages as needed for the image data, but not to assign these
2073  *	pages to specific tasks initially.  Instead, we just mark them as
2074  *	allocated and create a lists of "safe" pages that will be used
2075  *	later.  On systems with high memory a list of "safe" highmem pages is
2076  *	also created.
2077  */
2078 
2079 #define PBES_PER_LINKED_PAGE	(LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2080 
2081 static int
prepare_image(struct memory_bitmap * new_bm,struct memory_bitmap * bm)2082 prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2083 {
2084 	unsigned int nr_pages, nr_highmem;
2085 	struct linked_page *sp_list, *lp;
2086 	int error;
2087 
2088 	/* If there is no highmem, the buffer will not be necessary */
2089 	free_image_page(buffer, PG_UNSAFE_CLEAR);
2090 	buffer = NULL;
2091 
2092 	nr_highmem = count_highmem_image_pages(bm);
2093 	error = mark_unsafe_pages(bm);
2094 	if (error)
2095 		goto Free;
2096 
2097 	error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2098 	if (error)
2099 		goto Free;
2100 
2101 	duplicate_memory_bitmap(new_bm, bm);
2102 	memory_bm_free(bm, PG_UNSAFE_KEEP);
2103 	if (nr_highmem > 0) {
2104 		error = prepare_highmem_image(bm, &nr_highmem);
2105 		if (error)
2106 			goto Free;
2107 	}
2108 	/* Reserve some safe pages for potential later use.
2109 	 *
2110 	 * NOTE: This way we make sure there will be enough safe pages for the
2111 	 * chain_alloc() in get_buffer().  It is a bit wasteful, but
2112 	 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2113 	 */
2114 	sp_list = NULL;
2115 	/* nr_copy_pages cannot be lesser than allocated_unsafe_pages */
2116 	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2117 	nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2118 	while (nr_pages > 0) {
2119 		lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2120 		if (!lp) {
2121 			error = -ENOMEM;
2122 			goto Free;
2123 		}
2124 		lp->next = sp_list;
2125 		sp_list = lp;
2126 		nr_pages--;
2127 	}
2128 	/* Preallocate memory for the image */
2129 	safe_pages_list = NULL;
2130 	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2131 	while (nr_pages > 0) {
2132 		lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2133 		if (!lp) {
2134 			error = -ENOMEM;
2135 			goto Free;
2136 		}
2137 		if (!swsusp_page_is_free(virt_to_page(lp))) {
2138 			/* The page is "safe", add it to the list */
2139 			lp->next = safe_pages_list;
2140 			safe_pages_list = lp;
2141 		}
2142 		/* Mark the page as allocated */
2143 		swsusp_set_page_forbidden(virt_to_page(lp));
2144 		swsusp_set_page_free(virt_to_page(lp));
2145 		nr_pages--;
2146 	}
2147 	/* Free the reserved safe pages so that chain_alloc() can use them */
2148 	while (sp_list) {
2149 		lp = sp_list->next;
2150 		free_image_page(sp_list, PG_UNSAFE_CLEAR);
2151 		sp_list = lp;
2152 	}
2153 	return 0;
2154 
2155  Free:
2156 	swsusp_free();
2157 	return error;
2158 }
2159 
2160 /**
2161  *	get_buffer - compute the address that snapshot_write_next() should
2162  *	set for its caller to write to.
2163  */
2164 
get_buffer(struct memory_bitmap * bm,struct chain_allocator * ca)2165 static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2166 {
2167 	struct pbe *pbe;
2168 	struct page *page;
2169 	unsigned long pfn = memory_bm_next_pfn(bm);
2170 
2171 	if (pfn == BM_END_OF_MAP)
2172 		return ERR_PTR(-EFAULT);
2173 
2174 	page = pfn_to_page(pfn);
2175 	if (PageHighMem(page))
2176 		return get_highmem_page_buffer(page, ca);
2177 
2178 	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2179 		/* We have allocated the "original" page frame and we can
2180 		 * use it directly to store the loaded page.
2181 		 */
2182 		return page_address(page);
2183 
2184 	/* The "original" page frame has not been allocated and we have to
2185 	 * use a "safe" page frame to store the loaded page.
2186 	 */
2187 	pbe = chain_alloc(ca, sizeof(struct pbe));
2188 	if (!pbe) {
2189 		swsusp_free();
2190 		return ERR_PTR(-ENOMEM);
2191 	}
2192 	pbe->orig_address = page_address(page);
2193 	pbe->address = safe_pages_list;
2194 	safe_pages_list = safe_pages_list->next;
2195 	pbe->next = restore_pblist;
2196 	restore_pblist = pbe;
2197 	return pbe->address;
2198 }
2199 
2200 /**
2201  *	snapshot_write_next - used for writing the system memory snapshot.
2202  *
2203  *	On the first call to it @handle should point to a zeroed
2204  *	snapshot_handle structure.  The structure gets updated and a pointer
2205  *	to it should be passed to this function every next time.
2206  *
2207  *	On success the function returns a positive number.  Then, the caller
2208  *	is allowed to write up to the returned number of bytes to the memory
2209  *	location computed by the data_of() macro.
2210  *
2211  *	The function returns 0 to indicate the "end of file" condition,
2212  *	and a negative number is returned on error.  In such cases the
2213  *	structure pointed to by @handle is not updated and should not be used
2214  *	any more.
2215  */
2216 
snapshot_write_next(struct snapshot_handle * handle)2217 int snapshot_write_next(struct snapshot_handle *handle)
2218 {
2219 	static struct chain_allocator ca;
2220 	int error = 0;
2221 
2222 	/* Check if we have already loaded the entire image */
2223 	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2224 		return 0;
2225 
2226 	handle->sync_read = 1;
2227 
2228 	if (!handle->cur) {
2229 		if (!buffer)
2230 			/* This makes the buffer be freed by swsusp_free() */
2231 			buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2232 
2233 		if (!buffer)
2234 			return -ENOMEM;
2235 
2236 		handle->buffer = buffer;
2237 	} else if (handle->cur == 1) {
2238 		error = load_header(buffer);
2239 		if (error)
2240 			return error;
2241 
2242 		error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
2243 		if (error)
2244 			return error;
2245 
2246 		/* Allocate buffer for page keys. */
2247 		error = page_key_alloc(nr_copy_pages);
2248 		if (error)
2249 			return error;
2250 
2251 	} else if (handle->cur <= nr_meta_pages + 1) {
2252 		error = unpack_orig_pfns(buffer, &copy_bm);
2253 		if (error)
2254 			return error;
2255 
2256 		if (handle->cur == nr_meta_pages + 1) {
2257 			error = prepare_image(&orig_bm, &copy_bm);
2258 			if (error)
2259 				return error;
2260 
2261 			chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2262 			memory_bm_position_reset(&orig_bm);
2263 			restore_pblist = NULL;
2264 			handle->buffer = get_buffer(&orig_bm, &ca);
2265 			handle->sync_read = 0;
2266 			if (IS_ERR(handle->buffer))
2267 				return PTR_ERR(handle->buffer);
2268 		}
2269 	} else {
2270 		copy_last_highmem_page();
2271 		/* Restore page key for data page (s390 only). */
2272 		page_key_write(handle->buffer);
2273 		handle->buffer = get_buffer(&orig_bm, &ca);
2274 		if (IS_ERR(handle->buffer))
2275 			return PTR_ERR(handle->buffer);
2276 		if (handle->buffer != buffer)
2277 			handle->sync_read = 0;
2278 	}
2279 	handle->cur++;
2280 	return PAGE_SIZE;
2281 }
2282 
2283 /**
2284  *	snapshot_write_finalize - must be called after the last call to
2285  *	snapshot_write_next() in case the last page in the image happens
2286  *	to be a highmem page and its contents should be stored in the
2287  *	highmem.  Additionally, it releases the memory that will not be
2288  *	used any more.
2289  */
2290 
snapshot_write_finalize(struct snapshot_handle * handle)2291 void snapshot_write_finalize(struct snapshot_handle *handle)
2292 {
2293 	copy_last_highmem_page();
2294 	/* Restore page key for data page (s390 only). */
2295 	page_key_write(handle->buffer);
2296 	page_key_free();
2297 	/* Free only if we have loaded the image entirely */
2298 	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2299 		memory_bm_free(&orig_bm, PG_UNSAFE_CLEAR);
2300 		free_highmem_data();
2301 	}
2302 }
2303 
snapshot_image_loaded(struct snapshot_handle * handle)2304 int snapshot_image_loaded(struct snapshot_handle *handle)
2305 {
2306 	return !(!nr_copy_pages || !last_highmem_page_copied() ||
2307 			handle->cur <= nr_meta_pages + nr_copy_pages);
2308 }
2309 
2310 #ifdef CONFIG_HIGHMEM
2311 /* Assumes that @buf is ready and points to a "safe" page */
2312 static inline void
swap_two_pages_data(struct page * p1,struct page * p2,void * buf)2313 swap_two_pages_data(struct page *p1, struct page *p2, void *buf)
2314 {
2315 	void *kaddr1, *kaddr2;
2316 
2317 	kaddr1 = kmap_atomic(p1);
2318 	kaddr2 = kmap_atomic(p2);
2319 	copy_page(buf, kaddr1);
2320 	copy_page(kaddr1, kaddr2);
2321 	copy_page(kaddr2, buf);
2322 	kunmap_atomic(kaddr2);
2323 	kunmap_atomic(kaddr1);
2324 }
2325 
2326 /**
2327  *	restore_highmem - for each highmem page that was allocated before
2328  *	the suspend and included in the suspend image, and also has been
2329  *	allocated by the "resume" kernel swap its current (ie. "before
2330  *	resume") contents with the previous (ie. "before suspend") one.
2331  *
2332  *	If the resume eventually fails, we can call this function once
2333  *	again and restore the "before resume" highmem state.
2334  */
2335 
restore_highmem(void)2336 int restore_highmem(void)
2337 {
2338 	struct highmem_pbe *pbe = highmem_pblist;
2339 	void *buf;
2340 
2341 	if (!pbe)
2342 		return 0;
2343 
2344 	buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2345 	if (!buf)
2346 		return -ENOMEM;
2347 
2348 	while (pbe) {
2349 		swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2350 		pbe = pbe->next;
2351 	}
2352 	free_image_page(buf, PG_UNSAFE_CLEAR);
2353 	return 0;
2354 }
2355 #endif /* CONFIG_HIGHMEM */
2356