Linux Audio

Check our new training course

Loading...
v6.2
   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
  79
  80/*
  81 * The calls to set_direct_map_*() should not fail because remapping a page
  82 * here means that we only update protection bits in an existing PTE.
  83 * It is still worth to have a warning here if something changes and this
  84 * will no longer be the case.
  85 */
  86static inline void hibernate_map_page(struct page *page)
  87{
  88	if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
  89		int ret = set_direct_map_default_noflush(page);
  90
  91		if (ret)
  92			pr_warn_once("Failed to remap page\n");
  93	} else {
  94		debug_pagealloc_map_pages(page, 1);
  95	}
  96}
  97
  98static inline void hibernate_unmap_page(struct page *page)
  99{
 100	if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
 101		unsigned long addr = (unsigned long)page_address(page);
 102		int ret  = set_direct_map_invalid_noflush(page);
 103
 104		if (ret)
 105			pr_warn_once("Failed to remap page\n");
 106
 107		flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
 108	} else {
 109		debug_pagealloc_unmap_pages(page, 1);
 110	}
 111}
 112
 113static int swsusp_page_is_free(struct page *);
 114static void swsusp_set_page_forbidden(struct page *);
 115static void swsusp_unset_page_forbidden(struct page *);
 116
 117/*
 118 * Number of bytes to reserve for memory allocations made by device drivers
 119 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
 120 * cause image creation to fail (tunable via /sys/power/reserved_size).
 121 */
 122unsigned long reserved_size;
 123
 124void __init hibernate_reserved_size_init(void)
 125{
 126	reserved_size = SPARE_PAGES * PAGE_SIZE;
 127}
 128
 129/*
 130 * Preferred image size in bytes (tunable via /sys/power/image_size).
 131 * When it is set to N, swsusp will do its best to ensure the image
 132 * size will not exceed N bytes, but if that is impossible, it will
 133 * try to create the smallest image possible.
 134 */
 135unsigned long image_size;
 136
 137void __init hibernate_image_size_init(void)
 138{
 139	image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
 140}
 141
 142/*
 143 * List of PBEs needed for restoring the pages that were allocated before
 144 * the suspend and included in the suspend image, but have also been
 145 * allocated by the "resume" kernel, so their contents cannot be written
 146 * directly to their "original" page frames.
 147 */
 148struct pbe *restore_pblist;
 149
 150/* struct linked_page is used to build chains of pages */
 151
 152#define LINKED_PAGE_DATA_SIZE	(PAGE_SIZE - sizeof(void *))
 153
 154struct linked_page {
 155	struct linked_page *next;
 156	char data[LINKED_PAGE_DATA_SIZE];
 157} __packed;
 158
 159/*
 160 * List of "safe" pages (ie. pages that were not used by the image kernel
 161 * before hibernation) that may be used as temporary storage for image kernel
 162 * memory contents.
 163 */
 164static struct linked_page *safe_pages_list;
 165
 166/* Pointer to an auxiliary buffer (1 page) */
 167static void *buffer;
 168
 169#define PG_ANY		0
 170#define PG_SAFE		1
 171#define PG_UNSAFE_CLEAR	1
 172#define PG_UNSAFE_KEEP	0
 173
 174static unsigned int allocated_unsafe_pages;
 175
 176/**
 177 * get_image_page - Allocate a page for a hibernation image.
 178 * @gfp_mask: GFP mask for the allocation.
 179 * @safe_needed: Get pages that were not used before hibernation (restore only)
 180 *
 181 * During image restoration, for storing the PBE list and the image data, we can
 182 * only use memory pages that do not conflict with the pages used before
 183 * hibernation.  The "unsafe" pages have PageNosaveFree set and we count them
 184 * using allocated_unsafe_pages.
 185 *
 186 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
 187 * swsusp_free() can release it.
 188 */
 189static void *get_image_page(gfp_t gfp_mask, int safe_needed)
 190{
 191	void *res;
 192
 193	res = (void *)get_zeroed_page(gfp_mask);
 194	if (safe_needed)
 195		while (res && swsusp_page_is_free(virt_to_page(res))) {
 196			/* The page is unsafe, mark it for swsusp_free() */
 197			swsusp_set_page_forbidden(virt_to_page(res));
 198			allocated_unsafe_pages++;
 199			res = (void *)get_zeroed_page(gfp_mask);
 200		}
 201	if (res) {
 202		swsusp_set_page_forbidden(virt_to_page(res));
 203		swsusp_set_page_free(virt_to_page(res));
 204	}
 205	return res;
 206}
 207
 208static void *__get_safe_page(gfp_t gfp_mask)
 209{
 210	if (safe_pages_list) {
 211		void *ret = safe_pages_list;
 212
 213		safe_pages_list = safe_pages_list->next;
 214		memset(ret, 0, PAGE_SIZE);
 215		return ret;
 216	}
 217	return get_image_page(gfp_mask, PG_SAFE);
 218}
 219
 220unsigned long get_safe_page(gfp_t gfp_mask)
 221{
 222	return (unsigned long)__get_safe_page(gfp_mask);
 223}
 224
 225static struct page *alloc_image_page(gfp_t gfp_mask)
 226{
 227	struct page *page;
 228
 229	page = alloc_page(gfp_mask);
 230	if (page) {
 231		swsusp_set_page_forbidden(page);
 232		swsusp_set_page_free(page);
 233	}
 234	return page;
 235}
 236
 237static void recycle_safe_page(void *page_address)
 238{
 239	struct linked_page *lp = page_address;
 240
 241	lp->next = safe_pages_list;
 242	safe_pages_list = lp;
 243}
 244
 245/**
 246 * free_image_page - Free a page allocated for hibernation image.
 247 * @addr: Address of the page to free.
 248 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
 249 *
 250 * The page to free should have been allocated by get_image_page() (page flags
 251 * set by it are affected).
 252 */
 253static inline void free_image_page(void *addr, int clear_nosave_free)
 254{
 255	struct page *page;
 256
 257	BUG_ON(!virt_addr_valid(addr));
 258
 259	page = virt_to_page(addr);
 260
 261	swsusp_unset_page_forbidden(page);
 262	if (clear_nosave_free)
 263		swsusp_unset_page_free(page);
 264
 265	__free_page(page);
 266}
 267
 268static inline void free_list_of_pages(struct linked_page *list,
 269				      int clear_page_nosave)
 270{
 271	while (list) {
 272		struct linked_page *lp = list->next;
 273
 274		free_image_page(list, clear_page_nosave);
 275		list = lp;
 276	}
 277}
 278
 279/*
 280 * struct chain_allocator is used for allocating small objects out of
 281 * a linked list of pages called 'the chain'.
 282 *
 283 * The chain grows each time when there is no room for a new object in
 284 * the current page.  The allocated objects cannot be freed individually.
 285 * It is only possible to free them all at once, by freeing the entire
 286 * chain.
 287 *
 288 * NOTE: The chain allocator may be inefficient if the allocated objects
 289 * are not much smaller than PAGE_SIZE.
 290 */
 291struct chain_allocator {
 292	struct linked_page *chain;	/* the chain */
 293	unsigned int used_space;	/* total size of objects allocated out
 294					   of the current page */
 295	gfp_t gfp_mask;		/* mask for allocating pages */
 296	int safe_needed;	/* if set, only "safe" pages are allocated */
 297};
 298
 299static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
 300		       int safe_needed)
 301{
 302	ca->chain = NULL;
 303	ca->used_space = LINKED_PAGE_DATA_SIZE;
 304	ca->gfp_mask = gfp_mask;
 305	ca->safe_needed = safe_needed;
 306}
 307
 308static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
 309{
 310	void *ret;
 311
 312	if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
 313		struct linked_page *lp;
 314
 315		lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
 316					get_image_page(ca->gfp_mask, PG_ANY);
 317		if (!lp)
 318			return NULL;
 319
 320		lp->next = ca->chain;
 321		ca->chain = lp;
 322		ca->used_space = 0;
 323	}
 324	ret = ca->chain->data + ca->used_space;
 325	ca->used_space += size;
 326	return ret;
 327}
 328
 329/*
 330 * Data types related to memory bitmaps.
 331 *
 332 * Memory bitmap is a structure consisting of many linked lists of
 333 * objects.  The main list's elements are of type struct zone_bitmap
 334 * and each of them corresponds to one zone.  For each zone bitmap
 335 * object there is a list of objects of type struct bm_block that
 336 * represent each blocks of bitmap in which information is stored.
 337 *
 338 * struct memory_bitmap contains a pointer to the main list of zone
 339 * bitmap objects, a struct bm_position used for browsing the bitmap,
 340 * and a pointer to the list of pages used for allocating all of the
 341 * zone bitmap objects and bitmap block objects.
 342 *
 343 * NOTE: It has to be possible to lay out the bitmap in memory
 344 * using only allocations of order 0.  Additionally, the bitmap is
 345 * designed to work with arbitrary number of zones (this is over the
 346 * top for now, but let's avoid making unnecessary assumptions ;-).
 347 *
 348 * struct zone_bitmap contains a pointer to a list of bitmap block
 349 * objects and a pointer to the bitmap block object that has been
 350 * most recently used for setting bits.  Additionally, it contains the
 351 * PFNs that correspond to the start and end of the represented zone.
 352 *
 353 * struct bm_block contains a pointer to the memory page in which
 354 * information is stored (in the form of a block of bitmap)
 355 * It also contains the pfns that correspond to the start and end of
 356 * the represented memory area.
 357 *
 358 * The memory bitmap is organized as a radix tree to guarantee fast random
 359 * access to the bits. There is one radix tree for each zone (as returned
 360 * from create_mem_extents).
 361 *
 362 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
 363 * two linked lists for the nodes of the tree, one for the inner nodes and
 364 * one for the leave nodes. The linked leave nodes are used for fast linear
 365 * access of the memory bitmap.
 366 *
 367 * The struct rtree_node represents one node of the radix tree.
 368 */
 369
 370#define BM_END_OF_MAP	(~0UL)
 371
 372#define BM_BITS_PER_BLOCK	(PAGE_SIZE * BITS_PER_BYTE)
 373#define BM_BLOCK_SHIFT		(PAGE_SHIFT + 3)
 374#define BM_BLOCK_MASK		((1UL << BM_BLOCK_SHIFT) - 1)
 375
 376/*
 377 * struct rtree_node is a wrapper struct to link the nodes
 378 * of the rtree together for easy linear iteration over
 379 * bits and easy freeing
 380 */
 381struct rtree_node {
 382	struct list_head list;
 383	unsigned long *data;
 384};
 385
 386/*
 387 * struct mem_zone_bm_rtree represents a bitmap used for one
 388 * populated memory zone.
 389 */
 390struct mem_zone_bm_rtree {
 391	struct list_head list;		/* Link Zones together         */
 392	struct list_head nodes;		/* Radix Tree inner nodes      */
 393	struct list_head leaves;	/* Radix Tree leaves           */
 394	unsigned long start_pfn;	/* Zone start page frame       */
 395	unsigned long end_pfn;		/* Zone end page frame + 1     */
 396	struct rtree_node *rtree;	/* Radix Tree Root             */
 397	int levels;			/* Number of Radix Tree Levels */
 398	unsigned int blocks;		/* Number of Bitmap Blocks     */
 399};
 400
 401/* strcut bm_position is used for browsing memory bitmaps */
 402
 403struct bm_position {
 404	struct mem_zone_bm_rtree *zone;
 405	struct rtree_node *node;
 406	unsigned long node_pfn;
 407	int node_bit;
 408};
 409
 410struct memory_bitmap {
 411	struct list_head zones;
 412	struct linked_page *p_list;	/* list of pages used to store zone
 413					   bitmap objects and bitmap block
 414					   objects */
 415	struct bm_position cur;	/* most recently used bit position */
 416};
 417
 418/* Functions that operate on memory bitmaps */
 419
 420#define BM_ENTRIES_PER_LEVEL	(PAGE_SIZE / sizeof(unsigned long))
 421#if BITS_PER_LONG == 32
 422#define BM_RTREE_LEVEL_SHIFT	(PAGE_SHIFT - 2)
 423#else
 424#define BM_RTREE_LEVEL_SHIFT	(PAGE_SHIFT - 3)
 425#endif
 426#define BM_RTREE_LEVEL_MASK	((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
 427
 428/**
 429 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
 430 * @gfp_mask: GFP mask for the allocation.
 431 * @safe_needed: Get pages not used before hibernation (restore only)
 432 * @ca: Pointer to a linked list of pages ("a chain") to allocate from
 433 * @list: Radix Tree node to add.
 434 *
 435 * This function is used to allocate inner nodes as well as the
 436 * leave nodes of the radix tree. It also adds the node to the
 437 * corresponding linked list passed in by the *list parameter.
 438 */
 439static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
 440					   struct chain_allocator *ca,
 441					   struct list_head *list)
 442{
 443	struct rtree_node *node;
 444
 445	node = chain_alloc(ca, sizeof(struct rtree_node));
 446	if (!node)
 447		return NULL;
 448
 449	node->data = get_image_page(gfp_mask, safe_needed);
 450	if (!node->data)
 451		return NULL;
 452
 453	list_add_tail(&node->list, list);
 454
 455	return node;
 456}
 457
 458/**
 459 * add_rtree_block - Add a new leave node to the radix tree.
 460 *
 461 * The leave nodes need to be allocated in order to keep the leaves
 462 * linked list in order. This is guaranteed by the zone->blocks
 463 * counter.
 464 */
 465static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
 466			   int safe_needed, struct chain_allocator *ca)
 467{
 468	struct rtree_node *node, *block, **dst;
 469	unsigned int levels_needed, block_nr;
 470	int i;
 471
 472	block_nr = zone->blocks;
 473	levels_needed = 0;
 474
 475	/* How many levels do we need for this block nr? */
 476	while (block_nr) {
 477		levels_needed += 1;
 478		block_nr >>= BM_RTREE_LEVEL_SHIFT;
 479	}
 480
 481	/* Make sure the rtree has enough levels */
 482	for (i = zone->levels; i < levels_needed; i++) {
 483		node = alloc_rtree_node(gfp_mask, safe_needed, ca,
 484					&zone->nodes);
 485		if (!node)
 486			return -ENOMEM;
 487
 488		node->data[0] = (unsigned long)zone->rtree;
 489		zone->rtree = node;
 490		zone->levels += 1;
 491	}
 492
 493	/* Allocate new block */
 494	block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
 495	if (!block)
 496		return -ENOMEM;
 497
 498	/* Now walk the rtree to insert the block */
 499	node = zone->rtree;
 500	dst = &zone->rtree;
 501	block_nr = zone->blocks;
 502	for (i = zone->levels; i > 0; i--) {
 503		int index;
 504
 505		if (!node) {
 506			node = alloc_rtree_node(gfp_mask, safe_needed, ca,
 507						&zone->nodes);
 508			if (!node)
 509				return -ENOMEM;
 510			*dst = node;
 511		}
 512
 513		index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
 514		index &= BM_RTREE_LEVEL_MASK;
 515		dst = (struct rtree_node **)&((*dst)->data[index]);
 516		node = *dst;
 517	}
 518
 519	zone->blocks += 1;
 520	*dst = block;
 521
 522	return 0;
 523}
 524
 525static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
 526			       int clear_nosave_free);
 527
 528/**
 529 * create_zone_bm_rtree - Create a radix tree for one zone.
 530 *
 531 * Allocated the mem_zone_bm_rtree structure and initializes it.
 532 * This function also allocated and builds the radix tree for the
 533 * zone.
 534 */
 535static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
 536						      int safe_needed,
 537						      struct chain_allocator *ca,
 538						      unsigned long start,
 539						      unsigned long end)
 540{
 541	struct mem_zone_bm_rtree *zone;
 542	unsigned int i, nr_blocks;
 543	unsigned long pages;
 544
 545	pages = end - start;
 546	zone  = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
 547	if (!zone)
 548		return NULL;
 549
 550	INIT_LIST_HEAD(&zone->nodes);
 551	INIT_LIST_HEAD(&zone->leaves);
 552	zone->start_pfn = start;
 553	zone->end_pfn = end;
 554	nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
 555
 556	for (i = 0; i < nr_blocks; i++) {
 557		if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
 558			free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
 559			return NULL;
 560		}
 561	}
 562
 563	return zone;
 564}
 565
 566/**
 567 * free_zone_bm_rtree - Free the memory of the radix tree.
 568 *
 569 * Free all node pages of the radix tree. The mem_zone_bm_rtree
 570 * structure itself is not freed here nor are the rtree_node
 571 * structs.
 572 */
 573static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
 574			       int clear_nosave_free)
 575{
 576	struct rtree_node *node;
 577
 578	list_for_each_entry(node, &zone->nodes, list)
 579		free_image_page(node->data, clear_nosave_free);
 580
 581	list_for_each_entry(node, &zone->leaves, list)
 582		free_image_page(node->data, clear_nosave_free);
 583}
 584
 585static void memory_bm_position_reset(struct memory_bitmap *bm)
 586{
 587	bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
 588				  list);
 589	bm->cur.node = list_entry(bm->cur.zone->leaves.next,
 590				  struct rtree_node, list);
 591	bm->cur.node_pfn = 0;
 592	bm->cur.node_bit = 0;
 593}
 594
 595static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
 596
 597struct mem_extent {
 598	struct list_head hook;
 599	unsigned long start;
 600	unsigned long end;
 601};
 602
 603/**
 604 * free_mem_extents - Free a list of memory extents.
 605 * @list: List of extents to free.
 606 */
 607static void free_mem_extents(struct list_head *list)
 608{
 609	struct mem_extent *ext, *aux;
 610
 611	list_for_each_entry_safe(ext, aux, list, hook) {
 612		list_del(&ext->hook);
 613		kfree(ext);
 614	}
 615}
 616
 617/**
 618 * create_mem_extents - Create a list of memory extents.
 619 * @list: List to put the extents into.
 620 * @gfp_mask: Mask to use for memory allocations.
 621 *
 622 * The extents represent contiguous ranges of PFNs.
 623 */
 624static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
 625{
 626	struct zone *zone;
 627
 628	INIT_LIST_HEAD(list);
 629
 630	for_each_populated_zone(zone) {
 631		unsigned long zone_start, zone_end;
 632		struct mem_extent *ext, *cur, *aux;
 633
 634		zone_start = zone->zone_start_pfn;
 635		zone_end = zone_end_pfn(zone);
 636
 637		list_for_each_entry(ext, list, hook)
 638			if (zone_start <= ext->end)
 639				break;
 640
 641		if (&ext->hook == list || zone_end < ext->start) {
 642			/* New extent is necessary */
 643			struct mem_extent *new_ext;
 644
 645			new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
 646			if (!new_ext) {
 647				free_mem_extents(list);
 648				return -ENOMEM;
 649			}
 650			new_ext->start = zone_start;
 651			new_ext->end = zone_end;
 652			list_add_tail(&new_ext->hook, &ext->hook);
 653			continue;
 654		}
 655
 656		/* Merge this zone's range of PFNs with the existing one */
 657		if (zone_start < ext->start)
 658			ext->start = zone_start;
 659		if (zone_end > ext->end)
 660			ext->end = zone_end;
 661
 662		/* More merging may be possible */
 663		cur = ext;
 664		list_for_each_entry_safe_continue(cur, aux, list, hook) {
 665			if (zone_end < cur->start)
 666				break;
 667			if (zone_end < cur->end)
 668				ext->end = cur->end;
 669			list_del(&cur->hook);
 670			kfree(cur);
 671		}
 672	}
 673
 674	return 0;
 675}
 676
 677/**
 678 * memory_bm_create - Allocate memory for a memory bitmap.
 679 */
 680static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
 681			    int safe_needed)
 682{
 683	struct chain_allocator ca;
 684	struct list_head mem_extents;
 685	struct mem_extent *ext;
 686	int error;
 687
 688	chain_init(&ca, gfp_mask, safe_needed);
 689	INIT_LIST_HEAD(&bm->zones);
 690
 691	error = create_mem_extents(&mem_extents, gfp_mask);
 692	if (error)
 693		return error;
 694
 695	list_for_each_entry(ext, &mem_extents, hook) {
 696		struct mem_zone_bm_rtree *zone;
 697
 698		zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
 699					    ext->start, ext->end);
 700		if (!zone) {
 701			error = -ENOMEM;
 702			goto Error;
 703		}
 704		list_add_tail(&zone->list, &bm->zones);
 705	}
 706
 707	bm->p_list = ca.chain;
 708	memory_bm_position_reset(bm);
 709 Exit:
 710	free_mem_extents(&mem_extents);
 711	return error;
 712
 713 Error:
 714	bm->p_list = ca.chain;
 715	memory_bm_free(bm, PG_UNSAFE_CLEAR);
 716	goto Exit;
 717}
 718
 719/**
 720 * memory_bm_free - Free memory occupied by the memory bitmap.
 721 * @bm: Memory bitmap.
 722 */
 723static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
 724{
 725	struct mem_zone_bm_rtree *zone;
 726
 727	list_for_each_entry(zone, &bm->zones, list)
 728		free_zone_bm_rtree(zone, clear_nosave_free);
 729
 730	free_list_of_pages(bm->p_list, clear_nosave_free);
 731
 732	INIT_LIST_HEAD(&bm->zones);
 733}
 734
 735/**
 736 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
 737 *
 738 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
 739 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
 740 *
 741 * Walk the radix tree to find the page containing the bit that represents @pfn
 742 * and return the position of the bit in @addr and @bit_nr.
 743 */
 744static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
 745			      void **addr, unsigned int *bit_nr)
 746{
 747	struct mem_zone_bm_rtree *curr, *zone;
 748	struct rtree_node *node;
 749	int i, block_nr;
 750
 751	zone = bm->cur.zone;
 752
 753	if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
 754		goto zone_found;
 755
 756	zone = NULL;
 757
 758	/* Find the right zone */
 759	list_for_each_entry(curr, &bm->zones, list) {
 760		if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
 761			zone = curr;
 762			break;
 763		}
 764	}
 765
 766	if (!zone)
 767		return -EFAULT;
 768
 769zone_found:
 770	/*
 771	 * We have found the zone. Now walk the radix tree to find the leaf node
 772	 * for our PFN.
 773	 */
 774
 775	/*
 776	 * If the zone we wish to scan is the current zone and the
 777	 * pfn falls into the current node then we do not need to walk
 778	 * the tree.
 779	 */
 780	node = bm->cur.node;
 781	if (zone == bm->cur.zone &&
 782	    ((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
 783		goto node_found;
 784
 785	node      = zone->rtree;
 786	block_nr  = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
 787
 788	for (i = zone->levels; i > 0; i--) {
 789		int index;
 790
 791		index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
 792		index &= BM_RTREE_LEVEL_MASK;
 793		BUG_ON(node->data[index] == 0);
 794		node = (struct rtree_node *)node->data[index];
 795	}
 796
 797node_found:
 798	/* Update last position */
 799	bm->cur.zone = zone;
 800	bm->cur.node = node;
 801	bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
 802
 803	/* Set return values */
 804	*addr = node->data;
 805	*bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
 806
 807	return 0;
 808}
 809
 810static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
 811{
 812	void *addr;
 813	unsigned int bit;
 814	int error;
 815
 816	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 817	BUG_ON(error);
 818	set_bit(bit, addr);
 819}
 820
 821static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
 822{
 823	void *addr;
 824	unsigned int bit;
 825	int error;
 826
 827	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 828	if (!error)
 829		set_bit(bit, addr);
 830
 831	return error;
 832}
 833
 834static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
 835{
 836	void *addr;
 837	unsigned int bit;
 838	int error;
 839
 840	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 841	BUG_ON(error);
 842	clear_bit(bit, addr);
 843}
 844
 845static void memory_bm_clear_current(struct memory_bitmap *bm)
 846{
 847	int bit;
 848
 849	bit = max(bm->cur.node_bit - 1, 0);
 850	clear_bit(bit, bm->cur.node->data);
 851}
 852
 853static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
 854{
 855	void *addr;
 856	unsigned int bit;
 857	int error;
 858
 859	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 860	BUG_ON(error);
 861	return test_bit(bit, addr);
 862}
 863
 864static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
 865{
 866	void *addr;
 867	unsigned int bit;
 868
 869	return !memory_bm_find_bit(bm, pfn, &addr, &bit);
 870}
 871
 872/*
 873 * rtree_next_node - Jump to the next leaf node.
 874 *
 875 * Set the position to the beginning of the next node in the
 876 * memory bitmap. This is either the next node in the current
 877 * zone's radix tree or the first node in the radix tree of the
 878 * next zone.
 879 *
 880 * Return true if there is a next node, false otherwise.
 881 */
 882static bool rtree_next_node(struct memory_bitmap *bm)
 883{
 884	if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
 885		bm->cur.node = list_entry(bm->cur.node->list.next,
 886					  struct rtree_node, list);
 887		bm->cur.node_pfn += BM_BITS_PER_BLOCK;
 888		bm->cur.node_bit  = 0;
 889		touch_softlockup_watchdog();
 890		return true;
 891	}
 892
 893	/* No more nodes, goto next zone */
 894	if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
 895		bm->cur.zone = list_entry(bm->cur.zone->list.next,
 896				  struct mem_zone_bm_rtree, list);
 897		bm->cur.node = list_entry(bm->cur.zone->leaves.next,
 898					  struct rtree_node, list);
 899		bm->cur.node_pfn = 0;
 900		bm->cur.node_bit = 0;
 901		return true;
 902	}
 903
 904	/* No more zones */
 905	return false;
 906}
 907
 908/**
 909 * memory_bm_next_pfn - Find the next set bit in a memory bitmap.
 910 * @bm: Memory bitmap.
 911 *
 912 * Starting from the last returned position this function searches for the next
 913 * set bit in @bm and returns the PFN represented by it.  If no more bits are
 914 * set, BM_END_OF_MAP is returned.
 915 *
 916 * It is required to run memory_bm_position_reset() before the first call to
 917 * this function for the given memory bitmap.
 918 */
 919static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
 920{
 921	unsigned long bits, pfn, pages;
 922	int bit;
 923
 924	do {
 925		pages	  = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
 926		bits      = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
 927		bit	  = find_next_bit(bm->cur.node->data, bits,
 928					  bm->cur.node_bit);
 929		if (bit < bits) {
 930			pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
 931			bm->cur.node_bit = bit + 1;
 932			return pfn;
 933		}
 934	} while (rtree_next_node(bm));
 935
 936	return BM_END_OF_MAP;
 937}
 938
 939/*
 940 * This structure represents a range of page frames the contents of which
 941 * should not be saved during hibernation.
 942 */
 943struct nosave_region {
 944	struct list_head list;
 945	unsigned long start_pfn;
 946	unsigned long end_pfn;
 947};
 948
 949static LIST_HEAD(nosave_regions);
 950
 951static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
 952{
 953	struct rtree_node *node;
 954
 955	list_for_each_entry(node, &zone->nodes, list)
 956		recycle_safe_page(node->data);
 957
 958	list_for_each_entry(node, &zone->leaves, list)
 959		recycle_safe_page(node->data);
 960}
 961
 962static void memory_bm_recycle(struct memory_bitmap *bm)
 963{
 964	struct mem_zone_bm_rtree *zone;
 965	struct linked_page *p_list;
 966
 967	list_for_each_entry(zone, &bm->zones, list)
 968		recycle_zone_bm_rtree(zone);
 969
 970	p_list = bm->p_list;
 971	while (p_list) {
 972		struct linked_page *lp = p_list;
 973
 974		p_list = lp->next;
 975		recycle_safe_page(lp);
 976	}
 977}
 978
 979/**
 980 * register_nosave_region - Register a region of unsaveable memory.
 981 *
 982 * Register a range of page frames the contents of which should not be saved
 983 * during hibernation (to be used in the early initialization code).
 984 */
 985void __init register_nosave_region(unsigned long start_pfn, unsigned long end_pfn)
 
 986{
 987	struct nosave_region *region;
 988
 989	if (start_pfn >= end_pfn)
 990		return;
 991
 992	if (!list_empty(&nosave_regions)) {
 993		/* Try to extend the previous region (they should be sorted) */
 994		region = list_entry(nosave_regions.prev,
 995					struct nosave_region, list);
 996		if (region->end_pfn == start_pfn) {
 997			region->end_pfn = end_pfn;
 998			goto Report;
 999		}
1000	}
1001	/* This allocation cannot fail */
1002	region = memblock_alloc(sizeof(struct nosave_region),
1003				SMP_CACHE_BYTES);
1004	if (!region)
1005		panic("%s: Failed to allocate %zu bytes\n", __func__,
1006		      sizeof(struct nosave_region));
 
 
 
 
 
 
1007	region->start_pfn = start_pfn;
1008	region->end_pfn = end_pfn;
1009	list_add_tail(&region->list, &nosave_regions);
1010 Report:
1011	pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
1012		(unsigned long long) start_pfn << PAGE_SHIFT,
1013		((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1014}
1015
1016/*
1017 * Set bits in this map correspond to the page frames the contents of which
1018 * should not be saved during the suspend.
1019 */
1020static struct memory_bitmap *forbidden_pages_map;
1021
1022/* Set bits in this map correspond to free page frames. */
1023static struct memory_bitmap *free_pages_map;
1024
1025/*
1026 * Each page frame allocated for creating the image is marked by setting the
1027 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
1028 */
1029
1030void swsusp_set_page_free(struct page *page)
1031{
1032	if (free_pages_map)
1033		memory_bm_set_bit(free_pages_map, page_to_pfn(page));
1034}
1035
1036static int swsusp_page_is_free(struct page *page)
1037{
1038	return free_pages_map ?
1039		memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1040}
1041
1042void swsusp_unset_page_free(struct page *page)
1043{
1044	if (free_pages_map)
1045		memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1046}
1047
1048static void swsusp_set_page_forbidden(struct page *page)
1049{
1050	if (forbidden_pages_map)
1051		memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1052}
1053
1054int swsusp_page_is_forbidden(struct page *page)
1055{
1056	return forbidden_pages_map ?
1057		memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1058}
1059
1060static void swsusp_unset_page_forbidden(struct page *page)
1061{
1062	if (forbidden_pages_map)
1063		memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1064}
1065
1066/**
1067 * mark_nosave_pages - Mark pages that should not be saved.
1068 * @bm: Memory bitmap.
1069 *
1070 * Set the bits in @bm that correspond to the page frames the contents of which
1071 * should not be saved.
1072 */
1073static void mark_nosave_pages(struct memory_bitmap *bm)
1074{
1075	struct nosave_region *region;
1076
1077	if (list_empty(&nosave_regions))
1078		return;
1079
1080	list_for_each_entry(region, &nosave_regions, list) {
1081		unsigned long pfn;
1082
1083		pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1084			 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1085			 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1086				- 1);
1087
1088		for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1089			if (pfn_valid(pfn)) {
1090				/*
1091				 * It is safe to ignore the result of
1092				 * mem_bm_set_bit_check() here, since we won't
1093				 * touch the PFNs for which the error is
1094				 * returned anyway.
1095				 */
1096				mem_bm_set_bit_check(bm, pfn);
1097			}
1098	}
1099}
1100
1101/**
1102 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1103 *
1104 * Create bitmaps needed for marking page frames that should not be saved and
1105 * free page frames.  The forbidden_pages_map and free_pages_map pointers are
1106 * only modified if everything goes well, because we don't want the bits to be
1107 * touched before both bitmaps are set up.
1108 */
1109int create_basic_memory_bitmaps(void)
1110{
1111	struct memory_bitmap *bm1, *bm2;
1112	int error = 0;
1113
1114	if (forbidden_pages_map && free_pages_map)
1115		return 0;
1116	else
1117		BUG_ON(forbidden_pages_map || free_pages_map);
1118
1119	bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1120	if (!bm1)
1121		return -ENOMEM;
1122
1123	error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1124	if (error)
1125		goto Free_first_object;
1126
1127	bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1128	if (!bm2)
1129		goto Free_first_bitmap;
1130
1131	error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1132	if (error)
1133		goto Free_second_object;
1134
1135	forbidden_pages_map = bm1;
1136	free_pages_map = bm2;
1137	mark_nosave_pages(forbidden_pages_map);
1138
1139	pr_debug("Basic memory bitmaps created\n");
1140
1141	return 0;
1142
1143 Free_second_object:
1144	kfree(bm2);
1145 Free_first_bitmap:
1146	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1147 Free_first_object:
1148	kfree(bm1);
1149	return -ENOMEM;
1150}
1151
1152/**
1153 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1154 *
1155 * Free memory bitmaps allocated by create_basic_memory_bitmaps().  The
1156 * auxiliary pointers are necessary so that the bitmaps themselves are not
1157 * referred to while they are being freed.
1158 */
1159void free_basic_memory_bitmaps(void)
1160{
1161	struct memory_bitmap *bm1, *bm2;
1162
1163	if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1164		return;
1165
1166	bm1 = forbidden_pages_map;
1167	bm2 = free_pages_map;
1168	forbidden_pages_map = NULL;
1169	free_pages_map = NULL;
1170	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1171	kfree(bm1);
1172	memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1173	kfree(bm2);
1174
1175	pr_debug("Basic memory bitmaps freed\n");
1176}
1177
1178static void clear_or_poison_free_page(struct page *page)
1179{
1180	if (page_poisoning_enabled_static())
1181		__kernel_poison_pages(page, 1);
1182	else if (want_init_on_free())
1183		clear_highpage(page);
1184}
1185
1186void clear_or_poison_free_pages(void)
1187{
1188	struct memory_bitmap *bm = free_pages_map;
1189	unsigned long pfn;
1190
1191	if (WARN_ON(!(free_pages_map)))
1192		return;
1193
1194	if (page_poisoning_enabled() || want_init_on_free()) {
1195		memory_bm_position_reset(bm);
1196		pfn = memory_bm_next_pfn(bm);
1197		while (pfn != BM_END_OF_MAP) {
1198			if (pfn_valid(pfn))
1199				clear_or_poison_free_page(pfn_to_page(pfn));
1200
1201			pfn = memory_bm_next_pfn(bm);
1202		}
1203		memory_bm_position_reset(bm);
1204		pr_info("free pages cleared after restore\n");
1205	}
1206}
1207
1208/**
1209 * snapshot_additional_pages - Estimate the number of extra pages needed.
1210 * @zone: Memory zone to carry out the computation for.
1211 *
1212 * Estimate the number of additional pages needed for setting up a hibernation
1213 * image data structures for @zone (usually, the returned value is greater than
1214 * the exact number).
1215 */
1216unsigned int snapshot_additional_pages(struct zone *zone)
1217{
1218	unsigned int rtree, nodes;
1219
1220	rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1221	rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1222			      LINKED_PAGE_DATA_SIZE);
1223	while (nodes > 1) {
1224		nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1225		rtree += nodes;
1226	}
1227
1228	return 2 * rtree;
1229}
1230
1231#ifdef CONFIG_HIGHMEM
1232/**
1233 * count_free_highmem_pages - Compute the total number of free highmem pages.
1234 *
1235 * The returned number is system-wide.
1236 */
1237static unsigned int count_free_highmem_pages(void)
1238{
1239	struct zone *zone;
1240	unsigned int cnt = 0;
1241
1242	for_each_populated_zone(zone)
1243		if (is_highmem(zone))
1244			cnt += zone_page_state(zone, NR_FREE_PAGES);
1245
1246	return cnt;
1247}
1248
1249/**
1250 * saveable_highmem_page - Check if a highmem page is saveable.
1251 *
1252 * Determine whether a highmem page should be included in a hibernation image.
1253 *
1254 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1255 * and it isn't part of a free chunk of pages.
1256 */
1257static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1258{
1259	struct page *page;
1260
1261	if (!pfn_valid(pfn))
1262		return NULL;
1263
1264	page = pfn_to_online_page(pfn);
1265	if (!page || page_zone(page) != zone)
1266		return NULL;
1267
1268	BUG_ON(!PageHighMem(page));
1269
1270	if (swsusp_page_is_forbidden(page) ||  swsusp_page_is_free(page))
1271		return NULL;
1272
1273	if (PageReserved(page) || PageOffline(page))
1274		return NULL;
1275
1276	if (page_is_guard(page))
1277		return NULL;
1278
1279	return page;
1280}
1281
1282/**
1283 * count_highmem_pages - Compute the total number of saveable highmem pages.
1284 */
1285static unsigned int count_highmem_pages(void)
1286{
1287	struct zone *zone;
1288	unsigned int n = 0;
1289
1290	for_each_populated_zone(zone) {
1291		unsigned long pfn, max_zone_pfn;
1292
1293		if (!is_highmem(zone))
1294			continue;
1295
1296		mark_free_pages(zone);
1297		max_zone_pfn = zone_end_pfn(zone);
1298		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1299			if (saveable_highmem_page(zone, pfn))
1300				n++;
1301	}
1302	return n;
1303}
1304#else
1305static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1306{
1307	return NULL;
1308}
1309#endif /* CONFIG_HIGHMEM */
1310
1311/**
1312 * saveable_page - Check if the given page is saveable.
1313 *
1314 * Determine whether a non-highmem page should be included in a hibernation
1315 * image.
1316 *
1317 * We should save the page if it isn't Nosave, and is not in the range
1318 * of pages statically defined as 'unsaveable', and it isn't part of
1319 * a free chunk of pages.
1320 */
1321static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1322{
1323	struct page *page;
1324
1325	if (!pfn_valid(pfn))
1326		return NULL;
1327
1328	page = pfn_to_online_page(pfn);
1329	if (!page || page_zone(page) != zone)
1330		return NULL;
1331
1332	BUG_ON(PageHighMem(page));
1333
1334	if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1335		return NULL;
1336
1337	if (PageOffline(page))
1338		return NULL;
1339
1340	if (PageReserved(page)
1341	    && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1342		return NULL;
1343
1344	if (page_is_guard(page))
1345		return NULL;
1346
1347	return page;
1348}
1349
1350/**
1351 * count_data_pages - Compute the total number of saveable non-highmem pages.
1352 */
1353static unsigned int count_data_pages(void)
1354{
1355	struct zone *zone;
1356	unsigned long pfn, max_zone_pfn;
1357	unsigned int n = 0;
1358
1359	for_each_populated_zone(zone) {
1360		if (is_highmem(zone))
1361			continue;
1362
1363		mark_free_pages(zone);
1364		max_zone_pfn = zone_end_pfn(zone);
1365		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1366			if (saveable_page(zone, pfn))
1367				n++;
1368	}
1369	return n;
1370}
1371
1372/*
1373 * This is needed, because copy_page and memcpy are not usable for copying
1374 * task structs.
1375 */
1376static inline void do_copy_page(long *dst, long *src)
1377{
1378	int n;
1379
1380	for (n = PAGE_SIZE / sizeof(long); n; n--)
1381		*dst++ = *src++;
1382}
1383
1384/**
1385 * safe_copy_page - Copy a page in a safe way.
1386 *
1387 * Check if the page we are going to copy is marked as present in the kernel
1388 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1389 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1390 * always returns 'true'.
1391 */
1392static void safe_copy_page(void *dst, struct page *s_page)
1393{
1394	if (kernel_page_present(s_page)) {
1395		do_copy_page(dst, page_address(s_page));
1396	} else {
1397		hibernate_map_page(s_page);
1398		do_copy_page(dst, page_address(s_page));
1399		hibernate_unmap_page(s_page);
1400	}
1401}
1402
1403#ifdef CONFIG_HIGHMEM
1404static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1405{
1406	return is_highmem(zone) ?
1407		saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1408}
1409
1410static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1411{
1412	struct page *s_page, *d_page;
1413	void *src, *dst;
1414
1415	s_page = pfn_to_page(src_pfn);
1416	d_page = pfn_to_page(dst_pfn);
1417	if (PageHighMem(s_page)) {
1418		src = kmap_atomic(s_page);
1419		dst = kmap_atomic(d_page);
1420		do_copy_page(dst, src);
1421		kunmap_atomic(dst);
1422		kunmap_atomic(src);
1423	} else {
1424		if (PageHighMem(d_page)) {
1425			/*
1426			 * The page pointed to by src may contain some kernel
1427			 * data modified by kmap_atomic()
1428			 */
1429			safe_copy_page(buffer, s_page);
1430			dst = kmap_atomic(d_page);
1431			copy_page(dst, buffer);
1432			kunmap_atomic(dst);
1433		} else {
1434			safe_copy_page(page_address(d_page), s_page);
1435		}
1436	}
1437}
1438#else
1439#define page_is_saveable(zone, pfn)	saveable_page(zone, pfn)
1440
1441static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1442{
1443	safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1444				pfn_to_page(src_pfn));
1445}
1446#endif /* CONFIG_HIGHMEM */
1447
1448static void copy_data_pages(struct memory_bitmap *copy_bm,
1449			    struct memory_bitmap *orig_bm)
1450{
1451	struct zone *zone;
1452	unsigned long pfn;
1453
1454	for_each_populated_zone(zone) {
1455		unsigned long max_zone_pfn;
1456
1457		mark_free_pages(zone);
1458		max_zone_pfn = zone_end_pfn(zone);
1459		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1460			if (page_is_saveable(zone, pfn))
1461				memory_bm_set_bit(orig_bm, pfn);
1462	}
1463	memory_bm_position_reset(orig_bm);
1464	memory_bm_position_reset(copy_bm);
1465	for(;;) {
1466		pfn = memory_bm_next_pfn(orig_bm);
1467		if (unlikely(pfn == BM_END_OF_MAP))
1468			break;
1469		copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1470	}
1471}
1472
1473/* Total number of image pages */
1474static unsigned int nr_copy_pages;
1475/* Number of pages needed for saving the original pfns of the image pages */
1476static unsigned int nr_meta_pages;
1477/*
1478 * Numbers of normal and highmem page frames allocated for hibernation image
1479 * before suspending devices.
1480 */
1481static unsigned int alloc_normal, alloc_highmem;
1482/*
1483 * Memory bitmap used for marking saveable pages (during hibernation) or
1484 * hibernation image pages (during restore)
1485 */
1486static struct memory_bitmap orig_bm;
1487/*
1488 * Memory bitmap used during hibernation for marking allocated page frames that
1489 * will contain copies of saveable pages.  During restore it is initially used
1490 * for marking hibernation image pages, but then the set bits from it are
1491 * duplicated in @orig_bm and it is released.  On highmem systems it is next
1492 * used for marking "safe" highmem pages, but it has to be reinitialized for
1493 * this purpose.
1494 */
1495static struct memory_bitmap copy_bm;
1496
1497/**
1498 * swsusp_free - Free pages allocated for hibernation image.
1499 *
1500 * Image pages are allocated before snapshot creation, so they need to be
1501 * released after resume.
1502 */
1503void swsusp_free(void)
1504{
1505	unsigned long fb_pfn, fr_pfn;
1506
1507	if (!forbidden_pages_map || !free_pages_map)
1508		goto out;
1509
1510	memory_bm_position_reset(forbidden_pages_map);
1511	memory_bm_position_reset(free_pages_map);
1512
1513loop:
1514	fr_pfn = memory_bm_next_pfn(free_pages_map);
1515	fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1516
1517	/*
1518	 * Find the next bit set in both bitmaps. This is guaranteed to
1519	 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1520	 */
1521	do {
1522		if (fb_pfn < fr_pfn)
1523			fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1524		if (fr_pfn < fb_pfn)
1525			fr_pfn = memory_bm_next_pfn(free_pages_map);
1526	} while (fb_pfn != fr_pfn);
1527
1528	if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1529		struct page *page = pfn_to_page(fr_pfn);
1530
1531		memory_bm_clear_current(forbidden_pages_map);
1532		memory_bm_clear_current(free_pages_map);
1533		hibernate_restore_unprotect_page(page_address(page));
1534		__free_page(page);
1535		goto loop;
1536	}
1537
1538out:
1539	nr_copy_pages = 0;
1540	nr_meta_pages = 0;
1541	restore_pblist = NULL;
1542	buffer = NULL;
1543	alloc_normal = 0;
1544	alloc_highmem = 0;
1545	hibernate_restore_protection_end();
1546}
1547
1548/* Helper functions used for the shrinking of memory. */
1549
1550#define GFP_IMAGE	(GFP_KERNEL | __GFP_NOWARN)
1551
1552/**
1553 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1554 * @nr_pages: Number of page frames to allocate.
1555 * @mask: GFP flags to use for the allocation.
1556 *
1557 * Return value: Number of page frames actually allocated
1558 */
1559static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1560{
1561	unsigned long nr_alloc = 0;
1562
1563	while (nr_pages > 0) {
1564		struct page *page;
1565
1566		page = alloc_image_page(mask);
1567		if (!page)
1568			break;
1569		memory_bm_set_bit(&copy_bm, page_to_pfn(page));
1570		if (PageHighMem(page))
1571			alloc_highmem++;
1572		else
1573			alloc_normal++;
1574		nr_pages--;
1575		nr_alloc++;
1576	}
1577
1578	return nr_alloc;
1579}
1580
1581static unsigned long preallocate_image_memory(unsigned long nr_pages,
1582					      unsigned long avail_normal)
1583{
1584	unsigned long alloc;
1585
1586	if (avail_normal <= alloc_normal)
1587		return 0;
1588
1589	alloc = avail_normal - alloc_normal;
1590	if (nr_pages < alloc)
1591		alloc = nr_pages;
1592
1593	return preallocate_image_pages(alloc, GFP_IMAGE);
1594}
1595
1596#ifdef CONFIG_HIGHMEM
1597static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1598{
1599	return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1600}
1601
1602/**
1603 *  __fraction - Compute (an approximation of) x * (multiplier / base).
1604 */
1605static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1606{
1607	return div64_u64(x * multiplier, base);
1608}
1609
1610static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1611						  unsigned long highmem,
1612						  unsigned long total)
1613{
1614	unsigned long alloc = __fraction(nr_pages, highmem, total);
1615
1616	return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1617}
1618#else /* CONFIG_HIGHMEM */
1619static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1620{
1621	return 0;
1622}
1623
1624static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1625							 unsigned long highmem,
1626							 unsigned long total)
1627{
1628	return 0;
1629}
1630#endif /* CONFIG_HIGHMEM */
1631
1632/**
1633 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1634 */
1635static unsigned long free_unnecessary_pages(void)
1636{
1637	unsigned long save, to_free_normal, to_free_highmem, free;
1638
1639	save = count_data_pages();
1640	if (alloc_normal >= save) {
1641		to_free_normal = alloc_normal - save;
1642		save = 0;
1643	} else {
1644		to_free_normal = 0;
1645		save -= alloc_normal;
1646	}
1647	save += count_highmem_pages();
1648	if (alloc_highmem >= save) {
1649		to_free_highmem = alloc_highmem - save;
1650	} else {
1651		to_free_highmem = 0;
1652		save -= alloc_highmem;
1653		if (to_free_normal > save)
1654			to_free_normal -= save;
1655		else
1656			to_free_normal = 0;
1657	}
1658	free = to_free_normal + to_free_highmem;
1659
1660	memory_bm_position_reset(&copy_bm);
1661
1662	while (to_free_normal > 0 || to_free_highmem > 0) {
1663		unsigned long pfn = memory_bm_next_pfn(&copy_bm);
1664		struct page *page = pfn_to_page(pfn);
1665
1666		if (PageHighMem(page)) {
1667			if (!to_free_highmem)
1668				continue;
1669			to_free_highmem--;
1670			alloc_highmem--;
1671		} else {
1672			if (!to_free_normal)
1673				continue;
1674			to_free_normal--;
1675			alloc_normal--;
1676		}
1677		memory_bm_clear_bit(&copy_bm, pfn);
1678		swsusp_unset_page_forbidden(page);
1679		swsusp_unset_page_free(page);
1680		__free_page(page);
1681	}
1682
1683	return free;
1684}
1685
1686/**
1687 * minimum_image_size - Estimate the minimum acceptable size of an image.
1688 * @saveable: Number of saveable pages in the system.
1689 *
1690 * We want to avoid attempting to free too much memory too hard, so estimate the
1691 * minimum acceptable size of a hibernation image to use as the lower limit for
1692 * preallocating memory.
1693 *
1694 * We assume that the minimum image size should be proportional to
1695 *
1696 * [number of saveable pages] - [number of pages that can be freed in theory]
1697 *
1698 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1699 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1700 */
1701static unsigned long minimum_image_size(unsigned long saveable)
1702{
1703	unsigned long size;
1704
1705	size = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B)
1706		+ global_node_page_state(NR_ACTIVE_ANON)
1707		+ global_node_page_state(NR_INACTIVE_ANON)
1708		+ global_node_page_state(NR_ACTIVE_FILE)
1709		+ global_node_page_state(NR_INACTIVE_FILE);
1710
1711	return saveable <= size ? 0 : saveable - size;
1712}
1713
1714/**
1715 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1716 *
1717 * To create a hibernation image it is necessary to make a copy of every page
1718 * frame in use.  We also need a number of page frames to be free during
1719 * hibernation for allocations made while saving the image and for device
1720 * drivers, in case they need to allocate memory from their hibernation
1721 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1722 * estimate) and reserved_size divided by PAGE_SIZE (which is tunable through
1723 * /sys/power/reserved_size, respectively).  To make this happen, we compute the
1724 * total number of available page frames and allocate at least
1725 *
1726 * ([page frames total] - PAGES_FOR_IO - [metadata pages]) / 2
1727 *  - 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1728 *
1729 * of them, which corresponds to the maximum size of a hibernation image.
1730 *
1731 * If image_size is set below the number following from the above formula,
1732 * the preallocation of memory is continued until the total number of saveable
1733 * pages in the system is below the requested image size or the minimum
1734 * acceptable image size returned by minimum_image_size(), whichever is greater.
1735 */
1736int hibernate_preallocate_memory(void)
1737{
1738	struct zone *zone;
1739	unsigned long saveable, size, max_size, count, highmem, pages = 0;
1740	unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1741	ktime_t start, stop;
1742	int error;
1743
1744	pr_info("Preallocating image memory\n");
1745	start = ktime_get();
1746
1747	error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1748	if (error) {
1749		pr_err("Cannot allocate original bitmap\n");
1750		goto err_out;
1751	}
1752
1753	error = memory_bm_create(&copy_bm, GFP_IMAGE, PG_ANY);
1754	if (error) {
1755		pr_err("Cannot allocate copy bitmap\n");
1756		goto err_out;
1757	}
1758
1759	alloc_normal = 0;
1760	alloc_highmem = 0;
1761
1762	/* Count the number of saveable data pages. */
1763	save_highmem = count_highmem_pages();
1764	saveable = count_data_pages();
1765
1766	/*
1767	 * Compute the total number of page frames we can use (count) and the
1768	 * number of pages needed for image metadata (size).
1769	 */
1770	count = saveable;
1771	saveable += save_highmem;
1772	highmem = save_highmem;
1773	size = 0;
1774	for_each_populated_zone(zone) {
1775		size += snapshot_additional_pages(zone);
1776		if (is_highmem(zone))
1777			highmem += zone_page_state(zone, NR_FREE_PAGES);
1778		else
1779			count += zone_page_state(zone, NR_FREE_PAGES);
1780	}
1781	avail_normal = count;
1782	count += highmem;
1783	count -= totalreserve_pages;
1784
1785	/* Compute the maximum number of saveable pages to leave in memory. */
1786	max_size = (count - (size + PAGES_FOR_IO)) / 2
1787			- 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1788	/* Compute the desired number of image pages specified by image_size. */
1789	size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1790	if (size > max_size)
1791		size = max_size;
1792	/*
1793	 * If the desired number of image pages is at least as large as the
1794	 * current number of saveable pages in memory, allocate page frames for
1795	 * the image and we're done.
1796	 */
1797	if (size >= saveable) {
1798		pages = preallocate_image_highmem(save_highmem);
1799		pages += preallocate_image_memory(saveable - pages, avail_normal);
1800		goto out;
1801	}
1802
1803	/* Estimate the minimum size of the image. */
1804	pages = minimum_image_size(saveable);
1805	/*
1806	 * To avoid excessive pressure on the normal zone, leave room in it to
1807	 * accommodate an image of the minimum size (unless it's already too
1808	 * small, in which case don't preallocate pages from it at all).
1809	 */
1810	if (avail_normal > pages)
1811		avail_normal -= pages;
1812	else
1813		avail_normal = 0;
1814	if (size < pages)
1815		size = min_t(unsigned long, pages, max_size);
1816
1817	/*
1818	 * Let the memory management subsystem know that we're going to need a
1819	 * large number of page frames to allocate and make it free some memory.
1820	 * NOTE: If this is not done, performance will be hurt badly in some
1821	 * test cases.
1822	 */
1823	shrink_all_memory(saveable - size);
1824
1825	/*
1826	 * The number of saveable pages in memory was too high, so apply some
1827	 * pressure to decrease it.  First, make room for the largest possible
1828	 * image and fail if that doesn't work.  Next, try to decrease the size
1829	 * of the image as much as indicated by 'size' using allocations from
1830	 * highmem and non-highmem zones separately.
1831	 */
1832	pages_highmem = preallocate_image_highmem(highmem / 2);
1833	alloc = count - max_size;
1834	if (alloc > pages_highmem)
1835		alloc -= pages_highmem;
1836	else
1837		alloc = 0;
1838	pages = preallocate_image_memory(alloc, avail_normal);
1839	if (pages < alloc) {
1840		/* We have exhausted non-highmem pages, try highmem. */
1841		alloc -= pages;
1842		pages += pages_highmem;
1843		pages_highmem = preallocate_image_highmem(alloc);
1844		if (pages_highmem < alloc) {
1845			pr_err("Image allocation is %lu pages short\n",
1846				alloc - pages_highmem);
1847			goto err_out;
1848		}
1849		pages += pages_highmem;
1850		/*
1851		 * size is the desired number of saveable pages to leave in
1852		 * memory, so try to preallocate (all memory - size) pages.
1853		 */
1854		alloc = (count - pages) - size;
1855		pages += preallocate_image_highmem(alloc);
1856	} else {
1857		/*
1858		 * There are approximately max_size saveable pages at this point
1859		 * and we want to reduce this number down to size.
1860		 */
1861		alloc = max_size - size;
1862		size = preallocate_highmem_fraction(alloc, highmem, count);
1863		pages_highmem += size;
1864		alloc -= size;
1865		size = preallocate_image_memory(alloc, avail_normal);
1866		pages_highmem += preallocate_image_highmem(alloc - size);
1867		pages += pages_highmem + size;
1868	}
1869
1870	/*
1871	 * We only need as many page frames for the image as there are saveable
1872	 * pages in memory, but we have allocated more.  Release the excessive
1873	 * ones now.
1874	 */
1875	pages -= free_unnecessary_pages();
1876
1877 out:
1878	stop = ktime_get();
1879	pr_info("Allocated %lu pages for snapshot\n", pages);
1880	swsusp_show_speed(start, stop, pages, "Allocated");
1881
1882	return 0;
1883
1884 err_out:
1885	swsusp_free();
1886	return -ENOMEM;
1887}
1888
1889#ifdef CONFIG_HIGHMEM
1890/**
1891 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1892 *
1893 * Compute the number of non-highmem pages that will be necessary for creating
1894 * copies of highmem pages.
1895 */
1896static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1897{
1898	unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1899
1900	if (free_highmem >= nr_highmem)
1901		nr_highmem = 0;
1902	else
1903		nr_highmem -= free_highmem;
1904
1905	return nr_highmem;
1906}
1907#else
1908static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1909#endif /* CONFIG_HIGHMEM */
1910
1911/**
1912 * enough_free_mem - Check if there is enough free memory for the image.
1913 */
1914static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1915{
1916	struct zone *zone;
1917	unsigned int free = alloc_normal;
1918
1919	for_each_populated_zone(zone)
1920		if (!is_highmem(zone))
1921			free += zone_page_state(zone, NR_FREE_PAGES);
1922
1923	nr_pages += count_pages_for_highmem(nr_highmem);
1924	pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
1925		 nr_pages, PAGES_FOR_IO, free);
1926
1927	return free > nr_pages + PAGES_FOR_IO;
1928}
1929
1930#ifdef CONFIG_HIGHMEM
1931/**
1932 * get_highmem_buffer - Allocate a buffer for highmem pages.
1933 *
1934 * If there are some highmem pages in the hibernation image, we may need a
1935 * buffer to copy them and/or load their data.
1936 */
1937static inline int get_highmem_buffer(int safe_needed)
1938{
1939	buffer = get_image_page(GFP_ATOMIC, safe_needed);
1940	return buffer ? 0 : -ENOMEM;
1941}
1942
1943/**
1944 * alloc_highmem_pages - Allocate some highmem pages for the image.
1945 *
1946 * Try to allocate as many pages as needed, but if the number of free highmem
1947 * pages is less than that, allocate them all.
1948 */
1949static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1950					       unsigned int nr_highmem)
1951{
1952	unsigned int to_alloc = count_free_highmem_pages();
1953
1954	if (to_alloc > nr_highmem)
1955		to_alloc = nr_highmem;
1956
1957	nr_highmem -= to_alloc;
1958	while (to_alloc-- > 0) {
1959		struct page *page;
1960
1961		page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1962		memory_bm_set_bit(bm, page_to_pfn(page));
1963	}
1964	return nr_highmem;
1965}
1966#else
1967static inline int get_highmem_buffer(int safe_needed) { return 0; }
1968
1969static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1970					       unsigned int n) { return 0; }
1971#endif /* CONFIG_HIGHMEM */
1972
1973/**
1974 * swsusp_alloc - Allocate memory for hibernation image.
1975 *
1976 * We first try to allocate as many highmem pages as there are
1977 * saveable highmem pages in the system.  If that fails, we allocate
1978 * non-highmem pages for the copies of the remaining highmem ones.
1979 *
1980 * In this approach it is likely that the copies of highmem pages will
1981 * also be located in the high memory, because of the way in which
1982 * copy_data_pages() works.
1983 */
1984static int swsusp_alloc(struct memory_bitmap *copy_bm,
1985			unsigned int nr_pages, unsigned int nr_highmem)
1986{
1987	if (nr_highmem > 0) {
1988		if (get_highmem_buffer(PG_ANY))
1989			goto err_out;
1990		if (nr_highmem > alloc_highmem) {
1991			nr_highmem -= alloc_highmem;
1992			nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1993		}
1994	}
1995	if (nr_pages > alloc_normal) {
1996		nr_pages -= alloc_normal;
1997		while (nr_pages-- > 0) {
1998			struct page *page;
1999
2000			page = alloc_image_page(GFP_ATOMIC);
2001			if (!page)
2002				goto err_out;
2003			memory_bm_set_bit(copy_bm, page_to_pfn(page));
2004		}
2005	}
2006
2007	return 0;
2008
2009 err_out:
2010	swsusp_free();
2011	return -ENOMEM;
2012}
2013
2014asmlinkage __visible int swsusp_save(void)
2015{
2016	unsigned int nr_pages, nr_highmem;
2017
2018	pr_info("Creating image:\n");
2019
2020	drain_local_pages(NULL);
2021	nr_pages = count_data_pages();
2022	nr_highmem = count_highmem_pages();
2023	pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
2024
2025	if (!enough_free_mem(nr_pages, nr_highmem)) {
2026		pr_err("Not enough free memory\n");
2027		return -ENOMEM;
2028	}
2029
2030	if (swsusp_alloc(&copy_bm, nr_pages, nr_highmem)) {
2031		pr_err("Memory allocation failed\n");
2032		return -ENOMEM;
2033	}
2034
2035	/*
2036	 * During allocating of suspend pagedir, new cold pages may appear.
2037	 * Kill them.
2038	 */
2039	drain_local_pages(NULL);
2040	copy_data_pages(&copy_bm, &orig_bm);
2041
2042	/*
2043	 * End of critical section. From now on, we can write to memory,
2044	 * but we should not touch disk. This specially means we must _not_
2045	 * touch swap space! Except we must write out our image of course.
2046	 */
2047
2048	nr_pages += nr_highmem;
2049	nr_copy_pages = nr_pages;
2050	nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2051
2052	pr_info("Image created (%d pages copied)\n", nr_pages);
2053
2054	return 0;
2055}
2056
2057#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2058static int init_header_complete(struct swsusp_info *info)
2059{
2060	memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2061	info->version_code = LINUX_VERSION_CODE;
2062	return 0;
2063}
2064
2065static const char *check_image_kernel(struct swsusp_info *info)
2066{
2067	if (info->version_code != LINUX_VERSION_CODE)
2068		return "kernel version";
2069	if (strcmp(info->uts.sysname,init_utsname()->sysname))
2070		return "system type";
2071	if (strcmp(info->uts.release,init_utsname()->release))
2072		return "kernel release";
2073	if (strcmp(info->uts.version,init_utsname()->version))
2074		return "version";
2075	if (strcmp(info->uts.machine,init_utsname()->machine))
2076		return "machine";
2077	return NULL;
2078}
2079#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2080
2081unsigned long snapshot_get_image_size(void)
2082{
2083	return nr_copy_pages + nr_meta_pages + 1;
2084}
2085
2086static int init_header(struct swsusp_info *info)
2087{
2088	memset(info, 0, sizeof(struct swsusp_info));
2089	info->num_physpages = get_num_physpages();
2090	info->image_pages = nr_copy_pages;
2091	info->pages = snapshot_get_image_size();
2092	info->size = info->pages;
2093	info->size <<= PAGE_SHIFT;
2094	return init_header_complete(info);
2095}
2096
2097/**
2098 * pack_pfns - Prepare PFNs for saving.
2099 * @bm: Memory bitmap.
2100 * @buf: Memory buffer to store the PFNs in.
2101 *
2102 * PFNs corresponding to set bits in @bm are stored in the area of memory
2103 * pointed to by @buf (1 page at a time).
2104 */
2105static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2106{
2107	int j;
2108
2109	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2110		buf[j] = memory_bm_next_pfn(bm);
2111		if (unlikely(buf[j] == BM_END_OF_MAP))
2112			break;
2113	}
2114}
2115
2116/**
2117 * snapshot_read_next - Get the address to read the next image page from.
2118 * @handle: Snapshot handle to be used for the reading.
2119 *
2120 * On the first call, @handle should point to a zeroed snapshot_handle
2121 * structure.  The structure gets populated then and a pointer to it should be
2122 * passed to this function every next time.
2123 *
2124 * On success, the function returns a positive number.  Then, the caller
2125 * is allowed to read up to the returned number of bytes from the memory
2126 * location computed by the data_of() macro.
2127 *
2128 * The function returns 0 to indicate the end of the data stream condition,
2129 * and negative numbers are returned on errors.  If that happens, the structure
2130 * pointed to by @handle is not updated and should not be used any more.
2131 */
2132int snapshot_read_next(struct snapshot_handle *handle)
2133{
2134	if (handle->cur > nr_meta_pages + nr_copy_pages)
2135		return 0;
2136
2137	if (!buffer) {
2138		/* This makes the buffer be freed by swsusp_free() */
2139		buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2140		if (!buffer)
2141			return -ENOMEM;
2142	}
2143	if (!handle->cur) {
2144		int error;
2145
2146		error = init_header((struct swsusp_info *)buffer);
2147		if (error)
2148			return error;
2149		handle->buffer = buffer;
2150		memory_bm_position_reset(&orig_bm);
2151		memory_bm_position_reset(&copy_bm);
2152	} else if (handle->cur <= nr_meta_pages) {
2153		clear_page(buffer);
2154		pack_pfns(buffer, &orig_bm);
2155	} else {
2156		struct page *page;
2157
2158		page = pfn_to_page(memory_bm_next_pfn(&copy_bm));
2159		if (PageHighMem(page)) {
2160			/*
2161			 * Highmem pages are copied to the buffer,
2162			 * because we can't return with a kmapped
2163			 * highmem page (we may not be called again).
2164			 */
2165			void *kaddr;
2166
2167			kaddr = kmap_atomic(page);
2168			copy_page(buffer, kaddr);
2169			kunmap_atomic(kaddr);
2170			handle->buffer = buffer;
2171		} else {
2172			handle->buffer = page_address(page);
2173		}
2174	}
2175	handle->cur++;
2176	return PAGE_SIZE;
2177}
2178
2179static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2180				    struct memory_bitmap *src)
2181{
2182	unsigned long pfn;
2183
2184	memory_bm_position_reset(src);
2185	pfn = memory_bm_next_pfn(src);
2186	while (pfn != BM_END_OF_MAP) {
2187		memory_bm_set_bit(dst, pfn);
2188		pfn = memory_bm_next_pfn(src);
2189	}
2190}
2191
2192/**
2193 * mark_unsafe_pages - Mark pages that were used before hibernation.
2194 *
2195 * Mark the pages that cannot be used for storing the image during restoration,
2196 * because they conflict with the pages that had been used before hibernation.
2197 */
2198static void mark_unsafe_pages(struct memory_bitmap *bm)
2199{
2200	unsigned long pfn;
2201
2202	/* Clear the "free"/"unsafe" bit for all PFNs */
2203	memory_bm_position_reset(free_pages_map);
2204	pfn = memory_bm_next_pfn(free_pages_map);
2205	while (pfn != BM_END_OF_MAP) {
2206		memory_bm_clear_current(free_pages_map);
2207		pfn = memory_bm_next_pfn(free_pages_map);
2208	}
2209
2210	/* Mark pages that correspond to the "original" PFNs as "unsafe" */
2211	duplicate_memory_bitmap(free_pages_map, bm);
2212
2213	allocated_unsafe_pages = 0;
2214}
2215
2216static int check_header(struct swsusp_info *info)
2217{
2218	const char *reason;
2219
2220	reason = check_image_kernel(info);
2221	if (!reason && info->num_physpages != get_num_physpages())
2222		reason = "memory size";
2223	if (reason) {
2224		pr_err("Image mismatch: %s\n", reason);
2225		return -EPERM;
2226	}
2227	return 0;
2228}
2229
2230/**
2231 * load_header - Check the image header and copy the data from it.
2232 */
2233static int load_header(struct swsusp_info *info)
2234{
2235	int error;
2236
2237	restore_pblist = NULL;
2238	error = check_header(info);
2239	if (!error) {
2240		nr_copy_pages = info->image_pages;
2241		nr_meta_pages = info->pages - info->image_pages - 1;
2242	}
2243	return error;
2244}
2245
2246/**
2247 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2248 * @bm: Memory bitmap.
2249 * @buf: Area of memory containing the PFNs.
2250 *
2251 * For each element of the array pointed to by @buf (1 page at a time), set the
2252 * corresponding bit in @bm.
2253 */
2254static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2255{
2256	int j;
2257
2258	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2259		if (unlikely(buf[j] == BM_END_OF_MAP))
2260			break;
2261
2262		if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j])) {
2263			memory_bm_set_bit(bm, buf[j]);
2264		} else {
2265			if (!pfn_valid(buf[j]))
2266				pr_err(FW_BUG "Memory map mismatch at 0x%llx after hibernation\n",
2267				       (unsigned long long)PFN_PHYS(buf[j]));
2268			return -EFAULT;
2269		}
2270	}
2271
2272	return 0;
2273}
2274
2275#ifdef CONFIG_HIGHMEM
2276/*
2277 * struct highmem_pbe is used for creating the list of highmem pages that
2278 * should be restored atomically during the resume from disk, because the page
2279 * frames they have occupied before the suspend are in use.
2280 */
2281struct highmem_pbe {
2282	struct page *copy_page;	/* data is here now */
2283	struct page *orig_page;	/* data was here before the suspend */
2284	struct highmem_pbe *next;
2285};
2286
2287/*
2288 * List of highmem PBEs needed for restoring the highmem pages that were
2289 * allocated before the suspend and included in the suspend image, but have
2290 * also been allocated by the "resume" kernel, so their contents cannot be
2291 * written directly to their "original" page frames.
2292 */
2293static struct highmem_pbe *highmem_pblist;
2294
2295/**
2296 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2297 * @bm: Memory bitmap.
2298 *
2299 * The bits in @bm that correspond to image pages are assumed to be set.
2300 */
2301static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2302{
2303	unsigned long pfn;
2304	unsigned int cnt = 0;
2305
2306	memory_bm_position_reset(bm);
2307	pfn = memory_bm_next_pfn(bm);
2308	while (pfn != BM_END_OF_MAP) {
2309		if (PageHighMem(pfn_to_page(pfn)))
2310			cnt++;
2311
2312		pfn = memory_bm_next_pfn(bm);
2313	}
2314	return cnt;
2315}
2316
2317static unsigned int safe_highmem_pages;
2318
2319static struct memory_bitmap *safe_highmem_bm;
2320
2321/**
2322 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2323 * @bm: Pointer to an uninitialized memory bitmap structure.
2324 * @nr_highmem_p: Pointer to the number of highmem image pages.
2325 *
2326 * Try to allocate as many highmem pages as there are highmem image pages
2327 * (@nr_highmem_p points to the variable containing the number of highmem image
2328 * pages).  The pages that are "safe" (ie. will not be overwritten when the
2329 * hibernation image is restored entirely) have the corresponding bits set in
2330 * @bm (it must be uninitialized).
2331 *
2332 * NOTE: This function should not be called if there are no highmem image pages.
2333 */
2334static int prepare_highmem_image(struct memory_bitmap *bm,
2335				 unsigned int *nr_highmem_p)
2336{
2337	unsigned int to_alloc;
2338
2339	if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2340		return -ENOMEM;
2341
2342	if (get_highmem_buffer(PG_SAFE))
2343		return -ENOMEM;
2344
2345	to_alloc = count_free_highmem_pages();
2346	if (to_alloc > *nr_highmem_p)
2347		to_alloc = *nr_highmem_p;
2348	else
2349		*nr_highmem_p = to_alloc;
2350
2351	safe_highmem_pages = 0;
2352	while (to_alloc-- > 0) {
2353		struct page *page;
2354
2355		page = alloc_page(__GFP_HIGHMEM);
2356		if (!swsusp_page_is_free(page)) {
2357			/* The page is "safe", set its bit the bitmap */
2358			memory_bm_set_bit(bm, page_to_pfn(page));
2359			safe_highmem_pages++;
2360		}
2361		/* Mark the page as allocated */
2362		swsusp_set_page_forbidden(page);
2363		swsusp_set_page_free(page);
2364	}
2365	memory_bm_position_reset(bm);
2366	safe_highmem_bm = bm;
2367	return 0;
2368}
2369
2370static struct page *last_highmem_page;
2371
2372/**
2373 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2374 *
2375 * For a given highmem image page get a buffer that suspend_write_next() should
2376 * return to its caller to write to.
2377 *
2378 * If the page is to be saved to its "original" page frame or a copy of
2379 * the page is to be made in the highmem, @buffer is returned.  Otherwise,
2380 * the copy of the page is to be made in normal memory, so the address of
2381 * the copy is returned.
2382 *
2383 * If @buffer is returned, the caller of suspend_write_next() will write
2384 * the page's contents to @buffer, so they will have to be copied to the
2385 * right location on the next call to suspend_write_next() and it is done
2386 * with the help of copy_last_highmem_page().  For this purpose, if
2387 * @buffer is returned, @last_highmem_page is set to the page to which
2388 * the data will have to be copied from @buffer.
2389 */
2390static void *get_highmem_page_buffer(struct page *page,
2391				     struct chain_allocator *ca)
2392{
2393	struct highmem_pbe *pbe;
2394	void *kaddr;
2395
2396	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2397		/*
2398		 * We have allocated the "original" page frame and we can
2399		 * use it directly to store the loaded page.
2400		 */
2401		last_highmem_page = page;
2402		return buffer;
2403	}
2404	/*
2405	 * The "original" page frame has not been allocated and we have to
2406	 * use a "safe" page frame to store the loaded page.
2407	 */
2408	pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2409	if (!pbe) {
2410		swsusp_free();
2411		return ERR_PTR(-ENOMEM);
2412	}
2413	pbe->orig_page = page;
2414	if (safe_highmem_pages > 0) {
2415		struct page *tmp;
2416
2417		/* Copy of the page will be stored in high memory */
2418		kaddr = buffer;
2419		tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2420		safe_highmem_pages--;
2421		last_highmem_page = tmp;
2422		pbe->copy_page = tmp;
2423	} else {
2424		/* Copy of the page will be stored in normal memory */
2425		kaddr = safe_pages_list;
2426		safe_pages_list = safe_pages_list->next;
2427		pbe->copy_page = virt_to_page(kaddr);
2428	}
2429	pbe->next = highmem_pblist;
2430	highmem_pblist = pbe;
2431	return kaddr;
2432}
2433
2434/**
2435 * copy_last_highmem_page - Copy most the most recent highmem image page.
2436 *
2437 * Copy the contents of a highmem image from @buffer, where the caller of
2438 * snapshot_write_next() has stored them, to the right location represented by
2439 * @last_highmem_page .
2440 */
2441static void copy_last_highmem_page(void)
2442{
2443	if (last_highmem_page) {
2444		void *dst;
2445
2446		dst = kmap_atomic(last_highmem_page);
2447		copy_page(dst, buffer);
2448		kunmap_atomic(dst);
2449		last_highmem_page = NULL;
2450	}
2451}
2452
2453static inline int last_highmem_page_copied(void)
2454{
2455	return !last_highmem_page;
2456}
2457
2458static inline void free_highmem_data(void)
2459{
2460	if (safe_highmem_bm)
2461		memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2462
2463	if (buffer)
2464		free_image_page(buffer, PG_UNSAFE_CLEAR);
2465}
2466#else
2467static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2468
2469static inline int prepare_highmem_image(struct memory_bitmap *bm,
2470					unsigned int *nr_highmem_p) { return 0; }
2471
2472static inline void *get_highmem_page_buffer(struct page *page,
2473					    struct chain_allocator *ca)
2474{
2475	return ERR_PTR(-EINVAL);
2476}
2477
2478static inline void copy_last_highmem_page(void) {}
2479static inline int last_highmem_page_copied(void) { return 1; }
2480static inline void free_highmem_data(void) {}
2481#endif /* CONFIG_HIGHMEM */
2482
2483#define PBES_PER_LINKED_PAGE	(LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2484
2485/**
2486 * prepare_image - Make room for loading hibernation image.
2487 * @new_bm: Uninitialized memory bitmap structure.
2488 * @bm: Memory bitmap with unsafe pages marked.
2489 *
2490 * Use @bm to mark the pages that will be overwritten in the process of
2491 * restoring the system memory state from the suspend image ("unsafe" pages)
2492 * and allocate memory for the image.
2493 *
2494 * The idea is to allocate a new memory bitmap first and then allocate
2495 * as many pages as needed for image data, but without specifying what those
2496 * pages will be used for just yet.  Instead, we mark them all as allocated and
2497 * create a lists of "safe" pages to be used later.  On systems with high
2498 * memory a list of "safe" highmem pages is created too.
2499 */
2500static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2501{
2502	unsigned int nr_pages, nr_highmem;
2503	struct linked_page *lp;
2504	int error;
2505
2506	/* If there is no highmem, the buffer will not be necessary */
2507	free_image_page(buffer, PG_UNSAFE_CLEAR);
2508	buffer = NULL;
2509
2510	nr_highmem = count_highmem_image_pages(bm);
2511	mark_unsafe_pages(bm);
2512
2513	error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2514	if (error)
2515		goto Free;
2516
2517	duplicate_memory_bitmap(new_bm, bm);
2518	memory_bm_free(bm, PG_UNSAFE_KEEP);
2519	if (nr_highmem > 0) {
2520		error = prepare_highmem_image(bm, &nr_highmem);
2521		if (error)
2522			goto Free;
2523	}
2524	/*
2525	 * Reserve some safe pages for potential later use.
2526	 *
2527	 * NOTE: This way we make sure there will be enough safe pages for the
2528	 * chain_alloc() in get_buffer().  It is a bit wasteful, but
2529	 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2530	 *
2531	 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2532	 */
2533	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2534	nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2535	while (nr_pages > 0) {
2536		lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2537		if (!lp) {
2538			error = -ENOMEM;
2539			goto Free;
2540		}
2541		lp->next = safe_pages_list;
2542		safe_pages_list = lp;
2543		nr_pages--;
2544	}
2545	/* Preallocate memory for the image */
2546	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2547	while (nr_pages > 0) {
2548		lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2549		if (!lp) {
2550			error = -ENOMEM;
2551			goto Free;
2552		}
2553		if (!swsusp_page_is_free(virt_to_page(lp))) {
2554			/* The page is "safe", add it to the list */
2555			lp->next = safe_pages_list;
2556			safe_pages_list = lp;
2557		}
2558		/* Mark the page as allocated */
2559		swsusp_set_page_forbidden(virt_to_page(lp));
2560		swsusp_set_page_free(virt_to_page(lp));
2561		nr_pages--;
2562	}
2563	return 0;
2564
2565 Free:
2566	swsusp_free();
2567	return error;
2568}
2569
2570/**
2571 * get_buffer - Get the address to store the next image data page.
2572 *
2573 * Get the address that snapshot_write_next() should return to its caller to
2574 * write to.
2575 */
2576static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2577{
2578	struct pbe *pbe;
2579	struct page *page;
2580	unsigned long pfn = memory_bm_next_pfn(bm);
2581
2582	if (pfn == BM_END_OF_MAP)
2583		return ERR_PTR(-EFAULT);
2584
2585	page = pfn_to_page(pfn);
2586	if (PageHighMem(page))
2587		return get_highmem_page_buffer(page, ca);
2588
2589	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2590		/*
2591		 * We have allocated the "original" page frame and we can
2592		 * use it directly to store the loaded page.
2593		 */
2594		return page_address(page);
2595
2596	/*
2597	 * The "original" page frame has not been allocated and we have to
2598	 * use a "safe" page frame to store the loaded page.
2599	 */
2600	pbe = chain_alloc(ca, sizeof(struct pbe));
2601	if (!pbe) {
2602		swsusp_free();
2603		return ERR_PTR(-ENOMEM);
2604	}
2605	pbe->orig_address = page_address(page);
2606	pbe->address = safe_pages_list;
2607	safe_pages_list = safe_pages_list->next;
2608	pbe->next = restore_pblist;
2609	restore_pblist = pbe;
2610	return pbe->address;
2611}
2612
2613/**
2614 * snapshot_write_next - Get the address to store the next image page.
2615 * @handle: Snapshot handle structure to guide the writing.
2616 *
2617 * On the first call, @handle should point to a zeroed snapshot_handle
2618 * structure.  The structure gets populated then and a pointer to it should be
2619 * passed to this function every next time.
2620 *
2621 * On success, the function returns a positive number.  Then, the caller
2622 * is allowed to write up to the returned number of bytes to the memory
2623 * location computed by the data_of() macro.
2624 *
2625 * The function returns 0 to indicate the "end of file" condition.  Negative
2626 * numbers are returned on errors, in which cases the structure pointed to by
2627 * @handle is not updated and should not be used any more.
2628 */
2629int snapshot_write_next(struct snapshot_handle *handle)
2630{
2631	static struct chain_allocator ca;
2632	int error = 0;
2633
2634	/* Check if we have already loaded the entire image */
2635	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2636		return 0;
2637
2638	handle->sync_read = 1;
2639
2640	if (!handle->cur) {
2641		if (!buffer)
2642			/* This makes the buffer be freed by swsusp_free() */
2643			buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2644
2645		if (!buffer)
2646			return -ENOMEM;
2647
2648		handle->buffer = buffer;
2649	} else if (handle->cur == 1) {
2650		error = load_header(buffer);
2651		if (error)
2652			return error;
2653
2654		safe_pages_list = NULL;
2655
2656		error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
2657		if (error)
2658			return error;
2659
2660		hibernate_restore_protection_begin();
2661	} else if (handle->cur <= nr_meta_pages + 1) {
2662		error = unpack_orig_pfns(buffer, &copy_bm);
2663		if (error)
2664			return error;
2665
2666		if (handle->cur == nr_meta_pages + 1) {
2667			error = prepare_image(&orig_bm, &copy_bm);
2668			if (error)
2669				return error;
2670
2671			chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2672			memory_bm_position_reset(&orig_bm);
2673			restore_pblist = NULL;
2674			handle->buffer = get_buffer(&orig_bm, &ca);
2675			handle->sync_read = 0;
2676			if (IS_ERR(handle->buffer))
2677				return PTR_ERR(handle->buffer);
2678		}
2679	} else {
2680		copy_last_highmem_page();
2681		hibernate_restore_protect_page(handle->buffer);
2682		handle->buffer = get_buffer(&orig_bm, &ca);
2683		if (IS_ERR(handle->buffer))
2684			return PTR_ERR(handle->buffer);
2685		if (handle->buffer != buffer)
2686			handle->sync_read = 0;
2687	}
2688	handle->cur++;
2689	return PAGE_SIZE;
2690}
2691
2692/**
2693 * snapshot_write_finalize - Complete the loading of a hibernation image.
2694 *
2695 * Must be called after the last call to snapshot_write_next() in case the last
2696 * page in the image happens to be a highmem page and its contents should be
2697 * stored in highmem.  Additionally, it recycles bitmap memory that's not
2698 * necessary any more.
2699 */
2700void snapshot_write_finalize(struct snapshot_handle *handle)
2701{
2702	copy_last_highmem_page();
2703	hibernate_restore_protect_page(handle->buffer);
2704	/* Do that only if we have loaded the image entirely */
2705	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2706		memory_bm_recycle(&orig_bm);
2707		free_highmem_data();
2708	}
2709}
2710
2711int snapshot_image_loaded(struct snapshot_handle *handle)
2712{
2713	return !(!nr_copy_pages || !last_highmem_page_copied() ||
2714			handle->cur <= nr_meta_pages + nr_copy_pages);
2715}
2716
2717#ifdef CONFIG_HIGHMEM
2718/* Assumes that @buf is ready and points to a "safe" page */
2719static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2720				       void *buf)
2721{
2722	void *kaddr1, *kaddr2;
2723
2724	kaddr1 = kmap_atomic(p1);
2725	kaddr2 = kmap_atomic(p2);
2726	copy_page(buf, kaddr1);
2727	copy_page(kaddr1, kaddr2);
2728	copy_page(kaddr2, buf);
2729	kunmap_atomic(kaddr2);
2730	kunmap_atomic(kaddr1);
2731}
2732
2733/**
2734 * restore_highmem - Put highmem image pages into their original locations.
2735 *
2736 * For each highmem page that was in use before hibernation and is included in
2737 * the image, and also has been allocated by the "restore" kernel, swap its
2738 * current contents with the previous (ie. "before hibernation") ones.
2739 *
2740 * If the restore eventually fails, we can call this function once again and
2741 * restore the highmem state as seen by the restore kernel.
2742 */
2743int restore_highmem(void)
2744{
2745	struct highmem_pbe *pbe = highmem_pblist;
2746	void *buf;
2747
2748	if (!pbe)
2749		return 0;
2750
2751	buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2752	if (!buf)
2753		return -ENOMEM;
2754
2755	while (pbe) {
2756		swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2757		pbe = pbe->next;
2758	}
2759	free_image_page(buf, PG_UNSAFE_CLEAR);
2760	return 0;
2761}
2762#endif /* CONFIG_HIGHMEM */
v5.14.15
   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
  79
  80/*
  81 * The calls to set_direct_map_*() should not fail because remapping a page
  82 * here means that we only update protection bits in an existing PTE.
  83 * It is still worth to have a warning here if something changes and this
  84 * will no longer be the case.
  85 */
  86static inline void hibernate_map_page(struct page *page)
  87{
  88	if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
  89		int ret = set_direct_map_default_noflush(page);
  90
  91		if (ret)
  92			pr_warn_once("Failed to remap page\n");
  93	} else {
  94		debug_pagealloc_map_pages(page, 1);
  95	}
  96}
  97
  98static inline void hibernate_unmap_page(struct page *page)
  99{
 100	if (IS_ENABLED(CONFIG_ARCH_HAS_SET_DIRECT_MAP)) {
 101		unsigned long addr = (unsigned long)page_address(page);
 102		int ret  = set_direct_map_invalid_noflush(page);
 103
 104		if (ret)
 105			pr_warn_once("Failed to remap page\n");
 106
 107		flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
 108	} else {
 109		debug_pagealloc_unmap_pages(page, 1);
 110	}
 111}
 112
 113static int swsusp_page_is_free(struct page *);
 114static void swsusp_set_page_forbidden(struct page *);
 115static void swsusp_unset_page_forbidden(struct page *);
 116
 117/*
 118 * Number of bytes to reserve for memory allocations made by device drivers
 119 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
 120 * cause image creation to fail (tunable via /sys/power/reserved_size).
 121 */
 122unsigned long reserved_size;
 123
 124void __init hibernate_reserved_size_init(void)
 125{
 126	reserved_size = SPARE_PAGES * PAGE_SIZE;
 127}
 128
 129/*
 130 * Preferred image size in bytes (tunable via /sys/power/image_size).
 131 * When it is set to N, swsusp will do its best to ensure the image
 132 * size will not exceed N bytes, but if that is impossible, it will
 133 * try to create the smallest image possible.
 134 */
 135unsigned long image_size;
 136
 137void __init hibernate_image_size_init(void)
 138{
 139	image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
 140}
 141
 142/*
 143 * List of PBEs needed for restoring the pages that were allocated before
 144 * the suspend and included in the suspend image, but have also been
 145 * allocated by the "resume" kernel, so their contents cannot be written
 146 * directly to their "original" page frames.
 147 */
 148struct pbe *restore_pblist;
 149
 150/* struct linked_page is used to build chains of pages */
 151
 152#define LINKED_PAGE_DATA_SIZE	(PAGE_SIZE - sizeof(void *))
 153
 154struct linked_page {
 155	struct linked_page *next;
 156	char data[LINKED_PAGE_DATA_SIZE];
 157} __packed;
 158
 159/*
 160 * List of "safe" pages (ie. pages that were not used by the image kernel
 161 * before hibernation) that may be used as temporary storage for image kernel
 162 * memory contents.
 163 */
 164static struct linked_page *safe_pages_list;
 165
 166/* Pointer to an auxiliary buffer (1 page) */
 167static void *buffer;
 168
 169#define PG_ANY		0
 170#define PG_SAFE		1
 171#define PG_UNSAFE_CLEAR	1
 172#define PG_UNSAFE_KEEP	0
 173
 174static unsigned int allocated_unsafe_pages;
 175
 176/**
 177 * get_image_page - Allocate a page for a hibernation image.
 178 * @gfp_mask: GFP mask for the allocation.
 179 * @safe_needed: Get pages that were not used before hibernation (restore only)
 180 *
 181 * During image restoration, for storing the PBE list and the image data, we can
 182 * only use memory pages that do not conflict with the pages used before
 183 * hibernation.  The "unsafe" pages have PageNosaveFree set and we count them
 184 * using allocated_unsafe_pages.
 185 *
 186 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
 187 * swsusp_free() can release it.
 188 */
 189static void *get_image_page(gfp_t gfp_mask, int safe_needed)
 190{
 191	void *res;
 192
 193	res = (void *)get_zeroed_page(gfp_mask);
 194	if (safe_needed)
 195		while (res && swsusp_page_is_free(virt_to_page(res))) {
 196			/* The page is unsafe, mark it for swsusp_free() */
 197			swsusp_set_page_forbidden(virt_to_page(res));
 198			allocated_unsafe_pages++;
 199			res = (void *)get_zeroed_page(gfp_mask);
 200		}
 201	if (res) {
 202		swsusp_set_page_forbidden(virt_to_page(res));
 203		swsusp_set_page_free(virt_to_page(res));
 204	}
 205	return res;
 206}
 207
 208static void *__get_safe_page(gfp_t gfp_mask)
 209{
 210	if (safe_pages_list) {
 211		void *ret = safe_pages_list;
 212
 213		safe_pages_list = safe_pages_list->next;
 214		memset(ret, 0, PAGE_SIZE);
 215		return ret;
 216	}
 217	return get_image_page(gfp_mask, PG_SAFE);
 218}
 219
 220unsigned long get_safe_page(gfp_t gfp_mask)
 221{
 222	return (unsigned long)__get_safe_page(gfp_mask);
 223}
 224
 225static struct page *alloc_image_page(gfp_t gfp_mask)
 226{
 227	struct page *page;
 228
 229	page = alloc_page(gfp_mask);
 230	if (page) {
 231		swsusp_set_page_forbidden(page);
 232		swsusp_set_page_free(page);
 233	}
 234	return page;
 235}
 236
 237static void recycle_safe_page(void *page_address)
 238{
 239	struct linked_page *lp = page_address;
 240
 241	lp->next = safe_pages_list;
 242	safe_pages_list = lp;
 243}
 244
 245/**
 246 * free_image_page - Free a page allocated for hibernation image.
 247 * @addr: Address of the page to free.
 248 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
 249 *
 250 * The page to free should have been allocated by get_image_page() (page flags
 251 * set by it are affected).
 252 */
 253static inline void free_image_page(void *addr, int clear_nosave_free)
 254{
 255	struct page *page;
 256
 257	BUG_ON(!virt_addr_valid(addr));
 258
 259	page = virt_to_page(addr);
 260
 261	swsusp_unset_page_forbidden(page);
 262	if (clear_nosave_free)
 263		swsusp_unset_page_free(page);
 264
 265	__free_page(page);
 266}
 267
 268static inline void free_list_of_pages(struct linked_page *list,
 269				      int clear_page_nosave)
 270{
 271	while (list) {
 272		struct linked_page *lp = list->next;
 273
 274		free_image_page(list, clear_page_nosave);
 275		list = lp;
 276	}
 277}
 278
 279/*
 280 * struct chain_allocator is used for allocating small objects out of
 281 * a linked list of pages called 'the chain'.
 282 *
 283 * The chain grows each time when there is no room for a new object in
 284 * the current page.  The allocated objects cannot be freed individually.
 285 * It is only possible to free them all at once, by freeing the entire
 286 * chain.
 287 *
 288 * NOTE: The chain allocator may be inefficient if the allocated objects
 289 * are not much smaller than PAGE_SIZE.
 290 */
 291struct chain_allocator {
 292	struct linked_page *chain;	/* the chain */
 293	unsigned int used_space;	/* total size of objects allocated out
 294					   of the current page */
 295	gfp_t gfp_mask;		/* mask for allocating pages */
 296	int safe_needed;	/* if set, only "safe" pages are allocated */
 297};
 298
 299static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
 300		       int safe_needed)
 301{
 302	ca->chain = NULL;
 303	ca->used_space = LINKED_PAGE_DATA_SIZE;
 304	ca->gfp_mask = gfp_mask;
 305	ca->safe_needed = safe_needed;
 306}
 307
 308static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
 309{
 310	void *ret;
 311
 312	if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
 313		struct linked_page *lp;
 314
 315		lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
 316					get_image_page(ca->gfp_mask, PG_ANY);
 317		if (!lp)
 318			return NULL;
 319
 320		lp->next = ca->chain;
 321		ca->chain = lp;
 322		ca->used_space = 0;
 323	}
 324	ret = ca->chain->data + ca->used_space;
 325	ca->used_space += size;
 326	return ret;
 327}
 328
 329/**
 330 * Data types related to memory bitmaps.
 331 *
 332 * Memory bitmap is a structure consisting of many linked lists of
 333 * objects.  The main list's elements are of type struct zone_bitmap
 334 * and each of them corresponds to one zone.  For each zone bitmap
 335 * object there is a list of objects of type struct bm_block that
 336 * represent each blocks of bitmap in which information is stored.
 337 *
 338 * struct memory_bitmap contains a pointer to the main list of zone
 339 * bitmap objects, a struct bm_position used for browsing the bitmap,
 340 * and a pointer to the list of pages used for allocating all of the
 341 * zone bitmap objects and bitmap block objects.
 342 *
 343 * NOTE: It has to be possible to lay out the bitmap in memory
 344 * using only allocations of order 0.  Additionally, the bitmap is
 345 * designed to work with arbitrary number of zones (this is over the
 346 * top for now, but let's avoid making unnecessary assumptions ;-).
 347 *
 348 * struct zone_bitmap contains a pointer to a list of bitmap block
 349 * objects and a pointer to the bitmap block object that has been
 350 * most recently used for setting bits.  Additionally, it contains the
 351 * PFNs that correspond to the start and end of the represented zone.
 352 *
 353 * struct bm_block contains a pointer to the memory page in which
 354 * information is stored (in the form of a block of bitmap)
 355 * It also contains the pfns that correspond to the start and end of
 356 * the represented memory area.
 357 *
 358 * The memory bitmap is organized as a radix tree to guarantee fast random
 359 * access to the bits. There is one radix tree for each zone (as returned
 360 * from create_mem_extents).
 361 *
 362 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
 363 * two linked lists for the nodes of the tree, one for the inner nodes and
 364 * one for the leave nodes. The linked leave nodes are used for fast linear
 365 * access of the memory bitmap.
 366 *
 367 * The struct rtree_node represents one node of the radix tree.
 368 */
 369
 370#define BM_END_OF_MAP	(~0UL)
 371
 372#define BM_BITS_PER_BLOCK	(PAGE_SIZE * BITS_PER_BYTE)
 373#define BM_BLOCK_SHIFT		(PAGE_SHIFT + 3)
 374#define BM_BLOCK_MASK		((1UL << BM_BLOCK_SHIFT) - 1)
 375
 376/*
 377 * struct rtree_node is a wrapper struct to link the nodes
 378 * of the rtree together for easy linear iteration over
 379 * bits and easy freeing
 380 */
 381struct rtree_node {
 382	struct list_head list;
 383	unsigned long *data;
 384};
 385
 386/*
 387 * struct mem_zone_bm_rtree represents a bitmap used for one
 388 * populated memory zone.
 389 */
 390struct mem_zone_bm_rtree {
 391	struct list_head list;		/* Link Zones together         */
 392	struct list_head nodes;		/* Radix Tree inner nodes      */
 393	struct list_head leaves;	/* Radix Tree leaves           */
 394	unsigned long start_pfn;	/* Zone start page frame       */
 395	unsigned long end_pfn;		/* Zone end page frame + 1     */
 396	struct rtree_node *rtree;	/* Radix Tree Root             */
 397	int levels;			/* Number of Radix Tree Levels */
 398	unsigned int blocks;		/* Number of Bitmap Blocks     */
 399};
 400
 401/* strcut bm_position is used for browsing memory bitmaps */
 402
 403struct bm_position {
 404	struct mem_zone_bm_rtree *zone;
 405	struct rtree_node *node;
 406	unsigned long node_pfn;
 407	int node_bit;
 408};
 409
 410struct memory_bitmap {
 411	struct list_head zones;
 412	struct linked_page *p_list;	/* list of pages used to store zone
 413					   bitmap objects and bitmap block
 414					   objects */
 415	struct bm_position cur;	/* most recently used bit position */
 416};
 417
 418/* Functions that operate on memory bitmaps */
 419
 420#define BM_ENTRIES_PER_LEVEL	(PAGE_SIZE / sizeof(unsigned long))
 421#if BITS_PER_LONG == 32
 422#define BM_RTREE_LEVEL_SHIFT	(PAGE_SHIFT - 2)
 423#else
 424#define BM_RTREE_LEVEL_SHIFT	(PAGE_SHIFT - 3)
 425#endif
 426#define BM_RTREE_LEVEL_MASK	((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
 427
 428/**
 429 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
 
 
 
 
 430 *
 431 * This function is used to allocate inner nodes as well as the
 432 * leave nodes of the radix tree. It also adds the node to the
 433 * corresponding linked list passed in by the *list parameter.
 434 */
 435static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
 436					   struct chain_allocator *ca,
 437					   struct list_head *list)
 438{
 439	struct rtree_node *node;
 440
 441	node = chain_alloc(ca, sizeof(struct rtree_node));
 442	if (!node)
 443		return NULL;
 444
 445	node->data = get_image_page(gfp_mask, safe_needed);
 446	if (!node->data)
 447		return NULL;
 448
 449	list_add_tail(&node->list, list);
 450
 451	return node;
 452}
 453
 454/**
 455 * add_rtree_block - Add a new leave node to the radix tree.
 456 *
 457 * The leave nodes need to be allocated in order to keep the leaves
 458 * linked list in order. This is guaranteed by the zone->blocks
 459 * counter.
 460 */
 461static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
 462			   int safe_needed, struct chain_allocator *ca)
 463{
 464	struct rtree_node *node, *block, **dst;
 465	unsigned int levels_needed, block_nr;
 466	int i;
 467
 468	block_nr = zone->blocks;
 469	levels_needed = 0;
 470
 471	/* How many levels do we need for this block nr? */
 472	while (block_nr) {
 473		levels_needed += 1;
 474		block_nr >>= BM_RTREE_LEVEL_SHIFT;
 475	}
 476
 477	/* Make sure the rtree has enough levels */
 478	for (i = zone->levels; i < levels_needed; i++) {
 479		node = alloc_rtree_node(gfp_mask, safe_needed, ca,
 480					&zone->nodes);
 481		if (!node)
 482			return -ENOMEM;
 483
 484		node->data[0] = (unsigned long)zone->rtree;
 485		zone->rtree = node;
 486		zone->levels += 1;
 487	}
 488
 489	/* Allocate new block */
 490	block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
 491	if (!block)
 492		return -ENOMEM;
 493
 494	/* Now walk the rtree to insert the block */
 495	node = zone->rtree;
 496	dst = &zone->rtree;
 497	block_nr = zone->blocks;
 498	for (i = zone->levels; i > 0; i--) {
 499		int index;
 500
 501		if (!node) {
 502			node = alloc_rtree_node(gfp_mask, safe_needed, ca,
 503						&zone->nodes);
 504			if (!node)
 505				return -ENOMEM;
 506			*dst = node;
 507		}
 508
 509		index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
 510		index &= BM_RTREE_LEVEL_MASK;
 511		dst = (struct rtree_node **)&((*dst)->data[index]);
 512		node = *dst;
 513	}
 514
 515	zone->blocks += 1;
 516	*dst = block;
 517
 518	return 0;
 519}
 520
 521static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
 522			       int clear_nosave_free);
 523
 524/**
 525 * create_zone_bm_rtree - Create a radix tree for one zone.
 526 *
 527 * Allocated the mem_zone_bm_rtree structure and initializes it.
 528 * This function also allocated and builds the radix tree for the
 529 * zone.
 530 */
 531static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
 532						      int safe_needed,
 533						      struct chain_allocator *ca,
 534						      unsigned long start,
 535						      unsigned long end)
 536{
 537	struct mem_zone_bm_rtree *zone;
 538	unsigned int i, nr_blocks;
 539	unsigned long pages;
 540
 541	pages = end - start;
 542	zone  = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
 543	if (!zone)
 544		return NULL;
 545
 546	INIT_LIST_HEAD(&zone->nodes);
 547	INIT_LIST_HEAD(&zone->leaves);
 548	zone->start_pfn = start;
 549	zone->end_pfn = end;
 550	nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
 551
 552	for (i = 0; i < nr_blocks; i++) {
 553		if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
 554			free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
 555			return NULL;
 556		}
 557	}
 558
 559	return zone;
 560}
 561
 562/**
 563 * free_zone_bm_rtree - Free the memory of the radix tree.
 564 *
 565 * Free all node pages of the radix tree. The mem_zone_bm_rtree
 566 * structure itself is not freed here nor are the rtree_node
 567 * structs.
 568 */
 569static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
 570			       int clear_nosave_free)
 571{
 572	struct rtree_node *node;
 573
 574	list_for_each_entry(node, &zone->nodes, list)
 575		free_image_page(node->data, clear_nosave_free);
 576
 577	list_for_each_entry(node, &zone->leaves, list)
 578		free_image_page(node->data, clear_nosave_free);
 579}
 580
 581static void memory_bm_position_reset(struct memory_bitmap *bm)
 582{
 583	bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
 584				  list);
 585	bm->cur.node = list_entry(bm->cur.zone->leaves.next,
 586				  struct rtree_node, list);
 587	bm->cur.node_pfn = 0;
 588	bm->cur.node_bit = 0;
 589}
 590
 591static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
 592
 593struct mem_extent {
 594	struct list_head hook;
 595	unsigned long start;
 596	unsigned long end;
 597};
 598
 599/**
 600 * free_mem_extents - Free a list of memory extents.
 601 * @list: List of extents to free.
 602 */
 603static void free_mem_extents(struct list_head *list)
 604{
 605	struct mem_extent *ext, *aux;
 606
 607	list_for_each_entry_safe(ext, aux, list, hook) {
 608		list_del(&ext->hook);
 609		kfree(ext);
 610	}
 611}
 612
 613/**
 614 * create_mem_extents - Create a list of memory extents.
 615 * @list: List to put the extents into.
 616 * @gfp_mask: Mask to use for memory allocations.
 617 *
 618 * The extents represent contiguous ranges of PFNs.
 619 */
 620static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
 621{
 622	struct zone *zone;
 623
 624	INIT_LIST_HEAD(list);
 625
 626	for_each_populated_zone(zone) {
 627		unsigned long zone_start, zone_end;
 628		struct mem_extent *ext, *cur, *aux;
 629
 630		zone_start = zone->zone_start_pfn;
 631		zone_end = zone_end_pfn(zone);
 632
 633		list_for_each_entry(ext, list, hook)
 634			if (zone_start <= ext->end)
 635				break;
 636
 637		if (&ext->hook == list || zone_end < ext->start) {
 638			/* New extent is necessary */
 639			struct mem_extent *new_ext;
 640
 641			new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
 642			if (!new_ext) {
 643				free_mem_extents(list);
 644				return -ENOMEM;
 645			}
 646			new_ext->start = zone_start;
 647			new_ext->end = zone_end;
 648			list_add_tail(&new_ext->hook, &ext->hook);
 649			continue;
 650		}
 651
 652		/* Merge this zone's range of PFNs with the existing one */
 653		if (zone_start < ext->start)
 654			ext->start = zone_start;
 655		if (zone_end > ext->end)
 656			ext->end = zone_end;
 657
 658		/* More merging may be possible */
 659		cur = ext;
 660		list_for_each_entry_safe_continue(cur, aux, list, hook) {
 661			if (zone_end < cur->start)
 662				break;
 663			if (zone_end < cur->end)
 664				ext->end = cur->end;
 665			list_del(&cur->hook);
 666			kfree(cur);
 667		}
 668	}
 669
 670	return 0;
 671}
 672
 673/**
 674 * memory_bm_create - Allocate memory for a memory bitmap.
 675 */
 676static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
 677			    int safe_needed)
 678{
 679	struct chain_allocator ca;
 680	struct list_head mem_extents;
 681	struct mem_extent *ext;
 682	int error;
 683
 684	chain_init(&ca, gfp_mask, safe_needed);
 685	INIT_LIST_HEAD(&bm->zones);
 686
 687	error = create_mem_extents(&mem_extents, gfp_mask);
 688	if (error)
 689		return error;
 690
 691	list_for_each_entry(ext, &mem_extents, hook) {
 692		struct mem_zone_bm_rtree *zone;
 693
 694		zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
 695					    ext->start, ext->end);
 696		if (!zone) {
 697			error = -ENOMEM;
 698			goto Error;
 699		}
 700		list_add_tail(&zone->list, &bm->zones);
 701	}
 702
 703	bm->p_list = ca.chain;
 704	memory_bm_position_reset(bm);
 705 Exit:
 706	free_mem_extents(&mem_extents);
 707	return error;
 708
 709 Error:
 710	bm->p_list = ca.chain;
 711	memory_bm_free(bm, PG_UNSAFE_CLEAR);
 712	goto Exit;
 713}
 714
 715/**
 716 * memory_bm_free - Free memory occupied by the memory bitmap.
 717 * @bm: Memory bitmap.
 718 */
 719static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
 720{
 721	struct mem_zone_bm_rtree *zone;
 722
 723	list_for_each_entry(zone, &bm->zones, list)
 724		free_zone_bm_rtree(zone, clear_nosave_free);
 725
 726	free_list_of_pages(bm->p_list, clear_nosave_free);
 727
 728	INIT_LIST_HEAD(&bm->zones);
 729}
 730
 731/**
 732 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
 733 *
 734 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
 735 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
 736 *
 737 * Walk the radix tree to find the page containing the bit that represents @pfn
 738 * and return the position of the bit in @addr and @bit_nr.
 739 */
 740static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
 741			      void **addr, unsigned int *bit_nr)
 742{
 743	struct mem_zone_bm_rtree *curr, *zone;
 744	struct rtree_node *node;
 745	int i, block_nr;
 746
 747	zone = bm->cur.zone;
 748
 749	if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
 750		goto zone_found;
 751
 752	zone = NULL;
 753
 754	/* Find the right zone */
 755	list_for_each_entry(curr, &bm->zones, list) {
 756		if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
 757			zone = curr;
 758			break;
 759		}
 760	}
 761
 762	if (!zone)
 763		return -EFAULT;
 764
 765zone_found:
 766	/*
 767	 * We have found the zone. Now walk the radix tree to find the leaf node
 768	 * for our PFN.
 769	 */
 770
 771	/*
 772	 * If the zone we wish to scan is the current zone and the
 773	 * pfn falls into the current node then we do not need to walk
 774	 * the tree.
 775	 */
 776	node = bm->cur.node;
 777	if (zone == bm->cur.zone &&
 778	    ((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
 779		goto node_found;
 780
 781	node      = zone->rtree;
 782	block_nr  = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
 783
 784	for (i = zone->levels; i > 0; i--) {
 785		int index;
 786
 787		index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
 788		index &= BM_RTREE_LEVEL_MASK;
 789		BUG_ON(node->data[index] == 0);
 790		node = (struct rtree_node *)node->data[index];
 791	}
 792
 793node_found:
 794	/* Update last position */
 795	bm->cur.zone = zone;
 796	bm->cur.node = node;
 797	bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
 798
 799	/* Set return values */
 800	*addr = node->data;
 801	*bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
 802
 803	return 0;
 804}
 805
 806static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
 807{
 808	void *addr;
 809	unsigned int bit;
 810	int error;
 811
 812	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 813	BUG_ON(error);
 814	set_bit(bit, addr);
 815}
 816
 817static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
 818{
 819	void *addr;
 820	unsigned int bit;
 821	int error;
 822
 823	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 824	if (!error)
 825		set_bit(bit, addr);
 826
 827	return error;
 828}
 829
 830static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
 831{
 832	void *addr;
 833	unsigned int bit;
 834	int error;
 835
 836	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 837	BUG_ON(error);
 838	clear_bit(bit, addr);
 839}
 840
 841static void memory_bm_clear_current(struct memory_bitmap *bm)
 842{
 843	int bit;
 844
 845	bit = max(bm->cur.node_bit - 1, 0);
 846	clear_bit(bit, bm->cur.node->data);
 847}
 848
 849static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
 850{
 851	void *addr;
 852	unsigned int bit;
 853	int error;
 854
 855	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 856	BUG_ON(error);
 857	return test_bit(bit, addr);
 858}
 859
 860static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
 861{
 862	void *addr;
 863	unsigned int bit;
 864
 865	return !memory_bm_find_bit(bm, pfn, &addr, &bit);
 866}
 867
 868/*
 869 * rtree_next_node - Jump to the next leaf node.
 870 *
 871 * Set the position to the beginning of the next node in the
 872 * memory bitmap. This is either the next node in the current
 873 * zone's radix tree or the first node in the radix tree of the
 874 * next zone.
 875 *
 876 * Return true if there is a next node, false otherwise.
 877 */
 878static bool rtree_next_node(struct memory_bitmap *bm)
 879{
 880	if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
 881		bm->cur.node = list_entry(bm->cur.node->list.next,
 882					  struct rtree_node, list);
 883		bm->cur.node_pfn += BM_BITS_PER_BLOCK;
 884		bm->cur.node_bit  = 0;
 885		touch_softlockup_watchdog();
 886		return true;
 887	}
 888
 889	/* No more nodes, goto next zone */
 890	if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
 891		bm->cur.zone = list_entry(bm->cur.zone->list.next,
 892				  struct mem_zone_bm_rtree, list);
 893		bm->cur.node = list_entry(bm->cur.zone->leaves.next,
 894					  struct rtree_node, list);
 895		bm->cur.node_pfn = 0;
 896		bm->cur.node_bit = 0;
 897		return true;
 898	}
 899
 900	/* No more zones */
 901	return false;
 902}
 903
 904/**
 905 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
 906 * @bm: Memory bitmap.
 907 *
 908 * Starting from the last returned position this function searches for the next
 909 * set bit in @bm and returns the PFN represented by it.  If no more bits are
 910 * set, BM_END_OF_MAP is returned.
 911 *
 912 * It is required to run memory_bm_position_reset() before the first call to
 913 * this function for the given memory bitmap.
 914 */
 915static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
 916{
 917	unsigned long bits, pfn, pages;
 918	int bit;
 919
 920	do {
 921		pages	  = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
 922		bits      = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
 923		bit	  = find_next_bit(bm->cur.node->data, bits,
 924					  bm->cur.node_bit);
 925		if (bit < bits) {
 926			pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
 927			bm->cur.node_bit = bit + 1;
 928			return pfn;
 929		}
 930	} while (rtree_next_node(bm));
 931
 932	return BM_END_OF_MAP;
 933}
 934
 935/*
 936 * This structure represents a range of page frames the contents of which
 937 * should not be saved during hibernation.
 938 */
 939struct nosave_region {
 940	struct list_head list;
 941	unsigned long start_pfn;
 942	unsigned long end_pfn;
 943};
 944
 945static LIST_HEAD(nosave_regions);
 946
 947static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
 948{
 949	struct rtree_node *node;
 950
 951	list_for_each_entry(node, &zone->nodes, list)
 952		recycle_safe_page(node->data);
 953
 954	list_for_each_entry(node, &zone->leaves, list)
 955		recycle_safe_page(node->data);
 956}
 957
 958static void memory_bm_recycle(struct memory_bitmap *bm)
 959{
 960	struct mem_zone_bm_rtree *zone;
 961	struct linked_page *p_list;
 962
 963	list_for_each_entry(zone, &bm->zones, list)
 964		recycle_zone_bm_rtree(zone);
 965
 966	p_list = bm->p_list;
 967	while (p_list) {
 968		struct linked_page *lp = p_list;
 969
 970		p_list = lp->next;
 971		recycle_safe_page(lp);
 972	}
 973}
 974
 975/**
 976 * register_nosave_region - Register a region of unsaveable memory.
 977 *
 978 * Register a range of page frames the contents of which should not be saved
 979 * during hibernation (to be used in the early initialization code).
 980 */
 981void __init __register_nosave_region(unsigned long start_pfn,
 982				     unsigned long end_pfn, int use_kmalloc)
 983{
 984	struct nosave_region *region;
 985
 986	if (start_pfn >= end_pfn)
 987		return;
 988
 989	if (!list_empty(&nosave_regions)) {
 990		/* Try to extend the previous region (they should be sorted) */
 991		region = list_entry(nosave_regions.prev,
 992					struct nosave_region, list);
 993		if (region->end_pfn == start_pfn) {
 994			region->end_pfn = end_pfn;
 995			goto Report;
 996		}
 997	}
 998	if (use_kmalloc) {
 999		/* During init, this shouldn't fail */
1000		region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
1001		BUG_ON(!region);
1002	} else {
1003		/* This allocation cannot fail */
1004		region = memblock_alloc(sizeof(struct nosave_region),
1005					SMP_CACHE_BYTES);
1006		if (!region)
1007			panic("%s: Failed to allocate %zu bytes\n", __func__,
1008			      sizeof(struct nosave_region));
1009	}
1010	region->start_pfn = start_pfn;
1011	region->end_pfn = end_pfn;
1012	list_add_tail(&region->list, &nosave_regions);
1013 Report:
1014	pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
1015		(unsigned long long) start_pfn << PAGE_SHIFT,
1016		((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1017}
1018
1019/*
1020 * Set bits in this map correspond to the page frames the contents of which
1021 * should not be saved during the suspend.
1022 */
1023static struct memory_bitmap *forbidden_pages_map;
1024
1025/* Set bits in this map correspond to free page frames. */
1026static struct memory_bitmap *free_pages_map;
1027
1028/*
1029 * Each page frame allocated for creating the image is marked by setting the
1030 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
1031 */
1032
1033void swsusp_set_page_free(struct page *page)
1034{
1035	if (free_pages_map)
1036		memory_bm_set_bit(free_pages_map, page_to_pfn(page));
1037}
1038
1039static int swsusp_page_is_free(struct page *page)
1040{
1041	return free_pages_map ?
1042		memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1043}
1044
1045void swsusp_unset_page_free(struct page *page)
1046{
1047	if (free_pages_map)
1048		memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1049}
1050
1051static void swsusp_set_page_forbidden(struct page *page)
1052{
1053	if (forbidden_pages_map)
1054		memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1055}
1056
1057int swsusp_page_is_forbidden(struct page *page)
1058{
1059	return forbidden_pages_map ?
1060		memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1061}
1062
1063static void swsusp_unset_page_forbidden(struct page *page)
1064{
1065	if (forbidden_pages_map)
1066		memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1067}
1068
1069/**
1070 * mark_nosave_pages - Mark pages that should not be saved.
1071 * @bm: Memory bitmap.
1072 *
1073 * Set the bits in @bm that correspond to the page frames the contents of which
1074 * should not be saved.
1075 */
1076static void mark_nosave_pages(struct memory_bitmap *bm)
1077{
1078	struct nosave_region *region;
1079
1080	if (list_empty(&nosave_regions))
1081		return;
1082
1083	list_for_each_entry(region, &nosave_regions, list) {
1084		unsigned long pfn;
1085
1086		pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1087			 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1088			 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1089				- 1);
1090
1091		for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1092			if (pfn_valid(pfn)) {
1093				/*
1094				 * It is safe to ignore the result of
1095				 * mem_bm_set_bit_check() here, since we won't
1096				 * touch the PFNs for which the error is
1097				 * returned anyway.
1098				 */
1099				mem_bm_set_bit_check(bm, pfn);
1100			}
1101	}
1102}
1103
1104/**
1105 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1106 *
1107 * Create bitmaps needed for marking page frames that should not be saved and
1108 * free page frames.  The forbidden_pages_map and free_pages_map pointers are
1109 * only modified if everything goes well, because we don't want the bits to be
1110 * touched before both bitmaps are set up.
1111 */
1112int create_basic_memory_bitmaps(void)
1113{
1114	struct memory_bitmap *bm1, *bm2;
1115	int error = 0;
1116
1117	if (forbidden_pages_map && free_pages_map)
1118		return 0;
1119	else
1120		BUG_ON(forbidden_pages_map || free_pages_map);
1121
1122	bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1123	if (!bm1)
1124		return -ENOMEM;
1125
1126	error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1127	if (error)
1128		goto Free_first_object;
1129
1130	bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1131	if (!bm2)
1132		goto Free_first_bitmap;
1133
1134	error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1135	if (error)
1136		goto Free_second_object;
1137
1138	forbidden_pages_map = bm1;
1139	free_pages_map = bm2;
1140	mark_nosave_pages(forbidden_pages_map);
1141
1142	pr_debug("Basic memory bitmaps created\n");
1143
1144	return 0;
1145
1146 Free_second_object:
1147	kfree(bm2);
1148 Free_first_bitmap:
1149	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1150 Free_first_object:
1151	kfree(bm1);
1152	return -ENOMEM;
1153}
1154
1155/**
1156 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1157 *
1158 * Free memory bitmaps allocated by create_basic_memory_bitmaps().  The
1159 * auxiliary pointers are necessary so that the bitmaps themselves are not
1160 * referred to while they are being freed.
1161 */
1162void free_basic_memory_bitmaps(void)
1163{
1164	struct memory_bitmap *bm1, *bm2;
1165
1166	if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1167		return;
1168
1169	bm1 = forbidden_pages_map;
1170	bm2 = free_pages_map;
1171	forbidden_pages_map = NULL;
1172	free_pages_map = NULL;
1173	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1174	kfree(bm1);
1175	memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1176	kfree(bm2);
1177
1178	pr_debug("Basic memory bitmaps freed\n");
1179}
1180
1181static void clear_or_poison_free_page(struct page *page)
1182{
1183	if (page_poisoning_enabled_static())
1184		__kernel_poison_pages(page, 1);
1185	else if (want_init_on_free())
1186		clear_highpage(page);
1187}
1188
1189void clear_or_poison_free_pages(void)
1190{
1191	struct memory_bitmap *bm = free_pages_map;
1192	unsigned long pfn;
1193
1194	if (WARN_ON(!(free_pages_map)))
1195		return;
1196
1197	if (page_poisoning_enabled() || want_init_on_free()) {
1198		memory_bm_position_reset(bm);
1199		pfn = memory_bm_next_pfn(bm);
1200		while (pfn != BM_END_OF_MAP) {
1201			if (pfn_valid(pfn))
1202				clear_or_poison_free_page(pfn_to_page(pfn));
1203
1204			pfn = memory_bm_next_pfn(bm);
1205		}
1206		memory_bm_position_reset(bm);
1207		pr_info("free pages cleared after restore\n");
1208	}
1209}
1210
1211/**
1212 * snapshot_additional_pages - Estimate the number of extra pages needed.
1213 * @zone: Memory zone to carry out the computation for.
1214 *
1215 * Estimate the number of additional pages needed for setting up a hibernation
1216 * image data structures for @zone (usually, the returned value is greater than
1217 * the exact number).
1218 */
1219unsigned int snapshot_additional_pages(struct zone *zone)
1220{
1221	unsigned int rtree, nodes;
1222
1223	rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1224	rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1225			      LINKED_PAGE_DATA_SIZE);
1226	while (nodes > 1) {
1227		nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1228		rtree += nodes;
1229	}
1230
1231	return 2 * rtree;
1232}
1233
1234#ifdef CONFIG_HIGHMEM
1235/**
1236 * count_free_highmem_pages - Compute the total number of free highmem pages.
1237 *
1238 * The returned number is system-wide.
1239 */
1240static unsigned int count_free_highmem_pages(void)
1241{
1242	struct zone *zone;
1243	unsigned int cnt = 0;
1244
1245	for_each_populated_zone(zone)
1246		if (is_highmem(zone))
1247			cnt += zone_page_state(zone, NR_FREE_PAGES);
1248
1249	return cnt;
1250}
1251
1252/**
1253 * saveable_highmem_page - Check if a highmem page is saveable.
1254 *
1255 * Determine whether a highmem page should be included in a hibernation image.
1256 *
1257 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1258 * and it isn't part of a free chunk of pages.
1259 */
1260static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1261{
1262	struct page *page;
1263
1264	if (!pfn_valid(pfn))
1265		return NULL;
1266
1267	page = pfn_to_online_page(pfn);
1268	if (!page || page_zone(page) != zone)
1269		return NULL;
1270
1271	BUG_ON(!PageHighMem(page));
1272
1273	if (swsusp_page_is_forbidden(page) ||  swsusp_page_is_free(page))
1274		return NULL;
1275
1276	if (PageReserved(page) || PageOffline(page))
1277		return NULL;
1278
1279	if (page_is_guard(page))
1280		return NULL;
1281
1282	return page;
1283}
1284
1285/**
1286 * count_highmem_pages - Compute the total number of saveable highmem pages.
1287 */
1288static unsigned int count_highmem_pages(void)
1289{
1290	struct zone *zone;
1291	unsigned int n = 0;
1292
1293	for_each_populated_zone(zone) {
1294		unsigned long pfn, max_zone_pfn;
1295
1296		if (!is_highmem(zone))
1297			continue;
1298
1299		mark_free_pages(zone);
1300		max_zone_pfn = zone_end_pfn(zone);
1301		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1302			if (saveable_highmem_page(zone, pfn))
1303				n++;
1304	}
1305	return n;
1306}
1307#else
1308static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1309{
1310	return NULL;
1311}
1312#endif /* CONFIG_HIGHMEM */
1313
1314/**
1315 * saveable_page - Check if the given page is saveable.
1316 *
1317 * Determine whether a non-highmem page should be included in a hibernation
1318 * image.
1319 *
1320 * We should save the page if it isn't Nosave, and is not in the range
1321 * of pages statically defined as 'unsaveable', and it isn't part of
1322 * a free chunk of pages.
1323 */
1324static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1325{
1326	struct page *page;
1327
1328	if (!pfn_valid(pfn))
1329		return NULL;
1330
1331	page = pfn_to_online_page(pfn);
1332	if (!page || page_zone(page) != zone)
1333		return NULL;
1334
1335	BUG_ON(PageHighMem(page));
1336
1337	if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1338		return NULL;
1339
1340	if (PageOffline(page))
1341		return NULL;
1342
1343	if (PageReserved(page)
1344	    && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1345		return NULL;
1346
1347	if (page_is_guard(page))
1348		return NULL;
1349
1350	return page;
1351}
1352
1353/**
1354 * count_data_pages - Compute the total number of saveable non-highmem pages.
1355 */
1356static unsigned int count_data_pages(void)
1357{
1358	struct zone *zone;
1359	unsigned long pfn, max_zone_pfn;
1360	unsigned int n = 0;
1361
1362	for_each_populated_zone(zone) {
1363		if (is_highmem(zone))
1364			continue;
1365
1366		mark_free_pages(zone);
1367		max_zone_pfn = zone_end_pfn(zone);
1368		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1369			if (saveable_page(zone, pfn))
1370				n++;
1371	}
1372	return n;
1373}
1374
1375/*
1376 * This is needed, because copy_page and memcpy are not usable for copying
1377 * task structs.
1378 */
1379static inline void do_copy_page(long *dst, long *src)
1380{
1381	int n;
1382
1383	for (n = PAGE_SIZE / sizeof(long); n; n--)
1384		*dst++ = *src++;
1385}
1386
1387/**
1388 * safe_copy_page - Copy a page in a safe way.
1389 *
1390 * Check if the page we are going to copy is marked as present in the kernel
1391 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1392 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1393 * always returns 'true'.
1394 */
1395static void safe_copy_page(void *dst, struct page *s_page)
1396{
1397	if (kernel_page_present(s_page)) {
1398		do_copy_page(dst, page_address(s_page));
1399	} else {
1400		hibernate_map_page(s_page);
1401		do_copy_page(dst, page_address(s_page));
1402		hibernate_unmap_page(s_page);
1403	}
1404}
1405
1406#ifdef CONFIG_HIGHMEM
1407static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1408{
1409	return is_highmem(zone) ?
1410		saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1411}
1412
1413static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1414{
1415	struct page *s_page, *d_page;
1416	void *src, *dst;
1417
1418	s_page = pfn_to_page(src_pfn);
1419	d_page = pfn_to_page(dst_pfn);
1420	if (PageHighMem(s_page)) {
1421		src = kmap_atomic(s_page);
1422		dst = kmap_atomic(d_page);
1423		do_copy_page(dst, src);
1424		kunmap_atomic(dst);
1425		kunmap_atomic(src);
1426	} else {
1427		if (PageHighMem(d_page)) {
1428			/*
1429			 * The page pointed to by src may contain some kernel
1430			 * data modified by kmap_atomic()
1431			 */
1432			safe_copy_page(buffer, s_page);
1433			dst = kmap_atomic(d_page);
1434			copy_page(dst, buffer);
1435			kunmap_atomic(dst);
1436		} else {
1437			safe_copy_page(page_address(d_page), s_page);
1438		}
1439	}
1440}
1441#else
1442#define page_is_saveable(zone, pfn)	saveable_page(zone, pfn)
1443
1444static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1445{
1446	safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1447				pfn_to_page(src_pfn));
1448}
1449#endif /* CONFIG_HIGHMEM */
1450
1451static void copy_data_pages(struct memory_bitmap *copy_bm,
1452			    struct memory_bitmap *orig_bm)
1453{
1454	struct zone *zone;
1455	unsigned long pfn;
1456
1457	for_each_populated_zone(zone) {
1458		unsigned long max_zone_pfn;
1459
1460		mark_free_pages(zone);
1461		max_zone_pfn = zone_end_pfn(zone);
1462		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1463			if (page_is_saveable(zone, pfn))
1464				memory_bm_set_bit(orig_bm, pfn);
1465	}
1466	memory_bm_position_reset(orig_bm);
1467	memory_bm_position_reset(copy_bm);
1468	for(;;) {
1469		pfn = memory_bm_next_pfn(orig_bm);
1470		if (unlikely(pfn == BM_END_OF_MAP))
1471			break;
1472		copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1473	}
1474}
1475
1476/* Total number of image pages */
1477static unsigned int nr_copy_pages;
1478/* Number of pages needed for saving the original pfns of the image pages */
1479static unsigned int nr_meta_pages;
1480/*
1481 * Numbers of normal and highmem page frames allocated for hibernation image
1482 * before suspending devices.
1483 */
1484static unsigned int alloc_normal, alloc_highmem;
1485/*
1486 * Memory bitmap used for marking saveable pages (during hibernation) or
1487 * hibernation image pages (during restore)
1488 */
1489static struct memory_bitmap orig_bm;
1490/*
1491 * Memory bitmap used during hibernation for marking allocated page frames that
1492 * will contain copies of saveable pages.  During restore it is initially used
1493 * for marking hibernation image pages, but then the set bits from it are
1494 * duplicated in @orig_bm and it is released.  On highmem systems it is next
1495 * used for marking "safe" highmem pages, but it has to be reinitialized for
1496 * this purpose.
1497 */
1498static struct memory_bitmap copy_bm;
1499
1500/**
1501 * swsusp_free - Free pages allocated for hibernation image.
1502 *
1503 * Image pages are allocated before snapshot creation, so they need to be
1504 * released after resume.
1505 */
1506void swsusp_free(void)
1507{
1508	unsigned long fb_pfn, fr_pfn;
1509
1510	if (!forbidden_pages_map || !free_pages_map)
1511		goto out;
1512
1513	memory_bm_position_reset(forbidden_pages_map);
1514	memory_bm_position_reset(free_pages_map);
1515
1516loop:
1517	fr_pfn = memory_bm_next_pfn(free_pages_map);
1518	fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1519
1520	/*
1521	 * Find the next bit set in both bitmaps. This is guaranteed to
1522	 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1523	 */
1524	do {
1525		if (fb_pfn < fr_pfn)
1526			fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1527		if (fr_pfn < fb_pfn)
1528			fr_pfn = memory_bm_next_pfn(free_pages_map);
1529	} while (fb_pfn != fr_pfn);
1530
1531	if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1532		struct page *page = pfn_to_page(fr_pfn);
1533
1534		memory_bm_clear_current(forbidden_pages_map);
1535		memory_bm_clear_current(free_pages_map);
1536		hibernate_restore_unprotect_page(page_address(page));
1537		__free_page(page);
1538		goto loop;
1539	}
1540
1541out:
1542	nr_copy_pages = 0;
1543	nr_meta_pages = 0;
1544	restore_pblist = NULL;
1545	buffer = NULL;
1546	alloc_normal = 0;
1547	alloc_highmem = 0;
1548	hibernate_restore_protection_end();
1549}
1550
1551/* Helper functions used for the shrinking of memory. */
1552
1553#define GFP_IMAGE	(GFP_KERNEL | __GFP_NOWARN)
1554
1555/**
1556 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1557 * @nr_pages: Number of page frames to allocate.
1558 * @mask: GFP flags to use for the allocation.
1559 *
1560 * Return value: Number of page frames actually allocated
1561 */
1562static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1563{
1564	unsigned long nr_alloc = 0;
1565
1566	while (nr_pages > 0) {
1567		struct page *page;
1568
1569		page = alloc_image_page(mask);
1570		if (!page)
1571			break;
1572		memory_bm_set_bit(&copy_bm, page_to_pfn(page));
1573		if (PageHighMem(page))
1574			alloc_highmem++;
1575		else
1576			alloc_normal++;
1577		nr_pages--;
1578		nr_alloc++;
1579	}
1580
1581	return nr_alloc;
1582}
1583
1584static unsigned long preallocate_image_memory(unsigned long nr_pages,
1585					      unsigned long avail_normal)
1586{
1587	unsigned long alloc;
1588
1589	if (avail_normal <= alloc_normal)
1590		return 0;
1591
1592	alloc = avail_normal - alloc_normal;
1593	if (nr_pages < alloc)
1594		alloc = nr_pages;
1595
1596	return preallocate_image_pages(alloc, GFP_IMAGE);
1597}
1598
1599#ifdef CONFIG_HIGHMEM
1600static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1601{
1602	return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1603}
1604
1605/**
1606 *  __fraction - Compute (an approximation of) x * (multiplier / base).
1607 */
1608static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1609{
1610	return div64_u64(x * multiplier, base);
1611}
1612
1613static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1614						  unsigned long highmem,
1615						  unsigned long total)
1616{
1617	unsigned long alloc = __fraction(nr_pages, highmem, total);
1618
1619	return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1620}
1621#else /* CONFIG_HIGHMEM */
1622static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1623{
1624	return 0;
1625}
1626
1627static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1628							 unsigned long highmem,
1629							 unsigned long total)
1630{
1631	return 0;
1632}
1633#endif /* CONFIG_HIGHMEM */
1634
1635/**
1636 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1637 */
1638static unsigned long free_unnecessary_pages(void)
1639{
1640	unsigned long save, to_free_normal, to_free_highmem, free;
1641
1642	save = count_data_pages();
1643	if (alloc_normal >= save) {
1644		to_free_normal = alloc_normal - save;
1645		save = 0;
1646	} else {
1647		to_free_normal = 0;
1648		save -= alloc_normal;
1649	}
1650	save += count_highmem_pages();
1651	if (alloc_highmem >= save) {
1652		to_free_highmem = alloc_highmem - save;
1653	} else {
1654		to_free_highmem = 0;
1655		save -= alloc_highmem;
1656		if (to_free_normal > save)
1657			to_free_normal -= save;
1658		else
1659			to_free_normal = 0;
1660	}
1661	free = to_free_normal + to_free_highmem;
1662
1663	memory_bm_position_reset(&copy_bm);
1664
1665	while (to_free_normal > 0 || to_free_highmem > 0) {
1666		unsigned long pfn = memory_bm_next_pfn(&copy_bm);
1667		struct page *page = pfn_to_page(pfn);
1668
1669		if (PageHighMem(page)) {
1670			if (!to_free_highmem)
1671				continue;
1672			to_free_highmem--;
1673			alloc_highmem--;
1674		} else {
1675			if (!to_free_normal)
1676				continue;
1677			to_free_normal--;
1678			alloc_normal--;
1679		}
1680		memory_bm_clear_bit(&copy_bm, pfn);
1681		swsusp_unset_page_forbidden(page);
1682		swsusp_unset_page_free(page);
1683		__free_page(page);
1684	}
1685
1686	return free;
1687}
1688
1689/**
1690 * minimum_image_size - Estimate the minimum acceptable size of an image.
1691 * @saveable: Number of saveable pages in the system.
1692 *
1693 * We want to avoid attempting to free too much memory too hard, so estimate the
1694 * minimum acceptable size of a hibernation image to use as the lower limit for
1695 * preallocating memory.
1696 *
1697 * We assume that the minimum image size should be proportional to
1698 *
1699 * [number of saveable pages] - [number of pages that can be freed in theory]
1700 *
1701 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1702 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1703 */
1704static unsigned long minimum_image_size(unsigned long saveable)
1705{
1706	unsigned long size;
1707
1708	size = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B)
1709		+ global_node_page_state(NR_ACTIVE_ANON)
1710		+ global_node_page_state(NR_INACTIVE_ANON)
1711		+ global_node_page_state(NR_ACTIVE_FILE)
1712		+ global_node_page_state(NR_INACTIVE_FILE);
1713
1714	return saveable <= size ? 0 : saveable - size;
1715}
1716
1717/**
1718 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1719 *
1720 * To create a hibernation image it is necessary to make a copy of every page
1721 * frame in use.  We also need a number of page frames to be free during
1722 * hibernation for allocations made while saving the image and for device
1723 * drivers, in case they need to allocate memory from their hibernation
1724 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1725 * estimate) and reserved_size divided by PAGE_SIZE (which is tunable through
1726 * /sys/power/reserved_size, respectively).  To make this happen, we compute the
1727 * total number of available page frames and allocate at least
1728 *
1729 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1730 *  + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1731 *
1732 * of them, which corresponds to the maximum size of a hibernation image.
1733 *
1734 * If image_size is set below the number following from the above formula,
1735 * the preallocation of memory is continued until the total number of saveable
1736 * pages in the system is below the requested image size or the minimum
1737 * acceptable image size returned by minimum_image_size(), whichever is greater.
1738 */
1739int hibernate_preallocate_memory(void)
1740{
1741	struct zone *zone;
1742	unsigned long saveable, size, max_size, count, highmem, pages = 0;
1743	unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1744	ktime_t start, stop;
1745	int error;
1746
1747	pr_info("Preallocating image memory\n");
1748	start = ktime_get();
1749
1750	error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1751	if (error) {
1752		pr_err("Cannot allocate original bitmap\n");
1753		goto err_out;
1754	}
1755
1756	error = memory_bm_create(&copy_bm, GFP_IMAGE, PG_ANY);
1757	if (error) {
1758		pr_err("Cannot allocate copy bitmap\n");
1759		goto err_out;
1760	}
1761
1762	alloc_normal = 0;
1763	alloc_highmem = 0;
1764
1765	/* Count the number of saveable data pages. */
1766	save_highmem = count_highmem_pages();
1767	saveable = count_data_pages();
1768
1769	/*
1770	 * Compute the total number of page frames we can use (count) and the
1771	 * number of pages needed for image metadata (size).
1772	 */
1773	count = saveable;
1774	saveable += save_highmem;
1775	highmem = save_highmem;
1776	size = 0;
1777	for_each_populated_zone(zone) {
1778		size += snapshot_additional_pages(zone);
1779		if (is_highmem(zone))
1780			highmem += zone_page_state(zone, NR_FREE_PAGES);
1781		else
1782			count += zone_page_state(zone, NR_FREE_PAGES);
1783	}
1784	avail_normal = count;
1785	count += highmem;
1786	count -= totalreserve_pages;
1787
1788	/* Compute the maximum number of saveable pages to leave in memory. */
1789	max_size = (count - (size + PAGES_FOR_IO)) / 2
1790			- 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1791	/* Compute the desired number of image pages specified by image_size. */
1792	size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1793	if (size > max_size)
1794		size = max_size;
1795	/*
1796	 * If the desired number of image pages is at least as large as the
1797	 * current number of saveable pages in memory, allocate page frames for
1798	 * the image and we're done.
1799	 */
1800	if (size >= saveable) {
1801		pages = preallocate_image_highmem(save_highmem);
1802		pages += preallocate_image_memory(saveable - pages, avail_normal);
1803		goto out;
1804	}
1805
1806	/* Estimate the minimum size of the image. */
1807	pages = minimum_image_size(saveable);
1808	/*
1809	 * To avoid excessive pressure on the normal zone, leave room in it to
1810	 * accommodate an image of the minimum size (unless it's already too
1811	 * small, in which case don't preallocate pages from it at all).
1812	 */
1813	if (avail_normal > pages)
1814		avail_normal -= pages;
1815	else
1816		avail_normal = 0;
1817	if (size < pages)
1818		size = min_t(unsigned long, pages, max_size);
1819
1820	/*
1821	 * Let the memory management subsystem know that we're going to need a
1822	 * large number of page frames to allocate and make it free some memory.
1823	 * NOTE: If this is not done, performance will be hurt badly in some
1824	 * test cases.
1825	 */
1826	shrink_all_memory(saveable - size);
1827
1828	/*
1829	 * The number of saveable pages in memory was too high, so apply some
1830	 * pressure to decrease it.  First, make room for the largest possible
1831	 * image and fail if that doesn't work.  Next, try to decrease the size
1832	 * of the image as much as indicated by 'size' using allocations from
1833	 * highmem and non-highmem zones separately.
1834	 */
1835	pages_highmem = preallocate_image_highmem(highmem / 2);
1836	alloc = count - max_size;
1837	if (alloc > pages_highmem)
1838		alloc -= pages_highmem;
1839	else
1840		alloc = 0;
1841	pages = preallocate_image_memory(alloc, avail_normal);
1842	if (pages < alloc) {
1843		/* We have exhausted non-highmem pages, try highmem. */
1844		alloc -= pages;
1845		pages += pages_highmem;
1846		pages_highmem = preallocate_image_highmem(alloc);
1847		if (pages_highmem < alloc) {
1848			pr_err("Image allocation is %lu pages short\n",
1849				alloc - pages_highmem);
1850			goto err_out;
1851		}
1852		pages += pages_highmem;
1853		/*
1854		 * size is the desired number of saveable pages to leave in
1855		 * memory, so try to preallocate (all memory - size) pages.
1856		 */
1857		alloc = (count - pages) - size;
1858		pages += preallocate_image_highmem(alloc);
1859	} else {
1860		/*
1861		 * There are approximately max_size saveable pages at this point
1862		 * and we want to reduce this number down to size.
1863		 */
1864		alloc = max_size - size;
1865		size = preallocate_highmem_fraction(alloc, highmem, count);
1866		pages_highmem += size;
1867		alloc -= size;
1868		size = preallocate_image_memory(alloc, avail_normal);
1869		pages_highmem += preallocate_image_highmem(alloc - size);
1870		pages += pages_highmem + size;
1871	}
1872
1873	/*
1874	 * We only need as many page frames for the image as there are saveable
1875	 * pages in memory, but we have allocated more.  Release the excessive
1876	 * ones now.
1877	 */
1878	pages -= free_unnecessary_pages();
1879
1880 out:
1881	stop = ktime_get();
1882	pr_info("Allocated %lu pages for snapshot\n", pages);
1883	swsusp_show_speed(start, stop, pages, "Allocated");
1884
1885	return 0;
1886
1887 err_out:
1888	swsusp_free();
1889	return -ENOMEM;
1890}
1891
1892#ifdef CONFIG_HIGHMEM
1893/**
1894 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1895 *
1896 * Compute the number of non-highmem pages that will be necessary for creating
1897 * copies of highmem pages.
1898 */
1899static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1900{
1901	unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1902
1903	if (free_highmem >= nr_highmem)
1904		nr_highmem = 0;
1905	else
1906		nr_highmem -= free_highmem;
1907
1908	return nr_highmem;
1909}
1910#else
1911static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1912#endif /* CONFIG_HIGHMEM */
1913
1914/**
1915 * enough_free_mem - Check if there is enough free memory for the image.
1916 */
1917static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1918{
1919	struct zone *zone;
1920	unsigned int free = alloc_normal;
1921
1922	for_each_populated_zone(zone)
1923		if (!is_highmem(zone))
1924			free += zone_page_state(zone, NR_FREE_PAGES);
1925
1926	nr_pages += count_pages_for_highmem(nr_highmem);
1927	pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
1928		 nr_pages, PAGES_FOR_IO, free);
1929
1930	return free > nr_pages + PAGES_FOR_IO;
1931}
1932
1933#ifdef CONFIG_HIGHMEM
1934/**
1935 * get_highmem_buffer - Allocate a buffer for highmem pages.
1936 *
1937 * If there are some highmem pages in the hibernation image, we may need a
1938 * buffer to copy them and/or load their data.
1939 */
1940static inline int get_highmem_buffer(int safe_needed)
1941{
1942	buffer = get_image_page(GFP_ATOMIC, safe_needed);
1943	return buffer ? 0 : -ENOMEM;
1944}
1945
1946/**
1947 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1948 *
1949 * Try to allocate as many pages as needed, but if the number of free highmem
1950 * pages is less than that, allocate them all.
1951 */
1952static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1953					       unsigned int nr_highmem)
1954{
1955	unsigned int to_alloc = count_free_highmem_pages();
1956
1957	if (to_alloc > nr_highmem)
1958		to_alloc = nr_highmem;
1959
1960	nr_highmem -= to_alloc;
1961	while (to_alloc-- > 0) {
1962		struct page *page;
1963
1964		page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1965		memory_bm_set_bit(bm, page_to_pfn(page));
1966	}
1967	return nr_highmem;
1968}
1969#else
1970static inline int get_highmem_buffer(int safe_needed) { return 0; }
1971
1972static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1973					       unsigned int n) { return 0; }
1974#endif /* CONFIG_HIGHMEM */
1975
1976/**
1977 * swsusp_alloc - Allocate memory for hibernation image.
1978 *
1979 * We first try to allocate as many highmem pages as there are
1980 * saveable highmem pages in the system.  If that fails, we allocate
1981 * non-highmem pages for the copies of the remaining highmem ones.
1982 *
1983 * In this approach it is likely that the copies of highmem pages will
1984 * also be located in the high memory, because of the way in which
1985 * copy_data_pages() works.
1986 */
1987static int swsusp_alloc(struct memory_bitmap *copy_bm,
1988			unsigned int nr_pages, unsigned int nr_highmem)
1989{
1990	if (nr_highmem > 0) {
1991		if (get_highmem_buffer(PG_ANY))
1992			goto err_out;
1993		if (nr_highmem > alloc_highmem) {
1994			nr_highmem -= alloc_highmem;
1995			nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1996		}
1997	}
1998	if (nr_pages > alloc_normal) {
1999		nr_pages -= alloc_normal;
2000		while (nr_pages-- > 0) {
2001			struct page *page;
2002
2003			page = alloc_image_page(GFP_ATOMIC);
2004			if (!page)
2005				goto err_out;
2006			memory_bm_set_bit(copy_bm, page_to_pfn(page));
2007		}
2008	}
2009
2010	return 0;
2011
2012 err_out:
2013	swsusp_free();
2014	return -ENOMEM;
2015}
2016
2017asmlinkage __visible int swsusp_save(void)
2018{
2019	unsigned int nr_pages, nr_highmem;
2020
2021	pr_info("Creating image:\n");
2022
2023	drain_local_pages(NULL);
2024	nr_pages = count_data_pages();
2025	nr_highmem = count_highmem_pages();
2026	pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
2027
2028	if (!enough_free_mem(nr_pages, nr_highmem)) {
2029		pr_err("Not enough free memory\n");
2030		return -ENOMEM;
2031	}
2032
2033	if (swsusp_alloc(&copy_bm, nr_pages, nr_highmem)) {
2034		pr_err("Memory allocation failed\n");
2035		return -ENOMEM;
2036	}
2037
2038	/*
2039	 * During allocating of suspend pagedir, new cold pages may appear.
2040	 * Kill them.
2041	 */
2042	drain_local_pages(NULL);
2043	copy_data_pages(&copy_bm, &orig_bm);
2044
2045	/*
2046	 * End of critical section. From now on, we can write to memory,
2047	 * but we should not touch disk. This specially means we must _not_
2048	 * touch swap space! Except we must write out our image of course.
2049	 */
2050
2051	nr_pages += nr_highmem;
2052	nr_copy_pages = nr_pages;
2053	nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2054
2055	pr_info("Image created (%d pages copied)\n", nr_pages);
2056
2057	return 0;
2058}
2059
2060#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2061static int init_header_complete(struct swsusp_info *info)
2062{
2063	memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2064	info->version_code = LINUX_VERSION_CODE;
2065	return 0;
2066}
2067
2068static const char *check_image_kernel(struct swsusp_info *info)
2069{
2070	if (info->version_code != LINUX_VERSION_CODE)
2071		return "kernel version";
2072	if (strcmp(info->uts.sysname,init_utsname()->sysname))
2073		return "system type";
2074	if (strcmp(info->uts.release,init_utsname()->release))
2075		return "kernel release";
2076	if (strcmp(info->uts.version,init_utsname()->version))
2077		return "version";
2078	if (strcmp(info->uts.machine,init_utsname()->machine))
2079		return "machine";
2080	return NULL;
2081}
2082#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2083
2084unsigned long snapshot_get_image_size(void)
2085{
2086	return nr_copy_pages + nr_meta_pages + 1;
2087}
2088
2089static int init_header(struct swsusp_info *info)
2090{
2091	memset(info, 0, sizeof(struct swsusp_info));
2092	info->num_physpages = get_num_physpages();
2093	info->image_pages = nr_copy_pages;
2094	info->pages = snapshot_get_image_size();
2095	info->size = info->pages;
2096	info->size <<= PAGE_SHIFT;
2097	return init_header_complete(info);
2098}
2099
2100/**
2101 * pack_pfns - Prepare PFNs for saving.
2102 * @bm: Memory bitmap.
2103 * @buf: Memory buffer to store the PFNs in.
2104 *
2105 * PFNs corresponding to set bits in @bm are stored in the area of memory
2106 * pointed to by @buf (1 page at a time).
2107 */
2108static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2109{
2110	int j;
2111
2112	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2113		buf[j] = memory_bm_next_pfn(bm);
2114		if (unlikely(buf[j] == BM_END_OF_MAP))
2115			break;
2116	}
2117}
2118
2119/**
2120 * snapshot_read_next - Get the address to read the next image page from.
2121 * @handle: Snapshot handle to be used for the reading.
2122 *
2123 * On the first call, @handle should point to a zeroed snapshot_handle
2124 * structure.  The structure gets populated then and a pointer to it should be
2125 * passed to this function every next time.
2126 *
2127 * On success, the function returns a positive number.  Then, the caller
2128 * is allowed to read up to the returned number of bytes from the memory
2129 * location computed by the data_of() macro.
2130 *
2131 * The function returns 0 to indicate the end of the data stream condition,
2132 * and negative numbers are returned on errors.  If that happens, the structure
2133 * pointed to by @handle is not updated and should not be used any more.
2134 */
2135int snapshot_read_next(struct snapshot_handle *handle)
2136{
2137	if (handle->cur > nr_meta_pages + nr_copy_pages)
2138		return 0;
2139
2140	if (!buffer) {
2141		/* This makes the buffer be freed by swsusp_free() */
2142		buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2143		if (!buffer)
2144			return -ENOMEM;
2145	}
2146	if (!handle->cur) {
2147		int error;
2148
2149		error = init_header((struct swsusp_info *)buffer);
2150		if (error)
2151			return error;
2152		handle->buffer = buffer;
2153		memory_bm_position_reset(&orig_bm);
2154		memory_bm_position_reset(&copy_bm);
2155	} else if (handle->cur <= nr_meta_pages) {
2156		clear_page(buffer);
2157		pack_pfns(buffer, &orig_bm);
2158	} else {
2159		struct page *page;
2160
2161		page = pfn_to_page(memory_bm_next_pfn(&copy_bm));
2162		if (PageHighMem(page)) {
2163			/*
2164			 * Highmem pages are copied to the buffer,
2165			 * because we can't return with a kmapped
2166			 * highmem page (we may not be called again).
2167			 */
2168			void *kaddr;
2169
2170			kaddr = kmap_atomic(page);
2171			copy_page(buffer, kaddr);
2172			kunmap_atomic(kaddr);
2173			handle->buffer = buffer;
2174		} else {
2175			handle->buffer = page_address(page);
2176		}
2177	}
2178	handle->cur++;
2179	return PAGE_SIZE;
2180}
2181
2182static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2183				    struct memory_bitmap *src)
2184{
2185	unsigned long pfn;
2186
2187	memory_bm_position_reset(src);
2188	pfn = memory_bm_next_pfn(src);
2189	while (pfn != BM_END_OF_MAP) {
2190		memory_bm_set_bit(dst, pfn);
2191		pfn = memory_bm_next_pfn(src);
2192	}
2193}
2194
2195/**
2196 * mark_unsafe_pages - Mark pages that were used before hibernation.
2197 *
2198 * Mark the pages that cannot be used for storing the image during restoration,
2199 * because they conflict with the pages that had been used before hibernation.
2200 */
2201static void mark_unsafe_pages(struct memory_bitmap *bm)
2202{
2203	unsigned long pfn;
2204
2205	/* Clear the "free"/"unsafe" bit for all PFNs */
2206	memory_bm_position_reset(free_pages_map);
2207	pfn = memory_bm_next_pfn(free_pages_map);
2208	while (pfn != BM_END_OF_MAP) {
2209		memory_bm_clear_current(free_pages_map);
2210		pfn = memory_bm_next_pfn(free_pages_map);
2211	}
2212
2213	/* Mark pages that correspond to the "original" PFNs as "unsafe" */
2214	duplicate_memory_bitmap(free_pages_map, bm);
2215
2216	allocated_unsafe_pages = 0;
2217}
2218
2219static int check_header(struct swsusp_info *info)
2220{
2221	const char *reason;
2222
2223	reason = check_image_kernel(info);
2224	if (!reason && info->num_physpages != get_num_physpages())
2225		reason = "memory size";
2226	if (reason) {
2227		pr_err("Image mismatch: %s\n", reason);
2228		return -EPERM;
2229	}
2230	return 0;
2231}
2232
2233/**
2234 * load header - Check the image header and copy the data from it.
2235 */
2236static int load_header(struct swsusp_info *info)
2237{
2238	int error;
2239
2240	restore_pblist = NULL;
2241	error = check_header(info);
2242	if (!error) {
2243		nr_copy_pages = info->image_pages;
2244		nr_meta_pages = info->pages - info->image_pages - 1;
2245	}
2246	return error;
2247}
2248
2249/**
2250 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2251 * @bm: Memory bitmap.
2252 * @buf: Area of memory containing the PFNs.
2253 *
2254 * For each element of the array pointed to by @buf (1 page at a time), set the
2255 * corresponding bit in @bm.
2256 */
2257static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2258{
2259	int j;
2260
2261	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2262		if (unlikely(buf[j] == BM_END_OF_MAP))
2263			break;
2264
2265		if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2266			memory_bm_set_bit(bm, buf[j]);
2267		else
 
 
 
2268			return -EFAULT;
 
2269	}
2270
2271	return 0;
2272}
2273
2274#ifdef CONFIG_HIGHMEM
2275/*
2276 * struct highmem_pbe is used for creating the list of highmem pages that
2277 * should be restored atomically during the resume from disk, because the page
2278 * frames they have occupied before the suspend are in use.
2279 */
2280struct highmem_pbe {
2281	struct page *copy_page;	/* data is here now */
2282	struct page *orig_page;	/* data was here before the suspend */
2283	struct highmem_pbe *next;
2284};
2285
2286/*
2287 * List of highmem PBEs needed for restoring the highmem pages that were
2288 * allocated before the suspend and included in the suspend image, but have
2289 * also been allocated by the "resume" kernel, so their contents cannot be
2290 * written directly to their "original" page frames.
2291 */
2292static struct highmem_pbe *highmem_pblist;
2293
2294/**
2295 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2296 * @bm: Memory bitmap.
2297 *
2298 * The bits in @bm that correspond to image pages are assumed to be set.
2299 */
2300static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2301{
2302	unsigned long pfn;
2303	unsigned int cnt = 0;
2304
2305	memory_bm_position_reset(bm);
2306	pfn = memory_bm_next_pfn(bm);
2307	while (pfn != BM_END_OF_MAP) {
2308		if (PageHighMem(pfn_to_page(pfn)))
2309			cnt++;
2310
2311		pfn = memory_bm_next_pfn(bm);
2312	}
2313	return cnt;
2314}
2315
2316static unsigned int safe_highmem_pages;
2317
2318static struct memory_bitmap *safe_highmem_bm;
2319
2320/**
2321 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2322 * @bm: Pointer to an uninitialized memory bitmap structure.
2323 * @nr_highmem_p: Pointer to the number of highmem image pages.
2324 *
2325 * Try to allocate as many highmem pages as there are highmem image pages
2326 * (@nr_highmem_p points to the variable containing the number of highmem image
2327 * pages).  The pages that are "safe" (ie. will not be overwritten when the
2328 * hibernation image is restored entirely) have the corresponding bits set in
2329 * @bm (it must be uninitialized).
2330 *
2331 * NOTE: This function should not be called if there are no highmem image pages.
2332 */
2333static int prepare_highmem_image(struct memory_bitmap *bm,
2334				 unsigned int *nr_highmem_p)
2335{
2336	unsigned int to_alloc;
2337
2338	if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2339		return -ENOMEM;
2340
2341	if (get_highmem_buffer(PG_SAFE))
2342		return -ENOMEM;
2343
2344	to_alloc = count_free_highmem_pages();
2345	if (to_alloc > *nr_highmem_p)
2346		to_alloc = *nr_highmem_p;
2347	else
2348		*nr_highmem_p = to_alloc;
2349
2350	safe_highmem_pages = 0;
2351	while (to_alloc-- > 0) {
2352		struct page *page;
2353
2354		page = alloc_page(__GFP_HIGHMEM);
2355		if (!swsusp_page_is_free(page)) {
2356			/* The page is "safe", set its bit the bitmap */
2357			memory_bm_set_bit(bm, page_to_pfn(page));
2358			safe_highmem_pages++;
2359		}
2360		/* Mark the page as allocated */
2361		swsusp_set_page_forbidden(page);
2362		swsusp_set_page_free(page);
2363	}
2364	memory_bm_position_reset(bm);
2365	safe_highmem_bm = bm;
2366	return 0;
2367}
2368
2369static struct page *last_highmem_page;
2370
2371/**
2372 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2373 *
2374 * For a given highmem image page get a buffer that suspend_write_next() should
2375 * return to its caller to write to.
2376 *
2377 * If the page is to be saved to its "original" page frame or a copy of
2378 * the page is to be made in the highmem, @buffer is returned.  Otherwise,
2379 * the copy of the page is to be made in normal memory, so the address of
2380 * the copy is returned.
2381 *
2382 * If @buffer is returned, the caller of suspend_write_next() will write
2383 * the page's contents to @buffer, so they will have to be copied to the
2384 * right location on the next call to suspend_write_next() and it is done
2385 * with the help of copy_last_highmem_page().  For this purpose, if
2386 * @buffer is returned, @last_highmem_page is set to the page to which
2387 * the data will have to be copied from @buffer.
2388 */
2389static void *get_highmem_page_buffer(struct page *page,
2390				     struct chain_allocator *ca)
2391{
2392	struct highmem_pbe *pbe;
2393	void *kaddr;
2394
2395	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2396		/*
2397		 * We have allocated the "original" page frame and we can
2398		 * use it directly to store the loaded page.
2399		 */
2400		last_highmem_page = page;
2401		return buffer;
2402	}
2403	/*
2404	 * The "original" page frame has not been allocated and we have to
2405	 * use a "safe" page frame to store the loaded page.
2406	 */
2407	pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2408	if (!pbe) {
2409		swsusp_free();
2410		return ERR_PTR(-ENOMEM);
2411	}
2412	pbe->orig_page = page;
2413	if (safe_highmem_pages > 0) {
2414		struct page *tmp;
2415
2416		/* Copy of the page will be stored in high memory */
2417		kaddr = buffer;
2418		tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2419		safe_highmem_pages--;
2420		last_highmem_page = tmp;
2421		pbe->copy_page = tmp;
2422	} else {
2423		/* Copy of the page will be stored in normal memory */
2424		kaddr = safe_pages_list;
2425		safe_pages_list = safe_pages_list->next;
2426		pbe->copy_page = virt_to_page(kaddr);
2427	}
2428	pbe->next = highmem_pblist;
2429	highmem_pblist = pbe;
2430	return kaddr;
2431}
2432
2433/**
2434 * copy_last_highmem_page - Copy most the most recent highmem image page.
2435 *
2436 * Copy the contents of a highmem image from @buffer, where the caller of
2437 * snapshot_write_next() has stored them, to the right location represented by
2438 * @last_highmem_page .
2439 */
2440static void copy_last_highmem_page(void)
2441{
2442	if (last_highmem_page) {
2443		void *dst;
2444
2445		dst = kmap_atomic(last_highmem_page);
2446		copy_page(dst, buffer);
2447		kunmap_atomic(dst);
2448		last_highmem_page = NULL;
2449	}
2450}
2451
2452static inline int last_highmem_page_copied(void)
2453{
2454	return !last_highmem_page;
2455}
2456
2457static inline void free_highmem_data(void)
2458{
2459	if (safe_highmem_bm)
2460		memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2461
2462	if (buffer)
2463		free_image_page(buffer, PG_UNSAFE_CLEAR);
2464}
2465#else
2466static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2467
2468static inline int prepare_highmem_image(struct memory_bitmap *bm,
2469					unsigned int *nr_highmem_p) { return 0; }
2470
2471static inline void *get_highmem_page_buffer(struct page *page,
2472					    struct chain_allocator *ca)
2473{
2474	return ERR_PTR(-EINVAL);
2475}
2476
2477static inline void copy_last_highmem_page(void) {}
2478static inline int last_highmem_page_copied(void) { return 1; }
2479static inline void free_highmem_data(void) {}
2480#endif /* CONFIG_HIGHMEM */
2481
2482#define PBES_PER_LINKED_PAGE	(LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2483
2484/**
2485 * prepare_image - Make room for loading hibernation image.
2486 * @new_bm: Uninitialized memory bitmap structure.
2487 * @bm: Memory bitmap with unsafe pages marked.
2488 *
2489 * Use @bm to mark the pages that will be overwritten in the process of
2490 * restoring the system memory state from the suspend image ("unsafe" pages)
2491 * and allocate memory for the image.
2492 *
2493 * The idea is to allocate a new memory bitmap first and then allocate
2494 * as many pages as needed for image data, but without specifying what those
2495 * pages will be used for just yet.  Instead, we mark them all as allocated and
2496 * create a lists of "safe" pages to be used later.  On systems with high
2497 * memory a list of "safe" highmem pages is created too.
2498 */
2499static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2500{
2501	unsigned int nr_pages, nr_highmem;
2502	struct linked_page *lp;
2503	int error;
2504
2505	/* If there is no highmem, the buffer will not be necessary */
2506	free_image_page(buffer, PG_UNSAFE_CLEAR);
2507	buffer = NULL;
2508
2509	nr_highmem = count_highmem_image_pages(bm);
2510	mark_unsafe_pages(bm);
2511
2512	error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2513	if (error)
2514		goto Free;
2515
2516	duplicate_memory_bitmap(new_bm, bm);
2517	memory_bm_free(bm, PG_UNSAFE_KEEP);
2518	if (nr_highmem > 0) {
2519		error = prepare_highmem_image(bm, &nr_highmem);
2520		if (error)
2521			goto Free;
2522	}
2523	/*
2524	 * Reserve some safe pages for potential later use.
2525	 *
2526	 * NOTE: This way we make sure there will be enough safe pages for the
2527	 * chain_alloc() in get_buffer().  It is a bit wasteful, but
2528	 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2529	 *
2530	 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2531	 */
2532	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2533	nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2534	while (nr_pages > 0) {
2535		lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2536		if (!lp) {
2537			error = -ENOMEM;
2538			goto Free;
2539		}
2540		lp->next = safe_pages_list;
2541		safe_pages_list = lp;
2542		nr_pages--;
2543	}
2544	/* Preallocate memory for the image */
2545	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2546	while (nr_pages > 0) {
2547		lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2548		if (!lp) {
2549			error = -ENOMEM;
2550			goto Free;
2551		}
2552		if (!swsusp_page_is_free(virt_to_page(lp))) {
2553			/* The page is "safe", add it to the list */
2554			lp->next = safe_pages_list;
2555			safe_pages_list = lp;
2556		}
2557		/* Mark the page as allocated */
2558		swsusp_set_page_forbidden(virt_to_page(lp));
2559		swsusp_set_page_free(virt_to_page(lp));
2560		nr_pages--;
2561	}
2562	return 0;
2563
2564 Free:
2565	swsusp_free();
2566	return error;
2567}
2568
2569/**
2570 * get_buffer - Get the address to store the next image data page.
2571 *
2572 * Get the address that snapshot_write_next() should return to its caller to
2573 * write to.
2574 */
2575static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2576{
2577	struct pbe *pbe;
2578	struct page *page;
2579	unsigned long pfn = memory_bm_next_pfn(bm);
2580
2581	if (pfn == BM_END_OF_MAP)
2582		return ERR_PTR(-EFAULT);
2583
2584	page = pfn_to_page(pfn);
2585	if (PageHighMem(page))
2586		return get_highmem_page_buffer(page, ca);
2587
2588	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2589		/*
2590		 * We have allocated the "original" page frame and we can
2591		 * use it directly to store the loaded page.
2592		 */
2593		return page_address(page);
2594
2595	/*
2596	 * The "original" page frame has not been allocated and we have to
2597	 * use a "safe" page frame to store the loaded page.
2598	 */
2599	pbe = chain_alloc(ca, sizeof(struct pbe));
2600	if (!pbe) {
2601		swsusp_free();
2602		return ERR_PTR(-ENOMEM);
2603	}
2604	pbe->orig_address = page_address(page);
2605	pbe->address = safe_pages_list;
2606	safe_pages_list = safe_pages_list->next;
2607	pbe->next = restore_pblist;
2608	restore_pblist = pbe;
2609	return pbe->address;
2610}
2611
2612/**
2613 * snapshot_write_next - Get the address to store the next image page.
2614 * @handle: Snapshot handle structure to guide the writing.
2615 *
2616 * On the first call, @handle should point to a zeroed snapshot_handle
2617 * structure.  The structure gets populated then and a pointer to it should be
2618 * passed to this function every next time.
2619 *
2620 * On success, the function returns a positive number.  Then, the caller
2621 * is allowed to write up to the returned number of bytes to the memory
2622 * location computed by the data_of() macro.
2623 *
2624 * The function returns 0 to indicate the "end of file" condition.  Negative
2625 * numbers are returned on errors, in which cases the structure pointed to by
2626 * @handle is not updated and should not be used any more.
2627 */
2628int snapshot_write_next(struct snapshot_handle *handle)
2629{
2630	static struct chain_allocator ca;
2631	int error = 0;
2632
2633	/* Check if we have already loaded the entire image */
2634	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2635		return 0;
2636
2637	handle->sync_read = 1;
2638
2639	if (!handle->cur) {
2640		if (!buffer)
2641			/* This makes the buffer be freed by swsusp_free() */
2642			buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2643
2644		if (!buffer)
2645			return -ENOMEM;
2646
2647		handle->buffer = buffer;
2648	} else if (handle->cur == 1) {
2649		error = load_header(buffer);
2650		if (error)
2651			return error;
2652
2653		safe_pages_list = NULL;
2654
2655		error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
2656		if (error)
2657			return error;
2658
2659		hibernate_restore_protection_begin();
2660	} else if (handle->cur <= nr_meta_pages + 1) {
2661		error = unpack_orig_pfns(buffer, &copy_bm);
2662		if (error)
2663			return error;
2664
2665		if (handle->cur == nr_meta_pages + 1) {
2666			error = prepare_image(&orig_bm, &copy_bm);
2667			if (error)
2668				return error;
2669
2670			chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2671			memory_bm_position_reset(&orig_bm);
2672			restore_pblist = NULL;
2673			handle->buffer = get_buffer(&orig_bm, &ca);
2674			handle->sync_read = 0;
2675			if (IS_ERR(handle->buffer))
2676				return PTR_ERR(handle->buffer);
2677		}
2678	} else {
2679		copy_last_highmem_page();
2680		hibernate_restore_protect_page(handle->buffer);
2681		handle->buffer = get_buffer(&orig_bm, &ca);
2682		if (IS_ERR(handle->buffer))
2683			return PTR_ERR(handle->buffer);
2684		if (handle->buffer != buffer)
2685			handle->sync_read = 0;
2686	}
2687	handle->cur++;
2688	return PAGE_SIZE;
2689}
2690
2691/**
2692 * snapshot_write_finalize - Complete the loading of a hibernation image.
2693 *
2694 * Must be called after the last call to snapshot_write_next() in case the last
2695 * page in the image happens to be a highmem page and its contents should be
2696 * stored in highmem.  Additionally, it recycles bitmap memory that's not
2697 * necessary any more.
2698 */
2699void snapshot_write_finalize(struct snapshot_handle *handle)
2700{
2701	copy_last_highmem_page();
2702	hibernate_restore_protect_page(handle->buffer);
2703	/* Do that only if we have loaded the image entirely */
2704	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2705		memory_bm_recycle(&orig_bm);
2706		free_highmem_data();
2707	}
2708}
2709
2710int snapshot_image_loaded(struct snapshot_handle *handle)
2711{
2712	return !(!nr_copy_pages || !last_highmem_page_copied() ||
2713			handle->cur <= nr_meta_pages + nr_copy_pages);
2714}
2715
2716#ifdef CONFIG_HIGHMEM
2717/* Assumes that @buf is ready and points to a "safe" page */
2718static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2719				       void *buf)
2720{
2721	void *kaddr1, *kaddr2;
2722
2723	kaddr1 = kmap_atomic(p1);
2724	kaddr2 = kmap_atomic(p2);
2725	copy_page(buf, kaddr1);
2726	copy_page(kaddr1, kaddr2);
2727	copy_page(kaddr2, buf);
2728	kunmap_atomic(kaddr2);
2729	kunmap_atomic(kaddr1);
2730}
2731
2732/**
2733 * restore_highmem - Put highmem image pages into their original locations.
2734 *
2735 * For each highmem page that was in use before hibernation and is included in
2736 * the image, and also has been allocated by the "restore" kernel, swap its
2737 * current contents with the previous (ie. "before hibernation") ones.
2738 *
2739 * If the restore eventually fails, we can call this function once again and
2740 * restore the highmem state as seen by the restore kernel.
2741 */
2742int restore_highmem(void)
2743{
2744	struct highmem_pbe *pbe = highmem_pblist;
2745	void *buf;
2746
2747	if (!pbe)
2748		return 0;
2749
2750	buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2751	if (!buf)
2752		return -ENOMEM;
2753
2754	while (pbe) {
2755		swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2756		pbe = pbe->next;
2757	}
2758	free_image_page(buf, PG_UNSAFE_CLEAR);
2759	return 0;
2760}
2761#endif /* CONFIG_HIGHMEM */