Linux Audio

Check our new training course

Loading...
v5.9
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * linux/kernel/power/snapshot.c
   4 *
   5 * This file provides system snapshot/restore functionality for swsusp.
   6 *
   7 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
   8 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
   9 */
  10
  11#define pr_fmt(fmt) "PM: hibernation: " fmt
  12
  13#include <linux/version.h>
  14#include <linux/module.h>
  15#include <linux/mm.h>
  16#include <linux/suspend.h>
  17#include <linux/delay.h>
  18#include <linux/bitops.h>
  19#include <linux/spinlock.h>
  20#include <linux/kernel.h>
  21#include <linux/pm.h>
  22#include <linux/device.h>
  23#include <linux/init.h>
  24#include <linux/memblock.h>
  25#include <linux/nmi.h>
  26#include <linux/syscalls.h>
  27#include <linux/console.h>
  28#include <linux/highmem.h>
  29#include <linux/list.h>
  30#include <linux/slab.h>
  31#include <linux/compiler.h>
  32#include <linux/ktime.h>
  33#include <linux/set_memory.h>
  34
  35#include <linux/uaccess.h>
  36#include <asm/mmu_context.h>
 
  37#include <asm/tlbflush.h>
  38#include <asm/io.h>
  39
  40#include "power.h"
  41
  42#if defined(CONFIG_STRICT_KERNEL_RWX) && defined(CONFIG_ARCH_HAS_SET_MEMORY)
  43static bool hibernate_restore_protection;
  44static bool hibernate_restore_protection_active;
  45
  46void enable_restore_image_protection(void)
  47{
  48	hibernate_restore_protection = true;
  49}
  50
  51static inline void hibernate_restore_protection_begin(void)
  52{
  53	hibernate_restore_protection_active = hibernate_restore_protection;
  54}
  55
  56static inline void hibernate_restore_protection_end(void)
  57{
  58	hibernate_restore_protection_active = false;
  59}
  60
  61static inline void hibernate_restore_protect_page(void *page_address)
  62{
  63	if (hibernate_restore_protection_active)
  64		set_memory_ro((unsigned long)page_address, 1);
  65}
  66
  67static inline void hibernate_restore_unprotect_page(void *page_address)
  68{
  69	if (hibernate_restore_protection_active)
  70		set_memory_rw((unsigned long)page_address, 1);
  71}
  72#else
  73static inline void hibernate_restore_protection_begin(void) {}
  74static inline void hibernate_restore_protection_end(void) {}
  75static inline void hibernate_restore_protect_page(void *page_address) {}
  76static inline void hibernate_restore_unprotect_page(void *page_address) {}
  77#endif /* CONFIG_STRICT_KERNEL_RWX  && CONFIG_ARCH_HAS_SET_MEMORY */
  78
  79static int swsusp_page_is_free(struct page *);
  80static void swsusp_set_page_forbidden(struct page *);
  81static void swsusp_unset_page_forbidden(struct page *);
  82
  83/*
  84 * Number of bytes to reserve for memory allocations made by device drivers
  85 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
  86 * cause image creation to fail (tunable via /sys/power/reserved_size).
  87 */
  88unsigned long reserved_size;
  89
  90void __init hibernate_reserved_size_init(void)
  91{
  92	reserved_size = SPARE_PAGES * PAGE_SIZE;
  93}
  94
  95/*
  96 * Preferred image size in bytes (tunable via /sys/power/image_size).
  97 * When it is set to N, swsusp will do its best to ensure the image
  98 * size will not exceed N bytes, but if that is impossible, it will
  99 * try to create the smallest image possible.
 100 */
 101unsigned long image_size;
 102
 103void __init hibernate_image_size_init(void)
 104{
 105	image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
 106}
 107
 108/*
 109 * List of PBEs needed for restoring the pages that were allocated before
 110 * the suspend and included in the suspend image, but have also been
 111 * allocated by the "resume" kernel, so their contents cannot be written
 112 * directly to their "original" page frames.
 113 */
 114struct pbe *restore_pblist;
 115
 116/* struct linked_page is used to build chains of pages */
 117
 118#define LINKED_PAGE_DATA_SIZE	(PAGE_SIZE - sizeof(void *))
 119
 120struct linked_page {
 121	struct linked_page *next;
 122	char data[LINKED_PAGE_DATA_SIZE];
 123} __packed;
 124
 125/*
 126 * List of "safe" pages (ie. pages that were not used by the image kernel
 127 * before hibernation) that may be used as temporary storage for image kernel
 128 * memory contents.
 129 */
 130static struct linked_page *safe_pages_list;
 131
 132/* Pointer to an auxiliary buffer (1 page) */
 133static void *buffer;
 134
 135#define PG_ANY		0
 136#define PG_SAFE		1
 137#define PG_UNSAFE_CLEAR	1
 138#define PG_UNSAFE_KEEP	0
 139
 140static unsigned int allocated_unsafe_pages;
 141
 142/**
 143 * get_image_page - Allocate a page for a hibernation image.
 144 * @gfp_mask: GFP mask for the allocation.
 145 * @safe_needed: Get pages that were not used before hibernation (restore only)
 146 *
 147 * During image restoration, for storing the PBE list and the image data, we can
 148 * only use memory pages that do not conflict with the pages used before
 149 * hibernation.  The "unsafe" pages have PageNosaveFree set and we count them
 150 * using allocated_unsafe_pages.
 151 *
 152 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
 153 * swsusp_free() can release it.
 154 */
 155static void *get_image_page(gfp_t gfp_mask, int safe_needed)
 156{
 157	void *res;
 158
 159	res = (void *)get_zeroed_page(gfp_mask);
 160	if (safe_needed)
 161		while (res && swsusp_page_is_free(virt_to_page(res))) {
 162			/* The page is unsafe, mark it for swsusp_free() */
 163			swsusp_set_page_forbidden(virt_to_page(res));
 164			allocated_unsafe_pages++;
 165			res = (void *)get_zeroed_page(gfp_mask);
 166		}
 167	if (res) {
 168		swsusp_set_page_forbidden(virt_to_page(res));
 169		swsusp_set_page_free(virt_to_page(res));
 170	}
 171	return res;
 172}
 173
 174static void *__get_safe_page(gfp_t gfp_mask)
 175{
 176	if (safe_pages_list) {
 177		void *ret = safe_pages_list;
 178
 179		safe_pages_list = safe_pages_list->next;
 180		memset(ret, 0, PAGE_SIZE);
 181		return ret;
 182	}
 183	return get_image_page(gfp_mask, PG_SAFE);
 184}
 185
 186unsigned long get_safe_page(gfp_t gfp_mask)
 187{
 188	return (unsigned long)__get_safe_page(gfp_mask);
 189}
 190
 191static struct page *alloc_image_page(gfp_t gfp_mask)
 192{
 193	struct page *page;
 194
 195	page = alloc_page(gfp_mask);
 196	if (page) {
 197		swsusp_set_page_forbidden(page);
 198		swsusp_set_page_free(page);
 199	}
 200	return page;
 201}
 202
 203static void recycle_safe_page(void *page_address)
 204{
 205	struct linked_page *lp = page_address;
 206
 207	lp->next = safe_pages_list;
 208	safe_pages_list = lp;
 209}
 210
 211/**
 212 * free_image_page - Free a page allocated for hibernation image.
 213 * @addr: Address of the page to free.
 214 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
 215 *
 216 * The page to free should have been allocated by get_image_page() (page flags
 217 * set by it are affected).
 218 */
 219static inline void free_image_page(void *addr, int clear_nosave_free)
 220{
 221	struct page *page;
 222
 223	BUG_ON(!virt_addr_valid(addr));
 224
 225	page = virt_to_page(addr);
 226
 227	swsusp_unset_page_forbidden(page);
 228	if (clear_nosave_free)
 229		swsusp_unset_page_free(page);
 230
 231	__free_page(page);
 232}
 233
 234static inline void free_list_of_pages(struct linked_page *list,
 235				      int clear_page_nosave)
 236{
 237	while (list) {
 238		struct linked_page *lp = list->next;
 239
 240		free_image_page(list, clear_page_nosave);
 241		list = lp;
 242	}
 243}
 244
 245/*
 246 * struct chain_allocator is used for allocating small objects out of
 247 * a linked list of pages called 'the chain'.
 248 *
 249 * The chain grows each time when there is no room for a new object in
 250 * the current page.  The allocated objects cannot be freed individually.
 251 * It is only possible to free them all at once, by freeing the entire
 252 * chain.
 253 *
 254 * NOTE: The chain allocator may be inefficient if the allocated objects
 255 * are not much smaller than PAGE_SIZE.
 256 */
 257struct chain_allocator {
 258	struct linked_page *chain;	/* the chain */
 259	unsigned int used_space;	/* total size of objects allocated out
 260					   of the current page */
 261	gfp_t gfp_mask;		/* mask for allocating pages */
 262	int safe_needed;	/* if set, only "safe" pages are allocated */
 263};
 264
 265static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
 266		       int safe_needed)
 267{
 268	ca->chain = NULL;
 269	ca->used_space = LINKED_PAGE_DATA_SIZE;
 270	ca->gfp_mask = gfp_mask;
 271	ca->safe_needed = safe_needed;
 272}
 273
 274static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
 275{
 276	void *ret;
 277
 278	if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
 279		struct linked_page *lp;
 280
 281		lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
 282					get_image_page(ca->gfp_mask, PG_ANY);
 283		if (!lp)
 284			return NULL;
 285
 286		lp->next = ca->chain;
 287		ca->chain = lp;
 288		ca->used_space = 0;
 289	}
 290	ret = ca->chain->data + ca->used_space;
 291	ca->used_space += size;
 292	return ret;
 293}
 294
 295/**
 296 * Data types related to memory bitmaps.
 297 *
 298 * Memory bitmap is a structure consiting of many linked lists of
 299 * objects.  The main list's elements are of type struct zone_bitmap
 300 * and each of them corresonds to one zone.  For each zone bitmap
 301 * object there is a list of objects of type struct bm_block that
 302 * represent each blocks of bitmap in which information is stored.
 303 *
 304 * struct memory_bitmap contains a pointer to the main list of zone
 305 * bitmap objects, a struct bm_position used for browsing the bitmap,
 306 * and a pointer to the list of pages used for allocating all of the
 307 * zone bitmap objects and bitmap block objects.
 308 *
 309 * NOTE: It has to be possible to lay out the bitmap in memory
 310 * using only allocations of order 0.  Additionally, the bitmap is
 311 * designed to work with arbitrary number of zones (this is over the
 312 * top for now, but let's avoid making unnecessary assumptions ;-).
 313 *
 314 * struct zone_bitmap contains a pointer to a list of bitmap block
 315 * objects and a pointer to the bitmap block object that has been
 316 * most recently used for setting bits.  Additionally, it contains the
 317 * PFNs that correspond to the start and end of the represented zone.
 318 *
 319 * struct bm_block contains a pointer to the memory page in which
 320 * information is stored (in the form of a block of bitmap)
 321 * It also contains the pfns that correspond to the start and end of
 322 * the represented memory area.
 323 *
 324 * The memory bitmap is organized as a radix tree to guarantee fast random
 325 * access to the bits. There is one radix tree for each zone (as returned
 326 * from create_mem_extents).
 327 *
 328 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
 329 * two linked lists for the nodes of the tree, one for the inner nodes and
 330 * one for the leave nodes. The linked leave nodes are used for fast linear
 331 * access of the memory bitmap.
 332 *
 333 * The struct rtree_node represents one node of the radix tree.
 334 */
 335
 336#define BM_END_OF_MAP	(~0UL)
 337
 338#define BM_BITS_PER_BLOCK	(PAGE_SIZE * BITS_PER_BYTE)
 339#define BM_BLOCK_SHIFT		(PAGE_SHIFT + 3)
 340#define BM_BLOCK_MASK		((1UL << BM_BLOCK_SHIFT) - 1)
 341
 342/*
 343 * struct rtree_node is a wrapper struct to link the nodes
 344 * of the rtree together for easy linear iteration over
 345 * bits and easy freeing
 346 */
 347struct rtree_node {
 348	struct list_head list;
 349	unsigned long *data;
 350};
 351
 352/*
 353 * struct mem_zone_bm_rtree represents a bitmap used for one
 354 * populated memory zone.
 355 */
 356struct mem_zone_bm_rtree {
 357	struct list_head list;		/* Link Zones together         */
 358	struct list_head nodes;		/* Radix Tree inner nodes      */
 359	struct list_head leaves;	/* Radix Tree leaves           */
 360	unsigned long start_pfn;	/* Zone start page frame       */
 361	unsigned long end_pfn;		/* Zone end page frame + 1     */
 362	struct rtree_node *rtree;	/* Radix Tree Root             */
 363	int levels;			/* Number of Radix Tree Levels */
 364	unsigned int blocks;		/* Number of Bitmap Blocks     */
 365};
 366
 367/* strcut bm_position is used for browsing memory bitmaps */
 368
 369struct bm_position {
 370	struct mem_zone_bm_rtree *zone;
 371	struct rtree_node *node;
 372	unsigned long node_pfn;
 373	int node_bit;
 374};
 375
 376struct memory_bitmap {
 377	struct list_head zones;
 378	struct linked_page *p_list;	/* list of pages used to store zone
 379					   bitmap objects and bitmap block
 380					   objects */
 381	struct bm_position cur;	/* most recently used bit position */
 382};
 383
 384/* Functions that operate on memory bitmaps */
 385
 386#define BM_ENTRIES_PER_LEVEL	(PAGE_SIZE / sizeof(unsigned long))
 387#if BITS_PER_LONG == 32
 388#define BM_RTREE_LEVEL_SHIFT	(PAGE_SHIFT - 2)
 389#else
 390#define BM_RTREE_LEVEL_SHIFT	(PAGE_SHIFT - 3)
 391#endif
 392#define BM_RTREE_LEVEL_MASK	((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
 393
 394/**
 395 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
 396 *
 397 * This function is used to allocate inner nodes as well as the
 398 * leave nodes of the radix tree. It also adds the node to the
 399 * corresponding linked list passed in by the *list parameter.
 400 */
 401static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
 402					   struct chain_allocator *ca,
 403					   struct list_head *list)
 404{
 405	struct rtree_node *node;
 406
 407	node = chain_alloc(ca, sizeof(struct rtree_node));
 408	if (!node)
 409		return NULL;
 410
 411	node->data = get_image_page(gfp_mask, safe_needed);
 412	if (!node->data)
 413		return NULL;
 414
 415	list_add_tail(&node->list, list);
 416
 417	return node;
 418}
 419
 420/**
 421 * add_rtree_block - Add a new leave node to the radix tree.
 422 *
 423 * The leave nodes need to be allocated in order to keep the leaves
 424 * linked list in order. This is guaranteed by the zone->blocks
 425 * counter.
 426 */
 427static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
 428			   int safe_needed, struct chain_allocator *ca)
 429{
 430	struct rtree_node *node, *block, **dst;
 431	unsigned int levels_needed, block_nr;
 432	int i;
 433
 434	block_nr = zone->blocks;
 435	levels_needed = 0;
 436
 437	/* How many levels do we need for this block nr? */
 438	while (block_nr) {
 439		levels_needed += 1;
 440		block_nr >>= BM_RTREE_LEVEL_SHIFT;
 441	}
 442
 443	/* Make sure the rtree has enough levels */
 444	for (i = zone->levels; i < levels_needed; i++) {
 445		node = alloc_rtree_node(gfp_mask, safe_needed, ca,
 446					&zone->nodes);
 447		if (!node)
 448			return -ENOMEM;
 449
 450		node->data[0] = (unsigned long)zone->rtree;
 451		zone->rtree = node;
 452		zone->levels += 1;
 453	}
 454
 455	/* Allocate new block */
 456	block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
 457	if (!block)
 458		return -ENOMEM;
 459
 460	/* Now walk the rtree to insert the block */
 461	node = zone->rtree;
 462	dst = &zone->rtree;
 463	block_nr = zone->blocks;
 464	for (i = zone->levels; i > 0; i--) {
 465		int index;
 466
 467		if (!node) {
 468			node = alloc_rtree_node(gfp_mask, safe_needed, ca,
 469						&zone->nodes);
 470			if (!node)
 471				return -ENOMEM;
 472			*dst = node;
 473		}
 474
 475		index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
 476		index &= BM_RTREE_LEVEL_MASK;
 477		dst = (struct rtree_node **)&((*dst)->data[index]);
 478		node = *dst;
 479	}
 480
 481	zone->blocks += 1;
 482	*dst = block;
 483
 484	return 0;
 485}
 486
 487static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
 488			       int clear_nosave_free);
 489
 490/**
 491 * create_zone_bm_rtree - Create a radix tree for one zone.
 492 *
 493 * Allocated the mem_zone_bm_rtree structure and initializes it.
 494 * This function also allocated and builds the radix tree for the
 495 * zone.
 496 */
 497static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
 498						      int safe_needed,
 499						      struct chain_allocator *ca,
 500						      unsigned long start,
 501						      unsigned long end)
 502{
 503	struct mem_zone_bm_rtree *zone;
 504	unsigned int i, nr_blocks;
 505	unsigned long pages;
 506
 507	pages = end - start;
 508	zone  = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
 509	if (!zone)
 510		return NULL;
 511
 512	INIT_LIST_HEAD(&zone->nodes);
 513	INIT_LIST_HEAD(&zone->leaves);
 514	zone->start_pfn = start;
 515	zone->end_pfn = end;
 516	nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
 517
 518	for (i = 0; i < nr_blocks; i++) {
 519		if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
 520			free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
 521			return NULL;
 522		}
 523	}
 524
 525	return zone;
 526}
 527
 528/**
 529 * free_zone_bm_rtree - Free the memory of the radix tree.
 530 *
 531 * Free all node pages of the radix tree. The mem_zone_bm_rtree
 532 * structure itself is not freed here nor are the rtree_node
 533 * structs.
 534 */
 535static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
 536			       int clear_nosave_free)
 537{
 538	struct rtree_node *node;
 539
 540	list_for_each_entry(node, &zone->nodes, list)
 541		free_image_page(node->data, clear_nosave_free);
 542
 543	list_for_each_entry(node, &zone->leaves, list)
 544		free_image_page(node->data, clear_nosave_free);
 545}
 546
 547static void memory_bm_position_reset(struct memory_bitmap *bm)
 548{
 549	bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
 550				  list);
 551	bm->cur.node = list_entry(bm->cur.zone->leaves.next,
 552				  struct rtree_node, list);
 553	bm->cur.node_pfn = 0;
 554	bm->cur.node_bit = 0;
 555}
 556
 557static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
 558
 559struct mem_extent {
 560	struct list_head hook;
 561	unsigned long start;
 562	unsigned long end;
 563};
 564
 565/**
 566 * free_mem_extents - Free a list of memory extents.
 567 * @list: List of extents to free.
 568 */
 569static void free_mem_extents(struct list_head *list)
 570{
 571	struct mem_extent *ext, *aux;
 572
 573	list_for_each_entry_safe(ext, aux, list, hook) {
 574		list_del(&ext->hook);
 575		kfree(ext);
 576	}
 577}
 578
 579/**
 580 * create_mem_extents - Create a list of memory extents.
 581 * @list: List to put the extents into.
 582 * @gfp_mask: Mask to use for memory allocations.
 583 *
 584 * The extents represent contiguous ranges of PFNs.
 585 */
 586static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
 587{
 588	struct zone *zone;
 589
 590	INIT_LIST_HEAD(list);
 591
 592	for_each_populated_zone(zone) {
 593		unsigned long zone_start, zone_end;
 594		struct mem_extent *ext, *cur, *aux;
 595
 596		zone_start = zone->zone_start_pfn;
 597		zone_end = zone_end_pfn(zone);
 598
 599		list_for_each_entry(ext, list, hook)
 600			if (zone_start <= ext->end)
 601				break;
 602
 603		if (&ext->hook == list || zone_end < ext->start) {
 604			/* New extent is necessary */
 605			struct mem_extent *new_ext;
 606
 607			new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
 608			if (!new_ext) {
 609				free_mem_extents(list);
 610				return -ENOMEM;
 611			}
 612			new_ext->start = zone_start;
 613			new_ext->end = zone_end;
 614			list_add_tail(&new_ext->hook, &ext->hook);
 615			continue;
 616		}
 617
 618		/* Merge this zone's range of PFNs with the existing one */
 619		if (zone_start < ext->start)
 620			ext->start = zone_start;
 621		if (zone_end > ext->end)
 622			ext->end = zone_end;
 623
 624		/* More merging may be possible */
 625		cur = ext;
 626		list_for_each_entry_safe_continue(cur, aux, list, hook) {
 627			if (zone_end < cur->start)
 628				break;
 629			if (zone_end < cur->end)
 630				ext->end = cur->end;
 631			list_del(&cur->hook);
 632			kfree(cur);
 633		}
 634	}
 635
 636	return 0;
 637}
 638
 639/**
 640 * memory_bm_create - Allocate memory for a memory bitmap.
 641 */
 642static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
 643			    int safe_needed)
 644{
 645	struct chain_allocator ca;
 646	struct list_head mem_extents;
 647	struct mem_extent *ext;
 648	int error;
 649
 650	chain_init(&ca, gfp_mask, safe_needed);
 651	INIT_LIST_HEAD(&bm->zones);
 652
 653	error = create_mem_extents(&mem_extents, gfp_mask);
 654	if (error)
 655		return error;
 656
 657	list_for_each_entry(ext, &mem_extents, hook) {
 658		struct mem_zone_bm_rtree *zone;
 659
 660		zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
 661					    ext->start, ext->end);
 662		if (!zone) {
 663			error = -ENOMEM;
 664			goto Error;
 665		}
 666		list_add_tail(&zone->list, &bm->zones);
 667	}
 668
 669	bm->p_list = ca.chain;
 670	memory_bm_position_reset(bm);
 671 Exit:
 672	free_mem_extents(&mem_extents);
 673	return error;
 674
 675 Error:
 676	bm->p_list = ca.chain;
 677	memory_bm_free(bm, PG_UNSAFE_CLEAR);
 678	goto Exit;
 679}
 680
 681/**
 682 * memory_bm_free - Free memory occupied by the memory bitmap.
 683 * @bm: Memory bitmap.
 684 */
 685static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
 686{
 687	struct mem_zone_bm_rtree *zone;
 688
 689	list_for_each_entry(zone, &bm->zones, list)
 690		free_zone_bm_rtree(zone, clear_nosave_free);
 691
 692	free_list_of_pages(bm->p_list, clear_nosave_free);
 693
 694	INIT_LIST_HEAD(&bm->zones);
 695}
 696
 697/**
 698 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
 699 *
 700 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
 701 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
 702 *
 703 * Walk the radix tree to find the page containing the bit that represents @pfn
 704 * and return the position of the bit in @addr and @bit_nr.
 705 */
 706static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
 707			      void **addr, unsigned int *bit_nr)
 708{
 709	struct mem_zone_bm_rtree *curr, *zone;
 710	struct rtree_node *node;
 711	int i, block_nr;
 712
 713	zone = bm->cur.zone;
 714
 715	if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
 716		goto zone_found;
 717
 718	zone = NULL;
 719
 720	/* Find the right zone */
 721	list_for_each_entry(curr, &bm->zones, list) {
 722		if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
 723			zone = curr;
 724			break;
 725		}
 726	}
 727
 728	if (!zone)
 729		return -EFAULT;
 730
 731zone_found:
 732	/*
 733	 * We have found the zone. Now walk the radix tree to find the leaf node
 734	 * for our PFN.
 735	 */
 736
 737	/*
 738	 * If the zone we wish to scan is the the current zone and the
 739	 * pfn falls into the current node then we do not need to walk
 740	 * the tree.
 741	 */
 742	node = bm->cur.node;
 743	if (zone == bm->cur.zone &&
 744	    ((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
 745		goto node_found;
 746
 747	node      = zone->rtree;
 748	block_nr  = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
 749
 750	for (i = zone->levels; i > 0; i--) {
 751		int index;
 752
 753		index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
 754		index &= BM_RTREE_LEVEL_MASK;
 755		BUG_ON(node->data[index] == 0);
 756		node = (struct rtree_node *)node->data[index];
 757	}
 758
 759node_found:
 760	/* Update last position */
 761	bm->cur.zone = zone;
 762	bm->cur.node = node;
 763	bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
 764
 765	/* Set return values */
 766	*addr = node->data;
 767	*bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
 768
 769	return 0;
 770}
 771
 772static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
 773{
 774	void *addr;
 775	unsigned int bit;
 776	int error;
 777
 778	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 779	BUG_ON(error);
 780	set_bit(bit, addr);
 781}
 782
 783static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
 784{
 785	void *addr;
 786	unsigned int bit;
 787	int error;
 788
 789	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 790	if (!error)
 791		set_bit(bit, addr);
 792
 793	return error;
 794}
 795
 796static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
 797{
 798	void *addr;
 799	unsigned int bit;
 800	int error;
 801
 802	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 803	BUG_ON(error);
 804	clear_bit(bit, addr);
 805}
 806
 807static void memory_bm_clear_current(struct memory_bitmap *bm)
 808{
 809	int bit;
 810
 811	bit = max(bm->cur.node_bit - 1, 0);
 812	clear_bit(bit, bm->cur.node->data);
 813}
 814
 815static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
 816{
 817	void *addr;
 818	unsigned int bit;
 819	int error;
 820
 821	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 822	BUG_ON(error);
 823	return test_bit(bit, addr);
 824}
 825
 826static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
 827{
 828	void *addr;
 829	unsigned int bit;
 830
 831	return !memory_bm_find_bit(bm, pfn, &addr, &bit);
 832}
 833
 834/*
 835 * rtree_next_node - Jump to the next leaf node.
 836 *
 837 * Set the position to the beginning of the next node in the
 838 * memory bitmap. This is either the next node in the current
 839 * zone's radix tree or the first node in the radix tree of the
 840 * next zone.
 841 *
 842 * Return true if there is a next node, false otherwise.
 843 */
 844static bool rtree_next_node(struct memory_bitmap *bm)
 845{
 846	if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
 847		bm->cur.node = list_entry(bm->cur.node->list.next,
 848					  struct rtree_node, list);
 849		bm->cur.node_pfn += BM_BITS_PER_BLOCK;
 850		bm->cur.node_bit  = 0;
 851		touch_softlockup_watchdog();
 852		return true;
 853	}
 854
 855	/* No more nodes, goto next zone */
 856	if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
 857		bm->cur.zone = list_entry(bm->cur.zone->list.next,
 858				  struct mem_zone_bm_rtree, list);
 859		bm->cur.node = list_entry(bm->cur.zone->leaves.next,
 860					  struct rtree_node, list);
 861		bm->cur.node_pfn = 0;
 862		bm->cur.node_bit = 0;
 863		return true;
 864	}
 865
 866	/* No more zones */
 867	return false;
 868}
 869
 870/**
 871 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
 872 * @bm: Memory bitmap.
 873 *
 874 * Starting from the last returned position this function searches for the next
 875 * set bit in @bm and returns the PFN represented by it.  If no more bits are
 876 * set, BM_END_OF_MAP is returned.
 877 *
 878 * It is required to run memory_bm_position_reset() before the first call to
 879 * this function for the given memory bitmap.
 880 */
 881static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
 882{
 883	unsigned long bits, pfn, pages;
 884	int bit;
 885
 886	do {
 887		pages	  = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
 888		bits      = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
 889		bit	  = find_next_bit(bm->cur.node->data, bits,
 890					  bm->cur.node_bit);
 891		if (bit < bits) {
 892			pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
 893			bm->cur.node_bit = bit + 1;
 894			return pfn;
 895		}
 896	} while (rtree_next_node(bm));
 897
 898	return BM_END_OF_MAP;
 899}
 900
 901/*
 902 * This structure represents a range of page frames the contents of which
 903 * should not be saved during hibernation.
 904 */
 905struct nosave_region {
 906	struct list_head list;
 907	unsigned long start_pfn;
 908	unsigned long end_pfn;
 909};
 910
 911static LIST_HEAD(nosave_regions);
 912
 913static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
 914{
 915	struct rtree_node *node;
 916
 917	list_for_each_entry(node, &zone->nodes, list)
 918		recycle_safe_page(node->data);
 919
 920	list_for_each_entry(node, &zone->leaves, list)
 921		recycle_safe_page(node->data);
 922}
 923
 924static void memory_bm_recycle(struct memory_bitmap *bm)
 925{
 926	struct mem_zone_bm_rtree *zone;
 927	struct linked_page *p_list;
 928
 929	list_for_each_entry(zone, &bm->zones, list)
 930		recycle_zone_bm_rtree(zone);
 931
 932	p_list = bm->p_list;
 933	while (p_list) {
 934		struct linked_page *lp = p_list;
 935
 936		p_list = lp->next;
 937		recycle_safe_page(lp);
 938	}
 939}
 940
 941/**
 942 * register_nosave_region - Register a region of unsaveable memory.
 943 *
 944 * Register a range of page frames the contents of which should not be saved
 945 * during hibernation (to be used in the early initialization code).
 946 */
 947void __init __register_nosave_region(unsigned long start_pfn,
 948				     unsigned long end_pfn, int use_kmalloc)
 949{
 950	struct nosave_region *region;
 951
 952	if (start_pfn >= end_pfn)
 953		return;
 954
 955	if (!list_empty(&nosave_regions)) {
 956		/* Try to extend the previous region (they should be sorted) */
 957		region = list_entry(nosave_regions.prev,
 958					struct nosave_region, list);
 959		if (region->end_pfn == start_pfn) {
 960			region->end_pfn = end_pfn;
 961			goto Report;
 962		}
 963	}
 964	if (use_kmalloc) {
 965		/* During init, this shouldn't fail */
 966		region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
 967		BUG_ON(!region);
 968	} else {
 969		/* This allocation cannot fail */
 970		region = memblock_alloc(sizeof(struct nosave_region),
 971					SMP_CACHE_BYTES);
 972		if (!region)
 973			panic("%s: Failed to allocate %zu bytes\n", __func__,
 974			      sizeof(struct nosave_region));
 975	}
 976	region->start_pfn = start_pfn;
 977	region->end_pfn = end_pfn;
 978	list_add_tail(&region->list, &nosave_regions);
 979 Report:
 980	pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
 981		(unsigned long long) start_pfn << PAGE_SHIFT,
 982		((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
 983}
 984
 985/*
 986 * Set bits in this map correspond to the page frames the contents of which
 987 * should not be saved during the suspend.
 988 */
 989static struct memory_bitmap *forbidden_pages_map;
 990
 991/* Set bits in this map correspond to free page frames. */
 992static struct memory_bitmap *free_pages_map;
 993
 994/*
 995 * Each page frame allocated for creating the image is marked by setting the
 996 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
 997 */
 998
 999void swsusp_set_page_free(struct page *page)
1000{
1001	if (free_pages_map)
1002		memory_bm_set_bit(free_pages_map, page_to_pfn(page));
1003}
1004
1005static int swsusp_page_is_free(struct page *page)
1006{
1007	return free_pages_map ?
1008		memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1009}
1010
1011void swsusp_unset_page_free(struct page *page)
1012{
1013	if (free_pages_map)
1014		memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1015}
1016
1017static void swsusp_set_page_forbidden(struct page *page)
1018{
1019	if (forbidden_pages_map)
1020		memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1021}
1022
1023int swsusp_page_is_forbidden(struct page *page)
1024{
1025	return forbidden_pages_map ?
1026		memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1027}
1028
1029static void swsusp_unset_page_forbidden(struct page *page)
1030{
1031	if (forbidden_pages_map)
1032		memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1033}
1034
1035/**
1036 * mark_nosave_pages - Mark pages that should not be saved.
1037 * @bm: Memory bitmap.
1038 *
1039 * Set the bits in @bm that correspond to the page frames the contents of which
1040 * should not be saved.
1041 */
1042static void mark_nosave_pages(struct memory_bitmap *bm)
1043{
1044	struct nosave_region *region;
1045
1046	if (list_empty(&nosave_regions))
1047		return;
1048
1049	list_for_each_entry(region, &nosave_regions, list) {
1050		unsigned long pfn;
1051
1052		pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1053			 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1054			 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1055				- 1);
1056
1057		for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1058			if (pfn_valid(pfn)) {
1059				/*
1060				 * It is safe to ignore the result of
1061				 * mem_bm_set_bit_check() here, since we won't
1062				 * touch the PFNs for which the error is
1063				 * returned anyway.
1064				 */
1065				mem_bm_set_bit_check(bm, pfn);
1066			}
1067	}
1068}
1069
1070/**
1071 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1072 *
1073 * Create bitmaps needed for marking page frames that should not be saved and
1074 * free page frames.  The forbidden_pages_map and free_pages_map pointers are
1075 * only modified if everything goes well, because we don't want the bits to be
1076 * touched before both bitmaps are set up.
1077 */
1078int create_basic_memory_bitmaps(void)
1079{
1080	struct memory_bitmap *bm1, *bm2;
1081	int error = 0;
1082
1083	if (forbidden_pages_map && free_pages_map)
1084		return 0;
1085	else
1086		BUG_ON(forbidden_pages_map || free_pages_map);
1087
1088	bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1089	if (!bm1)
1090		return -ENOMEM;
1091
1092	error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1093	if (error)
1094		goto Free_first_object;
1095
1096	bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1097	if (!bm2)
1098		goto Free_first_bitmap;
1099
1100	error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1101	if (error)
1102		goto Free_second_object;
1103
1104	forbidden_pages_map = bm1;
1105	free_pages_map = bm2;
1106	mark_nosave_pages(forbidden_pages_map);
1107
1108	pr_debug("Basic memory bitmaps created\n");
1109
1110	return 0;
1111
1112 Free_second_object:
1113	kfree(bm2);
1114 Free_first_bitmap:
1115 	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1116 Free_first_object:
1117	kfree(bm1);
1118	return -ENOMEM;
1119}
1120
1121/**
1122 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1123 *
1124 * Free memory bitmaps allocated by create_basic_memory_bitmaps().  The
1125 * auxiliary pointers are necessary so that the bitmaps themselves are not
1126 * referred to while they are being freed.
1127 */
1128void free_basic_memory_bitmaps(void)
1129{
1130	struct memory_bitmap *bm1, *bm2;
1131
1132	if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1133		return;
1134
1135	bm1 = forbidden_pages_map;
1136	bm2 = free_pages_map;
1137	forbidden_pages_map = NULL;
1138	free_pages_map = NULL;
1139	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1140	kfree(bm1);
1141	memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1142	kfree(bm2);
1143
1144	pr_debug("Basic memory bitmaps freed\n");
1145}
1146
1147void clear_free_pages(void)
1148{
 
1149	struct memory_bitmap *bm = free_pages_map;
1150	unsigned long pfn;
1151
1152	if (WARN_ON(!(free_pages_map)))
1153		return;
1154
1155	if (IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) || want_init_on_free()) {
1156		memory_bm_position_reset(bm);
1157		pfn = memory_bm_next_pfn(bm);
1158		while (pfn != BM_END_OF_MAP) {
1159			if (pfn_valid(pfn))
1160				clear_highpage(pfn_to_page(pfn));
1161
1162			pfn = memory_bm_next_pfn(bm);
1163		}
1164		memory_bm_position_reset(bm);
1165		pr_info("free pages cleared after restore\n");
1166	}
 
 
 
1167}
1168
1169/**
1170 * snapshot_additional_pages - Estimate the number of extra pages needed.
1171 * @zone: Memory zone to carry out the computation for.
1172 *
1173 * Estimate the number of additional pages needed for setting up a hibernation
1174 * image data structures for @zone (usually, the returned value is greater than
1175 * the exact number).
1176 */
1177unsigned int snapshot_additional_pages(struct zone *zone)
1178{
1179	unsigned int rtree, nodes;
1180
1181	rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1182	rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1183			      LINKED_PAGE_DATA_SIZE);
1184	while (nodes > 1) {
1185		nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1186		rtree += nodes;
1187	}
1188
1189	return 2 * rtree;
1190}
1191
1192#ifdef CONFIG_HIGHMEM
1193/**
1194 * count_free_highmem_pages - Compute the total number of free highmem pages.
1195 *
1196 * The returned number is system-wide.
1197 */
1198static unsigned int count_free_highmem_pages(void)
1199{
1200	struct zone *zone;
1201	unsigned int cnt = 0;
1202
1203	for_each_populated_zone(zone)
1204		if (is_highmem(zone))
1205			cnt += zone_page_state(zone, NR_FREE_PAGES);
1206
1207	return cnt;
1208}
1209
1210/**
1211 * saveable_highmem_page - Check if a highmem page is saveable.
1212 *
1213 * Determine whether a highmem page should be included in a hibernation image.
1214 *
1215 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1216 * and it isn't part of a free chunk of pages.
1217 */
1218static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1219{
1220	struct page *page;
1221
1222	if (!pfn_valid(pfn))
1223		return NULL;
1224
1225	page = pfn_to_online_page(pfn);
1226	if (!page || page_zone(page) != zone)
1227		return NULL;
1228
1229	BUG_ON(!PageHighMem(page));
1230
1231	if (swsusp_page_is_forbidden(page) ||  swsusp_page_is_free(page))
1232		return NULL;
1233
1234	if (PageReserved(page) || PageOffline(page))
1235		return NULL;
1236
1237	if (page_is_guard(page))
1238		return NULL;
1239
1240	return page;
1241}
1242
1243/**
1244 * count_highmem_pages - Compute the total number of saveable highmem pages.
1245 */
1246static unsigned int count_highmem_pages(void)
1247{
1248	struct zone *zone;
1249	unsigned int n = 0;
1250
1251	for_each_populated_zone(zone) {
1252		unsigned long pfn, max_zone_pfn;
1253
1254		if (!is_highmem(zone))
1255			continue;
1256
1257		mark_free_pages(zone);
1258		max_zone_pfn = zone_end_pfn(zone);
1259		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1260			if (saveable_highmem_page(zone, pfn))
1261				n++;
1262	}
1263	return n;
1264}
1265#else
1266static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1267{
1268	return NULL;
1269}
1270#endif /* CONFIG_HIGHMEM */
1271
1272/**
1273 * saveable_page - Check if the given page is saveable.
1274 *
1275 * Determine whether a non-highmem page should be included in a hibernation
1276 * image.
1277 *
1278 * We should save the page if it isn't Nosave, and is not in the range
1279 * of pages statically defined as 'unsaveable', and it isn't part of
1280 * a free chunk of pages.
1281 */
1282static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1283{
1284	struct page *page;
1285
1286	if (!pfn_valid(pfn))
1287		return NULL;
1288
1289	page = pfn_to_online_page(pfn);
1290	if (!page || page_zone(page) != zone)
1291		return NULL;
1292
1293	BUG_ON(PageHighMem(page));
1294
1295	if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1296		return NULL;
1297
1298	if (PageOffline(page))
1299		return NULL;
1300
1301	if (PageReserved(page)
1302	    && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1303		return NULL;
1304
1305	if (page_is_guard(page))
1306		return NULL;
1307
1308	return page;
1309}
1310
1311/**
1312 * count_data_pages - Compute the total number of saveable non-highmem pages.
1313 */
1314static unsigned int count_data_pages(void)
1315{
1316	struct zone *zone;
1317	unsigned long pfn, max_zone_pfn;
1318	unsigned int n = 0;
1319
1320	for_each_populated_zone(zone) {
1321		if (is_highmem(zone))
1322			continue;
1323
1324		mark_free_pages(zone);
1325		max_zone_pfn = zone_end_pfn(zone);
1326		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1327			if (saveable_page(zone, pfn))
1328				n++;
1329	}
1330	return n;
1331}
1332
1333/*
1334 * This is needed, because copy_page and memcpy are not usable for copying
1335 * task structs.
1336 */
1337static inline void do_copy_page(long *dst, long *src)
1338{
1339	int n;
1340
1341	for (n = PAGE_SIZE / sizeof(long); n; n--)
1342		*dst++ = *src++;
1343}
1344
1345/**
1346 * safe_copy_page - Copy a page in a safe way.
1347 *
1348 * Check if the page we are going to copy is marked as present in the kernel
1349 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1350 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1351 * always returns 'true'.
1352 */
1353static void safe_copy_page(void *dst, struct page *s_page)
1354{
1355	if (kernel_page_present(s_page)) {
1356		do_copy_page(dst, page_address(s_page));
1357	} else {
1358		kernel_map_pages(s_page, 1, 1);
1359		do_copy_page(dst, page_address(s_page));
1360		kernel_map_pages(s_page, 1, 0);
1361	}
1362}
1363
1364#ifdef CONFIG_HIGHMEM
1365static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1366{
1367	return is_highmem(zone) ?
1368		saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1369}
1370
1371static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1372{
1373	struct page *s_page, *d_page;
1374	void *src, *dst;
1375
1376	s_page = pfn_to_page(src_pfn);
1377	d_page = pfn_to_page(dst_pfn);
1378	if (PageHighMem(s_page)) {
1379		src = kmap_atomic(s_page);
1380		dst = kmap_atomic(d_page);
1381		do_copy_page(dst, src);
1382		kunmap_atomic(dst);
1383		kunmap_atomic(src);
1384	} else {
1385		if (PageHighMem(d_page)) {
1386			/*
1387			 * The page pointed to by src may contain some kernel
1388			 * data modified by kmap_atomic()
1389			 */
1390			safe_copy_page(buffer, s_page);
1391			dst = kmap_atomic(d_page);
1392			copy_page(dst, buffer);
1393			kunmap_atomic(dst);
1394		} else {
1395			safe_copy_page(page_address(d_page), s_page);
1396		}
1397	}
1398}
1399#else
1400#define page_is_saveable(zone, pfn)	saveable_page(zone, pfn)
1401
1402static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1403{
1404	safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1405				pfn_to_page(src_pfn));
1406}
1407#endif /* CONFIG_HIGHMEM */
1408
1409static void copy_data_pages(struct memory_bitmap *copy_bm,
1410			    struct memory_bitmap *orig_bm)
1411{
1412	struct zone *zone;
1413	unsigned long pfn;
1414
1415	for_each_populated_zone(zone) {
1416		unsigned long max_zone_pfn;
1417
1418		mark_free_pages(zone);
1419		max_zone_pfn = zone_end_pfn(zone);
1420		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1421			if (page_is_saveable(zone, pfn))
1422				memory_bm_set_bit(orig_bm, pfn);
1423	}
1424	memory_bm_position_reset(orig_bm);
1425	memory_bm_position_reset(copy_bm);
1426	for(;;) {
1427		pfn = memory_bm_next_pfn(orig_bm);
1428		if (unlikely(pfn == BM_END_OF_MAP))
1429			break;
1430		copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1431	}
1432}
1433
1434/* Total number of image pages */
1435static unsigned int nr_copy_pages;
1436/* Number of pages needed for saving the original pfns of the image pages */
1437static unsigned int nr_meta_pages;
1438/*
1439 * Numbers of normal and highmem page frames allocated for hibernation image
1440 * before suspending devices.
1441 */
1442static unsigned int alloc_normal, alloc_highmem;
1443/*
1444 * Memory bitmap used for marking saveable pages (during hibernation) or
1445 * hibernation image pages (during restore)
1446 */
1447static struct memory_bitmap orig_bm;
1448/*
1449 * Memory bitmap used during hibernation for marking allocated page frames that
1450 * will contain copies of saveable pages.  During restore it is initially used
1451 * for marking hibernation image pages, but then the set bits from it are
1452 * duplicated in @orig_bm and it is released.  On highmem systems it is next
1453 * used for marking "safe" highmem pages, but it has to be reinitialized for
1454 * this purpose.
1455 */
1456static struct memory_bitmap copy_bm;
1457
1458/**
1459 * swsusp_free - Free pages allocated for hibernation image.
1460 *
1461 * Image pages are alocated before snapshot creation, so they need to be
1462 * released after resume.
1463 */
1464void swsusp_free(void)
1465{
1466	unsigned long fb_pfn, fr_pfn;
1467
1468	if (!forbidden_pages_map || !free_pages_map)
1469		goto out;
1470
1471	memory_bm_position_reset(forbidden_pages_map);
1472	memory_bm_position_reset(free_pages_map);
1473
1474loop:
1475	fr_pfn = memory_bm_next_pfn(free_pages_map);
1476	fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1477
1478	/*
1479	 * Find the next bit set in both bitmaps. This is guaranteed to
1480	 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1481	 */
1482	do {
1483		if (fb_pfn < fr_pfn)
1484			fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1485		if (fr_pfn < fb_pfn)
1486			fr_pfn = memory_bm_next_pfn(free_pages_map);
1487	} while (fb_pfn != fr_pfn);
1488
1489	if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1490		struct page *page = pfn_to_page(fr_pfn);
1491
1492		memory_bm_clear_current(forbidden_pages_map);
1493		memory_bm_clear_current(free_pages_map);
1494		hibernate_restore_unprotect_page(page_address(page));
1495		__free_page(page);
1496		goto loop;
1497	}
1498
1499out:
1500	nr_copy_pages = 0;
1501	nr_meta_pages = 0;
1502	restore_pblist = NULL;
1503	buffer = NULL;
1504	alloc_normal = 0;
1505	alloc_highmem = 0;
1506	hibernate_restore_protection_end();
1507}
1508
1509/* Helper functions used for the shrinking of memory. */
1510
1511#define GFP_IMAGE	(GFP_KERNEL | __GFP_NOWARN)
1512
1513/**
1514 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1515 * @nr_pages: Number of page frames to allocate.
1516 * @mask: GFP flags to use for the allocation.
1517 *
1518 * Return value: Number of page frames actually allocated
1519 */
1520static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1521{
1522	unsigned long nr_alloc = 0;
1523
1524	while (nr_pages > 0) {
1525		struct page *page;
1526
1527		page = alloc_image_page(mask);
1528		if (!page)
1529			break;
1530		memory_bm_set_bit(&copy_bm, page_to_pfn(page));
1531		if (PageHighMem(page))
1532			alloc_highmem++;
1533		else
1534			alloc_normal++;
1535		nr_pages--;
1536		nr_alloc++;
1537	}
1538
1539	return nr_alloc;
1540}
1541
1542static unsigned long preallocate_image_memory(unsigned long nr_pages,
1543					      unsigned long avail_normal)
1544{
1545	unsigned long alloc;
1546
1547	if (avail_normal <= alloc_normal)
1548		return 0;
1549
1550	alloc = avail_normal - alloc_normal;
1551	if (nr_pages < alloc)
1552		alloc = nr_pages;
1553
1554	return preallocate_image_pages(alloc, GFP_IMAGE);
1555}
1556
1557#ifdef CONFIG_HIGHMEM
1558static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1559{
1560	return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1561}
1562
1563/**
1564 *  __fraction - Compute (an approximation of) x * (multiplier / base).
1565 */
1566static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1567{
1568	return div64_u64(x * multiplier, base);
 
 
1569}
1570
1571static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1572						  unsigned long highmem,
1573						  unsigned long total)
1574{
1575	unsigned long alloc = __fraction(nr_pages, highmem, total);
1576
1577	return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1578}
1579#else /* CONFIG_HIGHMEM */
1580static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1581{
1582	return 0;
1583}
1584
1585static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1586							 unsigned long highmem,
1587							 unsigned long total)
1588{
1589	return 0;
1590}
1591#endif /* CONFIG_HIGHMEM */
1592
1593/**
1594 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1595 */
1596static unsigned long free_unnecessary_pages(void)
1597{
1598	unsigned long save, to_free_normal, to_free_highmem, free;
1599
1600	save = count_data_pages();
1601	if (alloc_normal >= save) {
1602		to_free_normal = alloc_normal - save;
1603		save = 0;
1604	} else {
1605		to_free_normal = 0;
1606		save -= alloc_normal;
1607	}
1608	save += count_highmem_pages();
1609	if (alloc_highmem >= save) {
1610		to_free_highmem = alloc_highmem - save;
1611	} else {
1612		to_free_highmem = 0;
1613		save -= alloc_highmem;
1614		if (to_free_normal > save)
1615			to_free_normal -= save;
1616		else
1617			to_free_normal = 0;
1618	}
1619	free = to_free_normal + to_free_highmem;
1620
1621	memory_bm_position_reset(&copy_bm);
1622
1623	while (to_free_normal > 0 || to_free_highmem > 0) {
1624		unsigned long pfn = memory_bm_next_pfn(&copy_bm);
1625		struct page *page = pfn_to_page(pfn);
1626
1627		if (PageHighMem(page)) {
1628			if (!to_free_highmem)
1629				continue;
1630			to_free_highmem--;
1631			alloc_highmem--;
1632		} else {
1633			if (!to_free_normal)
1634				continue;
1635			to_free_normal--;
1636			alloc_normal--;
1637		}
1638		memory_bm_clear_bit(&copy_bm, pfn);
1639		swsusp_unset_page_forbidden(page);
1640		swsusp_unset_page_free(page);
1641		__free_page(page);
1642	}
1643
1644	return free;
1645}
1646
1647/**
1648 * minimum_image_size - Estimate the minimum acceptable size of an image.
1649 * @saveable: Number of saveable pages in the system.
1650 *
1651 * We want to avoid attempting to free too much memory too hard, so estimate the
1652 * minimum acceptable size of a hibernation image to use as the lower limit for
1653 * preallocating memory.
1654 *
1655 * We assume that the minimum image size should be proportional to
1656 *
1657 * [number of saveable pages] - [number of pages that can be freed in theory]
1658 *
1659 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1660 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1661 */
1662static unsigned long minimum_image_size(unsigned long saveable)
1663{
1664	unsigned long size;
1665
1666	size = global_node_page_state_pages(NR_SLAB_RECLAIMABLE_B)
1667		+ global_node_page_state(NR_ACTIVE_ANON)
1668		+ global_node_page_state(NR_INACTIVE_ANON)
1669		+ global_node_page_state(NR_ACTIVE_FILE)
1670		+ global_node_page_state(NR_INACTIVE_FILE);
1671
1672	return saveable <= size ? 0 : saveable - size;
1673}
1674
1675/**
1676 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1677 *
1678 * To create a hibernation image it is necessary to make a copy of every page
1679 * frame in use.  We also need a number of page frames to be free during
1680 * hibernation for allocations made while saving the image and for device
1681 * drivers, in case they need to allocate memory from their hibernation
1682 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1683 * estimate) and reserved_size divided by PAGE_SIZE (which is tunable through
1684 * /sys/power/reserved_size, respectively).  To make this happen, we compute the
1685 * total number of available page frames and allocate at least
1686 *
1687 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1688 *  + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1689 *
1690 * of them, which corresponds to the maximum size of a hibernation image.
1691 *
1692 * If image_size is set below the number following from the above formula,
1693 * the preallocation of memory is continued until the total number of saveable
1694 * pages in the system is below the requested image size or the minimum
1695 * acceptable image size returned by minimum_image_size(), whichever is greater.
1696 */
1697int hibernate_preallocate_memory(void)
1698{
1699	struct zone *zone;
1700	unsigned long saveable, size, max_size, count, highmem, pages = 0;
1701	unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1702	ktime_t start, stop;
1703	int error;
1704
1705	pr_info("Preallocating image memory\n");
1706	start = ktime_get();
1707
1708	error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1709	if (error) {
1710		pr_err("Cannot allocate original bitmap\n");
1711		goto err_out;
1712	}
1713
1714	error = memory_bm_create(&copy_bm, GFP_IMAGE, PG_ANY);
1715	if (error) {
1716		pr_err("Cannot allocate copy bitmap\n");
1717		goto err_out;
1718	}
1719
1720	alloc_normal = 0;
1721	alloc_highmem = 0;
1722
1723	/* Count the number of saveable data pages. */
1724	save_highmem = count_highmem_pages();
1725	saveable = count_data_pages();
1726
1727	/*
1728	 * Compute the total number of page frames we can use (count) and the
1729	 * number of pages needed for image metadata (size).
1730	 */
1731	count = saveable;
1732	saveable += save_highmem;
1733	highmem = save_highmem;
1734	size = 0;
1735	for_each_populated_zone(zone) {
1736		size += snapshot_additional_pages(zone);
1737		if (is_highmem(zone))
1738			highmem += zone_page_state(zone, NR_FREE_PAGES);
1739		else
1740			count += zone_page_state(zone, NR_FREE_PAGES);
1741	}
1742	avail_normal = count;
1743	count += highmem;
1744	count -= totalreserve_pages;
1745
 
 
 
1746	/* Compute the maximum number of saveable pages to leave in memory. */
1747	max_size = (count - (size + PAGES_FOR_IO)) / 2
1748			- 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1749	/* Compute the desired number of image pages specified by image_size. */
1750	size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1751	if (size > max_size)
1752		size = max_size;
1753	/*
1754	 * If the desired number of image pages is at least as large as the
1755	 * current number of saveable pages in memory, allocate page frames for
1756	 * the image and we're done.
1757	 */
1758	if (size >= saveable) {
1759		pages = preallocate_image_highmem(save_highmem);
1760		pages += preallocate_image_memory(saveable - pages, avail_normal);
1761		goto out;
1762	}
1763
1764	/* Estimate the minimum size of the image. */
1765	pages = minimum_image_size(saveable);
1766	/*
1767	 * To avoid excessive pressure on the normal zone, leave room in it to
1768	 * accommodate an image of the minimum size (unless it's already too
1769	 * small, in which case don't preallocate pages from it at all).
1770	 */
1771	if (avail_normal > pages)
1772		avail_normal -= pages;
1773	else
1774		avail_normal = 0;
1775	if (size < pages)
1776		size = min_t(unsigned long, pages, max_size);
1777
1778	/*
1779	 * Let the memory management subsystem know that we're going to need a
1780	 * large number of page frames to allocate and make it free some memory.
1781	 * NOTE: If this is not done, performance will be hurt badly in some
1782	 * test cases.
1783	 */
1784	shrink_all_memory(saveable - size);
1785
1786	/*
1787	 * The number of saveable pages in memory was too high, so apply some
1788	 * pressure to decrease it.  First, make room for the largest possible
1789	 * image and fail if that doesn't work.  Next, try to decrease the size
1790	 * of the image as much as indicated by 'size' using allocations from
1791	 * highmem and non-highmem zones separately.
1792	 */
1793	pages_highmem = preallocate_image_highmem(highmem / 2);
1794	alloc = count - max_size;
1795	if (alloc > pages_highmem)
1796		alloc -= pages_highmem;
1797	else
1798		alloc = 0;
1799	pages = preallocate_image_memory(alloc, avail_normal);
1800	if (pages < alloc) {
1801		/* We have exhausted non-highmem pages, try highmem. */
1802		alloc -= pages;
1803		pages += pages_highmem;
1804		pages_highmem = preallocate_image_highmem(alloc);
1805		if (pages_highmem < alloc) {
1806			pr_err("Image allocation is %lu pages short\n",
1807				alloc - pages_highmem);
1808			goto err_out;
1809		}
1810		pages += pages_highmem;
1811		/*
1812		 * size is the desired number of saveable pages to leave in
1813		 * memory, so try to preallocate (all memory - size) pages.
1814		 */
1815		alloc = (count - pages) - size;
1816		pages += preallocate_image_highmem(alloc);
1817	} else {
1818		/*
1819		 * There are approximately max_size saveable pages at this point
1820		 * and we want to reduce this number down to size.
1821		 */
1822		alloc = max_size - size;
1823		size = preallocate_highmem_fraction(alloc, highmem, count);
1824		pages_highmem += size;
1825		alloc -= size;
1826		size = preallocate_image_memory(alloc, avail_normal);
1827		pages_highmem += preallocate_image_highmem(alloc - size);
1828		pages += pages_highmem + size;
1829	}
1830
1831	/*
1832	 * We only need as many page frames for the image as there are saveable
1833	 * pages in memory, but we have allocated more.  Release the excessive
1834	 * ones now.
1835	 */
1836	pages -= free_unnecessary_pages();
1837
1838 out:
1839	stop = ktime_get();
1840	pr_info("Allocated %lu pages for snapshot\n", pages);
1841	swsusp_show_speed(start, stop, pages, "Allocated");
1842
1843	return 0;
1844
1845 err_out:
 
1846	swsusp_free();
1847	return -ENOMEM;
1848}
1849
1850#ifdef CONFIG_HIGHMEM
1851/**
1852 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1853 *
1854 * Compute the number of non-highmem pages that will be necessary for creating
1855 * copies of highmem pages.
1856 */
1857static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1858{
1859	unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1860
1861	if (free_highmem >= nr_highmem)
1862		nr_highmem = 0;
1863	else
1864		nr_highmem -= free_highmem;
1865
1866	return nr_highmem;
1867}
1868#else
1869static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1870#endif /* CONFIG_HIGHMEM */
1871
1872/**
1873 * enough_free_mem - Check if there is enough free memory for the image.
1874 */
1875static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1876{
1877	struct zone *zone;
1878	unsigned int free = alloc_normal;
1879
1880	for_each_populated_zone(zone)
1881		if (!is_highmem(zone))
1882			free += zone_page_state(zone, NR_FREE_PAGES);
1883
1884	nr_pages += count_pages_for_highmem(nr_highmem);
1885	pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
1886		 nr_pages, PAGES_FOR_IO, free);
1887
1888	return free > nr_pages + PAGES_FOR_IO;
1889}
1890
1891#ifdef CONFIG_HIGHMEM
1892/**
1893 * get_highmem_buffer - Allocate a buffer for highmem pages.
1894 *
1895 * If there are some highmem pages in the hibernation image, we may need a
1896 * buffer to copy them and/or load their data.
1897 */
1898static inline int get_highmem_buffer(int safe_needed)
1899{
1900	buffer = get_image_page(GFP_ATOMIC, safe_needed);
1901	return buffer ? 0 : -ENOMEM;
1902}
1903
1904/**
1905 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1906 *
1907 * Try to allocate as many pages as needed, but if the number of free highmem
1908 * pages is less than that, allocate them all.
1909 */
1910static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1911					       unsigned int nr_highmem)
1912{
1913	unsigned int to_alloc = count_free_highmem_pages();
1914
1915	if (to_alloc > nr_highmem)
1916		to_alloc = nr_highmem;
1917
1918	nr_highmem -= to_alloc;
1919	while (to_alloc-- > 0) {
1920		struct page *page;
1921
1922		page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1923		memory_bm_set_bit(bm, page_to_pfn(page));
1924	}
1925	return nr_highmem;
1926}
1927#else
1928static inline int get_highmem_buffer(int safe_needed) { return 0; }
1929
1930static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1931					       unsigned int n) { return 0; }
1932#endif /* CONFIG_HIGHMEM */
1933
1934/**
1935 * swsusp_alloc - Allocate memory for hibernation image.
1936 *
1937 * We first try to allocate as many highmem pages as there are
1938 * saveable highmem pages in the system.  If that fails, we allocate
1939 * non-highmem pages for the copies of the remaining highmem ones.
1940 *
1941 * In this approach it is likely that the copies of highmem pages will
1942 * also be located in the high memory, because of the way in which
1943 * copy_data_pages() works.
1944 */
1945static int swsusp_alloc(struct memory_bitmap *copy_bm,
1946			unsigned int nr_pages, unsigned int nr_highmem)
1947{
1948	if (nr_highmem > 0) {
1949		if (get_highmem_buffer(PG_ANY))
1950			goto err_out;
1951		if (nr_highmem > alloc_highmem) {
1952			nr_highmem -= alloc_highmem;
1953			nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1954		}
1955	}
1956	if (nr_pages > alloc_normal) {
1957		nr_pages -= alloc_normal;
1958		while (nr_pages-- > 0) {
1959			struct page *page;
1960
1961			page = alloc_image_page(GFP_ATOMIC);
1962			if (!page)
1963				goto err_out;
1964			memory_bm_set_bit(copy_bm, page_to_pfn(page));
1965		}
1966	}
1967
1968	return 0;
1969
1970 err_out:
1971	swsusp_free();
1972	return -ENOMEM;
1973}
1974
1975asmlinkage __visible int swsusp_save(void)
1976{
1977	unsigned int nr_pages, nr_highmem;
1978
1979	pr_info("Creating image:\n");
1980
1981	drain_local_pages(NULL);
1982	nr_pages = count_data_pages();
1983	nr_highmem = count_highmem_pages();
1984	pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
1985
1986	if (!enough_free_mem(nr_pages, nr_highmem)) {
1987		pr_err("Not enough free memory\n");
1988		return -ENOMEM;
1989	}
1990
1991	if (swsusp_alloc(&copy_bm, nr_pages, nr_highmem)) {
1992		pr_err("Memory allocation failed\n");
1993		return -ENOMEM;
1994	}
1995
1996	/*
1997	 * During allocating of suspend pagedir, new cold pages may appear.
1998	 * Kill them.
1999	 */
2000	drain_local_pages(NULL);
2001	copy_data_pages(&copy_bm, &orig_bm);
2002
2003	/*
2004	 * End of critical section. From now on, we can write to memory,
2005	 * but we should not touch disk. This specially means we must _not_
2006	 * touch swap space! Except we must write out our image of course.
2007	 */
2008
2009	nr_pages += nr_highmem;
2010	nr_copy_pages = nr_pages;
2011	nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2012
2013	pr_info("Image created (%d pages copied)\n", nr_pages);
2014
2015	return 0;
2016}
2017
2018#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2019static int init_header_complete(struct swsusp_info *info)
2020{
2021	memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2022	info->version_code = LINUX_VERSION_CODE;
2023	return 0;
2024}
2025
2026static const char *check_image_kernel(struct swsusp_info *info)
2027{
2028	if (info->version_code != LINUX_VERSION_CODE)
2029		return "kernel version";
2030	if (strcmp(info->uts.sysname,init_utsname()->sysname))
2031		return "system type";
2032	if (strcmp(info->uts.release,init_utsname()->release))
2033		return "kernel release";
2034	if (strcmp(info->uts.version,init_utsname()->version))
2035		return "version";
2036	if (strcmp(info->uts.machine,init_utsname()->machine))
2037		return "machine";
2038	return NULL;
2039}
2040#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2041
2042unsigned long snapshot_get_image_size(void)
2043{
2044	return nr_copy_pages + nr_meta_pages + 1;
2045}
2046
2047static int init_header(struct swsusp_info *info)
2048{
2049	memset(info, 0, sizeof(struct swsusp_info));
2050	info->num_physpages = get_num_physpages();
2051	info->image_pages = nr_copy_pages;
2052	info->pages = snapshot_get_image_size();
2053	info->size = info->pages;
2054	info->size <<= PAGE_SHIFT;
2055	return init_header_complete(info);
2056}
2057
2058/**
2059 * pack_pfns - Prepare PFNs for saving.
2060 * @bm: Memory bitmap.
2061 * @buf: Memory buffer to store the PFNs in.
2062 *
2063 * PFNs corresponding to set bits in @bm are stored in the area of memory
2064 * pointed to by @buf (1 page at a time).
2065 */
2066static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2067{
2068	int j;
2069
2070	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2071		buf[j] = memory_bm_next_pfn(bm);
2072		if (unlikely(buf[j] == BM_END_OF_MAP))
2073			break;
 
 
2074	}
2075}
2076
2077/**
2078 * snapshot_read_next - Get the address to read the next image page from.
2079 * @handle: Snapshot handle to be used for the reading.
2080 *
2081 * On the first call, @handle should point to a zeroed snapshot_handle
2082 * structure.  The structure gets populated then and a pointer to it should be
2083 * passed to this function every next time.
2084 *
2085 * On success, the function returns a positive number.  Then, the caller
2086 * is allowed to read up to the returned number of bytes from the memory
2087 * location computed by the data_of() macro.
2088 *
2089 * The function returns 0 to indicate the end of the data stream condition,
2090 * and negative numbers are returned on errors.  If that happens, the structure
2091 * pointed to by @handle is not updated and should not be used any more.
2092 */
2093int snapshot_read_next(struct snapshot_handle *handle)
2094{
2095	if (handle->cur > nr_meta_pages + nr_copy_pages)
2096		return 0;
2097
2098	if (!buffer) {
2099		/* This makes the buffer be freed by swsusp_free() */
2100		buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2101		if (!buffer)
2102			return -ENOMEM;
2103	}
2104	if (!handle->cur) {
2105		int error;
2106
2107		error = init_header((struct swsusp_info *)buffer);
2108		if (error)
2109			return error;
2110		handle->buffer = buffer;
2111		memory_bm_position_reset(&orig_bm);
2112		memory_bm_position_reset(&copy_bm);
2113	} else if (handle->cur <= nr_meta_pages) {
2114		clear_page(buffer);
2115		pack_pfns(buffer, &orig_bm);
2116	} else {
2117		struct page *page;
2118
2119		page = pfn_to_page(memory_bm_next_pfn(&copy_bm));
2120		if (PageHighMem(page)) {
2121			/*
2122			 * Highmem pages are copied to the buffer,
2123			 * because we can't return with a kmapped
2124			 * highmem page (we may not be called again).
2125			 */
2126			void *kaddr;
2127
2128			kaddr = kmap_atomic(page);
2129			copy_page(buffer, kaddr);
2130			kunmap_atomic(kaddr);
2131			handle->buffer = buffer;
2132		} else {
2133			handle->buffer = page_address(page);
2134		}
2135	}
2136	handle->cur++;
2137	return PAGE_SIZE;
2138}
2139
2140static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2141				    struct memory_bitmap *src)
2142{
2143	unsigned long pfn;
2144
2145	memory_bm_position_reset(src);
2146	pfn = memory_bm_next_pfn(src);
2147	while (pfn != BM_END_OF_MAP) {
2148		memory_bm_set_bit(dst, pfn);
2149		pfn = memory_bm_next_pfn(src);
2150	}
2151}
2152
2153/**
2154 * mark_unsafe_pages - Mark pages that were used before hibernation.
2155 *
2156 * Mark the pages that cannot be used for storing the image during restoration,
2157 * because they conflict with the pages that had been used before hibernation.
2158 */
2159static void mark_unsafe_pages(struct memory_bitmap *bm)
2160{
2161	unsigned long pfn;
2162
2163	/* Clear the "free"/"unsafe" bit for all PFNs */
2164	memory_bm_position_reset(free_pages_map);
2165	pfn = memory_bm_next_pfn(free_pages_map);
2166	while (pfn != BM_END_OF_MAP) {
2167		memory_bm_clear_current(free_pages_map);
2168		pfn = memory_bm_next_pfn(free_pages_map);
2169	}
2170
2171	/* Mark pages that correspond to the "original" PFNs as "unsafe" */
2172	duplicate_memory_bitmap(free_pages_map, bm);
2173
2174	allocated_unsafe_pages = 0;
2175}
2176
2177static int check_header(struct swsusp_info *info)
2178{
2179	const char *reason;
2180
2181	reason = check_image_kernel(info);
2182	if (!reason && info->num_physpages != get_num_physpages())
2183		reason = "memory size";
2184	if (reason) {
2185		pr_err("Image mismatch: %s\n", reason);
2186		return -EPERM;
2187	}
2188	return 0;
2189}
2190
2191/**
2192 * load header - Check the image header and copy the data from it.
2193 */
2194static int load_header(struct swsusp_info *info)
2195{
2196	int error;
2197
2198	restore_pblist = NULL;
2199	error = check_header(info);
2200	if (!error) {
2201		nr_copy_pages = info->image_pages;
2202		nr_meta_pages = info->pages - info->image_pages - 1;
2203	}
2204	return error;
2205}
2206
2207/**
2208 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2209 * @bm: Memory bitmap.
2210 * @buf: Area of memory containing the PFNs.
2211 *
2212 * For each element of the array pointed to by @buf (1 page at a time), set the
2213 * corresponding bit in @bm.
2214 */
2215static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2216{
2217	int j;
2218
2219	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2220		if (unlikely(buf[j] == BM_END_OF_MAP))
2221			break;
2222
 
 
 
2223		if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2224			memory_bm_set_bit(bm, buf[j]);
2225		else
2226			return -EFAULT;
2227	}
2228
2229	return 0;
2230}
2231
2232#ifdef CONFIG_HIGHMEM
2233/*
2234 * struct highmem_pbe is used for creating the list of highmem pages that
2235 * should be restored atomically during the resume from disk, because the page
2236 * frames they have occupied before the suspend are in use.
2237 */
2238struct highmem_pbe {
2239	struct page *copy_page;	/* data is here now */
2240	struct page *orig_page;	/* data was here before the suspend */
2241	struct highmem_pbe *next;
2242};
2243
2244/*
2245 * List of highmem PBEs needed for restoring the highmem pages that were
2246 * allocated before the suspend and included in the suspend image, but have
2247 * also been allocated by the "resume" kernel, so their contents cannot be
2248 * written directly to their "original" page frames.
2249 */
2250static struct highmem_pbe *highmem_pblist;
2251
2252/**
2253 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2254 * @bm: Memory bitmap.
2255 *
2256 * The bits in @bm that correspond to image pages are assumed to be set.
2257 */
2258static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2259{
2260	unsigned long pfn;
2261	unsigned int cnt = 0;
2262
2263	memory_bm_position_reset(bm);
2264	pfn = memory_bm_next_pfn(bm);
2265	while (pfn != BM_END_OF_MAP) {
2266		if (PageHighMem(pfn_to_page(pfn)))
2267			cnt++;
2268
2269		pfn = memory_bm_next_pfn(bm);
2270	}
2271	return cnt;
2272}
2273
2274static unsigned int safe_highmem_pages;
2275
2276static struct memory_bitmap *safe_highmem_bm;
2277
2278/**
2279 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2280 * @bm: Pointer to an uninitialized memory bitmap structure.
2281 * @nr_highmem_p: Pointer to the number of highmem image pages.
2282 *
2283 * Try to allocate as many highmem pages as there are highmem image pages
2284 * (@nr_highmem_p points to the variable containing the number of highmem image
2285 * pages).  The pages that are "safe" (ie. will not be overwritten when the
2286 * hibernation image is restored entirely) have the corresponding bits set in
2287 * @bm (it must be unitialized).
2288 *
2289 * NOTE: This function should not be called if there are no highmem image pages.
2290 */
2291static int prepare_highmem_image(struct memory_bitmap *bm,
2292				 unsigned int *nr_highmem_p)
2293{
2294	unsigned int to_alloc;
2295
2296	if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2297		return -ENOMEM;
2298
2299	if (get_highmem_buffer(PG_SAFE))
2300		return -ENOMEM;
2301
2302	to_alloc = count_free_highmem_pages();
2303	if (to_alloc > *nr_highmem_p)
2304		to_alloc = *nr_highmem_p;
2305	else
2306		*nr_highmem_p = to_alloc;
2307
2308	safe_highmem_pages = 0;
2309	while (to_alloc-- > 0) {
2310		struct page *page;
2311
2312		page = alloc_page(__GFP_HIGHMEM);
2313		if (!swsusp_page_is_free(page)) {
2314			/* The page is "safe", set its bit the bitmap */
2315			memory_bm_set_bit(bm, page_to_pfn(page));
2316			safe_highmem_pages++;
2317		}
2318		/* Mark the page as allocated */
2319		swsusp_set_page_forbidden(page);
2320		swsusp_set_page_free(page);
2321	}
2322	memory_bm_position_reset(bm);
2323	safe_highmem_bm = bm;
2324	return 0;
2325}
2326
2327static struct page *last_highmem_page;
2328
2329/**
2330 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2331 *
2332 * For a given highmem image page get a buffer that suspend_write_next() should
2333 * return to its caller to write to.
2334 *
2335 * If the page is to be saved to its "original" page frame or a copy of
2336 * the page is to be made in the highmem, @buffer is returned.  Otherwise,
2337 * the copy of the page is to be made in normal memory, so the address of
2338 * the copy is returned.
2339 *
2340 * If @buffer is returned, the caller of suspend_write_next() will write
2341 * the page's contents to @buffer, so they will have to be copied to the
2342 * right location on the next call to suspend_write_next() and it is done
2343 * with the help of copy_last_highmem_page().  For this purpose, if
2344 * @buffer is returned, @last_highmem_page is set to the page to which
2345 * the data will have to be copied from @buffer.
2346 */
2347static void *get_highmem_page_buffer(struct page *page,
2348				     struct chain_allocator *ca)
2349{
2350	struct highmem_pbe *pbe;
2351	void *kaddr;
2352
2353	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2354		/*
2355		 * We have allocated the "original" page frame and we can
2356		 * use it directly to store the loaded page.
2357		 */
2358		last_highmem_page = page;
2359		return buffer;
2360	}
2361	/*
2362	 * The "original" page frame has not been allocated and we have to
2363	 * use a "safe" page frame to store the loaded page.
2364	 */
2365	pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2366	if (!pbe) {
2367		swsusp_free();
2368		return ERR_PTR(-ENOMEM);
2369	}
2370	pbe->orig_page = page;
2371	if (safe_highmem_pages > 0) {
2372		struct page *tmp;
2373
2374		/* Copy of the page will be stored in high memory */
2375		kaddr = buffer;
2376		tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2377		safe_highmem_pages--;
2378		last_highmem_page = tmp;
2379		pbe->copy_page = tmp;
2380	} else {
2381		/* Copy of the page will be stored in normal memory */
2382		kaddr = safe_pages_list;
2383		safe_pages_list = safe_pages_list->next;
2384		pbe->copy_page = virt_to_page(kaddr);
2385	}
2386	pbe->next = highmem_pblist;
2387	highmem_pblist = pbe;
2388	return kaddr;
2389}
2390
2391/**
2392 * copy_last_highmem_page - Copy most the most recent highmem image page.
2393 *
2394 * Copy the contents of a highmem image from @buffer, where the caller of
2395 * snapshot_write_next() has stored them, to the right location represented by
2396 * @last_highmem_page .
2397 */
2398static void copy_last_highmem_page(void)
2399{
2400	if (last_highmem_page) {
2401		void *dst;
2402
2403		dst = kmap_atomic(last_highmem_page);
2404		copy_page(dst, buffer);
2405		kunmap_atomic(dst);
2406		last_highmem_page = NULL;
2407	}
2408}
2409
2410static inline int last_highmem_page_copied(void)
2411{
2412	return !last_highmem_page;
2413}
2414
2415static inline void free_highmem_data(void)
2416{
2417	if (safe_highmem_bm)
2418		memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2419
2420	if (buffer)
2421		free_image_page(buffer, PG_UNSAFE_CLEAR);
2422}
2423#else
2424static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2425
2426static inline int prepare_highmem_image(struct memory_bitmap *bm,
2427					unsigned int *nr_highmem_p) { return 0; }
2428
2429static inline void *get_highmem_page_buffer(struct page *page,
2430					    struct chain_allocator *ca)
2431{
2432	return ERR_PTR(-EINVAL);
2433}
2434
2435static inline void copy_last_highmem_page(void) {}
2436static inline int last_highmem_page_copied(void) { return 1; }
2437static inline void free_highmem_data(void) {}
2438#endif /* CONFIG_HIGHMEM */
2439
2440#define PBES_PER_LINKED_PAGE	(LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2441
2442/**
2443 * prepare_image - Make room for loading hibernation image.
2444 * @new_bm: Unitialized memory bitmap structure.
2445 * @bm: Memory bitmap with unsafe pages marked.
2446 *
2447 * Use @bm to mark the pages that will be overwritten in the process of
2448 * restoring the system memory state from the suspend image ("unsafe" pages)
2449 * and allocate memory for the image.
2450 *
2451 * The idea is to allocate a new memory bitmap first and then allocate
2452 * as many pages as needed for image data, but without specifying what those
2453 * pages will be used for just yet.  Instead, we mark them all as allocated and
2454 * create a lists of "safe" pages to be used later.  On systems with high
2455 * memory a list of "safe" highmem pages is created too.
2456 */
2457static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2458{
2459	unsigned int nr_pages, nr_highmem;
2460	struct linked_page *lp;
2461	int error;
2462
2463	/* If there is no highmem, the buffer will not be necessary */
2464	free_image_page(buffer, PG_UNSAFE_CLEAR);
2465	buffer = NULL;
2466
2467	nr_highmem = count_highmem_image_pages(bm);
2468	mark_unsafe_pages(bm);
2469
2470	error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2471	if (error)
2472		goto Free;
2473
2474	duplicate_memory_bitmap(new_bm, bm);
2475	memory_bm_free(bm, PG_UNSAFE_KEEP);
2476	if (nr_highmem > 0) {
2477		error = prepare_highmem_image(bm, &nr_highmem);
2478		if (error)
2479			goto Free;
2480	}
2481	/*
2482	 * Reserve some safe pages for potential later use.
2483	 *
2484	 * NOTE: This way we make sure there will be enough safe pages for the
2485	 * chain_alloc() in get_buffer().  It is a bit wasteful, but
2486	 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2487	 *
2488	 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2489	 */
2490	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2491	nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2492	while (nr_pages > 0) {
2493		lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2494		if (!lp) {
2495			error = -ENOMEM;
2496			goto Free;
2497		}
2498		lp->next = safe_pages_list;
2499		safe_pages_list = lp;
2500		nr_pages--;
2501	}
2502	/* Preallocate memory for the image */
2503	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2504	while (nr_pages > 0) {
2505		lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2506		if (!lp) {
2507			error = -ENOMEM;
2508			goto Free;
2509		}
2510		if (!swsusp_page_is_free(virt_to_page(lp))) {
2511			/* The page is "safe", add it to the list */
2512			lp->next = safe_pages_list;
2513			safe_pages_list = lp;
2514		}
2515		/* Mark the page as allocated */
2516		swsusp_set_page_forbidden(virt_to_page(lp));
2517		swsusp_set_page_free(virt_to_page(lp));
2518		nr_pages--;
2519	}
2520	return 0;
2521
2522 Free:
2523	swsusp_free();
2524	return error;
2525}
2526
2527/**
2528 * get_buffer - Get the address to store the next image data page.
2529 *
2530 * Get the address that snapshot_write_next() should return to its caller to
2531 * write to.
2532 */
2533static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2534{
2535	struct pbe *pbe;
2536	struct page *page;
2537	unsigned long pfn = memory_bm_next_pfn(bm);
2538
2539	if (pfn == BM_END_OF_MAP)
2540		return ERR_PTR(-EFAULT);
2541
2542	page = pfn_to_page(pfn);
2543	if (PageHighMem(page))
2544		return get_highmem_page_buffer(page, ca);
2545
2546	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2547		/*
2548		 * We have allocated the "original" page frame and we can
2549		 * use it directly to store the loaded page.
2550		 */
2551		return page_address(page);
2552
2553	/*
2554	 * The "original" page frame has not been allocated and we have to
2555	 * use a "safe" page frame to store the loaded page.
2556	 */
2557	pbe = chain_alloc(ca, sizeof(struct pbe));
2558	if (!pbe) {
2559		swsusp_free();
2560		return ERR_PTR(-ENOMEM);
2561	}
2562	pbe->orig_address = page_address(page);
2563	pbe->address = safe_pages_list;
2564	safe_pages_list = safe_pages_list->next;
2565	pbe->next = restore_pblist;
2566	restore_pblist = pbe;
2567	return pbe->address;
2568}
2569
2570/**
2571 * snapshot_write_next - Get the address to store the next image page.
2572 * @handle: Snapshot handle structure to guide the writing.
2573 *
2574 * On the first call, @handle should point to a zeroed snapshot_handle
2575 * structure.  The structure gets populated then and a pointer to it should be
2576 * passed to this function every next time.
2577 *
2578 * On success, the function returns a positive number.  Then, the caller
2579 * is allowed to write up to the returned number of bytes to the memory
2580 * location computed by the data_of() macro.
2581 *
2582 * The function returns 0 to indicate the "end of file" condition.  Negative
2583 * numbers are returned on errors, in which cases the structure pointed to by
2584 * @handle is not updated and should not be used any more.
2585 */
2586int snapshot_write_next(struct snapshot_handle *handle)
2587{
2588	static struct chain_allocator ca;
2589	int error = 0;
2590
2591	/* Check if we have already loaded the entire image */
2592	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2593		return 0;
2594
2595	handle->sync_read = 1;
2596
2597	if (!handle->cur) {
2598		if (!buffer)
2599			/* This makes the buffer be freed by swsusp_free() */
2600			buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2601
2602		if (!buffer)
2603			return -ENOMEM;
2604
2605		handle->buffer = buffer;
2606	} else if (handle->cur == 1) {
2607		error = load_header(buffer);
2608		if (error)
2609			return error;
2610
2611		safe_pages_list = NULL;
2612
2613		error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
2614		if (error)
2615			return error;
2616
 
 
 
 
 
2617		hibernate_restore_protection_begin();
2618	} else if (handle->cur <= nr_meta_pages + 1) {
2619		error = unpack_orig_pfns(buffer, &copy_bm);
2620		if (error)
2621			return error;
2622
2623		if (handle->cur == nr_meta_pages + 1) {
2624			error = prepare_image(&orig_bm, &copy_bm);
2625			if (error)
2626				return error;
2627
2628			chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2629			memory_bm_position_reset(&orig_bm);
2630			restore_pblist = NULL;
2631			handle->buffer = get_buffer(&orig_bm, &ca);
2632			handle->sync_read = 0;
2633			if (IS_ERR(handle->buffer))
2634				return PTR_ERR(handle->buffer);
2635		}
2636	} else {
2637		copy_last_highmem_page();
 
 
2638		hibernate_restore_protect_page(handle->buffer);
2639		handle->buffer = get_buffer(&orig_bm, &ca);
2640		if (IS_ERR(handle->buffer))
2641			return PTR_ERR(handle->buffer);
2642		if (handle->buffer != buffer)
2643			handle->sync_read = 0;
2644	}
2645	handle->cur++;
2646	return PAGE_SIZE;
2647}
2648
2649/**
2650 * snapshot_write_finalize - Complete the loading of a hibernation image.
2651 *
2652 * Must be called after the last call to snapshot_write_next() in case the last
2653 * page in the image happens to be a highmem page and its contents should be
2654 * stored in highmem.  Additionally, it recycles bitmap memory that's not
2655 * necessary any more.
2656 */
2657void snapshot_write_finalize(struct snapshot_handle *handle)
2658{
2659	copy_last_highmem_page();
 
 
 
2660	hibernate_restore_protect_page(handle->buffer);
2661	/* Do that only if we have loaded the image entirely */
2662	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2663		memory_bm_recycle(&orig_bm);
2664		free_highmem_data();
2665	}
2666}
2667
2668int snapshot_image_loaded(struct snapshot_handle *handle)
2669{
2670	return !(!nr_copy_pages || !last_highmem_page_copied() ||
2671			handle->cur <= nr_meta_pages + nr_copy_pages);
2672}
2673
2674#ifdef CONFIG_HIGHMEM
2675/* Assumes that @buf is ready and points to a "safe" page */
2676static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2677				       void *buf)
2678{
2679	void *kaddr1, *kaddr2;
2680
2681	kaddr1 = kmap_atomic(p1);
2682	kaddr2 = kmap_atomic(p2);
2683	copy_page(buf, kaddr1);
2684	copy_page(kaddr1, kaddr2);
2685	copy_page(kaddr2, buf);
2686	kunmap_atomic(kaddr2);
2687	kunmap_atomic(kaddr1);
2688}
2689
2690/**
2691 * restore_highmem - Put highmem image pages into their original locations.
2692 *
2693 * For each highmem page that was in use before hibernation and is included in
2694 * the image, and also has been allocated by the "restore" kernel, swap its
2695 * current contents with the previous (ie. "before hibernation") ones.
2696 *
2697 * If the restore eventually fails, we can call this function once again and
2698 * restore the highmem state as seen by the restore kernel.
2699 */
2700int restore_highmem(void)
2701{
2702	struct highmem_pbe *pbe = highmem_pblist;
2703	void *buf;
2704
2705	if (!pbe)
2706		return 0;
2707
2708	buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2709	if (!buf)
2710		return -ENOMEM;
2711
2712	while (pbe) {
2713		swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2714		pbe = pbe->next;
2715	}
2716	free_image_page(buf, PG_UNSAFE_CLEAR);
2717	return 0;
2718}
2719#endif /* CONFIG_HIGHMEM */
v5.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * linux/kernel/power/snapshot.c
   4 *
   5 * This file provides system snapshot/restore functionality for swsusp.
   6 *
   7 * Copyright (C) 1998-2005 Pavel Machek <pavel@ucw.cz>
   8 * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
   9 */
  10
  11#define pr_fmt(fmt) "PM: " fmt
  12
  13#include <linux/version.h>
  14#include <linux/module.h>
  15#include <linux/mm.h>
  16#include <linux/suspend.h>
  17#include <linux/delay.h>
  18#include <linux/bitops.h>
  19#include <linux/spinlock.h>
  20#include <linux/kernel.h>
  21#include <linux/pm.h>
  22#include <linux/device.h>
  23#include <linux/init.h>
  24#include <linux/memblock.h>
  25#include <linux/nmi.h>
  26#include <linux/syscalls.h>
  27#include <linux/console.h>
  28#include <linux/highmem.h>
  29#include <linux/list.h>
  30#include <linux/slab.h>
  31#include <linux/compiler.h>
  32#include <linux/ktime.h>
  33#include <linux/set_memory.h>
  34
  35#include <linux/uaccess.h>
  36#include <asm/mmu_context.h>
  37#include <asm/pgtable.h>
  38#include <asm/tlbflush.h>
  39#include <asm/io.h>
  40
  41#include "power.h"
  42
  43#if defined(CONFIG_STRICT_KERNEL_RWX) && defined(CONFIG_ARCH_HAS_SET_MEMORY)
  44static bool hibernate_restore_protection;
  45static bool hibernate_restore_protection_active;
  46
  47void enable_restore_image_protection(void)
  48{
  49	hibernate_restore_protection = true;
  50}
  51
  52static inline void hibernate_restore_protection_begin(void)
  53{
  54	hibernate_restore_protection_active = hibernate_restore_protection;
  55}
  56
  57static inline void hibernate_restore_protection_end(void)
  58{
  59	hibernate_restore_protection_active = false;
  60}
  61
  62static inline void hibernate_restore_protect_page(void *page_address)
  63{
  64	if (hibernate_restore_protection_active)
  65		set_memory_ro((unsigned long)page_address, 1);
  66}
  67
  68static inline void hibernate_restore_unprotect_page(void *page_address)
  69{
  70	if (hibernate_restore_protection_active)
  71		set_memory_rw((unsigned long)page_address, 1);
  72}
  73#else
  74static inline void hibernate_restore_protection_begin(void) {}
  75static inline void hibernate_restore_protection_end(void) {}
  76static inline void hibernate_restore_protect_page(void *page_address) {}
  77static inline void hibernate_restore_unprotect_page(void *page_address) {}
  78#endif /* CONFIG_STRICT_KERNEL_RWX  && CONFIG_ARCH_HAS_SET_MEMORY */
  79
  80static int swsusp_page_is_free(struct page *);
  81static void swsusp_set_page_forbidden(struct page *);
  82static void swsusp_unset_page_forbidden(struct page *);
  83
  84/*
  85 * Number of bytes to reserve for memory allocations made by device drivers
  86 * from their ->freeze() and ->freeze_noirq() callbacks so that they don't
  87 * cause image creation to fail (tunable via /sys/power/reserved_size).
  88 */
  89unsigned long reserved_size;
  90
  91void __init hibernate_reserved_size_init(void)
  92{
  93	reserved_size = SPARE_PAGES * PAGE_SIZE;
  94}
  95
  96/*
  97 * Preferred image size in bytes (tunable via /sys/power/image_size).
  98 * When it is set to N, swsusp will do its best to ensure the image
  99 * size will not exceed N bytes, but if that is impossible, it will
 100 * try to create the smallest image possible.
 101 */
 102unsigned long image_size;
 103
 104void __init hibernate_image_size_init(void)
 105{
 106	image_size = ((totalram_pages() * 2) / 5) * PAGE_SIZE;
 107}
 108
 109/*
 110 * List of PBEs needed for restoring the pages that were allocated before
 111 * the suspend and included in the suspend image, but have also been
 112 * allocated by the "resume" kernel, so their contents cannot be written
 113 * directly to their "original" page frames.
 114 */
 115struct pbe *restore_pblist;
 116
 117/* struct linked_page is used to build chains of pages */
 118
 119#define LINKED_PAGE_DATA_SIZE	(PAGE_SIZE - sizeof(void *))
 120
 121struct linked_page {
 122	struct linked_page *next;
 123	char data[LINKED_PAGE_DATA_SIZE];
 124} __packed;
 125
 126/*
 127 * List of "safe" pages (ie. pages that were not used by the image kernel
 128 * before hibernation) that may be used as temporary storage for image kernel
 129 * memory contents.
 130 */
 131static struct linked_page *safe_pages_list;
 132
 133/* Pointer to an auxiliary buffer (1 page) */
 134static void *buffer;
 135
 136#define PG_ANY		0
 137#define PG_SAFE		1
 138#define PG_UNSAFE_CLEAR	1
 139#define PG_UNSAFE_KEEP	0
 140
 141static unsigned int allocated_unsafe_pages;
 142
 143/**
 144 * get_image_page - Allocate a page for a hibernation image.
 145 * @gfp_mask: GFP mask for the allocation.
 146 * @safe_needed: Get pages that were not used before hibernation (restore only)
 147 *
 148 * During image restoration, for storing the PBE list and the image data, we can
 149 * only use memory pages that do not conflict with the pages used before
 150 * hibernation.  The "unsafe" pages have PageNosaveFree set and we count them
 151 * using allocated_unsafe_pages.
 152 *
 153 * Each allocated image page is marked as PageNosave and PageNosaveFree so that
 154 * swsusp_free() can release it.
 155 */
 156static void *get_image_page(gfp_t gfp_mask, int safe_needed)
 157{
 158	void *res;
 159
 160	res = (void *)get_zeroed_page(gfp_mask);
 161	if (safe_needed)
 162		while (res && swsusp_page_is_free(virt_to_page(res))) {
 163			/* The page is unsafe, mark it for swsusp_free() */
 164			swsusp_set_page_forbidden(virt_to_page(res));
 165			allocated_unsafe_pages++;
 166			res = (void *)get_zeroed_page(gfp_mask);
 167		}
 168	if (res) {
 169		swsusp_set_page_forbidden(virt_to_page(res));
 170		swsusp_set_page_free(virt_to_page(res));
 171	}
 172	return res;
 173}
 174
 175static void *__get_safe_page(gfp_t gfp_mask)
 176{
 177	if (safe_pages_list) {
 178		void *ret = safe_pages_list;
 179
 180		safe_pages_list = safe_pages_list->next;
 181		memset(ret, 0, PAGE_SIZE);
 182		return ret;
 183	}
 184	return get_image_page(gfp_mask, PG_SAFE);
 185}
 186
 187unsigned long get_safe_page(gfp_t gfp_mask)
 188{
 189	return (unsigned long)__get_safe_page(gfp_mask);
 190}
 191
 192static struct page *alloc_image_page(gfp_t gfp_mask)
 193{
 194	struct page *page;
 195
 196	page = alloc_page(gfp_mask);
 197	if (page) {
 198		swsusp_set_page_forbidden(page);
 199		swsusp_set_page_free(page);
 200	}
 201	return page;
 202}
 203
 204static void recycle_safe_page(void *page_address)
 205{
 206	struct linked_page *lp = page_address;
 207
 208	lp->next = safe_pages_list;
 209	safe_pages_list = lp;
 210}
 211
 212/**
 213 * free_image_page - Free a page allocated for hibernation image.
 214 * @addr: Address of the page to free.
 215 * @clear_nosave_free: If set, clear the PageNosaveFree bit for the page.
 216 *
 217 * The page to free should have been allocated by get_image_page() (page flags
 218 * set by it are affected).
 219 */
 220static inline void free_image_page(void *addr, int clear_nosave_free)
 221{
 222	struct page *page;
 223
 224	BUG_ON(!virt_addr_valid(addr));
 225
 226	page = virt_to_page(addr);
 227
 228	swsusp_unset_page_forbidden(page);
 229	if (clear_nosave_free)
 230		swsusp_unset_page_free(page);
 231
 232	__free_page(page);
 233}
 234
 235static inline void free_list_of_pages(struct linked_page *list,
 236				      int clear_page_nosave)
 237{
 238	while (list) {
 239		struct linked_page *lp = list->next;
 240
 241		free_image_page(list, clear_page_nosave);
 242		list = lp;
 243	}
 244}
 245
 246/*
 247 * struct chain_allocator is used for allocating small objects out of
 248 * a linked list of pages called 'the chain'.
 249 *
 250 * The chain grows each time when there is no room for a new object in
 251 * the current page.  The allocated objects cannot be freed individually.
 252 * It is only possible to free them all at once, by freeing the entire
 253 * chain.
 254 *
 255 * NOTE: The chain allocator may be inefficient if the allocated objects
 256 * are not much smaller than PAGE_SIZE.
 257 */
 258struct chain_allocator {
 259	struct linked_page *chain;	/* the chain */
 260	unsigned int used_space;	/* total size of objects allocated out
 261					   of the current page */
 262	gfp_t gfp_mask;		/* mask for allocating pages */
 263	int safe_needed;	/* if set, only "safe" pages are allocated */
 264};
 265
 266static void chain_init(struct chain_allocator *ca, gfp_t gfp_mask,
 267		       int safe_needed)
 268{
 269	ca->chain = NULL;
 270	ca->used_space = LINKED_PAGE_DATA_SIZE;
 271	ca->gfp_mask = gfp_mask;
 272	ca->safe_needed = safe_needed;
 273}
 274
 275static void *chain_alloc(struct chain_allocator *ca, unsigned int size)
 276{
 277	void *ret;
 278
 279	if (LINKED_PAGE_DATA_SIZE - ca->used_space < size) {
 280		struct linked_page *lp;
 281
 282		lp = ca->safe_needed ? __get_safe_page(ca->gfp_mask) :
 283					get_image_page(ca->gfp_mask, PG_ANY);
 284		if (!lp)
 285			return NULL;
 286
 287		lp->next = ca->chain;
 288		ca->chain = lp;
 289		ca->used_space = 0;
 290	}
 291	ret = ca->chain->data + ca->used_space;
 292	ca->used_space += size;
 293	return ret;
 294}
 295
 296/**
 297 * Data types related to memory bitmaps.
 298 *
 299 * Memory bitmap is a structure consiting of many linked lists of
 300 * objects.  The main list's elements are of type struct zone_bitmap
 301 * and each of them corresonds to one zone.  For each zone bitmap
 302 * object there is a list of objects of type struct bm_block that
 303 * represent each blocks of bitmap in which information is stored.
 304 *
 305 * struct memory_bitmap contains a pointer to the main list of zone
 306 * bitmap objects, a struct bm_position used for browsing the bitmap,
 307 * and a pointer to the list of pages used for allocating all of the
 308 * zone bitmap objects and bitmap block objects.
 309 *
 310 * NOTE: It has to be possible to lay out the bitmap in memory
 311 * using only allocations of order 0.  Additionally, the bitmap is
 312 * designed to work with arbitrary number of zones (this is over the
 313 * top for now, but let's avoid making unnecessary assumptions ;-).
 314 *
 315 * struct zone_bitmap contains a pointer to a list of bitmap block
 316 * objects and a pointer to the bitmap block object that has been
 317 * most recently used for setting bits.  Additionally, it contains the
 318 * PFNs that correspond to the start and end of the represented zone.
 319 *
 320 * struct bm_block contains a pointer to the memory page in which
 321 * information is stored (in the form of a block of bitmap)
 322 * It also contains the pfns that correspond to the start and end of
 323 * the represented memory area.
 324 *
 325 * The memory bitmap is organized as a radix tree to guarantee fast random
 326 * access to the bits. There is one radix tree for each zone (as returned
 327 * from create_mem_extents).
 328 *
 329 * One radix tree is represented by one struct mem_zone_bm_rtree. There are
 330 * two linked lists for the nodes of the tree, one for the inner nodes and
 331 * one for the leave nodes. The linked leave nodes are used for fast linear
 332 * access of the memory bitmap.
 333 *
 334 * The struct rtree_node represents one node of the radix tree.
 335 */
 336
 337#define BM_END_OF_MAP	(~0UL)
 338
 339#define BM_BITS_PER_BLOCK	(PAGE_SIZE * BITS_PER_BYTE)
 340#define BM_BLOCK_SHIFT		(PAGE_SHIFT + 3)
 341#define BM_BLOCK_MASK		((1UL << BM_BLOCK_SHIFT) - 1)
 342
 343/*
 344 * struct rtree_node is a wrapper struct to link the nodes
 345 * of the rtree together for easy linear iteration over
 346 * bits and easy freeing
 347 */
 348struct rtree_node {
 349	struct list_head list;
 350	unsigned long *data;
 351};
 352
 353/*
 354 * struct mem_zone_bm_rtree represents a bitmap used for one
 355 * populated memory zone.
 356 */
 357struct mem_zone_bm_rtree {
 358	struct list_head list;		/* Link Zones together         */
 359	struct list_head nodes;		/* Radix Tree inner nodes      */
 360	struct list_head leaves;	/* Radix Tree leaves           */
 361	unsigned long start_pfn;	/* Zone start page frame       */
 362	unsigned long end_pfn;		/* Zone end page frame + 1     */
 363	struct rtree_node *rtree;	/* Radix Tree Root             */
 364	int levels;			/* Number of Radix Tree Levels */
 365	unsigned int blocks;		/* Number of Bitmap Blocks     */
 366};
 367
 368/* strcut bm_position is used for browsing memory bitmaps */
 369
 370struct bm_position {
 371	struct mem_zone_bm_rtree *zone;
 372	struct rtree_node *node;
 373	unsigned long node_pfn;
 374	int node_bit;
 375};
 376
 377struct memory_bitmap {
 378	struct list_head zones;
 379	struct linked_page *p_list;	/* list of pages used to store zone
 380					   bitmap objects and bitmap block
 381					   objects */
 382	struct bm_position cur;	/* most recently used bit position */
 383};
 384
 385/* Functions that operate on memory bitmaps */
 386
 387#define BM_ENTRIES_PER_LEVEL	(PAGE_SIZE / sizeof(unsigned long))
 388#if BITS_PER_LONG == 32
 389#define BM_RTREE_LEVEL_SHIFT	(PAGE_SHIFT - 2)
 390#else
 391#define BM_RTREE_LEVEL_SHIFT	(PAGE_SHIFT - 3)
 392#endif
 393#define BM_RTREE_LEVEL_MASK	((1UL << BM_RTREE_LEVEL_SHIFT) - 1)
 394
 395/**
 396 * alloc_rtree_node - Allocate a new node and add it to the radix tree.
 397 *
 398 * This function is used to allocate inner nodes as well as the
 399 * leave nodes of the radix tree. It also adds the node to the
 400 * corresponding linked list passed in by the *list parameter.
 401 */
 402static struct rtree_node *alloc_rtree_node(gfp_t gfp_mask, int safe_needed,
 403					   struct chain_allocator *ca,
 404					   struct list_head *list)
 405{
 406	struct rtree_node *node;
 407
 408	node = chain_alloc(ca, sizeof(struct rtree_node));
 409	if (!node)
 410		return NULL;
 411
 412	node->data = get_image_page(gfp_mask, safe_needed);
 413	if (!node->data)
 414		return NULL;
 415
 416	list_add_tail(&node->list, list);
 417
 418	return node;
 419}
 420
 421/**
 422 * add_rtree_block - Add a new leave node to the radix tree.
 423 *
 424 * The leave nodes need to be allocated in order to keep the leaves
 425 * linked list in order. This is guaranteed by the zone->blocks
 426 * counter.
 427 */
 428static int add_rtree_block(struct mem_zone_bm_rtree *zone, gfp_t gfp_mask,
 429			   int safe_needed, struct chain_allocator *ca)
 430{
 431	struct rtree_node *node, *block, **dst;
 432	unsigned int levels_needed, block_nr;
 433	int i;
 434
 435	block_nr = zone->blocks;
 436	levels_needed = 0;
 437
 438	/* How many levels do we need for this block nr? */
 439	while (block_nr) {
 440		levels_needed += 1;
 441		block_nr >>= BM_RTREE_LEVEL_SHIFT;
 442	}
 443
 444	/* Make sure the rtree has enough levels */
 445	for (i = zone->levels; i < levels_needed; i++) {
 446		node = alloc_rtree_node(gfp_mask, safe_needed, ca,
 447					&zone->nodes);
 448		if (!node)
 449			return -ENOMEM;
 450
 451		node->data[0] = (unsigned long)zone->rtree;
 452		zone->rtree = node;
 453		zone->levels += 1;
 454	}
 455
 456	/* Allocate new block */
 457	block = alloc_rtree_node(gfp_mask, safe_needed, ca, &zone->leaves);
 458	if (!block)
 459		return -ENOMEM;
 460
 461	/* Now walk the rtree to insert the block */
 462	node = zone->rtree;
 463	dst = &zone->rtree;
 464	block_nr = zone->blocks;
 465	for (i = zone->levels; i > 0; i--) {
 466		int index;
 467
 468		if (!node) {
 469			node = alloc_rtree_node(gfp_mask, safe_needed, ca,
 470						&zone->nodes);
 471			if (!node)
 472				return -ENOMEM;
 473			*dst = node;
 474		}
 475
 476		index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
 477		index &= BM_RTREE_LEVEL_MASK;
 478		dst = (struct rtree_node **)&((*dst)->data[index]);
 479		node = *dst;
 480	}
 481
 482	zone->blocks += 1;
 483	*dst = block;
 484
 485	return 0;
 486}
 487
 488static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
 489			       int clear_nosave_free);
 490
 491/**
 492 * create_zone_bm_rtree - Create a radix tree for one zone.
 493 *
 494 * Allocated the mem_zone_bm_rtree structure and initializes it.
 495 * This function also allocated and builds the radix tree for the
 496 * zone.
 497 */
 498static struct mem_zone_bm_rtree *create_zone_bm_rtree(gfp_t gfp_mask,
 499						      int safe_needed,
 500						      struct chain_allocator *ca,
 501						      unsigned long start,
 502						      unsigned long end)
 503{
 504	struct mem_zone_bm_rtree *zone;
 505	unsigned int i, nr_blocks;
 506	unsigned long pages;
 507
 508	pages = end - start;
 509	zone  = chain_alloc(ca, sizeof(struct mem_zone_bm_rtree));
 510	if (!zone)
 511		return NULL;
 512
 513	INIT_LIST_HEAD(&zone->nodes);
 514	INIT_LIST_HEAD(&zone->leaves);
 515	zone->start_pfn = start;
 516	zone->end_pfn = end;
 517	nr_blocks = DIV_ROUND_UP(pages, BM_BITS_PER_BLOCK);
 518
 519	for (i = 0; i < nr_blocks; i++) {
 520		if (add_rtree_block(zone, gfp_mask, safe_needed, ca)) {
 521			free_zone_bm_rtree(zone, PG_UNSAFE_CLEAR);
 522			return NULL;
 523		}
 524	}
 525
 526	return zone;
 527}
 528
 529/**
 530 * free_zone_bm_rtree - Free the memory of the radix tree.
 531 *
 532 * Free all node pages of the radix tree. The mem_zone_bm_rtree
 533 * structure itself is not freed here nor are the rtree_node
 534 * structs.
 535 */
 536static void free_zone_bm_rtree(struct mem_zone_bm_rtree *zone,
 537			       int clear_nosave_free)
 538{
 539	struct rtree_node *node;
 540
 541	list_for_each_entry(node, &zone->nodes, list)
 542		free_image_page(node->data, clear_nosave_free);
 543
 544	list_for_each_entry(node, &zone->leaves, list)
 545		free_image_page(node->data, clear_nosave_free);
 546}
 547
 548static void memory_bm_position_reset(struct memory_bitmap *bm)
 549{
 550	bm->cur.zone = list_entry(bm->zones.next, struct mem_zone_bm_rtree,
 551				  list);
 552	bm->cur.node = list_entry(bm->cur.zone->leaves.next,
 553				  struct rtree_node, list);
 554	bm->cur.node_pfn = 0;
 555	bm->cur.node_bit = 0;
 556}
 557
 558static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free);
 559
 560struct mem_extent {
 561	struct list_head hook;
 562	unsigned long start;
 563	unsigned long end;
 564};
 565
 566/**
 567 * free_mem_extents - Free a list of memory extents.
 568 * @list: List of extents to free.
 569 */
 570static void free_mem_extents(struct list_head *list)
 571{
 572	struct mem_extent *ext, *aux;
 573
 574	list_for_each_entry_safe(ext, aux, list, hook) {
 575		list_del(&ext->hook);
 576		kfree(ext);
 577	}
 578}
 579
 580/**
 581 * create_mem_extents - Create a list of memory extents.
 582 * @list: List to put the extents into.
 583 * @gfp_mask: Mask to use for memory allocations.
 584 *
 585 * The extents represent contiguous ranges of PFNs.
 586 */
 587static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
 588{
 589	struct zone *zone;
 590
 591	INIT_LIST_HEAD(list);
 592
 593	for_each_populated_zone(zone) {
 594		unsigned long zone_start, zone_end;
 595		struct mem_extent *ext, *cur, *aux;
 596
 597		zone_start = zone->zone_start_pfn;
 598		zone_end = zone_end_pfn(zone);
 599
 600		list_for_each_entry(ext, list, hook)
 601			if (zone_start <= ext->end)
 602				break;
 603
 604		if (&ext->hook == list || zone_end < ext->start) {
 605			/* New extent is necessary */
 606			struct mem_extent *new_ext;
 607
 608			new_ext = kzalloc(sizeof(struct mem_extent), gfp_mask);
 609			if (!new_ext) {
 610				free_mem_extents(list);
 611				return -ENOMEM;
 612			}
 613			new_ext->start = zone_start;
 614			new_ext->end = zone_end;
 615			list_add_tail(&new_ext->hook, &ext->hook);
 616			continue;
 617		}
 618
 619		/* Merge this zone's range of PFNs with the existing one */
 620		if (zone_start < ext->start)
 621			ext->start = zone_start;
 622		if (zone_end > ext->end)
 623			ext->end = zone_end;
 624
 625		/* More merging may be possible */
 626		cur = ext;
 627		list_for_each_entry_safe_continue(cur, aux, list, hook) {
 628			if (zone_end < cur->start)
 629				break;
 630			if (zone_end < cur->end)
 631				ext->end = cur->end;
 632			list_del(&cur->hook);
 633			kfree(cur);
 634		}
 635	}
 636
 637	return 0;
 638}
 639
 640/**
 641 * memory_bm_create - Allocate memory for a memory bitmap.
 642 */
 643static int memory_bm_create(struct memory_bitmap *bm, gfp_t gfp_mask,
 644			    int safe_needed)
 645{
 646	struct chain_allocator ca;
 647	struct list_head mem_extents;
 648	struct mem_extent *ext;
 649	int error;
 650
 651	chain_init(&ca, gfp_mask, safe_needed);
 652	INIT_LIST_HEAD(&bm->zones);
 653
 654	error = create_mem_extents(&mem_extents, gfp_mask);
 655	if (error)
 656		return error;
 657
 658	list_for_each_entry(ext, &mem_extents, hook) {
 659		struct mem_zone_bm_rtree *zone;
 660
 661		zone = create_zone_bm_rtree(gfp_mask, safe_needed, &ca,
 662					    ext->start, ext->end);
 663		if (!zone) {
 664			error = -ENOMEM;
 665			goto Error;
 666		}
 667		list_add_tail(&zone->list, &bm->zones);
 668	}
 669
 670	bm->p_list = ca.chain;
 671	memory_bm_position_reset(bm);
 672 Exit:
 673	free_mem_extents(&mem_extents);
 674	return error;
 675
 676 Error:
 677	bm->p_list = ca.chain;
 678	memory_bm_free(bm, PG_UNSAFE_CLEAR);
 679	goto Exit;
 680}
 681
 682/**
 683 * memory_bm_free - Free memory occupied by the memory bitmap.
 684 * @bm: Memory bitmap.
 685 */
 686static void memory_bm_free(struct memory_bitmap *bm, int clear_nosave_free)
 687{
 688	struct mem_zone_bm_rtree *zone;
 689
 690	list_for_each_entry(zone, &bm->zones, list)
 691		free_zone_bm_rtree(zone, clear_nosave_free);
 692
 693	free_list_of_pages(bm->p_list, clear_nosave_free);
 694
 695	INIT_LIST_HEAD(&bm->zones);
 696}
 697
 698/**
 699 * memory_bm_find_bit - Find the bit for a given PFN in a memory bitmap.
 700 *
 701 * Find the bit in memory bitmap @bm that corresponds to the given PFN.
 702 * The cur.zone, cur.block and cur.node_pfn members of @bm are updated.
 703 *
 704 * Walk the radix tree to find the page containing the bit that represents @pfn
 705 * and return the position of the bit in @addr and @bit_nr.
 706 */
 707static int memory_bm_find_bit(struct memory_bitmap *bm, unsigned long pfn,
 708			      void **addr, unsigned int *bit_nr)
 709{
 710	struct mem_zone_bm_rtree *curr, *zone;
 711	struct rtree_node *node;
 712	int i, block_nr;
 713
 714	zone = bm->cur.zone;
 715
 716	if (pfn >= zone->start_pfn && pfn < zone->end_pfn)
 717		goto zone_found;
 718
 719	zone = NULL;
 720
 721	/* Find the right zone */
 722	list_for_each_entry(curr, &bm->zones, list) {
 723		if (pfn >= curr->start_pfn && pfn < curr->end_pfn) {
 724			zone = curr;
 725			break;
 726		}
 727	}
 728
 729	if (!zone)
 730		return -EFAULT;
 731
 732zone_found:
 733	/*
 734	 * We have found the zone. Now walk the radix tree to find the leaf node
 735	 * for our PFN.
 736	 */
 
 
 
 
 
 
 737	node = bm->cur.node;
 738	if (((pfn - zone->start_pfn) & ~BM_BLOCK_MASK) == bm->cur.node_pfn)
 
 739		goto node_found;
 740
 741	node      = zone->rtree;
 742	block_nr  = (pfn - zone->start_pfn) >> BM_BLOCK_SHIFT;
 743
 744	for (i = zone->levels; i > 0; i--) {
 745		int index;
 746
 747		index = block_nr >> ((i - 1) * BM_RTREE_LEVEL_SHIFT);
 748		index &= BM_RTREE_LEVEL_MASK;
 749		BUG_ON(node->data[index] == 0);
 750		node = (struct rtree_node *)node->data[index];
 751	}
 752
 753node_found:
 754	/* Update last position */
 755	bm->cur.zone = zone;
 756	bm->cur.node = node;
 757	bm->cur.node_pfn = (pfn - zone->start_pfn) & ~BM_BLOCK_MASK;
 758
 759	/* Set return values */
 760	*addr = node->data;
 761	*bit_nr = (pfn - zone->start_pfn) & BM_BLOCK_MASK;
 762
 763	return 0;
 764}
 765
 766static void memory_bm_set_bit(struct memory_bitmap *bm, unsigned long pfn)
 767{
 768	void *addr;
 769	unsigned int bit;
 770	int error;
 771
 772	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 773	BUG_ON(error);
 774	set_bit(bit, addr);
 775}
 776
 777static int mem_bm_set_bit_check(struct memory_bitmap *bm, unsigned long pfn)
 778{
 779	void *addr;
 780	unsigned int bit;
 781	int error;
 782
 783	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 784	if (!error)
 785		set_bit(bit, addr);
 786
 787	return error;
 788}
 789
 790static void memory_bm_clear_bit(struct memory_bitmap *bm, unsigned long pfn)
 791{
 792	void *addr;
 793	unsigned int bit;
 794	int error;
 795
 796	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 797	BUG_ON(error);
 798	clear_bit(bit, addr);
 799}
 800
 801static void memory_bm_clear_current(struct memory_bitmap *bm)
 802{
 803	int bit;
 804
 805	bit = max(bm->cur.node_bit - 1, 0);
 806	clear_bit(bit, bm->cur.node->data);
 807}
 808
 809static int memory_bm_test_bit(struct memory_bitmap *bm, unsigned long pfn)
 810{
 811	void *addr;
 812	unsigned int bit;
 813	int error;
 814
 815	error = memory_bm_find_bit(bm, pfn, &addr, &bit);
 816	BUG_ON(error);
 817	return test_bit(bit, addr);
 818}
 819
 820static bool memory_bm_pfn_present(struct memory_bitmap *bm, unsigned long pfn)
 821{
 822	void *addr;
 823	unsigned int bit;
 824
 825	return !memory_bm_find_bit(bm, pfn, &addr, &bit);
 826}
 827
 828/*
 829 * rtree_next_node - Jump to the next leaf node.
 830 *
 831 * Set the position to the beginning of the next node in the
 832 * memory bitmap. This is either the next node in the current
 833 * zone's radix tree or the first node in the radix tree of the
 834 * next zone.
 835 *
 836 * Return true if there is a next node, false otherwise.
 837 */
 838static bool rtree_next_node(struct memory_bitmap *bm)
 839{
 840	if (!list_is_last(&bm->cur.node->list, &bm->cur.zone->leaves)) {
 841		bm->cur.node = list_entry(bm->cur.node->list.next,
 842					  struct rtree_node, list);
 843		bm->cur.node_pfn += BM_BITS_PER_BLOCK;
 844		bm->cur.node_bit  = 0;
 845		touch_softlockup_watchdog();
 846		return true;
 847	}
 848
 849	/* No more nodes, goto next zone */
 850	if (!list_is_last(&bm->cur.zone->list, &bm->zones)) {
 851		bm->cur.zone = list_entry(bm->cur.zone->list.next,
 852				  struct mem_zone_bm_rtree, list);
 853		bm->cur.node = list_entry(bm->cur.zone->leaves.next,
 854					  struct rtree_node, list);
 855		bm->cur.node_pfn = 0;
 856		bm->cur.node_bit = 0;
 857		return true;
 858	}
 859
 860	/* No more zones */
 861	return false;
 862}
 863
 864/**
 865 * memory_bm_rtree_next_pfn - Find the next set bit in a memory bitmap.
 866 * @bm: Memory bitmap.
 867 *
 868 * Starting from the last returned position this function searches for the next
 869 * set bit in @bm and returns the PFN represented by it.  If no more bits are
 870 * set, BM_END_OF_MAP is returned.
 871 *
 872 * It is required to run memory_bm_position_reset() before the first call to
 873 * this function for the given memory bitmap.
 874 */
 875static unsigned long memory_bm_next_pfn(struct memory_bitmap *bm)
 876{
 877	unsigned long bits, pfn, pages;
 878	int bit;
 879
 880	do {
 881		pages	  = bm->cur.zone->end_pfn - bm->cur.zone->start_pfn;
 882		bits      = min(pages - bm->cur.node_pfn, BM_BITS_PER_BLOCK);
 883		bit	  = find_next_bit(bm->cur.node->data, bits,
 884					  bm->cur.node_bit);
 885		if (bit < bits) {
 886			pfn = bm->cur.zone->start_pfn + bm->cur.node_pfn + bit;
 887			bm->cur.node_bit = bit + 1;
 888			return pfn;
 889		}
 890	} while (rtree_next_node(bm));
 891
 892	return BM_END_OF_MAP;
 893}
 894
 895/*
 896 * This structure represents a range of page frames the contents of which
 897 * should not be saved during hibernation.
 898 */
 899struct nosave_region {
 900	struct list_head list;
 901	unsigned long start_pfn;
 902	unsigned long end_pfn;
 903};
 904
 905static LIST_HEAD(nosave_regions);
 906
 907static void recycle_zone_bm_rtree(struct mem_zone_bm_rtree *zone)
 908{
 909	struct rtree_node *node;
 910
 911	list_for_each_entry(node, &zone->nodes, list)
 912		recycle_safe_page(node->data);
 913
 914	list_for_each_entry(node, &zone->leaves, list)
 915		recycle_safe_page(node->data);
 916}
 917
 918static void memory_bm_recycle(struct memory_bitmap *bm)
 919{
 920	struct mem_zone_bm_rtree *zone;
 921	struct linked_page *p_list;
 922
 923	list_for_each_entry(zone, &bm->zones, list)
 924		recycle_zone_bm_rtree(zone);
 925
 926	p_list = bm->p_list;
 927	while (p_list) {
 928		struct linked_page *lp = p_list;
 929
 930		p_list = lp->next;
 931		recycle_safe_page(lp);
 932	}
 933}
 934
 935/**
 936 * register_nosave_region - Register a region of unsaveable memory.
 937 *
 938 * Register a range of page frames the contents of which should not be saved
 939 * during hibernation (to be used in the early initialization code).
 940 */
 941void __init __register_nosave_region(unsigned long start_pfn,
 942				     unsigned long end_pfn, int use_kmalloc)
 943{
 944	struct nosave_region *region;
 945
 946	if (start_pfn >= end_pfn)
 947		return;
 948
 949	if (!list_empty(&nosave_regions)) {
 950		/* Try to extend the previous region (they should be sorted) */
 951		region = list_entry(nosave_regions.prev,
 952					struct nosave_region, list);
 953		if (region->end_pfn == start_pfn) {
 954			region->end_pfn = end_pfn;
 955			goto Report;
 956		}
 957	}
 958	if (use_kmalloc) {
 959		/* During init, this shouldn't fail */
 960		region = kmalloc(sizeof(struct nosave_region), GFP_KERNEL);
 961		BUG_ON(!region);
 962	} else {
 963		/* This allocation cannot fail */
 964		region = memblock_alloc(sizeof(struct nosave_region),
 965					SMP_CACHE_BYTES);
 966		if (!region)
 967			panic("%s: Failed to allocate %zu bytes\n", __func__,
 968			      sizeof(struct nosave_region));
 969	}
 970	region->start_pfn = start_pfn;
 971	region->end_pfn = end_pfn;
 972	list_add_tail(&region->list, &nosave_regions);
 973 Report:
 974	pr_info("Registered nosave memory: [mem %#010llx-%#010llx]\n",
 975		(unsigned long long) start_pfn << PAGE_SHIFT,
 976		((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
 977}
 978
 979/*
 980 * Set bits in this map correspond to the page frames the contents of which
 981 * should not be saved during the suspend.
 982 */
 983static struct memory_bitmap *forbidden_pages_map;
 984
 985/* Set bits in this map correspond to free page frames. */
 986static struct memory_bitmap *free_pages_map;
 987
 988/*
 989 * Each page frame allocated for creating the image is marked by setting the
 990 * corresponding bits in forbidden_pages_map and free_pages_map simultaneously
 991 */
 992
 993void swsusp_set_page_free(struct page *page)
 994{
 995	if (free_pages_map)
 996		memory_bm_set_bit(free_pages_map, page_to_pfn(page));
 997}
 998
 999static int swsusp_page_is_free(struct page *page)
1000{
1001	return free_pages_map ?
1002		memory_bm_test_bit(free_pages_map, page_to_pfn(page)) : 0;
1003}
1004
1005void swsusp_unset_page_free(struct page *page)
1006{
1007	if (free_pages_map)
1008		memory_bm_clear_bit(free_pages_map, page_to_pfn(page));
1009}
1010
1011static void swsusp_set_page_forbidden(struct page *page)
1012{
1013	if (forbidden_pages_map)
1014		memory_bm_set_bit(forbidden_pages_map, page_to_pfn(page));
1015}
1016
1017int swsusp_page_is_forbidden(struct page *page)
1018{
1019	return forbidden_pages_map ?
1020		memory_bm_test_bit(forbidden_pages_map, page_to_pfn(page)) : 0;
1021}
1022
1023static void swsusp_unset_page_forbidden(struct page *page)
1024{
1025	if (forbidden_pages_map)
1026		memory_bm_clear_bit(forbidden_pages_map, page_to_pfn(page));
1027}
1028
1029/**
1030 * mark_nosave_pages - Mark pages that should not be saved.
1031 * @bm: Memory bitmap.
1032 *
1033 * Set the bits in @bm that correspond to the page frames the contents of which
1034 * should not be saved.
1035 */
1036static void mark_nosave_pages(struct memory_bitmap *bm)
1037{
1038	struct nosave_region *region;
1039
1040	if (list_empty(&nosave_regions))
1041		return;
1042
1043	list_for_each_entry(region, &nosave_regions, list) {
1044		unsigned long pfn;
1045
1046		pr_debug("Marking nosave pages: [mem %#010llx-%#010llx]\n",
1047			 (unsigned long long) region->start_pfn << PAGE_SHIFT,
1048			 ((unsigned long long) region->end_pfn << PAGE_SHIFT)
1049				- 1);
1050
1051		for (pfn = region->start_pfn; pfn < region->end_pfn; pfn++)
1052			if (pfn_valid(pfn)) {
1053				/*
1054				 * It is safe to ignore the result of
1055				 * mem_bm_set_bit_check() here, since we won't
1056				 * touch the PFNs for which the error is
1057				 * returned anyway.
1058				 */
1059				mem_bm_set_bit_check(bm, pfn);
1060			}
1061	}
1062}
1063
1064/**
1065 * create_basic_memory_bitmaps - Create bitmaps to hold basic page information.
1066 *
1067 * Create bitmaps needed for marking page frames that should not be saved and
1068 * free page frames.  The forbidden_pages_map and free_pages_map pointers are
1069 * only modified if everything goes well, because we don't want the bits to be
1070 * touched before both bitmaps are set up.
1071 */
1072int create_basic_memory_bitmaps(void)
1073{
1074	struct memory_bitmap *bm1, *bm2;
1075	int error = 0;
1076
1077	if (forbidden_pages_map && free_pages_map)
1078		return 0;
1079	else
1080		BUG_ON(forbidden_pages_map || free_pages_map);
1081
1082	bm1 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1083	if (!bm1)
1084		return -ENOMEM;
1085
1086	error = memory_bm_create(bm1, GFP_KERNEL, PG_ANY);
1087	if (error)
1088		goto Free_first_object;
1089
1090	bm2 = kzalloc(sizeof(struct memory_bitmap), GFP_KERNEL);
1091	if (!bm2)
1092		goto Free_first_bitmap;
1093
1094	error = memory_bm_create(bm2, GFP_KERNEL, PG_ANY);
1095	if (error)
1096		goto Free_second_object;
1097
1098	forbidden_pages_map = bm1;
1099	free_pages_map = bm2;
1100	mark_nosave_pages(forbidden_pages_map);
1101
1102	pr_debug("Basic memory bitmaps created\n");
1103
1104	return 0;
1105
1106 Free_second_object:
1107	kfree(bm2);
1108 Free_first_bitmap:
1109 	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1110 Free_first_object:
1111	kfree(bm1);
1112	return -ENOMEM;
1113}
1114
1115/**
1116 * free_basic_memory_bitmaps - Free memory bitmaps holding basic information.
1117 *
1118 * Free memory bitmaps allocated by create_basic_memory_bitmaps().  The
1119 * auxiliary pointers are necessary so that the bitmaps themselves are not
1120 * referred to while they are being freed.
1121 */
1122void free_basic_memory_bitmaps(void)
1123{
1124	struct memory_bitmap *bm1, *bm2;
1125
1126	if (WARN_ON(!(forbidden_pages_map && free_pages_map)))
1127		return;
1128
1129	bm1 = forbidden_pages_map;
1130	bm2 = free_pages_map;
1131	forbidden_pages_map = NULL;
1132	free_pages_map = NULL;
1133	memory_bm_free(bm1, PG_UNSAFE_CLEAR);
1134	kfree(bm1);
1135	memory_bm_free(bm2, PG_UNSAFE_CLEAR);
1136	kfree(bm2);
1137
1138	pr_debug("Basic memory bitmaps freed\n");
1139}
1140
1141void clear_free_pages(void)
1142{
1143#ifdef CONFIG_PAGE_POISONING_ZERO
1144	struct memory_bitmap *bm = free_pages_map;
1145	unsigned long pfn;
1146
1147	if (WARN_ON(!(free_pages_map)))
1148		return;
1149
1150	memory_bm_position_reset(bm);
1151	pfn = memory_bm_next_pfn(bm);
1152	while (pfn != BM_END_OF_MAP) {
1153		if (pfn_valid(pfn))
1154			clear_highpage(pfn_to_page(pfn));
 
1155
1156		pfn = memory_bm_next_pfn(bm);
 
 
 
1157	}
1158	memory_bm_position_reset(bm);
1159	pr_info("free pages cleared after restore\n");
1160#endif /* PAGE_POISONING_ZERO */
1161}
1162
1163/**
1164 * snapshot_additional_pages - Estimate the number of extra pages needed.
1165 * @zone: Memory zone to carry out the computation for.
1166 *
1167 * Estimate the number of additional pages needed for setting up a hibernation
1168 * image data structures for @zone (usually, the returned value is greater than
1169 * the exact number).
1170 */
1171unsigned int snapshot_additional_pages(struct zone *zone)
1172{
1173	unsigned int rtree, nodes;
1174
1175	rtree = nodes = DIV_ROUND_UP(zone->spanned_pages, BM_BITS_PER_BLOCK);
1176	rtree += DIV_ROUND_UP(rtree * sizeof(struct rtree_node),
1177			      LINKED_PAGE_DATA_SIZE);
1178	while (nodes > 1) {
1179		nodes = DIV_ROUND_UP(nodes, BM_ENTRIES_PER_LEVEL);
1180		rtree += nodes;
1181	}
1182
1183	return 2 * rtree;
1184}
1185
1186#ifdef CONFIG_HIGHMEM
1187/**
1188 * count_free_highmem_pages - Compute the total number of free highmem pages.
1189 *
1190 * The returned number is system-wide.
1191 */
1192static unsigned int count_free_highmem_pages(void)
1193{
1194	struct zone *zone;
1195	unsigned int cnt = 0;
1196
1197	for_each_populated_zone(zone)
1198		if (is_highmem(zone))
1199			cnt += zone_page_state(zone, NR_FREE_PAGES);
1200
1201	return cnt;
1202}
1203
1204/**
1205 * saveable_highmem_page - Check if a highmem page is saveable.
1206 *
1207 * Determine whether a highmem page should be included in a hibernation image.
1208 *
1209 * We should save the page if it isn't Nosave or NosaveFree, or Reserved,
1210 * and it isn't part of a free chunk of pages.
1211 */
1212static struct page *saveable_highmem_page(struct zone *zone, unsigned long pfn)
1213{
1214	struct page *page;
1215
1216	if (!pfn_valid(pfn))
1217		return NULL;
1218
1219	page = pfn_to_online_page(pfn);
1220	if (!page || page_zone(page) != zone)
1221		return NULL;
1222
1223	BUG_ON(!PageHighMem(page));
1224
1225	if (swsusp_page_is_forbidden(page) ||  swsusp_page_is_free(page))
1226		return NULL;
1227
1228	if (PageReserved(page) || PageOffline(page))
1229		return NULL;
1230
1231	if (page_is_guard(page))
1232		return NULL;
1233
1234	return page;
1235}
1236
1237/**
1238 * count_highmem_pages - Compute the total number of saveable highmem pages.
1239 */
1240static unsigned int count_highmem_pages(void)
1241{
1242	struct zone *zone;
1243	unsigned int n = 0;
1244
1245	for_each_populated_zone(zone) {
1246		unsigned long pfn, max_zone_pfn;
1247
1248		if (!is_highmem(zone))
1249			continue;
1250
1251		mark_free_pages(zone);
1252		max_zone_pfn = zone_end_pfn(zone);
1253		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1254			if (saveable_highmem_page(zone, pfn))
1255				n++;
1256	}
1257	return n;
1258}
1259#else
1260static inline void *saveable_highmem_page(struct zone *z, unsigned long p)
1261{
1262	return NULL;
1263}
1264#endif /* CONFIG_HIGHMEM */
1265
1266/**
1267 * saveable_page - Check if the given page is saveable.
1268 *
1269 * Determine whether a non-highmem page should be included in a hibernation
1270 * image.
1271 *
1272 * We should save the page if it isn't Nosave, and is not in the range
1273 * of pages statically defined as 'unsaveable', and it isn't part of
1274 * a free chunk of pages.
1275 */
1276static struct page *saveable_page(struct zone *zone, unsigned long pfn)
1277{
1278	struct page *page;
1279
1280	if (!pfn_valid(pfn))
1281		return NULL;
1282
1283	page = pfn_to_online_page(pfn);
1284	if (!page || page_zone(page) != zone)
1285		return NULL;
1286
1287	BUG_ON(PageHighMem(page));
1288
1289	if (swsusp_page_is_forbidden(page) || swsusp_page_is_free(page))
1290		return NULL;
1291
1292	if (PageOffline(page))
1293		return NULL;
1294
1295	if (PageReserved(page)
1296	    && (!kernel_page_present(page) || pfn_is_nosave(pfn)))
1297		return NULL;
1298
1299	if (page_is_guard(page))
1300		return NULL;
1301
1302	return page;
1303}
1304
1305/**
1306 * count_data_pages - Compute the total number of saveable non-highmem pages.
1307 */
1308static unsigned int count_data_pages(void)
1309{
1310	struct zone *zone;
1311	unsigned long pfn, max_zone_pfn;
1312	unsigned int n = 0;
1313
1314	for_each_populated_zone(zone) {
1315		if (is_highmem(zone))
1316			continue;
1317
1318		mark_free_pages(zone);
1319		max_zone_pfn = zone_end_pfn(zone);
1320		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1321			if (saveable_page(zone, pfn))
1322				n++;
1323	}
1324	return n;
1325}
1326
1327/*
1328 * This is needed, because copy_page and memcpy are not usable for copying
1329 * task structs.
1330 */
1331static inline void do_copy_page(long *dst, long *src)
1332{
1333	int n;
1334
1335	for (n = PAGE_SIZE / sizeof(long); n; n--)
1336		*dst++ = *src++;
1337}
1338
1339/**
1340 * safe_copy_page - Copy a page in a safe way.
1341 *
1342 * Check if the page we are going to copy is marked as present in the kernel
1343 * page tables. This always is the case if CONFIG_DEBUG_PAGEALLOC or
1344 * CONFIG_ARCH_HAS_SET_DIRECT_MAP is not set. In that case kernel_page_present()
1345 * always returns 'true'.
1346 */
1347static void safe_copy_page(void *dst, struct page *s_page)
1348{
1349	if (kernel_page_present(s_page)) {
1350		do_copy_page(dst, page_address(s_page));
1351	} else {
1352		kernel_map_pages(s_page, 1, 1);
1353		do_copy_page(dst, page_address(s_page));
1354		kernel_map_pages(s_page, 1, 0);
1355	}
1356}
1357
1358#ifdef CONFIG_HIGHMEM
1359static inline struct page *page_is_saveable(struct zone *zone, unsigned long pfn)
1360{
1361	return is_highmem(zone) ?
1362		saveable_highmem_page(zone, pfn) : saveable_page(zone, pfn);
1363}
1364
1365static void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1366{
1367	struct page *s_page, *d_page;
1368	void *src, *dst;
1369
1370	s_page = pfn_to_page(src_pfn);
1371	d_page = pfn_to_page(dst_pfn);
1372	if (PageHighMem(s_page)) {
1373		src = kmap_atomic(s_page);
1374		dst = kmap_atomic(d_page);
1375		do_copy_page(dst, src);
1376		kunmap_atomic(dst);
1377		kunmap_atomic(src);
1378	} else {
1379		if (PageHighMem(d_page)) {
1380			/*
1381			 * The page pointed to by src may contain some kernel
1382			 * data modified by kmap_atomic()
1383			 */
1384			safe_copy_page(buffer, s_page);
1385			dst = kmap_atomic(d_page);
1386			copy_page(dst, buffer);
1387			kunmap_atomic(dst);
1388		} else {
1389			safe_copy_page(page_address(d_page), s_page);
1390		}
1391	}
1392}
1393#else
1394#define page_is_saveable(zone, pfn)	saveable_page(zone, pfn)
1395
1396static inline void copy_data_page(unsigned long dst_pfn, unsigned long src_pfn)
1397{
1398	safe_copy_page(page_address(pfn_to_page(dst_pfn)),
1399				pfn_to_page(src_pfn));
1400}
1401#endif /* CONFIG_HIGHMEM */
1402
1403static void copy_data_pages(struct memory_bitmap *copy_bm,
1404			    struct memory_bitmap *orig_bm)
1405{
1406	struct zone *zone;
1407	unsigned long pfn;
1408
1409	for_each_populated_zone(zone) {
1410		unsigned long max_zone_pfn;
1411
1412		mark_free_pages(zone);
1413		max_zone_pfn = zone_end_pfn(zone);
1414		for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
1415			if (page_is_saveable(zone, pfn))
1416				memory_bm_set_bit(orig_bm, pfn);
1417	}
1418	memory_bm_position_reset(orig_bm);
1419	memory_bm_position_reset(copy_bm);
1420	for(;;) {
1421		pfn = memory_bm_next_pfn(orig_bm);
1422		if (unlikely(pfn == BM_END_OF_MAP))
1423			break;
1424		copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
1425	}
1426}
1427
1428/* Total number of image pages */
1429static unsigned int nr_copy_pages;
1430/* Number of pages needed for saving the original pfns of the image pages */
1431static unsigned int nr_meta_pages;
1432/*
1433 * Numbers of normal and highmem page frames allocated for hibernation image
1434 * before suspending devices.
1435 */
1436static unsigned int alloc_normal, alloc_highmem;
1437/*
1438 * Memory bitmap used for marking saveable pages (during hibernation) or
1439 * hibernation image pages (during restore)
1440 */
1441static struct memory_bitmap orig_bm;
1442/*
1443 * Memory bitmap used during hibernation for marking allocated page frames that
1444 * will contain copies of saveable pages.  During restore it is initially used
1445 * for marking hibernation image pages, but then the set bits from it are
1446 * duplicated in @orig_bm and it is released.  On highmem systems it is next
1447 * used for marking "safe" highmem pages, but it has to be reinitialized for
1448 * this purpose.
1449 */
1450static struct memory_bitmap copy_bm;
1451
1452/**
1453 * swsusp_free - Free pages allocated for hibernation image.
1454 *
1455 * Image pages are alocated before snapshot creation, so they need to be
1456 * released after resume.
1457 */
1458void swsusp_free(void)
1459{
1460	unsigned long fb_pfn, fr_pfn;
1461
1462	if (!forbidden_pages_map || !free_pages_map)
1463		goto out;
1464
1465	memory_bm_position_reset(forbidden_pages_map);
1466	memory_bm_position_reset(free_pages_map);
1467
1468loop:
1469	fr_pfn = memory_bm_next_pfn(free_pages_map);
1470	fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1471
1472	/*
1473	 * Find the next bit set in both bitmaps. This is guaranteed to
1474	 * terminate when fb_pfn == fr_pfn == BM_END_OF_MAP.
1475	 */
1476	do {
1477		if (fb_pfn < fr_pfn)
1478			fb_pfn = memory_bm_next_pfn(forbidden_pages_map);
1479		if (fr_pfn < fb_pfn)
1480			fr_pfn = memory_bm_next_pfn(free_pages_map);
1481	} while (fb_pfn != fr_pfn);
1482
1483	if (fr_pfn != BM_END_OF_MAP && pfn_valid(fr_pfn)) {
1484		struct page *page = pfn_to_page(fr_pfn);
1485
1486		memory_bm_clear_current(forbidden_pages_map);
1487		memory_bm_clear_current(free_pages_map);
1488		hibernate_restore_unprotect_page(page_address(page));
1489		__free_page(page);
1490		goto loop;
1491	}
1492
1493out:
1494	nr_copy_pages = 0;
1495	nr_meta_pages = 0;
1496	restore_pblist = NULL;
1497	buffer = NULL;
1498	alloc_normal = 0;
1499	alloc_highmem = 0;
1500	hibernate_restore_protection_end();
1501}
1502
1503/* Helper functions used for the shrinking of memory. */
1504
1505#define GFP_IMAGE	(GFP_KERNEL | __GFP_NOWARN)
1506
1507/**
1508 * preallocate_image_pages - Allocate a number of pages for hibernation image.
1509 * @nr_pages: Number of page frames to allocate.
1510 * @mask: GFP flags to use for the allocation.
1511 *
1512 * Return value: Number of page frames actually allocated
1513 */
1514static unsigned long preallocate_image_pages(unsigned long nr_pages, gfp_t mask)
1515{
1516	unsigned long nr_alloc = 0;
1517
1518	while (nr_pages > 0) {
1519		struct page *page;
1520
1521		page = alloc_image_page(mask);
1522		if (!page)
1523			break;
1524		memory_bm_set_bit(&copy_bm, page_to_pfn(page));
1525		if (PageHighMem(page))
1526			alloc_highmem++;
1527		else
1528			alloc_normal++;
1529		nr_pages--;
1530		nr_alloc++;
1531	}
1532
1533	return nr_alloc;
1534}
1535
1536static unsigned long preallocate_image_memory(unsigned long nr_pages,
1537					      unsigned long avail_normal)
1538{
1539	unsigned long alloc;
1540
1541	if (avail_normal <= alloc_normal)
1542		return 0;
1543
1544	alloc = avail_normal - alloc_normal;
1545	if (nr_pages < alloc)
1546		alloc = nr_pages;
1547
1548	return preallocate_image_pages(alloc, GFP_IMAGE);
1549}
1550
1551#ifdef CONFIG_HIGHMEM
1552static unsigned long preallocate_image_highmem(unsigned long nr_pages)
1553{
1554	return preallocate_image_pages(nr_pages, GFP_IMAGE | __GFP_HIGHMEM);
1555}
1556
1557/**
1558 *  __fraction - Compute (an approximation of) x * (multiplier / base).
1559 */
1560static unsigned long __fraction(u64 x, u64 multiplier, u64 base)
1561{
1562	x *= multiplier;
1563	do_div(x, base);
1564	return (unsigned long)x;
1565}
1566
1567static unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1568						  unsigned long highmem,
1569						  unsigned long total)
1570{
1571	unsigned long alloc = __fraction(nr_pages, highmem, total);
1572
1573	return preallocate_image_pages(alloc, GFP_IMAGE | __GFP_HIGHMEM);
1574}
1575#else /* CONFIG_HIGHMEM */
1576static inline unsigned long preallocate_image_highmem(unsigned long nr_pages)
1577{
1578	return 0;
1579}
1580
1581static inline unsigned long preallocate_highmem_fraction(unsigned long nr_pages,
1582							 unsigned long highmem,
1583							 unsigned long total)
1584{
1585	return 0;
1586}
1587#endif /* CONFIG_HIGHMEM */
1588
1589/**
1590 * free_unnecessary_pages - Release preallocated pages not needed for the image.
1591 */
1592static unsigned long free_unnecessary_pages(void)
1593{
1594	unsigned long save, to_free_normal, to_free_highmem, free;
1595
1596	save = count_data_pages();
1597	if (alloc_normal >= save) {
1598		to_free_normal = alloc_normal - save;
1599		save = 0;
1600	} else {
1601		to_free_normal = 0;
1602		save -= alloc_normal;
1603	}
1604	save += count_highmem_pages();
1605	if (alloc_highmem >= save) {
1606		to_free_highmem = alloc_highmem - save;
1607	} else {
1608		to_free_highmem = 0;
1609		save -= alloc_highmem;
1610		if (to_free_normal > save)
1611			to_free_normal -= save;
1612		else
1613			to_free_normal = 0;
1614	}
1615	free = to_free_normal + to_free_highmem;
1616
1617	memory_bm_position_reset(&copy_bm);
1618
1619	while (to_free_normal > 0 || to_free_highmem > 0) {
1620		unsigned long pfn = memory_bm_next_pfn(&copy_bm);
1621		struct page *page = pfn_to_page(pfn);
1622
1623		if (PageHighMem(page)) {
1624			if (!to_free_highmem)
1625				continue;
1626			to_free_highmem--;
1627			alloc_highmem--;
1628		} else {
1629			if (!to_free_normal)
1630				continue;
1631			to_free_normal--;
1632			alloc_normal--;
1633		}
1634		memory_bm_clear_bit(&copy_bm, pfn);
1635		swsusp_unset_page_forbidden(page);
1636		swsusp_unset_page_free(page);
1637		__free_page(page);
1638	}
1639
1640	return free;
1641}
1642
1643/**
1644 * minimum_image_size - Estimate the minimum acceptable size of an image.
1645 * @saveable: Number of saveable pages in the system.
1646 *
1647 * We want to avoid attempting to free too much memory too hard, so estimate the
1648 * minimum acceptable size of a hibernation image to use as the lower limit for
1649 * preallocating memory.
1650 *
1651 * We assume that the minimum image size should be proportional to
1652 *
1653 * [number of saveable pages] - [number of pages that can be freed in theory]
1654 *
1655 * where the second term is the sum of (1) reclaimable slab pages, (2) active
1656 * and (3) inactive anonymous pages, (4) active and (5) inactive file pages.
1657 */
1658static unsigned long minimum_image_size(unsigned long saveable)
1659{
1660	unsigned long size;
1661
1662	size = global_node_page_state(NR_SLAB_RECLAIMABLE)
1663		+ global_node_page_state(NR_ACTIVE_ANON)
1664		+ global_node_page_state(NR_INACTIVE_ANON)
1665		+ global_node_page_state(NR_ACTIVE_FILE)
1666		+ global_node_page_state(NR_INACTIVE_FILE);
1667
1668	return saveable <= size ? 0 : saveable - size;
1669}
1670
1671/**
1672 * hibernate_preallocate_memory - Preallocate memory for hibernation image.
1673 *
1674 * To create a hibernation image it is necessary to make a copy of every page
1675 * frame in use.  We also need a number of page frames to be free during
1676 * hibernation for allocations made while saving the image and for device
1677 * drivers, in case they need to allocate memory from their hibernation
1678 * callbacks (these two numbers are given by PAGES_FOR_IO (which is a rough
1679 * estimate) and reserverd_size divided by PAGE_SIZE (which is tunable through
1680 * /sys/power/reserved_size, respectively).  To make this happen, we compute the
1681 * total number of available page frames and allocate at least
1682 *
1683 * ([page frames total] + PAGES_FOR_IO + [metadata pages]) / 2
1684 *  + 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE)
1685 *
1686 * of them, which corresponds to the maximum size of a hibernation image.
1687 *
1688 * If image_size is set below the number following from the above formula,
1689 * the preallocation of memory is continued until the total number of saveable
1690 * pages in the system is below the requested image size or the minimum
1691 * acceptable image size returned by minimum_image_size(), whichever is greater.
1692 */
1693int hibernate_preallocate_memory(void)
1694{
1695	struct zone *zone;
1696	unsigned long saveable, size, max_size, count, highmem, pages = 0;
1697	unsigned long alloc, save_highmem, pages_highmem, avail_normal;
1698	ktime_t start, stop;
1699	int error;
1700
1701	pr_info("Preallocating image memory... ");
1702	start = ktime_get();
1703
1704	error = memory_bm_create(&orig_bm, GFP_IMAGE, PG_ANY);
1705	if (error)
 
1706		goto err_out;
 
1707
1708	error = memory_bm_create(&copy_bm, GFP_IMAGE, PG_ANY);
1709	if (error)
 
1710		goto err_out;
 
1711
1712	alloc_normal = 0;
1713	alloc_highmem = 0;
1714
1715	/* Count the number of saveable data pages. */
1716	save_highmem = count_highmem_pages();
1717	saveable = count_data_pages();
1718
1719	/*
1720	 * Compute the total number of page frames we can use (count) and the
1721	 * number of pages needed for image metadata (size).
1722	 */
1723	count = saveable;
1724	saveable += save_highmem;
1725	highmem = save_highmem;
1726	size = 0;
1727	for_each_populated_zone(zone) {
1728		size += snapshot_additional_pages(zone);
1729		if (is_highmem(zone))
1730			highmem += zone_page_state(zone, NR_FREE_PAGES);
1731		else
1732			count += zone_page_state(zone, NR_FREE_PAGES);
1733	}
1734	avail_normal = count;
1735	count += highmem;
1736	count -= totalreserve_pages;
1737
1738	/* Add number of pages required for page keys (s390 only). */
1739	size += page_key_additional_pages(saveable);
1740
1741	/* Compute the maximum number of saveable pages to leave in memory. */
1742	max_size = (count - (size + PAGES_FOR_IO)) / 2
1743			- 2 * DIV_ROUND_UP(reserved_size, PAGE_SIZE);
1744	/* Compute the desired number of image pages specified by image_size. */
1745	size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1746	if (size > max_size)
1747		size = max_size;
1748	/*
1749	 * If the desired number of image pages is at least as large as the
1750	 * current number of saveable pages in memory, allocate page frames for
1751	 * the image and we're done.
1752	 */
1753	if (size >= saveable) {
1754		pages = preallocate_image_highmem(save_highmem);
1755		pages += preallocate_image_memory(saveable - pages, avail_normal);
1756		goto out;
1757	}
1758
1759	/* Estimate the minimum size of the image. */
1760	pages = minimum_image_size(saveable);
1761	/*
1762	 * To avoid excessive pressure on the normal zone, leave room in it to
1763	 * accommodate an image of the minimum size (unless it's already too
1764	 * small, in which case don't preallocate pages from it at all).
1765	 */
1766	if (avail_normal > pages)
1767		avail_normal -= pages;
1768	else
1769		avail_normal = 0;
1770	if (size < pages)
1771		size = min_t(unsigned long, pages, max_size);
1772
1773	/*
1774	 * Let the memory management subsystem know that we're going to need a
1775	 * large number of page frames to allocate and make it free some memory.
1776	 * NOTE: If this is not done, performance will be hurt badly in some
1777	 * test cases.
1778	 */
1779	shrink_all_memory(saveable - size);
1780
1781	/*
1782	 * The number of saveable pages in memory was too high, so apply some
1783	 * pressure to decrease it.  First, make room for the largest possible
1784	 * image and fail if that doesn't work.  Next, try to decrease the size
1785	 * of the image as much as indicated by 'size' using allocations from
1786	 * highmem and non-highmem zones separately.
1787	 */
1788	pages_highmem = preallocate_image_highmem(highmem / 2);
1789	alloc = count - max_size;
1790	if (alloc > pages_highmem)
1791		alloc -= pages_highmem;
1792	else
1793		alloc = 0;
1794	pages = preallocate_image_memory(alloc, avail_normal);
1795	if (pages < alloc) {
1796		/* We have exhausted non-highmem pages, try highmem. */
1797		alloc -= pages;
1798		pages += pages_highmem;
1799		pages_highmem = preallocate_image_highmem(alloc);
1800		if (pages_highmem < alloc)
 
 
1801			goto err_out;
 
1802		pages += pages_highmem;
1803		/*
1804		 * size is the desired number of saveable pages to leave in
1805		 * memory, so try to preallocate (all memory - size) pages.
1806		 */
1807		alloc = (count - pages) - size;
1808		pages += preallocate_image_highmem(alloc);
1809	} else {
1810		/*
1811		 * There are approximately max_size saveable pages at this point
1812		 * and we want to reduce this number down to size.
1813		 */
1814		alloc = max_size - size;
1815		size = preallocate_highmem_fraction(alloc, highmem, count);
1816		pages_highmem += size;
1817		alloc -= size;
1818		size = preallocate_image_memory(alloc, avail_normal);
1819		pages_highmem += preallocate_image_highmem(alloc - size);
1820		pages += pages_highmem + size;
1821	}
1822
1823	/*
1824	 * We only need as many page frames for the image as there are saveable
1825	 * pages in memory, but we have allocated more.  Release the excessive
1826	 * ones now.
1827	 */
1828	pages -= free_unnecessary_pages();
1829
1830 out:
1831	stop = ktime_get();
1832	pr_cont("done (allocated %lu pages)\n", pages);
1833	swsusp_show_speed(start, stop, pages, "Allocated");
1834
1835	return 0;
1836
1837 err_out:
1838	pr_cont("\n");
1839	swsusp_free();
1840	return -ENOMEM;
1841}
1842
1843#ifdef CONFIG_HIGHMEM
1844/**
1845 * count_pages_for_highmem - Count non-highmem pages needed for copying highmem.
1846 *
1847 * Compute the number of non-highmem pages that will be necessary for creating
1848 * copies of highmem pages.
1849 */
1850static unsigned int count_pages_for_highmem(unsigned int nr_highmem)
1851{
1852	unsigned int free_highmem = count_free_highmem_pages() + alloc_highmem;
1853
1854	if (free_highmem >= nr_highmem)
1855		nr_highmem = 0;
1856	else
1857		nr_highmem -= free_highmem;
1858
1859	return nr_highmem;
1860}
1861#else
1862static unsigned int count_pages_for_highmem(unsigned int nr_highmem) { return 0; }
1863#endif /* CONFIG_HIGHMEM */
1864
1865/**
1866 * enough_free_mem - Check if there is enough free memory for the image.
1867 */
1868static int enough_free_mem(unsigned int nr_pages, unsigned int nr_highmem)
1869{
1870	struct zone *zone;
1871	unsigned int free = alloc_normal;
1872
1873	for_each_populated_zone(zone)
1874		if (!is_highmem(zone))
1875			free += zone_page_state(zone, NR_FREE_PAGES);
1876
1877	nr_pages += count_pages_for_highmem(nr_highmem);
1878	pr_debug("Normal pages needed: %u + %u, available pages: %u\n",
1879		 nr_pages, PAGES_FOR_IO, free);
1880
1881	return free > nr_pages + PAGES_FOR_IO;
1882}
1883
1884#ifdef CONFIG_HIGHMEM
1885/**
1886 * get_highmem_buffer - Allocate a buffer for highmem pages.
1887 *
1888 * If there are some highmem pages in the hibernation image, we may need a
1889 * buffer to copy them and/or load their data.
1890 */
1891static inline int get_highmem_buffer(int safe_needed)
1892{
1893	buffer = get_image_page(GFP_ATOMIC, safe_needed);
1894	return buffer ? 0 : -ENOMEM;
1895}
1896
1897/**
1898 * alloc_highmem_image_pages - Allocate some highmem pages for the image.
1899 *
1900 * Try to allocate as many pages as needed, but if the number of free highmem
1901 * pages is less than that, allocate them all.
1902 */
1903static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1904					       unsigned int nr_highmem)
1905{
1906	unsigned int to_alloc = count_free_highmem_pages();
1907
1908	if (to_alloc > nr_highmem)
1909		to_alloc = nr_highmem;
1910
1911	nr_highmem -= to_alloc;
1912	while (to_alloc-- > 0) {
1913		struct page *page;
1914
1915		page = alloc_image_page(__GFP_HIGHMEM|__GFP_KSWAPD_RECLAIM);
1916		memory_bm_set_bit(bm, page_to_pfn(page));
1917	}
1918	return nr_highmem;
1919}
1920#else
1921static inline int get_highmem_buffer(int safe_needed) { return 0; }
1922
1923static inline unsigned int alloc_highmem_pages(struct memory_bitmap *bm,
1924					       unsigned int n) { return 0; }
1925#endif /* CONFIG_HIGHMEM */
1926
1927/**
1928 * swsusp_alloc - Allocate memory for hibernation image.
1929 *
1930 * We first try to allocate as many highmem pages as there are
1931 * saveable highmem pages in the system.  If that fails, we allocate
1932 * non-highmem pages for the copies of the remaining highmem ones.
1933 *
1934 * In this approach it is likely that the copies of highmem pages will
1935 * also be located in the high memory, because of the way in which
1936 * copy_data_pages() works.
1937 */
1938static int swsusp_alloc(struct memory_bitmap *copy_bm,
1939			unsigned int nr_pages, unsigned int nr_highmem)
1940{
1941	if (nr_highmem > 0) {
1942		if (get_highmem_buffer(PG_ANY))
1943			goto err_out;
1944		if (nr_highmem > alloc_highmem) {
1945			nr_highmem -= alloc_highmem;
1946			nr_pages += alloc_highmem_pages(copy_bm, nr_highmem);
1947		}
1948	}
1949	if (nr_pages > alloc_normal) {
1950		nr_pages -= alloc_normal;
1951		while (nr_pages-- > 0) {
1952			struct page *page;
1953
1954			page = alloc_image_page(GFP_ATOMIC);
1955			if (!page)
1956				goto err_out;
1957			memory_bm_set_bit(copy_bm, page_to_pfn(page));
1958		}
1959	}
1960
1961	return 0;
1962
1963 err_out:
1964	swsusp_free();
1965	return -ENOMEM;
1966}
1967
1968asmlinkage __visible int swsusp_save(void)
1969{
1970	unsigned int nr_pages, nr_highmem;
1971
1972	pr_info("Creating hibernation image:\n");
1973
1974	drain_local_pages(NULL);
1975	nr_pages = count_data_pages();
1976	nr_highmem = count_highmem_pages();
1977	pr_info("Need to copy %u pages\n", nr_pages + nr_highmem);
1978
1979	if (!enough_free_mem(nr_pages, nr_highmem)) {
1980		pr_err("Not enough free memory\n");
1981		return -ENOMEM;
1982	}
1983
1984	if (swsusp_alloc(&copy_bm, nr_pages, nr_highmem)) {
1985		pr_err("Memory allocation failed\n");
1986		return -ENOMEM;
1987	}
1988
1989	/*
1990	 * During allocating of suspend pagedir, new cold pages may appear.
1991	 * Kill them.
1992	 */
1993	drain_local_pages(NULL);
1994	copy_data_pages(&copy_bm, &orig_bm);
1995
1996	/*
1997	 * End of critical section. From now on, we can write to memory,
1998	 * but we should not touch disk. This specially means we must _not_
1999	 * touch swap space! Except we must write out our image of course.
2000	 */
2001
2002	nr_pages += nr_highmem;
2003	nr_copy_pages = nr_pages;
2004	nr_meta_pages = DIV_ROUND_UP(nr_pages * sizeof(long), PAGE_SIZE);
2005
2006	pr_info("Hibernation image created (%d pages copied)\n", nr_pages);
2007
2008	return 0;
2009}
2010
2011#ifndef CONFIG_ARCH_HIBERNATION_HEADER
2012static int init_header_complete(struct swsusp_info *info)
2013{
2014	memcpy(&info->uts, init_utsname(), sizeof(struct new_utsname));
2015	info->version_code = LINUX_VERSION_CODE;
2016	return 0;
2017}
2018
2019static char *check_image_kernel(struct swsusp_info *info)
2020{
2021	if (info->version_code != LINUX_VERSION_CODE)
2022		return "kernel version";
2023	if (strcmp(info->uts.sysname,init_utsname()->sysname))
2024		return "system type";
2025	if (strcmp(info->uts.release,init_utsname()->release))
2026		return "kernel release";
2027	if (strcmp(info->uts.version,init_utsname()->version))
2028		return "version";
2029	if (strcmp(info->uts.machine,init_utsname()->machine))
2030		return "machine";
2031	return NULL;
2032}
2033#endif /* CONFIG_ARCH_HIBERNATION_HEADER */
2034
2035unsigned long snapshot_get_image_size(void)
2036{
2037	return nr_copy_pages + nr_meta_pages + 1;
2038}
2039
2040static int init_header(struct swsusp_info *info)
2041{
2042	memset(info, 0, sizeof(struct swsusp_info));
2043	info->num_physpages = get_num_physpages();
2044	info->image_pages = nr_copy_pages;
2045	info->pages = snapshot_get_image_size();
2046	info->size = info->pages;
2047	info->size <<= PAGE_SHIFT;
2048	return init_header_complete(info);
2049}
2050
2051/**
2052 * pack_pfns - Prepare PFNs for saving.
2053 * @bm: Memory bitmap.
2054 * @buf: Memory buffer to store the PFNs in.
2055 *
2056 * PFNs corresponding to set bits in @bm are stored in the area of memory
2057 * pointed to by @buf (1 page at a time).
2058 */
2059static inline void pack_pfns(unsigned long *buf, struct memory_bitmap *bm)
2060{
2061	int j;
2062
2063	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2064		buf[j] = memory_bm_next_pfn(bm);
2065		if (unlikely(buf[j] == BM_END_OF_MAP))
2066			break;
2067		/* Save page key for data page (s390 only). */
2068		page_key_read(buf + j);
2069	}
2070}
2071
2072/**
2073 * snapshot_read_next - Get the address to read the next image page from.
2074 * @handle: Snapshot handle to be used for the reading.
2075 *
2076 * On the first call, @handle should point to a zeroed snapshot_handle
2077 * structure.  The structure gets populated then and a pointer to it should be
2078 * passed to this function every next time.
2079 *
2080 * On success, the function returns a positive number.  Then, the caller
2081 * is allowed to read up to the returned number of bytes from the memory
2082 * location computed by the data_of() macro.
2083 *
2084 * The function returns 0 to indicate the end of the data stream condition,
2085 * and negative numbers are returned on errors.  If that happens, the structure
2086 * pointed to by @handle is not updated and should not be used any more.
2087 */
2088int snapshot_read_next(struct snapshot_handle *handle)
2089{
2090	if (handle->cur > nr_meta_pages + nr_copy_pages)
2091		return 0;
2092
2093	if (!buffer) {
2094		/* This makes the buffer be freed by swsusp_free() */
2095		buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2096		if (!buffer)
2097			return -ENOMEM;
2098	}
2099	if (!handle->cur) {
2100		int error;
2101
2102		error = init_header((struct swsusp_info *)buffer);
2103		if (error)
2104			return error;
2105		handle->buffer = buffer;
2106		memory_bm_position_reset(&orig_bm);
2107		memory_bm_position_reset(&copy_bm);
2108	} else if (handle->cur <= nr_meta_pages) {
2109		clear_page(buffer);
2110		pack_pfns(buffer, &orig_bm);
2111	} else {
2112		struct page *page;
2113
2114		page = pfn_to_page(memory_bm_next_pfn(&copy_bm));
2115		if (PageHighMem(page)) {
2116			/*
2117			 * Highmem pages are copied to the buffer,
2118			 * because we can't return with a kmapped
2119			 * highmem page (we may not be called again).
2120			 */
2121			void *kaddr;
2122
2123			kaddr = kmap_atomic(page);
2124			copy_page(buffer, kaddr);
2125			kunmap_atomic(kaddr);
2126			handle->buffer = buffer;
2127		} else {
2128			handle->buffer = page_address(page);
2129		}
2130	}
2131	handle->cur++;
2132	return PAGE_SIZE;
2133}
2134
2135static void duplicate_memory_bitmap(struct memory_bitmap *dst,
2136				    struct memory_bitmap *src)
2137{
2138	unsigned long pfn;
2139
2140	memory_bm_position_reset(src);
2141	pfn = memory_bm_next_pfn(src);
2142	while (pfn != BM_END_OF_MAP) {
2143		memory_bm_set_bit(dst, pfn);
2144		pfn = memory_bm_next_pfn(src);
2145	}
2146}
2147
2148/**
2149 * mark_unsafe_pages - Mark pages that were used before hibernation.
2150 *
2151 * Mark the pages that cannot be used for storing the image during restoration,
2152 * because they conflict with the pages that had been used before hibernation.
2153 */
2154static void mark_unsafe_pages(struct memory_bitmap *bm)
2155{
2156	unsigned long pfn;
2157
2158	/* Clear the "free"/"unsafe" bit for all PFNs */
2159	memory_bm_position_reset(free_pages_map);
2160	pfn = memory_bm_next_pfn(free_pages_map);
2161	while (pfn != BM_END_OF_MAP) {
2162		memory_bm_clear_current(free_pages_map);
2163		pfn = memory_bm_next_pfn(free_pages_map);
2164	}
2165
2166	/* Mark pages that correspond to the "original" PFNs as "unsafe" */
2167	duplicate_memory_bitmap(free_pages_map, bm);
2168
2169	allocated_unsafe_pages = 0;
2170}
2171
2172static int check_header(struct swsusp_info *info)
2173{
2174	char *reason;
2175
2176	reason = check_image_kernel(info);
2177	if (!reason && info->num_physpages != get_num_physpages())
2178		reason = "memory size";
2179	if (reason) {
2180		pr_err("Image mismatch: %s\n", reason);
2181		return -EPERM;
2182	}
2183	return 0;
2184}
2185
2186/**
2187 * load header - Check the image header and copy the data from it.
2188 */
2189static int load_header(struct swsusp_info *info)
2190{
2191	int error;
2192
2193	restore_pblist = NULL;
2194	error = check_header(info);
2195	if (!error) {
2196		nr_copy_pages = info->image_pages;
2197		nr_meta_pages = info->pages - info->image_pages - 1;
2198	}
2199	return error;
2200}
2201
2202/**
2203 * unpack_orig_pfns - Set bits corresponding to given PFNs in a memory bitmap.
2204 * @bm: Memory bitmap.
2205 * @buf: Area of memory containing the PFNs.
2206 *
2207 * For each element of the array pointed to by @buf (1 page at a time), set the
2208 * corresponding bit in @bm.
2209 */
2210static int unpack_orig_pfns(unsigned long *buf, struct memory_bitmap *bm)
2211{
2212	int j;
2213
2214	for (j = 0; j < PAGE_SIZE / sizeof(long); j++) {
2215		if (unlikely(buf[j] == BM_END_OF_MAP))
2216			break;
2217
2218		/* Extract and buffer page key for data page (s390 only). */
2219		page_key_memorize(buf + j);
2220
2221		if (pfn_valid(buf[j]) && memory_bm_pfn_present(bm, buf[j]))
2222			memory_bm_set_bit(bm, buf[j]);
2223		else
2224			return -EFAULT;
2225	}
2226
2227	return 0;
2228}
2229
2230#ifdef CONFIG_HIGHMEM
2231/*
2232 * struct highmem_pbe is used for creating the list of highmem pages that
2233 * should be restored atomically during the resume from disk, because the page
2234 * frames they have occupied before the suspend are in use.
2235 */
2236struct highmem_pbe {
2237	struct page *copy_page;	/* data is here now */
2238	struct page *orig_page;	/* data was here before the suspend */
2239	struct highmem_pbe *next;
2240};
2241
2242/*
2243 * List of highmem PBEs needed for restoring the highmem pages that were
2244 * allocated before the suspend and included in the suspend image, but have
2245 * also been allocated by the "resume" kernel, so their contents cannot be
2246 * written directly to their "original" page frames.
2247 */
2248static struct highmem_pbe *highmem_pblist;
2249
2250/**
2251 * count_highmem_image_pages - Compute the number of highmem pages in the image.
2252 * @bm: Memory bitmap.
2253 *
2254 * The bits in @bm that correspond to image pages are assumed to be set.
2255 */
2256static unsigned int count_highmem_image_pages(struct memory_bitmap *bm)
2257{
2258	unsigned long pfn;
2259	unsigned int cnt = 0;
2260
2261	memory_bm_position_reset(bm);
2262	pfn = memory_bm_next_pfn(bm);
2263	while (pfn != BM_END_OF_MAP) {
2264		if (PageHighMem(pfn_to_page(pfn)))
2265			cnt++;
2266
2267		pfn = memory_bm_next_pfn(bm);
2268	}
2269	return cnt;
2270}
2271
2272static unsigned int safe_highmem_pages;
2273
2274static struct memory_bitmap *safe_highmem_bm;
2275
2276/**
2277 * prepare_highmem_image - Allocate memory for loading highmem data from image.
2278 * @bm: Pointer to an uninitialized memory bitmap structure.
2279 * @nr_highmem_p: Pointer to the number of highmem image pages.
2280 *
2281 * Try to allocate as many highmem pages as there are highmem image pages
2282 * (@nr_highmem_p points to the variable containing the number of highmem image
2283 * pages).  The pages that are "safe" (ie. will not be overwritten when the
2284 * hibernation image is restored entirely) have the corresponding bits set in
2285 * @bm (it must be unitialized).
2286 *
2287 * NOTE: This function should not be called if there are no highmem image pages.
2288 */
2289static int prepare_highmem_image(struct memory_bitmap *bm,
2290				 unsigned int *nr_highmem_p)
2291{
2292	unsigned int to_alloc;
2293
2294	if (memory_bm_create(bm, GFP_ATOMIC, PG_SAFE))
2295		return -ENOMEM;
2296
2297	if (get_highmem_buffer(PG_SAFE))
2298		return -ENOMEM;
2299
2300	to_alloc = count_free_highmem_pages();
2301	if (to_alloc > *nr_highmem_p)
2302		to_alloc = *nr_highmem_p;
2303	else
2304		*nr_highmem_p = to_alloc;
2305
2306	safe_highmem_pages = 0;
2307	while (to_alloc-- > 0) {
2308		struct page *page;
2309
2310		page = alloc_page(__GFP_HIGHMEM);
2311		if (!swsusp_page_is_free(page)) {
2312			/* The page is "safe", set its bit the bitmap */
2313			memory_bm_set_bit(bm, page_to_pfn(page));
2314			safe_highmem_pages++;
2315		}
2316		/* Mark the page as allocated */
2317		swsusp_set_page_forbidden(page);
2318		swsusp_set_page_free(page);
2319	}
2320	memory_bm_position_reset(bm);
2321	safe_highmem_bm = bm;
2322	return 0;
2323}
2324
2325static struct page *last_highmem_page;
2326
2327/**
2328 * get_highmem_page_buffer - Prepare a buffer to store a highmem image page.
2329 *
2330 * For a given highmem image page get a buffer that suspend_write_next() should
2331 * return to its caller to write to.
2332 *
2333 * If the page is to be saved to its "original" page frame or a copy of
2334 * the page is to be made in the highmem, @buffer is returned.  Otherwise,
2335 * the copy of the page is to be made in normal memory, so the address of
2336 * the copy is returned.
2337 *
2338 * If @buffer is returned, the caller of suspend_write_next() will write
2339 * the page's contents to @buffer, so they will have to be copied to the
2340 * right location on the next call to suspend_write_next() and it is done
2341 * with the help of copy_last_highmem_page().  For this purpose, if
2342 * @buffer is returned, @last_highmem_page is set to the page to which
2343 * the data will have to be copied from @buffer.
2344 */
2345static void *get_highmem_page_buffer(struct page *page,
2346				     struct chain_allocator *ca)
2347{
2348	struct highmem_pbe *pbe;
2349	void *kaddr;
2350
2351	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page)) {
2352		/*
2353		 * We have allocated the "original" page frame and we can
2354		 * use it directly to store the loaded page.
2355		 */
2356		last_highmem_page = page;
2357		return buffer;
2358	}
2359	/*
2360	 * The "original" page frame has not been allocated and we have to
2361	 * use a "safe" page frame to store the loaded page.
2362	 */
2363	pbe = chain_alloc(ca, sizeof(struct highmem_pbe));
2364	if (!pbe) {
2365		swsusp_free();
2366		return ERR_PTR(-ENOMEM);
2367	}
2368	pbe->orig_page = page;
2369	if (safe_highmem_pages > 0) {
2370		struct page *tmp;
2371
2372		/* Copy of the page will be stored in high memory */
2373		kaddr = buffer;
2374		tmp = pfn_to_page(memory_bm_next_pfn(safe_highmem_bm));
2375		safe_highmem_pages--;
2376		last_highmem_page = tmp;
2377		pbe->copy_page = tmp;
2378	} else {
2379		/* Copy of the page will be stored in normal memory */
2380		kaddr = safe_pages_list;
2381		safe_pages_list = safe_pages_list->next;
2382		pbe->copy_page = virt_to_page(kaddr);
2383	}
2384	pbe->next = highmem_pblist;
2385	highmem_pblist = pbe;
2386	return kaddr;
2387}
2388
2389/**
2390 * copy_last_highmem_page - Copy most the most recent highmem image page.
2391 *
2392 * Copy the contents of a highmem image from @buffer, where the caller of
2393 * snapshot_write_next() has stored them, to the right location represented by
2394 * @last_highmem_page .
2395 */
2396static void copy_last_highmem_page(void)
2397{
2398	if (last_highmem_page) {
2399		void *dst;
2400
2401		dst = kmap_atomic(last_highmem_page);
2402		copy_page(dst, buffer);
2403		kunmap_atomic(dst);
2404		last_highmem_page = NULL;
2405	}
2406}
2407
2408static inline int last_highmem_page_copied(void)
2409{
2410	return !last_highmem_page;
2411}
2412
2413static inline void free_highmem_data(void)
2414{
2415	if (safe_highmem_bm)
2416		memory_bm_free(safe_highmem_bm, PG_UNSAFE_CLEAR);
2417
2418	if (buffer)
2419		free_image_page(buffer, PG_UNSAFE_CLEAR);
2420}
2421#else
2422static unsigned int count_highmem_image_pages(struct memory_bitmap *bm) { return 0; }
2423
2424static inline int prepare_highmem_image(struct memory_bitmap *bm,
2425					unsigned int *nr_highmem_p) { return 0; }
2426
2427static inline void *get_highmem_page_buffer(struct page *page,
2428					    struct chain_allocator *ca)
2429{
2430	return ERR_PTR(-EINVAL);
2431}
2432
2433static inline void copy_last_highmem_page(void) {}
2434static inline int last_highmem_page_copied(void) { return 1; }
2435static inline void free_highmem_data(void) {}
2436#endif /* CONFIG_HIGHMEM */
2437
2438#define PBES_PER_LINKED_PAGE	(LINKED_PAGE_DATA_SIZE / sizeof(struct pbe))
2439
2440/**
2441 * prepare_image - Make room for loading hibernation image.
2442 * @new_bm: Unitialized memory bitmap structure.
2443 * @bm: Memory bitmap with unsafe pages marked.
2444 *
2445 * Use @bm to mark the pages that will be overwritten in the process of
2446 * restoring the system memory state from the suspend image ("unsafe" pages)
2447 * and allocate memory for the image.
2448 *
2449 * The idea is to allocate a new memory bitmap first and then allocate
2450 * as many pages as needed for image data, but without specifying what those
2451 * pages will be used for just yet.  Instead, we mark them all as allocated and
2452 * create a lists of "safe" pages to be used later.  On systems with high
2453 * memory a list of "safe" highmem pages is created too.
2454 */
2455static int prepare_image(struct memory_bitmap *new_bm, struct memory_bitmap *bm)
2456{
2457	unsigned int nr_pages, nr_highmem;
2458	struct linked_page *lp;
2459	int error;
2460
2461	/* If there is no highmem, the buffer will not be necessary */
2462	free_image_page(buffer, PG_UNSAFE_CLEAR);
2463	buffer = NULL;
2464
2465	nr_highmem = count_highmem_image_pages(bm);
2466	mark_unsafe_pages(bm);
2467
2468	error = memory_bm_create(new_bm, GFP_ATOMIC, PG_SAFE);
2469	if (error)
2470		goto Free;
2471
2472	duplicate_memory_bitmap(new_bm, bm);
2473	memory_bm_free(bm, PG_UNSAFE_KEEP);
2474	if (nr_highmem > 0) {
2475		error = prepare_highmem_image(bm, &nr_highmem);
2476		if (error)
2477			goto Free;
2478	}
2479	/*
2480	 * Reserve some safe pages for potential later use.
2481	 *
2482	 * NOTE: This way we make sure there will be enough safe pages for the
2483	 * chain_alloc() in get_buffer().  It is a bit wasteful, but
2484	 * nr_copy_pages cannot be greater than 50% of the memory anyway.
2485	 *
2486	 * nr_copy_pages cannot be less than allocated_unsafe_pages too.
2487	 */
2488	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2489	nr_pages = DIV_ROUND_UP(nr_pages, PBES_PER_LINKED_PAGE);
2490	while (nr_pages > 0) {
2491		lp = get_image_page(GFP_ATOMIC, PG_SAFE);
2492		if (!lp) {
2493			error = -ENOMEM;
2494			goto Free;
2495		}
2496		lp->next = safe_pages_list;
2497		safe_pages_list = lp;
2498		nr_pages--;
2499	}
2500	/* Preallocate memory for the image */
2501	nr_pages = nr_copy_pages - nr_highmem - allocated_unsafe_pages;
2502	while (nr_pages > 0) {
2503		lp = (struct linked_page *)get_zeroed_page(GFP_ATOMIC);
2504		if (!lp) {
2505			error = -ENOMEM;
2506			goto Free;
2507		}
2508		if (!swsusp_page_is_free(virt_to_page(lp))) {
2509			/* The page is "safe", add it to the list */
2510			lp->next = safe_pages_list;
2511			safe_pages_list = lp;
2512		}
2513		/* Mark the page as allocated */
2514		swsusp_set_page_forbidden(virt_to_page(lp));
2515		swsusp_set_page_free(virt_to_page(lp));
2516		nr_pages--;
2517	}
2518	return 0;
2519
2520 Free:
2521	swsusp_free();
2522	return error;
2523}
2524
2525/**
2526 * get_buffer - Get the address to store the next image data page.
2527 *
2528 * Get the address that snapshot_write_next() should return to its caller to
2529 * write to.
2530 */
2531static void *get_buffer(struct memory_bitmap *bm, struct chain_allocator *ca)
2532{
2533	struct pbe *pbe;
2534	struct page *page;
2535	unsigned long pfn = memory_bm_next_pfn(bm);
2536
2537	if (pfn == BM_END_OF_MAP)
2538		return ERR_PTR(-EFAULT);
2539
2540	page = pfn_to_page(pfn);
2541	if (PageHighMem(page))
2542		return get_highmem_page_buffer(page, ca);
2543
2544	if (swsusp_page_is_forbidden(page) && swsusp_page_is_free(page))
2545		/*
2546		 * We have allocated the "original" page frame and we can
2547		 * use it directly to store the loaded page.
2548		 */
2549		return page_address(page);
2550
2551	/*
2552	 * The "original" page frame has not been allocated and we have to
2553	 * use a "safe" page frame to store the loaded page.
2554	 */
2555	pbe = chain_alloc(ca, sizeof(struct pbe));
2556	if (!pbe) {
2557		swsusp_free();
2558		return ERR_PTR(-ENOMEM);
2559	}
2560	pbe->orig_address = page_address(page);
2561	pbe->address = safe_pages_list;
2562	safe_pages_list = safe_pages_list->next;
2563	pbe->next = restore_pblist;
2564	restore_pblist = pbe;
2565	return pbe->address;
2566}
2567
2568/**
2569 * snapshot_write_next - Get the address to store the next image page.
2570 * @handle: Snapshot handle structure to guide the writing.
2571 *
2572 * On the first call, @handle should point to a zeroed snapshot_handle
2573 * structure.  The structure gets populated then and a pointer to it should be
2574 * passed to this function every next time.
2575 *
2576 * On success, the function returns a positive number.  Then, the caller
2577 * is allowed to write up to the returned number of bytes to the memory
2578 * location computed by the data_of() macro.
2579 *
2580 * The function returns 0 to indicate the "end of file" condition.  Negative
2581 * numbers are returned on errors, in which cases the structure pointed to by
2582 * @handle is not updated and should not be used any more.
2583 */
2584int snapshot_write_next(struct snapshot_handle *handle)
2585{
2586	static struct chain_allocator ca;
2587	int error = 0;
2588
2589	/* Check if we have already loaded the entire image */
2590	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages)
2591		return 0;
2592
2593	handle->sync_read = 1;
2594
2595	if (!handle->cur) {
2596		if (!buffer)
2597			/* This makes the buffer be freed by swsusp_free() */
2598			buffer = get_image_page(GFP_ATOMIC, PG_ANY);
2599
2600		if (!buffer)
2601			return -ENOMEM;
2602
2603		handle->buffer = buffer;
2604	} else if (handle->cur == 1) {
2605		error = load_header(buffer);
2606		if (error)
2607			return error;
2608
2609		safe_pages_list = NULL;
2610
2611		error = memory_bm_create(&copy_bm, GFP_ATOMIC, PG_ANY);
2612		if (error)
2613			return error;
2614
2615		/* Allocate buffer for page keys. */
2616		error = page_key_alloc(nr_copy_pages);
2617		if (error)
2618			return error;
2619
2620		hibernate_restore_protection_begin();
2621	} else if (handle->cur <= nr_meta_pages + 1) {
2622		error = unpack_orig_pfns(buffer, &copy_bm);
2623		if (error)
2624			return error;
2625
2626		if (handle->cur == nr_meta_pages + 1) {
2627			error = prepare_image(&orig_bm, &copy_bm);
2628			if (error)
2629				return error;
2630
2631			chain_init(&ca, GFP_ATOMIC, PG_SAFE);
2632			memory_bm_position_reset(&orig_bm);
2633			restore_pblist = NULL;
2634			handle->buffer = get_buffer(&orig_bm, &ca);
2635			handle->sync_read = 0;
2636			if (IS_ERR(handle->buffer))
2637				return PTR_ERR(handle->buffer);
2638		}
2639	} else {
2640		copy_last_highmem_page();
2641		/* Restore page key for data page (s390 only). */
2642		page_key_write(handle->buffer);
2643		hibernate_restore_protect_page(handle->buffer);
2644		handle->buffer = get_buffer(&orig_bm, &ca);
2645		if (IS_ERR(handle->buffer))
2646			return PTR_ERR(handle->buffer);
2647		if (handle->buffer != buffer)
2648			handle->sync_read = 0;
2649	}
2650	handle->cur++;
2651	return PAGE_SIZE;
2652}
2653
2654/**
2655 * snapshot_write_finalize - Complete the loading of a hibernation image.
2656 *
2657 * Must be called after the last call to snapshot_write_next() in case the last
2658 * page in the image happens to be a highmem page and its contents should be
2659 * stored in highmem.  Additionally, it recycles bitmap memory that's not
2660 * necessary any more.
2661 */
2662void snapshot_write_finalize(struct snapshot_handle *handle)
2663{
2664	copy_last_highmem_page();
2665	/* Restore page key for data page (s390 only). */
2666	page_key_write(handle->buffer);
2667	page_key_free();
2668	hibernate_restore_protect_page(handle->buffer);
2669	/* Do that only if we have loaded the image entirely */
2670	if (handle->cur > 1 && handle->cur > nr_meta_pages + nr_copy_pages) {
2671		memory_bm_recycle(&orig_bm);
2672		free_highmem_data();
2673	}
2674}
2675
2676int snapshot_image_loaded(struct snapshot_handle *handle)
2677{
2678	return !(!nr_copy_pages || !last_highmem_page_copied() ||
2679			handle->cur <= nr_meta_pages + nr_copy_pages);
2680}
2681
2682#ifdef CONFIG_HIGHMEM
2683/* Assumes that @buf is ready and points to a "safe" page */
2684static inline void swap_two_pages_data(struct page *p1, struct page *p2,
2685				       void *buf)
2686{
2687	void *kaddr1, *kaddr2;
2688
2689	kaddr1 = kmap_atomic(p1);
2690	kaddr2 = kmap_atomic(p2);
2691	copy_page(buf, kaddr1);
2692	copy_page(kaddr1, kaddr2);
2693	copy_page(kaddr2, buf);
2694	kunmap_atomic(kaddr2);
2695	kunmap_atomic(kaddr1);
2696}
2697
2698/**
2699 * restore_highmem - Put highmem image pages into their original locations.
2700 *
2701 * For each highmem page that was in use before hibernation and is included in
2702 * the image, and also has been allocated by the "restore" kernel, swap its
2703 * current contents with the previous (ie. "before hibernation") ones.
2704 *
2705 * If the restore eventually fails, we can call this function once again and
2706 * restore the highmem state as seen by the restore kernel.
2707 */
2708int restore_highmem(void)
2709{
2710	struct highmem_pbe *pbe = highmem_pblist;
2711	void *buf;
2712
2713	if (!pbe)
2714		return 0;
2715
2716	buf = get_image_page(GFP_ATOMIC, PG_SAFE);
2717	if (!buf)
2718		return -ENOMEM;
2719
2720	while (pbe) {
2721		swap_two_pages_data(pbe->copy_page, pbe->orig_page, buf);
2722		pbe = pbe->next;
2723	}
2724	free_image_page(buf, PG_UNSAFE_CLEAR);
2725	return 0;
2726}
2727#endif /* CONFIG_HIGHMEM */