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