kernel/power/snapshot.c

   1 /*
   2  * linux/kernel/power/snapshot.c
   3  *
   4  * This file provide system snapshot/restore functionality.
   5  *
   6  * Copyright (C) 1998-2005 Pavel Machek <pavel@suse.cz>
   7  *
   8  * This file is released under the GPLv2, and is based on swsusp.c.
   9  *
  10  */
  11
  12
  13 #include <linux/version.h>
  14 #include <linux/module.h>
  15 #include <linux/mm.h>
  16 #include <linux/suspend.h>
  17 #include <linux/smp_lock.h>
  18 #include <linux/delay.h>
  19 #include <linux/bitops.h>
  20 #include <linux/spinlock.h>
  21 #include <linux/kernel.h>
  22 #include <linux/pm.h>
  23 #include <linux/device.h>
  24 #include <linux/bootmem.h>
  25 #include <linux/syscalls.h>
  26 #include <linux/console.h>
  27 #include <linux/highmem.h>
  28
  29 #include <asm/uaccess.h>
  30 #include <asm/mmu_context.h>
  31 #include <asm/pgtable.h>
  32 #include <asm/tlbflush.h>
  33 #include <asm/io.h>
  34
  35 #include "power.h"
  36
  37 struct pbe *pagedir_nosave;
  38 static unsigned int nr_copy_pages;
  39 static unsigned int nr_meta_pages;
  40 static unsigned long *buffer;
  41
  42 #ifdef CONFIG_HIGHMEM
  43 unsigned int count_highmem_pages(void)
  44 {
  45         struct zone *zone;
  46         unsigned long zone_pfn;
  47         unsigned int n = 0;
  48
  49         for_each_zone (zone)
  50                 if (is_highmem(zone)) {
  51                         mark_free_pages(zone);
  52                         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; zone_pfn++) {
  53                                 struct page *page;
  54                                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
  55                                 if (!pfn_valid(pfn))
  56                                         continue;
  57                                 page = pfn_to_page(pfn);
  58                                 if (PageReserved(page))
  59                                         continue;
  60                                 if (PageNosaveFree(page))
  61                                         continue;
  62                                 n++;
  63                         }
  64                 }
  65         return n;
  66 }
  67
  68 struct highmem_page {
  69         char *data;
  70         struct page *page;
  71         struct highmem_page *next;
  72 };
  73
  74 static struct highmem_page *highmem_copy;
  75
  76 static int save_highmem_zone(struct zone *zone)
  77 {
  78         unsigned long zone_pfn;
  79         mark_free_pages(zone);
  80         for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
  81                 struct page *page;
  82                 struct highmem_page *save;
  83                 void *kaddr;
  84                 unsigned long pfn = zone_pfn + zone->zone_start_pfn;
  85
  86                 if (!(pfn%10000))
  87                         printk(".");
  88                 if (!pfn_valid(pfn))
  89                         continue;
  90                 page = pfn_to_page(pfn);
  91                 /*
  92                  * This condition results from rvmalloc() sans vmalloc_32()
  93                  * and architectural memory reservations. This should be
  94                  * corrected eventually when the cases giving rise to this
  95                  * are better understood.
  96                  */
  97                 if (PageReserved(page))
  98                         continue;
  99                 BUG_ON(PageNosave(page));
 100                 if (PageNosaveFree(page))
 101                         continue;
 102                 save = kmalloc(sizeof(struct highmem_page), GFP_ATOMIC);
 103                 if (!save)
 104                         return -ENOMEM;
 105                 save->next = highmem_copy;
 106                 save->page = page;
 107                 save->data = (void *) get_zeroed_page(GFP_ATOMIC);
 108                 if (!save->data) {
 109                         kfree(save);
 110                         return -ENOMEM;
 111                 }
 112                 kaddr = kmap_atomic(page, KM_USER0);
 113                 memcpy(save->data, kaddr, PAGE_SIZE);
 114                 kunmap_atomic(kaddr, KM_USER0);
 115                 highmem_copy = save;
 116         }
 117         return 0;
 118 }
 119
 120 int save_highmem(void)
 121 {
 122         struct zone *zone;
 123         int res = 0;
 124
 125         pr_debug("swsusp: Saving Highmem");
 126         drain_local_pages();
 127         for_each_zone (zone) {
 128                 if (is_highmem(zone))
 129                         res = save_highmem_zone(zone);
 130                 if (res)
 131                         return res;
 132         }
 133         printk("\n");
 134         return 0;
 135 }
 136
 137 int restore_highmem(void)
 138 {
 139         printk("swsusp: Restoring Highmem\n");
 140         while (highmem_copy) {
 141                 struct highmem_page *save = highmem_copy;
 142                 void *kaddr;
 143                 highmem_copy = save->next;
 144
 145                 kaddr = kmap_atomic(save->page, KM_USER0);
 146                 memcpy(kaddr, save->data, PAGE_SIZE);
 147                 kunmap_atomic(kaddr, KM_USER0);
 148                 free_page((long) save->data);
 149                 kfree(save);
 150         }
 151         return 0;
 152 }
 153 #else
 154 static inline unsigned int count_highmem_pages(void) {return 0;}
 155 static inline int save_highmem(void) {return 0;}
 156 static inline int restore_highmem(void) {return 0;}
 157 #endif
 158
 159 static int pfn_is_nosave(unsigned long pfn)
 160 {
 161         unsigned long nosave_begin_pfn = __pa(&__nosave_begin) >> PAGE_SHIFT;
 162         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa(&__nosave_end)) >> PAGE_SHIFT;
 163         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 164 }
 165
 166 /**
 167  *      saveable - Determine whether a page should be cloned or not.
 168  *      @pfn:   The page
 169  *
 170  *      We save a page if it's Reserved, and not in the range of pages
 171  *      statically defined as 'unsaveable', or if it isn't reserved, and
 172  *      isn't part of a free chunk of pages.
 173  */
 174
 175 static int saveable(struct zone *zone, unsigned long *zone_pfn)
 176 {
 177         unsigned long pfn = *zone_pfn + zone->zone_start_pfn;
 178         struct page *page;
 179
 180         if (!pfn_valid(pfn))
 181                 return 0;
 182
 183         page = pfn_to_page(pfn);
 184         BUG_ON(PageReserved(page) && PageNosave(page));
 185         if (PageNosave(page))
 186                 return 0;
 187         if (PageReserved(page) && pfn_is_nosave(pfn))
 188                 return 0;
 189         if (PageNosaveFree(page))
 190                 return 0;
 191
 192         return 1;
 193 }
 194
 195 unsigned int count_data_pages(void)
 196 {
 197         struct zone *zone;
 198         unsigned long zone_pfn;
 199         unsigned int n = 0;
 200
 201         for_each_zone (zone) {
 202                 if (is_highmem(zone))
 203                         continue;
 204                 mark_free_pages(zone);
 205                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 206                         n += saveable(zone, &zone_pfn);
 207         }
 208         return n;
 209 }
 210
 211 static void copy_data_pages(struct pbe *pblist)
 212 {
 213         struct zone *zone;
 214         unsigned long zone_pfn;
 215         struct pbe *pbe, *p;
 216
 217         pbe = pblist;
 218         for_each_zone (zone) {
 219                 if (is_highmem(zone))
 220                         continue;
 221                 mark_free_pages(zone);
 222                 /* This is necessary for swsusp_free() */
 223                 for_each_pb_page (p, pblist)
 224                         SetPageNosaveFree(virt_to_page(p));
 225                 for_each_pbe (p, pblist)
 226                         SetPageNosaveFree(virt_to_page(p->address));
 227                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) {
 228                         if (saveable(zone, &zone_pfn)) {
 229                                 struct page *page;
 230                                 long *src, *dst;
 231                                 int n;
 232
 233                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 234                                 BUG_ON(!pbe);
 235                                 pbe->orig_address = (unsigned long)page_address(page);
 236                                 /* copy_page and memcpy are not usable for copying task structs. */
 237                                 dst = (long *)pbe->address;
 238                                 src = (long *)pbe->orig_address;
 239                                 for (n = PAGE_SIZE / sizeof(long); n; n--)
 240                                         *dst++ = *src++;
 241                                 pbe = pbe->next;
 242                         }
 243                 }
 244         }
 245         BUG_ON(pbe);
 246 }
 247
 248
 249 /**
 250  *      free_pagedir - free pages allocated with alloc_pagedir()
 251  */
 252
 253 static void free_pagedir(struct pbe *pblist, int clear_nosave_free)
 254 {
 255         struct pbe *pbe;
 256
 257         while (pblist) {
 258                 pbe = (pblist + PB_PAGE_SKIP)->next;
 259                 ClearPageNosave(virt_to_page(pblist));
 260                 if (clear_nosave_free)
 261                         ClearPageNosaveFree(virt_to_page(pblist));
 262                 free_page((unsigned long)pblist);
 263                 pblist = pbe;
 264         }
 265 }
 266
 267 /**
 268  *      fill_pb_page - Create a list of PBEs on a given memory page
 269  */
 270
 271 static inline void fill_pb_page(struct pbe *pbpage)
 272 {
 273         struct pbe *p;
 274
 275         p = pbpage;
 276         pbpage += PB_PAGE_SKIP;
 277         do
 278                 p->next = p + 1;
 279         while (++p < pbpage);
 280 }
 281
 282 /**
 283  *      create_pbe_list - Create a list of PBEs on top of a given chain
 284  *      of memory pages allocated with alloc_pagedir()
 285  */
 286
 287 static inline void create_pbe_list(struct pbe *pblist, unsigned int nr_pages)
 288 {
 289         struct pbe *pbpage, *p;
 290         unsigned int num = PBES_PER_PAGE;
 291
 292         for_each_pb_page (pbpage, pblist) {
 293                 if (num >= nr_pages)
 294                         break;
 295
 296                 fill_pb_page(pbpage);
 297                 num += PBES_PER_PAGE;
 298         }
 299         if (pbpage) {
 300                 for (num -= PBES_PER_PAGE - 1, p = pbpage; num < nr_pages; p++, num++)
 301                         p->next = p + 1;
 302                 p->next = NULL;
 303         }
 304 }
 305
 306 static unsigned int unsafe_pages;
 307
 308 /**
 309  *      @safe_needed - on resume, for storing the PBE list and the image,
 310  *      we can only use memory pages that do not conflict with the pages
 311  *      used before suspend.
 312  *
 313  *      The unsafe pages are marked with the PG_nosave_free flag
 314  *      and we count them using unsafe_pages
 315  */
 316
 317 static inline void *alloc_image_page(gfp_t gfp_mask, int safe_needed)
 318 {
 319         void *res;
 320
 321         res = (void *)get_zeroed_page(gfp_mask);
 322         if (safe_needed)
 323                 while (res && PageNosaveFree(virt_to_page(res))) {
 324                         /* The page is unsafe, mark it for swsusp_free() */
 325                         SetPageNosave(virt_to_page(res));
 326                         unsafe_pages++;
 327                         res = (void *)get_zeroed_page(gfp_mask);
 328                 }
 329         if (res) {
 330                 SetPageNosave(virt_to_page(res));
 331                 SetPageNosaveFree(virt_to_page(res));
 332         }
 333         return res;
 334 }
 335
 336 unsigned long get_safe_page(gfp_t gfp_mask)
 337 {
 338         return (unsigned long)alloc_image_page(gfp_mask, 1);
 339 }
 340
 341 /**
 342  *      alloc_pagedir - Allocate the page directory.
 343  *
 344  *      First, determine exactly how many pages we need and
 345  *      allocate them.
 346  *
 347  *      We arrange the pages in a chain: each page is an array of PBES_PER_PAGE
 348  *      struct pbe elements (pbes) and the last element in the page points
 349  *      to the next page.
 350  *
 351  *      On each page we set up a list of struct_pbe elements.
 352  */
 353
 354 static struct pbe *alloc_pagedir(unsigned int nr_pages, gfp_t gfp_mask,
 355                                  int safe_needed)
 356 {
 357         unsigned int num;
 358         struct pbe *pblist, *pbe;
 359
 360         if (!nr_pages)
 361                 return NULL;
 362
 363         pblist = alloc_image_page(gfp_mask, safe_needed);
 364         /* FIXME: rewrite this ugly loop */
 365         for (pbe = pblist, num = PBES_PER_PAGE; pbe && num < nr_pages;
 366                         pbe = pbe->next, num += PBES_PER_PAGE) {
 367                 pbe += PB_PAGE_SKIP;
 368                 pbe->next = alloc_image_page(gfp_mask, safe_needed);
 369         }
 370         if (!pbe) { /* get_zeroed_page() failed */
 371                 free_pagedir(pblist, 1);
 372                 pblist = NULL;
 373         } else
 374                 create_pbe_list(pblist, nr_pages);
 375         return pblist;
 376 }
 377
 378 /**
 379  * Free pages we allocated for suspend. Suspend pages are alocated
 380  * before atomic copy, so we need to free them after resume.
 381  */
 382
 383 void swsusp_free(void)
 384 {
 385         struct zone *zone;
 386         unsigned long zone_pfn;
 387
 388         for_each_zone(zone) {
 389                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 390                         if (pfn_valid(zone_pfn + zone->zone_start_pfn)) {
 391                                 struct page *page;
 392                                 page = pfn_to_page(zone_pfn + zone->zone_start_pfn);
 393                                 if (PageNosave(page) && PageNosaveFree(page)) {
 394                                         ClearPageNosave(page);
 395                                         ClearPageNosaveFree(page);
 396                                         free_page((long) page_address(page));
 397                                 }
 398                         }
 399         }
 400         nr_copy_pages = 0;
 401         nr_meta_pages = 0;
 402         pagedir_nosave = NULL;
 403         buffer = NULL;
 404 }
 405
 406
 407 /**
 408  *      enough_free_mem - Make sure we enough free memory to snapshot.
 409  *
 410  *      Returns TRUE or FALSE after checking the number of available
 411  *      free pages.
 412  */
 413
 414 static int enough_free_mem(unsigned int nr_pages)
 415 {
 416         struct zone *zone;
 417         unsigned int n = 0;
 418
 419         for_each_zone (zone)
 420                 if (!is_highmem(zone))
 421                         n += zone->free_pages;
 422         pr_debug("swsusp: available memory: %u pages\n", n);
 423         return n > (nr_pages + PAGES_FOR_IO +
 424                 (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE);
 425 }
 426
 427 static int alloc_data_pages(struct pbe *pblist, gfp_t gfp_mask, int safe_needed)
 428 {
 429         struct pbe *p;
 430
 431         for_each_pbe (p, pblist) {
 432                 p->address = (unsigned long)alloc_image_page(gfp_mask, safe_needed);
 433                 if (!p->address)
 434                         return -ENOMEM;
 435         }
 436         return 0;
 437 }
 438
 439 static struct pbe *swsusp_alloc(unsigned int nr_pages)
 440 {
 441         struct pbe *pblist;
 442
 443         if (!(pblist = alloc_pagedir(nr_pages, GFP_ATOMIC | __GFP_COLD, 0))) {
 444                 printk(KERN_ERR "suspend: Allocating pagedir failed.\n");
 445                 return NULL;
 446         }
 447
 448         if (alloc_data_pages(pblist, GFP_ATOMIC | __GFP_COLD, 0)) {
 449                 printk(KERN_ERR "suspend: Allocating image pages failed.\n");
 450                 swsusp_free();
 451                 return NULL;
 452         }
 453
 454         return pblist;
 455 }
 456
 457 asmlinkage int swsusp_save(void)
 458 {
 459         unsigned int nr_pages;
 460
 461         pr_debug("swsusp: critical section: \n");
 462
 463         drain_local_pages();
 464         nr_pages = count_data_pages();
 465         printk("swsusp: Need to copy %u pages\n", nr_pages);
 466
 467         pr_debug("swsusp: pages needed: %u + %lu + %u, free: %u\n",
 468                  nr_pages,
 469                  (nr_pages + PBES_PER_PAGE - 1) / PBES_PER_PAGE,
 470                  PAGES_FOR_IO, nr_free_pages());
 471
 472         if (!enough_free_mem(nr_pages)) {
 473                 printk(KERN_ERR "swsusp: Not enough free memory\n");
 474                 return -ENOMEM;
 475         }
 476
 477         pagedir_nosave = swsusp_alloc(nr_pages);
 478         if (!pagedir_nosave)
 479                 return -ENOMEM;
 480
 481         /* During allocating of suspend pagedir, new cold pages may appear.
 482          * Kill them.
 483          */
 484         drain_local_pages();
 485         copy_data_pages(pagedir_nosave);
 486
 487         /*
 488          * End of critical section. From now on, we can write to memory,
 489          * but we should not touch disk. This specially means we must _not_
 490          * touch swap space! Except we must write out our image of course.
 491          */
 492
 493         nr_copy_pages = nr_pages;
 494         nr_meta_pages = (nr_pages * sizeof(long) + PAGE_SIZE - 1) >> PAGE_SHIFT;
 495
 496         printk("swsusp: critical section/: done (%d pages copied)\n", nr_pages);
 497         return 0;
 498 }
 499
 500 static void init_header(struct swsusp_info *info)
 501 {
 502         memset(info, 0, sizeof(struct swsusp_info));
 503         info->version_code = LINUX_VERSION_CODE;
 504         info->num_physpages = num_physpages;
 505         memcpy(&info->uts, &system_utsname, sizeof(system_utsname));
 506         info->cpus = num_online_cpus();
 507         info->image_pages = nr_copy_pages;
 508         info->pages = nr_copy_pages + nr_meta_pages + 1;
 509         info->size = info->pages;
 510         info->size <<= PAGE_SHIFT;
 511 }
 512
 513 /**
 514  *      pack_orig_addresses - the .orig_address fields of the PBEs from the
 515  *      list starting at @pbe are stored in the array @buf[] (1 page)
 516  */
 517
 518 static inline struct pbe *pack_orig_addresses(unsigned long *buf, struct pbe *pbe)
 519 {
 520         int j;
 521
 522         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 523                 buf[j] = pbe->orig_address;
 524                 pbe = pbe->next;
 525         }
 526         if (!pbe)
 527                 for (; j < PAGE_SIZE / sizeof(long); j++)
 528                         buf[j] = 0;
 529         return pbe;
 530 }
 531
 532 /**
 533  *      snapshot_read_next - used for reading the system memory snapshot.
 534  *
 535  *      On the first call to it @handle should point to a zeroed
 536  *      snapshot_handle structure.  The structure gets updated and a pointer
 537  *      to it should be passed to this function every next time.
 538  *
 539  *      The @count parameter should contain the number of bytes the caller
 540  *      wants to read from the snapshot.  It must not be zero.
 541  *
 542  *      On success the function returns a positive number.  Then, the caller
 543  *      is allowed to read up to the returned number of bytes from the memory
 544  *      location computed by the data_of() macro.  The number returned
 545  *      may be smaller than @count, but this only happens if the read would
 546  *      cross a page boundary otherwise.
 547  *
 548  *      The function returns 0 to indicate the end of data stream condition,
 549  *      and a negative number is returned on error.  In such cases the
 550  *      structure pointed to by @handle is not updated and should not be used
 551  *      any more.
 552  */
 553
 554 int snapshot_read_next(struct snapshot_handle *handle, size_t count)
 555 {
 556         if (handle->page > nr_meta_pages + nr_copy_pages)
 557                 return 0;
 558         if (!buffer) {
 559                 /* This makes the buffer be freed by swsusp_free() */
 560                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 561                 if (!buffer)
 562                         return -ENOMEM;
 563         }
 564         if (!handle->offset) {
 565                 init_header((struct swsusp_info *)buffer);
 566                 handle->buffer = buffer;
 567                 handle->pbe = pagedir_nosave;
 568         }
 569         if (handle->prev < handle->page) {
 570                 if (handle->page <= nr_meta_pages) {
 571                         handle->pbe = pack_orig_addresses(buffer, handle->pbe);
 572                         if (!handle->pbe)
 573                                 handle->pbe = pagedir_nosave;
 574                 } else {
 575                         handle->buffer = (void *)handle->pbe->address;
 576                         handle->pbe = handle->pbe->next;
 577                 }
 578                 handle->prev = handle->page;
 579         }
 580         handle->buf_offset = handle->page_offset;
 581         if (handle->page_offset + count >= PAGE_SIZE) {
 582                 count = PAGE_SIZE - handle->page_offset;
 583                 handle->page_offset = 0;
 584                 handle->page++;
 585         } else {
 586                 handle->page_offset += count;
 587         }
 588         handle->offset += count;
 589         return count;
 590 }
 591
 592 /**
 593  *      mark_unsafe_pages - mark the pages that cannot be used for storing
 594  *      the image during resume, because they conflict with the pages that
 595  *      had been used before suspend
 596  */
 597
 598 static int mark_unsafe_pages(struct pbe *pblist)
 599 {
 600         struct zone *zone;
 601         unsigned long zone_pfn;
 602         struct pbe *p;
 603
 604         if (!pblist) /* a sanity check */
 605                 return -EINVAL;
 606
 607         /* Clear page flags */
 608         for_each_zone (zone) {
 609                 for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn)
 610                         if (pfn_valid(zone_pfn + zone->zone_start_pfn))
 611                                 ClearPageNosaveFree(pfn_to_page(zone_pfn +
 612                                         zone->zone_start_pfn));
 613         }
 614
 615         /* Mark orig addresses */
 616         for_each_pbe (p, pblist) {
 617                 if (virt_addr_valid(p->orig_address))
 618                         SetPageNosaveFree(virt_to_page(p->orig_address));
 619                 else
 620                         return -EFAULT;
 621         }
 622
 623         unsafe_pages = 0;
 624
 625         return 0;
 626 }
 627
 628 static void copy_page_backup_list(struct pbe *dst, struct pbe *src)
 629 {
 630         /* We assume both lists contain the same number of elements */
 631         while (src) {
 632                 dst->orig_address = src->orig_address;
 633                 dst = dst->next;
 634                 src = src->next;
 635         }
 636 }
 637
 638 static int check_header(struct swsusp_info *info)
 639 {
 640         char *reason = NULL;
 641
 642         if (info->version_code != LINUX_VERSION_CODE)
 643                 reason = "kernel version";
 644         if (info->num_physpages != num_physpages)
 645                 reason = "memory size";
 646         if (strcmp(info->uts.sysname,system_utsname.sysname))
 647                 reason = "system type";
 648         if (strcmp(info->uts.release,system_utsname.release))
 649                 reason = "kernel release";
 650         if (strcmp(info->uts.version,system_utsname.version))
 651                 reason = "version";
 652         if (strcmp(info->uts.machine,system_utsname.machine))
 653                 reason = "machine";
 654         if (reason) {
 655                 printk(KERN_ERR "swsusp: Resume mismatch: %s\n", reason);
 656                 return -EPERM;
 657         }
 658         return 0;
 659 }
 660
 661 /**
 662  *      load header - check the image header and copy data from it
 663  */
 664
 665 static int load_header(struct snapshot_handle *handle,
 666                               struct swsusp_info *info)
 667 {
 668         int error;
 669         struct pbe *pblist;
 670
 671         error = check_header(info);
 672         if (!error) {
 673                 pblist = alloc_pagedir(info->image_pages, GFP_ATOMIC, 0);
 674                 if (!pblist)
 675                         return -ENOMEM;
 676                 pagedir_nosave = pblist;
 677                 handle->pbe = pblist;
 678                 nr_copy_pages = info->image_pages;
 679                 nr_meta_pages = info->pages - info->image_pages - 1;
 680         }
 681         return error;
 682 }
 683
 684 /**
 685  *      unpack_orig_addresses - copy the elements of @buf[] (1 page) to
 686  *      the PBEs in the list starting at @pbe
 687  */
 688
 689 static inline struct pbe *unpack_orig_addresses(unsigned long *buf,
 690                                                 struct pbe *pbe)
 691 {
 692         int j;
 693
 694         for (j = 0; j < PAGE_SIZE / sizeof(long) && pbe; j++) {
 695                 pbe->orig_address = buf[j];
 696                 pbe = pbe->next;
 697         }
 698         return pbe;
 699 }
 700
 701 /**
 702  *      prepare_image - use metadata contained in the PBE list
 703  *      pointed to by pagedir_nosave to mark the pages that will
 704  *      be overwritten in the process of restoring the system
 705  *      memory state from the image ("unsafe" pages) and allocate
 706  *      memory for the image
 707  *
 708  *      The idea is to allocate the PBE list first and then
 709  *      allocate as many pages as it's needed for the image data,
 710  *      but not to assign these pages to the PBEs initially.
 711  *      Instead, we just mark them as allocated and create a list
 712  *      of "safe" which will be used later
 713  */
 714
 715 struct safe_page {
 716         struct safe_page *next;
 717         char padding[PAGE_SIZE - sizeof(void *)];
 718 };
 719
 720 static struct safe_page *safe_pages;
 721
 722 static int prepare_image(struct snapshot_handle *handle)
 723 {
 724         int error = 0;
 725         unsigned int nr_pages = nr_copy_pages;
 726         struct pbe *p, *pblist = NULL;
 727
 728         p = pagedir_nosave;
 729         error = mark_unsafe_pages(p);
 730         if (!error) {
 731                 pblist = alloc_pagedir(nr_pages, GFP_ATOMIC, 1);
 732                 if (pblist)
 733                         copy_page_backup_list(pblist, p);
 734                 free_pagedir(p, 0);
 735                 if (!pblist)
 736                         error = -ENOMEM;
 737         }
 738         safe_pages = NULL;
 739         if (!error && nr_pages > unsafe_pages) {
 740                 nr_pages -= unsafe_pages;
 741                 while (nr_pages--) {
 742                         struct safe_page *ptr;
 743
 744                         ptr = (struct safe_page *)get_zeroed_page(GFP_ATOMIC);
 745                         if (!ptr) {
 746                                 error = -ENOMEM;
 747                                 break;
 748                         }
 749                         if (!PageNosaveFree(virt_to_page(ptr))) {
 750                                 /* The page is "safe", add it to the list */
 751                                 ptr->next = safe_pages;
 752                                 safe_pages = ptr;
 753                         }
 754                         /* Mark the page as allocated */
 755                         SetPageNosave(virt_to_page(ptr));
 756                         SetPageNosaveFree(virt_to_page(ptr));
 757                 }
 758         }
 759         if (!error) {
 760                 pagedir_nosave = pblist;
 761         } else {
 762                 handle->pbe = NULL;
 763                 swsusp_free();
 764         }
 765         return error;
 766 }
 767
 768 static void *get_buffer(struct snapshot_handle *handle)
 769 {
 770         struct pbe *pbe = handle->pbe, *last = handle->last_pbe;
 771         struct page *page = virt_to_page(pbe->orig_address);
 772
 773         if (PageNosave(page) && PageNosaveFree(page)) {
 774                 /*
 775                  * We have allocated the "original" page frame and we can
 776                  * use it directly to store the read page
 777                  */
 778                 pbe->address = 0;
 779                 if (last && last->next)
 780                         last->next = NULL;
 781                 return (void *)pbe->orig_address;
 782         }
 783         /*
 784          * The "original" page frame has not been allocated and we have to
 785          * use a "safe" page frame to store the read page
 786          */
 787         pbe->address = (unsigned long)safe_pages;
 788         safe_pages = safe_pages->next;
 789         if (last)
 790                 last->next = pbe;
 791         handle->last_pbe = pbe;
 792         return (void *)pbe->address;
 793 }
 794
 795 /**
 796  *      snapshot_write_next - used for writing the system memory snapshot.
 797  *
 798  *      On the first call to it @handle should point to a zeroed
 799  *      snapshot_handle structure.  The structure gets updated and a pointer
 800  *      to it should be passed to this function every next time.
 801  *
 802  *      The @count parameter should contain the number of bytes the caller
 803  *      wants to write to the image.  It must not be zero.
 804  *
 805  *      On success the function returns a positive number.  Then, the caller
 806  *      is allowed to write up to the returned number of bytes to the memory
 807  *      location computed by the data_of() macro.  The number returned
 808  *      may be smaller than @count, but this only happens if the write would
 809  *      cross a page boundary otherwise.
 810  *
 811  *      The function returns 0 to indicate the "end of file" condition,
 812  *      and a negative number is returned on error.  In such cases the
 813  *      structure pointed to by @handle is not updated and should not be used
 814  *      any more.
 815  */
 816
 817 int snapshot_write_next(struct snapshot_handle *handle, size_t count)
 818 {
 819         int error = 0;
 820
 821         if (handle->prev && handle->page > nr_meta_pages + nr_copy_pages)
 822                 return 0;
 823         if (!buffer) {
 824                 /* This makes the buffer be freed by swsusp_free() */
 825                 buffer = alloc_image_page(GFP_ATOMIC, 0);
 826                 if (!buffer)
 827                         return -ENOMEM;
 828         }
 829         if (!handle->offset)
 830                 handle->buffer = buffer;
 831         if (handle->prev < handle->page) {
 832                 if (!handle->prev) {
 833                         error = load_header(handle, (struct swsusp_info *)buffer);
 834                         if (error)
 835                                 return error;
 836                 } else if (handle->prev <= nr_meta_pages) {
 837                         handle->pbe = unpack_orig_addresses(buffer, handle->pbe);
 838                         if (!handle->pbe) {
 839                                 error = prepare_image(handle);
 840                                 if (error)
 841                                         return error;
 842                                 handle->pbe = pagedir_nosave;
 843                                 handle->last_pbe = NULL;
 844                                 handle->buffer = get_buffer(handle);
 845                         }
 846                 } else {
 847                         handle->pbe = handle->pbe->next;
 848                         handle->buffer = get_buffer(handle);
 849                 }
 850                 handle->prev = handle->page;
 851         }
 852         handle->buf_offset = handle->page_offset;
 853         if (handle->page_offset + count >= PAGE_SIZE) {
 854                 count = PAGE_SIZE - handle->page_offset;
 855                 handle->page_offset = 0;
 856                 handle->page++;
 857         } else {
 858                 handle->page_offset += count;
 859         }
 860         handle->offset += count;
 861         return count;
 862 }
 863
 864 int snapshot_image_loaded(struct snapshot_handle *handle)
 865 {
 866         return !(!handle->pbe || handle->pbe->next || !nr_copy_pages ||
 867                 handle->page <= nr_meta_pages + nr_copy_pages);
 868 }