Linux Audio

Check our new training course

Loading...
v5.9
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * linux/kernel/power/snapshot.c
   4 *
   5 * This file provides system snapshot/restore functionality for swsusp.
   6 *
   7 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
   8 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
 
 
 
   9 */
  10
  11#define pr_fmt(fmt) "PM: hibernation: " fmt
  12
  13#include <linux/version.h>
  14#include <linux/module.h>
  15#include <linux/mm.h>
  16#include <linux/suspend.h>
  17#include <linux/delay.h>
  18#include <linux/bitops.h>
  19#include <linux/spinlock.h>
  20#include <linux/kernel.h>
  21#include <linux/pm.h>
  22#include <linux/device.h>
  23#include <linux/init.h>
  24#include <linux/memblock.h>
  25#include <linux/nmi.h>
  26#include <linux/syscalls.h>
  27#include <linux/console.h>
  28#include <linux/highmem.h>
  29#include <linux/list.h>
  30#include <linux/slab.h>
  31#include <linux/compiler.h>
  32#include <linux/ktime.h>
  33#include <linux/set_memory.h>
  34
  35#include <linux/uaccess.h>
  36#include <asm/mmu_context.h>
 
  37#include <asm/tlbflush.h>
  38#include <asm/io.h>
  39
  40#include "power.h"
  41
  42#if defined(CONFIG_STRICT_KERNEL_RWX) && defined(CONFIG_ARCH_HAS_SET_MEMORY)
  43static bool hibernate_restore_protection;
  44static bool hibernate_restore_protection_active;
  45
  46void enable_restore_image_protection(void)
  47{
  48	hibernate_restore_protection = true;
  49}
  50
  51static inline void hibernate_restore_protection_begin(void)
  52{
  53	hibernate_restore_protection_active = hibernate_restore_protection;
  54}
  55
  56static inline void hibernate_restore_protection_end(void)
  57{
  58	hibernate_restore_protection_active = false;
  59}
  60
  61static inline void hibernate_restore_protect_page(void *page_address)
  62{
  63	if (hibernate_restore_protection_active)
  64		set_memory_ro((unsigned long)page_address, 1);
  65}
  66
  67static inline void hibernate_restore_unprotect_page(void *page_address)
  68{
  69	if (hibernate_restore_protection_active)
  70		set_memory_rw((unsigned long)page_address, 1);
  71}
  72#else
  73static inline void hibernate_restore_protection_begin(void) {}
  74static inline void hibernate_restore_protection_end(void) {}
  75static inline void hibernate_restore_protect_page(void *page_address) {}
  76static inline void hibernate_restore_unprotect_page(void *page_address) {}
  77#endif /* CONFIG_STRICT_KERNEL_RWX  && CONFIG_ARCH_HAS_SET_MEMORY */
  78
  79static int swsusp_page_is_free(struct page *);
  80static void swsusp_set_page_forbidden(struct page *);
  81static void swsusp_unset_page_forbidden(struct page *);
  82
  83/*
  84 * Number of bytes to reserve for memory allocations made by device drivers
  85 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
  86 * cause image creation to fail (tunable via /sys/power/reserved_size).
  87 */
  88unsigned long reserved_size;
  89
  90void __init hibernate_reserved_size_init(void)
  91{
  92	reserved_size = SPARE_PAGES * PAGE_SIZE;
  93}
  94
  95/*
  96 * Preferred image size in bytes (tunable via /sys/power/image_size).
  97 * When it is set to N, swsusp will do its best to ensure the image
  98 * size will not exceed N bytes, but if that is impossible, it will
  99 * try to create the smallest image possible.
 100 */
 101unsigned long image_size;
 102
 103void __init hibernate_image_size_init(void)
 104{
 105	image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
 106}
 107
 108/*
 109 * List of PBEs needed for restoring the pages that were allocated before
 110 * the suspend and included in the suspend image, but have also been
 111 * allocated by the "resume" kernel, so their contents cannot be written
 112 * directly to their "original" page frames.
 113 */
 114struct pbe *restore_pblist;
 115
 116/* struct linked_page is used to build chains of pages */
 117
 118#define LINKED_PAGE_DATA_SIZE	(PAGE_SIZE - sizeof(void *))
 119
 120struct linked_page {
 121	struct linked_page *next;
 122	char data[LINKED_PAGE_DATA_SIZE];
 123} __packed;
 124
 125/*
 126 * List of "safe" pages (ie. pages that were not used by the image kernel
 127 * before hibernation) that may be used as temporary storage for image kernel
 128 * memory contents.
 129 */
 130static struct linked_page *safe_pages_list;
 131
 132/* Pointer to an auxiliary buffer (1 page) */
 133static void *buffer;
 134
 135#define PG_ANY		0
 136#define PG_SAFE		1
 137#define PG_UNSAFE_CLEAR	1
 138#define PG_UNSAFE_KEEP	0
 139
 140static unsigned int allocated_unsafe_pages;
 141
 142/**
 143 * get_image_page - Allocate a page for a hibernation image.
 144 * @gfp_mask: GFP mask for the allocation.
 145 * @safe_needed: Get pages that were not used before hibernation (restore only)
 146 *
 147 * During image restoration, for storing the PBE list and the image data, we can
 148 * only use memory pages that do not conflict with the pages used before
 149 * hibernation.  The "unsafe" pages have PageNosaveFree set and we count them
 150 * using allocated_unsafe_pages.
 151 *
 152 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
 153 * swsusp_free() can release it.
 154 */
 155static void *get_image_page(gfp_t gfp_mask, int safe_needed)
 156{
 157	void *res;
 158
 159	res = (void *)get_zeroed_page(gfp_mask);
 160	if (safe_needed)
 161		while (res && swsusp_page_is_free(virt_to_page(res))) {
 162			/* The page is unsafe, mark it for swsusp_free() */
 163			swsusp_set_page_forbidden(virt_to_page(res));
 164			allocated_unsafe_pages++;
 165			res = (void *)get_zeroed_page(gfp_mask);
 166		}
 167	if (res) {
 168		swsusp_set_page_forbidden(virt_to_page(res));
 169		swsusp_set_page_free(virt_to_page(res));
 170	}
 171	return res;
 172}
 173
 174static void *__get_safe_page(gfp_t gfp_mask)
 175{
 176	if (safe_pages_list) {
 177		void *ret = safe_pages_list;
 178
 179		safe_pages_list = safe_pages_list->next;
 180		memset(ret, 0, PAGE_SIZE);
 181		return ret;
 182	}
 183	return get_image_page(gfp_mask, PG_SAFE);
 184}
 185
 186unsigned long get_safe_page(gfp_t gfp_mask)
 187{
 188	return (unsigned long)__get_safe_page(gfp_mask);
 189}
 190
 191static struct page *alloc_image_page(gfp_t gfp_mask)
 192{
 193	struct page *page;
 194
 195	page = alloc_page(gfp_mask);
 196	if (page) {
 197		swsusp_set_page_forbidden(page);
 198		swsusp_set_page_free(page);
 199	}
 200	return page;
 201}
 202
 203static void recycle_safe_page(void *page_address)
 204{
 205	struct linked_page *lp = page_address;
 206
 207	lp->next = safe_pages_list;
 208	safe_pages_list = lp;
 209}
 210
 211/**
 212 * free_image_page - Free a page allocated for hibernation image.
 213 * @addr: Address of the page to free.
 214 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
 215 *
 216 * The page to free should have been allocated by get_image_page() (page flags
 217 * set by it are affected).
 218 */
 219static inline void free_image_page(void *addr, int clear_nosave_free)
 220{
 221	struct page *page;
 222
 223	BUG_ON(!virt_addr_valid(addr));
 224
 225	page = virt_to_page(addr);
 226
 227	swsusp_unset_page_forbidden(page);
 228	if (clear_nosave_free)
 229		swsusp_unset_page_free(page);
 230
 231	__free_page(page);
 232}
 233
 234static inline void free_list_of_pages(struct linked_page *list,
 235				      int clear_page_nosave)
 236{
 237	while (list) {
 238		struct linked_page *lp = list->next;
 239
 240		free_image_page(list, clear_page_nosave);
 241		list = lp;
 242	}
 243}
 244
 245/*
 246 * struct chain_allocator is used for allocating small objects out of
 247 * a linked list of pages called 'the chain'.
 248 *
 249 * The chain grows each time when there is no room for a new object in
 250 * the current page.  The allocated objects cannot be freed individually.
 251 * It is only possible to free them all at once, by freeing the entire
 252 * chain.
 253 *
 254 * NOTE: The chain allocator may be inefficient if the allocated objects
 255 * are not much smaller than PAGE_SIZE.
 256 */
 257struct chain_allocator {
 258	struct linked_page *chain;	/* the chain */
 259	unsigned int used_space;	/* total size of objects allocated out
 260					   of the current page */
 261	gfp_t gfp_mask;		/* mask for allocating pages */
 262	int safe_needed;	/* if set, only "safe" pages are allocated */
 263};
 264
 265static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
 266		       int safe_needed)
 267{
 268	ca->chain = NULL;
 269	ca->used_space = LINKED_PAGE_DATA_SIZE;
 270	ca->gfp_mask = gfp_mask;
 271	ca->safe_needed = safe_needed;
 272}
 273
 274static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
 275{
 276	void *ret;
 277
 278	if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
 279		struct linked_page *lp;
 280
 281		lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
 282					get_image_page(ca->gfp_mask, PG_ANY);
 283		if (!lp)
 284			return NULL;
 285
 286		lp->next = ca->chain;
 287		ca->chain = lp;
 288		ca->used_space = 0;
 289	}
 290	ret = ca->chain->data + ca->used_space;
 291	ca->used_space += size;
 292	return ret;
 293}
 294
 295/**
 296 * Data types related to memory bitmaps.
 297 *
 298 * Memory bitmap is a structure consiting of many linked lists of
 299 * objects.  The main list's elements are of type struct zone_bitmap
 300 * and each of them corresonds to one zone.  For each zone bitmap
 301 * object there is a list of objects of type struct bm_block that
 302 * represent each blocks of bitmap in which information is stored.
 303 *
 304 * struct memory_bitmap contains a pointer to the main list of zone
 305 * bitmap objects, a struct bm_position used for browsing the bitmap,
 306 * and a pointer to the list of pages used for allocating all of the
 307 * zone bitmap objects and bitmap block objects.
 308 *
 309 * NOTE: It has to be possible to lay out the bitmap in memory
 310 * using only allocations of order 0.  Additionally, the bitmap is
 311 * designed to work with arbitrary number of zones (this is over the
 312 * top for now, but let's avoid making unnecessary assumptions ;-).
 313 *
 314 * struct zone_bitmap contains a pointer to a list of bitmap block
 315 * objects and a pointer to the bitmap block object that has been
 316 * most recently used for setting bits.  Additionally, it contains the
 317 * PFNs that correspond to the start and end of the represented zone.
 318 *
 319 * struct bm_block contains a pointer to the memory page in which
 320 * information is stored (in the form of a block of bitmap)
 321 * It also contains the pfns that correspond to the start and end of
 322 * the represented memory area.
 323 *
 324 * The memory bitmap is organized as a radix tree to guarantee fast random
 325 * access to the bits. There is one radix tree for each zone (as returned
 326 * from create_mem_extents).
 327 *
 328 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
 329 * two linked lists for the nodes of the tree, one for the inner nodes and
 330 * one for the leave nodes. The linked leave nodes are used for fast linear
 331 * access of the memory bitmap.
 332 *
 333 * The struct rtree_node represents one node of the radix tree.
 334 */
 335
 336#define BM_END_OF_MAP	(~0UL)
 337
 338#define BM_BITS_PER_BLOCK	(PAGE_SIZE * BITS_PER_BYTE)
 339#define BM_BLOCK_SHIFT		(PAGE_SHIFT + 3)
 340#define BM_BLOCK_MASK		((1UL << BM_BLOCK_SHIFT) - 1)
 341
 342/*
 343 * struct rtree_node is a wrapper struct to link the nodes
 344 * of the rtree together for easy linear iteration over
 345 * bits and easy freeing
 346 */
 347struct rtree_node {
 348	struct list_head list;
 349	unsigned long *data;
 350};
 351
 352/*
 353 * struct mem_zone_bm_rtree represents a bitmap used for one
 354 * populated memory zone.
 355 */
 356struct mem_zone_bm_rtree {
 357	struct list_head list;		/* Link Zones together         */
 358	struct list_head nodes;		/* Radix Tree inner nodes      */
 359	struct list_head leaves;	/* Radix Tree leaves           */
 360	unsigned long start_pfn;	/* Zone start page frame       */
 361	unsigned long end_pfn;		/* Zone end page frame + 1     */
 362	struct rtree_node *rtree;	/* Radix Tree Root             */
 363	int levels;			/* Number of Radix Tree Levels */
 364	unsigned int blocks;		/* Number of Bitmap Blocks     */
 365};
 366
 367/* strcut bm_position is used for browsing memory bitmaps */
 368
 369struct bm_position {
 370	struct mem_zone_bm_rtree *zone;
 371	struct rtree_node *node;
 372	unsigned long node_pfn;
 373	int node_bit;
 374};
 375
 376struct memory_bitmap {
 377	struct list_head zones;
 378	struct linked_page *p_list;	/* list of pages used to store zone
 379					   bitmap objects and bitmap block
 380					   objects */
 381	struct bm_position cur;	/* most recently used bit position */
 382};
 383
 384/* Functions that operate on memory bitmaps */
 385
 386#define BM_ENTRIES_PER_LEVEL	(PAGE_SIZE / sizeof(unsigned long))
 387#if BITS_PER_LONG == 32
 388#define BM_RTREE_LEVEL_SHIFT	(PAGE_SHIFT - 2)
 389#else
 390#define BM_RTREE_LEVEL_SHIFT	(PAGE_SHIFT - 3)
 391#endif
 392#define BM_RTREE_LEVEL_MASK	((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
 393
 394/**
 395 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
 396 *
 397 * This function is used to allocate inner nodes as well as the
 398 * leave nodes of the radix tree. It also adds the node to the
 399 * corresponding linked list passed in by the *list parameter.
 400 */
 401static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
 402					   struct chain_allocator *ca,
 403					   struct list_head *list)
 404{
 405	struct rtree_node *node;
 406
 407	node = chain_alloc(ca, sizeof(struct rtree_node));
 408	if (!node)
 409		return NULL;
 410
 411	node->data = get_image_page(gfp_mask, safe_needed);
 412	if (!node->data)
 413		return NULL;
 414
 415	list_add_tail(&node->list, list);
 416
 417	return node;
 418}
 419
 420/**
 421 * add_rtree_block - Add a new leave node to the radix tree.
 422 *
 423 * The leave nodes need to be allocated in order to keep the leaves
 424 * linked list in order. This is guaranteed by the zone->blocks
 425 * counter.
 426 */
 427static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
 428			   int safe_needed, struct chain_allocator *ca)
 429{
 430	struct rtree_node *node, *block, **dst;
 431	unsigned int levels_needed, block_nr;
 432	int i;
 433
 434	block_nr = zone->blocks;
 435	levels_needed = 0;
 436
 437	/* How many levels do we need for this block nr? */
 438	while (block_nr) {
 439		levels_needed += 1;
 440		block_nr >>= BM_RTREE_LEVEL_SHIFT;
 441	}
 442
 443	/* Make sure the rtree has enough levels */
 444	for (i = zone->levels; i < levels_needed; i++) {
 445		node = alloc_rtree_node(gfp_mask, safe_needed, ca,
 446					&zone->nodes);
 447		if (!node)
 448			return -ENOMEM;
 449
 450		node->data[0] = (unsigned long)zone->rtree;
 451		zone->rtree = node;
 452		zone->levels += 1;
 453	}
 454
 455	/* Allocate new block */
 456	block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
 457	if (!block)
 458		return -ENOMEM;
 459
 460	/* Now walk the rtree to insert the block */
 461	node = zone->rtree;
 462	dst = &zone->rtree;
 463	block_nr = zone->blocks;
 464	for (i = zone->levels; i > 0; i--) {
 465		int index;
 466
 467		if (!node) {
 468			node = alloc_rtree_node(gfp_mask, safe_needed, ca,
 469						&zone->nodes);
 470			if (!node)
 471				return -ENOMEM;
 472			*dst = node;
 473		}
 474
 475		index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
 476		index &= BM_RTREE_LEVEL_MASK;
 477		dst = (struct rtree_node **)&((*dst)->data[index]);
 478		node = *dst;
 479	}
 480
 481	zone->blocks += 1;
 482	*dst = block;
 483
 484	return 0;
 485}
 486
 487static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
 488			       int clear_nosave_free);
 489
 490/**
 491 * create_zone_bm_rtree - Create a radix tree for one zone.
 492 *
 493 * Allocated the mem_zone_bm_rtree structure and initializes it.
 494 * This function also allocated and builds the radix tree for the
 495 * zone.
 496 */
 497static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
 498						      int safe_needed,
 499						      struct chain_allocator *ca,
 500						      unsigned long start,
 501						      unsigned long end)
 502{
 503	struct mem_zone_bm_rtree *zone;
 504	unsigned int i, nr_blocks;
 505	unsigned long pages;
 506
 507	pages = end - start;
 508	zone  = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
 509	if (!zone)
 510		return NULL;
 511
 512	INIT_LIST_HEAD(&zone->nodes);
 513	INIT_LIST_HEAD(&zone->leaves);
 514	zone->start_pfn = start;
 515	zone->end_pfn = end;
 516	nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
 517
 518	for (i = 0; i < nr_blocks; i++) {
 519		if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
 520			free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
 521			return NULL;
 522		}
 523	}
 524
 525	return zone;
 526}
 527
 528/**
 529 * free_zone_bm_rtree - Free the memory of the radix tree.
 530 *
 531 * Free all node pages of the radix tree. The mem_zone_bm_rtree
 532 * structure itself is not freed here nor are the rtree_node
 533 * structs.
 534 */
 535static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
 536			       int clear_nosave_free)
 537{
 538	struct rtree_node *node;
 539
 540	list_for_each_entry(node, &zone->nodes, list)
 541		free_image_page(node->data, clear_nosave_free);
 542
 543	list_for_each_entry(node, &zone->leaves, list)
 544		free_image_page(node->data, clear_nosave_free);
 545}
 546
 547static void memory_bm_position_reset(struct memory_bitmap *bm)
 548{
 549	bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
 550				  list);
 551	bm->cur.node = list_entry(bm->cur.zone->leaves.next,
 552				  struct rtree_node, list);
 553	bm->cur.node_pfn = 0;
 554	bm->cur.node_bit = 0;
 555}
 556
 557static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
 558
 559struct mem_extent {
 560	struct list_head hook;
 561	unsigned long start;
 562	unsigned long end;
 563};
 564
 565/**
 566 * free_mem_extents - Free a list of memory extents.
 567 * @list: List of extents to free.
 568 */
 569static void free_mem_extents(struct list_head *list)
 570{
 571	struct mem_extent *ext, *aux;
 572
 573	list_for_each_entry_safe(ext, aux, list, hook) {
 574		list_del(&ext->hook);
 575		kfree(ext);
 576	}
 577}
 578
 579/**
 580 * create_mem_extents - Create a list of memory extents.
 581 * @list: List to put the extents into.
 582 * @gfp_mask: Mask to use for memory allocations.
 583 *
 584 * The extents represent contiguous ranges of PFNs.
 585 */
 586static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
 587{
 588	struct zone *zone;
 589
 590	INIT_LIST_HEAD(list);
 591
 592	for_each_populated_zone(zone) {
 593		unsigned long zone_start, zone_end;
 594		struct mem_extent *ext, *cur, *aux;
 595
 596		zone_start = zone->zone_start_pfn;
 597		zone_end = zone_end_pfn(zone);
 598
 599		list_for_each_entry(ext, list, hook)
 600			if (zone_start <= ext->end)
 601				break;
 602
 603		if (&ext->hook == list || zone_end < ext->start) {
 604			/* New extent is necessary */
 605			struct mem_extent *new_ext;
 606
 607			new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
 608			if (!new_ext) {
 609				free_mem_extents(list);
 610				return -ENOMEM;
 611			}
 612			new_ext->start = zone_start;
 613			new_ext->end = zone_end;
 614			list_add_tail(&new_ext->hook, &ext->hook);
 615			continue;
 616		}
 617
 618		/* Merge this zone's range of PFNs with the existing one */
 619		if (zone_start < ext->start)
 620			ext->start = zone_start;
 621		if (zone_end > ext->end)
 622			ext->end = zone_end;
 623
 624		/* More merging may be possible */
 625		cur = ext;
 626		list_for_each_entry_safe_continue(cur, aux, list, hook) {
 627			if (zone_end < cur->start)
 628				break;
 629			if (zone_end < cur->end)
 630				ext->end = cur->end;
 631			list_del(&cur->hook);
 632			kfree(cur);
 633		}
 634	}
 635
 636	return 0;
 637}
 638
 639/**
 640 * memory_bm_create - Allocate memory for a memory bitmap.
 641 */
 642static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
 643			    int safe_needed)
 644{
 645	struct chain_allocator ca;
 646	struct list_head mem_extents;
 647	struct mem_extent *ext;
 648	int error;
 649
 650	chain_init(&ca, gfp_mask, safe_needed);
 651	INIT_LIST_HEAD(&bm->zones);
 652
 653	error = create_mem_extents(&mem_extents, gfp_mask);
 654	if (error)
 655		return error;
 656
 657	list_for_each_entry(ext, &mem_extents, hook) {
 658		struct mem_zone_bm_rtree *zone;
 659
 660		zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
 661					    ext->start, ext->end);
 662		if (!zone) {
 663			error = -ENOMEM;
 664			goto Error;
 665		}
 666		list_add_tail(&zone->list, &bm->zones);
 667	}
 668
 669	bm->p_list = ca.chain;
 670	memory_bm_position_reset(bm);
 671 Exit:
 672	free_mem_extents(&mem_extents);
 673	return error;
 674
 675 Error:
 676	bm->p_list = ca.chain;
 677	memory_bm_free(bm, PG_UNSAFE_CLEAR);
 678	goto Exit;
 679}
 680
 681/**
 682 * memory_bm_free - Free memory occupied by the memory bitmap.
 683 * @bm: Memory bitmap.
 684 */
 685static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
 686{
 687	struct mem_zone_bm_rtree *zone;
 688
 689	list_for_each_entry(zone, &bm->zones, list)
 690		free_zone_bm_rtree(zone, clear_nosave_free);
 691
 692	free_list_of_pages(bm->p_list, clear_nosave_free);
 693
 694	INIT_LIST_HEAD(&bm->zones);
 695}
 696
 697/**
 698 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
 699 *
 700 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
 701 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
 702 *
 703 * Walk the radix tree to find the page containing the bit that represents @pfn
 704 * and return the position of the bit in @addr and @bit_nr.
 705 */
 706static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
 707			      void **addr, unsigned int *bit_nr)
 708{
 709	struct mem_zone_bm_rtree *curr, *zone;
 710	struct rtree_node *node;
 711	int i, block_nr;
 712
 713	zone = bm->cur.zone;
 714
 715	if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
 716		goto zone_found;
 717
 718	zone = NULL;
 719
 720	/* Find the right zone */
 721	list_for_each_entry(curr, &bm->zones, list) {
 722		if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
 723			zone = curr;
 724			break;
 725		}
 726	}
 727
 728	if (!zone)
 729		return -EFAULT;
 730
 731zone_found:
 732	/*
 733	 * We have found the zone. Now walk the radix tree to find the leaf node
 734	 * for our PFN.
 735	 */
 736
 737	/*
 738	 * If the zone we wish to scan is the the current zone and the
 739	 * pfn falls into the current node then we do not need to walk
 740	 * the tree.
 741	 */
 742	node = bm->cur.node;
 743	if (zone == bm->cur.zone &&
 744	    ((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
 745		goto node_found;
 746
 747	node      = zone->rtree;
 748	block_nr  = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
 749
 750	for (i = zone->levels; i > 0; i--) {
 751		int index;
 752
 753		index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
 754		index &= BM_RTREE_LEVEL_MASK;
 755		BUG_ON(node->data[index] == 0);
 756		node = (struct rtree_node *)node->data[index];
 757	}
 758
 759node_found:
 760	/* Update last position */
 761	bm->cur.zone = zone;
 762	bm->cur.node = node;
 763	bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
 764
 765	/* Set return values */
 766	*addr = node->data;
 767	*bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
 768
 769	return 0;
 770}
 771
 772static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
 773{
 774	void *addr;
 775	unsigned int bit;
 776	int error;
 777
 778	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 779	BUG_ON(error);
 780	set_bit(bit, addr);
 781}
 782
 783static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
 784{
 785	void *addr;
 786	unsigned int bit;
 787	int error;
 788
 789	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 790	if (!error)
 791		set_bit(bit, addr);
 792
 793	return error;
 794}
 795
 796static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
 797{
 798	void *addr;
 799	unsigned int bit;
 800	int error;
 801
 802	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 803	BUG_ON(error);
 804	clear_bit(bit, addr);
 805}
 806
 807static void memory_bm_clear_current(struct memory_bitmap *bm)
 808{
 809	int bit;
 810
 811	bit = max(bm->cur.node_bit - 1, 0);
 812	clear_bit(bit, bm->cur.node->data);
 813}
 814
 815static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
 816{
 817	void *addr;
 818	unsigned int bit;
 819	int error;
 820
 821	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 822	BUG_ON(error);
 823	return test_bit(bit, addr);
 824}
 825
 826static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
 827{
 828	void *addr;
 829	unsigned int bit;
 830
 831	return !memory_bm_find_bit(bm, pfn, &addr, &bit);
 832}
 833
 834/*
 835 * rtree_next_node - Jump to the next leaf node.
 836 *
 837 * Set the position to the beginning of the next node in the
 838 * memory bitmap. This is either the next node in the current
 839 * zone's radix tree or the first node in the radix tree of the
 840 * next zone.
 841 *
 842 * Return true if there is a next node, false otherwise.
 843 */
 844static bool rtree_next_node(struct memory_bitmap *bm)
 845{
 846	if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
 847		bm->cur.node = list_entry(bm->cur.node->list.next,
 848					  struct rtree_node, list);
 849		bm->cur.node_pfn += BM_BITS_PER_BLOCK;
 850		bm->cur.node_bit  = 0;
 851		touch_softlockup_watchdog();
 852		return true;
 853	}
 854
 855	/* No more nodes, goto next zone */
 856	if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
 857		bm->cur.zone = list_entry(bm->cur.zone->list.next,
 858				  struct mem_zone_bm_rtree, list);
 859		bm->cur.node = list_entry(bm->cur.zone->leaves.next,
 860					  struct rtree_node, list);
 861		bm->cur.node_pfn = 0;
 862		bm->cur.node_bit = 0;
 863		return true;
 864	}
 865
 866	/* No more zones */
 867	return false;
 868}
 869
 870/**
 871 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
 872 * @bm: Memory bitmap.
 873 *
 874 * Starting from the last returned position this function searches for the next
 875 * set bit in @bm and returns the PFN represented by it.  If no more bits are
 876 * set, BM_END_OF_MAP is returned.
 877 *
 878 * It is required to run memory_bm_position_reset() before the first call to
 879 * this function for the given memory bitmap.
 880 */
 881static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
 882{
 883	unsigned long bits, pfn, pages;
 884	int bit;
 885
 886	do {
 887		pages	  = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
 888		bits      = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
 889		bit	  = find_next_bit(bm->cur.node->data, bits,
 890					  bm->cur.node_bit);
 891		if (bit < bits) {
 892			pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
 893			bm->cur.node_bit = bit + 1;
 894			return pfn;
 895		}
 896	} while (rtree_next_node(bm));
 897
 898	return BM_END_OF_MAP;
 899}
 900
 901/*
 902 * This structure represents a range of page frames the contents of which
 903 * should not be saved during hibernation.
 904 */
 905struct nosave_region {
 906	struct list_head list;
 907	unsigned long start_pfn;
 908	unsigned long end_pfn;
 909};
 910
 911static LIST_HEAD(nosave_regions);
 912
 913static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
 914{
 915	struct rtree_node *node;
 916
 917	list_for_each_entry(node, &zone->nodes, list)
 918		recycle_safe_page(node->data);
 919
 920	list_for_each_entry(node, &zone->leaves, list)
 921		recycle_safe_page(node->data);
 922}
 923
 924static void memory_bm_recycle(struct memory_bitmap *bm)
 925{
 926	struct mem_zone_bm_rtree *zone;
 927	struct linked_page *p_list;
 928
 929	list_for_each_entry(zone, &bm->zones, list)
 930		recycle_zone_bm_rtree(zone);
 931
 932	p_list = bm->p_list;
 933	while (p_list) {
 934		struct linked_page *lp = p_list;
 935
 936		p_list = lp->next;
 937		recycle_safe_page(lp);
 938	}
 939}
 940
 941/**
 942 * register_nosave_region - Register a region of unsaveable memory.
 943 *
 944 * Register a range of page frames the contents of which should not be saved
 945 * during hibernation (to be used in the early initialization code).
 946 */
 947void __init __register_nosave_region(unsigned long start_pfn,
 948				     unsigned long end_pfn, int use_kmalloc)
 949{
 950	struct nosave_region *region;
 951
 952	if (start_pfn >= end_pfn)
 953		return;
 954
 955	if (!list_empty(&nosave_regions)) {
 956		/* Try to extend the previous region (they should be sorted) */
 957		region = list_entry(nosave_regions.prev,
 958					struct nosave_region, list);
 959		if (region->end_pfn == start_pfn) {
 960			region->end_pfn = end_pfn;
 961			goto Report;
 962		}
 963	}
 964	if (use_kmalloc) {
 965		/* During init, this shouldn't fail */
 966		region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
 967		BUG_ON(!region);
 968	} else {
 969		/* This allocation cannot fail */
 970		region = memblock_alloc(sizeof(struct nosave_region),
 971					SMP_CACHE_BYTES);
 972		if (!region)
 973			panic("%s: Failed to allocate %zu bytes\n", __func__,
 974			      sizeof(struct nosave_region));
 975	}
 976	region->start_pfn = start_pfn;
 977	region->end_pfn = end_pfn;
 978	list_add_tail(&region->list, &nosave_regions);
 979 Report:
 980	pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
 981		(unsigned long long) start_pfn << PAGE_SHIFT,
 982		((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
 983}
 984
 985/*
 986 * Set bits in this map correspond to the page frames the contents of which
 987 * should not be saved during the suspend.
 988 */
 989static struct memory_bitmap *forbidden_pages_map;
 990
 991/* Set bits in this map correspond to free page frames. */
 992static struct memory_bitmap *free_pages_map;
 993
 994/*
 995 * Each page frame allocated for creating the image is marked by setting the
 996 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
 997 */
 998
 999void swsusp_set_page_free(struct page *page)
1000{
1001	if (free_pages_map)
1002		memory_bm_set_bit(free_pages_map, page_to_pfn(page));
1003}
1004
1005static int swsusp_page_is_free(struct page *page)
1006{
1007	return free_pages_map ?
1008		memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1009}
1010
1011void swsusp_unset_page_free(struct page *page)
1012{
1013	if (free_pages_map)
1014		memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1015}
1016
1017static void swsusp_set_page_forbidden(struct page *page)
1018{
1019	if (forbidden_pages_map)
1020		memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1021}
1022
1023int swsusp_page_is_forbidden(struct page *page)
1024{
1025	return forbidden_pages_map ?
1026		memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1027}
1028
1029static void swsusp_unset_page_forbidden(struct page *page)
1030{
1031	if (forbidden_pages_map)
1032		memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1033}
1034
1035/**
1036 * mark_nosave_pages - Mark pages that should not be saved.
1037 * @bm: Memory bitmap.
1038 *
1039 * Set the bits in @bm that correspond to the page frames the contents of which
1040 * should not be saved.
1041 */
1042static void mark_nosave_pages(struct memory_bitmap *bm)
1043{
1044	struct nosave_region *region;
1045
1046	if (list_empty(&nosave_regions))
1047		return;
1048
1049	list_for_each_entry(region, &nosave_regions, list) {
1050		unsigned long pfn;
1051
1052		pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1053			 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1054			 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1055				- 1);
1056
1057		for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1058			if (pfn_valid(pfn)) {
1059				/*
1060				 * It is safe to ignore the result of
1061				 * mem_bm_set_bit_check() here, since we won't
1062				 * touch the PFNs for which the error is
1063				 * returned anyway.
1064				 */
1065				mem_bm_set_bit_check(bm, pfn);
1066			}
1067	}
1068}
1069
1070/**
1071 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1072 *
1073 * Create bitmaps needed for marking page frames that should not be saved and
1074 * free page frames.  The forbidden_pages_map and free_pages_map pointers are
1075 * only modified if everything goes well, because we don't want the bits to be
1076 * touched before both bitmaps are set up.
1077 */
1078int create_basic_memory_bitmaps(void)
1079{
1080	struct memory_bitmap *bm1, *bm2;
1081	int error = 0;
1082
1083	if (forbidden_pages_map && free_pages_map)
1084		return 0;
1085	else
1086		BUG_ON(forbidden_pages_map || free_pages_map);
1087
1088	bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1089	if (!bm1)
1090		return -ENOMEM;
1091
1092	error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1093	if (error)
1094		goto Free_first_object;
1095
1096	bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1097	if (!bm2)
1098		goto Free_first_bitmap;
1099
1100	error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1101	if (error)
1102		goto Free_second_object;
1103
1104	forbidden_pages_map = bm1;
1105	free_pages_map = bm2;
1106	mark_nosave_pages(forbidden_pages_map);
1107
1108	pr_debug("Basic memory bitmaps created\n");
1109
1110	return 0;
1111
1112 Free_second_object:
1113	kfree(bm2);
1114 Free_first_bitmap:
1115 	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1116 Free_first_object:
1117	kfree(bm1);
1118	return -ENOMEM;
1119}
1120
1121/**
1122 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1123 *
1124 * Free memory bitmaps allocated by create_basic_memory_bitmaps().  The
1125 * auxiliary pointers are necessary so that the bitmaps themselves are not
1126 * referred to while they are being freed.
1127 */
1128void free_basic_memory_bitmaps(void)
1129{
1130	struct memory_bitmap *bm1, *bm2;
1131
1132	if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1133		return;
1134
1135	bm1 = forbidden_pages_map;
1136	bm2 = free_pages_map;
1137	forbidden_pages_map = NULL;
1138	free_pages_map = NULL;
1139	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1140	kfree(bm1);
1141	memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1142	kfree(bm2);
1143
1144	pr_debug("Basic memory bitmaps freed\n");
1145}
1146
1147void clear_free_pages(void)
1148{
 
1149	struct memory_bitmap *bm = free_pages_map;
1150	unsigned long pfn;
1151
1152	if (WARN_ON(!(free_pages_map)))
1153		return;
1154
1155	if (IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) || want_init_on_free()) {
1156		memory_bm_position_reset(bm);
1157		pfn = memory_bm_next_pfn(bm);
1158		while (pfn != BM_END_OF_MAP) {
1159			if (pfn_valid(pfn))
1160				clear_highpage(pfn_to_page(pfn));
1161
1162			pfn = memory_bm_next_pfn(bm);
1163		}
1164		memory_bm_position_reset(bm);
1165		pr_info("free pages cleared after restore\n");
1166	}
 
 
 
1167}
1168
1169/**
1170 * snapshot_additional_pages - Estimate the number of extra pages needed.
1171 * @zone: Memory zone to carry out the computation for.
1172 *
1173 * Estimate the number of additional pages needed for setting up a hibernation
1174 * image data structures for @zone (usually, the returned value is greater than
1175 * the exact number).
1176 */
1177unsigned int snapshot_additional_pages(struct zone *zone)
1178{
1179	unsigned int rtree, nodes;
1180
1181	rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1182	rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1183			      LINKED_PAGE_DATA_SIZE);
1184	while (nodes > 1) {
1185		nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1186		rtree += nodes;
1187	}
1188
1189	return 2 * rtree;
1190}
1191
1192#ifdef CONFIG_HIGHMEM
1193/**
1194 * count_free_highmem_pages - Compute the total number of free highmem pages.
1195 *
1196 * The returned number is system-wide.
1197 */
1198static unsigned int count_free_highmem_pages(void)
1199{
1200	struct zone *zone;
1201	unsigned int cnt = 0;
1202
1203	for_each_populated_zone(zone)
1204		if (is_highmem(zone))
1205			cnt += zone_page_state(zone, NR_FREE_PAGES);
1206
1207	return cnt;
1208}
1209
1210/**
1211 * saveable_highmem_page - Check if a highmem page is saveable.
1212 *
1213 * Determine whether a highmem page should be included in a hibernation image.
1214 *
1215 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1216 * and it isn't part of a free chunk of pages.
1217 */
1218static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1219{
1220	struct page *page;
1221
1222	if (!pfn_valid(pfn))
1223		return NULL;
1224
1225	page = pfn_to_online_page(pfn);
1226	if (!page || page_zone(page) != zone)
1227		return NULL;
1228
1229	BUG_ON(!PageHighMem(page));
1230
1231	if (swsusp_page_is_forbidden(page) ||  swsusp_page_is_free(page))
1232		return NULL;
1233
1234	if (PageReserved(page) || PageOffline(page))
1235		return NULL;
1236
1237	if (page_is_guard(page))
1238		return NULL;
1239
1240	return page;
1241}
1242
1243/**
1244 * count_highmem_pages - Compute the total number of saveable highmem pages.
1245 */
1246static unsigned int count_highmem_pages(void)
1247{
1248	struct zone *zone;
1249	unsigned int n = 0;
1250
1251	for_each_populated_zone(zone) {
1252		unsigned long pfn, max_zone_pfn;
1253
1254		if (!is_highmem(zone))
1255			continue;
1256
1257		mark_free_pages(zone);
1258		max_zone_pfn = zone_end_pfn(zone);
1259		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1260			if (saveable_highmem_page(zone, pfn))
1261				n++;
1262	}
1263	return n;
1264}
1265#else
1266static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1267{
1268	return NULL;
1269}
1270#endif /* CONFIG_HIGHMEM */
1271
1272/**
1273 * saveable_page - Check if the given page is saveable.
1274 *
1275 * Determine whether a non-highmem page should be included in a hibernation
1276 * image.
1277 *
1278 * We should save the page if it isn't Nosave, and is not in the range
1279 * of pages statically defined as 'unsaveable', and it isn't part of
1280 * a free chunk of pages.
1281 */
1282static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1283{
1284	struct page *page;
1285
1286	if (!pfn_valid(pfn))
1287		return NULL;
1288
1289	page = pfn_to_online_page(pfn);
1290	if (!page || page_zone(page) != zone)
1291		return NULL;
1292
1293	BUG_ON(PageHighMem(page));
1294
1295	if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1296		return NULL;
1297
1298	if (PageOffline(page))
1299		return NULL;
1300
1301	if (PageReserved(page)
1302	    && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1303		return NULL;
1304
1305	if (page_is_guard(page))
1306		return NULL;
1307
1308	return page;
1309}
1310
1311/**
1312 * count_data_pages - Compute the total number of saveable non-highmem pages.
1313 */
1314static unsigned int count_data_pages(void)
1315{
1316	struct zone *zone;
1317	unsigned long pfn, max_zone_pfn;
1318	unsigned int n = 0;
1319
1320	for_each_populated_zone(zone) {
1321		if (is_highmem(zone))
1322			continue;
1323
1324		mark_free_pages(zone);
1325		max_zone_pfn = zone_end_pfn(zone);
1326		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1327			if (saveable_page(zone, pfn))
1328				n++;
1329	}
1330	return n;
1331}
1332
1333/*
1334 * This is needed, because copy_page and memcpy are not usable for copying
1335 * task structs.
1336 */
1337static inline void do_copy_page(long *dst, long *src)
1338{
1339	int n;
1340
1341	for (n = PAGE_SIZE / sizeof(long); n; n--)
1342		*dst++ = *src++;
1343}
1344
1345/**
1346 * safe_copy_page - Copy a page in a safe way.
1347 *
1348 * Check if the page we are going to copy is marked as present in the kernel
1349 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1350 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1351 * always returns 'true'.
1352 */
1353static void safe_copy_page(void *dst, struct page *s_page)
1354{
1355	if (kernel_page_present(s_page)) {
1356		do_copy_page(dst, page_address(s_page));
1357	} else {
1358		kernel_map_pages(s_page, 1, 1);
1359		do_copy_page(dst, page_address(s_page));
1360		kernel_map_pages(s_page, 1, 0);
1361	}
1362}
1363
1364#ifdef CONFIG_HIGHMEM
1365static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1366{
1367	return is_highmem(zone) ?
1368		saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1369}
1370
1371static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1372{
1373	struct page *s_page, *d_page;
1374	void *src, *dst;
1375
1376	s_page = pfn_to_page(src_pfn);
1377	d_page = pfn_to_page(dst_pfn);
1378	if (PageHighMem(s_page)) {
1379		src = kmap_atomic(s_page);
1380		dst = kmap_atomic(d_page);
1381		do_copy_page(dst, src);
1382		kunmap_atomic(dst);
1383		kunmap_atomic(src);
1384	} else {
1385		if (PageHighMem(d_page)) {
1386			/*
1387			 * The page pointed to by src may contain some kernel
1388			 * data modified by kmap_atomic()
1389			 */
1390			safe_copy_page(buffer, s_page);
1391			dst = kmap_atomic(d_page);
1392			copy_page(dst, buffer);
1393			kunmap_atomic(dst);
1394		} else {
1395			safe_copy_page(page_address(d_page), s_page);
1396		}
1397	}
1398}
1399#else
1400#define page_is_saveable(zone, pfn)	saveable_page(zone, pfn)
1401
1402static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1403{
1404	safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1405				pfn_to_page(src_pfn));
1406}
1407#endif /* CONFIG_HIGHMEM */
1408
1409static void copy_data_pages(struct memory_bitmap *copy_bm,
1410			    struct memory_bitmap *orig_bm)
1411{
1412	struct zone *zone;
1413	unsigned long pfn;
1414
1415	for_each_populated_zone(zone) {
1416		unsigned long max_zone_pfn;
1417
1418		mark_free_pages(zone);
1419		max_zone_pfn = zone_end_pfn(zone);
1420		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1421			if (page_is_saveable(zone, pfn))
1422				memory_bm_set_bit(orig_bm, pfn);
1423	}
1424	memory_bm_position_reset(orig_bm);
1425	memory_bm_position_reset(copy_bm);
1426	for(;;) {
1427		pfn = memory_bm_next_pfn(orig_bm);
1428		if (unlikely(pfn == BM_END_OF_MAP))
1429			break;
1430		copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1431	}
1432}
1433
1434/* Total number of image pages */
1435static unsigned int nr_copy_pages;
1436/* Number of pages needed for saving the original pfns of the image pages */
1437static unsigned int nr_meta_pages;
1438/*
1439 * Numbers of normal and highmem page frames allocated for hibernation image
1440 * before suspending devices.
1441 */
1442static unsigned int alloc_normal, alloc_highmem;
1443/*
1444 * Memory bitmap used for marking saveable pages (during hibernation) or
1445 * hibernation image pages (during restore)
1446 */
1447static struct memory_bitmap orig_bm;
1448/*
1449 * Memory bitmap used during hibernation for marking allocated page frames that
1450 * will contain copies of saveable pages.  During restore it is initially used
1451 * for marking hibernation image pages, but then the set bits from it are
1452 * duplicated in @orig_bm and it is released.  On highmem systems it is next
1453 * used for marking "safe" highmem pages, but it has to be reinitialized for
1454 * this purpose.
1455 */
1456static struct memory_bitmap copy_bm;
1457
1458/**
1459 * swsusp_free - Free pages allocated for hibernation image.
1460 *
1461 * Image pages are alocated before snapshot creation, so they need to be
1462 * released after resume.
1463 */
1464void swsusp_free(void)
1465{
1466	unsigned long fb_pfn, fr_pfn;
1467
1468	if (!forbidden_pages_map || !free_pages_map)
1469		goto out;
1470
1471	memory_bm_position_reset(forbidden_pages_map);
1472	memory_bm_position_reset(free_pages_map);
1473
1474loop:
1475	fr_pfn = memory_bm_next_pfn(free_pages_map);
1476	fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1477
1478	/*
1479	 * Find the next bit set in both bitmaps. This is guaranteed to
1480	 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1481	 */
1482	do {
1483		if (fb_pfn < fr_pfn)
1484			fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1485		if (fr_pfn < fb_pfn)
1486			fr_pfn = memory_bm_next_pfn(free_pages_map);
1487	} while (fb_pfn != fr_pfn);
1488
1489	if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1490		struct page *page = pfn_to_page(fr_pfn);
1491
1492		memory_bm_clear_current(forbidden_pages_map);
1493		memory_bm_clear_current(free_pages_map);
1494		hibernate_restore_unprotect_page(page_address(page));
1495		__free_page(page);
1496		goto loop;
1497	}
1498
1499out:
1500	nr_copy_pages = 0;
1501	nr_meta_pages = 0;
1502	restore_pblist = NULL;
1503	buffer = NULL;
1504	alloc_normal = 0;
1505	alloc_highmem = 0;
1506	hibernate_restore_protection_end();
1507}
1508
1509/* Helper functions used for the shrinking of memory. */
1510
1511#define GFP_IMAGE	(GFP_KERNEL | __GFP_NOWARN)
1512
1513/**
1514 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1515 * @nr_pages: Number of page frames to allocate.
1516 * @mask: GFP flags to use for the allocation.
1517 *
1518 * Return value: Number of page frames actually allocated
1519 */
1520static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1521{
1522	unsigned long nr_alloc = 0;
1523
1524	while (nr_pages > 0) {
1525		struct page *page;
1526
1527		page = alloc_image_page(mask);
1528		if (!page)
1529			break;
1530		memory_bm_set_bit(&copy_bm, page_to_pfn(page));
1531		if (PageHighMem(page))
1532			alloc_highmem++;
1533		else
1534			alloc_normal++;
1535		nr_pages--;
1536		nr_alloc++;
1537	}
1538
1539	return nr_alloc;
1540}
1541
1542static unsigned long preallocate_image_memory(unsigned long nr_pages,
1543					      unsigned long avail_normal)
1544{
1545	unsigned long alloc;
1546
1547	if (avail_normal <= alloc_normal)
1548		return 0;
1549
1550	alloc = avail_normal - alloc_normal;
1551	if (nr_pages < alloc)
1552		alloc = nr_pages;
1553
1554	return preallocate_image_pages(alloc, GFP_IMAGE);
1555}
1556
1557#ifdef CONFIG_HIGHMEM
1558static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1559{
1560	return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1561}
1562
1563/**
1564 *  __fraction - Compute (an approximation of) x * (multiplier / base).
1565 */
1566static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1567{
1568	return div64_u64(x * multiplier, base);
 
 
1569}
1570
1571static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1572						  unsigned long highmem,
1573						  unsigned long total)
1574{
1575	unsigned long alloc = __fraction(nr_pages, highmem, total);
1576
1577	return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1578}
1579#else /* CONFIG_HIGHMEM */
1580static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1581{
1582	return 0;
1583}
1584
1585static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1586							 unsigned long highmem,
1587							 unsigned long total)
1588{
1589	return 0;
1590}
1591#endif /* CONFIG_HIGHMEM */
1592
1593/**
1594 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1595 */
1596static unsigned long free_unnecessary_pages(void)
1597{
1598	unsigned long save, to_free_normal, to_free_highmem, free;
1599
1600	save = count_data_pages();
1601	if (alloc_normal >= save) {
1602		to_free_normal = alloc_normal - save;
1603		save = 0;
1604	} else {
1605		to_free_normal = 0;
1606		save -= alloc_normal;
1607	}
1608	save += count_highmem_pages();
1609	if (alloc_highmem >= save) {
1610		to_free_highmem = alloc_highmem - save;
1611	} else {
1612		to_free_highmem = 0;
1613		save -= alloc_highmem;
1614		if (to_free_normal > save)
1615			to_free_normal -= save;
1616		else
1617			to_free_normal = 0;
1618	}
1619	free = to_free_normal + to_free_highmem;
1620
1621	memory_bm_position_reset(&copy_bm);
1622
1623	while (to_free_normal > 0 || to_free_highmem > 0) {
1624		unsigned long pfn = memory_bm_next_pfn(&copy_bm);
1625		struct page *page = pfn_to_page(pfn);
1626
1627		if (PageHighMem(page)) {
1628			if (!to_free_highmem)
1629				continue;
1630			to_free_highmem--;
1631			alloc_highmem--;
1632		} else {
1633			if (!to_free_normal)
1634				continue;
1635			to_free_normal--;
1636			alloc_normal--;
1637		}
1638		memory_bm_clear_bit(&copy_bm, pfn);
1639		swsusp_unset_page_forbidden(page);
1640		swsusp_unset_page_free(page);
1641		__free_page(page);
1642	}
1643
1644	return free;
1645}
1646
1647/**
1648 * minimum_image_size - Estimate the minimum acceptable size of an image.
1649 * @saveable: Number of saveable pages in the system.
1650 *
1651 * We want to avoid attempting to free too much memory too hard, so estimate the
1652 * minimum acceptable size of a hibernation image to use as the lower limit for
1653 * preallocating memory.
1654 *
1655 * We assume that the minimum image size should be proportional to
1656 *
1657 * [number of saveable pages] - [number of pages that can be freed in theory]
1658 *
1659 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1660 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
 
1661 */
1662static unsigned long minimum_image_size(unsigned long saveable)
1663{
1664	unsigned long size;
1665
1666	size = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B)
1667		+ global_node_page_state(NR_ACTIVE_ANON)
1668		+ global_node_page_state(NR_INACTIVE_ANON)
1669		+ global_node_page_state(NR_ACTIVE_FILE)
1670		+ global_node_page_state(NR_INACTIVE_FILE);
 
1671
1672	return saveable <= size ? 0 : saveable - size;
1673}
1674
1675/**
1676 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1677 *
1678 * To create a hibernation image it is necessary to make a copy of every page
1679 * frame in use.  We also need a number of page frames to be free during
1680 * hibernation for allocations made while saving the image and for device
1681 * drivers, in case they need to allocate memory from their hibernation
1682 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1683 * estimate) and reserved_size divided by PAGE_SIZE (which is tunable through
1684 * /sys/power/reserved_size, respectively).  To make this happen, we compute the
1685 * total number of available page frames and allocate at least
1686 *
1687 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1688 *  + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1689 *
1690 * of them, which corresponds to the maximum size of a hibernation image.
1691 *
1692 * If image_size is set below the number following from the above formula,
1693 * the preallocation of memory is continued until the total number of saveable
1694 * pages in the system is below the requested image size or the minimum
1695 * acceptable image size returned by minimum_image_size(), whichever is greater.
1696 */
1697int hibernate_preallocate_memory(void)
1698{
1699	struct zone *zone;
1700	unsigned long saveable, size, max_size, count, highmem, pages = 0;
1701	unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1702	ktime_t start, stop;
1703	int error;
1704
1705	pr_info("Preallocating image memory\n");
1706	start = ktime_get();
1707
1708	error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1709	if (error) {
1710		pr_err("Cannot allocate original bitmap\n");
1711		goto err_out;
1712	}
1713
1714	error = memory_bm_create(&copy_bm, GFP_IMAGE, PG_ANY);
1715	if (error) {
1716		pr_err("Cannot allocate copy bitmap\n");
1717		goto err_out;
1718	}
1719
1720	alloc_normal = 0;
1721	alloc_highmem = 0;
1722
1723	/* Count the number of saveable data pages. */
1724	save_highmem = count_highmem_pages();
1725	saveable = count_data_pages();
1726
1727	/*
1728	 * Compute the total number of page frames we can use (count) and the
1729	 * number of pages needed for image metadata (size).
1730	 */
1731	count = saveable;
1732	saveable += save_highmem;
1733	highmem = save_highmem;
1734	size = 0;
1735	for_each_populated_zone(zone) {
1736		size += snapshot_additional_pages(zone);
1737		if (is_highmem(zone))
1738			highmem += zone_page_state(zone, NR_FREE_PAGES);
1739		else
1740			count += zone_page_state(zone, NR_FREE_PAGES);
1741	}
1742	avail_normal = count;
1743	count += highmem;
1744	count -= totalreserve_pages;
1745
 
 
 
1746	/* Compute the maximum number of saveable pages to leave in memory. */
1747	max_size = (count - (size + PAGES_FOR_IO)) / 2
1748			- 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1749	/* Compute the desired number of image pages specified by image_size. */
1750	size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1751	if (size > max_size)
1752		size = max_size;
1753	/*
1754	 * If the desired number of image pages is at least as large as the
1755	 * current number of saveable pages in memory, allocate page frames for
1756	 * the image and we're done.
1757	 */
1758	if (size >= saveable) {
1759		pages = preallocate_image_highmem(save_highmem);
1760		pages += preallocate_image_memory(saveable - pages, avail_normal);
1761		goto out;
1762	}
1763
1764	/* Estimate the minimum size of the image. */
1765	pages = minimum_image_size(saveable);
1766	/*
1767	 * To avoid excessive pressure on the normal zone, leave room in it to
1768	 * accommodate an image of the minimum size (unless it's already too
1769	 * small, in which case don't preallocate pages from it at all).
1770	 */
1771	if (avail_normal > pages)
1772		avail_normal -= pages;
1773	else
1774		avail_normal = 0;
1775	if (size < pages)
1776		size = min_t(unsigned long, pages, max_size);
1777
1778	/*
1779	 * Let the memory management subsystem know that we're going to need a
1780	 * large number of page frames to allocate and make it free some memory.
1781	 * NOTE: If this is not done, performance will be hurt badly in some
1782	 * test cases.
1783	 */
1784	shrink_all_memory(saveable - size);
1785
1786	/*
1787	 * The number of saveable pages in memory was too high, so apply some
1788	 * pressure to decrease it.  First, make room for the largest possible
1789	 * image and fail if that doesn't work.  Next, try to decrease the size
1790	 * of the image as much as indicated by 'size' using allocations from
1791	 * highmem and non-highmem zones separately.
1792	 */
1793	pages_highmem = preallocate_image_highmem(highmem / 2);
1794	alloc = count - max_size;
1795	if (alloc > pages_highmem)
1796		alloc -= pages_highmem;
1797	else
1798		alloc = 0;
1799	pages = preallocate_image_memory(alloc, avail_normal);
1800	if (pages < alloc) {
1801		/* We have exhausted non-highmem pages, try highmem. */
1802		alloc -= pages;
1803		pages += pages_highmem;
1804		pages_highmem = preallocate_image_highmem(alloc);
1805		if (pages_highmem < alloc) {
1806			pr_err("Image allocation is %lu pages short\n",
1807				alloc - pages_highmem);
1808			goto err_out;
1809		}
1810		pages += pages_highmem;
1811		/*
1812		 * size is the desired number of saveable pages to leave in
1813		 * memory, so try to preallocate (all memory - size) pages.
1814		 */
1815		alloc = (count - pages) - size;
1816		pages += preallocate_image_highmem(alloc);
1817	} else {
1818		/*
1819		 * There are approximately max_size saveable pages at this point
1820		 * and we want to reduce this number down to size.
1821		 */
1822		alloc = max_size - size;
1823		size = preallocate_highmem_fraction(alloc, highmem, count);
1824		pages_highmem += size;
1825		alloc -= size;
1826		size = preallocate_image_memory(alloc, avail_normal);
1827		pages_highmem += preallocate_image_highmem(alloc - size);
1828		pages += pages_highmem + size;
1829	}
1830
1831	/*
1832	 * We only need as many page frames for the image as there are saveable
1833	 * pages in memory, but we have allocated more.  Release the excessive
1834	 * ones now.
1835	 */
1836	pages -= free_unnecessary_pages();
1837
1838 out:
1839	stop = ktime_get();
1840	pr_info("Allocated %lu pages for snapshot\n", pages);
1841	swsusp_show_speed(start, stop, pages, "Allocated");
1842
1843	return 0;
1844
1845 err_out:
 
1846	swsusp_free();
1847	return -ENOMEM;
1848}
1849
1850#ifdef CONFIG_HIGHMEM
1851/**
1852 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1853 *
1854 * Compute the number of non-highmem pages that will be necessary for creating
1855 * copies of highmem pages.
1856 */
1857static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1858{
1859	unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1860
1861	if (free_highmem >= nr_highmem)
1862		nr_highmem = 0;
1863	else
1864		nr_highmem -= free_highmem;
1865
1866	return nr_highmem;
1867}
1868#else
1869static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1870#endif /* CONFIG_HIGHMEM */
1871
1872/**
1873 * enough_free_mem - Check if there is enough free memory for the image.
1874 */
1875static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1876{
1877	struct zone *zone;
1878	unsigned int free = alloc_normal;
1879
1880	for_each_populated_zone(zone)
1881		if (!is_highmem(zone))
1882			free += zone_page_state(zone, NR_FREE_PAGES);
1883
1884	nr_pages += count_pages_for_highmem(nr_highmem);
1885	pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
1886		 nr_pages, PAGES_FOR_IO, free);
1887
1888	return free > nr_pages + PAGES_FOR_IO;
1889}
1890
1891#ifdef CONFIG_HIGHMEM
1892/**
1893 * get_highmem_buffer - Allocate a buffer for highmem pages.
1894 *
1895 * If there are some highmem pages in the hibernation image, we may need a
1896 * buffer to copy them and/or load their data.
1897 */
1898static inline int get_highmem_buffer(int safe_needed)
1899{
1900	buffer = get_image_page(GFP_ATOMIC, safe_needed);
1901	return buffer ? 0 : -ENOMEM;
1902}
1903
1904/**
1905 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1906 *
1907 * Try to allocate as many pages as needed, but if the number of free highmem
1908 * pages is less than that, allocate them all.
1909 */
1910static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1911					       unsigned int nr_highmem)
1912{
1913	unsigned int to_alloc = count_free_highmem_pages();
1914
1915	if (to_alloc > nr_highmem)
1916		to_alloc = nr_highmem;
1917
1918	nr_highmem -= to_alloc;
1919	while (to_alloc-- > 0) {
1920		struct page *page;
1921
1922		page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1923		memory_bm_set_bit(bm, page_to_pfn(page));
1924	}
1925	return nr_highmem;
1926}
1927#else
1928static inline int get_highmem_buffer(int safe_needed) { return 0; }
1929
1930static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1931					       unsigned int n) { return 0; }
1932#endif /* CONFIG_HIGHMEM */
1933
1934/**
1935 * swsusp_alloc - Allocate memory for hibernation image.
1936 *
1937 * We first try to allocate as many highmem pages as there are
1938 * saveable highmem pages in the system.  If that fails, we allocate
1939 * non-highmem pages for the copies of the remaining highmem ones.
1940 *
1941 * In this approach it is likely that the copies of highmem pages will
1942 * also be located in the high memory, because of the way in which
1943 * copy_data_pages() works.
1944 */
1945static int swsusp_alloc(struct memory_bitmap *copy_bm,
 
1946			unsigned int nr_pages, unsigned int nr_highmem)
1947{
1948	if (nr_highmem > 0) {
1949		if (get_highmem_buffer(PG_ANY))
1950			goto err_out;
1951		if (nr_highmem > alloc_highmem) {
1952			nr_highmem -= alloc_highmem;
1953			nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1954		}
1955	}
1956	if (nr_pages > alloc_normal) {
1957		nr_pages -= alloc_normal;
1958		while (nr_pages-- > 0) {
1959			struct page *page;
1960
1961			page = alloc_image_page(GFP_ATOMIC);
1962			if (!page)
1963				goto err_out;
1964			memory_bm_set_bit(copy_bm, page_to_pfn(page));
1965		}
1966	}
1967
1968	return 0;
1969
1970 err_out:
1971	swsusp_free();
1972	return -ENOMEM;
1973}
1974
1975asmlinkage __visible int swsusp_save(void)
1976{
1977	unsigned int nr_pages, nr_highmem;
1978
1979	pr_info("Creating image:\n");
1980
1981	drain_local_pages(NULL);
1982	nr_pages = count_data_pages();
1983	nr_highmem = count_highmem_pages();
1984	pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
1985
1986	if (!enough_free_mem(nr_pages, nr_highmem)) {
1987		pr_err("Not enough free memory\n");
1988		return -ENOMEM;
1989	}
1990
1991	if (swsusp_alloc(&copy_bm, nr_pages, nr_highmem)) {
1992		pr_err("Memory allocation failed\n");
1993		return -ENOMEM;
1994	}
1995
1996	/*
1997	 * During allocating of suspend pagedir, new cold pages may appear.
1998	 * Kill them.
1999	 */
2000	drain_local_pages(NULL);
2001	copy_data_pages(&copy_bm, &orig_bm);
2002
2003	/*
2004	 * End of critical section. From now on, we can write to memory,
2005	 * but we should not touch disk. This specially means we must _not_
2006	 * touch swap space! Except we must write out our image of course.
2007	 */
2008
2009	nr_pages += nr_highmem;
2010	nr_copy_pages = nr_pages;
2011	nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2012
2013	pr_info("Image created (%d pages copied)\n", nr_pages);
 
2014
2015	return 0;
2016}
2017
2018#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2019static int init_header_complete(struct swsusp_info *info)
2020{
2021	memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2022	info->version_code = LINUX_VERSION_CODE;
2023	return 0;
2024}
2025
2026static const char *check_image_kernel(struct swsusp_info *info)
2027{
2028	if (info->version_code != LINUX_VERSION_CODE)
2029		return "kernel version";
2030	if (strcmp(info->uts.sysname,init_utsname()->sysname))
2031		return "system type";
2032	if (strcmp(info->uts.release,init_utsname()->release))
2033		return "kernel release";
2034	if (strcmp(info->uts.version,init_utsname()->version))
2035		return "version";
2036	if (strcmp(info->uts.machine,init_utsname()->machine))
2037		return "machine";
2038	return NULL;
2039}
2040#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2041
2042unsigned long snapshot_get_image_size(void)
2043{
2044	return nr_copy_pages + nr_meta_pages + 1;
2045}
2046
2047static int init_header(struct swsusp_info *info)
2048{
2049	memset(info, 0, sizeof(struct swsusp_info));
2050	info->num_physpages = get_num_physpages();
2051	info->image_pages = nr_copy_pages;
2052	info->pages = snapshot_get_image_size();
2053	info->size = info->pages;
2054	info->size <<= PAGE_SHIFT;
2055	return init_header_complete(info);
2056}
2057
2058/**
2059 * pack_pfns - Prepare PFNs for saving.
2060 * @bm: Memory bitmap.
2061 * @buf: Memory buffer to store the PFNs in.
2062 *
2063 * PFNs corresponding to set bits in @bm are stored in the area of memory
2064 * pointed to by @buf (1 page at a time).
2065 */
2066static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2067{
2068	int j;
2069
2070	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2071		buf[j] = memory_bm_next_pfn(bm);
2072		if (unlikely(buf[j] == BM_END_OF_MAP))
2073			break;
 
 
2074	}
2075}
2076
2077/**
2078 * snapshot_read_next - Get the address to read the next image page from.
2079 * @handle: Snapshot handle to be used for the reading.
2080 *
2081 * On the first call, @handle should point to a zeroed snapshot_handle
2082 * structure.  The structure gets populated then and a pointer to it should be
2083 * passed to this function every next time.
2084 *
2085 * On success, the function returns a positive number.  Then, the caller
2086 * is allowed to read up to the returned number of bytes from the memory
2087 * location computed by the data_of() macro.
2088 *
2089 * The function returns 0 to indicate the end of the data stream condition,
2090 * and negative numbers are returned on errors.  If that happens, the structure
2091 * pointed to by @handle is not updated and should not be used any more.
2092 */
2093int snapshot_read_next(struct snapshot_handle *handle)
2094{
2095	if (handle->cur > nr_meta_pages + nr_copy_pages)
2096		return 0;
2097
2098	if (!buffer) {
2099		/* This makes the buffer be freed by swsusp_free() */
2100		buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2101		if (!buffer)
2102			return -ENOMEM;
2103	}
2104	if (!handle->cur) {
2105		int error;
2106
2107		error = init_header((struct swsusp_info *)buffer);
2108		if (error)
2109			return error;
2110		handle->buffer = buffer;
2111		memory_bm_position_reset(&orig_bm);
2112		memory_bm_position_reset(&copy_bm);
2113	} else if (handle->cur <= nr_meta_pages) {
2114		clear_page(buffer);
2115		pack_pfns(buffer, &orig_bm);
2116	} else {
2117		struct page *page;
2118
2119		page = pfn_to_page(memory_bm_next_pfn(&copy_bm));
2120		if (PageHighMem(page)) {
2121			/*
2122			 * Highmem pages are copied to the buffer,
2123			 * because we can't return with a kmapped
2124			 * highmem page (we may not be called again).
2125			 */
2126			void *kaddr;
2127
2128			kaddr = kmap_atomic(page);
2129			copy_page(buffer, kaddr);
2130			kunmap_atomic(kaddr);
2131			handle->buffer = buffer;
2132		} else {
2133			handle->buffer = page_address(page);
2134		}
2135	}
2136	handle->cur++;
2137	return PAGE_SIZE;
2138}
2139
2140static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2141				    struct memory_bitmap *src)
2142{
2143	unsigned long pfn;
2144
2145	memory_bm_position_reset(src);
2146	pfn = memory_bm_next_pfn(src);
2147	while (pfn != BM_END_OF_MAP) {
2148		memory_bm_set_bit(dst, pfn);
2149		pfn = memory_bm_next_pfn(src);
2150	}
2151}
2152
2153/**
2154 * mark_unsafe_pages - Mark pages that were used before hibernation.
2155 *
2156 * Mark the pages that cannot be used for storing the image during restoration,
2157 * because they conflict with the pages that had been used before hibernation.
2158 */
2159static void mark_unsafe_pages(struct memory_bitmap *bm)
2160{
2161	unsigned long pfn;
2162
2163	/* Clear the "free"/"unsafe" bit for all PFNs */
2164	memory_bm_position_reset(free_pages_map);
2165	pfn = memory_bm_next_pfn(free_pages_map);
2166	while (pfn != BM_END_OF_MAP) {
2167		memory_bm_clear_current(free_pages_map);
2168		pfn = memory_bm_next_pfn(free_pages_map);
2169	}
2170
2171	/* Mark pages that correspond to the "original" PFNs as "unsafe" */
2172	duplicate_memory_bitmap(free_pages_map, bm);
2173
2174	allocated_unsafe_pages = 0;
2175}
2176
2177static int check_header(struct swsusp_info *info)
2178{
2179	const char *reason;
2180
2181	reason = check_image_kernel(info);
2182	if (!reason && info->num_physpages != get_num_physpages())
2183		reason = "memory size";
2184	if (reason) {
2185		pr_err("Image mismatch: %s\n", reason);
2186		return -EPERM;
2187	}
2188	return 0;
2189}
2190
2191/**
2192 * load header - Check the image header and copy the data from it.
2193 */
2194static int load_header(struct swsusp_info *info)
2195{
2196	int error;
2197
2198	restore_pblist = NULL;
2199	error = check_header(info);
2200	if (!error) {
2201		nr_copy_pages = info->image_pages;
2202		nr_meta_pages = info->pages - info->image_pages - 1;
2203	}
2204	return error;
2205}
2206
2207/**
2208 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2209 * @bm: Memory bitmap.
2210 * @buf: Area of memory containing the PFNs.
2211 *
2212 * For each element of the array pointed to by @buf (1 page at a time), set the
2213 * corresponding bit in @bm.
2214 */
2215static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2216{
2217	int j;
2218
2219	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2220		if (unlikely(buf[j] == BM_END_OF_MAP))
2221			break;
2222
 
 
 
2223		if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2224			memory_bm_set_bit(bm, buf[j]);
2225		else
2226			return -EFAULT;
2227	}
2228
2229	return 0;
2230}
2231
2232#ifdef CONFIG_HIGHMEM
2233/*
2234 * struct highmem_pbe is used for creating the list of highmem pages that
2235 * should be restored atomically during the resume from disk, because the page
2236 * frames they have occupied before the suspend are in use.
2237 */
2238struct highmem_pbe {
2239	struct page *copy_page;	/* data is here now */
2240	struct page *orig_page;	/* data was here before the suspend */
2241	struct highmem_pbe *next;
2242};
2243
2244/*
2245 * List of highmem PBEs needed for restoring the highmem pages that were
2246 * allocated before the suspend and included in the suspend image, but have
2247 * also been allocated by the "resume" kernel, so their contents cannot be
2248 * written directly to their "original" page frames.
2249 */
2250static struct highmem_pbe *highmem_pblist;
2251
2252/**
2253 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2254 * @bm: Memory bitmap.
2255 *
2256 * The bits in @bm that correspond to image pages are assumed to be set.
2257 */
2258static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2259{
2260	unsigned long pfn;
2261	unsigned int cnt = 0;
2262
2263	memory_bm_position_reset(bm);
2264	pfn = memory_bm_next_pfn(bm);
2265	while (pfn != BM_END_OF_MAP) {
2266		if (PageHighMem(pfn_to_page(pfn)))
2267			cnt++;
2268
2269		pfn = memory_bm_next_pfn(bm);
2270	}
2271	return cnt;
2272}
2273
2274static unsigned int safe_highmem_pages;
2275
2276static struct memory_bitmap *safe_highmem_bm;
2277
2278/**
2279 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2280 * @bm: Pointer to an uninitialized memory bitmap structure.
2281 * @nr_highmem_p: Pointer to the number of highmem image pages.
2282 *
2283 * Try to allocate as many highmem pages as there are highmem image pages
2284 * (@nr_highmem_p points to the variable containing the number of highmem image
2285 * pages).  The pages that are "safe" (ie. will not be overwritten when the
2286 * hibernation image is restored entirely) have the corresponding bits set in
2287 * @bm (it must be unitialized).
2288 *
2289 * NOTE: This function should not be called if there are no highmem image pages.
2290 */
2291static int prepare_highmem_image(struct memory_bitmap *bm,
2292				 unsigned int *nr_highmem_p)
2293{
2294	unsigned int to_alloc;
2295
2296	if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2297		return -ENOMEM;
2298
2299	if (get_highmem_buffer(PG_SAFE))
2300		return -ENOMEM;
2301
2302	to_alloc = count_free_highmem_pages();
2303	if (to_alloc > *nr_highmem_p)
2304		to_alloc = *nr_highmem_p;
2305	else
2306		*nr_highmem_p = to_alloc;
2307
2308	safe_highmem_pages = 0;
2309	while (to_alloc-- > 0) {
2310		struct page *page;
2311
2312		page = alloc_page(__GFP_HIGHMEM);
2313		if (!swsusp_page_is_free(page)) {
2314			/* The page is "safe", set its bit the bitmap */
2315			memory_bm_set_bit(bm, page_to_pfn(page));
2316			safe_highmem_pages++;
2317		}
2318		/* Mark the page as allocated */
2319		swsusp_set_page_forbidden(page);
2320		swsusp_set_page_free(page);
2321	}
2322	memory_bm_position_reset(bm);
2323	safe_highmem_bm = bm;
2324	return 0;
2325}
2326
2327static struct page *last_highmem_page;
2328
2329/**
2330 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2331 *
2332 * For a given highmem image page get a buffer that suspend_write_next() should
2333 * return to its caller to write to.
2334 *
2335 * If the page is to be saved to its "original" page frame or a copy of
2336 * the page is to be made in the highmem, @buffer is returned.  Otherwise,
2337 * the copy of the page is to be made in normal memory, so the address of
2338 * the copy is returned.
2339 *
2340 * If @buffer is returned, the caller of suspend_write_next() will write
2341 * the page's contents to @buffer, so they will have to be copied to the
2342 * right location on the next call to suspend_write_next() and it is done
2343 * with the help of copy_last_highmem_page().  For this purpose, if
2344 * @buffer is returned, @last_highmem_page is set to the page to which
2345 * the data will have to be copied from @buffer.
2346 */
2347static void *get_highmem_page_buffer(struct page *page,
2348				     struct chain_allocator *ca)
2349{
2350	struct highmem_pbe *pbe;
2351	void *kaddr;
2352
2353	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2354		/*
2355		 * We have allocated the "original" page frame and we can
2356		 * use it directly to store the loaded page.
2357		 */
2358		last_highmem_page = page;
2359		return buffer;
2360	}
2361	/*
2362	 * The "original" page frame has not been allocated and we have to
2363	 * use a "safe" page frame to store the loaded page.
2364	 */
2365	pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2366	if (!pbe) {
2367		swsusp_free();
2368		return ERR_PTR(-ENOMEM);
2369	}
2370	pbe->orig_page = page;
2371	if (safe_highmem_pages > 0) {
2372		struct page *tmp;
2373
2374		/* Copy of the page will be stored in high memory */
2375		kaddr = buffer;
2376		tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2377		safe_highmem_pages--;
2378		last_highmem_page = tmp;
2379		pbe->copy_page = tmp;
2380	} else {
2381		/* Copy of the page will be stored in normal memory */
2382		kaddr = safe_pages_list;
2383		safe_pages_list = safe_pages_list->next;
2384		pbe->copy_page = virt_to_page(kaddr);
2385	}
2386	pbe->next = highmem_pblist;
2387	highmem_pblist = pbe;
2388	return kaddr;
2389}
2390
2391/**
2392 * copy_last_highmem_page - Copy most the most recent highmem image page.
2393 *
2394 * Copy the contents of a highmem image from @buffer, where the caller of
2395 * snapshot_write_next() has stored them, to the right location represented by
2396 * @last_highmem_page .
2397 */
2398static void copy_last_highmem_page(void)
2399{
2400	if (last_highmem_page) {
2401		void *dst;
2402
2403		dst = kmap_atomic(last_highmem_page);
2404		copy_page(dst, buffer);
2405		kunmap_atomic(dst);
2406		last_highmem_page = NULL;
2407	}
2408}
2409
2410static inline int last_highmem_page_copied(void)
2411{
2412	return !last_highmem_page;
2413}
2414
2415static inline void free_highmem_data(void)
2416{
2417	if (safe_highmem_bm)
2418		memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2419
2420	if (buffer)
2421		free_image_page(buffer, PG_UNSAFE_CLEAR);
2422}
2423#else
2424static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2425
2426static inline int prepare_highmem_image(struct memory_bitmap *bm,
2427					unsigned int *nr_highmem_p) { return 0; }
2428
2429static inline void *get_highmem_page_buffer(struct page *page,
2430					    struct chain_allocator *ca)
2431{
2432	return ERR_PTR(-EINVAL);
2433}
2434
2435static inline void copy_last_highmem_page(void) {}
2436static inline int last_highmem_page_copied(void) { return 1; }
2437static inline void free_highmem_data(void) {}
2438#endif /* CONFIG_HIGHMEM */
2439
2440#define PBES_PER_LINKED_PAGE	(LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2441
2442/**
2443 * prepare_image - Make room for loading hibernation image.
2444 * @new_bm: Unitialized memory bitmap structure.
2445 * @bm: Memory bitmap with unsafe pages marked.
2446 *
2447 * Use @bm to mark the pages that will be overwritten in the process of
2448 * restoring the system memory state from the suspend image ("unsafe" pages)
2449 * and allocate memory for the image.
2450 *
2451 * The idea is to allocate a new memory bitmap first and then allocate
2452 * as many pages as needed for image data, but without specifying what those
2453 * pages will be used for just yet.  Instead, we mark them all as allocated and
2454 * create a lists of "safe" pages to be used later.  On systems with high
2455 * memory a list of "safe" highmem pages is created too.
2456 */
2457static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2458{
2459	unsigned int nr_pages, nr_highmem;
2460	struct linked_page *lp;
2461	int error;
2462
2463	/* If there is no highmem, the buffer will not be necessary */
2464	free_image_page(buffer, PG_UNSAFE_CLEAR);
2465	buffer = NULL;
2466
2467	nr_highmem = count_highmem_image_pages(bm);
2468	mark_unsafe_pages(bm);
2469
2470	error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2471	if (error)
2472		goto Free;
2473
2474	duplicate_memory_bitmap(new_bm, bm);
2475	memory_bm_free(bm, PG_UNSAFE_KEEP);
2476	if (nr_highmem > 0) {
2477		error = prepare_highmem_image(bm, &nr_highmem);
2478		if (error)
2479			goto Free;
2480	}
2481	/*
2482	 * Reserve some safe pages for potential later use.
2483	 *
2484	 * NOTE: This way we make sure there will be enough safe pages for the
2485	 * chain_alloc() in get_buffer().  It is a bit wasteful, but
2486	 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2487	 *
2488	 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2489	 */
2490	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2491	nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2492	while (nr_pages > 0) {
2493		lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2494		if (!lp) {
2495			error = -ENOMEM;
2496			goto Free;
2497		}
2498		lp->next = safe_pages_list;
2499		safe_pages_list = lp;
2500		nr_pages--;
2501	}
2502	/* Preallocate memory for the image */
2503	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2504	while (nr_pages > 0) {
2505		lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2506		if (!lp) {
2507			error = -ENOMEM;
2508			goto Free;
2509		}
2510		if (!swsusp_page_is_free(virt_to_page(lp))) {
2511			/* The page is "safe", add it to the list */
2512			lp->next = safe_pages_list;
2513			safe_pages_list = lp;
2514		}
2515		/* Mark the page as allocated */
2516		swsusp_set_page_forbidden(virt_to_page(lp));
2517		swsusp_set_page_free(virt_to_page(lp));
2518		nr_pages--;
2519	}
2520	return 0;
2521
2522 Free:
2523	swsusp_free();
2524	return error;
2525}
2526
2527/**
2528 * get_buffer - Get the address to store the next image data page.
2529 *
2530 * Get the address that snapshot_write_next() should return to its caller to
2531 * write to.
2532 */
2533static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2534{
2535	struct pbe *pbe;
2536	struct page *page;
2537	unsigned long pfn = memory_bm_next_pfn(bm);
2538
2539	if (pfn == BM_END_OF_MAP)
2540		return ERR_PTR(-EFAULT);
2541
2542	page = pfn_to_page(pfn);
2543	if (PageHighMem(page))
2544		return get_highmem_page_buffer(page, ca);
2545
2546	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2547		/*
2548		 * We have allocated the "original" page frame and we can
2549		 * use it directly to store the loaded page.
2550		 */
2551		return page_address(page);
2552
2553	/*
2554	 * The "original" page frame has not been allocated and we have to
2555	 * use a "safe" page frame to store the loaded page.
2556	 */
2557	pbe = chain_alloc(ca, sizeof(struct pbe));
2558	if (!pbe) {
2559		swsusp_free();
2560		return ERR_PTR(-ENOMEM);
2561	}
2562	pbe->orig_address = page_address(page);
2563	pbe->address = safe_pages_list;
2564	safe_pages_list = safe_pages_list->next;
2565	pbe->next = restore_pblist;
2566	restore_pblist = pbe;
2567	return pbe->address;
2568}
2569
2570/**
2571 * snapshot_write_next - Get the address to store the next image page.
2572 * @handle: Snapshot handle structure to guide the writing.
2573 *
2574 * On the first call, @handle should point to a zeroed snapshot_handle
2575 * structure.  The structure gets populated then and a pointer to it should be
2576 * passed to this function every next time.
2577 *
2578 * On success, the function returns a positive number.  Then, the caller
2579 * is allowed to write up to the returned number of bytes to the memory
2580 * location computed by the data_of() macro.
2581 *
2582 * The function returns 0 to indicate the "end of file" condition.  Negative
2583 * numbers are returned on errors, in which cases the structure pointed to by
2584 * @handle is not updated and should not be used any more.
2585 */
2586int snapshot_write_next(struct snapshot_handle *handle)
2587{
2588	static struct chain_allocator ca;
2589	int error = 0;
2590
2591	/* Check if we have already loaded the entire image */
2592	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2593		return 0;
2594
2595	handle->sync_read = 1;
2596
2597	if (!handle->cur) {
2598		if (!buffer)
2599			/* This makes the buffer be freed by swsusp_free() */
2600			buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2601
2602		if (!buffer)
2603			return -ENOMEM;
2604
2605		handle->buffer = buffer;
2606	} else if (handle->cur == 1) {
2607		error = load_header(buffer);
2608		if (error)
2609			return error;
2610
2611		safe_pages_list = NULL;
2612
2613		error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
2614		if (error)
2615			return error;
2616
 
 
 
 
 
2617		hibernate_restore_protection_begin();
2618	} else if (handle->cur <= nr_meta_pages + 1) {
2619		error = unpack_orig_pfns(buffer, &copy_bm);
2620		if (error)
2621			return error;
2622
2623		if (handle->cur == nr_meta_pages + 1) {
2624			error = prepare_image(&orig_bm, &copy_bm);
2625			if (error)
2626				return error;
2627
2628			chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2629			memory_bm_position_reset(&orig_bm);
2630			restore_pblist = NULL;
2631			handle->buffer = get_buffer(&orig_bm, &ca);
2632			handle->sync_read = 0;
2633			if (IS_ERR(handle->buffer))
2634				return PTR_ERR(handle->buffer);
2635		}
2636	} else {
2637		copy_last_highmem_page();
 
 
2638		hibernate_restore_protect_page(handle->buffer);
2639		handle->buffer = get_buffer(&orig_bm, &ca);
2640		if (IS_ERR(handle->buffer))
2641			return PTR_ERR(handle->buffer);
2642		if (handle->buffer != buffer)
2643			handle->sync_read = 0;
2644	}
2645	handle->cur++;
2646	return PAGE_SIZE;
2647}
2648
2649/**
2650 * snapshot_write_finalize - Complete the loading of a hibernation image.
2651 *
2652 * Must be called after the last call to snapshot_write_next() in case the last
2653 * page in the image happens to be a highmem page and its contents should be
2654 * stored in highmem.  Additionally, it recycles bitmap memory that's not
2655 * necessary any more.
2656 */
2657void snapshot_write_finalize(struct snapshot_handle *handle)
2658{
2659	copy_last_highmem_page();
 
 
 
2660	hibernate_restore_protect_page(handle->buffer);
2661	/* Do that only if we have loaded the image entirely */
2662	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2663		memory_bm_recycle(&orig_bm);
2664		free_highmem_data();
2665	}
2666}
2667
2668int snapshot_image_loaded(struct snapshot_handle *handle)
2669{
2670	return !(!nr_copy_pages || !last_highmem_page_copied() ||
2671			handle->cur <= nr_meta_pages + nr_copy_pages);
2672}
2673
2674#ifdef CONFIG_HIGHMEM
2675/* Assumes that @buf is ready and points to a "safe" page */
2676static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2677				       void *buf)
2678{
2679	void *kaddr1, *kaddr2;
2680
2681	kaddr1 = kmap_atomic(p1);
2682	kaddr2 = kmap_atomic(p2);
2683	copy_page(buf, kaddr1);
2684	copy_page(kaddr1, kaddr2);
2685	copy_page(kaddr2, buf);
2686	kunmap_atomic(kaddr2);
2687	kunmap_atomic(kaddr1);
2688}
2689
2690/**
2691 * restore_highmem - Put highmem image pages into their original locations.
2692 *
2693 * For each highmem page that was in use before hibernation and is included in
2694 * the image, and also has been allocated by the "restore" kernel, swap its
2695 * current contents with the previous (ie. "before hibernation") ones.
2696 *
2697 * If the restore eventually fails, we can call this function once again and
2698 * restore the highmem state as seen by the restore kernel.
2699 */
2700int restore_highmem(void)
2701{
2702	struct highmem_pbe *pbe = highmem_pblist;
2703	void *buf;
2704
2705	if (!pbe)
2706		return 0;
2707
2708	buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2709	if (!buf)
2710		return -ENOMEM;
2711
2712	while (pbe) {
2713		swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2714		pbe = pbe->next;
2715	}
2716	free_image_page(buf, PG_UNSAFE_CLEAR);
2717	return 0;
2718}
2719#endif /* CONFIG_HIGHMEM */
v4.10.11
 
   1/*
   2 * linux/kernel/power/snapshot.c
   3 *
   4 * This file provides system snapshot/restore functionality for swsusp.
   5 *
   6 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
   7 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
   8 *
   9 * This file is released under the GPLv2.
  10 *
  11 */
  12
 
 
  13#include <linux/version.h>
  14#include <linux/module.h>
  15#include <linux/mm.h>
  16#include <linux/suspend.h>
  17#include <linux/delay.h>
  18#include <linux/bitops.h>
  19#include <linux/spinlock.h>
  20#include <linux/kernel.h>
  21#include <linux/pm.h>
  22#include <linux/device.h>
  23#include <linux/init.h>
  24#include <linux/bootmem.h>
 
  25#include <linux/syscalls.h>
  26#include <linux/console.h>
  27#include <linux/highmem.h>
  28#include <linux/list.h>
  29#include <linux/slab.h>
  30#include <linux/compiler.h>
  31#include <linux/ktime.h>
 
  32
  33#include <linux/uaccess.h>
  34#include <asm/mmu_context.h>
  35#include <asm/pgtable.h>
  36#include <asm/tlbflush.h>
  37#include <asm/io.h>
  38
  39#include "power.h"
  40
  41#ifdef CONFIG_DEBUG_RODATA
  42static bool hibernate_restore_protection;
  43static bool hibernate_restore_protection_active;
  44
  45void enable_restore_image_protection(void)
  46{
  47	hibernate_restore_protection = true;
  48}
  49
  50static inline void hibernate_restore_protection_begin(void)
  51{
  52	hibernate_restore_protection_active = hibernate_restore_protection;
  53}
  54
  55static inline void hibernate_restore_protection_end(void)
  56{
  57	hibernate_restore_protection_active = false;
  58}
  59
  60static inline void hibernate_restore_protect_page(void *page_address)
  61{
  62	if (hibernate_restore_protection_active)
  63		set_memory_ro((unsigned long)page_address, 1);
  64}
  65
  66static inline void hibernate_restore_unprotect_page(void *page_address)
  67{
  68	if (hibernate_restore_protection_active)
  69		set_memory_rw((unsigned long)page_address, 1);
  70}
  71#else
  72static inline void hibernate_restore_protection_begin(void) {}
  73static inline void hibernate_restore_protection_end(void) {}
  74static inline void hibernate_restore_protect_page(void *page_address) {}
  75static inline void hibernate_restore_unprotect_page(void *page_address) {}
  76#endif /* CONFIG_DEBUG_RODATA */
  77
  78static int swsusp_page_is_free(struct page *);
  79static void swsusp_set_page_forbidden(struct page *);
  80static void swsusp_unset_page_forbidden(struct page *);
  81
  82/*
  83 * Number of bytes to reserve for memory allocations made by device drivers
  84 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
  85 * cause image creation to fail (tunable via /sys/power/reserved_size).
  86 */
  87unsigned long reserved_size;
  88
  89void __init hibernate_reserved_size_init(void)
  90{
  91	reserved_size = SPARE_PAGES * PAGE_SIZE;
  92}
  93
  94/*
  95 * Preferred image size in bytes (tunable via /sys/power/image_size).
  96 * When it is set to N, swsusp will do its best to ensure the image
  97 * size will not exceed N bytes, but if that is impossible, it will
  98 * try to create the smallest image possible.
  99 */
 100unsigned long image_size;
 101
 102void __init hibernate_image_size_init(void)
 103{
 104	image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE;
 105}
 106
 107/*
 108 * List of PBEs needed for restoring the pages that were allocated before
 109 * the suspend and included in the suspend image, but have also been
 110 * allocated by the "resume" kernel, so their contents cannot be written
 111 * directly to their "original" page frames.
 112 */
 113struct pbe *restore_pblist;
 114
 115/* struct linked_page is used to build chains of pages */
 116
 117#define LINKED_PAGE_DATA_SIZE	(PAGE_SIZE - sizeof(void *))
 118
 119struct linked_page {
 120	struct linked_page *next;
 121	char data[LINKED_PAGE_DATA_SIZE];
 122} __packed;
 123
 124/*
 125 * List of "safe" pages (ie. pages that were not used by the image kernel
 126 * before hibernation) that may be used as temporary storage for image kernel
 127 * memory contents.
 128 */
 129static struct linked_page *safe_pages_list;
 130
 131/* Pointer to an auxiliary buffer (1 page) */
 132static void *buffer;
 133
 134#define PG_ANY		0
 135#define PG_SAFE		1
 136#define PG_UNSAFE_CLEAR	1
 137#define PG_UNSAFE_KEEP	0
 138
 139static unsigned int allocated_unsafe_pages;
 140
 141/**
 142 * get_image_page - Allocate a page for a hibernation image.
 143 * @gfp_mask: GFP mask for the allocation.
 144 * @safe_needed: Get pages that were not used before hibernation (restore only)
 145 *
 146 * During image restoration, for storing the PBE list and the image data, we can
 147 * only use memory pages that do not conflict with the pages used before
 148 * hibernation.  The "unsafe" pages have PageNosaveFree set and we count them
 149 * using allocated_unsafe_pages.
 150 *
 151 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
 152 * swsusp_free() can release it.
 153 */
 154static void *get_image_page(gfp_t gfp_mask, int safe_needed)
 155{
 156	void *res;
 157
 158	res = (void *)get_zeroed_page(gfp_mask);
 159	if (safe_needed)
 160		while (res && swsusp_page_is_free(virt_to_page(res))) {
 161			/* The page is unsafe, mark it for swsusp_free() */
 162			swsusp_set_page_forbidden(virt_to_page(res));
 163			allocated_unsafe_pages++;
 164			res = (void *)get_zeroed_page(gfp_mask);
 165		}
 166	if (res) {
 167		swsusp_set_page_forbidden(virt_to_page(res));
 168		swsusp_set_page_free(virt_to_page(res));
 169	}
 170	return res;
 171}
 172
 173static void *__get_safe_page(gfp_t gfp_mask)
 174{
 175	if (safe_pages_list) {
 176		void *ret = safe_pages_list;
 177
 178		safe_pages_list = safe_pages_list->next;
 179		memset(ret, 0, PAGE_SIZE);
 180		return ret;
 181	}
 182	return get_image_page(gfp_mask, PG_SAFE);
 183}
 184
 185unsigned long get_safe_page(gfp_t gfp_mask)
 186{
 187	return (unsigned long)__get_safe_page(gfp_mask);
 188}
 189
 190static struct page *alloc_image_page(gfp_t gfp_mask)
 191{
 192	struct page *page;
 193
 194	page = alloc_page(gfp_mask);
 195	if (page) {
 196		swsusp_set_page_forbidden(page);
 197		swsusp_set_page_free(page);
 198	}
 199	return page;
 200}
 201
 202static void recycle_safe_page(void *page_address)
 203{
 204	struct linked_page *lp = page_address;
 205
 206	lp->next = safe_pages_list;
 207	safe_pages_list = lp;
 208}
 209
 210/**
 211 * free_image_page - Free a page allocated for hibernation image.
 212 * @addr: Address of the page to free.
 213 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
 214 *
 215 * The page to free should have been allocated by get_image_page() (page flags
 216 * set by it are affected).
 217 */
 218static inline void free_image_page(void *addr, int clear_nosave_free)
 219{
 220	struct page *page;
 221
 222	BUG_ON(!virt_addr_valid(addr));
 223
 224	page = virt_to_page(addr);
 225
 226	swsusp_unset_page_forbidden(page);
 227	if (clear_nosave_free)
 228		swsusp_unset_page_free(page);
 229
 230	__free_page(page);
 231}
 232
 233static inline void free_list_of_pages(struct linked_page *list,
 234				      int clear_page_nosave)
 235{
 236	while (list) {
 237		struct linked_page *lp = list->next;
 238
 239		free_image_page(list, clear_page_nosave);
 240		list = lp;
 241	}
 242}
 243
 244/*
 245 * struct chain_allocator is used for allocating small objects out of
 246 * a linked list of pages called 'the chain'.
 247 *
 248 * The chain grows each time when there is no room for a new object in
 249 * the current page.  The allocated objects cannot be freed individually.
 250 * It is only possible to free them all at once, by freeing the entire
 251 * chain.
 252 *
 253 * NOTE: The chain allocator may be inefficient if the allocated objects
 254 * are not much smaller than PAGE_SIZE.
 255 */
 256struct chain_allocator {
 257	struct linked_page *chain;	/* the chain */
 258	unsigned int used_space;	/* total size of objects allocated out
 259					   of the current page */
 260	gfp_t gfp_mask;		/* mask for allocating pages */
 261	int safe_needed;	/* if set, only "safe" pages are allocated */
 262};
 263
 264static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
 265		       int safe_needed)
 266{
 267	ca->chain = NULL;
 268	ca->used_space = LINKED_PAGE_DATA_SIZE;
 269	ca->gfp_mask = gfp_mask;
 270	ca->safe_needed = safe_needed;
 271}
 272
 273static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
 274{
 275	void *ret;
 276
 277	if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
 278		struct linked_page *lp;
 279
 280		lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
 281					get_image_page(ca->gfp_mask, PG_ANY);
 282		if (!lp)
 283			return NULL;
 284
 285		lp->next = ca->chain;
 286		ca->chain = lp;
 287		ca->used_space = 0;
 288	}
 289	ret = ca->chain->data + ca->used_space;
 290	ca->used_space += size;
 291	return ret;
 292}
 293
 294/**
 295 * Data types related to memory bitmaps.
 296 *
 297 * Memory bitmap is a structure consiting of many linked lists of
 298 * objects.  The main list's elements are of type struct zone_bitmap
 299 * and each of them corresonds to one zone.  For each zone bitmap
 300 * object there is a list of objects of type struct bm_block that
 301 * represent each blocks of bitmap in which information is stored.
 302 *
 303 * struct memory_bitmap contains a pointer to the main list of zone
 304 * bitmap objects, a struct bm_position used for browsing the bitmap,
 305 * and a pointer to the list of pages used for allocating all of the
 306 * zone bitmap objects and bitmap block objects.
 307 *
 308 * NOTE: It has to be possible to lay out the bitmap in memory
 309 * using only allocations of order 0.  Additionally, the bitmap is
 310 * designed to work with arbitrary number of zones (this is over the
 311 * top for now, but let's avoid making unnecessary assumptions ;-).
 312 *
 313 * struct zone_bitmap contains a pointer to a list of bitmap block
 314 * objects and a pointer to the bitmap block object that has been
 315 * most recently used for setting bits.  Additionally, it contains the
 316 * PFNs that correspond to the start and end of the represented zone.
 317 *
 318 * struct bm_block contains a pointer to the memory page in which
 319 * information is stored (in the form of a block of bitmap)
 320 * It also contains the pfns that correspond to the start and end of
 321 * the represented memory area.
 322 *
 323 * The memory bitmap is organized as a radix tree to guarantee fast random
 324 * access to the bits. There is one radix tree for each zone (as returned
 325 * from create_mem_extents).
 326 *
 327 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
 328 * two linked lists for the nodes of the tree, one for the inner nodes and
 329 * one for the leave nodes. The linked leave nodes are used for fast linear
 330 * access of the memory bitmap.
 331 *
 332 * The struct rtree_node represents one node of the radix tree.
 333 */
 334
 335#define BM_END_OF_MAP	(~0UL)
 336
 337#define BM_BITS_PER_BLOCK	(PAGE_SIZE * BITS_PER_BYTE)
 338#define BM_BLOCK_SHIFT		(PAGE_SHIFT + 3)
 339#define BM_BLOCK_MASK		((1UL << BM_BLOCK_SHIFT) - 1)
 340
 341/*
 342 * struct rtree_node is a wrapper struct to link the nodes
 343 * of the rtree together for easy linear iteration over
 344 * bits and easy freeing
 345 */
 346struct rtree_node {
 347	struct list_head list;
 348	unsigned long *data;
 349};
 350
 351/*
 352 * struct mem_zone_bm_rtree represents a bitmap used for one
 353 * populated memory zone.
 354 */
 355struct mem_zone_bm_rtree {
 356	struct list_head list;		/* Link Zones together         */
 357	struct list_head nodes;		/* Radix Tree inner nodes      */
 358	struct list_head leaves;	/* Radix Tree leaves           */
 359	unsigned long start_pfn;	/* Zone start page frame       */
 360	unsigned long end_pfn;		/* Zone end page frame + 1     */
 361	struct rtree_node *rtree;	/* Radix Tree Root             */
 362	int levels;			/* Number of Radix Tree Levels */
 363	unsigned int blocks;		/* Number of Bitmap Blocks     */
 364};
 365
 366/* strcut bm_position is used for browsing memory bitmaps */
 367
 368struct bm_position {
 369	struct mem_zone_bm_rtree *zone;
 370	struct rtree_node *node;
 371	unsigned long node_pfn;
 372	int node_bit;
 373};
 374
 375struct memory_bitmap {
 376	struct list_head zones;
 377	struct linked_page *p_list;	/* list of pages used to store zone
 378					   bitmap objects and bitmap block
 379					   objects */
 380	struct bm_position cur;	/* most recently used bit position */
 381};
 382
 383/* Functions that operate on memory bitmaps */
 384
 385#define BM_ENTRIES_PER_LEVEL	(PAGE_SIZE / sizeof(unsigned long))
 386#if BITS_PER_LONG == 32
 387#define BM_RTREE_LEVEL_SHIFT	(PAGE_SHIFT - 2)
 388#else
 389#define BM_RTREE_LEVEL_SHIFT	(PAGE_SHIFT - 3)
 390#endif
 391#define BM_RTREE_LEVEL_MASK	((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
 392
 393/**
 394 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
 395 *
 396 * This function is used to allocate inner nodes as well as the
 397 * leave nodes of the radix tree. It also adds the node to the
 398 * corresponding linked list passed in by the *list parameter.
 399 */
 400static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
 401					   struct chain_allocator *ca,
 402					   struct list_head *list)
 403{
 404	struct rtree_node *node;
 405
 406	node = chain_alloc(ca, sizeof(struct rtree_node));
 407	if (!node)
 408		return NULL;
 409
 410	node->data = get_image_page(gfp_mask, safe_needed);
 411	if (!node->data)
 412		return NULL;
 413
 414	list_add_tail(&node->list, list);
 415
 416	return node;
 417}
 418
 419/**
 420 * add_rtree_block - Add a new leave node to the radix tree.
 421 *
 422 * The leave nodes need to be allocated in order to keep the leaves
 423 * linked list in order. This is guaranteed by the zone->blocks
 424 * counter.
 425 */
 426static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
 427			   int safe_needed, struct chain_allocator *ca)
 428{
 429	struct rtree_node *node, *block, **dst;
 430	unsigned int levels_needed, block_nr;
 431	int i;
 432
 433	block_nr = zone->blocks;
 434	levels_needed = 0;
 435
 436	/* How many levels do we need for this block nr? */
 437	while (block_nr) {
 438		levels_needed += 1;
 439		block_nr >>= BM_RTREE_LEVEL_SHIFT;
 440	}
 441
 442	/* Make sure the rtree has enough levels */
 443	for (i = zone->levels; i < levels_needed; i++) {
 444		node = alloc_rtree_node(gfp_mask, safe_needed, ca,
 445					&zone->nodes);
 446		if (!node)
 447			return -ENOMEM;
 448
 449		node->data[0] = (unsigned long)zone->rtree;
 450		zone->rtree = node;
 451		zone->levels += 1;
 452	}
 453
 454	/* Allocate new block */
 455	block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
 456	if (!block)
 457		return -ENOMEM;
 458
 459	/* Now walk the rtree to insert the block */
 460	node = zone->rtree;
 461	dst = &zone->rtree;
 462	block_nr = zone->blocks;
 463	for (i = zone->levels; i > 0; i--) {
 464		int index;
 465
 466		if (!node) {
 467			node = alloc_rtree_node(gfp_mask, safe_needed, ca,
 468						&zone->nodes);
 469			if (!node)
 470				return -ENOMEM;
 471			*dst = node;
 472		}
 473
 474		index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
 475		index &= BM_RTREE_LEVEL_MASK;
 476		dst = (struct rtree_node **)&((*dst)->data[index]);
 477		node = *dst;
 478	}
 479
 480	zone->blocks += 1;
 481	*dst = block;
 482
 483	return 0;
 484}
 485
 486static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
 487			       int clear_nosave_free);
 488
 489/**
 490 * create_zone_bm_rtree - Create a radix tree for one zone.
 491 *
 492 * Allocated the mem_zone_bm_rtree structure and initializes it.
 493 * This function also allocated and builds the radix tree for the
 494 * zone.
 495 */
 496static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
 497						      int safe_needed,
 498						      struct chain_allocator *ca,
 499						      unsigned long start,
 500						      unsigned long end)
 501{
 502	struct mem_zone_bm_rtree *zone;
 503	unsigned int i, nr_blocks;
 504	unsigned long pages;
 505
 506	pages = end - start;
 507	zone  = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
 508	if (!zone)
 509		return NULL;
 510
 511	INIT_LIST_HEAD(&zone->nodes);
 512	INIT_LIST_HEAD(&zone->leaves);
 513	zone->start_pfn = start;
 514	zone->end_pfn = end;
 515	nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
 516
 517	for (i = 0; i < nr_blocks; i++) {
 518		if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
 519			free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
 520			return NULL;
 521		}
 522	}
 523
 524	return zone;
 525}
 526
 527/**
 528 * free_zone_bm_rtree - Free the memory of the radix tree.
 529 *
 530 * Free all node pages of the radix tree. The mem_zone_bm_rtree
 531 * structure itself is not freed here nor are the rtree_node
 532 * structs.
 533 */
 534static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
 535			       int clear_nosave_free)
 536{
 537	struct rtree_node *node;
 538
 539	list_for_each_entry(node, &zone->nodes, list)
 540		free_image_page(node->data, clear_nosave_free);
 541
 542	list_for_each_entry(node, &zone->leaves, list)
 543		free_image_page(node->data, clear_nosave_free);
 544}
 545
 546static void memory_bm_position_reset(struct memory_bitmap *bm)
 547{
 548	bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
 549				  list);
 550	bm->cur.node = list_entry(bm->cur.zone->leaves.next,
 551				  struct rtree_node, list);
 552	bm->cur.node_pfn = 0;
 553	bm->cur.node_bit = 0;
 554}
 555
 556static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
 557
 558struct mem_extent {
 559	struct list_head hook;
 560	unsigned long start;
 561	unsigned long end;
 562};
 563
 564/**
 565 * free_mem_extents - Free a list of memory extents.
 566 * @list: List of extents to free.
 567 */
 568static void free_mem_extents(struct list_head *list)
 569{
 570	struct mem_extent *ext, *aux;
 571
 572	list_for_each_entry_safe(ext, aux, list, hook) {
 573		list_del(&ext->hook);
 574		kfree(ext);
 575	}
 576}
 577
 578/**
 579 * create_mem_extents - Create a list of memory extents.
 580 * @list: List to put the extents into.
 581 * @gfp_mask: Mask to use for memory allocations.
 582 *
 583 * The extents represent contiguous ranges of PFNs.
 584 */
 585static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
 586{
 587	struct zone *zone;
 588
 589	INIT_LIST_HEAD(list);
 590
 591	for_each_populated_zone(zone) {
 592		unsigned long zone_start, zone_end;
 593		struct mem_extent *ext, *cur, *aux;
 594
 595		zone_start = zone->zone_start_pfn;
 596		zone_end = zone_end_pfn(zone);
 597
 598		list_for_each_entry(ext, list, hook)
 599			if (zone_start <= ext->end)
 600				break;
 601
 602		if (&ext->hook == list || zone_end < ext->start) {
 603			/* New extent is necessary */
 604			struct mem_extent *new_ext;
 605
 606			new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
 607			if (!new_ext) {
 608				free_mem_extents(list);
 609				return -ENOMEM;
 610			}
 611			new_ext->start = zone_start;
 612			new_ext->end = zone_end;
 613			list_add_tail(&new_ext->hook, &ext->hook);
 614			continue;
 615		}
 616
 617		/* Merge this zone's range of PFNs with the existing one */
 618		if (zone_start < ext->start)
 619			ext->start = zone_start;
 620		if (zone_end > ext->end)
 621			ext->end = zone_end;
 622
 623		/* More merging may be possible */
 624		cur = ext;
 625		list_for_each_entry_safe_continue(cur, aux, list, hook) {
 626			if (zone_end < cur->start)
 627				break;
 628			if (zone_end < cur->end)
 629				ext->end = cur->end;
 630			list_del(&cur->hook);
 631			kfree(cur);
 632		}
 633	}
 634
 635	return 0;
 636}
 637
 638/**
 639 * memory_bm_create - Allocate memory for a memory bitmap.
 640 */
 641static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
 642			    int safe_needed)
 643{
 644	struct chain_allocator ca;
 645	struct list_head mem_extents;
 646	struct mem_extent *ext;
 647	int error;
 648
 649	chain_init(&ca, gfp_mask, safe_needed);
 650	INIT_LIST_HEAD(&bm->zones);
 651
 652	error = create_mem_extents(&mem_extents, gfp_mask);
 653	if (error)
 654		return error;
 655
 656	list_for_each_entry(ext, &mem_extents, hook) {
 657		struct mem_zone_bm_rtree *zone;
 658
 659		zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
 660					    ext->start, ext->end);
 661		if (!zone) {
 662			error = -ENOMEM;
 663			goto Error;
 664		}
 665		list_add_tail(&zone->list, &bm->zones);
 666	}
 667
 668	bm->p_list = ca.chain;
 669	memory_bm_position_reset(bm);
 670 Exit:
 671	free_mem_extents(&mem_extents);
 672	return error;
 673
 674 Error:
 675	bm->p_list = ca.chain;
 676	memory_bm_free(bm, PG_UNSAFE_CLEAR);
 677	goto Exit;
 678}
 679
 680/**
 681 * memory_bm_free - Free memory occupied by the memory bitmap.
 682 * @bm: Memory bitmap.
 683 */
 684static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
 685{
 686	struct mem_zone_bm_rtree *zone;
 687
 688	list_for_each_entry(zone, &bm->zones, list)
 689		free_zone_bm_rtree(zone, clear_nosave_free);
 690
 691	free_list_of_pages(bm->p_list, clear_nosave_free);
 692
 693	INIT_LIST_HEAD(&bm->zones);
 694}
 695
 696/**
 697 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
 698 *
 699 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
 700 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
 701 *
 702 * Walk the radix tree to find the page containing the bit that represents @pfn
 703 * and return the position of the bit in @addr and @bit_nr.
 704 */
 705static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
 706			      void **addr, unsigned int *bit_nr)
 707{
 708	struct mem_zone_bm_rtree *curr, *zone;
 709	struct rtree_node *node;
 710	int i, block_nr;
 711
 712	zone = bm->cur.zone;
 713
 714	if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
 715		goto zone_found;
 716
 717	zone = NULL;
 718
 719	/* Find the right zone */
 720	list_for_each_entry(curr, &bm->zones, list) {
 721		if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
 722			zone = curr;
 723			break;
 724		}
 725	}
 726
 727	if (!zone)
 728		return -EFAULT;
 729
 730zone_found:
 731	/*
 732	 * We have found the zone. Now walk the radix tree to find the leaf node
 733	 * for our PFN.
 734	 */
 
 
 
 
 
 
 735	node = bm->cur.node;
 736	if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
 
 737		goto node_found;
 738
 739	node      = zone->rtree;
 740	block_nr  = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
 741
 742	for (i = zone->levels; i > 0; i--) {
 743		int index;
 744
 745		index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
 746		index &= BM_RTREE_LEVEL_MASK;
 747		BUG_ON(node->data[index] == 0);
 748		node = (struct rtree_node *)node->data[index];
 749	}
 750
 751node_found:
 752	/* Update last position */
 753	bm->cur.zone = zone;
 754	bm->cur.node = node;
 755	bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
 756
 757	/* Set return values */
 758	*addr = node->data;
 759	*bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
 760
 761	return 0;
 762}
 763
 764static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
 765{
 766	void *addr;
 767	unsigned int bit;
 768	int error;
 769
 770	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 771	BUG_ON(error);
 772	set_bit(bit, addr);
 773}
 774
 775static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
 776{
 777	void *addr;
 778	unsigned int bit;
 779	int error;
 780
 781	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 782	if (!error)
 783		set_bit(bit, addr);
 784
 785	return error;
 786}
 787
 788static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
 789{
 790	void *addr;
 791	unsigned int bit;
 792	int error;
 793
 794	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 795	BUG_ON(error);
 796	clear_bit(bit, addr);
 797}
 798
 799static void memory_bm_clear_current(struct memory_bitmap *bm)
 800{
 801	int bit;
 802
 803	bit = max(bm->cur.node_bit - 1, 0);
 804	clear_bit(bit, bm->cur.node->data);
 805}
 806
 807static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
 808{
 809	void *addr;
 810	unsigned int bit;
 811	int error;
 812
 813	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 814	BUG_ON(error);
 815	return test_bit(bit, addr);
 816}
 817
 818static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
 819{
 820	void *addr;
 821	unsigned int bit;
 822
 823	return !memory_bm_find_bit(bm, pfn, &addr, &bit);
 824}
 825
 826/*
 827 * rtree_next_node - Jump to the next leaf node.
 828 *
 829 * Set the position to the beginning of the next node in the
 830 * memory bitmap. This is either the next node in the current
 831 * zone's radix tree or the first node in the radix tree of the
 832 * next zone.
 833 *
 834 * Return true if there is a next node, false otherwise.
 835 */
 836static bool rtree_next_node(struct memory_bitmap *bm)
 837{
 838	if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
 839		bm->cur.node = list_entry(bm->cur.node->list.next,
 840					  struct rtree_node, list);
 841		bm->cur.node_pfn += BM_BITS_PER_BLOCK;
 842		bm->cur.node_bit  = 0;
 843		touch_softlockup_watchdog();
 844		return true;
 845	}
 846
 847	/* No more nodes, goto next zone */
 848	if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
 849		bm->cur.zone = list_entry(bm->cur.zone->list.next,
 850				  struct mem_zone_bm_rtree, list);
 851		bm->cur.node = list_entry(bm->cur.zone->leaves.next,
 852					  struct rtree_node, list);
 853		bm->cur.node_pfn = 0;
 854		bm->cur.node_bit = 0;
 855		return true;
 856	}
 857
 858	/* No more zones */
 859	return false;
 860}
 861
 862/**
 863 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
 864 * @bm: Memory bitmap.
 865 *
 866 * Starting from the last returned position this function searches for the next
 867 * set bit in @bm and returns the PFN represented by it.  If no more bits are
 868 * set, BM_END_OF_MAP is returned.
 869 *
 870 * It is required to run memory_bm_position_reset() before the first call to
 871 * this function for the given memory bitmap.
 872 */
 873static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
 874{
 875	unsigned long bits, pfn, pages;
 876	int bit;
 877
 878	do {
 879		pages	  = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
 880		bits      = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
 881		bit	  = find_next_bit(bm->cur.node->data, bits,
 882					  bm->cur.node_bit);
 883		if (bit < bits) {
 884			pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
 885			bm->cur.node_bit = bit + 1;
 886			return pfn;
 887		}
 888	} while (rtree_next_node(bm));
 889
 890	return BM_END_OF_MAP;
 891}
 892
 893/*
 894 * This structure represents a range of page frames the contents of which
 895 * should not be saved during hibernation.
 896 */
 897struct nosave_region {
 898	struct list_head list;
 899	unsigned long start_pfn;
 900	unsigned long end_pfn;
 901};
 902
 903static LIST_HEAD(nosave_regions);
 904
 905static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
 906{
 907	struct rtree_node *node;
 908
 909	list_for_each_entry(node, &zone->nodes, list)
 910		recycle_safe_page(node->data);
 911
 912	list_for_each_entry(node, &zone->leaves, list)
 913		recycle_safe_page(node->data);
 914}
 915
 916static void memory_bm_recycle(struct memory_bitmap *bm)
 917{
 918	struct mem_zone_bm_rtree *zone;
 919	struct linked_page *p_list;
 920
 921	list_for_each_entry(zone, &bm->zones, list)
 922		recycle_zone_bm_rtree(zone);
 923
 924	p_list = bm->p_list;
 925	while (p_list) {
 926		struct linked_page *lp = p_list;
 927
 928		p_list = lp->next;
 929		recycle_safe_page(lp);
 930	}
 931}
 932
 933/**
 934 * register_nosave_region - Register a region of unsaveable memory.
 935 *
 936 * Register a range of page frames the contents of which should not be saved
 937 * during hibernation (to be used in the early initialization code).
 938 */
 939void __init __register_nosave_region(unsigned long start_pfn,
 940				     unsigned long end_pfn, int use_kmalloc)
 941{
 942	struct nosave_region *region;
 943
 944	if (start_pfn >= end_pfn)
 945		return;
 946
 947	if (!list_empty(&nosave_regions)) {
 948		/* Try to extend the previous region (they should be sorted) */
 949		region = list_entry(nosave_regions.prev,
 950					struct nosave_region, list);
 951		if (region->end_pfn == start_pfn) {
 952			region->end_pfn = end_pfn;
 953			goto Report;
 954		}
 955	}
 956	if (use_kmalloc) {
 957		/* During init, this shouldn't fail */
 958		region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
 959		BUG_ON(!region);
 960	} else {
 961		/* This allocation cannot fail */
 962		region = memblock_virt_alloc(sizeof(struct nosave_region), 0);
 
 
 
 
 963	}
 964	region->start_pfn = start_pfn;
 965	region->end_pfn = end_pfn;
 966	list_add_tail(&region->list, &nosave_regions);
 967 Report:
 968	printk(KERN_INFO "PM: Registered nosave memory: [mem %#010llx-%#010llx]\n",
 969		(unsigned long long) start_pfn << PAGE_SHIFT,
 970		((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
 971}
 972
 973/*
 974 * Set bits in this map correspond to the page frames the contents of which
 975 * should not be saved during the suspend.
 976 */
 977static struct memory_bitmap *forbidden_pages_map;
 978
 979/* Set bits in this map correspond to free page frames. */
 980static struct memory_bitmap *free_pages_map;
 981
 982/*
 983 * Each page frame allocated for creating the image is marked by setting the
 984 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
 985 */
 986
 987void swsusp_set_page_free(struct page *page)
 988{
 989	if (free_pages_map)
 990		memory_bm_set_bit(free_pages_map, page_to_pfn(page));
 991}
 992
 993static int swsusp_page_is_free(struct page *page)
 994{
 995	return free_pages_map ?
 996		memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
 997}
 998
 999void swsusp_unset_page_free(struct page *page)
1000{
1001	if (free_pages_map)
1002		memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1003}
1004
1005static void swsusp_set_page_forbidden(struct page *page)
1006{
1007	if (forbidden_pages_map)
1008		memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1009}
1010
1011int swsusp_page_is_forbidden(struct page *page)
1012{
1013	return forbidden_pages_map ?
1014		memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1015}
1016
1017static void swsusp_unset_page_forbidden(struct page *page)
1018{
1019	if (forbidden_pages_map)
1020		memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1021}
1022
1023/**
1024 * mark_nosave_pages - Mark pages that should not be saved.
1025 * @bm: Memory bitmap.
1026 *
1027 * Set the bits in @bm that correspond to the page frames the contents of which
1028 * should not be saved.
1029 */
1030static void mark_nosave_pages(struct memory_bitmap *bm)
1031{
1032	struct nosave_region *region;
1033
1034	if (list_empty(&nosave_regions))
1035		return;
1036
1037	list_for_each_entry(region, &nosave_regions, list) {
1038		unsigned long pfn;
1039
1040		pr_debug("PM: Marking nosave pages: [mem %#010llx-%#010llx]\n",
1041			 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1042			 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1043				- 1);
1044
1045		for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1046			if (pfn_valid(pfn)) {
1047				/*
1048				 * It is safe to ignore the result of
1049				 * mem_bm_set_bit_check() here, since we won't
1050				 * touch the PFNs for which the error is
1051				 * returned anyway.
1052				 */
1053				mem_bm_set_bit_check(bm, pfn);
1054			}
1055	}
1056}
1057
1058/**
1059 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1060 *
1061 * Create bitmaps needed for marking page frames that should not be saved and
1062 * free page frames.  The forbidden_pages_map and free_pages_map pointers are
1063 * only modified if everything goes well, because we don't want the bits to be
1064 * touched before both bitmaps are set up.
1065 */
1066int create_basic_memory_bitmaps(void)
1067{
1068	struct memory_bitmap *bm1, *bm2;
1069	int error = 0;
1070
1071	if (forbidden_pages_map && free_pages_map)
1072		return 0;
1073	else
1074		BUG_ON(forbidden_pages_map || free_pages_map);
1075
1076	bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1077	if (!bm1)
1078		return -ENOMEM;
1079
1080	error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1081	if (error)
1082		goto Free_first_object;
1083
1084	bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1085	if (!bm2)
1086		goto Free_first_bitmap;
1087
1088	error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1089	if (error)
1090		goto Free_second_object;
1091
1092	forbidden_pages_map = bm1;
1093	free_pages_map = bm2;
1094	mark_nosave_pages(forbidden_pages_map);
1095
1096	pr_debug("PM: Basic memory bitmaps created\n");
1097
1098	return 0;
1099
1100 Free_second_object:
1101	kfree(bm2);
1102 Free_first_bitmap:
1103 	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1104 Free_first_object:
1105	kfree(bm1);
1106	return -ENOMEM;
1107}
1108
1109/**
1110 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1111 *
1112 * Free memory bitmaps allocated by create_basic_memory_bitmaps().  The
1113 * auxiliary pointers are necessary so that the bitmaps themselves are not
1114 * referred to while they are being freed.
1115 */
1116void free_basic_memory_bitmaps(void)
1117{
1118	struct memory_bitmap *bm1, *bm2;
1119
1120	if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1121		return;
1122
1123	bm1 = forbidden_pages_map;
1124	bm2 = free_pages_map;
1125	forbidden_pages_map = NULL;
1126	free_pages_map = NULL;
1127	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1128	kfree(bm1);
1129	memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1130	kfree(bm2);
1131
1132	pr_debug("PM: Basic memory bitmaps freed\n");
1133}
1134
1135void clear_free_pages(void)
1136{
1137#ifdef CONFIG_PAGE_POISONING_ZERO
1138	struct memory_bitmap *bm = free_pages_map;
1139	unsigned long pfn;
1140
1141	if (WARN_ON(!(free_pages_map)))
1142		return;
1143
1144	memory_bm_position_reset(bm);
1145	pfn = memory_bm_next_pfn(bm);
1146	while (pfn != BM_END_OF_MAP) {
1147		if (pfn_valid(pfn))
1148			clear_highpage(pfn_to_page(pfn));
 
1149
1150		pfn = memory_bm_next_pfn(bm);
 
 
 
1151	}
1152	memory_bm_position_reset(bm);
1153	pr_info("PM: free pages cleared after restore\n");
1154#endif /* PAGE_POISONING_ZERO */
1155}
1156
1157/**
1158 * snapshot_additional_pages - Estimate the number of extra pages needed.
1159 * @zone: Memory zone to carry out the computation for.
1160 *
1161 * Estimate the number of additional pages needed for setting up a hibernation
1162 * image data structures for @zone (usually, the returned value is greater than
1163 * the exact number).
1164 */
1165unsigned int snapshot_additional_pages(struct zone *zone)
1166{
1167	unsigned int rtree, nodes;
1168
1169	rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1170	rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1171			      LINKED_PAGE_DATA_SIZE);
1172	while (nodes > 1) {
1173		nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1174		rtree += nodes;
1175	}
1176
1177	return 2 * rtree;
1178}
1179
1180#ifdef CONFIG_HIGHMEM
1181/**
1182 * count_free_highmem_pages - Compute the total number of free highmem pages.
1183 *
1184 * The returned number is system-wide.
1185 */
1186static unsigned int count_free_highmem_pages(void)
1187{
1188	struct zone *zone;
1189	unsigned int cnt = 0;
1190
1191	for_each_populated_zone(zone)
1192		if (is_highmem(zone))
1193			cnt += zone_page_state(zone, NR_FREE_PAGES);
1194
1195	return cnt;
1196}
1197
1198/**
1199 * saveable_highmem_page - Check if a highmem page is saveable.
1200 *
1201 * Determine whether a highmem page should be included in a hibernation image.
1202 *
1203 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1204 * and it isn't part of a free chunk of pages.
1205 */
1206static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1207{
1208	struct page *page;
1209
1210	if (!pfn_valid(pfn))
1211		return NULL;
1212
1213	page = pfn_to_page(pfn);
1214	if (page_zone(page) != zone)
1215		return NULL;
1216
1217	BUG_ON(!PageHighMem(page));
1218
1219	if (swsusp_page_is_forbidden(page) ||  swsusp_page_is_free(page) ||
1220	    PageReserved(page))
 
 
1221		return NULL;
1222
1223	if (page_is_guard(page))
1224		return NULL;
1225
1226	return page;
1227}
1228
1229/**
1230 * count_highmem_pages - Compute the total number of saveable highmem pages.
1231 */
1232static unsigned int count_highmem_pages(void)
1233{
1234	struct zone *zone;
1235	unsigned int n = 0;
1236
1237	for_each_populated_zone(zone) {
1238		unsigned long pfn, max_zone_pfn;
1239
1240		if (!is_highmem(zone))
1241			continue;
1242
1243		mark_free_pages(zone);
1244		max_zone_pfn = zone_end_pfn(zone);
1245		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1246			if (saveable_highmem_page(zone, pfn))
1247				n++;
1248	}
1249	return n;
1250}
1251#else
1252static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1253{
1254	return NULL;
1255}
1256#endif /* CONFIG_HIGHMEM */
1257
1258/**
1259 * saveable_page - Check if the given page is saveable.
1260 *
1261 * Determine whether a non-highmem page should be included in a hibernation
1262 * image.
1263 *
1264 * We should save the page if it isn't Nosave, and is not in the range
1265 * of pages statically defined as 'unsaveable', and it isn't part of
1266 * a free chunk of pages.
1267 */
1268static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1269{
1270	struct page *page;
1271
1272	if (!pfn_valid(pfn))
1273		return NULL;
1274
1275	page = pfn_to_page(pfn);
1276	if (page_zone(page) != zone)
1277		return NULL;
1278
1279	BUG_ON(PageHighMem(page));
1280
1281	if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1282		return NULL;
1283
 
 
 
1284	if (PageReserved(page)
1285	    && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1286		return NULL;
1287
1288	if (page_is_guard(page))
1289		return NULL;
1290
1291	return page;
1292}
1293
1294/**
1295 * count_data_pages - Compute the total number of saveable non-highmem pages.
1296 */
1297static unsigned int count_data_pages(void)
1298{
1299	struct zone *zone;
1300	unsigned long pfn, max_zone_pfn;
1301	unsigned int n = 0;
1302
1303	for_each_populated_zone(zone) {
1304		if (is_highmem(zone))
1305			continue;
1306
1307		mark_free_pages(zone);
1308		max_zone_pfn = zone_end_pfn(zone);
1309		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1310			if (saveable_page(zone, pfn))
1311				n++;
1312	}
1313	return n;
1314}
1315
1316/*
1317 * This is needed, because copy_page and memcpy are not usable for copying
1318 * task structs.
1319 */
1320static inline void do_copy_page(long *dst, long *src)
1321{
1322	int n;
1323
1324	for (n = PAGE_SIZE / sizeof(long); n; n--)
1325		*dst++ = *src++;
1326}
1327
1328/**
1329 * safe_copy_page - Copy a page in a safe way.
1330 *
1331 * Check if the page we are going to copy is marked as present in the kernel
1332 * page tables (this always is the case if CONFIG_DEBUG_PAGEALLOC is not set
1333 * and in that case kernel_page_present() always returns 'true').
 
1334 */
1335static void safe_copy_page(void *dst, struct page *s_page)
1336{
1337	if (kernel_page_present(s_page)) {
1338		do_copy_page(dst, page_address(s_page));
1339	} else {
1340		kernel_map_pages(s_page, 1, 1);
1341		do_copy_page(dst, page_address(s_page));
1342		kernel_map_pages(s_page, 1, 0);
1343	}
1344}
1345
1346#ifdef CONFIG_HIGHMEM
1347static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1348{
1349	return is_highmem(zone) ?
1350		saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1351}
1352
1353static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1354{
1355	struct page *s_page, *d_page;
1356	void *src, *dst;
1357
1358	s_page = pfn_to_page(src_pfn);
1359	d_page = pfn_to_page(dst_pfn);
1360	if (PageHighMem(s_page)) {
1361		src = kmap_atomic(s_page);
1362		dst = kmap_atomic(d_page);
1363		do_copy_page(dst, src);
1364		kunmap_atomic(dst);
1365		kunmap_atomic(src);
1366	} else {
1367		if (PageHighMem(d_page)) {
1368			/*
1369			 * The page pointed to by src may contain some kernel
1370			 * data modified by kmap_atomic()
1371			 */
1372			safe_copy_page(buffer, s_page);
1373			dst = kmap_atomic(d_page);
1374			copy_page(dst, buffer);
1375			kunmap_atomic(dst);
1376		} else {
1377			safe_copy_page(page_address(d_page), s_page);
1378		}
1379	}
1380}
1381#else
1382#define page_is_saveable(zone, pfn)	saveable_page(zone, pfn)
1383
1384static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1385{
1386	safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1387				pfn_to_page(src_pfn));
1388}
1389#endif /* CONFIG_HIGHMEM */
1390
1391static void copy_data_pages(struct memory_bitmap *copy_bm,
1392			    struct memory_bitmap *orig_bm)
1393{
1394	struct zone *zone;
1395	unsigned long pfn;
1396
1397	for_each_populated_zone(zone) {
1398		unsigned long max_zone_pfn;
1399
1400		mark_free_pages(zone);
1401		max_zone_pfn = zone_end_pfn(zone);
1402		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1403			if (page_is_saveable(zone, pfn))
1404				memory_bm_set_bit(orig_bm, pfn);
1405	}
1406	memory_bm_position_reset(orig_bm);
1407	memory_bm_position_reset(copy_bm);
1408	for(;;) {
1409		pfn = memory_bm_next_pfn(orig_bm);
1410		if (unlikely(pfn == BM_END_OF_MAP))
1411			break;
1412		copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1413	}
1414}
1415
1416/* Total number of image pages */
1417static unsigned int nr_copy_pages;
1418/* Number of pages needed for saving the original pfns of the image pages */
1419static unsigned int nr_meta_pages;
1420/*
1421 * Numbers of normal and highmem page frames allocated for hibernation image
1422 * before suspending devices.
1423 */
1424unsigned int alloc_normal, alloc_highmem;
1425/*
1426 * Memory bitmap used for marking saveable pages (during hibernation) or
1427 * hibernation image pages (during restore)
1428 */
1429static struct memory_bitmap orig_bm;
1430/*
1431 * Memory bitmap used during hibernation for marking allocated page frames that
1432 * will contain copies of saveable pages.  During restore it is initially used
1433 * for marking hibernation image pages, but then the set bits from it are
1434 * duplicated in @orig_bm and it is released.  On highmem systems it is next
1435 * used for marking "safe" highmem pages, but it has to be reinitialized for
1436 * this purpose.
1437 */
1438static struct memory_bitmap copy_bm;
1439
1440/**
1441 * swsusp_free - Free pages allocated for hibernation image.
1442 *
1443 * Image pages are alocated before snapshot creation, so they need to be
1444 * released after resume.
1445 */
1446void swsusp_free(void)
1447{
1448	unsigned long fb_pfn, fr_pfn;
1449
1450	if (!forbidden_pages_map || !free_pages_map)
1451		goto out;
1452
1453	memory_bm_position_reset(forbidden_pages_map);
1454	memory_bm_position_reset(free_pages_map);
1455
1456loop:
1457	fr_pfn = memory_bm_next_pfn(free_pages_map);
1458	fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1459
1460	/*
1461	 * Find the next bit set in both bitmaps. This is guaranteed to
1462	 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1463	 */
1464	do {
1465		if (fb_pfn < fr_pfn)
1466			fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1467		if (fr_pfn < fb_pfn)
1468			fr_pfn = memory_bm_next_pfn(free_pages_map);
1469	} while (fb_pfn != fr_pfn);
1470
1471	if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1472		struct page *page = pfn_to_page(fr_pfn);
1473
1474		memory_bm_clear_current(forbidden_pages_map);
1475		memory_bm_clear_current(free_pages_map);
1476		hibernate_restore_unprotect_page(page_address(page));
1477		__free_page(page);
1478		goto loop;
1479	}
1480
1481out:
1482	nr_copy_pages = 0;
1483	nr_meta_pages = 0;
1484	restore_pblist = NULL;
1485	buffer = NULL;
1486	alloc_normal = 0;
1487	alloc_highmem = 0;
1488	hibernate_restore_protection_end();
1489}
1490
1491/* Helper functions used for the shrinking of memory. */
1492
1493#define GFP_IMAGE	(GFP_KERNEL | __GFP_NOWARN)
1494
1495/**
1496 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1497 * @nr_pages: Number of page frames to allocate.
1498 * @mask: GFP flags to use for the allocation.
1499 *
1500 * Return value: Number of page frames actually allocated
1501 */
1502static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1503{
1504	unsigned long nr_alloc = 0;
1505
1506	while (nr_pages > 0) {
1507		struct page *page;
1508
1509		page = alloc_image_page(mask);
1510		if (!page)
1511			break;
1512		memory_bm_set_bit(&copy_bm, page_to_pfn(page));
1513		if (PageHighMem(page))
1514			alloc_highmem++;
1515		else
1516			alloc_normal++;
1517		nr_pages--;
1518		nr_alloc++;
1519	}
1520
1521	return nr_alloc;
1522}
1523
1524static unsigned long preallocate_image_memory(unsigned long nr_pages,
1525					      unsigned long avail_normal)
1526{
1527	unsigned long alloc;
1528
1529	if (avail_normal <= alloc_normal)
1530		return 0;
1531
1532	alloc = avail_normal - alloc_normal;
1533	if (nr_pages < alloc)
1534		alloc = nr_pages;
1535
1536	return preallocate_image_pages(alloc, GFP_IMAGE);
1537}
1538
1539#ifdef CONFIG_HIGHMEM
1540static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1541{
1542	return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1543}
1544
1545/**
1546 *  __fraction - Compute (an approximation of) x * (multiplier / base).
1547 */
1548static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1549{
1550	x *= multiplier;
1551	do_div(x, base);
1552	return (unsigned long)x;
1553}
1554
1555static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1556						  unsigned long highmem,
1557						  unsigned long total)
1558{
1559	unsigned long alloc = __fraction(nr_pages, highmem, total);
1560
1561	return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1562}
1563#else /* CONFIG_HIGHMEM */
1564static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1565{
1566	return 0;
1567}
1568
1569static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1570							 unsigned long highmem,
1571							 unsigned long total)
1572{
1573	return 0;
1574}
1575#endif /* CONFIG_HIGHMEM */
1576
1577/**
1578 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1579 */
1580static unsigned long free_unnecessary_pages(void)
1581{
1582	unsigned long save, to_free_normal, to_free_highmem, free;
1583
1584	save = count_data_pages();
1585	if (alloc_normal >= save) {
1586		to_free_normal = alloc_normal - save;
1587		save = 0;
1588	} else {
1589		to_free_normal = 0;
1590		save -= alloc_normal;
1591	}
1592	save += count_highmem_pages();
1593	if (alloc_highmem >= save) {
1594		to_free_highmem = alloc_highmem - save;
1595	} else {
1596		to_free_highmem = 0;
1597		save -= alloc_highmem;
1598		if (to_free_normal > save)
1599			to_free_normal -= save;
1600		else
1601			to_free_normal = 0;
1602	}
1603	free = to_free_normal + to_free_highmem;
1604
1605	memory_bm_position_reset(&copy_bm);
1606
1607	while (to_free_normal > 0 || to_free_highmem > 0) {
1608		unsigned long pfn = memory_bm_next_pfn(&copy_bm);
1609		struct page *page = pfn_to_page(pfn);
1610
1611		if (PageHighMem(page)) {
1612			if (!to_free_highmem)
1613				continue;
1614			to_free_highmem--;
1615			alloc_highmem--;
1616		} else {
1617			if (!to_free_normal)
1618				continue;
1619			to_free_normal--;
1620			alloc_normal--;
1621		}
1622		memory_bm_clear_bit(&copy_bm, pfn);
1623		swsusp_unset_page_forbidden(page);
1624		swsusp_unset_page_free(page);
1625		__free_page(page);
1626	}
1627
1628	return free;
1629}
1630
1631/**
1632 * minimum_image_size - Estimate the minimum acceptable size of an image.
1633 * @saveable: Number of saveable pages in the system.
1634 *
1635 * We want to avoid attempting to free too much memory too hard, so estimate the
1636 * minimum acceptable size of a hibernation image to use as the lower limit for
1637 * preallocating memory.
1638 *
1639 * We assume that the minimum image size should be proportional to
1640 *
1641 * [number of saveable pages] - [number of pages that can be freed in theory]
1642 *
1643 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1644 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages,
1645 * minus mapped file pages.
1646 */
1647static unsigned long minimum_image_size(unsigned long saveable)
1648{
1649	unsigned long size;
1650
1651	size = global_page_state(NR_SLAB_RECLAIMABLE)
1652		+ global_node_page_state(NR_ACTIVE_ANON)
1653		+ global_node_page_state(NR_INACTIVE_ANON)
1654		+ global_node_page_state(NR_ACTIVE_FILE)
1655		+ global_node_page_state(NR_INACTIVE_FILE)
1656		- global_node_page_state(NR_FILE_MAPPED);
1657
1658	return saveable <= size ? 0 : saveable - size;
1659}
1660
1661/**
1662 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1663 *
1664 * To create a hibernation image it is necessary to make a copy of every page
1665 * frame in use.  We also need a number of page frames to be free during
1666 * hibernation for allocations made while saving the image and for device
1667 * drivers, in case they need to allocate memory from their hibernation
1668 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1669 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1670 * /sys/power/reserved_size, respectively).  To make this happen, we compute the
1671 * total number of available page frames and allocate at least
1672 *
1673 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1674 *  + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1675 *
1676 * of them, which corresponds to the maximum size of a hibernation image.
1677 *
1678 * If image_size is set below the number following from the above formula,
1679 * the preallocation of memory is continued until the total number of saveable
1680 * pages in the system is below the requested image size or the minimum
1681 * acceptable image size returned by minimum_image_size(), whichever is greater.
1682 */
1683int hibernate_preallocate_memory(void)
1684{
1685	struct zone *zone;
1686	unsigned long saveable, size, max_size, count, highmem, pages = 0;
1687	unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1688	ktime_t start, stop;
1689	int error;
1690
1691	printk(KERN_INFO "PM: Preallocating image memory... ");
1692	start = ktime_get();
1693
1694	error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1695	if (error)
 
1696		goto err_out;
 
1697
1698	error = memory_bm_create(&copy_bm, GFP_IMAGE, PG_ANY);
1699	if (error)
 
1700		goto err_out;
 
1701
1702	alloc_normal = 0;
1703	alloc_highmem = 0;
1704
1705	/* Count the number of saveable data pages. */
1706	save_highmem = count_highmem_pages();
1707	saveable = count_data_pages();
1708
1709	/*
1710	 * Compute the total number of page frames we can use (count) and the
1711	 * number of pages needed for image metadata (size).
1712	 */
1713	count = saveable;
1714	saveable += save_highmem;
1715	highmem = save_highmem;
1716	size = 0;
1717	for_each_populated_zone(zone) {
1718		size += snapshot_additional_pages(zone);
1719		if (is_highmem(zone))
1720			highmem += zone_page_state(zone, NR_FREE_PAGES);
1721		else
1722			count += zone_page_state(zone, NR_FREE_PAGES);
1723	}
1724	avail_normal = count;
1725	count += highmem;
1726	count -= totalreserve_pages;
1727
1728	/* Add number of pages required for page keys (s390 only). */
1729	size += page_key_additional_pages(saveable);
1730
1731	/* Compute the maximum number of saveable pages to leave in memory. */
1732	max_size = (count - (size + PAGES_FOR_IO)) / 2
1733			- 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1734	/* Compute the desired number of image pages specified by image_size. */
1735	size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1736	if (size > max_size)
1737		size = max_size;
1738	/*
1739	 * If the desired number of image pages is at least as large as the
1740	 * current number of saveable pages in memory, allocate page frames for
1741	 * the image and we're done.
1742	 */
1743	if (size >= saveable) {
1744		pages = preallocate_image_highmem(save_highmem);
1745		pages += preallocate_image_memory(saveable - pages, avail_normal);
1746		goto out;
1747	}
1748
1749	/* Estimate the minimum size of the image. */
1750	pages = minimum_image_size(saveable);
1751	/*
1752	 * To avoid excessive pressure on the normal zone, leave room in it to
1753	 * accommodate an image of the minimum size (unless it's already too
1754	 * small, in which case don't preallocate pages from it at all).
1755	 */
1756	if (avail_normal > pages)
1757		avail_normal -= pages;
1758	else
1759		avail_normal = 0;
1760	if (size < pages)
1761		size = min_t(unsigned long, pages, max_size);
1762
1763	/*
1764	 * Let the memory management subsystem know that we're going to need a
1765	 * large number of page frames to allocate and make it free some memory.
1766	 * NOTE: If this is not done, performance will be hurt badly in some
1767	 * test cases.
1768	 */
1769	shrink_all_memory(saveable - size);
1770
1771	/*
1772	 * The number of saveable pages in memory was too high, so apply some
1773	 * pressure to decrease it.  First, make room for the largest possible
1774	 * image and fail if that doesn't work.  Next, try to decrease the size
1775	 * of the image as much as indicated by 'size' using allocations from
1776	 * highmem and non-highmem zones separately.
1777	 */
1778	pages_highmem = preallocate_image_highmem(highmem / 2);
1779	alloc = count - max_size;
1780	if (alloc > pages_highmem)
1781		alloc -= pages_highmem;
1782	else
1783		alloc = 0;
1784	pages = preallocate_image_memory(alloc, avail_normal);
1785	if (pages < alloc) {
1786		/* We have exhausted non-highmem pages, try highmem. */
1787		alloc -= pages;
1788		pages += pages_highmem;
1789		pages_highmem = preallocate_image_highmem(alloc);
1790		if (pages_highmem < alloc)
 
 
1791			goto err_out;
 
1792		pages += pages_highmem;
1793		/*
1794		 * size is the desired number of saveable pages to leave in
1795		 * memory, so try to preallocate (all memory - size) pages.
1796		 */
1797		alloc = (count - pages) - size;
1798		pages += preallocate_image_highmem(alloc);
1799	} else {
1800		/*
1801		 * There are approximately max_size saveable pages at this point
1802		 * and we want to reduce this number down to size.
1803		 */
1804		alloc = max_size - size;
1805		size = preallocate_highmem_fraction(alloc, highmem, count);
1806		pages_highmem += size;
1807		alloc -= size;
1808		size = preallocate_image_memory(alloc, avail_normal);
1809		pages_highmem += preallocate_image_highmem(alloc - size);
1810		pages += pages_highmem + size;
1811	}
1812
1813	/*
1814	 * We only need as many page frames for the image as there are saveable
1815	 * pages in memory, but we have allocated more.  Release the excessive
1816	 * ones now.
1817	 */
1818	pages -= free_unnecessary_pages();
1819
1820 out:
1821	stop = ktime_get();
1822	printk(KERN_CONT "done (allocated %lu pages)\n", pages);
1823	swsusp_show_speed(start, stop, pages, "Allocated");
1824
1825	return 0;
1826
1827 err_out:
1828	printk(KERN_CONT "\n");
1829	swsusp_free();
1830	return -ENOMEM;
1831}
1832
1833#ifdef CONFIG_HIGHMEM
1834/**
1835 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1836 *
1837 * Compute the number of non-highmem pages that will be necessary for creating
1838 * copies of highmem pages.
1839 */
1840static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1841{
1842	unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1843
1844	if (free_highmem >= nr_highmem)
1845		nr_highmem = 0;
1846	else
1847		nr_highmem -= free_highmem;
1848
1849	return nr_highmem;
1850}
1851#else
1852static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1853#endif /* CONFIG_HIGHMEM */
1854
1855/**
1856 * enough_free_mem - Check if there is enough free memory for the image.
1857 */
1858static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1859{
1860	struct zone *zone;
1861	unsigned int free = alloc_normal;
1862
1863	for_each_populated_zone(zone)
1864		if (!is_highmem(zone))
1865			free += zone_page_state(zone, NR_FREE_PAGES);
1866
1867	nr_pages += count_pages_for_highmem(nr_highmem);
1868	pr_debug("PM: Normal pages needed: %u + %u, available pages: %u\n",
1869		nr_pages, PAGES_FOR_IO, free);
1870
1871	return free > nr_pages + PAGES_FOR_IO;
1872}
1873
1874#ifdef CONFIG_HIGHMEM
1875/**
1876 * get_highmem_buffer - Allocate a buffer for highmem pages.
1877 *
1878 * If there are some highmem pages in the hibernation image, we may need a
1879 * buffer to copy them and/or load their data.
1880 */
1881static inline int get_highmem_buffer(int safe_needed)
1882{
1883	buffer = get_image_page(GFP_ATOMIC | __GFP_COLD, safe_needed);
1884	return buffer ? 0 : -ENOMEM;
1885}
1886
1887/**
1888 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1889 *
1890 * Try to allocate as many pages as needed, but if the number of free highmem
1891 * pages is less than that, allocate them all.
1892 */
1893static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1894					       unsigned int nr_highmem)
1895{
1896	unsigned int to_alloc = count_free_highmem_pages();
1897
1898	if (to_alloc > nr_highmem)
1899		to_alloc = nr_highmem;
1900
1901	nr_highmem -= to_alloc;
1902	while (to_alloc-- > 0) {
1903		struct page *page;
1904
1905		page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1906		memory_bm_set_bit(bm, page_to_pfn(page));
1907	}
1908	return nr_highmem;
1909}
1910#else
1911static inline int get_highmem_buffer(int safe_needed) { return 0; }
1912
1913static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1914					       unsigned int n) { return 0; }
1915#endif /* CONFIG_HIGHMEM */
1916
1917/**
1918 * swsusp_alloc - Allocate memory for hibernation image.
1919 *
1920 * We first try to allocate as many highmem pages as there are
1921 * saveable highmem pages in the system.  If that fails, we allocate
1922 * non-highmem pages for the copies of the remaining highmem ones.
1923 *
1924 * In this approach it is likely that the copies of highmem pages will
1925 * also be located in the high memory, because of the way in which
1926 * copy_data_pages() works.
1927 */
1928static int swsusp_alloc(struct memory_bitmap *orig_bm,
1929			struct memory_bitmap *copy_bm,
1930			unsigned int nr_pages, unsigned int nr_highmem)
1931{
1932	if (nr_highmem > 0) {
1933		if (get_highmem_buffer(PG_ANY))
1934			goto err_out;
1935		if (nr_highmem > alloc_highmem) {
1936			nr_highmem -= alloc_highmem;
1937			nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1938		}
1939	}
1940	if (nr_pages > alloc_normal) {
1941		nr_pages -= alloc_normal;
1942		while (nr_pages-- > 0) {
1943			struct page *page;
1944
1945			page = alloc_image_page(GFP_ATOMIC | __GFP_COLD);
1946			if (!page)
1947				goto err_out;
1948			memory_bm_set_bit(copy_bm, page_to_pfn(page));
1949		}
1950	}
1951
1952	return 0;
1953
1954 err_out:
1955	swsusp_free();
1956	return -ENOMEM;
1957}
1958
1959asmlinkage __visible int swsusp_save(void)
1960{
1961	unsigned int nr_pages, nr_highmem;
1962
1963	printk(KERN_INFO "PM: Creating hibernation image:\n");
1964
1965	drain_local_pages(NULL);
1966	nr_pages = count_data_pages();
1967	nr_highmem = count_highmem_pages();
1968	printk(KERN_INFO "PM: Need to copy %u pages\n", nr_pages + nr_highmem);
1969
1970	if (!enough_free_mem(nr_pages, nr_highmem)) {
1971		printk(KERN_ERR "PM: Not enough free memory\n");
1972		return -ENOMEM;
1973	}
1974
1975	if (swsusp_alloc(&orig_bm, &copy_bm, nr_pages, nr_highmem)) {
1976		printk(KERN_ERR "PM: Memory allocation failed\n");
1977		return -ENOMEM;
1978	}
1979
1980	/*
1981	 * During allocating of suspend pagedir, new cold pages may appear.
1982	 * Kill them.
1983	 */
1984	drain_local_pages(NULL);
1985	copy_data_pages(&copy_bm, &orig_bm);
1986
1987	/*
1988	 * End of critical section. From now on, we can write to memory,
1989	 * but we should not touch disk. This specially means we must _not_
1990	 * touch swap space! Except we must write out our image of course.
1991	 */
1992
1993	nr_pages += nr_highmem;
1994	nr_copy_pages = nr_pages;
1995	nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
1996
1997	printk(KERN_INFO "PM: Hibernation image created (%d pages copied)\n",
1998		nr_pages);
1999
2000	return 0;
2001}
2002
2003#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2004static int init_header_complete(struct swsusp_info *info)
2005{
2006	memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2007	info->version_code = LINUX_VERSION_CODE;
2008	return 0;
2009}
2010
2011static char *check_image_kernel(struct swsusp_info *info)
2012{
2013	if (info->version_code != LINUX_VERSION_CODE)
2014		return "kernel version";
2015	if (strcmp(info->uts.sysname,init_utsname()->sysname))
2016		return "system type";
2017	if (strcmp(info->uts.release,init_utsname()->release))
2018		return "kernel release";
2019	if (strcmp(info->uts.version,init_utsname()->version))
2020		return "version";
2021	if (strcmp(info->uts.machine,init_utsname()->machine))
2022		return "machine";
2023	return NULL;
2024}
2025#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2026
2027unsigned long snapshot_get_image_size(void)
2028{
2029	return nr_copy_pages + nr_meta_pages + 1;
2030}
2031
2032static int init_header(struct swsusp_info *info)
2033{
2034	memset(info, 0, sizeof(struct swsusp_info));
2035	info->num_physpages = get_num_physpages();
2036	info->image_pages = nr_copy_pages;
2037	info->pages = snapshot_get_image_size();
2038	info->size = info->pages;
2039	info->size <<= PAGE_SHIFT;
2040	return init_header_complete(info);
2041}
2042
2043/**
2044 * pack_pfns - Prepare PFNs for saving.
2045 * @bm: Memory bitmap.
2046 * @buf: Memory buffer to store the PFNs in.
2047 *
2048 * PFNs corresponding to set bits in @bm are stored in the area of memory
2049 * pointed to by @buf (1 page at a time).
2050 */
2051static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2052{
2053	int j;
2054
2055	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2056		buf[j] = memory_bm_next_pfn(bm);
2057		if (unlikely(buf[j] == BM_END_OF_MAP))
2058			break;
2059		/* Save page key for data page (s390 only). */
2060		page_key_read(buf + j);
2061	}
2062}
2063
2064/**
2065 * snapshot_read_next - Get the address to read the next image page from.
2066 * @handle: Snapshot handle to be used for the reading.
2067 *
2068 * On the first call, @handle should point to a zeroed snapshot_handle
2069 * structure.  The structure gets populated then and a pointer to it should be
2070 * passed to this function every next time.
2071 *
2072 * On success, the function returns a positive number.  Then, the caller
2073 * is allowed to read up to the returned number of bytes from the memory
2074 * location computed by the data_of() macro.
2075 *
2076 * The function returns 0 to indicate the end of the data stream condition,
2077 * and negative numbers are returned on errors.  If that happens, the structure
2078 * pointed to by @handle is not updated and should not be used any more.
2079 */
2080int snapshot_read_next(struct snapshot_handle *handle)
2081{
2082	if (handle->cur > nr_meta_pages + nr_copy_pages)
2083		return 0;
2084
2085	if (!buffer) {
2086		/* This makes the buffer be freed by swsusp_free() */
2087		buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2088		if (!buffer)
2089			return -ENOMEM;
2090	}
2091	if (!handle->cur) {
2092		int error;
2093
2094		error = init_header((struct swsusp_info *)buffer);
2095		if (error)
2096			return error;
2097		handle->buffer = buffer;
2098		memory_bm_position_reset(&orig_bm);
2099		memory_bm_position_reset(&copy_bm);
2100	} else if (handle->cur <= nr_meta_pages) {
2101		clear_page(buffer);
2102		pack_pfns(buffer, &orig_bm);
2103	} else {
2104		struct page *page;
2105
2106		page = pfn_to_page(memory_bm_next_pfn(&copy_bm));
2107		if (PageHighMem(page)) {
2108			/*
2109			 * Highmem pages are copied to the buffer,
2110			 * because we can't return with a kmapped
2111			 * highmem page (we may not be called again).
2112			 */
2113			void *kaddr;
2114
2115			kaddr = kmap_atomic(page);
2116			copy_page(buffer, kaddr);
2117			kunmap_atomic(kaddr);
2118			handle->buffer = buffer;
2119		} else {
2120			handle->buffer = page_address(page);
2121		}
2122	}
2123	handle->cur++;
2124	return PAGE_SIZE;
2125}
2126
2127static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2128				    struct memory_bitmap *src)
2129{
2130	unsigned long pfn;
2131
2132	memory_bm_position_reset(src);
2133	pfn = memory_bm_next_pfn(src);
2134	while (pfn != BM_END_OF_MAP) {
2135		memory_bm_set_bit(dst, pfn);
2136		pfn = memory_bm_next_pfn(src);
2137	}
2138}
2139
2140/**
2141 * mark_unsafe_pages - Mark pages that were used before hibernation.
2142 *
2143 * Mark the pages that cannot be used for storing the image during restoration,
2144 * because they conflict with the pages that had been used before hibernation.
2145 */
2146static void mark_unsafe_pages(struct memory_bitmap *bm)
2147{
2148	unsigned long pfn;
2149
2150	/* Clear the "free"/"unsafe" bit for all PFNs */
2151	memory_bm_position_reset(free_pages_map);
2152	pfn = memory_bm_next_pfn(free_pages_map);
2153	while (pfn != BM_END_OF_MAP) {
2154		memory_bm_clear_current(free_pages_map);
2155		pfn = memory_bm_next_pfn(free_pages_map);
2156	}
2157
2158	/* Mark pages that correspond to the "original" PFNs as "unsafe" */
2159	duplicate_memory_bitmap(free_pages_map, bm);
2160
2161	allocated_unsafe_pages = 0;
2162}
2163
2164static int check_header(struct swsusp_info *info)
2165{
2166	char *reason;
2167
2168	reason = check_image_kernel(info);
2169	if (!reason && info->num_physpages != get_num_physpages())
2170		reason = "memory size";
2171	if (reason) {
2172		printk(KERN_ERR "PM: Image mismatch: %s\n", reason);
2173		return -EPERM;
2174	}
2175	return 0;
2176}
2177
2178/**
2179 * load header - Check the image header and copy the data from it.
2180 */
2181static int load_header(struct swsusp_info *info)
2182{
2183	int error;
2184
2185	restore_pblist = NULL;
2186	error = check_header(info);
2187	if (!error) {
2188		nr_copy_pages = info->image_pages;
2189		nr_meta_pages = info->pages - info->image_pages - 1;
2190	}
2191	return error;
2192}
2193
2194/**
2195 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2196 * @bm: Memory bitmap.
2197 * @buf: Area of memory containing the PFNs.
2198 *
2199 * For each element of the array pointed to by @buf (1 page at a time), set the
2200 * corresponding bit in @bm.
2201 */
2202static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2203{
2204	int j;
2205
2206	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2207		if (unlikely(buf[j] == BM_END_OF_MAP))
2208			break;
2209
2210		/* Extract and buffer page key for data page (s390 only). */
2211		page_key_memorize(buf + j);
2212
2213		if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2214			memory_bm_set_bit(bm, buf[j]);
2215		else
2216			return -EFAULT;
2217	}
2218
2219	return 0;
2220}
2221
2222#ifdef CONFIG_HIGHMEM
2223/*
2224 * struct highmem_pbe is used for creating the list of highmem pages that
2225 * should be restored atomically during the resume from disk, because the page
2226 * frames they have occupied before the suspend are in use.
2227 */
2228struct highmem_pbe {
2229	struct page *copy_page;	/* data is here now */
2230	struct page *orig_page;	/* data was here before the suspend */
2231	struct highmem_pbe *next;
2232};
2233
2234/*
2235 * List of highmem PBEs needed for restoring the highmem pages that were
2236 * allocated before the suspend and included in the suspend image, but have
2237 * also been allocated by the "resume" kernel, so their contents cannot be
2238 * written directly to their "original" page frames.
2239 */
2240static struct highmem_pbe *highmem_pblist;
2241
2242/**
2243 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2244 * @bm: Memory bitmap.
2245 *
2246 * The bits in @bm that correspond to image pages are assumed to be set.
2247 */
2248static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2249{
2250	unsigned long pfn;
2251	unsigned int cnt = 0;
2252
2253	memory_bm_position_reset(bm);
2254	pfn = memory_bm_next_pfn(bm);
2255	while (pfn != BM_END_OF_MAP) {
2256		if (PageHighMem(pfn_to_page(pfn)))
2257			cnt++;
2258
2259		pfn = memory_bm_next_pfn(bm);
2260	}
2261	return cnt;
2262}
2263
2264static unsigned int safe_highmem_pages;
2265
2266static struct memory_bitmap *safe_highmem_bm;
2267
2268/**
2269 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2270 * @bm: Pointer to an uninitialized memory bitmap structure.
2271 * @nr_highmem_p: Pointer to the number of highmem image pages.
2272 *
2273 * Try to allocate as many highmem pages as there are highmem image pages
2274 * (@nr_highmem_p points to the variable containing the number of highmem image
2275 * pages).  The pages that are "safe" (ie. will not be overwritten when the
2276 * hibernation image is restored entirely) have the corresponding bits set in
2277 * @bm (it must be unitialized).
2278 *
2279 * NOTE: This function should not be called if there are no highmem image pages.
2280 */
2281static int prepare_highmem_image(struct memory_bitmap *bm,
2282				 unsigned int *nr_highmem_p)
2283{
2284	unsigned int to_alloc;
2285
2286	if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2287		return -ENOMEM;
2288
2289	if (get_highmem_buffer(PG_SAFE))
2290		return -ENOMEM;
2291
2292	to_alloc = count_free_highmem_pages();
2293	if (to_alloc > *nr_highmem_p)
2294		to_alloc = *nr_highmem_p;
2295	else
2296		*nr_highmem_p = to_alloc;
2297
2298	safe_highmem_pages = 0;
2299	while (to_alloc-- > 0) {
2300		struct page *page;
2301
2302		page = alloc_page(__GFP_HIGHMEM);
2303		if (!swsusp_page_is_free(page)) {
2304			/* The page is "safe", set its bit the bitmap */
2305			memory_bm_set_bit(bm, page_to_pfn(page));
2306			safe_highmem_pages++;
2307		}
2308		/* Mark the page as allocated */
2309		swsusp_set_page_forbidden(page);
2310		swsusp_set_page_free(page);
2311	}
2312	memory_bm_position_reset(bm);
2313	safe_highmem_bm = bm;
2314	return 0;
2315}
2316
2317static struct page *last_highmem_page;
2318
2319/**
2320 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2321 *
2322 * For a given highmem image page get a buffer that suspend_write_next() should
2323 * return to its caller to write to.
2324 *
2325 * If the page is to be saved to its "original" page frame or a copy of
2326 * the page is to be made in the highmem, @buffer is returned.  Otherwise,
2327 * the copy of the page is to be made in normal memory, so the address of
2328 * the copy is returned.
2329 *
2330 * If @buffer is returned, the caller of suspend_write_next() will write
2331 * the page's contents to @buffer, so they will have to be copied to the
2332 * right location on the next call to suspend_write_next() and it is done
2333 * with the help of copy_last_highmem_page().  For this purpose, if
2334 * @buffer is returned, @last_highmem_page is set to the page to which
2335 * the data will have to be copied from @buffer.
2336 */
2337static void *get_highmem_page_buffer(struct page *page,
2338				     struct chain_allocator *ca)
2339{
2340	struct highmem_pbe *pbe;
2341	void *kaddr;
2342
2343	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2344		/*
2345		 * We have allocated the "original" page frame and we can
2346		 * use it directly to store the loaded page.
2347		 */
2348		last_highmem_page = page;
2349		return buffer;
2350	}
2351	/*
2352	 * The "original" page frame has not been allocated and we have to
2353	 * use a "safe" page frame to store the loaded page.
2354	 */
2355	pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2356	if (!pbe) {
2357		swsusp_free();
2358		return ERR_PTR(-ENOMEM);
2359	}
2360	pbe->orig_page = page;
2361	if (safe_highmem_pages > 0) {
2362		struct page *tmp;
2363
2364		/* Copy of the page will be stored in high memory */
2365		kaddr = buffer;
2366		tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2367		safe_highmem_pages--;
2368		last_highmem_page = tmp;
2369		pbe->copy_page = tmp;
2370	} else {
2371		/* Copy of the page will be stored in normal memory */
2372		kaddr = safe_pages_list;
2373		safe_pages_list = safe_pages_list->next;
2374		pbe->copy_page = virt_to_page(kaddr);
2375	}
2376	pbe->next = highmem_pblist;
2377	highmem_pblist = pbe;
2378	return kaddr;
2379}
2380
2381/**
2382 * copy_last_highmem_page - Copy most the most recent highmem image page.
2383 *
2384 * Copy the contents of a highmem image from @buffer, where the caller of
2385 * snapshot_write_next() has stored them, to the right location represented by
2386 * @last_highmem_page .
2387 */
2388static void copy_last_highmem_page(void)
2389{
2390	if (last_highmem_page) {
2391		void *dst;
2392
2393		dst = kmap_atomic(last_highmem_page);
2394		copy_page(dst, buffer);
2395		kunmap_atomic(dst);
2396		last_highmem_page = NULL;
2397	}
2398}
2399
2400static inline int last_highmem_page_copied(void)
2401{
2402	return !last_highmem_page;
2403}
2404
2405static inline void free_highmem_data(void)
2406{
2407	if (safe_highmem_bm)
2408		memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2409
2410	if (buffer)
2411		free_image_page(buffer, PG_UNSAFE_CLEAR);
2412}
2413#else
2414static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2415
2416static inline int prepare_highmem_image(struct memory_bitmap *bm,
2417					unsigned int *nr_highmem_p) { return 0; }
2418
2419static inline void *get_highmem_page_buffer(struct page *page,
2420					    struct chain_allocator *ca)
2421{
2422	return ERR_PTR(-EINVAL);
2423}
2424
2425static inline void copy_last_highmem_page(void) {}
2426static inline int last_highmem_page_copied(void) { return 1; }
2427static inline void free_highmem_data(void) {}
2428#endif /* CONFIG_HIGHMEM */
2429
2430#define PBES_PER_LINKED_PAGE	(LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2431
2432/**
2433 * prepare_image - Make room for loading hibernation image.
2434 * @new_bm: Unitialized memory bitmap structure.
2435 * @bm: Memory bitmap with unsafe pages marked.
2436 *
2437 * Use @bm to mark the pages that will be overwritten in the process of
2438 * restoring the system memory state from the suspend image ("unsafe" pages)
2439 * and allocate memory for the image.
2440 *
2441 * The idea is to allocate a new memory bitmap first and then allocate
2442 * as many pages as needed for image data, but without specifying what those
2443 * pages will be used for just yet.  Instead, we mark them all as allocated and
2444 * create a lists of "safe" pages to be used later.  On systems with high
2445 * memory a list of "safe" highmem pages is created too.
2446 */
2447static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2448{
2449	unsigned int nr_pages, nr_highmem;
2450	struct linked_page *lp;
2451	int error;
2452
2453	/* If there is no highmem, the buffer will not be necessary */
2454	free_image_page(buffer, PG_UNSAFE_CLEAR);
2455	buffer = NULL;
2456
2457	nr_highmem = count_highmem_image_pages(bm);
2458	mark_unsafe_pages(bm);
2459
2460	error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2461	if (error)
2462		goto Free;
2463
2464	duplicate_memory_bitmap(new_bm, bm);
2465	memory_bm_free(bm, PG_UNSAFE_KEEP);
2466	if (nr_highmem > 0) {
2467		error = prepare_highmem_image(bm, &nr_highmem);
2468		if (error)
2469			goto Free;
2470	}
2471	/*
2472	 * Reserve some safe pages for potential later use.
2473	 *
2474	 * NOTE: This way we make sure there will be enough safe pages for the
2475	 * chain_alloc() in get_buffer().  It is a bit wasteful, but
2476	 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2477	 *
2478	 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2479	 */
2480	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2481	nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2482	while (nr_pages > 0) {
2483		lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2484		if (!lp) {
2485			error = -ENOMEM;
2486			goto Free;
2487		}
2488		lp->next = safe_pages_list;
2489		safe_pages_list = lp;
2490		nr_pages--;
2491	}
2492	/* Preallocate memory for the image */
2493	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2494	while (nr_pages > 0) {
2495		lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2496		if (!lp) {
2497			error = -ENOMEM;
2498			goto Free;
2499		}
2500		if (!swsusp_page_is_free(virt_to_page(lp))) {
2501			/* The page is "safe", add it to the list */
2502			lp->next = safe_pages_list;
2503			safe_pages_list = lp;
2504		}
2505		/* Mark the page as allocated */
2506		swsusp_set_page_forbidden(virt_to_page(lp));
2507		swsusp_set_page_free(virt_to_page(lp));
2508		nr_pages--;
2509	}
2510	return 0;
2511
2512 Free:
2513	swsusp_free();
2514	return error;
2515}
2516
2517/**
2518 * get_buffer - Get the address to store the next image data page.
2519 *
2520 * Get the address that snapshot_write_next() should return to its caller to
2521 * write to.
2522 */
2523static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2524{
2525	struct pbe *pbe;
2526	struct page *page;
2527	unsigned long pfn = memory_bm_next_pfn(bm);
2528
2529	if (pfn == BM_END_OF_MAP)
2530		return ERR_PTR(-EFAULT);
2531
2532	page = pfn_to_page(pfn);
2533	if (PageHighMem(page))
2534		return get_highmem_page_buffer(page, ca);
2535
2536	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2537		/*
2538		 * We have allocated the "original" page frame and we can
2539		 * use it directly to store the loaded page.
2540		 */
2541		return page_address(page);
2542
2543	/*
2544	 * The "original" page frame has not been allocated and we have to
2545	 * use a "safe" page frame to store the loaded page.
2546	 */
2547	pbe = chain_alloc(ca, sizeof(struct pbe));
2548	if (!pbe) {
2549		swsusp_free();
2550		return ERR_PTR(-ENOMEM);
2551	}
2552	pbe->orig_address = page_address(page);
2553	pbe->address = safe_pages_list;
2554	safe_pages_list = safe_pages_list->next;
2555	pbe->next = restore_pblist;
2556	restore_pblist = pbe;
2557	return pbe->address;
2558}
2559
2560/**
2561 * snapshot_write_next - Get the address to store the next image page.
2562 * @handle: Snapshot handle structure to guide the writing.
2563 *
2564 * On the first call, @handle should point to a zeroed snapshot_handle
2565 * structure.  The structure gets populated then and a pointer to it should be
2566 * passed to this function every next time.
2567 *
2568 * On success, the function returns a positive number.  Then, the caller
2569 * is allowed to write up to the returned number of bytes to the memory
2570 * location computed by the data_of() macro.
2571 *
2572 * The function returns 0 to indicate the "end of file" condition.  Negative
2573 * numbers are returned on errors, in which cases the structure pointed to by
2574 * @handle is not updated and should not be used any more.
2575 */
2576int snapshot_write_next(struct snapshot_handle *handle)
2577{
2578	static struct chain_allocator ca;
2579	int error = 0;
2580
2581	/* Check if we have already loaded the entire image */
2582	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2583		return 0;
2584
2585	handle->sync_read = 1;
2586
2587	if (!handle->cur) {
2588		if (!buffer)
2589			/* This makes the buffer be freed by swsusp_free() */
2590			buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2591
2592		if (!buffer)
2593			return -ENOMEM;
2594
2595		handle->buffer = buffer;
2596	} else if (handle->cur == 1) {
2597		error = load_header(buffer);
2598		if (error)
2599			return error;
2600
2601		safe_pages_list = NULL;
2602
2603		error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
2604		if (error)
2605			return error;
2606
2607		/* Allocate buffer for page keys. */
2608		error = page_key_alloc(nr_copy_pages);
2609		if (error)
2610			return error;
2611
2612		hibernate_restore_protection_begin();
2613	} else if (handle->cur <= nr_meta_pages + 1) {
2614		error = unpack_orig_pfns(buffer, &copy_bm);
2615		if (error)
2616			return error;
2617
2618		if (handle->cur == nr_meta_pages + 1) {
2619			error = prepare_image(&orig_bm, &copy_bm);
2620			if (error)
2621				return error;
2622
2623			chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2624			memory_bm_position_reset(&orig_bm);
2625			restore_pblist = NULL;
2626			handle->buffer = get_buffer(&orig_bm, &ca);
2627			handle->sync_read = 0;
2628			if (IS_ERR(handle->buffer))
2629				return PTR_ERR(handle->buffer);
2630		}
2631	} else {
2632		copy_last_highmem_page();
2633		/* Restore page key for data page (s390 only). */
2634		page_key_write(handle->buffer);
2635		hibernate_restore_protect_page(handle->buffer);
2636		handle->buffer = get_buffer(&orig_bm, &ca);
2637		if (IS_ERR(handle->buffer))
2638			return PTR_ERR(handle->buffer);
2639		if (handle->buffer != buffer)
2640			handle->sync_read = 0;
2641	}
2642	handle->cur++;
2643	return PAGE_SIZE;
2644}
2645
2646/**
2647 * snapshot_write_finalize - Complete the loading of a hibernation image.
2648 *
2649 * Must be called after the last call to snapshot_write_next() in case the last
2650 * page in the image happens to be a highmem page and its contents should be
2651 * stored in highmem.  Additionally, it recycles bitmap memory that's not
2652 * necessary any more.
2653 */
2654void snapshot_write_finalize(struct snapshot_handle *handle)
2655{
2656	copy_last_highmem_page();
2657	/* Restore page key for data page (s390 only). */
2658	page_key_write(handle->buffer);
2659	page_key_free();
2660	hibernate_restore_protect_page(handle->buffer);
2661	/* Do that only if we have loaded the image entirely */
2662	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2663		memory_bm_recycle(&orig_bm);
2664		free_highmem_data();
2665	}
2666}
2667
2668int snapshot_image_loaded(struct snapshot_handle *handle)
2669{
2670	return !(!nr_copy_pages || !last_highmem_page_copied() ||
2671			handle->cur <= nr_meta_pages + nr_copy_pages);
2672}
2673
2674#ifdef CONFIG_HIGHMEM
2675/* Assumes that @buf is ready and points to a "safe" page */
2676static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2677				       void *buf)
2678{
2679	void *kaddr1, *kaddr2;
2680
2681	kaddr1 = kmap_atomic(p1);
2682	kaddr2 = kmap_atomic(p2);
2683	copy_page(buf, kaddr1);
2684	copy_page(kaddr1, kaddr2);
2685	copy_page(kaddr2, buf);
2686	kunmap_atomic(kaddr2);
2687	kunmap_atomic(kaddr1);
2688}
2689
2690/**
2691 * restore_highmem - Put highmem image pages into their original locations.
2692 *
2693 * For each highmem page that was in use before hibernation and is included in
2694 * the image, and also has been allocated by the "restore" kernel, swap its
2695 * current contents with the previous (ie. "before hibernation") ones.
2696 *
2697 * If the restore eventually fails, we can call this function once again and
2698 * restore the highmem state as seen by the restore kernel.
2699 */
2700int restore_highmem(void)
2701{
2702	struct highmem_pbe *pbe = highmem_pblist;
2703	void *buf;
2704
2705	if (!pbe)
2706		return 0;
2707
2708	buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2709	if (!buf)
2710		return -ENOMEM;
2711
2712	while (pbe) {
2713		swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2714		pbe = pbe->next;
2715	}
2716	free_image_page(buf, PG_UNSAFE_CLEAR);
2717	return 0;
2718}
2719#endif /* CONFIG_HIGHMEM */